ANNALES

POLONICI MATHEMATICI
LVI.1 (1991)

On uniformly convex functions

by A. W. Goodman (Tampa, Fla.)

Abstract. We introduce a new class of normalized functions regular and univalent in

the unit disk. These functions, called uniformly convex functions, are defined by a purely
geometric property. We obtain a few theorems about this new class and we point out a
number of open problems.

1. Introduction. An earlier paper [3] introduced the class UST of uni-

formly starlike functions. We now consider the similar concept of uniformly
convex functions. Let CV denote the usual class of convex functions
∞
X
(1.1) f (z) = z + an z n .
n=2

These are normalized functions regular and univalent in E : |z| < 1, for
which f (E) is a convex domain.
Definition 1. A function f (z) is said to be uniformly convex in E if
f (z) is in CV and has the property that for every circular arc γ contained
in E, with center ζ also in E, the arc f (γ) is a convex arc. We let UCV
denote the class of all such functions.
A directed arc Γ (t), a < t < b, is said to be convex if the argument of
the tangent to Γ (t) is a nondecreasing function of t [2, Vol. I, pp. 109–110].
In our case the direction of Γ (t) = f (γ) is the one dictated by the direction
of γ which is the usual counterclockwise direction on a circle.
In [2, p. 110] we proved that if any arc γ is given by z(t), then f (γ) is
convex iff
 00
z (t) f 00 (z) 0

(1.2) Im 0 + 0 z (t) ≥ 0
z (t) f (z)
for all z on γ.

1991 Mathematics Subject Classification: Primary 30C45; Secondary 30C50.

Key words and phrases: univalent functions, convex functions, coefficient bounds.
88 A. W. Goodman

For a circular arc with center ζ, set z = ζ + reit . Then z 0 (t) = i (z − ζ)

and z 00 (t) = −(z − ζ). A brief computation using (1.2) will give
Theorem 1. Let f (z) have the form (1.1). Then f (z) is in UCV iff
 00 
f (z)
(1.3) 1 + Re (z − ζ) ≥ 0
f 0 (z)
for every pair (z, ζ) in the polydisc E × E.
Thus all the properties of functions in UCV are contained implicitly in
the relation (1.3). However, obtaining these properties is not always easy.

2. Functions with positive real part on the polydisc. Let P (2)

denote the set of functions
XX
(2.1) P (z, ζ) = 1 + bmn z m ζ n
m+n>0

that are regular in E ×E and satisfy the condition Re P ≥ 0 in that domain.

Such functions have been the subject of numerous investigations. However,
a representation formula for all functions in P (2) is still missing [4].
As my colleague V. Totik suggested, if we set ζ = eiα z in (2.1) we obtain
X∞
iα
(2.2) F (z) ≡ P (z, e z) ≡ 1 + Bn z n
n=1
where for n ≥ 1
n
X
(2.3) Bn = bn−k,k eikα .
k=0

If P (z, ζ) ∈ P (2) , the classical Carathéodory Theorem applied to F (z) gives

|Bn | ≤ 2 for all n ≥ 1, and all real α. Then on integrating Bn B n on a
suitable circle we obtain
Xn
(2.4) |bn−k,k |2 ≤ 4 , n ≥ 1 ,
k=0
(2)
for P (z, ζ) in P .
Let
f 00 (z)
(2.5) Q(z, ζ) ≡ 1 + (z − ζ)
f 0 (z)
and let Q(2) = {Q(z, ζ) | f (z) ∈ UCV}. Then Q(2) is properly contained
in P (2) and in fact if Q(z, ζ) has the form (2.1), then clearly bmn = 0 for all
m ≥ 0 and n ≥ 2. Further, if we put ζ = z we see that bm−1,1 = −bm,0 for
all m ≥ 1. Although these restrictions on Q(z, ζ) may be interesting, appar-
ently they are not very helpful in obtaining properties for the class UCV.
 On uniformly convex functions 89

3. Properties of uniformly convex functions. It is clear from the

definition of UCV that the class is invariant under the rotation eiα f (e−iα z).
Hence we may assume w.l.o.g. that a2 ≥ 0 in (1.1). We can let ζ → 1−
in (2.5) and obtain
 ∞
X 
(3.1) Re(Q(z, 1)) = Re 1 − 2a2 + cn z n ≥ 0 .
n=1

This gives 0 ≤ 1 − 2a2 or 0 ≤ a2 ≤ 21 . But if a2 = 12 , then cn = 0 for all

n > 0 and hence
f 00 (z)
(3.2) 1+z 0 (z − 1) ≡ 0 .
f (z)
Now (3.2) holds if and only if f (z) = − ln(1−z). For this function Q(z, ζ) =
(1 − ζ)/(1 − z) and for suitable selection of z and ζ we have Re Q(z, ζ) < 0.
We have proved
Theorem 2. If f (z) ∈ UCV, then |a2 | < 1/2.
The sharp upper bounds for |an | in the class UCV are not known, but
we have
Theorem 3. If f (z) is in UCV, then |an | ≤ 1/n for every n ≥ 2.
P r o o f. We use the symbol f (z)  g(z) to indicate that the power series
for f (z) is dominated by the power series for g(z) [2, Vol. I, pp. 81–83]. We
set ζ = −z in Q(z, ζ) and if f (z) ∈ UCV, then
f 00 (z) 1+z
(3.3) Q(z, ζ) = Q(z, −z) = 1 + 0
2z  .
f (z) 1−z
Then
f 00 (z) 2z
(3.4) 2z 0
 .
f (z) 1−z
Integration gives ln f 0 (z)  − ln(1 − z). Consequently f 0 (z)  1/(1 − z)
and hence |an | ≤ 1/n.
The following example will be useful.
Theorem 4. The function
∞
z X
(3.5) F1 (z) = =z+ An−1 z n
1 − Az n=2
is in UCV iff |A| ≤ 1/3.
P r o o f. By a rotation we may assume that 0 ≤ A in (3.5). A simple
computation shows that for this function
1 + Az − 2Aζ
Q(z, ζ) = .
1 − Az
90 A. W. Goodman

We set z = reiθ and ζ = ρeiϕ . Then Re Q(z, ζ) ≥ 0 iff

Re(1 + Areiθ − 2Aρeiϕ )(1 − Are−iθ ) ≥ 0
or
(3.6) 1 − 2Aρ cos ϕ − A2 r2 + 2A2 rρ cos(ϕ − θ) ≥ 0 .
It is clear that the minimum of the expression on the left side of (3.6) occurs
when r = ρ = 1, ϕ = 0 and θ = π. (Thus, ζ = 1 and z = −1.) These values
yield 1−2A−3A2 ≥ 0, and this is true for 0 ≤ A ≤ 1/3. Thus, the condition
is sufficient for F1 (z) to be in UCV. By a limit argument as z → −1+ and
ζ → 1− the condition is also necessary.
As a corollary of Theorem 4 we see that the set UCV has infinitely many
members.
It is natural to look for transformations which preserve the set UCV.
The rotation eiα f (e−iα z) is one such, but no other transformation seems
to be available. Pommerenke [2, Vol. II, p. 109] introduced the concept of
the linear-invariant family M and showed that numerous theorems about
the family M followed immediately once we have proved that M is a linear-
invariant family. By definition M is a linear-invariant family if
f (ϕ(z)) − f (ϕ(0)) z+c
(3.7) Λϕ [f ] ≡ , ϕ(z) = ,
ϕ0 (0)f 0 (ϕ(0)) 1 + cz
is also in M for every f in M and every c in E. If we apply (3.7) to
F1 (z) = z/(1 − Az) we find that
 
z z A−c
(3.8) Λϕ = , B= .
1 − Az 1 − Bz 1 − cA
Now set A = 1/4 < 1/3 and c = −1/2 . Then F1 (z) is in UCV, but B =
2/3 > 1/3, so Λϕ is not in UCV. We have proved
Theorem 5. The family UCV is not a linear-invariant family.

4. The sets UST and UCV. We recall the classic Alexander Theorem
that if f (z) is given by (1.1), then f (z) is in CV iff
(4.1) F (z) ≡ zf 0 (z)
is in ST, where CV and ST are the usual normalized families of convex and
starlike functions. What is the situation when we prefix the word “uni-
formly”?
To prove that (4.1) does not give a one-to-one correspondence between
the sets UST and UCV we need two examples.
 On uniformly convex functions 91
√
Lemma 1. The function F (z) ≡ z − Bz 2 is in UST iff |B| ≤ 3/4.

As we mentioned in [3], the proof of this lemma is a simple exercise in

calculus, but the computation is rather long so we omit the details.

Lemma 2. The function f (z) ≡ z − Az 2 is in UCV iff |A| ≤ 1/6.

P r o o f. If f (z) = z − Az 2 , then
 00 
f (z) −2A(z − ζ)
1 + Re (z − ζ) = 1 + Re .
f 0 (z) 1 − 2Az
But for 0 < A < 1/2,
−2A(z − ζ) 2A(z − ζ) 4A
(4.2) 1 + Re ≥1− ≥1−
1 − 2Az 1 − 2Az 1 − 2A
and hence in E × E
f 00 (z) 1 − 6A
1 + Re 0
(z − ζ) ≥ ≥0
f (z) 1 − 2A
iff A ≤ 1/6. But equality can occur when z = 1 and ζ = −1.
√
Now set F (z) = z−Bz 2 where B = √ 3/4 and hence F (z) is in
√UST. Then
(4.1) gives f (z) = z − Az 2 where A = 3/8 ≈ 0.216. Since 3/8 > 1/6,
the corresponding f (z) is not in UCV. The converse relation may hold. It
may be that f (z) in UCV implies that F (z) = zf 0 (z) is in UST, but I have
not been able to prove or disprove this statement.

5. A sufficient
P∞ condition. It is well-known [1] that if f (z) is given
by (1.1) and n=2 n|an | ≤ 1, then f (z) is in ST. In [3] we proved that if
∞ √
X 2
(5.1) n|an | ≤ ,
n=2
2

then
√ f (z) is in UST. However,
√ it was conjectured in [3] that the constant
2/2 can be replaced by 3/2 in (5.1).

Theorem 6. Suppose that f (z) is given by (1.1). If

∞
X 1
(5.2) n(n − 1)|an | ≤ ,
n=2
3

then f (z) is in UCV. Further , the constant 1/3 in (5.2) cannot be replaced
by a larger number.
92 A. W. Goodman

P∞
P r o o f. If (5.2) is satisfied, then n=2 n|an | ≤ 1/3. Hence in E × E
∞
X
n(n − 1)|an | |z n−2 |
00
f (z) n=2
(5.3) 1 + Re (z − ζ) ≥ 1 − ∞ |z − ζ|
f 0 (z) X
n−1
1− n|an | |z |
n=2
2/3
≥1− = 0.
1 − 1/3
But equality is attained in (5.3) when f (z) = z − z 2 /6, z = 1 and ζ = −1.

References

[1] A. W. G o o d m a n, Univalent functions and nonanalytic curves, Proc. Amer. Math.

Soc. 8 (1957), 598–601.
[2] —, Univalent Functions, Polygonal Publ. Co. Inc., Washington, N.J., 1983.
[3] —, On uniformly starlike functions, J. Math. Anal. Appl. 155 (1991), 364–370.
[4] W. R u d i n, Function Theory in Polydiscs, Benjamin, New York 1969.

DEPARTMENT OF MATHEMATICS
UNIVERSITY OF SOUTH FLORIDA
TAMPA, FLORIDA 33620-5700
U.S.A.

Reçu par la Rédaction le 1.9.1990