You are on page 1of 17

J. Math. Anal. Appl.

414 (2014) 693–709

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and Applications


www.elsevier.com/locate/jmaa

On the hyperbolic order


Hugo Arbeláez a,∗,1 , Diego Mejía a,1 , Willy Sierra b,2
a
Universidad Nacional de Colombia, Sede Medellín, Medellín, Colombia
b
Universidad del Cauca, Popayán, Colombia

a r t i c l e i n f o a b s t r a c t

Article history: We define the hyperbolic order of any locally injective holomorphic function between
Received 19 May 2013 arbitrary hyperbolic domains of the complex plane and study the relation between
Available online 18 January 2014 the hyperbolic order and the Schwarzian derivative for locally injective holomorphic
Submitted by E. Saksman
functions from the unit disk into itself.
Keywords: © 2014 Elsevier Inc. All rights reserved.
Hyperbolic order
Hyperbolic invariance
Schwarzian derivative
Hyperbolically linearly connected

1. Introduction

Let Δ and Ω be regions in the complex plane C. A holomorphic function h : Δ → Ω is called a covering

if each point w ∈ Ω has an open neighborhood W such that h−1 (W ) = {Uα : α ∈ Λ} is a disjoint union of
open sets such that for each α the restriction h|Uα is a conformal map of Uα onto W . A region in C is said to
be hyperbolic if its complement contains at least two points. We recall that there is a holomorphic covering
from the unit disk D onto any hyperbolic region but holomorphic coverings between arbitrary hyperbolic
regions may not exist. With this in mind, given two hyperbolic regions Δ and Ω and a locally injective
holomorphic map f from Δ into Ω we would like to measure the deviation of f from a covering when
such covering exists. To this end we define the hyperbolic order (h-order) of a locally injective holomorphic
function f : Δ → Ω by
 
 h  h   1 ∂ 
Af  := sup Af (z) = sup 
  log f Δ,Ω
(z),
Δ,Ω λΔ (z) ∂z
z∈Δ z∈Δ

where

* Corresponding author at: Escuela de Matemáticas, Universidad Nacional de Colombia, Sede Medellín, Calle 59A  63-20,
Bloque 43, Medellín, Colombia.
E-mail addresses: hjarbela@unal.edu.co (H. Arbeláez), dmejia@unal.edu.co (D. Mejía), wsierra@unicauca.edu.co (W. Sierra).
1
Partially supported by Colciencias code 1118-521-28160 and DIME.
2
The author wishes to thank the University of Cauca for providing time for this work through research project VRI ID 3357.

0022-247X/$ – see front matter © 2014 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jmaa.2014.01.029
694 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

λΩ (f (z))|f  (z)|
f Δ,Ω (z) :=
λΔ (z)

is the (local) hyperbolic distortion factor for f at z (see [3]) and λΩ and λΔ are the densities of the hyperbolic
metrics on Ω and Δ, respectively. When Ω = Δ we simply write Ahf Δ and f Δ . We also say that a family F
of locally injective holomorphic functions f from Δ into Ω is hyperbolically invariant or, in short, h-invariant,
if

f ∈ F, σ ∈ Cov(Ω), τ ∈ Cov(Δ) ⇒ σ ◦ f ◦ τ ∈ F,

where Cov(Ω) and Cov(Δ) are, respectively, the sets of holomorphic self-coverings of Ω and Δ. Hyperbolic
invariance was first considered by Ma and Minda [7] (see also [6]) as an analog in hyperbolic geometry
to (Euclidean) linear invariance, a concept introduced and extensively studied by Pommerenke [11]. More
specifically, Ma and Minda considered the important case Ω = Δ = D where Ahf takes the form

(1 − |z|2 ) f  (z) (1 − |z|2 )f (z)f  (z)


Ahf (z) = 
−z+ .
2 f (z) 1 − |f (z)|2

They proved [6, Th. 9] that Ahf D  1 if f is not a conformal automorphism of D, with equality precisely
when f is hyperbolically convex (h-convex), that is, when f maps D conformally onto a proper domain in D
for which any two points can be joined inside the domain by the arc of the hyperbolic geodesic between
the two points. Ma and Minda also proved [6, Th. 10] that the finiteness of the hyperbolic order for a
(locally injective) function f : D → D is equivalent to the existence of a positive number ρ such that for
all a in D, the image f (Dh (a, ρ)) is h-convex, where Dh (a, ρ) is the hyperbolic disk in D centered at a
with hyperbolic radius ρ. A function that satisfies this last condition is referred to as uniformly locally
hyperbolically convex.
The h-order of an h-invariant family F is given by
 
Oh (F) := sup Ahf Δ,Ω .
f ∈F

The results of Ma and Minda show that the family of h-convex functions in the unit disk has h-order equal
to one. An inequality of Pick [10] (see also [12, p. 99]) gives that the h-order of all univalent functions
in the unit disk is less or equal than two, and Example 3 in Section 5 shows that the bound two is
sharp.
This paper is organized as follows. Section 2 contains some facts about the hyperbolic metric. In Section 3
we study the main properties of the differential operator Ahf . Basically this operator measures the deviation
of f from a holomorphic covering. More precisely, Ahf identically vanishes if and only if f is a covering. The
boundedness of this operator is equivalent to a two-point distortion condition on the function f . In Section 4
we specialize to the case Δ = Ω = D and study the relation between the Schwarzian derivative, the local
distortion of f and Ahf . In particular Theorem 5 is the hyperbolic version of the corresponding result in the
Euclidean case due to Pommerenke [11, p. 133]. In Section 5 we give several examples to illustrate some of
the aspects considered above. We also mention that some of the results of this paper are hyperbolic analogs
of similar results in the spherical case obtained by the first two authors in [2].

2. Preliminaries

The hyperbolic metric on the unit disk D is given by λD (z)|dz| = (1 − |z|2 )−1 |dz|. On any hyperbolic
region Ω in the complex plane C the hyperbolic metric λΩ (z)|dz| is determined from λΩ (π(z))|π  (z)| =
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 695

λD (z), where π is any holomorphic covering of D onto Ω. The hyperbolic metric has Gaussian curvature −4
which means that
Δ log λΩ (z)
KλΩ (z) := − = −4.
λ2Ω (z)

It is immediate that if c > 0, then KcλΩ (z) = c−2 KλΩ (z).


Let Δ and Ω be hyperbolic regions and let f : Δ → Ω be a holomorphic function. The pull-back of
λΩ (w)|dw| by f is
    
f ∗ λΩ (w)|dw| = λΩ f (z) f  (z)|dz|.

The Gaussian curvature is invariant under the pull-back operation. More precisely, if the mapping f is
locally univalent then Kf ∗ (λΩ ) (z) = KλΩ (f (z)) for all z ∈ Δ.

3. The differential operator Ah


f

Suppose Ω and Δ are hyperbolic domains in C and let f : Δ → Ω be a locally injective holomorphic
function. We define the hyperbolic pre-Schwarzian of f by


Tf (z) = log f Δ,Ω (z).
∂z
The hyperbolic pre-Schwarzian satisfies a chain rule that can be checked with no difficulty. More precisely,
if g is a locally injective holomorphic function from a hyperbolic domain G into a hyperbolic domain Δ,
and f is a locally injective holomorphic function from Δ into a hyperbolic domain Ω, then

Tf ◦g = (Tf ◦ g)g  + Tg .

We will see shortly that Tf identically vanishes if and only if f is a holomorphic covering. Therefore it
follows from the chain rule that

Tf ◦g = Tg if and only if f is a holomorphic covering from Δ onto Ω.

Also, if g is a holomorphic covering from G onto Δ then Tf ◦g = (Tf ◦ g)g  .

Theorem 1. Suppose Ω and Δ are hyperbolic domains in C and let f : Δ → Ω be a locally injective
holomorphic function. Then Tf ≡ 0 if and only if f is a covering.

Proof. Suppose first that f is a covering. The invariance of the hyperbolic metric under coverings (see
for instance [3, Th. 10.4]) gives that the hyperbolic distortion for f is identically equal to 1 from which
follows that Tf ≡ 0. Conversely, suppose that Tf ≡ 0. Then there is a positive constant c such that
λΩ (f (z))|f  (z)| = cλΔ (z) for all z ∈ Δ. Hence, for all z ∈ Δ the Gaussian curvature satisfies Kf ∗ (λΩ ) (z) =
c−2 KλΔ (z). On the other hand, by the invariance of the Gaussian curvature under holomorphic functions
Kf ∗ (λΩ ) (z) = KλΩ (f (z)). Since the hyperbolic metric has constant Gaussian curvature −4 it follows that c
is equal to 1. The general version of Schwarz–Pick Lemma (see [3, Th. 10.5]) gives that f is a covering. 2

To some extent the sup-norm of Tf measures the deviation of f from a covering. However, it is more
convenient to use the slightly different operator

1 1 ∂
Ahf (z) = Tf (z) = log f Δ,Ω (z). (1)
λΔ (z) λΔ (z) ∂z
696 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

Then, the chain rule takes the form

λΔ ◦ g  h 
Ahf◦g = Af ◦ g g  + Ahg ,
λG

hence, by Theorem 1, Ahf◦g = Ahg if and only if f is a covering, and if g is a covering

λΔ ◦ g  h 
Ahf◦g = Af ◦ g g  . (2)
λG

It now follows the invariance property of the h-order:

Theorem 2. Let τ : G → Δ and σ : Ω → W be holomorphic coverings. If f : Δ → Ω is a locally injective


holomorphic function, then
 h   h
Aσ◦f ◦τ  = Af  .
G,W Δ,Ω

Theorem 3. Let f : Δ → Ω be a locally injective holomorphic function. Then, |Ahf (z)|  α for all z ∈ Δ if
and only if for all z0 , z ∈ Δ

  f Δ,Ω (z)  
exp −2αdh (z0 , z)  Δ,Ω  exp 2αdh (z0 , z) . (3)
f (z0 )

Proof. Suppose first that |Ahf (z)|  α for all z ∈ Δ. Fix z0 ∈ Δ and let z ∈ Δ. Let γ be an h-geodesic
in Δ from z0 to z. Parametrize γ by hyperbolic arc-length, that is, γ  (s) = eiθ(s) /λΔ (γ(s)), where s is the
h-length of γ from z0 to z. Set F (s) = f Δ,Ω (γ(s)). It follows that
 Δ,Ω 
∂f  
F  (s) = 2 Re γ(s) γ  (s)
∂z
 
1 ∂ Δ,Ω   iθ(s)
= 2F (s) Re f γ(s) e
λΔ (γ(s))f Δ,Ω (γ(s)) ∂z
 
1 ∂  
= 2F (s) Re log f Δ,Ω γ(s) eiθ(s) .
λΔ (γ(s)) ∂z

Thus,

d F  (s)  

log F (s) = = 2 Re Ahf γ(s) eiθ(s) ,


ds F (s)

and this implies, by hypothesis,

d
−2α  log F (s)  2α.
ds
Integrating between 0 and s we obtain

f Δ,Ω (z)
−2αdh (z0 , z)  log  2αdh (z0 , z).
f Δ,Ω (z0 )

For the converse, we set u(z) = log f Δ,Ω (z). Then by (3)

|u(z) − u(z0 )| dh (z0 , z)


 2α ,
|z − z0 | |z − z0 |
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 697

z0 , z ∈ Δ. Now let z approach z0 in the direction of maximum growth of u at z0 , to get |2 ∂u


∂z (z0 )| =
|∇u(z0 )|  2αλΔ (z0 ). 2

The left hand side inequality in (3) yields that the h-order of any noncovering locally injective holomorphic
function between hyperbolic domains is greater or equal than one. This was proved by Ma and Minda in
the case Δ = Ω = D [6, Th. 9], and the invariance of the h-order gives the result in the general situation.
Moreover, Ma and Minda also proved that the h-order is one precisely when the mapping is hyperbolically
convex. We summarize these results in the following proposition.

Proposition 1. Let f : Δ → Ω be a noncovering locally injective holomorphic function. Let π1 : D → Δ,


π2 : D → Ω be coverings. Then, Ahf Δ,Ω  1 with equality if and only if the lifting g of f to D, given by
π2 ◦ g = f ◦ π1 , is hyperbolically convex.

The above proposition suggests the importance of studying hyperbolically invariant families in the unit
disk.

4. Hyperbolically invariant families in D

Suppose f : D → D is holomorphic and locally univalent. For each a ∈ D let ρ(a, f ) be the hyperbolic
radius of the largest hyperbolic disk in D, centered at a, in which f is univalent. Define
 
ρh (a, f ) = sup ρ  ρ(a, f ): f Dh (a, ρ) is h-convex

and

ρh (f ) = inf ρh (a, f ): a ∈ D .

Let Aut(D) denote the group of conformal automorphisms of D. Ma and Minda [6, Th. 10] proved

Theorem 4. Let f : D → D be holomorphic and locally univalent and f ∈


/ Aut(D). Then ρh (f ) = ρ > 0 if
and only if Af D = coth(2ρ).
h

Let Fh (α) denote the family of locally injective holomorphic functions f : D → D with Ahf D  α.

Corollary 1. Let f ∈ Fh (α) and f ∈


/ Aut(D). Then, for all z ∈ D
   2
(i) β Ahf (z) + f D  β 2 , (4)
 2  2  
(ii) 3Ahf (z) + 1 − |z|2 Sf (z)  3β 2 , (5)

where β = α + α2 − 1.

The combination of (4) and (5) gives a nonsharp Wirths’ type inequality ([15, p. 49], see also [9, Th. 6])
for locally injective holomorphic functions from the unit disk into itself. We wonder what would be the
sharp version of this inequality.

Proof. (i) By invariance it suffices to prove the inequality at z = 0 with f (0) = 0. Then we must show
  
 f (0)    2
β    + f (0)  β 2 .

2f (0)
698 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

By Theorem 4 the function f is h-convex in the hyperbolic disk centered at 0 and hyperbolic radius ρ =
−1
1
2 coth α. Therefore the function g(z) := f (Rz) is h-convex in D where R = tanh ρ. It follows from [6,
Th. 5] that
  
 g (0)    2
   
 2g  (0)  + g (0)  1.

So,
  
 f (0)   2
R   + R2 f  (0)  1,

2f (0)

which is the required inequality since 1/R2 = β 2 .


(ii) As in part (i) we may assume that f (0) = 0 and must show the inequality
  
 f (0) 2  
3   + Sf (0)  3β 2 .
2f (0)

Theorem 5.1 of [8] applied to the function f (Rz), R = tanh ρ gives


  
 f (0) 2  
3R    + R2 Sf (0)  3
2
2f (0)

as desired. 2

Theorem 5. Let f : D → D be a locally injective holomorphic function with finite hyperbolic order α. Then
 h 2
Af   1 Sf D + 1. (6)
D 2
The bound is sharp for α ∈ [1, 2].

This theorem is the hyperbolic version of the corresponding result in the Euclidean case due to Pom-
merenke [11, p. 133]. We do not know whether the result is sharp for all α. At the end of the proof we show
that for α ∈ [1, 2] the sharpness is attained for univalent functions.

Proof. Let 0 < r < 1 and Dr = {z ∈ C | |z| < r}. Define h : D → Dr by h(z) = rz, fr := f |Dr and let
g = fr ◦ h. We will see first that the result holds for g, that is
 h 2
Ag   1 Sg D + 1. (7)
D 2

The function g has a locally injective analytic extension to D which we still denote by g. If the maximum
value of |Ahg (z)| is taken on the boundary of D then Ahg D = 1, therefore g is hyperbolically convex and
satisfies (7). Now suppose that the maximum value of |Ahg (z)| in D is reached inside D. Since all terms
in (7) are invariant if we change g by ϕ ◦ g ◦ ψ, with ϕ, ψ ∈ Aut(D) we may assume that g has the form
g(z) = β(z +a2 z 2 +a3 z 3 +· · ·), 0 < β  1 and that maxz∈D |Ahg (z)| = |Ahg (0)|. So, under these circumstances,
to show (7) we must prove that

 h 2  
Ag (0)  1 Sg (0) + 1. (8)
2

Observe that Ahg (0) = a2 and Sf (0) = 6(a3 − a22 ). The expansions of g, g  and g  , about 0 give
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 699

          
Ahg (z) = 1 − |z|2 a2 + 3a3 − 2a22 z + O z 2 − z + 1 − |z|2 β 2 z + O z 2
     
= a2 + 3a3 − 2a22 z − 1 − β 2 z + O |z|2 ,

near the origin. Therefore


 h 2    
 
Ag (z) = |a2 |2 + 2 Re a2 3a3 − 2a22 − a2 1 − β 2 z + O |z|2
  
Sg (0)   h  2  
h
= 2 Re Ag (0) + Ag (0) − 1 − β Ag (0) z + Ahg (0) + O |z|2
h 2 2
2

and under the assumption maxz∈D |Ahg (z)| = |Ahg (0)|, it follows that
  
Sg (0)    
2 Re Ahg (0) + Ahg (0)2 − 1 − β 2 Ahg (0) z + O |z|2  0,
2
S (0)
near the origin. Write C := Re[Ahg (0)( g2 + Ahg (0)2 ) − (1 − β 2 )Ahg (0)] and z = reiθ , θ ∈ R. Then
2 Re(Creiθ ) + O(r2 )  0. This implies that 2 Re(Ceiθ ) + O(r)  0, and if r → 0+ then Re(Ceiθ )  0.
By choosing θ = −arg C we get |C|  0, therefore

Sg (0)  
Ahg (0) + Ahg (0)2 − 1 − β 2 Ahg (0) = 0,
2

hence
 
 Sg (0)   h 2    h 2
    2 
> Ag (0) − 1.
 2  = Ag (0) − 1 − β

This shows (8). Now, given that g = fr ◦ h, 0 < r < 1, it follows from (2) and (7) that
 2
 h     
Af (w)2 = Ahg w   Ahg 2  1 Sg D + 1
r  r  D 2
1
 Sfr Dr + 1. (9)
2
The last inequality holds since for arbitrary z ∈ D and w = rz

 2    2   (r2 − |w|2 )2  
1 − |z|2 Sg (z) = r2 1 − |z|2 Sfr (rz) = Sf (w)
r
r2
|Sfr (w)|
=  Sfr Dr .
λDr (w)2

On the other hand we observe that

|Sfr (w)| |Sf (w)| |Sf (w)|


Sfr Dr = sup  sup  sup = Sf D ,
w∈Dr λDr (w)2 λ
w∈Dr D (w) 2
w∈D λD (w)
2

since λDr  λD and f = fr in Dr . Moreover, given w ∈ D there is 0 < r0 < 1 such that w ∈ Dr for all
r > r0 and by (1)

 h  1 ∂ λD (fr (w))|fr (w)|


Af (w) = log
r
λDr (w) ∂w λDr (w)
1 ∂ λD (f (w))|f  (w)|
= log .
λDr (w) ∂w λDr (w)
700 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

Therefore limr→1− |Ahfr (w)| = |Ahf (w)|. So, from (9) and the previous observations it follows that

 h 
Af (w)2  1 Sf D + 1,
2
for arbitrary w ∈ D. This shows (6).
For α ∈ [1, 2] the sharp bounds in (6) are attained by the univalent functions kα,β considered by Ma and
Minda in [7, p. 86]. Let β ∈ (0, 1] and kα,β be the holomorphic function on D normalized by kα,β (0) = 0,

kα,β (0) = β and defined from the identity
α  α 
1 + kα,β (z) 1+z
−1=β −1 .
1 − kα,β (z) 1−z

The function k1,β is a Möbius transformation that maps the unit disk properly into itself, it has Schwarzian
norm zero and h-order one since it is h-convex. This proves equality in the case α = 1.
For the other cases we first show that Ahkα,β D  α. A large but straightforward computation (see
Appendix A) gives

(α − 1)(1 − x) (α + 1)(1 + x)
Ahkα,β (x) = −x + +
2 2
2  
α−1 (1 − x )kα,β (x) α + 1 (1 − x2 )kα,β (x)
− − ,
2 1 − kα,β (x) 2 1 + kα,β (x)

for x ∈ (−1, 1). Thus Ahkα,β D  |Ahkα,β (x)| → α as x → −1. Now, as k2,β is a univalent function it is
known [5] that Sk2,β D  6. Hence 4  Ahk2,β 2D  12 Sk2,β D + 1  4 which proves the sharpness in the
case α = 2. For α ∈ (1, 2) we show in Appendix A that Skα,β D  2(α2 − 1) giving the sharp bound in all
these cases. 2

Remark 1. A Nehari function f : D → C is a univalent function for which Sf D  2. Hence, it also follows

from (6) that Ahf D  2 for any Nehari function f from the unit disk into itself of finite h-order.

Corollary 2. Let f : D → D be an injective holomorphic function with Ahf D = 2. Then Sf D = 6.


Furthermore, if f (D) is a quasidisk then Ahf D < 2.

Proof. Suppose that Ahf D = 2. Inequality (6) gives 6  Sf D . Also, it is known [5] that if f is injective
then Sf D  6.
Now, suppose that f (D) is a quasidisk, then (see [13, Th. 5.17]) f can be extended to a quasiconformal
map of C onto C. A result of Ahlfors and Weill [1], implies that Sf D < 6, it follows from (6) that
Af D < 2. 2
h

The examples of conformal mappings from D into itself with h-order two (see Example 3 and Example 4
in Section 5) are suggesting the presence of inward cusps on the boundary. The linearly connected geometric
condition on a domain prevents this to happen.

Definition 1. A simply connected domain G ⊂ D is hyperbolically linearly connected (h-linearly connected)


if there exists a constant b > 0 such that any two points w1 , w2 ∈ G can be connected by a curve Γ ⊂ Ω
with end points w1 and w2 such that

diamh (Γ )  bdh (w1 , w2 ) (10)

where diamh (Γ ) stands for the hyperbolic diameter of Γ .


H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 701

Theorem 6. Let f be a conformal mapping from D onto an h-linearly connected domain G ⊂ D. Then
Ahf  < 2.

Proof. We may suppose that f (0) = 0. Let b be as in the definition for G = f (D). We consider the family F
of univalent functions g : D → D with power series g(z) = a1 z + a2 z 2 + · · · , a1 > 0, and g(D) h-linearly
connected with constant b. The family F is normal. We will show that there exists M ∈ (0, 2) such that

|a2 |
M
a1

for all g ∈ F. Indeed if such M does not exist then there is a sequence (gn ) ∈ F such that

|a2 (n)|
→2 (11)
a1 (n)

as n → ∞. It follows from Pick’s inequality that |a2 (n)|/a1 (n) < 2 for all n and a1 (n) → 0 as n → ∞. Write
a2 (n) = |a2 (n)|eiθn and

eiθn   |a2 (n)| 2


fn (z) = gn e−iθn z = z + z + ···.
a1 (n) a1 (n)

The functions fn belong to the class S of normalized univalent functions in the unit disk. Then, we may
suppose that (fn ) converges locally uniformly to a function in S. Moreover, by (11) we have fn → k locally
uniformly where k(z) = z/(1 − z)2 is the Koebe function. Write g̃n (z) = eiθn gn (e−iθn z). Then fn = a11(n) g̃n
and g̃n (D) is h-linearly connected with the same constant b. By the Carathéodory convergence theorem
fn (D) → k(D) in the sense of kernel convergence. Since a2 (fn ) = |a2 (n)|/a1 (n) < 2, fn (D) contains a disk
centered at the origin with radius strictly larger than 1/4 so that −1/4 is an interior point of fn (D) for

all n. Since −1/4 ∈ ∂k(D), then −1/4 is not an interior point of fn (D). Hence, given δ > 0 and K > 0
there exists nm > K such that the disk D(−1/4, δ) is not contained in fnm (D).
On the other hand, let (yn ) be a sequence of real numbers in (0, 1) converging to zero as n → ∞. Write
An = (−2, yn ) and Bn = (−2, −yn ). Since a1 (n) → 0, then for sufficiently large n the points Pn = a1 (n)An
and Qn = a1 (n)Bn lie inside the disk D(0, 1/2). For convenience we may suppose that Pn and Qn lie all
inside D(0, 1/2). Therefore there exists an absolute constant C such that for all n

dh (Pn , Qn )  Cde (Pn , Qn ). (12)

For a fix positive integer m let Lm = [Am , 0] + [0, Bm ] and δm ∈ (0, 1/4) with D(−1/4, δm ) contained in
the interior of the triangle with vertices Am , 0, Bm . Since Lm is compact and it is inside k(D) (the kernel
of (fn (D))) there is Nm such that n  Nm ⇒ Lm ⊂ fn (D). Then we can choose nm > Nm and a point
wm ∈ D(−1/4, δm ) with Lm ⊂ fnm (D) and wm ∈ / fnm (D). From the construction we may take nm+1 > nm .
Since fnm (D) is simply connected it follows that any curve in fnm (D) with end points Am and Bm meets
some vertical line x = xm with xm > Re wm and therefore such a curve has Euclidean diameter greater
than one. We have therefore constructed a subsequence of (fn ) which we denote by (fm ) and points Am ,
Bm in fm (D) such that Pm = a1 (m)Am and Qm = a1 (m)Bm belong to the disk D(0, 1/2) for all m, and
every curve in fm (D) with end points Am , Bm has Euclidean diameter greater than one. We also notice
that de (Am , Bm ) → 0 as m → ∞.
Now, let pm , qm ∈ D be such that g̃m (pm ) = Pm and g̃m (qm ) = Qm . Then fm (pm ) = Am and
fm (qm ) = Bm . We have by (10)

diame (βm )  2diamh (βm )  2bdh (Pm , Qm )


702 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

for every curve βm ⊂ g̃m (D) with end points Pm , Qm . It follows from a result of Gehring and Hayman [4]
(see also [13, p. 88]), that there exists an absolute constant C1 such that for all m
 
diame g̃m (Sm )  C1 dh (Pm , Qm )

where Sm is the hyperbolic segment with end points pm , qm . It follows from (12)
 
diame g̃m (Sm )  C2 de (Pm , Qm )

for an absolute constant C2 . Therefore


 
1 1   C2 1 1
diame g̃m (Sm ) = diame g̃m (Sm )  de (Pm , Qm ) = C2 de Pm , Qm .
a1 (m) a1 (m) a1 (m) a1 (m) a1 (m)

Hence

diame (Γm )  C2 de (Am , Bm )

1
where Γm = a1 (m) g̃m (Sm ) = fm (Sm ) which is a curve in fm (D) with endpoints Am , Bm . Then we have

1  diame (Γm )  C2 de (Am , Bm ).

This contradicts the fact that de (Am , Bm ) → 0 as m → ∞. So, we conclude that


 
|a2 |
β = sup : g(z) = a1 z + a2 z 2 + · · · ∈ F < 2.
a1

Finally, given ξ ∈ D, let g = τ ◦ f ◦ σ where

ξ+z f (ξ) − w
σ(z) = and τ (w) = ,
¯
1 + ξz 1 − f (ξ)w

and h = e−iλ g, with λ = Arg g  (0). Then h ∈ F and |Ahf (ξ)| = | 12 hh (0)
(0)
|. We conclude that
 h 
Af (ξ)  β

for all ξ. Hence Ahf   β < 2. 2

5. Examples

Example 1. Let Gh be the family of the hyperbolically convex function f from D into D. Then Oh (Gh ) = 1
[6, Th. 9].

Any hyperbolically convex functions f from D into D is a conformal mapping. The next example shows
that if we get out of the family hyperbolically convex we can lose the univalence.

Example 2. Let a < b < 0, f be a conformal mapping from D onto [a, b]×[0, 3π] and g = ef . A straightforward
computation gives

1 − |z|2 f  (1 − |z|2 )f  (z) (1 − |z|2 )f  (z)|ef (z) |2


Ahg (z) = (z) − z + + .
2 f 2 1 − |ef (z) |2
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 703

Thus
 
 h   1 − |z|2 f   (1 − |z|2 )|f  (z)| 1 + |ef (z) |2
Ag (z)   (z) − z + .
 2 f   2 1 − |ef (z) |2

Since f is convex it follows [11] that

 h   
Ag (z)  1 + de f (z), ∂f (D) 1 + |e |
f (z) 2
1 + e2b
 1 + (b − a) .
1 − |ef (z) |2 1 − e2b

Let  > 0 be given. Fix b and choose a sufficiently close to b so that |Ahg (z)|  1 +  for all z ∈ D. Note that
g is not injective.

Example 3. Let  S be the family of analytic and univalent functions f : D → D. Then Oh (


S)  2 (see [7,
p. 76]). For n ∈ N, n  2, the polynomial

n 1 n
f (z) = z− z
n+1 n

maps D univalently into D (see [14]). Lengthy but straightforward computations give

(n − 1)(1 − |z|2 ) z n−2 (1 − |z|2 )( n+1


n 2
) (1 − z n−1 )(z − n1 z n )
Ahf (z) = − − z + .
2 1 − z n−1 1 − | n+1
n
(z − n1 z n )|2

For z = x real we have

(n − 1)(1 + x) xn−2 (1 − x2 )( n+1


n 2
) (1 − xn−1 )(x − n1 xn )
Ahf (x) = − − x + .
2 1 + x + · · · + xn−2 1 − | n+1
n
(x − n1 xn )|2

Therefore Ahf (x) → −2 as x → 1− . Hence Ahf D = 2. We note that the images of these polynomials have
inner cusps at those points where the derivative is zero.

Example 4. Let SL be the family of the slit functions from D into D, namely f ∈ SL if f maps D conformally
into D minus a set of Jordan analytic arcs emanating from the boundary of D. Since SL ⊂  S then Oh (SL )  2.
Indeed the order of this family is 2. To see this, let 0  p < 1 and h(z) = az + · · · , a > 0, be the map of D
onto D  (−1, −p]. It is clear that h ∈ SL . Then (see [12, p. 100])

h(z) 4p z
=
(1 − h(z))2 (1 + p) (1 − z)2
2

and thus

4p
h(z) = az + 2a(1 − a)z 2 + · · · , a= .
(1 + p)2

We have the inequality


  2
 h  h  1  h (0)  1−p
    
2  Ah D  Ah (0) =    = 2(1 − a) = 2 →2
2 h (0)  1+p

as p → 0.
704 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

Appendix A

Let α ∈ [1, 2], β ∈ (0, 1] and kα,β := k : D → D be such that k(0) = 0, k (0) = β and
α  α 
1 + k(z) 1+z
−1=β −1 .
1 − k(z) 1−z

Direct calculations show that

(1 + z)α−1 (1 − k(z))α−1
k (z) = β .
(1 − z)α+1 (1 + k(z))α+1

Since
 ∂   
Ahk (z) = 1 − |z|2 log 1 − |z|2 k (z)
∂z
we obtain

(1 − |z|2 ) (α − 1)(1 + z) (α + 1)(z − 1)
Ahf (z) = −z + −
2 |1 + z|2 |1 − z|2

(α − 1)k (z)(k(z) − 1) (α + 1)k (z)(k(z) + 1)
+ − .
|1 − k(z)|2 |1 + k(z)|2

For z = x ∈ (−1, 1) this expression is

(α − 1)(1 − x) (α + 1)(1 + x)
Ahk (x) = −x + +
2 2
α − 1 (1 − x )k (x) α + 1 (1 − x2 )k (x)
2 
− − .
2 1 − k(x) 2 1 + k(x)

Since k(x) > −1 and (1 − x2 )k (x)  dist(k(x), ∂k(D)) → 0 as x → −1 it follows Ahk D  |Ahk (x)| → α as
x → −1.
To show the inequality Ak D  α, α ∈ (1, 2) we will prove that Sk D  2(α2 − 1). Direct calculations
give
 2 
2(α2 − 1) (1 − z 2 )k (z)
Sk (z) = −1 .
(1 − z 2 )2 1 − (k(z))2

From the definition of k we have

(1 − z 2 )k (z) βwα


=
1 − k2 (z) βwα + (1 − β)
1+z
where w = 1−z . So

 2 
2(α2 − 1) βwα
Sk (z) = −1
(1 − z 2 )2 βwα + (1 − β)
or
 2 
2(α2 − 1) wα
Sk (z) = −1
(1 − z 2 )2 wα + δ
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 705

where δ = 1
β − 1. Hence

 2 
 2   (1 − |z|2 )2 (δ −1/α w)α
1 − |z|2 Sk (z) = 2 α2 − 1 −1 .
(1 − z 2 )2 (δ −1/α w)α + 1

w−1
Since z = w+1 we also have

1 − |z|2 Re(w) Re(δ −1/α w)


= = .
|1 − z 2 | |w| |δ −1/α w|

Combining the above equations it follows that


  2 
 2     Re(δ −1/α w) 2  (δ −1/α w)α 
1 − |z|2 Sk (z) = 2 α2 − 1 −1/α

 −1/α
− 1 .

|δ w| (δ α
w) + 1

Hence Sk D is the supremum of the function


  2 
 2  Re(w) 2  wα 
φ(w) = 2 α − 1  − 1, Re(w) > 0.
|w|  α
w +1

Rewriting w := wα , Sk D is the supremum of


  2 
 2  Re(w1/α ) 2  w 
ϕ(w) = 2 α − 1  − 1
|w| 1/α  w+1
 
 2  
Arg w  2w + 1    απ
= 2 α − 1 cos2  2  , Arg(w) < .
α (w + 1) 2

Thus, for w = reiθ , r  0 and 0  θ < απ


2 ,

 2
 2 θ 4r + 4r cos θ + 1
ϕ2 (r, θ) = 4 α2 − 1 cos4 .
α (r2 + 2r cos θ + 1)2

We will prove that ϕ2 (r, θ)  4(α2 − 1)2 for all r  0 and 0  θ < απ
2 .
For 0  θ < απ2 fixed, let

4r2 + 4r cos θ + 1
f (r) = .
(r2 + 2r cos θ + 1)2

Direct calculations give

4r  2  
f  (r) = − 2r + 3r cos θ + 2 cos2 θ − 1 . (A.1)
(r2 + 2r cos θ + 1)3

Thus, the critical points of f are


√ √
−3 cos θ − 8 − 7 cos2 θ −3 cos θ + 8 − 7 cos2 θ
r0 = 0, r1 = , r2 = .
4 4
Observe that

3π π
r1  0 for θ<π and r2  0 for  θ  π.
4 4
706 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

Since

π 3π
2 cos2 θ − 1  0 if and only if 0θ and  θ  π,
4 4

then f has a maximum at r0 = 0 if 0  θ  π


4 or if 3π
4 < θ  π. And f has a minimum at zero if π
4 <θ 3π
4 .
Consider the following cases for θ:

Case 1: If 0  θ  π4 , then r1 < 0 and r2 < 0. In this case the only critical point of f is zero, which is a
maximum. Thus 1 = f (0)  f (r), for all r  0. Then

 2 2 4 θ
 2
ϕ (r, θ) = 4 α − 1 cos
2
f (r)  4 α2 − 1 ,
α

for all r  0 and 0  θ  π4 .


We now give some general observations on the behavior of f for different values of θ, independent of the
values taken by α:
For π/4 < θ  3π/4, we know that f has a minimum at zero and take its maximum at r2 = r2 (θ). For
3π/4 < θ  π, 2 cos2 θ − 1 > 0 and therefore f has a maximum (relative) at zero. In this case, r2 > r1 > 0
and f takes a minimum at r1 . Again f reaches its maximum value at r2 . Thus, in all cases for which
π/4 < θ < π, f reaches its maximum value at r2 . Now we calculate the maximum value. Let r = r2 . As
f  (r) = 0, from (A.1) we obtain

3r cos θ + 2 cos2 θ − 1
r2 = −
2

and therefore

4r2 + 4r cos θ + 1 = −2r cos θ − 4 cos2 θ + 3

and

r cos θ + 3 − 2 cos2 θ
r2 + 2r cos θ + 1 = .
2

So, the maximum value of f is

(3 − 4 cos2 θ) − 2r cos θ
m(θ) = 4 , r = r2 .
(r cos θ + (3 − 2 cos2 θ))2

Replacing the value r2 , we get



(6 − 5 cos2 θ) − cos θ 8 − 7 cos2 θ n(θ)
m(θ) = 32 √ := 32 2 . (A.2)
((12 − 11 cos2 θ) + cos θ 8 − 7 cos2 θ )2 d (θ)
π απ
Then we must prove that given 1 < α < 2, for 4 <θ< 2


θ n(θ)
cos4 32 2  1.
α d (θ)

We get an upper bound for the quotient dn(θ) 4 θ


2 (θ) which we can control with the factor cos ( α ). Direct cal-

culations show that the critical points of n in [0, π] are θ = 0, π and a ∈ (0, π/2), b ∈ (π/2, π) such that
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 707

cos a = 0.2575894 and cos b = −0.2575894. We notice that n is increasing on [0, b] and decreasing on [b, π].
See the graph of n(θ).

Some values of n are

n(0) = 0, n(π/4) = 2, n(7π/24) = 2.731645654, n(π/3) = 7/2,


n(π/2) = 6, n(b) = 6.375345285, n(2π/3) = 6, n(3π/4) = 5, n(π) = 2.

Similar calculations for d show that the critical points in [0, π] are θ = 0, θ = π and c ∈ (0, π/2), e ∈ (π/2, π)
such that cos c = 0.12587562 and cos e = −0.12587562. We can observe that d is increasing in [0, c] and
decreasing in [c, π]. See the graph for d(θ). Some values of d are

d(0) = 2, d(π/4) = 8, d(7π/24) = 9.33998 . . . , d(π/3) = 21/2,


d(c) = 12.17926201, d(π/2) = 12, d(2π/3) = 8, d(3π/4) = 5, d(π) = 0.

Case 2: Suppose π/4 < θ  7π/24. In this case, π


8  θ
α  π
2. Then,

θ n(θ) sup{n(θ): π/4 < θ  7π/24}
cos4 32 2  32 cos4 (π/8)
α d (θ) inf{d2 (θ): π/4 < θ  7π/24}
2.731645654
 32 cos4 (π/8) < 1.
82

Case 3: Suppose 7π/24 < θ  π/3. In this case, 7π


48  θ
α  π
2. Then,

θ n(θ) sup{n(θ): 7π/24 < θ  π/3}
cos4 32 2  32 cos4 (7π/48)
α d (θ) inf{d2 (θ): 7π/24 < θ  π/3}
3.5
 32 cos4 (7π/48) < 1.
9.332

Case 4: Suppose π/3 < θ  π/2. In this case, π


6  θ
α  π
2. Then,

θ n(θ) sup{n(θ): π/3 < θ  π/2}
cos 4
32 2  32 cos4 (π/6)
α d (θ) inf{d2 (θ): π/3 < θ  π/2}
6
 32 cos4 (π/6) < 1.
(21/2)2

Case 5: Suppose π/2 < θ  2π/3, if 2π/3  απ/2. In other case we suppose π/2 < θ  απ/2 and the case
below 2π/3 < θ < απ/2 is not necessary.
708 H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709

In this case π
4  θ
α  π
2. Then

θ n(θ) sup{n(θ): π/2 < θ  2π/3}
cos 4
32 2  32 cos4 (π/4)
α d (θ) inf{d2 (θ): π/2 < θ  2π/3}
n(b)
 32 cos4 (π/4) < 1.
82

Finally, we consider 2π/3 < θ < απ/2. By rationalize at (A.2) we have


√ √
32 [(6 − 5 cos2 θ) − cos θ 8 − 7 cos2 θ ][(12 − 11 cos2 θ) − cos θ 8 − 7 cos2 θ ]2
m(θ) = 2
16 (8 cos4 θ − 17 cos2 θ + 9)2
 √ 2
1 n(θ) (12 − 11 cos2 θ) − cos θ 8 − 7 cos2 θ
=
8 sin4 θ 9 − 8 cos2 θ
  2
1 n(θ) sin2 θ cos θ (9 − 8 cos2 θ) − sin2 θ
= 1+3 − .
8 sin4 θ 9 − 8 cos2 θ 9 − 8 cos2 θ

As θ > π/2 we obtain the inequality


 2
1 n(θ) sin2 θ
m(θ)  2 + 3 . (A.3)
8 sin4 θ 9 − 8 cos2 θ
2
sin θ
Since 3 9−8 cos2 θ reaches its maximum value 1/3 at θ = π/2, it follows that

49 n(θ)
m(θ) 
72 sin4 θ

and therefore
  4
θ 49 cos( αθ ) 4 49 n(θ) cos( αθ ) 49 n(θ)
cos4
m(θ)  n(θ)   ,
α 72 sin θ 72 × 16 sin 2 cos 2
4 θ θ 1152 sin4 θ2

since 0 < θ/2 < θ/α < π/2. Now, the condition 2π/3 < θ < απ/2 implies that

θ π 3
sin > sin = and n(θ)  n(2π/3) = 6
2 3 2

for all 2π/3 < θ < απ/2. We conclude that for 2π/3 < θ < απ/2

θ 49 6 × 16
cos4 m(θ)  < 1.
α 1152 9

References

[1] L.V. Ahlfors, G. Weill, A uniqueness theorem for Beltrami equations, Proc. Amer. Math. Soc. 13 (1962) 975–978.
[2] H. Arbeláez, D. Mejía, On spherical invariance, Rev. Colombiana Mat. 45 (1) (2011) 97–112.
[3] A. Beardon, D. Minda, The hyperbolic metric and geometric function theory, in: S. Ponnusamy, T. Sugawa, M. Vourinen
(Eds.), Quasiconformal Mappings and Their Applications, Narosa Publishing House, New Delhi, India, 2007.
[4] F.W. Gehring, W.K. Hayman, An inequality in the theory of conformal mapping, J. Math. Pures Appl. (9) 41 (1962)
353–361.
[5] W. Krauss, Über den Zusammenhang einiger Charakteristiken eines einfach zusammenhängenden Bereiches mit der Kreis-
abbildung, Mitt. Math. Sem. Giessen 21 (1932) 1–28.
H. Arbeláez et al. / J. Math. Anal. Appl. 414 (2014) 693–709 709

[6] W. Ma, D. Minda, Hyperbolically convex functions, Ann. Polon. Math. 60 (1994) 81–100.
[7] W. Ma, D. Minda, Hyperbolic linear invariance and hyperbolic k-convexity, J. Aust. Math. Soc. Ser. A 58 (1995) 73–93.
[8] W. Ma, D. Minda, Hyperbolically convex functions II, Ann. Polon. Math. (1999) 273–285.
[9] D. Mejía, Ch. Pommerenke, On spherically convex univalent functions, Michigan Math. J. 47 (2000) 163–172.
[10] G. Pick, Über die konforme Abbildung eines Kreises auf ein schlichtes und zugleich beschränktes Gebiet, S.-B. Kaiserl.
Akad. Wiss. Wien, Math. – Naturwiss. Kl. 126 (1917) 247–263.
[11] Ch. Pommerenke, Linear-invariante Familien analytischer Funktionen I, Math. Ann. 155 (1964) 108–154.
[12] Ch. Pommerenke, Univalent Functions, Vandenhoeck and Ruprecht, Göttingen, 1975.
[13] Ch. Pommerenke, Boundary Behaviour of Conformal Maps, Springer, Berlin, 1992.
[14] T. Sheil-Small, Complex Polynomials, Cambridge University Press, New York, 2002.
[15] J. Wirths, Verallgemeinerungen eines Maximumprinzips, Bonner Math. Schriften, Univ. Bonn 51 (1971) 1–61.

You might also like