Entire and Meromorphic Functions

Walter Bergweiler
CAU Kiel
Winter term 2017/18

Version of February 6, 2018

Contents
1 An overview 1

2 Review of complex function theory 3

3 Jensen’s formula 5

4 The Nevanlinna characteristic and the first fundamental theorem 12

5 Properties of the Nevanlinna characteristic 17

6 Maximum modulus and characteristic of entire functions 21

7 The order of a meromorphic function 32

8 Weierstraß products 34

9 The Hadamard factorization theorem 38

10 Preparations for the second fundamental theorem 48

11 The lemma on the logarithmic derivative 51

12 The second fundamental theorem and the deficiency relation 59

13 Uniqueness theorems 64

14 An application to iteration theory 69

15 Differential equations in the complex domain 73

16 Exceptional values of derivatives 81

References 86
1 AN OVERVIEW 1

1 An overview
A function holomorphic in the complex plane C is called entire. Such a function f
has a power series expansion
∞
X
f (z) = cn z n
n=0
which converges for all z ∈ C.
Examples are
∞
z
X 1 n
f (z) = e = z ,
n=0
n!
1 iz
e − e−iz ,


f (z) = sin z =
2i
or
1 iz
e + e−iz .


f (z) = cos z =
2
A function which is holomorphic in a domain D except for poles is called mero-
morphic in D. (The precise definition and more details will be given later.) In
particular if D = C we just say the function is meromorphic. Functions mero-
morphic in C are precisely those which can be written as a quotient of two entire
functions. (A proof of this will be given later.)
An example is
sin z
tan z = .
cos z
The following result can be seen as the starting point of the mathematical area
we will discuss in this lecture.
Picard’s theorem (1879). Let f be an entire function and let a1 , a2 ∈ C be
distinct. Suppose that f (z) 6= aj for all z ∈ C and all j ∈ {1, 2}. Then f is
constant.
Example. Let f (z) = ez . Then f (z) 6= 0 for all z ∈ C.
The example shows that we need two values a1 and a2 in Picard’s theorem.
Let Cb = C ∪{∞} be the Riemann sphere. A meromorphic function f is thus a
function f : C → C, b with f (z0 ) = ∞ meaning that z0 is a pole. An entire function
f can also be considered as a meromorphic function f : C → C b with f (z) 6= ∞ for
all z ∈ C. This leads to the following result.
Picard’s theorem for meromorphic functions. Let f : C → C b be meromorphic
and a1 , a2 , a3 ∈ C distinct. Suppose that f (z) 6= aj for all z ∈ C and all j ∈
b
{1, 2, 3}. Then f is constant.
The proof is by reduction to the case of entire functions; that is, the case
a3 = ∞. In fact if aj 6= ∞ for all j, then g = 1/(f − a3 ) is entire and
1 1
g(z) = 6= bj :=
f (z) − a3 aj − a3
for j ∈ {1, 2}, so that g and hence f is constant.
2 1 AN OVERVIEW

Example. The tangent function satisfies tan z 6= ±i for all z ∈ C.

Around 1900, Borel and others found quantitative versions of Picard’s theorem
for entire functions. Let n(r, a) be the number of solutions of f (z) = a in the disk
{z ∈ C : |z| ≤ r} and let
M (r, f ) := max |f (z)|
|z|=r

be the maximum modulus of f . By Liouville’s theorem, M (r, f ) → ∞ as r → ∞ if

f is non-constant.
Borel showed that n(r, a) and log M (r, f ) are of the same order of magnitude
for all a ∈ C, with at most one exception.

Borel’s theorem.
log n(r, a) log log M (r, f )
lim sup = lim sup
r→∞ log r r→∞ log r

for all a ∈ C, with at most one exception.

Example. Consider f (z) = ez . Then M (r, f ) = er so that log M (r, f ) = r. Let

a ∈ C \{0}. Writing a = |a|eiϕ we see that the solutions of ez = a are given by
z = zk = log |a| + i(ϕ + 2πk), where k ∈ Z. It follows that
p
r2 − (log |a|)2
n(r, a) = + O(1)
π
and hence
r
n(r, a) ∼
π
as r → ∞. In particular,

log n(r, a) log log M (r, f )

lim sup = lim sup =1
r→∞ log r r→∞ log r

for all a ∈ C \{0}.

In 1924, Nevanlinna developed a theory giving quantitative estimates also for
meromorphic functions, replacing log M (r, f ) by a “characteristic function” T (r, f ).
This gives stronger results also for entire functions. The theory yields far-reaching
generalizations of the theorems of Picard and Borel.
The lecture will give an introduction to Nevanlinna theory, together with some
further results about entire and meromorphic functions (e.g. the Weierstraß and
Hadamard factorization theorems).
We will also give some applications of the theory, e.g. to differential equations
or iteration. (No knowledge of these subjects is required.)
The lecture will assume familiarity with the basic results of complex function
theory (as given in the course “Analysis IV”). Some results of complex function
theory will be recalled.
2 REVIEW OF COMPLEX FUNCTION THEORY 3

2 Review of complex function theory

Let D ⊂ C be a domain (i.e. open and connected) and let f : D → C be a function.
A substantial part of a basic complex function course consists of showing that the
following four properties are equivalent:
(i) f is complex differentiable at every point of D; that is, for all z0 ∈ D the
limit
f (z) − f (z0 )
f 0 (z0 ) := lim
z→z0 z − z0
exists;

(ii) with u = Re f and v = Im f the function given by

(x, y) 7→ (u(x + iy), v(x + iy))

is differentiable in the real sense and satisfies the Cauchy-Riemann differential

equations
ux = vy , uy = −vy ;

(iii) f is continuous and for every null-homologous, closed, piecewise smooth curve
γ in D we have Z
f (z)dz = 0,
γ

here γ is called null-homologous if the winding number

Z
1 dζ
n(γ, z) =
2πi γ ζ − z

satisfies n(γ, z) = 0 for all z ∈ C \D;

(iv) f can be developed into a power series around any point in D; that is, for all
z0 ∈ D there exists r > 0 and a sequence (cn )n≥0 in C such that
∞
X
f (z) = cn (z − z0 )n
n=0

for |z − z0 | < r.
A function having one (and hence all) of these properties is called holomorphic.
In (iv) we can take any r > 0 such that

D(z0 , r) := {z ∈ C : |z − z0 | < r} ⊂ D.

The radius of convergence of the power series may be greater than r. The coeffi-
cients have the representations
Z
1 (n) 1 f (ζ)
cn = f (z0 ) = dζ
n! 2πi |ζ−z0 |=ρ (ζ − z0 )n+1
4 2 REVIEW OF COMPLEX FUNCTION THEORY

if 0 < ρ < r.
In (iii), instead of considering all integrals over arbitrary nullhomologous curves,
it
R suffices to consider integrals over the boundaries of triangles in D; that is, if
∂∆
f (z)dz = 0 for every closed triangle ∆ contained in D, then f is holomorphic.
This result is known as Morera’s Theorem.
If K ⊂ C, b not necessarily open, and if f : K → C, then we say that f is
meromorphic (or holomorphic) in K if there exists a domain D ⊃ K such that f
extends to a meromorphic (or holomorphic) function f : D → K.
For K ⊂ C we denote by K the closure of K. For a ∈ C and r > 0 we put

D(a, r) = D(a, r) = {z ∈ C : |z − a| ≤ r}.

Cauchy’s integral formula says that if f : D → C is holomorphic and γ a null-

homologous, closed, piecewise smooth curve in D, then
Z
1 f (ζ)
n(γ, z)f (z) = dζ.
2πi γ ζ − z

In particular, if f is holomorphic in D(z0 , r), then

Z
1 f (ζ)
f (z) = dζ
2πi |z−z0 |=r ζ − z

for z ∈ D(z0 , r). Differentiating this we obtain

Z
(k) k! f (ζ)
f (z) = dζ
2πi |z−z0 |=r (ζ − z)k+1

If D is a domain, z0 ∈ D and f : D \ {z0 } → C is holomorphic, then z0 is called

an isolated singularity of f . It is called
– removable, if limz→z0 f (z) ∈ C. By putting f (z0 ) = limz→z0 f (z) the function
f extends to a holomorphic function f : D → C;
– a pole, if limz→z0 |f (z0 )| = ∞;
– essential otherwise.
In the case of a pole we write f (z0 ) = ∞ and with Cb := C ∪{∞} thus consider f
as a function f : D → C.b
Given a function f , we usually assume that removable singularities have already
been “removed” and that at poles the value ∞ has been assigned.
Example. Consider
b f (z) = sin(πz) .
f : C → C,
sin(πz 2 )
√ √
The expression defining f has isolated singularities at the points ± k and ±i k,
for all k ∈ Z. We have √ 
f 2 = ∞,
3 JENSEN’S FORMULA 5

sin(πz) π cos(πz)
f (0) = lim 2
= lim =∞
z→z sin(πz ) z→0 2πz cos(πz 2 )

and
sin(πz) 1
f (1) = lim 2
= .
z→1 sin(πz ) 2
Rational functions (i.e., quotients of polynomials) are meromorphic in C (even
in C). Rational functions of degree 1 are called Möbius transformations (or frac-
b
tional linear transformations). They are bijective maps from C b to C.
b

3 Jensen’s formula
Theorem 3.1. Let r > 0 and let f be holomorphic in D(0, r). Then
Z 2π
1
f (0) = f (reiθ )dθ,
2π 0
Z 2π
1
Re f (0) = Re f (reiθ )dθ,
2π 0
and Z 2π
1
Im f (0) = Im f (reiθ )dθ.
2π 0
Proof. Let
γ : [0, 2π] → C, γ(θ) = reiθ ,
be the standard parametrization of ∂D(0, r). Cauchy’s integral formula yields
Z
1 f (z)
f (0) = dz
2πi γ z
Z 2π
1 f (γ(θ)) 0
= γ (θ)dθ
2πi 0 γ(θ)
Z 2π
1 f (reiθ ) iθ
= ire dθ
2πi 0 reiθ
Z 2π
1
= f (reiθ )dθ.
2π 0
This is the first claim. The second and third claims follow by taking real and
imaginary parts.
Remark. The result is also known as the mean value property.
Theorem 3.2 (Jensen’s formula). Let r > 0 and f meromorphic in D(0, r). Let
a1 , . . . , am be the zeros of f and b1 , . . . , bn the poles of f in D(0, r)\{0}, counted
with multiplicity. Let
X∞
f (z) = ck z k
k=`
6 3 JENSEN’S FORMULA

be the Laurent series expansion of f near 0, with c` 6= 0. Then

Z 2π m n
1 iθ
X r X r
log |c` | = log |f (re )|dθ − log + log − ` log r.
2π 0 j=1
|aj | k=1 |bk |

Remark. If f (0) 6= 0, ∞, then we have ` = 0 and c` = log |f (0)| in Theorem 3.2.

We first state the following lemma.

Lemma 3.1. Let r > 0, a ∈ D(0, r) and

r(z − a)
ϕ : C → C,
b ϕ(z) = .
r2 − az

Then ϕ is holomorphic in D(0, r) and |ϕ(z)| = 1 for z ∈ ∂D(0, r).

The proof is left as an exercise.

Proof of Theorem 3.2. We will start with a special case and then increase the gen-
erality step by step.

(1) Suppose first that f has neither zeros or poles in D(0, r); that is, f is holo-
morphic in D(0, r) and f (z) 6= 0 for z ∈ D(0, r). Then there exists a
function g holomorphic in D(0, r) such that f = exp ◦g. (The existence
of such a function g is standard, but we repeat the argument: First choose
R > r such that f : D(0, R) → C is holomorphic and without zeros. Then
there exists G : D(0, R) → C with G0 = f 0 /f . For example, one may put
R z 0 (ζ)
G(z) = 0 ff (ζ) dζ. We then choose c ∈ C such that g(z) = G(z) + c satisfies
e g(0)
= f (0). The function h defined by h(z) = f (z)/eg(z) = f (z)e−g(z) then
satisfies h(0) = 1 and h0 = f 0 e−g − f g 0 e−g = 0 and thus h(z) = 1 for all
z ∈ D(0, R). Thus f = eg .)
It follows that |f | = eRe g and thus log |f | = Re g. The conclusion now follows
from Theorem 3.1.

(2) We allow zeros and poles in D(0, r)\{0}, but assume that f has no other zeros
and poles in D(0, r); that is, f (0) 6= 0, ∞ and f (z) 6= 0, ∞ for z ∈ ∂D(0, r).
Let ϕ = ϕa be as in Lemma 3.1, that is,

r(z − a)
ϕa (z) = ,
r2 − az

and consider the function h given by

Qn
ϕbk (z)
h(z) = f (z) Qk=1
m .
j=1 ϕaj (z)
3 JENSEN’S FORMULA 7

Then h is holomorphic in D(0, r) and has no zeros in D(0, r). Moreover, we

have |h(z)| = |f (z)| for z ∈ ∂D(0, r) by Lemma 3.1. It follows, using the
result of part (1) of the proof, that
Z 2π Z 2π
1 iθ 1
log |f (re )|dθ = log |h(reiθ )|dθ
2π 0 2π 0
= log |h(0)|
Xn m
X
= log |f (0)| + log |ϕbk (0)| − log |ϕaj (0)|
k=1 j=1
n m
X |bk | X |aj |
= log |f (0)| + log − log ,
k=1
r j=1
r

from which the conclusion follows.

(3) Now we allow that f (0) = 0 or f (z) = ∞, but still assume that f (z) 6= 0, ∞
for z ∈ ∂D(0, r). With ` and c` as in the statement of the theorem we consider

f (z)
h(z) = .
z`
Thus h(0) = c` 6= 0, ∞. Since f (z) = h(z)z ` , part (2) of the proof yields that
Z 2π Z 2π
1 iθ 1
log |f (re )|dθ = log |h(reiθ )|dθ + ` log r
2π 0 2π 0
n m
X r X r
= log |h(0)| − log + log + ` log r,
k=1
|bk | j=1 |aj |
n m
X r X r
= log |c` | − log + log + ` log r
k=1
|bk | j=1 |aj |

as claimed.

(4) We finally consider the general case, i.e., zeros and poles are also allowed on
∂D(0, r). Then
Z 2π
1
log |f (reiθ )|dθ
2π 0
is an improper integral. The existence of this integral is easy to show.
To prove the formula claimed, there are two possibilities:

– reduce to part (3) by multiplying f with factors z − bk or dividing by

z − aj , similarly as in part (2);
– replace r by r − ε and take the limit as ε → 0.

We omit the details.

8 3 JENSEN’S FORMULA

We want to rewrite the sums occurring in Jensen’s formula.

Lemma 3.2. Let r > 0 and c1 , . . . , cm ∈ D(0, r) \ {0}. For 0 ≤ t ≤ r let n(t)
denote the number of cj in D(0, t). Then
m Z r
X r n(t)
log = dt.
j=1
|c j | 0 t

Proof. Without loss of generality we may assume that 0 < |c1 | ≤ |c2 | ≤ . . . ≤
|cm | ≤ r. Using Z r
r dt
log = log r − log |cj | =
|cj | |cj | t

we find that
m m Z r m Z |cm | Z r
X r X dt X dt dt
log = = +m
j=1
|cj | j=1 |cj | t j=1 |cj | t |cm | t
m m−1
XZ |ck+1 | Z r
X dt dt
= +m
j=1 k=j |ck | t |cm | t
m−1 k Z |ck+1 | Z r
XX dt dt
= +m
k=1 j=1 |ck |
t |cm | t

m−1
X Z |ck+1 | dt Z r
dt
= k +m
k=1 |ck | t |cm | t
m−1
X Z |ck+1 | n(t) Z r
dt
= dt +
|ck | t |cm | t
Zk=1r Z r
n(t) n(t)
= dt = dt.
|c1 | t 0 t
Remark. The above considerations may be elegantly described using the Riemann-
Stieltjes integral. We discuss these integrals in a short excursion:
For a function f : [a, b] → C, a partition of the interval [a, b] given by a = x0 <
x1 < · · · < xn = b and points ξj ∈ [xj−1 , xj ], the corresponding Riemann sum is
given by
n
X
S(f, (xj ), (ξj )) = f (ξj )(xj − xj−1 ).
j=1

The Riemann integral is obtained by a limit process:

Z b
S(f, (xj ), (ξj )) → f (x)dx as max |xj − xj−1 | → 0.
a j

Let, in addition, a function g : [a, b] → C be given. The Riemann-Stieltjes sum is

defined by
n
X
S(f, g, (xj ), (ξj )) = f (ξj ) (g(xj ) − g(xj−1 )) .
j=1
3 JENSEN’S FORMULA 9

If these sums tend to a limit as maxj |xj − xj−1 | → 0, the limit is called Riemann-
Stieltjes integral and denoted by
Z b
f (x)dg(x).
a

It exists for example if f is continuous and g is of bounded variation, i.e.,

X
L(g) := sup |g(xj ) − g(xj−1 )| < ∞.
(xj ) j

This holds in particular if g is monotone.

If g ∈ C 1 [a, b] and f ∈ C[a, b], then
Z b Z b
f (x)dg(x) = f (x)g 0 (x)dx.
a a

But Riemann-Stieltjes integrals are of interest also if g is discontinuous. For exam-

ple, if f : [a, b] → C is continuous, x0 ∈ (a, b), g(x) = α for x < x0 and g(x) = β
for x ≥ x0 , then
Z b
f (x)dg(x) = f (x0 )(β − α).
a

The rule of integration by parts takes the following form:

Z b x=b Z b
f (x)dg(x) = f (x)g(x) − g(x)df (x).
a x=a a

For g(t) = n(t) and f (t) = log(r/t) = log r − log t this is precisely the formula from
Lemma 3.2.
Rr
Integrals of the form 0 a(t)
t
dt will occur repeatedly. ItR may be helpful to write
r
them (or at least think of them) as integrals of the form 0 a(t)d log t.

Definition 3.1. Let r > 0 and f meromorphic in D(0, r). For 0 ≤ t ≤ r let n(t, f )
denote the number of poles of f in D(0, t), counted according to multiplicity. Then
Z r
n(t, f ) − n(0, f )
N (r, f ) = dt + n(0, f ) log r
0 t

is called the (Nevanlinna) counting function of the poles of f .

Remark. (1) The zeros of f are poles of 1/f . For a ∈ C the zeros of f − a are
called a-points of f . They are the poles of 1/(f − a). Thus n(r, 1/(f − a))
and N (r, 1/(f − a)) count the a-points of f .
Sometimes, if the function f considered is clear from the context, we write
n(r, a) and N (r, a) instead of n(r, 1/(f − a)) and N (r, 1/(f − a)). Then we
also write n(r, ∞) and N (r, ∞) instead of n(r, f ) and N (r, f ).
10 3 JENSEN’S FORMULA

(2) If f (0) 6= ∞, then n(0, f ) = 0 and

Z r
n(t, f )
N (r, f ) = dt.
0 t

For 0 < r0 < r we always have

Z r0 Z r
n(t, f ) − n(0, f ) n(t, f ) − n(0, f )
N (r, f ) = dt + dt + n(0, ∞) log r
0 t r0 t
Z r
n(t)
= N (r0 , f ) + dt.
r0 t

Using Definition 3.1 Jensen’s formula takes the following form.

Theorem 3.3. Let r > 0 and f meromorphic in D(0, r). Let c` be the first non-
zero coefficient in the Laurent series of f as in Jensen’s formula (Theorem 3.2).
Then Z 2π  
1 iθ 1
log |f (re )|dθ = N r, − N (r, f ) + log |c` |.
2π 0 f

Proof. If f (0) 6= 0, ∞ so that c` = f (0) (and ` = 0) this follows directly from

Theorem 3.2 and Lemma 3.2.
Suppose that f (0) = ∞. Then ` < 0 and 0 is a pole of f of order −`. Thus
n(0, f ) = −`. Moreover, if b1 , . . . , bn are the poles of f in D(0, r) \ {0}, then
n r
n(t, f ) − n(0, f )
Z
X r
log = dt
k=1
|b k| 0 t

by Lemma 3.2. Thus

r
n(t, f ) − n(0, f )
Z
N (r, f ) = dt + n(0, f ) log r
0 t
n
X r
= log − ` log r
k=1
|bk |

and the conclusion follows with Theorem 3.2. The case that f (0) = 0 is analogous.
Here n(0, 1/f ) = `.

Definition 3.2. Let f be entire and r > 0. Then

M (r, f ) := max |f (z)|

|z|=r

is called the maximum modulus of f .

Remark. The maximum principle yields that M (r, f ) strictly increases with r if f
is non-constant.
3 JENSEN’S FORMULA 11

Theorem 3.4. Let f be entire, r > 0 and f (0) 6= 0. Then

 
1
N r, ≤ log M (r, f ) − log |f (0)|.
f
Proof. This follows directly from Theorem 3.3, since N (r, f ) = 0 and c` = f (0).
Theorems like Theorem 3.3 or Theorem 3.4 are the reason that we passed from
the “natural” function n(r, f ) counting the poles to the function N (r, f ). It is,
however, also possible to obtain estimates for n(r, f ) from this.
Lemma 3.3. Let 1 ≤ r ≤ R and f meromorphic in D(0, R). Then
N (r, f ) ≤ n(r, f ) log r + N (1, f )
and  
R
N (R, f ) ≥ log n(r, f ).
r
Proof. The first claim follows with
Z r
n(t, f )
N (r, f ) = dt + N (1, f )
1 t
Z r
dt
≤ n(r, f ) + N (1, f )
1 t
= n(r, f ) log r + N (1, f ),
the second one with
Z R
n(t, f )
N (R, f ) = dt + N (1, f )
1 t
Z R
n(t, f )
≥ dt
1 t
Z R
n(t, f )
≥ dt
r t
Z R
dt
≥ n(r, f )
r t
R
= n(r, f ) log .
r
Theorem 3.5. Let f be entire with f (0) 6= 0. Let r > 0 and K > 1. Then
 
1 1
n r, ≤ (log M (Kr, f ) − log |f (0)|) .
f log K
Proof. With R = Kr in Lemma 3.3 we deduce from Theorem 3.4 that
   
1 1 1
n r, ≤ N Kr,
f log K f
1
≤ (log M (Kr, f ) − log |f (0)|) .
log K
12 4 THE FIRST FUNDAMENTAL THEOREM

Remark. Suppose that log M (r, f ) = O(rρ ) as r → ∞ for some ρ > 0, say

log M (r, f ) ≤ τ rρ for r ≥ r0 .

Then
τ Kρ ρ
   
1 1 1
n r, ≤ N Kr, + O(1) ≤ r + O(1).
f log K f log K
First this follows only if f (0) 6= 0, but considering h(z) = f (z)/z p instead of f if
0 is a zero of f of multiplicity p and noting that n(r, 1/f ) = n(r, h) + p this also
holds if f (0) = 0.
It follows that  
1
n r, = O(rρ ) as r → ∞.
f
With K = e1/ρ we actually have
 
1
n r, ≤ eρτ rρ + O(1)
f
as r → ∞.
More generally,
 
1
n r, = O(rρ ) as r → ∞ for all a ∈ C .
f −a

So in some sense n(r, 1/(f − a)) does not grow faster than log M (r, f ). We will
see that for all a ∈ C, with at most one exception, n(r, 1/(f − a)) will grow of the
same order of magnitude as log M (r, f ).

4 The Nevanlinna characteristic and the first fun-

damental theorem
We define log+ : R → R by
(
log x, x ≥ 1,
log+ x =
0, x ≤ 1.

Definition 4.1. Let r > 0 and f meromorphic in D(0, r). Then

Z 2π
1
m(r, f ) = log+ |f (reiθ )|dθ
2π 0
is called the proximity function and

T (r, f ) = N (r, f ) + m(r, f )

is called the (Nevanlinna) characteristic of f .

4 THE FIRST FUNDAMENTAL THEOREM 13

Distinguishing the cases x ≥ 1 and x < 1 we see that

1
log x = log+ x − log+ .
x
This yields that
Z 2π  
1 iθ 1
log |f (re )|dθ = m(r, f ) − m r, .
2π 0 f
Inserting this into the formula of Theorem 3.3 yields
   
1 1
m(r, f ) − m r, = N r, − N (r, f ) + log |c` |
f f
and thus  
1
T (r, f ) = T r, + log |c` |.
f
We will also consider T (r, 1/(f − a)) for a ∈ C. Before doing so, we collect some
properties of the function log+ .
Lemma 4.1. Let x1 , . . . , xn ≥ 0 and x ≥ 0. Then
1
(i) |log x| = log+ x + log+ ;
x
n
! n
Y X
+
(ii) log xj ≤ log+ xj ;
j=1 j=1

n
! n
X X
+
(iii) log xj ≤ log+ xj + log n;
j=1 j=1

(iv) log+ x1 − log+ x2 ≤ log+ |x1 − x2 | + log 2.

by distinguishing the cases x ≥ 1 and x < 1.
Proof. Conclusion (i) follows Q
Conclusion (ii) is clear if nj=1 xj ≤ 1 and follows otherwise since then
n
! n
! n n
Y Y X X
+
log xj = log xj = log xj ≤ log+ xj .
j=1 j=1 j=1 j=1

Conclusion (iii) follows since

n
!
X
log+ xj ≤ log+ (n · max xj )
j
j=1

≤ log+ n + log+ (max xj )

j
n
!
X
≤ log n + log+ xj .
j=1
14 4 THE FIRST FUNDAMENTAL THEOREM

To prove (iv), we assume without loss of generality that x1 ≥ x2 . Then, by

(iii),
log+ x1 − log+ x2 = log+ x1 − log+ x2
= log+ (x1 − x2 + x2 ) − log+ x2
≤ log+ (x1 − x2 ) + log+ x2 + log 2 − log+ x2
= log+ |x1 − x2 | + log 2.

Theorem 4.1 (First Fundamental Theorem). Let r > 0, f meromorphic in D(0, r)

and a ∈ C. Let c(a) be the first non-zero coefficient in the Laurent series of f − a
around 0. Then
 
1
T r, = T (r, f ) − log |c(a)| + ϕ(r, a)
f −a
where |ϕ(r, a)| ≤ log+ |a| + log 2.
Proof. For a = 0 this is the formula stated before Lemma 4.1.
For a 6= 0 this formula yields that
 
1
T (r, f − a) = T r, + log |c(a)|.
f −a
Hence  
1
T r, = T (r, f − a) − log |c(a)|
f −a
= N (r, f − a) + m(r, f − a) − log |c(a)|
= N (r, f ) + m(r, f ) + ϕ(r, a) − log |c(a)|
= T (r, f ) + ϕ(r, a) − log |c(a)|
with
ϕ(r, a) = m(r, f − a) − m(r, f ).
By Lemma 4.1, (iv), we have
log+ |f (reiθ ) − a| − log+ |f (reiθ )| ≤ log+ |a| + log 2
and thus
ϕ(r, a) ≤ log+ |a| + log 2.
Remark. The main point is that the difference between T (r, f ) and T (r, 1/(f − a))
is bounded by a constant independent of r.
We will see soon that for f meromorphic in C and non-constant we have
T (r, f ) → ∞ as r → ∞. The first fundamental theorem says that
     
1 1 1
N r, + m r, = T r, = T (r, f ) + O(1)
f −a f −a f −a
as r → ∞.
The interpretation is as follows:
4 THE FIRST FUNDAMENTAL THEOREM 15

 
1
– N r, is large if f has many a-points;
f −a
 
1
– m r, is large if f is close to a on some part of ∂D(0, r).
f −a

If T (r, f ) is large we must have one of these two possibilities.

The second fundamental theorem will say that for “most” values of a the first
alternative holds; that is, the term m(r, 1/(f − a)) is small and thus N (r, 1/(f − a))
is large.
Example 4.1. Let f (z) = ez . Since f is entire, N (r, f ) = 0. Moreover,
Z 2π
1 iθ
m(r, f ) = log+ ere dθ
2π 0
Z 2π
1
log+ er cos θ dθ


=
2π 0
Z 2π
1
= max{0, r cos θ}dθ
2π 0
Z π/2
1
= r cos θ dθ
2π −π/2
r
= .
π
Thus
r
T (r, f ) = m(r, f ) = .
π
Analogously,    
1 1 r
N r, =0 and m r, = .
f f π
We saw in the introduction that if a ∈ C \{0}, then
  p
1 r2 − (log |a|)2
n r, − ≤K
f −a π

for some constant K if r ≥ | log |a||.

Since
√ r2 − (r2 − c2 ) c2 c2
0≤r− r2 − c2 = √ = √ ≤ ≤c
r + r 2 − c2 r + r 2 − c2 r
p
for c ≥ 0 we have r2 − (log |a|)2 /π = r/π + O(1) and thus
 
1 r
n r, = + O(1).
f −a π
16 4 THE FIRST FUNDAMENTAL THEOREM

It follows that
    Z r  
1 1 1 dt
N r, = N 1, + n t,
f −a f −a 1 f −a t
r
= + O(log r)
π
= T (r, f ) + O(log r).
The first fundamental theorem implies that
 
1
m r, = O(log r).
f −a
One can show that in fact m(r, 1/(f − a)) = O(1).
Example 4.2. Let f be a rational function, say f (z) = p(z)/q(z) with polynomials
p and q without common zeros. Let
p(z) = a` z ` + . . . + a0 ,
q(z) = bm z m + . . . + b0 ,
with a` 6= 0 and bm 6= 0. By the fundamental theorem of algebra, q has m zeros.
Thus f has m poles, so there exists r0 > 0 with n(r, f ) = m for r ≥ r0 . Hence
Z r
dt
N (r, f ) = N (r0 , f ) + m = m log r + C
r0 t
with a constant C.
We also have
|p(z)| ∼ |a` | · |z|` and |q(z)| ∼ |bm | · |z|m
and hence
|a` | `−m
|f (z)| ∼ |z|
|bm |
as |z| → ∞.
If ` ≤ m, then |f (z)| = O(1) as |z| → ∞ and hence m(r, f ) = O(1) as r → ∞.
Hence T (r, f ) = N (r, f ) + O(1) = m log r + O(1).
If ` > m, then, since
|a` |
log |f (z)| = (` − m) log |z| + log + o(1)
|bm |
as |z| → ∞, we have
m(r, f ) = (` − m) log r + O(1)
as r → ∞. Hence
T (r, f ) = N (r, f ) + m(r, f )
= m log r + (` − m) log r + O(1)
= ` log r + O(1).
5 PROPERTIES OF THE CHARACTERISTIC 17

In both cases we thus have

T (r, f ) = max{`, m} · log r + O(1) = deg(f ) · log r + O(1).

Had we only aimed at the last equation, we could have reduced the case ` > m to
the case ` ≤ m by passing from f to 1/f and using the first fundamental theorem.

5 Properties of the Nevanlinna characteristic

Theorem 5.1. Let f1 , . . . , fn be meromorphic and r ≥ 1. Then
n
! n
Y X
(i) T r, fj ≤ T (r, fj ),
j=1 j=1

n
! n
X X
(ii) T r, fj ≤ T (r, fj ) + log n.
j=1 j=1

Proof. The corresponding results for m(r, ·) follow directly from Lemma 4.1.
To prove the corresponding estimates for N (r, ·), we note that nj=1 fj can have
Q
a pole only where one of the function fj has a pole. One obtains
n
! n
Y X
n 0, fj ≤ n(0, fj )
j=1 j=1

and ! !
n
Y n
Y n
X
n t, fj − n 0, fj ≤ (n(t, fj ) − n(0, fj )) .
j=1 j=1 j=1

Integration yields !
n
Y n
X
N r, fj ≤ N (r, fj ).
j=1 j=1

Analogously we obtain
n
! n
X X
N r, fj ≤ N (r, fj ).
j=1 j=1

Together with the results for m(r, ·) this yields the conclusion.
Theorem 5.2. Let f be meromorphic and let M be a Möbius transformation (i.e.,
a rational function of degree 1). Then

T (r, M ◦ f ) = T (r, f ) + O(1).

Proof. Every Möbius transformation can be written as a composition of Möbius

transformation of the following three types:
18 5 PROPERTIES OF THE CHARACTERISTIC

(i) z 7→ 1/z (inversion)

(ii) z 7→ z + c, c ∈ C (translation)

(iii) z 7→ c · z, c ∈ C \{0} (dilation/rotation)

αz + β
For example, if M (z) = with γ 6= 0 (and αδ − βγ 6= 0 since M is non-
γz + δ
constant), then
αδ − βγ 1 α
M (z) = − · + ,
γ γz + δ γ
so M is composition of z 7→ γz, z 7→ z + δ, z 7→ −(αδ − βγ)/γz and z 7→ z + α/γ.
It then suffices to prove the conclusion for Möbius transformations of types (i),
(ii) and (iii). For type (i) this follows from the first fundamental theorem, and
cases (ii) and (iii) are easy.

Theorem 5.3. Let f be meromorphic and P a polynomial of degree d ≥ 1. Then

T (r, P ◦ f ) = d · T (r, f ) + O(1).

Proof. We have n(r, P ◦ f ) = d · n(r, f ) and thus

N (r, P ◦ f ) = d · N (r, f ).

Let P (z) = ad z d + . . . + a0 . Then there exists r0 > 1 with

1 3
1 < |ad | · |z|d ≤ |P (z)| ≤ |ad | · |z|d for |z| ≥ r0 .
2 2
Thus
|ad | 3 log d
log + d · log+ |z| ≤ log+ |P (z)| ≤ d · log+ |z| + log+
2 2
for |z| ≥ r0 . This implies that there exists a constant C such that

log+ |P (z)| − d · log+ |z| ≤ C

for all z ∈ C. Hence

|m(r, P ◦ f ) − d · m(r, f )| ≤ C
and thus
|T (r, P ◦ f ) − d · T (r, f )| ≤ C.

Theorem 5.4 (Cartan’s formula). Let f be meromorphic, f (0) 6= ∞ and r > 0.

Then Z 2π  
1 1
T (r, f ) = N r, dϕ + log+ |f (0)|.
2π 0 f − eiϕ
For the proof we will use the following lemma.
5 PROPERTIES OF THE CHARACTERISTIC 19

Lemma 5.1. Let a ∈ C. Then

Z 2π
+ 1
log |a| = log |a − eiϕ |dϕ.
2π 0

Proof. We apply Jensen’s formula for f (z) = a − z and r = 1. Distinguishing the

cases |a| > 1 and |a| ≤ 1 we obtain the conclusion.
Proof of Theorem 5.4. Theorem 3.3 yields that if f (0) 6= eiϕ and r > 0, then
Z 2π  
1 iθ iϕ 1
log f (re ) − e dθ = N r, − N (r, f ) + log |f (0) − eiϕ |.
2π 0 f − eiϕ

Integrating with respect to ϕ we obtain, using Lemma 5.1,

Z 2π Z 2π
1 1
log f (reiθ ) − eiϕ dθdϕ
2π 0 2π 0
Z 2π  
1 1
= N r, iϕ
dϕ − N (r, f ) + log+ |f (0)|.
2π 0 f −e

The order of integration on the left hand side can be interchanged, by Fubini’s
theorem. Here the measurability of

(ϕ, θ) 7→ log |f (reiθ ) − eiϕ |

is clear; one has to show the existence of

Z 2π Z 2π
log f (reiθ ) − eiϕ dθdϕ.
0 0

We omit this here.

Interchanging the order of integration and using Lemma 5.1 again we obtain
Z 2π Z 2π  
1 + 1 1
iθ
log f (re ) dθ = N r, dϕ − N (r, f ) + log+ |f (0)|,
2π 0 2π 0 f − eiϕ

and hence the conclusion.

Remark. (1) If f (0) = ∞, then the above proof gives
Z 2π  
1 1
T (r, f ) = N r, dϕ + log |c` |
2π 0 f − eiϕ

with the first non-zero Laurent coefficient c` .

(2) For ϕ ∈ [0, 2π] with f (0) 6= eiϕ we have

  Z r  
1 1 dt
N r, iϕ
= n t, iϕ
.
f −e 0 f −e t
20 5 PROPERTIES OF THE CHARACTERISTIC

Interchanging the order of integration as before we obtain with

Z 2π  
1 1
s(t, f ) = n t, dϕ
2π 0 f − eiϕ

that Z r
s(t, f )
T (r, f ) = dt + log+ |f (0)|.
0 t

We recall that for an interval I a function Φ : I → R is called convex if for all

x, x1 , x2 ∈ I with x1 ≤ x ≤ x2 we have

x − x1 x2 − x
Φ(x) ≤ Φ(x2 ) + Φ(x1 ).
x2 − x 1 x2 − x 1

Equivalently,
Φ(x) − Φ(x1 ) Φ(x2 ) − Φ(x)
≤ .
x − x1 x2 − x
A differentiable function Φ is convex if and only if Φ0 is non-decreasing. In fact,
if Ψ : I → R is non-decreasing and x0 ∈ I, then Φ : I → R,
Z x
Φ(x) = Ψ(t)dt
x0

is convex. (For continuous Ψ we of course have Φ0 = Ψ.)

For an interval I ⊂ (0, ∞) and Φ : I → R we say that Φ(r) is convex in log r
if the function Φ ◦ exp : log(I) → R is convex. This means that Φ is convex on a
logarithmic scale.
Analogously to the above, if Ψ : I → R is non-decreasing, then Φ : I → R,
Z r
Ψ(t)
Φ(r) = dt
r0 t

is convex in log r, for any fixed r0 ∈ I. In fact, with x0 = log r0 we have

Z ex Z x
x Ψ(t)
Φ(e ) = dt = Ψ(eu )du.
r0 t x0

We conclude from the above:

Theorem 5.5. Let f be meromorphic. Then N (r, f ) and T (r, f ) are non-de-
creasing and convex in log r.

Remark. In general, m(r, f ) is neither convex in log r nor increasing.

6 MAXIMUM MODULUS AND CHARACTERISTIC 21

6 Maximum modulus and characteristic of entire

functions
Our starting point for Jensen’s formula (Theorem 3.1) was the mean value property
Z 2π Z
1 iθ 1 dζ
f (0) = f (re )dθ = f (ζ)
2π 0 2πi |z|=r ζ

for a function f holomorphic in D(0, r), and the corresponding formula with f
replaced by Re f .
By Cauchy’s integral formula we have
Z
1 f (ζ)
f (z) = dζ,
2πi |z|=r ζ − z

but taking real parts does not give a formula for Re f (z) in terms of Re f (reiθ ).
Instead we use the following formula.

Theorem 6.1. Let f be holomorphic in D(0, r) and z ∈ D(0, r). Then

Z 2π
1
f (z) = f (reiθ )K(z, r, θ)dθ
2π 0
where
r2 − |z|2
K(z, r, θ) = .
|reiθ − z|2
Remark. The expression K(z, r, θ) is called Poisson kernel, the formula is called
Poisson integral formula.
Usually the Poisson integral formula is stated for harmonic functions. We recall
the definition. For a domain D a function u : D → R ( or u : D → C) is called
harmonic if it is twice continuously differentiable and

∂ 2u ∂ 2u
∆u := + = 0.
∂x2 ∂y 2
If f is holomorphic, then Re f is harmonic. This is an immediate consequence of the
Cauchy-Riemann equations. In the opposite direction, if u : D → R is harmonic and
D is simply connected, then there exists a holomorphic function f with u = Re f .
This fails if the domain is not simply connected, an example being u : C \{0} → R,
u(z) = log |z|. But in any simply connected subdomain D of C \{0} we may define
a branch arg : D → R of the argument, and log z = log |z| + i arg z defines a holo-
morphic function in D. Thus the theory of harmonic functions is closely connected
to that of holomorphic functions. We will not consider harmonic functions in detail
and thus have stated Theorem 6.1 for holomorphic functions. The above consider-
ations show, however, that it remains valid for harmonic functions, and in fact the
statement for harmonic functions is equivalent to that for holomorphic functions.
22 6 MAXIMUM MODULUS AND CHARACTERISTIC

Proof of Theorem 6.1. The function

r2 (ζ + z)
h : D(0, r) → D(0, r), h(ζ) =
zζ + r2

is bijective, with inverse function

ζ −z
k(ζ) := h−1 (ζ) = −r2 ,
zζ − r2

and we have h(∂D(0, r)) = ∂D(0, r); see Lemma 3.1 where h(ζ)/r is considered.
We apply the mean value property (Theorem 3.1) to g = f ◦ h and obtain

f (z) = f (h(0)) = g(0)

Z
1 g(w)
= dw
2πi |w|=r w
Z
1 f (h(w))
= dw
2πi |w|=r w
Z
1 f (ζ) 0
= k (ζ)dζ
2πi |ζ|=r k(ζ)
Z  
1 1 z
= f (ζ) − dζ
2πi |ζ|=r ζ − z zζ − r2

Now, for |ζ| = r,

1 z 1 z
− = −
ζ − z zζ − r2 ζ − z (z − ζ)ζ
(z − ζ)ζ − z(ζ − z)
=
(ζ − z)(z − ζ)ζ
ζζ − zz
=
(ζ − z)(z − ζ)ζ
r2 − |z|2 1
= ·
|ζ − z|2 ζ
and thus
2π
r2 − |z|2 dζ r2 − |z|2
Z Z
1 1
f (z) = f (ζ) 2
= f (reiθ ) dθ.
2πi |ζ|=r |ζ − z| ζ 2π 0 |reiθ − z|2

Corollary. Let f be holomorphic in D(0, r) and z ∈ D(0, r). Then

Z 2π
1
Re f (z) = Re f (reiθ )K(z, r, θ)dθ.
2π 0

Jensen’s formula was obtained from the mean value property. Applying similar
arguments to the Poisson integral formula, we obtain the following result.
6 MAXIMUM MODULUS AND CHARACTERISTIC 23

Theorem 6.2 (Poisson-Jensen-Nevanlinna formula). Let f be meromorphic in

D(0, r), with zeros a1 , . . . , am and poles b1 , . . . , bn in D(0, r). Let z ∈ D(0, r)
with f (z) 6= 0, ∞. Then
Z 2π
1
log f reiθ


log |f (z)| = K(z, r, θ)dθ
2π 0
n m
X r 2 − bk z X r 2 − aj z
+ log − log .
k=1
r(z − b k) j=1
r(z − a j )

Remark. Jensen’s formula is the special case z = 0.

Proof. We restrict, as in the proof of Jensen’s formula (Theorem 3.2) to the case
that f has no zeros and poles on ∂D(0, r), and we proceed as there.
With
r(z − a)
ϕa (z) = 2
r − az
we consider Qn
ϕbk (z)
h(z) = f (z) Qk=1
m .
j=1 ϕaj (z)

Then h is holomorphic in D(0, r) and has no zeros there. Thus log |h| is the real part
of a function holomorphic in D(0, r) and we have |h(ζ)| = |f (ζ)| for ζ ∈ D(0, r).
Hence
Z 2π
1
log f (reiθ ) K(z, r, θ)dθ
2π 0
Z 2π
1
= log h(reiθ ) K(z, r, θ)dθ
2π 0
= log |h(z)|
Xn Xm
= log |f (z)| + log |ϕbk (z)| − log |ϕaj (z)|
k=1 j=1
n 2 m
X r − bk z X r 2 − aj z
= log |f (z)| − log + log
k=1
r(z − bk ) j=1
r(z − aj )

as claimed.

Theorem 6.3. Let f be entire and 0 < r < R. Then

R+r
T (r, f ) ≤ log+ M (r, f ) ≤ T (R, f ).
R−r
Proof. The first inequality is obvious since
Z 2π
1
log+ f reiθ dθ.


T (r, f ) = m(r, f ) =
2π 0
24 6 MAXIMUM MODULUS AND CHARACTERISTIC

To prove the second one, let z ∈ C with |z| = r. By the Poisson-Jensen-Nevanlinna

formula,
Z 2π m
1 iθ

X R 2 − aj z
log |f (z)| = log f Re K(z, R, θ)dθ − log .
2π 0 j=1
R(z − aj )

Using the functions ϕaj from the previous proof, that is,

R(z − aj )
ϕaj (z) = ,
R 2 − aj z
we have ϕaj (D(0, R)) = D(0, 1) and thus

R 2 − aj z
− log = log |ϕaj (z)| ≤ 0.
R(z − aj )
We also have
R2 − |z|2 R2 − |z|2 R + |z| R+r
0 ≤ K(z, R, θ) = iθ 2
≤ 2
= = .
|Re − z| (R − |z|) R − |z| R−r
It follows that
Z 2π
1
log f Reiθ K(z, R, θ)dθ


log |f (z)| ≤
2π 0
Z 2π
1
log+ f Reiθ K(z, R, θ)dθ


≤
2π 0
Z 2π
1
R+r
≤ log+ f Reiθ dθ
2π 0 R−r
R+r
= T (R, f ).
R−r
The following result is a generalization of Liouville’s theorem that a bounded
entire function is constant.
Theorem 6.4. Let f be an entire function and suppose that
log M (r, f )
L := lim inf < ∞.
r→∞ log r
Then f is a polynomial and deg(f ) ≤ L.
Proof. Let
∞
X
f (z) = cn z n
n=1

be the Taylor series expansion of f . Then

Z
1 f (z)
cn = dz
2πi |z|=r z n+1
6 MAXIMUM MODULUS AND CHARACTERISTIC 25

and thus
1 M (r, f ) M (r, f )
|cn | ≤ · 2πr n+1 =
2π r rn
for all r > 0 and all n ∈ N. (The last inequality is known as Cauchy’s inequality.)
Given ε > 0 there exist arbitrarily large r with
log M (r, f )
≤L+ε
log r
and thus
M (r, f ) ≤ rL+ε .
It follows that
|cn | ≤ rL+ε−n
for arbitrarily large r and thus that cn = 0 for n > L+ε. The conclusion follows.
The following result is an analogue of Theorem 6.4 for meromorphic functions.
Theorem 6.5. Let f be meromorphic and suppose that
T (r, f )
L := lim inf < ∞.
r→∞ log r
Then f is a rational function (and deg(f ) = L).
Proof. Let C > 1. By Lemma 3.3, applied with R = rC , we have
1
n(r, f ) ≤ C
· N (rC , f )
log(r /r)
C N (rC , f )
= ·
C −1 log rC
C T (rC , f )
≤ ·
C − 1 log rC
C
≤ (L + ε)
C −1
for arbitrarily large r, for any given ε > 0. Thus f has at most L poles. Hence
there exists a polynomial P (of degree at most L) such that g := P f is entire.
Since T (r, P ) ∼ deg(P ) log r as r → ∞ (by Example 2 after Theorem 4.1) we find,
using Theorem 5.1, that

T (r, g) ≤ T (r, P ) + T (r, f ) ≤ (2L + 1) log r

for arbitrarily large r. Applying Theorem 6.3 with R = 2r we thus find that

log M (r, g) ≤ 3T (2r, g) ≤ 3(2L + 1) log(2r) ≤ 3(2L + 2) log r

for arbitrarily large r. Theorem 6.4 yields that g is a polynomial. Hence f is

rational. Using that T (r, f ) ∼ deg(f ) log r as r → ∞ we find that deg(f ) = L.
26 6 MAXIMUM MODULUS AND CHARACTERISTIC

A meromorphic function which is not rational is called transcendental. Theo-

rem 6.5 says that
T (r, f )
lim =∞
r→∞ log r

for a transcendental meromorphic function f .

Applying Theorem 6.3 with R = Kr where K > 1 we obtain
K +1
log M (r, f ) ≤ T (Kr, f ).
K −1
(For K = 2 we used this already in the above proof.) If

T (r, f ) ≤ τ rρ

for large r, with τ, ρ > 0, then

K +1 ρ ρ
log M (r, f ) ≤ K τr .
K −1
Choosing K = 1 + 1/ρ we thus have
   ρ  
1 1 ρ 1
log M (r, f ) ≤ 2 + ρ 1+ τr ≤ 2 + ρeτ rρ .
ρ ρ ρ

So up to the constant (2 + 1/ρ) ρe the maximum modulus does not grow faster.
The constant (2 + 1/ρ) ρe is not optimal. For ρ > 12 the sharp constant is πρ.
Theorem 6.3 compares log M (r, f ) with T (R, f ) for some R > r. The following
lemma will allow to compare log M (r, f ) with T (r, f ), but not for all values of r.

Lemma 6.1 (Borel). Let x0 , y0 > 0, δ > 0, u : [x0 , ∞) → [y0 , ∞) continuous and
non-decreasing and ϕ : [y0 , ∞) → (0, ∞) continuous and non-increasing. Suppose
that Z ∞
ϕ(x)dx < ∞.
y0

Put
E = {x ∈ [x0 , ∞) : u(x + ϕ(u(x))) ≥ u(x) + δ} .
Then there exist sequences (xk ) and (x0k ) in [x0 , ∞), with xk < x0k ≤ xk+1 for all
k ∈ N, such that
[∞
E⊂ [xk , x0k ]
k=1

and ∞
X
(x0k − xk ) < ∞.
k=1

Remark. The last two conditions imply that E has finite measure. The lemma says
that if x 6∈ E, then u(x + ϕ(u(x))) < u(x) + δ.
6 MAXIMUM MODULUS AND CHARACTERISTIC 27

Proof. We may assume that E is unbounded, since otherwise the result is obvious.
Put x1 = min E. Note that this minimum exists, since u and ϕ are continuous
and hence E is closed. Put x01 = x1 + ϕ(u(x1 )).
We define (xk ) and (x0k ) recursively by xk = min E ∩ [x0k−1 , ∞ ) and x0k =

xk + ϕ(u(xk )). By construction, xk < x0k ≤ xk+1 and

u(xk+1 ) ≥ u(x0k ) = u(xk + ϕ(u(xk ))) ≥ u(xk ) + δ
for all k ∈ N. We deduce that
u(xk+1 ) ≥ u(x1 ) + kδ
and thus u(xk ) → ∞. Hence xk → ∞.
By construction we thus have
∞
[
E⊂ [xk , x0k ] .
k=1

Finally, if k ≥ 2, then
x0k − xk = ϕ(u(xk ))
≤ ϕ(u(x1 ) + (k − 1)δ)
≤ ϕ(y0 + (k − 1)δ)
Z y0 +(k−1)δ
1
= ϕ(y0 + (k − 1)δ) · dx
δ y0 +(k−2)δ
Z y0 +(k−1)δ
1
≤ ϕ(x)dx
δ y0 +(k−2)δ
and thus ∞ Z ∞
X 1
(x0k − xk ) ≤ ϕ(x)dx < ∞.
k=2
δ y0

Remark. The result is often applied to the function v = eu . With K = eδ > 1 we

obtain
v(x + ϕ(log v(x))) < Kv(x) for x 6∈ E.
For example, choosing ϕ(x) = e−x we find that if v is increasing and continuous,
then  
1
v x+ < Kv(x) for x 6∈ E.
v(x)
In this form it is found in most standard books. Sometimes the result is also applied
to w = v ◦ log. Using that 1 + y ≤ ey for y ∈ R we find that
w(r(1 + ϕ(log w(r)))) ≤ w reϕ(log w(r))

= v(log r + ϕ(log v(log r)))

≤ Kv(log r)
= Kw(r)
28 6 MAXIMUM MODULUS AND CHARACTERISTIC

if log r 6∈ E.
Let
F = {r : log r ∈ E} .
0
Putting rk = exk and rk0 = exk the condition
∞
X
(x0k − xk ) < ∞
k=1

takes the form

∞
X rk0
log < ∞,
k=1
rk

and we have
∞
[
F ⊂ [rk , rk0 ] .
k=1

In particular, we have Z
dt
< ∞,
F t
corresponding to the condition
Z
dx < ∞.
E

We say that F has finite logarithmic measure.

Theorem 6.6. Let f be entire, r0 > 0 and ϕ : [r0 , ∞) → R continuous and de-
creasing with Z ∞
ϕ(t)dt < ∞.
r0

Then there exists a (closed) set F ⊂ [r0 , ∞) with

Z
dt
<∞
F t

such that
T (r, f )
log M (r, f ) ≤ for r 6∈ F.
ϕ(log T (r, f ))

Remark. Taking ϕ(t) = t−α where α > 1 we obtain

log M (r, f ) ≤ T (r, f ) [log T (r, f )]α for r 6∈ F.

Proof of Theorem 6.6. We apply Borel’s Lemma 6.1 and the remark following its
proof with w(r) = T (r, f ) and K = 2. Taking R = r(1 + ϕ(log T (r, f ))) in
6 MAXIMUM MODULUS AND CHARACTERISTIC 29

Theorem 6.3 we have

R+r
log M (r, f ) ≤ T (R, f )
R−r
2 + ϕ(log T (r, f ))
≤ · 2T (r, f )
ϕ(log T (r, f ))
T (r, f )
≤5
ϕ(log T (r, f ))
R∞
for large r 6∈ F . (Here we have used that r0 ϕ(t)dt < ∞ and the monotonicity of
ϕ implies that limt→∞ ϕ(t) = 0.)
Since the hypothesis remains valid if ϕ is replaced by 51 ϕ, the conclusion follows.

Theorem 5.5 says that T (r, f ) and N (r, f ) are convex in log r (and non-de-
creasing). The following result is an analogue for the maximum modulus.
Theorem 6.7 (Hadamard three circle theorem). Let f be entire, f 6= 0. Then
log M (r, f ) is convex in log r.
Proof. We have to show that if 0 < r1 < r < r2 , then
log r2 − log r log r − log r1
log M (r, f ) ≤ log M (r1 , f ) + log M (r2 , f ).
log r2 − log r1 log r2 − log r1
In order to do so, we consider for α ∈ R the function
φ : C \{0} → R, φ(z) = |z|α |f (z)|.
Then φ has no strict local maxima, since locally φ is the absolute value of a holo-
morphic function. Indeed, in any simply connected subdomain U of C \{0} there
exists a branch log of the logarithm, and
φ(z) = |z α f (z)| = |eα log z f (z)|
for z ∈ U .
We now choose α ∈ R such that
max φ(z) = max φ(z),
|z|=r1 |z|=r2

which is equivalent to
r1α M (r1 , f ) = r2α M (r2 , f ).
Solving the equation for α shows that this holds for
log M (r2 , f ) − log M (r1 , f )
α=− .
log r2 − log r1
Since φ has no local maxima we have
rα M (r, f ) ≤ r1α M (r1 , f ) = r1α M (r2 , f ).
Inserting the value of α found in this equation yields the conclusion.
30 6 MAXIMUM MODULUS AND CHARACTERISTIC

Instead of
M (r, f ) = max |f (z)|
|z|=r

we will sometimes consider

A(r, f ) := max Re f (z).

|z|=r

To relate the two quantities we will use the following theorem.

Theorem 6.8. Let f be holomorphic in D(0, r) and z ∈ D(0, r). Then

2π
reiθ + z
Z
1
f (z) = Re f (reiθ ) dθ + i Im f (0).
2π 0 reiθ − z

Proof. Putting ξ = reiθ we have

 iθ   
re + z ξ+z
Re = Re
reiθ − z ξ−z
 
(ξ + z)(ξ − z)
= Re
|ξ − z|2
 2
|ξ| − ξz + zξ − |z|2

= Re
|ξ − z|2
 2
|ξ| + 2i Im(zξ) − |z|2

= Re
|ξ − z|2
|ξ|2 − |z|2
=
|ξ − z|2
= K(z, r, θ).

Putting
2π
reiθ − z
Z
1
h(z) = Re f (reiθ ) dθ
2π 0 reiθ + z
we find that
Z 2π
1
Re h(z) = Re f (reiθ )K(z, r, θ)dθ = Re f (z)
2π 0

by Poisson’s integral formula (Theorem 6.1 and the corollary of it).

The above expression for h defines a holomorphic function h : D(0, r) → C. We
deduce that f = h + c with a constant c. The constant c is then computed by
considering the value for z = 0. We conclude that
Z 2π
1
c = f (0) − h(0) = f (0) − Re f (reiθ )dθ = f (0) − Re f (0) = i Im f (0)
2π 0

and the conclusion follows.

6 MAXIMUM MODULUS AND CHARACTERISTIC 31

Theorem 6.9. Let f be entire and 0 < r < R. Then

R+r
M (r, f ) ≤ 2 (max{A(R, f ), 0} + |f (0)|) .
R−r
Proof. Let z ∈ C with |z| = r. Theorem 6.7 yields that
Z 2π
1 Reiθ + z
|f (z)| = Re f (Reiθ ) iθ dθ + i Im f (0)
2π 0 Re − z
Z 2π
1 R + |z|
≤ Re f (Reiθ ) dθ + |Im f (0)|
2π 0 R − |z|
Z 2π
R+r 1
= · Re f (Reiθ ) dθ + |Im f (0)| .
R − r 2π 0
We also have Z 2π
1
Re f (0) = Re f (Reiθ )dθ
2π 0
and thus Z 2π
R+r R+r 1
Re f (0) = · Re f (Reiθ )dθ.
R−r R − r 2π 0
Adding this to the first inequality and noting that |x| + x = 2 max{x, 0} for x ∈ R
we obtain
Z 2π
R+r R+r 1
|f (z)| + Re f (0) ≤ · 2 max{Re f (Reiθ ), 0}dθ + |Im f (0)|
R−r R − r 2π 0
R+r
≤2 max{A(R, f ), 0} + |Im f (0)| .
R−r
We conclude that
R+r R+r
M (r, f ) ≤ 2 max{A(R, f ), 0} + |Im f (0)| − Re f (0)
R−r R−r
R+r
≤2 (max{A(R, f ), 0} + |f (0)|) .
R−r
Remark. If f is a non-constant entire function, then M (r, f ) → ∞ by Liouville’s
theorem. Theorem 6.9 implies that also A(r, f ) → ∞. In particular, A(r, f ) > 0
for large r so that we actually have
R+r
M (r, f ) ≤ 2 (A(R, f ) + |f (0)|) .
R−r
for 0 < r < R if R is large.
In analogy to Theorem 6.4 and Theorem 6.5 we have the following result.
Theorem 6.10. Let f be entire with
log A(r, f )
L := lim inf < ∞.
r→∞ log r
Then f is a polynomial with deg(f ) ≤ L.
The proof is a direct consequence of Theorem 6.4 and Theorem 6.9.
32 7 THE ORDER OF A MEROMORPHIC FUNCTION

7 The order of a meromorphic function

Definition 7.1. Let f be meromorphic. Then
log T (r, f )
ρ(f ) := lim sup
r→∞ log r
is called the order of f and
log T (r, f )
λ(f ) := lim inf
r→∞ log r
is called the lower order of f .
r
Example. Since T (r, exp) = , we have ρ(exp) = λ(exp) = 1.
π
Remark. Suppose that T (r, f ) ≤ rK for r ≥ r0 . Then
log T (r, f )
≤K for r ≥ r0
log r
and thus ρ(f ) ≤ K. The order is the infimum of the set of all K for which an
estimate T (r, f ) ≤ rK holds for all large r. An analogous remark applies to the
lower order.
Theorem 7.1. Let f be entire. Then
log log M (r, f )
ρ(f ) = lim sup
r→∞ log r
and
log log M (r, f )
λ(f ) = lim inf .
r→∞ log r
The proof follows easily from Theorem 6.4 and is omitted here.
Example. (1) Let
z
f (z) = ee .
r
Then M (r, f ) = ee , so log log M (r, f ) = r and hence ρ(f ) = λ(f ) = ∞.

(2) Let
∞
eiz + e−iz X (−1)k 2k
f (z) = cos z = = z .
2 k=0
(2k)!
Then
er + er
M (r, f ) ≤ = er
2
and
er + e−r er
M (r, f ) ≥ |f (ir)| = ≥ .
2 2
It follows that log M (r, f ) = r + O(1) and hence ρ(f ) = λ(f ) = 1.
7 THE ORDER OF A MEROMORPHIC FUNCTION 33

(3) Let
∞
√ X (−1)k k
f (z) = cos z = z .
k=0
(2k)!
Then √
√ r
r e
M (r, f ) ≤ e and M (r, f ) ≥ f (−r) ≥
2
1
and hence ρ(f ) = λ(f ) = 2
.
The following result is classical, but we omit the proof.

Theorem 7.2. Let f be entire with Taylor series expansion

∞
X
f (z) = cn z n .
n=0

Then
n log n
ρ(f ) = lim sup .
n→∞ − log |cn |
Here we put n log n/(− log |cn |) = 0 if cn = 0.

Remark. (1) The result gives an easy possibility to construct functions with
ρ(f ) = µ for any preassigned µ ≥ 1. E.g. for 0 < µ < ∞ we can take
cn = n−n/ρ .

(2) Taking the lower limit in the formula in Theorem 7.2 does in general not give
the lower order. However, we have

nk log nk−1
λ(f ) = max lim inf ,
(nk ) k→∞ − log |cnk |

with the maximum taken over all increasing sequences (nk ) in N.

Theorem 7.3. Let f be meromorphic and M a Möbius transformation. Then

ρ(M ◦ f ) = ρ(f ) and λ(M ◦ f ) = λ(f ).

The proof follows directly from Theorem 5.2.

Theorem 7.4. Let f1 and f2 be meromorphic. Then

ρ(f1 ± f2 ) ≤ max{ρ(f1 ), ρ(f2 )},

ρ(f1 · f2 ) ≤ max{ρ(f1 ), ρ(f2 )},

and
ρ(f1 /f2 ) ≤ max{ρ(f1 ), ρ(f2 )}.
If ρ(f1 ) 6= ρ(f2 ), then we have equality in the above estimates.
34 8 WEIERSTRASS PRODUCTS

Proof. We only consider the sum f = f1 + f2 , the other cases are analogous.
By Theorem 5.1 we have
T (r, f ) ≤ T (r, f1 ) + T (r, f2 ) + log 2 ≤ 2 max {T (r, f1 ), T (r, f2 )} + log 2
and thus
log T (r, f ) ≤ log+ max {T (r, f1 ), T (r, f2 )} + O(1)
which implies that ρ(f ) ≤ max{ρ(f1 ), ρ(f2 )}. Suppose that ρ(f1 ) 6= ρ(f2 ), say
ρ(f1 ) < ρ(f2 ). Then
ρ(f2 ) = ρ(f − f1 ) ≤ max{ρ(f1 ), ρ(f2 )} ≤ ρ(f )
since ρ(f1 ) < ρ(f2 ). Thus ρ(f ) = ρ(f2 ) = max{ρ(f1 ), ρ(f2 )}.

8 Weierstraß products
The fundamental theorem of algebra says that a polynomial p(z) = ad z d + · · · + a0
with a0 , ad 6= 0 has a factorization
d d  
Y Y z
p(z) = ad (z − zj ) = a0 1− ,
j=1 j=1
zj

with the zeros z1 , . . . , zd of p.

We consider the question whether entire functions with infinitely many zeros
can be written as infinite products in a similar way.
Let (aj )j∈N be a sequence of complex number. A naive definition for the con-
vergence of the infinite product ∞
Q Qk
j=1 aj would be the existence of limk→∞ j=1 aj .
This definition would have two disadvantages:
– if an = 0 for some n, then limk→∞ kj=1 aj = 0, regardless of the behaviour
Q
of aj as j → ∞;

– limk→∞ kj=1 aj = 0 is possible even if aj 6= 0 for all j. For example, this

Q
happens for aj = j/(j + 1).
Definition 8.1. Let (aj ) be a sequence in C. Then ∞
Q
j=1 aj is called convergent,
if there exists N ∈ N with aj 6= 0 for j ≥ N and if limk→∞ kj=N aj exists and
Q

limk→∞ kj=N aj 6= 0. In this case we put

Q

∞
Y k
Y
aj := a1 · a2 · . . . · aN −1 · lim aj .
k→∞
j=1 j=N

For a convergent infinite product ∞

Q Q∞
j=1 aj we easily see that j=1 aj = 0 if and
only if there exists j ∈ N with aj = 0. For k > N we have
Qk
j=N aj
ak = Qk−1
j=N aj
8 WEIERSTRASS PRODUCTS 35

and thus ak → 1 if the product ∞

Q
j=1 aj converges. This necessary condition for
convergence is not sufficient, as shown by the example aj = j/(j + 1) already
considered.
The condition Pak → 1 for infinite products corresponds to the condition ak → 0
for infinite series ∞j=1 aj . The
Qanalogue of the Cauchy criterion for infinite series
∞
says that the infinite product j=1 aj converges if and only if for every ε > 0 there
exists N ∈ N such that
n
Y
aj − 1 < ε for n > m ≥ N.
j=m

Since aj → 0 for a convergent infinite product ∞

Q
j=1 aj we write the factors aj
in the form aj = Q
1 + cj .
The product ∞
Q∞
j=1 (1 + cj ) is called absolutely convergent if j=1 (1 + |cj |) con-
verges. Since
n
Y Yn
(1 + cj ) − 1 ≤ (1 + |cj |) − 1
j=m j=m

for n ≥ m we see that absolutely convergent infinite products are convergent.

Moreover, the sequence !
Yn
(1 + |cj |)
j=1 n∈N

is non-decreasing, so it converges if and only if it is bounded.

Qn
Theorem 8.1. An infinite product j=1 (1 + cj ) converges absolutely if and only
if the series ∞
P
c
j=1 j converges absolutely.

Proof. Since log(1 + x) ≤ x for x > −1 we have

n
! n n
Y X X
log (1 + |cj |) = log(1 + |cj |) ≤ |cj |
j=1 j=1 j=1

so that the absolute convergence of the series implies that of the infinite product.
Suppose now that the infinite product converges absolutely. Then cj → 0 and thus
there exists N with |cj | ≤ 1 for j ≥ N . Since x ≤ 2 log(1 + x) for 0 ≤ x ≤ 1 we
deduce that
n n n
!
X X Y
|cj | ≤ 2 log(1 + |cj |) = 2 log (1 + |cj |)
j=N j=N j=N

for n ≥ N , from which the conclusion follows.

The above considerations extend to infinite products of functions. Definitions
like (locally) uniform convergence can be generalized to infinite products in an
obvious way. This yields the following result, the proof of which we omit.
36 8 WEIERSTRASS PRODUCTS

Theorem 8.2. Let D be a domain and let (uj ) be a sequence of functions holo-
morphic in D. Suppose that ∞
P
j=1 |u j | converges locally uniformly in D. Then

∞
Y
f (z) := (1 + uj (z))
j=1

converges locally uniformly in D and defines a holomorphic function f : D → C.

Moreover, for z ∈ D we have f (z) = 0 if and only if there exists j ∈ N such that
1 + uj (z) = 0.
When looking for an entire function whose zeros consist of a given sequence
(zj ), it is
Q∞ tempting to choose uj (z) = −z/zj and thus to consider the infinite
product j=1 (1 − z/zj ). However, in general this infinite product will diverge. For
example, this is the case for zj = j. The following definition will lead to a suitable
modification.
Definition 8.2. Let q ∈ N0 . The entire functions E(·, q) given by

E(u, 0) = 1 − u

and
q
!
X uk
E(u, q) = (1 − u) exp
j=1
k
for q ∈ N are called Weierstraß primary factors.
Lemma 8.1. Let q ∈ N0 and u ∈ D(0, 1). Then

|E(u, q) − 1| ≤ |u|q+1 .

Proof. For q = 0 this is obvious. So let q ∈ N. Then

q q
! ! q
d X uk X uk X k−1
E(u, q) = − exp + (1 − u) exp u
du k=1
k k=1
k k=1
q q−1
! !
Xu k X
= exp −1 + (1 − u) uk
k=1
k k=0
q q−1
! !
X uk X
k k+1


= exp −1 + u −u
k=1
k k=0
q
!
q
X uk
= −u · exp .
k=1
k

Since all coefficients of the Taylor series expansion of the right hand side are neg-
ative we conclude that ∞
X
E(u, q) = 1 − αj uj
j=q+1
8 WEIERSTRASS PRODUCTS 37

where αj ≥ 0 for all j ≥ q + 1. Since E(1, q) = 0 we find that

∞
X
αj = 1.
j=q+1

Hence
∞
X
|E(u, q) − 1| ≤ αj |u|j
j=q+1
∞
X
= |u|q+1 αj |u|j−q−1
j=q+1
X∞
≤ |u|q+1 αj
j=q+1
q+1
= |u| .
Theorem 8.3. Let (zj ) be a sequence in C \{0} with limj→∞ |zj | = ∞. Then there
exists a sequence (qj ) in N0 such that
∞ qj +1
X z
j=1
zj

converges locally uniformly in C.

If (qj ) has this property, then
∞  
Y z
E , qj
j=1
zj

converges locally uniformly in C and represents an entire function whose zeros are
precisely the zj .
Here multiplicities are counted in the sense that if a ∈ C appears m times in
the sequence (zj ), i.e., card{j ∈ N : zj = a} = m, then a is a zero of multiplicity m.
Proof. The first claim is satisfied for qj = j. To see this let R > 0. Then there
exists j0 ∈ N with |zj | ≥ 2R for j ≥ j0 . For z ∈ D(0, R) and j ≥ j0 we thus have
q +1 j+1  j+1
z j z R 1
= ≤ = j+1 .
zj zj 2R 2
P∞ qj +1
This implies that the sequence j=1 |z/zj | converges uniformly in D(0, R).
Since R > 0 was arbitrary this yields locally uniform convergence in C.
Let now (qj ) be a sequence such that this series converges locally uniformly.
Theorem 8.2 now yields that
∞  
Y z
E ,q
j=1
zj

converges locally uniformly and has the required properties.

38 9 THE HADAMARD FACTORIZATION THEOREM

Remark. It is easy to show that qj = [log j] is also an admissible choice. In Section 9

we will consider the case where (qj ) is constant.

Theorem 8.4 (Weierstraß factorization theorem). Let f be entire with infinitely

many zeros. If f (0) = 0, let m be the multiplicity of this zero, and put m = 0
otherwise. Let (zj ) be the sequence of zeros of f in C \{0}, counted according to
multiplicity, and let (qj ) be as in Theorem 8.3. Then there exists an entire function
g such that
∞  
g(z) m
Y z
f (z) = e z E , qj .
j=1
zj

Proof. The function

f (z)
z 7→  
m
Q ∞ z
z j=1 E zj , qj

is entire without zeros and thus has the form eg with an entire function g.
Remark. If f has only finitely many zeros, we have the same result, except that
the product occurring is finite.

Theorem 8.5. Let f be meromorphic in C. Then there are entire functions g and
h such that f = g/h.

Proof. Let (zj ) be the (finite or infinite) sequence of poles of f in C \{0}, counted
according to multiplicity. By Theorem 8.3 there exists an entire function k which
has precisely the zj as zeros.
If 0 is a pole of f , let m be its multiplicity and put m = 0 otherwise. Define h
by h(z) = z m k(z). Then h is entire and so is g = f h. The conclusion follows.

9 The Hadamard factorization theorem

Definition 9.1. Let (zj ) be a sequence in C \{0} with |zj | → ∞ as j → ∞. Then
( ∞
)
X 1
σ := σ((zj )) := inf µ ∈ R : <∞
j=1
|z|µ

is called the exponent of convergence of the sequence (zj ), with σ = ∞ if the series
diverges for all µ.
If σ < ∞, then
( ∞
)
X 1
q := q((zj )) := min m ∈ N : m+1
<∞
j=1
|zj |

is called the genus (German: Geschlecht) of the sequence (zj ).

Remark. 1. Clearly q ≤ σ ≤ q + 1.
9 THE HADAMARD FACTORIZATION THEOREM 39

2. The exponent of convergence is often denoted by λ in the literature, but we

reserve this for the lower order.

3. If (zj ) is a sequence in C with |zj | → ∞ as j → ∞, with zj = 0 for some j, the

exponent of convergence is defined by omitting these zj from the sequence.

4. Any (infinite) discrete subset M of C has the form M = {zj : j ∈ N} with

some sequence (zj ) satisfying |zj | → ∞. We define the exponent of conver-
gence of M as that of (zj ).

Theorem 9.1. Let (zj ) be as in Definition 9.1. Let n(r) be the number of zj in
D(0, r) and put Z r
n(t)
N (r) = dt.
0 t
Then
log n(r) log N (r)
σ = lim sup = lim sup .
r→∞ log r r→∞ log r
In the proof we will use the following lemmas.

Lemma 9.1. Let (zj ) and n(r) be as in Theorem 9.1, and let µ > 0. Then
X 1 Z r
n(t) n(r)
µ
=µ µ+1
dt + µ .
|zj | 0 t r
|zj |≤r

Proof. We proceed as in the proof of Lemma 3.2 and find that

Z r ! Z r
X 1 X dt 1 n(t) n(r)
µ
= µ µ+1
+ µ =µ µ+1
dt + µ .
|zj | |zj | t r 0 t r
|zj |≤r |zj |≤r

Remark. A proof using the Riemann-Stieltjes integral is also instructive:

X 1 Z r
1
µ
= dn(t)
|zj | ε tµ
|zj |≤r
r Z r  
1 1
= µ n(t) − n(t)d µ
t t
ε
Z rε
n(r) n(t)
= µ +µ µ+1
dt.
r ε t

Here 0 < ε < minj |zj |.

Lemma 9.2. Let (zj ) and n(r) be as in Theorem 9.1, and let µ > 0. Then
∞
X 1
µ
<∞
j=1
|zj |
40 9 THE HADAMARD FACTORIZATION THEOREM

if and only if Z ∞
n(t)
dt < ∞.
0 tµ+1
In this case
n(r)
lim = 0.
r→∞ r µ

Proof. Suppose first that the series converges. Lemma 9.1 implies that
R ∞the integral
µ
also converges and the limit limr→∞ n(r)/r = 0 exists. But since 1 dt/t = ∞,
the existence of the limit and the comparison test imply that the limit is actually
equal to 0.
Suppose now that the integral converges. Using the comparison test as before we
see that lim inf r→∞ n(r)/rµ = 0. Lemma 9.1 now yields that the series converges,
and as before this yields that in fact limr→∞ n(r)/rµ = 0.

Proof of Theorem 9.1. Suppose that σ < ∞ and choose µ with σ < µ < ∞. Then
∞
X 1
µ
< ∞.
j=1
|zj |

Lemma 9.2 implies that

n(r)
lim = 0.
r→∞ r µ

We thus have n(r) ≤ rµ and hence log n(r) ≤ µ log r for large r and hence

log n(r)
lim sup ≤ µ.
r→∞ log r

As µ ∈ (σ, ∞) was arbitrary, this yields that

log n(r)
lim sup ≤σ
r→∞ log r

Trivially this holds if σ = ∞.

Suppose now that
log n(r)
τ := lim sup < ∞.
r→∞ log r
We want to show that σ ≤ τ . In order to do so, choose ν, µ with τ < ν < µ < ∞.
For large r we then have log n(r) ≤ ν log r and hence n(r) ≤ rν . This implies that
Z ∞
n(t)
< ∞.
0 tµ+1

Lemma 9.2 now yields that

∞
X 1
µ
< ∞.
j=1
|zj |
9 THE HADAMARD FACTORIZATION THEOREM 41

Since µ can be chosen arbitrarily close to τ we conclude that τ ≤ σ. For σ = ∞

this is trivial. This proves the first equation claimed in Theorem 9.1.
The second equation follows easily from Lemma 3.3 which yields that
N (r) ≤ n(r) log r + O(1)
and
1
n(r) ≤ N (Kr)
log K
if K > 1. We omit the details.
Theorem 9.2. Let f be meromorphic, a ∈ C,
b and let (zj ) be the sequence of
a-points of f . Then σ((zj )) ≤ ρ(f ).
Proof. By Theorem 9.1 we have
log N (r, a)
σ((zj )) = lim sup .
r→∞ log r
By the first fundamental theorem, we have N (r, a) ≤ T (r, f )+O(1). The conclusion
follows.
Remark. We will see later that we have equality in Theorem 9.2 except for at
most two values of a. This is Borel’s theorem, and such values are called Borel
exceptional values.
Theorem 9.3. Let (zj ) be a sequence in C \{0}, with limj→∞ |zj | = ∞ and with
finite exponent of convergence. Let q be the genus of (zj ) and let
∞  
Y z
P (z) = E ,q .
j=1
zj

Then, with n(r) as in Theorem 9.1,

 Z r Z ∞ 
n(t) n(t)
log M (r, P ) ≤ L(q) rq dt + rq+1 dt ,
0 tq+1 r tq+2
where (
1 if q = 0,
L(q) =
(q + 1)(2 + log q) if q ≥ 1.
Remark. The product for P converges by Theorem 8.3 (and the definition of q).
Such a product P is also called canonical product.
We will first prove the following lemma.
Lemma 9.3. Let q ∈ N0 and u ∈ C. Then
log |E(u, p)| ≤ |u|q+1 for |u| ≤ 1, u 6= 1
and, if q ≥ 1,
log |E(u, p)| ≤ M (q)|u|q for |u| ≥ 1, u 6= 1
with
M (q) = 2 + log q.
42 9 THE HADAMARD FACTORIZATION THEOREM

Proof. Lemma 8.1 yields if |u| ≤ 1, then

log |E(u, p)| ≤ log(1 + |E(u, p) − 1|) ≤ |E(u, p) − 1| ≤ |u|q+1 .

This is the first inequality claimed. To prove the second one, let q ≥ 1 and |u| ≥ 1.
Then
q q q
!
X 1 j X 1 q X 1
log |E(u, p)| ≤ log |1 − u| + |u| ≤ |u| + |u| ≤ |u|q 1 + .
j=1
j j=1
j j=1
j

Since q Z q
X 1 dt
≤ = log q
j=2
j 1 t

the conclusion follows.

Proof of Theorem 9.3. We have, for z with |z| = r,
∞ 
X z
log |P (z)| = log E ,q
j=1
zj
   
X z X z
= log E ,q + log E ,q
zj zj
|zj |≤r |zj |>r

=: Σ1 + Σ2 .

We first estimate Σ2 . By Lemma 9.3 we have

X  r q+1 X 1
Σ2 ≤ = rq+1 .
|zj | |zj |q+1
|zj |>r |zj |>r

Lemma 9.1 says that if R > 0, then

Z R
X 1 n(t) n(R) n(r)
= (q + 1) dt + q+1 − q+1 .
|zj |q+1 r tq+2 R r
r<|zj |≤R

Since (zj ) has genus q, Lemma 9.2 yields that the integral
Z ∞
n(t)
dt
r tq+2

converges and
n(R)
lim = 0.
R→∞ Rq+1

It follows that Z ∞
X 1 n(t) n(r)
= (q + 1) dt − q+1
|zj |q+1 r tq+2 r
|zj |>r
9 THE HADAMARD FACTORIZATION THEOREM 43

and thus Z ∞
q+1 n(t)
Σ2 ≤ (q + 1)r dt − n(r).
r tq+2
If q = 0, we have
X z
Σ1 = log 1 −
zj
|zj |≤r
 
X r
≤ log 1 +
|zj |
|zj |≤r
 
X r
≤ log 2
|zj |
|zj |≤r
X X r
= log 2 + log
|zj |
|zj |≤r |zj |≤r
Z r
n(t)
= n(r) log 2 + dt
0 t
Z r
n(t)
≤ n(r) + dt
0 t
by Lemma 3.2. Combing the estimates for Σ1 and Σ2 gives the conclusion in this
case.
If q ≥ 1, then, by Lemmas 9.3 and 9.1,
X  r q
Σ1 ≤ M (q)
|zj |
|zj |≤r
X 1
= M (q)rq
|zj |q
|zj |≤r
 Z r 
q n(t) n(r)
= M (q)r q q+1
dt + q
0 t r
Z r
n(t)
= M (q)q rq q+1
dt + M (q)n(r).
0 t

Since M (q) ≥ 1, the estimate for Σ2 also yields

Z r
q+1 n(t)
Σ2 ≤ M (q)(q + 1)r q+2
dt − M (q)n(r).
0 t

Adding this to the estimate for Σ1 yields the conclusion, since we may replace
M (q)q also by M (q)(q + 1) before the first integral.

Remark. The constants L(q) and M (q) are not sharp.

Theorem 9.4. Let (zj ), q and P be as in Theorem 9.3; that is, P is the Weierstraß
product formed with the zj , of genus q. Then ρ(P ) = σ((zj )).
44 9 THE HADAMARD FACTORIZATION THEOREM

Proof. By Theorem 9.2 we have σ := σ((zj )) ≤ ρ(P ). So we only have to prove

the opposite inequality. As remarked after Definition 9.1, we have q ≤ σ ≤ q + 1.
We distinguish two cases:

Case 1: q ≤ σ < q + 1. Choose ε > 0 with σ + ε < q + 1. Then there exists

K > 0 such that n(r) ≤ Krσ+ε for r > 0, by Theorem 9.1. Theorem 9.3
implies that
 Z r Z ∞ 
q σ+ε−q−1 q+1 σ+ε−q−2
log M (r, P ) ≤ L(q) · K r t dt + r t dt
0 r
σ+ε−q
rσ+ε−q−1
 
q r q+1
= L(q) · K r +r
σ+ε−q q+1−σ−ε
 
1 1
= L(q) · K + rσ+ε .
σ+ε−q q+1−σ−ε

It follows that ρ(P ) ≤ σ + ε, and this yields ρ(P ) ≤ σ since ε can be chosen
arbitrarily small.

Case 2: σ = q + 1. Again we have n(r) ≤ Krσ+ε with a constant K, for any

given ε > 0. By Lemma 9.1, the integral
Z ∞
n(t)
dt
0 tq+2

converges. It follows from Theorem 9.3 that

 Z r Z ∞ 
q σ+ε−q−1 q+1 n(t)
log M (r, P ) ≤ L(q) Kr t dt + r dt
0 r tq+2
1
rσ+ε + o rq+1 .


= L(q) · K
σ+ε−q

and hence again that ρ(P ) ≤ σ.

Theorem 9.5 (Hadamard factorization theorem). Let f be meromorphic with

ρ(f ) < ∞. Let P0 and P∞ be the canonical products formed with the zeros and
poles of f in C \{0}, respectively. Let cm z m with cm 6= 0 be the first non-vanishing
term in the Laurent series of f near 0. Then there exists a polynomial Q with
deg Q ≤ ρ(f ) such that
P0 (z)
f (z) = z m eQ(z) .
P∞ (z)
Remark. 1. If f (0) = 0, then m is the order of this zero. If f (0) = ∞, then −m
is the order of this pole. Otherwise m = 0.

2. The main difference to the Weierstraß factorization theorem is the additional

hypothesis that f has finite order. This implies that Q is a polynomial (and
not just entire).
9 THE HADAMARD FACTORIZATION THEOREM 45

Proof of Theorem 9.5. The function h defined by

P∞ (z) −m
h(z) = f (z) z
P0 (z)
is entire and has no zeros. Then it is of the form

h(z) = eQ(z)

with an entire function Q. Moreover, by Theorem 7.4,

ρ(h) ≤ max ρ(f ), ρ(P∞ ), ρ(P0 ), ρ z −m .



Now ρ(z −m ) = 0 and denoting by (zj ) and (pj ) the sequences of zeros and poles in
C \{0} we have
ρ(P0 ) = σ((zj )) ≤ ρ(f )
and
ρ(P∞ ) = σ((pj )) ≤ ρ(f ).
Hence
ρ(h) ≤ ρ(f ).
It follows that for ε > 0 we have

log M (r, h) ≤ rρ(f )+ε

for large r. Since

log M (r, f ) = log max eQ(z) = log max eRe Q(z) = A(r, Q)
|z|=r |z|=r

we thus have
A(r, Q) ≤ rρ(f )+ε
for large r. Theorem 6.10 yields that Q is a polynomial and deg Q ≤ ρ(f ) + ε. The
conclusion follows since ε can be taken arbitrarily small.
Example 9.1. Let f (z) = sin πz. The zeros are 0, ±1, ±2, . . . . Since
X 1
<∞ for µ>1
|k|µ
k∈Z\{0}

while X 1
=∞ for µ ≤ 1,
|k|µ
k∈Z\{0}

the exponent of convergence and the genus of the sequence of zeros are both equal
to 1. Moreover, ρ(f ) = 1 and f has a simple zero at 0. By Hadamard’s factorization
theorem there exits a polynomial Q of degree at most 1 such that
 
Q(z)
Y z
f (z) = ze E ,1 .
zj
j∈Z\{0}
46 9 THE HADAMARD FACTORIZATION THEOREM

Writing Q(z) = az + b we have

Y  z

sin πz = ze az+b
1− e−z/j
j
j∈Z\{0}
∞    
az+b
Y z −z/j z
= ze 1− e 1+ ez/j
j=1
j j
∞
z2
Y  
az+b
= ze 1− 2 .
j=1
j
In order to determine a and b we write this as
sin πz 1
eaz+b = · Q∞ 2 2
.
z j=1 (1 − z /j )

We conclude that
eaz+b = ea(−z)+b = e−az+b
for all z ∈ C and thus a = 0. Hence
sin πz 1
eb = Q∞ 2 2
z j=1 (1 − z /j )

for all z ∈ C, where “for all z ∈ C” is understood to mean that if the expression
on the right hand side has a removable singularity, then it has to be replaced by
the appropriate value. It follows that
sin πz 1 sin πz
eb = lim · Q∞ 2 2
= lim = π.
z j=1 (1 − z /j ) z
z→0 z→0

Altogether we find that

∞ 
z2
Y 
sin πz = πz 1− 2 .
j=1
j
1
E.g. for z = 2
we obtain
sin 12 π
π= ∞  
1Y 1
1− 2
2 j=1 4j
2
= ∞
Y 4j 2 − 1
j=1
4j 2
∞
Y 4j 2
=2·
j=1
4j 2 − 1
∞
Y 2j 2j
=2· ·
j=1
2j − 1 2j + 1
2 2 4 4 6 6
=2· · · · · · · ....
1 3 3 5 5 7
9 THE HADAMARD FACTORIZATION THEOREM 47

This formula for π is known as Wallis’ product.

Example 9.2. Let f be meromorphic without zeros and poles at 0, −1, −2, . . . (and
no other poles). The exponent of convergence and genus of the poles are 1. Thus
ρ(f ) ≥ 1. Assuming in addition that ρ(f ) = 1, we see that the function f with the
above properties are precisely those which have the form

1
f (z) = z −1 eaz+b Q∞ z −z/j
j=1 (1 + j )e

eaz+b
= Q∞
z j=1 (1 + zj )e−z/j
∞
eaz+b Y jez/j
= ·
z j=1
j+z

with a, b ∈ C.
The functions zf (z) and f (z + 1) have the same poles (in −1, −2, −3, . . . ) and
no zeros. Their quotient zf (z)/f (z +1) is thus entire and without zeros. Moreover,
it has order at most 1 and is thus of the form ecz+d with c, d ∈ C. More precisely,
we have
∞
zf (z) eaz+d (z + 1) Y j j + z + 1 ez/j
=
f (z + 1) ea(z+1)+b j=1 j + z j e(z+1)/j
n
−a
Y j+z+1
= e (z + 1) lim e−1/j
n→∞
j=1
j+z
n
!
X 1
= e−a lim (z + n + 1) exp −
n→∞
j=1
j
n
!
z+n+1 X1
= e−a lim exp log n −
n→∞ n j=1
j
n
!
X 1
= e−a lim exp log n −
n→∞
j=1
j
= e−a eγ
where !
n
X 1
γ = lim log n − .
n→∞
j=1
j

The limit γ = 0.57721 . . . is called the Euler-Mascheroni constant (or Euler’s con-
stant).
For a detailed proof that the limit exists one may write
n Z n  
X 1 1 1 1
− log n = − dx +
j=1
j 1 [x] x n
48 10 PREPARATIONS

and note that, since

1 1 x − [x] 1 2
0≤ − = ≤ ≤ 2 for x ≥ 1
[x] x x[x] x[x] x

the integral Z ∞  
1 1
− dx
1 [x] x
converges.
Choosing a = −γ in the definition of f we thus have f (z +1) = zf (z). Choosing
b = 0 we have
e−γz
f (1) = lim zf (z) = lim Q∞ −z/j
= 1.
j=1 (1 + z/j)e
z→0 z→0

We conclude that f (1) = 1, f (2) = 1, f (3) = 2 · 1 = 2, f (4) = 3 · f (3) = 3 · 2 and

in general
f (n) = (n − 1)! = 1 · 2 · 3 · . . . · (n − 1)
for n ∈ N. The function f obtained is called the Gamma function and denoted
by Γ, that is
∞
e−γz e−γz Y jez/j
Γ(z) = ∞  = .
z j + z

Y z
z 1+ e−z/j j=1

j=1
j

As noted above, we have Γ(n) = (n − 1)! for n ∈ N and Γ(z + 1) = zΓ(z) for
z ∈ C \{0, −1, −2, . . . }.
Remark. For Re z > 0 the Gamma function is often defined by
Z ∞
Γ(z) = e−t tz−1 dt.
0

It can be shown that this agrees with our definition. The functional equation
Γ(z + 1) = zΓ(z) follows from this via integration by parts.

10 Preparations for the second fundamental the-

orem
The first fundamental theorem says that
   
1 1
T (r, f ) = N (r, f ) + m(r, f ) = N r, + m r, + O(1)
f −a f −a

as r → ∞, for all a ∈ C. So if T (r, f ) is large, then one of the terms N (r, 1/(f − a))
or m(r, 1/(f − a)) must be large, meaning that f has many a-points or f is close
to a on some part of the circle ∂D(0, r).
10 PREPARATIONS 49

The second fundamental theorem will say that for “most” values of a the first
alternative will hold. More precisely, we will see that if a1 , . . . , aq ∈ C, then
q  
X 1
m r, + m(r, f ) ≤ 2T (r, f ) + . . .
j=1
f − aj

where the dots indicate a “small error term”. Together with the first fundamental
theorem this yields
q  
X 1
(q − 1)T (r, f ) ≤ N r, + N (r, f ) + . . . .
j=1
f − aj

In particular, this leads to a contradiction if f (z) 6= a1 , a2 , ∞ for all z ∈ C so that

Picard’s theorem follows. However, the result gives much stronger results than just
Picard’s theorem.
We will first give an upper bound for the difference of the left and right side of
the above inequalities. Later we will then show that the upper bound obtained is
indeed small in a suitable sense.

Theorem 10.1. Let f be meromorphic (and non-constant) and let a1 , . . . , aq ∈ C

be distinct. Then
q  
X 1
m r, + m(r, f ) ≤ 2T (r, f ) − N1 (r) + S(r)
j=1
f − aj

where  
1
N1 (r) = N r, 0 + 2N (r, f ) − N (r, f 0 ) ≥ 0
f
for r ≥ 1 and
q
f0 f0
  X  
S(r) ≤ m r, + m r, + O(1)
f j=1
f − aj

as r → ∞.

Remark. A pole of order k is counted 2k times in 2N (r, f ) and k + 1 times in

N (r, f 0 ). Altogether it is then counted 2k − (k + 1) times; that is, k − 1 times. It
follows that N1 (r) ≥ 0 for r ≥ 1.
An a-points of multiplicity k ≥ 2 is a zero of f 0 of multiplicity k − 1 and thus
counted k − 1 times in N (r, 1/f 0 ). Together with the above considerations for poles
we see that N1 (r) is a term that counts the multiple values of f .
Proof of Theorem 10.1. Put
q
Y
P (w) = (w − aj ).
j=1
50 10 PREPARATIONS

By Theorem 5.3 we have

T (r, P ◦ f ) = q · T (r, f ) + O(1).

Since   q  
1 X 1
N r, = N r,
P ◦f j=1
f − aj
the first fundamental theorem yields that
   
1 1
m r, = T (r, P ◦ f ) − N r, + O(1)
P ◦f P ◦f
q  
X 1
= q · T (r, f ) − N r, + O(1)
j=1
f − a j
q   
X 1
= T (r, f ) − N r, + O(1)
j=1
f − a j
q  
X 1
= m r, + O(1).
j=1
f − aj

A partial fraction decomposition yields

q
1 X cj
=
P (w) j=1
w − aj

with constants cj ∈ C \{0}. (In fact, we have cj = 1/P 0 (aj ) for 1 ≤ j ≤ q.) It
follows, using Lemma 4.1, that
f0
   
1 1
m r, = m r, · + O(1)
P ◦f P ◦ f f0
f0
   
1
≤ m r, + m r, 0 + O(1)
P ◦f f
q
!
f0
 
X 1
= m r, cj + m r, 0 + O(1)
j=1
f − aj f
q q
f0
  X  
X
+ 1
≤ m r, + log |cj | + log q + m r, 0 + O(1).
j=1
f − a j j=1
f

Combining the above estimates we have

q    
X 1 1
m r, = m r, + O(1)
j=1
f − a j P ◦ f
q
f0
  X  
1
≤ m r, 0 + m r, + O(1).
f j=1
f − a j
11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE 51

The first fundamental theorem yields that

   
1 0 1
m r, 0 = T (r, f ) − N r, 0 + O(1)
f f
 
0 0 1
= N (r, f ) + m(r, f ) − N r, 0 + O(1)
f
0
= 2N (r, f ) − N1 (r) + m(r, f ) + O(1)
f0
 
= 2N (r, f ) − N1 (r) + m r, f + O(1)
f
f0
 
≤ 2N (r, f ) − N1 (r) + m r, + m(r, f ) + O(1)
f
Together with the previous inequality we thus have
q
f0
   
X 1
m r, ≤ 2N (r, f ) + m(r, f ) − N1 (r) + m r,
j=1
f − a j f
q
f0
X  
+ m r, + O(1).
j=1
f − a j

Adding m(r, f ) on both sides and noting that T (r, f ) = m(r, f ) + N (r, f ) by
definition yields the conclusion.

11 The lemma on the logarithmic derivative

Theorem 10.1 shows that the second fundamental theorem requires an estimate of
m(r, f 0 /f ). Since f 0 /f = (log f )0 , the term f 0 /f is called the logarithmic derivative.
In order to obtain the desired estimate, we begin with the following representation
of the logarithmic derivative.
Theorem 11.1. Let f be meromorphic in D(0, r) and let a1 , . . . , am be the zeros
and b1 , . . . , bn the poles of f in D(0, r). Let z ∈ D(0, r) with f (z) 6= 0, ∞. Then
Z 2π
f 0 (z) 1 iθ

2reiθ
= log f re dθ
f (z) 2π 0 (reiθ − z)2
m   X n  
X 1 aj 1 bj
+ + 2 − + 2 .
j=1
z − a j r − a j z j=1
z − b j r − b j z

Proof. Theorem 6.8 says that if g is holomorphic in D(0, r), then

Z 2π
1
reiθ + z
g(z) = Re g reiθ dθ + i Im g(0).
2π 0 reiθ − z
Differentiating this with respect to z we obtain
Z 2π
0 1
2reiθ
g (z) = Re g reiθ dθ.
2π 0 (reiθ − z)2
52 11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE

This yields the conclusion if f has no zeros and poles, since then f may be written in
the form f = eg with a holomorphic function f . Then f 0 /f = g and log |f | = Re g.
In the general case we assume again for simplicity that f has no zeros and poles
on ∂D(0, r) and, as in the proof of Theorem 3.2 and 6.2, consider
Qn
j=1 ϕbj (z)
h(z) = f (z) Qm
j=1 ϕaj (z)

with
r(z − a)
ϕa (z) = .
r2 − az
Then h has no zeros and poles in D(0, r) and |h(reiθ )| = |f (reiθ )| for θ ∈ R. Using
the result already proved we conclude that
Z 2π
1 2r h0 (z)
log f reiθ


dθ =
2π 0 (reiθ − z)2 h(z)
n 0 m 0
f 0 (z) X ϕbj (z) X ϕaj (z)
= + −
f (z) j=1
ϕbj (z) j=1 ϕaj (z)
n
f 0 (z) X
 
1 bj
= + +
f (z) j=1
z − bj r 2 − bj z
m  
X 1 aj
− +
j=1
z − aj r 2 − aj z

as claimed.

Remark. In the last equation we have used that if F = f1 · f2 and G = f1 /f2 , then

F0 f0 f0 G0 f0 f0
= 1+ 2 and = 1 − 2.
F f1 f2 G f1 f2

These rules can be proved (and remembered) by noting that locally we may write
log F = log f1 + log f2 with suitably chosen branches of the logarithm so that

F0 f0 f0
= (log F )0 = (log f1 )0 + (log f2 )0 = 1 + 2 ,
F f1 f2

with an analogous reasoning for G0 /G. Of course, the above rules can also be
checked with the product and quotient rules.

Theorem 11.2 (Jensen’s inequality). P Let I ⊂ R be an interval and f : I → R

convex. Let λ1 , . . . , λn ∈ (0, 1) with nj=1 λj = 1 and x1 , . . . , xn ∈ I. Then

n
! n
X X
f λj xj ≤ λj f (xj ).
j=1 j=1
11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE 53

Proof. The conclusion is trivial if all xj are equal. Suppose that this is not the
case. Then n
X
ξ := λj xj
j=1

is in the interior of I. Since f is convex, there exists an affine function g such that
g(ξ) = f (ξ) and g(t) ≤ f (t) for all t ∈ I. Here by affine we mean that g has the
form g(t) = at + b for certain a, b ∈ R. It follows that
n
!
X
f (ξ) = g(ξ) = aξ + b = a λj xj + b
j=1
n
! n
X X
=a λ j xj +b λj
j=1 j=1
n
X
= λj (axj + b)
j=1
n
X
= λj g(xj )
j=1
n
X
≤ λj f (xj ).
j=1

Theorem 11.3 (Jensen’s inequality, version for integrals). Let I ⊂ R be an inter-

val, f : I → R convex and g : [a, b] → I (Riemann) integrable. Then
 Z b  Z b
1 1
f g(t)dt ≤ f (g(t))dt.
b−a a b−a a
Remark. Since convex functions are continuous, it follows from the hypothesis of
the theorem that f ◦ g is also integrable.
Proof of Theorem 11.3. Let a ≤ t0 < t1 < . . . < tn = b be a partition of the
interval [a, b] and let τj ∈ [tj−1 , tj ] for j = 1, . . . , n. With
tj − tj−1
λj = and xj = g(τj )
b−a
we deduce from Theorem 11.2 that
n
! n
1 X 1 X
f (tj − tj−1 )g(τj ) ≤ (tj − tj−1 )f (g(τj )).
b − a j=1 b − a j=1

The conclusion follows by taking the limit through a sequence of partitions with
maximal interval length tending to 0, using also that f is continuous.
Remark. The result can be extended to Lebesgue integrable function, e.g. by ap-
proximation by continuous functions. Using f = exp and g = log h one obtains the
following result.
54 11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE

Corollary. Let h and log h be integrable over [a, b]. Then

Z b  Z b 
1 1
log h(t)dt ≤ log h(t)dt .
b−a a b−a a

The proof of the following simple lemma is omitted.

Lemma 11.1. Let 0 < α < 1 and x1 , . . . , xn ≥ 0. Then

n
!α n
X X
xj ≤ xαj .
j=1 j=1

Theorem 11.4. Let f be meromorphic with f (0) = 1 and let 0 < r < R and
0 < α < 1. Then
1+α !
f0
  
1 24 R T (R, f )
m r, ≤ log 1 + .
f α 1−α R−r rα

Proof. Let r < ρ < R and z ∈ D(0, ρ). Theorem 11.1 yields that if f (z) 6= 0, ∞,
then
Z 2π
f 0 (z) 1 2ρ X 1 aj
log f ρeiθ


≤ 2
dθ + + 2
f (z) 2π 0 (ρ − |z|) z − aj ρ − aj z
|aj |<ρ
X 1 bj
+ + 2 .
z − bj ρ − bj z
|bj |<ρ

Here the aj and bj are the zeros and poles of f . We combine the sequences of zeros
and poles into one sequence (cj ). Noting that

1
| log |w|| = log+ |w| + log+
|w|

for w ∈ C \{0} we thus conclude that

f 0 (z)
   X
2ρ 1 1 cj
≤ m(ρ, f ) + m ρ, + + .
f (z) (ρ − |z|)2 f z − cj ρ 2 − cj z
|cj |<ρ

For c ∈ D(0, ρ) we have

 
1 c 1 c̄ ρ(z − c)
+ 2 = 1+ .
z − c ρ − cz z−c ρ ρ2 − c̄z

Since the map given by

ρ(z − c)
z 7→
ρ2 − c̄z
11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE 55

maps D(0, ρ) onto D(0, 1) by Lemma 3.1, we deduce that

 
1 c 1 |c| 2
+ 2 ≤ 1+ ≤ .
z − c ρ − cz |z − c| ρ |z − c|

We conclude, noting that m(ρ, 1/f ) ≤ T (r, f ) since f (0) = 1, that

f 0 (z) 4ρ X 1
≤ 2
T (ρ, f ) + 2 .
f (z) (ρ − |z|) |z − cj |
|cj |<ρ

Together with Lemma 11.1 this yields

f 0 (z) f 0 (z)
 
+
log ≤ log 1 +
f (z) f (z)
α 
f 0 (z)

1
= log 1 +
α f (z)
 
α
1 (4ρ) X 1
≤ log1 + 2α
T (ρ, f )α + 2α .
α (ρ − |z|) |z − cj |α
|cj |<ρ

Integrating this, using Jensen’s inequality, and also noting that 4α < 4 and 2α < 2
we obtain
 
 0
 Z 2π α
f 1 1 4ρ X 1
m r, ≤ log1 + 2α
T (ρ, f )α + 2  dθ
f α 2π 0 (ρ − r) |re − cj |α
iθ
|cj |<ρ
 
α Z 2π
1 4ρ X 1 dθ
≤ log1 + 2α
T (ρ, f )α + 2 .
α (ρ − r) 2π 0 |re − cj |α
iθ
|cj |<ρ

Write cj = |cj |eiϕj . Then

Z 2π Z 2π
dθ dθ
=
0 |re − cj |α
iθ
0 |reiθ − |cj |eiϕj |α
Z 2π
dθ
= iθ−ϕ j − |c ||α
0 |re j
Z 2π
dt
=
0 |re − |cj ||α
it
Z 2π
dt
≤
0 r | sin t|α
α
Z π/2
4 dt
= α .
r 0 | sin t|α
56 11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE

Since sin t ≥ 2t/π for 0 ≤ t ≤ π/2 we thus have

2π π/2
4 2α
Z Z
dθ dt
≤ ·
0 |reiθ − cj |α rα π α 0 tα
t=π/2
4 2α t1−α
= α· α·
r π 1−α t=1
4 π
≤ α·
r 2(1 − α)
2π
= α .
r (1 − α)

Inserting this into the above estimate for m(r, f 0 /f ) we obtain

f0 4ρα
   
α 2
m r, ≤ log 1 + T (ρ, f ) + α (n(ρ, f ) + n (ρ, 1/f )) .
f (ρ − r)2α r (1 − α)

1 1
We now choose ρ = (R + r). Then ρ ≤ R and ρ − r = (R − r) and thus
2 2
4ρα α 16Rα
T (ρ, f ) ≤ T (R, f ).
(ρ − r)2α (R − r)α

We also have, by Lemma 3.3,

Z R
n(t, f ) R
N (R, f ) ≥ dt ≥ n(ρ, f ) log .
ρ t ρ

It is easy to see that

x−1
log x ≥ for x > 0.
x
This yields that
R R/ρ − 1 R−ρ R−r
log ≥ = =
ρ R/ρ R 2R
and hence that
1 2R 2R
n(ρ, f ) ≤ R
N (R, f ) ≤ N (R, f ) ≤ T (R, f )
log ρ R−r R−r

and also  
1 2R
n ρ, ≤ T (R, f ).
f R−r
Altogether we thus have

f0 16Rα
   
1 8R
m r, ≤ log 1 + T (R, f ) + T (R, f ) .
f α (R − r)2α (R − r)rα (1 − α)
11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE 57

The conclusion follows since

16Rα 16Rα rα (1 − α)
 
8R 1 8R
2α
+ α
= α +
(R − r) (R − r)r (1 − α) r (1 − α) (R − r)2α R−r
 2α !
1 R R
≤ α 16 +8
r (1 − α) R−r R−r
 1+α
1 R
≤ α · 24 .
r (1 − α) R−r

In the last estimate we simply used that 1 + α ≥ max{1, 2α}.

The term R − r in the denominator prevents us from taking R = r in Theo-
rem 11.4. However, the Borel lemma (Lemma 6.1) allows to estimate m(r, f 0 /f ) in
terms of T (r, f ) outside an exceptional set.
We recall a version of this lemma given in the remarks after Lemma 6.1: Let
x0 , y0 > 1, K > 1, ν : [x0 , ∞) → [ey0 , ∞) continuous and non-decreasing, and
ϕ : [y0 , ∞) → (0, ∞) continuous and non-increasing with
Z ∞
ϕ(y)dy < ∞.
y0

Then there exists a subset E of [x0 , ∞) of finite measure such that

ν(x + ϕ(log ν(x))) ≤ Kν(x) for x 6∈ E.

The choice ϕ(y) = 1/y 2 yields that

 
1
ν x+ ≤ Kν(x)
(log ν(x))2

for x 6∈ E.

Theorem 11.5. Let f be a transcendental meromorphic function and ε > 0. Then

there exists a subset E of [0, ∞) of finite measure such that

f0
 
m r, ≤ (1 + ε) (log T (r, f ) + log r)
f

for r 6∈ E.

Proof. Without loss of generality we may assume that f (0) = 1, since otherwise
we can replace f by g(z) = cz p f (z) with a suitable c ∈ C \{0} and p ∈ Z. Then
 0
f0
 
g
m r, = m r, + O(1)
g f

and
T (r, g) = T (r, f ) + O(log r) = (1 + o(1))T (r, f )
58 11 THE LEMMA ON THE LOGARITHMIC DERIVATIVE

by Theorem 6.5. Thus

log T (r, g) = log T (r, f ) + o(1),

which implies that f satisfies the conclusion if f does.

To apply the Borel lemma as in the remark above with ν(r) = T (r, f ) and
ϕ(y) = 1/y 2 we put
1
R=r+ .
(log T (r, f ))2
Then R ≤ 2r for large r so that Theorem 11.4 yields that if K > 1, then

f0
   
24 1+α 2+2α T (R, f )
α · m r, ≤ log 1 + R (log T (r, f ))
f 1−α rα
 
24 1+α 2+2α
≤ log 1 + 2 r(log T (r, f )) T (R, f )
1−α
 
24 1+α
≤ log 2 + log 2 + log r
1−α
+ (2 + 2α) log log T (r, f ) + log T (R, f )
 
24 1+α
≤ log 2 + log 2 + log r
1−α
+ (2 + 2α) log log T (r, f ) + log T (r, f ) + log K

for r 6∈ E, where E has finite measure. Thus

f0
  
ε
α · m r, ≤ 1+ log T (r, f ) + log r
f 2

for large r 6∈ E. Taking α close to 1 yields the conclusion.

Remark. If f is rational, then

f 0 (z)
 
1
=O
f (z) |z|

as z → ∞ and thus
f0
 
m r, =0
f
for all large r. Therefore it is no restriction to consider only transcendental func-
tions in Theorem 11.5.
If f has finite order ρ(f ), then Theorem 11.5 gives

f0
 
m r, ≤ (ρ(f ) + 1 + ε) log r for r 6∈ E.
f

The following result improves the constant on the right hand side and – more
importantly – says that in this case no exceptional set E is required.
12 THE SECOND FUNDAMENTAL THEOREM 59

Theorem 11.6. Let f be a meromorphic function of finite order. Then

m(r, f 0 /f )
lim sup ≤ max{ρ(f ) − 1, 0}.
r→∞ log r
Proof. We may again assume without loss of generality that f (0) = 1. We apply
Theorem 11.4 with R = 2r and obtain
f0
   
1 24 1+α T (2r, f )
m r, ≤ log 1 + 2 .
f α 1−α rα

Let ε > 0. Then T (2r, f ) ≤ (2r)ρ(f )+ε for large r and thus

f0
   
1 24 1+α ρ(f )+ε ρ(f )+ε−α
m r, ≤ log 1 + 2 2 r
f α 1−α
   
1 + 24 1+α+ρ(f )+ε + ρ(f )+ε−α


≤ log 2 + log r + log 2
α 1−α
max{ρ(f ) + ε − α, 0}
≤ log r + O(1)
α
as r → ∞. Considering the limits as α → 1 and ε → 0 we obtain the conclusion.

12 The second fundamental theorem and the de-

ficiency relation
The following result, called the second fundamental theorem of Nevanlinna theory,
is an immediate consequence of Theorem 10.1 and the result of Section 11.
Theorem 12.1. Let f be meromorphic and let a1 , . . . , aq ∈ C be distinct. Then
q  
X 1
m r, + m(r, f ) ≤ 2T (r, f ) − N1 (r) + S(r, f )
j=1
f − aj

where  
1
N1 (r) := N r, 0 + 2N (r, f ) − N (r, f 0 )
f
and the term S(r, f ) satisfies the following:
(i) There exists a subset E of [0, ∞) of finite measure such that

S(r, f ) = O(log T (r, f )) + O(log r)

as r → ∞, r 6∈ E.

(ii) If f has finite order, then S(r, f ) = O(log r) as r → ∞.

(iii) If f is rational, then S(r, f ) = O(1).

60 12 THE SECOND FUNDAMENTAL THEOREM

Remark. For transcendental f we have

T (r, f )
lim =∞
r→∞ log r

by Theorem 6.5. This implies that

S(r, f ) = o(T (r, f ))

as r → ∞, r 6∈ E. The conclusions (ii) and (iii) yield that this holds without the
exceptional set E if f has finite order.
Proof of Theorem 12.1. Theorem 10.1 says that the conclusion holds with
q
f0 f0
  X  
S(r, f ) = m r, + m r, + O(1).
f j=1
f − aj

Noting that T (r, f − aj ) = T (r, f ) + O(1) and f 0 /(f − aj ) = (f − aj )0 /(f − aj ) the

conclusion now follows from the results in Section 11.
The following result is another version of the second fundamental theorem,
using the counting functions instead of the proximity functions. Here we write
N (r, a) = N (r, 1/(f − a)) for a ∈ C and N (r, ∞) = N (r, f ).

Theorem 12.2. Let f be meromorphic and a1 , . . . , aq ∈ C

b be distinct. Then
q
X
(q − 2)T (r, f ) ≤ N (r, aj ) − N1 (r) + S(r, f ),
j=1

with S(r, f ) satisfying the conclusion of Theorem 12.1.

Proof. Suppose first that aj ∈ C for all j ∈ {1, . . . , q}. We add qj=1 N (r, aj ) to
P
the left and right hand side of the inequality in Theorem 12.1. Noting that
 
1
N (r, aj ) + m r, = T (r, f ) + O(1)
f − aj
by the first fundamental theorem we obtain
q
X
qT (r, f ) + m(r, f ) ≤ N (r, aj ) + 2T (r, f ) − N1 (r) + S(r, f ).
j=1

Subtracting 2T (r, f ) and noting that m(r, f ) ≥ 0 we obtain the conclusion.

Suppose now that aj = ∞ for some j. Without loss of generality we may assume
that aq = ∞. Then the above inequality holds with q replaced by q − 1. Thus
q−1
X
(q − 1)T (r, f ) + m(r, f ) ≤ N (r, aj ) + 2T (r, f ) − N1 (r) + S(r, f ).
j=1
12 THE SECOND FUNDAMENTAL THEOREM 61

Adding N (r, aq ) = N (r, ∞) = N (r, f ) on both sides we obtain

q
X
qT (r, f ) ≤ N (r, aj ) + 2T (r, f ) − N1 (r) + S(r, f ).
j=1

and thus the conclusion.

An alternative method to deal with the case that aj = ∞ for some j is to
consider 1/(f − c) instead of f for some c ∈ C with c 6= aj for all j.
Let n(r, f ) denote the number of distinct poles of a meromorphic function f
in the closed disk D(0, r); that is, multiplicities are not taken into account. In
analogy to the definition of N (r, f ) we put
Z r
n(t, f ) − n(0, f )
N (r, f ) = dt + n(0, f ) log r.
0 t
As before we also write n(r, ∞) and N (r, ∞) instead of n(r, f ) and N (r, f ) and we
put    
1 1
n(r, a) = n r, and N (r, a) = N r, .
f −a f −a
An a-point of multiplicity k is thus counted k times in N (r, a), but only once in
N (r, a). In N1 (r) it is counted k − 1 times. This implies that
q
X

N (r, aj ) − N (r, aj ) ≤ N1 (r)
j=1

for r ≥ 1. Theorem 12.2 then yields the following result.

Theorem 12.3. Let f be meromorphic and a1 , . . . , aq ∈ C be distinct. Then
q
X
(q − 2)T (r, f ) ≤ N (r, aj ) + S(r, f ),
j=1

with S(r, f ) as in Theorem 12.1.

Picard’s theorem is a simple consequence of Theorem 12.2 (or Theorem 12.3).
The following definitions will be used to state some generalizations.
Definition 12.1. Let f be meromorphic and a ∈ C. b Then
N (r, a)
δ(a) = δ(a, f ) = 1 − lim sup
r→∞ T (r, f )
is called the (Nevanlinna) deficiency of a, and
N (r, a) − N (r, a)
ε(a) = ε(a, f ) = lim inf
r→∞ T (r, f )
is called the ramification index of a. Moreover, we put
N (r, a)
Θ(a) = Θ(a, f ) = 1 − lim sup .
r→∞ T (r, f )
62 12 THE SECOND FUNDAMENTAL THEOREM

Remark. By the first fundamental theorem, we have 0 ≤ δ(a) ≤ 1. If f has no

a-points, then δ(a) = 1. In general, the deficiency δ(a) should be thought of as a
measure of how often f takes the value a. A “large” deficiency means that f takes
the value relatively “few” times, and vice versa.
A similar interpretation applies to Θ(a). Analogously, ε(a) > 0 means that f
has relatively “many” multiple a-points.
It follows easily from the definition that
0 ≤ δ(a) + ε(a) ≤ Θ(a) ≤ 1

for all a ∈ C.
b
In analogy to the terminology n(r, a) we also put
 
1
m(r, a) = m r,
f −a
if a ∈ C and m(r, ∞) = m(r, f ). The first fundamental theorem thus takes the
form
N (r, a) + m(r, a) = T (r, f ) + O(1)
for all a ∈ C.
b This implies that
m(r, a)
δ(a, f ) = lim inf .
r→∞ T (r, f )

Theorem 12.4. Let f be meromorphic. Then the set A = {a ∈ C

b : Θ(a, f ) > 0}
is countable and X X
Θ(a, f ) := Θ(a, f ) ≤ 2.
a∈C
b a∈A

Corollary. For a meromorphic function f we have

X
(δ(a, f ) + ε(a, f )) ≤ 2
a∈C
b

and, in particular, X
δ(a, f ) ≤ 2.
a∈C
b

The second inequality in the corollary is called the deficiency relation.

Proof of Theorem 12.4. Let a1 , . . . , aq ∈ C.
b Dividing the inequality
q
X
(q − 2)T (r, f ) ≤ N (r, aj ) + S(r, f )
j=1

from Theorem 12.3 by T (r, f ) and taking a limit as r → ∞ through a sequence of

r-values outside the exceptional set E we get
q
X N (r, aj )
q−2≤ lim sup
j=1
r→∞ T (r, f )
12 THE SECOND FUNDAMENTAL THEOREM 63

and thus
q
X
Θ(aj , f ) ≤ 2.
j=1

This implies that the set

 
2
Aq = a ∈ C : Θ(a, f ) ≥
q
S
has at most q elements. In particular, Aq is finite and thus A = q∈N Aq is count-
able. Writing A = {a1 , a2 , a3 , . . . } we obtain the conclusion.

Remark. Let f be a transcendental meromorphic function. Suppose that f takes

the values a1 , . . . , aq only finitely often. Then δ(aj ) = 1 for all j and thus q ≤ 2 by
the deficiency relation. Thus Picard’s theorem follows from the deficiency relation.

Theorem 12.5. Let f be a non-constant meromorphic function, a1 , . . . , aq ∈ C b

distinct and m1 , . . . , mq ∈ N. Suppose that all aj -points of f have multiplicity at
least mj , for j = 1, . . . , q. Then
q  
X 1
1− ≤ 2.
j=1
mj

Proof. By hypothesis we have

N (r, aj ) ≥ mj N (r, aj ).

Together with the first fundamental theorem we obtain

1 1
N (r, aj ) ≤ N (r, aj ) ≤ T (r, f ) + O(1)
mj mj

and thus
N (r, aj ) 1
Θ(aj , f ) = 1 − lim sup ≥1− .
r→∞ T (r, f ) mj
The conclusion now follows from Theorem 12.4.

Remark. If f does not take a value aj at all, that is, f (z) 6= aj for all z ∈ C, then
the hypothesis of Theorem 12.5 is satisfied for any mj ∈ N.
Taking the limit as mj → ∞ we may put 1/mj = 0 in Theorem 12.5 then.
Example. Let f (z) = cos z, a1 = 1, a2 = −1, a3 = ∞. We can take m1 = m2 = 2
and, in the sense of the previous remark, m3 = ∞ and 1/m3 = 0. Then we have
equality in Theorem 12.5.

Definition 12.2. Let f be meromorphic and a ∈ C.

b Then a is called totally
ramified if f has no simple a-points.
64 13 UNIQUENESS THEOREMS

Theorem 12.6. A non-constant meromorphic function has at most four totally

ramified values.
A non-constant entire function has at most two totally ramified values in C.

Proof. Let a1 , . . . , aq be totally ramified values. With m1 = m2 = . . . = mq = 2 we

deduce from Theorem 12.5 that q · 12 ≤ 2 and thus q ≤ 4.
If f is entire, we may choose aq = ∞ and mq = ∞. Applying Theorem 12.5
with m1 = m2 = . . . = mq−1 = 2 yields that q − 1 ≤ 2, so there are at most 2
totally ramified values in C.
The example f (z) = cos z considered above shows that an entire function may
have two totally ramified values in C. There are also meromorphic functions with
four totally ramified values. They come from the theory of elliptic (doubly periodic)
functions.

13 Uniqueness theorems
Our first application of the second fundamental theorem are some uniqueness the-
orems.

Definition 13.1. Let f and g be meromorphic and a ∈ C. b Then f and g are said
−1 −1
to share the value a if f (a) = g (a); that is, for z ∈ C we have f (z) = a if and
only if g(z) = a.
If, in addition, for every a-point z0 of f and hence g the multiplicity of z0 as
an a-point of f is the same as the multiplicity as an a-point of g, then f and g are
said to share the value a with multiplicity.

Example. (1) Let

1
f (z) = ez and g(z) = e−z = .
f (z)
Then f and g share 0, ∞, 1 and −1 with multiplicity.

(2) Let
ez + 1 (ez + 1)2
f (z) = and g(z) =
(ez − 1)2 8(ez − 1)
Then f and g share 0, ∞, 1 and − 81 , but without multiplicity. This is clear
for the values 0 and ∞. Moreover, we have

f (z) = 1 ⇔ ez + 1 = (ez − 1)2 ⇔ ez = e2z − 2ez ⇔ ez = 3

and
g(z) = 1 ⇔ (ez + 1)2 = 8(ez − 1) ⇔ e2z + 2ez + 1 = 8ez − 8
⇔ e2z − 6ez + 9 = 0 ⇔ (ez − 3)2 = 0 ⇔ ez = 3.

Thus f and g share 1. An analogous argument shows that they also share − 81 .
13 UNIQUENESS THEOREMS 65

Theorem 13.1. Let f and g be meromorphic and non-constant. If f and g share

5 values, then f = g.
Proof. Let a1 . . . , a5 be the values shared by f and g. Without loss of generality
we may assume that a1 . . . , a5 ∈ C, since otherwise we can consider 1/(f − c) and
1/(g − c) instead of f and g for a suitable c ∈ C.
Put h = f − g. We have to show that h = 0. Suppose that this is not the case.
Picard’s theorem says that f has an aj -point for some value of j. This aj -point is
a zero of h. Thus h is non-constant.
For j ∈ {1, . . . , 5} we put
 
1
N (r, aj ) = N r, .
f − aj
Since f and g share the values a1 . . . , a5 , we also have
 
1
N (r, aj ) = N r, .
g − aj
The first fundamental theorem yields that
5  
X 1
N (r, aj ) ≤ N r, ≤ T (r, h) + O(1) ≤ T (r, f ) + T (r, g) + O(1).
j=1
h

On the other hand, by the second fundamental theorem (more precisely, Theo-
rem 12.3) we have
5
X
3T (r, f ) ≤ N (r, aj ) + S(r, f )
j=1

with S(r, f ) as in Theorem 12.1. Thus

 X5
1
T (r, f ) ≤ + o(1) N (r, aj ) for r ∈
/ Ef ,
3 j=1

with a subset Ef of [0, ∞) of finite measure. Analogously,

 X 5
1
T (r, g) ≤ + o(1) N (r, aj ) for r ∈
/ Eg ,
3 j=1

with a subset Eg of [0, ∞) of finite measure. Combining the above estimates we

obtain
 X 5
2
T (r, f ) + T (r, g) ≤ + o(1) N (r, aj )
3 j=1
 
2
≤ + o(1) (T (r, f ) + T (r, g)) + O(1)
3
for r ∈
/ Ef ∪ Eg . This is a contradiction.
66 13 UNIQUENESS THEOREMS

Remark. (1) The examples given before Theorem 13.1 show that the number 5
in Theorem 13.1 cannot be replaced by 4.
(2) If f and g share 4 values a1 , . . . , a4 , the arguments in the proof of Theo-
rem 13.1 yield that
4
X
N (r, aj ) ≤ T (r, f ) + T (r, g) + O(1),
j=1

 X4
1
T (r, f ) ≤ + o(1) N (r, aj ) for r ∈
/ Eg
2 j=1

and  X4
1
T (r, g) ≤ + o(1) N (r, aj ) for r ∈
/ Eg .
2 j=1

With E = Ef ∪ Eg this yields that

T (r, f ) ∼ T (r, g) for r ∈

/E

as well as
4
X
N (r, aj ) ∼ 2T (r, f ) ∼ 2T (r, g) for r ∈
/ E.
j=1

Theorem 13.2. Let f and g be meromorphic and non-constant. If f and g share

4 values counting multiplicities, then there exists a Möbius transformation M such
that f = M ◦ g.
Proof. Without loss of generality we may assume that 0, 1 and ∞ are among the
values shared by f and g. Let c be the fourth value shared by f and g.
Let Ef and Eg be the exceptional sets arising from the second fundamental
theorem, (which in turn arise from the lemma of the logarithmic derivative), applied
to f and g, and put E = Ef ∪ Eg .
We note here that the exceptional in the second fundamental theorem does
not depend on the choice of the values a1 , . . . , aq . To see this recall that the
exceptional set arises from the application of Borel’s lemma (Lemma 6.1) in the
proof of the lemma on the logarithmic derivative. More precisely, this lemma was
used to estimate the term T (R, f ) occurring in Theorem 11.4 by KT (r, f ) with
a constant K. In the proof of the second fundamental theorem we also have to
estimate the terms m(r, f 0 /(f − aj )). Theorem 11.4 gives an upper bound in terms
of T (R, f − aj ). However, we have T (R, f − aj ) ≤ T (R, f ) + log+ |aj | + log 2 by
Theorem 5.1, (ii). This gives an estimate of m(r, f 0 /(f − aj )) in terms of T (R, f ).
Hence we obtain an estimate of m(r, f 0 /(f − aj )) in terms of T (r, f ) outside the
same exceptional set as before.
Thus, by the second fundamental theorem, there exist at most two values a ∈ C b
such that
N (r, a) = o(T (r, f )) for r ∈
/ E.
13 UNIQUENESS THEOREMS 67

So at least two of the shared values do not have this property, and we may assume
without loss of generality that this is the case for 0 and ∞; that is,
 
1
N r,
f N (r, f )
lim sup > 0 and lim sup >0
r→∞ T (r, f ) r→∞ T (r, f )
r∈E
/ r∈E
/

We consider the auxiliary function

f0 g0
H := − .
f (f − 1)(f − c) g(g − 1)(g − c)
We want to show that H = 0 and thus assume that this is not the case. We have
the partial fraction decomposition
f0 f0 f0 f0
=α +β +γ ,
f (f − 1)(f − c) f f −1 f −c

with α = 1/c, β = 1/(1 − c) and γ = 1/(c(c − 1)). The lemma on the logarithmic
derivative (Theorem 11.5) implies that

f0
 
m = o(T (r, f )) for r ∈
/ E.
f (f − 1)(f − c)
Analogously we have
g0
 
m = o(T (r, g)) for r ∈
/ E.
g(g − 1)(g − c)
As remarked before Theorem 13.2, we have

T (r, f ) ∼ T (r, g) for r ∈

/ E.

Altogether we thus obtain

m(r, H) = o(T (r, f )) for r ∈

/ E.

Next we show that H is entire. In order to do this we note that poles of H can
only occur at points where f and g take one of the shared values 0, 1, c and ∞.
Using the partial fraction decomposition mentioned above we have
 0
g0
 0
g0
 0
g0
  
f f f
H=α − +β − +γ − ,
f g f −1 g−1 f −c g−c

If z0 is a zero of f and hence g of multiplicity p, then f 0 /f and g 0 /g have simple

poles with residue p at z0 . This implies that H is holomorphic at z0 . The same
argument shows that H is holomorphic at the 1-points and c-points of H.
Suppose now that z0 is a pole of multiplicity p. Then f 0 has a pole of multiplicity
p + 1 at z0 while f (f − 1)(f − c) has a pole of multiplicity 3p at z0 . Thus the
68 13 UNIQUENESS THEOREMS

function f 0 /(f (f − 1)(f − c)) has a zero (of multiplicity 2p − 1) at z0 , and so does
g 0 /(g(g − 1)(g − c)). It follows that H has a zero at z0 . In particular, H is entire
as claimed.
Moreover, the argument shows that
 
1
N (r, f ) ≤ N r, .
H
Together with the first fundamental theorem we thus have
N (r, f ) ≤ T (r, H) + O(1).
Since H is entire we have N (r, H) = 0 and thus the estimate for m(r, H) obtained
above yields
T (r, H) = m(r, H) = o(T (r, f )) for r ∈
/ E.
Combining the last two equations we have
N (r, f ) = o(T (r, f )) for r ∈
/ E.
However, we assumed that
N (r, f )
lim sup > 0.
r→∞ T (r, f )
r∈E
/

This is a contradiction. Thus H = 0.

Next we define a second auxiliary function
f 0f g0g
K := − .
(f − 1)(f − c) (g − 1)(g − c)
Arguments analogous to the ones used above show that K is entire and, assuming
K 6= 0,    
1 1
N r, ≤ N r, ≤ T (r, K) + O(1)
f K
as well as
T (r, K) = m(r, K) = o(T (r, f )) for r ∈
/ E.
This contradicts the assumption that
 
1
N r,
f
lim sup > 0.
r→∞ T (r, f )
r∈E
/

Thus also K = 0. Combining this with the equation H = 0 obtained already we

find that
f 0f 2 g0g2
0=K= −
f (f − 1)(f − c) g(g − 1)(g − c)
f0 2 g0
= f − g2
f (f − 1)(f − c) g(g − 1)(g − c)
0
f
= (f 2 − g 2 )
f (f − 1)(f − c)
2 2
and thus f = g . Hence f = g or f = −g.
14 AN APPLICATION TO ITERATION THEORY 69

Remark. The second example after Definition 13.1 shows that the conclusion of
Theorem 13.2 need not hold if f and g share the values without multiplicities.
However, a result of Gundersen (1983) says that it suffices to assume that 2 of the
4 values are shared counting multiplicity. It is an open question whether this also
holds if only one of the 4 values is shared counting multiplicity.

14 An application to iteration theory

To motivate the application of Nevanlinna theory we are concerned with, we first
give the following application of Picard’s theorem.

Theorem 14.1. Let f be an entire function which is not if the form f (z) = z + c
for some c ∈ C. Then f ◦ f has a fixed point; that is, there exists z0 ∈ C with
f (f (z0 )) = z0 .

Proof. Suppose that f ◦ f has no fixed point. Then f has no fixed point. Thus
f (z) 6= z for all z ∈ C and this implies that f (f (z)) 6= f (z) for all z ∈ C. We
deduce that the function h defined by

f (f (z)) − z
h(z) =
f (z) − z

is entire and satisfies f (z) 6= 0 and f (z) 6= 1 for all z ∈ C. By Picard’s theorem, h
is constant, say h(z) = c for all z ∈ C, with c ∈ C \{0, 1}. Hence

f (f (z)) − z = c(f (z) − z)

for all z ∈ C. Differentiating this we obtain

f 0 (f (z))f 0 (z) − 1 = c(f 0 (z) − 1)

and thus
(f 0 (f (z)) − c)f 0 (z) = 1 − c.
Since c 6= 1 this yields that f 0 (z) 6= 0 for all z ∈ C. This implies that f 0 (f (z)) 6= 0
for all z ∈ C. Moreover, the last equation implies that f 0 (f (z)) 6= c for all z ∈ C.
Picard’s theorem implies that f 0 ◦ f is constant. Hence f 0 is constant. (Note that
the assumption that f has no fixed point implies that f is non-constant.) Thus
f has the form f (z) = az + b with a, b ∈ C. The hypothesis that f has no fixed
points no implies that a = 1.

Remark. If f is a transcendental entire function, one would expect that f ◦ f has

infinitely many fixed points. This is indeed the case, but it does not follow from
Picard’s theorem with the above method of proof. The point is that even if f has
only one fixed point a, the equation f (f (z)) = f (z) will in general have infinitely
solutions and thus the function h considered in the above proof will have infinitely
many 1-points. In fact, every a-point of f is a 1-point of h.
70 14 AN APPLICATION TO ITERATION THEORY

However, Nevanlinna theory can be used to show that h has “few” 1-points;
that is, N (r, 1/h) = o(T (r, h)). The proof can then be completed as before.
We will actually prove a more general result involving the iterates f n defined
by f 1 = f and f n = f ◦ f n−1 for n ≥ 2.

Theorem 14.2. Let f be a transcendental meromorphic function and let g be

entire. Then there exists a subset E of [0, ∞) of finite measure such that

T (r, f ◦ g)
lim
r→∞
=∞
r∈E
/
T (r, g)

with some subset E of [0, ∞) of finite measure.

Corollary. Let f be a transcendental entire function and m, n ∈ N with n > m.

Then
T (r, f n )
lim
r→∞ T (r, f m )
=∞
r∈E
/

Proof of Theorem 14.2. Let b ∈ C be such that f has infinitely many b-points
a1 , a2 , a3 , . . . . Then g(z) = aj implies that f (g(z)) = b. For q ∈ N and r ≥ 1 we
thus obtain   X q  
1 1
N r, ≥ N r, .
f ◦g−b j=1
g − a j

(In fact, this also holds with N (r, ·) replaced by N (r, ·), but we will not need this.)
By the second fundamental theorem we have
q  
X 1
N r, ≥ (q − 2 − o(1))T (r, g) for r ∈
/ E,
j=1
g − aj

with a subset E of [0, ∞) of finite measure depending only on g. By the first

fundamental theorem we have
 
1
N r, ≤ T (r, f ◦ g) + O(1).
f ◦g−b

Combining the last two inequalities we obtain

T (r, f ◦ g) ≥ (q − 2 − o(1))T (r, g) for r ∈

/E

so that
T (r, f ◦ g)
lim inf
r→∞
≥ q − 2.
r∈E
/
T (r, g)
Since q ∈ N was arbitrary, the conclusion follows.

Remark. The conclusion of Theorem 14.2 and the Corollary actually holds without
any exceptional set E.
14 AN APPLICATION TO ITERATION THEORY 71

Definition 14.1. Let f be entire, n ∈ N and z0 ∈ C. Then z0 is called a periodic

point of period n if f n (z0 ) = z0 and f k (z0 ) 6= z0 for 1 ≤ k < n.

The following result is due to Baker (1960).

Theorem 14.3. Let f be a transcendental entire function. Then there exists at

most one m ∈ N such that f has only finitely many periodic points of period m.

Proof. Suppose that there are two integers m, n ∈ N such that f has only finitely
many periodic points of periods m and n. Without loss of generality we may assume
that m < n. Let a1 , . . . , ap be the periodic points of period m and b1 , . . . , bq be the
periodic points of period n.
We put ` = n − m and consider the function h defined by

f n (z) − z
h(z) = .
f ` (z) − z

As in the proof of Theorem 14.1 we consider the zeros, 1-points and poles of h.
First we note that by the first fundamental theorem and Theorem 14.2 we have
 
1
N (r, h) ≤ N r, ` ≤ T (r, f ` (z) − z) + O(1)
f (z) − z
≤ T (r, f ` ) + O(log r)
= o(T (r, f n ))

as r → ∞, r ∈ / E. Here E is again a subset of [0, ∞) of finite measure.

Let z0 ∈ C with h(z0 ) = 0. Then f n (z0 ) = z0 . Thus z0 = bj for some
j ∈ {1, . . . , q} or there exists k ∈ N with 1 ≤ k < n such that f k (z0 ) = z0 . It
follows that
  n−1  
1 X 1
N r, ≤ q log r + N r, k + O(1).
h k=1
f (z) − z

By the first fundamental theorem and Theorem 14.2 we thus have

  n−1
1 X
N r, ≤ q log r + T (r, f k (z) − z) + O(1) = o(T (r, f n ))
h k=1

as r → ∞, r ∈ / E.
Let now z0 ∈ C with h(z0 ) = 1. Then f ` (z0 ) = f n (z0 ) = f m (f ` (z0 )) so that
f ` (z0 ) is a fixed point of f m . Thus f ` (z0 ) = aj for some j ∈ {1, . . . , p} or there
exists k ∈ N with 1 ≤ k < m such that f k+` (z0 ) = f k (f ` (z0 )) = f ` (z0 ). This
implies that
  p   m−1  
1 X 1 X 1
N r, ≤ N r, `
+ N r, k+` .
h−1 j=1
f (z) − aj k=1
f (z) − f ` (z)
72 14 AN APPLICATION TO ITERATION THEORY

The first fundamental theorem yields that

  m−1
1 `
X
N r, ≤ p · T (r, f ) + (T (r, f k+` ) + T (r, f ` )) + O(log r)
h−1 k=1

and as before we deduce from Theorem 14.2 that

 
1
N r, = o(T (r, f n ))
h−1
as r → ∞, r ∈ / E.
Altogether we thus have
   
n 1 n 1
N (r, h) = o(T (r, f )), N r, = o(T (r, f )) and N r, = o(T (r, f n ))
h h−1
as r → ∞, r ∈ / E.
Using Theorem 14.2 once more it is not difficult to deduce from the definition of
h that T (r, h) ∼ T (r, f n ) as r → ∞, r ∈
/ E. Together with the previous equations
this is a contradiction to the second fundamental theorem.
Remark. The example f (z) = ez + z shows that an entire functions need not have
fixed points. In other words, there need not be periodic points in period 1. Thus
m = 1 may arise as an exception in Theorem 14.2. One can show that this is
the only exception; that is, for n ≥ 2 there are infinitely many periodic points of
period n.
The second fundamental theorem concerns the a-points of a meromorphic func-
tion f ; that is, the zeros of f −a. In the above proof we used the second fundamental
theorem to obtain conclusions about the zeros of f (z) − z. More generally, one may
also consider the zeros of f (z) − a(z) for certain functions a.
Theorem 14.4. Let f be meromorphic and let a1 , a2 , aq be meromorphic functions
satisfying T (r, aj ) = o(T (r, f )) as r → ∞, for j = 1, 2, 3. Then
3  
X 1
T (r, f ) ≤ N r, + o(T (r, f ))
j=1
f − a j

as r → ∞, r ∈
/ E, for some subset E of [0, ∞) of finite measure.
Proof. We consider the meromorphic function h defined by
a3 (z) − a2 (z) f (z) − a1 (z)
h(z) = · .
a3 (z) − a1 (z) f (z) − a2 (z)
The hypothesis that T (r, aj ) = o(T (r, f )) and the first fundamental theorem yield
that
     
1 1 1
N r, ≤ N r, + N r, + N (r, a1 ) + N (r, a2 ) + N (r, a3 )
h f − a1 a3 − a1
 
1
≤ N r, + o(T (r, f )).
f − a1
15 DIFFERENTIAL EQUATIONS 73

The same reasoning shows that

 
1
N (r, h) ≤ N r, + o(T (r, f ))
f − a2
and    
1 1
N r, ≤ N r, + o(T (r, f )).
h−1 f − a3
Finally, using T (r, aj ) = o(T (r, f )) it is not difficult to see that

T (r, h) ∼ T (r, f ).

The conclusion now follows from the second fundamental theorem applied to h.
Remark. Theorem 14.4 has been extended to the case of q functions a1 , . . . , aq
satisfying T (r, aj ) = o(T (r, f )) by Yamanoi in 2004: given ε > 0 we have
q  
X 1
(q − 2 − ε)T (r, f ) ≤ N r,
j=1
f − aj

for r outside a set of finite measure. With N (r, ·) replaced by N (r, ·) this had been
proved before by Steinmetz in 1986.

15 Differential equations in the complex domain

We will give some application of Nevanlinna theory to the theory of differential
equations in the complex domain. In order to do so, we need to compare the
Nevanlinna characteristic and the other quantities of Nevanlinna theory of a mero-
morphic function with those of the derivatives.
Theorem 15.1. Let f be meromorphic and k ∈ N. Then, for some subset E of
[0, ∞) of finite measure,
(i) N (r, f (k) ) = N (r, f ) + kN (r, f ) ≤ (k + 1)N (r, f ),
(ii) m(r, f (k) ) ≤ m(r, f ) + o(T (r, f )) for r ∈
/ E,
(iii) T (r, f (k) ) ≤ (k + 1)T (r, f ) + o(T (r, f )) for r ∈
/ E.
Proof. Conclusion (i) follows immediately from the definition of N (r, ·) and N (r, ·).
We will prove (ii) and (iii) by induction. The case k = 0 is trivial. Assume
now that (ii) and (iii) hold for some k ≥ 0. Noting that
 (k+1)   (k+1) 
(k+1) f (k) f
m(r, f )=m (k)
f ≤m + m(r, f (k) ) + O(1)
f f (k)
we deduce from the lemma on the logarithmic derivative that

m(r, f (k+1) ) ≤ m(r, f (k) ) + o(T (r, f (k) ))

74 15 DIFFERENTIAL EQUATIONS

for r ∈ / E. Since by induction hypothesis (iii) holds for k, we deduce that

T (r, f (k) ) = O(T (r, f )) for r ∈
/ E and thus

m(r, f (k+1) ) ≤ m(r, f (k) ) + o(T (r, f ))

for r ∈
/ E. Together with the assumption that (ii) holds for k we deduce that (ii)
also holds with k replaced by k + 1. Together with (i) this implies that (iii) also
holds with k replaced by k + 1.
As a simple but instructive example of how Nevanlinna theory is applied to
differential equations is the following result.
Theorem 15.2. Let f be meromorphic and P a polynomial. If f satisfies the
differential equation f 0 = P (f ), then deg(P ) ≤ 2.
Proof. With d = deg(P ) we deduce from Theorems 5.3 and 15.1 that

d · T (r, f ) = T (r, P ◦ f ) + O(1) = T (r, f 0 ) + O(1) ≤ 2T (r, f ) + o(T (r, f ))

as r → ∞, r ∈
/ E. This implies that d ≤ 2.
Remark. A solution of f 0 = 1 + f 2 is given by f (z) = tan z.
Of course, the differential equation f 0 = P (f ) can be integrated directly:
Z z Z z Z f (z)
f 0 (ζ) dw
ζ= ζ= .
z0 z0 P (f (ζ)) f (z0 ) P (w)

Above we considered the differential

Pd equation f 0 = P (f ) with a polynomial P .
Thus P has the form P (w) = k=0 ak wk with ak ∈ C. Now we will study the more
general case where the ak are rational functions. First we extend Theorem 5.3 to
this context.
Theorem 15.3. Let f be meromorphic and
d
X
P (z, w) = ak (z)wk ,
k=0

where the ak are rational functions and ad 6= 0. Then

T (r, P (z, f (z))) = d · T (r, f ) + O(log r).

Proof. Choose r0 > 0 such that all zeros and poles of the coefficients a0 , . . . , ad are
contained in D(0, r0 ). Then a pole z0 of f of order m is a pole of P (z, f (z)) of
order dm. This implies that there exists M ∈ N such that

n(r, P (z, f (z))) = d · n(r, f ) + M

for r ≥ r0 . Hence

N (r, P (z, f (z))) = d · N (r, f ) + O(log r).

15 DIFFERENTIAL EQUATIONS 75

To obtain a corresponding estimate for the proximity function we note that

there exist cj ∈ C \{0} and mj ∈ Z such that

aj (z) ∼ cj · z mj

as |z| → ∞. Hence
aj (z) cj
≤ (1 + o(1)) · |z|mj −md
ad (z) cd
as |z| → ∞. Thus there exist R > 0 and L > 0 such that

aj (z) 1
≤ |z|L and ≤ |ad (z)| ≤ |z|L
ad (z) |z|d

for |z| ≥ R. For r ≥ R we consider

J = θ ∈ [0, 2π] : |f (reiθ )| ≥ rL+1 .



For z = reiθ with θ ∈ J and r ≥ R we then have

d−1
!
X ak (z)
|P (z, f (z))| = ad (z)f (z)d 1 + f (z)k−d
k=0
ad (z)
d−1
!
d
X ak (z)
≥ |ad (z)| · |f (z)| 1 − |f (z)|k−d
k=0
a d (z)
d−1
!
X
≥ |z|−L |f (z)|d 1 − |z|L |z|(k−d)(L+1)
k=0
 
−L d d
≥ |z| |f (z)| 1 − .
|z|

For r ≥ max{R, 2d} and θ ∈ J we thus have

1
|P (reiθ , f (reiθ ))| ≥ r−L |f (reiθ )|d .
2
Hence
Z Z
1 + 1
iθ iθ
− log 2 − L log r + d log+ |f (reiθ )| dθ


log |P (re , f (re ))|dθ ≥
2π J 2π J
Z
1
=d· log+ |f (reiθ )|dθ − log 2 − L log r.
2π J
Similarly,
|P (reiθ , f (reiθ ))| ≤ 2rL |f (reiθ )|d .
for r ≥ max{R, 2d} and θ ∈ J and thus
Z Z
1 + 1
iθ iθ
log |P (re , f (re ))|dθ ≤ d · log+ |f (reiθ )|dθ + log 2 + L log r.
2π J 2π J
76 15 DIFFERENTIAL EQUATIONS

For r ≥ R and θ ∈ [0, 2π]\J we have |f (reiθ )| ≤ rL+1 and

d
iθ iθ
X ak (z)
|P (re , f (re ))| = ad (z) f (z)k
k=0
ad (z)
d
X
≤ |z|L |z|L |z|(L+1)k ≤ (d + 1)|z|2L+(L+1)d
k=0

so that Z
1
log+ |f (reiθ )|dθ ≤ (L + 1) log r
2π [0,2π]\J

and
Z
1
log+ |P (reiθ , f (reiθ ))|dθ ≤ (2L + (L + 1)d) log r + log(d + 1).
2π [0,2π]\J

Combining the last estimates we deduce that

Z
1
m(r, P (z, f (z))) = log+ |P (reiθ , f (reiθ ))|dθ + O(log r)
2π J
Z
1
=d· log+ |f (reiθ )|dθ O(log r)
2π J
= d · m(r, f ) + O(log r).

Together with the estimate for N (r, ·) obtained already this yields

T (r, P (z, f (z))) = d · T (r, f ) + O(log r).

Let
d
X
P (z, w) = ak (z)wk ,
k=0

be as in Theorem 15.3. We say that P (z, w) is a polynomial in w with rational

coefficients and call d the degree with respect to w. We denote it by degw (P ).
These polynomials form a ring. The units in this ring are the polynomials of
degree 0 (in w); that is, the rational functions (of z). The corresponding field of
fractions (German: Quotientenkörper) consists of functions R of the form
Pm
ak (z)wk
R(z, w) = Pk=0n k
k=0 bk (z)w

with rational functions a0 , . . . , am and b0 , . . . , bn , and bk 6= 0 for some k. We may

actually assume that the ak and bk are polynomials, as this can be achieved by
multiplying them with a common multiple of their denominators. In other words,
R is a rational function in two variables, which means that R is a quotient of
polynomials in two variables. If m and n in the above representation are chosen
minimal, which means that numerator and denominator do not have a common
15 DIFFERENTIAL EQUATIONS 77

factor (with respect to w), then d := max{m, n} is called the degree of R with
respect to w and denoted by degw (R).
In elementary number theory, Bezout’s lemma says that the greatest common
divisor gcd(p, q) of two integers p and q has a representation ap + bq = gcd(p, q)
with integers a and b. (The proof is obtained from the Euclidean algorithm.) In
particular, if p and q are coprime (German: teilerfremd), then there exist a and b
such that ap + bq = 1. This result actually holds not only for the ring of integers,
but in any principal ideal domain (German: Hauptidealring). In particular, this is
true for the ring of polynomials in one variable over some field. We shall use this
for the field of rational functions.
Lemma 15.1. Let P (z, w) and Q(z, w) be two coprime polynomials in w with
rational functions in z as coefficients. Then there exist polynomials A(z, w) and
B(z, w) in w with rational functions in z as coefficients such that
A(z, w)P (z, w) + B(z, w)Q(z, w) = 1.
Theorem 15.4. Let f be meromorphic and let R(z, w) be rational in both variables.
Then
T (r, R(z, f (z))) = degw (R) · T (r, f ) + O(log r).
Proof. We write
m n
P (z, w) X
k
X
R(z, w) = with P (z, w) = ak (z)w and Q(z, w) = bk (z)wk
Q(z, w) k=0 k=0

with rational functions a0 , . . . , am , b0 , . . . , bn , where am 6= 0 and bn 6= 0. Assuming

that P and Q are coprime we have degw (R) = max{m, n}. We first show that
T (r, R(z, f (z))) ≤ degw (R) · T (r, f ) + O(log r).
by induction on degw (R). The conclusion is clear if degw (R) = 0. Suppose now
that d ∈ N and that the conclusion holds if degw (R) ≤ d − 1. We may assume
that d = m > n. In fact, if m < n this can be achieved by considering 1/R(z, w)
instead of R(z, w), noting that
 
1
T r, = T (r, R(z, f (z))) + O(1).
R(z, f (z))
And if m = n = d, then we consider
  −1
ad (z)

Pd−1 k
 k=0 ak (z) − bd (z) bk (z) w 
 −1
ad (z)
R(z, w) − = Pd 
bd (z)  b
k=0 k (z)w k 

Pd
b (z)wk
=  k=0 k 
Pd−1 ad (z)
k=0 ak (z) − bk (z) wk
bd (z)
78 15 DIFFERENTIAL EQUATIONS

instead of R(z, w), noting that

 −1 !
ad (z)
T r, R(z, f (z)) − = T (r, R(z, f (z))) + O(log r).
bd (z)

Thus we may assume that d = m > n. By long division (German: Polynomdivi-

sion) we find that R(z, w) has the form
m−n P`
X
k dk (z)wk
R(z, w) = ck (z)w + Pk=0
n k
k=0 k=0 bk (z)w

where 0 ≤ ` < n and c0 , . . . , cm−n , d0 , . . . , d` are rational functions. Thus

R(z, f (z)) = α(z) + β(z)

where
m−n P`
X
k dk (z)f (z)k
α(z) = ck (z)f (z) and β(z) = Pk=0
n k
.
k=0 k=0 bk (z)f (z)

By Theorem 15.3 we have T (r, α) = (m − n)T (r, f ) + O(log r) and by the induction
hypothesis we have T (r, β) ≤ n · T (r, f ) + O(log r). Combining the two estimates
we obtain the desired upper bound for T (r, R(z, f (z))).
Now we turn to the proof of the lower bound for T (r, R(z, f (z))); that is, we
prove that
T (r, R(z, f (z)) ≥ degw (R) · T (r, f ) + O(log r).
Again we may assume that m > n. Since the case n = 0 is covered by Theorem 15.3
we may thus assume that 1 ≤ n < m. Choose A and B according to Lemma 15.1
so that
A(z, w)P (z, w) + B(z, w)Q(z, w) = 1.
With s = degw (A) and t = degw (B) we have s + m = t + n. Since m > n this
yields t > s. We put C = A/B and write A∗ (z) = A(z, f (z)), B ∗ (z, f (z)), etc. In
particular, R∗ (z) = R(z, f (z)). Then

∗ 1 A∗ Q∗ 1
C + ∗ = ∗+ ∗ = ∗ ∗
R B P B Q

and hence
   
∗ 1 1
T r, C + ∗ = T r, ∗ ∗
R B Q
= T (r, B ∗ Q∗ ) + O(1)
= degw (BQ) log r + O(log r)
= (t + m)T (r, f ) + O(log r)
15 DIFFERENTIAL EQUATIONS 79

by the first fundamental theorem and Theorem 15.3. Moreover,

 
1
∗
T r, C + ∗ ≤ T (r, C ∗ ) + T (r, R∗ ) + O(1)
R
≤ degw (C) · T (r, f ) + T (r, R∗ ) + O(log r)
= t · T (r, f ) + T (r, R∗ ) + O(log r)

by the upper bound proved already. Combining the last two estimates we have
 
1
(t + m)T (r, f ) = T r, C + ∗ + O(log r) ≤ t · T (r, f ) + T (r, R∗ ) + O(log r)
∗
R
and hence

degw (R) · T (r, f ) = m · T (r, f ) ≤ T (r, R∗ ) + O(log r) = T (r, R(z, f (z))) + O(log r)

as claimed.
Theorem 15.5 (Malmquist’s theorem). Let f be a transcendental meromorphic
function and let R(z, w) be rational in both variables. Suppose that f satisfies the
differential equation
f 0 (z) = R(z, f (z)).
Then R is a polynomial in w and degw (R) ≤ 2.
Proof. By Theorem 15.1 we have

T (r, f 0 ) ≤ 2T (r, f ) + o(T (r, f ))

for r ∈
/ E while Theorem 15.4 says that

T (r, R(z, f (z))) = degw (R) · T (r, f ) + O(log r).

Together these results imply that degw (R) ≤ 2.

For c ∈ C we consider h = 1/(f − c). Then
f0
h0 = − = −h2 f 0 .
(f − c)2
Since f (z) = 1/h(z) + c we find that h satisfies the differential equation
 
0 2 1
h (z) = −h(z) R z, +c .
h(z)
Applying the result we have proved already we deduce that the function S given
by  
2 1
S(z, w) = w R z, + c
w
satisfies degw (S) ≤ 2. From this it is not difficult to deduce that R is a polynomial
in w.
80 15 DIFFERENTIAL EQUATIONS

The differential equation

f 0 (z) = a0 (z) + a1 (z)f (z) + a2 (z)f (z)2

is called Riccati differential equation. If a0 , a1 and a2 are polynomials, then all

solutions of this equation are meromorphic.
We only sketch the proof. Consider the following linear system of differential
equations:

u0 = a1 u + a0 v
v 0 = −a2 u

Standard techniques show that for every linear system of differential equations
with entire coefficients the solutions are also entire. Thus all solutions u and v
of the above system are entire. The quotient f = u/v of two solutions is thus
meromorphic, and we have

0
 u 0 u0 v − uv 0 (a1 u + a0 v)v − u(−a2 u)
f = = 2
= = a0 + a1 f + a2 f 2 .
v v v2
The following result is proved in the same way as the previous theorem.
Theorem 15.6 (Malmquist-Yosida theorem). Let f be a transcendental meromor-
phic function, R(z, w) be rational in both variables and n ∈ N. Suppose that f
satisfies the differential equation

f 0 (z)n = R(z, f (z)).

Then R is a polynomial in w and degw (R) ≤ 2n.

Remark. Actually there are much stronger restrictions on R than just degw (R) ≤
2n It was shown by Steinmetz as well as Bank and Kaufman in 1980 that the
differential equation considered can be reduced by linear transformations to one of
the following equations (or a power thereof):

f 0 (z) = a(z) + b(z)f (z) + c(z)f (z)2

f 0 (z)2 = a(z)(f (z) − b(z))2 (f (z) − τ1 )(f (z) − τ2 )
f 0 (z)2 = a(z)(f (z) − τ1 )(f (z) − τ2 )(f (z) − τ3 )(f (z) − τ4 )
f 0 (z)3 = a(z)(f (z) − τ1 )2 (f (z) − τ2 )2 (f (z) − τ3 )2
f 0 (z)4 = a(z)(f (z) − τ1 )2 (f (z) − τ2 )3 (f (z) − τ3 )3
f 0 (z)6 = a(z)(f (z) − τ1 )3 (f (z) − τ2 )4 (f (z) − τ3 )5 .

There the τj are distinct constants and a, b and c are rational functions.
Another generalization of Theorem 15.6 due to Eremenko says that if

f 0 (z)m Qm (z, f (z)) + · · · + f 0 (z)Q1 (z, f (z)) + Q0 (z, f (z)) = 0

with polynomials Qj (z, w) in w, then degw (Qj ) ≤ 2(m − j).

16 EXCEPTIONAL VALUES OF DERIVATIVES 81

16 Exceptional values of derivatives

Picard’s theorem says that if an entire function f satisfies f (z) 6= 0 and f (z) 6= 1
for all z ∈ C, then f is constant. Saxer showed in 1923 that this also holds if
f (z) 6= 0 and f 0 (z) 6= 1 for all z ∈ C. We will consider some more general results.
In fact, Saxer’s theorem is an immediate consequence of the following theorem.

Theorem 16.1. Let f be a transcendental meromorphic function and k ∈ N0 . Let

N0 (r) denote the counting function of the zeros of f (k+1) that are not 1-points of f .
Then
   
1 1
T (r, f ) ≤ N (r, f ) + N r, + N r, (k) − N0 (r) + o(T (r, f ))
f f −1

as r → ∞, r ∈
/ E, for some subset E of [0, ∞) of finite measure.

Proof. The second fundamental theorem, applied to g := f (k) , yields that

   
1 1
m(r, g) + m r, + m r, ≤ 2T (r, g) − N1 (r, g) + o(T (r, g))
g g−1

as r → ∞, r ∈
/ E, with
 
1
N1 (r, g) = N r, 0 + 2N (r, g) − N (r, g 0 ).
g

Here and in the following E always denotes a subset of [0, ∞) of finite measure.
The first fundamental theorem implies that
   
1 1
2T (r, g) − N1 (r, g) = m(r, g) + N (r, g) + m r, + N r,
g−1 g−1
   
1
− N r, 0 + 2N (r, g) − N (r, g 0 ) + O(1)
g
     
1 1 1
= m(r, g) + m r, + N r, − N r, 0
g−1 g−1 g
0
+ N (r, g ) − N (r, g) + O(1).

Since      
1 1 1
N r, − N r, 0 = N r, − N0 (r)
g−1 g g−1
and
N (r, g 0 ) − N (r, g) = N (r, g)
this yields
   
1 1
2T (r, g)−N1 (r, g) = m(r, g)+m r, +N r, −N0 (r)+N (r, g)+O(1).
g−1 g−1
82 16 EXCEPTIONAL VALUES OF DERIVATIVES

Together with the inequality obtained at the beginning of the proof this yields that
   
1 1
m r, ≤ N r, − N0 (r) + N (r, g) + o(T (r, g))
g g−1

as r → ∞, r ∈ / E.
By Theorem 15.1 we have T (r, g) = T (r, f (k) ) ≤ (k + 1)T (r, f ) as r → ∞,
r∈/ E. This implies that the o(T (r, g))-terms occurring above can be replaced by
o(T (r, f )). The lemma on the logarithmic derivative implies that

f (k)
   
1 1
m r, = m r, (k) · (k−1)
f (k−1) f f
f (k)
   
1
≤ m r, (k) + m r, (k−1)
f f
 
1
≤ m r, (k) + o(T (r, f (k−1) )
f
 
1
= m r, (k) + o(T (r, f )
f

and induction shows that

     
1 1 1
m r, ≤ m r, (k) + o(T (r, f ) = m r, + o(T (r, f )
f f g

as r → ∞, r ∈
/ E. Thus
   
1 1
m r, ≤ N r, + N (r, g) − N0 (r) + o(T (r, g))
f g−1

and hence
   
1 1
T (r, f ) = N r, + m r,
f f
   
1 1
≤ N r, + N r, + N (r, g) − N0 (r) + o(T (r, g))
f g−1
   
1 1
= N r, + N r, (k) + N (r, f ) − N0 (r) + o(T (r, f ))
f f −1

as r → ∞, r ∈
/ E.

Remark. The case k = 0 is a direct consequence of the second fundamental theorem.

The theorem of Saxer mentioned above follows from the case k = 1.
Theorem 16.1 implies that if f is meromorphic with only finitely many zeros
and poles such that f (k) has only finitely many 1-points for some k ∈ N, then
f is rational. To see this, we only have to note that Theorem 16.1 yields that
T (r, f ) = O(log r) under these hypotheses.
16 EXCEPTIONAL VALUES OF DERIVATIVES 83

One may replace the 1-points of f (k) by the a-points of f (k) for any a ∈ C \{0}.
This follows by considering f /a instead of f . However, one may not replace the
1-points of f (k) by the zeros of f (k) . In fact, f (z) = ez has no zeros and poles, and
f (k) (z) 6= 0 for all z ∈ C and all k ∈ N. And for f (z) = ez + 1 we have f (z) 6= 1
and f (k) (z) 6= 0 for all z ∈ C and all k ∈ N.
We will now prove a result of Hayman which implies that in order to conclude
that a meromorphic function f is rational it suffices to assume that f has only
finitely many zeros and f (k) has only finitely many 1-points for some k ∈ N. So no
hypothesis on the poles is required.
Theorem 16.2. Let f be a transcendental meromorphic function and k ∈ N. Then
       
1 1 2 1
T (r, f ) ≤ 2 + N r, + 2+ N r, (k) + o(T (r, f ))
k f k f −1
as r → ∞, r ∈
/ E.
The idea is to estimate the term N (r, f ) occurring on the right hand side of
the inequality in Theorem 16.1 in terms of the other counting functions occuring
there. In order to do so we write

N (r, f ) = N s (r, f ) + N m (r, f ),

where N s (r, f ) denotes the counting function of the simple poles and N m (r, f )
denotes the counting function of the multiple poles, counted without multiplicity.
First we prove the following lemma.
Lemma 16.1. Let f be a transcendental meromorphic function and let N0 (r) be
as in Theorem 16.1. Then
 
1
kN s (r, f ) ≤ N m (r, f ) + N r, (k) + N0 (r) + o(T (r, f ))
f −1
as r → ∞, r ∈
/ E.
Proof. Let g = f (k) be as in the proof of Theorem 16.1 and put

k+1
g 0 (z)k+1 f (k+1) (z)
h(z) = = k+2
.
(1 − g(z))k+2 (1 − f (k) (z))

Let z0 be a simple pole of f . Then f (k) and hence 1 − f (k) have a pole of order
k + 1 at z0 . In fact,
a a
1 − f (k) (z) = 1 + O (z − z0 )k+1



k+1
+ O(1) = k+1
(z − z0 ) (z − z0 )
as z → z0 for some non-zero constant a (depending on z0 ). It follows that
(k + 1)a (k + 1)a
f (k+1) (z) = k+2



+ O(1) = 1 + O (z − z0 )
(z − z0 )k+2 (z − z0 )k+2
84 16 EXCEPTIONAL VALUES OF DERIVATIVES

as z → z0 . Hence

(k + 1)k+1 ak+1 (z − z0 )(k+1)(k+2) k+1

h(z) = · 1 + O (z − z0 )
(z − z0 )(k+2)(k+1) ak+2
= b + O (z − z0 )k+1

as z → z0 , with b = (k + 1)k+1 /a ∈ C \{0}. Thus h(z0 ) = b 6= 0 and h(j) (z0 ) = 0

for 1 ≤ j ≤ k. Hence h/h0 has a pole of multiplicity at least k at z0 . It follows that
 
h
kN s (r, f ) ≤ N r, 0
h
 
h
≤ T r, 0
h
h0
 
= T r, + O(1)
h
h0 h0
   
= N r, + m r, + O(1)
h h
h0
 
≤ N r, + o(T (r, h))
h
 
1
= N (r, h) + N r, + o(T (r, h))
h
 
1
= N (r, h) + N r, + o(T (r, f ))
h

as r → ∞, r ∈ / E.
Next we note that if z0 is a pole of f of multiplicity p ≥ 2, then h has a zero of
multiplicity (p + k)(k + 2) − (p + k + 1)(k + 1) = p − 1 at z0 . Further zeros of h
can arise only from zeros of f (k+1) which are not 1-points of f (k) so that
 
1
N r, ≤ N m (r, f ) + N0 (r).
h

Poles of h can arise only from zeros of f (k) − 1 so that

 
1
N (r, h) ≤ N r, (k) .
f −1

Combining the last three inequalities we obtain the conclusion.

Proof of Theorem 16.2. We have

N s (r, f ) + 2N m (r, f ) ≤ N (r, f )

and thus, together with the first fundamental theorem,

N m (r, f ) ≤ N (r, f )− N s (r, f ) + N m (r, f ) = N (r, f )−N (r, f ) ≤ T (r, f )−N (r, f ).
16 EXCEPTIONAL VALUES OF DERIVATIVES 85

Theorem 16.1 says that

   
1 1
T (r, f ) ≤ N (r, f ) + N r, + N r, (k) − N0 (r) + o(T (r, f ))
f f −1
as r → ∞, r ∈
/ E. Together these inequalities yield that
   
1 1
N m (r, f ) ≤ N r, + N r, (k) − N0 (r) + o(T (r, f ))
f f −1
as r → ∞, r ∈
/ E. By Lemma 16.1 we have
 
1
kN s (r, f ) ≤ N m (r, f ) + N r, (k) + N0 (r) + o(T (r, f ))
f −1
as r → ∞, r ∈
/ E. Combining these two estimates we obtain
   
1 1
kN s (r, f ) ≤ N r, + 2N r, (k) + o(T (r, f ))
f f −1
and thus    
1 1 2 1
N s (r, f ) ≤ N r, + N r, (k) + o(T (r, f ))
k f k f −1
as r → ∞, r ∈
/ E. Adding the estimates for N m (r, f ) and N s (r, f ) yields that
N (r, f ) = N m (r, f ) + N s (r, f )
       
1 1 2 1
≤ 1+ N r, + 1+ N r, (k) + o(T (r, f ))
k f k f −1
as r → ∞, r ∈
/ E. Inserting this in the inequality of Theorem 16.1 yields the
conclusion.
Corollary. Let f be a meromorphic function and k ∈ N. Suppose that f (z) 6= 0
and f (k) (z) 6= 1 for all z ∈ C. Then f is constant.
To prove the corollary, we note that Theorem 16.2 implies that f is rational, say
f = P/Q with polynomials P and Q. Since f has no zeros, P is constant. Induction
shows that f (k) has the form f (k) = Rk /Qk+1 with a polynomial Rk which has no
common zeros with Q and satisfies deg Rk < (k + 1) deg Q = deg(Qk+1 ). This
implies that Rk − Qk+1 has zeros and thus f (k) has 1-points.
Theorem 16.3. Let g be an entire function. Suppose that g(z) 6= 0 and g 00 (z) 6= 0
for all z ∈ C. Then g has the form g(z) = eaz+b with constants a, b ∈ C.
Proof. We consider
g(z)
f (z) = .
g 0 (z)
Then f (z) 6= 0 and
g(z)g 00 (z)
f 0 (z) = 1 − 6= 1
g 0 (z)2
for all z ∈ C. Thus f is constant. Hence g 0 (z)/g(z) = a for some a ∈ C \{0}. This
implies that f has the form claimed.
86 REFERENCES

Remark. A result of Langley (1993) says that if g is a meromorphic function sat-

isfying g(z) 6= 0 and g 00 (z) 6= 0 for all z ∈ C, then g has the form g(z) = eaz+b or
the form g(z) = 1/(az + b)n for some n ∈ N. The proof of this result is much more
involved than that of Theorem 16.3.

References
[1] Anatoly A. Goldberg, Iossif V. Ostrovskii: Value distribution of meromorphic
functions. American Mathematical Society, Providence, RI, 2008.

[2] W. K. Hayman: Meromorphic functions. Clarendon Press, Oxford 1964

[3] Gerhard Jank, Lutz Volkmann: Einführung in die Theorie der ganzen und mero-
morphen Funktionen mit Anwendungen auf Differentialgleichungen. Birkhäuser
Verlag, Basel, 1985.

The above is only a small selection of the many books on the theory of entire and
meromorphic functions.