Uniformization of Julia sets of geometrically finite rational maps

Suppose that there exists a quasiconformal map g of \widehat \mathbb{C} that maps each Fatou component of f onto a disk. Condition ? is immediate. Each Fatou component is a quasidisk. By Proposition 5.2 and Remark 5.3, if f has a parabolic periodic point a_0, then its multiplicity must be at least 3. On the other hand, if the multiplicity is strictly greater than 3, then at least three Fatou components intersect at a_0, which is a contradiction. Thus ? holds. If a_0 lies in the boundary of a Fatou component U, then U is periodic by Lemma 4.2 ? and U is an immediate basin of a_0 by Proposition 5.2. This establishes condition ? of Theorem 1.1. Finally, suppose that U,V are Fatou components with \overline{U}\cap \overline{V}\neq \emptyset. Since the Julia set of f is not a Jordan curve, we have \partial U\neq \partial V. Since g(U) and g(V) are tangent disks, \overline{U}\cap \overline{V} must consist of a single point a_0. By Theorem 7.1, the point a_0 is mapped to a parabolic periodic cycle of multiplicity 3 by an iterate of f that is a local homeomorphism near a_0, as required in condition ? of Theorem 1.1.

Conversely, suppose that conditions ?, ?, ?, ? hold and that the Julia set of f is not homeomorphic to \mathbb S^1 or \mathbb S^2. If there exists only one Fatou component, then this component is a Jordan region by ?, so its complement, i.e., the Julia set, has non-empty interior. However, this forces the Julia set to be the entire sphere [Milnor:dynamics]*Corollary 4.11, a contradiction. Thus, there exist at least two Fatou components.

Our goal is to verify the assumptions of Theorem 7.1 ?. By assumption ?, there are no parabolic points of multiplicity 2. Let U be a Fatou component of f. By ?, the orbit of \partial U contains the orbit of at most one parabolic periodic point a, in which case an immediate basin of a is in the orbit of U. Thus, Proposition 5.2 applies and U is a quasidisk. By condition ?, if U,V are distinct Fatou components with \overline{U}\cap \overline{V}\neq \emptyset, then the intersection \overline{U}\cap \overline{V} consists of a single point a_0 that is not critical and it is eventually mapped to a parabolic periodic cycle of multiplicity 3. Note that f(U) and f(V) are Fatou components that are Jordan regions by ?, and they are distinct by Lemma 4.2 ?, given that a_0 is not a critical point. Applying condition ? to the pair f(U),f(V), we see that f(a_0) is the unique point in the intersection \overline{f(U)}\cap \overline{f(V)} and f(a_0) is not a critical point. Hence, inductively, the entire forward orbit of a_0 is not critical. Hence, a_0 is mapped to a parabolic periodic cycle of multiplicity 3 by an iterate of f that is a local homeomorphism near a_0, as in the assumptions of Theorem 7.1.

We now apply Theorem 7.1, which implies the existence of a quasiconformal homeomorphism g\colon \widehat \mathbb{C}\to \widehat \mathbb{C} that maps each Fatou component of f onto a disk and is 1-quasiconformal on the Julia set J(f). Since J(f) has measure zero [McMullen:Hausdorff_GF]*Corollary 6.2, the 1-quasiconformality condition holds vacuously for any such quasiconformal map g. By the last part of Theorem 7.1 we conclude that g|_{J(f)} is unique up to post-composition with Möbius transformations.

Suppose that the Julia set of f is a Jordan curve. Then the Fatou components of f can be either attracting or parabolic; see the discussion after Theorem 1.5 in [LuoNtalampekos:piecewise_qc]. This implies that f has no critical points on J(f). Let U be a Fatou component. By Proposition 5.2, U is a quasidisk if and only if either f has no parabolic periodic points or f has a parabolic periodic point of multiplicity at least 3. Since there are only two Fatou components the multiplicity of parabolic points is at most 3; see [Milnor:dynamics]*Section 10. Finally, note that f has no parabolic periodic points if and only if it is hyperbolic, given that it has no critical points on the Julia set [Milnor:dynamics]*Theorem 19.1.

Suppose that f is a unicritical geometrically finite polynomial. If there is no bounded Fatou component, the statement is vacuously true, so we assume that there exists a bounded Fatou component. All bounded periodic Fatou components are either hyperbolic or parabolic. Thus, the unique (finite) critical point of f is contained in the Fatou set and there are no critical points on the Julia set.

Suppose that there exists a quasiconformal map of \mathbb{C} that maps each bounded Fatou component of f onto a disk. Suppose first that J(f) is not homeomorphic to \mathbb S^1. By Theorem 1.2, conditions ? and ? hold. Suppose that f has no parabolic points. By ?, we conclude that bounded Fatou components have disjoint closures. Since there are no critical points on the Julia set, f is hyperbolic [Milnor:dynamics]*Theorem 19.1. Now, if J(f) is homeomorphic to \mathbb S^1, by Proposition 1.3, either f has a parabolic periodic point of multiplicity 3 or f is hyperbolic, as desired.

Conversely, suppose that f is hyperbolic and all bounded Fatou components have disjoint closures. If J(f) is homeomorphic to \mathbb S^1, then by Proposition 1.3 there exists a quasiconformal map g of \mathbb{C} that maps the bounded Fatou component of f onto a disk. If J(f) is not homeomorphic to \mathbb S^1, then conditions ?, ?, ? are vacuously true, so by Theorem 1.2 there exists a quasiconformal map g of \mathbb{C} that maps each bounded Fatou component of f onto a disk.

Finally, suppose that f has a parabolic periodic point of multiplicity 3. Since f is unicritical, there is a unique periodic cycle of bounded Fatou components, all of which are immediate basins of this parabolic point; see [Milnor:dynamics]*Theorem 10.15. Hence, conditions ? and ? hold. By Lemma 7.4 below, condition ? holds. Thus, by Theorem 1.2 we conclude that there exists a quasiconformal map g\colon \mathbb{C}\to \mathbb{C} that maps each bounded Fatou component of f to a disk.

Suppose that f(z) = z^d+c is hyperbolic, and the closures of any two distinct bounded Fatou components are disjoint. Let \Omega_1,\dots, \Omega_p be the unique cycle of periodic Fatou components of f. Then there exist disjoint open sets V_i, i\in \{1,\dots,p\}, and open sets U_i, i\in \{1,\dots,p\} such that \Omega_i \Subset U_i \Subset V_i and the map f^{\circ p}\colon U_i \to V_i is polynomial-like. By the straightening theorem [CarlesonGamelin:Dynamics]*Theorem VI.1.1, this polynomial-like map is hybrid equivalent to a unicritical polynomial with an attracting fixed point whose immediate basin contains the critical point. Thus, the corresponding straightening parameter lies in the main hyperbolic component of the Multibrot set. Since the neighborhoods V_i, i\in \{1,\dots,p\}, are chosen pairwise disjoint, the associated renormalization is primitive. Thus, c is contained in the main hyperbolic component of a primitive Multibrot copy. See [DudkoLyubichMLCFeigenbaum]*§ 2.2.3–2.2.4 and [MilnorOrbitPortraits]*§ 6 for the discussion for the Mandelbrot set, and the case for Multibrot set is similar.

Conversely, suppose that c lies in the main hyperbolic component of a primitive Multibrot copy. Then the distinct periodic bounded Fatou components have disjoint closures. We claim that the same is true for all bounded Fatou components. Suppose, to the contrary, that two distinct bounded Fatou components U, V have intersecting closures. Since f is hyperbolic, the critical point lies in the Fatou set. Therefore, the images f^{\circ n}(U), f^{\circ n}(V) remain distinct bounded Fatou components whose closures meet at a point (Lemma 4.2 ?). Since every bounded Fatou component is eventually periodic, this contradicts the disjointedness of the closures of distinct periodic bounded Fatou components. Hence all bounded Fatou components have pairwise disjoint closures.

Suppose f(z) = z^d +c has a parabolic periodic point of multiplicity 3. Let a be a parabolic periodic point with period p. Since f is unicritical, f has only one periodic cycle of bounded Fatou components. So the two attracting basins of the parabolic periodic point are contained in the same cycle. Thus, (f^{\circ p})'(a) = -1. In the family of unicritical polynomials the multiplier gives a branched covering map of a hyperbolic component to \mathbb{D} [MilnorOrbitPortraits]*Theorem 6.5 and Remark 6.4. Thus the period-doubling points correspond to exactly those with multiplier -1. Therefore, c is a period-doubling bifurcation point.

Conversely, suppose c is a period-doubling bifurcation point. Then there exists a parabolic periodic point a of period p with (f^{\circ p})'(a) = -1. Therefore, there are 2n attracting basins at a for some n\in \mathbb{N}; see [Milnor:dynamics]*Lemma 10.4. Since f is unicritical, there is a unique cycle of attracting basins; see [Milnor:dynamics]*Theorem 10.15. Thus, there are exactly two attracting basins at a, interchanged by f^{\circ p}. In particular, the parabolic multiplicity is 3.

Suppose that there exists a quasiconformal map \phi\colon \widehat \mathbb{C}\to \widehat \mathbb{C} that maps K_1\setminus \{0\} and K_2\setminus \{0\} to lines parallel to the real axis. Let g(z)=z+1 and f=\phi^{-1}\circ g\circ \phi(z). Then f satisfies the assumptions in the statement.

Conversely, by conjugating f near 0 with a quasiconformal map \psi as in the statement, we can reduce to the case that f is a parabolic germ. Note that the conjugacy \psi is only defined on subarcs K_1',K_2' of K_1,K_2 rather than on the entire curves, so we invoke Theorem 2.4 to extend \psi(K_i'), i=1,2 to suitable quasicircles. By Theorem 3.4, there exists a quasiconformal map of \widehat \mathbb{C} that maps these quasicircles to tangent circles. Finally, by Lemma 2.8 we obtain the desired conclusion for the initial quasicircles K_1,K_2.

If U=\widehat \mathbb{C}, then f is a Möbius transformation, so it is \eta-quasi-Möbius with \eta(t)=t. Suppose that \widehat \mathbb{C}\setminus U\neq \emptyset, so \widehat \mathbb{C}\setminus f(U)\neq \emptyset. Consider Möbius transformations \phi,\psi such that \phi(U)\subset \mathbb{C} and \psi(f(U))\subset \mathbb{C}, and let g=\psi \circ f\circ \phi^{-1}\colon \phi(U)\to \psi(f(U)). By covering \phi(E) with finitely many balls in \phi(U) and applying Koebe’s distortion theorem to each of them, we see that there exists a constant M>0 depending on g and a constant L\geq 1 that depends only on \phi(E) and \phi(U) such that

Thus, g\colon (\phi(E),|\cdot|)\to (\psi(f(E)), |\cdot|) is uniformly quasi-Möbius, depending only on U and E. The identity map from the Euclidean to the chordal metric preserves cross ratios, so g\colon (\phi(E),\chi)\to (\psi(f(E)), \chi) is uniformly quasi-Möbius, depending only on U and E. Finally, since \phi,\psi are Möbius transformations, the map f\colon (E,\chi)\to (f(E),\chi) is uniformly quasi-Möbius depending only on U and E, as desired.

First, we extend \phi to \widehat \mathbb{C} as follows. By composing \phi with isometries of \mathbb{C} we assume that z_1,z_2\in \mathbb{C}. By shrinking W_1, we may assume that W_1 is a disk centered at z_1 and \phi is a quasiconformal map from a neighborhood of \overline{W_1} into \mathbb{C}. By Theorem 2.1, \phi is quasisymmetric in \overline{W_1}. Thus, \phi(\partial W_1) is a quasicircle (see [Pommerenke:conformal]*Proposition 5.10) and W_2\coloneqq \phi(W_1) is a quasidisk by Theorem 2.3. By Proposition 2.5, \phi|_{W_1} extends to a quasiconformal map of \widehat \mathbb{C}. We denote the extension by \phi for simplicity.

Without loss of generality, suppose that the pair U_2,V_2 is quasiconformally circularizable. Thus, there exists a quasiconformal map \psi of \widehat \mathbb{C} such that \psi(U_2),\psi(V_2) are half-planes that are tangent at \psi(z_2)=\infty. Let g=\psi\circ \phi. Then g(\partial U_1) and g(\partial V_1) are Jordan curves in \widehat \mathbb{C} that are straight lines in a neighborhood of \infty. It is elementary to find a chord-arc curve J\subset \widehat \mathbb{C}\setminus (g(\overline{U_1})\cup g(\overline{V_1})) that is a straight line in a neighborhood of \infty and otherwise it is smooth; see Figure ?. By compactness, there exists a constant K\geq 1 such that N_{1/K}(J)\subset \widehat \mathbb{C}\setminus (g(\overline{U_1})\cup g(\overline{V_1}))\subset N_K(J). Thus, the conclusion of Lemma 2.7 is true. By Theorem 2.6 we conclude that the pair g(U_1),g(V_1) is quasiconformally circularizable. Equivalently, the pair U_1,V_1 is quasiconformally circularizable.

Note that U is a Fatou component and \partial U is contained in the Julia set [Milnor:dynamics]*Lemma 10.5. Thus, \partial U cannot intersect any petal corresponding to an attraction vector for 0. By the parabolic flower theorem [Milnor:dynamics]*Theorem 10.7, there exist arbitrarily small petals corresponding to the attraction and repulsion vectors of f whose union, together with 0, gives a neighborhood of 0. Also, these petals are Jordan regions cyclically ordered and each attracting petal intersects exactly two repelling petals, except in the case that the parabolic multiplicity is 2, where there exist only two petals, an attracting one and a repelling one. Since U cannot intersect any attracting petal other than the one corresponding to v, we conclude that \partial U contains two non-overlapping arcs E_1,E_2 contained in repelling petals P_1,P_2, respectively, which correspond to vectors adjacent to v. Moreover, since f(\partial U)=\partial U, if g is the local branch of f^{-1} that fixes 0, then g maps each arc E\subset \partial U near 0 onto an arc g(E)\subset \partial U.

If the multiplicity of 0 is not 2, then P_1,P_2 are disjoint attracting petals of g. For i\in \{1,2\}, each point of E_i converges to 0 under iteration of g within the petal P_i. Note that g(E_i) is an arc in \partial U\cap (P_i\cup \{0\}) with an endpoint at 0 and either it contains E_i or it is contained in E_i. If g(E_i)\supsetneq E_i then the endpoint z_0 of E_i that is different from 0 satisfies g^{\circ k}(z_0) \in (\partial U\cap P_i)\setminus E_i for each k\in \mathbb{N}. However, E_i is a neighborhood of 0 in \partial U\cap P_i, so \lim_{k\to\infty}g^{\circ k}(z_0)\neq 0, a contradiction. We conclude that g(E_i)\subset E_i, as desired.

Next, suppose that the multiplicity of 0 is 2. Then P_1=P_2\eqqcolon P. By shrinking P if necessary, we assume that g is injective in \overline{P} and \partial U\setminus \overline{P}\neq \emptyset, so \partial U \cap P has at least two components A_i\supset E_i\setminus \{0\}, i=1,2. Note that A_i is an arc in \partial U \cap P with an endpoint at 0 and another endpoint a_i\in \partial P. Hence, g(A_1)\subset A_1 and g(A_2)\subset A_2 or g(A_1)\subset A_2 and g(A_2)\subset A_1. The map g\colon \overline{P}\to g(\overline{P}) induces a bijection between the set \{0,a_1,a_2\}\subset \partial P and \{0,g(a_1),g(a_2)\}\subset g(\partial P). Since g is orientation-preserving, it preserves the cyclic order of the points 0,a_1,a_2. From this it follows that g(a_1)\in \overline{A_1} and g(a_2)\in \overline{A_2}. Therefore, g(A_1)\subset A_1 and g(A_2)\subset A_2. Arguing as in the previous paragraph, for i=1,2 we see that g(E_i)\subset E_i, otherwise, the endpoint of E_i that is different from 0 would not converge to 0 under iteration of g.

By Lemma 3.6, we may assume that K_1=\mathbb{R}\cup \{\infty\}. Without loss of generality suppose that K_2 is in the closure of the lower half-plane. Since the real line \mathbb{R} is locally preserved under f near 0, we conclude that P_1 and P_2 are parabolic petals in the direction of the positive and negative real axis, respectively. By replacing f with its inverse if necessary, we may assume that P_1 is an attracting petal in the direction of the positive real axis. Thus, f has an expansion at 0 of the form

where a<0 and n\in \mathbb{N}. We conjugate f by z\mapsto 1/z to the parabolic germ g at \infty with expansion

Let J be the image of K_2 under z\mapsto 1/z. Then J contains an arc J^+ with an endpoint at \infty that is invariant and is attracted to \infty by g in the direction of the positive real axis. Similarly, J contains an arc J^- with an endpoint at \infty that is invariant and is attracted to \infty by g or g^{-1} in the direction of the negative real axis. In order to prove that there exists a quasiconformal homeomorphism of \widehat \mathbb{C} that maps \mathbb{R} and J\setminus \{\infty\} to parallel lines, by Theorem 2.9 it suffices to prove that there exists L\geq 1 such that

It suffices to show the inclusion for the arcs J^+ and J^-. We focus on J^+.

Let w = \phi(z) = z^n/(-na), which maps the sector \Delta=\{re^{i\theta}: r>0,\,\, |\theta|<\pi/n\} conformally onto the slit plane \mathbb{C}\setminus (-\infty,0]. We set

as w\to\infty. The calculation of the expansion can be found in [Milnor:dynamics]*(10.2), p. 107. Let R>0 be sufficiently large so that

In particular, if w\in \mathbb H_{>R}, then \operatorname{Re}(G(w))>\operatorname{Re}(w)+3/4, so G(\mathbb H_{>R})\subset \mathbb H_{>R}. Note that for any k\in\mathbb{N} the map G^{\circ k}\colon \mathbb H_{>R}\to \mathbb H_{>R} is a conformal embedding. Since the real axis is locally invariant under G near \infty, we may consider an even larger R>0 so that G^{-1}(\mathbb H_{>R}\cap \mathbb{R})\subset \mathbb{R}. This implies that G(\mathbb H_{>R}\cap \mathbb H)\subset \mathbb H_{>R}\cap \mathbb H.

Since J^+ is invariant under g and is attracted to \infty in the direction of the positive real axis, by replacing it with an iterate if necessary (see Lemma 3.5), we may assume that J^+\subset \Delta. Moreover, each arbitrarily small sector at 0 that contains the positive real axis contains some iterate of J^+ under g. Thus, the same is true for iterates of \phi(J^+) under G (given that \phi is a power map and maps small sectors to small sectors). Thus, by passing to an iterate of J^+ under g, we may assume that \phi(J^+)\subset \mathbb H_{>R}.

Let p_0\in \mathbb{R}\cap \mathbb H_{>R} and p_k=G(p_{k-1}) = G^{\circ k}(p_0), k\in \mathbb{N}. Also, let q_0 \in \phi(J^+) and q_k=G(q_{k-1}) = G^{\circ k}(q_0), k\in \mathbb{N}. For k\in \mathbb{N} denote by J^+(k) the subarc of \phi(J^+) from q_{k-1} to q_k. Consider a ball B such that \overline{B}\subset \mathbb H_{>R} and [p_0,p_1]\cup J^+(1) \subset B. Note that there exists t>1 such that tB\subset \mathbb H_{>R}. By Koebe’s distortion theorem [Pommerenke:conformal]*Theorem 1.3, we have

for all distinct w_1,w_2\in B and k\in \mathbb{N}. Thus, for each w\in J^+(1) and k\in \mathbb{N} we have

By compactness, we have |p_0-w|\simeq 1 for w\in J^+(1), so

for all k\in \mathbb{N} and w\in J^+(1). Also, by (?), we have

for all k\in \mathbb{N}\cup\{0\}. Thus,

for all k\in \mathbb{N} and w\in J^+(1). This implies that \operatorname{Im} (w)\lesssim 1 for all w\in J^+(k), k\in \mathbb{N}.

Next, we prove that \operatorname{Im} (w) is also bounded from below. By (?) and (?), we have |(G^{\circ k})'(p_0)|\simeq 1 for all k\in \mathbb{N}. By Koebe’s distortion theorem [Pommerenke:conformal]*Theorem 1.3 we have |(G^{\circ k})'(w)|\simeq 1 for all k\in \mathbb{N} and for all w\in B. Let r>0 such that B(w,r)\subset \mathbb H_{>R}\cap \mathbb H for each w\in J^+(1). Let w\in J^+(1) be arbitrary. We apply Koebe’s distortion theorem [Pommerenke:conformal]*Corollary 1.4 to the conformal map G^{\circ k}|_{B(w,r)}, which takes values in \mathbb H_{>R}\cap \mathbb H, and we obtain

for every k\in \mathbb{N}. This proves that \operatorname{dist}(J^+(k),\mathbb{R})\gtrsim 1 for each k\in \mathbb{N}. Thus, \operatorname{dist}(\phi(J^+),\mathbb{R})\gtrsim 1, as desired.

Therefore, \phi(J^+) is between two horizontal lines \operatorname{Im}(w)=\tau_1 and \operatorname{Im}(w)=\tau_2, where 0<\tau_1<\tau_2. If \tau\in \mathbb{R} is fixed and \operatorname{Im} (\phi (x+iy))=\tau, note that as x\to \infty we have y=o(x). A direct computation shows that

This proves that there exists L\geq 1 such that

as illustrated in Figure ?. The proof is complete.

By replacing f with f^{-1}, we may suppose that P_1 is an attracting petal corresponding to an attracting vector v_1. For i=1,2 consider a Jordan arc E_{i1}\subset K_i\cap (P_1\cup \{0\}) with an endpoint at 0 such that f(E_{i1})\subset f(E_{i1}). Since f is a homeomorphism in P_1 and each point of P_1 is attracted to 0 under iteration, we conclude that for each z\in E_{i1}\setminus \{0\} the point f(z) lies in the open subarc of E_{i1} from 0 to z.

Let D be the Jordan region whose boundary is K_1 and K_2\subset \overline{D}. By Theorem 2.3, D is a quasidisk. By Proposition 2.5 there exists a quasiconformal map \psi of \widehat \mathbb{C} such that \psi(K_1)=\mathbb{R}\cup \{\infty\}, \psi(0)=0, \psi(D)=\mathbb H, and \psi|_D is conformal. Here \mathbb H denotes the upper half-plane. We conjugate f with \psi to obtain a holomorphic map g in a neighborhood of 0 in \mathbb H. Note that g extends continuously to the real line near 0 and maps it into itself. Thus, by Schwarz reflection we may extend g to a holomorphic map in a neighborhood of 0 that fixes 0.

Without loss of generality, suppose that E_{11} is mapped under \psi to an interval (0,s_0). By the behavior of f in E_{11}, we conclude that 0<g(x)<x for each x\in(0,s_0). This implies that 0\leq g'(0)\leq 1. Note that the petal P_2 for f is either attracting or repelling. By assumption, for i=1,2 there exists an arc E_{i2}\subset K_i\cap (P_2\cup \{0\}) with properties analogous to those of E_{i1} above. Let (-t_0,0)=\psi(E_{12}). We claim that g is a parabolic germ at 0. If P_2 is a repelling petal, then we have g(x)<x for each x\in (-t_0,0), which implies that g'(0)\geq 1. Thus, g'(0)=1. Clearly g is not the identity map, so it is a parabolic germ at 0.

Next, suppose that P_2 is an attracting petal. Fix j\in \{1,2\}. By Lemma 3.5, if S_j is an arbitrary sector at 0 that contains v_j then f^{\circ k}(E_{ij})\subset S_j\cup \{0\} for all sufficiently large k\in \mathbb{N} and for i=1,2. Thus, the arc E_{ij} is asymptotic to the attracting vector v_j. Since f is parabolic, there exists a repulsion vector v that is between the attraction vectors v_1,v_2 and a repelling petal P for the direction v that is disjoint from small sectors S_j, j=1,2, as above. In particular, we may find a sufficiently small repelling petal P as above that is contained in D. That is, each point of P is attracted to 0 under iteration of f^{-1}. This proves that there are points of \mathbb H that are attracted to 0 under iteration of g^{-1}. If |g'(0)|<1, then all points in a neighborhood of 0 would be attracted to 0 under iteration of g by [Milnor:dynamics]*Lemma 8.1, a contradiction. Thus, the only remaining case is that g'(0)=1. Hence, g is again a parabolic germ.

It remains to prove that the conditions of Theorem 3.4 remain true for g. Let F_{11}=(0,s_0)=\psi(E_{11}) and F_{21}=\psi(E_{21}), which satisfy g(F_{i1})\subset F_{i1} for i=1,2 by the dynamics of f. Clearly, each point of F_{11} is attracted to 0 in the direction of the positive real axis. We claim that the same is true for F_{21}. By the dynamics of f, each point of F_{21} is attracted to 0 under iteration of g. By Lemma 3.5 there exists an attraction vector v\in \overline{\mathbb H} for g such that F_{21} is asymptotic to v. If v is not in the direction of the positive real axis, then there exists a repulsion direction between the positive real axis and the vector v. In particular, there exists a repelling petal Q for g in \mathbb H that lies between (in the sense of cyclic order) F_{11} and F_{21}. This implies that there exists a repelling petal \psi^{-1}(Q) for f that is between E_{11} and E_{21}. Note that E_{11} and E_{21} are asymptotic to v_1, in view of Lemma 3.5. Thus, if Q is sufficiently small, then \psi^{-1}(Q) is necessarily contained in a small sector that contains v_1 and that does not contain repulsion vectors of f. So, points of \psi^{-1}(Q) cannot be attracted to 0 in the direction of a repulsion vector of f under iteration of f^{-1}. This is a contradiction. Thus, the claim is proved. Now, if we consider any attracting petal Q_1 for g for the direction of the positive real axis, then by replacing F_{i1} with an iterate, in view of Lemma 3.5, we may have F_{i1}\subset Q_1\cup \{0\} for i=1,2. In the same way we argue for F_{i2}=\psi(E_{i2}), i=1,2.

The map f_1 is locally m-to-1 near 0, so there exists a conformal map \phi in a neighborhood of 0 that fixes 0 and satisfies f_1(z)=\phi(z)^m. Next, we have

Since this map is locally m-to-1 near 0, there exists a conformal map \widetilde h in a neighborhood of 0 that fixes 0 and satisfies f_2(z^m)= \widetilde h(z)^m; there are m choices of \widetilde h, differing by multiplication with an m-th root of unity and we fix one such map. Thus,

for a choice \mu of the m-th root of \lambda. Finally, we set h=\phi^{-1}\circ \widetilde h\circ \phi, which satisfies h'(0)=\mu. Observe that h satisfies the claimed identity f_2\circ f_1=f_1\circ h. This proves ?.

If z\mapsto z^{1/m} is a branch of the m-th root defined in an appropriate open set not containing 0, then f_1^{-1}(z)=\phi^{-1}(z^{1/m}) is a branch of f_1^{-1}. Let g=f_1^{-1}\circ f_2\circ f_1 and \widetilde g= \phi\circ g\circ \phi^{-1}, defined in appropriate open sets. We have

This implies that \widetilde h differs from \widetilde g by multiplication with an m-th root of unity. So, \widetilde g and g may be extended in a neighborhood of 0. Thus, the chosen branch of f_1^{-1} yields a map g satisfying ?, as desired.

Next, suppose that \lambda=\mu=1. We have

We expand h(z)=z+ cz^{l}+O(z^{l+1}), where c\neq 0. For \delta=h(z)- z we have

Moreover,

Thus,

If b\neq 0, it follows that l=nm+1 and \frac{b}{m}\phi'(0)^{nm+1}=\phi'(0)c. If b=0, then l>nm+1. This proves ?.

By [LeiYongcheng:local_connectivity]*Lemma 2.1, \partial V is locally connected. By [TanYin96long]*Corollary 3.3, the point a_0 has finitely many prime ends in V. In particular, there exist finitely many cross cuts bounding disjoint regions Z_1,\dots,Z_k \subset U whose union is a neighborhood Z of a_0 in V. Note that a_0 cannot be a critical point. So, by taking Z sufficiently small, we assume that f is injective in Z. This implies that the regions f(Z_1),\dots,f(Z_k) are disjoint and are in correspondence with the prime ends of a_0; one can see this by conformally conjugating f|_V to a Blaschke product in \mathbb{D}. Therefore, f(Z) is a neighborhood of a_0 in V. Since f is injective near a_0, no point of U near a_0 can be mapped to a point of f(Z). Since f(Z) is a neighborhood of a_0 in V, no point of U near a_0 can be mapped to a point of V. Thus f(U)\neq V and ? is true.

Next, we prove ?. By the above argument, given that a_0,f(a_0) correspond to a unique prime end in V,f(V), respectively, we see that there exists a neighborhood Z of a_0 in V that is mapped to a neighborhood f(Z) of a_0 in f(V). Now we argue as above. Since f is injective near a_0, no point of U near a_0 can be mapped to a point of V near a_0. Thus, f(U)\neq f(V).

We prove ?. Suppose that U is a strictly pre-periodic Fatou component and a_0\in \partial U is a periodic point. By replacing f with an iterate, suppose that a_0 is fixed and V=f(U) is fixed. Since f(\partial U)=\partial V and \partial U is connected, the set \partial V is connected, so V is simply connected. However, this contradicts ?, so ? holds. Now we justify ?. By part ? and the finiteness of periodic Fatou components, each periodic point lies in the boundary of at most finitely many simply connected Fatou components. Thus, each pre-periodic point has the same property.

Let l be the pre-period of a_0 and let p be a multiple of the period of b_0=f^{\circ l}(a_0) so that f^{\circ p} fixes the attracting directions at the parabolic point b_0, i.e., (f^{\circ p})'(b_0)=1. We apply Lemma 4.1 for the holomorphic germs f_1=f^{\circ l} and f_2=f^{\circ p} at a_0. In both alternative assumptions the map f_1=f^{\circ l} is locally m-to-1 at a_0 and the map f_2=f^{\circ p} has a parabolic fixed point at b_0 with multiplicity n+1 for some n,m\in \mathbb{N}. By Lemma 4.1 we obtain a parabolic germ h at a_0 with multiplicity nm+1 such that f_2\circ f_1=f_1\circ h. Under assumption ?, we have m=1 and n=2, so the multiplicity of the parabolic point a_0 of h is 3. Under assumption ?, we have m=2 and n=1, so the multiplicity of the parabolic point 0 of h is again 3.

Under either assumption, we claim that f_1(U) is an immediate basin of the parabolic point b_0. If not, then f_1(U) is disjoint from all attracting petals at b_0. By Lemma 3.2, any arc of f_1(\partial U) that contains b_0 in its interior is not a quasiarc and \partial U\cap W is not a quasiarc, a contradiction. Similarly, f_1(V) is an immediate basin of b_0.

Suppose ? holds. The map f_1 is a local homeomorphism near a_0 and we have h=f_1^{-1}\circ f_2\circ f_1, where f_1^{-1} is a local inverse branch of f_1 that maps b_0 to a_0. Given that a_0 is not a critical point of f_1, by Lemma 4.2 ? the sets f_1(U) and f_1(V) are distinct Fatou components each contained in one of the two parabolic basins of the map f_2 for the point b_0. By Lemma 3.3 each of the boundaries f_1(\partial U) and f_1(\partial V) contains two non-overlapping arcs with a common endpoint at b_0, contained in opposite repelling petals of f_2, and each of these arcs is mapped into itself under a branch of f_2^{-1}. It follows that each of \partial U,\partial V contains two non-overlapping arcs with a common endpoint at a_0 contained in opposite repelling petals of h, and each of these arcs is mapped into itself under h^{-1}. By Theorem 2.4, we can extend the quasiarcs \partial U\cap \overline{W} and \partial V\cap \overline{W} to quasicircles K_1,K_2 as in Theorem 3.4. Therefore, by Theorem 3.4, there exists a quasiconformal map of \widehat \mathbb{C} that maps \partial U\cap \overline{W} and \partial V\cap \overline{W} onto two circular arcs, as desired.

Next, we assume that ? holds. Then f_1 is locally 2-to-1 at a_0. Consider a small repelling petal P of f_2 at b_0 that is a Jordan region. By Lemma 3.3, f_1(\partial U)\cap P contains two non-overlapping arcs with an endpoint at b_0 and each of these arcs is mapped into itself under a branch of f_2^{-1}. The set f_1^{-1}(P) has two components P_1,P_2, and in the closure of each of them f_1 topologically conjugates h to f_2. Thus, P_1,P_2 are repelling petals of h. Moreover, each of \partial U,\partial V contains two non-overlapping arcs with a common endpoint at a_0, contained in opposite repelling petals of h, and each of these arcs is mapped into itself under h^{-1}. Therefore, we are in the setting of Theorem 3.4 as before.

The lower bound follows from the Schwarz lemma; see [BeardonMinda:hyperbolic]*Theorem 8.1. To prove the upper bound, we use the estimate (see [BeardonMinda:hyperbolic]*Theorem 8.2)

We write \partial Q = J \cup \Gamma, where \Gamma = \overline{\partial Q \cap \mathbb{H}}. Let Q' be the double of Q along J and V' be the double of V along I. Let A = Q' \setminus \overline{V'}, which is an annulus. Since \operatorname{Mod} (Q\setminus \overline{V})\geq K^{-1}, we have

by the discussion in Section ?. Let z_0 \in V and w_0 \in \partial Q' so that

Note that A separates z_0, \overline{z_0} from w_0, \infty. By Teichmüller’s module theorem [LehtoVirtanen:quasiconformal]*Section II.1.3, there is an increasing function \Psi\colon (0,\infty) \to (0,\infty) so that

Therefore, there exists \delta = \delta(K)\geq 1 so that \operatorname{dist}(z_0, \Gamma) \geq \delta^{-1}\operatorname{Im}(z_0). Also, note that \operatorname{dist}(z_0,\partial Q)=\operatorname{dist}(z_0,\Gamma) or \operatorname{dist}(z_0,\partial Q)=\operatorname{Im}(z_0). Thus, by (Proof),

This proves the upper bound.

By Lemma 4.2 ?, there are finitely many simply connected Fatou components \Omega whose closures contain critical points. We choose k_0\geq 1 so that whenever a critical point of f lies in the closure of a simply connected Fatou component \Omega, the pre-period l_\Omega of \Omega is at most k_0. There are finitely many Fatou components U with l_U< k_0. Thus, if U is a quasidisk, then it is an L-quasidisk for some uniform L\geq 1.

Next, we suppose that l_U\geq k_0. Let V=f^{\circ m}(U) for m=l_U-k_0\geq 0, so l_V=k_0. By assumption, V is a Jordan region. We will show that U is a uniform quasidisk, depending only on V, but not on m. This is sufficient, since there are finitely many Fatou components V with l_V=k_0. Since V is not periodic, \overline{V} does not contain any parabolic periodic points by Lemma 4.2 ?. Thus, \overline{V} is disjoint from the accumulation points of the post-critical set. By the choice of k_0, \overline{V} has a positive distance from the post-critical set. Consider a Jordan region N\supset \overline{V} that is disjoint from the post-critical set of f.

The map g=f^{\circ m}\colon U\to V is conformal since V is simply connected and g has no critical points. Moreover, g^{-1} extends conformally to the neighborhood N of \overline{V}. Since U is a quasidisk, \partial U is a quasicircle by Theorem 2.3. The fact that g is a diffeomorphism implies that \partial V also satisfies the Ahlfors condition and is a quasicircle. There are only finitely many Fatou components of pre-period k_0, so \partial V is a uniform quasicircle. By Lemma 2.2, the map g^{-1}\colon (\partial V, \chi)\to (\partial U, \chi) is uniformly quasi-Möbius. Thus, \partial U is a uniform quasicircle, being the quasi-Möbius image of a quasicircle. Hence, U is a uniform quasidisk.

By Proposition 6.1, we have \operatorname{Mod} A(\overline{U},\overline{V})\geq L_1^{-1} for some L_1=L_1(f)\geq 1. Also, U and V are L_2-quasidisks for some L_2=L_2(f)\geq 1 by Proposition 5.1. Finally, by Proposition 6.2 there exists a K(L_1,L_2)-quasiconformal map of \widehat \mathbb{C} that maps U and V to disks.

We first prove the sufficiency. Suppose that U_i is a quasidisk for each i\in I, as in ?. By Theorem 2.11, it suffices to prove that for every pair of distinct i,j there exists a uniformly quasiconformal homeomorphism of \widehat \mathbb{C} that maps U_i and U_j to disks. If \overline{U_i}\cap \overline{U_j}=\emptyset, then the conclusion follows from Corollary 6.3. Suppose that \overline{U_i}\cap \overline{U_j}=\{a_0\}. By the assumptions, combined with Proposition 4.3 and Lemma 2.8, we see that the pair U_i,U_j is quasiconformally circularizable. By Proposition 4.5, the pair U_i, U_j is uniformly quasiconformally circularizable, as desired. The last uniqueness statement follows from the uniqueness part of Theorem 2.11.

We now prove the necessity. Suppose that there exists a quasiconformal map g of \widehat \mathbb{C} such that g(U_i) is a disk for each i\in I, as in ?. It is obviously necessary that U_i is a quasidisk for each i\in I. Also, if \overline{U_i}\cap \overline{U_j}\neq \emptyset for some i\neq j, then \overline{U_i}\cap \overline{U_j} consists of a single point a_0, given that the Julia set is not a Jordan curve.

By Lemma 7.3, a_0 is pre-periodic. Let b_0 be a periodic point in the orbit of a_0. Observe that the point b_0 must be repelling or parabolic; indeed, there are no Cremer points for geometrically finite rational maps [Milnor:dynamics]*Theorem 11.17. Let l be the pre-period of a_0 and p be a multiple of the period of b_0=f^{\circ l}(a_0) so that f^{\circ p} fixes all simply connected Fatou components that contain b_0 in their boundary; there are finitely many such Fatou components by Lemma 4.2 ?. Define f_1=f^{\circ l}, f_2=f^{\circ p}. By Lemma 4.1 there exists a parabolic germ h at a_0 such that f_2\circ f_1=f_1\circ h.

Suppose that b_0 is repelling for f_2, so by Lemma 4.1, a_0 is repelling for h. Given that f_1(U_i) and f_1(U_j) are Jordan regions that are fixed under f_2, we see that f_2 locally preserves their boundaries near b_0. Suppose first that f_1=f^{\circ l} is locally injective near a_0. Then we see directly that h locally preserves \partial U_i and \partial U_j near a_0. Now, suppose that f_1 has a critical point at a_0. If f_1^{-1} is a branch of the inverse of f_1 that maps a neighborhood of b_0 in f_1(U_i) onto a neighborhood of a_0 in U_i, then we may choose h so that h=f_1^{-1}\circ f_2\circ f_1; see Lemma 4.1 ?. Thus, we may have that h locally preserves \partial U_i near a_0. By Lemma 7.2, \partial U_j is also locally invariant under h near a_0. In both cases, by applying Lemma 3.1 we see that no quasiconformal map of the sphere can map U_i and U_j to two disks, a contradiction.

Therefore, b_0 is a parabolic point for f_2. Let m\geq1 be the local degree of f_1 at a_0 and n+1\geq 2 be the parabolic multiplicity of the point b_0 for the map f_2. Then the parabolic multiplicity of a_0 for the map h is nm+1 by Lemma 4.1 ?. We have nm+1=2 if and only if f_1 is locally injective near a_0 and the multiplicity of b_0 for the map f_2 is 2. In this case, any Fatou component that contains b_0 in its boundary, including the immediate basin, has a cusp; see Lemma 3.2. Therefore, the same is true for a_0 by the local injectivity of f_1. Thus, U_i and U_j are not quasidisks, a contradiction. We conclude that the parabolic multiplicity of a_0 for h is strictly greater than 2.

By Lemma 3.2 and since U_i,U_j are quasidisks, we see that f_1(U_i) and f_1(U_j) are immediate basins of b_0, not necessarily distinct. Note that in a neighborhood of a_0 in each of the sets U_i and U_j the map f_1 conjugates h to f_2. Thus, U_i,U_j contain attracting petals P_i,P_j, respectively, of h at a_0. In addition, points of U_i,U_j near a_0 are absorbed under iteration of h by P_i,P_j, respectively. If nm+1\geq 4, then h has at least 3 attracting directions, so there exists a third attracting direction and a corresponding petal P that is disjoint from U_i and U_j and can be taken to be a quasidisk. By assumption, there exists a quasiconformal map of the sphere that sends U_i and U_j to tangent disks. This contradicts Lemma 2.10. Therefore, nm+1 = 3. This implies that m = 2 and n=1 or m=1 and n=2, as required in the two alternatives of Theorem 7.1 ?.

The arc E has d preimages E_1,\dots,E_d under \phi. Each E_i is a Jordan arc with endpoints at 0 and at some point z_i \in \partial \mathbb{D}. The arcs are disjoint except at 0, which is their common endpoint. We assume that the points z_i\in \partial \mathbb{D}, i\in \{1,\dots,d\}, have the counter-clockwise order. Note that h acts transitively on these arcs in the sense that there exists a neighborhood U of 0 such that for each i\in \{1,\dots,d\} there exists a unique j\in \{1,\dots,d\} such that h(E_j\cap U)\subset E_i. Hence, h induces a bijection \tilde h\colon \mathbb{Z}_d\to \mathbb{Z}_d with the property that \tilde h(j)=i if h(E_j\cap U)\subset E_i. Since h is an orientation-preserving homeomorphism, the map \tilde h must be a translation. By assumption, a lift of E is locally invariant under h near 0. This implies that \tilde h has a fixed point, so it is the identity map. Thus, every lift of E is locally invariant under h.

Next, consider an arc F satisfying ? such that E\cap F=\{0\}. Then F has d lifts each of which is between two lifts of E and h acts transitively on the lifts of F. Since each lift of E is locally invariant near 0 under h, and h is an orientation-preserving homeomorphism, we see that each lift of F is also locally invariant near 0 under h.

By conjugating f with a Möbius transformation, we assume that \infty lies in the Fatou set. Let a_0\in \overline{U_1}\cap \overline{U_2} be an isolated point. If the orbit of a_0 contains critical points or parabolic points, then it is pre-periodic since f is geometrically finite. Suppose that the orbit of a_0 does not contain critical points or parabolic points. By Sullivan’s theorem, the components U_1,U_2 are eventually periodic. By passing to an iterate, we assume that f(U_1) and f(U_2) are fixed Fatou components, not necessarily distinct. Then by [McMullen:Hausdorff_GF]*Theorem 6.5 the point a_0 lies in the radial Julia set of f. By definition, this means that there exists r>0 such that for every \varepsilon>0 there exists a neighborhood \Omega of a_0 and n\in \mathbb{N} such that \operatorname{diam}(\Omega)<\varepsilon and the map

is conformal. By assumption, if \Omega is sufficiently small, then a_0 is the unique point of \overline{U_1}\cap \overline{U_2} in \Omega. Also, by shrinking r we may assume that \operatorname{diam} f(U_i)>2r, so f(U_i) is not contained in B(f^{\circ n}(a_0),r) for i=1,2.

By [LeiYongcheng:local_connectivity]*Lemma 2.1, U_1,U_2 and f(U_1),f(U_2) have locally connected boundary. Let I be the infinite set of indices n\in \mathbb{N} such that there exists a neighborhood \Omega_n of a_0 with \Omega_n\cap \overline{U_1}\cap \overline{U_2}=\{a_0\} and the map f_n=f^{\circ n}\colon \Omega_n\to B(a_n,r) is conformal, where a_n=f^{\circ n}(a_0). Consider conformal maps \phi_i\colon \mathbb{D}\to U_i, i=1,2, which extend continuously to the boundaries by Carathéodory’s theorem. For n\in I and i=1,2, let P_{i,n}\subset \partial \mathbb{D} be the union of the components of \phi_i^{-1}(\Omega_n\cap \partial U_i) that intersect \phi_i^{-1}(a_0). Also, let \psi_i\colon \mathbb{D}\to f(U_i), i=1,2, be conformal maps. Let i\in \{1,2\}. Consider the conformal map g_i=\psi_i^{-1}\circ f_n\circ \phi_i defined in a neighborhood of P_{i,n} in \mathbb{D}. Note that g_i extends continuously to P_{i,n} and let Q_{i,n}=g(P_{i,n}). Each component of Q_{i,n} intersects \psi_i^{-1}(a_n) and is mapped by \psi_i to a curve that passes through a_n and has endpoints on \partial B(a_n,r); see Figure ?. By the uniform continuity of \psi_i in \overline{\mathbb{D}}, there exists a uniform constant \delta>0 such that for J_{i,n}=\psi_i^{-1}(a_n)\cap Q_{i,n} we have

If a point b\in \partial f(U_i) satisfies \psi_i^{-1}(b) \cap Q_{i,n}\neq \emptyset for both i=1,2, then \phi_i^{-1}(f_n^{-1}(b))\cap P_{i,n}\neq \emptyset, so f_n^{-1}(b)\cap \overline{U_i}\neq \emptyset for both i=1,2. Thus, a_0=f_n^{-1}(b) and a_n=b. We conclude that a_n is the unique point of \partial f(U_i) that satisfies \psi_i^{-1}(a_n) \cap Q_{i,n}\neq \emptyset for both i=1,2. In particular, if a_m\neq a_n, then for some i\in \{1,2\} we have \psi_i^{-1}(a_m)\subset \partial \mathbb{D}\setminus Q_{i,n}, so by (?) we have

Consider the space X of compact subsets of \partial \mathbb{D}\times \partial \mathbb{D} equipped with the Hausdorff metric. Then for n,m\in I with a_n\neq a_m, any two points

have distance at least \delta. Since X is compact, we conclude that the set \{a_n:n\in I\} is finite. Thus, there exist n\neq m such that f^{\circ n}(a_0)=f^{\circ m}(a_0) and a_0 is pre-periodic, as desired.

Since f is a geometrically finite polynomial, all bounded Fatou components are Jordan regions [RoeschYin:boundary_fatou]. Let U,V be bounded Fatou components such that \overline{U}\cap \overline{V}\neq \emptyset. Since f is a polynomial, the Julia set is the boundary of the basin \mathcal A(\infty), so \overline{U}\cap \overline{V} consists of a single point a_0, which is not critical. Since there are no critical points on the Julia set, for each n\in \mathbb{N} the components f^{\circ n}(U) and f^{\circ n}(V) are distinct and touch at f^{\circ n}(a_0); see Lemma 4.2 ?. Since U and V are pre-periodic, the point a_0 is pre-periodic. Thus, by passing to iterates, it suffices to show that every periodic contact point between two periodic Fatou components is parabolic.

Let \{U_1,\dots, U_k\} be the collection of bounded periodic Fatou components, and let \{a_1,\dots, a_l\} be the collection of all contact points between U_i and U_j, i\neq j. We show that \{a_1,\dots,a_l\} is exactly the parabolic cycle. Since f is unicritical, \{U_1,\dots, U_k\} form a single orbit; see [Milnor:dynamics]*Theorem 10.15. In particular, f permutes the collection \{U_1,\dots, U_k\} and f^{\circ k}(U_i)=U_i for each i\in \{1,\dots,k\}. Moreover, f maps each contact point to a contact point and f^{\circ k} fixes contact points. Therefore f is a bijection on \{a_1,\dots,a_l\}.

We consider a combinatorial bipartite graph \Gamma with two types of vertices V_1, V_2: vertices in V_1 are in correspondence with components in \{U_1,\dots,U_k\} and vertices in V_2 are in correspondence with contact points \{a_1,\dots, a_l\}. A vertex v\in V_1 is connected to w\in V_2 if the corresponding contact point for w is on the boundary of the component for v. By construction, \Gamma is a finite union of trees, and the map f induces a natural graph isomorphism f_*\colon \Gamma \rightarrow \Gamma. By the above, f_* is a permutation on V_1.

Let T be a component of \Gamma. We claim that T contains exactly one vertex in V_2 that corresponds to a point in the parabolic periodic cycle. Indeed, let A = V_1 \cap T and B = V_2 \cap T. Since each vertex in V_2 has valence at least 2, every endpoint of T is in V_1. Since f_* acts transitively on V_1, all vertices of A have the same valence. Thus A is exactly the set of endpoints of T. Since T is bipartite, there is a unique vertex b\in B which connects to every vertex in A, i.e., T is a star. Since the components corresponding to the vertices of A are parabolic and the parabolic multiplicity is at least 3, the contact point corresponding to b is parabolic.

Let us sketch the proof of the lemma, since it is similar to the proof of [LuoNtalampekos:piecewise_qc]*Lemma 7.1. After appropriate normalization, we assume that \overline{U} \subset\mathbb{C}.

We first claim that there exists a continuum Q \subset \overline{U} so that

Let Q_0 \subset \overline{U} be a closed attracting petal if U is parabolic or a small forward invariant disk about its attracting fixed point if U is (super-)attracting. Choose arcs E_x \subset \overline{U} landing at the vertices of x\in \mathcal{A} that are dynamically compatible in the sense that

• if x is the parabolic fixed point, then E_x \subset Q_0 and f(E_x) \subset E_x and

• otherwise, E_{f(x)} \subset f(E_x).

This is possible by considering the Koenigs coordinate at repelling periodic points and taking pull-backs at pre-periodic points. Connect the interior endpoints of these arcs to Q_0 by compact arcs in U. Let T be their union with Q_0. Compatibility near the vertices ensures that f(T)\setminus T is contained in a compact set C\subset U. Since Q_0 attracts compact subsets of U, all sufficiently high forward images of C lie in Q_0. Therefore, the set

is a compact forward-invariant continuum and Q\cap\partial U=\mathcal A. This completes the proof of the claim.

Let X\coloneqq \widehat\mathbb{C} \setminus (Q \cup P(f)) and Y \coloneqq \widehat\mathbb{C} \setminus f^{-1}(Q \cup P(f)). For \epsilon>0 and a component \gamma_i of \partial U\setminus \mathcal A we set

Recall that h_X denotes the hyperbolic metric on X. We claim that for sufficiently small \epsilon, N_{i, \epsilon} is contained in some simply connected domain H_i \subset X such that the domains in \{H_i\} are pairwise disjoint.

Since Q is a non-degenerate continuum through each vertex v \in \mathcal{A}, by comparison and Koebe’s estimate (see [BeardonMinda:hyperbolic]*Section 8), we have

We construct H_i as follows. Let v be an endpoint of \gamma_i. We first construct two arcs landing at v that locally bound a sector containing \gamma_i such that the h_X-distance of the two arcs to \gamma_i is bounded from below. If v is a repelling periodic point, construct the two arcs of H_i in a Koenigs fundamental annulus and iterate them by the inverse return map. By compactness, there is a uniform lower bound for the distance between the two arcs in the metric |dz|/|z-v|. So, in view of (?) the distance between each of the two arcs and \gamma_i is bounded from below in the h_X-metric. If v is a parabolic fixed point, the existence of the arcs follows from the construction given in [LuoNtalampekos:piecewise_qc]*Lemma 7.1; see [LuoNtalampekos:piecewise_qc]*Figure 7.1. If v is a strictly pre-periodic point, then the two arcs are constructed by pull-backs. We can connect the arcs at the two endpoints of \gamma_i to obtain a closed curve \alpha which bounds a simply connected domain H_i \subset X and the h_X-distance between \alpha and \gamma_i is bounded from below. By choosing disjoint sectors at shared vertices, we may construct the regions \{H_i\} so that they are pairwise disjoint. The claim follows.

We choose \epsilon sufficiently small so that N_i\coloneqq N_{i,\epsilon} is contained in H_i for each i. Suppose that f(I_j) = \gamma_i. Let D_j be the component of f^{-1}(N_i) that contains I_j. Since H_i is simply connected and avoids critical values, the map f\colon D_j\rightarrow N_i is conformal. Since the neighborhoods \{N_i\} are pairwise disjoint, the sets \{D_j\} have the same property. Suppose that I_j \subset \gamma_{k}. Since Q \cup P(f) is forward invariant, the map f\colon Y \rightarrow X is a covering and the inclusion map \iota\colon Y \rightarrow X is distance-decreasing by the Schwarz–Pick lemma. Thus,

This completes the proof.

We adapt the construction given in the proof of Lemma A.3. As in Lemma A.3, construct a continuum Q\subset\overline V such that

We set

Forward invariance of Q and P(f) gives that Z\subset Y, and that the maps

are coverings. We equip these surfaces with their hyperbolic metrics. For \epsilon>0, let

We claim that for all sufficiently small \epsilon, the neighborhood N_{\epsilon, Y}(\gamma_i) (resp. N_{\epsilon, Z}(I_j)) is contained in a simply connected domain H_i \subset Y (resp. L_j \subset Z) such that the sets \{H_i\} (resp. \{L_j\}) are pairwise disjoint. The construction is the same as in Lemma A.3 away from vertices eventually mapped to parabolic points. If a vertex b is eventually mapped to a, let k be the local degree of an iterate taking b to a. By hypothesis k\ge2. Let \gamma^\pm be the two arcs from the family \{\gamma_i\} with an endpoint b. Then \gamma^\pm are separated by a positive angle at b, so we can choose local sectors bounded by two arcs with h_Y-distance to \gamma^{\pm} bounded from below. Indeed, in suitable local coordinates, f takes the form z\mapsto z^k, so the preimage of V near b consists of k components whose closures meet at b. Since V has a cusp at a, the two boundary arcs of each component form an interior angle of 2\pi/k at b. This proves the claim.

Set N_i=N_{\epsilon,Y}(\gamma_i) and D_j = N_{\epsilon, Z}(I_j) for sufficiently small \epsilon so that the previous claim holds. Then the sets \{N_i\} are pairwise disjoint and so are the sets \{D_j\}. Suppose that S(I_j)=\gamma_i. Since H_i and L_j are simply connected, we may analytically extend S\colon I_j \rightarrow \gamma_i and T \coloneqq S^{-1}\colon \gamma_i \rightarrow I_j to maps

By the invariance of the hyperbolic density under the coverings f and f^{\circ 2}, we conclude that S, T preserve hyperbolic lengths measured in Z, Y. Let z\in D_j=N_{\epsilon,Z}(I_j). Choose a rectifiable path in Z from a point of I_j to z with hyperbolic length less than \epsilon. Every point of this path lies in D_j\subset L_j, so its image under S starts on \gamma_i and has the same length. Hence S(z)\in N_i, proving S(D_j)\subseteq N_i. The same argument shows that T(N_i) \subseteq D_j. Since S\circ T, T \circ S are identity on \gamma_i and I_j respectively, we conclude that S(D_j) = N_i and the map S\colon D_j \rightarrow N_i is conformal. Finally, if I_j\subset\gamma_k, the Schwarz–Pick lemma for the inclusion Z\subset Y gives

This completes the proof.

Since f(U) is a quasidisk, there exists a quasiconformal map \phi of \widehat \mathbb{C} that maps f(U) onto a disk. By the measurable Riemann mapping theorem [AstalaIwaniecMartin:quasiconformal]*Theorem 5.3.4, there exists a quasiconformal map \psi of \widehat \mathbb{C} whose Beltrami coefficient is the same as that of \phi\circ f. Then the map \phi\circ f\circ \psi^{-1} is rational (see [AstalaIwaniecMartin:quasiconformal]*Theorem 5.5.1) and maps the Jordan region \psi(U) onto a disk.

Therefore, by Theorem 2.3, it suffices to prove that if g is a rational map and J is a Jordan curve that is mapped onto a circle, then J is a quasicircle. For this, it suffices to prove that the Ahlfors condition (?) holds near critical points of g on J. By precomposing g with a conformal map near a critical point z_0\in J and post-composing with a Möbius transformation, we may assume that g(z)=(z-z_0)^d in a neighborhood of z_0 for some d\geq 2 and g maps an arc A of J that passes through z_0 onto a line segment passing through 0. We observe that A is the union of two non-overlapping segments with a common endpoint at 0. Thus, the Ahlfors condition holds trivially on the arc A.