betaFeedback

WorksMathematics

A complete classification of isometries of quasihyperbolic domains

Abstract

We classify all surjective quasihyperbolic isometries between arbitrary proper planar domains. Apart from Euclidean similarities, the orientation-preserving isometries are reciprocal maps on sectors or punctured planes and explicit complex-power maps on logarithmic spiral sectors. In particular, orientation-preserving planar quasihyperbolic isometries need not be Möbius. As consequences, we obtain similarity rigidity when either domain has finite inradius, for multiply connected domains other than punctured planes, and for C1C^1 domains other than half-planes. The C1C^1 conclusion strengthens Hästö's corresponding result for planar C3C^3 domains. Our planar classification, together with Richard's classification in dimensions n≥3n\geq3, completely solves the quasihyperbolic isometry problem between proper subdomains of Rn\mathbb R^n in every dimension n≥2n\geq2. As an application, we extend to dimension two the characterization of pointed Gromov–Hausdorff convergence of quasihyperbolic domains obtained by Herron, Richard, and Snipes.

Dedicated to David A. Herron and Pekka Koskela for their pioneering contributions to quasihyperbolic geometry.

A. Pandey was supported by the Research Council of Finland, project number 360505.

Introduction and main result

Background and main result

For a domain Ω⊊Rn\Omega\subsetneq\mathbb{R}^{n}, n≥2n \ge2, let

δΩ(x)=dist⁡(x,∂Ω):=inf⁡p∈∂Ω∣x−p∣andRΩ=sup⁡x∈ΩδΩ(x).\delta_{\Omega}(x) = \operatorname{dist}(x, \partial\Omega) := \inf_{p \in\partial\Omega} |x-p| \quad\text{and}\quad R_{\Omega} = \sup_{x \in\Omega} \delta_{\Omega}(x).

The quasihyperbolic metric kΩk_{\Omega} in Ω\Omega is defined by the length element ∣dx∣/δΩ(x)|dx|/\delta_{\Omega}(x), where ∣dx∣|dx| denotes the Euclidean length element. Thus, kΩ(x,y)=inf⁡γℓk(γ)k_{\Omega}(x,y) = \inf_{\gamma} \ell_{k}(\gamma), where the infimum is taken over all rectifiable curves γ\gamma in Ω\Omega joining xx and yy, and ℓk(γ)=∫γ∣dx∣/δΩ\ell_{k}(\gamma) = \int_{\gamma} |dx|/\delta_{\Omega} is the quasihyperbolic length of a rectifiable curve γ\gamma in Ω\Omega. It is well-known that (Ω,kΩ)(\Omega, k_{\Omega}) is complete proper and geodesic metric space; see [2 Proposition 2.8]. The pair (Ω,kΩ)(\Omega, k_{\Omega}) is a quasihyperbolic domain.

The quasihyperbolic metric was introduced by Gehring and Palka [8] in their study of quasiconformally homogeneous domain. It played an important role in Jones’s classical characterization of BMO extension domains [19]. Soon after the characterizations of uniform domains by Gehring and Osgood [7], the metric becomes a basic tool in quasiconformal analysis. It has found numerous applications in Sobolev and quasiconformal removability [20], conformal rigidity [25], and the uniformization and Gromov hyperbolicity of metric spaces [2, 1, 3]. For further background and applications, see Koskela’s survey [21].

Martin [23 Corollary 4.8 and Example 4.11] proved that quasihyperbolic geodesic segments in arbitrary proper Euclidean domains are C1,1C^{1,1} when parametrized by Euclidean arclength, and showed that the regularity is sharp. Later, Martin and Osgood [24 Corollary 3.6] used the notion of generalized Gaussian curvature in the sense of Heins [14] and proved that the quasihyperbolic metric in every proper planar domain has nonpositive generalized Gaussian curvature.

In 2007, Hästö [10] observed that its intrinsic geometry had received comparatively little attention, partly because the density δΩ−1\delta_{\Omega}^{-1} need not be differentiable everywhere. Subsequent advances include Herron’s proof that the quasihyperbolic metric universal cover of a planar domain is a Hadamard space [15], and the theorem of Ikonen and the author [IP26] that quasihyperbolic domains are CAT(2), with the constant 2 optimal in dimensions n≥3n \ge3. This resolved several conjectures of Väisälä [Väi09] on the local structure of the quasihyperbolic metric and its geodesics.

Quasihyperbolic isometry problem. A complementary question concerns the maps preserving this metric. For domains Ω,Ω′⊊Rn\Omega, \Omega' \subsetneq\mathbb{R}^{n}, n≥2n \ge2, a surjection f:Ω→Ω′f : \Omega\to\Omega' is a quasihyperbolic isometry if f:(Ω,kΩ)→(Ω′,kΩ′)f : (\Omega, k_{\Omega}) \to(\Omega', k_{\Omega'}) is an isometry, i.e.,

kΩ′(f(x),f(y))=kΩ(x,y)(x,y∈Ω).(1)k_{\Omega'}(f(x), f(y)) = k_{\Omega}(x,y) \qquad(x,y \in\Omega). \tag*{(1)}

The quasihyperbolic isometry problem asks for a complete characterization of these maps, allowing both the source and target domains to vary [10]. Related isometry problems have been studied for the Apollonian metric [11], the distance-ratio and Seittenranta metrics [12], and the Kulkarni–Pinkall metric [16]; see also the survey [HIM⁺07].

In this paper, we give a complete classification of surjective quasihyperbolic isometries between proper planar domains, describing both the maps and the domains on which they occur; see Theorem 1.2. Together with Richard’s higher-dimensional classification (Theorem A), it completely solves the quasihyperbolic isometry problem for proper subdomains of Rn\mathbb{R}^{n} in every dimension n≥2n \ge2.

We first introduce a family of domains that appears in the classification of planar quasihyperbolic isometries; see Theorem 1.2.

Example 1.1. For σ,θ∈R\sigma,\theta\in\mathbb{R} and 0<α≤2π0 < \alpha\le2\pi, define

Uσ,θ,α={s+it∈C:θ<t−σs<θ+α},U_{\sigma,\theta,\alpha}=\{s+it\in\mathbb{C}:\theta<t-\sigma s<\theta+\alpha\},
Dσ,θ,α=exp⁡Uσ,θ,α,Lσ,θ,α=(exp⁡∣Uσ,θ,α)−1.(2)D_{\sigma,\theta,\alpha}=\exp U_{\sigma,\theta,\alpha},\qquad L_{\sigma,\theta,\alpha}=(\exp|_{U_{\sigma,\theta,\alpha}})^{-1}. \tag*{(2)}

The map exp⁡∣Uσ,θ,α\exp|_{U_{\sigma,\theta,\alpha}} is injective. Indeed if w,w+2πim∈Uσ,θ,αw,w+2\pi i m\in U_{\sigma,\theta,\alpha} for m∈Zm\in\mathbb{Z}, then 2π∣m∣<α≤2π2\pi|m|<\alpha\le2\pi, so m=0m=0. Thus Lσ,θ,αL_{\sigma,\theta,\alpha} is a holomorphic logarithm on Dσ,θ,αD_{\sigma,\theta,\alpha}.

For σ=0\sigma=0, D0,θ,αD_{0,\theta,\alpha} is a sector, including a slit plane when α=2π\alpha=2\pi. For σ≠0\sigma\ne0, Dσ,θ,αD_{\sigma,\theta,\alpha} is a logarithmic spiral sector. In both cases,

∂Dσ,θ,α={0}∪⋃β∈{θ,θ+α}{e(1+iσ)s+iβ:s∈R}.(3)\partial D_{\sigma,\theta,\alpha}=\{0\}\cup\bigcup_{\beta\in\{\theta,\theta+\alpha\}}\{e^{(1+i\sigma)s+i\beta}:s\in\mathbb{R}\}. \tag*{(3)}

When α=2π\alpha=2\pi, the two boundary curves in (3) coincide, since

e(1+iσ)s+i(θ+2π)=e(1+iσ)s+iθ,s∈R.e^{(1+i\sigma)s+i(\theta+2\pi)}=e^{(1+i\sigma)s+i\theta},\qquad s\in\mathbb{R}.

In this case,

Dσ,θ,2π=C∖({0}∪{e(1+iσ)s+iθ:s∈R}).(4)D_{\sigma,\theta,2\pi}=\mathbb{C}\setminus\left(\{0\}\cup\{e^{(1+i\sigma)s+i\theta}:s\in\mathbb{R}\}\right). \tag*{(4)}

For σ≠0\sigma\ne0, (4) describes a plane slit along a logarithmic spiral; see Figure 1 for the illustration.

Shaded portions of the domains inside the unit disk

Figure 1. Shaded portions of the domains inside the unit disk.

We now state our main result.

Theorem 1.2. Let Ω,Ω′⊊C\Omega,\Omega'\subsetneq\mathbb{C} be domains and let f:(Ω,kΩ)→(Ω′,kΩ′)f:(\Omega,k_{\Omega})\to(\Omega',k_{\Omega'}) be a surjective isometry. After postcomposition with z↦z‾z\mapsto\overline{z} if necessary, exactly one of the following holds.

(1) Similarity. f(z)=az+bf(z)=az+b for some a∈C∖{0}a\in\mathbb{C}\setminus\{0\} and b∈Cb\in\mathbb{C}.

(2) Reciprocal map. For some p∈∂Ωp\in\partial\Omega, A∈C∖{0}A\in\mathbb{C}\setminus\{0\}, and B∈CB\in\mathbb{C},

f(z)=B+Az−p,Ω=p+D0,θ,α or C∖{p},(5)f(z)=B+\frac{A}{z-p},\qquad\Omega=p+D_{0,\theta,\alpha}\text{ or }\mathbb{C}\setminus\{p\}, \tag*{(5)}

where θ∈R\theta\in\mathbb{R} and 0<α≤2π0<\alpha\le2\pi in the sector case Ω=p+D0,θ,α\Omega=p+D_{0,\theta,\alpha}.

(3) Spiral-sector map. For some p∈∂Ωp\in\partial\Omega, A∈C∖{0}A\in\mathbb{C}\setminus\{0\}, B∈CB\in\mathbb{C}, σ∈R∖{0}\sigma\in\mathbb{R}\setminus\{0\}, θ∈R\theta\in\mathbb{R}, and 0<α≤2π0<\alpha\le2\pi,

Ω=p+Dσ,θ,α,λσ=−1+iσ1+iσ,\Omega=p+D_{\sigma,\theta,\alpha},\qquad\lambda_\sigma=\frac{-1+i\sigma}{1+i\sigma},
f(z)=B+Aexp⁡(λσLσ,θ,α(z−p)).(6)f(z)=B+A\exp\bigl(\lambda_\sigma L_{\sigma,\theta,\alpha}(z-p)\bigr). \tag*{(6)}

Conversely, every map in above three cases is a quasihyperbolic isometry onto its image.

Remark 1.3. The term exp⁡(λσLσ,θ,α(z−p))\exp\bigl(\lambda_\sigma L_{\sigma,\theta,\alpha}(z-p)\bigr) in (6) is the complex-power map since it is a complex power of z−pz-p, defined using the holomorphic logarithm Lσ,θ,αL_{\sigma,\theta,\alpha}. Writing

(z−p)λσ:=exp⁡(λσLσ,θ,α(z−p)),(z-p)^{\lambda_\sigma}:=\exp\bigl(\lambda_\sigma L_{\sigma,\theta,\alpha}(z-p)\bigr),

the maps in (6) take the form f(z)=B+A(z−p)λσf(z)=B+A(z-p)^{\lambda_\sigma}.

For a locally univalent holomorphic map ff, its Schwarzian derivative is

Sf=f′′′f′−32(f′′f′)2=(f′′f′)′−12(f′′f′)2.(7)S_f=\frac{f'''}{f'}-\frac{3}{2}\left(\frac{f''}{f'}\right)^2=\left(\frac{f''}{f'}\right)'-\frac{1}{2}\left(\frac{f''}{f'}\right)^2. \tag*{(7)}

Recall that a locally univalent holomorphic function on a domain is the restriction of a Möbius transformation if and only if its Schwarzian derivative vanishes identically [26 pp. 275–276]. The maps in (6) satisfy

∣λσ∣=1,Sf(z)=1−λσ22(z−p)2≠0.(8)|\lambda_\sigma|=1,\qquad S_f(z)=\frac{1-\lambda_\sigma^2}{2(z-p)^2}\ne0. \tag*{(8)}

Hence they are not Möbius. See Example 4.1 for explicit computation with exponent λ1=i\lambda_1=i. To the best of the author’s knowledge, these are the first examples of orientation-preserving quasihyperbolic isometries between proper planar domains that are not restrictions of Möbius transformations.

It is known that every Euclidean similarity

S(x)=cQx+b,c>0,b∈Rn,Q∈O(n),S(x)=cQx+b,\qquad c>0,\qquad b\in\mathbb{R}^n,\qquad Q\in O(n),

satisfies (1) with Ω′=S(Ω)\Omega'=S(\Omega). Here O(n):={Q∈Rn×n:QTQ=In}O(n):=\{Q\in\mathbb{R}^{n\times n}:Q^TQ=I_n\} is the group of orthogonal linear automorphisms of Rn\mathbb{R}^n; equivalently, it is the group of Euclidean isometries of Rn\mathbb{R}^n that fix the origin. Each Q∈O(n)Q\in O(n) extends to Rn∪{∞}\mathbb{R}^n\cup\{\infty\} by setting Q(∞)=∞Q(\infty)=\infty. Examples of Euclidean similarities include translations, rotations, reflections, and dilations. Throughout the paper, Möbius transformations of R^n=Rn∪{∞}\widehat{\mathbb{R}}^n=\mathbb{R}^n\cup\{\infty\} are understood to include both orientations. In dimension two, these are the Möbius and anti-Möbius transformations in the usual complex-analytic terminology.

The function f:H→Hf:\mathbb{H}\to\mathbb{H} defined by f(z)=−1/zf(z)=-1/z is a Möbius quasihyperbolic self-isometry of H={z∈C:Im⁡z>0}\mathbb{H}=\{z\in\mathbb{C}:\operatorname{Im}z>0\}, but is not a Euclidean similarity. More generally, in every dimension n≥2n\ge2, a Möbius transformation of $\widehat{\mathbb{R}}^n restricts to a Euclidean similarity of Rn\mathbb{R}^{n} if and only if it fixes ∞\infty. In particular, a planar Möbius transformation

M(z)=az+bcz+dorM(z)=a‾z+b‾c‾z+d‾,a,b,c,d∈C,ad−bc≠0,M(z)=\frac{az+b}{cz+d}\quad\text{or}\quad M(z)=\frac{\overline{a}z+\overline{b}}{\overline{c}z+\overline{d}},\qquad a,b,c,d\in\mathbb{C},\quad ad-bc\ne0,

restricts to a Euclidean similarity of C\mathbb{C} if and only if c=0c=0. Every Möbius transformation that does not fix ∞\infty has the form

M(x)=b+cQx−p∣x−p∣2(x∈Rn∖{p}),(9)M(x)=b+cQ\frac{x-p}{|x-p|^{2}}\qquad(x\in\mathbb{R}^{n}\setminus\{p\}), \tag*{(9)}

where p,b∈Rnp,b\in\mathbb{R}^{n}, c>0c>0, and Q∈O(n)Q\in O(n). Here p=M−1(∞)p=M^{-1}(\infty) and b=M(∞)b=M(\infty).

Definition 1.4. We say that similarity rigidity holds for a domain Ω⊊Rn\Omega\subsetneq\mathbb{R}^{n}, n≥2n\ge2 if every quasihyperbolic isometry f:(Ω,kΩ)→(f(Ω),kf(Ω))f:(\Omega,k_{\Omega})\to(f(\Omega),k_{f(\Omega)}), where f(Ω)⊊Rnf(\Omega)\subsetneq\mathbb{R}^{n} is a domain, is the restriction of a Euclidean similarity. Similarly, we say that Möbius rigidity holds for Ω\Omega if every such isometry is the restriction of a Möbius transformation of R‾n\overline{\mathbb{R}}^{n}.

We say that Ω\Omega is self-similar rigid if every surjective quasihyperbolic isometry f:(Ω,kΩ)→(Ω,kΩ)f:(\Omega,k_{\Omega})\to(\Omega,k_{\Omega}) is the restriction of a Euclidean similarity. Similarly, we say that Ω\Omega is self-Möbius rigid if every such self-isometry is the restriction of a Möbius transformation of R‾n\overline{\mathbb{R}}^{n}.

Known results on quasihyperbolic isometry

Let Ω,Ω′⊊Rn\Omega,\Omega'\subsetneq\mathbb{R}^{n}, n≥2n\ge2, be domains, and let f:(Ω,kΩ)→(Ω′,kΩ′)f:(\Omega,k_{\Omega})\to(\Omega',k_{\Omega'}) be a surjective isometry. Martin and Osgood proved that, after postcomposition with a reflection if necessary, ff is conformal [24 Theorem 2.6]. Consequently, if n=2n=2, then ff is either holomorphic or antiholomorphic, and if n≥3n\ge3, then, by Liouville’s theorem, ff is the restriction of a Möbius transformation of R‾n\overline{\mathbb{R}}^{n}.

Hästö’s similarity rigidity results

We now review some of Hästö’s results from [10]. Hästö [10 p. 317] separates the isometry problem into conformality, Möbius rigidity, and similarity rigidity. His Example 2.4 gives nonsimilarity isometries on sectors and punctured planes. Moreover, Hästö proved the following similarity rigidity results for

  • planar C2C^{2} domains other than half-planes, strictly convex domains, and strictly concave domains; see [10 Corollary 4.2], where strict concavity means strict convexity of the complement;

  • planar C3C^{3} domains other than half-planes; see [10 Theorem 4.3].

His method uses generalized Gaussian curvature and its gradient;

For planar C1C^{1} domains other than half-planes, Hästö also stated similarity rigidity under the additional assumption that the quasihyperbolic isometry is Möbius [10 Proposition 2.2], and deduced similarity rigidity for C1C^{1} domains other than half-spaces in dimensions n≥3n\ge3 [10 Corollary 2.3]. Richard pointed out a gap in this argument; see [27 Section 7.1, Example 7.1.2], for a detailed discussion. Richard’s inversion classification [27 Theorem 7.1.5], which applies in every dimension n≥2n\ge2, yields an alternative proof of both conclusions.

A domain Ω\Omega is a CmC^{m} domain if its boundary is locally the graph of a CmC^{m} function after a Euclidean change of coordinates.

Richard’s higher-dimensional classification

Richard classifies quasihyperbolic isometries induced by spherical inversion in every dimension n≥2n \ge2 [27 Theorem 7.1.5]. For n≥3n \ge3, conformality and Liouville’s theorem reduce the general problem to that inversion case; see [27 Theorem 7.1.6]. We state a reformulation of these results which gives a complete classification in dimensions n≥3n \ge3. For completeness, we include a short proof of this reformulation in our notation in § 3.1.1.

Theorem A. Let n≥3n \ge3, let Ω,Ω′⊊Rn\Omega, \Omega' \subsetneq\mathbb{R}^n be domains, and let f:(Ω,kΩ)→(Ω′,kΩ′)f : (\Omega, k_\Omega) \to(\Omega', k_{\Omega'}) be a surjective map. Then ff is an isometry if and only if either ff is the restriction of a Euclidean similarity or there exist p∈∂Ωp \in\partial\Omega, b∈Rnb \in\mathbb{R}^n, c>0c > 0, and Q∈O(n)Q \in O(n) such that

f(x)=b+cQx−p∣x−p∣2(x∈Ω),p+t(Ω−p)=Ω(t>0).(10)f(x) = b + cQ \frac{x-p}{|x-p|^2} \qquad(x \in\Omega), \qquad p + t(\Omega-p) = \Omega\quad(t > 0). \tag*{(10)}

In the second case,

Ω′=b+cQ(Ω−p).(11)\Omega' = b + cQ(\Omega-p). \tag*{(11)}

Consequences of the main result

The classification in Theorem 1.2 gives the following similarity rigidity consequences, including similarity rigidity for domains of finite inradius and C1C^1 domains. Recall that RΩ=sup⁡z∈ΩδΩ(z)R_\Omega= \sup_{z \in\Omega} \delta_\Omega(z) is the inradius of Ω\Omega.

Corollary 1.5. Let Ω,Ω′⊊C\Omega, \Omega' \subsetneq\mathbb{C} be domains, and let f:Ω→Ω′f : \Omega\to\Omega' be a surjective quasihyperbolic isometry. Then the following statements hold.

(1) If min⁡{RΩ,RΩ′}<∞\min\{R_\Omega, R_{\Omega'}\} < \infty, then ff is a Euclidean similarity. In particular, the conclusion holds if one of the domains is bounded.

(2) If Ω\Omega is multiply connected and Ω≠C∖{p}\Omega\ne\mathbb{C} \setminus\{p\} for every p∈Cp \in\mathbb{C}, then ff is a Euclidean similarity. Thus, a nonsimilarity quasihyperbolic isometry can occur only when Ω\Omega is simply connected or a punctured plane.

(3) If Ω\Omega is a C1C^1 domain other than a half-plane, then ff is a Euclidean similarity.

Corollary 1.5(1) and Theorem A yield the following similarity rigidity result in every dimension n≥2n \ge2.

Corollary 1.6. Let n≥2n \ge2, let Ω,Ω′⊊Rn\Omega, \Omega' \subsetneq\mathbb{R}^n be domains, and let f:(Ω,kΩ)→(Ω′,kΩ′)f : (\Omega, k_\Omega) \to(\Omega', k_{\Omega'}) be a surjective isometry. If min⁡{RΩ,RΩ′}<∞\min\{R_\Omega, R_{\Omega'}\} < \infty, then ff is the restriction of a Euclidean similarity.

The next consequence tells us that the quasihyperbolic metric determines the planar domain up to Euclidean similarity, even though an individual isometry can be non-Möbius.

Corollary 1.7. If Ω,Ω′⊊C\Omega, \Omega' \subsetneq\mathbb{C} are quasihyperbolically isometric, then Ω′=S(Ω)\Omega' = S(\Omega) for a Euclidean similarity SS, possibly reversing orientation.

Corollary 1.7 allows us to test similarity rigidity and Möbius rigidity using only quasihyperbolic self-isometries; see Definition 1.4.

Corollary 1.8. Let Ω⊊C\Omega\subsetneq\mathbb{C} be a domain. Then

Ω has similarity rigidity⟺Ω is self-similar rigid,(12)\Omega\text{ has similarity rigidity} \Longleftrightarrow\Omega\text{ is self-similar rigid}, \tag*{(12)}

and

Ω has Mo¨bius rigidity ⟺ Ω is self-Mo¨bius rigid.(13)\Omega\text{ has Möbius rigidity}\ \Longleftrightarrow\ \Omega\text{ is self-Möbius rigid}. \tag*{(13)}

Remark 1.9. By Theorem 1.2 and Corollary 1.8, Ω\Omega is self-similar rigid if and only if Ω∉E\Omega\notin\mathcal{E}, where

E={C∖{p}:p∈C}∪{p+Dσ,θ,α:p∈C, σ,θ∈R, 0<α≤2π},\mathcal{E}=\{\mathbb{C}\setminus\{p\}:p\in\mathbb{C}\}\cup\{p+D_{\sigma,\theta,\alpha}:p\in\mathbb{C},\ \sigma,\theta\in\mathbb{R},\ 0<\alpha\leq2\pi\},

and it is self-Möbius rigid if and only if Ω∉N\Omega\notin\mathcal{N}, where

N={p+Dσ,θ,α:p∈C, σ∈R∖{0}, θ∈R, 0<α≤2π}.\mathcal{N}=\{p+D_{\sigma,\theta,\alpha}:p\in\mathbb{C},\ \sigma\in\mathbb{R}\setminus\{0\},\ \theta\in\mathbb{R},\ 0<\alpha\leq2\pi\}.

We next explain how Theorem 1.2 extends the known results.

Remark 1.10. Theorem 1.2 and its corollaries extend the known results as follows.

  • Theorem 1.2 gives a classification without boundary regularity assumptions. Corollary 1.5(3) strengthens Hästö’s similarity rigidity for planar C3C^3 domains other than half-planes to the corresponding result for C1C^1 domains. It also removes the Möbius assumption from [10 Proposition 2.2].

  • Theorem 1.2 completely characterizes the orientation-preserving non-Möbius quasihyperbolic isometries through the complex-power maps in (6).

  • Contrary to the expectation expressed in [10 Example 2.4] and [13 Section 2], sectors and punctured planes do not exhaust the domains admitting nonsimilarity quasihyperbolic isometries: the logarithmic spiral sectors in Theorem 1.2 provide additional examples.

Application to pointed Gromov–Hausdorff convergence

Using the classification of Richard in dimensions n≥3n\geq3, Herron, Richard, and Snipes [17 Theorem 5.8 and Corollary 5.9] characterized pointed Gromov–Hausdorff convergence of quasihyperbolic domains in terms of uniform convergence of distance-to-complement functions on bounded Euclidean subsets of the limit domain. The normalized convergence result already appears in Richard’s thesis [27 Theorem 7.2.1], which also contains a corresponding planar result when the limit domain has C3C^3 boundary [27 Theorem 7.2.2]. The implication from boundary-distance convergence to pointed Gromov–Hausdorff convergence follows from the more general convergence theorem [17 Theorem 4.4]. Using Theorem 1.2, we establish the converse implication in dimension two without boundary regularity assumptions, thereby extending the characterization of Herron, Richard, and Snipes to arbitrary proper planar domains.

For a proper domain U⊊CU\subsetneq\mathbb{C}, define dU:C→[0,∞)d_U:\mathbb{C}\to[0,\infty) by

dU(z)=dist⁡(z,C∖U)(z∈C).(14)d_U(z)=\operatorname{dist}(z,\mathbb{C}\setminus U)\quad(z\in\mathbb{C}). \tag*{(14)}

Thus dU=δUd_U=\delta_U on UU. We obtain the following characterization.

Theorem 1.11. Let Ω,Ωj⊊C\Omega,\Omega_j\subsetneq\mathbb{C} be domains, and let a∈Ωa\in\Omega and aj∈Ωja_j\in\Omega_j. The following are equivalent.

(1) The pointed quasihyperbolic spaces satisfy

(Ωj,kΩj;aj)→pGH(Ω,kΩ;a).(15)(\Omega_j,k_{\Omega_j};a_j)\xrightarrow{\mathrm{pGH}}(\Omega,k_\Omega;a). \tag*{(15)}

(2) There exist Qj∈O(2)Q_j\in O(2) such that the similarities

Sj(z)=a+δΩ(a)δΩj(aj)Qj(z−aj)(z∈C)(16)S_j(z)=a+\frac{\delta_\Omega(a)}{\delta_{\Omega_j}(a_j)}Q_j(z-a_j)\qquad(z\in\mathbb{C}) \tag*{(16)}

satisfy

lim⁡j→∞sup⁡z∈Ω∩B(a,R)∣dSj(Ωj)(z)−δΩ(z)∣=0(R>0).(17)\lim_{j\to\infty}\sup_{z\in\Omega\cap B(a,R)}\left|d_{S_j(\Omega_j)}(z)-\delta_\Omega(z)\right|=0\qquad(R>0). \tag*{(17)}

The convergence in (17) is equivalent to dSj(Ωj)(z)→δΩ(z)d_{S_j(\Omega_j)}(z)\to\delta_\Omega(z) for every z∈Ωz\in\Omega.

Idea of the proof and organisation of the paper

We prove Theorem 1.2 in § 3.1. The idea of the proof is as follows. Let f:Ω→Ω′f:\Omega\to\Omega' be a surjective quasihyperbolic isometry. After postcomposition with complex conjugation if necessary, we may assume that ff is holomorphic. Write P=f′′/f′P=f''/f' and u(z)=log⁡∣f′(z)∣u(z)=\log|f'(z)|. We divide the proof into the cases P≡0P\equiv0 and P≠0P\ne0. If P≡0P\equiv0, then f(z)=az+bf(z)=az+b with a≠0a\ne0.

Suppose that P≠0P\ne0. In § 2, we use the boundary-distance identity

δΩ′(f(z))=∣f′(z)∣δΩ(z)(z∈Ω)(18)\delta_{\Omega'}(f(z))=|f'(z)|\delta_\Omega(z)\qquad(z\in\Omega) \tag*{(18)}

and the eikonal identities ∣∇δΩ∣=1|\nabla\delta_\Omega|=1 and ∣∇δΩ′∣=1|\nabla\delta_{\Omega'}|=1 at their respective differentiability points to derive the key differential identity

∇δΩ⋅∇u=−δΩ2∣∇u∣2.(19)\nabla\delta_\Omega\cdot\nabla u=-\frac{\delta_\Omega}{2}|\nabla u|^2. \tag*{(19)}

This holds at every differentiability point of δΩ\delta_\Omega. See Remark 2.4 for the relation with earlier work. Introducing q=−2/Pq=-2/P on G={z∈Ω:P(z)≠0}G=\{z\in\Omega:P(z)\ne0\} gives its vector form

∇δΩ⋅q=δΩ,(20)\nabla\delta_\Omega\cdot q=\delta_\Omega, \tag*{(20)}

which holds at every differentiability point of δΩ\delta_\Omega in GG.

In § 3, we prove the key observation that restricting the vector identity (20) to nearest-boundary segments and differentiating yields Re⁡q′=1\operatorname{Re}q'=1 on GG. Holomorphicity and the identity theorem then imply that PP has no zeros in Ω\Omega, so G=ΩG=\Omega, and that there exist σ∈R\sigma\in\mathbb{R} and p∈C∖Ωp\in\mathbb{C}\setminus\Omega such that

P(z)=−2(1+iσ)(z−p),q(z)=(1+iσ)(z−p)(z∈Ω).(21)P(z)=-\frac{2}{(1+i\sigma)(z-p)},\qquad q(z)=(1+i\sigma)(z-p)\qquad(z\in\Omega). \tag*{(21)}

In particular, qq is affine. Since G=ΩG=\Omega and δΩ\delta_\Omega is differentiable a.e. in Ω\Omega, (20) and (21) give

∇δΩ(z)⋅(1+iσ)(z−p)=δΩ(z)for a.e. z∈Ω.\nabla\delta_\Omega(z)\cdot(1+i\sigma)(z-p)=\delta_\Omega(z)\qquad\text{for a.e. }z\in\Omega.

By Lemma 3.2, it follows that p∈∂Ωp\in\partial\Omega and that Ω\Omega is invariant under the flow

Φt(z)=p+e(1+iσ)t(z−p),t∈R.\Phi_t(z)=p+e^{(1+i\sigma)t}(z-p),\qquad t\in\mathbb{R}.

When σ=0\sigma=0, this flow consists of dilations about pp. When σ≠0\sigma\ne0, it combines rotations and dilations about pp, and its orbits in Ω\Omega are logarithmic spirals. Connectedness then gives either Ω=C∖{p}\Omega=\mathbb{C}\setminus\{p\} or Ω=p+Dσ,θ,α\Omega=p+D_{\sigma,\theta,\alpha} for some θ∈R\theta\in\mathbb{R} and 0<α≤2π0<\alpha\le2\pi; see Lemma 3.3. If σ=0\sigma=0, integrating the formula for P=f′′/f′P=f''/f' in (21) gives (5). If σ≠0\sigma\ne0, a winding-number argument rules out Ω=C∖{p}\Omega= \mathbb{C} \setminus\{p\}, and integration gives (6). For the converse, we verify that the listed maps are conformal bijections onto their images and satisfy (18). We prove Theorem 1.11 in Section 5.

Acknowledgements

Acknowledgements. The author thanks Kai Rajala for his comments on an earlier version of this manuscript, which helped improve the exposition and clarity of the manuscript. In particular, his question about whether the differential identity (30) had appeared previously prompted the literature search discussed in Remark 2.4. The author thanks David A. Herron and Pekka Koskela for their encouragement and discussion.

The differential identity

For a planar domain DD, write D∗={z‾:z∈D}D^* = \{\overline{z} : z \in D\}. Thus D∗D^* denotes reflection, whereas D‾\overline{D} denotes Euclidean closure. We identify vectors in R2\mathbb{R}^2 with complex numbers and use v⋅w=Re⁡(v‾w)v \cdot w = \operatorname{Re}(\overline{v}w).

Lemma 2.1. Let Ω,Ω′⊊C\Omega, \Omega' \subsetneq\mathbb{C} be domains, and let f:Ω→Ω′f : \Omega\to\Omega' be a holomorphic bijection. Then ff is a quasihyperbolic isometry if and only if

δΩ′(f(z))=∣f′(z)∣δΩ(z)(z∈Ω).(22)\delta_{\Omega'}(f(z)) = |f'(z)|\delta_{\Omega}(z) \qquad(z \in\Omega). \tag*{(22)}

Proof. Recall the infinitesimal formula in [10 p. 321]

lim⁡w→zkΩ(z,w)∣z−w∣=1δΩ(z).(23)\lim_{w \to z} \frac{k_{\Omega}(z,w)}{|z-w|} = \frac{1}{\delta_{\Omega}(z)}. \tag*{(23)}

If ff is a quasihyperbolic isometry, (23) gives (22). Conversely, (22) gives

∫f∘γ∣dx∣δΩ′=∫γ∣f′∣ ∣dx∣δΩ′∘f=∫γ∣dx∣δΩ(24)\int_{f \circ\gamma} \frac{|dx|}{\delta_{\Omega'}} = \int_{\gamma} \frac{|f'|\,|dx|}{\delta_{\Omega'} \circ f} = \int_{\gamma} \frac{|dx|}{\delta_{\Omega}} \tag*{(24)}

for every rectifiable curve γ⊂Ω\gamma\subset\Omega. Taking infimum in (24) over all rectifiable curves joining xx and yy, and applying the same argument to f−1f^{-1}, gives (1).

We recall the standard properties of δΩ\delta_{\Omega}; see [22 Section 3]. The function δΩ\delta_{\Omega} is 1-Lipschitz and is differentiable at z∈Ωz \in\Omega if and only if zz has a unique nearest boundary point. If ξ∈∂Ω\xi\in\partial\Omega is this point, then ∇δΩ(z)=(z−ξ)/∣z−ξ∣\nabla\delta_{\Omega}(z) = (z-\xi)/|z-\xi|. Consequently,

δΩ∈Lip⁡(Ω),∣∇δΩ(z)∣=1at every differentiability point z∈Ω.(25)\delta_{\Omega} \in\operatorname{Lip}(\Omega), \qquad|\nabla\delta_{\Omega}(z)| = 1 \quad\text{at every differentiability point } z \in\Omega. \tag*{(25)}

By Rademacher’s theorem, δΩ\delta_{\Omega} is differentiable a.e. in Ω\Omega. Our discussion is closely related to the geometry of the medial axis of Ω\Omega. For related work on medial axis and the regularity of distance functions we refer to [4, 5, 6].

The following lemma shows that δΩ\delta_{\Omega} is differentiable at every interior point of a segment joining a point of Ω\Omega to one of its nearest boundary points, and that its gradient is constant along this segment.

Lemma 2.2. Let Ω⊊Rn\Omega\subsetneq\mathbb{R}^n, n≥2n \ge2 be a domain. Let x∈Ωx \in\Omega, set r=δΩ(x)r = \delta_{\Omega}(x), and choose ξ∈∂Ω\xi\in\partial\Omega with ∣x−ξ∣=r|x-\xi| = r. Put ν=(x−ξ)/r\nu= (x-\xi)/r. For every 0<t<r0 < t < r, the point zt=ξ+tνz_t = \xi+ t\nu has ξ\xi as its unique nearest boundary point. Consequently,

zt∈Ω,δΩ(zt)=t,∇δΩ(zt)=ν(0<t<r).(26)z_t \in\Omega, \qquad\delta_{\Omega}(z_t) = t, \qquad\nabla\delta_{\Omega}(z_t) = \nu\qquad(0 < t < r). \tag*{(26)}

:::`

Proof. Since ∣x−zt∣=r−t<r|x-z_t|=r-t<r, we have zt∈B(x,r)⊂Ωz_t\in B(x,r)\subset\Omega. For every η∈∂Ω\eta\in\partial\Omega,

∣zt−η∣≥∣x−η∣−∣x−zt∣≥r−(r−t)=t.(27)|z_t-\eta|\ge|x-\eta|-|x-z_t|\ge r-(r-t)=t. \tag*{(27)}

Equality in (27) requires ∣x−η∣=r|x-\eta|=r and equality in the triangle inequality. Thus x−ztx-z_t and zt−ηz_t-\eta are nonnegative multiples of ν\nu, and zt−η=tνz_t-\eta=t\nu. Hence η=ξ\eta=\xi. The distance-function properties imply (26). □\square

Now, let f:Ω→Ω′f:\Omega\to\Omega' be a holomorphic quasihyperbolic isometry and write

u(z)=log⁡∣f′(z)∣,P(z)=f′′(z)f′(z).(28)u(z)=\log|f'(z)|,\qquad P(z)=\frac{f''(z)}{f'(z)}. \tag*{(28)}

Since ff is injective and holomorphic, f′(z)≠0f'(z)\ne0 for every z∈Ωz\in\Omega. Thus uu and PP are globally defined, and

Δu=0,P=ux−iuyin Ω.(29)\Delta u=0,\qquad P=u_x-iu_y\quad\text{in }\Omega. \tag*{(29)}

Comparing the eikonal identities (25) for δΩ\delta_\Omega and δΩ′\delta_{\Omega'} through the boundary-distance identity (22), we obtain the following differential identity.

Lemma 2.3. Let Ω,Ω′⊊C\Omega,\Omega'\subsetneq\mathbb{C} be domains, and let f:Ω→Ω′f:\Omega\to\Omega' be a surjective holomorphic quasihyperbolic isometry. Then at every differentiability point z∈Ωz\in\Omega of δΩ\delta_\Omega,

∇δΩ(z)⋅∇u(z)=−δΩ(z)2∣∇u(z)∣2.(30)\nabla\delta_\Omega(z)\cdot\nabla u(z)=-\frac{\delta_\Omega(z)}{2}|\nabla u(z)|^2. \tag*{(30)}

Let G={z∈Ω:P(z)≠0}G=\{z\in\Omega:P(z)\ne0\} and define q=−2/Pq=-2/P on GG. Then, at every differentiability point z∈Gz\in G of δΩ\delta_\Omega,

∇δΩ(z)⋅q(z)=δΩ(z).(31)\nabla\delta_\Omega(z)\cdot q(z)=\delta_\Omega(z). \tag*{(31)}

Proof. Fix a differentiability point z∈Ωz\in\Omega of δΩ\delta_\Omega. By (22), the identity δΩ′=(euδΩ)∘f−1\delta_{\Omega'}=(e^u\delta_\Omega)\circ f^{-1} holds on the open set f(Ω)f(\Omega). Hence δΩ′\delta_{\Omega'} is differentiable at f(z)f(z). By (25), we have ∣∇δΩ(z)∣=∣∇δΩ′(f(z))∣=1|\nabla\delta_\Omega(z)|=|\nabla\delta_{\Omega'}(f(z))|=1. The chain rule and conformality therefore give

∣f′(z)∣=∣∇(δΩ′∘f)(z)∣=eu(z)∣∇δΩ(z)+δΩ(z)∇u(z)∣.(32)|f'(z)|=|\nabla(\delta_{\Omega'}\circ f)(z)|=e^{u(z)}|\nabla\delta_\Omega(z)+\delta_\Omega(z)\nabla u(z)|. \tag*{(32)}

Since eu(z)=∣f′(z)∣e^{u(z)}=|f'(z)|, dividing (32) by eu(z)e^{u(z)} and squaring gives

1=1+2δΩ(z)∇δΩ(z)⋅∇u(z)+δΩ(z)2∣∇u(z)∣2.(33)1=1+2\delta_\Omega(z)\nabla\delta_\Omega(z)\cdot\nabla u(z)+\delta_\Omega(z)^2|\nabla u(z)|^2. \tag*{(33)}

Since δΩ(z)>0\delta_\Omega(z)>0, (33) implies (30). If z∈Gz\in G, then (29) gives

q(z)=−2P(z)=−2P(z)‾∣P(z)∣2=−2∇u(z)∣∇u(z)∣2.q(z)=-\frac{2}{P(z)}=-\frac{2\overline{P(z)}}{|P(z)|^2}=-\frac{2\nabla u(z)}{|\nabla u(z)|^2}.

Thus (30) implies (31). □\square

Remark 2.4. After completing the first version of this manuscript, we learned that Graham and Witten had obtained the same differential equation as (30) in their study of geodesic defining functions for conformally compact Einstein metrics. For two such defining functions related by r^=eωr\hat r=e^\omega r, their equation [9 Lemma 2.2, equation (2.8)] can be written as

⟨dr,dω⟩gˉ=−12∣dω∣gˉ2.\langle dr,d\omega\rangle_{\bar g}=-\frac{1}{2}|d\omega|_{\bar g}^{2}.

where gˉ\bar g is the compactified metric associated with rr. In our setting, gˉ\bar g is the Euclidean metric, and the corresponding functions are r=δΩr=\delta_\Omega, r^=δΩ′∘f\hat r=\delta_{\Omega'}\circ f,

and ω=log⁡∣f′∣\omega= \log|f'|. We also note that both (30) and [9 equation (2.8)] follow by differentiating the relation between the two functions and expanding the squared norm using their respective eikonal identities.

We are not aware of an earlier use of (30) in the study of quasihyperbolic isometries.

Classification of the maps and the domains

In this section we prove Theorem 1.2. We divide its proof into several lemmas.

Lemma 3.1. Let Ω,Ω′⊊C\Omega, \Omega' \subsetneq\mathbb{C} be domains, and let f:Ω→Ω′f : \Omega\to\Omega' be a surjective holomorphic quasihyperbolic isometry. Suppose that P=f′′/f′≠0P = f''/f' \ne0. Then PP has no zeros in Ω\Omega, and there exist σ∈R\sigma\in\mathbb{R} and p∈C∖Ωp \in\mathbb{C} \setminus\Omega such that

P(z)=−2(1+iσ)(z−p),q(z):=−2P(z)=(1+iσ)(z−p)(z∈Ω).(34)P(z) = -\frac{2}{(1+i\sigma)(z-p)}, \qquad q(z) := -\frac{2}{P(z)} = (1+i\sigma)(z-p) \qquad(z \in\Omega). \tag*{(34)}

Proof. Let G={P≠0}G = \{P \ne0\} and q=−2/Pq = -2/P on GG. Fix x∈Gx \in G and choose rr, ξ\xi, and nn as in Lemma 2.2. There exists t0∈(0,r)t_0 \in(0,r) such that ξ+tn∈G\xi+ tn \in G for t0<t<rt_0 < t < r. Equations (26) and (31) imply

Re⁡(q(ξ+tn)n‾)=t(t0<t<r).(35)\operatorname{Re}(q(\xi+ tn)\overline{n}) = t \qquad(t_0 < t < r). \tag*{(35)}

Differentiating (35) with respect to tt gives

Re⁡(q′(ξ+tn)n‾n)=Re⁡q′(ξ+tn)=1.(36)\operatorname{Re}(q'(\xi+ tn)\overline{n}n) = \operatorname{Re} q'(\xi+ tn) = 1. \tag*{(36)}

Letting t↑rt \uparrow r in (36), we obtain

Re⁡q′(x)=1(x∈G).(37)\operatorname{Re} q'(x) = 1 \qquad(x \in G). \tag*{(37)}

Choose a disk V⊂GV \subset G. Since q′q' is holomorphic, (37) implies that, for some σ∈R\sigma\in\mathbb{R} and b∈Cb \in\mathbb{C},

q′(z)=1+iσ,q(z)=(1+iσ)z+b(z∈V).(38)q'(z) = 1 + i\sigma, \qquad q(z) = (1+i\sigma)z+b \qquad(z \in V). \tag*{(38)}

By (38), the holomorphic function ((1+iσ)z+b)P(z)+2((1+i\sigma)z+b)P(z)+2 vanishes on VV. Therefore by the identity theorem we have

((1+iσ)z+b)P(z)=−2(z∈Ω).((1+i\sigma)z+b)P(z) = -2 \qquad(z \in\Omega).

This gives (34). In particular, PP has no zeros and p=−b/(1+iσ)∉Ωp=-b/(1+i\sigma)\notin\Omega.

Lemma 3.2. Let Ω⊊C\Omega\subsetneq\mathbb{C} be a domain, let p∈Cp \in\mathbb{C}, and let σ∈R\sigma\in\mathbb{R}. Suppose

∇δΩ(z)⋅(1+iσ)(z−p)=δΩ(z)for a.e. z∈Ω.(39)\nabla\delta_{\Omega}(z)\cdot(1+i\sigma)(z-p)=\delta_{\Omega}(z) \qquad\text{for a.e. } z\in\Omega. \tag*{(39)}

Then p∈∂Ωp\in\partial\Omega and, for Φt(z)=p+e(1+iσ)t(z−p)\Phi_t(z)=p+e^{(1+i\sigma)t}(z-p),

Φt(Ω)=Ω,δΩ(Φt(z))=etδΩ(z)(z∈Ω, t∈R).(40)\Phi_t(\Omega)=\Omega,\qquad\delta_{\Omega}(\Phi_t(z))=e^t\delta_{\Omega}(z) \qquad(z\in\Omega,\ t\in\mathbb{R}). \tag*{(40)}

Proof. Fix x∈Ωx\in\Omega and choose an open neighbourhood V∋xV\ni x with V‾⊂Ω\overline{V}\subset\Omega. Since Φt→id⁡\Phi_t\to\operatorname{id} uniformly on V‾\overline{V} as t→0t\to0, there exists h>0h>0 such that Φt(V)⊂Ω\Phi_t(V)\subset\Omega for ∣t∣<h|t|<h.

Let N⊂ΩN\subset\Omega be a Borel null set outside which δΩ\delta_{\Omega} is differentiable and (39) holds. For each fixed tt, the set Φ−t(N)\Phi_{-t}(N) has Lebesgue measure zero, since

Φ−t\Phi_{-t} is a Euclidean similarity. Therefore, by Fubini’s theorem it follows that for a.e. z∈Vz \in V, Φt(z)∉N\Phi_t(z) \notin N for a.e. t∈(−h,h)t \in(-h,h). The chain rule and (39) give

ddt(e−tδΩ(Φt(z)))=e−t(∇δΩ(Φt(z))⋅(1+iσ)(Φt(z)−p)−δΩ(Φt(z)))=0(41)\frac{\mathrm{d}}{\mathrm{d}t}\left(e^{-t}\delta_{\Omega}(\Phi_t(z))\right)=e^{-t}\left(\nabla\delta_{\Omega}(\Phi_t(z))\cdot(1+i\sigma)(\Phi_t(z)-p)-\delta_{\Omega}(\Phi_t(z))\right)=0 \tag*{(41)}

for a.e. t∈(−h,h)t \in(-h,h).

Since δΩ\delta_{\Omega} is 1-Lipschitz and Φt\Phi_t is smooth, for each fixed z∈Vz \in V, the function t↦e−tδΩ(Φt(z))t \mapsto e^{-t}\delta_{\Omega}(\Phi_t(z)) is locally Lipschitz. Thus (41) implies that this function is constant for a.e. z∈Vz \in V. Continuity in zz then gives

δΩ(Φt(z))=etδΩ(z)(z∈V, ∣t∣<h).(42)\delta_{\Omega}(\Phi_t(z))=e^t\delta_{\Omega}(z)\qquad(z\in V,\ |t|<h). \tag*{(42)}

Fix z∈Ωz \in\Omega and let IzI_z be the component of {t∈R:Φt(z)∈Ω}\{t\in\mathbb{R}:\Phi_t(z)\in\Omega\} containing 0. For every τ∈Iz\tau\in I_z, apply (42) at Φτ(z)\Phi_\tau(z). Using Φs∘Φτ=Φτ+s\Phi_s\circ\Phi_\tau=\Phi_{\tau+s}, we obtain

δΩ(Φτ+s(z))=esδΩ(Φτ(z))\delta_{\Omega}(\Phi_{\tau+s}(z))=e^s\delta_{\Omega}(\Phi_\tau(z))

for all sufficiently small ∣s∣|s|.

Hence t↦e−tδΩ(Φt(z))t\mapsto e^{-t}\delta_{\Omega}(\Phi_t(z)) is locally constant on the interval IzI_z. It is therefore constant on IzI_z, and evaluation at t=0t=0 gives

δΩ(Φt(z))=etδΩ(z)(t∈Iz).(43)\delta_{\Omega}(\Phi_t(z))=e^t\delta_{\Omega}(z)\qquad(t\in I_z). \tag*{(43)}

If IzI_z had a finite endpoint TT, continuity of the distance to the closed complement and (43) would imply

dist⁡(ΦT(z),C∖Ω)=lim⁡t→Tt∈IzδΩ(Φt(z))=eTδΩ(z)>0.(44)\operatorname{dist}(\Phi_T(z),\mathbb{C}\setminus\Omega)=\lim_{\substack{t\to T\\t\in I_z}}\delta_{\Omega}(\Phi_t(z))=e^T\delta_{\Omega}(z)>0. \tag*{(44)}

Thus ΦT(z)∈Ω\Phi_T(z)\in\Omega. Since Ω\Omega is open, this contradicts the definition of IzI_z. Hence Iz=RI_z=\mathbb{R}. Since Φ−t=Φt−1\Phi_{-t}=\Phi_t^{-1}, (43) proves (40).

Finally, fix z∈Ωz\in\Omega. Since Φt(z)→p\Phi_t(z)\to p as t→−∞t\to-\infty, we have p∈Ω‾p\in\overline{\Omega}. Moreover,

dist⁡(p,C∖Ω)=lim⁡t→−∞δΩ(Φt(z))=lim⁡t→−∞etδΩ(z)=0.\operatorname{dist}(p,\mathbb{C}\setminus\Omega)=\lim_{t\to-\infty}\delta_{\Omega}(\Phi_t(z))=\lim_{t\to-\infty}e^t\delta_{\Omega}(z)=0.

Therefore p∈∂Ωp\in\partial\Omega. □

Lemma 3.3. Let p∈Cp\in\mathbb{C}, σ∈R\sigma\in\mathbb{R}, and Ω⊊C\Omega\subsetneq\mathbb{C} be a domain. If Ω\Omega satisfies Φt(Ω)=Ω\Phi_t(\Omega)=\Omega for every t∈Rt\in\mathbb{R}, then either Ω=C∖{p}\Omega=\mathbb{C}\setminus\{p\} or

Ω=p+Dσ,θ,α(45)\Omega=p+D_{\sigma,\theta,\alpha} \tag*{(45)}

for some θ∈R\theta\in\mathbb{R}, 0<α≤2π0<\alpha\le2\pi.

Proof. We first prove that pp cannot be in Ω\Omega. If p∈Ωp\in\Omega, choose r>0r>0 with B(p,r)⊂ΩB(p,r)\subset\Omega. Since Φt(Ω)=Ω\Phi_t(\Omega)=\Omega for every t∈Rt\in\mathbb{R}, it follows that B(p,etr)⊂ΩB(p,e^t r)\subset\Omega for every t∈Rt\in\mathbb{R}, contrary to Ω≠C\Omega\ne\mathbb{C}. Thus p∉Ωp\notin\Omega. After translation, assume p=0p=0. Define S1={ζ∈C:∣ζ∣=1}S^1=\{\zeta\in\mathbb{C}:|\zeta|=1\} and the homeomorphism H:C∖{0}→R×S1H:\mathbb{C}\setminus\{0\}\to\mathbb{R}\times S^1 by

H(z)=(log⁡∣z∣,z∣z∣e−iσlog⁡∣z∣),H(z)=\left(\log|z|,\frac{z}{|z|}e^{-i\sigma\log|z|}\right),
H−1(s,ζ)=e(1+iσ)sζ.(46)H^{-1}(s,\zeta)=e^{(1+i\sigma)s}\zeta. \tag*{(46)}

The invariance Φt(Ω)=Ω\Phi_t(\Omega)=\Omega for every t∈Rt\in\mathbb{R}, together with (46), gives

H(Ω)=R×V,(47)H(\Omega)=\mathbb{R}\times V, \tag*{(47)}

where VV is a nonempty open subset of S1S^1. Since Ω\Omega is connected, so is H(Ω)H(\Omega), and hence its projection VV onto S1S^1. If V=S1V=S^1, then Ω=C∖{0}\Omega=\mathbb{C}\setminus\{0\}. Otherwise,

V={eiβ:θ<β<θ+α}(48)V=\{e^{i\beta}:\theta<\beta<\theta+\alpha\} \tag*{(48)}

for some θ∈R\theta\in\mathbb{R} and 0<α≤2π0<\alpha\le2\pi.

By equations (46), (47), and (48) it follows that Ω=Dσ,θ,α\Omega= D_{\sigma,\theta,\alpha}. □

Proof of the main result

Proof of Theorem 1.2. By [24 Theorem 2.6], after postcomposition with a reflection if necessary, we may assume that ff is holomorphic. We consider two cases: P≡0P \equiv0 and P≠0P \ne0. If P≡0P \equiv0, then f(z)=az+bf(z) = az + b with a≠0a \ne0. Thus, we have Theorem 1.2(1).

If P≠0P \ne0, we first provide the domain classification. Since P≠0P \ne0, G≠∅G \ne\varnothing, and by Lemma 3.1, there exist σ∈R\sigma\in\mathbb{R} and p∉Ωp \notin\Omega such that (34) holds. In particular, (34) gives G=ΩG = \Omega. Since δΩ\delta_\Omega is 1-Lipschitz, it is differentiable a.e. in Ω\Omega. Thus, by Lemma 2.3, (39) holds for a.e. z∈Ωz \in\Omega. Hence, by Lemma 3.2, we have p∈∂Ωp \in\partial\Omega and (40). Finally, Lemma 3.3 gives the classification of Ω\Omega, i.e., Ω\Omega is either C∖{p}\mathbb{C} \setminus\{p\} or p+Dσ,θ,αp + D_{\sigma,\theta,\alpha} for some θ∈R\theta\in\mathbb{R} and 0<α≤2π0 < \alpha\le2\pi.

Now we study two cases: σ=0\sigma= 0 and σ≠0\sigma\ne0. If σ=0\sigma= 0, (34) gives

(f′(z)(z−p)2)′=0,f′(z)=C(z−p)2,f(z)=B−Cz−p(49)(f'(z)(z-p)^2)' = 0,\qquad f'(z) = \frac{C}{(z-p)^2},\qquad f(z) = B - \frac{C}{z-p} \tag*{(49)}

for some C≠0C \ne0. This proves Theorem 1.2(2).

Assume σ≠0\sigma\ne0. We show that Ω≠C∖{p}\Omega\ne\mathbb{C} \setminus\{p\}. If Ω=C∖{p}\Omega= \mathbb{C} \setminus\{p\}, then, for the positively oriented circle γr={z:∣z−p∣=r}\gamma_r = \{z : |z-p| = r\},

Z∋12πi∫γrf′′(z)f′(z) dz=−21+iσ∉R,(50)\mathbb{Z} \ni\frac{1}{2\pi i}\int_{\gamma_r}\frac{f''(z)}{f'(z)}\,\mathrm{d}z = -\frac{2}{1+i\sigma} \notin\mathbb{R}, \tag*{(50)}

a contradiction. Here the integral divided by 2πi2\pi i is the winding number of f′∘γrf' \circ\gamma_r around 0. Hence Ω=p+Dσ,θ,α\Omega= p + D_{\sigma,\theta,\alpha}. For simplicity, write L=Lσ,θ,αL = L_{\sigma,\theta,\alpha} and take F(z)=exp⁡(λσL(z−p))F(z) = \exp(\lambda_\sigma L(z-p)), then

F′′(z)F′(z)=λσ−1z−p=−2(1+iσ)(z−p)=f′′(z)f′(z).(51)\frac{F''(z)}{F'(z)} = \frac{\lambda_\sigma-1}{z-p} = -\frac{2}{(1+i\sigma)(z-p)} = \frac{f''(z)}{f'(z)}. \tag*{(51)}

Thus (f′/F′)′=0(f'/F')' = 0 and f=AF+Bf = AF+B for some A≠0A \ne0. This proves (1.6).

Conversely since Euclidean similarities are quasihyperbolic isometries, it suffices to prove the converse in the sector cases with p=B=0p = B = 0 and A=1A = 1. Write

D=Dσ,θ,α,U=Uσ,θ,α,L=(exp⁡∣U)−1,F(z)=exp⁡(λσL(z)),(52)D = D_{\sigma,\theta,\alpha},\qquad U = U_{\sigma,\theta,\alpha},\qquad L = (\exp|_U)^{-1},\qquad F(z) = \exp(\lambda_\sigma L(z)), \tag*{(52)}

where σ=0\sigma= 0 is allowed. For z∈Dz \in D, write w=L(z)=s+itw = L(z) = s + it. It follows that

Im⁡(λσw)+σRe⁡(λσw)=−(t−σs),λσU=U∗.(53)\operatorname{Im}(\lambda_\sigma w) + \sigma\operatorname{Re}(\lambda_\sigma w) = -(t-\sigma s),\qquad\lambda_\sigma U = U^*. \tag*{(53)}

The exponential is injective on UU and U∗U^*, so F:D→D∗F:D \to D^* is a conformal bijection. Define

τ=−2(s+σt)1+σ2,λσw−w‾=(1−iσ)τ,F(z)=e(1−iσ)τz‾.(54)\tau= -\frac{2(s+\sigma t)}{1+\sigma^2},\qquad\lambda_\sigma w-\overline{w} = (1-i\sigma)\tau,\qquad F(z) = e^{(1-i\sigma)\tau}\overline{z}. \tag*{(54)}

Since U+(1+iσ)v=UU+(1+i\sigma)v=U for v∈Rv \in\mathbb{R},

e(1−iσ)vD∗=D∗,δD∗(e(1−iσ)vζ)=evδD∗(ζ)(ζ∈D∗, v∈R).(55)e^{(1-i\sigma)v}D^* = D^*,\qquad\delta_{D^*}\left(e^{(1-i\sigma)v}\zeta\right) = e^v\delta_{D^*}(\zeta)\quad(\zeta\in D^*,\ v \in\mathbb{R}). \tag*{(55)}

Equations (54) and (55) imply

δD∗(F(z))=eτδD∗(z‾)=eτδD(z),\delta_{D^*}(F(z)) = e^\tau\delta_{D^*}(\overline{z}) = e^\tau\delta_D(z),
∣F′(z)∣=∣λσF(z)z∣=eRe⁡(λσw)−s=eτ.(56)|F'(z)| = \left|\lambda_\sigma\frac{F(z)}{z}\right| = e^{\operatorname{Re}(\lambda_\sigma w)-s} = e^\tau. \tag*{(56)}

Lemma 2.1 applies to (56). For the punctured plane, we have

δC∖{0}(1/z)=1∣z∣=∣1z∣δC∖{0}(z),\delta_{\mathbb{C}\setminus\{0\}}(1/z)=\frac{1}{|z|}=\left|\frac{1}{z}\right|\delta_{\mathbb{C}\setminus\{0\}}(z),

so Lemma 2.1 proves the remaining case. Finally, substituting (51) into (7) gives (8). □

Quasihyperbolic isometry classification in dimensions n≥3n \ge3.

Proof of Theorem A. By [24 Theorem 2.6] and Liouville’s theorem, ff is the restriction of a Möbius transformation. If ff is not a similarity, its standard decomposition gives

f(x)=b+cQJ(x−p),J(y)=y∣y∣2,f(x)=b+cQJ(x-p),\qquad J(y)=\frac{y}{|y|^{2}},

where p∉Ωp\notin\Omega, b∈Rnb\in\mathbb{R}^{n}, c>0c>0, and Q∈O(n)Q\in O(n). Since similarities are quasihyperbolic isometries, J:Ω−p→J(Ω−p)J:\Omega-p\to J(\Omega-p) is a quasihyperbolic isometry. Thus [27 Theorem 7.1.5] gives p∈∂Ωp\in\partial\Omega and p+t(Ω−p)=Ωp+t(\Omega-p)=\Omega for every t>0t>0.

Conversely, suppose ff has the form (10) and write D=Ω−pD=\Omega-p. The identities tD=DtD=D and δD(tx)=tδD(x)\delta_D(tx)=t\delta_D(x) for t>0t>0 imply that J(D)=DJ(D)=D and

δD(J(x))=δD(x)∣x∣2,∣DJ(x)v∣=∣v∣∣x∣2(x∈D, v∈Rn).(57)\delta_D(J(x))=\frac{\delta_D(x)}{|x|^{2}},\qquad|DJ(x)v|=\frac{|v|}{|x|^{2}}\quad(x\in D,\ v\in\mathbb{R}^{n}). \tag*{(57)}

By (57), JJ preserves quasihyperbolic lengths. Since J−1=JJ^{-1}=J, it is a quasihyperbolic isometry of DD. Composition with similarities proves the converse, and J(D)=DJ(D)=D gives (11). The similarity case follows directly from the definition. □

Proof of Corollary 1.6. For n=2n=2, it follows from Corollary 1.5(1). For n≥3n\ge3, replacing ff by f−1f^{-1} if necessary, assume RΩ<∞R_{\Omega}<\infty. If ff is not a similarity, (10) gives

δΩ(p+t(x−p))=tδΩ(x)⟶∞(t→∞)\delta_{\Omega}(p+t(x-p))=t\delta_{\Omega}(x)\longrightarrow\infty\quad(t\to\infty)

for every x∈Ωx\in\Omega, a contradiction. □

Consequences and examples

Proof of Corollary 1.5(1). Suppose ff is not a similarity. Theorem 1.2 and (40) give, for any fixed z∈Ωz\in\Omega,

RΩ≥δΩ(Φt(z))=etδΩ(z)⟶∞(t→∞).(58)R_{\Omega}\ge\delta_{\Omega}(\Phi_t(z))=e^{t}\delta_{\Omega}(z)\longrightarrow\infty\quad(t\to\infty). \tag*{(58)}

The inverse f−1f^{-1} is also a nonsimilarity isometry, so the same argument gives RΩ′=∞R_{\Omega'}=\infty. □

Proof of Corollary 1.5(2). Each Dσ,θ,αD_{\sigma,\theta,\alpha} is conformally equivalent to the strip Uσ,θ,αU_{\sigma,\theta,\alpha} and hence is simply connected. The conclusion follows from Theorem 1.2. □

Proof of Corollary 1.5(3). Suppose that ff is not a similarity. By Theorem 1.2, Ω\Omega is a punctured plane or p+Dσ,θ,αp+D_{\sigma,\theta,\alpha}. The punctured plane has an isolated boundary point, while for σ≠0\sigma\ne0 the boundary has no tangent line at pp. For σ=0\sigma=0, the sector boundary is locally a C1C^{1} graph at pp if and only if α=π\alpha=\pi. Thus Ω\Omega is a half-plane, a contradiction. □

Proof of Corollary 1.7. In the sector cases, (53) and Theorem 1.2 give, for the holomorphic normalization,

Ω′=B+A(Ω−p)∗.(59)\Omega' = B + A(\Omega-p)^*. \tag*{(59)}

The map z↦B+Az−p‾z \mapsto B + A\overline{z-p} is a Euclidean similarity realizing (59). The punctured-plane case follows from f(C∖{p})=C∖{B}f(\mathbb{C} \setminus\{p\}) = \mathbb{C} \setminus\{B\}. The similarity case and the target reflection preserve the conclusion. □\square

Proof of Corollary 1.8. It is clear that each rigidity property implies the corresponding self-rigidity property. Conversely, let f:(Ω,kΩ)→(f(Ω),kf(Ω))f:(\Omega,k_{\Omega})\to(f(\Omega),k_{f(\Omega)}) be a quasihyperbolic isometry, where f(Ω)f(\Omega) is a proper planar domain. By Corollary 1.7, choose a Euclidean similarity SS with S(Ω)=f(Ω)S(\Omega)=f(\Omega). Then g=S−1∘fg=S^{-1}\circ f is a surjective quasihyperbolic self-isometry of Ω\Omega. If Ω\Omega is self-similar rigid, then f=S∘gf=S\circ g is a similarity. If Ω\Omega is self-Möbius rigid, then f=S∘gf=S\circ g is Möbius. □\square

Example 4.1. Let

U={s+it:0<t−s<π},Ω=exp⁡U,U=\{s+it:0<t-s<\pi\},\qquad\Omega=\exp U,
L=(exp⁡∣U)−1,f(z)=exp⁡(iL(z)).(60)L=(\exp|_U)^{-1},\qquad f(z)=\exp(iL(z)). \tag*{(60)}

Since iU=U∗iU=U^*, the map f:Ω→Ω∗f:\Omega\to\Omega^* is a conformal bijection. For L(z)=s+itL(z)=s+it and τ=−s−t\tau=-s-t,

f(z)=e(1−i)τz,f′(z)=if(z)z,∣f′(z)∣=eτ,f(z)=e^{(1-i)\tau}z,\qquad f'(z)=\frac{if(z)}{z},\qquad|f'(z)|=e^\tau,
δΩ∗(f(z))=eτδΩ∗(z)=∣f′(z)∣δΩ(z).(61)\delta_{\Omega^*}(f(z))=e^\tau\delta_{\Omega^*}(z)=|f'(z)|\delta_{\Omega}(z). \tag*{(61)}

Lemma 2.1 and (61) show that ff is a quasihyperbolic isometry. However,

f′′f′=i−1z,Sf(z)=−i−1z2−(i−1)22z2=1z2≠0.(62)\frac{f''}{f'}=\frac{i-1}{z},\qquad S_f(z)=-\frac{i-1}{z^2}-\frac{(i-1)^2}{2z^2}=\frac{1}{z^2}\ne0. \tag*{(62)}

Thus ff is not Möbius. The same construction also gives an explicit failure of self-Möbius rigidity. Indeed, g(z)=f(z)‾g(z)=\overline{f(z)} is an anticonformal quasihyperbolic self-isometry of Ω\Omega. Moreover,

L(g(z))=−iL(z)‾,g∘g=id⁡Ω.(63)L(g(z))=-i\overline{L(z)},\qquad g\circ g=\operatorname{id}_{\Omega}. \tag*{(63)}

If gg were Möbius, then so would be f=g‾f=\overline{g}, contradicting (62). Hence, gg is not Möbius. By (58), RΩ=RΩ∗=∞R_{\Omega}=R_{\Omega^*}=\infty.

Proof of Theorem 1.11

We use pointed Gromov–Hausdorff convergence in the sense of [17 Section 2.3]. We first strengthen the similarity conclusion for quasihyperbolically isometric domains by preserving a prescribed base point.

Lemma 5.1. Let Ω,Ω′⊊C\Omega,\Omega'\subsetneq\mathbb{C} be domains, let f:(Ω,kΩ)→(Ω′,kΩ′)f:(\Omega,k_{\Omega})\to(\Omega',k_{\Omega'}) be a surjective isometry, and let a∈Ωa\in\Omega. Then there exist a Euclidean similarity S:C→CS:\mathbb{C}\to\mathbb{C} and c>0c>0 such that ∣S(z)−S(w)∣=c∣z−w∣|S(z)-S(w)|=c|z-w| for all z,w∈Cz,w\in\mathbb{C}, and

S(Ω)=Ω′,S(a)=f(a),c=δΩ′(f(a))δΩ(a).(64)S(\Omega)=\Omega',\qquad S(a)=f(a),\qquad c=\frac{\delta_{\Omega'}(f(a))}{\delta_{\Omega}(a)}. \tag*{(64)}

Proof. The assertion is preserved under pre- and postcomposition by Euclidean similarities. If ff is a similarity, take S=fS=f. Otherwise, by Theorem 1.2, we may assume that ff is holomorphic and that p=B=0p=B=0 and A=1A=1. If Ω=C∖{0}\Omega=\mathbb{C}\setminus\{0\} and f(z)=1/zf(z)=1/z, take S(z)=z/∣a∣2S(z)=z/\lvert a\rvert^2. In the remaining cases, Ω=Dσ,θ,α\Omega=D_{\sigma,\theta,\alpha} and ff has the form (52), with σ=0\sigma=0 allowed. Write

w=Lσ,θ,α(a),t0=−2Re⁡w1+iσ,S(z)=e(1−iσ)t0z(z∈C).w=L_{\sigma,\theta,\alpha}(a),\qquad t_0=-2\operatorname{Re}\frac{w}{1+i\sigma},\qquad S(z)=e^{(1-i\sigma)t_0}z\quad(z\in\mathbb{C}).

Equations (54) and (55) give S(a)=f(a)S(a)=f(a) and S(Ω)=f(Ω)S(\Omega)=f(\Omega). Finally, if c>0c>0 is the scale factor of SS, then δΩ′(f(a))=cδΩ(a)\delta_{\Omega'}(f(a))=c\delta_{\Omega}(a).

In particular, for proper planar domains D,UD,U containing 00, Lemma 5.1 gives

(D,kD;0)≅(U,kU;0),δD(0)=δU(0)}⟹Q(D)=Ufor some Q∈O(2),(65)\left. \begin{aligned} (D,k_D;0)&\cong(U,k_U;0),\\ \delta_D(0)&=\delta_U(0) \end{aligned} \right\} \Longrightarrow Q(D)=U\quad\text{for some }Q\in O(2), \tag*{(65)}

where ≅\cong denotes pointed isometry.

Proof of Theorem 1.11. Write r=δΩ(a)r=\delta_{\Omega}(a), rj=δΩj(aj)r_j=\delta_{\Omega_j}(a_j),

T(z)=z−ar,Tj(z)=z−ajrj,U=T(Ω),Uj=Tj(Ωj).(66)T(z)=\frac{z-a}{r},\qquad T_j(z)=\frac{z-a_j}{r_j},\qquad U=T(\Omega),\qquad U_j=T_j(\Omega_j). \tag*{(66)}

Then δU(0)=δUj(0)=1\delta_U(0)=\delta_{U_j}(0)=1. Suppose (15) holds. Similarity invariance gives

(Uj,kUj;0)→pGH(U,kU;0).(U_j,k_{U_j};0)\xrightarrow{\mathrm{pGH}}(U,k_U;0).

In the proof of Theorem 5.8 in [17], the assumption n≥3n\geq3 is used only to deduce that two pointed quasihyperbolically isometric domains with equal boundary distances at the origin are related by an orthogonal transformation. Equation (65) provides this conclusion in dimension two. Therefore, following the argument in the proof of Theorem 5.8 in [17] for UjU_j and UU with (65), we obtain Qj∈O(2)Q_j\in O(2) such that

dQj(Uj)⟶δUuniformly on bounded subsets of U.(67)d_{Q_j(U_j)}\longrightarrow\delta_U\quad\text{uniformly on bounded subsets of }U. \tag*{(67)}

For the similarities in (16), we have

Sj=T−1∘Qj∘Tj,S_j=T^{-1}\circ Q_j\circ T_j,
dSj(Ωj)(z)=r dQj(Uj)(T(z)),δΩ(z)=r δU(T(z))(z∈Ω).(68)d_{S_j(\Omega_j)}(z)=r\,d_{Q_j(U_j)}(T(z)),\qquad\delta_{\Omega}(z)=r\,\delta_U(T(z))\quad(z\in\Omega). \tag*{(68)}

Equations (67) and (68) imply (17).

Conversely, since every closed quasihyperbolic ball KK in Ω\Omega is compactly contained in Ω\Omega, (17) gives

K⊂Vjfor all sufficiently large jandδΩdVj⟶1uniformly on K,K\subset V_j\quad\text{for all sufficiently large }j\quad\text{and}\quad\frac{\delta_{\Omega}}{d_{V_j}}\longrightarrow1\quad\text{uniformly on }K,

where Vj=Sj(Ωj)V_j=S_j(\Omega_j). By [17], Theorem 4.4, (Vj,kVj;a)→pGH(Ω,kΩ;a)(V_j,k_{V_j};a)\xrightarrow{\mathrm{pGH}}(\Omega,k_{\Omega};a). Each SjS_j is a quasihyperbolic isometry with Sj(aj)=aS_j(a_j)=a, so (15) follows. Finally, dVjd_{V_j} and δΩ\delta_{\Omega} are 1-Lipschitz on Ω\Omega. A finite-net argument on each bounded subset of Ω\Omega shows that pointwise convergence is equivalent to (17); compare [17], the proof of Proposition 3.8 and Remark 3.9(a).

Declaration of generative AI use

The author used ChatGPT (OpenAI, GPT-6 Astra) in preparing the proof of Theorem 1.11. Specifically, interactions with ChatGPT helped the author identify (65) as the planar replacement for the step in the argument of Herron, Richard, and Snipes that requires n≥3n \ge3. Comparing their proof with the planar argument with the help of ChatGPT allowed the author to shorten the proof, as explained in the final paragraph of the proof of Theorem 1.11. A literature search with ChatGPT for earlier occurrences of (30) helped the author locate the equation of Graham and Witten discussed in Remark 2.4. The author also used ChatGPT to assist with language editing and improving the exposition of the manuscript. Apart from the assistance described above, all mathematical statements and all other proofs were developed independently by the author. The author takes full responsibility for the content of the manuscript.

References

DEPARTMENT OF MATHEMATICS AND STATISTICS, UNIVERSITY OF JYVÄSKYLÄ, P.O. BOX 35, 40014 UNIVERSITY OF JYVÄSKYLÄ, FINLAND

Email address: [email protected]

References

  1. [1]Z. M. Balogh and S. M. Buckley, Geometric characterizations of Gromov hyperbolicity, Invent. Math. 153 (2003), no. 2, 261–301.DOI
  2. [2]M. Bonk, J. Heinonen, and P. Koskela, Uniformizing Gromov hyperbolic spaces, Astérisque, vol. 270, Société Mathématique de France, 2001.
  3. [3]S. M. Buckley and D. A. Herron, Quasihyperbolic geodesics are hyperbolic quasi-geodesics, J. Eur. Math. Soc (JEMS) 22 (2020), no. 6, 1917–1970.arxiv.org/abs/1704.07193
  4. [4]L. A. Caffarelli and A. Friedman, The free boundary for elastic-plastic torsion problems, Trans. Amer. Math. Soc. 252 (1979), 65–97.DOI
  5. [5]H. I. Choi, S. W. Choi, and H. P. Moon, Mathematical theory of medial axis transform, Pacific J. Math. 181 (1997), no. 1, 57–88.DOI
  6. [6]J. Damon, Smoothness and geometry of boundaries associated to skeletal structures I: sufficient conditions for smoothness, Ann. Inst. Fourier (Grenoble) 53 (2003), no. 6, 1941–1985.
  7. [7]F. W. Gehring and B. G. Osgood, Uniform domains and the quasi-hyperbolic metric, J. Anal. Math. 36 (1979), 50–74.DOI
  8. [8]F. W. Gehring and B. P. Palka, Quasiconformally homogeneous domains, J. Anal. Math. 30 (1976), 172–199.DOI
  9. [9]C. R. Graham and E. Witten, Conformal anomaly of submanifold observables in AdS/CFT correspondence, Nuclear Phys. B 546 (1999), no. 1–2, 52–64.DOI
  10. [10]P. Hästö, Isometries of the quasihyperbolic metric, Pacific J. Math. 230 (2007), no. 2, 315–326.DOI
  11. [11]P. Hästö and Z. Ibragimov, Apollonian isometries of planar domains are Möbius mappings, J. Geom. Anal. 15 (2005), no. 2, 229–237.DOI
  12. [12]P. Hästö, Z. Ibragimov, and H. Lindén, Isometries of relative metrics, Comput. Methods Funct. Theory 6 (2006), no. 1, 15–28.DOI
  13. [13]P. Hästö, Z. Ibragimov, D. Minda, S. Ponnusamy, and S. Sahoo, Isometries of some hyperbolic-type path metrics, and the hyperbolic medial axis, In the tradition of Ahlfors–Bers. IV, Contemp. Math., vol. 432, Amer. Math. Soc., Providence, RI, 2007, pp. 63–74.DOI
  14. [14]M. Heins, On a class of conformal metrics, Nagoya Math. J. 21 (1962), 1–60.DOI
  15. [15]D. A. Herron, Quasihyperbolic metric universal covers, Rev. Mat. Iberoam. 37 (2021), no. 4, 1277–1306.arxiv.org/abs/2007.00782
  16. [16]D. A. Herron, Z. Ibragimov, and D. Minda, Geodesics and curvature of Möbius invariant metrics, Rocky Mountain J. Math. 38 (2008), no. 3, 891–921.DOI
  17. [17]D. A. Herron, A. Richard, and M. A. Snipes, Chordal Hausdorff convergence and quasihyperbolic distance, Anal. Geom. Metr. Spaces 8 (2020), 36–67.DOI
  18. [19]P. W. Jones, Extension theorems for BMO, Indiana Univ. Math. J. 29 (1980), no. 1, 41–66.
  19. [20]P. W. Jones and S. K. Smirnov, Removability theorems for Sobolev functions and quasiconformal maps, Ark. Mat. 38 (2000), no. 2, 263–279.DOI
  20. [21]P. Koskela, Old and new on the quasihyperbolic metric, Quasiconformal mappings and analysis (P. Duren, J. Heinonen, B. Osgood, and B. Palka, eds.), Springer, New York, 1998, pp. 205–219.DOI
  21. [22]C. Mantegazza and A. C. Mennucci, Hamilton–Jacobi equations and distance functions on Riemannian manifolds, Appl. Math. Optim. 47 (2003), no. 1, 1–25.DOI
  22. [23]G. J. Martin, Quasiconformal and bi-Lipschitz homeomorphisms, uniform domains and the quasihyperbolic metric, Trans. Amer. Math. Soc. 292 (1985), no. 1, 169–191.DOI
  23. [24]G. J. Martin and B. G. Osgood, The quasihyperbolic metric and associated estimates on the hyperbolic metric, J. Anal. Math. 47 (1986), 37–53.DOI
  24. [25]D. Ntalampekos and M. Younsi, Rigidity theorems for circle domains, Invent. Math. 220 (2020), no. 1, 129–183.DOI
  25. [26]B. Osgood, Old and new on the Schwarzian derivative, Quasiconformal mappings and analysis (P. Duren, J. Heinonen, B. Osgood, and B. Palka, eds.), Springer, New York, 1998, pp. 275–308.DOI
  26. [27]A. Richard, Quasihyperbolic distance, pointed Gromov–Hausdorff distance, and bounded uniform convergence, Ph.D. thesis, University of Cincinnati, 2019.

Paper details

vlnc:W_2tuukjep
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

GPT-6 Astra (bare)

View statement
Looking for related papers…
Contents