A GEOMETRIC APPROACH TO TWO-POINT COMPARISONS FOR HYPERBOLIC AND EUCLIDEAN
GEOMETRY
SEONG-A KIM AND DAVID MINDA
ABSTRACT. Two-point comparison theorems between hyperbolic and euclidean geometry for convex regions in the complex plane <C are known ([5], [6]). We give new geometric proofs of sharp two-point comparison theorems for convex regions.
1. Introduction
Sharp two-point comparison theorems between hyperbolic and eu- clidean geometry are known for various types of regions in the complex plane C ([5], [6], [7], [8]). These comparison theorems were motivated by work of Blatter [1] dealing with a characterization of univalent functions.
The proofs of these results rely upon coefficient estimates for certain classes of functions defined on the unit disk IDJ. The purpose of this note is to present new geometric proofs of two-point comparison theorems for convex regions. Our proofs show that these comparisons can be derived from the fact that the reciprocal of the density of the hyperbolic metric is a concave function on convex regions [10]. This concavity property actu- ally characterizes convex regions [4]. Other characterizations of convex regions in terms of hyperbolic geometry are given in [3].
We recall the known comparisons for convex regions. We begin with a brief discussion of hyperbolic geometry. Suppose n is a convex region in C with n
=1=c. Let An(w)ldwl denote the hyperbolic metric on n. For the unit disk A])(z)ldzl = Idzl/(l -lzI
2).The density An is determined from An(f(z)) = 1j[(1-lzI
2)1f'(z)l]' where f is any conformal mapping
Received June 1, 1999.
1991 Mathematics Subject Classification: 30C45.
Key words and phrases: the hyperbolic metric, convex region.
This research was supported in part by a grant from the Korean Council for University Education, 1998.
of the unit disk onto n. The hyperbolic distance between A, BEn is given by
dn(A,B) = inf l An(w)ldwl,
where the infimum is taken over all rectifiable paths
Iin n joining A and B. A path 8 connecting A and B is called a hyperbolic geodesic if
dn(A, B) = 1 An(w)ldwl·
For the unit disk
I a-b I
dlIJ)(a, b) = artanh 1- ab
and hyperbolic geodesics are arcs of circles that are orthogonal to the unit circle aID. Hyperbolic geodesics exist in n and are the images of hyperbolic geodesics in
JI)lunder a conformal map f : ID
---+n.
In addition to supporting hyperbolic geometry, n carries the euclidean geometry that it inherits as a subset of C. Two-point comparison theo- rems bound the euclidean distance lA - BI above and below in terms of the hyperbolic distance dn(A, B) and the values An(A) and An(B). For a convex region n and any
p2: 1 the lower bound
sinh(dn(A, B)) [ 1 1 ]
l/pA B
[2cosh(Pdn(A,B))]1/p An(A)p + An(B)p ~ I - I
and the upper bound
lA - BI < [2cosh(pdn(A,B))Jl/psinh(dn(A,B)) - [An(A)p + An(B)p}1/p
are known ([5] and [6]) .. Both bounds are sharp: equality holds in either for distinct A, BEn if and only if n is a half-plane and the euclidean line through A and B is perpendicular to the edge of the half-plane.
The lower (upper) bound is a nonincreasing (riondecreasing) function of p 2: 1. Therefore, the bounds for p = 1, namely,
sinh(dn(A, B)) [ 1 1 ]
(1) 2 cosh(dn(A, Bn An(A) + An(B) ~ lA - BI
< 2cosh(dn(A, B)) sinh(dn(A, B))
- An(A) + An(B)
are the strongest. The limiting cases
p=
00are the weakest:
sinh(dn(A, B)) < lA _ BI
exp(dn(A, B)) min{>.n(A), An(B)} -
exp(dn(A, B)) sinh(dn(A, B))
< .
- max{An(A), An(B)}
These weakest comparisons are invariant versions of the classical growth theorem
Izl < I (z)1 < Izl
1 + Izl - 9 - 1 - Izl
for normalized (9(0)
=0,9'(0)
=1) convex univalent functions 9 de- fined on]])l. We shall provide a new, geometric proof of the strongest inequalities (1).
2. Preliminaries
Two quantities associated with the hyperbolic metric play an impor- tant role in our work. These quantities are the connection
fn(w) = 2aIOgAn(w)
aw
and the Schwarzian
Sn(w) = 2 [0' ~;''''(w) _ (81;;: An (w))']
of!] 1
2ow (w) - 2 fn (w) . Note that
of!] 1
2ow (w) = Sn(w) + 2 fn (w) . Also,
ofn (w) = 2An(W)2
&w
follows from the fact that An(w)ldwl has curvature -4; that is,
~ log An (w) 4
iPa:~~n ( w)
-4= An(W)2 An(w)2 .
IBn(w)1 2An(w)2'
Various characterizations of convexity are conveniently expressed in terms of the connection and the Schwarzian.
PROPOSITION
1. Suppose
Qis a hyperbolic region in C. Then the following are equivalent.
(i)
Qis convex.
(ii)
A~is concave on
Q.(iii) IBn(w)1 +
~lrn(wW ~2An(W)2 for wE
Q.(iv) Irn(w)1
~2An(w) for wE
Q.The equivalence of (i), (ii) and (iii) is established in [4], but in different notation. The simple identities
8
2(...!...)
1
AllAn(W) 8w2 (w)
Irn(w)1
= An(W) ,
show that Theorem 1 of [4] contains the equivalence of (i), (ii) and (iii).
Note that (iii) implies (iv). The equivalence of (i) and (iv) is given in [2]. See [9] for a geometric proof that (i) implies (iv), as well as for a proof that equality holds in (iv) if and only if
Qis a half-plane. In other words, Irnl- 2An when
Qis a half-plane.
In addition to these characterizations of convex regions we require an elementary result for a differential inequality.
PROPOSITION
2. Suppose u, v
EC2[- L, L], v"
~4v and u" = 4u. If u(L) = v(L) and u( -L) = v( -L), then either v = u on [-L, L] orv > u on (-L,L).
For a proof of this result see [6J.
3. Main result
THEOREM
1. Suppose
Qis a convex region in C with
Q-=I- C.
(i) For A,B
EQsinh(dn(A, B)) [1 1]
(2) 2cosh(dn(A, B)) An(A) + An(B) ~ lA - El·
(ii) For A,B
Ef2
(3) IA-BI < 2cosh(dn(A,B)) sinh(dn(A, B)).
- An(A) + An(B)
Equality holds in (2) or (3) for distinct A and B if and only if n is a
half-plane and the euc1idean line through A and B is perpendicular to the edge off2.
Proof. (i) Fix A, B E f2 with A
=1=B. Because f2 is convex, the straight line segment /
:=[A,B] is contained in f2. Let / : w
=w(s), -L ::; s ::; L, be a hyperbolic arclength parametrization of /. This means that w'(s) = e
i8jAn(w(s)), where e is the argument of B-A, and 2L is the hyperbolic length of /. Set
1
v(s)
:=An(w(s)) ' -L::; s ::; L.
Then
v'(s) 2 Re {OAn(W(s)) W'(S)}
An(W(S))2 OW
1 { iIJ}
= -
An(W(S)) Re fn(w(s))e -v
2(s)Re {fn(w(s))e
i8}.Next, we compute the second derivative of v.
v"(s) -2v(s)v'(s)Re {fn(w(s))e
i8} -v
2(s)Re {e
i8~ f(w(s)) }
2v
3(s)Re
2{fn(w(s))e
iO} -v
2(s)Re {e
iO~ fn(w(s)) } . Now,
ds fn(w(s)) d of n of n - -
~(w(s))w'(s)
+
~(w(s))w'(s)[ uw uw]
0 "81
2 e~ iOSn(w(s)) + "2fn(w(s)) An(W(S}) + 2An(w(s))e- ,
so that
v"(s) - 2v
3(s)Re 2{rn(w(s))eiO } v 2
(s) { [ 1 ( 2] 2iO}
- An(W(S)) Re Sn(w(s)) + 2fn w(s)) e -2v2(s )An( w(s))
= v
3(s) [2Re 2 {fn(w(s))eiO }
-Re { [Sn(w(s)) + ~fn(w(s))2] e 2iO }] - 2v(s)
_ v
3(s) [lfn(w(s))1
2-Re {[Sn(w(s)) - ~fn(w(s))2] e 2iO }] - 2v(s)
since
Because n is convex we obtain
v"(s) < v
3(s) [Irn(w(s))1 2 + ISn(w(s)) - ~fn(w(s))21] - 2v(s)
< v
3(s) [lrn(w(s))1 2
+ ISn(w(s))1 + ~ Ifn(w(s))1 2 ] - 2v(s)
< v
3(s) [(2An(w(s)))2 + 2An(w(s))2] - 2v(s)
= 4v(s).
Note that if v"(s) = 4v(s) , then Ifn(w(s))1 = 2An(w(s)), and so n must be a half-plane.
Let u(s) = ccosh(2s) + dsinh(2s) be the solution of u"(s) = 4u(s) . that satisfies the boundary conditions u( -L)
=v(-L) andu(L) =v(L).
Then
c d
1 1
v(L)+v(-L)
~+~2 cosh(2L)
-2 cosh(2L) , veL) - v(-L)
2sinh(2L) .
Proposition 2 implies that either v = u on [-L,L] or v > u on (-L,L).
Now,
lA - El
=!Idw
"I l =1 -L
LIw'(s)lds
-I: An(~(s)) = I: v(s)ds
> 1: u(s)ds
=csinh(2L)
sinh(2L) [ 1 1 ]
2cosh(2L) An(A) + An(B) - ~ tanh(2L) [An~A) + An~B)] .
Note that equality implies v = u and so v"(s) = 4v(s) on [-L, L]; this implies 0 is a half-plane. Since 2L ;::: dn(A, B) with equality if and only if 'Y is a hyperbolic geodesic, we conclude that
1 [ 1 1 ]
lA - BI > "2 tanh(dn(A, B)) An(A) + An(B)
= sinh(dn(A, B)) [ 1 1 ]
2cosh(dn(A, B)) An(A) + An(B) .
Equality implies that 0 is a half-plane and the straight line segment 'Y
=[A, Bl is a hyperbolic geodesic. In a half-plane the hyperbolic geodesics are arcs of circles orthogonal to the edge of the half-plane and segments of lines perpendicular to the edge of the half-plane. Hence, if equality holds, then 0 is a half-plane and [A, B)
ispart of a line perpendicular to the edge of the half-plane.
All that remains is to show that equality holds when 0 is a half-plane and [A, B] is perpendicular to aO. It suffices to consider the particular half-plane JHI = {w : Im{w} > O}. Then AJ8[(W) = 1/[2Im{w}]. Fix u
ElR and 0 < a < b. Then for A
=u+ia and B
=u+ib, IA-BI
=b-a, dJ8[(A, B) = pog~, and AJ8[(A) = 1/(2a), AJ8[(B) = 1/(2b). It is now straightforward to check that equality holds in (2).
(ii) Consider A, B E 0 with A
=1=B. Let 8 be the hyperbolic geodesic
joining A to B. Suppose 8 : w
=w(s), -L ::; s ::; L, is a hyperbolic
arclength parametrization of 8. Then 2L
=dn(A, B), the hyperbolic
length of 8, and w'(s)
= ei8(s) /An(W(S)), where
eiO(s)is a unit tangent to
oat w(s). Set
Then
V(s)
:=An(w(s)).
Kh(W(S),O)
V'(S) 2Re {OZ(W(S»W'(s)}
{
oAn ei8 (S)}
2Re Ow (w(s) An(w(s»
Re {rn(w(s»eill(S)}.
Because n is convex we obtain
(4) 1V'(s) I
~Irn(w(s»1
~2An(w(s»
=2V(s).
Next, we calculate the second derivative of V.
V"(s) Re {ei8(S) ~ rn(w(s» } + Re { rn(w(s» ~ ei8(S)}
An(~(s»Re {[Sn(w(s» + ~rn(w(s)?] e 2ill (S)}
+2An(w(s» + di:)Re {ieill(s)rn(w(s»)}.
The hyperbolic curvature of 0 is
Ke(W(S),o) + 2Im {~(w(s»ei8(s)}
An(w(s»
Ke(W(S),o) + III {rn(w(s»ei8(s)}
An(w(s»
where
1 {w"(s)}
Ke(W(S), 0)
=IW'(s)IIm w'(s).
is the euclidean curvature of'Y at w( s). Since 0 is a hyperbolic geodesic, Kh(W(S),o) =0, so
Im {rn(w(s))ei8(s)} = -Ke(W(S), 0).
From w'(s)
=eill(s) jAn(w(s)), we obtain Im {w"(s)}
=dOes)
w'(s) ds
D
=and so
dO(s)
Ke
(w(s),8) =
~AI1(W(S)).Thus,
dO(s) = _ 1 I {r ( ()) ill(s)}
ds AI1(W(S))
m 11w s e
and so
V"(s) = AI1(~(s))Re {[SI1(w(S)) + ~rl1(w(s))2] e 2i11 (S)}
+2AI1(w(s)) + AI1(~(S)) Im
2{rl1 (w(s))e ill (s)}
since
Finally,
V"(s) = AI1(~(S)) (~lrl1(w(s))12 + Re {SI1(w(s))e 2iO (s)} ) + 2AI1(w(s)), since
Because n is convex, we get
V"(s) ::; AI1(~(S)) (~lrl1(w(s))12 + ISI1(w(s))I) + 2An(w(s)) ::; 4AI1(w(s))
=4V(s).
Now, let U(s) = Ccosh(2s) + Dsinh(2s) be the solution of U"(s)
=4U(s) that satisfies the boundary conditions U(- L )
=V (- L) and U(L) = V(L). Then
C = V(L) + V( -L) = AI1(A) + AI1(B) > 0 2cosh(2L) 2cosh(dl1(A, B)) , V(L) - V(-L)
2sinh(2L) .
Proposition 2 implies V = U on [-L, L] or V > U on (-L, L). Now, if U > 0 on [-L,L]
lA - BI < lldwl a = 1
-LLIw'(s)lds 1
Lds 1
Lds
-L An(W(S)) = -L V(s)
< 1
-LLCcosh(2s) ds + Dsinh(2s)'
with equality if and only if the hyperbolic geodesic
~is the euclidean line segment [A, B]. We show that U > 0 on [-L, L]. Note that
1 C cosh(2s) - D sinh(2s) U(s) = C2 cosh2(2s) - D2 sinh
2(2s)
1 cosh(2s) - 7sinh(2s) - Cl + (1- 7
2)sinh2(2s)' where
D V(L) - V(-L) 1
7
:=C
= -=-=V-:--:=(Le'-)
-+-=-:V=-"-(_-L::7-) tanh(2L)
We will show that 171 :::; 1; in particular, this shows that U > 0 on [-L, L]. From (4) we have
-2 < V'(s) < 2 - V(s) -
for s
E [-L, L]. If we integrate these inequalities over the interval [-L,L]' then we obtain
V(L)
-4L :::; log V( -L) :::; 4L, or
e-4L < V(L) <e4L . - V(-L) -
The function h(t) = (t-1)j(t+ 1) is increasing for t > -1 since h'(t) =
2j(t + 1)2 > O. Thus,
h(e-4L ) < h ( V(L) ) < h(e4L)
- V(-L)
~,
or
l
-LLCcosh(2s) ds + Dsinh(2s)
V(L) - V(-L)
- tanh(2L):::; V(L) + V(L) :::; tanh(2L).
This demonstrates that 171 :::; 1. Also, we conclude that 7 = ±1 implies JV'(s) I = 2JV(s)l, or Ifn(w(s))1 = 2An(w(s)), which means n is a half- plane. Now
~ 1
Lcosh(2s) -
7sinh(2s) ds
C
-L1 + (1 -
7 2 )sinh
2(2s)
=
~ 1
Lcosh(2s) ds
C
-L1 + (1 -
7 2 )sinh
2(2s) since sinh(2s)/[1 + (1 -
72)sinh
2(2s)] is an odd function. Also,
j
L---'---",---- ds cosh(2s) < 1
Lcosh(2s) ds
-L
1 + (1 -
72)sinh
2(2s
-L= sinh(2L) - sinh(dn(A, B)),
and equality implies
7= ±l. By combining our inequalities, we obtain lA _ BI < sinh(dn(A, B)) = 2cosh(dn(A, B)) sinh(dn(A, B))
- C An(A) + An(B)
and equality implies n is a half-plane and 6 = [A, B] is a hyperbolic geodesic, so [A, B] must be perpendicular to an.
Conversely, it is routine to show that if n is a half-plane and [A, B]
lies on a line orthogonal to an, then equality holds in (3).
4. Acknowledgement
Both authors want to thank Professor Kang Tae Kim for the kind invitation to visit POSTECH May 2-3, 1997. This work was begun during the visit. The second author thanks Professor Jae Keol Park of Pusan National University for arranging his visit to Korea in May, 1997.
Also, the first author thanks Department of Mathematics of POSTECH
for their allowing her visit to the Department during the 1998 academic
year.
References
[1] C. Blatter, Ein Verzerrungssatz fUr schlichte Funktionen, Comment. Math.
Helv.
53(1978), 651-659.[2]
R.Harmelin, Hyperbolic metric, curvature of geodesics and hyperbolic discs in hyperbolic plane domains, Israel J. Math.
70 (1990), 111-128.[3]