V.

Ravichandran
Advanced Complex Analysis

Advanced Complex Analysis by V. Ravichandran

Advanced Complex Analysis by V. Ravichandran
 Contents

1. Convex Functions and Hadamard’s Three Circle Theorem . 5

1.1. Convex Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2. Hadamard’s Three Circles Theorem . . . . . . . . . . . . . . . 11
1.3. The Phragmen-Lindelöf Theorem . . . . . . . . . . . . . . . . 14

Advanced Complex Analysis by V. Ravichandran

2. The Space of Continuous Functions . . . . . . . . . . . . . . . . . . 19
2.1. The Space C (G, Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2. Normal Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3. The Space of Analytic Functions . . . . . . . . . . . . . . . . . 32
2.4. The Riemann Mapping Theorem . . . . . . . . . . . . . . . . . 37
2.5. The Weierstrass Factorization Theorem . . . . . . . . . . . . 40
3. Runge’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1. Runge’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2. Simple Connectedness . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3. Mittag-Leffler’s Theorem . . . . . . . . . . . . . . . . . . . . . . 58
4. Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.1. Basic Properties of Harmonic Functions . . . . . . . . . . . . 61
4.2. Harmonic Functions on a Disc . . . . . . . . . . . . . . . . . . . 65
5. The Range of an Analytic Function . . . . . . . . . . . . . . . . . . 73
5.1. Bloch’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2. The Little Picard Theorem . . . . . . . . . . . . . . . . . . . . . 79

3
Advanced Complex Analysis by V. Ravichandran
 1. C ONVEX F UNCTIONS AND H ADAMARD ’ S T HREE
C IRCLE T HEOREM

1.1. Convex Functions

Definition 1.1. Let a, b ∈ R, a < b. A function f : [a, b] → R is convex
if

Advanced Complex Analysis by V. Ravichandran

f (tx1 + (1 − t)x2 ) ≤ t f (x1 ) + (1 − t) f (x2 )
for each pair of points x1 , x2 ∈ [a, b] and for all t with 0 ≤ t ≤ 1.

Definition 1.2. A subset A ⊆ C is convex if, whenever z and w are in A,

the line segment joining z and w is also in A, that is, tz + (1 − t)w ∈ A
for all t with 0 ≤ t ≤ 1.

Theorem 1.3. A function f : [a, b] → R is convex if and only if the epi-

graph (or supergraph) of f given by A = {(x, y) : a ≤ x ≤ b, f (x) ≤ y}
is convex.
Proof. Suppose f : [a, b] → R be a convex function and (x1 , y1 ), (x2 , y2 ) ∈
A. Then f (x1 ) ≤ y1 , f (x2 ) ≤ y2 and so, for 0 ≤ t ≤ 1,
f (tx1 + (1 − t)x2 ) ≤ t f (x1 ) + (1 − t) f (x2 ) ≤ ty1 + (1 − t)y2 (1.1.1)
Now, in view of equation (1.1.1),
t(x1 , y1 ) + (1 − t)(x2 , y2 ) = (tx1 + (1 − t)x2 ,ty1 + (1 − t)y2 ) ∈ A.
Therefore, the epigraph A is a convex set.
Conversely, suppose that the epigraph A is a convex set and let x2 , x1 ∈
[a, b]. Since (x2 , f (x2 )) ∈ A and (x1 , f (x1 )) ∈ A, the convexity of the epi-
graph A shows that
(tx1 +(1−t)x2 ,t f (x1 )+(1−t) f (x2 )) = t(x1 , f (x1 ))+(1−t)(x2 , f (x2 )) ∈ A
for 0 ≤ t ≤ 1. By the definition of the epigraph, we have
f (tx1 + (1 − t)x2 ) ≤ t f (x1 ) + (1 − t) f (x2 )
or the epigraph A is convex.
5
6 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

Theorem 1.4. (a) A function f : [a, b] → R is convex if and

only if for any points x1 , x2 , . . . xn ∈ [a, b] and real numbers
t1 ,t2 , . . .tn ≥ 0 with ∑nk=1tk = 1,

f (∑nk=1tk xk ) ≤ ∑nk=1tk f (xk ) (1.1.2)

(b) A set A ⊆ C is convex if and only if, for any points

z1 , z2 , . . . , zn ∈ A and real numbers t1 ,t2 , . . .tn ≥ 0 with ∑nk=1tk =
1,

∑nk=1tk zk ∈ A (1.1.3)
Proof. (a) If the equation (1.1.2) holds, take n = 2 to get f (t1 x1 +t2 x2 ) ≤

Advanced Complex Analysis by V. Ravichandran

t1 f (x1 ) + t2 f (x2 ) for all x1 , x2 ∈ [a, b], and for all t1 ≥ 0,t2 ≥ 0 with
t1 + t2 = 1. By taking t1 = t,t2 = 1 − t, 0 ≤ t ≤ 1, we have f (tx1 + (1 −
t)x2 ) ≤ t f (x1 ) + (1 − t) f (x2 ) for all x1 , x2 ∈ [a, b], and for all 0 ≤ t ≤ 1.
This shows that f is convex.
Conversely, let f be convex. For n = 1, the equation (1.1.2) reduces
to f (x1 ) ≤ f (x1 ). For n = 2, the equation (1.1.2) holds by definition
of convexity of f . Assume that equation (1.1.2) holds for n ≤ m. Let
n = m + 1 and consider x1 , x2 , . . . , xm+1 ∈ [a, b] and t1 ,t2 , . . . ,tm+1 ≥ 0
with ∑m+1 m
k=1 tk = 1. Writing α = ∑k=1tk , and noting that α +tm+1 = 1, we
have

f (t1 x1 + . . .tm+1 xm+1 )

t1 tm



=f α α x 1 + . . . α x m + tm+1 xm+1
t1 tm


≤αf α x 1 + . . . α x m + tm+1 f (xm+1 )
t1 tm


≤α α f (x 1 ) + . . . α f (x m ) + tm+1 f (xm+1 )
= t1 f (x1 ) + . . .tm f (xm ) + tm+1 f (xm+1 ).

The first inequality above follows by convexity and the second inequal-
ity holds by using induction hypothesis. This proves that the equation
(1.1.2) holds for n = m + 1. This proves (a).
The other part can be proved in a similar manner.
Corollary 1.5. If f : [a, b] → R is convex, then f (x) ≤ max{ f (a), f (b)}
for all x ∈ [a, b].
Proof. By Theorem 1.2, the graph of a convex function lies below the
line joining the points (a, f (a)) and (b, f (b)) and the result follows. In-
deed, any x ∈ [a, b] can be written as x = ta + (1 −t)b for some 0 ≤ t ≤ 1
 1.1. CONVEX FUNCTIONS 7

and therefore

f (x) = f (ta + (1 − t)b) ≤ t f (a) + (1 − t) f (b)

≤ t max{ f (a), f (b)} + (1 − t) max{ f (a), f (b)} = max{ f (a), f (b)}.

Lemma 1.6. A function f : [a, b] → R is convex if and only if

(y − x) f (u) ≤ (y − u) f (x) + (u − x) f (y)

whenever a ≤ x < u < y ≤ b.

Proof. By definition, the function f : [a, b] → R if f (tx + (1 − t)y) ≤
t f (x) + (1 − t) f (y) for all x, y ∈ [a, b] and for all t with 0 ≤ t ≤ 1. Let

Advanced Complex Analysis by V. Ravichandran

a ≤ x < u < y ≤ b. Choose t such that u = tx + (1 − t)y. Then u − x =
(1 − t)(y − x) and y − u = t(y − x), we see that f is convex if and only if

(y − x) f (u) ≤ (y − u) f (x) + (u − x) f (y)

whenever a ≤ x < u < y ≤ b.

Theorem 1.7. If f : [a, b] → R is convex, then it is continuous on (a, b).
Proof. Let x0 ∈ (a, b) and choose α > 0 such that a < x0 − α < x0 +
α < b. Let x0 − α < x < x0 + α. Apply the previous lemma to f on
x0 − α < x0 < x to get

(x − x0 + α) f (x0 ) ≤ (x − x0 ) f (x0 − α) + α f (x)

or equivalently
f (x0 − α) − f (x0 )
(x − x0 ) ≤ f (x) − f (x0 ).
α
Similar appication of the previous lemma to f on x0 < x < x0 + α gives

α f (x) ≤ (x0 + α − x) f (x0 ) + (x − x0 ) f (x0 + α)

equivalently
f (x0 + α) − f (x0 )
f (x) − f (x0 ) ≤ (x − x0 ).
α
Hence, for x0 − α < x0 < x < x0 + α, we have
f (x0 − α) − f (x0 ) f (x) − f (x0 ) f (x0 + α) − f (x0 )
≤ ≤ .
α x − x0 α
8 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

This inequality also holds when x0 − α < x < x0 < x0 + α. Therefore

|x − x0 | < α, we have
 
f (x) − f (x0 ) | f (x0 − α) − f (x0 )| | f (x0 + α) − f (x0 )|
≤ max , := M
x − x0 α α
so that
| f (x) − f (x0 )| ≤ M|x − x0 |.
The continuity of f at x = x0 follows from this inequality. Given ε > 0,
choose δ = min{ε/M, α}.

Example 1.8. The function f : [0, 1] → R defined by f (x) = x for x ∈

Advanced Complex Analysis by V. Ravichandran

(0, 1), f (0) = 2 = f (1) is convex but not continuous at x = 0, 1.

Theorem 1.9. A differentiable function f : [a, b] → R is convex if and

only if f 0 is increasing.
Proof. Assume that f : [a, b] → R is convex. Let a ≤ x1 < x2 ≤ b and
0 < t < 1. By convexity of f , we have
f (tx1 + (1 − t)x2 ) − f (x1 ) ≤ (1 − t)( f (x2 ) − f (x1 ))
and this shows that, as (1 − t)(x2 − x1 ) > 0,
f (x1 + (1 − t)(x2 − x1 )) − f (x1 ) f (x2 ) − f (x1 )
≤
(1 − t)(x2 − x1 ) x2 − x1
By letting t → 1− in the above inequality, we get
f (x2 ) − f (x1 )
f 0 (x1 ) ≤ . (1.1.4)
x2 − x1
Similarly, the convexity of f shows that
f (tx1 + (1 − t)x2 ) − f (x2 ) ≤ t( f (x1 ) − f (x2 ))

and, as t(x1 − x2 ) < 0, we have

f (tx1 + (1 − t)x2 ) − f (x2 ) f (x1 ) − f (x2 ) f (x2 ) − f (x1 )
≥ = .
t(x1 − x2 ) x1 − x2 x2 − x1
By letting t → 0+ in the above inequality, we get
f (x1 ) − f (x2 )
f 0 (x2 ) ≥ . (1.1.5)
x1 − x2
 1.1. CONVEX FUNCTIONS 9

From equation (1.1.4) and (1.1.5), we see that f 0 (x1 ) ≤ f 0 (x2 ) for x1 < x2
and so f 0 is an increasing function.
Conversely, assume that f 0 is increasing. Let a ≤ x1 < x < x2 ≤ b.
Apply the mean-value theorem to find r, s with x1 < r < x < s < x2 such
that
f (x) − f (x1 ) f (x2 ) − f (x)
f 0 (r) = , and f 0 (s) = .
x − x1 x2 − x
Since f 0 is increasing and r < s, we have f 0 (r) ≤ f 0 (s), or
f (x) − f (x1 ) f (x2 ) − f (x)
≤ (1.1.6)
x − x1 x2 − x

Advanced Complex Analysis by V. Ravichandran

Since x1 < x < x2 , we see that x = tx1 + (1 −t)x2 for some t with 0 < t <
1. Since x − x1 = (1 − t)(x2 − x1 ) and x2 − x = t(x2 − x1 ), the inequality
(1.1.6) becomes

t( f (x) − f (x1 )) ≤ (1 − t)( f (x2 ) − f (x))

and, solving this for f (x), we get

f (tx1 + (1 − t)x2 ) = f (x) ≤ t f (x1 ) + (1 − t) f (x2 ).

This proves that the function f is convex.

Corollary 1.10. A twice differentiable function f : [a, b] → R is convex
if and only if f 00 (x) ≥ 0.
Proof. This follows from the previous theorem and the fact that a dif-
ferentiable function f is increasing if and only if f 0 (x) ≥ 0.

Example 1.11. The function f : [a, b] → R defined by f (x) = ex is con-

vex for any a, b ∈ R with a < b. Similarly, f : [a, b] → R defined by
f (x) = − log x is convex for any 0 < a < b.

Definition 1.12. A function f : [a, b] → R is log-convex (or logarithmi-

cally convex) if log f (x) is convex.
Note that log f (x) is defined if and only if f (x) > 0.
Example 1.13. A convex function need not be log-convex though every
log-convex function can be shown to be convex.
(a) Let 0 < a < b. The function f : [a, b] → R given by f (x) = x2
is convex while log f (x) = 2 log x is not convex.
10 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

2
(b) The functions ex , ex are convex as well as log-convex in any
interval [a, b].

Theorem 1.14. Every log-convex function is convex.

Proof. Let f : [a, b] → R be log-convex. Then, the function log f is

convex and, therefore,

log f (tx1 + (1 − t)x2 ) ≤ t log f (x1 ) + (1 − t) log f (x2 )

for 0 ≤ t ≤ 1. Since eu is an increasing function,

f (tx1 + (1 − t)x2 ) ≤ f (x1 )t f (x2 )1−t .

Advanced Complex Analysis by V. Ravichandran

We shall show that

f (x1 )t f (x2 )1−t ≤ t f (x1 ) + (1 − t) f (x2 ),

proving that f is convex. Let a = f (x1 ) and b = f (x2 ). Then a, b > 0

and the previous inequality is the same as at b1−t ≤ ta + (1 − t)b or xt ≤
tx + (1 − t) where x = a/b > 0. For a fixed t with 0 ≤ t ≤ 1, consider
the function f defined on (0, ∞) by

f (x) = xt − tx − 1 + t.

Then f 0 (x) = txt−1 − t = t(xt−1 − 1), f (1) = f 0 (1) = 0, f 0 (x) ≥ 0 for

0 < x ≤ 1, f 0 (x) ≤ 0 for x ≥ 1. Therefore, f (x) ≤ f (1) = 0 for all x ≥ 0.
Hence, xt ≤ tx + (1 − t) holds.

Corollary 1.15. Every twice differentiable log-convex is convex.

Proof. This is a trivial corollary of the previous theorem. Howver, a

different proof can be given using Corollary 1.10. Let f be twice differ-
entiable and log-convex. Then log f is twice differentiable and convex
and, by Corollary 1.10, we see that

f 00 (x) f (x) − ( f 0 (x))2

(log f (x))00 = ≥ 0.
( f (x))2

This shows that f 00 (x) f (x) ≥ f 0 (x)2 ≥ 0. Since f (x) > 0, it follows that
f 00 (x) ≥ 0 and, again by Corollary 1.10, the function f is convex.
 1.2. HADAMARD’S THREE CIRCLES THEOREM 11

1.2. Hadamard’s Three Circles Theorem

Bounded analytic functions on strip domains give rise to log-convex
function. We need the following lemma in proving it.
Lemma 1.16. Let G be the vertical strip {z ∈ C : a < Re z < b}. Sup-
pose f : G → C is continuous and f is analytic in G. If f is bounded in
G, and | f (z)| ≤ 1 for all z ∈ ∂ G, then | f (z)| ≤ 1 for all z ∈ G.
Proof. For each ε > 0, let gε : G → C be defined by gε (z) = 1/(1 +
ε(z − a)). Clearly, Re(1 + ε(z − a)) = 1 + ε(Re z − a) ≥ 1 for z ∈ G.
Therefore, gε is analytic in G and continuous on G. Also, for z ∈ G, we
have

Advanced Complex Analysis by V. Ravichandran

1 1 1
|gε (z)| ≤ ≤ = ≤ 1.
|1 + ε(z − a)| |Re(1 + ε(z − a))| 1 + ε(Re z − a)
In particular, | f (z)gε (z)| ≤ 1 for all z ∈ ∂ G. Since f is bounded (say, by
B) in G, for z ∈ G, we have
B B B
| f (z)gε (z)| ≤ ≤ = (1.2.1)
|1 + ε(z − a)| |Im(1 + ε(z − a))| ε|Imz|
Let R be the rectangle given by R = {z ∈ C : a ≤ Re z ≤ b, | Im z| ≤
B/ε}. The boundary ∂ R consists of part of the vertical lines Re z = a
and Re z = b where | Im z| ≤ B/ε and the two horizondal lines Im z =
±B/ε with a ≤ Re z ≤ b. Since | f (z)gε (z)| ≤ 1 for z ∈ ∂ G, it also holds
on the vertical lines of ∂ R. On the horizondal lines | Im z| = B/ε with
a ≤ Re z ≤ b, using the equation (1.2.1), we have | f (z)gε (z)| ≤ 1. Thus,
| f (z)gε (z)| ≤ 1 for z ∈ ∂ R.
By maximum principle, we have, for z ∈ R,
| f (z)gε (z)| ≤ max{| f (z)gε (z)| : z ∈ R} = max{| f (z)gε (z)| : z ∈ ∂ R} ≤ 1.
For z ∈ G\R, we have |y| > B/ε and the equation (1.2.1) gives | f (z)gε (z)| ≤
1 for z ∈ G \ R. Therefore, for z ∈ G, we have | f (z)gε (z)| ≤ 1 or
1
| f (z)| ≤ = |1 + ε(z − a)|.
|gε (z)|
By letting ε → 0, we have | f (z)| ≤ 1 for all z ∈ G.
Theorem 1.17. Let G be the vertical strip {z ∈ C : a < Re z < b}.
Suppose f : G → C is continuous and f is analytic in G. If f is
bounded in G and f 6≡ 0, then the function M : [a, b] → R defined by
M(x) := sup{| f (z)| : Re z = x, −∞ < Im z < ∞} is log-convex.
12 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

Proof. First, the function M(x) > 0. For, if M(x) = 0 for some x ∈ (a, b),
then by maximum modulus theorem, it follows that f ≡ 0. If M(a) or
M(b) is zero, then f (z) = 0 on Re z = a or Re z = b respectively. Apply-
ing the maximum modulus theorem to f on the rectangle R (defined in
the proof of Lemma 1.16), we see that f (z) = 0 on R and so f ≡ 0 on G
by identity theroem for analytic function.
Secondly, we show that log-convexity of M(x) is equivalent to the
condition
y−u u−x
M(u) ≤ M(x) y−x M(y) y−x (1.2.2)
whenever, a ≤ x < u < y ≤ b. By definition, the function f : [a, b] → R
is convex if f (tx + (1 −t)y) ≤ t f (x) + (1 −t) f (y) for all x, y ∈ [a, b] and

Advanced Complex Analysis by V. Ravichandran

for all t with 0 ≤ t ≤ 1. Let a ≤ x < u < y ≤ b. Choose t such that
u = tx + (1 − t)y. Then u − x = (1 − t)(y − x) and y − u = t(y − x), we
see that f is convex if and only if

(y − x) f (u) ≤ (y − u) f (x) + (u − x) f (y)

whenever a ≤ x < u < y ≤ b. The function M(x) is log-convex if and

only if
(y − u) (u − x)
log M(u) ≤ log M(x) + log M(y) (1.2.3)
(y − x) (y − x)
whenever a ≤ x < u < y ≤ b. The last inequality is equivalent to (1.2.2).
It is enough to prove
b−u u−a
M(u) ≤ M(a) b−a M(b) b−a (1.2.4)

whenever u ∈ (a, b). The general case follows by applying this result to
the interval (x, y). To prove this, define the function g : G → C where
b−z z−a
g(z) = M(a) b−a M(b) b−a .

The function Az = ez log A for A > 0 is analytic and non-vanishing in C.

Therefore, the function g is entire and non-vanishing. Also, as |Az | =
|ez log A | = eRe log A , we have
b−Re z Re z−a
|g(z)| = M(a) b−a M(b) b−a (1.2.5)

Since the right-hand side of equation (1.2.5) is non-vanishing, the func-

tion 1/|g| is continuous in G and therefore, 1/g bounded. This shows
 1.2. HADAMARD’S THREE CIRCLES THEOREM 13

that the function f /g is bounded on G. Also, |g(a + iy)| = M(a) and

|g(b + iy)| = M(b). On Re z = a, we have | f (z)/g(z)| ≤ M(a)/M(a) = 1
and, on Re z = b, | f (z)/g(z)| ≤ M(b)/M(b) = 1. Thus, | f (z)/g(z)| ≤ 1
on ∂ G. Lemma 1.16 applied to the function f /g shows that | f (z)/g(z)| ≤
1 on G. Therefore, for z ∈ G with Re z = u, we have
b−u u−a
| f (z)| ≤ |g(z)| = M(a) b−a M(b) b−a

and therefore
b−u u−a
M(u) = sup{| f (z)| : Re z = u} ≤ M(a) b−a M(b) b−a .

This proves that

Advanced Complex Analysis by V. Ravichandran

M(u)b−a ≤ M(a)b−u M(b)u−a .

Corollary 1.18. Let G be the vertical strip {z ∈ C : a < Re z < b}. Sup-
pose f : G → C is continuous and f is analytic in G. If f is bounded in
G, f 6≡ 0, and f is not constant, then | f (z)| < sup{| f (w)| : w ∈ ∂ G} for
all z ∈ G.
Proof. Since the function M(x) is log-convex, it is convex and hence

M(x) ≤ max{M(a), M(b)}.

For any z ∈ G, we have

| f (z)| ≤ M(x) ≤ max(M(a), M(b)) = sup{| f (w)| : w ∈ ∂ G}.

If, for some z0 ∈ G, the equality | f (z0 )| = sup{| f (w)| : w ∈ ∂ G} holds,

then | f (z)| ≤ | f (z0 )| for all z ∈ G. By maximum modulus theorem, the
function f is constant.
Theorem 1.19 (Hadamard’s Three Circles Theorem). Let 0 < R1 <
R2 < ∞ and suppose f is analytic on ann(0; R1 , R2 ). For R1 < r < R2 ,
define the function M by

M(r) = max{| f (z)| : |z| = r}.

Then, for R1 < r1 ≤ r ≤ r2 < R2 , the following inequality holds:

log r2 − log r log r − log r1
log M(r) ≤ log M(r1 ) + log M(r2 ).
log r2 − log r1 log r2 − log r1
(1.2.6)
14 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

In view of (1.2.3), the inequality (1.2.6) says that log M(r) is convex
function of log r.
Proof. The inequality in (1.2.6) is same as
r
log r2 r2 log rr
M(r) 1 ≤ M(r1 )log r M(r2 ) 1 . (1.2.7)
We prove equation (1.2.7). Let 0 < R1 < r1 < r < r2 < R2 and A =
ann(0; r1 , r2 ). Let G = {z ∈ C : log r1 < Re z < log r2 }. Then the function
g : G → C defined by g(z) = ez maps G onto A and ∂ G onto ∂ A. This
follows since |g(z)| = |ez | = eRe z implies the straight line Re z = log r1
goes to the circle |g| = elog r1 = r1 and the straight line Re z = log r2 goes
to the circle |g| = r2 . Since f is analytic on A, the function F : G → C

Advanced Complex Analysis by V. Ravichandran

defined by F(z) = ( f ◦ g)(z) is analytic in G and continuous on G. Since
f is continuous on the compact set A, it is bounded there and so F is
also bounded on G. For r1 ≤ r ≤ r2 , we have
M(r) = max{| f (w)| : |w| = r} = max{|F(z)| : Re z = log r}.
By Theorem 1.17, for r1 ≤ r ≤ r2 , we have
M(r)log r2 −log r1 ≤ M(r1 )log r2 −log r M(r2 )log r−log r1 .
This proves the inequality (1.2.7).

1.3. The Phragmen-Lindelöf Theorem

Remark 1.20. If f is an entire function and | f (z)| ≤ M(1 + |z|α ) for
some M ≥ 0, 0 ≤ α ≤ 1/2, then f is constant.

Proof. By Cauchy’s integral formula, we have

1 f (w)
Z
0
f (z) = dw
2πi (w − z)2
|w−z|=r

and, therefore,
1 | f (z + reiθ )|
| f 0 (z)| ≤ 2πr
2π r2
1 + |z + reiθ |α 1 + (|z| + r)α
≤M ≤M
r r
 α
M |z| 1
= + M 1/α + 1/α−1 →0 r→∞
r r r
Hence, f 0 = 0 or f is constant.
 1.3. THE PHRAGMEN-LINDELÖF THEOREM 15

Theorem 1.21 (Phragmen-Lindelöf Theorem). Let G be a simply

connected domain in C and ∂∞ G = A ∪ B. Let f , φ be functions ana-
lytic on G and let φ be non-vanishing and bounded on G. If there is a
positive constant M > 0 such that
(a) for every a ∈ A, lim supz→a | f (z)| ≤ M,
(b) for every b ∈ B and for every η > 0, lim supz→b | f (z)||φ (z)|η ≤
M,
then | f (z)| ≤ M for all z ∈ G.
Proof. Since φ is bounded on G, let |φ (z)| ≤ κ for all z ∈ G. Since
G is simply conncected, and φ (z) is non-vanishing, there is an analytic
branch of log φ (z) on G. Hence, g(z) = eη log φ (z) is an analytic branch of

Advanced Complex Analysis by V. Ravichandran

φ (z)η for η > 0. Also, |g(z)| = eηRe log φ (z) = eη log |φ (z)| = |φ (z)|η . De-
fine the function F : G → C by F(z) = f (z)g(z)κ −η . Then F is analytic
on G and as |φ (z)| ≤ κ,
|F(z)| ≤ | f (z)||g(z)|κ −η = | f (z)||φ (z)|η κ −η ≤ | f (z)|κ η κ −η = | f (z)|
By (a) and (b) of hypothesis, we have for a ∈ A,
lim sup |F(z)| ≤ lim sup | f (z)| ≤ M
z→a z→a

and for b ∈ B, and for every η > 0,

lim sup |F(z)| = lim sup | f (z)||φ (z)|η κ −η ≤ Mκ −η
z→b z→b

Therefore, for a ∈ ∂∞ G,
lim sup |F(z)| ≤ max(M, Mκ −η )
z→a

Thus, F satisifes the hypothesis of maximum modulus theorem (third

version). Hence |F(z)| ≤ max(M, Mκ −η ) for all z ∈ G. But f (z) =
κ η F(z)/g(z), and so for z ∈ G,
| f (z)| ≤ κ η |φ (z)|−η max(M, Mκ −η ) = |φ (z)|−η max(Mκ η , M).
Therefore, by letting η → 0+ , we have | f (z)| ≤ M for all z ∈ G.
Corollary 1.22. Let a ≥ 1/2. Suppose that f is analytic on G := {z ∈
C : | arg z| < π/(2a)} and satisfies lim supz→w | f (z)| ≤ M for all w ∈ ∂ G
for constant M. If there are positive constant P and 0 < b < a such
that | f (z)| ≤ P exp(|z|b ) for all z ∈ G with |z| sufficiently large, then
| f (z)| ≤ M for all z ∈ G.
16 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM

Proof. Let ∂∞ G = A ∪ B where A = ∂ G and B = {∞}. By hypothesis,

the function f satisfies the condition (a) of Phragmen-Lindelöf Theo-
rem. To apply Phragmen-Lindelöf Theorem, we need to show that the
conditon (b) of Phragmen-Lindelöf Theorem is satisfied at z = ∞ for
suitable choice of φ .
c
Let b < c < a and put φ (z) = e−z for z ∈ G. Since a ≥ 1/2, G ⊂
C \ {z ∈ C : z ≤ 0} and so log z is analytic in G. Thus, zc = ec log z is
analytic and hence the function φ is analytic in G and non-vanishing
in G. Let z = reiθ ∈ G. Then |θ | < π/(2a) and Re zc = Re(ec log z ) =
ec log r+ciθ = rc cos cθ . Therefore, we have
c c cos cθ
|φ (z)| = e−Re z = e−r .

Advanced Complex Analysis by V. Ravichandran

Since c < a we have |cθ | < ac π2 < π2 , and therefore cos cθ ≥ ρ > 0 for


c
all z ∈ G. Therefore, |φ (z)| ≤ e−ρr ≤ 1 for all z ∈ G. Thus φ is bounded
on G. Also, if η > 0, |z| = r is sufficiently large,
b −ρηrc c (rb−c −ρη)
| f (z)||φ (z)|η ≤ Per = Per

Since b < c, we see that rb−c − ρη → −ρη < 0 as r → ∞, and so

c b−c
rc (rb−c − ρη) → −∞ as r → ∞. Therefore, er (r ρη) → 0 as r → ∞.
Thus,

lim sup | f (z)||φ (z)|η = 0

z→∞

Therefore, the functions f and φ satisfies the hypothesis of Phragmen-

Lindelöf Theorem and hence | f (z)| ≤ M on G.

Corollary 1.23 (TO BE EDITED). Let a ≥ 1/2. Suppose that f is

analytic on G := {z ∈ C : | arg z| < π/(2a)}. Suppose that for every
w ∈ ∂ G, lim supz→w | f (z)| ≤ M. If, for every α > 0, there is a constant
a
P = P(α) such that | f (z)| ≤ Peα|z| for z ∈ G with |z| sufficiently large,
then | f (z)| ≤ M for all z ∈ G.
a
Proof. Fix ε > 0. Define the function F : G → C by F(z) = f (z)e−εz .
Note that G is simply connected and z 6= 0 in G and so log z is well-
defined analytic function and so za can be defined by za = ea log z so that
|za | = eaRe log z = ea log |z| = |z|a . The function F is clearly analytic in G.
If x > 0, and α is so chosen with 0 < α < ε, then there exists P such
a
that | f (x)| ≤ Peαx (for sufficiently large x). For such x > 0, |F(x)| ≤
 1.3. THE PHRAGMEN-LINDELÖF THEOREM 17
a a a
Peαx e−εx = Pe(α−ε)x → 0 as x → ∞ (for, α − ε < 0). Therefore, there
exists k > 0 such that |F(z)| ≤ 1 for x > k. Also,
lim sup | f (z)| ≤ M ⇒ lim sup{| f (z)| : z ∈ G ∩ B(0, r)} ≤ M
z→0 r→0

Thus, ∃ r = r0 such that sup | f (z)| ≤ M on G ∩ B(0, r0 ). This gives,

sup | f (x)| ≤ M on 0 < x < r0 . By continuity of |F| on [r0 , k], | f (x)| ≤ L
for some L in [r0 , k]. Hence, |F| is bounded in (0, ∞). Hence,
M1 := sup{|F(x)| : 0 < x < ∞} < ∞ (1.3.1)
Define M2 = max(M, M1 ) and

Advanced Complex Analysis by V. Ravichandran

H+ = {z ∈ G : 0 < arg z < π/2a}, and H− = {z ∈ G : −π/2a < arg z < 0}.
a a
For z = w with | arg w| = π/2a, since |F(z)| = | f (z)||e−εz | = | f (z)|e−ε|z| ,
we have
a
lim sup |F(z)| = lim sup | f (z)|e−ε|z| ≤ lim sup | f (z)| ≤ M
z→w z→w z→w

For z = x ∈ (0, ∞), as |F(z)| is continuous, we have

lim sup |F(z)| = lim = |F(x)| ≤ sup{|F(x)| : 0 < x < ∞} ≤ M1
z→x z→x

Therfore, for z ∈ ∂ H+ ∪ ∂ H− , we have

lim sup |F(z)| ≤ M2
z→w

Hence, by Corollary 1.22, we have, |F(z)| ≤ M2 for all z ∈ H+ ∪ H− .

Therefore, |F(z)| ≤ M2 for all z ∈ G.
The proof is complete if we show that M2 = M. If M2 6= M, as
M2 = max(M1 , M), we have M2 = M1 > M. Since limx→∞ |F(x)| = 0,
F(x) is closer to 0 near ∞. Also,
lim sup |F(x)| = lim sup | f (x)| ≤ M < M1
x→0 x→0

Thus, |F(x)|, being continuous (on some compact interval , sup = max),
we must have 0 < x0 < ∞ such that |F(x0 )| = M1 . Thus |F| assumes
maximum in (0, ∞) or F assumes maximum in G. Therefore, F is con-
stant. Thus, |F(z)| = M1 and sup |F(z)| = M1 = M2 . Hence, for w ∈
∂ G, M = limz→w sup |F(z)| = M1 . Therefore, M = M1 = M2 and hence,
a a
|F(z)| ≤ M for all z ∈ G. Therefore, | f (z)| ≤ M|eεz | = MeεRez → M as
ε → 0. Thus, | f (z)| ≤ M for all z ∈ G.
18 1. CONVEX FUNCTIONS AND HADAMARD’S THREE CIRCLE THEOREM
a a
Remark 1.24. If G = {z : | arg z| < π/2a}, f (z) = ez . Then | f | = e|z| cos aθ
a
where θ = arg z. For z ∈ ∂ G, θ = π/2a, and | f | = e|z| cos π/2 = e0 = 1.
a
But | f | is unbounded (For, z = x > 0, | f | = ex → ∞ as x → ∞). Thus,
the above corollary cannot be improved.

Advanced Complex Analysis by V. Ravichandran

 2. T HE S PACE OF C ONTINUOUS F UNCTIONS

2.1. The Space C (G, Ω)

Definition 2.1. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. Then class C (G, Ω) consists of all continuous func-
tions from G to Ω.

Advanced Complex Analysis by V. Ravichandran

Every analytic function defined on G belongs to C (G, C) and ev-
ery meromorphic functions defined on G belongs to C (G, C∞ ). Since
the constant functions are continuous, the class C (G, Ω) is non-empty.
However, it can contain only constant functions. For example, if Ω ⊆
N = {1, 2, . . .}, and G is connected, then, for any function f ∈ C (G, Ω),
the set f (G) is connected in Ω. Thus, f (G) = {n} for some n ∈ N and
so f is a constant function.
Theorem 2.2. Any open subset G ⊂ C can be written as G = ∪∞ n=1 Kn
for some sequence hKn i of compact subsets of G having the following
properties:
(a) Kn ⊂ int Kn+1 ,
(b) Each compact subset K ⊂ G is contained in Kn0 for some n0 ∈
N,
(c) Every component of C∞ \ Kn contains a component of C∞ \ G.
Proof. For each n, let

Kn = {z : |z| ≤ n} ∩ {z : d(z, C \ G) ≥ 1/n}

The function f (z) = d(z, C\G) is continuous and hence the set {z : d(z, C \ G) ≥
f −1 ([1/n, ∞)) is closed and so Kn , being intersection of two closed sets,
is closed. Since Kn ⊂ {z : |z| ≤ n}, the set Kn is bounded. Thus, the set
Kn is compact. To prove G = ∪∞ n=1 Kn , we first prove (a).
The set {z : |z| < n + 1} ∩ {z : d(z, C \ G) > 1/(n + 1)} is open, and
it contains Kn and is contained in Kn+1 . Therefore,

Kn ⊆ {z : |z| < n + 1} ∩ {z : d(z, C \ G) > 1/(n + 1)} ⊆ int Kn+1

This proves (a).

19
20 2. THE SPACE OF CONTINUOUS FUNCTIONS

We now prove that G = ∪∞ n=1 Kn . Let z ∈ G. Since {z} is compact and

disjoint from the closed set C \ G, we see that d(z, C \ G) = d({z}, C \
G) > 0 and so choose m1 such that d(z, C \ G) > 1/m1 . Choose m2 such
that |z| ≤ m2 . Let n = max(m1 , m2 ). Then d(z, C \ G) ≥ 1/n, and |z| < n
or z ∈ Kn . Therefore, z ∈ ∪Kn and so G ⊆ ∪Kn . But Kn ⊆ G and so
∪Kn ⊆ G. Therefore, G = ∪Kn .
To prove (b), let K be a compact subset of G. Since Kn ⊆ int Kn+1 ,
we have K ⊆ G = ∪Kn ⊆ ∪ int Kn and so {int Kn } is an open covering of
K. The compactness of K implies that there are nk (k = 1, 2 . . . , m) such
that K ⊆ ∪m k=1 int Kn ⊆ Kn0 where n0 = max{n1 , . . . , nm }. This proves
(b).
To prove (c), let D be an unbounded component of C∞ \ Kn . We shall

Advanced Complex Analysis by V. Ravichandran

show that D contains the unbounded component D1 of C∞ \ G. Since
D, D1 are unbounded, ∞ ∈ D ∩ D1 and so D ∪ D1 is a connected set in
C∞ \ Kn which contains the component D and therefore D ∪ D1 = D or
D1 ⊂ D.
We shall now consider a bounded component D of C∞ \ Kn . The
proof will be similar except for finding a point of D that also belongs to
C∞ \ G. (In case of unbounded component, this common point was ∞).
Since

C∞ \ Kn = {z : |z| > n} ∪ {z : d(z, C \ G) < 1/n}

and {z : |z| > n} is connected, the unbounded component of C∞ \ Kn

contains the set {z : |z| > n} and therefore D ⊂ {z : d(z, C \ G) < 1/n}.
There is a point z0 ∈ D with d(z0 , C \ G) < 1/n. If |z0 − w| ≥ 1/n for
every w ∈ C \ G, then d(z0 , C \ G) = inf{|z0 − w| : w ∈ C \ G} ≥ 1/n.
Thus, there exists w0 ∈ C \ G with |z0 − w0 | < 1/n. If z ∈ B(w0 , 1/n),
then d(z, C \ G) = inf{|z − w| : w ∈ C \ G} ≤ |z − w0 | < 1/n and so
z ∈ C∞ \ Kn . This shows that B(w0 , 1/n) ⊂ C∞ \ Kn . Since disks are
connected, z0 ∈ B(w0 , 1/n), D is a component of C∞ \ Kn containing z0 ,
B(w0 , 1/n) ⊆ D. Thus, we have a point w0 ∈ D ∩ (C∞ \ G). If D1 is
the component of C∞ \ G that contains w0 , then D1 ⊂ D. This follows
because D1 ∪D is a connected subset of C∞ \Kn which properly contains
D.

Let f , g ∈ C (G, Ω). Let G = ∪∞

n=1 Kn where Kn ’s are compact and
Kn ⊆ int Kn+1 . Define

ρn ( f , g) = sup{d( f (z), g(z)) : z ∈ Kn }.

 2.1. THE SPACE C (G, Ω) 21

Clearly ρn ( f , g) ≥ 0 and ρn ( f , g) = 0 if and only if f = g. It is also

trivial to see that ρn ( f , g) = ρn (g, f ). Since

d( f (z), h(z)) ≤ d( f (z), g(z)) + d(g(z), h(z)) ≤ ρn ( f , g) + ρn (g, h),

by taking supremum over all z ∈ Kn , the triangle inequality

ρn ( f , h) ≤ ρn ( f , g) + ρn (g, h)

follows. Thus, ρn is a metric on Kn . Using this metric, we introduce a

metric on C (G, Ω).
Definition 2.3. Let G be an open subset of C and let (Ω, d) be a com-

Advanced Complex Analysis by V. Ravichandran

plete metric space. Let G = ∪∞ n=1 Kn where Kn ’s are compact and Kn ⊆
int Kn+1 . Define ρ : C (G, Ω) × C (G, Ω) → R by
∞
1 n

ρn ( f , g)
ρ( f , g) = ∑ 2 . (2.1.1)
n=1 1 + ρn ( f , g)

Since t/(1 + t) ≤ 1 for all t ≥ 0, the series in equation (2.1.1) is

dominated by ∑ (1/2)n and therefore the series is convergent. To prove
that ρ is a metric, we first prove the following lemma.
Lemma 2.4. If (X, d) is a metric space, then

d(s,t)
µ(s,t) =
1 + d(s,t)

is also a metric in X. A set A is open in (X, d) if and only if it is open in

(X, µ). A sequence is a Cauchy sequence in (X, d) if and only if it is a
Cauchy sequence in (X, µ).
Proof. Clearly µ(s,t) ≥ 0. Also, µ(s,t) = 0 if and only if d(s,t) = 0
and this holds if and only if s = t. Also, µ(s,t) = µ(t, s). To prove the
triangle enquality, we first note thatd(r,t) ≤ d(r, s) + d(s,t). Using this,
we have
d(r,t) 1 1
µ(r,t) = = 1− ≤ 1−
1 + d(r,t) 1 + d(r,t) 1 + d(r, s) + d(s,t)
d(r, s) d(s,t)
= +
1 + d(r, s) + d(s,t) 1 + d(r, s) + d(s,t)
d(r, s) d(s,t)
≤ + = µ(r,t) + µ(s,t)
1 + d(r, s) 1 + d(s,t)
22 2. THE SPACE OF CONTINUOUS FUNCTIONS

If t ∈ Bd (s, ε), then d(s,t) ≤ ε, and

d(s,t)
µ(s,t) = ≤ d(s,t) < ε.
1 + d(s,t)
Thus, t ∈ Bµ (s, ε) or
Bd (s, ε) ⊆ Bµ (s, ε) (2.1.2)
Similarly, if µ(s,t) < ε < 1, then
µ(s,t) ε
d(s,t) = ≤ .
1 − µ(s,t) 1 − ε
This shows that

Advanced Complex Analysis by V. Ravichandran

 
ε
Bµ (s, ε) ⊆ Bd s, (2.1.3)
1−ε
In view of equations (2.1.2) and (2.1.3), the topologies on (X, d) and
(X, µ) are the same.
Theorem 2.5. With ρ as given in Definition 2.3, the space (C (G, Ω), ρ)
is a metric space.
Proof. Let G = ∪∞ n=1 Kn where Kn ’s are compact and ρn ( f , g) = sup{d( f (z), g(z)) :
z ∈ Kn }. Then, by previous lemma, the function µn defined on Kn by
ρn ( f , g)
µn ( f , g) =
1 + ρn ( f , g)
is a metric. The metric ρ is then defined by
∞ ∞
1 n

ρn ( f , g) 1 n


ρ( f , g) = ∑ 2 =∑ 2 µn ( f , g) (2.1.4)
n=1 1 + ρn ( f , g) n=1
Clearly, ρ( f , g) ≥ 0. If ρ( f , g) = 0, then equation (2.1.4) shows that
µn ( f , g) = 0 for all n and so, f (z) = g(z) for all z ∈ Kn . Since G = ∪Kn ,
f = g on G. If f = g, then ρ( f , g) = 0. Also, it is trivial to see that
ρ( f , g) = ρ(g, f ).
Since µn is a metric, it satisfies the triangle inequality for each n and
so
∞
n
ρ( f , g) = ∑ 21 µn ( f , g)
n=1
∞ ∞
1 n 1 n



≤ ∑ 2 µn ( f , h) + ∑ 2 µn (h, g) = ρ( f , h) + ρ(h, g).
n=1 n=1
 2.1. THE SPACE C (G, Ω) 23

Lemma 2.6. Let G be an open subset of C and let (Ω, d) be a complete

metric space. Let the metric ρ on C (G, Ω) be defined as in equation
(2.3). (a) If ε > 0 is given, then there is δ > 0 and a compact set K ⊂ G
such that, for f , g ∈ C (G, Ω),

sup{d( f (z), g(z)) : z ∈ K} < δ =⇒ ρ( f , g) < ε.

(b) Conversely, if δ > 0 and a compact set K ⊆ G are given, then there
is an ε > 0 such that, for f , g ∈ C (G, Ω),

ρ( f , g) < ε =⇒ sup{d( f (z), g(z)) : z ∈ K} < δ .

Proof. (a) Let Kn ’s be the compact sets used in Definition 2.3 to define

Advanced Complex Analysis by V. Ravichandran

the metric ρ. Then G = ∪∞ n=1 Kn . For given ε > 0, choose a positive
n
integer p such that ∑∞ n=p+1 (1/2) < ε/2 and put K = K p . Choose δ > 0
such that t/(1 + t) < 2 whenever 0 ≤ t < δ . Suppose f , g ∈ C (G, Ω)
ε

satisfy sup{d( f (z), g(z) : z ∈ K} < δ . Since K1 ⊂ K2 ⊂ K3 ⊂ · · · ⊂ K p =

K, we have
ρn ( f , g) = sup{d( f (z), g(z) : z ∈ Kn } ≤ sup{d( f (z), g(z) : z ∈ K} < δ
for n = 1, 2, . . . , p. The choice of δ shows that, for n = 1, 2, . . . , p,
ρn ( f , g) ε
< .
1 + ρn ( f , g) 2
Therefore,
∞ p ∞
1 n ρn ( f , g)

n 1 n
< ∑ ε2 12 + ∑



ρ( f , g) = ∑ 2 2
n=1 1 + ρn ( f , g) n=1 n=p+1

p  ∞
n
= 12 ε · 1 − 12 + ∑ 12 < 12 ε + 12 ε = ε
n=p+1

(b) Let K, δ be given. Since G = ∪∞n=1 Kn = ∪n=1 int Kn , K is com-

∞

pact, we have K ⊆ K p for some p ≥ 1. Then, for this p, we have

ρ p ( f , g) = sup{d( f (z), g(z) : z ∈ K p } ≥ sup{d( f (z), g(z)) : z ∈ K}.
(2.1.5)
Let ε > 0 be chosen such that s/(1−s) < δ whenever 0 ≤ s < 2 ε. Then, p

replacing s by t/(1 +t), we see that t < δ whenever 0 < t/(1 +t) < 2 p ε.
If ρ( f , g) < ε, then

1 p

ρ p ( f , g)
2 ≤ ρ( f , g) < ε
1 + ρ p ( f , g)
24 2. THE SPACE OF CONTINUOUS FUNCTIONS

and so, by the choice of δ , we have ρ p ( f , g) < δ . Using this in (2.1.5),

we have
sup{d( f (z), g(z)) : z ∈ K} < δ .

Theorem 2.7. Let G be an open subset of C and let (Ω, d) be a complete

metric space. A set O ⊆ C (G, Ω) is open in (C (G, Ω), ρ) if and only
if, for each f ∈ O, there is a compact set K and δ > 0 such that {g :
sup{d( f (z), g(z)) : z ∈ K} < δ } ⊂ O.
Proof. Let O be open in (C (G, Ω), ρ) and f ∈ O. Then, there is an
ε > 0 such that {g : ρ( f , g) < ε} ⊂ O. For this ε, using Lemma 2.6,
find a compact set K, and δ > 0 such that sup{d( f (z), g(z)) : z ∈ K} <

Advanced Complex Analysis by V. Ravichandran

δ ⇒ ρ( f , g) < ε. Hence,

{g : sup{d( f (z), g(z)) : z ∈ K} < δ } ⊂ {g : ρ( f , g) < ε} ⊂ O.

Conversely, if O satisfies the stated property, and f ∈ O, then Lemma

2.6 provides an ε > 0 such that ρ( f , g) < ε ⇒ sup{d( f (z), g(z)) : z ∈
K} < δ . Hence, {g : ρ( f , g) < ε} ⊆ {g : sup{d( f (z), g(z)) : z ∈ K} <
δ } ⊆ O or O is open in (C (G, Ω), ρ).
Corollary 2.8. The collection of open sets in (C (G, Ω), ρ) is indepen-
dent of the choice of the sets {Kn } used in defining the metric ρ.
Proof. This follows as the characterization of open sets given in the pre-
vious theorem does not depend on the sets Kn used to define the metric
ρ.
Theorem 2.9. Let G be an open subset of C. Let fn , f ∈ C (G, C). Then,
fn → f in C (G, C) if and only if fn (z) → f (z) uniformly on each com-
pact subset of G.
Proof. Let h fn i be a sequence in (C (G, Ω), ρ) convergent to f . Let K
be a compact subset of G and δ > 0. Using Lemma 2.6, choose ε > 0
such that, for f , g ∈ C (G, Ω),

ρ( f , g) < ε ⇒ sup{d( f (z), g(z)) : z ∈ K} < δ . (2.1.6)

For this ε > 0, find N > 0 such that ρ( fn , f ) < ε for all n ≥ N. Therefore,
using the implication in (2.1.6), we see that for n ≥ N, sup{d( fn (z), f (z)) :
z ∈ K} < δ . Thus, fn → f uniformly on K.
Conversely, let fn → f uniformly on compact subsets of G. Let ε > 0
be given. Using Lemma 2.6, find a compact set K and δ > 0 such that
 2.1. THE SPACE C (G, Ω) 25

sup{d( f (z), g(z)) : z ∈ K} < δ ⇒ ρ( f , g) < ε. Since fn → f uniformly

on compact sets, there is N > 0 such that sup{d( fn (z), f (z)) : z ∈ K} < δ
for all n ≥ N. Therefore ρ( fn , f ) < ε for all n ≥ N. Thus, fn → f in
C (G, C).
Since closed disks are compact, one way implication in the follow-
ing theorem is trivial.
Corollary 2.10. Let G be an open subset of C. Let fn , f ∈ C (G, C).
Then, fn → f in C (G, C) if and only if fn (z) → f (z) uniformly on each
closed disk inside G.
Proof. For the converse, assume that fn (z) → f (z) uniformly on each
closed disk inside G. Let K be a compact subset of G. Since K is

Advanced Complex Analysis by V. Ravichandran

disjoint from the closed set ∂ G, it follows that R := d(K, ∂ G) > 0. Then,
for each a ∈ K, the disk D(a, R) ⊂ G. Thus, the collection {D(a, R) :
a ∈ K} is open cover of K and, therefore, there are a1 , a2 , . . . , an ∈ K
such that K ⊂ ∪nk=1 D(ak , R). Let ε > 0 be given. For each k, there
is Nk > 0 such that | fn (z) − f (z)| < ε for each z ∈ D(ak , R). Let N =
max{N1 , . . . , Nn }. Then, for each z ∈ K, we have k such that z ∈ D(ak , R)
and so | fn (z) − f (z)| < ε. Therefore, fn (z) → f (z) uniformly on K. By
previous theorem, it follows that fn → f in C (G, C).
Theorem 2.11. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. Then, the space C (G, Ω) is a complete metric space.
Proof. Let h fn i be a Cauchy sequence in C (G, Ω). For a given δ > 0, a
compact subset K ⊆ G, Lemma 2.6 gives an ε > 0 such that

ρ( f , g) < ε ⇒ sup{d( f (z), g(z)) : z ∈ K} < δ (2.1.7)

For this ε > 0, choose N > 0 such that ρ( fn , fm ) < ε for all n, m ≥
N. Then, the implication in (2.1.7) shows that sup{d( fn (z), fm (z)) : z ∈
K} < δ for all n, m ≥ N. Thus, h fn (z)i is a Cauchy sequence in Ω for
each z ∈ K. Let fn (z) → f (z). This defines a function f : G → Ω.
We complete the proof by showing f is continuous and ρ( fn , f ) →
0. Let K be a compact subset of G, and δ > 0. Choose N so that
sup{d( fn (z), fm (z)) : z ∈ K} < δ for all n, m ≥ N. If z ∈ K is arbi-
trary, but fixed, there exists a fixed m ≥ N such that d( f (z), fm (z)) <
δ as fn (z) → f (z). This shows that d( f (z), fn (z)) ≤ d( f (z), fm (z)) +
d( fm (z), fn (z)) < 2δ for all n ≥ N. Since N is independent of z, sup{d( f (z), fn (z)
z ∈ K} ≤ 2δ and fn (z) → f (z) uniformly on compact set K. Thus, f is
continuous on each ball B contained in G as the convergence is uniform
26 2. THE SPACE OF CONTINUOUS FUNCTIONS

in B. Thus, f is continuous on G. Since fn , f ∈ C (G, Ω), Theorem 2.9

together with the fact that fn → f uniformly on compact sets in G shows
that fn → f in C (G, Ω) or ρ( fn , f ) → 0.

2.2. Normal Families

Definition 2.12. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. A set F ⊂ C (G, Ω) is normal if each sequence in
F has a subsequence which converges to a function f in C (G, Ω).
The limit of the subsequence is not required to be in F . If the limit
belongs to F , then F is sequentially compact. Since C (G, Ω) is a met-
ric space, sequentially compactness of F is equivalent to compactness

Advanced Complex Analysis by V. Ravichandran

of F . A set F is said to be relatively compact if F is compact.
Theorem 2.13. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. A set F ⊂C (G, Ω) is normal if and only if its closure
F is compact.

Proof. Let F be compact. Let h fn i ⊆ F be any sequence. Then h fn i ⊆

F and F is compact and so sequentially compact, there is a function
f ∈ F ⊆ C (G, Ω) such that fnk → f . Therefore, F is normal.
Let F be normal. We show that F is compact. Let h fn i be a se-
quence in F . Since fn ∈ F ), there is gn ∈ F such that ρ(gn , fn ) < 1/n
for each n. Since hgn i is a sequence in F it has a subsequence hgnk i
converging to g in C (G, Ω). Therefore, there exists a positive inte-
ger N such that ρ(gnk , g) < ε/2 for nk ≥ N. Using this together with
ρ(gnk , fnk ) < 1/nk , we have
ε ε
ρ( fnk , g) ≤ ρ( fnk , gnk ) + ρ(gnk , g) ≤ + for nk ≥ max (N, 2/ε)
2 2

Therefore, fnk converges to the function g in C (G, Ω). Since F is

closed, the limit function g ∈ F . This proves that F is sequentially
compact and hence compact.
A subset Y of a metric space X is totally bounded if for every ε > 0,
there are points y1 , y2 , . . . , yn ∈ Y such that

Y ⊂ ∪nk=1 B(yk , ε).

The following result about the total boundedness is required in the proof
of our next theroem.
 2.2. NORMAL FAMILIES 27

Lemma 2.14. A subset Y of a metric space X is totally bounded if and

only if Y is totally bounded.
Proof. If Y is totally bounded, then, for any ε > 0, there are y1 , y2 , . . . yn ∈
Y such that
Y ⊂ ∪nk=1 B yk , ε2

and therefore
Y = ∪nk=1 B yk , ε2 ⊆ ∪nk=1 B(yk , ε).

This proves Y is totally bounded.

Conversely, assume that Y is totally bounded and let ε > 0. Then,

Advanced Complex Analysis by V. Ravichandran

there are points y1 , y2 , . . ., yn ∈ Y such that

Y ⊆ ∪nk=1 B yk , ε2 .

Since yk ∈ Y for 1 ≤ k ≤ n, we have Y ∩ B (yk , ε/2) 6= 0. / Choose y0k ∈

Y ∩ B (yk , ε/2). We claim that Y ⊆ ∪nk=1 B(yk , ε) and this will prove that
Y is totally bounded. Let y ∈ Y . Then y ∈ Y and therefore, there is an
integer k, 1 ≤ k ≤ n, such that y ∈ B(yk , ε/2) and therefore d(y, yk ) <
ε/2. Also, by the choice of y0k , we have d(yk , y0k ) < ε/2. These two
together show that d(y, y0k ) ≤ d(y, yk ) + d(yk , y0k ) < ε/2 + ε/2 = ε, or
y ∈ B(y0k , ε) ⊆ ∪nk=1 B(y0k , ε). Therefore, Y ⊆ ∪nk=1 B(y0k , ε).
Theorem 2.15. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. A set F ⊆ C (G, Ω) is normal if and only if, for
every compact set K ⊆ G and δ > 0, there are functions f1 , f2 , . . . , fn ∈
F such that, for f ∈ F , there is atleast one k, 1 ≤ k ≤ n, with
sup{d( f (z), fk (z)) : z ∈ K} < δ .
Proof. Suppose F is normal. Let K ⊆ G be compact, and δ > 0 be
given. By Lemma 2.6, there is ε > 0 such that, for f , g ∈ C (G, Ω), we
have
ρ( f , g) < ε ⇒ sup{d( f (z), g(z)) : z ∈ K} < δ .
Since F is normal, F is compact and therefore, F is totally bounded.
By previous result, F is also totally bounded. Thus, there are functions
f1 , f2 , . . . , fn ∈ F such that

F ⊆ ∪nk=1 B( fk , ε) = ∪nk=1 { f : ρ( f , fk ) < ε}

⊆ ∪nk=1 { f : sup{d( f (z), fk (z)) : z ∈ K} < δ }.
28 2. THE SPACE OF CONTINUOUS FUNCTIONS

This shows that, if f ∈ F , there is k, 1 ≤ k ≤ n, such that sup{d( f (z), fk (z)) :

z ∈ K} < δ .
Conversely, assume that for every compact set K ⊆ G and δ > 0,
there are f1 , f2 , . . ., fn ∈ F such that

sup{d( f (z), fk (z)) : z ∈ K} < δ . (2.2.1)

We need to show that F is normal. By Theorem 2.13, it will follow

if F is compact. Recall that a metric space is compact if and only if
it is complete and totally bounded. Since F is a closed subset of the
complete metric space C (G, Ω), F is complete. Thus, we need to just
prove that F is totally bounded. Let ε > 0. Then, by Lemma 2.6, there

Advanced Complex Analysis by V. Ravichandran

exists a compact set K ⊆ G, and δ > 0 such that

sup{d( f (z), fk (z)) : z ∈ K} < δ ⇒ ρ( f , g) < ε.

Applying the hypothesis to this K, and δ > 0, we get f1 , f2 , . . . , fn ∈ F

such that, for f ∈ F , there is k, 1 ≤ k ≤ n, such that sup{d( f (z), fk (z)) :
z ∈ K} < δ . Therefore, we have ρ( f , fk ) < ε and hence f ∈ Bρ ( fk , ε)
or F ⊆ ∪nk=1 Bρ ( fk , ε). Hence, F is totally bounded and therefore F is
totally bounded.
Let (Xn , dn ) be a metric space for each n ≥ 1, and let X = ∏∞
n=1 Xn
be their Cartesian product. Any element is ξ = hxn i, xn ∈ Xn . For ξ =
hxn i, η = hyn i in X, define
∞
1 n

dn (xn , yn )
d(ξ , η) = ∑ 2 . (2.2.2)
n=1 1 + dn (xn , yn )

It is easy to see that d is a metric (see the proof of Theorem 2.5). There-
fore, the space (X, d) is a metric space.
Theorem 2.16. . Let Xn be a metric space for each n ≥ 1, and let
X = ∏∞ n=1 Xn be their Cartesian product with metric d defined by (2.2.2).
(a) If ξ k = hxnk i∞ k k
n=1 ∈ X, then ξ → ξ = hxn i if and only if xn → xn
for each n.
(b) Also, if (Xn , d) is compact, then X is compact.
Proof. (a) This proof is similar to the proof of Lemma 2.6. Let ξ k → ξ .
Then d(ξ k , ξ ) → 0. Thus, using equation (2.2.2), we get, for each n ≥ 1,

dn (xnk , xn )
0 ≤ Snk = k
≤ 2n d(ξ k , ξ ). (2.2.3)
1 + dn (xn , xn )
 2.2. NORMAL FAMILIES 29

Thus, for each n ≥ 1, Snk → 0 as k → ∞ and so

Snk
dn (xnk , xn ) = → 0 as k → ∞
1 − Snk
Therefore, for each n ≥ 1, xnk → xn as k → ∞.
Conversely, let xnk → xn for each n.To show ξ k → ξ as k → ∞, for a
given ε > 0, we need to find N such that d(ξ k , ξ ) < ε for all k ≥ N. Let
xnk → xn for
each n ≥ 1. Let ε > 0 be given. Let p be chosen such that
1 n
∑∞ n=p+1 2 < ε/2. Now, for 1 ≤ n ≤ p, choose Nn such that d(xnk , xn ) <
ε/2 for all k ≥ Nn . Now, for k ≥ max(N1 , N2 , . . . , N p ) = N,
p
k 1 n

d(xnk , xn ) ∞
1 n

d(xnk , xn )
d(ξ , ξ ) = ∑ + ∑

Advanced Complex Analysis by V. Ravichandran

2 1 + d(xnk , xn ) n=p+1 2 1 + d(xnk , xn )
n=1
p ∞
1 n k 1 n



≤ ∑ 2 d(xn , xn ) + ∑ 2
n=1 n=p+1
p
ε 1 n

ε ε ε
≤ ∑ 2 + < + =ε
2 n=1 2 2 2
Therefore, ξ k → ξ as k → ∞.
(b) Let (Xk , dk ) be compact for each k ≥ 1. Let X = ∏ Xk . Let hξ n i be
a sequence in X. Let ξ n = hxkn i∞ n k n ∞
k=1 , xk ∈ X . Since hxk ik=1 is a sequence
in X, there exists x1 ∈ X1 and an infinite subset N1 ⊆ N such that lim{x1k :
k ∈ N1 } = x1 . Now consider, hx2k ik∈N1 . This is an infinite sequence in
X2 and therefore, there exists x2 ∈ X2 , infinite set N2 ⊆ N1 such that
lim{x2k : k ∈ N2 } = x2 . Continuing this way, we get a sequence x1 , x2 , . . .
with xn ∈ Xn , and Nn ⊆ . . . ⊆ N2 ⊆ N1 ⊆ N of infinite sets such that
lim{xnk : k ∈ Nn } = xn (2.2.4)
Let k j be the jth integer in N j . We claim that ξ k j → ξ = (xn )∞
k=1 or,
k k
ξn j → xn as k j → ∞. This follows from equation (2.2.4) as hxn j : j ≥ ni
is a subsequence of hxnk : k ∈ Nn i.

Definition 2.17. Let G be an open subset of C and let (Ω, d) be a

complete metric space. A set F ⊆ C (G, Ω) is equicontinuous at a
point z0 ∈ G if and only if for every ε > 0, there is a δ > 0 such that
d( f (z), f (z0 )) < ε for all f ∈ F , and for all z ∈ G with |z − z0 | < δ . A
set F is equicontinuous over a set E ⊆ G if and only if for every ε > 0,
there is a δ > 0 such that d( f (z), f (z0 )) < ε for all f ∈ F , and for all
z, z0 ∈ E with |z − z0 | < δ .
30 2. THE SPACE OF CONTINUOUS FUNCTIONS

If F = { f }, the set F is equicontinuous at z0 is the same as f is

continuous at z0 , while F is equicontinuous over a set E is the same as
f is uniformly continuous on E.
Theorem 2.18. Let G be an open subset of C and let (Ω, d) be a com-
plete metric space. Suppose F ⊆ C (G, Ω) is equicontinuous at each
point of G. Then F is equicontinuous over each compact subset of G.
Proof. Let K ⊆ G be compact. Let ε > 0 be given. Then, for each
w ∈ K, there is δw > 0 such that d( f (w), f (w0 )) < ε/2 for all f ∈ F
and for all w ∈ K with |w − w0 | < δw . The balls {B(w, δw ) : w ∈ K}
form an open cover of K. By Lebesgue Covering Lemma, there is a
δ > 0 such that, for each z ∈ K, B(z, δ ) is contained in one of the balls
in {B(w, δw ) : w ∈ K}. If z, z0 ∈ K, |z − z0 | < δ , there is w ∈ K such

Advanced Complex Analysis by V. Ravichandran

that z0 ∈ B(z, δ ) ⊆ B(w, δw ). Therefore, |z − w| < δw and |z0 − w| < δw .
This implies that d( f (z), f (w)) < ε/2 and d( f (z0 ), f (w)) < ε/2 for all
f ∈ F . Therefore,

d( f (z), f (z0 )) ≤ d( f (z), f (w)) + d( f (z0 ), f (w)) < ε/2 + ε/2 = ε

for all f ∈ F . Hence, F is equicontinuous on K.

Theorem 2.19 (Arzela-Ascoli Theorem). Let G be an open subset of
C and let (Ω, d) be a complete metric space. A set F ⊆ C (G, Ω) is
normal if and only if the following conditions are satisfied:
(a) for each z ∈ G, { f (z) : z ∈ F } has compact closure in Ω, or it
is relatively compact.
(b) F is equicontinuous at each point of G.
Proof. Assume that F is normal. We first show that F z = { f (z) : z ∈
F } for any fixed z ∈ G has compact closure. First note that the function
F : C (G, Ω) → Ω defined by F( f ) = f (z) for any fixed z ∈ G is continu-
ous. For, if δ > 0 is given and K = {z}, by Lemma 2.6, there is an ε > 0
ρ(g, f ) < ε ⇒ d(g(z), f (z)) < δ or d(F(g), F( f )) < δ . Since contin-
uous functions maps compact sets to compact sets, F(F ) is compact.
But, F z = F(F )⊆ F(F ) and, therefore, F z ⊆ F(F ). Being closed
subset of a compact set F(F ), F z is compact.
We now show that F is equicontinuous at each point of G. Let z0 ∈
G be fixed and ε > 0 be given. Choose R > 0 so that K = B(z0 , R) ⊂ G.
Then K is compact. By Theorem 2.15, there are functions f1 , f2 , . . . , fn ∈
F such that, for each f ∈ F , there is at least one fk with

sup{d( f (z), fk (z)) : z ∈ K} < ε/3 (2.2.5)

 2.2. NORMAL FAMILIES 31

Since fk is continuous, there is δk with 0 < δk < R such that d( fk (z), fk (z0 )) <
ε/3 for all z ∈ K with |z − z0 | < δk . Let δ = min{δk : 1 ≤ k ≤ n}. Then,
d( fk (z), fk (z0 )) < ε/3 for all 1 ≤ k ≤ n and for all z ∈ K with |z−z0 | < δ .
Let z ∈ K satisfy |z − z0 | < δ , and let f ∈ F . Choose k satisfying the
inequality (2.2.5). Then

d( f (z), f (z0 )) ≤ d( f (z), fk (z)) + d( fk (z), fk (z0 )) + d( fk (z0 ), f (z0 ))

< ε3 + ε3 + ε3 = ε.

Therefore F is equicontinuous at z0 .
Conversely, suppose that the conditions (a) and (b) of the theorem
hold. We need to show that F is normal. Let G̃ := {zn ∈ G : Re zn , Im zn

Advanced Complex Analysis by V. Ravichandran

are rational numbers}. Since G̃ is dense in G, for a given δ > 0 and
z ∈ G, there is an zn ∈ G̃ with |z − zn | < δ . For each n ≥ 1, let Xn =
{ f (zn ) : f ∈ F } ⊆ Ω. By (a), (Xn , d) is compact metric space. There-
fore, by Theorem 2.16, X = ∏∞ n=1 Xn is also a compact metric space.
For f ∈ F , define f ∈ X by f = ( f (z1 ), f (z2 ), . . .). Let h fk i be a se-
˜ ˜
quence in F so that h f˜k i is a sequence in X. Since X is compact, it is
sequentially compact. Therefore, there is a subsequence of h f˜k i which
converges to some ξ ∈ X. Denote the subsequence by h f˜k i itself so that
limk→∞ f˜k = ξ = {wn }. By Theorem 2.16, we have wn = limk→∞ fk (zn ).
We claim that h fk i converges to a function f ∈ C (G, Ω). Since
C (G, Ω) is complete, it is enough to show that { fk } is a Cauchy se-
quence in C (G, Ω). By Lemma 2.6, for any given compact subset K ⊆ G
and ε > 0, it suffices to find N such that for k, j ≥ N, sup{d( fk (z), f j (z)) :
z ∈ K} < ε. Let Kn be compact subset of G used in the definition of the
metric ρ in Definition 2.1.1. Since K is compact subset of G, K ⊆ Kn for
some n ≥ 1 by Theorem 2.2 and so, K ∩ ∂ G = 0. / Thus, R := d(K, ∂ G) >
0 (using the result that if A and B are disjoint sets in X with B closed and
A compact then d(A, B) > 0). Let K1 = {z ∈ G : d(z, K) ≤ R/2}. (If
zn ∈ K1 , z = lim zn , then d(zn , K) ≤ R/2 ⇒ d(z, K) ≤ R/2 ⇒ z ∈ K1 ).
Then K1 is compact and K ⊆ int K1 ⊆ K1 ⊂ G. Since F is equicontin-
uous at each point of G, it is equicontinuous on K1 , by Theorem 2.18.
Choose 0 < δ < R/2 such that

d( f (z), f (z0 )) < ε/3 (2.2.6)

for all f ∈ F and for all z, z0 ∈ K1 with |z − z0 | < δ . Let D = G̃ ∩ K1 =

{zn ∈ G̃ : zn ∈ K1 }.
We now claim that K ⊆ ∪ni=1 B(wi , δ ) where w1 , w2 , . . . , wn ∈ D.
32 2. THE SPACE OF CONTINUOUS FUNCTIONS

Letz ∈ K. Then ∃ zn ∈ G̃ with |zn −z| < δ . Since δ < R/2, d(zn , K) ≤
|zn − z| < δ < R/2 or, zn ∈ K1 , Thus, zn ∈ D. Therefore, {B(z j , δ ) : z j ∈
D} is an open covering of K. Since K is compact, there are w1 , w2 , . . . wn ∈
D such that K ⊆ ∪ni=1 B(wi , δ ). Since limk→∞ fk (wi ) exists, for i = 1, 2, . . . , n,
there is an integer N such that for j, k ≥ N,

d( fk (wi ), f j (wi )) < ε/3 for all i = 1, 2, . . . , n. (2.2.7)

Let z be arbitrary point in K. Let wi be such that |wi − z| < δ . If k, j ≥ N,

then equations (11) and (12) give

d( fk (z), f j (z)) ≤ d( fk (z), fk (wi )) + d( fk (wi ), f j (wi )) + d( f j (wi ), f j (z))

Advanced Complex Analysis by V. Ravichandran

< ε/3 + ε/3 + ε/3 = ε.

Therefore, sup{d( fk (z), f j (z)) : z ∈ K} ≤ ε.

2.3. The Space of Analytic Functions

Let G be an open subset of C and let H (G) ⊂ C (G, C) be the subset
consisting of all analytic functions. The following theorem shows that
H (G) is closed subspace of C (G, C).
Theorem 2.20 (Weierstrass Theorem). If fn ∈ H (G) and fn → f in
(k)
C (G, C), then f ∈ H (G). Also, for each k ≥ 1, fn → f (k) in H (G).
Proof. We can assume that G is a region; otherwise, consider each com-
ponent separately. Consider any triangle T inside G. Since T is a com-
pact subset of C, and fn → f in C (G, C), it follows that fn (z) → f (z)
uniformly on T and hence
Z Z
fn (z)dz → f (z)dz.
T T
R R
Since each fn is an analytic, T fn (z)dz = 0 and hence T f (z) = 0. Since
f is continuous, by Morera’s theorem, it follows that f is analytic in G.
(k)
By Corollary 2.10, it is enough to show that fn (z) → f (k) (z) uni-
formly on any closed disk D(a, r) inside G. Since α := d(D(a, r), ∂ G) >
0, with R := r + α/2, we have D(a, R) ⊂ G. Let γ be the circle |w − a| =
R. Since fn → f , it follows that fn (z) → f (z) uniformly on γ. Thus, for
ε > 0, there is a positive integer N such that

| fn (w) − f (w)| < (R − r)k+1 ε/(k!R)

 2.3. THE SPACE OF ANALYTIC FUNCTIONS 33

for all w on γ and for all n ≥ N. By Cauchy’s integral formula for the
derivatives, we have, for each z ∈ D(a, r),
k! fn (w) − f (w)
Z
(k)
fn (z) − f (k) (z) = dw.
2πi γ (w − z)k+1

Therefore, we have, for n ≥ N and z ∈ D(a, r),

(k) k! (R − r)k+1 ε/(k!R)

| fn (z) − f (k) (z)| ≤ 2πR = ε.
2π (R − r)k+1

Corollary 2.21. The space H (G) is a complete subspace of C (G, C).

Proof. The subspace H (G) is a closed subspace of the complete space

Advanced Complex Analysis by V. Ravichandran

C (G, C) and hence it is complete.
Corollary 2.22. If fn ∈ H (G) and ∑∞ n=1 f n → f in C (G, C), then f ∈
(k)
H (G) and ∑∞ n=1 f n → f
(k) in H (G).

Proof. Apply Weierstrass Theorem to the sequence of partial sums of

∑∞
n=1 f n .

Theorem 2.23 (Hurwitz’s Theorem). Let G be a region and D(a, r) ⊆

G. Let fn ∈ H (G), fn → f in H (G) and f 6≡ 0. If f (z) 6= 0 on ∂ D(a, r),
then there is a positive integer N such that fn and f have the same
number of zeros in D(a, r) for each n ≥ N.
Proof. Since f (z) 6= 0 on ∂ D(a, r), and | f (z)| is continuous and non-
vanishing on the compact set ∂ D(a, r),

ε := min{| f (z)| : z ∈ ∂ D(a, r)} > 0.

Since convergence in H (G) is the same as the uniform convergence on

compact subsets of G, fn → f implies that there is a positive integer N
such that, for n ≥ N, we have

| fn (z) − f (z)| < ε/2 (z ∈ ∂ D(a, r)).

For n ≥ N and z ∈ ∂ D(a, r), we have

| fn (z) − f (z)| < ε/2 < ε ≤ | f (z)| ≤ | fn (z)| + | f (z)|.

From the last equation, it is easy to see that fn 6= 0 on ∂ D(a, r). Hence,
by Rouche’s theorem, it follows that fn and f have the same number of
zeros in D(a, r).
34 2. THE SPACE OF CONTINUOUS FUNCTIONS

Remark 2.24. Hurwitz’s theroem need not hold if the function f van-
ishes at some point of the circle ∂ D(a, r). Let G = D(0, 2), fn (z) =
z − 1 + 1/n and f (z) = z − 1. Then fn (z) → f (z) uniformly on any com-
pact subset of G and hence fn → f . The function f vanishes on ∂ D.
Each fn has a zero in D but f has no zero in D.

Corollary 2.25. Let fn ∈ H (G) and fn → f in H (G). If each fn is

non-vanishing in G, then either f ≡ 0 or f is non-vanishing in G.
Proof. Assume that f 6≡ 0 and f (a) = 0 for some a ∈ G. Since f ∈
H (G), its zeros are isolated, and therefore, there a closed neighbour-
hood D(a, r) ⊂ G such that f has no zeros in D(a, r) \ {a}. By Hurwitz’s
theorem, there is integer N such that fn and f have same number of ze-

Advanced Complex Analysis by V. Ravichandran

ros in D(a, r). Since f has a zero in D(a, r), it follows that fn has a zero
in D(a, r), contradicting the hypothesis.
We call an analytic one-to-one function as univalent. For univalent
functions, we have the following corollary.
Corollary 2.26. Let fn ∈ H (G) and fn → f in H (G). If each fn is
univalent in G, then either f is a constant function or f is univalent in
G.
Proof. Let a, b ∈ G and a 6= b. Choose r > 0 such that D(a, r) ⊂ G
and b ∈/ D(a, r). Since fn → f in H (G), it follows that fn (z) → f (z)
uniformly on D(a, r) and hence fn (z) − fn (b) → f (z) − f (b) uniformly
on D(a, r). Clearly fn (z) − fn (b) 6= 0 for z ∈ D(a, r) and hence, by
Hurwitz’s theorem, either f (z) − f (b) ≡ 0 on G or f (z) − f (b) 6= 0 on
D(a, r). In the first case, f is a constant function. In the second case,
f (a) − f (b) 6= 0, or f is one-to-one in G.
Recall that S ⊂ H (D) is the class of all analytic functions f that
are univalent in D and normalized by the conditions f (0) = 0, f 0 (0) = 1.
Corollary 2.27. If fn ∈ S and fn → f in H (D), then f ∈ S .
Proof. Since fn → f in H (D), we have fn0 → f 0 in H (D), and hence
1 = fn0 (0) → f 0 (0) or f 0 (0) = 1. Thus, f is non-constant and hence,
by the previous corollary, f is one-to-one. Clearly f (0) = 0 and so
f ∈S.

Definition 2.28. A family F ⊂ C (G, C) is uniformly bounded on A ⊂ G

if there exists a non-negative real number M such that | f (z)| ≤ M for all
z ∈ A and for all f ∈ F . The family F is locally bounded on G if,
 2.3. THE SPACE OF ANALYTIC FUNCTIONS 35

for each a ∈ G, there is a r > 0 such that F is uniformly bounded on

D(a, r).
Clearly, any uniformly bounded family F ⊂ C (G, C) is locally bounded
and a locally bounded family is not necessary uniformly bounded. How-
ever, a locally bounded family is uniformly bounded on every compact
subset of G as shown in the following lemma:
Lemma 2.29. A family F ⊂ C (G, C) is locally bounded on G if and
only if it is uniformly bounded on each compact subset K ⊂ G.
Proof. If F is uniformly bounded, and a ∈ G, chooser > 0 such that
D(a, r) ⊂ G. Then F is uniformly bounded in D(a, r) and, in particular,
F is uniformly bounded in D(a, r). Thus, F is locally bounded on G.

Advanced Complex Analysis by V. Ravichandran

For the converse, assume that the family F is locally bounded on G.
Let K ⊂ G be a compact set. As K is compact and is disjoint from
closed set ∂ G, α := d(K, ∂ G) > 0. Then, for each a ∈ K, the disk
D(a, α) ⊂ G. Thus, the collection {D(a, α) : a ∈ K} is open cover of K
and, therefore, there are a1 , a2 , . . . , an ∈ K such that K ⊂ ∪nk=1 D(ak , α).
Since F is locally bounded, for each 1 ≤ k ≤ n, there are rk < α and
Mk ≥ 0 such that | f (z)| ≤ Mk for all f ∈ F and for all z ∈ D(ak , rk ). Let
M := max{M1 , . . . , Mn }. Let z ∈ K. Choose rk such that z ∈ D(ak , rk ).
Then, for f ∈ F , | f (z)| ≤ Mk ≤ M. Thus, F is uniformly bounded on
K.
Theorem 2.30 (Montel’s theorem). A family F ⊂ H (G) is normal if
and only if it is locally bounded on G.
Proof. Let F be normal. If F is not locally bounded on G, then, by
previous lemma, it is not uniformly bounded on some compact subset
K ⊂ G. Then, for each n ∈ N, there is a function fn ∈ F such that
sup{| fn (z)| : z ∈ K} ≥ n. (For, if there is no such fn , then | f (z)| < n
for all f ∈ F and for all z ∈ K.) Since F is normal and fn ∈ F , there
is a subsequence fnk → f in C (G, C) and, in particular, fnk (z) → f (z)
uniformly on K. Thus, there is an N such that, for nk ≥ N, | fnk (z) −
f (z)| < 1 for all z ∈ K. Since f is continuous on K, there is an M ≥ 0
such that | f (z)| ≤ M for all z ∈ K and therefore, for all nk ≥ N,

nk ≤ sup{| fn (z)| : z ∈ K}
≤ sup{| fnk (z) − f (z)| : z ∈ K} + sup{| f (z)| : z ∈ K} ≤ 1 + M.

This contradiction shows that F should be locally bounded on G.

36 2. THE SPACE OF CONTINUOUS FUNCTIONS

We use Arzela-Ascoli Theorem to prove the converse. Since F is

locally bounded on G, for each point a ∈ G, there is r > 0 such that
sup{| f (z)| : z ∈ D(a, r)} ≤ M. In particular, the set { f (a) : f ∈ F }
and hence its closure is bounded. Thus, being closed set, the closure of
{ f (a) : f ∈ F } is compact for each a ∈ G.
The proof is complete if we show that the family F is equicontinu-
ous at each point of G. Let a ∈ G and ε > 0 be given. Since F is locally
bounded on G, choose r > 0, M ≥ 0 such that | f (z)| ≤ M for all f ∈ F
and for all z ∈ D(a, r). Let γ be the circle |w − a| = r. Then, for f ∈ F
and for all z ∈ D(a, r/2), by Cauchy’s integral formula, we have

1 f (w) 1 f (w)
Z Z
f (z) − f (a) = dw − dw

Advanced Complex Analysis by V. Ravichandran

2πi γ w − z 2πi γ w − a
1 f (w)(z − a)
Z
= dw.
2πi γ (w − z)(w − a)

For w on γ and z ∈ D(a, r/2), we have |w − a| = r and |w − z| ≥ |w −

a| − |a − z| ≥ r − r/2 = r/2. Using these, we have

1 M|z − a|2πr 2M
| f (z) − f (a)| ≤ = |z − a|.
2π (r/2)r r

If we choose α = min{r/2, rε/(2M)}, then it follows that, for all f ∈ F ,

and |z − a| < α, we have | f (z) − f (a)| < ε.
Corollary 2.31. A family F ⊂ H (G) is compact if and only if it is
closed and locally bounded on G.
Proof. If F ⊂ H (G) is compact, then it is closed and normal. By
the previous theorem, it is locally bounded. Conversely, if F ⊂ H (G)
is closed and locally bounded, then F is normal and so F = F is
compact.

Exercises 2

2.1. Show that the family {p ∈ H (D) : p(0) = 1 and Re(p(z)) > 0 for z ∈
D} is normal.
2.2. Show that the family { f ∈ H (D) : f (z) = z/(z − a), |a| ≥ 1} is nor-
mal.
2.3. Is the family {n/((n + 1)z) : n ∈ N, 0 < |z| < 1} normal? (Hint: this
sequence converges uniformly to 1/z on every compact subset)
 2.4. THE RIEMANN MAPPING THEOREM 37

f ∈ H (D) : f (0) = 0 and | f 0 (z)|(1 − |z|2 ) ≤ 1 for all z ∈ D


2.4. Is the family
normal?
2.5. If F ⊂ H (G) is normal, then show that { f 0 : f ∈ F } is normal. Is
the converse true?

2.4. The Riemann Mapping Theorem

We begin with the following definition:
Definition 2.32. Let G1 , G2 ⊂ C. Then G1 is conformally equivalent to
G2 if there is an univalent function f : G1 → G2 such that f (G1 ) = G2 .

Theorem 2.33. Conformal equivalence is an equivalence relation.

Advanced Complex Analysis by V. Ravichandran

Proof. The identity maps shows that conformal equivalence is reflexive.
The symmetry follows since the inverse of bijective analytic function is
analytic. The transitivity follows as the composition of bijective analytic
functions is a bijective analytic function.

Example 2.34. By Liouville’s theorem, there is no analytic function

that maps C onto bounded subset of C and so no bounded domain is
conformally equivalent to C.

Example 2.35. Consider the slit domain S := C \ {z : z ≤ 0}; this is

a simply connected domain in C. The function f : S → C given by
√
f (z) = z is univalent (analytic and one-to-one) and it maps S onto
the right half-plane H = {z ∈ C : Re z > 0}. The mapping g : H → C
given by g(z) = (z − 1)/(z + 1) is univalent and maps H onto D. Thus,
√ √
the composite mapping f ◦ g, given by ( f ◦ g)(z) = ( z − 1)/( z + 1)
is univalent and maps S onto D. Thus, the slit domain S is conformally
equivalent to the disk D.
In general, there are univalent functions that map a proper simply
connected domain of C onto D. To prove this, we need the following
lemma.
Lemma 2.36. Let G be a simply connected domain in C which is not the
whole complex plane C and a ∈ G. Then there is an univalent function
f : G → D with f (a) = 0, f 0 (a) > 0.
Proof. Since G is not the whole complex plane C, choose b ∈ C \ G.
Since z − b 6= 0 for z ∈ G, and G is simply connected, there is analytic
function g : G → C satisfying g2 (z) = z − b. Clearly, the function g is
38 2. THE SPACE OF CONTINUOUS FUNCTIONS

non-constant and univalent in G; indeed, if g(z1 ) = ±g(z2 ), then z1 = z2 .

By open mapping theorem, g(G) is an open subset of C and therefore
there is α > 0 such that D(g(a), α) ⊂ g(G).
We claim that D(−g(a), α) ∩ g(G) = 0. / If D(−g(a), α) ∩ g(G) 6= 0, /
there is w ∈ G such that g(w) ∈ D(−g(a), α) or |g(w) + g(a)| ≤ α. This
shows that −g(w) ∈ D(g(a), α) ⊂ g(G) or −g(w) = g(z) for some z ∈ G.
Then, w = z and so g(w) = 0 or w = b. Since w ∈ G and b ∈ C \ G, a
contradiction.
The bilinear transformation T (z) = α/(z + g(a)) is univalent and
maps C \ D(−g(a), α) onto D. Hence, the mapping h(z) = T (g(z)) is
univalent and maps G into D. Since h is univalent, h0 (a) 6= 0 and hence
the mapping

Advanced Complex Analysis by V. Ravichandran

|h0 (a)| h(z) − h(a)
f (z) = (2.4.1)
h0 (a) 1 − h(a)h(z)
is well-defined and analytic in G. It has the desired properties. Since f is
the composition of univalent function h with the bilinear transformation
|h0 (a)| w − h(a)
S(z) = ,
h(a) 1 − h(a)w

it is univalent. Since h maps G into D and S maps D onto D, f = S ◦ h

maps G into D. Also f (a) = 0 and

|h0 (a)|
f 0 (a) = > 0. (2.4.2)
1 − |h(a)|2

Theorem 2.37 (The Riemann Mapping Theorem). Any simply-

connected domain G in C which is not the whole complex plane C is
conformally equivalent to the unit disk D. In other words, there is an
univalent function f : G → D such that f (G) = D. If, for some a ∈ G,
f (a) = 0 and f 0 (a) > 0, then the mapping f is unique.
Proof. Fix a ∈ G. Consider the class F of all univalent function g :
G → D such that g(G) ⊂ D, g(a) = 0 and g0 (a) > 0. By previous lemma,
F 6= 0.
/ We show that F = F ∪ {0}. Since there is at least one g ∈ F ,
and g(z)/n converges uniformly to 0 on any compact subset of G, 0 ∈ F
and so F ∪ {0} ⊂ F . To prove the other way inclusion, let g ∈ F and
let gn ∈ F and gn → g in H (G). By Corollary 2.26, it follows that
either g is a constant function or g is univalent. If g is constant, then
g ≡ 0. For, gn → g and gn (a) = 0 implies g(a) = 0. If g is univalent,
 2.4. THE RIEMANN MAPPING THEOREM 39

then g0 (a) 6= 0. But g0n (a) > 0 implies that g0 (a) ≥ 0 and so g0 (a) > 0.
Thus, g ∈ F . In any case, g ∈ F ∪ {0} and this proves F ⊂ F ∪ {0}.
We shall now complete the proof by showing that there is a function
f ∈ F such that f (G) = D. If gn → g in H (G), then by Weierstrass
Theorem (Theorem 2.20), we have g0n → g and in particular g0n (a) →
g(a). This show that the function F : H (G) → R defined by F(g) =
g0 (a) is continuous. Since F is uniformly bounded on G, it is locally
bounded on G. By Montel’s Theorem (Theorem 2.30), it follows that
F is normal and so F is compact. Since F is continuous real valued
function on a compact set F , it has its maximum in F . Let f ∈ F
be the function such that f 0 (a) = F( f ) ≥ F(g) = g0 (a) for all g ∈ F ; in
other words, f is the function in F having maximum possible derivative

Advanced Complex Analysis by V. Ravichandran

at z = a. Since F 6= 0,/ choose a function g1 ∈ F . Since g01 (a) > 0, we
must have f 0 (a) > 0. Since F = F ∪ {0}, it follows that f ∈ F . The
proof is complete if we show f (G) = D.
Suppose on the contrary that f (G) ( D. Let w ∈ D \ f (G) and con-
sider
f (z) − w
g(z) = φw ( f (z)) = (2.4.3)
1 − w f (z)
where the bilinear transformation φw (z) = (z−w)/(1−wz) maps D onto
D. Clearly, the mapping g is analytic, univalent and non-vanishing in
G and maps G into D. Since G is simply connected, there is an ana-
lytic function h : G → C such that h(z)2 = g(z). Since the mapping g is
univalent in G and maps G into D, it follows that the mapping h is also
univalent in G and maps G into D. Now consider the mapping g : G → C
defined by the equation (see equation (2.4.1))
|h0 (a)| h(z) − h(a)
g(z) = .
h0 (a) 1 − h(a)h(z)

This function g is also univalent (being composition of a bilinear trans-

formation with a univalent function) and also maps G into D. Since
h(a)2 = g(a) = −w, and, from (2.4.3),

2h(a)h0 (a) = g0 (a) = f 0 (a)(1 − |w|2 ),

we see that, using equation (2.4.2) (with f replaced by g),

|h0 (a)| f 0 (a)(1 − |w|2 ) 1 (1 + |w|)

g0 (a) = 2
= p = f 0 (a) p > f 0 (a)
1 − |h(a)| 2 |w| 1 − |w| 2 |w|
40 2. THE SPACE OF CONTINUOUS FUNCTIONS

as |w| < 1. This contradicts the choice of f and hence f (G) = D.

To prove the uniqueness, let f and g be two univalent functions from
G onto D with positive derivatives at z = a that map a ∈ G to the origin.
Then, f ◦ g−1 maps D onto D, ( f ◦ g−1 )(0) = f (a) = 0. Thus, ( f ◦
g−1 )(w) = cw for some constant c with |c| = 1. Here we have used the
fact that the only univalent function that maps D onto D with f (0) = α
are of the form c(z − α)/(1 − αz) where |c| = 1. Since g is onto, writing
w = g(z), we get f (z) = cg(z). Since |c| = 1 and c = f 0 (a)/g0 (a) > 0,
we have c = 1 and hence f = g.

2.5. The Weierstrass Factorization Theorem

Advanced Complex Analysis by V. Ravichandran

An analytic functions with exactly a1 , . . . , an as its zeros of multi-
plicities m1 , . . . , mn respectively is (z − a1 )m1 · · · (z − an )mn . Any multi-
ple of this polynomial with exp( f (z)) for some analytic function f has
a1 , . . . , an as its zeros of multiplicities m1 , . . . , mn respectively. Let hak i
be a sequence of complex number with no limit point. The question of
interest now is to find an analytic function with exactly ak as its zeros.
Before we answer, we need to discuss convergence of infinite product.
Definition 2.38. Let zk ∈ C. The infinite product ∏∞ k=1 (1 + zk ) con-
n
verges to z if the sequence of partial product ∏k=1 (1 + zk ) converges to
z.
The infinite product ∏∞ k=1 (1 + zk ) converges if anyone of the term
(1 + zk ) is zero; in this case, the convergence does not depend on the
other terms of the product.
If none of the term is zero and if the product ∏∞k=1 (1 + zk ) converges
to a non-zero number z, then

∏nk=1 (1 + zk ) z
1 + zn = → =1
∏n−1
k=1 (1 + zk ) z

and so zn → 0. This is not true if infinite product ∏∞

k=1 (1 + zk ) converges
to 0. For example, let zk = a − 1, |a| < 1. Then ∏∞ k=1 (1 + zk ) converges
to 0 but zn 6→ 0.
Theorem 2.39. Let zk ∈ C satisfy Re zk > −1. The infinite product
∏∞ ∞
k=1 (1+zk ) converges to a non-zero number if and only if ∑k=1 log(1+
zk ) converges.
 2.5. THE WEIERSTRASS FACTORIZATION THEOREM 41

Proof. If ∑∞
k=1 log(1 + zk ) converges to `, then the sequence of partial
n
sums ∑k=1 log(1 + zk ) converges to ` and hence
n n
∏ (1 + zk ) = e∑k=1 log(1+zk ) → e`.
k=1

To prove the converse part, let us assume that the infinite product
∏∞
k=1 (1 + zk ) converges to a non-zero number z = reiθ . The logarithm
of a complex number is defined by log z = log |z| + i arg z. Let l(z) be the
branch of logarithm defined by l(z) = log |z| + i arg z where the argument
arg z is restricted by θ − π < arg z ≤ θ + π. Let pn = ∏nk=1 (1 + zk ) and
sn = ∑nk=1 log(1 + zk ) . Since esn = pn , we have sn = l(pn ) + 2πikn , kn

Advanced Complex Analysis by V. Ravichandran

is an integer. Since zn → 1, we have sn − sn−1 = log(1 + zn ) → 0. The
continuity of the branch of logarithm shows that

l(pn ) − l(pn−1 ) → l(z) − l(z) = 0.

Therefore,

kn − kn−1 = (2πi)−1 (sn − sn−1 − (l(pn ) − l(pn−1 ))) → 0.

Since kn ’s are integers, we have positive integer N and an integer k such

that kn = k for n ≥ N. Thus, sn = l(pn ) + 2πik for n ≥ N. Hence sn →
l(z) + 2πik. Thus, ∑∞ k=1 log(1 + zk ) converges.

Remark 2.40. If we choose arg(1 + zk ) arbitrarily to define log(1 +

zk ), then the series ∑∞k=1 log(1 + zk ) may not converge. For example,
if zk = 0, and log(1 + zk ) = 2πi, then the infinite product ∏∞
k=1 (1 + zk )
∞
converges to 1 but ∑k=1 log(1 + zk ) des not converge.
If an infinite series converges absolutely, then the series converges.
If we take zk = (−1)k − 1, then infinite product ∏∞ k=1 |1 + zk | converges
∞
to 1 but the infinite product ∏k=1 (1 + zk ) does not converge. If we wish
absolute convergence of the infinite product implies the convergence of
the infinite product, then we need to define absoulte convergence in a dif
and only if ernt way. Motivated by the previous theroem, we define the
absolute convergence of the infinite product as follows.
Definition 2.41. The infinite product ∏∞ k=1 (1+zk ) converges absolutely
∞
if and only if the series ∑k=1 log(1 + zk ) converges absolutely.
42 2. THE SPACE OF CONTINUOUS FUNCTIONS

The infinite product ∏∞

k=1 (1 + zk ) converges absolutely, then the se-
∞
ries ∑k=1 log(1 + zk ) converges absolutely and hence the series con-
verges. By the previous theroem, it converges to a non-zero complex
number. A condition for absolute convergence without use of logarithm
will be useful. Before proving such a conditon, we prove the following
lemmas.
Lemma 2.42. For |z| < 1/2, we have

1 3
|z| ≤ | log(1 + z)| ≤ |z|.
2 2
Proof. For |z| < 1/2, we have

Advanced Complex Analysis by V. Ravichandran

∞ ∞
(−1)n−1 n
| log(1 + z)| − |z| ≤ | log(1 + z) − z| = ∑ n z ≤ ∑ 1n |z|n
n=2 n=2
∞ ∞
|z| |z| |z|
≤ ∑ |z|n−1 ≤ ∑ ( 12 )n−1 ≤ .
2 n=2 2 n=2 2

Since |x − a| ≤ r is equivalent to a − r ≤ x ≤ a + r, we see that

1 |z| |z| 3
|z| = |z| − ≤ | log(1 + z)| ≤ |z| + = |z|.
2 2 2 2

Lemma 2.43. Let Re zk > −1. The series ∑∞ k=1 log(1 + zk ) converges
absolutely if and only if the series ∑∞ z
k=1 k converges absolutely.
Proof. The convergence of ∑∞ ∞
k=1 log(1 + zk ) or ∑k=1 zk immediately im-
plies that zk → 0 and so there is an integer N such that |zk | < 1/2 for
k ≥ N and, by Lemma 2.42,
1 3
|zk | ≤ | log(1 + zk )| ≤ |zk | (k ≥ N).
2 2
The desired result follows as both ∑∞ ∞
k=1 ak and ∑k=1 bk converge or di-
verge simultaneously whenever αbk ≤ ak ≤ β bk for k ≥ N.
Theorem 2.44. Let Re zk > −1. The series ∏∞ k=1 (1 + zk ) converges
∞
absolutely if and only if the series ∑k=1 zk converges absolutely.
Proof. By definition, the infinite product ∏∞ k=1 (1 + zk ) converges ab-
solutely if and only if the series ∑∞
k=1 log(1 + zk ) converges absolutely,
and, by the previous lemma, the later series converges absolutely if and
only if the series ∑∞
k=1 zk converges absolutely.
 2.5. THE WEIERSTRASS FACTORIZATION THEOREM 43

Lemma 2.45. Let (X, d) be a space and let f , fn : X → C be continuous.

If fn (x) → f (x) converges uniformly for x ∈ X and, for some constant a,
Re f (x) < a for all x ∈ X, then e fn (x) → e f (x) uniformly for x ∈ X.
Proof. Let ε > 0 be given. Since limw→0 ew = 1, we have α > 0 such
that |ew − 1| < e−a ε whenever |w| < α. Choose an integer N such that,
for all n 6= N, | fn (x) − f (x)| < α for all x ∈ X. Hence, for n ≥ N and for
all x ∈ X, we have
|e fn (x)− f (x) − 1| < e−a ε
and, therefore,

|e fn (x) − e f (x) | < e−a |e f (x) |ε = e−a eRe f (x) ε ≤ e−a ea ε = ε.

Advanced Complex Analysis by V. Ravichandran

Lemma 2.46. Let (X, d) be a compact metric space and let fn : X →
C be continuous. If ∑∞ k=1 | f k (x)| converges uniformly for x ∈ X, then
∞
f (x) = ∏k=1 (1 + fk (x)) converges absolutely as well as uniformly for
x ∈ X. In addition, f (x) = 0 if and only if there is an integer N such that
fk (x) = −1 for some k with 1 ≤ k ≤ N.
Proof. Let 0 < ε ≤ 3/4. Since ∑∞
k=1 | f k (x)| converges uniformly for
x ∈ X, there is an N such that
∞
2
∑ | fk (x)| < ε
k=N+1 3

for all x ∈ X and so, for k ≥ N + 1 and for all x ∈ X, we have

2 1
| fk (x)| < ε ≤ .
3 2
This show that, for k ≥ N + 1 and for all x ∈ X,
1
Re(1 + fk (x)) = 1 − | fk (x)| >
2
and, by Lemma 2.42, for k ≥ N + 1 and for all x ∈ X,
3
| log(1 + fk (x))| ≤ | fk (x)|. (2.5.1)
2
Since ∑∞ ∞
k=1 | f k (x)| converges in X, the above inequality shows that ∑k=N+1 | log(1
fk (x))| converges and hence, ∑∞ k=N+1 log(1 + f k (x)) converges abso-
lutely. By the definition of absolute convergence of infinite product, it
44 2. THE SPACE OF CONTINUOUS FUNCTIONS

follows that the infinite product ∏∞ ∞

k=N+1 (1 + f k (x)) and hence ∏k=1 (1 +
fk (x)) converges absolutely in X.
Let f (x) = ∏∞k=1 (1 + f k (x)). We now show that the convergence is
uniform in X. The inequality (2.5.1) shows that
∞ ∞
3 ∞
∑ log(1 + fk (x)) ≤ ∑ | log(1 + fk (x))| ≤ ∑ | fk (x)| < ε
k=N+1 k=N+1 2 k=N+1
and this show that ∑∞
k=N+1 log(1 + f k (x)) converges uniformly in X. Let
h(x) = ∑∞ k=N+1 log(1 + fk (x)). Since h is bounded and X is compact,
there is a real number a such that Re h(x) < a. Applying Lemma 2.45 to
the sequence of partial sums of ∑∞ k=N+1 log(1 + f k (x)), we see that

Advanced Complex Analysis by V. Ravichandran

∞
∑k=N+1 log(1+ fk (x))
∏∞
k=N+1 (1 + f k (x)) = e

converges uniformly to eh(x)
. Hence, ∏∞ k=1 (1 + f k (x)) converges uni-

N h(x)
formly to ∏k=1 (1 + fk (x)) e . The remaining part of the proof fol-
lows since f (x) = ∏N

h(x)
k=1 (1 + f k (x)) e .
Theorem 2.47. Let G be a region in C and let h fn i be a sequence in
H (G) such that no fn is identically zero. If ∑∞ k=1 | f k (z) − 1| converges
uniformly on compact subsets of G, then ∏k=1 fk (z) converges in H (G)
∞

to an analytic function f (z). If z = a is a zero of the function f , then there

is an integer N such that z = a is a zero of the functions f1 , . . . , fN and
the multiplicity of the zero of f at z = a is the sum of the multiplicities
of the zeros of these functions f1 , · · · , fN at z = a.
Proof. Since ∑∞ k=1 | f k (z) − 1| converges uniformly on compact subsets
of G, Lemma 2.46 shows that ∏∞ k=1 f k (z) converges uniformly on com-
pact subsets of G. Thus, the infinite product ∏∞ k=1 f k (z) converges in
H (G).
Choose R > 0 such that D(a, R) ⊂ G. By Lemma 2.46, f (z) =
f1 (z) · · · fn (z)h(z) where h is a non-vanishing analytic function in D(a, R).
The second half of the theorem follows from this.

Definition 2.48. For each integer p ≥ 0, the function E p : C → C de-

fined by
z2 zp
E0 (z) = 1 − z, E p (z) = (1 − z)ez+ 2 +···+ p (p ≥ 1)
is called an elementary factor. It can be written as
p zk ∞ zk
E p (z) = elog(1−z)+∑k=1 k = e− ∑k=p+1 k .
 2.5. THE WEIERSTRASS FACTORIZATION THEOREM 45

Lemma 2.49. The elementary function E p satisfies |E p (z) − 1| ≤ |z| p+1

for |z| ≤ 1.
Proof. For p = 0, the result is trivial and so assume p ≥ 1. Since
E p (0) = 1, write E p (z) = 1 + ∑∞ k
k=1 ak z . Then

∞ ∞
zp z2 zp
∑ kak zk−1 = E p0 (z) = −E p(z) ∑ zk−1 = − E p (z) = −z p ez+ 2 +···+ p .
k=1 k=p+1 1−z

The right-hand side function is of the form −z p (1 + c1 z + c2 z2 + · · · )

where ci > 0 and, hence, by comparing the coefficients of the last equa-
tion, we see that a1 = a2 = · · · = ak = 0 and ak+1 , ak+2 , . . . < 0. Thus,

Advanced Complex Analysis by V. Ravichandran

E p (z) = 1 + ∑∞ k ∞
k=p+1 ak z . Since E p (1) = 0, we have ∑k=p+1 |ak | =
− ∑∞ k=p+1 ak = 1. Therefore, for |z| ≤ 1, we have

∞ ∞ ∞
|E p (z)−1| ≤ ∑ |ak ||z|k ≤ |z| p+1 ∑ |ak ||z|k−p−1 ≤ |z| p+1 ∑ |ak | ≤ |z| p+
k=p+1 k=p+1 k=p+1

Theorem 2.50. Let han i be a sequence of non-zero complex numbers

with limn→∞ |an | = ∞. Let hpn i be a sequence of non-negative integers
pk +1
such that ∑∞ k=1 (r/|ak |) converges for each r > 0 and this condition
is satisfied if pk = k −1. Then the infinite product h(z) = ∏∞ k=1 E pk (z/ak )
converges in H (C). This entire function h has zeros exactly at ak and
the multiplicity of a zero is the same as the number of times it occur in
the sequence han i.
Proof. Since any arbitrary compact subset of C is contained in some
disk |z| ≤ r, it is enough to show uniform convergence of the infinite
product on each disk |z| ≤ r. Fix r > 0. Since limn→∞ |an | = ∞, there
is a positive integer N such that |ak | > r for k ≥ N. Let |z| ≤ r. Then
|z/ak | ≤ 1 for k ≥ N and so, by the previous lemma, we have

|E pk (z/ak ) − 1| ≤ |z/ak | pk +1 ≤ (r/|ak |) pk +1 for k ≥ N.

Thus, the series ∑∞k=N |E pk (z/ak ) − 1|, being dominated by convergent

series of number ∑∞ pk +1 , is uniformly convergent in |z| ≤ r.
k=N (r/|ak |)
Thus, by Theorem 2.47, the infinite product ∏∞ k=N E pk (z/ak ) converges
in H (C). Thus, ∏k=1 E pk (z/ak ) converges in H (C).
∞
k
To show that ∑∞ k=1 (r/|ak |) converges for each r > 0, first choose a
positive integer N such that |ak | > 2r for k ≥ N. Then (r/|ak |) < 1/2 for
46 2. THE SPACE OF CONTINUOUS FUNCTIONS

k ≥ N and hence
∞ ∞
∑ (r/|ak |)k < ∑ (1/2)k < ∞.
k=N k=N

k
Hence, ∑∞
k=1 (r/|ak |) converges for each r > 0.

Theorem 2.51 (The Weierstrass Factorization Theorem). Let han i

be the non-zero zeros, repeated according to multiplicity, of an entire
function f and suppose f has a zero at z = 0 of order m ≥ 0. Then there
is an entire function g and a sequence of non-negative integers hpn i such
that

Advanced Complex Analysis by V. Ravichandran

∞
m g(z)
f (z) = z e ∏ E pk (z/ak ).
k=1

Proof. Since an ’s are zeros of the analytic function f , {an } has no limit
points in C and, therefore, we must have limn→∞ |an | = ∞. Using Theo-
rem 2.50, choose non-negative integers pk such that the infinite product
h(z) = ∏∞ k=1 E pk (z/ak ) converges in H (C). The function h has the same
non-zero zeros as f with the same multiplicities. Thus, it follows that
f (z)/(zm h(z)) has removable singularities at z = 0 and z = an for n ≥ 1.
Thus, f (z)/(zm h(z)) is an non-vanishing entire function. Since C is sim-
ply connected and non-vanishing functions has analytic logarithm, there
is an entire function g such that f (z)/(zm h(z)) = eg(z) . This proves that
∞
f (z) = zm eg(z) h(z) = zm eg(z) ∏ E pk (z/ak ).
k=1
 3. RUNGE ’ S THEOREM

3.1. Runge’s theorem

Theorem 3.1. Let K be a compact subset of the region G then there are
straight line segments γ1 , γ2 , . . ., γn in G \ K such that for every function
f ∈ H (G)

Advanced Complex Analysis by V. Ravichandran

n
1 f (w)
Z
f (z) = ∑ dw
k=1 2πi γk w−z
for all z ∈ K. The line segments form a finite number of closed polygons.
Proof. By enlarging K, we can assume that K = int K. Let 0 < δ <
d(K, C \ G)/2. Place a ‘grid’ of horizontal and vertical lines in the
plane such that consecutive lines are less than a distance δ apart. Let
R1 , R2 , . . . Rm be the rectangles that intersects K. Since K is compact,
there are only finitely many. Let ∂ R j be the boundary of R j , 1 ≤ j ≤ m,
considered as a polygon with counter-clockwise direction.
Since the largest distance between two points√in a rectangle is the
diagonal distance, for z ∈ R j , we have d(z, K) ≤ 2δ and this implies
2δ < d(K, C \ G). Therefore z 6∈ C \ G which implies R j ⊆ G. Some
sides of a rectangle R j can be common with a side of some other Ri . Let
σ j , σi be the sides of R j and Ri that are common. Then σ j , σi are on
opposite directions. For any continuous function φ on {σ j } = {σi },
Z Z
φ+ φ = 0.
σj σi
Let γ1 , γ2 , . . ., γn are the directed line segments that constitute a side of
exactly one rectangle R j , 1 ≤ j ≤ m, then
n Z m Z
∑ φ= ∑ φ (3.1.1)
k=1 γk j=1 ∂ R j

for every continuous function on mj=1 ∂ R j .

S

We claim that γk ⊆ G \ K. If γk intersects K, then γk is a side of two

rectangles that meet K. But γk is choosen such that it is a side of just one
rectangle R j , 1 ≤ j ≤ m.
47
48 3. RUNGE’S THEOREM
Sm
If z ∈ K \ j=1 ∂ R j , then
 
1 f (w)
φ (w) =
2πi w−z

for f ∈ H (G). From equation (3.1.1), we

Sm
is continuous on j=1 ∂ R j
get
m n
1 f (w) 1 f (w)
Z Z
∑ 2πi dw = ∑ dw (3.1.2)
j=1 ∂Rj w−z k=1 2πi γk w−z

Since z belongs to the interior of exactly one R j ,

Advanced Complex Analysis by V. Ravichandran

(
1 f (w) f (z) if z ∈ int Ri ,
Z
dw =
2πi ∂ Ri w − z 0 if z 6∈ int Ri .

and by equation (3.1.2),

n
1 f (w)
Z
f (z) = ∑ dw (3.1.3)
k=1 2πi γk w−z

The functions on both sides of equation (3.1.3) are continuous and equal-
ity in equation (3.1.3) holds on K \ mj=1 ∂ R j which is dense in K, by
S

continuity, it holds for z ∈ K. That γ1 , γ2 , . . ., γn forms a finite number

of polygons is trivial.
Lemma 3.2. Let γ be a rectifiable curve and let K be a compact set
such that K ∪ {γ} = 0.
/ If f is a continuous function on {γ} and ε > 0,
then there is a rational function R(z) having all its poles on {γ} and
such that
f (w)
Z
dw − R(z) < ε
γ w−z
for all z ∈ K.
Proof. Since K and {γ} are disjoint, there is a number r with 0 < r <
d(K, {γ}). If γ is defined on [0, 1], then for 0 ≤ s,t ≤ 1 and z ∈ K,

f (γ(t)) f (γ(s)) 1
− ≤ 2 f (γ(t))γ(s) − f (γ(s))γ(t) − z[ f (γ(t)) − f (γ(s))]
γ(t) − z γ(s) − z r
1
= 2 f (γ(t))[γ(s) − γ(t)] − [ f (γ(t)) − f (γ(s))]γ(t)
r
 3.1. RUNGE’S THEOREM 49

− z[ f (γ(t)) − f (γ(s))]

1
≤ 2 | f (γ(t))||γ(s) − γ(t)| + | f (γ(t)) − f (γ(s))||γ(t)|
r

+ |z|| f (γ(t)) − f (γ(s))| (3.1.4)

Since K is compact, there exists c1 > 0 such that |z| ≤ c1 for all z ∈ K.
Since γ is constant on [0, 1], there exists c2 > 0 such that |γ(t)| ≤ c2 for
all 0 ≤ t ≤ 1. Since f is also continuous | f (γ(t))| ≤ c3 for c3 > 0, for all
0 ≤ t ≤ 1. Let c = max(c1 , c2 , c3 ). Then |z| ≤ c, for all z ∈ K, |γ(t)| ≤ c
for all 0 ≤ t ≤ 1, | f (γ(t))| ≤ 1, 0 ≤ t ≤ 1. Using this in equation (3.1.4)

Advanced Complex Analysis by V. Ravichandran

, we get

f (γ(t)) f (γ(s)) c 2c
− ≤ 2 γ(t) − γ(s) + 2 f (γ(s)) − f (γ(t))
γ(t) − z γ(s) − z r r
(3.1.5)
Since, both γ and f ◦ γ are uniformly continuous on [0, 1], for a given
ε > 0, there exists δ > 0 such that |s − t| < δ implies

γ 2ε γ 2ε
|γ(s) − γ(t)| < , | f (γ(s)) − f (γ(t))| < . (3.1.6)
2cV (γ) 4cV (γ)

Choose a partition {0 = t0 < t1 < · · · < tn = 1} such that max{ti −ti−1 } <
δ . Then from (3.1.5) and (3.1.6), we have

f (γ(t)) f (γ(t j )) c γ 2ε 2c γ 2 ε ε
− ≤ 2· + 2· = (3.1.7)
γ(t) − z γ(t j ) − z r 2cV (γ) r 4cV (γ) V (γ)

for t j−1 ≤ t ≤ t j , 1 ≤ j ≤ n and z ∈ K. Define the function R(z) by

n γ(t j ) − γ(t j−1 )
R(z) = ∑ f (γ(t j−1)) · γ(t j−1 ) − z
.
j=1

Then R is a rational function whose poles z = γ(t0 ), . . . , γ(tn−1 ) are in

{γ}. Now for z ∈ K,

f (w)
Z
dz − R(z)
γ w−z
Z 1
f (γ(t)) 0
= γ (t)dt − R(z)
0 γ(t) − z
50 3. RUNGE’S THEOREM

n Z tj n γ(t j ) − γ(t j−1 )

f (γ(t)) 0
= ∑ γ (t)dt − ∑ f (γ(t j−1 )) ·
j=1 t j−1 γ(t) − z j=1 γ(t j−1 ) − z
n Z tj 
f (γ(t))γ 0 (t) f (γ(t j−1 ))γ 0 (t)

= ∑ − dt
j=1 t j−1 γ(t) − z γ(t j−1 ) − z
n Z tj  
f (γ(t)) f (γ(t j−1 ))
= ∑ − dγ(t)
j=1 t j−1 γ(t) − z γ(t j−1 ) − z
n Z tj Z 1
ε ε
≤ ∑ |dγ(t)| = |dγ(t)|
V (γ) j=1 t j−1 V (γ) 0
ε
= ·V (γ) = ε.

Advanced Complex Analysis by V. Ravichandran

V (γ)

Let K be a compact subset of C on C(G, C), define

ρ( f , g) = sup{| f (z) − g(z)| : z ∈ K}

for f , g ∈ C(G, C). Then ρ( fn , f ) → 0 if and only if fn → f uniformly

on K. Therefore, a Cauchy sequence in C(G, C) is covergent and so
C(G, C) is a complete metric space.
Let K be a compact subset of C and E be a subset of C∞ \ K that
meets each component of C∞ \ K. Let B(E) be the set of all functions
f ∈ C(G, C) such that there is a sequence of rational functions {Rn } with
poles in E and f = u-lim Rn on K.
The set B(E) is an algebra if f , g ∈ B(E), α ∈ C, then α f , f + g,
f g ∈ B(E).
Lemma 3.3. The set B(E) is a closed subalgebra of C(G, C) that con-
tains every rational function with a pole in E.
Proof. Let Rn , Sn be rational functions with poles in E such that f =
u-lim Rn , g = u-lim Sn . Then α f = u-lim αRn , f + g = u-lim(Rn + Sn ),
f g = u-lim Rn Sn . Since αRn , Rn + Sn , Rn Sn has poles in E, α f , f + g,
f g ∈ B(E).
Note that, f g = u-lim Rn Sn as all these f , g, Rn , Sn are bounded,

|Rn Sn − f g| = |(Rn − f )Sn + f (Sn − g)|

≤ |Sn ||Rn − f | + | f ||Sn − g|
= M1 |Rn − f | + M2 |Sn − g|
→ 0 + 0 = 0.
 3.1. RUNGE’S THEOREM 51

Lemma 3.4. Let V and U be open subsets of C and V ⊆ U, ∂V ∩U = 0.

/
If H is a component of U and H ∩V 6= 0,
/ then H ⊂ V .
Proof. Let a ∈ H ∩V and let G be the component of V such that a ∈ G.
Then H ∪ G is connected (as, a ∈ H ∩ G) and H ⊆ U, G ⊆ V ⊆ U and so
H ∪ G ⊆ U. Since, H is a component of U, G ⊆ H. Thus, ∂ G ⊆ ∂ H and
by the hypothesis that ∂V ∩ V = 0,
/ ∂ G ∩ H ⊆ ∂V ∩ H ⊆ ∂V ∩ U = 0. /
This shows that

H \ G = H ∩ G0 = H ∩ [(C \ G) ∪ ∂ G]
= [(H ∩ (C \ G)] ∪ [H ∩ ∂ G]
= H ∩ (C \ G) ∪ 0/

Advanced Complex Analysis by V. Ravichandran

= H ∩ (C \ G).

Since U is open, H is open. Since C \ G is open, H \ G = H ∩ (C \ G) is

open in H. Since G is open implies H \ G = H ∩ (C \ G) is closed in H.
Therefore, H \ G is both open and closed in H. Since H is connected;
H \ G = 0/ or H. Since, a ∈ H ∩ G, a 6∈ H \ G. Thus, H \ G 6= H which
implies H \ G = 0.
/ Thus, G ⊆ H, H \ G = 0/ implies H = G ⊆ V .

Lemma 3.5. If a ∈ C \ K, then (z − a)−1 ∈ B(E).

Proof. Case 1. ∞ 6∈ E.
Let U = C \ K, and let V = {a ∈ C : (z − a)−1 ∈ B(E)}. Then, with
Rn (z) = (z − a)−1 , we see that (z − a)−1 = lim Rn (z), Rn (z) has a pole
in E if a ∈ E. Thus, if a ∈ E, then a ∈ V or E ⊆ V . If a ∈ V , then
(z − a)−1 = lim Rn (z) where the poles of Rn (z) are in E. Since a is a
pole of Rn for large n, a ∈ E ⊆ C∞ \K. Since ∞ 6∈ E, E ⊆ C \ K = U.
Thus V ⊆ U, i.e., E ⊆ V ⊆ U.
We claim that if a ∈ V , and |b − a| < d(a, K), then b ∈ V . Since,
|b − a| < d(a, K), there exists 0 < r < 1 such that |b − a| < rd(a, K) or,
|b − a| < r|z − a| for all z ∈ K. Now, if |b − a| < r|z − a|, then

1 1 1 1 ∞
(b − a)n
= = · = ∑ n+1
z − b z − a − (b − a) z − a 1 − b−a
z−a n=1 (z − a)

and

(b − a)n 1 ∞
n+1
≤ · rn+1 , ∑ rn+1 is convergent as r < 1.
(z − a) |b − a| n=0
52 3. RUNGE’S THEOREM

By Weierstrass M-test, the convergence of

∞
(b − a)n 1
∑ (z − a)n+1 = z − b
n=1
is uniform on K. Let
n
(b − a)k
Rn (z) = ∑ k+1
.
k=1 (z − a)

Then Rn → (z−b)−1 uniformly on K. Since, a ∈ V therefore, (z−a)−1 ∈

B(E). Since, B(E) is an algebra, Rn ∈ B(E). Since, B(E) is a closed
subalgebra, (z − b)−1 ∈ B(E). Therefore, b ∈ V . Our claim shows that
if a ∈ V , B(a, d(a, K)) ⊆ V and so V is open.

Advanced Complex Analysis by V. Ravichandran

Let b ∈ ∂V . Then there exists {an } ⊆ V with b = lim an . Since
b 6∈ V , then |b − an | ≥ d(an , K) for all n (using our claim). Let n → ∞.
Then, 0 ≥ d(b, K) which implies d(b, K) = 0. Therefore, b ∈ K = K and
hence, b 6∈ C \ K = U. Thus, ∂V ∩U = 0. /
If H is a component of U = C \ K, then H ∩ E 6= 0/ as E meets ever
component of C∞ \ K, ∞ 6∈ E. Since, E ⊂ V , H ∩V = 0. / By Lemma 3.4,
we have H ⊆ V . Since, H is an arbitrary component of U, U ⊆ V . Since,
V ⊆ U, we have U = V . Thus, if a ∈ C \ K = U, a ∈ V or (z − a)−1 ∈
B(E).
Case 2. ∞ ∈ E.
Let d be the metric on C∞ . Choose a0 in the unbounded component
of C \ K such that d(a0 , ∞) ≤ d(∞, K)/2 and |a0 | > 2 max{|z| : z ∈ K}.
Let E0 = (E \ {∞}) ∪{a0 }. Since E meets every component of C∞ \ K
and a0 belongs to the unbounded component of K, E0 also meets every
component of C∞ \ K and ∞ 6∈ E0 .
If a ∈ C \ K, then by Case 1, (z − a)−1 ∈ B(E0 ). If (z − a0 )−1 ∈
B(E) then there are rational functions Rn (z) with poles in E such that
(z − a0 )−1 = lim Rn (z). Let Rn (z) = R0n (z) · (z − a0 )−m , where m ≥ 0.
Since, if f ∈ B(E0 ), there exists Sn (z), rational functions with poles in
E0 such that f = lim Sn . Le t Sn (z) = Sn0 (z) · (z − z0 )−m , where m ≥ 0.
Then f = lim Sn = lim(Sn0 (z)Rn (z)m ). Since Sn0 and Rn have poles in E,
f ∈ B(E). That is, B(E0 ) ⊆ B(E) and so, (z − a)−1 B(E) for a ∈ C \ K.
We now show that (z − a0 )−1 ∈ B(E).
Since, |a0 | > 2 max{|z| : z ∈ K}, we have az0 < 12 where z ∈ K.
Hence, for z ∈ K,
 k
1 1 1 1 ∞ z
=− · z =− ∑ .
z − a0 a0 1 − a0 a0 k=0 a0
 3.1. RUNGE’S THEOREM 53

By Weierstrass M-test the series converges uniformly on K. Let

 k
1 n z
Qn (z) = − ∑ .
a0 k=0 a0
Then Qn is a polynomial with poles at ∞ ∈ E, Qn ∈ B(E). Since, Qn →
(z − a0 )−1 , (z − a0 )−1 ∈ B(E).
Theorem 3.6 (Runge’s Theorem). Let K be a compact subset of C and
let E be a subset of C∞ \ K. If f is analytic in an open set containing K
and ε > 0, then there is a rational function R(z) whose only poles lies in
E and such that | f (z) − R(z)| < ε for all z ∈ K.
Proof. Let f be an analytic function in G and K ⊆ G. By Proposition

Advanced Complex Analysis by V. Ravichandran

3.1, there are straight line segments γ1 , γ2 , . . ., γn in G \ K such that for
z∈K
n
1 f (w)
Z
f (z) = ∑ dw (3.1.8)
k=1 2πi γk w−z
By Lemma 3.2, there are rational functions Rk (z) with poles in {γk } and
such that for z ∈ K,
1 f (w)
Z
ε
dw − Rk (z) < . (3.1.9)
2πi γk w − z n
Using equations (3.1.8) and (3.1.9), we see that with R(z) = R1 (z) +
R2 (z) + · · · + Rn (z),
n
1 f (w)
Z
ε
| f (z) − R(z)| ≤ ∑ dw − Rk (z) ≤ n · = ε.
k=1 2πi γk w−z n
The function R is a rational function with poles in nk=1 γk ⊆ G \ K. By
S

Lemma 3.5 and the fact that B(E) is an algebra, R ∈ B(E).

Corollary 3.7. Let G be an open subset of the plane and let E be a sub-
set of C∞ \ G such that E meets every component of C \ G. Let R(G, E)
be the set of rational functions with poles in E and consider R(G, E) as
a subspace of H (G). If f ∈ H (G), then there exists Rn ∈ R(G, E) such
that f = lim Rn . That is, R(G, E) is dense in H (G).
Proof. Let K be a compact subset of G and ε > 0. Then there exists a
compact subset K1 such that K ⊆ K1 ⊆ G and each component of C∞ \K1
contains a component of C∞ \ G. Since, E meets every component of
C∞ \ G, E meets every component of C∞ \ K1 . By Runge’s theorem, for
every ε > 0, there exists a rational function R(z) whose poles in E and
such that | f (z) − R(z)| < ε. Therefore, R(G, E) is dense in H (G).
54 3. RUNGE’S THEOREM

Corollary 3.8. If G is open subset of C such that C∞ \ G is connected,

then for each f ∈ H (G), there exists polynomials {pn } such that f =
lim pn in H (G).
Proof. Let E = {∞} in previous corollary and note that a rational func-
tion whose only pole at ∞ is a polynomial.

3.2. Simple Connectedness

Definition 3.9. An open set G is simply connected if and only if every
closed rectifiable curve in G is homotopic to zero.
Recall that a closed rectifiable curve γ is homotopic to zero, written
γ ∼ 0, if γ is homotopic to a constant curve. Two closed rectifiable

Advanced Complex Analysis by V. Ravichandran

curves γ0 , γ1 : [0, 1] → G are homotopic in G if there exists a continuous
function Γ : [0, 1] × [0, 1] → G such that Γ(s, 0) = γ0 (s), Γ(s, 1) = γ1 (s),
for all 0 ≤ s ≤ 1 and Γ(0,t) = Γ(1,t), for all 0 ≤ t ≤ 1.
Definition 3.10. Let X and Ω be metric spaces. A homeomorphism
between X and Ω is a continuous bijection f : X → Ω such that f −1 :
Ω → X is also continuous. In such case, X and Ω are homeomorphic.
Recall that a continuous bijection f : X → Ω is a homeomorphism if
and only if f is a closed (or open) map.
Remark 3.11. The whole complex plane C and D = {z : |z| < 1} are
homeomorphic.

Proof. Consider f : C → D given by

z
f (z) = .
1 + |z|
Clearly, | f (z)| < 1. Consider g : D → C by
w
g(w) = .
1 − |w|

Then f (g(w)) = w and g( f (z)) = z. Thus f −1 = g and therefore, f is

one-one and onto. Both f and g are continuous funcions. Thus, f is a
homeomorphism between C and D. Thus C is homeomorphic to D.

Remark 3.12. If f is one-one analytic function on an open set G, Ω =

f (G), then G and Ω are homeomorphic.
 3.2. SIMPLE CONNECTEDNESS 55

Theorem 3.13. Let G be an open connected subset of C. Then the

following are equivalent:
(a) G is simply connected.
(b) n(γ; a) = 0 for all closed rectifiable curve γ in G and for all
a ∈ C \ G.
(c) C∞ \ G is connected.
(d) For f ∈ H (G), there exists a sequence of polynomials {pn }
such that pn → f .
(e) For f ∈ H (G), γ is a closed rectifiable curve, γ = 0.
R

(f) Every f ∈ H (G) has a primitive, i.e., there exists g such that
g = f 0.
(g) For f ∈ H (G), f (z) 6= 0 for all z ∈ G, there exists g ∈ H (G)

Advanced Complex Analysis by V. Ravichandran

such that f = eg .
(h) For f ∈ H (G), f (z) 6= 0 for all z ∈ G, there exists g ∈ H (G)
such that f = g2 .
(i) G is homeomorphic to D.
(j) If u : G → R is harmonic, then there exists a harmonic function
v : G → R such that f = u + iv is analytic in G.
Proof. We will show that (a) =⇒ (b) =⇒ . . . =⇒ (i) =⇒ (a) and (h)
=⇒ (j) =⇒ (g).
(a) =⇒ (b) Let G be simply connected. Then by Cauchy’s Theo-
rem, we have
1 dz
Z
n(γ; a) = =0
2πi γ z−a
for all a ∈ C \ G as (z − a)−1 is analytic in G, γ is a rectifiable curve in
G.
(b) =⇒ (c) Assume that n(γ; a) = 0 for all closed rectifiable curves
in G and for all a ∈ C \ G. We have to show that C∞ \ G is connected.
On the contrary, assume that C∞ \ G is not connected. Then there exist
A and B disjoint, non-empty open/closed sets of C∞ such that C∞ \ G =
A ∪ B. Since, ∞ ∈ C∞ \ G, this gives ∞ ∈ A or ∞ ∈ B. Suppose that
∞ ∈ B. Then A is compact in C, as A is compact in C∞ and ∞ 6∈ A. Now,
G1 = G ∪ A = C∞ \ B is an open set and contains A. Since, A is compact
and G1 is open set then by Proposition 3.1.1 there exist finite number of
closed polygons γ1 , γ2 , . . ., γn in G1 \ A ⊆ G such that
n
1 f (w)
Z
f (z) = ∑ dw
k=1 2πi γk w−z
56 3. RUNGE’S THEOREM

for all z ∈ A and for every f ∈ H(G1 ). In particular, with f (z) = 1,

n
1= ∑ n(γk ; z).
k=1

Thus, atleast one n(γk ; z) 6= 0 which is a contradiction.

(c) =⇒ (d) See Corollary 3.8 of Runge’s Theorem.
(d) =⇒ (e) Let γ be a closed rectifiable curve in G, let f be an
analytic function on G and let {pn } be a sequence of polynomials such
that f R= lim pn in H (G). Then since each pn is analytic in C and γ ∼ 0
in C, γ pn = 0 for all n. Since pn → f uniformly to {γ},

Advanced Complex Analysis by V. Ravichandran

Z Z
f = lim pn = 0.
γ γ

(e) =⇒ (f) Fix a ∈ G. Define

Z
F(z) = f (z)dz
γ

where γ is a rectifiable curve connecting a and z. The function F is

well defined, for if, γ1 is another rectifiable curve connecting a to z, then
γ + (−γ1 ) is a closed rectifiable curve and by (e),
Z Z Z Z Z
0= f= f+ f =⇒ f= f.
γ+(−γ1 ) γ −γ1 γ γ1

We claim that F is a primitive of f . If z0 ∈ G and r > 0 is such that

D(z0 , r) ⊂ G, then let γ be a path from a to z0 . For z ∈ D(z0 , r), let
γz = γ +[z0 , z]; that is, γz is the path followed by the straight line segment
from z0 to z. Hence,

F(z) − F(z0 ) 1
Z
= f.
z − z0 z − z0 [z0 ,z]

Now proceed as in the proof of Morera’s Theorem to show that F 0 (z0 ) =

f (z0 ).
(f) =⇒ (g) Let f ∈ H (G), f 6= 0 for all z ∈ G. Then f 0 (z)/ f (z) is
analytic in G. Therefore, by (f), there exists F ∈ H (G) such that

f 0 (z)
F(z) = .
f (z)
 3.2. SIMPLE CONNECTEDNESS 57

Now, ( f e−F )0 = f 0 e−F − f e−F F 0 = e−F ( f 0 − f F 0 ) = 0 and so, f e−F = A

and thus, f = AeF = eF A = eF+c = eg , where g = F + c is a constant.
(g) =⇒ (h) Let f ∈ H (G) for z ∈ G. Then there exists h ∈ H (G)
such that f = eh . Now g = eh/2 . Then g2 = (eh/2 )2 = eh = f .
z
(h) =⇒ (i) If G = C, then f (z) = 1+|z| is the homeomorphism of
C onto D. If G 6= C, then by Riemann Mapping Theorem, there exists
an analytic function f : G → D which is one-one and onto. By Open
Mapping Theorem, f −1 is analytic. Thus, both f and f −1 are continuous
and so f is a homeomorphism.
(i) =⇒ (a) Assume that G is homeomorphic to disk D = {z : |z| <
1}. We have to prove that G is simply connected. Let h : G → D be a
homeomorphism. Let γ be a closed curve in G. Then σ (s) = h(γ(s)) is

Advanced Complex Analysis by V. Ravichandran

a closed curve in D. Thus, if Λ : [0, 1] × [0, 1] → D is defined by

Λ(s,t) = (1 − t)h(γ(s)) = (1 − t)σ (s)

then Λ(s, 0) = σ (s), and Λ(s, 1) = 0, for all 0 ≤ s ≤ 1. Also, Λ(0,t) =

(1−t)σ (0) = (1−t)σ (1) = Λ(1,t) for all 0 ≤ t ≤ 1 and Λ is continuous.
Now let Γ = h−1 ◦ Λ. Then h−1 and Λ are continuous which give Γ is
continous map of [0, 1] × [0, 1] into G. Also,

Γ(s, 0) = h−1 (Λ(s, 0)) = h−1 (σ (s)) = h−1 (h(γ(s))) = γ(s)

Γ(s, 1) = h−1 (Λ(s, 1)) = h−1 (0)
Γ(0,t) = h−1 (Λ(0,t)) = h−1 (Λ(1,t)) = Γ(1,t)

Therefore, γ ∼ 0. Thus, G is simply connected.

(h) =⇒ (j) Assume that every non-vanishing analytic function in G
has square root. Then any harmonic function on C or D has a harmonic
conjugate. If G = C, then the result follows from this. Assume that
G 6= C. Then by hypothesis, and Riemann Mapping Theorem (which
holds under given hypothesis), there exists h : G → D which is one-
one and onto analytic function. Then, if u : G → R is harmonic, then
u1 : D → R defined by u1 (z) = u(h−1 (z)) is harmonic in D (Prove it).
Therefore, there exists v1 : D → R such that f1 (z) = u1 (z) + iv1 (z) is
analytic in D. Let f = f1 ◦ h. Then f is analytic on G and Re f (z) =
Re( f1 ◦ h)(z) = Re( f1 (h(z))) = u1 (h(z)) = u(h−1 (h(z))) = u(z). There-
fore, the harmonic conjugate of u is v = Im f = v1 ◦ h.
(j) =⇒ (g) Assume that harmonic function on G has harmonic con-
jugate in G. Let f ∈ H (G), f 6= G for all z ∈ G. Then let u = Re f ,
58 3. RUNGE’S THEOREM

v = Im f . Let U : G → R be defined by
1
U(x, y) = log | f (x + iy)| = log(u2 + v2 ).
2
Then, a calculation shows that U is harmonic. Therefore there exists
V which is harmonic such that g = U + iV is analytic on G. Then let
h(z) = eg(z) . Then, h 6= 0 on G,
f (z) | f (z)| |f| |f|
= Re g(z) = U = =1
h(z) e e |f|
on G. Therefore, f /h is analytic, range of f /h is |z| = 1 and it is not
open. Therefore, f /h = c or f = ch = ceg = eg+c .

Advanced Complex Analysis by V. Ravichandran

3.3. Mittag-Leffler’s Theorem
Theorem 3.14 (Mittag’s-Leffler’s Theorem). Let G be an open set,
{ak } a sequence of distinct points in G without a limit point in G and let
{Sk (z)} be the sequence of rational functions given by
mk
A jk
Sk (z) = ∑ (z − ak ) j
j=1

where, mk is some positive integer, A1k , A2k , . . . , Amk k are arbitrary com-
plex numbers. Then there is a meromorphic function f on G whose poles
are exactly the points {ak } and such that the singular part of f at ak is
Sk (z).
Proof. Since G is open in C, therefore we can find compact subsets Kn
of G such that (a) G = ∞ n=1 Kn (b) Kn ⊂ int Kn+1 (c) each component of
S

C∞ \ Kn contains a component of C∞ \ G (d) K ⊆ G, K is compact then

K ⊆ Kn for some n.
Since, Kn is compact, {ak } has no limit point in G, there are only
finite numbers of points ak in each Kn , as in a compact set every infinite
set has a limit point. Define the set In of integers as follows: I1 = {k :
ak ∈ K1 } and In = {k : ak ∈ Kn \ Kn−1 } for n ≥ 2. Each In is finite, as Kn
contains only finitely many ak ’s. Define the functions fn by

fn (z) = ∑ Sk (z).
k∈In

Note that fn = 0 if In 6= 0./ Then for n ≥ 2, fn is a rational function

(as In is finite) and the poles of fn are the points {ak : k ∈ In } ⊆ Kn \
 3.3. MITTAG-LEFFLER’S THEOREM 59

Kn−1 . Thus, fn (for n ≥ 2) has no poles in Kn−1 and so it is analytic

in a neighbourhood of Kn−1 . Since, Kn−1 is compact, C∞ \ G meets
every component of C∞ \ K, by Runge’s Theorem, there exists a rational
function Rn (z) with poles in C∞ \ G that satisfies
 n
1
| fn (z) − Rn (z)| <
2

for all z ∈ Kn−1 .

We claim that the function f given by
∞
f (z) = f1 (z) + ∑ [ fn (z) − Rn (z)] (3.3.1)

Advanced Complex Analysis by V. Ravichandran

n=2

is the desired meromorphic function. The series in equation (3.3.1) con-

verges uniformly on any compact subset of G \ {ak : k ≥ 1}. Let K be
a compact subset of G \ {ak : k ≥ 1}. Then K ⊆ KN for some N ≥ 1. If
n ≥ N, then
 n
1
| fn (z) − Rn (z)| <
2
for all z ∈ K ⊆ KN ⊆ Kn and so equation (3.3.1) converges uniformly
on K. Since, fn (z) − Rn (z) is analytic in G \ {ak : k ≥ 1}, the function
f is analytic in G \ {ak : k ≥ 1}. The function f is meromorphic if
ak ’s are poles of f . Consider a fixed integer k ≥ 1. Then there exists
Rk > 0 such that |a j − ak | > Rk for j 6= k (The Rk should be chosen so
that D(ak , Rk /2) has no other points of {a j : j ≥ 1} as this has no limit
points in G, ak ∈ G). Thus, for 0 < |z − ak | < Rk , f (z) = Sk (z) + g(z)
where g(z) is analytic in the disk D(ak , Rk ). Thus, z = ak is a pole of
f as the principal part of Laurent’s Series expansion of f at z = ak is
Sk (z).
Advanced Complex Analysis by V. Ravichandran
 4. H ARMONIC F UNCTIONS

4.1. Basic Properties of Harmonic Functions

Definition 4.1. Let G be an open subset of C. A function u : G → R
is harmonic if u has continuous second partial derivatives and satisfies
Laplace Equation

Advanced Complex Analysis by V. Ravichandran

∂ 2u ∂ 2u
∆u = 2 + 2 = 0
∂x ∂y
for all z = x + iy ∈ G.

Theorem 4.2. (a) A function f : G → C is analytic if and only if

Re f = u, Im f = v are harmonic functions which satisfy the
Cauchy Riemann Equations.
(b) A region G is simply-connected if and only if for each harmonic
function u on G, there is a harmonic function v on G such that
f = u + iv is analytic on G.

Definition 4.3. If f : G → C is an analytic function, then u = Re f is

harmonic conjugate of v = Im f .
If G is an open set, u : G → C is harmonic in G, then, by Theo-
rem 4.2, u has harmonic conjugate on every disc D ⊆ G. Thus, every
harmonic function has harmonic conjugate locally.
If v1 , v2 are harmonic conjugate of u, then i(v1 − v2 )=(u + iv1 ) −
(u + iv2 ) is analytic in G and takes only imaginary values. Thus, if G is
a region, v1 − v2 = c or, v1 = v2 + c. Any two harmonic conjugate of a
harmonic function u differs by a constant.
Theorem 4.4. If u : G → C is harmonic, then u is infinitely differen-
tiable.
Proof. Fix z0 = x0 + iy0 ∈ G and choose δ > 0 such that D(z0 , δ ) ⊆ G.
By Theorem 4.2, there exists a harmonic function v in D(z0 , δ ) such that
f = u + iv is analytic in D(z0 , δ ). Since f is analytic, the derivative of
all order of f and hence all partial derivatives of u exist at z0 .
61
62 4. HARMONIC FUNCTIONS

Theorem 4.5 (Mean Value Theorem). Let u : G → R be a harmonic

function and let D(a, r) be closed disc contained in G. Then,
Z 2π
1
u(a) = u(a + reit )dt.
2π 0

Proof. Let γ be the circle γ(t) = a+reit , 0 ≤ t ≤ 2π. Let δ = d(D(a, r), C\
G). Then δ > 0 and, with r0 = r + δ , we have D(a, r) ⊂ D(a, r0 ) ⊂ G.
Let f = u + iv be the analytic function on D(a, r0 ) (such function exists
by Theorem 4.2). Then

f (a + reit )
Z 2π Z 2π
1 f (z) 1 1
Z
it
f (a) = dz = re idt = f (a + reit )dt.

Advanced Complex Analysis by V. Ravichandran

2πi γ z−a 2πi 0 reit 2π 0

By taking real part, we have

Z 2π
1
u(a) = u(a + reit )dt.
2π 0

Definition 4.6. A continuous function u : G → R has Mean Value Prop-

erty (MVP), if whenever D(a, r) ⊂ G, we have
Z 2π
1
u(a) = u(a + reit )dt.
2π 0

Remark 4.7. Mean Value Theorem says that every harmonic function
has MVP. Every continuous function with MVP is also harmonic. This
will be proved later.

Theorem 4.8 (Maximum Principle (First Version)). Let G be a re-

gion and suppose that u is a continuous real-valued function on G with
the MVP. If there exists a point a ∈ G such that u(z) ≤ u(a) for all z ∈ G,
then u is a constant function.
Proof. Let A = {z ∈ G : u(z) = u(a)}. Clearly, a ∈ A and so A a is non-
empty set. Since u is continuous, A = u−1 ({u(a)}) is closed subset of G.
We will prove that A is open in G. Once this is proved, the connectedness
of G will imply that A = G and so u is constant in G.
To prove A is open, let z0 ∈ A. Let r be chosen such that D(z0 , r) ⊆ G.
We claim that D(z0 , r) ⊂ A and so A is open. If possible, let b ∈ D(z0 , r)
but b 6∈ A. Then u(b) 6= u(a). Since u(z) ≤ u(a) for all z ∈ G, we must
 4.1. BASIC PROPERTIES OF HARMONIC FUNCTIONS 63

have u(b) < u(a). Since z0 ∈ A, u(a) = u(z0 ) and so u(b) < u(z0 ). Since
u is continuous, there exists a neighbourhood D(b, δ ) of b in which

|u(z) − u(b)| < ε = u(z0 ) − u(b)

or
u(z) < u(z0 ) for all z ∈ D(b, δ ).

Let |z0 − b| = ρ. Then, b = z0 + ρeiβ , 0 ≤ β < 2π and there is a subin-

terval I of [0, 2π] with β ∈ I and u(z) < u(z0 ) for z = z0 + ρeit for all
t ∈ I. Thus, by MVP,

Advanced Complex Analysis by V. Ravichandran

1 2π
Z
u(z0 ) = u(z0 + ρeit )dt
2π 0
1 1
Z Z
= u(z0 + ρeit )dt + u(z0 + ρeit )dt
2π I 2π [0,2π]\I
< u(z0 )

This contradiction shows that D(z0 , r) ⊂ A and so A must be open.

Corollary 4.9 (Minimum Principle). Let G be a region and suppose
that u is a continuous real-valued function on G with the MVP. If there
exists a point a ∈ G such that u(a) ≤ u(z) for all z ∈ G, then u is a
constant function.
Proof. This immediately follows by applying Theorem 4.8 to the func-
tion −u.
Recall that, if f : G → R and a ∈ G or a = ∞, then the limit superior
of f (z) as z → a, denoted by lim supz→a f (z), is defined as

lim sup f (z) = lim sup{ f (z) : z ∈ G ∩ D(a, r)}

z→a r→0+

For a = ∞, D(a, r) is the ball in the metric of C∞ . If r < r0 , then

{ f (z) : z ∈ G ∩ D(a, r)} ⊆ { f (z) : z ∈ G ∩ D(a, r0 )}

and so

sup{ f (z) : z ∈ G ∩ D(a, r)} ≤ sup{ f (z) : z ∈ G ∩ D(a, r0 )}.

Thus, sup{ f (z) : z ∈ G ∩ D(a, r)} is increasing function of r.

64 4. HARMONIC FUNCTIONS

Theorem 4.10 (Maximum Principle (Second Version)). Let G be a

region and let u and v be two continuous real-valued functions on G
that have the MVP. If, for each point a in ∂∞ G,

lim sup u(z) ≤ lim inf v(z)

z→a z→a

then either u(z) < v(z) for all z ∈ G or u = v.

Proof. Fix a ∈ ∂∞ G, and for each δ > 0, let Gδ = G ∩ D(a, δ ). By

hypothesis, for each a ∈ ∂∞ G, we have

0 ≥ lim sup{u(z) : z ∈ Gδ } − lim inf{v(z) : z ∈ Gδ }

Advanced Complex Analysis by V. Ravichandran

z→a z→a

= lim [sup{u(z) : z ∈ Gδ } + sup{−v(z) : z ∈ Gδ }]

δ →0
= lim [sup{u(z) − v(z) : z ∈ Gδ }] = lim sup(u(z) − v(z)).
δ →0 z→a

Thus, it is enough to prove the theorem when v ≡ 0 in G.

So, assume that lim supz→a u(z) ≤ 0 for each a ∈ ∂∞ G. We shall
prove that either u(z) < 0 for all z ∈ G or u = 0. It is enough to prove
u(z) ≤ 0. Once this is proved and u(a) = 0 for some a ∈ G, then u(z) ≤
u(a) for all z ∈ G and by first version of maximum principle, u(z) =
u(a) = 0 for all z ∈ G or u = 0 on G. Thus, u = 0 or u(z) < 0 for all
z ∈ G.
We now prove u(z) ≤ 0. Assume, on the contrary that, u(b) > 0 for
some b ∈ G. Let ε > 0 be chosen so that u(b) > ε, and let B = {z ∈ G :
u(z) ≥ ε}. Clearly b ∈ B, B 6= φ . If a ∈ ∂∞ G, then lim supz→a u(z) ≤ 0
implies there is δ = δ (a) such that sup{u(z) : z ∈ Gδ } < ε, or u(z) < ε
for all z ∈ Gδ = G ∩ D(a, δ ). By Lebesgue Covering Lemma, a δ can be
chosen independent of a. That is, there exists δ > 0 such that, if z ∈ G,
and d(z, ∂∞ G) < δ , then u(z) < ε. Thus, B ⊆ {z ∈ G : d(z, ∂∞ G) ≥ δ }.
Therefore, B is bounded in C and B = u−1 {[ε, ∞)} is closed. Thus, B
is compact. Since B 6= φ , there exists z0 ∈ B such that u(z0 ) ≥ u(z) for
all z ∈ B. Since u(z) < ε for all z ∈ G \ B, we have u(z) ≤ u(z0 ) for all
z ∈ G. Therefore u is constant. This implies u(z) = u(z0 ) ≥ ε > 0 and
lim supz→a u(z) ≥ ε > 0 which is a contradiction.

Corollary 4.11. Let G be a bounded region and suppose that w : G → R

is a continuous function that satisfies MVP on G. If w(z) = 0 for all
z ∈ ∂ G, then w(z) = 0 for all z ∈ G.
 4.2. HARMONIC FUNCTIONS ON A DISC 65

Proof. By taking w = w and v = 0 in Theorem 4.10, we see that, either

w(z) < 0 for all z or w = 0. Similarly, by taking v = w and u = 0 in
Theorem 4.10, we see that, either w(z) > 0 for all z, or w = 0. Therefore,
we must have w = 0.

4.2. Harmonic Functions on a Disc

Definition 4.12. The function
∞
Pr (θ ) = ∑ r|n| einθ
n=−∞

for 0 ≤ r < 1, −∞ < θ < ∞, is called the Poisson Kernel.

Advanced Complex Analysis by V. Ravichandran

Since
1+z 2z
= 1+ = 1 + 2z + 2z2 + . . . (|z| < 1),
1−z 1−z
for 0 ≤ r < 1, we have
1 + reiθ ∞

1 − reiθ
= 1 + ∑ rneinθ (4.2.1)
n=1

and, therefore
1 + reiθ + e−inθ )
  ∞ ∞  inθ 
n n e
Re = 1 + 2 ∑ r cos nθ = 1 + 2 ∑ r
1 − reiθ n=1 n=1 2
∞ −∞ ∞
|n| inθ |n| inθ
= 1+ ∑ r e + ∑ r e = ∑ r|n| einθ
n=1 n=−1 n=−∞
= Pr (θ )
Therefore,
1 + reiθ 1 + reiθ 1 + re−iθ
   
1
Pr (θ ) = Re = +
1 − reiθ 2 1 − reiθ 1 − re−iθ
1 1 + reiθ − re−iθ − r2 + 1 + re−iθ − reiθ − r2
= ·
2 1 + r2 − 2r cos θ
1−r 2
=
1 + r2 − 2r cos θ
Hence, for 0 ≤ r < 1, −∞ < θ < ∞, we have
1 − r2 1 + reiθ
 
Pr (θ ) = = Re
1 + r2 − 2r cos θ 1 − reiθ
66 4. HARMONIC FUNCTIONS

Theorem 4.13. The Poisson kernel satisfies:

1 Rπ
(a) 2π −π Pr (θ )dθ = 1.
(b) Pr (θ ) > 0 for all θ , Pr (−θ ) = Pr (θ ), and Pr (θ ) periodic in θ
with period 2π.
(c) Pr (θ ) < Pr (δ ) if 0 < δ < |θ | ≤ π.
(d) For each δ > 0, limr→1−1 Pr (θ ) = 0 uniformly in θ for π ≥ |θ |
≥ δ.
Proof. Writing as a countour integral, we have

1 + reiθ
   
1 1 1 1+z 1
Z π Z π Z
Pr (θ )dθ = Re dθ = Re dz = 1.
1 − reiθ |z|=r 1 − z z

Advanced Complex Analysis by V. Ravichandran

2π −π 2π −π 2πi

(b)
(a) Clearly,

1 − r2
Pr (θ ) = > 0 for 0 ≤ r < 1,
|1 − reiθ |2
and
1 − r2 1 − r2
Pr (−θ ) = = = Pr (θ )
1 + r2 − 2r cos(−θ ) 1 + r2 − 2r cos θ
Since the Poisson kernel
1 − r2
Pr (θ ) =
1 + r2 − 2r cos θ
is a function of cos θ , it is periodic in θ with period 2π.
(c) The function

(1 − r2 ) · 2r sin θ
Pr0 (θ ) = −
(1 + r2 − 2r cos θ )2

satisfies Pr0 (θ ) < 0 for 0 < θ ≤ π, and Pr0 (θ ) > 0 for −π ≤ θ <
0. Therefore, Pr (θ ) is increasing in [−π, 0) and decreasing in
(0, π]. For δ < θ < π, Pr (θ ) < Pr (δ ). For, −π < θ < −δ ,
Pr (θ ) < Pr (−δ ) = Pr (δ ). Therefore, for 0 < δ < |θ | ≤ π,
Pr (θ ) < Pr (δ ).
(d) Since sup{Pr (θ ) : δ = |θ | ≤ π} ≤ Pr (δ ), we have

lim [sup{Pr (θ ) : δ ≤ |θ | ≤ π}] = lim Pr (δ ) = 0.

r→1− r→1−
 4.2. HARMONIC FUNCTIONS ON A DISC 67

A region G is Drichlet if, for any given continuous function f : ∂ G →

R, there exists continuous functions u : G → R such that u(z) = f (z) for
all z ∈ ∂ G, and u is harmonic in G. Drichlet Problem is the determina-
tion of whether a region is Drichlet or not. The next theroem shows that
the unit disk is a Drichlet region.
Theorem 4.14. If f : ∂ D → R is a continuous function, then there exists
a continuous function u : D → R such that
(a) u(z) = f (z) for all z ∈ D.
(b) u is harmonic in D.
This function u is unique and is given by
1
Z π

Advanced Complex Analysis by V. Ravichandran

iθ
u(re ) = Pr (θ − t) f (eit )dt
2π −π

for 0 ≤ r < 1, and 0 ≤ θ ≤ 2π.

Proof. Define u : D → R by
( R
1 π
iθ 2π −π Pr (θ − t) f (eit )dt 0 ≤ r < 1,
u(re ) =
f (eiθ ) r = 1.
Then (a) is satisfied. We need to show that u is harmonic in D, and is
continuous on D. If 0 ≤ r < 1, then
" #
1 π 1 + rei(θ −t)
Z
iθ
u(re ) = Re i(θ −t)
f (eit )dt
2π −π 1 − re
 Z π it
e + reiθ

1 it
= Re f (e )dt
2π −π eit − reiθ
= Re g(reiθ )
where
1 eit + z 1 η + z dη
Z π Z
it
g(z) = f (e ) it dt = f (η) · .
2π −π e −z 2πi |η|=1 η −z η
Then, the function g is analytic in D by Leibnitz’s Rule and so u = Re g
is harmonic in D. This proves (b).
Since the function u is harmonic in D, it is also continuous in D.
Thus, we need to show that u is continuous in ∂ D. Fix α ∈ [−π, π] so
that eiα ∈ ∂ D. Let ε > 0. We show that there exist 0 < ρ < 1 and an arc
A of ∂ D containing eiα such that
|u(reiθ ) − u(eiα )| = |u(reiθ ) − f (eiα )| < ε
68 4. HARMONIC FUNCTIONS

for all r ∈ (ρ, 1) and eiθ ∈ A. It is enough to prove this for α = 0. That
is, given ε > 0, there exists 0 < ρ < 1 and an arc A of ∂ D containing 1
such that |u(reiθ ) − f (1)| < ε for all r ∈ (ρ, 1) and eiθ ∈ A.
Since f is continuous at z = 1, there exists δ > 0 such that | f (eiθ ) −
f (1)| < ε/3 for all |θ | < δ . Using this inequality, we see that

1 1 ε
Z Z
it ε
Pr (θ − t)( f (e ) − f (1))dt ≤ · Pr (θ − t)dt ≤ .
2π |t|<δ 2π 3 |t|<δ 3
(4.2.2)
Let M := max{| f (eiθ ) : |θ | ≤ π}. Then M < ∞. If M = 0, u ≡ 0. Assume
M > 0. Therefore, 0 < M < ∞. By Theorem 4.13 (d), there exists ρ,

Advanced Complex Analysis by V. Ravichandran

0 < ρ < 1 such that Pr (θ ) < ε/3M for ρ < r < 1 and |θ | ≥ δ /2. If
|t| ≥ δ , then |t − θ | ≥ |t| − |θ | ≥ δ − δ /2 = δ /2. Hence,

1 1 2ε
Z
ε
Pr (θ − t)( f (eit ) − f (1))dt ≤
· 2M · · 2π =
2π
|t|≥δ 2π 3M 3
(4.2.3)
Let A = {eiθ : |θ | < δ /2}. Then, if eiθ ∈ A and ρ < r < 1,

1 π
Z
iθ
u(re ) − f (1) = Pr (θ − t) f (eit )dt − f (1)
2π −π
1 π
Z
= Pr (θ − t)( f (eit ) − f (1))dt
2π −π
1
Z
= Pr (θ − t)( f (eit ) − f (1))dt
2π |t|<δ
1
Z
+ Pr (θ − t)( f (eit ) − f (1))dt
2π |t|≥δ

Using (4.2.2) and (4.2.3), we have |u(reiθ ) − f (1)| ≤ ε for eiθ ∈ A and
ρ < r < 1.
To show the uniqueness of u, let v be a continuous function on D,
which is harmonic in D and v(eiθ ) = f (eiθ ) for all θ . Then u − v is
harmonic in D and (u − v)(z) = 0 for all z ∈ ∂ D. By Corollary 4.11,
u − v ≡ 0 on G or u = v on G. Therefore, u(z) = v(z) for all z ∈ D.
 4.2. HARMONIC FUNCTIONS ON A DISC 69

Corollary 4.15. If u : D → R is a continuous function that is harmonic

in D, then
1
Z π
iθ
u(re ) = Pr (θ − t)u(eit )dt
2π −π

for 0 ≤ r < 1 and θ ∈ R. Moreover, u is the real part of the analytic

function,
Z π it
1 e +z
f (z) = u(eit )dt.
2π −π eit −z
Proof. Since u : D → R is continuous, u is continuous on ∂ D and by
Theorem 4.14, the first part follows. The second part follows as f is

Advanced Complex Analysis by V. Ravichandran

analytic and u is continuous real part (as in proof of Theorem 4.14).

Example 4.16. Let u be harmonic and h be analytic. Then u(h(z)) is

harmonic.

Proof. Let h(z) = r(x, y) + it(x, y) where rx = ty and ry = −tx . We have

to show that u(r(x, y),t(x, y)) is harmonic. Now,

ux (x, y) = ux (r(x, y),t(x, y))rx (x, y) + uy (r(x, y),t(x, y))tx (x, y)

= ux (r,t)rx + uy (r,t)tx
uxx (x, y) = uxx (r,t)rx2 + uyy (r,t)tx2 + ux (r,t)rxx + uy (r,t)txx

Similarly,

uy (x, y) = ux (r,t)ry + uy (r,t)ty

uyy (x, y) = uxx (r,t)ry2 + uyy (r,t)ty2 + ux (r,t)ryy + uy (r,t)tyy

Therefore, uxx + uyy = 0.

Corollary 4.17. Let a ∈ C, ρ > 0 and suppose h is a continuous real-
valued function on {z : |z−a| = ρ}, then there exists a unique continuous
function w : D(a, ρ) → R such that w is harmonic in D(a, ρ) and w(z) =
h(z) on |z − a| = ρ.
Proof. Define f : ∂ D → R by f (eiθ ) = h(a + ρeiθ ). Then f is contin-
uous function such that u is harmonic in D and u(eiθ ) = f (eiθ ) (such
function exists by Theorem 4.14), then the function w(z) = u(z − a)/ρ
from D(a, ρ) to R is the desired function.
70 4. HARMONIC FUNCTIONS

Clearly, u is harmonic in D implies w is harmonic in D(a, ρ). Also,

w is a composition of harmonic function with analytic function, which
implies w is harmonic, and u : D → R is continuous implies w : D(a, ρ)
→ R is continuous. Finally, w(a + ρeiθ ) = u(eiθ ) = f (eiθ ) = h(a +
ρeiθ ) as w(z) = h(z) on |z − a| = ρ.
Theorem 4.18. If u : G → R is continuous function which has MVP,
then u is harmonic.
Proof. Let a ∈ G. Choose ρ > 0 such that D(a, ρ) ⊆ G. It is sufficient
to show that u is harmonic in D(a, ρ). Since u is continuous on G, u
is continuous on |z − a| = ρ. Thus, there exists w : D(a, ρ) → R which
is harmonic in D(a, ρ) and w(a + ρeiθ ) = u(a + ρeiθ ) for all θ . Since

Advanced Complex Analysis by V. Ravichandran

u − w satisfies MPV (use Theorem 4.5 and the fact that u has MVP),
and (u − w)(a + ρeiθ ) = 0 for all θ or (u − w)(z) = 0 for |z − a| = ρ.
Corollary 4.11 show that (u − w)(z) = 0 or u(z) = w(z) for all z ∈ G.
Since w is harmonic in D(a, ρ), u is also harmonic there.
Theorem 4.19 (Harnack’s Inequality). If u : D(a, R) → R is contin-
uous harmonic in D(a, R) and u ≥ 0, then for 0 ≤ r < R and for all
θ,
R−r R+r
u(a) ≤ u(a + reiθ ) ≤ u(a).
R+r R−r
Proof. Since u is continuous on D(a, R) and harmonic in D(a, R), the
function w : D → R given by w(z) = u(a + Rz) is continuous and w is
harmonic in D. Therefore, for 0 ≤ ρ < 1, we have by Corollary 4.15,
1
Z π
iθ
w(ρe ) = Pρ (θ − t)w(eit )dt
2π −π
or
1
Z π
u(a + Rρeiθ ) = Pρ (θ − t)u(a + Reit )dt.
2π −π

Since 0 ≤ ρ < 1, we have 0 ≤ Rρ < R. Replace Rρ by r. Then, for

0 ≤ r < R,
1
Z π
iθ
u(a + re ) = Pr/R (θ − t)u(a + Reit )dt.
2π −π

Since
1 − r2
Pr (θ ) = ,
|1 − reiθ |2
 4.2. HARMONIC FUNCTIONS ON A DISC 71

and 1 − r ≤ |1 − reiθ |2 ≤ 1 + r, we have

1−r 1 − r2 1 − r2 1+r
= 2
≤ Pr (θ ) ≤ 2
= .
1 + r (1 + r) (1 − r) 1−r
Hence,
R−r R+r
≤ Pr/R (θ − t) ≤
R+r R−r
and therefore we have
1 R−r 1 π
Z π Z
it
u(a + Re )dt ≤ Pr/R (θ − t)u(a + Reit )dt
2π −π R+r 2π −π
1 π R+r

Advanced Complex Analysis by V. Ravichandran

Z
≤ u(a + Reit )dt.
2π −π R − r
Using
1 π
Z
iθ
u(a + re ) = P (θ − t)u(a + Reit )dt,
2π −π r/R
1 π
Z
u(a) = u(a + Reit )dt
2π −π
we get
R−r R+r
u(a) ≤ u(a + reiθ ) ≤ u(a).
R+r R−r

Definition 4.20. If G is an open set in C, then Har(G) is the space

of harmonic functions on G. Since Har(G) ⊆ C(G, R), then Har(G) is
given by the metric inherited from C(G, R).

Theorem 4.21 (Harnack’s Theorem). Let G be a region. (a) The met-

ric space Har(G) is complete. (b) If hun i is a sequence of funtions in
Har(G) such that u1 ≤ u2 ≤ . . ., then either un (z) → ∞ uniformly on
compact subsets of G or, hun i converges in Har(G) to a harmonic func-
tion.
Proof. (a) Since C(G, R) is complete, Har(G) ⊆ C(G, R) is complete if
Har(G) is closed. Let hun i ⊆ Har(G) such that un → u in C(G, R). Then,
let a ∈ G be arbitrary, D(a, R) ⊆ G. Since un is harmonic in G, un is
continuous in D(a, R) and harmonic in D(a, R). Thus,
1 1
Z π Z
it
un (a) = un (a + Re )dt = un (z)dz
2π −π 2π γ
72 4. HARMONIC FUNCTIONS

where γ : |z − a| = R. Let n → ∞. Using the result that if Fn and F

are
R
continuous
R
functions on {γ} and Fn → F uniformly on {γ}, then
γ F = lim γ n we have
F ,
1
Z π
u(a) = u(a + Reit )dt.
2π −π

Therefore u has MVP and so by Theorem 4.18, u is harmonic in D(a, R)

and so harmonic in G.
(b) We may assume that u1 ≥ 0 (or else, consider un −u1 ). Let u(z) =
sup{un (z) : n ≥ 1} for each z ∈ G. Then either u(z) = ∞ or u(z) ∈ R and
un (z) → u(z). Define A := {z ∈ G : u(z) = ∞} and B := {z ∈ G : u(z) <
∞}. Then G = A ∪ B, A ∩ B = 0. /

Advanced Complex Analysis by V. Ravichandran

We first show that A and B are both open. Let a ∈ G and R be chosen
such that D(a, R) ⊆ G. By Harnack’s Inequality,
R − |z − a| R + |z − a|
un (a) ≤ un (z) ≤ un (a), (4.2.4)
R + |z − a| R − |z − a|
for z ∈ D(a, R) and n ≥ 1. If a ∈ A, then un (a) → ∞ and by (4.2.4),
un (z) → ∞ for all z ∈ D(a, R). Therefore, D(a, R) ⊆ A and so A is open.
If a ∈ B, un (a) ≤ u(a) < ∞ and so un (z) < Mu(a) for all z ∈ D(a, R).
Therefore u(z) = sup un (z) ≤ Mu(a). Thus D(a, R) ⊆ B and so B is
open.
Since G = A ∪ B, A ∩ B = 0,/ A, B are both open and G is connected,
we have either A = G or B = G. Suppose A = G, then u ≡ ∞ on G.
If D(a, R) ⊂ G, 0 < ρ < R, then, with M = (R − ρ)/(R + ρ) > 0, the
inequality (4.2.4) gives Mun (a) ≤ un (z) for all z with |z − a| < ρ. Let
ε > 0. Since un (a) → ∞, there exists N such that un (a) > ε/M for all
n ≥ N. Thus un (z) > M · ε/M = ε for all n ≥ N. Thus un (z) → ∞
uniformly on D(a, R). Therefore, un → ∞ uniformly on compact subset
of G (as any compact subset in contained in the finite union of balls
D(ai , ρi )).
If B = G, then u(z) < ∞ for all z ∈ G. For n ≥ m, un (z) − um (z) ≥ 0
for z ∈ G. If ρ < R, then, by the inequality (4.2.4), there exists N de-
pending on a and ρ only such that M(un (a) − um (a)) ≤ un (z) − um (z) ≤
N(un (a) − um (a)) for |z − a| ≤ ρ and for all n ≥ m. Therefore, {un (z)}
is uniformly Cauchy sequence on D(a, ρ). Hence, it follows that hun i
is a Cauchy sequence in Har(G). Since Har(G) is complete, hun i con-
verges in Har(G). Since un (z) → u(z), the harmonic function to which
hun i converges to u.
 5. T HE R ANGE OF AN A NALYTIC F UNCTION

5.1. Bloch’s Theorem

Lemma 5.1. Let f be analytic in D = {z : |z| < 1} and suppose that
f (0) = 0, f 0 (0) = 1 and | f (z)| ≤ M for all z ∈ D. Then M ≥ 1 and
f (D) ⊃ D(0; 1/(6M)).

Advanced Complex Analysis by V. Ravichandran

Proof. Let 0 < r < 1 and f (z) = ∑∞ n 0
n=1 an z . Then a1 = f (0) = 1. Since
| f (z)| ≤ M1 , Cauchy’s inequality shows that, for n ≥ 1,

f (n) (0) M n! M
|an | = ≤ · n = n.
n! n! r r

By letting r → 1, we see that |an | ≤ M. In particular, 1 = |a1 | ≤ M. We

first show that | f (z)| ≥ 1/(6M) for |z| = 1/4M. For |z| = 1/4M, we
have
∞ ∞
| f (z)| ≥ |z| − ∑ anzn c ≥ |z| − ∑ |an||z|n
n=2 n=2
1 −1
∞  n  
1 1 1 M
≥ −∑M = − 1−
4M n=2 4M 4M (4M)2 4M
1 M 4M 1 1 1
= − · = − ≥ .
4M (4M)2 4M − 1 4M 16M − 4 6M
That the last inequality holds when M ≥ 1 can be seen by the following
equivalences
1 1 1 1 1
− ≥ ⇐⇒ ≥ ⇐⇒ M ≥ 1.
4M 16M − 4 6M 6M 8M − 2
We now show that D(0, 1/(6M)) ⊆ f (D). For this purpose, let |w| <
1/(6M). It should be shown that w = f (z0 ) for some z − 0 ∈ D. We
show this by proving the function g defined by g(z) = f (z) − w has a
zero in D. For |z| = 1/(4M), we have
1
| f (z) − g(z)| = |w| < ≤ | f (z)|.
6M
73
74 5. THE RANGE OF AN ANALYTIC FUNCTION

Hence, by Rouche’s Theorem, the functions f and g have the same num-
ber of zeros in D(0, 1/(4M)). Since f (0) = 0, it follows that there is
atleast one zero for g in |z| < 1/(4M). In other words, there exists
z0 ∈ D(0, 1/(4M)) ⊆ D such that g(z0 ) = 0. This gives w = f (z0 ) ∈
f (D(0, 1/(4M))) ⊂ f (D). Hence, D(0, 1/(6M)) ⊆ f (D).

Lemma 5.2. If g is analytic on D(0, R) and g(0) = 0, |g0 (0)| = µ > 0

and |g(z)| ≤ M for all z, then

R2 µ 2
 
g(D(0, R)) ⊃ D 0, .
6M

Proof. The function f : D → C defined by f (z) = g(Rz)/Rg0 (0) is ana-

Advanced Complex Analysis by V. Ravichandran

lytic,
g(0) 0 Rg0 (0)
f (0) = 0 = 0, f (0) = 0 = 1,
Rg (0) Rg (0)
and, for all z ∈ D,
|g(Rz)| M
| f (z)| ≤ 0
≤ .
R|g (0)| Rµ
Lemma 5.1 applied to f shows that
 
Rµ
D 0, ⊆ f (D).
6M

Thus, if |w| < Rµ/(6M), then there is z ∈ D such that w = f (z). Thus,
if |w1 | < R2 µ 2 /(6M), then |w1 /(Rg0 (0))| = |w1 /(Rµ)| < Rµ/(6M) and
so w1 /(Rg0 (0)) = f (z) = g(Rz)/(Rg0 (0)) or w1 = g(Rz) ∈ g(D(0, R)) or

R2 µ 2
 
D 0, ⊆ g(D(0, R)).
6M

Lemma 5.3. Let f be analytic function on D(a, r) such that | f 0 (z) −

f 0 (a)| < | f 0 (a)| for all z ∈ D(a, r), z 6= a. Then f is one-to-one.
Proof. Suppose z1 , z2 ∈ D(a, r), z1 6= z2 . Let γ be the line segment join-
ing z1 and z2 . As {γ} is compact, and f 0 (z) − f 0 (a) is continuous, we
see that

M = max{| f 0 (z) − f 0 (a)| : z ∈ {γ}}

 5.1. BLOCH’S THEOREM 75

exists and hence M < | f 0 (a)|. Using this, we see that

Z
| f (z1 ) − f (z2 )| = f 0 (z)dz
γ
Z 
f 0 (a) + ( f 0 (z) − f 0 (a)) dz

=
γ
Z Z
0
≥ f (a)dz − ( f 0 (z) − f 0 (a))dz
γ γ
Z
0
≥ | f (a)||z1 − z2 | − | f 0 (z) − f 0 (a)||dz|
γ
≥ | f 0 (a)||z1 − z2 | − M|z1 − z2 |

Advanced Complex Analysis by V. Ravichandran

= (| f 0 (a)| − M)|z1 − z2 | > 0

Therefore, f (z1 ) 6= f (z2 ), or f is one-to-one.

Lemma 5.4 (Schwarz Lemma). If f : D(a, R) → D(α, M) is analytic,
and f (a) = α, then
M
| f (z) − α| ≤ |z − a|.
R
Proof. Apply the classical Schwarz lemma to the function g : D → D
defined by g(z) = ( f (a + Rz) − α)/M.
Theorem 5.5 (Bloch’s Theorem). If f is a function analytic on a re-
gion containing the closed disk D, f (0) = 0, and f 0 (0) = 1, then there
is a disk S ⊂ D such that f is one-to-one on S and f (S) contains a disk
of radius 1/72.
Proof. Let K, h : [0, 1] → R be defined by K(r) = max{| f 0 (z)| : |z| = r}
and h(r) = (1 − r)K(r). Since the function | f 0 | is uniformly continuous
on D, for ε > 0, there exists δ > 0 such that

| f 0 (z)| − | f 0 (z0 )| < ε whenever |z − z0 | < δ .

Thus, for |r − r0 | < δ , θ ∈ [0, 2π], we have

−ε + | f 0 (r0 eiθ )| < | f 0 (reiθ )| < ε + | f 0 (r0 eiθ )|.

Taking maximum over all θ ∈ [0, 2π], we get |K(r) − K(r0 )| ≤ ε. There-
fore K is continuous and hence h is continuous. Also h(0) = K(0)
= max{| f 0 (z)| : |z| = 0} = | f 0 (0)| = 1 and h(1) = 0. Let r0 = sup{r :
76 5. THE RANGE OF AN ANALYTIC FUNCTION

h(r) = 1}. By continuity of h, h(r0 ) = 1. Since, h(1) = 0, r0 < 1. Also,

for r > r0 , h(r) < 1. If not, h(r00 ) ≥ 1 for some r00 > r0 . By continu-
ity of h, this implies h(r1 ) = 1 for some r1 > r0 , a contradiction to the
definition of r0 .
Let a be chosen with |a| = r0 and | f 0 (a)| = K(r0 ). Let ρ0 = (1 −
r0 )/2. We show that the disk S := D(a, ρ0 /3) has the desired properties.
Since h(r0 ) = 1, we have
h(r0 ) 1 1
| f 0 (a)| = K(r0 ) = = = .
1 − r0 1 − r0 2ρ0

If z ∈ D(a, ρ0 ), then |z − a| ≤ ρ0 and so

Advanced Complex Analysis by V. Ravichandran

1 − r0 1 + r0
|z| ≤ |a| + |z − a| ≤ r0 + = .
2 2
Thus, D(a, ρ0 ) ⊆ D (0, (1 + r0 )/2). Since (1 + r0 )/2 < r0 , we have
h((1 + r0 )/2) < 1. By maximum modulus theroem, we have K(r) =
max{| f 0 (z)| : |z| ≤ r} and hence, for z ∈ D(a, ρ0 ), we have

| f 0 (z)| ≤ max{| f 0 (z)| : z ∈ D(a, ρ0 )}

≤ max{| f 0 (z)| : z ∈ D (0, (1 + r0 )/2)}
h((1 + r0 )/2)
= K (1 + r0 )/2) =
1 − (1 + r0 )/2
1 1
< = .
(1 − r0 )/2 ρ0
Using | f 0 (a)| = 1/(2ρ0 ), we have, for z ∈ D(a, ρ0 ),
1 1 3
| f 0 (z) − f 0 (a)| ≤ | f 0 (z)| + | f 0 (a)| ≤ + = .
ρ0 2ρ0 2ρ0
By Schwarz’s Lemma (Lemma 5.4), this implies that
3|z − a|
| f 0 (z) − f 0 (a)| ≤
2ρ02

for z ∈ D(a, ρ0 ). If z ∈ S := D(a, ρ0 /3),

1
| f 0 (z) − f 0 (a)| ≤ = | f 0 (a)|.
2ρ0
By Lemma 5.3, f is one-to-one on S.
 5.1. BLOCH’S THEOREM 77

It remains to show that f (S) contains a disk of radius 1/72. For this
define the function g : D(0, ρ0 /3) → C by g(z) = f (z + a) − f (a). Then
g(0) = f (a) − f (a) = 0, |g0 (0)| = | f 0 (a)| = 1/(2ρ0 ). If z ∈ D(0, ρ0 /3),
then the line segment γ = [a, z + a] lies in S = D(a, ρ0 /3). This follows
because any point in γ is w = a + tz, 0 ≤ t ≤ 1 and |w − a| ≤ t|z| ≤ |z| <
ρ0 /3. Also, S ⊆ D(a, ρ0 ). Since, | f 0 (z)| ≤ 1/ρ0 for z ∈ D(a, ρ0 ), we
have, for z ∈ D(0, ρ0 /3),
Z
|g(z)| = | f (z + a) − f (a)| = f 0 (w)dw
γ
1 1 1 1
Z
≤ | f 0 (w)||dw| ≤ · |z| < · ρ0 = .
γ ρ0 ρ0 3 3

Advanced Complex Analysis by V. Ravichandran

Since g : D(0, ρ0 /3) → D(0, 1/3), g(0) = 0, by Lemma 5.2, we have
( 13 ρ0 )2 (2ρ1 )2
!  
  ρ 
0 0 1
g D 0, ⊃ D 0, = D 0, .
3 6 · 13 72
Let z ∈ S. Then |z − a| < ρ0 /3. Therefore, g(z − a) ∈ D(0, 1/72).
But g(z − a) = f (z) − f (a) and so f (z) − f (a) ∈ D(0, 1/72) or f (z)
∈ D( f (a), 1/72). Thus, f (S) ⊃ D( f (a), 1/72).
Corollary 5.6. If f is analytic on a region containing D(0, R), then
f (D(0, R)) contains a disk of radius (R| f 0 (0)|)/72.
Proof. Consider the function g defined by
f (Rz) − f (0)
g(z) = .
R f 0 (0)
Then g is defined on a region containing D(0, 1) and g(0) = 0. Also,
R f 0 (0)
g0 (0) = = 1.
R f 0 (0)
Thus there exists a disk S ⊆ D on which g is one-to-one and g(S) ⊆
D(0, 1/72). Since f (Rz) = R f 0 (0)g(z) + f (0) for z ∈ D, we have f (D)
contains a disk of radius (R| f 0 (0)|)/72.

Example 5.7. Let f : D → C be given by f (z) = z + z2 /2. Clearly,

f (0) = 0, and f 0 (0) = 1. Also, f 0 (z) = 1 + z and hence | f 0 (z) − 1| =
|z| < 1 for z ∈ D. Thus f is one-to-one in D. Also,
1 1 1
| f (eiθ )| = 1 + eiθ ≥ 1 − = .
2 2 2
Therefore, f (D) contains a disk of radius 1/2.
78 5. THE RANGE OF AN ANALYTIC FUNCTION

Definition 5.8. Let F be the family of all functions f , analytic on a

region containing the closure of the disk D and satisfying f (0) = 0,
f 0 (0) = 1. For each f ∈ F , let β ( f ) be the supremum of all numbers r
such that there is a disk S in D on which f is one-to-one and f (S) con-
tains a disk of radius r. Then Bloch’s constant is the number B defined
by

B = inf{β ( f ) : f ∈ F }.

By Bloch’s Theorem, β ( f ) ≥ 1/72 and so B ≥ 1/72. Since the

function f defined by f (z) = z is in F , it is also clear that B ≤ 1. It is
known that 0.43 ≤ B ≤ 0.47. Also, it has been conjectured that

Advanced Complex Analysis by V. Ravichandran

Γ( 13 ) · Γ( 12
11
)
B= p √ .
1 + 3 · Γ( 41 )

Definition 5.9. Let F be the family as in Definition 5.8. For each f ∈

F , let λ ( f ) be the supremum of all numbers r such that f (D) contains
a disk of radius r. Then Landau’s constant is defined by

L = inf{λ ( f ) : f ∈ F }.

Clearly L ≥ B, and L ≤ 1. Again, is was proved that 0.50 ≤ L ≤ 0.56.

Thus L > B.
Theorem 5.10. If f is analytic on a region containing the closure of the
disk D, f (0) = 0, and f 0 (0) = 1, then f (D) contains a disk of radius L.

Proof. By definition of λ ( f ), f (D) contains a disk of radius r for each

r < λ := λ ( f ). The proposition is proved if we show that f (D) contains
a disk of radius λ . For each n, there is a point αn ∈ f (D) such that
 
1
D αn , λ − ⊆ f (D).
n

The set f (D) is compact and therefore it is sequential compact. Since

αn ∈ f (D), there is a subsequence {αnk } of {αn } such that αnk → α ∈
f (D). Since k ≤ nk , we have λ − 1/k ≤ λ − 1/nk and so
   
1 1
D αnk , λ − ⊆ D αnk , λ − ⊆ f (D).
k nk
 5.2. THE LITTLE PICARD THEOREM 79

Thus, rewriting αnk as αk , we can assume that lim αn = α. We show now

that D(α, λ ) ⊆ f (D). Let w ∈ D(α, λ ) so that |w − α| < λ . Choose n0
such that
1
|w − α| < λ − .
n0
Since λ − 1/n0 − |w − α| > 0, and lim αn = α, there exists n1 > n0 such
that
1
|αn − α| < λ − − |w − α|
n0
for n ≥ n1 . Hence, we have

Advanced Complex Analysis by V. Ravichandran

1 1
|w − αn | ≤ |w − α| + |α − αn | < λ − <λ−
n0 n
for n ≥ n1 . Thus,
 
1
w ∈ D αn , λ − ⊆ f (D).
n
Since w is arbitrary point in D(α, λ ), we have D(α, λ ) ⊆ f (D).
Corollary 5.11. If f is analytic on a region that contains D(0, R), then
f (D(0, R)) contains a disk of radius R| f 0 (0)|L.
Proof. If f 0 (0) = 0, then there is nothing to prove. If f 0 (0) 6= 0, apply
the previous proposition to the function g : D → C defined by g(z) =
( f (Rz) − f (0))/(R f 0 (0)).

5.2. The Little Picard Theorem

Lemma 5.12. Let G be a simply connected region and suppose that f is
an analytic function on G that does not assume the values 0 or 1. Then
there is an analytic function g on G such that

f (z) = − exp[πi cosh(2g(z))], z ∈ G.

Proof. Since f (z) 6= 0, there exists a branch l(z) of log f (z) defined on
G. Then el(z) = f (z). Let
1
F(z) = l(z).
2πi
If F(a) = n for some integer n, then

f (a) = el(a) = e2πiF(a) = e2πin = 1,

80 5. THE RANGE OF AN ANALYTIC FUNCTION

which cannot happen by hypothesis. Hence F does not assumes any

integer
p values.
p In particular, F does not assume 0, 1 and so H(z) =
F(z) − F(z) − 1 is well defined analytic function. Clearly H(z) 6= 0
(For, if H(z) = 0, F(z) = F(z) − 1 or 0 = −1). Since, H(z) 6= 0, there is
a branch g of log H on G and so eg(z) = H(z). Hence,

e2g(z) + e−2g(z)
cosh(2g(z)) + 1 = +1
2
e2g(z) + e−2g(z) + 2e2g(z) e−2g(z)
=
2
g(z)
(e + e −g(z) )2
=

Advanced Complex Analysis by V. Ravichandran

2
 2
1 1
= H(z) +
2 H(z)
!2
1 p p 1
= F(z) − F(z) − 1 + p p
2 F(z) − F(z) − 1
1 p p p p 2
= F(z) − F(z) − 1 + F(z) + F(z) − 1
2
4 1
= F(z) = 2F(z) = l(z).
2 πi
Thus,
f (z) = el(z) = eπi+πi cosh(2g(z)) = −eiπ cosh 2(g(z)) .
Lemma 5.13. Let G be simply connected region. If f is an analytic
function on G that does not assume the values 0 or 1 and g is analytic
function satisfying f (z) = − exp[πi cosh(2g(z))], then g(G) contains no
disk of radius 1.
Proof. Let m.n be integers, with n > 0. If there is a point a ∈ G with
√ √ 1
g(a) = ± log( n + n − 1) + imπ,
2
then

2 cosh[2g(z)] = e2g(a) + e−2g(a)

√ √ √ √
= eimπ ( n + n − 1)±2 + e−imπ ( n n − 1)±2
√ √ √ √
= (−1)m [( n + n − 1)2 + ( n − n − 1)2 ]
 5.2. THE LITTLE PICARD THEOREM 81

= (−1)m 2[n + (n − 1)]

= 2(−1)m (2n − 1)

and therefore
cosh[2g(a)] = (−1)m (2n − 1).
Therefore, f (a) = − exp[iπ(−1)m (2n − 1)] = 1 as (2n − 1)(−1)m is an
odd integer. Hence g cannot assume any of the values
√ √ 1
an,m = ± log( n + n − 1) + imπ, (n ≥ 1, m = 0, ±1, ±2, . . .).
2
These points form vertices of a grid of rectangles in the plane. The

Advanced Complex Analysis by V. Ravichandran

height of an arbitrary rectangle is
imπ i(m + 1)π π √
|an,m − an,m+1 | = − = < 3,
2 2 2
and the width is
√ √ √ √
|an,m − an+1,m | = log( n + 1 + n) − log( n + n − 1) > 0.
√ √ √
For
√ x ≥ 1, consider the function φ (x) = log( x + 1 + x) − log( x +
x − 1). Since
   
0 1 1 1 1 1 1
φ (x) = √ √ √ + √ −√ √ √ + √
x+1+ x 2 x+1 2 x x+ x−1 2 x 2 x−1
1 1
= √ √ − √ √
2 x x+1 2 x x−1
 
1 1 1
= √ √ −√
2 x x+1 x−1
√ √ 
1 x−1− x+1
= √ √ √
2 x x+1 x−1
<0

for x > 1, the function φ (x) is decreasing. Therefore, the width

√ of the
rectangle is less than or equal to φ (1) where φ (1) = log( 2 + 1) <
log e = 1. Thus the diagonal of the rectangle is less than 2. Hence, g(D)
cannot contain a disk of radius 1.
Theorem 5.14 (Little Picard Theorem). If f is an entire function that
omits two values, then f is a constant.
82 5. THE RANGE OF AN ANALYTIC FUNCTION

Proof. If f omits the values a and b, then the function ( f (z)−a)/(b−a)

omits the values 0 and 1. So, assume that f omits 0 and 1 (as, ( f (z) −
a)/(b − a) is constant if and only if f is constant). By Lemma 5.13, we
get that an entire function g such that g(C) contains no disk of radius 1.
If f is not constant, g is not constant. Thus, there exists z0 such that
g0 (z0 ) 6= 0. By considering g(z + z0 ), it may be supposed that g0 (0) 6=
0. By Corollary 5.11, g(D(0, R)) contains a disk of radius R|g0 (0)|L.
Choosing R large such that R|g0 (0)|L > 1, we see that g(C) contains a
disk of radius 1, which is a contradiction. Therefore, f is constant.

Advanced Complex Analysis by V. Ravichandran