PACIFIC JOURNAL OF MATHEMATICS

Vol. 20, No. 3, 1967

INEQUALITIES FOR FUNCTIONS REGULAR

AND BOUNDED IN A CIRCLE
CECIL CRAIG, J R . AND A. J. MACINTYRE

This paper is concerned with functions w = f(z) regular

and satisfying the inequality \f(z) \ < 1 in | z \ < 1. This class
of functions will be denoted E.
We determine conditions on zlt z2, zz and w1} w2, w% for

to be possible with an f(z) of E. In particular to map the

2kπίl3
vertices of the equilateral triangle Zk = re into the vertices
Zkπil
of another taken in the opposite direction Wk = pe~ * we
2
must have p ^ r . The extremal function associated with this
problem is w = z2. It seems convenient to discuss the fixed
point if any of the mapping of | z \ < 1 into | w \ < 1. We
include a simple proof of the theorem of Denjoy and Wolff
that if no such fixed point exists then there is a unique dis-
tinguished fixed point on | z | = 1. We give several results
restricting the position of the interior or distinguished
boundary fixed point in terms of the location of a zero of
f(z) or the value /(0).

The theorem of Pick asserts that if f(z) is in E then

S D(zu z2) where the nonEuclidean distance

+ y ' *2 with d(zl9 z2) =

λ
D(zu z2) = ± log 1 #2

2 1 - d(zu z2)

Equality holds if and only if / sets up a Mobius transformation. It

follows from Pick's theorem that there can be at most one fixed
point of w — f(z) in | z \ < 1 unless f(z) = z. It is usually sufficient
when / has an interior fixed point at z = a(Φ 0) to suppose 0 < a < 1.
Our first four theorems give information about the relative posi-
tions of zeros of /, an interior fixed point, and the value /(0). We
exclude the case where f(z) = z.

THEOREM 1. Let feE and /(0) Φ 0. Then f has no zeros in

\z\ < 1/(0)1; and has a zero on \z\ = |/(0)| if and only if f de-
termines a Mobius transformation.

Proof. The image of \z\^ |/(0)|, which we denote by C under

the transformation w = (z + /(0))/(l + f(O)z) is a circular disc C" having
nonEuclidean center /(0) with boundary passing through the origin.
The function w = f(z) takes the closed disc C inside C in the case /
449
450 CECIL CRAIG, JR. AND. A. J. MACINTYRE

is not a Mobius transformation so that f(z) Φ 0 for zeC. If / is

linear, nonEuclidean distances are preserved and /(0) is on the
boundary of C".

THEOREM 2. Let feE and let z = a be a fixed point of f with

2 2
0 < a < 1. Then f has no zeros inside \ z — α/(l + a ) \ = a/(l + a )
and has a zero on the boundary if and only if f determines a
Mobius transformation.

Proof. The conclusion follows directly from Pick's theorem since

the circle described is the nonEuclidean circle with nonEuclidean
center z — a.
If /(0) is known in addition to the existence of an interior fixed
point a(Φ 0), then these two results can be combined to give a larger
region which is zero-free, namely the union of the two closed discs.
The boundary zero of / occurs at z — /(0) when / is a Mobius trans-
formation.

THEOREM 3. If feE and /(0) Φ 0, then there can be no fixed

point interior to the circle Cx\ \ z \ — (1 — l/ 1 — |/(0) |2)/|/(0) |; and a
fixed point on the boundary at z0 = β ΐarg/(0) (l - i / 1 - |/(0) |2)/|/(0) | only
if f determines a Mobius transformation.

Proof. The nonEuclidean midpoint of the segment from 0 to /(0)

is z0 (See Figure 1). A displacement of all points inside Cx by w =
f(z) insures there can be no fixed point interior to Clm The boundary
case is clear.

Figure 1.

If / is known to have an interior fixed point, an improvement

over Theorem 3 can be made as to its location based on a knowledge
of /(0). This is indicated in:
 INEQUALITIES FOR FUNCTIONS REGULAR BOUNDED 451

THEOREM 4. Let feE. Iff has a fixed point z = a(Φ 0), in

z I < 1, then a lies inside the circle C2 passing through z0 (Figure 1)
with center at the geometric inverse of /(0), relative to the unit
circle and is on the boundary if and only if f sets up an elliptic
Mobius transformation.
Proof. This is a direct consequence of the inequality D(a, 0) Ξ>
D(a, /(0)) where the point z = a is considered variable and /(0) is
fixed. The assumed interior fixed point is nearer /(0) than the origin
in the nonEuclidean sense, except when / is linear. This requires an
investigation of the nonEuclidean perpendicular bisector of the radial
segment from 0 to /(0). Straight lines of the Poincare model are
Euclidean circles orthogonal to the unit circle. The Euclidean circle
C2 passing through the point zQ and orthogonal to | z \ — 1 is the one
described in the statement of the theorem.
Theorem 4 provides a simple proof of the Theorem of Denjoy on
the fixed points of analytic transformations of the unit circle into
itself [3]. It is convenient to develop the argument by formulating
several variants of Theorem 4. f(z) is supposed to belong to E.

THEOREM 4 A. // /(0) Φ 0 and arg/(0) = Θ, then any interior

fixed point must lie in the half plane B(eίθz) > 0.
The half plane evidently contains the circle C2 of Theorem 4.

THEOREM 4B. The nonEuclidean bisector of the nonEuclidean

segment joining x and f(x) divides the unit circle into two parts.
Any interior fixed point must lie in the part containing f(x) unless
the function sets up an elliptic linear transformation when the fixed
point must lie on the bisector.
This statement is equivalent to that of Theorem 4. We have
only to apply Theorem 4 to w = TfiT^z) where T is a linear trans-
formation of I z I < 1 into itself which carries x to the origin.

THEOREM 4C. If X and f(x) have the same argument, then any
interior fixed point must lie on the same side as f(x) of the circle
through x and orthogonal to the radius Ox and to \ z \ = 1.
This follows from Theorem 4 A. We consider w — Tf{T~xz) where
T carries x to the origin and the diameter through x into itself.
Now consider z = g(w) the solution of wf(z) = z. From Rouche's
theorem g(w) is regular and one valued for w in | w \ < 1. Let
0 < w < wf < 1. Apply Theorem 4 C to F(z) = w'f(z). Let a = g(w).
We know that F(a) = w'a/w. Any fixed point of F(z) and that is to
say g(wf) must lie in the smaller part of the unit circle partitioned
452 CECIL CRAIG, JR. AND A. J. MACINTYRE

as in Theorem 4 C. If g(w) does not tend to a fixed point of | w | < 1

as w~>1 by positive values, it must converge to a point of \z\ = 1.
This point on | z \ == 1 is the Denjoy distinguished fixed point. Calling
such points D fixed points it is clear that Theorem 4 applies to these
as well as to interior fixed points.
We shall next be concerned with special cases of three point
interpolation by feE. The problem first considered is that in which
we require the vertices of an isosceles triangle to be mapped by /
into vertices of another isosceles triangle.

THEOREM 5. A necessary and sufficient condition for the exist-

ence of a function feE taking points z0, 0, z0 into w, 0, w,
respectively, is that w = f(zQ) lies in lens B = {ί 11 = zQζ, ζ e A},
where A is the lens formed by the two circular arcs passing through
- 1, z0, + 1 and - 1, z0, + 1.

Proof. This follows from an inequality of G. Julia [4, 74-78]

which for our problem is expressed by D(w/z0, w/z0) ^ D(z0, z0). Since
D is a monotone increasing function of d, it is sufficient for our
purpose to use d and we shall refer to this as the nonEuclidean
distance.
Let 8 = I (z0 — zo)/(l — z20) | and ζ = w/zQ = x + iy. Then the basic
inequality becomes | (ζ - ζ)/(l - ζ2) | ^ δ or 2\y \/\ 1 - (x + iy)2 \ g δ.
On squaring and simplifying we have 4^/2(l — δ2) ^ δ2(l — {x2 + y2})2.
After taking square roots and rearranging we obtain

If y ^ 0, ζ lies on or below one circular arc; for y < 0, ζ lies on or

above the other arc, the reflection of the first in the real axis. These
arcs form the boundary of a lens. To see that the boundary curves
pass through zQ and z0, consider the case of equality | ζ - ζ |/| 1 - ζ21 =
\z0 — z0 I/I 1 — z\ I. This equation describes the locus of a point which
is a fixed nonEuclidean distance from its conjugate, in this case the
nonEuclidean distance being d(z0, z0). The lens just described is labeled
A in Figure 2. To complete the proof one notes that w — zoζ, for
ζ G A, is the set of points of lens B.
A slightly more general result than Theorem 5 can be obtained.
We require / to be real at a real point h as well as to take conjugate
values at the conjugate pair z0, zQ.

THEOREM 6. A necessary and sufficient condition for the exist-

ence of a function feE taking z0, h, z0 into w, h', w, respectively,
where h and K are real numbers, is that w = f(z0) lies in a lens
 INEQUALITIES FOR FUNCTIONS REGULAR BOUNDED 453

Figure 2.

Figure 3.

= {t\t = h')/(l + h'W), where

Ί
hz0
and A is the lens described in Theorem 5.

Proof. The proof depends on the fact that the composition of

functions in E is again in E. The transformation Z = (z - h)/(l - hz)
takes h to zero with zQ and z0 going to conjugate points Zo =
(So — h)/(l — hz0) and Zo. Since this transformation preserves nonEuclidean
distances, z0 is moved to Zo on circular arc C3 which passes through
— 1, z0, + 1. By Theorem 5, a necessary and sufficient condition for
the existence of a function of the class E taking 0 to 0 and Zo, Zo
into conjugate points, say W and W, is that W lies in lens B
described in the statement of the theorem. Denote by G the image
of B under the transformation w = (W + h')/(l + h'W). We conclude
that w = f(z0) must lie in G, the lens enclosing h' with end points
9
(Zo + h')/(l + h'ZQ) and (h - ZQ)/(1 - h'Z0) on C3 and C4, respectively.
454 CECIL CRAIG, JR. AND A. J. MACINTYRE

If f(z) is real for all h, — 1 < h < + 1, we have the Caratheodory

theorem [1, 53] which asserts that if feE and if, furthermore, / is
real for z real, then a point z inside lens A has its image f(z) also in
this lens.
Finally, we investigate the Julia inequality in the case of a
reversed equilateral triangle.

THEOREM 7. A necessary and sufficient condition for the exist-

ence of a function feE which maps the vertices of the equilateral
triangle, r, ra), rω2 into the vertices of the reversed equilateral
triangle p, pw2, pω, respectively, is that p <L r2.

Proof. The result is obtained by investigating the Julia condi-

tion: D(AΪ/al, Aί/ai) g D{a'2y αj), where
„, __ r ( ω - l ) n, __ r(ω2-!) A, _ p(ω* - 1) ,, _ p(ω - 1)
1 —r2ω 1 — r ω" 1 — p ω2 1 —pω

and simplifying the somewhat involved expression. The computation

is omitted.
In the extreme case p ~ r2, the function w = z2 performs the
required interpolation.

BIBLIOGRAPHY

1. C. Caratheodory, Conformal Representation, Cambridge Math. Tracts, No. 28.

2. , Theory of Functions, Vol. II, Chelsea, 1960.
3. A. Denjoy, Sur Viteration des fonctions analytiques, C. R. Acad. des Sc. Paris
182 (1926), 255-257.
4. G. Julia, Principes geometriques d'analyse, Premiere partie, Paris 1930.

Received January 3, 1966.

UNIVERSITY OF CINCINNATI
UNIVERSITY OF CALIFORNIA, DAVIS