Você está na página 1de 203

Lecture notes for

MATH 333 Analysis I


and MATH 334 Analysis II

M.McIntyre

August 14, 2013


2
Contents

Details of the courses v


0.1 Content of MATH 333 . . . . . . . . . . . . . . . . . . . . . . . . v
0.2 Content of MATH 334 . . . . . . . . . . . . . . . . . . . . . . . . vi
0.3 Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
0.4 Learning strategies . . . . . . . . . . . . . . . . . . . . . . . . . . viii
0.5 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

I Limits and Continuity of Functions 1


1 Vector Spaces 3
1.1 Definition and Examples . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Vector spaces for analysis . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Normed Vector Spaces. . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Geometry and the Cauchy-Schwartz inequality. . . . . . . . . . . 11
1.5 Subsets of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6 Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 Limits and Continuity 29


2.1 An Intuitive Approach . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 Theorems on Limits and Continuity . . . . . . . . . . . . . . . . 42
2.4.1 Limit of a composite. . . . . . . . . . . . . . . . . . . . . 42
2.4.2 The Squeeze Principle . . . . . . . . . . . . . . . . . . . . 44
2.4.3 Local Principle for Limits . . . . . . . . . . . . . . . . . . 44
2.4.4 Limits of Ordered Pairs of Functions. . . . . . . . . . . . 48
2.5 Some Algebraic Theorems . . . . . . . . . . . . . . . . . . . . . . 51
2.5.1 Algebra of Limits . . . . . . . . . . . . . . . . . . . . . . . 54

II Limits of Sequences 59
3 Limits of Sequences 61

i
ii CONTENTS

3.1 Intervals in R, countability, bounded subsets of R. . . . . . . . . 62


3.2 Some useful inequalities . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Definition-limit of a sequence . . . . . . . . . . . . . . . . . . . . 67
3.3.1 Limits of functions and limits of sequences-A comparison 70
3.4 Some theorems on limits of sequences . . . . . . . . . . . . . . . 72
3.4.1 The real number system- filling the line. . . . . . . . . . . 77
3.5 Existence theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5.1 Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.6 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.6.1 Real-valued series with positive terms . . . . . . . . . . . 87
3.7 Absolute convergence and Cauchy sequences . . . . . . . . . . . . 92
3.8 From Q to R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

III MATH 334, Part I- Sequences and Series in Func-


tion Spaces 105
4 Function Spaces 107
4.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2 Supremum norm. . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3 Balls in B[a, b] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

5 Sequences in B[a, b] 117


5.1 Uniform and pointwise convergence . . . . . . . . . . . . . . . . . 117
5.2 Series-partial sums of sequences of functions . . . . . . . . . . . . 122

6 Power series 131


6.1 Definition and examples . . . . . . . . . . . . . . . . . . . . . . . 131
6.2 Taylor (resp Maclaurin) series . . . . . . . . . . . . . . . . . . . . 136
6.3 Approximation by power series . . . . . . . . . . . . . . . . . . . 140

7 Contraction maps 143


7.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . 143
7.2 Contraction map theorem . . . . . . . . . . . . . . . . . . . . . . 144

IV Real Analysis 153


8 Limits 155
8.1 Real-valued functions on S R . . . . . . . . . . . . . . . . . . . 155
8.2 Order relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3 Right and left limits . . . . . . . . . . . . . . . . . . . . . . . . . 156

9 The Riemann Integral 159


9.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
9.2 Using the definition of the Riemann Integral . . . . . . . . . . . . 162
9.3 Properties of the Riemann integral . . . . . . . . . . . . . . . . . 164
CONTENTS iii

10 Major results used in calculus 169


10.1 The fundamental theorem of calculus . . . . . . . . . . . . . . . . 169
10.2 Zero derivative/mean value theorem . . . . . . . . . . . . . . . . 172
10.3 Bolzanos theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.4 Intermediate value theorem . . . . . . . . . . . . . . . . . . . . . 175
10.5 Brouwers fixed point theorem . . . . . . . . . . . . . . . . . . . . 176
10.6 Proof of Bolzanos theorem . . . . . . . . . . . . . . . . . . . . . 176
10.7 Maximum/minimum values/boundedness . . . . . . . . . . . . . 178
10.8 Rolles theorem/proof of mean value theorem . . . . . . . . . . . 181
10.9 IVT for derivatives/lH
opital . . . . . . . . . . . . . . . . . . . . 183

11 Appendix 187
iv CONTENTS
Details of the courses

Prerequisite: MATH 213 (and MATH 211)

0.1 Content of MATH 333


In analysis you will study limits of functions and limits of sequences. These
topics are not new to you as you have learned about limits in 100 and 200 level
calculus courses. The fundamental ideas of the derivative and the integral of
a function are both expressed in terms of limits. Derivatives involve limits of
functions and integrals involve limits of sequences. Calculus is the most widely
applied branch of mathematics and analysis provides the foundation on which
calculus rests.
PART I:

1. We show how the real line R can be generalized to give spaces of higher
dimension, by first recalling the definition of a vector space (MATH 211).
The emphasis is on the vector spaces useful to analysis, in particular Rn .
We define a norm on a vector space, an idea inspired by the idea of length
for the arrows studied in elementary vector algebra. In analysis we are
only interested in vector spaces with a norm because we want to be able
to measure how close two vectors are.

2. We define open and closed balls in a normed vector space and examine
some proofs using balls. We examine other subsets of Rn derived from
open (closed) balls.

3. In preparation for the study of limits we examine functions which have


subsets of Rn as their domain or codomain and see how to visualize such
functions.

4. Limits and continuity of functions are given a precise mathematical defi-


nition in the context of functions which map subsets of one normed vector
space to another normed vector space. We examine uniqueness of limit,
limit of a composite, the squeeze principle, the local principle for limits,
limits of ordered pairs of functions.

v
vi DETAILS OF THE COURSES

5. We examine the algebraic operations of sum, difference and product of


functions and develop the algebra of continuous functions.

PART II

1. Since this part of the course will look at the detail of convergence of se-
quences we will begin by discussing countability of subsets of the real
numbers and to examine the least upper bound (greatest lower bound)
property of the real numbers because this is the property of the real num-
bers upon which analysis depends. From this property we deduce that
there are enough real numbers to label every point on the line. There
are a number of ways in which this line-filling behaviour of the real num-
bers can be expressed formally. Three ways that will concern us in this
course are the bounded increasing sequence theorem, the nested intervals
theorem and the completeness theorem.

2. Next we want to examine the definition of limit in the context of se-


quences. Our examples can be real-valued sequences and initially we want
to prove, from the definition, whether or not a given(guessed) real num-
ber is the limit of a given sequence. We then give some basic theorems
about limits for vector-valued sequences. In particular: uniqueness of
limit, algebra of limits, limit of a composite, limit of an ordered pair and
the squeeze(sandwich) principle of limits. We also have invariance of the
limit of a sequence under finite alterations, invariance of limit to shifting
sequences and a definition of continuity via sequences.

3. Then we consider sequences for which it may not be possible to guess


the limit and examine the notion of existence theorems. If we cannot
establish what the limit is we can at least establish that a limit exists. We
begin with increasing sequences and give the bounded increasing sequence
theorem. (A bounded decreasing sequence theorem is indicated).

4. Series are introduced as a special sort of sequence. Given a sequence of real


numbers one defines a new sequence as the nth partial sum of the terms
of the original sequence and considers the limit of the new sequence. We
begin with series consisting of positive terms and discuss the comparison
test, the integral test, the ratio test and the root test.

5. Cauchy sequences. We work examples of series which are Cauchy and


define what it is for a sequence to be dominated by a Cauchy sequence.
We then have the result that every sequence which is dominated by a
Cauchy sequence is itself Cauchy. We define a complete space and look at
the proof that R is complete.

0.2 Content of MATH 334


PART I
0.2. CONTENT OF MATH 334 vii

1. We next consider sequences which contain non-positive terms and define


absolute convergence. Since we have that every Cauchy sequence in a
complete space is convergent, the proof that every absolutely convergent
sequence in a complete space is convergent can now be discussed.
2. We introduce the normed vector spaces in which uniform convergence can
be discussed and we define the supremum norm. In particular we define
the function spaces
B[a, b] = {f : [a, b] R and f is bounded}
C[a, b] = {f : [a, b] R and f is continuous}
D[a, b] = {f : [a, b] R and f is differentiable}
S[a, b] = {f : [a, b] R and f is a step function}
We define uniform convergence and compare it to pointwise convergence.
Our examples should include a sequence of functions which converges
pointwise to a function, but does not converge uniformly to any func-
tion. We prove that uniform convergence implies pointwise convergence.
We also give examples of sequences of functions which will demonstrate
that the first two normed vector spaces in the list above are complete
whilst the latter two are not complete for uniform convergence.
3. Convergence of that special class of sequences of functions referred to as
power series centred at some point, is the second topic of part I. We define
the radius of convergence and the interval of convergence. Our examples
are confined to real-valued functions. We review the power series approach
to the definition of the exponential function exp and its inverse the natural
logarithmic function log.
4. We give an application of the theory and techniques developed in the study
of sequences of functions for finding the solutions of equations. Using the
method of iteration we produce a sequence of functions which are succes-
sive approximations to the solution. We find in the guise of the contraction
map theorem, SUFFICIENT conditions for the successive approximations
to yield a solution. A particular example of using this technique could be
in establishing the existence of a solution to a differential equation which
does not yield to other methods of solution. The contraction map theorem
will also be used in MATH 421 Advanced Calculus, to prove the inverse
function theorem.
PART II:
1. We now return to the functions f : R R of elementary calculus. We
define the Riemann integral and see how the various properties of integrals
can be derived from the theorems about limits of sequences.
2. Our next aim is to understand the most useful Fundamental Theorem
of Calculus (FTC) which justifies the computation of integrals using an-
tiderivatives and indeed the method of solving first order linear differential
equations. We will focus on why the theorem is true.
viii DETAILS OF THE COURSES

3. To prove the second part of the FTC we will study, the zero derivative
theorem and the mean value theorem (proof of which will come towards
the end of the course).

4. Next we examine solutions of equations, in particular existence theorems


which were developed to tell us that a given equation will have a solution
if certain conditions are satisfied. We study Bolzanos theorem, derive the
Intermediate value theorem and examine a second theorem which gives
sufficient conditions for a map to have a fixed point, the Brouwer fixed
point theorem.

5. In this section we examine the idea of a real-valued function having a


maximum value on a domain and study the boundedness theorem and the
maximum value theorem. In addition, we prove an intermediate theorem
for derivatives.

6. We prove the mean value theorem, using Rolles theorem and we prove
Rolles theorem. This completes the proof of the fundamental theorem of
calculus. Finally, we can examine alternative versions of the mean value
theorem, the intermediate value theorem for derivatives and lH opitals
rule.

0.3 Assessment
There will be two class tests; the first in week 5 or 6 and the second in week
11. Exercises will be assessed from time to time. The class tests contribute 20%
and the exercises contribute 10% to the final grade, the final examination being
the other 70% of the final score.

0.4 Learning strategies


Strategies for effective learning

Attend all lectures. Notice there is no need to write down everything from
the board as much of it is already in the notes.

Attempt all exercises, practise writing down your solutions and submit
upon request.

Attend tutorials and take along your attempts at the relevant exercises
along with any remarks which may have been made by the one who read
your written solution.

Visit the library, browse through the available books and try additional
exercises relevant to your course.
0.5. INTRODUCTION ix

For every one hour of lecture time you should spend not less than three hours:

- independently trying the relevant exercises,

- discussing the lecture material and exercises within your study group and

- independently writing down your solutions.

It is important to try writing down solutions to problems. Otherwise you will


be trying to do this for the first time when you are faced with an assessment!

At times it may appear that the lecturer is going through the course material
rather rapidly. The extra 2-3 hours spent doing exercises and understanding
the material will ensure that you are not left behind. The lecturer will expect
that this is what you have been doing. You will also be in a position to seek
further clarification, either from the lecturer or from the assigned tutor, of any
material with which you are having trouble.

Join the Mathematical Society (TMS). Here you may find advise from your
seniors, teaching assistants and tutors, further motivation, interesting applica-
tions of the mathematics you are learning and a forum for expressing your own
ideas and opinions about the learning of mathematics.

0.5 Introduction

Analysis is a branch of mathematics which has grown out of calculus and indeed
provides the theoretical basis for calculus.
Aside from some isolated instances of finding solutions to area and tangent
problems, calculus is generally regarded as having been invented in the 17th
century by Newton and Leibniz. By realizing that differentiation and integra-
tion are mutually inverse processes, however, Newton and Leibniz opened the
way to a systematic approach to such problems. Solving a problem in calculus
came to mean finding a formula for the answer by mechanical manipulation of
rules, without much regard to their meaning. The emphasis on manipulation of
formulae reached its climax in the work of Euler in the 18th century.
During the 19th century, mathematicians began to take a critical look at
the work of their predecessors, who had often used the rules of calculus in
contexts where they were no longer valid. Thus analysis was conceived. Instead
of emphasising solution of problems by manipulation of formulae the aim was
to answer such questions as: does the problem have a solution? What does it
mean for the problem to have a solution? Under what conditions do the rules
apply?
A leading role in the genesis of analysis was played by Cauchy, who intro-
duced the ideas of limits and continuity as a basis for analysis, and Weierstrass,
x DETAILS OF THE COURSES

who was able to provide precise definitions of these concepts based on the idea
of inequalities between real numbers. Consequently a study of the real numbers
lies at the foundation of analysis.
Limits of functions form the foundation for differentiation. Given a function
f : R R the derivative of f at a point x R is defined as
f (x + h) f (x)
f 0 (x) = lim .
h0 h

The term f (x+h)fh


(x)
on the right hand side, is a function of h, it is called the
Newton quotient.
Geometrically the motivation for the definition comes from the problem of
defining the tangent to the graph of f at the point (x, f (x)). The Newton
quotient gives the slope of a typical chord to the graph. As h 0, we expect
the chord to approach the tangent. Thus if the limit exists we define the slope
f 0 (x) of the tangent as the limit of the slope of the chord.

6
6

f (x + h) f (x)

h -

-
x x+h

You must be familiar with functions which are not differentiable at some point
of their domain. These examples justify the caution of the 19th century math-
ematicians who emphasised the need to check the existence of limits.
Proving the various rules of differentiation involves being able to use the
corresponding rule for limits. One of the aims of analysis is to find proofs for
the validity of the rules for limits, based on the precise definition of the limit of
a function.
Limits of sequences form the foundation for integration. In calculus the
integral of a positive sufficiently smooth fuction over an interval is defined as
Z b
f = lim sn
a n

where sn is the sum of areas of n rectangles, formed under the graph of f , having
divided the interval [a, b] into n subintervals of equal length.
0.5. INTRODUCTION xi

6 f

...

-
a ... b

The various properties of integrals can be proved from this definition by using
corresponding properties of limits of sequences. Another of the aims of analysis
is to find proofs for the validity of the rules for integrals, based on the precise
definition of the limit of a sequence and the corresponding rules for limits.
Exercise
Give an example of a continuous function f : R R which is not differ-
entiable at the point 1, in its domain.
Explain the apparent paradox: The function f (x) = x2 is positive, hence
R 1 2
2
x dx > 0. On the other hand applying the rules of integration one
R 1 2
has 2
x dx = 1 12 = 1 1 < 0!
x 2
xii DETAILS OF THE COURSES
Part I

Limits and Continuity of


Functions

1
Chapter 1

Vector Spaces

1.1 Definition and Examples

In essence a vector space is to be an algebraic system which consists of an abelian


group (whose elements are to be called vectors) acted upon linearly by a field
(whose elements are to be called scalars), with the 1 of the field having trivial
action, the zero of the field sending any vector to the zero vector and the action
of the field respecting both the group operation and the field operations.

The term vector space thus has an axiomatic definition.

Definition 1.1.1 V is a vector space over a field F if:

(i) (V, +) is an abelian group and

(ii) there is an action of elements of F on elements of V satisfying:

(I) F , x V , !v V 1 such that x V


(i.e. V is closed under the action, which is well-defined),
(II) F , x, y V , (x + y) = x + y
(i.e. the action respects + in V )
(III) , F , x V , ( + )x = x + x
(i.e. the action respects + in the field)
(IV) , F , x V , ()x = (x)
(i.e. the action respects in the field) and
(V) x V and for 1 F , 1x = x
(i.e. the action of the 1 of the field is trivial.)
1 read this as there exists a unique v in V

3
4 CHAPTER 1. VECTOR SPACES

Another way of saying the same thing is given in the exercises. We call the
elements of V , vectors and the elements of F , scalars. Thus the action of F on
V can equally be thought of as a scalar multiplication.

The elements of V can be real numbers, complex numbers, ordered n-tuples,


matrices, polynomials, or functions with an appropriate vector addition and
scalar multiplication.

There are of course many laws or rules of vector algebra, but all of them
are derivable from the axioms. For example:

Proposition 1.1.2 Let V be a vector space over a field F , then


v V, 0v = 0.
Note: The zero on the left is the zero of the field, whilst the zero on the right is
the zero of the vector space.

Proof of 1.1.2: Let v V . 0 V , since (V, +) is an abelian group. Now


0 + 0v = 0v since 0 is the identity of the group (V, +)
= (0 + 0)v since 0 is the identity element of F
= 0v + 0v by axiom (III) of definition (1.1) thus
0 = 0v by right cancellation.

Proposition 1.1.3 Let V be a vector space over a field F , then


F, 0 = 0.

Proof: Let F , let v V then


0 = (0v) by proposition (1.2)
= (.0)v by axiom (IV)
= 0v n by 0 in F
=0 by proposition (1.2).

Proposition 1.1.4 Let V be a vector space over R, then


a V, a + a = 2a.

Proof: Let a V . Then


a + a = 1a + 1a by axiom (V)
= (1 + 1)a by axiom (III)
= 2a addition in the field R.
1.2. VECTOR SPACES FOR ANALYSIS 5

1.2 Vector spaces for analysis

Some examples of vector spaces which are important to analysis will be discussed
in this section.

Example 1.2.1 R2 is the set of ordered pairs of real numbers, i.e.

R2 = {(x1 , x2 ) : x1 R, x2 R},

with + defined for each (x1 , x2 ), (y1 , y2 ) R2 by

(x1 , x2 ) + (y1 , y2 ) = (x1 + y1 , x2 + y2 ).

(R2 , +) is an abelian group (check the details). In addition with scalar multi-
plication defined for each R, and for each (x1 , x2 ) R2 by

(x1 , x2 ) = (x1 , x2 ),

R2 is a vector space over R. (Again check the details.)

Example 1.2.2 We ask if it is possible for R2 to be a vectorspace over R if we


define the vector addition by, for each (x1 , x2 ), (y1 , y2 ) R2 ,
 
1 1
(x1 , x2 ) + (y1 , y2 ) = , ?
x1 + y1 x2 + y2

Notice now that (R2 , +) is not a group since the closure axiom fails. Consider
1
for example (1, 3), (1, 3) we have that x1 +y1
= 1/0 6 R, so R2 is not closed
under this operation.

To prove that a set with given vector addition and scalar multiplication is a
vector space we must check all of the axioms. But to demonstrate that it is not
a vector space we just need to find a specific example which does not satisfy one
of the axioms.

Example 1.2.3 The Vector Space Rn


Given real numbers x1 , x2 , . . . , xn then (x1 , x2 , . . . , xn ) is an ordered n-tuple
of numbers.
Another n-tuple (y1 , y2 , . . . , yn ) is equal to (x1 , x2 , . . . , xn ) xi = yi , i
{1, 2, . . . , n}. Given axes which are mutually perpendicular, we may think of an
n-tuple as
(i) a point (x1 , x2 ) in the plane, when n = 2
(ii) a point (x1 , x2 , x3 ) in 3-dimensional space, when n = 3.
6 CHAPTER 1. VECTOR SPACES

6 6 (x1 , x2 , x3 )
(x1 , x2 )
x2 . . . . . . . x3
..
. x2
.. ......
. -      -
x1    x 1

And in general Rn is the set of ordered n-tuples (x1 , x2 , . . . , xn ), with vector


addition defined for each x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ) by

x + y = (x1 , x2 , . . . , xn ) + (y1 , y2 , . . . , yn ) = (x1 + y1 , x2 + y2 , . . . , xn + yn )

and scalar multiplication defined for each R and each


x = (x1 , x2 , . . . , xn ) by

x = (x1 , x2 , . . . , xn ) = (x1 , x2 , . . . , xn ).

It is routine to check that Rn is a vector space.

In vector algebra courses, we use the vectors i, j and k (the arrows of unit length
along the Cartesian coordinate axes) and write

(x1 , x2 , x3 ) = x1 i + x2 j + x3 k.

In Rn we replace the vectors i, j and k with vectors

e1 = (1, 0, 0, . . . , 0)
e2 = (0, 1, 0, . . . , 0)
e3 = (0, 0, 1, . . . , 0)
..
.
en = (0, 0, 0, . . . , 1)

where each of the items on the right are n-tuples of real numbers. Notice that
in R3 we have
i = e1 = (1, 0, 0)
j = e2 = (0, 1, 0)
k = e3 = (0, 0, 1).
[The e stands for einheit the German word for unit.]
1.3. NORMED VECTOR SPACES. 7

Example 1.2.4 We want to write the vector 3e1 + 2e2 as a point in R2 :

3e1 + 2e2 = 3(1, 0) + 2(0, 1) by definition of ei in R2


= (3, 0) + (0, 2) by definition of scalar multiplication
= (3, 2) by definition of vector addition.

We could have performed the calculation geometrically by thinking as follows

(3, 2)
2e2
6 6 3





e2 6 6


- -  - -
e1 3e1

Notice that one can think of a vector in Rn either as an arrow and write
3e1 + 2e2 (for example) or as a point and write (3, 2).
For vector spaces other than Rn restudy your lecture notes for MATH 211,
where we discussed vector spaces of matrices like
  
a11 a12
M22 (R) = : aij R
a21 a22
and vector spaces of functions such as

F(S, R) = {f : S R : f is a function}.

1.3 Normed Vector Spaces.

A norm on a vector space V is a function which assigns to each element x


V a real number ||x|| 0, which measures how large the vector x is. If
V Rn and x Rn , then we can think of ||x|| as the distance of x from
0 = (0, 0, . . . , 0) Rn . (It is the length of the arrow x and alternatively the
distance from 0 to the point.)

The geometry of vector addition suggests that a norm should satisfy the 4-
inequality; a, b V, ||a + b|| ||a|| + ||b||. In addition the definition of the
scalar multiplication in Rn suggests that a norm should satisfy homogeneity;
R, x V , ||x|| = || ||x||, where || is the absolute value of R.
Thus
8 CHAPTER 1. VECTOR SPACES

Definition 1.3.1 A norm on a vector space V over the field R is a map || || :


V R satisfying a, b V, R,
(N1) (positivity) ||a|| 0 and ||a|| = 0 a = 0;
(N2) (the 4-inequality) ||a + b|| ||a|| + ||b|| and
(N3) (homogeneity) ||a|| = || ||a||.


In the above for R, we have || := 2 or equivalently

if 0
|| = ,
if < 0

so | | is just the absolute value function on R. In fact this function satisfies the
definition of being a norm on the vector space R over the field R.

A normed vector space is an ordered pair (V, || ||) with V a vector space and
|| || : V R a norm.

Example 1.3.2 We consider the pair (V, || ||) with V = R and || || = | |


defined by x R

x if x 0
2
|x| = x = .
x if x < 0
Geometrically |x| is the distance of x from 0 R, so for example | 3| = 3 = |3|.
length is | 3|
 | | | | | | | -
-3 -2 -1 0 1 2 3

Claim: | | : R R is a norm on R.
Proof: We will only prove
N1 and leave N2 and N3 as an exercise.
Let x R, then |x| := x2 0 by definition
of square root. If x = 0 then
|x| = |0| = 0 = 0. Conversely if |x| = x = 0 then x2 = 0 and so x = 0.
2 2

Hence |x| = 0 x = 0.

Example 1.3.3 We want to see whether or not (R, || ||) with || || : R R


defined by ||x|| = x2 is a normed vector space.

It should be clear to you that this map satisfies positivity N1 but probably not
N3 or N2. We choose a specific counter-example to demonstrate that N3 is not
1.3. NORMED VECTOR SPACES. 9

satisfied. Consider for example x = 1 R (the vector space) and = 2 R (the


field of scalars). Then ||x|| = ||2.1|| = 4, whilst || ||x|| = |2|.12 = 2 6= ||x||.
So (R, || ||) is not a normed vector space.

Example 1.3.4 In this example we will establish that (R2 , || ||) where
p
||x|| = ||(x, y)|| = x2 + y 2

is a normed vector space.

Proof: (N1) Let x = (x, y) R2 , then


p
2 2
p ||x|| = x + y 0 by definition
of square root. Next let ||x|| = 0 so x2 + y 2 = 0, thus x2 + y 2 = 0 and so
x, y = 0, since x2 , y 2 0. Thus ||x|| = 0 x = 0. Conversely if x = 0 = (0, 0),
then x = 0 = y and so ||x|| = ||(0, 0)|| = 0. So || || satisfies positivity.

(N3) Let x = (x, y) R2 and let R. Then

||x|| = ||(x, y)|| = ||(x, y)


2
by
p definition of the scalar multiplication in R
= p(x)2 + (y)2 by definition
p of || ||
2 2 2
= p(x + y ) = x + y 2 2 2

= || x2 + y 2 by definition of | |
= || ||x|| by definition of || ||

as required.

To prove the 4-inequality one can use the Cauchy-Schwartz inequality, which
says

Theorem 1.3.5 x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ) Rn

Pn
xy |x y| = | xi yi | = |x1 y1 + x2 y2 + + xn yn |
p i=1 p
x21 + + x2n y12 + + yn2 = ||x|| ||y||.

You can read more about the Cauchy-Schwartz inequality in section 1.4.
Be careful not to confuse the Cauchy-Schwartz inequality with the tri-
angle inequality. We are using theorem 1.3.5 to prove the triangle inequality
for the map given in example 1.3.4.

Proof of (N2): Let x = (x1 , x2 ), y = (y1 , y2 ) R2 then x+y = (x1 +y1 , x2 +y2 )
and so
||x + y|| = ((x1 + y1 )2 + (x2 + y2 )2 )1/2
10 CHAPTER 1. VECTOR SPACES

thus

||x + y||2 = (x1 + y1 )2 + (x2 + y2 )2


= x21 + y12 + x22 + y22 + 2x1 y1 + 2x2 y2
x21 + y12 + x22 + y22 + 2(x21 + x22 )1/2 (y12 + y22 )1/2
by the Cauchy-Schwartz inequality
2
= (x21 + x22 )1/2 + (y12 + y22 )1/2
= (||x|| + ||y||)2

consequently ||x + y|| ||x|| + ||y|| as required.

s
n
Example 1.3.6 More generally we have (Rn , || ||) with ||x|| = x2i for
P
i=1
x = (x1 , x2 , . . . , xn ), is a normed vector space.

The norms given in examples 1.3.2, 1.3.4 and 1.3.6 are variously referred to
as the

usual
standard


Euclidean
P ythagorean

norm. Geometrically || || : Rn R measures the Pythagorean distance


from
2
the zero of the vector space. For n = 1 x = x R and ||x|| = |x| = x .

 | | | -
x 0 x
p
For n = 2, x = (x1 , x2 ) and ||x|| = x21 + x22 .

6 6
(x1 , x2 )
x2 . . . . . . . (x1 , x2 , x3 )
... x3 *
length is ||x||  x2
.. 
 .....
. - 
    -
x1  x1
1.4. GEOMETRY AND THE CAUCHY-SCHWARTZ INEQUALITY. 11

1.4 Geometry and the Cauchy-Schwartz inequal-


ity.
Orthogonal is the greek term for at right angles to. The projection of a vector
(thought of here as an arrow) can be a scalar or a vector depending on what we
are interested in.

Given two vectors a and b 6= 0 in R2 , the (vector) orthogonal projection of a in


the direction of b is the vector shown below:

a a
 ..
.
..
.
..
.
- -
b vector projection

In the case shown, the (scalar) orthogonal projection of a is just the length of
the vector projection, i.e. vector projection of a = scalar projection times unit
vector in the direction of b, i.e. ab ||b|| . If the direction of b is reversed
||b||
then one attaches a minus sign to the length of the vector projection to get the
scalar projection. Thus the scalar projection is ||a|| cos , where is the angle
measured from the vector b to the vector a.

Another way to regard orthogonal projection, which does not explicitly mention
the angle , uses the idea of minimising the distance between the tip of a and
the tip of the arrow b to which one is projecting. The projection is orthogonal
when this distance is a minimum, as we illustrate below.

a
. HH
H
 ..

HH
 ..
 .
HH
 .
..
H
HH
H

 - H
12 CHAPTER 1. VECTOR SPACES

This way of viewing orthogonal projections has the advantage of using only
ideas which are already available in Rn .

Let a and b be vectors in Rn and suppose that ||b|| = 1. We want to choose


R so that the vector b is the (vector) orthogonal projection of a. Since
||b|| = 1, the length of b is ||. The distance between the tips of these two
arrows is ||a b|| = f () say. So f varies with .



a 

 ||a b|| = f ()
 


- 
b

Notice that we can express the Euclidean norm on Rn in terms of the dot prod-
uct of vectors.
Proposition 1.4.1 For each vector a = (a1 , a2 , . . . , an ) Rn

||a|| = a a.

Proof: Let a = (a1 , a2 , . . . , an ) Rn . Then


a a = a21 + a22 + + a2n = ||a||2 ,
where || || is the usual norm in Rn .

Definition 1.4.2 Let a and b be vectors in Rn and suppose that ||b|| = 1. The
(scalar) orthogonal projection of a in the direction of b is the number = 0
at which the distance ||a b|| = f () assumes its minimum value.

Theorem 1.4.3 Let a and b be vectors in Rn and suppose that ||b|| = 1. The
dot product a b is the orthogonal projection of a in the direction of b.

Proof: Since f 2 has its minimum at the same point as f and is simpler, consider
f 2 () = ||a b||2 = (a b) (a b)
1.4. GEOMETRY AND THE CAUCHY-SCHWARTZ INEQUALITY. 13

by proposition (1.4.1). And so


f 2 () = a a 2(a b) + 2 b b
= ||a||2 2(a b) + 2 .

Thus f 2 is a quadratic (in ) and has its minimum where the derivative is zero;
i.e. for 2(a b) + 2 = 0, i.e. = a b. Thus by the definition the dot product
a b is the required orthogonal projection.

The theorem leads to the following diagram, which demonstrates the Cauchy-
Schwartz inequality in the case ||b|| = 1.


a a


- -
b has length |a b|

The Cauchy-Schwartz inequality (in the special case with ||b|| = 1) says that
|a b| ||a||.

In fact proposition (1.4.1) can be used to give a very simple proof of the Cauchy-
Schwartz inequality. In terms of the dot product of vectors the Cauchy-Schwartz
inequality says: x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn )
|x y| = |x1 y1 + x2 y2 + + xn yn | ||x|| ||y||.

Proof: Let x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ) and R.

Consider the case y = 0, then


xy =x0=0
so |x y| = 0 and ||x|| ||y|| = 0.

Consider the case ||y|| = 1, then using proposition (1.4.1) we have


0 ||x + y||2 = (x y) (x y)
= x x 2x y + y y
= ||x||2 2x y + 2 since ||y|| = 1.
14 CHAPTER 1. VECTOR SPACES

So 0 ||x||2 2x y + 2 is true for all real numbers , so in particular it is


true for = x y. This gives
0 ||x||2 2x y + (x y)2
= ||x||2 (x y)2 .
Thus
|x y|2 ||x||2 .
Hence
|x y| ||x||
which is the Cauchy-Scwartz inequality with ||y|| = 1.

y
Finally in the case that y 6= 0, the vector has norm of 1 and so we can
||y||
y
apply the previous paragraph to the vector (in place of the vector y.) This
||y||
gives
x y ||x||,

||y||
thus
1
|x y| ||x||, using homogeneity of norm
||y||
i.e. |x y| ||x|| ||y|| as required.

The usual norm is not the only norm one can define on Rn . For example for
x = (x1 , x2 , . . . , xn ) Rn the maps
||x|| = |x1 | + |x2 | + + |xn | and ||x|| = max{|x1 |, |x2 |, . . . , |xn |}
each define a norm on Rn . It is routine to check the details.

1.5 Subsets of Rn .

In first year calculus the domains of functions are commonly subsets of R, indeed
most often domains are closed intervals [a, b] := {x R : a x b}. The notion
of a ball in Rn is the generalization of an interval in R.

Definition 1.5.1 Let (V, || ||) be a normed vector space over R. Let a V ,
r > 0, r R then
Br (a) = {x V : ||x a|| < r}
is the open ball centred on a V of radius r > 0 and
Br (a) = {x V : ||x a|| r}
is the closed ball centred on a V of radius r > 0.
1.5. SUBSETS OF RN . 15

||x a|| is the distance between the points x and a in the norm || ||.

Br (a) Br (a) V .

In different spaces, for example in Rn with different norms, the balls as-
sume different shapes.

Example 1.5.2 In (R, | |), where | | is the usual norm on R,

Br (a) = {x R : |x a| < r}.

Now if x a 0, then |x a| = x a so |x a| < r x a < r x < a + r.


And if x a < 0, then |x a| = a x so |x a| < r a x < r x > a r.
Thus
Br (a) = {x R : a r < x < a + r} = (a r, a + r)
the open interval from a r to a + r.

In particular B2 (5) = {x R : |x 5| < 2} = (3, 7).

 ( | ) -
3 5 7

The set B2 (5) is the set of points which are no further away from 5 than 2.

Example 1.5.3 In (R2p , || ||), with || || the usual norm on R2 , i.e. for x =
(x1 , x2 ) we have ||x|| = x21 + x22 . Let a = (a1 , a2 ) R2 and let r > 0 then

Br (a) = {x R2 : ||x a|| < r}


= {(x1 , x2 ) R2 : ||(x1 , x2 ) (a1 , a2 )|| < r}
= {(x1 , x2 ) R2 : ||(x1 a1 , x2 a2 )|| < r}
by definition of p the vector addition in R2
2
= {(x1 , x2 ) R : (x1 a1 )2 + (x2 a2 )2 < r}
by definition of || ||
= {(x1 , x2 ) R2 : (x1 a1 )2 + (x2 a2 )2 < r2 }

i.e. Br (a) is the set of points inside the circle centred on (a1 , a2 ) of radius r > 0.

In particular

B2 ((2, 1)) = {(x1 , x2 ) R2 : (x1 2)2 + (x2 1)2 < 4}

is the set of points inside the circle


16 CHAPTER 1. VECTOR SPACES
6 (2, 3)

(0, 1) (2, 1) (4, 1)


-

(2, 1)

Example 1.5.4 In (R2 , || ||), with || || the norm on R2 defined for x = (x1 , x2 )
by ||x|| = |x1 | + |x2 |. Let a = (a1 , a2 ) R2 and let r > 0 then

Br (a) = {x R2 : ||x a|| r}


= {(x1 , x2 ) R2 : ||(x1 , x2 ) (a1 , a2 )|| r}
= {(x1 , x2 ) R2 : ||(x1 a1 , x2 a2 )|| r}
by definition of the vector addition in R2
= {(x1 , x2 ) R2 : |x1 a1 | + |x2 a2 | r}.

Now we examine four cases: (i) x1 a1 0 and x2 a2 0. We have


|x1 a1 | + |x2 a2 | = x1 a1 + x2 a2 and so

x1 + x2 r + a1 + a2 .

(ii) x1 a1 0 and x2 a2 0 yields

x1 x2 r + a1 a2 ,

(iii) x1 a1 0 and x2 a2 0 yields

x2 x1 r a1 + a2

and (iv) x1 a1 0 and x2 a2 0 yields

x1 x2 r a1 a2 .

These four inequations describe regions of the plane, the intersection of which
gives Br (a).
1.5. SUBSETS OF RN . 17
(2, 3)
6 @
@
@
@
@
(0, 1) (2, 1) @ (4, 1)
@
@
@ -
@
@
@
(2, 1)

In particular for B2 ((2, 1)) from (i) we have x1 + x2 2 + 2 + 1 = 5 (i.e.


x2 5 x1 ), from (ii) we have x1 x2 2 + 2 1 = 3 (i.e. x2 x1 3), from
(iii) we have x2 x1 2 2 + 1 = 1 (i.e. x2 x1 + 1) and from (iv) we have
x1 + x2 2 + 1 2 = 1 (i.e. x2 1 x1 ). The intersection of these four regions
is B2 ((2, 1)).

Proofs using balls (subsets of Rn )

If n 3 we will be able to draw pictures , but when n 4 we cannot draw


in 4-dimensional space, although we will see that we can obtain a schematic
representation.

Typically we will want to prove statements of the type


AB
i.e. x V if x
| {z A} then x B} and
| {z
hypothesis conclusion
A 6 B
i.e. x A such that x 6 B.

Proof of the second type of statement involves finding a specific vector x which
satisfies the statement.

Example 1.5.5 We will prove that in R2 with the usual norm


B1 ((3, 1)) B2 ((2, 1)); i.e. x B1 ((3, 1)), x B2 ((2, 1)).

Proof: Let x B1 ((3, 1)), i.e. ||x (3, 1)|| < 1, where x = (x1 , x2 ).
18 CHAPTER 1. VECTOR SPACES

To Prove: ||x (2, 1)|| < 2.


Now ||x (2, 1)||
Notice that ||x (2, 1)||
= ||x (3, 1) + (3, 1) (2, 1)||
= ||x (3, 1) + (3, 1) (2, 1)||
||x (3, 1)|| + ||(3, 1) (2, 1)||
||x (3, 1)||+
<2
||(3, 1) (2, 1)||
since
||(3, 1) (2, 1)|| = 1. p ||(3, 1) (2, 1)|| =
and
(3 2)2 + 02 = 1.

So x B2 ((2, 1)) as required.


[Notice that the proof uses the norm on R2 , since the distance between the
centres of the balls is determined by the norm.]
Example 1.5.6 We will prove that B1 ((1, 1)) 6 B2 ((0, 0)) in (R2 , || ||), where
|| || is the usual norm i.e.
x B1 ((1, 1)) such that x 6 B2 ((0, 0)).

( Exploration) Find a point along the line from (0, 0) through (1, 1).
Notice q
||(1 12 , 1 21 ) (0, 0)|| = 2 49

3 2
= 2 >2
so the point (1 21 , 1 12 ) 6 B2 ((0, 0)).

And ||(1 21 , 1 12 ) (1, 1)|| = 1/ 2 < 1.

 

-


Proof: Choose x = (1 21 , 1 12 ), then ||x (1, 1)|| = ||( 12 , 12 )|| = 1/ 2 < 1 so
x B1 ((1, 1)) and ||x (0, 0)|| > 2 (by the box above) and so x 6 B2 ((0, 0)).
Hence B1 ((1, 1)) 6 B2 ((0, 0)) as required.
Example 1.5.7 We want to find a number r > 0 such that
Br ((1, 0)) B2 ((0, 0)); in R2 with the usual norm
1.5. SUBSETS OF RN . 19

and then prove that the choice of r works.

Exploration:

| | -
1 2

To
choose r > 0, notice that the distance between 1 and 2 on the hori-
zontal axis is 1, so can choose 1 (or any real number between 0 and
1).

Consider r = 1. We have to prove that

B1 ((1, 0)) B2 ((0, 0)), i.e. x B1 ((1, 0)), x B2 ((0, 0)).

Proof: Let x B1 ((1, 0)), then ||x (1, 0)|| < 1. Now
||x (0, 0)|| = ||x (1, 0) + (1, 0) (0, 0)||
||x (1, 0)|| + ||(1, 0) (0, 0)||
<1+1=2
and so x B2 ((0, 0)).

We will want to prove statements about balls which are more complicated than
those we have seen in examples 1.5.5, 1.5.6 and 1.5.7. A particular example
is given by an implication in which the conclusion is a universally quantified
implication; P A B is such a statement, since it says if P then a A, a
B. This can be seen as a nested implication in the following way; if P then
(if a A then a B). Thus the conclusion consists of an hypothesis and a
conclusion. The statement a A is the inner hypothesis (IH) and a B
is the inner conclusion (IC). To write a coherent proof of such a statement one
can follow the following steps:
i. Assume the hypothesis P (but dont say anything more about what it means
here.)
20 CHAPTER 1. VECTOR SPACES

0. Now we are to prove the conclusion that A B. At this point say what
this means; i.e. a A a B.

1. Assume the inner hypothesis; let a A, and say more to explain what this
means if the explanation is helpful to reach the inner conclusion.

2. To prove the inner conclusion: i.e. x B

3. Explore to see how to prove the inner conclusion that x B using the
hypothesis P and the inner hypothesis.

4. Write out the proof that x B, thereby proving that the conclusion A B
follows from the hypothesis P .

Example 1.5.8 We prove that in a normed vector space V


(r, s > 0), (a, b V )

Br (a) Bs (b) B2r (2a) B2s (2b).

Proof: Let r, s > 0, let a, b V .

i. Suppose Br (a) Bs (b).

0. To prove: B2r (2a) B2s (2b).

1. (Assume the inner hypothesis) Let x B2r (2a) i.e. ||x 2a|| < 2r.

2. To prove: x B2s (2b) i.e. ||x 2b|| < 2s.

3. We must use the hypothesis Br (a) Bs (b); i.e. y Br (a) y


Bs (b)
i.e. ||y a|| < r ||y b|| < s; as well as the inner hypothesis at
(1). We notice that we can use (1), homogeneity of norm then the
hypothesis and again homogeneity of norm, to conclude.

4. Now
||x 2a|| < 2r by the IH (1)
||1/2x a|| < r homogeneity of norm
||1/2x b|| < s by the hypothesis (i)
||x 2b|| < 2s homogeneity of norm

thus x B2r (2a) x B2s (2b)


i.e. B2r (2a) B2s (b) as required.
1.5. SUBSETS OF RN . 21

Other types of subsets of Rn

Open balls can be used to define other types of subsets of Rn .

By saying that a set is open we mean that it does not include its boundary.
In a diagram we can use a dotted line. Since an open set doesnt include its
boundary, every point in an open set is surrounded by points which also lie in
the set.
Definition 1.5.9 A subset S of Rn is open if and only if every point of S is the
centre of an open ball which also lies in S; i.e. (s S) (r > 0) : Br (s) S.
. Example An open ball is an open set.
We have to prove x Br (a), B (x) Br (a). Proof: Consider the open
ball Br (a), in normed vector space (V, || ||). Let x Br (a). Choose =
r ||x a|| > 0. Let y B (x) so ||x y|| < . Now
||y a|| = ||y x + x a||
||y x|| + ||x a||
< r ||x a|| + ||x a||
=r

so y Br (a).

Closed sets contain their boundary. In a diagram we can use a solid line. The
complement of a closed set T in Rn is an open set (it does not contain its
boundary) and likewise the complement of an open set in Rn is a closed set.
Definition 1.5.10 A subset T of Rn is closed if and only if its complement
Rn \T is open.
Example A closed ball is a closed set. We are to prove that V \Br (a) is an
open set, i.e. x V \Br (a), B (x) V \Br (a). Proof: Let x V \Br (a), so
||x a|| > r. Choose = ||x a|| r > 0. Let y B (x) so ||x y|| < . Now
||x y|| < ||x a|| r so
r < ||x a|| ||x y||
||y a||

from the exercises, so y V \Br (a).

Bounded sets are sets which never wander off to infinity.


Definition 1.5.11 A subset B of a vector space V is bounded if and only if it
is contained in some open ball in V ; i.e. (M > 0, a V ) : B BM (a).

Example Consider B = [3, 3] (2, 2) R2 , with the usual norm. We have


B B13 (0). So B is a bounded subset of R2 .
22 CHAPTER 1. VECTOR SPACES

Dense subsets of a vector space V are sets for which every open ball in V contains
a point of the set.

Definition 1.5.12 A subset D of V is dense in V if and only if every open


ball contains a point of D; i.e. (d V ), (r > 0) : Br (d) D 6= .

Example The set of rational numbers is dense in the set of real numbers with
the usual norm.

The systematic study of such properties is the subject matter of topology (in
particular point-set topology).

1.6 Mappings
The aim of this section is to give some geometric imagery to the principal object
of focus in the first part of MATH 333, namely functions between normed vector
spaces.
Functions are usually represented geometrically by their graphs or by arrow (or
mapping) diagrams. For functions f : R R, the graph of f is
{(x, f (x)) : x R} and so is a subset of R2 ; f : R2 R, the graph of f is
{(x, f (x)) : x R2 } R2 R = R3 since f (x) R; f : R R2 , the graph of
f is {(x, f (x)) : x R} R R2 since f (x) R2 , thus f (x) = (f1 (x), f2 (x))
where f1 , f2 are the real valued component functions of f . All of these graphs
can be drawn either in R2 or R3 .
It is not possible to sketch the graph of a function f : R2 R2 since the
graph of f is a subset of 4-dimensional space. For such functions the emphasis
shifts away from graphs to arrow diagrams. By studying arrow diagrams for such
functions one can understand something about the behaviour of the function.
These diagrams show the domain (or a subset of the domain) on the left and
the codomain (showing the range or image of the subset of the domain) on the
right.

Definition 1.6.1 Given a mapping f : X Y between two sets X and Y ,

f (A) = {f (a) : a A X}

is the image of the set A under the mapping f .

Examples of maps f : R R

For such maps we can either draw an arrow diagram or draw the graph.

Example 1.6.2 Consider the squaring function f : R R, with f (x) = x2


and let A = [1, 2] R. Then f (A) = [1, 4]. The graph is
1.6. MAPPINGS 23

u
4 6

f (A)

1 t

[ ]-
1 A 2 And the arrow diagram is

R R

6 6
2u u4
A |
P PP
1t PP
PP
Pq
P f (A)
0
1

1 u 
 t1

| 

2 t

Example 1.6.3 Consider the signum function f : R R, with


1 if x > 0
f (x) = 0 if x = 0 The graph of f is
1 if x < 0.

24 CHAPTER 1. VECTOR SPACES

1( -

 ) 1

And the arrow diagram is

R R
6 6
2
6X
XX
XXX
1 Xz
X 1

0^ - 0
_

1 :
 1


?
? which shows that points x (0, ) map to 1,

x (, 0) map to -1 and 0 maps to 0.

Examples of maps f : R2 R
A function of this type is a real-valued function of two variables. For f : R2 R
the graph of f is a subset of R2 R and so one can draw the graph in R3 . The
graph of f is typically a surface in R3 .

Example 1.6.4 The function f : R2 R with f (x1 , x2 ) = x21 + x22 .

Sketching an arrow diagram helps with sketching the graph. For this function,
since x21 + x22 0 for all values of x1 , x2 , we would consider the set of all points
1.6. MAPPINGS 25

in the domain which map to a given c 0 in the codomain. The set

{(x1 , x2 ) R2 : f (x1 , x2 ) = c} = {(x1 , x2 ) R2 : x21 + x22 = c}

is referred to as the level set of f at c.


For each c 0, the level set of f at c is a circle in the plane with centre (0, 0)
and radius c. The following is the arrow diagram for c = 0, 1, 2, 4.
6 6
6 6

- -

Now each point in R2 is on such a circle and so we can build up a picture of


how f maps the entire plane, by considering how it maps the circles. In order
to sketch the graph of f we take the vertical axis as codomain
and match its
origin with the (0, 0) of R2 . Above each circle of radius c in the domain of f
we then construct a circle in space at a height of c. We show this below.

Examples of maps f : R R2 A function of this type is usually referred to


as a vector valued function. For f : R R2 the graph of f is a subset of R R2
and so one can draw the graph in R3 . The graph of f is typically a curve in R3 .

Example 1.6.5 Consider the function f : R R2 with f (x) = (cos x, sin x).
Again it is useful to draw an arrow diagram as an aid to drawing the graph of
f . Notice that for any x R, the point (cos x, sin x) lies on a circle in R2 of
1
radius 1, since ||(cos x, sin x)|| = (cos2 x + sin2 x) 2 = 1.
26 CHAPTER 1. VECTOR SPACES

6 6

So as x moves along the line R, the point f (x) keeps moving around the unit
circle. For the choices x = 0, /2, , 3/2 we get the arrows indicated.
The function f has period 2 and so as x moves further up the x axis the
arrows keep coming back to the same points on the unit circle.
To sketch the graph of f , we make the domain axis the first axis, pointing
out of the page and the codomain is the plane of the page, the second and
third axes. As x increases, the point (x, f (x)) on the graph keeps coming out
towards us along the first axis, while f (x) keeps turning around in the plane
of the second and third axes. The graph looks like a stretched out spring, it is
called a helix.

Examples of maps f : R2 R2

Example 1.6.6 Given c R2 we define the constant map c : R2 R2 by


putting, for each x R2 , c(x) = c. So the constant map sends every point in
the plane to the single point c R2 . The arrow diagram is below.

domain codomain
6 c 6
c
- -

Example 1.6.7 The identity map Id : R2 R2 is defined by putting , for each


x R2 Id(x) = x. Thus the identity map leaves each point in R2 where it was.
For the subset B = [1, 1] [1, 1] of R2 the arrow diagram is below.
1.6. MAPPINGS 27

domain codomain

6 6
- -
Id

Example 1.6.8 An expansion map such as f : R2 R2 is defined by putting,


for each x R2 , f (x) = 2x. Thus f maps each arrow (thinking of the vectors
as arrows) to another arrow, with the same direction but with twice the length.
We say, that f is a uniform expansion by a factor of 2 away from the origin.
For the closed ball B1 (0, 0) the arrow diagram is below.
domain  codomain
6 f 6 2x

x | -
- -

Example 1.6.9 Given a vector a R2 , let f : R2 R2 be defined by putting,


for each x R2 , f (x) = x + a. The effect of f on the point x is to translate
it through the vector a for the closed square B = [1, 1] [1, 1] the arrow
diagram is below.

domain  codomain

6 f 6 
@
I 
* 
@ x
| - f (x)
a@ -  -

Example 1.6.10 Let f : R2 R2 be the function with f (x) = 2x.

To prove: f (B2 (0, 1)) B4 (0, 2).

Proof:
f (B2 (0, 1)) = {f (x) : x B2 (0, 1)} by definition 1.6.1
= {2x : ||x (0, 1)|| < 2} by definition of f and
of B2 (0, 1)
{2x : 2||x (0, 1)|| < 4}
{2x : ||2x (0, 2)|| < 4} by homogeneity of norm
= {y : y = 2x and ||2x (0, 2)|| < 4} y renames 2x
= {y : ||y (0, 2)|| < 4} = B4 (0, 2).

Thus f (B2 (0, 1)) B4 (0, 2) as required.


28 CHAPTER 1. VECTOR SPACES
Chapter 2

Limits and Continuity

2.1 An Intuitive Approach

The expression
lim f (x) = l
xa

means f (x) approaches l as x approaches a or


we can make f (x) close to l by making x close to a ( but x 6= a).
Of course we will want to make this more precise.
The following example is to help you recall one way in which limits arise in
elementary calculus.
From first principles we find the derivative, at the point 2, of the function
g : R R with g(x) = x2 + 3.
The Newton quotient is
g(x) g(2) (x2 + 3) (22 + 3) x2 22
= = (x 6= 2)
x2 x2 x2
which measures the slope of the line (or chord) joining the two points (2, g(2))
and (x, g(x)) on the graph. The condition that x 6= 2 is most important, both
from an algebraic and a geometric point of view. Algebraically division by zero
is not defined and geometrically, we need two distinct points to determine a line,
so the chord is undefined when x = 2.

To calculate g 0 (2), which is the slope of the tangent at the point (2, g(2)), we
now let x approach 2, giving
x2 2 2
g 0 (2) = lim = lim (x + 2) = 4. (2.1)
x2 x 2 x2

The essential step in the calculation is to cancel the factor x 2, which is valid

29
30 CHAPTER 2. LIMITS AND CONTINUITY

because we are supposing that x 6= 2. So this step really involves two functions
2
22
f1 : R\{2} with f1 (x) = xx2 and f2 (x) = x + 2. In (2.1) the second last = is
really asserting that lim f1 (x) = lim f2 (x), which is justified by the fact that
x2 x2
these functions have the same value at each point of R\{2} and f1 is not defined
at 2, whilst f2 (2) = 4. The last step lim f2 (x) = f2 (2) is justified by the fact
x2
that f2 is continuous at 2.
So a function f : R R is continuous at a point a
if a domf and lim f (x) = f (a).
xa

For each of the following, draw the graph of the function to convince yourself
of the truth of the statement:

1. f : R\{2} R with

3x if x < 2
f (x) =
x1 if x > 2

has the limit 1 at 2 but is not continuous at 2.

2. f : R R with 
3x if x 2
f (x) =
x1 if x > 2
has the limit 1 at 2 and is continuous at 2.

3. f : R R with
3x if x < 2
f (x) = 3 if x = 2
x1 if x > 2

has the limit 1 at 2 but is not continuous at 2.

4. f : R R with 
x if x < 2
f (x) =
x+1 if x 2
does not have a limit at 2 and is therefore not continuous at 2.

Limits in higher dimensions.

We will consider functions of types f : V W and f : V \{a} W , where


V, W are normed vector spaces.

We want to understand the statement

lim f (x) = l. (2.2)


xa
2.1. AN INTUITIVE APPROACH 31

The various quantities can be depicted as in the diagram. The idea which (2.2)
is supposed to convey is that

if x is close, but not equal to a, then f (x) is close to l. (2.3)

The closeness can be measured by the norms on the two spaces V and W .
Statement (2.3) still lacks precision.
Definition 2.1.1 A decentred ball of centre a and radius r > 0 is the set

Br (a)\{a}.
o o
We will denote the decentred ball by Br (a), i.e. Br (a)\{a} := Br (a).

In the following examples, we consider some functions for which it is intuitively


clear whether or not the statement lim f (x) = l is true. For each function
xa
we consider a collection of decentred balls in the domain, which shrink up to a
point a and ask how the images of the balls in the codomain behave.

1. Let f : R2 R2 be defined by f (x) = 2x for each x R2 . We want to see


whether or not lim f (x) exists, so we examine a collection of decentred
x0
balls in the domain, centred on 0.

Recall that this map f doubles the radius of each ball with centre at
0 = (0, 0). Check the series of mapping diagrams in which the decentred
ball in the domain gets smaller.
32 CHAPTER 2. LIMITS AND CONTINUITY

We see that as the ball A shrinks towards 0, so too does its image, in-
dicating that lim f (x) exists and is 0 in the codomain. (Note: in this
x0
example the conclusion holds whether or not we regard the balls as being
decentred. This is because f is continuous at 0. )

2. Let f : R2 R2 be defined by f (x) = 2x for each x R2 . We want to see


whether or not lim f (x) exists, so we examine a collection of decentred
x(1,1)
balls in the domain, centred on (1, 1).
The mapping f doubles the radius of each ball A and shifts its centre from
(1, 1) to (2, 2).
We see that as the ball A shrinks towards (1, 1), its image, shrinks towards
(2, 2) indicating that lim f (x) exists and is (2, 2) in the codomain.
x(1,1)
Again the result holds irrespective of whether we consider decentred balls
in the domain.

3. In this example, since we can draw the graph of f , we will demonstrate


the procedure using the graph, rather than a mapping diagram.
Let f : R\{2} R be defined by f (x) = x + 1 for each x R\{2}. We
want to see whether or not lim f (x) exists, so we examine a collection of
x2
decentred balls in the domain, centred on 2.
Clearly f preserves the radius of each decentred ball of centre 2 while
shifting its centre to the point 3.

We see that as the decentred ball A shrinks towards the point 2, its image
shrinks towards the point 3 in the codomain, indicating that lim f (x)
x2
2.2. DEFINITIONS AND EXAMPLES 33

exists and is 3 in the codomain. In this example we were forced to use


decentred balls because 2 is not in the domain of f .

4. The last example differs in that the given limit does not exist.


x if x < 2
Let f : R R be defined by f (x) =
x + 1 if x 2
We examine the effect of f on a collection of balls in the domain shrinking
up to the point 2.

6 6 6
7
 7
 7

  
  
  

  
  
  
 -  -  -

Each ball A in the domain of centre 2 contains the point 2 and a point
below 2. So the image f (A) contains the point 3 and another point which
is less than 2. Indeed for each ball A in the domain, f (A) contains at least
two points whose distance apart is bigger than 1. Thus as the ball A of
centre 2 shrinks up toward 2, its image f (A) cannot shrink up toward a
point in the codomain.

2.2 Definitions and Examples


Definition 2.2.1 The function f : S V W has a limit l at a

(B)(A) such that f (A) B,

where A is a decentred open ball, centred at a V , B is an open ball centred


at l W and S is either a subset of V or a subset of V \{a}.

To say that f has a limit at a we add the existentially quantified phrase l W ,


i.e.
34 CHAPTER 2. LIMITS AND CONTINUITY

f has a limit at a (l W ) such that (B)(A) such that f (A)


B, where A is a decentred open ball, centred at a V and B is an
open ball centred at l W .

Towards the end of the section we translate definition 2.2.1, into a statement
involving norms and inequalities.
Notice that given f : S V W , a V , l W in order to prove the statement
lim f (x) = l we must:
xa

1. fix an open ball B of centre l in the codomain,

2. choose a (decentred) ball A, centred on a in the domain and

3. prove the set containment


f (A) B.

Example 2.2.2 Let f : R2 R2 be defined by f (x) = 2x. We prove that


lim f (x) = (2, 0).
x(1,0)

We are to prove
(B)(A) such that f (A) B,
where A (decentred) and B are open balls of centre (1, 0) and (2, 0) respec-
tively.

Proof: Let B = B (2, 0) (= {x R2 : ||x (2, 0)|| < }).


o
Choose A = B/2 ((1, 0)).
Then
f (A) Exploration
= {f (x) : x A} To find a (decentred) open ball A.
= {2x : 0 < ||x (1, 0)|| < /2} Notice that f doubles the radius.
{2x : | 2|||x (1, 0)|| < } So choose A with radius /2.
= {2x : || 2x (2, 0)|| < }
= {y : ||y (2, 0)|| < }
= B.

Example 2.2.3 Let f : R\{0} R with f (x) = x2 . We will prove that f has
a limit of 0 at 0; i.e. lim f (x) = 0.
x0

We are to prove the statement:

(B)(A) such that f (A) B,

where A (decentred) and B are open balls of centre 0 (in the domain) and 0 in
the codomain respectively.
2.2. DEFINITIONS AND EXAMPLES 35

Proof: Let B = B (0) R.


o
Choose A = B (0).
Then
f (A) Exploration
= {f (x) : x A} To find a (decentred) open ball A

= {x2 : 0 < |x 0| < } so that f (A) B.
{x2 : |x|2 < } f is the squaring function

= {x2 : |x2 | < } so choose A with radius .
= {y : |y| < }
= B.

Example 2.2.4 Let f : R\{1} R with f (x) = 3x + 2. We will prove that


lim f (x) = 5.
x1

We are to prove the statement:

(B)(A) such that f (A) B,

where A (decentred) and B are open balls of centre 1 (in the domain) and 5 in
the codomain respectively.

Proof: Let B = B (5) R.

Exploration
To find a (decentred) open ball A
so that f (A) B.
Notice that f maps
o
Choose A = B/3 (1).
Let y f (A) 6 6
y = f (x)
for some x B/3 (1)\{1}
y = 3x + 2 1 + _ _5 + 3
for some x B/3 (1)\{1}.
f
Now /3 < x 1 < /3 1 - 5
so < 3x 3 <
< 3x + 2 5 <
1 ^ ^5 3
so |y 5| <
thus y B (5) = B
as required.
each ball of radius > 0
and centre 1 to a ball of
radius 3 and centre 5.
So choose A with radius /3.
36 CHAPTER 2. LIMITS AND CONTINUITY

Examples of disproofs. Given f : S V W we want to prove the


statement
f does not have a limit l at a i.e.
(B) such that (A) f (A) 6 B,
where A (decentred) and B are open balls of centre a (in the domain)
and l in the codomain respectively.
And to show that lim f (x) does not exist we must prove the statement
xa
(l W )(B) such that (A) f (A) 6 B,
where A (decentred) and B are open balls of centre a and l respectively.
The core of the proof is to find a suitable B. The main strategy used to find B is
to find two points in an arbitrary ball A whose images stay at least some fixed
distance d apart, regardless of how small the radius of A becomes. Suppose
|f (a1 ) f (a2 )| always exceeds d, then choose B to be an open ball having
diameter d and it is clear that f (a1 ) and f (a2 ) cannot both be in B and so one
has f (A) 6 B.

x+2 x1
Example 2.2.5 Let f : R R be defined by f (x) =
4x x > 1.
We are to prove the statement
(l R)(B) such that (A) f (A) 6 B,
where A (decentred) and B are open balls of centre 1 and l respectively.

Proof: Let l R.
Exploration
To find an open ball B
Choose B = B1/2 (l). so that for any (decentred)
Let A be a (decentred) open ball A centred on 1
ball of radius > 0 f (A) 6 B.
centred on 1. Consider Examine the diagram:
a1 = 1 + 12 , a2 = 1 21
a1 , a2 A, a1 > 1, a2 < 1.
So f (a1 ) = 4a1 = 4 + 2
6 6f (a1 )
and f (a2 ) = a2 + 2 = 3 12 . *

 4)
Thus |f (a1 ) f (a2 )| = 1 + _
gap of 1
|4 + 2 (3 12 )| = 1 + 52 a1
f
> 1, since > 0, 1 7 3
so at least one of a2 f (a2 )
f (a1 ), f (a2 ) is not in B. 1 ^ -
Hence f (A) 6 B
as required

So choose B with diameter 1.


2.2. DEFINITIONS AND EXAMPLES 37

Example 2.2.6 Let f : R R with



1 if x is rational
f (x) =
2 if x is not rational

We will prove that f does not have a limit at any point a R.

We are to prove the statement

(a R( domain ))(l R( codomain )(B) such that (A) f (A) 6 B,

where A (decentred) and B are open balls of centre a and l respectively.

Proof: Let a R and let l R.

Choose B = B1/2 (l). Exploration


Let A be a (decentred) To find an open ball B
ball of radius > 0 so that for any (decentred)
centred on a. Let open ball A centred on a,
a1 Q A, a2 (R\Q) A f (A) 6 B.
a1 , a2 A. Examine the diagram:

. . .. 2. .6. . . . . . . . . f (R\Q)


gap of 1
So f (a1 ) = 1 and f (a2 ) = 2 . . . . 1. . . . . . . . . . . f (Q)
thus |f (a1 ) f (a2 )| = 1
But the radius of B is 1/2 ( | ) -
a a a+
so at least one of
f (a1 ), f (a2 ) is not in B.
Hence f (A) 6 B Notice f (A) always contains the
as required points 1 and 2,
since A is an interval in R,
so it contains both rational
and irrational numbers.
So choose B with diameter 1.

Inequality form of definition 2.2.1.


We said in definition 2.2.1 that the function f : S V W has a limit l at a

(B)(A) such that f (A) B,

where A is a decentred open ball, centred at a V , B is an open ball centred


at l W and S is either a subset of V or a subset of V \{a}.
38 CHAPTER 2. LIMITS AND CONTINUITY

First we rewrite this as follows: f : S V W has a limit l at a


( open ball B of centre l and radius > 0),
( an open (decentred) ball A of centre a and radius > 0)
such that f (A) B.
Second we translate using the definition of open balls: we know B = B (l) =
o
{y W : ||y l|| < }, A = B (a) = {x S : ||x a|| < }\{a} (since A is a
decentred ball) and so the statement f (A) B is the statement if y f (A) then
y B; i.e. if y = f (x) for some x A then y = f (x) B; i.e. if ||x a|| <
then ||f (x) l|| < . So that definition 2.1.1 becomes if x A then f (x) B.

Definition 2.1.1 f : S V W has a limit l at a

( > 0), ( > 0) such that x S,


if 0 < ||x a|| < then ||f (x) l|| < .

2.3 Continuity
Generalising from what we had for real valued functions on a real domain, we
have that a function f : S V W is continuous at a point a if and only if
a S = domf and lim f (x) = f (a). There are two changes, the point a is in
xa
the domain of f and consequently we will not decentre the ball in the domain
and the limit of f at a is f (a), so the ball B in the codomain will be centred on
f (a). Thus in the form of definition 2.1.1 we have
Definition 2.3.1 The function f : S V W is continuous at a
(B)(A) such that f (A) B,
where A is an open ball, centred at a V and B is an open ball centred at
f (a) W .
And in terms of inequalities we have

Definition 2.3.1 f : S V W is continuous at a


( > 0), ( > 0) such that x S,
if 0 ||x a|| < then ||f (x) f (a)|| < .

To say that the function f : S V W is continuous we must add (to


the beginning of the statement) the quantified phrase (a S), so that the
statement holds at every point a S.
We only speak of continuity at points in the domain of f , so in example 2.2.3
we cannot speak of continuity at 0 (although we can speak of existence of limits
at 0). Of course, if we had defined the function as f : R R with f (x) = x2
then we can speak of continuity at 0. Indeed since the limit of f at 0 is 0 and
f (0) = 02 = 0 it is clear that f is continuous at 0.
2.3. CONTINUITY 39

Example 2.3.2 Given f : R R with



(x + 2)2 x > 2
f (x) = 2x + 4 x < 2
4 x = 2

We will show (using the definitions) that (i) lim f (x) = 0 and (ii) f is not
x2
continuous at -2.
(i) We are to prove the statement
(B)(A) such that f (A) B,
where A is a decentred open ball, centred at 2 R and B is an open ball
centred at 0 R.

Proof: Let B = B (0).


Exploration
To find a (decentred) open ball
A, so that f (A) B.
Notice that f maps

Choose = min{1, /2} and 6 6


o
A = B (2).
Now f (A) = {f (x) : x A}
2 + _ _
= {(x + 2)2 : 2 < x < 2 + }
{2x + 4 : 2 < x < 2} f
{(x + 2)2 : 2 < (x + 2)2 < 0} 2 b - 0
{2x + 4 : 2 < 2x + 4 < 0}
{y = f (x) : 2 < y < 0} 2 ^ ^
since 0 < 1 so2 < 2
{y : < y < }
since 2 2 to
= (, ) = B (0) = B 2(2 ) + 4 = 2
as required. and 2 + to
(2 + + 2)2 = 2 .
In addition, for 1,
2 < 2 .
So choose 1 and /2.
i.e. = min{1, /2}
Thus f has a limit at -2 (of 0).

(ii) We are to prove the statement


(B) such that (A), f (A) 6 B,
40 CHAPTER 2. LIMITS AND CONTINUITY

where B is a ball centred at f (2) (i.e. at 4) and A is a ball centred at -2.

Proof:

Exploration
Consider B = B4 (4). We first have to find a suitable
Let A = B (2). open ball B.
Now consider x = 2 /2 Notice that f maps
then x B (2) and 2 to 4 and
f (x) = 2(2 /2) + 4 = < 0 every other real number x to
a negative real number.

6 6
J
J 4
J  )
2 + _ J gap of 4
J
so f (x) 6 B4 (4) = B. f J
2 J 0
Thus f (A) 6 B.
^
-
2 ^

So choose the radius of B


to be 4 (or 0 < 4).

Example 2.3.3 Let f : R2 R be given by

x2 + y 2 + 3 (x, y) 6= (0, 0)

f (x, y) =
0 (x, y) = (0, 0)

We will prove that (i) lim f (x, y) = 3 and (ii) f (x, y) is not continuous at
(x,y)(0,0)
(0, 0).

(i) We are to prove the statement

(B)(A) such that f (A) B,

where A is a decentred open ball, centred on (0, 0) R2 and B is an open ball


centred at 3 R.
2.3. CONTINUITY 41

Proof: Let B = B (3).

Exploration
To find a (decentred) open ball A
so that f (A) B.

Choose = > 0 and 6 6
A = B ((0, 0))\{(0, 0)}.
Then f (A) = {f (x, y) : (x, y) A} 
{x2 + y 2 + 3 : x2 + y 2 < 2 = } - 3+
= {z R : 3 < z < 3 + } 

z replaces x2 + y 2 + 3 f
3
(3 , 3 + ) = B (3) = B
as required.
3


Choose = .

(ii) We are to prove the statement

(B) such that (A), f (A) 6 B,

where B is a ball centred at f (0, 0) (i.e. at 0) and A is a ball centred at (0,0).

Proof:
Consider B = B1 (0).
Let A = B (0, 0). Exploration
 
Now consider(x, y) = 1 , We first have to find a suitable
2 2 2 open ball B.
then (x, y) B ((0, 0)) and Notice that (x, y) R2
f (x, y) = 14 2 + 3 > 3 with (x, y) 6= (0, 0), f (x, y) > 3,
so f (x, y) 6 B1 (0) = B. so choose 3.
Thus f (A) 6 B.

In the previous example, suppose we had f (0, 0) = 4, instead of 0. This time


we need to control the in the choice of the point (x, y). For example we could
choose B = B1/2 (4), then let A
  = B ((0, 0)) and = min{1/2, }. Now consider
1
the point (x, y) = , . We have (x, y) A, since ||(x, y) (0, 0)|| =
2 2 2

2 < . And f (x, y) = 14 2 +


3 161
+ 3, since 1/2 thus f (x, y) < 3 12 ,
so f (x, y) 6 B1/2 (4) = B. The choice of B and is determined only by the
distance between 3 and 4 so it is not unique. You should try some other choices.
42 CHAPTER 2. LIMITS AND CONTINUITY

In fact after the next section you will realize that in the case f (0, 0) = 4, we
could have used the local principle for limits.

2.4 Theorems on Limits and Continuity


Theorem 2.4.1 (Uniqueness of limit). If a function f has two limits l and l0
at a point a, then l = l0 .

We will prove the theorem using the definition of limit given by 2.1.1.

Proof: Suppose f has two limits l and l0 at a.

Assume that l 6= l0 . (Exploration). To prove


Thus 1/2||l l0 || > 0. Choose open balls that l = l0 we will proceed
B, B 0 of centre l, l0 and radius 1/2||l l0 || by contradiction.
respectively. How to derive the false
Then B B 0 = . (Proof?) statement?
Since f has a limit l at a, A a If we choose two open balls
(decentred) open ball centred on a B, B 0 centred on l, l0
such that f (A) B. respectively with radius
Similarly since f has a limit l0 at a, 1/2||l l0 ||, then B
A0 a (decentred) open ball centred on a and B 0 are disjoint.
such that f (A0 ) B 0 . But since f (x) is supposed
Now A A0 is the smaller of the two to be close to both l and
balls, since they have the same centre. to l0 , when
Also f (A A0 ) B x is close to a,
and f (A A0 ) B 0 . f (x) must belong to both
So f (A A0 ) B B 0 = . balls. This is not possible.
But A A0 6=
and so f (A A0 ) 6= .
This is the required contradiction.
Hence l = l0 .

A consequence of theorem 2.4.1 is that instead of speaking of a limit at a we


speak of the limit at a. This is the tacit assumption when you write lim f (x).
xa

2.4.1 Limit of a composite.


Let f : U V and g : V W be maps between normed vector spaces. Then
x U , g f : U W is defined, since the range of f is a subset of the domain
of f
2.4. THEOREMS ON LIMITS AND CONTINUITY 43

U V W
f g
- -

@
@
gf

and (g f )(x) = g(f (x)).


Example 2.4.2 Let f : R2 R with f (x, y) = x + 2y and let g : R R with
g(x) = 3x2 + x + 1. Then f g is not defined, since the range of g is not a subset
of the domain of f . And g f : R2 R is the map with

(g f )(x, y) = g(f (x, y))


= g(x + 2y)
= 3(x + 2y)2 + (x + 2y) + 1.

Theorem 2.4.3 (Limit of a composite.)

We suppose f : U V (or U \{a} V ) and g : V W . If g is continuous at


l and lim f (x) = l then
xa

lim (g f )(x) = g( lim f (x)).


xa xa

Proof:
Assume the hypothesis i.e. g is continuous at l and lim f (x) = l.
xa

Let C be an open ball centred on Exploration


g(l) W . To prove
Let B be an open ball centred on lim (g f )(x) = g( lim f (x)).
xa xa
l V satisfying g(B) C i.e. given C a ball centred on
and let A be the decentred ball g(l) a ball A (decentred)
of centre a, such that f (A) B. of centre a such that
Now (g f )(A) = g(f (A)) (g f )(A) C. So to
g(B) C. start the proof write the inner
hypothesis. Now we are to prove
A of centre a and
(g f )(A) C.
By hypothesis we know B of
centre l such that g(B) C
and we know A (decentred)
of centre a such that
f (A) B.
44 CHAPTER 2. LIMITS AND CONTINUITY

Thus lim (g f )(x) = g(l) = g( lim f (x)) as required.


xa xa

Corollary 2.4.4 (Continuity of a composite). If f is continuous at a and g is


continuous at f (a) then (g f ) is continuous at a.

2.4.2 The Squeeze Principle

Suppose we have to find lim x sin( x1 ). Notice that


x0
   
1 1
0 x sin
= |x| sin |x|
x x

since sin x1 1, so that x sin( x1 ) is squeezed down to zero by |x|. We


exploit this observation in the squeeze principle.

Theorem 2.4.5 Let f : V W (or V \{a} W ), let a V and let l W . If


: V R+ {0} (or V \{a} R+ {0}) satisfying:

(i) ||f (x) l|| (x), x 6= a and

(ii) lim (x) = 0


xa

then lim f (x) = l.


xa

Proof: Suppose such that (i) and (ii) are true


Let B be an open ball centred on Exploration
l W , of radius r > 0. We are to prove lim f (x) = l
xa
Now choose A a decentred ball i.e. (IH) if B is a ball of centre l
of centre a such that (IC) then A: f (A) B
(A) Br (0) R. where A is an open decentred ball,
By (i) we have f (A) B, centred on a.
since To find A:
f (A) = {f (x) : x A} if is an open ball in R, centre 0
{f (x) : (x) Br (0)} and radius r then from (i)
by choice of A |(x) 0| < r ||f (x) l|| < r
{f (x) : f (x) Br (l)} i.e. (x) f (x) B
by (i) of hypothesis and by (ii) can choose A: (A)
=B so by (i) f (A) B.
Thus lim f (x) = l.
xa

2.4.3 Local Principle for Limits


When looking for lim f (x) we are only interested in x close to a (not x = a and
xa
not x far from a). This is reflected in the local principle for limits.
2.4. THEOREMS ON LIMITS AND CONTINUITY 45

Theorem 2.4.6 If (i) a decentred ball C of centre a such that


g(x) = f (x), x C and
(ii) lim g(x) exists and equals l
xa

then lim f (x) = l i.e. lim f (x) = lim g(x).


xa xa xa

The content of the theorem is that we can define a new function g such that
g(x) 6= f (x) for x outside a decentred ball A of centre a without affecting
lim f (x).
xa

Proof: Assume the hypothesis i.e. suppose

(i) a decentred ball C of centre a such that


g(x) = f (x), x C and
(ii) lim g(x) exists and equals l
xa

Let B be an open ball centred To prove lim f (x) = l.


xa
on l. (IH) B centred on l
Let A = A0 C, where C (IC) a decentred ball
is the decentred ball of centre a A of centre a such that
on which g(x) = f (x). f (A) B.
and A0 is the decentred ball To find A:
of centre a such that g(A0 ) B. we know from (i)
Now C of centre a:
f (A) = {f (x) : x A} f (x) = g(x) and by (ii)
= {f (x) : x A0 and x C} A0 decentred, of centre
{f (x) : g(x) B and g(x) = f (x)} a: g(A0 ) B
{y : y B} so choose A = A0 C.
=B
Hence lim f (x) = l.
xa

These theorems may be used in two ways (i) to help in the calculation of limits
of particular functions and (ii) to prove further theorems on limits.

Example 2.4.7 (Key example-squeeze principle)


Let : V \{a} R 0 with (x) = c||x a||, where c > 0, then lim (x) = 0.
xa

Proof: Let be an open ball in R of centre 0 and radius r > 0.


46 CHAPTER 2. LIMITS AND CONTINUITY

Let A be the open ball (decentred) Exploration


of centre a and radius To find A (decentred)
r/c. of centre a such that
Then (A) (A) .
= {(x) : x A}
= {c||x a|| : ||x a|| < r/c}
{c||x a|| : c||x a|| < r} _
6cr
{y : 0 y r}
. 0
Thus (), (A) 
6 7
such that (A) a
 - ^cr
where , A(decentred)are
balls of centre 0 R
and a V respectively.

Example 2.4.7 gives the idea of how to use the squeeze principle to prove
lim f (x) = l. We aim to find an inequality of the form
xa

||f (x) l|| c||x a||

and then choose (x) = c||xa||, since by the key example we have lim (x) = 0.
xa

Example 2.4.8 We will use  the squeeze principle to prove that if f : R\{0}
R with f (x) = 2 + x sin x1 then lim f (x) = 2.
x0

Proof:
Let x R\{0}. Now
1

|f (x) 2| = x sin
x 1 
= |x| sin x
1

|x| since sin x
1
= |x 0|.

So choose (x) = |x 0|, then by the key example lim (x) = 0 and so by the
x0
squeeze principle lim f (x) = 2.
x0

Example 2.4.9 We will prove that if f : R2 \{(0, 0} R with

3xy
f (x, y) = 4 + p
x2 + y 2

then lim f (x, y) = 4.


(x,y)(0,0)
2.4. THEOREMS ON LIMITS AND CONTINUITY 47

Proof: Let (x, y) R2 \{(0, 0)}. Now



|f (x, y) 4| = 3xy


2 2
x +y
= 3|x||y|
2 2
x +y
3 x2 +y 2 x2 +y 2 p
2 2 since |x|, |y| x2 + y 2
p x +y
= 3 x2 + y 2
= 3||(x, y)|| = 3||(x, y) (0, 0)||.

So choose (x, y) = 3||(x, y) (0, 0)||, then by the key example with a = (0, 0)
and l = 4 and the squeeze principle we have lim f (x, y) = 4.
(x,y)(0,0)

We will need the following projection inequalities. If (x, y), (a, b) R2 with the
usual norm (although the following statement is true for any norm on R2 ) then

|x a| ||(x, y) (a, b)|| = ||(x a, y b)|| and


|y b| ||(x a, y b)||.

These inequalities can be proved algebraically as follows:


p
|x a| = p(x a)2 defn | | on R
(x a)2 + (y b)2 since (y b)2 0
= ||(x a, y b)||.

It is also clear that the projection inequalities will be true on Rn , for n > 2.
The result is a direct consequence of positivity of the norm || || on Rn .

Example 2.4.10 We use the squeeze principle to prove that the projection
functions are continuous, where the k-th projection function Idk : Rn R is
defined by
Idk (x) = xk , where x = (x1 , x2 , . . . , xn ).

Proof: Let a = (a1 , a2 , . . . , an ) Rn . Let B = B (Idk (a)) = B (ak ).

Choose (x) = ||x a|| Exploration.


Then |xk ak | (x) We are to find :
by the projection inequalities (i) ||Idk (x) Idk (a)|| (x)
and lim (x) = 0 and (ii) lim (x) = 0.
xa xa
by the key example. Hence i.e. |xk ak | (x).
lim x = ak Choose (x) = ||x a||
xa k
by the squeeze principle. then (i) follows from the projection
inequalities and (ii)
from the key example.
48 CHAPTER 2. LIMITS AND CONTINUITY

Thus Idk is continuous at a Rn . And since a was arbitrary, we have Idk


continuous on Rn .

In the next example we use the local principle for limits.

Example 2.4.11 Let f : R R with



3x + 2 x > 0
f (x) =
x x0

We will prove that lim f (x) = 5.


x1

In example 2.2.4, we proved that if g(x) = 3x + 2, x 6= 1 then lim g(x) = 5. So


x1
with f as above we have f (x) = g(x) on C, where C is for example, the open
decentred ball of centre 1 and radius 1. We illustrate this in the diagram.

 
 
 
 
5 6  f 5 6  g
 
 
 
2  2 


( ) -  ( ) -
1 2   1 2

Proof: Let C = B1 (1)\{1} = (0, 2)\{1} R. Then g(x) = f (x), x C and


lim g(x) = 5, so by the local principle of limits lim f (x) = 5.
x1 x1

2.4.4 Limits of Ordered Pairs of Functions.


The theorem about limits of ordered pairs of real valued functions will be used
to derive results about some vector-valued (i.e. values in R2 ) functions and to
prove some theorems about the algebra of limits.

We introduce the idea of an ordered pair of functions (f1 , f2 ).


Let f1 : V W1 and f2 : V W2 then f = (f1 , f2 ) : V W1 W2 = W is
2.4. THEOREMS ON LIMITS AND CONTINUITY 49

defined by
(f1 , f2 )(v) = (f1 (v), f2 (v)), v V .

For example we have (f1 , f2 ) : R R2 ; x 7 (f1 (x), f2 (x)).

6 6

f2
f2 (x)
(f1 (x), f2 (x))
(f1 , f2 )
x
-
f1 (x)
f1

For a particular example we could take f1 = Id : R R and f2 = Id2 : R R


so that (f1 , f2 ) : R R2 at x is (Id(x), Id2 (x)) = (x, x2 ). As x moves along the
real axis, (f1 (x), f2 (x)) moves along a parabloa in R2 . The mapping diagram is

6 6

x (Id, Id2 ) x2
| - (x, x2 )

-
x

In this case we call the pair (Id, Id2 ) a parametrization of the parabola. The
next result is the key to proving the limit of an ordered pair theorem.

Lemma 2.4.12 (Each ball contains a box.)

If B is an open ball in R2 , then open balls 1 , 2 R such that 1 2 B.

In addition if
B has centre (l1 , l2 ) and radius r > 0 then 1 , 2 can be chosen
with radius r/ 2 and centre l1 , l2 respectively.

The idea is indicated in the diagram.


50 CHAPTER 2. LIMITS AND CONTINUITY

'$

l2 (l1 , l2 )
r
&%


l1

Proof: Let B = Br (l1 , l2 ).

Choose 1 = Br/2 (l1 ) and 2 = Br/2 (l2 ). Then


1 2 = {(x, y) : x Br/2 (l1 ), y Br/2 (l2 )}

= {(x, y) : |x l1 | < r/ 2, |y l2 | < r/ 2}
{(x, y) : ||(x l1 , y l2 )|| < r} since
||(x l1 , y l2 )|| = ((x l1 )2 + (y l2 )2 )1/2 < r
= {(x, y) : ||(x, y) (l1 , l2 )|| < r}
= B.
Theorem 2.4.13 (Limit of an ordered pair).

If lim f1 (x) = l1 and lim f2 (x) = l2 then


xa xa
 
lim (f1 , f2 )(x) = (l1 , l2 ) = ( lim f1 (x), lim f2 (x)) .
xa xa xa

Proof: Suppose lim f1 (x) = l1 and lim f2 (x) = l2 . (To prove: B, A :


xa xa
(f1 , f2 )(A) B, where B is an open ball centred on (l1 , l2 ) and A is a decentred
ball of centre a.)

Let B = B (l1 , l2 ). By lemma 2.4 12, there exist open balls 1 = B/2 (l1 ),
2 = B/2 (l2 ) such that 1 2 B. So since lim f1 (x) = l1 there is an open
xa
ball A1 , (decentred), of centre a, such that f1 (A1 ) 1 and since lim f2 (x) = l2
xa
there is an open ball A2 , (decentred), of centre a, such that f2 (A2 ) 2 .

Now choose A = A1 A2 so A A1 and A A2 . Then


(f1 , f2 )(A) = {(f1 (x), f2 (x)) : x A}
= {(y1 , y2 ) : y1 = f1 (x), y2 = f2 (x) and x A}
{(y1 , y2 ) : y1 f1 (A), y2 f2 (A)}
= f1 (A) f2 (A)
f1 (A1 ) f2 (A2 ) A A1 and A A2
1 2 by choice of A1 , A2
B
thus lim (f1 , f2 )(x) = (l1 , l2 ) as required.
xa
2.5. SOME ALGEBRAIC THEOREMS 51

2.5 Some Algebraic Theorems


We want to derive some theorems about sums, differences, products and quo-
tients of limits, so that we can calculate limits of more complicated functions,
without having to resort to using the definition.

Furthermore these algebraic theorems about limits can be used to prove the
basic rules of differentiation used in elementary calculus.

To begin we define functions for each of the operations of addition, subtraction,


product and quotient.

Definition 2.5.1
2
sum : R2 R

sum(x, y) = x + y
diff : R R diff(x, y) = x y

Define by

prod : R2 R
prod(x, y) = xy
quot : R2 \(R {0}) R quot(x, y) = x/y.

Proposition 2.5.2 The mapping sum : R2 R is continuous.

Proof: Let (a, b) R2 .

Choose c = 2 and let (x, y) R2 .


Then
|sum(x, y) sum(a, b)| c||(x, y) (a, b)||
as in the exploration.
Exploration
So by the key example and the squeeze principle
We will use the squeeze
with (x, y) = c||(x, y) (a, b)||,
principle: to find c > 0:
we have
|sum(x, y) sum(a, b)|
lim sum(x, y) = sum(a, b)
(x,y)(a,b) c||(x, y) (a, b)||
i.e. sum is continuous at (a, b) and therefore If (x, y) R2 , LHS
sum is continuous on R2 . = |x + y a b|
|x a| + |y b|
by 4-inequality
2||(x, y) (a, b)||
by projection
inequalities.
So can choose c = 2.

Proposition 2.5.3 The mapping diff : R2 R is continuous.

We leave the proof of propostion 2.5.3 to the exercises.

The next result will also be proved using the squeeze principle, however this
time the choice of c is not quite as straightforward. We have set the exploration
52 CHAPTER 2. LIMITS AND CONTINUITY

aside in a box, remember it is not supposed to be part of the proof, it is there


to help you understand how and why the choice of c was made.

Proposition 2.5.4 The mapping prod : R2 R is continuous.

Proof: Let (a, b) R2 .

Exploration We want to find c > 0 :


|prod(x, y) prod(a, b)| c||(x, y) (a, b)|| for all (x, y) R2 . Let (x, y)
R2 . Then
|prod(x, y) prod(a, b)| = |xy ab|
= |(x a)(y b) + a(y b) + b(x a)|
|x a||y b| + |a||y b| + |b||x a|
by the 4-inequality
||(x, y) (a, b)||2 + |a|||(x, y) (a, b)||+
|b|||(x, y) (a, b)||
by the projection inequalities
= (||(x, y) (a, b)|| + |a| + |b|)||(x, y) (a, b)||

which is not really what we want, because (||(x, y) (a, b)|| + |a| + |b|) is not a
constant; indeed it gets arbitrarily large as (x, y) ranges over R2 . However we
can use the local principle for limits. We change to a function g which agrees
with prod on some ball C of centre (a, b) and reduces the undesirable factor to
zero for (x, y) 6 C. Now for (x, y) C we have ||(x, y) (a, b)|| radius (C)
so
(||(x, y) (a, b)|| + |a| + |b|) radius(C) + |a| + |b|.

So we could choose radius(C) = 1, and c = 1 + |a| + |b|. Returning to the proof.

Let C = B1 (a, b) and choose c = 1 + |a| + |b|. Now define a function g : R2 R


by

prod(x, y) if (x, y) C
g(x, y) =
prod(a, b) otherwise.

Let (x, y) R2 . Then



|prod(x, y) prod(a, b)| if (x, y) C
|g(x, y) g(a, b)| =
 |prod(a, b) prod(a, b)| = 0 otherwise
c||(x, y) (a, b)|| if (x, y) C

0 otherwise
c||(x, y) (a, b)||

in both cases.
2.5. SOME ALGEBRAIC THEOREMS 53

So by this inequality and the key example with (x, y) = c||(x, y) (a, b)|| we
have lim (x, y) = 0, so by the squeeze principle
(x,y)(a,b)

lim g(x, y) = g(a, b).


(x,y)(a,b)

Now by the local principle

lim g(x, y) = lim prod(x, y)


(x,y)(a,b) (x,y)(a,b)

so lim prod(x, y) = g(a, b) = prod(a, b), since (a, b) C. Hence prod is


(x,y)(a,b)
continuous at (a, b) and therefore prod is continuous on R2 .

Notice that the domain of quot is R2 \(R {0}) = {(x, y) R2 : y 6= 0}.

Proposition 2.5.5 The mapping quot : R2 \(R {0}) R is continuous.

Proof: Let (a, b) R2 \(R {0}); i.e. (a, b) R2 , b 6= 0.

(Exploration).

We want to find c > 0 : |quot(x, y) quot(a, b)| c||(x, y) (a, b)|| for all
(x, y) R2 , y 6= 0.

Let (x, y) R2 . Then



|quot(x, y) quot(a, b)| = |x/y a/b| = xbya

yb

= (xa)b(yb)a

yb
|(xa)b(yb)a|
= |y||b|
|xa||b|+|yb||a|
|y||b|
||(x,y)(a,b)||(|a|+|b|)
|y||b|.

1
The factor |y| gets arbitrarily large as (x, y) ranges over R2 \(R {0}), in partic-
ular as y gets close to 0. So we need to keep y away from 0. With C = Br (a, b)
and r |b| |b|
2 we have if (x, y) C, then |y| > 2 because C is centred on (a, b)
with radius not more than |b| 1 2
2 . So |y| |b| , thus

|b| + |a| 2(|a| + |b|)


<
|y||b| |b|2

and we can choose the RHS of the inequality for c. Returning to the proof.
54 CHAPTER 2. LIMITS AND CONTINUITY

|b| 2(|a|+|b|)
Choose C = Br (a, b) with r = 2 and choose c = |b|2 . Now define a new
2
function g : R \(R {0}) R by

quot(x, y) if (x, y) C
g(x, y) =
quot(a, b) otherwise.

Let (x, y) R2 , y 6= 0. Then



|quot(x, y) quot(a, b)| if (x, y) C
|g(x, y) g(a, b)| =
 |quot(a, b) quot(a, b)| = 0 otherwise
c||(x, y) (a, b)|| if (x, y) C

0 otherwise
c||(x, y) (a, b)||

in both cases (by the exploration).

So by this inequality and the key example with (x, y) = c||(x, y) (a, b)|| we
have lim (x, y) = 0, so by the squeeze principle
(x,y)(a,b)

lim g(x, y) = g(a, b).


(x,y)(a,b)

Now by the local principle

lim g(x, y) = lim quot(x, y)


(x,y)(a,b) (x,y)(a,b)

so lim quot(x, y) = g(a, b) = quot(a, b), since (a, b) C. Hence quot is


(x,y)(a,b)
continuous at (a, b) and therefore quot is continuous on R2 \R {0}).

2.5.1 Algebra of Limits


Theorem 2.5.6 If f1 and f2 have limits l1 and l2 respectively, at a, then
f1
f1 + f2 , f1 f2 , f1 f2 and f2 have the respective limits
l1
l1 + l2 , l1 l2 , l1 l2 and l2 provided l2 6= 0, at a.

The main idea in each case is to decompose the given function into a composite
of functions whose limits we have already established. For example

f1 + f2 = sum (f1 , f2 ),

i.e.
(f1 ,f2 ) sum
x 7 (f1 (x), f2 (x)) 7 f1 (x) + f2 (x) = (f1 + f2 )(x)
| {z }
sum(f1 ,f2 )
2.5. SOME ALGEBRAIC THEOREMS 55

Proof: Suppose lim f1 (x) = l1 and lim f2 (x) = l2 then


xa xa

lim (f1 , f2 )(x) = (l1 , l2 ) by limit of an ordered pair and


xa
lim (sum (f1 , f2 ))(x) = sum(l1 , l2 ) by limit of a composite and continuity
xa
of sum. Thus
lim (f1 + f2 )(x) = sum(l1 , l2 ) = l1 + l2
xa

as required.

The proofs of the other 3 algebraic expressions are equally easy and are left as
an exercise.
Example 2.5.7 We use the algebra of limits and the local principle to prove
that any polynomial is continuous on R.
Let n N and let f : R R be

f = n Idn + n1 Idn1 + + 1 Id + 0 , i R, n 6= 0; i.e.


f (x) = n xn + n1 xn1 + + 2 x2 + 1 x + 0 ,

for x R.So f is a real-valued polynomial of degree n. In one of the exercises


we have proved that Idn is continuous at a R, n N. So once we prove
that the constant function is continuous, the result follows from the algebra of
limits.

To prove: (c W ), f : V W with f = c f is continuous on V .

Proof: Let c W and let a V . Let B = B (c). Choose A = B (a). Then

f (A) = {f (x) : x A}
= {c} since f (x) = c, x V
B (c) = B

as required. Hence c is continuous at a and so is continuous on V .

Now f = n Idn + n1 Idn1 + + 1 Id + 0 so f is the sum of continuous


functions thus by the algebra of limits f is continuous.

A rational function is the quotient of two polynomial functions p and q say,


where for q : R R we have q(x) 6= 0 for any x R.
Example 2.5.8 Every rational function is continuous.
Proof: Let p : R R and q : R R be polynomial functions with q(x) 6= 0
for any x R; then p/q is a rational function with domain R\{x : q(x) = 0}
(which is always an open set). By example 2.5.7, p and q are continuous on V
and so by the algebra of limits p/q is continuous on V .
56 CHAPTER 2. LIMITS AND CONTINUITY

Example 2.5.9 We use limit of a composite to prove the theorem translation


of the origin in the domain; i.e. for f : V W, a V, l W

lim f (x) = l lim f (a + h) = l.


xa h0

Proof: Notice that f (a + h) = (f (a + Id))(h) and for x = a + h V we have

x a h 0.

Now a + Id is a polynomial function (of degree 1) and so by example 2.5.7 it is


continuous at h.

f (x) x 6= a
Suppose lim f (x) = l and put f(x) =
xa l x = a.

Then f is continuous at a, since by the local principle lim f(x) = lim f (x) so
xa xa
lim f(x) = l = f(a). Now by the theorem on composites we have
xa

lim (f (a + Id))(h) = lim f(a + h)


h0 h0
= f( lim )(a + Id)(h))
h0
= f(a) = l.

We leave the converse as an exercise.

Recall the inequalities for trigonometric functions:

sin(h) h tan(h), (h 0), h [0, /2). (2.4)

The inequalities are derivable from a diagram which compares the lengths of the
sides of some triangles with the length of the arc of the circle of radius 1. For
example, in the following, for angle h radians we have length of arc is h which
is greater than the length sin h of the opposite side of the triangle.

Example 2.5.10 With the aid of the squeeze principle and the local principle
we will prove lim sin(h) = 0.
h0

Proof: If h (/2, /2), then h (/2, 0] or [0, /2).


2.5. SOME ALGEBRAIC THEOREMS 57

If h [0, /2) then we have


| sin(h) 0| = sin(h) h = |h 0|.
And if h (/2, 0] then since sin(h) and h are both positive we have
| sin(h) 0| = sin(h) = sin(h) h = |h 0|.

sin(h) h (/2, /2)
Thus in both cases | sin(h) 0| |h|. Put f (h) =
0 h=0
then for every h we have |f (h) 0| |h 0| and lim |h| = 0 and so by the
h0
squeeze principle
lim f (h) = 0.
h0

But f = sin on (/2, /2), so by the local principle lim sin(h) = 0.


h0

Example 2.5.11 We prove that lim cos(h) = 1.


h0

Proof: Let h (/2, /2), then cos(h) 0. From cos2 (h) + sin2 (h) = 1 we
have cos2 (h) 1 = sin2 (h) so that

sin2 (h)
cos(h) 1 = .
cos(h) + 1
Thus
sin2 (h)
| cos(h) 1| =
cos(h)+1

2

sin 1(h) since cos(h) + 1 1

|h|| sin(h)| by example 2.5.10


|h|.

cos(h) 1 h (/2, /2)
Now put f (h) = then by the squeeze princi-
0 h=0
ple lim f (h) = 0 and since f = cos 1 on (/2, /2) by the local principle
h0
lim f (h) = lim (cos(h) 1) = lim cos(h) 1 and so by the algebra of limits we
h0 h0 h0
have lim cos(h) = 1.
h0

Example 2.5.12 sin is continuous at a R.


Proof: Let a R. We have
sin(a + h) = sin(a) cos(h) + cos(a) sin(h)
so
lim sin(a + h) = lim (sin(a) cos(h)) + lim (cos(a) sin(h))
h0 h0 h0
by the algebra of limits
= sin(a) lim cos(h) + cos(a) lim sin h
h0 h0
by the algebra of limits again
= sin(a) 1 + cos(a) 0 = sin(a)
58 CHAPTER 2. LIMITS AND CONTINUITY

by the previous examples. Now by the result on translation of the origin we


have
lim sin(x) = sin(a).
xa

Hence sin is continuous at a R.


Part II

Limits of Sequences

59
Chapter 3

Limits of Sequences

Recall the established notation for some subsets of R. We had

N the set of natural numbers i.e. {1, 2, 3 . . . }.


Z the set of integers i.e. {0, 1, 2, . . . }.
Q the set of rational numbers i.e. {p/q : p Z, q N}.
R the set of real numbers in addition

we had C for the set of complex numbers of which R is a subset.

Recall that a number is rational if its infinite decimal representation either


terminates i.e. finishes with an infinite string of zeroes, or has a repeating
subsequence for example 1/8,1/11 and 1/7 are rational numbers,

1/8 = 0.125000 . . . and so terminates, whilst


1/11 = 0.090909 . . . , 1/7 = 0.142857142857 . . . and so each have a
repeating subsequence.

In this part of the course we will examine the limit of a sequence. A sequence is
a map s : N V which assigns to each natural number n an element s(n) V ,
where V is a normed vector space.

In some settings we may wish to think of the elements s(1), s(2), . . . as the
sequence (rather than the map s) we should make it clear when this is what we
mean by the sequence.

61
62 CHAPTER 3. LIMITS OF SEQUENCES

3.1 Intervals in R, countability, bounded subsets


of R.
Proposition 3.1.1 Every non-degenerate interval I R contains a rational
number, where a degenerate interval is either the emptyset , or a singleton set
{x}, x R.

Proof: (Sketch) Let I R be a non-degenerate interval, with endpoints a < b,


a, b R.
Let the respective decimal representations of a, b be

a = a0 .a1 a2 a3 . . . and b = b0 .b1 b2 b3 . . .

where a0 , b0 Z and ai , bi {0, 1, 2, . . . , 9}, i N. We will consider only the


case a0 = b0 , since if a0 < b0 ( resp. b0 < a0 ) then a0 +b 2
0
I Q and so is a
rational number in the interval. Since a < b and a0 = b0 there is a least natural
number i such that ai < bi . We truncate b at bi , i.e. set q = b0 .b1 b1 . . . bi 000 . . .
then
a = a0 .a1 a2 . . . ai < a0 .a1 a2 . . . bi 000 . . .
= b0 .b1 b2 . . . bi 000 = q
b0 .b1 b2 . . . bi bi+1 . . .

and q Q because it is a terminating decimal.

We say a subset S R is dense in R non-degenerate interval I R, x S


such that x I. In other words the intersection of any non-degenerate interval
of real numbers with S contains some real number. Proposition 3.1.1 thus says
that Q is dense in R. On the other hand N is not dense in R. We have to
demonstrate that I R such that I N = . Let n N then the interval
(n, n + 1) contains no natural number, so satisfies the statement I N = .

Definition 3.1.2 A set S is countable if and only if either it is finite or it has


the same cardinality as N; i.e. a bijection f : S N (resp. N S).

Proposition 3.1.3 Q+ is countable.

The strategy of proof is to demonstrate a way to list all of the elements of Q+ .


Labelling the elements of the list then gives a bijection from the set of natural
numbers N; i.e let the first element of the list be s1 , the second element s2 , the
third element s3 , . . . and define a function f : N Q+ be f (i) = si . It is easy
then to check that f is a bijection.

Q+ = m

n : m, n N , so we could set out the elements in an array by putting
those elements with: n = 1 in the first column, n = 2 in the second column,
3.1. INTERVALS IN R, COUNTABILITY, BOUNDED SUBSETS OF R. 63

n = 3 in the third column . . . and so on.


..
. ...
5 5 5
1 2 3 ...
4 4 4 4
1 2 3 4 ...
3 3 3 3 3
1 2 3 4 5 ...
2 2 2 2 2 2
1 2 3 4 5 6 ...
1 1 1 1 1 1 1
1 2 3 4 5 6 7 ...

Of course this array repeats some elements of Q+ , since for example


1 2 3
= = = ...,
2 4 6
so we must take account of this repetition when we construct a list from the
array. To construct the list, start at 11 , go up to 12 and down the diagonal to 12 ,
across to 13 , up the diagonal jumping 22 since we already have 1 in the list, to
3 4 3 2 1 1
1 , up to 1 and down the diagonal through 2 , 3 to 4 , across to 5 and up the
diagonal jumping 4 , 3 and 2 (since 2 , 1, 2 are already in the list) to 51 and so
2 3 4 1

on.

The proof of the following result is left as an exercise.


Proposition 3.1.4 If A, B are countable sets then A B is countable.
So by a finite inductive argument one has that a finite union of countable sets
is countable. From which one can deduce that Q is countable. Of course count-
ability would not be an interesting notion if all sets were countable.
Proposition 3.1.5 The set of real numbers, R, is uncountable.
We use the strategy of Cantors first proof of this result and prove that the
subset [0, 1] of R is uncountable.

Proof: Suppose [0, 1] is countable and let


0.a11 a12 a13 a14 . . .
0.a21 a22 a23 a24 . . .
0.a31 a32 a33 a34 . . .
..
.
be a list of the real numbers in [0, 1] in their infinite decimal representation; i.e.
aij {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}, for all i, j N.

Now construct a number b = 0.b1 b2 b3 b4 . . . by putting



2 if ann 6= 2
bn =
1 otherwise
64 CHAPTER 3. LIMITS OF SEQUENCES

For example given the list


0.1234 . . .
0.0133 . . .
0.2423 . . .
0.00135 . . .
..
.
b = 0.2212 . . . . Now the motivation for this construction is that the number b
is clearly in [0, 1], since it is a string of ones and twos, but it cannot be in the
list since it differs from
0.a11 a12 a13 a14 . . . because b1 6= a11
0.a21 a22 a23 a24 . . . because b2 =6 a22
0.a31 a32 a33 a34 . . . because b3 = 6 a33
..
.
Thus [0, 1] is not countable and so R is not countable.

The previous result highlights one of the many properties of the real numbers,
which the rational numbers do not have. The rational numbers are not ade-
quate for analysis. The basic geometric difference is that there are enough real
numbers to label every point on the line. This is referred to as the line-filling
behaviour of R. There are various ways of expressing this line-filling behaviour
and some will occur in later sections.
Definition 3.1.6 (Upper/lower bound)

b R is an upper bound for S R s S, s b and


b R is a lower bound for S R s S, b s.
Thus S has an upper (resp. lower ) bound b R such that s S, s b
(resp. b s).
Proposition 3.1.7 (The Archimedian Property of R) The set N of natural
numbers is not bounded above in R; i.e. b R, s N such that s > b.
Proof: Let b R and first suppose b 1. Then choose s = bbc + 1. We have
s N and bbc + 1 b.
Next suppose b < 1, then choose s = 1 N and s > b. Hence we have b R,
s N such that s > b, as required.
Example 3.1.8 Any real number is an upper bound for the empty set .
Proof: Notice that (1) b R, x , x b is true
(2) b R such that x , x > b is false.

But (2) is clearly false since there is no x in the empty set. Hence (1) is true.
3.1. INTERVALS IN R, COUNTABILITY, BOUNDED SUBSETS OF R. 65

1
Example 3.1.9 Let S = {1 n : n N}. Then 1 is an upper bound of S.

We are to prove: s S, s 1.
Proof: Let s S, so s = 1 n1 for some n N and 1 1
n < 1, since 1
n > 0. So
s < 1, as required.

Example 3.1.10 Let S = { x1 : x R+ }. Then S has no upper bound.

We are to prove that b R, s S, such that s > b.


Proof: Let b R. Now if b 0, then we can choose any element of S, since
they are all positive; i.e. s S, s > b.
1
And if b > 0 then choose x = b+1 and s = x1 = b + 1. Then s S and s > b.

Definition 3.1.11 (Least Upper/Greatest Lower bound)


l is the least upper bound (lub) or supremum (sup) of S

(i) l is an upper bound; i.e. x S, x l and


(ii) there is no upper bound less than l; i.e. > 0, s S such that s > l .
s

 -
l l

g is the greatest lower bound (glb) or infimum (inf ) of S

(i) g is a lower bound; i.e. x S, g x and


(ii) there is no lower bound greater than g; i.e. > 0, s S such that s < g + .
s

 -
g g+

Example 3.1.12 Let S = {2+ n1 : n N}, by inspection we guess that glb(S) =


2.

We are to prove that

(i) 2 is a lower bound; i.e. x S, 2 x and i.e. > 0, s S such that


(ii) there is no lower bound greater than 2;
s < 2 + .

1 1
Proof: (i) Let s S, so s = 2 + n for some n N and n > 0 for every n N,
so 2 + n1 > 2. Hence s 2.
66 CHAPTER 3. LIMITS OF SEQUENCES

(ii) Let > 0. Choose n = b 1 c + 1 Find s S:


then n N, since s < 2 + , i.e.
n 1 and 2 + n1 < 2 +
n > 1 i.e. 1
<
n
1
i.e. n > .

1 1
i.e. > n, so s = 2 + n < 2 + as required.

In addition, by inspection we guess that lub(S) = 3.


We are to prove that
(i) 3 is an upper bound; i.e. x S, x 3 and
(ii) there is no upper bound less than 3;
i.e. > 0, s S such that s > 3 .

Proof: (i) Let s S, so s = 2 + n1 for some n N and n1 1 for every n N,


so s 2 + 1 = 3. Hence s 3.
(ii) Let > 0. Choose s = 3 > 3 Find s S:
s > 3 , i.e.
s = 2 + n1 > 3
so can take s = 3.
Thus > 0, s S, such that s > s .
Example 3.1.13 Let S = (1/2, 0) = {x R : 1/2 < x < 0}, by inspection
we guess that lub(S) = 0.
We are to prove that
(i) 0 is an upper bound; i.e. x S, x 0 and
(ii) there is no upper bound less than 0;
i.e. > 0, s S such that s > 0 .

Proof: (i) Let s S, so 1/2 < s < 0, hence s 0.


(ii) Let > 0. If 1/2 then Find s S:
choose s = 1/4 then s > 0 = .
< 1/2 < 1/4 = s. Notice if 1/2, then
And if 0 < < 1/2 then we can take any s S and
choose s = /2 > . if 0 < < 1/2, then can take
s = /2 > .
The determination of the two cases depends on the length of the interval and is
to ensure that /2 S.

A related idea is that of the maximum and the minimum of a set. The maximum
max S is defined to be a point M S, such that x S, x M . And the
minimum min S is defined to be a point m S, such that x S, m x. Thus
a set may not have a maximum or a minimum. For example in example 3.1.12
max S = 3, whilst in example 3.1.13, S has no maximum.
3.2. SOME USEFUL INEQUALITIES 67

Proposition 3.1.14 (The least upper bound (resp. greatest lower bound)
property of R.) For every non-empty set S of real numbers which is bounded
above (resp. below), l R (resp. g R), such that l = sup S (resp. g = inf S).

Many of the analysis text books contain a proof, for example Rudin, Principles
of Mathematical Analysis.

3.2 Some useful inequalities


1 1
1. n N, 2n < n.
n 2
2. n N, 2n < n.
1 1
3. n N, n > 3, n! < 2n .

4. n N, ln(n) < n.

5. n N, n 3, 1 < ln(n).

To prove the first two inequalities we can use the binomial theorem, for example
in the case of 2:
Let n N, then
1 1
(1+1)n = n(n1) n(n1)(n2)
1+n+ 2 + 3! +...
1
< n(n1)
1+n+ 2
1 1 2
= n n2
< n2
= n2
1+ 2 + 2 2

n 2
consequently 2n < n as required.

Inequality 3 can be proved by induction, the first case being n = 4. And one
can use calculus for the inequalities in 4 and 5, for example to prove 4: let n N
then,
ln(n) < n n = eln(n) < en
so it suffices to prove n < en , for every n N. Now for f (n) = n we have
f 0 (n) = 1 and for g(n) = en we have g 0 (n) = en , thus f 0 (n) < g 0 (n) for every
n, in addition f 0 , g 0 0. Thus f, g are both increasing with g increasing faster
than f . Hence n = f (n) < g(n) = en for every n.

3.3 Definition-limit of a sequence

A sequence is a map s : N V , where V is some set (for our purposes equipped


with the structure of a normed vector space). Such a function assigns to each
natural number n an element s(n) V . In some instances we may think of the
sequence as the list of elements. s(1), s(2), . . . . This is really the image of N
68 CHAPTER 3. LIMITS OF SEQUENCES

under s; i.e. s(N).


The pertinant question is; what happens to s(n) as n gets large?
We consider sequences where V = R. Such sequences have a graphical repre-
sentation in R2 .

1
Example 3.3.1 The sequence s : N R with s(n) = 2 n has the following
graph.

2


1

| | | | | -
1 2 3 4 5
From the graph it seems that
lim s(n) = 2.
n

Example 3.3.2 The sequence s : N R with s(n) = 2 + (1)n has the


following graph.

3 6

| | | | | -
1 2 3 4 5
The graph oscillates between 1
and 3, so it seems not to converge to a limit. We have that lim s(n) does not
n
exist.

Example 3.3.3 The sequence s : N R with s(n) = n has the following


graph.
3.3. DEFINITION-LIMIT OF A SEQUENCE 69
3 6

| | | -
1 2 3
It increases in proportion with n
and so again lim s(n) does not exist.
n

We want to say what it means for a sequence to have a limit. Graphically we


have that if s has a limit l then after some natural number , i.e. for all n > ,
all of the values of s(n) lie between l and l + , as illustrated below.


l + 6


l
s()

l

| | | | -
1 2 ...

Definition 3.3.4 Let (V, || ||) be a normed vector space. The sequence s :
N V has a limit l V

> 0, N such that n N, n > ,


we have ||s(n) l|| < ; i.e. s(n) B (l).

We write lim s(n) = l.


n

Using the definition

Example 3.3.5 To prove: for s : N R defined by

3
s(n) = 2 we have lim s(n) = 2.
n n
70 CHAPTER 3. LIMITS OF SEQUENCES

Proof: Let > 0. Choose = b 3 c + 1 then Find N:


N and > 3 so |s(n) 2| < , i.e.
let n N, n > then |2 n3 2| = | n3 | =
> n3 = |s(n) 2|. 3
n < ,
so we need n > 3 .
Hence > 0, N such that n N, n > we have |s(n) 2| < .
Example 3.3.6 To prove: for s : N R defined by
s(n) = 1 + 2n we have lim s(n) = 1.
n

Proof: Let > 0. Choose = b 1 c + 1 then Find N:


N and > 1 so |s(n) 1| < , i.e.
let n N, n > then |1 + 2n 1| = |2n | =
1
> n1 > 21n = |s(n) 1|. 2n < .
Now 21n < n1 so
if suffices to have n > 1 .
Hence > 0, N such that n N, n > we have |s(n) 1| < . Notice
in this last example from 21n < , we have 1 < 2n so log2 1 <  n which sug-
gests we could choose > log2 1 . The problem is that log2 1 may not be a
1 1 1

positive real number; for example if = 4, then = 4 and log 2 = 2, so in
1

general if we took = blog2 c+?, we wont know what to add to ensure that
N. The inequalities play a role here.

3.3.1 Limits of functions and limits of sequences-A com-


parison
We have seen two types of limits, the limit of a function at a point and the limit
of a sequence. There are some obvious differences but also some useful analogies
between these two notions of limit.

An important difference between the two types of limit, is that in our definition
of
lim f (x) = l
xa
we could use a decentred ball A, of centre a, in the domain, to measure the
closeness of x to a. In our definition
lim s(n) = l
n

we could not use a ball to measure how large the integer n is and so we could
not use the nice geometric approach to limits as we used in the case of limits of
functions. We can recast the definition in a more geometric form.
Recall our definition of lim f (x) = l; it said
xa
(B)(A) : f (A) B
3.3. DEFINITION-LIMIT OF A SEQUENCE 71

where A is a decentred ball centred on a and B is a ball centred on l.

On the other hand our definition of lim s(n) = l said


n

( > 0)( N) : (n N), n > ||s(n) l|| < .


We aim to recast the definition so that we can have a mapping diagram showing
how s maps the natural numbers. Now the phrase ||s(n) l|| < can be
expressed in terms of balls. It says s(n) is in the (open) ball of centre l and
radius > 0 so we could write
(B)( N) : (n N)n > s(n) B.
Examine the part of the definition which involves n. We see that the condition
has to hold for all integers n such that n > ; that is, for all integers in the set
{ + 1, + 2, + 3, . . . }. We can represent this as a subset of the real line

..
.

+ 3
+ 2
+ 1

The rest of the definition says that s maps this set into the ball B as shown
below


6 6 B

.. #
.
s
- l
+ 3 "!
+ 2
+ 1 -

Thus if we define the symbol
:= {all integers > } we have

Def (3.3.1)
lim s(n) = l
n
72 CHAPTER 3. LIMITS OF SEQUENCES

means that
(B)(
) : s( ) B,

where
is the set of all integers greater than and B is an open ball of centre
l.

So the set
replaces the decentred ball A of centre a.
With this view in mind we can think of the theorems: Uniqueness of limit, limit
of a composite theorem, the squeeze principle and the local principle for limits
as geometric theorems about limits.

3.4 Some theorems on limits of sequences


Theorem 3.4.1 (Uniqueness of limits) If a sequence s : N V has two limits
l and l0 then l = l0 .

Proof: Suppose s : N V has limits l, l0 V. Thus > 0 we have

1 N : n N, n > 1 we have ||s(n) l|| < /2 (i)


2 N : n N, n > 2 we have ||s(n) l0 || < /2 (ii).

Now let > 0 and suppose 1 , 2 satisfy (i) and (ii). Consider = max{1 , 2 }.
Let n N, n > , then n > 1 and n > 2 . So

||l l0 || ||s(n) l|| + ||s(n) l0 || (by the 4-inequality)


< /2 + /2 = .

This says lim l(n) = l0 , where l is a constant sequence; i.e. l = lim l(n) = l0
n n
as required.

Theorem 3.4.2 (Limit of a composite) If lim s(n) = l and if g is continuous


n
at l then
lim (g s)(n) = g( lim s(n)) i.e.
n n

lim g(s(n)) = g( lim s(n)) = g(l) (interchange of lim with a continuous func-
n n n
tion.)

Proof: Suppose lim s(n) = l and g is continuous at l. [We are to prove


n
lim (g s)(n) = g(l), i.e. > 0, N such that n N, n > we have
n
||(g s)(n) g(l)|| < .]

Let > 0.
3.4. SOME THEOREMS ON LIMITS OF SEQUENCES 73

Choose N such that n N Find N:


n > , s(n) B (l) g(s(n)) B (g(l)) ||(g s)(n) g(1)|| < , i.e.
( exists because g is continuous at l (g s)(n) B (g(l));
and exists because lim s(n) = l.) which is true if
n
Let n N, n > then ||s(n) l|| < ||s(n) l|| <
and so ||g(s(n)) g(l)|| < where g(B (l)) B (g(l))
as required. by continuity of g at l.

Theorem 3.4.3 (Squeeze principle) Let s : N V . If there exists a real-valued


sequence : N R such that
(i) ||s(n) l|| (n), n N and
(ii) lim (n) = 0
n

then lim s(n) = l.


n

Theorem 3.4.4 (Local principle for limits) Suppose there exists N such
that s(n) = t(n) for all n > . If lim s(n) = l (resp. lim t(n) = l) then
n n
lim t(n) = l (resp. lim s(n) = l.)
n n

The local principle for limits says that by changing a finite number of elements
of the sequence you do not change the existence of or the value of the limit
of the sequence; in other words the local principle for limits of sequences says
that if two sequences are eventually equal then the two sequences must have the
same limit, if it exists.

Proofs of the last two theorems will be on the exercises. Using the squeeze prin-
ciple will involve finding an inequality satisfying the hypothesis. We illustrate
the technique with some examples.
Example 3.4.5 Let s : N R be defined by
1
s(n) = 2 + sin(n2 ).
n
To prove: lim s(n) = 2 using the squeeze principle.
n

[We are to find a function


satisfying

(i) n N, |s(n) 2| = n1 sin(n2 ) (n) and
(ii) lim (n) = 0. ]
n

Notice that
|s(n) 2| = n1 sin(n2 )
= n1 | sin(n2 )|
n1
1 1
and so since lim = 0, we can take (n) = n.
n n
74 CHAPTER 3. LIMITS OF SEQUENCES

Proof: Let (n) = n1 , then by the above we have |s(n) 2| (n) and
lim (n) = 0; hence satisfies the hypothesis of the squeeze principle and so
n
the conclusion is that lim s(n) = 2 as required.
n

4
Example 3.4.6 Let s : N R be defined by s(n) = n! .
To prove: lim s(n) = 0, using the squeeze principle.
n

[We are to find a function satisfying (i) n N, |s(n) 0| = |s(n)| (n)


and (ii) lim (n) = 0. ]
n

Notice that 4 4 4
|s(n)| = n! =
n! n
4 4
for every n N and lim = 0 and so we can take (n) = n.
n n

Proof: Let (n) = n4 , then by the above we have |s(n) 0| (n) and
lim (n) = 0; hence satisfies the hypothesis of the squeeze principle and so
n
the conclusion is that lim s(n) = 0 as required.
n

1
Example 3.4.7 Let s : N R be defined by s(n) = 2n .
To prove: lim s(n) = 0, using the squeeze principle.
n

[We are to find a function satisfying (i) n N, |s(n) 0| = |s(n)| (n)


and (ii) lim (n) = 0. ]
n

Notice that
|s(n)| = 21n = 1
2n 1
n
1 1
for every n N and lim = 0. and so we can take (n) = n.
n n

Proof: Let (n) = n1 , then by the above we have |s(n) 0| (n) and
lim (n) = 0; hence satisfies the hypothesis of the squeeze principle and so
n
the conclusion is that lim s(n) = 0 as required.
n

Proposition 3.4.8 a R, |a| < 1 lim an = 0.


n

Proof: (We will use the squeeze principle.)


Let a R with |a| < 1. Then |a| = 1b for some b > 1. Thus b = 1 + d for some
d R, d > 0. Now

bn = (1 + d)n  
n
= 1 + nd + d2 + . . .
2
dn
3.4. SOME THEOREMS ON LIMITS OF SEQUENCES 75

i.e.
1 1 c
|a|n = = , where c = 1
d > 0.
bn dn n
Now choose (n) = nc , c > 0 then

|s(n)| = |a|n (n), for all n N and lim (n) = 0.


n

So by the squeeze principle lim an = 0 as required.


n

Next we give a theorem on ordered pairs of sequences, from which we can derive
an algebra of limits of sequences as we did for limits of functions.

Theorem 3.4.9 (Limit of an ordered pair) Let s : N V , t : N W be


two sequences and suppose lim s(n) = l V and lim t(n) = m W then
n n
lim (s(n), t(n)) = (l, m) V W ; i.e. lim (s(n), t(n)) = ( lim s(n), lim t(n)).
n n n n

When V = R = W the proof is very similiar to the proof concerning ordered


pairs of limits of functions and uses the every ball contains a box lemma. More
generally one has to say something about an appropriate norm on V W .

Indeed theorem 3.4.9 can be generalized to the limit of an ordered k-tuple. If


s1 , s2 , . . . , sk : N R are k real-valued sequences and for each i : 1 i k
lim si (n) = li then
n

lim (s1 (n), s2 (n), . . . , sk (n)) = (l1 , l2 , . . . , lk ).


n

Theorem 3.4.10 (Algebra of limits) If each of the real-valued sequences


s : N R and t : N R have limits then

(i) lim (s + t)(n) = lim (s(n) + t(n)) = lim s(n) + lim t(n).
n n n n

(ii) lim (s t)(n) = lim (s(n) t(n)) = lim s(n) lim t(n).
n n n n

(iii) lim (s.t)(n) = lim s(n).t(n) = lim s(n) lim t(n).


n n n n

(iv) lim (s/t)(n) = lim s(n)/t(n) = lim s(n)/ lim t(n), provided n N,
n n n n
t(n) 6= 0 and lim t(n) 6= 0.
n

Proof: Again the idea is to decompose the given sequence into a composite of
one of the continuus functions sum, diff, prod and quot and the given sequence.
We give the proof of (i) and leave proof of the other parts to the exercises.

(sum (s, t))(n) = sum((s, t)(n)) by definition of composite


= sum((s(n), t(n)) = s(n) + t(n) = (s + t)(n),
by definition of sum.
76 CHAPTER 3. LIMITS OF SEQUENCES

Now suppose lim s(n) = l and lim t(n) = m, then by theorem 3.4.9 (limit of
n n
an ordered pair) we have lim (s(n), t(n)) = (l, m) and by theorem 3.4.2, limit
n
of a composite we have

lim (sum (s, t))(n) = sum( lim (s(n), t(n))


n n
= sum(l, m) = l + m

as required.

We can use the results in theorem 3.4.10 along with some basic examples, which
we have already proved, to deduce the limits of other more complicated se-
quences. This is easier than trying to establish the limit of a sequence using the
definition, for which we need to be able to guess the limit.

Basic examples

1. Constant sequences: lim c(n) = c.


n

+ 1
2. Powers: p R , lim p = 0.
n n

3. Exponentials: a R, |a| < 1, lim an = 0.


n

To complete this section we give three extra theorems on limits of sequences


which can be useful either in calculating other limits or in establishing the
existence of a limit.

Theorem 3.4.11 (Shifting sequences) Given a sequence s : N V and a


number p N we can define a new sequence t : N V by t(n) = s(n+p). If one
of the sequences has a limit then the other has a limit and lim s(n) = lim t(n).
n n

A sequence is said to be bounded if there is an open ball which contains all of


its elements.

Theorem 3.4.12 Every sequence which has a limit is bounded.

The contrapositive of theorem 3.4.12 is of interest since it says that if a sequence


is not bounded it does not have a limit. Recall the example 3.3.3.

Theorem 3.4.13 (Preserving inequalities) If s : N R and t : N R are two


real-valed sequences which converge to limits and if s(n) t(n) for every n N,
then
lim s(n) lim t(n).
n n

2 1
Theorem 3.4.13 does not extend to strict inequality, since for example n > n
for every n N but
2 1
0 = lim 6> lim = 0.
n n n n
3.4. SOME THEOREMS ON LIMITS OF SEQUENCES 77

3.4.1 The real number system- filling the line.

All of the numbers which can be represented in a modern electronic computer


are rational numbers. But as the Greeks realised, it is not possible to solve
even the simplest nonlinear equation within the rational numbers. For example,
there is no rational solution to the equation x2 = 2. A consequence of this is
that not all points along a line can be labelled, if only rational numbers are
permitted. The real numbers overcome these difficulties.

Rational Numbers
Recall that a rational number has the form p/q where p and q are integers, q 6= 0.
Rational numbers may be used as labels for certain points on a line, once a unit
of length has been chosen.

The points which we can label with rational numbers cover the line densely in
the sense that every interval, no matter how small, contains a rational number.
To see this, divide the interval between two consecutive integers into 10 equal

| | | | | | | | | | |
0 1
parts:

In the example illustrated above, the endpoints of the resulting intervals rep-
resent the rational numbers 0, 0.1, 0.2, . . . , 0.9, 1.0. Clearly we can repeat the
process and subdivide each of the smaller intervals into 10 equal parts;

| | | | | ||||||||||| | | | |
0 0.5 0.6 1

In the example the endpoints of the resulting intervals represent the rational
numbers
0.5, 0.51, 0.52, . . . , 0.58, 0.59, 0.6. By proceeding in this way we can cover the line
as densely as we please with points labelled with rational numbers.

Contrast with the Integers


The behaviour on the line, of the rationals, is quite different to the behaviour
of the integers. The integers are not dense on the line, there are gaps between
them.

One of the properties of the integers which the rationals do not have is that
there is always a next integer. Thus the integer after 3 is 4, the next after 1008
78 CHAPTER 3. LIMITS OF SEQUENCES

is 1009, and so on. For the rationals the concept of a next rational number is
quite meaningless because between any two distinct rational numbers, there is
another rational number.

Even though the rationals cover the line densely, there are still points on the
line which cannot be labelled by a rational number.

The Greeks constructed points on the line which cannot be labelled with rational
numbers. One such point is shown in the following construction:

2 |x|

-

0 |x| 2

From the theorem of Pythagoras, the number x must satisfy the equation x2 +
x2 = 22 i.e. x2 = 2. The Greeks proved that there is no rational number whose
square is 2.
Proof: Suppose there are integers p and q 6= 0 such that (p/q)2 = 2. We can
assume that p and q have no common factor other than 1 (otherwise we could
just divide through by the common factor and the resultant numerator and
denominator would have no common factor other than 1).

It follows that p2 = 2q 2 , hence p2 is even. But if p2 is even, then p is even, so


there is an integer k such that p = 2k. Thus 4k 2 = 2q 2 so q 2 = 2k 2 which says
that q 2 is even and so q is even. But this contradicts our choice of p and q,
since p and q were to have no common factor other than 1. Thus the original
assumption is false.

Real Numbers

The real numbers can be regarded as decimals. The decimals may terminate,
they may consist of blocks which repeat themselves indefinitely, or they may
continue indefinitely without repetition of any block for example:

0.25 (terminating)
0.36791197197197 . . . (repeating)
0.10100100010000100000 . . . (neither)

Decimals which terminate or repeat represent the rational numbers, those which
do neither represent the irrational numbers.
3.4. SOME THEOREMS ON LIMITS OF SEQUENCES 79

Since every decimal representation represents either a rational or an irrational


number
R = {irrationals} Q. In addition the reals and the rationals are each a field,
but the irrationals are not a field.

The set of rational numbers is clearly infinite. The set of irrational numbers is
also infinite. Like the rationals the irrationals are also dense on the line and
there is no next irrational number.

Recall the definition of a countable set and that the rationals are countable,
whilst the reals are not countable and so the set of irrational numbers is also
uncountable.

Once we have the extra numbers (i.e. the irrationals) we can solve the equation
x2 = 2 by a process of successive approximation. By trial and error we find that

1.42 < 2 < 1.52


and so 1.4 is too small 1.5 is too big
1.412 < 2 < 1.422
and so 1.41 is too small 1.42 is too big
1.4142 < 2 < 1.4152
and so 1.414 is too small 1.415 is too big
..
.

the process continues indefinitely


to produce the non terminating, non repeating
infinite decimal which is 2.

The line-filling property

The big advantage of the reals over the rationals is that they provide enough
numbers to label all of the points on the line. Although we shall not prove this
here, we have already illustrated the
ideas involved in the proof by showing how
to construct the infinite decimal for 2.

To prove the claim that our approximation process produces a solution to


x2 = 2.
Proof: The process of successive approximations produced two sequences of
numbers. Put

s1 = 1.4, s2 = 1.41, s3 = 1.414, . . . and


t1 = 1.5, t2 = 1.42, t3 = 1.415, . . .

If l is the infinite decimal produced by the successive approximations, then

|sn l| < 10n and |tn l| < 10n ,


80 CHAPTER 3. LIMITS OF SEQUENCES

since they differ only in the nth place.


But since 10n > n it follows that

|sn l| < 1/n and |tn l| < 1/n.

From these inequalities and the squeeze principle, we deduce that

lim sn = l and lim tn = l.


n n

But limits respect products, by the algebra of limits and so

lim s2n = lim sn lim sn = l2 and lim t2n = lim tn lim tn = l2 .


n n n n n n

Finally by going back to the construction process we see that, for all n N
s2n < 2 < t2n thus lim s2n 2 lim t2n , by preserving inequalities. This means
n n
l2 2 l2 , so l2 2 and l2 2 and so l2 = 2, thus l is the solution to the
original equation.

3.5 Existence theorems


In sections 3.3 and 3.4 we have seen that one may be given a sequence and asked
to find the limit using (a) the definition of limit of a sequence or (b) using the
given theorems.

Often sequences are too obscure to guess a limit and so proving, using the
definition, that some real number is the limit of the sequence, is tedious. Math-
ematicians have developed existence theorems which enable one to prove that
the limit of a sequence exists, even if you cannot say what the limit is. These
theorems depend on the line-filling behaviour which we referred to in section
3.1 and 3.4.1. The first existence theorem will apply to bounded monotone
sequences.
Definition 3.5.1 A sequence s : N R is said to be increasing if n N,
s(n) s(n + 1) and s is said to be decreasing if n N, s(n + 1) s(n). When
we have strict inequality, we will say the sequence is strictly increasing (resp.
strictly decreasing).

If a sequence is either increasing or decreasing we say it is monotone.


The sequences relevant to the first theorem are either those which are increas-
ing and bounded above (hence bounded, since s(1) is a lower bound) or those
which are decreasing and bounded below (hence bounded, since s(1) is an upper
bound).
Theorem 3.5.2 (BIST) If s : N R is a real-valued sequence which is increas-
ing and bounded above, then l R such that lim s(n) = l; i.e. the sequence
n
converges to l.
3.5. EXISTENCE THEOREMS 81

The acronym BIST stands for bounded increasing sequence theorem. Clearly
one can have a BDST, bounded decreasing sequence theorem by replacing in-
creasing with decreasing and bounded above with bounded below.

In fact we can replace R in the theorem with any normed vector space which
has an order relation, once we understand that a subset S of V is bounded if
and only if r > 0, a V , such that S Br (a).

Proof: (of 3.5.2)


Let s : N R be a sequence which is bounded above and increasing. Let
S = s(N) = {s(1), s(2), . . . } =
6 .
S is a non-empty set of real numbers which by hypothesis is bounded above, so
by the least upper bound theorem 3.1.14, sup S exists.
Let l = sup S. [We are to prove that l = lim s(n) and so establish the existence
n
of a limit of the sequence.]

Let > 0 and choose N such that l < s() (such a exists by the
definition of least upper bound).
Let n N, n > then
l < s() s(n) l since l is an upper bound and s is increasing thus

l < s(n) < l + i.e. |s(n) l| <

as required.

The BIST and BDST provide a formal way of expressing the line-filling be-
haviour of R. We work an example using the theorem.

Example 3.5.3 We consider the sequence


r q

q
2, 2 2, 2 2 2, . . .

and decide on its convergence.

We have
r q
p p
q p
2 2 2 2 2 2 2 2 2 2 ...
1/2 1+1/2 1/2 1+3/4 1/2 1+7/8 1/2
2 (2 ) (2 ) (2 ) ...
1/2 3/4 7/8 15/16
2 2 2 2 ...
s(1) s(2) s(3) s(4) ...

so
2n 1 1
s(n) = 2 2n = 2(1 2n ) .
82 CHAPTER 3. LIMITS OF SEQUENCES

Now the sequence is increasing, since for any n N,

2n 1 1 1 2n+1 1
= 1 < 1 =
2n 2n 2n+1 2n+1
and since 2 > 1 we have
2n 1 2n+1 1
2 2n <2 2n+1 .
And the sequence is bounded above, since n N, we have 1 21n < 1 so
1
s(n) = 2(1 2n ) < 2. Hence by the bounded increasing sequence theorem BIST,
we know that the sequence converges in R.
Furthermore in this case we can prove that the limit lim s(n) = 2. We will use
n
the theorems of section 3.3.

Consider f : R R with f (x) = 2x which is continuous on R (for the moment


we appeal to calculus, we have f differentiable and so continuous.)
Let t : N R be the sequence t(n) = 1 21n . Then lim t(n) = 1, since for
n
any > 0, we can choose = b 1 c + 1, then for n > , we have n > 1 so
> n1 > 21n = |t(n) 1|.
1
Now s(n) = 21 2n = 2t(n) = (f t)(n). So by limit of a composite theorem
3.4.3,
lim s(n) = lim (f t)(n) = f ( lim t(n)) = f (1) = 2.
n n n

The next theorem provides another precise version of the line-filling behaviour
of R. It follows easily from the BIST.

First a sequence [a1 , b1 ], [a2 , b2 ], [a3 , b3 ], . . . is said to be nested if each interval


in the sequence contains the next interval i.e.

[a1 , b1 ] [a2 , b2 ] [a3 , b3 ] . . .

i.e. on the line we have

| | | | | |-
a1 a2 a3 . . . b3 b2 b1

For a nested sequence of intervals, the sequence (an )nN of left endpoints is
increasing whilst the sequence (bn )nN of right endpoints is decreasing.

In addition notice that each element of (bn )nN is an upper bound for the
sequence (an )nN , thus (an )nN is increasing and bounded above whilst each
element of (an )nN is a lower bound for the sequence (bn )nN , so (bn )nN is
3.5. EXISTENCE THEOREMS 83

decreasing and bounded below, thus by the BIST, l = lim an and since for each
n
fixed m N we have an bm , for all n N we have lim an lim bm = bm ,
n n
thus l bm .
In addition, n N, an l (since an increasing sequence approaches its limit
from below) so for each fixed m N we have
n N, an l bm
consequently
m N, n N, an l bm .
This means that l is common to all of the intervals and indeed
Theorem 3.5.4 (Nested intervals theorem) for each sequence of closed nested
intervals
l R such that l is in every interval of the sequence.
Theorem 3.5.4 is false if you omit closed. (Think about an example of a
sequence of nested intervals (not necessarily closed) which do not have a
point in common.)
Nested intervals arise in the practical problem of finding solutions of an
equation. Suppose we know that the values of a function f , at each end-
point of an interval, have opposite signs. If f is continuous throughout
the interval, then the graph makes it seem likely that there is a solution
to the equation f (x) = 0 in the interval. The solution is the point l
of the theorem. Compare
also to the technique used to find successive
approximations to 2 in 3.4.1.

3.5.1 Subsequences
Definition 3.5.5 Suppose s : N V is a vector-valued sequence and suppose
j : N N is a sequence in N which is strictly increasing so j(1) = n1 , j(2) =
n2 , . . . and m N, nm < nm+1 . Then s j : N V is a subsequence of s.
Example 3.5.6 Let s : N R be the sequence defined by
 1
s(n) = 2 (n 1) if n is odd
n2 if n is even

so s(N) = Z, since s({odd n}) maps onto 0 Z+ and s({even n}) maps onto
Z .
1. Consider j : N N with j(m) = 2m 1. Then j is strictly increasing since
for any m N, j(m) = 2m 1 < 2m + 1 = 2(m + 1) 1 = j(m + 1).
The subsequence s j : N R sends
j s 1
m 7 2m 1 7 (2m 1 1) = m 1
2
i.e. (s j)(m) = m 1 and so (s j)(N) = {0, 1, 2, . . . }.
84 CHAPTER 3. LIMITS OF SEQUENCES

2. Consider j : N N with j(m) = 3m. Then j is strictly increasing since for


any m N, j(m) = 3m < 3m + 3 = j(m + 1).
(s j)(1) = s(j(1)) = s(3) = 21 (3 1) = 1
(s j)(2) = s(j(2)) = s(6) = 3
(s j)(3) = s(j(3)) = s(9) = 4
we find (
1
2 (3m 1) if m is odd
(s j)(m) =
3m
2 if m is even.
We have (s j)(N) = {1, 3, 4, 6, 7, . . . }.
Theorem 3.5.7 Suppose lim s(n) = l. Let s j be any subsequence of s, then
n
lim (s j)(n) = l.
n

i.e. if the limit of a sequence exists then the limit of any subsequence exists and
must be equal to the limit of the sequence.

It is the contrapositive of theorem 3.5.7 which is useful since it gives a criterion


for the non-existence of a limit. If we can find subsequences s j1 , s j2 and
lim (s j1 )(n) 6= lim (s j2 )(n) then lim s(n) does not exist.
n n n

Proof: (of 3.5.7)


Suppose lim s(n) = l and let s j be a subsequence of s. [We are to prove that
n
lim (s j)(n) = l.]
n
Let > 0. Choose N satisfying n > , |s(n) l| < . (*)(
exists by the
assumption that lim s(n) = l.) Now choose = j( ) .
n
Now let n N, n > then j(n) > j(
), since j is increasing, so
|s(j(n)) l| < , by (*).
Notice that the fact that j : N N which is increasing and such that
lim (s j)(n) = l, does not imply that lim s(n) exists; i.e. the converse of
n n
theorem 3.5.7 is false. Consider for example sequence s : N R given by

1 if n is odd
s(n) = n
2 + 1 if n is even
Then for the increasing sequence j : N N given by j(m) = 2m 1, we have
(s j)(n) = 1 for all n N, so lim (s j)(n) = 1, but lim s(n) does not exist,
n n
since the sequence is not bounded above.
Example 3.5.8 Consider the sequence for which s(N) = {1, 1, 1, 1, . . . }.
We have for j1 : N N with j1 (m) = 2m, (s j1 )(m) = 1, for every m
and for j2 : N N with j2 (m) = 2m 1, (s j2 )(m) = 1, for every m. So
lim (s j1 )(n) = 1 6= 1 = lim (s j2 )(n). Thus by the contrapositive of
n n
theorem 3.5.7, we have lim s(n) does not exist.
n
3.5. EXISTENCE THEOREMS 85

Lemma 3.5.9 Any sequence s : N R contains a subsequence which is either


increasing or it is decreasing.

Proof: Call a natural number n a peak point of the sequence if m > n,


s(m) < s(n). For example:


| | | | |
| | | | | | -
2 3 4 6

The peak points are marked by .

Consider two cases (i) infinitely many peak points and (ii) finitely many peak
points. In case (i), suppose the peak points are n1 n2 n3 . . . Then by
the definition of a peak point we have s(n1 ) s(n2 ) s(n3 ) . . . and so there
is a decreasing subsequence. In case (ii) suppose there are finitely many peak
points, say n1 , n2 , . . . , nm and let p > max{n1 , n2 , . . . , nm }. Then there exists a
point p2 > p1 such that s(p2 ) s(p1 ), since p1 is not a peak point. And there
exists a point p3 > p2 such that s(p3 ) s(p2 ), since p2 is not a peak point. In
this way one can construct an increasing subsequence.

Theorem 3.5.10 (Bolzano-Weierstrass) Every bounded sequence has a conver-


gent subsequence.

 
increasing
Proof: Let s be a sequence. Then by lemma 3.5.9, s has an
decreasing
subsequence.  
bounded above
In addition since the sequence is bounded it is and thus
bounded below
by
 
BIST
, there exists l R such that l = lim (s j)(n).
BDST n

Notice this doesnt tell you about the convergence of the sequence s and so is
of limited value.
86 CHAPTER 3. LIMITS OF SEQUENCES

3.6 Series
Our study of sequences now specialises to series. We use the language of se-
quences to define a series.

Let V be a normed vector space. Given a sequence u : N V we want to make


sense of the infinite sum
u(1) + u(2) + . . .
The first step is to define a new sequence s : N V by putting for each n N
n
X
s(n) = u(1) + u(2) + + u(n) = u(i).
i=1

We call s the sequence of partial sums of the terms u(i) and we interpret the
infinite sum as the point l V given by
n
X
X
l = lim s(n) = lim u(i) := u(i)
n n
i=1 i=1

thus for a series we really have an ordered pair of sequences (u, s).

Definition 3.6.1 Let u : N V be a sequence and define a sequence s : N V


by
Xn
s(n) = u(i) = u(1) + + u(n),
i=1

where + is the vector addition in V . Then the ordered pair (u, s) is a series.

u(i) is the ith term of the series

s(n) is the nth partial sum of the series.

By saying the series converges we mean variously



lim s(n)
n



P

u(i)




i=1
l R : P
u(i) = l.

nN


n
P



lim u(i)

n i=1

Example 3.6.2 We will show that 0.999 = 1. We have

9 9 9 9
0.999 = + + 3 + + n + ...
10 100 10 10
3.6. SERIES 87

9
We put u(n) = 10n and
n
9 1 1
P 
s(n) = u(i) = 10 1 + 10 + + 10n1
i=1 
1 n
9 1( 10 )
= 10 1 1
using the sum of a geometric series
10
1
=1 10n .

Now lim s(n) = lim (1 101n ) = 1, by the algebra of limits and the fact that
1 n n
< 1 and so lim 1n = 0. Thus 0.999 =
P
10 u(i) = 1 as required.
n 10 nN

3.6.1 Real-valued series with positive terms


Let u : N R be a sequence of positive terms, i.e. u(n) > 0, for every n N
and let s : N R be the sequence of partial sums of the terms u(i). Then s is
strictly increasing since if n N then

s(n) = u(1) + u(2) + + u(n) < u(1) + u(2) + + u(n) + u(n + 1) = s(n + 1),

since u(n + 1) > 0. For such series if we can prove that the infinite sum is
bounded above, then we can use the BIST to demonstrate the existence of a
limit of the series. Next we give two tests which are useful in demonstrating
that the infinite sum is bounded above.

TEST I- The Comparison test.


Suppose s : N R and t : N R are series with positive terms and suppose
0 < s(n) t(n) for each n N. Then if t(n) converges to a limit l R, s(n) is
bounded above, by l and so by the BIST we have that lim s(n) exists.
n

To use the test, one looks for a series t which satisfies the hypotheses of the test
and which is known (from previous results) to converge.

Example 3.6.3 Let s : N R be the series with

9 9 1 9 1 9 1
s(n) = + 2 + 3 + + n ,
10 10 2 10 3 10 n
we are to show that lim s(n) exists
n

Proof: Let t : N R be the series


9 9 9
t(n) = + + + n
10 102 10
9 1 9
then t(n) is a sequence of positive terms and for every i N, we have 10i i 10i
so for every n N, we have s(n) t(n).
88 CHAPTER 3. LIMITS OF SEQUENCES

In addition, by example 3.6.2, we know that lim t(n) = 1. Hence s(n) is


n
bounded above and so by the comparison test lim s(n) exists.
n

TEST II-The integral test.


Let (u, s) be a series. If f : U R R is a positive, decreasing function on
[1, ) and
Rif for every nR N we have f (n) = u(n) then s(n) converges to a limit
n
l R 1 f := lim 1 f exists.
n

1
Example 3.6.4 Let u : N R be the sequence u(n) = n2 and let s : N R
be the sequence of partial sums of the terms u(i), i.e.

n
X 1 1
s(n) = u(i) = 1 + + + 2.
i=1
22 n

We will show that lim s(n) exists.


n

Proof: Series (u, s) is strictly increasing since u(n) > 0 for every n N. Let
f : [1, ) R be the function f (x) = x12 which is positive and decreasing since
for any y > x, we have y12 < x12 [Alternatively: one has f 0 (x) = x13 < 0 for
x [1, ).]

1
In addition f (n) = n2 = u(n), for every n N. So by test II,

Z n
lim s(n) exists lim f exists.
n n 1

Now

Z n Z n
1 1
lim f = lim dx = lim (x1 )|n1 = lim (1 ) = 1
n 1 n 1 x2 n n n

by the algebra of limits. Hence, by the integral test, lim s(n) exists.
n

We will use this example to see why the integral test works.
3.6. SERIES 89

1 6

1
4
1
9
| | | | | -
1 2 3 ... n 1 n

Notice that
s(n) = 1 + 212 + 312 + + n12
= sumRof the area of the shaded rectangles
n
< 1 + 1 x2 dx
= 1 + (x1 )|n1 = 1 + 1 n1 = 2 n1 < 2.
We give two more easy-to-use tests for convergence of a series with positive
terms.

TEST III- The ratio test.


If lim u(n+1)
u(n) = l < 1 then lim s(n) exists, i.e. the series converges.
n n
If l > 1 the series does not converge and if l = 1 the test does not apply (one
cannot conclude).

Proof: Suppose lim u(n+1) = l < 1 thus > 0, N such that n N,


n u(n)
n > we have
u(n + 1)
u(n) l < .

Since l < 1, s R+ , such that l < s < 1 and


0 < u(n+1)
u(n) s, n >
i.e. u(n + 1) s.u(n), n >
u( + 1) s.u()
and u( + 2) s.u( + 1) s2 .u()
and u( + 3) s.u( + 2) s3 .u()
..
.
u( + k) sk .u().
Now

X
X
sk .u() = u() sk
k=1 k=1
90 CHAPTER 3. LIMITS OF SEQUENCES

a constant times a geometric series with common ratio 0 < s < 1. Hence

sk .u() converges.
P
k=1
Now

X
X
X
u(n) = u( + k) u() sk
n=+1 k=1 k=1

and

X
X
s(n) = u(n) + u(n)
n=1 n=+1

P
so s(n) differs from u(n) by finitely many terms. Thus lim s(n) exists.
n=+1 n

n
1 1
P
Example 3.6.5 Let s(n) = u(i) and u(n) = n2n1 , so u(n + 1) = (n+1)2n .
i=1
Thus
u(n + 1) n2n1 n
= n
= .
u(n) (n + 1)2 2(n + 1)
And  
u(n + 1) n 1 1 1
lim = lim = lim 1 =
n u(n) n 2(n + 1) n 2 1+ 2
n

by the algebra of limits. Since 1/2 < 1 by the ratio test, we have that the series
converges, i.e. lim s(n) exists.
n

We introduce some notation before giving test IV.


Let s : N R be a bounded sequence in R and let

Sn = {s(n), s(n + 1), . . . } = {s(i) : i n}

so Sn is a bounded set. Define

bn = sup Sn = supin {s(i)} and


.
an = inf Sn = inf in {s(i)}

So n N, an bn and
both (an )nN and (bn )nN are bounded sequences.

Claim: (bn )nN is a decreasing sequence and


(an )nN is an incresing sequence.

Proof: Let n N, bn = supin {s(i)} and bn+1 = supin+1 {s(i)} so bn+1 is


the supremum over a smaller set than bn , thus bn+1 bn , so the sequence is
decreasing.
Let n N, an = inf in {s(i)} and an+1 = inf in+1 {s(i)} so an+1 is the infimum
3.6. SERIES 91

over a smaller set than an , thus an+1 an , so the sequence is increasing.


Thus by the BIST lim an exists and by the BDST lim bn exists. We have
n n

lim bn = inf nN {bn } = inf nN (supin {s(i)})


n
lim an = supnN {an } = supnN (inf in {s(i)}).
n

Definition 3.6.6

lim inf s(n) = lim an := limit inferior


n
lim sup s(n) = lim bn := limit superior.
n

Consider for example s(n) = (1)n + n1 . We have

Sn = {s(n), s(n + 1), . . . } = {s(i) : i n}


bn = sup Sn
an = inf Sn so
b1 = sup{1 + 1, 1 + 12 , 1 + 13 , . . . } = 1 21
b2 = sup{1 + 12 , 1 + 31 , . . . } = 1 12
b3 = sup{1 + 13 , 1 + 14 , . . . } = 1 14
..
.
so lim bn = 1 and
n
a1 = inf{1 + 1, 1 + 12 , 1 + 31 , . . . } = 1
a2 = inf{1 + 21 , 1 + 13 , . . . } = 1
a3 = inf{1 + 13 , 1 + 41 , . . . } = 1
..
.
so lim an = 1 thus
n

lim sup s(n) = 1 and lim inf s(n) = 1.

TEST IV- Root test p


Let (u, s) be a series. If lim sup n |u(n)| = l < 1 then lim s(n) exists.
n
As in the ratio test if l > 1, the series diverges and if l = 1 the test does not
apply.

A modification of the root test is as follows:


If l : 0 < l < 1 and N such that
p
n
|u(n)| l, for every n >

then lim s(n) exists.


n
92 CHAPTER 3. LIMITS OF SEQUENCES

Proof: (of the modified root test)


Suppose l (0, 1) R and N such that n N, n > we have
p
n
p
|u(n)| = n u(n) l for series with positive terms
i.e. u(n) l . Now
n
P
lim s(n) = lim u(i)
n n i=1
P
= u(i)
i=1
P
P
= u(i) + u(i)
i=1 i=+1
P P
= u(i) + u( + i)
i=1 i=1

i
P P
u(i) + l l.
i=1 i=1

li is a geometric series so it converges, thus lim s(n) exists as required.
P
Now
i=1 n

Example 3.6.7 Given s(n) = 21 + 13 + 212 + 312 + 213 + . . . we have u(2m 1) =


1/2m , u(2m) = 1/3m .
Now the ratio test cannot apply since if n is even, say n = 2m then
 m
u(n + 1) u(2m + 1) 1 3
= = ,
u(n) u(2m) 2 2
m
but 23 , diverges as m . On the other hand the root test applies.
If n = 2m then p p
2m
u(2m) = 2m 1/3m = 1/ 3 < 1.
And if n = 2m 1 then
2m1
p 1 m/(2m1)
u(2m 1) = < 1.
2
Thus the root test guarantees convergence.

In fact example 3.6.7 is a bit artificial in that s(n) is clearly the sum of two geo-
metric progressions and so there is a closed form for the finite sum of the terms
in each progression, thus s(n) is bounded above and increasing, so existence of
lim s(n) follows from the BIST.
n

3.7 Absolute convergence and Cauchy sequences


Definition 3.7.1 A series (u, s) (in which the terms u(n) are not necessarily
n
P
positive) is absolutely convergent the sequence t, given by t(n) = |u(i)|
i=1
is convergent.
3.7. ABSOLUTE CONVERGENCE AND CAUCHY SEQUENCES 93

(1)n+1
Example 3.7.2 Consider the series (u, s) with u(n) = n2 .

Claim: s is absolutely convergent.


Proof: Let t : N R be the sequence
n
(1)n+1

X 1 1
t(n) = n2 = 1 + 2 2 + + n2 .

i=1

By example 3.6.4, the series t converges, thus s is absolutely convergent.


Theorem 3.7.3 Every absolutely convergent real-valued series is convergent.
We will prove theorem 3.7.3 after we have examined the notion of a Cauchy
sequence. Suppose we have a sequence such that s(1), s(2), l, as n ,
what else can we say about sequence s without explicitly mentioning l?
Definition 3.7.4 Let (V, || ||) be a normed vector space. A sequence s : N
V is a Cauchy sequence > 0, N such that m, n > we have
||s(n) s(m)|| < .
1
Example 3.7.5 The sequence s : N R given by s(n) = n is a Cauchy
sequence.

Proof: Let > 0.

Choose = b 1 c + 1 N. Find N:
Let m, n > , with m > n m, n > , |s(n) s(m)| <
then n > 1 , i.e. i.e. | n1 m
1
| < ;
> n1 > |s(n) s(m)|. now for m > n, 1 1
m < n so
| n1 m
1
| = n1 m
1
< n1
1
so we want n > .

On the other hand



Example 3.7.6 The sequence s : N R with s(n) = n is not a Cauchy
sequence.
To prove: > 0 such that N, m, n > and |s(n) s(m)| .
Proof:

Consider = 1. Find > 0: N


Let N and choose can find m, n > and
m = + 1 and n = 4( + 1) then |s(n) s(m)|
,
m, n > and i.e. | n m| ;
|s(n) s(m)| = + 1 > 1 = . now for n =4m, n m = m
so we want m . Take = 1.
Theorem 3.7.7 Every convergent sequence is a Cauchy sequence.
94 CHAPTER 3. LIMITS OF SEQUENCES

Proof: Suppose s : N V is convergent and let l = lim s(n).


n
Let > 0, choose to be the natural number for which ||s(n) l|| < /2,
n N, n > . Let m, n N, m, n > then

||s(n) s(m)|| ||s(m) l|| + ||s(n) l|| < /2 + /2 = .

Hence s is a Cauchy sequence.

The contrapositive of theorem 3.7.7, is referred to as the Cauchy criterion.


It gives a sufficient condition for the non-existence of lim s(n).
n
For example in 3.7.6 we had s(n) = n is not a Cauchy sequence, hence by
theorem 3.7.7 it cannot converge.
Definition 3.7.8 Let s : N V , t : N W be two sequences mapping into
normed vector spaces (V, || ||V ) and (W, || ||W ) respectively.
We say t dominates s

m, n N, ||s(m) s(n)||V ||t(m) t(n)||W .

Theorem 3.7.9 (Domination by a Cauchy sequence)


Let s : N V , t : N W be two sequences such that t dominates s. If t is a
Cauchy sequence then s is a Cauchy sequence.

The proof of this theorem is quite straight forward and so we will leave it to the
exercises.

The contrapositive of theorem 3.7.9 is a useful test for showing that a given
sequence is NOT a Cauchy sequence. It avoids the difficulties involved in trying
to prove this directly from the definition. (Recall example 3.7.6, where we have
to make a choice of before we can begin the proof.)
Lemma 3.7.10 Every real-valued Cauchy sequence is bounded above and below.
This result can be extended to vector-valued sequences we just have to use the
correct definition of being bounded. Here we will confine ourselves to real-valued
sequences.

Proof: Let s : N R be a Cauchy sequence. So > 0, N, such that


m, n N, m, n > we have |s(n) s(m)| < .
In particular, for = 1 we can choose N: m, n > |s(n) s(m)| < 1. So
for m = + 1
|s(n) s( + 1)| < 1 thus

s( + 1) 1 < s(n) < s( + 1) + 1, n > + 1

i.e. there are infinitely many terms

s( + 1), s( + 2), s( + 3), . . .


3.7. ABSOLUTE CONVERGENCE AND CAUCHY SEQUENCES 95

that are bounded above by s( + 1) + 1 and below by s( + 1) 1. In addition


there are finitely many terms s(1), s(2), . . . , s() for which there is a least l and
a greatest u, i.e.;

l = min{s(1), s(2), . . . , s()} max{s(1), s(2), . . . , s()} = u.

Thus an upper bound for {s(n) : n N} is max{s( + 1) + 1, s(u)} and a lower


bound for {s(n) : n N} is min{s( + 1) 1, s(l)}.

Lemma 3.7.11 Every Cauchy sequence in R is convergent; i.e. converges to a


real number.

The proof uses the least upper bound property of R.


Proof: Let s : N R be a Cauchy sequence.[To prove l R such that
l = lim s(n).]
n
Define S = {x R : x < s(n) for infinitely many values of n} R. [The
idea is to construct a set S which has a least upper bound and prove that
sup S = lim s(n).]
n
By lemma 3.7.10, we have that s is bounded i.e. L, U R such that L
s(n) U for every n N.

Now for any x R such that L < x < U we have x S, since x R and
x < s(n) for infinitely many values of n, so S is bounded and non-empty, thus
sup S exists.
Let l = sup S.
Claim: l = lim s(n).
n

Proof: Let > 0, since l = sup S, l is an upper bound of S and so there are
infinitely many n N for which s(n) (l /2, l + /2). Since s is Cauchy
it means that from some N onwards, all elements of the sequence lie in
(l /2, l + /2) i.e.

n N, n > , |s(n) l| < /2 < .

Hence l = lim s(n) as required.


n

Lemma 3.7.11 is not true for a Cauchy sequence in an arbitrary normed vector
space. Indeed it is not true for subsets of R, for example
Q. The sequences sn
and tn constructed in section 3.4 as approximations
to 2 are Cauchy sequences
of rational numbers which both converge to 2 which is not rational. Thus
sequences sn and tn are not convergent in Q. We have

Definition 3.7.12 A normed vector space V is said to be complete every


Cauchy sequence converges to a point l V .
A Banach space is a complete normed vector space.
96 CHAPTER 3. LIMITS OF SEQUENCES

We are now in a position to prove theorem 3.7.3, that every absolutely conver-
gent sequence in R is convergent (converges in R).

Proof: Let (u, s) be an absolutely convergent series in R. So the series (|u|, t)


converges (in R). Thus by theorem 3.7.7, (|u|, t) is a Cauchy sequence.

Claim: Sequence t dominates sequence s.


Proof: Let m, n N, m < n then

||s(m) s(n)|| = || u(m + 1) u(m + 2) u(n)||


= | 1||u(m + 1) + + u(n)| by homogeneity
|u(m + 1)| + + |u(n)| by the 4-inequality
= |t(n) t(m)|.

Thus s is dominated by t, which is a Cauchy sequence. Thus by theorem 3.7.9


s is a Cauchy sequence. Now by theorem 3.7.11, s is convergent i.e. lim s(n)
n
exists. Now it is easy to see that one can generalise theorem 3.7.3 to

Theorem 3.7.13 In a Banach space, every absolutely convergent sequence is


convergent.

The converse of theorem 3.7.13 is false i.e. there exist convergent sequences
which are not absolutely convergent.

Example 3.7.14 Consider the sequence (u, s) defined by

1 1 (1)n+1
s(n) = 1 + + ,
2 3 n

n
(1)n+1 P
i.e. u(n) = n and s(n) = u(i).
i=1

Claim: The sequence s converges.


We will prove that s is a Cauchy sequence and so since it is in R a complete
space, the result follows.

Proof: Let m, n N with n > m. There are cases to argue according as to


whether m, n are odd or even.
3.7. ABSOLUTE CONVERGENCE AND CAUCHY SEQUENCES 97

(i) Suppose n, m are both even. Then


n m

P (1)i+1 P (1)i+1

|s(n) s(m)| =
i i
i=1 i=1
= m+1 m+2 + + n1 n1
1 1 1
1 1 1
(m+1) 2 + (m+3)2 + + (n1)2 (since m, n are even, there

are an even number of terms and so can consider them pair-


1 1 1 1
wise. Notice that m+j m+j+1 = (m+j)(m+j+1) (m+j)2)

1 1 1 1
2 + (m+2)2 + (m+3)2 + + n2
R(m+1)
n
m+1 x2 dx
= x1 |nm+1
1 1 1 1
= m+1 n < m+1 , since n >0
1
< m.

So, fix > 0 and choose = b 1 c + 1 N. Let n > m > , then > 1
so

1 1
> > > |s(n) s(m)|.
m
Hence by lemma 3.7.11, lim s(n) exists.
n
(ii) If n, m are both odd we have

1 1 1 1
|s(n) s(m)| = | 1| + +
m+1 m+2 n 1 n

and so (i) applies.


(iii) If n even and m odd then

1 1 1
|s(n) s(m)| = m+1 + m+2 + n1 n1

(an odd number


 of terms
  
1 1 1 1
= m+1 + m+2 m+3 + + n1 n1

n
1
+ m+2 x2 dx
R
m+1
1 1
= m+1 + m+2 n1
2m+3
(m+1)2
3m
(m+1) 2 for m > 3

3m
m2 = 3
m

and so we can again proceed as in (i) with = b 3 c + 1.

n
P (1)i+1
But the sequence s(n) = i is not absolutely convergent.
i=1
n n
P (1)i+1 P 1
Proof: Let t(n) = i = i.
i=1 i=1
98 CHAPTER 3. LIMITS OF SEQUENCES

Claim: t is not a Cauchy sequence.


To prove the claim, we must find > 0 such that N we can find n, m >
and |s(n) s(m)| . Proof: Consider = 1/2. Let N and choose
m = + 1 and n = 2( + 1). Then n > m > . And
1 1
|t(n) t(m)| = m+1 + m+2 + + n1
1 1
(n m) n since m+1 , . . . , n1 1
n
m
=1 n
= 1 1/2 = 1/2 = .

Hence t is not a Cauchy sequence and so by the Cauchy criterion, it does not
converge. Hence s is not absolutely convergent.

In fact there is a more general result for alternating sequences.

Theorem 3.7.15 (Alternating sequence theorem) Let s : N R be a sequence


n
P
with s(n) = u(i), where u : N R is a sequence satisfying
i=1

(i) u is an alternating sequence

(ii) lim u(n) = 0 and


n

(iii) n, |u(n + 1)| < |u(n)|, (unless |u(n)| = 0 for some n and then u(m) = 0
m > n)

then the series converges and



X
u(n) = | lim s(n)| |u(1)|.


n
n=1

Proof: Suppose u(1) > 0 so we can write

u(1) + u(2) + . . . in the form


b(1) c(1) + b(2) c(2) + where
b(1) = u(1); b(n), c(n) 0 and b(n), c(n)
are subsequences of u.

Let r(n) = b(1) c(1) + + b(n 1) c(n 1) + b(n) and let


t(n) = b(1) c(1) + + b(n 1) c(n 1) + b(n) c(n)
then r(n + 1) = r(n) c(n) + b(n + 1) so r(n + 1) r(n), since
0 b(n + 1) < c(n) < b(n) ( |u(n)| is strictly decreasing by hypothesis) thus

r(1) r(2) r(3) . . .

and similarly
t(1) t(2) t(3) . . .
3.7. ABSOLUTE CONVERGENCE AND CAUCHY SEQUENCES 99

In addition t(n) = r(n) c(n) and c(n) 0 so t(n) r(n) for every n N. i.e.
r(1) r(2) t(n) t(2) t(1).
Let > 0. [We must find N such that n > , 0 r(n) t(n) = c(n).] We
can choose : |c(n)| < , for all n > . Then 0 |r(n) t(n)| = |c(n)| < , for
all n > .
Now let m > n > , then
|r(n) r(m)| |r(n) t(n)| <
so r is a Cauchy sequence in R, thus r converges and lim s(n) = inf nN {r(n)} =
n
supnN {t(n)}. In addition,
0 b(1) c(1) = r(1) c(1) = t(1) lim s(n) r(1) = b(1) = u(1).
n

Hence
X
u(n) = | lim s(n)| |u(1)|.


n
n=1

Now for u(1) < 0, consider the series formed using the terms u(n), n N,
and apply the above argument.
Example 3.7.16 We want to decide whether or not the series

X n+1 n
(1)n
n
nN

(i) converges and (ii) converges absolutely.


1 1
Notice n + 1 n = n+1+ n
and lim n+1+ n
= 0 by the algebra of
n
limits. So

n+1 n 1
0< < n+1 n=
n n+1+ n

n+1 n
thus lim n = 0 by the squeeze principle.
n


n+1+ n
Similiarly lim n = 0.
n

In addition, the sequence of terms is alternating and its easy to check that
n N
n + 1 n n + 2 n + 1
>
n n+1
so by the alternating sequence test we have that

X n+1 n
(1)n converges.
n
nN
100 CHAPTER 3. LIMITS OF SEQUENCES

In addition the series converges absolutely, since


P
n+1 n n+1 n
(1)n
P
n = n
nN nN
1
P
=
n n+1+n n
nN
1
P
2n n
.
nN

Now consider f : [1, ) R given by f (x) = 12 x3/2 . Then f is positive and


decreasing and f (n) = 2n1n . So f satisfies the hypothesis of the integral test.
In addition Z n n
1 3/2 1
x dx = x1/2 = 1 ,

1 2 1 n
Rn P 1
thus lim 1 f (x)dx = 1. By the integral test the series converges
2n n
n nN
and so by the comparison test, the original series converges absolutely.
[Notice the ratio test does not apply here since the limit of the ration of con-
secutive terms is 1.]
Lemma
P 3.7.17 ForPany sequence (an )
0 in a normed vector space V
n
if n=0 an := lim i=0 ai = l V, then lim an = 0.
n n

Pn
Proof: Let (an )nN be a sequence in V and suppose lim i=0 ai = l V .
n

Let > 0 and let N satisfy


n
X
|| ai l||V < for every n + 1.
i=0
2

Then Pn Pn1
||an ||V = || i=0 ai l + l ai ||V
i=0
Pn Pn1
|| i=0 ai l||V + ||l i=0 ai ||V
< 2 + 2 =
for every n + 1. Thus lim an = 0.
n

3.8 From Q to R
In Math 211, we saw how to construct the integers from the natural numbers
and then how to construct the rational numbers from the integers. In addition,
we examined how to construct the complex numbers from the real numbers. The
missing link is how to construct the real numbers from the rational numbers.

Instead of taking the Cartesian product of two copies of Q, we consider the


countable product of the rational numbers. Let = #N, the cardinality of the
3.8. FROM Q TO R 101

natural numbers. Then

Q = Q Q = {(x1 , x2 , . . . ) : xi Q}

i.e. Q is the set of all sequences of rational numbers.


Now consider the subset S Q consisting of all of the Cauchy sequences in
Q .

We define a relation R3 on S by

s, t S sR3 t > 0( Q), M N


such that |s(n) t(n)| < , n > M.

Claim: R3 is an equivalence relation on S.


It is reflexive since |s(n) s(n)| = 0 <
it is symmetric since |s(n) t(n)| = |t(n) s(n)| and
it is transitive, since if s(n) t(n)| < /2 and |t(n) r(n)| < /2 then
|s(n) r(n)| < by the triangle inequality.

Now define + by s, t S

s + t is the sequence (si + ti )iN and

define by
s t is the sequence (si ti )iN .
Now we need to check
(i) S is closed under + and and
(ii) the operations are independent of the class name
and then we need to understand the sense in which we have constructed the real
numbers from the rational numbers.

(i) Let s, t S, i.e. s, t are Cauchy sequences. Let > 0 and choose =
max{1 , 2 } where 1 N satisfies
m, n > 1 we have |s(n) s(m)| < /2 (1)
and 2 N satisfies
m, n > 2 we have |s(n) s(m)| < /2. (2)

Now let m, n > so both (1) and (2) hold. Then

|(s + t)(n) (s + t)(m)| = |s(n) + t(n) s(m) t(m)|


|s(n) s(m)| + |t(n) t(m)|
< /2 + /2 = .
102 CHAPTER 3. LIMITS OF SEQUENCES

So s + t is a Cauchy sequence, i.e. s + t S.


Again suppose s, t S. Let > 0 and choose = max{1 , 2 } where 1 N
satisfies
m, n > 1 we have |s(n) s(m)| < 2l2 (1)
and 2 N satisfies
m, n > 2 we have |s(n) s(m)| < 2l1 , (2)
where l1 is the upper bound for s and l2 is the upper bound for t. ( l1 , l2 exist
by lemma 3.7.10).

Let n, m > so (1) and (2) hold and


|s(n)t(n) s(m)t(m)| |s(n)||t(n) t(m)| + |t(m)||s(n) s(m)|
< l1 2l1 + l2 2l2
=
thus (s.t)nN is a Cauchy sequence.

(ii) To see that the operations +, are independent of the class name, suppose
sR3 s0 and t S.
Claim: (s + t)R3 (s0 + t) (resp. (s t)R3 (s0 t)).
Proof: Let s, s0 t S and suppose sR3 s0 , so > 0, M N such that n
N, n > M , we have |s(n) s0 (n)| < .
Let > 0 and choose = M N which satisfies n N, n > M , we have
|s(n) s0 (n)| < .
Let n N, n > M then
|(s + t)(n) (s0 + t)(n)| = |s(n) s0 (n)| <
as required. So (s + t)R3 (s0 + t).

Again let > 0 and choose = M N which satisfies n N, n > M , we have


|s(n) s0 (n)| < l ,
where l is an upper bound for the sequence t.
Let n N, n > M then

|(s t)(n) (s0 t)(n)| = |t(n)|s(n) s0 (n)| < l =
l
as required. So s tR3 s0 t.

The result is the field R of real numbers. Any given rational number q cor-
responds to that class of Cauchy sequences which has the constant sequence
{q, q, q, . . . } as a member. R also containsnon-rational numbers (recall the se-
quences of successive approximations to 2 is section 3.4.1. These sequences
are convergent and so they are Cauchy sequences.

In addition, R contains non-rational numbers which are not algebraic (i.e. not
the solution to any polynomial equation) for example e corresponds to the class
3.8. FROM Q TO R 103

1 1 1 1
containing the sequence of rational numbers {1, 2, 2 + 2! , 2 + 2! + 3! , 2 + 2! +

1 1
P 1
3! + 4! , . . . } i.e. e = n! . is also expressible as a sum of rational numbers
n=0

4
(1)n 2n+1
P
for example = . The numbers e and are non-algebraic, non-
n=0
rational numbers.
104 CHAPTER 3. LIMITS OF SEQUENCES
Part III

MATH 334, Part I-


Sequences and Series in
Function Spaces

105
Chapter 4

Function Spaces

4.1 Definitions
We can also define sequences and series of functions. We create normed vector
spaces (metric spaces or topological spaces) from sets of functions and refer to
them as function spaces.

Let
F[a, b] = {f : [a, b] R, a < b, a, b R},
so F[a, b] is a set of real-valued functions with domain the closed interval [a, b]
R. First we add the vector space structure: we define an addition of functions
and an action of the field, usually R, (also thought of as the scalar multiplication)
pointwise, i.e. by stating the value of the sum of functions at each point in the
domain.

f, g F[a, b], define f + g by x [a, b], (f + g)(x) = f (x) + g(x) and


f F[a, b], R define f by x [a, b], (f )(x) = f (x) (a product of
real numbers).

With these operations it is routine to check that F[a, b] is a vector space over
the field of real numbers.

We are interested in certain vector subspaces of F[a, b] i.e. subsets of F[a, b]


which are vector spaces in their own right.

In particular we define

B[a, b] = {f : [a, b] R and f is bounded}

C[a, b] = {f : [a, b] R and f is continuous}.

107
108 CHAPTER 4. FUNCTION SPACES

We have C[a, b] B[a, b] because any continuous function on a bounded closed


interval is bounded. (The result is the maximum/minimum value theorem which
we examine in the last part of the course). We define

D[a, b] = {f : [a, b] R and f is differentiable}

and you know from elementary calculus that differentiable functions are contin-
uous so we have the set containment

D[a, b] C[a, b] B[a, b].

And the other set of interest will be


S[a, b] =
{f |[a,b] | where f : R R and f is a step function, with f = 0 on R\[a, b]},

where

Definition 4.1.1 (Step function)


f : R R is a step function if and only if there exists a partition
a = x1 < x2 < < xn = b of [a, b] such that f is constant on each open
subinterval of the partition and zero on R\[a, b].

Example: Let : R R be the function with




0 if x 1
1 if 1 < x 1




2 if 1 < x < 2

(x) =

0 if x = 2
1 if 2<x3




0 if 3 < x

The most obvious partition compatible with is {1, 1, 2, 3}, since is constant
on each of the open subintervals (1, 1), (1, 2), (2, 3) and it is zero on R\[1, 3].
Hence is a step function.

Clearly a step function only takes finitely many real values and so it is bounded.
But a step function is certainly not continuous, in fact it has finitely many points
of discontinuity.

We want to examine the convergence of sequences in these function spaces and


so we will need to define a norm on the vector space B[a, b]. For every f B[a, b],
the set {f (x) : x [a, b]} is a non-empty bounded set of real numbers and so
by the least upper bound propery of R this set has a least upper bound i.e.
sup{f (x) : [a, b]} exists. (Also inf{f (x) : x [a, b]} exists.) But neither of these
will provide a norm as they may not satisfy positivity.
4.2. SUPREMUM NORM. 109

4.2 Supremum norm.

Proposition 4.2.1 Let || || : B[a, b] R be the map


||f || := sup{|f (x)| : x [a, b]}.
Then || || is a norm on B[a, b].

 f is a real valued function |f (x)| refers to the usual norm on R, i.e.


Notice since
f (x) if f (x) 0
|f (x)| =
f (x) if f (x) < 0.

Proof: (N1 positivity)


Let x [a, b] then |f (x)| 0, since | | is a norm on R and so ||f || is the
supremum of a set of non-negative numbers, hence ||f || 0.
Suppose ||f || = 0. Then for every x [a, b], |f (x)| = 0, since ||f || is the
supremum of a set of non-negative numbers. Hence f (x) = 0 for every x [a, b],
since | | is a norm and so satisfies positivity, thus f = 0; i.e. ||f || = 0 f = 0,
where the first zero is a real number and the second zero is the zero function on
[a, b].

Conversely, suppose f = 0 on [a, b], then for every x [a, b] f (x) = 0, thus
|f (x)| = 0 and so ||f || = 0, since ||f || is the supremum of the set containing
zero. So f = 0 ||f || = 0, completing the proof of positivity.

(N2- the 4-inequality)


Let f, g B[a, b]. We have |f (x) + g(x)| |f (x)| + |g(x)| for every x [a, b],
since | | satisfies the triangle inequality. So
x [a, b], |f (x) + g(x)| ||f || + ||g||
since ||f || (resp. ||g||) is an upper bound for {|f (x)| : x [a, b]} (resp.
{|g(x)| : x [a, b]}). Thus ||f || + ||g|| is an upper bound for
{|f (x) + g(x)| : x [a, b]} = {|(f + g)(x)| : x [a, b]}
(by definition of the vector addition).

But ||f + g|| is by definition of the map || || the least upper bound of
{|(f + g)(x)| : x [a, b]} and so
||f + g|| ||f || + ||g||
as required.

(N3- homogeneity)
Let f B[a, b] and let R. For each x [a, b], we have |f (x)| = |||f (x)|
110 CHAPTER 4. FUNCTION SPACES

since | | satisfies homogeneity being a norm on R. Also for each x [a, b],
|f (x)| sup{|f (x)| : x [a, b]} = ||f || thus

|||f (x)| ||||f ||

i.e. ||||f || is an upper bound for

||{|f (x)| : x [a, b]}


= {|||f (x)| : x [a, b]}
= {|(f )(x)| : x [a, b]}

by definition of the scalar multiplication and the homogeneity of | |.


But ||f || is the least upper bound of {|(f )(x)| : x [a, b]} so

||f || ||||f ||()

On the other hand, ||f || is an upper bound for {|(f )(x)| : x [a, b]} and
||||f || is the least upper bound of {|||f (x)| : x [a, b]} = {|(f )(x)| : x
[a, b]}, thus
||||f || ||f ||.()
By () and () we have the required equality.

Definition 4.2.2 For X = B[a, b] or C[a, b] or D[a, b] or S[a, b] the map || || :


X R defined by
||f || := sup{|f (x)| : x [a, b]}
is called the supremum norm on X.

Remarks:

(i) A useful inequality which follows from the definition is

x [a, b], |f (x)| ||f ||.

(ii) ||f || is not necessarily defined for f F[a, b], we need the function to be
bounded so that the required least upper bound exists.

Proposition 4.2.3 (a) B[a, b] and C[a, b] are complete normed vector spaces
i.e. every Cauchy sequence converges to a point in the space.

(b) D[a, b] and S[a, b] are not complete normed vector spaces i.e. there exists
a Cauchy sequence which does not converge to a point in the space.

There is a proof for B[a, b] in the appendix, theorem 5.2.2 is the proof for C[a, b]
and we will examine some examples to demonstrate (b).
4.3. BALLS IN B[A, B] 111

4.3 Balls in B[a, b]


Definition 4.3.1 Let f B[a, b] and let r > 0, then the open ball centred on
f of radius r is the set

Br (f ) = {g B[a, b] : ||f g|| < r}

and the closed ball

Br (f ) = {g B[a, b] : ||f g|| r}.

So Br (f ) is the set of all bounded functions g : [a, b] R such that the graph
of g lies between the graphs of f + r and f r, where for every x [a, b],
(f r)(x) = f (x) r(x) = f (x) r, by definition of the vector addition and
the constant map r.

We can think of Br (f ) as the set of all bounded functions g such that the graph
of g lies in the r-strip around the graph of f .

f +r
f
f (a) + r . . . . . . f r
f (a) . . . . . .
f (a) r . . . . . .

-
a ... b

Examples:

1.


x if 1 x < 2
(a) f : [1, 2] R with f (x) =
1 x = 2.
112 CHAPTER 4. FUNCTION SPACES

6
2

-
1 2

We have
||f || = supx[1,2] {|f (x)|}
.
= sup[1, 2) = 2
[Think about how to prove the last equality.]
||f || is the supremum over x [1, 2], of the distance that f (x) is from
zero.
(b) We will show that g : [1, 2] R with

(x 1)2 + 2

x [1, 2)
g(x) =
1/2 x=2

is in the closed ball B1 (f ), where f is the function in (a).

Recall (i) B1 (f ) = {g B[1, 2] : ||g f || 1} and


(ii) for bounded functions on bounded intervals extreme values (i.e.
local maxima and minima) occur either at the endpoints of the in-
terval or at points x such that f 0 (x) = 0.
We are required to prove that ||f g|| 1, where

||f g|| = sup {|(f g)(x)|} = sup {|f (x) g(x)|}.


x[1,2] x[1,2]

Now
x ((x 1)2 + 2) x [1, 2)

f (x) g(x) =
 1/2 2 x=2
x + x 2x + 1 2 x [1, 2)
=
 1/2 x=2
x2 x 1 x [1, 2)
=
 1/2 x=2
(x 1/2)2 5/4 x [1, 2)
=
1/2 x=2

so the graph of f g is
4.3. BALLS IN B[A, B] 113

6
1

1+ 5

2 & -
1 2
-1

and the graph of |f g| is


6
1


-
1+ 5
2
-1

from which one has supx[1,2] {|f (x) g(x)|} = 1.


(c) Suppose now that h = f on [1, 2) and h(2) = 2 and show that
supx[1,2] {|h(x) g(x)|} = 3/2.

2. Let f, g B[0, ] withf (x) = sin x and g(x) = cos x.


Claim: ||f g|| = 2 in B[0, ] with supremum norm.
Proof:
||f g|| = supx[0,] {| sin x + cos x|}. Now sin and cos are differentiable so
(sin + cos)0 exists and extreme values occur either at the endpoints 0 and
or at points x such that (sin + cos)0 (x) = 0.
We have sin 0 + cos 0 = 1, sin + cos = 1 and
(sin + cos)0 (x) = (cos sin)(x) so

(sin + cos)0 (x) = 0 cos(x) = sin(x)


x = /4 since x [0, ].

In addition,
sin /4 + cos /4 = 2.1/ 2 = 2,

thus ||f g|| = supx[0,] {|f (x) g(x)|} = 2 as claimed.

3. In B[0, 1] with supremum norm we want to find a number r > 0 such that
B1/4 ( 12 Id) Br (Id), where
1
2 Id : [0, 1] R Id : [0, 1] R
and
x 7 x2 x 7 x.
114 CHAPTER 4. FUNCTION SPACES

To find an appropriate r > 0 we can try using the graphs of 21 Id and Id:

1
6
3
4

1
2 need r 3/4

1
4

-
1

or we can do the usual thing and think of open balls as discs in the plane
(not including the boundary) and then calculate the distance between the
centres of the balls i.e.
1 1 x
|| Id Id|| = || Id|| = sup {| |} = 1/2
2 2 x[0,1] 2

because Id2 is monotonic increasing on [0, 1] and so the supremum occurs


at the right end point of the interval. Now

1
2
z}|{
1
2 Id Id so choose r = 3/4.

Now we must prove the set containment

1
B 14 ( Id) B 34 (Id).
2

Let f B 14 ( 12 Id), so ||f 21 Id|| < 14 . Now

||f Id|| = ||f 21 Id + 12 Id Id||


||f 12 Id|| + || 12 Id Id||
1 1 3
< 4 + 2 = 4

so f B 34 (Id).
4.3. BALLS IN B[A, B] 115

4. With f, g as in example 1, find r > 0 so that B1 (f ) Br (g). We have


||f g|| = 1 and so would need r 2. Now you must prove the set
containment B1 (f ) B2 (g). Is it possible to find r > 0 so that Br (f )
B1 (g)?
116 CHAPTER 4. FUNCTION SPACES
Chapter 5

Sequences in B[a, b]

A sequence (fn )nN in B[a, b] is a sequence of functions f1 , f2 , . . . indexed by


N such that for each i N fi : [a, b] R is bounded.

The rules for limits of sequences in normed vector spaces and the theorems
developed in chapter 3, section 4 are valid in this setting with the exception of
those results which specifically referred to real-valued sequences, for example
limit of an ordered pair and products and quotients. In the case of limit of
an ordered pair, in order to generalise the result, we would need to discuss an
appropriate norm for the product of normed vector spaces.

5.1 Uniform and pointwise convergence


Definition 5.1.1 (Uniform limit). Given a sequence (fn )nN in B[a, b] the
uniform limit of fn is f
> 0, N such that n N, n > we have ||fn f || < .
The expression ||fn f || < means that for every x [a, b], |fn (x) f (x)| < .
One interprets it to mean that the graph of fn lies in an -strip around the
graph of f for all values of n > and so for all but finitely many values of n.
Proposition 5.1.2 Given a sequence (fn )nN in B[a, b] the uniform limit of
fn is f
the sequence (dn )nN of real numbers
dn = supx[a,b] {|fn (x) f (x)|} converges to zero.
Proof of the proposition is on the exercises.

This type of convergence is referred to as uniform convergence because the choice


of applies uniformly to every point x [a, b]. We write lim fn = f (uniform)
n
u
or fn f .

117
118 CHAPTER 5. SEQUENCES IN B[A, B]

Given > 0, the graphs of functions f1 , f2 , . . . f may have points which lie
outside the -strip around f , but the point is that for all n > , the graph of fn
lies inside the -strip.
Examples
1. Let (fn )nN be the sequence with fn = n1 cos (nId) in B[, ] i.e.

1
x [, ], fn (x) = cos(nx).
n

We sketch the first few graphs of the sequence in order to guess the uniform
limit.

1 1
f1 (x) = cos x f2 (x) = cos 2x f3 (x) = cos 3x.
2 3

1 u
So as n , we have n 0 and so we guess that fn 0.

Claim: 0 : [, ] R is the uniform limit of the given sequence. We are to


prove

> 0, N such that


n N, n > we have ||fn 0|| = ||fn || < .
5.1. UNIFORM AND POINTWISE CONVERGENCE 119

Proof: Let > 0.

Exploration
To find N : n >
||fn 0|| = ||fn ||
= || n1 cos nId|| < .
Now | n1 cos nx| 1
so
Choose = b 1 c + 1 N. Let n
1
n N, n > then n > 1 and n is an upper bound
||fn || = || n1 cos nId|| = n1 of {| n1 cos nx| : x [, ]}
by (*). So ||fn || < 1
In addition fn (0) = n so
1
n {| n1 cos nx| : x [, ]}
Thus n1 is the supremum
of the set(*).
So we need n > 1 .

Graphically we assert that for all but finitely many values of n, the graph of fn
lies in the -strip around the zero function 0.

Definition 5.1.3 (Pointwise limit) Given a sequence (fn )nN in B[a, b] the
pointwise limit of fn is f

x [a, b], > 0, x N such that


n N, n > x we have |fn (x) f (x)| < .

This type of convergence is referred to as pointwise convergence and we write


p
lim fn (x) = f (x) (pointwise) or fn (x) f (x).
n

So the order of quantification

for uniform limit , . . . x


for pointwise limit x, , x . . .

thus the in the definition of the uniform limit does not depend on x, whilst the
x in the definition of the pointwise limit can depend on the x. An immediate
consequence of the order of quantification in these two definitions is

Proposition 5.1.4 Let (fn )nN be a sequence in B[a, b]. If (fn )nN is uni-
formly convergent then (fn )nN is pointwise convergent.

Proof: Suppose (fn )nN is uniformly convergent to a function f , so > 0,


N such that n N, n > we have ||fn f || < .

Now let x [a, b] and let > 0. Then choose the which satisfies the statement
of uniform convergence. Let n N, n > , then |fn (x) f (x)| ||fn f || < .
Hence (fn )nN converges to f pointwise.
120 CHAPTER 5. SEQUENCES IN B[A, B]

The converse of proposition 5.1.4 is false (otherwise there would be no difference


between the two definitions), a sequence may have a pointwise limit which is
not a uniform limit. For example
2. consider the sequence fn : [0, 2] R with
 n
x 0x<1
fn (x) =
1 1 x 2.

Again we sketch the first few graphs of the sequence to gain some insight into
what the limit could be.

( ( (
x x [0, 1) x2 x [0, 1) x3 x [0, 1)
f1 (x) = f2 (x) = f3 (x) =
1 x [1, 2] 1 x [1, 2] 1 x [1, 2]

6 6 6
1 1 1

| |- | |- | |-
2 2 2

Since when 0 x < 1, fn (x) = xn and lim xn = 0 (Proposition 3.4.8) we


n
expect the limit to be zero on [0, 1).

Claim: 
0 x [0, 1)
f (x) =
1 x [1, 2]
is the pointwise limit.

To prove: x [0, 2], > 0, x N such that n > x , |fn (x) f (x)| < .
In fact as noted above we can use some previous results and so we wont need
to prove it using the definition.

Proof: Let x [1, 2] then fn (x) = 1 = f (x) for every n N and so lim fn (x) =
n
f (x).

In addition, we had from Proposition 3.4.8, that for 0 x < 1, lim xn = 0 =


n
f (x). So for every x [0, 2], lim fn (x) = f (x).
n

But this limit is not a uniform limit, notice that the graph of fn comes outside
5.1. UNIFORM AND POINTWISE CONVERGENCE 121

the -strip for every : 0 < 12 , since it has to get from 1


2n + to 1 .

For example with = 41 , let N,choose n = + 1, so n N and n > and


1
choose x = 41 n (0, 1). Then
1
|fn (x) f (x)| = |xn 0| = .
4
So > 0 such that N, n N, n > and x [0, 2] such that |fn (x)
f (x)| , so f is not the uniform limit of (fn )nN .

1
Now try a similiar example with fn (x) = 1+x n on the domain [0, 2]. You should

find that the pointwise limit is



1 x [0, 1)
f (x) = 1/2 x = 1
0 x (1, 2].

In the next example we examine a sequence of functions in B(R) that is we have


for every n N, fn : R R with fn bounded.
3. Consider
1

1 x n

fn (x) = nx n1 x n1

1 x n1

As usual we sketch the first few graphs of the sequence to understand a possible
pointwise limit.

f1 (x) f2 (x) f3 (x)


1 1

1 x 2 1 x 3

1 x 1
= x x [1, 1] = 2x x [ 21 , 12 ] = 3x x [ 13 , 31 ]
1 x1

1 x 12 1 x 13

1 1 1
6 6 6
 
| |- |  |- |  |-
-1 1 -1  1 -1  1
 
-1 -1 -1

Since n1 0 as n , we have for x < 0, fn (x) = 1 for large enough n


and similiarly since n1 0 as n , we have for x > 0, fn (x) = 1 for large
enough n, whilst fn (0) = 0 for all n.
122 CHAPTER 5. SEQUENCES IN B[A, B]

Claim: The function


1 x<0
f (x) = 0 x=0
1 x>0

is the pointwise limit.

Proof: Let x R and let > 0. First suppose x < 0 and choose = b x1 c + 1,
which is in N since x1 > 0. Let n N, n > then n > x1 ; i.e. x > n1 (since
x > 0), so x < n1 thus fn (x) = 1 = f (x). Thus |fn (x) f (x)| = 0 < .

Next suppose x = 0, then, as we observed above fn (x) = 0 = f (x) for every


n N.

Finally suppose x > 0, then choose = b x1 c + 1 N since x1 > 0. Let n


N, n > so n > x1 thus fn (x) = 1 = f (x). Hence in all cases we have N
such that n N,n > , |fn (x) f (x)| < , completing the proof of the claim.

Again this is not a uniform limit since the graphs of fn come outside any -strip
for < 1.

Consider for example = 41 . For N we can choose n = + 1 > and


1
x = 2(+1) then fn (x) = 21 and so

1 1 1
|fn (x) f (x)| = | + 1| = > = .
2 2 4
Thus > 0 such that N, n > and x R such that |fn (x) f (x)| 0.

5.2 Series-partial sums of sequences of functions


Given (gn )nN a sequence in B[a, b] we can define a sequence (fn )nN of partial
sums of the gn ; i.e. fn : [a, b] R with
n
X
fn = gi ; i.e. x [a, b], fn (x) = g1 (x) + g2 (x) + + gn (x).
i=1

As for sequences of partial sums of real numbers, we refer to the pair (g, f ) as
a series, where g is the sequence of terms and f is the sequence of partial sums
of the terms.

The next two examples will serve to demonstrate the claim in Proposition 4.2.3
(b) that the space D[a, b] is not complete. The first example is also an example
of a series whilst the second is a sequence. For example
5.2. SERIES-PARTIAL SUMS OF SEQUENCES OF FUNCTIONS 123

1. n N, let gn : [, ] R be the sequence of functions

cos (2n 1)Id


gn = ,
(2n 1)2

cos((2n1)x)
i.e. x [, ], gn (x) = (2n1)2 . And let fn : [, ] R be the sequence
defined by
n
4X
fn = gi ,
2 i=1

i.e. x [, ],
4
Pn
fn (x) = gi (x)
2


4
Pi=1
n cos((2i1)x)
=
2  i=1 (2i1)2 
cos(x)
=
2 4
1 + cos(3x)
32 + + cos((2n1)x)
(2n1)2

so (g, f ) is a series.

Claim: (fn )nN is a Cauchy sequence.

Remind yourself of the definition of a Cauchy sequence, although we wont use


it here, we will rather use the result domination by a Cauchy sequence. Recall
Theorem 3.7.9: If s : N V is a sequence in V and t : N W is a sequence
in W and m, n N, ||s(n) s(m)||V ||t(n) t(m)||W (we say sequence t
dominates sequence s), if t is a Cauchy sequence then s is a Cauchy sequence.
The proof was quite straightforward, check it.

Now to prove the claim we find a Cauchy sequence which dominates the given
sequence (fn )nN .
Proof of the claim: Consider the sequence tn given below. We have
 
tn = 4 1 + 312 + 512 + + (2n1)
1
2
 
< 4 1 + 212 + 312 + 412 + + (2n1)
1
2
R 2n1 1
< 4 (1 + 1 x2 dx)
4 1 2n1 4 1
= (1 + x |1 ) = (2 2n1 ).

8
Thus lim tn by the algebra of limits.
n

So tn is a convergent sequence and so by theorem 3.7.7 it is a Cauchy sequence.

To prove that the sequence tn dominates the sequence fn , let m, n N, m < n


124 CHAPTER 5. SEQUENCES IN B[A, B]

say. Then

 
(2m+1)id (2n1)id
||fn fm || = || 4 cos(2m+1) 2 + + cos(2n1) 2 ||
 
4 || cos(2m+1)
(2m+1)id||
2 + + || cos(2n1)
(2n1)id||
2

by the 4-inequality, homogeneity and the fact that


|(2n 1)2 | = (2n 1)2 

1
4 (2m+1) 2 + +
1
(2n1)2
= |tn tm |

and so by the theorem, domination by a Cauchy sequence, the sequence fn is a


Cauchy sequence.
We check that (fn )nN is in B[, ].

 
8 4 cos(x) cos(3x) cos((2n 1)x)
fn (x) = + + +
2 2 1 32 (2n 1)2 2

thus sequence (fn )nN is in B([, ]), a complete space and so (fn )nN con-
verges uniformly to some f B([, ]).

In fact using the definition of Fourier series, one can show that the f above is the
absolute value function.(See the appendix). Now it is easy to see that sequence
(fn )nN is a sequence of differentiable functions and so it is in D([, ]), but
the absolute value function is not in D([, ]), because it is not differentiable
at zero (Proof?), hence we have an example of the fact that D([, ]) is NOT
a complete space.

q
1
2. Consider the sequence (fn )nN given by fn (x) = x2 + n2 . Sequence
2 1 +
(fn )nN is in D[1, 1], since the term x + R and is differentiable on
n2

[1, 1], since is differentiable on R+ . (Recall is not differentiable at 0,
+
but 0 6 R , so does not cause a problem).
For n = 1 we have f1 (x) = x2 + 1, so f10 (x) = 21 22x 1 = 2 x 1 , and
(x +1) 2 (x +1) 2
f10 (0) = 0. In addition

1 x2 1
f100 (x) = 1 3 = 3 >0
(x2 + 1) 2 (x2 + 1) 2 (x2 + 1) 2

so the graph of
f1 (x) is concave up. At the endpoints of the interval [1, 1] we
have f1 (1) = 2. Thus the graph of f1 is
5.2. SERIES-PARTIAL SUMS OF SEQUENCES OF FUNCTIONS 125


6
2
1

| | -
-1 1

q
1 5
f2 (x) = x2 + 4 with graph, for c = 2 .

6
c
1
2
| | -
-1 1

q
1 10
f3 (x) = x2 + 9 with graph, for c = 3 .

6
c
1
3
| | -
-1 1

1 1

As n both n and n2 0 so fn (1) 1, fn (x) x2 = |x| and
fn (0) 0. Hence

claim: The uniform limit lim fn = | | := abs, the absolute value function.
n

To prove: > 0, N such that n N, n > we have ||fn abs|| < ,


i.e. x [1, 1], |fn (x) abs(x)| = |fn (x) x2 | < .
126 CHAPTER 5. SEQUENCES IN B[A, B]

Proof: First we prove that


1
||fn abs|| = sup {|fn (x) x2 |} = .
x[1,1] n

Let x [1, 1] then


r
1 1
|fn (x) x2 | 2 2
= x + 2 x

n n
q q
since x2 + n12 x2 + n12 . So n1 is an upper bound for

{|fn (x) x2 | : x [1, 1]} and ||fn abs|| is the least upper bound of the
same set, so
||fn abs|| n1 .

In addition 0 [1, 1] and fn (0) = n1 so n1 {|fn (x) x2 | : x [1, 1]}, thus
1
n = ||fn abs||.

Now let > 0 and choose = b 1 c + 1, so N. Let n > then n > 1 , so


> n1 = ||fn abs||. Hence lim fn = abs (uniform).
n
As in the previous example, abs is not differentiable at 0 [1, 1] and so
abs 6 D[1, 1], thus (fn )nN is a sequence in D[1, 1] which converges, but not
to a point (i.e. function) in D[1, 1].
Theorem 5.2.1 Suppose (fn )nN is a sequence of functions which are inte-
grable on [a, b] and that (fn )nN converges uniformly on [a, b] to a function f
which is integrable, then
Z b Z b Z b
f= lim fn = lim fn .
a a n n a
R
[So one can interchange and lim .]
n

u
Proof: Let > 0. Since (fn )nN f , N such that n N, n > we
have
|f (x) fn (x)| < , x [a, b].
So for n > we have
Rb Rb Rb R
| a f (x)dx a fn (x)dx| = | a (f (x) fn (x))dx| by linearity of
Rb
a |f (x) fn (x)|dx
R absolute value
property of
Rb
a dx = (b a).
Since the above statement is true for every > 0, we have
Rb Rb
a
f (x)dx = lim a fn (x)dx as required.
n
5.2. SERIES-PARTIAL SUMS OF SEQUENCES OF FUNCTIONS 127

Theorem 5.2.2 Suppose (fn )nN is a sequence of functions which are continu-
ous on [a, b] and suppose (fn )nN f uniformly on [a, b]. Then f is continuous
on [a, b].

Proof: For each x [a, b] we must prove that f is continuous at x. Here we will
deal only with x (a, b), since when x = a or b, we need to discuss one-sided
limits.
So let x (a, b). Let > 0. Since (fn )nN f uniformly on [a, b] there is some
n such that |f (y) fn (y)| < 3 , for every y [a, b]. In particular, for all h such
that x + h [a, b], we have

1. |f (x) fn (x)| < 3


2. |f (x + h) fn (x + h)| < 3

and since fn is continuous > 0 such that for |h| < we have

3. |fn (x) fn (x + h)| < .
3
Thus if |h| <

|f (x + h) f (x)| =
|f (x + h) fn (x + h) + fn (x + h) fn (x) + fn (x) f (x)|
|f (x + h) fn (x + h)| + |fn (x + h) fn (x)| + |fn (x) f (x)|
< 3 3 =

so x (a, b), > 0, > 0, such that


h with |x + h x| = |h| < we have |f (x +h) f (x)| < . Thus f is continuous
on (a, b).

Notice the contrapositive of theorem 5.2.2 is:


if f is not continuous on [a, b] and (fn )nN is a sequence of functions in C[a, b],
u
then (fn )nN 6 f.

The situation for a sequence of differentiable functions is comparatively disap-


pointing. If each fn is differentiable and if (fn )nN f uniformly, it is not
necessarily true that f is differentiable. We have already seen an example of
this.

We can use the fundamental theorem of calculus 1 to obtain a result but we


need the additional hypotheses that the sequence (fn0 )nN of derivatives of the
functions fn converges uniformly to a continuous function.

1 The fundamental theorem of calculus asserts that under certain hypotheses the operations

of differentiation and integration are inverse to each other. See Chapter 10 for a statement of
the theorem.
128 CHAPTER 5. SEQUENCES IN B[A, B]

Theorem 5.2.3 Suppose (fn )nN is a sequence of functions in D[a, b] and that
(fn )nN f pointwise. In addition, suppose (fn0 )nN converges uniformly
to a continuous function g. Then f is differentiable and lim fn0 (x) = f 0 (x)
n
(pointwise).
q
1
Notice that for our example with fn (x) = x2 + n2 we have

x
fn0 (x) = q .
1
x2 + n2

Thus  x
x 6= 0
as n , fn0 (x) |x|
0 x = 0.
0
so the function g which is the limit of sequence (fn (x))nN would be g(x) =
And
1 x>0
1 x < 0 which is not continuous at zero. In addition this convergence
0 x=0

cannot be uniform because fn0 is continuous and so for example, for every < 21 ,
its graph comes outside the -strip around g.

Proof of Theorem 5.2.3: Suppose (fn )nN is a sequence of functions in


D[a, b], thus fn is differentiable and therefore continuous for each n. Now con-
tinuous functions are integrable on closed intervals and so applying theorem
5.2.1 to the interval [a, x], for each x [a, b] we have
Z x Z x
g = lim fn0 = lim (fn (x) fn (a))
a n a n

by the fundamental theorem of calculus. So


Z x
g = lim fn (x) lim fn (a) = f (x) f (a),
a n n

since by hypothesis lim fn (x) = f (x) (pointwise). Now since g is assumed to


n
be continuous, it follows that

f 0 (x) = g(x) = lim fn0 (x), x [a, b].


n

We can apply these theorems to uniformly convergent series.


Pn
Corollary 5.2.4 Suppose ( i=1 fi )nN = (f1 + f2 + + fn )nN converges
Pn u
uniformly to f on [a, b]; i.e. ( i=1 fi )nN f

(1) If fn is continuous on [a, b] for each n, then f is continuous on [a, b].


5.2. SERIES-PARTIAL SUMS OF SEQUENCES OF FUNCTIONS 129

(2) If f and for each n, fn is integrable on [a, b], then


Z b Z
X b
f= fn .
a n=1 a

And
Pn p Pn
(3) if ( i=1 fi )nN f on [a, b] and i=1 fi0 converges uniformly on [a, b] to
some continuous function, then

X
f 0 (x) = fn0 (x), x [a, b], i.e. pointwise.
n=1

Pn so too is f1 + f2 + + fn and f is
Proof: (1) Since each fn is continuous
the uniform limit of the sequence ( i=1 fi )nN , so f is continuous by theorem
5.2.2. Pn
(2) Since ( i=1 fi )nN f uniformly it follows from theorem 5.2.1 that
Rb Rb
a
f = lim (f1 + f2 + + fn )
n Ra
b Rb Rb
= lim ( a f1 + a f2 + + a fn )
n R
P b
= n=1 ( a fn ).

Pn u
(3) Since ( i=1 fi0 )nN g and g is continuous so it follows from theorem 5.2.3
that

X
f 0 (x) = lim (f10 (x) + f20 (x) + + fn0 (x)) = fn0 (x).
n
n=1

In practice, unfortunately, thePcorollary is not very useful, as it is not generally


n
easy to see whether or not ( i=1 fi )nN converges uniformly. The following
theorem provides a sufficient condition.

Theorem 5.2.5 (The Weierstrass M-test)


Let (fn )nN be a sequence of functions defined on A R and suppose (Mn )nN
is a sequence of real numbers such that

(*) |fn (x)| Mn , for every x A and


P
(**) ( n=1 Mn )nN converges.
P P
Then x A, n=1 fn (x) converges absolutely, i.e. n=1 |fn (x)| converges
and

u
X
fn f on A,
n=1
P
where f (x) = n=1 fn (x), for all x A.
130 CHAPTER 5. SEQUENCES IN B[A, B]

Proof:
P By the comparison test applied to (*) for every x A P the series

n=1 |fn (x)| converges. So by the definition of absolute convergence, n=1 fn (x)
converges absolutely.

In addition, let > 0 and choose N satisfying



X
X P
Mi Mi < , i.e. i=+1 Mi < .



i=1 i=1

Now let n > and let x A, then


P Pn P
| i=1 fi (x) i=1 fi (x)| = | i=n+1 fi (x)|
P
i=n+1 |fi (x)| by the 4-inequality
P P
i=n+1 Mi < i=+1 Mi <
n n
P P P u P
so || fi fi || < , i.e. ( fi )nN fi = f.
i=1 i=1 i=1 i=1

The Weierstrass M -test is an ideal tool for analysing functions which are well-
behaved, for example power series.
Chapter 6

Power series

6.1 Definition and examples


P Pn
Definition 6.1.1 A function F (x) = n=0 fn (x) = lim i=1 fi (x), where
n
fi (x) = bi (x a)q(i) , a, bi and x R and q : N N a subsequence is a power
series centred on a.

Conventionally the summation in a power series is from n = 0, but one should


check that the sequence of terms is defined for n = 0.
Notice F D(R) C(R). For x R and fn (x) R, the set of values of x
for which the given power series can converge will be a real interval; i.e. a ball
BR (a) = (a R, a + R) possibly including one or both endpoints, where R will
be referred to as the radius of convergence.
When we replace the real numbers x, a, bi and fn (x) by complex numbers we
have a complex power series.

Theorem 6.1.2 The set of values of x for which the power series

X
X
fn (x) = bn (x a)q(n) ,
n=0 n=0

converges is BR (a) (possibly including endpoints.)

The strategy of proof is to use the fact that if the power series converges for y
then it must converge for all x satisfying |x a| < |y a|.

q(n)
Pn
Proof: If x = a, then n (x a)
Pnfor each n, bq(i) = 0, so i=1 bi (x a)q(i) = 0 for
every n thus lim
n i=1 bi (x a) = 0. Consequently a power series always
converges at its centre. We recall lemma 3.7.17:
forPany sequence (anP)0 in a normed vector space V
n
if n=0 an := lim
n i=0 ai = l V, then lim an = 0. [So if a series converges
n

131
132 CHAPTER 6. POWER SERIES

then the sequence of terms converges to zero. The converse is false. Think of
some examples.]

Now suppose the power series converges for y R, then by lemma 3.7.17, we
have
lim bn (y a)q(n) = 0 and so the sequence (bn (y a)q(n) )nN is a bounded
n
sequence i.e.
M R, n N : |bn (y a)q(n) | M.(*)
|xa| |xa|
Let x B|ya| (a) then |x a| < |y a|; i.e. |ya| < 1. Set = |ya| so
0 < 1 and

|bn (x a)q(n) | = |bn ||x a|q(n) by homogeneity


q(n)
= |bn | |xa|
|ya|q(n)
|y a|q(n)
= |bn |q(n) |y a|q(n) q(n) M by (*).

Now Pn
0 P| i=0 bi (x a)q(i) |
n
i=0 a)q(i) |P
Pn|bi (xq(i) n
M i=0 M i=0 i ,
the last of which is a geometric progression with common ratio 0 < < 1 and
so it converges.
Pn Thus by the comparison test,Pn
s(n) = i=0 |bi (x a)q(i) | converges and so i=0 (bi (x a)q(i) ) converges abso-
lutely and is thus convergent for all x B|ya| (a), where y is a point at which
the power series converges.

In addition, if we suppose that the subsequence q satisfies q(n + 1) q(n) is a


constant, say r, then, since the series converges for all x B|ya| (a), by the
ratio test for convergence of a sequence, we have

|bn+1 (xa)q(n+1) |
lim q(n) | <1
n |bn (xa)
q(n+1)
|x a|r lim |b|bn+1 n|
|
< 1 since (xa)
(xa)q(n)
= |x a|r
n
lim |b|bn+1
n|
|
< |xa| 1
r ,
n
 1/r
lim |b|bn+1n|
| 1
< |xa|
n
 1/r
lim |b|bn+1n|
| > |x a|
n

 1/r
so the sequence converges for all x BR (a) for R = |b|bn+1
n|
| . This gives a
formula for the radius of convergence. We can also derive a formula from the
root test for convergence. Since the series converges for all x B|ya| (a), by
the root test
p it must be that
lim sup q(n) |bn (x a)q(n) | < 1, i.e.
6.1. DEFINITION AND EXAMPLES 133
p 1
lim sup q(n)
|bn | < and so
|xa|
 p 1
|x a| < lim sup q(n) |bn | . So
1 1
R= p = lim sup p .
lim sup |bn | q(n) q(n)
|bn |
P
In particular, for power series n=0 bn (xa)n the radius of convergence satisfies
|bn | 1
(1) R = lim and (2) R = p .
n |bn+1 | lim sup n |bn |
Examples
P (x2)n
1. Consider the power series n=1 n! centred at 2 R. We have
|bn+1 | n! 1 1
= = and lim = 0.
|bn | (n + 1)! n+1 n n + 1

Since
1 |bn+1 |
= lim =0
R n |bn |
we say R = (which is not a real number) and we mean that the series
converges for every x R; i.e. the radius of convergence is and the
interval of convergence is R.
P
2. Consider the power series n=0 nn (x + 1)n centred at 1 R. To find R
we examine
|bn+1 | (n + 1)n+1 |bn+1 |
= n
> 1 and so lim
|bn | n n |bn |
p
does not exist.p[ Alternatively we could examine n
|bn | = n nn = n but
again lim sup n |bn | does not exist as N is not bounded above.]
In this case we say R = 0 and so the power series converges only at its
centre. We have that the radius of convergence is 0 and the interval of
convergence is the degenerate interval {1}.
P n
3. n=0 x is a power series centred at 0 R. We have |b|bn+1
n|
|
= 1 since bn = 1
for every n N. Hence
1 |bn+1 |
= lim = 1,
R n |bn |
thus the radius of convergence is 1 and the power series converges on
(1, 1).
It remains to examine
P convergence at the endpoints of this interval.
n
If x = 1 we have n=0 1P which is not bounded above and so does not
n
converge
P 2nand if Px= 1,n
n=0 (1) has a subsequence
P n
n=0 x = n=0 1 which is not finite and so n=0 (1) does not
converge.
Thus the interval of convergence is (1, 1).
134 CHAPTER 6. POWER SERIES

P (x1)n
4. Consider the power series n=1 n centred at 1 R. We have

|bn+1 | n 1
lim = lim = lim 1 =1
n |bn | n n + 1 n 1 +
n

by the algebra of limits. So R1 = 1 the radius of convergence is 1, thus the


series converges on B1 (1) = (0, 2).
P n
Now if x = 0 we have n=1 (1) n which we have seen is a Cauchy sequence
and
P 1 since R is complete it must converge in R. And if x = 2 we have
n=1 n (the harmonic series) which is not Cauchy and does not converge.
Thus the interval of convergence is [0, 2).
P (x+2)n
5. n=1 n2 is a power series centred at -2 in R.

|bn+1 | n2 1
lim = lim = lim =1
n |bn | n (n + 1)2 n 1 + 2 + 1
n n2

by the algebra of limits. So R1 = 1 the radius of convergence is 1, thus the


series converges on B1 (2) = (3, 1).
Now when x = 3 we have

X (1)n X (1)n
2
<
n=1
n n=1
n

and the series on the right converges, so by the comparison test


P (1)n
n=1 n2 converges.
P (1)n P 1
[Alternatively: n=1 n2 = n=1 n2 which converges by the integral
P (1)n
test, thus n=1 n2 is absolutely
P convergent and thus convergent.]
And when x = 1, we have n=1 n12 which we have seen converges (by the
integral test). Thus the given series converges with radius of convergence
1 and interval of convergence [3, 1].
P n
Notice that for a power series n=0 bn (x a) each term in the sum is a
polynomial in x and as such is differentiable with respect to x. Hence f 0 (x)
exists. Indeed

X
f 0 (x) = nbn (x a)n1 ,
n=1

where the summation starts at n = 1 since the n = 0 term of f (x) is a constant,


thus the n = 0 term of f 0 (x) is zero.
P
Lemma 6.1.3 If 0 r < 1, then n=1 nrn converges.

Proof: We have
(n + 1)rn+1
 
1
lim = lim r 1 + =r<1
n nrn n n
6.1. DEFINITION AND EXAMPLES 135

so the result follows by the ratio test.

Claim: f 0 (x) converges for all x BR (a).

Proof: Let x BR (a) then


|nbn (x a)n1 | = n|bn ||(x a)n1 |
n|bn ||(z a)n1 | where as in the proof of 6.1.2,
|x a| < |z a| < |y a| R
n|bn ||ya|n |za|n
= |za| |ya|n
n
nM |za|
|za| |ya|n where M |bn (y a)n |.
|za|n
Since |ya|n < 1 by lemma 6.1.3, the series

X nM |z a|n M X |z a|n
= n
n=1
|z a| |y a|n |z a| n=1 |y a|n
converges.

Now by the Weierstrass M -test since



X M X |z a|n
|nbn (x a)n1 | n ,
n=1
|z a| n=1 |y a|n

the
Platter series being convergent in R, we have that
n1
n=1 |nb n (x a) | converges on BR (a).

It is clear that we can iterate the previous to obtain a series



X
00
f (x) = n(n 1)bn (x a)n2
n=2

which converges on BR (a) and indeed for all k N,



X
f (k) (x) = n(n 1) . . . (n k + 1)bn (x a)nk
n=k

converges on BR (a), where f (k) is the kth derivative of f . Thus we have proved
Theorem 6.1.4 A power series which converges on an interval (a R, a + R)
is -ly differentiable on (a R, a + R) and for every k N

X
f (k) (x) = n(n 1) . . . (n k + 1)bn (x a)nk
n=k

converges for all x BR (a).


136 CHAPTER 6. POWER SERIES

Notice, in addition, that

f 0 (x) = b1 + 2b2 (x a) + 3b3 (x a)2 + . . .

so f 0 (a) = b1 . And

f 00 (x) = 2b2 + 3!b3 (x a) + 4!b4 (x 2)2 + . . .

so f 00 (a) = 2b2 . In general

f (k) (x) = k!bk (x a)kk + (k + 1)!bk+1 (x a) + . . . ,

thus
f (k) (a)
bk =
k!
,since n > k, (a a)nk = 0. You have already met a power series which has
coefficients of this form.

6.2 Taylor (resp Maclaurin) series


Definition 6.2.1 If there exists

X f (n) (a)
f (x) = (x a)n
n=0
n!

then the right hand side is referred to as the Tayor (resp. Maclaurin) series of
f at a 6= 0 (resp. a = 0).

So theorem 6.1.4 says that a convergent power series centred at a point a R


is always the Taylor series at a of the function which it defines.
Theorem 6.2.2 Let f : R R be n times differentiable then there exists
exactly one polynomial of degree at most n such that the polynomial and its first
n derivatives agree at a with f and its first n derivatives.

Proof: Consider the polynomial


Pn f (k) (a)
Tn f (x) := k=0 k! (x a)k
f (1) (a) f (2) (a) f (n) (a)
= f (a) + 1! (x a) + 2! (x a)2 + + n! (x a)n

(i.e. (Tn f )nN is the sequence of partial sums of the terms in the Taylor series
for f .)
We are to show Tn f (a) = f (a), Tn f 0 (a) = f 0 (a) and for all k,
Tn f (k) (a) = f (k) (a).

We have
Tn f (a) = f (a)
6.2. TAYLOR (RESP MACLAURIN) SERIES 137

i.e. the polynomial agrees with f at a.


Pn (k)
Tn f 0 (x) = k=1 kf k! (a) (x a)k1
2f (2) (a) nf (n) (a)
= f (1) (a) + 2! (x a) + + n! (x a)n1
so
Tn f 0 (a) = f 0 (a)
since (a a)k1 = 0 for every k > 1; i.e. the first derivative of Tn f agrees with
the first derivative of f at a.
Pn k.(k1)f (k) (a)
(Tn f )(2) (x) := k=2 k! (x a)k2
2.1f (2) (a) 3.2f (3) (a) n(n1)f (n) (a)
= 2! + 3! (x a) + + n! (x a)n2
so
(Tn f )(2) (a) = f (2) (a)
so that the second derivatives agree at a.

In general
n
X k.(k 1) . . . (k m + 1)f (k) (a)
(Tn f )(m) (x) = (x a)km
k!
k=m
so
(Tn f )(m) (a) = f (m) (a)
so that the m th derivatives agree at a.

Thus Tn f is a polynomial satisfying the requirements of the theorem. It remains


to see that Tn f is unique.
Suppose there is a polynomial g of degree n which also satisfies g (k) (a) =
f (k) (a) = Tn f (k) (a), 0 k n.

Let g(x) = a0 + a1 (x a) + + an (x a)n so we have g(a) = f (a) = a0 ;


g 0 (x) = a1 + 2a2 (x a) + + nan1 (x a)n1 so we have g 0 (a) = f 0 (a) = a1 ;
g 00 (x) = 2a2 + 3.2.a3 (x a) + + n(n 1)an (x a)n2 so we have g 00 (a) =
f 00 (a) = 2a2 .
(2)
f (3) (a) f (k) (a)
Thus a0 = f (a), a1 = f (1) (a), a2 = f 2!(a) ,a3 = 3! and ak = k! . Hence
the polynomial g is the polynomial Tn f .

The sequence (Tn f )nN is a sequence of polynomials of increasing degree which


approximates the function f at a.
Definition 6.2.3 Define a function En f : R R by
En f = f Tn f ; i.e. x R, En f (x) = f (x) Tn f (x).
We refer to En f as the error (or remainder) function.
138 CHAPTER 6. POWER SERIES

The function En f measures the difference between


n
X f (k) (a) X f (n) (a)
(x a)k and (x a)n := f (x).
k! n=0
n!
k=0

Theorem 6.2.4 (Taylors Theorem)


If f : I R R is (n + 1) times differentiable on I, a I and if Mn R
such that
|f (n+1) (x)| Mn , x I then
Mn
(*) |f (x) Tn f (x)| (n+1)! |x a|n+1 ,
Mn
i.e. |En f (x)| (n+1)! |x a|n+1 .

The inequality (*) says


Mn Mn
Tn f (x) |x a|n+1 f (x) Tn f (x) + |x a|n+1 .
(n + 1)! (n + 1)!

Proof: First we prove by induction that


Z x
(x t)n
En f (x) = f (n+1) (t) dt.
a n!
Suppose a < x,
E0 f (x) = f (x) T0 f (x)
P0 (k)
= f (x) k=0 f k!(a) (x a)k
= fR (x) f (a)
x
= a f 0 (t)dt by the FTC, since f 0 is continuous (A)
E1 f (x) = f (x) T1 f (x)
P1 (k)
= f (x) k=0 f k!(a) (x a)k
= fR (x) f (a) f 0 (a)(x a)
x
= a f 0 (t)dt f 0 (a)(x a) by (A).
Now,
R x 0 since f (2)
R xis continuous
a
f (t)dt = a f 0 (t).1dt so with u(t) = f 0 (t), v(t) = t x
Rx
= f 0 (t)(t x)|xa a f (2) (t)(t x)dt
Rx
= f 0 (a)(a x) + a f (2) (t)(x t)dt
Rx
= f 0 (a)(x a) + a f (2) (t)(x t)dt
Rx
so E1 f (x) = a f (2) (t)(x t)dt.
Let k N, k > 1 and suppose
Z x
(x t)k
Ek f (x) = f (k+1) (t) dt the inductive hypothesis. Now
a k!
6.2. TAYLOR (RESP MACLAURIN) SERIES 139

Ek+1 f (x) = f (x) Tk+1 f (x)


f (k+1) (a)
= f (x) Tk f (x) (k+1)! (x a)k+1
(k+1)
f (a)
= Ek f (x) (k+1)! (x a)k+1 .
So, by the inductive hypothesis
Z x
(x t)k f (k+1) (a)
Ek+1 f (x) = f (k+1) (t) dt (x a)k+1 . Now
a k! (k + 1)!
R x (k+1) (xt)k k+1

a
f (t) k! dt with u(t) = f (k+1) (t), v(t) = (xt) (k+1)!
 k+1
 x R
x (xt)k+1
= f (k+1) (t). (xt) + a f (k+2) (t) (k+1)! dt

(k+1)!
a
k+1 R x (k+2) (xt)k+1
= f (k+1) (a). (xa)
(k+1)! + a f (t) (k+1)! dt so
R x (k+2) (xt)k+1
Ek+1 f (x) = a f (t) (k+1)! dt as required.
So by the principal of induction, if f (n+1) is continuous on [a, x]
Rx n
En f (x) = a f (n+1) (t) (xt)n! dt. Thus
R n

x
|En f (x)| = a f (n+1) (t) (xt) dt

n!
R x (n+1) (xt)n
a f (t) n! dt
R x Mn
a n! (x t)n dt by assumption |f (n+1) (t)| Mn

Rx
= Mn!n a |(x t)n |dt
n+1 x
 
= Mn!n |(xt)n+1
|

 n+1
 a
= Mn!n |(xa)n+1
| Mn
= (n+1)! |(x a)|n+1

as required.

With x < a we just change the order of endpoints of integration to arrive at the
same result.

The theorem is useful in finding the number of terms to be taken so that the
error is bounded in other words the approximation of f by Tn f is sufficiently
accurate.

For example with f = cos, a = 0 and I any interval containing 0, using Taylors
theorem, we could choose Mn = 1, since

| cos(x)|
| cos(n+1) (x)| = 1
| sin(x)|
on the whole of R. So the theorem gives
|x|n+1
|En f (x)| 1. .
(n + 1)!
140 CHAPTER 6. POWER SERIES

For x = 0.1 suppose we want to find the number of terms n we should take so
that |En f (x)| 106 ; i.e. the error in the approximation is 0.000001. We
n+1
want (0.1)
(n+1)! 10
6
and so by trial and error using small values of n we find

(0.1)4 (0.1)5
0.000004 > 106 and 0.0000008 < 106 .
4! 5!
So we need n 4 (i.e. n + 1 5) and we will have that
cos(0.1) T4 cos(0.1) 106 .

6.3 Approximation by power series

We can also make use of power series to derive series approximations to known
functions other than the Taylor series. For example we derive a series approxi-
mation centred on 0 for arctan (the inverse of tan).

We know from the algebra of continuous functions in the first part of math 333,
1
that the function f (t) = 1+t 2 is continuous on an interval containing 0. We

have
2t >0 t<0
f 0 (t) = =0 t=0 and
(1 + t2 )2
<0 t>0
2
2 8t 2+6t
f (2) (t) = (1+t2 )2 + (1+t2 )3 = (1+t2 )3

1 1
<0 3 <t< 3


= 0 t = 13 .
> 0 t > 1 and t < 1

3 3
The graph of f is

6
1

| 1 |1 -
3
3
6.3. APPROXIMATION BY POWER SERIES 141

We also know that arctan0 (t) = 1


1+t2 = f (t) so
Z x
1
arctan(x) = dt by the FTC.
0 1 + t2
Recall that for every p R andn |x| < 1 we have

p(p 1) 2
(1 + x)p = 1 + px + x + ...
2!
so for |t2 | = t2 < 1 we have
1.2 2 2 (1)(2)...(n) 2n
(1 + t2 )1 = 1 + (1)t2 + 2! (t ) + + n! t + ...
2 4 6 n 2n
= 1 t + t t + + (1) t + ...

So integrating term by term we have


Rx 1 Rx
0 1+t2
dt = 0 (1 t2 + t4 + (1)n t2n + . . . )dt
(1)n+1 2n+1
= x 13 x3 + 15 x5 + + 2n+1 x + ...

(1)n+2 n+1
Now bn+1 = 2n+1 and bn = (1)
2n1 so

bn+1 2n 1 1
bn = 2n + 1 = 1 n + 1/2


so lim bn+1
bn = 1 by the algebra of limits, thus the radius of convergence is 1.

n
(Notice this is compatible with the condition t2 < 1.)

P n+1
At the endpoints for x = 1 we have n=1 (1)2n1 . Consider the sequence of
partial sums
n n
X (1)r+1 X
s(n) = = u(r).
r=1
2r 1 r=1

We claim that s(n) is Cauchy. The proof is very similiar to our proof in math
Pn r
333 that r=1 (1)
r is Cauchy, look at cases n > m with n, m even, n, m odd
and n odd and m even. Otherwise we can use the alternating sequences test.
The hypotheses of the alternating sequences test were:
1 1
1. lim u(r) = 0 since for r even we have u(r) = 2r1 = 12r 0 as r
r
1
and for r odd, u(r) = 2r1 0 as r .
2. Clearly u is alternating and
1 1 1
3. |u(r + 1)| = 2(r+1)1 = 2r+1 < 2r1 = |u(r)| so the sequence of absolute
values of the terms is decreasing.
142 CHAPTER 6. POWER SERIES

In addition, when x = 1 we have



X (1)n+1 (1)2n1 X (1)n
=
n=1
2n 1 n=1
2n 1

which also satisfies the alternating sequences test. Check the details.
Chapter 7

Contraction maps

Our final application of convergence examines a technique of producing a se-


quence of approximations to the solution of a given equation and we give a
theorem which states sufficient conditions for the method to work.

7.1 Definitions and examples


Definition 7.1.1 Let S V be a normed vector space. A map g : S S is a
contraction map k < 1 such that x, y S, ||g(x) g(y)||V k||x y||V .

We refer to the right hand side of as the contraction condition.


We refer to k as the contraction constant.

Examples:
1. Let g : [0, 1/2] R be defined by g(x) = 21 (1 x3 ).
Claim: g is a contraction map.
To prove: g([0, 1/2]) [0, 1/2] and k < 1 such that x, y [0, 1/2],
|g(x) g(y)| k|x y|.
Proof: g 0 (x) = 23 x2 < 0 on [0, 1/2] so g is decreasing from g(0) = 21 to
g( 12 ) = 16
7
thus g([0, 1/2]) [0, 1/2].

Let x, y [0, 1/2] then

|g(x) g(y)| = | 12 (1 x3 ) 21 (1 y 3 )|
= 12 |x3 y 3 |
= 12 |x y||x2 + xy + y 2 |
83 |x y| since 0 x2 , xy, y 2 1
4

3
so we can take k = 8 < 1.

143
144 CHAPTER 7. CONTRACTION MAPS

2. In this example we will see that it is not sufficient to check the contrac-
tion condition alone, we must also check that the image of the domain is
contained in the domain. Consider g : [0, 1/2] R with g(x) = 2 x3
then g satisfies the contraction condition since following the argument
in example 1, one could choose k = 83 < 1, and we have x, y R,
|g(x) g(y)| 83 |x y|.
But g is not a contraction map since for example g(0) = 2 6 [0, 1/2].
Proposition 7.1.2 Let S V a normed vector space. Let g : S S and let
a S. If g is a contraction map then g is continuous at a.

Proof: Suppose g : S S is a contraction map and let k < 1 satisfy the


contraction condition; i.e. x, y S, ||g(x) g(y)||V k||x y||V .

Let B = B (g(a)) be an open ball centred on g(a) in the codomain. Then


choose A = B (a) as the ball centred on a in the domain. Then
g(A) = {g(x) : x A} = {g(x) : ||x a||V < }
{g(x) : ||g(x) g(a)||V < }
since ||g(x) g(a)||V k||x a||V < ||x a||V <
= B (g(a)) = B.

7.2 Contraction map theorem


Theorem 7.2.1 Let S be a complete subset of a normed vector space and let
x1 S. If g is a contraction map then the sequence of iterates of x1 under g
converges to an element x S satisfying x = g(x). There is no other element
of S which satisfies this equation.

Since x1 S and g is a contraction map, g(x1 ) = x2 S and g(x2 ) = x3 S.


The sequence (xn )nN is referred to as the sequence of iterates of x1 under g.
One can also express the conclusion by saying that g has a unique fixed point
in S.

To use the theorem we need to be able to prove that a given map is a contraction
map and we need to be able to tell whether a subset of a normed vector space
is complete.
Theorem 7.2.2 If S is a closed ball in a complete normed vector space then S
is itself complete; i.e. every Cauchy sequence in S converges to a point in S.
The theorem gives for example
1. since R is a complete normed vector space and any closed interval [a, b] in
R is a closed ball Br ( a+b ba
2 ), where r = 2 , then the closed interval [a, b]
is complete.
7.2. CONTRACTION MAP THEOREM 145

2. In R2 , any closed ball Br (x) is complete.

Theorem 7.2.2 is easy to prove once we have done some more topology and
learned about limit points. The result is that every closed set contains all of its
limit points.

Application of theorem 7.2.2: In example 1 we had g : [0, 1/2] R with


g(x) = 21 (1 x3 ) is a contraction map and since [0, 1/2] is a closed interval
in R it is complete, so the hypotheses of theorem 7.2.1 are satisfied. Thus the
conclusion follows; it says !x [0, 1/2] such that x = lim xn , where for every
n
n N, xn = g(xn1 ) and x is a fixed point of g; i.e. x = 12 (1 x3 ).
Furthermore this means !x [0, 1/2] which is a solution to the equation x3 +
2x 1 = 0. The importance of this result is that there is no rational solution
to the equation and so no computational method can produce a solution. The
theorem guarantees that a solution exists.

Notice that by the rational roots test,


a rational solution pq (p Z, q N) p divides -1 and q divides 1 p = 1
and q = 1 so pq = 1 neither of which are a solution to the given equation. So
there are no rational solutions.

Proof of theorem 7.2.1: Let g : S S be a contraction map with contraction


constant k < 1.
By construction of the sequence (xn )nN we have xn = g(xn1 ), n 2 and
xn+1 = g(xn ) so
xn+1 xn = g(xn ) g(xn1 ) thus
||g(xn ) g(xn1 )|| = ||xn+1 xn ||
k||xn xn1 ||
k 2 ||xn1 xn2 ||
..
.
k n1 ||x2 x1 ||.
To prove: (xn )nN is a Cauchy sequence.
Proof: Let m < n and note that

(1)
||xn xm || = ||xn xn1 + xn1 + xm+2 xm+1 + xm+1 xm ||
||xn xn1 || + ||xn1 xn2 || + + ||xm+2 xm+1 || + ||xm+1 xm ||
k n2 ||x2 x1 || + k n3 ||x2 x1 || + . . . k m ||x2 x1 || + k m1 ||x2 x1 ||
= (k n2 + k n3 + + k m + k m1 )||x2 x1 ||.
Pn2 P2
Now let tn = i=1 k i ||x2 x1 || = ||x2 x1 || i=1 k i , which is a constant times
a geometric progression with common ratio k < 1. In addition

|tn tm | = (k n2 + + k m1 )||x2 x1 ||.


146 CHAPTER 7. CONTRACTION MAPS

So from (1) we have

||xm xn || |t(n) t(m)| i.e.

sequence t dominates sequence (xn )nN .


Now (tn )nN converges and so it is a Cauchy sequence, thus by the theorem
domination by a Cauchy sequence, (xn )nN is a Cauchy sequence. In addition
it is a Cauchy sequence in S, a complete space and so (xn )nN converges to a
point x say, in S.

Next we want to see that the point x = lim xn is a fixed point of the map g.
n
We have xn = g(xn1 ), n 2 so

x = lim xn = lim g(xn1 )


n n
= g( lim xn1 ) since g is continuous
n
= g(x) by shifting sequences lim xn1 = lim xn .
n n

Thus x is a fixed point of g.

Finally we have to see that x is unique.


Let x, y S and suppose both are fixed points of g; i.e. g(x) = x and g(y) = y.
So ||g(x) g(y)|| = ||x y|| and so x = y since otherwise we contradict the
assumption that g is a contraction map (we have k = 1, rather than k < 1).

The theorem is useful to establish the existence of fixed points and to estab-
lish the existence of solutions to various polynomial, rational and differential
equations.

Examples

1. Suppose we want to establish that a solution to

1
x4 x3 + 3x =0
2

exists in the interval [0, 41 ].

Step (i) express the equation in the form x = g(x).


We have
1
x4 x3 + 3x 2 = 0 3x = 12 + x3 x4 
x = 31 12 + x3 x4 .

There is no unique way of performing this operation the important


requirement is that the expression on the right hand side defines a
contraction map.
7.2. CONTRACTION MAP THEOREM 147

Step (ii) Set g(x) = 13 12 + x3 x4 . We are to check that g is a con-




traction map on [0, 41 ]. If this step fails then we redo step (i). Now

x2
g 0 (x) = 13 (3x2 4x3 ) = 3 (3 4x)
so for
0 x 14
0 4x 1
1 4x 0
2 3 4x 3

so g 0 (x) > 0 on [0, 1/4]; i.e. g is monotonic increasing from

g(0) = 61 to
g( 41 ) = 13 21 + 1 1 131
< 14 .

64 256 = 768

So g([0, 1/4]) [0, 1/4]. In addition, let x, y [0, 1/4] then

||g(x) g(y)|| = | 13 12 + x3 x4 13 21 + y 3 y 4 |
 

= 31 |x3 y 3 (x4 y 4 )|
31 |x y||x2 + xy + y 2 (x3 + xy 2 + yx2 + y 3 )|.

Now
1 1
0 x2 , xy, y 2 and 0 x3 , xy 2 , yx2 , y 3
16 64
so
1 3 1 1
|g(x) g(y)| |x y| + = |x y|
3 16 16 12
so g is a contraction map on [0, 1/4].
Step (iii) We check that the domain of g is a complete space. But [0, 1/4]
is a closed interval and so by the observation following theorem 7.2.2,
we have that the domain is complete. Thus g satisfies the hypotheses
of the contraction map theorem and so the conclusion follows.
Step (iv) We interpret the conclusion. !x [0, 1/4] such that
 
1 1
x = g(x) = + x3 x4
3 2

i.e. x is a solution to the original equation


1
x4 x3 + 3x = 0 in the interval [0, 1/4].
2

Again there are no rational solutions to this equation.

2. It can be proved that the rather innocent looking differential equation

40 = 2 + Id2
148 CHAPTER 7. CONTRACTION MAPS

cannot be solved by known means. The contraction map theorem can be


applied to establish that a solution exists. First we give a result which
shows how the norm of an indefinite integral relates to the norm of f . To
follow the proof, lets be clear about the difference between functions f
and |f | and real number ||f ||.

f |f |
6 6
6 ||f ||
.

?
| |- | |-
a b a b

Lemma 7.2.3 Let f C[a, b] and let c [a, b] then


Z
|| f || (b a)||f ||;
c

i.e. the area under the graph of f from c is less than or equal to the area of the
rectangle with side lengths b a and ||f ||.

Proof: By definition
Z Z t
|| f || := sup {| f |}.
c t[a,b] c

For t c,
Z t Z t Z b
| f| |f | |f | (b a)||f ||
c c a

since ||f || := supx[a,b] {|f (x)|} and for t < c


Z t Z c Z b
| f| = | f| |f | (b a)||f ||.
c t a
7.2. CONTRACTION MAP THEOREM 149

Rt R
Hence (b a)||f || is an upper bound of {| c f | : t [a, b]} and || c f || is the
least upper bound of the same set so
Z
|| f || (b a)||f || as required.
c

We need to be able to write the differential equation in the form = G() for
some function G.

For example, consider the elementary differential equation


0 = 0, (0) = 1
which is solvable by elementary means, for example the integrating factor method.
It is clear that = exp is a solution to the equation, since 0 = exp and
(0) = exp(0) = 1. To use the contraction map theorem, we want to write the
equation in the form = G().
We have Z
G() = (0) +
0
since then for = G() differentiating each side gives
0 = 0 + ( 0 )0 =
R

i.e. 0 = 0.
Next we iterate using the function G, since (0) = 1 we take 1 = 1 then
R
2 = G(1 ) = 1 (0) + 0 1
R
= 1 + 0 1 = 1 + Id
R
3 = G(2 ) = 2 (0) + 0 2
= 1 + 0 (1 + Id) = 1 + Id + 12 Id2
R
R
4 = G(3 ) = 3 (0) + 0 3
= 1 + 0 (1 + Id + 21 Id2 )
R

= 1 + Id + 21 Id2 + 1 3
3.2 Id . In general
n = G(n1 )
1 2 1 3 1 n1
= 1 + Id + 2! Id + 3! Id + + (n1)! Id .
And so we have lim n = exp on any interval containing 0. So the sequence of
n
iterates of 1 under G converges to the solution = exp and satisfies (0) = 1.
But as we noted earlier, it is not always possible to find an explicit solution to
a differential equation and so to prove that a solution exists we appeal to the
contraction map theorem.

Suppose C[1/2, 1/2], we want to find out whether or not a solution to the
equation
1
0 = (2 + Id2 ), (0) = 0
4
150 CHAPTER 7. CONTRACTION MAPS

exists. First we write the equation as an integral equation; i.e. an equation of


the form = G().
Claim: The given differential equation and initial condition is equivalent to the
single integral equation
Z
1
= (2 + Id2 ).
4 0
Z
1 1
To prove: 0 = (2 + Id2 ), (0) = 0 = (2 + Id2 ).
4 4 0

Proof: Suppose (0) = 0 and 0 = 14 (2 + Id2 ), so by the algebra of continuous


functions 0 C[1/2, 1/2], since , Id C[1/2, 1/2]. We have 0 [1/2, 1/2]
and so by the FTC (i)
Rx
0 )(x) 0 (t)dt = (x) (0) = (x)
R
( 0
=
R0x
i.e. (x) = 0 4
1
(2 + Id2 )(t)dt
x
1
+ Id2 )(t)dt = 1
2 + Id2 (x)
R 2
R 
= 4 0 ( 4 0

1
(2 + Id2 ) thus
R
so = 4 0
Z
1 2 1
0 = ( + Id2 ), (0) = 0 = (2 + Id2 ).
4 4 0

1
(2 + Id2 ). Then
R
Conversely, suppose = 4 0

0
0 1
(2 + Id2 )
R
= 4 0
= 14 ( 0 (2 + Id2 ))0
R

= 41 (2 + Id2 ) by the FTC (i)


since 2 + Id2 is continuous
R0
In addition (0) = 14 0 (2 + Id2 ) = 0.
Z
1 1 2
So = (2 + Id2 ) 0 = ( + Id2 ), (0) = 0.
4 0 4

So set G() = 41 0 (2 + Id2 ) for C[1/2, 1/2] which is a complete normed


R

vector space.
Next we show that G satisfies the hypotheses of the contraction map theorem
on the closed ball
 
1
B1/2 (0) = f C[1/2, 1/2] : ||f 0|| C[1/2, 1/2],
2

which is a closed subset of a complete space and so it is complete.


To prove: G maps B1/2 (0) into itself.
7.2. CONTRACTION MAP THEOREM 151

1
Proof: Suppose B1/2 (0), so |||| 2 then

= || 14 (2 + Id2 )||
R
||G()||
R0
= 14 || 0
(2 + Id2 )|| by homogeneity
14 || + Id2 || since we are integrating
2

over [1/2, 1/2] and so in lemma 7.2.3 we have


(b a) = 1
14 (||2 || + ||Id2 ||) by the triangle inequality
14 ( 14 + 14 ) since |||| 1/2,
||Id|| = sup[1/2,1/2] {x2 } = 41
= 18 12

so G() B1/2 (0).

To prove: G satisfies the contraction condition.


Proof: Let , B1/2 (0) C[1/2, 1/2] then

||G() G()|| = || 14 R0 (2 + Id2 ) R41 0 ( 2 + Id2 )||


R R

= 14 || R0 (2 + Id2 ) 0 ( 2 + Id2 )||R


= 41 || 0 (2 2 )||by linearity of 0
14 ||2 2 )|| by lemma 7.2.3 and b a = 1
41 || |||| + ||
14 || ||(|||| + ||||)
41 || || since ||||, |||| 1/2

Hence G is a contraction map.

The conclusion of the contraction map theorem thus asserts the existence of a
unique point B1/2 (0) such that
Z
1
= G() = (2 + Id2 ).
4 0

So by the first claim is a solution to the initial value problem


0 = 41 (2 + Id2 ), (0) = 0.
152 CHAPTER 7. CONTRACTION MAPS
Part IV

Real Analysis

153
Chapter 8

Limits

We call this part real analysis because we study the theoretical background for
the rules, methods and results used in elementary calculus to differentiate and
integrate functions f : S R R. We will study and prove for example the
fundamental theorem of calculus which gives conditions under which we can find
an antiderivative to integrate certain functions; i.e. the conditions on function
f such that differentiation and integration are related.

8.1 Real-valued functions on S R


In Analysis I, we learned the definitions of limit and continuity respectively, for
functions f : V W , where V and W were arbitrary normed vector spaces. So
in this section we interpret the definitions in the case where V = R = W .

We said a function f : V W has a limit at a V l W such that B,


A with f (A) B, where B is an open ball centred on l and A is an open
decentred ball, centred on a. Recall that we decentred the ball because we are
only interested in what happens as x approaches a and not in what happens at
a.

And we said a finction f : V W is continuous at a V B, A with


f (A) B, where B is an open ball centred on f (a) and A is an open ball
centred on a.

For V = R the open ball A centred on a is an open interval (a , a + ), for


some > 0 (the radius of the ball A). The decentred ball centred on A is the
union of the two open intervals we obtain from (a , a + ) by removing a; i.e.
(a , a) (a, a + ). And for W = R the open ball centred on l (resp. f (a)) is
the open interval (l , l + ) (resp. (f (a) , f (a) + )), for some > 0, the
radius of the ball B.

155
156 CHAPTER 8. LIMITS

We assume in the case of functions whose domain S is not the whole of R that
S {a} is an open set.

Definition 8.1.1 A function f : S R R has a limit l at a R


> 0, > 0 such that x 6= a, x S, if x (a , a) (a, a + ) then
f (x) (l , l + ).

And for continuity

Definition 8.1.2 a function f : S R R is continuous at a S


> 0, > 0 such that x S, if x (a , a + ) then
f (x) (f (a) , f (a) + ).

8.2 Order relation


R is an ordered field, so it has a lot more algebraic structure than a vector space.
(Recall that R is a vector space over the field R). The symbol < commonly
denotes an order relation.

Definition 8.2.1 An order relation on a set S satisfies

O1 (comparability) if x 6= y then x < y or y < x,

O2 (nonreflexivity) if x < y then x 6= y and if x > y then x 6= y and

O3 (transitivity) if x < y and y < z then x < z.

O2 is the converse of O1. An equivalent expression for O2 is


O20 For no x S does x < x hold.

One can check that the natural order in R is an order relation.

8.3 Right and left limits


An immediate consequence of the presence of an order relation is the notion of
right and left limits.

Definition 8.3.1 f : S R has a right limit l at a R


> 0, > 0 such that if x (a, a + ) then f (x) (l , l + ).

and

Definition 8.3.2 f : S R has a left limit l at a R


> 0, > 0 such that if x (a , a) then f (x) (l , l + ).

In a similiar way one can define the idea of right and left continuity at a point.
u
Recall in the proof of 5.2.2 the hypothesis was (fn )nN f and fn C[a, b]
for each n N and we proved that f C(a, b), but we wanted to prove that
8.3. RIGHT AND LEFT LIMITS 157

f C[a, b]. We can use the definition of left continuity at a and right continuity
at b to complete the proof.

For example, if f : R R is defined by



1 x<0
f (x) = 0 x=0
1 x>0

then f has a left limit of -1 at 0 R.


Proof: Let > 0 and choose = and let x (, 0). Then f (x) = 1 so
f (x) (1 , 1 + ).

And f has a right limit of 1 at 0 R.


Proof: Let > 0 and choose = and let x (0, ). Then f (x) = 1 so
f (x) (1 , 1 + ).
[Notice that in both cases one can make any choice of .]

Clearly f is not continuous at 0.


[We are to prove > 0 such that > 0, x (, ) such that f (x) 6
(f (0) , f (0) + ) = (, ).]
Proof: Consider = 12 . Let > 0 and let x = 2 , so x (, ). But since
> 0, f (x) = 1 and 1 6 ( 12 , 21 ).

As you know from elementary calculus the derivative of a function f : R R


at a point a R is also a limit. The derivative is the gradient of a tangent to
the graph of the function at the point (a, f (a)) i.e.

6
-slope of the line= lim f (a+)f

(a)
 0



| | -
a a+

Definition 8.3.3 Let f : R R and let a R. Let


f (x) f (a)
g(x) = , x R, x 6= a
xa
then
f is differentiable at a lim g(x) exists;
xa
158 CHAPTER 8. LIMITS

i.e. if l R such that B, A such that g(A) B, where B is a ball centred


on l and A is a decentred ball centred on a, then l = f 0 (a).

Equivalently with = |x a| we
have
lim f (a+)f

(a)
exists, when x>a
0
f is differentiable at a R f (a)f (a)
lim exists, when x < a.
0

We leave the algebra of derivatives to advanced calculus (Math 421) where you
will study the algebra of derivatives in a more general setting. Instead we will
turn to integration.
Chapter 9

The Riemann Integral

9.1 Definitions
We used the idea of a partition of a closed interval to define a step function.
Recall that:

Definition 9.1.1 a partition of a closed interval [a, b] is a finite set of points


P = {a0 , a1 , . . . , an } with a = a0 < a1 < a2 < < an = b. The points ai are
the points of the partition and the intervals (a0 , a1 ), (a1 , a2 ), . . . , (an1 , an ) are
the open subintervals of the partition.

Now given a function f : S R R where [a, b] S, we choose a partition


of [a, b] and construct rectangles with the open subintervals of the partition as
base such that the rectangles enclose the set of points
{(x, y) : a x b, 0 y f (x)}.
The sum of the areas of such a collection of rectangles is called an upper sum
for f on [a, b].

-
a b
Lower sums are defined analogously. This time we require that the rectangles
lie inside the set of points {(x, y) : a x b, 0 y f (x)}.

159
160 CHAPTER 9. THE RIEMANN INTEGRAL

The sum of the areas of such a collection of rectangles is called a lower sum for
f on [a, b].

-
a b

Definition 9.1.2 A function f is said to be Riemann integrable on an interval


[a, b] contained in its domain if the following condition holds: For each  > 0,
there is an upper sum U and a lower sum L such that U L < .

If f is Riemann integrable on [a, b] then it can be shown that there is a unique


number A satisfying

( upper sum U )( lower sum L) L A U.

This number A is then called the Riemann integral of f on [a, b] and is denoted
Rb
a
f . Next well see how to write A as the limit of a sequence.

Let f be a function from [a, b] into R.


Let P = {x0 , x1 , . . . , xn } be a partition of [a, b].
If f is bounded then we can define real numbers MiP and mP i as
sup{f (x) : x [xi1 , xi ]} and inf{f (x) : x [xi1 , xi ]} respectively. Then
n
MiP (xi xi1 )
P
define the upper sum S(P ) of f for the partition P by S(P ) =
i=1
n
mP
P
and the lower sum s(P ) of f for the partition P by s(P ) = i (xi xi1 ).
i=1

Theorem 9.1.3 If P is any partition of [a, b] and


m = inf{f (x) : x [a, b]} = inf{mP
i : i {1, . . . , n}} and
M = sup{f (x) : x [a, b]} = sup{MiP : i {1, . . . , n}} then m(b a)
s(P ) S(P ) M (b a).
9.1. DEFINITIONS 161

Proof: For each i: m mP P P P


i Mi M by definition of m, mi , Mi and M .
Thus
n
P n
P
m(b a) = m (xi xi1 ) = m(xi xi1 )
i=1 i=1
n
mP
P
i (xi xi1 )
i=1
n
MiP (xi xi1 )
P

i=1
Pn n
P
M (xi xi1 ) = M (xi xi1 ) = M (b a).
i=1 i=1

Definition 9.1.4 A partition P 0 is said to be finer than P if every point of P


is also in P 0 ; i.e. P P 0 .
The partition P = {1, 1, 2, 3} P 0 = {1, 0, 1, 2, 3}, so P 0 is finer than P; P 0
is said to refine P.
Lemma 9.1.5 If P2 refines P1 then S(P2 ) S(P1 ) and s(P2 ) s(P1 ).
(i.e. a refinement can neither increase an upper sum, nor decrease a lower sum.)
Proof: It suffices to consider the case where P2 contains one point more than P1 .
Suppose P1 = {x0 , x1 , . . . , xn }. Suppose that y [xj1 , xj ] and P2 = P1 {y}.
Let MyP12 = sup{f (x)| x [xi1 , y]} and MyP22 = sup{f (x)| x [y, xi ]}. Then
P P1
S(P1 ) S(P2 ) = Mi (xi xi1 ) + MjP1 (xj xj1 )
i6=j
P P1
Mi (xi xi1 ) + MyP12 (y xj1 ) + MyP22 (xj y)
i6=j
= MjP1 (xj xj1 ) MyP12 (y xj1 ) MyP22 (xj y)
MjP1 (xj xj1 ) MjP1 (y xj1 ) MjP1 (xj y)
since MjP1 MyP12 , MyP22
=0
i.e. S(P1 ) S(P2 ). Apply a similar argument to complete the proof.

Rb
Now the lower integral f of f over [a, b] is defined to be
a
sup{s(P )| P is a partition of [a, b]} and the upper integral
Rb
a
f of f over [a, b] is defined to be inf{S(P )| P is a partition of [a, b]}. Thus
for every partition P we have
Z b Z b
s(P ) f, S(P ) f.
a a

Lemma 9.1.6 If P1 , P2 are any two partitions of [a, b], then s(P1 ) S(P2 ).
Proof: Let P be a common refinement of P1 and of P2 (for example P = P1 P2
would suffice). Then by the previous result
S(P ) S(P1 ), S(P ) S(P2 ), s(P1 ) s(P ) and s(P2 ) s(P ).
162 CHAPTER 9. THE RIEMANN INTEGRAL

In addition we have s(P ) S(P ). Hence

s(P1 ) s(P ) S(P ) S(P2 )

as required. And so
Theorem 9.1.7 For every function f : [a, b] R
Z b Z b
f f.
a a

Proof: Fix a partition P1 , then s(P ) S(P1 ) for every partition P . So


Z b
() f S(P1 )
a
Rb
since S(P1 ) is an upper bound of the lower sums and f is the least upper
a
bound of the lower sums. Now allow P1 to vary over all partitions. Since (*)
holds for every partition we have
Z b Z b
f f,
a a

the least upper bound of the lower sums cannot be greater than the greatest
lower bound of the upper sums.

9.2 Using the definition of the Riemann Integral


Consider the function f : R R defined by f (x) = x2 , in the interval [0, 1].
(For simplicity we have chosen an f such that f 0.) Let

n : 0 = x0 < x1 < x2 < < xn = 1

be a subdivision of [0, 1] into subintervals of equal length, i.e. xi = ni for every


i = 0, 1, . . . , n. For every interval [xi1 , xi ] we have

(i 1)2
 
i1
min {f (x)} = min {x2 } = f =
x[xi1 ,xi ] i1
n x n
i n n2

and
i2
 
i
max {f (x)} = max {x2 } = f = 2.
x[xi1 ,xi ] i1 i
n x n
n n
i2
It follows that in constructing an upper sum we can use rectangles of height n2
2
(i1)
and to construct a lower sum we can use rectangles of height n2 .
9.2. USING THE DEFINITION OF THE RIEMANN INTEGRAL 163

Sketch the graph of f on the interval [0, 1] and verify these statements.

Let Un be the upper sum constructed from rectangles of width (xi xi1 ) = n1
2
and height ni 2 . Let Ln be the lower sum constructed from rectangles of width
1 (i1)2
(xi xi1 ) = n and height n2 .

1 i2
The area of any rectangle in the upper sum Un is then n n2 and the area of
2
1 (i1)
any rectangle in the lower sum Ln is Let Ai be the area of the region
n n2
bounded by the graph of f , the x-axis the line x = xi1 = i1 n and the line
i
x = xi = n . We have
(i 1)2 i2
3
Ai 3 .
n n
Let A be the area bounded by the graph of f , the x-axis, the line x = 0 and
n n
i2
P P
the line x = 1. Clearly A = Ai . The upper sum Un = n3 , the lower sum
i=1 i=1
n
P (i1)2
Ln = n3 and
i=1
n n
X (i 1)2 X i2
A .()
i=1
n3 i=1
n3
n n
P i2 n(n+1)(2n+1) P (i1)2 (n1)n(2n1)
We have n3 = 6n3 () and similarly n3 = 6n3 .
i=1 i=1

Notice now that limn Ln = 13 (i.e. ( > 0), (n N) such that |Ln 13 | < ).
Also limn Un = 13 (i.e. ( > 0), (n N) such that |Un 31 | < ). (**)

And so there is a choice of n for which Un Ln is arbitrarily small. Thus f is


Riemann integrable on [0, 1]. We said that if f was Riemann integrable then
there exists a unique number A, such that for all upper sums U and for all lower
sums L, L A U . A is the Riemann integral of f on [0, 1].
R1
In fact by (*) and (**) we have A = 13 , i.e. 0
x2 dx = 31 .

Pn n(n+1)
Proof of (): We have i=1 i = 2 . (There are several ways to prove
this...try something). Then
(i + 1)3 i3
= 3i2 + 3i + 1 so that
Pn Pn
(n + 1)3 = 3 3
i=0 ((i + 1) i ) =
2
i=0 (3i + 3i + 1)
Pn 2 Pn Pn
= 3 i=0 i + 3 i=0 i + i=0 1.
Thus
n
3 2 3n(n + 1) X
n + 3n + 3n + 1 (n + 1) = 3 i2
2 i=0
164 CHAPTER 9. THE RIEMANN INTEGRAL

Pn n(n+1)(2n+1)
i.e. i=0 i2 = 6 . Similarly
Pn (i1)2 (n1)n(2n1)
i=1 n3 = 6n3
(n1)(2n1)
= 6n2 = Ln .

So
2n2 3n+1
lim Ln = lim 6n2
n n
23/n+1/n2 1
= lim 6 = 3
n

by the algebra of limits and

2n2 +3n+1
lim Un = lim 6n2
n n
2+3/n+1/n2 1
= lim 6 = 3
n

so lim (Un Ln ) = 0 as stated above. So > 0, N such that n N


n
n > we have Un Ln < .

9.3 Properties of the Riemann integral

(1) Linearity Let f, g be integrable functions on [a, b] and let k R. Then


Rb Rb
(a) kf is integrable and a
kf = k a
f and
Rb Rb Rb
(b) f + g is integrable and a
(f + g) = a
f+ a
g.

Proof: (a) f is integrable on [a, b] so let P = {x0 , . . . , xn } be a partition and


let
n
X n
X
S(f ) = MiP (f )(xi xi1 ) and s(f ) = mP
i (f )(xi xi1 ),
i=1 i=1

where MiP (f ) = sup{f (x) : x [xi1 , xi ]} and


Rb
mP
i (f ) = inf{f (x) : x [xi1 , xi ]}. We have lim S(f ) = a f = lim s(f ).
n n

Suppose k > 0. Now

MiP (kf ) = sup{(kf )(x) : x [xi1 , xi ]}


= k sup{f (x) : x [xi1 , xi ]}
= kMiP (f )
9.3. PROPERTIES OF THE RIEMANN INTEGRAL 165

and mP P
i (kf ) = kmi (f ) so
Rb
lim S(kf ) = lim kS(f ) = k lim S(f ) = k a f and
n n n Rb
lim s(kf ) = lim ks(f ) = k lim s(f ) = k a f.
n n n

Rb Rb
So kf is integrable and a
kf = k a
f.

For k < 0, if MiP (f ) = supx[xi1 ,xi ] {f (x)} then kMiP (f ) = inf x[xi1 ,xi ] {kf (x)} =
Rb
mP lim S(kf ) = k lim s(f ) = k a f . In additon kmP
i (kf ) so n
P
i (f ) = Mi (kf )
n
Rb
and so lim s(kf ) = k lim S(f ) = k a f .
n n

(b) We have
MiP (f + g) MiP (f ) + MiP (g) and
mP P P
i (f + g) mi (f ) + mi (g),

(proof is on the exercises).

Let > 0, since f, g are integrable

P1 : S(f, P1 ) s(f, P1 ) < /2 and


P2 : S(f, P2 ) s(f, P2 ) < /2.

Let P = P1 P2 then by Lemma 9.1.5, S(f, P) S(f, P1 ) and s(f, P) s(f, P1 )


and S(g, P) S(g, P2 ) and s(g, P) s(g, P2 ) so

S(f, P) s(f, P) < /2 and


S(g, P) s(g, P) < /2.

Thus
S(f + g, P) S(f, P) + S(g, P)
< s(f, P) + s(g, P) +
s(f + g, P) +
so S(f + g, P) s(f + g, P) < . Hence f + g is integrable. Now
Rb
a
(f + g) = lim s(f + g, P)
n
lim s(f, P) + s(g, P)
n
Rb Rb
= a f + a g and
Rb
a
(f + g) = lim S(f + g, P)
n
lim S(f, P) + S(g, P)
n
Rb Rb
= a f + a g thus
166 CHAPTER 9. THE RIEMANN INTEGRAL

Rb Rb Rb
a
(f + g) = a
f+ a
g.

(2) Comparison Property Suppose f, g are integrable on [a, b] and f (x)


g(x), x [a, b].
Let P = {x0 , . . . , xn } be any partition of [a, b] then

MiP (f ) MiP (g) and mP P


i (f ) mi (g),

since f (x) g(x).

Rb Rb
Now a f = lim S(f, P) = lim s(f, P) and a g = lim S(g, P) = lim s(g, P)
n n n n
but since
n
X n
X
S(f, P) = MiP (f )(xi xi1 ) MiP (g)(xi xi1 ) = S(g, P)
i=1 i=1

Rb Rb
we have a
f a
g as required.

(3) Absolute Value Property Suppose f is integrable on [a, b]. Then |f | is


integrable on [a, b] and
Z b Z b
| f| |f |.
a a

Proof: Define functions Define functions


 
f (x) f (x) 0 f (x) f (x) 0
f + (x) = f (x) =
0 otherwise and 0 otherwise

Then |f | = f + + f since x [a, b] we have


(
f (x) f (x) 0
|f |(x) = |f (x)| =
f (x) f (x) 0
(
f + (x) + f (x) f (x) 0 since f (x) = 0
=
f + (x) + f (x) f (x) 0 since f + (x) = 0
= f (x) + f (x), x [a, b].
+

Now suppose f is integrable on [a, b]. Let P = {x0 , . . . , xn } be a partition of


[a, b].

Now MiP (f ) = sup{f (x) : x [xi1 , xi ]} and so


MiP (f + ) = sup{f + (x) : x [xi1 , xi ]} and if f (x) < 0 on [xi1 , xi ] then
MiP (f + ) = 0; whilst if P +
Pfn(x) P 0 on [xi1 , xi ] then Mi (f P) = MiP (f ) so since
n P +
lim S(f, P) = lim i=1 Mi (f )(xi xi1 ) exists lim i=1 Mi (f )(xi
n n n
xi1 ) exists.
9.3. PROPERTIES OF THE RIEMANN INTEGRAL 167

Similarly since
inf{f (x) : x [xi1 , xi ]} = sup{f (x) : x [xi1 , xi ]} either MiP (f ) coincides
with mP P
i (f ) or mi (f ) = 0, so again, since

lim S(f, P) = lim s(f, P)


n n

exists we have lim S(f , P) (or lim s(f , P)) exists. Thus |f | is integrable by
n n
the linearity property.

Now to establish the inequality of integrals we have

|f (x)| f (x) |f (x)|, x [a, b]; i.e. |f | f |f |

so by the comparison property


Z b Z b Z b
|f | f |f |
a a a
Rb Rb
which is to say | a
f| a
|f | as required.
168 CHAPTER 9. THE RIEMANN INTEGRAL
Chapter 10

Major results used in


calculus

10.1 The fundamental theorem of calculus


Theorem 10.1.1 The fundamental theorem of calculus (FTC)
(i) Let f : I R be continuous, where I is an interval. Then for any a I,
Z 0
f = f and
a

(ii) let : I R be differentiable with 0 continuous, where I is an interval.


Then for any points a, x I,
Z x
0 = (x) (a).
a
R R 
Here the notation a f denotes the function whose value at x is a f (x) =
Rx Rx
a
f = a f (t)dt.
You should already be convinced of the usefulness of the theorem as you have
used it in your calculus courses to compute integrals using antiderivatives and
to solve differential equations. In this course we focus on why it is true.

Thinking geometrically, the first


R part of the FTC is almost obvious. Let us
denote the indefinite integral a f by F , so
Z 
f (x) = F (x).
a

Now Z x+h Z x Z x+h


F (x + h) F (x) = f f= f
a a x

169
170 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

for each x I and each h R such that x + h I. Thus F (x + h) F (x) is the


area of the shaded region (for h > 0). When h is sufficiently small this area is
approximately equal to the area of the rectangle of height f (x) and width h.
6

-
x x+h
i.e. F (x + h) F (x) hf (x) and so it seems plausible that

F (x + h) F (x)
lim = f (x),
h0 h
i.e. given > 0 we must find > 0 such that whenever 0 < |h| < , we have
Z Z
1 x+h 1 x+h
f f (x) < i.e. (f (y) f (x))dy < .

h x h x

To prove FTC(i):
Proof: We have two cases , case A, h > 0 and case B, h < 0.
Case A: h > 0.

Exploration By the continuity of f at x we can ensure that by making h small,


we can also make the difference between the shaded area F (x + h) F (x) and
hf (x) very small.
We have > 0, > 0 such that y I

|x y| < |f (x) f (y)| <

so given > 0 if we choose |h| < then for each y [x, x + h], |x y| < and
so |f (x) f (y)| < so since the integrand is small, the area difference is small.
Now, using the observations in the exploration, we prove

x I and > 0, > 0 such that


R x+h
0 < h < h x (f (y) f (x))dy h i.e.
R x+h
(*)0 < h < h(f (x) ) x f (y)dy h(f (x) + ).

Proof: Let x I, let > 0 and choose > 0 satisfying

y I, if |x y| < then |f (x) f (y)| <


10.1. THE FUNDAMENTAL THEOREM OF CALCULUS 171

then if 0 < h < we have

< f (y) f (x) < for all y [x, x + h]

so by the comparison property of integrals


R x+h R x+h R x+h
x x (f (y) f (x))dy x
R x+h
i.e. h x f (y)dy f (x)h h
R x+h
i.e. h + f (x)h x f (y)dy h + f (x)h

which is (*).
R
Now using the notation F = a
f we have
Z x+h Z x+h
f (y)dy = f = F (x + h) F (x).
x x

F (x+h)F (x)
To prove: lim h = f (x).
h0+

Proof: Let > 0, by the previous pargraph > 0 such that h with 0 < h <

h(f (x) ) F (x + h) F (x) h(f (x) + )


F (x+h)F (x)
i.e. h, 0 < h < , f (x) h f (x) +
F (x+h)F (x)
so h f (x)
F (x+h)F (x)
i.e. lim h = f (x).
h0+

[Notice we could have chosen : 2 < f (x) f (y) < 2 to get


< 2 lim+ F (x+h)F
h
(x)
2 < .]
h0
We have F 0 (x) := lim+ F (x+h)F (x)
= f (x) so F 0 = f or ( a f )0 = f as required.
R
h
h0

Case B: h < 0.
We have < h < 0 for > 0 and
Z x
h (f (x) f (y))dy h
x+h

and so h( f (x)) F (x) F (x + h) h( f (x))


thus, remembering that inequalities reverse when dividing by h < 0

F (x) F (x + h)
f (x) f (x)
h
172 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

so multiplying through by -1

f (x) F (x+h)F
h
(x)
f (x) +
F (x+h)F (x)
i.e. h f (x)

and so replacing by 2 we have

F (x + h) F (x)
lim = f (x).
h0 h
Geometrically

6
f (x) +
f
f (x)

f (x)

-
x x x+h x+
On (x , x + ) the graph of f is between the horizontal lines f (x) and
R x+h
f (x) + so for 0 < h < , A1 x f A2 , where A1 is the area of the
rectangle with base [x, x + h] and height f (x) and A2 is the area of the
rectangle with base [x, x + h] and height f (x) + .

Proof of the second and most useful part of the fundamental theorem of calculus
is more involved. We use the zero derivative theorem and the mean value
theorem.

10.2 Zero derivative/mean value theorem


Theorem 10.2.1 (Zero derivative theorem)
If F 0 = 0 on an interval I, then F is constant on I.

In one of the exercises you will be asked to prove FTC (ii), using FTC (i) and
the zero derivative theorem.

But proof of the zero derivative theorem is not trivial, the trouble being that
from F 0 (x) = 0 we cannot deduce that F is constant close to the point x. For
example, F = Id2 and x = 0 (indeed n N, n > 1 with F = Idn ) we have
F 0 = nIdn1 , so F 0 (0) = 0 but F is not a constant function on any interval
containing zero.
10.2. ZERO DERIVATIVE/MEAN VALUE THEOREM 173

We will be able to prove the zero derivative theorem using the mean value
theorem.

Theorem 10.2.2 (Mean value theorem)


Let f : [a, b] R be continuous (where a < b) and let f be differentiable on
(a, b). Then x (a, b) with

f (b) f (a)
f 0 (x) = .
ba

f (b) 


 f

f (a)  


x

-
a b
Geometrically the mean value theorem says that at some point (x, f (x)) on
the graph of f the slope of the tangent equals the slope f (b)f
ba
(a)
of the chord
between the points (a, f (a) and (b, f (b)).

Proof: (of 10.2.1 assuming that 10.2.2 is true)


Suppose F 0 = 0 on I. Let s be some fixed element of I [we must show that
t I, F (t) = F (s) i.e. F is constant on I.]
Let t I, by the mean value theorem 10.2.2, x (s, t) (if s < t otherwise
(t, s)) with

F (t) F (s)
F 0 (x) = .
ts

But F 0 = 0 on I, so F 0 (x) = 0 hence F (t) = F (s).


Since t was arbitrary, this is true for every point t I so F is a constant
function.

And so we have the following scheme of results:


174 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS


FTC (i)



mean value

zero FTC (ii)
theorem derivative



theorem

So once we have proved the mean value theorem we will have a proof of
FTC(ii). The proof of 10.2.2 will come towards the end of the course. We will
examine some other useful results of real analysis first.

10.3 Bolzanos theorem


The next theorem can be used to establish the existence of solution(s) to equa-
tions.

Theorem 10.3.1 (Bolzanos theorem)


Let f be a real valued function on a domain which includes the closed interval
[a, b]. If

(i) f is continuous on [a, b] and

(ii) f (a) < 0 and f (b) > 0 then

x (a, b) such that f (x) = 0.

In terms of graphs the theorem says, if the graph of a continuous function has
a point below the x axis and a point above the x-axis then the graph must
cross the x-axis, i.e. there is a point x such that f (x) = 0. It is clear that the
result is false if we omit continuity.

We give two examples in which we use the theorem and we prove the theorem
in section 10.6.
(1) We show that x R satisfying x2 = 2.
Let f (x) = x2 2 then (i) f is continuous. [We use the definition of continuity
to prove this, although we could also argue from the algebra of continuous
functions.]
Let a R and let B = B (f (a)) = B (a2 2). Choose = min{1, 2a+1 } and
2
A = B (a). Now let x A. We have a < x < a + and f (a ) = a 2a +
2 2 > a2 2(2a+1), whilst f (a+) = a2 +2a + 2 2 < a2 2+(2a+1),
since 2 < , thus

a2 2 (2a + 1) < f (x) < a2 2 + (2a + 1),


10.4. INTERMEDIATE VALUE THEOREM 175

so f (x) B. Hence f is continuous at a. But a was an arbitrary point in R


and so f is continuous on R.

And (ii) f (0) = 2 < 0 whilst f (2) = 2 > 0. So the hypotheses of Bolzanos
theorem hold and so the conclusion is that there is a point x (0, 2) such that
f (x) = 0; i.e. x2 2 = 0 and so there is a point x (0, 2) such that x2 = 2.

(2) We show that cos x = x has atP least one solution between 0 and /2.
Let f (x) = cos x x then (i) f = (Id1 cos) and so is the sum of continuous
functions, so by the algebra of continuous functions f is continuous on R. And
(ii) f (0) = 1 < 0 whilst f (/2) = /2 > 0 thus Bolzanos theorem says there
is x (0, /2) such that f (x) = 0, i.e. cos x = x. We next state two results
which can be easily deduced from Bolzanos theorem.

10.4 Intermediate value theorem


Theorem 10.4.1 Let f be a real valued function which is continuous on the
closed interval [a, b]. If y is a number such that f (a) < y < f (b) (or f (b) < y <
f (a)) then x (a, b) such that y = f (x).

The following graph illustrates the meaning of the theorem:

f (b) 6 f
y
f (a)

x
| -
a b
so for each y between f (a) and f (b) the horizontal line at height y intersects
the graph in at least one point.

It is clear that we cannot omit the condition that f is continuous, for example

f (b) 6
y


f (a) 
-
a b
To use the theorem it suffices to establish that the function is continuous on
the relevant interval. For example, the function sin : R R is continuous and
sin (/2) = 1 and sin (/2) = 1. It follows from theorem 10.4.1, that sin
176 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

assumes every value in the interval [1, 1] i.e. y [1, 1], x (/2, /2)
such that sin x = y.

10.5 Brouwers fixed point theorem


Theorem 10.5.1 Let f : Rn Rn be continuous. If f maps a closed ball
B Rn into itself then there is an x B such that f (x) = x, i.e. f has a fixed
point x B.

In the special case n = 1 this theorem is an easy consequence of Bolzanos


theorem. We have f : R R continuous. A closed ball in R is a closed
interval and so the theorem has the following interpretation in terms of graphs.
If continuous f maps [a, b] into [a, b] then the graph of f lies in the square
[a, b] [a, b]. Now the graph of the identity function passes through the diagonal
of the square from point (a, a) to point (b, b). The conclusion is the observation
that the graph of f must cross the graph of the identity function.

b 6

-
a b

10.6 Proof of Bolzanos theorem


We prove Bolzanos theorem in the following way. The number x such tht
f (x) = 0 will be found as the least upper bound of a certain set S. Several
choices of S are possible leading to technical differences in the resulting proof.
Some of the possibilities may be explored on an exercise.
The following lemma contains all of the information about the continuity of f
needed in the proof of Bolzanos. We will prove the lemma at the end of the
section.

Lemma 10.6.1 (No-jump lemma)


Let D R. If f : D R is continuous and l D is such that f (l) > 0 then
there exists > 0 such that for all x D, if |x l| then f (x) > 0.

The lemma says that if f (l) > 0 then because f is continuous it cannot suddenly
take negative values arbitrarily close to l; i.e. f (x) > 0, x sufficiently close to
10.6. PROOF OF BOLZANOS THEOREM 177

l.
(*) Of couse a similar result holds for f (l) < 0.

Proof of Bolzanos theorem 10.3.1:


Let f be a real valued function whose domain D includes the closed interval
[a, b] and suppose that

(i) f is continuous and

(ii) f (a) < 0 and f (b) > 0.

Define a set S by putting S = {x [a, b] : f (x) < 0}.


Then S 6= since a S. In addition S is bounded above (for example b is an
upper bound). Hence S has a least upper bound l R.

Next we show that if we assume that f (l) < 0 we get a contradiction and if we
assume that f (l) > 0 we also get a contradiction, so the only possibility is that
f (l) = 0.

Suppose that f (l) < 0.


Then l 6= b since f (b) > 0. Also l b, since b is an upper bound and so l < b.

| || | -
a ll+ b

Now by the no-jump lemma > 0 such that x [l , l + ], f (x) < 0. But
since f (b) > 0 we have b 6 [l , l + ], so l + < b. Now l a and l + [a, b].
In addition f (l + ) < 0 so l + S. This is the required contradiction, since
l = sup S and so l is an upper bound of S. Hence f (l) 6< 0.

Next suppose that f (l) > 0.


Then l 6= a since f (a) < 0. Also l a, since a is a lower bound and so l > a.

| || | -
a l b
l
178 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

Now by the no-jump lemma > 0 such that x [l , l + ], f (x) > 0. (1)
But since l = sup S, x S such that l < x,(there is no upper bound less
than l). Since x S, f (x) < 0 and x l so

x S such that f (x) < 0 and l x l + . (2)

But (2) contradicts (1). Hence f (l) 6> 0. Thus f (l) = 0, completing the proof
of Bolzanos theorem.

Proof of the no-jump lemma:


Suppose f : D R is continuous and l D with f (l) > 0. By the continuity
of f we have

> 0, > 0 such that x D, |x l| < |f (x) f (l)| < ,

i.e. f (l) < f (x) < f (l) + .


So setting = f (l)
2 > 0 (since f (l) > 0) we know

f (l) 3f (l)
> 0 such that x D |x l| < < f (x) <
2 2
so f (x) > 0.

Notice this is not quite the no-jump lemma. If we now put 0 = /2 we have

x D, |x l| 0 f (x) > 0

as required.

Again, the no-jump lemma clearly fails as soon as we remove the continuity
requirement.
For example 
1 if x < 0
f (x) =
1 if x 0
with l = 0 then f (l) = 1 > 0 but 6 > 0 such that

x D, l x l + f (x) > 0

since > 0 x = l < l = 0 so f (x) = 1 < 0.

10.7 Maximum/minimum values/boundedness


There are a variety of practical problems which lead to the desire to find the
maximum (or minimum) value of some real valued function f . Our concern will
be with the basic issue of whether or not a maximum exists.
10.7. MAXIMUM/MINIMUM VALUES/BOUNDEDNESS 179

Let f be a real valued function and let S domf . If there is an element c S


such that x S f (x) f (c) then we say that f (c) is the maximum value of f
on S.
Examples.
1. Let f : R R with f (x) = 2x + 1 and let S = [0, 1]. Then max f (S) = 3
since 1 [0, 1] and x [0, 1] f (x) 3 = f (1).
2. Let f : R R with f (x) = 2x + 1 and let S = (0, 1). Then f has no
maximum value on S since c S we have c + 1c 1c
2 S and c < c + 2 so since
1c
f is monotonic increasing f (c) < f (c + 2 ).
3. Let f : R2 R with f (x, y) = x + 2y and let S = {(x, y) : 0 x 1, 0 y
1}. So the maximum value of f on S is 3 when (x, y) = (1, 1) since (1, 1) S
and (x, y) S, f (x, y) f (1, 1).

We recall the notion of a bounded function.


Theorem 10.7.1 Boundedness theorem
Let f be a real valued function which is continuous on the closed interval [a, b].
Then f is bounded (above and below) on [a, b].
The condition that the interval be closed cannot be omitted. We will prove this
result shortly, we need another result first. This is our main result concerning
the existence of maximum values of functions.
Theorem 10.7.2 (Maximum value theorem)
Let f be a real valued function which is continuous on the closed interval [a, b].
Then f assumes a maximum value on [a, b].
[There is a similar result for the existence of a minimum value.]

In analysis it can be useful to distinguish between properties of functions which


are local and those which are global. A local property tells you about what
happens near some point in domf . A global property tells you what happens
over the whole of some interval in its domain. There is a local form of the
boundedness theorem, which is easy to prove.
Lemma 10.7.3 (Local boundedness lemma)
Let f be a real valued function with domain D R. If f is continuous at l then
there exists > 0 such that f is bounded above (and below) on (l , l + ) D.
Proof: Let f be continuous at l, then > 0, > 0 such that x D
|x l| < |f (x) f (y)| < .
So with = 1 for example, > 0 such that x D ,
x (l , l + ) f (x) (f (l) , f (l) + )
and so for this , f (l) + 1 is an upper bound for f on (l , l + ) D.
[Similarly f (l) 1 is a lower bound.]
And so we go on to prove the boundedness theorem 10.7.1.
180 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

Proof of 10.7.1: Let f be a real valued function which is continuous on the


closed interval [a, b].
Let S = {x [a, b] : f is bounded above on [a, x]}. Then S 6= because a S.
S is bounded above (by b for example), hence sup S exists.
Let l = sup S.

Now suppose l < b (we want a contradiction).


(i) By the local lemma 10.7.3, for some > 0, f is bounded above on (l , l +)
and
(ii) since l is the least upper bound of S x S such that l < x so f is
bounded above on [a, x] for some x satisfying x > l .

By (i) and (ii) f is bounded above on [a, l +) and hence on [a, l + 2 ] so l + 2 S


and l + 2 > l. This is a contradiction since l = sup S and so must be an upper
bound. Hence l = b.

Finally, since we have f bounded above on [a, l + ), we have f bounded above


on [a, l] and so f bounded above on [a, b], since l = b.
Similarly, one can prove that f is bounded below on [a, b].
And so we prove the maximum value theorem. We use the boundedness theorem.

Proof of theorem 10.7.2: Let f be a real valued function which is continuous


on the closed interval [a, b].
By the boundedness theorem 10.7.1, f is bounded (above and below) on [a, b]
so
{f (x) : x [a, b]} is bounded above.
It is also non-empty, since it contains f (a) for example. Hence by the least
upper bound property of R this set has a least upper bound say l R.
Since l is an upper bound we must have
x [a, b] f (x) l. (1)
If we can show that for some x the equality in (1) is attained then we have the
result that l is the maximum value of f on [a, b].
Geometrically we are asking whether or not the graph of f touches the graph
of l.
l 6

| | -
a b
10.8. ROLLES THEOREM/PROOF OF MEAN VALUE THEOREM 181

Suppose it doesnt. Then strict inequality obtains throughout [a, b] so that we


may write f < l on [a, b]. (Now we find a contradiction).
We supposed that f < l on [a, b]
i.e. l f > 0 on [a, b]
1
so lf is continuous on [a, b].
1
By the boundedness theorem lf must have an upper bound, say > 0 on [a, b]
1
thus lf < on [a, b]
i.e. l f > 1 on [a, b]
i.e. f < l 1 , which says that l 1
is an upper bound for the values of f on
[a, b] contradicting the assumption that l was the least such upper bound.
Hence x [a, b] such that f (x) = l, i.e. l is the maximum value of f on [a, b].

And so we will now return to proving the mean value theorem which we still
had to prove to complete the proof of the FTC (ii).

10.8 Rolles theorem/proof of mean value theo-


rem
First we give Rolles theorem, which is a special case of the mean value theorem.

Theorem 10.8.1 (Rolles theorem)


Let f be a real valued function which is continuous on [a, b] and differentiable
on (a, b), where a < b.
If f (a) = f (b) = 0 then x (a, b) such that f 0 (x) = 0.

Proof: If f is constant on [a, b] then clearly f 0 (x) = 0, x [a, b] (since


f (a) = f (b) = 0 = f (x), x [a, b]).
We suppose that f takes only positive values at some points of (a, b) since if
this is not true then we can replace f by f in the following discussion.
By the maximum value theorem 10.7.2, x [a, b] such that
f (x) = max{f (y) : y [a, b]}. Clearly f (x) > 0 as f takes some positive
values by assumption, so x 6= a and x 6= b, i.e. x (a, b) such that f (x) =
maxy[a,b] {f (y)} (*). To complete the proof of 10.8.1 we will prove that for x
satisfying (*), f 0 (x) = 0.

Lemma 10.8.2 For a continuous, real valued function which satisfies the hy-
potheses of theorem 10.8.1, we have that for x (a, b) with f (x) = maxy[a,b] {f (y)},
f 0 (x) = 0.

Proof of the lemma: Now

f (x + h) f (x) f (x + h) f (x)
lim = f 0 (x) = lim
h0+ h h0 h
182 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

and for all sufficiently small values |h| (in particular satisfying x + h [a, b]) we
have f (x + h) f (x), since f (x) = maxy[a,b] {f (y)}. So
f (x+h)f (x)
f 0 (x) = lim+ h 0 and
h0
f (x+h)f (x)
f 0 (x) = lim h 0
h0

consequently f 0 (x) = 0.

We are now in a position to prove the mean value theorem 10.2.2.


Proof: Suppose f : [a, b] R is continuous and f is differentiable on (a, b).
Let g : [a, b] R be defined by
(x a)(f (b) f (a))
g(x) = f (x) f (a) .(*)
ba
Then by the algebra of continuous functions, g is continuous on [a, b] and dif-
ferentiable on (a, b), since f has these properties. Notice that

g(a) = f (a) f (a) (aa)(fba


(b)f (a))
= 0 and
ba
g(b) = f (b) f (a) ba (f (b) f (a)) = 0
so by Rolles theorem (applied to g)
x (a, b) such that g 0 (x) = 0, i.e.
f (b) f (a)
0 = g 0 (x) = f 0 (x) differentiating (*) with respect to x
ba
f (b)f (a)
thus f 0 (x) = ba as required.

So to complete the diagram of results we have

local maximum
lemma
boundedness boundedness value
10.8.2
lemma theorem theorem
| {z }

Rolles theorem




FTC (i)


FTC (ii) zero mean value theorem


derivative
theorem



10.9. IVT FOR DERIVATIVES/LHOPITAL 183

completing the proof of the fundamental theorem of calculus (ii).

10.9 IVT for derivatives/lHopital


Theorem 10.9.1 (Intermediate value theorem for derivatives)
Let f be differentiable on [a, b] and suppose k is a number between f 0 (a) and
f 0 (b). Then there is a point c (a, b) such that f 0 (c) = k.
Proof: Suppose f 0 (a) < k < f 0 (b). Let x [a, b] and define g(x) = f (x) kx.
Then g is differentiable on [a, b] since f is and g 0 (x) = f 0 (x) k thus
g 0 (a) = f 0 (a) k < 0 and g 0 (b) = f 0 (b) k > 0.
Now g is continuous on [a, b] (since it is differentiable) and so by the minimum
value theorem, g assumes a minimum at some point c [a, b], i.e. x [a, b],
g(c) g(x).
Claim: c (a, b).
Now g 0 (b) = lim g(x)g(b)
xb and g 0 (b) > 0, thus g(x)g(b)
xb > 0 on a deleted neigh-
xb
bourhood U of b. So for x < b, g(x) < g(b). Thus g(b) is not the minimum.
And g 0 (a) = lim g(x)g(a)
xa and g 0 (a) < 0, thus g(x)g(a)
xa < 0 on a deleted neigh-
xa
bourhood V of a. So for x > a, g(x) < g(a). Thus g(a) is not the minimum.
Hence c (a, b) as claimed.
Now using the result of lemma 10.8.3 (applied to minimum rather than maxi-
mum) we have g 0 (c) = 0. But g 0 (c) = f 0 (c)k, so c (a, b) such that f 0 (c) = k.

Theorem 10.9.2 (The Cauchy Mean Value Theorem) If f, g C[a, b] and f, g


are differentiable on (a, b) then x (a, b) such that
(f (b) f (a))g 0 (x) = (g(b) g(a))f 0 (x).
Notice that if g(x) = x for all x [a, b], then g 0 (x) = 1, g(b) = b and g(a) = a
and the Cauchy mean value theorem reduces to the mean value theorem.
Given the hypothesis, if we apply the mean value theorem separately to f and
g, we have y (a, b) such that f (b) f (a) = (b a)f 0 (y) and z (a, b)
such that g(b) g(a) = (b a)g 0 (z) and so the content of the theorem is that
x (a, b) which satisfies the statement for both f and g.

Proof: Let f, g be continuous on [a, b] and differentiable on (a, b) and set


h(x) = f (x)(g(b) g(a)) g(x)(f (b) f (a)). Then h is continuous on [a, b]
and differentiable on (a, b), being a linear combination of the functions f and g.

h(a) = f (a)(g(b) g(a)) g(a)(f (b) f (a))


= f (a)g(b) g(a)f (b)
= f (b)(g(b) g(a)) g(b)(f (b) f (a))
= h(b)
184 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS

and so applying Rolles theorem, x (a, b) such that h0 (x) = 0. And so


0 = h0 (x) = f 0 (x)(g(b) g(a)) g 0 (x)(f (b) f (a)),
which is the required result.

If g(b) g(a) 6= 0 and g 0 (x) 6= 0 for x [a, b] satisfying the conclusion to the
Cauchy mean value theorem, then one can express the conclusion as
f (b)f (a) f 0 (x)
g(b)g(a) = g 0 (x) .

Theorem 10.9.3 (lHopitals rule) If f, g : S R R, where S contains the


0
closed interval [a, b] and if f (a) = g(a) = 0 and lim fg0 (t)
(t)
exists then
ta
0
f (t) f (t)
lim = lim .
ta g(t) ta g 0 (t)
In fact, we dont need f, g continuous at a and so we could replace the condition
f (a) = g(a) = 0 by lim f (x) = 0 = lim g(x) and use the local principle for
xa xa
limits.

Proof: Suppose f, g : S R R, where S contains the closed interval [a, b],


0
suppose f (a) = g(a) = 0 and suppose lim fg0 (t)
(t)
exists. The last condition implies
ta
two things, that interval (a , a + )
(1) such that for all x (a , a + ), f 0 (x) and g 0 (x) exist (except possibly at
a itself) and
(2) on (a , a + ), g 0 (x) 6= 0 (except possibly at x = a).
Now if a < x < a + , then both the mean value theorem and the Cauchy mean
value thoerem apply to f and g on the interval [a, x].
[Similarly, if a < x < a, both theorems apply on [x, a].]
Applying the mean value theorem to g on [a, x] we have y (a, x) such that
g(x) g(a) g(x)
g 0 (y) = = , since g(a) = 0.
xa xa
So by (2), g(x) 6= 0, since g 0 (y) 6= 0.
Now applying the Cauchy mean value theorem to f and g we have x (a, x)
such that
(f (x) f (a))g 0 (x ) = (g(x) g(a))f 0 (x ), so
f (x)g 0 (x ) = g(x)f 0 (x ), i.e.
f (x) f 0 (x )
= 0 .
g(x) g (x )
f 0 (y)
Now since x (a, x), as x a, x a and since by hypothesis lim 0
ya g (y)
exists we have
f (x) f 0 (x ) f 0 (x)
lim = lim 0 = lim 0 .
xa g(x) xa g (x ) xa g (x)

10.9. IVT FOR DERIVATIVES/LHOPITAL 185

For a < x < a, we have from the mean value theorem y (x, a) such that
g 0 (y) = g(a)g(x)
ax
g(x)
= xa and the proof proceeds as in the case a < x < a + .
186 CHAPTER 10. MAJOR RESULTS USED IN CALCULUS
Chapter 11

Appendix

THEOREM: The normed vector space B[a, b] is complete.

Proof: Let {fn } n=1 be a Cauchy sequence in B[a, b]. We are to prove that
there is a function f in B[a, b] which is the limit of this sequence.
The first step is to find some way of constructing a function f which has a
reasonable chance of being the required limit. We now show how to define the
value of such a function f at each point x in [a, b].
To this end, let x [a, b]. Since for all m, n, N
|fm (x) fn (x)| ||fm fn ||
it follows that the real-valued sequence {fn (x)}n=1 is dominated by a Cauchy
sequence; hence it is itself a Cauchy sequence.
But R is complete (see Analysis I). Hence this Cauchy sequence {fn (x)} n=1
of real numbers must have a limit in R. It is therefore meaningful to define f (x)
by putting
f (x) = lim fn (x).
n
It remains to show that the function f defined in the previous paragraph is
indeed the limit of the sequence of functions {fn }
n=1 in the space B[a, b].
Let > 0.
Since {fn }
n=1 is a Cauchy sequence in B[a, b] it follows that

( N)(m, n N) m, n > ||fm fn || < .


Let be this natural number.
Now let x [a, b] and let m, n N, we have
|fm (x) fn (x)| < . ()
We can now fix m N and let n . Since inequalities are respected by
limits we have
lim |fm (x) fn (x)| .
n

187
188 CHAPTER 11. APPENDIX

But the absolute value function is continuous. So by the theorem continuity via
sequences we may take lim inside:

|fm (x) limn fn (x)| remembering that fm (x) is fixed.


i.e. |fm (x) f (x)| .

But this is true for all x [a, b]. Hence

||fm f || .

Thus we have shown that

( > 0)( N)|(m N) m > ||fm f || . ()

From this it would seem that we have reached our goal, except that we have
rather than < . We can fix this by using the 2 trick at line (); i.e.
go back to (*) and replace with 2 .
More seriously we have not yet shown that f B[a, b]. So we do this next.
Let x [a, b], and let n N. By the triangle inequality for R,

|f (x)| = |f (x) fn (x) + fn (x)|


|f (x) fn (x)| + |fn (x)|
||f fn || + ||fn ||

where by (), ||f fn || is certainly defined for some n and where ||fn || is defined
since fn B[a, b]. The right hand side is thus an upper bound for the set whose
typical member appears on the left. Hence f B[a, b], as required.

Finally we will demonstrate the claim that abs is the limit of the sequence
fn : [, ] R defined by
n
4X
fn = gi ,
2 i=1

where
cos (2i 1)Id
gi = .
(2i 1)2
The form of the Fourier series of a function g is

1
P
2 a0 + (an cos nx + bn sin nx), where
n=1
1
R
g(t) cos ntdt and bn = 1 g(t) sin ntdt.
R
an =

Now given g : [, ] R defined by g(x) = |x|


R we have g is an integrable
even function and so bn = 0. In addition, an = 2 0 g(t) cos ntdt, since g(x) =
189

R R
g(x), for all x [0, ]. Thus a0 = 2 0 g(t)dt = 2 0 ( x)dx, since |x| = x
on [0, ]. Thus a0 = . And
R
an = 2 h0( x) cos nxdx
R i
(x) sin nx
= 2 n + 0 n1 sin nxdx
0 
nx
= 2 0 + cosn
 2
0
n
= 2 1(1)
n2

4
and so for even n, an = 0 and for odd n = 2m 1 say, we have an = (2m1)2 ,
thus

4 X cos(2i 1)x
h(x) = +
2 i=1 (2i 1)2

is the Fourier series for g(x) = |x|. Thus



4 X cos(2i 1)x
|x| = = lim fn (x).
2 i=1 (2i 1)2 n

Você também pode gostar