PACIFIC JOURNAL OF MATHEMATICS

Vol. 138, No. 1, 1989

INTEGRAL LOGARITHMIC MEANS FOR

REGULAR FUNCTIONS

C. N. LINDEN

For a function / , regular in the unit disc, integral logarithmic

means are defined by the formulae
2π ι
Γ 1 ί2π t
r J ) = l0g lf{reiθ
\Έt] '
for 0 < p < oo. These are related to

Moo(r,/) = s u p | l o g | / ( z ) | | (0<r<l)
|z|=r

when the latter increases sufficiently rapidly. Thus when λoo(f) > 1
the orders
, / Λ .. logΛf»(r,/)
^(/) = ll
? i " P log 1/(1-r)
are continuous at infinity in the sense that

a property which does not generally hold when λoo(f) < 1. It tran-
spires that in the extreme cases λoo(f) = λ i ( / ) + 1, and λoo{f) =
λ\(f) > 1, λp(f) is uniquely determined for 1 < p < oo.

1. Introduction, For a given function / , regular in the unit disc

Z)(0,l) = {z: \z\< 1} let

Mp(r,f) = l±J^ \log\ f(reiθWdθ\ (0<jp<oo),

Moo(r,/) = suplog|/(z)|
\z\=r

for 0 < r < 1. We consider the asymptotic values of these quantities

as r —• 1 in terms of the orders
^ogMp(rff)

Note that λ\(f) is equal to the Nevanlinna order of / , and λoo(f) is

the maximum modulus order of / , related by the classical inequalities

(i.i)

119
120 C. N. LINDEN

which are readily obtained from the Poisson-Jensen formula [6, p.

205].
Certain properties of λp{f) follow immediately from the Holder
inequalities (see [4, pp. 9 and 15] for the corresponding properties of
Mp(r9f))\ for example
(A) λp(f) is an increasing function of p (0 < p < oo),
(B) pλp(f) is convex on the interval (0, oo).
In contrast, AQO(/) does not generally fit naturally into this context.
For elementary calculations show that if 0 < a < 1 then
a 1
F(z) = exp{(l + z)' - (1 - z)- } (|z| < 1)
satisfies λoo(F) = α, while λp(F) = 1 - l/p for p > 1. Nevertheless,
we will show that
(1.2) λ
provided that λoo(/) is sufficiently large by proving the following re-
sult.
THEOREM 1. Iff is regular in D(0,l) and λco(f) > 1, then
(1.3) λp(f)<λoo(f),
(1.4) A o o ( / ) < W ) + l/p
forO< p<oo. Thus (1.2) λβ/έfc wλeΛ ^ ( Z ) > 1.
The following corollary is deduced readily from Theorem 1 in §4.
COROLLARY 1. Iff is regular in D(0,l) and λ^f) > 1, then
(ί) p(Aoo(/) -λp(f)) is an increasing function ofp on (0, oo), with
range contained in [0,1],
(ii) λp(f) + l/p is a decreasing function ofp on (0, oo].
When p = 1, the inequalities (1.3) and (1.4) are equivalent to (1.1);
in the case p = 2 they have been obtained by Sons [5], As far as
one extreme case of the inequalities (1.1) is concerned, it is readily
observed that condition (A) shows that the equality λoo(f) = λ\(f)
implies that λp(f) = λ\(f) for p > 1. In the other extreme case
represented by
(1.5) λO0(f) = λι(f)+l
λp(f) is also completely determined when p > 1, since Corollary 1 (ii)
implies
λoo(f) < λp(f) + l/p < W ) + 1 ( ! < / > < oo).
Thus we obtain a second corollary.
 INTEGRAL LOGARITHMIC MEANS 121

COROLLARY 2. / / / is regular in D(0,1) and λoo(/) = h{f) + 1,

then

2. Preliminaries for the proof of Theorem 1. In this section, we

assemble some background material needed for the proof of Theorem
1. We put λoo(/) = λ and, when λ is finite, let μ be the integer
satisfying
λ<μ<λ+\.
Then for each given positive number ε we have
λ ε
(2.1) log|/(r^)|<(l~r)- - (r0 < r < 1, 0 < θ < 2π)
for some r$ in (0,1).
We later seek lower bounds for log|/(re ί(9 )| by considering a fac-
torisation based on the zero sequence {am} off in D(0, l)\{0}, each
zero being counted according to multiplicity. Let

This leads to the factorisation

(2.2) f(z) = g(z)zsB(z,{am},μ),
where
(2.3) B{z) = B{z, {am}, μ) = f[ b(z, am,μ),
m
s is a nonnegative integer, and g(z) is regular and nonzero in D(0,1).
The result (1.4), is readily obtained for g(z) by a simple application
of a known theorem [1, p. 84]. We need to show that it also applies
to the factor B(z). We require some known results, the first being a
theorem of Tsuji [6, p. 224].
THEOREM A. For the canonical product B(z) defined by (2.3), and
positive ε we have
1-1/7 |2 ^ + 1 + ε
(2 4) IΛΛID/^IX Γ " \^m\
l-zaκ
and, if Cm denotes the disc D(am> (1 - \am\2Y+*) then

(2.5) i - '
1 - zam
1
when\<\z\< 1, z$\JmCm.
122 C. N. LINDEN

The constant K in (2.4) and (2.5) depends on ε, μ and {am}, or

on ε and / if we regard B(z) as defined by (2.2). As here, we will
subsequently use K to denote a positive constant, not necessarily the
same at each occurrence, but depending on parameters which will
normally be stipulated as appropriate. The symbol r0 will be used
similarly, but always restricted to the interval (0,1).
We require some information regarding the zero distribution of /
when λoo(f)=λ>l. Let the disc D(0,1) be covered by sets of the
form
q x q
S(q,k) = {z: 1 -2~« < \z\ < 1 -2~ ~ , πk2~ < argz < π{k+ 1)2"*}
for integers q and k satisfying
(2.6) q = 0,1,2,..., -2q < k < 2q - 1.
For the given function / let N(q, k, f) denote the number of zeros of
/ in S(q, k). Then for any positive ε there is a number #o> such that
(2.7) N(q,k,f)<2^q (q>q0),
for all relevant k in (2.6) [3, p. 21]. This inequality gives rise to a
bound to the sums occurring in (2.4) and (2.5), as estimated in [3, pp.
23-25].

THEOREM B. Let f be regular in D(0,1) with factorisation (2.2).

Then for each positive ε, and a > λ = λoo(f) > I, we have

H-l
< K(l - \z\)-λ~ε
m

for some r0 in (0,1).

As a final preliminary to the proof of (1.3) of Theorem 1, we esti-

mate Mp(r, f) according to the following lemma.

LEMMA 1. Let f be regular in Z)(0,1) and λ^f) = λ > 1. Then, if

ε > 0 and 1 < p < oc we have

(2.8) / π I log \B(reiθ, {am}, μ)\ \* dθ < K{\ - r)

Jo
for some constant K and 0 < r < 1.
 INTEGRAL LOGARITHMIC MEANS 123

We deal with the integral in (2.8) by covering the range of integra-

tion by [π/(l - r)] + 1 intervals of the form [τ + r - 1, τ + 1 - r] for
τ = 2k(l - r) and k = 0,1,2,..., [π/(l - r)], showing that
Γ
(2.9) Γ I log \B(reiθ)\ \p dθ < K(\ - r)1
Λ+r-1
for each τ. The method of proof indicates that the constant K need not
depend on τ. However, for convenience and without loss of generality,
we suppose that τ = 0 in the following proof. Thence we obtain (2.8),
as stated.
Without loss of generality, we assume
(2.10) \<r<\, l<\am\<\,
since the contribution to the integral (2.8), due to any zeros not satis-
fying this latter inequality is clearly bounded. For given r, let E denote
the set of integers m for which the exceptional discs Cm of Theorem
A intersect γr = {z: z = reiθ, r-\<θ<\-r}. By application of
(2.7), we have
(2.11) #(E)<K(l-r)-λ~ε,
where #(E) denotes the number of elements in the set E. We consider
the factorisation B = BιB2B3, where
b{z9am,μ),

~z )
T ^ ^ Γ \ - z a m
•
m m
meE meE
First we note that for any positive number ε, Theorems A and B
give

/Vog|Λ1(^)||^fl<^(lr)logf^
(2.12) 7r_i \l - r

where the constants K in (2.12) can be chosen to depend only on

e,μ, p, and the whole sequence {am}.
Next, the inequality
2
\l-zam\ ^
124 C. N. LINDEN

yields
1 i - Kl 2 < K#(E).
|log|* 2 (z)| l< V 1 ~zam
meE^
Hence (2.11) implies

(2.13) f ' \\og\B2{reiθ)\\p dθ < K{\ - r)x

Jr-\
It remains to consider Bj,.
Given z = re"9 in D(0,1) we have

(2.14) 1>

where am = rmeiθm. For each m in E we can find w with \w\ = r such

that

\w-am\ < (1 - K | 2 r 4 < ( ^ ) 3 ( 1 -r2m) < 1(1 - r2).

Thus
1 - r 2 < 2 ( l - r w ) < 2 ( l - r + |w-a O T |),
from which we obtain

Since
2
I* - am\|2 >
> 4rr sin2 2^(0
4rrm sin sin2 ^(θ
1(0 -- 00OT)) >> ^| sin^ 1 - θm),
and in (2.14),

Minkowski's inequality yields

/ ' ^ogφidre^WPdθ
Jr-\
<K#(E)p(l-r)+K f ' [ V; log 1 + —
λ V^r 9i 2 2
f- { ( \6(l-r ) \Y
< K#(E)p(l -r)+ K#(E)P / log 1+ — ^ — ^ - dt
Λ-i V V « • " - • '
p
<K#{E) {\-r).
 INTEGRAL LOGARITHMIC MEANS 125

The inequality (2.9) with τ = 0 now follows from (2.11), (2.12), (2.13)
and this last inequality, so that Lemma 1 is proved.
3. The Proof of Theorem 1. We begin the proof of Theorem 1 by
using the results of the last section to verify (1.3). The property (A)
shows that, without any loss of generality, we may assume p > 1.
Let ε be a given number in the interval (0, μ - λ). Then in applying
Tsuji's Theorem A, we note

r<\am\<l r<|α m |<l

where this latter sum converges. Therefore, there is an integer q$ such

that each interval [1 — 2~q, \—2~q~λ) contains a number Rq for which
the circle {z: \z\ = Rq} does not intersect any of the exceptional discs
of Theorem A when q >qo. An application of Theorem A implies
(3.1) \log\B(z,{am},μ)\\<K(l-\z\Γλ-elog(l/(l-\z\))
(\z\ = Rq,q>q0).
By using the factorisation (2.2), we now have

|z|)) (|z| =Rq,q> q0).

q q x
Hence, for any r in [1 - 2~ , 1 - 2~ ~ ), the maximum modulus prin-
ciple implies

<K(l-r)-λ-εlog(ί/(l-r))
when q > q$. Since λ > 1, and g has no zeros in D(0,1), it follows
[2] that
(3.2) \log\g(z)\\<K(l -\z\)-λ-elog(l/(l -\z\))

The inequality (3.2) leads to

λ ε
Mp(r, g)<K(l- rΓ - log(l/(l - r)),
from which the Minkowski inequalities and Lemma 1 yield
Mp(r, f) < Mp(r, g) + Mp(r, B)
<K(l-r)-λ-ε\og(l/(l-r))
126 C. N. LINDEN

when r is sufficiently close to 1. We now have λp(f) < λ + ε for all

positive ε, so that

(3.3) λp(f)<λ = λoo(f).

The inequality (1.3) has been proved.

The proof of (1.4) when p > 1 is obtained by applying the method
of proof of Theorem 5.9 [1, p. 84]. The Poisson-Jensen formula,
together with Holder's inequality, yields

log\f(rew)\<^^π\log\f(Re^)\\P(Rfrfθ-φ)dφ

< Mp(r, f) I — J P(R, r,θ- φγlp-χ dφ

π

for 0 < r < R < 1, 0 < θ < 2π. We put R = ^(1 + r), and use a
standard estimate [1, p. 84] for the Poisson kernel to obtain

Moo(r,f)<KMp(r,f){l-rΓι">.

The inequality (1.4) follows for 1 < p < oo, and so does (1.2).
We have already noted that (1.1) implies (1.4) when p = 1, so it
remains to consider 0 < p < 1. The property (B) shows that

p(λs(f) - λp{f)) < q ( ~ ) (W) - W ) ) (0 < P < Q <

with limiting form

(3.5) p(λoo(f) - λp{f)) < qiλooif) - λq{f)) (0<p< q),

obtained from (1.2). But we have already seen that the right-hand
side of this latter inequality has upper bound 1 when q > 1. Hence
0 < p < 1 < q implies

(3.6) P{λoo{f)~λp{f))<\

for 0 < p < 1, and (1.4) follows for all positive p.

4. The proof of Corollary 1. Corollary 1 follows readily from the

proof of Theorem 1. The inequality (3.5) shows that p(λoo(f)—λp(f))
is increasing on (0, oo), and (1.3) and (1.4) show that the range of this
function is included in [0,1]. The inequalities (3.5) and (1.4) also
 INTEGRAL LOGARITHMIC MEANS 127

imply

> qλq{f) -{q- P)(λq(f) + l/q) = pλq{f) - 1 + p/q.

Corollary 1 (ii) follows immediately for 0 < p < oo, and for p — oc by
taking limits.

REFERENCES

[1] P. Duren, Theory ofHp Spaces, Academic Press, 1970.

[2] C. N. Linden, Functions regular in the unit circle, Proc. Camb. Phil. Soc, (1956),
49-60.
[3] , The representation of regular functions, J. London Math. Soc, 39 (1964),
19-30.
[4] J. E. Littlewood, Theory of Functions, Oxford University Press, 1944.
[5] L. R. Sons, Zero distribution of functions with slow or moderate growth in the
unit disc, Pacific J. Math., 99 (1982), 473-481.
[6] M. Tsuji, Potential Theory, Tokyo, 1959.

Received November 23, 1987.

UNIVERSITY COLLEGE OF SWANSEA

SINGLETON PARK
SWANSEA SA2 8PP WALES