FOUNDATIONS

of CALCULUS

John Hutchinson
email: [email protected]
 Contents

Introduction 5

Chapter 1. The Real Numbers 9

1.1. Preliminary remarks 9
1.1.1. Decimal expansions 9
1.1.2. Geometric representation 9
1.1.3. Different decimal expansions for the same number 11
1.1.4. Density of the rationals and the irrationals 11
1.2. Algebraic and Order Properties 12
1.2.1. Algebraic and Order Axioms 12
1.2.2. Algebraic consequences 13
1.2.3. Order consequences 18
1.2.4. Our approach henceforth 19
1.2.5. Natural and Rational numbers 19
1.2.6. FFields 19
1.2.7. FGroups 20
1.3. Completeness Axiom 21
1.3.1. Statement of the Axiom 22
1.3.2. An equivalent formulation 22
1.3.3. Interpretation of the Completeness Axiom 23
1.3.4. Archimedean Property 24
1.3.5. FFHyperreals 25
1.4. Sets 26
1.4.1. Notation for sets 26
1.4.2. Ordered pairs of real numbers 28

Chapter 2. Sequences 31
2.1. Examples and Notation 31
2.2. Convergence of Sequences 31
2.3. Properties of Sequence Limits 34
2.4. Proofs of limit properties 35
2.5. More results on sequences 40
2.6. Bolzano Weierstrass Theorem 42
2.7. FCauchy Sequences 44

Chapter 3. Continuous Functions 47

3.1. The Approach in these Notes 47
3.2. Definition via Sequences and Examples 47
3.3. Basic Properties of Continuous Functions 50
3.3.1. Combining Continuous Functions 50
3.3.2. Analogous Results at a Point 51
3
4 CONTENTS

3.3.3. Removable and Non-Removable Singularities 51

3.4. Continuous Functions on a Closed Bounded Interval 53
3.5. FFunctions of two or more variables 55
Chapter 4. Limits 57
4.1. Definition via Sequences, and Examples 57
4.2. Calculating Limits 58
4.3. Limits and Continuity 58
4.4. Definitions of Limit and Continuity via  − δ 59
4.5. Uniform Continuity 62
Chapter 5. Differentiation 65
5.1. Introduction 65
5.2. The Derivative of a Function 65
5.3. Computing Derivatives 68
5.3.1. Sums, Products and Quotients 68
5.3.2. The Chain Rule 70
5.4. Maximum and Minimum Values 72
5.5. Mean Value Theorem 73
5.6. FPartial derivatives 76
Chapter 6. Integration 77
6.1. Introduction 77
6.2. The Riemann integral 77
6.3. Riemann sums 84
6.4. Properties of the Riemann integral 85
6.5. Fundamental Theorem of Calculus 87
Chapter 7. FDifferential Equations 91
7.1. Overview 91
7.2. Outline of proof of the Existence and Uniqueness theorem 93
7.3. FRigorous proof of the Existence and Uniqueness theorem 96
 Introduction

These notes present the theoretical foundations of Calculus. As such

they are an introduction to the mathematical field of Analysis. More gener-
ally, they are an introduction to the methods used in modern mathematics.
In the process of working through this material you will prove the major
theorems used in the Calculus section of MATH1115 and to a lesser extent
in MATH1116, and thereby obtain a more fundamental understanding of
that material.
Some of the material here will be part of your first year courses, some
will be supplementary and covered in later courses. It is all important
mainstream mathematics.

There are essentially two parts to the Calculus section of MATH1115

— theoretical foundations on the one hand; methods, techniques and appli-
cations on the other. The former is treated here, more thoroughly than in
Adams. The latter is covered in Adams, and is an extension of the material
in the higher level school/college calculus courses.

Mathematics is the study of pattern and structure. It is studied both

for its universal applicability and its internal beauty. In mathematics we
make certain specific assumptions (or axioms) about the objects we study
and then develop the consequences of these assumptions in a precise and
careful manner. The axioms are chosen because they are “natural” in some
sense; it usually happens that these axioms also describe phenomena in other
subjects, in which case the mathematical conclusions we draw will also apply
to these phenomena.
Areas of mathematics developed for “mathematical” reasons usually
turn out to be applicable to a wide variety of subjects; a spectacular recent
example being the applications of differential geometry to understanding
the fundamental forces of nature studied in physics, and another being the
application of partial differential equations and geometric measure theory to
the study of visual perception in biology and robotics. There are countless
other examples in engineering, economics, and the physical and biological
sciences. The study of these disciplines can usually only be done by apply-
ing the techniques and language of mathematics. On the other hand, their
study often leads to the development of new fields of mathematics and gives
insights into old fields.

In these notes you will study the real number system, the concepts of
limit and continuity, differentiability and integrability, and differential equa-
tions. While most of these terms will be familiar from high school in a more
or less informal setting, you will study them in a much more precise way.
5
6 INTRODUCTION

This is necessary both for applications and as a basis for generalising these
concepts to other mathematical settings.
One important question to be investigated in the last chapter is: when
do certain types of differential equations have a solution, when is there ex-
actly one solution, and when is there more than one solution? The solution
of this problem uses almost all the earlier material. The study of differential
equations is of tremendous importance in mathematics and for its applica-
tions. Any phenomenon that changes with position and/or time is usually
represented by one or more differential equations.
The ideas developed here are basic to further developments in mathe-
matics. The concepts generalise in many ways, such as to functions of more
than one variable and to functions whose variables are themselves functions
(!); all these generalisations are fundamental to further applications.
At the end of the first semester you should have a much better under-
standing of all these ideas.
These notes are intended so that you can concentrate on the relevant
lectures rather than trying to write everything down. There may occcasion-
ally be lecture material on this part of the course which is not mentioned in
the notes here, in which case that will be indicated.
There are quite a few footnotes. This can be annoying. You should
read the footnotes when you initially study the material. But after you have
noted and understood the footnotes, you do not need to reread them every
time. Instead, you should concentrate on the main ideas in the main body
of the notes.
References are to the seventh edition of the text Calculus text by Adams,
and occasionally to the book Calculus by Michael Spivak.
Why go to the lectures? Because the notes are frequently rather formal
(this is a consequence of the precision of mathematics) and it is often very
difficult to see the underlying concepts. In the lectures the material is ex-
plained in a less formal manner, the key and underlying ideas are singled
out and discussed, and generally the subject is explained and discussed in a
manner which it is not possible to do efficiently in print. It would be a very
big mistake to skip lectures.
Do not think that you have covered any of this material in school; the
topics may not appear new, but the material certainly will be. Do the
assignments, read the lecture notes before class. Mathematics is not a body
of isolated facts; each lecture will depend on certain previous material and
you will understand the lectures much better if you keep up with the course.
In the end this approach will be more efficient as you will gain more from
the lectures.
Throughout these notes I make various digressions and additional re-
marks, marked clearly by a star F. This would generally be non-examinable
and is (even) more challenging material. But you should still read and think
about it. It is included to put the subject in a broader perspective, to pro-
vide an overview, to indicate further directions, and to generally “round
out” the subject. In addition, studying this more advanced material will
help your understanding of the examinable material.
 INTRODUCTION 7

There are a number of places where I ask why? Don’t just convince
yourself informally that it is indeed so. Write down a careful proof and then
copy it into the margin of these notes.
Some of the proofs of theorems are quite tricky, certainly at first. In this
case just try to understand what the theorem is saying, think about some
examples, and think why the various hypotheses are necessary, and think
about how they are used in the proof. There are examples which discuss
some of these points before or after some of the more difficult theorems.
Studying mathematics is not like reading a book in other subjects. It
may take many hours to understand just one sentence or one paragraph.
When you get stuck, it will often help to eventually continue on, and then
later come back to the difficult points. Also, ask your tutor, your lecturer,
a fellow student, or an assistant in the mathematics “drop in” centre. Do
not let things slide!
The study of Mathematics is not easy, but it is challenging, rewarding
and enjoyable.
 CHAPTER 1

The Real Numbers

You should begin by first reviewing the material in Chapter P1 of Adams,

pages 3–9; and particularly pages 3 and 4.
We begin with a brief discussion of a few properties of the real numbers.
We then discuss the algebraic and order properties and indicate how they
follow from 13 basic properties called the Algebraic and Order Axioms. Fi-
nally we discuss the Completeness Axiom. All properties of the real numbers
follow from these 14 axioms.

1.1. Preliminary remarks

1.1.1. Decimal expansions. Real numbers have decimal expansions,
for example:
2 = 2.000 . . .
1
1 = 1.5 = 1.5000 . . .
2
π = 3.14159 . . .
.45271̇46̇, also written .4527146.
The “. . . ” indicate the expansions go on forever, and the 1̇46̇ indicate that
the pattern 146 is repeated forever. In the first two cases the expansion
continues with zeros and in the third case one can compute the expansion
to any required degree of accuracy.
Instead of counting with base 10, we could count with any other integer
base b ≥ 2. In this case, for integers we write
b1 b2 . . . bn = bn−1 b1 + · · · + b2 bn−2 + bbn−1 + bn ,
where each bi takes values in the set {0, 1, . . . , b − 1}.(Some societies did
count with other than base 10.)
How is 123 written in base 16 and in base 2?
We can also write nonintegers using base b. In particular,
b1 b2 bn
·b1 b2 . . . bn = + 2 + ··· + n.
b b b
1.1.2. Geometric representation. Real numbers can be represented
geometrically as points on an infinite line.

Figure 1. Geometric representation of real numbers as

points on an (infinite) line.
9
10 1. THE REAL NUMBERS

The ancient Greeks thought of real numbers as lengths of lines, and

they knew that if x is the length of the hypotenuse of the following right
angled triangle, then√its square must satisfy x2 = 12 + 12 = 2 (Pythagoras’s
theorem). We write 2 for this number x.

Figure 2. Is the length of the hypotenuse given by a rational

number? No!
The Greeks also thought of real numbers as ratios of integers (or what
we now call rational numbers).
So when they discovered the following theorem, they were very upset.
√
Theorem 1.1.1. 2 is not rational.
Proof. We argue by contradiction. That is, we assume
√
2 = m/n
where m and n are integers.
Multiplying numerator and denominator by −1 if necessary, we can take
m and n to be positive. By cancelling if necessary, we can reduce to the
situation where m and n have no common factors. Squaring both sides of
the equation, we have for these new m and n that
2 = m2 /n2
and hence
m2 = 2n2 .
It follows that m is even, since the square of an odd number is odd.
(More precisely, if m were odd we could write m = 2r + 1 for some integer
r; but then m2 = (2r + 1)2 = 4r2 + 4r + 1 = 2(2r2 + 2r) + 1, which is odd,
not even.)
Since m is even, we can write
m = 2p
for some integer p, and hence
m2 = 4p2 .
Substituting this into m2 = 2n2 gives
4p2 = 2n2 ,
and hence
2p2 = n2 .
But now we can argue as we did before for m, and deduce that n is also
even. Thus m and n both have the common factor 2, which contradicts the
fact that they have no common factors.
√This contradiction implies that our original assumption was wrong, and
so 2 is not rational. 
 1.1. PRELIMINARY REMARKS 11
√
Use a similar argument to prove 3 is irrational. HINT: Instead of
considering even and odd integers (i.e. remainder 0 or 1 after dividing by 2),
you will need to consider integers with remainders 0, 1 or 2 after dividing
by 3.
1.1.3. Different decimal expansions for the same number. There
is one point that sometimes causes confusion. Is it the case that
1 = .9̇ ?,
or is it that .9̇ is a “little” less than one? By .9̇ we mean, as usual, .999 . . . ,
with the 9’s repeated forever.
Each of the following approximations to .9̇,
9 99 999 9999
.9 = , .99 = , .999 = , .9999 = , ...
10 100 1000 10000
is certainly strictly less than one.
On the other hand, .9̇ is defined to be the “limit” of the above infinite
sequence (we discuss limits of sequences in a later chapter). Any mathemat-
ically useful way in which we define the limit of this sequence will in fact
imply that .9̇ = 1. To see this, let
a = .9̇ = .999 . . . .
Then, for any reasonable definition of infinite sequence and limit, we would
want that

10a = 9.999 . . . .
Subtracting gives 9a = 9, and hence a = 1.
The only way a real number can have two decimal expansions is for it
to be of the form
.a1 a2 . . . an−1 an = .a1 a2 . . . an−1 (an − 1)9̇ .
For example,
.2356 = .23559̇ .
1.1.4. Density of the rationals and the irrationals. We claim that
between any two real numbers there is a rational number (in fact infinitely
many of them) and an irrational number (in fact infinitely many).
To see this, first suppose 0 < a < b.
Choose an integer n such that n1 < b − a. Then at least one member m n
of the sequence
1 2 3 4 5
, , , , ,...
n n n n n
will lie between a and b. To see this, take the first integer m such that
a< m m
n . It follows that n < b. Why?
1

Since we can similarly obtain another rational between mn and b, and yet
another rational between this rational and b, etc., etc., there is in fact an
infinite number of rationals between a and b.
1First try to understand this geometrically. Then write out an algebraic proof, which
should only be a couple of lines long!
12 1. THE REAL NUMBERS

Figure 3. Since b − a > 1/n, some integer multiple of 1/n

must lie between a and b.

If a < 0, a similar argument works with the sequence

1 2 3 4 5
− ,− ,− ,− ,− ,...
n n n n n
√
Finally, choosing n so n2 < b − a and applying a similar argument to
the sequence √ √ √ √ √
2 2 2 3 2 4 2 5 2
, , , , ,...
n n n n n
gives the result for irrational2 numbers.

1.2. Algebraic and Order Properties

We introduce the Algebraic and Order Axioms for the real
number system and indicate how all the usual algebraic
and order properties follow from these.
In a later section we discuss the Completeness Axiom.

1.2.1. Algebraic and Order Axioms. The real number system con-
sists of the real numbers, together with the two operations addition (denoted
by +) and multiplication (denoted by ×) and the less than relation (denoted
by <). One also singles out two particular real numbers, zero or 0 and one
or 1.
If a and b are real numbers, then so are a + b and a × b. We say that
the real numbers are closed under addition and multiplication. We usually
write
ab for a × b.
For any two real numbers a and b, the statement a < b is either true or
false.
We will soon see that one can define subtraction and division in terms
of + and ×. Moreover, ≤, > etc. can be defined from <.

Algebraic Axioms. For all real numbers a, b and c:

(1) a + b = b + a (commutative axiom for addition)
(2) (a + b) + c = a + (b + c) (associative axiom for addition)
(3) a + 0 = 0 + a = a (additive identity axiom)
(4) there is a real number, denoted −a, such that
a + (−a) = (−a) + a = 0 (additive inverse axiom)
(5) a × b = b × a (commutative axiom for multiplication)
(6) (a × b) × c = a × (b × c) (associative axiom for multiplication)
√
2The number m 2
is irrational, because if it were rational then on multiplying by n
√n m
we would get that 2 is rational, which we know is not the case.
 1.2. ALGEBRAIC AND ORDER PROPERTIES 13

(7) a × 1 = 1 × a = a, moreover 0 6= 1 (multiplicative identity axiom)

(8) if a 6= 0 there is a real number, denoted a−1 , such that
a × a−1 = a−1 × a = 1 (multiplicative inverse axiom)
(9) a × (b + c) = a × b + a × c (distributive axiom)

Order Axioms. For all real numbers a, b and c:

(10) exactly one of the following holds:
a < b or a = b or b < a (trichotomy axiom)
(11) if a < b and b < c, then a < c (transitivity axiom)
(12) if a < b then a + c < b + c (addition and order axiom)
(13) if a < b and 0 < c, then a × c < b × c (multiplication and order
axiom)

There are a number of points that need to be made at this stage, before
we proceed to discuss the consequences of these axioms.
• By the symbol “=” for equality we mean “denotes the same thing
as”, or equivalently, “represents the same real number as”. We take
“=” to be a logical notion and do not write down axioms for it.3
Instead, we use any properties of “=” which follow from its logical
meaning. For example: a = a; if a = b then b = a; if a = b and
b = c then a = c; if a = b and something is true of a then it is also
true of b (since a and b denote the same real number!).
When we write a 6= b, we just mean that a does not denote the
same real number as b.
• We are not really using subtraction in the algebraic Axiom 4; we
are merely asserting that a real number, with a certain property,
exists. It is convenient to denote this number by −a. A similar
remark applies for Axiom 8.
• The assertion 0 6= 1 in Axiom 7 may seem silly. But it does not
follow from the other axioms, since all the other axioms hold for
the set containing just the number 0.
• Parts of some of the axioms are redundant. For example, from
Axiom 1 and the property a + 0 = a it follows that 0 + a = a.
Similar comments apply to Axiom 4; and because of Axiom 5 to
Axioms 7 and 8.

1.2.2. Algebraic consequences. All the usual algebraic properties of

the real numbers follow from Axioms 1–9. In particular, one can solve si-
multaneous systems of linear equations. We will not spend much time on
indicating how one deduces algebraic properties from these axioms, but will
continue to use all the usual properties of addition, multiplication, subtrac-
tion and division that you have used in the past.
Nonetheless, it is useful to have some idea of the methods involved in
making deductions from axioms.

3FOne can write down basic properties, i.e. axioms, for “=” and the logic we use.
See later courses on the foundations of mathematics.
14 1. THE REAL NUMBERS

1.2.2.1. Sum of three or more numbers. The expression a + b + c is at

first ambiguous. Does it mean (a + b) + c or a + (b + c)? The first expression
means add a to b, then add c to the result; the second means add a to (the
result of adding b to c). By the associative axiom, the result is the same
in either case, and so we can define a + b + c to be either (a + b) + c or
a + (b + c).4
Using also the commutative axiom we also have
(a + b) + c = (b + a) + c = b + (a + c) = b + (c + a),
etc.
Similar remarks apply to the product of three or more numbers.
1.2.2.2. Subtraction and Division. We use the axioms to define these
operations in terms of addition and multiplication by
a − b = a + (−b)
a ÷ b = a × b−1 for b 6= 0.
This may look like a circular definition; it may appear that we are defining
“subtraction” in terms of “subtraction”. But this is not the case. Given b,
from Axiom 4 there is a certain real number, which we denoted by −b, with
certain properties. We then define a − b to be the sum of a and this real
number −b.
Similar comments apply to the definition of division. We also write a/b
or ab for a ÷ b. Division by 0 is never defined.
1.2.2.3. Other definitions. We can also now define other numbers and
operations. For example, we define 2 = 1 + 1, 3 = 2 + 1, etc.

2
Also, we define x2 = x × x, x3 = x × x × x, x−2 = x−1 , etc. etc.
1.2.2.4. Cancellation property of addition. As an example of how to use
Axioms 1–9 to derive other algebraic properties, we prove the cancellation
property of addition, which says informally that if a + c = b + c then we can
“cancel” the number c.
Theorem 1.2.1 (Cancellation Theorem). If a, b and c are real numbers
and a + c = b + c, then a = b. Similarly, if c + a = c + b then a = b.

Proof. Assume
a + c = b + c.
Since a + c and b + c denote the same real number, we obtain the same result
if we add −c to both; i.e.
(a + c) + (−c) = (b + c) + (−c).
(This used the existence of the number −c from Axiom 4.) Hence
a + (c + (−c)) = b + (c + (−c))
from Axiom 2 applied twice, once to each side of the equation. Hence
a+0=b+0
4A similar remark is not true for subtraction, since (a − b) − c and a − (b − c) are in
general not equal.
 1.2. ALGEBRAIC AND ORDER PROPERTIES 15

from Axiom 4 again applied twice. Finally,

a=b
from Axiom 3.
If c + a = c + b, then from the commutative axiom a + c = b + c, and we
have just seen that this implies a = b. 
1.2.2.5. Characterisation of 0 and of −a. One of the axioms is that 0
has the property:
(1) a+0=0+a=a
for every real number a. Does this property characterise 0? In other words,
is there any other real number x with the property that
(2) a+x=x+a=a
for every real number a?
Of course we know the answer is NO, but the point here is that the
answer follows from the axioms.
In fact from (1) and (2) we have a + 0 = a + x, and so from (the second
part of) Theorem 1.2.1 with c, a, b there replaced by a, 0, x respectively, it
follows that 0 = x.
One of the axioms asserts that for each number a there is a number
(denoted by −a) which when added to a gives zero. But could there be
another number which when added to a also gives zero? We know the
answer is NO, but this fact does not need to be asserted as a separate
axiom, because it also follows from the existing axioms.
In fact, suppose x + a = 0. Because we already know (−a) + a = 0 for
some specific number −a, it follows that x + a = (−a) + a. We can “cancel”
the a by Theorem 1.2.1, giving x = −a.
1.2.2.6. More algebraic consequences. Certain not so obvious “rules”,
such as “the product of minus and minus is plus” and the rule for adding
two fractions, follow from the axioms. If we want the properties given by
Axioms 1–9 to be true for the real numbers (and we do), then there is no
choice other than to have (−a)(−b) = ab and (a/c) + (b/d) = (ad + bc)/cd
(see the following theorem).
We will not emphasise the idea of making deductions from the axioms,
and for this reason I have marked the proofs of the assertions in the following
theorem as F material. Nonetheless, you should have some appreciation of
the ideas involved, and so you should work through a couple of the proofs.
Theorem 1.2.2. If a, b, c, d are real numbers and c 6= 0, d 6= 0 then
(1) ac = bc implies a = b.
(2) a0 = 0
(3) −(−a) = a
(4) (c−1 )−1 = c
(5) (−1)a = −a
(6) a(−b) = −(ab) = (−a)b
(7) (−a) + (−b) = −(a + b)
(8) (−a)(−b) = ab
(9) (a/c)(b/d) = (ab)/(cd)
16 1. THE REAL NUMBERS

(10) (a/c) + (b/d) = (ad + bc)/cd

Proof. F Each line in the following proofs will be

(1) an example of one (or occasionally more) of axioms 1–9;
(2) a previously proved result;
(3) or follow from previously proved results by rules of logic5 (which
include the properties of equality).
Fill in any missing steps. Go through the proofs line by line and indicate
what is used to justify each step.

1. Write out your own proof, following the ideas of the proof of the similar
result for addition.

2. The trick here is to use the fact 0 + 0 = 0 (from A3), together with the
distributive axiom. The proof is as follows:
One has a(0 + 0) = a0
But the left side equals a0 + a0
and the right side equals 0 + a0
Hence a0 + a0 = 0 + a0
Hence a0 = 0.

3. We want to show −(−a) = a.

By −(−a) we mean the negative of −a, and hence by Axiom 4 we know
that6
(−a) + (−(−a)) = 0.
But from Axiom 3
(−a) + a = 0
Hence
(−a) + (−(−a)) = (−a) + a
and so −(−a) = a from the Cancellation Theorem.

4. Write out your own proof, along similar lines to the preceding proof.
You should first prove a cancellation theorem for multiplication.

5. (As in the proof of 2.) it is sufficient to show (−1)a + a = 0, because

then
(−1)a + a = (−a) + a (additive inverse axiom)
and so
(−1)a = −a
by the Cancellation Theorem.

5For example, if we prove that some statement P implies another statement Q, and
if we also prove that P is true, then it follows from rules of logic that Q is true.
6Since a can represent any number in Axiom 4, we can replace a in Axiom 4 by −a.
This might seem strange at first, but it is quite legitimate.
 1.2. ALGEBRAIC AND ORDER PROPERTIES 17

The proof is as follows:

(−1)a + a = (−1)a + 1a
= a((−1) + 1) (two axioms were used for this step)
= a0
=0
This completes the proof.

6.
a(−b) = a((−1)b)
= (a(−1))b
= ((−1)a)b
= (−1)(ab)
= −(ab)
Prove the second equality yourself.

7. Prove this yourself using, in particular, use 4 and Axiom 9.

8.
(−a)(−b) = ((−1)a)(−b)
= (−1)(a(−b))
= −(a(−b))
= −(−(ab))
= ab

9. First note that (a/c)(b/d) = (ac−1 )(bd−1 )

and (ab)/(cd) = (ab)(cd)−1 .
But (ac−1 )(bd−1 ) = (ab)(c−1 d−1 )
(fill in the steps to prove this equality; which involve a number of applications
of Axioms 5 and 6).
If we can show that c−1 d−1 = (cd)−1 then we are done.
Since, by Axiom 8, (cd)−1 is the unique real number such that (cd)(cd)−1 =
1, it is sufficient to show7 that (cd)(c−1 d−1 ) = 1.
Do this; use A5–A8.
This completes the proof.
Important Remark : There is a tricky point in what we havew just done
that is easy to overlook; but will introduce some important ideas about
logical reasoning.
We used the number (cd)−1 .
To do this we need to know that cd 6= 0.
We know that c 6= 0 and d 6= 0 and we want to prove that cd 6= 0.
This is equivalent to proving that if cd 6= 0 is false, i.e. if cd = 0, then at
least one of c 6= 0 and d 6= 0 is false, i.e. at least one of c = 0 or d = 0 is
7When we say “it is sufficient to show . . . ” we mean that if we can show . . . then the
result we want will follow.
18 1. THE REAL NUMBERS

true.
In other words, we want to show that if cd = 0 then either c = 0 or d = 0
(possibly both).
The argument is written out as follows:
Claim: If c 6= 0 and d 6= 0 then cd 6= 0
We will establish the claim by proving that if cd = 0 then c = 0 or
d = 0.8
There are two possibilities concerning c;
either c = 0, in which case we are done
or c 6= 0. But in this case, since cd = 0, it follows
c−1 (cd) = c−1 0 and so
d=0
why?; fill in the steps.

Thus we have shown that if cd = 0 then c = 0 or d = 0. Equivalently, if

c 6= 0 and d 6= 0, then cd 6= 0. This completes the proof of the claim.

10. Exercise
HINT: We want to prove
ac−1 + bd−1 = (ad + bc)(cd)−1 .
First prove that
(ac−1 + bd−1 )(cd) = ad + bc.
Then deduce the required result.


1.2.3. Order consequences. All the standard properties of inequali-

ties for the real numbers follow from Axioms 1–13.
1.2.3.1. More definitions. One defines “>”, “≤” and “≥” in terms of <
as follows:
a > b if b < a,
a ≤ b if (a < b or a = b),
a ≥ b if (a > b or a = b).
(Note that the statement 1 ≤ 2, although it is not one we are likely to
make, is indeed√true, why?)
We define b, for b ≥ 0, to be that√number a ≥ 0 such that a2 = b.
Similarly, if n is a natural number, then n b is that number a ≥ 0 such that
an = b. To prove there is always such a number a requires the “completeness
axiom” (see later). (To prove that there is a unique such number a requires
the order axioms.)
If 0 < a we say a is positive and if a < 0 we say a is negative.

8Note; in mathematics, if we say P or Q (is true) then we always include the possibility
that both P and Q are true.
 1.2. ALGEBRAIC AND ORDER PROPERTIES 19

1.2.3.2. Some properties of inequalities. The following are consequences

of the axioms, although we will not stop to prove them.
Theorem 1.2.3. If a, b and c are real numbers then
(1) a < b and c < 0 implies ac > bc
(2) 0 < 1 and −1 < 0
(3) a > 0 implies 1/a > 0
(4) 0 < a < b implies 0 < 1/b < 1/a
(5) |a + b| ≤ |a| + |b| (triangle inequality)
(6) | |a| − |b| | ≤ |a − b| (a consequence of the triangle inequality)

1.2.4. Our approach henceforth. From now on, unless specifically

noted or asked otherwise, we will use all the standard algebraic and order
properties of the real numbers that you have used before. In particular, we
will use any of the definitions and results in Adams Section P1.

1.2.5. Natural and Rational numbers. The set N of natural num-

bers is defined by
N = {1, 2, 3, . . . }.
Here 2 = 1+1, 3 = 2+1, . . . . (Thus N is described by listing its members.)
The set Z of integers is defined by
Z = {. . . , −3, −2, −1, 0, 1, 2, 3, . . . }.
The set Q of rational numbers is defined by
Q = { m/n | m, n ∈ Z, n 6= 0 }.
(We read the right side after the equality as “the set of m/n such that m, n
are members of Z and n 6= 0.)
A real number is irrational if it is not rational. It can be proved that π
and e are irrational, see Calculus by M. Spivak.

The set N is not a model of Axiom 3, as 0 is not a member. The set Z is

a model of all of Axioms 1–13, except for Axiom 8, since the multiplicative
inverse a−1 of an integer is not usually an integer.
The set Q is a model of all of Axioms 1–13 but not of the Completeness
Axiom (see later).

1.2.6. FFields.
The real numbers and the rationals, as well as the integers
modulo a fixed prime number, form a field.
Any set S, together with two operations ⊕ and ⊗ and two members 0⊕
and 1⊗ of S, which satisfies the corresponding versions of Axioms 1–9, is
called a field.
Thus R (together with the operations of addition and multiplication and
the two real numbers 0 and 1) is a field. The same is true for Q, but not for
Z since Axiom 8 does not hold, why?.
An interesting example of a field is the set
Fp = {0, 1, . . . , p − 1}
20 1. THE REAL NUMBERS

for any fixed prime p, together with addition and multiplication defined
“modulo p”; i.e. one performs the usual operations of addition and multi-
plication, but then takes the “remainder” after dividing by p.
Thus for p = 5 one has:
⊕ 0 1 2 3 4 ⊗ 0 1 2 3 4
0 0 1 2 3 4 0 0 0 0 0 0
1 1 2 3 4 0 1 0 1 2 3 4
2 2 3 4 0 1 2 0 2 4 1 3
3 3 4 0 1 2 3 0 3 1 4 2
4 4 0 1 2 3 4 0 4 3 2 1
It is not too hard to convince yourself that the analogues of Axioms 1–9
hold for any prime p. The axiom which fails if p is not prime is Axiom
8, why?. Note that since Fp is a field, we can solve simultaneous linear
equations in Fp .
The fields Fp are very important in coding theory and cryptography.

1.2.7. FGroups. Any set S, together with an operation ⊗ and a par-

ticular member e ∈ S, which satisfies:
for all a, b, c ∈ S:
(1) (a ⊗ b) ⊗ c = a ⊗ (b ⊗ c) (associative axiom)
(2) a ⊗ e = e × a = a (identity axiom)
(3) there is a member of S, denoted a−1 , such that a⊗a−1 = a−1 ⊗a = e
(inverse axiom)
Examples are the reals or rationals, with ⊗ replaced by × and e replaced
by 1, or with ⊗ replaced by + and e replaced by 0. Another example is Z
with ⊗ and e replaced by + and 0.
The notion of a group pervades much of mathematics and its applica-
tions. Important examples are groups of transformations and their applica-
tion to classification of crystals.
As a simple case, consider a square.

Figure 4. Reflection in any of the axes D, D0 , V, H maps

the shaded square onto itself.
 1.3. COMPLETENESS AXIOM 21

If the square is reflected in any of the four axes V (vertical axis), H

(horizontal axis), D or D0 (two diagonals), then the image coincides with
the original square. We denote these four reflection transformations by
V, H, D, D0 respectively.
We could also rotate the square through 900 , 1800 or 2700 ; these oper-
ations are denoted by R, R0 and R00 respectively. Finally, we could rotate
through 3600 , which has the same effect as doing nothing; this operation is
called the identity operation and is denoted by I.
We say the square is invariant under the 8 transformations I, V , H, D,
D0 , R, R0 and R00 .
If we first apply R and then apply H, the result is written as H ⊗ R
(read from right to left, just as for the composition of functions in general).
This is not a new operation — it has the same effect as applying D0 . One
way to see this is to check what happens to each of the four vertices of the
square. For example, if we apply H ⊗ R the top right vertex is first mapped
to the top left vertex and then to the bottom right vertex. This is the same
as applying D0 . Similarly for the other three vertices.
However, if we apply H and R in the reverse order, i.e. R ⊗ H, then the
result is different. This time it is the same as D. In particular, the operation
⊗ is not commutative.
If we draw up a table we obtain
⊗ I V H D D0 R R0 R00
I I V H D D0 R R0 R00
V V I R0 R R00 D H D0
H H R0 I R00 R D0 V D
D D R00 R I R H D0 V
0 0
D D R R R 00 0 I V D H
R R D D V H R0 R00 I
0

R0 R0 H V D0 D R00 I R
R00 R00 D D0 H V I R R0

A crystal is classified by the group of transformations which leaves it

invariant. These ideas are treated in your later Algebra courses.
Other important examples of groups are certain sets of 2 × 2 matrices,
where
 multiplication is matrix multiplication and the identity is the matrix
1 0
. In fact this is related to the previous example, the 8 operations
0 1
correspond to certain 2×2 matrices (Can you find them? See Linear Algebra
by Lay.). More generally, important examples are various groups of n × n
matrices.
See “A Survey of Modern Algebra” by Birkhofff and MacLane for an
introduction to fields and groups. (An old classic.)

1.3. Completeness Axiom

The Completeness Axiom is introduced. It is true for the
real numbers, but the analogous axiom is not true for the
rationals. We define the notion of upper bound (lower bound)
22 1. THE REAL NUMBERS

and least upper bound (greatest lower bound) of a set of real

numbers.
See Adams page 4 for a few brief remarks, page A22 in the Appendices,
and an application to sequences on page A23.
The Completeness Axiom is the final axiom for the real number system,
and is probably not one you have met before. It is more difficult to under-
stand than the other properties, but it is essential in proving many of the
important results in calculus.
1.3.1. Statement of the Axiom.
Axiom 14 (Completeness Axiom): If A is any non-empty set of real
numbers with the property that there is some real number x such that a ≤ x
for every a ∈ A, then there is a smallest (or least) real number x with this
same property.
A is non-empty means that A contains at least one number.
Note that the number x in the axiom need not belong to A. For example,
if A is the interval [0, 1) then the smallest (or “least”) number x as above is
1, but 1 6∈ A. On the other hand, if A = [0, 1] then the smallest number x
as above is again 1, but now 1 ∈ A.
There is some useful notation associated with the Completeness Axiom.
Definition 1.3.1. If A is a set of real numbers and x is a real number
such that a ≤ x for every a ∈ A, then x is called an upper bound for A. If
in addition x ≤ b for every upper bound b, then x is called the least upper
bound or supremum of A. In this case one writes
x = lub A or x = sup A.
If x ≤ a for every a ∈ A, then x is called a lower bound for A. If also
x ≥ c for every lower bound c, then x is called the greatest lower bound or
infimum of A. In this case one write
x = glb A or x = inf A.

Thus we have:
Axiom 14 (Completeness Axiom): If a non-empty set A has an upper
bound then it has a least upper bound.
(Remember that when we say A “has” an upper bound or a least upper
bound x we do not require that x ∈ A.)
See Adams page A22, Example 1.
1.3.2. An equivalent formulation. There is an equivalent form of
the axiom, which says: If A is any non-empty set of real numbers with the
property that there is some real number x such that x ≤ a for every a ∈ A,
then there is a largest real number x with this same property. In other words
if a non-empty set A has a lower bound then it has a greatest lower bound.
It is not too hard to see that this form does indeed follow from the
Completeness Axiom. The trick is to consider, instead of A, the set
A∗ := { −x : x ∈ A },
which is obtained by “reflecting” A about 0.
 1.3. COMPLETENESS AXIOM 23

Figure 5. Reflecting A through the origin sends A to A∗ ,

lower bounds of A to upper bounds of A∗ and the g.l.b. of A
to the l.u.b. of A∗ .

Lower bounds for A correspond under reflection to upper bounds for A∗ ,

and a glb corresponds to a lub. If A is bounded below then A∗ is bounded
above, and so by the Completeness Axiom has a lub. After reflection, this
lub for A∗ gives a glb for A. (To actually write this out carefully needs some
care—you need to check from the relevant definitions and the properties
of inequalities that the first three sentences in this paragraph are indeed
correct.)
(Similarly, the Completeness Axiom follows from the above version.)
Unlike in the case of Axioms 1–13, we will always indicate when we use
the Completeness Axiom.

1.3.3. Interpretation of the Completeness Axiom. The Complete-

ness Axiom implies there are no “gaps” in the real numbers.
For example, the rational numbers are not a model of the corresponding
version of Axiom 14. In other words, Axiom 14 is not true if, in the first
statement of the axiom, the three occurence of the word “real” are replaced
by “rational”.
For example, let
√
A = { a ∈ Q | 0 ≤ a and a2 < 2 } = { a ∈ Q | 0 ≤ a < 2 }.
(The first definition for A has the advantage that A is defined
√ without
actually referring to the existence of the irrational number 2.) There are
certainly rational numbers x which are upper bounds for A, i.e. such that
a ≤ x for every a ∈ A, just take x = 23. But we claim there is no rational
number b which is a least upper bound for A.

Figure 6. In each of these diagrams the set A is represented

by the thin line. In the left diagram we see b is not an upper
bound for A as there are members of A greater than b. In
the right diagram we see b is not the l.u.b. for A as there are
other upper bounds that are less than b.

√
Proof of claim. Since 2 is not rational it cannot be the required
rational number b.
24 1. THE REAL NUMBERS
√
On the other√ hand, if b < 2, since there is always a rational
√ number
between b and 2 this gives a member of A between b and 2, and so b
cannot be an upper√bound.
√ Finally, if b > 2, there is always a smaller rational number between
2 and b, and so b cannot be the least rational number which is an upper
bound for A. √ √ √
We have ruled out the three possibilities b = 2, b < 2 and b > 2.
This completes the proof of the claim. Hence there is no rational number
which is a least upper bound for A. 

1.3.4. Archimedean Property. The following property of the real

numbers is not surprising, but it does not follow from the algebraic and
order axioms alone. It says, informally, that there are no real numbers
beyond all the natural numbers.
Theorem 1.3.2 (Archimedean Property). For every real number x there
is a natural number n such that x < n. Equivalently, the set N is not bounded
above.

F Proof. Suppose that the theorem were false. Then there would be
a real number x with the property that n < x for all n ∈ N. This implies N
is bounded above and so there must be a least upper bound b (say) by the
Completeness Axiom.
In other words,
n ≤ b for every n ∈ N.
It follows that
n + 1 ≤ b for every n ∈ N,
since n + 1 ∈ N if n ∈ N. But this implies
n ≤ b − 1 for every n ∈ N.
In other words, b − 1 is also an upper bound for N, which contradicts the
fact that b is the least upper bound.
Since we have obtained a contradiction by assuming the statment of the
theorem is false, the statement must in fact be true. 
The only surprising thing about the Archimedean property is that it
needs the Completeness Axiom to prove it. But there are in fact models of
the algebraic and order axioms in which the Archimedean property is false.
They are sometimes called the Hyperreals! See the next starred section.
The following corollary says that between zero and any positive number
(no matter how small) there is always a number of the form 1/n, where
n ∈ N. This is the same type of result as in Section 1.1.4, which stated that
between any two different real numbers there is always a rational number.
But in Section 1.1.4 we were not very careful, and did not go back to the
axioms to prove the result (as we actually do need to!).
The symbol ε in the following is called “epsilon” and is a letter of the
Greek alphabet. You could replace ε by any other symbol such as x or a,
and the Corollary would have exactly the same meaning. However, it is
traditional in mathematics to use ε when we are thinking of a very small
 1.3. COMPLETENESS AXIOM 25

positive number. Sometimes we use the symbol δ or “delta”, another letter

of the Greek alphabet, in the same way.

Corollary 1.3.3. For any real number ε > 0 there is a natural number
n such that n1 < ε.

Proof. By the Archimedean Property there is a natural number n such

that n > 1ε . But then n1 < ε (by standard properties for manipulating
inequalities). 
Note that this corollary was actually used in the proof of the results in
Section 1.1.4, where?
It is probably confusing as to why the Archimedean property and its
Corollary should rely on the Completeness Axiom for their proofs. What is
the difference between the Archimedean Property and the previous Corollary
on the one hand, and the other standard properties of inequalities such as
in Theorem 1.2.3?
Well, the main difference is that the usual properties of inequalities, such
as in Theorem 1.2.3, essentially tell us how to manipulate inequalities. The
Archimedean Property is essentially a non-existence property concerning
the infinite set N of real numbers, namely that N has no upper bound.
Don’t worry! There will not be any more surprises like this. There will
be important situations where we rely on the Completeness Axiom, such as
in proving certain properties of continuous functions, but these applications
will not be so surprising.
1.3.5. FFHyperreals. Part of any model of the hyperreals looks like
a “fattened up” copy of R, in the sense that it contains a copy of R together
with “infinitesimals” squeezed between each real a and all reals greater than
a. (In particular there are hyperreals bigger than 0 and less than any positive
real number!) This part is followed and preceded by infinitely many “copies”
of itself, and between any two copies there are infinitely many other copies.
See the following crude diagram.
In particular there are hyperreals bigger than any natural number, as
the natural numbers will all lie on the fattened up copy of R.
The hyperreals were discovered by Abraham Robinson in the 1960’s.
They turn out to be quite useful in proving results about the usual real
numbers and have been used in many areas of mathematics (for example
in probability theory and stochastic processes). They are one way of giving
a rigorous meaning to the notion of an infinitesimal — i.e. it is possible to
dy
interpret the expression dx as the ratio of two hyperreal numbers dy and dx.
There are even courses on Calculus that are based on using the hyper-
reals (also called “nonstandard” numbers). See “Elementary Calculus” by
H. Jerome Keisler, and the corresponding Instructors Manual “Foundations
of Infinitesimal Calculus”. However, this approach to teaching Calculus has
not been particularly popular!
A full analysis of the hyperreals and their properties requires a study of
the underlying formal logic and set theory used in mathematics.
26 1. THE REAL NUMBERS

Figure 7. The line containing 0 and 1 is a “fattened up”

copy of R. It contains the infinitesimals, and for all each
numbers a it contains the hyperreals a +  where  is an
arbitrary infinitesimal. All points on each line are less than
all points on any higher line.

You will not use the hyperreals. You certainly should not refer to them
in any of your proofs.

1.4. Sets
The notion of a set is basic in mathematics. We will not need to study
the theory of sets, but we will need to know some notation and a few basic
properties.
1.4.1. Notation for sets. By a set (sometimes called a class or family)
we mean a collection, often infinite, of (usually mathematical) objects.9
Members of a set are often called elements of the set. If a is a member
(i.e. element) of the set S, we write
a ∈ S.
If a is not a member of S we write
a 6∈ S.

If a set is finite, we may describe it by listing its members. For example,

A = { 1, 2, 3 }.
Note that {1, 2, 3}, {2, 3, 1}, {1, 1, 1, 2, 3} are different descriptions of exactly
the same set. Some infinite sets can also be described by listing their mem-
bers, provided the pattern is clear. For example, the set of even positive
integers is
E = { 2, 4, 6, 8, . . . }.
We often use the notation
S = { x | P (x) },
9F There is a mathematical theory of sets, and in fact all of mathematics can be
formulated within the theory of sets. However, this is normally only useful or practical
when considering fundamental questions about the foundations of mathematics.
 1.4. SETS 27

where P (x) is some statement involving x. We read this as “S is the set

of all x such that P (x) is true”. It is often understood from the context of
the discussion that x is restricted to be a real number. But if there is any
possible ambiguity, then we write
S = { x ∈ R | P (x) },
which we read as “S is the set of elements x in R such that P (x) is true”.
Note that this is exactly the same set as
{ y | P (y) } or equivalently { y ∈ R | P (y) }.
The variables x and y are sometimes called “dummy” variables, they are
meant to represent any real number with the specified properties.
One also sometimes uses “ : ” instead of “ | ” when describing sets.
The union of two or more sets is the set of numbers belonging to at least
one of the sets. The intersection of two or more sets is the set of numbers
belonging to all of the sets. We use ∪ for union and ∩ for intersection. Thus
if A and B are sets, then
A ∩ B = { x | x ∈ A and x ∈ B },
A ∪ B = { x | x ∈ A or x ∈ B }.

Figure 8. A Venn diagram representing the sets A, B, A∩B

an A ∪ B.
The set A is a subset of B, and we write A ⊆ B, if every element of A
is also an element of B. In symbols
A⊆B iff x ∈ A ⇒ x ∈ B.
(Where ⇒ is shorthand for “implies”.)
For example
N ⊆ Z ⊆ Q ⊆ R.
It is also true that N ⊆ N, etc., although we would not normally make this
statement.
In fact it is very common to write A ⊂ B instead of A ⊆ B. The most
common convention nowadays is that A ⊂ A is a true statement. If you
want to indicate that every element in A is also in B, and that there is at
least one element in B that is not in A, then you should write A ( B.
Two sets are equal iff they have the same elements. It follows that
A=B iff A ⊆ B and B ⊆ A.
28 1. THE REAL NUMBERS

In particular, we frequently prove two sets A and B are equal by first proving
that every member of A is a member of B (i.e. A ⊆ B) and then proving
that every member of B is a member of A (i.e. B ⊆ A).
For example, using the standard notation for intervals of real numbers
in Section ??,
{ x | 0 < x < 2 and 1 ≤ x ≤ 3 } = (0, 2) ∩ [1, 3] = [1, 2),
{ x | 0 < x < 1 or 2 < x ≤ 3 } = (0, 1) ∪ (2, 3].

Also
(0, 2) = (0, 1) ∪ [1, 2) = (0, 1] ∪ [1, 2) = (0, 1) ∪ ( 21 , 2),
etc.
1.4.2. Ordered pairs of  real numbers. In the Linear Algebra course
a
you use both the notation and (a, b) to represent vectors in R2 , which we
b
also regard as ordered pairs, or 2-tuples, of real numbers. Of course, (a, b)
and (b, a) are distinct, unless a = b. This is different from the situation for
the set containing a and b; i.e. {a, b} and {b, a} are just different ways of
describing the same set.
We also have ordered triples (a, b, c), and more generally ordered n-tuples
(a1 , . . . , an ), of real numbers.
Remark 1.4.1. F It is sometimes useful to know that we can define
ordered pairs in terms of sets. The only property we require of ordered pairs
is that
(3) (a, b) = (c, d) iff (a = c and b = d).
We could not define (a, b) = {a, b}, because we would not be able to
distinguish between (a, b) and (b, a). But there are a number of ways that
we can define ordered pairs in terms of sets. The standard definition is
(a, b) := {{a}, {a, b}}.
To show this is a good definition, we need to prove (3).
Proof. It is immediate from the definition that if a = c and b = d then
(a, b) = (c, d).
Next suppose (a, b) = (c, d), i.e. {{a}, {a, b}} = {{c}, {c, d}}. We con-
sider the two cases a = b and a 6= b separately.
If a = b then {{a}, {a, b}} contains exactly one member, namely {a},
and so {{c}, {c, d}} also contains exactly the one member {a}. This means
{a} = {c} = {c, d}. Hence a = c and c = d. In conclusion, a = b = c = d.
If a 6= b then {{a}, {a, b}} contains exactly two (distinct) members,
namely {a} and {a, b}. Since {{a}, {a, b}} = {{c}, {c, d}} it follows {c} ∈
{{a}, {a, b}} and so {c} = {a} or {c} = {a, b}. The second equality cannot
be true since {a, b} contains two members whereas {c} contains one member,
and so {c} = {a}, and so c = a.
Since also {c, d} ∈ {{a}, {a, b}} it now follows that {c, d} = {a, b} (oth-
erwise {c, d} = {a}, but since also {c} = {a} this would imply {{c}, {c, d}}
and hence {{a}, {a, b}} has only one member, and we have seen this is not
 1.4. SETS 29

so). Since a and b are distinct and {c, d} = {a, b}, it follows c and d are
distinct; since a = c it then follows b = d. In conclusion, a = c and b = d.
This completes the proof. 

Remark 1.4.2. In Adams page 24 there is the definition: a function f

on a set D into a set S is a rule that assigns a unique element f (x) in S to
each element x in D.
This is indeed the way to think of a function. But there is a problem.
What do we mean by a “rule”? Is it something that can be ultimately written
down by some algorithmic procedure, or in some sense can be programmed?
This underestimates the collection of all possible functions.10
F If we take the notion of a set as basic then we can define the notion
of a function as follows: A function f on a set D into a set S is a set f of
ordered pairs of the form (x, y) satisfying the conditions
(1) if (x, y) ∈ f then x ∈ D and y ∈ S;
(2) for each x ∈ D there is exactly one y ∈ S such that (x, y) ∈ f —
this y is denoted by f (x).
This definition is useful because it shows that the notion of a function
can be reduced to that of a set. In fact all of mathematics can be formulated
within set theory. This is fundamental for the study of the foundations of
mathematics, which is done later in a third year honours level course.

10F In a manner that can be made precise, and will be in later courses, it is possible to
define a notion of countably infinite and uncountably infinite. The integers, the rationals,
and the set of possible rules, are all countably infinite. The reals, the irrationals and the
set [0, 1], are uncountably infinite. The set of functions from R into R, or even from [0, 1]
into [0, 1], are both uncountably infinite. In fact, these sets of functions can be shown to
be, in a precise manner, a larger infinity than the infinity of real numbers.
 CHAPTER 2

Sequences

The reference here is Adams Section 9.1, and the material on pages A-22
and A-23, but we do considerably more. Another reference is Calculus by
M. Spivak.

2.1. Examples and Notation

A sequence is an infinite list of numbers with a first, but no last, element.
Simple examples are
1, 2, 1, 3, 1, 4, . . .
1 1
1, , , . . .
2 3
1, −1, 1, −1, 1, . . .
A sequence can be written in the form
a1 , a2 , a3 , . . . , an , . . . .
More precisely, a sequence is a function f whose domain is the set of natural
numbers, where in the above example f (n) = an . We often just write (an )
or (an )n≥1 to represent the sequence.
See Adams, page 496, Example 1.
See Adams pages 496 and 497 for the definitions of the following terms:
(1) bounded below, lower bound; bounded above, upper bound; bounded;
(2) positive, negative;
(3) increasing, decreasing, monotonic;
(4) alternating;
(5) ultimately (or “eventually”).
See Examples 2 and 3 in Adams page 497.

2.2. Convergence of Sequences

The most fundamental concept in the study of sequences is the notion
of convergence of a sequence.
The informal idea is that a sequence (an ) converges to a, and we write
lim an = a,
if no matter how small a positive number is chosen, the distance between an
and a, i.e. |an − a|, will ultimately be less than this positive number. (The
smaller the positive number, the further out in the sequence we will need to
go.)
31
32 2. SEQUENCES

It is important to note that this is a condition that must be satisfied by

any positive number. For example, we may have
|an − a| < .1 if n > 50, (here the positive number is .1)
|an − a| < .01 if n > 300, (here the positive number is .01)
|an − a| < .001 if n > 780, (here the positive number is .001)
etc.
Definition 2.2.1. We say that the sequence (an ) converges to a limit a,
and write
lim an = a, lim an = a, an → a, or lim an = a,
n n→∞
if for every positive number ε there exists an integer N such that
(4) n>N implies |an − a| < ε.

See Adams page 498, Figure 9.1, for a graphical illustration of limit.
Example 2.2.2. Show that the sequence given by an = 1 + n12 converges
to 1 according to the definition.
Solution. Let ε > 0 be given.
We want to find N such that (4) is true with a = 1.
We have
1
|an − 1| = 2 .
n
Since
1 1
2
< ε if n2 > ,
n ε
i.e.
1
if n > √ ,
ε
we can take  
1
N= √ ,
ε
or any larger integer, where [ ] denotes “the integer part of”. 
√
Thus if ε = .1 we can take any integer N > 1/ .1, for example N = 4
(or anything larger). If ε = .01 we can take N = 10 (or anything larger).
If ε = .001 we can take N = 32 (or anything larger). But the above proof
works of course for any ε > 0.1

h i h i
1* In the example we took N = 1
√ , the integer part of √1ε , or equivalently the
ε
1
smallest integer greater than √
ε
− 1.
The general statement
∀z > 0 ∃N ∈ N (N > z),
is just the Archimedean Property, which follows from the Completeness Axiom as we saw
before. Thus we are usually using the Archimedean Property when we prove the existence
of limits.
A similar remark applies to the following examples, but we will not usually explicitly
state this fact.
 2.2. CONVERGENCE OF SEQUENCES 33

Example 2.2.3. Consider the sequence defined by a1 = 1, and an+1 =

1
2 an + 2 for n ≥ 1.
It is easy to calculate that the first few terms are
1, 2.5, 3.25, 3.625, 3.8125, 3.90625, 3.953125, 3.9765625, . . . .
It seems reasonable that the sequence is converging to 4. One way to prove
this is as follows.
Proof. Let ε > 0 be given.
We want to find N such that2
(5) n>N ⇒ |an − 4| < ε.
We have a formula for an+1 in terms of an , and we first use this to get
a formula for |an+1 − 4| in terms of |an − 4|. Thus
1 1 1 1
|an+1 − 4| = an + 2 − 4 = an − 2 = (an − 4) = |an − 4|.
2 2 2 2
Thus |a1 − 4| = 3, |a2 − 4| = 3/2, |a3 − 4| = 3/22 , |a4 − 4| = 3/23 , . . . . In
general3 |an − 4| = 3/2n−1 .
It follows that
3
|an − 4| < ε for those n such that n−1
< ε.
2
This last inequality is equivalent to 2n−1 /3 < 1/ε, i.e. 2n−1 > 3/ε, i.e.
(n − 1) ln 2 > ln(3/ε), i.e. n > 1 + ln(3/ε)/ ln 2.4
Hence (5) is true for " #
ln 3ε
N =1+ .
ln 2

You may object that we used ln, the natural logarithm, in the previous
example, but we have not yet shown how to define logarithms and establish
their properties from the axioms. This is a valid criticism. But in order to
have interesting examples, we will often do this sort of thing.
However, we will not do it when we are establishing the underlying
theory. In particular, the development of the theory will not depend on the
examples.
See Adams page 498 Example 4.
Definition 2.2.4. If a sequence (an ) does not converge, then we say
that it diverges.
We say that the sequence (an ) diverges to +∞ (or to ∞),5 and write
lim an = ∞, lim an = ∞, an → ∞, or lim an = ∞,
n n→∞

2We will often write “⇒” for “implies”.

3This could easily be proved by induction, but it is not necessary to do so.
4F As in the previous example, to prove the existence of such an n requires, strictly
speaking, the Archimedean Property, which is a consequence of the Completeness Axiom.
5Note that +∞ is not a real number. In fact we do not here define an object denoted
by “∞”. We just define a certain concept which we denote by “lim an = ∞” or by
something similar.
34 2. SEQUENCES

if for each real number M there exists an integer N such that

n>N implies an > M.

(The interesting situation is M large and positive.)

Similarly, we say (an ) to diverges to −∞ if for each real number M there
exists an integer N such that
n>N implies an < M.
(The interesting situation is M large and negative.)
A sequence may diverge, without diverging to ±∞. See Adams page 498
Example 5.

2.3. Properties of Sequence Limits

It is normally not very efficient to use the definition of a limit in order
to prove that a sequence converges. Instead, we prove a number of theorems
which will make things much easier.
The first theorem shows that if two sequences converge, then so does
their sum, and moreover the limit of the new sequence is just the sum of
the limits of the original sequences. Similar results are true for products
and quotients, and if we multiply all terms in a sequence by the same real
number.
In (9) we need to assume b 6= 0. This will imply that ultimately bn 6= 0
(i.e. bn 6= 0 for all sufficiently large n), and hence that the sequence (an /bn )
is defined for all sufficiently large n.
Theorem 2.3.1. Suppose
lim an = a, lim bn = b,
and c is a real number. Then the following limits exist and have the given
values.
(6) lim(an ± bn ) = a ± b,
(7) lim can = ca,
(8) lim an bn = ab,
an a
(9) lim = , assuming b 6= 0.
bn b

We will give the proofs in the next section. The theorem is more justi-
fication that Definition 2.2.1 does indeed capture the informal notion of a
limit.
The results are not very surprising. For example, if an is getting close
to a and bn is getting close to b then we expect that an + bn is getting close
to a + b.
 2
Example 2.3.2. Let an = 1 + √1n − (1 + 2−n ).
We can prove directly from the definition of convergence that √1 → 0
n
and 2−n → 0. It then follows from the previous theorem that 1 + √1n →
1 (since we can think of 1 + √1 as obtained by adding the term 1 from
n
 2.4. PROOFS OF LIMIT PROPERTIES 35

the constant sequence (1) to the term √1n ). Applying the theorem again,
 2
1 + √1n → 1. Similarly, 1 + 2−n → 1.
Hence (again from the theorem) an → 1 − 1 = 0.

−12
Example 2.3.3. Let an = 3n2n
2 −7n+1 .
Write
2n2 − 1 2 − n12
= .
3n2 − 7n + 1 3 − n7 + n12
Since the numerator and denominator converge to 2 and 3 respectively, it
follows an → 2/3.
See also Adams, pages 499, Example 6.
Before we prove Theorem 2.3.1 there is a technical point. We should
prove that a convergent sequence cannot have two different limits. This is
an easy consequence of the definition of convergence.
This is done in the next section, but try it yourself first of all.
Theorem 2.3.4. If (an ) is a convergent sequence such that an → a and
an → b then a = b.
The next easy result is useful in a number of situations.
Theorem 2.3.5. Suppose an → a. Then the sequence is bounded; i.e.
there is a real number M such that |an | ≤ M for all n.
The next theorem is not true if we replace both occurrences of “≤” by
“<”. For example −1/n < 1/n for all n, but the sequences (1/n) and (−1/n)
have the same limit 0.
Theorem 2.3.6. Suppose an → a, bn → b, and an ≤ bn ultimately.
Then a ≤ b.
The following theorem says that if a sequence is “squeezed” between two
sequences which both converge to the same limit, then the original sequence
also converges, and it converges to the same limit.
Theorem 2.3.7 (Squeeze Theorem). Suppose an ≤ bn ≤ cn ultimately.
Suppose an → L and cn → L. Then bn → L.

Example 2.3.8. Consider the sequence 3 + (sin cos n)/n. Since −1 ≤

sin x ≤ 1, it follows that 3 − 1/n ≤ 3 + (sin cos n)/n ≤ 3 + 1/n. But
3 − 1/n → 3 and 3 + 1/n → 3. Hence 3 + (sin cos n)/n → 3.

2.4. Proofs of limit properties

I have starred some of these proofs, as they are a bit technical. But you
should aim to have some understanding of the ideas involved.
36 2. SEQUENCES

Figure 1.

F Proof of Theorem 2.3.4. Suppose an → a and an → b.

Assume (in order to obtain a contradiction) that a 6= b.
Take ε = |a − b|/3 in the definition of a limit, Definition 2.2.1. (For
motivation, look at the following diagram).
Since an → a, it follows that
(10) an ∈ (a − ε, a + ε)
for all sufficiently large n, say for n > N1 .
Since an → b, it follows that
(11) an ∈ (b − ε, b + ε)
for all sufficiently large n, say for n > N2 .
But this implies
an ∈ (a − ε, a + ε) and an ∈ (b − ε, b + ε)
for all n > max{N1 , N2 }.
However this is impossible since ε = |a−b|/3, as we can see geometrically
or show algebraically.
Thus the assumption a 6= b leads to a contradiction, and so a = b. 

F Proof of Theorem 2.3.5. Assume an → a.

From the definition of convergence, taking ε = 1, there is an integer N
such that
(12) a − 1 < an < a + 1 for all n > N.
Fix this N . Since the set of terms
a1 , a2 , . . . , aN
is finite, it follows that there exist real numbers M1 and M2 such that
(13) M1 ≤ an ≤ M2 for all n ≤ N.
(Just take M1 = min{a1 , a2 , . . . , aN } and M2 = max{a1 , a2 , . . . , aN }.)
From (12) and (13),
M1∗ ≤ an ≤ M2∗ for all n,
where M1∗ = min{a − 1, M1 }, M2∗ = max{a + 1, M2 }.
Hence |an | ≤ M for all n where M = max{|M1∗ |, |M2∗ |}. 
 2.4. PROOFS OF LIMIT PROPERTIES 37

Proof of Theorem 2.3.1 (6) (“+” case). Suppose an → a and bn →

b.
Let ε > 0 be given.
Since an → a there exists an integer N1 (by Definition 2.2.1) such that
(14) n > N1 implies |an − a| < ε/2.
Since bn → b there exists an integer N2 (again by Definition 2.2.1) such
that
(15) n > N2 implies |bn − b| < ε/2.
It follows that if n > max{N1 , N2 } then
|(an + bn ) − (a + b)| = |(an − a) + (bn − b)|
< |an − a| + |bn − b| by the triangle inequality
ε ε
< + by (14) and (15)
2 2
= ε.
It follows from Definition 2.2.1, with N = max{N1 , N2 }, that (an + bn )
converges and the limit is a + b. 

Notice in the proof how the definition of a limit is used three times; once
to get information from the fact an → a, once to get information from the
fact bn → b, and finally to deduce that an + bn → a + b.
By the way, why do we use ε/2 in (14) and (15), and why is this justifiable
by Definition 2.2.1?

Proof of Theorem 2.3.1 (7). Suppose an → a and c is a real num-

ber.
Let ε > 0 be any positive number. We want to show
|can − ca| < ε
for all sufficiently large n.
Since an → a there exists an integer N such that
|an − a| < ε/|c| for all n > N.
(This assumes c 6= 0. But if c = 0, then the sequence (can ) is the sequence
all of whose terms are 0, and this sequence certainly converges to ca = 0.)
Multiplying both sides of the inequality by |c| we see
|c| |an − a| < ε for all n > N,
i.e.
|can − ca| < ε for all n > N,
and so can → ca by the definition of convergence. 
38 2. SEQUENCES

Proof of Theorem 2.3.1 (6) (“−” case). Suppose an → a and bn →

b.
We can write
an − bn = an + (−1)bn .
But (−1)bn → (−1)b by the previous result with c = −1, and so the result
now follows from (6) for the sum of two sequences. 

F Proof of Theorem 2.3.1 (8). Suppose an → a and bn → b.

As usual, let ε > 0 be an arbitrary positive number.
We want to show there is an integer N such that
|an bn − ab| < ε
for all n > N .
To see how to choose N , write
|an bn − ab| = |an bn − an b + an b − ab|
= |an (bn − b) + b(an − a)|
(16)
≤ |an (bn − b)| + |b(an − a)|
= |an | |bn − b| + |b| |an − a|.
(This trick of adding and subtracting the same term, here it is an b, is often
very useful.) We will show that both terms are < ε/2 for all sufficiently
large n.
For the second term |b| |an − a|, the result is certainly true if b = 0, since
the term is then 0. If b 6= 0, since an → a, we can choose N1 such that
ε
|an − a| < for all n > N1 ,
2|b|
and so
ε
(17) |b| |an − a| < for all n > N1 .
2
For the first term |an | |bn − b|, we use Theorem 2.3.5 to deduce for some
M that |an | ≤ M for all n. By the same argument as for the second term,
we can choose N2 such that
ε
M |bn − b| < for all n > N2 ,
2
and so
ε
(18) |an | |bn − b| < for all n > N2 .
2
Putting (16), (17) and (18) together, it follows that if n > N , where
N = max{N1 , N2 }, then
ε ε
|an bn − ab| < + = ε.
2 2
Since ε > 0 was arbitrary, this proves an bn → ab. 
 2.4. PROOFS OF LIMIT PROPERTIES 39

F Proof of Theorem 2.3.1 (9). Suppose an → a and bn → b where

b 6= 0.
We will prove that an /bn → a/b by first showing 1/bn → 1/b and then
using the previous result about products of sequences.
We first prove
(19) |bn | > |b|/2 ultimately.
The proof is similar to that in Theorem 2.3.5, and goes as follows:
First assume b > 0. Choose ε = |b|/2 (> 0) in the definition of conver-
gence and deduce that for some integer N ,
n > N ⇒ |bn − b| < b/2,
and so in particular
n > N ⇒ |bn | > b/2.
This proves (19) in case b > 0.
In case b < 0 we similarly prove that ultimately bn < b/2, and so ulti-
mately |bn | > |b|/2. Ths completes the proof of (19).
We now proceed with the proof that 1/bn → 1/b. For this let ε > 0 be
any positive number.
In order to see how to choose N in the definition of convergence, we
compute
1 1 |b − bn | 2|b − bn |
(20) − = ≤ ,
bn b |bn | |b| |b|2
ultimately (this uses (19)). (The only reason for “≤” instead of “<” is that
perhaps |b − bn | = 0.)
Since bn → b we can find an integer N such that for all n > N ,
|b|2
|b − bn | < ε.
2
It follows from (20) that if n > N then
1 1
− < ε.
bn b
Since ε > 0 was arbitrary, it follows that 1/bn → 1/b.
Since an → a, it now follows from the result for products that an /bn →
a/b. 

Proof of Theorem 2.3.6. (The proof is similar to that for Theo-

rem 2.3.4.)
Suppose an → a, bn → b, and ultimately an ≤ bn .
Assume (in order to obtain a contradiction) that a > b. Let ε = 13 (a−b).
Then
an ∈ (a − ε, a + ε) ultimately,
and in particular
an > a − ε ultimately.
Similarly
bn < b + ε ultimately.
40 2. SEQUENCES

Figure 2.

(Draw a diagram.) Since ε = 13 (a − b), this implies

an > bn ultimately.
But this contradicts an ≤ bn , and so the assumption is false. Thus a ≤ b. 

Proof of Theorem 2.3.7. Suppose an ≤ bn ≤ cn ultimately. Sup-

pose an → L and cn → L.
Let ε > 0 be given. (For motivation, look at the following diagram).
Since an → L there is some integer N1 such that
(21) n > N1 ⇒ an ∈ (L − ε, L + ε).
Since cn → L there is some integer N2 such that
(22) n > N2 ⇒ cn ∈ (L − ε, L + ε).
Let N = max{N1 , N2 }. Then since an ≤ bn ≤ cn it follows from (21)
and (22) that
n > N ⇒ bn ∈ (L − ε, L + ε).
But ε was an arbitrary positive number, and so it follows that bn →
L. 

2.5. More results on sequences

We have seen that a convergent sequence is bounded.
The converse is false. For example, the sequence
1, −1, 1, −1, . . .
is bounded but does not converge.
However, a bounded monotone sequence does converge. The proof needs
the Completeness Axiom in order to give a “candidate” for the limit. (See
also Adams, Appendix III page A-23, Theorem 2.)
Theorem 2.5.1. If a sequence is bounded and ultimately monotone (i.e.
either ultimately increasing or ultimately decreasing), then it converges.

Proof. We do the “ultimately decreasing case”, the other is similar.

Suppose the sequence decreases from the N th term onwards. Consider-
ing just those terms from this point, and changing notation, we may assume
a1 ≥ a2 ≥ a3 ≥ . . . an ≥ · · · .
This set of terms is bounded below, and so by the Completeness Axiom
this set has a glb which we denote by L. We will prove that lim an = L.
Suppose  > 0 is an arbitrary positive number.
Because L is a lower bound, L ≤ an for all n.
 2.5. MORE RESULTS ON SEQUENCES 41

Because L is the greatest lower bound, L +  is not a lower bound, and

so there is a k (possibly depending on )6 such that ak < L + .7
By the decreasing property of the sequence, it follows an < L +  for all
n ≥ k.
We have shown that L ≤ an < L +  for all n ≥ k, where k may depend
on . Since  > 0 was arbitrary, it follows from the definition of convergence
that an → L as n → ∞. 

Example 2.5.2. Prove the sequence (an ) defined by

√
a1 = 1, an+1 = 6 + an
is convergent, and find the limit.

Solution. (See Adams Example 8, page 501, for details).

The idea is to show by induction that
(1) (an ) is monotone increasing,
(2) an ≤ 3.
It follows from the previous theorem that an → a, say.
a 8 and
√ In order√to find 9a, we use the facts that if an → a then an+1 → √
6 + an → 6 + a . By uniqueness of the limit of a sequence, a = 6 + a.
Solving gives a = −2 or 3. The former is impossible, as it it easy to see
an ≥ 1 for every n. 
The following limits are often useful.
Theorem 2.5.3.
(1) If |x| < 1 then lim xn = 0.
n
(2) If x is any real number, then lim xn! = 0.

Proof. (Adams gives a proof (see Theorem 3 section 9.1 page 501)
which uses continuity and properties of ln for the first part — here is another
proof that does not use this.)
Since |x| < 1 the sequence |x|n is decreasing10 and all terms are ≥ 0.
Hence |x|n → a (say) by Theorem 2.5.1.
Since |x|n → a, also |x|n+1 → a (Why? ). But |x|n+1 = |x| |x|n → |x| a.
Hence a = |x|a by uniqueness of limits, and so a = 0 as |x| = 6 1.
6The statement “possibly depending on ” is redundant. We include it here for
emphasis, but normally would not include it. Whenever we introduce a constant  and
then say there exists a k such that “blah blah involving k and  is true” we always mean
that k may, and indeed it almost always does, depend on .
7In fact there are infinitely many such k, as follows from the decreasing property
which we next use. But at this point we are just using the properties of greatest lower
bound, and so just get the existence of one k.
8It follows easily from the definition of convergence that if a → a, then also a
n n+1 → a
(Exercise). This is frequently a useful fact.
9This can either be proved from the definition of a limit (Exercise). Later it will
√
follow easily from the fact that the function f given by f (x) = 6 + x is continuous.
10We could prove this by induction, but that is not really required at this level as it
is routine and assumed you can give a proof if asked.
42 2. SEQUENCES

Because −|x|n ≤ xn ≤ |x|n and since both |x|n → 0 and −|x|n → 0, it

follows from the Squeeze Theorem that xn → 0.
The second result follows from the first, see Adams. 
n +4n +5n
Example 2.5.4. Find lim 3 5n .
Solution. Example 10 page 502 of Adams. 
The following theorem is easy and useful.
Theorem 2.5.5. Suppose S is a set of real numbers which has lub (glb)
equal to L. Then there is a sequence (xn ) from S such that limn xn = L.

Proof. We do the glb case.

If L ∈ S then we can take the constant sequence L, L, L, . . . . In general,
for each positive integer n, since L is a lower bound and L + 1/n is not a
lower bound, there is an element xn ∈ S such that L ≤ xn < L + 1/n. It
follows that xn → L. 

Remark 2.5.6. It is easy to obtain a sequence (xn ) as in the theorem,

such that (xn ) is decreasing in the glb case and increasing in the lub case.
Exercise.

2.6. Bolzano Weierstrass Theorem

Definition 2.6.1. A subsequence of the sequence a1 , a2 , a3 , . . . , an , . . .
is a sequence ai1 , ai2 , ai3 , . . . , ain , . . . , where i1 < i2 < i3 < · · · < in < · · · .
Subsequences of 1, 1/2, 1/3, . . . , 1/n, . . . are the sequence 1, 1/3, 1/5, . . . ,
the original sequence itself, and infinitely many other possibilities. Subse-
quences of the sequence 1, −1, 1, −1, . . . are the sequence 1, 1, 1, . . . , the
sequence −1, −1, −1, . . . and infinitely many other possibilites.
If a sequence does not converge, it may or may not be the case that some
subsequence converges.
For example, the sequence 1, 2, 3, 4, . . . , does not converge, nor does any
subsequence.11
The sequence 1, −1, 1, −1, . . . does not converge, but the subsequence
1, 1, 1, . . . does converge, as does any subsequence of the original sequence
for which all terms are eventually equal to 1. Similarly, the subsequence
−1, −1, −1, . . . converges as does any subsequence of the original sequence
for which all terms are eventually equal to −1.
The following theorem is needed to prove Theorem 4.5.2 on uniform
continuity, and this in turn is used to prove Theorem 6.2.10 showing that
a continuous function on a closed bounded interval is Riemann integrable.
These results are all very important.
11This is clear. More precisely, if (a ) = a , a , a , . . . , a , . . . is a subsequence
in i1 i2 i3 in
then ain ≥ n as we could show by induction. This rules out limn→∞ ain = a for any real
number a, why?.
 2.6. BOLZANO WEIERSTRASS THEOREM 43

Theorem 2.6.2 (Bolzano Weierstrass Theorem). Let (cn ) be a sequence

of real numbers all of which are contained in the closed bounded interval
[a, b].12
Then some subsequence converges, and the limit also belongs to [a, b].

Remark 2.6.3. The conclusion of the theorem is not true for the non-
bounded interval [1, ∞) or for the nonclosed interval [0, 1). In the first case
consider the sequence 1, 2, 3, . . . , n, . . . and in the second consider the se-
quence 1 − 21 , 1 − 31 , . . . , 1 − n1 , . . . .
Where does the following proof break down in each of these two cases?

Remark 2.6.4. The idea behind the following proof is straightforward.

Divide the interval [a, b] into halves, keeping one subinterval which contains
an (infinite) subsequence of (cn ). Keep subdividing the subintervals into
halves, so that each new subinterval contains an (infinite) subsequence of
the previous (infinite) subsequence. Then define a new subsequence (xn ) of
the original sequence in such a way that xn is in the nth subinterval.
Checking that this all works is a little tedious, but not difficult.

Proof of Theorem. Divide the interval [a, b] into two closed bounded
intervals [a, (a + b)/2] and [(a + b)/2, b] each of equal length and with the
common endpoint (a + b)/2. At least one of these two subintervals contains
an (infinite)13 subsequence of the original sequence (cn ). Choose one such
subinterval and denote it by [a1 , b1 ].
Similarly subdivide [a1 , b1 ] and chose a subinterval which contains an
(infinite) subsequence of the infinite subsequence in [a1 , b1 ]. Denote this
interval by [a2 , b2 ].
Similarly subdivide [a2 , b2 ] to obtain [a3 , b3 ] which contains a subse-
quence of the subsequence of the original sequence. Etc., etc.
Now define a convergent subsequence (xn ) from the original sequence
(cn ) as follows. First choose x1 to be any element from the (infinite) subse-
quence corresponding to [a1 , b1 ]. Next choose some x2 from the subsequence
corresponding to [a2 , b2 ] which occurs in the sequence (cn ) after x1 . (Why is
this possible? ) Next choose some x3 from the subsequence corresponding to
[a3 , b3 ] which occurs in the sequence (cn ) after x2 . (Why is this possible? )
Etc., etc.
We now have
a1 ≤ b1 , a1 ≤ a2 ≤ b2 ≤ b1 , a1 ≤ a2 ≤ a3 ≤ b3 ≤ b2 ≤ b1 , ... ,
x1 ∈ [a1 , b1 ], x2 ∈ [a2 , b2 ], x3 ∈ [a3 , b3 ], ... ,
12A bounded interval is one for which there is both an upper and lower bound, not
necessarily in the interval. In particular, (0, 1), (0, 1] and [0, 1] are all bounded. However,
(−∞, 0], and [1, ∞) are not bounded.
A closed interval is one which contains all of its “finite” endpoints. Thus [0, ∞) and
[0, 1] are both closed. The only closed bounded intervals are those of the form [a, b], where
a and b are both real numbers (and a < b in cases of interest!).
13We use the word “infinite” only for emphasis. By our definition, any sequence is
infinite in the sense it contains an infinite number of terms. Of course, some or even all
of the terms may be equal. Consider the sequence 1, 1, 1, . . . , 1, . . . .
44 2. SEQUENCES

and (xn ) is a subsequence of (cn ).

Since the sequence (an ) is increasing and bounded above it has a lub
L by Theorem 2.5.1. Similarly, (bn ) has a glb M . Since each an is a lower
bound for every bm (why? )14, it follows that an ≤ M for all n. Hence L ≤ M
(why? ).
Since an ≤ L ≤ M ≤ bn for every n, and since bn − an = 2−n (b − a) → 0,
it follows that L = M .
It follows that lim xn = L(= M ), why? (And lim an = lim bn = L,
why? ) 

2.7. FCauchy Sequences

You might think that if a sequence (an )n≥1 satisfies limn (an − an+1 ) = 0
then the sequence converges (to a finite limit). This is false. Consider the
sequence (an )n≥1 given by
0, 21 , 1, 32 , 13 , 0, 41 , 42 , 34 , 1, 45 , 35 , 25 , 15 , 0, 61 , 26 , 63 , 46 , 56 , 1, 67 , . . . .
The sequence does not converge although an+1 − an → 0.
√ √
Another example is given by an = n. Clearly n → ∞ as n → ∞.
However,
√ √ (n + 1) − n 1
n+1− n= √ √ =√ √ →0
n+1+ n n+1+ n
as n → ∞. (To obtain
√ the √
first equality
√ we√“rationalised the numerator” by
multiplying by ( n + 1 + n)/( n + 1 + n).)

However, there is a stronger requirement than limn (an −an+1 ) = 0 which

does imply a sequence (an )n≥1 converges, see Theorem 2.7.2. First we need
a definition.
Definition 2.7.1. A sequence (an )n≥1 is a Cauchy sequence if for every
 > 0 there exists an integer N such that
n, m > N =⇒ |an − am | < .
We write limm,n (an − am ) = 0, or write an − am → 0 as m, n → ∞.
The difference between an − an+1 → 0 and an − am → 0 is that the
first requires the distance between every two consecutive members beyond
a certain point should be < , whereas the second requires the distance
between every two members (not necessarily consecutive) beyond a certain
point should be < .
Theorem 2.7.2. A sequence (an )n≥1 converges if and only if it is Cauchy.

Proof. First assume (an ) converges and so an → a for some a.

Suppose  > 0. Then there exists an integer N such that
n > N =⇒ |an − a| < /2.
Hence if m, n > N then
|an − am | = |an − a + a − am | ≤ |an − a| + |a − am | < /2 + /2 = .
14Think first about a . Then try a . Then a .
1 2 3
 2.7. FCAUCHY SEQUENCES 45

Hence (an )n≥1 is Cauchy.

Next assume (an ) is Cauchy.

The first problem is to identify an a to which the sequence might con-
verge. For this note that the sequence (an ) is bounded. To see this apply
the Cauchy definition with  = 1. It follows there exists an integer N such
that
m, n > N =⇒ |am − an | < 1.
Hence for any n,

|an | ≤ max{|a1 |, . . . , |aN |, |aN +1 | + 1}.

Why? This shows the sequence is bounded.

By the Bolzano Weierstrass Theorem 2.6.2 there is a subsequence (ain )
such that ain → a. We claim that for the full sequence we similarly have
an → a.
To see this suppose  > 0.
Since (an ) is Cauchy there is an integer N such that

m, n > N =⇒ |an − am | < /2.

Since ain → a there is an integer k such that

ik > N and |aik − a| < /2.

Using both the previous inequalities it follows that

n > N =⇒ |an − a| ≤ |an − aik | + |aik − a| < /2 + /2 = .

Since  > 0 was arbitrary, it follows that an → a. 

Remark 2.7.3. The idea behind the second part of the above proof,
from “By the Bolzano Weierstrass ...” onwards, is as follows:
(1) if we go out far enough in the original sequence (an ) then every two
elements of the sequence are within /2 of one another;
(2) if we go out far enough in the subsequence then every element is
within /2 of a.
Putting these two facts together, if we go out far enough in the original
sequence then every element is within  of a.

The following gives a condition on consecutive members of a sequence

which implies the sequence is Cauchy, and hence converges.

Theorem 2.7.4. Suppose a sequence (an )n≥1 satisfies

|an − an+1 | ≤ Krn for all n,

where K is a positive real and 0 ≤ r < 1. Then the sequence is Cauchy and
hence converges.
46 2. SEQUENCES

Proof. Suppose m > n. Then

|an − am | ≤ |an − an+1 | + |an+1 − an+2 | + · · · + |am−1 − am |
≤ K(rn + rn+1 + · · · + rm−1 )
≤ K(rn + rn+1 + · · · + rm−1 + . . . )
rn
=K .
1−r
Hence |an − am | → 0 as m, n → ∞, and so the sequence is Cauchy. 
Note that if a is the limit of the sequence (an ) in the previous theorem
then |an − a| ≤ Arn where A = K/(1 − r). Why?
 CHAPTER 3

Continuous Functions

The intuitive idea of a continuous function is one whose graph can be

drawn without lifting pen from paper. But this is too vague to develop a
useful theory.

3.1. The Approach in these Notes

In this Chapter we first use sequences, and properties of their limits,
in order to define continuity and establish its properties. This approach is
more intuitive and easier to follow for most students that what is called the
 − δ approach used in Adams,1 see also Section 4.4 here. The sequence
approach also enables us to proceed quickly to the proof of the major deep
theorems about continuous functions in Section 3.4.
In the next Chapter we first treat limits in a similar manner to continuity,
by again using sequences. In fact, the properties of continuity follow easily
from the properties of limits, as we see in the next chapter, but the reason
for treating continuity first is to make this Chapter brief and self-contained.
In the next Chapter we also show that the definitions of continuity and
of limit via sequences are equivalent to the analogous −δ definitions.2 Thus
we get to the same point in the theory by either route. Understanding both
approaches will give you much more insight into the subject.
There is one other small difference between the approach here and that
in Adams. In the latter the domain of a function is always an interval or the
union of a finite number of intervals. This is the example case you should
always think about, but here we consider more general domains.

3.2. Definition via Sequences and Examples

Recall3 that the domain of a function f , denoted by D(f ), is the set
of numbers x such that f (x) is defined. We will usually be interested in
functions whose domains are intervals [a, b], (a, b), (a, ∞)4, etc. These are
the cases you should think of when you study the material in these notes,
unless it is indicated otherwise.

1See Adams Chapter 1 and Appendix III.

2In Adams Appendix III page A-20 the fact that the  − δ definitions of limit and
continuity imply the sequence definitions is proved. However, the converse direction, that
the sequence definitions implies the  − δ definitions is not proved.
3
See Adams page 24.
4Note that ∞ is not a number, and that for us the symbol ∞ has no meaning by
itself. The interval (a, ∞) is just the set of real numbers strictly greater than a.

47
48 3. CONTINUOUS FUNCTIONS

But it is possible for the domain to be a more complicated set of real

numbers. In fact, a sequence is just a function whose domain is the set N of
natural numbers.
We will define the notion of continuity of a function in terms of conver-
gence of sequences. The informal idea of “continuity of a function f at a
point c” is that “as x approaches c then f (x) approaches f (c)”.
More precisely, we have the following natural definition.
Definition 3.2.1. A function f is continuous at a point c ∈ D(f ) if for
every sequence (xn )n≥1 from D(f ) such that xn → c, we have f (xn ) → f (c).
The function f is continuous (on its domain) if f is continuous at every
point in its domain.

The first paragraph in the definition can be slightly rewritten as follows:

f is continuous at c ∈ D(f ) if:
xn ∈ D(f ) and xn → c =⇒ f (xn ) → f (c).
We will often not specifically write xn ∈ D(f ), although this
is always
understood in order that f (xn ) be defined. Note that f (xn ) n≥1 is also a
sequence of real numbers.
In order to show f is continuous at c, we have to show that for every
sequence (xn )n≥1 (from D(f )) such that xn → c, one has f (xn ) → f (c).
In order to show f is not continuous at c, we only have to show there is
one (“bad”) sequence (from D(f )) such that xn → c and f (xn ) 6→ f (c).5
Example 3.2.2. Suppose
(
x 0≤x<1
f (x) = 1 2
2x 1 ≤ x ≤ 32
The domain of f is [0, 23 ]. The following is an attempt to sketch the graph
of f .

Figure 1. Sketch of the graph of f .

It is clear that f is not continuous at 1. For example, take the sequence

xn = 1 − n1 . Then xn → 1 but f (xn ) (= 1 − n1 ) 6→ f (1) since f (1) = 12 .
5If there is one, there will in fact be many such “bad” sequences — we can always
change the first million or so terms — but the point of logic is that to show continuity
fails it is sufficient to just prove there is one “bad” sequence.
 3.2. DEFINITION VIA SEQUENCES AND EXAMPLES 49

On the other hand, if c 6= 1 and c ∈ D(f ) then

xn → c ⇒ f (xn ) → f (c).
To see this, first suppose xn → c and 1 < c ≤ 23 . Then xn ≥ 1 for all
sufficiently large n, and so f (xn ) = 21 x2n for all sufficiently large n. From
properties of sequences if xn → c then x2n → c2 and so 21 x2n → 21 c2 . But
f (xn ) = 12 x2n for all sufficiently large n, and so lim f (xn ) = lim 12 x2n = 12 c2 .
The case 0 ≤ c < 1 is similar, and easier.
In particular, f is not continuous on its domain since it fails to be con-
tinuous at c = 1.
If we vary this example a little, and define
(
x 0≤x<1
g(x) = 1 2
2x 1 < x ≤ 32 ,
then the domain of g is [0, 1) ∪ (1, 23 ]. The function g is continuous at each
c ∈ D(g), and so is continuous on its domain.
However, there is no extension of g to a continuous function defined on
all of [0, 3/2].

Example 3.2.3. The absolute value function f (given by f (x) = |x|) is

continuous.
We first show continuity at 0. For this, suppose xn → 0. Then |xn | → 0
(this is immediate from the definition of convergence, since |xn − 0| ≤  iff
| |xn | − 0 | ≤ ), i.e. f (xn ) → f (0).
To prove continuity at c 6= 0 is similar to the previous example.
The following result is established directly from the properties of con-
vergent sequences.
Proposition 3.2.4. Every polynomial function is continuous.

Proof. Let
f (x) = a0 + a1 x + a2 x2 + · · · + ak xk .
To show that f is continuous at some point c, suppose xn → c.
Then x2n → c2 , x3n → c3 , etc., by the theorem about products of con-
vergent sequences. It follows that a1 xn → a1 c, a2 x2n → a2 c2 , a3 x3n → a3 c3 ,
etc., by the theorem about multiplying a convergent sequence by a constant.
Finally,
a0 + a1 x + a2 x2 + · · · + ak xk → a0 + a1 c + a2 c2 + · · · + ak ck
by repeated applications of the theorem about sums of convergent sequences
(a0 is here regarded as a constant sequence). 

Example 3.2.5.F Here is a surprising example.

Let (
x x rational
f (x) =
−x x irrational.
50 3. CONTINUOUS FUNCTIONS

Figure 2. An attempt to sketch the graph of f .

The following diagram is misleading, since between any two real numbers
there is both a rational and an irrational number.
The function f is continuous at 0. To see this, suppose xn → 0. Then
|xn | → 0 (this follows from the definition of a limit). Since −|xn | ≤ f (xn ) ≤
|xn |, it follows from the Squeeze Theorem that f (xn ) → 0, i.e. f (xn ) → f (0).
On the other hand, f is not continuous at c if c 6= 0. For example if c is
irrational then we can choose a sequence of rationals xn such that xn → c
(by repeated applications of the remark above in italics). It follows that
f (xn ) = xn → c 6= f (c). Similarly if c is irrational.
We will later define the exponential, logarithm, and trigonometric func-
tions, and show they are continuous. Meanwhile, we will use them in exam-
ples (but not in the development of the theory).

3.3. Basic Properties of Continuous Functions

These properties follow easily from the analogous properties of sequences.
3.3.1. Combining Continuous Functions.
Theorem 3.3.1. Let f and g be continuous functions and let D = D(f )∩
D(g). Then
(1) f + g is continuous on D,
(2) f g is continuous on D,
(3) αf is continuous on D(f ) (α any real number),
(4) f /g is continuous at any point c ∈ D such that g(c) 6= 0.

Proof. Suppose c ∈ D. Let (xn ) be any sequence such that xn → c

(and as usual, xn ∈ D).
Then f (xn ) → f (c) and g(xn ) → g(c), since f and g are continuous at
c. It follows
f (xn ) + g(xn ) → f (c) + g(c)
by Theorem 2.3.1 about sums of convergent sequences. That is,
(f + g)(xn ) → (f + g)(c).
It follows that f + g is continuous at c.
 3.3. BASIC PROPERTIES OF CONTINUOUS FUNCTIONS 51

The proof in the other cases is similar. Just note for the case f /g that
if xn → c and g(c) 6= 0, then g(xn ) 6= 0 for all sufficiently large n.6 
The composition of two continuous functions is continuous. (See Adams
page 35 for a discussion about the composition of functions.)
Theorem 3.3.2. Suppose f and g are continuous. Then f ◦ g is contin-
uous.

Proof. The domain D of f ◦ g is the set of numbers x such that both

x ∈ D(g) and g(x) ∈ D(f ).
Suppose c ∈ D. Let xn → c and xn ∈ D. It follows that g(xn ) → g(c)
since g is continuous at c. It then follows that f (g(xn )) → f (g(c)) since f is
continuous at g(c) (note that g(xn ) ∈ D(f )). In other words, (f ◦ g)(xn ) →
(f ◦ g)(c), and so f ◦ g is continuous at c. 
It follows from our results so far that rational functions (quotients of
polynomials) and in general functions defined from other continuous func-
tions by means of algebraic operations and composition, will be continuous
on their domains.
3.3.2. Analogous Results at a Point. As in the previous section,
we will usually be interested in functions that are continuous everywhere
on their domain. However, occasionally we may not have continuity at
every point in the domain. However, the following analogues of the previous
theorems are proved with almost exactly the same proofs as before. Check
this yourself.
Theorem 3.3.3. Suppose the functions f and g are both continuous at
a, and c is any real number. Then the following are also continuous at a:
f + g, f g, cf, f /g provided g(a) 6= 0.
If g is continuous at a and f is continuous at g(a) then f ◦ g is continuous
at a.

3.3.3. Removable and Non-Removable Singularities. The func-

tion
1
f1 (x) = sin
x
is the composition of the two continuous functions sin(x) and 1/x 7 and so is
continuous. The domain of f1 is the set of real numbers x such that x 6= 0.
That is, D(f1 ) = { x | x 6= 0 }.
6If g(c) > 0, by continuity of g at c and the definition of convergence of a sequence,
g(xn ) ∈ [ 21 g(c), 23 g(c)] for all sufficiently large n and so it is positive. The argument in
case g(c) < 0 is similar.
7The notation may seem a bit confusing. You may ask “is it the same x in both
cases”? But this is not the right way to look at it. By the function sin x, is meant the
function which assigns to each real number x (say) the real number sin x. If we said the
function sin y, or just sin, we would mean the same thing.
Similarly, the function 1/x, or 1/y, or “the reciprocal function”, all mean the same
thing.
52 3. CONTINUOUS FUNCTIONS

Similarly, the function

1
f2 (x) = x sin
x
is continuous on its domain, which is the same domain as for f1 .

Figure 3. Graphs of f1 (x) = sin(1/x) and f2 (x) = x sin(1/x).

However, there is an important difference between f1 and f2 , even though

they both have the same domain and are continuous on this domain. In the
case of f2 we can define a new function g2 by
(
x sin x1 x 6= 0
g2 (x) =
0 x = 0.

Then D(g2 ) = R and g2 (x) = f2 (x) if x 6= 0, i.e. if x ∈ D(f2 ). Moreover, g2

is continuous on its domain R.
To show continuity of g2 at x 6= 0, take any sequence xn → x. For all
sufficiently large n, xn ∈ D(f2 ), and so g2 (xn ) = f2 (xn ). It follows that
g2 (xn ) → g2 (x) since f2 (xn ) → f (x) by the continuity of f . This means g2
is continuous at x if x 6= 0.
To show continuity of g2 at x = 0, take any sequence xn → 0. Then

−|xn | ≤ g2 (xn ) ≤ |xn |

since | sin t| ≤ 1, and so g2 (xn ) → 0 (= g2 (0)) by the Squeeze Theorem. (We

need to be a bit careful since some of the xn may equal zero.) This means
g2 is continuous at 0.

In the case of f1 there is no way of extending the function to a continuous

function g1 defined on all of R. This is essentially because there is no number
y such that f1 (xn ) → y for every sequence xn → 0 (with xn 6= 0.)

We sometimes say that f2 has a removable singularity at 0, and that the

singularity of f1 at 0 is not removable.
 3.4. CONTINUOUS FUNCTIONS ON A CLOSED BOUNDED INTERVAL 53

3.4. Continuous Functions on a Closed Bounded Interval

The following two theorems are “deep” and require the Completeness
Axiom for their proof.
Adams gives different proofs in Appendix III. But here we have easier
proofs using sequences and the Bolzano Weierstrass Theorem.
See also Adams pp 80–85 for some discussion of these matters.
Theorem 3.4.1 (Boundedness and Max-Min Theorems). If f is contin-
uous on [a, b], then it is bounded there (i.e. there exists a constant K such
that |f (x)| ≤ K if a ≤ x ≤ b).
Moreover, there exist points u, v ∈ [a, b] such that for any x ∈ [a, b] we
have
f (v) ≤ f (x) ≤ f (u).
That is, f assumes maximum and minimum values on [a, b].

Proof. Suppose in order to obtain a contradiction that f is not bounded

above.
Then for each positive integer n there is some xn ∈ [a, b] such that
f (xn ) > n. By the Bolzano Weierstrass Theorem there is a subsequence of
(xn ), which we denote by (x0n ), such that (x0n ) converges to some c ∈ [a, b].
By continuity, f (x0n ) → f (c). But f (x0n ) → ∞ since f (xn ) → ∞.
This is a contradiction. Hence f is bounded above. Similarly f is
bounded below.
Now let
L = lub {f (x) : x ∈ [a, b]}.
By what we have proved, L is finite. It follows from Theorem 2.5.5
applied to the set S = {f (x) : x ∈ [a, b]} that there is a sequence (xn ) for
which f (xn ) → L.
By the Bolzano Weierstrass Theorem there is a subsequence of (xn ),
which we denote by (x0n ), such that x0n → u for some u ∈ [a, b]. By the
continuity of f , f (x0n ) → f (u).
Because f (xn ) → L it follows f (x0n ) → L. (Why? ) By uniqueness of
limits (see Theorem 2.3.4) it follows f (u) = L.
Since f (x) ≤ L for all x ∈ [a, b], it immediately follows that f (x) ≤ f (u)
for all x ∈ [a, b].
Similarly there is a minimum point v. Write out the proof. 

Remark 3.4.2. The Completeness Axiom is used in the proof, since it

is used in the proof of the Bolzano Weierstrass Theorem.
In fact the Completeness Axiom is required for the theorem to hold.
To see this, first note that all the other axioms hold if we restrict to the
“universe of rational numbers”.
3
Now consider the function √ f (x) = x − x on the domain [0, 1]. The
maximum occurs at x = 1/ 3, as a little calculus shows. (One can also
show this directly, how? ) It follows there is no rational maximum point, and
so the previous theorem is not true in the “universe of rational numbers”.
54 3. CONTINUOUS FUNCTIONS

Remark 3.4.3. The next theorem implies that if a continuous function

defined on an interval I (not necessarily closed or bounded) takes two par-
ticular values, then it must take all values between. In other words, for any
two points a, b ∈ I and any γ between f (a) and f (b), there is a c ∈ [a, b]
such that f (c) = γ.
In order to understand the proof, look at Figure 4. It is clear in this
simple case that f (c) = γ. But the function f needs to be continuous,
the proof needs to follow from the original defintion of continuity, and the
Completeness Axiom has to come into the proof. It also should be noted
that continuous functions can be pretty wild, there are examples which are
not differentiable at any point!

Figure 4. f is continuous on [a, b], A = {x ∈ [a, b] : f (x) ≤

γ}, c = lub A.

Theorem 3.4.4 (Intermediate Value Theorem). Suppose f is continuous

on [a, b]. Then for any γ between f (a) and f (b) there exists c ∈ [a, b] such
that f (c) = γ.

Proof. We do the case where f (a) < γ < f (b).

Let
A = {x ∈ [a, b] | f (x) ≤ γ }.
Because A is bounded, and nonempty since a ∈ A, it follows that it has
a supremum c, say.
We want to show that f (c) = γ.
There is a sequence xn ∈ A such that xn → c (Theorem 2.5.5). By
continuity, f (xn ) → f (c). Since f (xn ) ≤ γ for all n, it follows f (c) ≤ γ
(special case of Theorem 2.3.6). So, in particular, c ∈ A.
Since c 6= b (because f (c) ≤ γ but f (b) > γ), there is a sequence x0n → c
and c < x0n < b. But f (x0n ) > γ (since otherwise x0n ∈ A, which contradicts
c = sup A) and so f (c) ≥ γ (Theorem 2.3.6 again).
Because f (c) ≤ γ and f (c) ≥ γ it follows f (c) = γ. 
 3.5. FFUNCTIONS OF TWO OR MORE VARIABLES 55

Example 3.4.5. See Example 11 in Section 1.4 of Adams. Here the

Intermediate Value Theorem is used to justify the existence of a solution of
the equation x3 − x − 1 = 0.
One can also prove the existence of a number x such that x2 = 2 in this
manner. Just note that if f (x) = x2 then f (1) = 1, f (2) = 4, and since f is
continuous it follows by the Intermediate Value Theorem that f (x) √ = 2 for
some x between 1 and 2. Thus we have justified the existence of 2, i.e. a
positive number whose square is 2.

Example 3.4.6. In order to improve/test your understanding give ex-

amples to show that all the hypotheses in the previous two theorems are
necessary. Make your examples as simple as possible.
A first example might be of a simple function f which satisfies all the hy-
potheses of Theorem 3.4.1 except that f is not continuous and the theorem’s
conclusion of being bounded does not hold. Another example would give
a discontinuous f which is bounded but does not have a maximum value.
Others would give examples where the domain is (0, 1] (hence not a closed
bounded interval) and the various conclusions of the theorem do not hold.

Example 3.4.7. An interesting problem. Use the Intermediate Value

Theorem to prove that at any fixed point in time, there are two antipodal
points on the equator with the same temperature. Assume the temperature
is a continuous function of position.

3.5. FFunctions of two or more variables

Suppose f : A (⊂)R2 → R. Think of the case of a “closed bounded
rectangle”
A = {(x, y) : a ≤ x ≤ b, c ≤ y ≤ d}.


We say f is continuous at (u, v) ∈ A if, for every sequence (xn , yn ) n≥1
from A such that (xn , yn ) → (u, v), we have f (xn , yn ) → f (u, v).8 This is
completely analogous to Definition 3.2.1.
Analogues of Theorems 3.3.1, 3.3.2 and 3.3.3 hold, with similar proofs.
In particular, if h and g are functions of one variable which is continuous
at a (and hence defined at a), and f is a function of two variables which is
continuous at (h(a), g(a)) (and hence defind there), then f (h(x), g(x)) is a
function of one variable which is continuous at a.
Analogues of the Bolzano Weierstrass Theorem 2.6.2, Boundedness and
Max-Min Theorem 3.4.1 and the Uniform Continuity Theorem 4.5.2 hold
for continuous functions on any closed bounded rectangle A. The proofs are
similar.

8We define (x , y ) → (u, v) to mean that |(x , y ) − (u, v)| → 0. This can be easily
n n n n
shown to be equivalent to xn → u and yn → v. Exercise.
 CHAPTER 4

Limits

4.1. Definition via Sequences, and Examples

We often need to consider limits of a function f at a point c where f
may not be continuous or even defined. For example, consider the two cases
( (
2x if x 6= 1 2x if x 6= 1
f (x) = , f (x) = .
undefined if x = 1 3 if x = 1
In either case we want the limit of f at c = 1 to equal 2, that is we want
a definition of limit such that limx→1 f (x) = 2.
Important points to take into account when defining limx→c f (x) are
that
(1) c may not be in D(f );
(2) if c ∈ D(f ) then the value of f (c) is not relevant to the existence
or value of the limit at c;
(3) we need at least one sequence (xn ) from D(f ) such that xn → c
and such that xn 6= c for all n.
We now first define the set of points at which we consider the existence
of a limit, and then we define what we mean by a limit. The following
definition is a natural consequence of the previous discussion.
Definition 4.1.1. Suppose f is a function with domain D(f ). Then c
is a limit point of D(f ) if there exists a sequence (xn ) from D(f ) such that
xn → c and such that xn 6= c for all n.
Suppose c is a limit point of D(f ). If f (xn ) → L for every sequence (xn )
from D(f ) such that xn → c and xn 6= c for all n, then we say that f has
limit L at c. We write
lim f (x) = L.
x→c

Remark 4.1.2. If c ∈ D(f ) and c is not a limit point of D(f ) we say c

is an isolated point of D(f ). For example, if D(f ) = [0, 1] ∪ {3} then 3 is an
isolated point of D(f ). See Example 4.1.3. 

Example 4.1.3. For the two examples at the beginning of this Chapter
it follows from Definition 4.1.1 that limx→1 f (x) = 2. Why? In fact we have
in both cases that limx→a f (x) = 2a for every real number a. Why?
Other simple examples are
f (x) = 2x for x ∈ D(f ) = [0, 1], f (x) = 2x for x ∈ D(f ) = [0, 1).
In both cases limx→1 f (x) = 2.
57
58 4. LIMITS

Finally, if
f (x) = 2x for x ∈ D(f ) = [0, 1] ∪ {3},
then limx→3 f (x) is not defined, why? In particular, limx→3 f (x) = 6 is not
a true statement, since there is no limit of f at 3.

4.2. Calculating Limits

The following rules are listed on page 68 of Adams.
The proofs for us are easy, because we defined limits via sequences and
we have already proved similar properties for sequences in Theorem 2.3.1
and Section 2.4.
Since Adams defines limits using the  − δ approach, as we later also do
here in Section 4.4, he needs to prove the following Theorem from “scratch”.
See Adams page 89 for the sum and the Exercises on page 92 for the other
cases.
Theorem 4.2.1. Suppose that a is a limit point of D(f ) ∩ D(g).1
Suppose limx→a f (x) = L, limx→a g(x) = M and k is a constant. Then


(1) limx→a f (x) + g(x)

= L + M;
(2) limx→a f (x)g(x)

= LM ;
(3) limx→a kf (x) = kL;


(4) limx→a f (x)/g(x) = L/M, if M 6= 0.

Proof. (We just do the quotient. The other cases are similar and
slightly easier. Do them yourself ! )
Suppose xn → a where (xn )n≥1 is any sequence such that for all n,
xn ∈ D(f ) ∩ D(g).
Then f (xn ) → L and g(xn ) → M . (Why? ) Moreover, since M 6= 0 it
follows that g(xn ) 6= 0 for all sufficiently large n (why? ) and so f (xn )/g(xn )
is a real number for such n.
Since M 6= 0 it follows from Theorem 2.3.1 that f (xn )/g(xn ) → L/M .
The fourth claim in the Theorem now follows from Definition 4.1.1. 

4.3. Limits and Continuity

In the following Proposition, particularly the second paragraph, we see
that once we have the definition of the limit of a function at a point c, we
can define the notion of continuity at c in terms of such a limit.
The proof is a straightforward matter of using the relevant definitions.
It is just a little messy and tedious to write out! The main point is that
in Definition 3.2.1 of continuity we consider all sequences xn → c, whereas
in Definition 4.1.1 of a limit we only consider sequences (xn )n≥1 satisfying
xn 6= c. (In both cases, as usual, xn ∈ D(f ).)

1It follows that a is a limit point of both D(f ) and D(g). Why? The important and
simple case to keep in mind is where f and g have the same domain and this domain is
an interval.
 4.4. DEFINITIONS OF LIMIT AND CONTINUITY VIA  − δ 59

Note that Adams does not need to consider the first paragraph in the
Proposition, since the type of domain he considers does not contain isolated
points.2
Proposition 4.3.1. Suppose f is a function and c ∈ D(f ). If c is not
a limit point of D(f ), i.e. c is an isolated point of D, then f is continuous
at c.
If c is a limit point of D(f ) then f is continuous at c iff limx→c f (x) =
f (c).

Proof. First suppose c ∈ D(f ) and c is an isolated point in D(f ).

(Think of the last example in Example 4.1.3.)
In order to apply Definition 3.2.1 let (xn ) be any sequence from D(f )
such that xn → c. It follows that xn = c for all sufficiently large n (why? ).
Hence f (xn ) = f (c) for all sufficiently large n, and so in particular f (xn ) →
f (c). Hence f is continuous at c.
Next suppose c ∈ D(f ) and c is a limit point of D(f ).
First assume f is continuous at c. By Definition 3.2.1, for any sequence
(xn ) from D(f ) such that xn → c (even if xn = c for some n) it follows that
f (xn ) → f (c). In particular it follows that limx→c f (x) = f (c).
Next assume that limx→c f (x) = f (c). (Remember that Definition 4.1.1
only considers sequences (xn ) satisfying xn 6= c.) Now consider any sequence
(xn ) from D(f ) such that xn → c. We want to show that f (xn ) → f (c).
Split the sequence (xn ) into one subsequence (x0n ) consisting of those terms
xn 6= c and the remaining subsequence (x00n ) consisting of those terms xn = c.
At least one of these subsequences is infinite.
If (x0n ) is an infinite sequence, we have that f (x0n ) → f (c) by the Defi-
nition 4.1.1 of a limit. If (x00n ) is an infinite sequence, we have f (x00n ) → f (c)
since f (x00n ) = f (c) for all n. Putting these two facts together it follows that
for the original sequence (xn ) we have f (xn ) → f (c). 

4.4. Definitions of Limit and Continuity via  − δ

See Adams, Section 1.5 pages 87 and 88. The following definition is just
a little different from Definition 8 on page 88 in Adams. It is written in
exactly the same format. What is the difference?
Note that Definition 8 in Adams applies only to domains which are a
finite union of intervals, and does not usually apply to endpoints of such an
interval.3 Adams uses Definition 9 on page 90 of a right limit (and similarly
for left limit), to take care of endpoints.
Definition 4.4.1. We say that f (x) approaches the limit L as x ap-
proaches a, and we write
lim f (x) = L,
x→a
if the following conditions are satisfied:
2Isolated points are just a bit of a nuisance and not particularly significant when we
consider limits and continuity.
3If the domain is (0, 1), [0, 1], (0, 1] or [0, 1), then Adams’ definition of a limit does
not apply to either 0 or 1, whereas the (standard) definition given here does apply. If the
domain is [0, 1) ∪ (1, 2] then Adams’ definition does apply at 1, as does the one here.
60 4. LIMITS

(1) a is a limit point of D(f ),

(2) for every number  > 0 there exists a number δ > 0, possibly
depending on , such that if 0 < |x − a| < δ and x ∈ D(f ), then
|f (x) − L| < .

The idea of the definition is that for any given “tolerance”  > 0, there
is a corresponding “tolerance” δ > 0, such that any x in the domain of f
within distance δ of a, and not equal to a, gives an output (or value) f (x)
which is within  of L.

Inputs and Outputs Analogy. For any allowable x (i.e. x ∈ D(f )), input
x into the f machine gives output f (x).
Suppose there are allowable inputs arbitrarily close to a. Then limx→a f (x) =
L is equivalent to the following:
For every output tolerance  > 0 for the deviation of f (x) from L, there
is a corresponding input tolerance δ > 0 (normally depending on ) for the
deviation of x from a, such that whenever the input tolerance is satisfied by
x 6= a, then the output tolerance is satisfied by f (x).

Figure 1. limx→a f (x) = L means: for each  > 0 there is

a corresponding δ > 0 such that if x 6= a and x is δ-tolerant
from a, then f (x) is -tolerant from L.

Now we come to the important theorem showing the approach to limits

in Definition 4.1.1 via sequences, and the  − δ approach in Definition 4.4.1,
are equivalent.
The proof in one direction is easy.4 The other direction is more subtle
and is not covered in Adams, even in the Appendices. If you find it tricky,
don’t be too concerned. (Just think about it for a couple of hours each day.)
Theorem 4.4.2. Suppose f is a function and c is a limit point of D(f ).
Then limx→c f (x) = L according to Definition 4.1.1 iff limx→c f (x) = L
according to Definition 4.4.1.
4See Adams Theorem 1(b) and Theorem 4 in Appendix III for similar results. But it
is easier in Adams as he just covers special cases.
 4.4. DEFINITIONS OF LIMIT AND CONTINUITY VIA  − δ 61

Proof. First suppose limx→c f (x) = L according to Definition 4.4.1.

We want to show that
(23) xn 6= c & xn → c =⇒ f (xn ) → L.
In order to do this, suppose  > 0 is given. We need to show there is a
corresponding N such that
(24) n>N =⇒ |f (xn ) − L| < .
First note from Definition 4.4.1 that there exists δ > 0 such that
0 < |x − c| < δ =⇒ |f (x) − L| < .
But if xn 6= c and xn → c, then there is an N corresponding to δ such that
n>N =⇒ 0 < |xn − c| < δ.
Combining the last two implications gives (24). Since  > 0 was ar-
bitrary, this gives (23). In other words, f is continuous at c according to
Definition 4.1.1.

Conversely, suppose f is continuous at c according to Definition 4.1.1.

Suppose  > 0 is given. We want to show there is some corresponding δ > 0
such that
(25) 0 < |x − c| < δ ⇒ |f (x) − L| < .
Assume there is no such δ (in order to obtain a contradiction). This
means that for each δ > 0 there is a real number x (with x ∈ D(f )) such
that
0 < |x − c| < δ but |f (x) − L| ≥ .
Let δ = n1 and denote some x as above by xn . Thus we have for each
natural number n a real number xn such that
1
0 < |xn − c| < but |f (xn ) − L| ≥ .
n
It follows that xn 6= c and xn → c but f (xn ) 6→ L.
This contradicts the fact that f is continuous at c according to Definition
3.2.1, and so the assumption is false. In other words, there is a δ > 0
corresponding to  such that (25) is true. Since  > 0 was arbitrary, it
follows that f is continuous at c according to Definition 4.4.1. 

There is also an  − δ definition of continuity at c and a corresponding

theorem showing the equivalence with the sequence definition. The ideas
are essentially the same as for limits, but easier because we do not need to
restrict to x 6= c and to xn 6= c.
Definition 4.4.3. A function f is continuous at a point c ∈ D(f ) if for
every  > 0 there is a corresponding δ > 0 such that
x ∈ D(f ) and |x − c| < δ ⇒ |f (x) − f (c)| < .
The function f is continuous (on its domain) if f is continuous at every
point in its domain.
62 4. LIMITS

Figure 2. In this diagram we see that

|x − c| < δ =⇒ |f (x) − f (c)| < .
For  as shown, I have drawn the largest δ for which this
implication holds. Can you explain to a neighbour why a
larger δ will not work? Any smaller δ > 0 will also work.

See Figure 2 for a diagram similar to that in Adams on page 87 for

limits.
Finally, we have the following theorem.
Theorem 4.4.4. Suppose f is a function and c ∈ D(f ).
Then f is continuous at c according to Definition 3.2.1 iff it is continuous
at c according to Definition 4.4.3.

Proof. Almost exactly the same as for the proof of Theorem 4.4.2.
Write it out yourself to consolidate your understanding. 

4.5. Uniform Continuity

In the Definition 4.4.3 of continuity at a point c ∈ D(f ), we require that
for every  > 0 there is a corresponding δ > 0 with a certain property. The
number δ will normally depend on , but δ may also depend on c in an
essential manner.
As a simple example, consider the function f (x) = x2 defined on R. For
simplicity consider the case where c is large and positive and x ∈ (c−1, c+1).
Then
(
2 2 ≤ |x − c| (2c + 1)
|f (x) − f (c)| = |x − c | = |x − c| |x + c| .
≥ |x − c| (2c − 1)
This implies that the δ > 0 needed in Definition 4.4.3 will be roughly of
the order /2c or less. More precisely, any δ less than /(2c + 1) will work,
but δ must not be more than /(2c − 1).
Thus for a given  > 0, no single δ will work for every c. For a fixed
 > 0, the larger we take c the smaller we need to take δ.
 4.5. UNIFORM CONTINUITY 63

If for each  > 0 there is a δ > 0 which works for all c ∈ D(f ) then we
say that f is uniformly continuous on its domain.
Definition 4.5.1. A function f is uniformly continuous on its domain if
for every  > 0 there is a corresponding δ > 0 such that, for every c ∈ D(f ),
x ∈ D(f ) and |x − c| < δ =⇒ |f (x) − f (c)| < .

Theorem 4.5.2. If f is continuous on a closed bounded interval [a, b]

then it is uniformly continuous on [a, b].

Proof. We argue by contradiction.

Assume f is not uniformly continuous on its domain. By Definition 4.5.1
this is equivalent to claiming there is a “bad”  > 0, for which there is no
δ > 0 such that for every c ∈ D(f ),
x ∈ D(f ) and |x − c| < δ =⇒ |f (x) − f (c)| < .
In particular, for each positive integer n, we see by taking δ = 1/n that
there exists cn ∈ D(f ) and xn ∈ D(f ), such that
1
(26) |xn − cn | < and |f (xn ) − f (cn )| ≥ .
n
By the Bolzano Weierstrass Theorem there is a subsequence of (cn ),
which we denote by (c0n ), such that c0n → c for some c ∈ [a, b]. Let (x0n )
be the corresponding subsequence of (xn ). It follows that x0n → c since
x0n − c0n → 0 (why? ) and so
x0n = c0n + (x0n − c0n ) → c + 0 = c.
Since f is continuous at c, and since both x0n → c and c0n → c, it follows
that f (x0n ) → f (c) and f (c0n ) → f (c). Hence |f (x0n ) − f (c0n )| → 0, which
contradicts the fact from (26) that |f (x0n ) − f (c0n )| ≥  for all n.
This contradiction shows our assumption is false and so f is uniformly
continuous. 

Remark 4.5.3. We saw an example at the beginning of this section

showing the theorem would not hold if we dropped just the assumption that
the domain of f is bounded.
If we drop just the assumption that the domain be closed, the theorem
will not hold. A counterexample is obtained by defining f (x) = 1/x for
x ∈ (0, 1]. Try and write out the details.
√
Remark 4.5.4. If f (x) = x for x ∈ [0, 1] then the function f is uni-
formly continuous on its domain by the Theorem. This may seem surprising,
since f 0 (x) → ∞ as x → 0.
The fact that f (x) = x2 is not uniformly continuous on R is related to
the fact that f 0 (x) → ∞ as x → ∞. But this is not a complete explanation!
 CHAPTER 5

Differentiation

The main references in Adams is Chapter 2.

Convention and Notation In this and the following chapters, unless

stated otherwise, the domain of a function is an interval. The interval may
be open or closed at either end, and may be bounded or unbounded.1
A point in the domain is an interior point if it is not an endpoint.
The results we prove usually extend in a straightforward manner to more
general cases, in particular if the domain is a finite union of intervals.

5.1. Introduction
The theory of differentiation allows us to analyse the concept of the slope
of the tangent to the graph of a function. Similarly, it allows us to find the
best linear approximation to a function near a given point.
If we write y = f (x) then we can interpret f 0 (a) in the following way: “for
x near a, y is changing approximately f 0 (a) times as fast as x is changing”.
Alternatively, for x ≈ a (“x approximately equal to a”) we have
f (x) ≈ f (a) + f 0 (a)(x − a).
See Adams Chapter 4.9, where this is made more precise.
There are many problems that can then be analysed using the ideas of
differentiation and extensions of these ideas. For example, anything that
changes with time or position, as well as optimisation problems (e.g. in
economics or engineering) and approximation problems. See Adams Chapter
3 for a number of examples.

5.2. The Derivative of a Function

Derivatives are defined and the fact that differentiability implies continuity
is proved.
The idea from Figure 1 is that the derivative f 0 (a) of f at a should be
the slope of the tangent to the graph of f at the point (a, f (a)) on the graph.
We make this precise by considering the slope of the line through the
two points (a, f (a)) and (a + h, f (a + h)) and considering the limit (if it
exists) as h → 0 (where h is allowed to be either positive or negative, except
at endpoints a of an interval from the domain of f ).

1Thus the allowable domains are (a, b), [a, b], (a, b], [a, b), (the four bounded possibil-
ities); (a, ∞), [a, ∞), (−∞, b), (−∞, b], (−∞, ∞), (the five unbounded possibilities).

65
66 5. DIFFERENTIATION

Figure 1. Draw in the tangent line at (a, f (a)).

Definition 5.2.1. If a is an interior point of the domain of f 2 and

f (a + h) − f (a)
lim
h→0 h
exists, or if a ∈ D(f ) and is an endpoint and the corresponding one-sided
limit
f (a + h) − f (a) f (a + h) − f (a)
lim or lim
h→0+ h h→0− h
exists, then we say f is differentiable at a.
The limit is denoted by f 0 (a) (sometimes f+0 (a) or f−0 (a) in the case of
endpoints) and is called the derivative of f at a.
The derivative of f is the function f 0 whose value at a is the number
0
f (a) defined above, with domain consisting of all a such that the derivative
f 0 (a) exists.
The function f is differentiable if it is differentiable at every point in its
domain.
An alternative way of writing the same limit is
f (x) − f (a)
f 0 (a) = lim ,
x→a x−a
and similarly for the one-sided limits. Why is this equivalent?3
The tangent to the graph of f at a has slope f 0 (a). It follows that the
equation of the tangent line is
y = f (a) + f 0 (a)(x − a).

See the examples on pages 96 and

√ 97 of Adams, where the derivatives of
some simple functions x2 , 1/x and 3 x are calculated directly from the above
definition. But we do not usually need to do this. Instead we normally can
use the methods in Section 5.3.
Notation 5.2.2. If y = f (x) then we use the dependent variable y to
represent the function, and the derivative is denoted in the following various
ways:
dy d
y0, , f (x), f 0 (x),
dx dx
2This, of course, implies that a ∈ D(f ).
3Just write out the corresponding ε − δ definition in each case and checking that each
limit means the same thing.
 5.2. THE DERIVATIVE OF A FUNCTION 67

which we read as “y prime”, “the derivative of y with respect to x” or “dy

dx” for short, “the derivative with respect to x of f (x) or “d dx of f (x)” for
short, and “f prime of x”, respectively.
In particular, we often write
d 3
x = 3x2 ,
dx
d 4
t = 4t3 ,
dt
d
etc., and regard dx as a “differential operator” which maps one function to
another function; such as the function f given by f (x) = x3 to the function
g given by g(x) = 3x2 .
(F Thus a differential operator is a function which sends functions to
functions, rather than numbers to numbers!)
The value of the derivative of a function at a fixed real number a can
also be written in various ways:
dy d
y 0 (a), y 0 , , f (x) , f 0 (a).
a dx a dx a

The symbol is the evaluation symbol, and signifies that the function pre-
ceding it should be evaluated at a. If there is any doubt as to what is the
dependent variable, one replaces by .
a x=a
dy
The dx type notation is called Leibniz notation after its inventor. It is
very good for computations and for motivating some results. If one thinks
of
∆y = f (x + h) − f (x)
as being the increment in y and
∆x = (x + h) − h = h
as being the increment in x, then
dy ∆y
= lim .
dx ∆x→0 ∆x
However, the Leibniz notation should not be used when proving theorems
rigorously. It is often ambiguous in more complicated situations, and this
can easily lead to logical errors. See the discussion before Theorem 5.3.4 for
a good example of what can go wrong.
See Adams pp 103-105 for more discussion of notation.

The following theorem is important.

Theorem 5.2.3. If f is differentiable at a then f is continuous at a.

Proof. Assume f is differentiable at a, and a is an interior point of

D(f ). We want to show4 that
lim f (a + h) = f (a).
h→0

4f is continuous at a means lim

x→a f (x) = f (a). This is the same as limh→0 f (a+h) =
f (a). See Footnote 3 for a similar but slightly more complicated situation.
68 5. DIFFERENTIATION

But
f (a + h) − f (a)
f (a + h) = f (a) + h .
h
Taking the limit as h → 0 of the right side, we see this limit exists and hence
so does the limit of the left side, and both are equal. That is
lim f (a + h) = f (a) + 0f 0 (a) = f (a).
h→0
A similar proof applies if a is an endpoint of the domain of f . 

5.3. Computing Derivatives

The standard rules for differentiation, including the chain rule, are dis-
cussed. Examples are given.

5.3.1. Sums, Products and Quotients. The next result is easy to

check from the definitions, and is obvious from the relevant diagram. It
implies that the slope of the straight line, which is the graph of the function
f (x) = cx + d, is c.
Theorem 5.3.1. If f (x) = cx + d then f 0 (x) = c.

Proof.
f (x + h) − f (x) (c(x + h) + d) − (cx + d) ch
lim = lim = lim = c.
h→0 h h→0 h h→0 h

The next theorem follows in a fairly straightforward way from the prop-
erties of limits given in Theorem 4.2.1.
Theorem 5.3.2. If f and g are differentiable at x and c is a real number,
then the following functions are differentiable at x with derivatives as shown.
(f ± g)0 (x) = f 0 (x) ± g 0 (x)
(cf )0 (x) = cf 0 (x)
(f g)0 (x) = f 0 (x)g(x) + f (x)g 0 (x)
 0
1 −g 0 (x)
(x) =
g (g(x))2
 0
f f 0 (x)g(x) − f (x)g 0 (x)
(x) =
g (g(x))2
In the last two cases we also assume g(x) 6= 0.

Proof. See Adams, Section 3.3. 

The following now follows from the product rule and the Principle of
Induction.
Theorem 5.3.3. If f (x) = xn then f 0 (x) = nxn−1 .
 5.3. COMPUTING DERIVATIVES 69

Proof. The result is true for n = 1.

Assume it is true for some integer n, i.e. (xn )0 = nxn−1 .
Then
(xn+1 )0 = (xxn )0 = x0 xn + xnxn−1 = xn + xnxn−1 = (n + 1)xn .
Thus the corresponding result is true for n + 1.
This gives the result for all natural numbers n by the Principle of In-
duction. 

It follows that if f (x) = a0 + a1 x + a2 x2 + · · · + an xn then f 0 (x) =

a1 + 2a2 x + · · · + nan xn−1 .
We can also now compute derivatives of rational functions.
√
One can also show directly (as we noted before for x), that the de-
1
rivative of x1/n for x > 0 and n a natural number, is n1 x n −1 , see Adams
p107 Question 54. By using induction on m, one can then show that the
m
−1
derivative of xm/n for x > 0 and m, n natural numbers, is m nx
n . One
can also show directly by induction, using the derivative of 1/x, that for n
a natural number the derivative of x−n = (x−1 )n is −nx−n−1 , see Adams
p112, just before Example 8.
In a similar way, one can prove the general rule (xr )0 = rxr−1 for any
rational number r whenever the function xr is defined. The same result is
also true for any real number r, as we would expect by taking a sequence
of rational numbers rn → r. But this is best proved by first developing the
theory of logarithms and exponential functions. See in particular Adams
p179, just before Example 6.
In fact, natural (i.e. to base e) logarithms are defined as areas under
the graph of the function 1/x, see Adams p174.5 Then the derivative of
loge x is proved to be 1/x. From this one can then derive the usual rules for
derivatives of exponentials ax and general power functions xa , see Adams
p179 between Examples 5 and 6.
The proofs of the usual rules for the derivatives of the trigonometric
functions are given in Adams Section 2.5. They are not completely rigorous,
since the definition of sin and the other trigonometric functions was only
given informally, using diagrams, in Section P7.6
At this stage, we only use derivatives of such functions in the examples,
but not in the rigorous development of the subject.

5In Chapter 5 the area is defined more precisely as a Riemann integral.

6FTo do things rigorously, without using diagrams other than for motivation, one
approach is to define sin x by means of the power series in Adams p 540. Another is to
use complex numbers and define sin x as in Adams Exercise 17 page A-19.
The reason for rigour is not a matter of being obsessive compulsive about such things.
Diagrams and intuition are extremely useful and important, but they can be misleading.
Moreover, passing into the realm of complex numbers turns out to be extremely useful.
The analysis of periodic behaviour is fundamental throughout physics, engineering, mete-
orology, ..., and is essentially intractable if one does not pass through the world of complex
numbers!
70 5. DIFFERENTIATION

5.3.2. The
√ Chain Rule. In order to compute the derivatives of func-
tions such as 1 + x2 we need the Chain Rule. You have probably seen the
Chain Rule in the form
dy dy du
= ,
dx du dx
where u = g(x), y = f (u) and y = f (g(x)).
The informally stated motivation is that for a given value of x = a and
corresponding value u = g(a),
du
at x = a, u is changing times as fast as x,
dx
and
dy
at u = g(a), y is changing times as fast as u
du
du dy
(where is evaluated at x = a and is evaluated at u = g(a)). So that
dx du
dy du
at x = a, y is changing × times as fast as x.
du dx
In functional notation
(f ◦ g)0 (a)) = f 0 (g(a)) g 0 (a) or (f ◦ g)0 (x)) = f 0 (g(x)) g 0 (x).

An incorrect “proof” along these lines is often given for the chain rule
by writing
dy ∆y ∆y ∆u ∆y ∆u
= lim = lim = lim lim
dx ∆x→0 ∆x ∆x→0 ∆u ∆x ∆x→0 ∆u ∆x→0 ∆x
∆y ∆u dy dy
= lim lim =
∆u→0 ∆u ∆x→0 ∆x du dx
The second last step is “justified” by saying that ∆u → 0 as ∆x → 0. This
is all rather sloppy, because it is not clear what depends on what.
When one tries to fix it up, there arises a serious difficulty. Namely,
the increment ∆u = u(x + ∆x) − u(x) (which depends on ∆x) may be
zero although ∆x 6= 0. A trivial example is if u is the constant function.
There is the same difficulty when u is not constant, but there are points
x + ∆x arbitrarily close to x such that u(x + ∆x) = u(x) (such as with
u(x) = x2 sin(1/x) for x 6= 0 — see Example 5.3.5).
This becomes clearer when we write out the argument in a more precise
functional notation. See Adams Q46 p119.
We now state the Chain Rule precisely, and refer to Adams for a (correct)
proof.
Theorem 5.3.4 (Chain Rule). Assume the function f is differentiable at
g(x) and the function g is differentiable at x. Then the composite function
f ◦ g is differentiable at x and
(f ◦ g)0 (x) = f 0 (g(x)) g 0 (x).
(We also assume that g is defined in some open interval containing x and
f is defined in some open interval containing g(x), although this can be
generalised a bit.)
 5.3. COMPUTING DERIVATIVES 71

Proof. See Adams p118. To help understand the proof, note that the
“error” function E(k) is the difference between the slope of the line through
the points (u, f (u)) and (u + k, f (u + k)) on the graph of f , and the slope
of the tangent at the point (u, f (u)) and (u + k, f (u + k)) on the graph of
f . (Draw a diagram like the first one in this chapter.) 

Example 5.3.5.

Figure 2. The graph of f if f (x) = x2 sin(1/x) at x 6= 0

and f (0) = 0.

We can now compute the derivative of the function

(
x2 sin x1 x 6= 0
f (x) = .
0 x=0
If x 6= 0 then by the Product and Chain Rules (and using the fact sin0 y =
cos y)
   0
0 2 0 1 2 0 1 1
f (x) = (x ) sin + x sin
x x x
  
1 2 1 1
= 2x sin + x cos − 2
x x x
1 1
= 2x sin − cos .
x x
We see that f 0 (x) has no limit as x → 0, since the first term approaches
zero but the second “oscillates” between ±1.
However, f is differentiable at 0, and in fact f 0 (0) = 0. This is in fact not
surprising if we look at the graph. Any line passing through the points (0, 0)
and (h, h2 sin h1 ) on the graph lies in the region between the two parabolas
corresponding to ±x2 . it is thus geometrically clear that the slope of this
line approaches 0 as h → 0.
Analytically,
f (h) − f (0) h2 sin h1 − 0 1
f 0 (0) = lim = lim = lim h sin = 0.
h→0 h h→0 h h→0 h
72 5. DIFFERENTIATION

The last limit follows easily from the Squeeze Theorem in Adams p69 applied
with ±x. (This Squeeze Theorem for limits follows easily from the Squeeze
Theorem 2.3.7 here, for sequences. Exercise.)
Thus f is differentiable for all x, but the derivative is not continuous at
0.

5.4. Maximum and Minimum Values

The relationship between derivatives and maximum and
minimum points is given.
Definition 5.4.1. A function f has a maximum value (minimum value)
f (x0 ) at the maximum point (minimum point) x0 ∈ D(f ) if
f (x) ≤ f (x0 ) (f (x) ≥ f (x0 ))
for all x ∈ D(f ).
The function has a local maximum value (local minimum value) f (x0 ) at
the local maximum point (local minimum point) x0 ∈ D(f ) if there exists
an open interval N containing x0 such that
f (x) ≤ f (x0 ) (f (x) ≥ f (x0 ))
for all x in N ∩ D(f ).

Figure 3. Which are the (local) maximum and minimum points?

In Figure 3, f has a maximum at x3 , a minimum at b, local maxima at

a, x1 , x3 and local minima at x0 , x2 , b.
We saw in Theorem 3.4.1 that a continuous function f defined on a
closed bounded interval always has a maximum and a minimum value.
If the domain of f is an interval and f has a local maximum or minimum
at x there are three logical possibilities:
• x is an endpoint of the domain of f ;
• x is not an endpoint and f 0 (x) does not exist ;
• x is not an endpoint and f 0 (x) does exist.
Theorem 5.4.2. Suppose f has a local maximum or minimum at an
interior point x0 and that f 0 (x0 ) exists. Then f 0 (x0 ) = 0.
 5.5. MEAN VALUE THEOREM 73

Proof. Suppose f has a local maximum at the non endpoint x0 (the

proof for a local minimum is similar). Then for some h0 > 0,
|h| < h0 implies f (x0 ) ≥ f (x0 + h).
Hence,
f (x0 + h) − f (x0 )
(27) ≤ 0 if 0 < h < h0 .
h
and
f (x0 + h) − f (x0 )
(28) ≥ 0 if − h0 < h < 0.
h
We know that the derivative at x0 exists and hence
f (x0 + h) − f (x0 ) f (x0 + h) − f (x0 )
lim and lim
h→0+ h h→0− h
both exist and are equal. But the first limit is ≤ 0 from(28) and the second
is ≥ 0 from(27). Hence the derivative must be 0. 
It is not true that if f 0 (x0 ) = 0 then f must have a local maximum or
minimum at x0 . Consider f (x) = x3 at 0.
We say a function f has a critical value f (x0 ) at the critical point x0 ∈
D(f ) if f 0 (x0 ) = 0.
It is not true that if f has a local maximum or minimum at an endpoint
then the derivative is zero there. Just consider f (x) = x on [0, 1].

5.5. Mean Value Theorem

The Mean Value Theorem is proved. This is used to bound
the difference between values of a function, and to prove
the Constancy Theorem and Rolle’s Theorem. The rela-
tionship between the sign of the derivative and the mono-
tone behaviour of a function is developed.
The Mean Value Theorem says that the slope of the line joining two
points (a, f (a)) and (b, f (b)) on the graph of a differentiable function f is
equal to the slope of the tangent at the point (c, f (c)) for some c between a
and b. This is geometrically clear for any reasonable function whose graph
we can draw. We want to show that it follows rigorously from the definition
of differentiable (this then will be another justification that our definition
correctly captures our informal notions of differentiability).
From the diagram, we expect that c will correspond to some point on the
graph of f at maximum vertical distance from the line joining (a, f (a)) and
(b, f (b)). Since the equation of this line is y = f (a) + f (b)−f
b−a
(a)
(x − a), this
f (b)−f (a)
vertical distance is given by f (x) − f (a) − b−a (x − a). This motivates
the following proof.
Theorem 5.5.1 (Mean Value Theorem). Suppose f is continuous on a
closed bounded interval [a, b] and is differentiable on the open interval (a, b).
Then there exists c ∈ (a, b) such that
f (b) − f (a)
f 0 (c) = .
b−a
74 5. DIFFERENTIATION

Figure 4. Example of the Mean Value Theorem.

Proof. Consider the function g given by

f (b) − f (a)
g(x) = f (x) − f (a) − (x − a).
b−a
Then g(a) = g(b) = 0 and g is continuous on [a, b]. (Why? )
We claim that g 0 (c) = 0 for some interior point c. To prove the claim
consider three cases, at least one of which must occur:
(1) g(x) = 0 for all x in [a, b]. Since g is a constant functon it follows
immediately from the definition of differentiation that g 0 (c) = 0 for
every c in (a, b).
(2) g(x) > 0 for some x in [a, b]. By Theorem 3.4.1 the function g has
at least one maximum point c in [a, b]. Since g(c) > 0 (why? ), and
since g(a) = g(b) = 0, if follows that c 6= a and c 6= b. Hence c is
an interior point and so g 0 (c) = 0 by Theorem 5.4.2.
(3) g(x) < 0 for some x in [a, b]. Then g has at least one minimum
point c in [a, b], this is an interior point, and g 0 (c) = 0, by a similar
argument to case (2). Write it out yourself !
Since g 0 (c) = 0, and by the rules for differentiation
f (b) − f (a)
g 0 (c) = f 0 (c) − ,
b−a
this proves the theorem. 

Remark 5.5.2. The proof used the Max-Min Theorem 3.4.1, which in
turn required the Completeness Axiom. Give an example that shows the
result would not be true if we did not assume the Completeness Axiom.
(The fact that our proof relied on the Completeness Axiom does answer the
question, why? )
HINT: See Remark 3.4.2. 

Remark 5.5.3. Rolle’s Theorem, see Adams p140, says that if f is

continuous on [a, b] and differentiable on (a, b), and f (a) = f (b) = 0, then
f 0 (x0 ) = 0 for some x0 ∈ (a, b). It is a particular case of the Mean Value
Theorem.
We gave a direct proof of the Mean Value Theorem, whereas Adams first
proved Rolle’s Theorem. 
 5.5. MEAN VALUE THEOREM 75

Corollary 5.5.4. Suppose f is continuous on an interval and |f 0 (x)| ≤

K at every interior point in the interval. (The interval may be open, closed,
bounded or unbounded.) Then
|f (x1 ) − f (x2 )| ≤ K|x1 − x2 |
for all x1 , x2 in the interval.

Proof. Suppose x1 < x2 . (The proof is similar if x2 < x1 and the result
is trivial if x1 = x2 .)
By the Mean Value theorem there exists a number c between x1 and x2
such that
f (x1 ) − f (x2 ) = f 0 (c)(x1 − x2 ),
and so
|f (x1 ) − f (x2 )| = |f 0 (c)| |x1 − x2 | ≤ K |x1 − x2 |.


Corollary 5.5.5 (Constancy Theorem). If f is continuous on an in-

terval and f 0 (x) = 0 at every interior point in the interval then f is constant
on the interval. (The interval I may be open, closed or unbounded, at either
end.)

Proof. Choose a point c ∈ I and let C = f (c). We want to show

f (x) = C for every x ∈ I.
But |f (x) − f (c)| = 0 by the previous corollary, and so f (x) = C. 

The corollary is not true if the domain of f is a finite union of more than
one interval. In this case the function is constant on each interval, but the
constant may depend on the interval.
A useful application of Corollary 5.5.5 is to prove that complicated ex-
pressions are equal. For example, to prove that f (x) = g(x) for all x in
some interval, it is sufficient to prove that the functions f and g are equal
at a single point c and that their derivatives are equal everywhere.
To see this apply the corollary to the function f (x)−g(x). The derivative
is zero and so the function is constant; but the constant is zero since f (c) −
g(c) = 0.
The Mean Value Theorem leads to a result which enables us to decide
where a function is increasing or decreasing.
Definition 5.5.6. We say a function f is
increasing if x1 < x2 implies f (x1 ) < f (x2 ),
decreasing if x1 < x2 implies f (x1 ) > f (x2 ),
non-decreasing if x1 < x2 implies f (x1 ) ≤ f (x2 ),
non-increasing if x1 < x2 implies f (x1 ) ≥ f (x2 ).

See Adams p139 for some diagrams.

76 5. DIFFERENTIATION

Theorem 5.5.7. Suppose that f is continuous on an interval I and

differentiable at every interior point of I . Then
f 0 (x) > 0 for every interior point x implies f is increasing on I,
f 0 (x) < 0 for every interior point x implies f is decreasing on I,
f 0 (x) ≥ 0 for every interior point x implies f is non-decreasing on I,
f 0 (x) ≤ 0 for every interior point x implies f is non-increasing on I.

Proof. Suppose x1 < x2 are points in I. By the Mean Value Theorem,

f (x2 ) − f (x1 ) = f 0 (x0 )(x2 − x1 )
for some x0 between x1 and x2 .
If f 0 (x) > 0 for all x ∈ I then this implies f (x1 ) < f (x2 ), and similarly
for the other three cases. 
Note that if a function is increasing on an interval then it does not follow
that f 0 (x) > 0 for every interior point x. For example, if f (x) = x3 then f
is increasing on the interval R, but f 0 (0) = 0.
However, if f is increasing, or even just non-decreasing, on an interval,
then it does follow that f 0 (x) ≥ 0 for all x in the interval. This also follows
from the Mean Value Theorem. (Exercise).
Note also that the first of the four cases in the theorem applies to f (x) =
x3 on the interval [0, 1]. Why?

5.6. FPartial derivatives

Suppose, for simplicity, we have a function f (x, y) defined on an open
rectangle A, where
A = (a, b) × (c, d) = (x, y) ∈ R2 : a < x < b, c < y < d .


Draw a diagram.
The partial derivative with respect to y at (x0 , y0 ) ∈ A is defined by
∂f f (x0 , y0 + h) − f (x0 , y0 )
(x0 , y0 ) = lim .
∂y h→0 h
Think of the line parallel to the y-axis through the point (x0 , y0 ), and
think of f as a function with domain restricted to this line, i.e. f is a function
of y with x fixed to be x0 . Draw a diagram. Then ∂f /∂y(x0 , y0 ) is just the
ordinary derivative with respect to y.
If we know that
∂f
(x, y) ≤ K
∂y
at every point (x, y) in the open rectangle A, then it follows from Corol-
lary 5.5.4 that
|f (x, y1 ) − f (x, y2 )| ≤ K|y1 − y2 |
for every (x, y1 ), (x, y2 ) ∈ A.
We will use this in the proof of Theorem 7.3.1.
 CHAPTER 6

Integration

The main references in Adams are Sections 5.1–5.5 and Appendix IV.
Integration allows us to find areas and volumes bounded by curves and surfaces.
It is rather surprising at first, but there is a close relationship between inte-
gration and differentiation; each is the inverse of the other. This is known as the
Fundamental Theorem of Calculus. It allows us to find areas by doing the reverse
of differentiation.
Integrals are also used to express lengths of curves, work, energy and force,
probabilities, and various quantities in economics, for example.

6.1. Introduction
The topic of this chapter is the concept of “area under a graph” in a quantitative
sense and the elucidation of some of its properties.
Everyone would be happy with the definition: “the area of a rectangle is the
product of its length and breadth”. The problem is more difficult with more com-
plicated plane figures. The circle, for example, has “area πr2 ”; but is this “area”
the same concept as that applied to rectangles?
In everyday life one often needs only an approximation to the area of, say, a
country or a field. If pressed one would calculate it approximately by filling it as
nearly as possible with rectangles and summing their area. This is very close to
what we do here in giving a precise definition of the concept of area.

6.2. The Riemann integral

The (definite) Riemann integral is defined in terms of upper and
lower sums. It is shown that continuous functions on closed
bounded intervals are integrable.
Throughout this section,1 unless stated otherwise, f is a continuous function
defined on a closed bounded interval [a, b].
We aim to define the “area under the graph of f ”. That is we wish to attach
a number to the shaded region in the following diagram, which is its “area”, and
which has the properties that we normally associate with “area”.
The basic properties that we want of this “area” are
• the area of a rectangle should be “length times breadth”;
• the area of non-overlapping regions is the sum of their areas;
• if one region is contained in another the area of the first is ≤ the area of
the second.
Before we begin, a preliminary comment: a given function f on [a, b] may take
values both positive and negative, as in the next diagram.
The concept of area which we are about to define will treat the regions below
the x-axis as negative. The concept we define is in this sense not quite what one
1Some of the material in this section closely follows notes of Bob Bryce from a previous

first year honours level course.

77
78 6. INTEGRATION

Figure 1. Suppose the domain of f is the interval [a, b].

Rb
Then a f is the area of the shaded region.

Figure 2. Suppose the domain of f is the interval [a, b].

Rb
Then a f is the area of the shaded region above the axis
minus the area below the axis.

might expect, though it agrees with our intuition in the case when f (x) ≥ 0 for all
x ∈ [a, b].

We begin by defining a partition of [a, b]; this is simply a finite set of points
from [a, b] which includes a and b. Thus P = {0, 1/4, 1} is a partition of [0, 1], and
P = {−1, − 1/2, − 1/4, 3/4, 1} is a partition of [−1, 1].
The general notation for a partition P of [a, b] is

P = {a = x0 , x1 , x2 , . . . , xn = b }.

We will assume always that a = x0 < x1 < . . . < xn = b. The ith subinterval is
[xi−1 , xi ] and its length is defined to be

∆xi := xi − xi−1 .
 6.2. THE RIEMANN INTEGRAL 79

Figure 3. In this case the partition P divides the domain

into three intervals. The upper sum U (P, f ) is the sum of the
areas of the three rectangles, but with the first two counted
positively and the third negatively. What about L(P, f )?

With each partition P of [a, b] we associate the so-called upper and lower sums.
To define these we need the following notation: write

Mi = max { f (x) : xi−1 ≤ x ≤ xi }, 1 ≤ i ≤ n;

mi = min { f (x) : xi−1 ≤ x ≤ xi }, 1 ≤ i ≤ n.

That is, Mi is the maximum value and mi the minimum value of f on the ith
sub-interval [xi−1 , xi ] of the partition. These exist because f is continuous on the
closed bounded interval [xi−1 , xi ].
The upper sum of f over P is defined by
n
X
U (P, f ) = Mi ∆xi ,
i=1

and the lower sum of f over P is defined by

n
X
L(P, f ) = mi ∆xi .
i=1

(See Adams p289 for a discussion of the summation notation.) Roughly speak-
ing L(P, f ) is the sum of the areas of all the rectangles whose bases are the sub-
intervals [xi−1 , xi ] and which just fit under the graph of f . Similarly U (P, f ) is the
sum of the areas of all the rectangles whose bases are the sub-intervals [xi−1 , xi ]
and which just contain the graph of f . At least this is the case when f (x) ≥ 0. In
other cases the interpretation is less simple. Various possibilities are illustrated in
the diagrams below.

Example 6.2.1. Let f (x) = 1 − 2x on [0, 1], and let P = {0, 1/4, 1/3, 2/3, 1}.
Find L(P, f ) and U (P, f ).
80 6. INTEGRATION

Figure 4. Sketch the graph of the function f (x) = 1 − 2x.

Here
M1 = 1 m1 = 1/2 ∆x1 = 1/4
M2 = 1/2 m2 = 1/3 ∆x2 = 1/12
M3 = 1/3 m3 = −1/3 ∆x3 = 1/3
M4 = −1/3 m4 = −1 ∆x4 = 1/3
and so
4  
X 1 1 1 1 1 −1 −1 7
L(P, f ) = mi ∆xi = · + · + + =−
i=1
2 4 3 12 3 3 3 24
4  
X 1 1 1 1 1 1 1 7
U (P, f ) = Mi ∆xi = 1 · + · + · + − =
i=1
4 2 12 3 3 3 3 24

Exercise 6.2.2. Let f (x) = cos x on [−π/2, π] and P = {−π/2, −π/4, 0, π/2, π}.
Show that
1 π π
L(P, f ) = √ · −
2 4 2
and
1 π 3π
U (P, f ) = √ · + .
2 4 4

We now develop the properties of upper and lower sums that we need.
Lemma 6.2.3. Let f be a continuous function on [a, b] and P be a partition of
[a, b]. Then L(P, f ) ≤ U (P, f ).

Proof. Since mi ≤ Mi for all i, and since xi − xi−1 > 0,

mi (xi − xi−1 ) ≤ Mi (xi − xi−1 ).
Summing,
L(P, f ) ≤ U (P, f )
as required. 

Draw a diagram and you will see how obvious this result is.
The following lemma is obvious from a few diagrams. We need a proof that
does not rely on particular examples. This is straightforward.
 6.2. THE RIEMANN INTEGRAL 81

Lemma 6.2.4. Let f be a continuous function on [a, b]. Let P1 , P2 be two

partitions of [a, b] with P1 ⊂ P2 . (We say that P2 is a refinement of P1 .) Then
L(P1 , f ) ≤ L(P2 , f ) and U (P2 , f ) ≤ U (P1 , f ).

Proof. We can get P2 from P1 by successively adding one new point at a time.
If therefore, we can show that adding one new point to a partition has the effect of
not decreasing the lower sum and not increasing the upper sum, we will be done.
In other words we might as well suppose that P2 is obtained from P1 by adding one
more point.
Suppose therefore that P1 = {a = x0 , x1 , x2 , . . . , xn = b} and that P2 =
P1 ∪ {x} 2 with x ∈ (xi−1 , xi ). Let Mj , mj (1 ≤ j ≤ n) be the maximum and
minimum values of f on [xj−1 , xj ]. Let M 0 , m0 be the maximum and minimum
values for f on [xi−1 , x]; and M 00 , m00 be the maximum and minimum values for f
on [x, xi ]. Note that
m0 ≥ mi , m00 ≥ mi
and
M 0 ≤ Mi , M 00 ≤ Mi
because when passing from an interval to a subinterval, the minimum value cannot
decrease and the maximum value cannot increase.

Figure 5. The minimum and maximum values of f on

[xi−1 , x] are m0 and M 0 respectively, and on [x, xi ] are m00
and M 00 respectively.

Then
L(P2 , f ) − L(P1 , f ) = m0 (x − xi−1 ) + m00 (xi − x) − mi (xi − xi−1 )
≥ mi (x − xi−1 ) + mi (xi − x) − mi (xi − xi−1 )
= 0,
and
U (P1 , f ) − U (P2 , f ) = Mi (xi − xi−1 ) − M 0 (x − xi−1 ) − M 00 (xi − x)
≥ Mi (xi − xi−1 ) − Mi (xi − x) − Mi (xi − x)
= 0.
That is
L(P1 , f ) ≤ L(P2 , f ) and U (P2 , f ) ≤ U (P1 , f ).

2This notation just means that P is the union of the set P and the set {x} containing
2 1
the single point x.
82 6. INTEGRATION

In words: refining a partition increases lower sums and decreases upper sums.
Corollary 6.2.5. If f is continuous on [a, b] and if P1 , P2 are arbitrary par-
titions of [a, b], then L(P1 , f ) ≤ U (P2 , f ).

Proof. The partition P obtained by using all the points of P1 and P2 together,
i.e. P is the union of P1 and P2 , is a refinement of both P1 and P2 . Hence
L(P1 , f ) ≤ L(P, f ) ≤ U (P, f ) ≤ U (P2 , f ),
by Lemma 6.2.3 and Lemma 6.2.4. 
In other words : every lower sum is less than or equal to every upper sum.
The important consequence we need is this : since the lower sums L(P , f ) are
all bounded above (by every upper sum in fact) the set of lower sums has a least
upper bound. Similarly the set of upper sums is bounded below (by any lower sum)
so the set of upper sums has a greatest lower bound. We define the lower integral
of f from a to b and the upper integral of f from a to b by
Z b
L f := lub{ L(P, f ) : P is a partition of [a, b] }
a
Z b
U f := glb{ U (P, f ) : P is a partition of [a, b] }
a
respectively.
The next lemma just uses the fact that every lower sum is ≤ every upper sum.
It will soon be replaced by the stronger result that (for continuous functions) the
lower and upper integrals are in fact equal.
Rb Rb
Lemma 6.2.6. Let f be a continuous function on [a, b]. Then L a f ≤ U a f .

Proof. Let P be a partition of [a, b].

Rb
Since U (P, f ) is an upper bound for all lower sums, and since L a f is the least
upper bound, it follows that
Z b
L f ≤ U (P, f ).
a
Rb
Since this is true for every partition P , L a f is thus a lower bound for the set of
Rb
all upper bounds. Since U a f is the greatest lower bound, it follows that
Z b Z b
L f ≤U f.
a a


Remark 6.2.7. F Everything we have done so far can also be done with an
arbitrary bounded 3 function f defined on [a, b], except that we must define
Mi = lub{ f (x) : xi−1 ≤ x ≤ xi }, 1 ≤ i ≤ n,
mi = glb{ f (x) : xi−1 ≤ x ≤ xi }, 1 ≤ i ≤ n.
Lemma 6.2.3, Lemma 6.2.4, Corollary 6.2.5 and Lemma 6.2.6 are still valid, with
similar proofs as for continuous functions, but with “min” replaced by “glb” and
“max” replaced by “lub”.
3A function is bounded if there exist numbers A and B such that A ≤ f (x) ≤ B for
every x in the domain of f . Thus any continuous function defined on [a, b] is bounded.
But the function f , with f (x) = 1/x for x 6= 0 and f (0) = 0, is not bounded on its domain
R.
 6.2. THE RIEMANN INTEGRAL 83

Definition 6.2.8. A bounded function f defined on [a, b] is integrable (in the

sense of Riemann) if
Z b Z b
L f =U f.
a a
Rb Rb
We call L a f and U a f respectively the lower and upper integral of f over [a, b].
For an integrable function we denote the common value of the upper and lower
Rb
integrals by a f and call it the (definite) integral of f over [a, b].
Rb Rb
Note that L a f and U a f are numbers.
Remark 6.2.9. F There is another type of integral called the Lebesgue integral.
This is much more difficult to define, but it is much more powerful (more functions
are integrable) and it has better properties (under very general conditions, if a se-
quence of functions fn (x) converges to f (x) for every x, then the Lebesgue integrals
of fn converge to the Lebesgue integral of f ). Such a convergence result is true
for Riemann integration only if the functions converge in a rather strong sense. If
a function is Riemann integrable then it is Lebesgue integrable (and the integrals
agree), but the converse is not true.
For many applications, Riemann integration is sufficient, but for more sophis-
ticated applications one needs the Lebesgue integral. There is a third year course
on measure theory and Lebesgue integration.
The case we will be mainly interested in is when f is continuous. In The-
orem 6.2.10 we prove the important result that every continuous function on a
closed bounded interval is integrable.
In general it is not the case that upper and lower integrals are equal. For
example, consider the function f defined on [0, 1] by
(
0 x is irrational,
f (x) =
1 x is rational.

Then, whatever partition P of [0, 1] we have, Mi = 1 and mi = 0 for every i, since

every interval [xi−1 , xi ] contains both rational and irrational points. Hence
Z 1 Z 1
L f = 0 and U f = 1.
0 0

Theorem 6.2.10. Let f be continuous on [a, b]. Then f is integrable.

Proof. Suppose f is continuous on [a, b]. Suppose ε > 0.

We will first show there exists some partition P (which may depend on ε) such
that
(29) U (P, f ) − L(P, f ) < ε.
Since f is uniformly continuous on [a, b] by Theorem 4.5.2, there exists δ > 0
such that
ε
|x1 − x2 | < δ =⇒ |f (x1 ) − f (x2 )| < .
b−a
(We will see the reason for taking ε/(b − a) in a moment.)
Now let P = {a = a0 , a1 , a2 , . . . , aN = b} be any partition of [a, b] such
that the difference between consecutive points in P is < δ. Then by the above
84 6. INTEGRATION

implication the difference between the maximum value Mi and the minimum mi of
the function f on the ith interval must be < ε/(b − a). Hence
N
X N
X
U (P, f ) − L(P, f ) = Mi ∆xi − mi ∆xi
i=1 i=1
N
X
= (Mi − mi ) ∆xi
i=1
ε
< (∆x1 + · · · + ∆xN )
b−a
ε
= (b − a) = ε.
b−a
This proves (29).
From the definition of the lower and upper integrals, and Lemma 6.2.6,
Z b Z b
L(P, f ) ≤ L f ≤U f ≤ U (P, f ).
a a

Since the difference between the outer two terms is < ε by (29), the difference
between the inner two terms is also < ε. That is
Z b Z b
U f −L f < ε.
a a
Rb Rb
Since this holds for every ε > 0 it follows that U a
f =L a
f. 

6.3. Riemann sums

The connection between Riemann sums and the Riemann integral is established.

If f is a continuous function on [a, b] and P is a partition, then the upper and

lower sums can be written in the form
n
X
U (P, f ) = f (ui ) ∆xi ,
i=1
Xn
L(P, f ) = f (li ) ∆xi .
i=1

where ui and li are points in the ith interval [xi−1 , xi ] for which f takes its maximum
and minimum values respectively. More generally, we can define a general Riemann
sum corresponding to the partition P by
N
X
R(P, f ) = f (ci ) ∆xi ,
i=1

where each ci is an arbitrary point in [xi−1 , xi ]. Note that this notation is a little
imprecise, since R(P, f ) depends not only on the partition P , but also on the points
ci chosen in each of the intervals given by P .
Note that
(30) L(P, f ) ≤ R(P, f ) ≤ U (P, f ).
Let the maximum length of the intervals in a partition P be denoted by kP k.
 6.4. PROPERTIES OF THE RIEMANN INTEGRAL 85

Theorem 6.3.1. Z b
lim R(P, f ) = f.
kP k→0 a
More precisely, for any ε > 0 there exists a number δ > 0 (which may depend on
ε) such that
Z b
whenever kP k < δ then R(P, f ) − f < ε.
a

Proof. The proof of Theorem 6.2.10 in fact showed that if kP k < δ then
U (P, f ) − L(P, f ) < ε.
Since
L(P, f ) ≤ R(P, f ) ≤ U (P, f )
and Z b
L(P, f ) ≤ f ≤ U (P, f )
a
it follows that Z b
R(P, f ) − f < ε.
a


Notation 6.3.2. We often use the notation

Z b Z b
f (x) dx for f.
a a
Rb
Note that f (x) dx is a number, not a function of x. It has exactly the same
a
Rb PN PN
meaning as f (y) dy, just as i=1 f (ci ) ∆xi and j=1 f (cj ) ∆xj mean the same
a
thing. We say x is a “dummy” variable.
Rb
You can informally think of a f (x) dx as the sum of the “areas” of an infinite
number of rectangles of height f (x) and “infinitesimal” width “dx”. More precisely,
from the previous theorem,
Z b N (P )
X
f (x) dx = lim f (ci ) ∆xi .
a kP k→0
i=1
(We write N (P ) to emphasise the fact that the number of points in the partition
depends on P .)

6.4. Properties of the Riemann integral

The basic linearity and order properties of the Riemann integal
are developed. The mean value theorem for integrals is proved.
The extension of these results to piecewise continuous functions
is noted.

Remark 6.4.1. The (easy) theorems in this section apply more generally with
minor modifications, provided the functions are Riemann integrable, and not neces-
sarily continuous. For example, in Theorem 6.4.3 we need to replace the minimum
m by the glb of the set of values, and similarly for the maximum M .
In particular, piecewise continuous functions (see Adams p309) on a closed
bounded interval are integrable, and have the same properties as below. This
essentially follows from writing each integral as a sum of integrals over intervals on
which all the relevant functions are continuous.
86 6. INTEGRATION

Theorem 6.4.2. If f, g are continuous functions on [a, b] and c, d are real

numbers, then
Z b Z b Z b
(31) (cf + dg) = c f +d g,
a a a
Z b Z b
(32) f (x) ≤ g(x) for all x ∈ [a, b] =⇒ f≤ g,
a a
Z b Z b
(33) f ≤ |f |.
a a

Proof. The main point in the proofs of (31) and (32) is that similar properties
are true for the Riemann sums used to define the integrals. See Adams, A-30,
Exercise 6.
To prove (33), note that
−|f (x)| ≤ f (x) ≤ |f (x)|
for all x. From (32)
Z b Z b Z b
−|f | ≤ f≤ |f |.
a a a
From (31) this gives
Z b Z b Z b
− |f | ≤ f≤ |f |.
a a a
This implies (33). 

Theorem 6.4.3. If f is continuous on [a, b] with minimum and maximum val-

ues m and M then
Z b
(34) m(b − a) ≤ f ≤ M (b − a).
a

Proof. Consider the partition P = {a, b} containing just the two points a
and b. Since
L(P, f ) = m(b − a), U (P, f ) = M (b − a),
and
Z b Z b Z b
L(P, f ) ≤ L f= f =U f ≤ U (P, f ),
a a a
the result follows. 

Theorem 6.4.4. Suppose a ≤ c ≤ b. Then

Z a Z b
(35) f =− f,
b a
Z a
(36) f = 0,
a
Z c Z b Z b
(37) f+ f= f.
a c a
 6.5. FUNDAMENTAL THEOREM OF CALCULUS 87

Proof. TheR a first is really a definition. It also follows if we use the same
definition of b f as in the case b < a, but allow “decreasing” partitions where
∆ xi < 0.
The second is again by definition. It also follows if we use the same definition
of the integral as when the endpoints are distinct, except that now the points in
any “partition” are all equal and so ∆ xi = 0.
The third is straightforward. See **** for details. 

Remark 6.4.5. If we allow b ≤ a as well as a < b, then (33) should be replaced

by
Z b Z b
(38) f ≤ |f |
a a
Exercise.

Theorem 6.4.6 (Mean Value Theorem for Integrals). If f is continuous on

[a, b] then there exists c ∈ [a, b] such that
Z b
(39) f = (b − a)f (c).
a

Proof. Choose l and u to be minimum and maximum points for f on [a, b].
Then from (34) it follows that
Rb
f
f (l) ≤ a ≤ f (u),
b−a
By the Intermediate Value Theorem applied to the function f on the interval [l, u]
or [u, l] (depending on whether l ≤ u or u ≤ l), there exists c between l and u such
that Rb
f
f (c) = a .
b−a
This gives the result. 

6.5. Fundamental Theorem of Calculus

The relationship between integration and differentiation is developed.

The following theorem essentially says that differentiation and integration are
reverse processes. Rx
In the first part of the theorem we consider the integral a f 4 as a function of
the endpoint x (we allow x ≤ a as well as x > a) and prove: the derivative of the
integral of f gives back f .
Rb
In the second part, we are saying that in order to compute a f it is sufficient
to find a function G whose derivative is f and then compute G(b) − G(a).

4In the theorem we could also write

Z x
d
f (t) dt = f (x).
dx a
The variable t is a dummy variable, and we could have used y or anything else instead.
But it is “good practice” not to use x instead of t in this case, since we are already using
x here to represent the endpoint of the interval of integration.
88 6. INTEGRATION

To put the second assertion in a form that looks more like the “reverse” of the
first, we could write it in the form
Z x
G0 = G(x) − G(a),
a

provided G0 is a continuous function on I. We could even use f instead of G and

then get
Z x
f 0 = f (x) − f (a),
a

provided f 0 is continuous on I. The integral of the derivative of f gives back f (up

to the constant f (a)).

Theorem 6.5.1 (Fundamental Theorem of Calculus). Suppose that f is con-

tinuous on some interval I (not necessarily closed and bounded) and that a ∈ I.
Then
Z x
d
f = f (x).
dx a

If G0 (x) = f (x) for all x ∈ I then

Z b
f = G(b) − G(a).
a

Proof.

R x+h
Figure 6. The area of the shaded region is x f.
 6.5. FUNDAMENTAL THEOREM OF CALCULUS 89

For the first assertion we have

Z x R x+h Rx
d a
f− a f
f = lim
dx a h→0 h
R x+h
f
= lim x from (37)
h→0 h
hf (c(h))
= lim for some c = c(h) between x and x + h,
h→0 h
depending on h, by the Mean Value Theorem
for integrals.
= lim f (c(h))
h→0
= f (x) since f is continuous at x
and c lies between x and x + h.

d
For the second assertion, suppose G(x) = f (x) on the interval I.
Z xdx
d
But we have just seen that f = f (x). It follows that the derivative of
dx a
the function, given by Z x
G(x) − f,
a
is G0 (x) − f (x) = 0 on the interval I. Thus this function is constant on I by
Corollary 5.5.5.
Setting x = 0 we see that the constant is G(a). Hence
Z x
G(x) − f = G(a)
a
for all x ∈ I. Taking x = b now gives the second assertion. 
 CHAPTER 7

FDifferential Equations

7.1. Overview
The differential equation

dy
(40) = f (x, y)
dx

requires that the gradient of the function y = y(x) at each point (x, y) on its graph
should equal f (x, y) for the given function f .
Suppose that at each point (x, y) on the x − y plane we draw a little line whose
slope is f (x, y); this is the slope field. Then at every point on the graph of any
solution to (40), the graph should be tangent to the corresponding little line. In
the following diagram we have shown the slope field for f (x, y) = y + cos x and the
graph of three functions satisfying the corresponding differential equation.

Figure 1. Slope field for the function y + cos x, and the

dy
graph of three funcions y = y(x) satisfying = y + cos x.
dx

It is plausible from the diagram that for any given point (x0 , y0 ) there is exactly
one solution y = y(x) satisfying y(x0 ) = y0 . This is indeed the case here, and is
true under fairly general conditions.
91
92 7. FDIFFERENTIAL EQUATIONS

But it is not always true. For example, if f (x, y) = y 2/3 then there is an infinite
set of solutions satisfying y(0) = 0. Namely, for any real numbers a ≤ 0 ≤ b,
 (x−a)3
 27
 x≤a
y= 0 a≤x≤b
 (x−b)3

27 x ≥b

is a solution, (check it). See the following diagram. The problem here is that
although f (x, y) is continuous everywhere, (∂/∂y)f (x, y) = 2y −1/3 /3 is not contin-
uous on the x-axis. Notice that the slope lines on the x-axis are horizontal.

Figure 2. Slope field for the function (x, y) 7→ y 2/3 . (In this
case the slope field does not depend on x.)

If the function f has even worse behaviour, there may be no solution at all.

In the two simple examples we just gave, we could write out the solutions in
terms of standard functions. But in practice, this is almost never the case. The
solutions of differential equations almost invariably cannot be expressed in terms
of standard functions. In fact, one of the most useful ways to introduce new and
useful functions is to define them as the solutions of certain differential equations.
But in order to do this, we first need to know that the differential equations have
unique solutions if we specify certain “initial conditions”. This is the main result
in this chapter.

The point to this chapter is to prove the Fundamental Existence and Uniqueness
Theorem for differential equations of the form (40). Such an equation is called first-
order, since only the first order derivative of y occurs in the differential equation.
Differential equations of the form (40) are essentially the most general first order
differential equation.

The following remark justifies that we are about to prove a major result in
mathematics!!.
 7.2. OUTLINE OF PROOF OF THE EXISTENCE AND UNIQUENESS THEOREM 93

Remark 7.1.1. F A system of first order differential equations for the depen-
dent variables y1 , . . . , yn is a set of differential equations of the form
dy1
= f1 (x, y1 , . . . , yn )
dx
dy2
= f2 (x, y1 , . . . , yn )
dx
..
.
dyn
= fn (x, y1 , . . . , yn )
dx
which are meant to be satisfied simultaneously by functions y1 = y1 (x), y2 =
y2 (x), . . . , yn = yn (x). Here the functions f1 , f2 , . . . , fn are given. If n = 2 you
can visualise this as in the one dimensional case, by considering three axes la-
beled x, y1 , y2 . The solution to a differential equation in this case will be repre-
sented by the graph (curve) over the x axis which for each point x gives the point
(x, y1 (x), y2 (x)).
A very similar proof as for a single differential equation, gives the analogous
Fundamental Existence and Uniqueness Theorem for a system of first-order differ-
ential equations.
A differential equation which involves higher derivatives can be reduced to a
system of differential equations of first order (essentially by introducing new vari-
ables for each of the higher order derivatives). Thus the Existence and Uniqueness
Theorem, suitably modified, applies also to higher order differential equations. In
fact it even applies to systems of higher order differential equations in a similar
manner!.

7.2. Outline of proof of the Existence and Uniqueness theorem

Since I am realistic enough to know that not everyone is going to study the
proof in Section 7.3 in detail (but it is not that difficult to follow), I will provide
you here with an overview. (After that, hopefully you will then be inspired to work
through the details.)
We want to prove that the the initial value problem
dy
= f (x, y)
(41) dx
y(x0 ) = y0
has exactly one solution under certain (general) assumptions.
The assumptions are that f (x, y), and f2 (x, y) = (∂/∂y)f (x, y), are both con-
tinuous in some fixed (closed) rectangle R in the x−y plane. Note that in particular,
we are assuming that the partial derivative (∂/∂y)f (x, y) exists in R.
We want to prove there is a unique solution passing through any point (x0 , y0 )
in R. In fact the solution will go all the way to the boundary of R — top, bottom
or one of the sides.
We will prove there is a solution in some smaller rectangle centred at (x0 , y0 ),
which passes through both of its sides. By then taking a new small rectangle centred
at some point further along the solution for the first small rectangle, we can extend
the solution. In fact, one can continue this process all the way1 to the boundary of
R.
1F We will be able to compute the size of these small rectangles, and in this way one

can show that only a finite number of them are needed to “reach” the boundary of R.
94 7. FDIFFERENTIAL EQUATIONS

Figure 3. Slope field for f in the rectangle R.

Figure 4. 6 small rectangles here get us all the way to the

boundary of R. The point (x0 , y0 ) is indicated by a small
square black dot with an arrowhead below it.

Thus the main point is to first show that in some (small) rectangle Rδ , whose
base is of length 2δ and which is centred at the point (x0 , y0 ), there is a solution
which extends to both sides of this small rectangle.
The proof proceeds in 7 steps.
Step A Problem (41) is equivalent to showing the “integral equation”
Z x
(42) y(x) = y0 + f (t, y(t)) dt
x0

has a solution. One sees this by integrating both sides of (41) from x0 to x. Con-
versely, differentiating the integral equation (42) gives back the differential equation,
and clearly y(x0 ) = x0 also follows from the integral equation.
 7.2. OUTLINE OF PROOF OF THE EXISTENCE AND UNIQUENESS THEOREM 95

For our first example

dy
(43) = y + cos x, y(0) = 0,
dx
we get
Z x
(44) y(x) = (y(t) + cos t) dt.
0

Step B To find the solution of (42) we begin with the constant function
y(x) = y0
and plug it into the right side of (42) to get a new function of x. We plug this
again into the right side to get yet another function of x. And so on.
For example, with (43), substituting the constant function y = 0 in the right
side of (44), and then repeating by plugging in the new function obtained after
each step, we get
Z x
cos t dt −→ sin x
0
Z x 
sin t + cos t dt −→ − cos x + 1 + sin x
0
Z x 
− cos t + 1 + sin t + cos t dt −→ x − cos x + 1
0
Z x  1 2
t − cos t + 1 + cos t dt −→ x +x
0 2
Z x
1 2  1 3 1 2
t + t + cos t dt −→ x + x + sin x
0 2 6 2
Z x
1 3 1 2  1 4 1 3
t + t + sin t + cos t dt −→ x + x − cos x + 1 + sin x
0 6 2 24 6
We call this sequence of functions a “sequence of approximate solutions”. We see
from the diagram that this sequence is converging, at least near (0, 0).

Figure 5. Apologies for the poor quality diagram.

In general, if yn (x) is the nth approximate solution, then

Z x
(45) yn+1 (x) = y0 + f (t, yn (t)) dt
x0
96 7. FDIFFERENTIAL EQUATIONS

is the (n + 1)th approximate solution.

Step C The next step is to show that on some small rectangle around (x0 , y0 )
this sequence of “approximate solutions” does indeed converge. The main point
in the proof is showing that if the rectangle is sufficiently small then the distance
between the nth and (n + 1)th approximate solutions is < r times the distance
between the (n − 1)th and the nth approximate solutions, for some fixed r < 1.
Thus the distance between consecutive solutions is decreasing “geometrically”
fast. This is the main idea in the proof.
Step D Let the limit function for the approximate solutions be denoted by y =
y(x). The next step is to show that this limit function is continuous. This is
not obvious, although it is not hard to show that the approximate solutions are
themselves continuous. The problem is that a sequence of continuous functions can
in fact converge to a non-continuous function, as in the following diagram. But in
our case the fact that the approximate solutions converge “geometrically” fast is
enough to ensure that the limit is continuous

Figure 6. A sequence of continuous functions converging at

each point to a discontinuous limit function.

Step E The next step is to show that the limit function y = y(x) satisfies the
integral equation. The fact it is continuous implies we can integrate the right
side of (42). And the fact that the approximate solutions converge to the function
y = y(x) geometrically fast enables us to prove that we can take the limit as n → ∞
on both sides of (45) and deduce (42)
Step F The next step is to show that any two solutions are equal. We show that
if d is the distance between two solutions of (42) then d ≤ rd for some r < 1, by an
argument like the one in Step C. This implies that d = 0 and so the two solutions
agree.
Step G The final step is to extend the solution from the small rectangle in step
C up to the boundary of R. We do this by essentially starting the process again at
a new point on the graph near the boundary of the small rectangle, getting a new
rectangle centred at the new point, and extending the solution out into the new
rectangle. We can show this process stops after a finite number of steps, when the
solution reaches the boundary of R.

7.3. FRigorous proof of the Existence and Uniqueness theorem

Theorem 7.3.1. Suppose that f (x, y) and f2 (x, y) = (∂/∂y)f (x, y) are both
continuous in the rectangle R consisting of all points (x, y) of the form a ≤ x ≤
b, c ≤ y ≤ d. Suppose (x0 , y0 ) is in the interior of R.
 7.3. FRIGOROUS PROOF OF THE EXISTENCE AND UNIQUENESS THEOREM 97

Then there exists a number δ > 0 and a unique function φ(x), defined and
having a continuous derivative on the interval (x0 − δ, x0 + δ), such that
(46) φ0 (x) = f (x, φ(x))
(47) φ(x0 ) = y0 .
In other words, φ(x) solves (i.e. satisfies) the initial value problem
dy
= f (x, y)
dx
y(x0 ) = y0
on the interval (x0 − δ, x0 + δ).
Remark : Let

∂ o
M = max{ |f (x, y)| : (x, y) ∈ R }, K = max f (x, y) : (x, y) ∈ R .
∂y
We will see in the proof that if we define Rδ (x0 , y0 ) to be the (open) rectangle
consisting of all those (x, y) such that x0 −δ < x < x0 +δ and y0 −M δ < y < y0 +M δ,
i.e.
n o
(48) Rδ (x0 , y0 ) = (x, y) : x ∈ (x0 − δ, x0 + δ), y ∈ (y0 − M δ, y0 + M δ) ,

then any δ > 0 for which Rδ (x0 , y0 ) ⊂ R and δ < K −1 , will work for the above
theorem.

Figure 7. The rectangle Rδ from (48). (The small “wiggles”

in the graph of φ should not be present. The shafts of the
arrow heads are barely discernible.)

Proof.
Step A We first claim that if φ(x) is a continuous function defined on some
interval (x0 −δ, x0 +δ), and (x, φ(x)) ∈ R for all x, then the following two statements
are equivalent:
(1) φ(x) has a continuous derivative on the interval (x0 − δ, x0 + δ) and solves
the given initial value problem there, i.e. (46) and (47) are true;
(2) φ(x) satisfies the integral equation
Z x
(49) φ(x) = y0 + f (t, φ(t)) dt.
x0

Assume the first statement is true. Then both φ0 (t), and f (t, φ(t)) by Sec-
tion 3.5, are continuous on (x0 − δ, x0 + δ) (it is convenient to use t here instead
98 7. FDIFFERENTIAL EQUATIONS

of x for the dummy variable). Thus for any x in the interval (x0 − δ, x0 + δ) the
following integrals exist, and from (46) they are equal:
Z x Z x
0
φ (t) dt = f (t, φ(t)) dt.
x0 x0

From the Fundamental Theorem of Calculus it follows that

Z x
φ(x) − φ(x0 ) = f (t, φ(t)) dt,
x0

which implies the second statement (since we are assuming φ(x0 ) = y0 ).

Next assume the second statement is true. Note that since φ(t) is continuous,
so is f (t, φ(t)) by Section 3.5, and so the integral does exist. Setting x = x0 we
immediately get (47)
Since the right side of (49) is differentiable and the derivative equals f (x, φ(x))
(by the Fundamental Theorem of Calculus), the left side must also be differentiable
and have the same derivative. That is, (46) is true for any x in the interval (x0 −
δ, x0 + δ). Moreover, we see that the derivative φ0 (x) is continuous since f (x, φ(x))
is continuous.
Thus the first statement is true.

Step B We now define a sequence of approximations to a solution of (49) as

follows:
φ0 (x) = y0
Z x
φ1 (x) = y0 + f (t, φ0 (t)) dt
x0
Z x
φ2 (x) = y0 + f (t, φ1 (t)) dt
x0
..
.
Z x
φn+1 (x) = y0 + f (t, φn (t)) dt
x0
..
.

The functions in the above sequence will be defined for all x in some interval
(x0 −δ, x0 +δ), where the δ has yet to be chosen. We will first impose the restriction
on δ that

(50) Rδ (x0 , y0 ) ⊂ R,

where Rδ (x0 , y0 ) was defined in (48).

The function φ0 (x) is just a constant function.

Since the points (t, φ0 (t)) certainly lie in Rδ (x0 , y0 ) if t ∈ (x0 − δ, x0 + δ), it
follows that f (t, φ0 (t)) makes sense. Also, f (t, φ0 (t)) is a continuous fnction of t
from Section 3.5, being a composition of continuous functions. It follows that the
integral used to define φ1 (x) exists if x ∈ (x0 − δ, x0 + δ). In other words the
definition of φ1 (x) makes sense for x ∈ (x0 − δ, x0 + δ).
 7.3. FRIGOROUS PROOF OF THE EXISTENCE AND UNIQUENESS THEOREM 99

Next, for x ∈ (x0 − δ, x0 + δ), we show that (x, φ1 (x)) ∈ Rδ (x0 , y0 ) and hence
∈ R. This follows from the fact that
Z x
|φ1 (x) − y0 | = f (t, φ0 (t)) dt
x0
Z x
≤ |f (t, φ0 (t))| dt from (38)
x
Z x0
≤ M dt since |f | ≤ M in R
x0
≤ Mδ since |x − x0 | ≤ δ.

It follows as before that the definition of φ2 (x) makes sense for x ∈ (x0 −
δ, x0 + δ). (We also need the fact that f (t, φ1 (t)) is continuous. This follows from
the fact φ1 (t) is in fact differentiable by the Fundamental Theorem of Calculus, and
hence continuous; and the fact that f (t, φ1 (t)) is thus a composition of continuous
functions and hence continuous.)
Etc. etc. (or proof by induction, to be rigorous; but it is clear that it will work).
In this way we have a sequence of continuous functions φn (x) defined on the
interval (x0 − δ, x0 + δ), and for x in this interval we have (x, φn (x)) ∈ Rδ (x0 , y0 ).

Step C The next step is to prove there exists a function φ(x) defined on the
interval (x0 − δ, x0 + δ) such that

φn (x) → φ(x)

for all x ∈ (x0 − δ, x0 + δ). Let (for n ≥ 0)

dn = max |φn (x) − φn+1 (x)|,

2
where the maximum is taken over the interval (x0 − δ, x0 + δ).

Figure 8. dn is the distance between the functions φn and φn+1 .

2We should be a little more careful here. Since the points (x, φn (x)) and (x, φn+1 (x))
both lie in Rδ (x0 , y0 ), it follows that |φn (x) − φn+1 (x)| < 2M δ. But the maximum may
be “achieved” only when x = x0 ± δ, which is not actually a point in the (open) interval
(x0 − δ, x0 + δ). To make the argument rigorous, we should replace “max” by “lub” in
Step C.
100 7. FDIFFERENTIAL EQUATIONS

Then for n ≥ 1
dn = max |φn (x) − φn+1 (x)|
x∈(x0 −δ,x0 +δ)
Z x
= max f (t, φn−1 (t)) − f (t, φn (t)) dt
x∈(x0 −δ,x0 +δ) x0
Z x
≤ max f (t, φn−1 (t)) − f (t, φn (t)) dt
x∈(x0 −δ,x0 +δ) x0
Z x
≤ max K φn−1 (t) − φn (t) dt by Section 5.6
x∈(x0 −δ,x0 +δ) x0
Z x
≤ max Kdn−1 dt from the definition of dn−1
x∈(x0 −δ,x0 +δ) x0
= Kδ dn−1
Repeating this argument we obtain
dn ≤ Kδ dn−1 ≤ (Kδ)2 dn−2 ≤ (Kδ)3 dn−3 ≤ · · · ≤ (Kδ)n d0 .
We now make the further restriction on δ that
(51) Kδ < 1.
Since
|φn (x) − φn+1 (x)| ≤ dn ≤ d0 (Kδ)n ,
it follows from Theorem 2.7.4 that the sequence φn (x) converges for each x ∈
(x0 − δ, x0 + δ). We define the function φ(x) on (x0 − δ, x0 + δ) by
φ(x) = lim φn (x).
n→∞
Moreover, by the commt following that theorem,
(52) |φn (x) − φ(x)| ≤ Arn ,
where A = d0 /(1 − Kδ) and r = Kδ < 1. (Note that this is saying that the graph
of φn lies within distance Arn of the graph of φ, see Figure 9.)

Step D (See Figure 9.) We next claim that φ(x) is continuous on the interval
(x0 − δ, x0 + δ).
To see this let a be any point in the interval (x0 − δ, x0 + δ); we will prove that
φ is continuous at a.
Let ε > 0 be an arbitrary positive number.
First choose n so that Arn < ε/3 and hence from (52)
(53) x ∈ (x0 − δ, x0 + δ) implies |φn (x) − φ(x)| ≤ ε/3.

Figure 9. Figure for Step D. (The small squiggles in the

graph are the fault of the graphics program!)
 7.3. FRIGOROUS PROOF OF THE EXISTENCE AND UNIQUENESS THEOREM 101

By continuity of φn there exists η > 0 (which may depend on n and hence on

ε) such that
(54) |x − a| < η implies |φn (x) − φn (a)| < ε/3.
(We also choose η sufficiently small that if |x − a| < η then x ∈ (x0 − δ, x0 + δ).
From (53) (applied with x and again with x replaced by a) and (54) it follows
that if |x − a| < η then
|φ(x) − φ(a)| = |(φ(x) − φn (x)) + (φn (x) − φn (a)) + (φn (a) − φ(a))|
≤ |φ(x) − φn (x)| + |φn (x) − φn (a)| + |φn (a) − φ(a)|
ε ε ε
≤ + + = ε.
3 3 3
Since a was any point in (x0 − δ, x0 + δ), and ε was any positive number, this proves
the claim that φ is continuous on the interval (x0 − δ, x0 + δ).

Step E We defined
Z x
(55) φn+1 (x) = y0 + f (t, φn (t)) dt.
x0

We have shown in Step C that3

φn+1 (x) → φ(x)
for each x in the interval (x0 − δ, x0 + δ). We next claim that for the right side
of (54) we have
Z x Z x
y0 + f (t, φn (t)) dt → y0 + f (t, φ(t)) dt.
x0 x0

It then follows from the claim that

Z x
φ(x) = y0 + f (t, φ(t)) dt,
x0

which establishes (49) and hence proves the theorem by Step A.

To prove the claim we compute
Z x Z x Z x
f (t, φn (t)) dt − f (t, φ(t)) dt ≤ f (t, φn (t)) − f (t, φ(t)) dt
x0 x0 x0
Z x
≤ K φn (t) − φ(t) dt by Section 5.6
x0
Z x
≤ KArn dt by (52)
x0
≤ KδArn .
Since 0 ≤ r < 1 this establishes the claim and hence the theorem.

Step F Next, we must show that any two solutions of (46) and (47), or equiva-
lently of (49), are equal.
Suppose that φ(x) and ψ(x) are any two solutions. Then if
d = max |φ(x) − ψ(x)|,

3If an → a for a sequence, then it follows that an+1 → a, why?

102 7. FDIFFERENTIAL EQUATIONS

where the maximum is taken over the interval (x0 − δ, x0 + δ).4 Then for any
x ∈ (x0 − δ, x0 + δ),
Z x


|φ(x) − ψ(x)| = f (t, φ(t)) − f (t, ψ(t)) dt
x0
Z x
≤ f (t, φ(t)) − f (t, ψ(t)) dt
x
Z x0
≤ K|φ(t)) − ψ(t)| dt from Section 5.6
x0
≤ Kδd
Since this is true for any x ∈ (x0 − δ, x0 + δ), it follows that
d ≤ Kδd.
Since Kδ < 1, this implies d = 0!!
Hence φ(x) = ψ(x) for all x ∈ (x0 − δ, x0 + δ). 

Step G So far we have a solution, and it is unique, whose graph lies in a rectangle
Rδ . The dimensions of Rδ depend only on K an M , and otherwise not on the initial
point, except that we also require Rδ ⊂ R. By starting the process again at a new
point close to the boundary of Rδ we can extend the solution outside Rδ , and after
a finite number of steps extend the solution up to the boundary of R.

End of proof, end of chapter, end of semester. Have a good holiday.

4As in Step C we should really write “lub” instead of “max”. The proof is essentially

unchanged.