Overview

  • How does Capelli’s lemma reduce irreducibility of a composition?
  • Which critical-orbit conditions imply stability?
  • Can irreducibility fail after several successful iterates?
  • When is the number of irreducible factors bounded?

Stability and eventual stability

Fix a field KK, a polynomial f∈K[x]f\in K[x] of degree d≥2d\ge2, and a target a∈Ka\in K. The pair (f,a)(f,a) is stable if f∘n(x)−af^{\circ n}(x)-a is irreducible over KK for every n≥1n\ge1. It is eventually stable if the number of irreducible factors, counted with multiplicity, is bounded independently of nn. When a=0a=0, one speaks of stability of ff.

If f∘n(x)−a=∏igi(x)f^{\circ n}(x)-a=\prod_i g_i(x), then f∘(n+1)(x)−a=∏igi(f(x))f^{\circ(n+1)}(x)-a=\prod_i g_i(f(x)). Each old factor contributes at least one new factor. The factor count is therefore nondecreasing; boundedness means it eventually becomes constant.

References 1

Capelli’s lemma

Lemma

Let F∈K[x]F\in K[x] be irreducible of degree mm, let g∈K[x]g\in K[x] have degree d≥1d\ge1, and let α\alpha be a root of FF. Then F(g(x))F(g(x)) is irreducible over KK if and only if g(x)−αg(x)-\alpha is irreducible over K(α)K(\alpha).

Proof. Choose β\beta with g(β)=αg(\beta)=\alpha. Since α∈K(β)\alpha\in K(\beta),

[K(β):K]=[K(β):K(α)] [K(α):K]=m[K(β):K(α)].[K(\beta):K]=[K(\beta):K(\alpha)]\,[K(\alpha):K]=m[K(\beta):K(\alpha)].

The right side equals md=deg⁡(F∘g)md=\deg(F\circ g) exactly when β\beta has degree dd over K(α)K(\alpha). This is the required equivalence.

References 1

A quadratic criterion

Assume char⁡K≠2\operatorname{char}K\ne2 and put f=x2+cf=x^2+c. If

−c,f∘2(0),f∘3(0),…-c,\quad f^{\circ2}(0),\quad f^{\circ3}(0),\ldots

are all nonsquares in KK, then ff is stable over KK.

To prove this, the first condition makes ff irreducible. Suppose F=f∘(n−1)F=f^{\circ(n-1)} is irreducible, with n≥2n\ge2, and let α\alpha be a root. Capelli’s lemma asks whether x2−(α−c)x^2-(\alpha-c) is irreducible over K(α)K(\alpha). Since deg⁡F\deg F is even,

NK(α)/K(α−c)=F(c)=f∘n(0).N_{K(\alpha)/K}(\alpha-c)=F(c)=f^{\circ n}(0).

A square has square norm. A nonsquare value on the right therefore proves the required irreducibility. Over general fields this is a sufficient criterion; a square norm alone does not prove that α−c\alpha-c is a square.

References 1

Example

For f=x2+1f=x^2+1, let an=f∘n(0)a_n=f^{\circ n}(0). Then a1=1a_1=1 and an=an−12+1a_n=a_{n-1}^2+1. Since an−1≥1a_{n-1}\ge1, for every n≥2n\ge2 we have

an−12<an<(an−1+1)2.a_{n-1}^2<a_n<(a_{n-1}+1)^2.

The second inequality follows from (an−1+1)2−an=2an−1>0(a_{n-1}+1)^2-a_n=2a_{n-1}>0. Thus ana_n lies strictly between consecutive integer squares. An integer that is a square in Q\mathbb Q is already a square in Z\mathbb Z, so every ana_n, n≥2n\ge2, is a nonsquare in Q\mathbb Q. Together with the nonsquare −1-1, the quadratic criterion proves that all iterates of x2+1x^2+1 are irreducible over Q\mathbb Q.

References 1

Eisenstein’s criterion at every level

Let pp be prime and f=xd+pf=x^d+p, with d≥2d\ge2. Modulo pp,

f∘n(x)≡xdn.f^{\circ n}(x)\equiv x^{d^n}.

Also f∘n(0)≡p(modp2)f^{\circ n}(0)\equiv p\pmod{p^2}: this holds at n=1n=1, and follows at the next step from (pk)d+p≡p(modp2)(pk)^d+p\equiv p\pmod{p^2}. Every iterate is therefore Eisenstein at pp. This proves stability over Q\mathbb Q, and even irreducibility of every iterate over Qp\mathbb Q_p.

References 1

Irreducibility can fail later

For f=x2−x−1f=x^2-x-1, the first two iterates reduce modulo 22 to x2+x+1x^2+x+1 and x4+x+1x^4+x+1, both irreducible. But

f∘3(x)=(x4−3x3+4x−1)⋅(x4−x3−3x2+x+1).\begin{aligned}f^{\circ3}(x)={}&(x^4-3x^3+4x-1)\\&\cdot(x^4-x^3-3x^2+x+1).\end{aligned}

The quadratic reduction has no root. The quartic reduction has no root and is not divisible by the only irreducible binary quadratic, x2+x+1x^2+x+1; hence it is irreducible. Expanding the displayed product verifies the third-iterate factorization. Two successful levels do not prove stability for an arbitrary quadratic polynomial.

References 1

Local criteria for eventual stability

Jones–Levy prove a useful extension of the Eisenstein argument. Let vv be a discrete valuation on KK, let ff have good reduction at vv, and suppose that the reduction has only one geometric preimage of the reduction of aa, namely aˉ\bar a. If f(a)≠af(a)\ne a, then (f,a)(f,a) is eventually stable over KK. Good reduction means that the degree does not fall after reduction.

In particular, if f=xd+cf=x^d+c, c≠0c\ne0, and v(c)>0v(c)>0, then the reduction is xdx^d and the criterion applies at a=0a=0. For quadratic maps x2+1/cx^2+1/c over Q\mathbb Q, with odd integer c≠−1c\ne-1, their finite-residue-field criterion also proves eventual stability at 00.

References 1

Quadratic polynomials over Q

For x2+rx^2+r, with nonzero r∈Qr\in\mathbb Q not the reciprocal of an integer, eventual stability is known. The remaining family is x2+1/cx^2+1/c, where c∈Z∖{0,−1}c\in\mathbb Z\setminus\{0,-1\}. DeMark–Hindes–Jones–Misplon–Stoll–Stoneman prove that all iterates have at most four irreducible factors when ∣c∣≤109|c|\le10^9. Their argument also verifies, for ∣c∣≤101000|c|\le10^{1000}, that irreducibility of the second iterate implies stability.

The all-parameter statements remain conjectural in that work. The bound four is attained: for c=−64c=-64, every iterate of depth at least three has four irreducible factors. The methods combine critical values, Diophantine equations, and bounds that reduce infinitely many depths to finitely many checks.

References 2

A nonperiodic target need not suffice

The 2017 eventual-stability conjecture asked whether nonperiodicity of the target suffices over number fields. Tucker–Zhu’s October 2026 preprint gives a counterexample with f=x6f=x^6 and target −1-1. Here −1↦1↦1-1\mapsto1\mapsto1, so the target is not periodic. Yet

f∘n(x)+1=x6n+1=∏j=0nΦ2n+13j(x),f^{\circ n}(x)+1=x^{6^n}+1=\prod_{j=0}^{n}\Phi_{2^{n+1}3^j}(x),

where Φm\Phi_m is the mm-th cyclotomic polynomial. The n+1n+1 factors are distinct and irreducible over Q\mathbb Q. The identity follows from the divisors of 2⋅6n2\cdot6^n that do not divide 6n6^n, and proves unbounded factor count directly.

This example concerns a strictly preperiodic target. It does not disprove a statement restricted to targets with infinite forward orbit.

References 2

New reducible fibers under iteration

Write Red⁡f(Z)={a∈Z:f(x)−a is reducible over Q}\operatorname{Red}_f(\mathbb Z)=\{a\in\mathbb Z:f(x)-a\text{ is reducible over }\mathbb Q\}. Behajaina–König–Neftin’s March 2026 preprint claims that, if f∈Q[x]f\in\mathbb Q[x], deg⁡f>1\deg f>1, and ff has no composition factor of degree 22 or 44, then for each fixed n≥2n\ge2,

Red⁡f∘n(Z)∖Red⁡f(Z)is finite.\operatorname{Red}_{f^{\circ n}}(\mathbb Z)\setminus\operatorname{Red}_f(\mathbb Z)\quad\text{is finite}.

The finite exceptional set may depend on nn. This statement does not assert stability for a fixed target at every depth.

References 1

Related topics

References

  1. Eventually stable rational functions

    Jones, Rafe and Levy, Alon. International Journal of Number Theory, 13(9), 2299–2318, 2017.

    Read the source
  2. Polynomials with Special Regard to Reducibility

    Schinzel, Andrzej. Cambridge University Press, Encyclopedia of Mathematics and its Applications, vol. 77, 2000.

    Read the source
  3. An iterative construction of irreducible polynomials reducible modulo every prime

    Jones, Rafe. Journal of Algebra, 369, 114–128, 2012.

    Read the source
  4. Galois representations from pre-image trees: an arboreal survey

    Jones, Rafe. Publications mathématiques de Besançon. Algèbre et théorie des nombres, 107–136, 2013.

    Read the source
  5. The density of primes in orbits of z^d+c

    Hamblen, Spencer and Jones, Rafe and Madhu, Kalyani. International Mathematics Research Notices, 2015(7), 1924–1958, 2015.

    Read the source
  6. Eventually stable quadratic polynomials over Q

    DeMark, David and Hindes, Wade and Jones, Rafe and Misplon, Moses and Stoll, Michael and Stoneman, Michael. New York Journal of Mathematics, 26, 526–561, 2020.

    Read the source
  7. A Counterexample to the Eventual Stability Conjecture

    Tucker, Thomas J. and Zhu, Zheng. arXiv:2610.28688 [math.NT], 2026.

    Read the source
  8. The Davenport–Lewis–Schinzel problem on the reducibility of f(X)−g(Y)

    Behajaina, Angelot and König, Joachim and Neftin, Danny. arXiv, 2026.

    Read the source