Overview

  • Why does x+1x+1 divide An(x)A_n(x) exactly when nn is even?
  • Why are ApA_p and Ap+1/(x+1)A_{p+1}/(x+1) irreducible for odd primes pp?
  • How does reciprocal symmetry restrict their Galois groups?

Definition and normalization

For nβ‰₯1n\ge1, define An(x)=βˆ‘ΟƒβˆˆSnxdes⁑(Οƒ)A_n(x)=\sum_{\sigma\in S_n}x^{\operatorname{des}(\sigma)}, where a descent is an index ii with Οƒ(i)>Οƒ(i+1)\sigma(i)>\sigma(i+1). Their generating function is

βˆ‘kβ‰₯1knxkβˆ’1=An(x)(1βˆ’x)n+1.\sum_{k\ge1}k^n x^{k-1}=\frac{A_n(x)}{(1-x)^{n+1}}.

Complementing the values of a permutation gives xnβˆ’1An(xβˆ’1)=An(x)x^{n-1}A_n(x^{-1})=A_n(x). Set A~n=An/(x+1)\widetilde A_n=A_n/(x+1) for even nn, and A~n=An\widetilde A_n=A_n for odd nn. This notation distinguishes the normalized Eulerian polynomial from the Appell Euler polynomial EnE_n. For nβ‰₯3n\ge3, A~n\widetilde A_n is monic, reciprocal, and has positive even degree.

References 1

Roots and reciprocity

The recurrence

An=(1+(nβˆ’1)x)Anβˆ’1+x(1βˆ’x)Anβˆ’1β€²,A1=1,A_n=(1+(n-1)x)A_{n-1}+x(1-x)A'_{n-1},\qquad A_1=1,

gives simple negative roots by induction. At each negative root of Anβˆ’1A_{n-1}, the factor x(1βˆ’x)x(1-x) reverses the sign of its derivative. Together with the signs at 00 and at negative infinity, this places one root of AnA_n in each intervening interval and one on each outside interval. Reciprocity then pairs roots by inversion. Since An(1)=n!A_n(1)=n!, the only possible fixed point of inversion is βˆ’1-1. It occurs exactly when nn is even, and simplicity makes its multiplicity one.

References 1

Limiting zero distribution

Let ρn,1,…,ρn,nβˆ’1<0\rho_{n,1},\ldots,\rho_{n,n-1}<0 be the roots of AnA_n. Then

1nβˆ’1βˆ‘j=1nβˆ’1Ξ΄βˆ’Οn,j ⟹ 1t>0t(Ο€2+log⁑2t) dt.\frac1{n-1}\sum_{j=1}^{n-1}\delta_{-\rho_{n,j}}\ \Longrightarrow\ \frac{\mathbf1_{t>0}}{t(\pi^2+\log^2t)}\,dt.

Melotti gives a combinatorial proof of this log-Cauchy limit, using exact stabilized moments after the transformation u=(1βˆ’Ο)βˆ’1u=(1-\rho)^{-1}. His convention uses xAn(x)xA_n(x), including a zero root; deleting that one atom and changing the normalization leaves the weak limit unchanged. The limiting distribution does not determine the Galois orbits or settle the irreducibility conjecture.

References 1
  • Melotti2025Eulerian Β· Theorems 1–2; arXiv:2507.15908v2The limiting law is older; the paper supplies a new proof via stabilized moments.Source

Prime indices

Theorem

For every odd prime pp, ApA_p is irreducible over Q\mathbb Q.

Fermat's congruence in the generating identity gives

Ap(x)(1βˆ’x)p+1β‰‘βˆ‘kβ‰₯1kxkβˆ’1=1(1βˆ’x)2(modp).\frac{A_p(x)}{(1-x)^{p+1}}\equiv\sum_{k\ge1}kx^{k-1}=\frac1{(1-x)^2}\pmod p.

Therefore Ap(1+t)≑tpβˆ’1(modp)A_p(1+t)\equiv t^{p-1}\pmod p. The shifted polynomial is monic, while its constant term is Ap(1)=p!A_p(1)=p!, divisible by pp but not p2p^2. Eisenstein applies. This is the r=1r=1 instance of Sharma–Straus, Theorem 6.

References 1

A large irreducible factor

Theorem

For nβ‰₯3n\ge3, let pp be the largest prime at most nn. The normalized polynomial A~n\widetilde A_n has an irreducible factor over Q\mathbb Q of degree at least pβˆ’1p-1. In particular, A~p\widetilde A_p and A~p+1\widetilde A_{p+1} are irreducible for every odd prime pp.

Put r=nβˆ’p+1r=n-p+1. Bertrand's postulate gives 1≀r<p1\le r<p and n<2pn<2p. Fermat's congruence and the generating identity imply

An(1+t)≑tpβˆ’1Ar(1+t)(modp).A_n(1+t)\equiv t^{p-1}A_r(1+t)\pmod p.

Since Ar(1)=r!A_r(1)=r! is a p-adic unit, the coefficient of tpβˆ’1t^{p-1} is a unit and every preceding coefficient is divisible by pp. The constant term n!n! has valuation one. The first Newton edge is therefore from (0,1)(0,1) to (pβˆ’1,0)(p-1,0). It gives an irreducible factor of degree pβˆ’1p-1 over Qp\mathbb Q_p. Some rational irreducible factor must contain it, hence has at least this degree. At even nn, removing x+1x+1 removes only a local linear factor with slope zero, because pp is odd.

Ta Thi Hoai An and Ngo Quoc Hoan attribute this bound to Heidrich; the original full text was not accessible in this review. It does not exclude an additional smaller factor at every composite index.

References 3

Example

Example

A5=x4+26x3+66x2+26x+1=x2q(x+xβˆ’1),q(y)=y2+26y+64.A_5=x^4+26x^3+66x^2+26x+1=x^2q(x+x^{-1}),\quad q(y)=y^2+26y+64.

The splitting field of qq is M=Q(105)M=\mathbb Q(\sqrt{105}). If its roots are y1,y2y_1,y_2, then

(y12βˆ’4)(y22βˆ’4)=q(2)q(βˆ’2)=1920=64β‹…30.(y_1^2-4)(y_2^2-4)=q(2)q(-2)=1920=64\cdot30.

Each radicand has norm 19201920, a nonsquare in Q\mathbb Q, so neither is a square in MM. Their product is also a nonsquare in MM: a rational number is a square in Q(105)\mathbb Q(\sqrt{105}) only if it or its quotient by 105105 is a rational square; neither condition holds for 19201920. The two classes are independent, so the splitting field has degree 2β‹…4=82\cdot4=8. Thus Gal⁑(A5/Q)=C2≀S2\operatorname{Gal}(A_5/\mathbb Q)=C_2\wr S_2, the dihedral group of order eight.

Modular factorizations

Modulo 2

(x+1)4(x + 1)^{4}
Factor degrees
1 + 1 + 1 + 1
Squarefree reduction
No
Possible proper factor degrees
1, 2, 3

The reduction has repeated factors, so the Frobenius cycle criterion does not apply. It still restricts factor degrees because the degree is preserved.

Proof by reduction modulo primes

The irreducible reduction modulo 11 supplies a 4-cycle and proves irreducibility over Q\mathbb Q. The factorization modulo 23 supplies a transposition.

Reciprocity bounds the group by C2≀S2C_2\wr S_2, of order 8. A 4-cycle generates a subgroup of order 4, and a single transposition lies outside that cyclic subgroup. The generated subgroup has order 8, so it is the full wreath product.

References 1

General indices

The general question is whether A~n\widetilde A_n is irreducible over Q\mathbb Q for every nβ‰₯3n\ge3. The prime and prime-plus-one cases establish this in two infinite sets of indices. The largest-prime bound gives a large irreducible factor at every index, but can leave room for smaller factors.

Determining the Galois group requires two further steps: find the group of the trace polynomial, then determine the square-class relations in the quadratic extensions above its roots.

References 1

Questions

OpenPolynomial families

Is the normalized Eulerian polynomial irreducible for every index nβ‰₯3?

Scope. The classical r=1 Euler–Frobenius specialization over Q; remove x+1 at even indices.

The prime and prime-plus-one cases and the general large-factor bound are proved above. The all-index statement is a conjecture in the cited 1974 source; this status does not exclude subsequent advances.

References 2
Discuss this question β†’
Research directionPolynomial families

For which indices is the group the full signed-permutation group?

Scope. The separable reciprocal polynomial A~n\widetilde A_n, of degree 2m.

The group is determined by the action on the degree-m trace polynomial and the rank of the square classes of the quadratic radicands. The index-five example computes both; this computation does not give a uniform argument.

References 1
Discuss this question β†’

Related topics

References

  1. ThΓ©orie gΓ©omΓ©trique des polynΓ΄mes eulΓ©riens

    Foata, Dominique and SchΓΌtzenberger, Marcel-Paul. Springer, Lecture Notes in Mathematics, vol. 138, 1970.

    Authorized digital retypesetting
  2. Distribution of the roots of Eulerian polynomials

    Melotti, Paul. Electronic Communications in Probability, 30, 1–7, 2025.

    Read the source
  3. On the irreducibility of a class of Euler Frobenius polynomials

    Sharma, A. and Straus, Ernst G. Canadian Mathematical Bulletin, 17(2), 265–273, 1974.

    Read the source
  4. On the factorization of Eulerian polynomials

    Heidrich, ArnoΕ‘t J. J. Journal of Number Theory, 18(2), 157–168, 1984.

    Read the source
  5. Multiple Eulerian polynomials

    Ta, Thi Hoai An and Ngo, Quoc Hoan. The Journal of Analysis, 34, 21–34, 2026.

    Read the source
  6. Galois groups of reciprocal polynomials and the van der Waerden-Bhargava theorem

    Anderson, Theresa C. and Bertelli, Adam and O'Dorney, Evan M. Algebra & Number Theory, 20(3), 603–628, 2026.

    Read the source