This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Quasi-self-similar fractals containing “Y” have dimension larger than one

Insung Park The Institute for Computational and Experimental Research in Mathematics, Providence, RI, 02903 USA insung_park@brown.edu  and  Angela Wu Department of Mathematics, Indiana University, Bloomington, IN 47405 USA angelawu0312@gmail.com
Abstract.

Suppose XX is a compact connected metric space and f:XXf:X\to X is a metric coarse expanding conformal map in the sense of Haïssinsky-Pilgrim. We show that if XX contains a homeomorphic copy of the letter “Y”, then the Hausdorff dimension of XX is greater than one. As an application, we show that for a semi-hyperbolic rational map ff its Julia set 𝒥f\mathcal{J}_{f} is quasi-symmetric equivalent to a space having Hausdorff dimension 1 if and only if 𝒥f\mathcal{J}_{f} is homeomorphic to a circle or a closed interval.

1. Introduction

In the study of dynamical systems for rational maps on the Riemann sphere, there are fractals called Julia sets on which the dynamics is more chaotic than on the complements, called Fatou sets. If a Julia set is connected, then its Hausdorff dimension is often greater than one, see [Zdu90], [Prz06], [PZ21]. In this vein, we also discuss a family of Julia sets whose Hausdorff dimensions are greater than one. In fact, we obtain a stronger assertion in the following sense: Any quasi-symmetric deformations of those Julia sets still have Hausdorff dimension greater than one.

Let XX denote a compact connected metric space and f:XXf:X\to X a continuous map, so that the iterations of ff define a topological dynamical system on XX. For e1=(1,0,0),e2=(0,1,0),e_{1}=(1,0,0),e_{2}=(0,1,0), and e3=(0,0,1)e_{3}=(0,0,1), let

𝕐=[0,e1][0,e2][0,e3]3\mathbb{Y}=[0,e_{1}]\cup[0,e_{2}]\cup[0,e_{3}]\subset\mathbb{R}^{3}

denote the metric space which is the union of three unit Euclidean segments meeting at the point o=(0,0,0)o=(0,0,0), equipped with its length metric. If XX contains a homeomorphic copy of 𝕐\mathbb{Y}, we write 𝕐X\mathbb{Y}\hookrightarrow X.

The following is our main theorem. We denote by Hdim(X)\operatorname{Hdim}(X) the Hausdorff dimension of XX.

Theorem A.

Let XX be a compact connected locally connected metrizable space. Suppose f:XXf:X\to X is a topological coarse expanding conformal dynamical system. If 𝕐X\mathbb{Y}\hookrightarrow X, then Hdim(X,d)>1\operatorname{Hdim}(X,d)>1 for any metric dd in the canonical gauge 𝒢(f)\mathcal{G}(f).

From Theorem A we have the following corollary.

Corollary B.

Let XX and ff be as in Theorem A. If 𝕐X\mathbb{Y}\hookrightarrow X and Cdim((X,𝒢(f)))=1\operatorname{Cdim}((X,\mathcal{G}(f)))=1 then the conformal dimension is not attained by any metric in 𝒢(f)\mathcal{G}(f).

Let us define the terminologies sued in the statement of Theorem A.

Quasi-symmetries and conformal dimensions

Let (X,dX)(X,d_{X}) and (Z,dZ)(Z,d_{Z}) be compact metric spaces. A homeomorphism h:XZh:X\to Z is a quasi-symmetry if there is a homeomorphism η:[0,)[0,)\eta:[0,\infty)\to[0,\infty) such that for every x,y,zXx,y,z\in X with xzx\neq z, we have

dZ(h(x),h(y))dZ(h(x),h(z))η(dX(x,y)dX(x,z)).\frac{d_{Z}(h(x),h(y))}{d_{Z}(h(x),h(z))}\leq\eta\left(\frac{d_{X}(x,y)}{d_{X}(x,z)}\right).

One can think of quasi-symmetries as homeomorphisms that do not distort shapes much.

We say that two metric spaces XX and ZZ are quasi-symmetrically equivalent, or simply quasi-symmetric, and write XqsZX\sim_{\mathrm{qs}}Z, if there is a quasi-symmetric homeomorphism between XX and ZZ. Quasi-symmetries define an equivalence relation on the collection of metric spaces.

We often consider different metrics on the same topological space MM. In this case, we say that two metrics d1d_{1} and d2d_{2} on MM are quasi-symmetric if the identity map id:(M,d1)(M,d2)\mathrm{id}:(M,d_{1})\to(M,d_{2}) is a quasi-symmetry.

For a compact metric space XX, the conformal dimension of XX, denoted by Cdim(X)\operatorname{Cdim}(X), is defined as the infimal Hausdorff dimension in the quasi-symmetric equivalence class containing XX, i.e.,

Cdim(X)=inf{Hdim(Z):XqsZ}.\operatorname{Cdim}(X)=\inf\{\operatorname{Hdim}(Z):X\sim_{\mathrm{qs}}Z\}.

We say that XX attains its conformal dimension if the infimum in the definition of the conformal dimension of XX is realized as the minimum.

Coarse expanding conformal dynamical systems

We abbreviate coarse expanding conformal as “cxc”. The idea of cxc maps were introduced by P. Haïssinsky and K. Pilgrim in [HP09]. We give precise definitions in Section 2.

We say that a topological dynamical system f:XXf:X\to X on a topological space XX is topological cxc if it satisfies 3 properties (Expansion), (Degree), and (Irreducibility): (Expansion) and (Irreducibility) are standard conditions for topological dynamical systems. The property (Degree) prohibits periodic and more generally recurrent periodic branch points, which characterizes semi-hyperbolic rational maps in complex dynamics. We remark that we do not use a metric to define topological cxc maps.

Now let us assume that XX is equipped with a metric dXd_{X}. We say that f:XXf:X\to X is metric cxc if (1) ff is topological cxc, and (2) ff satisfies metric conditions defined in such a way that the Sullivan’s Principle of the Conformal Elevator holds. It guarantees that the metric space (X,dX)(X,d_{X}) is quasi-self-similar in a sense that any arbitrarily small piece of XX looks similar to a large piece of XX up to some bounded error. More precisely, a metric space XX is called quasi-self-similar if there is a k1k\geq 1 and r0>0r_{0}>0 such that for any ball BXB\subset X with radius r<r0r<r_{0} there exists a map EB:BXE_{B}:B\to X satisfying

1kr0rdX(x,y)dX(EB(x),EB(y))kr0rdX(x,y)\frac{1}{k}\cdot\frac{r_{0}}{r}\cdot d_{X}(x,y)\leq d_{X}(E_{B}(x),E_{B}(y))\leq k\cdot\frac{r_{0}}{r}\cdot d_{X}(x,y)

for every x,yBx,y\in B [Sul82, p. 42].

Although the definition of cxc maps takes motivation in complex dynamics, cxc maps also include self-maps of manifolds whose iterates are uniformly quasi-regular. Being topological or metric cxc is preserved by quasi-symmetric conjugacies and taking products. See [HP12] for more examples of cxc maps.

Canonical gauge and visual metrics of topological cxc maps

Let f:XXf:X\to X be a topological cxc map. If ff is metric cxc for two different metrics dd and dd^{\prime} on XX, then (X,d)(X,d) and (X,d)(X,d^{\prime}) are quasi-symmetric. See Proposition 2.4. Hence, there exists a unique quasi-symmetric class 𝒢(f)\mathcal{G}(f), called the canonical gauge of ff, consisting of metrics for which ff is metric cxc. In this sense, the conformal dimension for 𝒢(f)\mathcal{G}(f), i.e., the infimal Hausdorff dimension of metrics in 𝒢(f)\mathcal{G}(f), is a natural invariant for topological cxc maps.

For a topological cxc map f:XXf:X\to X, we can construct metrics, called visual metrics, which are contained in 𝒢(f)\mathcal{G}(f). One advantage of having visual metrics is the following: Equipping XX with a visual metric, we may assume that the nn-th preimages of a ball are quasi-balls with uniformly bounded distortion whose radii decay exponentially fast about n>0n>0.

Obstruction for hdim \leq 1

J. Azzam proved that if a compact connected metric space XX is antenna-like, i.e., having “sufficiently many” copies of 𝕐\mathbb{Y} with uniformly bounded distortion, then Hdim(X)>1\operatorname{Hdim}(X)>1 [Azz15]. See Section 2.3 for precise definitions.

Since being antenna-like is preserved by quasi-symmetries, to show Theorem A it suffices to prove that XX with a visual metric is antenna-like. Starting from one copy of 𝕐\mathbb{Y} in XX, we first find many other copies of 𝕐\mathbb{Y} by using properties of topological cxc maps, and then we show these copies are not distorted much by using the properties of metric cxc maps.

Families of rational maps

In complex dynamics with one variable, we consider rational maps f:^^f:\hat{\mathbb{C}}\to\hat{\mathbb{C}}, which are holomorphic maps from the Riemann sphere ^\hat{\mathbb{C}} to itself. In this article, we discuss three sub-families of rational maps. For a rational map ff,

  • ff is semi-hyperbolic if ff does not have any parabolic periodic point nor any recurrent critical point;

  • ff is sub-hyperbolic if any critical point is periodic, preperiodic, or attracted to an attracting periodic cycle; and

  • ff is post-critically finite if any critical point is periodic or preperiodic.

Post-critically finiteness implies sub-hyperbolicity, and sub-hyperbolicity implies semi-hyperbolicity. Every Fatou component of a semi-hyperbolic rational map is in an attracting or a super-attracting basin [Mañ93]. Any sub-hyperbolic rational map can be obtained from a unique post-critically finite rational map by quasi-conformal deformation on the Fatou set, which deforms each super-attracting basin to an attracting basin [McM88].

Attainment of (Ahlfors-regular) conformal dimensions

For applications of conformal dimensions to geometric group theory or complex dynamics, we use a slight variant of the conformal dimension, called the Ahlfors-regular conformal dimension. A metric space XX is Ahlfors-regular if round sets of XX well behave with the δ\delta-Hausdorff measure δ\mathcal{H}^{\delta} where δ=Hdim(X)\delta=\operatorname{Hdim}(X), in a sense that there exists C>1C>1 such that for any 0<r<12diam(X)0<r<\frac{1}{2}\cdot\operatorname{diam}(X) and xXx\in X we have

1Crδδ(B(x,r))Crδ.\frac{1}{C}\cdot{r^{\delta}}\leq\mathcal{H}^{\delta}(B(x,r))\leq C\cdot{r^{\delta}}.

For a compact metric space XX, the Ahlfors-regular conformal dimension, denoted by ARConfdim(X)\operatorname{ARConfdim}(X), is defined by

ARConfdim(X)=inf{Hdim(Z):XqsZ,Zis Ahlfors-regular}\operatorname{ARConfdim}(X)=\inf\{\operatorname{Hdim}(Z):X\sim_{\mathrm{qs}}Z,Z~{}\text{is~{}Ahlfors-regular}\}

M. Bonk and B. Kleiner showed that the attainment of conformal dimensions characterizes lattices of hyperbolic isometry groups.

Theorem 1.1 (Bonk-Kleiner [BK02]).

Suppose that GG is a hyperbolic group and the boundary at infinity G\partial_{\infty}G has topological dimension nn. If ARConfdim(G)=n\operatorname{ARConfdim}(\partial_{\infty}G)=n and attained, then GG is (up to finite index) the fundamental group of a closed hyperbolic nn-manifold.

A similar result for self-maps of the 2-sphere was established by P. Haïssinsky and K. Pilgrim.

Theorem 1.2 (Haïssinsky-Pilgrim [HP14]).

Suppose f:XXf:X\to X is a metric cxc map for a metric space XX homeomorphic to the 2-sphere. If ARConfdim(X)=Hdim(X)\operatorname{ARConfdim}(X)=\operatorname{Hdim}(X), then ff is topologically conjugate to (1) a semi-hyperbolic rational map if ARConfdim(X)=2\operatorname{ARConfdim}(X)=2 or (2) a Latté’s map (with additional properties) if ARConfdim(X)>2\operatorname{ARConfdim}(X)>2.

A semi-hyperbolic rational map ff acts on its Julia set 𝒥f\mathcal{J}_{f} as a metric cxc system. If the semi-hyperbolic rational map is a polynomial and the Julia set 𝒥f\mathcal{J}_{f} is connected, then the conformal dimension of 𝒥f\mathcal{J}_{f} equals 1, see [CP12], [Kin17].

Theorem C.

Let ff be a semi-hyperbolic rational map of degree dd with a connected Julia set 𝒥f\mathcal{J}_{f}. Suppose Cdim(𝒥f)=1\operatorname{Cdim}(\mathcal{J}_{f})=1. Then the conformal dimension is attained if and only if 𝒥f\mathcal{J}_{f} is homeomorphic to S1S^{1} or [0,1][0,1]. Moreover, if Cdim(𝒥f)=1\operatorname{Cdim}(\mathcal{J}_{f})=1 is attained, then ff is a sub-hyperbolic rational map whose corresponding post-critically finite rational map is zdz^{d}, 1/zd1/z^{d}, the degree-dd Chebyshev polynomial or the degree-dd negated Chebyshev polynomial up to conjugation by Möbius transformation.

The degree-dd negated Chebyshev polynomials are (-1) times the degree-dd Chebyshev polynomials.

Since ARConfdimCdim\operatorname{ARConfdim}\geq\operatorname{Cdim} in general, Theorem C holds even if we change the conformal dimension to Ahlfors-regular conformal dimension. Theorem C may be considered as a complex dynamical analogue of Theorem 1.1 for one dimension.

Julia sets of semi-hyperbolic quadratic polynomials are related to fractals generated by classical iterated function systems [ERS10]. For instance, the Julia set of z2+iz^{2}+i is shown to be quasi-symmetric equivalent to a universal object called the continuum self-similar tree, see [BM20] and [BM22]. The following is then an immediate corollary of Theorem A.

Corollary D.

The conformal dimension of the continuum self-similar tree, which is equal to 11, is not attained.

Acknowledgements

The authors are grateful to Kevin Pilgrim and Dylan Thurston for useful conversation.

The first author was supported by the Simons Foundation Institute Grant Award ID 507536 while the author was in residence at the Institute for Computational and Experimental Research in Mathematics in Providence, RI.

2. Background

2.1. Finite branched coverings (fbc’s)

In this subsection, we assume spaces XX and ZZ are compact, Hausdorff, connected, and locally connected topological spaces, and f:XZf:X\to Z is a finite-to-one continuous map. We define the degree of ff by

deg(f):=sup{#f1(z):zZ}\deg(f):=\sup\{\#f^{-1}(z):z\in Z\}

and the local degree of ff at xXx\in X by

deg(f;x):=infUsup{#f1(z)U:zf(U)}\deg(f;x):=\inf_{U}\sup\{\#f^{-1}(z)\cap U:z\in f(U)\}

where the infimum is taken over all neighborhoods UU of xx.

Definition 2.1 (Finite branched covering).

The map f:XZf:X\to Z is a finite branched covering (abbreviated fbc) provided deg(f)<\deg(f)<\infty and

  • (i)
    xf1(z)deg(f;x)=deg(f)\sum_{x\in f^{-1}(z)}\deg(f;x)=\deg(f)

    holds for each zZz\in Z;

  • (ii)

    for every x0Xx_{0}\in X, there are compact neighborhoods UU and VV of x0x_{0} and f(x0)f(x_{0}) respectively such that

    xU,f(x)=zdeg(f;x)=deg(f;x0)\sum_{x\in U,f(x)=z}\deg(f;x)=\deg(f;x_{0})

    for all zVz\in V.

The following properties of fbc’s are shown in [HP09, Lemma 2.1.2 and Lemma 2.1.3]: For an fbc f:XZf:X\to Z,

  • ff is open, closed, onto, and proper;

  • the set of branch points Bf:={xX:deg(f;x)>1}B_{f}:=\{x\in X:\deg(f;x)>1\} is nowhere dense in XX;

  • the set of branch values Vf:=f(Bf)V_{f}:=f(B_{f}) is nowhere dense in ZZ;

  • if UZU\subset Z is open and connected, then its inverse image f1(U)f^{-1}(U) is a union of disjoint open subsets f1(U)=U~1U~mf^{-1}(U)=\widetilde{U}_{1}\sqcup\ldots\sqcup\widetilde{U}_{m} where f:U~iUf:\widetilde{U}_{i}\to U is an fbc of degree did_{i}, and d1++dk=deg(f)d_{1}+\ldots+d_{k}=\deg(f).

We refer the reader to [Edm76] and [HP09] for more details on finite branched coverings.

Lemma 2.1 (Path-lifting for fbc’s).

Finite branched covers f:XZf:X\to Z have the path-lifting property: For a continuous map γ:[0,1]Z\gamma:[0,1]\to Z and x0f1(γ(0))x_{0}\in f^{-1}(\gamma(0)), there exists a continuous map γ~:[0,1]X\widetilde{\gamma}:[0,1]\to X with γ~(0)=x0\widetilde{\gamma}(0)=x_{0} and fγ~=γf\circ\widetilde{\gamma}=\gamma.

Proof.

An fbc is “interior” (sending open sets to open sets) and “light” (fibers being totally disconnected); cf. [Wal40]. The lemma follows from [Flo50, Theorem 2]. ∎

Recall that oo denotes the center of the space 𝕐\mathbb{Y}, i.e., the point where the three arms meet.

Corollary 2.2.

If f:XZf:X\to Z is an fbc and γ:𝕐Z\gamma:\mathbb{Y}\hookrightarrow Z and x0f1(γ(o))x_{0}\in f^{-1}(\gamma(o)) then there is a lift γ~:𝕐X\widetilde{\gamma}:\mathbb{Y}\hookrightarrow X with γ~(o)=x0\widetilde{\gamma}(o)=x_{0} and fγ~=γf\circ\widetilde{\gamma}=\gamma.

2.2. Coarse expanding conformal systems

When we consider a topological cxc system f:XXf:X\to X, we always assume that XX is non-singleton, compact, connected, locally connected, and metrizable.

Let 𝒰0\mathcal{U}_{0} be a cover of XX consisting of finitely many open connected subsets. For any n>0n>0, we inductively define a cover 𝒰n\mathcal{U}_{n} by the collection of connected components of inverse images of elements of 𝒰n\mathcal{U}_{n} under ff. We define 𝐔:=n0𝒰n\mathbf{U}:=\bigcup_{n\geq 0}\mathcal{U}_{n} the collection of all open sets in the covers 𝒰n\mathcal{U}_{n}’s. If U𝒰nU\in\mathcal{U}_{n}, then we say that the level of UU is nn and write |U|=n|U|=n.

Topological cxc systems

The dynamical system f:XXf:X\to X is topological cxc if there is a finite open cover 𝒰0\mathcal{U}_{0} satisfying the following three properties:

  • (Expansion) mesh(𝒰n)0\operatorname{mesh}(\mathcal{U}_{n})\to 0 as nn\to\infty. More precisely, for any open cover 𝒱\mathcal{V} of XX, there is N>0N>0 such that for any U𝒰nU\in\mathcal{U}_{n} with n>Nn>N there is V𝒱V\in\mathcal{V} satisfying UVU\subset V. If XX is a metric space, it is equivalent to max{diamU:U𝒰n}0\max\{\operatorname{diam}U:U\in\mathcal{U}_{n}\}\to 0 as nn\to\infty.

  • (Degree)

    max{deg(fk:U~U):k,U~𝒰k,U𝒰0}<.\max\{\deg(f^{k}:\widetilde{U}\to U):k\in\mathbb{N},~{}\widetilde{U}\in\mathcal{U}_{k},~{}U\in\mathcal{U}_{0}\}<\infty.
  • (Irreducibility) For any open set WXW\subset X, there exists nNn\in N with fn(W)=Xf^{n}(W)=X.

Metric cxc systems

For a metric space (X,dX)(X,d_{X}), a dynamical system f:XXf:X\to X is metric cxc if it is topological cxc about a finite open cover 𝒰0\mathcal{U}_{0} and, in addition, the following conditions hold for the same 𝒰0\mathcal{U}_{0}: There exist

  • continuous, increasing embeddings ρ±:[1,)[1,)\rho_{\pm}:[1,\infty)\to[1,\infty), called the forward and backward roundness distortion functions, and

  • increasing homeomorphisms δ±:[0,1][0,1]\delta_{\pm}:[0,1]\to[0,1], called the forward and backward relative diameter distortion functions,

such that the following two properties hold.

  • (Roundness distortion) For all n,kn,k\in\mathbb{N} and for all U𝒰n,U~𝒰n+k,y~U~U\in\mathcal{U}_{n},~{}\widetilde{U}\in\mathcal{U}_{n+k},~{}\tilde{y}\in\widetilde{U}, and yUy\in U, if fk(U~)=Uf^{\circ k}(\widetilde{U})=U and fk(y~)=yf^{\circ k}(\tilde{y})=y, then we have the backward roundness bound

    round(U~,y~)<ρ(round(U,y)),\mathrm{round}(\widetilde{U},\tilde{y})<\rho_{-}(\mathrm{round}(U,y)),

    and the forward roundness bound

    round(U,y)<ρ+(round(U~,y~)),\mathrm{round}(U,y)<\rho_{+}(\mathrm{round}(\widetilde{U},\tilde{y})),

    where for an interior point aa of AA the roundness of AA about aa is defined as

    round(A,a)=sup{|xa|:xA}sup{r:B(a,r)A}.\mathrm{round}(A,a)=\frac{\sup\{|x-a|:x\in A\}}{\sup\{r:B(a,r)\subset A\}}.
  • (Diameter distortion) For all n0,n1,kn_{0},n_{1},k\in\mathbb{N} and for all U𝒰n0,U𝒰n1,U~𝒰n0+kU\in\mathcal{U}_{n_{0}},~{}U^{\prime}\in\mathcal{U}_{n_{1}},~{}\widetilde{U}\in\mathcal{U}_{n_{0}+k}, and U~𝒰n1+k\widetilde{U}^{\prime}\in\mathcal{U}_{n_{1}+k} with U~U~\widetilde{U}^{\prime}\subset\widetilde{U} and UUU^{\prime}\subset U, if fk(U~)=Uf^{k}(\widetilde{U})=U and fk(U~)=Uf^{k}(\widetilde{U}^{\prime})=U^{\prime}, then

    diamU~diamU~<δ(diamUdiamU)\frac{\operatorname{diam}\widetilde{U}^{\prime}}{\operatorname{diam}\widetilde{U}}<\delta_{-}\left(\frac{\operatorname{diam}U^{\prime}}{\operatorname{diam}U}\right)

    and

    diamUdiamU<δ+(diamU~diamU~).\frac{\operatorname{diam}U^{\prime}}{\operatorname{diam}U}<\delta_{+}\left(\frac{\operatorname{diam}\widetilde{U}^{\prime}}{\operatorname{diam}\widetilde{U}}\right).

From [HP09], we have that the property of being metric cxc is preserved by quasi-symmetric conjugacies.

Visual metrics

A dynamical system f:XXf:X\to X that is topological cxc with respect to a finite open cover 𝒰0\mathcal{U}_{0} determines a class of so-called visual metrics. Visual metrics are compatible with the dynamics and 𝒰0\mathcal{U}_{0} in a sense that any element in 𝐔\bf{U} can be thought of as a ball whose size exponentially decays as the level increases. The following results summarize key properties of the visual metrics; see [HP09, Chapter 3].

Proposition 2.3 (Visual metric, [HP09, Proposition 3.3.2 and Proposition 3.2.3]).

Let (X,dX)(X,d_{X}) be a metric space. Suppose that f:XXf:X\to X is metric cxc with respect to an open cover 𝒰0\mathcal{U}_{0}. There exist ϵ>0\epsilon>0 and a metric dd, called a visual metric, on XX that is quasi-symmetric to dXd_{X} such that the following estimates hold for (X,d)(X,d):

  1. (1)

    (Nearly balls, I) There is some constant C>1C>1 such that, for all W𝐔W\in\mathbf{U}, there is a point xWx\in W so that

    B(x,(1/C)eϵ|W|)WB(x,Ceϵ|W|).B(x,(1/C)e^{-\epsilon|W|})\subset W\subset B(x,Ce^{-\epsilon|W|}).
  2. (2)

    (Nearly balls, II) There is a radius r1>0r_{1}>0 such that, for any n1n\geq 1 and for any xXx\in X, there is some W𝒰nW\in\mathcal{U}_{n} so that B(x,r1eϵn)WB(x,r_{1}e^{-\epsilon n})\subset W.

  3. (3)

    (Nearly balls, III) There exists r0r_{0} such that, for any r(0,r0)r\in(0,r_{0}) and any xXx\in X, there exist WW and WW^{\prime} in 𝐔\mathbf{U} such that |W||W|=O(1)|W|-|W^{\prime}|=O(1),

    xWB(x,r)W,x\in W^{\prime}\subset B(x,r)\subset W,

    and

    max{round(W,x),round(W,x)}=O(1).\max\{\mathrm{round}(W,x),\mathrm{round}(W^{\prime},x)\}=O(1).
  4. (4)

    (Homothety) For every x,yXx,y\in X, d(x,y)eεd(f(x),f(y))d(x,y)\geq e^{-\varepsilon}\cdot d(f(x),f(y)).

In fact, we have d(x,y)=eεd(f(x),f(y))d(x,y)=e^{-\varepsilon}\cdot d(f(x),f(y)) for x,yBx,y\in B where BB is a sufficiently small ball that does not contain a branch point [HP09, Proposition 3.2.3].

The following proposition implies that a visual metric is a canonical metric, in some sense, with which the topological cxc map becomes metric cxc.

Proposition 2.4 ([HP09, Theorem 3.5.3 and Corollary 3.5.4]).

Let f:XXf:X\to X be a topological cxc dynamical system. There is a quasi-symmetric equivalence class 𝒢(f)\mathcal{G}(f) of metrics on XX, called the canonical gauge, such that

  • visual metrics are in 𝒢(f)\mathcal{G}(f), and

  • d𝒢(f)d\in\mathcal{G}(f) if and only if ff is metric cxc with respect to dd.

Hence, if two metric cxc dynamical systems f:XXf:X\to X and g:ZZg:Z\to Z are topologically conjugate by a homeomorphism h:XZh:X\to Z, then hh is a quasi-symmetric conjugacy.

We call the quasi-symmetric equivalence class 𝒢(f)\mathcal{G}(f) of metrics on XX that contains visual metrics the canonical gauge (or the conformal gauge) of ff. The conformal dimension Cdim((X,𝒢(f)))\operatorname{Cdim}((X,\mathcal{G}(f))) (resp. the Ahlfors-regular conformal dimension ARConfdim((X,𝒢(f)))\operatorname{ARConfdim}((X,\mathcal{G}(f)))) denotes the infimal Hausdorff dimension of metrics (resp. Ahlfors-regular metrics) on XX in the canonical gauge 𝒢(f)\mathcal{G}(f).

2.3. Antenna-like spaces

We refer the reader to [Azz15] for a detailed account of this subsection.

Definition 2.2 (Antenna-like space).

Suppose XX is a compact connected metric space. For 0<c<10<c<1 and an open subset UU of XX, we say that UU has a cc-antenna if there is a homeomorphism h:𝕐Uh:\mathbb{Y}\hookrightarrow U such that for all permutations (i,j,k)(i,j,k) of (1,2,3)(1,2,3), the distance between h(ei)h(e_{i}) and h([0,ej])h([0,ek])h([0,e_{j}])\cup h([0,e_{k}]) is at least cdiam(U)c\cdot\operatorname{diam}(U). The space XX is called cc-antenna-like if for each r<12diam(X)r<\frac{1}{2}\operatorname{diam}(X), every ball B(x,r)B(x,r) has a cc-antenna.

The following theorem is a combination of the results in [Azz15].

Theorem 2.5 (Dimension of antenna-like spaces [Azz15]).

Suppose XX is a compact connected metric space. Suppose that XX is cc-antenna-like for some c(0,1)c\in(0,1). Then the following are hold.

  • If a metric space ZZ is quasi-symmetric to XX, then ZZ is also a cc^{\prime}-antenna-like for some c(0,1)c^{\prime}\in(0,1).

  • There is a uniform constant b>0b>0 that is independent of cc such that Hdim(X)>1+bc2\operatorname{Hdim}(X)>1+bc^{2}.

  • Hence, Cdim(X)>1\operatorname{Cdim}(X)>1 or Cdim(X)=1\operatorname{Cdim}(X)=1 but not attained.

3. No exotic 1-dimensional cxc systems

Cxc maps on a circle S1S^{1} or a closed interval II are quasi-symmetrically conjugate to the well-known dynamical systems.

Theorem 3.1.

Let f:XXf:X\to X be a metric cxc map of degree-dd.

  • (i)

    If X=S1X=S^{1}, then ff is quasi-symmetrically conjugate to the map zzdz\mapsto z^{d} on the unit circle 𝕊1\mathbb{S}^{1} [HP09, Theorem 4.1.1].

  • (ii)

    If X=IX=I, then ff is quasi-symmetrically conjugate to the degree-dd Chebyshev or negated Chebyshev polynomial on [1,1][-1,1].

As an immediate corollary, we can obtain a rigidity theorem, promoting semi-hyperbolic rational maps to sub-hyperbolic rational maps, i.e., every critical point in Julia sets is preperiodic.

Corollary 3.2.

Let ff be a semi-hyperbolic rational map with the Julia set 𝒥f\mathcal{J}_{f} homeomorphic to S1S^{1} or II. Then ff is a sub-hyperbolic rational map.

Chebyshev and negated Chebyshev polynomials

For d1d\geq 1, the degree-dd Chebyshev polynomial TdT_{d} is defined by Td(cosθ)=cos(dθ)T_{d}(\cos{\theta})=\cos(d\cdot\theta). The Julia set of TdT_{d} is a closed interval [1,1][-1,1] so that Td:[1,1][1,1]T_{d}:[-1,1]\to[-1,1] is well-defined. We always have Td(1)=1T_{d}(1)=1 but Td(1)=1T_{d}(-1)=-1 if dd is odd and Td(1)=1T_{d}(-1)=1 if dd is even.

Case (i): d=2n+1d=2n+1.

We have

f1(1)={1,x1,x2,,xn}f^{-1}(-1)=\{-1,x_{1},x_{2},\dots,x_{n}\}

and

f1(1)={1,y1,y2,,yn}f^{-1}(1)=\{1,y_{1},y_{2},\dots,y_{n}\}

so that deg(f;xi)=deg(f;yi)=2\deg(f;x_{i})=\deg(f;y_{i})=2 for any i,ji,j. By reordering the indices, we have

1<y1<x1<y2<x2<<yn<xn<1.-1<y_{1}<x_{1}<y_{2}<x_{2}<\cdots<y_{n}<x_{n}<1.

Then ff homeomorphically maps each connected component of [1,1]f1({1,1})[-1,1]\setminus f^{-1}(\{-1,1\}) to (1,1)(-1,1).

Case (ii): d=2nd=2n.

We have

f1(1)={x1,x2,,xn}f^{-1}(-1)=\{x_{1},x_{2},\dots,x_{n}\}

and

f1(1)={1,1,y1,y2,,yn1}f^{-1}(1)=\{-1,1,y_{1},y_{2},\dots,y_{n-1}\}

so that deg(f;xi)=deg(f;yj)=2\deg(f;x_{i})=\deg(f;y_{j})=2 for any i,ji,j. By reordering the indices, we have

1<x1<y1<x2<y2<<yn1<xn<1.-1<x_{1}<y_{1}<x_{2}<y_{2}<\cdots<y_{n-1}<x_{n}<1.

Again ff homeomorphically maps each connected component of [1,1]f1({1,1})[-1,1]\setminus f^{-1}(\{-1,1\}) to (1,1)(-1,1).

Now we consider the negated Chebyshev polynomial Td:=TdT^{\prime}_{d}:=-T_{d}. If d=2nd=2n then TdT^{\prime}_{d} and TdT_{d} are conjugate each other by the negation zzz\mapsto-z. Suppose d=2n+1d=2n+1. Since TdT^{\prime}_{d} swaps the two end points {1,1}\{-1,1\}, TdT^{\prime}_{d} is not conjugate to TdT_{d}. With the notations used in Case (i) above, we have f1(1)={1,y1,y2,,yn}f^{-1}(-1)=\{1,y_{1},y_{2},\dots,y_{n}\} and f1(1)={1,x1,x2,,xn}f^{-1}(1)=\{-1,x_{1},x_{2},\dots,x_{n}\}.

One can construct the degree-dd Chebyshev polynomial Td:[1,1][1,1]T_{d}:[-1,1]\to[-1,1] by projecting the map zzdz\mapsto z^{d} on the unit circle 𝕊1\mathbb{S}^{1} onto the diameter [1,1][-1,1] on the real axis. Suppose d=2n+1d=2n+1. The map exp(iθ)exp(i(πθ))\exp(i\theta)\mapsto\exp(i(\pi-\theta)) is the reflection of 𝕊1\mathbb{S}^{1} in the imaginary axis. It follows from

exp(i(2n+1)(πθ))=exp(i(π(2n+1)θ))\exp(i\cdot(2n+1)\cdot(\pi-\theta))=\exp(i\cdot(\pi-(2n+1)\theta))

that the projection of the map zzdz\mapsto z^{d} on 𝕊1\mathbb{S}^{1} onto the diameter on the imaginary axis [i,i][-i,i] is well-defined. This vertical projection gives rises to the negated Chebyshev polynomial dynamics.

Proof of Theorem 3.1.

See [HP09, Theorem 4.1.1] for the proof of Theorem 3.1-(i). We prove Theorem 3.1-(ii) here. Both proofs show the existence of a semi-conjugacy first and then show the semi-conjugacy is indeed a conjugacy.

By Proposition 2.4, it suffices to show that ff is topologically conjugate to the dynamics of a Chebyshev or negated Chebyshev polynomial on the interval.

Let I=[0,1]I=[0,1]. It follows from the topology of II that for every branch point xx of ff, we have deg(f;x)=2\deg(f;x)=2, xint(I)x\in\operatorname{int}(I), and f(x){0,1}f(x)\in\{0,1\}. Also, since ff is an open map, f({0,1}){0,1}f(\{0,1\})\subset\{0,1\}. Hence every branch point is prefixed.

There are three cases: (i) f(0)=0f(0)=0 and f(1)=1f(1)=1, (ii) f(0)=1f(0)=1 and f(1)=0f(1)=0, and (iii) f(1)=f(0)=0or1f(1)=f(0)=0~{}\mathrm{or}~{}1.

Let x1,x2,,xn(0,1)f1(0)x_{1},x_{2},\dots,x_{n}\in(0,1)\cap f^{-1}(0) and y1,y2,,ym(0,1)f1(1)y_{1},y_{2},\dots,y_{m}\in(0,1)\cap f^{-1}(1) so that xi<xi+1x_{i}<x_{i+1} and yj<yj+1y_{j}<y_{j+1} for any ii and jj. Then deg(f;xi)=deg(f;yj)=2\deg(f;x_{i})=\deg(f;y_{j})=2 for any i,ji,j.

Case (i): Since

deg(f;0)+i=1ndeg(f;xi)=deg(f;1)+i=1mdeg(f;yi),\deg(f;0)+\sum_{i=1}^{n}\deg(f;x_{i})=\deg(f;1)+\sum_{i=1}^{m}\deg(f;y_{i}),

we have n=mn=m and deg(f)=2n+1\deg(f)=2n+1. It follows that

0<y1<x1<y2<x2<<yn<xn<1.0<y_{1}<x_{1}<y_{2}<x_{2}<\cdots<y_{n}<x_{n}<1.

The map ff is then a homeomorphism onto (0,1)(0,1) on each connected component JJ of (0,1)i=1n{xi,yi}(0,1)\setminus\bigcup_{i=1}^{n}\{x_{i},y_{i}\}; if not, then JJ has a branch point zz which must be mapped to either 0 or 11 so that zz is xix_{i} or yiy_{i} for some ii.

One can compute the asymptotic growth rate ss of the number of laps for fnf^{n} (simply called the growth number), which is equal to the leading eigenvalue of the incidence matrix for the Markov partition of f:IIf:I\to I by i=1n{xi,yi}\bigcup_{i=1}^{n}\{x_{i},y_{i}\}, and check that s>1s>1. Then there is one and only one piecewise linear map g:IIg:I\to I with |g(x)|=s|g^{\prime}(x)|=s for every xx where g(x)g^{\prime}(x) is well-defined such that there is a semi-conjugacy hh from f:IIf:I\to I to g:IIg:I\to I [MT88, Theorem 7.4]. Note that (Irreducibility) for topological cxc maps implies topological transitivity, i.e., for any pair of open sets U,VIU,V\subset I, we have fn(U)Vf^{n}(U)\cap V\neq\emptyset for some n0n\geq 0. Then the semi-conjugacy hh is a topological conjugacy [ALM00, Proposition 4.6.9].

Since ff and the Chebyshev polynomial T2n+1T_{2n+1} have the same transition matrices of Markov partitions, the growth numbers of ff and T2n+1T_{2n+1} are equal. It follows from the same argument used for ff that T2n+1T_{2n+1} is also topologically conjugate to gg. Hence ff and T2n+1T_{2n+1} are topologically conjugate.

Case (ii): By a similar argument in Case (i), we can show that ff is topologically conjugate to the negated Chebyshev polynomial T2n+1T^{\prime}_{2n+1} of degree 2n+12n+1.

Case (iii): Without loss of generality, suppose f(0)=f(1)=1f(0)=f(1)=1. Since

i=1ndeg(f;xi)=deg(f;0)+deg(f;1)+i=1mdeg(f;yi),\sum_{i=1}^{n}\deg(f;x_{i})=\deg(f;0)+\deg(f;1)+\sum_{i=1}^{m}\deg(f;y_{i}),

we have n=m+1n=m+1 and deg(f)=2n\deg(f)=2n. It follows that

0<x1<y1<x2<<yn1<xn<1.0<x_{1}<y_{1}<x_{2}<\cdots<y_{n-1}<x_{n}<1.

Then by a similar argument used in Case (i), one can show that ff is topologically conjugate to a Chebyshev polynomial T2nT_{2n} of degree 2n2n.

4. Proof of Theorem A and Theorem C

In this section we prove Theorems A and C.

4.1. Proof of Theorem A

Suppose f:XXf:X\to X is metric cxc with respect to 𝒰0\mathcal{U}_{0} and XX is equipped with properties (1)-(4) in Proposition 2.3. We also use the constants C,r0C,r_{0}, and ε\varepsilon for used in Proposition 2.3. Suppose that there is an embedding 𝕐X\mathbb{Y}\hookrightarrow X.

By Theorem 2.5, it suffices to show that XX is c′′c^{\prime\prime}-antenna-like for some c′′>0c^{\prime\prime}>0. We divide the proof into three steps.

  • (1)

    We first show in Lemma 4.1 that there is c>0c>0 so that every U𝒰0U\in\mathcal{U}_{0} has a cc-antenna.

  • (2)

    Then we can use Lemma 4.2 to show that there exists c>0c^{\prime}>0 so that every U𝐔U\in\mathbf{U} has a cc^{\prime}-antenna.

  • (3)

    Then Lemma 4.3 applies to show that XX is c′′c^{\prime\prime}-antenna-like for some c′′>0c^{\prime\prime}>0.

Let us show the three lemmas.

Lemma 4.1.

There exists c>0c>0 so that every U𝒰0U\in\mathcal{U}_{0} has a cc-antenna.

Proof.

By the assumption, we have an embedding 𝕐X\mathbb{Y}\hookrightarrow X. There exists U0𝒰0U_{0}\in\mathcal{U}_{0} containing the center oo of 𝕐\mathbb{Y}. By restricting the copy of 𝕐\mathbb{Y} to U0U_{0} if necessary, we may assume that 𝕐U0\mathbb{Y}\hookrightarrow U_{0}. Since 𝒰0\mathcal{U}_{0} is a finite cover, it suffices to show that every V𝒰0V\in\mathcal{U}_{0} has an antenna.

Let V𝒰0V\in\mathcal{U}_{0}. We show that there is a connected component UN(V)U_{N(V)} of fN(V)(U0)f^{-N(V)}(U_{0}) so that UN(V)VU_{N(V)}\subset V. It follows from Proposition 2.3 that there is xVx\in V so that B(x,1/C)VB(x,1/C)\subset V. We take an rr-ball B(x,r)VB(x,r)\subset V for r<1/Cr<1/C. By (Irreducibility) of topological cxc maps, there is K>0K>0 so that fK(B(x,r))=Xf^{K}(B(x,r))=X. Then, there is a connected component UKU_{K} of fK(U0)f^{-K}(U_{0}) so that UKB(x,r)U_{K}\cap B(x,r)\neq\emptyset. By Proposition 2.3-(1) we have

diam(UK)<2CeεK.\operatorname{diam}(U_{K})<2Ce^{-\varepsilon K}.

Hence, by triangle inequality, we have

UKB(x,r+2CeεK).U_{K}\subset B(x,r+2Ce^{-\varepsilon K}).

Note that we can make rr arbitrarily small. Moreover, if we decrease rr, then KK should increase. Hence we can make KK arbitrarily large. For a sufficiently small rr and a sufficiently large KK, we have r+2CeϵK<1/Cr+2Ce^{-\epsilon\cdot K}<1/C so that UKVU_{K}\subset V. We write K=K(V)K=K(V) to indicate the dependence on VV.

Since fK(V):UK(V)U0f^{K(V)}:U_{K(V)}\to U_{0} is a fbc, using Corollary 2.2, we can find a lift 𝕐UK(V)V\mathbb{Y}\hookrightarrow U_{K(V)}\subset V of 𝕐U0\mathbb{Y}\hookrightarrow U_{0}. Thus any V𝒰0V\in\mathcal{U}_{0} has an antenna. ∎

Lemma 4.2.

There exists c(0,1)c^{\prime}\in(0,1) such that every U𝐔U^{\prime}\in\mathbf{U} has a cc^{\prime}-antenna.

Proof.

It suffices to show that for any U𝒰0U\in\mathcal{U}_{0} and n>0n>0, every U𝒰nU^{\prime}\in\mathcal{U}_{n} with fn(U)=Uf^{n}(U^{\prime})=U has a cc^{\prime}-antenna.

Suppose h:𝕐Uh:\mathbb{Y}\hookrightarrow U is an embedding such that for all permutations (i,j,k)(i,j,k) of (1,2,3)(1,2,3), the distance between h(ei)h(e_{i}) and h([0,ej])h([0,ek])h([0,e_{j}])\cup h([0,e_{k}]) is at least cdiam(U)c\cdot\operatorname{diam}(U). Suppose U𝒰nU^{\prime}\in\mathcal{U}_{n} is a connected component of fn(U)f^{-n}(U). Let y=h(o)y=h(o) where oo is the branch point 𝕐\mathbb{Y}. Let xUx\in U^{\prime} be a point such that fn(x)=yf^{n}(x)=y. Since fn:UUf^{n}:U^{\prime}\to U is an fbc, it follows from Corollary 2.2 that there is an embedding h~:𝕐U\tilde{h}:\mathbb{Y}\to U^{\prime} which is a lift of h:𝕐Uh:\mathbb{Y}\hookrightarrow U, i.e., fnh~=hf^{n}\circ\tilde{h}=h and h~(o)=x\tilde{h}(o)=x. Proposition 2.3-(4) implies that for any t1,t2𝕐t_{1},t_{2}\in\mathbb{Y},

d(h~(t1),h~(t2))eεnd(h(t1),h(t2)).d(\tilde{h}(t_{1}),\tilde{h}(t_{2}))\geq e^{-\varepsilon n}d(h(t_{1}),h(t_{2})).

Thus

d(h~(e1),h~([0,e2][0,e3]))\displaystyle\hskip 12.0ptd(\tilde{h}(e_{1}),\tilde{h}([0,e_{2}]\cup[0,e_{3}]))
=inf{d(h~(e1),h~(t)):t[0,e2][0,e3]}\displaystyle=\inf\{d(\tilde{h}(e_{1}),\tilde{h}(t)):t\in[0,e_{2}]\cup[0,e_{3}]\}
eεninf{d(h(e1),h(t)):t[0,e2][0,e3]}\displaystyle\geq e^{-\varepsilon n}\inf\{d(h(e_{1}),h(t)):t\in[0,e_{2}]\cup[0,e_{3}]\}
=eεnd(h(e1),h([0,e2][0,e3]))\displaystyle=e^{-\varepsilon n}d(h(e_{1}),h([0,e_{2}]\cup[0,e_{3}]))
eεncdiamU.\displaystyle\geq e^{-\varepsilon n}c\cdot\operatorname{diam}U.

Likewise, we have

d(h~(e2),h~([0,e3][0,e1])),d(h~(e3),h~([0,e1][0,e2]))eεncdiamU.\displaystyle d(\tilde{h}(e_{2}),\tilde{h}([0,e_{3}]\cup[0,e_{1}])),d(\tilde{h}(e_{3}),\tilde{h}([0,e_{1}]\cup[0,e_{2}]))\geq e^{-\varepsilon n}c\cdot\operatorname{diam}U.

It follows from Proposition 2.3-(1) that

diamU2CeεnC2eεndiamU.\operatorname{diam}U^{\prime}\leq 2Ce^{-\varepsilon n}\leq C^{2}e^{-\varepsilon n}\cdot\operatorname{diam}U.

We conclude that UU^{\prime} has a cc^{\prime}-antenna for c=c/C2c^{\prime}=c/{C^{2}}. ∎

Lemma 4.3.

If every U𝐔U\in\mathbf{U} has a cc-antenna for some c>0c>0, then XX is cc^{\prime}-antenna-like for some c(0,1)c^{\prime}\in(0,1).

Proof.

Suppose B(x,r)B(x,r) be a ball of radius rr Let r0r_{0} is as defined in Proposition 2.3.

If r<r0r<r_{0}, then by Proposition 2.3-(3) there exist W,W𝐔W,W^{\prime}\in\mathbf{U} such that |W||W|=O(1)|W|-|W^{\prime}|=O(1) and

WB(x,r)W.W^{\prime}\subset B(x,r)\subset W.

Since |W||W|=O(1)|W|-|W^{\prime}|=O(1), Proposition 2.3-(1) yields

diamWC2eε(|W||W|)diamW2rC2eεO(1).\operatorname{diam}W^{\prime}\geq C^{-2}e^{-\varepsilon(|W|^{\prime}-|W|)}\operatorname{diam}W\geq 2rC^{-2}e^{-\varepsilon\cdot O(1)}.

Let c=cC2eεO(1)c^{\prime}=c\cdot C^{-2}\cdot e^{-\varepsilon\cdot O(1)}. By the assumption in the statement of the lemma, there exists an embedding h:𝕐WB(x,r)h:\mathbb{Y}\to W^{\prime}\subset B(x,r) such that for all permutations (i,j,k)(i,j,k) of (1,2,3)(1,2,3), the distance between h(ei)h(e_{i}) and h([0,ej])h([0,ek])h([0,e_{j}])\cup h([0,e_{k}]) is at least cdiam(W)c2r=cdiamB(x,r)c\cdot\operatorname{diam}(W^{\prime})\geq c^{\prime}\cdot 2r=c^{\prime}\cdot\operatorname{diam}B(x,r), i.e., B(x,r)B(x,r) has a cc^{\prime}-antenna. If r0>12diamXr_{0}>\frac{1}{2}\operatorname{diam}X, then this completes the proof of the lemma.

Suppose now r0<r<12diamXr_{0}<r<\frac{1}{2}\operatorname{diam}X. Then 2rr0diamX<r0\frac{2rr_{0}}{\operatorname{diam}X}<r_{0} so that there exists rr^{\prime}\in\mathbb{R} such that 2rr0diamX<r<r0\frac{2rr_{0}}{\operatorname{diam}X}<r^{\prime}<r_{0}. By the previous paragraph, there exists an embedding h:𝕐B(x,r)h:\mathbb{Y}\to B(x,r^{\prime}) such that for all permutations (i,j,k)(i,j,k) of (1,2,3)(1,2,3), the distance between h(ei)h(e_{i}) and h([0,ej])h([0,ek])h([0,e_{j}])\cup h([0,e_{k}]) is at least c2r>c(2r0/diamX)2rc^{\prime}\cdot 2r^{\prime}>c^{\prime}\cdot(2r_{0}/\operatorname{diam}X)\cdot 2r. Replacing cc^{\prime} by c(2r0/diamX)c^{\prime}\cdot(2r_{0}/\operatorname{diam}X), we can make XX cc^{\prime}-antenna-like. ∎

4.2. Proof of Theorem C

We prove the following proposition first and then complete the proof of Theorem C.

Proposition 4.4.

Suppose 𝒥f\mathcal{J}_{f} is the Julia set of a semi-hyperbolic rational map ff. If 𝒥f\mathcal{J}_{f} is connected, and 𝒥f\mathcal{J}_{f} is homeomorphic to neither the circle S1S^{1} or a closed interval [0,1][0,1], then 𝕐J\mathbb{Y}\hookrightarrow J.

Proof.

Let ff be a semi-hyperbolic rational map with a connected Julia set 𝒥f\mathcal{J}_{f}. Then 𝒥f\mathcal{J}_{f} is locally connected [Mih11, Theorem 2]. A connected and locally connected compact subset of 2\mathbb{R}^{2} is arcwise connected [Mil06, Lemmas 17.17 and 17.18], i.e., for every x,y𝒥fx,y\in\mathcal{J}_{f}, there exists a homeomorphic embedding, called an arc, γ:[0,1]𝒥f\gamma:[0,1]\to\mathcal{J}_{f} such that γ(0)=x\gamma(0)=x and γ(1)=y\gamma(1)=y.

Let us assume that 𝒥f\mathcal{J}_{f} does not contain a homeomorphic copy of 𝕐\mathbb{Y} and show that 𝒥f\mathcal{J}_{f} is either S1S^{1} or [0,1][0,1].

If 𝒥f\mathcal{J}_{f} contains a simple closed cure CC that omits a point x𝒥fx\in\mathcal{J}_{f}, then xx is joined to some point in CC by an arc γ:[0,1]𝒥f\gamma:[0,1]\to\mathcal{J}_{f} so that γ(0)=x\gamma(0)=x and γ(1)C\gamma(1)\in C. Let t:=min{t[0,1]:γ(t)Ct:=\min\{t\in[0,1]:\gamma(t)\in C. Let II be a subarc of CC containing γ(t)\gamma(t) in its interior. Then Iγ([0,t])I\cup\gamma([0,t]) is the image of an embedding 𝕐𝒥f\mathbb{Y}\hookrightarrow\mathcal{J}_{f}, which contradicts to the assumption that 𝒥f\mathcal{J}_{f} does not contains 𝕐\mathbb{Y}. Hence 𝒥f\mathcal{J}_{f} is homeomorphic to S1S^{1}.

If 𝒥f\mathcal{J}_{f} does not contain a simple closed loop, then 𝒥f\mathcal{J}_{f} is a locally connected continuum that contains no simple closed curves, i.e., a dendrite in the sense of [CC98]. Pick x𝒥fx\in\mathcal{J}_{f}. Then xx is either a cut point of 𝒥f\mathcal{J}_{f} or an end point of 𝒥f\mathcal{J}_{f} [CC98, Theorem 1.1 (3)].

If 𝒥f\{x}\mathcal{J}_{f}\backslash\{x\} has more than 2 connected components, then one can find a homeomorphic copy of 𝕐\mathbb{Y} centered at xx, contradicting our assumption. If 𝒥f\{x}\mathcal{J}_{f}\backslash\{x\} has two connected components, we denote by L1L_{1} and L2L_{2} the closures of the connected components. If xx is an end point of 𝒥f\mathcal{J}_{f}, we let L1=𝒥fL_{1}=\mathcal{J}_{f} and L2=L_{2}=\emptyset.

Every point yL1y\in L_{1} with yxy\neq x can be joined to xx by an unique arc [CC98, Theorem 1.2 (20)]. Let y1y_{1} and y2y_{2} be two distinct points in L1\{x}L_{1}\backslash\{x\}. For i=1,2i=1,2, let γi\gamma_{i} be the unique arc joining xx and yiy_{i}. Since xγ1γ2x\in\gamma_{1}\cap\gamma_{2}, γ1γ2\gamma_{1}\cap\gamma_{2} is non-empty. By hereditary unicoherency of 𝒥f\mathcal{J}_{f}([CC98, Theorem 1.1 (18)]), γ1γ2\gamma_{1}\cap\gamma_{2} is connected. Then either γ1γ2=γ1\gamma_{1}\cap\gamma_{2}=\gamma_{1} or γ1γ2=γ2\gamma_{1}\cap\gamma_{2}=\gamma_{2}, for otherwise γ1γ2\gamma_{1}\cup\gamma_{2} would contain a homeomorphic copy of 𝕐\mathbb{Y}. This proves that L1L_{1} is homeomorphic to [0,1][0,1] if they are non-empty, with xx as one end point. The same proof shows that L2L_{2} is either empty or homeomorphic to [0,1][0,1] with xx as one end point, and hence 𝒥f\mathcal{J}_{f} is homeomorphic to [0,1][0,1]. ∎

Proof of Theorem C.

Suppose ff is a degree-dd semi-hyperbolic rational map with connected Julia set 𝒥f\mathcal{J}_{f} such that Cdim(𝒥f)=1\operatorname{Cdim}(\mathcal{J}_{f})=1. By Theorem A and Proposition 4.4, the conformal dimension is attained if and only if 𝒥f\mathcal{J}_{f} is homeomorphic to S1S^{1} or [0,1][0,1]. If 𝒥f\mathcal{J}_{f} is homeomorphic to S1S^{1} or [0,1][0,1], then, by Theorem 3.1 and Corollary 3.2, ff is sub-hyperbolic whose dynamics restricted to 𝒥f\mathcal{J}_{f} is quasi-symmetrically conjugate to zdz^{d}, the degree-dd Chebyshev polynomial, or the degree-dd negated Chebyshev polynomial. We can also have 1/zd1/z^{d} if the quasi-symmetric conjugacy reverses the orientation of on the circle. Then the unique post-critically rational map corresponding to the sub-hyperbolic rational map ff, by [McM88], is one of these rational maps up to conjugation by Möbius transformations. ∎

References

  • [ALM00] Lluís Alsedà, Jaume Llibre, and MichałMisiurewicz. Combinatorial dynamics and entropy in dimension one, volume 5 of Advanced Series in Nonlinear Dynamics. World Scientific Publishing Co., Inc., River Edge, NJ, second edition, 2000.
  • [Azz15] Jonas Azzam. Hausdorff dimension of wiggly metric spaces. Ark. Mat., 53(1):1–36, 2015.
  • [BK02] Mario Bonk and Bruce Kleiner. Rigidity for quasi-Möbius group actions. J. Differential Geom., 61(1):81–106, 2002.
  • [BM20] Mario Bonk and Daniel Meyer. Quasiconformal and geodesic trees. Fund. Math., 250(3):253–299, 2020.
  • [BM22] Mario Bonk and Daniel Meyer. Uniformly branching trees. Trans. Amer. Math. Soc., 375(6):3841–3897, 2022.
  • [CC98] Janusz J. Charatonik and Włodzimierz J. Charatonik. Dendrites. In XXX National Congress of the Mexican Mathematical Society (Spanish) (Aguascalientes, 1997), volume 22 of Aportaciones Mat. Comun., pages 227–253. Soc. Mat. Mexicana, México, 1998.
  • [CP12] Matias Carrasco Piaggio. Conformal dimension and combinatorial modulus of compact metric spaces. C. R. Math. Acad. Sci. Paris, 350(3-4):141–145, 2012.
  • [Edm76] Allan L. Edmonds. Branched coverings and orbit maps. Michigan Math. J., 23(4):289–301 (1977), 1976.
  • [ERS10] Kemal Ilgar Eroğlu, Steffen Rohde, and Boris Solomyak. Quasisymmetric conjugacy between quadratic dynamics and iterated function systems. Ergodic Theory Dynam. Systems, 30(6):1665–1684, 2010.
  • [Flo50] E. E. Floyd. Some characterizations of interior maps. Ann. of Math. (2), 51:571–575, 1950.
  • [HP09] Peter Haïssinsky and Kevin M. Pilgrim. Coarse expanding conformal dynamics. Astérisque, (325):viii+139 pp., 2009.
  • [HP12] Peter Haïssinsky and Kevin M. Pilgrim. Examples of coarse expanding conformal maps. Discrete Contin. Dyn. Syst., 32(7):2403–2416, 2012.
  • [HP14] Peter Haïssinsky and Kevin M. Pilgrim. Minimal Ahlfors regular conformal dimension of coarse expanding conformal dynamics on the sphere. Duke Math. J., 163(13):2517–2559, 2014.
  • [Kin17] Kyle Kinneberg. Conformal dimension and boundaries of planar domains. Trans. Amer. Math. Soc., 369(9):6511–6536, 2017.
  • [Mañ93] Ricardo Mañé. On a theorem of Fatou. Bol. Soc. Brasil. Mat. (N.S.), 24(1):1–11, 1993.
  • [McM88] Curt McMullen. Automorphisms of rational maps. In Holomorphic functions and moduli, Vol. I (Berkeley, CA, 1986), volume 10 of Math. Sci. Res. Inst. Publ., pages 31–60. Springer, New York, 1988.
  • [Mih11] Nicolae Mihalache. Julia and John revisited. Fund. Math., 215(1):67–86, 2011.
  • [Mil06] John Milnor. Dynamics in one complex variable, volume 160 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, third edition, 2006.
  • [MT88] John Milnor and William Thurston. On iterated maps of the interval. In Dynamical systems (College Park, MD, 1986–87), volume 1342 of Lecture Notes in Math., pages 465–563. Springer, Berlin, 1988.
  • [Prz06] Feliks Przytycki. On the hyperbolic Hausdorff dimension of the boundary of a basin of attraction for a holomorphic map and of quasirepellers. Bull. Pol. Acad. Sci. Math., 54(1):41–52, 2006.
  • [PZ21] Feliks Przytycki and Anna Zdunik. On Hausdorff dimension of polynomial not totally disconnected Julia sets. Bull. Lond. Math. Soc., 53(6):1674–1691, 2021.
  • [Sul82] Dennis Sullivan. Seminar on conformal and hyperbolic geometry (Note by M. Baker and J. Seade). Preprint IHES, 1982. URL: http://www.math.stonybrook.edu/ dennis/publications/.
  • [Wal40] A. D. Wallace. Quasi-monotone transformations. Duke Math. J., 7:136–145, 1940.
  • [Zdu90] Anna Zdunik. Parabolic orbifolds and the dimension of the maximal measure for rational maps. Invent. Math., 99(3):627–649, 1990.