Metric Spaces and Complex Analysis

Richard Earl

Michaelmas Term 2014

(Notes for lectures 1—16, and last year’s notes for lectures 17 — 32)
 SYLLABUS

Metric Spaces (10 lectures)

Basic definitions: metric spaces, isometries, continuous functions (ε − δ definition), homeo-

morphisms, open sets, closed sets. Examples of metric spaces, including metrics derived from a
norm on a real vector space, particularly l1 , l2 , l∞ norms on Rn , the sup norm on the bounded
real-valued functions on a set, and on the bounded continuous real-valued functions on a metric
space. The characterization of continuity in terms of the pre-image of open sets or closed sets.
The limit of a sequence of points in a metric space. A subset of a metric space inherits a metric.
Discussion of open and closed sets in subspaces. The closure of a subset of a metric space. [3.5]

Completeness (but not completion). Completeness of the space of bounded real-valued

functions on a set, equipped with the norm, and the completeness of the space of bounded
continuous real-valued functions on a metric space, equipped with the metric. Lipschitz maps
and contractions. Contraction Mapping Theorem. [2]

Connected metric spaces, path-connectedness. Closure of a connected space is connected,

union of connected sets is connected if there is a non-empty intersection, continuous image of
a connected space is connected. Path-connectedness implies connectedness. Connected open
subset of a normed vector space is path-connected. [2]

Compactness. Heine-Borel theorem. The image of a compact set under a continuous map
between metric spaces is compact. The equivalence of continuity and uniform continuity for
functions on a compact metric space. Compact metric spaces are sequentially compact. State-
ment (but no proof) that sequentially compact metric spaces are compact. Compact metric
spaces are complete. [2.5]

Reading

1. W. A. Sutherland, Introduction to Metric and Topological Spaces (Second Edition, OUP,

2009).

1
1. Metric Spaces. Convergence and Continuity.
Topology, loosely speaking, is the study of continuity. Consequently the objects of interest of
topology are those properties that remain invariant under continuous deformations: so angle
and area are out (being more rigid geometrical notions), but ideas of shape (e.g. connectedness
= being in one piece) are in.
Quite what the rigorous definition of a continuous function f : R → R should be wasn’t
properly understood until the 19th century, with the work of Weierstrass and others, but the
importance of having calculus and analysis on a rigorous footing was becoming very clear.
Generalizing the definition of continuity to maps f : Rn → Rm was relatively straightforward
but it would take much effort to fully generalize these definitions to metric and topological
spaces (around the start of the 20th century), and to properly appreciate which properties of a
"closed, bounded interval" had led to continuous functions being bounded and achieving those
bounds, or to the Intermediate Value Theorem.
We first revisit the definition of convergence and continuity in Euclidean space.

Definition 1 Given x, y ∈ Rn the distance between x and y is

 n

|x − y| =  (xi − yi )2 .
i=1

Now, with this notion of distance in Rn , we can easily generalize the notions of convergence
and continuity that we met in the first year. Namely:

Definition 2 Let (xk ) be a sequence in Rn and L ∈ Rn . We say that (xk ) converges to L if

|xk − L| → 0 as k → ∞

or equivalently if
∀ε > 0 ∃K ∈ N ∀k
K |xk − L| < ε.
 
(k) (k) (k)
If xk = x1 , x2 , . . . , xn and L = (L1 , L2 , . . . , Ln ) then (xk ) converging to L is equivalent
(k)
to each sequence of coordinates xi converging to Li as k → ∞.

We can similarly generalize our definition of continuity:

Definition 3 A function f : Rn → Rm is said to be continuous at x ∈ Rn if

lim f(y) = f(x)

y→x

or equivalently if

∀ε > 0 ∃δ > 0 such that whenever |y − x| < δ then |f(y) − f(x)| < ε.

METRIC SPACES. CONVERGENCE AND CONTINUITY. 2

 From a topological point of view we will identify C with R2 . (Though of course there are
other important ways in which the two are different, for example when it comes to their algebra
or geometry.) We start with the following simple fact:
Proposition 4 f : C → C is continuous at z if and only if Re f and Im f are continuous at z.
Proof Suppose that f is continuous at z = x + iy and let ε > 0. Then there exists δ > 0 such
that
|z − w| < δ =⇒ |f(z) − f(w)| < ε.
Noting |Re ζ|  |ζ| and |Im ζ|  |ζ| for any ζ ∈ C, we have
|z − w| < δ =⇒ |(Re f )(z) − (Re f )(w)| = |Re(f(z) − f(w))| < ε,
with a similar inequality for Im and hence Re f and Im f are both continuous at z. Conversely
suppose that Re f and Im f are both continuous at z and let ε > 0. Then there exist positive
δ R and δ I such that
|z − w| < δ R =⇒ |Re f(z) − Re f (w)| < ε/2; |z − w| < δ I =⇒ |Im f (z) − Im f (w)| < ε/2.
If we set δ = min {δ R , δ I } then we have by the Triangle Inequality
|z − w| < δ =⇒ |f (z) − f (w)| = |(Re f (z) − Re f(w)) + i(Im f (z) − Im f (w))|
 |Re f (z) − Re f(w)| + |Im f (z) − Im f(w)|  ε/2 + ε/2 = ε
and hence f is continuous at z.

Here’s a result from first year analysis that we see generalizes:

Proposition 5 (Uniqueness of Limits) Let (xk ) be a sequence in Rn which converges to L
and also to M. Then L = M.
Proof Suppose for a contradiction that L = M. Then we may set ε = |L − M| /2 > 0.
As xk → L then there exists K1 such that |xk − L| < ε when k
K1 ;
as xk → M then there exists K2 such that |xk − M| < ε when k
K2 .
Hence for k
max{K1 , K2 } we have
|L − M| = |(L − xk ) + (xk − M)|  |L − xk | + |xk − M| (1.1)
= |L − xk | + |M − xk | < ε + ε = |L − M| , (1.2)
which is our required contradiction.

In the hope of generalizing the definitions of continuity and convergence still further, we
might ask ourselves this question: what underlying properties of distance are important? What-
ever the setting, it seems important that limits should still be unique. We might not expect
an "algebra of limits" or "a sandwich lemma"; depending on context these might not make
sense. If we are interested in the question "does the Magnetic North Pole move continuously
with time?" then algebra makes no sense as we can’t reasonably add points on a sphere. But
the above proposition is more fundamental topologically.

METRIC SPACES. CONVERGENCE AND CONTINUITY. 3

 What properties of distance does the proof use? The following stand out:
• Distances are nonnegative and distinct points are a positive distance apart. We use this
immediately in the first line.
• Symmetry. We make use of |xk − M| = |M − xk | in (1.2).
• Triangle Inequality. We make use of this to get the first inequality in (1.1).
In light of this we define:
Definition 6 Let M be a set. A metric on M is a map d : M × M → [0, ∞) such that
(M1) d(m1 , m2 ) = 0 if and only if m1 = m2 .
(M2) d(m1 , m2 ) = d(m2 , m1 ) for any m1 , m2 ∈ M.
(M3) d(m1 , m3 )  d(m1 , m2 ) + d(m2 , m3 ) for any m1 , m2 , m3 ∈ M.
In generalizing the distance properties to Rn we have met our first example of a metric space.
Example 7 (Euclidean Space) Let M = Rn and define

d2 (x, y) = (x1 − y1 )2 + · · · + (xn − yn )2 .
Example 8 Let M = Rn and
d1 (x, y) = |x1 − y1 | + · · · + |xn − yn | .
Example 9 Let M = Rn and
d∞(x, y) = max{|x1 − y1 | , . . . , |xn − yn |}.
Proposition 10 Let x, y ∈ Rn . Show that
√ 1
d2 (x, y)  d1 (x, y)  nd2 (x, y); √ d2 (x, y)  d∞ (x, y)  d2 (x, y).
n
Proof This is seen most easily diagrammatically. Let ε > 0 and consider the sets
B1 = {x ∈ Rn : d1 (x, 0) = ε} ; B2 = {x ∈ Rn : d2 (x, 0) = ε} ; B∞ = {x ∈ Rn : d∞ (x, 0) = ε} .
These sets are sketched below for n = 2.
H0,ΕL d¥= Ε

d2 = Ε

d1 = Ε

O HΕ,0L

METRIC SPACES. CONVERGENCE AND CONTINUITY. 4

Measuring with
√ d2 , we see that the farthest B1 is from 0 is ε at (ε, 0, . . . , 0) (say) and the
nearest is ε/ n at
√ (ε/n, ε/n, . . . , ε/n) (say). Again measuring with d2 , we see that the farthest
B∞ is from 0 is nε at (ε, ε, . . . , ε) (say) and the nearest is ε at (ε, 0, . . . , 0) (say). The result
follows as the geometry of these sets is not affected by translation — i.e. we can assume y = 0
without any loss of generality.
Example 11 (Discrete Metric) Let M be any set and define d on M 2 by

0 if x = y;
d(x, y) =
1 if x = y.
Then d is a metric on M , referred to as the discrete metric. M1 and M2 are easily seen to
be met and M3 can be verified on a case-by-case basis
Case d(x, z) d(x, y) + d(y, z) Case d(x, z) d(x, y) + d(y, z)
x=y=z 0 0 x = z = y 0 2
x = y = z 1 1 x = y = z = x 1 2
x = y = z 1 1
Example 12 Let V be a real inner product space. Distance is defined in V by
d(x, y) = ||x − y||
where ||v||2 = v, v for any v ∈ V and this gives rise to a metric. We refer to ||v|| as the
norm of v.
Example 13 Let C[a, b] denote the space of continuous functions f : [a, b] → R. As
b
f, g = f (t)g(t) dt
a

defines an inner product on C[a, b] then we have an associated metric

b
d2 (f, g) = (f(t) − g(t))2 dt.
a

Example 14 A second metric on C[a, b] is given by

d∞ (f, g) = sup {|f (x) − g(x)| : a  x  b} .
M1 and M2 are clear. To show M3, note that for any x ∈ [a, b] we have
|f (x) − h(x)|  |f(x) − g(x)| + |g(x) − h(x)|
 sup |f(x) − g(x)| + sup |g(x) − h(x)| .
axb axb

As sup respects weak inequalities then

sup |f (x) − h(x)|  sup |f (x) − g(x)| + sup |g(x) − h(x)|
axb axb axb

and so M3 follows for d∞ . More generally d∞ defines a metric on the space of bounded real-
valued functions on [a, b] .

METRIC SPACES. CONVERGENCE AND CONTINUITY. 5

Definition 15 Let V be a real vector space. A norm on V is a map || || : V → [0, ∞) such
that

(N1) ||v|| = 0 if and only if v = 0V .

(N2) ||λv|| = |λ| × ||v|| for v ∈ V and λ ∈ R.

(N3) ||x + y||  ||x|| + ||y|| for x, y ∈ V.

A vector space with a norm is called a normed vector space. As the norm associated with
an inner product satisfies N1, N2, N3 then inner product spaces are normed vector spaces.
For any normed vector space, d(x, y) = ||x − y|| defines a metric on V.

Example 16 (Sequence Spaces) We define now the following spaces of real sequences.
   
l1 = (xn ) : |xn | converges , ||(xn )||1 = |xn | .
   
l2 = (xn ) : |xn |2 converges , ||(xn )||2 = |xn |2 .
l∞ = {(xn ) : |xn | is bounded} , ||(xn )||∞ = sup |xn | .
n

These spaces are all normed vector spaces. In fact l2 is an inner product space with its norm
coming from the inner product 
(xn ) , (yn ) = xn yn .

Example 17 (Subspace metrics) Let (M, d) be a metric space and A ⊆ M. Then d naturally
induces a metric dA on A by setting

dA (a, b) = d(a, b) for a, b ∈ A.

Recall that our aim here was to put our notions of continuity and convergence on a more
general footing. We can see that Definitions 2 and 3 can be naturally generalized as follows:

Definition 18 Let (xk ) be a sequence in a metric space (M, d) and x ∈ M . We say that (xk )
converges to x if
d(xk , x) → 0 as k → ∞
or equivalently if
∀ε > 0 ∃K ∈ N ∀k
K d(xk , x) < ε.

Definition 19 A function f : M → N between metric spaces (M, dM ) and (N, dN ) is said to

be continuous at x ∈ M if

∀ε > 0 ∃δ > 0 such that whenever dM (x, y) < δ then dN (f (x), f (y)) < ε.

Example 20 Consider the sequence of functions fn (x) = xn in C[0, 1]. Show that the sequence
(fn ) converges in (C[0, 1], d2 ) but not in (C[0, 1], d∞) .

METRIC SPACES. CONVERGENCE AND CONTINUITY. 6

Solution (i) Pointwise fn (x) converges to

0 0x<1
f (x) =
1 x=1
but this is not continuous. However g = 0 is continuous and differs from f at only one point
(x = 1). Note that
1 1
2 2 1
d2 (fn , g) = (fn (x) − g(x)) dx = x2n dx = →0
0 0 2n + 1
demonstrating that the sequence does indeed converge to g in the d2 -metric.
(ii) On the other hand suppose that there were a continuous function h such that fn con-
verges to h in the d∞ metric. Then
sup |fn (x) − h(x)| → 0 as n → ∞.
0x1

In particular this means that for any fixed x we have |fn (x) − h(x)| → 0 as n → ∞. So
h(x) = lim fn (x) = f(x). However as f is not continuous we see that (fn ) is not a convergent
sequence in (C[0, 1], d∞ ).

Remark 21 You may note that a sequence (gn ) converges to g in (C[0, 1], d∞) if and only if
gn converges uniformly to g.

Example 22 Show that the map f : l∞ → l2 given by f ((xn )) = (xn /n) is well-defined and
continuous.

Solution If (xn ) ∈ l∞ then xn is bounded, by M say. So


 |xn |2  1 Mπ
||(xn /n)||2 = 2
 M 2
= √ <∞
n n 6
√
and (xn /n) ∈ l2 . Now let ε > 0 and set δ = 6ε/π. If ||(yn ) − (xn )||∞ < δ then |yn − xn | < δ
for each n. Hence


 |yn − xn |2  1 δπ
||f ((yn )) − f ((xn ))||2 = ||(yn /n) − (xn /n)||2 = 2
< δ 2
= √ = ε.
n n 6

Proposition 23 Let (M, d) be a metric space and a ∈ M . Define f : M → R by f (x) = d(x, a).
Then f is continuous.

Proof Let ε > 0 and x ∈ M. Set δ = ε and take y ∈ M such that d(x, y) < δ. Then
|f (y) − f(x)| = |d(y, a) − d(x, a)|  d(x, y) < δ = ε
by the triangle inequality.

We finish by generalizing a result of first year analysis.

METRIC SPACES. CONVERGENCE AND CONTINUITY. 7

Theorem 24 Let f : M → N be a map between metric spaces. Then f is continuous at x ∈ M
if and only if whenever xk → x in M then f (xk ) → f (x) in N.

Proof Suppose firstly that f is continuous. Let ε > 0. By continuity there exists δ > 0 such
that
dM (y, x) < δ =⇒ dN (f(y), f (x)) < ε.
As xk → x then there exists K such that dM (xk , x) < δ when k
K. So dN ((f(xk ), f (x)) < ε
when k
K and hence f (xk ) → f (x) in N .
Conversely, suppose that f is not continuous at x ∈ M . Then there exists ε > 0 such that
for every k there is xk ∈ M with

dM (xk , x) < 1/k and yet dN (f (xk ), f (x))
ε.

In this case xk → x in M and yet f (xk )  f (x) in N .

METRIC SPACES. CONVERGENCE AND CONTINUITY. 8

2. Open and Closed Sets
Throughout the following (M, d) will be taken to be a metric space. We introduce here the
notion of open balls — and, in due course, open sets — which will allow us to rephrase our
definition of continuity.

Notation 25 Let x ∈ M and ε > 0. Then the open ball B(x, ε) and closed ball B̄(x, ε) are
defined as

B(x, ε) = {y ∈ M : d(x, y) < ε} , B̄(x, ε) = {y ∈ M : d(x, y)  ε} .

If we wish to stress that we are working in M we will write BM (x, ε) and B̄M (x, ε).

We can see that Definition 19, regarding the continuity of a function f : M → N between
metric spaces can be rephrased as: f is continuous at x if

∀ε > 0 ∃δ > 0 f (BM (x, δ)) ⊆ BN (f (x), ε). (2.1)

This rephrasing will become still more streamlined and elegant once we have introduced the
more general idea of open sets.

Definition 26 U ⊆ M is said to be open in M (or simply open) if

∀x ∈ U ∃ε > 0 B(x, ε) ⊆ U.

We say that x is an interior point of U if there exists ε > 0 such that B(x, ε) ⊆ U , so saying
U is open is equivalent to saying all its points are interior points.

Remark 27 We will refer to such U as open sets but really it would be better to think of U as
an open subset of M . Because the ambient space M will usually be obvious it’s usually safe
to suppress the "in M " bit, but we will see instances where the same set is open as a subset of
one ambient set, but not open as a subset of another. So take some care.

So (2.1) can be rephrased as: f : M → N is continuous at x if

whenever f (x) is an interior point of U ⊆ N then x is an interior point of f −1 (U) ⊆ M

or can be rephrased still more cleanly when we are talking about a function which is continuous
everywhere on M as:

Theorem 28 Let f : M → N be a map between metric spaces (M, dM ) and (N, dN ). Then f
is continuous if and only if whenever U ⊆ N is open then f −1 (U ) ⊆ M is open.

Remark 29 Hopefully this helps motivate the importance of open sets as, with this result, we
see they are sufficient to determine which functions are continuous.

OPEN AND CLOSED SETS 9

Proof Suppose that f is continuous and that U is an open subset of N. Let x ∈ f −1 (U ) so
that f(x) is an interior point of U (as U is open). So there exists ε > 0 such that

BN (f (x), ε) ⊆ U.

As f is continuous at x there exists δ > 0 such that

f (BM (x, δ)) ⊆ BN (f (x), ε) ⊆ U

and hence BM (x, δ) ⊆ f −1 (U ), showing that x is an interior point of f −1 (U ). As x was an

arbitrary point, then f −1 (U) is open.
Conversely suppose that the pre-image of every open set in N is open in M . Let ε > 0 and
x ∈ M. As BN (f (x), ε) is open in N , by assumption x is an interior point of f −1 (BN (f (x), ε))
and so there exists δ > 0 such that

BM (x,δ) ⊆ f −1 (BN (f (x), ε)) and hence f (BM (x,δ)) ⊆ BN (f(x), ε),

which is just another way of saying the f is continuous at x.

Definition 30 We refer to the open sets of M as the topology of M , or the topology induced
by the metric d.

Proposition 31 (a) The intersection of finitely many open sets is open.

(b) An arbitrary union of open sets is open.
m

Proof (a) Let U1 , . . . , Um be (finitely many) open subsets of M . Let x ∈ Ui . Then, for
i=1
each i, there exists εi > 0 such that B(x, εi ) ⊆ Ui . Let ε = min {ε1 , ε2 , . . . , εm } > 0 and then
m
B(x, ε) ⊆ Ui .
i=1 
(b) For i ∈ I (a not-necessarily finite indexing set), let Ui be an open subset. Let x ∈ Ui .
i∈I
Then there exists j ∈ I such that x ∈ Uj and ε > 0 such that B(x, ε) ⊆ Uj . Then

B(x, ε) ⊆ Uj ⊆ Ui .
i∈I

Remark 32 (Off-syllabus) We saw in Theorem 28 that knowledge of the open sets is sufficient
to determine which functions are continuous. So the open sets are more fundamental to conti-
nuity than the actual metric (many different metrics will lead to the same family of open sets).
In line with this thinking a topological space is defined as follows.
A topological space (X, T ) is a set X together with a family T of subsets of X with the
following properties:
(i) ∅ ∈ T and X ∈ T ;

OPEN AND CLOSED SETS 10

 (ii) T is closed under finite intersections;
(iii) T is closed under arbitrary unions.
T is called the topology and U is said to be open when U ∈ T .
So all metric spaces are topological spaces though not all topologies are induced by metrics.
However a topological space (which is a much more general notion) is sufficient to describe many
aspects of continuity and indeed the proofs are often neater and more natural in that context
without the unnecessary and messy inequalities associated with a metric. Those interested in
this should continue with the Hilary Term long option in Topology.

Example 33 An infinite collection of open sets need not have an open intersection. For ex-
ample, consider Ui = (−i, i) for i > 0 which have intersection {0} which is not open.

Definition 34 C ⊆ M is said to be closed in M (or simply closed) if M \C is open.

Remark 35 Note that open and closed are not opposites of one another! Sets may be open,
closed, neither or both, as we shall see.

Example 36 (a) ∅ and M are both open and closed in M . That ∅ is open is "vacuously true"
and M is clearly open by definition.
(b) The interval [0, 1) is neither open nor closed in R. To see this, note it is not open as 0
is not an interior point of [0, 1) and that the set is not closed as 1 is not an interior point of
the complement.
(c) The subset [0, ∞) × (0, ∞) ⊆ R2 is neither open nor closed. The point (0, 1) is not
interior to the subset and (1, 0) is not interior to its complement.
(d) The subset Z2 is closed but not open in R2 .
(e) The set {(x, y) : x2 + xy + y 2 < 1} is open. To see this we can note that f : R2 → R
given by f(x, y) = x2 +xy +y 2 is continuous and then the set in question is f −1 ((−∞, 1)) which
is the preimage by a continuous function of an open set (Theorem 28).
(f) Let M = [0, 1] ∪ [2, 3]. Then [0, 1] is both open and closed in M . Likewise [2, 3] . (We
will comment more on such examples in the chapter Subspaces.)

Proposition 37 Let x ∈ M and ε > 0. Then B(x, ε) is open and B̄(x, ε) is closed.
1
Proof Firstly if y ∈ B(x, ε) then let δ = 2
(ε − d(y, x)) > 0. By the triangle inequality, if
z ∈ B(y, δ) then

ε + d(y, x) ε+ε
d(z, x)  d(z, y) + d(y, x) < δ + d(y, x) = < = ε.
2 2
Hence B(y, δ) ⊆ B(x, ε) and B(x, ε) is open.
To prove the second result, take y ∈ M \B̄(x, ε), Then d(x, y) > ε and set δ = 21 (d(x, y)−ε) >
0. If d(z, y) < δ then by the triangle inequality

d(z, x)
d(x, y) − d(z, y) > d(x, y) − δ = ε + δ > ε

and hence z ∈ M\B̄(x, ε). That is M \B̄(x, ε) is open and B̄(x, ε) is closed.

OPEN AND CLOSED SETS 11

Proposition 38 Singleton sets are closed.

Proof Let x ∈ M . If y = x we may set ε = d(x, y) > 0. Then x ∈

/ B(y, ε) and hence y is
interior to M\{x}.

Example 39 Note, when M = (0, 1) ∪ {2}, that BM (2, 1) = B̄M (2, 1) = {2}. In this case this
open ball is also closed.

Definition 40 Let S ⊆ M and x ∈ M . We say that x is a limit point of S or an accumu-

lation point of S if for any ε > 0

(B(x, ε) ∩ S) \{x} = ∅.

i.e. there are points of S arbitrarily close to x other than x, which may or may not be in S.
We will denote the set of limit points of S as S ′ .

Limit Point in set

Interior Point Not a

Limit
Point

Limit Point not in set

Interior and limit points of a set

Example 41 (a) Let M = R and S = [0, 1). Then S ′ = [0, 1] . But if M = S = [0, 1), then
S ′ = [0, 1).
(b) Let M = R and S = (−∞, 1) ∪ {2}. Then S ′ = (−∞, 1]. So a point of S need not be a
point of S ′ .

Proposition 42 C ⊆ M is closed if and only if it contains all its limit points.

Proof Note that x is an interior point of M \C if and only if x is not a limit point of C. So

C is closed ⇐⇒ M\C is open

⇐⇒ every point of M\C is an interior point
⇐⇒ no point of M\C is a limit point of C
⇐⇒ C contains all its limit points.

The results corresponding to Proposition 31 for closed sets are:

OPEN AND CLOSED SETS 12

Proposition 43 (a) The union of finitely many closed sets is closed.
(b) An arbitrary intersection of closed sets is closed.

Proof These simply follow from applying De Morgan’s laws to the equivalent properties of
open subsets shown in Proposition 31.

And we also have:

Corollary 44 (To Theorem 28) Let f : M → N be a map between metric spaces M and N.
Then f is continuous if and only if whenever C is closed in N then f −1 (C) is closed in M.

Proof Noting that for any S ⊆ N we have f −1 (N\S) = M \f −1 (S), we have

f is continuous ⇐⇒ f −1 (U ) is open in M whenever U is open in N

⇐⇒ f −1 (N\C) is open in M whenever C is closed in N
⇐⇒ M \f −1 (C) is open in M whenever C is closed in N
⇐⇒ f −1 (C) is closed in M whenever C is closed in N.

Corollary 45 Open balls are open and closed balls are closed.

Proof We showed this in Proposition 37. However this is now an easy application of Theorem
28 and Corollary 44 to the continuous function f (x) = d(x, a) from Proposition 23 as

B(a, ε) = f −1 ((−∞, ε)) and B̄(a, ε) = f −1 ((−∞, ε]).

Example 46 The sphere x2 + y 2 + z 2 = a2 is closed in R3 . This is because {a2 } is closed in R

and f : R3 → R is continuous where

f(x, y, z) = x2 + y 2 + z 2 .

The sphere is then f −1 ({a2 }).

Example 47 Let c denote the subset of l∞ consisting of all convergent sequences. Then c is a
closed subset of l∞ .
 
(k)
Solution Say that xn is a sequence in c which converges to (Xn ) in l∞ as k → ∞. We aim
to show that (Xn )is inc; we will do this by showing (Xn ) is Cauchy and so convergent.
(k)
Let ε > 0. As xn → (Xn) in l∞ then
 (k)    
 xn − (Xn )
∞
= sup x(k) 
n − Xn → 0 as k → ∞.
n

OPEN AND CLOSED SETS 13

So there exists K such that
 (k) 
x − Xn  < ε/3 for k
K and all n.
n
 
(K)
As xn is convergent then it is Cauchy. So there exists N such that
 (K) 
x − x(K)  < ε/3 for n, m
N.
n m

Thus for m, n
N we have

   (K)   (K) 
|Xm − Xn |  Xm − x(K)
m
 + xm − x (K) 
n + x n − Xn
 < ε/3 + ε/3 + ε/3 = ε.

Example 48 Given a metric space M, x ∈ M and ε > 0, the limit points of B(x, ε) need not
include B̄(x, ε).

Solution Take the discrete metric on a set M and x ∈ M . Then B(x, 1) = {x} whilst
B̄(x, 1) = M. However given any y = x we see that x ∈ / B(y, 1) and so y is not a limit point of
B(x, 1). In fact there are no limit points of B(x, 1).

OPEN AND CLOSED SETS 14

3. Subspaces. Homeomorphisms. Isometries.
When describing any type of structure in mathematics, it is natural to discuss the possible
substructures and isomorphisms.

Again throughout this (M, d) will be a metric space. As we have already noted (Example
17) the metric d on M induces a metric dA on any subset A of M so that (A, dA ) is a metric
space in its own right.

Recall the following Definitions 26 and 34.

• Say S ⊆ T ⊆ M. We say S is open in T if for every s in S there exists ε > 0 such that

BT (s, ε) = {x ∈ T : dT (x, s) < ε} ⊆ S.

But as dT (x, s) = d(x, s) for x, s ∈ T then

BT (s, ε) = {x ∈ T : d(x, s) < ε} = BM (s, ε) ∩ T.

• We say that S is closed in T if T \S is open in T.

Definition 49 We refer to (T, dT ) as a subspace of (M, d) and refer to the open sets of T as
the subspace topology of T.

As (T, dT ) is a metric space in its own right, it remains the case that:

• Given an arbitrary collection of open subsets of T, their union is open in T.

• Given finitely many open subsets of T, their intersection is open in T.

• Given an arbitrary collection of closed subsets of T, their intersection is closed in T.

• Given finitely many closed subsets of T, their union is closed in T.

Example 50 (a) [0, 1) is open in [0, 1]. Note that, for suitably small ε > 0, we have

B[0,1] (x, ε) = (x − ε, x + ε)

for 0 < x < 1 and

B[0,1] (0, ε) = [0, ε).
So given any x in the range 0  x < 1, we can find ε > 0 such that B[0,1] (x, ε) ⊆ [0, 1)
(b) Given any X ⊆ M then X is open and closed in itself. So the empty set is also an open
and closed subset of X.
(c) Every subset of Z is open and closed in Z. Note that for any integer n we have
BZ (n, 1/2) = {n}. As singleton points are open in Z, and an arbitrary union of open sets
is open, then any subset is open in Z.
(d) Consequently any map f : Z → M, to any metric space M, is continuous.

SUBSPACES. HOMEOMORPHISMS. ISOMETRIES. 15

Example 51 Let X = [0, 1] ∪ [2, 3]. The map f : X → Z defined by

0 0x1
f(x) =
1 2x3

is continuous. Given S ⊆ Z then f −1 (S) equals

∅, [0, 1] , [2, 3] , X,

depending on whether neither, 0, 1 or both lie in S. All these possible pre-images are open in
X.

Proposition 52 Let S ⊆ T ⊆ M.
(a) S is open in T if and only if there is an open set U (in M ) such that S = T ∩ U.
(b) S is closed in T if and only if there is a closed set C (in M ) such that S = T ∩ C.

Proof (a) Suppose that S is open in T . Then for each s ∈ S there exists εs > 0 such that
BT (s, εs ) ⊆ S. So
 
  
S= BT (s, εs ) = (B(s, εs ) ∩ T ) = B(s, εs ) ∩ T
s s s

and we set U = B(s, εs ). Conversely if S = T ∩ U for some open U then for any s ∈ S ⊆ U
s
there exists ε > 0 such that B(s, ε) ⊆ U and then BT (s, ε) ⊆ U ∩ T = S.
(b) Note that

S is closed in T ⇐⇒ T \S is open in T
⇐⇒ T \S = T ∩ U for some open U in M
⇐⇒ S = T ∩ (M\U ) for some open U in M
⇐⇒ S = T ∩ C for some closed C in M.

Example 53 (a) (a, b) ∩ Q is open in Q for all choices of a, b and is closed in Q if and only
a and b are irrational.
By the above (a, b) ∩ Q is always open in Q. Further if a and b are irrational then

(a, b) ∩ Q = [a, b] ∩ Q

is closed in Q. If, say, a is rational then a is not an interior point of the complement and the
set is not closed in Q.
(b) Let S = (0, 1] ∪ [2, 3]. Then (0, 1] and [2, 3] are open and closed subsets of S.

(0, 1] = (0, 2) ∩ S = [0, 1] ∩ S; [2, 3] = (1, 4) ∩ S = [2, 3] ∩ S.

SUBSPACES. HOMEOMORPHISMS. ISOMETRIES. 16

Example 54 Let A ⊆ B ⊆ C ⊆ M.
(a) Show that if A is open in B and B is open in C, then A is open in C.
(b) Show that if A is closed in B and B is closed in C, then A is closed in C.

Solution (a) As B is open in C then there exists open U such that B = U ∩ C. As A is open
in B then there exists open V such that A = V ∩ B. Hence

A = V ∩ B = (V ∩ U) ∩ C

and V ∩ U is open. Part (b) follows similarly.

Proposition 55 Let A ⊆ M and f : M → N be a continuous map between metric spaces M

and N. Then the restriction f |A of f to A is continuous.

Proof Let U be an open subset of N. Then

(f|A )−1 (U) = f −1 (U) ∩ A

which is open in A as f −1 (U ) is open in M.

Definition 56 Given A ⊆ M the closure of A, written A, is the smallest closed subset of M

which contains A. This is well-defined as it equals

A= {C : C closed in M and A ⊆ C} .

We saw in Sheet 1, Exercise 5(ii), that A = A ∪ A′ where A′ is the set of limit points of A.
M B
Proposition 57 Let A ⊆ B ⊆ M. Let A and A respectively denote the closures of A in M
and of A in B. Then
B M
A = A ∩ B.

Proof We have that

B 
A = {C : C closed in B and A ⊆ C}

= {D ∩ B : D closed in M and A ⊆ D}

= {D : D closed in M and A ⊆ D} ∩ B
M
= A ∩ B.

The notion of isomorphism for metric spaces is that of an isometry.

Definition 58 An isometry between metric spaces (M, dM ) and (N, dN ) is a bijection f : M →

N such that
dN (f(x), f (y)) = dM (x, y) for all x, y ∈ M. (3.1)
In such a case we would say that M and N are isometric. Clearly f −1 is also an isometry.

SUBSPACES. HOMEOMORPHISMS. ISOMETRIES. 17

Remark 59 Sometimes a map f : M → N will be termed an isometry if (3.1) holds. This is
sufficient to guarantee that f is 1-1 but such an "isometry" need not be onto.

Proposition 60 (a) An isometry is continuous.

(b) The isometries of a metric space form a group under composition.
(c) The isometry group of Rn is the so-called Euclidean group. This consists of all maps
of the form x → Ax + b where A ∈ O(n) and b ∈ Rn .
(d) The isometry group of the unit sphere S n−1 ⊆ Rn is O(n).

Proof (a) and (b) are trivial. (c) and (d) are results from (or using) the first year geometry
course.

Example 61 Let (M, dM ) be a metric space, N be a set and f : M → N be a bijection. We

can then use f to induce a metric dN on N in such a way that M and N are isometric. We
simply define
dN (x, y) = dM (f −1 (x), f −1 (y)).
For example, this immediately means that

d(x, y) = |ln x − ln y|

is a metric on (0, ∞) as this is the metric induced from the bijection exp : R → (0, ∞) .

From a point of view of continuity, though, isometries really ask too much. We are aware
of non-isometric metrics on the same space leading to the same topology (i.e. the same family
of open sets) and so to the same family of continuous functions. What we really need is the
idea of a bijection between two spaces that induces a bijection between the topologies. This is
the idea of a homeomorphism.

Definition 62 Let M and N be metric spaces. A homeomorphism f : M → N is a bijection

f such that f and f −1 are continuous. Note, in particular, that this induces a bijection between
the two spaces’ topologies. (i.e. if U is open in M then f (U ) is open in N and if V is open in
N then f −1 (V ) is open in M ).

Example 63 An isometry is a homeomorphism.

Example 64 (0, 1) and R are homeomorphic (and not isometric).

Solution The map f : (0, 1) → R given by

2x − 1
f (x) =
x(1 − x)

is a homeomorphism. (This is clear from a sketch of its graph.) The two spaces are clearly not
isometric as 0 and 2 are distance 2 apart in R and no such points exist in (0, 1) .

Example 65 N and Z are homeomorphic. Both spaces have the discrete topology (all subsets
are open) and so any bijection between the two is automatically a homeomorphism.

SUBSPACES. HOMEOMORPHISMS. ISOMETRIES. 18

Example 66 The homeomorphisms of R are the continuous, strictly monotone maps which
are neither bounded above nor below.

Example 67 R and R2 are not homeomorphic.

How would one show this last result? The simple answer is that we can’t prove this just yet,
but posing the question motivates some of our later concepts. If we were faced with a similar
question in group theory — for example: show Z and Q are not isomorphic groups — then we
would note that Z is cyclic and that Q is not. "Being cyclic" is an algebraic invariant preserved
by isomorphisms. What we need are similar topological invariants to help us distinguish between
these two spaces.

SUBSPACES. HOMEOMORPHISMS. ISOMETRIES. 19

4. Completeness
Throughout the following (M, d) is a metric space.

Definition 68 We say that a sequence (xn ) in M is Cauchy if d(xm , xn ) → 0 as m, n → ∞

or equivalently
∀ε > 0 ∃N ∀m, n
N d(xm , xn ) < ε.

Proposition 69 A convergent sequence is Cauchy.

Proof Say xn → x in M . Let ε > 0. Then there exists N such that d(xn , x) < ε/2 for all
n
N. By the triangle inequality

d(xn , xm )  d(xn , x) + d(x, xm ) < ε/2 + ε/2 = ε for all m, n
N.

Example 70 In R and C every Cauchy sequence is convergent. In Q this is not the case, for
example
3, 3.1, 3.14, 3.141, 3.1415, . . .
is a Cauchy sequence in Q which does not converge in Q. Note 1/n is a Cauchy sequence in
(0, 1) which does not converge in (0, 1) .

Definition 71 A metric space M is said to be complete if every Cauchy sequence in M

converges in M.

Proposition 72 Let (X, d) be a complete metric space and Y ⊆ X. Then Y is complete if and
only if Y is closed in X.

Proof Suppose that Y is closed in X and let (yn ) be a Cauchy sequence in Y . Then (yn ) is
also Cauchy in X and so convergent to x ∈ X as X is complete. As Y is closed then x ∈ Y
and we see that Y is complete. Conversely suppose that Y is complete and y ∈ X is a limit
point of Y . Then there is a sequence (yn ) in Y converging to y. But convergent sequences are
Cauchy and so y ∈ Y as Y is complete. Hence Y contains all its limit points and so is closed
in X.

Example 73 Note that (0, 1) and R (with the usual topologies) are homeomorphic and yet R
is complete and (0, 1) is not. This shows that completeness is not a topological invariant (i.e. is
not preserved by homeomorphisms). Completeness is preserved by isometries but the equivalent
topological invariant is that of complete metrizability.
For example, the map f : (0, 1) → R given by f (x) = tan (π(x − 1/2)) is a homeomorphism.
So if we define the metric d on (0, 1) by d(x, y) = |f (x) − f(y)| then ((0, 1) , d) and R (with
the usual metric) are isometric. In particular ((0, 1) , d) is complete whilst still having the usual
topology.

COMPLETENESS 20
Example 74 The spaces l1 , l2 , l∞ are all complete.
     
(1) (2) (3)
Solution We shall prove this only for l1 . Let xn , xn , xn , . . . be a Cauchy sequence
in l1 . This means that given any ε > 0 there exists K such that
∞

 (k)   (l)   (k) 
 xn − xn  = xn − x(l)
n
 < ε for k, l
K (4.1)
1
n=1
 
 (k) (l)  (k)
In particular this means that xn − xn  < ε for k, l
K and so xn is a Cauchy sequence
which converges to some Xn as k → ∞. Note that for any N and k
K we have
N
 N
 N
  
(k) 
 (k)   
|Xn |  Xn − xn + xn   ε +  x(k)
n

1
n=1 n=1 n=1

and so (Xn ) is absolutely summable. Further letting l → ∞ in (4.1) we have

 (k)  
 xn − (Xn ) < ε for k
K
1
 
(k)
so that xn converges to (Xn ) as k → ∞.

Theorem 75 Let X be a set and let B(X) denote the set of all bounded real-valued functions
on X. Then
δ(f, g) = sup {|f (x) − g(x)| : x ∈ X}
defines a metric on B(X). Further B(X) is complete.

Proof Clearly δ(f, g)
0 and if δ(f, g) = 0 then we have |f (x) − g(x)| = 0 for all x and hence
f = g. The symmetry of δ is evident. Finally if f, g, h ∈ B(X) and x ∈ X then:
|f (x) − h(x)|  |f(x) − g(x)| + |g(x) − h(x)|  δ(f, g) + δ(g, h).
If we take the supremum of the LHS over all x ∈ X then we have the required triangle inequality.
Finally let (fn ) be a Cauchy sequence in B(X). This means
sup {|fn (x) − fm (x)| : x ∈ X} → 0 as m, n → ∞.
For a particular x0 ∈ X, it follows that |fn (x0 ) − fm (x0 )| → 0 as m, n → ∞ and so (fn (x0 ))
is a real Cauchy sequence. By completeness, this converges which we’ll denote the limit f(x0 ).
Let ε > 0. As (fn ) is a Cauchy sequence, then there exists N such that
sup {|fn (x) − fm (x)| : x ∈ X} < ε when m, n
N
and so
δ(fn , f) = sup {|fn (x) − f(x)| : x ∈ X}  ε when n
N
as the modulus is continuous and sup respects weak inequalities. As fn is bounded then so is
f and further fn converges to f.
If X now has a metric d, then we can discuss the bounded continuous real-valued functions
on X and we have the following result.

COMPLETENESS 21
Theorem 76 Let (X, d) be a metric space and let C(X) denote the subset of B(X) consisting
of all continuous real-valued functions. Then C(X) is complete.

Proof Let (fn ) be a Cauchy sequence in C(X); this is then a Cauchy sequence in B(X) and so
converges to some bounded function f by the previous theorem. It remains to show that f is
continuous. Let ε > 0. As (fn ) is Cauchy there exists N such that

sup {|fn (x) − fm (x)| : x ∈ X} < ε/3 when m, n
N

and hence |fN (x) − f(x)| < ε/3 for all x ∈ X. As fN is continuous at x then there exists δ > 0
such that
|fN (x) − fN (y)| < ε/3 when d(x, y) < δ.
Hence when d(x, y) < δ then

|f (x) − f (y)|  |f (x) − fN (x)| + |fN (x) − fN (y)| + |fN (y) − f (y)| < ε/3 + ε/3 + ε/3 = ε

and the result follows.

Remark 77 Note that we could have equally shown that C(X) is a closed subset of B(X)
and used Proposition 72. This approach is tantamount to showing that the uniform limit of
continuous functions is continuous, a result from first year analysis.

Definition 78 A contraction (or a contraction mapping) on a metric space M is a map

f : M → M such that there is some K < 1 with

d(f (x), f (y))  Kd(x, y) for all x, y ∈ M.

Contractions are special cases of Lipschitz maps. A map f : (M, dM ) → (N, dN ) is said to be
Lipschitz if there exist K > 0 such that

dN (f(x), f (y))  KdM (x, y) for all x, y ∈ M.

Proposition 79 Lipschitz maps (and hence contractions) are uniformly continuous.

Proof Let ε > 0 and set δ = ε/K. Then for any x, y and with dM (y, x) < δ we have

dN (f (x), f(y))  KdM (x, y) < Kδ = ε.

Theorem 80 Contraction Mapping Theorem (Banach 1922) Let f : X → X be a con-

traction on a complete non-empty metric space (X, d). Then there is a unique fixed point x ∈ X
such that f (x) = x.

COMPLETENESS 22
Proof Let 0  K < 1 be such that d(f(x), f (x′ ))  Kd(x, x′ ) for x, x′ ∈ X..Let x0 ∈ X and
define the sequence (xn ) by xn+1 = f(xn ) for n
0. We will first show that (xn ) is Cauchy. If
n > m > 0 then
d(xn , xn−1 ) = d(f (xn−1 ), f (xn−2 ))  Kd(xn−1 , xn−2 )  · · ·  K n−1 d(x1 , x0 )
and
d(xn , xm )  d (xn, xn−1 ) + d(xn−1 , xn−2 ) + · · · + d(xm+1 , xm ) [triangle inequality]
n−1 n−2 m−1
 (K +K + ··· + K )d(x1 , x0 )
m n
K −K
= d(x1 , x0 ) → 0 as m, n → ∞.
1−K
Hence, as X is complete, the sequence (xn ) is convergent, and set x = lim xn . For n
0 we
have f (xn ) = xn+1 and if we let n → ∞ then
x = lim xn+1 = lim f (xn ) = f(lim xn ) = f (x)
as f is a contraction and so continuous. Finally if x and x′ are two fixed points of f we have
d(x, x′ ) = d(f (x), f (x′ ))  Kd(x, x′ )
which is a contradiction unless d(x, x′ ) = 0 and x = x′ .
Example 81 Prove that there exists a unique continuous function f : [0, 1] → R such that

1 1
f (x) = x + f (y) sin(xy) dy for all x ∈ [0, 1].
2 0
Proof We know from Theorem 76 that C[0, 1], the space of all real-valued continuous functions
on [0, 1] with the supremum metric
d(f, g) = sup{|f (x) − g(x)| : x ∈ [0, 1]}
is a complete metric space. Consider F : C[0, 1] → C[0, 1] defined by

1 1
(F f )(x) = x + f(y) sin(xy) dy.
2 0
For f, g ∈ C[0, 1] we have
d(F f, F g) = sup{|F f (x) − F g(x)| : x ∈ [0, 1]}
 1 
1 
= sup{  (f (y) − g(y)) sin(xy) dy  : x ∈ [0, 1]}
2 0

1 1
 sup{ |(f (y) − g(y)) sin(xy)| dy : x ∈ [0, 1]}
2 0

1 1
 sup{ |(f (y) − g(y))| dy : x ∈ [0, 1]}
2 0

1 1
= |(f (y) − g(y))| dy
2 0
1 1
 sup{|f (y) − g(y)| : y ∈ [0, 1]} = d(f, g).
2 2
So F is a contraction on the complete space C[0, 1] and hence has a unique fixed point.

COMPLETENESS 23
5. Connectedness
We meet now our first topological invariant, connectedness, which rigorously gives definition to
the idea of a space being "in one piece".

Definition 82 A metric space M is disconnected if there exist A, B, disjoint, nonempty,

open subsets of M such that M = A ∪ B. We say that S is connected if it is not disconnected.

Remark 83 Note connectedness is a property that a space itself has, or hasn’t. "Being con-
nected" has nothing to do with being connected "in" a larger space.

Proposition 84 The following three statements are equivalent definitions for a space M being
connected.
(a) There is no partition of M into nonempty, disjoint open subsets of M.
(b) The only open and closed subsets of M are ∅ and M .
(c) Any continuous function f : M → Z is constant.
Proof (a) ⇐⇒ (c) This is left to Exercise 6(i) on Sheet 2.
¬(a) =⇒ ¬(b) If M = A ∪ B is a partition of M into non-empty open subsets, then A is
open but also closed as A = M \B.
¬(b) =⇒ ¬(a) If A is a non-empty proper open and closed subset of M then M = A∪(M\A)
is a partition of M into non-empty open subsets.

Remark 85 Many authors use the word clopen for "open and closed", but the term will not
be used in these notes.

Example 86 (a) Q is disconnected as

Q = ((−∞, π) ∩ Q) ∪ ((π, ∞) ∩ Q).

The two subsets on the right hand side are open and closed in Q as

(−∞, π) ∩ Q = (−∞, π] ∩ Q and (π, ∞) ∩ Q = [π, ∞) ∩ Q.

(b) X = (0, 1] ∪ (2, 3) is disconnected as (0, 1] and (2, 3) are both open and closed in X. To see
this note
(0, 1] = [0, 1] ∩ X = (0, 2) ∩ X; (2, 3) = (2, 3) ∩ X = [2, 3] ∩ X.

Proposition 87 Let a, b ∈ R with a  b. Then [a, b] is connected.

Proof Let C be an open and closed subset of [a, b] . Without any loss of generality we may
assume a ∈ C; if not we could work with [a, b] \C. Set

W = {x ∈ [a, b] : [a, x] ⊆ C} and c = sup W

which is well-defined as a ∈ W = ∅.

CONNECTEDNESS 24
 (a) Let ε > 0. By the Approximation Property there exists x ∈ W such that c − ε < x  c
and in particular [a, c − ε] ⊆ C. Hence

[a, c − ε] = [a, c) ⊆ C.
ε>0

As C is closed then [a, c] ⊆ C and hence that c ∈ W .

(b) Now suppose that x ∈ W and that x < b. Then [a, x] ⊆ C, which is open, and so there
exists ε > 0 such that (x − 2ε, x + 2ε) ⊆ C. Thus

[a, x] ∪ (x − 2ε, x + 2ε) = [a, x + 2ε) ⊆ C

and hence x + ε ∈ W.
(c) Combining (a) and (b), if c < b we would have c + δ ∈ W for some δ > 0 contradicting
the fact that c = sup W . Hence b = c ∈ W and [a, b] = C.

Corollary 88 (Intermediate Value Theorem) Let f : [a, b] → R be continuous with

f (a) < 0 < f (b).

Then there exists c ∈ (a, b) such that f (c) = 0.

Proof Suppose for a contradiction that f (x) = 0 for all x ∈ (a, b). Then

A = {x ∈ [a, b] : f (x) > 0} = f −1 (0, ∞) and B = {x ∈ [a, b] : f(x) < 0} = f −1 (−∞, 0)

partition [a, b] into disjoint, non-empty open subsets, contradicting its connectedness.

Remark 89 So we see that it is the connectedness of [a, b] that is the driving force behind
the Intermediate Value Theorem. Indeed we see more generally from this that a continuous
real-valued function on any connected space which takes positive and negative values must have
a root. It is just as easy to see that this isn’t generally the case for continuous functions on
disconnected subsets.

Proposition 90 The connected subsets of R are the intervals.

Proof The proof of the previous proposition can be adapted to show that any interval, be it
open, closed, half-open/half-closed, bounded or unbounded, is connected. Conversely, suppose
for a contradiction that C is a connected subset of R, with a < c < b, a, b ∈ C, c ∈
/ C. Then

C = [(−∞, c) ∩ C] ∪ [(c, ∞) ∩ C]

is a partition of C into disjoint, non-empty open subsets of C.

Example 91 Let X = S ∪ T be connected where S and T are closed subsets of X. Show that
if S ∩ T is connected then S and T are also connected.

CONNECTEDNESS 25
Proof Note that S and T are not disjoint or otherwise they would form a partition for the
connected S ∪ T into disjoint closed subsets. Take x ∈ S ∩ T and let C be a non-empty open
and closed subset of S containing x; we will aim to show C = S. Now C ∩ T is nonempty (it
contains x) and is an open and closed subset of S ∩ T ; as S ∩ T is connected then C ∩ T = S ∩T
or equivalently S ∩ T ⊆ C. Note then that S\C and C ∪ T partition S ∪ T into disjoint closed
subsets. As S ∪ T is connected, and as C ∪ T = ∅ then it follows that S\C = ∅ and so C = S.

Proposition 92 If f : M → N is continuous and C is a connected subset of M, then f (C) is

connected.

Proof Suppose that A and B provide a partition of f(C) into non-empty, disjoint sets which
are open in f (C). Then, as f is continuous, the preimages f −1 (A) and f −1 (B) provide a
partition of C into non-empty, disjoint sets which are open in C. As C is connected then one
of these preimages is empty, say f −1 (A) = ∅. As f maps onto f (C) then A = f f −1 (A) = ∅
showing f (C) is connected.

Corollary 93 Connectedness is a topological invariant. i.e. it is preserved under homeomor-

phisms.

Proposition 94 Let M and N be metric spaces. Let M × N be their product space with metric
d((m1 , n1 ), (m2 , n2 )) = dM (m1 , m2 ) + dN (n1 , n2 ).
Then M × N is connected if and only if M and N are connected.

Proof The projection maps

π1 : M × N → M and π2 : M × N → N
are continuous. (Check!) If M ×N is connected then M = π 1 (M ×N ) is connected and likewise
N. Conversely say that M and N are connected and f : M × N → Z is continuous.
For any m ∈ M, {m} × N is connected and so f is constant on {m} × N.
For any n ∈ N, M × {n} is connected and so f is constant on M × {n} .
So for any (m1 , n1 ) and (m2 , n2 ) in M × N we have by constancy on {m1 } × N and on M × {n2 }
f (m1 , n1 ) = f (m1 , n2 ) = f (m2 , n2 ).
Hence f is constant on M × N which we see to be connected.

Definition 95 A set S ⊆ Rn is said to be path-connected if given any a, b ∈ S there exists a

continuous map γ : [0, 1] → S such that γ(0) = a and γ(1) = b.

Example 96 Rn , l1 , l2 , l∞ are all path-connected. In fact they are all convex. That is given
points v, w in any one of these spaces, the line segment connecting v and w is also in that space.
The straight line path connecting v and w is given by
γ(t) = tw + (1 − t)v 0  t  1.

CONNECTEDNESS 26
Proposition 97 A path-connected set is connected.
Proof Let U be a path-connected set and f : U → Z be continuous. If a, b ∈ U then there
exists a continuous map γ : [0, 1] → U connecting a to b. Then f ◦γ : [0, 1] → Z is a continuous,
integer-valued map on the connected set [0, 1] and so constant. In particular,
f(a) = f(γ(0)) = f (γ(1)) = f(b).
As a and b were arbitrary then f is constant on U and hence (by Proposition 84) U is connected.

Proposition 98 An open connected subset of Rn is path-connected.

Proof Let U be an open connected subset of Rn and let x ∈ U . Let X denote the path
component of x, that is all those points of U that can be connected to x by a continuous path.
If u ∈ X then there is a continuous path γ connecting x to u. Further, as U is open, there is
ε > 0 such that B(u,ε) ⊆ U. Clearly B(u, ε) ⊆ X as the path γ can be extended along a radius
of the ball to any point of B(u, ε). In particular, X is open.
We have shown that the path component X is open. Similarly any other path components
are open and hence so is their union. As X is the complement of this open union then X is
also closed. As X is open and closed and nonempty, and as U is connected, then X = U and
so U is path-connected.
Example 99 (Topologist’s Sine Curve)

Consider the set

X =A∪B where A = {(x, sin(1/x)) : x > 0} and B = {(0, y) : y ∈ R} ⊆ R2 .
We shall see that X is connected but not path-connected. Firstly connectedness: consider an
open and closed subset C of X which contains the origin. Then C ∩ B is a non-empty open
and closed subset of B. As B is connected then C ∩ B = B and so B ⊆ C. But as C is open
then the origin is an interior point of C and C also includes some points from A. Again C ∩ A
is a non-empty open and closed subset of A, and as A is connected (being a continuous image
of (0, ∞)) then C ∩ A = A and A ⊆ C. Hence C = A ∪ B = X and X is connected.
However X is not path-connected. Demonstrating this is a little fiddly, but a proof appears
in Sutherland pp.101—102. Unsurprisingly the path components of X are A and B.

CONNECTEDNESS 27
6. Compactness and Sequential Compactness
Definition 100 An open cover U = {Ui : i ∈ I} for a space M is a collection of sets Ui ,
which are open in M, and such that 
M= Ui .
i∈I

A subcover of U is a collection {Ui : i ∈ J} where J ⊆ I such that


M= Ui
i∈J

and we say this subcover is finite if J is finite.

Definition 101 A space M is said to be compact if every open cover of M has a finite
subcover.

Remark 102 Let A ⊆ M and let U = {Ui : i ∈ I} be an open cover of A. When determining
the compactness or not of A, we might question whether it matters whether the Ui are open in
A or open in M . In fact, it does not matter whether the Ui are open in A and

A= Ui
i∈I

or whether the Ui are open in M and


A⊆ Ui .
i∈I

Remark 103 At this point this may seem a very odd definition and will perhaps seem moreso
when we see the compact subsets of Rn are simply those that are closed and bounded. Compact-
ness is certainly important and we shall see that:

• On a compact space a continuous real-valued function is bounded and attains its bounds.
• On a compact space a continuous function is uniformly continuous.
• A space is compact if and only if the Bolzano-Weierstrass Theorem holds — i.e. a sequence
has a convergent subsequence.

So why not simply focus on the closed and bounded subsets; why this extra definition? More
generally, for metric spaces other than Rn , compact subsets and closed-and-bounded subsets
differ and the three bullet points above all remain true of compact subsets of metric spaces, and
*aren’t* generally true of closed-and-bounded subsets.
Further when we generalize to topological spaces — so that we have only a notion of open sets,
no notion of a metric and so no notion of bounded — then the above definition of compactness,
being solely in terms of open subsets, generalizes to topological spaces with most of the theory
of compact sets remaining true.

COMPACTNESS AND SEQUENTIAL COMPACTNESS 28

Example 104 R and (0, 1] are not compact.

Solution {(−r, r) : r > 0} is an open cover of R which has no finite subcover. Any finite
collection would be of the form {(−ri , ri ) : i = 1, . . . , n} whose union would be (−R, R) where
R = max ri .
{(1/k, 1) : k = 1, 2, 3, . . .} is an open cover of (0, 1] with no finite subcover. Any finite
collection would be of the form {(1/ki , 1) : k = 1, . . . , n} which would only cover (1/ max ki , 1].

Example 105 Let M be a set and d the discrete metric on M . Then A ⊆ M is compact if
and only if A is finite. Note that all subsets, though, are closed and bounded.

Proposition 106 The closed interval [a, b] is compact.

Proof Let U be an open cover of [a, b]. Define W to be the set

W = {x ∈ [a, b] : a finite subcover from U for [a, x] exists} and let c = sup(W ).

Note that c is well-defined as a ∈ W = ∅.

(a) c ∈ W as follows: As a is in some open subset in U then c > a. If 0 < δ < c − a then,
by the Approximation Property, there exists w ∈ W with c − δ < w and so c − δ ∈ W . Say
c ∈ U ∈ U. As U is open then (c − 2δ, c + 2δ) ⊆ U for some δ > 0 and so a finite subset of U
covers [a, c − δ] ∪ (c − 2δ, c + 2δ) ⊇ [a, c]. In particular, c ∈ W .
(b) c = b as follows: Say x ∈ W and x < b. There exists V ∈ U such that x ∈ V and δ > 0
such that (x − 2δ, x + 2δ) ⊆ V . So a finite subset of U covers

[a, x] ∪ (x − 2δ, x + 2δ) ⊇ [a, x + δ]

and hence x + δ ∈ W . This certainly means x = sup W = c and so c = b remains the only
possibility. This shows that [a, b] is compact.

Proposition 107 A closed "hypercuboid" [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ] is compact.

Proof This is an adaptation of the previous proof. We will prove this first for closed rectangles
in R2 . Let U = {Ui : i ∈ I} be an open cover of X = [a, b] × [c, d]. Let

W = {x ∈ [a, b] : a finite subcover from U for [a, x] × [c, d] exists} .

(a) We firstly note that a ∈ W as {a} × [c, d] is compact given the previous result.
(b) Now define e = sup W which is well-defined as a ∈ W =  ∅. For each y ∈ [c, d] there is
an open set Uy ∈ U containing (e, y) and so some δ y > 0 such that the square

(e − δ y , e + δ y ) × (y − δ y , y + δ y ) ⊆ Uy .

The intervals (y − δ y , y + δ y ) form an open cover of [c, d] which is compact; hence, there are
finite y1 , . . . , yn such that the (yi − δ yi , yi + δ yi ) cover [c, d] . Let δ = min δ yi > 0. Then

{Uy1 , . . . , Uyn }

COMPACTNESS AND SEQUENTIAL COMPACTNESS 29

is an open cover for (e − δ, e + δ) × [c, d]. Recall that e = sup W and so by the Approximation
Property there exists w ∈ W with e − δ < w  e. This means there is a finite subcover V of U
for [a, w] × [c, d]. Hence
V ∪ {Uy1 , . . . , Uyn }
is a finite subcover of U for [a, e + δ) × [c, d]. In particular this means that e ∈ W .
(c) In fact, part (b) shows more than e ∈ W . It shows that if x ∈ W and x < b, then
x + δ/2 ∈ W for some δ > 0. As e ∈ W we would have a contradiction unless e = b. Hence X
has a finite subcover from U and we see that X is compact.
(d) With an inductive proof based on the above we can see that closed bounded hypercuboids
in Rn are compact.

Proposition 108 Let M be a metric space and A ⊆ Ṁ . If A is compact then A is closed (in
M ).

Proof Let x ∈ M \A. For any y ∈ A there exists εy = 12 d(x, y) > 0 such that

B(y, εy ) ∩ B (x, εy ) = ∅.

Now U = {BA (y, εy ) : y ∈ A} is an open cover of A and so, by compactness, there exist finitely
many y1 , y2 , . . . , yn such that

A = BA (y1 , εy1 ) ∪ · · · ∪ BA (yn , εyn ).

Let ε = min εyi > 0 and then we see

  c
B(x, ε) = B(x, εyi ) ⊆ B(yi , εyi ) ⊆ Ac

and hence Ac is open, i.e.A is closed.

Proposition 109 Let M be a metric space and A ⊆ Ṁ . If A is compact then A is bounded.

Proof Note that for any a ∈ A,

U = {BA (a, n) : n ∈ N}

is an open cover of A. As A is compact then there exist n1 < n2 < · · · < nk such that
k

A⊆ BA (a, ni ) = BA (a, nk ).
i=1

Hence A is bounded.

Proposition 110 A closed subset of a compact space is compact.

COMPACTNESS AND SEQUENTIAL COMPACTNESS 30

Proof Let M be a compact space and A ⊆ M closed. We aim to show that A is compact (in
its own right as a metric space). Let

U = {Ui : i ∈ I}

be an open cover of A. As A is closed then M\A is open. So

{Ui : i ∈ I} ∪ {M \A}

is open cover of M . By the compactness of M, there is a finite subcover for M . Say

M = Ui1 ∪ · · · ∪ Uin ∪ {M \A}.

Then
A = Ui1 ∪ · · · ∪ Uin
and so there is a finite subcover of U for A.

Theorem 111 (Heine-Borel) Let C ⊆ Rn . Then C is compact if and only if C is closed and
bounded.

Proof We have shown generally in metric spaces that compact subsets are closed and bounded.
Conversely let C be a closed and bounded subset of Rn . As C is bounded there exist real ai , bi
with
C ⊆ [a1 , b1 ] × [a2 , b2 ] × · · · × [an , bn ] .
This hypercuboid is compact and C is a closed subset of it, so by the previous proposition C
is compact.

Proposition 112 If f : M → N is continuous and C is a compact subset of M then f (C) is

compact.

Proof Let {Ui : i ∈ I} be an open cover of f(C). Then, as f is continuous, {f −1 (Ui ) : i ∈ I}

is an open cover of C. As C is compact then there is a finite subcover

 −1 
f (Ui1 ), f −1 (Ui2 ), . . . , f −1 (Uin )

but, in that case, {Ui1 , Ui2 , . . . , Uin } is a finite subcover of {Ui : i ∈ I} .

Corollary 113 Compactness is a topological invariant — i.e. it is preserved by homeomor-

phisms.

Corollary 114 A continuous real-valued function f : C → R on a compact subset C of Rn is

bounded and attains its bounds.

Proof By the previous proposition f (C) is a compact subset of R. So by the Heine-Borel

Theorem f(C) is bounded (i.e. f is a bounded function) and f (C) is closed, so f attains its
bounds as the supremum and infimum of a set are either in the set or are limit points thereof.

COMPACTNESS AND SEQUENTIAL COMPACTNESS 31

Definition 115 We say that a map f : M → N between metric spaces is uniformly contin-
uous if

∀ε > 0 ∃δ ε > 0 ∀m ∈ M dM (m, m′ ) < δ =⇒ dN (f (m), f(m′ )) < ε.

Note how this differs from continuity

∀ε > 0 ∀m ∈ M ∃δ ε,m > 0 dM (m, m′ ) < δ =⇒ dN (f (m), f(m′ )) < ε.

Example 116 The function f (x) = (1 − x)−1 is not uniformly continuous on (0, 1). Let ε = 1.
For 0 < a < 1 set b = (2 − a)−1 > a. Note that f (b) − f (a) = 1 = ε and yet

1 (1 − a)2
δ =b−a = −a = → 0 as a → 1.
2−a 2−a
The function f (x) = x2 is uniformly continuous on (0, 1); in fact, it’s Lipschitz as by the Mean
Value Theorem we have

|f (b) − f(a)| = |f ′ (c)| |b − a| = 2c |b − a| < 2 |b − a| .

Theorem 117 Let f : M → N be a continuous map between metric spaces. If M is compact

then f is uniformly continuous.

Proof Let ε > 0. As f is continuous then for every x ∈ M there exists δ x > 0 such that
f (B(x, δ x )) ⊆ B(f (x), ε/2). The collection
 
δx
U = B(x, ) : x ∈ M
2
form an open cover for M and so, by compactness, this has a finite subcover

{B(xi , δ xi /2) : i = 1. . . . , n} .

We set  
δ xi
δ = min > 0.
i 2
Take x ∈ M . Then x ∈ B(xk , δ xk /2) for some k as the sets form a subcover. If dM (x, y) < δ
then we have
δ xk δx δx
dM (xk , y)  dM (xk , x) + dM (x, y) < + δ  k + k = δ xk .
2 2 2
Hence x and y both lie in B(xk , δ xk ) and
ε ε
dN (f(x), f (y))  dN (f(x), f (xk )) + dN (f (xk ), f (y)) < + = ε.
2 2

Definition 118 Let M be a metric space. We say that M is sequentially compact if every
sequence in M has a convergent subsequence.

COMPACTNESS AND SEQUENTIAL COMPACTNESS 32

Example 119 [a, b] is sequentially compact. This is just a restating of the Bolzano-Weierstrass
Theorem.
(0, 1) is not sequentially compact as the sequence 1/n has no convergent subsequence.
Likewise Q is not sequentially compact as the terminating decimal expansions of π do not
have a convergent subsequence.

Theorem 120 Compact spaces are sequentially compact.

Proof Let (xk ) be a sequence in the compact metric space M . Let

Xn = {xk : k
n} and Un = M\Xn .

As Xn is a decreasing sequence ofclosed sets then U n is an increasing sequence of open sets.

Suppose for a contradiction that Xn = ∅ so that Un = M . That is, the Un form an open
cover for M . As M is compact then there is a finite subcover so that there are i1 < i2 < · · · < in
such that
M = Ui1 ∪ Ui2 ∪ · · · ∪ Uin = Uin

which means that Xin = ∅, a contradiction. Hence there exists x ∈ Xn . This means that

B(x, 1/n) ∩ {xk : k
n} = ∅ for each n

so that we can pick a subsequence (xkn ) which converges to x.

Remark 121 The converse — that sequentially compact metric spaces are compact — is also
true. This fact is on the A2 syllabus but its proof is beyond this course.

Lemma 122 If a Cauchy sequence has a convergent subsequence then the sequence is conver-
gent.

Proof Let (xn ) be a Cauchy sequence in a metric space M and suppose that the subsequence
(xnk ) converges to x. Let ε > 0. As (xn ) is Cauchy there exists N such that

d(xn , xm ) < ε/2 for m, n
N.

As xnk converges to x there exists K such that

d (xnk , x) < ε/2 for k
K.

So for n
N and k
K such that nk
N ,

d (xn , x)  d(xn, xnk ) + d(xnk , x) < ε/2 + ε/2 = ε.

Hence (xn ) converges to x.

Corollary 123 Compact spaces are complete.

Proof Let (xn ) be a Cauchy sequence in a compact metric space M . By Theorem 120 (xn ) has
a convergent subsequence converging to a limit x. By the previous lemma, the entire sequence
converges to x.

COMPACTNESS AND SEQUENTIAL COMPACTNESS 33

Example 124 The unit cube in l∞

C = {(xk ) : |xk |  1}

is not compact.

Solution The sequence e1 = (1, 0, . . .) , e2 = (0, 1, 0, . . .) , e3 = (0, 0, 1, 0, . . .) has no convergent

subsequence as ||ei − ej ||∞ = 1 for i = j.

Remark 125 Here then is an example of a closed and bounded metric space which is not
compact.

Remark 126 (Off-syllabus) A metric space is compact if and only if it complete and totally
bounded. A space M is said to be totally bounded if for every ε > 0 there exist finite points
x1 , . . . , xn such that
n

M⊆ B(xi , ε).
i=1

COMPACTNESS AND SEQUENTIAL COMPACTNESS 34

 SYLLABUS

Complex Analysis (22 lectures)

Basic geometry and topology of the complex plane, including the equations of lines and
circles. [1]

Complex differentiation. Holomorphic functions. Cauchy-Riemann equations (including z, z̄

version). Real and imaginary parts of a holomorphic function are harmonic. [2]

Recap on power series and differentiation of power series. Exponential function and loga-
rithm function. Fractional powers – examples of multifunctions. The use of cuts as method
of defining a branch of a multifunction. [3]

Path integration. Cauchy’s Theorem. (Sketch of proof only – students referred to various
texts for proof.) Fundamental Theorem of Calculus in the path integral/holomorphic situation.
[3]

Cauchy’s Integral formulae. Taylor expansion. Liouville’s Theorem. Identity Theorem.

Morera’s Theorem. [3]

Laurent’s expansion. Classification of isolated singularities. Calculation of principal parts,

particularly residues. [2]

Residue Theorem. Evaluation of integrals by the method of residues (straightforward exam-

ples only but to include the use of Jordan’s Lemma and simple poles on contour of integration).
[3]

Extended complex plane, Riemann sphere, stereographic projection. Möbius transforma-

tions acting on the extended complex plane. Möbius transformations take circlines to circlines.
[2]

Conformal mappings. Riemann mapping theorem (no proof): Möbius transformations,

exponential functions, fractional powers; mapping regions (not Christoffel transformations or
Joukowski’s transformation). [3]

Reading

1. H. A. Priestley, Introduction to Complex Analysis (second edition, OUP, 2003).

Further Reading

1. L. Ahlfors, Complex Analysis (McGraw-Hill, 1979).

2. Reinhold Remmert, Theory of Complex Functions (Springer, 1989) (Graduate Texts in

Mathematics 122).

COMPACTNESS AND SEQUENTIAL COMPACTNESS 35

7. Complex Algebra and Geometry
In the first year course Introduction to Complex Numbers the algebra and geometry of the
complex numbers were defined. Further, in the Analysis I course the complex exponential and
trigonometric functions were defined. We introduce some examples here by way of a reminder,
and also move on the prove Apollonius’ Theorem.

Example 127 Express in the form f(z) = az̄ + b reflection in the line x + y = 1.

Remark 128 The isometries of C are the maps f (z) = az +b and f(z) = az̄ +b where |a| = 1.
The former are orientation-preserving and rotations. The latter are orientation-reversing.

Solution Knowing from the remark that the reflection has the form f (z) = az̄ + b we can find
a and b by considering where two points go to. As 1 and i both lie on the line of reflection then
they are both fixed. So

a1 + b = a1̄ + b = 1, −ai + b = ai + b = i.

Substituting b = 1 − a into the second equation we find

1−i
a= = −i,
1+i
and b = 1 + i. Hence f (z) = −iz + 1 + i.

Example 129 Prove the Parallelogram Law which states that:

"A quadrilateral is a parallelogram if and only if the sum of the squares of the diagonals
equals the sum of the squares of the sides."

Solution If a quadrilateral has vertices 0, a, b, c then we have

parallelogram law holds

⇐⇒ |b|2 + |a − c|2 = |a|2 + |c|2 + |a − b|2 + |b − c|2
⇐⇒ bb̄ + aā + cc̄ − ac̄ − āc = aā + cc̄ + aā − ab̄ − āb + bb̄ + bb̄ − bc̄ − b̄c + cc̄
⇐⇒ aā + ac̄ + āc − ab̄ − āb + bb̄ − bc̄ − b̄c + cc̄ = 0
 
⇐⇒ (a + c − b) ā + c̄ − b̄ = 0
⇐⇒ |b − a − c|2 = 0
⇐⇒ b = a + c
⇐⇒ 0abc is a parallelogram.

Example 130 Find the image of each given region Ai ⊆ C under the given map fi : Ai → C.

COMPLEX ALGEBRA AND GEOMETRY 36

 • A1 is the line Re z = 1 and f1 (z) = sin z;
A general point on the line has the form z = 1 + it and we have
f1 (z) = sin (1 + it) = sin 1 cos it + cos 1 sin it = sin 1 cosh t + i cos 1 sinh t.
So the image is given parametrically by
x(t) = sin 1 cosh t, y(t) = cos 1 sinh t.
(Note in particular that sin z is unbounded here!) Eliminating t by means of the identity
cosh2 t − sinh2 t = 1 we see the image is one branch of the hyperbola
x2 y2
− = 1.
sin2 1 cos2 1
• A2 is the region Im z > Re z > 0 and f2 (z) = z 2 ;
A general point in A2 can be written in the form z = reiθ where π/4 < θ < π/2 and r > 0. As
z 2 = r2 ei2θ has double the argument of z then
f2 (A2 ) = {z ∈ C : π/2 < arg z < π} .
• A3 is the unit disc |z| < 1 and f3 (z) = (1 + z)/(1 − z).
The map f3 is a bijection from the set C\ {1} to C\ {−1}, with f3−1 (z) = (z − 1)/(z + 1). To
see this note
1+z w−1
w= ⇐⇒ w − wz = 1 + z ⇐⇒ z(1 + w) = w − 1 ⇐⇒ z = .
1−z 1+w
So  
 z − 1 
z ∈ f3 (A3 ) ⇐⇒ f3−1 (z) ∈ A3 ⇐⇒   < 1 ⇐⇒ |z − 1| < |z + 1|
z + 1
which is the half-plane of points closer to 1 than −1, or equivalently the half-plane Re z > 0.
Example 131 Let ABC be a triangle. Prove the cosine rule
|BC|2 = |AB|2 + |AC|2 − 2 |AB| |AC| cos Â. (7.1)
Solution We can choose complex coordinates in the plane so that A is at the origin and B is
at 1. Let C be at the point z. So in terms of our co-ordinates:
|AB| = 1, |BC| = |z − 1| , |AC| = |z| , Â = arg z.
So
Re z
RHS of (7.1) = |z|2 + 1 − 2 |z| cos arg z = z z̄ + 1 − 2 |z| ×
|z|
(z + z̄)
= z z̄ + 1 − 2 × = z z̄ + 1 − z − z̄
2
= (z − 1)(z̄ − 1) = |z − 1|2 = LHS of (7.1).

COMPLEX ALGEBRA AND GEOMETRY 37

Example 132 Let ω = e2πi/3 . Show that the triangle abc in C (with the vertices taken in
anticlockwise order) is equilateral if and only if
a + ωb + ω 2 c = 0
Solution Firstly note that 1 + ω + ω 2 = 0, as 0 = ω 3 − 1 = (ω − 1) (ω 2 + ω + 1). The triangle
abc is equilateral if and only if c − b is the side b − a rotated through 2π/3 anticlockwise – i.e.
if and only if
c − b = ω (b − a)
⇐⇒ ωa + (−1 − ω) b + c = 0
⇐⇒ ωa + ω 2 b + c = 0
⇐⇒ a + ωb + ω 2 c = 0.

Proposition 133 The equation

Az z̄ + B z̄ + B̄z + C = 0, (7.2)
where A, C ∈ R and B ∈ C represents:-
(i) a line when A = 0;
2
(ii) a circle if A = 0 and |B|
AC with centre −B/A and radius 1
|A|
|B|2 − AC;
or otherwise has no solutions. All circles and lines can be represented in this way.
Proof If A = 0 then we can rewrite (7.2) as
 2
B B̄ C 
 B 
 |B|2 − AC
z z̄ + z̄ + z + = 0, z +  =
A A A A A2
which is the given circle described in (ii) unless |B|2 < AC in which case there are no solutions.
If A = 0 then our equation reads
B z̄ + B̄z + C = 0 ⇐⇒ 2ux + 2vy + C = 0
where B = u + iv and z = x + iy and this is clearly a line. Clearly any line is of this form and
likewise any circle |z − a|2 = r2 can be put into the form of (7.2).
Apollonius of Perga (c. 260-190 BC) wrote a book Conics, which was considered second
only to the Elements in its importance. We meet here a description of circles due to him. All
circles can be represented in this way. In this form a circle is often referred to as a Circle of
Apollonius.
Theorem 134 (Apollonius’ Theorem) Let k be positive, with k = 1, and let α, β be distinct
complex numbers. Then the locus of points satisfying the equation
|z − α| = k |z − β|
is a circle with centre c and radius r where
k2 β − α k |α − β|
c= 2 , r= .
k −1 |k 2 − 1|
Further α and β are inverse points – i.e. α and β are collinear with c and |c − α| |c − β| = r2 .

COMPLEX ALGEBRA AND GEOMETRY 38

Proof Squaring up the equation we have
 
(z − α) (z̄ − ᾱ) = k 2 (z − β) z̄ − β̄

and rearranging this becomes

 2     
k − 1 z z̄ + ᾱ − k 2 β̄ z + α − k 2 β z̄ = αᾱ − k 2 β β̄
  2 2 2

 k β − α  αᾱ − k 2 β β̄ |k 2 β − α|
⇐⇒ z − = +
k2 − 1  k2 − 1 (k 2 − 1)2
  2 2    

 k β − α  αᾱ − k 2 β β̄ (k 2 − 1) + (k 2 β − α) k 2 β̄ − ᾱ
⇐⇒ z − =
k2 − 1  (k 2 − 1)2
  2 2

 k β − α  k 2 αᾱ + k 2 β β̄ − k 2 αβ̄ − k 2 ᾱβ
⇐⇒ z − =
k2 − 1  (k 2 − 1)2
  2 2
 k β − α  k 2 |α − β|2
⇐⇒ z −  = ,
k2 − 1  (k 2 − 1)2

which is a circle with centre c and radius r as given above. Note that c is collinear with α and
β as
k2 1
c=α+ 2 (β − α) = β + 2 (β − α)
k −1 k −1
and from these formulae we see that
k2 2 2
|c − α| |c − β| = 2 |β − α| = r .
2
(k − 1)

COMPLEX ALGEBRA AND GEOMETRY 39

8. Holomorphic Functions
We move on from algebra and geometry now to look at what it means for a complex function
to be differentiable. It will turn out that being differentiable on the complex plane is a stronger
requirement than it is on the real line. Consequently we will be able to prove a much richer
analytical theory than is possible on the real line, whilst still considering a class of functions
general enough to include all the important everyday functions of mathematics.

Notation 135 In line with Priestley’s notation, for a ∈ C and r > 0, we will write

open disc : D(a, r) = {z ∈ C : |z − a| < r} ;

closed disc : D̄(a, r) = {z ∈ C : |z − a|  r} ;
punctured disc : D′ (a, r) = {z ∈ C : 0 < |z − a| < r} .

Definition 136 Let U be an open subset of C and f : U → C. Let z0 ∈ U. Then:

(a) f is said to be differentiable at z0 if
f (z0 + h) − f (z0 )
lim
h→0 h
exists, in which case this limit is denoted f ′ (z0 ). Note that, precisely, this means that
 
′
 f(z0 + h) − f (z0 ) 
∃L ∈ C ∀ε > 0 ∃δ > 0 ∀h ∈ D (0, δ)   − L < ε.
h
Note, in particular, that this is sufficient to guarantee f is continuous at z0 as

f (z0 + h) = f (z0 ) + hf ′ (z0 ) + o(h) → f(z0 ) as h → 0.

(b) f is said to be holomorphic (or analytic) on U if it is differentiable at every point of U.

Remark 137 Note for the limit in (a) to exist, h must be able to approach 0 from any direction;
this is a stronger requirement than the partial derivatives ∂f/∂x and ∂f /∂y existing which would
just require that the limit to exist as h approaches 0 along either axis.

Example 138 (a) z 2 is holomorphic on C

(b) z̄ is not differentiable anywhere in C.
(c) z −1 is holomorphic on C\{0}.

Solution (a) For any z, h ∈ C, h = 0,

(z + h)2 − z 2
= 2z + h → 2z as h → 0.
h
(b) For any z, h ∈ C, h = 0,

z + h − z̄ h̄ 1 as h → 0 along positive real axis
= →
h h −1 as h → 0 along positive imaginary axis

HOLOMORPHIC FUNCTIONS 40
so that the required limit exists nowhere.
(c) For any z, h = 0,
 
1 1 1 −1 −1
− = → 2 as h → 0
h z+h z z (z + h) z

by the Algebra of Limits, provided z = 0.

Remark 139 (Algebra of Differentiation) The addition, product, quotient and chain rules
of real calculus all apply in a similar fashion when doing complex differentiation. This is
entirely down to the fact that the Algebra of Limits for real sequences and functions carries
over to complex sequences and functions.

It is important to note that differentiability, in the sense of Definition 136, of a function

f (x + iy) requires more than just the partial derivatives ∂f /∂x and ∂f /∂y existing. We see
this straight away now in the Cauchy-Riemann equations.

Theorem 140 (Cauchy-Riemann Equations) Let f : U → C be differentiable at z0 ∈ U.

For z ∈ U let x = Re z and y = Im z, and let u = Re f, v = Im f . Then, at z0 ,

ux = vy , uy = −vx .

Proof The proof follows by approaching z0 horizontally and vertically and equating the limits.
Let z0 = x0 + iy0 ∈ C. Then

f(z0 + h) − f (z0 )
f ′ (z0 ) = lim
h→0, h∈R h
[u(x0 + h, y0 ) + iv(x0 + h, y0 )] − [u(x0 , y0 ) + iv(x0 , y0 )]
= lim
h→0
 h  
u(x0 + h, y0 ) − u(x0 , y0 ) v(x0 + h, y0 ) − v(x0 , y0 )
= lim +i
h→0 h h
= ux (x0 , y0 ) + ivx (x0 , y0 ).

Likewise
f (z0 + ih) − f(z0 )
f ′ (z0 ) = lim
h→0, h∈R ih
[u(x0 , y0 + h) + iv(x0 , y0 + h)] − [u(x0 , y0 ) + iv(x0 , y0 )]
= lim
h→0
  ih   
1 u(x0 , y0 + h) − u(x0 , y0 ) v(x0 , y0 + h) − v(x0 , y0 )
= lim +
h→0 i h h
= −iuy (x0 , y0 ) + vy (x0 , y0 ).

Hence, comparing real and imaginary parts, ux = vy and uy = −vx .

Example 141 Show directly that the real and imaginary parts of z 3 satisfy the CREs.

HOLOMORPHIC FUNCTIONS 41
Solution As z 3 = (x + iy)3 = (x3 − 3xy 2 ) + i (3x2 y − y 3 ) then we see
   
ux = x3 − 3xy 2 x = 3x2 − 3y 2 ; vy = 3x2 y − y 3 y = 3x2 − 3y 2 .
   
uy = x3 − 3xy 2 y = −6xy; −vx = y 3 − 3x2 y x = −6xy.

Example 142 Show that Re z and Im z are differentiable nowhere in C.

Solution For Re z we have u = x, v = 0 and so ux = 1 = 0 = vy . For Im z we have u = y, v = 0

and so uy = 1 = 0 = −vx. (Note that we could also have demonstrated these facts directly
from the definition of differentiability.)

Another way of thinking about differentiability is via the Wirtinger derivatives. These are
notationally convenient, but also reinforce the idea that if f is differentiable then it is a function
of z alone and independent of z̄. Any function of x = Re z and y = Im z can instead be written
in terms of z and z̄ via the equations
z + z̄ z − z̄
x= , y= .
2 2i
The Cauchy-Riemann equations are then equivalent to ∂f /∂ z̄ = 0.

Definition 143 The Wirtinger derivatives, for a complex valued function on C, are defined
as    
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= +i ; = −i .
∂ z̄ 2 ∂x ∂y ∂z 2 ∂x ∂y

Remark 144 Note that these derivatives are formally what one expect from the chain rule as
we’d have, for example,
 
∂ ∂x ∂ ∂y ∂ 1 ∂ 1 ∂y ∂ 1 ∂ ∂
= + = + = −i ,
∂z ∂z ∂x ∂z ∂y 2 ∂x 2i ∂z ∂y 2 ∂x ∂y

but the expression ∂x/∂z has no natural meaning except via the Wirtinger derivatives.

Corollary 145 Let f : U → C be a complex-valued function on an open set U ⊆ C. The

Cauchy-Riemann equations are equivalent to
∂f
= 0,
∂ z̄
and if f is differentiable then
df ∂f
= .
dz ∂z

HOLOMORPHIC FUNCTIONS 42
Proof If u = Re f and v = Im f then
 
∂f 1 ∂ ∂ 1
= +i (u(x, y) + iv(x, y)) = [(ux − vy ) + i (uy + vx )] .
∂ z̄ 2 ∂x ∂y 2
Comparing real and imaginary parts, we see ∂f /∂ z̄ = 0 if and only if ux = vy and uy = −vx .
If ∂f /∂ z̄ = 0 then
 
∂f 1 ∂ ∂ 1 df
= −i (u + iv) = [(ux + vy ) + i (−uy + vx)] = ux + ivx = .
∂z 2 ∂x ∂y 2 dz

Example 146 Note for the non-differentiable functions we have met so far
∂ z̄ ∂ (Re z) 1 ∂ (Im z) i
= 1; = ; = .
∂ z̄ ∂ z̄ 2 ∂ z̄ 2
Example 147 Show that the function
 5
z / |z|4 z = 0
f(z) =
0 0
is continuous at 0, satisfies the Cauchy-Riemann equations at z = 0, but is not differentiable
at 0. (So the Cauchy-Riemann equations holding at a point are not sufficient to guarantee
differentiability at that point.)
Solution Continuity at 0 follows from the fact that |f (z)| = |z| for all z. Secondly
f (0 + h) − f (0) h5 / |h|4 h4
= = 4
h h |h|
has no limit as h → 0. To see this we can approach 0 along the ray arg z = ±π/8 and get the
different limits of ±i. Finally for z = 0,
x5 − 10x3 y 2 + 5xy 4 5x4 y − 10x2 y 3 + y 5
u = Re f = ; v = Im f = .
(x2 + y 2 )2 (x2 + y 2 )2
So at (0, 0) we have
u(ε, 0) − u(0, 0) ε−0 u(0, ε) − u(0, 0) 0−0
ux = lim = lim = 1; uy = lim = lim = 0;
ε→0 ε ε→0 ε ε→0 ε ε→0 ε
v(ε, 0) − v(0, 0) 0−0 v(0, ε) − v(0, 0) ε−0
vx = lim = lim = 0; vy = lim = lim = 1.
ε→0 ε ε→0 ε ε→0 ε ε→0 ε
In fact there are more pathological examples — the function

exp(−z −4 ) z = 0
f (z) =
0 z=0
satisfies the Cauchy-Riemann equations everywhere on C but is not even continuous at 0.

But by way of a partial converse to the Cauchy-Riemann equations, we can prove the
following result.

HOLOMORPHIC FUNCTIONS 43
Proposition 148 (Goursat) Given f : U → C, let u = Re f, v = Im f . Suppose that ux , uy , vx , vy
exist, are continuous and satisfy the Cauchy-Riemann Equations at z ∈ U . Then f is differen-
tiable at z.
Proof Let z = x + iy ∈ U and ε > 0 be such that D(z, ε) ⊆ U . Let h = p + iq where |h| < ε.
Then (f (z + h) − f (z))/h equals
 
p u (x + p, y + q) − u (x, y + q) + iv (x + p, y + q) − iv(x, y + q)
h p
 
q u (x, y + q) − u (x, y) + iv (x, y + q) − iv(x, y)
+
h q
p q
= (ux (x + θ1 p, y + q) + ivx (x + θ2 p, y + q)) + (uy (x, y + θ 3 q) + ivy (x, y + θ4 q))
h h
by the Mean-value Theorem and where each θi ∈ (0, 1) . Using the Cauchy-Riemann Equations,
the above can be rewritten as
[pux (x + θ1 p, y + q) + iqux (x, y + θ4 q)] i [pvx (x + θ2 p, y + q) + iqvx (x, y + θ3 q)]
+ .
p + iq p + iq
Finally, using the continuity of ux and vx we see that
 
′ f (z + h) − f(z)
f (z) = lim = ux (x, y) + ivx (x, y)
h→0 h
exists.
Remark 149 There is a stronger result and something closer to a full converse; this is the
Looman-Menchoff Theorem (1923) which states that if f is continuous on an open set
U and satisfies the Cauchy-Riemann equations on all of U, then f is holomorphic. But this
theorem is beyond the scope of this introductory course.
Remark 150 (Local behaviour of a holomorphic map) Consider a holomorphic map f
defined on an open set U with z0 = x0 + iy0 ∈ U . We will consider a nearby point z0 + h where
|h| < ε The existence of f ′ (z0 ) means that
f(z0 + h) = f(z0 ) + hf ′ (z0 ) + o(h).
So, viewing h as a coordinate on D(a, ε), we see that the linear approximation of f on D(z0 , ε)
is multiplication by f ′ (z0 ), which geometrically represents scaling by |f ′ (z0 )| and rotation by
arg f ′ (z0 ), followed by translation by f (z0 ). If we instead use two real coordinates, say h = α+iβ,
and write u = Re f, v = Im f, then we see
   
u (x0 + α, y0 + β) u(x0 , y0 ) + ux (x0 , y0 )α + uy (x0 , y0 )β
≈
v(x0 + α, y0 + β) v(x0 , y0 ) + vx (x0 , y0 )α + vy (x0 , y0 )β
    
u(x0 , y0 ) ux (x0 , y0 ) uy (x0 , y0 ) α
= +
v(x , y ) v (x
x 0 0 , y ) v (x
y 0 0, y ) β
 0 0    
u(x0 , y0 ) ux (x0 , y0 ) uy (x0 , y0 ) α
= + [by CREqs]
v(x0 , y0 ) −uy (x0 , y0 ) ux (x0 , y0 ) β
    
u(x0 , y0 ) cos θ sin θ α
= +λ
v(x0 , y0 ) − sin θ cos θ β

HOLOMORPHIC FUNCTIONS 44
  2
where λ = ux + u2y = |f ′ (z0 )| and θ = arg f ′ (z0 ). Again we see that the matrix
 
cos θ sin θ
λ
− sin θ cos θ

denotes a scaling and rotation. Those who go on to do Multivariable Calculus will recognize
this as the derivative of f at z0 (essentially the Jacobian).

We shall see, in due course, that the derivative of a holomorphic function is itself holomor-
phic; however for now it seems sensible just to assume this so that we can state the following.

Corollary 151 Let f be holomorphic on an open set U . Then u = Re f and v = Im f are

harmonic functions (i.e. they satisfy Laplace’s equation).

Proof As f is holomorphic then we have ux = vy and uy = −vx in the usual way. But, with
our assumption, the real and imaginary parts of

f ′ = ux + ivx

also satisfy the Cauchy-Riemann equations. Hence

(ux )x = (vx )y ; (ux )y = − (vx )x

with all these second-derivatives themselves being continuous. So

uxx = (vx )y = (−uy )y = −uyy =⇒ uxx + uyy = 0;

vxx = − (ux )y = − (vy )y = −vyy =⇒ vxx + vyy = 0.

Example 152 Show directly that the following functions u(x, y) and U(x, y) are harmonic on
C.
u(x, y) = x3 − 3xy 2 ; U (x, y) = sin x cosh y
Find holomorphic functions f and F such that u = Re f and U = Re F.

Solution Firstly

∇2 u = (6x) + (−6x) = 0; ∇2 U = (− sin x cosh y) + (sin x cosh y) = 0.

By inspection we note that

 
Re (x + iy)3 = Re x3 + 3ix2 y − 3xy 2 − iy 3 = x3 − 3xy 2 = u;
Re sin (x + iy) = Re (sin x cosh y + i cos x sinh y) = sin x cosh y = U.

Definition 153 If u, v are harmonic functions on the open set U ⊆ C such that u + iv is
holomorphic on U , then v is said to be a harmonic conjugate of u..

HOLOMORPHIC FUNCTIONS 45
9. Power Series. Complex Exponential.
Recall the following from the first year course.

Definition 154 Let a ∈ C. A power series centred on a is a series of the form

∞

an (z − a)n (9.1)
n=0

where (an ) is a complex sequence and z ∈ C. We here treat (an ) as given and z as a complex
variable so that the series, where it converges, determines a function in z.

Theorem 155 Given a power series ∞ n
n=0 an (z − a) , there exists a unique R in the range
0  R  ∞ called the radius of convergence of the series such that:

(9.1) converges absolutely when |z − a| < R,

(9.1) diverges when |z − a| > R.
∞
Further, within the disc of convergence |z − a| < R, the power series n=0 an (z − a)n
defines a differentiable function f(z) and
∞

f ′ (z) = nan (z − a)n−1 .
n=1

(i.e. term-by-term differentiation is valid within the disc of convergence.)

∞
Proposition 156 Given a power series n=0 an (z − a)n , if
an
L = lim
n→∞ an+1

exists then R = |L| .

Example 157 (a) Using the previous proposition, the radius of convergence of
∞

zn
n=0

is 1 and so the series defines a holomorphic function on D(0, 1). The function it defines is
(1 − z)−1 . The series converges nowhere on the boundary |z| = 1.
(b) By differentiating term by term in D(0, 1) we get
∞
 ∞

−2 n−1
(1 − z) = nz = (n + 1) z n.
n=1 n=0

POWER SERIES. COMPLEX EXPONENTIAL. 46

We can also arrive at this by multiplying series; so on D(0, 1) we can validly write
 ∞  ∞ 
−2 −1 −1
 
(1 − z) = (1 − z) (1 − z) = zk zl
k=0 l=0
∞ 
 ∞
= z k+l
k=0 l=0
∞
 
= zn 1
n=0 k+l=n
k,l
0
∞

= (n + 1) z n .
n=0

(c) We can find a power series for the function (1 − z)−1 centred at any a = 1. For example,
with a = i, we see

(1 − z)−1 = (1 − i − (z − i))−1
  −1
−1 z−i
= (1 − i) 1−
1−i
∞ n
(z − i)
=
n=0
(1 − i)n+1
√
and this has radius of convergence 2.
−1
Example 158 Determine the power series, centred at 0, for (1 + z + z 2 ) .

Solution Note that, for |z| < 1, we have

1 1−z
2
=
1+z+z 1 − z3  
= (1 − z) 1 + z 3 + z 6 + · · ·
= 1 − z + z3 − z4 + z6 − z7 + · · ·

Remark 159 We can start now to appreciate the radius of convergence of a power series as
something naturally linked to the function it defines. In due course we will  prove Taylor’s
Theorem that a holomorphic function on D(a, r) has a convergent power series ∞ n
0 an (z − a) .
A function f (z) which is holomorphic at a has a power series, and the radius of convergence is
R where R is the distance from a to the nearest singularity. If f is holomorphic on D(a, R)
then by Taylor’s Theorem f has a convergent power series on D(a, R); on the other hand if the
power series converged on D(a, S) where S > R then f would be holomorphic at the singularity
as power series define differentiable functions.

POWER SERIES. COMPLEX EXPONENTIAL. 47

Remark 160 Consider the set of power series with disc of convergence D (0, R) where R is
finite. It is still an open problem as to which subsets of the boundary are sets of convergence
for some power series. It was shown by Piranian and Herzog (1949,1953) that any countable
union of closed subsets of the boundary is the set of convergence for some power series.
However there is a theorem of Abel’s which goes a little way to describing behaviour of a
power series on the boundary of convergence.
Theorem 161 (Abel) Assume that we have, for fixed θ,
∞

f (r) = an (reiθ )n converges for − R < r < R.
n=0

If the series also converges at r = R then the limit limr→R− f (r) exists and
∞

lim f (r) = an (Reiθ )n .
r→R−
n=0

Remark 162 This theorem is off-syllabus but we shall make use of it in Exercise Sheet 4,
Questions 7 and 8. For a proof see, for example, Apostol’s Mathematical Analysis.
Further it was shown in Analysis I that
Theorem 163 The following power series define the complex exponential and trigonometric
functions on the entire complex plane.
∞ ∞ ∞
 zn  (−1)n z 2n+1  (−1)n z 2n
exp z = ; sin z = ; cos z = .
n=0
n! n=0
(2n + 1)! n=0
(2n)!

Further these functions have the following properties, that for any z, w ∈ C:
(a) exp′ z = exp z, sin′ z = cos z and cos′ z = − sin z.
(b) exp(z + w) = exp z × exp w.
(c) exp(iz) = cos z + i sin z.
(d)
exp(iz) + exp(−iz) exp(iz) − exp(−iz)
cos z = ; sin z = .
2 2i
(e) sin2 z + cos2 z = 1.
(f) cos z and sin z have period 2π and no smaller period.
(g) exp z has period 2πi and no smaller period.
(h) exp r = er for real values of r (and so we will often write ez for exp z).
Example 164 Show that ez takes all non-zero values.
Solution Let z = x + iy so that
ez = ex+iy = ex eiy .
We see that ex is the modulus of ez and can take any positive value. y is the argument of ez
and so ez can take any argument (for infinitely many y in fact) and hence the image of exp is
C\{0}.

POWER SERIES. COMPLEX EXPONENTIAL. 48

Example 165 Show that the only solutions of ez = 1 are 2nπi where n ∈ Z. Find all the
solutions of sin z = 1.

Solution If ex+iy = 1, where x and y are real, then equating moduli we see that ex = 1 and
x = 0. Then cos y + i sin y = 1 and we see that cos y = 1, sin y = 0 so that y = 2nπ.
We also have
eiz − e−iz
sin z = =1
2i
⇐⇒ e2iz − 2ieiz − 1 = 0
 iz 2
⇐⇒ e −i =0
⇐⇒ eiz = i.

If we set z = x + iy we then have

eix e−y = i ⇐⇒ y = 0 and cos x = 0, sin x = 1

π
⇐⇒ y = 0 and x = + 2nπ for some n ∈ Z.
2
(So sin z = 1 has no further solutions beyond the real ones that we are accustomed to.)

Example 166 Let z = x + iy. Then

sin(x + iy) = sin x cosh y + i cos x sinh y.

Solution Note that

  y   ix  y 
eix − e−ix e + e−y e + e−ix e − e−y
RHS = +i
2i 2 2 2
 ix −ix
 y −y
  ix −ix
 y 
e −e e +e e +e e − e−y
= −
2i 2 2i 2
ix−y −ix+y
2e − 2e
= = LHS.
4i

When it comes to the complex logarithm and other multifunctions (or multivalued functions)
defined in terms of the complex logarithm (such as square root) then some considerable care
needs to be taken. Certainly reckless assumptions that rules previously true of positive numbers
remain so for complex or negative numbers leads quickly to issues.

Example 167 Clearly

−1 1
= .
1 −1
Taking square roots we find  
−1 1
=
1 −1

POWER SERIES. COMPLEX EXPONENTIAL. 49

  √ √
and as a/b = a/ b then √
√
i −1 1 1
= √ =√ = .
1 1 −1 i
This rearranges to givei2 = 1, √
which 2
√ is plainly false as i = −1. The error of course is in
assuming that the rule a/b = a/ b still holds.

Proposition 168 (a) Any z ∈ C\(−∞, 0] can be written as z = reiθ where r > 0, θ ∈ (−π, π)
in a unique fashion.
(b) The function L : C\(−∞, 0] → C given by

L(z) = log r + iθ

satisfies exp(L(z)) = z and is holomorphic with L′ (z) = 1/z.

Proof (a) follows from choosing r = |z| and θ = arg z, which takes a unique principal value in
the given range. For (b) we firstly note that

exp(L(z)) = elog r eiθ = reiθ = z

and that  1
u(x, y) = log x2 + y 2 = log(x2 + y 2 ); v(x, y) = tan−1 (y/x)
2
(at least when x > 0). Then
1
x x x
ux = ; vy = 2 = ;
x + y2
2
1 + xy 2 x2 + y2
y − xy2 −y
uy = ; vx = 2 = .
x2 + y2 1 + xy2 x2
+ y2

So the Cauchy-Riemann equations are satisfied, and ux , uy , vx , vy are clearly continuous, so by

Proposition 148 we have that L is holomorphic. Further
x − iy 1 1
L′ (z) = ux + ivx = 2 2
= = .
x +y x + iy z

(Some care should really be taken with different formulas for argument separately for y > 0
and y < 0 but all the above follows in a near-identical fashion.)

Corollary 169 Let α ∈ C. Then

z α := exp(αL(z)) (9.2)
defines a holomorphic function on C\(−∞, 0] and

d α
(z ) = αz α−1 .
dz

POWER SERIES. COMPLEX EXPONENTIAL. 50

Proof As L and exp are holomorphic then so is z α . Then by the chain rule
d α d
(z ) = exp(αL(z)) = αL′ (z) exp(αL(z))
dz dz
α
= exp(αL(z))
z
= α exp(αL(z)) exp(−L(z))
= α exp((α − 1)L(z))
= αz α−1 .

Remark 170 The function L(z) defined on C\(−∞, 0] is a holomorphic branch (or just
branch) of the complex logarithm. The other holomorphic branches of the logarithm on this
cut plane are L(z) + 2nπi. (See Exercise Sheet 4, Question 5(i)).
We consider now L(z) near the cut (−∞, 0] and do similarly for z 1/2 and z 1/3 as defined in
the above corollary.
Im
3

L@z+D » logr+Πi L@1D = 0

Re
-3 -2 -1 1 2 3

L@z-D » logr-Πi -1

-2

-3

If we were to take points z+ and z− , respectively just above and below the cut (−∞, 0] then we
would have
z+ = reiθ+ where θ+ ≈ π; z− = reiθ− where θ− ≈ −π.
So
L(z+ ) ≈ log r + iπ; L(z− ) ≈ log r − iπ,
hence there is a discontinuity of 2πi across the cut. If we consider similarly z 1/2 as defined in
(9.2) we see √ √ √ √
(z+ )1/2 ≈ reiπ/2 = i r; (z− )1/2 ≈ re−iπ/2 = −i r.
We see this time that there is a sign change as we cross the cut.
The only other holomorphic function on C\(−∞, 0] which satisfies w 2 = z is w = −z 1/2 (in
the sense of (9.2)) and these two functions, z 1/2 and −z 1/2 are the two holomorphic branches

POWER SERIES. COMPLEX EXPONENTIAL. 51

 √
of z. We see that as we cross the cut we move from one branch to the other.
Im Im
3 3

2 2

1 1

Hz+L12 » i r 1 12
=1 Hz+L12 » -i r 112 = -1
Re Re
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3

Hz-L12 » -i r -1
Hz-L12 » i r -1

-2 -2

-3 -3

z 1/2 −z 1/2
Something similar happens with z 1/3 but now there are three holomorphic branches on C\(−∞, 0]
namely z 1/3 , ωz 1/3 and ω 2 z 1/3 where ω = e2πi/3 is a cube root of unity. Again we see that as
we cross the cut we move from one branch to another. Specifically if we cross the cut in a
downward direction then we move from z 1/3 to ωz 1/3 , from ωz 1/3 to ω 2 z 1/3 , from ω 2 z 1/3 to z 1/3 .
Im Im Im
3 3 3

2 2 2

1 1 1
3 3 3
Hz+L13 » - r 113 = 1 Hz+L13 » -Ω r 113 = Ω Hz+L13 » -Ω2 r 113 = Ω2
Re Re Re
-3 -2 -1 1 2 3 -3 -2 -1 1 2 3 -3 -2 -1 1 2 3

13 2 3 13 3 13 3

Hz-L » -Ω -1
r Hz-L »- r -1
Hz-L » -Ω r -1

-2 -2 -2

-3 -3 -3

z 1/3 ωz 1/3 ω 2 z 1/3

Remark 171 If we had wanted to define a holomorphic branch of the logarithm on a different
cut plane, for example on X = C\{negative imaginary axis}, then this could have been done
similarly. Any z ∈ X can be uniquely written as z = reiθ where θ ∈ (−π/2, 3π/2) and we would
then set log z = log r + iθ as before. We could again prove that this function is holomorphic or
we could simply note that this choice of logarithm is simply L(z/i) + iπ/2.
Given any simple (i.e. non-intersecting) curve C which runs from 0 to ∞ there are infinitely
many branches of the logarithm that can be defined on C\C which differ from each other by a
constant which is an integer multiple of 2πi. If the curve C wraps around the origin then such
branches may share values with many or indeed all of the branches L(z) + 2nπi that we defined
for the cut plane C\(−∞, 0].

Example 172 Note that

L(zw) = L(z) + L(w) z, w ∈ C\(−∞, 0]

POWER SERIES. COMPLEX EXPONENTIAL. 52

in general. In fact it might even be the case that L(zw) is not defined (e.g. z = w = i) Or we
may find cases like z = w = e3πi/4 where

L(zw) = −πi/2 = 3πi/2 = L(z) + L(w).

Of course what has happened is that zw has in effect moved into the domain of a different
holomorphic branch.

Example 173 Show that the power series

∞
 zn
(9.3)
1
n

converges to −L(1 − z) in D(0, 1). Hence evaluate the two series

1 1 1 1 1 1 1 1
2
− 4
+ 6
− + ··· , − + − + ··· .
2×2 4×2 6×2 8 × 28 1×2 3×2 3 5×2 5 7 × 27
Solution The power series is easily seen to have radius of convergence 1; denote by f (z) the
function it defines on D(0, 1). Differentiating term by term we see that
∞
 ∞
 d
′
f (z) = z n−1
= z n = (1 − z)−1 = − L (1 − z) .
1 0
dz

A holomorphic function with zero derivative on a path-connected subset is constant. (We

haven’t proven this general result yet but will do in the next chapter.) So f and −L (1 − z)
differ by a constant; as they agree at z = 0 then this constant is 0 and f (z) = −L(1 − z) on
D(0, 1).
If we put z = i/2 into the series we get
∞
 √ 
 in 5 1
n
= −L(1 − i/2) = − log − i tan−1 .
1
n2 2 2

Taking the negative real part of both sides we get

1 1 1 1 1
2
− 4
+ 6
− 8
+ · · · = log 5 − log 2
2×2 4×2 6×2 8×2 2
and taking the imaginary part we get
1 1 1 1 −1 1
− + − + · · · = tan .
1 × 2 3 × 23 5 × 25 7 × 27 2

Remark 174 The series in (9.3) has radius of convergence R = 1. It diverges at z = 1 (as
this is just the harmonic series) but converges for all other z with |z| = 1. See Exercise Sheet
4, Question 7(ii).

POWER SERIES. COMPLEX EXPONENTIAL. 53

 Instead of focussing on a principal set of values for a multivalued function we might instead
seek to treat all values at once. Following notation introduced in Priestley we will write:

Notation 175 Given z ∈ C, z = 0, a ∈ C we define (using the notation of Priestley)

[arg z] = {θ ∈ R : z = |z|eiθ };
[log z] = {w ∈ C : ew = z};
[z a ] = exp(a[log z]).

Proposition 176 If ew0 = z, show that

[log z] = w0 + 2πiZ and that [log z] = log |z| + i[arg z].

Proof Note that

w ∈ [log z] ⇐⇒ ew = z
⇐⇒ ew = ew0
⇐⇒ ew−w0 = 1
⇐⇒ w − w0 ∈ 2πiZ
⇐⇒ w ∈ w0 + 2πiZ.

Also

θ ∈ [arg z] ⇐⇒ eiθ = z/ |z|

⇐⇒ eiθ+log|z| = z
⇐⇒ log |z| + iθ ∈ [log z] .

Proposition 177 For z, w, a, b ∈ C with z, w = 0

(a) [log z] + [log w] = [log(zw)];
(b) [z a ] · [wa ] = [(zw)a ]
(c) in general [z a ] · [z b ] does not equal [z a+b ].

Proof (a) If w0 ∈ [log w] and z0 ∈ [log z] then

ew0 +z0 = ew0 ez0 = wz =⇒ w0 + z0 ∈ [log(zw)] .

So

[log z] + [log w] = (z0 + 2πiZ) + (w0 + 2πiZ)

= z0 + w0 + (2πiZ + 2πiZ)
= z0 + w0 + 2πiZ
= [log(zw)] .

POWER SERIES. COMPLEX EXPONENTIAL. 54

(b)

[z a ] · [wa ] = exp(a[log z]) exp(a[log w])

= exp (a [log z] + a [log w])
= exp (a ([log z] + [log w]))
= exp (a [log(zw)])
= [(zw)a ] .

(c) But if we take z = w = 1 and a = b = 1/2 then we see that

 1/2   1/2   
1 · 1 = {1, −1} · {1, −1} = {1, −1} = {1} = 11 .

Example 178 For z in the cut plane C\(−∞, 1] we will let

θ1 denote the value of arg (z + 1) in the range (−π, π) ,

θ2 denote the value of arg (z − 1) in the range (−π, π) ,

as in the diagram below.

Im

1.0
0.8
z
0.6
0.4

Θ1 0.2
Θ2
Re
-1.0 -0.5 0.5 1.0 1.5 2.0

So we have
(z + 1) (z − 1) = |z + 1| eiθ1 |z − 1| eiθ2
and 
w= |z + 1| |z − 1|ei(θ1 +θ2 )/2
is a holomorphic function on C\(−∞, 1] which satisfies

w 2 = z 2 − 1.

What about the continuity, or otherwise, of w over the cut? Firstly let r be a real number in
the range −1 < r < 1 and let r+ and r− be complex numbers just above and just below r in the
complex plane. Then

for r+ we have θ 1 ≈ 0 and θ2 ≈ π;

for r+ we have θ 1 ≈ 0 and θ2 ≈ −π.

POWER SERIES. COMPLEX EXPONENTIAL. 55

So
√ √
w+ ≈ 1 − r2 ei(0+π)/2 = i 1 − r2 ;
√ √
w− ≈ 1 − r2 ei(0−π)/2 = −i 1 − r2 .

So we see that we have a sign discontinuity across (−1, 1).

However if we take r be a real number in the range r < −1 and let r+ and r− be complex
numbers just above and just below r in the complex plane. Then

for r+ we have θ1 ≈ π and θ 2 ≈ π;

for r+ we have θ1 ≈ −π and θ2 ≈ −π.

So
√ √
w+ ≈ r2 − 1ei(π+π)/2 = − r2 − 1;
√ √
w− ≈ r2 − 1ei(−π−π)/2 = − r2 − 1.

We see that w is actually continuous across (−∞, −1) and we can in fact extend w to a
holomorphic function on all of C\ [−1, 1] . √ √
√ Note the behaviour of w near
√ the points −1 and 1. If z ≈ −1 then w ≈ 2i z √
+ 1√where
z + 1√is a standard branch of z + 1 on√ the cut plane C\[−1, ∞). If z ≈ 1 then w ≈ 2 z−1
where z − 1 is a standard branch of z − 1 on the cut plane C\(−∞, 1].
√
Remark 179 To properly consider the multifunction z 2 − 1 (or any similar multi-valued
function) it helps to consider its Riemann Surface. In this case the Riemann surface is the
set of points  
Σ = (z, ζ) ∈ C2 : ζ 2 = z 2 − 1 .
in R2 . The curve y 2 = x2 − 1 is a hyperbola. Above (1, ∞) and
Firstly consider the situation √
(−∞, −1) sit branches y = ± x2 − 1 and these two branches meet at (±1, 0) . So most of the
curve is in one or other of the sets
 √    √  
C+ = x, x2 − 1 : |x| > 1 ; C− = x, − x2 − 1 : |x| > 1 .

In fact C+ ∪ C− excludes only the branch points (±1, 0) and we also see that as we cross the
branch points we move from C+ to C− (or vice versa).
In the complex case, for z ∈ / [−1, 1] there are two values of ζ, namely ±w. For z = ±1 the
only value of√ζ is 0. The points (z, w) and (z, −w) have already been described as two different
branches of z 2 − 1 but we need to take some care to see how these branches fit together as
subset of Σ. If we set as above

Σ+ = {(z, w) : z ∈
/ [−1, 1]} and Σ− = {(z, −w) : z ∈
/ [−1, 1]} .

Then Σ+ ∪ Σ− is most of Σ missing only those points associated with z ∈ [−1, 1]. We can note,
as with previous branches, that as z crosses the cut [−1, 1] then (z, w) moves continuously to

POWER SERIES. COMPLEX EXPONENTIAL. 56

the other branch Σ− and likewise (z, −w) moves continuously to the other branch Σ+ .
Im Im

S+ S-

-1 A 1 Re
-1 B 1 Re

B A

So Σ+ and Σ− fit together on Σ by gluing the either side of [−1, 1] as shown in the diagram
above. We can then see that topologically Σ is a cylinder in C2 .

However it often makes sense to include a point at infinity to the complex numbers, something
we will meet in more detail later in the course. Likewise it makes sense to introduce some points
at infinity to Σ. As z becomes large then the points (z, w) and (z, −w) move to the different
ends of the cylinder. So it is fairly natural to include points at infinity at either end of the
cylinder to reflect this behaviour. Topologically, with these points included, Σ is a sphere (in
what is called complex projective space).
The theory of Riemann surfaces is a particularly rich one involving much algebra, topology
and complex analysis with the topology of the surface (primarily) being determined by the degree
of the defining polynomial.

POWER SERIES. COMPLEX EXPONENTIAL. 57

 Intermission
In the second half of this course we shall begin to see how very different Complex Analysis is
from last year’s Real Analysis. One immediate difference we have, when working in the complex
plane, is that extra dimension and this really makes a difference when it comes to integration.
Rather than just integrating over an interval in R we can now consider path integrals in C.
It proves (as some of you will have already seen in Multivariable Calculus) natural to focus on
simple closed curves (deformed circles) and our first major result is

• Cauchy’s Theorem: the integral around a simple closed curve, of a function which is
holomorphic in and on that curve, is zero.

This theorem has natural links with Green’s Theorem, but we will be able to build up a much
richer theory because of the complex element of the course.

As we have seen with the Cauchy-Riemann equations, a function f : C → C being differen-

tiable in the complex sense is more demanding than being differentiable R2 → R2 . Holomorphic
functions, it turns out, are not just once differentiable but infinitely so; in fact, more than this,
holomorphic functions are analytic and can be defined locally by a power series. Whilst being
holomorphic then is somewhat more restrictive, it is still inclusive enough that most functions
that we might be interested in are holomorphic or at least approachable within this theory.
But we will also see some rather alien results such as Liouville’s Theorem which states that
bounded holomorphic functions on C are constant.

Because of Cauchy’s Theorem, path integrals around simple closed curves have a robust
topological nature. Deforming a path integral to include a region where the integrand is holo-
morphic does not affect the integral. Consequently the only contributing elements come from
the integrand’s singularities. We shall see that if a function f (z) has an isolated singularity
at a point a ∈ C (the function is locally holomorphic except at a) then Laurent’s Theorem
applies and we can write f (z) as a doubly infinite series involving positive and negative powers
of z − a. In fact, we shall see that most of those powers do not contribute to the integral. In
fact, if the path only includes the singularity a we have
∞

f (z) dz = cn (z − a)n dz = 2πic−1 .
n=−∞

The coefficient c−1 is called the residue of f at a.

Given that path integrals are complex numbers you might wonder what physical meaning
they might possibly have. But their real or imaginary parts are real integrals and we will meet
an array of techniques to calculate such integrals and infinite sums such as
∞ 2π ∞

cos2 x π cos2 x 2π 1 π
2
dx = (1 + e−2 ), dx = √ , = .
0 1+x 4 0 2 + sin x 2+ 3 n=−∞
n2 +a 2 a tanh(πa)

POWER SERIES. COMPLEX EXPONENTIAL. 58

10. Cauchy’s Theorem
Definition 180 A subset U ⊆ C is said to be a domain if U is non-empty, open and connected.
(Note that Priestley refers to such subsets as "regions".)

Definition 181 A path or contour γ is a continuous and piecewise continuously differentiable

function γ : [a, b] → C.
(Here "piecewise continuously differentiable" means that γ is continuously differentiable
except at finitely many points). We shall also use the same notation γ to denote the image of
γ in C and also refer to that image as a path or contour.
A path γ : [a, b] → C is said to be simple if γ is 1—1 with the possible exception that γ(a)
may equal γ(b) and is said to be closed if γ(a) = γ(b).

We will need to assume in this course an important theorem. However intuitively obvious
the result might be, its proof is subtle and well beyond the course, usually being the subject of
algebraic topology courses at fourth year or graduate level.

Theorem 182 (Jordan Curve Theorem, Jordan 1887) Let γ be the image of a simple closed
path in C. The complement C\γ has exactly two connected components. One of these, known
as the interior, is bounded and the other, known as the exterior, is unbounded.

Example 183 Given a ∈ C and r > 0, we shall denote by

γ(a, r)
the circle with centre a and radius r with a positive orientation (i.e. oriented anticlockwise)
parametrized by
z = a + reiθ 0  θ  2π.
We shall denote by γ + (a, r) the upper semicircle of γ(a, r) and shall denote by γ − (a, r) the
lower semicircle of γ(a, r), both positively oriented.
γ(a, r) is simple and closed; its interior is D(a, r).

Example 184 For w1 , w2 ∈ C we shall denote by [w1 , w2 ] the line segment in C from w1 to w2
with parametrization
z = w1 + t(w2 − w1 ) 0  t  1.

Definition 185 Given a path γ : [a, b] → C a reparametrization of γ is a second map

Γ : [c, d] → C such that
Γ=γ◦ψ
where ψ : [c, d] → [a, b] is a bijection with positive continuous derivative. ψ should be viewed as
a change of coordinates taking a point’s Γ-coordinate to its γ-coordinate.

Definition 186 Let γ : [a, b] → C be a path. The length L(γ) of γ is defined to be

b
L(γ) = |γ ′ (t)| dt.
a

CAUCHY’S THEOREM 59
Proposition 187 The length of a path γ is invariant under reparametrization.
Proof Let γ 1 : [a1 , b1 ] → C and γ 2 : [a2 , b2 ] → C be two parametrizations of the same curve γ.
Let α : [a1 , b1 ] → [a2 , b2 ] be the change of coordinate map so that γ 2 (α(t)) = γ 1 (t).
As α(a1 ) = a2 and α(b1 ) = b2 then
b1 b1
′
|γ 1 (t)| dt = |γ ′2 (α(t))| |α′ (t)| dt
a1 a1
b1 b2
= |γ ′2 (α(t))| ′
α (t) dt = |γ ′2 (u)| du.
a1 a2

Definition 188 Let f : U → C be a continuous function defined on a domain U and let

γ : [a, b] → U be a path in U . We define the path integral
b
f (z) dz = f (γ(t)) γ ′ (t) dt.
γ a

Remark 189 Note that we are integrating here a complex function over a real interval [a, b]
but the definition of the such integrals is very much what you would expect. Given continuous
real functions x(t) and y(t) on [a, b] we defined
b b b
(x(t) + iy(t)) dt = x(t) dt + i y(t) dt.
a a a

It is an easy check that integration remains linear: that is, for complex scalars λ, µ and complex
integrands w(t) and z(t) we have
b b b
(λw(t) + µz(t)) dt = λ w(t) dt + µ z(t) dt.
a a a

Remark 190 Given two paths γ 1 : [a1 , b1 ] → C and γ 2 : [a2 , b2 ] → C with γ 1 (b1 ) = γ 2 (a2 ) we
define their join γ 1 ∪ γ 2 : [a1 , b1 + b2 − a2 ] → C by

γ 1 (t) a1  t  b1
γ 1 ∪ γ 2 (t) =
γ 2 (t − b1 + a2 ) b1  t  b1 + b2 − a2
In case it is not clear then γ 1 ∪ γ 2 follows traces out γ 1 and then traces out γ 2 . Unsurprisingly
for any continuous f defined on both paths

f (z) dz = f (z) dz + f (z) dz.
γ 1 ∪γ 2 γ1 γ2

In a similar fashion for a path γ : [a, b] → C we can also define the reverse path −γ : [a, b] → C
by
−γ(t) = γ(b + a − t).
Unsurprisingly
f(z) dz = − f (z) dz.
−γ γ

CAUCHY’S THEOREM 60
 
Proposition 191 The path integral γ
f (z) dz is invariant under reparametrization.

Proof Let γ i : [ai , bi ] → C for i = 1, 2 be two parametrizations of γ and let α : [a1 , b1 ] → [a2 , b2 ]
be the change of coordinate map so that γ 2 (α(t)) = γ 1 (t). Then
b2 b1 b1
′ ′ ′
f (γ 2 (t)) γ 2 (t) dt = f (γ 2 (α(t)) γ 2 (α (t)) α (t) dt = f (γ 1 (t)) γ ′1 (t) dt
a2 a1 a1

by the chain rule.

Example 192 Rewrite the integral

2π
dθ
0 (3 + cos θ)2

as a path integral around γ(0, 1).

Solution If we set z = eiθ then z wraps positively around γ(0, 1) and we also have

eiθ + e−iθ z + z −1
cos θ = = ; dz = ieiθ dθ = iz dθ.
2 2
So
2π
dθ dz/(iz) 4 z dz
= = .
0 (3 + cos θ)2 γ(0,1)
z+z −1 2
(3 + 2 ) i γ(0,1) (z 2 + 6z + 1)2

Example 193 Evaluate the following path integrals:

dz dz
. z 2 dz, .
γ + (0,1) z [a,b] [1,−i] z

Solution Writing z = eiθ where 0  θ  π we see that

π iθ π
dz ie dθ
= =i dθ = πi.
γ + (0,1) z 0 eiθ 0

Parametrizing [a, b] by z = a + t (b − a) with 0  t  1, we see

1
2
z dz = (a + t (b − a))2 d (a + t (b − a))
[a,b] 0
1
= (b − a) (a + t (b − a))2 dt
0
1 1 !
2 2 2
= (b − a) a + 2a (b − a) t dt + (b − a) t dt
0 0
1   b3 a3
= (b − a) a2 + ab + b2 = − .
3 3 3
CAUCHY’S THEOREM 61
An oddly familiar answer!!
Finally we will parametrize [1, −i] by z = 1 − (1 + i)t where 0  t  1 and find
1
dz −(1 + i) dt
=
[1,−i] z 0 1 − (1 + i)t
1
[(1 − t) + it] dt
= −(1 + i)
0 (1 − t)2 + t2

− (1 + i) 1 [(1 − t) + it] dt
= 2
2 0 (t − 1/2) + 1/4
     1 
− (1 + i) π/4 12 − 21 tan θ + i 12 + 12 tan θ 2
sec2
θ dθ
= 2
2 −π/4 (1/4) sec θ

where the last integral is arrived at by substituting t = (1 + tan θ) /2. With some simplifying
the above becomes

− (1 + i) π/4
[(1 + i) + (i − 1) tan θ] dθ
2 −π/4

− (1 + i) π/4
= (1 + i) dθ [as tan θ is odd]
2 −π/4
− (1 + i) π −πi
= × (1 + i) × = .
2 2 2

Proposition 194 For a ∈ C, r > 0 and k ∈ Z,


k 0 k = −1,
(z − a) dz =
γ(a,r) 2πi k = −1.

Proof We set z = a + reiθ where 0  θ  2π and then

2π
k  iθ k  iθ 
(z − a) dz = re ire dθ
γ(a,r) 0
2π
k+1
= ir ei(k+1)θ dθ
0 2π
= irk+1 [cos(k + 1)θ + i sin(k + 1)θ] dθ.
0

This last integral is clearly zero when k = −1. When k = −1 we have

2π
−1 0
(z − a) dz = ir dθ = 2πi.
γ(a,r) 0

CAUCHY’S THEOREM 62
Remark 195 Whilst the above calculation may look rather trifling, it will prove to be central
to the theory of residues and contour integration. In fact, Preistley refers to it as the Funda-
mental Integral (though this is a phrase of her coining and not widely used). It is at least the
reason why you will see the number 2πi appearing in a great many theorems. We shall see in
due course that a function f (z) which is holomorphic on D′ (a, R) can be written in the form
∞

f (z) = cn (z − a)n
n=−∞

for unique cn . This is Laurent’s Theorem. If 0 < r < R and providing convergence allows
it then we can see that
 ∞  ∞  
 
n n
f (z) dz = cn (z − a) dz = cn (z − a) dz = 2πic−1 .
γ(a,r) γ(a,r) n=−∞ n=−∞ γ(a,r)

So the function’s integral depends only on the coefficient c−1 , known as the residue of f at
a. Of course, we will need to carefully demonstrate that the sum and integral can indeed be
interchanged.

Proposition 196 (Fundamental Theorem of Calculus for path integrals) Let f be a

holomorphic function on a domain U, and let γ be a path in U from p to q. Then

f ′ (z) dz = f (q) − f (p).
γ

Proof Let γ : [a, b] → U be a parametrization of γ with γ(a) = p and γ(b) = q. Then

b
′
f (z) dz = f ′ (γ(t))γ ′ (t) dt
γ a
b
d
= f (γ(t)) dt
a dt
= [f(γ(t))]t=b
t=a
= f(γ(b)) − f (γ(a))
= f(q) − f(p)

as required. (The application of the usual Fundamental Theorem of Calculus relies on deriv-
atives to be continuous, but we will see with Corollary 225 that holomorphic functions have
continuous derivatives.)

Corollary 197 If f ′ = 0 on a domain U then f is constant.

Example 198 If we return to the three integrals in Example 193, we see that each of them can
be determined simply using the above version of the Fundamental Theorem.

CAUCHY’S THEOREM 63
 • If we take a branch of logarithm on C\ {negative imaginary axis} defined by

log z = log |z| + i arg z − π/2 < arg z < 3π/2

then this branch of log z is defined and holomorphic on all of γ + (0, 1) and we see

dz
= log (−1) − log 1 = πi − 0 = πi.
γ + (0,1) z

• We can define f (z) = z 3 /3 on all of C (and so in particular on [a, b]) with f ′ (z) = z 2 and
so !b
2 z3 b3 a3
z dz = = − .
[a,b] 3 a 3 3

• If we take a branch of logarithm on C\(−∞, 0] defined by

log z = log |z| + i arg z − π < arg z < π,

then this branch of log z is defined and holomorphic on all of [1, −i] and we see

dz
= log(−i) − log 1 = −πi/2 = 0 = −πi/2.
[1,−i] z

Proposition 199 (Estimation Theorem) Let U be a domain, γ : [a, b] → U a path in U

and f : U → C. Then   b
 
 f(z) dz  
  |f (γ(t))γ ′ (t)| dt.
γ a

In particular if |f(z)|  M on γ then

 
 
 f (z) dz   M L(γ).
 
γ

Proof The proof that  b  b

 
 z(t) dt  |z(t)| dt

a a

for a complex integrand follows in a similar fashion to that for real integrands (being essentially
a continuous version of the triangle inequality). We then have
   b  b
   ′

 f(z) dz  = 
   f(γ(t))γ (t) dt  |f (γ(t))γ ′ (t)| dt.
γ a a

Further if |f (z)|  M on γ then

  b
 
 f (z) dz  
  M |γ ′ (t)| dt = M L(γ).
γ a

CAUCHY’S THEOREM 64
Proposition 200 (Uniform Convergence) Suppose that the functions fn (z) uniformly con-
verge to f(z) on the path γ. Then

lim fn (z) dz = f (z) dz.
n→∞ γ γ

Proof Let ε > 0. There exists N such that

ε
|fn (z) − f(z)| < for n
N and z ∈ γ.
L(γ)
Hence for n
N we have
     
   
 fn (z) dz − f(z) dz  =  (fn (z) − f (z)) dz  < L(γ) × ε = ε
    L(γ)
γ γ γ

and the result follows.

Example 201 Show that

dz
→0 as R → ∞.
γ + (0,R) z2 + 1

Solution On γ + (0, R) we have

 2     
z + 1 = z 2 − (−1)
z 2  − |−1| = R2 − 1

using |z − w|
|z| − |w|. Hence

 
 dz 
   L(γ + (0, R)) × 1 =
πR
→0 as R → ∞.
 + 2  R2 − 1 R2 − 1
γ (0,R) z + 1

We now move to address the major theorem of the course: Cauchy’s Theorem (or perhaps
more properly the Cauchy-Goursat Theorem). This states:

Theorem 202 (Cauchy 1825, Goursat 1900) Suppose that f (z) is holomorphic inside and
on a closed path γ. Then
f(z) dz = 0.
γ

Remark 203 Whilst Cauchy’s Theorem is one of the most important in analysis, it may well
also not be that surprising at this stage in your studies. Those who has met Green’s and Stokes’
Theorems have seen similar results before, and a weaker version of Cauchy’s Theorem does
indeed follow from Green’s Theorem (see Exercise Sheet 5, Question 1). Further, in those
cases when f is the derivative of some function F then Cauchy’s Theorem just follows from
the Fundamental Theorem of Calculus for a closed curve. In fact we shall in due course see
that f is indeed the derivative of such F — of course we will have needed to make much use of
Cauchy’s Theorem to demonstrate the existence of such F .

CAUCHY’S THEOREM 65
 It is beyond the scope of this course to give a full rigorous proof of this result. We shall
first prove this result for a triangle.

Theorem 204 (Cauchy’s Theorem for a triangle) Let f be holomorphic on a domain

which includes a closed triangular region T . Let ∆ denote the boundary of the triangle, positively
oriented. Then
f(z) dz = 0.
∆

Proof STEP ONE: Creating nested triangles and their intersection. If we join the three mid-
points of the sides of ∆ to make four new triangular paths A, B, C, D, and orient them as in
the diagram below,

D
C
B

then we see that

f (z) dz = f (z) dz + f (z) dz + f (z) dz + f (z) dz
∆ A B C D

as the contributions to the RHS from the interior edges cancel out. Of the four summands on
the RHS, one has maximal modulus, call it ∆1 , and then by the triangle inequality
   
   
 f (z) dz   4  f (z) dz .
   
∆ ∆1

We can then apply this repeatedly, producing triangle ∆1 , ∆2 , . . . such that

   
  n
 
 f (z) dz   4  f (z) dz  for any n.
  
∆ ∆n

As the closed triangular regions (boundaries and interiors) are nested compact subsets then
their intersection ∞

∆n
n=1

is non-empty (this is left as a lemma at the end of the theorem), but can contain no more than
one point as the triangles lie in discs whose radii tend to 0. Let ζ denote the single complex
number in the intersection.

CAUCHY’S THEOREM 66
 STEP TWO: Applying the differentiability of f at ζ. As f ′ (ζ) exists, then for any ε > 0
there exists δ > 0 such that
 
 f (z) − f(ζ) ′

 − f (ζ)  < ε whenever |z − ζ| < δ
 z−ζ 

or equivalently that

f(z) = f (ζ) + (z − ζ) f ′ (ζ) + (z − ζ)w where |w| < ε.

If we choose n sufficiently large that ∆n ⊆ D(ζ, δ) then we see

′
f(z) dz = f(ζ) dz + f (ζ) (z − ζ) dz + (z − ζ)w(z) dz = (z − ζ)w(z) dz
∆n ∆n ∆n ∆n ∆n

as the first two integrals are zero by the Fundamental Theorem of Calculus. Note, from the
nature of the triangles’ construction, thatL(∆n ) = 12 L(∆n−1 ) = 21n L(∆) and that for any
z ∈ ∆n we have |z − ζ| < L(∆n ). So, by the Estimation Theorem,
   
    ε


 
f (z) dz  =  (z − ζ)w(z) dz   L(∆n ) × L(∆n ) × ε = n L(∆)2 .
∆n ∆n 4

Finally, then,    
  n
 
 f (z) dz   4 
   f (z) dz   εL(∆)2
∆ ∆n

and as ε was arbitrarily chosen then it follows that

f(z) dz = 0.
∆

Lemma 205 Let M  be a compact metric space and Cn a decreasing sequence of closed non-
empty subsets. Then Cn = ∅.

Proof Let Un = M \C n so that the Un form an increasing sequence of open subsets. Suppose
that the intersection Cn is empty and then, by De Morgan’s Law
  "
Un = (M \Cn ) = M Cn = M,

shows that the Un form an open cover for M . By compactness there are n1 < n2 < · · · < nk
such that
Unk = Un1 ∪ · · · ∪ Unk = M
but this leads to the conclusion that Cnk = ∅, a contradiction.

Recall that a domain U is called convex if for any a, b ∈ U it follows that [a, b] ⊆ U. We
now have:

CAUCHY’S THEOREM 67
Theorem 206 (Antiderivative Theorem) Let f be a function which is holomorphic on a
convex domain U. Then there exists a holomorphic function F on U such that F ′ (z) = f(z).

Proof Fix a ∈ U and for any z ∈ U let [a, z] denote the oriented line segment from a to z.
Define
F (z) = f (w) dw.
[a,z]

Let ε > 0 be such that D(z, ε) ⊆ U and take h with |h| < ε. By Cauchy’s Theorem for triangles

f (w) dw + f (w) dw + f (w) dw = 0
[a,z] [z,z+h] [z+h,a]

which may be rearranged to

F (z + h) − F (z) = f (w) dw.
[z,z+h]

So
      
 F (z + h) − F (z)   1  1 
 − f (z) = f (w) dw − f (z) = [f(w) − f(z)] dw 
 h   h  h 
[z,z+h] [z,z+h]

as [z,z+h] k dw = kh for any constant function of w. Finally, by the Estimation Theorem, we
have
   
 F (z + h) − F (z)  1 
 − f (z)  =  [f(w) − f(z)] dw 
 h  h 
[z,z+h]
1
 × |h| × sup |f (w) − f(z)|
|h| w∈[z,z+h]

= sup |f (w) − f (z)|

w∈[z,z+h]

which tends to zero as h → 0 by the continuity of f at z, showing that F ′ (z) = f (z) as required.

Corollary 207 (Cauchy’s Theorem for a convex region) Let f be holomorphic on a

convex domain U. Then for any closed path γ in U we have

f(z) dz = 0.
γ

Proof This follows from the Antiderivative Theorem and the Fundamental Theorem of Calcu-
lus.

We will now proceed to state, though not prove, two more general versions of Cauchy’s
Theorem. For these we will need to introduce some further topological notions.

CAUCHY’S THEOREM 68
Definition 208 Let γ 1 : [0, 1] → U and γ 2 : [0, 1] → U be two closed paths in a domain U.
We say that γ 1 and γ 2 are homotopic if there is a continuous function H : [0, 1]2 → C such
that for each u ∈ [0, 1], then H(_, u) : [0, 1] → U is a closed path in U and for all t ∈ [0, 1] ,

H(t, 0) = γ 1 (t), H(t, 1) = γ 2 (t).

We might think of the second variable u as a time parameter. The homotopy H can then be
viewed as a continuous deformation from γ 1 (when u = 0) to γ 2 (when u = 1). Whilst the full
rigour of the following result is somewhat difficult, it should not be a surprising result. Indeed
γ 1 ∪ (−γ 2 ) might be viewed as the boundary to a domain U on which f is holomorphic and
thus, by an appropriate version of Cauchy’s Theorem, we might expect

0= f (z) dz = f (z) dz − f (z) dz.
γ 1 ∪(−γ 2 ) γ1 γ2

Theorem 209 (Deformation Theorem) Let f be holomorphic on a domain U and let γ 1

and γ 2 be homotopic closed paths in U . Then

f (z) dz = f (z) dz.
γ1 γ2

Example 210 Any γ(a1 , r1 ) and γ(a2 , r2 ) are homotopic in C.

Solution We may use

H(θ, u) = ua2 + (1 − u) a1 + (ur2 + (1 − u) r1 ) e2πiθ .

Example 211 γ(0, 1) and γ(2, 1) are not homotopic in C\{0}.

Solution Let f (z) = z −1 which is homotopic on C\{0}. Note that

dz dz
= 2πi and = 0,
γ(0,1) z γ(2,1) z

(the first by direct calculation, the second by Cauchy’s Theorem). Hence, by the Deformation
Theorem, γ(0, 1) and γ(2, 1) cannot be homotopic in C\{0}.

Definition 212 We say that a domain U is simply-connected if every closed curve in U is

homotopic to a point (i.e. to a constant path).

Example 213 C, (open and closed) discs, (open and closed) half-planes, line segments and
cut-planes are all simply-connected domains.
Punctured discs, circles and C\{0} are domains which are not simply-connected.

The following are then corollaries to the Deformation Theorem.

CAUCHY’S THEOREM 69
Corollary 214 (Cauchy’s Theorem on a Simply Connected Domain) Let f be holo-
morphic on a simply connected domain U and let γ be a closed path in U. Then

f(z) dz = 0.
γ

Proof γ is homotopic to a constant path Γ and so by the Deformation Theorem

f (z) dz = f(z) dz = 0.
γ Γ

Corollary 215 Let f be holomorphic on a simply-connected domain U . Let a, b ∈ U, and let

γ 1 and γ 2 be two paths from a to b. Then

f (z) dz = f (z) dz.
γ1 γ2

Proof Apply the previous corollary to γ 1 ∪ (−γ 2 ).

Corollary 216 (Antiderivative Theorem for a simply-connected domain) Let f be
holomorphic on a simply-connected domain U . Then there exists F on U such that F ′ = f.
Proof Fix a ∈ U and for z ∈ U let γ be a path from a to z in U . Define

F (z) = f(w) dw.
γ

By the previous corollary F is independent of the choice of path and so a function of z alone.
By an argument identical to that for the Antiderivative Theorem in a convex domain, it follows
that F ′ (z) = f(z).
Corollary 217 (Logarithm for a simply connected domain) Let U be a simply-connected
domain not containing zero. Then there exists a holomorphic function l(z) on U such that
exp(l(z)) = z.
Proof Let a ∈ U and c ∈ [log a] so that exp c = a. For z ∈ U, define

dw
l(z) = c +
γ w

where γ is a path from a to z. By the previous corollary l′ (z) = z −1 . Define

exp(l(z))
G(z) = .
z
Then
zl′ (z) exp(l(z)) − exp(l(z)) exp(l(z)) − exp(l(z))
G′ (z) = = =0
z2 z2
and so G is constant on U by connectedness. At a we see
exp(l(a)) exp c a
G(a) = = = =1
a a a
and hence exp(l(z)) = z as required.

CAUCHY’S THEOREM 70
11. Consequences of Cauchy’s Theorem
Let γ : [0, 1] → C be a simple path in C so that, in particular, γ is a homeomorphism onto
its image. If Γ : [0, 1] → C is also a homeomorphism onto the image of γ then γ ◦ Γ−1 is a
homeomorphism of [0, 1]. This means γ ◦ Γ−1 is strictly monotone and so either increasing or
decreasing.

Definition 218 We say that γ and Γ have the same orientation if γ ◦ Γ−1 is increasing and
have reverse orientations if γ ◦ Γ−1 is decreasing.

If γ and Γ have the same orientation then

f (z) dz = f (z) dz
γ Γ

and if they have reverse orientations then

f (z) dz = − f (z) dz.
γ Γ

Now let γ be a simple closed path in C. By the Jordan Curve Theorem C\γ has precisely
two connected components, one of which, the interior, is bounded. If a is in the interior of
γ, and is such that the circle γ(a, r) is interior to γ also, then γ is homotopic to γ(a, r) or to
−γ(a, r). If γ is homotopic to γ(a, r) then by the Deformation Theorem

dz dz
= = 2πi,
γ z−a γ(a,r) z − a

and otherwise the integral is −2πi.

Definition 219 Let γ be a simple closed curve and a a point in the interior of γ. We say that
γ is positively oriented if
dz
= 2πi.
γ z−a

Remark 220 More generally if γ is a (not necessarily simple) closed curve, and a is a point
not on γ, then
1 dz
2πi γ z − a
is an integer, called the winding number of γ about a.

Theorem 221 (Cauchy’s Integral Formula) Let f be holomorphic on and inside a posi-
tively oriented, simple, closed curve γ and let a be a point inside γ. Then

1 f (w)
dw = f(a).
2πi γ w − a

CONSEQUENCES OF CAUCHY’S THEOREM 71

Proof By the Deformation Theorem we know that

1 f (w) 1 f (w)
dw = dw
2πi γ w − a 2πi γ(a,r) w − a

for any circle γ(a, r) inside γ. From Proposition 194, we know that

dw
= 2πi
γ(a,r) w − a

so that
1 f(a) dw
f (a) = .
2πi γ(a,r) w−a
Hence
    
 1 f(w)   1 f (w) − f (a) 
 dw − f(a) =  dw 
 2πi w−a 2πi γ(a,r) w−a
γ(a,r)
 2π 
1  f (a + reiθ ) − f (a) iθ

=  iθ
ire dθ 
2π 0 re
 2π 
1   iθ
 
= f (a + re ) − f(a) dθ
2π  0
1  
 × 2π sup f (a + reiθ ) − f(a)
2π 0θ2π
 
= sup f (a + reiθ ) − f (a) → 0 as r → 0

0θ2π

by the continuity of f at a. The result follows.

Example 222 Evaluate, using Cauchy’s Integral Formula only,

dz
2
,
γ(0,1) z − 4z + 1

where γ(0, 1) denotes the positively oriented unit circle centred at 0. By means of the substitution
z = eiθ , or otherwise, prove that
2π
dθ 2π
=√ .
0 2 − cos θ 3

Solution Note that z 2 − 4z + 1 = 0 only at α and β where

√ √
α = 2 + 3 and β = 2 − 3.

As |α| > 1 then (z − α)−1 is holomorphic in the disc D(0, 1) and so by the Integral Formula

dz (z − α)−1 dz 2πi −πi
2
= = = √ .
γ(0,1) z − 4z + 1 γ(0,1) z−β β −α 3

CONSEQUENCES OF CAUCHY’S THEOREM 72

If we set z = eiθ in the integral we see
2π
dz ieiθ dθ
2
=
γ(0,1) z − 4z + 1 0 e2iθ − 4eiθ + 1
2π
dθ
= i
0 eiθ + e−iθ − 4
2π
−i dθ
=
2 0 2 − cos θ
and hence 2π
dθ 2 −πi 2π
= × √ =√ .
0 2 − cos θ −i 3 3

Example 223 Use Cauchy’s Integral Formula to determine

Im z exp z
dz and 2
dz.
γ(0,1) 2z − 1 γ(0,1) 4z + 1

Solution The function Im z is not holomorphic so we cannot immediately apply Cauchy’s

Integral Formula. However, for z ∈ γ(0, 1) we have
1 1  
Im z = (z − z̄) = z − z −1 .
2i 2i
Hence

Im z 1 z − z −1
dz = dz
γ(0,1) 2z − 1 2i γ(0,1) 2z − 1

1 z2 − 1
= dz
2i γ(0,1) z (2z − 1)
 
1 1 1 3/4
= + − dz
2i γ(0,1) 2 z z − 1/2
  
2πi 3  π
= 1|0 −  = .
2i 4 1/2 4
For the second integral there are two points z = ±i/2 at which the integrand is not holomorphic.
However, using partial fractions we can argue as follows.

exp z exp z
2
dz = dz
γ(0,1) 4z + 1 γ(0,1) (2z + i)(2z − i)
!
1 exp z exp z
= − dz
4i γ(0,1) z − i/2 z + i/2
   !
2πi i −i
= exp − exp
4i 2 2
π 1 1
= × 2i × sin = πi sin .
2 2 2

CONSEQUENCES OF CAUCHY’S THEOREM 73

Theorem 224 (Taylor’s Theorem) Let a ∈ C, ε > 0 and let f : D(a, ε) → C be a holomor-
phic function. Then there exist unique cn ∈ C such that
∞

f(z) = cn (z − a)n , z ∈ D(a, ε). (11.1)
n=0

Further
f (n) (a) 1 f (w)
cn = = n+1 dw (0 < r < ε) . (11.2)
n! 2πi γ(a,r) (w − a)

Equation (11.2) is often referred to as Cauchy’s Formula for Derivatives. We refer to the
cn as the Taylor coefficients.
Proof Existence: Choose r such that |z − a| < r < ε. By Cauchy’s Integral Formula we have

1 f (w)
f (z) = dw
2πi γ(a,r) w − z

1 f (w)
= dw
2πi γ(a,r) (w − a) − (z − a)

1 f (w)/ (w − a)
=  z−a  dw.
2πi γ(a,r) 1 − w−a
For w ∈ γ(a, r) we have |z − a| < |w − a| = r, and so
  −1  ∞  
z−a z−a n
1− = .
w−a n=0
w−a
Hence # $
∞
1  (z − a)n
f (z) = f (w) dw.
2πi γ(a,r) n=0
(w − a)n+1
As γ(a, r) is compact and f (w) is continuous, then there exists M > 0 such that |f(w)| < M
on γ(a, r). So, for w ∈ γ(a, r),
 n  n  n
 (z − a) |z − a| M |z − a|
f(w) <M = =: Mn .
 (w − a)n+1  rn+1 r r

As Mn converges — being a convergent geometric series — then, by the Weierstrass M—Test,
∞
 (z − a)n
f (w)
n=0
(w − a)n+1
converges uniformly and we can interchange the series and integral to find
#∞ $
1  (z − a)n
f (z) = f (w) dw
2πi γ(a,r) n=0 (w − a)n+1
∞ !
1 f (w) dw n
= n+1 (z − a) .
n=0 %
2πi γ(a,r) (w − a)
&' (
cn

CONSEQUENCES OF CAUCHY’S THEOREM 74

 Uniqueness: To demonstrate uniqueness we can make use of the fact that, within the disc
of convergence, the term-by-term derivative of a power series converges to the derivative of the
function. Hence if (11.1) holds, then
∞
 n!cn
f (k) (z) = (z − a)n−k
n=k
(n − k)!

and in particular f (k) (a) = k!ck .

Corollary 225 Let f be holomorphic on a domain U. Then f (n) exists and is holomorphic for
all n
0.
Proof
∞ Let a ∈ U and ε > 0 such that D(a, ε) ⊆ U . By Taylor’s Theorem we know that f (z) =
n
0 an z is defined by a power series. From first year analysis we know that a power series
defines a differentiable function
∞ on its disc of convergence and that term-by-term differentiation
is valid so that f (z) = 1 nan z n−1 By induction, f has derivatives of all orders.
′

Definition 226 Let f be holomorphic on some domain U and let a ∈ U . Let the Taylor series
of f at a be
∞
f (z) = cn (z − a)n .
n=0
We say that a is a zero of f if f(a) = c0 = 0 and the order of the zero to be N where N is
the least n such that cn = 0.
Example 227 Find the orders of the following zeros.
(a) sin2 z at 0.
(b) (cos z + 1)3 at π.
Solution (a) For sin2 z we could either argue that
sin z = z + O(z) =⇒ sin2 z = z 2 + O(z 3 )
or note that
 
2 d  d2   2 
sin 0 = 0;  sin2 z = 2 sin 0 cos 0 = 0; sin2
z = 2 cos 0 − sin2
0 = 2.
dz 0 dz 2 0
Either way we see that the order is 2.
(b) Let w = z − π. Then
(cos z + 1)3 = (cos(w + π) + 1)3
= (1 − cos w)3
  3
w2 3
= 1− 1− + O(w )
2
 2 3
w 3
= + O(w )
2
w6
= + O(w 7 ).
8
Hence the order of the zero is 6.

CONSEQUENCES OF CAUCHY’S THEOREM 75

Theorem 228 (Liouville’s Theorem) Let f : C → C be a holomorphic function which is
bounded. Then f is constant.

Proof As f is bounded 
there exists M such that |f (z)| < M for all z ∈ C. By Taylor’s
Theorem we have f(z) = cn z n for z ∈ C. So, for n
1 and r > 0 we have
 
 1 f(w) 
|cn | =  dw 

2πi γ(0,r) w n+1
 
1  f(w) 
 × 2πr × sup  n+1 
2π |w|=r w
1 M
 × 2πr × n+1
2π r
= M r−n → 0 as r → ∞.

Hence cn = 0 for n
1 and f(z) = c0 is constant.

Corollary 229 Let f be holomorphic on C and non-constant. Then f (C) is dense in C — that
is f (C) = C.

Proof Take a ∈ C and δ > 0 and suppose for a contradiction that D(a, δ) ⊆ C\f(C). Then
for all z ∈ C
|f(z) − a|
δ
and so
1 1
 .
|f (z) − a| δ
Hence (f (z) − a)−1 is a bounded holomorphic function on C and so, by Liouville, constant.
Hence f (z) is constant, a contradiction.

Remark 230 Liouville’s Theorem is considerably weaker than Picard’s Little Theorem
which is off-syllabus, but which states that a non-constant holomorphic function on C has image
either C or C\ {a single value} .

Theorem 231 (Fundamental Theorem of Algebra) (Gauss 1799) Let p be a non-constant

polynomial with complex coefficients. Then there exists α ∈ C such that p(α) = 0.

Proof Say that p(z) = an z n + · · · + a0 where n
1, ai ∈ C and an = 0. As p(z)/z n → an as

z → ∞ then there exists R > 0 such that
 
 p(z)  |an |
 
 zn  > 2 for |z| > R.

CONSEQUENCES OF CAUCHY’S THEOREM 76

Suppose for a contradiction that p has no roots so that 1/p is holomorphic. Then, by Cauchy’s
Integral Formula, with r > R, we have
   
 1   1 dw 
0 =  =
 

p(0) 2πi γ(0,r) wp(w) 
 
1  1 
 × 2πr × sup  

2π |w|=r wp(w)
1 2
 × 2πr ×
2π |an | rn+1
2
= →0 as r → ∞,
|an | rn
which is the required contradiction.
Theorem 232 (Morera’s Theorem) Let f : U → C be a continuous function on a domain
such that
f (z) dz = 0
γ
for any closed path γ. Then f is holomorphic.
Proof Let z0 ∈ U . As U is open and connected then it is path-connected and so for any z ∈ U
there is a path γ(z) connecting z0 to z. We will then define

F (z) = f (w) dw.
γ(z)

Note that if γ 1 and γ 2 are two such paths then γ 1 ∪ (−γ 2 ) is a closed path and hence by
hypothesis
0= f (w) dw = f (w) dw − f(w) dw
γ 1 γ −1
2 γ1 γ2

and we see that F (z) is well-defined.

Now take z0 ∈ U and r > 0 such that D(z0 , r) ⊆ U and h ∈ D(z0 , r). Then

F (z0 + h) = F (z0 ) + f (w) dw.
[z0 ,z0 +h]

Hence
    
 F (z0 + h) − F (z0 )   1 
 − f (z0 ) =  f(w) dw − f (z0 )
 h h
 [z0 ,z0 +h] 
1 
=  (f (w) − f (z0 )) dw
h [z0 ,z0 +h]
1
 |h| sup |f(w) − f(z0 )| [by the Estimation Theorem]
|h| [z0 ,z0 +h]
= sup |f(w) − f (z0 )| → 0 as h → 0 by the continuity of f at z0 .
[z0 ,z0 +h]

Hence F is holomorphic and F ′ = f . By Corollary 225 f is holomorphic.

CONSEQUENCES OF CAUCHY’S THEOREM 77

Theorem 233 (Identity Theorem) Let f be holomorphic on a domain U. Then the following
are equivalent:
(a) f (z) = 0 for all z ∈ U.
(b) The zero set f −1 (0) has a limit point in U.
(c) There exists a ∈ U such that f (k) (a) = 0 for all k
0.

Proof We shall demonstrate (a) =⇒ (c). =⇒ (b) =⇒ (a)

• (a) =⇒ (c): This implication is obvious.

• (c). =⇒ (b): As U is open then there exists ε > 0 such that D(a, ε) ⊆ U. By Taylor’s
Theorem, for z ∈ D(a, ε) we have
∞
 f (k) (a)
f (z) = (z − a)k = 0.
k=0
k!

Then D(a, ε) ⊆ f −1 (0) and so f −1 (0) has a limit point in U .

• (b) =⇒ (a): Let a be a limit point of f −1 (0) in U. Let ε > 0 be such that D(a, ε) ⊆ U
and let ∞

f (z) = ck (z − a)k
k=0

be the Taylor expansion of f centred about a. Suppose for a contradiction that not all ck
are zero and let K be the smallest k such that cK = 0. Then
∞

K
f (z) = (z − a) g(z) where g(z) = ck (z − a)k−K .
k=K

Note that g(z) is holomorphic on D(a, ε) and g(a) = cK = 0. By continuity, there exists
δ > 0 such that g(z) = 0 on D(a, δ). Thus, in D(a, δ), we see f (z) = 0 only holds at a,
contradicting the fact that a is a limit point of f −1 (0). Hence ck = 0 for all k
0, and so
f (z) = 0 on D(a, ε).
Now let S denote the set of limit points of f −1 (0). By assumption S = ∅. As f is
continuous, then f −1 (0) is closed and so S ⊆ f −1 (0). By the previous part of the argument
if a ∈ S then D(a, ε) ⊆ S for some ε > 0 and hence S is open. Finally if z ∈ U \S then z
is not a limit point of f −1 (0) and so there exists r > 0 such that D(z, r) ⊆ U\S and we
see that U \S is open and S is closed. As S is open and closed, and as U is connected,
then U = S ⊆ f −1 (0).

Corollary 234 Let f and g be holomorphic on a domain U. Then the following are equivalent:
(a) f (z) = g(z) for all z ∈ U.
(b) f (z) = g(z) for all z ∈ S where S ⊆ U has a limit point in U.
(c) There exists a ∈ U such that f (k) (a) = g (k) (a) for all k
0.

CONSEQUENCES OF CAUCHY’S THEOREM 78

Proof Apply the Identity Theorem to f − g.

Remark 235 The Identity Theorem is far from being true on the real line for differentiable,
even infinitely differentiable, functions. The function
  
exp −1x2
x = 0,
f(x) =
0 x = 0,

is a commonly given example to show that an infinitely differentiable function need not have a
locally convergent Taylor series. In this case f (n) (0) = 0 for all n and yet the function is clearly
zero only at zero. This contradicts the Identity Theorem. Another example is
 2  
x sin πx x = 0,
g(x) =
0 x = 0,
 
which is differentiable everywhere. The zero set includes 0, 1, 21 , 13 , . . . which has a limit point
in R (namely 0) but the function isn’t identically zero. The Identity Theorem gives a sense of
how rigid holomorphic functions are compared with even infinitely differentiable real functions.

Example 236 Find all the functions f(z) which are holomorphic on D(0, 1) and which satisfy

f (1/n) = n2 f(1/n)3 n = 2, 3, 4, . . .

Solution We can rewrite this as

       
1 1 1 1 1
f f − f + = 0.
n n n n n

So, at each n, one of the following holds

     
1 1 1 1 −1
f = 0, f = , f = .
n n n n n

At least one of these three equations must hold for infinitely many n, say n1 , n2 , . . . This means
that one of
f (z) = 0, f (z) = z, f (z) = −z
holds on S = {1/n1 , 1/n2 , . . .}. Applying the Identity Theorem we see that f(z) = 0 or z or
−z on all of D(0, 1).

Proposition 237 (Counting Zeros) Let f be holomorphic inside and on a positively oriented
closed path; assume further than f is non-zero on γ. Then
′
1 f (z)
number of zeros of f in γ (counting multiplicities) = dz
2πi γ f (z)

CONSEQUENCES OF CAUCHY’S THEOREM 79

Proof Let the zeros of f be a1 , . . . , ak with multiplicities m1 , . . . , mk . Then f ′ (z)/f (z) is
holomorphic inside γ except at the ai . By Taylor’s Theorem we know that in an open disc
D(ai , ri ) around ai we can write
∞
 ∞

k
f(z) = ck (z − ai ) = (z − ai ) mi
ck (z − ai )k−mi = (z − ai )mi g(z).
k=mi k=mi
% &' (
g(z)

where g(z) is holomorphic and g(ai ) = 0. So in D′ (ai , ri ) we have

f ′ (z) mi (z − ai )mi −1 g(z) + (z − ai )mi g ′ (z) mi g ′ (z)

= = + .
f (z) (z − ai )mi g(z) (z − ai ) g(z)

In particular we see that

f ′ (z) mi g ′ (z)
− =
f(z) (z − ai ) g(z)
is holomorphic in D(a, r). So we similarly see that
k
f ′ (z)  mi
F (z) = −
f (z) i=1
(z − ai )

is holomorphic inside and on γ, having been suitably adjusted at each zero ai . By Cauchy’s
Theorem
′ k   ′ k
f (z) mi f (z)
0 = F (z) dz = dz − dz = dz − 2πi mi
γ γ f (z) i=1 γ (z − ai ) γ f (z) i=1

and the result follows.

CONSEQUENCES OF CAUCHY’S THEOREM 80

12. Laurent’s Theorem
Theorem 238 (Laurent’s Theorem) Let f be holomorphic on the annulus
A = {z ∈ C : R < |z − a| < S} .
Then there exist unique ck (k ∈ Z) such that
∞

f(z) = ck (z − a)k (z ∈ A)
k=−∞

where
1 f (w)
ck = dw (R < r < S) .
2πi γ(a,r) (w − a)k+1
Proof Existence: Let z ∈ A and choose P, Q such that
R < P < |z − a| < Q < S.
Let γ 1 and γ 2 be as given in the diagram below.

Γ1

0
R
P

Q Γ2
S

By Cauchy’s Integral Formula and Cauchy’s Theorem respectively, we know that

1 f(w) 1 f (w)
f (z) = dw, 0= dw.
2πi γ 1 w − z 2πi γ 2 w − z
As the path integrals along the internal line segments cancel then we have

1 f (w) 1 f(w)
f (z) = dw + dw
2πi γ 1 w − z 2πi γ 2 w − z

1 f (w) 1 f (w)
= dw − dw.
2πi γ(a,Q) w − z 2πi γ(a,P ) w − z

LAURENT’S THEOREM 81
For w ∈ γ(a, Q) note |z − a| < |w − a| and for w ∈ γ(a, P ) note |z − a| > |w − a|. Hence

1 f (w)/(w − a) 1 f(w)/(z − a)
f (z) = (z−a)
dw + dw
2πi γ(a,Q) 1 − 2πi γ(a,P ) 1 − (w−a)
(w−a) (z−a)
 ∞ k ∞
1 f(w)(z − a) 1 f (w)(w − a)k
= dw + dw.
2πi γ(a,Q) k=0 (w − a)k+1 2πi γ(a,P ) k=0 (z − a)k+1

Arguing as in Taylor’s Theorem with the Weierstrass M—Test, we may show that these sums
converge uniformly. Hence we may change the order of the integral and summation to obtain
∞
  ∞  
 1 f (w)  1
k
f (z) = dw (z − a) + k
f(w)(w − a) dw (z − a)−k−1
2πi γ(a,Q) (w − a)k+1 2πi γ(a,P )
k=0 k=0
∞
  ∞  
 1 f (w)  1 k
k
= k+1
dw (z − a) + f (w) (w − a) dw (z − a)−k−1
2πi γ(a,r) (w − a) 2πi γ(a,r)
k=0 k=0
∞
 
 1 f (w)
= dw (z − a)k
2πi γ(a,r) (w − a)k+1
k=−∞

as required.

Uniqueness: Suppose now that

∞

f(z) = dk (z − a)k (z ∈ A) .
k=−∞

For R < r < S we have

f (w)
2πicn = n+1 dw
γ(a,r) (w − a)
∞

= dk (w − a)k−n−1 dw
γ(a,r) k=−∞
 ∞
  ∞

 k−n−1
 −k−n−1
= dk (w − a) dw + d−k (w − a) dw .
γ(a,r) k=0 γ(a,r) k=1

As the separate power series converge uniformly on γ(a, r) then we may exchange integration
and summation and find that
∞  
k−n−1
2πicn = dk (w − a) dw = 2πidn .
k=−∞ γ(a,r)

Remark 239 We will almost never use this integral expression for cn in order to determine cn .
Rather we will instead use standard power series for familiar functions to determine Laurent
series.

LAURENT’S THEOREM 82
Remark 240 If f is holomorphic at a then, by uniqueness, the Laurent series of f centred at
a is simply the Taylor series of f centred at a.

Example 241 Find the Laurent series of:

(a) (1 − z)−1 about a = 1.
−1
(b) (1 + z 2 ) exp z about a = i.

Solution (a) We just note

1 −1
=
1−z z−1
so that c−1 = −1 and cn = 0 for n = −1. This is obviously convergent on C\ {1} .
(b) For ease of notation write w = z − i. Then
∞  ∞ 
exp z exp(w + i) ew ei  w −1 ei 1  wk  (−w)l
= = 1+ = .
z2 + 1 w(w + 2i) 2iw 2i 2i w k=0 k! l=0
(2i)l

Hence cn = 0 for n < −1 and for n
−1

ei  (−1)l ei  (i/2)l
cn = = .
2i k+l=n+1 k! (2i)l 2i k+l=n+1 k!

This is convergent on D′ (i, 2).

Example 242 Find cn , where n  3, for f (z) = cot z and a = 0.

Solution Using the standard series for sine and cosine we see, for suitably small z,

cos z 1 − 12 z 2 + 24
1 4
z − ···
cot z = =
sin z z − 16 z 3 + 120
1 5
z − ···
 2 4
 −1
1 z z z2 z4
= 1− + − ··· 1− + − ···
z 2 24 6 120
   2   2 2 
1 z2 z4 z z4 z z4
= 1− + − ··· 1+ − + ··· + − + ··· + ···
z 2 24 6 120 6 120
    
1 z2 z4 z2 1 1 4
= 1− + − ··· 1+ + − z + ···
z 2 24 6 36 120
     
1 1 1 2 1 1 1 1 4
= 1+ − z + − + − z + ···
z 6 2 24 12 36 120
 
1 1 2 1 4
= 1 − z − z + ··· .
z 3 45
So cn = 0 for n  3 except
−1 −1
c−1 = 1, c1 = , c3 = .
3 45

LAURENT’S THEOREM 83
Example 243 Find the Laurent expansion of (z (z − 1))−1 in the following annuli:
 √ 
A1 = {z ∈ C : 0 < |z| < 1} ; A2 = {z ∈ C : 1 < |z − 2| < 2} ; A3 = z ∈ C : |z − i| > 2 .

Solution In A1 = {z ∈ C : 0 < |z| < 1} we have

1 −1 −1   1
= (1 − z)−1 = 1 + z + z2 + · · · = − − 1 − z − z2 − · · · .
z (z − 1) z z z

In A2 = {z ∈ C : 1 < |z − 2| < 2} , writing w = z − 2, we have

1 1 1 1
= = −
z (z − 1) (1 + w) (2 + w) 1+w 2+w
1/w 1/2  
= + [noting w −1  < 1 and |w/2| < 1 in A2 ]
1 + 1/w 1 + w/2
   
1 1 1 1 w w2
= 1 − + 2 − ··· + 1− + − ···
w w w 2 2 4
∞ 

n (−1)n 2−n−1 n
0
= cn w where cn =
(−1)n+1 n<0
−∞

 √ 
In A3 = z ∈ C : |z − i| > 2 , writing w = z − i, we have

1 1 −1 1
= = +
z (z − 1) (w + i) (w − 1 + i) w+i w−1+i
−1/w 1/w  
= + [noting w −1  < |(i − 1) /w| < 1 in A3 ]
1 + i/w 1 + (i − 1) /w
    2 
−1 i i2 i3 1 i−1 i−1
= 1 − + 2 − 3 + ··· + 1+ + + ···
w w w w w w w
∞ 
n 0 n
0
= cn w where cn =
(i − 1)n−1 − (−i)n−1 n < 0
−∞

Definition 244 Let f : U → C be defined on a domain U .

(a) We say that a ∈ U is a regular point if f is holomorphic at a.
(b) We say that a ∈ U is a singularity if f is not holomorphic at a but a is a limit point
of regular points.
(c) We say that a singularity a ∈ U is isolated if f is holomorphic on some D′ (a, r) ⊆ U.

Definition 245 Let a be an isolated singularity of f . By Laurent’s Theorem there exist unique
cn such that
∞
f(z) = cn (z − a)n z ∈ D′ (a, ε).
n=−∞

LAURENT’S THEOREM 84
(i) The principal part of f at a is
−1

cn (z − a)n .
n=−∞

(ii) The residue of f at a is c−1 .

(iii) a is said to be a removable singularity of f if cn = 0 for all n < 0.
(iv) a is said to be a pole of order k if c−k = 0 and cn = 0 for all n < −k. Poles of order
1, 2, 3 are respectively referred to as simple, double and triple poles.
(v) a is said to be an essential singularity if cn = 0 for infinitely many negative n.
(vi) We will simply classify a singularity which is not isolated as non-isolated.

Proposition 246 Let f, g : U → C be holomorphic and a ∈ U. Suppose that f has a zero of

order m at a and g has a zero of order n at a. Then

a pole of order n − m at a if m < n,
f/g has
a removable singularity at a if m
n.

Proof Given the hypotheses, f (z) = (z − a)m F (z) and g(z) = (z − a)n G(z) where F and G
are holomorphic on U and F (a) = 0 = G(a). Hence

f(z) F (z)
= (z − a)m−n (z = a)
g(z) G(z)

where F/G is holomorphic about a and non-zero. The result follows. Note that if m
n then
there is still a singularity at a as the function is undefined, but that if this singularity were
removed then f /g would have a zero of order m − n at a.

Example 247 Locate and classify the singularities of the following functions:
z+1 1 1 1
f(z) = , g(z) = − , h(z) =
z3 + 1 z sin z e1/z + 1

Solution As z 3 + 1 = 0 at z = −1, eπi/3 , e−πi/3 then these are the singularities of f and they
are clearly isolated. Further when z 3 + 1 = 0 we see
1
f (z) = .
(z − eπi/3 )(z − e−πi/3 )

So f has a removable singularity at −1 and simple poles at eπi/3 and e−πi/3 .

Note that
sin z − z
g(z) = .
z sin z
This has singularities at z = nπ where n ∈ Z. At z = 0 we note

z3
sin z − z = − + ··· , z sin z = z 2 + · · · .
3!

LAURENT’S THEOREM 85
So sin z − z has a triple zero and z sin z has a double zero, so that g(z) has a removable
singularity at z = 0. At z = nπ where n = 0 then z sin z has a single zero whilst sin z − z = 0.
Hence g(z) has a simple pole.
We see that h(z) has a singularity when z = 0 and when e1/z = −1, i.e. at zn = (2nπi)−1 . As
zn → 0 as n → ∞, then z = 0 is a non-isolated singularity. At zn , looking at the denominator,
we have 
d   1/z  −1
 e − 1 = 2 e1/zn = − (2nπi)2 (−1) = −4n2 π 2 = 0.
dz z=zn zn
Thus e1/z − 1 has a simple zero at z = zn and hence h(z) has a simple pole at z = zn .
Proposition 248 (Residues at Simple Poles) Suppose that f (z) has a simple pole at a.
Then
res(f (z); a) = lim (z − a)f (z).
z→a
If f(z) = g(z)/h(z), where g and h are holomorphic at a and h has a simple zero at a, then
g(a)
res(f (z); a) = .
h′ (a)
Proof Let R = res(f(z); a). Then there exists a holomorphic function h such that
R
f(z) = + h(z) on some D′ (a, ε).
z−a
So
(z − a) f (z) = R + (z − a) h(z) → R as z → a.
If f (z) = g(z)/h(z) as given then by Taylor’s Theorem,
g(z) = g(a) + O(z − a), h(z) = h′ (a) (z − a) + O((z − a)2 ),
so that
(z − a)g(z)
lim (z − a)f(z) = lim
z→a z→a h(z)
(z − a) [g(a) + O(z − a)]
= lim
z→a h′ (a) (z − a) + O((z − a)2 )

g(a) + O(z − a)
= lim ′
z→a h (a) + O(z − a)

g(a)
= ′ .
h (a)

Proposition 249 (Residues at Overt Multiple Poles) Suppose that

g(z)
f (z) = where g is holomorphic at a and g(a) = 0,
(z − a)n
(so that g has an overt pole of order n at a). Then
g (n−1) (a)
res(f(z); a) = .
(n − 1)!

LAURENT’S THEOREM 86
Proof By Taylor’s Theorem we can write, on some D(a, ε),
∞
g(z) = ck (z − a)k
k=0

where ck = g (k) (a)/k!. So on D′ (a, ε) we have

∞

f (z) = ck (z − a)k−n
k=0

and we see that res(f (z); a) = cn−1 as required.

Example 250 Determine the following residues.
res((z 2 + 1)−1 ; i), res((z 2 − 1)−3 ; 1), res((z − sin z)−1 ; 0), res((z − sin z)−2 ; 0).
Solution The first example is a (relatively overt) simple pole and we see
  
2 −1 1 1  1
res((z + 1) ; i) = res ;i = = .
(z − i)(z + i) z + i z=i 2i
The second example is a (relatively overt) triple pole. We can write
1 (z + 1)−3
=
(z 2 − 1)3 (z − 1)3
and then see

2 −3 1 d2   −3  (−3) (−4) −5 12 3
res((z − 1) ; 1) =  (z + 1) = (1 + 1) = = .
2! dz 2 z=1 2! 2 × 32 16
The third example is a covert triple pole and we have no alternative but to calculate the
principal part of the Laurent series. We have a Taylor expansion at 0 of
 
z3 z5 z3 z5
z − sin z = z − z − + − ··· = − + ··· .
6 120 6 120
So, by the binomial theorem, and for suitably small z, we have
 3 −1
−1 z z5
(z − sin z) = − + ··· .
6 120
 −1
6 z2
= 3 1− + ··· .
z 20
  2   2 2 
6 z z
= 3 1+ + ··· + + ··· + ···
z 20 20
6 3
= + + ···
z 3 10z
so that
3
res((z − sin z)−1 ; 0) =
.
10
The fourth example is a covert pole of order six. At first glance calculating the residue looks
intimidating, but note that the function is even (about 0) and so its Laurent series will only
involve even powers. Hence the residue is zero.

LAURENT’S THEOREM 87
13. Residue Theorem & Applications.
Theorem 251 (Cauchy’s Residue Theorem) Let f be holomorphic inside and on a simple
closed, positively oriented path γ except at points a1 , . . . , an inside γ. Then
n

f (z) dz = 2πi res(f (z); ak ).
γ k=1

Proof On a disc D(ak , εk ) about ak we can expand f(z) as a Laurent expansion

∞
 −1

f(z) = c(k)
n (z − ak ) + n
c(k) n
n (z − ak ) = gk (z) + hk (z),

%n=0 &' ( n=−∞

% &' (
gk (z) hk (z)

so that hk (z) is the principal part of f about ai . Note also that the sum defining hk (z) converges
except at ak . Hence
n
F (z) = f(z) − hk (z)
k=1

is holomorphic in γ except for removable singularities at a1 , . . . , an . As we have seen each

F (ak ) may be assigned a value so as to extend F to a holomorphic function in γ. By Cauchy’s
Theorem and the Deformation Theorem we have
n n

f (z) dz = hk (z) dz = hk (z) dz
γ k=1 γ k=1 γ(ak ,rk )

where 0 < rk < εk . As we saw in the proof of Laurent’s Theorem, the sum defining hk (z)
converges uniformly on γ(ak , rk ) and hence
−1
 −1

(k)
hk (z) dz = c(k)
n (z
n
− ak ) dz = c(k)
n (z − ak )n dz = 2πic−1
γ(ak ,rk ) γ(ak ,rk ) n=−∞ n=−∞ γ(ak ,rk )

and the result follows.

Remark 252 Standard Contours. As we shall see, the Residue Theorem is a powerful
means of calculating certain real integrals. There are three standard contours that we will use
and we shall also introduce certain embellishments here and there when slightly more finesse is
needed. If ever a contour is needed other than the standard ones below then you will typically
be advised what contour to use.

• Integrals between 0 and 2π. We have already seen integrals of this type: as θ moves
from 0 to 2π then z = eiθ moves positively around the unit circle γ(0, 1).

RESIDUE THEOREM & APPLICATIONS. 88

 • Integrals from 0 to ∞ or −∞ to ∞. We will standardly use a semicircular contour
in the upper half-plane made up of γ + (0, R) and the line segment [−R, R], as pictured
below,
Im

Re
-R R

with the aim of showing that as R → ∞ the contribution from γ + (0, R) tends to 0. The
contribution from [−R, R] then tends to the integral on (−∞, ∞) which we are seeking.
If the integrand is even then the integral on (0, ∞) will be half that on (−∞, ∞) .

• Infinite sums. We will use a square contour ΓN with vertices at (N + 1/2) (±1 ± i)
where N is a positive integer and aim to show that the integral around the contour tends
to 0 as N → ∞.
Full details follow in examples below.

13.1 Integrals on R
Example 253 Evaluate ∞
dx
2 4
.
−∞ 1 + x + x

−1
Solution We will consider f(z) = (1 + z 2 + z 4 ) . As
  
1 + z2 + z4 z2 − 1 = z6 − 1

then f (z) has simple poles at eπi/3 , e2πi/3 , e4πi/3 , e5πi/3 . If we use the contour Γ which consists
of the positively-oriented contour consisting of the interval [−R, R] and the upper semicircle
γ + (0, R) then only the simple poles α = eπi/3 and α2 = e2πi/3 lie inside Γ and we have

1  1 1 1 1
res(f (z); α) = 3 
 = 3 = ; res(f (z); α2 ) = 6 2
= .
4z + 2z z=α 4α + 2α −4 + 2α 4α + 2α 4 + 2α2
√
By Cauchy’s Residue Theorem, and noting α2 + α = i 3, α2 − α = −1 we see
 
1 1 1 2πi (2α2 + 2α) π
2 4
dz = 2πi + 2
= 2
=√ . (13.1)
Γ 1+z +z 2α − 4 2α + 4 4(−1 − 2α + 2α − 4 3

INTEGRALS ON R 89
Note that if |z| = R then
         
1 + z 2 + z 4 
z 4  − z 2 + 1
z 4  − z 2  − |1| = R4 − R2 − 1,

so that, by the Estimation Theorem, we have

   
 dz  πR 1
  =O → 0 as R → ∞.
 + 2 4  4 2 R3
γ (0,R) 1 + z + z R −R −1

Finally, letting R → ∞ in (13.1) we have

∞
dx π
2 4
=√ .
−∞ 1+x +x 3

Example 254 Let a
0. Evaluate

∞
cos ax
dx.
0 1 + x2
What is the value of the integral when a < 0?

Remark 255 At first glance this looks no more complicated than the previous example. How-
ever note that
 iz 
e  = e− Im z eiz + e−iz
and cos z = .
2
So on γ + (0, R) we see that e−iz and hence cos z are very large — we would, in due course,
have found that the integral around γ + (0, R) did not converge. So rather than integrating
cos az/ (1 + z 2 ) we will use eiaz / (1 + z 2 ), noting that eiaz is small on γ + (0, R). When done we
can note cos ax = Re eiax for real x.

Solution We will consider the function

eiaz
f (z) = ,
1 + z2
which has simple poles at i and −i, and use the contour Γ which denotes the positively-oriented
contour consisting of the interval [−R, R] and the upper semicircle γ + (0, R). Only the simple
pole i lies inside Γ and we have

eiaz  e−a
res(f (z); i) = = .
z + i z=i 2i

By Cauchy’s Residue Theorem

 −a 
eiaz e
dz = 2πi = πe−a .
Γ 1 + z2 2i

INTEGRALS ON R 90
Note that
 
if z = Reiθ then eiaz  = e−a Re z = e−aR sin θ  1;
     
if |z| = R then z 2 + 1 = z 2 − (−1)
z 2  − |−1| = R2 − 1,

so that, by the Estimation Theorem, we have

 iaz
  
 e  πR 1
 dz   2 =O → 0 as R → ∞.
 + 1+z 2 R −1 R
γ (0,R)

(Note that this part of the argument relies on a being positive so that e−aR sin θ  1.) Now,
letting R → ∞ in the identity
R iaz
e eiaz
2
dz + 2
dz = πe−a
−R 1 + z +
γ (0,R) 1 + z

we arrive at ∞
eiax
dx = πe−a .
−∞ 1 + x2
Comparing real parts we have ∞
cos ax
2
dx = πe−a .
−∞ 1+x
Using the evenness of the integrand we arrive at
∞
cos ax π
2
dx = a .
0 1+x 2e
We clearly couldn’t have argued quite this way if a had been negative. One alternative would
have been to use the lower semicircle γ − (0, R) and repeat a similar calculation, but we can be
smarter than that and note, because of the evenness of cos ax as a function of a that
∞ ∞
cos(−a)x cos ax π
2
dx = 2
dx = a
0 1+x 0 1+x 2e
and so for general a ∈ R we have
∞
cos ax π −|a|
dx = e .
0 1 + x2 2

13.2 Infinite Series

Example 256 Determine
∞ ∞
 1  (−1)n
and .
n=1
1 + n2 n=1
1 + n2

INFINITE SERIES 91
 Before we start this example we will need to introduce some relevant theory. The contour
we shall use is the positively-oriented square contour ΓN with vertices N + 12 (±1 ± i) and
we shall use the integrands
π π
f (z) = 2
, g(z) = 2
.
(1 + z ) tan πz (1 + z ) sin πz
Quite why becomes apparent with the next two lemmas.
Lemma 257 Suppose that the function φ(z) is holomorphic at z = n ∈ Z with φ(n) = 0. Then
(a) πφ(z) cot πz has a simple pole at n with residue φ(n);
(b) πφ(z) csc πz has a simple pole at n with residue (−1)n φ(n).
Proof Note that tan πz and sin πz have simple zeros at z = n and hence πφ(z) cot πz and
πφ(z) csc πz have simple poles there. So
   
πφ(z) πφ(n) πφ(z) πφ(n)
res ;n = = φ(n); res ; n = = (−1)n φ(n).
tan πz π sec2 πn sin πz π cos πn

Lemma 258 There exists a C > 0 such that for all N and for all z ∈ ΓN
 π   π 
   
 C and    C.
tan πz sin πz
Proof For the square’s top and bottom we note
 π   
   cos (πx ± iπ (N + 1/2)) 
  = π  
tan πz sin (πx ± iπ (N + 1/2)) 
 
 exp (iπx ∓ π (N + 1/2)) + exp (−iπx ± π (N + 1/2)) 
= π  
exp (iπx ∓ π (N + 1/2)) − exp (−iπx ± π (N + 1/2)) 
exp (∓π (N + 1/2)) + exp (±π (N + 1/2))
 π
|exp (∓π (N + 1/2)) − exp (±π (N + 1/2))|
= π coth((N + 1/2) π)
 π coth (3π/2) .
On the sides we have (using the periodicity of tan)
 π   π
 
  π 

      = π |tanh πy|  π.
 = =
tan πz tan(±π/2 + πiy)   cot πiy 
Similarly
 π  π
 
  =
sin πz |sin (πx ± iπ (N + 1/2))|
2π
=
|exp (iπx ∓ π (N + 1/2)) − exp (−iπx ± π (N + 1/2))|
2π

|exp (∓π (N + 1/2)) − exp (±π (N + 1/2))|
2π
=
sinh((N + 1/2) π)
2π

sinh(3π/2)

INFINITE SERIES 92
and     
 π   π   π 
   = π |sechπy|  π.
 = =
sin πz sin(±π/2 + πiy)   cos πiy 
Set C to be the maximum of these three numbers.

Solution
  (To Example 256) Consider the positively-oriented square contour ΓN with vertices
1
N + 2 (±1 ± i) and the functions
π π
f (z) = , g(z) = .
(1 + z 2 ) tan πz (1 + z 2 ) sin πz

We know from Lemma 257 that f and g have simple poles at n ∈ Z and that

1 (−1)n
res (f(z); n) = , res (g(z); n) = .
1 + n2 1 + n2
Further there are simple poles at ±i with
π −π
res (f (z); i) = res (f(z); −i) = = ;
2i tan πi 2 tanh π
π −π
res (g(z); i) = res (g(z); −i) = = .
2i sin πi 2 sinh π
Hence, by Cauchy’s Residue Theorem, we have
N
 N
1 1 π 1 π
f(z) dz = − = 2 + 1 − ;
2πi ΓN n=−N
1 + n2 tanh π n=1
1 + n2 tanh π
N N
1  (−1)n π  (−1)n π
g(z) dz = 2
− =2 2
+1− .
2πi ΓN n=−N
1+n sinh π n=1
1+n sinh π

Now by the Estimation Theorem and Lemma 258 there exists C > 0 such that
     
   1  4 (2N + 1) C 1
 
f (z) dz   4 (2N + 1) C sup     =O → 0 as N → ∞.
 z∈ΓN 1 + z
2  2
N
ΓN N + 12 − 1
     
   1  4 (2N + 1) C 1
 g(z) dz   4 (2N + 1) C sup     = O → 0 as N → ∞.
 z∈ΓN 1 + z
2  2
N
ΓN N + 12 − 1

Letting N → ∞ we then have

∞
 ∞
1 π 1 π coth π − 1
2 2
+1− = 0 =⇒ 2
= .
n=1
1+n tanh π n=1
1 + n 2
∞ ∞
 (−1)n π  (−1)n π csch π − 1
2 2
+1− = 0 =⇒ 2
= .
n=1
1+n sinh π n=1
1+n 2

INFINITE SERIES 93
Remark 259 Should we wish to sum
∞ ∞
 1  (−1)n
4
, 2−1
,
n=1
n n=2
n
we would need to note that our chosen φ would have poles that mingle with the poles of cot πz
and csc πz. This is not in itself problematic save that it means such poles need to be treated
separately.

13.3 Refinements
Remark 260 For integrals such as
∞ ∞
x sin x cos x
2 4
dx 6
dx
−∞ 1 + x + x −∞ 1 + x

we would have no great trouble arguing along the lines of Example 254. These integrals converge
relatively quickly and we would see that the contribution from γ + (0, R) was O (R−2 ) and O(R−5 )
respectively. For integrals such as
∞ ∞
sin x x sin x
dx, dx
−∞ x 0 1 + x2
without subtler reasoning we would estimate the contribution from γ + (0, R) as O(1). The prob-
lem is that these integrals do not converge fast enough. (To be technical these integrals in fact
only exist as improper integrals.) We need the following inequality due to Jordan to strengthen
our estimate.
Lemma 261 (Jordan’s Lemma)
2 sin θ
< <1 for 0 < θ < π/2.
π θ
Proof Let f(θ) = sin θ/θ on (0, π/2) . Note that
θ cos θ − sin θ
f ′ (θ) = .
θ2
Now, for 0 < θ < π/2 let
g(θ) = θ cos θ − sin θ,
so that
g ′ (θ) = −θ sin θ + cos θ − cos θ = −θ sin θ < 0,
so that g is decreasing on (0, π/2). As g(0) = 0 then g(θ) < 0 for 0 < θ < π/2. So
g(θ)
f ′ (θ) = <0 for 0 < θ < π/2
θ2
so that f is decreasing on (0, π/2) . As f (θ) → 1 as θ → 0 then
2
< f(θ) < 1 for 0 < θ < π/2
π

REFINEMENTS 94
Example 262 Determine ∞
x sin x
dx.
0 1 + x2

Solution We will consider the function

zeiz
f (z) = ,
1 + z2
which has simple poles at i and −i, and use the contour Γ which denotes the positively-oriented
contour consisting of the interval [−R, R] and the upper semicircle γ + (0, R). Only the simple
pole i lies inside Γ and we have

zeiz  ie−1 1
res(f (z); i) =  = = .
z + i z=i 2i 2e

By Cauchy’s Residue Theorem

zeiz 2πi πi
2
dz = = . (13.2)
Γ 1+z 2e e
Note that
 
if z = Reiθ then eiz  = e− Re z = e−R sin θ ;
     
if |z| = R then z 2 + 1 = z 2 − (−1)
z 2  − |−1| = R2 − 1,

so that, by the Estimation Theorem, we have

   
 ze iz   π Reiθ eiReiθ 
 dz  =  iRe iθ
dθ


 + 1 + z 2   1 + R 2 e2iθ 
γ (0,R) 0
π
Re−R sin θ
 R dθ
0 R2 − 1
π/2
2R2
= e−R sin θ dθ
R2 − 1 0
π/2
2R2
 2
e−2Rθ/π dθ [by Jordan’s Lemma]
R −1 0
 
2R2 −π  −2Rθ/π π/2
= × × e 0
R2 − 1 2R
   
πR2 1 − e−R 1
= 2
=O → 0 as R → ∞.
R(R − 1) R

Letting R → ∞ in (13.2) we obtain

∞
xeix πi
2
dx = .
−∞ 1 + x e

REFINEMENTS 95
Taking imaginary parts we find ∞
x sin x π
2
dx = .
−∞ 1 + x e
Finally using the even-ness of the integrand we obtain
∞
x sin x π
2
dx = .
0 1+x 2e

Remark 263 With some integrals the natural integrand to take may leave us with a singularity
on the contour. This isn’t a problem when that singularity is a simple pole as we may indent
at such singularities (see lemma below). It is impossible to sensibly indent at other
singularities and have any hope of convergence; it is only possible to indent at simple poles
because the contour has length O(ε) and the integrand grows as O(ε−1 ).

Lemma 264 Let f be a holomorphic function on D′ (a, ε) with a simple pole at a. Let 0 < r < ε,
and let γ r be the positively oriented arc

γ r (θ) = a + reiθ α < θ < β.

Then
lim f(z) dz = (β − α) i × res(f (z); a).
r→0 γr

Proof Let R = res(f(z); a). There is a holomorphic function h on D′ (a, ε) such that
R
f(z) = + h(z).
z−a
Now
β β
dz ireiθ dθ
= =i dθ = i (β − α) .
γr z−a α reiθ α

Further, as D̄(a, ε/2) is compact and h is continuous, then h is bounded on D̄(a, ε/2), and so
on every γ r with r < ε/2, say by M. So, by the Estimation Theorem
 
 
 h(z) dz   L(γ r ) × M = M (β − α) r → 0 as r → 0.

γr

Finally then

R dz
lim f (z) dz = lim + lim h(z) dz = iR (β − α) + 0 = iR (β − α)
r→0 γr r→0 γr z − a r→0 γr

as required.

Example 265 Determine ∞

sin2 x
dx.
0 x2

REFINEMENTS 96
Solution Note that
sin2 x 1 − cos 2x
2
=
x 2x2
and so for an integrand we will use (1 − e2iz ) / (2z 2 ). This has a simple pole at 0 as
 
(2iz)2
1−e 2iz 1 − 1 + 2iz + 2!
+ · · · −i
2
= 2
= + 2 + ···
2z 2z z
So we shall indent our usual contour Γ so that we now have a contour made up of line segments
[−R, −ε] and [ε, R], a positively oriented semicircle γ + (0, R) and a negatively oriented indent
γ + (0, ε). As the integrand is holomorphic inside the contour Γ then, by Cauchy’s Theorem, we
have
1 − e2iz
dz = 0.
Γ 2z 2
On the large semicircle  
 1 − e2iz  1 − e−2R sin θ 1
 
 2z 2  = 2R 2

2R2
so that, by the estimation theorem,
 2iz
  
 1 − e  πR 1
 
dz   =O → 0 as R → ∞.
 + 2z 2 2R 2 R
γ (0,R)

We also have from the previous lemma that

 
1 − e2iz 1 − e2iz
− dz = −πi × res ; 0 = (−πi) × (−i) = −π.
γ + (0,ε) 2z 2 2z 2

We have the identity

−ε R
1 − e2iz 1 − e2ix 1 − e2iz 1 − e2ix
dz + dx − dz + dx = 0
γ + (0,R) 2z 2 −R 2x2 γ + (0,ε) 2z 2 ε 2x2

in R and ε. If we let R → ∞, ε → 0 and take real parts then we obtain

R
1 − cos 2x
−π + 2 lim 2
dz = 0.
R→∞ ε 2x
ε→0

Hence ∞
sin2 x π
dx = .
0 x2 2

REFINEMENTS 97
13.4 Further Contours
Example 266 Let −1 < a < 2. Use a keyhole contour to determine
∞
xa
dx.
0 1 + x3
Solution We will consider the function
za
f (z) = .
1 + z3
We need to define a branch of z a on the cut-plane C\[0, ∞). Any z in the cut-plane can be
written z = reiθ where 0 < θ < 2π; we then set
z a = ra eiaθ .
Our keyhole contour Γ comprises two line segments just above and just below [0, R]. Alterna-
tively, we can consider these two line segments both as [0, R] but with one on the lower branch
with θ = 0 and one of the upper branch with θ = 2π. (Cauchy’s Residue Theory is valid on
Riemann surfaces so this viewpoint is perfectly valid.) We also have as part of our keyhole
contour a small cut circle around the origin but as the integrand is O(εa ) and the contour has
length O(ε) then the contribution from this is zero in the limit.
Im

ãΠi3

Re
-1

ã-Πi3

Note that f is holomorphic inside Γ except at −1, eπi/3 and e−πi/3 . Note also that we should
write e−πi/3 = e5πi/3 for the purpose of the branch’s definition. So

z a  1
res(f ; −1) = 2  = eπia ;
3z z=−1 3

πi/3 z a  1
res(f ; e ) = 2  = eπi(a−2)/3 ;
3z z=eπi/3 3

−πi/3 z a  1
res(f; e ) = 2  = e5πi(a−2)/3 .
3z z=e5πi/3 3

FURTHER CONTOURS 98
By the Residue Theorem we have
a
z dz 2πi  πia 
3
= e + eπi(a−2)/3 + e5πi(a−2)/3 .
Γ 1+z 3
The contribution from the outside circle γ(0, R) satisfies
 
 z a dz  2πR1+a

 3
 3
= O(Ra−2 ) → 0 as R → ∞.
γ(0,R) 1 + z R

Hence we have, letting R → ∞, that

∞ a ∞ a
x dx 2πia x dx 2πi  πia 
3
−e 3
= e + eπi(a−2)/3 + e5πi(a−2)/3 .
0 1+x 0 1+x 3
Rearranging we have
∞  πia 
xa dx 2πi e + eπi(a−2)/3 + e5πi(a−2)/3
= ×
0 1 + x3 3 1 − e2πia
 
2πi 1 + eπi(−2a−2)/3 + eπi(2a+2)/3
= ×
3 e−πia − eπia
 πi(2(a+1))/3 
2πi e + 1 + eπi(−2(a+1))/3
= ×
3 eπi(a+1) − e−πi(a+1)
 πi(2(a+1))/3 
2πi e + 1 + eπi(−2(a+1))/3
= ×
3 eπi(a+1) − e−πi(a+1)
 
2πi 1
= ×
3 eπi(a+1)/3 − e−πi(a+1)/3
 
π (a + 1) π
= csc .
3 3

The cancellation to the penultimate line uses the identity (x2 + xy + y 2 ) / (x3 − y 3 ) = 1/ (x − y) .

Remark 267 The above integral could also have been done by using the contour in the next
example which goes around a sector.

Example 268 Evaluate ∞

dx
.
0 1 + x3 + x6
Solution The denominator has roots at

e2πi/9 , e4πi/9 , e8πi/9 , e10πi/9 , e14πi/9 , e16πi/9

as 1 + x3 + x6 = (x9 − 1) / (x3 − 1) . Let R > 1 and let α = e2πi/9 . We will consider the
positively oriented contour Γ made up of the line segment from 0 to R, the circular arc centred
−1
at 0 connecting R to Rα3 and the line segment from Rα3 to 0. Let f (z) = (1 + z 3 + z 6 ) .

FURTHER CONTOURS 99
Then f(z) has six simple poles at the above roots with only α and α2 being inside Γ. By
Cauchy’s Residue Theorem
   !
dz 1 1 2
3 6
= 2πi res ; α + res ;α
Γ 1+z +z 1 + z3 + z6 1 + z3 + z6
!
2πi 1 1
= +
3 α2 + 2α5 α4 + 2α
!
2πi (2α2 + 2α−2 + α + α−1 )
= .
3 (1 + 2α3 ) (α3 + 2)

If we write β = 2π/9 then this simplifies as

! !
2πi 4 cos(2β) + 2 cos β 2πi 4 cos(2β) + 2 cos β 4πi
3 −3 3
= 3 2 = 3 [2 cos(2β) + cos β] .
3α (α + 2) (α + 2) 3α |2 + α3 | 9α

The contribution to this integral from the circular arc satisfies

   
 dz  2πR  1  2πR  −5 
  sup   = O R → 0 as R → ∞.
 3 6 3 arc  1 + z 3 + z 6  3 (R6 − R3 − 1)
arc 1 + z + z

Hence, letting R → ∞, we have

∞ 0
dx α3 dx 4πi
+ = [2 cos(2β) + cos β] ,
0 1 + x3 + x6 3
∞ 1+x +x
6 9α3
giving
∞
dx 4πi [2 cos(2β) + cos β]
=
0 1 + x3 + x6 9 α3 − α6
4π
= √ [2 cos(2β) + cos β]
9 3

Example 269 Show that

∞√ ∞ √
3
x log x π2 3
x π
dx = and dx = √ .
0 1 + x2 6 0 1+x 2
3
Remark 270 Note that it’s not necessary to use a keyhole contour here as along the negative
real axis our integrand at x = −t < 0 becomes
√ √
3
−t log(−t) eπi/3 3 t (log t + πi)
=
1 + (−t)2 1 + t2

which contributes, in its real and imaginary parts, towards both integrals.

FURTHER CONTOURS 100

Solution Let R > 1 and let U be the cut complex plane with the negative imaginary axis
removed. Consider the positively oriented contour Γ comprising [−R, R] and γ + (0, R) and the
function √3
z log z
f(z) = (z ∈ U )
1 + z2
where  
√ 1
log z = log |z| + i arg z, 3
z = exp log z
3
√
and − π2 < arg z < 3π
2
. By these definitions 3
i = eπi/6 and log i = πi/2. So
√ √ 
3
z log z 3
i log i π 2 i πi/6
2
dz = 2πires (f ; i) = 2πi = e .
Γ 1+z 2i 2

Now
 √  √ 
 3
z log z   3 z log z  πR4/3 (log R + π)
 dz   πR sup   → 0 as R → ∞.
 2 1 + z2  R2 − 1
γ + (0,R) 1 + z z∈γ + (0,R)
√ √
For z = −x where x > 0 we have 3 −x = 3 xeπi/3 and log (−x) = log x + iπ. Hence, letting
R → ∞ we have
∞√ ∞√
3
xeπi/3 (log x + iπ) 3
x log x π 2 i πi/6
dx + dx = e
0 1 + x2 0 1 + x2 2
which simplifies to
√ ∞ √
πi/3
3
x  πi/3
 ∞ 3 x log x π 2 i πi/6
iπe dx + 1 + e dx = e .
0 1 + x2 0 1 + x2 2
    √
As eπi/6 / 1 + eπi/3 = 1/ e−πi/6 + eπi/6 = 1/ (2 cos π/6) = 1/ 3 then
√ ∞√
iπeπi/6 ∞ 3 x 3
x log x π2i
√ dx + dx = √ . (13.3)
3 0 1 + x2 0 1 + x2 2 3
Taking imaginary parts we find
√ ∞ √
π ∞ 3x π2 3
x π
2
dx = √ =⇒ 2
dx = √
2 0 1+x 2 3 0 1+x 3
and then taking real parts in (13.3) we find
∞√   ∞ √   
3
x log x iπeπi/6 3
x π π π2
dx = Re − √ dx = √ √ = .
0 1 + x2 3 0 1 + x2 2 3 3 6

Example 271 By considering a rectangular contour Γ with corners at R, R + iπ, −R + iπ,

−R, and with an appropriate indentation, determine
∞
x
dx.
0 sinh x

FURTHER CONTOURS 101

Solution We will consider the function f(z) = z/ sinh z on the suggested contour. Then f (z)
has a removable singularity at 0 (which we can remove, and so treat the function as holomorphic
there) and also a simple pole at z = πi which we need to indent around.

Im
-R+Πi R+Πi

Re
-R R

As f(z) is holomorphic in the rectangle then

f (z) dz = 0
Γ

by Cauchy’s Theorem. The contribution from the rectangle’s sides satisfy

 π   π   π 
 ±R + iy   2 (±R + iy)   2 (±R + iy)  Rπ + π 2 /2
 i dy = dy  dy  →0
   ±R+iy − e∓R−iy   ±R − e∓R | 
0 sinh(±R + iy) 0 e 0 |e sinh R

as R → ∞. The residue at πi is
πi
res(f(z); πi) = = −πi.
cosh πi
Also note along the top of the rectangle that the function is
x + πi x + πi
f (x + πi) = =− .
sinh(x + πi) sinh x

(This identity, coming from the periodicity of sinh x, is the very reason for choosing the rec-
tangular contour.) Hence letting R → ∞ and shrinking the indentation to zero, we see
∞  ε −∞ 
x dx −(x + πi) dx −(x + πi) dx
+ lim + + (−π) × ι × (−πi) = 0.
−∞ sinh x ε→0 ∞ sinh x −ε sinh x

Comparing real parts we find ∞

x dx π2
= .
−∞ sinh x 2

Remark 272 We have not discussed in any detail precisely what integration theory we are
working within when calculating these integrals. In Prelims Analysis III, all integrals discussed

FURTHER CONTOURS 102

were on closed bounded intervals. For those who go on to take Part A Integration they will see
that many of the integrals such as
∞ ∞ ∞
x dx dx cos ax
, 3 6
, dx
−∞ sinh x 0 1+x +x 0 1 + x2

exist as proper Lebesgue integrals. However integrals (usually those that require Jordan’s
Lemma to evaluate them) such as
∞ ∞
cos πx x sin x
dx, dx
−∞ 2x − 1 0 1 + x2

only exist as improper Lebesgue integrals — that is the limits

R R
cos πx x sin x
lim dx, lim dx
R→∞ −R 2x − 1 R→∞ 0 1 + x2

but the integrals are not proper Lebesgue integrals. This is closely related to the difference
between absolutely summable and conditionally summable series. The limit
R
x |sin x|
lim dx
R→∞ 0 1 + x2
does not exist and if the signed area in these improper integrals were calculated in a different
order to the usual one then different answers could be determined. The values we have assigned
to such an improper Lebesgue integral is called the principal value integral.

FURTHER CONTOURS 103

14. Riemann Sphere. Möbius Transformations
Definition 273 Let S 2 denote the unit sphere x2 + y 2 + z 2 = 1 in R3 and let N = (0, 0, 1)
denote the "north pole" of S 2 . Given a point M ∈ S 2 , other than N , then the line connecting N
and M intersects the xy-plane at a point P . If with identify the xy-plane with C in the natural
way, then stereographic projection is the map

π : S 2 \ {N } → C defined by M → P.

Proposition 274 The map π is given by

a + ib
π(a, b, c) = .
1−c
The inverse map is given by

−1 (2x, 2y, x2 + y 2 − 1)
π (x + iy) = .
1 + x2 + y 2
Proof Say M = (a, b, c). Then the line connecting M and N can be written parametrically as

r(t) = (0, 0, 1) + t (a, b, c − 1) .

This intersects the xy-plane when 1 + t (z − 1) = 0, i.e. when t = (1 − z)−1 . Hence

   
1 a b
P =r = ,
1−c 1−c 1−c
which is identified with
a + ib
∈ C.
1−c

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 104

On the other hand, if π(a, b, c) = x + iy then

a + ib
= x + iy and a2 + b2 + c2 = 1.
1−c
Hence (a − ib)/(1 − c) = x − iy and so
  
2 2 a + ib a − ib a 2 + b2 1 − c2 1+c 2
x +y = = 2 = 2 = = −1 + ,
1−c 1−c (1 − c) (1 − c) 1−c 1−c

giving
2
= 1−c
1 + x2 + y 2
and
x2 + y 2 − 1
c= .
x2 + y 2 + 1
Then
2 (x + iy)
a + ib =
x2 + y 2 + 1
and we may compare real and imaginary parts for the result.

Definition 275 If we identify, via stereographic projection, points in S 2 \ {N} with points in
the complex plane we see that points near N correspond to distant points in all directions. It
makes a certain sense then to include a single point ∞ which is "out there" in all directions.
The North Pole N then corresponds with ∞ under stereographic projection.

Definition 276 The extended complex plane is the set C ∪ {∞} and we denote this by C̃.

Definition 277 When we identify S 2 with C̃ using stereographic projection, S 2 is known as

the Riemann sphere.

Corollary 278 If M corresponds to z ∈ C̃ then the antipodal point −M corresponds to −1/z̄.

Proof Say M = (a, b, c) which corresponds to z = (a + ib) / (1 − c). Then −M corresponds to

−a − ib
w= ,
1+c
and
(a − ib) (−a − ib) −a2 − b2 c2 − 1
wz̄ = = = = −1.
(1 − c) (1 + c) 1 − c2 1 − c2

Theorem 279 Circles and lines in the complex plane correspond to circles on the Riemann
sphere and vice-versa.

Remark 280 Note that lines in C̃ simply correspond to circles on S 2 that pass through N.

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 105

Proof Consider the plane Π with equation Aa + Bb + Cc = D. This plane will intersect with
S 2 in a circle if A2 + B 2 + C 2 > D2 . Recall that the point corresponding to z = x + iy is

(2x, 2y, x2 + y 2 − 1)
(a, b, c) =
1 + x2 + y 2
which lies in the plane Aa + Bb + Cc = D if and only if
   
2Ax + 2By + C x2 + y 2 − 1 = D 1 + x2 + y 2 .

This can be rewritten as

 
(C − D) x2 + y 2 + 2Ax + 2By + (−C − D) = 0.

This is the equation of a circle in C if C = D. The centre is (A/ (D − C) , B/ (D − C)) and

the radius is √
A2 + B 2 + C 2 − D2
.
C −D
Furthermore all circles can be written in this form – we can see this by setting C − D = 1 and
letting A, B, C + D vary arbitrarily. On the other hand if C = D then we have the equation

Ax + By = C

which is the equation of a line and moreover any line can be written in this form. Note that
C = D if and only if N = (0, 0, 1) lies in the plane. So under stereographic projection lines in
the complex plane correspond to circles on S 2 which pass through the north pole.

Definition 281 We will use the term circline to denote any subset which is a circle or a line
in the extended complex plane.

Proposition 282 Stereographic projection is conformal (i.e. angle-preserving).

Proof Without loss of generality we can consider the angle defined by the real axis and an
arbitrary line meeting it at the point p ∈ R and making an angle θ. So points on the two lines
can be parametrized as
z = p + t, z = p + teiθ ,
where t is real. These points map onto the sphere as
   
2 (p + t) , 0, (p + t)2 − 1 2 (p + t cos θ) , 2t sin θ, (p + t cos θ)2 + t2 sin2 θ − 1
r(t) = , s(t) = .
1 + (p + t)2 1 + (p + t cos θ)2 + t2 sin2 θ

Then

′ (2 (p2 − 1) , 0, 4p) ′ (2 (p2 − 1) cos θ, 2 (1 + p2 ) sin θ, 4p cos θ)

r (0) = , s (0) = .
(1 + p2 )2 (p2 + 1)2

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 106

So the angle φ between these tangent vectors is given by
 
2
4 (p2 − 1) cos θ + 0 + 16p2 cos θ
cos φ =
4 (p − 1) + 16p 4 (p2 − 1)2 cos2 θ + 4 (1 + p2 )2 sin2 θ + 16p2 cos2 θ
2 2 2

2
4 (p2 + 1) cos θ
=
{2 (p2 + 1)} {2 (p2 + 1)}
= cos θ.

Hence stereographic projection is conformal as required.

Definition 283 A Möbius transformation is a map f : C̃ → C̃ of the form

az + b
f (z) = where ad = bc.
cz + d
We define  a
c
if c = 0,
f(∞) =
∞ if c = 0,
and also f (−d/c) = ∞ if c = 0.

Remark 284 The Möbius transformations clearly have a "matrix" look to them and there
are good reasons for this, though the full reason is beyond this course and more a matter for
Trinity’s Projective Geometry course. The extended complex plane is the complex projective
line P (C2 ) and the Möbius transformations are its projective transformations P GL (2, C).
The complex projective line is as follows. Suppose that we consider the equivalence classes
of non-zero complex pairs C2 \ {(0, 0)} under the equivalence relation

(z1 , w1 ) ∼ (z2 , w2 ) if and only there is a non-zero λ ∈ C\ {0} such that z1 = λw1 , z2 = λw2 .

Then each equivalence class has a representative of the form (z, 1) where z ∈ C except (1, 0).
The former can be thought of as z and the latter as ∞.
Matrices don’t act on these pairs in a well-defined way, but equivalence classes of invertible
matrices do. If we identify 2 × 2 complex matrices under the equivalence relation

M1 ∼ M2 if and only if there is a non-zero λ ∈ C\ {0} , such that M1 = λM2

then the Möbius transformations can be identified with the equivalence classes of invertible ma-
trices which is denoted P GL (2, C), the Projective General Linear group. The Möbius trans-
formation
 
az + b a b
is then identified with the equivalence class of .
cz + d c d

Proposition 285 Möbius transformations form the group of transformations C̃ → C̃ generated

(under composition) by:

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 107

 • translations – maps of the form z → z + k where k ∈ C;

• scalings or dilations – maps of the form z → kz where k ∈ C\ {0};

• inversion – the map z → 1/z. (Note this map is not an actual inversion in the sense
of inverting in a circle.)

Proof Note that if c = 0 then

az + b a bc − ad
= − 2
cz + d c c z + dc
is a composition of various translations, scalings and inversion. If c = 0, then d = 0 and clearly
z → (a/d) z + (b/d) is a composition of a scaling and a translation.
This shows that the set of Möbius transformations are a subset of the group generated by
translations, scalings and inversion. It is also clear that translations, scalings and inversion are
all special types of Möbius transformations. Further if f (z) = (az + b)/(cz + d) is a Möbius
transformation then
az + (ak + b)
f (z + k) = where a (ck + d) − c (ak + b) = ad − bc = 0;
cz + (ck + d)
akz + b
f (kz) = where (ak) d − b (ck) = k (ad − bc) = 0 if k = 0;
  ckz + d
1 bz + a
f = where bc − ad = 0.
z dz + c

Hence Möbius transformations composed with these generators yield further Möbius transfor-
mations and the result follows.

Proposition 286 The Möbius transformations are bijections from C̃ to C̃.

Proof

• The translation z → z + k, ∞ → ∞ is clearly a bijection with inverse z → z − k, ∞ → ∞.

• The scaling z → kz, ∞ → ∞ is clearly a bijection with inverse z → z/k, ∞ → ∞.

• And inversion z → 1/z, 0 → ∞, ∞ → 0 is a bijection which is its own inverse.

Hence any composition of these maps, i.e. the Möbius transformations, are all bijections.

Remark 287 In fact, though we shall not prove this, the Möbius transformations correspond
to the conformal bijections of S 2 . Given any bijection f of S 2 there is a corresponding bijec-
tion πf π −1 of C̃. Identified this way, the conformal bijections of S 2 correspond to the Möbius
transformations.

Proposition 288 Given two triples of distinct points z1 , z2 , z3 ∈ C̃ and w1 , w2 , w3 ∈ C̃ then

there is a unique Möbius transformation f such that f (zi ) = wi for i = 1, 2, 3.

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 108

Proof Note that the map
(z2 − z3 ) (z − z1 )
f(z) =
(z2 − z1 ) (z − z3 )
is a Möbius transformation which maps z1 , z2 , z3 to 0, 1, ∞. There is a similar Möbius transfor-
mation g which sends w1 , w2 , w3 to 0, 1, ∞. By Proposition 285 g −1 f is a Möbius transformation
which maps each zi to wi .
To show uniqueness suppose h is another such map. Then ghf −1 is a Möbius transformation
which maps 0, 1, ∞ to 0, 1, ∞. If we write
az + b
ghf −1 (z) =
cz + d
then 0 → 0 means b = 0, 1 → 1 means a + b = c + d and ∞ → ∞ means c = 0. Hence
ghf −1 (z) = z for all z and h = g −1 f as required.

Remark 289 Recall from Proposition 133 that the circlines are the solutions sets of the equa-
tions
Azz̄ + B̄z + B z̄ + C = 0 (14.1)
where A, C ∈ R and B ∈ C.

Proposition 290 The Möbius transformations map circlines to circlines.

Proof Inversion z → 1/z maps the circline (14.1) to the circline with equation

Czz̄ + B̄ z̄ + Bz + A = 0,

and it is clear that translations also map circlines to circlines. Also the image of (14.1) under
the scaling z → kz has equation

Azz̄ + Bkz + Bkz̄ + |k|2 C = 0

which is another circline. Hence a Möbius transformation, which can be written as a composi-
tion of these maps, also maps circlines to circlines.

Example 291 Show that the maps

eiθ (z − a)
f (z) = where θ ∈ R and |a| < 1
1 − āz
maps the unit circle |z| = 1 to itself and a to 0.

Solution Clearly a maps to 0. Suppose now that |z| = 1. Then

 iθ   
 e (z − a)  |z − a| |z̄| |1 − az̄|  1 − az̄ 
|f(z)| =  = = = =1
1 − āz  |1 − āz| |1 − āz|  1 − āz 

as the numerator is the conjugate of the denominator.

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 109

Example 292 Find a Möbius transformation which maps the unit circle |z| = 1 to the real
axis and the real axis to the imaginary axis.

Solution Initially it might well seem that the two diagrams – a circle divided by a line and
two perpendicular lines meeting at the origin – are very different. In one, the circle meets the
real axis at right angles at two points; in the second diagram this intersection is still there but
"hidden" at infinity. So if we take one of these intersections, say 1, to ∞ and the other, −1, to
the origin, the diagrams will look much more similar. Let’s start then with
z+1
f (z) = .
z−1
As required f takes −1 to 0 and 1 to ∞. Also f(0) = −1 and so f takes the real axis to a
circline containing 0, −1, ∞, that is the real axis (rather than the imaginary axis as desired).
It takes the unit circle to a circle containing 0, ∞ and f(i) = (i + 1)/(i − 1) = −i, that is the
imaginary axis (rather than the real axis as desired). To unmuddle our images we can rotate
by π/2 about the origin – hence we see the map
 
z+1
z → i
z−1

solves the given problem.

RIEMANN SPHERE. MÖBIUS TRANSFORMATIONS 110

15. Conformal Maps
Definition 293 A holomorphic map f : U → C is said to be conformal if f ′ (z) = 0 for all
z ∈ U.

Proposition 294 A conformal map is angle-preserving and sense-preserving.

Proof Let f : U → C be a holomorphic map on an open set U. Let z0 ∈ U and let γ 1 : [−1, 1] →
U and γ 2 : [−1, 1] → U be two paths which meet at z0 = γ 1 (0) = γ 2 (0). The original curves
meet at z0 in the (signed) angle
′ γ ′2 (0)
θ = arg γ ′2 (0) − arg γ 1 (0) = arg .
γ ′1 (0)
The images of the curves f(γ 1 ) and f (γ 2 ) meet at f (z0 ) at angle

φ = arg (f γ 2 )′ (0) − arg (f γ 1 )′ (0)

(f γ 2 )′ (0)
= arg
(f γ 1 )′ (0)
f ′ (γ (0)) γ ′2 (0)
= arg ′ 2 [by the chain rule]
f (γ 1 (0)) γ ′1 (0)
f ′ (z0 ) γ ′2 (0)
= arg ′ [as γ 2 (0) = γ 1 (0) = z0 ]
f (z0 ) γ ′1 (0)
γ ′ (0)
= arg 2′ = θ.
γ 1 (0)

For various reasons, relating to areas as diverse as harmonic analysis and fluid dynamics, it
is important to know what domains can be mapped to other domains by conformal bijections.
Such domains would then be said to be conformally equivalent. We meet now our main
examples of conformal maps.

Example 295 There are three main examples of conformal maps which we shall use.

• Möbius Transformations.
The Möbius transformations are conformal. These are maps of the form
az + b
f (z) = (ad = bc)
cz + d
so that
a (cz + d) − c (az + b) ad − bc
f ′ (z) = 2 = = 0.
(cz + d) (cz + d)2
In fact, the Möbius transformations are precisely the group of conformal bijections of the
Riemann Sphere. These maps are particularly useful for conformally mapping regions
which are bounded by circlines.

CONFORMAL MAPS 111

 • Power Maps. Maps of the form
z → zα
are conformal except when z = 0. These are very useful for changing the angle at the
origin — for example, the map z → z 2 maps the quadrant {z : Re z > 0, Im z > 0} to the
upper half-plane. This is not a cheat as the origin is on the boundary of the quadrant but
crucially not in it.
• Exponential. This is particularly useful for mapping semi-infinite and infinite bars, see
the example below.

Remark 296 Recall that:

(a) Möbius transformations map the set of circlines (circles and lines) to the set of circlines.
(b) A circline is uniquely determined by three points.
(c) The Möbius transformations are bijections on C̃ = C ∪ {∞} .

Example 297 Determine the image of the semi-infinite rectangles

A = {z : 0 < Im z < π, Re z < 0} , B = {z : 0 < Im z < π, Re z > 0}

under the maps

f (z) = exp z and g(z) = exp(iz).

Solution An arbitrary element of A is of the form z = x + iy where y ∈ (0, π) and x < 0. So

f (z) = ex+iy = ex eiy

is a general element of the half-disc D(0, 1) ∩ H + as ex < 1 and Im eiy > 0. We likewise see that
the image of B under f is H + \D̄(0, 1).
We can similarly note that when y ∈ (0, π) and x < 0 then g(z) = e−y eix is an arbitrary
complex number in the annulus {z : e−π < |z| < 1}. Note that g is not bijective on A repeatedly
wrapping A around the annulus. Again g maps B to this annulus repeatedly.
Im
Im

fHBL
A B −→
f
fHAL
Re Re

Remark 298 From the Exercise sheets we know that we could use sin to take

{z : −π/2 < Re z < π/2, Im z > 0}

to the upper half-plane. We could suitably adjust sin and instead use

sin(−iz − π/2) or sin(iz + π/2)

to respectively take A and B to the upper half-plane.

CONFORMAL MAPS 112

Example 299 Determine the image of the quadrant Q = {z : Re z > 0, Im z > 0} and the half-
plane H + = {z : Re z > 0} under the Möbius transformation
z−i
f(z) = .
z+i

Solution Note that both Q and H + are bounded by circlines (lines in fact) and so their images
will be bounded by circlines also. Note further that
1 − 2i −1 − 2i
f(0) = −1, f(1) = −i, f (∞) = 1, f(i) = 0, f(1 + i) = , f (−1 + i) = .
5 5
Hence the positive imaginary axis maps to the interval (−1, 1) and the positive real axis maps
to the lower unit semicircle γ − (0, 1) (the circline connecting −1, −i and 1). As 1 + i maps into
the lower unit semi-disc that then that is f (Q). The image of H + \Q makes the remainder of
the unit disc as shown in the diagram below.
Im
Im

fHH +\QL

Re

H +\Q Q f fHQL
−→
Re

Remark 300 It is in fact quite easy to see why the map f above takes the upper half-plane
onto the unit disc. The upper half-plane can be characterized as those points closer to i than to
−i. Hence  
z − i
Im z > 0 ⇐⇒ |z − i| < |z + i| ⇐⇒    < 1.
z + i
We can see that other maps (z − a)/(z − ā) would also work for any a ∈ H + and that more
generally any half-plane can be mapped to the unit disc by a map (z − α)/(z − β) where α is a
point in the half-plane and β is the mirror image of α across the bounding line.

Example 301 Find a conformal map which maps the half-disc

U = {z ∈ C : |z| < 1, Im z > 0}

bijectively onto D(0, 1). Explain why it is not possible to do this using only Möbius transfor-
mations.

Remark 302 An important tip. If faced with a domain U bounded by two circline arcs or
segments which meet at α and β then the best thing to do first, almost without any further
consideration, is to apply the map (z − α)/(z − β). This map sends α to 0 and β to ∞ and so
takes the two circline arcs or segments to half-lines meeting at the origin.

CONFORMAL MAPS 113

Solution In light of the above remark, the domain U is best thought of as being bounded by
two circlines, namely γ(0, 1) and the real axis, which meet at ±1, and so we will begin with
z−1
f (z) = .
z+1
As
i−2 −3 + 4i
f (1) = 0, f (−1) = ∞, f (i) = i, f (0) = −1, f (i/2) = =
i+2 5
then γ(0, 1) maps to the imaginary axis and the real axis maps to the real axis. Given the
position of f (i/2) then we know f (U ) to be the second quadrant. As Mobius maps are bijections
on the extended complex plane then f is a bijection onto its image.
The map g(z) = z 2 is conformal (except at the origin) and bijective on f (U ) as it includes
no point z and its negative −z. We have g(f (U )) is the lower half-plane.
Finally the lower half-plane can be identified as precisely those points closer to −i than i
(mathematically |z + i| < |z − i|) and hence
z+i
h(z) =
z−i
maps the lower half-plane bijectively onto the unit disc. A map that then meets the require-
ments of the question is
 z−1 2
+i
h(g(f (z))) =  z+1 2
z−1
z+1
−i
(z − 1)2 + i (z + 1)2
=
(z − 1)2 − i (z + 1)2
(1 + i) (z 2 + 1) + 2z (i − 1)
=
(1 − i) (z 2 + 1) − 2z (1 + i)
 2 
z + 2iz + 1
= i 2 .
z − 2iz + 1
Im Im
Im Im
Re

fHUL
gfHUL Re

f g h
U −→ −→ −→ hgfHUL

Re Re

That there is no Möbius transformation which performs this is clear because any such map
would keep the two circlines bounding U as two distinct circlines, whereas D(0, 1) is bounded
by just one circline.

Example 303 Find a conformal bijection mapping U = {z ∈ C : |z| < 1, Im z > 0} to D(0, 1)
and such that i/2 maps to 0.

CONFORMAL MAPS 114

Solution In our previous solution, we see that hgf (i/2) = −i/7. However we don’t need
to reinvent the wheel here as we know that for any a in the lower half-plane then the map
(z − a)/(z − ā) will suffice as a replacement for h. So note
 2  2  2
i/2 − 1 −2 + i −3 + 4i −7 − 24i
gf(i/2) = = = = .
i/2 + 1 2+i 5 25

If we then instead use

25z + (7 + 24i)
h̃(z) =
25z + (7 − 24i)
then h̃gf will do the job.

Example 304 Show that D(0, 1) and C are not conformally equivalent.

Solution Let f : C → D(0, 1) be a holomorphic map. In particular it is bounded and so, by

Liouville’s Theorem, f is constant and so certainly not a bijection.

One might ask what domains are conformally equivalent to D(0, 1). As conformal equiva-
lence is a particular type of topological isomorphism (the technical term for a topological iso-
morphism being homeomorphism) then any such domain would have to be simply-connected.
But there is a wide variety of such domains, many rather nasty boundaries, and so the following
result may come as striking.

Theorem 305 (Riemann Mapping Theorem) Let U be a simply-connected domain with

U = C. Then U is conformally equivalent to D(0, 1). In the case the the boundary of U is
smooth then the conformal equivalence can be extended between U ∪ ∂U and D(0, 1).

Proof The proof of this is well beyond this course. Interested readers might try Conway
Functions of One Complex Variable, §7.4.

An important use of conformal maps is their use in harmonic analysis and begins with the
following result. A conformal equivalence between domains U and V induces an isomorphism
between the space of harmonic functions on each domain. So if we are seeking (for example) to
solve Laplace’s Equation in a simply-connected proper domain of C with a smooth boundary
then the Riemann Mapping Theorem tells us (in principle at least) that this is no harder than
solving the problem on the unit disc or upper half-plane.

Proposition 306 Let U and V be open subsets of C, let f : U → V be holomorphic and let
φ : V → R be harmonic. Then φ ◦ f is harmonic.

Proof Let u = Re f and v = Im f and we will also write ψ = φ ◦ f . If z = x + iy then we can

write ψ as
ψ(x + iy) = φ(u(x, y), v(x, y)).
By the chain rule we have

ψ x = φu ux + φv vx , ψ y = φu uy + φv vy

CONFORMAL MAPS 115

and applying it again we get

ψ xx = (φuu ux + φuv vx ) ux + φu uxx + (φvu ux + φvv vx ) vx + φv vxx

= φu uxx + φv vxx + φuu u2x + 2φuv ux vx + φvv vx2
ψ yy = (φuu uy + φuv vy ) uy + φu uyy + (φvu uy + φvv vy ) vy + φv vyy
= φu uyy + φv vyy + φuu u2y + 2φuv uy vy + φvv vy2

Remember that as φ is harmonic then φxx + φyy = 0. Also, by the Cauchy-Riemann equations,
we have
ux = vy , uy = −vx , uxx + uyy = 0, vxx + vyy = 0.
Hence

ψ xx + ψ yy = φu (uxx + uyy ) + φv (vxx + vyy ) + φuu (u2x + u2y ) + 2φuv (ux vx + uy vy ) + φvv (vx2 + vy2 )
= φuu (u2x + u2y ) + φvv (vx2 + vy2 ) [as consequences of the CREqs].

Now, also by the Cauchy-Riemann equations, we have

2
|f ′ (z)| = u2x + u2y = vx2 + vy2 .

So
2
ψ xx + ψ yy = |f ′ (z)| (φuu + φvv ) = 0
as φ is harmonic. Hence ψ is harmonic also.

Remark 307 You may recall from the first year meeting parabolic coordinates
x2 − y 2
u= , v = xy.
2
If g(u, v) = f (x, y) then it is the case that ∇2 f = 0 if and only if ∇2 g = 0. This does not seem
surprising now in light of the above result with f = u + iv = z 2 /2.

Definition 308 Given an open subset U of C, we will denote by H(U ) the vector space of
harmonic functions on U.

Corollary 309 Let U and V be conformally equivalent open subsets of C. Then H(U ) and
H(V ) are isomorphic as vector spaces.

Proof Let f : U → V be a conformal equivalence. This then induces map

H(V ) → H(U ) given by φ → φ ◦ f,

H(U) → H(V ) given by φ →  φ ◦ f −1 ,

which are clearly linear (in φ) and inverses of one another.

This introduces powerful techniques for handling the Dirichlet Problem (after the German
mathematician Lejeune Dirichlet 1805-1859). The Dirichlet Problem is typically stated as
follows:

CONFORMAL MAPS 116

 • Given a domain U and a function g defined on the boundary ∂U, is there a unique function
f such that f = g on ∂U and ∇2 f = 0 in U .

The Dirichlet problem was rigorously demonstrated by Hilbert in 1900 for suitably nice
boundaries (after Weierstrass has shown Dirichlet’s own proof lacked rigour). The problem has
the following solution in the upper half-plane.

Proposition 310 (Dirichlet Problem in the Half-Plane) (Off syllabus) The function

y ∞ f (t)
u(x, y) = dt
π −∞ (t − x)2 + y 2

is harmonic in the upper half-plane and satisfies u(x, 0+ ) = f(x).

Proof See Priestley 23.14 and 23.16.

Example 311 Use the above proposition to solve Dirichlet problem in the half-plane when

1 t<0
f (t) =
0 t>0

Solution We have, for y > 0

y 0 dt
u(x, y) =
π −∞ (t − x)2 + y 2

y −x dt
=
π −∞ t2 + y 2
 !−x
1 −1 t
= tan
π y −∞
 
1 −1 x π
= − tan +
π y 2
1 y
= tan−1
π x
arg (x + iy)
=
π
with arg(x + iy) taken in the range (0, π) .

Example 312 By means of conformal maps, solve the Dirichlet problem in:
(a) the quadrant {z : Re z > 0, Im z > 0} subject to the boundary conditions

u(x, 0) = 0 (x > 0), u(0, y) = 1 (y > 0).

(b) the half-disc {z : Im z > 0, |z| < 1} subject to the boundary conditions

u(z) = 0 (|z| = 1), u(x, 0) = 1 (|x| < 1),

CONFORMAL MAPS 117

Solution (a) We know z → z 2 is a conformal equivalence between the quadrant and the upper
half-plane which matches up the boundary conditions correctly. So, by Proposition 306, we see
that we should assign to the value z (in the quadrant) the value of the solution at z 2 (in the
half-plane). Hence our solution in the quadrant is
 
arg (x + iy)2 2 arg (x + iy)
u(x, y) = = .
π π
(b) We saw in Example 301 that the upper half-disc is mapped to the upper half-plane by
 
z−1 2
w(z) = − ,
z+1
and that this map takes the bottom of the half-disc to the negative real axis and the semicircular
arc to the positive real axis. So, again by Proposition 306, we need to assign to z the solution
at w. Hence our solution is #  $
1 z−1 2
u(z) = arg −
π z+1
with arg taking values in the range (0, π) .

Example 313 (Off-syllabus) (Fluid Flow in the Plane) Let ω = φ + iψ be a holomorphic

function on a domain U. Such a function can be considered as a complex potential of a steady,
irrotatonal, incompressible, inviscid fluid with velocity potential φ and streamfunction ψ.
The velocity of the flow is then
u = (φx , φy ).
That the flow is incompressible means div u = 0; this is true here as

div u = φxx + φyy = 0

and we know φ = Re ω to be harmonic. That the flow is irrotational means curl u = 0 which
holds true as  
curl u = φyx − φxy k = 0.
The streamlines of the flow are the curves ψ = c (a constant).

Example 314 An example of a uniform flow in half-plane H = {z : Im z > 0} is u = (U, 0)

where U is a (real) constant, and we can see that this corresponds to the complex potential

ω = U z = Ux + iU y.

The streamlines are (unsurprisingly) the curves y = c (a positive constant).

We know from the exercise sheets that sine is a conformal equivalence between
U = {z : Im z > 0, −π/2 < Re z < π/2}

and H. Any complex potential ω(z) on H corresponds to a complex potential ω(sin z) on U

(and vice versa). Recall that

sin(x + iy) = sin x cosh y + cos x sinh y.

CONFORMAL MAPS 118

The above uniform flow in H corresponds to a flow in U with streamlines

cos x sinh y = c (where c > 0).

These streamlines are sketched below.

-1.5 -1.0 -0.5 0.5 1.0 1.5

CONFORMAL MAPS 119