LECTURE 0

Assumed Knowledge

This is a review of basic facts about complex numbers that ought to be familiar:

• the definition of complex numbers,

• their arithmetic,
• Cartesian and polar representations,
• the Argand diagram,
• de Moivre’s theorem, and
• extracting nth roots of complex numbers.

Students who do not feel confident about this material need to do lots of exercises
about these, such as those in the MATH1141 notes.

1. Complex numbers
Definition 0.1. A complex number is an expression of the form x + iy, where
x and y are real numbers. The real part of x + iy is x and the imaginary part of
x + iy is y. We denote this by Re(x + iy) = x and Im(x + iy) = y. The set of all
complex numbers is denoted C.

We often write w = u + iv and z = x + iy, and work with w and z rather than
u + iv and x + iy. In this case, we write, for instance, Re(z) = x and Im(w) = v.
We abbreviate x + i0 and 0 + iy to x and iy, and 0 + i1 to i.

Definition 0.2. Suppose that w = u + iv and z = x + iy, where u, v, x, y ∈ R.

Then we define the complex numbers w + z, −z, wz and, if z 6= 0, z −1 and w/z, as
follows:

w + z = (u + x) + i(v + y)
−z = −x + i(−y)
wz = (ux − vy) + i(uy + vx)
z −1 = (x2 + y 2 )−1 (x − iy)
w/z = wz −1 = (x2 + y 2 )−1 [(ux + vy) + i(vx − uy)].

Then i2 = −1 and −z = (−1)z.

We may combine these operations to make sense of more complicated expressions
such as wm − z n , where m and n are integers.
1
2 0. ASSUMED KNOWLEDGE

Proposition 0.3. Complex numbers have the following properties:

z1 + z2 = z2 + z1 ∀z1 , z2 ∈ C
(z1 + z2 ) + z3 = z1 + (z2 + z3 ) ∀z1 , z2 , z3 ∈ C
0+z =z ∀z ∈ C
(−z) + z = 0 ∀z ∈ C
z1 z2 = z2 z1 ∀z1 , z2 ∈ C
(z1 z2 )z3 = z1 (z2 z3 ) ∀z1 , z2 , z3 ∈ C
1z = z ∀z ∈ C
zz −1 = 1 ∀z ∈ C \ {0}
z1 (z2 + z3 ) = z1 z2 + z1 z3 ∀z1 , z2 , z3 ∈ C
The symbol ∀ is read “for all”. This proposition shows that the complex numbers
form a field.
Definition 0.4. If z = x + iy, then z, the (complex) conjugate of z, and |z|, the
modulus of z, are defined to be x − iy and (x2 + y 2 )1/2 .
Some further properties of complex numbers relate conjugates and moduli.
Proposition 0.5. For all z, z1 , z2 ∈ C,
z+z z−z
(a) Re(z) = (b) Im(z) =
2 2i
(c) z1 + z2 = z 1 + z 2 (d) z1 z2 = z 1 z 2
(e) −z = −z (f) z −1 = (z)−1
(g) |z1 + z2 | ≤ |z1 | + |z2 | (h) |z1 z2 | = |z1 | |z2 |
(i) zz = |z| 2
(j) z −1 = |z|−2 z.

Inequality (g) is called the triangle inequality. If |z| = 1, then z −1 = z, from (j).
Proof. We only prove the triangle inequality, because this is hardest.
First, here is an algebraic proof. Recall that 2ab ≤ a2 + b2 for real numbers a
and b. Taking a and b to be x1 y2 and x2 y1 , we deduce that
2x1 x2 y1 y2 ≤ x21 y22 + x22 y12 ,
and hence
x21 x22 + y12 y22 + 2x1 x2 y1 y2 ≤ x21 x22 + y12 y22 + x21 y22 + x22 y12 = (x21 + y12 )(x22 + y22 ),
so, taking square roots,
x1 x2 + y1 y2 ≤ |z1 | |z2 | .
Finally, z1 + z2 = (x1 + x2 ) + i(y1 + y2 ), and so
|z1 + z2 |2 = (x1 + x2 )2 + (y1 + y2 )2
= x21 + x22 + y12 + y22 + 2(x1 x2 + y1 y2 )
≤ |z1 |2 + |z2 |2 + 2|z1 | |z2 |

2
= |z1 | + |z2 | ;
the triangle inequality follows by taking square roots.
 2. EULER’S FORMULA 3

Alternatively, here is a geometric version. Consider the triangle whose vertices

are w, z, and w + z, and the parallelogram with vertices 0, w, z, and w + z. The
side of the parallelogram that joins w to w + z is congruent to the side joining 0 to
z, and the side of the parallelogram that joins z to w + z is congruent to the side
joining 0 to w. So we are just asserting the obvious fact that the length of one side
of a triangle is less than the sum of the other two sides. □

2. Euler’s formula
The usual exponential function has a Taylor series:
x x2 x3 x4 x5
ex = 1 + + + + + + ··· ,
1! 2! 3! 4! 5!
so at least formally, for a real number θ,
θ θ2 θ3 θ4 θ5
eiθ = 1 + i − −i + + i + ···
1! 2! 3! 4! 5!
 θ 2
θ 4   θ θ3 θ5 
= 1− + + ··· + i − + + ···
2! 4! 1! 3! 5!
= cos θ + i sin θ,
from the Taylor series for the cosine and sine functions.
Later we will make this rigorous and use power series very effectively. This
observation leads us to make the following definition.
Definition 0.6. We define eiθ to be cos θ + i sin θ, for any real number θ.

Suppose that z = x + iy 6= 0. Write r instead of |z|. Then (x/r, y/r) lies on the
unit circle in the Cartesian plane, so (x/r, y/r) = (cos θ, sin θ) for an appropriate
choice of θ, and hence
x y

z=r +i = r cos θ + i sin θ = reiθ .
r r
Definition 0.7. The Cartesian form of a complex number z is its representation
in the form x + iy, where x and y are real. The polar form of a complex number z is
its representation in the form reiθ , where r ≥ 0 and θ ∈ R. The number θ is called
the argument of z, and is written arg(z).
Lemma 0.8. Suppose that r, s ∈ R+ and θ, ϕ ∈ R. Then
r(cos θ + i sin θ) = s(cos ϕ + i sin ϕ)
if and only if r = s and θ − ϕ = 2kπ for some k ∈ Z.

This lemma follows immediately from trigonometry. It tells us that the argument
of a nonzero complex number z is ambiguous. The next definition is to avoid this
ambiguity.
Definition 0.9. The principal value of the argument of a nonzero complex
number z, written Arg(z), is the unique number θ such that z = |z| eiθ and −π <
θ ≤ π.

We do not define the argument of 0.

4 0. ASSUMED KNOWLEDGE

3. The Argand Diagram

Suppose that w = u + iv. Then to the complex number w we associate the point
in the Cartesian plane with Cartesian coordinates (u, v). When we do this, we call
the axes the real axis and the imaginary axis. See Figure 0.1.
Geometrically, |w| is the length of the line joining w to O, and arg(w) is the
angle between this line and the positive real axis (taking the anticlockwise direction
to be positive). Further, |w − z| is the length of the line joining w and z.
Adding two complex numbers corresponds to vector addition in the plane. Mul-
tiplying by r in R+ dilates by a factor of r, and multiplying by the complex number
eiθ (where θ ∈ R) rotates (anticlockwise) through the angle θ.

4. De Moivre’s formula
Theorem 0.10. If θ, ϕ ∈ R, then
(cos θ + i sin θ)(cos ϕ + i sin ϕ) = cos(θ + ϕ) + i sin(θ + ϕ).f
Proof. For all θ, ϕ ∈ R,
(cos θ + i sin θ)(cos ϕ + i sin ϕ)
= (cos θ cos ϕ − sin θ sin ϕ) + i(cos θ sin ϕ + sin θ cos ϕ)
= cos(θ + ϕ) + i sin(θ + ϕ),
as required. □
Corollary 0.11 (de Moivre’s formula). If n ∈ Z and θ ∈ R, then
(cos θ + i sin θ)n = cos(nθ) + i sin(nθ). (0.1)
Proof. This is obviously true if n = 0 or 1. The result may be proved for
n ∈ Z+ by induction. Suppose that k ∈ Z+ and formula (0.1) holds when n = k,
i.e.,
(cos θ + i sin θ)k = cos kθ + i sin kθ.
Then by Theorem 0.10,
(cos θ + i sin θ)k+1 = (cos θ + i sin θ)(cos θ + i sin θ)k
= (cos θ + i sin θ)(cos kθ + i sin kθ)
= cos(θ + kθ) + i sin(θ + kθ)
= cos(k + 1)θ + i sin(k + 1)θ,

Im
w v w = u + iv = seiϕ
b b

s
ϕ Re
b

u O

Figure 0.1. The complex plane

 5. ROOTS OF COMPLEX NUMBERS 5

so the result holds when n = k + 1. By induction, the result holds for all n ∈ Z+ .
To prove the result when n ∈ Z− , we use the fact that if |z| = 1, then zz = 1, so
z = z −1 . That is,
(cos θ + i sin θ)−1 = cos θ − i sin θ = cos(−θ) + i sin(−θ),
as required. □

In polar notation, Theorem 0.10 and Corollary 0.11 become

ei(θ+ϕ) = eiθ eiϕ and (eiθ )n = einθ .
These are more “obvious” and easier to remember than the trigonometric formulae.

5. Roots of complex numbers

We use complex exponentials to find roots of complex numbers. Fix w ∈ C and
n ∈ Z+ , and suppose that w = z n for some z ∈ C. Then z is called an nth root of
w. Write z as reiθ and w as seiϕ . Then
seiϕ = w = z n = rn einθ .

From Lemma 0.8, r = s1/n and nθ = ϕ + 2kπ for some k ∈ Z. Therefore z = s1/n eiθ ,
where
ϕ 2kπ
θ= +
n n
for some k ∈ Z. If k = nl + r, where l ∈ Z and r = 0, 1, 2, . . . , n − 1, then
e(ϕ/n+2kπ/n) = e(ϕ/n+2rπ/n+2lπ) = e(ϕ/n+2rπ/n) ,
so we get the same value of z by taking k = nl + r as by taking k = r. Thus, to
get all n possible values of z, it suffices to take the first n values of k, or any n
consecutive values, or indeed any n values of k which give the n possible different
remainders when divided by n.
The nth roots of any nonzero complex number w are uniformly spaced around
the circle with centre 0 and radius |w|1/n . For example, Figure 0.2 shows the seventh
roots of unity. A symmetry argument shows that the sum of all these numbers is 0.

Im
e4πi/7 b

b
e2πi/7
e6πi/7 b

b
1 Re
b

e−6πi/7 b

b
b
e−2πi/7
e−4πi/7

Figure 0.2. The seventh roots of unity

6 0. ASSUMED KNOWLEDGE

6. History†
Leonhard Euler found the formula that bears his name in 1748; he was one of
the most prolific mathematicians ever. Some mathematicians did not believe that
the geometric representation afforded by the Argand diagram was legitimate. The
story of de Moivre’s death is very curious. For more information, see
http://www-groups.dcs.st-and.ac.uk/~history/Biographies/
and then find Euler, Argand, and de Moivre.
 LECTURE 1

Inequalities and Sets of Complex Numbers

In complex analysis, we consider functions whose domains or ranges or both are

regions in the complex plane. So to be able to discuss functions, we need to be able
to describe regions. Curves and regions in the complex plane are often described by
equalities and inequalities involving | · |, Arg, Re, Im, ….
In the first part of this lecture, we review some equalities and inequalities. Then
we discuss different types of regions. Finally, we consider some examples.

1. Equalities and inequalities

We begin with a lemma that is related to the cosine rule. It implies that Re(wz̄)
is the inner product of the vectors represented by the complex numbers w and z.
Lemma 1.1. For all complex numbers w and z,
|w + z|2 = |w|2 + 2 Re(wz̄) + |z|2 .
Proof. Observe that
|w + z|2 = (w + z)(w̄ + z̄) = ww̄ + wz̄ + w̄z + z z̄
= |w|2 + wz̄ + (wz̄) + |z|2 = |w|2 + 2 Re(wz̄) + |z|2 ,
as required. □
The triangle inequality is one of the most useful results about complex numbers.
It states:
|w + z| ≤ |w| + |z| ∀w, z ∈ C.
Here are a variation on the triangle inequality, sometimes called the circle inequality.
Lemma 1.2. For all complex numbers w and z,
|w| − |z| ≤ |w − z|.
Proof. Observe that w = (w − z) + z, so |w| ≤ |w − z| + |z| by the triangle
inequality, and hence
|w| − |z| ≤ |w − z|. (1.1)
Interchanging the roles of w and z in (1.1),
|z| − |w| ≤ |z − w|.
Combining these inequalities and recalling that |w − z| = |z − w|, we see that
|w| − |z| = max{|w| − |z|, |z| − |w|} ≤ |w − z|,
as required.
Alternatively, consider points on circles of radii |z| and |w|, and compare the
distance between the points with the difference of the radii. □
7
8 1. INEQUALITIES AND SETS OF COMPLEX NUMBERS

Recall that if z = x + iy, then ez is defined to be ex (cos y + i sin y). Then

ew ez = ew+z for all complex numbers w and z. Here is another very useful result.
Lemma 1.3. If z ∈ C, then
|ez | = eRe(z) .
Proof. See Problem Sheet 1. □
Lemma 1.4. For all real numbers θ,
eiθ − 1 ≤ |θ|.
Proof. See Problem Sheet 1. □

2. Properties of sets
Definition 1.5. The open ball with centre z0 and radius ε, written B(z0 , ε), is
the set {z ∈ C : |z − z0 | < ε}.
The punctured open ball with centre z0 and radius ε, written B◦ (z0 , ε), is the set
{z ∈ C : 0 < |z − z0 | < ε}.
Sometimes these sets are called discs rather than balls.
Definition 1.6. Suppose that S ⊆ C. For any point z0 in C, there are three
mutually exclusive and exhaustive possibilities:
(1) When the positive real number ε is sufficiently small, B(z0 , ε) is a subset of
S, that is, B(z0 , ε) ∩ S = B(z0 , ε). In this case, z0 is an interior point of S.
(2) When the positive real number ε is sufficiently small, B(z0 , ε) does not meet
S, that is, B(z0 , ε) ∩ S = ∅. In this case, z0 is an exterior point of S.
(3) No matter how small the positive real number ε is, neither of the above
holds, that is, ∅ ⊂ B(z0 , ε) ∩ S ⊂ B(z0 , ε). In this case, z0 is a boundary
point of S.
These definitions are illustrated in Figure 1.1. We consider points z1 , z2 and z3 .
If the radius of the ball centred at z1 is small enough, then the ball lies inside
the set S, and B(z1 , ε) ∩ S = B(z1 , ε). Thus z1 is an interior point.
If the radius of the ball centred at z2 is small enough, then the ball lies outside
the set S, and B(z2 , ε) ∩ S is empty. Thus z2 is an exterior point.

Im
z1 b

B(z0 , ε)
z0 ε
b z2 b

z3 Re
b

Figure 1.1. The ball with centre z0 and radius ε, and interior, ex-
terior and boundary points z1 , z2 and z3 of the set S
 3. DESCRIBING SETS IN THE COMPLEX PLANE 9

No matter how small the radius of the ball centred at z3 is, part of the ball lies
inside S, and part lies outside S, and B(z3 , ε) ∩ S is neither empty nor all of B(z3 , ε).
Thus z3 is a boundary point.
Definition 1.7. Suppose that S ⊆ C.
(1) The set S is open if all its points are interior points.
(2) The set S is closed if it contains all of its boundary points, or equivalently,
if its complement C \ S is open.
(3) The closure of S, written S̄, is the set consisting of all the points of S
together with all its boundary points.
(4) The set S is bounded if S ⊆ B(0, R) for some positive real number R.
(5) The set S is compact if it is both closed and bounded.
(6) The set S is a region if it is an open set together with none, some, or all of
its boundary points.
For example, the dashed boundary lines of the set S in Figure 1.1 indicate that
it does not contain any boundary points. Consequently, this set is open.
Note that open and closed are not exclusive nor exhaustive. There are sets that
are open and closed, such as the whole plane, and sets that are neither open nor
closed, such as {z ∈ C : Re(z) ≥ 0, Im(z) > 0}. In complex analysis, we focus on
open sets. We often write Ω for an open set.
We now present connectedness.
Definition 1.8. A polygonal path is a finite sequence of finite line segments,
where the end point of one line segment is the initial point of the next one. A simple
closed polygonal path is a polygonal path that does not cross itself, but the final point
of the last segment is the initial point of the first segment. The complement of a
simple closed polygonal path is made up of two pieces: one, the interior of the path,
is bounded, and the other, the exterior, is not.
Definition 1.9. Let X ⊆ C be a subset of the complex plane.
(1) The set X is polygonally path-connected if any two points of X can be joined
by a polygonal arc lying inside X.
(2) The set X is simply polygonally connected if it is polygonally path-connected
and if the interior of every simple closed polygonal arc in X lies in X, that
is, if “X has no holes”.
(3) The set X is a domain if it is open and polygonally path-connected.
The set Ω in Figure 1.2 is polygonally path-connected, because any two points in
Ω (such as z1 and z4 ) can be joined by a polygonal path. However, Ω is not simply
polygonally connected, because part of the interior of the closed path shown going
through z5 is not in Ω.

3. Describing sets in the complex plane

Exercise 1.10. Suppose that a, b, c, d ∈ C. Show that the set
{z ∈ C : |az + b| = |cz + d|}
may be empty, a point, a line, a circle, or the whole complex plane, and all these
possibilities occur for suitable values of the parameters a, b, c, d.
10 1. INEQUALITIES AND SETS OF COMPLEX NUMBERS

Answer. If |az + b| = |cz + d|, then |az + b|2 = |cz + d|2 , so (az + b)(az + b) =
(cz + d)(cz + d) , whence
¯ + (āb − c̄d)z̄ + (cc̄ − dd)
(aā − cc̄)z z̄ + (ab̄ − cd)z ¯ = 0.

We may rewrite this in the form e|z|2 + f z + f¯z̄ + g = 0, where e and g are real,
while f may be complex. Now we pass to Cartesian coordinates:
ex2 + ey 2 + (f + f¯)x + i(f − f¯)y + g = 0;
all the coefficients are now real since f + f¯ = 2 Re f while i(f − f¯) = −2 Im f .
The desired result follows from coordinate geometry. For instance, if e 6= 0 and
eg − |f |2 < 0, then we have a circle with centre −f¯/e, while if e = 0 and f 6= 0,
then we have a straight line.
The other types of solutions arise when many of the coefficients are 0. If a = c = 0
and |b| = |d|, then the equation holds in the whole plane; if a = c = 0 while |b| 6= |d|,
then there are no solutions at all. If a = b = 0 and c 6= 0, then |z − d/c| = 0, and
the radius of the circle is 0, which gives a point solution. The situation is similar if
c = d = 0. 4

Exercise 1.11. Sketch the set {z ∈ C : |z − 3 − 2i| < 4, Re(z) > 0} in the
complex plane. Is it open, closed, bounded, compact, polygonally path-connected,
simply polygonally connected, a region, or a domain?

Answer. When |z − 3 − 2i| < 4, the distance of z from 3 + 2i is at most 4, that

is, z lies inside the circle with centre 3 + 2i and radius 4. This is the region shaded
with lines with slope −1 in the figure. Further, if Re(z) > 0, then z lies to the right
of the imaginary axis. This is the region shaded with lines with slope 1 in the figure.
The desired set is the region where both inequalities hold, that is, the region with
cross-hatching.
The set does not include points on the circumference of the circle, nor does it
include the points on the imaginary axis that also lie inside the circle.
The set is open, connected, simply connected and bounded, and hence it is also
a domain and a region. It is not closed, so not compact. 4

Im
z1 b

Re
z5 b

z4
Ω

Figure 1.2. A polygonally path-connected, but not simply polygo-

nally connected, set
 3. DESCRIBING SETS IN THE COMPLEX PLANE 11

Im Im

4 4
b b 3
3 + 2i Re 3 + 2i Re

Figure 1.3. Two regions defined by inequalities

Exercise 1.12. Sketch the set {z ∈ C : 4 ≤ |z − 3 − 2i| ≤ 5} in the complex

plane. Is it open, closed, bounded, compact, polygonally path-connected, simply
polygonally connected, a region or a domain?

Answer. The set includes points inside or on the circle with centre 3 + 2i and
radius 5, but outside or on the circle with centre 3 + 2i and radius 4. It includes the
points on the two concentric circles. The sketch is on the right in Figure 1.3.
The set {z ∈ C : 4 < |z − 3 − 2i| < 5} is open, because all points are interior
points. The set we are considering is this set together will all its boundary points,
so it is closed, and also a region. It is not open, because some points are boundary
points, and hence not a domain. It is closed and bounded, hence also compact. It
is connected, but not simply connected, because there are points in the complement
of the set inside a polygonal path around the annulus. 4

Here is a more complicated example, related to conic sections.

Exercise 1.13. Sketch the set {z ∈ C : |z + i| + |z − i| = 4} in C. Is it open or
bounded? Describe the set {z ∈ C : |z +i|+|z −i| < 4}. Is it open, closed, bounded,
compact, polygonally path-connected, simply polygonally connected, a region or a
domain?

Answer. First we describe the set {z ∈ C : |z + i| + |z − i| = 4} using algebra.

|z + i| + |z − i| = 4
|z + i| = 4 − |z − i|
|z + i| = 16 − 8|z − i| + |z − i|2
2

x2 + (y + 1)2 = 16 − 8|z − i| + x2 + (y − 1)2

8|z − i| = 16 − 4y
2|z − i| = 4 − y
4 (x + (y − 1)2 ) = 16 − 8y + y 2
2

x2 y 2
+ = 1.
3 4
This is the circumference of an ellipse, as shown in Figure 1.4. The ellipse is the set
of points for which the sum of the distances from i and −i is exactly 4. It is closed
but not open, and bounded.
12 1. INEQUALITIES AND SETS OF COMPLEX NUMBERS

Im
b

2i
b

√
3 b
Re

Figure 1.4. An ellipse

The sets
{z ∈ C : |z + i| + |z − i| > 4} and {z ∈ C : |z + i| + |z − i| < 4}
are the exterior and interior of the ellipse. 4

(1) Note that connected is not defined for closed sets, but there are questions
about closed sets being connected.
(2) Note that simply connected does not imply connected, or vice versa.
(3) Point out that it is best to use Cartesian coordinates late in a calculation.
(4) It is not important to do the last example, but if it is done, then it is nice
to make a comment that the sign of |z + i| + |z − i| − 4 has to be constant
on connected components of the complement of the set where it is 0.
 LECTURE 2

Functions of a complex variable

In this lecture, we introduce functions of a complex variable, and recall concepts

such as domain and range. We examine some examples and consider the problem
of estimating the size of a complex function.

1. Functions
In Mathematics, we often think of functions as machines: you give the machine
a number, x say, press a button, and out comes f (x). We sometimes write x 7→ f (x)
to indicate that x is the input and f (x) is the output.
• The domain of a function f , written Domain(f ), is the set of all the numbers
you are allowed to put in. Sometimes this is restricted in some way. If there
is no explicit restriction, you should consider the natural domain, that is,
the largest domain possible.
• A codomain is a set of numbers that includes all the numbers that you can
get out, and perhaps more.
• The range or image of a function f , written Range(f ), is the set of the
numbers that you can get out, and no others.
• The image of a subset S of the domain of a function f , sometimes written
f (S), is the set of all possible values f (s) as s varies over S.
• The preimage of a subset T of the codomain of a function f , sometimes
written f −1 (T ), is the set of all x in Domain(f ) such that f (x) ∈ T .
Definition 2.1. A complex function is one whose domain, or whose range, or
both, is a subset of the complex plane C that is not a subset of the real line R.
To emphasize that the domain is complex, not real, the expression function of a
complex variable may be used. To emphasize that the range is complex, not real,
the expression complex-valued function may be used.
Remark 2.2. The word ”domain” is then used in two different ways in this
course. We say that a subset of C is a domain if it is open and connected which has
nothing to do with a function. We just saw that the domain of a function f is the
set of points where f is defined. Be careful to not confuse them and note that the
domain of a function is not necessarily a domain in the sense of open and connected:
for instance consider f : {0, 1} → C, 0 7→ 0, 1 7→ 1 and not that its domain is equal
to two points which forms a nonconnected and nonopen subset of C.

2. Examples of functions
Examples of functions of a complex variable include the real part function Re,
the imaginary part function Im, the modulus function z 7→ |z|, and the principal
value of the argument Arg; these are all real-valued. Complex conjugation z 7→ z̄ is
an example of a complex-valued function of a complex variable.
13
14 2. FUNCTIONS OF A COMPLEX VARIABLE

In this course, we are going to learn about a number of useful complex functions.
Shortly we will define complex polynomials and then rational functions. In future
lectures, we will define log z, sin z, and cosh z for a complex number z, and there
are many other functions in the menagerie of complex functions.
Exercise 2.3. Suppose that f (z) = 1/z for all z ∈ C \ {0}, and that g(z) = z
for all z ∈ C. Show that f ◦ f (z) = z for all z ∈ C \ {0}. Is f ◦ f = g?
Answer. By definition,
f ◦ f (z) = f (1/z) = 1/(1/z) = z = g(z).
However, the domain of f ◦ f is C \ {0} and the domain of g is C, so these functions
are different. 4

Definition 2.4. A complex polynomial is a function p : C → C of the form

p(z) = ad z d + · · · + a1 z + a0 ,
where ad , . . . , a1 , a0 ∈ C. If ad 6= 0, we say that p is of degree d. A rational function
is a quotient of polynomials.
Sums, differences, products and compositions of polynomials are polynomials.
Theorem 2.5 (The fundamental theorem of algebra). Every nonconstant com-
plex polynomial p of degree d factorizes uniquely: there exist α1 , α2 , . . . , αd and c in
C such that
Yd
p(z) = c (z − αj ) ∀z ∈ C.
j=1
Equivalently, every nonconstant complex polynomial has at least one root.
In the factorisation above, the roots αj may occur more than once. Thus we
could also write
Ye
p(z) = c (z − αj )mj ,
j=1

where the αj are all distinct, and the multiplicities mj add to give the degree of p.
Theorem 2.6 (Polynomial division and partial fractions). Suppose that p and q
are polynomials of degrees m and n. Then the rational function p/q may be written
as a sum
p(z) r(z)
= s(z) + ,
q(z) q(z)
where r and s are polynomials, and the degree of r is strictly less than the degree of
q. Further, if
Y e
q(z) = c (z − βj )mj ,
j=1

then we may decompose the term r/q into partial fractions:

r(z) X X
e mj
ajk
= .
q(z) j=1 k=1
(z − βj )k
 3. REAL AND IMAGINARY PARTS 15

At this stage, we do not prove these results, which should be familiar, though
perhaps not in this generality; we will give proofs later.
The natural domain of any complex polynomial is C. Sometimes we cannot
determine the range of a real polynomial exactly, because we cannot find maxima
or minima exactly. However, for complex polynomials, things are easier.
Exercise 2.7. Suppose that p is a nonconstant complex polynomial. Show that
the range of p is C.
Answer. Take a nonconstant complex polynomial p, and a complex number w.
We need to show that there is z ∈ C such that p(z) = w. Define q(z) = p(z) − w.
Then q is also a nonconstant complex polynomial, so has a root by the fundamental
theorem of algebra. That is, there exists z ∈ C such that q(z) = 0. It follows that
p(z) = w, as required. 4

3. Real and imaginary parts

To a function f : S → C, where S ⊆ C, we associate two real-valued functions u
and v of two real variables:
f (x + iy) = u(x, y) + iv(x, y).
Then u(x, y) = Re f (x + iy) and v(x, y) = Im f (x + iy). It is very useful and very
important to be able to view a complex-valued function of a complex variable in
this way.
Exercise 2.8. Suppose that f (z) = z and that g(z) = z 2 . Find the real and
imaginary parts of f and g.
Answer. Clearly, Re f (x + iy) = x and Im f (x + iy) = y. Further, if z = x + iy,
then z 2 = x2 + 2ixy − y 2 , so Re g(x + iy) = x2 − y 2 and Im g(x + iy) = 2xy. 4

Exercise 2.9. Suppose that f (z) = z 3 + z − 2. Write the real and imaginary
parts of this function as functions u and v of (x, y), where z = x + iy.
Answer. Observe that
f (x + iy) = (x + iy)3 + x + iy − 2
= x3 + 3ix2 y − 3xy 2 − iy 3 + x − iy − 2
= (x3 − 3xy 2 + x − 2) + i(3x2 y − y 3 − y).
Thus u(x, y) = x3 − 3xy 2 + x − 2 and v(x, y) = 3x2 y − y 3 − y. 4

Exercise 2.10. Suppose that f (z) = 1/z. Write the real and imaginary parts
of this function as functions of x and y, where z = x + iy.
Answer. Note that
 
1 x − iy x −y
f (x + iy) = = 2 = 2 +i .
x + iy x + y2 x + y2 x2 + y 2
x −y
Thus Re f (x + iy) = 2 and Im f (x + iy) = . 4
x + y2 x2 + y 2
16 2. FUNCTIONS OF A COMPLEX VARIABLE

Exercise 2.11. Write ez in the form u(x, y) + iv(x, y), where z = x + iy.
Answer. Observe that
ez = ex (cos(y) + i sin(y)) = ex cos(y) + iex sin(y) = u(x, y) + iv(x, y),
where u(x, y) = ex cos(y) and v(x, y) = ex sin(y). 4
Sometimes we view the complex number z in polar coordinates, that is, we write
z = reiθ . In this case, we consider
f (z) = u(r, θ) + iv(r, θ).
Exercise 2.12. Write ez in the form u(r, θ) + iv(r, θ), where z = reiθ .
Answer. Observe that
ez = er cos θ+ir sin θ
= er cos θ (cos(r sin θ) + i sin(r sin θ))
= er cos θ cos(r sin θ) + i er cos θ sin(r sin θ).
Thus u(r, θ) = er cos θ cos(r sin θ) and v(r, θ) = er cos θ sin(r sin θ). 4

4. The function z 7→ 1/z

It is obvious that if w = 1/z, then z = 1/w, and the function z 7→ 1/z is one-to-
one (injective). Further, the domain and the range of the function are both equal
to C \ {0}.
Exercise 2.13. Suppose that z varies on the line x = 1, and let w = 1/z. Show
that w varies on the circle |w − 21 | = 12 .
Answer. We set w = u + iv. If z is on the line x = 1, then Re z = 1, whence
Re(1/w) = 1 (and w 6= 0). Now
 
w̄ u
1 = Re = 2 .
|w| 2 u + v2
It follows that u2 + v 2 = u, whence (u − 21 )2 + v 2 = 41 , and the result follows. 4
We can reverse the argument and show that every point on the circle except 0
arises in this way. Thus the image of the line is the circle with the point 0 removed.

5. Fractional linear transformations

The fractional linear transformations form an important family of complex func-
tions. These are the functions of the form
az + b
f (z) = ,
cz + d
where a, b, c, d ∈ C and ad − bc 6= 0. We will study these functions in more detail
later, but at the moment we just point out that if f is a fractional linear transfor-
mation and z varies on a line, then f (z) varies on a line or on a circle. The same
holds if z varies along a circle. Note that when z → −d/c, then cz + d → 0 and
f (z) → ∞ (we will define limits formally later). We can tell whether f (z) varies
on a line or on a circle as follows: if the points where z varies include −d/c, then
 6. ESTIMATING THE SIZE OF THE VALUES OF A FUNCTION 17

the points where f (z) varies will include ∞, and this means that f (z) must vary
on a line. Conversely, if the points where z varies do not include −d/c, then f (z)
will stay bounded, and this means that f (z) must vary on a circle. Once we know
whether f (z) varies on a line or on a circle, we may find the equation of the line or
the circle quite easily by finding a few values of f (z).
Example 2.14. Let f (z) = 1/z. As z varies on the line x = 1, its image f (z)
varies on a circle, because z stays away from 0 and so 1/z stays away from ∞.
This circle passes through the points 1 and 0 (since f (z) → 0 as z → ∞), and is
symmetric about the real axis, since 1/(1 − it) = (1/(1 + it)) . This must be the
circle that we found above.

6. Estimating the size of the values of a function

We will need to use what we know about inequalities to estimate how large the
values of a complex function are.
1
Exercise 2.15. Suppose that f (z) = for all z ∈ C \ {±1, ±i}. Show that
z4 −1
1
|f (z)| ≤
15
if |z| ≥ 2 (that is, if z lies on or outside the circle with centre 0 and radius 2).
Answer. If |z| ≥ 2, then by a variant of the triangle inequality from Lecture 1,
|z 4 − 1| ≥ |z 4 | − |1| = |z|4 − 1 ≥ 16 − 1 = 15.
Hence
1 1 1
|f (z)| = = 4 ≤ ,
z4 −1 |z − 1| 15
as required. 4

Exercise 2.16. Suppose that p(z) = 10z 4 − 3z 3 + z − 10. Show that when |z|
is large enough, |p(z)| ≤ 11|z|4 .
Answer. Write p(z) = z 4 (10 − 3z −1 + z −3 − 10z −4 ). If |z| > 10, then, by the
triangle inequality applied several times,
3 1 10
|10 − 3z −1 + z −3 − 10z −4 | ≤ 10 + + + ≤ 11,
10 1000 10000
and so |p(z)| ≤ 11|z|4 when |z| > 10. 4
 LECTURE 3

Sketching complex functions

We often use graphical methods to gain useful intuition about complex functions,
and we spend some time investigating these. It is hard to represent complex func-
tions, because there are up to four variables involved. Just as we often write y = f (x)
for a real function, it is common to consider a function in the form w = f (z), and
to use real variables x and y to describe the domain and u and v to describe the
range. Typically, we draw “elementary” curves in the z plane, such as lines parallel
to the axes, or concentric circles around and rays exiting from the origin, and then
examine their images in the w plane, or we draw similar elementary curves in the w
plane and then examine their preimages.

1. Linear and affine mappings

Suppose that a 6= 0, and consider the bijective linear map z 7→ az. We write
a = c + id and z = x + iy; then
az = (c + id)(x + iy) = (cx − dy) + i(cy + dx).
In Cartesian coordinates, the map may be written
    
x c −d x
7→ .
y d c y
Note that    
c −d cos θ − sin θ
=r ,
d c sin θ cos θ
√
where r = c2 + d2 , while cos θ = c/r and sin θ = d/r. Hence multiplying by the
matrix  
c −d
d c
is the same as rotating through the angle θ and then dilating by the real number
r. This corresponds to the representation of a in the form reiθ , where r = |a| and
θ = Arg(a).
An affine map of the complex plane is a map of the form z 7→ az + b; such
mappings are also bijective (as long as a 6= 0). We may also represent this as a map
from R2 to R2 . We write a = c + id and b = e + if . Then in Cartesian coordinates,
we have       
x c −d x e
7→ + .
y d c y f
It is an exercise in algebra to see that the image of a line under an affine mapping
is a line, and the image of a circle under an affine mapping is a circle.
19
20 3. SKETCHING COMPLEX FUNCTIONS

Im Im

Re Re

Figure 3.1. The translation z 7→ z + 1

Im Im

Re Re

Figure 3.2. The multiplication z 7→ (1 + i)z

The inverse of an affine mapping is an affine mapping. It follows that the pre-
images of lines are lines and preimages of circles are circles; the preimage of a grid
parallel to the axes is a rectangular grid, but not necessarily parallel to the axes.
In Figures 3.1 and 3.2, we illustrate the images of lines parallel to the axes and
circles around the origin under affine mappings.
This is a good way to show how the function behaves, although a lot of space is
used and care is needed to choose the points that are moved in a way that is not
ambiguous. Indeed, it might be better to draw a very asymmetrical figure in the z
plane and then its image in the w plane.
Sometimes we just draw the right hand figure of the two drawn above, labelling
the curves with the corresponding curve in the domain of the function. In Figure 3.1,
these are the circles r = 1, r = 2, and r = 3; in Figure 3.2, they are the horizontal
lines x = 0, x = ±1, x = ±2, and the vertical lines y = 0, y = ±1, y = ±2.
On the other hand, we may look for curves in the xy plane whose images in the
uv plane are the lines u = c and v = d. So we are finding the level curves of the
 3. THE FUNCTION w = 1/z 21

real and imaginary parts of the function. People who are used to map reading can
build a picture in their mind of terrain, just knowing the contours that represent
different heights. They imagine a surface above the page at the height indicated by
the contour, and then fill in the gaps.

2. Quadratic functions
Now we consider the function z 7→ z 2 . Notice that this function is two-to-one in
B◦ (0, ∞).
On the one hand, we may represent the images in the uv plane of the curves in
the xy plane given by x = a and y = b, or by r = a and θ = b. For instance, if x = a
and y = t, where a is fixed and t varies, then
w = z 2 = (a + it)2 = a2 − t2 + 2iat.
That is, u = a2 − t2 and v = 2at. We eliminate t to show that
v2
u = a2 − .
4a2
Alternatively, if y = b and x = t, where b is fixed and t varies, then
w = z 2 = (t + ib)2 = t2 − b2 + 2ibt.
That is, u = t2 − b2 and v = 2bt. We eliminate t to show that
v2
u= − b2 .
4b2
See, for example, Figure 3.3.
Exercise† 3.1. Find the focus and the directrix of the parabola u = v 2 /4b2 − b2
in the uv plane.
On the other hand, we may look for the values of x and y so that Re(z 2 ) or
Im(z 2 ) takes a fixed value. For instance, if Re(z 2 ) = a, then
x2 − y 2 = a,
and this is a hyperbola opening to the left and right, or up and down, depending on
the sign of a. Similarly, if Im(z 2 ) = b, then
2xy = b,
and this is a right hyperbola in the first and third quadrants, or in the second and
fourth quadrants, depending on the sign of b. See, for example, Figure 3.3.
Exercise 3.2. How would you “sketch the graph” of w = (z − 1)2 − 1?

3. The function w = 1/z

As we saw in the last lecture, the function w = 1/z is one-to-one. Further, it
sends lines through the origin to lines through the origin. Actually, this is easiest
to see using polar coordinates: as r varies in R+ , the point z = reiθ varies along a
ray from the origin. Now w = (1/r)e−iθ , and 1/r also varies in R+ and this point
too varies along a ray. However the new ray is the reflection of the old ray in the
real axis. Since moreover diametrically opposed rays are mapped to diametrically
opposed rays, lines through the origin do indeed go to lines through the origin.
22 3. SKETCHING COMPLEX FUNCTIONS

Im Im

Re Re

Figure 3.3. Images of lines x = c and y = d for w = z 2

Im Im

Re Re

Figure 3.4. Images of curves r = c and θ = d for w = z 2

Im Im

Re Re

Figure 3.5. The level curves for Re(z 2 ) and Im(z 2 )

Similarly, as θ varies, the point reiθ moves around a circle centred at the origin.
Now w = (1/r)e−iθ , and this point moves around the same circle, but in the opposite
direction.
 5. MORE ON GRAPHICAL REPRESENTATIONS OF COMPLEX FUNCTIONS 23

Im Im

Re Re

Figure 3.6. Images of curves r = c and θ = d for w = 1/z

We represent this graphically in Figure 3.6.

For completeness, we now describe the images of lines and circles under the map
w = 1/z in more detail.
Lemma 3.3. Consider the mapping w = 1/z.
(1) The image of a line through 0 (not including 0) is a line through 0 (not
including 0).
(2) The image of a line that does not pass through 0 is a circle through 0, with
0 removed. If p is the closest point on the line to 0, then the line segment
between 0 and 1/p is a diameter of the circle.
(3) The image of a circle that passes through 0 is a line. If q is the furthest
point on the circle from 0, then the closest point on the line to 0 is 1/q.
(4) The image of a circle that does not pass through 0 is a circle. If p and q
are the points on the circle closest to and furthest from 0, then the points
on the image circle closest to and furthest from 0 are 1/q and 1/p.
Proof. See the exercise sheet. □
Exercise 3.4. Suppose that z varies on the line ax + by = c, where a, b, c ∈ R,
and let w = 1/z. Show that w varies on a line when c = 0 and on a circle otherwise.
Answer. We leave this exercise to the reader. 4

4. The exponential function

The exponential function w = ez is ∞-to-1; that is, infinitely many different
points in the xy plane are sent to the same point in the uv plane. See Figure 3.7 for
a graphical representation.

5. More on graphical representations of complex functions

There are many good web-sites that explore different ways to represent complex
functions.
24 3. SKETCHING COMPLEX FUNCTIONS

Im Im

Re Re

Figure 3.7. Images of curves x = c and y = d for w = ez

 LECTURE 4

Fractional linear transformations

In this lecture, we study a particular type of function: fractional linear trans-

formations. These are easy to handle because we can use linear algebra to simplify
computations.

1. Domains and ranges of fractional linear transformations

Let M be a 2 × 2 complex matrix:
 
a b
M= .
c d
We write TM for the associated fractional linear transformation:
az + b
TM (z) = . (4.1)
cz + d
First, we need to assume, and will always do so, that (c, d) 6= (0, 0), otherwise
the denominator is always 0. If det M = 0, then (a, b) = λ(c, d) for some λ ∈ C,
whence az + b = λ(cz + d) and so f (z) = λ for all z ∈ Domain(f ). Thus fractional
linear transformations associated to matrices with determinant 0 are essentially just
constant functions, and we do not consider them further.
Observe now that TλM = TM if λ 6= 0, and recall that det(λM ) = λ2 det(M ).
This means that when det M 6= 0, if we set M ′ = (det M )−1/2 M , then det M ′ = 1
and TM = TM ′ . Thus we may and shall henceforth restrict our attention to fractional
linear transformations associated to matrices with determinant 1.
First we find the domain of TM . If c = 0 and d 6= 0, then the denominator
is never 0, so TM (z) is defined for all a, that is, Domain(TM ) = C; otherwise, the
denominator is nonzero when z 6= −d/c, whence Domain(TM ) = C \ {−d/c}.
Now suppose that w ∈ Range(TM ). Then
az + b
w= =⇒ wcz + wd = az + b =⇒ z(wc − a) = b − wd,
cz + d
and as long as wc 6= a, this in turn implies that
b − wd (−d)w + b dw − b
z= = = = TM ′ (w) , (4.2)
wc − a cw + (−a) (−c)w + a
where  
′ d −b
M = .
−c a
Then M ′ is exactly M −1 , since det M = 1. We conclude that
(
C \ {a/c} if c 6= 0
Range(TM ) =
C if c = 0.

25
26 4. FRACTIONAL LINEAR TRANSFORMATIONS

Our discussion of the domain and range had several cases; we can simplify the
statements by enlarging the set of complex numbers by adding ∞. Indeed, we define
az + b az + b a
f (−d/c) = lim = ∞ and f (∞) = lim =
z→−d/c cz + d z→∞ cz + d c
(we will define limits formally in the next lecture) and now we can just write
TM : C ∪ {∞} → C ∪ {∞}.
We often call C ∪ {∞} the Riemann sphere, and write it S: we imagine a unit
sphere in three dimensions with centre at 0. We can define a function σ : C → S
geometrically by joining a point p in the plane to the north pole n of the sphere by a
straight line. The line will cut the sphere at n and at one other point, which we call
σ(p). Then we may think of n as being σ(∞). The function σ is called stereographic
projection.

2. Matrix products and composition of mappings

Exercise 4.1. Suppose that M, N ∈ M2,2 (C). Show that TM TN = TM N (on
the Riemann sphere). Deduce that the transformation TM is bijective and that
(TM )−1 = TM −1 .
   
a b e f
Answer. Write M = and N = . Then we compute
c d g h
    
a b e f ae + bg af + bh
MN = = .
c d g h ce + dg cf + dh
Further,
ez + f
a +b
aTN (z) + b gz + h
TM TN (z) = =
cTN (z) + d ez + f
c +d
gz + h
a(ez + f ) + b(gz + h) (ae + bg)z + (af + bh)
= = .
c(ez + f ) + d(gz + h) (ce + dg)z + (cf + dh)
This formula holds for all z ∈ C for which none of the denominators is 0, and then
extends to all z ∈ S by continuity. It follows that
TM TN (z) = TM N (z) ∀z ∈ S.
In particular, TM TM −1 = TI , and TI is the identity mapping. Hence TM is
invertible, with inverse TM −1 . 4

3. Factorisations of fractional linear transformations

Matrices may be factorised, and hence fractional linear transformations may be
factorised too.
Theorem 4.2. Every 2 × 2 complex matrix with determinant 1 may be written
as a product of at most three matrices of the following special types:
   
a b 0 1
and .
0 d −1 0
 3. FACTORISATIONS OF FRACTIONAL LINEAR TRANSFORMATIONS 27
 
a b
Proof. Consider the matrix . We consider three cases, according to
c d
whether any of a or c is 0. since the determinant is 1, both cannot be 0.
If c = 0, the matrix itself is of the desired form.
Now suppose that a = 0. Then
    −1  
0 b 0 1 0 1 0 b
=
c d −1 0 −1 0 c d
   
0 1 0 −1 0 b
=
−1 0 1 0 c d
  
0 1 −c −d
= ,
−1 0 0 b
which is of the desired form.
Finally, if neither a nor c is 0, then we take x = a/c, and write
    −1  
a b 1 x 1 x a b
=
c d 0 1 0 1 c d
   
1 x 1 −x a b
=
0 1 0 1 c d
  
1 x a − cx b − dx
= ;
0 1 c d
since a − cx = 0, combining with the previous case shows that the factorisation
holds in this case too. □

This factorisation simplifies a number of arguments; the next result is an exam-

ple.
Theorem 4.3. Let TM be a fractional linear transformation. Then the image of
a line under TM is a line or a circle, and the image of a circle under TM is also a
line or a circle.
 
a b
Proof. If M is of the form , then TM is an affine transformation, and
0 d
the theorem holds in this case. Otherwise, TM is composed of some affine transfor-
mations and the inversion map z 7→ −1/z; it therefore suffices to treat the inversion
map.
We may write the equation of the circle with centre c and radius r in the form
|z − c|2 = r2 . Note that this is the same as |z|2 − 2 Re(zc̄) + |c|2 − |r|2 = 0.
If we set w = −1/z, then we find that
−1
− c = r2 ,
w
whence
|wc + 1| = r2 |w|2 ,
and


|c|2 − r2 |w|2 + 2 Re(wc) + 1 = 0,
28 4. FRACTIONAL LINEAR TRANSFORMATIONS

which is the equation of a circle, unless |c|2 = r2 , in which case we get

2 Re (wc) = −1,
which is the equation of a straight line.
The argument to show that the inversion mapping sends straight lines to straight
lines or circles is similar, and we omit it; the starting point is that the equation of
a straight line may be written as |z − p| = |z − q|. □
Remark 4.4. We are slightly imprecise here. Given a fractional linear trans-
formation TM (e.g. TM (z) = −1/z) and X a circle or a line (e.g. X = C(1, 1) the
circle centre 1 radius 1), it might happen that TM is defined on X \ {x} where x is
precisely the point that will make vanish the denominator of TM (e.g. x = 0). We
are then looking at TM (X \ {x}) rather than TM (X). The image is then alway a
line as we will see in the next corollary.
a b
Corollary 4.5. Let TM be a fractional linear transformation with M =
c d
satisfying that c 6= 0. If X ⊂ C is a line or a circle containing the point −d/c, then
TM (X \ {−d/c}) is a line.
Proof. By the previous theorem we have that TM (X \ {−d/c}) is either a line
or a circle. Note that limz→−d/c TM (z) = ∞ (see the next section on limits) implying
that TM (X \ {−d/c}) is unbounded. Therefore, TM (X \ {−d/c}) must be a line. □
z−i
Exercise 4.6. Show that the fractional linear transformation T : z 7→
z+i
sends the upper half plane onto the unit disc {w ∈ C : |w| < 1}.
Find the image under T of the sector {z ∈ C : φ < Arg(z) < π − φ}.
Answer. We know that T is invertible and may check easily that T −1 is the
w+1
fractional linear transformation w 7→ −i .
w−1
Suppose that z = x + iy, where y > 0. Then
2
x + iy − i x2 + (y − 1)2
|T z| =
2
= < 1,
x + iy + i x2 + (y + 1)2
so T maps into the unit disc. On the other hand, if |w| < 1, then
w+1 (w + 1)(w̄ − 1) i(w − w̄) + i(1 − ww̄)
T −1 w = −i = −i = ,
w−1 (w − 1)(w̄ − 1) |w − 1|2
so Im(T −1 w) = (1 − ww̄)/(|1 − w|2 ) > 0; this implies that w is in the image of the
upper half plane under T , so T maps onto the unit disc. Further, T (0) = −1 and
T (∞) = 1.
Let Rθ denote the ray {reiθ : r ∈ R+ }. Each ray Rθ is mapped into the part of a
line or a circle that lies inside the unit disc and starts at −1 and ends at 1. Further,
eiθ − i eiθ − eiπ/2 eiθ/2−iπ/4 − eiπ/4−iθ/2
T eiθ = = = = i tan(θ/2 − π/4).
eiθ + i eiθ + eiπ/2 eiθ/2+iπ/4 − eiπ/4−iθ/2
It follows that the image of the ray is Aθ , the arc of the circle with centre (0, tan(θ))
and radius sec θ between (−1, 0) and (1, 0).
Hence the image of the sector is the “eye-shaped” region between the arcs Aθ
and Aπ−θ . 4
 4. SPECIAL CLASSES OF FRACTIONAL LINEAR TRANSFORMATIONS 29

4. Special classes of fractional linear transformations

Exercise 4.7. Suppose that M ∈ M2,2 (R) and det(M ) = 1. Show that if
Im(z) > 0, then Im(TM z) > 0. Deduce from this and Exercise 4.1 that TM maps
the upper half plane {z ∈ C : Im(z) > 0} onto itself bijectively.
 
a b
Answer. Suppose that M = , where a, b, c, d ∈ R. Then
c d
az + b (az + b)(cz + d)
TM (z) = =
cz + d (cz + d)(cz + d)
(az + b)(cz̄ + d) ac|z|2 + adz + bcz̄ + bd
= = .
|cz + d|2 |cz + d|2
Suppose that z = x + iy, where y > 0. Then
 ac|z|2 + adz + bcz̄ + bd  Im ac|z|2 + adz + bcz̄ + bd

Im(TM (z)) = Im =
|cz + d|2 |cz + d|2
Im(ad(x + iy) + bc(x − iy)) Im((ad + bc)x + i(ad − bc)y)
= =
|cz + d| 2 |cz + d|2
(ad − bc)y y
= = > 0.
|cz + d| 2 |cz + d|2
Thus Im(TM (z)) > 0, as required.
From Exercise 4.1, TM is bijective. It remains to show that if Im(w) > 0, then
the complex number z such that TM (z) = w satisfies Im(z) > 0. But TM (z) = w
is the same as z = (TM )−1 w = TM −1 w; by linear algebra, M −1 ∈ M2,2 (R) and
det(M −1 ) = 1. The first part of this answer, applied to M −1 rather than M , shows
that Im(TM −1 w) > 0, and thus Im(z) > 0, as required. 4
 LECTURE 5

Limits and continuity

In this lecture, we outline the key ideas and facts about limits and continuity,
as a preliminary to defining differentiability.

1. Limits
We define limits for complex functions much as for real functions.
Recall that, given a set S, we define its closure S̄ or S to be the set consisting
of all points of S together with all boundary points.
Definition 5.1. Suppose that f is a complex function, ℓ ∈ C, and z0 is in
Domain(f ) . We say that f (z) tends to ℓ as z tends to z0 , or that ℓ is the limit of
f (z) as z tends to z0 , and we write f (z) → ℓ as z → z0 , or
lim f (z) = ℓ,
z→z0

if, for every ε ∈ R+ , there exists δ ∈ R+ such that |f (z) − ℓ| < ε provided that z is
in Domain(f ) and 0 < |z − z0 | < δ.
Suppose also S is a subset of Domain(f ) and that z0 ∈ S̄. We say that f (z)
tends to ℓ as z tends to z0 in S, or that ℓ is the limit of f (z) as z tends to z0 in S,
and we write f (z) → ℓ as z → z0 in S, or
lim f (z) = ℓ,
z→z0
z∈S

if, for every ε ∈ R+ , there exists δ ∈ R+ such that |f (z) − ℓ| < ε provided that z ∈ S
and 0 < |z − z0 | < δ.
Informally, f (z) tends to ℓ if we can make f (z) arbitrarily close to ℓ by taking z
close to, but not equal to, z0 .
Most of what follows about limits of the form limz→z0 f (z) also applies to re-
stricted limits, that is, limits of the form limz→z0 f (z).
z∈S
We may rewrite the conditions 0 < |z − z0 | < δ and |f (z)−ℓ| < ε as z ∈ B◦ (z0 , δ)
and f (z) ∈ B(ℓ, ε). We define limits involving infinity in a similar way by defining
balls centred at infinity, and extending our previous definition slightly.
Definition 5.2. Suppose that ε > 0. We define both B(∞, ε) and B◦ (∞, ε) to
be the set {z ∈ C : |z| > 1/ε}.

Definition 5.3. Suppose that f is a complex function, that ℓ ∈ C ∪ {∞}, and that
either z0 ∈ Domain(f ) or Domain(f ) is unbounded and z0 = ∞. We say that f (z)
tends to ℓ as z tends to z0 , or that ℓ is the limit of f (z) as z tends to z0 , and we
write f (z) → ℓ as z → z0 , or
lim f (z) = ℓ,
z→z0

31
32 5. LIMITS AND CONTINUITY

if for all ε ∈ R+ , there exists δ ∈ R+ such that f (z) ∈ B(ℓ, ε) provided that
z ∈ Domain(f ) ∩ B◦ (z0 , δ).
With this definition, the following lemma holds; we omit the proof.
Lemma 5.4 (Standard limits). Suppose that α, c ∈ C. Then
lim c = c lim c = c
z→α z→∞
lim z − c = α − c lim z − α = ∞
z→α z→∞
1 1
lim =∞ lim = 0.
z→α z − α z→∞ z − α

As the statement of these limits indicates, we are sometimes allowed to consider

∞ as a limit in this course.
The next results follows from the definition; we omit the proofs, which generalise
arguments from calculus.
Lemma 5.5. Suppose that f is a complex function, that T ⊆ S ⊆ Domain(f ),
and that z0 ∈ T̄ . If limz→z0 f (z) exists, then so does limz→z0 f (z), and these limits
z∈S z∈T
are equal.
Lemma 5.6. Suppose that f is a complex function, and that z0 ∈ Domain(f ) .
If limz→z0 f (z) exists, then it is unique.
We may break complicated limits up into sums, products, and so on, of simpler
limits.
Theorem 5.7. Suppose that f and g are complex functions and that c ∈ C.
Then
lim cf (z) = c lim f (z)
z→z0 z→z0
lim f (z) + g(z) = lim f (z) + lim g(z)
z→z0 z→z0 z→z0
lim f (z) g(z) = lim f (z) lim g(z)
z→z0 z→z0 z→z0
f (z) limz→z0 f (z)
lim = ,
z→z0 g(z) limz→z0 g(z)
in the sense that if the right hand side exists, then so does the left hand side, and
they are equal. In particular, for the quotient, we require that limz→z0 g(z) 6= 0.
We also omit the proof of this theorem, which is very similar to that of the
corresponding theorem for limits of functions of a real variable.
Limits respect complex conjugation and related operations.
Theorem 5.8. Suppose that f is a complex function and that either Domain(f )
is unbounded and z0 = ∞ or z0 ∈ Domain(f ) . Then
lim f (z) = lim f (z)
z→z0 z→z0
lim Re(f (z)) = Re lim f (z)
z→z0 z→z0
lim Im(f (z)) = Im lim f (z)
z→z0 z→z0
lim f (z) = lim Re(f (z)) + i lim Im(f (z)),
z→z0 z→z0 z→z0
 2. EXAMPLES OF LIMITS 33

Im

δ π − sin−1 (δ/a) Re
z b

−a

Figure 5.1. The trigonometry of the argument function

in the sense that if the right hand side exists, then so does the left hand side, and
they are equal. In particular, f (z) tends to ℓ as z tends to z0 if and only if Re(f (z))
tends to Re(ℓ) and Im(f (z)) tends to Im(ℓ) as z tends to z0 .
Proof. The proof of part (1) uses that fact that
f (z) − ℓ̄ = |f (z) − ℓ| .
The rest follows from the first part and the first two parts of Theorem 5.7. □

2. Examples of limits
Exercise 5.9. Show from first principles that limz→z0 z = z0 .
Answer. Given ε ∈ R+ , take δ = ε. Then if 0 < |z − z0 | < δ, it follows
immediately that |z − z0 | < ε, and the result is proved. 4

Exercise 5.10. Suppose that f (z) = z 2 − z̄ + i. Does limz→2i f (z) exist: if so

find it, and if not, explain why not.
Answer. By the algebra of complex limits,

2

lim z 2 − z̄ + i = lim z − lim z + i = −4 + 2i + i = −4 + 3i. 4
z→2i z→2i z→2i

We now consider an important example.

Exercise 5.11. Suppose that a > 0. Show that
lim Arg(z) = π and lim Arg(z) = −π.
z→−a z→−a
Im(z)≥0 Im(z)<0

Does limz→−a Arg(z) exist, and if so, what is it?

Answer. We consider the limit where Im z ≥ 0. If δ < a/2, then 0 ∈ / B̄(−a, δ)

−1
and sin (δ/a) < π/6; so we may represent the exercise as in the diagram. In this
case, if z ∈ B(−a, δ) and Im(z) ≥ 0, then
π − sin−1 (δ/a) < Arg(z) ≤ π.
To be sure that |Arg(z) − π| < ε, it suffices to ensure that sin−1 (δ/a) < ε; to
do this, and ensure that δ < a/2, take δ = a sin(min{ε, π/6}). This shows that
lim z→−a Arg(z) = π.
Im(z)≥0
34 5. LIMITS AND CONTINUITY

Similarly, lim z→−a Arg(z) = −π.

Im(z)<0
By the uniqueness of limits, limz→−a Arg(z) does not exist. 4

We may also show that

lim Arg(z) = θ.
z→0
Arg(z)=θ

Thus Arg is not continuous at any point of (−∞, 0]. The function Arg is one of
many important discontinuous complex functions.
Exercise 5.12. Suppose that f (z) = z̄/z and g(z) = z 2 /z̄. Does limz→0 f (z) or
limz→0 g(z) exist? If so, find the limit; otherwise, explain why it does not exist.

Answer. On the one hand, if z = reiθ , then f (z) = e−2iθ . As we approach 0

along a ray from the origin, θ is fixed and r → 0+, so the limit along the ray is
e−2iθ ; different rays give different limits. By the uniqueness of the limit, limz→0 f (z)
does not exist.
On the other hand, |g(z) − 0| = |z|, and as z → 0, |z| → 0. Hence g(z) → 0 as
z → 0. 4

3. Stereographic projection and the Riemann sphere

To explain the role of ∞, we imagine a unit sphere S in xyz space, with center at
0, and we identity the xy plane with the complex plane. We may define a function
σ : C → S geometrically by joining a point p in the plane to the north pole n of the
sphere by a straight line. The line will cut the sphere at n and at one other point,
which we call σ(p). Then we may think of n as being σ(∞).
The function σ is called stereographic projection, and the sphere is called the
Riemann sphere. Then balls B(0, ε) correspond to spherical caps in the Riemann
sphere, centred at σ(0), while balls B(z0 , ε) correspond to spherical caps in the
Riemann sphere containing the point σ(z0 ), and the “punctured balls” B◦ (∞, ε)
correspond to “punctured” spherical caps in the Riemann sphere centred at σ(∞).
These spherical caps shrink down towards the points σ(0), to σ(z) and to σ(∞) as
ε tends to 0.

4. Continuity
Definition 5.13. Suppose that f is a complex function. We say that f is continuous
at a point z0 if f (z0 ) is defined and limz→z0 f (z) = f (z0 ). We say that f is continuous
in a set S if it is continuous at all points of S. We say that f is continuous if it is
continuous at all points of its domain.

The functions z 7→ z, z 7→ z, z 7→ |z|, z 7→ Re(z), and z 7→ Im(z) are all

continuous. The function Arg is continuous in the set {z ∈ C \ {0} : Arg(z) 6= π}.
Properties of limits lead to similar properties of continuous functions.
Theorem 5.14. Suppose that c ∈ C, and that f : S → C and g : S → C are
continuous complex functions in S ⊆ C. Then cf , f + g, |f |, f , Re f , Im f and f g
are continuous in S, as is f /g provided that g(z) 6= 0 for any z in S.
 5. EXAMPLES OF CONTINUOUS FUNCTIONS 35

Theorem 5.15. Suppose that f : S → C and g : T → C are continuous complex

functions in S ⊆ C and T ⊆ C. Then f ◦ g is continuous where it is defined, that
is, in {z ∈ T : g(z) ∈ S}.
By using the theorems above, it follows that functions that are composed of the
standard functions (except Arg), such as
Re(z 2 ) + i Im(z 3 )
z 7→ ,
|z| + 1 + z
are also continuous where they are defined (with this example, the tricky bit is
finding the domain of definition; the natural domain is actually C).
Generally speaking, any function that can be written down using the standard
functions, and without choices in the definition, is continuous in its domain of defi-
nition, except when Arg is involved. Where there are choices in the definition, the
difficulties usually lie where the different definitions match up.
Continuity is useful for two reasons. First, when functions are continuous, we
do not have to worry about limits much. Next, continuous functions have some
important properties.
Theorem 5.16. Suppose that the set S ⊆ C is compact (i.e., closed and bounded)
and that f is a continuous complex function defined on S. Then there exists a point
z0 in S such that
|f (z0 )| = max{|f (z)| : z ∈ S}.
One says that the modulus of a continuous function attains its maximum in a
compact set. As a consequence, if f is a continuous complex function defined in a
compact set S ⊆ C, then there is a number R such that
|f (z)| ≤ R ∀z ∈ S.
Thus f is bounded in S.
Last but not least, continuous functions in compact sets are uniformly continu-
ous. We will not explain this now.

5. Examples of continuous functions

Exercise 5.17. Show from first principles that z 7→ |z| is a continuous function
in C.
Answer. Take z0 in C. We claim that limz→z0 |z| = |z0 |. Given ε ∈ R+ , take
δ = ε. If 0 < |z − z0 | < δ, then |z| − |z0 | ≤ |z − z0 | < ε, proving our claim. Since
limz→z0 |z| = |z0 | for any z0 ∈ C, z 7→ |z| is continuous in C. 4

Exercise 5.18. Show that Arg(z) is continuous in C \ (−∞, 0].

Answer. Fix z0 in C \ (−∞, 0], and suppose that ε ∈ R+ . Suppose that δ =
|z0 | sin min{ε, π − | Arg(z0 )|}. Then trigonometry shows that if 0 < |z − z0 | < δ,
then |Arg(z) − Arg(z0 )| < ε, as required. 4

Definition 5.19. The function Log : C \ {0} → C is defined by

Log(z) = ln |z| + i Arg(z).
36 5. LIMITS AND CONTINUITY

Exercise 5.20. Show that Log(z) is continuous in C \ (−∞, 0], and is not
continuous at any point on (−∞, 0].
Answer. We know that z 7→ |z| is continuous, and from calculus, ln is also
continuous. Hence z 7→ ln |z| is continuous.
We claim that Log is continuous where Arg is continuous. Indeed, if Arg is
continuous, then so is Log, as it is the sum of continuous functions; conversely, if
Log is continuous, then so is z 7→ Log(z) − ln |z|, for the same reason, that is, Arg
is continuous. 4

(1) Standard real-valued functions of a real variable may be considered as com-

plex functions; the definitions from complex analysis then coincide with
those from calculus.
(2) Unless problems use expressions like “from first principles”, then you may
use standard limits and the algebra of limits to solve them.
 LECTURE 6

Complex Differentiability

In this lecture, we investigate the differentiability of a function of a complex

variable, defined much as for functions of a real variable. Many calculations are
similar to the real-variable case; however, some functions that we might expect to
be differentiable are not.

1. Definition
Definition 6.1. Suppose that S ⊆ C and that f : S → C is a complex function.
Then we say that f is differentiable at the point z0 in S if
f (z) − f (z0 )
lim (6.1)
z→z0 z − z0
exists and is finite. If it does, it is called the derivative of f at z0 , and is written
df (z0 )
or f ′ (z0 ).
dz
We say that f is differentiable in a set S if it is differentiable at all points of S,
and that f is differentiable if it is differentiable at all points of its domain.

Remark 6.2. The limit (6.1) may also be written as

f (z0 + h) − f (z0 )
lim , (6.2)
h→0 h
and the definition may be given with this limit instead.

2. Examples
Exercise 6.3. Suppose that f1 (z) = z 2 + iz + 2. Is f1 differentiable at z0 in C?
If so, find f1′ (z0 )?
Answer. We use formula (6.1): if z 6= z0 , then
f1 (z) − f1 (z0 ) z 2 + iz + 2 − z02 − iz0 − 2
=
z − z0 z − z0
(z − z0 )(z + z0 ) + i(z − z0 )
=
z − z0
= z + z0 + i.
From the algebra of limits and standard limits,
f1 (z) − f1 (z0 )
lim = lim z + z0 + i = 2z0 + i.
z→z0 z − z0 z→z0
′
Hence f1 is differentiable at z0 and f1 (z0 ) = 2z0 + i. This holds for any point z0 , so
f1 is differentiable in C, and f1′ (z) = 2z + i. 4
37
38 6. COMPLEX DIFFERENTIABILITY

The computation above is almost identical to that to find the derivative of the
real function x2 + x + 2. Indeed, many formulae from the real case also hold in the
complex case when x is replaced by z. So do many theorems.

3. More examples of differentiation of complex functions

The following examples show that there is a twist to the story.
Exercise 6.4. Suppose that f2 (z) = z. Is f2 differentiable at z0 in C? If so,
find f2′ (z0 )?
Answer. If w 6= 0, then
f2 (z0 + w) − f2 (z0 ) z0 + w − z0 w
= = .
w w w
Now, if θ ∈ (−π, π], then
lim w/w = e−2iθ ,
w→0
Arg(w)=θ

which depends on θ. By the uniqueness of limits, limw→0 w/w does not exist, and
f2 (z0 + w) − f2 (z0 )
lim
w→0 w
fails to exist. As z0 is arbitrary, f2 is not differentiable anywhere in C. 4

Exercise 6.5. Suppose that f3 (z) = |z|2 . Is f3 differentiable at z0 in C? If so,

find f3′ (z0 )?
Answer. Fix z0 , and recall that |z|2 = zz. If w 6= 0, then
f3 (z0 + w) − f3 (z0 ) (z0 + w)(z0 + w) − z0 z0
=
w w
z0 w + wz0 + ww
=
w
w
= z0 + w + z0 .
w
If z0 = 0, then the right hand side is equal to w, and limw→0 w = 0. Hence f3
is differentiable at 0 with derivative 0. If z0 6= 0, then limw→0 z0 + w = z0 , but
limw→0 w/w does not exist. Hence f3 is not differentiable at z0 if z0 6= 0. 4
These examples show that a function may be differentiable everywhere, or no-
where, or at some points and not others. The nondifferentiable examples involved
the complex conjugate, explicitly or implicitly.

4. The Cauchy–Riemann equations

We will now investigate differentiability using theoretical tools. If limw→0 q(w)
exists, then
lim q(w) = lim q(w) = lim q(w),
w→0 w→0 w→0
w∈R w∈iR
in the sense that the first two limits also exist, and are equal to the third. This
allows us to relate the complex derivative to partial derivatives.
 4. THE CAUCHY–RIEMANN EQUATIONS 39

Theorem 6.6. Suppose that Ω is an open subset of C, that f is a complex

function defined in Ω, that f (x + iy) = u(x, y) + iv(x, y), where u and v are real-
valued functions of two real variables, and that f is differentiable at z0 ∈ Ω. Then
the partial derivatives
∂u ∂u ∂v ∂v
(x0 , y0 ), (x0 , y0 ), (x0 , y0 ) and (x0 , y0 )
∂x ∂y ∂x ∂y
all exist, and
∂u ∂v ∂u ∂v
(x0 , y0 ) = (x0 , y0 ) and (x0 , y0 ) = − (x0 , y0 ) . (6.3)
∂x ∂y ∂y ∂x
Further,
∂u ∂v  ∂u ∂v 
′
f (z0 ) = (x0 , y0 ) + i (x0 , y0 ) = −i (x0 , y0 ) + i (x0 , y0 ) . (6.4)
∂x ∂x ∂y ∂y

Remark 6.7. The pair of equations (6.3), which relate the partial derivatives of
u and v, are known as the Cauchy–Riemann equations.
Proof. If f ′ (z0 ) exists, then
f (z0 + w) − f (z0 )
f ′ (z0 ) = lim
w→0
w∈R
w
u(x0 + w, y0 ) + iv(x0 + w, y0 ) − u(x0 , y0 ) − iv(x0 , y0 )
= lim
w→0
w∈R
w
u(x0 + w, y0 ) − u(x0 , y0 ) v(x0 + w, y0 ) − v(x0 , y0 )
= lim + i lim
w→0
w∈R
w w→0
w∈R
w
∂u ∂v
= (x0 , y0 ) + i (x0 , y0 ),
∂x ∂x
because a limit exists if and only if its real and imaginary parts do. Thus
∂u ∂v
(x0 , y0 ) = Re(f ′ (z0 )) and (x0 , y0 ) = Im(f ′ (z0 )).
∂x ∂x
′ ′ ′
Since f (z0 ) = Re(f (z0 )) + i Im(f (z0 )), part of (6.4) follows.
Similarly, if f ′ (z0 ) exists, then
f (z0 + w) − f (z0 )
f ′ (z0 ) = lim
w→0
w∈iR
w
u(x0 , y0 + h) + iv(x0 , y0 + h) − u(x0 , y0 ) − iv(x0 , y0 )
= lim
h→0 ih
h∈R
v(x0 , y0 + h) − v(x0 , y0 ) u(x0 , y0 + h) − u(x0 , y0 )
= lim − i lim
h→0 h h→0 h
h∈R h∈R
∂v ∂u  ∂u ∂v 
= (x0 , y0 ) − i (x0 , y0 ) = −i (x0 , y0 ) + i (x0 , y0 ) .
∂y ∂y ∂y ∂y
Thus
∂v ∂u
(x0 , y0 ) = Re(f ′ (z0 )) and (x0 , y0 ) = − Im(f ′ (z0 )).
∂y ∂y
40 6. COMPLEX DIFFERENTIABILITY

The Cauchy–Riemann equations follow by equating the two expressions for the real
part of f ′ (z0 ) and the two expressions for the imaginary part of f ′ (z0 ), and the
remaining part of (6.4) also follows.. □
One consequence of the previous theorem is that if f is differentiable at every
point of an open set Ω in C, then the Cauchy–Riemann equations hold at every
point of Ω. Later on, we will see that in addition the four partial derivatives ∂u/∂x,
∂v/∂x, ∂u/∂y and ∂v/∂y are all continuous. For open sets Ω, the converse is true.
Theorem 6.8. If the four partial derivatives ∂u/∂x, ∂v/∂x, ∂u/∂y and ∂v/∂y are
all continuous in an open set Ω, then f is complex differentiable at z0 ∈ Ω if and
only if the Cauchy–Riemann equations hold at z0 , and if so, then
∂u ∂v
f ′ (x0 + iy0 ) = (x0 , y0 ) + i (x0 , y0 ).
∂x ∂x
We will justify this result later. When the partial derivatives are continuous in
a set that is not open, the function might be differentiable, or it might not be.

5. Examples
We revisit our previous examples, and add another, using the Cauchy–Riemann
equations.
Examples 6.9. (1) Suppose that f1 (z) = z 2 + iz + 2. Then
u(x, y) = x2 − y 2 − y + 2 and v(x, y) = 2xy + x,
so
∂u ∂u ∂v ∂v
= 2x, = −2y − 1, = 2y + 1 and = 2x,
∂x ∂y ∂x ∂y
and hence the Cauchy–Riemann equations hold for all (x, y) in R2 . Since the partial
derivatives are continuous and C is open, f1 is differentiable in C, and
f1′ (z) = 2x + i(2y + 1) = 2z + i.
(2) Suppose that f2 (z) = z. Then
u(x, y) = x and v(x, y) = −y,
so
∂u ∂u ∂v ∂v
= 1, = 0, = 0 and = −1,
∂x ∂y ∂x ∂y
and hence the Cauchy–Riemann equations do not hold for any (x, y) in R2 .
Hence f2 is not differentiable at any point in C.
(3) Suppose that f3 (z) = |z|2 . Then
u(x, y) = x2 + y 2 and v(x, y) = 0,
so
∂u ∂u ∂v ∂v
= 2x, = 2y, = 0 and = 0,
∂x ∂y ∂x ∂y
and hence the Cauchy–Riemann equations hold if and only if x = y = 0. The partial
derivatives are continuous in C, which is open, and hence f3 is differentiable at 0.
Finally, f is not differentiable at any other point than 0, since the Cauchy–Riemann
equations do not hold at any other point.
 6. PROPERTIES OF THE DERIVATIVE 41

(4) Suppose that f4 (z) = ez . Then

u(x, y) = ex cos y and v(x, y) = ex sin y,
so
∂u ∂u ∂v ∂v
= ex cos y, = −ex sin y, = ex sin y and = ex cos y,
∂x ∂y ∂x ∂y
and hence the Cauchy–Riemann equations hold for all (x, y) in R2 . Since the partial
derivatives are continuous and C is open, f4 is differentiable in C, and
∂u ∂u
f4′ (z) = (x, y) + (x, y) = ex (cos y + i sin y) = ez .
∂x ∂y
Remark 6.10. Doing (4) using limits would be rather messy!

6. Properties of the derivative

Theorem 6.11. Suppose that z0 ∈ C, that the complex functions f and g are
differentiable at z0 , and that c ∈ C. Then the functions cf , f + g and f g are
differentiable at z0 , and
(cf )′ (z0 ) = c f ′ (z0 ),
(f + g)′ (z0 ) = f ′ (z0 ) + g ′ (z0 ),
(f g)′ (z0 ) = f ′ (z0 ) g(z0 ) + f (z0 ) g ′ (z0 ).
Further, if g(z0 ) 6= 0, then the function f /g is differentiable at z0 , and
f ′ (z0 ) g(z0 ) − f (z0 ) g ′ (z0 )
(f /g)′ (z0 ) = .
g(z0 )2
Theorem 6.12. Suppose that z0 ∈ C, that the complex function f is differentiable
at g(z0 ), and that the complex function g is differentiable at z0 . Then the function
f ◦ g is differentiable at z0 , and
(f ◦ g)′ (z0 ) = f ′ (g(z0 )) g ′ (z0 ).
Theorem 6.13. Suppose that f is a complex function and that z0 ∈ Domain(f ).
If f is differentiable at z0 , then f is continuous at z0 .
Proof. The proof of these results are very similar to those of the corresponding
results for real functions, and we omit them. □
Theorem 6.14 (l’Hôpital’s rule). Suppose that z0 ∈ C ∪ {∞} and that the com-
plex functions f and g are differentiable at z0 . If limz→z0 f (z)/g(z) is indeterminate,
that is, of the form 0/0 or ∞/∞, and if limz→z0 f ′ (z)/g ′ (z) exists, then
f (z) f ′ (z)
lim = lim ′ . (6.5)
z→z0 g(z) z→z0 g (z)
We do not prove l’Hôpital’s rule at this time. It follows from results that we
prove later about Taylor series and Laurent series.
(1) If you write the matrix of partial derivatives of u and v with respect to x
and y, then the terms that would multiply together in the determinant are
equal, up to sign; the product that has the negative sign in the determinant
marks the partial derivatives that are equal to each others negative.
42 6. COMPLEX DIFFERENTIABILITY

(2) We use d, ∂ and δ in this course and it is important to write these clearly
and correctly. Marks will be deducted for incorrectly written letters!
(3) It is common to use ux instead of ∂∂xu ; this is fine.
 LECTURE 7

Connections with calculus in the plane† (Not examinable)

In this lecture, we link limits, continuity and differentiability of complex functions

with their analogues for functions of two real variables.

1. Limits and continuity

The definitions and properties of limits and continuity for functions of two real
variables are essentially the same as limits for functions of one complex variable. To
save space, we represent vectors in R2 as row vectors rather than column vectors.
Definition 7.1. Suppose that u is a real-valued function of two real variables,
and that (x0 , y0 ) ∈ Domain(u) . We say that u(x, y) tends to ℓ as (x, y) tends to
(x0 , y0 ), or that ℓ is the limit of u(x, y) as (x, y) tends to (x0 , y0 ), and we write
u(x, y) → ℓ as (x, y) → (x0 , y0 ), or
lim u(x, y) = ℓ,
(x,y)→(x0 ,y0 )

if, for every ε ∈ R+ , there exists δ ∈ R+ such that |u(x, y) − ℓ| < ε provided that
(x, y) ∈ Domain(f ) and 0 < |(x, y) − (x0 , y0 )| < δ. The same definition applies to a
vector-valued function F , provided that we interpret |F (x, y) − ℓ| as a vector length.
Given a complex function f , we associate real-valued functions u and v and an
R2 -valued function F of two real variables by the formulae
f (x + iy) = u(x, y) + iv(x, y)
F (x, y) = (u(x, y), v(x, y)).
The definitions of limit imply the following link between the real and complex
functions.
Connection 1. Let functions f , u, v and F be related as above. Then the
following are equivalent:
(1) f (z) → ℓ as z → z0 ;
(2) u(x, y) → Re ℓ and v(x, y) → Im ℓ as (x, y) → (x0 , y0 );
(3) F (x, y) → (Re ℓ, Im ℓ) as (x, y) → (x0 , y0 ).
This means that to most theorems about limits of real functions such as u and v,
there is a corresponding theorem about complex-valued functions f , and vice versa.
For example, a theorem about vector-valued functions states that a vector-valued
function tends to a limit ℓ if and only if each component of the function tends to the
corresponding component of ℓ. This is the analogue of the theorem that a complex-
valued function tends to a complex limit if and only if the real and imaginary parts
of the function tend to the real and imaginary parts of the limit.
Continuity for functions of two real variables is defined much as for functions of
a complex variable.
43
44 7. CONNECTIONS WITH CALCULUS IN THE PLANE† (NOT EXAMINABLE)

Definition 7.2. Suppose that u is a real-valued function of two real vari-

ables, and that (x0 , y0 ) ∈ Domain(u) . We say that u is continuous at (x0 , y0 )
if lim(x,y)→(x0 ,y0 ) u(x, y) and u(x0 , y0 ) both exist and are equal. We say that u is
continuous if it is continuous at all points of Domain(u).
The same definition applies to vector-valued functions.
Connection 2. Let functions f , u, v and F be related as in (1). Then the
following are equivalent:
(1) f is continuous;
(2) u and v are continuous;
(3) F is continuous.

2. Differentiability
For functions of two real variables, we cannot define the derivative as for functions
of a real variable, because this would involve dividing by a vector. But we note that
an equivalent definition of the derivative of a function of one (real or complex)
variable is that f is differentiable at z0 and f ′ (z0 ) = D if
|f (z0 + h) − f (z0 ) − Dh|
lim = 0.
h→0 |h|
To see this, note that
|f (z0 + h) − f (z0 ) − Dh| f (z0 + h) − f (z0 ) − Dh
lim = lim
h→0 |h| h→0 h
f (z0 + h) − f (z0 )
= lim −D .
h→0 h
We define the derivative by extending this modified definition.
Definition 7.3. A real- or vector-valued function u of two real variables is
differentiable at (x0 , y0 ), and its derivative is the linear transformation D, if
|u((x0 , y0 ) + (h, k)) − u(x0 , y0 ) − D(h, k)|
lim = 0,
(h,k)→(0,0) |(h, k)|
or equivalently, if
|u(x, y) − u(x0 , y0 ) − D(x − x0 , y − y0 )|
lim = 0.
(x,y)→(x0 ,y0 ) |(x − x0 , y − y0 )|
The linear transformation D sends R2 to R if u is real-valued, and sends R2 to R2
if u is R2 -valued.

It turns out that a vector-valued function is differentiable if and only if each of

its components is differentiable, so that much of the theory may be developed for
real-valued functions and then extended to vector-valued functions component by
component.
One of the first steps in the development of calculus in several real variables is the
identification of the linear transformation D. If u is real-valued and differentiable,
 2. DIFFERENTIABILITY 45

then the partial derivatives exist and

∂u ∂u
D(h, k) = (x0 , y0 )h + (x0 , y0 )k
∂x ∂y
 
∂u ∂u
= (x0 , y0 ), (x0 , y0 ) · (h, k)
∂x ∂y
  
∂u ∂u h
= (x0 , y0 ) (x0 , y0 )) ,
∂x ∂y k

where the second line is a dot product of vectors, and the third is a product of two
matrices. The plane z = u(x0 , y0 ) + D(x − x0 , y − y0 ) in R3 is the tangent plane to
the surface z = u(x, y) in R3 at the point (x0 , y0 , u(x0 , y0 )).
To avoid having to deal with limits all the time, most treatments of multivariable
calculus prove the next result as soon as possible. For ease of notation, we state it
for a real-valued function, but it also holds for vector-valued functions.

Theorem 7.4. Suppose that u is a real-valued function of two real variables, and
that the partial derivatives ∂u/∂x and ∂u/∂y exist and are continuous in an open
set Ω. Then u is differentiable in Ω.

Proof. Take (x0 , y0 ) ∈ Ω. To show that u is differentiable at (x0 , y0 ), we need

to make the quotient
|u(x, y) − u(x0 , y0 ) − D(x − x0 , y − y0 )|
|(x − x0 , y − y0 )|

small, by taking (x, y) close to (x0 , y0 ). To make this quantitative, we take ε ∈ R+ ,

and make the quotient less than ε. Recall that
∂u ∂u
D(x − x0 , y − y0 ) = (x0 , y0 )(x − x0 ) + (x0 , y0 )(y − y0 ).
∂x ∂y

Choose δ such that B((x0 , y0 ), δ) ⊂ Ω and

∂u ∂u ε ∂u ∂u ε
(x1 , y1 ) − (x0 , y0 ) < and (x1 , y1 ) − (x0 , y0 ) <
∂x ∂x 2 ∂y ∂y 2

whenever (x1 , y1 ) ∈ B((x0 , y0 ), δ). This is possible because Ω is open and because
both the partial derivatives are continuous at (x0 , y0 ). If (x, y) ∈ B((x0 , y0 ), δ),
then the line segments joining (x0 , y0 ) to (x, y0 ) and joining (x, y0 ) to (x, y) lie in
B((x0 , y0 , δ), by the geometry of balls in R2 .
The fundamental theorem of calculus and the mean value theorem for integrals
implies that there exist y1 between y and y0 and x1 between x and x0 such that
u(x, y) − u(x0 , y0 ) = u(x, y) − u(x, y0 ) + u(x, y0 ) − u(x0 , y0 )
Z y Z x
∂u ∂u
= (x, t) dt + (s, y0 ) ds
y0 ∂y x0 ∂x
∂u ∂u
= (x, y1 )(y − y0 ) + (x1 , y0 )(x − x0 ),
∂y ∂x
46 7. CONNECTIONS WITH CALCULUS IN THE PLANE† (NOT EXAMINABLE)

and this in turn implies that

|u(x, y) − u(x0 , y0 ) − D(x − x0 , y − y0 )|
∂u ∂u
= (x, y1 )(y − y0 ) − (x0 , y0 )(y − y0 )
∂y ∂y
∂u ∂u
+ (x1 , y0 )(x − x0 ) − (x0 , y0 )(x − x0 )
∂x ∂x
∂u ∂u ∂u ∂u
≤ (x, y1 ) − (x0 , y0 ) |y − y0 | + (x1 , y0 ) − (x0 , y0 ) |x − x0 |
∂y ∂y ∂x ∂x
ε ε
< |y − y0 | + |x − x0 | ,
2 2
whence
|u(x, y) − u(x0 , y0 ) − D(x − x0 , y − y0 )| ε
|y − y0 | + 2ε |x − x0 |
< 2 ≤ ε,
|(x − x0 , y − y0 )| |(x − x0 , y − y0 )|
as required. □
Here is another theorem about differentiation of functions of two real variables;
we omit the proof.
Theorem 7.5. Suppose that u is a twice continuously differentiable real-valued
function of two real variables. Then
∂ 2u ∂ 2u
= .
∂x ∂y ∂y ∂x
A consequence of this theorem is that the derivative of u, which is the vector-
valued function (∂u/∂x, ∂u/∂y), satisfies the condition that the derivative with
respect to y of the first component is equal to the derivative with respect to x
of the second component. In general, a vector-valued function (p, q) is said to be
closed or conservative if ∂p/∂y = ∂q/∂x; and we have just seen that a derivative
is conservative. An important question in multivariable calculus is whether every
conservative vector-valued function (p, q) is a derivative of a real-valued function,
called the potential; whether this is always true or not depends on whether the
domain of definition of the function (p, q) is simply connected or not. We will return
to this later.
Theorem 7.6. Suppose that Ω is a simply connected subset of R2 , that p and
q are differentiable functions from Ω to R, and that ∂p/∂y = ∂q/∂x. Then there
exists a function f : Ω → R such that ∂f /∂x = p and ∂f /∂y = q.
We omit the proof of this result, as we will prove an equivalent result for complex
functions later.
Example 7.7. Consider the vector-valued function on R2 \ {(0, 0)} given by
 
−y x
F (x, y) = , = (p(x, y), q(x, y)),
x2 + y 2 x2 + y 2
say. It is easy to check that
∂p y 2 − x2 ∂p y 2 − x2
(x, y) = 2 and (x, y) = 2 ,
∂y (x + y 2 )2 ∂y (x + y 2 )2
 3. DIFFERENTIABILITY IN R2 AND COMPLEX DIFFERENTIABILITY 47

and so ∂p/∂y = ∂q/∂x. However there is no function on R2 \{(0, 0)} whose gradient
is F . When we integrate, we get the argument function (plus a constant), and this
cannot be defined continuously on R2 \ {(0, 0)}.

Observe that R2 \ {(0, 0)} is not simply connected.

3. Differentiability in R2 and complex differentiability

Suppose that f is a complex function. As before, we associate to f the functions
u, v and F :
f (x + iy) = u(x, y) + iv(x, y)
F (x, y) = (u(x, y), v(x, y)).
Then f is differentiable at z0 with derivative f ′ (z0 ) if and only if
|f (z0 + h + ik) − f (z0 ) − f ′ (z0 )(h + ik)|
lim = 0.
h+ik→0 |h + ik|
And F is differentiable at (x0 , y0 ) with derivative F ′ (x0 , y0 ) if and only if
|F ((x0 , y0 ) + (h, k)) − F (x0 , y0 ) − F ′ (x0 , y0 )(h, k)|
lim = 0.
(h,k)→0 |(h, k)|
Now F ′ (x0 , y0 ) is a linear transformation that may be identified with multipli-
cation by the matrix
 
∂u ∂u
 (x0 , y0 ) (x0 , y0 )
 ∂x ∂y 
 .
 ∂v 
 ∂v 
(x0 , y0 ) (x0 , y0 )
∂x ∂y
Multiplication by f ′ (z0 ) may be identified with multiplication by the matrix
 
b −c
,
c b
where b = Re f ′ (z0 ) and c = Im f ′ (z0 ).
These two operations correspond provided that
∂u ∂v ∂u ∂v
(x0 , y0 ) = (x0 , y0 ) and (x0 , y0 ) = − (x0 , y0 ),
∂x ∂y ∂y ∂x
and then
∂u ∂u ∂u ∂u
f ′ (z0 ) = (x0 , y0 ) + i (x0 , y0 ) = (x0 , y0 ) − i (x0 , y0 ) + . . . .
∂x ∂x ∂x ∂y
These conditions are exactly the Cauchy–Riemann equations.
In conclusion, we have established the following link between complex functions
and R2 -valued functions of two real variables.
Connection 3. A complex function f is complex differentiable if and only if
the associated vector-valued function of two real variables F is differentiable and the
derivative of F corresponds to multiplication by a complex number.
48 7. CONNECTIONS WITH CALCULUS IN THE PLANE† (NOT EXAMINABLE)

A consequence of this connection and Theorem 7.4 is that a complex function

f is differentiable in an open set if the Cauchy–Riemann equations hold for the
associated functions u and v and the partial derivatives of u and v are continuous.
Here is another. As we have seen, multiplication by the nonzero complex number
c corresponds to scalar multiplication by the modulus |c|, which preserves angles,
and then rotating through Arg c, which also preserves angles.
It can be shown that each linear transformation of R2 that preserve angles is the
composition of a nonzero scalar multiplication and a rotation. Thus differentiable
complex functions whose derivatives do not vanish in some open set correspond to
differentiable real functions on R2 whose derivatives preserve angles at every point
in the same open set. Such functions will be called conformal.
Later we will see the noteworthy result that functions that are complex differ-
entiable in open sets in C are infinitely differentiable. This contrasts with what
happens in the theory of real functions, where a function can be differentiable k
times but not k + 1 times. Thus a small difference in definition can lead to very
large differences in behaviour.
 LECTURE 8

Properties of differentiable functions

We explore some of the consequences of the Cauchy–Riemann equations.

1. Examples
The Cauchy Riemann equations enable us to define new functions and show that
they are complex differentiable. For instance, recall the definition of the hyperbolic
functions:
ex + e−x ex − e−x
cosh(x) = and sinh(x) = ∀x ∈ R.
2 2
Recall also that the derivative of cosh is sinh and the derivative of sinh is cosh, and
that
cosh(x + y) = cosh(x) cosh(y) + sinh(x) sinh(y)
and
sinh(x + y) = sinh(x) cosh(y) + cosh(x) sinh(y).
We define two new functions of a complex variable as follows:
ch(x + iy) = cosh(x) cos(y) + i sinh(x) sin(y)
and
sh(x + iy) = sinh(x) cos(y) + i cosh(x) sin(y).
Exercise 8.1. Is ch differentiable? What is its derivative? What about sh?

Answer. Write ch(x + iy) = u(x, y) + iv(x, y). Then,

u(x, y) = cosh(x) cos(y) v(x, y) = sinh(x) sin(y)
∂u ∂u
(x, y) = sinh(x) cos(y) (x, y) = − cosh(x) sin(y)
∂x ∂y
∂v ∂v
(x, y) = cosh(x) sin(y) (x, y) = sinh(x) cos(y).
∂x ∂y
Hence the Cauchy–Riemann equations hold in the whole plane, which is open, so ch
is differentiable in the whole plane. Moreover,
∂u ∂v
ch′ (x+iy) = (x, y)+i (x, y) = sinh(x) cos(y)+i cosh(x) sin(y) = sh(x+iy). 4
∂x ∂x

Recall that we defined Log(z) = ln |z| + i Arg(z) for all z ∈ C \ {0}.

Exercise 8.2. Where is Log differentiable? What is its derivative?
49
50 8. PROPERTIES OF DIFFERENTIABLE FUNCTIONS

Answer. As usual, we may write Log(x + iy) = u(x, y) + iv(x, y), where
u(x, y) = ln(x2 + y 2 )1/2 and v(x, y) = Arg(x + iy). Note that Log cannot be differ-
entiable at 0 since it is not defined there, nor on the negative real axis (−∞, 0), as
it is not continuous there.
In C \ (−∞, 0], which is open, we may apply the Cauchy–Riemann equations.
Observe that
∂u(x, y) 1 1 x
= 2 2x = ,
∂x (x + y 2 )1/2 2(x2 + y 2 )1/2 x2 + y 2
∂u(x, y) 1 1 y
= 2 2y = .
∂y (x + y 2 )1/2 2(x2 + y 2 )1/2 x2 + y 2
In the right half-plane where x > 0, we may write Arg(x + iy) = tan−1 (y/x),
and so
∂v(x, y) 1 −y −y
= 2 2
= 2 ,
∂x 1 + (y/x) x x + y2
∂v(x, y) 1 1 x
= = .
∂y 1 + (y/x)2 x x2 + y 2
In the upper half-plane where y > 0, we may write Arg(x+iy) = − tan−1 (x/y)+π/2;
in the lower half plane where y < 0, we may write Arg(x + iy) = − tan(x/y) − π/2.
Two very similar calculations show that the same formulae hold for the partial
derivatives in these cases too.
Hence the Cauchy–Riemann equations hold in the open set C \ (−∞, 0], and so
Log is differentiable in this set.
The derivative of Log is given by
∂u(x, y) ∂v(x, y) x − iy z̄ 1
Log′ (x + iy) = +i = 2 2
= =
∂x ∂x x +y z z̄ z
in C \ (−∞, 0]. 4

2. Geometry of the Cauchy–Riemann equations (Not examinable)

The Cauchy–Riemann equations are important in physical applications because
they imply that the functions u and v are related in a very significant way.
Corollary 8.3. Suppose that f is a differentiable function in an open set Ω,
and that f (x + iy) = u(x, y) + iv(x, y) for all x + iy ∈ Ω. Then the gradients ∇u
and ∇v satisfy
k∇uk = k∇vk and ∇u · ∇v = 0,
that is, the two gradient vectors are of equal length and perpendicular to each other.
Consequently when the derivative is nonzero, the contour lines for u are perpendicular
to the contour lines for v.
Proof. The gradient vectors ∇u and ∇v are given by
 ∂u ∂u   ∂v ∂v 
∇u = , and ∇v = , .
∂x ∂y ∂x ∂y
The Cauchy–Riemann equations imply that the second vector may be obtained from
the first by rotating through 90 degrees, and the result follows. □
 3. THE CAUCHY–RIEMANN EQUATIONS AND COMPLEX DIFFERENTIABILITY 51

Im Im

0
Re Re

Figure 8.1. Sketches of level curves for two functions

When we sketched complex functions, we drew level curves for the functions u
and v. The sketches suggested that the level curves for u and for v are perpendicular
to each other; the corollary explains why this is so. It is also true that the images
of the lines x = c and y = d are perpendicular, at least when f ′ (z) 6= 0.
There are examples in physics where u is a potential (electrical or gravitational),
so the contour lines for u are lines of equipotential; the contour lines for v are then
the curves along which particles (with charge or mass) move under the influence of
the field.
Example 8.4. Suppose that f (z) = z 2 , and that f (x + iy) = u(x, y) + iv(x, y).
Then u(x, y) = x2 −y 2 and v(x, y) = 2xy. The contours for u, that is, the level curves
given by an equation of the form u(x, y) = c, for some constant c, are hyperbolae
asymptotic to the lines y = ±x. The contours for v, that is, the level curves given by
an equation of the form v(x, y) = d, for some constant d, are hyperbolae asymptotic
to the x and y axes. See the left hand sketch in Figure 8.1.
Example 8.5. The next example is related to ch, and we are going to study this
in detail later. It corresponds to the electric field between two “infinite” charged
plates, represented by the infinite intervals (−∞, −1] and [1, ∞): the equipotential
lines are hyperbolae, while the paths followed by charged particles are elliptical. See
the right hand sketch in Figure 8.1.

3. The Cauchy–Riemann equations and complex differentiability

Before we look at other implications of complex differentiability, we consider a
“topological” question: if f is defined in an open set Ω, and f is constant along all
horizontal and vertical line segments contained in Ω, must f be constant? It is easy
to see that if Ω is not connected, then f need not be constant. But what if Ω is
connected? Recall that a connected open subset of C is called a domain.
Proposition 8.6. Suppose that f is a function defined on a domain Ω in C,
and f is constant along all horizontal and vertical line segments contained in Ω.
Then f is constant in Ω.
52 8. PROPERTIES OF DIFFERENTIABLE FUNCTIONS

Proof. Two distinct points of Ω may always be connected by a polygonal path,

involving finitely many line segments, so it will suffice to know that f is constant
along line segments. Take a line segment ℓ, with endpoints z0 and z1 , and write
zt = (1 − t)z0 + tz1 , where t ∈ [0, 1]. Then zt moves along the line segment ℓ from
z0 to z1 as t varies between 0 and 1.
Suppose that z ∈ Ω and z lies on ℓ. Since Ω is open, there is a ball B(z, ε)
contained in Ω. We claim that f is constant on B(z, ε) ∩ ℓ. Indeed, by elementary
geometry, if z ′ ∈ B(z, ε) ∩ ℓ, then the line segments joining x + iy and x′ + iy and
joining x′ + iy and x′ + iy ′ both lie inside B(z, ε) and hence in Ω. By hypothesis, f
is constant along both these segments, and so f (z) = f (z ′ ), and our claim is proved.
To show that f is constant along the whole line segment ℓ, we argue by contra-
diction. If f is not constant on ℓ, then we define t∗ = inf{t ∈ [0, 1] : f (zt ) 6= f (z0 )}.
Then 0 ≤ t∗ ≤ 1. But f is constant on B(zt∗ , εt∗ ) ∩ ℓ for some positive εt∗ , so f is
constant near t∗ , contradicting the definition of t∗ , and showing that f is constant
on ℓ. □

Theorem 8.7. Suppose that f is differentiable in a domain Ω in C. Then

(a) if f ′ = 0 in Ω, then f is constant on Ω;
(b) if |f | is constant, then f is constant on Ω;
(c) if Re(f ) or Im(f ) is constant, then f is constant on Ω.

Proof. As usual, write f (x + iy) = u(x, y) + iv(x, y).

First, suppose that f ′ = 0. Then ∂u/∂x, ∂u/∂y, ∂v/∂x, and ∂v/∂y are all 0.
By Proposition 8.6, f is constant.
Next, suppose that |f | is a constant, C say, in Ω. If C = 0, then f = 0, a
constant, so without loss of generality we may suppose that C 6= 0. Then
u2 + v 2 = C 2 > 0. (8.1)
Differentiating (8.1) with respect to x and with respect to y gives
∂u ∂v
2u + 2v =0 (8.2)
∂x ∂x
∂u ∂v
2u + 2v = 0. (8.3)
∂y ∂y
Using the Cauchy–Riemann equations with (8.3), we get
∂u ∂v
2v − 2u = 0. (8.4)
∂x ∂x
Eliminating ∂v/∂x from (8.2) and (8.4) shows that 2(u2 + v 2 ) ∂u/∂x = 0. From
(8.1), ∂u/∂x = 0. Similarly, ∂v/∂x = 0, so f ′ = 0, and f is constant.
Finally, suppose that Re(f ) is constant. Then ∂u/∂x = ∂u/∂y = 0, so from the
Cauchy–Riemann equations, ∂v/∂x = ∂v/∂y = 0, whence f ′ = 0 and f is constant.
The argument if Im(f ) is constant is similar. □

In fact, it is possible to show that if f is complex differentiable in a connected

open set Ω and nonconstant, then the range of f is open, which means that f cannot
satisfy any equations that restrict its range to lie in a one-dimensional set, such as
a curve.
 4. POLAR COORDINATES 53

4. Polar coordinates
The Cauchy–Riemann equations are a very powerful tool. Consequently it is
worth stating them in polar coordinates as well.
Theorem 8.8. Suppose that the complex function f is differentiable at a point
z0 in C \ {0}, and that z0 = r0 eiθ0 . Then
∂u ∂v ∂v ∂u
(r0 , θ0 ) = −r0 (r0 , θ0 ) and (r0 , θ0 ) = r0 (r0 , θ0 ).
∂θ ∂r ∂θ ∂r
Further,
   
′ −iθ0 ∂u ∂v −ie−iθ0 ∂u ∂v
f (z0 ) = e (r0 , θ0 ) + i (r0 , θ0 ) = (r0 , θ0 ) + i (r0 , θ0 ) .
∂r ∂r r ∂θ ∂θ
Proof. Write z in polar coordinates as reiθ . Much as argued to prove the
Cauchy–Riemann equations, since f is differentiable at z0 ,
∂f f ((r0 + s)eiθ0 ) − f (r0 eiθ0 )
(r0 eiθ0 ) = lim
∂r s→0 s
f (r0 e + seiθ0 ) − f (r0 eiθ0 )
iθ0
= eiθ0 lim
s→0 seiθ0
f (z0 + seiθ0 ) − f (z0 )
= eiθ0 lim
s→0 seiθ0
iθ0 ′
= e f (z0 ),
and
∂f f (r0 ei(θ0 +φ) ) − f (r0 eiθ0 )
(r0 eiθ0 ) = lim
∂θ φ→0 φ
f (r0 ei(θ0 +φ) ) − f (r0 eiθ0 ) r0 ei(θ0 +φ) − r0 eiθ0
= lim
φ→0 r0 ei(θ0 +φ) − r0 eiθ0 φ
f (r0 e i(θ0 +φ)
) − f (r0 e )
iθ0
r0 e i(θ0 +φ)
− r0 eiθ0
= lim lim
φ→0 r0 ei(θ0 +φ) − r0 eiθ0 φ→0 φ
iθ0 ′
= ir0 e f (z0 ),
where we used trigonometric limits to evaluate the last limit.
We now have two formulae, both involving eiθ0 f ′ (z0 ); we equate these to get
∂f ∂f
(r0 eiθ0 ) = ir0 (r0 eiθ0 ).
∂θ ∂r
Now we write f in u + iv form, and see that
 
∂u ∂v ∂u ∂v
(r0 , θ0 ) + i (r0 , θ0 ) = ir0 (r0 , θ0 ) + i (r0 , θ0 ) .
∂θ ∂θ ∂r ∂r
Equating the real and imaginary parts gives the Cauchy–Riemann equations.
The formula for f ′ (z0 ) is found similarly. □
Remark 8.9. It is easiest to remember this version of the Cauchy–Riemann
equations in the form
∂f (reiθ ) ∂f (reiθ )
= ir .
∂θ ∂r
Corollary 8.10. The function Log is differentiable in C \ (−∞, 0].
54 8. PROPERTIES OF DIFFERENTIABLE FUNCTIONS

Proof. Write Log(reiθ ) in the form u(r, θ) + iv(r, θ). Then u(r, θ) = ln(r) and
v(r, θ) = θ; v is discontinuous when θ = π. Hence
∂u 1 ∂u
(r0 , θ0 ) = (r0 , θ0 ) = 0
∂r r0 ∂θ
∂v ∂v
(r0 , θ0 ) = 0 (r0 , θ0 ) = 1,
∂r ∂θ
unless θ = π. Hence the Cauchy–Riemann equations in polar form are satisfied
in C \ (−∞, 0]. Further, the partial derivatives are continuous in the open set
C \ (−∞, 0], so Log is differentiable in this set. □

5. Inverse functions
Suppose that Ω and Υ are open subsets of C, and that f is one-to-one from Ω
onto Υ. Then f has an inverse function, usually written f −1 , from Υ to Ω: we define
f −1 (w) = z if f (z) = w.
Theorem 8.11. Suppose that Ω and Υ are open subsets of C, that f : Ω → Υ is
one-to-one, and that f (z0 ) = w0 . If f is differentiable at z0 and f −1 is differentiable
at w0 , then (f −1 )′ (w0 ) = 1/f ′ (z0 ).
Proof. By definition, z = f −1 (f (z)); the chain rule implies that
df −1
1= (w0 ) f ′ (z0 ). □
dw
Later, we will investigate whether the inverse function is differentiable.

6. Definition
The examples above show that functions that are complex differentiable in an
open set have special properties; we are going to study them in much greater detail.
This justifies giving them a name.
Definition 8.12. Suppose that Ω is an open subset of C and f : Ω → C is a
function. If f is differentiable in Ω, that is, if it is differentiable at every point of Ω,
then we say that f is holomorphic or complex analytic or analytic or regular in Ω,
and we write f ∈ H(Ω).
If Ω = C and f is differentiable in Ω, then we say that f is entire.
 LECTURE 9

Harmonic functions

In this lecture, we introduce harmonic functions, and we discuss their signifi-

cance. We see how to find them using holomorphic functions, and conversely, how
to find holomorphic functions using harmonic functions. One of the reasons why
complex variable theory is an important area of mathematics is that it enables us
to find harmonic functions.

1. Harmonic functions
We begin with a definition.
Definition 9.1. Suppose that u : Ω → R is a function, where Ω is an open
subset of R2 , and that u is twice continuously differentiable, that is, all the partial
derivatives ∂u/∂x, ∂u/∂y, ∂ 2 u/∂x2 , ∂ 2 u/∂x ∂y, ∂ 2 u/∂y ∂x and ∂ 2 u/∂y 2 exist and
are continuous. Then we say that u is harmonic in Ω if it satisfies Laplace’s equation:
∂ 2u ∂ 2u
+ = 0.
∂x2 ∂y 2
Remark 9.2. In three dimensions, where the three real variables are labelled
x, y and z, we say that a real-valued twice continuously differentiable function u is
harmonic if
∂ 2u ∂ 2u ∂ 2u
+ + = 0.
∂x2 ∂y 2 ∂z 2
If a function u on R3 is independent of the variable z, then we may consider it as
a function of x and y only, and then it is harmonic in the two-dimensional sense
above.
How do we find harmonic functions? The next theorem tells us that it is enough
to find holomorphic functions.
Theorem 9.3. Suppose that f ∈ H(Ω), where Ω is an open subset of C, that f
is twice continuously differentiable, and that
f (x + iy) = u(x, y) + iv(x, y)
for all x + iy in Ω, where u and v are real-valued. Then u and v are harmonic
functions.
Remark 9.4. Later we will see that f ∈ H(Ω) implies that f is actually in-
finitely differentiable, so the “twice continuously differentiable” hypothesis is not
really necessary.
Proof. Suppose that f ∈ H(Ω), where Ω is an open subset of C, and that
f (x + iy) = u(x, y) + iv(x, y)
55
56 9. HARMONIC FUNCTIONS

in Ω, where u and v are real-valued. Since f is twice continuously differentiable,

the partial derivatives ∂u/∂x, ∂u/∂y, ∂ 2 u/∂x2 , ∂ 2 u/∂x ∂y, ∂ 2 u/∂y ∂x and ∂ 2 u/∂y 2
and ∂v/∂x, ∂v/∂y, ∂ 2 v/∂x2 , ∂ 2 v/∂x ∂y, ∂ 2 v/∂y ∂x and ∂ 2 v/∂y 2 all exist and are
continuous.
From vector calculus, this implies that
∂ 2u ∂ 2u ∂ 2v ∂ 2v
= and = .
∂x ∂y ∂y ∂x ∂x ∂y ∂y ∂x
Now, from the Cauchy–Riemann equations,
∂ 2u ∂ 2u ∂ ∂u ∂ ∂u
2
+ 2 = +
∂x ∂y ∂x ∂x ∂y ∂y
∂ ∂v ∂ ∂v
= −
∂x ∂y ∂y ∂x
∂ 2v ∂ 2v
= − = 0,
∂x ∂y ∂y ∂x
so u is harmonic. Similarly, v is harmonic. □
Now we ask whether we get all harmonic functions in this way. Let us first
consider some examples.

2. Examples
Exercise 9.5. Suppose that u(x, y) = x3 − 3xy 2 . Show that u is harmonic in C,
and find a function v such that the function f , given by f (x+iy) = u(x, y)+iv(x, y),
is holomorphic in C.
Answer. The partial derivatives of u are given by:
∂u ∂u
= 3x2 − 3y 2 = − 6xy
∂x ∂y
∂ 2u ∂ 2u
= 6x = − 6x.
∂x2 ∂y 2
Hence
∂ 2u ∂ 2u
+ = 6x − 6x = 0,
∂x2 ∂y 2
and u is harmonic.
If v exists such that the function f given by
f (x + iy) = u(x, y) + iv(x, y)
is holomorphic, then the Cauchy–Riemann equations must hold, so we must have
∂v
= 3x2 − 3y 2 (9.1)
∂y
∂v
= 6xy. (9.2)
∂x
Fix y ∈ R. From (9.2), ∂v(x, y)/∂x = 6xy, and integrating with respect to x
shows that
v(x, y) = 3x2 y + C1 ,
 2. EXAMPLES 57

where C1 does not depend on x. However, C1 may depend on y. Thus the general
solution to equation (9.2), when we allow y to vary, is
v(x, y) = 3x2 y + c(y),
where c is an unknown function.
The function v must also satisfy (9.1), and this happens as long as
∂v
3x2 − 3y 2 = = 3x2 + c′ (y), (9.3)
∂y
i.e., c′ (y) = −3y 2 , so c(y) = −y 3 + C, where C is a constant, and
v(x, y) = 3x2 y − y 3 + C.
Set f (x + iy) = u(x, y) + iv(x, y). Then
f (z) = f (x + iy) = x3 − 3xy 2 + i(3x2 y − y 3 ) + iC = z 3 + iC,
which is holomorphic. 4
Note that, in the solution above, if we had just used the constant C1 rather than
the function c(y), then we would not have been able to satisfy (9.3).
Exercise 9.6. Suppose that u(x, y) = x2 − y 2 . Show that u is harmonic in C.
Find a function v such that the function f , given by f (x + iy) = u(x, y) + iv(x, y),
is holomorphic in C.
Answer. The partial derivatives of u are given by:
∂u ∂u
= 2x = − 2y
∂x ∂y
∂ 2u ∂ 2u
= 2 = − 2.
∂x2 ∂y 2
Hence
∂ 2u ∂ 2u
+ = 2 − 2 = 0,
∂x2 ∂y 2
and u is harmonic.
If v exists such that the function f given by
f (x + iy) = u(x, y) + iv(x, y)
is holomorphic, then the Cauchy–Riemann equations must hold, so we must have
∂v ∂u
= = 2x (9.4)
∂y ∂x
∂v ∂u
=− = 2y. (9.5)
∂x ∂y
The general solution to equation (9.4) is
v(x, y) = 2xy + c(x),
where c is an unknown function.
For (9.5) to hold as well, we require that
∂v
2y = = 2y + c′ (x)
∂x
58 9. HARMONIC FUNCTIONS

for all x. This tells us that c′ (x) = 0 for all x and c has to be a constant, C say.
Set f (x + iy) = u(x, y) + iv(x, y). Then

f (z) = f (x + iy) = x2 − y 2 + 2ixy + iC = z 2 + iC,

which is holomorphic. 4

These examples suggest that it might always be possible to find a function v

such that u + iv is holomorphic. In fact we have the following theorem.

Theorem 9.7. If Ω is a simply connected domain, and u : Ω → R is harmonic,

then there exists a harmonic function v : Ω → R such that f , given by

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

in Ω, is holomorphic. Any two such functions v differ by an additive constant.

Remark 9.8. The function v is called a harmonic conjugate of u. The function

f may often be determined using the fact that

f ′ (x + iy) = ux (x, y) + ivx (x, y) = ux (x, y) − iuy (x, y);

in fact, often when we write out ux (x, y) − iuy (x, y), we can express this in terms of
z and hence find f .

We give a partial proof of this theorem in Section 4.

To see why it might be necessary to require the simple connectedness of Ω in the
theorem, consider the following exercise.

Exercise 9.9. Suppose that u(x, y) = ln(x2 + y 2 )1/2 in R2 \ {(0, 0)}. Show that
u is harmonic. Can you find a function v such that the function f in C \ {0}, defined
by f (x + iy) = u(x, y) + iv(x, y), is holomorphic.

Exercise 9.10. Suppose that u(x, y) = x2 − y 2 . Show that u is harmonic in C.

Find a function v such that the function f , given by f (x + iy) = u(x, y) + iv(x, y),
is holomorphic in C.

Answer. The partial derivatives of u are given by:

∂u ∂u
= 2x and = − 2y.
∂x ∂y
Hence
∂u ∂u
−i = 2(x + iy) = 2z.
∂x ∂y
Clearly, if f (z) = z 2 , then f ′ (z) = 2z, so we just observe that Re f (x + iy) = u(x, y),
so u is the real part of a holomorphic function and hence is harmonic. Further,
f = u + iv, where v(x, y) = 2xy. 4

The trouble with this method is that it may be hard to express ux − iuy as a
function of z, or hard to find the integral of this.
 4. PROOF OF THE EXISTENCE THEOREM† (NOT EXAMINABLE) 59

u is harmonic here
u is known here

Figure 9.1. The Dirichlet problem: find u inside the body

3. Applications of harmonic functions† (Not examinable)

Harmonic functions are important because they appear in many physical ap-
plications. For instance, electrical and gravitational potentials are harmonic, and
harmonic functions appear in fluid flow too. Similarly, components of electrical and
gravitational fields in fixed directions are harmonic. Physicists often talk of infinite
plates, or infinite wires, so that the associated potentials and fields do not depend
on one of the variables, and may be modelled in two dimensions.
Here is a basic problem. We can measure a potential on the boundary of a body.
How do we determine the potential inside the body? Mathematically, the problem
is to find a harmonic function inside which, on the boundary, is equal to a given
function. This is called the Dirichlet problem.

4. Proof of the existence theorem† (Not examinable)

Proof. We give the proof for the case where Ω is a rectangle with sides parallel
to the axes. In general, we need to be more sophisticated.
It is enough to find v such that the Cauchy–Riemann equations hold, for if we
can do this, then u + iv will be holomorphic in Ω.. Choose (x0 , y0 ) in Ω. For any
differentiable function v in Ω,
Z y Z x
∂v ∂v
v(x, y) = (x0 , t) dt + (s, y) ds + v(x0 , y0 ).
y0 ∂y x0 ∂x

But if the Cauchy–Riemann equations hold, then

Z y Z x
∂u ∂u
v(x, y) = (x0 , t) dt − (s, y) ds + v(x0 , y0 ).
y0 ∂x x0 ∂y

So we simply define v by the formula

Z y Z x
∂u ∂u
v(x, y) = (x0 , t) dt − (s, y) ds + C.
y0 ∂x x0 ∂y

Now
Z y Z x 
∂v ∂ ∂u ∂u ∂u
(x, y) = (x0 , t) dt − (s, y) ds + C =− (x, y)
∂x ∂x y0 ∂x x0 ∂y ∂x
60 9. HARMONIC FUNCTIONS

as needed, and, since u is harmonic,

Z y Z x 
∂v ∂ ∂u ∂u
(x, y) = (x0 , t) dt − (s, y) ds + C
∂y ∂y y0 ∂x x0 ∂y
Z x 2
∂u ∂ u
= (x0 , y) − 2
(s, y) ds
∂x x0 ∂y
Z x 2
∂u ∂ u
= (x0 , y) + 2
(s, y) ds
∂x x0 ∂x
∂u ∂u s=x
= (x0 , y) + (s, y)
∂x ∂x s=x0
∂u ∂u ∂u
= (x0 , y) + (x, y) − (x0 , y)
∂x ∂x ∂x
∂u
= (x, y),
∂x
as required.
In summary, the function v, given by an integral, has partial derivatives so that
the Cauchy–Riemann equations hold in Ω, and hence f is holomorphic. □
In general, we may prove this theorem by appealing to Theorem 7.6.
(1) The main issue that is not explored is why it is possible to find a harmonic
conjugate in a nonrectangular simply connected region.
(2) Students are quite good at finding harmonic conjugates. However, there
may be confusion about “constants of integration” which are not constants.
Solutions of the form v(x, y) = F (x, y) + C and v(x, y) = G(x, y) + C (de-
rived by integrating two equations) and “by inspection” v(x, y) = H(x, y)
are less mathematically satisfactory, especially if the same C is used! How-
ever a solution of the form “Try v(x, y) = H(x, y). By differentiation, this
satisfies both equations, so is the solution up to an additive constant” is
more acceptable. It is important to lay down clear guidelines.
 LECTURE 10

Power series

In this lecture, we define and study complex power series. The proofs of many
theorems about complex power series are almost identical to the proof of the corre-
sponding theorem for real power series, and so we omit most proofs. Power series
are important for several reasons.
(a) There are formulae for manipulating them, so they may be used for calculations,
for instance, in MAPLE.
(b) Holomorphic functions may be expressed in power series, and, as we will see
later, vice versa.

1. Definition and convergence

A (complex) power series is an expression of the form
X
∞
an (z − z0 )n , (10.1)
n=0
where the centre z0 and the coefficients an are all fixed complex numbers, and the
variable z is complex. We take (z − z0 )0 to be 1 for all z, even when z = z0 .
The first problem is whether the sum (10.1) makes sense. If z = z0 , then the
sum converges trivially; for other values of z, the series may or may not converge.
P
Theorem 10.1. Every power series ∞ n=0 an (z −z0 ) has a radius of convergence
n

ρ, given by the formulae

 −1  −1
ρ = lim sup |an |1/n = lim sup |an |1/n .
n→∞ k→∞ n≥k

The radius of convergence ρ ∈ [0, +∞] has the following properties:

P
(a) ∞ n=0 an (z − z0 ) converges if |z − z0 | < ρ
n
P
(b) ∞ an (z − z0 )n does not converge if |z − z0 | > ρ
Pn=0
(c) ∞ n=0 an (z − z0 ) may converge for no, some or all z such that |z − z0 | = ρ.
n

If ρ = 0, then the series converges only when

P∞z = z0 , whilen if ρ = +∞, then the
series converges for all z ∈ C. We often write n=0 an (z − z0 ) converges in B(a, r)
to mean that the sum converges for all z in B(a, r). When a = z0 , this means that
r ≤ ρ.
There are several tests for convergence that carry over to complex power series
from the theory of real power series.
Lemma 10.2 (The ratio test). The radius of convergence is given by
|an |
ρ = lim ,
n→∞ |an+1 |

61
62 10. POWER SERIES

as long as the limit exists or is +∞.

Lemma 10.3 (The root test). The radius of convergence is given by
1
ρ = lim ,
n→∞ |an |1/n

as long as the limit exists or is +∞.

In order to use the root test when the coefficients involve factorials, the following
facts, collectively known as Stirling’s formula, may sometimes be useful:
ln(n!) n!
lim =1 and lim = 1.
n→∞ ln((2π)1/2 e−n nn+1/2 ) n→∞ (2π)1/2 e−n nn+1/2
There are variations on this that may also be useful.

2. Examples
X
∞
Exercise 10.4. Find the centre and radius of convergence of 3n−1 (z + 1)n .
n=0

Answer. The centre is −1. Take an = 3n−1 for all n ∈ N. Now |an |/|an+1 | =
1/3, so limn→∞ |an |/|an+1 | = 1/3, and ρ = 1/3 by the ratio test. 4
 −1
This is a geometric series whose sum is 3(1 − 3(z + 1)) .
X
∞
(z − 2)n
Exercise 10.5. Find the centre and radius of convergence of .
n=0
n!

Answer. The centre is 2. Take an = 1/n! for all n ∈ N . Then

|an | 1/n! (n + 1)!
= = =n+1→∞ as n → ∞,
|an+1 | 1/(n + 1)! n!
and ρ = +∞ by the ratio test. 4
This is an exponential series whose sum is ez−2 .
Exercise 10.6. Fix α ∈ C. Find the centre and radius of convergence of the
X
∞
series an (z − 1)n , where a0 = 1 and, for all n ∈ Z+ ,
n=0

α(α − 1) . . . (α − n + 1)
an = .
n!
Answer. The centre is 1. By the ratio test,
|an | |α(α − 1) . . . (α − n + 1)| (n + 1)!
=
|an+1 | |α(α − 1) . . . (α − n + 1)(α − n)| n!
n+1 n+1 1
= = →1
|α − n| n |α/n − 1|
as n → ∞. Thus ρ = 1, and the series converges in B(1, 1). 4
This is a binomial series, which converges to (1 + (z − 1))α , that is, z α .
 2. EXAMPLES 63

Exercise 10.7. Find the centre and radius of convergence of the series
X
∞
k(k + 1)(k 2 + 2)
(z + 3)k .
k=1
2k

Answer. The centre is −3. Take ak = k(k + 1)(k 2 + 2)/2k . Now

|ak | k(k + 1)(k 2 + 2)2k+1 k k2 + 2

= = 2
|ak+1 | (k + 1)(k + 2)((k + 1)2 + 2)2k k + 2 (k + 1)2 + 2
k k2 + 2
= 2→2
k + 2 k 2 + 2k + 3
as k → ∞, and the radius of convergence is 2, by the ratio test. 4

Exercise 10.8. Find the centre and radius of convergence of the series
X
∞
(−j)j
(z − 5)j .
j=1
j!

Answer. The centre is 5. Take aj = (−j)j /j!. Now

 −1
|aj | jj j + 1 j
= (j + 1) = → e−1 as j → ∞,
|aj+1 | (j + 1) j+1 j
and the radius of convergence of the series is 1/e. 4

Exercise 10.9. Find the centre and radius of convergence of the series
X zm
.
m∈N
2m
m even

Answer. The centre is 0. Take am = 0 when m is odd while am = 2−m when

m is even. We cannot use the ratio test, as am /am+1 is not defined when m is odd.
The root test also fails because |am |−1/m is equal to 2 when m is even and 0 when m
is odd, so limm→∞ |am |−1/m does not exist. The easiest way to deal with this series
is to take w = z 2 , and rewrite the series:
X z m X z 2n X wn
m
= 2n
= n
.
m∈N
2 n∈N
2 n∈N
4
m even

This is as a geometric series with ratio z 2 /4, whose sum is 1/(1 − z 2 /4). The series
converges when |z 2 /4| < 1, that is, |z| < 2, so the radius of convergence is 2. 4

Note that |am |−1/m is equal to 0 if m is odd and to 2 if m is even, and

lim sup |am |−1/m = 2.

m→∞
64 10. POWER SERIES

3. The algebra and calculus of power series

P
P∞Theorem 10.10. Suppose that the complex power series ∞n=0 an (z − z0 ) and
n

n=0 bn (z − z0 ) both converge in B(z0 , ρ), and that c ∈ C. Then the following
n

series also converge in B(z0 , ρ):

X∞ X
∞
(a) c an (z − z0 ) , and its sum is c
n
an (z − z0 )n ;
n=0 n=0

X
∞ X
∞ X
∞
(b) (an + bn )(z − z0 )n , and its sum is an (z − z0 )n + bn (z − z0 )n ;
n=0 n=0 n=0

X
∞ X
n
(c) cn (z − z0 ) , where c0 = a0 b0 , c1 = a0 b1 + a1 b0 , …, and cn =
n
aj bn−j ,
n=0 j=0
X
∞ X
∞
and its sum is an (z − z0 ) ×
n
bn (z − z0 )n .
n=0 n=0

Proof. We omit the proof. □

P∞
Theorem 10.11. Suppose that f (z) = n=0 an (z − z0 )n in B(z0 , ρ), and that
ρ > 0. Then f ′ (z) exists in B(z0 , ρ), and
X∞ X
∞
f ′ (z) = an n(z − z0 )n−1 = am+1 (m + 1)(z − z0 )m
n=0 m=0

in B(z0 , ρ).
Proof. We omit the proof. □
This theorem allows us to differentiate power series term by term.
P
Corollary 10.12. Suppose that f (z) = ∞ n=0 an (z − z0 ) in B(z0 , ρ). Then f
n

may be differentiated as many times as desired, and

X
∞
(k)
f (z) = n(n − 1) . . . (n − k + 1)an (z − z0 )n−k .
n=k
In particular,
f (k) (z0 ) = k! ak .
Further, the real-valued functions u and v, such that f (x + iy) = u(x, y) + iv(x, y),
may be differentiated as many times as desired, and all their partial derivatives are
continuous.
Proof. We omit the proof, but point out that induction may be used. □
P∞
Corollary 10.13. Suppose that g(z) = n=0 an (z − z0 )n in B(z0 , ρ), and that
ε > 0. If g(z0 + t) = 0 for all real t in (−ε, ε), then g(z) = 0 for all z in B(z0 , ρ).
Proof. First, we prove by induction that g (k) (z0 + t) = 0 for all real t in (−ε, ε)
and all natural numbers n. This is true when k = 0, by hypothesis. Suppose now
that g (k) (z0 + t) = 0 for all real t in (−ε, ε) for some natural number k, and write
f (z) = g (k) (z) for all z ∈ B(z0 , ρ), so that f (z0 + t) = 0 for all real t in (−ε, ε). By
Theorem 10.11, applied as many times as necessary, f is holomorphic in B(z0 , ρ).
 3. THE ALGEBRA AND CALCULUS OF POWER SERIES 65

As usual, we write z0 = x0 + iy0 , and f (x + iy) = u(x, y) + iv(x, y). Then

u(x0 + t, y0 ) = 0 and u(x0 + t, y0 ) = 0 for all t ∈ (−ε, ε). Hence ∂u/∂x(x0 + t, y0 ) = 0
and ∂v/∂x(x0 + t, y0 ) = 0 for all t ∈ (−ε, ε). Since
∂u ∂v
f ′ (z0 + it) = (x0 + t, y0 ) + i (x0 + t, y0 ),
∂x ∂x
′
we see that f (z0 + it) = 0 for all t ∈ (−ε, ε). This implies that g (k+1) (z0 + t) − 0 for
all t ∈ (−ε, ε), and the inductive step is established.
To conclude, we recall that the power series for g (k) is of the form
X∞
n(n − 1) . . . (n − k + 1)an (z − z0 )n−k ,
n=k
(k)
whence
P∞ g (z0 ) = k! ak , and so ak = 0. This is true for all k, and hence g(z) =
n=0 0(z − z0 ) = 0 for all z ∈ B(z0 , ρ). □
n

P∞
Corollary
P∞ 10.14. Suppose that f (z) = n=0 an (z − z0 )n and moreover that
g(z) = n=0 bn (z − z0 )n in B(z0 , ρ). If f (z0 + t) = g(z0 + t) for all t ∈ (−ε, ε), then
f (z) = g(z) for all z ∈ B(z0 , ρ).
Proof. We apply the previous corollary to f − g. □
There is a stronger version of Corollary 10.13 that says that if f is holomorphic
in a domain Ω, and f (zn ) = 0 for distinct points zn ∈ Ω such that zn → z0 ∈ Ω as
n → ∞, then f = 0. This leads to a stronger version of the last corollary.
These last corollaries will later lead us to the concept of analytic continuation:
if a function is defined in a domain Ω, then it is determined by its values in a small
set. In particular, if f is an entire function, then it is determined by its values on R.
This explains why, in finding an analytic function with certain properties, it suffices
to find it on R; this fact is useful in dealing with harmonic functions.
 LECTURE 11

Exponential, Hyperbolic and Trigonometric functions

1. The exponential function

Definition 11.1. We define the exponential series by the formula
X∞
zn
exp(z) = ∀z ∈ C.
n=0
n!

This is the only possible power series extension of the real exponential into the
whole P
complex plane. Indeed, we saw in the previous lecture that a complex power
series ∞ n
n=0 an z is determined by its values on any interval (−ε, ε).

Theorem 11.2. The exponential series has the following properties:

(1) exp(0) = 1;
(2) exp(z + w) = exp(z) exp(w) for all z, w ∈ C;
(3) exp(−z) = exp(z)−1 for all z ∈ C;
(4) exp(z) 6= 0 for all z ∈ C;
(5) exp′ (z) = exp(z) for all z ∈ C;
(6) if a function f : C → C satisfies f (0) = 1 and f ′ (z) = f (z) for all z ∈ C,
then f (z) = exp(z) for all z ∈ C;
(7) exp(x + iy) = ex (cos(y) + i sin(y)) for all x, y ∈ R.
Proof. Item (1) is obvious.
Item (2) follows from the binomial theorem and manipulation of sums:
X n  
(z + w)n X 1 X n j n−j X X z j wn−j X
∞ ∞ ∞ n ∞ X ∞
z j wm
= z w = = .
n=0
n! n=0
n! j=0 j n=0 j=0
j! (n − j)! j=0 m=0
j! m!
Item (3) and (4) follows from the first two: exp(z) exp(−z) = exp(z + (−z)) = 1,
and so exp(z) cannot be 0 and moreover exp(z)−1 = exp(−z).
Item (5) follows from the calculus of power series.
To prove (6), we consider the derivative of the quotient function f / exp:
 f ′ f ′ (z) exp(z) − f (z) exp′ (z) f (z) exp(z) − f (z) exp(z)
(z) = 2
= = 0.
exp exp (z) exp2 (z)
Thus f / exp is constant; when z = 0 its value is 1, so it is identically 1, and
f (z) = exp(z) for all z ∈ C.
Finally, we prove (7). From part (2), exp(z) = exp(x) exp(iy); using results for
real power series, we see that exp(x) = ex and exp(iy) = cos(y) + i sin(y). □
Corollary 11.3. The exponential exp maps C onto C \ {0}, and exp(z1 ) =
exp(z2 ) if and only if z1 − z2 ∈ 2πiZ.
67
68 11. EXPONENTIAL, HYPERBOLIC AND TRIGONOMETRIC FUNCTIONS

Proof. From the theorem, exp(z) 6= 0 for all z ∈ C. If w 6= 0, then take

z = ln(|w|) + i Arg(w); we may check that exp(z) = w. Hence exp maps C onto
C \ {0}. If exp(z1 ) = exp(z2 ), then exp(z1 − z2 ) = 1, and so Re(z1 − z2 ) = 0 and
Im(z1 − z2 ) ∈ 2πZ, by trigonometry. □
The fact that exp(z) = exp(z + 2πik) for all k ∈ Z is called the periodicity of
exp. Often we write ez rather than exp(z).

2. The hyperbolic functions

Definition 11.4. We define the hyperbolic cosine and sine by the formulae
ez + e−z X z 2n ez − e−z X z 2n+1
cosh(z) = = and sinh(z) = =
2 n∈N
(2n)! 2 n∈N
(2n + 1)!
for all z ∈ C.
It follows from the definitions that these are the only power series extensions of
the real functions cosh and sinh into the whole complex plane.
Theorem 11.5. The hyperbolic sine and cosine have the following properties:
(i) cosh(−z) = cosh(z) (ii) sinh(−z) = − sinh(z)
(iii) cosh′ (z) = sinh(z) (iv) sinh′ (z) = cosh(z)
(v) cosh(z + 2πik) = cosh(z) (vi) sinh(z + 2πik) = sinh(z)
for all z ∈ C and all k ∈ Z. Further, for all w, z ∈ C,
(vii) cosh(z + w) = cosh(z) cosh(w) + sinh(z) sinh(w)
(viii) sinh(z + w) = sinh(z) cosh(w) + cosh(z) sinh(w)
(ix) cosh2 (z) − sinh2 (z) = 1.
Finally, for all x, y ∈ R,
(x) cosh(x + iy) = cosh(x) cos(y) + i sinh(x) sin(y)
(xi) sinh(x + iy) = sinh(x) cos(y) + i cosh(x) sin(y).
Proof. Items (i) and (ii) follow from the definition. Items (iii) and (iv) follow
from the definition and the calculus of power series. Items (v) and (vi), often called
the “periodicity of cosh and sinh”, follow from the periodicity of exp.
To prove item (vii), observe that
4(cosh(z) cosh(w) + sinh(z) sinh(w))
= (ez + e−z )(ew + e−w ) + (ez − e−z )(ew − e−w )
= ez+w + ez−w + e−z+w + e−z−w + ez+w − ez−w − e−z+w + e−z−w
= 2(ez+w + e−z−w ) = 4 cosh(z + w),
and divide by 4. Items (viii) and (ix) may be proved similarly. Items (x) and (xi)
follow from the identifications sinh(iy) = i sin(y) and cosh(iy) = cos(y), which may
be seen using power series. □
Exercise 11.6. Expand cosh(z + iπ/2) and sinh(z + iπ/2) in terms of cosh(z)
and sinh(z). Hence find cosh(z + iπ).
 3. THE TRIGONOMETRIC FUNCTIONS 69

Answer. From the theorem,

cosh(z + iπ/2) = cosh(z) cosh(iπ/2) + sinh(z) sinh(iπ/2)
= cosh(z) cos(π/2) + i sinh(z) sin(π/2)
= i sinh(z).
Similarly,
sinh(z + iπ/2) = i cosh(z).
Hence
cosh(z + iπ) = cosh((z + iπ/2) + iπ/2) = i sinh(z + iπ/2) = i2 cosh(z) = − cosh(z),
as required. 4

3. The trigonometric functions

Definition 11.7. We define the cosine and sine by the formulae
eiz + e−iz X (−1)n z 2n
cos(z) = =
2 n∈N
(2n)!
eiz − e−iz X (−1)n z 2n+1
sin(z) = =
2i n∈N
(2n + 1)!
for all z ∈ C.
It follows from the definitions that these are the only power series extensions of
the real functions cos and sin into the whole complex plane.
Exercise 11.8. Show that the complex sine and cosine have the following prop-
erties:
(i) cos(−z) = cos(z) (ii) sin(−z) = − sin(z)
(iii) cos′ (z) = − sin(z) (iv) sin′ (z) = cos(z)
(v) cos(z + 2πk) = cos(z) (vi) sin(z + 2πk) = sin(z)
for all z ∈ C and all k ∈ Z. Show also that
(vii) cos(z + w) = cos(z) cos(w) − sin(z) sin(w)
(viii) sin(z + w) = sin(z) cos(w) + cos(z) sin(w)
(ix) cos2 (z) + sin2 (z) = 1
for all w, z ∈ C.
Exercise 11.9. Express the following as functions of z:
(a) sinh(iz), (b) cosh(iz), (c) sin(iz), (d) cos(iz).

Answer. We do (a), using power series:

iz (iz)3 (iz)5 (iz)7 z z3 z5 z7
sinh(iz) = + + + + · · · = i − i + i − i + · · · = i sin(z).
1! 3! 5! 7! 1! 3! 5! 1!
The others are similar. 4
70 11. EXPONENTIAL, HYPERBOLIC AND TRIGONOMETRIC FUNCTIONS

y
y

cosh x
x

Figure 11.1. The map cosh

4. Graph sketching
Example 11.10. Find the images of the lines Re(z) = c and Im(z) = d under
the function cosh, and sketch these for various values of c and d.

Answer. As usual, we write w = cosh(z) as well as w = u + iv and z = x + iy.

If Re(z) = c, then z = c + iy, for some y ∈ R, and
w = cosh(z) = cosh(c + iy) = cosh(c) cosh(iy) + sinh(c) sinh(iy)
= cosh(c) cos(y) + i sinh(c) sin(y),
so u = cosh(c) cos(y) while v = sinh(c) sin(y). We eliminate y from these two
formulae, and obtain
u2 v2
+ = cos2 (y) + sin2 (y) = 1.
cosh2 (c) sinh2 (c)
This curve is an ellipse, unless c = 0. The periodicity of cosh means that each time
y changes by 2π, the point w moves once around the ellipse. In the exceptional case
when c = 0, we see that w = cos(y), and so w varies on the real axis between −1
and 1.
If Im(z) = d, then z = x + id, for some x ∈ R, and
w = cosh(z) = cosh(x + id) = cosh(x) cosh(id) + sinh(x) sinh(id)
= cosh(x) cos(d) + i sinh(x) sin(d),
so u = cosh(x) cos(d) while v = sinh(x) sin(d). We eliminate x from these two
formulae, and obtain
u2 v2
2
− 2 = cosh2 (x) − sinh2 (x) = 1.
cos (d) sin (d)
This curve is a hyperbola, unless d = 0, ±π/2, ±π, . . . . The periodicity of cosh means
that horizontal lines at a distance 2kπ apart are mapped to the same hyperbola. In
the exceptional cases, if sin(d) = 0, then w = ± cosh(x), and so w varies on the real
 5. HARMONIC FUNCTIONS 71

axis between −∞ and −1, or between 1 and +∞; if cos(d) = 0, then w = ±i sinh(x),
and so w varies along the imaginary axis.
Some of these curves are sketched in Figure 11.1. In this figure, the blue lines
on the left are the lines Im(z) = d, where d = 0.5, 1.0, 1.5, 2.0, 2.5, and 3, and
the half-hyperbolae on the right are the images of these lines; as the value of d
increases from 0.5 to 3.0, the half-hyperbola moves to the left. The images of the
lines Im(z) = d where d = 0 or π would be the half-lines [1, +∞) and (−∞, −1]; it
seems evident, and it may be proved, that the images of the lines Im(z) = d where
0 < d < π fill the plane apart from the two half-lines mentioned above. The red line
segments on the left are the parts of the lines Re(z) = c, where c = 0.5, 1.0, 1.5 and
2.0, where the imaginary part varies from 0 t π. Note that we could complete the
red ellipses in the figure in two different ways: we could either include segments of
the lines Re(z) = c, where c is negative and the imaginary part varies from 0 to π,
or we could allow the imaginary part of the line segments already present to range
from −π to π. 4

Exercise 11.11. Find the images of the lines Re(z) = c and Im(z) = d under
the functions sinh, sin and cos, and sketch these for various values of c and d.

5. Harmonic functions
Exercise 11.12. Suppose that u(x, y) = cos(x) sinh(y) for all x, y ∈ R. Show
that u is harmonic and find its harmonic conjugate.

Answer. First,
∂ 2 u(x, y) ∂ 2 u(x, y) ∂ 2 cos(x) ∂ 2 sinh(y)
+ = sinh(y) + cos(x)
∂x2 ∂y 2 ∂x2 ∂y 2
= − cos(x) sinh(y) + cos(x) sinh(y)
= 0.
Thus u is harmonic.
The harmonic conjugate v of u must satisfy:
∂v(x, y) ∂u(x, y)
=− = − cos(x) cosh(y)
∂x ∂y
and
∂v(x, y) ∂u(x, y)
= = − sin(x) sinh(y).
∂y ∂x
It follows that
v(x, y) = − sin(x) cosh(y) + c(y),
where c(y) depends only on y, and then
∂v(x, y)
sin(x) sinh(y) + c′ (y) = = − sin(x) sin(y),
∂y
so c′ (y) = 0, that is, c(y) = C, a constant. Then v(x, y) = − sin(x) cosh(y) + C.
72 11. EXPONENTIAL, HYPERBOLIC AND TRIGONOMETRIC FUNCTIONS

As a check, observe that

u(x, y) + iv(x, y) = cos(x) sinh(y) − i sin(x) cosh(y) + iC
= cosh(ix) sinh(y) + sinh(−ix) cosh(y) + iC
= sinh(y − ix) + iC
= sinh(−iz) + iC
= −i sin(z) + iC,
where z = x + iy. Since this is a differentiable function of z, the function v that we
found must be correct. 4
 LECTURE 12

Logarithms and roots

In this lecture, we investigate some inverse functions, in particular, roots and

logarithms.
This part of the theory can become quite complicated, and we start by just
considering a few examples.

1. Some algebra
Many complex functions are not one-to-one. This means that inverse functions
are not well-defined. One solution is to deal with “multi-valued functions”, and
another is to restrict the domain of the original function. The archetypical example
is the argument function, and we usually write arg(z) to indicate the multi-valued
function and Arg(z) to indicate the particular choice where the values lie in the
range (−π, π].
√
Example 12.1. Suppose that w = z 2 . Then we may write z = w or w1/2 ; the
question is what this means. Unfortunately, different writers mean different things.
Some writers mean that z may be any of the two possible values; others mean that
a particular choice has been made. We will try to use the expressions above for the
multi-valued function, and add the words “principal value” (symbolically, PV) or
“principal branch” to indicate that a particular choice is being made. In particular,
we define
(
1/2 |w|1/2 ei Arg(w)/2 if w 6= 0
PV w =
0 if w = 0.
In words, we might say “the principal branch of the square root”.

We may think of the graph of the multi-valued function w1/2 as being like two
copies of the plane, slit along the branch cut, and then joined together cleverly. It
is not possible to do this in three dimensions, but it is possible in four dimensions.
Example 12.2. Suppose that w = ez and z = x + iy. Then w = ex eiy , so
|w| = ex and Arg w = Arg eiy .
Then x = ln |w|, and x is single-valued, but y = Arg(w) + 2πk, where k ∈ Z; and
y is multiple-valued. When w 6= 0, we write z = log(w) to indicate that z may be
any one of the infinitely many complex numbers such that ez = w and we write
z = Log(w) to indicate the choice that z = ln |w| + i Arg(w).

It is important to be very clear in what you write about multi-valued functions;

always include the words “multi-valued” or ”principal value” or the symbol PV.
73
74 12. LOGARITHMS AND ROOTS

Im Im

z 7→ z 6
Ω
Re Υ Re

z 7→ z 1/6

Im Im

z 7→ z 6

Re Υ Re
Ω

z 7→ z 1/6

Figure 12.1. Two branches of the 6th root function

2. Defining nth roots

Suppose that f (z) = z n , where n ≥ 2. Then f is not one-to-one. However,
if we restrict f to a region Ω such as {z ∈ C : | Arg(z)| < π/n}, then f becomes
one-to-one, and we may define an inverse function, z 7→ z 1/n . If we choose a different
region, we get a different inverse function. The principal value of the nth root is
given by
 
Log(z)
PV z 1/n
= exp = |z|1/n ei Arg(z)/n .
n

Proposition 12.3. The function PV z 1/n is differentiable in C \ (−∞, 0].

Proof. This function is the composition of Log, followed by division by n,

followed by the exponential. Compositions of differentiable functions are differen-
tiable. □

Note that where this function is differentiable and where it is continuous are
nearly the same; the only difference is that PV z 1/n is continuous at 0 but not
differentiable.
All the possible inverse functions are constant multiples of each other on con-
nected sets where both are defined, where the multiplying factors are nth roots of
unity. The different possible inverse functions are called branches of the nth root
function.
 3. SQUARE ROOTS OF POLYNOMIALS 75

Figure 12.1 illustrates some possibilities in the case where n = 6. In the upper
figure, we usually choose the inverse function on the negative real axis to map to
the upper side of the cone. It is given by f −1 (w) = |w|1/6 ei Arg(w)/6 .
In the lower figure, we usually choose the inverse function to map the positive
real axis to the lower side of the cone. It is given by f −1 (w) = |w|1/6 ei Arg (w)/6 ,
0

where Arg0 denotes the choice of argument in the range [0, 2π).
The first inverse function is the “standard” choice; it is called the principal value
of the 6th root; the notation PV z 1/6 should be used.
The regions Ω that we chose are relatively simple, but we could have chosen
more complicated versions, such as a region between two curves coming out of the
origin. This would have led to a curved branch cut.
In general, when we restrict a function f to a smaller domain in order to define
an inverse function for f , we try to make the smaller domain Ω as large as possible,
so that the domain Υ of the inverse function in as large as possible. The boundary
of Υ is called the branch cut (or cuts, as it may have a number of connected pieces).
By choosing Ω and Υ carefully, we may usually avoid having an inverse function
with discontinuities where we want to work. Although we may vary these sets by
varying the branch cut, there are some points, called branch points, which must
appear in any branch cut. For the case of the function f (z) = z n , the only branch
point is 0.

3. Square roots of polynomials

The next level of difficulty with inverse functions is with defining square roots
of polynomials.
First, we consider the problem of defining (z ± 1)1/2 . The principal branch of
(z − 1)1/2 is given by
(
|z − 1|1/2 ei Arg(z−1)/2 if z 6= 1
PV(z − 1) =
1/2
0 if z = 1.
This function is continuous in C \ (−∞, 1); the branch cut is the infinite interval
(−∞, 1] and the branch point is 1. In much the same way, the principal branch of
(z + 1)1/2 is given by
(
|z + 1|1/2 ei Arg(z+1)/2 if z 6= −1
PV(z + 1)1/2 =
0 if z = −1.

Now we consider the problem of defining (z 2 − 1)1/2 . We could define

(
|z 2 − 1|1/2 ei Arg(z −1)/2 when z 6= ±1
2

(z − 1) = PV(z − 1) =
2 1/2 2 1/2
(12.1)
0 when z = ±1.

Alternatively, since z 2 − 1 = (z − 1)(z + 1), we could define

(z 2 − 1)1/2 = PV(z − 1)1/2 PV(z + 1)1/2
(
|z 2 − 1|1/2 ei Arg(z−1)/2+i Arg(z+1)/2 when z 6= ±1 (12.2)
=
0 when z = ±1.
76 12. LOGARITHMS AND ROOTS

Im
Arg(z 2 − 1) → −π Arg(z 2 − 1) → π

bc bc
Re

Arg(z 2 − 1) → π Arg(z 2 − 1) → −π

Figure 12.2. The discontinuities of Arg(z 2 − 1)

Let us now determine whether these possible definitions coincide and where they
are continuous and where they are differentiable.
First, we consider |z 2 − 1|1/2 ei Arg(z −1)/2 . Since |z 2 − 1|1/2 is continuous, the
2

question is linked to where ei Arg(z −1)/2 is continuous. The possible discontinuities

2

are when Arg(z 2 − 1) is discontinuous, that is, when z 2 − 1 is real and z 2 − 1 ≤ 0.

Thus the possible discontinuities are when 0 ≤ z 2 ≤ 1, that is, −1 ≤ z ≤ 1, or when
z 2 ≤ 0, that is, z ∈ iR. Clearly
lim |z 2 − 1|1/2 ei Arg(z−1)/2+i Arg(z+1)/2 = 0,
z→±1

so the function is continuous at ±1. There are many possibilities to consider, best
represented in a diagram—see Figure 12.2. We deduce that this function is contin-
uous in C \ (iR ∪ (−1, 1)).
We may check that this function is differentiable in C \ (iR ∪ [−1, 1]): indeed, it
is the composition of differentiable functions there.
Next, we consider |z 2 −1|1/2 ei Arg(z−1)/2+i Arg(z+1)/2 . Since |z 2 −1|1/2 is continuous,
the question is linked to where ei Arg(z−1)/2+i Arg(z+1)/2 is continuous. The possible
discontinuities are when Arg(z − 1) or Arg(z + 1) are discontinuous, that is, when
z is real and z ≤ 1. Clearly
lim |z 2 − 1|1/2 ei Arg(z−1)/2+i Arg(z+1)/2 = 0,
z→±1

so the function is continuous at ±1. Observe that, on the one hand, if −1 < x0 < 1,
then
lim Arg(z − 1)/2 + Arg(z + 1)/2 = π/2
z→x 0
Im(z)>0

lim Arg(z − 1)/2 + Arg(z + 1)/2 = −π/2,

z→x0
Im(z)<0

and it follows that

lim ei Arg(z−1)/2+i Arg(z+1)/2 = i
z→x0
Im(z)>0

lim ei Arg(z−1)/2+i Arg(z+1)/2 = −i,

z→x0
Im(z)<0
 3. SQUARE ROOTS OF POLYNOMIALS 77

so that there is a discontinuity at x0 . On the other hand, if x0 < −1, then

lim Arg(z − 1)/2 + Arg(z + 1)/2 = π
z→x0
Im(z)>0

lim Arg(z − 1)/2 + Arg(z + 1)/2 = −π,

z→x0
Im(z)<0

and it follows that

lim ei Arg(z−1)/2+i Arg(z+1)/2 = −1
z→x0
Im(z)>0

lim ei Arg(z−1)/2+i Arg(z+1)/2 = −1,

z→x0
Im(z)<0

so that there is no discontinuity at x0 . Thus this function is continuous in C\(−1, 1).

We will see later that it is differentiable in C \ [−1, 1].
We could describe the choice (12.2) in words as the “branch of (z 2 − 1)1/2 that is
continuous in C\(−1, 1) and takes positive values on (1, ∞)”. This second definition
leads to a continuous inverse function with a larger domain, and is preferable.
Exercise 12.4. Find an explicit formula for the “branch of (1 − z 2 )1/2 that is
continuous except on the real intervals (−∞, −1) and (1, ∞) and that takes the
value 1 at 0”. Where is this function differentiable?
Answer. The allowed discontinuities for this function are on the real axis but
not on [−1, 1],
The function PV(1 − z)1/2 is discontinuous when 1 − z is a negative real number,
that is, when z ∈ (1, ∞). Similarly, the function PV(1 + z)1/2 is discontinuous when
1 + z is a negative real number, that is, when z ∈ (−∞, −1).
It follows that PV(1 − z)1/2 PV(1 + z)1/2 is continuous except when z ∈ (1, ∞)
or z ∈ (−∞, −1). Further, PV(1 − z)1/2 PV(1 + z)1/2 = 1 when z = 0. So this is
the solution. We may also write
1/2 i(Arg(1−z)+Arg(1+z))/2
1 − z2 e .
Alternatively, the function PV(1−z 2 )1/2 is discontinuous when 1−z 2 is a negative
real number, that is, when z 2 ∈ (1, ∞), that is, when z ∈ (−∞, −1) or z ∈ (1, ∞).
Since this function takes the value 1 at 0, it is also the solution. We may also write
1/2 i Arg(1−z 2 )/2
1 − z2 e . 4

The next step up in complication is the square root of a cubic polynomial. For
instance, we might take the multi-valued function w = (z(z 2 − 1))1/2 . “Graphs” of
functions like this are called elliptic curves. They are important in number theory—
they are central to Andrew Wiles’ proof of Fermat’s last theorem—and they are
important in cryptography—they give rise to “better” codes.
Exercise 12.5. Show that there is a choice of definition of (z(z 2 − 1))1/2 that
is continuous except on the intervals (−∞, −1) and (0, 1).
The graph of the multi-valued function is like two copies of the plane, with slits
along the branch cuts, joined together appropriately.
78 12. LOGARITHMS AND ROOTS

4. More examples
The logarithm log is multi-valued because the argument arg is multi-valued.
There are two common choices of argument, one between −π and π, and the other
between 0 and 2π. Both of these give logarithms with the property that log(i) = π/2,
but the first has its branch cut along the negative real axis and the second has its
branch cut along the positive real axis. If we want to deal with a logarithm that is
continuous around 1, the first choice is better, but if we want to arrange continuity
around −1, then the second is better.
Soon we will need to look at expressions such as log(w + (w2 − 1)1/2 ). We try to
choose a branch of log and a branch of the square root to make the function con-
tinuous in as large a domain as possible. Usually this function will be differentiable
where it is continuous, with some possible exceptions.
(1) While the arg–Arg notation seems fairly√standard, there does not seem to
be any standard about whether z 1/n or n z is uniquely defined. We use the
word “multi-valued” wherever appropriate to try to avoid any ambiguity.
 LECTURE 13

Inverses of exponential and related functions

In this lecture, we discuss the inverse functions for the exponential, hyperbolic
and trigonometric functions introduced in the last lecture.

1. Inverse functions
The following exercise is a warm-up.
Exercise 13.1. Fix w in C. Find all z in C such that
(a) exp(z) = w, (b) cosh(z) = w, (c) sinh(z) = w.

Answer. (a) If ez = w, then we may write

z = Log (w) + 2πik,
where k ∈ Z. Alternatively, we may write
z = log (w) ,
where log is multi-valued.
(b) If cosh(z) = w, then ez + e−z = 2w, so (ez )2 − 2w(ez ) + 1 = 0, whence
p
2w ± PV (2w)2 − 4 √
ez = = w ± PV w2 − 1.
2
√ √
Noting that (w + PV w2 − 1)(w − PV w2 − 1) = 1, we conclude that
√
z = ± Log(w + PV w2 − 1) + 2πik
for some k ∈ Z. Each step of this argument is reversible, so every z of this form
satisfies cosh(z) = w. We could also write
√  √ 
z = ln w ± PV w2 − 1 + i Arg w ± PV w2 − 1 + 2πik,

where k ∈ Z and we take the same choice of ± in both ln and Arg, or

 √ 
z = log w + w − 1 ,
2

where both log and square root are multi-valued.

Part (c) is similar. 4

Note that cosh is even, so if cosh(z) = w, then cosh(−z) = w. Ideally, the

solution set should be chosen so that this fact is evident.
79
80 13. INVERSES OF EXPONENTIAL AND RELATED FUNCTIONS

Im Im
i(θ + 2π) b

Sθ

Re Re

b
Rθ
iθ

exp

Figure 13.1. The exponential map

2. The exponential function

Now we consider inverse functions for the exponential function. Here is one
candidate.
Definition 13.2. The principal branch of the complex logarithm is the function
Log from C \ {0} to C, given by
Log(z) = ln |z| + i Arg(z),
where Arg(z) takes values in the range (−π, π].
By properties of the exponential, if z = reiθ , then
eLog(z) = eln(r)+iθ = reiθ = z;
however, if z = x + iy, then
Log(ez ) = ln |ez | + i Arg(ez ) = ln(ex ) + i Arg(eiy ) = x + iy + 2πik,
where the integer k is such that −π < y + 2πk ≤ π, and it is possible that Log(ez ) 6=
z.
Other candidates for inverse function are sometimes more useful than Log. We
define the ray Rθ and the horizontal strip Sθ (where θ ∈ R) in the complex plane by
Rθ = {w ∈ C : Arg(w) − θ ∈ 2πZ}
Sθ = {z ∈ C : θ < Im(z) < θ + 2π}.
The exponential map exp takes horizontal lines to rays, and is one-to-one and onto
from the open horizontal strip Sθ to C \ Rθ .
We may define an inverse function logθ from C \ Rθ to Sθ :
logθ (w) = ln |w| + i argθ (w),
where argθ (w) is the argument in the range (θ, θ + 2π).
As θ varies, we get different inverse functions. These inverse functions are called
branches of the complex logarithm, and the rays Rθ are called branch cuts. Different
branches of the logarithm differ by a constant in connected open sets where they
 3. COMPLEX POWERS 81

are both defined, The point 0, which is common to all the branch cuts, is called a
branch point.
Lemma 13.3. For any branch logθ of the complex logarithm,
1
log′θ (w) =
w
for all w ∈ C \ Rθ .

Proof. We apply the Cauchy–Riemann equations. □

The notation logθ is not standard, and we will not use it any more. Rather, we
use the expression “the branch of the logarithm with imaginary part in (θ, θ + 2π)”.
A different sort of “inverse function” of the exponential function is the “multi-
function” log from C \ {0} to C given by
log(z) = ln |z| + i arg(z).
This is a “multifunction” in the sense that it takes multiple values, because arg(z)
takes multiple values.

3. Complex powers
We define complex powers of complex numbers using exponentials and loga-
rithms.
Definition 13.4. Given z ∈ C \ {0} and α ∈ C, we define
z α = exp(α log(z)).
The principal branch of z α is found by using Log, the principal branch of the loga-
rithm. That is, PV z α = exp(α Log(z)).

The possible values of z α are exp(α Log(z) + 2πikα) where k ∈ Z. Different

values of k may give very different values of z α .
Lemma 13.5. The function z 7→ PV z α is differentiable in C \ (−∞, 0], with
derivative α PV z α /z.

Proof. This follows because Log is differentiable in C \ (−∞, 0]. □

Exercise 13.6. Compute the possible values of ii ; which is the principal value?

Answer. By definition, the possible values are

π π
ii = exp(i log(i)) = exp(i(i + 2πik)) = exp(− − 2πk),
2 2
where k ∈ Z. The principal value is when k = 0, and so it is e−π/2 . 4

Often we write ez rather than exp(z). Note that this is ambiguous, since complex
powers are multi-valued! So arguably exp(z) is better notation.
82 13. INVERSES OF EXPONENTIAL AND RELATED FUNCTIONS

4. The inverse hyperbolic sine

Last lecture, we defined the hyperbolic sine and cosine, and established some of
their properties. Now we consider the inverse function(s) of sinh.
Exercise 13.7. Find all z in C such that sinh(z) = w .

Answer. If sinh(z) = w, then ez − e−z = 2w, so (ez )2 − 2wez − 1 = 0, whence

2w ± PV(4w2 + 4)1/2
ez = = w ± PV(w2 + 1)1/2
2
z = Log(w ± PV(w2 + 1)1/2 ) + 2πik
= ln w ± PV(w2 + 1)1/2 + i Arg(w ± PV(w2 + 1)1/2 ) + 2πik
where k ∈ Z. Note that we must take the same choice of ± in both the ln part and
in the Arg part.
Alternatively, we may note that
(w + PV(w2 + 1)1/2 )(w − PV(w2 + 1)1/2 ) = w2 − (w2 + 1) = −1;
hence ez is either w + PV(w2 + 1)1/2 or −(w + PV(w2 + 1)1/2 )−1 , and either
z = Log(w + PV(w2 + 1)1/2 ) + 2πik
or
z = − Log(w + PV(w2 + 1)1/2 ) + 2πik + πi.
This is a better answer because there is less room for ambiguity.
We may also write
z = log(w + PV(w2 + 1)1/2 ) and z = − log(w + PV(w2 + 1)1/2 ) + πi,
where log is multi-valued. 4

The principal branch of the inverse hyperbolic sine function is given by

PV sinh−1 w = Log(w + PV(w2 + 1)1/2 ).

It is easy to show that for any complex number w,

sinh PV sinh−1 w = w;
however, it is not necessarily true that PV sinh−1 sinh w = w.
Exercise 13.8. Find the possible values of PV sinh−1 sinh w − w, where w ∈ C.

Answer. We leave this to the reader. 4

Both the logarithm and the square root are possible causes of discontinuity. The
function PV(w2 + 1)1/2 is continuous as long as w2 + 1 is not in the interval (−∞, 0],
and the logarithm is continuous as long as w + PV(w2 + 1)1/2 is not in the interval
(−∞, 0].
On the one hand, if w2 + 1 is not in (−∞, 0], then w2 is not in (−∞, −1]. So
one possible discontinuity is when w = iv, where |v| ≥ 1.
 4. THE INVERSE HYPERBOLIC SINE 83

On the other hand, we may try to solve the equation w + PV(w2 + 1)1/2 = −t
for t ∈ [0, ∞); we get
PV(w2 + 1)1/2 = −t − w
w2 + 1 = t2 + w2 + 2tw
1 − t2
w= ,
2t
and so w is real. But if w is real, then
w + PV(w2 + 1)1/2 = w + (w2 + 1)1/2 > 0,
and so w + PV(w2 + 1)1/2 is not in the interval (−∞, 0]. Thus the only possible
discontinuities are when w = iv, where v is real and |v| ≥ 1.
Lemma 13.9. The principal branch of the inverse hyperbolic sine function is
differentiable in C \ ([i, +i∞) ∪ (−i∞, −i]). Further,

′ 1
PV sinh−1 (w) = √ .
PV w2 + 1
Proof. We compute the derivative:

d sinh−1 (w) d Log(w + PV(w2 + 1)1/2 )

=
dw dw
1 + w/PV(w2 + 1)1/2
= √
w + PV w2 + 1
1
= ,
PV(w2 + 1)1/2
as required. □
The inverse hyperbolic cosine may be treated in a similar way. We define
PV cosh−1 (w) = Log(w + PV(w + 1)1/2 PV(w − 1)1/2 ).
Exercise 13.10. Show that
d PV cosh−1 (w) 1
=
dw PV(w + 1) PV(w − 1)1/2
1/2

for most w ∈ C. Where is cosh−1 not differentiable?

Answer. We compute the derivative:

d cosh−1 (w) d Log(w + PV(w + 1)1/2 PV(w − 1)1/2 )

=
dw dw
1 + 12 PV(w + 1)1/2 / PV(w − 1)1/2 + 21 PV(w − 1)1/2 / PV(w + 1)1/2
=
w + PV(w + 1)1/2 PV(w − 1)1/2
1
= .
PV(w2 + 1)1/2
Aa far as continuity is concerned, the potential problems are when w + 1 or w − 1
or w + PV(w + 1)1/2 PV(w − 1)1/2 lies in (−∞, 0]. Observe that if
w + PV(w + 1)1/2 PV(w − 1)1/2 = −a,
84 13. INVERSES OF EXPONENTIAL AND RELATED FUNCTIONS

where a ≥ 0, then PV(w + 1)1/2 PV(w − 1)1/2 = −a − w, so w2 − 1 = w2 + 2aw + a2 ,

whence w = −(a2 + 1)/a, that is, w is a negative real number. So the potential
discontinuities are when w ∈ (−∞, +1].
By studying what happens as w tends to a, we may show that each point in
(−∞, +1] is a discontinuity.
It turns out that cosh−1 is not differentiable on (−∞, +1]. 4

5. The inverse trigonometric functions

We may define the inverse trigonometric functions using the formulae cos(iz) =
cosh(z) and sin(iz) = i sinh(z). For example, if z = cos−1 (w), then w = cos(z) =
cosh(iz), and so iz = cosh−1 (w).
Exercise 13.11. What are the ranges of sinh−1 , cos−1 , and sin−1 ? Where are
the branch cuts for cos−1 and sin−1 ?
 LECTURE 14

Paths and path integrals† (Not examinable)

In this lecture, we define paths and path integrals, and see a key theorem about
these. This material should be familiar to students who have studied multi-variable
calculus.
The main question underlying this lecture is the relation between integration
and differentiation. As before, we represent points in R2 by row vectors rather than
by column vectors.

1. Curves
Definition 14.1. A curve γ in R2 is a continuous function from an interval [a, b]
of real numbers into R2 . We may write γ(t) = (γ1 (t), γ2 (t)), where γ1 and γ2 are
real-valued; then the continuity of γ is equivalent to the continuity of both γ1 and
γ2 .
The initial point of the curve is γ(a) and the final point of the curve is γ(b). The
range of the curve is the set of points {γ(t) : t ∈ [a, b]}.
A curve γ : [a, b] → R2 is said to be closed if γ(a) = γ(b), and simple if
γ(s) 6= γ(t) when s < t, except perhaps if s = a and t = b.

Examples 14.2. In Figure 14.1, the curve α moves from p to q, the curve β
moves up the parabolic arc, the curve γ moves from the right to the left of the
line segment and then back again, the curve δ moves once around the circle in the
anticlockwise direction, starting at the right-most point, the curve ε moves twice
around the “infinity-shaped” figure, starting at the right-most point and moving
upwards, and the curve ζ moves along the “alpha-shaped” figure.
The curves α and β are simple but not closed; γ and ε are closed but not simple;
δ is both simple and closed; and ζ is neither closed not simple. The curve γ in the
figure “goes back on itself”, while the curves ε and ζ intersect themselves. The
curve ε repeats itself.

Definition 14.3. Suppose that α : [a, b] → R2 and β : [c, d] → R2 are curves such
that the final point of [a, b] is the initial point of [c, d] and the final point of α is the
initial point of β, that is, α(b) = β(c). The join α t β of α and β is the curve
(
α(t) when a ≤ t ≤ b
α t β(t) =
β(t) when c ≤ t ≤ d.

Suppose that γ : [a, b] → R2 is a curve. The reverse curve γ ∗ : [−b, −a] → R2 of

γ is given by
γ ∗ (t) = γ(−t) where −b ≤ t ≤ −a.

85
86 14. PATHS AND PATH INTEGRALS† (NOT EXAMINABLE)

y
α(t) = p + (q − p)t b q
= p(1 − t) + qt ∀t ∈ [0, 1]
α
β(t) = ( 23 + t, t ) ∀t ∈ [0, 1]
2
p b

ε
b

β δ
γ(t) = (1 + 2t , −3) ∀t ∈ [−1, 1]
2
x
b b b

δ(t) = (1 + 22
10
cos(t), 22
10
sin(t)) ∀t ∈ [0, 2π]
ε(θ) = (cos(θ), sin(θ) cos(θ)) ∀θ ∈ [0, 4π]
ζ
ζ(θ) = (cos(θ) − 2, sin(θ) cos(θ) − 3) b

γ
∀θ ∈ [π/4, 7π/4]
b

Figure 14.1. Examples of curves

y
γ
α(t) = (t, 0) ∀t ∈ [0, 1] 1
β(t) = (1, t − 1) ∀t ∈ [1, 2] δ β
γ(t) = (3 − t, 1) ∀t ∈ [2, 3]
δ(t) = (0, 4 − t) ∀t ∈ [3, 4] α 1 x

Figure 14.2. Example of a join of curves

Example 14.4. In Figure 14.2, the join αtβ tγ tδ moves anticlockwise around
the perimeter of the square with vertices (0, 0), (1, 0), (1, 1) and (0, 1), starting and
ending at (0, 0), as t varies from 0 to 4. The curve α∗ moves from 1 to 0 as t varies
between 0 and 1. What does (α t β t γ t δ)∗ do?

Definition 14.5. Suppose that γ : [a, b] → R2 is a curve, and that h is a continuous

bijection from [c, d] to [a, b] such that h(c) = a and h(d) = b. Then γ ◦h : [c, d] → R2
is also a curve, called a reparametrisation of γ.
A reparametrised curve γ ◦ h has the same initial and final point as the original
curve, and the same range; further it is traversed in the same direction and it goes
through each point the same number of times. What may change is the interval of
definition, and the speed.
There are some very strange curves, such as space-filling curves and snowflake
curves that violate some of our intuitions about curves. It is helpful to restrict to
certain “nice” curves that are more tractable.
Definition 14.6. Suppose that γ : [a, b] → R2 is a curve and γ = (γ1 , γ2 ), where
γ1 , γ2 : [a, b] → R. Then we define
γ ′ (t) = (γ1′ (t), γ2′ (t)),
provided that both γ1′ (t) and γ2′ (t) exist. We say that γ is continuously differen-
tiable if the derivative γ ′ exists and is continuous in [a, b], and γ is smooth if it is
continuously differentiable and moreover γ ′ (t) 6= 0 for all t ∈ [a, b].
 2. ORIENTATION 87

α(t) = (cos(t), sin(t)) ∀t ∈ [0, 2π] 1

α′ (t) = (− sin(t), cos(t)) ∀t ∈ [0, 2π] α
β
β(t) = (cos3 (t), sin3 (t)) ∀t ∈ [0, 2π] 1 x
′ 2 2
β (t) = (−3 sin(t) cos (t), 3 sin (t) cos(t))
∀t ∈ [0, 2π]

Figure 14.3. Examples of differentiable curves

Examples 14.7. In Figure 14.3, both curves α and β are continuously differ-
entiable. Only α is smooth.

Definition 14.8. A curve is piecewise smooth if it is a join of finitely many smooth

curves. The length of a piecewise smooth curve γ : [a, b] → R2 is given by
Z b
Length(γ) = |γ ′ (t)| dt.
a
′
Note that γ (t) may not be defined at finitely many points, where different
smooth curves are joined. This is just a formal difficulty; we take γ ′ (t) to be 0
where it is not defined, and then the integral exists as a Riemann integral. For this
it is important that γ ′ is bounded.

2. Orientation
If γ is a simple curve that is not closed, then it has an initial point and an
endpoint and a “direction of motion”, and even if we reparametrise it, these do not
change. Similarly, simple closed curves have a “direction of motion”. We call this
orientation. For simple closed curves, the next theorem lets us find an orientation
that does not depend on the parametrisation.
Theorem 14.9 (The Jordan curve theorem). If γ : [a, b] → R2 is a simple closed
curve, then the complement of the range of γ is the union of two disjoint domains.
One of these is bounded and the other is not. The bounded component is called the
interior of γ and written Int(γ), and the unbounded component is called the exterior
of γ and written Ext(γ).
Proof. We do not prove this, but remark that it relies on approximating a
simple curve by a simple polygonal curve. □
This theorem seems obvious but is actually quite difficult, particularly when
curves such as snowflakes are concerned. We will consider only curves for which it
is easy to identify the interior and exterior.
If γ is a simple closed curve, then as we travel around γ, the interior of γ will
always lie on our left, or always on our right. The standard orientation of γ is the
direction of motion such that the interior is always on our left. With the standard
orientation, we move around the perimeter of Int(γ) in an anti-clockwise direction.
88 14. PATHS AND PATH INTEGRALS† (NOT EXAMINABLE)

Exercise 14.10. What is the standard orientation of each of the simple closed
curves described above? Can you define orientation for closed curves that are not
simple?

We will use the expression oriented range of a curve to describe the image of a
simple curve in R2 and a “direction of motion” along the curve.

3. Vector fields and line integrals

A vector field V on an open subset Ω of R2 is a function from Ω to R2 . A vector
field may be viewed as a pair of functions (v, w), where v, w : Ω → R.
Definition 14.11. Given a piecewise smooth curve γ : [a, b] → R2 and
R a con-
tinuous vector field V defined on Range(γ), we define the line integral γ V (s) ds
by
Z Z b
V (s) ds = V (γ(t)) · γ ′ (t) dt.
γ a

Again, γ ′ (t) may not be defined for finitely many points, where different con-
tinuously differentiable curves are joined. We take γ ′ (t) to be 0 wherever it is not
defined, and then the integral exists as a Riemann integral.
We now consider the approximation of piecewise smooth curves by polygonal
curves. If γ : [a, b] → R2 is a piecewise smooth curve, we may define “approximating
curves” γ N (where N ∈ Z+ ) as follows. Fix N , and subdivide the interval [a, b] into
N equal subintervals of equal length, [an−1 , an ] say, where n = 1, . . . , N (we do
this by choosing an = (N − n)a/N + nb/N ). Now let γ N be the polygonal curve
composed of the N line segments from γ(an−1 ) to γ(an ), in order.
Lemma 14.12. Suppose also V is a continuous vector field in a domain Ω in R2 .
Suppose that γ is a piecewise smooth curve in Ω, and that the polygonal curves γ N
defined above also lie in Ω. Then the curves γ N approximate γ, in the sense that
γ N (t) → γ(t) for all t ∈ [a, b] and Length(γ N ) → Length(γ) as N → ∞; further,
Z Z
V (s) ds → V (s) ds.
γN γ

Proof. Omitted. □

The point of this lemma is that we may prove many results for integration along
general curves by proving easier results for integration along polygonal curves.
The line integral has the usual properties of integration.
Theorem 14.13. Suppose that λ, µ ∈ R, that γ : [a, b] → R2 is a piecewise
smooth curve, and that V and W are vector fields defined on Range(γ). Then the
following hold.
(a) The integral is linear:
Z Z Z
λV (s) + µW (s) ds = λ V (s) ds + µ W (s) ds.
γ γ γ
 3. VECTOR FIELDS AND LINE INTEGRALS 89

(b) The integral is independent of parametrisation: if δ is a reparametrisation of γ

that is also a piecewise smooth curve, then
Z Z
V (s) ds = V (s) ds.
γ δ

(c) The integral is additive for joins: if γ = α t β, then

Z Z Z
V (s) ds = V (s) ds + V (s) ds.
γ α β

(d) The integral depends on the orientation:

Z Z
V (s) ds = − V (s) ds.
γ∗ γ

(e) The size of the integral is controlled by the size of the vector field V and the
length of the curve γ:
Z
V (s) ds ≤ M L,
γ

where L is the length of γ and M is a number such that |V (s)| ≤ M for all
s ∈ Range(γ).

Proof. We omit this proof. □

Definition 14.14. We define a path Γ to be the oriented range of a piecewise

smooth curve γ. The theorem above implies that the line integral depends only on
Γ, and hence we define
Z Z
V (s) ds = V (s) ds,
Γ γ

where γ is any parametrisation of Γ.

We may extend the notation for joins and reverse curves to paths.
Note that for paths that are not closed, different parametrisations differ by a
change of variable, and part (d) of the theorem applies. But for closed paths,
parametrisations might have different initial points and endpoints. To show that
the integral around a closed path does not depend on where we start, we break up
the closed path Γ into two parts: Γ = A t B, and observe that
Z Z Z
V (s) ds = V (s) ds + V (s) ds
A⊔B
Z A
Z B
Z
= V (s) ds + V (s) ds = V (s) ds.
B A B⊔A

This implies that the integral around Γ starting at the initial point of A is the same
as the integral around Γ starting at the initial point of B.
90 14. PATHS AND PATH INTEGRALS† (NOT EXAMINABLE)

4. Closed and exact vector fields

Suppose that the vector field V : Ω → R2 is given in coordinates by (v, w).
Then V is said to be curl-free or closed if ∂w/∂x = ∂v/∂y. Further, V is said to
be conservative or exact if there is a function u : Ω → R such that ∂u/∂x = v and
∂u/∂y = w; this is often written as V = ∇u. If the function u is twice continuously
differentiable, then
∂w ∂ 2u ∂ 2u ∂v
= = = .
∂x ∂x ∂y ∂y ∂x ∂y
This leads to the conclusion that a continuously differentiable exact vector field is
closed. The converse of this is very useful.
Theorem 14.15. Suppose that Ω is a simply connected domain, and that V is a
closed continuously differentiable vector field in Ω. Then
Z
V (s) ds = 0
γ
for all closed piecewise smooth curves γ whose range lies in Ω.
Proof. We omit this proof, but mention that Green’s theorem is often used
when γ is simple. We will prove a more general result about complex integrals,
whose proof may be modified to establish this result.
Then the result is extended to closed polygonal curves, and finally an approxi-
mation argument is used to extend to general closed piecewise smooth curves. □
Theorem 14.16. Suppose that Ω is a simply connected domain, and that V is a
closed continuously differentiable vector field in Ω. Then V is exact.

R Proof. We omit this proof, but mention that the previous theorem implies that
γ
V (s) ds depends only on the initial and final points of the curve γ, which allows
us to define Z
u(q) = V (s) ds
γ
for any curve γ from a fixed point p to q. □
This theorem has variants; in particular, it is equivalent to the statement that a
line integral depends only on the initial point and the end point.
(1) We have seen that not all vector fields can be integrated “nicely”. However,
the integral of a derivative is the function that we started with, up to a
constant of integration.
(2) We will prove similar results in the context of complex analysis, where line
integrals become contour integrals, vector fields become complex-valued
functions, and the condition for a vector field to be closed becomes the
Cauchy–Riemann equations.
(3) The material in this lecture is not examinable, except that we re-use without
further mention definitions such as closed and simple (for curves) in future
lectures.
 LECTURE 15

Contour integrals

In this lecture, we define contour integrals, and compute some examples.

1. Curves and contours

We define a curve in C much as in R2 : a curve t 7→ (γ1 (t), γ2 (t)) is replaced by
a curve t 7→ γ1 (t) + iγ2 (t). The definitions of simple and closed curves are almost
identical, as are joins, reverse curves, reparametrisations, and orientations.
Definition 15.1. Suppose that γ : [a, b] → C is a curve and γ(t) = γ1 (t)+iγ2 (t),
where γ1 , γ2 : [a, b] → R, that is, γ1 (t) = Re(γ(t)) and γ2 (t) = Im(γ(t)) for all
t ∈ [a, b]. Then we define
γ ′ (t) = γ1′ (t) + iγ2′ (t),
provided that both γ1′ (t) and γ2′ (t) exist. We say that γ : [a, b] → C is continuously
differentiable if γ ′ is defined and continuous on [a, b], smooth if γ is continuously
differentiable and moreover γ ′ (t) 6= 0 for all t ∈ [a, b], and piecewise smooth if it is
the join of finitely many smooth curves.
Note that
1
γ ′ (t) = lim (γ(t + h) − γ(t)) .
h→0 h
We may differentiate many complex-valued functions of a real variable as if they
were real-valued. For instance, if α(t) = eit , then α′ (t) = ieit . The usual rules
of differentiation hold for complex curves and functions defined on complex curves,
and we do not always need to break things up into their real and imaginary parts in
order to differentiate. In particular, if f is a holomorphic function and α is a curve
in the complex plane, then
d
f (α(t)) = f ′ (α(t)) α′ (t).
dt
We can prove this by breaking everything up into real and imaginary parts, and
applying the chain rule from multi-variable calculus, or by copying the proof from
calculus of one real variable.
Definition 15.2. A contour is an oriented range of a piecewise smooth curve in the
complex plane.
A contour is the analogue in the complex plane of a path in R2 ; usually we deal
with simple contours. We may extend the definition of joins and reverse curves in
R2 to curves in C and hence to contours.
Examples 15.3. Figure 15.1 illustrates some contours in the complex plane. In
the absence of other information, we assume that closed contours have the standard
orientation.
91
92 15. CONTOUR INTEGRALS

α(t) = cos(t) + i sin(t) = eit ∀t ∈ [0, 2π] i

α
β(t) = (1 − t) + it ∀t ∈ [0, 1] γ β
γ(t) = −t + i(1 − t) ∀t ∈ [0, 1] δ ε 1

δ(t) = (t − 1) − it ∀t ∈ [0, 1]
ε(t) = t + i(t − 1) ∀t ∈ [0, 1]

Figure 15.1. Examples of parametrised contours

2. Integrals with complex integrands

Definition 15.4. Suppose that u, v : [a, b] → R are real-valued functions, and
that f : [a, b] → C is given by f = u + iv. We define
Z Z Z Z
b b
b b
f (t) dt = u(t) + iv(t) dt = u(t) dt + i v(t) dt,
a a a a

provided that the two real integrals on the right hand side exist. In other words,
Z b  Z b Z b  Z b
Re f (t) dt = Re (f (t)) dt and Im f (t) dt = Im (f (t)) dt.
a a a a

Integration of complex-valued functions on an interval has similar properties to

standard integration. For instance, it is linear, we may integrate by substitution
and by parts, and exponentials integrate as we would expect: for a, b, c, d ∈ R,
λ, µ ∈ C, a real-valued differentiable function h : [c, d] → [a, b] such that h(c) = a
and h(d) = b, and complex-valued functions f and g,
Z b Z b Z b
λf (t) + µg(t) dt = λ f (t) dt + µ g(t) dt
a a a
Z d Z b
′
f (h(t)) h (t) dt = f (s) ds
c a
Z b Z b
′


f (t) g(t) dt = f (b) g(b) − f (a) g(a) − f (t) g ′ (t) dt
a a
Z b h eλt it=b e −e
λb λa
eλt dt = =
a λ t=a λ
Z b Z b
f (t) dt ≤ |f (t)| dt .
a a

We prove only the last part: if the left hand side is 0, there is nothing to prove,
otherwise we take θ to be the argument of the left hand side. Then
Z b Z b Z b
−iθ
f (t) dt = e f (t) dt = e−iθ f (t) dt.
a a a
 3. CONTOUR INTEGRALS 93

Since the left-hand side is real, so is the right-hand side, and so

Z b Z b 
−iθ
f (t) dt = Re e f (t) dt
a a
Z b

= Re e−iθ f (t) dt
a
Z b
≤ e−iθ f (t) dt
a
Z b
= |f (t)| dt.
a
Rπ
Exercise 15.5. Evaluate 0
t eit dt.
Answer. We integrate by parts:
Z π Z  Z π 
1 π deit 1  it t=π dt it
it
te dt = t dt = te t=0 − e dt
i 0 dt i 0 dt
0
 Z π   
1 1 1  it t=π
= πe −
iπ it
e dt = −π − e t=0
i i i
 0

1 1 iπ

= −π − e − 1 = iπ − 2. 4
i i
Rπ
Rπ We take real and imaginary parts and deduce that 0
t cos(t) dt = −2 and
R (1+i)t
0
t sin(t) dt = π. It is much easierR t to find an integral like e dt by using
complex powers than to evaluate e cos(t) dt.

3. Contour integrals
Definition 15.6. Given a piecewise smooth curve γ : [a, b] → C and a contin-
uous (not necessarily differentiable)
R function f defined on the range of γ, we define
the complex line integral γ f (z) dz by
Z Z b
f (z) dz = f (γ(t)) γ ′ (t) dt,
γ a

provided that the integral on the right hand side exists.

dz
To remember this formula, write z(t) rather than γ(t): then dz = dt, and
dt
Z Z b
dz
f (z) dz = f (z(t)) dt.
γ a dt
If we now convert back to writing γ(t), which is a good idea, because we use z to
mean too many different things, then we get the formula in the definition.
Note that in general,
Z Z
Re f (z) dz 6= Re(f (z)) dz,
γ γ

and similarly for the imaginary part.

94 15. CONTOUR INTEGRALS

Theorem 15.7. Suppose that λ, µ ∈ C, that γ : [a, b] → C is a piecewise smooth

curve, and that f and g are complex functions defined on Range(γ). Then the
following hold.
(a) The integral is linear:
Z Z Z
λf (z) + µg(z) dz = λ f (z) dz + µ g(z) dz.
γ γ γ

(b) The integral is independent of parametrisation: if δ is a reparametrisation of γ

that is also a piecewise smooth curve, then
Z Z
f (z) dz = f (z) dz.
γ δ

(c) The integral is additive for joins: if γ = α t β, then

Z Z Z
f (z) dz = f (z) dz + f (z) dz.
γ α β

(d) The integral depends on the orientation:

Z Z
f (z) dz = − f (z) dz.
γ∗ γ

(e) The size of the integral is controlled by the size of the function f and the length
of the curve γ: Z
f (z) dz ≤ M L,
γ
where L is the length of γ and M is a maximiser for |f | on the curve, that is, a
number such that |f (z)| ≤ M for all z ∈ Range(γ).

Proof. We omit the proof of parts (a) to (d).

For part (e), observe that
Z Z b Z b
′
f (z) dz = f (γ(t)) γ (t) dt ≤ f (γ(t)) γ ′ (t) dt
γ a a
Z b
≤ M γ ′ (t) dt = M L,
a

by the formula for the length of a curve. □

Part (e) of the theorem is often called the M L Lemma. Note that M is not
necessarily the maximum of |f | on the curve.
Recall that a contour Γ is the oriented range of a piecewise smooth curve γ. The
theorem above implies that the complex line integral depends only on Γ, and not
on the parametrisation γ.
Definition 15.8. We define
Z Z
f (z) dz = f (z) dz,
Γ γ

where γ is any parametrisation of Γ.

 4. EXAMPLES OF CONTOUR INTEGRALS 95

4. Examples of contour integrals

We consider some examples of contour integration.
Exercise 15.9. Suppose thatZ p, q ∈ C, that Γ is the line segment from p to q,
and that λ1 , λ2 , λ3 , λ4 ∈ C. Find λ1 + λ2 z + λ3 z + λ4 ez dz.
Γ

Answer. We parametrise Γ: let γ(t) = (1 − t)p + tq = p + t(q − p), where

0 ≤ t ≤ 1. Then γ ′ (t) = q − p. We treat the summands separately and then add,
using linearity. First
Z Z 1 Z 1
′
dz = γ (t) dt = (q − p) dt = q − p.
γ 0 0

Next
Z Z 1 Z 1
′
z dz = γ(t) γ (t) dt = [p + t(q − p)] (q − p) dt
γ 0 0
q 2 − p2
= p(q − p) + 12 (q − p)2 = .
2
Third,
Z Z 1 Z 1 
′
z dz = γ(t) γ (t) dt = p + t(q − p) (q − p) dt
γ 0 0
(q + p)(q − p)
= p(q − p) + 12 (q − p)(q − p) = .
2
Finally,
Z Z 1 Z 1
′
z
e dz = e γ(t)
γ (t) dt = ep et(q−p) (q − p) dt
γ 0 0
h it=1  
= ep et(q−p) = ep e(q−p) − 1 = eq − ep .
t=0

Hence
Z
λ1 + λ2 z + λ3 z + λ4 ez dz
γ
q 2 − p2 (q + p)(q − p)
= λ1 (q − p) + λ2 + λ3 + λ4 (eq − ep ). 4
2 2
Alternatively, if λ3 = 0, then we may write
Z Z 1 
z
λ1 + λ2 z + λ4 e dz = λ1 + λ2 γ(t) + λ4 eγ(t) γ ′ (t) dt
γ 0
Z 1
dγ(t) λ2 dγ 2 (t) deγ(t)
= + λ1 + λ4 dt
0 dt 2 dt dt
h λ2 it=1
= λ1 γ(t) + γ 2 (t) + λ4 eγ(t)
2 t=0
q −p
2 2
= λ1 (q − p) + λ2 + λ4 (eq − ep ).
2
96 15. CONTOUR INTEGRALS

This approach fails when λ3 6= 0, because we cannot find a function of t whose

derivative is γ(t) γ ′ (t), or at least not easily. At this point, we do not know whether
such a function might exist, but later we will see that this is problematic.
Exercise 15.10. Suppose that Γ is the circle {z ∈ZC : |z| = r} with the standard
orientation, traversed k times, and n ∈ Z. Compute z n dz.
Γ

Answer. First, we parametrise Γ: define γ(t) = reit where 0 ≤ t ≤ 2kπ, so

′
γ (t) = ireit . Now
Z Z Z 2kπ Z 2kπ
n n n ′
z dz = z dz = γ(t) γ (t) dt = rn eint ireit dt
Γ γ 0 0
Z 2kπ h ei(n+1)t it=2kπ
= irn+1 ei(n+1)t dt = irn+1
0 i(n + 1) t=0
 e2kπi(n+1) − e0 
= irn+1 = 0,
i(n + 1)
unless n = −1, in which case
Z Z 2kπ Z 2kπ
−1 −1 −it
z dz = r e it
ire dt = i dt = 2kπi. 4
γ 0 0

We would get the same answer if the parameter t varied from −kπ to kπ.
Alternatively, we may write, when n 6= 0,
Z Z 2π Z 2π h i2π
n n ′ 1 dγ n+1 (t) 1−1
z dz = γ (t) γ (t) dt = dt = γ n+1 (t) = = 0.
γ 0 0 n+1 dt 0 n+1
When n = −1,
Z Z 2π ′ Z 2π h i2π
1 γ (t) d log γ(t)
dz = dt = dt = log γ(t) = 2πi,
γ z 0 γ(t) 0 dt 0

where we take a branch of log that is continuous along the curve.

(1) The M in the M L Lemma is a maximiser, not necessarily the maximum.
But it can be the maximum!
 LECTURE 16

The Cauchy–Goursat theorem

In this lecture, we begin with an exercise, and then state and discuss one of the
key theorems of complex analysis.

1. An exercise

Z 16.1. Suppose that Γ is a simple closedZ contour in C and c0 , c1 ∈ C.

Exercise
Show that (c1 z + c0 ) dz = 0. Is it always true that Re(z) dz = 0?
Γ Γ

Answer. We parametrise the contour Γ by a simple closed piecewise smooth

curve γ : [a, b] → C. By definition,
Z Z b
(c1 z + c0 ) dz = (c1 γ(t) + c0 ) γ ′ (t) dt
γ a
Z b
d γ 2 (t) d
= c1
+ c0 γ(t) dt
a dt 2 dt
h γ 2 (t) it=b
= c1 + c0 γ(t)
2 t=a
2
γ (b) γ 2 (a)
= c1 + c0 γ(b) − c1 − c0 γ(a)
2 2
= 0,
since γ is closed.
The second part of the question needs a counterexample. Take γ(t) = eit , where
0 ≤ t ≤ 2π. By definition,
Z Z Z 2π
Re(z) dz = x dz = cos(t) i eit dt
γ γ 0
Z 2π
=i cos2 (t) + i cos(t) sin(t) dt = iπ.
Z 0

The integral Re(z) dz can be nonzero! 4

Γ

2. Simply connected domains

First, we consider the case of a simply connected domain.
Theorem 16.2 (Cauchy–Goursat). Suppose that Ω is a simply connected domain,
that f ∈ H(Ω), and that Γ is a closed contour in Ω. Then
Z
f (z) dz = 0.
Γ
97
98 16. THE CAUCHY–GOURSAT THEOREM

Proof. We prove a more general version of this result later for the case in which
Γ is simple. To extend from the simple case to the general case, we argue as we did
for path integrals in Lecture 13. □

Corollary 16.3 (Independence of contour). Suppose that Ω is a simply connected

domain in C, that f ∈ H(Ω), and that Γ and ∆ are contours with the same initial
point p and the same final point q. Then
Z Z
f (z) dz = f (z) dz.
Γ ∆

Proof. The contour Γ t ∆∗ is closed, and so

Z Z Z
0= f (z) dz = f (z) dz − f (z) dz,
Γ⊔∆∗ Γ ∆
and the desired equality follows. □

Corollary 16.4 (Existence of primitives). Suppose that Ω is a simply connected

domain in C, and that f ∈ H(Ω). Then there exists a function F on Ω such that
Z
f (z) dz = F (q) − F (p)
Γ
for all contours Γ in Ω from p to q. Further, F is differentiable, and F ′ = f .
Finally, if F1 is any other function such that F1′ = f , then F1 − F is a constant and
Z
f (z) dz = F1 (q) − F1 (p),
Γ
where p and q are the initial and final points of Γ.
R
Proof. Fix a “base point” b in Ω, and for p ∈ Ω, define F (p) to be Γ f (z) dz,
where Γ is any contour in Ω with initial point b and final point p. This definition
makes sense in light of the previous corollary.
Given p and q in Ω, and a contour Γ from p to q, take contours Γp from b to p
and Γq from b to q. Then Γ t (Γq )∗ t Γp is a closed contour for which
Z
f (z) dz = 0.
Γ⊔(Γq )∗ ⊔Γp

Writing this as a combination of integrals, we see that

Z
f (z) dz = F (q) − F (p).
Γ

Now we take p ∈ Ω, and show that F ′ (p) = f (p). To do so, we need to make
F (q) − F (p)
− f (p)
q−p
small by taking q sufficiently close to p.
Take q ∈ Ω close to p, and let ∆ be the line segment from p to q. On the one
hand, Z
F (q) − F (p) 1
= f (z) dz;
q−p q−p ∆
 3. MULTIPLY CONNECTED DOMAINS 99

this is correct
R provided that ∆ is contained in Ω. On the other hand, by direct
calculation, ∆ dz = q − p, whence
Z
1
f (p) = f (p) dz.
q−p ∆
Thus
Z Z
F (q) − F (p) 1 1
− f (p) = f (z) dz − f (p) dz
q−p q−p ∆ q−p ∆
Z
1

= f (z) − f (p) dz .
q−p ∆
We can make this small by making ∆ small enough that ∆ is contained in Ω and f
does not vary much on ∆(f is differentiable and hence continuous), and then using
the M L Lemma.
More precisely, take any small positive ε. Since Ω is open and f is continuous at
p, there exists δ such that B(p, δ) ⊂ Ω and |f (z) − f (p)| < ε when z ∈ B(p, δ). Take
q ∈ B ◦ (p, δ) and let ∆ be the straight line segment from p to q. Then ∆ ⊂ B(p, δ)
and so |f (z) − f (p)| < ε for all z ∈ ∆. Thus
Z
F (q) − F (p) 1

− f (p) = f (z) − f (p) dz
q−p q−p ∆
1
≤ max{|f (z) − f (p)| : z ∈ ∆} |q − p|
|q − p|
= max{|f (z) − f (p)| : z ∈ ∆} < ε,
by the M L lemma and the fact that max{|f (z) − f (p)| : z ∈ ∆} = f (z ∗ ) for some
z ∗ ∈ ∆ (the maximum is attained). It follows that F is differentiable at p, with
derivative f (p), as required.
If F1 is another function such that F1′ = f , then (F1 − F )′ = 0, so F1 − F is a
constant, C say. This means that
F1 (q) − F1 (p) = (F (q) + C) − (F (p) + C) = F (q) − F (p),
R
so that F1 can also be used to compute Γ f (z) dz. □

We call a function F such that F ′ = f a primitive or an anti-derivative of f .

In some of our earlier computations, there are hints that it might be possible to
compute contour integrals using primitives; now we have the proof of this, at least
when f is holomorphic.

3. Multiply connected domains

Perhaps unfortunately, many domains are not simply connected. For these do-
mains, the discussion is more complicated. Suppose that Ω is a bounded domain
whose boundary ∂Ω consists of finitely many disjoint contours. Then one of the
contours, Γ0 say, is outside Ω while the others, Γ1 , …, Γn say, are inside Ω. We
orient these contours in the standard way. This is illustrated in Figure 16.1.
Exercise 16.5. Determine the orientations of the contours in Figure 16.1.
100 16. THE CAUCHY–GOURSAT THEOREM

Im

Γ2

Re
Γ0
Γ1

Figure 16.1. A multiply connected domain

Theorem 16.6 (Cauchy–Goursat). Suppose that Ω is a bounded domain whose

boundary ∂Ω consists of finitely many contours Γ0 , Γ1 , …, Γn . Suppose also that
f ∈ H(Υ), where Ω ⊂ Υ. Then
Z n Z
X
f (z) dz = f (z) dz = 0.
∂Ω j=0 Γj

Proof. We will see this next lecture. □

Corollary 16.7. Suppose that Υ is a simply connected domain, that Γ is a
simple closed contour in Υ, and that f is a differentiable function in Υ. Then
Z
f (z) dz = 0.
Γ

Proof. We let Ω be the interior of Γ and apply the previous result. □

Corollary 16.8. Suppose that Ω is a bounded domain whose boundary ∂Ω
consists of finitely many contours
R Γ0 , Γ1 , …, Γn , that Ω ⊂ Υ, and R that f is a
differentiable function in Υ. If Γj f (z) dz = 0 when j = 1, . . . , n, then Γ f (z) dz = 0
when Γ is a closed curve in Ω, and there is a differentiable function F in Ω such
that F ′ = f , and Z
f (z) dz = F (q) − F (p)
∆
whenever ∆ is a contour in Ω from p to q.
Proof. We omit this proof. □

4. History† (Not examinable)

The Cauchy–Goursat theorem is mathematically important because it will lead
to the Cauchy integral formula, one of the most useful formulae in complex analysis.
One of the uses of this formula is to compute integrals. Another is to show that a
holomorphic function in a domain Ω has continuous partial derivatives, and indeed
is infinitely differentiable. We stated several theorems earlier about holomorphic
 5. CONNECTION WITH VECTOR FIELDS† (NOT EXAMINABLE) 101

functions which include the hypotheses that f is holomorphic and that f ′ is con-
tinuous, and it is useful to know that the continuity hypothesis is automatically
true. At least in principle, we should check that the hypotheses of a theorem are
satisfied before we apply the theorem, and so it is good to make these hypotheses
unnecessary.
Augustin Cauchy was one of the finest French mathematicians of the first half of
the 1800s, and he developed much of what is in a course on complex analysis today, as
well as making precise the idea of limit that had been worrying mathematicians and
philosophers of mathematics since the time of Newton and Leibnitz. The Cauchy–
Goursat theorem is Goursat’s 1884 improvement of a theorem of Cauchy from 1829—
it says something of Goursat’s ability that he could improve the work of Cauchy.
Some of Cauchy’s ideas were being developed simultaneously by George Green,
an “uneducated miller” from Nottingham, who gave us Green’s theorem in 1828.

5. Connection with vector fields† (Not examinable)

Take a simple piecewise smooth curve γ : [a, b] → C, and write γ(t) = γ1 (t) +
iγ2 (t) and f (z) = u(x, y) + iv(x, y). Let γ(t) be the analogue of γ in R2 , that is,
γ(t) = (γ1 (t), γ2 (t)). Then
Z Z b
f (z) dz = (u(γ(t)) + iv(γ(t))) (γ1′ (t) + iγ2′ (t)) dt
γ a
Z b
= (u(γ(t)) γ1′ (t) − v(γ(t))) γ2′ (t)) dt
a
Z b
+i (u(γ(t)) γ2′ (t) + v(γ(t) γ1′ (t)) dt
a
Z b Z b
′
= U (γ(t)) · γ (t) dt + i V (γ(t)) · γ ′ (t) dt
Za Z a

= U (s) ds + i V (s) ds,

γ γ
where the vector fields U and V are given by
U (x, y) = (u(x, y), −v(x, y)) and V (x, y) = (v(x, y), u(x, y)).
We know that, in a simply connected domain, the line integrals depend only on
the initial point and the final point when the vector fields are closed. The condition
that U and V be closed is the condition
∂u ∂v ∂v ∂u
=− and = .
∂y ∂x ∂y ∂x
These are exactly the Cauchy–Riemann equations. So the complex line integrals
depend only on the initial point and the final point when the Cauchy–Riemann
equations hold.
 LECTURE 17

Cauchy’s integral formula

In this lecture, we begin with a sketch proof of the Cauchy–Goursat Theorem.

Then we state and prove Cauchy’s integral formula. Finally we see some applica-
tions.

1. Proof of the Cauchy–Goursat Theorem

First we restate the Cauchy–Goursat Theorem.

Theorem. Suppose that Ω is a bounded domain whose boundary ∂Ω consists of

finitely many contours Γ0 , Γ1 , …, Γn . Suppose also that f ∈ H(Υ), where Ω ⊂ Υ.
Then
XZ
f (z) dz = 0.
j Γj

Proof. The proof of the theorem involves three steps. First, we prove it in the
case where Ω is a triangle. Second, we consider the case where Ω is a domain whose
boundary is made up of finitely many closed polygonal contours. Third, we treat
the general case.
Step one. Suppose that Ω is a triangle in the complex plane. We write T0 for Ω
and ∂T0 for its boundary. Suppose that f ∈ H(Υ), where T0 ⊂ Υ, and let
Z
f (z) dz = I.
∂T0

We have to show that I = 0, and we suppose towards a contradiction that I 6= 0.

We may subdivide T0 into four congruent sub-triangles, T′ , T′′ , T′′′ and T′′′′ say, by
taking the midpoint of each side, and joining these midpoints; see Figure 17.1.

T′′′′
′′
T
T′
T′′′

Figure 17.1. Subdividing a triangle

103
104 17. CAUCHY’S INTEGRAL FORMULA

Now
Z
I= f (z) dz
Z ∂T0
Z Z Z
= f (z) dz + f (z) dz + f (z) dz + f (z) dz.
∂T′ ∂T′′ ∂T′′′ ∂T′′′′

At least one of the triangles T′ , T′′ , T′′′ and T′′′′ , which we call T1 , must satisfy
Z Z
1 |I|
f (z) dz ≥ f (z) dz = .
∂T1 4 ∂T0 4
We now subdivide T1 into 4 congruent triangles, and argue in the same way that
there must be one of these, T2 say, with the property that
Z Z
1 |I|
f (z) dz ≥ f (z) dz ≥ .
∂T2 4 ∂T1 16
Continuing inductively in this way, we find a sequence (Tn )n∈N of nested trian-
gles, such that
Z
|I|
f (z) dz ≥ n . (17.1)
∂Tn 4
Write Length(∂Tn ) for the perimeter of Tn . Then
Length(∂Tn ) = 2−n Length(∂T0 ).

By compactness, there is a point z0 that lies in each of the closed triangles Tn ,

and by hypothesis, f is differentiable at z0 . If z ∈ ∂Tn , then |z − z0 | is less than
half the perimeter of Tn , that is,
|z − z0 | ≤ 1
2
Length(∂Tn ) = 2−n−1 Length(∂T0 ),
and this tends to 0 as n → ∞.
Since f is differentiable at z0 , we may write
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + E(z),
where the error term E(z) satisfies
|E(z)|
→0 as z → z0 .
|z − z0 |
In particular, we can ensure that
|E(z)| |I|
≤ ∀z ∈ ∂Tn (17.2)
|z − z0 | Length(∂T0 )2
by taking n large enough. In what follows, we take such an n.
Recall that
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + E(z).
This means that
Z Z Z Z
′
f (z) dz = f (z0 ) dz + f (z0 )(z − z0 ) dz + E(z) dz.
∂Tn ∂Tn ∂Tn ∂Tn
 2. CAUCHY’S INTEGRAL FORMULA 105

The first two integrals on the right hand side are 0, by calculation, and hence by
(17.1), the last equation, the M L Lemma, properties of maxima, and (17.2),
Z
|I| ≤ 4 n
f (z) dz
Z ∂Tn

n
=4 E(z) dz
∂Tn
≤ 4 max{|E(z)| : z ∈ ∂Tn } Length(∂Tn )
n
 
|E(z)|
n
= 4 max |z − z0 | : z ∈ ∂Tn Length(∂Tn )
|z − z0 |
 
|E(z)|
≤ 4 max
n
: z ∈ ∂Tn max {|z − z0 | : z ∈ ∂Tn } Length(∂Tn )
|z − z0 |
|I| Length(∂Tn )2
≤ 4n
Length(∂T0 )2 2
|I|
= ,
2
which is absurd. Hence I = 0.
Step 2. The next step is to deal with a domain Ω with a polygonal boundary.
Any such domain may be subdivided into triangles Tn (although this may seem
obvious, it is hard to prove), in such a way that
Z XZ
f (z) dz = f (z) dz;
∂Ω n ∂Tn

by the result of the previous step,

Z
f (z) dz = 0.
∂Ω

Step 3. Finally, we have to deal with a domain whose boundary is the union of
finitely many disjoint closed contours. This can be done by approximating unions of
closed contours by unions of polygonal contours; the integral is 0 for all the unions of
approximating polygonal contours, and so the integral around the union of general
contours that we want is also 0. □

2. Cauchy’s integral formula

Theorem 17.1 (Cauchy’s integral formula). Suppose that Ω is a simply con-
nected domain in C, that f ∈ H(Ω), that Γ is a simple closed contour in Ω and that
w ∈ Int(Γ). Then
Z
1 f (z)
f (w) = dz. (17.3)
2πi Γ z − w

Proof. Let Γε be the circle with centre w and radius ε, traversed clockwise,
and take ε small enough that Γε ⊂ Int(Γ). We consider the domain Υ consisting
of Int(Γ) ∩ Ext(Γε ), the domain between Γ and Γε , whose boundary consists of
Γ, traversed anti-clockwise, and Γε , traversed clockwise. The quotient function
z 7→ f (z)/(z − w) is holomorphic in Ω \ {w}, a domain that contains Υ ∪ ∂Υ.
106 17. CAUCHY’S INTEGRAL FORMULA

By the Cauchy–Goursat theorem,

Z Z Z
f (z) f (z) f (z)
dz = dz + dz = 0;
∂Υ z − w Γ z −w Γε z − w

that is,
Z Z
f (z) f (z)
dz = dz.
Γ z−w Γ∗ε z−w
The left hand side of this equality does not depend on ε, so the limit as ε tends to
0 of the right hand side exists.
To compute the limit, we parametrise Γ∗ε . Define γε∗ (θ) = w + εeiθ , where 0 ≤
θ ≤ 2π, and observe that
Z Z
f (z) f (z)
dz = lim dz
Γ∗ε z − w ε→0 Γ∗ z − w
Z 2π
ε

f (w + εeiθ )
= lim iεeiθ dθ
ε→0 0 εeiθ
Z 2π
= lim f (w + εeiθ ) i dθ
ε→0 0
Z 2π
=i lim f (w + εeiθ ) dθ
0 ε→0
Z 2π
=i f (w) dθ
0
= 2πif (w).

We can move the limit inside the integral because limε→0 f (w + εeiθ ) = f (w) uni-
formly in θ, since limz→w f (z) = f (w). Formula (17.3) follows. □

3. Two corollaries of Cauchy’s integral formula

Corollary 17.2 (independence of contour). Suppose that w lies in a simply
connected domain Ω, and that f ∈ H(Ω). If Γ and ∆ are simple closed contours
such that w ∈ Int(Γ) and w ∈ Int(∆), then
Z Z
f (z) f (z)
dz = dz.
Γ z −w ∆ z −w

Proof. By Cauchy’s integral formula, both sides are equal to 2πif (w). □
Z
f (z)
This means that if we need to compute the integral dz, we may change
Γ z −w
the contour to make the calculation easier.
Corollary 17.3 (mean value formula). Suppose that Ω is a simply connected
domain in C, that f ∈ H(Ω), and that w ∈ Ω. If B(w, r) ⊂ Ω, then
Z 2π
1
f (w) = f (w + reiθ ) dθ. (17.4)
2π 0
 4. A COMPUTATION 107

Proof. This formula is virtually proved in the course of the proof of the Cauchy
integral formula; let γ(θ) = w + reiθ , where 0 ≤ θ ≤ 2π. Then
Z
1 f (z)
f (w) = dz
2πi γ z − w
Z 2π
1 f (w + reiθ )
= ireiθ dθ
2πi 0 reiθ
Z 2π
1
= f (w + reiθ ) dθ,
2π 0
as required. □
The Cauchy integral formula may be considered as a way to write f (w) as a
weighted average of the values of f (z) around any contour surrounding w.

4. A computation
Cauchy’s integral formula enables us to compute some integrals without inte-
grating!
Z
sin z
Exercise 17.4. Compute dz, where Γ is the circle with centre 0 and
Γ z
radius R.
Answer. Take f (z) = sin z and z0 = 0, and apply Cauchy’s integral formula:
Z Z
sin z f (z)
dz = dz = 2πif (z0 ) = 0. 4
Γ z Γ z − z0

(1) Precise statements of the Cauchy–Goursat theorem and of Cauchy’s integral

formula may be examined.
(2) The proofs of the results might be examined.
 LECTURE 18

Cauchy’s generalised integral formula

Cauchy’s integral formula,

Z
1 f (z)
f (w) = dz, (18.1)
2πi Γ z−w
where w lies inside a simple closed contour Γ in a simply connected domain Ω, and
f ∈ H(Ω), is perhaps the most important formula in complex analysis.
In this lecture, we establish various consequences of Cauchy’s integral formula.
These include both theoretical results and explicit computations.

1. Cauchy’s generalised integral formula

The most important theoretical application of Cauchy’s integral formula is the
following extension.

Corollary 18.1. Suppose that f ∈ H(B(z0 , R)), and that Γ is a simple closed
contour in B(z0 , R) such that z0 ∈ Int(Γ). Then
X
∞
f (w) = cn (w − z0 )n ∀w ∈ B(z0 , R),
n=0

where
Z
1 f (z)
cn = dz.
2πi Γ (z − z0 )n+1
The radius of convergence of the power series is at least R.

Remark 18.2. This corollary, combined with the fact that f (n) (z0 ) = n! cn ,
implies that
Z
(n) n! f (z)
f (z0 ) = dz.
2πi Γ (z − z0 )n+1
This is often called Cauchy’s generalised integral formula.
Notice that we just assumed that f is differentiable once; Corollary 18.1 implies
that f is actually infinitely differentiable.

Proof. Write Γr for the circle with centre z0 and radius r, where r < R. By
independence of contour,
Z Z
f (z) f (z)
dz = dz,
Γ (z − z0 ) Γr (z − z0 )
n+1 n+1

109
110 18. CAUCHY’S GENERALISED INTEGRAL FORMULA

so we may assume that Γ = Γr . By the Cauchy integral formula, if w ∈ B(z0 , r),

then
Z
1 f (z)
f (w) = dz
2πi Γr z − w
Z
1 f (z)
= dz
2πi Γr (z − z0 ) − (w − z0 )
Z
1 f (z) 1
= dz.
2πi Γr (z − z0 ) (1 − (w − z0 )/(z − z0 ))
Observe that |w − z0 | < |z − z0 | = r for all z ∈ Γr , so
1 X (w − z0 )n
∞
= ,
1 − (w − z0 )/(z − z0 ) n=0 (z − z0 )n
and, for fixed w, this series converges uniformly for z ∈ Γr . This means that
Z
f (z) X (w − z0 )n
∞
1
f (w) = dz
2πi Γr (z − z0 ) n=0 (z − z0 )n
∞ Z
1 X f (z) (w − z0 )n
= dz
2πi n=0 Γr (z − z0 ) (z − z0 )n
X
∞
= cn (w − z0 )n ,
n=0

where Z Z
1 f (z) 1 f (z)
cn = dz = dz,
2πi Γr (z − z0 )n+1 2πi Γ (z − z0 )n+1
by independence of contour. (Here we have exchanged the order of summation and
integration.) Once we know that f has this power series representation, it follows
(n)
that cn = f n!(z0 ) from results on power series in Lecture 9.
We chose r and w such that |w − z0 | < r < R. If we take any w ∈ B(z0 , R),
then we may choose r such that these inequalities hold, so the series converges at
w. Since w is an arbitrary element of B(z0 , R), the series converges in B(z0 , R), and
the radius of convergence is at least R. □

Here is an example of the use of Cauchy’s generalised integral formula.

Z
ez
Exercise 18.3. Compute n+1
dz, where Γ is the circle with centre 0 and
Γ z
radius 1.

Answer. Take f (z) = ez and z0 = 0, and apply Cauchy’s generalised integral

formula:
Z Z
ez f (z) 2πi (n) 2πi
dz = dz = f (0) = ,
Γ (z − z0 )
n+1 n+1
Γ z n! n!
since the derivative, and hence also the nth derivative, of the exponential function
is the exponential function itself, and e0 = 1. 4
 2. LIOUVILLE’S THEOREM AND THE FUNDAMENTAL THEOREM OF ALGEBRA 111

2. Liouville’s theorem and the fundamental theorem of algebra

Cauchy’s generalised integral formula has some very surprising consequences.
Here is one.

Theorem 18.4 (Liouville’s Theorem). Suppose that f is a bounded entire function.

Then f is constant.

Proof. Since f is bounded, we may choose a positive constant C such that

|f (z)| ≤ C for all z ∈ C.
Since f is entire, we may take ΓR to be the circle centre 0 and radius R, and use
Cauchy’s generalised integral formula to find the power series for f inside ΓR :
X
∞
f (n) (0)
f (z) = zn, (18.2)
n=0
n!

where
Z
(n) n! f (z)
f (0) = dz.
2πi ΓR z n+1
The power series (18.2) converges inside ΓR , so its radius of convergence ρ is at least
R; as R may be made arbitrarily large, ρ = ∞.
Further, when |z| = R,
f (z) C
n+1
≤ n+1 ,
z R
and so, by the M L Lemma,
Z
n! f (z) n! C n! C
f (n)
(0) = n+1
dz ≤ n+1
2πR = n .
2π ΓR z 2π R R
If n ≥ 1, then the left hand side of the formula above is 0, since R may be made
arbitrarily large, and so f (n) (0) = 0. Thus the power series (18.2) simplifies to show
that f (z) = f (0). □

Corollary 18.5 (Fundamental theorem of algebra). Suppose that f is a non-

constant complex polynomial. Then f has at least one root, and hence f may be
factorised as a product of a constant and finitely many linear factors.

Sketch proof. Suppose that f has no root. First, f (z) → ∞ as z → ∞,

and so there exists R such that |f (z)| ≥ 1 when |z| ≥ R. Next, in the compact
set B(0, R), the function |f | is continuous and takes positive values, so it has a
minimum value, m say, which cannot be 0 as f has no root. Thus |f (z)| ≥ m when
|z| ≤ R. It follows that the function 1/f is bounded and entire, so 1/f is constant,
and f is constant. Since f is not constant by hypothesis, f must have a root.
The complete factorisation of a polynomial follows by dividing out a factor of
z − r for each root r. If the quotient is a nonconstant polynomial, we can find
another root, and keep on dividing out and finding more roots until the quotient is
constant and we have a complete factorisation. □
112 18. CAUCHY’S GENERALISED INTEGRAL FORMULA

3. Behaviour of a holomorphic function near a zero

A zero of a function f that is holomorphic in an open set Ω is a point w ∈ Ω such
that f (w) = 0. For such a point w, Cauchy’s generalised integral formula implies
that
X
∞
f (z) = an (z − w)n ∀z ∈ B(w, r),
n=0
for some open ball B(w, r). Note that a0 = f (w) = 0. If all an are 0, then f (z) = 0
for all z ∈ B(w, r). Otherwise, we define N = min{n ∈ N : an 6= 0}; then an = 0
when n < N and an 6= 0.
One idea that we will use in several ways in this course is that f (z) behaves like
aN (z − w)N near to w. Here is one way in which this is true.
P
Proposition 18.6. Suppose that f (z) = ∞ n=0 an (z − w) for all z ∈ B(w, r),
n

and that an 6= 0 for some n ∈ N. Let N = min{n ∈ N : an 6= 0}. Then

f (z)
lim = 1.
z→w aN (z − w)N

Proof† . We shall show that, given ε ∈ R+ , there exists rε ∈ R+ such that

X
∞
an (z − w)n < ε aN (z − w)N ∀z ∈ B(w, rε ). (18.3)
n=N +1

Since an = 0 when n < N , it follows that

P∞ P∞
n=0 an (z − w) n=N +1 an (z − w)
n n
f (z)
= =1+ ,
aN (z − w)N aN (z − w)N aN (z − w)N
and the lemma follows.
◦
Now the argument involves P∞power series. nTake z0 ∈ B (w, r) such that |z0 − r| is
close to r. Since the series n=0 an (z0 − w) converges, there is a constant C such
that |an (z0 − w)n | ≤ C for all n ∈ N, so |an | ≤ C |z0 − w|−n . It follows that, when
|z − w| < |z0 − w|,
X∞ X |z − w|n |z − w|N |z − w|
|an | |z − w|n ≤ C = C .
n=N +1 n=N +1
|z 0 − w|
n
|z 0 − w| N
|z 0 − w| − |z − w|

By taking rε small enough, we can ensure that

1 rε
C < ε |aN | ,
|z0 − w| |z0 − w| − rε
N

and then when z ∈ B(w, rε ), it follows that

|z − w|N |z − w|
C < ε aN (z − w)N ,
|z0 − w| N
|z0 − w| − |z − w|
and then (18.3) holds. □
Later we will use this fact to prove l’Hôpital’s rule for analytic functions.
Corollary 18.7. Suppose that Ω is an open set, that f ∈ H(Ω), and that
f (w) = 0 for some w ∈ Ω. Then there exists r ∈ R+ such that either f (z) = 0 for
all z ∈ B(w, r) or f (z) 6= 0 for all z ∈ B ◦ (w, r).
 4. EXAMPLES 113
P
Proof† . Write f (z) = ∞n=0 an (z − w) for all z ∈ B(w, r), and suppose that
n

an 6= 0 for some n ∈ N. Then f (n) (w) 6= 0, and so f is not identically equal to 0

near w. Take N as in Proposition 18.6. Then there exists r ∈ R+ such that
f (z) 1
−1 < ∀z ∈ B ◦ (w, r),
aN (z − w) N 2
and for these z it follows that f (z) 6= 0. □

In summary, if a holomorphic function f is not constant, then the zeroes of f

are isolated.

4. Examples
Z 2π
4
Exercise 18.8. Compute dθ.
0 5 + 3 cos(θ)

Answer. First we “reverse engineer” this into an integral that arises from inte-
grating around a parametrised circle: take γ(θ) = eiθ , where 0 ≤ θ ≤ 2π.
Z 2π Z 2π
4 8
dθ = dθ
0 5 + 3 cos(θ) 0 10 + 3e + 3e−iθ
iθ
Z 2π
8eiθ
= dθ
0 3(eiθ )2 + 10eiθ + 3
Z
1 2π 8
= ieiθ dθ
i 0 3(e ) + 10eiθ + 3
iθ 2
Z
1 2π 8
= 2
γ ′ (θ) dθ
i 0 3(γ(θ)) + 10γ(θ) + 3
Z
1 8
= 2
dz.
i γ 3z + 10z + 3

Now we use partial fractions. Clearly 3z 2 + 10z + 3 = (3z + 1)(z + 3), and it is easy
to check that
8 3 1
2
= − .
3z + 10z + 3 3z + 1 z + 3
Thus
Z 2π Z
4 1 3 1
dθ = − dz
0 5 + 3 cos(θ) i γ 3z + 1 z + 3
Z Z
1 1 1 1
= dz − dz
i γ z + 1/3 i γ z+3
Z
1 1
= dz,
i γ z + 1/3

because z 7→ 1/(z +3) is holomorphic in B(0, 3) and Int(γ) ⊂ B(0, 3), so the second
integral is zero by the Cauchy–Goursat theorem. Finally, we apply Cauchy’s integral
formula, where f (z) is identically 1 and w = −1/3, which lies inside the curve γ. It
114 18. CAUCHY’S GENERALISED INTEGRAL FORMULA

follows that
Z 2π
4 2πi
dθ = f (−1/3)
0 5 + 3 cos(θ) i
= 2π. 4
This integral may also be computed using the substitution t = tan(θ/2).
 LECTURE 19

Morera’s Theorem and analytic continuation

In this lecture, we prove Morera’s theorem, which completes a logical circle

relating differentiability and independence of contour for integrals. Then we discuss
analytic continuation for holomorphic functions. Finally, we compute more examples
of contour integrals.

1. Morera’s theorem
Theorem 19.1 (Morera’s theorem). Suppose that Ω is a domain, that the function
f : Ω → C is continuous, and that
Z
f (z) dz = 0,
Γ
whenever Γ is a closed contour in Ω. Then f is holomorphic in Ω.

Proof. This proof is in two steps.

R
Step 1. We fix a base point b in Ω, and for all w ∈ Ω, define F (w) to be Γ f (z) dz,
where Γ is a contour from b to w. Then F ′ (w) = f (w) for all w ∈ Ω, and hence F
is holomorphic. The details are in the proof of the corollary of the Cauchy–Goursat
theorem on existence of primitives.
Step 2. We deduce that f is holomorphic from the fact that F is holomorphic. To
do this, we appeal to the corollary to Cauchy’s integral formula from the preceding
lecture. We take an arbitrary point z0 ∈ Ω, and then B(z0 , r) ⊆ Ω for some r ∈ R+
because Ω is open. Now F ∈ H(B(z0 , r)), and so by the corollary,
X
∞
F (z) = cn (z − z0 )n
n=0

in B(z0 , r). We may differentiate a power series term by term in any ball in which
it converges, so
X
∞ X
∞
f (z) = F ′ (z) = cn n(z − z0 )n−1 = cm+1 (m + 1)(z − z0 )m
n=0 m=0

in B(z0 , r), and hence f is holomorphic in B(z0 , r). Now f is holomorphic in Ω

because z0 was an arbitrary point in Ω. □
Corollary 19.2. Suppose that Λ is a (possibly infinite) line segment in an open
set Ω and Ω\Λ is open. If function f : Ω → C is continuous in Ω and is holomorphic
in Ω \ Λ, then f is holomorphic in Ω.
Sketch proof. R By Morera’s theorem, and an approximation argument, it suf-
fices to show that Γ f (z) dz = 0 for all closed polygonal contours Γ in Ω. We can
115
116 19. MORERA’S THEOREM AND ANALYTIC CONTINUATION

break such an integral into a sum of integrals over closed contours in Ω \ Λ together
with an error term that may be made arbitrarily small. □

2. The logic of the theorems on contour integration

We have seen a number of results about contour integration, involving various
hypotheses and conclusions, and we now summarise these results. We suppose that
f is a continuous function defined in a simply connected domain Ω, and Γ denotes
a contour in Ω. Consider the following conditions:

(a) f is holomorphic in Ω
Z
(b) f (z) dz = 0 for every closed Γ
Γ
Z
′
(b ) f (z) dz depends only on the start and end of Γ
Γ
Z
′′
(b ) there is a function F in Ω such that f (z) dz = F (q) − F (p), where Γ goes
Γ
from p to q, and F ′ = f
Z
1 f (z)
(c) f (w) = dz for any simple closed Γ such that w ∈ Int(Γ)
2πi Γ z − w
X
∞
(d) f (z) = cn (z − z0 )n in any open ball B(z0 , r) contained in Ω.
n=0
We have seen that:
(a) =⇒ (b) (the Cauchy–Goursat theorem)
′
(b) =⇒ (b ) (independence of contour)
(b′ ) =⇒ (b′′ ) (existence of primitives)
(a) =⇒ (c) (Cauchy’s integral formula)
(c) =⇒ (d) (corollary to Cauchy’s integral formula)
(d) =⇒ (a) (power series)
′
(b ) =⇒ (a) (Morera’s theorem)
′′
(b ) =⇒ (a) (proof of Morera’s theorem)

What is important in all this is that:

(a) for holomorphic functions in simply connected domains, integrals depend only
on the initial and final points of contours,
(b) to integrate a holomorphic function f along a contour from p to q, we may
find a primitive F and compute F (q) − F (p)
(c) integrals of functions around closed contours where the function is not holo-
morphic at finitely many points inside the contour may be calculated in terms
of the values of the function or its derivatives at these points.
 4. FURTHER EXAMPLES OF CONTOUR INTEGRALS 117

3. Analytic continuation
We have already seen that the values of a holomorphic function f on an interval
determine f in an open ball. The key to this is showing that if f is 0 on an interval,
then f is 0 on the ball. Now we see that the values of a holomorphic function in an
open subset of a domain determine the values in the whole domain.

Lemma 19.3. Suppose that B(z1 , r1 ) ⊂ B(w, R), that f ∈ H(B(w, R)), and that
f (z) = 0 for all z ∈ B(z1 , r1 ). Then f (z) = 0 for all z ∈ B(w, R).

Proof. We can find a finite sequence of open balls, B(z1 , r1 ), B(z2 , r2 ), …,

B(zJ , rJ ), say, with the properties that

B(z1 , r1 ) ⊂ B(z2 , r2 ) ⊂ · · · ⊂ B(zJ , rJ ) = B(w, R),

and the centre of B(zj+1 , rj+1 ) is contained in B(zj , rj ). We will show that f (z) = 0
for all z ∈ B(zJ , rJ ) by induction.
By hypothesis, f (z) = 0 for all z ∈ B(z1 , r1 ).
Suppose that f (z) = 0 for all z ∈ B(zj , rj ). Then f (z) = 0 for all z near the
centre of B(zj+1 , rj+1 ). Hence f (n) (zj+1 ) = 0 for all n ∈ N. We may expand f in a
power series in B(zj+1 , rj+1 ), and the coefficients in the power series are multiples
of the derivatives f (n) (zj+1 ). It follows that f (z) = 0 for all z ∈ B(zj+1 , rj+1 ).
By induction, f (z) = 0 for all z ∈ B(zJ , rJ ). □

Theorem 19.4. Suppose that Υ is a nonempty open subset of a domain Ω in C,

and that f ∈ H(Ω). If f (z) = 0 for all z in Υ, then f (z) = 0 for all z in Ω.

Proof† . Take a base point b ∈ Υ. Since Ω is connected, any point w ∈ Ω

may be joined to b by a polygonal contour Γ in Ω. Each point z on Γ is in the
open set Ω, and so there is a ball B(z, r) with centre z that is contained in Ω.
Because Γ is compact, there are finitely many of these balls, B(zj , rj ) say, such
that B(zj , rj ) ∩ B(zj+1 , rj+1 ) is not empty, b ∈ B(z1 , r1 ) and w ∈ B(zJ , rJ ). Since
f (z) = 0 for all z near b, the lemma implies that f (z) = 0 for all z ∈ B(z1 , r1 ).
Now f (z) = 0 for all z in an open subset of B(z2 , r2 ), and the lemma implies that
f (z) = 0 for all z ∈ B(z2 , r2 ). Continuing inductively, f (z) = 0 for all z ∈ B(zJ , rJ )
and hence f (w) = 0. □

Corollary 19.5. Suppose that Υ is a nonempty open subset of a domain Ω in

C, and that f, g ∈ H(Ω). If f (z) = g(z) for all z in Υ, then f (z) = g(z) for all z
in Ω.

Proof. Apply the previous result to f − g. □

4. Further examples of contour integrals

Z 2π
cos θ
Exercise 19.6. Compute dθ.
0 5 + 3 cos θ
118 19. MORERA’S THEOREM AND ANALYTIC CONTINUATION

Answer. Then
Z 2π Z 2π
cos θ 2 cos θ
dθ = dθ
0 5 + 3 cos θ 0 10 + 6 cos θ
Z 2π
eiθ + e−iθ
= dθ
0 10 + 3(eiθ + e−iθ )
Z 2π
e2iθ + 1
= dθ
0 3e2iθ + 10eiθ + 3
Z
1 2π e2iθ + 1
= ieiθ dθ.
i 0 eiθ (3e2iθ + 10eiθ + 3)
Let γ(θ) = eiθ , where 0 ≤ θ ≤ 2π. Then
Z 2π Z
cos θ 1 z2 + 1
dθ = dz
0 5 + 3 cos θ i γ z(3z 2 + 10z + 3)
Z
1 1 5 5
= − + dz
i γ 3z 4(3z + 1) 12(z + 3)
Z Z Z
1 1 5 1 5 1
= dz − dz + dz
3i γ z 12i γ z + 1/3 12i γ z + 3
2πi 10πi
= − +0
3i 12i
−π
= ,
6
by the Cauchy integral formula for the first two integrals and the Cauchy–Goursat
theorem for the last integral. 4
Z 2π
sin θ
Exercise 19.7. Compute dθ.
0 cos θ
Answer. This is a trick question: the integral is not defined. 4
In all the following exercises, Γ is the circle of radius 3 and centre 0, traversed
in the usual anticlockwise direction, and f ∈ H(B(0, π)).
Z
f (z)
Exercise 19.8. Compute dz.
Γ z −1
2

Answer. We may use partial fractions:

1 1 (1/2) (−1/2)
= = + ,
z2 − 1 (z − 1)(z + 1) z−1 z+1
so
Z Z Z
f (z) 1 f (z) 1 f (z)
dz = dz − dz
Γ z −1
2 2 Γ z−1 2 Γ z+1
2πi 2πi
= f (1) − f (−1)
2 2
= πi(f (1) − f (−1)),
by the Cauchy integral formula. 4
 5. REMARKS 119
Z
f (z)
Exercise 19.9. Compute dz.
Γ (z − 1)2
Answer. By the Cauchy integral formula for higher derivatives,
Z
f (z)
dz = 2πi[f ′ (1)]. 4
Γ (z − 1)
2

Z
f (z)
Exercise 19.10. Compute dz.
Γ (z 2 − 1)2
Answer. We may use partial fractions again:
1 A B C D
= + + + ,
(z − 1)2 (z + 1)2 (z − 1)2 z − 1 (z + 1)2 z + 1
from which it follows that
A(z + 1)2 + B(z − 1)(z + 1)2 + C(z − 1)2 + D(z − 1)2 (z + 1) = 1.
Looking at the coefficients of z 3 tells us that B + D = 0. Putting z = 1 and then
putting z = −1 tells us first that 4A = 1 and then that 4C = 1. Putting z = 0 tells
us that A − B + C + D = 1. It follows that
1
A = −B = C = D = .
4
We conclude that
Z Z Z Z Z
f (z) 1h f (z) f (z) f (z) f (z) i
dz = dz − dz + dz + dz
Γ (z − 1) 4 Γ (z − 1)2 Γ z −1
2 2 2
Γ (z + 1) Γ z +1
2πi  ′ 
= f (1) − f (1) + f ′ (−1) + f (−1)
4
πi  ′ 
= f (1) − f (1) + f ′ (−1) + f (−1) ,
2
by Cauchy’s (generalised) integral formula. 4

5. Remarks
The exercises that we have done suggest that we will be able to compute “arbi-
trary” integrals around closed contours of functions of the form f (z)/p(z), where p
is a polynomial. To do this, we will need to be able to factorise p, and then expand
1/p into partial fractions, that is, a sum of terms of the form 1/(z − α)k . We will
need to be able to compute with partial fractions!
 LECTURE 20

Taylor series

We recall some facts about power series, Cauchy’s integral formula, and Taylor
series. We discuss computation with and manipulation of Taylor series.

1. Power series, Taylor series, and Maclaurin series

Definition 20.1. A power series (with centre z0 ) is an expression of the form
X∞
an (z − z0 )n ,
n=0

where the coefficients an , the centre z0 , and the variable z are complex. A Taylor
series (with centre z0 ) for a function f is a series of the form
X∞
f (n) (z0 )
(z − z0 )n .
n=0
n!
A Maclaurin series for a function f is a Taylor series for f with centre 0, that is, a
series of the form
X∞
f (n) (0) n
z .
n=0
n!

Maclaurin series and Taylor series are particular examples of power series.
Recall from Lecture 10 that a power series has a radius of convergence, ρ, which
can often be found using the ratio test or the root test. The power series converges
inside B(z0 , ρ) and fails to converge outside B(z0 , ρ); if a power series converges in
B(z0 , r), then r ≤ ρ, but it is possible that r < ρ.
We say that (or write)
X∞
f (z) = an (z − z0 )n in B(z0 , r)
n=0

to mean that the domain of f includes B(z0 , r), the power series converges in B(z0 , r)
and that f (z) is the sum of the power series for each z ∈ B(z0 , r). If this holds, then
f is holomorphic in B(z0 , r), and moreover
f (n) (z0 )
an = , (20.1)
n!
that is, this power series is the Taylor series for f with centre z0 . Conversely, from
the lecture on Cauchy’s generalised integral formula, if a function
P∞ f is holomorphic
in B(z0 , r), then it can be represented as a power series n=0 an (z − z0 )n in the
same ball.
When we ask how large a ball with centre z0 in which a function f is represented
by a power series can be, we often find that the maximum value of the radius r
121
122 20. TAYLOR SERIES

of the ball is equal to the distance of z0 from the set of points where f fails to be
holomorphic.
Exercise 20.2. Show that the function f defined by f (z) = 1/z can be rep-
resented as a power series in a ball B(z0 , r), where z0 6= 0. Find the radius of
convergence of this power series.

Answer.
1 1 1
f (z) = = =
z z0 + z − z0 z0 (1 + (z − z0 )/z0 )
1 X
∞
(z − z0 )n
= (−1)n
z0 n=0 z0n
X
∞
(−1)n
= (z − z0 )n .
n=0
z0n+1
By the ratio test, the radius of convergence of this series is |z0 |. 4

The function f fails to be holomorphic at 0, which is on the edge of the ball

B(z0 , |z0 |). We say that “the singularity of f at 0 is a barrier to the convergence of
the power series in any larger ball with centre z0 ”.

2. The algebra and calculus of power series

To determine a Taylor series, we need to be able to find all the derivatives of a
function. There are only a few basic examples, and variations on these, for which
finding all the derivatives is possible. Geometric series and exponential series are
particularly important. Putting together these basic series to find more complicated
series is our next topic.
First, we can add and multiply power series.
Theorem 20.3. Suppose that c ∈ C, and that
X
∞ X
∞
f (z) = an (z − z0 )n and g(z) = bn (z − z0 )n in B(z0 , r).
n=0 n=0

Then
X
∞
(a) c f (z) = c an (z − z0 )n in B(z0 , r);
n=0
X
∞
(b) (f + g)(z) = (an + bn )(z − z0 )n in B(z0 , r);
n=0
X
∞
Pn
(c) (f g)(z) = cn (z − z0 )n in B(z0 , r), where cn = j=0 aj bn−j .
n=0

Proof. First, f, g ∈ H(B(z0 , r)), and hence c f, f + g, f g ∈ H(B(z0 , r)) from

properties of holomorphic functions. Thus these new functions may also be repre-
sented by power series in B(z0 , r). To determine the coefficients, we observe that
(c f )(n) (z0 ) = (c f )(n) (z0 ), (f + g)(n) (z0 ) = f (n) (z0 ) + g (n) (z0 ),
 3. EXAMPLES 123

and
n  
X
(n) n
(f g) (z0 ) = f (k) (z0 ) g (n−k) (z0 ),
k=0
k
and the theorem follows from this and (20.1). □

We can differentiate and integrate power series term by term.

X
∞
Theorem 20.4. Suppose that f (z) = an (z − z0 )n in B(z0 , r), where r > 0.
n=0
Then f is differentiable in B(z0 , r), and
X
∞ X
∞
f ′ (z) = nan (z − z0 )n−1 = (m + 1)am+1 (z − z0 )m in B(z0 , r).
n=0 m=0

Further, let
(
c when m = 0
bm = am−1
when m ≥ 1.
m
X
∞
Then bn (z − z0 )n converges in B(z0 , r), to a function F say, such that F ′ = f
n=0
and F (z0 ) = c.

Proof. We omit this proof. □

It is sometimes possible to divide power series, since the quotient of holomorphic

functions is homomorphic, at least when the denominator does not vanish, and
holomorphic functions may be represented as power series. It is also sometimes
possible to compose power series, since the composition of holomorphic functions is
homomorphic. Rather than try to state theorems about these operations, we will
look at some examples.

3. Examples
Exercise 20.5. Consider the function Log. Determine its Taylor series with
centre i − 1. What is the radius of convergence ρ of this series. Does the series
represent Log in the ball B(i − 1, ρ)?

Answer. We compute the derivatives of Log: as long as z is not a nonpositive

real number, then
1 −1
Log(0) (z) = Log(z), Log(1) (z) = , Log(2) (z) = ,
z z2
2 −6 24
Log(3) (z) = , Log(4) (z) = 4 , Log(5) (z) = 5 ,
z3 z z
and we may guess that
(−1)n−1 (n − 1)!
(n)
Log (z) = ,
zn
when n > 0. It is easy to verify this guess using induction.
124 20. TAYLOR SERIES

We deduce that the Taylor series for Log around z0 is

X
∞
(−1)n−1 (n − 1)!
Log(z) = Log(z0 ) + (z − z0 )n
n=1
n! z0n
X
∞
(−1)n−1
= Log(z0 ) + (z − z0 )n ,
n=1
nz0n

and hence, when z0 = i − 1, we get

X∞
(−1)n−1
Log(z) = Log(i − 1) + (z + 1 − i)n
n=1
n(i − 1) n

ln 2 3 X∞
(−1)n−1
= + iπ + (z + 1 − i)n .
2 4 n=1
n(i − 1) n

Alternatively, taking z0 = 1 − i in the first exercise, we see that

1 X
∞
(z + 1 − i)n
= (−1)n ,
z n=0
(i − 1) n+1

√
in the ball B(i − 1, 2).
We look for a function F such that F ′ (z) = 1/z in this same ball. According to
the previous theorem, if
X
∞
(z + 1 − i)n+1 X∞
m−1 (z + 1 − i)
m
n
F (z) = (−1) = (−1) ,
n=0
(n + 1)(i − 1) n+1
m=1
m(i − 1) m

√
then F ′ (z) = 1/z. In the connected set {z ∈ C : |z + 1 − i| < 2, Im(z) > 0}, it is
also true that Log′ (z) = 1/z, and hence (Log −F )′ (z) = 0 and Log(z) − F (z) is a
constant. By considering this expression at i − 1, we see that
√ 3
Log(z) − F (z) = Log(i − 1) − 0 = ln( 2) + iπ,
4
and hence
X∞
(z + 1 − i)m ln 2 3 X∞
Log(z) = (−1)m+1 + + iπ = am (z + 1 − i)m ,
m=1
m(i − 1) m 2 4 m=0

ln 2 3 (z + 1 − i)m
where a0 = + iπ and am = (−1)m+1 when m ≥ 1. From the ratio
2 4 m(i − 1)m √
test, we see that the radius of convergence of the series is 2. The Taylor series
represents Log above the real axis, but does not represent Log on the other side of
the branch cut along the negative real axis. However, it does represent the branch
of log where the imaginary parts of the values lie in [0, 2π). 4

sin(z)
Exercise 20.6. Define the function f by f (z) = if z 6= 0 and f (0) = 1.
z
Does this function have a Maclaurin series? If so, then what is its radius of conver-
gence?
 3. EXAMPLES 125

Answer. By definition, for all z ∈ C,

z3 z5 X (−1)n ∞
sin(z) = z − + − ··· = z 2n+1 .
3! 5! n=0
(2n + 1)!

In particular the series converges for all z ∈ C. Thus, when z 6= 0,

sin(z) z2 z4 X (−1)n ∞
=1− + − ··· = z 2n .
z 3! 5! n=0
(2n + 1)!

When z = 0, the power series on the right hand side of this last expression is equal
to 1. It follows that

z2 z4 X (−1)n
∞
f (z) = 1 − + − ··· = z 2n ∀z ∈ C.
3! 5! n=0
(2n + 1)!

This power series converges for all z in C. Thus the radius of convergence of the
Maclaurin series for f (z) is infinite. 4

We note that it follows that f (2n) (0) = (−1)n /(2n + 1) and f (2n+1) (0) = 0 for
all n ∈ N. It would also be possible to compute the derivatives of the function
z 7→ sin(z)/z “by hand”, and to find a formula for these using induction, but this
would be much longer.
When we discuss singularities in more detail, we will say that “the function
z 7→ sin(z)/z has a removable singularity at 0”.
z
Exercise 20.7. Define the function g : C → C by g(z) = when z 6= 0
sin(z)
and g(0) = 1. Does this function have a Maclaurin series? If so, then what is its
radius of convergence?

Answer. The function g is 1/f , where f is as in the previous exercise. Since

f is entire, g is holomorphic in C \ {±π, ±2π, ±3π, . . . }, that is, where f does not
vanish. So g has a Maclaurin series in B(0, π).
To determine the series, we mayP argue 2n in several ways. Note that f is even, so
g is even. Assuming that g(z) = ∞ n=0 b 2n z , we may write
 
z2 z4 z6

1− + − + . . . b0 + b2 z 2 + b4 z 4 + b6 z 6 + . . . = 1,
3! 5! 7!

whence, equating the coefficient of 1 to 0, and the coefficients of z 2 , z 4 , and so on,

to 0, we get
1 7 31
b0 = 1, b2 = , b4 = , b6 = ,
6 360 15120
and we could continue if we wished. In practice, this kind of computation would be
done using a computer. 4
126 20. TAYLOR SERIES

4. Compositions of power series

We begin with an easy exercise.
2
Exercise 20.8. Find the Maclaurin series for ez .
P P
Answer. We know that ew = ∞ wn
n=0 n! , and so e
z2
= ∞ z 2n
n=0 n! . 4
P∞ P
More generally, if f (z) = m=0 am (z − z0 )m and g(w) = ∞ n=0 bn (w − w0 ) , then
n
!n
X
∞ X∞
g(f (z)) = bn am (z − z0 )m − w0 ;
n=0 m=0
at least in principle we can expand the inner power and then gather terms. Usually
this is only done if a0 = w0 ; then the first nonzero term of the expanded series is
[a1 (z − z0 )]n , and only finitely many terms are involved in the coefficient of each
of the powers (z − z0 )k . Using this, it is possible to determine the power series
corresponding to the inverse function of a holomorphic function.
Theorem† 20.9 (Lagrange inversion theorem). Suppose that f ∈ H(Ω), and that
f (a) = b and f ′ (a) 6= 0. Then there is a holomorphic function g, defined in an open
set Υ that contains b, such that g ◦ f (z) P = z for all z near to a, and f ◦ g(w) = w
for all w near to b. Further, g(w) = a + ∞ n=1 cn (w − b) , where
n
  n
1 dn−1 z−a
cn = lim
n! z→a dz n−1 f (z) − f (a)
when n ≥ 1.
Proof. Omitted. □
Exercise 20.10. Define the function f by f (z) = e1/z when z 6= 0. Find a series
that represents f in B(1, 1).
Answer. The function is holomorphic in B(1, 1), and so there is a convergent
power series in powers of (z − 1) that represents the function in the ball. To find
the coefficients, we have to compute the derivatives of f : this is tricky.
But we can also observe that
X∞
wn
w
e =
n=0
n!
whenever w 6= 0, and so
X
∞
z −n
1/z
e = .
n=0
n!
This series certainly answers the question! 4
 LECTURE 21

Laurent series

In this lecture, we prove Laurent’s theorem about representing holomorphic func-

tions in an annulus by a series, which we now call a Laurent series, and we present
some examples of these series.
An annulus is a set of the form {z ∈ C : R1 < |z − z0 | < R2 }; we allow R1 to be
0 or R2 to be ∞. We also call the set {z ∈ C : 0 < |z − z0 | < R} a punctured ball,
and we denote it by B ◦ (z0 , R).

1. Laurent’s theorem
Theorem 21.1. Suppose that A is the annulus {z ∈ C : R1 < |z − z0 | < R2 } and
R1 < r < R2 . If f ∈ H(A), then
X
∞
f (w) = cn (w − z0 )n ∀w ∈ A,
n=−∞

where Z
1 f (z)
cn = dz.
2πi ∂B(z0 ,r) (z − z0 )n+1

Proof. First, take r1 and r2 such that R1 < r1 < r2 < R2 , and let Ω1 be the
domain B(z0 , r2 ) \ B(z0 , r1 ), as shown in Figure 21.1.
R
Suppose that g ∈ H(A). By the Cauchy–Goursat theorem, ∂Ω1 g(z) dz = 0. It
follows that Z Z
g(z) dz = g(z) dz.
∂B(z0 ,r1 ) ∂B(z0 ,r2 )

R2 r2

R1 b z0 r1

A Ω1

Figure 21.1. The annulus A and the first region Ω1

127
128 21. LAURENT SERIES

b
w

Ω2

Figure 21.2. The second region Ω2

In particular, by taking g(z) = f (z)/(z − z0 )n , where f ∈ H(A), we see that the

expression defining cn does not depend on r.
Now suppose that f ∈ H(A) and w ∈ A. Choose r1 and r2 such that
R1 < r1 < |w − z0 | < r2 < R2 ,
and take a small ball B(w, r) whose closure lies inside B(z0 , r2 ) \ B(z0 , r1 ), as shown
in Figure 21.2. Let Ω2 = Ω1 \ B(w, r).
By the Cauchy–Goursat theorem,
Z
1 f (z)
dz = 0.
2πi ∂Ω2 z − w
Breaking the integral over the boundary of Ω2 up into its three constituent integrals,
we see that
Z Z Z
1 f (z) 1 f (z) 1 f (z)
dz = dz − dz.
2πi ∂B(w,r) z − w 2πi ∂B(z0 ,r2 ) z − w 2πi ∂B(z0 ,r1 ) z − w
From the Cauchy integral formula,
Z Z
1 f (z) 1 f (z)
f (w) = dz − dz.
2πi ∂B(z0 ,r2 ) z − w 2πi ∂B(z0 ,r1 ) z − w

If z lies on ∂B(z0 , r2 ), then |z − z0 | = r2 and |w − z0 | < r2 , so

Z Z
f (z) f (z)
dz = dz
∂B(z0 ,r2 ) z − w ∂B(z0 ,r2 ) (z − z0 ) − (w − z0 )
Z
f (z)
= dz
∂B(z0 ,r2 ) (z − z0 )(1 − (w − z0 )/(z − z0 ))
Z
f (z) X (w − z0 )n
∞
= dz
∂B(z0 ,r2 ) (z − z0 ) n=0 (z − z0 )
n

X∞ Z
f (z)
= (w − z0 ) n
dz.
∂B(z0 ,r2 ) (z − z0 )
n+1
n=0
 2. FINDING LAURENT SERIES 129

Similarly, if z lies on ∂B(z0 , r1 ), then |z − z0 | = r1 and |w − z0 | < r1 , so

Z Z
f (z) f (z)
− dz = dz
∂B(z0 ,r1 ) z − w ∂B(z0 ,r1 ) (w − z0 ) − (z − z0 )
Z
f (z)
= dz
∂B(z0 ,r1 ) (w − z0 )(1 − (z − z0 )/(w − z0 ))
Z
f (z) X (z − z0 )m
∞
= dz
∂B(z0 ,r1 ) (w − z0 ) m=0 (w − z0 )
m

X∞ Z
1
= f (z) (z − z0 )m dz
m=0
(w − z 0 ) m+1
∂B(z0 ,r1 )

X−1 Z
f (z)
= (w − z0 )n dz
∂B(z0 ,r1 ) (z − z0 )
n+1
n=−∞

(at the last step, we substituted n = −m − 1). It follows that

X
∞
f (w) = cn (w − z0 )n ∀w ∈ A,
n=−∞

where
 Z


1 f (z)
if n ≥ 0
 dz
2πi Z∂B(z0 ,r2 ) (z − z0 )n+1
cn =

 1 f (z)
 dz if n ≤ −1.
2πi ∂B(z0 ,r1 ) (z − z0 )n+1

To conclude, we recall that the integrals defining cn around ∂B(z0 , r) do not depend
on r in the range (R1 , R2 ). □

A series in powers of (z−z0 ) that converges in an annulus with centre z0 is called a

Laurent series. We do not always find Laurent series by computing integrals; rather,
often we compute integrals by finding Laurent series.

2. Finding Laurent series

Exercise 21.2. Suppose that

1 1 −1 1/2 1/2
f (z) = = = + + .
z(z 2 − 1) z(z − 1)(z + 1) z z−1 z+1

Find Laurent series for f in the largest annuli with centre 0 in which f is holomor-
phic.

Answer. The function f has singularities at 0, ±1. The largest annuli with
centre 0 in which f is holomorphic are {z ∈ C : 0 < |z| < 1} and {z ∈ C : 1 < |z|}.
130 21. LAURENT SERIES

In the annulus {z ∈ C : 0 < |z| < 1}, also known as the punctured ball B ◦ (0, 1),
we may write
−1 1/2 1/2
f (z) = − +
z 1−z 1+z
−1 1  1 
= − 1 + z + z2 + z3 + z4 + . . . + 1 − z + z2 − z3 + z4 − . . .
z 2 2
−1  
= − z + z3 + z5 + . . .
z  
= (−1) z −1 + z + z 3 + z 5 + . . .

In the annulus {z ∈ C : 1 < |z| < ∞}, we may write

−1 1 1 1 1
f (z) = + · + ·
z 2z 1 − 1/z 2z 1 + 1/z
−1 1 1 1 1 1 
= + 1 + + 2 + 3 + 4 + ...
z 2z z z z z
1 1 1 1 1 
+ 1 − + 2 − 3 + 4 − ...
2z z z z z
−1 1  1 1 1 
= + 1 + 2 + 4 + 6 + ...
z z z z z

1 1 1 1 
= + + + ...
z z2 z4 z6
1 1 1 
= 3 + 5 + 7 + ... . 4
z z z

Remarks. This form of presentation is legitimate as long as there are sufficiently

many terms that the pattern is clear.
Note that both series have only odd powers, reflecting the symmetry of f ; indeed,
z n is an odd or even function as n is an odd or even integer. Note also that the
radius of convergence of the first power series is 1; this corresponds to the fact that
f is holomorphic in B ◦ (0, 1) but not in any larger punctured ball with centre 0.
The second series converges when |z| > 1; indeed, it is a geometric series with ratio
|1/z 2 |.
Note also that the first Laurent series has only one negative power of z, while
the second has many, starting with z −3 . This is because f (z) behaves like z −1 when
z is small and like z −3 when z is large.
Exercise 21.3. Suppose that
1 1 −1 1/2 1/2
f (z) = = = + + .
z(z 2− 1) z(z − 1)(z + 1) z z−1 z+1
Find Laurent series for f in the largest annuli with centre 1 in which f is holomor-
phic.

Answer. The function f has singularities at 0, ±1. The three largest annuli
with centre 1 in which f is holomorphic are {z ∈ C : 0 < |z − 1| < 1}, {z ∈ C : 1 <
|z − 1| < 2}, and {z ∈ C : 1 < |z − 1|}.
 2. FINDING LAURENT SERIES 131

In the annulus {z ∈ C : 0 < |z − 1| < 1}, we may write

−1 1/2 1/2
f (z) = + +
z z−1 z+1
−1 1 1 1 1
= + · + ·
z−1+1 2 z−1 2 z−1+2
−1 1 1 1 1
= + · + ·
1 + (z − 1) 2 z − 1 4 1 + (z − 1)/2
X∞
1 1 1 X (−1)n (z − 1)n
∞
= (−1) (z − 1) + ·
n+1 n
+
n=0
2 z − 1 4 n=0 2n
1 1 X
∞  1 
= · + (−1)n −1 + n+2 (z − 1)n
2 z − 1 n=0 2
X
∞  n+2 − 1 
n+1 2
= (−1) n+2
(z − 1)n .
n=−1
2

In the annulus {z ∈ C : 1 < |z − 1| < 2}, we may write

−1 1/2 1/2
f (z) = + +
z z−1 z+1
−1 1 1 1 1
= + · + ·
z−1+1 2 z−1 2 z−1+2
−1 1 1 1 1 1
= · + · + ·
z − 1 1 + 1/(z − 1) 2 z − 1 4 1 + (z − 1)/2
−1 X (−1)n 1 X (−1)n (z − 1)n
∞ ∞
1 1
= + · +
z − 1 n=0 (z − 1)n 2 z − 1 4 n=0 2n
X∞
(−1)n 1 1 X
∞
(−1)n (z − 1)n
= + · +
n=1
(z − 1)n 2 z − 1 n=0 2n+2
X
∞
= cn (z − 1)n ,
n=−∞

where cn = (−1)n if n ≤ −2 and cn = (−1)n 2−n−2 if n ≥ −1.

In the annulus {z ∈ C : 2 < |z − 1| < ∞}, we may write
−1 1/2 1/2
f (z) = + +
z z−1 z+1
−1 1 1 1 1
= + · + ·
z−1+1 2 z−1 2 z−1+2
−1 1 1 1 1 1
= · + · + ·
z − 1 1 + 1/(z − 1) 2 z − 1 2(z − 1) 1 + 2/(z − 1)
−1 X (−1)n X
∞ ∞
1 1 1 (−2)n
= + · +
z − 1 n=0 (z − 1)n 2 z − 1 2(z − 1) n=0 (z − 1)n
X∞
(−1)n+1 1 1 X
∞
(−1)n 2n−1
= + · +
n=0
(z − 1)n+1 2 z − 1 n=0 (z − 1)n+1
132 21. LAURENT SERIES

X∞
(−1)m−1 (2m−2 − 1)
= . 4
m=3
(z − 1)m

Remarks. These series involve both odd and even powers, reflecting the fact
that f is neither even nor odd about 1.
None of these series converges in any larger annulus, since otherwise f would be
holomorphic in a larger domain.
Note also that the last Laurent series starts with z −3 . This corresponds to
the fact that f (z) behaves like z −3 at infinity, and that (z − 1)−3 and z −3 behave
similarly there. Thus we should have expected cancellations in the terms (z − 1)−1
and (z − 1)−2 .
 LECTURE 22

Singularities

In this lecture, we study holomorphic functions in a punctured ball, and illustrate

this study with some examples. Recall that the punctured ball B ◦ (z0 , R) is defined
to be {z ∈ C : 0 < |z − z0 | < R}; we allow R to be ∞.

1. Laurent series at an isolated singularity

Definition 22.1. A function f has an isolated singularity at z0 in C if f is holomor-
phic in the punctured ball B ◦ (z0 , R) for some R ∈ R+ , and f is not differentiable
at z0 , perhaps because f (z0 ) is not defined or f is not continuous at z0 .
Suppose that f ∈ H(B ◦ (z0 , R)). Take r such that 0 < r < R. By Laurent’s
theorem,
X
∞
f (z) = cn (z − z0 )n ∀z ∈ B ◦ (z0 , R),
n=−∞
where Z
1 f (z)
cn = dz.
2πi ∂B(z0 ,r) (z − z0 )n+1
If cn = 0 for all n ∈ Z, then f is identically zero in B ◦ (0, R). In this case we
can define f (z0 ) to be 0 and then f ∈ H(B(z0 , R)). Otherwise, are three mutually
exclusive and exhaustive possibilities:
(i) There are infinitely many n ∈ Z− such that cn = 6 0. In this case, we say
that f has an (isolated) essential singularity at z0 .
(ii) There are no n ∈ Z− such that cn 6= 0. In this case, we say that f has a
removable singularity at z0 . If there exists M ∈ Z+ such that cM 6= 0 and
cn = 0 for all n < M , then we say that f has a zero of order (or multiplicity)
M at z0 . Zeros of order 1 are also known as simple zeros.
(iii) There are at least one and finitely many n ∈ Z− such that cn 6= 0. In this
case, there exists M ∈ Z− such that cM 6= 0 and cn = 0 for all n < M . In
this case, we say that f has a pole at z0 of order −M . Poles of order 1 are
also known as simple poles.
Examples 22.2. The following examples illustrate different types of singulari-
ties.
sin(z)
1. Suppose that f (z) = . The natural domain for f is C \ {0}, because f (0)
z
is not defined. As we have seen, it is possible to extend the definition of f to 0, by
setting f (0) = 1, and then the extended function (still written f ) is entire, and the
singularity has been removed.
(1 − cos(z))2
2. Suppose that f (z) = . The natural domain for f is C \ {0}, because
z
f (0) is not defined. As we will see, it is possible to extend the definition of f to 0,
133
134 22. SINGULARITIES

by setting f (0) = 0, and then f becomes an entire function, and the singularity has
been removed. This extended function has a zero of order three at 0.
1
3. Suppose that f (z) = 3 . The natural domain for f is C \ {0, ±1}, because
z −z
f (0), f (1) and f (−1) are not defined. As we will see, limz→0 f (z) = ∞, so we will
leave f (0) undefined. For this function,
X
∞
f (z) = (−1)n+1 z 2n−1
n=0
◦
in B (0, 1), and there is a pole of order 1, that is, a simple pole, at 0.
4. Suppose that f (z) = e1/z . The natural domain for f is C \ {0}, because f (0)
is not defined. As we will see, limz→0 f (z) does not exist, and f has an essential
singularity at 0.
5. Suppose that f (z) = Log(z). The domain of f is C \ {0}, and we say that f has
a singularity at 0. However, f is not differentiable at all points on the negative real
axis, due to the jump in Arg there, and so the singularity at 0 is not isolated.
1
6. Suppose that f (z) = . Then f is defined in C\({0}∪{1/n : n ∈ Z\{0}}).
sin(π/z)
In this example, f has isolated singularities at the points 1/n, where n ∈ Z \ {0},
and a nonisolated singularity at 0.
Theorem 22.3. Suppose that f ∈ H(B ◦ (z0 , R)), and that
X
∞
f (z) = cn (z − z0 )n ∀z ∈ B ◦ (z0 , R).
n=−∞

Then the following are equivalent:

(i) cn = 0 for all n < 0
(ii) lim f (z) exists in C, and by defining f (z0 ) to be lim f (z), we may extend
z→z0 z→z0
f to a holomorphic function on B(z0 , R)
(iii) lim f (z) exists in C
z→z0
(iv) there exists C ∈ R+ and r ∈ (0, R) such that
|f (z)| ≤ C ∀z ∈ B ◦ (z0 , r).
P
Proof. Suppose that (i) holds. Then f (z) = ∞ ◦
n=0 cn (z − z0 ) in B (z0 , R),
n
◦
and the power series converges in B (z0 , R), so its radius of convergence is at least
R. Hence limz→z0 f (z) = c0 , and by defining f (z0 ) to be c0 , we extend f to agree
with the power series in B(z0 , R), which is holomorphic there, that is, (ii) holds.
Suppose that (ii) holds. Then (iii) holds trivially.
Suppose that (iii) holds, i.e., that limz→z0 f (z) = c for some c ∈ C. Take ε = 1
in the definition of the limit. Then there exists r such that 0 < r < R and
|f (z) − c| < 1
when z ∈ B ◦ (z0 , r). For such z,
|f (z)| = |f (z) − c + c| ≤ |f (z) − c| + |c| < 1 + |c|,
so taking C to be 1 + |c|, we have |f (z)| < C, and (iv) holds.
 1. LAURENT SERIES AT AN ISOLATED SINGULARITY 135

Suppose that (iv) holds. We use an argument like that to prove Liouville’s
theorem. According to Laurent’s theorem, if 0 < r < R, then
Z
1 f (z)
cn = dz .
2πi ∂B(z0 ,r) (z − z0 )n+1
Clearly when z lies on ∂B(z0 , r),
f (z) |f (z)| C
= n+1 ≤ n+1 .
(z − z0 ) n+1 r r
The length of ∂B(z0 , r) is 2πr, and so, by the M L Lemma,
Z
1 f (z) 1 C
|cn | ≤ dz ≤ 2πr = 2πCr−n .
2π ∂B(z0 ,r) (z − z0 ) n+1 2π rn+1
If n < 0, then −n > 0, and r−n → 0 as r → 0. In this case, we can make r−n
arbitrarily small, so cn = 0, i.e., (i) holds. □

The reason for the terminology “removable singularity” should now be clear:
by defining, or perhaps redefining, f (z0 ) appropriately, we may extend f to be
holomorphic in B(z0 , R), that is, we remove the singularity of f at z0 .
It is possible to generalize the proof of the theorem above to prove the following
result.
Theorem 22.4. Suppose that f ∈ H(B ◦ (z0 , R)), and that
X
∞
f (z) = cn (z − z0 )n ∀z ∈ B ◦ (z0 , R).
n=−∞

Then the following are equivalent:

(i) cn = 0 for all n < M and cM 6= 0
(ii) there exists F ∈ H(B(z0 , R)) such that f (z) = (z − z0 )M F (z) in B ◦ (z0 , R)
and F (z0 ) 6= 0
(iii) lim (z − z0 )−M f (z) exists and is in C \ {0}
z→z0
(iv) there exists C ∈ R+ and r ∈ (0, R) such that
|f (z)| ≤ C|z − z0 |M ∀z ∈ B ◦ (z0 , r),
but lim (z − z0 )−M f (z) 6= 0 (either because the limit does not exist or be-
z→z0
cause it exists and is not 0).

We omit the details of the proof, but observe that it involves defining the function
F by F (z) = (z − z0 )−M f (z), applying the techniques of the proof of the previous
theorem to F , and being careful about when limits are 0.
We write f (z) ∼ c(z − z0 )N as z → z0 , where c ∈ C \ {0} and N ∈ Z, to mean
that
f (z)
lim = 1.
z→z0 c(z − z0 )N

Note that if f (z) ∼ c(z − z0 )N as z → z0 , then there is a punctured ball B(z0 , r) in

which f does not vanish.
136 22. SINGULARITIES

Summary. If f has a zero of order N at z0 , then f (z) ∼ cN (z − z0 )N as z → z0 ,

where cN 6= 0, and vanishes at z0 ; when the order of the zero is higher, f (z) vanishes
more rapidly.
If f has a pole of order N at z0 , then f (z) ∼ cN (z − z0 )−N as z → z0 , where c 6= 0,
and diverges to ∞; when the order of the pole is higher, the divergence is more
rapid.
If f has a removable singularity at z0 , then f is bounded near z0 , and vice versa.
Corollary 22.5. Suppose that f, g, h ∈ H(B ◦ (z0 , R)) and that f has a zero of
order M at z0 while g has a zero of order N at z0 , while h has a pole of order P at
z0 . Then:
(i)1/h has a zero of order P at z0 ;
(ii)1/f has a pole of order M at z0 ;
(iii)f g has a zero of order M + N at z0 ;
(iv) if M ≥ N , then f /g has a removable singularity at z0 , and if M > N , then
f /g has a zero of order M − N ;
(v) if M < N , then f /g has a pole of order N − M .
Proof. We prove only (v). If f has a zero of order M at z0 while g has a
zero of order N at z0 , then f (z) ∼ c(z − z0 )M and g(z) ∼ d(z − z0 )N as z → z0 ,
where c, d ∈ C \ {0}. If follows that (f /g)(z) ∼ (c/d)(z − z0 )M −N as z → z0 , where
(c/d) 6= 0. □
We may now prove l’Hôpital’s rule.
Theorem 22.6. Suppose that f, g ∈ H(Ω) and z0 ∈ Ω. Suppose also that
limz→z0 f (z)/g(z) is of the form 0/0. If limz→z0 f ′ (z)/g ′ (z) exists, then so does
limz→z0 f (z)/g(z), and the limits are equal.
Proof. By hypothesis, f and g have Taylor series in some ball centred at z0 ,
and f (z) ∼ a(z − z0 )M while g(z) ∼ b(z − z0 )N as z → z0 , where a, b ∈ C \ {0}. Then
f ′ (z) ∼ aM (z − z0 )M −1 and g ′ (z) ∼ bN (z − z0 )N −1 as z → z0 . If limz→z0 f ′ (z)/g ′ (z)
exists, then
f (z) a(z − z0 )M N aM (z − z0 )M −1 N f ′ (z)
lim = lim = lim = lim .
z→z0 g(z) z→z0 b(z − z0 )N z→z0 M bN (z − z0 )N −1 M z→z0 g ′ (z)
Since the right hand side exists, so does the left hand side. Further, the right hand
side may be 0, in which case the left hand side is too, and the desired equality holds,
or a nonzero complex number, in which case M = N and then the desired equality
also holds. □

2. Examples
Exercise 22.7. Suppose that f (z) = (1 − cos(z))2 /z. What kind of singularity
does f have at 0?
Answer. Clearly
(1 − cos(z))2 2(1 − cos(z)) sin(z)
lim f (z) = lim = lim =0
z→0 z→0 z z→0 1
by l’Hôpital’s rule. So f has a zero of some currently unknown order at 0.
 3. MORE ABOUT SINGULARITIES† (NOT EXAMINABLE) 137

There are various ways to proceed. Here is one:

1 + cos(2z)
(1 − cos(z))2 = 1 − 2 cos(z) + cos2 (z) = 1 − 2 cos(z) + ,
2
whence
z2 z4 z6 
(1 − cos(z))2 = 1 − 2 1 − + − + ...
2! 4! 6!
1  4z 2
16z 4 64z 6 
+ 1+1− + − + ...
2 2! 4! 6!
1 1
= 0z 0 + 0z 2 + z 4 − z 6 + . . . ;
4 24
by dividing by z, we obtain the series for f , which has a zero of order 3.
1 − cos(z) 1
Here is another: by L’Hôpital’s rule, limz→0 2
= , whence
z 2
(1 − cos(z)) 2
1
lim 4
= ,
z→0 z 4
and so (1 − cos(z)) has a zero of order 4 at 0, whence f has a zero of order 3. 4
2

Exercise 22.8. Suppose that f (z) = e1/z . How does f behave near 0?
Answer. From Taylor series,
w2 w3
ew = 1 + w + + + ....
2! 3!
Hence
z −2 z −3
e1/z = 1 + z −1 + + + ....
2! 3!
Since there are infinitely many nonzero negative powers of z, the singularity at 0 is
essential. 4

3. More about singularities† (Not examinable)

We can say more about the behaviour of a function near a singularity.
Theorem 22.9 (Picard’s theorem). If f has an essential singularity at z0 , then
for all δ > 0, the set C \ f (B ◦ (z0 , δ)) has at most one element.
We shall not prove this result. But consider Exercise 22.8. Given any c in C\{0},
we can find w such that ew = c. Let wn be w + 2πin, for all n ∈ N. Then ewn = c.
Now take zn = 1/wn ; this sequence tends to 0. Thus there is a zn as close to 0 as
we like for which e1/zn = c. This illustrates Picard’s theorem.
 LECTURE 23

Residues and the residue theorem

In this lecture, we define residues, and prove Cauchy’s residue theorem. We

show how to find residues. This will enable us to find integrals over closed contours
efficiently.

1. Residues
Definition 23.1. Suppose that the function f has an isolated singularity at z0 .
Then
P∞ f is holomorphic in some punctured ball B ◦ (z0 , r), and has a Laurent series
n=−∞ cn (z − z0 ) there. The residue of f at z0 , written Res(f, z0 ) or Res(f (z), z =
n

z0 ), is defined to be c−1 , the coefficient of (z − z0 )−1 in this series.

From Laurent’s theorem, we see that if f is holomorphic in B ◦ (z0 , r), and Γ is a
simple closed contour in B ◦ (z0 , r) around z0 , then
Z
1
Res(f, z0 ) = f (z) dz. (23.1)
2πi Γ
Observe that Res(g, z0 ) = 0 if g is holomorphic in B(z0 , r); this can be seen from
the definition above, or from the Cauchy–Goursat theorem and (23.1).
We are interested in residues because of the following generalization of Cauchy’s
integral formula.
Theorem 23.2 (Cauchy’s residue theorem). Suppose that Γ is a simple closed con-
tour with the standard orientation in a domain Ω and f ∈ H(Ω), and further that
Int(Γ) ∩ Ω = Int(Γ) \ {z1 , z2 , . . . , zK }. Then
Z XK
f (z) dz = 2πi Res(f, zk ).
Γ k=1

We prove this theorem later. This theorem enables us to calculate integrals over
closed curves as long as we can calculate residues.
Exercise 23.3. Suppose that
α1 β1 β2
f (z) = + + .
z − a z − b (z − b)2
R
Find the residues of f at a and b, and hence find Γ f (z) dz, where Γ is a simple
closed contour surrounding a and b.
Answer. Both the singularities a and b lie inside Γ, and both may contribute
to the sum of residues.
To find Res(f, a), we observe that the second and third terms are holomorphic
inside a very small ball B(a, ε) centred at a, so only contribute terms in (z − a)n
where n ≥ 0 to the Laurent series for f in B ◦ (a, ε), and the residue is α1 .
139
140 23. RESIDUES AND THE RESIDUE THEOREM

To find Res(f, b), we observe that the first term is holomorphic inside a very
small ball B(b, ε) centred at b, so only contributes terms in (z − b)n where n ≥ 0 to
the Laurent series for f in B ◦ (b, ε), and the residue is β1 .
Consequently,
Z



f (z) dz = 2πi Res(f, a) + Res(f, b) = 2πi α1 + β1 . 4
Γ

This example shows that, for rational functions, residues are coefficients of cer-
tain terms in the partial fraction expansions, and that not all the coefficients are
needed to compute contour integrals (in the example above, β2 does not matter at
all). If we can compute residues efficiently, then we will not need to find partial
fraction expansions.

2. Computing residues
Suppose that f has an isolated singularity at z0 . There are three ways to compute
Res(f, z0 ), according to the type of singularity,
First, if the singularity is removable, that is, if f is bounded near z0 , the residue
is 0.
Second, if f has a pole of order N at z0 , then
f (z) = c−N (z − z0 )−N + c1−N (z − z0 )1−N + . . .
+ c−1 (z − z0 )−1 + c0 (z − z0 )0 + c1 (z − z0 )1 + . . .
so
(z − z0 )N f (z) = c−N (z − z0 )0 + c1−N (z − z0 )1 + . . .
+ c−1 (z − z0 )N −1 + c0 (z − z0 )N + c1 (z − z0 )N +1 + . . .
and
dN −1  
(z − z0 ) f (z)
N
dz N −1
dN −1 
= N −1 c−N (z − z0 )0 + c1−N (z − z0 )1 + . . .
dz 
+ c−1 (z − z0 )N −1 + c0 (z − z0 )N + c1 (z − z0 )N +1 + . . .
(N + 1)!
= (N − 1)! c−1 (z − z0 )0 + N ! c0 (z − z0 )1 + c1 (z − z0 )2 + . . . ,
2!
whence
dN −1
lim (z − z0 )N f (z) = (N − 1)! c−1 .
z→z0 dz N −1
Thus
1 dN −1
Res(f, z0 ) = lim N −1 (z − z0 )N f (z). (23.2)
(N − 1)! z→z0 dz

Note that this formula is not the definition of a residue; that definition is at the
start of this lecture. To use this formula, we have to know the order of the pole,
and it does not apply at all singularities.
Third, if f has an essential singularity at z0 then we need a different approach,
usually involving integration or series.
 2. COMPUTING RESIDUES 141

If f has a singularity at z0 , we could proceed by looking at

lim (z − z0 )n f (z)
z→z0

for increasing values of n, starting at 0, until such time as we find a finite number;
this is then the order of the pole, N . Once we know N , we use formula (23.2). This
would be very inefficient if the order of the pole is large, and would not give a result
at all for an essential singularity! It is therefore important to know how to find the
order of the pole. The key information about the orders of zeros and poles is in the
previous lecture.
z − π/2
Exercise 23.4. Suppose that f (z) = . Find Res(f, π/2).
1 − sin(z)

Answer. By l’Hôpital’s rule,

z − π/2 1
lim f (z) = lim = lim ,
z→π/2 z→π/2 1 − sin(z) z→π/2 − cos(z)
which does not exist, so the singularity at π/2 cannot be removed.
By l’Hôpital’s rule again,
(z − π/2)2 2(z − π/2) 2
lim (z − π/2)f (z) = lim = lim = lim = 2,
z→π/2 z→π/2 1 − sin(z) z→π/2 − cos(z) z→π/2 sin(z)

and so the pole is simple and Res(f, π/2) = 2. 4

We could have worked out that the pole was simple by working out that the zero
of the numerator is of order 1 and that of the denominator is of order 2, and then
doing the second calculation only.
The following result is often useful.
Proposition 23.5 (The p/q ′ formula). Suppose that f (z) = p(z)/q(z) in Ω, and
that p(z0 ) 6= 0 while q(z0 ) = 0. If z0 is a simple zero of q, that is, a zero of order 1,
then
p(z0 )
Res(f, z0 ) = ′ .
q (z0 )

Proof. Since the pole of f is simple,

p(z) (z − z0 )
Res(f, z0 ) = lim (z − z0 ) = lim p(z)
z→z0 q(z) z→z 0 q(z)
(z − z0 ) 1 p(z0 )
= p(z0 ) lim = p(z0 ) ′ = ′ ,
z→z0 q(z) q (z0 ) q (z0 )
by l’Hôpital’s rule. □
Exercise 23.6. Suppose that
f (z) = tan(z).
Find the residue of f at (2k + 1)π/2, where k ∈ Z.
142 23. RESIDUES AND THE RESIDUE THEOREM

Answer. This function has singularities at (2k + 1)π/2, where k ∈ Z, because

tan = sin / cos and cos is zero at these points. The zeros of cos are of order 1. By
the p/q ′ formula,
sin((2k + 1)π/2)
Res(f, (2k + 1)π/2) = = −1. 4
− sin((2k + 1)π/2)

Of course, we could also have done this calculation directly, without using the
formula.
Exercise 23.7. Suppose that
f (z) = 2z sin(z −2 ).
Find the residue of f at 0.
Answer. By Taylor series,
w3 w5
sin(w) = w − + − ...,
3! 5!
and so, replacing w by z −2 , we see that
 z −6 z −10 
2z sin(z −2 ) = 2z z −2 − + − ...
3! 5!
 z −5
z −9 
−1
=2 z − + − ... .
3! 5!
Hence Res(f, 0) = 2. 4

Exercise 23.8. Suppose that

f (z) = sin(z) e1/z .
Find the residue of f at 0.
Answer. Observe that
X∞
(−1)m 2m+1 X 1 −n
∞
1/z
f (z) = sin(z) e = z z
m=0
(2m + 1)! n=0
n!
X∞ X ∞
(−1)m 1 2m−n+1
= z .
m=0 n=0
(2m + 1)! n!
To get terms in z −1 , we require that 2m − n + 1 = −1, that is, that n = 2m + 2.
Thus
X∞
(−1)m
Res(f, 0) = . 4
m=0
(2m + 1)! (2m + 2)!

The expression for the residue in the last example does not simplify but may be
computed numerically to any desired degree of accuracy. This behaviour is common
with essential singularities.
Essential singularities usually appear because there is a “transcendental func-
tion” (such as exp or sinh or tan) which can be written as a power series, with
something like 1/z in its argument. Residues involving transcendental functions
 3. PROOF OF CAUCHY’S RESIDUE THEOREM 143

Im

× z3 × z2 Γ

× z1

× z4 × z5
Re

Figure 23.1. Curves around the singularities inside Γ

can often be expressed as series which are derived from the corresponding power
series.

3. Proof of Cauchy’s Residue Theorem

For convenience, we restate the theorem.
Theorem (Cauchy’s residue theorem). Suppose that Γ is a simple closed contour
with the standard orientation in a (multiply connected) domain Ω, that f ∈ H(Ω),
and that Int(Γ) ∩ Ω = Int(Γ) \ {z1 , z2 , . . . , zK }. Then
Z X
K
f (z) dz = 2πi Res(f, zk ).
Γ k=1

Proof. We take balls B(zk , ε) centred at the singularities zk , where the ε are
chosen small enough that the closed balls B̄(zi , ε) and B̄(zj , ε) are disjoint if i 6= j,
and such that each closed ball B̄(zk , ε) is contained in Int(Γ). This ensures that
the contours around the perimeters of the balls B(zk , ε) are well defined and do not
meet each other or Γ. Define
[ K 
Υ = Int(Γ) \ B(zk , ε) .
k=1

See Figure 23.1. Then f is holomorphic on the open set Ω which contains Ῡ, and ∂Υ
is made up of Γ, traversed anti-clockwise, together with the boundaries ∂B(zk , ε)
traversed clockwise.
The Cauchy–Goursat theorem implies that
Z Z XK Z
0= f (z) dz = f (z) dz + f (z) dz,
∂Υ Γ k=1 ∂B(zk ,ε)∗

that is,
Z K Z
X
f (z) dz = f (z) dz.
Γ k=1 ∂B(zk ,ε)
144 23. RESIDUES AND THE RESIDUE THEOREM

From (23.1), Z
1
Res(f, zk ) = f (z) dz,
2πi ∂B(zk ,ε)
so
Z
f (z) dz = 2πi Res(f, zk ),
∂B(zk ,ε)

and hence
Z K Z
X X
K
f (z) dz = f (z) dz = 2πi Res(f, zk );
Γ k=1 ∂B(zk ,ε) k=1
the theorem is proved. □
 LECTURE 24

Computing integrals. 1

In this lecture, we give three examples of the use of Cauchy’s residue theorem
to calculate integrals.

1. Trigonometric integrals
Z π
cos(θ)
Exercise 24.1. Evaluate dθ.
−π 5 − 4 cos(θ)

Answer. The first step is to convert this to a contour integral. We take γ(θ) =
eiθ , where −π ≤ θ ≤ π. Then γ ′ (θ) = ieiθ = iγ(θ) and
eiθ + e−iθ γ(θ) + γ(θ)−1
cos(θ) = = .
2 2
As θ varies over its domain, γ(θ) travels anticlockwise around the unit circle in the
complex plane. Thus
Z π Z
cos(θ) (z + z −1 )/2 1
dθ = dz
−π 5 − 4 cos(θ) γ 5 − 2(z + z ) iz
−1
Z
1 z2 + 1 1
= dz
2i γ (−2z + 5z − 2) z
2
Z
i z2 + 1
= dz
2 γ z(2z 2 − 5z + 2)
Z
i z2 + 1
= dz
2 γ z(2z − 1)(z − 2)
Z
i
= f (z) dz,
2 γ
where
z2 + 1
f (z) = dz.
z(2z − 1)(z − 2)
The second step is to evaluate the contour integral by evaluating the residues of f .
Clearly, f has singularities at 0, 21 and 2. Only 0 and 12 lie inside γ. Because each
of the factors in the denominator is of degree 1, and the numerator does not vanish
at these points, each of the singularities is a simple pole. Further
z2 + 1 1
Res(f, 0) = lim (z − 0)f (z) = lim zf (z) = lim =
z→0 z→0 z→0 (2z − 1)(z − 2) 2
and
z2 + 1 5/4 5
Res(f, 1/2) = lim (z − 1/2)f (z) = lim = =− .
z→1/2 z→1/2 2z(z − 2) −3/2 6
145
146 24. COMPUTING INTEGRALS. 1

By the residue theorem,

Z π Z
cos(θ) i z2 + 1
dθ = dz
−π 5 − 4 cos(θ) 2 γ z(2z − 1)(z − 2)
i 1 5
= 2πi −
2 2 6
π
= . 4
3

Remarks on this calculation. If the integrand had contained a sin term, then
we could have used the fact that
eiθ − e−iθ z − z −1
sin(θ) = = .
2i 2i
Similarly, expressions such as sin(2θ) and tan(θ) may be expressed in terms of eiθ
and hence of z. In summary, any integral that can be put in the form
Z π
f (sin(θ), cos(θ)) dθ
−π

where f is a rational function of two variables, can be tackled in this way, and
becomes an integral of a different rational function around a closed contour, which
can be evaluated, at least in principle, and often in practice.
There are some other integrals that may be converted to integrals of this form.
For instance, since cos is an even function, so is θ 7→ cos(θ)/(5 − 4 cos(θ)), and
Z π Z 0
cos(θ) cos(θ)
dθ = dθ
0 5 − 4 cos(θ) −π 5 − 4 cos(θ)

and so Z Z
π
cos(θ) 1 π cos(θ)
dθ = dθ.
0 5 − 4 cos(θ) 2 −π 5 − 4 cos(θ)
If the integrand involves expressions like cos(θ/2), this method does not work: square
roots appear, and these mess up the holomorphy.

2. A rational function on the real line

Z ∞
1
Exercise 24.2. Evaluate 4
dx.
−∞ x + 1

Answer. The real line is not a simple closed contour, and it is not obvious how
the residue theorem can be used.
Define f by
1 1
f (z) = 4 = ,
z +1 (z − ω1 )(z − ω2 )(z − ω3 )(z − ω4 )
where ωk = exp( 12 πik − 14 πi).
R The function f has four singularities,
R at the points
ωk . We will integrate f , as f (z) dz should be related to f (x) dx.
We will take Ω to be the semicircular region Ω above the interval [−R, R], and
integrate f around the boundary ∂Ω. We suppose that R > 1, so that all the
singularities of the integrand in the upper half plane lie in Ω.
 2. A RATIONAL FUNCTION ON THE REAL LINE 147

Im

Ω
ω2 × × ω1 Re
−R ω3 × × ω4 R

Figure 24.1. The region Ω

We defined f by
1 1
f (z) = = ,
z4 + 1 (z − ω1 )(z − ω2 )(z − ω3 )(z − ω4 )
where ωk = exp( 12 πik − 14 πi). The only singularities of f that lie in Ω are ω1 and ω2 .
Since the power of (z − ωk ) in the denominator is 1, these are simple poles. Further,
ωk4 = −1, and so ωk3 = −1/ωk . We deduce that
z − ωk 1 1 1
Res(f, ωk ) = lim (z − ωk )f (z) = lim 4
= lim 3
= 3
= − ωk .
z→ωk z→ωk z +1 z→ωk 4z 4ωk 4
Now 2πi multiplied by the sum of the residues is given by
2πi  1 + i −1 + i 
2πi(Res(f, ω1 ) + Res(f, ω2 )) = − √ + √
4 2 2
  √
πi 2i π 2
=− √ = .
2 2 2
The boundary contour ∂Ω has two parts: the circular arc and the diameter.
Parametrise the diameter by the function λ : [−R, R] → C, defined by λ(x) = x.
Then one component of the integral around the contour is
Z Z R Z R
′
f (z) dz = f (λ(x)) λ (x) dx = f (x) dx
λ −R −R
Z R Z ∞ (24.1)
1 1
= 4
dx → 4
dx
−R x + 1 −∞ x + 1

as R → ∞; this limit is exactly the integral that we were asked to compute.

Parametrise the semicircular arc by the function γ : [0, π] → C, defined by
γ(t) = Reit . When z ∈ Range(γ), then |z 4 + 1| ≥ |z|4 − 1 = R4 − 1 by the triangle
inequality; thus
1 1
≤ 4 .
|z + 1|
4 R −1
By the M L lemma,
Z
1 πR
f (z) dz ≤ 4 length(γ) = 4 →0
γ R −1 R −1
148 24. COMPUTING INTEGRALS. 1

Im
iR
Γ2
Γ1
Γ3
ir Re
× ×
−ir 1
Γ4

−iR

Figure 24.2. An indented contour

as R → ∞, that is,
Z
f (z) dz → 0. (24.2)
γ

By the residue theorem,

Z √

π 2
f (z) dz = 2πi Res(f, ω1 ) + Res(f, ω2 ) = .
∂Ω 2
That is,
Z √ Z
π 2
f (z) dz = − f (z) dz. (24.3)
λ 2 γ

We conclude from (24.1), (24.3) and (24.2) that

Z ∞
1
Z  π √2 Z  π √2
4
dx = lim f (z) dz = lim − f (z) dz = . 4
−∞ x + 1 R→∞ λ R→∞ 2 γ 2

3. An integral involving a root

Exercise 24.3. Compute the contour integral
Z
z 1/2
dz,
Γ z −1
2

where Γ is the join of the contours Γ1 , Γ2 , Γ3 and Γ4 shown in Figure 24.2, and z 1/2
denotes the principal branch of the square root. What happens to the component
integrals when r → 0 and R → ∞? Find
Z +∞ 1/2
x
dx.
0 x2 + 1

Answer. We suppose that r < 1 < R. First of all, this is a simple closed
contour, and the integrand is holomorphic on and inside the contour, except for a
 3. AN INTEGRAL INVOLVING A ROOT 149

simple pole at 1, so
Z  z 1/2 
z 1/2
dz = 2πi Res , z = 1
Γ z2 − 1 z2 − 1
z 1/2 (z − 1)
= 2πi lim
z→1 z2 − 1
= πi.
Now, by the M L lemma,
Z n z 1/2 o
z 1/2
dz ≤ max : |z| = r, Re(z) ≥ 0 πr
Γ3 z − 1 z2 − 1
2

r1/2
≤ πr → 0 as r → 0.
1 − r2
Similarly, by the M L lemma,
Z n z 1/2 o
z 1/2
dz ≤ max : |z| = R, Re(z) ≥ 0 πR
Γ1 z − 1 z2 − 1
2

R1/2
≤ πR → 0 as R → ∞.
R2 − 1
Parametrise Γ2 by γ2 (y) = iy, where y goes from R to r, and Γ4 by γ4 (y) = −iy,
where r ≤ y ≤ R. Then,
Z Z R Z R 1/2
z 1/2 (iy)1/2 y
dz = − i dy = i(+i) 1/2
dy,
Γ2 z − 1 r −y − 1
2 2 2
r y +1

and Z Z Z
R R
z 1/2 (−iy)1/2 y 1/2
dz = (−i) dy = i(−i)1/2 dy.
Γ4 z2 − 1 r −y 2 − 1 r y2 + 1
As r → 0 and R → ∞,
Z R Z ∞
y 1/2 y 1/2
dy → dy.
r y2 + 1 0 y2 + 1
Now
Z Z Z Z
z 1/2 z 1/2 z 1/2 z 1/2
dz + dz + dz + dz
Γ1 z2 − 1 Γ2 z2 − 1 Γ3 z − 1
2
Γ4 z − 1
2
Z
z 1/2
= dz = πi ,
Γ z −1
2

whence
Z z 1/2
Z
z 1/2
Z
z 1/2
Z
z 1/2 
lim dz + dz + dz + dz
r→0
R→∞ Γ1 z2 − 1 Γ2 z2 − 1 Γ3 z2 − 1 Γ4 z2 − 1
= πi,
so
 Z ∞
y 1/2
1/2 1/2
i(i ) + i(−i) dy = πi,
0 y2 + 1
150 24. COMPUTING INTEGRALS. 1

and hence Z ∞
y 1/2 π
2
dy = √ . 4
0 y +1 2

We can compute integrals involving functions with branches as long as the branch
cut is outside the contour. The process of avoiding a singularity, such as 0 here, by
adding a small circular arc to the contour, is called indenting.
 LECTURE 25

Computing Integrals. 2

In this lecture, we compute more definite integrals, and introduce Jordan’s

lemma, which can be useful in this context.

1. An integral involving an exponential

Z ∞ iξx
e
Exercise 25.1. Evaluate 2
dx, where ξ ∈ R+ .
−∞ x + 1

Answer. Define f by

eiξz eiξz
f (z) = = .
z2 + 1 (z − i)(z + i)

The function f has two singularities, at the points ±i. We integrate around the
boundary of the semicircular region Ω above the interval [−R, R] shown in Figure
25.1. We suppose that R > 1, so that all the singularities of the integrand in the
upper half plane lie inside the boundary contour ∂Ω.
Since the power of (z − i) in the denominator of the fraction defining f is 1, the
pole at i is simple. From l’Hôpital’s rule,

z−i 1 1
Res(f, i) = lim(z − i)f (z) = lim eiξz 2
= e−ξ lim = e−ξ .
z→i z→i z +1 z→i 2z 2i

The contour ∂Ω has two components: the circular arc and the diameter.

Im

Ω
×i
Re
−R × −i R

Figure 25.1. The region Ω

151
152 25. COMPUTING INTEGRALS. 2

Parametrise the diameter by the function λ : [−R, R] → C, given by λ(x) = x.

Then
Z Z R
f (z) dz = f (λ(x)) λ′ (x) dx
λ −R
Z R
= f (x) dx
−R
Z R Z ∞
eiξx eiξx
= dx → dx
−R x2 + 1 −∞ x2 + 1

as R → ∞; this is exactly the integral that we were asked to compute. We have

shown that
Z ∞ iξx Z
e
2
dx = lim f (z) dz. (25.1)
−∞ x + 1 R→∞ λ

When z ∈ Range(γ), then z = x + iy where x ∈ R and y ≥ 0, so

|eiξz | = |eiξx−ξy | = |eiξx | |e−ξy | = |e−ξy | ≤ 1,

and further, by the triangle inequality, |z 2 + 1| ≥ |z|2 − 1 = R2 − 1; thus

eiξz 1 1
≤ 2 ≤ 2 = M,
|z + 1|
2 |z + 1| R −1
say.
From the M L lemma,
Z
1 πR
f (z) dz ≤ M length(γ) = πR = 2 →0
γ R2 −1 R −1

as R → ∞, that is,
Z
f (z) dz → 0. (25.2)
γ

By the residue theorem,

Z Z
e−ξ
f (z) dz + f (z) dz = 2πi Res(f, i) = 2πi = πe−ξ ,
λ γ 2i

that is,
Z Z
−ξ
f (z) dz = πe − f (z) dz. (25.3)
λ γ

We conclude from (25.1), (25.3) and (25.2) that

Z ∞ iξx Z  Z 
e −ξ
2+1
dx = lim f (z) dz = lim πe − f (z) dz = πe−ξ . 4
−∞ x R→∞ λ R→∞ γ
 3. AN EXAMPLE OF THE USE OF JORDAN’S LEMMA 153

Remarks on this calculation. There are many integrals on the whole line R
that may be evaluated in this way. Integrals between 0 and ∞ may also be tackled
if the integrand is even. When the integrand involves cos(ξx) or sin(ξx), then we
write the trigonometric function in terms of exponentials, and if necessary change
variables to get an integral involving eiξx , where ξ ≥ 0. Then |eiξz | ≤ 1 when
z ∈ Range(γ). Sometimes this inequality is not enough and we have to use Jordan’s
lemma (see below).
For example, by the changes of variable x′ = −x and then x = x′ ,
Z ∞ Z Z ∞ −iξx
cos(ξx) 1  ∞ eiξx e 
2
dx = dx + dx
−∞ x + 1 2 −∞ x2 + 1 2
−∞ x + 1
Z Z ∞
1  ∞ eiξx 
′
eiξx ′
= dx + ′ 2
dx
2 −∞ x2 + 1 −∞ (x ) + 1
Z ∞ iξx
e
= 2
dx,
−∞ x + 1

which we have just computed. Since cos(x) and x2 are even functions of x,
Z ∞ Z
cos(ξx) 1 ∞ cos(ξx) πe−ξ
dx = dx = .
0 x2 + 1 2 −∞ x2 + 1 2

2. Jordan’s lemma
Lemma 25.2. Suppose that f is continuous in {z ∈ C : Im(z) ≥ 0, |z| ≥ S}, for
some positive S, that ΓR is the upper half of the circle with centre 0 and radius R,
and that |f (z)| ≤ MR for all z ∈ ΓR where limR→∞ MR = 0.
Then for any ξ > 0, Z
lim eiξz f (z) dz = 0.
R→∞ ΓR

Proof. We give the proof at the end of this lecture. □

3. An example of the use of Jordan’s lemma

Z ∞
x eiξx
Exercise 25.3. Evaluate 2+1
dx, where ξ ∈ R+ .
−∞ x
Answer. We integrate around the boundary of the same semicircular region Ω
above the interval [−R, R] as before, as shown in Figure 25.1.
Define f by
z eiξz z eiξz
f (z) = 2 = .
z +1 (z − i)(z + i)
The function f has two singularities, at the points ±i; only the singularity at i lies
in the region Ω. Since the power of (z − i) in the denominator is 1, this is a simple
pole. We deduce that
z−i 1 1
Res(f, i) = lim(z − i)f (z) = lim z eiξz 2 = ie−ξ lim = e−ξ .
z→i z→i z +1 z→i 2z 2
The boundary contour has two components: the circular arc and the diameter.
154 25. COMPUTING INTEGRALS. 2

Parametrise the diameter by the function λ : [−R, R] → C, given by λ(x) = x.

Then
Z Z R
f (z) dz = f (λ(x)) λ′ (x) dx
λ −R
Z R
= f (x) dx (25.4)
−R
Z R
x eiξx
= dx.
−R x2 + 1
It is not obvious what happens to this integral when R → ∞.
Parametrise the circular arc by the function γ : [0, π] → C, given by γ(t) = Reit .
When z ∈ Range(γ), then z = x + iy where x ∈ R and y ≥ 0, so, as before,

|eiξz | ≤ 1,
and further, by the triangle inequality, |z 2 + 1| ≥ |z|2 − 1 = R2 − 1; thus
z eiξz |z| R
≤ ≤ .
|z 2 + 1| |z 2 + 1| R2 − 1
From the M L lemma,
Z
R πR2
f (z) dz ≤ length(γ) = 2 →π
γ R2 − 1 R −1
as R → ∞. This is not helpful. However, since M → 0 as R → ∞, Jordan’s lemma
still tells us that Z
f (z) dz → 0. (25.5)
γ

By the residue theorem,

Z Z
e−ξ
f (z) dz + f (z) dz = 2πi Res(f, i) = 2πi = iπe−ξ ,
λ γ 2
that is,
Z Z
−ξ
f (z) dz = iπe − f (z) dz. (25.6)
λ γ

We conclude from (25.4), (25.6) and (25.5) that, as R → ∞,

Z R
x eiξx
2
dx → iπe−ξ . 4
−R x + 1

We may write the integral in the previous exercise as

Z Z
x cos(ξx) x sin(ξx)
2
dx + i 2
dx.
R x +1 R x +1

The result implies that the first integral is 0, as it should be since the integrand is
odd, while the second integral is πe−ξ .
 4. PROOF OF JORDAN’S LEMMA† 155

y
y = 2θ/π
1 b

y = sin(θ)
θ
π/2 π

Figure 25.2. The graphs y = sin θ and y = πθ/2.

4. Proof of Jordan’s lemma†

For convenience, we restate the result.

Lemma. Suppose that f is continuous in the set {z ∈ C : Im(z) ≥ 0, |z| ≥ S},

for some positive S, that γR is the upper half of the circle with centre 0 and radius
R, and that |f (z)| ≤ MR for all z ∈ γR where limR→∞ MR = 0.
Then for any ξ > 0,
Z
lim eiξz f (z) dz = 0.
R→∞ γR

Proof. We parametrise γR by setting γR (θ) = Reiθ . Then

Z Z π
iξz iθ
e f (z) dz = eiξRe f (Reiθ ) iReiθ dθ
γR 0
Z π
≤ eiξR(cos(θ)+i sin(θ)) f (Reiθ ) iReiθ dθ
Z0 π
= e−ξR sin(θ) f (Reiθ ) R dθ (25.7)
Z0 π
≤ e−ξR sin(θ) MR R dθ
0
Z π/2
=2 e−ξR sin(θ) MR R dθ,
0

because sin is symmetric about π/2.

It is clear from the graph in Figure 25.2, and could be proved by using either
the mean value theorem or convexity, that sin(θ) ≥ 2θ/π when 0 ≤ θ ≤ π/2, so
−ξR sin(θ) ≤ −2ξRθ/π and

e−ξR sin(θ) ≤ e−2ξRθ/π .

156 25. COMPUTING INTEGRALS. 2

Thus, substituting t = 2ξRθ/π, we see from (25.7) that

Z Z π/2
e f (z) dz ≤ 2MR
iξz
e−2ξRθ/π R dθ
γR 0
ZξR
π
= MR e−t dt
ξ
Z0 ∞
π
≤ MR e−t dt
ξ 0
π
= MR
ξ
→0
as R → ∞, as required. □
 LECTURE 26

The Theory of Functions

In this lecture, we count the number of times a closed curve winds around a
point, and use this to count the number of zeros and poles of a function. This leads
to some surprising facts about holomorphic functions.
We start with several definitions that could have come earlier.
Definition 26.1. A point z of a subset S of a set Ω is isolated if there exists
ε ∈ R+ such that B(z, ε) ∩ S = {z}. We say that S is discrete if every point of S is
isolated.
Definition 26.2. A function is said to be meromorphic in the open set Ω if it
is holomorphic in Ω \ ∆, where ∆ is a discrete subset of Ω, and the singularity at
each point of ∆ is a pole.

1. The winding number, zeros and poles

If γ : [a, b] → C is a closed curve that does not pass through 0, then γ(b) = γ(a)
and so log γ(b) = log γ(a). We may choose arg(γ(t)) to vary continuously with
t. When we do this, arg(γ(b)) − arg(γ(a)) is 2π multiplied by an integer, and the
integer is the total number of times γ winds about 0 in the anti-clockwise sense. If
we now define log using this version of arg, then
log(γ(b)) − log(γ(a)) = ln |γ(b)| + i arg(γ(b)) − ln |γ(a)| − i arg(γ(a))
= i(arg(γ(b)) − arg(γ(a))),
and so (log(γ(b)) − log(γ(a))) /(2πi) is the number of times γ winds about 0.
Moreover, if γ : [a, b] → C is a closed curve that does not pass through w, then
the number of times that γ winds about w should be the same as the number of
times that γ − w winds about 0. This, together with the fact that log(z − w) is the
integral of 1/(z − w), inspires the following definition.
Definition 26.3. The winding number of a closed curve γ about a point w is
defined to be Z
1 1
dz.
2πi γ z − w
Now we see how to count zeros and poles inside a simple closed curve.
Theorem 26.4 (Cauchy’s argument principle). Suppose that f is meromorphic in
the simply connected domain Ω, with zeros of order mj at the points aj and poles of
order nk at the points bk . Suppose also that Γ is a simple closed contour in Ω that
does not pass through any zero or pole of f . Then
Z ′ X X
1 f (z)
dz = mj − nk . (26.1)
2πi Γ f (z)
j: aj ∈Int(Γ) k: bk ∈Int(γ)

157
158 26. THE THEORY OF FUNCTIONS

Proof. From Cauchy’s residue theorem, the left hand side of the equality above
is equal to the sum of the residues of f ′ /f at its singularities inside γ. The function
f ′ /f has a singularity at w if and only if w is a zero or a pole of f , so it suffices to
compute the residues at these points.
Suppose that f has a zero or a pole at w. Then, from the lecture on singularities,
we know that there exist ε ∈ R+ and m ∈ Z such that f (z) = (z − w)m g(z), where
g, 1/g ∈ H(B(w, ε)). If w is a zero, then the order of the zero is m, while if w is a
pole, then the order of the pole is −m. Hence
f ′ (z) m(z − w)m−1 g(z) + (z − w)m g ′ (z) m g(z) + (z − w) g ′ (z)
= = .
f (z) (z − w)m g(z) (z − w) g(z)
Then f ′ /f has a simple pole at w, and
m g(z) + (z − w) g ′ (z)
Res(f ′ /f, w) = lim (z − w)
z→w (z − w) g(z)
m g(z) + (z − w) g ′ (z)
= lim
z→w g(z)
= m. □
This result has been used to provide experimental confirmation of the Riemann
hypothesis, which states that the zeros of the Riemann zeta function lie on the line
{z ∈ C : Re(z) = 21 }. We will discuss this later.
Theorem 26.4 may also be used to show that functions have zeros in certain
regions.
Corollary 26.5. Suppose that Ω is a domain, that f ∈ H(Ω), that γ : [a, b] → Ω
is a simple closed curve, and that f (γ(t)) 6= 0 for all t ∈ [a, b]. Then the number of
zeros of f in Int(γ) (counting multiplicities) is equal to the number of times f ◦ γ
winds around 0.
Proof. Let δ = f ◦ γ. Then δ : [a, b] → C is also a closed curve, though not
necessarily simple. Further, δ ′ (t) = f ′ (γ(t)) γ ′ (t), and so
Z ′ Z b ′ Z b ′ Z
f (z) f (γ(t)) ′ δ (t) 1
dz = γ (t) dt = dt = dz. □
γ f (z) a f (γ(t)) a δ(t) δ z

2. Rouché’s theorem
The next result gives us a useful tool for counting zeros.
Theorem 26.6 (Rouché’s theorem). Suppose that γ : [a, b] → Ω is a closed curve
in a simply connected domain Ω, that f, g ∈ H(Ω), and that |f (z)| < |g(z)| for all
z ∈ Range(γ). Then the number of zeros of f + g inside γ is equal to the number of
zeros of g inside γ.

Proof. From the hypotheses, no zero of f + g or g lies on γ; indeed, if t ∈ [a, b],

then
|(f + g)(γ(t))| = |f (γ(t)) + g(γ(t))| ≥ |g(γ(t))| − |f (γ(t))| > 0
and
|g(γ(t))| > |f (γ(t))| ≥ 0.
 3. MORE CONSEQUENCES OF ROUCHÉ’S THEOREM 159

We need to show that the winding number of (f +g)◦γ is the same as the winding
number of g ◦ γ. We choose the argument arg(g(γ(t)) so that it varies continuously
with t. Since |f (γ(t))| < |g(γ(t))|, we may choose arg(f (γ(t)) + g(γ(t))) in such a
way that
π
|arg((f + g)(γ(t))) − arg(g(γ(t)))| < ;
2
when we do this, arg((f + g)(γ(t))) varies continuously with t, and
|arg((f + g)(γ(b))) − arg((f + g)(γ(a))) − arg(g(γ(b))) + arg(g(γ(a)))| < π .
Since the left-hand side is equal to 2πk for some integer k, it must be 0, and so
arg((f + g)(γ(b))) − arg((f + g)(γ(a))) = arg(g(γ(b))) − arg(g(γ(a))),
which means that the winding numbers are equal. □
Here is a corollary.
P P
Corollary 26.7. Suppose that p(z) = nj=0 cj z j , and that |cm | > j̸=m |cj |.
Then p has m zeros (counting multiplicities) inside the unit circle.
P
Proof. Take γ(t) = eit , where 0 ≤ t ≤ 2π, and define f (z) = j̸=m cj z j and
g(z) = cm z m . The hypothesis on the coefficients of p means that |f (eit )| < |g(eit )|,
and by Rouché’s theorem, (f + g) has the same number of zeros inside γ as g, that
is, m zeros. □
Exercise 26.8. Suppose that n ≥ 3, and let p(z) = 2z n + z 2 − 3z − 1. Using
Rouché’s theorem, show that all the zeros of p lie inside the circle of radius 2.
Answer. Take g(z) = 2z n , and f (z) = z 2 − 3z − 1; then f + g = p. If |z| = 2,
then |g(z)| = 2n+1 ≥ 16, while |f (z)| ≤ 4 + 6 + 1, by the triangle inequality. By
Rouché’s theorem, the number of zeros of p inside the circle of radius 2 is equal to
the number of zeros of g inside the circle of radius 2, that is, n.
Since p has n zeros, all the zeros of p are inside the circle of radius 2. 4

3. More consequences of Rouché’s theorem

Rouché’s theorem has many other corollaries that help us understand the be-
haviour of holomorphic functions.
Theorem 26.9. Suppose that R > 0, and that
X∞
h(z) = an (z − z0 )n in B(z0 , R).
n=0

Let M be the smallest positive integer for which am 6= 0. Then there exist δ, ε ∈ R+
such that for all w ∈ B ◦ (a0 , ε), there are exactly M points z in B ◦ (z0 , δ) such that
f (z) = w (counting multiplicities).
The proof of this result may be found at the end of this lecture.
P
Remark 26.10. Suppose that f (z) = ∞ n=M an (z − z0 ) in B(z0 , r), where aM 6=
n

0. We have already remarked that f (z) “behaves like aM (z − z0 )M near z0 ”, in the

sense that
f (z)
lim = 1.
z→z0 aM (z − z0 )M
160 26. THE THEORY OF FUNCTIONS

Another way in which this assertion is true is that both f and the leading term of
the Taylor series are both M -to-1 near z0 .
In the case where a1 6= 0, Theorem 26.9 implies that f is one-to-one near z0 .
This is the hard part of the inverse function theorem.
Corollary 26.11 (Inverse function theorem). Suppose that f is a holomorphic
function in Ω, that f (z0 ) = w0 , and that f ′ (z0 ) 6= 0. Then there is an open subset
Υ of Ω, containing z0 , and an open ball B containing w0 such that the restriction of
f to Υ is a bijection from Υ onto B, and the inverse function F −1 from B to Υ is
holomorphic. Further,
1
(f −1 )′ (w0 ) = ′ .
f (z0 )
Proof. By Theorem 26.9 above, there exist δ, ε ∈ R+ such that for every point
w ∈ B(w0 , ε), there exists exactly one point z in B(z0 , δ) such that f (z) = w. We
define f −1 (w) = z.
It may be shown (but we do not do so here) that a continuous bijection on
a compact set has a continuous inverse. From this result, it follows that f −1 is
continuous, and so f −1 (w) → f −1 (w0 ) as w → w0 .
Now, taking w = f (z) and w0 = f (z0 ), we see that
 −1
f −1 (w) − f −1 (w0 ) w − w0
lim = −1 lim−1
w→w0 w − w0 f (w)→f (w0 ) f −1 (w) − f −1 (w0 )
 −1
f (z) − f (z0 )
= lim = f ′ (z0 )−1 ,
z→z0 z − z0
which shows both that the derivative exists and finds it. □
Corollary 26.12. Suppose that Ω and Υ are domains, and f is a holomorphic
and bijective function from Ω to Υ. Then f ′ (z) 6= 0 for all z ∈ Ω, and hence the
inverse function f −1 is also holomorphic.
Proof. If f ′ (z0 ) = 0 for some z0 ∈ Ω, then f is not one-to-one near z0 , by
Theorem 26.9 above. This is a contradiction, and hence f ′ never vanishes in Ω.
Now the inverse function f −1 is holomorphic by the inverse function theorem. □
Theorem 26.13 (The open mapping theorem). Suppose that Ω is a domain in
C and f ∈ H(Ω) is nonconstant. Then Range(f ) is open in C.
Proof. Take w0 ∈ Range(f ); then w0 = f (z0 ) for some z0 ∈ Ω.
From the theorem above, there exists ε ∈ R+ such that every point in B(w0 , ε)
also lies in Range(f ). Hence Range(f ) is open. □
Exercise 26.14. Find an example of a holomorphic function defined on an open
set Ω that is not constant, but whose range is not open.
Answer. In light of the open mapping theorem, Ω cannot be connected. We
may take Ω to be B(0, 1) ∪ B(3, 1), and define
(
z when z ∈ B(0, 1)
f (z) =
3 when z ∈ B(3, 1).
Then the range of f is B(0, 1) ∪ {3}, which is not open. 4
 4. A PROOF† 161

4. A proof†
In this section, we prove Theorem 26.9, whose statement we first recall.
Theorem. Suppose that h is a nonconstant holomorphic function on B(z0 , R),
where R > 0, and that
X
∞
h(z) = an (z − z0 )n in B(z0 , R).
n=0

Let M be the smallest positive integer for which am 6= 0. Then there exist δ, ε ∈ R+
such that for all w ∈ B ◦ (a0 , ε), there are exactly M points z in B ◦ (z0 , δ) such that
f (z) = w (counting multiplicities).
Proof† . Since h is not constant, it is not possible that an = 0 for all positive
n. Thus M exists. Note that a0 need not be 0.
We may write
X∞ X
∞
′
h(z) − a0 = an (z − z0 ) and h (z) =
n
nan (z − z0 )n−1
n=M n=M

in B(z0 , R). In the lecture on Taylor series, we showed that

h′ (z)
lim = 1.
z→z0 M aM (z − z0 )M −1

Thus there exists δ1 ∈ R+ such that

h′ (z) 1
−1 <
M aM (z − z0 ) M −1 2
for all z ∈ B(z0 , δ1 ) \ {z0 }; this implies that h′ (z) 6= 0 for all such z. Similarly,
h(z) − a0
lim = 1,
z→z0 aM (z − z0 )M

and there exists δ0 ∈ R+ such that h(z) − a0 6= 0 for all z ∈ B(z0 , δ0 ) \ {z0 }. Take
δ = min{δ0 , δ1 }.
The function |h(z) − a0 | is continuous and nonvanishing on ∂B(z0 , δ). Define
ε = min{|h(z) − a0 | : z ∈ ∂B(z0 , δ)}; then ε > 0, by compactness.
Suppose that w ∈ B ◦ (a0 , ε). Take f (z) = a0 − w, and g(z) = h(z) − a0 ; then
|f (z)| < ε ≤ |g(z)| for all z ∈ ∂B(z0 , δ). Rouché’s Theorem implies that f + g and g
have the same number of zeros in B(z0 , δ), counting multiplicities. But z0 is a zero
of multiplicity M for g, and so f + g has M zeros. The derivative of f + g does not
vanish in B ◦ (z0 , δ), and so these zeros all have order one. Since (f +g)(z) = h(z)−w,
there are M distinct points in B(z0 , δ) such that h(z) = w. □
 LECTURE 27

The Riemann zeta function

In this lecture, we show how residue calculus can be used to evaluate sums. We
also introduce the Riemann ζ function, a focus of current mathematical research.

1. Computing sums
We begin with an exercise.
Exercise 27.1. Denote by ΓN the perimeter of the square with vertices at the
points ±(N + 1/2) ± i(N + 1/2), where N ∈ N, as in Figure 27.1, and define
cot(πz)
f (z) = .
z2

(a) Find the singularities of f inside ΓN , classify them, and compute the
residues of f at the singularities.
(b) Show that
| cot(±(N + 1/2)π + iy)| ≤ 1 ∀y ∈ R,
and
| cot(x ± i(N + 1/2)π)| ≤ coth(π/2) ∀x ∈ R.
(c) Deduce that if Λ is a vertical side and ∆ is a horizontal side of ΓN , then
Z Z
4 4 coth(π/2)
f (z) dz ≤ and f (z) dz ≤ .
Λ 2N + 1 ∆ 2N + 1
X∞
1
(d) Hence find .
n=1
n2

Im
ΓN (N + 1/2) + i(N + 1/2)

Re
+
+
+
+
+
+
+
+
+
+
+

−(N + 1/2) − i(N + 1/2)

Figure 27.1. A contour

163
164 27. THE RIEMANN ZETA FUNCTION

Answer. (a) The function f has singularities when z = 0 and when sin(πz) =
0, that is, when z ∈ Z. By L’Hôpital’s rule,
(z − n) cos(πz) 1
lim (z − n) cot(πz) = lim = 6= 0
z→n z→n sin(πz) π
for every integer n, and we deduce that cot(πz) has a simple pole at every integer
n ∈ Z. Moreover z 7→ z 2 has a zero of order 2 at 0, and it follows that the
singularities of f are at the points n, where n ∈ Z; all the singularities are poles;
the poles are simple except at 0, where the pole is of order 3.
We compute the residues using the appropriate formulae: if n ∈ Z \ {0}, then
cot(πz) 1
Res(f, n) = lim (z − n)f (z) = lim (z − n) 2
= .
z→n z→n z πn2
Further,
1 d2 3 1 d2
Res(f, 0) = lim z f (z) = lim z cot(πz)
z→0 2! dz 2 z→0 2 dz 2
1 d d2 
= lim 2 cot(πz) + z 2 cot(πz) .
z→0 2 dz dz
Now
d
cot(πz) = −π cosec2 (πz) ,
dz
and
d2
cot(πz) = 2π 2 cot(πz) cosec(πz).
dz 2
By l’Hôpital’s rule,
d z d2
Res(f, 0) = lim cot(πz) + cot(πz)
z→0 dz 2 dz 2
−π π 2 cos(πz)
= lim + z
z→0 sin2 (πz) sin3 (πz)
−π sin(πz) + π 2 z cos(πz)
= lim
z→0 sin3 (πz)
−π sin(πz) + π 2 z cos(πz) π 3 z 3
= lim
z→0 π3z3 sin3 (πz)
−π sin(πz) + π 2 z cos(πz)
= lim
z→0 π3z3
−π cos(πz) + π 2 cos(πz) − π 3 z sin(πz)
2
= lim
z→0 3π 3 z 2
sin(πz) π
= lim − =− .
z→0 3z 3
(b) By the addition formula for the sine and cosine,
cos(x + iy)
cot(x + iy) =
sin(x + iy)
cos(x) cosh(y) − i sin(x) sinh(y)
=
sin(x) cosh(y) + i cos(x) sinh(y)
 1. COMPUTING SUMS 165

Taking x = ±(N + 1/2)π and y ∈ R, then

|cot(±(N + 1/2)π + iy)|
cos((N + 1/2)π) cosh(y) − i sin((N + 1/2)π) sinh(y)
=
sin((N + 1/2)π) cosh(y) + i cos((N + 1/2)π) sinh(y)
sinh(y)
= ≤ 1,
cosh(y)

Further, taking x ∈ R and y = ±(N + 1/2)π, we see that

|cot(x ± i(N + 1/2)π)|2
2
cos(x) cosh((N + 1/2)π) ∓ i sin(x) sinh((N + 1/2)π)
=
sin(x) cosh((N + 1/2)π) ± i cos(x) sinh((N + 1/2)π)
cos2 (x) cosh2 ((N + 1/2)π) + sin2 (x) sinh2 ((N + 1/2)π)
=
sin2 (x) cosh2 ((N + 1/2)π) + cos2 (x) sinh2 ((N + 1/2)π)
cos2 (x) cosh2 ((N + 1/2)π) + sin2 (x) cosh2 ((N + 1/2)π)
≤
sin2 (x) sinh2 ((N + 1/2)π) + cos2 (x) sinh2 ((N + 1/2)π)
cosh2 ((N + 1/2)π)
= 2 = coth2 ((N + 1/2)π)
sinh ((N + 1/2)π)
≤ coth2 ((1/2)π).
as required.
(c) If z ∈ Λ, then
coth(z) 1 4
|f (z)| = ≤ ≤ .
z2 |z|2 (2N + 1)2
Similarly, if z ∈ ∆, then
coth(π/2) 4 coth(π/2)
|f (z)| ≤ ≤ .
|z| 2 (2N + 1)2
By the M L lemma,
Z
4 4
f (z) dz ≤ 2
(2N + 1) = ,
Λ (2N + 1) 2N + 1
and
Z
4 coth(π/2) 4 coth(π/2)
f (z) dz ≤ 2
(2N + 1) = ,
∆ (2N + 1) 2N + 1
as required.

(d) By the residue theorem,

Z  π X
X 2 
N
f (z) dz = 2πi Res(f, n) = 2πi − + .
ΓN n∈Z
3 n=1 πn2
|n|≤N
166 27. THE RIEMANN ZETA FUNCTION

By the estimates that we have proved for the integrals along the sides of ΓN ,
 π 2 X Z
1
N
16 coth(π/2)
2πi − + = f (z) dz ≤ .
3 π n=1 n2 ΓN 2N + 1

As N → ∞, the right hand side tends to 0, so the left hand side does too. Thus

π 2 X 1
N
− + → 0,
3 π n=1 n2
or in other words,
X∞
1 XN
1 π2
= lim = .
n=1
n2 N →∞ n=1 n2 6
4

X∞
1
Exercise 27.2. Compute 2k
, where k = 2, 3, 4, . . . .
n=1
n

2. The Riemann zeta function

Exercise 27.3. Suppose that s ∈ C and s = σ + it, where σ > 1. Show that
X∞
1 X∞
ln n
<∞ and < ∞.
n=1
ns n=1
ns

Answer. Observe that ns = exp((σ + it) log(n)), so

ns = | exp(σ log(n))| | exp(it log(n))| = | exp(σ log(n))| = nσ .
Then
X∞
1 X∞
1
s
= .
n=1
n n=1
nσ
From the p test for the convergence of a sum, we know that if p > 1, then
X∞
1
< ∞.
n=1
np

Replacing p by σ gives the desired result for the first sum. The second one is treated
similarly. 4

The Riemann zeta function is defined by an infinite sum.

Definition 27.4. The function ζ is given by
X∞
1
ζ(s) = .
n=1
ns

This sum converges absolutely when Re(s) > 1, and continues meromorphically into
C.
 2. THE RIEMANN ZETA FUNCTION 167

It may be shown that ζ is holomorphic in the half plane {s ∈ C : Re s > 1}, and
X
∞
ln n
′
ζ (s) = − .
n=1
ns

The argument is similar to that needed to show that we may differentiate inside an
integral, which we will see shortly.

2.1. The Euler product formula. The zeta function is connected to prime
numbers via the Euler product formula:
Y
ζ(s) = (1 − p−s )−1 .
p

Indeed, when Re(s) > 1, we see that

(1 − p−s )−1 = 1 + p−s + p−2s + p−3s + . . . ,
and the summands in the expanded product are therefore all products over all primes
of the form p−k
1
1 s −k2 s −k3 s
p2 p3 . . . . Each positive integer n has a unique factorisation
k1 k 2 k 3
of the form p1 p2 p3 . . . (in which most of the powers k are 0), so the summands in
the expanded product are therefore the powers n−s , for all positive integer n.

2.2. The meromorphic continuation. Riemann showed that ζ extends holo-

morphically to C\{1}, and has a simple pole at 1. The first step in this meromorphic
continuation is to show that
∞  
1 X n n−s
ζ(s) = − (27.1)
s − 1 n=1 (n + 1)s ns

when Re s > 0 (and s 6= 1). To do this, we observe first that when Re(s) > 2, then
X∞  
n n−s
s
−
n=1
(n + 1) ns
X∞
n+1 X
∞
1 X
∞
n X s
∞
= − − +
n=1
(n + 1)s n=1 (n + 1)s n=1 ns n=1 ns
= ζ(s − 1) − 1 − ζ(s) + 1 − ζ(s − 1) + s ζ(s)
= (s − 1)ζ(s).
However, it may be seen that the sum on the right hand side of (27.1) converges
when Re(s) > 0, and then analytic continuation (or more precisely meromorphic
continuation) gives the desired result.
We may find other ways to rewrite (s − 1)ζ(s) that converge when Re(s) > −1,
or when Re(s) > −2, and so on.

2.3. More on the zeta function. It is possible to express the zeta function
as an integral:
Z ∞ s−1
1 x
ζ(s) = dx.
Γ(s) 0 ex − 1
168 27. THE RIEMANN ZETA FUNCTION

To see this, assume that Re(s) > 1, and write

Z ∞ s−1 Z ∞ Z ∞X ∞
x xs−1 xs−1 e−nx
dx = dx = dx
0 ex − 1 0 ex (1 − e−x 0 n=0
e x

X∞ Z ∞ X∞
s−1 −(n+1)x 1
= x e dx = Γ(s) s
.
n=0 0 n=0
(n + 1)
Taking the infinite summation out of the integral requires exchanging the order of
a limit and an integral.
Another interesting formula concerning the zeta function is the following func-
tional equation  πs 
s s−1
ζ(s) = 2 π sin Γ(1 − s) ζ(1 − s).
2
2.4. The Riemann hypothesis. Riemann hypothesised that if ζ(z) = 0, then
either Re(z) = 1/2 or z = −1, −3, −5, . . . . The so-called Riemann hypothesis is
one of the key problems in pure mathematics today. The first person to prove—or
disprove—it will earn a prize of USD 1 million from the Clay Mathematics Institute.
For more information, see
http://www.claymath.org/millennium/.
It is known that, apart from the so-called trivial zeros at −1, −3, −5, …, all
the zeros lie inside the strip {z ∈ C : 0 < Re(z) < 1}, and that the zeta function
has infinitely many zeros on the line {z ∈ C : Re(z) = 1/2}. It can be shown by
numerical computation that all the nontrivial zeros z such that | Im(z)| ≤ 10N lie
on the line for some large N (in fact, finding zeros numerically is one of the test
problems used by computer scientists to demonstrate their virtuosity).
To do this, one takes a rectangular contour which surrounds a segment of the
vertical axis, above the horizontal axis; the vertical sides are where Re(z) = 0 or
Re(z) = 1. Then there are no poles inside the contour; there are formulae which
give the number of zeros on the axis. It is possible to compute the integral on the
left hand side numerically, because the value must be an integer, and if the error is
strictly less than 21 , then the real value must be the integer closest to the numerical
value. If the number of zeros inside the contour is equal to the predicted number of
zeros on the line {z ∈ C : Re(z) = 12 }, then there cannot be any zeros not on the
line inside the contour.
In some sense, the Riemann hypothesis is equivalent to “randomness” of prime
numbers. For instance it tells us that the number of primes in a large interval of
integers can be given by a formula which amounts to saying that the probability of
a large number n being prime is about 1/ log(n). If we could compute enough prime
numbers, then we could test this computationally. However, it would require us to
compute more prime numbers than there are atoms in the universe, so we would
never be able to store all the necessary data.
The Riemann hypothesis is also of interest in theoretical physics, as the distri-
bution of the zeros of a random matrix are related to those of the Riemann zeta
function.
 LECTURE 28

Interlude: Integrals†

In this lecture, we consider integrals in which the integrand contains a parameter,

and consider the question of whether
Z ∞ Z ∞
lim F (s, t) ds = lim F (s, t) ds. (28.1)
t→t0 −∞ −∞ t→t0

In our theorem, we suppose that F is defined on R × P , where the parameter space

P is either R or C. To cover both cases, let us agree that the notation B(t, r)
means the open interval {x ∈ R : |x − t| < r} when P is R, and the open ball
{z ∈ C : |z − t| < r} when P is C.
We will often write
Z Z ∞
|f (x)| dx rather than |f (x)| dx,
R −∞
Z Z R
|f (x)| dx rather than |f (x)| dx,
[−R,R] −R

and
Z Z −R Z ∞
|f (x)| dx rather than |f (x)| dx + |f (x)| dx.
R\[−R,R] −∞ R

So when the improper integral exists,

Z
|f (x)| dx → 0 as R → ∞.
R\[−R,R]

Other obvious variations of this notation will also be used.

1. An example
Just to be sure that there is a real issue to discuss, we consider an example.

Exercise 28.1. Define the function F on R × (0, ∞) by

F (s, t) = t−3/2 s2 exp(−s2 /t).

Find limt→0+ F (s, t) and limt→+∞ F (s, t), and decide whether or not (28.1) holds
when t0 = 0 and when t = ∞.

Answer. Fix s ∈ R. If s = 0, then F (s, t) = 0 for all t, so limt→0+ F (s, t) = 0

and limt→+∞ F (s, t) = 0. Otherwise, by the change of variables u = t−1 , and
169
170 28. INTERLUDE: INTEGRALS†

l’Hôpital’s rule, applied twice, we see that

s2 u3/2
lim F (s, t) = lim t−3/2 s2 exp(−s2 /t) = lim
t→0+ t→0+ u→+∞ exp(us2 )

3 s2 u1/2 3 u1/2
= lim = lim
u→+∞ 2 s2 exp(us2 ) u→+∞ 2 exp(us2 )

3 u−1/2
= lim = 0.
u→+∞ 4 s2 exp(us2 )

Moreover, exp(−s2 /t) → 1 as t → +∞, and so

lim F (s, t) = lim t−3/2 s2 exp(−s2 /t) = 0 × s2 × 1 = 0.
t→+∞ t→+∞

Therefore, on the one hand,

Z ∞ Z ∞
lim F (s, t) ds = 0 ds = 0
−∞ t→0+ −∞

and Z ∞ Z ∞
lim F (s, t) ds = 0 ds = 0.
−∞ t→+∞ −∞

On the other hand, for any fixed t ∈ R+ , by setting u = t−1/2 s, we see that
Z ∞ Z ∞ Z ∞ √
−3/2 2 2 1/2 2 2 π
F (s, t) ds = t tu exp(−u )t du = u exp(−u ) du = ,
−∞ −∞ −∞ 2
which is independent of t, and so
Z ∞ √
π
lim F (s, t) ds =
t→0+ −∞ 2
and Z √
∞
π
lim . F (s, t) ds =
t→+∞ −∞ 2
Thus (28.1) does not hold when t0 = 0 and does not hold when t0 = +∞. 4
It is a good idea to draw the graphs of the function s 7→ F (s, t) for different
values of t to try to understand what is going on in the exercise above.

2. A theorem on limits
Now we prove a theorem giving sufficient conditions for the validity of exchanging
the orders of limits and integrals. In this theorem P may be R or C.
Theorem 28.2. Suppose that F : R × P → R and t0 ∈ P . Suppose also that
there exists µ ∈ R+ such that
(a) F is continuous on R × B(t0 , µ), and
(b) there is a nonnegative function M ∈ L1 (R) such that
|F (s, t)| ≤ M (s) ∀s ∈ R ∀t ∈ B(t0 , µ).
Then Z Z Z
lim F (s, t) ds = lim F (s, t) ds = F (s, t0 ) ds.
t→t0 R R t→t0 R
 2. A THEOREM ON LIMITS 171

Proof. Suppose that ε ∈ R+ . We need to find η ∈ R+ such that

Z Z
F (s, t) ds − F (s, t0 ) ds < ε
R R
whenever t ∈ B(t0 , η).
We will break the difference into two parts and make each part less than ε/2:
Z Z
F (s, t) ds − F (s, t0 ) ds
R R
Z
= (F (s, t) − F (s, t0 )) ds
R
Z (28.2)
≤ |F (s, t) − F (s, t0 )| ds
R
Z Z
= |F (s, t) − F (s, t0 )| ds + |F (s, t) − F (s, t0 )| ds.
R\[−R,R] [−R,R]

We make the first integral (which actually has two parts) small by making R big
and using condition (b), and we make the second integral small by using condition
(a).
To make the first integral small, recall that |F (s, t)| ≤ M (s) when t ∈ B(t0 , µ).
Hence, for such t,
Z Z
|F (s, t) − F (s, t0 )| ds ≤ |F (s, t)| + |F (s, t0 )| ds
R\[−R,R] R\[−R,R]
Z
≤2 M (s) ds.
R\[−R,R]
R R
Since the improper integral R M (s) ds converges, we may make R\[−R,R]
M (s) ds
small by taking R large enough. Take R such that
Z
ε
M (s) ds < .
R\[−R,R] 4
Then
Z
ε
|F (s, t) − F (s, t0 )| ds < . (28.3)
R\[−R,R] 2
Further, from (b), F is uniformly continuous on compact subsets of R × B(t0 , µ).
Hence there exists δ ∈ R+ such that
ε
|F (s, t) − F (s, t0 )| < ∀s ∈ [−R, R]
4R
provided that t ∈ B(t0 , δ). For such t,
Z Z
ε ε
|F (s, t) − F (s, t0 )| ds ≤ ds = . (28.4)
[−R,R] [−R,R] 4R 2
Set η = min{δ, µ}. Putting everything together, (28.3) and (28.4) hold when
t ∈ B(t0 , η). Combining with (28.2), we see that
Z Z
ε ε
F (s, t) ds − F (s, t0 ) ds < + = ε
R R 2 2
when t ∈ B(t0 , η), as required. □
172 28. INTERLUDE: INTEGRALS†

3. Remarks
Though we state our results for real-valued integrands F , they also hold for
complex-valued integrands, by an identical argument.
The arguments also apply when the parameter t is constrained to lie in a subset
of R or C; we just have to add this constraint to the statement of the theorem, and
to each step of the proof. By replacing t by 1/t, we can also establish a version of
the theorem for limits as t tends to ∞.
Our arguments also apply to integrals where one or both the limits of integration
is finite, that is, to integrals such as
Z
H(s, t) ds;
[a,b]

in this case, the proof of Theorem 28.2 simplifies, as we replace [−R, R] by [a, b] and
do not need to consider the other component of the integral.
Finally, the result also extends to integration in Rn ; we replace the integrals over
[−R, R] and over R \ [−R, R] by integrals over B(0, R) and over R \ B(0, R).

4. Exchanging integration and differentiation

Now we consider integrals in which the integrand contains a parameter, and
consider the question of whether
Z ∞ Z ∞
∂ ∂
F (s, t) ds t=t0 = F (s, t) t=t0 ds. (28.5)
∂t −∞ −∞ ∂t

Just to be sure that there is a real issue to discuss, we consider an example.

Exercise 28.3. Define F : R × R → R by
 3
 st if (s, t) 6= (0, 0)
F (s, t) = (s2 + t2 )2

0 if (s, t) = (0, 0).
Z 1 Z
∂ d 1
Compute F (s, t) t=0 ds and F (s, t) ds t=0 , and show that they are not
0 ∂t dt 0
equal.

Answer. We omit the details. 4

Finally, we state a theorem giving sufficient conditions for the validity of ex-
changing the order of differentiation and integration.
Theorem 28.4. Suppose that G : R × P → R and t0 ∈ P , where P ⊆ R, and
that the function s 7→ G(s, t0 ) ∈ L1 (R). Suppose also that there exists µ ∈ R+ such
that
(a) (s, t) 7→ ∂t
∂
G(s, t) is continuous on R × B(t0 , µ), and
(b) there is a nonnegative function M ∈ L1 (R) such that
∂
G(s, t) ≤ M (s) ∀s ∈ R ∀t ∈ B(t0 , µ).
∂t
 5. APPLICATIONS† 173

Then Z Z
∞ ∞
∂ ∂
G(s, t) ds t=t0
= G(s, t) t=t0
ds. (28.6)
∂t −∞ −∞ ∂t
Proof. The idea of the proof is to write
Z ∞ Z ∞
∂
G(s, t) ds = lim F (s, h) ds,
∂t −∞ t=t0 h→0 −∞

where  Z t0 +h

1 ∂
 G(s, t) dt if h 6= 0
F (s, h) = h t 0
∂t


 ∂ G(s, t) if h = 0.
∂t t=t0

Then we show that F satisfies the hypotheses of the previous theorem (where we
suppose that h → 0). It follows that
Z ∞ Z ∞
∂
G(s, t) ds = lim F (s, h) ds
∂t −∞ t=t0 h→0 −∞
Z ∞
= lim F (s, h) ds
−∞ h→0
Z ∞
∂
= G(s, t0 ) ds,
−∞ ∂t

as required. □
Remark 28.5. This result also applies when t is a complex variable, and differ-
entiation with respect to t is complex differentiation.
Differentiation with respect to a parameter may be used to compute integrals.
For instance, to compute
Z 1 2
x −1
dx,
0 ln x
we may find
Z Z 1 t
d 1 xt − 1 x −1
dx and dx t=0 .
dt 0 ln x 0 ln x

5. Applications†
The gamma function Γ is defined by an integral.
Definition 28.6. The function Γ is given by
Z ∞
Γ(z) = sz−1 e−s ds.
0
This integral converges absolutely when Re z > 0, and continues meromorphically
into C.
It follows from Theorem 28.4 that the gamma function is holomorphic, and
Z ∞
′
Γ (z) = sz−1 ln(s) e−s ds.
0
174 28. INTERLUDE: INTEGRALS†

It is easy to check that that Γ(z) = (z−1)Γ(z−1), by integration by parts, and hence
Γ(n) = (n − 1)! when n ∈ Z+ . This formula allows to extend Γ meromorphically
into C.
Now we consider some integrals that may be evaluated in terms of this function.
Exercise 28.7. Find changes of variables or other transformations such that
the following integrals may be converted into an integral of the form
Z π/2
cosa (θ) sinb (θ) dθ.
0
Z 1 Z 1
(a) (1 − x ) x dx;
2 α β
(b) (1 − y)α y β dy;
Z 0 π/2 0
tanc (φ)
(c) dφ;
Z0 ∞ secd (φ) Z ∞
tγ sε
(d) dt; (e) ds.
0 (1 + t2 )δ 0 (1 + s)ζ
Answer. Omitted 4
Observe that
Z ∞ Z ∞
a −x2 ′
x e dx = 1
2
(x′ )(a−1)/2 e−x dx′ = 12 Γ( a+1
2
);
0 0
similarly, Z ∞
y b e−y dy = 12 Γ( b+1
2
2
),
0
and Z ∞
ra+b+1 e−r dr = 12 Γ( a+b+2
2
2
).
0
By using polar coordinates, we see that
Z ∞ Z ∞
a −x2
y b e−y dy
2
1 a+1 1 b+1
2
Γ( 2 ) 2 Γ( 2 ) = x e dx
Z0 ∞ Z ∞ 0

xa y b e−x −y dx dy
2 2
=
0 0
Z π/2 Z ∞
ra cosa (θ) rb sinb (θ) e−r r dr dθ
2
=
0 0
Z ∞ Z π/2
a+b+1 −r2
= r e dr cosa (θ) sinb (θ) dθ
0 0
Z π/2
= 12 Γ( a+b+2
2
) cosa (θ) sinb (θ) dθ.
0
Hence Z π/2
Γ( a+1 )Γ( b+1 )
cosa (θ) sinb (θ) dθ = 2
a+1
2
b+1
.
0 2Γ( 2 + 2 )
 LECTURE 29

The Fourier transformation

In this lecture we introduce the Fourier transform fb of an integrable function

f on R; we then compute some examples. We will assume that it is legitimate to
exchange limits and integrals; in a future lecture, we will examine whether this is
really so.

1. Locally integrable and integrable functions on R

We say that a function f : R → C is locally integrable if f is Riemann inte-
grable on all finite intervals [−R, R]. For example, continuous functions are locally
integrable.
We write L1 (R) for the collection of all locally integrable functions f : R → C
such that the improper integral
Z ∞ Z
|f (x)| dx = lim |f (x)| dx
−∞ R→∞ [−R,R]
RR
converges (and is finite). Note that −R |f (x)| dx increases as R increases, so the
limit exists (and is finite) if and only if the integrals are bounded, uniformly in R.
1
Example 29.1. Define the Gaussian φ : R → C by φ(x) = √ e−x /2 for all
2

R 2π
x ∈ R. Then φ ∈ L1 (R) and R φ(x) dx = 1.

2. The Fourier transform

Definition 29.2. Suppose that f ∈ L1 (R). The Fourier transform of f is the
function fb : R → C given by
Z Z
b
f (ξ) = f (x) e −ixξ
dx = lim f (x) e−ixξ dx.
R R→∞ [−R,R]

The convergence of the improper integral follows from that fact that | e−ixξ | = 1
for all x, ξ ∈ R, and properties of the Riemann integral.

3. Examples
Exercise 29.3. Define f : R → R by f (x) = e−|x| . Show that
2
fb(ξ) = ∀ξ ∈ R.
1 + ξ2
Answer. First of all, it is easy to check that f ∈ L1 (R). By definition,
Z
fb(ξ) = lim e−|x| e−ixξ dx.
R→∞ [−R,R]
175
176 29. THE FOURIER TRANSFORMATION

Now, Z Z Z
−|x| −ixξ x −ixξ
e e dx = e e dx + e−x e−ixξ dx
[−R,R] [−R,0] [0,R]
Z 0 Z R
= x(1−iξ)
e dx + e−x(1+iξ) dx
−R 0
−R(1−iξ)
1 e e−R(1+iξ) 1
− = − + ,
1 − iξ 1 − iξ 1 + iξ 1 + iξ
and since | e | = 1, it follows that
iRξ

e−R(1−iξ) e−R(1+iξ) e−R

= =p →0
1 − iξ 1 + iξ 1 + ξ2
as R → ∞, so
1 1 2
fb(ξ) = + = ,
1 − iξ 1 + iξ 1 + ξ2
as required. 4

Exercise 29.4. Define f : R → R by f (x) = 1/(1 + x2 ). Show that

fb(ξ) = πe−|ξ| ∀ξ ∈ R.
Answer. We saw in Lecture 24 that
Z ∞ iξx
e
2+1
dx = πe−ξ
−∞ x
when ξ ≥ 0. Further, regardless of the sign of ξ,
Z ∞ −iξx Z ∞ iξx
e e
2
dx = 2
dx.
−∞ x + 1 −∞ x + 1

Hence Z Z
∞ ∞
e−iξx eiξx
dx = dx
−∞ x2 + 1 −∞ x2 + 1
Z ∞
ei|ξ|x
= dx
−∞ x2 + 1
−|ξ|
= πe ,
as required. 4

1
Exercise 29.5. The Gaussian φ is defined by φ(x) = √ e−x /2 . Show that
2

2π
b = e−ξ
2 /2
φ(ξ) .
Answer. First, by Example 29.1,
Z ∞
φ(x) dx = 1. (29.1)
−∞

By completing the square, we see that

√ Z Z R
−x2 /2 −ixξ −ξ 2 /2
e−(x+iξ)
2 /2
b = lim
2π φ(ξ) e e dx = lim e dx. (29.2)
R→∞ [−R,R] R→∞ −R
 3. EXAMPLES 177

Since the function z 7→ e−z /2 is holomorphic, the Cauchy–Goursat Theorem, ap-

2

plied to the rectangular contour with vertices at −R, R, R + iξ and −R + iξ, implies
that
Z R Z R Z R+iξ Z −R+iξ
−(x+iξ)2 /2 −x2 /2 −z 2 /2
e−z /2 dz. (29.3)
2
e dx = e dx + e dz −
−R −R R −R

We now make two observations.

√
First, as R → ∞, the first term on the right hand side of (29.3) tends to 2π,
by (29.1) above.
Second, the terms
Z R+iξ Z −R+iξ
−z 2 /2
e−z /2 dz
2
e dz and
R −R

tend to 0 when R tends to infinity. We show how to handle the first term; the second
term may be handled similarly. We parametrise the line segment between R and
R + iξ by writing z = R + iy, where 0 ≤ y ≤ ξ. For such z,
e−z = e−(R+iy) = e−(R = e−(R
2 /2 2 /2 2 +2iRy−y 2 )/2 2 −y 2 )/2 2 −R2 2 −R2
= ey ≤ eξ .
By the M L Lemma,
Z R+iξ
e−z
2 /2 2 −R2
dz ≤ eξ |ξ| → 0 as R → ∞.
R

b =
Combining these two observations with (29.2) and (29.3), we deduce that φ(ξ)
−ξ 2 /2
e for all ξ ∈ R, as required. 4

Exercise 29.6. Let f (x) = 1 if x ∈ [−1, 1] and f (x) = 0 otherwise. Show that
fb ∈
/ L1 (R).

Answer. We compute that fb : ξ 7→ 2 sin(ξ)/ξ ∈

/ L1 (R). 4
The unfortunate fact that the Fourier transform of an integrable function need
not be integrable causes some problems. However, Fourier transforms of integrable
functions do have some good properties.
Lemma 29.7 (The Riemann–Lebesgue Lemma). If f ∈ L1 (R), then the function
fb is bounded and continuous, and vanishes at infinity.
Proof. For the boundedness, observe that
Z Z Z
b
f (ξ) = f (x) e −ixξ
dx ≤ f (x) e −ixξ
dx = | f (x)| dx < ∞,
R R R

since f ∈ L (R). Further, assuming that it is legitimate to exchange limits and

1

integrals, we see that

Z ∞ Z ∞
b
lim f (ξ) = lim f (x) e −ixξ
dx = lim f (x) e−ixξ dx
ξ→ξ0 ξ→ξ0 −∞ −∞ ξ→ξ 0
Z ∞
= f (x) e−ixξ0 dx = fb(ξ0 ),
−∞

and so f is continuous.
178 29. THE FOURIER TRANSFORMATION

To justify this last step, write F (x, ξ) = f (x) e−ixξ for all z, ξ ∈ R. Observe that
|F (x, ξ)| = |f (x)| ∈ L1 (R),
by hypothesis, and so condition (b) of Theorem 28.2 holds. Further, if f is contin-
uous, then F is continuous, and condition (a) of Theorem 28.2 holds.
We will not consider the vanishing at infinity. □

4. The inversion formula

Now we show how to recover the original function f from fb. In other words, we
find the inverse of the linear transformation F given by F(f ) = fb.
To simplify the statement of the inversion theorem, we make an additional defi-
nition.
Definition 29.8. We define M(R) to be the set of “moderately nice” functions
on R, that is, of continuous bounded functions whose absolute value is integrable.

Theorem 29.9 (The inversion formula for the Fourier transform). If f ∈ M(R)
and fb ∈ M(R), then Z
1
f (x) = fb(ξ) eixξ dξ.
2π R

Proof. We give the proof in the final section. □

The inversion formula implies that the Fourier transformation is bijective on the
vector space of functions on M(R) whose Fourier transform lies in M(R).
The inversion theorem does not√ hold without a continuity hypothesis. For ex-
ample, we have shown that φ b = 2πφ, where φ is the Gaussian. Now define
(
φ(x) if x 6= x0
f (x) =
0 if x = x0 .
Since changing the value of a function at a point does not change the integral of the
function, it follows that fb = φ, b so
Z Z
b
f (ξ) e iξx0
dξ = b eiξx0 dξ = φ(x0 ) 6= f (x0 ),
φ(ξ)
R R
and the inversion formula for f fails at x0 .
However, counterexamples of this rather artificial type are the only problem.
It may be shown that, if f ∈ L1 (R) and fb ∈ L1 (R), then f may be modified
on a “small set” in order to become continuous without changing fb, and then the
inversion formula holds.

5. Proof of the inversion formula†

This proof requires a preliminary lemma.
Lemma 29.10. Suppose that g, ψ ∈ L1 (R). Then
Z ∞ Z ∞
b
g(x) ψ(x) dx = gb(ξ) ψ(ξ) dξ.
−∞ −∞
 5. PROOF OF THE INVERSION FORMULA† 179

Proof. Observe that

Z Z Z  Z Z
−iξx
f (x) gb(x) dx = f (x) g(ξ) e dξ dx = f (x) g(ξ) e−iξx dξ dx
R
ZR Z R
Z ZR R 
−ixξ −ixξ
= f (x) e g(ξ) dx dξ = f (x) e dx g(ξ) dξ
ZR R R R

= fb(ξ) g(ξ) dξ,

R

as required. □

The change in the order of integration in the double integral is justified by the
fact that Z Z
|g(x)| |ψ(ξ)| dx dξ < ∞.
R R

And now we give the proof of the inversion formula.

Proof. Suppose that ψδ (x) = e−δ , where δ ∈ R+ . Then by a variation of

2 x2 /2

Exercise 29.5, we may show that

√
2π −ξ2 /2δ2
ψbδ (ξ) = e .
δ
By the change of variables x = δy and Lemma 29.10, if g, gb ∈ M(R), then
√
√ Z ∞ Z ∞
2π −x2 /2δ2
Z ∞
−y /2
g(x) ψbδ (x) dx
2
2π g(δy) e dy = g(x) e dx =
−∞ −∞ δ −∞
Z ∞ Z ∞
gb(ξ) e−δ ξ /2 dξ.
2 2
= gb(ξ) ψδ (ξ) dξ =
−∞ −∞

Now we claim that we may exchange limits and integrals to show that
√ Z ∞ −y 2 /2
√ Z ∞
lim g(δy) e−y /2 dy
2
lim 2π g(δy) e dy = 2π
δ→0 −∞ −∞ δ→0
√ Z ∞
g(0) e−y /2 dy = 2π g(0),
2
= 2π
−∞

since g is continuous, and

Z ∞ Z ∞ Z ∞
−δ 2 ξ 2 /2 −δ 2 ξ 2 /2
lim gb(ξ) e dξ = lim gb(ξ) e dξ = gb(ξ) dξ.
δ→0 −∞ −∞ δ→0 −∞

To justify the first of these, write F (y, δ) = g(δy) e−y /2 . Since g is continuous,
2

so is F , and condition (a) of Theorem 28.2 holds. Further, since g is bounded,

that is, there is a constant C such that |g(y)| ≤ C for all y ∈ R, it follows that
|F (y, δ)| ≤ C e−y /2 for all y, δ ∈ R, and condition (b) of Theorem 28.2 also holds.
2

Thus we may exchange the limit and the integral. The second formula may be
justified similarly.
Hence Z ∞
1
g(0) = gb(ξ) dξ. (29.4)
2π −∞
180 29. THE FOURIER TRANSFORMATION

Now suppose that x ∈ R, f ∈ M(R) and g(y) = f (y + x). Then, by the change
of variable u = y + x, we see that
Z Z ∞
−iyξ
gb(ξ) = g(y) e dy = f (y + x) e−iyξ dy
ZR∞ −∞
Z ∞
= f (u) e −i(u−x)ξ
du = e ixξ
f (u) e−iuξ du = eixξ fb(ξ)
−∞ −∞
for all ξ ∈ R, and by applying (29.4) to this g, we obtain
Z ∞ Z ∞
1 1
f (x) = g(0) = gb(ξ) dξ = fb(ξ) eixξ dξ,
2π −∞ 2π −∞
completing the proof of Theorem 29.9. □
 LECTURE 30

The Laplace Transform

In this lecture, we introduce the Laplace transformation. The first step is to

define the class of functions on which it will act.

1. Functions of exponential type

Definition 30.1 (Functions of exponential type). Suppose that A ∈ R. A function
f : [0, ∞) → C is said to be of exponential type A if there exists a constant C such
that
|f (t)| ≤ C eAt ∀t ∈ [0, ∞).
A function f : [0, ∞) → C is of exponential type A+ if it is exponential type A + ε
for all ε ∈ R+ .
The definition of type A+ is not standard, but is convenient.
The following examples illustrate these concepts.
Exercise 30.2. In this exercise, a, b ∈ C and p denotes a polynomial. Show
that:
(a) t 7→ a ebt is of exponential type Re(b);
(b) t →7 p(t) ebt is of exponential type Re(b)+;
(c) t → 7 et is not of exponential type A for any A ∈ R.
2

Answer. For part (a), we observe that ebt = eRe(bt) = eRe(b)t for all t ∈
[0, +∞); we then take C = |a| in the definition of exponential type.
For (b), observe first that if ε > 0, then
p(t)
lim p(t) e−εt = lim = 0,
t→+∞ t→+∞ eεt

from l’Hôpital’s rule. This means that |p(t) e−εt | < 1 if t is big enough, say t > R.
The function t 7→ p(t) e−εt is continuous in the compact interval [0, R], and so it
is bounded there, that is, there exists B ∈ R+ such that |p(t) e−εt | ≤ B for all
t ∈ [0, R]. We deduce that |p(t) e−εt | ≤ B + 1 for all t ∈ [0, R) and all t ∈ [R, +∞),
whence
|p(t)| ≤ C eεt ∀t ∈ [0, +∞),
where C = B + 1, and finally
p(t) ebt ≤ C eεt ebt = C e(Re(b)+ε)t ∀t ∈ [0, +∞).
Finally, for part (c), suppose with a view to a contradiction that there is a real
number A such that
2
et ≤ C eAt ∀t ∈ [0, +∞).
Then taking logarithms, we see that
t2 ≤ ln(C) + At ∀t ∈ [0, +∞),
181
182 30. THE LAPLACE TRANSFORM

which is impossible, and so the real number A cannot exist. 4

Note that if f is of exponential type A, then it is of exponential type A+ and
also of exponential type B for all B ∈ [A, ∞). For computing exponential type, the
following lemma may be useful.
Lemma 30.3. Suppose that the functions f, g : [0, +∞) → C are of exponential
types A and B respectively. Then the functions af + bg and af g are of exponential
type max{A, B} and A + B respectively, for all a, b ∈ C. Similarly, if the functions
f and g are of exponential types A+ and B+, then the functions af + bg and af g
are of exponential types max{A, B}+ and (A + B)+.
Proof. We leave this as an exercise. □
We will also require a certain type of integrability of the functions to which we
will apply the Laplace transformation, which we now define.
Definition 30.4. A function f : [0, +∞) → C is said to be locally integrable if
it is Riemann integrable on all intervals [0, R] where R ∈ R+ .
Finally, we will deal with many sets of the form {z ∈ C : Re(z) > A}; for brevity,
we will denote this (right) half-plane by HA .

2. The Laplace transform

Definition 30.5. Suppose that f : [0, ∞) → C is locally integrable and of expo-
nential type A. The Laplace transform Lf : HA → C of f is the function given
by Z Z
∞ R
Lf (z) = f (t) e−zt dt = lim f (t) e−zt dt.
0 R→∞ 0
R∞
The integral in the definition converges by comparison with 0 C e(A−Re(z))t dt.
Notice that L is linear. More precisely, if f and g are locally integrable and of
exponential types A and B, and a, b ∈ C, then af + bg is also locally integrable and
is of exponential type C, where C = max{A, B}. Moreover, if Re(z) > C, then
L(af + bg)(z) = aLf (z) + bLg(z).
Exercise 30.6. Find the Laplace transforms of the functions
(a) t 7→ a (b) t →
7 p(t) (c) t →
7 eat (d) t 7→ p(t) eat .
Answer. Parts (a) to (c) are all particular cases of part (d), and so we solve
P
(d) only. Suppose that p(t) = K k
k=0 ck t .
If f : [0, ∞) → R is given by f (t) = p(t) eat , then f is locally integrable and
of exponential type Re(a)+. Hence the Laplace transform of f is defined in the
half-plane HRe(a) and, when z ∈ HRe(a) ,
Z ∞ Z RX K
at −zt
Lf (z) = p(t) e e dt = lim ck tk e(a−z)t dt
0 R→∞ 0 k=0
X
K Z R X
K
k!
= ck lim tk e(a−z)t dt = ck ;
k=0
R→∞ 0 k=0
(z − a)k+1
we computed the integral by parts and induction on k. 4
 2. THE LAPLACE TRANSFORM 183

Theorem 30.7. If f : [0, ∞) → C is locally integrable and of exponential type

A+, then Lf is holomorphic on HA . Further,
d
Lf (z) = Lg(z) ∀z ∈ HA ,
dz
where g : [0, +∞) → C is given by g(t) = −t f (t).
Proof. We give the proof at the end of this lecture. □
Next, here are a number of properties of the Laplace transformation.
Proposition 30.8. Suppose that f : [0, ∞) → C is locally integrable and of
exponential type A+. Then the following hold:
(a) if a ∈ C and g(t) = e−at f (t) for all t ∈ [0, +∞), then
Lg(z) = Lf (z + a) ∀z ∈ HA−Re(a) ;
(b) if a ∈ R+ and g(t) = f (t/a) for all t ∈ [0, +∞), then
Lg(z) = aLf (az) ∀z ∈ HA/a ;
(c) if g(t) = tf (t) for all t ∈ [0, +∞), then
d
Lg(z) = − Lf (z) ∀z ∈ HA ;
dz
(d) if f is differentiable and f ′ is also of exponential type A+, then
L(f ′ )(z) = zLf (z) − f (0) ∀z ∈ HA .
Proof. To prove (a), observe that
Z ∞ Z ∞
−zs
Lg(z) = g(s) e ds = e−as f (s) e−zs ds
Z0 ∞ 0

= f (s) e−(z+a)s ds = Lf (z + a).

0

To prove (b), we make the change of variables s = at:

Z ∞ Z ∞
−zs
Lg(z) = g(s) e ds = f (s/a) e−zs ds
Z0 ∞ 0

= f (t) e−azt a dt = aLf (az).

0

We prove something very close to (c) in Theorem 30.7, and omit this.
To prove (d), we integrate by parts:
Z ∞ Z ∞
−zs
Lg(z) = g(s) e ds = f ′ (s) e−zs ds
0
Z ∞ 0
= [ f (s) e−zs ]∞
0 − f (s) (−z) e−zs ds = −f (0) + zLf (z).
0
This last argument is not really complete—in the integration by parts, we really
ought to deal with integrals over [0, R] and then let R tend to +∞. □
Remembering these properties can save time; for example, by part (a), since the
Laplace transform of t3 is 6/z 4 , the Laplace transform of t3 e−2t is just 6/(z + 2)4 .
184 30. THE LAPLACE TRANSFORM

3. The inversion formulae for the Laplace transform

Theorem 30.9. If f : [0, ∞) → C is continuous and of exponential type A+, then
Z
1
f (t) = lim Lf (z) etz dz ∀t ∈ R+ , (30.1)
R→∞ 2πi λ

where λ is the line segment from σ − iR to σ + iR and σ ∈ (A, ∞).

Suppose further that Lf extends to a holomorphic function on C \ {a1 , . . . , an },
and that there are positive constants M and k such that

|Lf (z)| ≤ M |z|−k

whenever |z| is sufficiently large. Then for any t ∈ R+ ,
X
n
f (t) = Res(Lf (z) ezt ; z = aj ) ∀t ∈ R+ . (30.2)
j=1

The integral on the right-hand-side of (30.1) is known as the Bromwich Integral.

Proof. We give a sketch of the argument.

The first inversion formula may be deduced from the inversion formula for the
Fourier transformation rather quickly. If we write z = σ + iy, then
Z ∞ Z ∞
−(σ+iy)t
Lf (z) = f (t) e dt = gσ (t) e−iyt dt = (gσ b(y),
0 −∞

where gσ (t) = f (t) e−σt if t ≥ 0 and gσ (t) = 0 if t < 0. By the inversion formula for
the Fourier transform, if t > 0 then
Z ∞ Z σ+iR
−σt 1 −σt 1
f (t) e = Lf (σ + iy) e dy = e
iyt
lim Lf (z) ezt dz.
2π −∞ R→∞ 2πi σ−iR

Cancelling the factor of e−σt now leads to formula (30.1).

For formula (30.2), we consider the semicircular contour γ t λ, where γ(θ) =
σ + R eiθ and π/2 ≤ θ ≤ 3π/2. The result will follow from the Cauchy Residue
Theorem once we show that
Z
lim g(z) etz dz = 0.
R→∞ γ

This may be done using a Jordan’s lemma-type argument. □

Strictly speaking, this is not a proof, because we proved the Fourier inversion
formula for a more limited class of functions. However, with a little more work the
Fourier inversion formula may be proved for more general functions, justifying the
above argument.
The key fact is that the Laplace transformation is invertible. Thus to find the
inverse Laplace transform of g, it suffices to find a continuous function f of expo-
nential type such that Lf = g; then f is the desired inverse transform.
 5. A PROOF† 185

4. Examples
Exercise 30.10. Find the continuous function f : [0, ∞) → C of exponential
1
type 1+ such that Lf (z) = .
(z − 1)2
1
Answer. We have seen that the Laplace transform of t 7→ t is z 7→ 2 . Now
z
1
the Laplace transform of t 7→ t e is z 7→
t
, by Proposition 30.8. 4
(z − 1)2

Exercise 30.11. Find the continuous function f : [0, ∞) → C of exponential

1
type 1+ for which Lf (z) = 2 .
z −1
1
Answer. We rewrite 2 in partial fractions:
z −1
 
1 1 1 1
= − .
z2 − 1 2 z−1 z+1
By Proposition 30.8, this is the Laplace transform of 1
2
( et − e−t ) = sinh t, and
f (t) = sinh(t). 4

5. A proof†
For convenience, we restate the result that we are going to prove.
Theorem. If f : [0, ∞) → C is locally integrable and of exponential type A+,
then Lf is holomorphic on HA . Further,
d
Lf (z) = Lg(z) ∀z ∈ HA ,
dz
where g : [0, +∞) → C is given by g(t) = −t f (t).
Proof. We do this by applying Theorem 26.4, which states that we may ex-
change the order of differentiation with respect to a parameter and integration when
three conditions are satisfied. By definition,
Z ∞
Lf (z) = f (s) e−zs ds.
0

Consider the function G : [0, +∞) × HA → C given by G(s, z) = f (s) e−zs .

Observe that
dG
(s, z) = −s f (s) e−zs ,
dz
which is a continuous function on [0, +∞) × C. Hence the continuity condition of
Theorem 26.4 is satisfied.
Take z ∈ HA , so that Re(z) > A. Set ε = (Re(z)−A)/2; then Re(z)−A = 2ε > ε,
that is, z ∈ HA+ε . Since f is of exponential type A+, there exists C such that
|f (s)| ≤ C e(A+ε/2)s and |s f (s)| ≤ C e(A+ε/2)s
for all s ∈ [0, ∞). If z ∈ HA+ε , then
|G(s, z)| ≤ |f (s)| e−zs = |f (s)| e− Re(z)s ≤ C e(A+ε/2)s e−(A+ε)s = C e−εs/2
186 30. THE LAPLACE TRANSFORM

and similarly
dG
(s, z) ≤ C e−εs/2 ,
dz
and the integrability conditions of Theorem 26.4 are also satisfied. Thus we may
differentiate under the integral and obtain the desired result. □
 LECTURE 31

Applications to differential and integral equations

We may use the Laplace transform to solve differential and integral equations
on [0, ∞).

1. Differential equations
The Laplace transform may be used to solve differential equations on [0, +∞).
Exercise 31.1. Use the Laplace transform to solve the ordinary differential
equation
d2 u du
2
(t) − 2 (t) + u(t) = t ∀t ∈ [0, +∞),
dt dt
du
subject to the initial conditions u(0) = (0) = 0.
dt
Answer. Recall from Proposition 30.8 that L(f ′ )(z) = zLf (z) − f (0). We take
the Laplace transform of both sides of the equation and use this result and the initial
conditions to obtain
1
z 2 Lu(z) − 2zLu(z) + Lu(z) = 2 .
z
Thus
1 1
Lu(z) = 2 2 = 2 .
z (z − 2z + 1) z (z − 1)2
Now it is just a matter of inverting the Laplace transform. First, by partial fractions,
1 2 2 1 1
Lu(z) = = − + 2+ ;
z 2 (z − 1) 2 z z−1 z (z − 1)2
n!
Now the Laplace transform of tn is n+1 for all n ∈ N, and the Laplace transform
z
−at n n!
of e t is for all a ∈ C by Proposition 30.8, and so
(z + a)n+1
u(t) = 2 − 2 et + t + t et ,
and we are done. 4

Notice that this solution could have been found by more elementary methods,
and the use of the Laplace transform here is using a sledgehammer to crack a nut.
The same cannot be said so easily of the next examples.
Example 31.2. Suppose that someone wiggles one end of a very long straight
piece of string up and down for a certain time. How does the string move?
187
188 31. APPLICATIONS TO DIFFERENTIAL AND INTEGRAL EQUATIONS

Let us suppose that the vertical displacement of the string at position x and
time t is given by u(x, t). The string satisfies the wave equation
∂ 2u ∂ 2u
= 2,
∂x2 ∂t
and we may take the initial conditions u(x, 0) = ∂∂tu (x, 0) = 0 when x > 0 and
u(0, t) = f (t), where f (t) = 0 unless a < t < b, where b > a > 0. In addition, we
shall make other assumptions along the way to get a solution.
We write Lu(x, z) for the Laplace transform of u in the t variable; this means
that x 6= Re(z). Then (assuming that there are no problems with exchanging the
order of differentiation and integration)
∂ 2 Lu ∂ 2u ∂ 2u
(x, z) = L (x, z) = L (x, z).
∂x2 ∂x2 ∂t2
Further, from Proposition 30.8,
∂ 2u ∂u ∂u ∂u
L 2
(x, z) = zL (x, z) − (x, 0) = z 2 Lu(x, z) − u(x, 0) − (x, 0).
∂t ∂t ∂t ∂t
The initial conditions imply that u(x, 0) = ∂∂tu (x, 0) = 0, and so
∂ 2 Lu
(x, z) = z 2 Lu(x, z).
∂x2
The solution to this equation is
Lu(x, z) = A(z)ezx + B(z)e−zx .
It is physically sensible to suppose that u(x, t) → 0 as x → +∞, so it seems
reasonable to suppose that A(z) = 0. Thus, taking x = 0, we see that
B(z) = Lu(0, z) = Lf (z).
Finally,
Lu(x, z) = e−zx Lf (z).
To unravel the Laplace transform, observe that if g(t) = f (t − c), then g(t) = 0 if
t < c, and so
Z ∞ Z ∞
−tz
Lg(z) = g(t)e dt = g(t)e−tz dt
Z0 ∞ c
Z ∞
−tz
= f (t − c)e dt = f (t)e−(t+c)z dt = e−cz Lf (z).
c 0
Hence
u(x, t) = f (t − x) ∀t ∈ R+ .
Example 31.3. Use the Laplace transform to find the solution u : [0, ∞) ×
[0, ∞) → R of the partial differential equation
∂u ∂ 2u
(s, t) = 2 (s, t) ∀(s, t) ∈ R+ × R+ ,
∂t ∂s
subject to the initial conditions u(s, 0) = 0 for all s ∈ R+ and u(0, t) = 1 for all
t ∈ R+ .
We omit the answer to this problem, which is known as the heat equation on the
half-line [0, ∞).
 2. CONVOLUTION ON [0, ∞) 189

2. Convolution on [0, ∞)
Throughout this section, f and g will denote locally integrable functions on
[0, ∞).
Definition 31.4 (Convolution on [0, ∞)). The convolution of f and g is the
function f ∗ g : [0, ∞) → C given by
Z t
f ∗ g(t) = f (s) g(t − s) ds.
0

Now we examine the properties of convolution.

Proposition 31.5. Suppose that a, b ∈ C and f, g, h are locally integrable func-
tions of exponential type A+. Then f ∗ g, f ∗ h and f ∗ (a g + b h) are also of
exponential type A+. Further
(a) f ∗ (a g + b h) = a f ∗ g + b f ∗ h;
(b) f ∗ g = g ∗ f ;
(c) L(f ∗ g) = Lf Lg.

Proof. Fix ε ∈ R+ . Since f and g are of exponential type A + ε, there exist

constants C and D such that |f (t)| ≤ C e(A+ε)t and |g(t)| ≤ D e(A+ε)t for all t ≥ 0.
Hence
Z t
CD (A+2ε)t
|f ∗ g(t)| ≤ C e(A+ε)s D e(A+ε)(t−s) ds = CD t e(A+ε)t ≤ e .
0 ε
Since ε is arbitrary, f ∗g is of exponential type A+. Similarly, f ∗h and f ∗(a g +b h)
are of exponential type A+.
Item (a) follows from the definition of convolution as an integral, and linearity
of that integral.
To prove item (b), take an arbitrary t ∈ [0, +∞). The change of variables
u = t − s shows that
Z t Z t
f ∗ g(t) = f (s) g(t − s) ds = f (t − u) g(u) du = g ∗ f (t).
0 0

Item (c) is a consequence of the fact that the exponential function is multiplica-
tive (that is, ea+b = ea eb ). Again, by the change of variables u = t − s, we see
that
Z +∞ Z t 
L(f ∗ g)(z) = f (s) g(t − s) ds e−zt dt
0 0
Z +∞ Z t
= f (s) e−zs g(t − s) e−z(t−s) ds dt
0 0
Z +∞ Z +∞
= f (s) e−zs g(u) e−zu ds du
0 0
= Lf (z) Lg(z)
for all z ∈ HA+ . □
190 31. APPLICATIONS TO DIFFERENTIAL AND INTEGRAL EQUATIONS

3. Integral equations
Exercise 31.6. Use the Laplace transform to find a function u satisfying the
integral equation
Z
2 t
u(t) = t −
2
(t − s)3 u(s) ds.
3 0
This is an example of a Volterra integral equation; these arise in many areas of
application, including finance and biology.

Answer. We first note that this is a “convolution equation”, in the sense that
it may be expressed as
u(t) = t2 − h ∗ u(t),
where h(t) = 23 t3 .
Taking the Laplace transform of this equation and using Proposition 31.5, we
obtain
2 4Lu(z)
Lu(z) = 3 − ,
z z4
which may be rearranged to give
2z
Lu(z) = .
z4+4
Again, u may be found by applying the inversion formula for the Laplace transform
or by the method of partial fractions and inspection. This time we use the inversion
formula.
First observe that Lu has four simple poles, at the fourth roots of −4. These
roots are (1 + i)ik , where k = 1, 2, 3, 4. If w is any one of these roots, then, by
l’Hôpital’s rule,
2z etz z−w
Res(L u(z) etz , z = w) = lim (z − w) = 2w e tw
lim
z→w z4 + 4 z→w z 4 + 4
1 1
= 2w etw lim 3 = 2w etw 3
z→w 4z 4w
1
= etw 2 .
2w
k k
e(1+i)i t e(1+i)i t
k
Thus the residue at (1 + i)i is if k is odd and − if k is even.
4i 4i
Using the inversion formula for the Laplace transform, we obtain
e(1+i)t e(−1+i)t e(−1−i)t e(1−i)t
u(t) = − + −
4i  4i  4i  4i 
−t
1 e −e
t
1 e − e−t
t
= eit − e−it
2i 2 2i 2
 t −t
  it −it

e −e e −e
=
2 2i
= sinh t sin t,
and we are done. 4
 3. INTEGRAL EQUATIONS 191

Suppose that X is a positive random variable, described by a probability density

function f ; then f (t) = 0 when t < 0. Let F be the cumulative distribution function
associated to f ; that is, Z t
F (t) = f (s) ds.
−∞
The renewal equation associated to X is the Volterra integral equation
Z t
m(t) = F (t) + f (s) m(t − s) ds. (31.1)
0

Exercise 31.7. Use the Laplace transform to find a function m satisfying the
renewal equation (31.1), when f (t) = e−t for all t ∈ R+ .
Answer. We note that the renewal equation is a convolution equation, and
apply the Laplace transform, to get
Lm(z) = LF (z) + Lf (z) Lm(z).
Then
LF (z)
Lm(z) = .
1 − Lf (z)
Now we compute the Laplace transforms of f and F . Clearly,
1
Lf (z) = ∀z ∈ H−1 ,
z+1
z
and hence 1 − Lf (z) = . Further,
z+1
Z +∞ Z t Z +∞ Z t
−zt
LF (z) = f (s) ds e dt = e−s e−zt ds dt
0 0 0 0
Z +∞ Z ∞ Z +∞ h −zt i∞
−s −zt e
= e e dt ds = e−s ds
0 s 0 −z s
Z +∞ −zs Z +∞ −(z+1)s
e e
= e−s ds = ds
0 z 0 z
h e−(z+1)s i∞ 1
= = .
−z(z + 1) 0 z(z + 1)
Thus
1
LF (z) z(z + 1) 1
Lm(z) = = z = 2,
1 − Lf (z) z
z+1
and so m(t) = t. 4
 LECTURE 32

The Dirichlet problem

In this lecture, we introduce the Dirichlet problem. We show that the solution to
this problem in a domain is unique, if it exists, using the so-called maximum prin-
ciple; later we consider the questions of existence and of finding explicit solutions.
Given a subset S of C, we write C(S) for the collection of all continuous functions
on S. At the risk of some confusion, we will write Ω both for an open subset of C
and for the corresponding subset of R2 .
Recall that a real-valued function h of two real variables u and v is said to be
harmonic in Υ its partial derivatives of order one and two are continuous in Υ, and
∆h(u, v) = 0 for all (u, v) ∈ Υ, where
 2 
∂ ∂2
∆h(u, v) = + h(u, v).
∂u2 ∂v 2
The differential operator ∆ is known as the Laplacian or Laplace operator.
Problem 32.1 (Dirichlet problem). Suppose that Ω is a bounded domain and
that ∂Ω is a finite union of contours. Given a function b ∈ C(∂Ω), find a harmonic
function u on Ω such that
∆u(x, y) = 0 for all (x, y) ∈ Ω
lim u(x, y) = b(x0 , y0 ) for all (x0 , y0 ) ∈ ∂Ω.
(x,y)→(x0 ,y0 )

1. The maximum modulus principle

Recall from Lecture 26 that the open mapping theorem states that if Ω is a
domain and f ∈ H(Ω) is nonconstant, then Range(f ) is open.
Theorem 32.2 (Maximum modulus principle). Suppose that Ω is a bounded domain
in C, that f ∈ H(Ω), and that f extends continuously to the compact set Ω. If there
is a point z0 ∈ Ω such that
|f (z0 )| = max{|f (z)| : z ∈ Ω},
then f is constant.

Proof. If f were not constant, then Range(f ) would contain an open ball
around f (z0 ), by the open mapping theorem, and |f (z0 )| could not be the maxi-
mum value. □
Exercise 32.3. Find an example of a nonconstant holomorphic function defined
on a bounded open set Ω for which there is a point z0 ∈ Ω such that
|f (z0 )| = max{|f (z)| : z ∈ Ω}.
193
194 32. THE DIRICHLET PROBLEM

Answer. We may take Ω to be B(0, 1) ∪ B(3, 1), and define

(
z when z ∈ B(0, 1)
f (z) =
3 when z ∈ B(3, 1).

Then 3 ∈ Ω and |f (3)| = max{|f (z)| : z ∈ Ω}. 4

Exercise 32.4. Find the maximum value of |sin(z)| as z varies over the closed
unit disc B(0, 1).
Answer. By the maximum modulus principle, the maximum value of |sin(z)|
is attained on the boundary of the ball, where x2 + y 2 = 1. Now
|sin(x + iy)| = |sin(x) cosh(y) + i cos(x) sinh(y)|

1/2
= sin2 (x) cosh2 (y) + cos2 (x) sinh2 (y)

1/2
= sin2 (x) + sin2 (x) sinh2 (y) + cos2 (x) sinh2 (y)

1/2
= sin2 (x) + sinh2 (y)

1/2
= sin2 (x) − x2 + sinh2 (y) − y 2 + 1 .
Changing the sign of x or y does not change the value, so it will suffice to look
for a point where x ≥ 0 and y ≥ 0.
It is easy to check that, for such x and y,
d

sin2 (x) − x2 = 2 sin(x) cos(x) − 2x ≤ 2x cos(x) − 2x
dx
= −2x(1 − cos(x)) ≤ 0
and similarly
d

sinh2 (y) − y 2 = 2 sinh(y) cosh(y) − 2y ≥ 2y cosh(y) − 2y
dy
= 2y(cosh(y) − 1) ≥ 0.
It follows that we can make |sin(x + iy)| as large as possible by making y as big
and x as small as possible. Hence the maximum value is sinh(1), attained when
z = ±i. 4

2. The maximum principle for harmonic functions

Theorem 32.5 (Maximum principle for harmonic functions). Suppose that Ω is a
bounded domain in R2 , that h : Ω → R is harmonic and that h extends continuously
to the compact set Ω. If there is a point (x0 , y0 ) ∈ Ω such that
h(x0 , y0 ) = max{h(x, y) : (x, y) ∈ Ω},
then h is constant.

Proof. We will first prove this for a set Ω1 that is simply connected. Suppose
that there is a point (x0 , y0 ) ∈ Ω1 such that |h(x0 , y0 )| = max{|h(x, y)| : (x, y) ∈ Ω1 }.
Since Ω1 is simply connected, we may find a harmonic conjugate of h, and hence
 4. THE DIRICHLET PROBLEM AND HOLOMORPHIC MAPPINGS 195

write h(x, y) = Re(f (x + iy)), for some f ∈ H(Ω1 ). Consider the holomorphic
function exp ◦f . Evidently
|exp ◦f (x + iy)| = exp(Re(f (x + iy))) = exp(h(x, y)),
and this attains its maximum at x0 + iy0 , which lies in the interior of Ω1 . By the
maximum modulus principle, exp ◦f is constant in Ω1 , that is, f is constant in Ω1 ,
and so h is constant in Ω1 .
To deal with the general case of a domain Ω that need not be simply connected,
we take a arbitrary point (x1 , y1 ) of Ω, and show that h(x1 , y1 ) = h(x0 , y0 ), and
conclude that h is constant. Since Ω is connected, there is a polygonal path joining
(x1 , y1 ) and (x0 , y0 ); by omitting any loops, we may assume that this is simple. Now
by “fattening up” this path a little, we may find a connected and simply connected
open subset Ω1 of Ω that contains both points. Observe that
h(x0 , y0 ) ≤ max{h(x, y) : (x, y) ∈ Ω1 } ≤ max{h(x, y) : (x, y) ∈ Ω} ≤ h(x0 , y0 ),
since (x0 , y0 ) is just one of the points considered in the first maximum, since Ω1 ⊆ Ω,
and by hypothesis. In particular, h(x0 , y0 ) = max{h(x, y) : (x, y) ∈ Ω1 }, and
so we may apply the result for the simply connected set Ω1 , and conclude that
h(x0 , y0 ) = h(x1 , y1 ); hence h is constant. □

3. Uniqueness for the Dirichlet problem

Corollary 32.6 (Uniqueness for the Dirichlet problem). Suppose that Ω is a
bounded domain, that b ∈ C(∂Ω), and that u1 and u2 are solutions of the Dirichlet
problem
∆u(x, y) = 0 for all (x, y) ∈ Ω
lim u(x, y) = b(x0 , y0 ) for all (x0 , y0 ) ∈ ∂Ω.
(x,y)→(x0 ,y0 )

Then u1 = u2 .

Proof. By linearity, u1 − u2 is a harmonic function in Ω that tends to 0 as

we approach the boundary of Ω; it suffices to show that u1 − u2 = 0. We write
u = u1 − u2 , and extend u to Ω by continuity; then u ∂Ω = 0.
If there were a point (x, y) ∈ Ω such that u(x, y) > 0, then there would be a
point (x0 , y0 ) ∈ Ω such that
u(x0 , y0 ) = max{u(x, y) : (x, y) ∈ Ω} > 0.
By the maximum principle, (x0 , y0 ) ∈ ∂Ω. However, by definition, u ∂Ω = 0, and so
u(x0 , y0 ) = 0, which is a contradiction. Thus u ≤ 0 in Ω, that is u1 ≤ u2 .
Similarly, by taking u = u2 − u1 , we may show that u2 ≤ u1 . Hence u2 = u1 , as
required. □

4. The Dirichlet problem and holomorphic mappings

We will spend some time looking at the Dirichlet problem later, but here is a
very important idea, which is one of the reasons that complex analysis is useful.
Theorem 32.7. Suppose that Ω is a domain, that f : Ω → Υ is holomorphic, and
that h : Υ → R is harmonic. Write f (x + iy) = u(x, y) + iv(x, y), where u and v
196 32. THE DIRICHLET PROBLEM

are real-valued, and let F (x, y) = (u(x, y), v(x, y)) be the corresponding vector-valued
function. Then h ◦ F is harmonic in Ω.

Proof. It is evident that h ◦ F is defined in Ω, so it suffices to show that it is

harmonic. To do this, it suffices to take an arbitrary point (x0 , y0 ), and show that
the function is harmonic in B((x0 , y0 ), r) for some small positive r. Let z0 = x0 +iy0 ,
and (u0 , v0 ) = (u(x0 , y0 ), v(x0 , y0 )).
Since h is harmonic, it has a conjugate harmonic function k, at least in some
ball centred at (u0 , v0 ) and lying in Ω. Let g(u + iv) = h(u, v) + ik(u, v), for (u, v)
in this ball.
From the definitions, we see that
h(u(x, y), v(x, y)) = Re(h(u(x, y), v(x, y)) + ik(u(x, y), v(x, y)))
= Re(g(f (x + iy))).
Since g and g and holomorphic, so is g ◦ f , and hence (x, y) 7→ h(u(x, y), v(x, y)) is
harmonic. □
The Dirichlet problem is important. If we can find holomorphic functions from
a “harder” region to an “easier” one, we will be able to solve the Dirichlet problem
in the “harder” case. Indeed, we have the following theorem.
Theorem 32.8. Suppose that Ω is a domain in C, that B is a subset of ∂Ω,
and that b : B → R is a continuous function. Suppose also that f : Ω → Υ is
holomorphic, that f extends continuously to B, in the sense that if z0 ∈ B and z →
z0 , then f (z) → f (z0 ) ∈ ∂Υ, and that c : f (B) → R is given by c(w) = b(f −1 (w))
for all w ∈ f (B). If h : Υ → R solves the Dirichlet problem
∆h(u, v) = 0 for all (u, v) ∈ Υ
lim h(u, v) = b(f −1 (u0 , v0 )) for all (u0 , v0 ) ∈ f (B),
(u,v)→(u0 ,v0 )

then (x, y) 7→ h(u(x, y), v(x, y)) solves the Dirichlet problem
∆u(x, y) = 0 for all (x, y) ∈ Ω
lim u(x, y) = b(x0 , y0 ) for all (x0 , y0 ) ∈ B.
(x,y)→(x0 ,y0 )

If moreover f : Ω → Υ is bijective, and extends to a continuous bijection of the

boundaries, then the solution to every Dirichlet problem in Ω arises by transferring
a solution to the Dirichlet problem in Υ.
Proof. Omitted. □
The next result is one of the most important theorems in complex analysis.
Theorem 32.9 (Riemann mapping theorem). Given any simply connected do-
main Ω in C, other then C itself, there exists a bijective holomorphic mapping
f : Ω → B(0, 1).
The proof of this is too long and hard for this course. But it relies on being able
to find a harmonic function on Ω such that u(z) = − ln |z − z0 | on the boundary of
Ω, where z0 is a point in Ω. It seems to be circular to use the Riemann mapping
theorem to solve the Dirichlet problem and then use the Dirichlet problem to prove
the Riemann mapping theorem, but it is not!
 6. SOLVING THE DIRICHET PROBLEM IN THE UPPER HALF PLANE 197

5. Examples
In the exercises below, we set H = {(x, y) ∈ R2 : y > 0}, that is, H is the upper
half plane. Boundary values should be interpreted as continuous limits at points of
continuity on the boundary.
Exercise 32.10. Find bounded harmonic functions h in H such that
(a) h(x, 0) = 1 for all x ∈ R;
(b) h(x, 0) = 0 if x > 0 and h(x, 0) = π if x < 0;
(c) h(x, 0) = sgn(x) for all x ∈ R;
(d) h(x, 0) = 1 if 0 < x < 1 and h(x, 0) = 0 if x < 0 or x > 1.
Answer. These are all combinations of constants and argument functions. 4

6. Solving the Dirichet problem in the upper half plane

The preceding exercise suggests that we can find harmonic functions in the upper
half plane whose boundary values are “step functions”. But actually we can do more.
The key is the following: let H be the “Heaviside function”: H(x) = 0 if x < 0 and
H(x) = 1 if x ≥ 0. If g : R → R is continuously differentiable and both g and g ′
tend to 0 at ±∞, then
Z x Z ∞
′
g(x) = g (s) ds = H(x − s) g ′ (s) ds
Z−∞
∞
−∞
1
= (1 − Arg(x − s)) g ′ (s) ds
π
Z−∞
∞ Z Z
′ 1 ∞ ′ 1 ∞
= g (s) ds − Arg(x − s) g (s) ds = − Arg(x − s) g ′ (s) ds.
−∞ π −∞ π −∞
Now Arg extends into the upper half plane as a harmonic function, and so we
examine the function h, defined on the upper half plane, by
Z
1 ∞
h(z) = − Arg(z − s) g ′ (s) ds
π −∞
Z
1 ∞
=− Arg(x − s + iy) g ′ (s) ds
π −∞
Z
1 ∞ ∂
= Arg(x − s + iy) g(s) ds.
π −∞ ∂s
Finally,
∂ ∂ π 
Arg(x − s + iy) = − tan−1 ((x − s)/y)
∂s ∂s 2
1
=
((x − s) /y 2 + 1)y
2

y
= .
(x − s)2 + y 2
Theorem 32.11. Suppose that g is a bounded continuous function on R, and
define h on the upper half-plane by the formula
Z
1 ∞ y
h(x + iy) = g(s) ds ∀x ∈ R ∀y ∈ R+ .
π −∞ (x − s)2 + y 2
198 32. THE DIRICHLET PROBLEM

Then the integral defining h converges, and h has the following properties:
(a) h is harmonic
(b) h is bounded
(c) limx+iy→x0 h(x + iy) = g(x0 ) for all x0 ∈ R.
Further, h is the only function with these properties.
Proof. See next lecture. □

7. Solving the Dirichet problem in the unit ball

We may also compute an explicit solution to the Dirichlet problem in the unit
disc.
Theorem 32.12. Suppose that g is a bounded continuous function on ∂B(0, 1),
and define h inside B(0, 1) by the formula
Z π
iθ 1 1 − r2
h(re ) = g(ei(θ−φ) ) dφ
2π −π 1 − 2r cos(φ) + r2
for all reiθ ∈ B(0, 1). Then the integral defining h converges, and moreover
(a) h is harmonic;
(b) h is bounded;
(c) limreiθ →eiθ0 h(reiθ ) = g(eiθ0 ) for all θ0 ∈ [−π, π].
Further, h is the only function with these properties.
Proof. Omitted. □
We can now solve the Dirichlet problem in several different regions.
 LECTURE 33

Conformal mappings and harmonic functions

In this lecture, we first show that our proposed solution to the Dirichlet problem
in the upper half plane has the desired properties. Next, we examine conformal
mappings: mappings that preserve “form”, at least locally. Finally we solve the
Dirichlet problem in some other regions.

1. The Dirichlet problem in the upper half plane

Now we prove our theorem on harmonic functions in the upper half plane. First
we recall the theorem for completeness.
Theorem. Suppose that g is a bounded continuous function on R, and define h
on the upper half-plane by the formula
Z
1 ∞ y
h(x + iy) = g(s) ds ∀x ∈ R ∀y ∈ R+ .
π −∞ (x − s) + y
2 2

Then the integral defining h converges, and h has the following properties:
(a) h is harmonic
(b) h is bounded
(c) limx+iy→x0 h(x + iy) = g(x0 ) for all x0 ∈ R.
Further, h is the only function with these properties.
Proof. To show that h is harmonic, we observe that
Arg(z − s) = Re(−i Log(z − s)),
and so
 
∂ ∂ ∂ 1
Arg(z − s) = Re(−i Log(z − s)) = Re (−i Log(z − s)) = Re i ,
∂s ∂s ∂s z−s
which is harmonic, so if we can differentiate under the integral sign, the harmonicity
will follow. It is possible to verify that it is legitimate to exchange the order of
differentiation and integration; we omit the details.
To show that h is bounded, we observe that
Z Z
1 y 1 y
g(s) ds ≤ | g(s)| ds
R π (x − s) + y R π (x − s) + y
2 2 2 2
Z
1 y
≤ sup{|g(s)| : s ∈ R} ds
R π (x − s) + y
2 2
Z
1 1
= sup{|g(s)| : s ∈ R} 2
dt
R πt +1
= sup{|g(s)| : s ∈ R},
by the change of variables t = (x − s)/y.
199
200 33. CONFORMAL MAPPINGS AND HARMONIC FUNCTIONS

Finally, to establish the boundary limiting behaviour, change variables to see

that
Z
1 ∞ y
h(x + iy) = g(x − s) ds
π −∞ s + y 2
2

whence, changing variable again (s = yt), we see that

Z
1 ∞ y
h(x + iy) − g(x0 ) = (g(x − s) − g(x0 )) ds
π −∞ s2 + y 2
Z
1 ∞ 1
= (g(x − yt) − g(x0 )) dt
π −∞ t2 + 1
Z
1 ∞ 1
= (g(x − yt) − g(x)) dt
π −∞ t2 + 1
Z
1 ∞ 1
+ (g(x) − g(x0 )) dt.
π −∞ t2 + 1

Yet another “limit inside the integral” argument shows that this goes to 0 as x + iy
goes to x0 , that is, as x → x0 and y → 0+. □

2. Affine mappings and affine approximations

Consider the translation mapping f : C → C, given by z 7→ z + b, where b ∈ C.
This is a congruence, that is, it preserves distances, so that if T is a triangle in C,
then so is f (T ), and f (T ) is congruent to T , and the corresponding angles are equal.
But more is true: the sides of f (T ) are parallel to those of T .
Now consider the linear mapping f : C → C, given by z 7→ az, where a ∈ C\{0}.
This is composed of a rotation (through the angle Arg(a)) and a dilation (by a factor
of |a|); it is not a congruence unless |a| = 1, but it still preserves angles and ratios of
lengths. More precisely, if T is a triangle in C, then so is f (T ), and f (T ) is similar
to T ; corresponding angles are equal, and ratios of side lengths are preserved. Note
that the orientation of triangles is also preserved: the equal angles of T and f (T )
appear in the same order as we go around the boundaries in the usual anti-clockwise
direction.
We now combine these two observations, and consider the affine mapping f :
C → C, given by z 7→ az + b, where a ∈ C \ {0} and b ∈ C. This mapping also
preserves angles, orientation, and ratios of lengths.
Consider now a differentiable function f : Ω → C, defined in a domain Ω, and
fix z0 ∈ Ω. The definition of differentiability tells us that

f (z) ≈ f (z0 ) + f ′ (z0 )(z − z0 ),

at least when z is close to z0 . What does this tell us about the geometric properties
of f ? Because we are only dealing with an approximation, with errors that increase
the further z is from z0 , it seems unlikely that f will preserve ratios of lengths.
However, if might still be true that f preserves angles. But since f (T ) might have
curved sides, even if T has straight sides, we need to be able to measure angles
between curves.
 3. DIFFERENTIABILITY AND ANGLES BETWEEN TWO CURVES 201

3. Differentiability and angles between two curves

Suppose that γ : I → C and δ : J → C are curves that both pass once through
z0 , where I and J are open intervals in R; we suppose that γ(s0 ) = z0 and δ(t0 ) = z0 .
The tangent vectors to the curves at z0 are γ̇(s0 ) and δ̇(t0 ), where the dot denotes
differentiation (with respect to a real variable). If these tangent vectors are both
nonzero, then the angle between the tangent vectors is well-defined. Because we
are working in the complex plane, we may define the angle between γ and δ to be
Arg(γ̇(s0 )/δ̇(t0 )). This means that, if we align a pointer in the direction in which δ is
moving at z0 , and turn the pointer anti-clockwise through the angle Arg(γ̇(s0 )/δ̇(t0 )),
then the pointer indicates the direction in which γ is moving. Note that this is an
oriented angle, and the angle between δ and γ is Arg(δ̇(t0 )/γ̇(s0 )), which has the
opposite sign (unless the argument is 0 or π).
Now suppose also that f : Ω → C is differentiable at z0 . Then f ◦ γ(s0 ) = f (z0 )
and f ◦ δ(t0 ) = f (z0 ), and the curves f ◦ γ and f ◦ δ pass through f (z0 ) at s0 and t0 .
Further, the chain rule implies that f ◦ γ and f ◦ δ are differentiable at s0 and t0 , and
(f ◦ γ) ˙ (s0 ) = f ′ (z0 ) γ̇(s0 ) while (f ◦ δ) ˙ (t0 ) = f ′ (z0 ) δ̇(t0 ). It follows immediately
that
(f ◦ γ) ˙ (s0 ) f ′ (z0 ) γ̇(s0 ) γ̇(s0 )
= = ,
(f ◦ δ) ˙ (t0 ) f ′ (z0 ) δ̇(t0 ) δ̇(t0 )
and so    
(f ◦ γ) ˙ (s0 ) γ̇(s0 )
Arg = Arg ;
(f ◦ δ) ˙ (t0 ) δ̇(t0 )
that is, f preserves the (oriented) angles between curves that intersect at z0 .
Note that if f is differentiable but f ′ (z0 ) = 0, then f does not preserve angles.
Indeed, if f is constant, then all curves map to the same point in C, and there
are no angles at all to measure; otherwise, by Taylor series, we may write f (z) =
f (z0 ) + cn (z − z0 )n + . . . , where n > 1 since f ′ (z0 ) = 0; it is then an exercise to show
that angles are not preserved.
Exercise 33.1. Suppose that f (z) = z n for all z ∈ C, where n > 1. Let
I = (−1, 1), and consider the curves γθ : I → C given by γθ (t) = teiθ . Show that
the curves f ◦ γ0 and f ◦ γ2π/n coincide and are horizontal, and deduce that f does
not preserve the angles between curves at 0.
Exercise 33.2. Suppose that f (z) = z for all z ∈ C. Show that f changes the
sign of angles between curves.
Definition 33.3. We say that a mapping f from a domain Ω of C to C is
conformal if it is differentiable and f ′ (z0 ) 6= 0 for all z0 ∈ Ω. If the mapping is one-
to-one, then the condition that f ′ (z0 6= 0 for all z0 in Ω tells us that f is invertible,
and its inverse is also differentiable.
Exercise 33.4. Suppose that T : R2 → R2 is a linear mapping. Show that T
preserves the (oriented) angles between vectors if and only if T is composed of a
rotation and a dilation. (Hint: write T as RU , where R is a rotation and U is upper
triangular.)
Deduce that a mapping from an open subset of R2 into R2 that preserves (ori-
ented) angles between curves arises as the “real version” of a holomorphic mapping.
202 33. CONFORMAL MAPPINGS AND HARMONIC FUNCTIONS

Conformal mappings arose when mathematicians started to analyse the ways

in which map-makers present a piece of the earth’s curved surface on a plane.
Map-makers’ projections cannot preserve distance (or even scale), but some map
projections are conformal.
Conformal mappings became important in applied mathematics because com-
posing with conformal mappings preserves harmonic functions, and this allows us
to solve many physical problems.
They are also used to explain morphology in biology.

4. Examples of conformal mappings

In the following exercises, we do not distinguish between points in the complex
plane and points in R2 .
Exercise 33.5. Find a conformal mapping f from the upper half plane H to
the unit ball B.
Answer. The boundary of the upper half plane, that is the x axis, will be
mapped to the boundary of the unit ball, that is the unit circle. We have mentioned
that fractional linear transformations send lines to circles or lines.
z−i
Consider the mapping f (z) = . A point z lies in the upper half plane if
z+i
and only if |z − i| < |z + i|, and this holds if and only if |f (z)| < 1. 4

Exercise 33.6. Let Σα be the sector {z ∈ C : | Arg(z)| < α} in the com-

plex plane. Find a conformal mapping g from the sector Σα (where 0 < α < π
to the upper half plane H. Hence find a harmonic function h in Σα such that
h(r cos α, −r sin α) = 0 and h(r cos α, r sin α) = 1 for all positive r.
Answer. Complex powers map sectors to sectors, and change the angles. Thus
w = z π/2α maps the sector Σα to the sector Σπ/2 , which is the same as the right half
plane. Composing with a rotation will have the desired effect: we set g(z) = iz π/2α .
To find h, we have to solve a boundary value problem. Note that h ◦ g −1 takes
the value 0 on (0, +∞) and 1 on (−∞, 0). We may find a harmonic function in the
upper half plane with these boundary values by modifying Arg: we would like to
have
1
h ◦ g −1 (z) = Arg z.
π
Hence
1 1
h(w) = Arg g(w) = Arg(iwπ/2α )
π π
1 π  1 π π 
= + Arg(wπ/2α ) = + Arg w
π 2 π 2 2α
1 1
= + Arg w
2 2α
has the desired property. 4
Observe that the composed mapping
iz π/2α − i z π/2α − 1
w= =
iz π/2α + i z π/2α + 1
 4. EXAMPLES OF CONFORMAL MAPPINGS 203

takes the sector to the unit ball in the complex plane.

Exercise 33.7. Let Sb be the infinite strip {z ∈ C : 0 < Im z < b}. Find a
conformal mapping h from the infinite strip Sb to the upper half plane H.
Answer. We have already seen that the exponential function exp maps the
strip Sb into the sector {w ∈ C : 0 < Arg(w) < b}. By adjusting this, we can easily
check that g, given by  πz 
h(z) = exp ,
b
has the desired property. 4
Other simply connected regions for which it is possible to find “explicit” confor-
mal mappings from the region to the unit ball include polygons (the mappings that
arise are called Schwarz–Christoffel mappings).
It is also possible to solve the Dirichlet problem in a doubly connected domain (a
domain with one hole, which must not be just a missing point). First, a generalisa-
tion of the Riemann mapping theorem shows that any such domain may be mapped
conformally onto an annulus {z ∈ C : r < |z| < 1}. Next, it is possible to solve the
problem in this kind of annulus, using separation of variables. More precisely, we
may express a harmonic function in the annulus in a “generalised Laurent series”:
X∞ X∞
n
h(z) = an z + ān z̄ n ,
n=−∞ n=−∞

and choose the coefficients an to get the desired boundary value functions.
Note that a + b ln(| · |) is a radial harmonic function in the annulus, which may
be given arbitrary boundary values that are constant on the inner and the outer
circles. This kind of problem arises in electrostatics.
 LECTURE 34

Loose ends

In this last lecture, we consider an application in aerodynamics.

1. An application to fluid mechanics

In the theory of steady flow of an ideal fluid, it is shown that the “streamlines”
of the flow, that is, the lines that a particle in the flow will follow, are level curves
of a harmonic function.
Example 34.1. The streamlines for a 2-dimensional ideal flow around a solid
obstacle in the shape of the unit ball are the parts of the level sets {(x, y) ∈ R2 :
ψ(x, y) = d} that lie outside the unit ball, where
y
ψ(x, y) = y − 2 .
x + y2
To understand the flow around a more complicated object, we may employ a
Joukowsky transformation: this is the composition of an affine map close to the
identity with the map z 7→ z + 1/z.

205