Lecture notes on Complex Analysis

Sourov Roy
IACS, Kolkata

1 Introduction

A complex number is an ordered pair of two real numbers, (a, b) or a + ib. i = (−1)1/2 .
A complex variable is an ordered pair of two real variables, z = (x, y) = x + iy.

We can represent complex numbers by points in the xy-plane called the complex plane or
Argand plane. Given a complex number z = x + iy, its real and imaginary parts define
an element (x, y) of R2 .

z = x + iy

In fact, this identification is one of real vector spaces., in the sense that adding complex
numbers and multiplying them with real scalars mimic similar operations one can do in
R2 . If α ∈ R, then to αz = (αx) + i(αy) there corresponds the pair (αx, αy) = α(x, y).

Similarly, if z1 = x1 + iy1 and z2 = x2 + iy2 are complex numbers, then z1 + z2 =

(x1 + x2 ) + i(y1 + y2 ), whose associated pair is (x1 + x2 , y1 + y2 ) = (x1 , y1 ) + (x2 , y2 ).

In fact, the identification is even one of Euclidean spaces. Given a complex number
p
z = x + iy, its modulus |z|, defined by |z|2 = zz ∗ , is given by x2 + y 2 , which is the norm
||(x, y)|| of the pair (x, y).

The complex conjugate of a complex number z = x1 +iy1 is z ∗ = x1 −iy1 (or z̄ = x1 −iy1 ).

Similarly, if z1 = x1 + iy1 and z2 = x2 + iy2 , then

Re(z1∗ z2 ) = Re[(x1 − iy1 )(x2 + iy2 )]

= Re[x1 x2 + y1 y2 + i(x1 y2 − y1 x2 )]
= x1 x 2 + y 1 y 2 (1.1)

This is the dot product of the pairs (x1 , y1 ) and (x2 , y2 ).

1
We shall denote the complex plane, that is the set of all complex numbers, by C and any
other set, unless otherwise qualified, is assumed to be a subset of C. The real line will be
denoted by R.

The complex plane is assumed to be endowed with a metric induced by the norm,

|·| : C → R, z 7→ |z| (1.2)

The distance between two points z1 = x1 + iy1 and z2 = x2 + iy2 in the complex plane
p
is given by |z1 − z2 | = (x1 − x2 )2 + (y1 − y2 )2 . Any -neighbourhood is defined with
respect to this metric.

1.1 Polar form and the argument function

Points in the plane can also be represented using polar coordinates, and this representation
in turn translates into a representation of the complex numbers. Let (x, y) be a point in

z = reiθ
r
θ

p
the plane. If we define r = x2 + y 2 and θ = tan−1 (y/x), then we can write (x, y) =
(r cos θ, r sin θ) = r(cos θ, sin θ). The complex number z = x + iy can then be written as
z = r(cos θ + i sin θ).

By assuming ex = 1 + x + x2 /! + x3 /3! + ... holds when x = iθ, we can arrive at the result

eiθ = cos θ + i sin θ (1.3)

The real number r is the modulus |z| of z and the complex number cos θ + i sin θ has unit
modulus. The angle θ is called the argument of z and is written arg(z). Therefore, we
have the following polar form for a complex number z :

z = |z|ei arg(z)
(1.4)

2
Being an angle, the argument of a complex number is only defined up to the addition
of integer multiples of 2π. In other words, it is a multiple-valued function. This
ambiguity can be resolved by defining the principal value Arg of the argument function
to take values in the interval (−π, π]; that is for any complex number z, one has

−π < Arg(z) ≤ π (1.5)

Notice, however, that Arg is not a continuous function: it has a discontinuity along the
negative real axis. Approaching a point on the negative real axis from the upper half
plane, the principal value of its argument approaches π, whereas if we approach it from
the lower half plane, the principal value of its argument approaches −π.

Notice that the modulus is a multiplicative function: |zw| = |z||w|, the argument is
additive:

arg(z1 z2 ) = arg(z1 ) + arg(z2 ) (1.6)

provided that we understand the equation to hold up to integer multiples of 2π. Also mod-
ulus is invariant under conjugation |z ∗ | = |z| , the argument changes sign arg(z ∗ ) = −arg(z) ,
again up to integer multiples of 2π.

2 Some important subsets of the complex plane

2.1 Open and closed sets

Definition For a point z0 in C, the set

D(z0 , r) = {z ∈ C; |z − z0 | < r, r > 0} (2.1)

is called an open disk of radius r, centered at z0 .

The set

D̄(z0 , r) = {z ∈ C; |z − z0 | ≤ r, r > 0} (2.2)

is called a closed disk of radius r, centered at z0 , which consists of the open disk of
radius r and the circle |z − z0 | = r, which forms its boundary.

Definition A subset U of C is said to be open if given any z ∈ U , there exists a positive

real number  > 0 (which can depend on z) such that the open disk of radius  centred
at z, also belongs to U .

3
A set C is said to be closed if its complement C c = {z ∈ C|z 6= C}- that is, all those
points not in C - is open.

By a neighbourhood of a point z0 in the complex plane, we will mean any open set
containing z0 . For example, any open disk of radius , centered at z0 , is a neighbourhood
of z0 . |z − z0 | < .

0 < |z − z0 | <  is called deleted neighbourhood, consisting of all points z in an 

neighbourhood of z0 except for the point z0 itself.

A point z0 is said to be an interior point of a set S whenever there is some neighbourhood

of z0 that contains only point of S. It is called an exterior point of S when there exists
a neighbourhood of it containing no points of S. If z0 is neither of these, it is a boundary
point of S.

A boundary point is, therefore, a point all of whose neighbourhoods contain at least one
point in S and at least one point not is S. The totality of all boundary points is called
the boundary of S. The circle |z| = 1, for example, is the boundary of each of the sets

|z| < 1 and |z| ≤ 1

Example The set U = {z : |z| < 1} is an open set. The set U = {z : |z| ≤ 1} is closed,
√
since the point P = (1 + i)/ 2, for example, does not have any neighbourhood contained
in U .

Example
Only the void set (∅) and the universal set are both open and closed. Universal set: set
of all complex numbers, has no boundary points.

One should keep in mind that generic subsets of the complex plane are neither closed nor
open. For example, 0 < |z| ≤ 1 is neither open nor closed. For a set to be not open, there
must be a boundary point that is contained in the set; and if a set is not closed, there
exists a boundary point not contained in the set.

Definition:
A point z in a subset U of C is said to be an adherent point of U if every open disk
D(z, r) contains some elements of U .

Proposition (Triangle inequality)

Let z1 , z2 ∈ C. Then the following inequality holds

4
 |z1 + z2 | ≤ |z1 | + |z2 | (2.3)

Proof:
From the definition of modulus

|z1 + z2 |2 = (z1 + z2 )(z1 + z2 )

= (z1 + z2 )(z1∗ + z2∗ )
= |z1 |2 + z1 z2∗ + z1∗ z2 + |z2 |2
= |z1 |2 + 2Re(z1 z2∗ ) + |z2 |2
≤ |z1 |2 + 2|z1 z2∗ | + |z2 |2
= |z1 |2 + 2|z1 ||z2 | + |z2 |2
= (|z1 | + |z2 |)2 (2.4)

The inequality follows by taking positive roots on both sides.

3 Complex functions

A symbol, such as z, which can stand for any one of a set of complex numbers is called a
complex variable.

If to each value which a complex variable z can assume there corresponds one or more
values of a complex variable w, we say, that w is a function of z and write w = f (z).

For example, if f (z) = z 2 , then f (2i) = (2i)2 = −4.

3.1 Single and multiple-valued functions

If only one value of w corresponds to each value of z, we say that w is a single-valued

function of z or that f (z) is single-valued.

If more than one value of w corresponds to each value of z, w is a multiple-valued function

of z.

5
Traditionally, one considers two planes: the z-plane whose points have coordinates (x, y)
corresponding to the real and imaginary parts of z = x+iy, and the w-plane whose points
have coordinates (u, v) corresponding to w = u + iv.

Any complex-valued function f of the complex variable z maps points in the z-plane to
points in the w-plane via w = f (z). A lot can be learned from a complex function by
analysing the image in the w-plane of certain sets in the z-plane.

A multiple-valued function can be considered as a collection of single-valued functions,

each member of the collection is called a branch of the function.

It is customary to consider one particular member as a principal branch of the multiple-

valued function and the value of the function corresponding to this branch as the prin-
cipal value.

Example 1:
w = f (z) = z 2 is a single-valued function of z.

Example 2:
w = f (z) = z 1/2 is a multiple-valued (two-valued) function of z.

We will learn more about multiple-valued functions later.

Inverse functions
If w = f (z), then z = f −1 (w). The function f −1 is called the inverse function corre-
sponding to f .

Transformations
If w = u + iv (u and v are real) is a single-valued function of z = x + iy (x and y are
real), we can write u + iv = f (x + iy).

Equating real and imaginary parts this is seen to be equivalent to

u = u(x, y), v = v(x, y) (3.1)

Thus given a point (x, y) in the z-plane, there corresponds a point (u, v) in the w-plane.
The set of equations (3.1) (equivalently w = f (z)) is called a transformation. We say that
point P is mapped or transformed into point P 0 by means of the transformation and call
P 0 the image of P .

Example:

6
w = z 2 , then

u + iv = (x + iy)2
= x2 − y 2 + 2ixy
=⇒ u(x, y) = x2 − y 2 , v(x, y) = 2xy (3.2)

The image of point (1, 2) in the z-plane is the point (−3, 4) in the w-plane.

4 Limits

Let f (z) be defined and single-valued in a neighbourhood of z = z0 with the possible

exception of z = z0 itself.

We say that the number f0 is the limit of f (z) as z approaches z0 and write

lim f (z) = f0 (4.1)

z→z0

if for any positive number  (however small) we can find some positive number δ, such
that

|f (z) − f0 | <  whenever 0 < |z − z0 | < δ (4.2)

In such case we also say that f (z) approaches f0 as z approaches z0 and write f (z) → f0
as z → z0 . The limit must be independent of the manner in which z approaches z0 .

Geometrically, if z0 is a point in the complex plane, then limz→z0 f (z) = f0 if the absolute
value of the difference between f (z) and f0 can be made as small as we wish by choosing
points z sufficiently close to z0 (excluding z = z0 itself).

Example:
If f (z) = z 2 , prove that limz→z0 f (z) = z02 .

Proof: We must show that given any  > 0, we can find δ (depending in general on )
such that |z 2 − z02 | <  whenever 0 < |z − z0 | < δ.

If δ ≤ 1, then 0 < |z − z0 | < δ implies that

|z 2 − z02 | = |z − z0 ||z + z0 |
< δ|z − z0 + 2z0 |
< δ{|z − z0 ||2z0 |}
< δ(1 + 2|z0 |) (4.3)

7
Take δ as 1 or /(1 + 2|z0 |), whichever is smaller.

Then we have |z 2 − z02 | <  whenever |z − z0 | < δ, and the required result is proved.

4.1 Continuity

Let f (z) be defined and single-valued in a neighbourhood of z = z0 . The function f (z) is

said to be continuous at z = z0 if

lim f (z) = f (z0 ) (4.4)

z→z0

Note that this implies three conditions which must be met in order that f (z) be continuous
at z = z0 .

1. limz→z0 f (z) = f0 must exist

2. f (z0 ) must exist, i.e., f (z) be defined at z0

3. f0 = f (z0 )

4.1.1 Continuity in a region

Any neighbourhood is a domain. A domain together with some, none or all of its boundary
points is referred to as a region.

A function f (z) is said to be continuous in a region if it is continuous at all points of the

region.

A function is said to be continuous if it is continuous at every point of C.

Example:
The polynomial function

f (z) = a0 + a1 z + a2 z 2 + · · · + an z n (4.5)

is continuous in C.

8
4.1.2 Uniform Continuity

Let f (z) be continuous in a region. Then by definition, at each point z0 of the region
and for any  > 0, we can find δ > 0 (which will, in general, depend on both  and the
particular point z0 ) such that |f (z) − f (z0 )| <  whenever |z − z0 | < δ.

If we can find δ depending on  but not on the particular point z0 , we say that f (z) is
uniformly continuous in that region.

[For a function to be continuous in a domain it suffices for it to be continuous at each

point of the domain, which allows the δ in the definition of limit to vary from one point
in U (domain) to another. If δ can be chosen independently of the point, there is some
sort of uniformity in the function. ]

Theorem:
If f (z) is continuous in a closed region, it is uniformly continuous there.

5 Derivatives

If f (z) is single-valued in some region R of the z-plane, the derivative of f (z) is defined
as
f (z + ∆z) − f (z)
f 0 (z) = lim (5.1)
∆z→0 ∆z
provided that the limit exists independent of the manner in which ∆z → 0. In such case
we say that f (z) is differentiable at z.

Here the limit ∆z → 0 is a double limit inasmuch as both the real part ∆x and the
imaginary part ∆y of ∆z must each tend separately to zero.

Since there are an infinite number of ways by which ∆z can tend to zero, there are, in
general, an infinite number of possible values that the limits can assume.

For example, to achieve the limit in Eq.(5.1) we could let |∆z| → 0 for any value of
arg(∆z)

(|∆z| → 0, arg(∆z) takes on arbitrary values)

But what is more likely is that the resulting derivative will depend on the particular value
of arg(∆z). Similarly, the limit in Eq.(5.1) will, in general, depend on the order in which
∆x, ∆y tend to zero.

9
 y

Consider as a simple example the function f (z) = x+2iy. We will show that this function
does not have a well defined derivative at the origin.

Its derivative at z = 0 is given by

f (z) − f (0)
f 0 (z) z=0
= lim
z→0 z
x + 2iy x2 + 2y 2 + ixy
= lim = lim (5.2)
x→0 x + iy
y→0
x→0
y→0
x2 + y 2

The value of f 0 (z) will obviously depend on the order in which the two limits are taken.

If x is first held fixed while y → 0, we obtain

f 0 (z) z=0
=1 (5.3)

whereas if y is first held fixed while x → 0, we obtain

f 0 (z) z=0
=2 (5.4)

Again, suppose that x and y tend to zero along some arbitrary line, y = αx. Then

0 1 + 2α2 + iα
f (z) z=0
= (5.5)
1 + α2
and the value of the derivative depends in this case on arg(z) : α = xy .

We shall say that a function of a complex argument z is differentiable at a given point

z = z0 if the limit
f (z0 + ∆z) − f (z0 )
lim (5.6)
∆z→0 ∆z

10
exists, is finite, and does not depend on the way one approaches the point z = z0 .

For real variables we require that the right hand limit (x → x0 from above) and the left
hand limit (x → x0 from below) be equal for the derivative f 0 (x) to exist at x = x0 .

Now, with z (or z0 ) some point in a plane, our requirement that the limit be independent
of the direction of approach is very restrictive. Differentiation in C thus entails a notion
of isotropy or “rotational invariance”.

An infinite number of such requirements have to be satisfied in order to ensure the dif-
ferentiability of a function of a complex argument. One can expect, therefore, that the
property of being differentiable is, in the case of functions of a complex variable, very
much more restrictive than it is for functions of a real variable.

Although differentiability implies continuity, the reverse is not true.

6 Analytic functions

A function is said to be analytic at a point z0 if there exists a neighbourhood |z − z0 | < δ

at all points of which f 0 (z) exists.

If the derivative f 0 (z) exists at all points z of a region R, then f (z) is said to be analytic
in R and is referred to as an analytic function in R or a function analytic in R. The
terms regular and holomorphic are used as synonyms for analytic.

It is important to have a criterion that will allow us to decide whether a function is

differentiable. The necessary condition for a function to be differentiable at a given point
is that, at that point, it obeys the Cauchy-Riemann conditions.

6.1 Cauchy-Riemann Conditions

Theorem

In order that a function w = f (z) = u(x, y) + iv(x, y) be analytic in a region R, the

∂u ∂u ∂v ∂v
partial derivatives , , ,
∂x ∂y ∂x ∂y
of u and v with respect to x and y should exist at every
point z ∈ R and satisfy the Cauchy-Riemann equations
∂u ∂v ∂u ∂v
= , =− (6.1)
∂x ∂y ∂y ∂x

11
This is a necessary condition that f (z) be analytic in a region R.

Proof: In order for f (z) to be analytic, the limit

f (z + ∆z) − f (z)
lim = f 0 (z)
∆z→0 ∆z
{u(x + ∆x, y + ∆y) + iv(x + ∆x, y + ∆y)} − {u(x, y) + iv(x, y)}
= lim (6.2)
∆x→0 ∆x + i∆y
∆y→0

must exist independent of the manner in which ∆z (∆x and ∆y) approaches zero.

We take the limit in two different ways, namely, along the real axis and the imaginary
axis.

1. First let ∆y = 0, ∆x → 0

(  )
u(x + ∆x, y) − u(x, y) v(x + ∆x, y) − v(x, y)
f 0 (z) = lim +i
∆x→0 ∆x ∆x
∂u ∂v
= +i (6.3)
∂x ∂x

2. ∆x = 0, ∆y → 0

(  )
u(x, y + ∆y) − u(x, y) v(x, y + ∆y) − v(x, y)
f 0 (z) = lim +i
∆y→0 i∆y i∆y
1 ∂u ∂v ∂u ∂v
= + = −i + (6.4)
i ∂y ∂y ∂y ∂y

Now f (z) cannot possibly be analytic unless these two limits are identical. Thus a neces-
sary condition that f (z) be analytic is
∂u ∂v ∂u ∂v
+i = −i + (6.5)
∂x ∂x ∂y ∂y

Equating real and imaginary parts

∂u ∂v ∂v ∂u
= =− (6.6)
∂x ∂y ∂x ∂y

We have obtained Cauchy-Riemann equations.

If the partial derivatives in (6.6) are continuous in R, then the Cauchy-Riemann equations
are sufficient conditions that f (z) be analytic in R.

12
Theorem: The necessary and sufficient condition that w = f (z) = u(x, y) + iv(x, y) be
∂u ∂v ∂u ∂v
analytic in a region R is that the Cauchy-Riemann equations ∂x
= ∂y
, ∂y
= − ∂x are
satisfied in R and the partial derivatives are continuous in R.

Theorem: If f (z) is analytic in a region R, so also are f 0 (z), f 00 (z), . . . , analytic in R.

(Proof will be discussed later).

6.2 Harmonic functions

Definition: A function f : U → R defined on a domain U ⊂ C is said to be harmonic

in U if all of its second partial derivatives with respect to x and y exist and satisfy the
2-dimensional Laplace equation

∂ 2f ∂ 2f
f ≡ + = 0. (6.7)
∂x2 ∂y 2

Example:
f (z) = |z|2 = x2 + y 2 is not harmonic in any domain of the complex plane since

∂ 2f ∂ 2f
+ = 4. (6.8)
∂x2 ∂y 2

Example:
The function f (z) = Rez 2 = x2 − y 2 is harmonic in C.

Properties of analytic functions

1. If f (z) = u(x, y) + iv(x, y) is analytic on a region R then both u and v are harmonic
functions on R.
∂u ∂v ∂v
Proof: Since ∂x
= ∂y
and ∂x
= − ∂u
∂y
, we have
∂2u ∂2v ∂2u 2
∂ v
∂x2
= ∂x∂y
and ∂y 2
= − ∂y∂x

Since an analytic function is infinitely differentiable, u and v have the required two con-
tinuous partial derivatives. This also ensures that the mixed partial derivatives agree, i.e.
∂2v ∂2v
∂x∂y
= ∂y∂x
.

Therefore,

∂ 2u ∂ 2u
+ =0 (6.9)
∂x2 ∂y 2

13
Similarly, we can show v is also harmonic function on R.

The converse is not true, however; a pair of harmonic functions does not, in general, define
a differentiable function. For example: f (z) = x + 2iy.

The real and imaginary parts of this function trivially satisfy Laplace’s equation. However,
f (z) does not have a well-defined derivative at the origin. Does not satisfy the Cauchy-
Riemann equations.

Given a harmonic function u we say that another harmonic function v is its harmonic
conjugate if the complex-valued function f = u + iv is analytic.

2. If a function is holomorphic in R, then

∂f
= 0. (6.10)
∂ z̄

Proof:

Let us consider z to be a function of x and y in R2 , z = z(x, y) and f (z) = u(x, y)+iv(x, y).
Then

1 1
x = (z + z̄), y= (z − z̄) (6.11)
2 2i

Now,
∂f ∂u ∂x ∂u ∂y ∂v ∂x ∂v ∂y
= + +i +i (6.12)
∂ z̄ ∂x ∂ z̄ ∂y ∂ z̄ ∂x ∂ z̄ ∂y ∂ z̄
From (6.11), we have
∂x 1 ∂y 1 i
= , =− = (6.13)
∂ z̄ 2 ∂ z̄ 2i 2
From (6.12), we have
∂f ∂u 1 ∂u i ∂v 1 ∂v i
= + +i +i
∂ z̄ ∂x 2 ∂y 2 ∂x 2 ∂y 2
   
1 ∂u ∂v i ∂u ∂v
= − + + (6.14)
2 ∂x ∂y 2 ∂y ∂x

Using Cauchy-Riemann equations we get

∂f
= 0. (6.15)
∂ z̄

14
Entire function: If f (z) is analytic everywhere in the (finite) complex plane, we call it
an entire function.

Remark: f 0 (z) = ∂u
∂x
∂v
+ i ∂x along with Cauchy-Riemann condition (6.6) lets us take
derivatives of functions of z operationally as we do for real variables.

Example 1:

f (z) = z 2
u(x, y) = x2 − y 2 , v(x, y) = 2xy
∂u ∂v
= 2x =
∂x ∂y
∂u ∂v
= −2y = − (6.16)
∂y ∂x
f (z) = z 2 satisfies the Cauchy-Riemann equations throughout the complex plane. Partial
derivatives are clearly continuous. f (z) = z 2 is analytic.

Example 2:

f (z) = z̄
u = x, v = −y
∂u ∂v
= 1 6= (6.17)
∂x ∂y
Cauchy-Riemann equations are not satisfied anywhere on C and f (z) = z̄ is not an analytic
function of z.

f (z) = z̄ is continuous everywhere but nowhere differentiable.

Example 3:

f (z) = x2 + y + i(y 2 − x)
∂u ∂u
= 2x, =1
∂x ∂y
∂v ∂v
= −1, = 2y (6.18)
∂x ∂y
Cauchy-Riemann equations are satisfied only along the line y = x in R2 . The function is
differentiable along this line but it is not analytic.

It is important to realise that analyticity, unlike differentiability, is not a property of a

function at a point, but on an open set of points. The reason for this is to able to eliminate
from the class of interesting functions, functions which may be differentiable at a point
(or along a line) but nowhere else.

15
6.3 Some more properties of Analytic Function

In any analytic function w = u(x, y) + iv(x, y) the families of curves given by u(x, y) = k1
and v(x, y) = k2 , where k1 and k2 are constants, are mutually orthogonal.

Each member of one family is perpendicular to each member of the other family at their
point of intersection.

Proof: Let w(x, y) be a real function of two real arguments x and y.

One has
∂w ∂w
dw = dx + dy (6.19)
∂x ∂y

→
− ∂ →
− ∂
( ∇)x = , ( ∇)y = , (6.20)
∂x ∂y

→
− ∂w →
− ∂w
( ∇w)x = , ( ∇w)y = (6.21)
∂x ∂y

→
−
The vector ∇w is called the gradient of w at a given point. dx and dy can also be
considered as components of a vector d~r

(d~r)x = dx, (d~r)y = dy (6.22)

Eq.(6.19) can be written as

→
− →
−
dw = ( ∇w)x (d~r)x + ( ∇w)y (d~r)y
→
−
= ∇w · d~r (6.23)
→
−
(usual scalar product of ∇w and d~r)

In the case when the points x, y lie on the same curve

w(x, y) = constant (6.24)

One has
→
−
dw = ( ∇w) · d~r = 0 (6.25)

→
−
Since d~r is now tangent to the curve (6.24), Eq.(6.25) shows that the gradient ∇w at a
point (x0 , y0 ) is perpendicular to the curve

w(x, y) = w(x0 , y0 ).

16
Consider now a differentiable function

f (z) = u(x, y) + iv(x, y)

→
− →
−
The gradients ∇u and ∇v are perpendicular at an arbitrary point (x0 , y0 ) to the curves

u(x, y) = u(x0 , y0 ) and v(x, y) = v(x0 , y0 ),

respectively.
→
− →
−
Furthermore, ∇u and ∇v are perpendicular to each other (using Cauchy-Riemann con-
ditions)
→
− →
− ∂u ∂v ∂u ∂v
( ∇u) · ( ∇v) = +
∂x ∂x ∂y ∂y
= 0 (6.26)

Hence, the curves

u(x, y) = u(x0 , y0 ) (6.27)

v(x, y) = v(x0 , y0 ) (6.28)

make a right angle with each other at the point (x0 , y0 ).

In other words, the tangents to the curves

Ref (z) = Ref (z0 ) (6.29)

Imf (z) = Imf (z0 ) (6.30)

are perpendicular at the point z0 if f (z) is differentiable in a neighbourhood of this point.

6.4 Polynomials and rational functions

Let f (z) be an analytic function on some open subset U ⊂ C, and let α be a complex
number. Then the function αf (z) is also analytic on U . Indeed, from the definition (5.1)
of the derivative, we see that

(αf )0 (z0 ) = αf 0 (z0 ), (6.31)

which exists whenever f 0 (z0 ) exists.

17
Let f (z) and g(z) be analytic functions on the same open subset U ⊂ C. Then the
functions f (z) + g(z) and f (z)g(z) are alos analytic. Again from the definition (5.1) of
the derivative,

(f + g)0 (z0 ) = f 0 (z0 ) + g 0 (z0 ) (6.32)

(f g)0 (z0 ) = f 0 (z0 )g(z0 ) + f (z0 )g 0 (z0 ), (6.33)

which exists whenever f 0 (z0 ) and g 0 (z0 ) exist.

Therefore, we see that finite sums and products of analytic functions are analytic with
the same domain of analyticity. In particular, sums and products of entire functions are
entire.

We can easily verify that the function f (z) = z is entire. Thus, we see that any polynomial
P (z) = N n
P
n=0 an z of finite degree N is also an entire function. Its derivative is given by
N
X
0
P (z0 ) = nan z0n−1 . (6.34)
n=1

This follows from formulae (6.32) and (6.33) for the derivatives of sums and products.

We will see that to some extent we will be able to describe all analytic functions (at
least locally) in terms of polynomials, provided that we allow the polynomials to have
arbitrarily high degree; in other words, in terms of power series.

Let f (z) and g(z) be analytic functions on some open subset U ⊂ C. If g(z0 ) 6= 0, then
from the definition of the derivative (5.1), it follows that
 0
f f 0 (z0 )g(z0 ) − f (z0 )g 0 (z0 )
(z0 ) = . (6.35)
g g(z0 )2

By a rational function we mean the ratio of two polynomials. Let P (z) and Q(z) be
two polynomials. Then the rational function
P (z)
R(z) =
Q(z)
is analytic away from the zeros of Q(z).

Finally, let g(z) be analytic in an open subset U ⊂ C and let f (z) be analytic in some
open subset containing g(U ), the image of U under g. Then the composition f ◦ g defined
by
(f ◦ g)(z) = f (g(z))
is also analytic in U . Its derivative can be computed using the chain rule,

(f ◦ g)0 (z0 ) = f 0 (g(z0 ))g 0 (z0 ) (6.36)

18
6.5 The complex exponential and related functions

Let z = x + iy be a complex variable and define the complex exponential exp(z) (also
written ez ) to be the function

exp(z) = exp(x + iy) ≡ ex (cos y + i sin y). (6.37)

We will first check that this function is entire. Decomposing it into real and imaginary
parts, we see that

u(x, y) = ex cos y and v(x, y) = ex sin y. (6.38)

It is easy to check that the Cauchy-Riemann equations (6.1) are satisfied everywhere on
the complex plane:
∂u ∂v ∂v ∂u
= ex cos y = and = ex sin y = − . (6.39)
∂x ∂y ∂x ∂y
Therefore, the function is entire and its derivative is given by
∂u ∂v
exp0 (z) = +i
∂x ∂x
= e cos y + iex sin y
x

= exp(z). (6.40)

The exponential function obeys the following addition property

exp(z1 + z2 ) = exp(z1 ) exp(z2 ) , (6.41)

which has as a consequence the celebrated De Moivre’s Formula:

(cos θ + i sin θ)n = cos(nθ) + i sin(nθ), (6.42)

obtained simply by noticing that exp(inθ) = exp(iθ)n .

The exponential is also a periodic function, with period 2πi. In fact, from the periodicity of
trigonometric functions, we see that exp(2πi) = 1 and hence, using the addition property
(6.41), we find

exp(z + 2πi) = exp(z). (6.43)

This means that the exponential is not one-to-one, in sharp contrast with the real expo-
nential function. It follows from the definition of the exponential function that

exp(z1 ) = exp(z2 ) if and only if z1 = z2 + 2πik for some integer k.

19
We can divide up the complex plane into horizontal strips of height 2π in such a way
that in each strip the exponential function is one-to-one. To see this define the following
subsets of the complex plane

Sk ≡ {x + iy ∈ C | (2k − 1)π < y ≤ (2k + 1)π},

for k = 0, ±1, ±2, . . . , as shown in the figure below.

S2
3π

S1
π

S0

−π
S−1

−3π
S−2

Figure 6.1: Fundamental regions of the complex exponential function

Then it follows that if z1 and z2 belong to the same set Sk , then exp(z1 ) = exp(z2 ) implies
that z1 = z2 . Each of the set Sk is known as a fundamental region for the exponential
function. The periodicity of the complex exponential will have as a consequence that the
complex logarithm will not be single-valued.

6.6 Complex trigonometric functions

We can also define complex trigonometric functions starting from the complex exponential.
Let z = x+iy be a complex variable. Then we define the complex sine and cosine functions
as
eiz − e−iz eiz + e−iz
sin(z) ≡ and cos(z) ≡ (6.44)
2i 2

20
Being linear combinations of the entire functions exp(±iz), sine and cosine functions are
entire. Their derivatives are

sin0 (z) = cos(z) and cos0 (z) = − sin(z). (6.45)

The complex trigonometric functions obey many of the same properties of the real sine
and cosine functions. For example,

cos2 (z) + sin2 (z) = 1,

and they are periodic with period 2π. However, there is one important difference. The
complex sine and cosine functions are not bounded whereas the real sine and cosine
functions are bounded. We can see this by breaking the complex sine and cosine functions
into real and imaginary parts:

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

cos(x + iy) = cos x cosh y − i sin x sinh y. (6.46)

Since on the right hand side hyperbolic functions appear, this means that the complex
sine and cosine functions are not bounded.

Finally, let us define the complex hyperbolic functions. If z = x + iy, then let

ez − e−z ez + e−z
sinh(z) ≡ and cosh(z) ≡ .
2 2

In contrast with the real hyperbolic functions, they are not independent from the trigono-
metric functions.

sinh(iz) = i sin(z) and cosh(iz) = cos(z). (6.47)

In a similar way we can also define other complex trigonometric functions, namely,
tan(z), cot(z), sec(z) and csc(z) in the usual way, as well as their hyperbolic counter-
parts.

L’HOSPITAL’S RULE
Let f (z) and g(z) be analytic in a region containing the point z0 and suppose that f (z0 ) =
g(z0 ) = 0 but g 0 (z0 ) 6= 0.. Then L’Hospital’s rule states that

f (z) f 0 (z0 )
lim = 0 . (6.48)
z→z0 g(z) g (z0 )

21
 Home assignment 10
(For your practice only)

Using the ( − δ) criterion discussed in the notes

Q.1 Show
lim Rez = Rez0
z→z0

Q.2 Show
lim z̄ = z̄0
z→z0

Q. 3 Show
lim (az + b) = az0 + b,
z→z0

where a and b are complex constants. [First consider a = 0 and then consider a 6= 0]

Q.4 Show
lim [x + i(2x + y)] = 1 + i
z→1−i

[Hint: Let  > 0. Assume z = x + iy and |z − (1 − i)| < δ. Then |x − 1| < δ (real part)
and |y + 1| < δ (imaginary part). ]

Q.5 Prove that 

z 2 z 6= z0
f (z) =
0 z = z0 where z0 6= 0
is discontinuous at z = z0 . What happens when z0 = 0?

Q.6 Suppose u(x, y) = xy − x + y. Find a function v such that f (z) = u + iv is analytic.

Express f (z) in terms of z.

Q.7 (a) Prove that u(x, y) = x3 − 3xy 2 is harmonic. (b) Find v such that f (z) = u + iv
is analytic.

7 The complex logarithm

Now we will discuss the logarithm of a complex number. We will use the notation ‘log’
for the natural logarithm. One could also use the notation ‘ln’.

We define the complex logarithm as an inverse to the complex exponential function.

This means a logarithm of a nonzero complex number z, is any complex number w such

22
that exp(w) = z. Therefore, we define the function log(z) by

w = log(z) if exp(w) = z. (7.1)

We already know
exp(z + 2πi) = exp(z).

Therefore, if w = log(z) so is w + 2πik for any integer k.

=⇒ log(z) is a multiple-valued function. The imaginary part of the log(z) function is

arg(z). To see this, let us write

z = |z| exp(i arg(z)) and w = log(z) = u + iv.

Now, we have
exp(u + iv) = eu eiv = |z|ei arg(z) .

Comparing we get
eu = |z| and eiv = ei arg(z)

Since u is a real number and |z| is a positive real number, the first equation for u can be
uniquely solved using the real logarithmic function:

u = Log|z|,

where we have written real logarithm as Log, in order to distinguish it from the complex
function log(z). Similarly, the second equation gives v = arg(z). Therefore, we can write

log(z) = Log|z| + i arg(z). (7.2)

Thus we see that the complex logarithm is a multiple-valued function because arg(z) is a
multiple-valued function. In terms of the principal value Arg(z) of the argument function,
we can also write the log(z) as follows:

log(z) = Log|z| + iArg(z) + 2πik for k = 0, ±1, ±2, . . . , (7.3)

This manifestly shows that the complex logarithm is a multiple-valued function.

For example, whereas the real logarithm of 1 is simply 0, the complex logarithm is given
by
log(1) = Log|1| + i arg(1) = 0 + i2πk for any integer k.

23
The real logarithm function is only defined for positive real numbers. The complex loga-
rithm is defined for all nonzero complex numbers. For example,

log(−1) = Log| − 1| + i arg(−1) = 0 + iπ + i2πk for any integer k.

The complex logarithm obeys many of the algebraic identities that we expect from the
real logarithm.

The identity

log(z1 z2 ) = log(z1 ) + log(z2 ), (7.4)

holds up to integer multiples of 2πi, i.e., modulo 2πi.

log(z1 z2 ) − log(z1 ) − log(z2 ) = 2πik for some integer k.

Similarly, we have

log(z1 /z2 ) = log(z1 ) − log(z2 ), (7.5)

in the same sense as before, for any two nonzero complex numbers z1 and z2 .

7.1 Choosing a branch for the logarithm

Let us now discuss the analyticity properties of the complex logarithm function and for
this we need to take its derivative at some point z0 as in (5.1):
log(z0 + ∆z) − log(z0 )
log0 (z0 ) = lim .
∆z→0 ∆z
However, we immediately encounter a problem in doing this. This is because the function
log(z) is multiple-valued and so we have to make sure that the two log functions in the
numerator tend to the same value in the limit, otherwise the limit will not exist. In
other words, we have to choose one of the infinitely many values for the log function in a
consistent way.

This way of restricting the values of a multiple-valued function to make it single-valued

in some region (in the above example in some neighbourhood of z0 ) is called choosing
a branch of the function. For example, we define the principal branch Log of the
logarithm function to be
Log(z) = Log|z| + iArg(z),

where Arg(z) is the principal value of arg(z).

24
Although we are using Log both for real logarithm and the principal branch of the complex
logarithm, the notation is consistent. If z is a positive real number, then z = |z| and
Arg(z) = 0, whence Log(z) = Log|z|.

The function Log(z) is single-valued but it is not continuous in the whole complex plane,
since Arg(z) is not continuous in the whole complex plane. Let z± = −x ± i where x
and  are positive numbers. In the limit  → 0, z+ and z− tend to the same point on
the negative real axis from the upper and lower half-planes respectively. Hence, whereas
lim→0 z± = −x, the principal value of the logarithm obeys

lim Log(z± ) = Log(x) ± iπ,

→0

so that it is not a continuous function anywhere on the negative real axis, or at the origin,
where the function itself is not defined. The non-positive real axis is known as a branch
cut for this function and the origin is known as a branch point.

Let D denote all the points in the complex plane except for those which are real and
non-positive. In other words, D is the complement of the non-positive real axis. D is
an open subset of the complex plane and by construction, Log(z) is single-valued and
continuous for all points in D.

Let z0 ∈ D be any point in D and consider w0 = Log(z0 ). Let ∆z = z −z0 . The derivative
of w = Log(z) at z0 can be written as
w − w0
Log0 (z0 ) = lim
z→z0 z − z0
1
= lim
z→z0 z−z0
w−w0
1
= lim z−z0 (7.6)
w→w0
w−w0

To reach the third line of the above equation we used the continuity of Log(z) in D,

25
which means w → w0 as z → z0 . Remember that w = Log(z) =⇒ exp(w) = z and the
exponential function is single-valued.

Now using z = ew we see that

1 1 1 1
Log0 (z0 ) = lim w w0 = = = . (7.7)
w→w0 e −e exp0 (w 0) exp(w0 ) z0
w−w0

Since this is well-defined everywhere but for z0 = 0, which does not belong to D, we see
that Log(z) is analytic in D.

D0 Dτ

Figure 7.2: Two other branches of the complex logarithm.

We can also choose a different branch for the argument function. For example, let us
consider the function Arg0 (z)
0 ≤ Arg0 (z) < 2π.
This is another popular choice. This function, like Arg(z), is single-valued but discontin-
uous. However, the discontinuity is now along the positive real axis, since approaching a
positive real number from the upper half-plane we would see that its argument tends to
0 whereas approaching it from the lower half-plane the argument would tend to 2π. We
can, therefore, define a branch Log0 (z) of the logarithm by

Log0 (z) = Log|z| + iArg0 (z).

This branch then has a branch cut along the non-negative real axis but it is continuous in
its complement D0 as shown in the figure 7.2. The same argument as before shows that
Log0 (z) is analytic in D0 with derivative given by
1
Log)00 (z0 ) = for all z0 in D0 .
z0

There are also many other branches. For example, let τ be any real number and define
the branch Argτ (z) of the argument function to take values

τ ≤ Argτ (z) < τ + 2π.

26
This gives rise to a branch logτ (z) of the logarithm function defined by

Logτ (z) = Log|z| + iArgτ (z),

which has a branch cut starting from the origin and consisting of all those points z with
arg(z) = τ modulo 2π. Logτ (z) is analytic everywhere on the complement Dτ of the
branch cut, as shown in figure 7.2. Its derivative is
1
Log0τ (z0 ) = for all z0 in Dτ .
z0

The choice of branch is immaterial for many properties of the complex logarithm, although
it is important that a choice be made. Different applications may require choosing one
branch over another. Provided one is consistent this should not cause any problems.

As an example, let us compute the derivative of the function f (z) = log(z 2 + 2iz + 2)
at the point z = i. We need to choose a branch of the logarithm, which is analytic in a
region containing a neighbourhood of the point i2 + 2i i + 2 = −1. The principal branch is
not analytic there, so we have to choose another branch. Suppose that we choose Log0 (z).
Then, by the chain rule

2z + 2i 2i + 2i
f 0 (i) = = = −4i.
z2 + 2iz + 2 i2 + 2i2 + 2
z=i

Any other valid branch would give the same result.

7.2 Complex powers

Let α be a complex number. Then for all z 6= 0, we define the α-th power z α of z by

z α ≡ eα log(z) = eαLog|z|+iα arg(z) . (7.8)

Rewriting the above expression, we get

z α = eαLog|z|+iαArg(z)+iα2πk = eαLog(z) eiα2πk , (7.9)

for k = 0, ±1, ±2, . . . .

Depending on α we will have either one, finitely many or infinitely many values of
exp(i2παk). Suppose that α is real. If α = n is an integer then so is αk = nk and
exp(i2παk) = exp(i2πnk) = 1.

=⇒ only one value for z n .

27
 p
If α = q
is a rational number, where the integers p and q are coprime (have no com-
mon factors), then z p/q will have a finite number of values. To see this let us consider
exp(i2πkp/q) as k ranges over the integers.

ei2π(k+q)p/q = ei2π(k(p/q)+p)
= ei2πk(p/q)+i2πp = ei2πkp/q (7.10)

=⇒ ei2πkp/q takes the same values for k and k + q.

=⇒ z p/q will have q distinct values, say, k = 0, 1, 2, . . . q − 1.

For example, let us compute 11/q . According to (7.9),

11/q = eLog(1)/q ei2π(k/q) = ei2π(k/q) ,

where k ranges over the integers.

Only the values k = 0, 1, 2, . . . , q − 1 will be different (i.e., a total of q number of values).

The values of 11/q are known as q-th roots of unity.

Let ω = exp(i2π/q) correspond to the k = 1 value of 11/q . Then the q-th roots of unity
are 1, ω, ω 2 , . . . , ω q−1 , and there are q of them. The q-th roots of unity lie in the unit circle
|z| = 1 in the complex plane and define the vertices of a regular q-gon.

In the above figure we show the q-th roots of unity for q = 3, 5.

Let z be a nonzero complex number and we want to find its q-th roots. Writing z as
z = |z| exp(iθ), we have

z 1/q = |z|1/q eiθ/q ω k for k = 0, 1, 2, . . . , q − 1.

If p is any integer, we can then take the p-th power of the above formula:

z p/q = |z|p/q eipθ/q ω pk for k = 0, 1, 2, . . . , q − 1.

If p and q are coprime, the ω pk for k = 0, 1, 2, . . . , q − 1 are different. To see this, suppose
0 0
that ω pk = ω pk , for k and k 0 between 0 and q − 1. Then ω p(k−k ) = 1, which means that

28
p(k − k 0 ) has to be a multiple of q. Because p and q are coprime, this can only happen
when k = k 0 . Therefore, we see that indeed a rational power p/q (with p and q coprime)
of a complex number has precisely q values.

Let us now consider complex powers. If α = a + ib is complex, then z α will always have
an infinite number of values.

eiα2πk = ei(a+ib)2πk = ei2πka e−2πkb .

The last term of the above expression takes a different value for each integer k.

For example, let us compute ii . By definition,

ii = ei log(i) = ei(Log(i)+i2πk) = ei(iπ/2+i2πk) = e−π/2 e−2πk

for k = 0, ±1, ±2, . . . , which is real.

7.2.1 Choosing a branch for the complex power

Every branch of the logarithm gives rise to a branch of z α . In particular we define the
principal branch of z α to be exp(αLog(z)). Since the exponential function is entire, the
principal branch of z α is analytic in the domain D where Log(z) is analytic. Its derivative
can be computed for any point z0 in D using the chain rule (6.36):
d αLog(z)
α
e = eαLog(z0 ) .
dz z=z0 z0
Given any nonzero z0 in the complex plane, we can choose a branch of the logarithm so
that the function z α is analytic in a neighbourhood of z0 . We can compute its derivative
there and we see that the following equation holds
d α 1
(z ) z=z0 = αz0α , (7.11)
dz z0
α
provided that we use the same branch of z on both sides of the equation.

Let us finally note that the function z z is analytic wherever the chosen branch of the
logarithm function is defined. z z = exp(z log(z)) and its principal branch is defined to be
the function exp(zLog(z)). This is analytic in D. Taking the derivative we see that
d zLog(z)

e = ez0 Log(z0 ) (Log(z0 ) + 1), (7.12)
dz z=z0
which exists everywhere on D. Again a similar result holds for any other branch provided
we are consistent and take the same branches of the logarithm in both sides of the following
equation:
d z
(z ) = z0z0 (log(z0 ) + 1). (7.13)
dz z=z0

29
 Home assignment 11

Q.1 Find all distinct values of z for which z 5 = −32 and locate these values in the complex
plane.

Q.2 Find all the seventh roots of unity.

√
Q.3 Find each of the indicated roots: (a) (−1+i)1/3 , (b) (−2 3−2i)1/4 , (c) (−15−8i)1/2 ,
(d) cube roots of 8.

Q.4 Using the chain rule (6.36) and the result in (6.40), show that

exp0 (az) = a exp(az).

Q.5 Using the results of Q. 4 and the definitions (6.44), show that

sin0 (z) = cos(z) and cos0 (z) = − sin(z).

Now with (6.46) verify that the Cauchy-Riemann equations are satisfied. Using
∂u ∂v
f 0 (z) = +i
∂x ∂x
again show that
sin0 (z) = cos(z) and cos0 (z) = − sin(z).

Q.6 Determine the values of log(1 − i) using (7.3). What is the value corresponding to
the principal branch ?

8 Complex Integration

In real analysis differentiation and integration are roughly speaking inverse operations.
We will see that something similar also happens in the complex domain, but in addition,
and this is unique to complex analytic functions, differentiation and integration are also
roughly equivalent operations, in the sense that we will be able to take derivatives by
performing integrals.

Let us discuss the integration of complex-valued functions of a complex variable. We want

to perform an integration of the type
Z z1
f (z)dz,
z0

30
where z0 and z1 are complex numbers. We are immediately faced with a difficulty. Unlike
the case of an interval [a, b] on the real line, where it is obvious how to go from a to b,
here z0 and z1 are points in the complex plane and there are many ways to go from one
point to the other. Therefore, the above integral is ambiguous. The way out is to specify
a path joining z0 to z1 and then integrate the function along the path.

Parametrised curve

Let z0 and z1 be two points in the complex plane. A parametrised curve joining z0
and z1 can be defined by a continuous function z(t) taking points t in the interval [t0 , t1 ]
to points z(t) in the complex plane in such a way that z(t0 ) = z0 and z(t1 ) = z1 .

z1

z0

We can decompose z into its real and imaginary parts, and this is equivalent to two
continuous real valued functions x(t) and y(t) defined on the interval [t0 , t1 ] such that
x(t0 ) = x0 and x(t1 ) = x1 and similarly for y(t) : y(t0 ) = y0 and y(t1 ) = y1 , where
dz(t)
z0 = x0 + iy0 and z1 = x1 + iy1 . We say that the curve is smooth if ż(t) = dt
is a
continuous function [t0 , t1 ] → C, which is never zero.

In other words, if x(t) and y(t) are real functions of the real variable t assumed continuous
in t0 ≤ t ≤ t1 , the parametric equation

z(t) = x(t) + iy(t) t0 ≤ t ≤ t1

defines a continuous curve or arc in the z-plane joining points z0 = z(t0 ) and z1 = z(t1 ).

For example, suppose we consider

z(t) = cos(t) + i sin(t) (0 ≤ t ≤ π/2)

1 − t2 2t
≡ 2
+i (0 ≤ t ≤ 1). (8.1)
1+t 1 + t2

31
y

These curves have certain specific properties.

(i) For a particular value of t, z(t) has a unique value, i.e., if z(t0 ) = z(t00 ) then t0 = t00 ,
which implies that the curve is simple. A simple curve never crosses itself.

(ii) The curves are smooth, differentiable curves.

z(t), t0 ≤ t ≤ t1 =⇒ This represents a part of the curve z(t) in the interval [t0 , t1 ].

If t0 6= t1 while z(t0 ) = z(t1 ), i.e., z0 = z1 , the endpoints coincide and the curve is said to
be closed.
y

−→ a simple closed curve

A closed curve which does not intersect itself anywhere is called a simple closed curve.

A curve which is composed of a finite number of smooth arcs is called a piecewise or

sectionally smooth curve or a contour.

32
y

−→ not a simple closed curve

8.1 Complex Integration

Let f (z) be a complex-valued function of the complex variable z = x + iy and let us

consider a smooth curve C in the complex plane. Let f (z) be continuous at all points of
C and let us assume C has a finite length. C has end points a and b. C may be described
by a parametric equation

z(t) = x(t) + iy(t) ta ≤ t ≤ tb .

t is a real parameter and x(t), y(t) are real single-valued functions that have continuous
derivatives.

Subdivide C into n parts by introducing n + 1 points

z0 , z1 , z2 , . . . , zn−1 , zn ,

chosen arbitrarily and a = z0 , b = zn .

On each arc joining zk−1 to zk [where k goes from 1 to n ] choose a point ξk .

ξk
ξn
C zk−1 zk b
zn−1

ξ2 z
2

ξ1
z1
a
x

33
We now form the sum

Sn = f (ξ1 )(z1 − a) + f (ξ2 )(z2 − z1 ) + · · · + f (ξn )(b − zn−1 )

Let zk − zk−1 = ∆zk . Therefore,

n
X
Sn = f (ξk )(zk − zk−1 )
k=1
Xn
= f (ξk )∆zk (8.2)
k=1

Take the limit n → ∞ in such a way that the largest of the chord lengths |∆zk | → 0.

This limit, provided it exists and provided it does not depend on the way we have chosen
zj and ξj , is called the complex line integral of f (z) along curve C. It is written as
Z Z b
I= f (z)dz or I = f (z)dz.
C a

In such case f (z) is said to be integrable along C. This is also called contour integral.

Separating f (z) and z into their real and imaginary parts, we can also write the sum
n 
X 
 
 
Sn = u(ξk )(xk − xk−1 ) − v(ξk )(yk − yk−1 ) + i v(ξk )(xk − xk−1 ) + u(ξk )(yk − yk−1 ) .
k=1

|∆zk | → 0 implies

|xk − xk−1 | → 0, |yk − yk−1 | → 0, for all k.

Therefore, I can be expressed as

Z Z
I = (udx − vdy) + i (vdx + udy).
C C

Assuming for definiteness that t increases from an initial value ta to a final value tb as we
go along C from a to b, we can write
Z tb   Z tb  
dx dy dx dy
I= u −v dt + i v +u dt.
ta dt dt ta dt dt

Since
dz(t) dx(t) dy(t)
= +i ,
dt dt dt

34
we have
Z tb
dz
I = (u + iv)
dt
ta dt
Z tb
dz(t)
= f (z(t)) dt. (8.3)
ta dt
Note that we have assumed that f (z) is continuous and C is a smooth curve, therefore
the integrand is itself continuous and hence the integral exists.

Assume in a region of the complex plane, which includes the contour C that the function
f (z) can be expressed as the derivative of another function F (z). The function F (z) is
called the primitive function of f (z)
dF (z)
f (z) = .
dz
Notice that F (z) is therefore analytic in the region.

Then for a z situated on the contour

dz dF (z) dz dF (z(t))
f (z) = = . (8.4)
dt dz dt dt

Inserting Eq.(8.4) in Eq.(8.3), we get

Z Z tb
dF [z(t)]
f (z)dz = dt = F (b) − F (a). (8.5)
C ta dt

Eq.(8.5) forms the content of the so-called fundamental theorem of integral calculus.

Let us consider an example. Consider the function f (z) = x2 + iy 2 integrated along the
smooth curve parametrised by z(t) = t + it for 0 ≤ t ≤ 1. This is the straight line segment
joining the origin and the point 1 + i as shown in the figure below.

1+i

Decomposing z(t) = x(t) + iy(t), we see that x(t) = y(t) = t. Therefore, f (z(t)) = t2 + it2
and ż(t) = 1+i and using complex linearity of the integral and performing the elementary

35
real integral we find the following result
1 1 1
t3
Z Z Z
2 2 2i
f (z)dz = (t + it )(1 + i)dt = (1 + i)2 t2 dt = 2i = .
C 0 0 3 0 3

Another example: consider the function f (z) = 1/z integrated along the smooth curve C
parametrised by z(t) = R exp(i 2π t) for 0 ≤ t ≤ 1, where R 6= 0. As shown in the figure,
the resulting curve is the circle of radius R centred about the origin.

Here f (z(t)) = (1/R) exp(−i 2π t) and ż(t) = 2π i R exp(i 2π t). Therefore, we obtain

1 1
2πi Rei2πt
Z Z Z
f (z)dz = dt = 2πi dt = 2πi. (8.6)
C 0 Rei2πt 0

Notice that the result is independent of the radius.

Some further properties of the integrals along a curve

R
Let us discuss some general properties of the integrals C f (z)dz.

(i) The first important property is that the integral is complex linear. That is, if α and β
are complex numbers and f and g are functions, which are continuous on C, then
Z Z Z
(αf (z) + βg(z))dz = α f (z)dz + β g(z)dz.
C C C

R
(ii) The integral C
f (z)dz does not depend on the actual parametrisation of the curve C
(Reparametrisation Invariance - I will not discuss the proof here. ) It can be shown
that because of reparametrisation invariance, we can always parametrise a curve in such
a way that the initial t = 0 and the final t = 1.

(iii) Let us notice that parametrised curves C (sometimes I will use Γ in place of C) have a
natural notion of direction: this is the direction in which we traverse the curve. Choosing

36
a parametrisation z(t) for 0 ≤ t ≤ 1, as we go from z(0) to z(1), we trace the points in
the curve in a given order, which we depict by an arrowhead on the curve indicating the
direction along which t increases, as in the curves shown earlier. A curve with such a
choice of direction is said to be directed.

Given any directed curve C, we let −C denote the directed curve with the opposite
direction; that is, with the arrow pointing in the opposite direction. A very interesting
R
property of the integral C f (z)dz is that
Z Z
f (z)dz = − f (z)dz. (8.7)
−C C

To prove this, let z(t) for 0 ≤ t ≤ 1 is a parametrisation for C. Then z 0 (t) = z(1−t) for 0 ≤
t ≤ 1 is a parametrisation for −C. Indeed, we can see that z 0 (0) = z(1) and z 0 (1) = z(0)
and they trace the same set of points. Let us compute
Z Z 1
f (z)dz = f (z 0 (t))ż 0 (t)dt
−C 0
Z 1
= − f (z(1 − t))ż(1 − t)dt
0
Z 0
= f (z(t0 ))ż(t0 )dt0
1
Z 1
= − f (z(t0 ))ż(t0 )dt0
Z0
= − f (z)dz. (8.8)
C

A useful estimate of the complex integral - the Darboux’s inequality / ML

inequality

It is useful to consider an estimate of the upper bound of a complex integral along a

parametrised curve.

We can show that if f (z) is integrable over a curve C having finite length L, and if there
exists a positive number M such that |f (z)| ≤ M on C, then
Z
f (z)dz ≤ M L
C

R
Proof: As discussed in Section 8.1, the integral C f (z)dz is the limit, as n → ∞, of the
sum nk=1 f (ξk )∆zk [see (8.1)]
P

Z Xn
f (z)dz = lim f (ξk )∆zk
C n→∞
k=1

37
Now,
n
X n
X
f (ξk )∆zk ≤ |f (ξk )||∆zk |
k=1 k=1
n
X
≤ max|f (z)| |∆zk |
k=1

≤ ML

where max|f (z)| = M is the maximum modulus of f (z) on C. |f (z)| ≤ M for all points
z on C and that nk=1 |∆zk | is not greater than the arc length L of the curve C.
P

In particular, taking n → ∞ on both sides we get

Z
f (z)dz ≤ max|f (z)| · L (8.9)
C

This is known as Darboux’s inequality / ML inequality. Results of this type are very
important for analysis and we will use this particular result in our subsequent discussions.

Let us also see the following. If Γ is a curve parametrised by z(t) = x(t) + iy(t) for t ∈
[t0 , t1 ], consider the real integral

Z Z t1
|dz| ≡ |ż(t)|dt
Γ t
Z 0t1 p
= ẋ(t)2 + ẏ(t)2 dt,
t0

p
which is the integral of the infinitesimal line element dx2 + dy 2 along the curve. There-
fore, the integral is the arc length L of the curve Γ:
Z
|dz| = L (8.10)
Γ

Piecewise smooth curves and contour integrals

R
Finally we have to generalise the integral C f (z)dz to curves which are not necessarily
smooth but which are made out of smooth curves. Curves can be composed: if C1 is a
curve joining z0 to z1 and C2 is a curve joining z1 to z2 , then we can make a curve C
joining z0 to z2 by first going to the intermediate point z1 via C1 and then from there via
C2 to our destination z2 . The resulting curve C is still continuous but it will generally fail

38
 z2
C1 C2
z0 z1

to be smooth, since ż(t) need not be continuous at the intermediate point z1 , as shown in
the figure.

However, such a curve is piecewise smooth. This means that it is made out of smooth
components by the composition procedure mentioned above. In terms of parametrisations,
if z1 (t) and z2 (t), for 0 ≤ t ≤ 1, are smooth parametrisations for C1 and C2 respectively,
then 
1
z1 (2t) for 0 ≤ t ≤ 2
z(t) =
z (2t − 1) 1
2 for 2
≤t≤1
1
is a parametrisation for C. Note that it is well-defined and continuous at t = 2
pre-
cisely because z1 (1) = z2 (0). However, it need not be smooth there since ż1 (1) 6= ż2 (0)
necessarily.

We can construct curves which are made out of a finite number of smooth curves: one
curve ending where the next curve starts. Such a piecewise smooth curve will be called
a contour. If a contour C is made out of composing a finite number of smooth curves
{Cj } we will say that each Cj is a smooth component of C.

Let C be a contour with n smooth components {Cj } for j = 1, 2, . . . , n. If f (z) is a

function continuous on C, then the contour integral of f (z) along C is defined as
Z n Z
X Z Z Z
f (z)dz = f (z)dz = f (z)dz + f (z)dz + · · · + f (z)dz,
C j=1 Cj C1 C2 Cn

R
with each of the Ci
f (z)dz is defined by (8.2).

8.2 Path independence

Connected set: We will say that an open subset U of the complex plane is connected,
if every pair of points in U can be joined by a contour.

We have seen that when the contour of integration lies in a region R where f (z) possesses
a primitive function F (z), the value of the contour integral is determined by the values of
that primitive function at the end points of the path of integration. It is independent of
the choice of the contour. F (z) is also said to be the antiderivative of f (z) in region R.

39
In the general case, let C is a contour with smooth components {Cj } for j = 1, 2, . . . , n.
The curve C1 starts in z0 and ends in some intermediate point τ1 , C2 starts in τ1 and ends
in a second intermediate point τ2 , and so on until Cn which starts in the intermediate
point τn−1 and ends in z1 . Then
Z n Z
X
f (z)dz = f (z)dz
C j=1 Cj
Z Z Z
= f (z)dz + f (z)dz + · · · + f (z)dz
C1 C2 Cn
= F (τ1 ) − F (z0 ) + F (τ2 ) − F (τ1 ) + · · · + F (z1 ) − F (τn−1 )
= F (z1 ) − F (z0 ). (8.11)

Path independence can also be rephrased in terms of closed contour integrals. We say that
a contour is closed if its endpoints coincide. The contour integral along a closed contour
H
C in the anti-clockwise direction is denoted as C whereas in the clockwise direction it is
H
denoted as −C .

As a corollary of the above result, we see that if Γ is a closed contour in some domain D
and f : D → C has an antiderivative in D, then
I
f (z)dz = 0.
Γ

This is clear because if the endpoints coincide, so that z0 = z1 , then F (z1 ) − F (z0 ) = 0.
(Note here I have used the notation Γ for the contour.)

In general, an arbitrary function of a complex variable will not possess a single-valued

primitive function and the result of the contour integration will almost invariably depend
upon the path linking the two endpoints.

However, Cauchy’s integral theorem will tell us that, under some conditions, analytic
functions always possess single-valued primitive functions, so that the results of integration
is independent of the choice of path. These conditions refer to the topology of the domain,
so we have to first introduce a little bit of notation.

8.3 Simply and multiply connected regions

Let us first remind ourselves that a contour is simple if it has no self-intersections. We

define a loop to be a closed simple contour. We start by mentioning the celebrated
Jordan curve lemma, a version of which states that any loop in the complex plane

40
separates the plane into two domains with the loop as common boundary: one of which is
bounded and is called the interior and the other is unbounded and is called the exterior.

A domain D (or a region R) is simply-connected if the interior domain of every loop

in D lies completely in D. Stated differently, a domain is simply-connected if it is such
that all closed paths within it contain only points that belong to the domain. Hence, for
example, a disk is simply-connected, while a punctured disk is not. Any circle around the
puncture contains the puncture in its interior, but this has been excised from the disk.

In other words, a domain D (or a region R) is simply-connected if any loop in the domain
can be continuously shrunk to a point without any point of the loop ever leaving the
domain.
y

|z| = 2
Γ

Example: Suppose D is the domain defined by |z| < 2. If Γ is any simple closed curve
lying in D [i.e., whose points are in D] we see that it can be shrunk to a point which lies
in D, and thus does not leave D, so that D is simply-connected.

|z| = 2 Γ
1
=
|z |
x

On the other hand, if D is the domain defined by 1 < |z| < 2, then there is a simple

41
closed curve Γ lying in D, which cannot be shrunk to a point without leaving D. In this
case, D is multiply-connected.

9 Cauchy’s Integral Theorem

Let f (z) be an analytic function in a simply-connected region R, then for any loop Γ in
that region, the contour integral vanishes:
I
f (z)dz = 0.
Γ

It was first proved by use of Green’s Theorem which is valid in the complex plane, with
the added restriction that f 0 (z) be continuous in R. However, Goursat gave a proof which
removed this restriction. This is called Cauchy-Goursat Theorem.

Proof: We shall prove the Cauchy Integral Theorem which requires the hypothesis that
f 0 (z) be continuous in R.

The proof uses Green’s theorem which states that if P (x, y) and Q(x, y) be real functions
of x and y, continuous and have continuous partial derivatives in a simply-connected
region R in the complex plane, and if Γ is any positively oriented loop in R, then
I ZZ  
∂Q ∂P
(P (x, y)dx + Q(x, y)dy) = − dxdy, (9.1)
Γ ∂x ∂y
Int(Γ)

where Int(Γ) is the interior of the loop Γ.

Since f (z) = u + iv is analytic and has a continuous derivative

∂u ∂v ∂v ∂u
f 0 (z) = +i = −i ,
∂x ∂x ∂y ∂y
it follows that the partial derivatives
∂u ∂v ∂v ∂u
(1) = (2) =−
∂x ∂y ∂x ∂y
are continuous inside and on Γ.

Thus Green’s theorem can be applied and we have

I I I
f (z)dz = (u(x, y)dx − v(x, y)dy) + i (v(x, y)dx + u(x, y)dy)
Γ Γ
ZZ   Z ZΓ  
∂v ∂u ∂u ∂v
= − − dxdy + i − dxdy
∂x ∂y ∂x ∂y
Int(Γ) Int(Γ)

= 0 Using Cauchy-Riemann equations. (9.2)

42
Cauchy-Goursat Theorem removes the restrictions that f 0 (z) be continuous.

A consequence of Cauchy’s Integral Theorem

Suppose that Γ1 and Γ2 are two contours in a region D that have the same initial and
final endpoints, say z0 and z1 , respectively. Suppose further that f (z) is analytic in the
region enclosed by Γ1 and Γ2 .

Γ1 + (−Γ2 ) is a closed contour.

Γ2
Γ1

Therefore, using Cauchy’s theorem

Z Z Z
0= f (z)dz = f (z)dz + f (z)dz
Γ1 +(−Γ2 ) Γ1 −Γ2
Z Z
= f (z)dz − f (z)dz.
Γ1 Γ2

Hence, Z Z
f (z)dz = f (z)dz.
Γ1 Γ2

Path independence of integration of analytic functions.

43
 Home assignment 12

Q.1 Evaluate Z Z
f (z)dz = z 2 dz
Γ Γ

where Γ ≡ Γ1 (OB) or Γ2 (OA + AB), as shown in the figure and verify path indepen-
dence of the integration.

B(1, 1)
Γ1
Γ2

O Γ2 A(1, 0) x

[Hint: Write z 2 as the derivative of the primitive function and do the integration.]

Q.2 Evaluate
Z (2,4)
(2y + x2 )dx + (3x − y)dy
(0,3)

along
 
2 33
a) x = 2t, y = t + 3 Ans:
2
 
44 5 103
b) straight lines from (0,3) to (2,3) and then (2,3) to (2,4) Ans: + =
3 2 6
 
97
c) straight line from (0,3) to (2,4) Ans: .
6

Q.3 Evaluate Z
z̄ dz
Γ

from z = 0 to z = 4 + 2i along the curve Γ given by

 
2
a) z = t + it Ans: 10 − 8i/3
 
b) Line from z = 0 to z = 2i and then line from z = 2i to z = 4 + 2i Ans: 2 + (8 − 8i).

44
Consequences of Cauchy Integral Theorem

The Cauchy integral Theorem has a very important consequence for the computation of
contour integrals. Let f (z) be analytic in a region bounded by two simple closed curves C1
and C2 (where C2 lies inside C1 ) and on these curves (see left panel of the figure below).
Then I I
f (z)dz = f (z)dz
C1 C2

P
C1 C1

C2 −C2

Proof: In the right panel of the figure above, consider the total contour C consisting of
C1 , P R, −C2 , RP . Since the region bounded by this total contour C is simply connected
and f (z) is analytic in that region and also on its boundary C, we have by Cauchy’s
Integral Theorem
I
f (z)dz = 0. (9.3)
C

However, the integration along RP cancels the integration along P R. Therefore, we are
left with only integration along C1 and −C2 and thus from (9.2), we get
I I
f (z)dz + f (z)dz = 0.
C1 −C2
I I
=⇒ f (z)dz = f (z)dz.
C1 C2

Deforming the contour without changing the integral

The Cauchy integral Theorem says that contour can be deformed without changing the
result of the integral, provided that in doing so we never cross a point where the integral
ceases to be analytic. Let us discuss an example.

Let us compute the contour integral

I
1
dz,
E z

45
 E

where E is the positively-oriented ellipse x2 + 4y 2 = 1 as shown in the figure.

Earlier we computed the same integral around a circular contour C of radius 1, centred
at the origin, and we obtained I
1
dz = 2πi.
C z
We will argue, using the Cauchy Integral Theorem, that we get the same result whether
we integrate along E or along C.

Γ+

Γ−

Consider the two domains in the interior of the circle C but in the exterior of the ellipse
E. The integrand is analytic everywhere in the complex plane except for the origin, which
lies outside these two regions. Let us call these contours Γ± as shown in the two figures.
The Cauchy Integral Theorem says that the contour integral vanishes along either of the
two contours which make up the boundary of these domains.

Then we can clearly see that

I I I I
1 1 1 1
dz = dz + dz + dz.
C z Γ+ z Γ− z E z

1
Since the interior of Γ± is simply-connected and the integrand z
is analytic in and on Γ± ,
the Cauchy Integral Theorem says that
I
1
dz = 0,
Γ± z

46
and thus we get I I
1 1
dz = dz = 2πi.
E z C z

In other words, we could deform the contour from E to C without altering the result of
the integral because in doing so we do not pass over any point where the integrand ceases
to be analytic.

Let us illustrate this with another example. Let Γ be any positively-oriented loop in the
complex plane and let z0 be any complex number, which does not lie on Γ. Consider the
following contour integral I
1
dz.
Γ z − z0
There are two possibilities that we must distinguish: z0 is in the interior of Γ or in
the exterior. In the latter case, the integral is zero because the integrand is analytic
everywhere except for z0 , hence if z0 lies outside Γ, Cauchy’s Integral Theorem applies
and the integral vanishes.

On the other hand, if z0 is in the interior of Γ we expect that we should obtain a nonzero
answer. This is because if Γ were the circle |z − z0 | = R > 0, then the same calculation
as in (8.5) yields a value of 2πi for the integral. In fact, this is the answer that we will
get for any positively-oriented loop containing z0 in its interior.

z0
P4 P3 P2 P1
C

Figure 9.3: The contours Γ and C.

In Fig.9.3 we have shown the contour Γ and a circular contour C of radius r about the
point z0 . We have also shown two pairs of points (P1 , P2 ) and (P3 , P4 ); each pair of points
having one point in each contour. In addition, the straight line segments joining the
points in each pair are also shown.

47
Now let us consider the loop Γ1 starting and ending at P1 , as shown in Fig.9.4. We start
at P1 and go to P4 via the top half of Γ, which we call Γ+ ; then we go to P3 along the
straight line segment −γ34 ; then to P2 via −C+ ; and then back to P1 via the straight line
segment −γ12 . The interior of this contour is simply-connected and does not contain the
point z0 . Therefore, using Cauchy’s Integral Theorem we get
I Z Z Z Z !
1 1
dz = + + + dz
Γ1 z − z0 Γ+ −γ34 −C+ −γ12 z − z0
Z Z Z Z !
1
= − − − dz
Γ+ γ34 C+ γ12 z − z0
= 0,

and from this we get

Z Z Z Z !
1 1
dz = + + dz.
Γ+ z − z0 γ34 C+ γ12 z − z0

Γ+

Γ1
−C+
γ34 γ12
P4 −γ34 −γ12
−C−
Γ2

Γ−

Figure 9.4: The contours Γ1 and Γ2 .

Similarly, let us consider the loop Γ2 starting and ending at P4 . It starts at P4 and go to
P1 along the lower half of Γ, which we call Γ− ; then to P2 along γ12 ; and from there to
P3 via −C− ; and then finally back to P4 along γ34 . Using the same argument as above,
the interior of Γ2 is simply-connected and z0 lies in its exterior domain, we get, by the

48
Cauchy Integral Theorem
I Z Z Z Z !
1 1
dz = + + + dz
Γ2 z − z0 Γ− γ12 −C− γ34 z − z0
Z Z Z Z !
1
= + − + dz
Γ− γ12 C− γ34 z − z0
= 0,

and from this we get

Z Z Z Z !
1 1
dz = − + − dz.
Γ− z − z0 γ12 C− γ34 z − z0

Putting the two results together, we find that

I Z Z
1 1 1
dz = dz + dz
Γ z − z0 Γ+ z − z0 Γ− z − z0
Z Z
1 1
= dz + dz
C+ z − z0 C− z − z0
Z
1
= dz
C z − z0

Now, on C, |z − z0 | = r. Let us parametrise C by z(t) = z0 + rexp(i 2π t) for 0 ≤ t ≤ 1.

Then 1 1
2πi rei2πt
I Z Z
dz
= dt = 2πi dt = 2πi.
C z − z0 0 rei2πt 0
Therefore, I Z
1 1
dz = dz = 2πi.
Γ z − z0 C z − z0

In summary, we find that if Γ is any positively-oriented loop in the complex plane and z0
a point not in Γ, then

I
1 2πi for z0 in the interior of Γ; and
dz = (9.4)
Γ z − z0 0 otherwise.

In the next section we will generalise this formula.

10 Cauchy’s Integral Formula

Statement: If f (z) be a function which is analytic inside and on a positively-oriented

simple closed curve (contour) C in a simply-connected region and if z0 is any point in the

49
interior of C then the value of the function f (z0 ) is given by the expression
I
1 f (z)
f (z0 ) = dz. (10.1)
2πi C z − z0

Proof of Cauchy’s Integral Formula

Suppose C2 be a circle of radius r with center at z = z0 , so that C2 is inside C (see the

figure below). C is any positively-oriented simple closed contour.

Now, I I
dz dz
= = 2πi.
C z − z0 C2 z − z0

z0
C2 r

Also, we have I I
f (z) f (z)
dz = dz.
C z − z0 C2 =|z−z0 |=r z − z0

Since f (z) is continuous at z = z0 , we can take

|f (z) − f (z0 )| <  provided |z − z0 | = r < δ,

where  is a pre-assigned small positive number.

Therefore,
I I
f (z) f (z)
dz = dz
C z − z0 C2 =|z−z0 |=r z − z0
I  
f (z0 ) f (z) − f (z0 )
= + dz
C2 z − z0 z − z0
f (z) − f (z0 )
I
= 2πif (z0 ) + dz.
C2 z − z0

From this, we get

  I  
f (z) − f (z0 )
I
1 f (z) 1
dz − f (z0 ) = dz (10.2)
2πi C z − z0 2πi C2 z − z0

50
Now, for z on C2 , we have
f (z) − f (z0 ) 
< .
z − z0 r
Hence, using Darboux’s inequality, we get
I  
1 f (z) − f (z0 ) 1 
dz ≤ · · 2πr = .
2πi C2 z − z0 2π r

Therefore, in the limit  → 0, the R.H.S of (10.1) tends to zero and since the L.H.S does
not depend on , the L.H.S. is also zero, i.e.,
 I 
1 f (z)
dz − f (z0 ) = 0.
2πi C z − z0

From this, we get

I
1 f (z)
f (z0 ) = dz.
2πi C z − z0

It is a remarkable formula. It says that an analytic function in a simply-connected domain

is determined by its behaviour on the boundary. In other words, if two analytic functions
f (z) and g(z) agree on the boundary of a simply-connected domain they agree everywhere
on the domain.

Cauchy’s Integral Formula is a mathematical analogue of a notion that is very much in

use in today’s theoretical physics, namely ‘holography’. The idea of an optical hologram
is that it is a two-dimensional film which contains enough information to reconstruct
(optically) a three-dimensional object. In theoretical physics, the celebrated formula of
Bekenstein-Hawking says that the entropy of a black hole is proportional to the area of
the horizon. In simple terms, the horizon of the black hole is the surface within which
light can no longer escape the gravitational attraction of a black hole.

On the other hand, we know from Boltzmann’s formula that the entropy of a statistical
mechanical system is a measure of the density of states of the system. The entropy formula
of Bekenstein-Hawking is holographic because it tells us that the degrees of freedom of
a three-dimensional object like a black hole is determined from the properties of a two-
dimensional system: the horizon, just like with the optical hologram.

The Cauchy Integral Formula is holographic in the sense that an analytic function in the
plane (which is two-dimensional) is determined by its behaviour on contours (which are
one-dimensional).

51
10.1 The generalised Cauchy Integral Formula

Using Cauchy’s Integral Formula one can show that the n-th derivative of f (z) at z = z0
is given by
I
(n) n! f (z)
f (z0 ) = dz, n = 1, 2, 3 . . . (10.3)
2πi C (z − z0 )n+1

Cauchy’s integral formula can be considered a special case of (10.3) with n = 0, if we

define 0! = 1 and understand the zeroth derivative f (0) as the function f itself.

These results are quite remarkable because they show that if an analytic function f (z) is
known on the simple-closed curve C in a simply-connected region, then the values of the
function and all its derivatives can be found at all points inside C.

Thus if a function of a complex variable has a first derivative, i.e., is analytic, in a simply-
connected region R, all its higher derivatives exist in R. That is to say, the derivatives of
all order of an analytic function are themselves analytic.

Let us now prove the above formula for n = 1.

If f (z) is analytic in a simply-connected region R and if C is a positively-oriented loop in

R, then I
0 1 f (z)
f (a) = dz,
2πi C (z − a)2
where a is a point in the interior of C.

Proof: If a and a + h lie in R in the interior of C, we have

 
f (a + h) − f (a)
I
1 1 1 1
= − f (z)dz [using (9.4)]
h 2πi C h z − (a + h) z − a
I
1 f (z)dz
=
2πi C (z − a − h)(z − a)
{(z − a − h) + h}f (z)dz
I
1
=
2πi C (z − a − h)(z − a)2
I I
1 f (z)dz h f (z)dz
= 2
+
2πi C (z − a) 2πi C (z − a − h)(z − a)2

The result follows if we take the limit h → 0 and show that the last term on the R.H.S. of
the above equation approaches zero. Note that we have to justify taking the limit inside
the integral.

To show this we use the fact that if Γ is a circle of radius  and centre at a, which lies
entirely in the interior of C (see figure below). Then, we have

52
 C

Γ
a


a+h

I I
h f (z)dz h f (z)dz
2
= .
2πi C (z − a − h)(z − a) 2πi Γ (z − a − h)((z − a)2

Since we are taking the limit h → 0, we can choose |h| ≤ 12  so that a + h lies in the
interior of Γ. Now by the triangle inequality

|z1 − z2 | ≥ |z1 | − |z2 |,

(this is another form of triangle inequality which can be obtained from (2.3))

and the fact that Γ has equation |z − a| = , we have

|z − a − h| ≥ |z − a| − |h| ≥  − /2 = /2.

Also, since f (z) is analytic in the simply-connected region R, we can find a positive
number M such that |f (z)| < M.

Then since the length of Γ is 2π, we have

|h| M (2π)
I
h f (z)dz 2|h|M
2
≤ 2
= .
2πi Γ (z − a − h)(z − a) 2π (/2)( ) 2

It follows that L.H.S of the above equation approaches zero as h → 0. Thus we have
proved that I
0 1 f (z)
f (a) = dz.
2πi C (z − a)2
It is interesting to note that the result is equivalent to taking derivative under the integral
sign.

53
In a similar manner we can prove that
I
(n) n! f (z)
f (a) = dz, n = 2, 3, . . . .
2πi C (z − a)n+1

The generalised Cauchy integral formula can also be turned around in order to compute
contour integrals. Hence, if f (z) is analytic in and on a positively-oriented loop C, and if
z0 is a point in the interior of C, then
I
f (z) 2πi (n)
n+1
dz = f (z0 ). (10.4)
C (z − z0 ) n!

Morera’s Theorem

Finally, we discuss a converse of the Cauchy Integral Theorem, known as Morera’s The-
orem. Let f (z) be continuous in a simply-connected region R and suppose that
I
f (z)dz = 0,
C

around every simple closed curve C in R. Then f (z) is analytic in R.

Home assignment 13

Q.1 Evaluate the contour integral

e5z
I
dz,
Γ z3
where Γ is the positively oriented unit circle |z| = 1. (Use generalised Cauchy Integral
Formula.)

Q.2 Evaluate the contour integral

e2z
I
dz,
Γ (z + 1)4
where Γ is the circle |z| = 3. (Use generalised Cauchy Integral Formula.)

Q.3 Evaluate the contour integral

e3z
I
dz,
C z − πi
where C is the circle |z − 1| = 4. (Use Cauchy Integral Formula.)

Q.4 Evaluate the contour integral

Z
2z + 1
dz,
Γ z(z − 1)2

54
 Γ Γ0 Γ1
0 +1 0 +1

where Γ is the contour shown in the left panel of the figure above.

[Hint: Two things prevent us from applying the generalised Cauchy Integral Formula.
The contour is not a simple closed curve and the integrand is not of the form f (z)/(z−z0 )n
where f (z) is analytic inside the contour. Notice that the smooth contour Γ can be written
as a piecewise smooth contour with two smooth components: both starting and ending
at the point of self-intersection of Γ. The first such contour is the left lobe of Γ, which is
a negatively oriented loop about z = 0, and the second is the right lobe of Γ, which is a
positively oriented loop about z = 1. Because the integrand is analytic everywhere except
at z = 0 and z = 1, the Cauchy Integral Theorem tells us that we get the same result
by integrating around the circular contours Γ0 and Γ1 as shown in the right panel of the
above figure. For the integrand use partial fractions.]

10.2 Cauchy’s inequality and Liouville’s Theorem

An immediate application of the generalised Cauchy Integral Formula is the so-called

Cauchy inequality or Cauchy estimates for the derivatives of an analytic function.

If f (z) is analytic in some domain D containing a circle C of radius R centred about z0 ,

then Cauchy’s inequality tells that
M · n!
|f (n) (z0 )| ≤ , n = 0, 1, 2, 3, . . .
Rn
where M is a constant such that |f (z)| ≤ M for all z on the circle C.

Proof: We have by the generalised Cauchy’s Integral Formula,

I
(n) n! f (z)
f (z0 ) = dz, n = 0, 1, 2, 3, . . .
2πi C (z − z0 )n+1

Since, |z − z0 | = R on C and the length of C is 2πR,

I
(n) n! f (z) n! M
|f (z0 )| = n+1
dz ≤ · n+1 · 2πR
2π C (z − z0 ) 2π R
M · n!
=
Rn

55
Therefore, we have
M · n!
|f (n) (z0 )| ≤ . (10.5)
Rn
This inequality is known as the Cauchy estimate.

Let n = 1 and we see that at any point z0 in the complex plane, the derivative of f (z) is
bounded by
|f 0 (z0 )| ≤ M/R.

As an immediate corollary of this estimate suppose that f (z) is analytic in whole complex
plane (i.e. f (z) is an entire function) and that it is bounded, so that |f (z)| ≤ M for all
z. Because the function is entire, we can take R as large as we wish.

Now, given any number ε > 0, however small, there is always an R large enough for which
M/R < ε, so that |f 0 (z0 )| < ε.

Therefore, |f 0 (z0 )| = 0, whence f 0 (z0 ) = 0. Since this is true for all z0 in the complex
plane, we get f (z) = constant, as required. Thus we have proven Liouville’s theorem:

a bounded entire function is constant.

This does not violate our experience since the only entire functions we have met are poly-
nomials and the exponentials and functions we can make out of them by multiplication,
linear combinations and compositions, and these functions are clearly not bounded.

Using the above result one can prove the Fundamental Theorem of Algebra which
states that

every nonconstant polynomial has at least one zero.

However, I will not discuss the proof here.

11 Series expansion for analytic functions

Let u1 (z), u2 (z), . . . , un (z), . . . denoted by {un (z)}, be a sequence of functions of z (infinite
set of complex functions) defined and single-valued in some region of the z-plane.

We call U (z) the limit of un (z) as n → ∞, and write

lim un (z) = U (z),

n→∞

56
if given any positive number  we can find a number N [depending in general on both 
and z] such that
|un (z) − U (z)| <  for all n > N.

In such case the sequence converges or is convergent to U (z).

If a sequence converges for all values of z in a region R, we call R the region of con-
vergence of the sequence. A sequence which is not convergent at some point z is called
divergent at Z.

11.1 Series of functions

From the sequence of functions {un (z)} let us form a new sequence {Sn (z)} defined by

S1 (z) = u1 (z)
S2 (z) = u1 (z) + u2 (z)
.. ..
. .
Sn (z) = u1 (z) + u2 (z) + · · · + un (z),

where Sn (z), called the n-th partial sum, is the sum of the first n terms of the sequence
{un (z)}.
∞
X
u1 (z) + u2 (z) + · · · = un (z) (11.1)
n=1

is called an infinite series.

If limn→∞ Sn (z) = S(z), the series is called convergent and S(z) is its sum; otherwise the
series is called divergent.

A necessary condition that the series (11.1) converges is

lim un (z) = 0,
n→∞

but this is not sufficient.

If a series converges for all values of z in a region R, we call R the region of convergence
of the series.

Absolute convergence

A series ∞
P
n=1 un (z) is called absolutely convergent if the series of absolute values, i.e.,
P∞
n=1 |un (z)|, converges.

57
 P∞ P∞ P∞
If n=1 un (z) converges but n=1 |un (z)| does not converge, we call n=1 un (z) condi-
tionally convergent.

Uniform convergence of sequence and series

If we can find a number N such that

|un (z) − U (z)| < , for all n > N,

where the same number N holds for all z in a region R [i.e., N depends only on  and
not on the particular value of z in the region], we say un (z) converges uniformly, or is
uniformly convergent, to U (z) for all z in R.

If the sequence of partial sums {Sn (z)} converges uniformly to S(z) in a region, we say
that the infinite series (11.1) converges uniformly, or is uniformly convergent, to S(z) in
the region.

If we call Rn (z) = un+1 (z) + un+2 (z) + · · · = S(z) − Sn (z) the remainder of the infinite
series (11.1) after n-terms, we can equivalently say that the series is uniformly convergent
to S(z) in R if given any  > 0 we can find a number N such that for all z in R,

|Rn (z)| = |S(z) − Sn (z)| < 

for all n > N.

Power series

A series having the form

∞
X
2
a0 + a1 (z − a) + a2 (z − a) + · · · = an (z − a)n , (11.2)
n=0

is called a power series in (z − a). Clearly, the power series (11.2) converges for z = a,
and this may indeed be the only point for which it converges.

In general, however, the series converges for other points as well. In such a case we can
show that there exists a positive number R such that (11.2) converges for |z − a| < R and
diverges for |z − a| > R, while for |z − a| = R it may or may not converge.

Geometrically, if Γ is a circle of radius R with center at z = a, then the series (11.2)

converges at all points inside Γ and diverges at all points outside Γ, while it may or may
not converge on the circle Γ.

R is called the radius of convergence of (11.2) and the corresponding circle is called
the circle of convergence.

58
We consider the special cases R = 0, and R = ∞, respectively to be the cases where
(11.2) converges only at z = a or converges for all (finite) values of z.

Some important theorems

Let us state a few important theorems without proof.

Theorem 1: If a sequence has a limit, the limit is unique [ i.e., it is the only one].

Theorem 2: A necessary and sufficient condition that {un } converges is that given any
 > 0, we can find a number N such that |up − uq | <  for all p > N , q > N . This result
is called Cauchy’s convergence criteria.

Theorem 3: A necessary condition that ∞

P
n=1 un (z) converges is that limn→∞ un = 0.

Theorem 4: If ∞
P P∞
n=1 |un | converges, then n=1 un converges. In words, an absolutely
convergent series is convergent.

Theorem 5: The sum of a uniformly convergent series of continuous functions is contin-

uous, i.e., if un (z) is continuous in R and S(z) = ∞
P
n=1 un (z) is uniformly convergent in
R, then S(z) is continuous in R,
P∞
Theorem 6: If {un (z)} are continuous in R, S(z) = n=1 un (z) is uniformly convergent
in R and if C is a curve in R, then
Z Z Z
S(z)dz = u1 (z)dz + u2 (z)dz + . . .
C C C

or ∞
Z X  XZ
un (z) dz = un (z)dz.
C n=1 C

In words, a uniformly convergent series of continuous functions can be inte-

grated term by term.

Theorem 7: If u0n (z) = dz d

P 0
un (z) exists in R, un (z) converges uniformly in R and
P
un (z) converges in R, then
d X X
un (z) = u0n (z).
dz

Theorems on Power Series

Theorem 8: A power series converges uniformly and absolutely in any region which lies
entirely inside its circle of convergence.

Theorem 9: (a) A power series can be differentiated term by term in any region which
lies entirely inside its circle of convergence.

59
(b) A power series can be integrated term by term along any curve C which lies entirely
inside its circle of convergence

(c) The sum of a power series is continuous in any region which lies entirely inside its
circle of convergence.

Uniform convergence of the Geometric Series

Let us take an example to see how, in practice, one checks that a sequence {Sn (z)} of
functions converges uniformly in a subset U to a function S(z). Let us see this for the
Geometric series: ∞
X
zj .
j=0

Let
Sn (z) = 1 + z + z 2 + · · · + z n−1

Then
zSn (z) = z + z 2 + · · · + z n−1 + z n

Subtracting, we get
1 − zn
(1 − z)Sn (z) = 1 − z n =⇒ Sn (z) = . (11.3)
1−z

We claim that this geometric series converges uniformly to the function 1/(1 − z) on every
closed disk |z| ≤ R with R < 1. Indeed, we have the following estimate for the remainder:

1 |z n | |z|n Rn
− Sn (z) = = ≤ .
1−z |1 − z| |1 − z| |1 − z|

Now, using the triangle inequality

1 1 1
|z − 1| = |1 − z| ≥ 1 − |z| whence ≤ ≤ .
|1 − z| 1 − |z| 1−R

In other words,
1 |z|n Rn
− Sn (z) = ≤ .
1−z |1 − z| 1−R
This bound is independent of z and can be made as small as desired since R < 1, whence
the convergence is uniform.

Thus,
∞
X 1
z j = lim Sn (z) = if |z| < 1.
j=0
n→∞ 1−z

60
11.2 Taylor Theorem

If f (z) is analytic inside a circle C of radius R, with centre at a, then for all z inside C,
f 00 (a) f 000 (a)
f (z) = f (a) + f 0 (a)(z − a) + (z − a)2 + (z − a)3 + . . .
2! 3!
The series is the Taylor series around a of the function f (z). If a = 0, the series is also
called the Maclaurin series of f (z).

The Taylor series for f (z) around a converges to f (z) for all z in the disk |z − a| < R and
moreover the convergence is uniform on any closed subdisk |z − a| ≤ r < R.

Proof: Let z be any point inside C. Construct a circle C1 with radius r1 , centre at a and
enclosing z, where r < r1 < R.

z
a
C1
r1

Then by Cauchy’s integral formula

I
1 f (w)
f (z) = dw. (11.4)
2πi C1 w−z
Now, we have
 
1 1 1 1
= = ,
w−z (w − a) − (z − a) (w − a) 1 − (z − a)/(w − a)
where
z−a
< 1,
w−a
because w is on the circumference of C1 and z in the interior of C1 .

Therefore,
    2  n−1  n 
1 1 z−a z−a z−a z−a 1
= 1+ + +· · ·+ + .
w−z w−a w−a w−a w−a w − a 1 − (z − a)/(w − a)

This is because, if
Sn = 1 + z + z 2 + · · · + z n−1 , |z| < 1,

61
then
zn 1
Sn + = see (11.3).
1−z 1−z

Therefore,
n
(z − a)2 (z − a)n−1

1 1 z−a z−a 1
= + + + · · · + + (11.5)
w−z w − a (w − a)2 (w − a)3 (w − a)n w−a w−z

Multiplying both sides of (11.5) by f (w) and using (11.4). we have

(z − a) (z − a)n−1
I I I
1 f (w) f (w) f (w)
f (z) = dw + 2
dw + · · · + n
dw + Un ,
2πi C1 w − a 2πi C1 (w − a) 2πi C1 (w − a)
(11.6)

where  n
z−a
I
1 f (w)
Un = dw.
2πi C1 w−a w−z
Using generalised Cauchy’s integral formula
I
(n) n! f (w)
f (a) = dw, n = 0, 1, 2, 3 . . .
2πi C1 (w − a)n+1

(11.6) becomes

f 00 (a) f (n−1) (a)

f (z) = f (a) + f 0 (a)(z − a) + (z − a)2 + · · · + (z − a)n−1 + Un .
2! (n − 1)!

In order to prove uniform convergence of the Taylor series, we only have to show that
limn→∞ Un = 0.

To do this we note that since w is on C1 ,

z−a
= γ < 1, (γ is some constant).
w−a

Also, we have |f (w)| ≤ M , where M is a constant, and

|w − z| = |(w − a) − (z − a)| ≥ r1 − |z − a|,

where r1 is the radius of C1 .

Hence, we have
I  n
1 z−a f (w)
|Un | = dw
2π C1 w − a w − z
1 γ nM γ n M r1
≤ · 2πr1 = .
2π r1 − |z − a| r1 − |z − a|

62
Thus,
lim Un = 0.
n→∞

Notice that this result implies that the Taylor series will converge to f (z) everywhere
inside the largest open disk, centred at a, over which f (z) is analytic.

As an example, let us compute the Taylor series for the function Log z around a = 1.
The derivatives of the principal branch of the logarithm are:
dj Log z 1
j
= (−1)j+1 (j − 1)! j .
dz z
Evaluating at z = 1 and constructing the Taylor series, we have
∞
X (−1)j+1
Log z = (z − 1)j .
j=1
j

This series is valid for |z − 1| < 1 which is the largest open disk centred at z = 1 over
which Log z is analytic, as seen in the left panel of the figure below. Similarly, let us

1 0 1

Figure 11.5: Analyticity disks for the Taylor series of Log z and 1/(1 − z).

compute the Taylor series for the function 1/(1 − z) around a = 0.

dj 1 j!
= ,
dz j 1 − z (1 − z)j+1
whence evaluating at z = 0 and building the Taylor series we find the geometric series
∞
1 X
= zj ,
1−z j=0

which is valid for |z| < 1 since that is the largest open disk around the origin over which
1/(1 − z) is analytic as seen in the right panel of the figure above.

Let us notice something very interesting. We have a priori two different series represen-
tations for the function 1/(1 − z) around the origin: one is the Taylor series and another
is the geometric series. Yet we have shown that these two series are the same. This is

63
not a coincidence, series representations for analytic functions are unique: they are all
essentially Taylor series. We will also state another result without proof. Any power
series is the Taylor series of a function analytic in the disk of convergence |z − a| < R.

This is a very useful result, because it says that in order to compute the Taylor series of
a function it is enough to produce any power series which converges to that function. For
example, let us compute the Taylor series of the function
1
(z − 1)(z − 2)
in the disk |z| < 1. This is the largest disk centred at the origin where we could hope to
find a convergent power series for this function, since the function fails to be analytic at
z = 1 and z = 2. The naive solution to this problem would be to take derivatives and
evaluate them at the origin and build the Taylor series this way. However, it is enough to
exhibit any power series which converges to this function in the specified region. Let us
write
1 1 1
= − .
(z − 1)(z − 2) 1−z 2−z
Now we use geometric series for each of them. For the first fraction we have
∞
1 X
= zj valid for |z| < 1;
1−z j=0

whereas for the second fraction we have

∞ ∞
−1 −1/2 1 X zj X zj
= =− = − ,
2−z 1 − (z/2) 2 j=0 2j j=0
2j+1

which is valid for |z| < 2, which contains the region of interest. Therefore, putting the
two series together,
∞  
1 X 1
= 1 − j+1 z j , for |z| < 1.
(z − 1)(z − 2) j=0
2

11.3 Laurent’s Theorem

A function which is analytic in some neighbourhood of a point a can be approximated by

its Taylor series about that point. However, if a function f (z) is not analytic throughout
the whole interior of a circle centred at a, but only throughout the annular region between
two concentric circles C1 and C2 , it is possible to generalise the Taylor series of f (z), which
now consist of a sum of two power series with both positive and negative powers of (z −a).
This is known as Laurent series and we state this as a theorem.

64
Let f (z) be analytic on two concentric circles C1 and C2 centred at a and in the annular
region between C1 and C2 . Then at any point z within the annular region, the value of
f (z) is given by the uniformly convergent series
∞
X ∞
X
f (z) = aj (z − a) +j
a−j (z − a)−j , (11.7)
j=0 j=1

where
I
1 f (ζ)
an = dζ, n = 0, ±1, ±2, . . . , (11.8)
2πi C (ζ − a)n+1

and C is any positively-oriented contour within the annular region encircling the point a.

This is Laurent’s Theorem and (11.7) with coefficients (11.8) is called a Laurent series.

The part a0 + a1 (z − a) + a2 (z − a)2 + . . . is called the analytic part of the Laurent

series, while the remainder of the series which consists of inverse power of (z − a) is called
the principal part. If the principal part is zero, the Laurent series reduces to a Taylor
series.

The Laurent series around a converges uniformly to f (z) in the closed annular region
R2 ≤ |z − a| ≤ R1 , where R1 and R2 are radii of the concentric circles C1 and C2
respectively and R1 > R2 .

Proof:

C1
z
γ

C2 −C2

R1 C0

R2 a

65
Let us draw a small circle γ centred at z and contained entirely within the annular region
of analyticity of f (z). Consider the closed contour C 0 , shown in the figure above.

According to Cauchy’s theorem

I
f (w)
dw = 0, (11.9)
C0 (w − z)

since the integrand is analytic in the region enclosed by C 0 . However, the parallel straight
segments of the path can lie arbitrary close to each other and thus give contributions
equal in magnitude but of opposite sign.

Therefore, (11.9) can be rewritten as

I I I
f (w) f (w) f (w)
dw = dw − dw (11.10)
γ (w − z) C1 (w − z) C2 (w − z)

By Cauchy’s integral formula,

I
f (w)
dw = 2πif (z).
γ (w − z)

I I
1 f (w) 1 f (w)
=⇒ f (z) = dw − dw (11.11)
2πi C1 (w − z) 2πi C2 (w − z)

Consider the first integral in (11.11). We have

1 1
=  
w−z
(w − a) 1 − (z − a)/(w − a)
n
(z − a)2 (z − a)n−1

1 z−a z−a 1
= + + + · · · + +
w − a (w − a)2 (w − a)3 (w − a)n w−a w−z
(11.12)

so that
I
1 f (w)
dw
2πi C1 w−z
(z − a) (z − a)n−1
I I I
1 f (w) f (w) f (w)
= dw + 2
dw + · · · + n
dw + Un
2πi C1 w−a 2πi C1 (w − a) 2πi C1 (w − a)
= a0 + a1 (z − a) + . . . + an−1 (z − a)n−1 + Un
(11.13)

66
where
I I
1 f (w) 1 f (w)
a0 = dw, a1 = dw,
2πi C1 w − a 2πi C1 (w − a)2
I
1 f (w)
. . . , an−1 = dw (11.14)
2πi C1 (w − a)n
I  n
1 z−a f (w)
and Un = dw.
2πi C1 w − a w − z

Let us consfder the second integral in (11.11)

1 1
− =  
w−z
(z − a) 1 − (w − a)/(z − a)
n
(w − a)n−1

1 w−a w−a 1
= + + · · · + +
z − a (z − a)2 (z − a)n z−a z−w
so that
I
1 f (w)
− dw
2πi C2 w−z
(w − a) (w − a)n−1
I I I
1 f (w) 1 1
= dw + f (w)dw + · · · + f (w)dw + Vn
2πi C2 z−a 2πi C2 (z − a)2 2πi C2 (z − a)n
a−1 a−2 a−n
= + 2
+ ... + + Vn (11.15)
z−a (z − a) (z − a)n
where
I I
1 1
a−1 = f (w)dw, a−2 = (w − a)f (w)dw,
2πi C2 2πi C2
I
1
. . . , a−n = (w − a)n−1 f (w)dw (11.16)
2πi C2
I  n
1 w−a f (w)
and Vn = dw.
2πi C2 z − a z−w

From (11.11), (11.13), (11.15), we have

 
n−1
f (z) = a0 + a1 (z − a) + · · · + an−1 (z − a)
 
a−1 a−2 a−n
+ + + ··· +
z − a (z − a)2 (z − a)n
+ Un + Vn (11.17)

We have to show (for uniform convergence)

(a) lim Un = 0 (b) lim Vn = 0.

n→∞ n→∞

67
(a) can be shown in a similar fashion as in the case of Taylor series. To prove (b) we first
note that since w is on C2 ,

w−a
= k < 1, k is constant.
z−a

Also, |f (w)| < M , M is a constant and

|z − w| = |(z − a) − (w − a)| ≥ |z − a| − R2 .

I  n
1 w−a f (w)
|Vn | = dw
2π C2 z − a z−w
1 knM k n M R2
≤ 2πR2 =
2π |z − a| − R2 |z − a| − R2
Then
lim Vn = 0, (proof of uniform convergence.)
n→∞

The series (11.17) converges uniformly to f (z) for

R2 < |z − a| < R1

R1 and R2 denoting respectively the radii of the circles C1 and C2 , since this is the
condition for simultaneous convergence of the series in (11.13) and (11.15).

We are almost done, except that in the statement of the theorem the coefficients an are
given by contour integrals along C and what we have shown is that they are given by
contour integrals along C1 or C2 . Here C is any positively-oriented loop lying in the
annular region and containing a in its interior. However, notice that the integrand in
(11.14) is analytic in the region bounded by the contours C and C1 and similarly for the
integrand in (11.16) in the region bounded by the contours C and C2 . Therefore, we can
deform the contours C1 and C2 to C in the integrals and thus the coefficients an can be
written in a single formula
I
1 f (w)
an = dw, n = 0, ±1, ±2, . . . . (11.18)
2πi C (w − a)n+1
which proves the theorem.

It turns out that Laurent series representation of a function analytic in an annular region
is unique. The proof follows from the uniqueness of the power series.

Let us compute the Laurent series of the rational function (z 2 − 2z + 3)/(z − 2) in the
region |z − 1| > 1.

68
Let us first rewrite the numerator as a power series in (z − 1):

z 2 − 2z + 3 = (z − 1)2 + 2.

Now we do the same with the denominator:

1 1 1 1
= = 1 ,
z−2 (z − 1) − 1 z − 1 1 − z−1

where we see that we should try a geometric series in 1/(z − 1), which converges in the
specified region |z − 1| > 1. Therefore, we have
∞ ∞
1 1 1 X 1 X 1
1 = j
= j+1
.
z − 1 1 − z−1 z − 1 j=0 (z − 1) j=0
(z − 1)

Putting the two series together, we get

∞
z 2 − 2z + 3
 X
2 1
= (z − 1) + 2
z−2 j=0
(z − 1)j+1
∞
X 3
= (z − 1) + 1 + .
j=0
(z − 1)j+1

[Take u = z − 1 and then expand the sum in the top line of the above equation and
multiply. You will get the result in the bottom line of the above equation.]

By the uniqueness of the Laurent series, this is the Laurent series for the function in the
specified region.

As another example, consider the function 1/(z − 1)(z − 2). Let us find its Laurent series
in the regions: |z| < 1, 1 < |z| < 2 and |z| > 2, which we have labelled I, II and III in
the figure.

III

II
I 0 1 2

Now,
1 1 1
= − .
(z − 1)(z − 2) z−2 z−1

69
In region I, we have the following geometric series:
∞
1 1 X
− = = zj valid for |z| < 1; and
z−1 1−z j=0

∞  j X ∞
1 − 12 1X z −1 j
= =− = z valid for |z| < 2.
z−2 1 − (z/2) 2 j=0 2 j=0
2j+1
Therefore, in their common region of convergence, namely region I, we have
∞  
1 X 1
= 1 − j+1 z j .
(z − 1)(z − 2) j=0
2

In region II, the first of the geometric series above is not valid, but the second one is.
Because in region II, |z| > 1, this means that |1/z| < 1, whence we should try and use a
geometric series in 1/z.
∞  j X ∞
1 1 1 1X 1 −1
− =− =− = valid for |z| > 1.
z−1 z [1 − (1/z)] z j=0 z j=0
z j+1

Therefore in region II we have

X −1 ∞ X −1 ∞
1
= j+1
+ j+1
zj .
(z − 1)(z − 2) j=0
z j=0
2

Finally in region III, we have that |z| > 2, so that we will have to find another series
converging to 1/(z − 2) in this region. Again, since now |2/z| < 1 we should try to use a
geometric series in 2/z. Thus we have
∞  j X ∞
1 1 1 1X 2 2j
= = = valid for |z| > 2.
z−2 z [1 − (2/z)] z j=0 z j=0
z j+1

Therefore, in region III we have

∞
1 X 1
= (−1 + 2j ) j+1 .
(z − 1)(z − 2) j=0
z

Again by the uniqueness of the Laurent series, we know that these are the Laurent series
for the function in the specified regions.

12 Classification of Singularities

A point z0 at which f (z) fails to be analytic is called a singular point or singularity

of f (z).

70
Isolated singularities

The point z = z0 is called an isolated singularity or isolated singular point of f (z) if

we can find δ > 0 such that the circle |z − z0 | = δ encloses no singular point other than z0
(i.e., there exists a deleted δ neighbourhood of z0 containing no singularity). If no such δ
can be found, we call z0 a non-isolated singularity.

In other words, if f (z) is analytic in some punctured disk around the singularity; that is,
in 0 < |z − z0 | < δ for some δ > 0, then the point z0 is an isolated singularity for the
function f (z).

We have of course already encountered isolated singularities, e.g., the function 1/(z − z0 )
has an isolated singularity at z0 .

Singularities need not be isolated, of course. For example, any point −x in the non-positive
real axis is a singularity for the principal branch Log z of the logarithm function which is
not isolated, since any disk around −x, however small, will contain other singularities.

In this section we will concentrate on isolated singularities.

It is possible to classify the singularities of f (z) by examination of its Laurent series.

This is because if z0 be an isolated singularity for a function f (z), this means that f (z)
is analytic in some punctured disk 0 < |z − z0 | < δ, for some δ > 0. By the results of the
previous section, we know that f (z) has a Laurent series expansion there.

• Isolated Singularity: If f (z) has a Laurent series expansion at every point in the
neighbourhood of z0 except at the point z = z0 then the point z0 is an isolated singularity
of f (z).

1. Poles If Laurent series of f (z) around ‘a’ has the form

a−1 a−2 a−n
f (z) = a0 + a1 (z − a) + a2 (z − a)2 + · · · + + 2
+ ··· +
(z − a) (z − a) (z − a)n

in which the principal part has only a finite number of terms

a−1 a−2 a−n
+ 2
+ ··· + ,
(z − a) (z − a) (z − a)n

where a−n 6= 0, then z = a is called a pole of order n. If n = 1, it is called a simple

pole.

2. Removable singularities If a single-valued function f (z) is not defined at z = a but

lim f (z)
z→a

71
exists, then z = a is called a removable singularity. In such case we define f (z) at
z = a as equal to
lim f (z).
z→a

In other words, we say that ‘a’ is a removable singularity of f (z), if the Laurent expansion
of f (z) around ‘a’ has no negative powers, i.e.,
∞
X
f (z) = aj (z − a)j .
j=0

Clearly, for a removable singularity the function is bounded as z → a, since the power
series representation
f (z) = a0 + a1 (z − a) + . . .

certainly has a well-defined limit as z → a; namely, a0 .

This is not the same as saying that f (a) = a0 . If this were the case, then the function
would not have a singularity at a, but it would be analytic there as well.
sin z
Example 1: If f (z) = z
, then z = 0 is a removable singularity since f (0) is not defined
but
sin z
lim = 1.
z→0 z

Note that in this case

z3 z5 z7
 
sin z 1
= z− + − + ...
z z 3! 5! 7!
z2 z4 z6
=1− + − + ...
3! 5! 7!
which has a well-defined limit as z → 0; namely, 1.
 
z 2 −1 1 2
Example 2: z−1 = z−1 (z − 1) + 2(z − 1) = (z − 1) + 2.

In summary, at a removable singularity the function can be redefined at the singularity

so that the new function is analytic there, in effect removing the singularity.

3. Essential singularities If f (z) is single-valued, then any singularity which is not a

pole or removable singularity is called an essential singularity.

If z = a is an essential singularity of f (z), the principal part of the Laurent expansion

around ‘a’ has infinitely many terms.

Example: Since
1 1 1
e1/z = 1 + + 2
+ + ...,
z 2!z 3!z 3

72
z = 0 is an essential singularity.

If a function is single-valued and has a singularity, then the singularity is either a pole or
an essential singularity.

Zeros of analytic functions

Let f (z) be analytic in a neighbourhood of a point z0 . This means that there is an open
disk |z − z0 | < R in which f (z) is analytic. We say that z0 is a zero of order m, for
m = 1, 2, . . . , if

f (z0 ) = f 0 (z0 ) = f 00 (z0 ) = · · · = f (m−1) (z0 ) = 0 but f (m) (z0 ) 6= 0.

If m = 1, z0 is called a simple zero.

Because f (z) is analytic in the disk |z − z0 | < R, it has a power series representation
there: namely the Taylor series:
∞
X f (j) (z0 )
f (z) = (z − z0 )j .
j=0
j!

Since z0 is a zero of order m, the first m terms in the Taylor series vanish, whence
∞
X f (j) (z0 )
f (z) = (z − z0 )j = (z − z0 )m g(z),
j=m
j!

where g(z) has a power series representation

∞
X f (j+m) (z0 )
g(z) = (z − z0 )j
j=0
(j + m)!

in the disk, whence it is analytic there. Moreover, since f (m) (z0 ) 6= 0, we have g(z0 ) =
f (m) (z0 )/m! 6= 0.

It follows from this that the zeros of an analytic function are isolated. Because g(z) is
analytic, and hence continuous, in the disk |z − z0 | < R and g(z0 ) 6= 0, it means that there
is a disk |z − z0 | < ε < R in which g(z) 6= 0, and hence neither is f (z) = (z − z0 )m g(z)
zero there.

Therefore, we can say that if f (z) = (z − z0 )m g(z), where g(z) is analytic and g(z0 ) 6= 0
and m is a positive integer, then z = z0 is called a zero of order m of f (z). If m = 1,
z0 is called a simple zero.

In such case z0 is a pole of order m of the function 1/f (z).

73
Around a pole z0 of order m, the Laurent series for f (z) looks like
1
f (z) = h(z),
(z − z0 )m
where h(z) has a series expansion around z0 given by
∞
X
h(z) = aj−m (z − z0 )j = a−m + a−m+1 (z − z0 ) + . . .
j=0

Let us now discuss the singularities of a rational function. Let f (z) = P (z)/Q(z) be a
rational function. Then we claim that f (z) has either a pole or a removable singularity
at the zeros of Q(z). Suppose that z0 is a zero of Q(z) and assume that it is a zero of
order m. This means that
Q(z) = (z − z0 )m q(z),

where q(z) is an analytic function around z0 and such that q(z0 ) 6= 0. If z0 is not a zero
of P (z), then z0 is a pole of f (z) of order m.

If z0 is a zero of order k of P (z), then we have

P (z) = (z − z0 )k p(z),

where p(z) is analytic and p(z0 ) 6= 0. Therefore, we have that

(z − z0 )k p(z) 1 p(z)
f (z) = m
= m−k
;
(z − z0 ) q(z) (z − z0 ) q(z)
whence f (z) has a pole of order m − k if m > k and has a removable singularity otherwise.

Suppose a function f (z) has an isolated singular point at z = z0 . If we can find a positive
integer m such that
lim (z − z0 )m f (z) = A 6= 0,
z→z0

then z = z0 is called a pole of order m for f (z). If m = 1, z0 is called a simple pole.

Example 1:
1
f (z) =
(z − 2)3
has a pole of order 3 at z = 2.

Example 2:
3z − 2
f (z) =
(z − 1)2 (z + 1)(z − 4)
has a pole of order 2 at z = 1 and simple poles at z = −1 and z = 4.

Singularities at infinity

74
 1
By letting z = w
in f (z), we obtain the function f (1/w) = F (w).

Then the nature of the singularity at z = ∞ is defined to be the same as that of F (w) at
w = 0.
1
Example: Suppose f (z) = z 3 . Then F (w) = f (1/w) = w3
has a pole of order 3 at
3
w = 0. So f (z) = z has a pole of order 3 at z = ∞.

Similarly, let f (z) = ez . Since F (w) = f (1/w) = e1/w has an essential singularity at
w = 0, so f (z) = ez has an essential singularity at z = ∞.

Meromorphic functions

A function which is analytic everywhere in the finite complex plane except at a finite
number of poles is called a meromorphic function.

Example:
z
f (z) =
(z − 1)(z + 3)2
which is analytic everywhere in the finite complex plane except at the poles z = 1 (simple
pole) and z = −3 (pole of order 2) is a meromorphic function.

Now let us discuss an example where the function f (z) has a non-isolated singularity. Let
us consider
f (z) = sec(1/z)

Now,
1
sec(1/z) =
cos(1/z)
1
Singularities when cos(1/z) = 0, i.e., z
= (2n + 1)π/2. Or,

2
z= , n = 0, ±1, ±2, ±3, . . .
(2n + 1)π

In addition, f (z) is not defined at z = 0, so z = 0 is also a singularity.

Now,
 
2 z − 2/(2n + 1)π
lim z− f (z) = lim
z→2/(2n+1)π (2n + 1)π z→2/(2n+1)π cos(1/z)
1
= lim
z→2/(2n+1)π − sin(1/z){−1/z 2 }

{2/(2n + 1)π}2 4(−1)n

= = 6= 0.
sin{ (2n+1)π }
2
(2n + 1)2 π 2

75
 2
Thus the singularities z = (2n+1)π
, n = 0, ±1, ±2, . . . are poles of order 1, i.e. simple
poles. These poles are located on the real axis at z = ±2/π, ±2/3π, ±2/5π, . . . and that
there are infinitely many in a finite interval which includes 0.

−2/5π 2/5π
−2/π −2/3π 2/3π 2/π x

We can surround each of these by a circle of radius δ, which contains no other singularity,
it follows that they are isolated singularities. Note that the δ required is smaller the closer
the singularity is to the origin.

z = 0 is an essential singularity of f (z), which can be seen by Laurent series expansion

of f (z) = sec(1/z).

[Since we cannot find any positive integer n such that limz→0 (z − 0)n f (z) = A 6= 0, it
follows that z = 0 is an essential singularity.]

Also, since every circle of radius δ with centre at z = 0 contains singular points other than
z = 0, no matter how small we take δ, we see that z = 0 is a non-isolated singularity.

Home assignment 14

Q.1 Find Laurent series about the indicated singularity for each of the following functions.
Name the singularity in each case.

e2z
(a) ; z=1
(z − 1)3
1
(b) (z − 3) sin ; z = −2
z+2
z − sin z
(c) ; z=0
z3
z
(d) ; z = −2 (Give the region of convergence.)
(z + 1)(z + 2)

76
Q.2 Expand
1
f (z) =
(z + 1)(z + 3)
in a Laurent series valid for (a) 1 < |z| < 3, (b) |z| > 3.

Q.3 For the following function locate and name the singularities in the finite complex
plane and determine whether they are isolated singularities or not.
z
f (z) =
(z 2 + 4)2

Q.4 Evaluate I
dz
, n = 2, 3, 4, . . .
C (z − a)n
where C is a positively-oriented simple closed curve as shown in the figure and z = a is
inside C.

Γ
a


13 The residue calculus and its application

In this section we will study the theory of residues. Most of the sections are applications
of the residue theory to the computation of the real integrals and infinite sums.

The Cauchy Residue Theorem

Let us consider the behaviour of an analytic function around an isolated singularity. To

be precise let z0 be an isolated singularity for an analytic function f (z).

77
The function is analytic in some punctured disk 0 < |z − z0 | < R, for some R > 0, and
has a Laurent series there of the form
∞
X
f (z) = aj (z − z0 )j .
j=−∞

Consider the contour integral of the function f (z) along a positively oriented loop Γ
contained in the punctured disk and having the singularity z0 in its interior.

Because the Laurent series converges uniformly, we can integrate the series term by term:
I X∞ I
f (z)dz = aj (z − z0 )j dz.
Γ j=−∞ Γ

From the generalised Cauchy Integral Formula or simply by deforming the contour to a
circle of radius ρ < R, we have

2πi for j = −1 and
I
(z − z0 )j dz =
Γ 0 otherwise;

so only the j = −1 term contributes to the sum, so that

I
f (z)dz = 2πia−1 .
Γ

This singles out the coefficient a−1 in the Laurent series and hence we give it a special
name. We say that a−1 is the residue of f (z) at z0 and we write this as Res(f ; z0 ) or
simply Res(z0 ).

Example: Consider the function f (z) = z exp(1/z). This function has an essential
singularity at the origin and is analytic everywhere else. The residue can be computed
from the Laurent series:
∞
1/z
X 1 1
ze =z
j=0
j! z j
∞
X 1 1 1
= j−1
=z+1+ + ...,
j=0
j! z 2z

whence the residue is given by Res(0) = 21 .

It is often not necessary to calculate the Laurent expansion in order to extract the residue
of a function at a singularity.

For example, the residue of a function at a removable singularity vanishes, since there are
no negative powers in the Laurent expansion.

78
On the other hand, if the singularity happens to be a pole, we will see that the residue
can be computed by differentiation.

Suppose, for simplicity, that f (z) has a simple pole at z0 . Then the Laurent series of f (z)
around z0 has the form
a−1
f (z) = + a0 + a1 (z − z0 ) + . . . ,
z − z0
whence the residue can be computed by

Res(f ; z0 ) = lim (z − z0 )f (z)

z→z0

= lim (a−1 + a0 (z − z0 ) + a1 (z − z0 )2 + . . . )
z→z0

= a−1 + 0.

ez
Example: The function f (z) = z(z+1)
has simple poles at z = 0 and z = −1; therefore

ez
Res(f ; 0) = lim zf (z) = lim =1
z→0 z→0 z + 1

ez 1
Res(f ; −1) = lim (z + 1)f (z) = lim =− .
z→−1 z→−1 z e
p(z)
Suppose that f (z) = q(z)
where p and q are analytic at z0 and q has a simple zero at z0
whereas p(z0 ) 6= 0.

Clearly, f (z) has a simple pole at z0 , whence the residue is given by

p(z)
Res(f ; z0 ) = lim (z − z0 )
z→z0 q(z)
p(z) p(z0 )
= lim = ,
z→z0 q(z)−q(z0 ) q 0 (z0 )
z−z0

where we have used that q(z0 ) = 0 and the definition of the derivative, which exists since
q is analytic at z0 .

Example: Let us compute the residues at each singularity of the function f (z) = cot z.

Since cot z = cos z/ sin z, the singularities occur at the zeros of the sine function: z =
nπ, n = 0, ±1, ±2, . . . These zeros are simple because sin0 (nπ) = cos(nπ) = (−1)n 6= 0.

Therefore, we can apply the above formula to deduce that

cos z cos(nπ)
Res(f ; nπ) = = = 1.
(sin z)0 z=nπ cos(nπ)
This result will be crucial for the applications concerning infinite series.

79
Now, let us suppose that f (z) has a pole of order m at z0 . The Laurent expansion is then
a−m a−1
f (z) = m
+ ··· + + a0 + a1 (z − z0 ) + . . .
(z − z0 ) z − z0
Let us multiply this by (z − z0 )m to obtain

(z − z0 )m f (z) = a−m + · · · + a−1 (z − z0 )m−1 + a0 (z − z0 )m + . . . ,

whence taking m − 1 derivatives, we have

dm−1
[(z − z0 )m f (z)] = (m − 1)! a−1 + m! a0 (z − z0 ) + . . . .
dz m−1
If we evaluate this at z = z0 , we obtain (m − 1)!a−1 , which then gives a formula for the
residue of f (z) at a pole of order m:

1 dm−1
Res(f ; z0 ) = lim [(z − z0 )m f (z)]. (13.1)
z→z0 (m − 1)! dz m−1

Let us now find out the formula for the integral of a function f (z) which is analytic on
a positively-oriented loop Γ and has only a finite number of isolated singularities {zk } in
the interior of the loop.

Because of the analyticity of the function and using a contour deformation argument,
we can express the integral of f (z) along Γ as the sum of the integrals of f (z) along
positively-oriented loops Γk , each one encircling one of the isolated singularities.

However, we have seen above that the integral along each of these loops is given by 2πi
times the residue of the function at the singularity. In mathematical expressions, we have
I XI X
f (z)dz = f (z)dz = 2πiRes(f ; zk ).
Γ k Γk k

In other words, we arrive at the Cauchy Residue Theorem, which states that the
integral of f (z) along Γ is equal to 2πi times the sum of the residues of the singularities
in the interior of the contour:
I X
f (z)dz = 2πi Res(f ; zk ). (13.2)
Γ singularities
zk ∈ Int Γ

14 Applications: Evaluation of real integrals

Since contour integration is the integration of the restriction of a complex function over
a curve, definite and improper integrals over the reals can be evaluated using contour

80
integration by identifying the real domain of integration as part of the contour in the
complex plane. We consider a few generic types.

Type-I

Integrals involving trigonometric functions

Z 2π
I= R(cos θ, sin θ)dθ,
0

where R is a rational function of its arguments and such that it is finite in the range
0 ≤ θ ≤ 2π. We want to turn this into a complex contour integral so that we can apply
the residue theorem.

The integral can be converted into an integral of an analytic function of a complex variable
over a closed contour Γ parametrised by z = eiθ for θ ∈ [0, 2π]. The closed contour Γ is
the unit circle traversed once in the positive sense.

|z| = 1
O

On this contour, we have

1
z = cos θ + i sin θ, = cos θ − i sin θ.
z
Therefore,    
1 1 1 1
cos θ = z+ and sin θ = z− .
2 z 2i z
dz
Similarly, dz = iz dθ, whence dθ = iz
. Putting it all together we have

z + z1 z − z1
I  
1
I= R , dz,
|z|=1 iz 2 2i

which is the contour integral of a rational function of z and hence can be computed using
the residue theorem.
X
I = 2π Res(f ; zk ), (14.1)
singularities
|zk |<1

81
where f (z) is the rational function
z + z1 z − z1
 
1
f (z) ≡ R , . (14.2)
z 2 2i

Example:
2π
(sin θ)2
Z
I= dθ.
0 5 + 4 cos θ
First of all, notice that the denominator never vanishes, so that we can go ahead.

The rational function f (z) given in (14.2) is

1 ( 2i1 (z − z1 ))2 1 (z 2 − 1)2 1 (z 2 − 1)2
f (z) = =− 2 2 =− 2 ,
z 5 + 4 12 (z + z1 ) 4 z (2z + 5z + 2) 8 z (z + 12 )(z + 2)
whence it has a double pole at z = 0 and single poles at z = − 21 and z = −2.

Only the poles at z = 0 and z = − 12 lie inside the unit disk. Therefore,
 
1
I = 2π Res(f ; 0) + Res(f ; − ) .
2
Now,
1 (z 2 − 1)2
 
d 5
Res(f ; 0) = lim − 1 = ,
z→0 dz 8 (z + 2 )(z + 2) 16
1 (z 2 − 1)2
 
1 3
Res(f ; − ) = lim1 − 2
=− .
2 z→− 2 8 z (z + 2) 16
Therefore,
2π
(sin θ)2
Z  
5 3 π
I= dθ = 2π − = .
0 5 + 4 cos θ 16 16 4

14.1 Improper Integrals

In this section we consider improper integrals of rational functions and of products of

rational and trigonometric functions.

Let f (x) be a function of a real variable, which is continuous in 0 ≤ x < ∞. Then by the
R∞
improper integral 0 f (x)dx, we mean the limit
Z ∞ Z R
f (x)dx ≡ lim f (x)dx,
0 R→∞ 0

if such a limit exists. Similarly, if f (x) is continuous in −∞ < x ≤ 0, then the impropoer
R0
integral −∞ f (x)dx is defined by the limit
Z 0 Z 0
f (x)dx ≡ lim f (x)dx,
−∞ r→−∞ r

82
again provided the limit exists.

If f (x) is continuous on the whole real line and both of the above limit exists, we define
Z ∞ Z R
f (x)dx ≡ lim f (x)dx. (14.3)
−∞ R→∞ r
r→−∞

If such limits exist, then we get the same result by symmetric integration:
Z ∞ Z ρ
f (x)dx = lim f (x)dx. (14.4)
−∞ ρ→∞ −ρ

Notice, however, that the symmetric integral may exist even if the improper integral (14.3)
does not.
R∞ R0
For example, consider the function f (x) = x. Clearly, the integrals 0 xdx and −∞ xdx
Rρ
do not exist, yet because x is an odd function, −ρ xdx = 0 for all ρ, whence the limit is
0.

In cases like this we say that equation (14.4) defines the Cauchy principal value of the
integral and we denote this by
Z ∞ Z ρ
p.v. f (x)dx ≡ lim f (x)dx.
−∞ ρ→∞ −ρ

We emphasise to point out that whenever the improper integral (14.3) exists it agrees
with its principal value.

Type-II: Improper integrals of rational functions over (−∞, ∞)

Let us consider as an example the improper integral

Z ∞ Z ρ
dx dx
I = p.v. 2
= lim 2
.
−∞ x + 4 ρ→∞ −ρ x + 4

The integral for finite ρ can be interpreted as the complex integral of the function f (z) =
1/(z 2 + 4), Z
dz
,
γρ z2 + 4
where γρ is the straight line segment on the real axis: y = 0 and −ρ ≤ x ≤ ρ.

In order to use the residue theorem we need to close the contour; that is, we must produce
a closed contour along which we can apply the residue theorem.

Of course, in doing so we are introducing a further integral, and the success of the method
depends on whether the extra integral is computable. We will see that in this case, the
extra integral, if chosen judiciously, vanishes.

83
Let us therefore complete the contour γρ to a closed contour. One suggestion is to consider
the semicircular contour Cρ+ in the upper half plane, parametrised by z(t) = ρ exp(it), for
t ∈ [0, π]. Let Γρ be the composition of both contours: it is a closed contour as shown in
the figure.

Cρ+

2i

-ρ γρ ρ
-2i

Then, according to the residue theorem,

I Z Z
dz dz dz X
2
= 2
+ 2
= 2πi Res(f ; zk );
Γρ z + 4 γρ z + 4 Cρ+ z + 4 singularities
zk ∈Int Γρ

whence Z Z
dz X dz
= 2πi Res(f ; zk ) − .
γρ z2 + 4 singularities C + z2 + 4
ρ
zk ∈Int Γρ

We will now argue that the integral along Cρ+ vanishes in the limit ρ → ∞.

We know that
|dz|
Z Z
dz
2
≤ . (14.5)
Cρ+ z +4 Cρ+ |z 2+ 4|

Using the triangle inequality, we have that on Cρ+ ,

|z 2 + 4| ≥ |z 2 | − 4 = |z|2 − 4 = ρ2 − 4,

whence
1 1
≤ 2 .
+ 4| |z 2
ρ −4
Plugging this into (14.5) and taking into account that the length of the semicircle Cρ+ is
πρ, Z
dz πρ
≤ 2 →0 as ρ → ∞.
Cρ+ z2 +4 ρ −4
Therefore, in the limit ρ → ∞
Z
dz X
= 2πi Res(f ; zk ).
γρ z2 + 4 singularities
zk ∈Int Γρ

84
The function f (z) has poles at z = ±2i, of which only the one at z = 2i lies inside the
closed contour Γρ , for large ρ. Computing the residue there, we find that
 
1 1
Res(f ; 2i) = lim =
z→2i z + 2i 4i
and hence the integral is given by
Z ∞
dx 1 π
I= 2
= 2πi = .
−∞ x + 4 4i 2

• Note that we could have chosen to close the contour using the bottom semicircle Cρ−
parametrised by z(t) = ρ exp(it) for t ∈ [π, 2π]. The same argument shows that in the
limit ρ → ∞ the integral along Cρ− vanishes. It is now the pole at −2i that we have to
take into account and one has that Res(f ; −2i) = −1/4i. Notice, however, that the closed
contour is negatively-oriented, which produces an extra − sign from the residue formula,
in such a way that the final answer is again
−1 π
I = −2πi = .
4i 2
More general result

The technique employed in the calculation of the above integral can be applied in more
general situations.

Let R(x) = P (x)/Q(x) be a rational function of a real variable satisfying the following
two criteria:

• Q(x) 6= 0; and

• deg Q - deg P ≥ 2.

Then the improper integral of R(x) along the real line is given by considering the residues
of the complex rational function R(z) at its singularities in the upper half plane. Being
a rational function the only singularities are either removable singularities or poles and
only poles contribute to the residue. In summary
Z ∞ X
p.v. R(x)dx = 2πi Res(R; zk ). (14.6)
−∞ poles zk
Im(zk )>0

Let us do this in detail. The integral to be computed is

Z ∞
I = p.v. R(x)dx
−∞
Z ρ Z
= lim R(x)dx = lim R(z)dz.
ρ→∞ −ρ ρ→∞ γρ

85
Closing the contour with Cρ+ to Γρ , we have
Z I Z
R(z)dz = R(z)dz − R(z)dz.
γρ Γρ Cρ+

In the limit ρ → ∞, one finds

I X
lim R(z)dz = 2πi Res(R; zk ).
ρ→∞ Γρ poles zk
Im(zk )>0
R
All that remains is to show that the integral Cρ+
R(z)dz vanishes in the limit ρ → ∞.
We can estimate it in the usual way as shown below
|P (z)|
Z Z Z
R(z)dz ≤ |R(z)||dz| = |dz|.
Cρ+ Cρ+ Cρ+ |Q(z)|

Let the degree of the polynomial P (z) be p and that of Q(z) be q, where by hypothesis
q − p ≥ 2.

For large |z| a polynomial P (z) of degree N behaves like |P (z)| ∼ c|z|N for some c.
P (z)
The rational function R(z) = Q(z)
with q = deg Q > deg P = p obeys
c
|R(z)| ≤ ,
|z|q−p
for some constant c independent of |z|. Using this into the estimate of the integral along
Cρ+ and using that the semicircle has length πρ, we have
Z
cπρ
R(z)dz ≤ q−p .
Cρ+ ρ

Since q − p ≥ 2, we have that this goes to zero in the limit ρ → ∞.

Example: Let us compute the following integral

Z ∞
x2
I = p.v. 2 2
dx.
−∞ (x + 1)

The integrand is rational and obeys the two criteria above:

1. it is always finite and

2. the degree of the denominator is 4 whereas that of the numerator is 2.

In order to compute the integral it is enough to compute the residues of the rational
function
z2
f (z) = ,
(z 2 + 1)2

86
at the poles in the upper half-plane. This function has poles of order 2 at the points
z = ±i, of which only z = +i is in the upper half-plane. Hence, from (14.6) we have

z2
 
d
I = 2πi lim
z→i dz (z + i)2
 
2iz −i π
= 2πi lim 3
= 2πi = .
z→i (z + i) 4 2

Home assignment 15

Q.1 Compute the residues of the function

cos z
f (z) =
z 2 (z − π)3

at the isolated singularities.

Q.2 Evaluate the integral

1 − 2z
I
dz
Γ z(z − 1)(z − 3)
along the positively oriented circle of radius 2: |z| = 2, using the Cauchy residue theorem.

Q.3 Using the techniques of contour integration evaluate the following integrals
Z 2π
cos 3θ
i) I = dθ
0 5 − 4 cos θ
Z 2π
dθ
ii) I = , || < 1
0 1 +  cos θ
Z π
dθ
iii) I =
2 − cos θ
Z0 ∞
dx
iv) I = 6
.
−∞ x + 1

Hint: Note that the integral in Q. 3(iii) is only over [0, π], so that we cannot immediately
use the residue theorem. However, in this case we notice that because cos(2π − θ) = cos θ,
we have Z 2π Z π
dθ dθ
= (Show this.)
π 2 − cos θ 0 2 − cos θ
Therefore, Z 2π
1 dθ
I=
2 0 2 − cos θ
1
Hint: In Q.3 (iv), first find the zeros of z 6 + 1, which are (simple ?) poles of z 6 +1
. Then
check which of these poles lie inside the closed contour. Use L’Hospital’s rule to calculate
the residues.

87
Type-III: Improper integrals of rational and trigonometric functions

The next type of integrals whch can be handled by the method of residues are of the kind
Z ∞ Z ∞
p.v. f (x) cos(αx)dx and p.v. f (x) sin(αx)dx,
−∞ −∞

where f (x) is a rational function which is continuous everywhere in the real line and α is
a nonzero real number.

As an example, consider the integral

Z ∞ Z R
cos(3x) cos(3x)
I= 2
dx = lim 2
dx.
−∞ x + 4 R→∞ −R x + 4

[Note that we shall not write p.v. from now on]

From the discussion in the previous section, we are tempted to try to express the integral
over [−R, R] as a complex contour integral, close the contour and use the residue theorem.
Notice, however, that we cannot use the function cos(3z)/(z 2 + 4) because | cos(3z)| is
not bounded for large values of |Im(z)|. Instead we notice that we can write the integral
as the real part of a complex integral I = Re(I0 ), where
R
ei3x
Z
I0 = lim dx.
R→∞ −R x2 + 4

In general, we shall compute integrals of the form

Z ∞
f (x)eiαx dx, (14.7)
−∞

where α is a nonzero real number.

In order to evaluate integrals of this type we shall prove a lemma, which goes by the name
of the Jordan lemma.

Jordan’s Lemma

Let ΓR denote a semicircle in the upper half-plane (Imz ≥ 0) of the complex plane, of
radius R, and centred at the origin.

Let f (z) is analytic in the upper half-plane except for a finite number of poles and f (z)
does not have a singularity on the real line, and f (z) tends uniformly to zero with respect
to arg z as |z| → ∞ when arg z lies in the interval

0 ≤ arg z ≤ π.

88
Then, if α is a real and α > 0, then
Z
0
lim IR ≡ lim eiαz f (z 0 )dz 0 = 0.
R→∞ R→∞ ΓR

Proof: In polar coordinates

Z π
IR = i f (Reiθ )eiαR cos θ−αR sin θ+iθ Rdθ.
0

Since, by hypothesis, f (z) tends uniformly to zero as R → ∞, we must have

|f (Reiθ )| < (R)

where (R) is some positive number, which depends on R only and which tends to zero
as R → ∞. Therefore,
Z π Z π/2
−αR sin θ
|IR | ≤ (R) · R e dθ = 2(R)R e−αR sin θ dθ,
0 0
π
where in the last step we have used the symmetry of sin θ about θ = 2

y
ΓR

-R R x

However, notice that

2θ
sin θ ≥ when 0 ≤ θ ≤ π/2.
π
Hence,
Z π/2
|IR | ≤ 2(R) · R e−(2αR θ/π)
dθ
0
1 − e−αR π(R)
= 2(R) · R = (1 − e−αR ),
2αR/π α
and therefore,
π
lim |IR | ≤ (R).
R→∞ α
Now, since
(R) → 0 as R → ∞.

89
we have
lim |IR | = 0.
R→∞

and hence Z
0
lim IR = lim eiαz f (z 0 )dz 0 = 0.
R→∞ R→∞ ΓR
which proves the lemma.

If α < 0, the lemma remains valid, provided the semicircle ΓR is taken in the lower half
of the complex plane and provided f (z) tends uniformly to zero for π ≤ arg z ≤ 2π.

Therefore, we see that the improper integral (14.7) is given by considering the residues
of the function h(z) = f (z)eiαz at the poles in the upper half-plane (if α > 0) or lower
half-plane (if α < 0).

Considering the figure above, we see that

I Z R Z X
iαx
h(z)dz = f (x)e dx + f (z)eiαz dz = 2πi Res(h; zk ), α > 0,
C −R ΓR poles zk
Im(zk )>0

where C is the closed contour composed of the straight line segment on the real axis:
−R ≤ x ≤ R and the semicircular contour ΓR in the upper half-plane.

In the limit R → ∞, we get

Z ∞ X
f (x)eiαx dx + lim IR = 2πi Res(h; zk ) α > 0, upper half-plane
−∞ R→∞
poles zk
Im(zk )>0

and using Jordan’s lemma we get

Z ∞ X
f (x)eiαx dx = 2πi Res(h; zk ) α > 0, upper half-plane
−∞ poles zk
Im(zk )>0

Similarly, for α < 0 we have

Z ∞ X
f (x)eiαx dx = −2πi Res(h; zk ) α < 0. lower half-plane
−∞ poles zk
Im(zk )<0

Example: Z ∞
cos mx
I= dx, m > 0.
−∞ x 2 + a2
Here
eimz
h(z) = = f (z)eimz .
z 2 + a2

90
Therefore, we have (see figure)
Z R
eimx eimz
I Z X
h(z)dz = 2 2
dx + 2 2
dz = 2πi Res(h; zk )
C −R x + a ΓR z + a poles z k
Im(zk )>0

y
ΓR

-R R x

We also see that

1
|f (z)| = .
z 2 + a2
Now,
|z 2 + a2 | ≥ |z 2 | − a2 = |z|2 − a2 = R2 − a2 ,

whence,
1 1
|f (z)| = ≤ 2 → 0 as R → ∞.
z2 +a 2 R − a2
Therefore,
lim f (z) = 0, uniformly.
|z|→∞

Then using Jordan’s lemma we see that

eimz
Z
lim IR = 0, where IR = dz.
R→∞ ΓR z 2 + a2
Hence,
∞
eimx
Z X
dx = 2πi Res(h; zk ).
−∞ x 2 + a2 poles z k
Im(zk )>0

The function h(z) has simple poles at z = ±ia, of which only the pole at z = +ia
contributes.
eimz
Res(h, ia) = lim (z − ia)
z→ia (z + ia)(z − ia)
eimz
e−ma
= lim = .
z→ia z + ia 2ia

91
Therefore,
∞
eimx e−ma
Z
π
I0 = 2 2
dx = 2πi × = e−ma .
−∞ x +a 2ia a
Thus, we have
Z ∞
cos mx π
I= 2 2
dx = Re(I0 ) = e−ma .
−∞ x +a a

Type-IV: Improper integrals with poles

We have considered contour integrals of functions in two different cases.

In the first case, the integral was taken along a curve inside and on which there were no
singularities of the function. This led us to Cauchy’s theorem.

In the second case, the boundary curve enclosed, but did not go through, singularities of
the function and this led us to the theorem of residues.

There remains to be considered the case when the path of integration actually passes
through a singularity of the integrand.

In this case, strictly speaking, the integral does not exist.

To give it meaning, one must choose a path that circumvents the singularity. We shall
show that the result of the integration will then depend on how the path is chosen to
avoid singularity. We shall restrict ourselves to the case where the singularity is a simple
pole on the real axis.

Suppose that we want to compute the principal value integral

Z ∞
sin x
I = p.v. dx.
−∞ x

This integral should converge: the singularity at x = 0 is removable, so that the integrand
is continuous for all x. Following the discussions in the previous section, we would write
Z ∞ ix
e
I = Im(I0 ) where I0 = p.v. dx, (14.8)
−∞ x

and compute I0 using Jordan’s lemma. However, notice that now the integrand of I0 has a
pole at x = 0. Until now we have always assumed that integrands have no poles along the
contour, so the method developed until now are not immediately applicable to perform
the above integral. We, therefore, need to make sense out of integrals whose integrands
are not continuous everywhere in the region of integration.

92
Let f (x) be a function of a real variable, which is continuous everywhere on the real line
except at the point x0 , we define the principal value integral by
Z ∞  Z x0 − Z R 
p.v. f (x)dx ≡ lim f (x)dx + f (x)dx , (14.9)
−∞ R→∞ −R x0 +
→+0

provided the limits R → ∞ and  → +0 exists independently. Note that  approaches 0

from above; that is,  > 0 as we take the limit.

It turns out that principal value integrals of this type can often be evaluated using the
residue theorem. The residue theorem applies to closed contour, so in computing a prin-
cipal value integral we need to close the contour, not just R to −R as in the previous
section, but also x0 −  to x0 + . One way to do this is to consider a small semicircle S ,
of radius  around the singular point x0 , as in the figure.

S

x0
x0 −  x0 + 

Figure 14.6: Indenting the contour around a singularity

Because we are interested in the limit  → +0, we will have to consider the integral
Z
lim f (z)dz.
→+0 S

When the singularity of f (z) at z = x0 is a simple pole, this integral can be evaluated
using the following result.

A


θ1 − θ0

x0

Figure 14.7: A small circular arc

Let f (z) has a simple pole at z = x0 and let A be the circular arc in Fig. 14.7,
parametrised by z(θ) = x0 +  exp(iθ) with θ0 ≤ θ ≤ θ1 . Then
Z
lim f (z)dz = i(θ1 − θ0 ) Res(f ; x0 ).
→+0 A

93
Therefore, for the semicircle S in Fig 14.6, we have
Z
lim f (z)dz = −iπ Res(f ; x0 ). (14.10)
→+0 S

Let us prove this result. Since f (z) has a simple pole at x0 , its Laurent expansion in a
punctured disk 0 < |z − x0 | < R has the form
∞
a−1 X
f (z) = + ak (z − x0 )k ,
z − x0 k=0

where ∞
X
g(z) ≡ ak (z − x0 )k
k=0

defines an analytic function in the disk |z − x0 | < R. Now, let 0 <  < R and consider
the integral Z Z Z
dz
f (z)dz = a−1 + g(z)dz.
A A z − x0 A

Because g(z) is analytic it is, in particular, bounded on some neighbourhood of x0 , so

that |g(z)| ≤ M for some M and all |z − x0 | < R. Then we can estimate its integral
Z Z
g(z)dz ≤ |g(z)||dz| ≤ M `(A ) = M (θ1 − θ0 ),
A A

whence Z
lim g(z)dz = 0.
→+0 A

On the other hand,

θ1 θ1
ieiθ
Z Z Z
dz
= dθ = i dθ = i(θ1 − θ0 ).
A z − x0 θ0 eiθ θ0

Therefore, Z
lim f (z)dz = i(θ1 − θ0 )a−1 + 0 = i(θ1 − θ0 ) Res(f ; x0 ).
→+0 A

Having discussed the basic theory, let us go back to the original problem: the computation
of the integral I0 given in (14.8):
− R
eix eix
Z Z 
I0 = lim dx + dx , (14.11)
R→∞ −R x  x
→+0

which for finite R and nonzero  can be understood as a contour integral in the complex
plane along the subset of the real axis consisting of the intervals [−R, −] and [, R]. In
order to use the residue theorem we must close this contour. The Jordan lemma forces
us to join R and −R via a large semicircle ΓR of radius R in the upper half-plane. In

94
 ΓR

S

0
-R −  R x

Figure 14.8: The contour in the calculation of I0 in (14.8).

order to join − and  we choose a small semicircle S also in the upper half-plane. The
resulting closed contour is shown in Fig. 14.8.

Because the function is analytic on and inside the contour, the Cauchy Integral Theorem
says that the contour integral vanishes (Cauchy residue theorem also tells the same).
Splitting this contour integral into its different pieces, we have
 Z − Z Z R Z  iz
e
+ + + dz = 0, (14.12)
−R S  ΓR z

which remains true in the limits R → ∞ and  → 0. By the Jordan’s lemma, the integral
along ΓR vanishes in the limit R → ∞, whence using (14.10),
Z ∞ ix
e
I0 = p.v. dx
−∞ x
 Z − ix Z R ix 
e e
= lim dx + dx
R→∞ −R x  x
→+0
Z iz
eiz
Z
e
= − lim dz = lim dz
→+0 S z →0 −S z
 

= iπ Res(0) = iπ,

since the residue of eiz /z at z = 0 is equal to 1. Therefore, we have

Z ∞
sin x
p.v. dx = Im(iπ) = π.
−∞ x

Now, let us make an appropriate change of variables in the first integral in (14.12), x →
−x, and combine with the third integral, we get from (14.12)
Z R ix
e − e−ix
lim dx = iπ.
→+0  x
R→∞

95
Therefore,
Z ∞
sin x π
dx = .
0 x 2

In general, let us consider the integral

Z ∞
f (x)
p.v. dx,
−∞ x − x0
where x0 is on the path of integration, i.e., on the real axis and f (x) is analytic at x0 .
This means the singularity is a simple pole on the real axis.

As discussed above, we may indent the path and follow the contour which encircle, rather
than goes through, the singularity at x0 . The integral along this indented contour can be
written as the sum of three integrals.

S

x0
x0 −  x0 + 

Z Z − Z R  Z
f (z) f (x) f (x) f (z)
dz = lim dx + dx + lim dz (14.13)
Ω z − x0 R→∞ −R x − x0  x − x0 →+0 S z − x0

→+0

We have seen above that the integral over S can be evaluated to be

Z
f (z)
lim dz = −iπf (x0 ) (14.14)
→+0 S z − x0


f (z)
[Note that the residue of z−x0
at z = x0 is f (x0 ). ]

Hence,
Z Z ∞
f (z) f (x)
dz = p.v. dx − iπf (x0 ). (14.15)
Ω z − x0 −∞ x − x0

On the other hand, if S had been taken below the singularity instead of above it, (14.15)
would have been replaced by
Z Z ∞
f (z) f (x)
dz = p.v. dx + iπf (x0 ). (14.16)
Ω z − x0 −∞ x − x0

Equations (14.15) and (14.16) display the very important point that the value of integrals
such as those considered here depends upon the path chosen to circumvent the singularity.
In applications, the physical situation will always determine the path that must be chosen.

96
An equivalent way of getting around the singularity

There is another equivalent way of getting around the difficulty of having a pole on the
path of integration.

Instead of lowering the contour, say, we can also give the singularity an infinitesimally
small, positive, imaginary part. Similarly, instead of raising the contour, we can give the
singularity an infinitesimally small, negative, imaginary part. That is to say, we raise or
lower the singularity.
x0 x0
x axis
 

(a) (b)

Figure 14.9: Two equivalent ways of avoiding the singularity

To see this, suppose for simplicity that f (z) is analytic in an environment of the real axis.
Let us integrate the function
f (z)
,
z − x0
where x0 is real, along a path that goes along the real axis and circumvents x0 from below,
as in the Fig. (a) above.

According to the results derived above, we have

Z Z ∞
f (z) f (x)
dz = p.v. dx + iπf (x0 ).
Ω z − x0 −∞ x − x0

On the other hand, the contour of Fig. 14.9(a) is completely equivalent to the contour of
Fig. 14.9(b), which is obtained by stretching out the former contour below the singularity.
Hence, we may write
Z Z +∞−i
f (z) f (z)
dz = lim dz. (14.17)
Ω z − x0 →+0 −∞−i z − x0
We now perform the change of integration variable

z 0 = z + i,

in the integral on the R.H.S. of (14.17), and obtain

Z ∞
f (z 0 − i)
Z
f (z)
dz = lim 0 − x − i
dz 0
z − x 0 →+0 z 0
Ω
Z−∞
∞
f (x)
= lim dx. (14.18)
→+0 −∞ x − x0 − i

97
The foregoing relation expresses the equivalence between the prescription of lowering the
contour around a singularity and that of raising the singularity, which now appears at
z = x0 + i.

Comparing (14.16) and (14.18), we find

Z ∞ Z ∞
f (x) f (x)
p.v. dx + iπf (x0 ) = lim dx. (14.19)
−∞ x − x0 →+0 −∞ x − x0 − i

Similarly, one can show that raising the contour around a singularity is equivalent to
giving the singularity an infinitesimally small, negative, imaginary part.
Z ∞ Z ∞
f (x) f (x)
p.v. dx − iπf (x0 ) = lim dx. (14.20)
−∞ x − x0 →+0 −∞ x − x0 + i

Gauss’ Integral

Let us show that Z ∞

2
I= e−(x+ib) dx
−∞
is independent of b and deduce that
Z ∞
2 √ 2
e−x cos 2bx dx = πe−b .
−∞

Let us take z = x + ib and assume that b > 0.

Therefore, Z +∞+ib
2
I= e−z dz.
−∞+ib

2
The function e−z can be integrated along the contour shown in the figure below. Since
2
e−z is entire, it has no poles inside the contour.

Therefore, we have
I Z Z Z Z 
−z 2 2
0= e dz = + + + e−z dz. (14.21)
C C1 C2 C3 C4
2
e−z dz,
R
Let Jk = Ck
k = 1, 2, 3, 4

Then,
Z Z b
−z 2 2
|J2 | = e dz = e−(R+iy) idy
C2 0
Z b
2 −2iRy+y 2
≤ e−R dy
0
Z b
2 2 2 2
= e−R ey dy ≤ e−R · eb · b
0

98
 y

C3 : z = x + ib
dz = dx

C4 : z = −R + iy C2 : z = R + iy
dz = idy b dz = idy

−R C1 : z = x R x
dz = dx

Thus, for fixed b, J2 → 0 as R → ∞. Similarly, J4 → 0.

Therefore, as R → ∞, we have from (14.21),

Z ∞ Z −∞+ib
−z 2 2
e dz + e−z dz = 0.
−∞ +∞+ib

Or, Z ∞ Z −∞
−x2 2
e dx + e−(x+ib) dx = 0.
−∞ +∞

Thus, we have
Z ∞ Z ∞
−(x+ib)2 2
e dx = e−x dx, (14.22)
−∞ −∞

Now,
Z ∞ Z ∞ 1/2  Z ∞ 1/2
−x2 −x2 −y 2
e dx = e dx e dy .
−∞ −∞ −∞

Or,
Z ∞ Z ∞ Z ∞ 1/2
−x2 −(x2 +y 2 )
e dx = e dxdy
−∞ −∞ −∞
 Z 2π Z ∞ 1/2
1 −r2 2
= e d(r )dθ
2 0 0
√
= π

Hence,
Z ∞ √
2
e−(x+ib) dx = π. (14.23)
−∞

99
From (14.23), we have
Z ∞ √
b2 2
e e−x [cos 2bx − i sin 2bx]dx = π.
−∞

Therefore,
Z ∞ √
2 2
e−x cos 2bx dx = πe−b . (14.24)
−∞

Home assignment 16

Q.1 Using the techniques of contour integration evaluate

Z ∞
x sin x
(a) 2 2
dx (a > 0).
−∞ x + a

Z ∞
x sin πx
(b) dx.
−∞ x2+ 2x + 5

Q.2 Show that

∞
Z r
iax−bx2 π −(a2 /4b)
I= e dx = e a, b real, b > 0.
−∞ b

ia
[Hint: Take z = x − 2b
and write I as
ia
Z +∞− 2b
−(a2 /4b) 2
I=e e−bz dz.
ia
−∞− 2b

Then use the same technique as discussed in the study material. However, take the
contour in the lower half-plane.]

15 Evaluation of an integral involving a multivalued

function

We shall evaluate the integral

∞
xp−1
Z
I= dx 0 < p < 2. (15.1)
0 x2 + 1

100
Let us consider the contour integral
z p−1
Z
0
I = dz, (15.2)
C z2 + 1

where C is the path shown in the figure below. When p is not an integer (p 6= 1), the
integrand has a branch point at z = 0. We take the cut along the positive real axis. (More
discussions on branch points and branch cuts can be found later in section 18.)

The path C consists of a small circle γ of radius  surrounding the branch point, of a large
circle Γ of radius r, and of two straight lines L1 and L2 lying respectively just above and
just below the branch cut.
y

γ L1

x
L2

r C

Thus, C does not cross the cut and the integrand in (15.2) is single-valued within this
contour. There remains to choose a particular branch of the function. We choose the
branch of the function z p−1 as

z p−1 = |z|p−1 ei(p−1)θ 0 ≤ θ < 2π. (15.3)

We have specified everything properly and we can now proceed with the evaluation of I 0 .
We will first show that the circles γ and Γ do not contribute to I 0 as  → 0 and r → ∞.

Let the parametric equation of either one of these circles be

z = ρeiθ

Then for either circles

2π
z p−1 ρp−1 ei(p−1)θ (iρeiθ )
Z Z
dz = dθ (15.4)
γ or Γ z2 + 1 0 ρ2 ei2θ + 1

101
and thus, when either ρ → 0 (circle γ) or when ρ → ∞ (circle Γ), (15.4) tends to zero.

There remains the evaluation of the contribution to I 0 from the integrations along L1 and
L2 . Because a branch cut separates L1 from L2 , the values of the integrand along these
lines will not be the same.

Along L1 , z = x, and therefore

z p−1 = xp−1 along L1 .

Along L2 , z = xei2π , and therefore

z p−1 = xp−1 ei2π(p−1) along L2 .

Thus, we have (as  → 0, r → ∞)

Z ∞
z p−1
Z
1
2
dz = [xp−1 − xp−1 ei2π(p−1) ] 2 dx
C z +1 0 x +1
Z ∞ p−1
iπ(p−1) −iπ(p−1) iπ(p−1) x
= e [e −e ] dx
0 x2 + 1
Z ∞ p−1
iπ(p−1) x
= e [−2i sin π(p − 1)] dx
0 x2 + 1
Z ∞ p−1
iπ(p−1) x
= 2ie sin πp dx. (15.5)
0 x2 + 1

On the other hand, the contour integral in (15.5) is equal to 2πi times the sum of the
residues of the integrand at the simple poles z = ±i.

Since

ip−1 = eiπ(p−1)/2
(−i)p−1 = ei3π(p−1)/2

we have
z p−1
 iπ(p−1)/2
ei3π(p−1)/2
Z 
e
dz = 2πi − (15.6)
C z2 + 1 2i 2i

Combining (15.5) and (15.6) we obtain the result

Z ∞ p−1  
x π πp
2
dx = csc . (15.7)
0 x +1 2 2

Integrals of rational functions on (0, ∞)

102
The next type of integrals which can be tackled using the residue theorem are integrals
of rational functions of the form Z ∞
R(x)dx,
0

where R(x) is continuous for x ≥ 0. Of course, if R(x) were an even function, i.e.,
R∞ R∞
R(−x) = R(x), then we would have 0 R(x)dx = 21 −∞ R(x)dx, and we could use the
method discussed previously. However, for general integrands, this does not work and we
have to do something different.

Let R(x) = P (x)/Q(x) be a rational function of a real variable satisfying the following
two conditions
• Q(x) 6= 0; and
• deg Q − deg P ≥ 2.

Further let f (z) = log(z)R(z) with the branch of the logarithm chosen to be analytic at
the poles {zk } of R; for example, we can choose the branch Log0 (z) which has the cut
along the positive real axis, since Q(x) has no zeros there. Then we have the general
result
Z ∞ X
R(x)dx = − Res(f ; zk ), for f (z) = log(z)R(z). (15.8)
0 poles zk

Let us take an example

Z ∞
dx
I= (15.9)
0 1 + x3

The integrand is not even, so we cannot extend it to −∞. Consider the integral
Z
0 log(z)dz
I = 3
,
C 1+z

where C is the path shown in the figure below. It consists of a small circle γ of radius r
surrounding the branch point, of a large circle Γ of radius R, and of two straight lines L1
and L2 lying respectively just above and just below the branch cut.

We interpret log(z) as ln(x) just above the cut (in L1 ) and we interpret log(z) as ln(x)+2πi
just below the cut (in L2 ). [Note that here we are using the notation ln for real logarithm]

The function f (z) = log(z)/(z 3 + 1) has three simple poles at the cube roots of -1, which
are eiπ/3 , eiπ and e5iπ/3 . To find the residues we can use the p(z0 )/q 0 (z0 ) method, with

103
 y

γ ×
r L1
×
x
L2
×
R C

p(z) = log(z) and q(z) = z 3 + 1. (Strictly speaking we should write p(z) = Log0 (z) since
we have chosen the branch such that 0 ≤ Arg0 (z) < 2π.) This gives the values
 √ 
iπ/3 iπ −1 3
Res(f ; e ) = − i
9 2 2
iπ
Res(f ; eiπ ) =
3 √ 

5iπ/3 5iπ −1 3
Res(f ; e )= + i
9 2 2
−2π
The sum of these is √ .
3 3

Hence, by the residue theorem, we have

Z R Z R  
−2π
Z Z
ln(x) ln(x) + 2πi
3
dx − dx + f (z)dz + f (z)dz = 2πi √ . (15.10)
r 1+x r 1 + x3 γ Γ 3 3
Now, for a point z on Γ, | log(z)| is at most ln R + 2π for R ≥ 1. Also, we have on Γ,

|z 3 + 1| ≥ |z 3 | − 1 = |z|3 − 1 = R3 − 1,

Hence,
2πR ln R + 4π 2 R
Z
log(z)
dz ≤ → 0 as R → ∞.
Γ z3 + 1 R3 − 1

By a similar reasoning, for a point z ∈ γ, | log(z)| ≤ | ln r| + 2π while |z 3 + 1| is close to

1 if r is very small. Then, we have
Z
log(z)
3+1
dz ≤ 2πr ln r + 4π 2 r → 0 as r → 0,
γ z

104
since r ln r → 0 as r → 0.

So, taking limits as R → ∞ and r → 0 in (15.10), we get

Z ∞
dx 4π 2 i
−2πi = − √ .
0 x3 + 1 3 3
Hence, we have
Z ∞
dx 2π
3
= √ .
0 1+x 3 3

Let us take another example of an integral using the same contour as in the
previous two examples

Evaluate ∞
(ln x)2
Z
dx
0 1 + x2
by integrating
(log z)3
f (z) =
1 + z2
around the contour as shown below
y

×i
γ L1
r
x
L2
×−i
R C

The calculations are similar to the previous example. One can prove in a similar fashion
that the integrals of f (z) over both the inner and outer circular arcs tend to 0 as r → 0
and R → ∞, respectively [Show this].

105
The function f (z) has poles at ±i with residues

(log i)3
Res(f ; i) =
2i
(iπ/2)3 π3
= =−
2i 16
(log(−i))3
Res(f ; −i) =
−2i
(i3π/2)3 27π 3
= =
−2i 16
Therefore, using the residue theorem and taking the limits r → 0 nd R → ∞, we get
Z ∞ Z ∞
(ln x)3 (ln x + 2πi)3 13π 3 13π 4
dx − dx = 2πi × = i. (15.11)
0 1 + x2 0 1 + x2 8 4

Expanding the numerator of the second integral and taking the imaginary parts of both
sides we get
∞ ∞
(ln x)2
Z Z
dx 13 4
−6π dx + 8π 3 = π . (15.12)
0 1 + x2 0 1 + x2 4

The second integral on the L.H.S. is already known to be π2 . Substituting this result into
(15.12) gives
∞
(ln x)2 π3
Z
dx = .
0 1 + x2 8

Evaluation of integrals: exponential forms

With exponential or hyperbolic functions present in the integrand, instead of a general

overall prescription, the contour must be chosen to fit the specific integral.

As an example, let us evaluate

∞
eax
Z
I= dx. 0 < a < 1.
−∞ 1 + ex

The limits on a are necessary (and sufficient) to prevent the integral from diverging as
x → ±∞.

The integral may be handled by replacing the real variable x by the complex variable z
and considering the integral
eaz
I
dz
1 + ez
around the contour shown below.

106
 y

−R + 2πi R + 2πi

−R R x

If we take the limit as R → ∞, the real axis leads to the desired integral. The return
path along y = 2π is chosen to leave the denominator of the integral invariant, at the
same time introducing a constant factor ei2πa in the numerator. We have in the complex
plane,
R Z 2π a(R+iy)
eaz eax
I Z
e
dz = lim dx + idy
1 + ez R→∞ −R 1 + e
x
0 1 + e(R+iy)
Z R Z 2π a(−R+iy)
eax

ia2π e
− e dx − idy (15.13)
−R 1 + e
x
0 1 + e−R+iy

Now, we have
2π 2π
e±aR eiay e±aR eiay
Z Z
± idy ≤ idy
0 1 + e±R+iy 0 1 + e±R+iy
2π
e±aR
Z
≤ dy.
0 |1 + e±R+iy |

In the limit R → ∞, on the vertical line on left of the contour, the minus signs are chosen
in the integrand so that the numerator tends to zero (a > 0) and the denominator to unity,
taking the integrand tending to zero. On the line on right of the contour the integrand
tends to zero since the numerator goes as eaR and the denominator as eR and 0 < a < 1.
Thus, the two vertical sections (0 ≤ y ≤ 2π) vanish (exponentially) as R → ∞.

Therefore, we have
R R
eaz eax eax
I Z Z 
dz = lim dx − eia2π dx
1 + ez R→∞ −R 1 + e x
−R 1 + ex
Z ∞
i2πa eax
= (1 − e ) x
dx (15.14)
−∞ 1 + e

107
We have a pole when

ez = −1. (15.15)

Equation (15.15) is satisfied at z = iπ, which is inside the contour. By a Laurent expansion
in powers of (z − iπ) the pole is seen to be a simple pole. The residue is given by
eaz
Res(f ; iπ) = −eiπa , where f (z) = .
1 + ez
z − iπ (z − iπ)2
  
z z−iπ iπ z−iπ
1+e =1+e e =1−e = −(z − iπ) 1 + + + ··· .
2! 3!
Then, applying the residue theorem, we have
Z ∞
i2πa eax
(1 − e ) x
dx = 2πi(−eiπa ). (15.16)
−∞ 1 + e

Simplifying, we get
∞
eax
Z
π
dx = , 0 < a < 1.
−∞ 1 + ex sin aπ

Home assignment 17

Q.1 Using the techniques developed in Section 15, pages 105-106, show that
Z ∞
ln x π ln 2
2
dx =
0 x +4 4

Q.2 (a) Evaluate the integral Z ∞

ln x
dx
0 1 + x4
(log z)3
by integrating the function f (z) = 1+z 4
using the techniques developed in pages 105 -
√
−π 2 2
106. [Ans: 16
]

(b) Using the same contour integration technique as in Q. 2(a), prove that
Z ∞ √
(ln x)2 3π 3 2
dx =
0 1 + x4 64

16 Summation of infinite series

In this section we shall discuss a beautiful application of the theory of residues to the
computation of infinite sums.

108
We shall use contour integration in order to calculate sums like the following one:
∞
X 1
2
. (16.1)
n=1
n

The idea is to exhibit this sum as part of the right-hand side of the Cauchy Residue
Theorem. For this we need a function F (z) which has only simple poles at the integers
and whose residue is 1 there. We already met a function which has an infinite number
of poles which are integrally spaced. The function cot z has simple poles for z = nπ, n =
0, ±1, ±2, · · · with residues equal to 1. Therefore, the function F (z) = π cot(πz) has
simple poles at z = n, n an integer, and the residue is still 1:
π cos(πz) π cos(πz)
Res(F ; n) = lim 0
= lim = 1.
z→n (sin(πz)) z→n π cos(πz)

Now, let R(z) = P (Z)/Q(z) be any rational function such that deg Q − deg P ≥ 2.
Consider the function f (z) = π cot(πz)R(z) and let us integrate this along the contour
ΓN , for N a positive integer, defined as the positively oriented square with vertices (N +
1
2
)(1 + i), (N + 12 )(−1 + i), (N + 12 )(−1 − i), (N + 12 )(1 − i), as shown in Figure 16.10.
Notice that the contour misses the poles of π cot(πz). Assuming that N is taken to be
large enough, and since R(z) has a finite number of poles, one can also guarantee that
the contour will miss the poles of R(z).

(N + 12 )(−1 + i) (N + 12 )(1 + i)

N
N +1

(N + 21 )(−1 − i) (N + 12 )(1 − i)

Figure 16.10: The contour ΓN .

Let us compute the integral of the function f (z) along this contour,
Z
π cot(πz)R(z)dz,
ΓN

109
in two ways. On the one hand, we can use the residue theorem to say that the integral
will be (2πi) times the sum of the residues of the poles of f (z). These poles are of two
types: the poles of R(z) and the poles of π cot(πz), which occur at the integers.

Let us assume for simplicity that R(z) has no poles at integer values of z, so that the
poles of R(z) and π cot(πz) do not coincide. Therefore, we see that
Z N
!
X X
π cot(πz)R(z)dz = 2πi Res(f ; n) + Res(f ; zk ) .
ΓN n=−N poles zk of R
inside ΓN

The residue of f (z) at z = n is easy to compute. Since by assumption R(z) is analytic

there and π cot(πz) has a simple pole with residue 1, we see that around z = n, we have
 
1 R(z)
f (z) = R(z)π cot(πz) = R(z) + ··· = + h(z),
z−n z−n
where h(z) is analytic at z = n. Therefore,

Res(f ; n) = lim [(z − n)f (z)] = R(n) + 0,

z→n

and as a result
Z N
!
X X
π cot(πz)R(z)dz = 2πi R(n) + Res(f ; zk ) . (16.2)
ΓN n=−N poles zk of R
inside ΓN

On the other hand, we can estimate the integral for large enough N as follows. First of
all because of the condition on R(z), we have that for large |z|,
c
|R(z)| ≤ .
|z|2
Similarly, it can be shown that the function π cot(πz) is bounded along the contour, so
that |π cot(πz)| ≤ K for some K independent of N .

To prove this we will consider the parts of ΓN where y > 21 , − 12 ≤ y ≤ 1

2
and y < − 21 .

Case 1: y > 12 . Let z = x + iy, then

eπiz + e−πiz eπix−πy + e−πix+πy |eπix−πy | + |e−πix+πy |
| cot(πz)| = = ≤
eπiz − e−πiz eπix−πy − e−πix+πy |e−πix+πy | − |eπix−πy |
e−πy + eπy 1 + e−2πy 1 + e−π
= πy = ≤ = K1
e − e−πy 1 − e−2πy 1 − e−π

Case 2: y < − 12 . Here, similar to Case 1, we have

|eπix−πy | + |e−πix+πy | e−πy + eπy 1 + e2πy 1 + e−π
| cot(πz)| ≤ = = ≤ = K1
|eπix−πy | − |e−πix+πy | e−πy − eπy 1 − e2πy 1 − e−π

110
Case 3: − 12 ≤ y ≤ 12 . In this case let us consider z = N + 12 + iy. Then we have
 
1 π
| cot(πz)| = cot(π(N + + iy)) = | cot(π/2 + πiy)| = | tanh(πy)| ≤ tanh = K2
2 2
Similarly, if z = −N − 12 + iy, we have
 
1 π
| cot(πz)| = cot(π(−N − + iy)) = | tanh(πy)| ≤ tanh = K2 .
2 2
So, if we choose K such that K > max{K1 , K2 }, then we have | cot(πz)| < K on ΓN
with a K independent of N .

Since the length of the contour ΓN is given by 4(2N + 1), we have the following result
Z
Kc
π cot(πz)R(z)dz ≤ 4(2N + 1),
ΓN (N + 21 )2
which vanishes in the limit N → ∞. Therefore, taking the limit N → ∞ of equation
(16.2) and using that the left-hand side vanishes, one finds
∞
X X
R(n) = − Res(f ; zk ).
n=−∞ poles zk of R

More generally, if R(z) does have some poles for integer values of z, then we have to take
care not to over-count these poles in the sum of the residues. We will count them as poles
of R(z) and not as poles of π cot(πz) and the same argument as above yields the general
formula:
∞
X X
R(n) = − Res(f ; zk ), for f (z) = π cot(πz)R(z). (16.3)
n=−∞ poles
n6=zk zk of R

Let us now compute the sum (16.1). Notice that

∞ ∞
X 1 1 X 1
= .
n=1
n2 2 n=−∞ n2
n6=0

Let us consider the function R(z) = 1/z 2 . Hence, by (16.3)

∞
X 1 1
2
= − Res(f ; 0).
n=1
n 2

Now, we know that cot(z) has a simple pole at z = 0 because tan(z) has a simple zero
there. Hence, if the Laurent expansion of cot(z) around z = 0 is
b1
cot(z) = + a0 + a1 z + a2 z 2 + a3 z 3 + . . . ,
z

111
then
z2 z4 z6 z8 z3 z5 z7 z9
    
b1 2 3
1− + − + −. . . = z− + − + −. . . · +a0 +a1 z+a2 z +a3 z +. . . ,
2! 4! 6! 8! 3! 5! 7! 9! z
If we multiply, collect terms and then equate coefficients, we find that b1 = 1, a0 = 0,
1
a1 = − 31 , a2 = 0, a3 = − 45 .
π cot(πz)
Thus, the Laurent expansion of z2
around z = 0 is given by
 
1 πz π3 z3
π πz − − 45 + . . .
π cot(πz) 3 1 π2 π4z
= = − − + .... (16.4)
z2 z2 z 3 3z 45
π cot(πz) −π 2
Hence, the residue of z2
at z = 0 is 3
. Therefore, we have
∞
1 −π 2 π2
 
X 1
= − = .
n=1
n2 2 3 6

Similarly, we can compute the sum

∞
X 1 1
4
= − Res(f ; 0),
n=1
n 2
π cot(πz)
where f (z) = z4
, whose Laurent series about z = 0 can be read off from (16.4) above:
π cot(πz) 1 π2 π4
= − − + ...,
z4 z 5 3z 3 45z
whence
∞
X 1 π4
= .
n=1
n4 90

Infinite alternating sums

The techniques above can be extended to the computation of infinite alternating sums of
the form ∞
X
(−1)n R(n),
n=−∞

where R(z) = P (z)/Q(z) is a rational function with degQ − degP ≥ 2. Now what is
needed is a function G(z) which has a simple pole at z = n, for n an integer and whose
residue there is (−1)n . We claim that this function is π csc(πz).

The poles of csc(πz) are simple at z = n, for n an integer. The residues of G(z) = π csc(πz)
at z = n, n = 0, ±1, ±2, . . . , are
 
z−n
Res(G; n) = lim (z − n)π csc(πz) = lim π = (−1)n .
z→n z→n sin(πz)

112
The trigonometric identity

(csc(πz))2 = 1 + (cot(πz))2 ,

implies that csc(πz) is also bounded along the contour ΓN , with a bound which is inde-
pendent of N just like for cot(πz).

Just as was done above for the cotangent function, we can prove that the integral of the
function f (z) = π csc(πz)R(z) along ΓN vanishes in the limit N → ∞. Therefore, we can
conclude that
∞
X X
(−1)n R(n) = − Res(f ; zk ), forf (z) = π csc(πz)R(z). (16.5)
n=−∞ poles
n6=zk zk of R

As an example, let us compute the alternating sums

∞ ∞
X (−1)n X (−1)n
S1 = and S2 = .
n=1
n2 n=1
n4
For the first sum we have that
1
S1 = − Res(f ; 0),
2
where f (z) = π csc(πz)/z 2 .

Now, the Laurent expansion of π csc(πz) about z = 0 is given by

1 π 2 z 7π 4 z 3
π csc(πz) = + + + .... (16.6)
z 6 360

Therefore, we can read off the Laurent expansion of f (z) = π csc(πz)/z 2 about z = 0
from (16.6):
1 π 2 7π 4 z
f (z) = + + + ....
z 3 6z 360
Hence, the residue of f (z) at z = 0 is π 2 /6 and the sum
π2
S1 = − .
12

For the second sum we also have that

1
S2 = − Res(f ; 0),
2
where the function f (z) = π csc(πz)/z 4 has now a Laurent series
1 π2 7π 4
f (z) = + + + ...,
z 5 6z 3 360z
whence the residue is 7π 4 /360 and the sum
7π 4
S2 = − .
720

113
 Home assignment 18

Q.1 Using the techniques developed in Section 16 (in particular, equation (16.3) ), prove
that ∞
X 1 π
= coth(πa)
n=−∞
n2 +a 2 a
where a > 0. Hence, show that
∞
X 1 π 1
= coth(πa) − 2 .
n=1
n2 +a 2 2a 2a

Q.2 Show that

∞
X 1 2
1 2 = π .
n=−∞
(n − 2
)

17 Analytic Continuation

Suppose that a function f (z) is analytic within a certain region D. Which subsets of
D have the property that, specifying the values of f (z) over these subsets only, f (z) is
determined throughout the whole of D ? In this connection let us state a very important
theorem.

Theorem. Let f1 (z) and f2 (z) be two functions of the complex variable z that are
analytic within a region D. If the two functions coincide in the neighbourhood of a point
z ∈ D, or on a segment of a curve lying in D, or more generally on a point set with an
accumulation point belonging to D, then they coincide throughout D.

[Note: A point z0 is said to be an accumulation point of a set S if each deleted neigh-

bourhood of z0 contains at least one point of S. ]

The theorem states that two different analytic functions cannot coincide in the neigh-
bourhood of an arbitrary point.

This theorem is useful in extending into the complex plane functions defined on the real
axis. For example,
1 2
ez = 1 + z + z + ...
2!
is the unique function f (z) which is equal to ex on the real axis.

This theorem (called identity theorem or uniqueness theorem) forms the basis for the
procedure of analytic continuation. A power series about z1 represents an analytic

114
function f1 (z) within its circle of convergence, which extends to the nearest singularity.
If an expansion of this function is made about a new point z2 inside this circle of conver-
gence, the resulting series will converge in a circle which may extend beyond the circle of
convergence of f1 (z). The values of f2 (z) in the extended region are uniquely determined
bt f1 (z) - in fact, by the values of f1 (z) in the common region of convergence of f1 (z)
and f2 (z). f2 (z) is said to be the analytic continuation of f1 (z) into the new region. This
process may be repeated until the entire plane is covered except for singular points by
these elements of a single function F (z).

Example
f1 (z) = 1 + z + z 2 + . . .

converges in a circle of radius 1 to

1
F (z) = .
1−z
However, F (z) is analytic everywhere except at the simple pole z = 1 and no other
function analytic outside |z| = 1 can coincide with f1 (z) within |z| < 1. F (z) is the
unique analytic continuation of f1 (z) into the entire plane.

Let us have a more detailed discussion. Consider a function f (z) that is analytic in some
region D. We can expand f (z) in a Taylor series about an arbitrary point z0 in D
∞
X
f (z) = an (z − z0 )n (17.1)
n=0

This series will be convergent within some circle γ0 . Let this circle be defined by

|z − z0 | = r0 . (17.2)

Suppose that we know the behaviour of f (z) in the neighbourhood of the point z = z0
or, more precisely, that we know the coefficients an (n = 0, 1, 2 . . . ) of the Taylor series
(17.1). We show that this knowledge is sufficient to determine the behaviour of f (z) in
the neighbourhood of an arbitrary z00 ∈ D, which may be far removed from z0 .

By the definition of a region in the complex plane, it is possible to connect z0 and z00 by
a continuous path C lying entirely within D. Let us take a point z1 ∈ C such that

|z1 − z0 | < r0 (17.3)

i.e., z1 is within γ0 .

Since the power series converges uniformly for |z − z0 | < r0 , it can be differentiated term
by term, and therefore, one may find all derivatives of f (z) from its Taylor expansion

115
at all points within the circle γ0 and, in particular, at the point z1 . Thus, we know the
values of
df (z) dn f (z)
f (z1 ), ,..., ,... (17.4)
dz z=z1 dz n z=z1

(1)
However, these are, apart from a factor 1/n!, just the values of the coefficients an (n =
0, 1, 2, . . . ) of the Taylor expansion of f (z) about the point z1 . Hence, the coefficients in
∞
X
f (z) = a(1)
n (z − z1 )
n
(17.5)
n=0

are known.

Suppose that the sum in (17.5) converges within a circle γ1 defined by

|z − z1 | = r1 (17.6)

z00

z1

z0 γ1

γ0

Figure 17.11: Analytic continuation

Since f (z) is analytic on C ⊂ D, r1 is always nonzero and it is always possible to choose

the point z1 (which can be anywhere within γ0 ) in such a way that the circle γ1 lies partly
outside the circle γ0 (Fig.17.11). This shows that now we are able to calculate f (z) not
only within the original region γ0 , where it was assumed that f (z) was known, but also
within the larger region consisting of the union of the interiors of γ0 and γ1 . We say in
this case that f (z) has been extended analytically beyond γ0 and call the process analytic
continuation.

This larger region will contain a segment of the curve C, which will be outside γ0 but within
γ1 . Thus, repeating over and over the argument, we cover the path C by overlapping

116
circles γ0 , γ1 , γ2 , . . . , which approach the point z00 . After a certain number of steps, one of
these circles will cover the point z00 and thus we shall be able to find the Taylor expansion
of f (z) about this point. Consequently, we can determine the behaviour of f (z) in the
neighbourhood of z = z00 .

The process described above, which consists in determining the behaviour of an analytic
function outside the region where it was originally defined (in the present case, within the
circle γ0 ), is called analytic continuation.

If one knows the behaviour of f (z) at z0 , then one can deduce its behaviour at z00 but
only if one can connect z0 and z00 by a path lying entirely within the domain of analyticity
of f (z). This means that each point of the path together with its neighbourhood must
belong to the domain of analyticity of f (z) and this guarantees that each circle γj has a
finite size. Hence, z0 and z00 must belong to the same region of analyticity of f (z).

The techniques of analytic continuation presented here determines also the location of the
singular points of f (z), since the radius of convergence of the Taylor expansion of f (z)
at a given point is equal to the distance from the point to the nearest singularity of the
function. If we continue analytically a function along a path going through a singular
point of the function, the radii of circles γj tend to zero as we approach the singularity.
Hence, the process of analytic continuation stops naturally at the singular points.

One can choose different paths and obtain the same series representation valid in the
neighbourhood of z = z00 so long as the region bounded by paths C and C 0 has no
singularity.

Suppose that the function we consider is single valued. Then continuing this function
from an arbitrary point where it is analytic along all possible paths, we determine the
entire region where it is analytic. The “natural boundaries” of this region, if they exist,
are the singular points of the function.

Suppose now that two functions f1 (z) and f2 (z) which have different functional forms,
are analytic within regions D1 and D2 , respectively, which overlap. If f1 (z) and f2 (z) are
identical within the intersection D1 ∩ D2 of the two regions, the result of the analytical
continuation of f1 (z) in D2 (which is unique) must be identical with f2 (z) and the result
of the analytical continuation of f2 (z) in D1 must coincide with f1 (z). Thus, in fact we
may regard f1 (z) and f2 (z) as corresponding to a unique function f (z)

f1 (z) z ∈ D1
f (z) = (17.7)
f (z)
2 z ∈ D2

117
which is analytic throughout the union D1 ∪ D2 of the regions D1 and D2 and which is
uniquely determined by f1 (z) or f2 (z), for z ∈ D1 ∩ D2 .

One say that f1 (z) and f2 (z) are analytic continuations of each other.

As an illustration we take
∞ ∞
X zn X (z − i)n
(a) f1 (z) = and (b) f2 (z) =
n=0
2n+1 n=0
(2 − i)n+1

√
|z − i| = 5

|z| = 2

(a) The series converges for |z| < 2 (see figure above). In this circle the series [which
1
is a geometric series with first term 2
and ratio z/2] can be summed and represents the
function
1/2 1
= .
1 − z/2 2−z

z−i
(b) The series converges for 2−i
< 1 (see figure above). In this circle the series [which is
a geometric series with first term 1/(2 − i) and ratio (z − i)/(2 − i) can be summed and
represents the function
1/(2 − i) 1
= .
1 − (z − i)/(2 − i) 2−z
1
Since the power series f1 (z) and f2 (z) represent the same function f (z) = 2−z in the
√
regions common to the interiors of the circles |z| = 2 and |z − i| = 5, it follows that
they are analytic continuations of each other.

Some more discussions

1. Suppose F (z) be analytic in a region D and suppose that F (z) = 0 at all points on an
arc P Q inside D (see figure). Prove that F (z) = 0 throughout D.

Proof: Choose any point, say z0 , on arc P Q. Then in some circle of convergence C with
centre at z0 [ this circle extending at least to the boundary of D where a singularity may

118
 y

C
P z0 Q

exist], F (z) has a Taylor series expansion

1
F (z) = F (z0 ) + F 0 (z0 )(z − z0 ) + F 00 (z0 )(z − z0 )2 + . . .
2
But by hypothesis F (z0 ) = F 0 (z0 ) = F 00 (z0 ) = · · · = 0. Hence, F (z) = 0 inside C.

By choosing another arc inside C, we can continue the process. In this manner we can
show that F (z) = 0 throughout D.

2. Given that the identity sin2 z + cos2 z = 1 holds for real values of z, prove that it also
holds for all complex values of z.

Proof: Let F (z) = sin2 z + cos2 z − 1 and let D be a region of the z plane containing a
portion of the x-axis [see figure].

119
Since sin z and cos z are analytic in D, it follows that F (z) is analytic in D. Also, F (z) = 0
on the x-axis. Hence, from the discussion in (1) above, F (z) = 0 identically in D, which
shows that sin2 z + cos2 z = 1 for all z in D. Since D is arbitrary, we obtain the required
result.

This method is useful in proving for complex values many of the results true for real
values.

3. Let F1 (z) and F2 (z) be analytic in a region D [see figure] and suppose that on an arc
P Q in D, F1 (z) = F2 (z). Prove that F1 (z) = F2 (z) in D.

P
Q

Proof: This follows from (1) above by choosing F (z) = F1 (z) − F2 (z).

18 Multivalued function and Riemann surface

Let us recall our discussion on complex logarithm in section 7. We know that

log z = ln r + i(θ + 2πn) (n = 0, ±1, ±2, . . . )

Here we denote the real logarithm function as ln, θ is the principal value Arg(z) of the
argument function and r = |z|. To the same point in the complex plane correspond
different values of the function log z.

A point of the complex plane having the property that, after the completion of any cycle
around it, a given function is not restored to its initial value is called a branch point. The
point z = 0 is a branch point of log z.

We also discussed that the non-positive real axis is the branch cut of the logarithm function
when we consider principal value of the argument function, because in this case the log

120
function is not continuous anywhere on the negative real axis, or at the origin, where the
function itself is not well-defined. Depending on the branch of the argument function we
could have chosen other branch cuts as well for the logarithm function.

Let us assume that the z-plane is cut along the negative half of the real axis. One can
then define the single-valued function
!
−π < θ < π
f0 (z) = f0 (r, θ) = ln r + iθ (18.1)
r>0

One can also define the single-valued functions

!
−π < θ < π
f1 (z) = f1 (r, θ) = ln r + i(θ + 2π) (18.2)
r>0

and
!
−π < θ < π
f−1 (z) = f−1 (r, θ) = ln r + i(θ − 2π) (18.3)
r>0

In their range of definition, f1 (z) and f−1 (z) take on the same values the logarithm takes
in the polar angle range: π < θ < 3π and −3π < θ < −π, respectively.

In general we can define an infinite series of single-valued functions

f0 (z), f±1 (z), f±2 (z), . . . , f±|n| (z), . . .

where
!
−π < θ < π
fn (z) = fn (r, θ) = ln r + i(θ + 2πn) (18.4)
r>0

so that fn (z) takes on the same values for −π < θ < π that the logarithm takes in the
polar angle range
(2n − 1)π < θ < (2n + 1)π

We have now replaced the multivalued logarithmic function by a series of different func-
tions that are analytic in the cut z-plane.

Let us now observe that each function fn (z) suffers a discontinuity across the cut; for
example, the value of the function fn (z) just above the cut is very different from its value
just below it. Above the cut we have

fn (r, π − ) = ln r + i[(2n + 1)π − ]  > 0

121
while below the cut

fn (r, −π + ) = ln r + i[(2n − 1)π + ]  > 0.

The discontinuity of fn (z) across the cut is therefore given by

 
lim fn (r, π − ) − fn (r, −π + ) = 2πi (18.5)
→+0

On the other hand, the value of the function fn (z) just above the cut is the same as the
value of the function fn+1 (z) just below the cut:

lim fn (r, π − ) = lim fn+1 (r, −π + ) (18.6)

→+0 →+0

[fn+1 (r, −π + ) = ln r + i(2n + 1)π + ]

This suggest the following geometrical construction: we superpose an infinite series of cut
complex planes one on top of the other, each plane corresponding to a different value of
n(= 0, ±1, ±2, . . . ).

The adjacent planes are connected along the cut; the upper lip of the cut in the nth
plane is connected to the lower lip of the cut in the (n + 1)st plane; the branch points are
common to all the planes.

Hence, a crossing of the cut is equivalent to going to one of the two adjacent complex
planes.

The geometrical surface obtained from this helix-like superposition of planes is called a
Riemann surface and each plane is called a Riemann sheet of the function.

A single-valued function defined on a Riemann sheet is called a branch of the complete

multivalued function.

What we have achieved by this construction is the following: from a sequence of single-
valued functions defined in a single complex plane, we have obtained one continuous
(see (18.6)) single-valued function defined on a Riemann surface. In fact, throughout the
Riemann surface we have constructed, the logarithmic function is analytic except at the
branch points, which play the role of singular points.

As for the notion of a branch point, it now gets a simple geometrical interpretation.
Performing a complete cycle around a branch point, we move to another Riemann sheet
where the function takes on different values. On the other hand a complete cycle that
does not include a branch point brings us to the starting point on the Riemann surface
and hence restores the function to its initial value.

122
We say that a branch point is of the nth order if after making (n + 1) complete cycles
around it, we restore the function to its initial value. Otherwise, a branch point is said to
be of an infinite order, as is the case of the logarithmic function. By performing successive
rotations around the origin, we move farther and farther away from the initial Riemann
sheet.

18.1 The functions f (z) = z 1/n and their Riemann surface

Setting z = reiθ in f (z) = z 1/2 , we have

f (z) = f (r, θ) = r1/2 eiθ/2 (18.7)

One easily finds that z = 0 is the branch point of f (z) = z 1/2 . (One can say that z = ∞
is another branch point by considering z = 1/w in f (z)).

A cycle around z = 0 changes θ by 2π and from (18.7) we see that this results in changing
the sign of the function at a given point

f (r, θ) = −f (r, θ + 2π)

The branch cut of f (z) = z 1/2 can be chosen to be the non-negative real axis.

We define the first Riemann sheet by fixing the values of f (z) on the upper lip of the cut

f (z) = f (r, 0) = r1/2 for Re z > 0, Im z = +0.

Then on the lower lip of the cut we have

f (z) = f (r, 2π) = −r1/2 for Re z > 0, Im z = −0.

Crossing the cut, we move to the next sheet, where the values of f (z) on the upper lip of
the cut are the same as the values of f (z) on the lower lip on the first sheet, while the
values of f (z) on the lower lip are obtained by adding 2π

f (z) = f (r, 4π) = r1/2 for Re z > 0, Im z = −0

on the second sheet. Hence, a second crossing of the cut does not yield a new value for
f (z), since

f (r, 0) = f (r, 4π). (18.8)

123
Therefore, the Riemann surface of f (z) = z 1/2 , constructed so as to make this function
continuous everywhere, has two sheets that are connected along the cut; because of (18.8).
The lower lip of the second sheet must be reconnected to the upper lip of the first sheet,
i.e., if the cut is crossed from the second sheet, the function returns to the first sheet. In
other words, the Riemann surface is closed.

In general, the Riemann surface of the function f (z) = z 1/n (n = 2, 3, 4, . . . ) is a closed,

n-sheeted surface, the nth sheet being reconnected to the first sheet.

124