Hyperbolic entire functions and the Eremenko-Lyubich class: Class $\mathcal{B}$ or not class $\mathcal{B}$?

Hyperbolicity plays an important role in the study of dynamical systems, and is a key concept in the iteration of rational functions of one complex variable. Hyperbolic systems have also been considered in the study of transcendental entire functions. There does not appear to be an agreed definition of the concept in this context, due to complications arising from the non-compactness of the phase space. In this article, we consider a natural definition of hyperbolicity that requires expanding properties on the preimage of a punctured neighbourhood of the isolated singularity. We show that this definition is equivalent to another commonly used one: a {\tef} is hyperbolic if and only if its postsingular set is a compact subset of the Fatou set. This leads us to propose that this notion should be used as the general definition of hyperbolicity in the context of entire functions, and, in particular, that speaking about hyperbolicity makes sense only within the \emph{Eremenko-Lyubich class} $\mathcal{B}$ of transcendental entire functions with a bounded set of singular values. We also considerably strengthen a recent characterisation of the class $\mathcal{B}$, by showing that functions outside of this class cannot be expanding with respect to a metric whose density decays at most polynomially. In particular, this implies that no transcendental entire function can be expanding with respect to the spherical metric. Finally we give a characterisation of an analogous class of functions analytic in a hyperbolic domain.


Introduction
It is a general principle in the investigation of dynamical systems that hyperbolic systems (also known as "Axiom A", following Smale [Sma67]) are the first class to understand in a given setting: they exhibit the simplest behaviour, yet their study leads to a better picture in greater generality.In this article, we consider (non-invertible) 2010 Mathematics Subject Classification.Primary 37F10; Secondary 30D05, 30D15, 30D20, 37F15.The first author was supported by a Philip Leverhulme Prize.The second author was supported by Engineering and Physical Sciences Research Council grant EP/J022160/1.dynamics in one complex variable.For rational maps, hyperbolic behaviour was understood, in general terms, already by Fatou [Fat20,, though, of course, he did not use this terminology.
More precisely, a rational map f : Ĉ → Ĉ is said to be hyperbolic if one of the following, equivalent, conditions holds [Bea91, Section 9.7] (see below for definitions): (a) the function f is expanding with respect to a suitable conformal metric defined on a neighbourhood of its Julia set; (b) every critical value of f belongs to the basin of an attracting periodic cycle; (c) the postsingular set is a subset of the Fatou set.
Moreover [McM94,Theorem 4.4] every hyperbolic rational map satisfies (d) f is stable; in other words, any nearby rational map is topologically conjugate to f on its Julia set, with the conjugacy depending continuously on the perturbation.
The famous Hyperbolicity Conjecture asserts that condition (d) is also equivalent to hyperbolicity; this question essentially goes back to Fatou.See the final sentence of Chapitre IV in [Fat20,p. 73], and also compare [McM94,Section 4.1] for a historical discussion.
The iteration of transcendental entire functions f : C → C also goes back to Fatou [Fat26], and has received considerable attention in recent years.As in the rational case, the Fatou set F (f ) ⊂ C of such a function is defined as the set of z ∈ C such that the iterates {f n } n∈N form a normal family in a neighbourhood of z.Its complement, the Julia set J(f ) . .= C \ F (f ), is the set where the dynamics is "chaotic".The role played by the set of critical values in rational dynamics is now taken on by the set S(f ) of (finite) singular values, i.e. the closure of the set of critical and asymptotic values of f (see Section 2).
In the transcendental setting -due to the effect of the non-compactness of the phase space and the essential singularity at infinity -it is not clear how "hyperbolicity" should be defined.Accordingly, there is no accepted general definition; see Appendix A for a brief historical discussion.We propose the following notion of "expanding" in this setting.

Definition (Expanding entire functions).
A transcendental entire function f is expanding if there exist a connected open set W ⊂ C, which contains J(f ), and a conformal metric ρ = ρ(z)|dz| on W such that: (1) W contains a punctured neighbourhood of infinity, i.e. there exists R > 0 such that z ∈ W whenever |z| > R; (2) f is expanding with respect to the metric ρ, i.e. there exists λ > 1 such that whenever z, f (z) ∈ W ; and (3) the metric ρ is complete at infinity, i.e. dist ρ (z, ∞) = ∞ whenever z ∈ W .
We make three remarks about this definition.First we note that the final condition holds whenever the metric ρ is complete.Indeed, as we shall see below, we could require that ρ is complete without changing the class of expanding functions.We prefer instead to use the weaker condition (3), which allows ρ to be the Euclidean metric.
Second, one might ask, in analogy to hyperbolicity for rational maps, for the metric ρ to be defined and conformal on a full (rather than punctured) neighbourhood of ∞.However, due to the nature of the essential singularity at infinity, such expansion can never be satisfied; see Corollary 1.6 below.
Finally, one might require only expansion on a neighbourhood of the Julia set, as a subset of the complex plane.This is too weak a condition.Due to the lack of uniformity, functions with this property may have infinitely many attractors, wandering domains or Baker domains: invariant domains of normality in which all orbits converge to infinity.(Compare Figure 1.)Such behaviour is not compatible with the usual picture of hyperbolicity and, moreover, usually not stable even under simple perturbations.
Our key observation about expanding entire functions is the following: 1.2.Theorem (Expansion only in the Eremenko-Lyubich class).
If f is expanding in the sense of Definition 1.1, then S(f ) is bounded.
In other words, every expanding function belongs to the Eremenko-Lyubich class Introduced in [EL92], B is a large and well-studied class.In particular, for functions f ∈ B there is a natural and well-established notion of hyperbolicity; see [RS99,Sta04].Theorem 1.2 implies that this definition coincides precisely with our notion of expansion.This suggests that only these entire functions ought to be classed as hyperbolic.
Moreover, each hyperbolic function is stable within its quasiconformal equivalence class.
Remark.Quasiconformal equivalence classes form the natural parameter spaces of transcendental entire functions.(See Proposition 3.2 for the formal meaning of the final statement in the theorem.)These classes were defined implicitly by Eremenko and Lyubich [EL92], and explicitly by the second author [Rem09], and provide the appropriate context in which to consider stability.
Observe that this quantity is precisely the derivative of f with respect to the cylindrical metric |dz|/|z| on the punctured plane.Recently, the second author [Six14] was able to show that the converse also holds: no function outside of class B exhibits this type of expansion near infinity.More precisely, he obtained the following characterisation of the Eremenko-Lyubich class.
Theorem A (Characterisation of class B).Suppose that f is a transcendental entire function, and consider the quantity Then, either η(f ) = ∞ and f ∈ B, or η(f ) = 0 and f / ∈ B.
Theorem 1.2 follows from the following strengthening of Theorem A.

Theorem (Shrinking property near infinity).
Let f be a transcendental entire (or meromorphic) function and s ∈ S(f ).Let U be an open neighbourhood of s.Suppose that ρ(z)|dz| is a conformal metric defined on Note that one obtains the second half of Theorem A by applying Theorem 1.4 to the cylindrical metric and a sequence of singular values tending to infinity.A similar application of Theorem 1.4 gives Theorem 1.2.
Clearly some condition must be imposed upon the metric ρ when U contains no critical values.Indeed, if we let ρ be the pull-back of the Euclidean metric under f , then (1.2) fails by definition.However, it appears plausible that such a pull-back must decay rapidly on some approach to infinity, and so the conclusion of Theorem 1.4 holds for certain metrics that are not complete at infinity.The following theorem, another considerable strengthening of Theorem A, shows that this is indeed the case.

Theorem (Metrics with polynomially decaying densities).
Let f be a transcendental entire (or meromorphic) function and s ∈ S(f ).If U is an open neighbourhood of s, then One obtains the second half of Theorem A from Theorem 1.5 by setting τ = 1.Similarly, one obtains Theorem 1.4 with ρ being the spherical metric by setting τ = 2.Moreover, if f is transcendental entire, then we can apply Theorem 1.5 to the function 1/f , the singular value s = 0 and τ = 2.As claimed above, this implies that the function f cannot be expanding with respect to the spherical metric, on the preimage of a full neighbourhood of ∞.

Corollary (No expansion for the spherical metric).
Let f be a transcendental entire function, and let R > 0. Then 1.7.Remark.Corollary 1.6 can also be seen to follow from Wiman-Valiron theory, which shows that there are points of arbitrarily large modulus near which f behaves like a monomial of arbitrarily large degree.It is straightforward to deduce (1.3) from this fact.(More precisely, the claim follows e.g. from formulae (3) and (4) in the statement of the Wiman-Valiron theorem in [Ere89,p. 340].)We are grateful to Alex Eremenko for this observation.
The Eremenko-Lyubich class on a hyperbolic domain.Finally, we consider the analogue of the class B for functions analytic in a hyperbolic domain.Suppose that Ω ⊂ C is a hyperbolic domain (or, more generally, a hyperbolic Riemann surface), and that f : Ω → C is analytic.We say that f belongs to the Eremenko-Lyubich class, B, if the set of singular values of f is bounded.We modify the definition (1.1) as follows: (1.4) η Ω (f ) . .= lim where the norm of the derivative is evaluated using the hyperbolic metric on Ω and the cylindrical metric on the range.
For the second part of Theorem 1.8, we could replace the hyperbolic metric by any complete metric on Ω, as in Theorem 1.4.However, Theorem 1.8 as stated provides an appealing dichotomy in terms of the conformally natural quantity η Ω (f ).
Ideas of the proofs.Our proof of Theorem 1.4 considerably simplifies the original proof of Theorem A, and can be summarized as follows.The set V must contain either a critical point of f or an asymptotic curve γ whose image is a line segment in U ; this is a classical and elementary fact, but its connection to the questions at hand appears to have been overlooked.Since the ρ-length of γ is infinite, while the Euclidean length of f (γ) is finite, the conclusion is immediate.This argument clearly does not apply to metrics that are not complete at infinity, and hence a more detailed analysis is required for the proof of Theorem 1.5.Once again, we rely on elementary mapping properties of functions near a singular value, s, that is not the limit of critical values.We show that there are infinitely many, pairwise disjoint, unbounded simply-connected domains on which the function in question is univalent, and which are mapped to round discs near s; see Corollary 2.9.Using a similar idea as in the classical proof of the Denjoy-Carleman-Ahlfors theorem, it follows that, within some of these domains, the function must tend very quickly towards a corresponding asymptotic value.This leads to the desired conclusion.
Finally, the second part of Theorem 1.8 follows in the same manner as Theorem 1.4.The first part, on the other hand, can be deduced in a similar way to the proof of [EL92, Lemma 1]), although we adopt a slightly different approach using basic properties of the hyperbolic metric.
Structure of the article.In Section 2 we give background on the notion of singular values, and prove some basic results.In Section 3, we deduce Theorems 1.2, 1.3 and 1.4.Sections 4 and 5 prove Theorems 1.5 and 1.8.Appendix A contains historical remarks concerning notions of hyperbolicity for entire functions, and Appendix B concerns the functions in Figure 1.
Basic background and notation.We denote the complex plane, the Riemann sphere and the unit disc by C, C and D respectively.For Euclidean discs, we use the notation If X is a Riemann surface, then we denote by ∞ X the added point in the one-point compactification of X.For example, if (z n ) is a sequence of points of X which eventually leaves any compact subset of X, then lim n→∞ z n = ∞ X .(Recall that, if X is already compact, then ∞ X is an isolated point of the one-point compactification.This allows for uniformity of statements and definitions.) Closures and boundaries are always taken in an underlying Riemann surface X; which X is meant should be clear from the context.
Suppose that X is a Riemann surface, and U ⊂ X is open.A conformal metric on U is a tensor that takes the form ρ(z)|dz| in local coordinates, with ρ a continuous positive function.When X = C, which is the case of most interest to us, we can express the metric globally in this form, and we do not usually distinguish between the metric and its density function ρ(z).
If γ ⊂ U is a locally rectifiable curve, then we denote the length of γ with respect to the metric ρ by ρ (γ).If z, w ∈ U , then we denote the distance from z to w with respect to the metric ρ by dist ρ (z, w); i.e. dist ρ (z, w) = inf γ ρ (γ), where the infimum is taken over all curves connecting z and w.Note that, by definition, this distance is infinite if z and w belong to different components of U .Abusing notation slightly, we define dist ρ (z, ∞ X ) . .= lim inf w→∞ X dist ρ (z, w).By definition, this quantity is infinite if U is relatively compact in X.
If X is a hyperbolic surface, then we write ρ X for the hyperbolic metric in X and, in local coordinates, denote its density function by ρ X (z).which we refer to [BE95], [BE08], [Ive14] and also [Ere].Instead, we only use some basic mapping properties of functions having singular values, which we derive here from first principles.

Definition (Singular values).
Let X and Y be Riemann surfaces, let f : X → Y be analytic, and let s ∈ Y .(a) s is called a regular value of f if there is an open neighbourhood U of s with the following property The sets of singular, critical and asymptotic values of f are denoted by S(f ), CV (f ) and AV (f ) respectively.

Remark (Comments on the definition).
Clearly CV (f ) ∪ AV (f ) ⊂ S(f ).On the other hand, critical and asymptotic values are dense in S(f ) (see Corollary 2.7 below), so that, in fact, S(f Equivalently to the definition above, S(f ) is the smallest closed subset S of Y such that the restriction f : Observe that any value in Y \ f (X) is a regular value of f .For example, if X = D, Y = C and f (z) . .= z, then the set of singular values of f coincides with the unit circle ∂D.
The following observation shows that, near any singular value, we can find inverse branches defined on round discs, and containing either critical or asymptotic values on their boundaries.

Lemma (Asymptotic values with well-behaved inverse branches).
Let X be a Riemann surface, and let f : X → D be analytic, but not a conformal isomorphism.Then there exists a round disc D with D ⊂ D, a branch ϕ : D → X of f −1 defined on D, and a singular value s ∈ ∂D such that either (a) lim z→s ϕ(z) is a critical point of f in X, and s is a critical value of f , or (b) lim z→s ϕ(z) = ∞ X .In particular, s is an asymptotic value of f , and there is an asymptotic curve γ that maps one-to-one onto a straight line segment f (γ) ending at s.
Proof.Fix some point z 0 ∈ X which is not a critical point of f .By postcomposing with a Möbius transformation, we may assume without loss of generality that f (z 0 ) = 0. Let ϕ be the branch of f −1 taking 0 to z 0 , and let r > 0 be the greatest value such that ϕ can be continued analytically to the disc D of radius r around 0. Then r < 1, since f is not a conformal isomorphism.It follows that there is a point s ∈ ∂D ∩ D such that ϕ cannot be continued analytically into s.
If lim z→s ϕ(z) = ∞ X , then we can take γ = ϕ(L), where L is the radius of D ending at s, and the proof of case (b) is complete.Otherwise, there is a sequence z n ∈ D with z n → s and ϕ(z n ) → c, for some c ∈ X.By continuity of f , we have f (c) = s.Moreover c is a critical point, since otherwise the local inverse of f that maps s to c would provide an analytic continuation of ϕ into s by the identity theorem.
Discs as in Lemma 2.3, on which branches of the inverse are defined and which have asymptotic values on their boundary, have appeared previously in the study of indirect asymptotic values in the sense of Iversen; see [BE95, Proof of Theorem 1] or [Zhe10, Theorem 6.2.3].To simplify subsequent discussions, we introduce the following terminology.

Definition (Discs of univalence).
Let X and Y be Riemann surfaces, and let f : X → Y be analytic.Suppose that D is an analytic Jordan domain such that D ⊂ Y , and that ϕ : D → X is a branch of f −1 defined on D. Suppose furthermore that there is an asymptotic value s ∈ ∂D such that lim z→s ϕ(z) = ∞ X .
Then we call D a disc of univalence at the asymptotic value s.We also call the domain V . .= ϕ(D) ⊂ X a tract over the disc of univalence D.

Remark (Asymptotic curves with well-behaved images).
Observe that, if s is an asymptotic value for which there exists a disc of univalence, then in particular there is an asymptotic curve γ for s that is mapped one-to-one to an analytic arc compactly contained in Y , with one endpoint at s.
The following observation is frequently useful.

Observation (Discs of univalence for a restriction).
Let X and Y be Riemann surfaces, and let f : X → Y be analytic.Let U ⊂ Y be a domain, and let W be a connected component of f −1 (U ).Then any disc of univalence of the restriction f : W → U at an asymptotic value s ∈ U is also a disc of univalence of f : X → Y at s.
Proof.Let ϕ : D → V ⊂ W be a branch of f −1 as in the definition of the disc of univalence D. Then lim z→s ϕ(z) = ∞ W . Since W was chosen to be a connected component of f −1 (U ), and so f (∂W ) ⊂ ∂U , we deduce that lim z→s ϕ(z) = ∞ X .

Corollary (Critical and asymptotic values are dense).
Let X and Y be Riemann surfaces, and let f : X → Y be analytic.Denote by AV (f ) the set of those asymptotic values at which there exists a disc of univalence.Then Proof.Let s ∈ S(f ), and let U be a simply-connected neighbourhood of s.By definition, there exists a connected component , where ϕ : U → D is a Riemann map, we find that either U contains a critical point of f , or that F , and hence f : V → U , has a disc of univalence.The claim follows from Observation 2.6.
We also note the following well-known fact.

Lemma (Isolated singular values).
Let X and Y be Riemann surfaces, and let f : X → Y be analytic.Let s be an isolated point of S(f ), let U be a simply-connected neighbourhood of s with U ∩ CV (f ) = ∅, and let V be a connected component of f −1 (U \ {s}).Then either (a) V is a punctured disc, and f : V → U \ {s} is a finite-degree covering map, or (b) V is simply-connected, and f : V → U \ {s} is a universal covering map.
Proof.By definition, f : V → U \ {s} is a covering map, and the only analytic coverings of the punctured disc are as stated [For99, Theorem 5.10].
We can now establish the following, which is crucial for the proof of Theorem 1.5.

Corollary (Disjoint tracts over discs of univalence).
Let X and Y be Riemann surfaces, and let f : X → Y be analytic, with no removable singularities at any punctures of X.Let s ∈ S(f ), and suppose that s has an open neighbourhood U with U ∩ CV (f ) = ∅.Then there exist infinitely many discs of univalence D i ⊂ U of f such that the corresponding tracts T i are pairwise disjoint.
Proof.If s is not an isolated point of S(f ), then the claim follows immediately from Corollary 2.7.Otherwise, let D be a round disc around s in a given local chart, and apply Lemma 2.8.By assumption, s is not a critical value, and f does not have any removable singularities.Hence, if V is any connected component of f −1 ( D), then f : V → D \ s is a universal covering.We may thus choose a single disc D ⊂ D that is tangent to s, set D i = D for all i, and let the T i be the infinitely many different components of f −1 (D) in X.

Proofs of Theorems 1.2, 1.3 and 1.4
The following is a general result concerning behaviour of analytic functions on sets that map close to singular values.

Proposition (Analytic functions contract when mapping near singular values).
Let X and Y be Riemann surfaces, let f : X → Y be analytic, let s ∈ S(f ), and let U be an open neighbourhood of s.Suppose that V is a connected component of f −1 (U ) such that f : V → U is not a conformal isomorphism, and that ρ is a conformal metric on V such that dist ρ (z, ∞ X ) = ∞ for all z ∈ V .Also let σ be any conformal metric on U .Then inf where the norm of the derivative is measured with respect to the metrics ρ and σ.
Proof.Suppose, by way of contradiction, that inf z∈V Df (z) > 0. Then f has no critical points in V .By Corollary 2.7 and Remark 2.5, there is an analytic curve γ ⊂ V to infinity such that γ is mapped one-to-one to an analytic arc L compactly contained in U .Then Proof of Theorem 1.4.This follows immediately from Proposition 3.1, by taking σ to be the Euclidean metric.
Remark.In particular, Theorem 1.4, remains true if we replace the full preimage V by any component on which f is not a conformal isomorphism.
Proof of Theorems 1.2 and 1.3.Let f be a transcendental entire function, and let ρ be a conformal metric, defined on an open neighbourhood W of J(f ) that contains a punctured neighbourhood of ∞.Suppose that dist ρ (z, ∞) = ∞ for all z ∈ W , and that f is expanding with respect to the metric ρ.If c is a critical point of f , then either c / ∈ W or f (c) / ∈ W .It follows that W contains only finitely many critical values, none of which lie in J(f ).By shrinking W , we may hence assume that W contains no critical values of f .Claim 1.We have W ∩ S(f ) = ∅.In particular, S(f ) is bounded.
Proof.Let w ∈ W and let R > 0 be such that A R . .= {z ∈ C : |z| > R} ⊂ W . Since w is not a critical value of f , if D ⊂ W is a sufficiently small disc around w, then every component of f −1 (D) that is not contained in A R is mapped to D as a conformal isomorphism.On the other hand, every component V of f −1 (D) that is contained in A R ⊂ W is also mapped as a conformal isomorphism, by Proposition 3.1 and the expanding property of f .Observe that this proves Theorem 1.2.
Proceeding inductively, we see that every branch of f −n can be defined on ∆ 0 , for all n ≥ 0. Hence ∆ 0 ∩ P (f ) = ∅, and in particular w / ∈ P (f ), as required.
Recall that every parabolic periodic cycle of a transcendental entire function, and the boundary of every Siegel disc, lies J(f ) ∩ P (f ) [Ber93,Theorem 7].Furthermore, the iterates in any wandering domain of f converge uniformly to (J(f

Proposition (Stability of hyperbolic functions).
Suppose that Λ is a complex manifold, and that (f λ ) λ∈Λ is a family of entire functions of the form λ , where ϕ λ , ψ λ : C → C are quasiconformal homeomorphisms depending analytically on the parameter λ and f is transcendental entire.If λ 0 ∈ Λ is a parameter for which f λ 0 is hyperbolic, then f λ 0 and f λ are quasiconformally conjugate on their Julia sets whenever λ is sufficiently close to λ 0 .Moreover, this conjugacy depends analytically on the parameter λ.
Proof.Let us first observe that hyperbolicity is an open property in any such family.Indeed, (c) in Theorem 1.3 is equivalent to the existence of a compact set K with f (K) ∪ S(f ) ⊂ int(K); see [BFR14, Proposition 2.1].Clearly, any function f that is sufficiently close to f in the sense of locally uniform convergence satisfies f (K) ⊂ int(K).If, furthermore, S( f ) is sufficiently close to S(f ) in the Hausdorff metric, then S( f ) ⊂ int(K), and hence f is also hyperbolic.
Therefore, if f belongs to any analytic family as in the statement of the theorem, then f has a neighbourhood in which all maps are hyperbolic.In particular, no function in this neighbourhood has any parabolic cycles, and it follows that the repelling periodic points of f move holomorphically over this neighbourhood.By the "λ-lemma" [MSS83], it follows that the closure of the set of repelling periodic points of f , i.e. the Julia set, also moves holomorphically.This yields the desired result.
We note that this argument uses the fact that the analytic continuation of a repelling periodic point encounters only algebraic singularities within a family as above.This is proved in [EL92, Section 4] for maps with finite singular sets; the same argument applies in our setting.Alternatively, the claim can also be deduced formally from the results of [Rem09], where it is shown that, for a given compact subset of the quasiconformal equivalence class, the set of points whose orbits remain sufficiently large moves holomorphically.

Metrics decaying at most polynomially
Proof of Theorem 1.5.Let f be a transcendental entire or meromorphic function, let s ∈ S(f ) be a finite singular value of f , and let U be an open neighbourhood of s.We may assume that there is a disc D ⊂ U around s that contains no critical values of f , as otherwise there is nothing to prove.
Let K be any positive integer.By Corollary 2.9, we can find K discs of univalence D 1 , . . ., D K ⊂ D, having pairwise disjoint tracts G 1 , . . ., G K .Let a 1 , . . .a K be the associated asymptotic values.Also, for 1 ≤ n ≤ K, let Γ n be the preimage in G n of the radius of D n ending at a n .
Then Γ n is an asymptotic curve for a n , f (Γ n ) is a straight line segment, and f (z) → a n as z → ∞ in Γ n .This is precisely the setting of the proof of [BE95, Theorem 1], and formula (12) in that paper shows that there exist an integer n and a sequence (w j ) of points tending to infinity on Γ n such that where p is any positive integer with 4p + 3 < K. (In [BE95], the function f is required to have order less than p − 3, and the values a j are assumed to be pairwise distinct. However, neither of these assumptions are required for the proof of formula (12).)Since K was arbitrary, the claim of the theorem follows.
The argument in [BE95] is essentially the same as in the proof of the classical Denjoy-Carleman-Ahlfors theorem: since the tracts G n are unbounded and pairwise disjoint, some of them must have a small average opening angle.By the Ahlfors distortion theorem, it follows that f must approach a n rapidly along Γ n , which is only possible if the derivative becomes quite small along this curve.
For the reader's convenience, we present a self-contained proof of Theorem 1.5, following the same idea.Since we are not interested in precise estimates, we replace the use of the Ahlfors distortion theorem by the standard estimate on the hyperbolic metric in a simply-connected domain.Let R 0 > 1 be sufficiently large to ensure that each Γ n contains a point of modulus R 0 .For each n, and each z ∈ Γ n , let us denote by Γ + n (z) the piece of Γ n connecting z to ∞, and the complementary bounded piece by Γ − n (z).Suppose that the conclusion of the theorem did not hold.Then for some τ > 1 and all z ∈ Γ n .This implies We now prove that, if K was chosen large enough, depending on τ , then such an estimate cannot hold for all Γ n .Indeed, for x ≥ R 0 , let ϑ n (x) denote the angular measure of the set {ϑ : We are interested in the reciprocals 1/ϑ n (x), since these allow us to estimate the density of the hyperbolic metric in G m .Indeed, for |z| ≥ R 0 , it follows from the fact that G m is simply connected and [CG93, Theorem I.4.3] that By (4.2) and the Cauchy-Schwarz inequality, we have Thus there must be a choice of n and a sequence w j → ∞ in Γ n such that ) is a radial segment connecting the centre of D n to f (w j ), we deduce that Hence we see that (4.1) cannot hold for τ < 1 + K/(4π).Since K was arbitrary, this completes the proof.

Remark.
We could have formulated a more general version of Theorem 1.5, where f is an analytic function between Riemann surfaces X and Y , and ρ is a conformal metric on X that is complete except at finitely many punctures of X, where the metric is allowed to decay at most polynomially.

The Eremenko-Lyubich class on a hyperbolic surface
Proof of Theorem 1.8.Let Ω be a hyperbolic surface, and let f : Ω → C be analytic.First suppose that S(f ) is unbounded.For every R > 0, we may apply Proposition 3.1 to f , taking s to be an element of S(f ) with |s| > R, taking ρ to be the hyperbolic metric on Ω, and σ the cylindrical metric on C. The fact that η Ω (f ) = 0 follows.
On the other hand, suppose that S(f ) is bounded, and let R 0 > max s∈S(f ) |s|.Consider the domain U . .= {z ∈ C : |z| > R 0 }, and its preimage V . .= f −1 (U ).If V is a connected component of V, then f : V → U is an analytic covering map, and hence a local isometry of the corresponding hyperbolic metrics.
By the Schwarz lemma [CG93, Theorem I.4.2],we know that ρ V ≥ ρ Ω .On the other hand, up to a normalisation factor, the density of the hyperbolic metric of U is given by [Hay89, Example 9.10] Hence, the derivative of f , measured with respect to the hyperbolic metric on Ω and the cylindrical metric on C, satisfies as |f (z)| → ∞ (where the second term should be understood in a local coordinate for Ω near z).Hence η Ω (f ) = ∞, as claimed.

Appendix A: Definitions of hyperbolicity
It seems that the first discussion of hyperbolicity in the transcendental case was given by Rippon and Stallard [RS99].They argued that functions in the class B that satisfy (c) in Theorem 1.3 deserve to be called hyperbolic.To do so, they proved [RS99, Theorem C] that such functions have an expanding property at points in the Julia set, with respect to the cylindrical metric.However, they left open the possibility that other functions might also be classed as "hyperbolic".
Mayer and Urbanski [MU10] gave a definition of hyperbolicity that relies on the Euclidean metric.Specifically, they require both uniform expansion on the Julia set and that the postsingular set be a definite (Euclidean) distance away from the Julia set.Without additional requirements, this includes such functions as our examples in the introduction, which exhibit "non-hyperbolic" phenomena and are not stable under simple perturbations; see Appendix B.
With an additional strong regularity assumption near the Julia set, Mayer and Urbanski obtained strong and striking results concerning the measurable dynamics of "hyperbolic" functions in the above sense; further properties of functions with these properties are described in [Bad08].We are, however, not aware of any examples outside of the class B where these assumptions are known to hold.On the other hand, if f ∈ B is hyperbolic in the sense of Mayer and Urbanski, then it is also hyperbolic in the sense of Theorem 1.3.
We note that all these definitions consider, more generally, transcendental meromorphic functions.In this setting, there is a third, and strongest, notion of hyperbolicity, studied by Zheng [Zhe15].This requires, in addition, that infinity is not a singular value, and hence can never be satisfied when f is transcendental entire.An example of a function with this property is f (z) = λ tan z, for λ ∈ (0, 1).(We note that Zheng also studied the definitions given by Rippon and Stallard and by Mayer and Urbanski; we refer to his paper for further details.) Zheng's definition of "hyperbolicity on the Riemann sphere" may be considered to correspond most closely to the case of hyperbolic rational functions.In particular, Zheng shows that a meromorphic function is hyperbolic in this sense if and only if it satisfies a certain uniform expansion property with respect to the spherical metric.
We note that our results also apply in the setting of meromorphic functions, with appropriate modifications of definitions to correctly handle prepoles.Once again, they indicate that the notion of hyperbolicity does not make sense outside of the class B.
We recall also our earlier comment that, although our applications are to meromorphic functions on subsets of the complex plane, our basic results regarding the properties of singularities are given for analytic functions between Riemann surfaces.A generalisation of the study of transcendental dynamics into this setting are the classes of "finite type maps", and more generally "Ahlfors islands maps", suggested by Epstein [Eps95,EO].If W is a Riemann surface and X is a compact Riemann surface, then an analytic function f : W → X is called a finite-type map if S(f ) is finite and f has no removable singularities at any punctures of W .In the case where W ⊂ X, Epstein develops an iteration theory that carries over all of the basic results from the theory of rational dynamics and of entire and meromorphic functions with finitely many singular values.
Similarly to the case of meromorphic functions, we could again consider two different notions of hyperbolicity in this setting: one, analogous to that of Zheng, where all singular values must lie in attracting basins or map outside of the closure of W ; and a weaker notion, in analogy to that of Rippon and Stallard, where we allow singular values to lie on (or map into) the boundary of the domain of definition.All the standard results for the case of meromorphic functions should extend to this setting also.
The larger class of Ahlfors islands maps includes all transcendental meromorphic functions, as well as all finite type maps.It is tempting to define an "Eremenko-Lyubich class" of such maps, consisting of those for which S(f ) ∩ W is a compact subset of the domain of definition W .Our results still apply in this setting, and imply that hyperbolicity is only to be found within this class.However, it is no longer clear that functions within this class have suitable expansion properties near the boundary, and hence dichotomies such as that of Theorem A break down.Finding a natural class that extends both the Eremenko-Lyubich class of entire functions and all finite type maps appears to be an interesting problem.
Appendix B: Non-hyperbolic functions that are expanding near the Julia set In this section, we briefly discuss the three functions mentioned in the introduction.All three are well-studied, and have properties that are not compatible with what would normally be considered hyperbolic behaviour.
B.1.Proposition (Dynamical properties of the functions f p ).
We have the following: (a) f 1 has a Baker domain containing the right half-plane H . .= {z : Re(z) > 0}.
Proof.The function f 1 was first studied by Fatou.The stated property is well-known and can be verified by an elementary calculation.
The function f 2 is precisely Newton's method for finding the points where e z = 1; it was studied in detail by Weinreich [Wei90].Clearly it follows directly from the definition that the points z n are indeed superattracting fixed points.
The function f 3 is a well-known example of a transcendental entire functions with wandering domains, first described by Herman; see [Bak84, Example 2 on p. 564] and [Ber93, Section 4.5].The fact that J(f 3 ) = J(f 2 ) follows from the relations f 2 (z +2πi) = f 2 (z) + 2πi and f 3 (z) = f 2 (z) + 2πi.Since the points z n all belong to different Fatou components for f 2 , the same is true for f 3 .Since f 3 (z n ) = z n+1 , they do indeed belong to an orbit of wandering domains for f 3 .
We now justify the claim, made in the introduction, that these functions are expanding on a complex neighbourhood of the Julia set.

B.2. Proposition (Expansion properties of the functions f p ).
For each p ∈ {1, 2, 3}, there is an open neighbourhood U of J(f p ) in C with f −1 (U ) ⊂ U and a conformal metric ρ on U such that f p is uniformly expanding with respect to ρ.
Proof.Writing w = −e −z , the function f 1 is semi-conjugate to  This is a hyperbolic entire function, since its unique asymptotic value 0 is an attracting fixed point, and its unique critical point −1 belongs to the basin of attraction of this fixed point.By [Rem09, Lemma 5.1], the function F 1 is expanding on a neighbourhood of its Julia set, with respect to a suitable hyperbolic metric.Pulling back this metric under the semiconjugacy, we obtain the desired property for f 1 .
(We remark that, alternatively, one can show directly that f 1 is expanding with respect to the Euclidean metric, when restricted to a suitable neighbourhood of the Julia set.) The argument for f 2 and f 3 is analogous.Both functions are semi-conjugate to the map This function is not hyperbolic, as the asymptotic value 0 is a repelling fixed point.However, we note that this asymptotic value does not correspond to any point in the z-plane under the semiconjugacy.One can hence think of F 2 as being hyperbolic as a self-map of C * = C \ {0}, since the critical point −1 is a superattracting fixed point.The same proof as in [Rem09] yields a neighbourhood Ũ of J(F 2 ) \ {0} and a conformal metric on U such that F 2 is expanding, and the claim follows.

B.3. Proposition (Instability of the functions f p ).
Let p ∈ {1, 2, 3}.Then f p is not stable in the family (λf p ) λ∈C ; more precisely, there exist values of λ arbitrarily close to 1 such that λf p and f p are not topologically conjugate on their Julia sets.

1. 3 .
Theorem and Definition (Hyperbolic entire functions).All of the following properties of a transcendental entire function f : C → C are equivalent.If one, and hence each, of these conditions holds, then f is called hyperbolic.(a) f is expanding in the sense of Definition 1.1; (b) f is expanding in the sense of Definition 1.1, and (for suitably chosen W ) the metric ρ can be chosen to be the hyperbolic metric on W ; (c) f ∈ B and every singular value belongs to the basin of an attracting periodic cycle; (d) the postsingular set

Figure 1 .
Figure 1.Entire functions that are expanding on a neighbourhood of the Julia set, but should not be considered hyperbolic.The functions f 1 (z) = z +1+e −z , f 2 (z) = z −1+e −z and f 3 (z) = z −1+2πi+e −z can be seen to be expanding on neighbourhoods (in C) of their Julia sets (shown in black).f 1 has a Baker domain, f 2 has infinitely many superattracting fixed points and f 3 has wandering domains.None of the functions f p is stable under perturbations within the family z → λf p , λ ∈ C. See Appendix B for details, and compare also Figure 2.
has no Fatou component on which the iterates converge uniformly to infinity [EL92, Theorem 1].It follows that f has no wandering domains or Baker domains.Thus we conclude that F (f ) is a union of attracting basins, and hence every point of S(f ) ⊂ C \ W ⊂ F (f ) belongs to an attracting basin.This completes the proof that (a) implies (c) in Theorem 1.3.The equivalence of (c) and (d) is well-known; see e.g.[RS99] or [BFR14, Section 2].That these in turn imply expansion in the sense of Definition 1.1, with ρ the hyperbolic metric on a suitable domain W , was shown in [Rem09, Lemma 5.1].The final claim regarding stability is made precise by Proposition 3.2 below.

Figure 2 .
Figure 2. Instability of f 1 , illustrating Proposition B.3.Shown are the Julia sets (in black) of J(λf 1 ) for two parameter values close to λ = 1, illustrating Proposition B.3.The second parameter is chosen so that f 2 λ (z 1000 ) = ϕ(λ), as in the proof of the proposition.
Expansion near infinity.It is well-known that functions in the Eremenko-Lyubich class have strong expansion properties near ∞.Indeed, if f ∈ B, then Eremenko and