MATH2015 Analysis 2, 200910

Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009

Contact Details (Semester 1)
Lecturer: Prof J C Wood
Oce: School of Mathematics, 8.20e
E-mail: [email protected]
Module website: http://www.maths.leeds.ac.uk/Pure/staff/wood/MATH2015/MATH2015.htm
All handouts will be made available on the above webpage or from the VLE
https://vlebb.leeds.ac.uk/webapps/portal/frameset.jsp
Introduction
The algebra of complex numbers enable us to do many things we couldnt do before. Some examples
you have already met are nding formulae for cos n and sin n from deMoivres theorem, e.g.
cos
4
= 8 cos
4
8 cos
2
; by the same method we can nd formulae for cos
n
x and sin
n
x which
enable you to integrate them. For example, can you integrate sin
4
x?
Here is a new example you will meet in the rst semester. In applications of mathematics, many
quantities, for example steady state temperature distribution on a region of the plane are given in
Cartesian coordinates (x, y) by functions u = u(x, y) which satisfy Laplaces equation
u
xx
+ u
yy
= 0 .
Can you nd any solutions to this equation? We will see how to nd all solutions using complex
analysis. There will be lots more applications in the second semester.
Complex analysis is the theory of functions of a complex variable. It is rather dierent from
functions of a real variable, for example once a function has a complex derivative, it is dierentiable
any number of times. To discover this theory, well need both a theory of dierentiation and a theory
of integration.
1 Revision of complex numbers
1.1 Arithmetic The standard Cartesian form of a complex number is z = x + iy where x, y
are real numbers. x is called the real part of x + iy and y is called the imaginary part. We write
Re (z) = x, Im(z) = y.
Addition, subtraction, multiplication and division are done algebraically, using i
2
= 1. The
complex conjugate of z = x + iy is z = x iy. Do you remember how to do division?
1.2 Argand Diagram or complex plane We represent z = x + iy by the point (x, y) in the
plane, or by the vector from the origin to that point; then addition and subtraction are like addition
and subtraction of vectors. To interpret multiplication and division, we need to discuss the polar
form of a complex number.
1.3 Polar form Given a complex number z = x +iy, its modulus r is the distance of (x, y) from
the origin, so r =
_
x
2
+ y
2
. If z is non-zero, its argument is any value of the angle from the
positive real axis to the vector, considered positive when measured in the anticlockwise direction.
Analytically,
cos = x/r , sin = y/r ;
1
see the following picture:
Note that we get other possible values of the argument by adding any integer multiple of 2; the
principal value of the argument is the one in the range < . Note that tan = y/x, but this
only determines up to multiples of ; you must plot the complex number on the Argand diagram
to get the correct answer. More analytically, the principal value of is only given by arctan(y/x)
when x > 0. If x = 0 then
=
_
/2 when y > 0,
/2 when y < 0.
If x < 0 then
=
_
arctan(y/x) + when y 0,
arctan(y/x) when y < 0.
Given the modulus r and any value of the argument of a complex number z, we can recover
the complex number z = x + iy as follows. From the formula above, a = r cos and y = r sin so
that
z = r(cos + i sin)
This is often written as z = re
i
by using Eulers formula:
e
i
= cos + i sin .
But why is Eulers formula true? What does the LHS mean?
1.4 Arithmetic in polar form Recall that (i) the modulus of a product = the product of its
moduli; (ii) a value of the argument of the product of two non-zero numbers is the sum of arguments,
though adding principal arguments may not give the principal argument.
1.5 Note that [e
it
[ = 1 for any real t (Why?). Also [Re z[, [Imz[ [z[.
1.6 Triangle inequality for moduli [z
1
+ z
2
[ [z
1
[ +[z
2
[
Proof: The geometrical proof is based on the result that the sum of the lengths of the two sides
of a triangle is greater than or equal to the length of the third side. Question: What happens when
the triangle collapses, i.e., z and w are in the same direction? A simple analytical proof is the
following; see the module website for other proofs.
Multiply by a with [a[ = 1, if necessary, so that z
1
+ z
2
is real and positive. Then [z
1
+ z
2
[ =
Re z
1
+ Re z
2
[z
1
[ +[z
2
[.
Now revise the rest of your notes on complex numbers and do the Example Sheet 1!
2
2 Complex Functions
In this section we will dene what we mean by a complex function. We will then generalize the
denitions of the exponential, sine and cosine functions to a complex variable by using complex
power series. To deal with complex power series we dene the notions of convergence and absolute
convergence, and see how to use the ratio test from rst year analysis to determine convergence and
the radius of convergence for these complex series.
We start by dening the type of subset in the complex plane on which are functions will be
dened. This requires a preliminary denition.
Denition 2.7 The -neighbourhood of a
complex number z is the set of complex num-
bers B

(z) = w C : [z w[ < where is a

positive number.
.
.
.
.
.
.
Thus the -neigbourhood of a point z is just the set of points lying within the circle of radius
centred at z. Note that it doesnt contain the circle.
Denition 2.8 An open set is a non-empty subset D of C such that for every point z in D there
exists an -neighbourhood of z contained in D.
Example 2.9 Determine whether the following subsets of C are open sets:
(a) D = C is clearly a open set: given any z D,
we can choose, for example, = 36. Then
B

(z) C = D. .
.
.
.
.
.
(b) D = R, however, is not: 1 D, but for any
> 0, B

(1) contains 1 + i/2 which is not in

D. .
.
.
.
.
.
(c) D = C0
We can now dene the basic object of study.
Denition 2.10 Let D be an open set in C. A complex function f : D C is a map which
assigns to each z in D an element of C, this value being denoted by f(z).
Note that f is the function and f(z) is the value the function assigns to the input z D. Technically,
it is wrong to say that f(z) is a function, what we mean is the function is given by z f(z).
3
Example 2.11 (i) Dene f : C C by f(z) = z
2
.
(ii) Dene f : C C by f(z) = [z[. Note that here we have a complex function for which every
value is real.
(iii) Dene f : C C by f(z) = 3z
4
(5 2i)z
2
+z 7. More generally, any complex polynomial
denes a complex function.
(iv) Dene f : C0 C by f(z) = 1/z. Note that this is not dened on the whole complex
plane.
Remark 2.12 Functions such as sinx for x real are not complex functions since the real line in C
is not an open set. Later we will see how to extend the concept of the sine so that it is a complex
function on the whole of the complex plane.
Obviously, if f and g are complex functions on open sets D
f
and D
g
respectively, then f + g,
f g, and fg are functions on D
f
D
g
given by (f +g)(z) = f(z)+g(z), (f g)(z) = f(z)g(z), and
(fg)(z) = f(z)g(z), respectively. We can also dene (f/g)(z) = f(z)/g(z) provided that g(z) ,= 0
on D
f
D
g
. Thus we can build up lots of new functions by these elementary operations.
2.1 The aim of complex analysis
We wish to study complex functions. Can we dene dierentiation? Can we integrate? Which
theorems from real analysis can be extended to complex analysis? For example, is there a version
of the mean value theorem? Complex analysis is essentially the attempt to answer these questions.
The theory will be built upon real analysis but in many ways it is easier than real analysis. For
example if a complex function is dierentiable (see below for the denition), then its derivative is
also dierentiable. This is not true for real functions do you know an example of a dierentiable
real function with non-dierentiable derivative?
Our main objective will be Cauchys Residue Theorem, an extraordinarily powerful result on
integration of complex functions which turns out to be very useful for computing real integrals too.
Using this, well be able to make short work of seemingly intractable integrals like
_

cos x
x
2
2x + 2
dx.
Imagine trying to compute this using standard real methods. (The answer turns out to be

e
cos 1.)
2.2 Real and imaginary parts of functions
We will often use z to denote a complex number and we will have z = x+iy where x and y are both
real. The value f(z) is a complex number and so also has a real and an imaginary part. We often use
u to denote the real part and v to denote the imaginary part. Thus, f(x + iy) = u(x, y) + iv(x, y).
Note that u and v are both real-valued functions of two real variables, x and y.
Example 2.13 (i) Dene f : C C by f(z) = z
2
. Then,
f(x + iy) = (x + iy)
2
= x
2
y
2
+ 2ixy
u(x, y) = x
2
y
2
v(x, y) = 2xy.
(ii) Dene f : C C by f(z) = [z[. Then,
f(x + iy) =
_
x
2
+ y
2
u(x, y) =
_
x
2
+ y
2
v(x, y) = 0.
(iii) Dene f : C0 C by f(z) = 1/z. Then
u(x, y) = v(x, y) =
4
2.3 Dening e
z
, cos z and sin z
First we will try to dene some elementary complex functions to play with. How shall we dene
functions such as e
z
, cos z and sin z? We require that their denition should coincide with the
real version when z is a real number, and we would like them to have properties similar to those
real versions, e.g. sin
2
z + cos
2
z = 1 would be nice. However, sine and cosine are dened using
trigonometry and so are hard to generalise: for example, what does it mean for a triangle to have a
hypotenuse of length 2 + 3i? The exponential is dened using dierential calculus and we have not
yet dened dierentiation of complex functions.
However, we know from real analysis that the functions can be described using a power series,
e.g.,
sin x = x
x
3
3!
+
x
5
5!
=

n=0
(1)
n
x
2n+1
(2n + 1)!
.
Thus, for z C, we will dene the exponential, sine and cosine of z as follows:
e
z
:=

n=0
z
n
n!
,
sin z :=

n=0
(1)
n
z
2n+1
(2n + 1)!
,
cos z :=

n=0
(1)
n
z
2n
(2n)!
.
Thus,
e
3+2i
=

n=0
(3 + 2i)
n
n!
= 1 + (3 + 2i) +
(3 + 2i)
2
2!
+
(3 + 2i)
3
3!
+ . . .
These denitions obviously satisfy the requirement that they coincide with the denitions we know
and love for real z, but how can we be sure that the series converges?
To answer this we will have to study complex series and, as the theory of real series was built
on the theory of real sequences, we had better start with complex sequences.
5
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
3 Complex Sequences and Series
3.1 Complex Sequences
The denition of convergence of a complex sequence is the same as that for convergence of a real
sequence, as follows.
Denition 3.14 A complex sequence (c
n
) converges to c C, if given any > 0, there exists
N N such that [c
n
c[ < for all n N. We write c
n
c or lim
n
c
n
= c, and call c the limit
of the sequence.
Exercise 3.15 Prove that the limit of a convergent complex sequence is unique.
Example 3.16 The sequence c
n
= (n
2
+i)/(n
2
i) converges to 1. Indeed,

n
2
+ i
n
2
i
1

=
2
[n
2
i[
=
2
n
2
+ 1

2
n
2
. Hence, given > 0,

n
2
+ i
n
2
i
1

< provided n >

2

.
So, given any > 0 we can choose N to be any natural number greater than 2/

. Thus the
sequence converges to zero.
Remark 3.17 Notice that a
n
= [c
n
c[ is a real sequence, and that c
n
c if and only if the real
sequence [c
n
c[ 0. Hence, we can say something about a complex sequence using real analysis.
The remark above gives a good example of the paradigm
1
we will be using. We can often apply
results from real analysis to produce results in complex analysis. In this case we take the modulus,
but we can also take real and imaginary parts. This is a key observation; note it well!
Lets apply the paradigm. The next proposition shows that a sequence converges if and only if
its real and imaginary parts do.
Proposition 3.18 Let c
n
= a
n
+ ib
n
where a
n
and b
n
are real sequences, and c = a + ib. Then
c
n
c a
n
a and b
n
b.
Proof: [] If c
n
c, then [c
n
c[ 0. But
0 [a
n
a[ = [Re(c
n
) Re(c)[ = [Re(c
n
c)[ [c
n
c[.
So by the squeeze rule [a
n
a[ 0, i.e. a
n
a. Similarly, b
n
b.
[] Suppose a
n
a and b
n
b, then [a
n
a[ 0, and [b
n
b[ 0. We have
0 [c
n
c[ = [(a
n
a) + i(b
n
b)[ [a
n
a[ +[b
n
b[.
The last inequality follows from the triangle inequality applied to z = a
n
a and w = i(b
n
b).
Because [a
n
a[ 0 and [b
n
b[ 0 we deduce [c
n
c[ 0, i.e. c
n
c, again by the squeeze
rule.
Tip Try not to use the denition of convergence to prove that a sequence converges.
1
Paradigm: a conceptual model underlying the theories and practice of a scientic subject. (Oxford English
Dictionary).
6
Example 3.20 (Example 3.19 only in lectures)
n
2
+ in
3
n
3
+ 1
=
n
1
+ i
1 + n
3
=
n
1
1 + n
3
+ i
1
1 + n
3
0 + i.1 = i.
Remark 3.21 Many basic results on convergence of complex sequences can be transferred immedi-
ately from real analysis using Proposition 3.18. For example, you should now be able to give quick
proofs (using analogous results from Real Analysis I) of the following:
(i) If a
n
a and b
n
b, then a
n
+ b
n
a + b and a
n
b
n
ab;
(ii) If a
n
a then a
n
is bounded (meaning there exists K R such that [a
n
[ K for all n);
(iii) If a
n
0 and b
n
is bounded then a
n
b
n
0;
(iv) If a
n
is bounded then a
n
has a convergent subsequence.
3.2 Complex Series
Now that we have dened convergence of complex sequences we can dene convergence of complex
series. In fact a series can be considered to be a sequence in the following way.
Denition 3.22 A complex series

k=0
w
k
can be thought of as the sequence (s
n
) of its partial
sums s
n
=

n
k=0
w
k
. That is, it is the sequence:
s
0
= w
0
s
1
= w
0
+ w
1
s
2
= w
0
+ w
1
+ w
2
s
3
= w
0
+ w
1
+ w
2
+ w
3
.
.
.
The series converges if and only if the sequence (s
n
) converges (in the sense of Denition 3.14). If
it does converge we denote its limit, rather confusingly, by the same symbol,

k=0
w
k
. To avoid
confusion (and save ink) we will often denote the series by just

w
k
and reserve

k=0
w
k
for its
limit. The complex number w
k
is called the k
th
term of the series. Be careful not to confuse it with
s
k
, the k
th
partial sum.
Example 3.23 Write down the sequence of partial sums dened by the complex series

k=0
i
k
.
s
0
= 1, s
1
= 1 + i, s
2
= 1 + i + i
2
= i,
s
3
= i + i
3
= 0, s
4
= 1, s
5
= 1 + i.
Since i
k+4
= i
k
i
4
= i
k
, it follows that the sequence (s
n
) just keeps cycling through these 4 values
endlessly. Clearly, this complex series does not converge.
We can now transfer Proposition 3.18 immediately to complex series:
Corollary 3.24 Let w
k
= x
k
+ iy
k
where x
k
and y
k
are real for all k. Then,

w
k
converges

x
k
and

y
k
converge.
In this case

k=0
w
k
=

k=0
x
k
+ i

k=0
y
k
.
7
Proof: Let a
n
=

n
k=0
x
k
, b
n
=

n
k=0
y
k
, and s
n
=

n
k=0
w
k
, and apply Proposition 3.18.
Example 3.25 The series

k=0
(1)
k
i/k! converges.
Thus

k=0
(1)
k
i/k! converges to i/e.
In real analysis we have some great ways to tell if a series is convergent, for example, the ratio
test and the integral test. Can we use the real analysis tests in complex analysis? The next theorem
says we can, but rst let us make a denition.
Denition 3.26 We say

w
k
is absolutely convergent if the real series

[w
k
[ converges.
This denition is really the same as in real analysis, it has merely been extended to complex numbers
in a natural way.
We now prove a very useful theorem which says that if a series is absolutely convergent, then it
is convergent. Note that the hypothesis concerns the real series

[a
k
[, to which we can apply the
full battery of tests from real analysis, while the conclusion concerns the complex series

a
k
.
Theorem 3.27 (Absolute convergence implies convergence) .
If

[w
k
[ converges, then

w
k
converges.
Proof: Let w
k
= x
k
+ iy
k
, with x
k
and y
k
real. Then

[w
k
[ convergent implies that

[x
k
[
is convergent (because 0 [x
k
[ = [Re(w
k
)[ [w
k
[ and we can apply the comparison test). So
the real series

x
k
converges absolutely and we know from Real Analysis I that this implies that

x
k
converges. Similarly, the series

y
k
converges. Thus, by Corollary 3.24,

w
k
converges to

k=0
x
k
+ i

k=0
y
k
.
Tip When asked to show a series converges, try to show it absolutely converges.
Remark 3.28 The converse to Theorem 3.27 is not true. We already know this from real analysis.
For example,

k=1
(1)
k
k
converges but

k=1

(1)
k
k

k=1
1
k
diverges.
So if

[w
k
[ converges, so does

w
k
. Is there any relationship between their limits? The next
proposition, which can be thought of as an innite version of the triangle inequality, says yes!
Proposition 3.29 Suppose that

w
k
converges absolutely. Then

k=0
w
k

k=0
[w
k
[.
To prove this carefully, well need a couple of lemmas. First
8
Lemma 3.30 For any complex sequence (a
n
), a
n
a implies [a
n
[ [a[.
Proof: A consequence of the triangle inequality is the inequality
[z w[

[z[ [w[

for any z, w C.
Can you prove this?
Hence 0 [[a
n
[ [a[[ [a
n
a[ 0 so [[a
n
[ [a[[ 0 by the squeeze rule.
Lemma 3.31 For real sequences a
n
a and b
n
b, if a
n
b
n
for all n, then a b.
Proof: Assume, to the contrary, that a > b, and consider the sequence c
n
= b
n
a
n
0. This
converges to b a < 0, a contradiction (let = (a b)/2 > 0, so convergence implies there is some
N for which n N implies c
n
< (b a) + < 0 contradicting c
n
0).
Proof of Proposition 3.29: Let s
n
=

n
k=0
w
k
and t
n
=

n
k=0
[w
k
[. Note that s
n
s :=

k=0
w
k
and t
n
t :=

k=0
[w
k
[. Then [s
n
[ [s[ by Lemma 3.30. Furthermore, for all n, [s
n
[ t
n
by
induction on n:
Base step: [s
0
[ = t
0
t
0
Inductive step: if [s
n
[ t
n
then
[s
n+1
[ = [s
n
+ w
n+1
[ [s
n
[ +[w
n+1
[ t
n
+[w
n+1
[ = t
n+1
.
Hence [s[ t by Lemma 3.31, which was to be proved.
3.3 Power Series and Radius of Convergence
Denition 3.32 A complex power series is a series of the form

k=0
c
k
z
k
, where c
k
C. Note
that such a series may be a nite sum (this is the case if there is an m such that c
k
= 0 for all
k m).
2
A power series is a function of z. Much of the theory of dierentiable complex functions is con-
cerned with power series, because as we shall see later, any dierentiable complex function can be
represented locally as a power series.
Just as with real power series we can have complex power series that do not converge on the
whole of the complex plane.
Example 3.33 Consider the series

n=1
z
n
, where z C. We know for z = 1 this series does not
converge because then we have

1
1
n
=

1
1 = 1 + 1 + 1 + . . . .
We also know it converges for z = 0, because

1
0
n
=

1
0 = 0 + 0 + 0 + = 0. Hopefully,
you remember from Real Analysis I that for real z the power series converges only for 1 < z < 1.
So, for which complex values of z does it converge? Lets use the ratio test. Let a
n
= [z
n
[. Then

a
n+1
a
n

=
[z
n+1
[
[z
n
[
= [z[.
As n we have [z[ [z[, because there is no dependence on n. So by the ratio test the series

a
n
converges if [z[ < 1, diverges if [z[ > 1 and for [z[ = 1 we dont know what will happen. So

z
n
converges absolutely, and hence converges, for [z[ < 1. For [z[ > 1, the ratio test tells us that

[z
n
[ is divergent and hence

z
n
is not absolutely convergent. However, we cannot conclude (yet)
that

z
n
is divergent in this case.
2
We take z
0
= 1, even when z = 0. Dont fall into the trap of thinking 0
0
is equal to 0.
9
Given a power series

a
k
z
k
, we can ask: what does the set of points z C for which the series
converges look like? Since the sequence of coecients a
k
could be absolutely anything, you might
expect that the answer could be arbitrarily nasty. Actually it turns out to be always nice and simple:
either the series converges only for z = 0, or it converges for all z, or it converges in the interior
of some disk (centred on 0) of nite radius. In the last case, the only complication is that it may
converge on all, none, or part of the boundary circle of the disk. To prove this, we must dene the
radius of convergence of a complex power series.
Denition 3.34 The radius of convergence of a power series

a
k
z
k
is
R := sup[z[ :

[a
k
z
k
[ converges.
Note the set [z[ :

[a
k
z
k
[ converges R always contains 0 (since

a
k
z
k
converges to a
0
at
z = 0), so is certainly nonempty. If the set is unbounded, we say R = .
Lemma 3.35 Let

a
k
z
k
converge at z = z
1
. Then it converges absolutely for all z such that
[z[ < [z
1
[.
Proof: Since

a
k
z
k
1
converges, so do its real and imaginary parts (Corollary 3.24), and hence their
sequences of terms, Re(a
k
z
k
1
) and Im(a
k
z
k
1
) must converge to 0. Thus a
k
z
k
1
0 (Proposition 3.18).
Hence there exist K, N > 0 such that k N implies [a
k
z
k
1
[ < K (convergent implies bounded). But
then for all k N,
[a
k
z
k
[ = [a
k
z
k
1
[
[z[
k
[z
1
[
k
< K
_
[z[
[z
1
[
_
k
so if [z[ < [z
1
[,

[a
k
z
k
[ converges by comparison with the convergent (geometric) series

([z[/[z
1
[)
k
.

Theorem 3.36 Let

a
n
z
n
have radius of convergence R. Then

a
n
z
n
_
converges absolutely for [z[ < R,
diverges for [z[ > R.
Proof: Let A := [z[ :

[a
k
z
k
[ converges R, so that R = supA.
If [z[ < R then there exists z
1
with [z[ < [z
1
[ < R such that

a
k
z
k
1
is absolutely convergent (else
R isnt the least upper bound on A, [z[ being smaller), hence convergent, so

a
k
z
k
is absolutely
convergent by Lemma 3.35.
If [z[ > R and

a
k
z
k
converges then

a
k
z
k
2
, where z
2
= ([z[ + R)/2, converges absolutely by
Lemma 3.35, since [z
2
[ < [z[. But [z
2
[ > R, a contradiction (R isnt an upper bound on A at all!),
so we conclude that

a
k
z
k
diverges.
Tip Given a power series, immediately ask, What is its radius of convergence?
Note that Theorem 3.36 says nothing about the convergence of the series on the circle of radius R.
Exercise 3.37 Show that

1
z
n
/n has radius of convergence 1.
In the last exercise note that for z = 1 the series converges, but for z = 1 the series diverges
both these facts should be well known from real analysis).
10
3.4 Sine, cosine, and exponential are dened for all complex numbers
Let us now return to showing that the sine, cosine and exponential functions are dened on the
whole of C (they each have innite radius of convergence).
Proposition 3.38 For all z C, the series
e
z
=

n=0
z
n
n!
, sinz =

n=0
(1)
n
z
2n+1
(2n + 1)!
, cos z =

n=0
(1)
n
z
2n
(2n)!
converge.
Proof: We will prove convergence of e
z
and leave sinz and cos z as an exercise. For any z C let
a
n
=

z
n
n!

. We want

n=0
a
n
to converge (since then the series for e
z
will converge by the result
absolute convergence = convergence). So, we use the ratio test on this real series. We have

a
n+1
a
n

z
n+1
z
n
n!
(n + 1)!

=
[z[
n + 1
0.
The last part is true because for xed z the real number [z[ is a nite constant in the denition of
a
n
. Hence, by the ratio test,

n=0
a
n
converges so by Theorem 3.27, the series

n=0
z
n
n!
converges for
all z C.
Exercise 3.39 Complete the proof of Proposition 3.38.
So, e
z
, sin z and cos z are all well dened: whatever z I use, all three series converge to nite
complex numbers. Excellent!
3.5 Properties of the exponential
Why do we write the sum of the rst series in Proposition 3.38 as e
z
?
Let us rst of all consider what e
x
, or indeed a
x
, means for real x and a > 0. When x is rational,
say p/q, a
x
means the qth root of a
p
. When x is irrational, we can write x as the limit of a sequence
r
i
of rational numbers, and dened a
x
as the limit of a
ri
(some work is required to show this is
well-dened).
Now, for the special number e, from real analysis, we know that the function e
x
is given by the
sum of the series e
x
=

n=0
x
n
/n!, which is convergent for all x.
Now the series for e
z
reduces to the series for e
x
when z = x +i0 is real, further e
z
continues to
enjoy the index properties, as the next result shows. For this reason, we continue to use the notation
e
z
. An alternative notation is exp(z).
Theorem 3.40 (i) e
z
= e
z
, for all z C.
(ii) e
iz
= cos z + i sinz, for all z C.
(iii) e
z+w
= e
z
e
w
, for all z, w C.
(iv) e
z
,= 0, for all z C.
(v) e
z
= 1/e
z
, for all z C.
(vi) e
nz
= (e
z
)
n
, for all z C and n Z.
(vii) [e
z
[ = e
Re(z)
, for all z C.
11
(viii)

e
iy

= 1, for all y R.
Proof: (i) Let e
z
= a + ib. Now e
z
is the limit of s
n
:=
n

k=0
z
k
/k! =
n

k=0
z
k
/k!. But the latter
converges to a ib by Proposition 3.18.
(ii) Exercise. (Just put iz into the power series for the exponential function.)
(iii) This will be delayed until we deal with dierentiability.
(iv) Note that e
z
and e
z
both exist. We have
e
z
e
z
= e
zz
by (iii),
= e
0
= 1, by calculation.
Thus, e
z
cannot be zero.
(v) This follows from the proof of (iv).
(vi) Follows from repeated application of (iii) (or by induction on n, if you want to be picky).
(vii) We have
[e
z
[
2
= e
z
e
z
by denition,
= e
z
e
z
by (i),
= e
z+z
by (iii),
= e
2Re(z)
=
_
e
Re(z)
_
2
by (vi).
As both [e
z
[ and e
Re(z)
are real and positive we deduce that (vii) is true.
(viii) From (vii) we get [e
iy
[ = e
Re(iy)
= e
0
= 1.
You should know these properties and their proofs by heart.
Warning 3.41 We have not shown that e
zw
= (e
z
)
w
for all z, w C. This is because we have
not yet dened a
b
for all complex a and b. How do you think we might dene a
b
when a and b are
complex?
Exercise 3.42 Prove that
sin z =
e
iz
e
iz
2i
and cos z =
e
iz
+ e
iz
2
.
Unlike its real counterpart, the complex exponential is not one-to-one. In fact, it is periodic.
Theorem 3.43 For any complex numbers z and w we have
e
z
= e
w
z w = 2in for some n Z.
12
Proof: [] Let z w = x + iy where x and y are real. Then,
e
z
= e
w
e
z
/e
w
= 1
e
zw
= 1
e
x+iy
= 1 ()
= [e
x+iy
[ = 1, (the implication does not reverse!)
[e
x
e
iy
[ = 1
[e
x
[[e
iy
[ = 1
[e
x
[ = 1
e
x
= 1, since the exponential of a real number is positive,
x = 0.
By () we know that e
x+iy
= 1, so e
iy
= 1 as x = 0. Then,
e
iy
= 1 cos y + i siny = 1
cos y = 1 and sin y = 0
y = 2n for some n Z.
So z w = x + iy = 0 + i.2n = 2in.
[] Suppose that z w = 2in for some n Z. Then, e
zw
= e
2in
=
_
e
2i
_
n
= 1
n
= 1. But
from the working in the earlier part of the proof we know this is equivalent to e
z
= e
w
.
Example 3.44 Probably, the simplest example is that e
0
= 1 and e
2i
= 1.
Recall that, in real analysis, the natural logarithm function log : (0, ) R is dened to be
the inverse function of the exponential function. This makes sense because given any positive real
number y, there exists a unique real number x such that e
x
= y, so we can dene log y to be x. Note
that injectivity (one-to-one-ness) of the exponential function is crucial for this.
In complex analysis, given any non-zero complex number w, there are innitely many complex
numbers z such that e
z
= w, as we now see.
Proposition 3.45 For complex numbers z and w ,= 0 we have
e
z
= w z = log [w[ + i(arg(w) + 2k), for k Z.
Proof: [] Write z = x+iy, so we get e
x+iy
= e
x
(cos y +i siny) = w. Now let us take the modulus
of both sides:
[w[ = [e
x+iy
[ = [e
x
[[e
iy
[ = [e
x
[ = e
x
log [w[ = x.
Hence
w = [w[e
i arg(w)
= e
x
e
iy
= [w[e
iy
e
i arg(w)
= e
iy
iy i arg(w) = 2ik
for some k Z by Theorem 3.43. Hence z must have the form stated.
[] If z = log [w[ + i(arg(w) + 2k) where k Z then
e
z
= e
log |w|+i(arg(w)+2k)
= e
log |w|
e
i arg(w)
e
2ik
= [w[e
i arg(w)
_
e
2i
_
k
= [w[e
i arg(w)
= w.
13

Many authors write z = log w for all the solutions to e

z
= w, thus log w = log [w[+i(arg(w)+2k)
for any k Z; however, note that this is not a function as it is multivalued. There is a way to consider
such multivalued functions by using Riemann surfaces but that is beyond the scope of this course.
To get a function, we must make a choice, the usual one is called the principal value of log w.
This is the unique solution z of e
z
= w with < Im(z) , it is thus given by the formula:
principal value of log w = log [w[ +i arg(w) where arg(w) is the principal argument of w. Note that
the principal value of log w jumps as w crosses the negative real axis, so that it isnt continuous. For
that reason, we shall avoid the notation log w for a complex number w in the following examples.
Example 3.46 Solve e
z
= 1 + i

3.
Let w = 1 + i

3. The modulus of w is [w[ =

_
1
2
+

3
2
=

4 = 2. By drawing a picture (or

through careful use of a calculator) we can see that the principal value of arg(w) is

3
. Hence
z = log 2 + i
_

3
+ 2k
_
, k Z.
Common Error: Dont forget the 2k.
Having worked out how to solve e
z
= w, its not hard to solve sinz = w and cos z = w too.
Example 3.47 Solve the equation sinz = 2.
Solution: We can rewrite this as
e
iz
e
iz
2i
= 2. Let w = e
iz
. Then the equation becomes
1
2i
_
w
1
w
_
= 2. Equivalently, w
2
4iw 1 = 0, which has solutions w = 2i

3. So,
writing z = x+iy, we have e
iz
= e
ix
e
y
= 2i i

3 which holds i e
y
=

7 and x = /2 +2k
giving z = /2 + 2k i ln

7.
14
Tip Note that in the above example we replaced e
iz
with another complex number w, because we
could then get a polynomial equation.
Exercise 3.48 Show that
sinz = 0 z = k, k Z,
cos z = 0 z =
1
2
(2k + 1), k Z.
These mean that the zeros of sine and cosine lie on the real line. These results will be used later.
Learn them and make your lecturer happy.
Summary
We dene exponential, sine and cosine by power series.
If

n=0
[w
k
[ converges, then

n=0
w
k
converges.
Apply the ratio test, comparison test, etc, to the modulus of terms of a complex series to
determine convergence.
For power series use the ratio test to nd radius of convergence.
15
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
4 Continuity of Complex Functions
The denition of continuity of a complex function is identical to that of a real function, as follows.
Denition 4.1 f : D C is continuous at w D if for all > 0 there exists > 0 such that
[z w[ < [f(z) f(w)[ < .
We say f is continuous if it is continuous at every w D.
We want to know that if f, g are continuous, so are f +g, fg, Re(f), Im(f), [f[ etc. etc. Its not
hard to prove these directly from the denition, but it is time consuming. Luckily, we can get all
this stu for free (almost) using the following denition and theorem:
Denition 4.2 Let D C be an open set, w D and f : D C. f is sequentially continuous
at w if given any sequence z
n
D such that z
n
w, the image sequence f(z
n
) f(w).
Theorem 4.3 (Sequential continuity is equivalent to continuity) .
f : D C is continuous at w D if and only if f is sequentially continuous at w.
Proof: [] Let f be continuous at w and z
n
be any sequence converging to w. Then given any
> 0, there exists > 0 such that [f(z) f(w)[ < for all w with [z w[ < . But since z
n
w,
there exists N > 0 such that n N implies [z
n
w[ < and hence, [f(z
n
) f(w)[ < . Hence
f(z
n
) f(w).
[] We will prove the contrapositive. Assume that f is discontinuous at w D. Then there
exists some > 0 such that for all > 0, there is at least one z B

(w) with f(z) / B

(f(w)).
3
In
particular, this is true for = 1/n where n Z
+
. So, for each n Z
+
let z
n
be one of the elements
of B
1/n
(w) for which f(z
n
) / B

(f(w)). Then 0 < [z

n
w[ < 1/n so z
n
w by the squeeze rule,
but [f(z
n
) f(w)[ for all n so f(z
n
) does not converge to f(w).
Theorem 4.3 has three immediate corollaries:
Corollary 4.4 Let f : D D

C be continuous at w, and g : D

C be continuous at f(w).
Then their composition g f is continuous at w.
Proof: Let z
n
w in D. Then f(z
n
) f(w) in D

, by continuity of f, so g(f(z
n
)) g(f(w)) by
continuity of g. This is true for all such sequences z
n
, so g f is continuous at w.
Corollary 4.5 Let f : D C be continuous at w. Then Re(f), Im(f) and [f[ are continuous at
w.
Proof: Let z
n
w in D. Then f(z
n
) f(w), so Re(f(z
n
)) Re(f(w)) and Im(f(z
n
))
Im(f(w)) by Proposition 3.18 and [f(z
n
)[ [f(w)[ by Lemma 3.30. This is true for all such
sequences z
n
, so Re(f), Im(f) and [f[ are all continuous at w.
Corollary 4.6 Let f : D C and g : D

C be continuous at w. Then f + g and fg are

continuous at w.
3
Recall that Br(w) = {z : |z w| < r}, the open disk of radius r centred on w.
16
Proof: Exercise.
Example 4.7 Clearly f(z) = z and f(z) = c are trivially continuous, by Theorem 4.3. Hence, all
polynomials are continuous, by Corollary 4.6.
Exercise 4.8 Is f : C0 C such that f(z) = z/[z[ continuous? Be careful!
One thing we cannot yet prove is that functions like e
z
, sin z and cos z, dened by power series,
are continuous. For the time being, you must take this on trust.
17
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
5 Complex Dierentiation
5.1 Denitions
If we were inventing the theory of dierentiation of complex functions for the rst time, then we
might be tempted to dene complex dierentiable to mean that the real and imaginary parts of
the function are dierentiable with respect to x and y. This would give us a theory, but not a great
one: it would be the same as the theory of dierentiable maps from R
2
to R
2
. Instead we develop
a theory that really uses the complex numbers. This is much richer. It looks a lot better, but more
importantly, it allows us to solve some hard problems concerning real functions in a simple manner.
As in real analysis, the derivative of f(z) will be a limit.
Denition 5.1 Suppose that f : D C is a complex function on an open set D. We say f has
limit L at c, written lim
zc
f(z) = L, if for all > 0, there exists > 0 such that
[f(z) L[ < whenever 0 < [z c[ < .
Remark 5.2 (i) This is entirely analogous to the denition of lim
xa
f(x) in real analysis (as
usual).
(ii) If lim
zc
f(z) exists it is unique. Proof by contradiction: assume L and L

,= L both satisfy
Denition 5.1. Then take = [L L

[
_
4 > 0. There must exist > 0 such that for z
B

(c)c, f(z) is in both B

(L) and B

(L

). But these disks are disjoint.

(iii) The denition of lim
zc
f(z) concerns the behaviour of f arbitrarily close to c, but not at c
(if, indeed, c D): note [z c[ > 0 in Denition 5.1.
(iv) There is a strong link with continuity (Denition 4.1): f is continuous at c D if and only
if lim
zc
f(z) = f(c).
Given Remark 5.2(iv) and Theorem 4.3 (sequential continuity is equivalent to continuity) its
not surprising that we can give a sequential criterion for f to have limit L at c:
Proposition 5.3 Let f : D C. Then lim
zc
f(z) exists if and only if f(z
n
) converges for every
sequence (z
n
) in Dc with z
n
c. In this case f(z
n
) converges to the same limit for all such
(z
n
), and this is lim
zc
f(z).
Proof: Exercise: straightforward modication of the proof of Theorem 4.3.
It follows immediately from this result and previous work on limits of sequences that
lim
zc
(f(z) + g(z)) = lim
zc
f(z) + lim
zc
g(z),
lim
zc
f(z)g(z) = lim
zc
f(z) lim
zc
g(z)
etc. etc.
We are now ready to dene complex dierentiation. At rst sight, it doesnt very dierent from
the real situation, but the consequences are far more profound.
18
Denition 5.4 Let f : D C be a complex function. Then, f is (complex) dierentiable at c
if
lim
h0
f(c + h) f(c)
h
exists. We write f

(c) for this limit.

We say f : D C is complex dierentiable or holomorphic (on D) if it is complex dier-
entiable at c for all c D.
The denition looks the same as in the real case, but the fact that h can go to zero from any
direction in the complex plane makes a huge dierence; it imposes considerable restrictions, as we
shall see. Holomorphic functions are also called analytic, or even regular, functions by some
authors. We will see why later.
Example 5.5 (a) The function f : C C given by f(z) = z
2
is holomorphic: let c C. Then
That is, f

(c) = 2c, as expected.

(b) The function f : C C such that f(z) = [z[
2
is not holomorphic, however. In fact it is complex
dierentiable only at 0:
f

(0) = lim
h0
[h[
2
0
h
= lim
h0

h = 0.
At any c ,= 0, lets dene the dierence quotient : C0 C,
(h) =
f(c + h) f(c)
h
=
1
h
([c + h[
2
[c[
2
) =
1
h
( ch + c

h + h

h) = c + c

h
h
+

h.
We can show that f

(c) = lim
h0
(h) does not exist by exhibiting a sequence h
n
0 such that
(h
n
) does not converge. One such sequence is
h
n
=
i
n
n
.
Note that [h
n
[ =
1
n
0, and
(h
n
) = c + c
h
n
h
n
+ h
n
= c + c
(i)
n
i
n
+ h
n
= c + c(1)
n
+ h
n
which does not converge (unless c = 0). Hence f

(c) does not exist by Proposition 5.3.

More generally, we have that, if f(z) = bz
n
for some complex constant b and integer n, then f

(z) =
nbz
n1
. One can easily prove this from rst principles (i.e. from Denition 5.4). Alternatively, it
follows by induction on n from the following general properties.
Proposition 5.6 Let f and g be complex functions on an open set D.
(i) If f and g are dierentiable at c D, then so are
(a) f + g, and (f + g)

(c) = f

(c) + g

(c);
(b) fg, and (fg)

(c) = f(c)g

(c) + f

(c)g(c);
(c) f/g, and (f/g)

(c) =
g(c)f

(c) f(c)g

(c)
g(c)
2
provided g(c) ,= 0.
19
(ii) (Chain Rule) Suppose f : D C and g : E C are complex functions with f(D) E.
If f is dierentiable at c and g is dierentiable at f(c), then g f is dierentiable at c with
(g f)

(c) = g

(f(c))f

(c).
Proof: These are proved in the same way as the real case.
Its clear that f, g : C C such that f(z) = z and g(z) = c, constant, are holomorphic, and that
f

(z) = 1, g

(z) = 0. It follows from Proposition 5.6, therefore, that every polynomial function
p(z) = a
0
+ a
1
z + + a
n
z
n
is holomorphic.
Warning 5.7 Unlike continuity, complex dierentiability isnt the same as being dierentiable with
respect to the two real variables x, y (where z = x +iy). In fact, it is much stronger as we shall see
shortly. In fact it is this dierence that makes complex analysis so dierent from real analysis.
Exercise 5.8 Show that if f : D C is dierentiable at c, then f is continuous at c.
We dont yet know that we can dierentiate series term-by-term and so cant immediately prove
that e
z
, cos z and sin z are dierentiable, or that their derivatives are what we expect them to be.
Exercise 5.9 Where are the following functions dierentiable?
z, z + z, z
2
,
z
[z[
2
.
For f(z) = z we have f

(z) = lim
h0
_
(z + h z)
_
h
_
= lim
h0
(h/h), which does not exist, so
f is not dierentiable anywhere.
If z + z were dierentiable at a point then, since z is, z would be dierentiable at that point,
which it is not; hence z + z is not dierentiable anywhere.
For f(z) = z
2
we have f

(z) = lim
h0
_
(z + h
2
z
2
)
_
h
_
= lim
h0
2zh
_
h+h
2
_
h which doesnt
exist unless z = 0. Hence f is dierentiable only at z = 0.
Since z/[z[
2
= 1/z (z ,= 0), this function is dierentiable at all points z ,= 0.
lim
zc
f(z) = L means that for all > 0 there is some > 0 such that
[f(z) L[ < whenever 0 < [z c[ < .
The complex function f is (complex) dierentiable at c if
lim
h0
f(z + h) f(z)
h
exists.
If so we denote this limit f

(c).
Dierentiablity = continuity.
Complex dierentiation satises all the familiar rules of dierential calculus (sum rule, product
rule, quotient rule, chain rule).
5.2 The CauchyRiemann Equations
If f is a holomorphic (complex dierentiable) function, then the real and imaginary parts are dier-
entiable as real functions (but not vice versa, see earlier warning). Futhermore the partial derivatives
of u and v are related to one another.
Theorem 5.10 Suppose f(x+iy) = u(x, y) +iv(x, y), where z = x+iy, and that f is holomorphic
at c = a + ib. Then the partial derivatives u
x
, u
y
, v
x
and v
y
all exist at (a, b) and
f

(c) = u
x
(a, b) + iv
x
(a, b) = v
y
(a, b) iu
y
(a, b).
20
Proof: Write (h) = [(f(c + h) f(c)]/h and let h
n
,= 0 be any real sequence converging to 0.
Then by Proposition 5.3, (h
n
) f

(c) and (ih

n
) f

(c). But
(h
n
) =
1
h
n
[u(a + h
n
, b) + iv(a + h
n
, b) u(a, b) iv(a, b)]
=
u(a + h
n
, b) u(a, b)
h
n
+ i
v(a + h
n
, b) v(a, b)
h
n
.
Hence, by Proposition 3.18,
u(a + h
n
, b) u(a, b)
h
n
Ref

(c) and
v(a + h
n
, b) v(a, b)
h
n
Imf

(c).
But h
n
0 is any real sequence. Thus
u
x
(a, b) = lim
h0
u(a + h, b) u(a, b)
h
= Ref

(c)
and v
x
(a, b) = lim
h0
v(a + h, b) v(a, b)
h
= Imf

(c).
This establishes the rst equality claimed. The second follows from an identical argument using
(ih
n
) f

(c) (write this out!).

An immediate corollary is the celebrated CauchyRiemann equations, named after two of the
founders of complex analysis.
Corollary 5.11 (CauchyRiemann equations) If f(z) = u(x, y) + iv(x, y) is complex dieren-
tiable at z = a + ib, then
u
x
(a, b) = v
y
(a, b) and v
x
(a, b) = u
y
(a, b).
Proof: Equate real and imaginary parts in the theorem above.
Example 5.12 (a) Let f(z) = z
2
= (x + iy)
2
. Then,
u = x
2
y
2
and v = 2xy so that u
x
= 2x = v
y
and u
y
= 2y = v
x
,
so the C-R equations hold.
(b) Let f(z) = [z[
2
= x
2
+ y
2
. Then, u = x
2
+ y
2
and v = 0 so that u
x
= 2x, u
y
= 2y, v
x
= 0,
v
y
= 0.
The only place where the C-R equations are satised is x = y = 0. It follows that f is not complex
dierentiable at any c ,= 0. Actually we already knew that (Example 5.5).
Exercise 5.13 Show that the function f(z) = [z[ is not dierentiable anywhere in C. Note that it
is continuous everywhere! HARD: Can you think of a real function f : R R which is continuous
everywhere but dierentiable nowhere? (Its OK to give up after 5 minutes.)
21
Remark 5.14 Corollary 5.11 says that if f = u+iv is holomorphic, then u and v satisfy the coupled
pair of linear partial dierential equations
u
x
=
v
y
,
u
y
=
v
x
. (CR)
What about the converse? Is every u + iv satisfying (CR) holomorphic? Not quite. A cheap and
nasty counterexample is given by f : C C such that
f(z) =
_
0, if x = 0 or y = 0,
1, otherwise.
This satises (CR) at z = 0 (check!), but f is not continuous at z = 0, so cant be complex
dierentiable there. You might think this counterexample is rather fake, and that any reasonable
function satisfying (CR) should be holomorphic. Youd be right.
Theorem 5.15 Let f : D C such that f(x+iy) = u(x, y)+iv(x, y) where u and v are continuously
dierentiable and satisfy the CauchyRiemann equations for all x+iy D. Then f is holomorphic,
and f

(x + iy) = u
x
(x, y) + iv
x
(x, y).
Proof: Let w = x + iy D, and z
n
= x
n
+ iy
n
0 (z
n
,= 0). We wish to show that [f(w + z
n
)
f(w)]/z
n
converges. But
f(w + z
n
) f(w)
z
n
=
1
z
n
_
u(x + x
n
, y + y
n
) u(x, y + y
n
) + i[v(x + x
n
) v(x, y + y
n
)]
+u(x, y + y
n
) u(x, y) + i[v(x, y + y
n
) v(x, y)]
_
=
x
n
z
n
_
u(x + x
n
, y + y
n
) u(x, y + y
n
)
x
n
+ i
v(x + x
n
, y + y
n
) v(x, y + y
n
)
x
n
_
+
y
n
z
n
_
u(x, y + y
n
) u(x, y)
y
n
+ i
v(x, y + y
n
) v(x, y)
y
n
_
.
Consider the real-valued function of one real variable g(t) = u(x + t, y + y
n
). It is continuously
dierentiable on the interval from 0 to x
n
, so by the Mean Value Theorem, there exists some
number, call it
n
, between 0 and x
n
such that
g

(
n
) =
g(x
n
) g(0)
x
n
0
.
Then, by the denition of g,
u
x
(x +
n
, y + y
n
) =
u(x + x
n
, y + y
n
) u(x, y + y
n
)
x
n
and 0 < [
n
[ < [x
n
[, so
n
0. By identical reasoning, there exist real sequences
n
,
n
,
n
0
such that
v(x + x
n
, y + y
n
) v(x, y + y
n
)
x
n
= v
x
(x +
n
, y + y
n
),
u(x, y + y
n
) u(x, y)
y
n
= u
y
(x, y +
n
),
v(x, y + y
n
) v(x, y)
y
n
= v
y
(x, y +
n
).
22
Hence
f(w + z
n
) f(w)
z
n
=
x
n
z
n
u
x
(x +
n
, y + y
n
) + iv
x
(x +
n
, y + y
n
) +
y
n
x
n
u
y
(x, y +
n
) + iv
y
(x, y +
n
)
= u
x
(x +
n
, y + y
n
) + iv
x
(x +
n
, y + y
n
) +
i
y
n
z
n
_
[v
y
(x, y +
n
) v
y
(x +
n
, y + y
n
)] +
[u
y
(x, y +
n
) u
y
(x +
n
, y + y
n
)]
_
since u, v satisfy the CR equations. But u
x
, u
y
, v
x
, v
y
are all continuous functions so
u
x
(x +
n
, y + y
n
) u
x
(x, y),
v
x
(x +
n
, y + y
n
) v
x
(x, y),
v
y
(x, y +
n
) v
y
(x +
n
, y + y
n
) 0,
and u
y
(x, y +
n
) u
y
(x +
n
, y + y
n
) 0,
and y
n
/z
n
is bounded ([y
n
/z
n
[ 1) so
y
n
z
n
[v
y
(x, y +
n
) v
y
(x +
n
, y + y
n
) + [u
y
(x, y +
n
) u
y
(x +
n
, y + y
n
)] 0,
also. Hence
f(w + z
n
) f(w)
z
n
u
x
(x, y) + iv
x
(x, y)
for all sequences z
n
0, and it follows that f is dierentiable at w, with f

(w) as claimed.
23
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
6 Harmonic functions and conformal transformations
Denition 6.1 Let w(x, y) be a C
2
-function of two real variables
4
. Then w is a harmonic func-
tion if it satises Laplaces equation:
w
xx
+ w
yy
= 0.
Example 6.2 Let w(x, y) = x
2
y
2
. Then w
xx
= 2 and w
yy
= 2, so w
xx
+ w
yy
= 0.
Laplaces equation is important in potential theory and many other areas, and so, since they
are solutions of it, harmonic functions are of particular interest. Complex analysis provides many
examples of harmonic functions, as the next theorem shows.
Theorem 6.3 If f : D C is holomorphic, and u and v are C
2
-functions, then u and v are both
harmonic:
u
xx
+ u
yy
= 0 and v
xx
+ v
yy
= 0.
Proof: We use the C-R equations:
u
xx
+ u
yy
= (u
x
)
x
+ (u
y
)
y
= (v
y
)
x
+ (v
x
)
y
= v
xy
v
yx
= 0.
Similarly for v.
Remark 6.4 We will see later that the hypothesis that u and v are C
2
is actually redundant,
because u and v are always C
2
when f is holomorphic. In fact, theyre always smooth (i.e. innitely
dierentiable)!
Conversely to Theorem 6.3, given a harmonic function u there is locally (i.e. in some -neighbourhood
of any point in the domain of u) a harmonic function v such that f(x + iy) = u(x, y) + iv(x, y) is
holomorphic. Such a v is called a harmonic conjugate of u.
Example 6.5 Let u(x, y) = 2xy, then u is harmonic. We can construct v using the C-R equations
(theyre very useful, arent they!). Because
2y = u
x
= v
y
= v = y
2
+ g(x) where g is a function of x,
and
2x = u
y
= v
x
= v = x
2
+ h(y) where h is a function of y,
we can deduce that v = y
2
x
2
+ C where C is a constant.
Example 6.6 Sometimes an easier strategy than solving the CR equations directly is to compute
f

(z) = u
x
+ iv
x
= u
x
iu
y
,
write it in terms of z, then try to spot the holomorphic f(z) whose derivative is f

(z).
4
A C
k
-function is a function that is k times continuously dierentiable.
24
Lets try this on u(x, y) =
x
x
2
+y
2
, which is harmonic on R
2
(0, 0):
f

(z) = u
x
iu
y
=
(y
2
x
2
) + i2xy
(x
2
+ y
2
)
2
=
z
2
z
2
z
2
=
1
z
2
hence f(z) =
1
z
.
Summary
f : D C is holomorphic means f is complex dierentiable at every point in D.
(Cauchy-Riemann equations) If f(z) = u(x, y) + iv(x, y) is holomorphic, then
u
x
= v
y
and v
x
= u
y
. (CR)
Conversely, if u, v satisfy (CR) and are continuously dierentiable, then
f(x + iy) = u(x, y) + iv(x, y) is holomorphic.
The function w : R
2
R is a harmonic function if it satises Laplaces equation:
w
xx
+ w
yy
= 0.
For every harmonic u there is a harmonic function v (called a harmonic conjugate of u) such
that f(x + iy) = u(x, y) + i(v, y) is holomorphic.
25
A wonderful consequence of Theorem 6.3 and its converse, as stated above, is the following.
Proposition 6.7 (Conformal invariance) Let : V R be a harmonic function on an open set
of C and let f : D C be a holomorphic function on an open set of C with image in V . Then the
composite function f is harmonic.
Proof: Locally, is the real part of a holomorphic function g, say. Then f is the real part of
the composite function g f. But this is the composition of two holomorphic functions and so, by
the Chain Rule (Chapter 5) is holomorphic, hence, by Theorem 6.3, f is harmonic.
Suppose now that we want to nd a harmonic function which is zero on the boundary of some
open set, e.g., on the boundary of the sector 0 < arg z < /3. The mapping w = f(z) given by
w = z
3
transforms this region in the z-plane onto the half plane Imw > 0 in the w-plane. A
harmonic function which is zero on the boundary of that half plane is (u, v) = v = Imw. Hence
the composition f is given by (x, y) = 3x
2
y y
3
. This is a harmonic function which is zero on
the boundary of the sector.
Thus we can transform nding harmonic functions on a dicult region into nding them on a
simpler region where we can often solve our problem.
What do the transformations given by holomorphic functions f look like?
Example 6.8 Almost the simplest holomorphic function w = f(z) is the map w = az where a is a
non-zero complex number. Since [w[ = [a[[z[ and arg w = arg a + arg z ( mod 2) for z ,= 0, f is a
combination of magnication by [a[ and rotation about the origin through an angle arg a. . Note
how the orthogonal grid of coordinate lines x = const and y = const maps to another orthogonal
grid of lines.
As special cases, if a is real, f is just magnication by [a[; if a = e
i
then f is rotation about the
origin through .
Example 6.9 Consider the holomorphic function w = z
2
. We shall draw the images of some
coordinate lines.
Example 6.10 Consider the holomorphic function w = 1/z (z ,= 0). This time, we draw the
images of the circles and radii.
26
In both cases, the holomorphic functions map an orthogonal grid to an orthogonal grid. Why? In
the sequel, let f : D C be a complex function which is complex dierentiable at z
0
= x
0
+iy
0
D.
Explanation 1: Why the coordinate lines map to an orthogonal grid. The coordinate
line y = y
0
maps to the curve x
_
u(x, y
0
), v(x, y
0
)
_
which has tangent vector (u/x, v/x) and
the coordinate line x = x
0
maps to the curve y
_
u(x
0
, y) which has tangent vector (u/y, v/y);
here all partial derivatives are evaluated at (x
0
, y
0
). Let us form the dot product of the two tangent
vectors:
(u/x, v/x) (u/y, v/y) =
=
u
x
u
y
+
v
x
v
y
=
u
x
v
x
+
v
x
u
x
= 0.
From the Cauchy-Riemann equations, this is zero, so the two tangent vectors are orthogonal; in
fact the second is obtained from the rst by rotation through +/2. Note this argument only works
if f

(z
0
) ,= 0, otherwise the two vectors (u/x, v/x), (u/y, v/y) are zero.
Explanation 2: Other curves. What does f do to other curves? With explanations to follow,
consider a curve in the z-plane which has a tangent line at z
0
. Let z ,= z
0
be a nearby point.
Then as z z
0
the direction of the chord from z
0
to z tends to the tangent direction at z
0
. Write
w = f(z) and w
0
= f(z
0
). Then, by denition, the quotient (w w
0
)/(z z
0
) tends to f

(z
0
).
If f

(z
0
) ,= 0, the quotient is non-zero for z close enough to z
0
, so we may take arguments to give
arg(ww
0
)/(zz
0
) arg f

(z
0
) (mod 2), i.e., arg(ww
0
)arg(zz
0
) arg f

(z
0
) (mod 2).
This shows that f rotates the tangent to the curve at z
0
through an angle of arg f

(z
0
), in particular,
the angle between the tangents to two curves through z
0
is equal to the angle between the tangents
of the images of the two curves under f. (Here we measure angle from the rst curve to the second
curve, as usual counting it positive in the anticlockwise direction.)
To make this rigorous, we must dene what we mean bycurve and its tangent. By a parametrized
curve or path in C we mean a continuous mapping from an interval I of R to C, for z I,
we shall write z = (t). Let t
0
I and write z
0
= (t
0
) . We say is dierentiable at t
0
if

(t
0
) = lim
_
f(t) f(t
0
)
__
(t t
0
) = lim(z z
0
)
_
(t t
0
) exists. Equivalently, writing in terms of
real and imaginary parts as (t) = x(t) +iy(t), then is dierentiable at t
0
if and only if x

(t
0
) and
y

(t
0
) exist, and then

(t
0
) = x

(t
0
) + iy

(t
0
).
Then if f

(t
0
) is non-zero, it is called the tangent (vector) and the line through z
0
= f(t
0
) in
direction f

(t
0
) is called the tangent line at z
0
. Note that, for any t ,= t
0
if t is close to t
0
then
z ,= z
0
and arg(z z
0
) = arg(z z
0
)/(t t
0
) arg f

(z
0
), so the limit of arg(z z
0
) gives the
direction of the tangent, clarifying the above argument.
An alternative way to clarify it is to note that we have a chain rule:
(f )

(t
0
) = f

(z
0
)

(t
0
) .
Comparing with Example 6.8, this shows that the tangent at w
0
= f(z
0
) to the image f of is
obtained from that of at z
0
by rotating through arg f

(z
0
) and magnifying by [f

(z
0
)[.
27
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
7 The Riemann integral
7.1 Denitions
We now switch attention to real-valued functions of a real variable. Since the start of the calculus, in
the 17th century, two processes have been involved. Geometrically we may regard them as (i) nding
the tangent to a graph (dierentiation) and (ii) nding the area between a graph and the x-axis
(integration). The subject really got going when the relationship between the two was discovered,
in the form of the Fundamental Theorem of the Calculus which states
d
dt
_
t
a
f(x) dx = f(t)
when f is continuous. Here, the meaning of
_
t
a
f(x) dx is the signed area between the graph of f and
the horizontal axis. The rst thing well do is to make this notion precise and to dene a class of
functions, the Riemann integrable functions on a closed interval, for which the theory works. This
class includes, for instance, all functions continuous on that interval and all step functions.
In what follows, [a, b] will be a closed interval in R with a < b and f : [a, b] R will be a
bounded function, that is, K > 0 such that [f(x)[ K for all x [a, b]. This includes functions
like x x
2
on [1/2, 1] but not on [0, 1].
We wish to dene the denite integral
_
b
a
f(x)dx in such a way that it measures area. But how
do we do that? The key is to chop or dissect the interval into small (sub-)intervals and add up
the areas of the rectangles we get. There are two sensible choices of rectangles, the ones which
overestimate the area, and the ones which underestimate it. The give estimates for the area known
as the upper and lower Riemann sums. Precisely,
7.1 Denitions A dissection of a closed interval [a, b] R into n (sub-)intervals (not necessarily
of equal width) is the ordered set of distinct points (forming the end-points of the intervals) T =
a
0
, a
1
, a
2
, . . . , a
n
, where a = a
0
< a
1
< a
2
< < a
n
= b. For a bounded function f[a, b] : R,
the upper and lower Riemann sums, u
D
(f) and l
D
(f) corresponding to T, are given by
u
D
(f) =
n1

j=0
M
j
(a
j+1
a
j
) and l
D
(f) =
n1

j=0
m
j
(a
j+1
a
j
)
where
M
j
= supf(x) : x (a
j
, a
j+1
) and m
j
= inff(x) : x (a
j
, a
j+1
).
7.2 Examples
(i) Its often easiest to take the points equally spaced, i.e. all intervals are of equal width
such a dissection is called a regular dissection. For example, suppose that a = 0, b = 1 and
f(x) = x. For any n 1, 2, . . ., let T
n
denote the dissection into n equal intervals, thus
T
n
= 0, 1/n, 2/n, . . . , (n 1)/n, 1. Then u
Dn
=

n
j=1
j/n
2
= n(n + 1)/(2n
2
) = (n + 1)/2n
and l
Dn
= (n 1)/2n. Note that, in the limit, these tend to 1/2, which is the area of under
the graph from the formula for the area of a triangle. We draw a picture:
28
(ii) Again consider f(x) = x with now with b > a arbitrary. As before, we take a regular dissection
T
n
= a, a + (b a)/n, . . . , a + j(b a)/n, . . . , b. Then M
j
= (a + (j + 1)(b a)/n) and
m
j
= (a + j(b a)/n), giving u
Dn
= ((b a)/n)

n1
j=0
(a + (j + 1)(b a)/n) and l
Dn
=
((b a)/n)

n1
j=0
(a + j(b a)/n).
As n
u
Dn
(b
2
a
2
)/2 , see Ex 7.9(ii)
l
Dn
(b
2
a
2
)/2 .
(iii) Let a < c < b be points in R. Suppose f(x) = d when x = c and 0 otherwise. Then, it no
longer makes sense to take a regular dissection. Instead, we take T = a, c, b and we get
u
D
= l
D
= 0. So the spike of zero width at c does not contribute to the area; you would be
worried if it did.
(iv) Let a < c < b be points in R and let f(x) = d
1
when a < x < c and f(x) = d
2
when c < x < b.
Then, with the same dissection as before, u
D
= l
D
= d
1
(c a) + d
2
(b c).
(v) Let a = 0, b = 1 and dene f by f(x) = 1 if x is irrational, f(x) = 0 if x is rational. For any
dissection T, u
D
= 1 and l
D
= 0. So can we make any sense of the area under this function?
Now, to get a better estimate of the area, it makes sense to take more points in our dissection.
We see what this does. First, we add one more point.
7.3 Lemma Let [a, b] be a closed interval in R, let f : [a, b] R be a bounded function and let T
be a dissection of [a, b]. Let T

= T c. Then u
D
u
D
and l
D
l
D
.
Proof: If c T, there is nothing to prove. Otherwise, a
j
< c < a
j+1
for some j. Then u
D

u
D
= M
j
(a
j+1
a
j
)
_
M

j
(c a
j
) + M

j+1
(a
j+1
c)
_
, where M
j
= supf(x) : x (a
j
, a
j+1
),
M

j
= supf(x) : x (a
j
, c), and M

j+1
= supf(x) : x (c, a
j+1
). Clearly, M

j
, M

j+1
M
j
, so
u
D
u
D
M
j
(a
j+1
a
j
) M
j
(a
j+1
c + c a
j
) = 0. So u
D
u
D
. Similarly l
D
l
D
.
Next, we replace T by a ner dissection T

T.
7.4 Lemma Let [a, b] be a closed interval in R, let f : [a, b] R be a bounded function and let
T, T

be dissections of [a, b] satisfying T T

. Then u
D
u
D
and l
D
l
D
.
Proof: We obtain T

from T by adding a nite number of points to T, so this lemma follows from

7.3 by induction.
7.5 Corollary Let f : [a, b] R satisfy m f(x) M for all x [a, b]. Then, for any dissections
T, T

, we have
m(b a) l
D
u
D
M(b a).
Proof: First, note that T T T

and T

T T

.
Let T
0
:= a, b. Then T
0
T, so
m(b a) l
D0
(since l
D0
= m
0
(b a) = inff(x) : x (a, b)(b a) and m m
0
)
l
D
by 7.4
l
DD
by 7.4
u
DD

u
D
by 7.4
u
D0
by 7.4
M(b a) since u
D0
= M
0
(b a) M(b a).

29
7.6 Corollary For [a, b] and f as in 7.5, if we put l = sup l
D
and u = inf u
D
, where the supremum
and innum are taken over the set of all dissections of [a, b], then both l and u exist and m(b a)
l u M(b a).
Proof: The set of real numbers l
D
: T a dissection is bounded below by m(b a) and bounded
above by u
D
, for any T so l = sup l
D
exists and m(b a) l u
D
M(b a), for all T. Hence
u = inf u
D
exists and m(b a) l u M(b a).
7.7 Denition Let a > b and let f : [a, b] R be a bounded function. We say f is Riemann
integrable on [a, b] if sup l
D
= inf u
D
(i.e., l = u in the notation of 7.6), and we denote their common
value by
_
b
a
f(x) dx. (We may shorten this to
_
b
a
f.)
7.8 Corollary If f is Riemann integrable and m f(x) M for all x [a, b] then m(b a)
_
b
a
f M(b a).
Proof: This follows immediately from the denition and 7.6.
7.9 Examples We use the examples of 7.2
(i) f(x) = x with a = 0 and b = 1. For the regular dissection into n intervals, we have
u
Dn
= (n+1)/(2n) 1/2 ; l
Dn
= (n1)/(2n) 1/2 .
(ii) f(x) = x with b > a arbitrary. Again we take a regular dissection T
n
= a, a+(ba)/n, . . . , a+
j(b a)/n, . . . , b. Then M
j
= a + (j + 1)
b a
n
and m
j
= a + j
b a
n
, giving
u
Dn
= ((b a)/n)

n1
j=0
(a + (j + 1)(b a)/n) = a(b a) + ((b a)
2
/n
2
)

n1
j=0
(j + 1) =
a(b a) + (n + 1)(b a)
2
/(2n) and, similarly, l
Dn
= a(b a) + (n 1)(b a)
2
/(2n). Both
u
n
and l
n
a(b a) + (b
2
2ab + a
2
)/2 = (b
2
a
2
)/2, so f is Riemann integrable and
_
b
a
xdx = (b
2
a
2
)/2.
(iii) Let a < c < b be points in R. Let f(x) = d
1
when a < x < c and f(x) = d
2
when c < x < b.
then, with the same dissection as before u
D
= l
D
= d
1
(c a) +d
2
(b c). Hence f is Riemann
integrable and
_
b
a
f(x) dx = d
1
(c a) + d
2
(b c).
(iv) Let a < c < b be points in R. Suppose f(x) = d when x = c and 0 otherwise. Then, taking
T = a, c, b we get u
D
= l
D
= 0. Hence f is Riemann integrable and
_
b
a
f(x) dx = 0.
(v) a = 0, b = 1. f(x) = 1 if x is irrational, f(x) = 0 if x is rational. For any dissection T,
u
D
= 1 and l
D
= 0. So u = 1 and l = 0, and f is not Riemann integrable. Remark for lovers
of analysis only. There is, however, a more powerful, but much harder, theory of Lebesgue
integration which allows us to ignore values on countable sets such as the rational numbers,
and indeed, on the class of so-called null sets which includes the Cantor middle third set. All
reasonable real-valued bounded functions on an interval [a,b] are Lebesgue-integrable: to get
something not Lebesgue-integrable requires the axiom of choice.
7.10 Theorem (Linearity of the integral) Let f, g : [a, b] R be Riemann integrable functions.
Then (i) f + g is Riemann integrable, (ii) for any c R, cf is Riemann integrable, and
_
b
a
(f + g) =
_
b
a
f +
_
b
a
g ,
_
b
a
cf = c
_
b
a
f.
30
Proof: (i) First recall that, for all x [a, b] we dene (f +g)(x) := f(x)+g(x). Now l
D
(f)+l
D
(g) =

n1
j=0
(m
j
(f) + m
j
(g))(a
j+1
a
j
)

n1
j=0
m
j
(f + g)(a
j+1
a
j
) = l
D
(f + g). (This is because,
on any interval, inf f(x) + inf g(x) f(x) + g(x) which gives inf f + inf g inf(f + g).) Similarly
u
D
(f + g) u
D
(f) + u
D
(g).
Thus, for dissections T

, T

and their union T = T

, we have
l
D
(f) + l
D
(g) l
D
(f) + l
D
(g) by 7.4
l
D
(f + g)
l(f + g)
u(f + g)
u
D
(f + g)
u
D
(f) + u
D
(g)
u
D
(f) + u
D
(g) by 7.4.
So, taking suprema over all dissections T

, T

, we get
l(f) + l(g) l(f + g) u(f + g) u(f) + u(g) = l(f) + l(g),
and the two outer terms are equal, which proves that f + g is Riemann integrable and
_
(f + g) =
_
f +
_
g.
(ii) Again (cf)(x) := c(f(x)), for all x [a, b]. For c = 0 everything reduces to 0, so the result is
trivially true. For c > 0, l
D
(cf) = cl
D
(f) and u
D
(cf) = cu
D
(f), so cf is Riemann integrable and
_
b
a
cf = c
_
b
a
f. Then, for c = 1, l
D
(f) = u
D
(f) and u
D
(f) = l
D
(f), so f is Riemann
integrable and
_
b
a
(f) =
_
b
a
f. So, nally, for negative c, cf = [c[(f) and the result follows by
combining the other two cases.
For a positive function, the Riemann integral
_
b
a
f gives (indeed denes) the area between the
graph of the function, the x-axis and the two vertical lines x = a and x = b. Also, if M = supf(x) :
x [a, b] we have seen that 0
_
b
a
f M(b a) for positive f. Here is a further property that is
suggested by this interpretation.
7.11 Theorem Let f, g : [a, b] be Riemann integrable functions such that 0 f(x) g(x), for all
x [a, b]. Then
0
_
b
a
f
_
b
a
g.
Proof: Certainly 0
_
b
a
f, since 0 is a lower bound for f. Now g(x) f(x) 0, for all x [a, b], so
_
b
a
g
_
b
a
f =
_
b
a
(g f) 0.
Thus
0
_
b
a
f
_
b
a
g.

7.12 Remarks (i) The signicance of all this is that for positive bounded functions, the denition
of area under the graph satises properties that we expect of area. (ii) If f(x) 0 for all x [a, b],
then f 0 so
_
(f) =
_
f 0 and we have
_
f 0. So if the graph is below the axis,
(Riemann) integration gives a negative (or zero) quantity.
31
8 Integrability of step functions and continuous functions
8.1 Denition Let I be a bounded interval in R. The characteristic function
I
of the interval is
the function which is one on I and zero elsewhere. Here is a graph of such a function.
A step function is a function of the form

k
j=1
a
j

Ij
for some collection of bounded intervals I
j
,
which may be open or closed or half-open. We can choose the intervals to be non-overlapping
but this is not necessary for what follows. If they are non-overlapping then the step function has
the value a
j
on I
j
.
8.2 Example (i) With I
1
= [0, 1], I
2
= [1, 2], f = 2
I1
+ 5
I2
looks like ...
(ii) With I
1
= [0, 1], I
2
= [0, 2], f = 2
I1
+ 5
I2
looks like ...
8.3 Proposition Let f : [a, b] R be a step function. Then f is Riemann integrable.
In fact, if f =

l
k=1
a
j

I
k
for some intervals I
k
[a, b], then
_
b
a
f =

l
k=1
a
k
[I
k
[, where [I
k
[
denotes the length of I
k
.
Proof: We rst show that if I [a, b], then
I
is integrable and
_
b
a

I
= [I[. Indeed if the end points
of the interval I are c, d with c d, we just take the dissection T = a, c, d, b (amend if necessary
to accommodate any coincidence of points) which has l
D
= u
D
= [I[. So
_
b
a

I
= [I[. The result for
a general step function follows from the linearity of the integral (Theorem 7.10).
This result shows that you can nd a dissection T such that l
D
(f) = u
D
(f) =
_
b
a
f if and only
if f is a step function. For other functions l
D
(f) < u
D
(f) and they only approximate the integral.
32
We now proceed to show that continuous functions are Riemann integrable. The rst step is to
provide a test for integrability in terms of upper and lower sums l
D
(f), u
D
(f).
8.4 Integrability Criterion Let f : [a, b] R be a bounded function. Then f is Riemann
integrable if and only if, for each > 0 there is a dissection T such that u
D
l
D
< .
Proof: If f is integrable, then, given > 0 there exist dissections T
0
, T
1
such that
_
b
a
f l
D0
< /2
and u
D1

_
b
a
f < /2. Putting T = T
0
T
1
, by 7.4, l
D0
l
D

_
b
a
f, so
_
b
a
f l
D
< /2. Similarly,
u
D

_
b
a
f < /2, so u
D
l
D
< .
Conversely, if the criterion is satised, then u l < , for every > 0, because for any dissection T
we have ul u
D
l
D
. Thus ul = 0, giving u = l which means that f is Riemann integrable.
Note that u
D
l
D
=

n
j=0
(M
j
m
j
)(a
j+1
a
j
). If we can nd T such that M
j
m
j
< /(ba),
for each j we get u
D
l
D
<

n
j=0
(a
j+1
a
j
)/(b a) = . Now the denition of continuity almost
does this for us: if f is continuous then, given > 0, there exists for each x a = (, x) > 0 such
that [y x[ < (, x) [f(y) f(x)[ < /(b a). If we could nd independent of x then we could
choose T so that a
j+1
a
j
< for j = 0, 1 . . . , n 1 and everything would work.
In fact we can do this, as the next result shows.
8.5 Theorem (Uniform Continuity) Let f : [a, b] R be continuous. Given > 0 there exists
> 0, depending only on and f, such that, for all x, y [a, b] we have [xy[ < [f(x)f(y)[ < .
Proof: We use the compactness of [a, b] and argue by contradiction.
Suppose the result is not true. Then, there is an > 0 such that for each n N, we have
x
n
, y
n
[a, b] with [x
n
y
n
[ < 1/n but [f(x
n
) f(y
n
)[ . Since [a, b] is compact, by Bolzano
Weierstrass and the subsequent theorem, we can nd a convergent subsequence (x
n
k
) of (x
n
), with
x
n
k
x [a, b] , say. Since f is continuous at x, there is a > 0 such that y [a, b] and
[x y[ < [f(x) f(y)[ < /2. Choose N so that 1/N < /2. Choose k such that n
k
> N and
[x
n
k
x[ < /2. Then, on the one hand [f(x
n
k
) f(y
n
k
)[ .
But, on the other hand, [x x
n
k
[ < /2 so [f(x) f(x
n
k
)[ < /2 and [x y
n
k
[ [x x
n
k
[ +
[x
n
k
y
n
k
[ < /2 + /2 = , giving [f(x) f(y
n
k
)[ < /2. Thus [f(x
n
k
) f(y
n
k
)[ [f(x
n
k
)
f(x)[ +[f(x) f(y
n
k
)[ < /2 + /2 = . We have a contradiction, so the theorem is true.
The argument preceding 8.5 gives us the following corollary.
8.6 Theorem Let f : [a, b] R be continuous. Then f is Riemann integrable.
The property of 8.5 is so important that it is given a name.
8.7 Denition Let f be a real-valued function of a real variable and let S R. Then f is called
uniformly continuous on S if for each > 0 there is a > 0 such that for all x, y S, we have
[x y[ < [f(x) f(y)[ < .
Of course we can make the same denition for functions of a complex variable. We can now
restate 8.5 as
8.8 Theorem Let f : [a, b] R be continuous. Then it is uniformly continuous on [a, b].
8.9 Examples Some functions are even uniformly continuous on R, for example f(x) = sin x. Since
f is dierentiable, we can use the mean value theorem to show that [ sin x sin y[ = [f

(c)[[x y[
for some c between x and y. But [f

(c)[ = [ cos c[ 1 so [ sin x sin y[ [x y[. and [x y[ < <

[ sin x sin y[ < .
This argument works for any dierentiable function whose derivative is bounded on an interval.
33
Some extra material for lovers of analysis only
In contrast to the last examples, f(x) := 1/x and g(x) := sin(1/x), though continuous on the
half-open interval (0, 1] are not uniformly continuous there.
For f we can see this by considering the derivative: f

(x) = x
2
. For 0 < x < y < 1,
f(x) f(y) = (x y)f

(c), for some x < c < y and f

(c) < f

(y), so [f(x) f(y)[ = f(x) f(y) >

(xy)(y
2
) = (y x)y
2
. Taking = 1, we see that [f(x) f(y)[ < 1 [y x[ < y
2
and y
2
may
be arbitrarily small, so f is not uniformly continuous.
For g, we note that sin(1/x) = 0 when x = 1/n and sin(1/x) = 1 whenx = 1/((n+1/2)). So
[g(1/n) g(1/(n + 1/2))[ = 1 and 1/n 1/(n + 1/2) = (2n(n + 1/2))
1
, which is arbitrarily
small as n increases. So g is not uniformly continuous.
8.10 Question Is
f(x) =
_
xsin(1/x) x ,= 0
0 x = 0
uniformly continuous on [0, 1]? Can you prove this directly?
As if this werent enough complication, in certain contexts it turns out to be necessary to go
further than this and look precisely how is related to . For instance, the Lipschitz functions are
those for which, for some xed constant K > 0 we have [f(x) f(y)[ K[x y[. This condition
plays an important role in the theory of dierential equations.
34
8.11 Proposition Let f : [a, b] R be a bounded function. Let a < c < b. Then f is Riemann
integrable on [a, b] if and only it is Riemann integrable on [a, c] and [c, b]. In that case, we have
_
b
a
f =
_
c
a
f +
_
d
c
f.
Proof: If f is Riemann integrable on [a, b], then, by the integrability criterion, given > 0, there
is a dissection T = a
0
, . . . , a
n
of [a, b] such that u
D
l
D
< . Now consider the corresponding
dissection T

= T [a, c] = a
0
, . . . , a
k
, c of [a, c]. Now u
D
l
D
u
D
l
D
, because on each
(a
j
, a
j+1
) with j = 1, . . . , k 1, M
j
m
j
is the same and on the last interval (a
j
, c) we have
sup f(x) inf f(x) M
k
m
k
and c a
k
a
k+1
a
k
. So u
D
l
D
< , satisfying the integrability
criterion, and thus f is integrable on [a, c]. Similarly, f is integrable on [c, b].
Conversely, suppose that f is integrable on both [a, c] and [c, b]. Given > 0, let T and T

be
dissections of [a, c] and [b, c], respectively, such that u
D
l
D
< /2 and u
D
l
D
< /2. Since
l
D

_
c
a
f u
D
we get
_
c
a
f /2 l
D
u
D

_
c
a
f + /2, and similarly
_
b
c
f /2 l
D
u
D

_
b
c
f + /2.
Consider the dissection of [a, b] given by T T

. We get
_
c
a
f(x) dx +
_
b
c
f(x) dx l
D
+ l
D

= l
DD

u
DD

= u
D
+ u
D

_
c
a
f(x) dx +
_
b
c
f(x) dx + ,
so f is integrable on [a, b] and
_
b
a
f =
_
c
a
f +
_
b
c
f.
8.12 Remark It follows now that If f 0 on [a, c] and f 0 on [c, b] then
_
b
a
f =
_
c
a
f
_
b
c
(f).
Extending this to functions which are alternatively positive and negative on adjacent bounded
intervals, we see that the integral represents the signed area between the graph and the x-axis
(positive above, negative below). So we get results like
_

cos xdx = 0 and

_

sin xdx = 0
35
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
9 The Fundamental Theorem of the Calculus
We have dened integration in a way that measures area. The Fundamental Theorem of the Calculus
makes precise in what sense integration and dierentiation are inverse operations when applied to
continuous functions. The rst form of the theorem explains how dierentiation is a left inverse of
integration. This means
9.1 Fundamental Theorem of the Calculus (rst form) Let f : [a, b] R be continuous.
Then,
_
t
a
f is a dierentiable function of t for t (a, b), and
d
dt
_
t
a
f(x) dx = f(t).
Proof: For h ,= 0, consider
1
h
_
_
t+h
a
f
_
t
a
f
_
=
_

_
1
h
_
t+h
t
f if h > 0,
1
h
_
t
t+h
f if h < 0.
Since f is continuous, given > 0, we can nd > 0 such that x [t , t +] [f(x) f(t)[ < .
For h with [h[ < , we have
f(t) <
1
h
_
t+h
t
f < f(t) +
if h > 0, and
f(t) <
1
h
_
t
t+h
f < f(t) +
if h < 0. Thus

1
h
_
_
t+h
a
f
_
t
a
f
_
f(t)

< ,
which proves the result.
The second form of the theorem explains in what sense dierentiation is a right inverse of
integration, up to a constant. This means
36
9.2 Fundamental Theorem of the Calculus (2nd form) Let I R be an open interval. Let
f : I R be a continuous function and suppose that F : I R is a function such that F

= f (we
call F an antiderivative of f). Then for any a < b I, we have
_
b
a
f(x) dx = F(b) F(a).
Proof: Let c I with c < a. Put G(t) =
_
t
c
f F(t). Then G

(t) = 0, by 9.1 and the denition

of F. By the Mean Value Theorem, it follows that G(b) G(a) = (b a)G

(t
0
) = 0 for some t
0
between a and b. So G(b) = G(a), that is,
_
b
c
f F(b) =
_
a
c
f F(a), so
_
b
a
f =
_
b
c
f
_
a
c
f = F(b) F(a).

We can drop the restriction a < b if we make the following denitions:

9.3 Denitions
_
a
a
f = 0 and, for a < b,
_
a
b
f =
_
b
a
f.
9.4 Examples
(i)
_
2
1
[x[ dx =
_
0
1
(x) dx +
_
2
0
xdx =
_
x
2
2
_
0
1
+
_
x
2
2
_
2
0
= 0 +
1
2
+ 0 + 2 =
5
2
.
(ii)
_

[x[ cos xdx =

_
0

xcos xdx +
_

0
xcos xdx = 2
_

0
xcos xdx
=
_
2xsin x
_

0
2
_

0
sin x = 2
_
cos x
_

0
= 4.
(iii) By 9.1,
d
dt
_
t
0
sin xdx = sin t .
(iv) By 9.1,
d
dt
_
t
0
e
x
2
dx = e
t
2
.
(v)
d
dt
_
t
2
0
e
x
2
dx = 2te
t
2
.
37
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
10 Improper Integrals
The Fundamental Theorem applies to integration over a nite range of integration and to functions
continuous on that range. We can relax these conditions to make sense of many important integrals.
Lets rst deal with the case of an innite range.
10.1 Denition Let f : [a, ) R be Riemann integrable on every nite interval [a, b], for
a < b < . We dene
_

a
f := lim
X
_
X
a
f ,
provided that the limit exists, in which case we say that the integral converges. If the limit does not
exist, we say the integral diverges or doesnt converge.
Geometrically, for a positive (or even, non-negative) function, we are saying that the integral
converges if the area under the (innitely long) curve is nite. For a function which changes sign,
we dont have this interpretation as areas below the axis might cancel with areas above the axis.
10.2 Examples
(i) Investigate the convergence of
_

1
x
t
dx for t R.
_
X
1
dx
x
t
=
_
log X if t = 1,
1/(t 1) X
1t
/(t 1) if t ,= 1.
So
_

1
dx
x
t
diverges when t 1, and converges for t > 1 with
_

1
dx
x
t
=
1
t 1
.
(ii)
_

0
1
1 + x
2
= lim
X
tan
1
X 0 =

2
.
10.3 Denition Suppose now that f is Riemann integrable over any nite interval. We say
_
a

f
converges if
_
a

f = lim
Y
_
a
Y
f exists. We say that
_

f converges if both
_

0
f and
_
0

f
converge; then
_

f =
_
0

f +
_

0
f.
10.4 Examples
(i)
_

sin xdx doesnt exist because

_

0
sin xdx doesnt exist. However,
_
X
X
sin xdx = 0, so its
not enough to investigate
_
X
X
f in general: we have to look more closely.
(ii)
_

1
1 + x
2
= lim
X,Y
_
X
Y
dx
1 + x
2
= lim
X
tan
1
X lim
Y
tan
1
(Y ) =

2

_

2
_
= .
We now give a test for convergence of improper integrals which is analogous to a test for conver-
gence of innite series.
38
10.5 The comparison test Let f : [a, ) R
+
be a positive function which is Riemann
integrable on every nite subinterval.
(i) If there is a Riemann integrable function g such that g(x) f(x) for all x [a, ) and
_

a
g
converges, then
_

a
f converges.
(ii) If there is a positive Riemann-integrable function h such that f(x) h(x) for all x [a, )
and
_

a
h diverges, then
_

a
f diverges.
Proof: Part (ii) is easier than the rst; indeed, we have
_
X
a
f
_
X
a
h. Since
_
X
a
h it follows
that
_
X
a
f , (To prove this, given K > 0 there exists X(K) such that for all X > X(K), we
have
_
X
a
f
_
X
a
h > K: note that
_
X
a
h is an increasing function of X, because h 0.]
Now for part (i). The condition on f means that
_
X
a
f
_

a
g. But
_
X
a
f is an increasing
function of X, and weve shown its bounded above, therefore it tends to a limit. [The limit is
sup
_
X
a
f = l, say: this exists, since the integral is bounded above: then the standard property of
suprema says that, given > 0, there exists X() such that
_
X()
a
f > l : then X > X(), we
have
_
X
a
f > l : hence
_
X
a
f l.]
10.6 Example
_

0
e
x
2
dx. We rst note that e
x
2
is continuous and therefore integrable on
every nite interval. Now e
x
2
= 1/(1 + x + x
2
/2 + ) 1/(1 + x
2
/2) < 2/(1 + x
2
). We have
seen that
_

0
1/(1 + x
2
) converges, thus
_

0
2/(1 + x
2
) dx converges and by the comparison test
_

0
e
x
2
dx converges.
There is a second sort of improper integral: that of an unbounded function on a nite interval :
10.7 Denition Let f : (a, b] R be unbounded, but bounded and integrable on every interval
[c, b] for a < c < b. Then
_
b
a
f = lim
0+
_
b
a+
f, if the limit exists, in which we case we say the
interval converges.
10.8 Example
_
1
0
x
t
dx =
So the integral converges when
It diverges otherwise.
10.9 Comparison test Suppose f : (a, b] R
+
is as in 10.7. If g : (a, b] R
+
is such that
g(x) f(x) 0 for all x (a, b], and
_
b
a
g converges, then
_
b
a
f converges. On the other hand, if
h : (a, b] R
+
is such that 0 h(x) f(x) for all x (a, b], and
_
b
a
h diverges, then
_
b
a
f diverges.
Proof: Omitted. Its very similar to the proof of 10.5.
10.10 Denition Let f : [a, b) R be unbounded, but bounded and integrable on every interval
[a, c], with a < c < b. Then
_
b
a
f := lim
0+
_
b
a
f, provided this limit exists, in which case we say
that the integral converges.
10.11 Denition Let f : [a, b] R and let a < c < b be such that f, though unbounded, is
bounded on every closed interval not including c. Then
_
b
a
f converges if and only if both
_
c
a
f and
_
b
c
f converge, in which case
_
b
a
f :=
_
c
a
f +
_
b
c
f.
39
10.12 Example 1/x
2
has a singularity at 0, so
_
b
a
1
x
2
dx =
_
0
a
1
x
2
dx +
_
b
0
1
x
2
dx
diverges for a, b > 0. Note however that if we blindly apply the fundamental theorem, we get
_
2
1
1
x
2
dx =
_

1
x
_
2
1
=
1
2
1 =
3
2
,
which is nonsensical, because the area under a positive curve must always be positive.
The next test compares the integral with an innite series and can be used to prove the conver-
gence of the one given the convergence of the other.
10.13 Integral Test Let f : [0, ) R
+
be a positive, decreasing function (that is, x
1
< x
2

f(x
1
) f(x
2
) 0) which is integrable on every nite interval. Then
_

0
f converges if and only if

n=0
f(n) converges.
Proof: Let k > 1 be a positive integer. Consider the regular dissection T
k
= 0, 1, 2, . . . , k of the
interval [0, k]. Then the lower and upper Riemann sums for this dissection satisfy
l
D
k
=
k1

n=0
f(n + 1)
_
k
0
f u
D
k
=
k1

n=0
f(n).
If the integral converges, the left-hand side shows that

k
n=1
f(n) is bounded above by
_

0
f. Hence

n=1
f(n) converges, being a series of positive terms, so the partial sums are increasing and bounded
above. Thus

0
f(n) converges.
Conversely, if
_

0
f diverges, then the right-hand side of the inequality shows that

n=0
f(n)
diverges.
10.14 Example Consider

n=1
1
n
t
.
We compare with
_

0
f where f(x) := 1/(1+x)
t
; this converges if and only if t > 1, hence the series
converges i t > 1 (this is clearer than taking 1/x
t
).
40
MATH2015 Analysis 2, 200910
Notes c _ K Houston, D L Salinger, J M Speight and J C Wood 2009
Appendix: More on convergence of sequences. The GPC,
compactness and BolzanoWeierstrass
We here present two important properties of sequences of complex numbers, which will give us new
information even for sequences of real numbers. First recall that a sequence (z
n
) of complex numbers
converges to a limit w if and only if, given a real number > 0, there exists N = N() such that
n > N() [z
n
w[ < .
We now run into a problem: in the real case, every increasing sequence which is bounded above
converges to its least upper bound. This important principle expresses the completeness of the real
numbers, from which important consequences ow, for example the Intermediate Value Theorem and
Rolles Theorem. We want corresponding property holds for complex numbers, but it is meaningless
to talk about increasing sequences of complex numbers. So we need a new way of looking at
completeness. The fundamental idea is the following.
Denition 0.1 A Cauchy sequence of complex numbers is a sequence (z
n
) such that given > 0,
N() R such that
n, m > N() [z
n
z
m
[ <
The idea (due to A.A. Cauchy) is that as you go further out along the sequence, the members
of the sequence get closer and closer together.
We give some examples of sequences which are or are not Cauchy.
(i) Let s
n
= 1/n. Then [s
n
s
m
[ =

1
n

1
m

<
2
N
if n, m > N. So given > 0, choose N > 2/,
then [s
n
s
m
[ < for all n, m > N. Hence the sequence is Cauchy.
(ii) Let s
n
= (1)
n
. Then [s
n
s
m
[ = [(1)
n
(1)
m
[ = 0 if n m is even and 2 if n m is
odd. If we take = 1, say, then there exists no N such that n, m > N implies [s
n
s
m
[ < .
Proposition 0.2 Every convergent sequence of complex numbers is a Cauchy sequence.
Proof: Let z
n
z. Given > 0, N = N() R such that n > N [z
n
z[ < /2.
Then [z
n
z
m
[ = [z
n
z + z z
m
[ [z
n
z[ + [z z
m
[, by the triangle inequality. Hence
n, m > N [z
n
z
m
[ < /2 + /2 = . So (z
n
) is a Cauchy sequence.
We now set out to prove the converse. We rst reduce it to the real case.
Proposition 0.3 . Let (z
n
) be a Cauchy sequence. Then (Re z
n
) and (Im z
n
) are Cauchy sequences.
Proof: We have [Re z
n
Re z
m
[ [z
n
z
m
[ and similarly for the imaginary part, so the result
follows.
Denition 0.4 A sequence (z
n
) in the complex plane is bounded if K > 0 such that n N,
[z
n
[ < K.
i
Proposition 0.5 Every Cauchy sequence is bounded.
Proof: Let (z
n
) be a Cauchy sequence. Then N = N(1) N such that n > N [z
n
z
m
[ <
1 [z
n
z
N+1
[ < 1 [z
n
[ < [z
N+1
[ + 1. Now put K = max[z
0
[, [z
1
[, . . . [z
N
[, [z
N+1
[ + 1, Then
[z
n
[ K n.
Corollary 0.6 Every convergent sequence of complex numbers is bounded.
Noting that for real numbers [x
n
[ < K is equivalent to K < x
n
< K, we have
Corollary 0.7 A Cauchy sequence of real numbers is bounded.
Theorem 0.8 Let (x
n
) be a Cauchy sequence of real numbers. Then (x
n
) converges.
Proof: Given a sequence (x
n
), we dene the sequence (X
j
) by X
j
= infx
n
: n j. The X
j
exist by 0.7 (and the completeness of the real numbers) and the sequence (X
j
) is increasing (by the
denition) is bounded above, again by 0.7, so X
j
l = sup(X
j
), by the completeness of the real
numbers.
We claim that x
n
l. Indeed, given > 0, let N be an integer such that [X
N
l[ < /2 and
[x
m
x
n
[ < /2, n, m N. Then [X
N
x
n
[ /2, when n > N, so [x
n
l[ = [x
n
X
N
+X
N
l[
[x
n
X
N
[ +[X
N
l[ < , when n > N.
Theorem 0.9 (General Principle of Convergence for sequences of complex numbers) Every
Cauchy sequence of complex numbers converges.
Proof: We have seen (0.3) that if a sequence of complex numbers (z
n
) is a Cauchy sequence, then
so are the sequences of Real and Imaginary parts, (x
n
) and (y
n
), where z
n
= x
n
+ iy
n
. Hence, by
0.8, x
n
x and y
n
y, say. It follows that z
n
x + iy.
Three remarks for fans of analysis only
The General Principle of Convergence for real numbers is in fact equivalent to the completeness
axiom (that every increasing sequence which is bounded above converges). This follows from the
fact that the following result can be proved without using the completeness axiom.
Proposition 0.10 Let (x
n
) be an increasing sequence of real numbers which is bounded above. Then
(x
n
) is a Cauchy sequence.
The proof (by contradiction) is omitted.
Theorem 0.11 On R, The completeness axiom is equivalent to the General Principle of Conver-
gence.
Proof: We have seen that the completeness axiom implies the general principle of convergence.
Conversely, suppose the GPC holds. Then, by 0.10, an increasing sequence bounded above is a
Cauchy sequence and hence converges, so the completeness axiom holds.
Remark 0.12 For subsets S of R, if (x
n
) is an increasing sequence in S which is bounded above
but doesnt tend to a limit in S, then (x
n
) is a Cauchy sequence in S which doesnt converge to a
point of S. However, the converse does not necessarily hold: for example take S =]0, 1] and (1/2
n
),
which is a Cauchy sequence tending to 0 / S. Yet every increasing sequence in S converges to a
point of S.
ii
Non-fans please resume reading from here.
We give an application of the GPR. Recall what is means for a subset of C to be open.
0.13 Denition A set F C is called closed if its complement F
c
= C F is open.
0.14 Examples C is closed. is closed. The real and imaginary axes are closed.

D
r
(z) = w
C : [w z[ r is closed for every r > 0, z C.
Its useful to have an intrinsic denition of a closed set.
0.15 Proposition A set F is closed i for every sequence of points (z
n
) F such that z
n
w C
we have w F.
Proof: Suppose F is closed and z
n
w with z
n
F for all n. Suppose that w / F. Since the
complement of F is open, then > 0 such that D

(w) F
c
. Now z
n
w, so [z
n
w[ < for all
n large enough. Such z
n
lies in D

(w) which is in F
c
, contradicting the choice of (z
n
) as a sequence
in F. Thus w F.
On the other hand, if the criterion is satised and w / F, we have for some n N, D1
n
(w)F =
and hence F
c
is open. If not we would have a sequence z
n
F D1
n
(z) and z
n
w.
0.16 Denition A subset F of the plane is bounded if K > 0 such that z F, [z[ < K. F is
compact if F is closed and bounded.
0.17 Examples of compact sets: [a, b] R,

D
r
(z).
0.18 Theorem (Bolzano-Weierstrass) Let F C be bounded. Then every innite subset of
F contains an innite sequence of distinct points which converges to a point of C.
Proof: We may suppose that F is contained in the square centre 0, side length 2K (by taking the
smallest square containing the circle centre 0, radius K, which, by denition contains F). Cut this
square into 4 equal squares and select one, call it S
1
, which contains innitely many points of F.
Select z
1
F S
1
. Now divide S
1
into 4 equal squares. Select one which contains innitely many
points of F, call it S
2
and select z
2
F S
2
, with z
2
,= z
1
. Continuing in this way, we get a
sequence of distinct points (z
n
) and squares S
n
of side 2K/2
n
= K/2
n1
, such that z
n
F S
n
and S
1
S
2
S
n1
S
n
.
Claim (z
n
) is a Cauchy sequence. Indeed, given > 0, choose N N such that K

2/2
N
< . Then,
for m, n > N, we have z
m
, z
n
S
N+1
, so [z
n
z
m
[ diameter S
N+1
=

2K/2
N
< . So (z
n
) is a
Cauchy sequence, and hence converges by the GPC.
Remark This beautiful proof of a fundamental result illustrates the importance of drawing pictures.
To remember the proof, all you need is the picture.
This result can be applied to sequences as follows.
0.19 Denition Let S be a set and let (s
n
) be a sequence in S. Formally, (s
n
) is a mapping
s : N S. A subsequence (s
n
k
)
kN
of (s
n
)
nN
is what the notation suggests: a subset of the points
of the sequence, indexed by the appropriate integers n
k
and such that n
k+1
> n
k
. Formally, a
subsequence is the composition s r, where r : N N : k n
k
is some strictly increasing sequence.
0.20 Examples (2
2n
), (2
n
2
) are subsequences of (2
n
, as is (2
n
k
) where (n
k
) is any strictly
increasing sequence of integers. Notice that a sequence is a subsequence of itself. Note also that
(1, 1, 1, 1, 1 . . .) is a subsequence of (sin n/2) (take n
k
= 4k + 1).
iii
0.21 Theorem Let F C. F is compact if and only if every sequence in F has a subsequence
which converges to a point in F.
Proof: [] Suppose F is compact. Let (z
n
) be a sequence in F. If the set of points of the sequence
is nite, then one, call it w, must occur innitely often in the sequence and we can choose the
subsequence (w, w, w, w, . . . ) which clearly converges w F.
Otherwise, apply the Bolzano-Weierstrass theorem. This gives you an innite subset of distinct
points of z
n
: n N, and thus a subsequence, which converges to w C. Since F is closed, w F.
[] Conversely, suppose that every sequence in F has a subsequence converging to a point of F.
In particular, if you start with a sequence (z
n
) in F which converges to w C, then any subsequence
converges to w (proof left to reader) and w F. So F is closed.
Finally, suppose that F is not bounded. Then we can nd a sequence z
n
such that [z
n
[ > n. Let
(z
n
k
) be a convergent subsequence, z
n
k
z, say. But [z[ < N for some integer N, and if k > N,
then [z
n
k
[ > n
k
k N + 1 and giving [z
n
k
z[ > 1 for all k > n, which is a contradiction. So F
is bounded.
Further remarks for fans of analysis only
We know that a continuous function takes convergent sequences to convergent sequences: it
would be nice if this meant that it maps a closed set to a closed set. This is unfortunately not the
case: z arctan(Re z) maps the whole plane to the interval (/2, /2) on the real axis, which is
not closed. However, we have the following result.
0.22 Theorem Let F C be compact and let f : F C be continuous. Then f(F) is compact.
Proof: Let (w
n
) be a sequence in f(F). Then (z
n
) F such that f(z
n
) = w
n
. By compactness,
there is a subsequence (z
n
k
) which converges to F. Then w
n
k
f() f(F), by the continuity
of F. By 0.21, f(F) is compact.
This gives another proof of a result from rst year analysis.
0.23 Corollary Let a b R. Let f : [a, b] R be continuous. Then f is bounded and attains
its bounds.
Proof: f is bounded just means that f([a, b]) is bounded, which is true because f([a, b]) is compact.
Let k = sup f([a, b]). Then there is a sequence w
n
in f([a, b]) such that w
n
k. Hence k f([a, b]),
since f([a, b]) is closed, that is, there is x [a, b] with f(x) = k. Similarly there is y [a, b] with
f(y) = inf f([a, b]).
iv