Stable Components in the Parameter Plane of Transcendental Functions Of Finite Type

We study the parameter planes of certain one-dimensional, dynamically-defined slices of holomorphic families of entire and meromorphic transcendental maps of finite type. Our planes are defined by constraining the orbits of all but one of the singular values, and leaving free one asymptotic value. We study the structure of the regions of parameters, which we call shell components, for which the free asymptotic value tends to an attracting cycle of non-constant multiplier. The exponential and the tangent families are examples that have been studied in detail, and the hyperbolic components in those parameter planes are shell components. Our results apply to slices of both entire and meromorphic maps. We prove that shell components are simply connected, have a locally connected boundary and have no center, i.e., no parameter value for which the cycle is superattracting. Instead, there is a unique parameter in the boundary, the virtual center, which plays the same role. For entire slices, the virtual center is always at infinity, while for meromorphic ones it maybe finite or infinite. In the dynamical plane we prove, among other results, that the basins of attraction which contain only one asymptotic value and no critical points are simply connected. Our dynamical plane results apply without the restriction of finite type.


Introduction
Our starting point is a holomorphic family F = { f (x, z) = f x (z) : X × C → C} of transcendental entire or meromorphic functions defined over a complex analytic manifold, the parameter space. By a meromorphic map we mean a map with at least one pole which is not omitted. Iterating f x for a given value of x gives rise to a dynamical system. A fundamental problem is to understand what bifurcations occur in the dynamics as one varies the parameter x. In this generality, the problem is very difficult and no results exist. Because the dynamics of these systems are controlled by the orbits of the singular values of the maps f x , one effective way to approach the problem is to define one-dimensional subspaces, or "slices" through the manifold X that constrain all but one of these singular values and analyze the structure within these slices. For the family of quadratic polynomials this analysis has been carried out more or less completely. (See for example [18,27,28]). For polynomials of higher degree and rational maps, this is also true to a lesser extent (see e.g., [7,12,13,49]), and this analysis shows that behavior for quadratics is not special, but is a paradigm for rational dynamics as well. To wit, one often sees copies of Mandelbrot sets in one-dimensional slices of rational or transcendental families; the slice of degree two rational maps with an attracting fixed point and fixed multiplier, or Newton's method for cubic polynomials are examples of this [29,36]).
The exponential and the tangent family were the first examples of transcendental entire and meromorphic functions respectively to be systematically studied (e.g. [17,26,41,50]). The essential new feature for transcendental functions is that they have asymptotic values, which are, in effect, images of infinity along certain paths. This introduces new phenomena into the structure of the parameter spaces that are not seen for rational maps. In the same way that the Mandelbrot-like sets code the influence of the critical values of the dynamics on rational maps, these new phenomena code the influence of the asymptotic values on the dynamics of our families of transcendental maps.
This work is the first to find a context for studying parameter spaces of large classes of both transcendental entire and meromorphic maps. Our goal is to show that in holomorphic families of such functions, there are natural one-dimensional slices in the parameter space in which there are always certain kinds of components whose properties are characterized by the existence of asymptotic values.
We define a one-dimensional slice by allowing only one "free" asymptotic value to vary. Our main results concern properties of the parameter regions for which this free asymptotic value is attracted to an attracting cycle. These regions, which we call "shell components", are the analogues in this setting of the hyperbolic components of the interior of the Mandelbrot-like sets for rational maps.
To make these ideas precise, consider the dynamical system formed by iterating an entire or meromorphic transcendental map of the complex plane Because of the essential singularity at infinity, the global degree of f is infinite and the dynamics is richer than for rational maps. The exponential and the tangent maps are in this class. Other examples are generated by applying Newton's method to an entire function, which almost always yields a transcendental meromorphic map as the root finder function to be iterated.
The dynamical plane of f splits into two disjoint completely invariant subsets: the Fatou set, or points for which the family of iterates is well defined and form a normal family in some neighborhood; and its complement, the Julia set. The Fatou set is open and contains, among other types of components, all basins of attraction of attracting periodic orbits. By contrast, the Julia set is closed in the sphere and it is either the whole sphere or has no interior. It can be characterized as the closure of the repelling periodic points or, if the map is meromorphic, as the closure of the set of poles and prepoles. For general background on meromorphic dynamics we refer to the survey [5] and the references there.
A key role in the dynamics of a transcendental map f is played by points in the set S( f ) of singular values of f ; these are either critical values (images of zeroes of f , the critical points), or asymptotic values (points v = lim t→∞ f (ω(t)) where ω(t) → ∞ as t → ∞), or accumulations thereof. Indeed every periodic connected component of the Fatou set is, in some sense, associated to the orbit of a point in S( f ). For example, basins of attraction of an attracting or parabolic cycle must actually contain a singular value, but weaker conditions associating a singular value to Siegel disks or Baker domains [6,10,45] and even to wandering domains [11] also exist. Note that S( f ) coincides with the set of singularities of the inverse map.
Transcendental entire or meromorphic functions with a finite number of singular values have finite dimensional parameter spaces [33,35]. These are known as finite type maps and denoted by S := { f : C → C ∪ {∞} entire or meromorphic | #S( f ) < ∞}.
Maps in the class S have several special properties such as the absence of wandering domains [16,33,35,53] or the existence of at most a finite number of non-repelling cycles.
Entire transcendental functions, whose dynamics have attracted a great deal of attention in the last few years, may be thought of as special cases of transcendental meromorphic maps in a wider sense, namely those with all their poles at infinity. The simplest example is the exponential family E λ (z) = e z + λ. By Iversen's theorem [40] entire functions (and meromorphic functions with finitely many poles), all have infinity as an asymptotic value, which may be taken to mean that infinity is a multiple pole. By contrast, meromorphic maps for which infinity is not an asymptotic value are at the other end of the spectrum: they necessarily have infinitely many poles and this difference in behavior at infinity has consequences for the dynamical and parameter spaces (see e.g., [14]). The tangent family of maps T λ (z) = λ tan(z) [30,37,41] is the simplest example of these. This class, which we denote as M ∞ := { f ∈ S | ∞ is not an asymptotic value}, in some sense captures the essence of meromorphic maps since the condition that infinity is not an asymptotic value persists under small perturbations of f in S.
To center the discussion, we consider dynamically natural families (or slices) of functions in S which, roughly speaking, consist of one-dimensional slices of larger holomorphic families of entire or meromorphic maps, for which only one singular value is allowed to bifurcate at any given parameter value (see Definition 4.5 for details). We concentrate our attention on the behavior of a marked asymptotic value called the free asymptotic value, and in particular, on the connected components of parameters for which this marked point is attracted to an attracting cycle whose basin contains no other singular value. We call these sets shell components and they are the transcendental version of the hyperbolic components one finds in parameter planes of rational maps. Note, however, that because we allow parabolic cycles to exist and attract a non-free singular value, the maps here need not be hyperbolic. While shell components occur in the parameter planes for many families of entire functions, they are better understood in the broader context which includes meromorphic functions.
The first part of the paper deals with the properties of attracting basins in the dynamical plane of maps belonging to a shell component. Our main results about the dynamical plane are proved in Sect. 3, and are summarized (in a slightly weaker form) in the following statement.
Theorem A Let f be an entire or meromorphic function. Suppose that a 0 , . . . , a p−1 is an attracting cycle of period p ≥ 1 whose basin of attraction A contains exactly one asymptotic value v and no critical points. For i = 0, . . . , p − 1 let A i be the component of A containing a i and labeled so that f Note that the only assumption on f in Theorem A is that it is transcendental. In particular, it may have infinitely many singular values.
The rest of the paper concentrates on properties of parameter spaces. A dynamically natural slice is represented by a one-dimensional manifold, ⊂ X , isomorphic to the complex plane with perhaps a discrete set of punctures called parameter singularities, where, for example, the function may reduce to a constant. For simplicity we require the parametrization to trace the position of the marked asymptotic value in an affine fashion. Most of the best known one-parameter slices such as quadratic rational maps with a super attracting cycle (quadratic polynomials), cubic polynomials with a super attracting fixed point, exponential or tangent functions, and so on, are dynamically natural slices of maps in different classes where the parameter is an affine function of a singular value (see Sect. 7).
If is a dynamically natural slice for the family F of entire or meromorphic maps in S, we call the restriction of the family to a "dynamically natural family" and denote it by { f λ } λ∈ . The slice contains different types of "distinguished" parameter values which are solutions of algebraic or transcendental equations. Slightly abusing the usual terminology we call those parameter values for which the free singular value lands on a repelling cycle Misiurewicz parameters. Centers, parameters for which a critical point is periodic, are also distinguished parameters. Meromorphic transcendental maps have an additional type of distinguished parameter which does not occur for rational maps. We call a parameter λ a virtual cycle parameter if some iterate of the asymptotic value is a pole for f λ . This name is motivated by the existence of a virtual cycle, morally a "cycle" that contains the point at infinity and the asymptotic value (see Definition 6.7). Virtual cycle parameters are important in our discussion and they are generally quite abundant in the following sense (see Proposition 5.6).
Proposition B Let { f λ } λ∈ be a dynamically natural family of entire or meromorphic maps in S, and let v = v(λ) be the free asymptotic value. Both the virtual cycle parameters, when they exist, and the Misiurewicz parameters are dense in the set is not a normal family in any neighborhood of λ 0 }.

Moreover B(v) has no isolated points in .
The set B(v) is known as the activity locus of the free asymptotic value and is part of the bifurcation locus B, or the set of parameters for which at least one of the singular values bifurcates. It follows from our definition of a dynamically natural slice, that the boundary of any shell component is a subset of B(v).
General results of this nature have been shown for parameters spaces of rational maps using bifurcation currents (see for example [7,25,31,34]).
By definition, functions in a shell component of a dynamically natural slice have no critical points in the basin of the attracting cycle, so the multiplier of the cycle is never zero, and therefore the component has no center. We see, however, that some virtual cycle parameters play the role of centers in our setting (see Theorem 6.10). Definition 1.1 (Virtual center) Let be a shell component of a dynamically natural family { f λ } λ∈ of entire or meromorphic maps. We say that λ * ∈ (∂ ∩ ) ∪ {∞} is a virtual center of , if there exists a sequence of λ n ∈ such that λ n → λ * and the multiplier ρ (λ n ) of the attracting cycle for λ n tends to 0.
Here and throughout the paper, when we say a point is finite we mean that it is not the point at infinity. Although maps in a shell component might be non-hyperbolic, a shell component shares many properties with hyperbolic components of known families like the exponential or tangent families, [41,50]. Indeed, shell components can be parametrized by the multiplier map ρ = ρ : −→ D * Fig. 1 Sketch of two shell components and their internal rays. On the left (resp. right) the virtual center is at infinity (resp. at a finite point) and therefore the component is unbounded (resp. bounded) where D * := D \ {0}, and ρ(λ) is the the multiplier of the attracting cycle to which v(λ) is attracted. Since the multiplier is never zero, ρ is a covering (see Theorem 6.4) and hence shell components are either simply connected and ρ has infinite degree or else they are homeomorphic to D * , where the puncture is a parameter singularity (see Theorem 2.2 and Corollary 6.5).
Our last result describes the internal structure and the boundary of a shell component and establishes the existence and uniqueness of the virtual center (see Theorem 6.13, and corollaries 6.14, 6.14 and 6.18 ).
Then, all internal rays have one end at the virtual center and the other end at a point in ∂ (which a priori could be infinity). If the virtual center λ * is finite no internal ray has both ends at λ * (Fig. 1).
When the shell component is not simply connected then the puncture, which is a parameter singularity, plays the role of the virtual center. An example of a doubly connected shell component can be seen in Example 1(d) in Sect. 7.
Numerical experiments show that for meromorphic functions in M ∞ shell components of period larger than one are always bounded, that is, they have their virtual center at a finite parameter value, so that it seems reasonable to conjecture that this is so in general. At present, we can only prove this fact for particular families like the tangent family. (See [23]).
The paper is structured as follows. In Sect. 2 we recall the definition of singular values of transcendental functions and we state some standard theorems on the covering and connectivity properties of their Fatou components. In Sect. 3 we discuss the properties of the dynamical planes of these functions, proving Theorem A. In Sect. 4 we define the dynamically natural slices of parameter spaces that are the main subject of the paper and in Sect. 7 we give a number of examples. In Sect. 5 we define various types of distinguished parameter values that lie in the bifurcation locus and prove Proposition B. Section 6 is the heart of the paper where, after some preliminary results, we prove Theorems C and D. We conclude with an Appendix where we discuss a theorem of Nevanlinna that allows us to characterize certain functions of finite type. We also extend this theorem to a somewhat larger class of functions.

Singularities of the Inverse Function
Let f be a transcendental entire or meromorphic map. A point v ∈ C is a singular value of f if some branch of f −1 fails to be well defined in every small enough neighborhood of v. Singular values may be critical, asymptotic or accumulations thereof. If c is a critical point, that is, a zero of f then its image v = f (c) is a critical value. If there is a path ω(t) such that lim t→1 ω(t) = ∞ and lim t→1 f (ω(t)) = v then the limit v is an asymptotic value of f . Observe that the point at infinity is always an essential singularity, but it may be or not be a singular value. It is a critical value if and only if it is the image of a multiple pole of f and it is an asymptotic value if some unbounded curve has an image that tends to infinity. For example, infinity is not a singular value for the tangent map.
Singular values are classified in terms of the branches of f −1 as follows (see [3,39]). If an asymptotic value v is isolated, the radius r in the above proposition can be chosen small enough so that f : U r → B(v, r ) \ {v} is a universal covering map. In this case U r is called an asymptotic tract for the asymptotic value v and v is called a logarithmic singular value or a logarithmic singularity. The number of distinct asymptotic tracts of a given asymptotic value is called its multiplicity. Note that this notion is not the same as the multiplicity of a critical value, that is, the sum of the multiplicities (local degree minus one) of f over all its preimages. Using this term in these two ways should not cause confusion.
We denote the set of singular values of f by S( f ) and the post-singular set by By Iversen's Theorem [40] an entire transcendental function f always has an asymptotic value at infinity.

Mapping Properties
The following lemma is well known in algebraic topology. See for example [43] for the general theory of coverings or [56, Thm. 6.1.1.] for a proof of the lemma below.
If f is unbranched and U is simply connected we say that f is a universal covering.

Dynamical Plane: Proof of Theorem A
In this section we describe some properties of the dynamical plane for general entire or meromorphic maps. In particular, we do not assume our maps to have a finite number of singular values. Theorem A in the introduction follows from the three propositions in this section, which are slightly stronger than the theorem itself. We are interested in basins of attraction of attracting cycles containing a unique asymptotic value v. Since v is the only singular value in the basin, it is isolated and is thus a logarithmic singularity (see Sect. 2.1). It follows that if V * is a punctured neighborhood of v, there exists at least one asymptotic tract among the components of f −1 (V * ).
We shall prove that under this weak assumption, a great deal can be said about the attracting basin. We begin with connectivity properties. Proof For k = 0, . . . , p − 1, let A k denote the component of A which contains a k where labels are defined so that f (A k ) = A k+1 (mod p) . Let U 0 ⊂ A 0 be a topological open disk with a smooth boundary containing a 0 in its interior and such that f p is one to one in U 0 and f p (U 0 ) ⊂ U 0 . This set exists by Koenig's Linearization Theorem (see e.g., [45]). Consider the successive preimages of U 0 which follow the cycle in reverse order; that is for every k ≥ 1, recursively define the set U k ⊂ A ( p−k) mod p as the component of f −1 (U k−1 ) that contains the point a ( p−k) mod p . This defines a nested sequence of open sets around each of the points in the cycle. That is, for every 0 ≤ k ≤ p − 1 we have U k ⊂ U k+ p ⊂ U k+2 p · · · and the infinite union of these nested sets equals A k . By definition, since every component A k is the infinite union of a collection of nested open sets, the components of the immediate basin are simply connected if and only if U k is simply connected for every k.
We will use induction to show every U k is simply connected. For k = 0, this is true by construction. We assume now that U k is simply connected and prove that the same is true for U k+1 .
First suppose that v / ∈ U k . Then U k has no singular value and so U k+1 has no critical point and thus f : U k+1 → U k is a covering map. Since U k is conformally equivalent to a disk by hypothesis, it follows from Lemma 2.2 that U k+1 is simply connected and f is conformal on this set. Now suppose that v ∈ U k . Then U k+1 either contains an asymptotic tract of v and f | U k+1 has infinite degree, or it does not. Suppose it does. By hypothesis f : U k+1 \{ f −1 (v)} → U k \{v} is a covering map. Since U k \{v} is conformally equivalent to a punctured disk, it follows from Lemma 2.2 that f −1 (v) ∩ U k+1 contains at most one point. But if it contained a point, f | U k+1 would have finite degree, and that is not the case. Hence f −1 (v) ∩ U k+1 = ∅ and U k+1 is simply connected and we are done. Now suppose U k+1 does not contain an asymptotic tract. In this case, f : is still a covering, but now of finite degree, and thus f −1 (v) is a single point, a finite preimage v of v. Note that this degree must be one because there are no critical values in the basin. Eventually, however, U k+qp+1 must contain an asymptotic tract of v for some q > 0 because the immediate basin must contain a singular value and v is the only such by hypothesis. Since the sets are nested, U k+qp+1 would then contain not only v but also the asymptotic tract of v so that f : U k+1+qp → U k+qp has infinite degree and would contradict part (ii) of Lemma 2.2.
We conclude that every component of the immediate basin is simply connected and v has no finite preimages in A.
If furthermore A contains no critical points, the same arguments applied to preimages of every A k for every k, show that at every component A is simply connected.
contains a simply connected unbounded set T in A 0 (an asymptotic tract) whose boundary is a open curve which tends to infinity in both directions, showing that infinity is accessible from A 1 . Moreover, f : T → B has infinite degree, hence f : A 0 → A 1 has infinite degree. This shows (a).
Since v ∈ A 1 is the only singular value in the basin, f : A j → A j+1 (mod p) is a covering map for all j = 0. Consequently, since all components of the basin are simply connected, these maps are all univalent.
Let γ : [0, 1) → A 0 be any curve tending to infinity as t → 1, and let σ : [0, 1) → A p−1 be a path satisfying γ = f • σ . By continuity, any accumulation point of σ (t) when t → 1 must be infinity or a pole. Since the set of poles is discrete, it follows that σ must actually land at infinity or at a pole of f . If σ were to land on infinity, infinity would be an asymptotic value. Therefore, if we assume that ∞ is not an asymptotic value of f , it follows that it lands at a pole in ∂ A p−1 ; it is accessible from A p−1 (i.e., the limit of an arc σ (t) ⊂ A p−1 ) by construction. This ends the proof of (b).
The strategy for proving (c) relies on showing that if A 0 contains two asymptotic tracts of v, it must also contain a critical point of f , contradicting the hypotheses. Although this is probably well known, we did not find a reference for it and therefore we include a proof. We need to set up some notation. Let δ : (Recall that most boundary points are accessible [48,Sect. 6.4]). Since f : A 0 → A 1 \ {v} is a covering map, the set of preimages of δ in A 0 consists of countably many disjoint simple curves No δ i can have both ends at infinity, because otherwise w would be an asymptotic value in ∂ A 1 which contradicts the hypotheses. The curves δ i divide A 0 into infinitely many fundamental domains D i , each of which maps conformally onto A 1 \ δ. Hence the inverse branches ψ i : A 1 \ δ → D i are well defined and univalent.
Let B ∈ A 1 be a simply connected neighborhood of v such that δ crosses ∂ B exactly once. Suppose that f −1 (B) ∩ A 0 has two (and for simplicity, only two) different components T and T ; that is, there exist two asymptotic tracts of v in A 0 . Our goal is to show that this cannot happen under the assumption that there no singular values in ∂ A 1 . To that end, start by observing that T and T are both unbounded simply connected sets whose boundary is an open curve that tends to infinity in both directions. Together, T and T form the full preimage of B in A 0 . Hence the preimages of δ ∩ B are unbounded subsets of the curves δ i which map to B, and thus belong either to T or to T . Now let {δ T j } j∈Z and {δ T k } k∈Z denote the two bi-infinite subsets of {δ i } i∈Z that tend to infinity within T or T , respectively, and indexed in such a way that δ T j is in between δ T j−1 and δ T j+1 (respectively for T ). Also relabel the domains D i and the points w i accordingly, so that D T j is bounded by δ T j and δ T j+1 , and w T j ∈ ∂ A 0 is the finite endpoint of δ T j (respectively for T ). See Fig. 2. Now consider a point b ∈ ∂ B, b / ∈ δ, and let b T and b T be two arbitrary preimages (out of the infinitely many) in ∂ T and ∂ T , respectively. Let γ be a simple curve joining b T and b T through the open set A 0 \ (T ∪ T ). Then γ divides this set into two unbounded components, say C 1 and C 2 , whose union must contain the whole boundary of A 0 . In particular, the points w T i and w T i , each belong either to C 1 or to C 2 .
Our first claim is that not all these points can belong to the same component C 1 or C 2 . Indeed if they did, infinitely many curves δ T j and δ T j would intersect γ . This would in turn imply that f (γ ) ∈ A 1 \ B crosses δ infinitely many times and hence runs around v infinitely many times and has infinite length. But this cannot happen because f (γ ) is a finite length closed curve with initial and endpoint b ∈ ∂ B. We conclude that γ crosses only finitely many curves δ i and thus there are infinitely many points w T j and w T k in each component C 1 or C 2 , as claimed. Choose one of the components, say C 1 , and let J , K ∈ Z be the "last" indices for which w T J and w T K belong to C 2 or, in other words, w T J , w T K ∈ C 2 but w T j , w T k ∈ C 1 for all j > J and k < K . This in turn implies that the curves δ T j and δ T k do not intersect γ for any j > J or k < K ; note that if they do, because both their endpoints lie on C 1 , γ can be modified slightly to avoid such intersections.
Finally observe that the domains D T J and D T K −1 must then be part of the same fundamental domain, which is "double" in some sense. Indeed, one could join the This would imply that f (γ ) ⊂ A 1 \ δ and therefore γ = ψ i ( f (γ )) for some fixed i ∈ Z. But γ contains at least two preimages of the point b which contradicts the fact that the inverse branch ψ is univalent, and ends the proof of part (c).
This concludes the proof of Theorem A.

Boundaries of Basins of Attraction
In the next proposition, we make the additional restrictions that ∂A is locally connected and contains no singular values to obtain properties of the component of the basin that maps to a neighborhood of the asymptotic value. Under the additional assumption that f ∈ M ∞ , we get a complete understanding of the topology and dynamics of the whole basin for these maps.

Proposition 3.3 (Bounded components) Let f be an entire or meromorphic function with an attracting cycle of period p ≥ 1. Let A be its immediate attracting basin and suppose it contains a unique asymptotic value v and no critical points. Assume ∂A is locally connected and S( f
Proof Recall from the propositions above that all components of A are simply connected, A 0 is unbounded and maps infinite to one to A 1 , and f : The assumption of local connectivity of A, implies that A i is bounded if and only if ∂ A i is a single closed curve. Otherwise ∂ A i is a countable union of continuous unbounded curves.
We first show that ∂ A 0 is a single continuous curve with both ends at infinity. Indeed, if ∂ A 0 had at least two components, an argument analogous to the proof of Proposition 3.2 would show that at least one of the fundamental domains D j would be bi-infinite and thus map 2 to 1 onto A 1 , contradicting the injectivity of f on D j .
Thus ∂ A 0 is connected and since A 0 is simply connected, A 0 has only one access to infinity; that is, if ) must converge to a pole, say P ∈ ∂ A p−1 or otherwise ∞ would be an asymptotic value. Let N be the neighborhood of P which maps onto C \ B(0, R) for some large R. Then N ∩ A p−1 contains the full preimage of C. Now suppose there is a second pole P ∈ ∂ A p−1 and choose a curve σ (t) → P in A p−1 . Then f (σ (t)) must tend to infinity in C since there is no other access to infinity. Let N be a preimage of C \ B(0, R) containing P . Choosing R larger if necessary, N and N are disjoint. But N ∩ A p−1 must contain preimages of all points in C, which implies f : A p−1 → A 0 is not one to one. Thus there is no second pole on ∂ A p−1 . Now suppose that A p−1 is unbounded and let γ (t) → ∞ within A p−1 . Since we can choose N so that γ / ∈ N , f (γ ) cannot enter C and so must converge to a finite boundary point which is therefore an asymptotic value, contradicting the hypothesis that f ∈ M ∞ . This proves that A p−1 is bounded. By the same argument, since ∂A contains no asymptotic values, it follows that each component A i , i = 1, . . . , p − 2 is bounded by a continuous closed curve without poles.

Remark 3.4
In this proposition we assumed local connectivity of the boundary of the immediate basin and concluded that all its components except A 0 are bounded. It is plausible that a partial converse statement is also true. If we assume that A 1 is bounded, and add some hyperbolicity condition, it is very possible that ∂ A 1 is locally connected. Then the same would be true for the remaining components.

Parameter Space: Dynamically Natural Slices
Families of functions defined by explicit formulas such as rational functions, tangent or exponential functions clearly have "natural" embeddings into C n for an appropriate n in terms of their coefficients or singular values. Up to affine conjugation, the images of these embeddings can be thought of as parameter spaces and we can study how the dynamics depends on these parameters. We do not necessarily have closed forms for the families we are discussing here. Nevertheless, there is a sense in which every transcendental function f with N < ∞ singular values belongs to an N dimensional complex analytic manifold [16,33,36].
To make this precise we use the theory of holomorphic motions for holomorphic families of entire and meromorphic maps. For rational maps, the theory is explained in [44] and it is adapted to meromophic functions in [41]. We state the results from the latter reference that we use here.

Definition 4.1 (Holomorphic family)
A holomorphic family of entire or meromorphic maps over a complex manifold X is a map F : is meromorphic for all x ∈ X and x → f x (z) is holomorphic for all z ∈ C. If X has dimension n, we say the family has dimension n and X is the parameter space for F.
A holomorphic motion of a set V respects the dynamics of the holomorphic family Because maps in F that are conjugate by an affine map have the same dynamics, we want to restrict ourselves to one representative for each conjugacy class. We do this by choosing an appropriate normalization for the functions and restricting F to those normalized functions. Below, we always assume the family is normalized in some way; the normalization depends on the family.
The following equivalencies are proved for rational maps in [44] and extended to the transcendental setting in [41].

Definition 4.4 (J −stability)
A parameter x 0 ∈ X is a J -stable parameter for the normalized family F if it satisfies any of the above conditions. We denote by X stab , the set of J -stable parameters for the family F. Its complement is known as the bifurcation locus of the family F, and the elements in B X are the bifurcation parameters.
In families of maps with more than one singular value, it makes sense to consider subsets of the bifurcation locus where only some of the singular values are bifurcating, in the sense that the families {g i n (x) := f n x (s i (x))} n are normal in a neighborhood of x 0 for some values of i, but not for all. We define This is also known as the activity locus (in X ) of the singular value s i (see [25,34]). In this paper we investigate one-dimensional slices of holomorphic families in which the activity loci of the different singular values are disjoint. This means that at any given parameter on our slice only one of the singular values is allowed to bifurcate at a given parameter value. More precisely, if the maps in our slice have N distinct singular values, we require the existence of N − 1 persistent attracting or parabolic cycles of fixed multiplier throughout the slice. Singular values can take turns in "serving" these basins but only one of them at a time is allowed to be free. These and other conditions are collected in the definition below of a dynamically natural slice.
Definition 4.5 (Dynamically natural slice) Let F be a holomorphic family of normalized entire or meromorphic maps over X . Assume S( f x ), the set of finite singular values of f x has cardinality N < ∞ for all x ∈ X . A one-dimensional subset ⊂ X is a dynamically natural slice with respect to F if the following conditions are satisfied.
is embedded or conformally equivalent to C minus a discrete set. The removed points are called parameter singularities. (By abuse of notation we denote its image in C by again, and denote the variable in C by λ, and the function f x by The singular values are given by distinct holomorphic functions s i (λ), i = 1, . . . , N −1, and an asymptotic value v λ : that is an affine function of λ. We further assume that no preimage of v λ is a critical point for all λ. We call v λ the free asymptotic value, and we require that There are N −1 distinct attracting or parabolic cycles whose period and multiplier are constant for all λ ∈ . (e) Suppose there exists a parameter λ 0 such that v λ 0 is the only singular value in A λ 0 , the basin of attraction of an attracting cycle whose multiplier is not a constant function of λ. Then the slice contains, up to affine conjugacy, all meromorphic maps g : C → C that are quasiconformally conjugate to f λ 0 in C and conformally conjugate to then does not satisfy at least one of the conditions above.
From now on, as long as it is understood from the context, we will drop the index . In other words B := B . Definition 4.6 (Dynamically natural family) We define the subfamily of the holomorphic family F of normalized entire or meromorphic functions, F = { f λ } λ∈ , as those f λ ∈ F for which λ lies in the dynamically natural slice . We call F (or { f λ } λ∈ ) a dynamically natural family for the slice . When referring to the parameters we use and when referring to the functions we use F .
Let us make some remarks on the conditions above.
Remark 4.7 (i) A given family F may contain both entire and meromorphic functions. Condition (c) implies that in a dynamically natural slice the functions are either all entire or all meromorphic. (ii) Condition (d) could actually be weakened for our purposes. As an example, we could require only N − 2 attracting or parabolic cycles of constant multiplier plus an orbit relation between two distinct singular values (e.g., symmetry, or the two orbits coinciding after some iterates.) The tangent family is an example. Although most of our results hold in these more general situations, for simplicity of exposition we choose to require condition (d) as stated. We will, however, also consider more general dynamically natural slices in the examples in Sect. 7. (iii) Condition (e) is generally easy to verify in concrete families. (iv) Condition (f) is imposed to avoid artificial parameter singularities. In general, these singularities occur because the functions become constant or some of the singular values coalesce and/or some poles coalesce and become critical points; either of these events will create parameter singularities.
(v) Suppose is a dynamically natural slice. For a given parameter value λ 0 ∈ , if the free asymptotic value v λ 0 is attracted to an attracting cycle, it follows that λ 0 ∈ stab (indeed, if any of the cycles is parabolic, it must be persistent and therefore the map is J −stable, although not necessarily hyperbolic). Because there are N − 1 attracting or parabolic basins which need to attract N − 1 different singular values, only one of the singular values can be active at any given parameter. That is, if we set B 0 := B(v λ ), then for i, j = 0, 1, . . . , N − 1, If we relax the definition of a dynamically natural slice as indicated in Remark (ii) above, then we must also allow the possibility that B i = B j for some i = j. This is the case, for example, in the tangent family.
Most of the slices which have been systematically studied in the literature, like the exponential family e z + λ or the tangent family λ tan z, are dynamically natural slices of the larger family of functions with two asymptotic values and no critical values. In the Appendix we show that many other slices can be constructed by pre-or postcomposing functions with finitely many asymptotic values with rational maps, and some examples are shown in Sect. 7.

Distinguished Parameter Values: Proof of Proposition B
Let be a dynamically natural slice for a family of entire or meromorphic functions and let F = { f λ } λ∈ be the corresponding dynamically natural family. Recall that the definition implies that #S( f λ ) is constant and finite for all λ ∈ . Let v λ denote the free asymptotic value of f λ . Our goal is to discuss special parameter values for which the forward orbit of the free singular value is finite. We restrict our discussion to parameters varying in and to the bifurcation locus B(v λ ), although the definitions could be adapted for the full holomorphic family defined over X .

Definition 5.1 (Misiurewicz parameters)
A parameter λ ∈ (or the map f λ ) is called a slice Misiurewicz parameter if the free singular value v λ has the property that f n λ (v λ ) is a repelling or parabolic periodic point for some n ≥ 0. Remark In the definition above we assumed the periodic point f n λ (v λ ) to be repelling or parabolic, because we want our Misiurewicz parameters belong to B(v λ ). Indeed, otherwise f n λ (v λ ) would be attracting, Siegel or Cremer. In the first case the parameter λ would be J −stable while the last two cases cannot occur since by definition of , they would imply the presence of N non-repelling cycles, and therefore N singular values apart from v λ .
For simplicity of exposition, throughout this paper we will refer to slice Misiurewicz parameters as Misiurewicz parameters.
By definition, Misiurewicz parameters are solutions of for some m > n ∈ N. If f λ is conjugate to a Misiurewicz map f λ which satisfies Equation (5.1) for certain values of m, n ∈ N, then f λ satisfies the same equation for the same m, n.
The second kind of distinguished parameter value that we discuss is specific to meromorphic maps. A meromorphic function has at least one pole that is not an omitted value. The preimages of the poles have a finite forward orbit and play an important role in the dynamics.
With this definition a pole is a prepole of order 1.
For every λ ∈ , the poles form a discrete set so they can only accumulate at infinity. Unless there are at most two poles and both are omitted values, which is never the case for functions of finite type, Picard's theorem implies that the prepoles of order 2 form an infinite set. Their accumulation set is the set of poles and the point at infinity. Prepoles of order n ≥ 3 accumulate both at prepoles of order n − 1 and at infinity. It is well known that prepoles are dense in the Julia set of f λ [5,15].
We now look at the parameter plane and consider parameters for which some iterate of the free asymptotic value is a pole.
In other words a virtual cycle is the forward orbit of an asymptotic value which is a prepole. For λ ∈ , let {P i (λ)} 0≤i≤M≤∞ denote the set of poles of f λ . Recall that since we are working in a dynamically natural slice, each P i (λ) is a distinct holomorphic function of λ throughout . By definition, virtual cycle parameters of order p ≥ 2 in are λ−values that are solutions of for some 0 ≤ i ≤ M, where as usual v(λ) denotes the free asymptotic value of f λ .
Remark Virtual cycle parameters are parameter values for which an orbit relation exists in the sense of (5.2). Hence if λ is a virtual cycle parameter and f λ and f λ are topologically conjugate, then λ must be virtual cycle parameters of the same order. These exist since the set of prepoles of order 3 or larger is infinite. Now observe that because of the assumption that λ 0 ∈ B(v λ ), {g n } is not normal in U , and so there must be some λ ∈ U such that g n (λ) = f n λ (v λ ) takes one of the values P i (λ), for some i = 1, 2 or 3 and some n.
To show that Misiurewicz parameters are dense we use a repelling periodic point a(λ) of period 3 (for example) for all λ ∈ U , and apply the above arguments to the functions {a(λ), f λ (a(λ)), f 2 λ (a(λ))}, and to g n (λ) := f n λ (v λ ). Note that these proofs actually show that the sets of virtual cycle parameters of order 4 or Misiurewicz parameters of preperiod 1 and period 3 are dense in B(v λ ). By choosing prepoles of a different order or periodic points of a different period, the same proof would show that fixing these numbers arbitrarily, the corresponding subsets are also dense in B(v λ ). Since these subsets are disjoint, it follows that they must accumulate on each other and therefore B(v λ ) has no isolated points.

Shell Components: Proof of Theorems C and D
Let be a dynamically natural slice of a normalized family of entire or meromorphic maps and let F = { f λ } λ∈ be the associated dynamically natural family. Let v λ denote the free asymptotic value of f λ which, we recall, depends affinely on λ.
We are interested in studying those stable components in the slice that reflect the behavior of the free asymptotic value. Observe that if λ ∈ H, then all remaining singular values must belong to the N − 1 basins of parabolic or attracting cycles with constant multiplier, so that v λ is the only singular value in the basin of attraction of the cycle with non-constant multiplier. As a consequence, parameters in H are J −stable in ; that is where as above, B = B is the bifurcation locus of the slice .
Since ∂ is in B, it follows that is also a connected component of \ B. Shell components may be hyperbolic components in the usual sense -or they may not be; for example, if some of the cycles with constant multiplier are parabolic then no map in is hyperbolic. Conversely, capture components, where one of the persistent cycles attracts v λ and another singular value, are hyperbolic components but are not shell components.
The following property follows directly from the definition.

Lemma 6.2 (The period is constant) Let F be a dynamically natural family and let ⊂ be a shell component. Then the period of the attracting cycle to which v λ is attracted is constant throughout . It is called the period of .
Since the dynamics of all but one of the attracting cycles of f λ are fixed as λ varies in a shell component, the attracting cycle for this shell component is the one whose multipler varies with λ. Below, when we talk about an attracting cycle or its multiplier, in relation to a shell component, it is these that we mean.
Every hyperbolic component of the parameter plane of the well known exponential family E λ = z → e z + λ is an example of a shell component. In this case, the multiplier of the attracting cycle of E λ is never zero. This is a general phenomenon for shell components.

Lemma 6.3 (Nonzero multipliers) Let F be a dynamically natural family and let ⊂ be a shell component. Then the multiplier of the attracting cycle is nonzero for all λ ∈ .
Proof If the multiplier of the attracting cycle were zero for λ * ∈ , the cycle would contain a critical point. Therefore, since v λ is not persistently critical, in a neighborhood of λ * there would be two singular values in the basin. This is impossible because we are in a shell component.
The following theorem explains how shell components are parameterized by the multiplier of the attracting cycle. Let be a shell component of period k > 0. We define the multiplier map as where ρ(λ) is the multiplier of the attracting cycle which attracts v λ . Proof The proof uses quasiconformal surgery on the basin of attraction of the attracting cycle that attracts v λ . For details see [9]. Note that because f λ is a holomorphic family both ρ and its derivative are holomorphic functions of λ.
Choose a parameter λ 0 ∈ such that ω 0 := ρ(λ 0 ) ∈ D * . Let k > 0 be the period of and a 0 = {a 0 0 , · · · , a 0 k−1 } the attracting periodic orbit to which the free asymptotic value v λ 0 is attracted. Let A 0 denote its full basin of attraction and let A 0 i be the connected component of A 0 containing a 0 i for i = 0, · · · , k − 1. Assume without loss of generality that v λ 0 ∈ A 0 1 . Let W ⊂ D * be a (simply connected) neighborhood of ω 0 . For any ω ∈ W we define a Beltrami form μ ω in A 0 1 invariant under f k λ 0 as follows. Since a 0 is attracting, there is a biholomorphic map ψ from a neighborhood U of a 0 1 to the unit disk D such that ψ(a 0 1 ) = 0 and ψ( f k λ 0 (z)) = ω 0 ψ(z) for every z ∈ U . In D choose an annular fundamental domain R 0 = {|ω 0 | < |z| < 1} for the map ψ(z) → ω 0 ψ(z). Now define the surgery using standard arguments: Construct a quasiconformal mapφ ω : D → D such that it fixes the origin and the unit circle and it maps R 0 to R ω = {|ω| < |w| < 1}, conjugating multiplication by ω 0 to multiplication by ω. In addition, it should satisfỹ φ ω 0 = I d andφ ω should depend holomorphically on ω. To do this, first define the map on R 0 and then extend it by the dynamics to all of D. The Beltrami coefficientμ ω ofφ ω lifts to a Beltrami coefficient μ ω on U invariant under the map ψ that satisfies = μ ω (z). Extend μ ω to the full backward orbit of U so that for any inverse branch g n of f −n λ 0 defined on a component of A 0 , μ ω (g n (z)) g n (z) g n (z) = μ ω (z). Extend it to be identically zero on the complement of A 0 in C. A Beltrami coefficient with this property is said to be compatible with f λ 0 . Notice that the Beltrami coefficient μ ω depends holomorphically on ω.
By the Measurable Riemann Mapping Theorem [1], there is a quasiconformal homeomorphism φ ω : C → C with Beltrami coefficient μ ω . It is unique up to postcomposition by an affine map and depends holomorphically on ω. Moreover, φ ω 0 is conformal.
At this point we recall that we fixed the affine conjugacy class by normalizing all the functions in the family F in the same way. Different normalizations work and/or are more convenient for different slices-for example, in some, but not all slices, one could ask that f λ fix two given points in C, or ask that it fix one point and also fix its multiplier.
We can chose a uniform normalization for φ ω that preserves the normalization for F and define the new entire or meromorphic map This map respects the dynamics: it has an attracting periodic cycle a(ω) = φ ω (a 0 ) of multiplier ω. Moreover, F ω is quasiconformally conjugate to f λ 0 in the respective basins of attraction and conformally conjugate to f λ 0 everywhere else. Since φ ω 0 is conformal and fixes two points it follows that it is the identity map. Hence F ω 0 = f λ 0 . Since f λ is in a dynamically natural slice, property (d) ensures that F ω = f λ(ω) for some λ(ω) ∈ . Observe that the asymptotic value of f λ(ω) is v ω = φ ω (v λ 0 ) and it depends holomorphically on ω. Since v λ = v(λ) is affine, λ = λ(v) is also affine and hence λ(ω) = λ(v ω ) is holomorphic. Since the multiplier changes with ω, the map is non-constant and thus open. Finally, λ(ω 0 ) = λ 0 and therefore λ(W ) ⊂ . This construction defines a holomorphic local inverse of the multiplier map ρ in a neighborhood of any point ω 0 = ρ(λ 0 ) ∈ D * . It follows that ρ is a covering map from to D * .

Corollary 6.5 (Connectivity of shell components) Let be a shell component in a dynamically natural slice . Then one of the two following possibilities occurs:
(a) is simply connected and ρ : → D * is a universal covering (hence of infinite degree), or (b) is conformally equivalent to D * and lim |ω|→0 ρ −1 (ω) = λ * where λ * / ∈ is a parameter singularity.
Proof Since ρ : → D * is a covering, it follows from Lemma 2.2 that either ρ is a universal covering and hence has infinite degree and is simply connected, or ρ has finite degree and is conformally equivalent to D * . In this case, the point λ * := lim |ω|→0 ρ −1 (ω) cannot belong to B(v λ ) because, by Proposition 5.6, B(v λ ) has no isolated points. It follows from Lemma 6.3 that ρ cannot extend continuously to λ * because f λ has no superattracting cycles. Therefore λ * is a parameter singularity.

Virtual Centers of Shell Components
In standard rational dynamics, hyperbolic components have a distinguished point called the center of the component at which the attracting cycle is superattracting. This is not the case for the shell components we are studying, where the multiplier of a finite cycle cannot vanish. Nevertheless we shall see that there is a unique point in the boundary of the shell component that plays the role of the center. Our goal in this section is to describe the relation between virtual cycle parameters and virtual centers. To do this we need a preliminary result which will also be useful later on.

Proposition 6.8 (Bounded or unbounded cycle) Let ⊂
be a shell component of period p ≥ 1 for a dynamically natural family F . Let λ n ∈ be such that λ n → λ * ∈ ∂ ∩ . Let a n = {a n 0 , . . . , a n p−1 } be the attracting cycle of period p for f n := f λ n , attracting the asymptotic value v(λ n ). (a) Suppose the cycle a n stays bounded as n → ∞. Then, f λ * has an indifferent cycle a whose period divides p and a n → a as n → ∞. (b) Suppose that |a n j | → ∞ as n → ∞ for some 0 ≤ j < p. Then, p > 1 and Proof of Proposition 6.8 (a) Since the cycle stays bounded and λ * ∈ , it follows that for all 0 ≤ i ≤ p − 1, there exists a subsequence of the sequence a n i that has a limit a i and that this limit point is a fixed point of f p λ * . This cycle must be non-repelling. Since λ * ∈ ∂ , it cannot be attracting. It also cannot have multiplier 0 because if it did, a * would be superattracting and some point in the limit cycle would be a critical point. But then, for every λ in a neighborhood of λ * , the analytic continuation of this cycle would be attracting and its basin would contain a critical point, and thus λ * would be in the interior of and not on its boundary. Hence a * is an indifferent cycle and part (a) is proved. (b) We will give a proof by contradiction. We will show that if the conclusion in (b) does not hold, then for n large enough the multiplier of the cycle must be larger than one in modulus. Set f * = f λ * , v n = v(λ n ) and v * = v(λ * ).
Let c * be a periodic point of f * which can be analytically continued to c(λ) in a neighborhood of λ * , and choose R 0 so that | f k n (c(λ n ))|, | f k * (c * )| < R 0 for all k ≤ p and all n > 0.
Let us first deal with the case p = 1 and show that the fixed point a n 0 of f n cannot tend to infinity with n. Suppose it does. Let R be large enough to satisfy R > e 8π and also large enough that for n > N , B(0, R) contains v n and all the remaining singular values of f n . This is possible because λ * ∈ so that f * is well defined as a limit of f n , and thus v n → v * and all the other singular values (finite in number) also depend holomorphically on λ. Now choose N > 0 such that for n > N , |a n 0 | > R 2 and recall that since p = 1, f n (a n 0 ) = a n 0 . Then Lemma 6.9 implies that | f n (a n 0 )| > which contradicts the assumption that λ n ∈ n for all n > 0. Next assume that p > 1. Suppose that the conclusion in part (b) is not true so that f k * (v * ) is not a pole for any k < p − 1. Then, for all 0 Since all the singular values of f n and f * , other than the free asymptotic value, vary continuously with λ and are attracted to parabolic or attracting cycles of fixed multiplier, there exists R > max{R 0 , e 8π } large enough, and independent of n, such that for all n, B(0, R) contains the first p − 1 iterates of all these singular values; that is 0≤k< p f k (S( f n )) ⊂ B(0, R), for all n > 0.
Since by hypothesis |a n j | → ∞ as n → ∞, and f p n (a n j ) = a n j , there exists N > 0 such that |a n j | > R 2 for all n > N . Hence the hypotheses of Lemma 6.9 are satisfied for every n > N and we conclude that It follows that the multiplier of the cycle a n is larger than 1 for n > N which contradicts the hypothesis that λ n ∈ for all n > 0.
With these results in hand, we can prove that virtual centers are always virtual cycle parameters. Consequently, if is a slice of entire maps, ∩ has no virtual centers or, in other words, every virtual center is at infinity.
Proof Suppose that λ * ∈ ∩ is a virtual center. Then there exists a sequence of parameters λ n ∈ such that λ n → λ * ∈ ∂ ∩ and the multipliers ρ(λ n ) tend to 0. Let a n = {a n 0 , . . . , a n k−1 } be the cycle to which the free asymptotic value v n = v(λ n ) is attracted by the iterates of f λ n .
We claim that, taking a subsequence if necessary, one of the points in the cycle must tend to infinity as n → ∞. Indeed, otherwise the cycle would converge to a finite cycle a * of f * λ with multiplier 0, which is impossible by Proposition 6.8 (a). Hence one of the points in the cycle tends to infinity. It now follows from Proposition 6.8 (b) that λ * is a virtual cycle parameter.

Remark 6.11
It seems plausible to believe that a partial converse is true. More precisely, if a virtual cycle parameter λ * ∈ ∂ is accessible from , then λ * is a virtual center. The idea for a proof would be to find a new path in ,λ(t) → λ * along which the multiplier would tend to 0. This could be accomplished if a path through a preasymptotic tract of v(t) such that v(t) → v * could be found. See Conjecture 6.17. Propositions 6.8 and 6.10 above have an immediate consequence for components of period one. Corollary 6.12 (Period one shell components) Let F be a dynamically natural family and let ⊂ be a shell component. Let be a shell component of period one. Then, for any sequence λ n → λ * ∈ ∂ ∩ , the attracting fixed point a n converges to an indifferent fixed point a * of f λ * . Moreover, ∂ ∩ C has no virtual cycle parameters and consequently no virtual centers.

The Boundary of the Shell Component
Let be a simply connected shell component and as usual, let H l denote the left half plane. Recall from Corollary 6.5 that the multiplier map ρ : → D * is a universal covering. Thus, there exists a biholomorphic map ϕ : H l → , unique up to precomposition by a Möbius transformation, such that

Theorem 6.13 (Extension to the boundary) Let F be a dynamically natural family and let ⊂ be a shell component. Let be a simply connected shell component in
and let ρ and ϕ be as above. Then, ϕ extends continuously to a map ϕ : H l → , where the closures are taken in C. Moreover, the point ϕ(∞) ∈ C is a virtual center.
Given a < b ≤ ∞, we say that an arc γ : (a, b) → C lands at z 0 ∈ C if lim t→b γ (t) = z 0 in the spherical metric.
Proof of Theorem 6.13 Let p be the period of . We first deal with the case p ≥ 2.
Let ω(t), t ∈ (0, 1), be an arbitrary arc in H l such that lim t→1 ω(t) = iθ * for some θ * ∈ R. We want to show that as t → 1, the accumulation set of λ(t) = ϕ(ω(t)) in ∂ = \ is a single point. That is, λ(t) lands at a point in the boundary (which could be infinity) when t → 1.
To this end, assume that λ(t) does not land at infinity and let λ * be a finite accumulation point of λ(t) as t → 1. This means that there exists a sequence t n → 1 as n → ∞, such that λ n = λ(t n ) ∈ satisfies λ n → λ * . Let a n denote the attracting cycle corresponding to λ n and set f * = f λ * .
If the attracting cycle a n remains bounded for all n > 0, it follows from Proposition 6.8 that f * has an indifferent cycle whose period divides p.
On the other hand if the cycle does not remain bounded, let n k → ∞ be a subsequence for which some point in the attracting cycle tends to infinity. Since λ n k → λ * , it follows by Proposition 6.8 that λ * is a virtual center.
Thus, all accumulation points of the curve λ(t) are either parameters with a neutral cycle of period dividing p or virtual cycle parameters, and both these sets are discrete subsets of . Since, the set of accumulation points must be connected [48, p.33], it follows that λ(t) accumulates at a single point, i.e., that λ(t) actually lands at a point λ * ∈ ∂ (which might be infinity).
Since ω(t) was arbitrary among all curves tending to iθ , all of them have the same property. By Lindelöf's theorem [48,Prop. 2.14], all their images by ϕ must land at the same point λ * . Hence ϕ has an unrestricted limit at every point iθ in the imaginary axis, and ϕ(iθ), θ ∈ R, is a continuous curve running along ∂ . Note that this curve might pass through infinity several times.
It remains to consider the case that ω(t) → ∞ in H l . Using exactly the same arguments as above, λ(t) = ϕ(ω(t)) must land at a single point λ c ∈ ∂ and hence the limit of ϕ at ∞ exists and is independent of the choice of ω(t). In particular, if we choose ω(t) to be the negative real axis, λ(t) → λ c as t → 1, and the multiplier γ (λ(t)) tends to zero. This shows that λ c is a virtual center and, by Theorem 6.10, also a virtual cycle parameter.
Since the unrestricted limit exists at every point of ∂H l , it follows that ϕ extends continuously to H l . If any of the limits is the point at infinity, continuity has to be understood in the spherical metric.
Finally suppose that p = 1. The same arguments as above show that ϕ has unrestricted limits at all points in ∂H l . Nevertheless, since in this case has no finite virtual centers, λ(1) = ∞.

Remark 6.15
If a shell component is not simply connected, the arguments above show that ϕ extends continuously to all points in on the boundary but the the role of the virtual center may be taken by a parameter singularity. An example of this can be found in the tangent family z → λ tan(z), where D * is a shell component of period one and the origin is a parameter singularity. In this example, the multiplier map is biholomorphic.

Remark 6.16
Our methods do not show that ∂ has a unique virtual cycle parameter because we have not discarded the possibility that while λ(1) = ϕ(iθ) = λ * ∈ C is a virtual cycle parameter, λ * = λ c , the virtual center of . If such a parameter were to exist, it would need to be isolated. Nevertheless, in view of Remark 6.11 we believe that this should never happen. We state this as a conjecture.

Conjecture 6.17
In the setup of Corollary 6.14, ∂ has a unique virtual cycle parameter which is the virtual center.
Theorem 6.13 also allows us to define an internal structure in the simply connected shell component .

Corollary 6.18 (Internal rays)
Under the hypotheses of Theorem 6.13 and Corollary 6.14, define the internal ray in of angle θ by Then, all internal rays have one end at the virtual center and the other end at a point in ∂ (which could be infinity). If the virtual center λ * is finite no internal ray has both ends at λ * .
Notice that the internal rays foliate the shell component. The rays landing at parameters corresponding to parabolic cycles of multiplier one, that is the preimages of {0, 1} under the multiplier map extended to ∂ , divide into fundamental domains which are mapped one to one to the punctured disk by the multiplier map ρ. See Fig. 1.
We have shown that simply connected shell components of period one are always unbounded. In the next section we look at some examples of dynamically natural slices of parameter spaces for specific families. The first is the exponential family {E λ = e z + λ} λ∈C where all hyperbolic components are unbounded. Other examples are dynamically natural slices of meromorphic families F ⊂ M ∞ . In these parameter planes we observe that all of the simply connected shell components of period greater than one are bounded. These observations lead us to the following conjecture.

Examples
Our results apply to very general families of entire and meromorphic functions. To find examples, we need to look at families whose singular sets depend holomorphically on the parameters. We also want to be sure that if we perturb a hyperbolic map in our family by a holomorphic motion, the resulting map is still in the family. And finally, to compute a dynamically natural slice, we would like to have an explicit formula for the maps that contains the parameters as constants. In the appendix, we show how to find fairly large classes of families with these properties.
The examples in this section are all families of entire and meromorphic functions that have two asymptotic values and either no critical values or just one critical value. That the formulas we use for the functions are valid under perturbation follows from the material in the appendix.

Example 1 Any meromorphic function with exactly two asymptotic values and no critical points belongs to the family
See the Appendix and the references there. The asymptotic values are a/c and b/d. If b or c is equal to zero, the function is the exponential function and infinity is an asymptotic value. Otherwise both asymptotic values are finite. Any function in the family is determined by three complex constants. We will look at several dynamically natural slices of F 2 and use different normalizations for each.
(a) The exponential family E λ (z) = e z + λ is one of the best understood dynamically natural slices of F 2 . It is the slice defined by fixing one of the asymptotic values of the functions f λ ∈ F 2 at infinity by setting c = 0 and taking the other as the parameter λ = b/d. We can put these functions into the form above, normalizing by an affine conjugation that replaces z by 2z. Choosing the coefficients as a = 1/2, d = 2, c = 0 and b = λ we get E λ (0) = 1 + λ. The infinite asymptotic value has neither a forward nor backward images (poles). Thus the dynamics of the functions in this slice are determined by the orbit of the other asymptotic value, v λ = λ ∈ = C. Since every meromorphic map with these properties is affine conjugate to a member of this family, property (e) holds. Note that another normalization for {E λ } λ in common use is one that fixes the second asymptotic value at the origin. The family then takes the form ae z and the parameter is an affine function of the first iterate of the finite asymptotic value. Standard references include [17,26,50,52]. Figure 3 shows the parameter plane for E λ . It is known that all hyperbolic components in are unbounded and simply connected and that the multiplier goes to zero as the parameter goes to infinity inside the component; that is, infinity acts as a virtual center for all the hyperbolic components. In the figure, components are colored according to their respective period, yellow for 1, cyan-blue for 2, red for 3, brownish green for 4, etc. (b) Another dynamical slice that can be extracted from F 2 is formed by requiring the origin to be a fixed point with persistent multiplier ρ 0 , |ρ 0 | < 1. The resulting slice is = C \ {0, ρ 0 /2} and the maps have the form with μ = ρ 0 λ/(2λ − ρ 0 ). The asymptotic values are λ and −μ and at least one of them is attracted by the origin. See [21,22] for a discussion of the topology of the capture component for maps in the slice where ρ = 2/3 and [37] for a discussion of properties of maps in the slice where ρ 0 = 1/2. To see that condition (e) is satisfied, observe that any conjugacy which is conformal in the basin of attraction of 0 keeps the multiplier of this fixed point unchanged. Since the resulting conjugate map again has two asymptotic values and no critical points, it belongs to F 2 . Using an affine conjugacy, we may require that the new fixed point be the origin again; thus the new map belongs to the slice. Figure 4 shows the parameter plane for for ρ 0 = 2/3 and hence μ = −λ/(1 − 3λ) and = C \ {0, 1/3}. As in Fig. 3, the color represents the period of the attracting orbit that the free asymptotic value λ is attracted to. It does not reflect the behavior of the orbit of the other asymptotic value −μ. In the unbounded green capture component on the left Fig. 3 The parameter plane of the exponential family E λ (z) = exp(z) + λ. Components are colored according to their respective period, yellow for 1, cyan-blue for 2, red for 3, brownish green for 4, etc (Color figure online) λ is attracted to the origin. As we see below, −μ may or may not also be attracted to the origin in this component. In the yellow shell component on the right, λ is attracted to an attracting fixed point different from zero while −μ is attracted to zero. It follows from Theorem C (c) that this component is indeed unbounded. In the cyan-blue bounded shell components, λ is attracted to a cycle of period 2; in the red ones the attracting cycle has period three, etc. Bifurcations occur at parabolic parameters as usual, giving rise to shell components of all periods attached to the boundary of any given shell component. In Fig. 5 we again plot the λ−plane but we show the dynamics of the second asymptotic value −μ. On the unbounded green region of Fig. 5, −μ is attracted to the origin. This figure should be superimposed on top of Fig. 4 near the origin in the large green component on the left in Fig. 4; it is relatively so small that it almost vanishes. In intersection of the green regions of the the combined figure both asymptotic values are attracted to the origin. On the complement of the green region in Fig. 4, λ is attracted to the origin and −μ is free. The yellow region is a shell component of period one which should have a virtual center at the parameter λ = 1/3, (where μ = ∞) but this is a parameter singularity. Observe that μ is not an affine function of the parameter λ; this would explain the numerics which seem to indicate that this period one component is bounded.
In fact, if we were to reparametrize the family so that μ depended affinely on the parameter, we would see the same picture as in Fig. 5 because of the symmetry in the equation connecting μ and λ. = C * and v λ = λi [30,41,42]. This is a dynamically natural slice in the relaxed sense of Remark 4.7 because the second asymptotic value is s 1 (λ) = −v λ = −λi and T λ is symmetric. It follows that the bifurcation locus is the same for both singular values. Condition (e) should be replaced by the condition that the relation s 1 (λ) = −v λ = −λi must persist under deformation. Figure 6 shows the parameter plane of T λ . 1 There is a lot of symmetry in this family. We have

In addition
This says, for example, that if for some λ 0 , z 1 is a fixed point of T λ 0 , then T −λ 0 (z 1 ) = −z 1 so that z 1 is a period 2 point of T −λ 0 . It also says that complex conjugate values of λ have conjugate periodic cycles. The figure shows the λ plane where we can observe these symmetries. We only follow the orbit of one asymptotic value, λi, and we color the components based on the period of the cycle it is attracted to. The colors do not reflect the behavior of the orbit of the other asymptotic value. Thus, we see the same color for a λ value for which there are two separate attracting periodic cycles of period 2k as we do when there is a single attracting periodic cycle of period 2k attracting both asymptotic values. There is a single non-simply connected component, the punctured unit disk, for which both asymptotic values are attracted to the origin. There are two unbounded components, one on the right (yellow) and one on the left (cyan-blue). In the one on the right, λi is attracted to a fixed point a 0 with multiplier ρ 0 , and −λi is attracted to −a 0 with the same multiplier. It has the same color as the unit disk. In the unbounded component on the left, both λi and −λi are attracted to the attracting period two cycle a 0 , −a 0 and its multiplier is ρ 2 0 . Neither of these components has a finite virtual center. Although the left one is colored for period 2, it cannot have a finite virtual center because if λ * were a finite virtual center, −λ * would be a virtual center for the unbounded component on the right of period 1, contradicting Corollary 6.14. Note that, except for the punctured disk, each bounded component is paired with another bounded component; the two have a common virtual center and are tangent there. The two unbounded components can be thought of as paired at infinity. The relationship between these component pairs is discussed in in [41] and in detail for λ on the imaginary axis in [20]. (a) If we start with an an f ∈ F 2 with finite asymptotic values and pre-compose by a degree two polynomial, Q(z) = az 2 + bz + c, the resulting function

Example 2 Functions in
has the same two asymptotic values as f does. Since each asymptotic tract of f has two preimages under Q, each asymptotic value of g has two asymptotic tracts and so has multiplicity two. Since f has no critical points, g has a single critical point at −b/2a and a single critical value at f (−b/2a). If we assume the origin is fixed, then c = 0. If we assume it is also a critical value then b = 0. Now making one of the asymptotic values land on the fixed point 0, we can form a dynamically natural slice. This determines the coefficient a. This is a slice in the relaxed sense of Remark 4.7, because one of the singular values is preperiodic. The functions in the slice can then be written The asymptotic values are λ and − √ πi. The singular parameter values are at 0 and − √ πi. Theorem 8.3 implies that condition (d) holds. Figure 7 shows the λ-plane. The green components are capture components where the orbit of λ falls into the basin of attraction of the origin. In the unbounded yellow shell components, λ is attracted to an attracting (not super attracting) fixed point. In the cyan-blue (bounded) components it is attracted to a cycle of period two, in the red ones to a cycle of period three and in the brownish green ones to a cycle of period four. The components come in pairs because the asymptotic values have multiplicity two. This is discussed in [23]. (b) We can form a dynamically natural slice, in the relaxed sense, with the functions h = a f 2 + b f + c, for f ∈ F 2 by assuming the origin is a super attracting fixed point and that the two asymptotic values coincide. In this case the functions in the slice each have one asymptotic value of multiplicity two. We can write these functions as The parameter λ is the free (double) asymptotic value and so determines a dynamically natural slice of the parameter space = C \ {0}. Condition (e) is satisfied in the modified sense.
In the left plot of Fig. 8 we see the λ−plane. In the center of the figure we see a green capture component formed by parameters for which the asymptotic value belongs to the immediate basin of 0. This component is doubly connected because of the puncture at λ = 0. The remaining green components are capture components for higher iterates.
In the two yellow shell components the period is one. They are unbounded in accordance with Proposition 6.12. The remaining shell components are all of higher period and numerics indicate that they are all bounded. We see that they are grouped into quadruples that share a virtual center. This is because the asymptotic value has two asymptotic tracts and each asymptotic tract has two pre-asymptotic tracts.
Compare this figure to the right plot in Fig. 8 which shows the parameter plane for the family λ tanh z 2 . It is essentially the "square root" of this slice, since both maps are semiconjugate by z 2 . The asymptotic values now have multiplicity 2. These examples are investigated further in [23] where it is proved that all shell components of period greater than 1 are bounded.

Example 3
Our last example is a slice in the family R(z)e Q(z) (see the Appendix) where R(z) is rational and Q(z) is a polynomial. Choosing the degrees of R and Q to be 1, we obtain a function with two asymptotic values, one of which is infinity. It also has two critical points and two critical values. For our slice, we fix one of the critical points at infinity so it is both a critical point and an asymptotic value. We let the second asymptotic value a be free. We normalize so that the origin is an attracting fixed point with fixed multiplier |ρ| < 1; the origin attracts the critical value that is the image of the finite critical point which we take as ρ/(ρ + a). Note there is then a single pole at −a/(ρ + a). With this normalization, the functions depend only on a and have the form The previous examples contained only two types of components, shell components and capture components. This example also contains very small yellow Mandelbrot sets at the ends of the Cantor bouquets inside the big green capture component in the blow-up (Fig. 9). They correspond to parameters where the critical point behaves independently, while the asymptotic value a is attracted to 0. There are also both bounded and unbounded shell components in the picture. In the blow-up, we see bounded shell components bifurcating from the unbounded yellow fixed one on the left. The virtual center of the period two component on the right is at infinity. This example has only one pole, so infinity must be an asymptotic value. It could be thought as a "hybrid" between a slice in an entire family and a meromorphic slice in M ∞ . The other examples have either infinitely many poles or none.

Further Examples
Other one-dimensional slices recording the position of a free asymptotic value while keeping the other ones under control have been studied in the literature. These are mostly entire functions and we refer the reader to [8,24] for these examples.

Appendix
Any meromorphic function that is a branched cover of the sphere of finite degree d is a rational function and so can be expressed as a quotient of relatively prime polynomials at least one of which has degree d. The 2d +1 coefficients of these polynomials define a natural embedding of the space of such functions into C 2d+1 . Up to affine conjugation then, these functions are represented as a complex analytic manifold of dimension 2d − 2. This family is denoted by Rat d in the literature.
Not all holomorphic families of meromorphic functions have such an obvious representation as a complex manifold. In this appendix we consider some transcendental families that do. These are the families that most of our examples are drawn from. The functions have explicit expressions that involve complex constants. We show that under quasiconformal deformation, the deformed function has a similar expression in which only the constants are changed. Thus the constants determine the embedding of the manifold into C n for the appropriate n.
We begin with families of meromorphic functions with p < ∞ asymptotic values, no critical points and a single essential singularity at infinity. We denote these by F p . (See [30,32,41] for further discussion.) These functions have a particularly nice characterization. Recall that the Schwarzian derivative of a function g is defined by Nevanlinna, [46,47], Chap X1, proved Theorem 8.1 Every meromorphic function g with p < ∞ asymptotic values and q < ∞ critical points has the property that its Schwarzian derivative is a rational function of degree p + q − 2. If q = 0, the Schwarzian derivative is a polynomial P(z). In the opposite direction, for every polynomial function P(z) of degree p − 2, the solution to the Schwarzian differential equation S(g) = P(z) is a meromorphic function with exactly p asymptotic values and no critical points.
Since the theorem is classical and the proof is not well known, we sketch the proof here and refer the reader to the literature for details.

Proof
The proof follows from the construction of a function with p asymptotic values and no critical points as a limit of rational functions with p branch points. Letting the order of the branching at some or all of these p points increase, one obtains a sequence of rational functions. In the limit, the images of the branch points whose order goes to infinity become logarithmic singularities and the limit function has finitely many branch points and finitely many logarithmic singularities. The limit function is a parabolic covering map from the plane to itself. The Schwarzian derivatives of the rational functions in the sequence are again rational functions with degree determined only by the number of branch points, not their order. The limit of the Schwarzian derivatives is the Schwarzian derivative of the limit and so must be rational.
If all the branch points become asymptotic values in the limit, the limit function has no critical points or critical values and hence its derivative never vanishes. It follows from the definition of the Schwarzian that in the limit, it must be a polynomial.
It is classical, ( see e.g., [38]), that solutions to the Schwarzian equation S(g) = P(z) are related to solutions of the linear second degree ordinary differential equation Such an equation has a two dimensional space of holomorphic solutions. If w 1 , w 2 are a pair of linearly independent solutions, and g = aw 2 + bw 1 cw 2 + dw 2 , a, b, c, d ∈ C, ad − bc = 1, it is easy to check that S(g) = P(z). Moreover, if g is any solution of the Schwarzian equation, w = 1/g is a solution of the linear equation.

Remark 8.2
One of the basic features of the Schwarzian derivative is that it satisfies the following cocycle relation: if f , g are meromorphic functions then S(g( f ))(z) = S(g( f )) f (z) 2 + S( f (z)). (8.3) In particular, if T is a Möbius transformation, S(T (z)) = 0 and S(T •g(z)) = S ((g(z)) so that post-composing by T does not change the Schwarzian. Under pre-composition by a Möbius transformation the Schwarzian behaves like a quadratic differential. In particular, pre-composing by an affine transformation multiplies the Schwarzian by a constant.
This means that if we find a specific pair (w * 1 , w * 2 ) of solutions to the second order linear equation and set g * = w * 2 /w * 1 , then every solution of the Schwarzian equation S(g) = P(z) has the formula g = ag * + b cg * + d , a, b, c, d ∈ C, ad − bc = 1, In particular, if p = 2, P(z) is identically constant. Using one of the constants in an affine conjugation, we may assume P ≡ −1/2; then a specific pair of solutions to equation (8.2) is w * 1 = e − z 2 , w * 2 = e z 2 so we have g * = w * 2 /w * 1 = e z . The functions in the family F 2 have the form g(z) = ae z + b ce z + d , a, b, c, d ∈ C, ad − bc = 1.
Since there is one more degree of freedom from the affine conjugation, we see this is a three dimensional family. We have discussed several dynamically natural slices of F 2 in this paper. For F 3 , P(z) is linear and two specific solutions to the second order linear equation w + 1 2 ζ w = 0 are given by the Airy functions 2 Ai(ζ ), Bi(ζ ). Setting g * (ζ ) = Ai(ζ )/Bi(ζ ) we obtain a solution with three asymptotic values, 0, i and −i. Since we are interested in the dynamics, we may conjugate by the affine transformation ζ = r z + s. We may thus transform any function whose Schwarzian is linear, and hence any function in F 3 , up to affine conjugation, to one given by the formula g(ζ ) = ag * (ζ ) + b cg * (ζ ) + d , a, b, c, d ∈ C, ad − bc = 1.
The asymptotic values of g are b d , ai+b ci+d , −ai+b −ci+d . We could define a dynamically natural slice, for example, by choosing the coefficients a, b and d so that g(1) = 1, and g (1) = 1/2.
It is not hard to compute that the remaining coefficient c is an affine function of the asymptotic value b d . There are standard functions that solve the second order equation when the coefficient polynomial is of degree two or three, and they give formulas for functions in F 4 and F 5 .
The following families of holomorphic functions have rational Schwarzian derivatives and are determined by their topological covering properties. We list them here and refer the reader to the cited literature for further discussion of them.
• Functions of type Re Q for R rational and Q polynomial. See [19,55] for discussions with R polynomial. • Functions of type

Compositions
The following theorem shows how we can find more families for which the functions have explicit formulas. (See [23] for further discussion of these functions).

Theorem 8.3
Let f 0 ∈ F p 1. Suppose g 0 = f 0 • Q 0 is a function such that Q 0 is a polynomial of degree d and suppose that g is a meromorphic function quasiconformally conjugate to g 0 . Then g can be expressed as g = f • Q for some function f ∈ F p and some polynomial Q of degree d. 2. Suppose R 0 ∈ Rat d is rational of degree d and h 0 = R 0 • f 0 . Then if h(z) is quasiconformally conjugate to h 0 , h can be expressed as h = R • f for functions R ∈ Rat d and f ∈ F p .

Remark 8.4
The proof of part 1 of the theorem works if Q 0 is replaced by a rational function R 0 ∈ Rat d but then the composed function g 0 has essential singularities at the poles of R 0 .

Remark 8.5
Note that in part 1, g = f • Q is a function with rational Schwarzian. For each asymptotic value of f of multiplicity m, g has an asymptotic value of multiplicity dm; moreover g has the same critical points as Q, namely 2d −2 critical points counted with multiplicity, d − 1 of which are at infinity. In part 2, however, it is no longer true that the Schwarzian of the composed function h = R • f is a rational function. For example, it may have infinitely many critical points.

Remark 8.6
Observe that if f ∈ F p and Q is a degree d polynomial then Q • f and f • Q are semiconjugate by a degree d polynomial -in fact by Q.
Proof The proof of both parts of the theorem is essentially the same. We therefore carry it out only for part 1.
In part 1, let φ μ be a quasiconformal homeomorphism with Beltrami coefficient μ such that is meromorphic. We can use f 0 to pull back the complex structure defined by μ := ∂φ μ /∂φ μ to obtain a complex structure ν = f * 0 μ such that the map is meromorphic. Note that this is not a conjugacy since it involves two different homeomorphisms.
We can now write Again this is not a conjugacy but it is meromorphic since g μ and f μ are homeomorphisms.
The main point here is that although f μ is not a conjugate of f 0 , since the quasiconformal maps φ μ and φ ν are homeomorphisms, the map f μ is a meromorphic map with the same topology as f 0 ; that is, it has p asymptotic values and no critical values. By Nevanlinna's theorem, Theorem 8.1, f μ belongs to F p . Similarly, although Q μ is not defined as a conjugate of Q 0 , since the quasiconformal maps φ ν and φ η are homeomorphisms, the map Q μ is a meromorphic map with the same topology as Q 0 ; that is, it is a degree d branched covering of the Riemann sphere with the same number of critical points and the same branching as Q 0 and thus it must be a polynomial of degree d.