---
title: Escaping Set in Transcendental Dynamics
url: https://www.emergentmind.com/topics/escaping-set
type: topic
---

# Escaping Set in Transcendental Dynamics

In iteration theory, an escaping set records points whose forward orbit tends to infinity or, more generally, leaves every compact region of phase space. For a transcendental entire function \(f\), with \(f^n\) denoting the \(n\)-th iterate, the escaping set is
\[
I(f)=\{z\in\mathbb C: f^n(z)\to\infty \text{ as } n\to\infty\}.
\]
This set is fundamental in transcendental dynamics: Eremenko proved that \(I(f)\neq\varnothing\) and that \(J(f)=\partial I(f)\), where \(J(f)\) is the Julia set. More recent work shows that its topology is substantially more intricate than a countable compact exhaustion would suggest: for every transcendental entire function, \(I(f)\) is not \(\sigma\)-compact, equivalently not \(F_\sigma\) in \(\mathbb C\) [2507.11370] [2006.16946].

## 1. Classical formulation and foundational results

For an entire function \(f\), iteration means repeated composition:
\[
f^{n}=f\circ f\circ \cdots \circ f \quad (n \text{ times}),
\]
and \(f^{n}(z)\) denotes the \(n\)-th iterate evaluated at \(z\). The Fatou set \(F(f)\) is the set where \(\{f^n\}\) is a normal family, and the Julia set is \(J(f)=\mathbb C\setminus F(f)\). In transcendental entire dynamics, the escaping set
\[
I(f)=\{z\in\mathbb C:f^{n}(z)\to\infty\}
\]
is a basic dynamical object because its elementary definition already yields global information about \(J(f)\) [2507.11370].

Two standard auxiliary growth functions are the maximum modulus
\[
M(r,f)=\max_{|z|=r}|f(z)|
\]
and its radial iterates \(M^{n}(R,f)\). Choosing \(R>0\) so that \(M(r,f)>r\) for \(r\ge R\), one obtains \(M^{n}(R,f)\to\infty\). This growth scale organizes finer escaping subsets, above all the fast escaping set.

In the Eremenko–Lyubich class
\[
B=\{f\text{ entire}: \mathrm{Sing}(f^{-1}) \text{ is bounded}\},
\]
a stronger inclusion holds: \(I(f)\subset J(f)\). Outside \(B\), escaping Fatou components may occur, including Baker domains and wandering domains, so the relation between \(I(f)\), \(F(f)\), and \(J(f)\) becomes more delicate [2507.11370].

## 2. Topological complexity and non-\(\sigma\)-compactness

A set \(S\) is \(\sigma\)-compact if it can be written as a countable union of compact sets,
\[
S=\bigcup_{j=1}^{\infty}K_j,
\]
with each \(K_j\) compact. For transcendental entire \(f\), the decisive theorem is that \(I(f)\) is not \(\sigma\)-compact. In \(\mathbb C\), this is equivalent to saying that \(I(f)\) is not an \(F_\sigma\) set [2006.16946].

The result is stronger than the bare non-\(\sigma\)-compactness of \(I(f)\). Let
\[
UO(f)=\{z\in\mathbb C:\{f^n(z):n\ge 0\}\text{ is unbounded}\}
\]
be the set of points with unbounded orbit, and let
\[
BU(f)=UO(f)\setminus I(f)
\]
be the bungee set. Then every \(\sigma\)-compact subset of \(UO(f)\) omits some points of \(I(f)\cap J(f)\) and of \(BU(f)\cap J(f)\). In particular, \(I(f)\), \(UO(f)\), \(BU(f)\), and their intersections with \(J(f)\) are all not \(\sigma\)-compact. Moreover, \(I(f)\cap J(f)\), \(UO(f)\cap J(f)\), and \(BU(f)\cap J(f)\) are nowhere \(\sigma\)-compact: any \(\sigma\)-compact subset has empty interior relative to these sets [2006.16946].

The proof combines two mechanisms. The first is a slow-escape theorem built from annular itineraries. Fix \(R>0\) so that \(M(r,f)>r\) for \(r\ge R\). Rippon–Stallard’s annular itinerary machinery allows orbits to linger for long blocks at a given modulus scale before rising. Consequently, for any sequence \((a_m)\) with \(a_m\to\infty\) and \(a_m\ge M_0\), there exist points \(\zeta\in I(f)\cap J(f)\) and \(\omega\in BU(f)\cap J(f)\) with
\[
R_0\le |f^m(\zeta)|\le a_m,\qquad R_0\le |f^m(\omega)|\le a_m
\]
for all \(m\ge 0\). The second mechanism is an exit-time obstruction for compact exhaustions: if \(X=\bigcup K_j\subset UO(f)\) is \(\sigma\)-compact, upper semicontinuity of exit times from discs \(D(0,M_j)\) yields a growth bound \(a_n\) that every \(K_j\) must violate, while the slow-escape theorem produces escaping points respecting that bound. Hence some escaping points avoid every \(K_j\) [2006.16946].

This topological complexity is visible already at the descriptive-set-theoretic level:
\[
I(f)=\bigcap_{M=0}^{\infty}\ \bigcup_{N=0}^{\infty}\ \bigcap_{n=N}^{\infty} f^{-n}(\mathbb C\setminus D(0,M)),
\]
so \(I(f)\) is an \(F_{\sigma\delta}\) set. The same work also shows that \(I(f)\) is never a \(G_\delta\) set [2006.16946].

## 3. Fast escape, spider’s webs, and Eremenko points

The fast escaping set is defined by
\[
A(f)=\{z\in\mathbb C:\exists \ell\in\mathbb N \text{ such that } |f^{n+\ell}(z)|\ge M^n(R,f)\text{ for all }n\in\mathbb N\},
\]
for sufficiently large \(R\). Its closed core levels are
\[
A_R(f)=\{z: |f^n(z)|\ge M^n(R)\text{ for }n\in\mathbb N\}.
\]
One has \(A(f)\subset I(f)\). In striking contrast with \(I(f)\), \(A(f)\) is \(F_\sigma\), hence \(\sigma\)-compact in \(\mathbb C\) [2006.16946].

A connected set \(E\subset\mathbb C\) is a spider’s web if there exists a sequence of bounded simply connected domains \((G_n)\) such that
\[
\partial G_n\subset E,\qquad G_n\subset G_{n+1},\qquad \bigcup_{n\in\mathbb N}G_n=\mathbb C.
\]
When \(A_R(f)\) is a spider’s web, the consequences are strong: \(A(f)\) and \(I(f)\) are also spider’s webs, Eremenko’s conjecture holds for that \(f\), and all Fatou components are bounded [1301.2676]. If \(f\) has a multiply connected Fatou component, then \(A_R(f)\) is a spider’s web; in this situation the geometry of \(A_R(f)\) is organized by fundamental holes \(H_R\) and fundamental loops \(L_R=\partial H_R\), and the rate-of-escape function
\[
R_A(z)=\max\{R:z\in A_R(f)\}
\]
links these loops to harmonic level sets inside multiply connected Fatou components [1301.2676].

Rippon–Stallard’s refined Eremenko-point construction sharpens the component picture. If \(I(f)\) is disconnected, then for any open disc \(D\) meeting \(J(f)\), the set \(I(f)\setminus D\) has uncountably many unbounded components. For the core \(A_R(f)\), there is a stronger dichotomy: for some \(R>0\), either \(A_R(f)\) is connected and has the structure of an infinite spider’s web, or it has uncountably many components, each of which is unbounded [1703.11001]. These results are obtained by combining Wiman–Valiron-based constructions of Eremenko points with separation arguments in the plane.

## 4. Geometric models: hairs, bouquets, wandering domains, dimension, and measure

In many classes of entire functions, escaping dynamics is organized by curves to infinity. For exponential maps \(f(z)=\lambda e^z\), trigonometric maps, and more generally finite-order functions in class \(B\) and finite compositions thereof, every escaping point lies on a hair along which \(f^n\to\infty\) uniformly. The resulting topology is often a Cantor bouquet: uncountably many disjoint curves to infinity together with their endpoints [2507.11370].

This bouquet geometry can coexist with spider’s-web geometry. For the family
\[
f(z)=\sum_{k=0}^{p-1}\exp(\omega^k z),\qquad \omega=\exp(2\pi i/p),\ p\ge 3,
\]
each of \(A(f)\), \(I(f)\), \(J(f)\cap A(f)\), \(J(f)\cap I(f)\), and \(J(f)\) is a spider’s web, and \(J(f)\) contains a Cantor bouquet; the curves minus the endpoints lie in \(A(f)\) [1908.07260]. This places dynamic-ray structure inside a globally connected web.

Multiply connected wandering domains provide a different source of escaping geometry. Every multiply connected Fatou component \(U\) is wandering, bounded, and lies in \(I(f)\); in fact \(U\subset A(f)\). Their forward images contain absorbing annuli, and the global consequences include spider’s-web structure for \(A(f)\), \(A_R(f)\), and \(I(f)\) [2507.11370]. The fine topology can nevertheless be complicated: multiply connected wandering domains can have complementary components with no interior, indeed uncountably many [1703.11001].

Quantitative size varies widely. For transcendental entire functions, \(\dim_H I(f)\ge 1\), and every \(d\in[1,2]\) occurs. For finite-order functions in class \(B\), \(\dim_H I(f)=2\). At the level of Hausdorff measure, slow escaping sets
\[
\mathrm{Esc}(f,(p_n))=\{z\in I(f): |f^n(z)|\le p_n \text{ for all large } n\}
\]
can have infinite \(H^h\)-measure for gauge functions with
\[
\lim_{t\to 0}\frac{\log h(t)}{\log t}=1,
\]
while definitive-speed escaping sets can have \(H^h\)-measure \(0\) for suitable \(f\in B\) when
\[
\lim_{t\to 0}\frac{h(t)}{t}=0
\]
[1203.0190]. In transcendental meromorphic dynamics, the range of escaping-set dimensions is even broader: within the Speiser class with at most four singular values,
\[
\{\dim_H(I(f)): f\in S_4\}=[0,2]
\]
[2011.08267].

## 5. Extensions beyond transcendental entire maps

For transcendental self-maps of the punctured plane \(\mathbb C^*=\mathbb C\setminus\{0\}\), where both \(0\) and \(\infty\) are essential singularities, the escaping set is defined by
\[
I(f)=\{z\in\mathbb C^*: \omega(z,f)\subset\{0,\infty\}\}.
\]
Escape is refined by an essential itinerary \(e\in\{0,\infty\}^{\mathbb N}\), with
\[
e_n=0 \text{ if } |f^n(z)|\le 1,\qquad e_n=\infty \text{ if } |f^n(z)|>1.
\]
There are corresponding fast escaping sets \(A_e(f)\), and for every itinerary \(e\),
\[
J(f)=\partial A_e(f)=\partial I_e(f).
\]
Moreover, every connected component of \(A_e(f)\) is unbounded, and there is an uncountable collection of disjoint sets of fast escaping points each of which has the Julia set as its boundary [1412.1032]. In the bounded-type class \(B^*\), escaping points lie in the Julia set, and for finite-order compositions every escaping point can be connected to \(0\) or \(\infty\) by a dynamic ray tail; for each essential itinerary \(e\), \(I_e(f)\) contains a Cantor bouquet [1603.03311].

Semigroup dynamics introduces several non-equivalent escaping-set notions. One definition is
\[
I(S)=\{z\in\mathbb C: f^n(z)\to\infty \text{ as } n\to\infty \text{ for all } f\in S\},
\]
which yields \(I(S)\subset I(f)\) for all \(f\in S\) and, for finitely generated bounded-type semigroups, \(I(S)\subset J(S)\) and \(J(S)=\overline{I(S)}\). If the semigroup is abelian, bounded type, and each generator is hyperbolic, then all components of \(I(S)\) are unbounded [1803.10381]. Another definition requires that every sequence in the semigroup admit a subsequence diverging to \(\infty\) at the given point; with this choice, \(I(G)\) is forward invariant and satisfies \(J(G)=\partial I(G)\) [1401.0425]. For non-abelian semigroups, a completely invariant escaping core \(K(S)\) has been introduced to recover complete invariance [1804.11252].

In higher-dimensional quasiregular dynamics, for a transcendental-type quasiregular map \(f:\mathbb R^m\to\mathbb R^m\), the escaping set remains
\[
I(f)=\{x\in\mathbb R^m: f^n(x)\to\infty\},
\]
and the fast escaping set admits equivalent definitions analogous to the plane:
\[
A(f)=A_1(f)=A_2(f).
\]
It is nonempty, every component of \(A(f)\) is unbounded, and under explicit minimum-modulus control \(A(f)\) is a spider’s web [1308.2860].

Topological dynamics broadens the notion further. For a flow \((X,T,\Phi)\) on a Hausdorff, first countable space, an escaping point is one whose orbit eventually stays outside every compact set in forward time, or in backward time when the flow is invertible. In a proper metric space, this compact-escaping definition is equivalent to divergence to infinity; it is a topological conjugacy invariant and is characterized by emptiness of the \(\omega\)-limit set [1904.12333]. For continuous maps \(f:\mathbb R^d\to\mathbb R^d\), the escaping set
\[
I(f)=\{x\in\mathbb R^d:\|f^n(x)\|\to\infty\}
\]
can be open, closed, or countable in dimensions \(d\ge 2\), in marked contrast with the transcendental entire setting [1601.04010].

In polynomial automorphisms of \(\mathbb C^2\), especially generalized Hénon maps, the forward escaping set is
\[
U^+=\{z\in\mathbb C^2:\|H^n(z)\|\to\infty\},
\]
and the non-escaping set is \(K^+=\mathbb C^2\setminus U^+\). Here the Green function
\[
G_H^+(z)=\lim_{n\to\infty}\frac{1}{d^n}\log^+\|H^n(z)\|
\]
satisfies \(U^+=\{G_H^+>0\}\) and \(K^+=\{G_H^+=0\}\). Recent rigidity results show that for polynomial automorphisms of positive entropy, every holomorphic automorphism of \(\mathbb C^2\) preserving \(U^+\) has the form \(L\circ H^s\), where \(L\) belongs to a finite cyclic group of affine maps preserving the escaping set [2601.07681].

## 6. Conjectures, counterexamples, and current directions

The central historical question was Eremenko’s conjecture: every connected component of \(I(f)\) is unbounded. Positive cases remain extensive. The conjecture holds for postsingularly bounded maps, including hyperbolic functions in class \(B\), and for finite-order maps in class \(B\), where dynamic-ray theory provides curves through escaping points [2507.11370].

At the same time, the modern picture is more nuanced. The strong form of the conjecture—that every point of \(I(f)\) lies on a curve to infinity—fails in general. More significantly, the conjecture itself fails in general: there exist transcendental entire functions with bounded components of \(I(f)\), even singleton components, although the known counterexamples lie outside class \(B\) and have infinite order [2507.11370]. This places recent structural theorems, such as the non-\(\sigma\)-compactness of \(I(f)\), in a setting where unboundedness of components can no longer be taken as universal [2006.16946].

Several problems remain central. The survey literature isolates the status of Eremenko’s conjecture in class \(B\) or \(S\), the finite-order case outside \(B\), the possibility of Jordan spider’s webs for canonical examples, the regularity of hairs, and sharp growth criteria for dimension and area of escaping sets [2507.11370]. A plausible implication is that the modern theory now splits into two complementary programs: one studies rigidity and geometric organization inside structured classes such as \(B\), \(B^*\), quasiregular fast-escape settings, and Hénon-type dynamics; the other studies how slowly escaping or topologically pathological orbits obstruct classical compact, connected, or ray-based models.

The escaping set therefore remains both a definition and a research program: a set given by the elementary condition \(f^n(z)\to\infty\), but one whose topology, geometry, and quantitative size continue to organize large parts of transcendental and non-compact dynamical systems.

Source: https://www.emergentmind.com/topics/escaping-set