Introductory Complex Analysis

Dipolardog

January 11, 2023

1 Complex Valued Functions

A complex number can be thought as a mathematical object that “store” two real numbers. An element z

of the set of complex numbers C can be written as:

z = x + iy

√
where i ≡ −1 is called the imaginary unit and x, y ∈ R. With this, one can think of constructing

complex valued functions – namely, a function f : C → C:

f (z) = f (x + iy) = w(x, y) + iu(x, y) (1)

Since complex numbers encode two real numbers x and y, we can consider complex valued functions yielding

two real valued functions w(x, y) and u(x, y) that depend on x, y; note the last part of the sentence – two

real valued functions. w and u, being real valued multivariable functions, allows us to extend calculus on Rn

to C. This constitutes the study of complex analysis.

2 Defining Elementary Functions

Defining complex polynomials follows trivially from its real counterpart. But what about exponentials?

Logarithms? Trigonometric functions? We’ll be extending the latter three in this section.

1
2.1 Exponentials

Exponentials turn out to have very useful properties in the study of complex analysis. We begin by recalling

the Taylor expansion of ex :

∞
X xn 1 1 1 1
ex = = 1 + x + x2 + x3 + x4 + x5 + ...
n=0
n! 2! 3! 4! 5!

We try plugging in, crudely, a purely imaginary number iξ:

1 1
eiξ = 1 + iξ + (iξ)2 + (iξ)3 + ...
2! 3!

Let’s go ahead and expand all the i terms:

1 2 1 1 1
eiξ = 1 + iξ − ξ − iξ 3 + ξ 4 + iξ 5 − ...
2! 3! 4! 5!

We group the real and imaginary parts to find something remarkable:

   
iξ 1 2 1 4 1 3 1 5
e = 1 − ξ + ξ − ... + i ξ − ξ + ξ − ... = cos(ξ) + i sin(ξ)
2! 4! 3! 5!

We get sinusoidal real and imaginary parts! This result is known as the Euler formula. Observe that eiξ

encodes the angular position of a point on the complex plane, at an angle ξ. Therefore, we can express all

points on the complex plane by specifying the radial position r as well:

z = reiξ

Note that we can use the Euler formula to define cosine and sine functions in a more illuminating way:

eiξ + e−iξ
cos ξ =
2
eiξ − e−iξ
sin ξ =
2i

This introduces the hyperbolic cosine and sine functions as well:

eξ + e−ξ
cosh ξ = cos(iξ)
2
eξ − e−ξ
sinh ξ = −i sin(iξ) =
2

2
2.2 Logarithms

As we’ve defined complex exponentials now, it’s natural for us to define logarithms using it:

ln(z) = ln reiξ = ln r + iarg(z)

where arg(z) is the angular coordinate of z. Note that we haven’t specified the range of arg(z) yet. Does it

reset to 0 once we go over 2π? Does it run from −π to π? Turns out such seemingly trivial questions hold

great significance in complex analysis, as we shall see in the next section.

3 Multiple Valued Functions and Branch Cuts

3.1 Multiple Valued Functions

√
Consider the function f (z) = z. We have no ambiguity in determining the value of such function at some

point on the Cartesian plane. However, on the complex plane, we run into a problem. We begin by noting

that:
1 1 1 1
f (z) = z 2 = e 2 ln(z) = e 2 ln |z| ei 2 arg(z)

Say we start at some point z, go around a full circle counter-clockwise to return to z so that its argument

increases by 2π. Since the point that returns remains at the same radial distance, the first exponential part

doesn’t change. However, the second exponential part gives:

1 1 1
ei 2 [arg(z)+2π] = ei 2 arg(z) eiπ = −ei 2 arg(z)

Therefore, going around a circle, back to (seemingly) the same point, picks up a negative sign! In other

words, functions on the complex plane can have multiple values at (seemingly) the same point. This obviously

poses some issues when we’re integrating, differentiating, etc as we won’t know which value we’re on. Math-

ematicians have several ways of avoiding this issue, but the most popular methods involve Riemann surfaces

or branch cuts. We’ll be focusing on the latter as the former requires an understanding of topology and

differential geometry.

3.2 Branch Cuts

The problem seems to arise when z crosses a certain line on the complex plane. In our case above, f (z) is

single valued as long as z doesn’t travel over 2π radians. Regions in which a function is single valued are

3
called branches. When z travels over 2π, the function changes branches. The most obvious way to solve the

problem, then, is to not allow arg(z) to change over 2π. We draw a line from the origin, out to infinity in

one direction, so whenever we integrate or something, we just keep in mind that we must not cross over the

line – this way, we allow the argument to change by 2π at max. Such line is called a branch cut. Though

the branch cut for this function is pretty straightforward, branch cuts for other functions may be hard to
√
find. Let us study another example. Consider the function f (z) = z 2 − 1. If we let z − 1 = r1 eθ1 and

z + 1 = r2 eθ2 , we get:
p p √ θ1 +θ2
f (z) = z2 − 1 = (z − 1)(z + 1) = r1 r2 e 2

As shown in the diagram below, as z circles around z1 = 1, z − 1 loops around a circle of equal radius

centered at the origin, and z + 1 loops around the same circle centered at (2, 0) on the complex plane.

Therefore, θ1 increases by 2π and θ2 remains unchanged as z circles around z1 . We can write the new value

of f (z) after one loop around z1 as:

√ θ1 +2π+θ2 √ θ1 +θ2
f (zz1 ) = r1 r2 ei 2 = r1 r2 e 2 eiπ = −f (z)

Note that we’ve distinguished the z before and after looping around z1 , but they are obviously the same

complex number. We therefore find that f (z) changes signage, from its original value, after looping around

z1 . Looping around it twice will cause an increment of 4π on θ1 , and f (z) will return to its original value.

When z circles around z2 = −1, z + 1 circles around the origin, and z − 1 circles around a circle centered at

4
(−2, 0). This time, θ2 increases by 2π every loop around z2 . We get that f (z) again changes signage every

loop around z2 . Thus, studying the function, we get that there’s a sign ambiguity anytime z loops around

either 1 or −1 – z shouldn’t be able to fully loop around either point. To that end, we suggest drawing the

branch cut (colored gray) below:

To actually verify that this branch cut works, let’s consider the following loop – circles around z1 , moves to

z2 through a path C3 that’s infinitesimally close to the branch cut, circles around z2 , and comes back to its

original point through C4 that’s also infinitesimally close to the branch cut. Note that as this happens, z − 1

traverses a copy of the loop shifted 1 unit to the left (along the real axis) and z + 1 traverses a copy shifted

1 unit to the right. Let’s first observe what happens to z − 1:

1. C1 : θ1 gains 2π

2. C3 : r1 increases from ϵ, the radius of C1 and C2 , to 2 − ϵ

3. C2 : θ1 remains the same, as discussed in previous sections

4. C4 : r1 returns to ϵ

Thus, z − 1 only experiences a 2π increase in θ1 . We can do the same thing for z + 1 and arrive at the

conclusion that θ2 also increases by 2π as z goes through the loop. Now, calculating the value of f (z) after

moving through the loop once

√ θ1 +2π+θ2 +2π √ θ1 +θ2 √ θ1 +θ2

r1 r2 ei 2 = r1 r2 ei 2 e2πi = r1 r2 ei 2

5
giving its original value, thus verifying that there’s no longer a value ambiguity. Note that this isn’t the

only valid branch cut – we can also consider drawing cuts from z1 out to positive infinity and from z2 out

to negative infinity – either way, the branch cut prevents looping around z1 or z2 .

4 Derivatives of Complex Valued Functions

df
We notice a problem when we take the derivative dz – we have multiple paths through which we can take

the derivative. From calculus on reals,


df  f (z0 + δz) − f (z0 )
= lim (2)
dz z0 δz→0 δz

but δz can be given in infinitely many directions – we can change x while keeping y constant, change y while

keeping x constant, etc. If the notion of differentiability carries onto complex analysis, the limit in (2) must

give the same value regardless of the direction in which δz is given for f to be differentiable at z0 . This gives

us an idea – let’s try differentiating f along the “real” direction and along the “imaginary” direction. If we

give δz by changing x – δz = δx. The limit in (2) becomes:

f (x0 + δx + iy0 ) − f (x0 + iy0 )

lim
δx→0 δx

where z0 = x0 + iy0 . Using (1), the limit then becomes:

 
w(x0 + δx, y0 ) − w(x0 , y0 ) u(x0 + δx, y0 ) − u(x0 , y0 )
lim +i
δx→0 δx δx

but we see that these are just partial derivatives evaluated at z0 :

 
∂w ∂u
+i (3)
∂x ∂x z0

Now, let’s give δz by changing y, so δz = iδy. The limit in (2) becomes:

f (x0 + i(y0 + δy)) − f (x0 + iy0 )

lim
δy→0 iδy

Following a similar procedure from before, the limit becomes:

 
w(x0 , y0 + δy) − w(x0 , y0 ) u(x0 , y0 + δy) − u(x0 , y0 )
lim +i
δy→0 iδy iδy

6
 1
Since i = −i, the limit can be written as:

 
∂u ∂w
−i (4)
∂y ∂y z0

If f is differentiable at z0 , (3) and (4) should be equal. Equating real and imaginary parts of (3) and (4),

we find a remarkable result:

∂u ∂w
=−
∂x ∂y
∂w ∂u
= (5)
∂x ∂y

The two equations in (5) are known as the Cauchy-Riemann Conditions – necessary criteria for a function

to be complex-differentiable at some point. If f is to be analytic, that is, to have a convergent power

series expansion at z0 , the four partial derivatives in (5) must exist and be real-continuous at z0 . The

additional requirement of continuity makes a sufficient condition for complex-analyticity. As you’ll see, (5)

has important implications in the study of complex analysis – I’d even go as to say it’s the most crucial

result of complex analysis!

5 Power Series on the Complex Plane

Say we are looking for the power series expansion of the complex valued function:

1
f (z) =
1−z

From calculus on reals, we’re familiar with the expansion:

∞
1 X
= 1 + x + x2 + ... = xn
1−x n=0

which is convergent for |x| < 1. We carry this straight to complex analysis and argue that:

∞
X
f (z) = zn
n=0

7
is convergent inside the open unit disk centered at the origin on the complex plane (ie. |z| < 1). What about
1
the region outside? There, |z| > 1 and |z| < 1, so we can think about doing the expansion:

  ∞
1 1 X 1
f (z) = − =− n+1
z 1 − 1/z n=0
z

If the previous expansion looked familiar, this should look very odd – a power series in negative powers! Yet

this is indeed a convergent power series. We shouldn’t be surprised – check for yourself that the analyticity

condition is met for f (z) in the region outside of the unit disk. It turns out that we need to extend the

Taylor series in ordinary calculus to negative powers in complex analysis. This generalized series expansion

is called the Laurent series. As we saw above, a Laurent series of the form:

∞
X
f (z) = an z n (6)
n=−∞

is convergent when the real valued power series:

∞
X
|an ||z|n
n=−∞

is convergent. If a Taylor series converges on a disk, a Laurent series converges on an annulus. For the

expansion outside of the unit disk from before, the series converges on an annulus of inner radius Rin = 1

and outer radius Rout = ∞. In many cases, the Laurent series can be found from the relevant Taylor series
1
expansion. For instance, the Laurent series expansion of e1/z can be found by plugging in z into the Taylor

expansion of ex :
∞
X 1
e1/z =
n=0
n!z n

which converges on the annulus Rin → 0 (note that e1/z isn’t analytic at 0) and Rout = ∞ since the expansion

for ez is convergent everywhere on the complex plane.

6 Complex Integration of Analytic Functions

We diverge for a bit to the topic of complex integrals. As was the case with differentiation, integration of

complex valued functions can be evaluated along different curves – we don’t necessarily have to integrate

along an axis, as we would in single variable calculus. It turns out closed loop integrals have some nice

properties in complex analysis – we’ll see why in this section. A function f (z) is analytic in the region S

8
bounded by a curve C. Let’s evaluate the closed loop integral:

I
f (z)dz
C

We can write the integral over x and y:

I I I
[w(x, y) + iu(x, y)][dx + idy] = w(x, y)dx − u(x, y)dy + i w(x, y)dy + u(x, y)dx
C C C

Using Green’s theorem, this is equivalent to:

ZZ ZZ
∂u ∂w ∂w ∂u
− + dA + i − dA
S ∂x ∂y S ∂x ∂y

Applying the Cauchy Riemann conditions in (5), we discover that both integrals equal zero! We conclude

that the closed loop integral of a function f equals zero for all loops on a region f is analytic. Closed loop

integrals on the complex plane are often referred to as contour integrals.

7 Cauchy Integral Formula

Say a function f is analytic in the neighborhood of a point z. Let’s evaluate the contour integral:

f (z ′ ) ′
I
dz (7)
C z′ − z

around an infinitesimally small circular loop C of radius ϵ, centered at z. Due to the denominator, the

integrand isn’t analytic at z, which is a point enclosed by C, so we can’t argue that the integral evaluates

to 0 right away. For all points z ′ on C, we can write:

z ′ − z = ϵeiθ

where we’ve used Euler’s formula. With this, we write:

dz ′ = ϵieiθ dθ

Since ϵ is very small, the value of f (z ′ ) can be taken to be constant at f (z) around the loop. Therefore, the

integral in (7) can be evaluated as:

Z 2π
f (z) idθ = 2πif (z)
0

9
We therefore get the remarkable result that:

f (z ′ ) ′
I
1
f (z) = dz (8)
2πi C z′ − z

We can even evaluate derivatives as a line integral now. We just take the derivative of (8) with respect to z:

∂ f (z ′ ) f (z ′ )
I   I
df 1 ′ 1
= ′
dz = dz ′
dz 2πi C ∂z z − z 2πi C (z ′ − z)2

We can generalize this to the nth derivative:

dn f f (z ′ )
I
n!
= dz ′ (9)
dz n 2πi C (z ′ − z)n+1

8 Extension of Cauchy Integral Formula

What if C in the previous section isn’t a small loop? Say we want to evaluate the integral in (8) around the

loop C shown in the figure below. Though C itself isn’t small enough to carry out the calculation we did

in the previous section, we can extend the contour as shown in the figure – do a “nearly” full loop around

C, travel towards z0 along γ1 , loop around a small circle around z0 , then return to C along γ3 , a line that’s

infinitesimally close to γ1 .

10
Line integrals have the useful property that a change in the direction of integration results in a sign change.

Since γ1 and γ3 are very close, we can almost treat them as the same line. Therefore,

Z Z
f (z)dz = − f (z)dz
γ1 γ3

The integrand in (8) is actually analytic in the region enclosed by this new contour as we’ve cleverly avoided

going around z0 , so the integral around the whole thing should just be zero:

I Z Z I I I
+ + + = + =0
C γ1 γ2 γ3 C γ3

where we’ve omitted the integrands for clarity. We note that the second term is just f (z0 ) as we’ve found

in the previous section. Now, note that the direction of integration over C is opposite to that circling γ2 , so

the contour integral over C picks up a negative sign. Thus,

I
1 f (z)
dz = f (z0 )
2πi C z − z0

regardless of the shape and size of C. The technique used in this section is used widely across complex

analysis, as we shall see in the following sections.

9 Laurent Series

Say we want to find the Laurent series expansion of a function f (z) that is analytic everywhere on the

complex plane except at a finite number of points. Let’s say the expansion converges on an annulus centered

at a point z0 , with Rin < |z − z0 | < Rout . Consider the Taylor expansion formula:

∞
X
an (z − z0 )n
n=0

where the coefficients are:

1 (n)
an = f (z0 )
n!

Turns out the Laurent series on this annulus can be found by substituting the derivative with (9) and

extending the series to negative powers. Hence, our Laurent series is:

∞
X
f (z) = bn (z − z0 )n (10)
n=−∞

11
where the coefficients are:
f (z ′ )
I
1
bn = dz ′ (11)
2πi C (z ′ − z0 )n+1

Negative powers of (z − z0 ) constitute the principal part of the Laurent series and the rest constitute the

analytic part.

10 Regular Points and Singularities

Points at which f is analytic are called regular points and points at which it isn’t are called singularities.

The Laurent series expansion can be used to classify the point z0 we’re expanding about. If z0 is a regular

point, all negative powers of (z − z0 ) must vanish, which makes intuitive sense since we want ‘nice’ behavior

at z = z0 . If z0 is a singularity, two cases arise:

1. The principal part terminate at some n = −p

2. The principal part continues indefinitely

Case 1 is a pole of order p at z = z0 . If case 2 happens for the series expansion in the annulus closest to z0 ,

z = z0 is called a essential singularity.

11 Integration Around Singularities

Thus far, we’ve only dealt with contour integrals of analytic functions. Let us now study contour integrals

over regions where the integrand isn’t analytic. We begin with the Laurent series expansion about an

arbitrary singularity z0 :

b−2 b−1
f (z) = ... + + + b0 + b1 (z − z0 ) + b2 (z − z0 )2 + ... (12)
(z − z0 )2 z − z0

H
say the series expansion converges in the vicinity of z0 . Notice if we evaluate the integral C
f (z)dz around

a small loop around z0 ,

I I I I
b−2 b−1
f (z)dz = ... + dz + dz + [b0 + b1 (z − z0 ) + b2 (z − z0 )2 + ...]dz
C C (z − z0 )2 C z − z0 C

we find that the integral over the analytic part goes to zero since the integrand is analytic everywhere in the

region bounded by C. Now, by (9), all terms of the principal part save the b−1 term would involve some

12
derivative of b−n , thus equating to zero as well:

I I
b−1
f (z)dz = dz = 2πib−1
C C z − z0

b−1 in a Laurent series expansion is so important that we give it a name – residue. A contour integral around

a singularity returns nothing but 2πi times the residue of the singularity.

12 Residue Calculus

We can combine the results of sections 6 and 9 to develop a powerful tool that proves useful in areas far

beyond complex analysis – residue calculus. Now, we wish to integrate around multiple (yet a countable

number of) singularities. Consider a contour C that encloses 4 singularities:

As explained in section 6, the lines that run from C to the singularity contribute nothing to the full contour

integral. Now, if we do a Laurent series expansion about each singularity to find the residues of each

singularity, the integral will pick out the residues to give:

I
f (z)dz − 2πi[Res(z1 ) + ... + Res(z2 )] = 0 (13)
C

where the negative sign accounts for the fact that the loops around the singularities are in opposite direction

of C. We therefore find that the contour integral over a function is equal to 2πi times the sum of the residues

of all singularities bounded by the contour.

13
13 Finding Residues
1
The manual way to find residues is to directly find the expansion coefficient of the z−z0 term. However, that

involves the contour integral:

I
1
b−1 = f (z ′ )dz ′ (14)
2πi C

which can be hard to do and, from our discussion in section 12, kind of defeats the purpose of finding the

residues in the first place – we wish to find the residues to evaluate the integral in (14), not the other way

around. Thankfully, we can use some basic calculus to find b−1 for poles of order p ∈ Z+ . We start with the

series expansion in (12) with a pole of order p at z = z0 :

b−p b−1
f (z) = p
+ ... + + b0 + b1 (z − z0 ) + ...
(z − z0 ) z − z0

Notice if we multiply (z − z0 )p to both sides,

f (z)(z − z0 )p = b−p + ... + b−1 (z − z0 )p−1 + b0 (z − z0 )p + b1 (z − z0 )p+1 + ...

take p − 1 derivatives with respect to z,

dp−1 (p + 1)!
f (z)(z − z0 )p = 0 + ... + (p − 1)!b−1 + p!b0 (z − z0 ) + b1 (z − z0 )2 + ...
dz p−1 2

and finally take the limit z → z0 :

dp−1
lim f (z)(z − z0 )p = (p − 1)!b−1
z→z0 dz p−1

Therefore, for poles of order p at z0 ,

1 dp−1
Res[z0 ] = lim f (z)(z − z0 )p
(p − 1)! z→z0 dz p−1

such trick doesn’t work when we have an essential singularity, so we would have to resort to finding the

Laurent series, which can be do-able if we know the relevant Taylor expansion.

14
14 Residue Integration for Real Valued Integrals

Suppose we want to evaluate the real valued integral:

Z ∞
cos(x)
dx
−∞ x2 + 4

which is pretty tough with the methods we know in single-variable calculus. Turns out residue integration

can be used to evaluate such integral. Consider the contour integral:

eiz
I
dz
C z2 + 4

whose real part is equal to the integrand of the real valued integral. Let us first find the singularities and

residues. We can write the complex integrand as:

eiz eiz
f (z) = =
z2 + 4 (z + 2i)(z − 2i)

Therefore, our singularities are at ±2i. We find that since

e∓2
lim (z ∓ 2i)f (z) =
z→±2i ±4i

Both poles are of order p = 1 and the residues are:

e∓2
Res[±2i] =
±4i

With this, let’s choose our contour C to be the top half circle of radius R → ∞ that encloses 2i:

15
Why not the bottom half-circle? We’ll answer that in a second. We first note that the integral over γ gives

the integral we wish to find since, along γ, z = x ∈ R:

 Z ∞
eix
Z
cos(x)
Re 2
dx = 2
dx
γ x +4 −∞ x + 4

What about the part along Γ? Along this curve, z = Reiθ = R cos θ + iR sin θ, so:

π
e−R sin θ+iR cos θ
Z Z
I= = lim iReiθ dθ
Γ R→∞ 0 R2 ei2θ + 4

Let’s take a look at the magnitude of the value of the integral, which would be a complex number. We first

use the integral inequality:

Z π  −R sin θ+iR cos θ
e

iθ 

|I| ≤ lim 
 R2 ei2θ + 4 iRe  dθ
R→∞ 0

The modulus of the product of two complex numbers zw is equal to the product of their respective moduli:

π  −R sin θ+iR cos θ π

 −R sin θ+iR cos θ  π
e−R sin θ
Z  Z e Z
e iθ 

|iReiθ |dθ =


 R2 ei2θ + 4 iRe dθ = Rdθ
0

0 |R2 ei2θ + 4| 0 |R2 ei2θ + 4|

Next, we use the reverse triangle inequality:

|R2 ei2θ − (−4)| ≥ ||R2 ei2θ | − | − 4|| = |R2 − 4|

Therefore,
π
Re−R sin θ
Z
|I| ≤ lim dθ
R→∞ 0 R2 − 4

Applying the limit, we find that the right hand side of the inequality is simply zero since sin θ is positive in

the range 0 ≤ θ ≤ π – the exponential vanishes, as well as the rational part. We see here that if we had

taken the bottom half circle, the Γ integral wouldn’t have vanished and the calculation might’ve been very

complicated. We also have that the magnitude of any complex number has to be greater than or equal to

zero, where equality holds when the complex number is zero. Since we have 0 ≤ |I| ≤ 0, I must vanish as R

approaches infinity. Therefore,

eiz πe−2
I Z Z Z
dz = + = = 2πiRes[2i] =
C z2 + 4 γ Γ γ 2

16
 R
Since the last value is purely real, we have that the imaginary part of γ
vanishes, which is an obvious result

since the integrand of

Z  Z ∞
sin x
Im = dx
γ −∞ x2 + 4

is odd. Equating the real part, we finally find:

Z ∞
cos x π
dx = 2
−∞ x2 + 4 2e

The contour used for this example was fairly straightforward. However, in most practical applications of

complex analysis, finding the appropriate contour takes ingenuity. There are literally hundreds of areas in

mathematics and physics where residue integration is applied – integral approximations, integral transforms,

series summations, etc. I recommend looking for more complicated examples in textbooks and videos online.

I haven’t introduced another important tool in complex analysis – conformal transformations – which are

crucial to the study of various fields in physics (potential theory, fluid dynamics, etc). Hopefully, I’ll have

time to do so on the next handout.

17