Lectures on

Mathematical Analysis

Dr J. Zacharias
Dr J. F. Feinstein
University of Nottingham
Open Course:
https://www.youtube.com/watch?v=a0JNGx0Da8k&list=PL58984C080F2B0575
G12MAN Mathematical Analysis

Based on notes by Dr J. Zacharias

Modified by Dr J. F. Feinstein

1
About these notes
These notes include all of the definitions and theorems of
the module.
Most of them are examinable.
The notes omit a number of proofs and explanations of
examples given during the lectures.
There are gaps left where these should be filled in by you
during the lectures.
Technology permitting, Dr Feinstein’s annotated slides
from the lectures will be made available from the module
web page, along with screencasts of the lectures (video
and audio).

2
Examinable and non-examinable material
Certain material in this module will be clearly marked as
Not examinable as bookwork, or NEB for short.
This does not mean that this material is irrelevant to
exam questions.
Some sections of some of the exam questions are likely to
be ‘unseen’.
It is possible that some of the NEB material could be
relevant to some unseen portions of exam questions.
For more information on the style of exam questions, see
the G12MAN Module Information page and the
G12MAN Module Feedback page.

3
1 Introduction to Rd

In this chapter we look at subsets of

Rd = {(x1 , x2 , . . . , xd ) | x1 , x2 , . . . , xd ∈ R} and
establish some of the basic notation and terminology that
we will need in the rest of the module.
You have met R1 = R, R2 , R3 and to some extent Rd
already, for instance when solving larger systems of linear
equations.

4
1.1 Set notation

We use the standard set theoretical notations from

G11ACF. In particular we denote the set of
• positive integers by N = {1, 2, 3, 4, . . . };
• integers by Z = {. . . , −2, −1, 0, 1, 2, . . . };
• rational numbers (fractions) by
a
Q = { | a ∈ Z, b ∈ N};
b
• real numbers by R;
• non-negative real numbers by

R+ = {x ∈ R | x ≥ 0} = [0, ∞).

5
Subsets are often specified by a certain condition shared
by all its elements.
For instance certain subsets of R can be written as

{x ∈ R | x2 < x}

or
{x ∈ R | x2 + 1 = 0}.
The former is another way to write

{x ∈ R | 0 < x < 1}.

The latter is another way to write the empty set ∅.

Many different looking conditions or properties define the
same set.

6
A set A is a subset of a set B, written A ⊆ B, if each
element in A lies also in B, i.e. x ∈ A implies x ∈ B (we
can write: x ∈ A ⇒ x ∈ B).
Thus A = B if and only if A ⊆ B and B ⊆ A.
For sets C and D their intersection is

C ∩ D = D ∩ C = {x | x ∈ C and x ∈ D}

= {x | x ∈ D, x ∈ C}
and their union is

C ∪ D = D ∪ C = {x | x ∈ D or x ∈ C}.

7
Similarly if D1 , D2 , . . . , Dn are sets then
n
\
D1 ∩ D2 ∩ · · · ∩ Dn = Di
i=1

= {x | x ∈ D1 , x ∈ D2 , . . . , x ∈ Dn }
and
n
[
D1 ∪ D2 ∪ · · · ∪ Dn = Di
i=1

= {x | x ∈ D1 or x ∈ D2 or . . . or x ∈ Dn }
Intersections and unions can also be defined for infinitely
many Di ’s, see later.

8
The difference between C and D is

C\D = {x ∈ C | x ∈
/ D}.

where x ∈
/ D means that x is not an element of D.
Note that this is often different from

D\C = {x ∈ D | x ∈
/ C} .

If A is a subset of a larger set B then its complement (in

B) is Ac = B\A.
Usually it is clear from the context what B is.
Examples: Complement of a subset of R.
Gap to fill in

9
1.2 Cartesian products

The Cartesian product of two sets A and B is the set

of all ordered pairs of elements in A and B, i.e.

A × B = {(a, b) | a ∈ A, b ∈ B}.

(a1 , b1 ) = (a2 , b2 ) if and only if a1 = a2 and b1 = b2 .

Example 1.2.1 R × R is the set of ordered pairs of real

numbers.
Each (x, y) ∈ R × R can be visualised as a point in a
coordinate system.

Gap to fill in

10
The Cartesian product of two subsets A, B ⊆ R can be
visualised as well, e.g. if A = [1, 2] and B = [2, 4], then
A × B is represented by
Gap to fill in

11
We can produce a variety of subsets of R2 by taking
products of subsets of R, but there are many others such
as S = {(x, y) | 0 ≤ x ≤ 1, 0 ≤ y ≤ x} or
R = {(x, y) | x2 + y 2 = 1} which can not be written as
Cartesian products.
(See question sheets for details.)
Gap to fill in

12
R × R is also denoted by R2 .

Similarly A × B × C is the set of ordered triples

A × B × C = {(a, b, c) | a ∈ A, b ∈ B, c ∈ C}

etc.
We denote R × R × R by R3 and more generally

Rd = R × R × · · · × R
| {z }
d

= {(x1 , x2 , . . . , xd ) | x1 , x2 , . . . , xd real numbers} ,

the set of all d-tuples of real numbers.
Here d is a positive integer. R1 is identified with R.

13
1.3 Revision of Rd

Elements of Rd may be regarded as points which are

positioned in a certain way or as vectors which we can
add or multiply.
The former is Rd as an analytic or geometrical object,
the latter is Rd as a vector space: an algebraic object.
We will not worry about the distinction in this module.
Thus elements in Rd are denoted by (for example)

x, y, z, p, q,

etc. where x = (x1 , x2 , . . . , xd ), y = (y1 , y2 , . . . , yd ) etc.

14
Sometimes we will write them as d-dimensional column
vectors.
     
x1 y1 z1
     
 x2   y2   z2 
x= .  , y= .  , z =  .  etc.
     
 ..   ..   .. 
     
xd yd zd

Column vectors may be added

 
x1 + y1
 
 x2 + y2 
x+y =
 
.. 

 . 

xd + yd

Column vectors may be multiplied by scalars λ ∈ R

   
x1 λx1
   
 x2   λx2 
λx = λ  .  =  . 
   
 ..   .. 
   
xd λxd

15
  
x1
..
 
Since we identify (x1 , x2 , . . . , xd ) and   we may
 
 . 
xd
well also write

(x1 , x2 , . . . , xd ) + (y1 , y2 , . . . , yd )

= (x1 + y1 , x2 + y2 , . . . , xd + yd )
etc.

16
Recall the standard inner product on Rd :
   
* x1 y1 +
d
 .   . 
    X
hx, yi =  ..  ,  ..  = xi yi .
    i=1
xd yd

The norm corresponding to it is the Euclidean norm

defined by
 
! 1 x1
d 2
.
p X  
kxk = hx, xi = |xi |2 , for x =  .. 
 
i=1
 
xd

and may be regarded as the length of a vector.

Recall the Cauchy–Schwarz inequality

|hx, yi| ≤ kxk kyk

which holds for all x, y ∈ Rd .

17
Proposition 1.3.1 The function || · || : Rd → R+ has the
following properties, for all x and y in Rd and all λ ∈ R:
(i) kx + yk ≤ kxk + kyk (triangle inequality);
(ii) kλxk = |λ|kxk (homogeneity);
(iii) kxk = 0 ⇐⇒ x = 0.

The statement and applications of this result are

examinable as bookwork.
The proof of this result is NEB (not examinable as
bookwork), but is available on request.
Gap to fill in

18
The three properties above are all you usually need to
know about the Euclidean norm kxk in this module.

Using our Euclidean norm k · k, we define the Euclidean

distance between points x and y by

d(x, y) = kx − yk = ky − xk = d(y, x).

If z is another point in Rd we have

d(x, y) ≤ d(x, z) + d(z, y)

(triangle inequality for the Euclidean distance), since

Gap to fill in

19
Throughout this module, the Euclidean distance

d(x, y) = kx − yk
p
= (x1 − y1 )2 + · · · + (xd − yd )2
will play an important role.
In this module, unless otherwise specified, when we
refer to distance in Rd we will always mean the
Euclidean distance.

20
2 Boundedness of subsets of Rd

In this module we will meet, among others, the following

terms: bounded, boundary, closed.
These terms each have their own precise definitions in this
module.
It is essential that you are confident with these definitions
and can work with them.
In particular, you should be able to investigate whether or
not particular examples of subsets of Rd are closed or
bounded, and to determine the boundaries of such sets.
Warning! For those of you taking the module G12VEC,
there is an unavoidable conflict in terminology.
The standard terminology in these two rather different
areas of mathematics is long-established.
The term ‘bounded’ will have the same meaning in both
modules, but the G12VEC notions of a ‘closed surface’ and
a ‘surface with boundary’ involve different usages of the
words ‘closed’ and ‘boundary’.
You should ensure that you work with the appropriate
definitions in each module.

1
We now introduce some standard subsets of Rd .

2.1 Open and closed balls in Rd

Convention. In this module, when we say ‘Let r > 0’, we

mean that r is a positive real number (and the same goes
for other letters).
Let r > 0 and x ∈ Rd .
The (d-dimensional) open ball with radius r and centre x
is the set

Br (x) = {y ∈ Rd | ky − xk < r} .

The closed ball with radius r and centre x is the set

B̄r (x) = {y ∈ Rd | ky − xk ≤ r}.

The open unit ball is the set

B1 (0) = {x ∈ Rd | kxk < 1}

and the closed unit ball is the set

B̄1 (0) = {x ∈ Rd | kxk ≤ 1}.

2
For d = 2, Br (x) is the disc with radius r about x without
its ‘boundary’.
B̄r (x) is the same disc including its ‘boundary’.
Gap to fill in

3
We will give a formal definition for the term ‘boundary’
later when we we investigate the topology of Rd .
Exercise. What are the open/closed balls when d = 1?
Gap to fill in

4
2.2 Bounded sets and unbounded sets

In this section, we discuss bounded sets and unbounded

sets.
As with all concepts in this module, it is very important
that you understand these definitions and examples.
Informal/intuitive notions can be useful, but you will be
expected to be able to work confidently and reliably with
definitions and examples.
The word ‘bounded’ can be particularly confusing (see the
module FAQ) until you have worked with and understood
some examples.
Our first examples of bounded sets will be the open balls
and closed balls introduced above.

5
Definition 2.2.1 A subset S ⊆ Rd is said to be bounded
if there exists a real number R > 0 such that S ⊆ B̄R (0),
i.e., for all x ∈ S, we have ||x|| ≤ R.
A set which is not bounded is said to be unbounded.

Note that R must not, of course, depend on x here!

(Otherwise every set would be bounded.)
Gap to fill in

6
Example 2.2.2 (1) Open or closed balls centred on the
origin 0 are bounded.
Gap to fill in

7
(2) All open/closed balls in Rd are bounded, whether or
not they are centred on the origin. Let x ∈ Rd and let
r > 0. Then B̄r (x) and Br (x) are both bounded.
Gap to fill in

8
(3) The empty set ∅ is bounded, but Rd is unbounded.
Gap to fill in

9
(4) Every subset of a bounded set is also bounded.
Gap to fill in

10
(5) In the special case of R = R1 (i.e. when d = 1), we
note that a set S ⊆ R is bounded if and only if it is
both bounded above and bounded below in the sense
discussed in G11ACF.
Note that these concepts from G11ACF do not
make sense directly in higher dimensions.

Note that there is no connection at all between the

terms ‘bounded’ and ‘boundary’.
In particular, a bounded set need not include its
boundary.

11
Some further important examples of bounded and
unbounded sets are given below, when we discuss d-cells in
Rd .
There are many equivalent definitions of the term
‘bounded’.
For example, a set S is bounded if and only if the set of all
possible distances between pairs of points of S is bounded
above in R.
Thus non-empty bounded sets can also be described as
sets which have finite diameter.
See question sheets for details.
Gap to fill in

12
2.3 Intervals and d-cells

We begin by discussing bounded intervals in R.

Let a and b be real numbers, with a ≤ b.
Then we have four types of bounded interval associated
with a and b.
First we have the closed interval,

[a, b] = {x ∈ R | a ≤ x ≤ b} ⊆ R

and the open interval

]a, b[ = (a, b) = {x ∈ R | a < x < b} .

We may use either of these two notations for open

intervals.
The new notation is useful in order to avoid confusion
of the ordered pair (a, b) with the open interval ]a, b[.
We also have the half-open intervals

]a, b] = (a, b] = {x ∈ R | a < x ≤ b}

and
[a, b[ = [a, b) = {x ∈ R | a ≤ x < b} .

13
For a = b, [a, b] is just the single-point set {a}, and the
other intervals are empty in this case.
Warning! You should not attempt to apply the term
‘half-open’ to a set which is not an interval.
Most subsets of R are not intervals!
Gap to fill in

14
Analogues of intervals in Rd can be obtained by taking
cartesian products of d intervals to form d-cells.
This gives us rectangles in R2 , cuboids in R3 and
hyper-cuboids in higher dimensions.
Note. In 2006-7 (and before), d-cells were described as
intervals in Rd .
Let a1 ≤ b1 , a2 ≤ b2 , . . ., ad ≤ bd be real numbers.
Corresponding to these we have closed intervals [ai , bi ] and
open intervals ]ai , bi [ (1 ≤ i ≤ d).
The corresponding closed d-cell in Rd is defined to be

[a1 , b1 ] × [a2 , b2 ] × · · · × [ad , bd ]

= {x ∈ Rd | a1 ≤ x1 ≤ b1 , . . . , ad ≤ xd ≤ bd }
and the corresponding open d-cell is

]a1 , b1 [ × ]a2 , b2 [ × · · · × ]ad , bd [

= {x ∈ Rd | a1 < x1 < b1 , . . . , ad < xd < bd }.

15
If d > 1 then there are many more ‘half-open’
combinations possible than just 2.
For example, when d = 2 and a1 = a2 = 1, b1 = b2 = 2 we
have the closed 2-cell [1, 2] × [1, 2] and the open 2-cell
]1, 2[ × ]1, 2[
Gap to fill in

16
and 14 other ‘half-open’ 2-cells such as
Gap to fill in

17
In the same way, there are 64 types of bounded 3-cell in
R3 and 22d types of bounded d-cell in Rd .

We have not proved that the above d-cells are bounded.

You may check this claim directly from the definitions.
Alternatively, you can deduce it from results on the
question sheets.

18
We may also form unbounded d-cells by replacing some of
the ak by −∞ and/or some of the bk by ∞ (which should
be thought of as meaning +∞ here).
Note: ±∞ are not real numbers, so that these must be
excluded from the intervals.
Gap to fill in

19
3 Open subsets of Rd

In this chapter, we will single out some ‘nice’ subsets for

doing analysis.
The only notion we need is that of a distance between two
points in Rd .
Much of what we are doing works similarly for all
dimensions d.
Actually the concepts we are going to describe can be
studied in far greater generality: we only need a distance
concept called a metric.
This is done in a branch of mathematics called (analytic)
topology.
You can find out more about topology in the level 3
module G13MTS: Metric and Topological Spaces.

1
3.1 Interior points and non-interior points

We begin by discussing a fairly intuitive notion: which

points of a set lie in its ‘interior’, and which points do not.
This is based on the idea that a point at the centre of a
ball should clearly be regarded as being in the interior of
that ball: there is room to move a bit in all directions
without leaving the ball.
Gap to fill in

2
Definition 3.1.1 Let S be a subset of Rd .
We classify the points of S as either interior points of S
or non-interior points in S as follows.
Let x ∈ S. Then x is an interior point of S if there is an
r > 0 such that Br (x) ⊆ S.
Otherwise x is a non-interior point in S.
The set of all interior points of S is called the interior of
S, and is denoted by int S or S̊.
The set of non-interior points in S is denoted by nint S, so
that
nint S = S \ int S .

Gap to fill in

3
Remarks.
The terms interior point of S and interior are entirely
standard.
The term non-interior point in S and the notation nint S
are not standard, but they are very helpful tools in
understanding this material.
If you use this non-standard terminology and notation in
another module, you must explain what you are doing:
other lecturers are unlikely to have come across nint S.
Note that a non-interior point in S must be an element
of S.
Although points of S c = Rd \ S are certainly not interior
points of S, such points do not count as non-interior
points in S and so they are not in nint S
Gap to fill in

4
In 2006-7 and 2007-8, the term ‘non-interior point of S’
was used (instead of ‘. . . in S’), but this did not make it
quite as obvious that the points concerned must actually
be in S.
Exercise. By negating the definition of interior in the
usual way, write down an explicit definition of what it
means for a point of S to be a non-interior point in S.
Gap to fill in

5
Recall that, in R = R1 , open balls are just open intervals:
for a ∈ R and r > 0, Br (a) = ]a − r, a + r[ .

Example 3.1.2 Determine, with full justification, which

points (if any) are the interior points and which points (if
any) are the non-interior points in the following subsets of
R:
S1 = [2, 5] ; S2 = ]3, 7[ ; S3 = [1, 4[ .

Gap to fill in

6
Exercise 3.1.3 Recall that the set of interior points of a
set S is called the interior of S.
Write down, without justification, what you believe the
interiors of each of the following subsets of R are, and
which points (if any) are the non-interior points in these
sets:
∅; [0, ∞[ ; Q; Qc = R \ Q ; R.

Gap to fill in

7
3.2 Open sets

We now define a very important class of subsets of Rd : the

open sets.
The study of open sets is fundamental to understanding
the topological nature of Rd .
In complex analysis, open sets are of key importance when
investigating analytic functions: see G12COF for details.

Definition 3.2.1 Let U be a subset of Rd .

We say that U is an open set (or U is open in Rd , or U
is an open subset of Rd ) if every point of U is an interior
point of U , i.e., the interior of U is equal to U .

Remarks. Using our notation for interior, this means that

U is open if and only if U = int U (or U = Ů ).
Note that the set U is not open if and only if U has at
least one non-interior point.

8
You can write out the definition of open set in full as
follows.
A subset U ⊆ Rd is open if, for each x ∈ U , there is an
r > 0 such that Br (x) ⊆ U .
Note, here, that the r you need usually depends on x.
Examples 1) An open interval in R is an open set. But a
closed interval or a half-open interval is not open (unless
empty).
Gap to fill in

9
2) An open ball in Rd is open!
This sounds so obvious it is almost silly, but it really does
need a proof: see the module FAQ document for a
discussion of this issue.
In R2 we may illustrate the result using the diagram below.
The same diagram helps us to give a careful proof of the
result in Rd (the diagram alone is insufficient).
Gap to fill in

10
By contrast a closed ball is not open.
Gap to fill in

3) Both the empty set ∅ and Rd are open in Rd .

Gap to fill in

11
4) The set ]0, 1[ ∪ ]2, 3[ is open in R.
Gap to fill in

In fact we will see later that whenever you take a union of

open sets, you get another open set.
5) A non-empty, finite subset of Rd is not open.
Gap to fill in

12
6) The set Q is not open in R, and neither is Qc .
Gap to fill in

7) Of the non-empty bounded d-cells discussed in the last

chapter, the ones we called the open d-cells really are are
open, and the others are not open.
You should know this fact and be able to apply it, but
the proof of this fact is NEB.
Gap to fill in

13
4 The topology of Rd

In this chapter we look at some properties of open sets,

and introduce the closed sets.

4.1 Properties of open sets

We begin with an elementary lemma.

Lemma 4.1.1 Let A and B be subsets of Rd . Suppose

that A ⊆ B. Then every interior point of A is also an
interior point of B, i.e., int A ⊆ int B.

Gap to fill in

1
Theorem 4.1.2 Let U, V ⊆ Rd be open. Then U ∪ V and
U ∩ V are also open.

Gap to fill in

2
It follows (by an easy induction) that finite unions and
intersections of open sets are open.
In fact, for unions, more is true.
Infinite unions of open sets are still open sets.
Let us first say what we mean by infinite unions.
Recall that if A1 , A2 , . . . , An are sets then
A1 ∪ A2 ∪ · · · ∪ An can be written as
n
[
A1 ∪ A2 ∪ · · · ∪ An = Ai
i=1

= {x | x ∈ Ai for at least one i ∈ {1, . . . , n}}.

Suppose we have a sequence A1 , A2 , A3 , . . . (i.e. an
ordered list) of sets.
We simply take the above reformulation of a finite union
and define
[ ∞
[
An = An = {x | x ∈ An for at least one n ∈ N }.
n∈N n=1

Such unions are called countable unions.

3
Theorem 4.1.3 Let (Un ) be a sequence of open sets in R.
S∞
Then n=1 Un is open.
In other words: countable unions of open sets are again
open.

Gap to fill in

4
One can also define countable intersections by
\ ∞
\
An = An = {x | x lies in all of the sets An }
n∈N n=1

= {x | x ∈ An for all n ∈ N}.

In general it is not true that countable intersections of
open sets are again open, even though finite intersections
are!
Gap to fill in

5
4.2 Closed sets

Complements of open sets are usually not open (though

there are some exceptions).
In fact these complements form an important class of
subsets as well.

Definition 4.2.1 A subset E of Rd is said to be closed

(or closed in Rd , or a closed subset of Rd ) if its
complement E c = Rd \E is open.

Example 4.2.2 1. Closed intervals in R and (more

generally) closed d-cells in Rd are closed subsets, and so
are closed balls in Rd .
Gap to fill in

6
2. Finite subsets of Rd are closed.
Gap to fill in

3. Both ∅ and Rd are closed in Rd .

Gap to fill in

7
Warnings!
1. Most sets are neither open nor closed.
Gap to fill in

2. Some sets are both open and closed.

Gap to fill in

3. It is very important that you remember the

distinction between closed and not open.

8
Since closed sets are complements of open sets many of
their properties follow from properties of open sets. If
A, B ⊆ Rd then

(A ∪ B)c = Ac ∩ B c .

(de Morgan’s law, as you checked on the first question

sheet).
Similarly
(A ∩ B)c = Ac ∪ B c .

We even have
∞
!c ∞
[ \
An = Acn
n=1 n=1

for any sequence (An ) of subsets of Rd , and also

∞
!c ∞
\ [
An = Acn .
n=1 n=1

We leave the verification as a coursework exercise.

9
The statement and applications of the next result are
examinable, but the proof is NEB (not examinable as
bookwork).

Theorem 4.2.3 (i) Finite or countable intersections of

closed subsets of Rd are closed.
(ii) Unions of finitely many closed subsets of Rd are
closed.

10
5 Sequences in Rd
In this chapter we revisit the notion of a convergent
sequence, extending this to the setting of Rd .
As with the topology of Rd , the notion of convergence
comes from the notion of distance that we are working
with.

5.1 The definition of convergence

You have already met sequences in R.

We now wish to investigate sequences in Rd .
We use the following notation.
Notation: Let A be a subset of Rd .
We write (xn ) ⊆ A to mean that (xn ) is a sequence
x1 , x2 , x3 , . . . all of whose terms xn lie in A.
In particular, (xn ) ⊆ Rd just means that (xn ) is a
sequence of points in Rd .
Sequences may or may not include repeated terms.
For example, xn = (1, (−1)n ) gives a sequence (xn ) ⊆ R2
with multiply repeated terms.

1
We already know what it means for a sequence to
converge in R.
Here is the generalization to sequences in Rd .

Definition 5.1.1 Let (xn ) be a sequence of points in Rd .

We say that (xn ) converges to a ∈ Rd if the sequence of
real numbers kxn − ak converges to 0 in the usual sense
in R.
Equivalently (exercise), for every ε > 0 there exists
n0 ∈ N s.t. for all natural numbers n ≥ n0 we have
kxn − ak < ε.

By convention, n is usually an integer, and so it is

common to omit the words ‘natural numbers’ in this
definition and just write ‘for all n ≥ n0 ’.
Notation: Given such a convergent sequence (xn )
converging to a ∈ Rd , we may also write
limn→∞ xn = lim xn = a or xn → a as n → ∞.
Note that this notation is NOT a definition of
convergence.

2
5.2 Absorption of sequences by sets

We introduce the following non-standard terminology

which can help to understand convergence.

Definition 5.2.1 Let (xn ) ⊆ Rd and let A ⊆ Rd . We say

that the set A absorbs the sequence (xn ) if there exists
N ∈ N such that the following condition holds:
for all n ≥ N , we have xn ∈ A, (∗)
i.e., all terms of the sequence from xN onwards lie in the
set A.

Gap to fill in

3
Some authors use terms such as xn eventually lies in A
or say at most finitely many terms of the sequence lie
in Ac to mean the same thing.
If N satisfies condition (∗) above, then we also say that A
absorbs the sequence (xn ) by stage N .
In this case, it is clear that, if N 0 ≥ N , then the same set
A absorbs the sequence (xn ) by stage N 0 too.
Also, if A ⊆ B, then it is clear that B also absorbs the
sequence (xn ) by stage N .
Gap to fill in

4
Example 5.2.2 Consider the sequence of real numbers
(−1)n
xn = √ (n ∈ N).
n

Clearly this sequence converges to 0 in R.

How far do you have to go along this sequence before it
gets stuck in a given small interval?
We can see, for example, that for n ≥ 106 , we have
|xn | ≤ 1/1000.
1 1
So, for n ≥ 106 , xn ∈ [− 1000 , 1000 ].
In our new terminology, we can say that the set
[−1/1000, 1/1000] absorbs the given sequence (xn ) by
stage 1000000.

Gap to fill in

5
The following result is immediate from the definitions
above.
See question sheets for further discussion of this and other
results concerning convergence and absorption of
sequences.

Proposition 5.2.3 Let a ∈ Rd and let (xn ) ⊆ Rd . Then

the following statements are equivalent:
(a) the sequence (xn ) converges to a;
(b) for all ε > 0, the open ball Bε (a) absorbs the
sequence (xn ).

Thus (xn ) converges to a if and only if every open ball

centred on a absorbs the sequence (xn ).

6
5.3 Standard convergence results

Given a sequence (xn ) ⊆ Rd each xn has d coordinates.

Say xn = (xn1 , xn2 , . . . , xnd ).
This means that we have d sequences of real numbers
corresponding to the sequence (xn ) (one for each of the d
coordinates).
Gap to fill in

7
The next result can often save work when looking at
examples.
This statement and its applications are examinable, but
the proof is NEB (not examinable as bookwork).
This result may be described as saying that convergence in
Rd is the same as coordinatewise convergence.

Proposition 5.3.1 Suppose that (xn ) is a sequence in

Rd , with xn = (xn1 , xn2 , . . . , xnd ) ( for n ∈ N), and let
a = (a1 , . . . , ad ) ∈ Rd . Then the sequence (xn ) converges
to a if and only if the following holds:
for all i ∈ {1, 2, . . . , d} we have xni → ai as n → ∞
(or, in other words, xn1 → a1 , xn2 → a2 , .... and
xnd → ad as n → ∞).

Gap to fill in

8
  n

n 3
Example: Consider xn = 4n+2 , n! .
Determine, with justification, whether this sequence
converges and, if so, what the limit is.
Gap to fill in

9
Next recall the sandwich theorem (or squeeze rule) from
G11ACF.
As we will see later, this can be very useful in applications.

Proposition 5.3.2 (Sandwich Theorem) Let a ∈ R.

If (an ), (bn ), (cn ) are sequences of real numbers s.t.
an ≤ bn ≤ cn for each n, and such that an → a and
cn → a as n → ∞, then bn → a as n → ∞.

Exercise. Can you think of a way to generalize the

sandwich theorem to work in Rd ?
Gap to fill in

10
There is the following ‘algebra of limits’ result for
sequences in Rd :

Theorem 5.3.3 (Algebra of limits in Rd ) Let (xn ), (y n )

be sequences in Rd and (λn ) a sequence in R s.t. xn → x,
y n → y and λn → λ as n → ∞. Then
(i) xn + y n → x + y
(ii) ||xn || → ||x||
(iii) λn xn → λx
as n → ∞.

Proof. We prove (i) and (ii). See question sheets for (iii).
Gap to fill in

11
We conclude this chapter with a very important
characterization of closedness of subsets in terms of
convergent sequences.

Theorem 5.3.4 (Sequence criterion for closedness)

For a subset C ⊆ Rd the following conditions are
equivalent:
(i) C is closed.
(ii) For every sequence (xn ) ⊆ C which converges in Rd ,
the limit limn→∞ xn = x must also lie in C.

Proof.
Gap to fill in

12
6 Subsequences and sequential
compactness

6.1 Nested intervals and nested d-cells

Recall the nested intervals principle from G11ACF:

Let Ik = [ak , bk ] ⊆ R be closed (non-empty) intervals in R
with I1 ⊇ I2 ⊇ I3 ⊇ . . . (such a sequence of sets is said to
be nested decreasing).
Then there exists at least one x ∈ R such that x is in all
T∞
of the intervals Ik , i.e., such that x ∈ k=1 Ik .
(This fails for non-closed intervals in general. See question
sheets.)
Moreover, if bk − ak → 0 as k → ∞, then this point x is
unique.
Gap to fill in

1
Before discussing the d-dimensional analogue of this result,
let us first look at the condition that bk − ak → 0 as
k → ∞.
Note that bk − ak = max{|x − y| : x, y ∈ [ak , bk ] } .
In general, for a non-empty, bounded subset A of Rd , we
define the diameter of A, diam (A), by

diam (A) = sup{kx − yk : x, y ∈ A }

the supremum of the possible distances between pairs of

points of A.
From earlier results we know that 0 ≤ diam (A) < ∞ for
non-empty bounded sets.
For (non-empty, bounded) closed d-cells, this
supremum is actually a maximum, and this maximum is
achieved by considering any pair of diametrically opposite
corners of the d-cell.
Gap to fill in

2
We now come to the d-dimensional analogue of the nested
intervals principle.
From now on, whenever we discuss CLOSED intervals
or d-cells, we will assume that they are non-empty
and bounded.

Theorem 6.1.1 (Nested d-cells principle) Let d ∈ N,

and let Ik be a nested decreasing sequence of closed
d-cells (so I1 ⊇ I2 ⊇ I3 ⊇ . . . ).
Then there exists at least one point x ∈ Rd such that x is
T∞
in all of the d-cells Ik , i.e., such that x ∈ k=1 Ik .
Moreover, if diam (Ik ) → 0 as k → ∞ then this point x is
unique.

3
Typical pictures for d = 2:
Gap to fill in

The proof of this d-dimensional principle is an NEB

exercise.
It is, in fact, fairly easy to deduce the result from the
nested intervals principle in R.

4
6.2 The Bolzano–Weierstrass Theorem

Let x1 , x2 , x3 , x4 , x5 , . . . be a sequence in Rd .
We can form a new sequence by omitting some of the
sequence members, e.g.
(i) x1 , x3 , x5 , x7 , . . .
(ii) x2 , x4 , x6 , x8 , . . .
(iii) x3 , x4 , x5 , x1304 , . . .
(All three examples are meant to continue indefinitely.)

5
In (i) we have formed the new sequence (y n ), where
y n = x2n−1 .
In (ii) we have formed (z n ), where z n = x2n .
In (iii) it is less clear what the pattern is but in any case
we have chosen a strictly increasing sequence of positive
integers

k1 = 3 < k2 = 4 < k3 = 5 < k4 = 1304 < . . .

and formed a new sequence (pn ) given by pn = xkn .

Definition 6.2.1 A subsequence of a sequence (xn ) in Rd

is a sequence of the form (y n ) = (xkn ), where
k1 < k2 < k3 < . . . is an increasing sequence of positive
integers.

6
This means: when a subsequence is formed the order of
the sequence members is kept and repetitions of the index
are excluded.
There may be repeated values, but only if there were
repeated values in the original sequence.
Why is this notion important?
It happens often that divergent (i.e. non-convergent)
sequences have convergent subsequences.
For example, xn = (−1)n defines a divergent sequence in
R but (x2n ) is convergent.
What about the sequence xn = sin(n) in R?
The Bolzano–Weierstrass Theorem (below) will show, in
particular, that this sequence also has at least one
convergent subsequence.

7
In order to help us find subsequences, we note the
following elementary lemma.

Lemma 6.2.2 Let A and B be subsets of Rd , and

suppose that (xn ) ⊆ A ∪ B, i.e., (xn ) is a sequence of
elements of A ∪ B.
Then at least one of the following two statements is true.
• Infinitely many terms of the sequence are in A, i.e.,
{n ∈ N | xn ∈ A} is an infinite subset of N.
• Infinitely many terms of the sequence are in B, i.e.,
{n ∈ N | xn ∈ B} is an infinite subset of N.
(Of course it is possible for both statements to be true.)

Gap to fill in

8
We say that a sequence (xn ) in Rd is bounded if the set
{xn | n ∈ N} ⊆ Rd of all sequence members is bounded,
i.e. there exists M ≥ 0 such that ||xn || ≤ M for all n ∈ N.
Now we can formulate the famous

Theorem 6.2.3 (Bolzano–Weierstrass Theorem) Every

bounded sequence (xn ) in Rd has at least one convergent
subsequence.

[ So, in particular, (sin(n)) ⊆ R has a convergent

subsequence!]
The statement of the Bolzano–Weierstrass Theorem
and its applications are examinable.
The proof, which we sketch, is NEB.
Gap to fill in

9
6.3 Sequential compactness

The notion of sequential compactness and other related

notions of compactness are extremely important in analysis
and mathematics in general and have far-reaching
applications.

Definition 6.3.1 A subset E ⊆ Rd is said to be

sequentially compact if every sequence (xn ) ⊆ E has at
least one subsequence which converges to a point of E.

We can characterize sequentially compact subsets of Rd as

follows.

10
Theorem 6.3.2 (Heine-Borel Theorem, sequential
compactness version) Let E ⊆ Rd be any subset. Then
the following conditions are equivalent:
(i) E is sequentially compact;
(ii) E is closed and bounded.

Gap to fill in

11
Example 6.3.3 1. Closed intervals and closed balls in
Rd are sequentially compact.
2. The set {0} ∪ { n1 | n ∈ N} is sequentially compact.

Gap to fill in

12
Sequentially compact sets are ‘nice’ sets in many respects,
as we will see.
It also turns out that, for subsets of Rd , sequential
compactness is equivalent to a topological condition
involving coverings by open sets.
This related condition, which is called simply
compactness, is of great importance throughout
mathematics.
See books and/or the module G13MTS if interested.

13
7 Functions, Limits and Continuity

7.1 Revision and Examples

A function f : A → B from a set A (the domain of f ) to a

set B (the co-domain of f ) is a rule assigning to each
a ∈ A a unique element f (a) ∈ B.
We write a 7→ f (a) to stress this.
If S ⊆ A is a subset then we define the restriction f |S of
f to S to be the function f |S : S → B given by
(f |S )(s) = f (s) for all s ∈ S.
It is still the same rule but only applied to elements in S so
we only change the domain of the function (which is part
of its definition).
Given a function f : A → B, the image of S under f is
the set

f (S) = {f (s) | s ∈ S}
= {b ∈ B | there exists s ∈ S such that f (s) = b} ⊆ B .

The image of f is simply the image of the whole of the

domain of f under f , i.e., f (A) (since the domain of our
function is A).

1
In G11ACF you encountered mainly functions
f : [a, b] → R or g : R → R i.e. functions of a single (real)
variable defined on a closed subinterval or the whole of R
(i.e. with domain equal to [a, b] or R) taking values in R
(i.e. co-domain equal to R).
Gap to fill in

In G11CAL you saw functions f : R2 → R or g : R3 → R

or ones defined on subsets of R2 or R3 respectively
(functions of two or three variables).
Gap to fill in

2
We will consider functions functions of the type

f : Rd → Rl or f : D → Rl ,

where d, l are positive integers and D ⊆ Rd .

In the sequel d and l denote fixed positive integers
unless otherwise specified.
Given f : R → Rl , f (x) has l coordinates, and so f is
given by an l-tuple of real-valued functions in the following
way: we may write

f (x) = (f1 (x), f2 (x), . . . , fl (x))

for l real-valued functions f1 , f2 , . . . , fl (one for each

coordinate in Rl ).
Gap to fill in

3
In the same way, any function f : Rd → Rl or g : D → Rl
is given by an l-tuple of real valued functions
f1 , f2 , . . . , fl : Rd → R respectively g1 , g2 , . . . gl : D → R,
where
f (x) = (f1 (x), f2 (x), . . . , fl (x))
for x ∈ Rd , and

g(y) = (g1 (y), g2 (y), . . . , gl (y))

for y ∈ D.
You saw a large number of examples in G11ACF and
G11CAL, often made up from polynomials, rational
functions, exponential functions, logarithmic functions and
trigonometric functions.
Gap to fill in

4
You should bear these examples in mind, and remind
yourselves of their properties. We shall assume without
further proof that these standard functions have the
properties claimed in G11CAL.
Gap to fill in

See books for formal proofs of facts concerning the

continuity and differentiability of these standard functions.

5
7.2 Limits and Continuity

We recall the notions of limit and continuity for functions

of one variable from G11ACF and G11CAL.
Let a ∈ R and let f be a real-valued function defined (at
least) at all points of R \ {a} (but not necessarily at a).
Then the limit limx→a f (x) exists and equals L ∈ R if,
for every sequence (xn ) ⊆ R \ {a} which converges to a,
we have f (xn ) → L as n → ∞.
The notation
lim f (x) = L
x→a

means that the limit exists and is equal to L.

Note that f (a) was not necessarily defined above, and so
was not involved when investigating the above limit.
Now suppose that we have f : R → R.
Then f is said to be continuous at a ∈ R if

lim f (x) = f (a) .

x→a

6
Since we know what convergence of sequences in Rd and
Rl means, we can generalize the definition of limits to
functions f : Rd → Rl .

Definition 7.2.1 Let a ∈ Rd ,let q ∈ Rl , and let f be a

function taking values in Rl which is defined (at least) at
all points of Rd \ {a}.
We say that the limit limx→a f (x) of f at a exists and
equals q if, for every sequence (xn ) ⊆ Rd \ {a} which
converges to a, we have f (xn ) → q (in Rl ) as n → ∞.
In this case, we may also write f (x) → q as x → a.
If there is no such q in Rl , then the limit above does not
exist.
As before, if we write either limx→a f (x) = q or f (x) → q
as x → a, we mean that the limit exists and is equal to q.

7
Now suppose that we have a function f : Rd → Rl , and let
a ∈ Rd .
Then f is said to be continuous at a if

lim f (x) = f (a) .

x→a

Otherwise we say that f is discontinuous at a: this

means that either the limit above does not exist or it does
exist, but is not equal to f (a).

Note that f : Rd → Rl is continuous at a if and

only if the following condition holds: for every se-
quence (xn ) ⊆ Rd which converges to a, we have
f (xn ) → f (a) as n → ∞.

This time we do not insist on xn 6= a.

[Exercise: check the details.]

8
We now wish to consider functions defined on a non-empty
subset D of Rd .
However, when discussing limits, we will meet an obstacle
when it comes to ‘isolated’ points.
A point a of D is isolated if there is an r > 0 such that a
is the only point in D ∩ Br (a): informally, there are no
other points of D near to a.
Gap to fill in

9
Let D be a non-empty subset of Rd , and suppose that we
have a function f : D → Rl or f : D \ {a} → Rl .
We define limits and continuity in a similar way to above.
For such f , and a ∈ D, we say that limx→a f (x) exists
and equals q ∈ Rl if, for every sequence (xn ) ⊆ D \ {a}
which converges to a, we have f (xn ) → q as n → ∞.
If there is no such q in Rl , then the limit above does not
exist.
Again, if we write either limx→a f (x) = q or f (x) → q as
x → a, we mean that the limit exists and is equal to q.
The function f : D → Rl is said to be continuous at
a ∈ D if
lim f (x) = f (a) .
x→a

Otherwise, we say that f is discontinuous at a.

There is one problem with this definition of limit.
If a is an isolated point of D then there are no such
sequences (xn )!
[Exercise. What do the above definitions mean if a is an
isolated point?]

10
Again, we may reformulate the definition of continuity in
terms of sequences in the following way. This will be one
of the most useful forms of the definition of
continuity for us.

The function f is continuous at a if and only if the

following condition holds: for every sequence (xn ) ⊆
D which converges to a, we have f (xn ) → f (a) as
n → ∞.

Here some or all of the xn may be equal to a.

Note that this time we do not need to worry about isolated
points, as we always have at least the constant sequence
a, a, a, . . . available.
[Exercise. What does this condition mean if a is an
isolated point?]

11
Definition 7.2.2 A function f : D → Rl is said to be
continuous if it is continuous at every a ∈ D.
Otherwise the function f is said to be discontinuous.

Thus a function is discontinuous if there is at least one

point of its domain at which it is discontinuous.
We shall use without further proof the fact that the
standard functions met in G11CAL are continuous on
their domains of definition.
In particular, ‘coordinate projections’ are continuous.
Co-ordinate projections are maps such as the map from R2
to R given by (x, y) 7→ x or (x, y) 7→ y.
More generally, for D ⊆ Rd and 1 ≤ i ≤ d, the ith
coordinate projection pi : D 7→ R is defined by
pi ((x1 , x2 , . . . , xd )) = xi .
Gap to fill in

12
Warning! It does not make sense to ask whether or not a
function is continuous or discontinuous at a point where it
is undefined! These questions only make sense at
points of the domain of f .
Gap to fill in

13
7.3 Meaning of continuity for functions of
several variables

We now discuss some of the many different ways there are

to ‘approach’ a point in R2 .
This can be helpful when we want to establish that a
function is discontinuous.
Establishing that a function is continuous can be trickier!
We will return to this issue in the next chapter.
Example Let f : R2 → R be defined by

 2xy if x2
+ y 2
> 0,
2
x +y 2
f (x, y) =
 0 otherwise.

Does the limit lim(x,y)→(0,0) f (x, y) exist?

Is f continuous at (0, 0)?
Gap to fill in

14
Notice that f defined above is continuous if we fix x and
regard only y as a variable or fix y and regard only x as
variable.
This continuity in each variable of a function of two (or
more variables) is called separate continuity whereas the
type of continuity defined in 7.2.1 and 7.2.2 is sometimes
called joint continuity.
By the above example a function of several variables may
well be separately continuous without being (jointly)
continuous.
Example. Determine whether or not the following
function g : R2 → R is continuous at the point (0, 0):

 xy2 if x2 + y 2 > 0,
x2 +y 4
g(x, y) =
 0 otherwise.

Gap to fill in

15
When investigating the continuity of specific examples, the
following is very useful.
The proof of this result is NEB, but the statement and
applications are examinable.
Let D ⊆ Rd and f : D → Rl be a function.
As usual, write f (x) = (f1 (x), f2 (x), . . . , fl (x)) for
x ∈ D, where f1 , f2 , . . . , fl are functions from D to R.

Lemma 7.3.1 The function f : D → Rl is continuous at

a ∈ D if and only if all the real valued functions f1 , f2 , ....
fl are continuous at a.

Examples.
Gap to fill in

16
8 Further theory of function limits
and continuity

8.1 Algebra of limits and sandwich

theorem for real-valued function limits

The following results give versions of the algebra of limits

and sandwich theorem for real-valued functions defined
on subsets of Rd .
Don’t forget that there is a big difference between limits
of sequences and function limits, even though there are
connections between them.
You should ensure that you understand the different
nature of these different types of limit.
Suppose that D ⊆ Rd , λ ∈ R, and f : D → R,
g : D → R are functions.
We may form new functions in the usual way (using what
are often described as pointwise operations, e.g.
pointwise addition, etc.).

1
 (f + g)(x) = f (x) + g(x) , (f g)(x) = f (x)g(x) ,

(λf )(x) = λf (x) and |f |(x) = |f (x)| ,

where x ∈ D.
These new functions also have domain D.
The pointwise product f g should not be confused with
the composite of the two functions f and g.
Finally if g(x) 6= 0 then the quotient
f (x)
(f /g)(x) =
g(x)
makes sense and defines a function f /g, at least at those
points of D where g is non-zero.

2
Theorem 8.1.1 (Algebra of real-valued function
limits) Let D be a non-empty subset of Rd and suppose
that a is a non-isolated point of D. Let f and g be
functions from D to R (or from D \ {a} to R), and let
λ ∈ R.
Suppose that limx→a f (x) exists and is L1 ∈ R and that
limx→a g(x) exists and is L2 ∈ R.
Then

lim (f (x)+g(x)) = L1 +L2 , lim (f (x)g(x)) = L1 L2 ,

x→a x→a

lim |f (x)| = |L1 | and lim (λf (x)) = λL1 .

x→a x→a

Moreover, provided that g does not take the value 0

anywhere on D \ {a}, and that L2 6= 0, we have

lim (f (x)/g(x)) = L1 /L2 .

x→a

Gap to fill in

3
Theorem 8.1.2 (Sandwich Theorem for real-valued
function limits) Let D be a non-empty subset of Rd and
suppose that a is a non-isolated point of D. Let f , g
and h be functions from D to R (or from D \ {a} to R),
and let L ∈ R. Suppose that, for all x ∈ D \ {a}, we
have f (x) ≤ g(x) ≤ h(x), and that both
limx→a f (x) = L and limx→a h(x) = L. Then
limx→a g(x) also exists, and is equal to L.

The proof of this is an exercise.

Corollary 8.1.3 Let D be a non-empty subset of Rd and

suppose that a is a non-isolated point of D. Let f be a
function from D to R (or from D \ {a} to R). Then

lim |f (x)| = 0 ⇔ lim f (x) = 0 .

x→a x→a

Gap to fill in

4
Example. Prove that
3 4
 
x y
lim = 0.
(x,y)→(0,0) x6 + y 6
Gap to fill in

5
8.2 New continuous functions from old!

The following results allow us to show that many types

of functions are continuous.
For instance, we can use it to show that the two
examples we looked at earlier of functions on R2 which
were discontinuous at (0, 0) are, in fact, continuous
everywhere else in R2 .

Theorem 8.2.1 Let λ ∈ R. If f : D → R and

g : D → R are both continuous at a point a ∈ D, then
the following functions are also continuous at a:

f +g; λf ; fg ; |f | .

Moreover, if g does not take the value 0 on D, then f /g

is also continuous at a.

Gap to fill in

6
Corollary 8.2.2 Let λ ∈ R.
If f : D → R and g : D → R are continuous (i.e.
continuous at every point of D) then so are f + g, λf ,
f g and |f |.
Also, the function fg is continuous, provided that g does
not take the value 0 on D.

Gap to fill in

7
Another result of this type concerns composition of
continuous functions.

Theorem 8.2.3 (Continuity of composite functions)

Let f : Rd → Rl and g : Rl → Rk be two continuous
functions, where d, l, k are positive integers.
Then the composite function g ◦ f : Rd → Rk defined by
(g ◦ f )(x) = g(f (x)) for x ∈ Rd is also continuous.

Proof.
Gap to fill in

8
This result also works point by point, provided that the
domains match up.
If f is continuous at a and g is continuous at f (a) then,
restricting attention to a domain where g ◦ f can be
defined, g ◦ f is continuous at a.
(Exercise. Fill in the details.)
Gap to fill in

9
8.3 Equivalent Definitions of Limits and
Continuity (in terms of ε and δ)

We begin by recalling two results from the module

G11ACF.

Proposition 8.3.1 (Characterization of the limit of

f : R → R)
Let f : R → R be a function and a ∈ R.
Then the limit limx→a f (x) of f (x) as x tends to a
exists and equals L ∈ R if and only if the following
condition holds:
for every ε > 0 there is δ > 0 such that, for x ∈ R, we
have

0 < |x − a| < δ ⇒ |f (x) − L| < ε. (1)

Note that δ will usually depend on ε.

The smaller ε is, the smaller we will need to make δ in
order for (1) to hold.

10
Correspondingly, for continuity, we have the following
result.

Proposition 8.3.2 (Characterization of continuity of

f : R → R at a)
Let f : R → R be a function and a ∈ R.
Then f is continuous at a if and only if the following
condition holds:
for every ε > 0 there is δ > 0 such that, for x ∈ R, we
have
|x − a| < δ ⇒ |f (x) − f (a)| < ε.

Now if D ⊆ Rd and f : D → Rl then we have the same

pair of propositions for the function f .
The proofs are omitted, as they are essentially identical
to those of the results from G11ACF which we recalled
above.
(Exercise. Fill in the details here.)

11
Proposition 8.3.3 (Characterization of the limit of
f : D → Rl )
Let f : D → Rl be a function and suppose that a is a
non-isolated point of D.
Then the limit limx→a f (x) of f (x) as x tends to a
exists and equals q ∈ Rl if and only if the following
condition holds:
for every ε > 0 there is δ > 0 such that, for x ∈ D,

0 < kx − ak < δ ⇒ kf (x) − qk < ε.

Proposition 8.3.4 (Characterization of continuity of

a function f : D → Rl at a ∈ D) Let f : D → Rl be a
function and a ∈ D.
Then f is continuous at a if and only if the following
condition holds:
for every ε > 0 there is δ > 0 such that, for x ∈ D,

kx − ak < δ ⇒ kf (x) − f (a)k < ε.

12
The following lemma, which is often useful, can be
shown using the above characterization of continuity.

Lemma 8.3.5 Let a ∈ D ⊆ Rd , f : D → R continuous.

Suppose f (a) > 0. Then there is δ > 0 such that
f (x) > 0 whenever x ∈ D and kx − ak < δ.
If D = Rd , this means that, with δ > 0 as above, our
function f is strictly positive throughout the open ball
Bδ (a).

Proof.
Gap to fill in

13
8.4 Images under continuous functions

The following is a very important property of sequentially

compact sets, and may be summarized as follows.
The continuous image of a sequentially compact set
is always sequentially compact.

Theorem 8.4.1 Let D be a non-empty, sequentially

compact subset of Rd , and suppose that f : D → Rl is a
continuous function.
Then the image of D under f , f (D) = {f (x) | x ∈ D},
is a sequentially compact subset of Rl .

Proof.
Gap to fill in

14
The situation is different for continuous images of other
types of sets.

1) Continuous images of open sets need not be open.

Gap to fill in

2) Continuous images of closed sets need not be closed.

Gap to fill in

15
3) Continuous images of bounded sets need not be
bounded.
Gap to fill in

Exercise. Find some more examples for yourself!

16
The special case of the above theorem where D ⊆ Rd is
sequentially compact and f : D → R is continuous is
particularly important.
Before looking at this, we prove an important lemma.

Lemma 8.4.2 Every non-empty, sequentially compact

subset of R has both a maximum element and a
minimum element.

Gap to fill in

17
Theorem 8.4.3 (The Boundedness Theorem for
sequentially compact subsets of Rd .) Let D be a
non-empty, sequentially compact subset of Rd , and let f
be a continuous function from D to R.
Then there are p and q ∈ D such that
f (p) ≤ f (x) ≤ f (q) for all x ∈ D.

Remark. Thus, in this setting, f is bounded above and

below on D and, among the values f takes on D, f
attains a maximum value at some point q of D and a
minimum value at some point p of D.
Proof.
Gap to fill in

18
In particular: If f : [a, b] → R is continuous then there
are p, q ∈ [a, b] such that f (p) ≤ f (x) ≤ f (q) for all
x ∈ [a, b].
(This is the version of the Boundedness Theorem which
you met in G11ACF.)

19
9 Sequences of Functions

9.1 Pointwise and uniform convergence

Let D ⊆ Rd , and let fn : D → R be functions (n ∈ N).

We may think of the functions f1 , f2 , f3 , . . . as forming a
sequence of functions.
This is very different from a sequence of numbers but it is still
possible to define the concept of a limit of such a sequence.
Such a notion is quite important.
For instance when solving differential equations or other
problems one is often able to produce a sequence of
approximate solutions and then needs to know in which sense
the approximate solutions converges to the exact one.
We shall discuss two of the main notions of convergence of
sequences of functions fn : D → R.
To help us, recall the following non-standard terminology,
introduced earlier.

1
 Definition 9.1.1 Let (xn ) ⊆ Rd and let A ⊆ Rd .
We say that the set A absorbs the sequence (xn ) if there
exists N ∈ N such that the following condition holds:
for all n ≥ N , we have xn ∈ A, (∗)
i.e., all terms of the sequence from xN onwards lie in the
set A.

The following is a slight variation of an exercise on a question

sheet.
The proof is an exercise.

Proposition 9.1.2 Let a ∈ R and let (xn ) ⊆ R. Then the

following statements are equivalent:
(a) the sequence (xn ) converges to a;
(b) for all ε > 0, the closed interval [a − ε, a + ε] absorbs the
sequence (xn ).

2
 Definition 9.1.3 The sequence (fn ) converges point-
wise (on D) to the function f if, for every x ∈ D, the se-
quence (fn (x))∞
n=1 converges to f (x), i.e., fn (x) → f (x)
as n → ∞.

Gap to fill in

The notion of uniform convergence is more subtle.

To explain this, we first extend our notions of closed ball and
of sets absorbing sequences.
We need to consider sets and sequences of functions.

3
Definition 9.1.4 Let D be a non-empty subset of Rd , let
f : D 7→ R, and let (fn ) be a sequence of functions from D
to R.
For ε > 0, we define the closed ball centred on f and with
radius ε, B̄ ε (f ), by

B̄ ε (f ) = {g : D → R | |g(x) − f (x)| ≤ ε for all x ∈ D }

= {g : D → R | g(x) ∈ [f (x) − ε, f (x) + ε] for all x ∈ D } .

Note that this closed ball is a set of functions from D to R.
Gap to fill in

4
As in (∗) above, we say that the closed ball B̄ ε (f ) absorbs
the sequence of functions (fn ) if there exists N ∈ N such that
the following condition holds:
for all n ≥ N , we have fn ∈ B̄ ε (f ), (∗∗)
i.e., all terms of the sequence of functions from fN onwards
lie in B̄ ε (f ).

The sequence (fn ) converges uniformly (on D) to the

function f if, for every ε > 0, the closed ball B̄ ε (f ) absorbs
the sequence (fn ).

In full, this means the following:

For all ε > 0, there exists an N ∈ N such that, for all

n ≥ N and all x ∈ D, we have

|fn (x) − f (x)| ≤ ε .

The N in the full definition of uniform convergence depends

only on ε; the same N works for all x ∈ D.
Roughly this means that the sequences of numbers (fn (x))
converge to f (x) at the same rate.

5
Pointwise convergence on the other hand means simply that
the sequence of numbers (fn (x)) converges to f (x) for each
x ∈ D.
At different points the speed of convergence could be very
different.
The next result follows directly from the definitions.
(Exercise. Convince yourself that this is correct.)

Lemma 9.1.5 If (fn ) converges uniformly to f (on D)

then it converges pointwise to f on D.

The converse of this lemma is NOT true.

It is time for some examples to illustrate this.

6
Examples 1) Let D = [0, 1] and fn (x) = xn . Then (fn )
converges pointwise, but NOT uniformly, to the function

 0 if 0 ≤ x < 1
f (x) =
 1 if x = 1.

Gap to fill in

7
2) Suppose D = [0, 1/2] and again fn (x) = xn . Now (fn )
converges uniformly to the function which is identically 0 i.e.
the function given by f (x) = 0 for all x ∈ [0, 1/2] (we say it
converges uniformly to 0).
Gap to fill in

8
3) Suppose D = R+ and fn : R+ → R is defined by

 x/n if 0 ≤ x ≤ n
fn (x) =
 1 if x > n.

Then the sequence of functions (fn ) converges to 0 pointwise,

but not uniformly, on R+ .
Gap to fill in

9
We will discuss a variety of methods to investigate the
convergence of sequences of functions.
For an alternative approach, involving the uniform norm, see
question sheets.

The above examples indicate that uniform convergence is

much stronger than pointwise convergence and more difficult
to establish.
One of the reasons that uniform convergence is important is
that it has much better properties.

In the above examples, all of the functions fn are continuous.

Unfortunately, as Example 1), shows the limit of a pointwise
convergent sequence of continuous functions need not be
continuous.
For uniformly convergent sequences the situation is much
better.

10
The proof of the next theorem is NEB (not examinable as
bookwork): see Wade’s book, Theorem 7.9, if you are
interested.
The statement and applications of this theorem are
examinable.

Theorem 9.1.6 Let D be a non-empty subset of Rd , let

f : D → R, and let (fn ) be a sequence of continuous
functions from D to R.
Suppose that the functions fn converge uniformly on D to
f.
Then f must also be continuous.

From now on you may quote this theorem as standard.

It may be summarized as follows:

Uniform limits of sequences of continuous functions

are always continuous.

11
This theorem is one of the main reasons why uniform
convergence is so important.
It sometimes gives you a quick way to see that certain
pointwise convergent sequences do not converge uniformly.
If all the fn are continuous but f isn’t then your sequence
cannot converge uniformly! (e.g. Example 1)
However, this trick does not always work.
Example 3) shows that a pointwise but non-uniform limit can
sometimes be continuous.
We conclude this section with some additional standard facts
which can sometimes help to establish that uniform
convergence fails.
The proofs of these are an exercise.

12
Proposition 9.1.7 Let D be a non-empty subset of Rd and
suppose that f is a bounded function from D to R, i.e.,
f (D) is a bounded subset of R.
(i) Let ε > 0 and suppose that g ∈ B̄ ε (f ). Then g is also
bounded on D.
(ii) Let (fn ) be a sequence of functions from D to R.
Suppose that all of the functions fn are unbounded on D.
Then (fn ) can not converge uniformly on D to f .

Another way to state this last result is:

It is impossible for a sequence of UNBOUNDED, real-

valued functions from D to R to converge uniformly
on D to a BOUNDED real-valued function.

13
 Since g ∈ B̄ ε (f ) ⇔ f ∈ B̄ ε (g), a similar proof shows the
following:

It is impossible for a sequence of BOUNDED func-

tions from D to R to converge uniformly on D to an
UNBOUNDED real-valued function.

14
10 Rigorous Differential Calculus

10.1 Differentiable functions

Let D be a subset of R such that int D 6= ∅.

Let f be a real valued function defined on D, and suppose
that a ∈ int D.

Definition 10.1.1 The function f is differentiable at

a if the limit
f (x) − f (a)
lim
x→a x−a
(or, equivalently, limh→0 f (a+h)−f
h
(a)
)
exists and is finite.
In this case we define the derivative of f at a, f 0 (a),
to be this limit.

1
In this setting, we obtain a new real-valued function f 0
defined at all those points of int D at which f is
differentiable.
Now let U be a non-empty, open subset of R with
U ⊆ int D.
The function f is said to be differentiable on U if it is
differentiable at every point of U .
In this case, the function f 0 is defined at all points of U
(and possibly also at some other points of int D).
In this setting, if D itself is open, and f is differentiable on
D (i.e., at every point of its domain), we simply say that f
is differentiable.
Gap to fill in

2
In words, the next result may be stated as follows.

Differentiability at a implies continuity at a, and

so differentiability implies continuity.

Theorem 10.1.2 Let f : ]b, c[ → R and let a ∈ ]b, c[.

If f is differentiable at a, then f is continuous at a.
Thus, if f is differentiable on ]b, c[, then f is continuous
on ]b, c[.

Proof.
Gap to fill in

3
Warning. The converse of this theorem is not true. For
example, the function f (x) = |x| defined on R is
continuous, but is not differentiable at 0.
Gap to fill in

4
The following more extreme example was found by
Weierstrass (using the very useful Weierstrass M -test for
uniform convergence: see books for details).
The infinite series
∞
X
f (x) = 2−n cos((21)n πx)
n=0

defines a continuous, real-valued function on R.

Weierstrass showed that it is nowhere differentiable on R!
Gap to fill in

5
You should be very familiar with the following pair of
standard results. We omit the proofs of these.
In fact the first is fairly easy to prove rigorously
(Exercise), but the second is a bit tricky (there is a
standard false proof available).
See books for more details.

Theorem 10.1.3 Let f , g be functions from ]a, b[ to R,

let x ∈ ]a, b[ and suppose that f and g are both
differentiable at x. Then f + g and f g are also
differentiable at x, and we have the following:
• (f + g)0 (x) = f 0 (x) + g 0 (x);
• (f g)0 (x) = f 0 (x)g(x) + f (x)g 0 (x) (product rule);
• if g does not take the value 0 on ]a, b[, then f /g is
also differentiable at x and
0 f 0 (x)g(x) − f (x)g 0 (x)
(f /g) (x) = (quotient rule).
g(x)2

6
Theorem 10.1.4 (chain rule) Let f : ]b, c[ → R and let
a ∈ ]b, c[. Let g be a real-valued function defined on an
open set containing f ( ]b, c[ ) (the image of f ).
Suppose that f is differentiable at a and that g is
differentiable at f (a).
Then the composite function G = g ◦ f : ]b, c[ → R, given
by
G(x) = g(f (x))
for all x ∈ ]b, c[ , is differentiable at a and

G0 (a) = g 0 (f (a))f 0 (a) .

From now on, you may assume that all the standard
differentiable functions (polynomials, etc.) you met in
G11CAL have the derivatives and properties discussed
there.

7
10.2 Fermat’s Theorem, Rolle’s Theorem
and the Mean Value Theorem

In this section we look at some very important theorems

and their applications.
We prove these results in full, as they form the basis of the
rigorous theory of differentiable functions.
We begin with a rigorous proof of an intuitively obvious
result due to Fermat.

Theorem 10.2.1 (Fermat’s Theorem for Station-

ary Points) Let a and b be real numbers with a < b,
let c ∈ ]a, b[, and let f : ]a, b[ → R. Suppose that f is
differentiable at c.
(a) If f (x) ≤ f (c) for all x ∈ ]a, b[, then f 0 (c) = 0.
(b) Similarly, if f (x) ≥ f (c) for all x ∈ ]a, b[, then
f 0 (c) = 0.

Gap to fill in

8
9
For the next theorem, we need to recall the Boundedness
Theorem.
If f : [a, b] → R is a continuous function then there are
p, q ∈ [a, b] such that, for all x ∈ [a, b], we have
f (p) ≤ f (x) ≤ f (q).

Theorem 10.2.2 (Rolle’s Theorem) Let a and b be

real numbers with a < b.
Let f : [a, b] → R be continuous on [a, b] and differen-
tiable on the open interval ]a, b[.
Suppose that f (a) = f (b).
Then there exists c ∈ ]a, b[ such that f 0 (c) = 0.

Proof.
Gap to fill in

10
11
Using Rolle’s Theorem, we can now prove the Mean Value
Theorem (MVT).

Theorem 10.2.3 (Mean Value Theorem) Let a and

b be real numbers with a < b.
Let f : [a, b] → R be continuous on [a, b] and differen-
tiable on the open interval ]a, b[.
Then there exists c ∈ ]a, b[ such that

f (b) − f (a) = f 0 (c)(b − a) .

Before giving the formal proof, we illustrate the result with

a diagram.
Gap to fill in

12
Proof. Define α = (f (b) − f (a))/(b − a) and let
g(x) = f (x) − αx.
Then g is a continuous function on [a, b] which is
differentiable on ]a, b[ (because f is) and

g 0 (x) = f 0 (x) − α

for all x ∈]a, b[.

Also,
g(b) − g(a) = f (b) − αb − (f (a) − αa)
= f (b) − f (a) − α(b − a) = 0 ,
i.e. g(a) = g(b).
Thus g satisfies the conditions of Rolle’s Theorem.
By Rolle’s Theorem there exists c ∈ ]a, b[ such that
g 0 (c) = 0.
But g 0 (c) = f 0 (c) − α so, for this c, we have f 0 (c) = α as
required. 

13
 Definition 10.2.4 Let I be some interval (open, half-
open or closed) and f : I → R a function.
The function f is non-decreasing (or increasing) on
I if,for all x, y ∈ I with x < y, we have f (x) ≤ f (y).
Similarly, f is non-increasing (or decreasing) on I if,
for all x, y ∈ I with x < y, we have f (x) ≥ f (y).

Functions which are non-decreasing on I or non-increasing

on I are also said to be monotonic (or monotone) on I.

Similarly, f is said to be strictly increasing (on I)

if ≤ can be replaced by < above, and f is strictly
decreasing (on I) if ≥ can be replaced by > above.

Gap to fill in

14
Theorem 10.2.5 Let a, b ∈ R with a < b.
Let f be a differentiable function from ]a, b[ to R. Then
1. f 0 (t) ≥ 0 for all t ∈ ]a, b[ =⇒ f is non-decreasing on
]a, b[;
2. f 0 (t) = 0 for all t ∈ ]a, b[ =⇒ f is constant on ]a, b[;
3. f 0 (t) ≤ 0 for all t ∈ ]a, b[ =⇒ f is non-increasing on
]a, b[;
4. Similar results hold for strict monotonicity if the
inequalities above are replaced by strict inequalities
( > or < ).
5. Similar results hold for differentiable functions defined
on unbounded open intervals of R.

These can all be proved using the Mean Value Theorem

carefully.

15
 The key is to note that, for each pair x < y in ]a, b[ we
are able to find at least one point c ∈ ]x, y[ such that

f (y) − f (x) = (y − x)f 0 (c) . (∗)

See the comments on the 2005-6 exam perfor-

mance for some warnings on this!

For example, if we know that f 0 (t) ≥ 0 for all t ∈ ]a, b[ ,

then in particular f 0 (c) ≥ 0 and so, by (∗),
f (y) − f (x) ≥ 0.
The remaining cases are exercises.
We conclude this chapter with another application of the
MVT.

16
Example 10.2.6 Let b > 0. Apply the mean value
theorem to the function f (t) = log(1 + t) on the interval
[0, b] to prove that
b
< log(1 + b) < b .
1+b
Gap to fill in

17
11 An introduction to Riemann
Integration

The PROOFS of the standard lemmas and theorems

concerning the Riemann Integral are NEB, and you
will not be asked to reproduce proofs of these in full
in the examination in January 2010.
However, you ARE expected to know the definitions
and the statements of the results, and to know how
to apply these results.
In particular, the EXAMPLES discussed in lectures
ARE examinable.

1
 Note. In 2003-4, 2004-5 and 2005-6 the notation and def-
initions used for the Riemann integral were somewhat dif-
ferent, and used a somewhat technical, formal definition
of step function.
Since 2006-7, we have used a slightly more standard ap-
proach.

11.1 Integration and antidifferentiation

From the modules G11CAL and G11ACF you will be familiar

with integration as a form of antidifferentiation, and as ‘area
under the curve’.
The problem is: how do you know which functions have
antiderivatives?
An antiderivative for a function f is a differentiable
function F which satisfies F 0 = f .
Antiderivatives are also called primitives or indefinite
integrals.

2
Recall that the characteristic function of a set E, χE , is
defined by χE (x) = 1 if x ∈ E while χE (x) = 0 if x ∈
6 E.
On Question Sheet 5 there is an example of a function
which has no antiderivative: the characteristic function of a
set with just one point.
This shows that some care is needed!
One of the main results of this chapter is the (first)
Fundamental Theorem of Calculus, one implication of
which is that every continuous, real-valued function on an
interval has an antiderivative.
Of course, some discontinuous functions do have
antiderivatives (can you think of an example?).

3
11.2 Partitions, areas and Riemann sums

Let a and b be real numbers with a < b.

For a bounded, real-valued function f defined on [a, b], we
now discuss partitions P of [a, b] and the corresponding
Riemann upper sum and Riemann lower sum for f on
[a, b] (denoted by U (P, f ) and L(P, f ) respectively).
The bounded function f and the interval [a, b] will be fixed
throughout the following definitions.

Definition 11.2.1 A partition of [a, b] is a finite set of

points P = {x0 , x1 , . . . , xn } ⊆ [a, b], where
a = x0 < x1 < · · · < xn = b. The points x0 , x1 , . . . , xn are
called the vertices of P .

Gap to fill in

4
For 1 ≤ k ≤ n, we set

Mk (P, f ) = sup{f (t) | xk−1 ≤ t ≤ xk }

and
mk (P, f ) = inf{f (t) | xk−1 ≤ t ≤ xk } .

Gap to fill in

5
The Riemann upper sum for f corresponding to P ,
U (P, f ), and the Riemann lower sum for f corresponding
to P , L(P, f ), are defined by

n
X
U (P, f ) = Mk (P, f )(xk − xk−1 )
k=1

and
n
X
L(P, f ) = mk (P, f )(xk − xk−1 ) .
k=1

Gap to fill in

6
It is obvious that we always have L(P, f ) ≤ U (P, f ).
With a bit more work, we can prove the following fact (see
books for details).

Lemma 11.2.2 Let P and Q be partitions of [a, b].

Then L(P, f ) ≤ U (Q, f ).

In words, this tells us the following.

The Riemann lower sum for f corresponding to a partition

P of [a, b] can not be greater than the Riemann upper sum
for f corresponding to a partition Q of [a, b], even if P and
Q are different.

7
11.3 The Riemann integral

With f and [a, b] as above, the preceding lemma allows us to

define the Riemann lower and upper integrals of f over the
interval [a, b].

Definition 11.3.1 The Riemann lower integral of f over

Rb
the interval [a, b], a f (x) dx and the Riemann upper
Rb
integral of f over the interval [a, b], a f (x) dx are defined
by

Z b
f (x) dx = sup{L(P, f ) : P is a partition of [a, b] }
a

and
Z b
f (x) dx = inf{U (Q, f ) : Q is a partition of [a, b] } .
a

8
This in turn allows us to define Riemann integrability for f .

Definition 11.3.2 The bounded function f is Riemann

integrable on [a, b] if
Z b Z b
f (x) dx = f (x) dx ,
a a

i.e., if the Riemann lower integral is equal to the Riemann

upper integral.
In this case we define the Riemann integral of f from a
to b to be the common value:

Z b Z b Z b
f (x) dx = f (x) dx = f (x) dx.
a a a

UNBOUNDED functions on an interval [a, b] are declared

NOT to be Riemann integrable.

However they may have ‘improper’ integrals (as discussed in

G11ACF).

9
Note the following facts for bounded, real-valued functions on
an interval.

• The Riemann lower integral is always less than or equal

to the Riemann upper integral.
• Every Riemann lower sum is less than or equal to the
lower integral.
• Every Riemann upper sum is greater than or equal to
the upper integral.

It is easy to show that constant functions are Riemann

integrable, with the obvious integral (exercise).
We will see below that the family of Riemann-integrable
functions is fairly large.
However, not all bounded functions are Riemann integrable.

10
Example. Let f be the characteristic function of the
rationals, χQ .
Then f is not Riemann integrable on [0, 1].
Remark: Strictly speaking here, we mean that the restriction
of f to [0, 1] is not Riemann integrable on [0, 1].
Gap to fill in

11
However, continuous functions are well-behaved.

Theorem 11.3.3 Every continuous, real-valued function

on an interval [a, b] is Riemann integrable on [a, b].

The Riemann integral behaves as you expect a sensible notion

of integration to behave.

Theorem 11.3.4 Let f, g : [a, b] → R be Riemann integrable

and λ ∈ R. Then f + g, λf and |f | are also Riemann
integrable and the following hold.
Z b Z b Z b
(a) (f (x) + g(x)) dx = f (x) dx + g(x) dx
a a a
(additivity).
Z b Z b
(b) λf (x) dx = λ f (x) dx.
a a

(c) If f (x) ≤ g(x) for all x ∈ [a, b], then

Z b Z b
f (x) dx ≤ g(x) dx.
a a

12
(d) We always have

Z  Z
 b  b
f (x) dx ≤ |f (x)| dx .
 

 a  a

(e) For any c ∈ ]a, b[, we have that f is also Riemann

integrable on [a, c] and on [c, b] and
Z b Z c Z b
f (x) dx = f (x) dx + f (x) dx .
a a c

We now come to the final (and main) results of this chapter.

13
Theorem 11.3.5 (Fundamental Theorem of Calculus,
also known as the First Fundamental Theorem of Cal-
culus)
Let f be a continuous, real-valued function on [a, b].
For x ∈ [a, b] define
Z x
F (x) = f (t) dt.
a

Then F is continuous on [a, b], and is differentiable on

]a, b[, with F 0 (x) = f (x) for all x ∈ ]a, b[.

Gap to fill in

14
From this it follows easily that continuous, real-valued
functions on intervals always have antiderivatives.
It also shows that antidifferentiation is the correct way to
integrate continuous functions.
See Question Sheet 5 for more details.

Theorem 11.3.6 (Mean Value Theorem of Integral

Calculus) Let a and b be real numbers with a < b and
let f : [a, b] → R be continuous.
Then there exists x0 ∈ [a, b] such that
Z b
f (x) dx = f (x0 )(b − a).
a

Gap to fill in

15
There is a more powerful integration theory due to Henri
Lebesgue.
In this theory, χQ is integrable on [0, 1] (with
R1
χ (x) dx = 0), and so are many other strange functions.
0 Q

The Lebesgue integral is beyond the scope of this module, but

it is an important tool in more advanced analysis and in
probability theory.
THE END
Have a good holiday!

16