Вы находитесь на странице: 1из 421

trim

ss

Analysis

>

Terence

Tao

QA

300

.T325

2006

v.l

u ihju

ILJOIHI

UM

HINDUSTAN

BOOK

AGENCY

TEXTS

AND

RteDWSS

IN MATHEMATICS

Analysis

Texts and Readingsin Mathematics


Editor

Advisory

C.

S. Seshadri,

Chennai Mathematical Inst.,

ManagingEditor
Bhatia,

Rajendra

Indian

Statistical

Inst.,

Chennai.

New Delhi.

Editors

V. S.

Borkar,

Tata

Inst,

of Fundamental

Research, Mumbai.

Kolkata.
Probal Chaudhuri,Indian
Statistical
Inst.,
R. L. Karandikar, Indian Statistical
Inst.,New Delhi.
M. Ram Murty, Queen's University, Kingston.

V.S.

Sunder,

Inst,

of Mathematical

Sciences,

M. Vanninathan,
TIFR
Centre,
Bangalore.
T. N. Venkataramana, Tata Inst, of Fundamental

Chennai.
Research,

Mumbai.

Analysis

Terence
University

of

Los

IUdIIdI

HINDUSTAN

1\320\233\320\250\320\237
BOOK

AGENCY

Tao

California

Angeles

Published by
Book Agency (India)
Park Extension

Hindustan
\316\24119
Green

New Delhi 110016


India

email:

hba@vsnl.com

http://www.hindbook.com

Copyright
No part

of

the

material

mechanical,
storage and

including
retrieval

in

system,

translation

All export

rights

edition vest

for this

Agency (India). Unauthorized


and is subject to legal action.
Produced

from camera

ready

ISBN 81-85931-62-3

written permission from the


right to grant licences for

without

owner, who has also the


into other languages and

copyright

(India)

protected
by this copyright notice may be
electronic or
any form or by any means,
or
recording
photocopying,
by any information

or utilized

reproduced

Book Agency

by Hindustan

\302\251
2006

export

copy

sole

publication

thereof.

exclusively

with

is a violation

supplied

by the

Hindustan

of

Copyright

Author.

Book
Law

To

my

parents,

for

everything

This

page

intentionally

left

blank

Contents

Volume

xiii

Preface

Introduction

1.1

What

1.2

Why do

is

analysis?

analysis?

2 The natural numbers

2.1
2.2

The

16

axioms

Peano

27

Addition

33

2.3 Multiplication

3 Set theory

37

37

3.1

Fundamentals

3.2

Russell's paradox

3.3 Functions . .
3.4

and

Images

inverse

(Optional)
images

64
70

of sets

76

rationals

84

Cardinality

4 Integers and

52
55

3.5 Cartesianproducts
3.6

14

4.1

The

4.2

The rationals

integers

4.3 Absolute value and exponentiation

84

92

98

CONTENTS

viii

4.4

rational numbers

in the

Gaps

107

5 The real numbers

5.1

Cauchy

5.2

Equivalent

5.3 The
5.4

5.5
5.6

Cauchy sequences

least

upper

property

117
133

139
145

sequences

Convergence and limit laws

6.2 The

114

127

bound

Real exponentiation,part I
of

number

extended

real

6.3

Suprema and

infima

6.4

Limsup, liminf, and limit

6.5

Some

6.6

Subsequences

of

system
sequences

points

145
153
157

160
170

limits

standard

171

6.7 Real exponentiation,


part
7

numbers

real

the

of

construction

6 Limits
6.1

109

sequences

Ordering the reals


The

103

II

175

Series

179

7.1 Finite series

179

7.2

Infinite

189

7.3

Sums of

series

non-negative numbers

7.4

Rearrangement

7.5

The root

of

200

series

and ratio tests

8 Infinitesets
Countability

Summation

208

8.3

Uncountable

on

sets

infinite

9.2
9.3

Subsets

227

232

sets

9 Continuous functions

9.1

of

the

real

on R
line

The algebraof real-valuedfunctions


Limiting

values

216

224

sets

8.4 The axiomof choice


Ordered

204
208

8.1

8.2

8.5

195

of functions

242
243
250

253

ix

CONTENTS

9.4

Continuous

9.5

Left and

9.6 The

266

right limits

maximum

269

principle

9.7

The intermediate

9.8

Monotonic

9.9

Uniform

273

value theorem

276

functions

279

continuity

9.10 Limitsat
10

261

functions

286

infinity

288

of functions

Differentiation

10.1 Basic definitions


10.2

Local

local minima,

maxima,

10.3

and

functions

Monotone

10.4 Inverse
10.5

288

and derivatives . .

298

derivatives

functions and derivatives

L'Hopital's

11 The Riemann

11.1

300

rule

303

integral

306
307

Partitions

11.2

11.3 Upper and lower

Riemann

317

integrals

Riemann
Riemann

integrability

of continuous

integrability

of monotone

non-Riemann

integrable

functions

Riemann-Stieltjes integral

11.9The

two

theorems

A.l

Mathematical

A.2

Implication

332

of calculus

338
343

349
357

364

A.3 The structure of proofs


and

A.4

Variables

A.5

Nested quantifiers
examples

A.7 Equality

logic

....

350

statements

A.6 Some

. 326
. 330

334

of mathematical

basics

the

Appendix:

. .
. .

function

11.8 The

fundamental

functions

ll.lOConsequences of the fundamentaltheorems


A

321

properties of the Riemannintegral

11.4 Basic

11.5

312

functions

constant

Piecewise

11.6
11.7

. 295

367

quantifiers

of

372
proofs

and

quantifiers

375

377

CONTENTS

\317\207

decimal

the

\316\222Appendix:

system

B.l The decimalrepresentation


B.2 The decimalrepresentation

of

natural

of

real

. .

numbers
numbers

....

380
381

385

Index

Volume

xiii

Preface

12 Metric

389

spaces

12.1

Definitions

12.2

Some point-set

and

12.3

Relative

examples

389

topology of metric spaces

400.
405

topology

12.4

Cauchy

12.5

Compact

sequences and complete metric spaces.


metric

13.1

on metric spaces

13.3

Continuity

and product

Continuity

and

13.4 Continuity

423

spaces

427

compactness

and connectedness

429

13.5 Topological
spaces

433

(Optional)

14

Uniform

440

convergence

441

14.1 Limiting values of functions

14.2

Pointwise

and

uniform

444

convergence

14.3 Uniform.convergence and continuity


14.4
The metric of uniform convergence

14.5 Seriesof
14.6 Uniform

14.7

Uniform

14.8 Uniform

functions;

the

Weierstrass

M-test

convergenceand integration
convergence

420
420

functions

Continuous

13.2

408

412

spaces

13 Continuous functions

..

and

approximation by

derivatives
polynomials

449

452
455

458
461
464

xi

CONTENTS

15 Power

474

series

15.1

Formal

474

series

power

analytic functions

15.2 Real

477

15.3 Abel's

483

theorem

15.4

Multiplication

15.5 The
15.6

and

exponential

on

digression

487

series

of power

490

functions

logarithm

494

numbers

complex

503

15.7 Trigonometric functions

510

16 Fourierseries
16.1

products on periodicfunctions

16.2 Inner
16.3

511

functions

Periodic

514

518

polynomials

Trigonometric

521

16.4 Periodic convolutions

16.5

The

17 Several variable

17.1

533

in several

Derivatives

17.3 Partial

540

variable calculus

544

derivatives

directional

and

17.4 The

533

differential calculus

transformations

Linear

17.2

526

theorems

Plancherel

and

Fourier

several variable calculuschain


rule
derivatives and Clairaut's theorem

552

555

Double

17.5

17.6

contraction

17.7 The

inverse functiontheorem

The

17.8

The

18 Lebesgue

573

18.3

Outer

575

measure

Lebesgue

First

567

measure

18.1The goal:

18.2

561

theorem

function

implicit

558

theorem

mapping

577

Outer measure

attempt:

is not

measure

587

additive

18.4 Measurable sets

590

18.5

597

functions

Measurable

19 Lebesgue

602

integration

19.1 Simplefunctions
19.2

602

Integration

of non-negative

Integration

of

19.3

absolutely

measurable functions .

integrable

functions

...

608

617

CONTENTS

xii

19.4

with the

Comparison

19.5Fubini's

theorem

Index

Riemann integral

622
624

Preface

This text originatedfrom


honours

real

undergraduate-level

of California,LosAngeles,
was
here, real analysis
to

courses
introduced

but

etc.),
of

the

with the

the

in

course

face

and
of

the

bright

perception of

order

to

difficulty,

the level
reducing
it easier, or to maintain

make

of many
and enthusiastic
ones,
the

was

one

either

of

choice

difficult

course

standards

even many

of this

Because

course.

of rigour in the

strict

in

viewed
as being one of the most difficult
not
of the abstract conceptsbeing
because
learn,
only
for the first time (e.g., topology, limits, measurability,
of the level of rigour and proofdemanded
also because

faced

often

I gave teaching the


sequence at the University
analysis
the undergraduates
2003.
Among
notes

lecture

the

prospect

undergraduates,
struggling with

material.

this dilemma, I trieda somewhat


unusual
approach
an
i
n
real
sequence
subject. Typically,
introductory
analysisassumes
that
the students
are already familiar with the real
with
mathematical
with elementary calculus,
numbers,
induction,
Faced with

to the

and with the basicsofset theory,


the heart of the subject,for instance
Normally,

students

entering

this

and

sequence

then
the

quickly

launches

into

of a limit.
concept
do indeed have a fair

bit

exposure
prerequisite
topics, though in most casesthe
in a thorough
materialis not covered
For instance, very
manner.
few
students
were able to actually define a real number,or even
an integer,
even though they could visualizethese
properly,
numbers intuitively
them algebraically.
This seemed
and
manipulate
of

to

these

xiv

to

Preface
to

me

student

truly rigorous mathematical proof.


an excellent chance to go backto the

of a

offered

and in particular
of the

of mathematics,
and

with

encounters,

subtleties

course

opportunity. Real analysis is oneofthe first


linear algebra and abstract algebra)that
in which
one truly has to grapplewith
the

a missed

(together

subjects

be

construction

thorough

Thus the course

was

I described

some

well-known

structured

As

such,

the

foundations

the opportunity to doa proper


real numbers.
as

\"paradoxes\"

follows.

In the

first week,

in

analysis,

in which

of limits
and sums,
standard laws of the subject(e.g.,interchange
or sums and integrals)wereappliedin a non-rigorous
way to give
the need
to go
nonsensical results such as 0 = 1. This motivated
even to the very
of the subject,
back to the very beginning
and check all the foundations
definition of
the
natural
from
numbers,
For instance,
one of the first homework
was
scratch.
assignments
was
that
addition
to check (using only the Peanoaxioms)
for
numbers
associative
natural
(i.e., that (a + b) + c = a+ (b + c) for
in the
all
natural
numbers
see Exercise 2.2.1). Thus even
a,b,c:
first week, the students had to write rigorousproofsusing
mathematical
induction.
After
we had derived all the basic properties
of the natural
we
then
moved on to the integers
numbers,
as formal
defined
differences of natural numbers); oncethe
(initially
of the integers, we
all
the
basic properties
students had verified
on to the rationale
moved
(initially defined as formal quotientsof
and
from
then
there
we moved on (via formal
of
limits
integers);
to
the
reals.
Around
the
same
we
covered
time,
Cauchy sequences)
for
the uncountthe basics of set theory,
instance
demonstrating
of
about
reals.
then
ten
the
ability
Only
(after
lectures) did we
of
considers
the heart
begin what one normally
undergraduate real

and
so forth.
analysis - limits,continuity,
differentiability,
The response to this format was quite
In the first
interesting.
few weeks, the students found
the
material
very easy on a
with
as
we
were
the basicproperties
conceptual
level,
dealing only
of the
it
standard
number systems. But on an intellectuallevel
was very challenging,
as one was analyzing
number
these
systems
from a foundational viewpoint, in order to rigorously
the
derive

xv

Preface

facts about thesenumber


primitive ones. One student told me

more advanced
to

explain

friends

his

he was

why

in the

still learning

are either positive,negative,


non-honours sequencewas
and

convergent

conditionally

he thought

this,

how

show

to

zero

or

it was

analysis

sequence

his homework

why

all rational
4.2.4),

(Exercise

series,

more

to
(a)

numbers
while the

absolutely

distinguishing

already

convergent

the

difficult

non-honours real

how

from

systems

and (b)

was significantly

harder

despite

why,

than

that

to me, quite wryly,


of his friends. Another student commented
one
divide
could
that while she could obviously see why
always
into
a positive
to
a
a natural number \316\267
integer q
give quotient
r less than q (Exercise 2.3.5), she still had,
a and a remainder
much difficulty in writing down a proofofthis
to her
frustration,
told
to prove
her
later in the course she would
have
that
fact.
(I
for which it would not be as obvious
see that
the
to
statements
consoled
statements were true; shedidnot seemtobeparticularly
these students greatly enjoyed the
Nevertheless,
by this.)
homework,
an

as

when

intuitive

they did persevere


it solidifed the
fact,

and obtain a rigorousproofof

link in their

minds betweenthe

abstract

of formal
mathematics
and their informal
manipulations
in a very
often
intuition of mathematics (and of the real world),
the
of
the
time
were
task
satisfying way. By
assigned
they
infamous
and delta\" proofs in real analysis, they
\"epsilon
giving the
had
had
so
much
experience with formalizingintuition,
already
in discerning
as the
the subtleties of mathematical logic(such
and
distinction
and the \"there exists\"
between the \"for all\" quantifier

quantifier), that the transitionto these proofs was fairly smooth,


and we were able to cover
both
and rapidly.
material
thoroughly
the
tenth
we
had
with
the
non-honours
week,
class,
By
caught
up
and the studentswereverifying
of variables
the
formula for
change
that
piecewise
Riemann-Stieltjes integrals,and showing
continuous
functions
were
Riemann
the
conclusionof the
integrable.
By
sequencein the twentieth
week, we had covered (both in lecture
and
in homework)
the convergence
theory of Taylor and Fourier

series,the

inverse

and

implicit

function

differentiable functionsof severalvariables,

theorem
and

for continuously
established

the

xvi

Preface

dominated convergencetheoremfor the Lebesgue


integral.
of
In order to cover this much
the key
material,
many
left to the student to prove as homework;
foundational
were
results
this
was an essential aspect of the course, as it ensured
indeed,
the concepts as they
were
the
students
truly appreciated
being
This
has been retained in this text; the majority
introduced.
format

of the exercises
consistof

lemmas,

proving

propositions

and

would strongly recommendthat


as possible - and this includes
many
- if one wishes to use
statements
exercises
\"obvious\"
proving
real
this is not a subject whose
learn
analysis;
are
subtleties
just from passive reading. Most of
easily
appreciated
the chaptersectionshave a number of exercises, which are listed
at the end of the section.
Tothe expert mathematician,
the pace of this book may
seem
in early chapters, as there is a heavy
somewhat
slow, especially
on
emphasis
rigour
(except for those discussionsexplicitlymarked
and
steps that would ordinarily be
justifying
many
\"Informal\,")
theorems

in

the

one doas
those
this text to

text.

main

of

these

Indeed, I
exercises

The first few chapters


of
of the
\"obvious\"
the
develop
properties
(in painful detail) many
standard
number systems, for instance
that
the sum of two
positive real
numbers
is again positive (Exercise 5.4.1), or that given
two distinct
real numbers, one can find
number
rational
any
In these foundational chapters, there
between
them
(Exercise
5.4.5).
- not using later, more
on non-circularity
is alsoan emphasis
advanced
to prove
results
earlier, more primitive ones. In particular,
are
derived
of algebra are not used until they
the
laws
usual
(and
for
the
have
to
be
derived
natural
numbers,
separately
they
and
The reason for this is that it allows
the
rationals,
integers,
reals).
to learn the art of abstract reasoning,deducingtrue facts
students
from
a limited
and
set of assumptions, in the friendly
intuitive
of
for
the
number
this
comeslater,
practice
setting
payoff
systems;
of reasoning
when
one has to utilize the same type
to
techniques
grapple with more advancedconcepts
(e.g.,the Lebesgue
integral).
The text here evolved from
notes on the subject,
lecture
my
and thus is very much oriented
towards
a pedagogical perspecquickly

passed

over

as being

self-evident.

xvu

Preface
tive; much of the key
in many cases I have

material

is

inside

contained

exercises, and

to give a lengthy and tedious, but


slick
abstract proof. In moreadvanced
proof
instructive,
will
see shorter and more conceptually
the student
textbooks,
this
and with more emphasison
of
treatments
material,
coherent
I
on
feel it is important to know
than
however,
rigour;
intuition
in order to truly
to do analysis
first,
how
rigorously and \"by hand\"
the
more
intuitive
and
to
abstract
modern,
approach
appreciate
chosen

of a

instead

level and beyond.


analysis that one usesat the graduate
and
The exposition in this book heavily
emphasizes
rigour
this
does
however
m
ean
that
not
lectures
formalism;
necessarily
in my
based on this book have to proceedthe sameway.
Indeed,
to present
the intuition
own teaching I have used the lecturetime
informal
and giving
behind the concepts (drawingmany
pictures
to the
viewpoint
examples), thus providing a complementary
in the text.
The exercises assigned as homework
formal presentation
an essential
bridge between the two, requiringthe student
provide
both
and formal understanding together in
intuition
combine
to
correct proofs for a problem. This I found
to locate
to
be
order
as it requires the subject
the most difficult task for the students,
rather
than
to be genuinely
learnt^
merely memorized or vaguely

I received
from
absorbed. Nevertheless,the feedback
the students
for this
was that the homework, while very
reason,
demanding
was also very rewarding, as it allowed
to connect
them
the rather
their
abstract manipulations of formal mathematicswith
innate
on such basic concepts as numbers,sets, and
intuition
functions.
Of course,
the aid of a good teaching assistantis invaluable
in
achieving

this

connection.

With regard

to examinationsfor

on this

text,
examination
with problems similarto the exercises
in the text (but
given
with
no
unusual
perhaps
shorter,
trickery involved), or else a takehome
examination
that involves problems comparable to the more
in the text.
The subject matter is too vast
intricate
exercises
to
force
to memorize the definitions and theorems,so
the students
I would not recommend
a closed-book
examination, or an exami-

I would

recommend

either

an open-book,

course

based

open-notes

xviii

Preface

extracts

on regurgitating

based

nation

book.

the

from

in

(Indeed,

examinations I gave a supplemental


sheet listing the key
were
to the examination
relevant
definitions and theoremswhich
similar
to
examinations
the homework
the
problems.) Making
in the
course
will also help motivate the students to work
assigned
and
understand
their homework problems as thoroughly
through
as possible(as opposedto, say, using flash cards or other such
memorize
which is good preparation not only
devices
to
material),
for
examinations
but for doing mathematics in general.
my own

Someof the material


as are the number

to analysis

(Chapters 3,
less rigour, or be
on logic and the

as

given

decimal

course

concurrently

16 Fourier
be omitted.
(on

series)

of

reasons

For

system

thirty lectures

the

if

abridged. The second


other

from

who

It also

sources.

at

takes about

feedback.I
referees

who

made

am

also

several

improvementsto

the

appendices

or

be

two

covered

is omitted

or

to the first, but can


a first course in analysis

times

thirty

who

several

to

lectures

over

errors

cover.

the progression
in the lectures

is derived, and gave other valuable


to the many anonymous
grateful

text

this

which

from

can

material

have had

I am deeply indebted to my students,


of the real analysis
course
corrected

notes

The

as optional

has been split into

peripheral
refers

volume

students

to

assignments.
intended

is slightly longer,but

first volume

The

in about

taught

chapters

textbook

this

length,

volumes.

alsobe

time

fundamental

probably not be coveredin the main


on logic is particularly suitable for
appendix
with the first few chapters. Also, Chapter
is not needed elsewhere in the text and can

the

lectures;

are

of

reasons

as

on set theory
and with substantially

the

more quickly

reading

not

would

and

reading

supplemental

is

theory

systems,

covered

be

can

8)

reading

as set

For instance,

constraints.

is somewhat

be omitted for

and may

theme

main

the

to

textbook

this

in

peripheral

very

corrections

and suggested many

important

text.

Terence

Tao

Chapter

Introduction

1.1 What is analysis?


the

analysis:
real

is

analysis
and

numbers,

complex analysis, which concerns the analysis


numbers
and complex functions, harmonicanalysis,

distinct

the

from,

complex

the

concerns

which

how

and

waves,

they

on functions

(and

forth.

is the

Analysis
on

and

trying

quantitative

collection

of

which
form

they

much more heavily


like vector spaces), and so

focuses

things

such objects, with a


and
precisely
accuratelythe qualitative
of these objects. Real analysis
is the
which underlies calculus, which is the
which one uses to manipulate
algorithms
study of

rigourous

to pin down
behavior

foundation

theoretical

(waves) such as sine


synthesize other functions via the Fourier

analysis,
how

of

of harmonics

analysis

functional

transform,

focus

introduction to real
undergraduate
of the real numbers, sequences and seriesof
real-valued
functions.
This is related to, but

honours-level

an

is

text

This

computational

functions.

In
familiar

this
to

we will be

text
you

from

limits,

functions,

definite

already have a great deal


objects; however here we
theory for these objects.
as

the

following:

studying many objects which will be


calculus: numbers, sequences, series,
and so forth. You
derivatives,
integrals,

freshman

of
will

experience
be

We will

focused

of computing

more on

be concerned

with

these

the underlying

with questionssuch

1. Introduction

1.

real

is

What

0, what

After

number?
is the \"next\"

smallest positivereal
into piecesinfinitely

as 2

have

not? If

rationals,

rational
2.

square
are

there

Can

number)?

many

times?

root,

while
many
are

there

a largest real number?


(i.e., what is the
cut a real number

number

real

infinitely

come

how

Is there

you

does a number such


a number such as -2 does
reals and infinitely many
\"more\" real numbers than
Why

numbers?

take the limit of a sequenceof real numbers?


Which
have limits and which ones don't? If you
sequences
can
does
a
this mean
sequence from escaping to infinity,
stop
that it must eventually settle down
and
Can you
converge?
add infinitely many real numbers togetherand
a
still
get
finite real number? Can you
add
rational
infinitely
many
numbers together and end up with a non-rational
number?
If you rearrange the elements
of an infinite sum, is the sum
How

do you

still the same?

3.

What

is

a function?

What

does it mean

for

a function

to be

differentiable? integrable? bounded? can you


infinitely
many functions
together? What about taking
limitsof sequences
of functions?
an
Can you differentiate
infinite series of functions? What
If
a
about
integrating?
= 0 and 5 when
function f(x) takes the value
\317\207
3 when
= 1
\317\207
to take
every
(i.e., /(0) = 3 and /(1) = 5),doesit have
intermediate
value between 3 and 5 when \317\207
between
0
goes
and 1? Why?
continuous?
add

You may already know

how

to

answer

some of

these questions

most likely these sorts of issues


to those courses; the emphasis
only
importance
was
on
getting
you to perform computations, such as computing
= 1. But now that
= 0 to \317\207
the
of rrsin(a:2) from \317\207
integral
you
are comfortablewith these objects and already know how to do all
the
we will go back to the theory and try
to
computations,
really
understand
what is going on.
from

were

your

calculus

of secondary

classes,

but

1.2.

do

Why

1.2

do analysis?

Why

It is a

analysis?

fair

to

question

ask,

\"why

knowing why

to know how

only needs

calculus
adequate

for

to begin

you

economics,

biology,

or whatever elseyou
like

things

chain

the

work

things

end

to do

in introductory

one

that

argue

real-life problems. The


classes is certainly

many problems in physics,


science, finance, engineering,
computer
- and you can certainly use
up doing

solving

rule, or integrationby

rule, L'Hopital's

there
work, or
knowing
to these rules. However,onecan get into
exceptions

why these rules

without

to

comes
in

satisfaction

pragmatic person may

receive

you

training

chemistry,

but a

work,

things

it

when

bother?\",

analysis. There is a certainphilosophical

any
if one

trouble

from
came
rules without knowing where they
applies
the limits of their applicability are. Let me give some
in which several of these familiar rules, if applied
blindly
of
the
can
lead
to
knowledge
underlying analysis,

parts

are

whether

and

what

examples
without

disaster.

Example1.2.1

to

you:

the

when

=
\321\201

if one

blindly

0.

which

1.2.2
series

work
but

true,

case it was obvious that one was


other cases it can be more
hidden.

geometric

very familiar one

law
cancellation
=> a = does not
For instance, the identity 1x0 = 2x0 is
is
cancels the 0 then one obtains1 = 2,

ac = be

In this

Example

is a

This

zero).

by

(Division

such

dividing

(Divergent
series). You
as the infinite sum

by

zero;

false.

but in

have probably seen

trick
to sum this series: if
You have probably seenthe following
we call the above
if we multiply both sides by 2, we
sum
5, then

obtain

2 + 1+

25 =
and

hence

S =

^
2

74 +

2, so the seriessums

the same trick to the

5=1

^8 +
to

... = 2 + 5
2.

However,

series

+ 2

+ 4 + 8 +16 +

...

if you

apply

one

Introduction

1.

results:

gets nonsensical

25 = 2 + 4 + 8 + 16+

...

same reasoningthat shows


that 1 + 2 + 4 + 8 + ... =

So the
gives
first

not

but

equation

the series

5=

S-1

=\302\273

1-1 + 1-1

similar

1-1

-l.

+ ...

we trust

that

is it

\342\200\2241.
Why

the second?

| +j

l +

that

arises

example

also

the
with

...;

we can write

5=1- (1-1
and hence

S = (1- 1)

and hence

5 =

that

5 = 1+ (-1
and hence
an

limit

1)

one is

Which

1.2.3

0 +

0 +

...

... = 1

0 +

0 +

...

correct? (See Exercise7.2.1

previous

variation
(Divergent sequences). Hereis a slight
and
L be the
Let
\320\266
be
a
real
let
number,
example.

=
Changing variables \316\267

m +

lim xm+1=

L=

m+l\342\200\224\321\216\320\276

if m

+ 1

oo,

lim

xm =

\342\200\224\342\226\272
then

m+1\342\200\224>oo

and

+ 1) +

(-1

But

(1 - 1) + ... =

1) +

answer.)

Example
of the

1-5

instead we can write

0; or

5=1.

that

for

(1

...) =

instead we can write

1/2; or

5 =

that

1-1

Urn xn.

n->oo

1, we have
lim

\317\207
\317\207
xm

\317\207 lim

m+1\342\200\224>oo

\342\200\224>

oo,

thus

lim xm = lim
m\342\200\224>oo

thus

rrL =

L.

xm.

m+1\342\200\224>oo

xn = L,

n\342\200\224>oo

1.2.

do

Why

analysis?

At this point we could cancelthe


which
for an arbitrary real number\320\266,

already aware of the division

smarter
particular

lim xn

we are

=
\317\207

either

that

shown

have

to

seem

we

=
\317\207

we could be a little
1, or L = 0. In

problem,

that

instead

conclude

and

is

zero

by

conclude that
absurd.
But since

and

L's

= 0 for all \317\207\317\206\\.

\320\277-\321\216\320\276

instance

for
\320\266,

specializing

by
the

that

conclude

specializing

1,-1,1,-1,... also
is
to be absurd;

=
\317\207

\342\200\224

the

what

Exercise6.3.4

an

for

with the

problem

appear

above argument? (See

answer.)

the
(Limiting values of functions). Start with
=
make
the change of variable \317\207 \321\203
+ \317\200
\320\230\321\202\320\266_,\320\276\320\276
sin(rr),

expression

sin(y +

that

recall

lim sin(a:)

lim

\342\200\224

sin(y+

lim

\342\200\224

(\342\200\224sin(y))
\320\243\342\200\224\320\256\320\236

thus

we

\320\230\321\202\321\205_\321\216\320\276
sin(y)
sin(a:)
lim^-,\316\277\316\277

lim sin(rr)

obtain

to

sin(y)

\317\200)

\317\200)
X\342\200\224\320\256\320\236 1/+7\320\223\342\200\224\320\256\320\236

Since

conclusions

1.2.4

Example
and

These

zero.

to

converges

of

values

could

2 we

conclude

we

certain

converges to zero, and by


that the sequence

1,2,4,8,...

sequence
the case

to

if we apply it to
=
to the case \317\207

is absurd

conclusion

this

But

\342\200\224

lim

lim

sin(y).

\321\203\342\200\224\320\256\320\236

have

sin(a:)

x\342\200\224\321\216\320\276

x\342\200\224\342\226\272po

hence

and

lim

sin(a:) =

0.

X\342\200\224\320\256\320\236

If

make

then

we

=
the change of variables \317\207
we conclude that
cos(^)

\342\200\224
and
\316\266
\317\200/2

recall

\342\200\224
=
8\316\220\316\267(\317\200/2
2)

that

lim cos(rr) =

0.

i-\302\273oo

Squaring

both

of these

lim

(sin2(:z)
x\342\200\224>oo

On

the

other

hand,

limits and

adding we seethat
= 02

+ cos2(:z))

we have

sin2(\320\266)+

we have shown that 1 = 0! What

is

the

cos2

+ 02 = 0.
=

(\320\266)

difficulty

1 for

here?

all x.

Thus

1. Introduction

Example 1.2.5
fact of

arithmetic. Considerany

the

Consider

sums).

(Interchanging

of numbers,

matrix

following

e.g.

2 3\\
4 5 6
\\ 7 8 9 /
/1

and compute

the

then

and

columns,

In both

sums.

of all

rows and

the

row sumsand totalall the column


- the
total sum
will get the samenumber

matrix:

the

9/

\\

15

24

12 15 18
To

put

matrix
first or sum

way, if you want to add all the


together, it doesn't matter whetheryou

the columnsfirst, you

of computers,
(Beforethe invention
would use this fact to guard
against
their books.) In seriesnotation,
m

i=l

if a,ij denoted
infinite

45

it another

mxn

Now

the sums of all the

all the

total

cases you

the entries in

all

of

sums

one

errors

fact

would

in

sum

the same

the

an

rows

answer.

and book-keepers
when balancing
be expressed

as

\316\267\317\204\316\267

j=l

j=l

that

up

making
this

\316\267

think

with

accountants

the entry in the ith


might

end

entries

i=l
and

row

this rule

column of the matrix.


should extend easily to

jth

series:

oo

oo

oo

oo

sum

\316\243\316\243\316\261^'\302\267

i=l

j=l

j=l

i=l

infinite seriesa lot in your work, you will find


switch summationslikethis fairly
often.
of saying
this fact is that in an infinite
the
way
matrix,
of the row-totals should equal the sum of the column-totals.

Indeed, if you use


yourself having to
Another

\316\243\316\243\316\261^'

1.2.

do

Why

analysis?

is

despite the reasonableness of this statement,it


Here
is a counterexample:

However,
false!
actually

-1

...

...

...

-1

0
0
0

...

0
0

-1

...

0 -1

\\

rows, and then add up all the row totals,


up all the columns, and add up all the
0!
column
totals, you get
So, doesthis mean that summations
for infinite series should not be swapped,and that any
argument
a
such
should
be
distrusted?
8.2.2
Theorem
using
swapping
(See
If

sum

you

all the

up

you get 1; but

for

if

you

sum

an answer.)

1.2.6 (Interchanging integrals). The interchanging


of
trick
which
occurs
in
as
mathematics
integrals
just commonly
of sums. Suppose one wants
to
the
as the
interchanging
compute
=
of
volume under a surface \316\266
us
the
limits
ignore
/(\320\266,\321\203)
(let
for
the
can
it
One
do
integration
by slicing parallel
moment).
to the \320\266-axis: for each fixed value of y, we can computean area
variable
to
/ /(\320\266,\321\203)dx, and then we integrate the area in the \321\203
Example

is a

the volume

obtain

v = /

Or

could

we

compute

an

area

slice

to the

parallel

J f(x,y)

/ f(x,y)dxdy.
y-axis

each

for

fixed

dy, and then integratein

\317\207
and

the

\320\266-axis

to

obtain

V =

This seems to suggest that


integral signs:

f(x,y)dydx.

one

should

/ / f(x,y) dxdy= / /

always

f(x,

y)

be

dydx.

able

to swap

Introduction

1.

all the time, because


signs
to integrate in first than the other.
sums sometimes cannot be swapped,
as infinite
just
to swap. An example is with
also sometimes
dangerous
\342\200\224
e~xy
we believe that we can swap
Suppose
xye~xy.

And indeed,
However,

are

integrals

people swap

the

integrand

the

integrals:

pi

poo

Jo

Jo

xye~xy)

(e~xy

integral

is easier

variable

one

sometimes

dydx

pi poo

/ (e~xy
Jo Jo

dx

xye~xy)

dy.

Since
(\320\265-*\320\243
xye-*\302\273)

dy

e~x
dx
side is /0\302\260\302\260

J
the left-hand

_
(e-\302\253w

Jo
/o

\342\200\224
e~x|o\302\260

dx =

xye-xv)

the right-hand side is J0 0 dx = 0.


find
error somewhere; but you won't

step where we
to

when
for

partial

plausible

0,

one

1 \317\206
0,

so

anywhere
So how

is an
in
the
except

integrals.

(Interchanging limits). Supposewe


looking statement
x->0

we

xe-xv\\xxZ$>

Clearly

since

1. But

there

answer.)

x2
lim lim -=
But

1.2.7

Example
the

the interchange

trust

e~x,

do we know
of integrals? (SeeTheorem19.5.1

the

interchanged

\321\203\320\265-*\320\243\\|g

y->0

^
\316\247\316\224
+

lim -o

lim

x2

o-

+ \321\2031
2/->0\317\207->0\316\247\316\224

yA

start

with

(1-1)

have

x2

x2

so the left-handsideof (1.1)is 1;on

_^ x2
\320\231\320\276\321\2172
+ 2/2

the

other

hand,

we have

02

02 +

~\302\260'
\321\2032

side of (1.1) is 0. Since1 is clearly


not
equal
this suggests that interchangeof limitsis untrustworthy.
But
are there
any other circumstances in which the interchange
of limitsis legitimate?
Exercise
13.2.9 for a partial answer.)
(See
so

the

to zero,

right-hand

1.2.

do

Why

the
limits, again). Consider

1.2.8 (Interchanging

Example

statement

looking

plausible

analysis?

lim xn

lim

= lim lim xn

71\342\200\224\320\256\320\236
X\342\200\224\302\273171\342\200\224\320\256\320\236\321\205_>1-

notation

the

where

\342\200\224>
1~
\317\207

means

that

\320\266
is

1 from

approaching

is to the left of 1, then limn_>oo


xn = 0, and hence
the left. When \317\207
= 1 for
is
zero.
But
we
also
have
side
the left-hand
\316\227\317\200^^-xn
limit
side
is 1. Does this demonstrate
all n, and so the right-hand
of
limit
is
this
that
interchange
always untrustworthy? (See
type

Proposition14.3.3

for

an

limits and integrals).Forany

1.2.9

Example
number

y,

(Interchanging
we have

00

dx

j_ool

(x-y)2-

limits

Taking

answer.)

as

should

\342\200\224\342\226\272
we
\321\203 oo,

f\302\260\302\260

lim

cte

\342\200\224-\317\201
\316\2232
1
2/-00
\320\243_\320\276\320\276
(\320\266 \321\203)2

But

for every

- yJI^.^
= -\317\200
*\320\273*=-\320\276\320\276
2

arctan(:z

\342\204\242v-

we
\320\266,

have

lim

\317\200

\316\275
(-77)

\317\200.

obtain
\320\22350
1
\342\200\224\342\200\224

/
2/-\320\276\320\276

\321\203^^

liniy-^o

real

1+/\321\2171_

1 +

(\317\207

r^

dx

\317\200.

yy

= \320\236.So
\321\207\320\264

we

seem

to

What
was
the problem
with the
have concluded that 0 =
one
abandon
the (very useful) technique
above argument?
14.6.1
for a
of interchanging limits and integrals? (See
\317\200.

Should

Theorem

answer.)

partial

Example

1.2.10 (Interchanginglimitsand

that

0, then

if

\316\265
>

_d_
dx

and

in

particular

\317\2073

+
\\\316\2652

that

\\

=
\"
\317\2072)

derivatives).

3\317\2072(\316\2652+\317\2072)-2\317\2074
(\316\2652+\317\2072)2

Observe

1. Introduction

10

Takinglimitsas

\342\200\224>
\316\265
0, one

might

^ {oT^J
the

But

side is

right-hand
14.7.1

-^x = 1.

f(x,y)

interchange

two

\302\260-

Doesthis

is

it

that

mean

derivatives?

(See

for an

Example 1.2.11 (Interchanging


3
function

u=0

to interchange limits and


answer.)

illegitimate

always
Theorem

expect that

then

:=

x%+ji

maneuvre

be

/(\320\266,\321\203)

in analysis

the

is to

one expects

thus

derivatives,

partial

Let1

derivatives).

common

\302\267
A

\320\2602/(\302\260>0)=<\320\271-(\302\260>0)\302\267
\320\264\321\205\320\264\321\203
\320\264\321\203\320\264\321\205

the

from

But

rule we have

quotient

df,
V
in

and

2xy4

3xy2

'

\321\2052
+ \320\2432

(\321\2052+\320\2432)2

particular

ar
\317\207\316\224

ay
Thus

\316\2642/-(\316\277,\316\277)\316\277.
\320\264\321\205\320\264\321\203

the

On

other

hand,

from the

quotient rule again we

df.

have

2aV

y3

\320\2662
\320\264\321\205\320\272,\321\203'
+ \321\2032

(x2

+ y2)2

and hence
\317\207 \320\2433

df(n

xOne

might

set /(0,0)
for all
entiable
continuous and
if we

object

that

function

this

:= (0,0) then
and
(\320\266,\321\203),
differentiable

this

in fact
for

all

\320\236

is not

function

defined at (x,y) = (0,0),but


continuous and differ-

becomes

both partial derivatives


(\320\266,\321\203)!

are

|\302\243,|\302\243

also

11

do analysis?

Why

1.2-

Thus

0,
Since 1 \317\206
derivatives

is

which

the

have shown that interchangeof


are there any other circumstancesin
of derivatives is legitimate? (See Theorem

seem to

thus

we

But

untrustworthy.

interchange

17.5.4

for some

17.5.1

Exercise

and

answers.)

Example 1.2.12 (L'Hopital'srule). We


beautifully simple L'Hopital'srule

lim m

lim

all

are

with the

familiar

/\316\256,
gf(x)'

\317\207->\317\207\316\277
\317\207->\317\207\316\277
g(x)

get led to incorrectconclusions


For
it to f(x)
instance, applying
incorrectly.
and
xo := 0 we would obtain
but

one

can still

if

:=

it
applies
:= 1 + \320\266,

one

\320\266,
g(x)

= lim - = 1,
x->o
1

lim

\317\207->\316\277
1 + \317\207

=
= 0\302\267
incorrect answer, sincelimx_>o
j^q
j^
all that is going on here is that L'Hopital'srule
is
Of
course,
\342\200\224\342\226\272
as
\317\207
when
both
and
to
zero
\320\266\320\276>
applicable
only
g(x) go
f(x)
was violated in the above example. But even
which
a condition
\342\200\224> there
and g(x) do go to zero as \317\207
is still a
when
xq
f(x)
For instance, consider the
conclusion.
possibility for an incorrect
limit
but

this

is the

hm

x2sm(x~4)
-.

x->0
Both

pretty safe to
hm

go to

denominator

and

numerator

rule,

L'Hopital's

apply

x2 sin(rr~4)

_.

= hm

2x

lim 2a:sin(:E~4)

x\342\200\224>0

The
function

first

limit

2xsin(a:~4)

converges
is

to zero

bounded

it seems

\342\200\224
4\320\266~3cos(rr~4)
\342\200\224

sin(rr~4)

\317\207\342\200\224\342\226\272\316\277
x\342\200\224>o
\317\207

\342\200\224>
zero as \317\207
0, so

to obtain

\342\200\224

lim

4x~3cos(a:~4).

x\342\200\224\302\2730

test(since
by the squeeze

above

by

2\\x\\

and

below

the

by

\342\200\224

2|sc|,

both

x~3

(because

does not
diverges.
limx_>o

S1

\342\200\224-

also

g
limx_>o

squeeze

test again.

goes

to

\342\200\224\342\226\272
\317\207
and
0,

as

infinity

it

presented

2a;sin(a;\"4)-^\"2cos(a;\"4)
rule
that
L'Hopital's

that

rule

L'Hopital's

is

rigourous; see Section 10.5), but

when applied.
and

(Limits

how

was

goes

is quite

some care

requires

cos(rr~4)

we can clearly rewrite this


\342\200\224\342\226\272
to zero when \317\207
0 by the

however

This doesnot show


it

second limit is

the

But

0).

which

:rsin(:E~4),

Example 1.2.13
integration and
curve

at

diverges;

(indeed,

untrustworthy

probably

zero

go to zero). So the limit limx^0


One might then concludeusing

limit as

still

to

of which go

divergent

it

Introduction

1.

12

relates

When

lengths).

to the

you learn

area under a

about

curve, you

picture in which the areaunder


a
bunch
of rectangles,whose
area
by

were

with some

approximated

the

was

\"took
limits\"
to
Riemann sum, and then onesomehow
an
which
then
integral,
replace that Riemann sum with
area
under
the
matched
actual
curve.
the
Perhaps a little
presumably
a
how to compute the length of a curve
learnt
later,
by
you
a
similar method
the
curve
bunchof
line
approximate
segments,
by
of all the line segments, then takelimits
the
compute
length
again
to see what you get.

given

by a

However, it should comeas no surprise


approach

also

can

lead

by

if used

to nonsense

the right-angledtriangle
vertices
(0,0),
suppose we wanted to compute length
this triangle. Pythagoras' theoremtells

(1,0),
of the

the

about Pythagoras'
using calculusmethods.
horizontal
the
by
theorem,

some
and

that

reason
wanted

to

and (0,1),
hypotenuse
this

that

us

but suppose for

this

incorrectly. Consider

with

has length \\/2,

that

now

of

hypotenuse

we did

compute

and

not know

the length

to approximate
and vertical
hypotenuse
edges. Pick a large
and
the
number
JV,
approximate
hypotenuse by a \"staircase\"
of N
horizontal
N
consisting
edges of equal length, alternating with
vertical
edges of equal length. Clearly these edgesall have
length
= 2. If one takes
is 2N/N
1/iV, so the total length of the staircase
the
limits as N goesto infinity,
staircase
the
clearly
approaches
and
so
in
the
limit
we
should
the
of
hypotenuse,
get
length the
Well,

one

way

to do

so is

1.2.

do

Why

the limit of 2N/N is 2, not \320\273/2,


oo,
However, as N \342\200\224\342\226\272
an incorrect value for the length
of the hypotenuse.

hypotenuse.
so

we

13

analysis?

have

How did this happen?

The

resolve these
you
(and others)
and
when
arejustified,
they are illegal, thus separating the
of these
rules from the nonsense. Thus they
can
useful applications
these
place
you from making mistakes, and can help you
prevent
learn
rules in a wider context. Moreover,as you
analysis
you
of thinking\",
which will help you
will develop an \"analyticalway
wheneveryou come into contact with any new rules of
with situations which are not quite covered
or when
dealing
mathematics,

questions,and

by

the

know

let

will

standard

in this

learn

you

analysis

For instance,

rules,

instead of

complex-valued

sphere instead of the plane?


not continuous, but are insteadthings

limits

of

are

summation,

senseof
how

if your

What

to

why

it to

adapt

are; this will

learnt more

confidently

like

functions,

occasionally

to

apply
and

functions

your

are

are

working
are
if your
functions
waves and delta
square
if

What

What

you

of integration, or
You will developa
(e.g., the chain rule) works,

infinite?

new situations, and what

allow you

if

or limits

in mathematics

rule

what

real-valued?

on the

functions?

text will help you


when these rules

the

correctly.

its

mathematics

limitations

you have

(if any)

already

Chapter

Starting

at the

the natural

beginning:

numbers

in high
have
learnt
this
we will review the material you
text,
school and in elementary calculusclasses,but as rigourously
as
To do so we will have
to begin
at the very basics possible.
indeed,we will go back to the concept of numbers and what their
are.
dealt
with
for over
Of course, you have
numbers
properties
ten years and you know how to manipulate
the rules of algebra
to simplify
but we will now
any
numbers,
expression
involving
In

turn to a more

fundamental

which

issue,

is:

why

do

the

rules

of

is it true
that a(b + c)
algebra work at all? For instance,why
is equal to ab + ac for any three numbers \320\260,
This
is not an
6, \321\201?
from
more
arbitrary choice of rule; it can be proven
primitive,
and more fundamental, properties of the number
This
system.
will teach you a new skill how
to
complicated
prove
properties
from simpler ones. You will find that even though a statement
be
it may not be easy to prove;
the
material
here
\"obvious\",
may
will give you plenty of practice in doing
in the process
so, and
will lead you to think about why an obvious statement really is
obvious.One skillin particularthat you will pick up here is the
use of mathematical
which
is a basic tool in proving
induction,

things in many
So in the

areas

first

few

of mathematics.

numbersystemsthat are
of sophistication, they

we will

chapters

are

used

the

in

real

natural

you with various


In increasing order
analysis.
numbers N; the integers Z;
re-acquaint

15
numbers R. (Thereareother
number
such as the complex numbers C, but
we
will
not study
systems
are
them until Section 15.6.) The natural numbers
{0,1,2,...}
of
the
number
but
are
used
to
most
the
primitive
systems,
they
are
used
to build the rationals.
build the integers, which in turn
the
rationals
are
used
to
build
the real numbers,
Furthermore,
to
the
numbers.
Thus to
which are in turn used build
complex
the
we
must
look
at
the
natural
numbers.
at
very
beginning,
begin
the following
consider
will
We
question: how does one actually
natural
numbers?
the
different
question
(This is a very
define
is something
from how to use the natural numbers,which
you of
to
do
well.
It's
like
the
know
how
difference
between
course
very
knowinghow to use, say, a computer, versus knowing how to build
rationals

the

that

Q,

computer.)

This question
basic

the real

and

problem

is moredifficult

is that

it looks.

than

answer

to

you have used

the natural numbersfor

The
so

long

mathematical
that they are embedded deeply into your
thinking,
and you can make various implicitassumptionsabout these
that
a + b is always equal to b + a) without
even aware
numbers (e.g.,

that you are doing so; it

this number system as if


in

what

I will

follows

is

it

is the

have to

task: try to set aside,for

the

Furthermore,

your

first time

ask you
moment,

go and

let

to

difficult

to

try to

inspect

you have seen it. So

perform

everything

difficult
you know about
to count, to add,
a rather

the natural numbers;forgetthat you know how


the rules of algebra, etc. We will try to
to multiply,
to manipulate
what
introduce these concepts one at a timeand identify
explicitly
our assumptions are as we goalong- and not allow ourselves
to use
more \"advanced\"
tricks
such
as the rules of algebra until we have
them.
This
actually
proven
may seem likean irritatingconstraint,
a
which
as
we
will
lot of time proving
statements
especially
spend
are \"obvious\", but it is necessary to dothis suspension
of known
facts to avoid circularity (e.g., using an advanced
a
fact
to prove
more elementary fact, and then
the elementary
fact to
later
using
this
exercise
will be an excellent
prove the advancedfact). Also,
to
affirm
the foundations
of your mathematical knowledge.
way
practicing

proofs

and

abstract

thinking

here

2.

16

The

numbers

natural

advanced
be invaluable when we move on to more
concepts,
as real numbers, functions, sequencesand series,
and so forth. In short, the resultshere
differentials and
may
integrals,

will

such

than
the
journey is much moreimportant
the
number
are
constructed
systems
destination,
(Once
we can resume
using the laws of algebra etc. without
properly,
them each time.)
to rederive
having
that we know the decimalsystem, which
We
will
also
forget
is an extremely convenient way to manipulatenumbers,
of course
which is fundamental to what
numbers
are.
but
it is not something
of
one could use an octalor binary
instead
system
(For instance,
numeral
and
still
the decimal system, or eventhe Roman
system,
if one tries to fully
Besides,
get exactly the same setof numbers.)
as natural
is, it isn't
system
explain what the decimalnumber
the
is
00423
same
numberas
as you might think.
423,but
Why
is 123.4444...
a real
32400
isn't
the same number as 324? Why
...
444.321
i
s
not?
And
do
we
have
to
while
number,
why
carry
is 0.999...
of digits when adding or multiplying?
the same
Why
number as 1? What
is the
smallest
positive real number? Isn't it
these
we will not try
0.00...
001?
aside
Sotoset
just
problems,
of the decimal
to assumeany knowledge
system, though we will
of course
still refer to numbers by their familiar namessuch
as
etc.
other
or
instead
of
notation
such
as
1,2,3,
using
\316\231,\316\231\316\231,\316\231\316\240
0++,
(0++)++,
(see below) so as not to be needlessly
((0-H-)++)++
artificial.
For
we review the decimal system in an
completeness,

seem

Appendix

(\302\247B).

We now
in

Guiseppe

axioms

Peano

The

2.1

numbers,

but the
for now.

trivial,

present one standardway


of the

terms

Peano (1858-1932).

that

five

set.

elements

We

and

the

another

for

instance

only
approach

natural
out

laid

way

by

to define
is to

one could

talk

take a

5 to be the numberof elements


in
this alternate approachin Section
3.6.

define

shall discuss

the

which were first

This is not

the natural numbers. For instance,


about the cardinalityof finite
sets,

set of

define

to

Peano axioms,

stick

shall

we

However,

17

axioms

The Peano

2.1.

with

the Peano

axiomatic approach for

now.

are

How

could

we

Informally,

we to

define what the natural numbers

say

2.1.1. (Informal) A

Definition

are?

the set

natural

number

is any

element of

N:= {0,1,2,3,4,...},

with
0 and
all the numbers created by starting
N
the
forward
We
call
set
natural
then counting
of
indefinitely.
the

is

which

set of

numbers.

Remark 2.1.2.In
of

instead

0,. but

some

this is

texts

the

natural

numbers

a matter of notationalconvention

start at
more

1
than

as
this text we shall referto the set {1,2,3,...}
Z+
the
numbers.
Natural
rather
than
natural
integers
as whole
numbers.
numbers are sometimes also known
anything

else. In

the positive

the
sense, this definition solvesthe problemof what
are:
a natural
number is any element of the set1
numbers
natural
it is not
that
N. However,
really
satisfactory, because it begs the
what
N
This
of
is.
definition of \"start at 0 and count
question
it
seems like an intuitive enough definition of N, but
indefinitely\"
it
because
leaves
is not entirely
questions
acceptable,
many
For
instance:
how do we know we can keepcounting
unanswered.
without
back to 0? Also, how
do you
indefinitely,
cycling
perform
such
as
o
r
addition, multiplication, exponentiation?
operations
can answer
We
the latter question first: we can define
in terms
of simpler operations. Exponentiation
complicated operations
is nothingmorethan repeated
53 is nothing more
multiplication:
than three fives multiplied
is nothing
together.
Multiplication
more than repeated addition;5 \317\207
3 is nothing
more than three
fives
added
and division will not be
together.
(Subtraction
covered
because
are
not
here,
operations which are well-suited
they

In a

Strictly

we have not
rest of this

speaking,
yet
chapter

is another problem with


this
informal
definition:
what a \"set\" is, or what \"element of\" is. Thus for the
we shall avoid mention of sets and their
elements
as much
there

defined

as possible,except in

informal

discussion.

18
will
to the natural numbers;
And
and rationale,
than the repeatedoperationof

have

they

respectively.)

2.

The natural

numbers

to

wait

for the

integers

you

three

add

to five,

what

or

forward,

counting

If
incrementing.

are

you

more

is nothing

It

addition?

is incrementing
seems
to be

doing

five three times. On the other hand,


incrementing
to any simpler
a fundamental operation, not reducible
it is the first operation one learnson numbers,
indeed,
to add.
learning

operation;
before

even

Thus, to define the natural numbers,we will use two


the zero number
0, and the incrementoperation.
concepts:
In deference
to
modern
we will use n++ to
languages,
computer
= 4,
or successor
of n, thus for instance
denotethe increment
3++
different
from that in
usage
(3++)++ = 5, etc. This is a slightly
fundamental

n++
actually
redefines the
computer languages such as C, where
to be its successor;
however in mathematics we try not
value of \316\267
a variable more than oncein any given setting,
as it can
to define
of the statements
which were true
often lead to confusion;
many
and vice
for
the
old value
of the variable can now
become
false,

versa.

So, it seems like we

to

want

N consists of 0

that

say

everything which can be obtained


should consist of the objects

from

incrementing:

by

0,0++,(0++)++,
start

numbers,

we

concerning

writing down what this meansabout


see that we should have the following

and

the

increment

Axiom 2.1. 0 is a
Axiom

the

thus
0

2.2. If

\316\267
is

etc.

((0++)++)++,

If we

and

natural
axioms

++:

operation
number.

natural

a natural

then n-H-

number,

is alsoa natural

number.

Thus

for

instance,

2.2, we see that


this notation will begin

Axiom

to write these numbers

from Axiom

2.1 and

(0++)++ is a

to get unwieldy,
in

more

familiar

number.

natural

so

two

we

adopt

notation:

of

applications

Of

course,

a convention

19

The Peano axioms

2.1.

Definition2.1.3.
number

the

(0++)++,

3 to

1 :=

0++, 2

words,

other

use

\342\204\242

Thus

for

be the

:= 1++, 3 :=

the statement

denote

2/\" to

\"x

be the number 0++, 2 to be


number ((0++)++)++, etc. (In

1 to

define

We

we

instance,

2++,etc.

that

\317\207
is

defined

In

this

text

to equal

y.)

have

2.1.4. 3 is a natural number.

Proposition

Axiom
2.1, 0 is a natural number. By Axiom
2.2,
=
a
number.
Axiom
2.2
is
natural
2
1++
0-HBy
again,
=
2.2 again,
3 is a natural
2++
is a natural number.By Axiom
By

Proof.

= 1

number.

It
numbers.

we

However,

is enough
not pinned

this

that

seem

may

have

to describe the natural


down completely the behavior

ofN:

Consider a number system which consistsofthe


in which the increment operation wraps back

2.1.5.

Example
numbers

0,1,2,3,

isequal to 2,
precisely 0++ is equal to 1, 1-Hbut
to
is
to
0
also
to 4, by
3++ equal
2-H- is equal
3,
equal
(and
of thing
in real life,
definition of 4). This type
actually
happens
to try
to store a natural number: if one
when one usesa computer
startsat 0 and performs the increment operation repeatedly,
the
will overflow its memory and the number
will
eventually
computer
around
back
to
this
take
a
number
0
quite
wrap
may
large
(though
a two-byte
of incrementation operations,for instance
representation
of an integer
will wrap around only after 65,536increments).
Note
that
this
type of number system obeys Axiom 2.1 and
Axiom
even
it clearly does not correspondto what
we
2.2,
though
natural
the
to
be
believe
numbers
like.
intuitively
from

To

to

0. More

prevent

sort

this

of

another axiom:
not the

Axiom

2.3.

0 is

we have

n++

0 for
\317\206

every

\"wrap-around

successor of any
natural

issue\" we will

natural

number n.

number;

impose
i.e.,

2. The natural numbers

20

we

instance

for

that certain types of wrap-arounddo not


of behavior
in
can now rule out the type

show

can

we

Now

occur:

2.1.5 using

Example

2.1.6. 4 is

Proposition
Don't

not

to

equal

Because of the

laugh!

way

0.
we

the incrementof the incrementof the increment


a priori
that
true
of 0 - it is not necessarily

the sameas

- it

\"a

posteriori\"
is

argument

indeed

was

refers

this

number

is \"obvious\",

The
one begins a proofor argument.
opposite
- what
one knows to be true after the proofor
Note for instance that in Example 2.1.5,
concluded.)
to
and
that in a standard two-byte
equal
0,
computer

true before

to be
is

is

increment

of the

is not
is
Latin
for
(\"a priori\"
or
what
one
knows
assumes
to
already

if it

even

zero,

\"beforehand\"

4 - it

defined

have

natural number,

of a

representation

0 (using our definition


thousand,

Proof,

of

hundred

five

definition,

By

Thus

number.

natural

to

is equal

65536

instance,

as equal to 0 incremented sixty-five


65536
and thirty-six times).
=

3++.
by Axiom

However, even with our new

numbersystem

for

axiom,

other

in

behaves

By Axioms 2.1 and 2.2, 3


2.3, 3++ \317\206
0, i.e.,
4^0.
it is

that our

still possible

pathological

is a

ways:

2.1.7. Consider a number


of five
Example
consisting
system
numbers 0,1,2,3,4, in which the increment
operationhits a
at
4.
More
that
0-H- = 1, 1++ =
\"ceiling\"
suppose
precisely,
=
=
=
words
that
5 = 4,
2-H3, 3++
4, but 4-H- 4 (or in other

2,

and hence 6 = 4,
Axioms

2.1,2.2,2.3.

is one

in

suppose
There

from
axiom:

4,

= 1

4++

are

happening,

many

This does not contradict


system with a similar problem
around, but not to zero,e.g.
wraps

etc.).
number

incrementation

which
that

Another

(so that 5 =

ways

to prohibit

but one of the

1, then 6 = 2, etc.).
the above types

simplestis to

assume

of
the

behavior
following

Axiom 2.4.

Thus,

for

\320\263/n-H-

2.1.8. 6 is not equal

Proposition

we

then

\321\202++,

different
n-H-

have n

must

\317\206

= m.

have

we

instance,

have

and \320\277\321\204\321\202,
then

natural numbers

m are

Equivalently*,

m++-

must

numbers

natural

Different

ifn,

i.e.,

successors;

21

axioms

The Peano

2.1.

2.

to

Then
5++
Suppose for sake of contradiction that 6 =
=
=
Axiom
2.4
we
have
5
that
so
4++
0-H-. By
1,
by
1-H-,
we then have 4 = 0, which
contradicts
our previous
Axiom 2.4
again

2.

Proof.

so

proposition.

can see from

As one

keepall of

this

distinct

numbers

natural

the

looks like
from each other.
it now

proposition,

we can
There

while the axioms (particularly


problem:
allow
to
us
confirm
that 0,1,2,3,... are
Axioms
2.2)
of N, there is the problem that theremay
be
other
elements
distinct
in
our
number
which
ofthis
form:
elements
arenot
system
\"rogue\"

is however

more

one

still

and

2.1

Example 2.1.9.
consisted of the

of integers

collection

following

our

that

Suppose

(Informal)

number system N
and half-integers:

N := {0,0.5,1,1.5,2,2.5,3,3.5,...}.

numbers,

2.1-2.4

difficult

without

in

order

using

Fortunately,
this

be

this

that

set.

says

that

the only

from 0 and
elements
such as

obtained

exclude

to

what

quantify

already

define.

capture

see

to

can

are usingreal

use yet.) One can check

is some axiomwhich

are those which

operationto

not supposed
to
are still satisfied for

we want

What

in N

we're

which

Axioms

since we

\"informal\"

is marked

example

(This

numbers

the increment
0.5. But it is

by \"can be obtained from\"


numbers, which we are trying
is an ingenious solution to try

we mean

the natural
there

fact:

2This is an exampleof reformulating


Section
A.2 for more details.

an implication

using its

contrapositive;

to

2. The natural numbers

22
2.5

Axiom

pertaining

property

any

is true,

true.

of mathematical induction). Let P(n) be


to a natural number n. Supposethat
P(0)

(Principle

and suppose that whenever P(n)


natural
Then P(n) is true for every

is

P(n-H-)

true,

number

is

also

n.

what \"property\" means


of P(n) might be \"n
possible
examples
\"n is equal to 3\"; \"n solves
the
is even\";
equation
(n + l)2 =
n2 + 2n + 1\";and so forth. Of course we haven't defined many
of
we
Axiom
will
to
these
but
when
2.5
these
do,
concepts
yet,
apply
refers
not just
properties. (A logical remark: Becausethis axiom
it is of a different
to variables^ but also properties^
nature than
the other four axioms;
Axiom
2.5 should technically be
indeed,
an axiom - it is a template
calledan axiomschema
rather
than
for
an (infinite)
number of axioms, rather than being a
producing
in its own right.
axiom
To discuss this distinctionfurther
single
the scope of this text, though,
falls in the realm
is far beyond
and
of logic.)
2.1.10.

Remark

The
Suppose

P{n)

are

We

at this point, but

intuition

informal

is

a little

vague on

some

such

that

behind

this

axiom is

P(0) is true, and

the following.

such

that

whenever

then P(n-H-) is true. Then sinceP(0) is true,


P(n)
=
is true. Since P(l) is true, P(l++) = P(2)is
P(0++)
P(l)
true.
we see that P(0), P(l), P(2),
this
Repeating
indefinitely,
P(3),etc.are all true however this line of reasoning will never
let us conclude
that P(0.5), for instance,
Thus Axiom
is true.
2.5 should not hold for number
which
contain
systems
elements
such
of this
as 0.5. (Indeed, one can give a \"proof\"
\"unnecessary\"
=
a
fact.
Axiom
the
\316\267
not
half2.5
to
\"is
Apply
property P(n)
0.5.
Then
is true, and if P{n)
integer\", i.e., an integerplus
P(0)
is true,
then P(n++)
is true. Thus Axiom
that
2.5
asserts
P(n)
is true for all natural
no
numbers
natural
number
canbea
n, i.e.,
In particular,
be a natural number.This
0.5 cannot
half-integer.
is
not
we have not defined such
because
quite
\"proof\"
genuine,
notions as \"integer\",
and
\"0.5\" yet, but it should
\"half-integer\",
of induction
is supposed
give you some idea as to how the principle
is true,

to prohibit any

numbers

other

than

the

\"true\"

natural

numbers

axioms

The Peano

2.1.

in

appearing

from

The principle
is true
property P(n)

we will

this text

N.)

of induction gives us a way to prove


for every natural numbern. Thus
see
which have a form
proofs
many

2.1.11.

Proposition

23

certain

P(n)

property

that
in

of

this:

like

is true

rest

the

for every

natural number n.

Proof.

base case \316\267=

P(0). (Insert proofof P(0)here).

i.e., we prove
that

inductively

verify the

first

We

induction.

use

We

natural

is
\316\267

number,

We
now
been
prove
P(n-H-).
This
that
assuming
P(n) is
here).
all numbers
n.
thus P{n).is true
proven.

true,

Now

suppose

and P(n) has already


(Insert proof of P(n++),
closes the induction, and
D

for

Of course
wording,

will

also

some

later,

other

induction

backwards

as

such

and

2.2.14),

(Proposition

8.5.15).
Axioms 2.1-2.5are known

numbers.

all

are

They

Assumption2.6.
2.1-2.5are
will

make

notation

number

possibility

have

could

Roman

the

we

there

that

really

systems

as

numbers,

for the

natural

we shall make

a number

system N;
Axioms
which
for

the

wanted

sets

for

refer

will

We

system.
is more

number system N as
of course considerthe

to this

One could

the

than one natural numbersystem,e.g.,we


number system {0,1, 2,3,...} and

Hindu-Arabic

number

different.

a bit morepreciseonce we have


in the next chapter.
and
functions

this assumption

Remark 2.1.12.
natural

exists

There

natural

call

axioms

and so

plausible,

very

will

Peano

strong
(Lemma

true.

down our

laid

exact template,

2.2.6),
(Exercise
transfinite
induction

the

as

{Informal)

whose elementswe
We

use the

necessarily

above type of proof, but the proofsusing


be something
like the above form. Thereare
variants
of induction which we shall encounter

generally

induction

not

in the

order

or

induction

we will

0,

system

to be
But

{O, J,/J,

JJJ, JV, F, V7,...},

annoying we couldview

these number

these

and

number

systems are clearlyequivalent

if

The

2.

24

(the technical term is isomorphic),

because

1 <->
to-one correspondence0 <-\342\226\272
O,
the zero of the Hindu-Arabic
system

J,

<->

numbers

natural

one

can

II,

etc.

create a onewhich maps

the Roman
is preserved
by the increment operation (e.g.,
system, and which
if
2 corresponds
to II, then 2++ will correspond to //++). For
statement
of this type of equivalence,see
a more
precise
of the natural number system are
all versions
Exercise 3.5.13.
Since
in
is
no
there
having distinct natural number
equivalent,
point
systems,and we will just use a single natural numbersystem to
with

the

zero of

do

mathematics.

not

will

We

prove

Assumption

include

it in

our axioms

be the

only

assumption

remarkable

create

used

we

accomplishment

from

axioms

set

theory,

see Axiom

ever make about


of modern
analysis

will

eventually

3.7), and it will

our numbers.

is that just

by

five very primitive axioms, and someadditional


we can build all the othernumber
theory,
systems,

these

from

starting

for

2.6 (though we will

set

functions,

and do

all the algebraand calculusthat

we

are

to.

(Informal) One interestingfeature about the


is that
natural
numbers
while each individual natural number is
the
set
of
natural
numbers is infinite; i.e., N is infinite
finite,
of
finite elements. (The wholeis greater
but
consists
individually
than
There are no infinite
natural
one
numbers;
any of its parts.)
one
is
comfortable
can even prove this usingAxiom
2.5,
provided
with the notionsof finite
infinite.
0 is finite. Also,
and
(Clearly
if
then
is
also
finite.
Hence
Axiom
\316\267
is finite,
n++
clearly
by
the
can
all
natural
are
natural
numbers
numbers
2.5,
finite.) So
but never actually reach it; infinity
one
is not
approach
infinity,
which
of the natural numbers. (There are other
number
systems
as the cardinals, ordinals, and padmit \"infinite\"
such
numbers,
adics,but they do not obey the principle of induction, and in any
event
are beyond
the scope of this text.)
Remark

2.1.13.

Remark 2.1.14.

Note

numbers

is

cmomatic

rather

definition

that

our

than

constructive.

of the natural
We have not told you

The

2.1.

Peano

25

axioms

what the natural numbers are (so we do not address such


the numbers
are made of, are they physical
what
questions as
objects,
some
listed
do
what
they measure, etc.) - we have only
things
do
with
the only operation
we
have
defined
them
can
fact,
(in
you
on them right now is the increment one) and some of the
have.
This is how mathematics works - it treats
that
properties
they
what properties the
about
its objectsabstractly,
only
caring
what
the
mean.
If one
not
are
or
what
have,
they
objects
objects
a natural
wants to do mathematics, it does not matter whether
means a certain arrangement of beads on an abacus,or
number
in a computer's memory, or some
of bits
a certain
organization
with no physical substance; as long as you
abstract
more
concept
if two of them are equal,and
on do
later
see
increment
can
them,
other arithmetic operations such as add and
multiply,
they
for
numbers
mathematical
as
they
purposes (provided
obey the
qualify
of
It
is
to
the
natural
construct
possible
requisiteaxioms,
course).
from sets, for instance
from
other
mathematical
numbers
objects
- but there are multiple ways to constructa working
model
of the
a
natural numbers, and it is pointless,at least from
the
as
which
model
is
one
to
about
\"true\"
mathematician's
standpoint,
argue
- as
all
the
and
all
the
as
it
axioms
does
long
obeys
right things,
that's good enoughto domaths.
the realization that numberscould
is very
be
recent, not much more than a
axiomatically
hundred
old.
Before
years
then, numbers were generally
be inextricably
to some external concept,such
understood to
connected
of a
as
the cardinality
of a set, measuringthe length
counting
line segment,
This
etc.
worked
or the mass of a physical
object,
from one number
to move
reasonably well, until one was forced
in terms
numbers
system to another;for instance,
understanding
Remark

2.1.15.

Historically,

treated

of counting beads, for

is

instance,

numbers 3 and 5, but doesn't


3+4i; thus eachgreatadvance

work
in

the

the
for conceptualizing
great
so well for \342\200\2243
or 1/3
or y/2 or
theory

of numbers

- negative

even the number


numbers,irrational
complex
numbers,
zero led to a lotof
The great
philosophical
anguish.
of
the
late
was
that
numbers
can be
nineteenth
discovery
numbers,

unnecessary

century

2. The natural numbers

26
understood

models

and
his or her intuition
can also be just as easily discardedwhen
but
they
in
the
way.
get
it

when

understanding,

to

begin

to aid

is convenient,

One consequenceof the axiomsis that

:=

ao

by first

of numbers

c&2,...

for
\321\201
of

function

ao,

we want

Suppose

recursively.

sequences

necessarily needing a
can use any of these

without

axioms,

a mathematician

course

of

model;

they

via

abstractly

concrete

some

number

a\\ :=

/o(ao)>

define

now

can

we

to build

a sequence\316\261\316\277,
\316\261\317\207,

defining ao to be somebasevalue,
c, and then by letting a\\ be some
of ai,

function

be
some
\320\276>\321\207

e.g.,

:=

a<i

/i(a^),

for some
and so forth. In general,we set an++ := /n(an)
from
we
will
N to N. By using all the axiomstogether
function fn
now conclude
that this procedure will give a single value to the
sequence elementan for each natural number n. More precisely3:

Proposition 2.1.16
number n,

natural

natural

the

number.

we

some

have

natural number n,

ao =

that

such

fn : N

function

numbers to the natural


Then we can assign a unique

numbers. Let

\321\201
and

an++

natural numbern.

Proof.
procedure

a single

gives

of Axiom 2.3.) Now

a singlevalue
:=

to

fn(a>n)\302\267

suppose

an.

Then

(None

induction.

to ao,

value

an++ := /n(an)

definitions

an++

use

We

(Informal)

of

will

We first

namely
the

redefine

that

inductively

completes

to each

an

for each

/n(an)

observe that this


of the

\321\201

(None

o^.

other

because
the procedure gives

value

the value of an++, becauseof Axiom


and so an is defined for each
induction,

a single value assignedto each

n, with

from

a natural

\321\201
be

of ao,

it gives a single value to an++,


the other definitions 0\342\204\242++
:=

will redefine
the

\342\200\224>
N

number

natural

each

for

Suppose

definitions).

(Recursive

2.4.)

natural

namely

/m(am)

This

number
D

one to define the notion


of
Strictly speaking, this proposition
requires
which we shall do in the next chapter.
However, this will not
be circular, as the conceptof a function
does not require the Peano axioms.
can be formalized more rigourously
in the language
2.1.16
of set
Proposition
a function,

theory;

see

Exercise

3.5.12.

27

Addition

2.2.

be used here.In a system


some sort of wrap-around, recursive definitions would
had
which
some elements of the sequencewould
because
work
constantly
not
For instance, in Example 2.1.5, in which 3++ = 0,
be redefined.
be (at least) two conflicting definitions for ao,
would
there
then
In
a
which had superfluous elements
or
either\321\201
/\320\267(\320\260\320\267)). system
Note how

of

all

suchas 0.5,

the

had to

axioms

the

element

ao.5

one

for

powerful;

multiplication,

instance,

we can

to which

we now

Addition

2.2
The

be defined.

never

would

Recursive definitions are very


addition
and
use them to define
turn.

number

natural

system
- increment

operation

is very bare
- and a handful

right now: we have only


of

now we

But

axioms.

can build up more complexoperations,such as addition.


The way it works is the following. To add threeto five should
- this is one increment
three
times
be the same as incrementing five
to five, which is one incrementmorethan
two
morethan adding
zero
one to five, which is one incrementmorethan
adding
adding
So we give a recursive definition
to five, which should just give five.
for

as

addition

follows.

(Addition of natural numbers). Let m be a


zero to m, we define 0 + m := m. Now
natural
number.
Then
we have defined how to add \316\267
to m.
that
suppose inductively
:=
we can add n++ to m by defining
m
+
+
(n++)
(n m)++.
2.2.1

Definition

To add

Thus

mism,

1 +

(1-H-)+m = (m-H-)-H-;
and

(3++)++= 4++ = 5.

Prom

m = (0++) + m
so

forth;
our

is

for instance
discussion

m++;

2 +

we have

of recursion

m =

2+3 =
in the

\316\267
+ m for every
previous section we see that we have defined
the previous
integer n. Herewe are specializing
general discussion
= n + m and
= an++.
to the settingwhere
Note that
an
fn(a>n)
this definition is asymmetric:3 + 5 is incrementing
5 three times,
while 5 + 3 is incrementing3 five times.
Of course,
they both
it
a
fact
the
value
of
More
is
same
8.
yield
generally,
(which we

2.

28

The

numbers

natural

prove shortly) that a + b = b + a for all natural numbersa, b,


the
definition.
is not immediately clear from
this
although
Axioms
that we can prove easily,
Notice
2.1, 2.2, and
using

shall

numbers
is
induction (Axiom 2.5), that the sum of two natural
a
number
natural
again
(why?).
now
we only have two facts about addition:that 0+m =
Right
= (n + m)++. Remarkably,
this
turns
and
that
out
m,
(n++)+m
we know
about addition.
to be enough to deduceeverything
else
We begin with somebasiclemmas4.

deduce this

we cannot

that

Note

natural number \316\267,\316\267


+ 0 =

For any

2.2.2.

Lemma

because we do not know

Proof.

The

m = m

0 +

that

know

we

yet

induction.

use

We

is a natural number. Now


show

to

wish

We

addition,
\316\267
+

immediately from
+ b = b + a.

base case

for

0 + 0 = 0 follows
that

inductively

m =

since

m, and 0

number

natural

every

suppose

0 +

\316\267
+

0 =

\316\267.

that (n++) + 0 = n++. But


definition
of
0 is equal to + 0)++, which is equal to n++
by

since

that

\316\267.

(n++)
This
0 = \316\267.

(n

the

closes

Lemma 2.2.3. For any

natural

induction.
numbers

\316\267
and

n +

m,

(m++)

(n + m)++.
we

Again,
because

cannot

deduce

not

know yet

do

we

Proof. We

base case

induct on \316\267
(keeping

=
\316\267

0. In this

from (n++)+m =
that a + b = b + a.

this yet

fixed).

case we have to prove

(n+m)++

consider the
= (0 +
(m++)

We first
0 +

of view, there is no differencebetween


a lemma,
or
proposition,
theorem,
corollary- they are all claims waiting to be proved.
we use these terms to suggest different
of importance
and
levejs
However,
A lemma
is an easily proved claim which
is helpful
for proving
difficulty.
other propositions and theorems,but is usually not particularly interesting
in
4From

logical

point

is a statement which
in its own right,
is interesting
right. A proposition
a theorem is a more important
statement
than a proposition which
says
to prove than
definitive on the subject, and often takes more
effort
something
A corollary
a propositionor lemma.
is a quick consequence of a proposition
or theorem
that was proven recently.
its own

while

29

2.2* Addition

(m++) = m++ and 0 +


and
are thus equal to each
= ra, so both sides are equalto m++
m
we
n
assume
that
Now
+
inductively
other.
(m++) = (n + m)++\\
=
+ (m-H-)
we now have to show that (n++)
((n++) +m)-H-. The
is
definition
of addition, which
side
+
by
left-hand
(n
(m++))++

m)++. But by

0+

of addition,

definition

the
inductive
by
hypothesis.
Similarly,
is equalto ((n+ m)++)++
=
we have (n++)+m
(n+m)++ by the definition of addition, and
to ((n + m)++)++.
Thus both
so the right-handsideisalsoequal
to
sidesare equal each other, and we have closed the induction. D
As

see

a particular

n++

that

of
corollary
\316\267
+ 1 (why?).

earlier, we can now

As promised

Proposition 2.2.4
numbers

\316\267
and

shall

We

Proof.

Lemma 2.2.2

m,

is

(Addition

n +

m =

prove

and Lemma2.2.3we
a + b

that

For

commutative).

m + n.

use induction

on

\316\267

(keeping

fixed).

= b+

a.

any natural

First

we do

=
the basecase\316\267
0, i.e., we show 0+m = m+0. By the definition
= m, while by Lemma 2.2.2, m + 0 = m. Thus
m
0
of addition,
+
= m+n,
case is done. Now suppose inductively
that
base
the
n+m
now we have to prove that (n++)
+ m = m+ (n++) to closethe
the definition
of addition, (n++)+m = (n+m)++.
induction.
By
Lemma
m + (n++) = (m + n)-H-,but
to
this
is equal
2.2.3,
By
=
the
n
m
m
Thus
inductive
+
+
+n.
hypothesis
m)++ by
(n
D
have
the induction.
closed
+ m = m + (n++) and we
(n++)

2.2.5 (Addition is associative).For any


= a +(b + c).
b, c, we have
(a + b) + c

Proposition
numbers

a,
See

Proof.

Exercise

2.2.1.

Because of this associativity we canwritesumssuch


having to worry about which orderthe numbers

without
added

as
are

together.

Now we

law.
develop a cancellation

2.2.6

Proposition
numbers

natural

such

that

a + b

(Cancellation

= a

+ c.

Then

law). Let
we

have

a, b,

\321\201
be

b =

c.

natural

a+b+c
being

2. The natural numbers

30

Note that
to

this

prove

we

concepts

In

yet.

us

becauseit
subtractionis
We

Proof.

case

integers)
of 'Virtual

not developed these

subtraction\" even before

defined.

officially

the
induction on a. First consider
base
=
Then we have 0 + b
0 + c, which by definition of

prove

a = 0.

the

(and
a sort

for

allows

numbers yet

law is crucial in letting


later on in these notes,

cancellation

this

fact,

subtraction

define

we have

because

proposition,

or negative

subtraction

use

cannot

this by

addition implies

b =

that

\321\201
as

Now

desired.

inductively

suppose

for a (so that


law
that we have the cancellation
a+b \342\200\224
a+c
implies
=
In
we
now
have
cancellation
law
for
to
the
other
b
a++.
prove
c);
=
we
to
b
and
need
show
assume
that
\321\201
+
+
words,
(a++)
(a++)
=
=
of addition,
that b c. By the definition
(a++) + b (a + b)++
= (a + c)++.
=
we
have
and
so
and
+
(a
c)++
(a + b)++
(a++)+c
=
a
\321\201
we
have the
Axiom
we
have
a
b
Since
+
+
2.4,
already
By
=
as desired.
This closes
cancellation law for a, we thus have b \321\201
the induction.
D

discuss

now

We

Definition 2.2.7
said

\316\267
is

for

\"if

to be

and

natural

(Positive

it

positive

number

A natural

equal to 0.

shorthand

is

(\"iff\"

A.l).

Proposition 2.2.8. If a is
then a + is positive
{and hence
b

with positivity.

numbers).

is not

positive
if\" - see Section
iff

only

interacts

addition

how

and

b +

b is

a natural

a is also,

by

number,

Proposition

2.24)-

We use induction on b. If b = 0, then a+b=a+0=


is positive,so this proves
which
the
case.
Now suppose
base
a,
\342\200\224
a + (b++)
Then
inductively that a + b is positive.
(a + b)++,
which cannot be zero by Axiom
and
is
hence
2.3,
positive. This
Proof.

closesthe
Corollary

0,

induction.

2.2.9.

then a = 0

and

If a

and b are natural numbers

6 =

0.

such

that

a+b

31

Addition

2.2.

Proof. Suppose for


If

then

0
\316\261
\317\206

\316\261
is

2.2.8,

positive,

and again

a+

exactly

or

0.
\317\206

Similarly if b \317\2060 then b is


0 is positive by Proposition 2.2.8, a
b must both be zero.
D

Then there exists


=
a.
b++

positive number.

b such that

number

natural

Proof.See

2.2.2.

Exercise

Once

be a

Let

2.2.10\302\267

one

a and

Thus

Lemma

hence

and

positive,

\316\261
0
\317\206

a + b = 0 is positiveby

a contradiction.

Proposition

contradiction.

that

contradiction

of

sake

a notion of addition, we

we have

can

defining

begin

order.

notion of

Definition 2.2.11

natural

the

of

(Ordering

Let

numbers).

\316\267
and

is greater
numbers. We say that \316\267
than or equal to
= m + a for some
> m or m < n, iff we
\316\267
have
\316\267
write
and
ra,
is strictly
than m, and
natural number a. We say that \316\267
greater

be natural

write

\316\267
>

< n,

or m

iff

>
\316\267

and

\320\277\321\204\321\202.

Also
5, because 8 = 5 + 3 and 8^5.
for
thus
there
is
no
natural
note that n-H- > \316\267
n;
any
largest
the next number n++ is always
still.
number
n, because
larger
for

Thus

2.2.12

Proposition
numbers).

Let

8 >

instance

be

a,b,c

(Basic properties of order for


natural numbers. Then

(a) (Order is reflexive)


(b) (Order
(c)

(Order

(d)

(Addition

a>

is anti-symmetric)

(e) a <b if

only

if a-H-

(f) a <b if and

only

ifb

See Exercise

2.2.3.

and

If a>b

b >

c, then

and b>a,

order) a>b if and

and

Proof.

a.

is transitive) If a>b

preserves

natural

= a

only

a>

\321\201

then

if a

b.

c> b +

c.

< b.

+ d

for somepositive

number

d.

2.

32

2.2.13 (Trichotomy of order


b be natural numbers. Then
exactly
statements is true: a < b, a = b, or a> b.

Proposition
Let a and

Proof This is only

First we show that


a

then

<

by

\316\261
b
\317\206

by

> b and a

If a

one

numbers).

of the

gaps will

following

in

filled

be

2.2.4.

Exercise
statements

natural

for

proof; the

of the

a sketch

numbers

natural

The

b>

cannot

we

>

definition,

<

then

one of the
same time. If a

than

more

have

b holding

at the

and if a

> b then

by

\316\261
b
\317\206

2.2.12

Proposition

by

<

definition.

a=

we have

b,

than one of the statementsis true.


contradiction.
Thus
show
that
Now
we
at least one of the statementsis true. We
fixed
and
on a. When a = 0 we have 0 < b for
b
induct
keep
all b (why?), so we have either 0 = b or 0 < b, which
the
proves
for
and
base case. Now suppose we have proven the proposition
a,
Prom
the trichotomy
now we prove the proposition for a++.
for
a, there are three cases:a < b, a = b, and a > b. If a > b, then
If a = b, then a-H- > b (why?).
Now
> b (why?).
\316\261-\320\235suppose
that a < b. Then by Proposition
we have a-H- < b. Thus
2.2.12,
eithera-H-= b or a-H- < b, and in either case we are done.This
no more

closesthe
The
of the

induction.

properties

principle

allow one to obtaina stronger


of induction:
of order

Proposition 2.2.14 (Strong


principle
a natural number, and Let P(m) be
arbitrary natural numberm. Suppose

of
a

induction).

property
that

for

version

Let mo be
to an

pertaining
each m >

mo, we

if P(mf) is true for all natural


implication:
< m, then P(m) is also true.
particular,
(In
this means that P(mo) is true, sincein this
is
case
the hypothesis
is true for all natural
vacuous.) Then we can concludethat P(m)

have

numbers

the

following

mo

< m!

numbersm > mo.

Remark2.2.15.
In

=
with mo = 0 \320\276\321\202
mo

Proof.

See Exercise

applications

we

usually

use this

principle

1.

2.2.5.

2.3.

33

Multiplication

Prove Proposition

2.2.1.

Exercise

and induct on the

2.2.5. (Hint: fix

variables

of the

two

third.)

Exercise2.2.2.

2.2.10.

Lemma

Prove

use induction.)

(Hint:

2.2.12.
Proposition
(Hint: you will
Exercise 2.2.3. Prove
propositions, corollaries,and lemmas.)
2.2.4. Justify the three statementsmarked
Exercise
(why?)
of Proposition 2.2.13.

need

many

of

the preceding

Exercise2.2.5.Prove
that

property

P{m)

vacuously true when


Let

2.2.6.

Exercise

to the

pertaining

induction.

backwards

2.3

note

proof

to be the

that

Q{ri)

is

\316\267
< mo\302\267)

\316\267
be

natural

all natural

for

define Q{n)

(Hint:

< m < n;

for all mo

a natural

number,

and let

numbers such that

is true. Supposethat

then P(m)

is true

2.2.14.

Proposition

is true

in the

P{n)

ra <

numbers

apply

(Hint:

is also

n; this

is

induction

P(m) be a property

whenever

is true,

P(ra-H-)

true.

known

to the

Prove that P{m)


as the principle of
variable

n.)

Multiplication

previous section we have proven all the basicfacts that we


true about addition and order.
To save
space and
the
we will now allow ourselves to
to avoid
obvious,
belaboring
addition
and order that we
useall the rulesof algebra
concerning
In the

to be

know

for
instance
are familiar with, without further comment.Thus
=
c + b + a without
we may write things like a + b + c
supplying
Now
we
further
introduce
multiplication.Just
any
justification.

as additionis the iterated

increment

iterated

operation,

is

multiplication

addition:

Definition

2.3.1 (Multiplicationof natural numbers).Let

be

m := 0.
number. To multiply
zero
to m, we define 0 \317\207
Now suppose inductively that we have defined
how to multiply \316\267
to m.
\317\207
m :=
Then we can multiply n++ to m by defining
(n++)
m.
+
(nxm)

a natural

Thus for

etc.
natural

By

instance

induction

numbers

0\317\207m

one

is a

= 0+m,
easily verify that

can

natural

0, lxm

number.

2xm = 0+m+m,
the product of two

2.

34
2.3.2

Lemma

(Multiplication
Then

numbers.

natural

\317\207
\316\267
m = m

natural

The

numbers

is commutative). Let \316\267


be
^ \317\204\316\267
x n.

Proof. See Exercise2.3.1.


that

convention

ab +

usual notational conventions of precedencefor


when they are defined later, to
operations
all
the
parentheses
time.)

using

Lemma 2.3.3
be

\317\207
(\316\261 b)

arithmetic

other
on

\321\201
means

Then

numbers.

natural

no zero

have

numbers

(Natural

= 0 if and
nxm
and
if \316\267
particular,

thus
will

Proof.See

have

numbers
(Distributive law). For any natural
=
=
and
ab
ac
ba
ca.
+
+
c)
(b + c)a

2.3.4

b, c, we

least one of
both positive,

at

2.3.2.

Exercise

Proposition

if

are

the
save

Let n, m

divisors).

only

In
to zero.
n, m is equal
then nm is alsopositive.

a,

usual

over addition,
precedence
+ c, not a x (b + c). (We

the

use

also

takes

use the

and

\317\207
\316\267
\317\204\316\267
as nm,

multiplication

instance

for

abbreviate

now

will

We

a(b

is commutative
we only need to show
Proof. Since
=
the
ab + ac.
a and
b fixed,
identity
keep
a(b + c)
=
and use inductionon c. Let's
the
base
case
0, i.e.,
=
The
a0.
the
left-hand
side
is
while
+
+
ab,
0)
a(b
is ab + 0 = ab, so
side
are
with the base case. Now let us
done
= ab + ac, and let
that
that
suppose
prove
a(b + c)
=
ab + a(c-H-).The
side
is a((b + c)++) =
a(b + (c-H-))
= a(b+c) + a
side
is ab+ac+a
a(b+c)+a, while the
the
induction
can
and
so
we
closethe
induction. D
hypothesis,
multiplication

first

We

\321\201

prove

ab

right-hand

we

us

inductively

left-hand

right-hand

Proposition 2.3.5
natural

numbers

Proof.

See

a, b,c,

natural

2.3.6
numbers^

is

(Multiplication

we have

\317\207
(\316\261 b)

For

associative).

x c

= a

any

x (b \317\207
\321\201).

2.3.3.

Exercise

Proposition

by

such

(Multiplication
that a < b,

\342\226\241

preserves

and

\321\201
is

order). If a,b

positive,

then

ac <

are
be.

35

2.3. Multiplication

proof. Sincea <


Multiplying

is

Since

ac <

\321\201
and

by

we

using

b =

a +
the distributive
have

and

positive,

\321\201
is

some

for

law we
dc is

positive,

d.

positive

obtain bc = ac + dc.
positive, and hence
D

desired.

as

be

b,

2.3.7 (Cancellation law). Let a, b, \321\201


be natural
Corollary
=
=
a
ac
be
and
is
b.
\321\201
Then
non-zero.
that
such

2.2.6will
Remark 2.3.8. Just as Proposition
\"virtual

which

subtraction\"
this

subtraction,

neededto

corollary

define

cases:

trichotomy of order (Proposition


2.2.13),we
a < b, then

contradiction when a

a similar

that a =

b,

on.

later

division

= b, \316\261
a < b, \316\261
first that
> b. Suppose
2.3.6 we have ac < be, a contradiction.

Proposition

for

allow

let us define genuine


eventually
a \"virtual division\" which will be
provides
will

genuine

Proof. By. the


three

numbers

> b.

Thus

the

only

can

We

have

by

obtain

possibility

desired.

as

is

these propositions it is easy to deduceall the familiar


involving addition and multiplication,seefor

With

of algebra

rules

2.3.4.

Exercise

instance

the familiar operationsof additionand


notion of increment will begin to
primitive
multiplication,
the wayside,
on.
In any
fall
and we will see it rarely from now
by
event we can always use additiontodescribe
since
incrementation,
we have

that

Now

more

the

n-H-

\316\267
+

Proposition
number,
numbers

Remark
\316\267
by

1.

(Euclidean algorithm). Let \316\267be a


and let q be a positive number. Then
there
exist
=
r
such
<r
that
0
and
\316\267
r.
<
+
m,
q
mq
2.3.9

2.3.10. In otherwords,

a positive

natural

number)

number

and a

and

this text.

important

can

divide

a natural

obtain a quotientm (which


is less
than
remainder r (which

q to

algorithm marks the beginning


beautiful

we

subject

of

number

but one

theory,

which is beyond

is

natural
natural

number
another

q). This
which is a
the

scope

of

2. The natural numbers

36

2.3.5.

Exercise

See

Proof.

to recursively
Just like one usesthe increment
operation
define
addition, and additionto recursively
multiplication,
use multiplication to recursively define exponentiation:

Definition 2.3.11
be a natural number.To

natural

for

(Exponentiation

Now suppose recursively that mn


natural
number n, then we define mn++
1.

one can

Let m

numbers).

power 0, we define mP

to the

raise

define

has been
mn

:=

for

defined

:=

some

\317\207
m.

x1 =

=
= 1 \317\207
\317\207
\317\207
\317\207
x\302\260
2.3.12.
Thus for instance
Examples
\317\207\\
=
=
=
=
\317\207
\317\207
\317\207
\317\207
\317\207
x2
\317\2071
\317\207 \317\207
\317\207
so forth.
\317\2073 \317\2072
\317\207 \317\207
and
\320\226;
\320\266;
By

this recursive

induction we see that


natural numbers n.
not

will

We

here,

but

rational

develop
wait

instead

Exercise2.3.1.

Prove

2.2.2,

Exercise2.3.3.

Prove

Proposition

2.3.2.

Lemma

Exercise2.3.4.

2.3.3.

Proposition

2.3.5.

modify the

proofs of

Lemmas

prove the

(Hint:
2.3.5.

second statement

modify the

(Hint:

proof

of

the distributive law.)

Prove

the

Prove

Proposition

identity

numbers a, b.
Exercise

(Hint:

2.2.4.)

Lemma

and use

2.2.5

of exponentiation too deeply


have defined the integersand
Proposition 4.3.10.

particular

2.2.3 and Proposition

Exercise2.3.2.Prove
first.)

all

for

xn

the theory
until after we

see in

numbers;

defines

definition

(a +

6)2 =

a2

2.3.9. (Hint:

+ 2ab+b2

fix

q and

for all

natural

induct

on

\317\200.)

Chapter

Set

theory

is concerned
Modern analysis, like mostof modern
mathematics,
and geometry. We have already
introduced
system, the natural numbers.We will

with numbers, sets,


one type of number

number
systems shortly, but for now we pause to
the
and
notationofset theory,
as they
will be
introduce
concepts
used increasingly heavily in later chapters. (We will not pursue a
rigourous description of Euclidean geometryin this text,
to describe
that geometry in terms of the real number
preferring instead
means
of
the
Cartesian co-ordinate system.)
by
system
set
is not the main focus of this text, almost
While
theory
of
as part
of
other
branch
mathematics relies on set theory
every
in
its foundation, so it is important to get at least some grounding
In
areas
of mathematics.
set theory before doingother advanced
this chapter we present the moreelementary
of
axiomatic
aspects
set theory, leaving more advancedtopicssuch
as a discussion
of
infinite sets and the axiomofchoice
to Chapter
8. A full treatment
of the finer subtletiesof set theory
are many!) is
which
there
(of
introduce

the

other

unfortunately well beyond


3.1

the

scope

of this

text.

Fundamentals

In this section

we

shall

set

for the natural


numbers.
somewhat
overcomplete

out

some axioms

for sets, just as we

did

For pedagogical reasons,we will use a


list of axioms for set theory,
in the sense

3. Set theory

38
of the axioms can be usedto deduce
harm in doingthis. We begin with an
of what sets should
be.
some

that

no real

3.1.1.

Definition

an
in

of

collection

unordered

3 e {1,2,3,4,5} but
definition

This

be

considered

to

defines

operations
no axioms

do.
the

we say that

be

to

any

\317\207
G

A.

\317\207
#

if

\317\207
is

\317\207
lies

For instance,

7 \302\243
{1,2,3,4,5}.

is intuitive enough, but it doesn't answer


a
such as which collectionsof objectsare
how
which sets are equal to othersets,and
one
on sets (e.g., unions,intersections,
etc.).Also,

of questions,

number

we

of A or

element

an

is

description

is a set. If

{3,8,5,2}

e.g.,

objects,

is
we say that \317\207
object,
the collection;
otherwise

a set A

We define

(Informal)

there

but

others,

informal

sets,

their
what sets do, or what
elements
and
these
these axioms
defining
operationswill be
Obtaining
of the remainder of this section.
purpose
first
We
one point: we consider sets themselvesto be a
clarify

have

of

type

yet on

object.

3.1 (Sets

Axiom

object In

are objects). If A

particular,

ask whetherA

is

also

Example 3.1.2.

two

an

element

elements,

set

which

of

one

then

A is

also an

B, it is meaningful to

of B.

The

(Informal)

distinct

a set,

A and

sets

given

is

4} is a set of three
to itself be a set of two

{3, {3,4},

happens

more formal versionof this


for instance, we
example.However,
objects
do
not
consider
a natural number such as 3 to be a set.
typically
more
accurate
statement
is that natural numbers can be the
(The
than necessarily being sets themselves.
cardinalities
of sets, rather
See

elements.

3.1.10

Example

for a

are sets;

all

not

SeeSection3.6.)

Remark3.1.3.

There

set

theory\",

in which

is

a special

all objects

might be identifiedwith the empty

be
with

with

identified

{0,1}

{0}

{{},{{}}}>

case of

are sets; for


set

set theory, called \"pure


instance

0 = {},

the

number

the number

1 might

the number 2 might be identified


and so forth. Prom a logicalpoint of

{{}},

3.1.

39

Fundamentals

is a simplertheory,

set theory

pure

view,

deal with sets and not

point of view
in

some

which

types

or less equivalent for the


and so we shall take an agnostic

are more

doing mathematics,
to whether all objects
so

summarize

To

of the

some

mathematics,

are setsor not.

objects

statement3

meaningless,

simply

to

not a

4 is

since

of

position

as

objects studied in
is an
happen to be sets; and if \320\266

a set, then either\317\207


or \317\207
G A is true
G A
we
A
the
\317\207
leave
statement
G
set,
undefined;

not

purposes

all the

among

far,

and A is

we consider the

has to

only

from a conceptual
however,
objects;
easier to deal with impure set theories
are not considered to be sets. The two

objects

theories

of

one

with

often

is

it

since

(If A is
for instance,

be true

neither

object

is false.

or false, but

set.)

when are two sets


not
the orderof the
consider
equal?
inside
a
set
to
think of {3,8,5,2}
be
thus
we
elements
important;
as the same set. On the other hand,
and
{2,3,5,8}
{3,8,5,2}
are different sets, becausethe latter
contains
set
and {3,8,5,2,1}
that the former one doesnot, namely
1.
element
the
an element
We
sets.
For similar reasons {3,8,5,2} and {3,8,5}are different
formalize this as a definition:

Next,
to

considered

We do

be

Definition3.1.4

of

(Equality

A =

B,

iff

of equality:

notion

the

define

we

every

of A

element

To put it another way,


also to B, and
belongs

sets).

is an

\320\222
if

and

element

every

3.1.5. Thus, for instance,


are the same set, since
contain
they

Example

sets A and

\320\222
are

equal,

element of \320\222and

vice

versa.

Two

only

if every

element

of \320\222
\321\203
belongs

can

symmetric,
and

A = B,

element

that

verify

easily
and

transitive

exactly

then

of\" relation

\317\207
G

this notion of

(Exercise
B,

G obeys

3.1.1).

\317\207
of

to A.

and {3,4, 2,1,5}


the same elements.

{1,2,3,4,5}

to
(The set {3,3,1,5,2,4,2}is alsoequal
{1,2,3,4,5};
of 3 and
2 is irrelevant as it does not further
repetition
status
of 2 and 3 being elementsof the set.)
One

also

the
change

the

equality is reflexive,

Observe

that if

\317\207
G

Definition
3.1.4. Thus the \"is an
the axiom of substitution(seeSection

by

3. Set

40

theory

operation we define on sets will


the axiom of substitution, as long as we can define
also
that
obey
instance
in
terms
of
the
Thisis
for
relationG.
the
operation
purely
the
case for the remaining definitions in this section. (On
other
the
or
the
of
\"first\"
\"last\"
we
notion
cannot
use
element
hand,
not
in a set in a well-definedmanner, becausethis would
respect
of substitution;

axiom

the

any new

of this,

Because

A.7).

sets

the

instance

for

{1,2,3,4,5}

and

first elements.)
{3,4,2,1,5} are the sameset,but have different
which
are sets
Next, we turn to the issueof exactly
objects
is analogous
to how we
and which objects are not. The situation
we started
defined the natural numbersin the previous
chapter;
more
with a single natural number,0, and
started
building
of 0 using the increment operation. We
will
numbers out
try something
similar here, starting with a singleset, the empty
set, and building
more sets out of the empty
set
We begin
operations.
by various
of
the
set.
the
existence
empty
by postulating

set,

There

3.2 (Empty set).

Axiom

no

contains

which

elements,

i.e., for

as the empty

0; known

a set

exists

we
every object \317\207

have

\317\207&\316\246.

be

The

empty

one

empty
then

empty,

were two

set; if there
by

set is not

equalto

the

following statement is very

empty
simple,

3.1.6 (Singlechoice).Let

Lemma

such
there existsan object\317\207

that

0 and

sets

there

0' which

can

worth
be

A
\317\207
G

other

non-empty. The
stating nevertheless:

we call it

set,
but

only

were both

would be equalto each

3.1.4 they

Definition

(why?).

If a

denoted {}. Note that

is also

set

set.

a non-empty

Then

A.

by contradiction. Supposethere doesnot exist


that
Then
for all objects \320\266,
we
\317\207
have
G A.
any
object
A.
Axiom
3.2
have
\317\207
Thus
\317\207
A
\317\207
\317\207
we
4=>
G
G 0
Also, by
g
& 0.
=
statements
are equally false), and so A
0 by Definition
(both
a
contradiction.
D
3.1.4,
Proof.

We prove

\317\207
such

Remark 3.1.7.
nonempty

set

The

A, we are

above

Lemma

asserts

allowed to \"choose\"an

that

given any

element

\317\207
of

A which

41

Fundamentals

3.1.

3.5.12)
demonstrates this non-emptyness. Lateron (in Lemma
that
finite
number
of
show
will
any
sets,
given
say
non-empty
we
i*
IS
to
one
choose
element
from
\302\267
\302\267
\302\267
possible
x\\,..., xn
> -^n>
\320\233\321\214
set

each

axiomof

Remark 3.1.8. Notethat

the

as the natural number 0.


it

choice\".

\"finite

elements from an infinite

an additional axiom,the
Section 8.4.

However,

as

is known

this

\316\221\317\207,...,
An\\

order to choose

is true

is

the other

a set;

the cardinality of

that

the same thing


is a number.

is not

set

empty

One

we need
will discuss in

we

which

choice,

in

However,
of sets,

number

the

is

set

empty

0; see

Section 3.6.

If Axiom 3.2 was


in

as there might

quite
boring,
the empty set.

existence,

We

now

had, then set

set theory

that

axiom

only

be

could

theory

the

be just a single
further

present

set

to

axioms

enrich the classof sets available.


exists a

there

have

we

y,

set

objects,

then

and b;
=

if and only
element is

whose

there

i.e., for

a or

and pair sets). If a is an

set {a} whose only

G {a}
\321\203

singleton

\321\203

(Singleton sets

3.3

Axiom

\321\203

exists a

y,

refer to

i.e., for every object


we refer to {a} as the

is a,

=
\321\203 a;

if
a. Furthermore, if a and

set {a,b} whose

object

every

b; we

element

we have

this set as

then

object,

G {a,
\321\203
the

elements

only

pair

b}
set

are

are

if and only
formed

if

by a

and b.

Remarks 3.1.9. Just as there

is only
one empty
set, there is
to
thanks
Definition
each
3.1.4
only
singleton
object a,
is only
Similarly,
given any two objects a and b, there
(why?).
one pair set formed by a and b. Also, Definition
3.1.4 also ensures
= {a} (why?). Thus the
that {a,b} = {b,a}(why?)
and
{a, a}
is in fact
singletonset axiom
redundant,
being a consequence of
the pairset axiom.
axiom will follow
from
the
set
Conversely,
pair
the singleton
set axiom and the pairwiseunion
axiom
below
(see
Lemma 3.1.13). One may wonder why
we
don't
further
and
go
one

set for

will
create triplet axioms,quadrupletaxioms,etc.;however
there
be no need for this once we introduce
the
union axiom
pairwise
below.

3. Set

42

theory

hence
3.1.10.
an object),
so is
Since 0 is a set (and
Examples
singleton set {0}, i.e., the set whose only element is 0, is a set
it is not the same set as 0, {0}\317\206
the
0 (why?).
Similarly,
(and
are
also
sets.
These
singleton set {{0}} and the pair set {0,{0}}
three sets are not equal to each
other
3.1.2).
(Exercise

above examples show, we

As the

sets; however,the sets

allowsus to build

somewhat

3.4 (Pairwiseunion).

Given

next axiom

Axiom

exists a set

two

so

elements,

any

two

\320\222
of

which
elements consists of all the elements
In
both.
other words, for any object x,

than

sets

larger

the union

called

B,

quite

fairly small (each

more than

of no

consists

build

can

we

create

now

can

still

are

make

we

sets A,
A

a few

set that

far).

The

before.

B, there

B, whose
to A or \320\222or

and

belong

4=>
\321\205\320\265\320\220\320\270\320\262

(xeAorxeB).

Recall

or:

or

\"X

that

\"or\"

\316\245
is

true\"

by

in mathematics
to inclusive
is true,
\"either X is true, or \316\245

default

that

means

See SectionA.l.

are true\".

or both

refers

Example3.1.11.

set

The

or

{2,3}

the elementsof this set


simply
we denotethis set as {1,2} {2,3}
are

3.1.12. If A, B, Af

Remark
A

U \320\222
is equal

Definition

3.1.4).

\320\222
is

A\\J

axiom

equal

to A! U
to A U

now

We

give

some

Lemma 3.1.13. If a
If A^B,C
A\\JB
Also,

=
we

if B1

B1. Thus

is a

3. Because of

this,

{1,2,3}.

A is

and

sets,

needs

One

(why?

Similarly

of substitution,

are

\320\222

in both,
2, and

1,
=

of those elements
or in other words

consists

{1,2}U{2,3}

which either lieon {1,2}or in

equal to A', then


Axiom 3.4 and

to use

set which is equal to B, then

the operationof union

obeys

the

and is thus well-definedon sets.


basic properties
and

are

objects,

of unions.
then {a,

b} = {a} U

{b}.

then the union operationis commutative


{i.e.,
=
and
associative
B\\JA)
{i.e., {AuB)UC AU{BUC)).
have
Au A = A U 0 = 0 U A = A.
are sets,

43

3J. Fundamentals
We

proof\302\273

the

just

prove

identity (A U B)

associativity

=
\320\241

and leave the remainingclaimsto Exercise


3.1.3.
By
we
to
show
need
that
\317\207
of
element
3.1.4,
every
Definition
(A(JB)U
of A U (B U C), and vice versa.So suppose
first
is an element
\320\241
an
of
is
element
U
Axiom
this
U
\320\241
means
\317\207
3.4,
By
that
(A B)
A U \320\222
or
\317\207
\320\241
is true.
We now divide
\342\202\254
\342\202\254
that at least one of \317\207
ytU (BUC),

\320\222
then
and
3.4 again \317\207
U \320\241,
\342\202\254
\342\202\254
\320\241,
into two cases. If \317\207
by Axiom
3.4
A
we
have
\317\207
U
Now
Axiom
U
\342\202\254
so by
again
suppose
(B
C).
A U B,
then by Axiom 3.4 again \317\207
A or \317\207
\320\222.
\342\202\254
\342\202\254
instead \317\207\342\202\254
A U (B U C) by Axiom 3.4, while
\320\222
A then
\317\207
if \317\207
\342\202\254
\342\202\254
\342\202\254
If \317\207
\320\222
Axiom
consecutive
of
3.4
we
have
\317\207
\320\241
U
\342\202\254
then
applications
by
A U (B U C).
\317\207
Thus in all caseswe see that every
hence
\342\202\254
and
of
U
A U (B U C). A similar
U
\320\241
lies
in
element
argument
(A B)
shows that every element of A U (B U C) lies in (A U B) U C, and

so (A

=
U \320\241
U \320\222)

A U

U
(\320\222

Because of the above lemma,we


for
thus
to denote multipleunions,

insteadof

U B)

(A

or
\320\241

Remark 3.1.14.
with

also meaningless

(B U C).

the

operation

Similarly

of four

unions

for

has some
are not identical. For

of union

two operations
and 2 + 3 =
{2,3}
to numbers,
pertains

5, whereas{2}+ {3}is
and 2 U

not sets)

(union pertains to sets, not

3 is

numbers).

allows us to define triplet sets,quadruplet


and
sets,
:=
if a, b, \321\201
are
three
we
define
objects,
{a, b, c}
{a}U{b}U
b, c, d are four objects, then we define
{a, b, c, d} := {a} U

a,

{b}U{c}U{d},and soforth.

positionto

define

number n; this
\"n

parentheses
write AuBuC

we can

axiom

This

so forth:
if

A U

to use

need

not

the

{3}

(addition

meaningless

{c};

While

addition,

instance,{2}

do

instance

etc.

sets, AUBUCUD,

similarities

desired.

as

\320\241)

times\",

but

sets
would

the

On

of

consisting

other

of

\316\267-fold

we are not

hand,
for

\316\267
objects

iterating

require

concept

the

the
iteration

any

given natural

construction
yet been

above
has

rigourously defined. For similarreasons,we cannot


consisting of infinitely many objects, becausethat
the axiom of pairwise union infinitely
iterating

yet in a

not
yet

would
often,

define

sets

require
and

it is

3. Set

44

theory

Later
one can dothis rigourously.
which
allow
one
we will introduce other axioms of set theory
sets.
construct arbitrarily large, and even
infinite,
others.
One
some sets seem to be largerthan
Clearly,
way
of a subset
notion
the
formalize this concept is through

stage that

at this

clear

not

3.1.15 (Subsets).Let A,

Definition

denoted

of B,

a subset

\320\241
\320\220
iff
\320\222,

We say

sets.

\320\222
be

every

that

to
to

is

also an

of A is

element

on,

element of 5, i.e.
For

We

say

and

=$>

\320\222.
\317\207
\342\202\254

\320\241
if
a proper subsetof B, denoted\320\220
\316\262,

A is

that

A
\342\202\254
\317\207, \317\207

object

any

\316\222.
\316\221
\317\206

notions

Because these

3.1.16.

Remark
of

the

and

equality

\"is

an

definitions involve
of

element

A =

\320\241
\320\222
\320\220
and

A', then

relation,

both of which
of subset
for

also

instance

\320\241
\320\220'
\320\222.

We have

3.1.17.

Examples

the

only

already obey the axiomof substitution,the notion


automatically obeys the axiomof substitution.Thus
if

\320\241
\320\220
\320\222

{1,2,4}

because

\320\241

{1,2,3,4,5},

Infact
is alsoan elementof {1,2,3,4,5}.
\320\241
since the two sets {1,2,4}
{1,2,3,4,5},
not
Given
equal.
any set A, we always have

every element of {1,2,4}


we

also

have

{1,2,4}

and {1,2,3,4,5}

are

and

\320\220\320\241\320\220

(why?)

\320\241
A

(why?).

of
notion of subset in set theory is similar to the notion
than
or
\"less
as the following Proposition
equal to\" for numbers,

The

demonstrates (for a moreprecisestatement,

3.1.18 (Sets are partially


ordered
\320\241
\320\241
\320\222
\320\222
be sets. If \320\220
and
\320\241
then

Proposition

Let A,B,C
and

see

then

\320\241
\320\222
A,

A =

\320\222. Finally,

if

Definition

by
AQ

\320\241
\320\220
\320\222
and

8.5.1):

set inclusion).
\320\241.If

\320\241
\320\222
\320\220

\320\241
\320\222
then
\320\241

\321\201
\320\220
\320\241.

Proof
and

We

BCC.

element of

prove the first claim.


\320\241
To prove that \320\220
have
we
\320\241,
is an
of C. So, let
element

shall

just

Supposethat
to prove

\320\241
\320\220
\320\222

that every

us pick an arbitrary

45

Fundamentals
\316\231.
\316\262

of
\317\207

element

\316\221.Then,

0f 2?. But then


element of A is

unions: see for

5,

\320\241
\320\222
\320\266
is
\320\241,

since

is

There

Exercise

instance

then

\317\207
must

an

element

of C,

an element

indeed

Remark 3.1.19.

A C

since

relationship
3.1.7.

be

an element

Thus every

of C.

as claimed.
between

subsets

and

the
There is one importantdifference
between
\320\241
two
and
the
less than relation <. Given any
numbers n, m, we know
that
one of them is smaller

3.1.20.

Remark
subset

relation

distinct

natural

two
distinct
than the other (Proposition
given
2.2.13);however,
in
one
of
is
a
subsetof
the
not
true
that
is
them
it
general
sets,
A
:=
:
to
of
For
take
\316\267
be
the
set
even
\342\202\254
other.
instance,
\316\235}
{2n
to be the set of odd
G N}
natural numbers, and \320\222:= {2n + 1 : \316\267

natural numbers.Then neitherset is a subsetof the


is why we say that sets are only
ordered,
partially
ordered
Definitions
natural numbers are totally
(see
Remark 3.1.21.

other.

whereas

This

the

8.5.1, 8.5.3).

also caution
that the subset relation
2 is
element relation \342\202\254.
The
number
2 \342\202\254
but
an element of {1,2,3} but not a subset;thus
{1,2,3},
2
is
not
a
set.
even
while
2 \302\243
Indeed,
Conversely,
{2}
{1)2,3}.
\320\241
thus
is a subset of {1,2,3}, it is not an element;
{2}
{1,2,3}
2 and
the set
but {2} ^ {1,2,3}. The point is that the number
from
It
is
to
sets
are
distinct
objects.
important
distinguish
{2}
as
For
can
have
different
their elements,
instance,
they
properties.
of finite numbers
an
set consisting
it is possibleto have
infinite
(theset N of natural numbers is one such example), and it is also
possibleto have a finite set consisting of infinite objects (consider
\320\241
is

for

the

not

instance

the

should

We

same

as the

finite

set

Q,
{\316\235,\316\226,

R},

which

has four

elements,

all of which are infinite).

out

We

now

of

larger

give an
sets.

axiom which

easily allowsus to createsubsets

3.5 (Axiom of specification). Let A be a set, and for each


\317\207
is either
let P(x)
\342\202\254
be a property pertaining to \317\207
A,
(i.e.,
P(x)
a true statement or a false statement).
a set,
there
exists
Then
Axiom

3. Set theory

46
called

is true} (or simply

: P(x)

G A

{x

whose elementsare precisely


is true. In otherwords,
for
e
\321\203

e A:

{x

as

known

{x

: P(x)

is true}

be as

large as

verify

that

if

then {x G

\317\207
in

A for

(y

the

axiom

is true).

and P(y)

Note that

of separation.

a subset of A

small

for short),
which P(x)

works for specification,


(why?).
{x G A! : \320\240(\320\266)}

: P(x)}

it

though

(why?),

empty set. One can

as the

of substitution

axiom

the

A = A!

as

or

: P(x)}

G A

<==>

is always
A

G A

y,

object

any

This axiomisalso
could

elements

the

is true}

P(x)

{x

thus

:= {1,2,3,4,5}. Then the set {n G S :


\316\267
in S for which
those elements\316\267
< 4 is true,
=
the
set {n\302\243 S :n<7}
i.e., {n G 5 : n< 4}
{1,2,3}. Similarly,
:
5
is
is the same as S itself,while
\316\267
<
the empty set.
G
{n
1}
\316\267
<

set of

4} is the

We

Let S

3.1.22.

Example

write

sometimes

instead of

{x G A\\P(x)}

{x G

P(x)};

colon
\":\" to
when we are using
the
denote
something
for
instance
to
denote
and
domain
of
a
the
function
else,
range
/ :X
Y).
can
use this axiom of specification to define
We
some
further
on
and
intersections
difference
sets.
sets, namely
operations

is useful

this

3.1.23

Definition

sets is

two

defined to be the set

Si
In

other

Si

words,

to both Si

and

3.1.24.

obeys the

\316\240
S2

\316\240
S2

Thus,

S2.

\317\207
G

Remark

The intersection Si

(Intersections).

Si

:=

{x G Si

consists

for all

\316\240
<=>
\316\257>2

:x

of all

which belong

\320\266,

and
\342\202\254
Si

Note that this definition

\317\207
G

is

S2.

well-defined

axiom of substitution, seeSectionA.7)

known
future

definitions

explicitly again.

in this

chapter and will

usually

not

(i.e.,

it

it

is

because

in terms of more primitive operationswhich


to obey the axiomof substitution. Similar
remarks

defined

of

S2}.

the elements

objects
x

\316\240
S2

were

already
apply

be mentioned

to

47

Fundamentals
\316\231.
\316\262

3.1.25. We have {1,2,4} \316\240


{2,3,4}
=
and
U0
{2,3}
{2,3},
{2,3}\316\2400=

Examples
=

0,

{3j4}

By the

3.1.26.

Remark

way,

{2,4},

{1,2}

\316\240

0.

with the
either union

careful

be

should

one

it can
mean
Englishword \"and\":rather confusingly,
on
For
context.
if one talks
intersection,
instance,
depending
or
of
and
one
of a set of
means
the
union
set
a
\"boys
girls\",
about
a set of girls, but if one talks
about
the set of people who
with
boys
of the set of
and
then
one
means
the
intersection
male,
are single
set
with
the
of
work
male people.(Can you
out the
single people
\"and\"
means
when
union
and
rule of grammar that determines

when
is

means

\"and\"

could say that


of {2} and the
in

elements

of

to denote addition, thus


for
instance
one
while
also
that
\"the
elements
5\",
saying
and
\"the
elements of {3} form the set {2,3}\"
and {3} form the set 0\". This can certainlyget
reason
we resort to mathematical symbols instead

One

is that
as \"and\"
and unambiguous

such

mathematical

carefully

meaning, whereas one must


out
what
at the context in order to work

words

English

always

have

a precise

often

look

very

Two sets \320\220,\320\222are said to


is not the sameconcept
the
of

and

disjoint

Definition 3.1.27
the

define

=
\316\222
\316\221
\316\240

Note

0.

\316\221
\316\222.For
\317\206

distinct,

being

set

intersection is non-empty).
but not distinct (why?).

their

are

as

sets {1,2,3} and {2,3,4} are distinct(there


are
set which are not elements of the other) but

(because

disjoint

sets0

one

if

be disjoint

that this
elements

symbols

word means.

an English

instance,

\"and\"

3 is

and

\"2

{2}

confusing!

is that

problem

in English

used

also

Another

intersection?)

sets).

Given

to be

the set

(Difference

\342\200\224
\320\222
or

A\\B

two
A

with

not

Meanwhile,the

sets A and
any

i?, we

elements

of

\320\222
removed:

A\\B
for

instance,

be a

subset

We now
and

difference

:= {x

{1,2,3,4}\\{2,4,6}

of

A,

but

not

A : \317\207
\342\202\254
# \320\257};

= {1,3}.

In many cases \320\222will

necessarily.

give some basic propertiesof unions,


sets.

intersections,

3. Set

48
3.1.28

Proposition
be

let X

and

sets,

theory

\320\241
(Sets form a boolean algebra). Let \320\220,\320\221,
as subsets.
be a set containingA^B^C

We have

element)

(a)

(Minimal

(b)

(Maximal element)

(c)

(Identity)

(d)

(Commutativity)

(e)

(Associativity)

We have

=
\320\220\320\223\\\320\220

A and

have

We

and

Au

\316\221
\316\240
0 =

and
=

(A U B) U \320\241=

0.

=
\316\221\316\240\316\247

A.

A.

and AnB

= BUA

AUB

have

We

= A
= X

AUX

have

We

U 0

A U

\320\222\320\237\320\220.

(B U C)

and

(\320\220\320\237\320\222)\320\237\320\241
\320\220\320\237(\320\222\320\237\320\241).

and

\320\233
U

(\316\233)(Be

and

Remark
of the

X\\(A

De

Morgan

basic laws

laws

U
\316\240
(\316\221 \316\222)

assert

this

the

after

them

as

3.1.6.

The reader may


U and

between

relation

complementation
that

= (X\\A) \316\240
(\316\247\\\316\222)

who identified

(1806-1871),

0.

of set theory.

a certain

observe

symmetry
X and 0. This

between

and
\316\240,

- two
an exampleof duality
distinct
properties
dual to eachother. In this case, the duality
the

=
\316\221\316\240
(\316\247\\\316\221)

named

are

laws

\320\237
(\320\220
\320\241)

\320\241).

= X and

AU (X\\A)

3.1.29. The de Morgan

3.1.30.

above

\316\240
\316\262)
(A

\320\241)

We have X\\(A U B)
=
U (X\\B).
\320\237\320\222) (X\\A)

Exercise

the

\316\221
U
\316\240
(\316\222

laws)

Morgan

Proof. See
Remark

have

We

Augustus

logician

(A U

\316\240
(\320\222 \320\241)

(\317\201)(Partition)

one

We have

(Distributivity)

(/)

relation

i-\302\273

X\\A;

converts unions

the

in

is

or objects being
is

de

manifested

Morgan

into intersectionsand

by
laws,

vice

The
above laws
(It also interchanges X and the empty
set.)
are collectively known as the laws
Boolean
after the
algebra,
of
mathematician
Boole
and
arealso
applicable
George
(1815-1864),
of other objects other than sets;it plays
to a number
a particularly

versa.

important

role in logic.

49

Fundamentals

q 1.

basic things

of the

One

yet.

many things we are not ableto do


we wish to dowith a set is take each of

are still

there

but

sets,

about

the objects of that set, and


into a new object; for instance
of

{3,5,9},
This is not

set

{4,6,10}.

we

may

such object

start with a

wish to

set

each one, creatinga new

something we can do directly

so we need a new

axioms we already have,

the

each

transform

somehow

and increment

say

numbers,

of axioms and results

a number

accumulated

now

have

We

using

only

axiom:

Let A be a set For any object \317\207


\342\202\254
(Replacement).
a statement
and
P(x, y)
any object y, suppose we have
A
that
each
\317\207
A
such
there
is at most
and
\302\243
\317\207
to
for
y,
pertaining
is true.
Then there exists a set {y :
which
P{x,y)
one \321\203
for
3.6

Axiom

^rue
\342\204\242
\321\203)

P(x>

4=>

statement

\321\203

:=

is the
sc-H-, i.e., \321\203
one
is exactly

{y :

\321\203

same

the

clearly

for some

X++

Example 3.1.32. Let A


statement

for

1.

\321\203

which
:

is true
\320\240(\320\266,
\321\203)

{\321\203 \321\203

1 for

some

which

\317\207
A.
\342\202\254

is
\320\240(\320\266,
\321\203)

true

exists;

in this

every

above axiom assertsthat

the

case, it is

(why?).
and let

{3,5,9},

again for

Then

z,
object
\320\260\320\277\320\243

and let P(x, y) be the


of x. Observe that for

\317\207
\342\202\254
{3,5,9}}

as {4,6,10}

set

some

for

{3,5,9},
successor

Thus the

of x.

successor

the

specifically,

true

for
\321\203

for

some \317\207
\342\202\254
A}

true for

there
\317\207
\342\202\254
A,

set

^la^

such

is
\320\240(\320\266,
\316\266)

Let

3.1.31.

Example

\342\202\254
A} *

y) is

: P(x,

\316\266
e{y

some

for

specifically,
is

\317\207
\342\202\254
{3,5,9}}

P(x,y) be the

is exactly one \321\203


the number 1. In this case
just the singleton set {1};

G A,
every \317\207

there

3,5,9 of the originalset A by the


1.
Thus
this rathersilly example
shows
that
object, namely
the set obtained by the above axiomcan
be
\"smaller\"
than
the
set.
original

we

have

replaced

each

element

same

We

often

abbreviate

a set
=

:
{\320\243\320\243

f(x)

of the

form

f\302\260r
some

\317\207
\342\202\254
A}

Set theory

3.

50

as {f(x) :
=

A
can

of

axiom

of

{f(x)

by starting with the set A,

is true}

\342\202\254
A\\P(x)

of specification

set {x

to create the

A
:
\342\202\254

We

as

the

is true},

:x

{f(x)

is true}.

P(x)

if

using

P(x)

to create
applying the axiomof replacement
Thus for instance {n-H-: \316\267
\342\202\254
{3,5,9};

and then
A]

instance,

combine

specification,

axiom

for

Thus

\342\202\254
A}.

{f(x)\\x

set {4,6,10}.
is the
\302\243
A}
{x++ : \317\207
the axiom of replacement with the
thus for instance we can create sets such

then

{3,5,9},
course
: x

or

\317\207
\342\202\254
A}

\316\267
<

\316\225

6} =

{4,6}.

In many of our exampleswe


are

numbers

natural

in fact

have

that

assumed

implicitly

objects. Let us formalizethis as follows.

There exists a set N; whose


elements
are
an
in
an
as
as
0
and
well
numbers,
object N;
object
such
that the Peano
\342\202\254
n++
N;
assigned to every natural number\316\267
axioms (Axioms 2.1 - 2.5)hold.
3.7

Axiom

of infinity

axiom

infinite

an

of

axiom of

infinity

union
as

{3,5,9}
to

has

One

set; for

of that

we

with

and infinite mean in


see

Section

such as

numbers

that

(We

will

Prom

the

N.

3.6.)

3, 5, 7, etc.

instance, the set {n+ 3


expression

: \316\267
0 <
\342\202\254
\316\235,

function

or

n +

3.

<
\316\267
We

5} is

not

emphasize

an example:

3.1.33. (Informal)This examplerequires


the
of subtraction,
which has not yet been formally
introduced.
following two sets are equal,
Example

{n+ 3 :

\316\267
0
\342\202\254
\316\235,

(see below),
never

are

and
theory, and so (from the pair set axiom
axiom) we can indeed legitimatelyconstructsets
as we have been doing in our examples.
the
elements
keep the concept of a set distinctfrom

the same thing as the


this

the set of natural numbers

in set

objects

pairwise
such

set, namely

what finite

formalize
indeed

version of Assumption 2.6. It is called


because it introduces the most basicexample

formal

more

the

is

This

the

(Infinity).

natural

called

equal

<

<
\316\267

5}

= {8

even though the


to each other for

: \316\267
\316\267
0
\342\202\254
\316\235,

expressions +

\316\267

any

natural

<

<
\316\267

3 and

number

n.

notion

The

5},

(3.1)

\342\200\224
\316\267
are

Thus, it

51

3.1. Fundamentals

is a goodideato
you

One

elements.
the

\316\267
is

letter

those

use

braces {}

curly

accidentally confuse a

set

when

sets,

lest you

reason

for this counter-intuitive situation is that


used in two different ways on the two
sides

about

talk

to

remember

being

its

with

the situation, let us rewrite


:
the
set {8 - \316\267
<
<
\316\267
\316\267
the
the
letter
thus
letter by
0
\342\202\254
m,
\316\267
\316\235,
5} by replacing
\342\200\224
<
<
:
m
m
m
0
This
is
the
\342\202\254
sameset
N,
exactly
5}.
giving {8
so we can rewrite (3.1) as
before
as
(why?),
of

To clarify

(3.1).

{n+ 3 :

0 <

N,

to see

is easy

it

Now

\316\267
e

<
\316\267

= {8

5}

- m :m

(using (3.1.4)) why

this

\342\202\254
N, 0

< 5}.

< m

is true:

identity

every

is a natural
number between
number of the
3, where \316\267
m where
m := 5 \342\200\224
m
\316\267
that
0 and 5, is alsoof the form 8 \342\200\224
(note
a
natural
0
and
also
numberbetween
therefore
is
conversely,
5);
8 \342\200\224
of the form
\316\267
is a natural
number
where
m,
every number
\342\200\224
is
of
form
n
also
the
n:=
m
and
where
5
+
between \316\260
5,
3,
(note
a natural
is therefore
number between 0 and 5). Observe
that \316\267
the above explanation of (3.1) would
much
more
how
confusing
if we had not changed oneof the n's to an m first!
been
have
\316\267
+

form

Exercise 3.1.1.

Show that

and

transitive.

symmetric,

the

Exercise 3.1.2. Using only Definition


prove that the sets 0, {0}, {{0}},and
two

them

of

Exercise3.1.3.

A\\JB

the other

the

claims

previously in
Exercise
Furthermore,

a similar

that

remaining
sets.

the three statementsACB,

Show that

logically equivalent (any

one of

them

implies

two).

Exercise 3.1.6. Prove


these

in Lemma

in Proposition

A are

is reflexive,

3.1.13.
3.1.18.

claims
claims

\320\222
be
\320\220,

=
\316\221
\316\222
\316\240
\316\222,

{0,

remaining

the

Prove

in (3.1.4)

Axiom 3.2, and Axiom


3.3,
{0}} are all distinct
(i.e., no

3.1.4,

to each other).

are equal

Exercise3.1.4.Prove
Exercise3.1.5.Let

of equality

definition

Lemma

3.1.7. Let
show

others.

Some of

(Hint:
claims

the

one can use some of


have also appeared

3.1.13.)
be
\320\241
\320\220,
\320\221,

that

CCA

spirit, show that

\320\241
\320\233
and
\320\241

3.1.28.

Proposition

to prove

\320\241
\320\221
if
\320\241

sets.

and

Show that
\320\241
\320\221
if
\320\241

AC.AUB and
and

only if A

\320\241
A and
\320\220\320\237\320\222

and

only if

\320\241.
\320\222
AUB,
\320\241
\320\222
\320\241.

AnB

\320\241
\320\222.

\320\241
\320\221.
\320\241
\320\220
In
\320\237

and

furthermore

Set

theory

the absorption laws An(A\\JB)

3.

52

be
\320\222
Exercise 3.1.8. Let \320\220,

A U

\320\220
and

\316\240
5)
{\316\221

Exercise 3.1.9.
that

Show

Let A,
X\\B

=
\320\221

Exercise 3.1.11. Show that


of

A U

B.

implies the

of replacement

axiom

the

is

union

axiom

specification.

Russell's

3.2

of

Many

similar

paradox

elements
one

but

introducing

previous section have a


a set consistingof all the
form
they
which
have a certain property. They are both plausible,
think
that they could be unified, for instance
might
by
the following axiom:

every object

a property

have

is true}

P(x)

This axiomisalso

known

as

the

<==> P(y)
axiom

talk about the setofall

blue

discovered

by

(1872-1970)
in

1901.

the

philosopher
The

paradox

set; if

4=>

is
\"\317\207

every object

It

the

objects,

This

into

set of

axiom

(Exercise
as

that

assumed

we

set

all

also

3.2.1).
theory,

RusselVs

and logician Bertrand Russell


runs as follows. Let P(x) be

statement

P(x)

(so

statement).

is true.

natural numbers,the setofall sets,and so forth.


section
implies most of the axiomsin the previous
this
be
axiomcannot
introduced
Unfortunately,
because it creates a logical contradictionknown
paradox,

that

\317\207

of comprehension.

to a

corresponds

property

every

axiom, we could

P(x) pertaining to

is eithera true statementor a false


a set {x : P(x) is true}
such
that for

\342\202\254
\321\203
{x :

for

Suppose

(Dangerous!)

specification).

(Universal
\317\207
we

for every x, P(x)


Then there exists

that

in the

us to

allow

both

Axiom 3.8

asserts

(Optional)

introduced

axioms

the

flavor:

the

0.

three sets A\\B,

that the

Show

sets.

and that their

are disjoint,

and
\320\220\320\237\320\222,
B\\A

=
\316\222
\316\221\316\240

and

X\\A.

\320\222
be

and

Al)B = X

sets such that

X be

B,

and

3.1.10. Let

Exercise

Prove

sets.

= A.

a set,

and

\317\207
\320\266\";
\302\243

Russell7s

3.2*

paradox

i.e., P(x)IS true only

when

\320\266
is

For instance,P({2,3,4})is
of

elements

three

the

53

(Optional)

2, 3,

be the set of all sets


of
axiom
the
specification),
so P(S)
it is an elementof 5,

:=
\316\251

the

of all

set

the

i.e.,

itself,
a set and

\316\251
contain

other

the

\316\251
^ \316\251.On

is

i.e.

itself,

and

\317\207
&

\316\251
\316\251?If
\342\202\254

this meansthat

by definition

then

set

\320\266},

sets which do not containthemselves.


Now

does

question:

: \317\207
is a

= {x

is true}

{\317\207 P(x)

the axiomof

the set

create

to

Now use

is false.

and

from
itself a set,

S is

since

then

if we

to exist

know

would

we

(which

universal

specification

set

the

since

4 of {2,3,4}. On the otherhand,

let S

universal

not contain itself.


{2,3,4} is not one

which does

a set

true,

if

hand,

ask

contain

\316\251
did

is
\316\241(\316\251) true,

\316\251
did

is
i.e., \316\251
contain
itself,
in either
case

not

\316\251.Thus
would
be true,
and hence \316\251
\342\202\254
then \316\241(\316\251)
\316\251
\316\251
and
which
is
absurd.
\342\202\254
we have both \316\251
^ \316\251,
The problem with the above axiomis that it createssetswhich
far too
are
\"large\" - for instance, we can use that axiom to talk

about the set of all objects(a so-called


sets are themselves objects (Axiom
3.1),
which
is
allowedto contain themselves,
One

affairs.

way

to

\"universal
this

sets are
silly state of

that

means

a somewhat

resolve this

informally

Since

set\.")

issue is to

of

think

of the
arranged in a hierarchy. At the bottom
are
not
are
the
the
that
sets1,
primitiveobjects objects
hierarchy
Then
on the next rung of the
37.
such as the natural number
as being

objects

hierarchy there are setswhose


objects, such as {3,4,7}or the

consist

elements
empty

set

only

0; let's call

of primitive

these

sets whoseelementsconsistonly
of primitive
and
primitive sets, such as {3,4,7,{3,4,7}}.
objects
The
Then
we
can
form sets out of these objects, and so forth.
sets
that
at
each
of
the
we
see
whose
is
point
hierarchy
stage
only
of the hierarchy,
elements consist of objects at lower
and so
stages
at no stage do we ever construct
a set which contains itself.
To actually formalizethe above
of a hierarchy
of
intuition
is actually
rather
and we will not do so here.
objects
complicated,
\"primitivesets\"

*In

pure

for

set

now.

theory,

one primitive set

0 on

there are

Then

there
the

next

will

be

no primitive

rung of the

hierarchy.

objects,

but

there

will

be

3.

54

we shall simply
postulate
absurdities such as Russell'sparadox

Instead,

3.9 (Regularity).

Axiom

at leastone element\317\207
of

If

is

set, then

a non-empty

is either

A which

occur.

not

do

theory

ensures that

which

axiom

an

Set

there is

not a set, or is disjoint

A.

from

the
axiom (which is also known
as
axiom
is that it is assertingthat at least one of the
of foundation)
A is so low on the hierarchy of objects that it does
elements
of

The

not

of this

point

contain

other

the

of

any

of

elements

For

A,

instance,

if

{{3,4}, {3,4, {3,4}}},then the element{3,4} A does not


contain any of the elementsof
3 nor 4 lies in A),
(neither
the
element
being somewhat higher in the
{3,4, {3,4}},
A =

\342\202\254

although

does

hierarchy,
particular

to

themselves

(Exercise

of

{3,4}.

namely

Ay

One

sets areno longerallowed

is that

axiom

of this

consequence

contain

an element

contain

3.2.2).

we
need this axiom
One can legitimatelyask whether
really
than
our
in our set theory, as it is certainly lessintuitive
other
For the purposes of doing analysis,
it turns
out in fact
axioms.
all the sets we consider in analysis
that this axiomis never
needed;
low
on the hierarchy of objects, for
are typically
instance
very
sets
of
of
or
setsof
sets
primitive objects,
being
primitive
objects,
or at worst sets of sets of sets of primitive
However
it is
objects.
this
axiom
in
to
to
include
order
more
advanced
necessary
perform
set theory, and so we have included
this axiom in the text (but in

an optionalsection)

for

Exercise 3.2.1.
if

assumed

to be

Show that

sake

the

of completeness.

universal

specification

true, would imply Axioms

axiom, Axiom

3.2, 3.3, 3.4,3.5,and

3.6.

3.8,
(If

natural numbers are objects,we also obtain Axiom


would
this
axiom, if permitted,
simplify the foundations of
3.7.) Thus,
be
set theory
can
viewed
as
one basisfor an intuitive
tremendously
(and
as \"naive
set theory\.") Unfortunately,
model of set theory known
as we
have seen,Axiom
3.8 is \"too good to be true\"!
we assume

that all

Exercise3.2.2.Usethe axiom of regularity (and the singleton set axiom)


show that if A and
to show that if A is a set, then A\302\243A. Furthermore,
\320\222
are

two

sets,

then either

or \320\222
& \320\222
& A

(or both).

55

Functions

3.3.

Exercise3.2.3.Show
the

universal

axiom,

specification

postulating

of a

existence

the

we
(i.e., for all objects \320\266,

a universal

then

true,

then Axiom
axiom of universal

3.8

is true.

Axiom

3.8,

that

theory)

axiom

to an

of all objects
if
Axiom 3.8 is
words,
if a universal
set exists,
Axiom
3.8 is called the

\316\251
consisting

other

\317\207
G \316\251).In

set

is equivalent

set\"

\"universal

have

axioms of

other

the

(assuming

set exists, and conversely,


(This may explain why

Note that if a universal set \316\251


existed,
specification).
\316\251
\316\251
have
Axiom
Exercise 3.2.2.
would
\316\225
we
3.1, contradicting
by
of foundation specifically rules out the axiom of universal
the axiom

then
Thus

specification.

Functions

3.3

In orderto do

it is not
useful to just have
particularly
from
one
also needthe notionof a function
\342\200\224\342\226\272
:
\320\243
from
set
a
function
X
one
another.
to
set
Informally,
/
is an operation
which assigns to each element
X to another set \320\243
\317\207
in
a
element
X,
single
(or \"output\") f(x) in \320\243;we
(or \"input\")
used this informal concept in the previous
have
already
chapter
is
we discussed
the natural numbers. The formaldefinition
when
analysis,

of a

notion

the

set; we

as follows.

(Functions). Let

3.3.1

Definition

be a property pertainingto an
for every

that

such

\317\207
G

X,

\316\245
be
\316\247,
\317\207
G

object

(this is sometimesknown
P(x,y)
: X
Then we define the function
f
is true

domain
\317\207
G

X,

X and
assigns

object f(x)

for

range

\316\245
to

an output
which

be

the

f(x)

an object
\320\243
for
\342\202\254
\321\203

vertical

\342\200\224>
\320\243
defined

object
G

and

one
the

as

by

G \320\243,
\321\203

which

line test).
\316\241
on

the

which, given any input


to be the unique
G X and
Thus, for any \317\207

\320\243,defined

is true.

P{x,f{x))

is exactly

there

let P(x,y)

and

sets,

yeY,
=

\321\203

f(x)

Functionsarealso

referred

depending
phisms,

on
although

the

to

as

They are
be more precise,

context.
to

<==> P(x,y)
maps

class

of object,

or transformations,

also sometimes called mora morphism refersto a more

which may or may


the context.
on
functions, depending
general

is true.

not

correspond

to actual

3. Set

56

theory

property that
which

for
\321\203

this

to this

there

is exactly

Thus

we can

as

output.

g : N

y-H-

natural

\342\200\224>

to the property
\320\266,
i.e.,
P(x,y) defined by y-Hbe the number whose incrementis x. Unfortunately
= 0 there is no
not define a function, because when \317\207

N\\{0}
=

forth.

so

and

is equal to \317\207
whose
natural number\321\203
increment
(Axiom
a decrement
the other hand, we can legitimatelydefine

h :

takes

which

function

decrement

one

define

so that f(x)

property,

f(2n + 3) = 2n+

define

be the

P(x,y)

would
does

\321\203 x-\\-+.

namely

/(4) = 5,

might also hope to

associated

g(x)

\317\207
G

let

is the increment function on N,


as input and returns its increment

number

for instance

Thus

\316\235,and

for each

\342\200\224>
N
associated

x\\ this

all

=
\316\235,\316\245

Then

sc-H-.

\321\203

: N

a natural

One

is
\320\240(\320\266,
\321\203) true,

function

for
\320\266-\320\235-

Let X

3.3.2.

Example

when

because
\320\266,

\317\207
G

that

such
\321\203

number

the

to

\342\200\224>
N
associated

N\\{0}

y++

Thus for instance \320\233(4) 3


undefined since0 is not in
=

and h(2n

+ 3) =

domain

the

defined by
exactly one

is indeed
to

thanks
\320\266,

On

function

P(x,y)

property
there

2.3).

2.2.10.

Lemma
2), but

+
\316\233(2\316\267

h(0) is

JV\\{0}.

(Informal) This examplerequiresthe real


define in Chapter 5. One couldtry to define
a square root function
R by associating
it to the property
^j : R \342\200\224>
= x, i.e., we would want
defined
to be the
by y2
y/x
\320\240(\320\266,
\321\203)
=
number \321\203
such
that
x. Unfortunately there are two problems
y2
this definition from actually creating a function.
which
prohibit
is that there exist real numbers\317\207
The
first
for
which
is
P{x,y)
\342\200\224
=
1 then
never true, for instance
there
is no real number
if \317\207
=
that
x.
such
however
can
This
be solved
\321\203
y2
problem
by
the
domain
from R to the right half-line [0,+oo). The
restricting
secondproblem
is that
even when \317\207
it is possible for
G [0,+oo),
=
for
there to be more than one \321\203
in the
R
which
x, for
range
y2
3.3.3.

Example
numbers

R,

we will

which

instance if

=
\317\207

both
i.e.,
\320\240(\320\266,\321\203),

can
Once

does

one

function

y/x

be

however

^j

is the

: [0,

4 then
+2
solved

this,
+oo)

both
and

\321\203

\342\200\2242
are

2 and
square

by restricting

then

one can

\342\200\224>

unique number

[0,

+oo)

G
\321\203

\321\203

\342\200\224
2

roots

obey

of 4.

the

property

This problem

the range of R to [0,+oo).


correctly define a squareroot

using

[0, +oo)

the relation
such that y2

\321\2032

= x.

thus
\320\266,

57

3.3. Functions
One common

a function

define

to

way

is simply

to

its

specify

one generatesthe output


from
f(x)
known
as
an
a
d
efinition
of
function.
explicit
each input;
be
defined
For instance, the function / in Example 3.3.2could
that
has
domain
and
to N,
/
range
equal
by saying
explicitly
^
:=
N.
In
f\302\260r
aU
other
cases
we
define
a
G
\320\266-\320\235and f(x)
only
the input
\317\207
links
function / by specifying what property \320\240(\320\266,
\321\203)
this
is
an
a
definition
of
function.
the
with
output
implicit
f(x)\\
the square root function y/x in Example3.3.3was
For
instance,
an implicit
defined
by the relation (y/x)2 = x. Notethat
implicitly
if
is
valid
we
know
that
for
definition
only
every input there is
which
the implicit
relation. In many
obeys
exactly one output
the
domain
and range of a function for
cases we omit specifying
function
and
thus for instance we could refer to the
/ in
brevity,
\\->
\"the
\317\207
function
Example 3.3.2 as \"thefunctionf(x) := sc-H-\",
its

domain,

and how

range,

this is

function

\"the
\321\217-\320\235-\",

too

\302\267
However,

\"-H-\"

it is
sometimes

of this

much

to

important

the

even

or
\320\266-\320\235-\",

know

extremely

abbreviated

abbreviation can be dangerous;


the domain and range of
what

the function is.

inputs

not

that

observe

We

=
\317\207

functions
=

obey

the axiom of

substitution:

In other words, equal


f(xf) (why?).
f(x)
On the other hand, unequalinputs
equal
outputs.
imply
ensure
unequal outputs, as the following
necessarily
example
\320\266',then

if

do

shows:

Let X =

3.3.4.

Example

=
\316\235,\316\245

\316\235,and

let

P(x,y)

be the

is
there
G N
property that \321\203 7. Then certainly for every \317\207
the
Thus
is
number
one
for
which
7.
namely
exactly
\321\203
\320\240(\320\266,
\321\203) true,
N
we
can
a function
to this property;
create
associated
/ : N \342\200\224>
it is simply the constant function
the output
of
which
assigns
=
N.
\317\207
it is certainly possible for
7 to each input
Thus
G
f(x)
the same output.
different
to generate
inputs
=

Remark
different

clarify

14

with

() to denote several
are using them to
we
hand,
things
2 + (3 x 4) =
the order of operations (comparefor instance
use
4 =
20), but on the other hand we also
(2 + 3) \317\207
3.3.5\302\273

We

are

now using

in mathematics;

parentheses

on one

3.

58

Set theory

of a function or of a
parentheses to enclosethe argument
f(x)
the
two usages of parentheses
However,
property such as \320\240(\320\266).
if \316\261
is a
are
For instance,
from
context.
usually
unambiguous
ax (b + c), whereas
then a(b + c) denotesthe expression
number,
the
denotes
if / is a function,
then
+
output of / when the
c)
f(b
is denoted
the
of
a function
is b + c. Sometimes
input
argument
a sequence
for
of
instance,
by subscripting insteadof parentheses;
a
natural numbers ao, ai, a2> \320\260\320\267,...
function
is,
speaking,
strictly
\316\271\342\200\224>
\\-> an rather
than \316\267
from N to N, but is denoted
by \316\267
a(n).

Strictly speaking, functions are not sets,and sets


an object
are not functions;
it does not make senseto ask whether
\317\207
sense
element
of a function /, and it doesnot make
is an
to
a
set
A
to
an
\317\207
to
create
On
the
an
output
apply
input
A(x).
\316\245
other hand,
it is possible
to start with a function / : X \342\200\224>
: x G X},
and
which describes, the
construct
its graph {(\320\266,f(x))
3.3.6.

Remark

functioncompletely:
seeSection3.5.
first

The

is that of

notion

Definition 3.3.7
\320\243,
g
equal,

: X
f

and g(x)

the

3, if and

agree

not consider/

only

of

functions).

same

domain

and

if

f(x)

of

values

some

for

to

and notions for

functions.

equality.

(Equality

\342\200\224>
\316\245
with

basic concepts

some

define

now

We

Two

functions

/ :X

\342\200\224>

and range are said to be


X.
\342\202\254
(If f(x)
g(x) for all \317\207
not
then we do
but
\320\266,
others,

be equal2.)

\316\271->
\316\271-\302\273
3.3.8. The functions \317\207
x2 + 2x +1 and \317\207
+1)2
(\317\207
\316\271->
\316\271->
are equal on the domain R. The functions
\317\207
\317\207
and
\317\207
|\320\266|
on the positive real axis, but are not
are
on R; thus
equal
equal
the concept of equality of functions can dependon the choice
of

Example

domain.

3.3.9.

Example
empty

set X.
2In

functions

function

Since

the

Chapter

being

: 0

rather

empty

function is the
set
to
an arbitrary
empty
has no elements, we do not need

boring

\342\200\224>
X

set

from

example

shall introduce a weaker notion


almost
everywhere.
equal

19, we

of a

the

of equality,

that of

two

59

Functions

3.3.

what / does to

to specify

as the

just

Nevertheless,

input.

any

function
is a function, albeit not
empty set is a set, the empty
Note
one.
that
for each set X, there is
a particularly interesting
function
from 0 to X, since Definition 3.3.7assertsthat
one
only
from
0 to X are equal (why?).
functions
all

fundamental

usual axioms(Exercise
3.3.1).
for functions is composition.

obeys the

of equality

notion

This

available

operation

Definition 3.3.10 (Composition). /

: X

Let

be

two

such that

functions,
of g.

domain

g :

\316\245
and

define the composition


f :
g \320\276

We then

g and / to be the

two functions

->

->
\316\245

\316\226

the range of / is the sameset as the


defined

function

the

\342\200\224>
\316\226
of

by the

explicitly

formula

(g \320\276
f)(x)
range of / does not

If the

compositiongo f
substitution

and

5, we

of

that composition obeys the

Let / : N

3.3.11.

let g : N

domain

the

leave the

axiomof

3.3.1).

(Exercise

Example

match

undefined.

to check

is easy

It

:= g(f(x)).

-> N be

function

the

g(n) :=

function

the

\342\200\224>
N
be

:=

f(n)

Then

3.

\316\267
+

2n,

\320\276 is

the function
=

g(f(n))

g \320\276
/(1)
the function

= 5,

9
thus

Meanwhile,

thus

instance

for

\320\276is
\320\264

\320\276

for

\320\276

f(n)

g(n)

instance

composition

and

\320\276

/(2)

= 7,

3,

and so

f(g(n))

= f(n

+ 3) = 2(n + 3) =

= 8,

=
\320\276
\320\264{2)

\320\276

g(l)

The above exampleshows


commutative:
fog

= 2n +

=.g(2n)

that

10, and

so forth.

the samefunction.

Lemma 3.3.12 (Compositionis associative).


Let
h:

\342\200\224\342\226\272
\316\245
and
\316\226,

\316\226
\342\200\224>W
be

functions.

2n + 6,

is not

composition

gof are not necessarily


is still associative:

forth.

Then

fo(goh)

However,

: X

\342\200\224>Y,
g:

= (fog)

oh.

3. Set

60
Since

Proof.

\320\276
ft

to W. Similarly /

from X

function

and hence(/

from

a function

is

W,

a function

\320\276is

from X

a function

\320\276 \320\276
h is
\320\264)

\316\245
to

to W. Thus /

verify that (fo(go

h))(x)

((/

for all

\320\276 \320\276
\320\264) h)(x)

to Z,

from X

\320\276 \320\276
(\320\264 h)

to

(fog)

is

\320\276 \320\276
(\320\264 h)

oh have the same domainand range.In order


3.3.7
that
are
equal, we seefrom Definition
they

and

that

theory

check

to

we have

x\342\202\254X.

But

by

3.3.10

Definition

= f((goh)(x))

(fo(goh))(x)

= f{g(h(x))

= (fog)(h(x))

= ((fog)oh)(x)

as

desired.

Note that

3.3.13.

Remark

expression go/, the


/ first, before
because traditionally

of / in the
the
go f applies
right-most function
is often confusing at first; it arises
g. This
a
function / to the left of its input \317\207
place

applying

we

than

rather

to the
in which

notations

we would

thus

while g appearstotheleft

function

right. (There are somealternatemathematical


of the
the function is placed to the right

write xf insteadof /(\320\266),

to be more
than
become
particularly
yet
popular.)
often proven

We

now

functions,

certain

describe
onto

clarifying,

confusing

functions,

3.3.14 (One-to-onefunctions).
if different
to-one (or injective)
elements
=\342\226\272
\321\205\321\204\321\2051

Equivalently,

\317\206
f^).

fix')

=\342\226\272

x'.

as

one-to-one

function

map to different

is one-to-one if

a function

f(x)

f{x)

has

has not

and

special types of functions:


functions.
and invertible

Definition

input,

notation

this

but

/ is

one-

elements:

61

Functions

3.3.

Example 3.3.15.

to the natural numbers,defining


function
the
:=
is
now
a
then
one-to-one
function.
g
n2,
g(n)
of a one-to-one function depends not just on what
the notion
function
does, but also what its domainis.

g : N
Thus
the

\342\200\224>
\316\226
by

can

then one
=
\320\224\321\217)

distinct

find

/(x7),

output. Becauseof this,


one-to-one.

we

Definition3.3.17
if

jective)

/(-X\")

/ to some

For

in the
A :=
defined

not

depends
its

what

just

\317\207
G

The

(Informal)

function

if

we

restrict

are

y.

\342\200\224*
\316\226
\316\226
defined

the range

\316\226
to

are

not

the

set

\342\200\224>
A
function g : \316\226
numbers,
onto
now
onto.
Thus
the notion of an
on what the function does, but also

then the

is.

range

3.3.19.

Definition 3.3.20
which

numbers

The concepts of injectivity


dual
to eachother;seeExercises
ways
some evidence of this.

Remark
in many

applying

f(x) =

that

such

because the negative

not onto

n2 is

image of /. However,
of
\342\202\254
{n2 : \316\267
\316\226} square
:=
n2 is
by g{n)

function

exists

there
e \316\245,
\321\203

Example 3.3.18.
/0\316\214

onto (or sur-

from

\316\245
comes

to one

instead of

element in X:

every

:=

in

such that

map

/ is

A function

element

every

\320\243,
i.e.,

which

/ is two-to-one

that

say

one-to-one,

domain X

inputs

functions).

(Onto

two

find

not

\342\200\224*
\316\245
is

in the

xf

\317\207
and

one can

thus

/ : X

function

If a

3.3.16.

Remark

for

\342\200\224*
\316\226
\316\226
defined

this function

restrict

by

because the distinct elements


1. On the other hand,
if we

one-to-one

same element

the

to

1
\342\200\2241, map

not

is

:\342\200\224
^2

f(n)

by

function

The

(Informal)

both

and onto

are

surjectivity
3.3.4,

3.3.2,

/ :

Functions

functions).

(Bijective

one-to-one

and

3.3.5

\342\200\224>
\316\245

are also calledbijective

or

invertible.

Example
/(0)
bijective

:=
because

3.3.21.
3, /(1)
if we

Let / :

:= 3, /(2)
set

\321\203

{0,1,2} -> {3,4}

be

:= 4.

3, then

This

function

there is

the

is

function

not

more than one \317\207


in

Set theory

3.

62

{0,1,2} such that

let g :

be

which

because

bijective

g(x)

of injectivity). Now
function
the
g(0) := 2, g(l) := 3;
= 4, then there is no
if we set \321\203
is a failure
of surjectivity). Now let

\321\203
(this

failure

is a

\321\203
(this

{0,1}-> {2,3,4}

then 3 is not
\317\207
for

f(x) =

:=
h : {0,1,2} -> {3,4,5}bethe function
\320\233(0)
:=
h
of
the
5.
is
because
each
Then
bijective,
h(2)
from 0, 1, 2.
5 comes from exactly
one element
: N
Example 3.3.22. The
/
:=
is
a
this
n++
fact,
bijection
f(n)
Axioms 2.2, 2.3,2.4). the other hand,
defined

Thus

by

definition g(n) := n++ is not


not
of a bijectivefunction
depends

the function does, but also what


Remark 3.3.23. If a function
sometimes call / a perfect
function),

(not
and

denote

of

its

(and

range

is

\316\271->
\317\207

f(x)

or

matching

bijection.

on what

just

are.

domain)

we

then

bijective,

a one-to-one

with the notion of a one-to-one


<->
action of / using the notation \317\207
f(x)
for instance
the function h in the
Thus

confused

be

to

instead

restating
g : N -> N

the function

same

the

the notion

correspondence

is simply

fact

(in

On

N\\{0} defined by

->

function

3, \320\233(1) := 4,
elements
3, 4,

the

\320\270
\320\260:
/(\320\266).

above exampleis the one-to-one


correspondence

<->

3,

1 <-> 4,

2~5.

Remark 3.3.24.
\342\200\224>
\316\245
is

in
\321\203

to

/ to

= 1

is exactly

there

X,

and also

map

one cannot

instance

/(0) = 2. Thefunctions

one

what it means for

is not

cannot

function

for

elements,

/(0)

f(x)\"

\321\203

\317\207
in

This

rather, this is merely stating what

be a function.

different

which

that

be bijective;

\"for every

iff

bijective

\320\243
such

a function / :

say that

is to

error

common

it

for

means

one element to two


have a function / for
/,

previous example are not bijective,


they
since each input still
one output.
exactly
If / is bijective, then
there
G
every
but

are

g given in the
still functions,

gives

for

such

that

f(x)

and at mostone
/_1(y);

thus

inverse of /.

is

\321\203
(there

because

f\"1 is

\321\203 \320\243,

of

at least

injectivity).

a function

from

is exactly

one

\317\207

one because of surjectivity,


is denoted
This value of \317\207
\316\245
to

X.

We

call /_1

the

63

\320\276 Functions
\302\243

the definition
and
transitive.
reflexive, symmetric,
3.3.1. Show that

Exercise

property:
and

~^
\316\247
'\302\267

if />/

then

\302\243=<?,

\316\245
an(^

:Y

g,g

the

verify

= /

that /

such

functions

\342\200\224>
\316\226
are

3.3.7 is

in Definition
substitution

of equality
Also

= fog.

fog

\342\200\224>
\320\243
\316\226
and
be functions.
Show that
g : \320\243
Exercise 3.3.2. Let / : X \342\200\224>
\320\276
both
then
so
is
show
that if /
are
injective,
g /; similarly,
if / and g
\320\276
then
so
is
both
are
surjective,
g f.
and g
When
is
the
function
3.3.3.
injective?
surjective?
bijecempty
Exercise

tive?

Exercise 3.3.4. In this

: X

composition.

Exercise3.3.5.Let /

J,

g :

\342\200\224>
\320\243
and

all

:X

Let /
its inverse.

3.3.6.

Exercise

X
\320\266
\302\243

invertible,

Verify

and

f{f~1{y))

and

has

/ as

its

all

for
\321\203
inverse

map from

to

all

:= \317\207
for
^\321\205_\321\203(\320\266)
identity

map

is a

of

subset

defined
\320\243,

by

The map

iGl.

\320\241
\320\241
\316\245
\316\226
then

Show

\320\221
is a bijective
that, if / : A \342\200\224>
=
and /_1 \320\276
\320\261\320\273-\320\273.
/

that

if

if X

\342\200\224>
\316\245
\316\226
are
\342\200\224>
\320\243
\316\226
such

: A

\302\260
\316\220\316\263^\316\266
*-x->y

(c)

g
X U

{gof)_1
all

op-1.

inclusion

the

\320\266
\342\202\254
X,

i.e.,

called the

particular

that

= /_1

is in
\316\257\317\207->\317\207

Show

Show

Show that if

functions.

\321\214->
\320\260:
for
\320\266

(b)

(d)

is also

/_1

on X.

(a) Show that if

lb^b

that

= /).

\320\243
: X \342\200\224>
be
^\317\207->\316\263

let
\320\243,

that

let /_1 :
= x for

function, and
f~1{f{x))

(/_1)_1

mapping

is not

Show

laws

\342\200\224>
\320\243
: \320\243
\316\226
and
be
Exercise 3.3.7. Let / : X \342\200\224>
\317\201
are
so
is
and
we
have
then
and
bijective,
f,
g
go
/

Exercise3.3.8.If X

f
if /

true

\320\243.Conclude
\342\202\254
\321\203

(thus

= /. Is
= go f

also
that g must
then g must be surjective.

cancellation

the

if go

Is it true

a bijective

\342\200\224>
\320\243
be

\342\200\224
\316\245
\316\226
be

then /

functions.

\342\200\224>
\316\245
\316\226
be

be injective.
then / must
injective,
g
is
Show that if g \320\276
be injective?
f
surjective,
Is it true that / must also be surjective?
\320\276
f is

\342\200\224>
X be
\320\243

g :

\342\200\224

= gof and g is injective,


functions.
true
if
is
not injective? Show that
statement
same
the
g
=
then
is
surjective,
g g. Is the same statement
and
/
surjective?
if

laws for

some cancellation
\342\200\224 and
\316\226,
g : \316\245

we give

section

\320\243,
/ : X \342\200\224
/
Show that \\igof

Let

\342\200\224>
\320\221
is

and

\320\243
are

any

*<\317\207->\316\266\302\267

then /

function,

disjoint

then

function,
sets,

there

and /

is a

= /

=
\320\276
*,\320\260_>\320\273

then /

:X

\316\277

/_1

\342\200\224>
\316\226
and

/\316\271
:
functions,
unique
=
=
\316\277
\316\277
/\316\271
that
/\316\271
and
\316\260\316\263^\317\207\317\205\316\263
\316\257,\317\207_>\317\207\317\205\316\263
/
\320\264.

function

Set

3.

64

3.4

and inverse images

Images

elements

individual

take

f :X

a function

that

know

We

can also

\317\207
G

X to

in

subsets

take

theory

X to

a set

\342\200\224>Y
from

a set

f(x) G Y.

to elements

\320\243
can

Functions

subsets in Y:

\316\245
is a function
Definition 3.4.1 (Imagesof sets).If / : X \342\200\224*
from X to \320\243,and S is a set in X, we define /(5) to be the set

:=

f(S)

subset of

is a

set

this

under the

\342\202\254
S};

the

call f(S)

sometimes

the image of

called

is sometimes

and
\320\243,

We

/.

map

: *

{/(\316\266)

distinguish it from the conceptof the inverse


S, which is defined below.

S to
of

Note

set

the

that

One

(Axiom

3.5)

(Axiom

forward

but we

of replacement,

instead

3.4.2. If / : N -> N is the


image of {1,2,3} is {2,4,6}:

Example

also define

reader.

the

challenge

can

/_1(5)

thanks to the axiom


f(S) using the

is well-defined

f(S)

of replacement
3.6).
axiomof specification
to
leave this as a

image

5
of

image

forward

map

f(x)

then
2\320\266,

the

/({1,2,3})= {2,4,6}.
More

and

to compute
/(5), we take
to
element
each
individually,
/

informally,

apply

resulting objects together to form

new

above example, the imagehad


set.
But sometimes
can
the
original
image
is not one-to-one
3.3.14):
(seeDefinition
3.4.3.

Example

be

the

define

map

(Informal)

rigourously

f(x)

= x2,

Let

in the

\316\226
be

then

put

the

same

size

the

be smaller,

set

next section)

that

/ is

of integers

then

not one-to-one

5,

all the

because /(\342\200\224!)=

as the
because /

and let /

/({-1,0,1,2}) = {0,1,4}.
Note

\317\207
of

and

set.

In the

we will

element

every

/(!).

(which
\342\200\224*
\316\226
\316\226

inverse

and

\316\271
\316\277
Images

65

images

Note that
\317\207
e

but

in

&

\342\202\254
f(S)

\317\207
\342\202\254
S]

above informal example, /(\342\200\2242)lies in


but -2 is not in {-1,0,1,2}.
The correct

in the

instance

/({-1,0,1,2}),
statement

f(S)

general

f(x)
for

f(x) e

=*

the

set

is
e
\321\203

4=>

f(S)

\321\203

for some

f(x)

\317\207
G

(why?).

3.4.4 (Inverse images).


f\"l{U) to be the set

Definition
the

set

:=

{x

\320\223\\\320\270)

If U is a subsetofF, we

X :
\342\202\254

/(a)

G U}.

words, f~l(U) consistsof all the elements


map into U:
In other

f(x) e U

We

call

Example 3.4.5. If /

2\320\266,then

\317\206=* \317\207
e \320\223\320\263(\320\270).

\342\200\224>
N

is

the

map

f(x)

{2,4,6},
but/^({l,
2,3}) = {1}. Thus the forward
and
the
backwards
imageof {1,2,3}are quite
{1,2,3}
=

/({1,2,3})

of

image

which

of

image of U.

the inverse

f\"l{U)

define

sets.

different

Also note

that
{1,2,3}

\320\257\320\2231\302\273!,
2,3}))^

(why?).

Example

3.4.6. (Informal) If /

the

\342\200\224>
\316\226
\316\226
is

map

f(x)

\320\2662,

then

/^({0,1,4})

Note that /

doesnot

to makesense.Also
quite

invert

each

other,
/^(/({-\316\271,

(why?).

have

note

to

that

{-2,-1,0,1,2}.
be

invertible

images

for instance

we have

0,1,2}))

74-1,0,1,2}

for /^(U)
images do not

in order

and inverse

3. Set theory

66
If /

3.4.7.

Remark

f\"1 in two

both

is a bijectivefunction,

slightly different
are

definitions

earlier, functions are not sets. However,

As remarked
should

to be a
functions
to consider
be able

should

be

consider

of

type

set Y. To dothis we
theory:
to a

need

consists

which

is a

function

3.4.8. Let X =
Yx consistsof four functions:
Example

we

a set

from

axiom to set

Then

there

functions

from

sets.

all the

\316\271-\302\273
the

0;

if

that

that

show

One

Lemma

function

\316\245
has

Yx

=
\316\245

and

{4,7}

that

i->

{0,1}.

that

function

the

domain

with

maps 4
1 and
that maps 4 \316\271\342\200\224>
7 i-> 0; and
and
1. The
we use the
7 \316\271\342\200\224>
reason
is

of

\316\245
be

we

thus
\316\245',

feYx*=>(f

and

Let X

(Power

do

particular,

another

introduce

to

set
Axiom 3.10
axiom).
exists a set, denotedYx,

X to

object,

we

in particular

and

sets of functions. In
to consider the set of all functions

able

defined

an issue because

3.4.1).

(Exercise

equivalent

is not

this

but

ways,

have

we

then

and

and

Y).

range

the set
\316\271\342\200\224\342\226\27
4
0 and
maps
Then

\316\271\342\200\224>
the

1;

function

the function that maps4 \316\271\342\200\224>


1
Yx
notation
to denotethis set
\316\267
and
X has m elements, then one can
elements
has nm elements; see Proposition3.6.14(f).
of this

consequence

3.4.9. Let X

be

axiom
set

is

Then

the set

[Y :Y is a subsetof X]

is

Proof

set

See Exercise

Remark 3.4.10. The set {\316\245:

the powersetof

distinct

we have

objects,

3.4.6.

and

is denoted

\316\245
is

a subset

2X. For

2KM = {0,{a},{b},{c},
{a,b},

{a,

c},

of X}

instance,

{b, c},

is known as
if

a,

b,

{a, b, c}}.

\321\201
are

inverse

and

\316\271
\316\277
Images

Note that while {a, b,


This

elements.

as

2X\\

set

our

allow

to

Axiom3.11 (Union).

Let

those

precisely

thus

of X

further axiom to

add one

whose elementsare
exists a set \\JA whose elementsare
are elements of the elements of A,
be

a set,

all of

\317\207

A 4=>

If A

3.4.11.

Example

which

objects

all objects

for

there

Then

sets.

themselves

set

power

the axiomof pairwiseunion


sets.
collectionsof
larger

of much

unions

we enhance

in which

theory,

now

us

let

completeness,

23

has

2^\321\212^

as to why we refer to the


this issue in Chapter 8.

to

For sakeof

3 elements,

has

c}

a hint

gives
return

we

67

images

(x e S for some

= {{2,3},

A)\302\267

{3,4}, {4,5}}, then

\\JA

(why?).

{2,3,4,5}

The axiom of

union, combinedwith

of pair set,
Another
3.4.8).
has some
one
set

axiom

the

(Exercise
implies the axiom of pairwise union
of
this
axiomis
that if
consequence
important
I
a
for
element
we
some
set Aa,
E
have
every
/, and
form the union set [jaej Aa by defining

(J

Aa :=

then

can

we

G /},
\\J{Aa : \316\261

aei

which is

a set thanks to the axiom

of union. Thus for


{3,4},
generally,

instance,

and A3 := {4,5},
we see that for
G
\320\243

(J

Aa

if

of

and

replacement

/ = {1,2,3},

A\\

then LU{i,2,3}4>=
any object y,

4=>

\320\265
Aa
(\321\203

for

some

:=

the axiom

{2,3},

{2,3,4,5}.

a e

I).

(3.2)

to J as an index set, and the


refer
as labels;the sets Aa are then called
the
a G A. Note that
labels
by
family of sets, and are indexed

In situationslikethis,
elements

if

:=

More

aei

A<i

J was

(why?).

of this

empty,

we

often

index set

then

Aa
\\Ja\342\202\254l

would

automatically

also be

empty

3. Set

68

theory

of familiesof sets, as long


More
given
any
specifically,
non-empty.
a
to
\316\261
and
of
each
an
set
set
\342\202\254
Aa
J, we
J,
nonempty
given
assignment
first
some
can define the intersection \320\237\320\260\320\265
element
^a
choosing
by
J
and
we can do sinceJ is non-empty),
of I (which
\316\262
setting
can

We

form intersections

similarly

set is

index

the

as

Aa :=

p|

{xe

for
all
\316\221\316\261
\316\221\316\262:\317\207\342\202\254

a G /},

(3.3)

aei

which is a set by

on the

that

3.4.9). Observe

G
\320\243

for

choice of /?, but

\320\244=> (\321\203
e

Remark 3.4.12. The axioms


3.1-3.11,

(Axioms
as

known

does

not

may

(Exercise

for all

Aa

\320\260
\342\202\254
J)

(3.4)

with (3.2)).

(compare

introduced

it

y,v

object

any

P| Aa
aei

This definition

of specification.

axiom

the

it depends

like

look

theZermelo-Fraenkel

set

of

Zermelo(1871-1953)
is one further axiom
will
eventually
rise
of choice (see Section8.4),
Choice (ZFC) axioms of set
and

need,
the

we

to

giving

we

but

theory,

axiom for

that

theory

we have

the dangerous Axiom 3.8) are


excluding
axioms
of set theory*, after Ernest
Praenkel
There
Abraham
(1891-1965).
the famous

axiom

Zermelo-Fraenkelwill not need this

time.

some

3.4.1. Let / : X \342\200\224>


\320\243
be a bijective
function, and let /_1 :
forward
its inverse. Let V be any subset of Y. Prove that
the
V
V
of
under
sameset
inverse
of
under
is
as
the
the
/_1
image
image
will
lead
thus
the
fact
are
to
that
denoted
not
both
sets
/;
by /_1(V)
Exercise

\342\200\224>
\316\245
X be

any inconsistency.

Exercise 3.4.2. Let /


set

let
\320\243,

general,

be a

can one

andi/?
3These

axioms

formulations

can

\342\200\224>
\316\245
be

subset of X,

say about /-1


are formulated

be shown

to be

a function

and

let

U be

(/(\302\243)) and

slightly
equivalent

from one
a subset

5?

in other

differently

to

What

each

other.

set X to another
of \320\243.What,
about

texts,

in

/(/-1(t/))
but all the

Let

3.4.3.

Exercise

f(A\\B)>

f(A

the

\320\241
relation

that

true

Exercise3.4.4.Let /
let

and
\320\243,

set

DB)C f(A)

HV) = f~l(U)

f~l(U

\320\241

f(A)\\f(B)

is it

to =?

from one set X


Show that f-1(U\\JV)=

of Y.

\342\200\224>
\316\245

two statements,

first

a function

\342\200\224>
\320\243
be

/ :

let

and

that

\316\240
f(B),

the

f(B)\302\267 For

subsets

set X,

of a

subsets

be improved

can

\320\243
be

t/,

that

ri(\\>),

two

\320\222
be
\320\220,

Show that f(A


u B) = f(A) u

a function.

be

69

and inverse images

'. Images

\316\240
and
that
\320\223\320\263(\320\243),

to another
U
/~\320\263(\320\270)

/^OW

\316\223\316\271(\317\205)\\\316\2234\316\275).

Let /

3.4.5.

Exercise

that

\320\243.Show

set

:X

= 5

/(/\"H5))

surjective.Show that

/-1

(/(\302\243))

everv
&>\320\263

= S

set X to another

from one

a function

\342\200\224>
\320\243
be

for every

\320\241
\320\243
if

\320\241
X

and

if /

is

only if /

is

only

ii and

injective.

Prove Lemma 3.4.9. (Hint: start with


set {0,1}X
the
the replacement axiom, replacingeach function
/ with the
Seealso
Exercise
3
.5.11.
/\"1({1})0
object
\316\245
X to
be sets.
Define a partial function from
Let \316\247,
3.4.7.
Exercise
Y' whose
domain X1 is a subset of X,
to be any function
\320\243
/ : X1 \342\200\224>
3.4.6.

Exercise

and apply

whose

and

the

set

power

to

of

the collection of all


that
(Hint: use Exercise3.4.6,

\320\243.Show

itself

\320\243
is

a set.

axiom, and the

the replacement

axiom,

3.4.8.

Exercise

a subset

is

range.\320\243

partial functions from

Show that Axiom 3.4 can

be deduced

union

axiom.)

3.3

Axiom

from

and Axiom 3.11.

Exercise3.4.9.Show
to

each

: \317\207
\302\243
\302\243
Aa
{\317\207
\316\221\316\262

and so
Also

that

why

all a

for

the definition

explain

if

of

AQ,

\302\243
/}

two

are
\316\262'

and
\316\262

assign a set

aG/we

Aa

be

a set.

of a

: \317\207
for
\302\243
\342\202\254
Aa
{\317\207
\316\221\316\262>

set /,

and

defined

Aa
\320\237\320\260\320\265/

in (3.3)

all a

\302\243
/},

does not depend on \316\262.

(3.4) is true.

Exercise3.4.10.Suppose
let

elements

then

/ and J are two


that
Show that (\\Jaei A<*) u (UaeJ

are non-empty, show that (f|e\342\202\254J


Aa)
Exercise
3.4.11.
Let X be a set, let Jbea
a \342\202\254
/ let Aa be a subset of X.
Show that
and J

X\\\\]Aa=[\\{X\\Aa)
and

X\\f]Aa=\\J(X\\Aa).

ael

ael

for all
=
A<*)
Uae/uJ
=
\316\240
A^
sets,

and

a G /U

If
A<*\302\267

J
J A\"\302\267
(\320\237\320\260\320\265
\320\237\320\260\320\265/\320\270

non-empty

set,

and

for all

3. Set

70
be compared

should

This

(although one cannot


as /

laws,

gan's

could be

Cartesian

3.5

de

with

the

derive

laws

Morgan's

identities

above

differencing,

another

fundamental

and

intersection,

operation on sets is that of the

product

3.5.1 (Ordered

Definition

we

equal),

(possibly

of

consisting

if both

only

the

define
its

as
\317\207

ordered

Two

component.

first

and
are
any objects
\321\203
pair). If \320\266
ordered pair (x,y) to be a new

and

component

pairs (x,y) and (x'\\y')


their components match,
=

(x9y)

This

3.1.28
directly from de \320\234\320\276\320\26

products

to the basicoperationsof union,

and

in Proposition

infinite).

In addition
Cartesian

theory

(x',y')

(x =

4=>

as
\321\203

its

object,

second

are consideredequalif

i.e.

=
x1 and \321\203
\321\203').

(3.5)

axioms of equality (Exercise3.5.3).Thus


for
but
to
the
is
+
is
+
equal
pair (2 1,3 2),
pair (3,5)
is
in
from the pairs (5,3), (3,3),and (2,5).
contrast
(This
the usual

obeys

the

instance,

distinct

to sets, where {3,5}and

{5,3}

are

equal.)

an
is partly
Strictly speaking, this definition
that given any two objects
because
we have simply postulated
axiom,
and
that
an
of
form
\317\207
the
exists.
However, it is
object
y,
(x,y)
axioms
of set theory
to
an
ordered
the
define
pair
possible
using
in such a way that we do not need any further postulates (see

3.5.2.

Remark

Exercise3.5.1).

Remark3.5.3.

We

symbols

once

()

ordered

pairs.

determine

now

\"overloaded\"

the parenthesis

used
to denote
they now are not only
grouping
and argumentsof functions,
but
also
to enclose
This
is usually not a problem in practice as onecan
what
usage the symbols () wereintendedfor from

again;

of operators
still

have

context.

Definition

we define

3.5.4 (Cartesian product). If X


\316\245
to
the Cartesian product \316\247\317\207

and

be

\316\245
are

the

sets,

collection

then

of

71

3.5. Cartesian products


first

whose

pairs,

ordered

in

lies

and second

lies in X

component

thus
\320\243,

component

= {(x,y):xeX,y\302\243Y}

XxY

or equivalently
x

G (X
\316\261

\320\243)<=>

(a

some

for

(\320\266,
\321\203)

\320\266
X
and
\342\202\254

\342\202\254
\321\203
\320\243).

of
simply assume that our notion
is such that whenever X and \320\243
the Cartesian
are
sets,
ordered
pair
x
\320\243
is also
a set. This is however not a problemin
X
product

We

3.5.5.

Remark

3.5.1.

Exercise

see

practice;

shall

3.5.6. If X :=

Example

{1,2}

:=
\320\243

and

=
\320\245\321\205\320\243

(1,5), (2,3),

{(1,3),

(1,4),

{(3,1),

(4,1), (5,1),

then

{3,4,5},

(2,4), (2,5)}

and

\317\207
\316\245
X

speaking,

strictly

Thus,

are

although

they

same

number

Let / :
Cartesian

very

\317\207
\320\243
\316\247
and

\342\200\224>
\317\207
\316\226
\316\247
\320\243
be

product

of

they

whose

a function

input
are

technically

domain

one

and

variable,

\320\243.
Then

mapping

defined
One
ordered

simultaneously as a function

for
now

can

by

triples,

or
\316\226,

another

and

\317\207
X
\342\202\254

one

as

notions

two

the

variable

two,

with

of two variableswith domainsX


instance
the addition operation + on the natural
\342\200\224\342\226\272
\317\207
\316\235
as a function + : \316\235
be re-interpreted
\316\235,
as

a function

\316\271->
\317\207
+ \321\203.
(\320\266,
\321\203)

can

in

/(\320\266,\321\203)

of

\317\207
\316\247
\320\243
is

output /(x, y) in Z. While the


different, we will not bother to distinguish

\316\247
\317\207
\320\243
and

\320\243.
Thus

and
numbers

the

/ can either be
the single input

a single

of /

think

and

always

domain

to an output
of an orderedpair(x,y) in \316\247\317\207
\320\243
a function
of two variables, mappingan input
\320\243
to
\342\202\254
\321\203

sets,
have

(Exercise 3.6.5).

of two other sets X

thought of as a function

different

are

\321\205
\320\243
X

For instance,

similar.

of elements

(3,2), (4,2), (5,2)}.

of course
ordered

generalize
quadruples,

the concept of ordered pairs to


etc:

3. Set

72
3.5.7

Definition
product).

Let

(also

denoted

every

natural

number

is an

ordered

ordered

Two

of

Xi
product \320\237\320\272\320\263<\320\277

denoted
\342\204\226so

:=
\316\247\316\271

(xi)i<i<n

\316\267-tuples

one

the

for
i\342\204\242

(xi)i<i<n

all 1 < i < n. If


define their Cartesian

yi for

\317\207\316\271

we

sets,
\320\251=1

^orli

\302\267
Xi \302\243
Xi

{(xi)i<i<n

Cartesian

\316\267-tuple

collection of objects x^
1 and n; we refer to Xi as

\316\267-tuple.

\316\267-tuple

\316\267-fold

An ordered

to be equal iff

are said

{yi)i<i<n

(Xi)i<i<n

is a

\320\263
between

ordered

and

\316\267-tuple

number.

(rri,...,a:n))
of the

component

and

(Ordered
a natural

\316\267
be

theory

x...xXn)by
all 1

for

<

<
\320\263

\316\267}.

1<\320\263<\320\277

this

Again,
tuple

the

of

axioms

3.5.8. One can show

Remark
from

functions

the

i->

-^, and

Ui<i<n

tion to

all 1 <

postulates

simply

3.5.2).

(Exercise

Indeed,

that an ordered nexist


when
needed,but using
product always
set theory one can explicitly
construct
these
objects
definition

a Cartesian

and

that

power

set

axiom

from

the

domain

Xi

then we can restrictusing

restrict to thosefunctions

i < n.

One

can

generalize

Cartesian products, seeDefinition


Let

3.5.9.

Example

\317\207
\316\247\316\271
X2

x X3

this

and

a set.

consider the set of all


<
\320\263

to

\316\267}

the

axiom

for

which

the
range
of specificaXi G Xi for

to infinite

construction

8.4.1.

be
\320\2543
\316\261\317\212,
bi, \320\260\320\263,
\320\2603,
\320\254\320\263,

'-= {^2)^2},
\316\2472
{\316\261\316\271,\316\264\316\271},

*s indeed

Y[i<i<nXi

we can
{1 <

X3 :=

objects,

and

Then
we
{\320\260\320\267,&\320\267}\302\267

let

X\\

:=

have

={(ai,a2,a3),(ai,a2,b3),(ai,b2,a3),(ai,b2,b3),

(bi,

\320\260\320\263,
\320\260\320\263,
(bi,
63)? (bi,
\320\260\320\267),

\320\254\320\263,
\320\254\320\263,
\320\260\320\267),
(bi,
\320\2543)}

=
\317\207
(\316\247\316\271
\320\2452)\321\205\320\245\320\263
\320\254\320\267),
\316\2612),
\316\2613),((\316\261\316\271,
\320\2602),
((\316\261\317\212,
\320\253,\320\260\320\267),
\320\2542),
{((\316\261\317\212,
((ai,
\320\254\320\267),
\320\2602),
((\320\254\321\214
\320\2602),
\320\2543),
\320\2542),
\320\2603),
\320\2603),((\320\254\321\214
((\320\254\321\214
((\320\254\321\214
\320\2542),
\320\2543)}

=
\317\207
\321\205
\320\245\321\205
(\320\2452
\320\2453)
(\320\2602,
\320\2603)),(\316\261\317\212,
\320\2543)),(\316\261\317\212,
{(\316\261\317\212,
(\320\2602,
(\320\2542,
\320\2603)),(\320\260\321\214
(\320\2542,
\320\2543)),
(\320\254\321\214
(\320\2602,
\320\2543)),(\320\254\321\214
(\320\2542,
\320\2603)),(\320\254\321\214
(\320\2542,
\320\2603)),(\320\254\321\214
(\320\2602,
\320\2543))}.

73

products

Cartesian

3.5*

Thus, strictly

speaking, the setsX\\

to each other

instance,

(for

and
\321\205\320\245\320\267,
\321\205^\320\267,
{X\\ \320\245\320\2452)

XX2

clearly very related


bijections between

However,
they are
there
are obvious

distinct.

are
x
\316\247
(\316\2472 -^\320\267)
\316\247\316\271

to neglect
sets), and it is commonin practice
sets
distinctions
between
these
and
that
minor
pretend
they
the
\342\200\224*
:
\317\207
\317\207
\316\245
Thus
a
can
be
function
\316\247<\316\271
Xz
X\\
/
are in fact equal.
G X\\ XX2 \320\245-^\320\267?
of as a function of onevariable(sci,X2, \321\205\320\263)
thought
of
three
variables
function
G
a
X2
x\\
X2, \302\2433G I3,
Gli,
or as
and so
G X3,
or as a function of two variables x\\ G X\\, (\321\2052,\321\205\320\263)
will
these
not
bother
to
between
different
we
distinguish
forth;
of the

two

any

three

perspectives.

3.5.10. An ordered \316\267-tuple

Remark

calledan ordered
short. In Chapter

of

sequence
5

of

an

3.5.11. If

shall

or

with

identify

an

object,

\317\207
itself

(even

\320\266
is

then
though

speaking,not the sameobject).


is just
Cartesianproduct
Cartesian product
gives,
seF
whose
rather the
{()}
known

If

objects is also
sequence for

useful concept

(x) is a 1-tuple, which


the two are, strictly

set, then the


the empty
Also,
(why?).
\316\240\316\271<\316\257<\316\271^
set
the
not
empty
{}, but
\320\237\320\272\320\263<\320\276^
singleton
only element is the 0-tuple(),
as the empty tuple.
if

Then

also

the very

also introduce

shall

a finite

sequence.

infinite

Example
we

we

xn of

sci,...,

\316\267
elements,

\316\267
is

natural

number,

X\\

is any

X\\

we often

write Xn as

shorthand

X1 is
between
ignore
(if
an object \317\207
and
the 1-tuple
while
X2 is the Cartesian product
(\320\266)),
\317\207
\316\247
X.
set X\302\260is a singleton
The
set {()} (why?).
We
can
now generalize
the single choice lemma (Lemma3.1.6)
to allow for multiple (but finite) number of choices.
for

the

\316\267-fold

Cartesian

product

essentiallythe samesetas

Lemma 3.5.12
and for each

(Finite

there exists

set.

Then

all 1

< i <

set

n. In
is

other

also

an

\316\267-tuple

words,

non-empty.

\316\240\316\271<\316\271<\316\267\316\247\316\257

the distinction

Let
1 <

if

Thus

[]1<KnI.

we

choice).
number

natural

X\320\271:=

>
\316\267

i <

1 be

such

(xi)i<i<n

each

a natural

n, let Xi
Xi is

be

that

number,

non-empty
Xi

G X%

for

non-empty, then the

on

induct

We

Proof.

interesting

that

in

(why?).

already

been

proven

Let Xi,...,

be

Xn++

all 1

< i<

may

find

that a

setting

yi :=

by

(yi)i<i<n++

when i

Remark 3.5.13.Itis
done

cannot

be

axiom

of choice.

allow

for

an

automatically;
See Section

Exercise3.5.1.Suppose

it for

prove

n++.

sets. By induction
that

G Xi
\317\207\316\271

for

by Lemma 3.1.6we
If we

thus

when
\317\207\316\271

1 <

for all

G Xi

define the
< \316\267
\320\263
and

1<

<
\320\263

n++}

0
lemma

of choices,

number

infinite

this

that

plausible

intuitively

be extended to

yi

1; the

that the claimhas

such

Xn++-

= n-H- it is clearthat
closing the induction.

:= a

ls non-empty,

Xn++

=
\316\267

is not particularly
the claim follows from

{xi)i<i<n

\316\267-tuple

a such

an object

n++-tuple

an

since

\320\277.
Also,

case

theory

0 but

Now suppose inductively


now
for some n; we will
of non-empty
a collection

find

can

we

hypothesis,

=
\316\267

When

case).

3.1.6

=
\316\267

with

true

Lemma

thus

(starting

base

the

with

\316\267

also vacuously

claim is

yi

Set

3.

74

should

but

this

an additional axiom,the

it requires

8.4.

the ordered pair (x,y) for any objects


:=
\321\203
by
(x,y)
{{#}, {x,y}} (thus using several
applications
of Axiom
Thus
for
instance (1,2) is the set {{1},{1,2}},
3.3).
is
the
set
and
(1,1) is the set {{1}}.Show that such
(2,1)
{{2}, {2,1}},
X and \316\245
a definition indeed obeys the property (3.5),and also whenever
are sets,
the Cartesian product XxFis
also a set.Thus this definition
can be validly used as a definition
of an ordered
pair. For an additional
the
show
that
alternate
definition
challenge,
(x,y) := {x, {x,y}} also
also
verifies
and
is
thus
an
definition
of ordered
acceptable
pair.
(3.5)
one
axiom
of
and
in particular
this
latter
task
needsthe
regularity,
(For
Exercise 3.2.2.)
the

\317\207
and

we

Exercise3.5.2.Suppose
function

: {i
\317\207

X (so different

we then write

define

formula

GN

ordered
Xi

we

:1 <i
for

define

\316\267-tuples

x(i),

an ordered

<n} -^ X
and

are

whose

\316\267-tuple

range

to

allowed

also write

\317\207
as

to

be a

surjective

arbitrary set
have different ranges);
is some

{xi)i<i<n\302\267

Using

this

if and only if Xi =
verify that we have {xi)i<i<n = {\320\243\320\263)\321\205<\320\263<\320\277
<
<
\320\263\320\277.
for all 1
axe an orderedn-tuple
show that if (Xi)i<i<n
Also,
\320\243\320\263
then the Cartesian product, as defined
of sets,
in Definition
3.5.7, is

definition,

indeed a

set. (Hint: use Exercise3.4.7and

the

axiom

of specification.)

75

products

Cartesian

5#

Exercise 3.5.3. Show that


obey

\317\200-tuple

ordered

Let

3.5.4.

Exercise
(A x C),

/\320\264
\317\207\316\262)\\(AC).

rolesof

symmetry,

sets.

Show that

be
\320\241
\320\222,
\320\220,

=
\316\221
\317\207
\316\240
\320\241)
(\316\222

that
left

the

reflexivity,

and

axioms.

transitivity

Ax (B\\JC)
that

\317\207
\316\240
\317\207
and
[\316\221
\316\222)
(\316\221
\320\241),

similar
identities
(One can of course prove
and right factors of the Cartesianproduct

Exercise 3.5.5. Let A, B, C, D be sets. Show that


Is it true that (AxB)U(CxD)
(AnC)x(BCiD).
=
x
that
true
B)\\(C \317\207
D)
(A
(A\\C) \317\207
(B\\D)?
Isit

pair and

for ordered

of equality

definitions

the

the

(Ax

= (Ax B)U

A x (B\\C) =
in which the

are

reversed.)

\317\207
\320\222)\320\237(\320\241
D)

= (AUC)x(BUD)?

\320\241
Let \316\221,\316\222,\316\237,\316\214
be non-empty
sets. Show that Ax \320\222
\320\241
\320\241
\320\241
\320\222
and that Ax\302\243 = CxL>if
and
D,
only if \320\220
= D. What happens if the
= \320\241
\320\222
and
hypothesesthat
and only if A
are removed?
the A, 5, C, D are all non-empty

3.5.6.

Exercise

D
\317\207
\320\241

if and

Let X,F be sets,and

Exercise 3.5.7.
and

:
\317\200\317\207\317\207\316\275-y

->
\320\243

:= \320\243'>
these
maps
7\320\223\321\205\321\205\320\243-+\321\203(\320\266>2/)

that

Show

onXxY-

for

the

\320\243
be

:1\321\205\320\243->1
\317\200\317\207\317\207\316\263->\317\207

/ :

and

a:
7\320\263\320\245\321\205\320\243^\321\205(\321\2051\321\203)
:=

as the co-ordinate functions

are known

functions

any

let

maps

\342\200\224>
\316\226
X and

there

\342\200\224>
: \316\226
F,
\317\201

such that \317\200\317\207\317\207\316\263^\317\207


oh = f and
exists a unique function /i:Z->XxF
=
oh
and
3.3.8,
g. (Compare this to the last part of Exercise
xy_>y
7\316\223\317\207
h is known
as the direct sum of / and
to Exercise 3.1.7.) This function
= f \316\230
\320\264.
g and is denoted h
Let

3.5.8.

Exercise

Exercise

3.5.9.

Aa be a

set,

^d

for

if at

only

that /

Suppose
and

be sets.

Xn
\316\221\316\212,...,

*s emPty if

\316\240\316\223=\316\271

all

\302\243
J
\316\262

least

Show that the Cartesian product


one of the Xi is empty.

and J are two


let

be a
\316\222\316\262

and for all a \342\202\254


I let
Show that (\\JaeI Aa) \316\240

sets,

set.

=
(U/\302\273ejS/O

Exercise
be

the

U(aJJ)6Jxj(4\302\273n*0)\302\267

If /

3.5.10.
subset

of

:X

\342\200\224>
\320\243
is

\316\245
\317\207
\316\247
defined

a function,

\320\252\321\203
{(\317\207,
f(x))

define the prap/i


: \317\207
Show
\342\202\254
X}.

of / to

that

two

if and only if they have the


f
equal
same
if
is
X xY with
G
subset
of
the
Conversely,
graph.
any
property
\316\245
:
that for each \317\207
the
set
has
one
\302\243
\302\243
\302\243
X,
exactly
{y
(x,y)
G}
element (or in other words, G obeys
the
vertical
line test), show that
there is exactly one function
\316\245
whose
is equal to G.
/ : X \342\200\224*
graph
functions

: X

\342\200\224>Y,f:X\342\200\224>Y are

Exercise 3.5.11. Show

that

Axiom

3.10

can

in

fact

be

deduced

from

Lemma 3.4.9 and the other axioms of set theory, and thus Lemma 3.4.9
can
be used
as an alternate formulation of the power set axiom. (Hint:
for any
two sets X and F, useLemma
3.4.9
and the axiom of specification
to construct
\317\207
\316\245
the set of all subsets of \316\247
which
obey the vertical line
test. Then use Exercise
3.5.10and the axiom of replacement.)

3. Set

76
This exercise

3.5.12.

Exercise

sition 2.1.16.Let

a rigourous
version of Propo.
a function, and let \321\201
be a natural
\342\200\224>
:
N
N
\316\261
such
function
that

establish

will

be

N
/:NxN\342\200\224\342\226\272

Show that there

number.

theory

existsa

a(0) =

\321\201

and
=

all

for
\320\260(\320\277++) /(\317\200,\316\261(\317\200))

this function is unique. (Hint: first show


of the proof of Lemma
that for every
3.5.12,
'
N \342\202\254
exists a unique function \316\261
there
\302\243
\316\275 {\316\267
\316\235:
N,
that

furthermore

and

a modification

by

inductively,

number

natural

that

such

without

result

\316\267
< \316\235.)

an

For

for all
\320\260^(\320\277-\320\235-)
/(\317\200,\316\261(\317\200))

additional

of the

any properties

using

= \321\201
and
\316\261^\316\275(\316\237)

that

< \316\235}\342\200\224>
\316\235
\316\267
such
N
\316\267
\302\243

\316\267
\302\243
\316\235,

prove this

challenge,

other

numbers

natural

than

the

the ordering
Peano axioms directly (in particular, without
of the
using
and without appealing to Proposition2.1.16).(Hint:
natural
numbers,
show inductively,
first
using only the Peano axioms and basicset theory,
exists a unique pair An, Bn
that
for every natural number N \302\243
N, there
of subsets
of N which obeys the following
Bn = 0,
properties:
(a) An \316\240
=
\316\267
\302\243
\302\243
N-HBNl
(e) Whenever
(b) \316\221\316\235\317\205^
\316\235,(\317\213)0\316\266^, (d)
\316\222\316\235}

n-H-

have

we

n-H-

\302\243
Bn\302\267

< \316\235}in
: \316\267
\316\235
\302\243

{n

Whenever

purposeof

this

use

\316\267
\316\235,we
\317\206

number

have

a substitute

as

An

is to

exercise

one version of the natural


discussion
in Remark 2.1.12).
only

the

and

\316\267
\302\243
An

these sets,
previous
argument.)

obtains

the

Exercise 3.5.13. The


essentially
(cf.

(f)

one

Once

\302\243
An\302\267

for

show that there

is

in set theory
have
a set N'
an
0', and

of

system

Supposewe

an \"alternative zerp\"
numbers\",
\"alternative increment
which takes any alternative
natural
number
operation\"
N'
n' \342\202\254
and returns
another alternative natural number
\302\243
n'-H-'
N;,
such that the Peano axioms (Axioms
all hold with the natural
2.1-2.5)
alternative
numbers,
zero, and increment replaced by their
counterparts.
N'
Show that there exists a bijection
from
the natural
/ : N \342\200\224>
natural

\"alternative

the

numbers

to

that

for

/(n-hh)

3.6
In
cally,

the

natural

alternative

any

\316\267
N
\302\243

= n'-H-'.

and

n'

(Hint: use

numbers such that


\302\243
N',

we

have

/(\317\200)

0', and such


/(0)
= \316\267!
if and
only if

Exercise 3.5.12.)

Cardinality of sets
previous
assuming

we defined the
they were equipped

chapter
that

natural numbersaxiomatiwith a 0 andan increment

\320\276

gt

77

of sets

Cardinality

operation, and assumingfive


Philosophically)

of

conceptualizations

approach

natural

treats

things

many

kow
Second,

numbers.

more

numbers

like ordinals

than cardinals.

to
First,

count
One, Two, Three, ..., and are used
there are in a set. The ordinalsare
and are used to order a sequenceof objects.There

are

cardinals

(The

is

natural

elements

many

on these

axioms

different from one of our main


- that of cardinality, or measuringhow
numbers
there
are in a set. Indeed, the Peanoaxiom

is quite

this

...,

Third,

difference

a subtle

cardinals

infinite
comparing

between the
infinite

with

two, especiallywhen
ordinals, but this is beyond

the

number
paid a lot of attentionto what
scope
number \316\267 which is an operation which
came
for
but
natural
less so for cardinals - but did not
ordinals,
is quite
the issue of whether these numbers couldbeusedto count
address
of this section is to addressthisissueby noting
The
sets.
purpose
natural
numbers
can be used to count
the
of
the
that
cardinality
sets, as long as the set is finite.
first
is to work out when two
sets
the same
have
The
thing
and
have the
size: it seems clear that the sets {1,2,3}
{4,5,6}
a different
size from {8,9}.
One
same size, but that both have
that
the
same
size
define
this
is
to
two
sets
if
have
to
say
they
way
the
number
of elements, but we have not yet defined
same
have
the \"number of elements\" in a set is. Besides,
thisruns into
what
when a set is infinite.
problems

We
of this text).
next after a given

The

right

same size\"

the concept of
to define
way
is not immediately obvious,but

\"two

can

sets

having

the

be

worked

out

reason why the sets {1,2,3} and


is
that
the
of
one can match
elements
{4,5,6}
the first set with the elementsin the second
set in a one-to-one
this is how we
6.
correspondence: 1 <-\342\226\272
4, 2 <-+ 5, 3 <-\342\226\272
(Indeed,
first
to
learn
to count
a set: we correspond the set we are trying
count with another set, such as a set of fingers
on
your
hand).
We will use this intuitive understanding as our rigourousbasisfor
with

some

thought.

have

\"having

the

Definition

and \316\245
have

the

same

One

intuitive

same size

size\".

3.6.1 (Equal cardinality). We


equal

cardinality

iff

there

exists

say

that

a bijection

two

sets

/ :X

\342\200\224*
\316\245

Set

3.

78

to Y.

from X

Example 3.6.2. The sets{0,1,2}

and

not yet

we do

whether

know

have

{3,4,5}

equal

bijection between the two

find a

can

we

since

cardinality,

that

theory

and

{0,1,2}

cardinality; we know that one of


we
{3,4} is not a bijection,but
other
be
still
some
bijection
might

sets.

the functions

Note

have equal

{3,4}
from

to

{0,1,2}

proven yet that there


one set to the other. (It

not

have

from

equal cardinality, but we will


Note that this definition
makes
this
sense
prove
is
finite
we
of
whether
X
or
infinite
haven't
fact,
regardless
(in
even defined what finite meansyet).
out

turns

do not have

that

they
a little later).

Remark 3.6.3.
fact
doesnot precludeoneof
instance, if X is the setof

from

sets

the

natural numbers, then the map /


is a bijection from X to \316\245(why?),
despite
cardinality,
as if it should
only

have

:X

and

the

set of

by

f(n)

and

For
even

:= 2n

\316\245
have

equal

and seeming intuitively

the

elements

The notion of having equal cardinality

is

of

\"half\"

\316\245
is

so X

of X

cardinality.

the other.

\342\200\224>
\316\245
defined

and

a subset

have equal

containing

numbers

natural

\316\245
being

sets

two

that

The

of X.
an

equivalence

relation:

cardinality with

If X

X.

be
\316\247,\316\245,\316\226

has equal

with X.IfX
with Z,
cardinality

cardinality

equal
has

equal

Let

\316\267
be

X has
=

\316\267}

=
\316\267

{1,2,...,

notion

set

to
\316\267}
{\320\263

of equal

have
: 1

\316\267
elements.

<

<
\320\263

cardinality,

Definition 3.6.5.

Let

\316\267
be

iff

it has

n,

Now we want
the set

we want

have

cardinality

has equal
then

Y,
with

\316\245
has

\316\245
and

\316\245

Z.

with

cardinality

Certainly

Then

cardinality with

has equal cardinality


then X has equal

number.

a natural

\316\267
elements.

0; the

sets.

3.6.1.

See Exercise

Proof.

our

Let

3.6.4.

Proposition

0} is

(This

just the

to

a set

when

say

G N
{\320\263

is true

: 1

<

<
\320\263

even when

empty set.)

Using

we thus define:

a natural

number.

A set

equal cardinality with

X is saidto

N
:
\342\202\254
{\320\263

1 <

79

of sets

Cardinality

gm

also

\342\200\242
< n}\302\267We

X has

that

say

iff

\316\267
elements

it has

cardinality

ft.

< i

: 1

\320\270
e
o\302\243

can use the set {i


< n}, since thesetwo

One

3.6.6.

Remark

:
N
\342\202\254
sets

n} instead
equal

<

have

clearly

is the
bijection?)
cardinality. (Why? What
3.6.7. Let a, b, c, d be distinct objects. Then {a, b, c, d}
Example
: t <
N
same
cardinality as {t \342\202\254
4} = {0,1,2,3} or
has the
=
<
<
i
4.
: 1
N
4}
{1,2,3,4} and thus has cardinality
ii \302\243
1.
the
set
has
cardinality
{a}
Similarly,
There

different

two

have

be

might

Proposition 3.6.8
some

with

prove

3.6.9.

Lemma

Then X

that

any

\321\202\321\204\320\277.

>
\316\267

1,

X has

and

is
and if \317\207
is non-empty,
element
any
\342\200\224
X
with
\317\207
the
element
(i.e.,
removed)
{x}

Then

set X

a set

other

we need a lemma.

this proposition,

Suppose

X be

Let

cardinality).

cannot have
have cardinality m for any

cannot
we

Before

of

(Uniqueness

n.

cardinality
X

i.e.,
cardinality,

with this definition: a set might


But this is not possible:

one problem
cardinalities.

cardinality n.
of X, then the
has

cardinality

n-1.

Proof. If X is
as

cardinality

is no

as
cardinality
X

to

N
:
\342\202\254
{\320\263
N
\342\202\254
{\320\263

number

between

{ieN:

1<

we
if

define
>

f(y)

from the
an

\317\207
be

empty

of X.

element
1 <

<
\320\263

iV},

<

i <

n}.

1 and

clearly

set {i

non-empty

bijection
let

Now

it

then

empty

the

n. Now

cannot

\316\235:
\316\225

<
\320\263

there

as

\316\267},

to a non-empty
set (why?).
Since X has the same
we thus have a bijection/ from
set

In particular, f(x)
define the function
g

the following rule: for any


1}
:=
if
g(y)
g(y)
f(y)
f(y) < f(x), and define
f(x).
(Note that f(y) cannot equal f(x)
\342\200\224
< \316\267
\320\263
by

same

have the
<

is a

: X

natural

\342\200\224

{x}

to

\342\200\224
\316\247
\302\243
\321\203
{\317\207},

:=
since

f(y)

- 1

\321\205
\321\203
\321\204

and
this
is also
/ is a bijection.) It is easy to checkthat
map
a bijection (why?), and so X \342\200\224
with
has
equal
cardinality
{x}
\342\200\224
has
{ieN:l<i<n\342\200\224 1}. In particular X
cardinality
{x}
n\342\200\224 as

1,

desired.

3. Set

80
we

Now

theory

the proposition.

prove

Proof of Proposition 3.6.8. We


0. Then X must be empty,

First

suppose that
have any non-zero

cannot

so X

and

n.

on

induct

=
\316\267

that the proposition is already proven


Now
suppose
Let X have cardinality n++\302\267
n; we now prove it for n++.
m \317\206
also
has
some
that
X
n+f,
suppose
cardinality

cardinality.

some

for
and

then X

m\342\200\224
1,

by

=
\316\267

3.6.4
know, thanks to Propositions
sets {0,1,2} and {3,4} do not have
equal
the first set has cardinality 3 and the secondset

since

we now

2.

cardinality

3.6.10

(Finite sets). A set is finite iff it has cardinality


n; otherwise, the set is calledinfinite.

number

natural

some

finite set,

is a

we use #(X)

3.6.11. The sets

Example

set (0 is a natural
=
2,and#(0) = O.
#({3,4})
we

give

an example

Theorem 3.6.12.
Proof. Suppose
numbersN
finite,
for

was

is

there

set

The

of

sake

so it

/ from

a bijection

to denote

the

{0,1,2}

and

the empty

Now

for

all

1 <

< \316\267
\320\263

(Exercise

equal

are

and

as is
= 3,

finite,

#({0,1,2})

of an infinite set.

:
N
\342\202\254
{\320\263

3.6.3).

infinite.

the set of

that

contradiction

had some

of the

N is

numbers

of natural

natural

cardinality #(N) = n.
1 <

the sequence/(1), /(2),...,


that there exists a natural number

M +1 isnot
to any
that / is a bijection.

of X.

cardinality

{3,4}

number),

show that
precisely

cardinality

\342\226\241

instance,
that
the

Definition
If

element

any

has

By induction hypothesis, this meansthat


that m = ra-H-, a contradiction. This
implies

for

cardinality,

\316\267
for

cardinality

\317\207
is

also

\316\267
and

induction.

3.6.8,

has

has

3.6.9.

which

1,

Thus,
and

{x}

Lemma

the

closes

\342\200\224

\342\200\224

3.6.4,

Proposition

By

of X,

other
is non-empty, and if

But

<
\320\263

then

/(\320\263), contradicting

or

bounded,

\316\234
such

that

more
<

/(\320\263)

the natural
the

can

One

N.

to
\316\267}

f(n) is

Then

\316\234

number

hypothesis

81

of sets

\316\277 Cardinality
\316\262.

to show
3.6.13. One can alsouse similar
arguments
Remark
set
is
for
the
instance
rationale
unbounded
infinite;
Q
that any
we
will
in
later
are
R
construct
reals
(which
chapters)
and the
it is possible
be
for
some
to
\"more\"
infinite
sets
However,
infinite.
8.3.
Section
see
others;
than

Now we relate cardinality

of natural

arithmetic

the

with

numbers.

3.6.14

Proposition

X be a finite
element of X.

(a) Let

(b) Let X
<

(d)

#P0

be a

Let

\316\245
be

and

=
\316\245
\316\247
\316\240

(i.e.,
(c)

finite,

is

a proper

is

If

in

and

and

\316\245
be

Then

{x})

finite

using

recursively

the

product

sets.
Then the set Yx
and #(YX) = #(\320\243)#(\320\245).

f is

\316\247
\317\207
\316\245

(defined in
D

to

defined

If in addition

Cartesian

3.6.15. Proposition 3.6.14suggests


define

\316\245

then f(X)

3.6.4.

Exercise

\316\245

\317\207
#(\320\243).

Remark

instead

not an

#(X).

xY) = #(X)

is finite

a function,

#(X).

another

way

= #(X) + #(Y).

\342\200\224>
\320\243
is

<

sets.

finite

#(X

3.10)

See

U Y)

#(/(X))

X and \316\245be

Axiom

not

and #(X

finite set, and let \316\245be a subset of X. Then


X (i.e.,
\317\206
#(Y) < #(X)- If in addition\316\245
subset of X), then we have
< #(X)\302\267
#(Y)

is a

finite

Let

which is

object

Then XUY is finite and #(XU


addition
X and \316\245are disjoint

0), then #(X

finite set, and f


is a finite set with #(/(X))

(e) Let

an

is finite

and

is

If X

\317\207
be

U {x}

sets.

#00\302\267

then

Proof.

finite

one-to-one,

(/)

and let

set,
Then

+ 1.

#(X)

Y)

arithmetic).

(Cardinal

that

there

is

the arithmetic operations of natural numbers;


as in Definitions 2.2.1, 2.3.1, 2.3.11,but

notions

of union,

Cartesian product, and

power

Set theon

3.

82

which
is an alternative
set. This is the basisof cardinal
arithmetic,
we have devel\302\267
foundation to arithmeticthan the Peano arithmetic
in
this
this
we
not
arithmetic
will
but we
text,
here;
develop
oped
of how one would work with this arithmetic
some
examples
give
Exercises

in

3.6.6.

3.6.5,

This concludes our discussionof finite


infinite sets in Chapter 8, oncewe have

examples of infinite sets (suchas the


Exercise 3.6.1.

3.6.3. Let

Exercise

<

f(i)

\316\234
for

all 1

peek
bounded.

3.6.4. Prove

Exercise

Exercise3.6.5.

Let

have
two

a set X has cardinality

reals),

and

only if X

a natural

number,

and let

existsa natural

<

induct

< \316\267.
\320\263
(Hint:

N
\302\243

number

on n. You may
of the natural

subsets

finite

/ : {%

: 1

<

3.6.6.

Exercise

a, 6, c.

also want
numbers

Proposition 3.6.14.
and

\320\222
be

Let

the

two

be
\320\241
\320\220,
\320\222,

cardinality

sets.

Usea similar

Conclude
argument

3.6.7. Let A and


if
to \320\222
equal cardinality
B.
A
if
Show
that
and
to
Exercise

to

cardinality

3.6.8.

Exercise

: A

\320\263
<

sets.

that

Show

\316\221
\317\207
\316\222
and

\316\222
\317\207
A

the
of

2.3.2.

\320\264\320\262\321\205\321\201
have
eqUal
between

is

\316\234
such

equal cardinality by constructing an explicitbijectionbetween


sets. Then use Proposition 3.6.14to concludean alternate proof

Lemma

0 if

Show that there

5.1.14.) Thus

at Lemma

are

\316\267
be

a function.

\342\200\224>
N be
\316\267}

that

more

set.

empty

to

and

rationale

integers,

discuss

a few

constructed

3.6.4.

Proposition

3.6.2. Show that

Exercise
the

Prove

shall

We

sets.

\320\222
if

and

Let A

\342\200\224\342\226\272
\320\222
A
from

to

\320\222
be

\320\222
are

finite

only if #{A)
\320\222
be

\320\222

(i.e.,

that

sets
A has

the sets (AB)Cand


explicitbijection
any
\317\207
ac

natural
numbers
= a6+c.

Let us say that A has lesser


\320\222
an injection / : A \342\200\224>
from
A
has
then
lesser
or equal
sets,

sets.

exists

there

and

Show

sets.

by constructing an
that (ab)c = abc for
to also conclude ab

<

or
A

#(\320\221).

such that

lesser or

there exists an

equal cardinality

injection

to

B).

\342\200\224*\302\267
A from
there then existsa surjectiong : \320\222
\320\222
to A.
(The
of choice;
converse to this statement requiresthe axiom
see Exercise
8.4.3.)
sets. Show that AuB and \316\221\316\240
\316\222
be finite
Exercise 3.6.9. Let A and \320\222
are
also finite sets, and that #{A)+ #{B) = #{AU\320\222)+ #(\320\220\320\237
\320\222).

Show that

$.6.

of sets

Cardinality

An
Exercise 3.6.10. Let \316\221\317\207,...,
Show
\316\263\316\271.
known

as

that
the

there

exists

pigeonhole

sets such that #(|Jie


,n} such that #(A*) >

be finite

\320\263
G {1,...

principle.)

Chapter

Integers

and rationals

4.1 The integers


In

built up most
but we have

2 we

Chapter

number

system,

do with

just addition

of the basicproperties
ofthe natural
reached the limitsof what
one

and multiplication.We

now

would

a new operation, that of subtraction,but


number
the
natural
we will have to pass from

introduce
properly

a larger numbersystem,that of the

to

that

do

to

can

like

system

to

integers.

you can get by subtracting


\342\200\224
for
5 should
be an integer, as
3
numbers;
instance,
\342\200\224
should 6 2. This is not a complete definition of the integers,
it doesn't
because
say when two differencesare equal(for instance
(a)
is not equal to 1\342\200\2246),
is equal
to 2\342\200\2244,
but
we should
know why 3\342\200\2245
the

Informally,

two

and

are

integers

what

natural

(b)

it doesn't
one add

does

(how

say how to do arithmeticon thesedifferences


5 to 6 \342\200\2242?).
3 \342\200\224
Furthermore,
(c) this definition

which
because it requiresa notionof subtraction,
we
once
define
the
a
re
constructed.
only adequately
integers
of our prior experience
because
with integers we know
Fortunately,
to these questions should be. To answer
what
the answers
we
(a),
\342\200\224
b = \321\201
d
know from our advanced knowledgein algebrathat a \342\200\224

is circular
can

exactly

happens

when a+d

of differencesusingonly
answer

and

(b)

that

we

know

the

concept

from algebra
=

\321\201+6, so

(a\342\200\224b)(c\342\200\224d) (ac+bd)

that

we

can characterize

of addition.

equality
Similarly, to
=

(a\342\200\224b)+(c\342\200\224d)

\342\200\224

(ad+bc).

So we

(a+c)

will take

\342\200\224

(b+d)

advan-

85

integers

The

la

all this into the definition


our foreknowledgeby building
as we shall do shortly.
integers,
of the
to resolve (c). To get aroundthis problem
have
we
will
still
We
work-around: we will temporarily write integers
use the following
\342\200\224
\320\260 b,
instead
use a new notation \316\261\342\200\224b
but
difference
a
as
not

tage of

to

similar to the commain


for

is in
\320\260\342\200\224b

that

see

the notation

the

\342\200\224

is

Cartesian

plane. Later

in the

points

the

where

integers,

define

when we define subtractionwe

equal to
only needed right

\342\200\224
\316\261 b,
and

fact

it is
\342\200\224;

place-holder,
notation
(\320\266,
\321\203)

meaningless
co-ordinate

so we

now to

avoid

will

can discard
circularity.

the scaffoldingused to construct


a
(These
essential to makesurethe building
they
building;
but
once
are
the
built
is
correctly,
building is completedthey
seem
thrown
away and never used again.) This may
unnecessarily
in
order
to
we
define
that
are very
already
something
complicated
but
will
to
we
use
this
device
construct
the
familiar with,
again
and knowing these kinds of constructionswill be very
rationale,
are similar to
are temporarily

devices

in later

helpful

chapters.

Definition 4.1.1 (Integers).An


where
\320\260\342\200\224b,

form

equal,

We let

the

\316\226
denote

Thus

for

3 +

because

set

3\342\200\2245is

an

integer,

and is

5. On the
hand,
5.
This
notationis
+
3^2

3 +

it is not of the

only

form

\320\260
+

d =

\321\201
+ b.

strange

not

2\342\200\2244,

equal

looking,

to
and

not yet an integer, because


these problems later.
rectify

3 is

instance,

\320\260\342\200\224b!
We

equal to

3\342\200\2245is

other

for

if

integers.

4 = 2+

few deficiencies;

1Inthe

of all

of the
Two integers are

expression1

numbers.

natural

= \321\201\342\200\224d,
\320\260\342\200\224b
if and

instance

2\342\200\2243
because

has a

b are

\316\261
and

to be
considered

is an

integer

will

with
the
theory, what we are doing here is starting
Then
we place an
numbers.
(a, b) of natural
equivalence relation ~ on these pairs by declaring
(a, b) ~ (c,d) iff a+d = c+b.
The set-theoretic interpretation
is that it is the space of all
of the symbol a\342\200\224b
to (a, b): a\342\200\224b
:= {(c, d) \342\202\254
\316\235
\317\207
\316\235
: (a, b) ~ (c, d)}. However,
pairs equivalent
this
the integers and we
interpretation
plays no role in how we manipulate
will
not
refer to it again. A similar
can be given
set-theoretic
interpretation
to the construction
of the rational
numbers
later in this chapter, or the real
in the next chapter.
numbers

space

language

\316\235
\317\207
\316\235
of

of set

ordered

pairs

4- Integers and rationale

86

substitution

axiom.

equations

together

we obtain
cancel the

a\342\200\224b=

can

we

2.2.6

Proposition

e\342\200\224/.

\321\201
and

any operations on
yet defined
do define our basic operations
and order j

multiplication,

axiomat that time


will

(We
advanced

be

only

to

this

we

ensure

not

as

such

we

addition,

to verify the substitution


that the definition is valid.
basic

the

have
when

However,

more

operations;

integers,

of the

in terms

because

integers,

for

the

for

As

such as exponentiation,
basic ones, and so we do not need

the

on

operations

defined

order

to do

in

need

two

By

f = b + e, i.e.,
law was needed to make

the integers.
on the
have

e\342\200\224-/.

a+

d, obtaining

will

we

and

the

Adding

= c+b+d+e.

a+d+c+f

and

transitivity
=
\321\201\342\200\224d

d + e.

/ =

of equality is sound.
verify it at this stage

cannot

we

substitutionaxiom,

\321\201
+

and

the

\321\201\342\200\224d
and

cancellation

the

Thus

notion

our

that

sure

+ d = c+ b

have

we

a\342\200\224b=

that

know

we

Suppose

Then

4.1.1 and insteadverify

to Exercise

symmetry

a legitimatenotion of

this is

that

to verify the reflexivity, symmetry, transitivity,


axioms
reflexivity
(see Section A.7). We leave

need

equality. We

check

to

have

We

will

to

re-verify the substitution axiomfor the advanced


operations.)
Now we define two basic arithmetic operationson integers:
addition

and

4.1.2. The sum of two

Definition
defined

multiplication.

by the

(a\342\200\224b)

(c\342\200\224d),

is

formula
(a\342\200\224b)

The product

integers,

(c\342\200\224d)

:=

(a +

c)\342\200\224(b +

of two integers, (a\342\200\224b) x


(a\342\200\224b)

(c\342\200\224d)

:=

(ac +

(c\342\200\224d),

bd)\342\200\224{ad

d).
defined

is

by

+ be).

for instance, (3\342\200\2245) + (1\342\200\2244) is equal to (4\342\200\2249). There


is however
one thing we have to checkbeforewe can accept
these
definitions - we have to checkthat if we replace
one of the integers
Thus

by an equalinteger,

instance,
to

have

not give
the case:

that

is
(3\342\200\2245)

the same

the

equal

value as

to

sum

or product

so
(2\342\200\2244),

(2\342\200\2244) +

does not change. For

(3\342\200\2245)

otherwise
(1\342\200\2244),

a consistent definition of addition.

Fortunately,

(1\342\200\2244)

this

ought

would
this

is

87

The integers

/J.

4.1.3

JjemxnsL

a b>a!,bf,c,dbe
(a-\342\200\224b)+

(c\342\200\224d)

and

(c\342\200\224d)

(a'\342\200\224b')

also

and

(c\342\200\224d),
\320\270'.\342\200\224bf)

numbers.

natural

(a\342\200\224b)

If

(c\342\200\224d) +

(a\342\200\224b)

(a\342\200\224b)

(c\342\200\224d)

(c\342\200\224d)+

(a'\342\200\224b')

and

addition

Thus

\317\207
(c\342\200\224d)
(a7\342\200\224bf).

then

(af\342\200\224bf),

x
(a\342\200\224b)

and
(c\342\200\224d)

are well-defined

multiplication

are well-defined). Let

and multiplication

(Addition

operations {equalinputs

equal

give

outputs).

that

To prove

Proof.

(a\342\200\224b)

as (a

sides

evaluate

both

Thus we

need to show that

since

(a\342\200\224b)

adding

\321\201
+

(c\342\200\224d)

a + c+

b'

a +

have

b' =

a' +

we obtain the claim.

both sides

(a7\342\200\224b')

(c\342\200\224d),

we

d) and (a7 + c)\342\200\224(b7 + d).


a' + \321\201
+ b + d. But
+ d=

c)\342\200\224(b +

we

(a7\342\200\224b7),

d to

so

and

b,

by-

show

we

Now

Both
sides
evaluate
that (a-^-b) x (c\342\200\224d) = (a7\342\200\224b7) \317\207
(\321\201\342\200\224d).
to (ac + bd)\342\200\224(ad + be) and (a'c + b7d)\342\200\224(a7d + b'c), so we have
= a'c + b'd + ad + be. But the
b'c
+
to show that ac

+ bd+a'd

as

+ d(a +

c(a' + b)

equal.The

are

identities

behave
The integers \316\267\342\200\2240
numbers n; indeedonecan check
m)\342\200\2240

is

and

to

equal

numbers

since

equality

they
number

natural

3\342\200\2240, thus

integer

and

integers

3 =

3 is

=
\316\267

Furthermore,

\321\202.

(n +

(n\342\200\2240)

mathematical

(The

natural

the

between

form

the natural

(m\342\200\2240)

isomorphism

of the

3\342\200\2240.

equal to 1\342\200\2240. Of
also be equal to any

instance

nm\342\200\2240.

if

only

(n\342\200\2240)

Thus

\316\267\342\200\2240.)

we

may

=
numbers with integersby setting
\316\267 \316\267\342\200\2240;
our definitions of addition or multiplication
or
with each other. For instance
are consistent
3 is now considered to be the same
as the

1 is

it will
for

affect

not

(m\342\200\2240)

if

those

\316\267
and

does

way as

same

the
that

factors

sides are

similarly.

proven

in

is that there is an

the natural

identify
this

and

(m\342\200\2240)

for this

term

the

x
(n\342\200\2240)

\316\261
+ b7

Since

b7).

two

other

+ b), while the right


a7 + b, the two

as c(a+b,)+d(a/

factors

side

left-hand

equal not

course,
other

In

0 is

particular

if we
integer

set

which

only to 3\342\200\2240,but

equal to
to

\316\267
equal

also

then
\316\267\342\200\2240,

is equal
to

0\342\200\2240

to

\316\267\342\200\2240,

4\342\200\2241,5\342\200\2242,

etc.

We
x++

:=

can

incrementation on the integersby


any integer x; this is of courseconsistent

now define

\317\207
+

1 for

defining
with

88

and rationale

Integers

4-

our definition of the incrementoperationfor natural


numbers.
this is no longer an importantoperationfor us, as it has
However,
been now superceded by the more general notion of addition.
basic
on the integers.
Now we considersomeother
operations
4.1.4 (Negationof integers).If (a\342\200\224b) is an integer,
to
be the
we define the negation \342\200\224(a\342\200\224b)
integer (b\342\200\224a). In
= \316\267\342\200\2240
if
\316\267
is
we can define
a
natural
number,
positive
particular
Definition

its

= 0\342\200\224n.
\342\200\224n

negation

instance

For

\342\200\224(3\342\200\2245)
(5\342\200\2243).

One

this definition

check

can

is well-defined(Exercise4.1.2).
can
now show that the integers correspond exactlyto what
We
we

expect.

4.1.5 (Trichotomy of integers).

Lemma

Let \317\207
be

an

Then

integer.

istrue:(a) x is zero;
exactly one of the following threestatements
\317\207
is
a
natural
to
number
equal
positive
n; or (\321\201)\321\205is the
(b)
\342\200\224n
a
natural
n.
number
positive
of
negation
Proof

= a\342\200\224bfor
\317\207

definition,
cases:
a > b, a = b, or
number
natural
positive
which
is (b).
If a = b,

is (a). If
number
is

a<

<

c, which meansthat

then

previous

a\342\200\224b=

a, so

that

reasoning,

a\342\200\224b=

a\342\200\224a=
b\342\200\224a=

some

=
\321\201\342\200\2240
\321\201,

0\342\200\2240=

0, which
some
natural

\316\267
for

and thus

By
three

=
a\342\200\224r-b

which
\342\200\224n,

(c).

Now we

show that no morethan

at a time. By

so

natural

some

b >

then

b,

the

\316\267
by

one of (a), (b), (c) is true.


numbers a, b. We have
If
a
b.
for
> b then a = b + \321\201

at least

that

show

first

We

(a)

and

definition,

(0\342\200\2240)

that
exactly

for
\342\200\224m
\316\267
+

If (b) and

some

m = 0

one

(b), (c) can hold


number is non-zero,

(a),

be true. If (a)
simultaneously
=
for
then
0
\342\200\224n
some
true,
positive
= 0 +
so
that
0 + \316\267
0, so
(0\342\200\224n),

a contradiction.
=
\316\267

of

cannot

(b)

simultaneously

thus

one

natural

a positive

positive

(c)

n,m,

were

(n\342\200\2240)

(c)

were

natural

that
true,

simultaneously

so that

and

n;

=
\316\267

0,

then

(0\342\200\224m),

so

+ 0, which contradictsProposition2.2.8.Thus
x.
D
(b), (c) is true for any integer

of (a),

89

The integers

Ll.

If

is
\316\267

natural

a positive

a negative
number, we call \342\200\224n

Thus every integer is positive,zero,or negative,


one of these at a time.
well
ask why we don't use
could
One
i.e., why didn't we just say
integers;
the

which is
a natural

but

not

Lemma4.1.5to
an

times

or

positive,
to

and

of

the rules

for

split into many different


equals
positive; negative plus

positive

negative,

negative

would

integers

multiplying

negative

define

is anything

integer

either a positivenatural number,zero,or the


number. The reason is that if we did so,

adding and
cases (e.g.,
positive is either
term is larger,etc.)
messier.
being

integer.
more than

verify

on which
would end up

depending

zero,

all the

properties

much

We

now

Proposition4.1.6

(Laws

integers. Then we

properties of the integers.

the algebraic

summarize

of

for

algebra

integers).

Let x,y,z

be

have

x+y=y+x

(x + y) + z =
x + 0= Q + x =
\317\207
+

=
=
+ \317\207
(\342\200\224\317\207)
(\342\200\224\317\207)

x+

(y

+ z)

\317\207

xy = yx

(xy)z =
rrl

=
1\320\266

z)x

\317\207

xy +

xz

\342\200\224+

zx.

+ \316\266)
\320\245(\320\243
+
(\321\203

x(yz)

yx

a name;
identities
have
4.1.7. The above set of nine
are
that
the
form
a
commutative
ring. (If
they
asserting
integers
=
one deleted the identity
yx, then they would only assert
xy
that the integersform a ring). Note that some of these identities
were
for the natural numbers, but this doesnot
already
proven
because
that
mean
automatically
they also hold for the integers
the integers are a larger set than
the
numbers.
On the
natural
other hand, this proposition
of the propositions
supercedes
many
derived earlier for natural
numbers.

Remark

90

4-

Proof. There
Lemma4.1.5

two

axe
and

x,

\316\266
are

y,

shorter
for

identities

in

and

((ac
((ace

+ be))

bd)\342\200\224(ad

+ bde

(xy)z

natural

few

lines.

x(yz):

+ adf +

(e\342\200\224/)

bef)\342\200\224(acf

+ bdf

+ ade

+ bee));

+ ade

+ bee

+ bdf))

(a\342\200\224b)((c\342\200\224d)(e\342\200\224/))

(a\342\200\224b) ((ce

and so one

see

can

identities

are

proven

bef +

bde)\342\200\224(acf

x(yz) are equal. The other


fashion; see Exercise 4.1.4. \316\240

and

that

(xy)z

in

a similar

the operation

define

+ de))

df)\342\200\224(cf

+ adf +

= ((ace

now

(e\342\200\224/)

((a\342\200\224b)(c\342\200\224d))(e\342\200\224f)

messy.

=
\316\266
these

expand

namely

x(yz) =

We

a, b, c, d, e, /,

just prove the longestone,

(xy)z

and
(a\342\200\224b), \321\203 (\321\201\342\200\224d),

c, d, e, / and use the algebraof the


in a
allows each identity to be proven

This

We shall

This becomes very

of a, b,

terms

numbers.

on

=
\317\207

numbers

natural

to prove these identities. One is


a lot of cases depending
whether

or negative.

to write

way is

some

into

split

rationals

to use

ways

positive,

zero,

Integers and

of two
of subtraction x\342\200\224\321\203

integers

by the formula

+ (-y).
do not
this
need to verify the substitution axiom
operation,
other
since we have defined subtraction in termsof
operations
on integers,
have
already
namely addition and negation,and
verified
that those operations are well-defined.
now that
if a and b are natural numbers,
One can
check
x-y:=x

for

We

two

we

easily

then

a-b
and

so

a\342\200\224bis

= a

+ -b

just

the

can now discard the


of subtraction instead.

(a\342\200\2240)

same

thing as

\342\200\224

and

notation,
(As

remarked

(0\342\200\224b)

a\342\200\224b,

a \342\200\224
b.
Because
use

of this

the familiar

before,

we

could

we

operation
not use

subtraction immediatelybecauseit would


be
circular.)
We can now generalize Lemma 2.3.3and Corollary2.3.7from
the

natural

numbers

to the

integers:

91

11. The integers

Proposition4.1.8

Corollary
such

integers

Proof. See

and b

both).

D
//
a =

a, b,
b.

\321\201
are

Exercise 4.1.6.

of order, which was defined on the


integers by repeating the definition

notion

the

extend

the

to

numbers,

natural

0 orb = 0 (or

4.1.9 (Cancellationlaw for integers).


that ac = be and \321\201
is \320\277\320\276\320\277-zero,
then

now

We

Let a

divisors).

4.1.5.

Exercise

See

Proof-

zero

that ab = 0. Then eithera =

such

be integers

no

have

(Integers

verbatim:

4.1.10

Definition

integers.
or

> m
\316\267

We

a.

m <

say

that

iff

>
\316\267

n,

Clearly
natural

numbers,

laws

the

the

and

greater
= m
\316\267

n^m.
5 =

because
\342\200\2243,

b if

(b)

(Addition

(c)

(Positive
positive,

and

(f)

only

preserves

if

\342\200\224

b is

is transitive)

(Order

trichotomy)
or a = b is

<b,

\342\200\2243.
\317\206

natural number.

a positive

then a

order)

preserves

b,

then

and b

Exactly

one of the

>

\321\201
>

If a >

b +
b

\321\201

and

is
\321\201

\342\200\224a
< \342\200\224b.

If a> b

true.

be integers.

a^b^c

Ifa>b,

order)

multiplication
then ac> be.

(Order

8 and

of order:

(d) (Negation reversesorder)If a>


(e)

3 +

with the

Lemma 4.1.11 (Propertiesof order). Let


(a) a>

\342\200\224

notion of orderon the


since
we are using the same definition.
of algebra in Proposition
4.1.6 it is not hard to
properties

following

than

greater

strictly

is consistent

definition

this

show

have

we

iff

\316\267
is

\316\267
is

instance 5 >

Thus for

Using

that

and
m be
integers). Let \316\267
or equal to m, and write
+ a for some natural number
than m, and write \316\267
> m or

of the

(Ordering

We say
m < n,

c,

then

a >

\321\201

statements a >

b,

92
See Exercise

Proof.

4.1.1.

Exercise

that the definition

sense

the

that

Exercise 4.1.4. Prove

the

the

on

is

integers

definition
of negation on the integers is
if (a\342\200\224b)= {a'\342\200\2246'), then
=
\342\200\224(a\342\200\224b)
have equal negations).

\317\207=
for
(\342\200\2241)\316\261 \342\200\224a

every

a.

integer

in Proposition 4.1.6. (Hint:


identities
some identities to prove
others.
For
once you know that xy = yx, you get for free that xl = 1\320\266,
you also prove x(y + z) = xy + xz, you automatically
get
remaining

save some work by using

can

instance,
and

that

integers

\342\200\224(af\342\200\2246')
(so equal

Exercise4.1.3.Show
one

of equality

the

that

Show

in

symmetric.

Exercise4.1.2.
well-defined

rationale

4.1.7.

Verify

both reflexiveand

and

Integers

4-

once

(y + z)x = yx +
Exercise 4.1.5.

for free.)

zx
Prove

4.1.8.

Proposition

(Hint:

while this proposition

is not quite the same as Lemma


it is certainly legitimate to use
2.3.3,
Lemma 2.3.3in the course of proving Proposition 4.1.8.)
Exercise4.1.6.Prove Corollary
4.1.9.
(Hint: there are two ways to do
b must
be zero.
this. One is to use Proposition 4.1.8 to concludethat a \342\200\224
Another

combine

is to

way

4.1.7. Prove

Exercise

lemma to

apply directly

with

the

Lemma

of induction

principle

to the integers. More

a property P(n) pertaining


that P(n) impliesP(n-\\-+)
for all integers n. Thus
the integers as it is with
worse

with

4.1.5.)

Lemma 4.1.11. (Hint: use the first

that

Show

even

2.3.7

of this

part

prove all the others.)

Exercise4.1.8.
not

Corollary

the

precisely,

(Axiom

give

2.5)

does

an example

of

such
that
integer \316\267
P(0) is true, and
for all integers
n, but that P(n) is not true
induction
is not as useful a tool for
with
dealing
the
natural
numbers.
(The situation becomes
we shall
rational and real numbers, which
define
to

an

shortly.)

The rationals

4.2
We
addition,

have

now
subtraction,

constructed

the integers,
and

multiplication,

expected algebraic order-theoretic


use a similar constructionto
to our mix of operations.
and

build

with the operationsof


order and verified all the
propertiesr

the

rationals,

Now
adding

we will
division

93

rationale

\316\271
J7&e
\302\253g

the integers wereconstructed


two
subtracting
by
the
be
rationale
can
two
constructed
by
dividing
natural
of
we
to
course
makethe
have
usual
caveat
that
though
integers,
should be non-zero2. Of course,just as two
denominator
the
c\342\200\224d
can
be equal
b and
if a+d = c+b, we know
a \342\200\224
(from
differences
that
two
advanced
quotients a/b and c/d can be
knowledge)
more
with
the integers,
we create a
if ad = be. Thus, in analogy
equal
will
be
symbol
meaningless
eventually
superceded
// (which
new
Just like
numbers,

define

and

division),

by

4.2.1.

Definition

is an

number

rational

expression of the

a and b are integers and b is


where
number.
Two
considered to be a rational
=
be
to
if
equal,a//b c//d, and
considered

non-zero;

a//b,

rational
only

form

is not
a//0
numbers
are
if ad = cb. The

set of all rationalnumbersis denotedQ.


This

instance

for

Thus

= 6//8

3//4

valid definition of

is a

= -3//

- 4,

(Exercise

equality

but

3//4

4.2.1).

\317\206
4//3.

Now

we

notion of addition, multiplication,


and
negation.
Again,
of
our
take
which
tells
we will
advantage
pre-existingknowledge,
that
and
us that a/b + c/d should equal (ad + bc)/(bd)
a/b * c/d
Motivated
should equal ac/bd, while
by
\342\200\224(a/b) equals
(\342\200\224a)/b.
need a

this

we define

foreknowledge,

Definition 4.2.2. If a//b


define

their

+ (c//d)

(a//b)
their

and

c//d

are

rational

numbers, we

sum

:= (ad +

bc)//(bd)

product

(o//6)*(c//d):=(oc)//(M)

and the

negation

-ia//b) := (-a)//ft.
is no reasonable
have
we can divide by zero, since one cannot
way
the identities (a/b) *6 = a and c*0 = 0 hold simultaneously
if b is allowed

2There

both

to be zero. However,
by a quantity which
10.5),

which

we

can

approaches

suffices for

doing

eventually
zero
things

get a reasonablenotion
of L'Hopital's rule
defining differentiation.

- think
like

of

dividing

(see Section

4- Integers and rational

94

Note

if

that

by Proposition

a//b

with

then

the

claims

to

abf =

remaining

b and

+ c//d

a//b

(ad+bc)//bd
to

have

show

and

that

bb'cd = a'bd2 +

abf

alb, the

bbfcd.

claim follows. Similarly if

one

replaces

by d//d'.

c//d

bc)b'd=(a!d + b'c)bd,

ab'd2 +
since

on

to

expands

But

side is

left-hand

the

that

show

now

We

+ b'c)//b'd, so we

is (a'd

{ad +

the

we leave

addition;

a'b.

definition,

By

side

right-hand

which

operations

negation

in the

Suppose a//b = a'//b', so that

4.2.2.

and

non-zero

for

this

verify

Exercise

af//bf + c//d.
the

number

c//d.

We just

are

non-zero

rational

sense that if one replaces


number a'//b' which is equal to \316\261/\316\262
rational
the above operations remains unchanged,
and

of

output

Proof
b1

sum,
product,
are well-defined,

another

for

similarly

and

The

4.2.3\302\267

numbers

rational

also

is

rational number.

remains a

Lemma

d are non-zero, then bd


b and
of a
4.1.8, so the sum or product

We note

that the rationalnumbers

identical to the integersa:


+

{a//\\)

= (a

(6//1)

behave

a//1

in a

manner

+ b)//l;

{a/II) x (b//l) = (ab//l);

-(a//l)
Also,
Because

a//1
of

and 6//1
this,

we

are

only

will identify

equal

(-a)//l.
when

a and

a with a//1

for

b are

equal.

each

integer

arithmetic
the
a//l; the above identitiesthen guaranteethat
of the integers is consistent with the arithmeticofthe rationale.
as we embedded the natural numbersinsidethe integers,
Thus
just
the integers inside the rational numbers.In particular,
we embed

a =

a:

1 is

and

rational

are

numbers

natural
\320\260\320\224

to 0//1

95

rationale

The

4.2-

for instance 0 is equal

numbers,

to 1//1.

equal

to 0 = 0//1 if
is equal
a//b
Observe that a rationalnumber
=
b \317\207
if the numerator
a is equal to 0.
if a x 1
0, i.e.,
only
and
are
non-zero
then
so
b
is
and
a
if
a//b.
Thus
If
a new operation on the rationale:reciprocal.
define
now
We
we define
is a non-zero rational (so that a, b \317\206
0) then
x = a//b
:= b//a.
of \320\266
to be the rational
It
number \317\207~\316\273
the reciprocal \317\207~\316\273
with
our notion
is easy to checkthat this operationis consistent
two
then
if
rational
numbers
are
equal,
a//b, a1//V
of equality:
are
a
n
as
also
such
equal. (In contrast, operation
their reciprocals
are
\"numerator\" is not well-defined: the rationale3//4 and
6//8
to
s
o
we
have
be
careful
when
have
but
unequal numerators,
equal,

referring to such termsas \"the numerator


leave the reciprocal of 0 undefined.
now

We

the

summarize

Proposition4.2.4
rationals. Then the

of algebra

We

however

of the rationals.

for rationals).

algebra
laws

following

\320\266\".)

properties

algebraic
of

(Laws

of

Let x^y^z be

hold:

x+y=y+x

(x + y)+z =
x + 0 = 0+ =
x

\317\207
+

=
=
+ \317\207
(\342\200\224\317\207)
(\342\200\224\317\207)

xy =

(xy)z =
=

xl

+ z)

x(y
(y

Ifxis non-zero,

we

also

lx =

+ z)x

x~lx

they
better

are
than

4.2.5. The
asserting
being

above

set

of

the rationals
a commutative
ring
that

(y +

z)

\317\207

yx
x(yz)
\317\207

= xy
= yx

+ xz
+ zx.

have
=
\321\205\321\205~\320\263

Remark

= 1.
ten

identities

have a

name;

Q form a field. This is


because of the tenth identity

96

xx l =

lx
\317\207

that this

Note

I.

sition4.1.6.
To

Proof.

=
\316\266
and
We

this

prove

proposition supercedesPropo.

one writes

identity,

=
\317\207

a//b,

\321\203

c//d

integers a,c,e and non-zerointegers


b,d,/
each identity in turn using the algebraof the integers.
=
just prove the longest one, namely
(x+y)+z
x+(y+z)\\

e//f
verifies

shall

for some

y) + z={(a//b)

(x +

=
x+

+ bcf

{adf

+ bde)//bdf-

+ ((c//d) +
+

and so one

can

(e//f))

de)//df)

+ bde)//bdf

\317\207
and
+
(x + y) + \316\266
are proven in a similarfashion

that

see

identities

other

+ bcf

(adf

(e//f)

+ (e//f)

((ad + bc)//bd)

= (a//b)
z)

(y

+ (c//d))

= (a//b) + ((cf

The

and rational^

Integers

4-

are

equal.

are

left to

+ \316\266)
(\321\203
and

Exercise 4.2.3.
can

We
\317\207
and

y,

D
that

provided

quotient x/y of two rational


by the formula

the

define

now

x/y :=

\317\207
\317\207
\321\203~\320\263.

instance

for

Thus,

numbers

is non-zero,
\321\203

x (6//5) = (18//20)= (9//10).


this
it is easy to see that a/b = a//b for every
Using
formula,
a
and
non-zero
integer
every
integer b. Thus we can now discard
of a//b.
the // notation, and usethe morecustomary
instead
a/b
of
normal
rules
us
to
use
all
the
4.2.4allows
Proposition
we
will
now
further
comment.
to
so
without
do
algebra;
proceed
In the previous
we organized
the integers into positive,
section
= (3//4)

(3//4)/(5//6)

zero,and
Definition
we

have

be

negative

=
\317\207

4.2.6.
=
\317\207

We

numbers.

negative

a/b

iff we

for
(\342\200\224a)/b

have

some

same

\317\207
is

said

the

rationals.

to be

positive

for

iff

positive integers a and b. It is said

=
\317\207

positive

do the

number

rational

for some

now

to

for
some
\342\200\224\321\203

integers

positive

a and

b).

rational

\321\203
(i.e.,

/ 2.

97

rationale

The

every positive integer is a positive


rational
integer is a negative
with
is
old
consistent
our
one.
definition
new
for instance,

Thus
number,

our

% w

(a)

(Trichotomy
one of
exactly

Then

number.

to 0.

equal

rational

a positive

a rational

three statements is true:

the following

\317\207
is

number,

number,

(c)

\317\207
is

Proof See Exercise4.2.4.


4.2.8

Definition

number, and

iff
\321\203

or

\317\207
<y,
\321\203,

that

yiSx

a negative
=
\317\207
and
\321\203,

rational

positive

rational

number.

We

< \321\203.
\317\207

define

similarly

be
\321\203

\317\207
and

of order on the rationals).


the following properties hold.

one of

=
the three statements\317\207

true.

\317\207
is
> \321\203

is anti-symmetric)

(Order

(b)

\317\207
>

\342\200\224
is a
\321\203

Exactly

trichotomy)

(Order

rationale). Let

(Basic
properties
numbers.
Then

be rational
x, y, \316\266

(a)

of the

\342\200\224
is
\317\207. \321\203

\317\207
or
> \321\203

4.2.9

Proposition
Let

We say

\317\207
<

> \321\203
iff either
\317\207

write

(Ordering

numbers.

rational

number.

rational

negative

(b)

Let \317\207
be

of rationale).

4.2.7

Lemma

rational

every negative

and

One has \317\207<

\321\203
if

and

only

if

y>x.

(c) (Order is transitive)


preserves

(Addition

(d)

If

(e) (Positive

order)

Ifx<y,

then
< \316\266,
\321\203

then x +

\317\207
< \316\266.

\316\266
<

<
multiplication preservesorder)If \317\207

y +

z.

and
\321\203

Proof. See Exercise4.2.5.


Remark

4.2.10.

combined

with

they

assert

important to

is positive,

keep

that

above

The
the

\316\266
is

xz <yz.

then

positive,

\317\207
and
< \321\203

field

the

in mind

axioms

rationals

five properties in Proposition 4.2.9,


in Proposition 4.2.4, have a name:
Q form an orderedfield.It is

that Proposition
4.2.6.

see Exercise

4.2.9(e) only

works

when

\316\266

4- Integers and

98

for the rational


the definition of equality
transitive.
and
symmetric,
(Hint: for transitivity,

Show that

4.2.1.

Exercise

is reflexive,

bers

Exercise 4.2.2.

the

Prove

Exercise4.2.3.Prove
(Hint: as with Proposition
to prove

identities

Exercise 4.2.4.
to

have

Prove

prove

(b), (c)is true; and


Exercise 4.2.5. Prove

use

the

4.2.3.

of Lemma

components

remaining

of Proposition

components
4.2.4.
you can save some work by using some

remaining
4.1.6,

others.)
Lemma
4.2.7.
(Note that, as in
that
two different things: firstly,
at
one
of
most
that
secondly,
(a),

Proposition 2.2.^
at least one of (a),
(c) is

true.)

such that

\317\207
<

(b),

4.2.9.

Proposition

4.2.6. Show that if \320\266,


yy
then xz > yz.

Exercise

real

are
\316\266

numbers

\321\203

is negative,
\316\266

and

Absolute value and

4.3
have

We

basic arithmetic operations

the four

introduced

already

exponentiation

of addition,

\342\200\224
:=
\317\207 \321\203

\317\207
+

of order <, and have


the

rationale,

One can

basic

these

use

operations. There are many


but we shall just introduce
two
value and exponentiation.

Definition 4.3.1
absolute
:=

value

of

\\x\\

4.3.2

quantity

\342\200\224
\\x

sometimes

= 2.

y\\

denoted

such

is

called

the

\320\266
is

as follows.

defined

then

useful

particularly

If

\\x\\

:=

a rational

If

\317\207
is

\317\207
and

be
\321\203

d{x,y) :=

then

\317\207
and

\342\200\224
\\x

then

positive,
\\x\\

:=

numbers.

real

between

the

number,

\342\200\224\320\266.
\316\231\316\220\317\207
is zero,

distance

d(rr,y), thus

can construct,
ones: absolute

we

operations

(Distance). Let

Definition

d(3,5)

\317\207
is

to construct more

operations

value).

(Absolute

\316\231\316\257\317\207
is negative,
\317\207.

positive
In short, we have shown

an orderedfield.

Q form

now

the formulae
have a notion

into the

rationale

and zero.

rationale,

also

We

y_1.)

the

by

reciprocal

the

from

came

division

\317\207
\317\207

organized

negative

rationale

the

that

:=

and
(\342\200\224\321\203) x/y

on the rar

division

and

multiplication,

subtraction,

tionals. (Recall that subtractionand


more primitive notions of negationand

\\x\\

\320\277\320\270\321\21

2.3.7.)

Corollary

you

rationale

y\\.

For

0.

The

and
\321\203

instance,

is

99

exponentiation

4.3.3 (Basicpropertiesof absolutevalue

Proposition
Let

distance).

and

value

\316\271 Absolute
\316\262

(a) (Non-degeneracyof absolute value)


= 0 if and only if \317\207
is 0.
\\x\\
(b)

(Triangle
N + lvl-

(c)

We

the

In

inequalities

we have

particular,

In

particular,

\\

(Symmetry

(g)

(Triangle

Definition4.3.4

textbooks;

we

of

they

larger

positive

y\\

ify>

<

\\x\\.

have

We

\\xy\\

\\x\\

\\y\\.

have

>

d(x,y)

0.

Also,

d(y,x).
< d(x,y) + d(y,z).

d(x,z)

two

artificial definition:

Let

is \316\265-close
\321\203

\"close\"

how

measuring

to

\316\265
>

0,

iff
\317\207

we

and
have

be rational
d(y,x) < \316\265.

x,y

definition
is not standard in mathematics
it as \"scaffolding\" to construct the more

use

sequences) later on,


the
more advanced notions we will discard

limits

those

(and

of Cauchy

\316\265-close.

4.3.6.

Examples

are

only

\\x +

(\316\265-closeness).

of

have

we

once

notion

and

Also,

y.

a somewhat

This
will

notions

important

We

distance)

is useful for

that

say

4.3.5.

Remark

ifx

for

us make

numbers. We

have

4.3.1.

Exercise

Let

are.

We

0.

\\x\\.

of distance) d(x,y) =

Absolute value
numbers

only

>

\\x\\

< \317\207
<
\342\200\224\\x\\
\\x\\.

of distance)

inequality

See

Proof

is

x\\

d(x,y) = 0 if and

(f)

and

\342\200\224
=

(Non-degeneracy

(e)

\342\200\224y<x<yif

of absolute value)

(Multiplicativity

(d)

have

We

absolute value)

inequality for

have

and

numbers.

be rational

X)V)Z

not
than
\316\265.

numbers
0.99 and 1.01 are 0.1-close,
but
because d(0.99,1.01) = |0.99-1.01|= 0.02
2 and 2 are \316\265-close for every
The numbers

The

0.01 close,
0.01.

100

We do not botherdefining

of

a notion

and rationak

Integers

4-

when

\316\265-close

\316\265
is zero

or

then \317\207
and
are
is zero
\316\265-close
because if \316\265
\321\203
only
when
then \317\207
is negative
and
are
are equal, and when \316\265
never
\316\265\321\203
that
(In any event it is a long-standing tradition in analysis
small positive numbers.)
\316\264
should
denote
Greek letters \316\265,
only

negative,
they
close.

the

Some basic propertiesof \316\265-closeness


Let x, \321\203,
z, w

4.3.7.

Proposition

=
y, then
(a) If \317\207

if

\317\207
is \316\265-close

(b) Let
(c)

Let

\316\265
>

Let

are

also

Let

\316\265
>

Let

+
(\316\265

0.

and
\321\203

0.

0.

We

Proof.

If

\\a\\

then

<

If

0.

yw are

<

if

and

are
\321\203

then

z,

\316\265-close,

+
(\316\265

and

\316\266
and

and

6)-close,

are

\316\264-close,

\342\200\224
\317\207 \316\266
and

\342\200\224

\316\265-close,

also

are

they

\316\265'-close

for

both
\316\266
or

\316\265-close

to

< w
\316\266

w is

and

x,

between

then w is also

< y),

\316\265-close,

and

\316\266
is

then

non-zero,

<

\316\265-close,

and

\316\266
and

are

\316\264-close,

+ \316\264\\\317\207\\
+ \316\265\316\264)-close.
(\316\265\\\316\266\\

5
\316\265,

\316\266
and
\\b\\

are
\321\203

the most

Let
a :=

write

\316\266
+ b

\317\207
and

prove

\316\265.
Similarly,

w =

to

\316\265\\z\\-close.

If

and

only
we

\317\207
and

are

to Exercise 4.3.2.
\316\265-close.

x.

to

is \316\264-close
\321\203

y.

x.

0.

xz

\316\266
are

<
\321\203 w

(i.e.,

to

\316\264
>
\316\265,

then

are
\321\203

and
\321\203

If

\316\266

\316\265
>

Let

are
\321\203

+ w are

\317\207
and

If

xz andyz
(h)

=
\317\207

\316\264)-close.

\317\207
and

If

is \316\265-close
\321\203

and

y,

Conversely

we have

then

then

y,
to

0,

0.

\316\2651
> \316\265.

\316\265
>

Let

to

\317\207
is \316\265-close

If

\316\265
>

\316\265
>

every

\321\203
for

every

\321\203
for

numbers.

be rational

to

following.

+ \316\264)-close.
(\316\265

\316\265-close

(g)

0.

\317\207
and
+ \316\266

every

(/)

to

is \316\265-close
\317\207

If

\316\266
are

\316\264
>
\316\265,

then

(e)

0.

\316\264
>
\316\265,

\317\207
and

(d)

\317\207
is \316\265-close

the

are

>

difficult

\342\200\224
then
\321\203 x,

w are
\316\264.

one,

0, and

we have

5-close,

we leave

(h);

suppose

that

\317\207
and

and
+ \316\261
\321\203 \317\207

and we define b

:=w

(a)-(g)
are
\321\203

that
\342\200\224

z,

\316\271
3.

and w =

+ a

= x
Since \321\203

yw =

101

and exponentiation

value

Absolute

we have

\316\266
+ b,

(x + a)(z + b)

xz +

az +

xb + ab.

Thus

Since

\\a\\

<

<

\\az+bx+ab\\

lyw-a^l

and
\316\265

<

\\b\\

yw and

that

thus

\\bx\\

5, we

thus have

\342\200\224

<

xz\\

\\yw

and

\\az\\

\317\207
\316\266
are

\\ab\\

+ |b||sc|+

|a||s|

|a||b|.

+ \316\264\\\317\207\\
+ \316\265\316\264
\316\265\\\316\266\\

+ \316\264\\\317\207\\
+
(\316\265|\316\266|

s5)-close.

4.3.8. One should comparestatements


of this
(a)-(c)
with the reflexive, symmetric,and transitive
axioms
proposition
It
is
useful
to
think
of
the
notion
as
often
of
\"\316\265-close\"
of equality.
substitute
for that of equality in analysis.
an approximate
Remark

we

Now

define

recursively

exponents,extending

the

To raise

number.

a rational

be

\317\207
to

the

natural

power

Now suppose inductively that xn has beendefined


\317\207.
number n, then we define xn+l := xn \317\207
4.3.10

Proposition
rational

numbers,

have

(a)

We

(b)

We have

xn = 0 if and

Proof

We

See

have

Exercise

of exponentiation,

\317\207

1.

natural

some

I). Let x,y

be

be natural numbers.

only

0, then xn >

\\xn\\

define x\302\260:=

for

(xn)m = xnm, and {xy)n

= xn+m,

xnxm

(c) Ifx>y>
then xn>yn>

(d)

(Properties
and let n, m

Let

number).

0, we

number
2.3.11.

in Definition

definition

previous

4.3.9 (Exponentiationto a

Definition

for natural

exponentiation

yn

if
>

=
\317\207
0.

xnyn.

0.
If

>
\317\207
> \321\203

and

\316\267
>

0,

0.
=

\\x\\n.

4.3.3.

4- Integers and

102
we

Now

define

Definition 4.3.11
define

we
\342\200\224n,

integer exponents.

a negative

to

(Exponentiation

Then

rational
number.
x~n := l/xn.

a non-zero

\320\266
be

for negative

exponentiation

for

be

\320\277\320\276\320\277-zero
rational

(a)

We

If

(b)

>
\317\207

>
\321\203

(xn)m = xnm,

(c)

>
J/\320\266,\321\203

(d)

We have

0,

Exercise4.3.1.

Prove

can

See Exercise

Proof

be proven

or zero,

letn^m

>

yn

and

xn = yn,

the

following

Let

if

and (xy)n =

is
\316\267

xnyn.

then

=
\317\207

y.

4.3.4.

4.3.3.

Proposition

several parts of

the

(Hint:

such as when
can be

proposition

while all

of these claims

\317\207
is positive,

proven

negative,

Exercise4.3.2.

the

Exercise 4.3.3. Prove

Proposition

4.3.10.

Proposition

4.3.12.

here.

Instead,

Exercise 4.3.5. Prove


use induction.)

remaining

use Proposition
that

in Proposition

claims

(Hint:
(Hint:

such

without

division into cases. For instanceonecan use earlierparts


to prove later ones.)
proposition

Exercise 4.3.4. Prove

0 <

and

positive,

tedious

suitable

x,j/

\\x\\n.

by dividing into cases,

Prove

have
or

negative,

negative.

\316\267
0,
\317\206

\\xn\\

xn >

now

II).

= xn+m,
then

0,

integer

be integers.

numbers,

\317\207\316\267
<\321\203\320\277
^\320\277 ^5

We

exponentiation,

and

xnxm

have

Let

number).
negative

any

=
\317\207
\317\207
\317\207
Thus for instance
x~3 = 1/\321\205\320\263
x).
l/(x
\316\267
xn defined for any integer
is positive,
n, whether
zero. Exponentiation with integer exponentshas
properties
(which supercede Proposition 4.3.10):

Proposition 4.3.12 (Propertiesof

rationals

of

the

4.3.7.

use induction.)
induction

is not

4.3.10.)

2N > N for all

positive integers N.

(Hint:

Gaps in the

of

right

if x
\321\203

not

have

we

we

that

Imagine

rigourous

thus

line. Now we work

on the

arranged with respect to the

be

rational

number.

n+1.

In fact,

< \317\207
<
\316\267

one

is only

a
as

sometimes

=
\316\267

to

referred

are

rationals

that

which is

is unique (i.e., for each \317\207


there
+ l). In particular, there exists

\316\235
>

is no

there

\317\207

(i.e.,

larger than all the

<
The integer \316\267
which
\316\267
for
as the integer part of \317\207
is
and

4.4.2.

Remark

< \317\207
\316\267
<n

such

number

a rational

this integer

which

\316\267
for

number

natural

the

how

out

integers.

of integers by rationals). Let


(Interspersing
Then there exists an integer \316\267
such
that

4.4.1

proposition

\317\207

arranging

but

the

arranged

also

to the
line,
arrangement,since
defined the concept of a line,
this
discussion
to motivate the more
propositions
rationals
we have the integers, which are

yet

Inside

below.)

the rationals on a
arrange
> y- (This is a non-rigourous

intended

is only

numbers

in the rational

Gaps

4.4

103

numbers

rational

natural

such

thing

numbers).

\317\207
< \316\267
+

1 is

denoted

sometimes

[x\\.

See Exercise

Proof.

4.4.1.

Also, betweenevery

there

numbers

rational

two

least one

is at

additionalrational:

Proposition4.4.3
two

are
\321\203

rational

\316\266
such

We set

Proof.

such that

rationals
that

to

\321\203/2+

obtain

both

:=
\316\266

y/2,
x/2

sides

i.e.,
+ x/2

\317\207
then
< \321\203,

by rationals).
If \317\207
exists a third

there

\317\207
< \316\266
<y.

+ \321\203)/2.
(\317\207

positive, we have from


\321\203/2

rationals

of

(Interspersing

and

\316\266
<

<

4.2.9

Proposition

using

\317\207
<

Since

Proposition
instead

\321\203.If

we

\321\203/2+

\320\266/2,

i.e.,

that

and
\321\203,
\317\207/2<

1/2

1//2

If we
\321\203/2.

4.2.9 we obtain x/2 + \321\203/2<


add x/2 to both sides we
\317\207
<

\316\266.
Thus

\317\207
<

\316\266
as
< \321\203

desired.

the rationals having


there are still
incomplete;

Despite
still

is
add

this densenessproperty,

an

infinite

number

they

of

\"gaps\"

are

or

4- Integers and rationak

104
the

between

\"holes\"

does ensure

holes

these

that

this denseness
although
are in some sense infinitely

rationale,

property
small.

show that the rationalnumbers


instance,
contain
any square root of two.
we will now

For

Proposition 4.4.4.
which

for

Proof.

x2 =

We

Exercise

not

number

any rational

exist

not

does

There

do

\317\207

2.

4.4.3.

of a

a sketch

give

only

Suppose

of

sake

for

proof; the gaps will

in

we had a

that

contradiction

filled

be

zero.
x2 = 2. Clearly \317\207
is not
for
which
rational number \317\207
We
were
then we could
is positive,
for if \317\207
negative
may assume that \317\207
=
=
Thus
\317\207
\317\207
\342\200\224x
x2
replace
by
just
(since
(\342\200\224x)2).
p/q for some
=
which
we
can
so
2,
rearrange as
positive integers p, g,
(p/q)2
= 2k for some
= 2q2. Define a natural number \317\201
to be even if \317\201
p2
= 2fc + 1 for some natural number
natural number fc, and odd if \317\201
not
natural
number
\320\272.Every
is either even or odd, but
both
which
If
is
then
odd
is
also
contradicts
\317\201
odd,
p2
(why?),
(why?).
p2

2q2.

Thus

is
\317\201

is positive,
\320\272
must
Since \317\201
=
=
we
obtain
4fc2
2q2
p2

To summarize,

be positive.

also

with

integers
(why?).

If we

that

one solution (p,q)


the
same
(p'iQf) t\302\260
then

we

sequence

can

p' := q

rewrite

repeat

(p77,?77)?

to

Inserting

\317\201

2k into

= 2k2.
pair (p, q) of positive integers

p2
2q2, and ended up with
such that q2 = 2k2. Sincep2

such

number k.

natural

2q2, so that q2

started

we

some

for
2k
\317\201\342\200\224

i.e.,

even,

the

and q' := fc,


p2

equation

equation

{pm,q\"'),

etc.

of

(\317\201,
\320\272)

positive

2<?2, we have q
we

\342\200\224

has a

which

this procedure

a pair

2q2

<

\317\201

pass from

thus

can

to

a new

solution

smaller value of p. But

again and again, obtaining

of solutions to

p2

2q2,

each

a smaller value
than
the
and each one
of \317\201
previous,
of
B
ut
this
contradicts
the
consisting positiveintegers.
principle
of infinite descent (see Exercise 4.4.2). This contradiction
shows
we could
that
not have had a rational \317\207
which
for
x2 = 2.
D
one with

On the other hand,


arbitrarily closeto a square

we

can
root

get

rational

of 2:

numbers

which are

proposition 4.4.5. For


a non-negativerational
is no

that

2<

+ \316\265)2.
(\317\207

Suppose for sake of contradiction


x2 <
rational number \317\207
for
which

rational.

0 be

\316\265
>

Let
there

x2 <

that

\317\207
such

there exists

> 0,

numbers

rational

every

number

Proof.

105

numbers

rational

in the

Gaps

L
\316\271

non-negative

This
means
that
whenever
\317\207
and
is non-negative
2 < (x + \316\265)2\302\267
< 2 (note that (\317\207
cannot
+ \316\265)2
+ \316\265)2
(\317\207
x2 < 2, we must alsohave
Since
we
\316\2652
02
thus
have
<
< 2,
2,
2, by Proposition
4.4.4).
equal
which then implies (2\316\265)2 < 2, and indeed a simple inductionshows
is non\316\267\316\265
< 2 for every natural number n. (Notethat
that
(\316\267\316\265)2
\316\267
natural
number
for
But, by Proposition
every
why?)
negative

4.4.1 we can find


that

claimthat
the

gives

an

If3

4.4.6.

Proposition
does

to

that

\316\267
>

2/\316\265,

> 2,

> 4
(\316\267\316\265)2

which

implies

contradicting the

natural numbers n. This contradiction


D

x2 = 1.999396and

wish

that

proof.

Example

not

\316\267
such

integer

which

implies
< 2 for all
(\316\267\316\265)2

> 2,
\316\267\316\265

actually
\320\273/2.For

=
\316\265

we
0.001,
= 2.002225.
+
(\317\207 \316\265)2

since

1.414,

a member,

as
\320\273/2

the

instance,

=
\317\207

that, while the set Q of rationale


we can get as closeas we

4.4.5 indicates
have

can take

sequence

of rationale

1.4,1.41,1.414,1.4142,1.41421,...

seemto get closer

closer

and

to

\320\273/2,as

their

indicate:

squares

1.96,1.9881,1.99396,1.99996164,1.9999899241,...

Thus it seems that


of

\"limit\"

the real

a sequence

we

can

create

of rationals.

This is how

numbers in the next chapter. (There

using somethingcalled\"Dedekind
pursue here. One canalsoproceedusing
do so,

expansions,

but

root of 2

a square

there

are

some

sticky issues

cuts\",
infinite

by taking a
construct

we

shall

is

another

which

way

we will

to

not

decimal

when doing so, e.g.,one

will use the decimal system for defining


3We
decimals, for
terminating
instance 1.414 is defined to equal
number 1414/1000. We defer
the rational
the
formal
on the decimal system to an Appendix
discussion
(\302\247B).

106

has to make 0.999... equal


despite

other

approaches;

Prove

4.4.1.

Exercise

4.4.2.
Exercise
if we

and

1.000...,

have an >

Proposition

4.4.1. (Hint:

the principle of infinite


descent
have a sequenceof natural
numbers
assume for sake of contradiction
(Hint:
sequence

natural

of

the an are

(b) Does the

that it is not possible to


which is in infinite
descent.
that
you can find a

which is in
know

to show in fact that


a contradiction.)

an >

Since

descent.

infinite

that

an > 0 for
\320\272
for

all

all n.

N
\320\272
\302\243

and

all
Now

all

obtain

principleof

work if the sequence \316\261\317\212,


descent
\320\260\320\263,
number
integer values insteadof natural
What about if it is allowed to take positive rational values
of natural
numbers? Explain.

is allowed
\320\260\320\267,...

values?

numbers

natural numbers, you

induction

and
\316\267
\302\243
\316\235,

infinite

to take

Exercise 4.4.3. Fill in the gaps marked


Proposition4.4.4.

use Proposition2.3.9.)

definition:

(a) Prove

instead

approach,

integers,

numbers,

use

this

.\302\267
a sequence
of numbers
ao, ai, \320\260\320\263,..
or reals) is saidto be in infinite
rationale,
descent
numbers \316\267
for all natural
ao > a\\ > a<i > ...).
\316\261\316\267+\316\271
(i.e.,
A

(natural

to

familiar, is actually more complicatedthan


see the Appendix \302\247B.)

most

the

being

and rationale

Integers

4-

(why?)

in the

proof of

Chapter

numbers

real

The

to date, we have rigourously constructed


our
review
progress
the natural
number
number system
three
systems:
the
the natural
rationale
We
defined
Q1.
N, the integersZ,
and postulated
Peano
that such a
numbers usingthe
axioms,
this
is
the
natural
number system existed;
plausible,
to the very intuitive
and fundamental notionof
correspond
that
number
counting.
system one could then
Using
To

fundamental

and

five

since

numbers

sequential

define

recursively

obeyed

the

addition

usual

laws of

and

multiplication,

algebra.

We

then

and verify
constructed

that

they

the integers

formal2 differencesof the natural numbers,a\342\200\224b. We


the rationale by taking formal quotients of the
constructed
then
we need
to exclude division by zero in
integers,a//b, although
laws
of
are
of course
the
reasonable.
free
order to
keep
algebra
(You
one
where
to design your own numbersystem,possiblyincluding
will
to
zero
is
have
division
but
by
permitted;
give up one
you
by taking

and \"real\"
\"quotient\",
Q, and R stand for \"natural\",
for \"Zahlen\", the German word
for number.
There is
also the complex numbers
for
stands
C, which
\"complex\".
obviously
of our construction
the form of\"; at the beginning
2Formal
means \"having
the expression a\342\200\224b
did
not actually
the
mean the difference a \342\200\224
6, since
was
It only had the form of a difference.
Later
on
symbol \342\200\224
meaningless.
we defined subtraction and verified that the formal difference was equal to the
actual difference,so this eventually
a non-issue, and our symbol
for
became
formal differencing was discarded. Somewhat confusingly,
use of the term
this
formal\" is unrelated to the notions
of a formal argument and an informal
\321\205\320\242\320\254\320\265
N,
symbols

respectively.

argument.

\316\226
stands

5.

108

or more of the

things, and

you

do any

to

which

field

will

from

axioms

probably

much

among other

useful number

system

a lot

of

\321\211

real-world problems.)

The rational system is already sufficient


mathematics

4.2.4,

Proposition

get a less

The real numbers

of high

school algebra,

to
for

do

instance,

works

just

there
is a
only knows about the rationale.However,
fundamental
area of mathematics wherethe rationalnumber
of lengths,
system does
not suffice - that of geometry (the study
areas,
sides
both
equal to
etc.). For instance, a right-angledtrianglewith
1 gives a hypotenuseof \320\273/2,which
is an irrational
number, i.e., not
see
4.4.4.
a rational
number;
Proposition
Things get even worse
when
one starts to deal with the sub-field of geometryknown
as
when
such
as
\317\200
or
one
sees
numbers
which
trigonometry,
cos(l),
more\"
irrational
than
turn out to be in some sense \"even
y/2.
numbers
as
transcendental
but
are
known
to
numbers,
(These
this
discuss
further
would be far beyond the scopeofthis text.) Thus,
in order
to have a number system which can adequatelydescribe
- or even
as simple as measuringlengths
on
something
geometry
one needs to replacethe rationalnumber,
with
a line
the
system
and
real number system. Since differential
calculus
is also
integral
of slopes
of tangents,
or
intimately tied up with geometry- think
areas under a curve
also requires the real number system
calculus

fine if one

in

order

to function

properly.

the reals from the rar


However, a rigourousway to construct
tionals
turns
out to be somewhat difficult
a bit more
requiring
than
was
from
the
what
needed
to
naturals
to the
machinery
pass
In those two
to the rationals.
integers,or the integers
the
task
was
to
introduce
more
one
constructions,
algebraicoperationto
the
number
one
can
system
by
e.g.,
get integersfrom naturals
and get the rationals from
the
introducing
subtraction,
integers
the
division. But to get the realsfrom
rationals
by introducing
is to pass from a \"discrete\"system to a \"continuous\"
and
one,
- that
the introduction
of
of a somewhatdifferent
notion
requires
a limit The limit is a concept
which
on one level is quite
but
to pin down rigourously
turns out to be quite
difficult.
intuitive,
such
as
Euler
and
had
diffimathematicians
Newton
great
(Even

109

sequences

\316\266
\316\271,Cauchy

concept. It was only


and
mathematicians such as Cauchy
limits
rigourously.)
deal with
culty with this

In Section4.4

we

now

the

explored

these gaps

fill in

shall

we

century that
out how to
figured

nineteenth

the

in

Dedekind

rational numbers;

in the

\"gaps\"

using limits to

real

the

create

The
system
up being a lot
but will have some new operationsnumbers,
rational
which can then be used to
limits
of supremum,
that
else
that
calculus
needs.
thence to everything

real

numbers.

will end

number

the

like
notably

and

define

The
the

we

procedure

limit

it

However,

complicated.

useful

procedure,

another;

see

5.1

Cauchy

here of

give

obtaining the real numbersas

of rational numbers may seem rather


is in fact an instance of a very
general

of sequences

that of

completingone metric

space

sequences

Let
Definition 5.1.1 (Sequences).

of

greater

a sequence

numbers

rational

>
: \316\267
to
\316\226
Q,
\317\204\316\267}

than

or equal

to m,
of

(an)\302\243Lm

this

be

notion

the
function from
set
which assigns to eachinteger

numbers

sequence

(n2)\302\243L0

sequences

make

sequence
natural

More

an.

is a

{n

G
\316\267

informally,

collection of rationale
collection

the

(3)\302\243L0

other

A sequence

an integer.

a rationalnumber

rational

concept of
formally, let us

on the

rely

is any

a mapping

i.e.,

shall

is
Example 5.1.2. The
9,... of natural numbers;the sequence
3, 3,... of natural numbers.These
from 0, but we can of course
sequences
any

and

form

12.4.8.

Exercise

Our construction of the real numbers


a Cauchy sequence.Beforewe define
the concept
of a sequence.
define
first

(on)\302\243Lm

to

is

are

the

indexed

starting

0, 1,

collection
starting
from 1

4,
3,

or

the
the sequence (\320\260\320\277)^\320\267
denotes
so (n2)^=3 is the collection
9,16,25,...of

for instance,

number;

as,...,

a^
\320\260\320\267,
numbers.

We want to define the real numbersas the


limits
of rational numbers. To do so, we have
to distinguish

of sequences

which

se-

5. The realnumbers

110
of

quences

and which ones are not. p0r

are convergent

rationale

the

instance,

sequence

1.4,1.41,1.414,1.4142,1.41421,...

looks like it

is trying

to

as does

to something,

converge

0.1,0.01,0.001,0.0001,...
such
while other sequences

as

1,2,4,8,16,...

or

1,0,1,0,1,...

do not. To do this we
are
\321\203

5.1.3

Definition

to be

said

is

for

\316\265-close

each

iff

\316\265-steady

natural

is

Let

(\316\265-steadiness).

every

a2,...

ai,

ao,

if

\316\265-close

of

definition

the

use

from Definition 4.3.4 that


= \\x \342\200\224
< \316\265.
y\\
d(x, y)

Recall

earlier.

\316\265-steady

pair

\316\265
>

\316\261^,

j, fc. In

iff

<
\316\261&)

d(aj,

0.

of
\316\261&

number

\316\265-closeness

other words, the sequence

for
\316\265

all

j, k.

5.1.4. This definition is not standardin


will not need it outside of this section;
similarly

of \"eventual
for
a

notion

similar

\316\265
for

all

a sequence
j,

We

have

the
for

defined

literature;
the

concept

\316\265-steadiness

at 0, but clearly we can make


other
for sequences whose indices start from
any
index

whose

sequences

number:

below.

\316\265-steadiness\"

x)

A sequence
(\316\261\316\267)^=\316\277
elements
is
sequence

Remark
we

defined

numbers

rational

two

starts

is \316\265-steady
\316\261^\316\275,
\316\261\317\204\316\275+\316\271?...

if one

has d(a.j,

<

\316\261&)

> N.
\320\272

The sequence 1,0,1,0,1,... is 1-steady,but is


is 0.1sequence 0.1,0.01,0.001,0.0001,...
1/2-steady.
but
The
is
not
steady,
0.01-steady
sequence 1, 2, 4, 8,
(why?).
The
16,...isnot \316\265-steady for any \316\265
sequence
2,2,2,2,.. .
(why?).
Example

5.1.5.

The

not

is \316\265-steady

for

every

\316\265
>

0.

\320\2721.

of \316\265-steadiness

The notion

but does not

is simple,

a sequence

behavior
limiting
initial
members of

the

to

sensitive

of

the

capture

really

111

sequences

Cauchy

of a sequence, becauseit is too


the sequence. For instance,the

sequence

10,0,0,0,0,0,...

members

initial

Definition 5.1.6
(an)\302\243Lo

sequence

t\302\260
be

eventually

\302\267
\302\267
\302\267
is \316\265-steady
for
\320\260#,\320\260\320\273\320\275-\321\212\320\260\320\273\320\263+2?

Let

the

sequence

N >

number

In otherwords, the sequenceao,ai, a2,... is eventually


<
for
iff there exists an N > 0 suchthat d(a,j?,
\316\261&) \316\265

the

0.

\316\265
>

natural

need

about

care

iff

\316\265-steady

some

despite
\316\265,

to zero. So we

that doesnot

\316\265-steadiness).

(Eventual

said

is

value of

smaller

any

almost immediately

notion of steadiness
of a sequence.

robust

a more

for

\316\265-steady

converging

sequence

the

not

but is

is 10-steady,

0.

\316\265-steady

N.

j, k>

all

defined
by an := 1/n,
Example 5.1.7. The sequence\316\261\316\271,\316\2612,...
is
not
but
is
the
sequence1,1/2,1/3,1/4,...)
0.1-steady,
(i.e.,
the
because
sequence \316\261\316\271\316\277>\316\261\316\271\316\271>\316\26
eventually
0.1-steady,
(i-e.,
is
The
10,0,0,0,0,...
sequence
0.1-steady.
1/10,1/11,1/12,...)
for
than
less
is not
10, but it is eventually \316\265-steady
\316\265-steady
any \316\265
for

every

\316\265
>

Now we
a sequence

for

0 (why?).

can

to

of rationals

5.1.8

Definition

the

define

finally

correct

notion

to

converge.

\"want\"

(Cauchy sequences). A sequence


iff
to be a Cauchy sequence

is said

rational

numbers

\316\265
> 0,

the sequence

is

(an)\342\204\242=0

eventually

5.1.9.

a positive

to
rational; we cannot take \316\265

At

the

present,

real number, becausethe


constructed.

shall
to

be

However,

see

real

that

once

the above

instead

real

we

is
\316\265

numbers

do construct

an

be
have

all

words,
\316\265
>

every

j,

0,

> N.
\320\272

to

restricted

yet

be

positive
been

the real numbers, we

definition will not change if

of rational

other

for

arbitrary
not

rational

every
In

\316\265-steady.

parameter

of

(an)\302\243L0

for

the sequence \316\261\316\277,


is a Cauchy
\316\261<\316\271,...
\316\261\317\207,
sequence iff
>
there exists an N
for
0 suchthat d(aj,\316\261&)< \316\265
Remark

means

of what it

(Proposition 6.1.4). In

we
other

require
words,

\316\265

The real numbers

5.

112

we will eventually prove that

is

sequence

eventually

\316\265-steady

f0r

only if it is eventually \316\265-steady f0r


real \316\265
see
6.1.4. This rather subtle distino.
> 0;
Proposition
not to be
and
turns
out
\316\265
a real \316\265
between
a rational
very
rational

every
every
tion

the

in

important

and

0 if

\316\265
>

and the reader is advised

long run,

attention as to what

too much

pay

be.

should
\316\265

(Informal) Consider the sequence

5.1.10.

Example

of number

type

to

not

1.4,1.41,1.414,1.4142,...
This

earlier.

mentioned

the remaining sequence

1.4, then

element

first

the

discards

1-steady. If one

is already

sequence

1.41,1.414,1.4142,...
is

now

this

that

is

is in fact

sequence

not

precisely
example

to

ajv+i,...

a&) <

\\l/j

\316\261\317\212,
\316\2612,
\320\260\320\267,...
defined

for
\316\265

> 1

us

see

every

j,

>
\320\272

\342\200\224

l/k\\

<

know

let

\316\265
for

\316\265
>

the

0,
0

:=

by an

sequence.

Cauchy

\316\265
>

every

Let

> \316\235,we
\320\272

l/fc| < 1/N.

for

So

number

\316\265-steady.

\\l/j

sequence

\316\265-steady.

find

that

Now since j,

The

to show that

eventually
is

d(a.j,

seems

this

However,

sequence.

sequence 1,1/2,1/3,...) is a

We have

now have

0, which

reasons, for instance we have not


rigourous
defined
what
this sequence 1.4,1.41,1.414,...
really is.
of a rigourous
treatment follows next.

1/n (i.e., the

\316\2612,... is

\316\265
>

every

for several

Proposition 5.1.11.
Proof.

for

\316\265-steady

that this is a Cauchy

to suggest
discussion

Continuing

0.01-steady.

eventually

An

1.414,1.4142,...;

sequence

steady

original sequence was


element gives the 0.01thus the original sequence was
in this way it seems plausible

Discarding

0.1-steady.

eventually

that the
the next

means

which

0.1-steady,

be

sequence

arbitrary.

\316\261\317\212,

We

such that the sequence\316\261#,


this means. This means

what
iV,

i.e.
j, k>

every

that

So in order to force

N.

1/j, 1/fc < 1/N, so that


\\l/j
l/fc| to be lessthan

0 <

0r equal

to

it
\316\265,

to be lessthan

for 1/N
that

than

is less

1/N

can

this

But
l/\316\265.

\316\265.
So

or
\316\265,

be done

Proposition 4.4.1.

thanks to

can

you

from first

verifying

see,

a Cauchy

simple as 1/n.

The

principles (i.e., without

etc.) that a

machinery of limits,
sequence
requires some effort,
of the

any

using

is choose an N such
that N is greater than

words

As

be sufficient

would

to do

need

we
\320\264\320\224

in other

113

sequences

Cauchy

51.

for

even

sequence

is

a sequence

as

an N can be
in reverse,
has
to
think
beginners
on
N
conditions
would
suffice
to
the sequence
force
what
out
working
to be \316\265-steady,
and
an N which
then
finding
aN)0>N+uaN+2,...
conditions.
will
some
limit laws
Later
we
those
develop
obeys
us to determine
when a sequence is Cauchy
more
allow
which
part

about

selecting

- one

for

difficult
particularly

easily.

the notion of a Cauchy


basic notion, that of a boundedsequence.
now relate

We

Definition5.1.12
A

finite

<

all 1 < i
<

\\a,i\\

is bounded
\316\261\317\212,
\316\2612,..., \316\261\316\267

sequence

\317\204\316\271.
An
infinite

all

\316\234for

bounded

by

>
\320\263

1.

some

\316\234for

sequence

The finite
is bounded by
equal to 4). But the infinite

this case, it
or

unbounded.
1,

sequence

(an)^=1

sequence
4,

or

1,

indeed

another

0 be

rational.

<
|\316\261^|

\316\234iff

by

A sequence
is said
rational
\316\234>
0.

5.1.13.

Example

>
\316\234

Let

sequences).

(Bounded

to

sequence

is bounded

to be

\316\234for
\316\234iff

by

iff

bounded

is bounded
\342\200\2244
\342\200\2242,3,

by any
1,

sequence

is

it

(in
than

\316\234greater

\342\200\2242,3,
\342\200\2244,5,
\342\200\2246,... is

(Can you prove this? Use Proposition4.4.1.)The


\342\200\2241,1,
\342\200\2241,...is bounded
(e.g., by 1), but is not a Cauchy

sequence.

Lemma
sequence

Proof.
sequence

clearly

that

5.1.14

(Finite

\316\261\317\212,
\316\2612,...,

is clearly
\316\261\317\212

we

have

bounded).

this by inductionon

We prove

we have

sequences are

Every

finite

is bounded.
\316\261\316\267

bounded,

<
|\316\261^|

\316\234for

n.

=
\316\267

When

choose

for if we
1 <
all

<
\320\263

already proved the lemma

for

some

\316\234:=

\320\277.

Now

>
\316\267

1 the
then

|\316\261\316\271|

suppose
1; we

now

5. The real number^

114

\320\260\320\263,...,
1, i.e., we prove every sequence \316\261\317\212,
\316\261\316\267+1
^
know
we
that
the
induction
\320\273
a\\,
hypothesis
By
\316\2612,...,
it must be bounded
by some \316\234> 0; in particular,

it for

prove

bounded.
is

bounded

by

\316\234+

\316\267
+

On

|\316\270\316\267+\316\271|.

other

the

hand,

Thus
\320\260\320\263,...,
\316\261\316\267,
anH_j|\316\261\316\267+\316\271|. \316\261\317\212,

\316\234
+

and is

hence bounded. This closes


that

is bounded,
say

anything

not;

infinity

Proof.

bounded.

See Exercise

5.1.1.

Prove

eventually

1-steady,

sequence.

1-steady

there is nothing
number

5.2
Consider

sequences

(Cauchy

is
(an)\342\204\242=1

5.1.1.

Exercise

by

\316\234
+

|an+i|,
q

finite
while this argument shows that every
sequence
how
the
finite
no matter
sequence
is, it does not
long
about
whether
an infinite sequence is boundedor
is not a natural number. However,we have

Lemma 5.1.15
sequence

also boundedby

induction.

the

Note

an+i is

is bounded

special

would have

are

bounded).

Every

Cauchy

Lemma 5.1.15. (Hint: use the fact


that
an is
and thus can be split into a finite sequence and
Then use Lemma 5.1.14for the finite part. Note
about
the number 1 used here;any other
positive

sufficed.)

Equivalent Cauchy sequences


the

two

Cauchy

sequences

of rational

numbers:

1.4,1.41,1.414,1.4142,1.41421,...
and

1.5,1.42,1.415,1.4143,1.41422,...
seem to be convergingto the
=
same
the square root y/2
1.41421... (though
this
number,
is not
statement
because
we
not
have
defined
real
yet rigourous
numbers
If
to
we
are
define
the
real
numbers
from
the rationals
yet).
as limits of Cauchy sequences,we have to know when two Cauchy
sequences of rationals give, the same limit, without first defining
Informally,

both

of these

sequences

5 2a real

number (sincethat
set

similar

in the

is

this we

To do

circular).

used to define a Cauchy

use a

sequence

first place.

to

words,

We

an

is

\316\265-close

the sequence

&0)bi,b2,\302\267\302\267\302\267|On

<

bn\\

5.2.2. The

Example

0.

two

to

=
\316\267

all

^e

(M\302\243Lo

sequence (a>n)\342\204\242=o
each

for

bn

ao, ai, a2,... is


\316\265
for

ancl

(an)\302\243L0

say that the

\316\265
>

iff
(bn)\342\204\242=0

iff

Let

sequences).

(\316\265-close

and let

sequences,

\316\265-close

other

be

would

to those

of definitions

Definition 5.2.1
two

115

sequences

Cauchy

Equivalent

\316\267
N.
\342\202\254

the

to

\316\265-close

In

sequence

0,1,2,....

sequences

1,-1,1,-1,1,...

and

1.1,-1.1,1.1,-1.1,1.1,...
are 0.1-close to eachother.
are 0.1-steady).

Definition 5.2.3
(bn)%Lo ke
(an)\302\243L0

IS

\316\265-close

(Eventually

two sequences, and let


\316\265-close

eventually

to

words,
\316\261\316\277,
\316\261\317\212,
\316\2612,...
exists an N > 0

is

\\an

\342\200\224 <

bn\\

are
(bn)\342\204\242=N
to

\316\265
for

5.2.4. Again, the notations for

Remark
eventually

\316\265-close

and we will

not

use

sequences

them

Example 5.2.5. The

two

are

outside

not

exists

there

of this

n>

\316\265-close

standard

sequence
an N

\316\265-close.

bo, bi,
all

and

(an)^=Q

that the

We say

\316\265-close

eventually

suchthat

0.

iff
(bn)\342\204\242^

and
suchthat the sequences
(an)\342\204\242=N

Let

sequences).

\316\265
>

of them

neither

that

however

(Note

In

iff
\320\254\320\263,...

> 0

other
there

N.
sequences

in the

and

literature,

section.

sequences

1.1,1.01,1.001,1.0001,...

and

0.9,0.99,0.999,0.9999,...

are not
are not

0.1-close (becausethe first elements


0.1-close to eachother).However,

of
the

both

sequences

sequences
are still

5. The real numbers

116

the

in

5.2.6

Definition

and
sequences

(an)\342\204\242=0

\316\265
>

>
\316\267

\316\265
>

> 0 suchthat

an N

other

for

iff

equivalent

the
In

\316\265-close.

eventually

(an)^

0,

every

\342\200\224 <

bn\\

\\an

\316\265
f or

N.

5.2.7. As with Definition 5.1.8, the


restricted to be a positiverational,rather

Remark

currently
real. However,

we shall eventually

whether \316\265
ranges

the

over

see

than

0 is

a positive

no difference

it makes

that

positive reals; see

rationale or

positive

\316\265
>

quantity

6.1.10.

Exercise

to be

5.2.5 appear

Example

that the

it seems

5.2.6

Definition

Prom

in

are

\302\267
\302\267
are
\320\254\321\212
&2> \302\267

bo>

exists

there

0,

and

rational

f\302\260r
each

(bn)\302\243Lo

\316\261\317\212,
\316\2612,...
\316\261\316\277,

rational

iff

eQuivalent
and

Two sequences

sequences).

(Equivalent

are

(bn)\302\243L0

words,
all

sequences are eventually


0.01-close
the third element onwards),and so forth.
the two

that

from

starting

(by

these

sequence,

shows

argument

start from the secondelements


sequences are 0.1-close. A similar

if we

because

0.1-close,

eventually
onwards

two

given

sequences

now

We

equivalent.

prove

this

rigourously.

5.2.8. Let

Proposition
an = 1 + 10~n

and

and

(an)\302\243Li

bn = 1

\342\200\224

the

Then

10~n.

be the

(bn)^=1

sequences

sequences
an, bn are

equivalent

that

= 0.9999...;

1.0000...

Proof.

\316\265
>

0.

need

We

\316\265
>

eventually

to find an

N > 0 such

other

words,

\342\200\224 <

\316\265
for

0,

the
each

to

\316\265-close

in

\316\265-close;

for every

that

prove

are
(bn)\342\204\242=1

are

(bn)^=N

to

need

We

and

(an)\302\243Li

fix an

This Proposition, in decimal notation, asserts


see Proposition B.2.3.

5.2.9.

Remark

two sequences
other.
So we

that

to

we need

an

such that
{an

bn\\

all

n>

N.

However, we have
\\an

bn\\

|(1 +

10\"n)

- (1 -

10\"n)|
=

and

(an)^!^
find

\317\207
10\"\316\267.

> 0

The

5 S.

10~n

Since

>

m
whenever

is a

2 x 10_n

have

is easily

< 2
-

bn| <

to obtain

in order

10_N for

\342\200\224

<

bn\\

\\an

and

>
\316\267

\316\235,we

we have

Thus

2 x

< 10~n

10~m

by induction),

proven

\317\207
10_N.

\320\232

Thus

function of \316\267
(i.e.,

decreasing

this
\320\251

117

real numbers

of the

construction

>
\316\267

all

>
\316\267

all

\316\265
for

N.

AT, it

will be

2 \317\207
10~\316\235 <
is easy
\316\265.This
to
sufficient to chooseN so that
logarithms
do using logarithms, but we have not yet developed
a
cruder
is
will
method.
we
we
use
observe
10^
so
First,
yet,
>
than
N
for
N
1
Thus
Exercise
any
(see
always greater
4.3.5).

< 2/N.
2 \317\207
Thus to get 2 \317\207
10\"\316\235 < \316\265,
10\"N
N so that 2/N < \316\265,
or
that
equivalently
4.4.1 we can always choosesuch
an
jV > 2/\316\265. But
by Proposition
the
claim
follows.
D
and
jV,

and so

<

1/N,
suffice

10\"N

it will

to choose

Exercise5.2.1. Show

that

of

then

rationals,
ls a Cauchy
if (&\317\200)\317\204\302\243=\316\271

sequences

Exercise 5.2.2.

The

and
(\316\261\316\267)^=\316\271
(bn)^=i

is a
(\316\261\316\267)^=\316\271Cauchy

are

equivalent
sequence if and

only

sequence.

Let \316\265
> 0.

Show that if

is bounded
(\316\261\316\267)^=\316\271

\316\265-close, then
eventually

5.3

if

and
are
(\320\254\320\277)^=1
(\320\260\320\277)\342\204\242=1

if and

only if

of the real

construction

^\316\267)^=\316\271 bounded.

is

numbers

the real numbers.We shall


to the formalnotations\342\200\224
similar
symbol LIM,
this will
and
// defined earlier; as the notation suggests,
of
which
the familiar
at
lim,
eventuallymatch
point the formal
operation
We

now

are

introduce

a new

limit

symbol

Definition

to construct

ready
formal

can

be discarded.

5.3.1 (Real numbers). A

number

real

be an object of the form LIMn_>ooan,


where
of rational numbers. Two real numbers

sequence

said
are
LIMn-\321\216\320\276\320\254\320\277
equivalent

R.

Cauchy

to be

sequences.

equal iff
The

(an)\302\243Li

set of

and

to

is defined
is

(an)\342\204\242=1

a Cauchy

LIMn_>ooan

(bn)^)=1

and

are

all real numbersis denoted

5.

118

numbers

(Informal) Let oi, 02,03,... denotethe

5.3.2.

Example

real

The

sequence

1.4,1.41,1.414,1.4142,1.41421,...

let bi,

and

\302\267
\302\267
denote
62,63, \302\267

the

sequence

1.5,1.42,1.415,1.4143,1.41422,...

then

real

number,

because
\342\200\224\321\216\320\276\320\254\321\202\321\214
(a^^-i

LIMn

and is

is a
\316\231\316\231\316\234\316\267-,\316\277\316\277\316\237\316\267

and

(b^^-i

the samerealnumber
as
are equivalent Cauchy

sequences:

= LIMn
\320\253\320\234\320\277\342\200\224\321\216\320\276\320\260\320\277

We will refer
(an)\302\243Li-

to

formal

we will

define a

that
the formal limit of a Cauchy
the limit of that sequence;after that,
limits ever again. (The situation is much
\342\200\224

subtraction

and

that the notionof equality


of

laws

the

like

what

as

need

formal
we did with

//.)
is valid,

definition

we need

obeys

to check

the first

three

equality:

5.3.3

Proposition

(Formal

limits are well-defined).Let \317\207=


real

\342\200\224
\342\200\224
\316\266
and
Cn be
LIMn\342\200\224\321\216\320\276
\342\200\224\321\216\320\276\320\260\320\277;
\320\243 LIMn\342\200\224\321\216\320\276\320\254\320\277,

LIMn
numbers.

we

in

not

will

and

same

is the

sequence
we

division

formal

to ensurethat this definition

In order

the sequence

genuine notion of limit,

show

formal

limit of

as the
\317\205\316\234\316\267_>\316\277\316\277\316\261\316\267

on

Later

with

Then,

= \317\207.
\317\207
Also,

have

See

if

=
\317\207
then
\321\203,

of equality for real


=
\321\203 \317\207.
Finally,

if

numbers,

=
\317\207
and
\321\203

\342\200\224
\317\207
\316\266.

\342\200\224then
\321\203 \316\266,

Proof

above definition

the

Exercise

5.3.1.

Because of this proposition,we


equality between two real numbers

know

that

our

definition

of

is legitimate.
Of course, when
we define other operationson the reals, we have to check that they
two real number inputs which are
obey the law of substitution:
when applied to any operation on
equalshould
give
equal
outputs

the real
Now
arithmetic

with

numbers.

we
operations,

addition.

want

to define
such

on the real numbers


all the
usual
and multiplication. We

as addition

begin

The construction of the

5.3.

Definition5.3.4

of

(Addition

\321\203

to

Example 5.3.5. The


is LlMn^oo3 + 4/n.

of

sum

need

the

is to confirm that

to do

and

LIMn_>ool+l/n

this definition

that

check

now

We

and

LIMn_>ooan

the sum x + y

we define

Then

+ bn).

LIM n->oo (an

:=
\317\207
+ \321\203

be

=
\317\207

Let

reals).

numbers.

be
real
LIMn-\321\216\320\276\320\254\320\277

119

real numbers

LIMn_,oo2+3/n

is valid. The first

we

thing

is in

numbers

real

two

of

sum

fact a real number:


5.3.6

Lemma

and

LIMn-^oo^n

\321\203

LIMn_,oobn

real number (i.e., (an

also a

=
sequences is Cauchy). Let \317\207
be real numbers. Then \317\207
+ y is
is
a
sequence
of
Cauchy
bn)^=1

of Cauchy

(Sum

rationals).

Proof

is

show

to

need

We

eventually

bn)\342\204\242=i

eventually
(\316\261\316\267)\342\204\242=\316\271

sequence
(an +
we know that

0, the

hypothesis
is

eventually

{bn)\342\204\242=l

this is not quite enough

out that

turns

it

and

\316\265-steady,

\316\265
>

from

Now

\316\265-steady.

is

but

for every

that

\316\265-steady,

(this

but
to imply that (an + bn)\342\204\242=1is eventually
2\316\265-steady,
which
what we want). So we need to doa littletrick,

know

We

it

that

is also

eventually

\316\265/2-steady.

Since

exists an

(an

bn)\342\204\242=1

N > 1 such

that

for

that

every

{bn)\342\204\242=M

the

turn

will

eventually

play

sequence

out to
we

IS \316\265/2-steady,

(bn)(^=1

be enough to

know

i.e.,

\316\265/2-steady,
(\316\261\316\267)\302\2431\316\235

N.

but

\316\265-steady,

is

n,m>

value of

\316\265-steady.

\316\265/2-steady,

Similarly there
i.e., bn and bm

that
there
an and am

exists an \316\234>
are

\316\265/2-close for

M.

n,m>

Let max(iV,M) be the larger


2.2.13 that one has
Proposition
the other). If n, m > max(iV,
M),

are

eventually

Similarly,
This

is

is
(an)\342\204\242=1

is eventually

(an)^Li

are \316\265/2-close

\316\265/2-close,

Proposition4.3.7

is to

for every

5-steady

eventually

eventually \316\265/2-steady.
that

every

is

only that

conclude

such

(an)(\302\243=1

not

implies

is also

not

that's

value of \316\265.

with the
This
\316\264.

used

be

can

we

and

see

bn

that

and

an +

bm are
bn

and

of

to be
then

and

(we

than

greater

we know

\316\265/2-close,

am

\316\234

+ bm

and

are

equal

that an
so

from

know
or

to

and am

by

\316\265-close

for

every

n, m >
is

This

M).

max(iV,

eventually

this
equal to \321\217,
we substitute

Lemma 5.3.7
=
\317\207

Let

Proof.
sequences

\316\265
>

(an)\342\204\242=N

see

each

for

\316\265-close
from

4.3.7

each

Remark 5.3.8. The


substitution

for

the

from

+ x,

the

in

bn

bn +

one

\317\207
that
+ \321\203

that

Definition
and
xy

and

are

are

bn)\342\204\242=l

axiom

the

verifies

can

of
prove

similarly
(A quick

we

way is
have

certainly

an.)

We can define multiplicationof realnumbers


to

afn

done.

we are

\"y\" variable.

of
=

(a'n
\320\254\320\277)\342\204\242=1

\317\207
but
+ \321\203,

the

definition

and

0, and

lemma

above

since an

(an

and

an

0-close to bn, we thus


afn + bn are \316\265-close for

and

bn

that

i.e.,

of course

an +

\316\265
>

the axiom of substitution for


x + y=

is

bn

that

\"\320\266\"
variable

to observe

\316\265-close,

implies that
for

\316\265-close

eventually

are

N. Since

n>

N. This

n>

each

(a'n)\342\204\242=N

Proposition

Cauchy

so in other words
need to show that

We

0.

xf+y.

\316\265-close
bn)^=1are eventually
already know that there is an N > 1such

we

and

\316\265
>

have x+y =

that the

equivalent,

each

if

similarly

and {a'n +

+ bn)^=1

But

0.

are

for

\316\265-close

Then we

xf.

number

are

sequences

we know

equal,

{o!n)^=l

eventually

each

=
\317\207

are

x1

and

the sequences(an
that

number

equivalent
Cauchy
=
\320\253\320\234\320\277_\321\216\320\276\320\260\320\277>
\320\243 LIMn

(an)\342\204\242=1

are

number \317\207
by another
the sum \317\207
+ \321\203
(and
equal to y).

of

\317\207
and

Since

for

change

another

(Sums

bn)\302\243_l

of substitution

axiom

the

is

a real

replace

not

numbers. Suppose that

be real

they

we

should
by
\321\203

thing we need to check

(see Section A.7): if

equivalent).

the sequence (an

that

implies

numbers

desired.

as

\316\265-close,

The other

similar

The real

5.

120

in

a manner

of addition:
= LIMn_>ooan
5.3.9 (Multiplication of reals). Let \317\207
be real numbers. Then we define the product

=
\321\203 LIMn_>oobn
to

be

xy :=

LIMn_>ooanbn.

The following Proposition ensuresthat this definition


that the product of two real numbersis in fact a

is

real

valid,

number:

The construction of the real numbers

5.5.

=
is well defined). Let \317\207
(Multiplication
=
and xf = LIMn_>ooc4 be real
\320\243 LIMn_>oobn;
=
is
also
a
real
number.
Furthermore,if \317\207
xf, then
xy

5.3.10

proposition

LlMn-^ofln,
numbers.

xy =

Then
x'y\302\267

5.3.2.

Exercise

See

Proof

Of course we can prove


by a real number
replaced
this

At

if

instance,

the rationale
q with the

number

rational

is
\316\261\317\212,
\316\2612,
\320\260\320\267,...

the

is
\321\203

to y.

is equal

which

y1

rule when

substitution

similar

we embed

point
every

equating
For

121

back into the reals,by


real numberLIMn_>oo9\302\267

sequence

0.5,0.5,0.5,0.5,0.5,...

then we set limn_>oo

of addition

a, b we

numbers

rational

This embedding is consistent


and multiplication, since for any

to 0.5.

equal

an

definitions

our

with

+
(LIMn-\316\277\316\277\316\261)

have
= LIMn_>oo(a

(LIMn_>oob)

\317\207
(LIMn-,\316\277\316\277\316\261)(LIMn_>oob)

that when one wants

this means

numbers a, b

consistent

now

can

We

the

to

LIMn_>oo(ab);
or multiply

add

whether one

two rational

thinks of these

real numbers \320\253\320\234\320\277_\321\216\320\276\320\260,


LIMn-\321\216\320\276\320\254.
is
of rational numbers and real numbers
identification
with our definitions of equality (Exercise5.3.3).

this

Also,

matter

not

as the

or

rationale

as

numbers

does

it

+ b) and

easily

define

negation

\342\200\224x
for

real

numbers

\317\207
by

formula
\342\200\224\317\207
:= (\342\200\2241)
\317\207
\320\266,

since

\342\200\2241
is a

is clearly

we have
definitions

rational

consistent
\342\200\224
=

it

number

with our

\317\207 for
(\342\200\2241) q

is clear

and is

hence real. Notethat

negation for

all rational

rational

numbers q.

that

-LIMn_>ooan =

LIMn_>oo(-an)

numbers

this

since

Also, from our

5.

122
(why?).

Once we

subtraction

as

have addition and negation,we

usual

real

The

numbers

define

can

by

+ (-y),

x-y:=x

note that this implies


-

= LIMn-oo(an

LIMn-oobn

LIMn-,\316\277\316\277\316\261\316\267

now

can

We

division,

(except
address shortly):

5.3.11.
hold
not
4-1-6
for
only

Proposition

Proof
Let

We

=
\317\207

LIMn

the

integers,

LIMn_>ooan,

=
and
\316\266
\320\253\320\234\320\277-^\320\254\320\277,

\321\203

Then

line of reasoning shows that


we

are

algebra

x(y

Cn,

first

guess

for how

there are

be
the
\320\260\320\267,...

few

xy +

This

operation

one

to proceed

is a

be

A similar

+ 0\"ncn)-

= LIMn_>ooan(bn

is equal to

Onbn

claim follows.

+ Cn). But

+ anCn

for the

The other laws

we need to define is
little more subtle. On obvious

would be define
:=

with

problems

xz.

LIMn-^^

(LIMn^ooan)-1

but

similarly.

proven

\342\200\224>
\317\207
x~l.

the

and

The last basicarithmetic


reciprocation:

+ z)

Proposition

as well

and
xz =
LlMn-\321\216\320\276\320\260\320\277\320\254\320\277

LIMn_>oo{a>nbn

an(bn + Cn)

that

know

already

rational numbers an, bn,


of

+ xz

xy

definition,

by

\320\260,\321\2101
so xy
-\321\216\320\276\320\230\320\277\320\241\320\277)

involving

rule: x(y + z)

one such

with

laws

of algebra from
but for the reals

laws

the

All

this

illustrate

real numbers.

real numbersobey all the

perhaps for the

shall

we

which

that the

show

easily

of algebra

rules

usual

- bn).

LIMn^a\"1,
this.

For instance,

let

\316\261\317\212,
\316\2612,

sequence

Cauchy

0.1,0.01,0.001,0.0001,...,

and let
LIMn

:=
\317\207

LIMn_,ooan.

-\302\273oo&n> where

&i,

Then
62,63,...

by this definition, x~l


is the sequence

10,100,1000,10000,...

would

be

$.3.

construction

The

of

a
course, the problem

but this is not

sequence

Cauchy

(an)%Li

equivalent

hencethat our real

our
in fact

\317\207
was

number

original

zero sequence

to the

Of
bounded).
Cauchy sequence

even

isn't

(it

that

is

here

was

123

real numbers

the

(why?), and
to 0. So we should

(0)^

equal

of reciprocal when \317\207


is non-zero.
we restrict ourselvesto non-zero
real
a
because
a
non-zero
number
we
real
have
slight
problem,
numbers,
the
formal
limit
of
a
which
contains
be
sequence
Cauchy
might
is rational
and
1, which
zero elements. For instance, the number
=
is
the
formal
limit
1
of
the
LIMn_>ooan
hence real,
Cauchy
the operation
only
even when
However,
allow

sequence

0,0.9,0.99,0.999,0.9999,...

our naive definition of reciprocal,we


first
the
real number 1, because we can't invert
but

using

the

invert

cannot

0 of this

element

Cauchy sequence!

To get

away

sequence

(an)\342\204\242=1

zero

from

from

5.3.12

Definition
sequence

these

around

for all

iff

>
\316\267

there

of

we

problems

zero.

To do

need

this we first

(Sequences bounded
numbers

rational
exists

a rational

our

to keep

from

away

is said to
number \321\201
> 0

Cauchy

a definition.

need

be

zero).
bounded

such

away

that

\\an\\

>c

1.

is
5.3.13. The sequence 1, \342\200\2241,
1,
1, \342\200\2241,
\342\200\2241,
1,...
Examples
value
have
absolute
bounded away from zero (all the coefficients
bounded
at least 1). But the sequence
The
from
and
neither
is
zero,
0,0.9,0.99,0.999,0.9999,....
away

is not
0.1,0.01,0.001,...

is
sequence10,100,1000,...

bounded

away

from

zero,

but is

not

bounded.
now

We
limit

of

5.3.14.

Lemma

LIMn-,ooOn
away

show

a Cauchy

from

that

sequence
Let

for some

zero.

every

be
\317\207

non-zero real number


bounded away from
zero:
a

non-zero

real number.

Cauchy sequence (an)^=1which

is the formal
Then
is

=
\317\207

bounded

5. The realnumbers

124
Since

Proof.

\317\207
is

that

know

we

real,

=
\317\207

for soiue

LIMn_>oobn

Cauchy sequence (bn)^=1.But we are not yet done, because we


that
do not know
bn is bounded
away from zero. On the other
sequence

be

\316\265-close

to

for

an

\316\265
>

the

Thus

(0)^!\302\267

that

0 such

that

which means
for

to

\316\265-close

this

0.

no*

(bn)\302\243Lils

the

sequence
\316\265
>

every

(0)\342\204\242=1

0.

eventually

is an

a Cauchy

is

(bn)\342\204\242=1

it is

Moreover,

there

Thus

that

know

\316\265.
We

eventually \316\265-steady.

\316\265/2>

all

equivalent

to

(0)^.

Let us fix

so it is

LIMn_^oO,

eventually

we can find

Therefore

=
\317\207
\320\236
\317\206

not

is

(bn)\302\243Li

cannot

(bn)\342\204\242=1

since

that

are given

we

hand,

eventually

N > 1 such

that

sequence,
\316\265/2-steady,
\342\200\224 <
bm\\
\316\265/2

\\bn

N.

n,m>

< \316\265
> iV, since
for all \316\267
On the other hand, we cannot
have
\\bn\\
\316\265-close to
is eventually
this would imply that (bn)(^=1
Thus
(0)^.
>
we
N for which |bno| > \316\265.
Since
there must be someno
already
<
> \316\233\316\223,
we
\316\267
thus
conclude
know that |bno \342\200\224
for
all
from
bn|
\316\265/2
> N.
the triangle inequality (how?)that \\bn\\ > \316\265/2for all \316\267
This almost proves that (bn)\342\204\242=1is bounded
away from zero.
is
it
what
show
that
doesis
Actually,
eventually bounded
(br^^i

away from

an :=

\316\261\317\200)
by
setting

if
\316\265/2

which

sequence

Cauchy

sequences

are eventually

since

>

>
\316\267

\\bn\\

fixed, by defininga new


>
and an := bn if \316\267

1 (splitting

Once

of

reciprocal without

iV,

but
\316\265,

Cauchy sequence.

still

and so

limit,

away

>

\\an\\

and

\316\267
<

is

bounded

OK since

And
all

for

\316\265/2

N separately).
away
\316\265/2>

from
and

0),

we are done.
from

we

zero,

take

can

its

difficulty:

Suppose

is bounded away from

two

the

(because

that

we know

that's

is bounded
any

Since

= LIMn_,ooan.
the same), and so \317\207

Cauchy sequencewhich
formal

bn

sequence

N.

that an is also

verify

to

two cases n>

into the

a sequence

5.3.15.

to

hard

is equivalent

>
\316\267

all

for

\316\265/2

Thus we have a
zero (by \316\265/2instead
which has \320\266
as a

Lemma

\316\267
<

a Cauchy sequence,it is not

bn is

is easily

this

But

zero.

that

(an)\302\243Li

zero. Then the

is

a Cauchy

sequence

sequence

(a\"1)^!

which

is also

5 3-

proof Since (an)\302\243Li


\321\201
\320\260
>

is

there

skow that

that

0 such

\316\265
>

\342\200\224 <
\316\261^1!

all

\316\265
for

(since |amU^n|
it will
equal to \316\265,
find

i.e.,

for

this

fln

'

aman

make

to

~~

\316\271
^ \\Q\"m

that

an\\

c2
\342\200\224
less
\316\261^*|

\\a~l

than

or

>
\316\267

all

\342\200\224

\\an

am\\

<

eventually
have
that

all

(a~1)(\302\243=1

we

Since
is

and hence the

> N,

m,n

\316\265-steady.

said above,

we have

By what

N.

for
\316\265

is

we
\316\265,

every

\342\200\224

suffice

(a~1)(^>=1

sequence

'

so

and

<?\316\265
for

that

shows

this

<

an\\

|flm

> c),

'

to

Thus

0.

1 such

>

an

find

\316\265
>

or equal to
to make \\am
less than
an\\
and
<?\316\265
is
a
>
0, we can
Cauchy sequence,
(an)^Li
N such that the sequence (an)^=N is c^-steady,
an

since

certainly

am
m

each

know that
we need

Now

1.

for

\316\265-steady

flm
\342\200\224li\316\271

\"

\\an

n,m>

>
\316\267

all

\321\201
for

is now to
N. But

task

our

0;

i\342\200\224l
1 n

But
&\316\265.

>

\\an\\

zero, we

from

away

is eventually
{\316\277>\316\2671)\342\204\242=\316\271

let us fix an
\\a\"1

bounded

is

125

real numbers

of the

construction

The

a Cauchy

have

proven

as

sequence,

desired.
now

are

We

to define

ready

reciprocation:

ofreal
Definition 5.3.16 (Reciprocals
Lemma

is a

x~l

=
\317\207

to

the reciprocal x~l by


Lemma
5.3.15we know
(Prom

what
and

the

that

give

and LIMn-.oob\"1.

reciprocals

Fortunately,

LIMn

LIM^^a\"1

their

formal

limit,

\321\217-1,

LIMn-x^a\"1

namely

this never happens:

(Reciprocation

two

\317\207
as

The above definition might

Lemma5.3.17
^

have

which

different

two

before we are sure this


are two different
Cauchy

(bn)\342\204\242=1

= LIMn_>oobn.

LIMn_>ooan

conceivably

Then

bounded
a
sequence exists
(such
sequence

Cauchy

LIMn_>ooan

one thing
if there

check

sense:

(an)^=1

(Mt\302\243=i

real number.)

makes

sequences

be

\320\266
be

we define

:= LIMn-^a\"1.

need

We
definition

=
\317\207

Then

5.3.14).

x~l

formula

(an)\342\204\242=1

that

such

zero

from

away
by

Let

number.

real

nonzero

Let

numbers).

is

well

defined).

Let

(an)(\302\243=1

and

Cauchy sequences bounded away from zero suchthat


are equivalent).
the two sequences
LIMn \342\200\224>oobn
(i-e.,
= LIMn-oob\"1.

5. The real numbers

126
the

Consider

Proof.

:=
\316\241

=
\316\241

another

in
\316\241

=
\316\241

LIMn

write

for

formulae

(\320\253\320\234^^\320\254\"1)

things out

Multiplying

LIMn^ooa-^nb\"1

different

our

, as
\342\200\224>oobn

LIMn

= LIMn^oob\"1.

as

5.3.10).

Comparing

numbers:

obtain

\317\207
(LIMn^oobn)
(\320\253\320\234\320\277-\321\216\320\276\320\260\"1)

Proposition

(cf.

real

we
can
\317\205\316\234\316\267_>\316\277\316\277\316\261\316\267
\342\200\224\321\216\320\276\320\236\320\277,

since

way

=
\316\241

three

x (UMn^fe\"1).

(LIM^^n)

LIMn^ooa-^nb\"1

On the other hand,

\316\241
of

\317\207

(LIMn^ooa\"1)

multiply this out, we

If we

product

following

again, we get

LIMn-,\316\277\316\277\316\261^1.

\316\241
we

see

that

=
\320\253\320\234\320\277-\321\216\320\276\320

desired.

is well-defined(for each non-zeroreal number


one definition of the reciprocal\320\266-1). Note
=
it is clear from the definition that xx~l = x~lx
1 (why?);
thus all the field axioms
apply to the reals as
4.2.4)
(Proposition
well
We of course cannot give 0 a reciprocal,
as to the rationals.
1. Also note that
since 0 multiplied by anything
0, not
gives
if q is a non-zero
rational, and hence equal to the real number
Thus reciprocal
have

we
\320\266,

LIMn_>oo<7,
thus
the

exactly

the

operation
of
operation

Once one
numbers

real

= q~l\\
is LIMn-^og-1
reciprocal of LIMn_>oo<7
of reciprocal on real numbersis consistent
with
on
rational
numbers.
reciprocal
one
can define division x/y of two
reciprocal,

then the

has

is
\321\203

\320\266,
provided
\321\203,

non-zero,

x/y :=xx

just

as

we

cancellation

is

that

non-zero,

this

did
law:

with
if

the rationals.

\316\266
are
\320\266,
\321\203,

real

by the

formula

y\"1,
In particular, we have the
such that xz = yz, and

numbers

=
\317\207

we
conclude
then
that
by dividing by \316\266
law does not work when \316\266
is zero.
cancellation

Note
\321\203.

\316\266

\320\272
/

We now have

127

reals

the

Ordering

all four

of

the

arithmetic

basic

reals: addition,subtraction,
of algebra. Next
the usual rules
the

to

turn

we

on the

operations

and

multiplication,

on

of order

notion

the

all

with

division,

reals.

5.3.1.

Exercise

Prove Proposition

5.3.3. (Hint: you

may

find

Proposition

4.3.7 to be useful.)

Exercise5.3.2. Prove
may

be

5.3.10.

Proposition

(Hint:

again, Proposition

4.3.7

useful.)

Exercise

5.3.3. Let

a, 6 be rational

that

Show

numbers.

=
\316\261

and

b if

sequences a, a, a,a,...
only
=
if and
if
This
\320\260
allows us to embed
b...
equivalent
only
6,
6).
6, b,
and
real
numbers
a
inside
the
in
well-defined
manner.
numbers
rational
the
of rational numbers which
is
5.3.4. Let (an)\302\243L0 be a sequence
Exercise
is
Let (6n)\302\243Lo be another
sequence of rational numbers which
bounded.
that
to (an)\302\243L0\302\267
Show
(bn)%L0 is also bounded.
equivalent
= 0.
Exercise5.3.5. Show that LIMn_>ool/n
= LIMn_>oob
if \320\253\320\234\320\277_\321\216\320\276\320\260
(i.e.,

5.4

Ordering

We

know

We

now

that
want

the Cauchy

the reals

number
every rational
to say the same thing for

is positive, negative,orzero.
the

each

reals:

real number

or zero.
Since a real number \317\207
is just
should be positive,negative,
the
a formal limit of rationale an, it is temptingto make
following
= LIMn_>ooan
a real number \317\207
is positive if all of the
definition:
if all of the an are negative
and negative
On, are
positive,
(and
one
soon
realizes
some
zero if all of the an are zero). However,
the
sequence
problems with this definition.Forinstance,
(an)\302\243Li
:=
thus
defined
an
10~n,
by

0.1,0.01,0.001,0.0001,...
consists entirely of positivenumbers,but
equivalent

to

the

zero

sequence

0,0,0,0,...

this

though all the rationalewere positive,


limit of these rationale was zero rather than
Thus even

example

sequence

and thus

is

0.1,-0.01,0.001,-0.0001,...;

is
=

LIMn_,ooan

the
positive.

real

0.

formal
Another

5. The realnumber

128
this
again

The trick, as with

the

previous section, is to
which are boundedaway
&\320\276\321\211
in the

reciprocals

to sequences

attention

one's

limit

positive and negativenumbers,but

zero.

limit is

formal

the

hybrid of

is a

sequence

zero.

a sequenceof rationale.We
zero iff we
that
this sequence
is positively boundedaway
from
say
for
> \\
that an > \321\201
all \316\267
have a positive rational \321\201
0 such
>
the sequence is entirely positive). The sequence
is
(in particular,
Definition

5.4.1.

negatively

bounded

0 such

\342\200\224\321\201
<

(an)^L1 be

Let

away from

<

that an

zero

a negative

have

we

iff

>
\342\200\224\321\201
for all \316\267

1 (in

rational
the sequence

particular,

is entirely negative).

Examples 5.4.2.

The

positively

bounded

to 1).

The sequence

bounded

bounded

away

zero nor

from

but is neither positively


bounded
from
away
negatively

5.4.3.
as

written

be

=
\317\207

positively

to be

it

negative

sequence

can

bounded
zero

number

real

(an)%Li which is
iff

bounded

to

be written

the

(b)

positive,

Proof.

then

negative

so are

See Exercise

from

away

as

=
\317\207

if and

only if

\317\207
and
+ \321\203

5.4.1.

\317\207
is

away from

\342\200\224x
is positive.

zero.

For

reals).

three

every

statements

negative.

If

said

for some

LIMn_>ooan

bounded

it

iff

sequence

zero,

from

\321\205
is
(\321\201)

\317\207

Also, a
zero and

be positive
Cauchy

bounded

positive

\317\207
is

negatively

for
some
\320\253\320\234\320\277_\321\216\320\276\320\260\320\277

is negatively

which
(an)\342\204\242=1

or

zero.

same time.

said

\317\207
is

away

zero.

away

the

at

Proposition 5.4.4 (Basicpropertiesof


real number x, exactly one of
following
true: (a) x is zero;
is positive;
number

equal

1, \342\200\2241,1,\342\200\2241,1,\342\200\2241,...is

from zero,

sequence cannot be both positively


negatively bounded, away from

Definition

or
negatively

The sequence

zero.

clear that any sequence which is positively


from
away from zero, is bounded away

It is
bounded

can

terms are greaterthan

is
-1.1, -1.01,-1.001,-1.0001,...

from

away

zero (all

from

away

is

1.1,1.01,1.001,1.0001,...

sequence

\317\207
and

is

A real
are
\321\203

xy.

129

reals

the

\316\271
Ordering
\316\262

positive rational number, then the Cauchy


is
from
and
positively bounded away
zero,
sequence
=
is
a
real
number.
Thus
the
g
positive
hence \320\253\320\234\320\277_\321\216\320\2769
for rationals
is consistent with that for reals.
of
positivity
notion
notion
of
the
is consistent
with
negativity for rationals
Similarly?
that for reals.

Note that

g, g,

if

is a

g,...

Once we have defined positiveand negative


define absolute value and order.

Definition5.4.5

define the absolutevalue


x is

when

\342\200\224
is
\321\203

\317\207
to

a real

\320\266
be

equal

\320\266
if \320\266
is

can

number.

We

positive,

\342\200\224x

is zero.
when \317\207

and
be
\321\203
(Ordering of the real numbers).Let \317\207
than
and
write
\317\207
if
>
greater
y,
\321\203,
\342\200\224
iS \317\207 \321\203
is a negative
and \317\207
< \321\203
=
> \321\203
iS \317\207
\316\277\317\204
\317\207
and
We define \317\207
> \321\203
y,
similarly

5.4.6

Definition
real numbers.
x

of

\\x\\

and 0

negative,

Let

value).

(Absolute

we

numbers,

We say that \317\207


is
real number,
positive

real number.
< y.
define \317\207

this with

Comparing
from

Definition

with

order

4.2.8

on the

we

the definition of order on the rationals


see that order on the realsis consistent
i.e.,

rationals,

if

numbers

rational

two

g,

qf

are

number
that
q is less than q' in the rational
system, then
in
less
the
and
still
than
real
number
is
q'
system,
similarly
q
than\".
In the same way we see that the definition
for
\"greater
value given here is consistent with that in Definition
of absolute
such

4.3.1.

All the claims in Proposition


4-2.9which
continue to hold for real numbers.
5.4.7.

Proposition
for

rationals,
We

Proof.

Exercise 5.4.2.

just

to conclude
by Proposition

xz <
As

prove

Suppose

one of
we have

that xz <

yz.

the claims and leave


\317\207
and
< \321\203

Since

5.4.4 we have (y

x)z

yz

is positive,

hence

hence

yz.
an

and want

positive,

\342\200\224

xz

to

rest

real,

\342\200\224
is
\321\203,
\321\203 \317\207

\317\207
<
\342\200\224

the

\316\266
a positive

held

application

of these

propositions,

we prove

5.

130

is

also

then

Proof. Let
it is

\317\207
be

is another
\321\203

easy to seethat

= 1,

\317\207-\316\271

Proposition

a negative

times

number

\317\207-\316\271

> \321\203

the real number

from

Also,

\317\206
1).

positive

Then
and x

number

positive

xx~l

Since

positive.

nurnben

real number.

a positive

\317\207
be

(since xO = 0

zero

be

if

Also,

positive.
x~l
< y~l.

cannot

Let

5.4.8.

Proposition

real

The

5.4.4
number

cannot be negative,
since
this
=
1 is negative, a contradiction. Thus,
would imply that \321\205\321\205~\320\263
by
the only possibility left is that x~l is positive.
5.4.4,
Proposition
let \321\203
be positive
Now
as well, so x~l and y~l are alsopositive.
> y_1, then by Proposition 5.4.7we have
xx~l
If x~l
> yx~~l >
1
we
thus
a
contradiction.
Thus
must
which
is
>
have
1,
yy_1,
is

this

negative;

that x~l

shows

x~l<y~l.

Another
(Proposition

rationale

have

already

need

not

negative rationale (i.e., rationalethat


is non-negative.

a2,

non-negative

Eventually, we will
set

Proof.

We

argue

by

contradiction,

:=
number \317\207

LIMn_>ooan

negative

real number,

bn which

a better
of this fact: the
explanation
reals
closed, whereas the setofpositive

12.2.

See Section

is open.

see

reals is

of non-negative

is a

we

and suppose

that the real

negative number. Then by


have

of

definition

= \320\253\320\234\320\277_\321\216\320\276\320\254\320\277
\317\207
for
some

is negatively bounded away

sequence

from

< 0 such that bn <


negative rational \342\200\224\321\201
> 1,
we have an > 0 for all \316\267

other hand,

a\\,

real number.

is a
\320\253\320\234\320\277-\321\216\320\276^\320\277

Then

Let

closed).

of non-negative rational numbers.

sequence

Cauchy

or zero)

positive

are

reals

non-negative

(The

either

are

of non-

limit

formal

the

However,

positive

be zero, as the example

it could

positive;

showed.

be a
\320\260\320\267,...

rationale, are also

the formal limit of

that

0.1,0.01,0.001,...

Proposition 5.4.9

exponentiation

5.6.

seen

be

of

proven for

previously

see Section

reals;

We

were

that

4.3.12)

for

true

is that the laws

application

zero,

\342\200\224c
for

all

i.e.,

there

>
\316\267

1.

by hypothesis.

is a

On the

Thus the

numbers an and

never

are

bn

(an)^
c/2 > 0) this impliesthat
the
fact
that
contradicts
this
gut

not

are not

(bn)^=1
these

both

Since

c/2-close.

eventually

and

the

Thus

c.

\321\201/2<

equivalent,
have

sequences

\317\207
as

limit.

formal

their

since

c/2-close,

are
(bn)\342\204\242=1

and
(an)\342\204\242=1

sequences

131

reals

the

Ordering

A.
\316\262

sequences
Corollary 5.4.10. Let (an)^=l and (bn)\302\243Li be Cauchy
>
>
1. Then \317\205\316\234\316\267_>\316\277\316\277\3
all n>
such
that
bn
an
for
ofrationals
LlMn-oobn\302\267

5.4.11.

Remark

5.4.9 to the sequence


an

Proposition

Apply

Proof.

Note that the

above

now

define

distance

d(x, y)

:=

not work
:= 1 + 1/n

greater than

\342\200\224
\\x

4.3.3
rationals.In fact, Propositions

and

but

the rationals,
numbers

if an

limit

formal

but

bn,

of bn,

are equal.

instead they
We

bn.

does

Corollary

if the
signs are replaced by >: for instance
then an is always strictly
md bn := 1 \342\200\224
1/n,
limit
the
of
formal
an is not greaterthan
the

>

\342\200\224

all

obey

for

the

the

the

reals;

laws of

y\\ just
4.3.7

as we
hold

proof is identical,

did for

not

the

only for

since the real

algebra and orderthat the rationals

do.
now

We
arbitrarily

large

integers,

or

that

observe
or
smaller

small,
in

positive real numbers can be


they cannot be than all of the positive
than all of the positive
magnitude
while

rationals:

5.4.12
of reals by rationals). Let \317\207
be a
(Bounding
real number. Then there existsa positive
rational
number
N such that
q such that q < x, and there existsa positive
integer
Proposition

positive

x<N.
is
Proof. Since \320\266
sequence

Cauchy

zero. Also, by
have

rationals

But by

positive

real,

it is

the formal limit of some

is positively bounded away from


(an)^=1
this
Lemma
sequence is bounded. Thus we
5.1.15,
> 1.
such
that
q > 0 and \320\263
q < an < r for all \316\267
which

Proposition 4.4.1 we know

that

there

is some

integer N

5.

132
r <

that

such

obtain that

5.4.10 we

iV,

>
\316\267

all

for

r <

see

that

Corollary

as desired.

real numbers.

positive

number^

we

JV,

1. Applying

5.4.13 (Archimedeanproperty).Let
Then there existsa positive

Corollary
that

<
\317\207

<

and q <

positive

an < N

q <

Thus

is positive.
\316\233\316\223

q is

since

N\\

real

The

\317\207
and

\316\265
be

any

\316\234

integer

such

\316\234\316\265
> \317\207.

number

The

Proof.

we set
This

property

is
\317\207

large

N +

:=
\316\234

and

1, then

< \316\234.Now
\317\207/\316\265

now

any

that

such

If

\316\265.

keeps adding

to
\316\265

itself,

one

two

numbers

real

\317\207
<

q <

\317\207
<

y,

we can

y.

that

the

operations,

laws

the

our construction of the real


the rationale, and has almost

completed

number

everything

by

multiply

Exercise 5.4.5.

have
This

if one

\316\265
is,

Given

We

<

\317\207/\316\265N.

overtake x.

Proposition 5.4.14.
find a rational number

numbers.

by Proposition

that

important; it says that no matterhow

is quite
how small

will eventually

Proof. See

such

integer

positive

hence

and

is

\317\207/\316\265positive,

5.4.12 there existsa

system
rational

contains

system has: the arithmetic


laws of order. However, we

number
Ihe
algebra,

of

yet demonstrated
any advantages that the realnumbers
have
over
the rationals;
we have
so far, even after much effort,
all
done is shown that they
are
at least as good as the rational
not

have

number

But

system.
can

numbers

take square

do

more

roots in a real number

5.4.15.

Remark

next few sections we show that


things than rationals: for example,

in the

Up until

numbers can be expressed


instance, the formal limit of

we

can

system.

now, we have

that real

the real

using

not

the

the

fact

system.

For

addressed

decimal

1.4,1.41,1.414,1.4142,1.41421,...
is more

represented

conventionally

will

address

this

in an

Appendix

as

the

decimal

but
(\302\247B),

for

1.41421
now

let us

We

just

$.5. The least upper bound

and 1.000... are in

instance 0.9999...

Exercise 5.4.1. Prove

be eventually \316\265-close to
the
Use this to show that

zero sequence

the

the

Prove

Exercise5.4.3.Show
such that

integer N

Show

5.4.4.

Exercise

a positive integer N
Exercise 5.4.5.
also

may

<

Prove

that

\317\207
>

if and

[x\\.)
real number

\317\207
> 0

use Exercise

(Hint:

and let

numbers
only if

\316\265
> 0

be a

\342\200\224
\342\200\224
if
\316\265
< \317\207
< y+\316\265, and
y
\\x
y\\ < \316\265and
\342\200\224
<\317\207
\316\265.
\316\271\317\212
+
\320\271\321\203
\321\203\316\265

that

only

and
be real
numbers.
Let \317\207
Show that
5.4.7.
\321\203
<
if
and
if
\317\207
that
numbers
0
real
\316\265
>
Show
all
only
\321\203.
|rc
=
and
if
\317\207
numbers
\316\265
0
if
>
real
only
\321\203.

exists

5.4.4. You

<

\\x\342\200\224y\\ \316\265

< \321\203
for
\317\207
+ \316\265
\342\200\224
<
for
\317\204/| \316\265

all

Let (an)^=1 be a Cauchy sequenceof rationale,


and let \317\207
> 1, then LIMn-,\316\277\316\277\316\261\316\267
<
for all \316\267
number. Show that if an < \317\207
>
>
>
if
all
\316\267
x.
show
that
\317\207
for
then
an
LIMn_>ooan
1,
Similarly,
5.4.8.

Exercise

x.

there

positive real.

that

Exercise

be

one

is exactly

is calledthe integer

contradiction.)

be real
Exercise5.4.6.Letx,\321\203
Show

5.4.7.

> 0.

\\/N

5.4.14.

Proposition

\317\207

cannot

either positively
from zero.)

\317\207
there

integer N

N =

for any positive

that
such

away

real number
(This

denoted

to argue by

need

every

and

zero,

in Proposition

claims

\317\207
< N+l.

is sometimes

and
\320\266,

of

part

for

that
N

not

sequence

for every single \316\265


> 0.

(0)^!

bounded

remaining

real number.

is
\317\207
this

is eventually
(\316\261\316\267)5\302\243=\316\271

sequence

bounded away from zero or negatively


Exercise 5.4.2.

if

(Hint:

for

system,

same

the

fact

5.4.4.

Proposition

decimal

the

some sequence(an)^=uthen

limit of

formal

is the

some subtletiesin

there are

that

remark

133

property

a real

(Hint: prove

by

5.4.14 to

Use Proposition

contradiction.

find

a rational

use Proposition5.4.9.)

LIMn_>ooan and x, and then

between

5.5 The least upperbound

property

We

now

over the
any

give

one of

the most basic advantages

rationale; one can take

subset

\316\225
of

the

real

the

numbers

least

of

upper

we

have

< \316\234for
\317\207

every

element

real

numbers

bound sup(J5)

of

R.

Definition 5.5.1 (Upperbound). Let \316\225be


let \316\234be a real number. We say that \316\234is an
iff

the

\317\207
in

E.

a subset

upper

of R, and
bound for \302\243\",

The real numbers

5.

134
be
Example 5.5.2. Let \316\225

indeed

and

interval

:=
\316\225

for

since
\302\2437,

every

to 1. It is alsotrue

or equal

than

bound

an upper

1 is

Then

the

greater or

number

every

for E.

On the

other hand, any

upper

bound,

because

other

such

not larger than

0.5 is

every

of \316\225is
(Merely being larger than some elements
make
to
an
0.5
upper bound.)
enough
Example 5.5.3.
:

\317\207
>

0}.

\316\225

is not

element

not

for

bound

upper

as 0.5,

\\\316\263

less

in

an

E.

necessarily

be the set of positive reals: R+ := {\317\207


\316\276
does not have any upper bounds3at all

R+

Let

R+

Then

\317\207
<
\316\225
is

bound

equal to 1is an
number,

0 <

of

element

an upper

2 is

that

R :
\342\202\254
{\317\207

(why?).

an

It is

Mf

than

following

if

an

upper

also

an

upper

also

\316\234to

greater

vacuously true

\316\234is

\316\234is

so clear

is not

it

smaller

>

Then

set.

empty

because \316\234is

is a

(this

that

clear

number

the

set

empty

be the

for 0,

bound

upper

of the

hand,

Let 0

5.5.4.

Example
is

every number \316\234


than every element

statement, but still true).

of E, then any larger


bound of E. On the other

bound

whether it is also possiblefor any number


be an upper bound of E. This .motivates

definition:

be
a subset
of R,
5.5.5 (Least upper bound). Let \316\225
and \316\234be a real number. We say that \316\234is a least upper bound
for \316\225iff (a) M is an upper bound for \302\2437,
and
also
(b) any other
M'
bound
for
\316\225
must
than
or
be
upper
equal to M.
larger

Definition

Example 5.5.6.

Let

Then,

as noted

before,

number greater than


is the

\316\225
be

least upper

or

interval

the

\316\225
has

equal

upper

many

to 1

:=
\316\225

is an

R :
\342\202\254
{\317\207

0 <

<
\317\207

1}.

indeed every

bounds,

upper bound. But only

bound; all other upperbounds

are

larger

than

1.

Example

5.5.7. The empty

set

does

not

have a

least upper bound

(why?).
R+ has no upper bounds which
3More
In
are
real numbers.
precisely,
Section 6.2 we shall introduce the extended real number
R*, which
system
allows one to give the upper bound of +oo for sets such as R+.

The least upper bound

5.5.

5.5.8

proposition

jlf2.

is a

M\\

have

\316\225
can

Proof. Let Mi and M<i


Since

of least upper bound). Let \316\225be


at most one least upper bound.

(Uniqueness

Then

of R.

a subset

be

135

property

two

least upper

least

say M\\ and

bounds,

upper

bound and M<i

is

an

bound,

upper

we
have
M<i > M\\. Since
then by definition of least upperbound
bound
and M\\ is an upper bound, we similarly
yi^ is a leastupper
= M2. Thus there is at most one least
>
Thus
M<i.
M\\
M\\
have

bound.

upper

Theorem

to an importantproperty of the real numbers:

we come

Now

of least upper bound). Let \316\225be a


(Existence
subset of R. If \316\225has an upper bound, (i.e., \316\225has
bound M), then it must have
one
least upper
upper
exactly
5.5.9

non-empty
some

bound.

Proof. This theorem will


many

By

the

of

will be

steps

Let Bbea

5.5.8,

Proposition

\316\225
is

Let

be a

\316\225
has

at

We

most

at

least

choose some

positive integer.

an upper

R with

of

that

\316\225
has

we can

non-empty,

> 1
\316\267

to prove, and

left as exercises.

we know
that

of effort

a bit

quite

subset

non-empty

bound; we have to show

Since

take

bound M.

one least

one least

upper
upper bound.

element xo in
that

know

\316\225
has

E.
an

upper

By the Archimedeanproperty (Corollary5.4.13),we


> M, and hence K/n is also
find an integer
\320\232
that
can
such
K/n
for E. By the Archimedean property again,there
an
bound
upper
L such that L/n < x$. Sincexo lies in E,
another
exists
integer
we see that L/n is not an upperbound
for
E.
is an
Since K/n
upper bound but L/n is not, we see that \320\232> L.
and
is not, we can find
Since K/n is an upper bound for \316\225
L/n
an integer
L < mn < \320\232
the property
that mn/n is an upper
with
\342\200\224
bound
for
but
is
not
Exercise
E,
5.5.2). In fact,
(mn
l)/n
(see
bound

this

to

M.

integers,

of the

is unique

the

fact that

(Exercise 5.5.3).

(mn

We

subscript

this integer m dependson the

well-defined(and unique)
mi,
sequence
with each of the mnjn being upperbounds

gives a

n. This
of

mn

integer

emphasize

\342\200\224

l)/n

not

being upper bounds.

mn
choice

7712,
and

by

\316\267

of

\320\2503,...

each

5.

136

> 1 be

let N

real

The

numbers

> N

be
an
is
upper
integerslarger
bound
\342\200\224
\342\200\224
\316\225
for
and
is not, we must have mn/n > (mn>
(mnf
l)/n'
l)/nf
After a little algebra, this impliesthat
(why?).
Now

or

than

nynf

Since mn/n

to N.

equal

let

and

integer,

positive

~~
nf

\316\267

since

Similarly,

we have

is not,

is an

mn'/nf

>

mnf/nf

(mn

upper bound
\342\200\224

and

l)/n,

bounds

two

these

Putting

\316\267

\316\267'

for

\316\225
and

(rrtn

\342\200\224

l)/n

hence

~~
\316\267

\316\235'

\316\267'

\316\235

we see that

together,

|^-^|<^fcralln,n'>JV>l.
JN
rv
that ^ is a
implies
sequence
rational
are
we
can now
^
\316\267

This

Cauchy

the

numbers,

Sas

5.5.4).

(Exercise

define the

Since

real number

5:=\320\253\320\234\320\277^\320\276\320\276^.
\316\267

we conclude

5.3.5

Exercise

Prom

that

5 = LIMn^oo^^.
\316\267
To

the

finish

least upper

Let
for

\317\207
be

any

<
\317\207

we conclude that
bound

upper

E.

of

element

have

we
\302\243?,

theorem, we need to show that S is the


we show that it is an upper bound.
E. Then, since mn/n is an upperbound

of the

proof

bound for

mn/n

<
\317\207

First

for all

>
\316\267

LIMn_,oo^n/^

1.

Applying
= S. Thus

Exercise

5.4.8,

S is indeed an

for E.

Now we show it is a leastupper


bound.
is an upper
Suppose
\321\203
\342\200\224
bound for E. Since (mn l)/n
is not an upper bound, we conclude

that

>

\321\203

(rn'n

conclude that

\342\200\224

l)/n

>
\321\203

bound S is less

than

is thus

a least

for

all

>
\316\267

LIMn_,oo(mn
or

equal

1.

Exercise
Applying
= S. Thus
l)/n

\342\200\224

to every

upper boundof E.

upper bound of

5.4.8, we
the upper
and
\302\243?,

S
D

least

The

5.5.

Definition 5.5.10
is
numbers. If \316\225

137

bound property

upper

Let

(Supremum).

real

of the

a subset

\302\243
be

has some

and

non-empty

upper bound, we

define

least upper bound of \316\225(this is well-defined


We
introduce
two
additional
+oo
symbols,
by Theorem 5.5.9).
\316\225
is
and
no
has
we
set
\342\200\224oo.If
non-empty
upper bound,
\320\260\320\264\320\260
:=
if
\316\225
is
we
set
\342\200\224oo.
:\342\200\224
We
refer
tp
+\302\260\302\260;
empty,
sup(E)
s\\xp(E)
the
of
and
E.
alsodenoteit
as
supremum
E,
by
sup
svp(E)
to

svp(E)

be the

5.5.11.

Remark

involving

results

and

+oo

present,

\342\200\224oo
are

to the
is

not

because

6.2

many

we

but

system,

+oo

\342\200\224oo

this
system,

it

instance,
this

\316\227\342\200\224oo;
setting

of how the least upperbound

an example

give

and

+oo

with as the realnumber


the laws of algebra break down.
For

is not a good idea to try to define


to 0 causes some problems.
Now

add

we

to work

as convenient

of

meaningless

at present, and noneof our


to +oo and \342\200\224
because
oo,

are not real numbers. In Section


reals to form the extended real number

these

system

At

on them
no operations
numbers
real
apply

have

symbols; we

equal

property

is useful.

5.5.12.

Proposition

There exists a positive real number

5.5.13. Comparing this

Remark

that certain

of this

proposition alsoshows

least

upper

bound

to construct a
possible.

Proof.

Let

set of

that

square

the

root

set

all non-negative

the

of 2,

{y

with

not

4.4.4,

Proposition

The

rational.

Q do

rationale

proof

not obey the

use that property


which by Proposition 4.4.4 is not

otherwise

property,

be
\316\225

result

numbers are real but

we see

the

\317\207
such

= 2.

thatx2

one could

>
R : \321\203
\342\202\254

0 and

real numberswhose

y2 <

2}; thus

square

is less

\316\225
is

than

2. Observe that \316\225has an upper bound of 2 (because if \321\203


>
2,
then y2 > 4 > 2 and hence \321\203
\316\225
is
Also,
non-empty
g E).
(for
1 is an elementof E). Thus
the
least upper bound
instance,
by
\317\207
:= sup(E)
which is the least
property, we have a real number

upper

bound

of

E.

Then

\317\207
is

greater

than

or equal

to 1 (since

5.

138
1

and

less than

\317\207
is

positive.

\342\202\254
E)

So

E).

We

this

argue

x2

0<

\316\265
<

by

lead to contradictions. First supposethat


be a small number; then we have

(x +

<
\316\2652

the fact that

\317\207
is

an

Now supposethat x2 > 2. Let


then we have

(x since

such

that

\342\200\224
\317\2072 4\316\265
>

bound

for

>
\316\2652

<

2,

which
\302\2437,

0.
2,

Since

<

\317\2072
>

thus

and
all

\316\265
<

a small

1 be

number;

choose 0

we can

2,

\342\200\224
(\317\207 \316\265)2>

Thus
that

\316\265
<

this

\317\207
is

the

an

upper

least

see that

upper

x2 = 2,

as desired.
to

property
more

many

6 we will use the leastupper


bound
develop the theory of limits, which allows us to do
than just take square roots.
things

5.5.14.

Remark

In Chapter

Remark 5.5.15.

We

greatest

lower

set

also

or

\316\225
is

inf

E.

can

of course

of sets E\\
bounds,
as the infimum4
known

Everything

4Supremum

means

we say
\"highest\"

having

mostly

replaced

talk about
the

lower

greatest
of

lower bounds, and

\316\225
and

bound

is denoted

about suprema has a counterpart


and

are suprema and infima.


Supremum
maximum is to major, and minimum
which
means
\"above\", and \"infer\",
in a few rare English
survives
words
\"sub\"

\342\200\224
\317\207 \316\265
< \321\203
then

\342\200\224
\317\207 \316\265
is

we

contradictions

<

then

But

2.
If

\316\225.
\342\202\254
\321\203
(Why?

a contradiction.)
the fact
contradicts

E. Prom thesetwo

bound of

By

of E.

\342\200\224
> \321\203
for
\317\207 \316\265
\321\2032

2.

\316\265)2<

this contradicts

but

> \317\2072 4\316\265


> \317\2072 2\316\265\317\207
\317\2072 2\316\265\317\207
+ \316\2652

2 and

\342\200\224
(\317\207 \316\265)2<

+
(\317\207

\317\207
G E\\
+ \316\265

\316\265)2

<
\317\207

implies that

2. Let

we can choose

see that

2, thus

5\316\265
<

bound

upper

2, we

\317\2072
<

\320\2662
+

this meansthat

of E,

construction

\316\265.
Since

that

1 such

\316\265
<

<

x2

= \317\2072
< \317\2072
\317\2072
+ \316\2652
+ \316\265
+ 2\316\265\317\207
+ 4\316\265
+ 5\316\265

\316\265)2

2 and

<
\317\207

an 0 <

numbers

or equal to 2 (since2 is an upper


bound
for
Now we show that x2 = 2.
contradiction.
We show that both x2 < 2

> 2

and

since

real

The

\"infer\"

infimum

means

\"lowest\",

and

of a
inf (E)
for

the plurals

as
to inferior,
superior, and infimum
to minor.
The root words are \"super\",
\"below\" (this usage only
which
means
with
the Latin prefix
such as \"infernal\",

is to

in English).

the reader.

leave such statements to


usually
between
the two notions is given by

will

we

infinia;

relationship

precise

Exercise5.5.1.Let
\342\200\224
\316\225
be

Let

SUp(i?).

the

\342\200\224
\316\234
is

that

Let

5.5.2.

Exercise

the

:= {-x

and

real number,

suppose

i.e.,

=
\316\234

:x\342\202\254E}.

\320\232
be

mi(\342\200\224E).

> 1
of R, let \316\267
is
that
an
Suppose
K/n
E.
bound
for
Without
upper

integers.

is not

for E, but that L/n


Theorem
5.5.9, show that

bound

\342\200\224
=
\316\234

an

be an

subset

a non-empty

\316\225
be

of \342\200\224
E, i.e.,

lower bound

greatest

L <

integer, and let

is a

\316\234
which

set

-E
Show

real numbers R,

of the

subset

Ebea

upper bound

a least

has
\316\225

that

Exercise

Section 6.2.

See also

5.5.1.

139

part I

Real exponentiation,

5.0.

upper
using

there exists an integer L <m< \320\232


such
that
\342\200\224
an
for
but
is
not
bound
bound
that
is an upper
E,
upper
(m l)/n
\317\204\316\267/\316\267
It may
also
E.
for
prove by contradiction, and use induction.
(Hint:
a
the
of
to
draw
picture
situation.)
help
Let

5.5.3.

Exercise

subset

a non-empty

\316\225
be

m,m' be integerswith

integer, and let

the

of R, let

properties

1 be

>
\316\267

that

m/n

an

and

\342\200\224 are
\342\200\224 and
for E, but
not
(m/
l)/n
l)/n
(m
m'/n are upper bounds
= m'. This shows that the integer m
E.
for
that
m
bounds
Show
upper
constructedin Exercise5.5.2is unique.
drawing a picture
again,
(Hint:

be

will

helpful.)

5.5.4. Let

Exercise
the

property

n,n' > M.
if

:=

qnf\\

\302\267
\302\267
\302\267
be a

<

jj

\302\267
\302\267
\302\267
is a
41,42\302\2734\320\267\302\273

show that

LIMn^oo^n,

sequence

whenever

\\qM

S\\

\316\267
is

\316\267
is

an

the

properties

for the

-fa

1 is

sequence.
for

every

numbers with
an integer and

Furthermore,
>
\316\234

1.

(Hint:

5.4.8.)

4.3

Section

and

of rational
\316\234>

Cauchy

\342\200\224
<

Real exponentiation, part

5.6

on

\\qn

that

Show

use Exercise

In

#3i
4i, 42\302\273
\342\200\224

that

xn when \317\207
is rational
exponentiation
or
and
rational
when \317\207
is a non-zero
number,
Now that we have all the arithmeticoperations

we defined

a natural

integer.
reals (and
of

the

5.4.7 assures us that the arithmetic


that we are used to, continue
to
hold

Proposition
rationale

reals) we can similarly

define

exponentiation

of the

reals.

5. The realnumber^

140
5.6.1

Definition

Let

\320\266
be

x\302\260
:=

some

Now suppose
natural
number n,
1.

5.6.2

Definition
be

To raise

number.

real

x~n

define

the

\317\207
to

a natural number).
0, we define
power

recursively that xn has beendefined


\317\207
:\342\200\224
xn
\317\207.
then we define \320\266\320\277+1
Then

for

integer

negative

any

we
\342\200\224n,

:= l/xn.

the
Clearly these definitions are consistentwith
rational exponentiation
given earlier. We can then

All the

5.6.3.

Proposition

valid

remain

4-3.12

for

by an integer). Let \317\207

a real

(Exponentiating

real number.

a non-zero

real by

(Exponentiating^

if

definition

of

assert

properties in Propositions
4-3.10and

\317\207
and

be real

to

assumed

are
\321\203

numbers

instead of rationalnumbers.

Insteadof
give a meta-proof

an

giving

actual

proof

to

appealing

argument

(an

proposition, we shall
the nature of proofs,

of this

the
nature
of real and rational numbers).
rather
4.3.10 and
Meta-proof.If one inspects proof of Propositions
of
on
the
laws
and
the
4.3.12 we see that
algebra
rely
of order
for the rationale (Propositions 4.2.4 and 4.2.9). But
5.4.7, and the identity xx~l = x~lx = 1
5.3.11,
Propositions
know
that
all these laws of algebra and order
to hold
for
real numbers as well as rationals.
we
can modify
the proof
of Proposition 4.3.10and 4.3.12tohold the case when
and
than

the

laws

they

by

we

continue

Thus

in

are

\317\207

Now we

consider exponentiation to exponentswhich


with the notionof an nth root, which

We begin

integers.

define using our notion


Definition

a positive
\320\266,
by

the

5.6.4.

integer.

of

Let

0 be

a positive

write

\\fx

for

x1/2.

not

we can

\320\2661/71,also

>
real, and let \316\267

known

as the

formula

sup{y

are

supremum.

\317\207
>

define

We

x1/n :=

We often

\321\203

real.

>
R : \321\203
\342\202\254

0 and

yn <

x}.

nth

root

1 be
of

exponentiation,

Real
\302\243.

Note we do not

jo

the

nth

define

141

at this point, nor

of zero

root

a negative number. The former


issue
as
the
we
will
the
nth
for
leave
latter,
presently;
numbers undefined for the rest of the text (one
of negative
these nth roots once one defines the complex
define
numbers,
refrain
from
shall
we
doing so).
root of

nth

the

define

we

be addressed

^U
roots
can
but

Lemma 5.6.5
real, and let >

yn

The set

and

x < 1

First

1.

To see this, supposefor


\316\225
for
which
\342\202\254
element \321\203
hence

>

yn

\317\207
>

Since

have

1,

>

yn

an upper

(why?),

basic

some

bound.
upper
1.
Then
we

by x.

To see this,
>
\321\203

Thus

If

(b)

Conversely,

\321\203

(c) xlln
We

xlln,

of

properties

then

if

yn

yn

in both

\317\207
> \321\203
if

for

> \317\207.
\321\203

>
\321\203

cases

1,

we

\316\225
has

be

root

nth

reals,

positive

below.

and

x,

x.

then

\321\203

xl/n.

is a positive real number.


have

\317\207
and

Now
claim

xxln is finite.

integers.

(a)

and

suppose

which

\316\225
for
\342\202\254
\321\203

Since

1.

cases:

1 (why?),

\317\207
>

element

>
\321\203

Let x,y > 0

5.6.6.

positive

(d)

have

and so

bound,

Lemma

above

was an

an

\316\225
has

>

yn

where

a contradiction.

\317\207

list

We

we thus

case

the

there

that

contradiction

Thus

a contradiction.

x,

suppose that we are in


that the set \316\225is bounded

that

then

But

1.

>
\321\203

empty.

two

above
by 1.
there was an

bounded

contradiction

of

sake

into

are in the casewhere

that we

suppose

certainly not

divide

We

that the set \316\225is

we claim

Then

1.

<

\317\207
>

so it is

0 (why?),

an upper bound.

it has

a positive

0 be

number.

real

\316\225
contains

show

we

Now

\317\207
>

integer. Then the set \316\225:= {\321\203\342\202\254


and is also bounded above.
non-empty

is
\317\207}

In particular,xxln is a
Proof.

Let

roots).

a positive

1 be

<

nth

of

(Existence

\316\267

> \320\236
and
: \321\203
\320\233

part

and

only

if xl/n

> yl/n.

let n,

m > 1

be

5.

142
If

(e)

\317\207
>

xi/k _

If

j j0T \320\260\321\206
^
=

(/)

We have

(g)

Wehave(x1/n)l/m

\342\226\241

reader

observant

1, but it is

may

inconsistent

be

possibly

might

= xl/nm.

5.6.1.

Exercise

=
\316\267

xllnyl'n.

{\321\205\321\203)1'\320\277

Proof See
The

numben

xl/k is a decreasingfunction of k.
\317\207
< \\
=
is an increasing function of k. If \317\207
1, then

then

1,

xllk

then

real

The

note that this


with our previous

easy to checkthat

=
\317\207

\317\2071/1=

definition of x1/*
notion of xn when
so there

x1 (why?),

is no inconsistency.

One consequence
of
if

cancellationlaw:

\316\266
are

follow from

this

does
when

and
\321\203

and
\321\203

\316\266
are

=
\321\203 \316\266.
(Why

positive;

works

32, but we

for instance,
= 3.
\342\200\2243

(\342\200\2243)2

define how to raise a positive

\320\266
to

number

integera

and

0 be

cannot

a rationed

b, and

integer

positive

x* :=

Notethat
can be

positive (why?).
in the form a/b

in

form

the

write

define

(xl/b)a.

g, whether

rational

every

written

\316\261
is

However,the rationalnumber
in more than one way, for
check

the same formula for

to

So

that

ensure

that

different

zero,

and b is
integer
can be expressed

instance

Let a, a' be integers and b, b'


=
such that a/b
a'/bf, and let \317\207be a positive
=

an

this

1/2 can also


definition is

expressions

a/b give

xq:

5.6.8.

(xllb')a'

negative, or

positive,

a/b where

be expressedas 2/4 or 3/6.


well-defined, we need to

real number, and let


q = a/b for some

a positive

To define \320\2669,we

number.

a rational

we have

then

Lemma 5.6.6(b)?) Notethat this only

5.6.7. Let \317\207


>

Definition

Lemma

following

yn = zn,

and

q.

exponent

q be

is the

5.6.6(b)

positive

conclude from this that


Now we

Lemma

(\321\205\320\263/\321\214)\320\260.

be positive

integers
real number. Then

R>eal

$.6.

equal

ba'. Write
=

{xllb')l,a
= xxlh' and

and

\321\203
ya

\320\2661/(<**>') =

:=

\321\203

0. Then

a >

that

suppose

Now

a =

0. If

<

0, then

(\320\2661/6)0

are

are done.

so we

to 1,

0, a > 0, a

(why?) and so both {xl/b')a'and

a' = 0

have

must

We

a =

cases:

three

are

There

proof-

143

part

exponentiation,

by

\321\203 (\320\2661/6)1/0';

= (ya)a'

suppose

that

positive,

so

we have

5.6.6(g)

we thus

5.6.6(a)

have

have

we

\320\2661/6. Thus

(a-i/*y

Lemma

ab'

and

(why?),

\320\260\320\224/(\320\254\320\276')\320\262
Lemma
\320\222\321\203

ya' =

a' > 0

= y-' =

(/)a = [x^f

desired.

as

Finally,
is
\342\200\224a

But

(xllb')~a'

the

have

(\342\200\224a)/b

case applies and

previous

(\342\200\224af)/b.

have

we

Taking the reciprocal of both sideswe

(xl/b)~a.

result.

the

obtain

0. Then we

a <

Thus xq is

well-defined
with

consistent

is

definition

for

that this

q. Note

rational

every

our old

definition

for

xlln

(why?)

new
and

for
xn (why?).
is also consistent with our old definition
Some basic facts about rational exponentiation:

> 0 be positive reals, and

Let x,y

5.6.9.

Lemma

let

g, r

be ratio-

nals.

(a)

(b) xq+r
(c) x~q

Proof
We

have

then

then

1,

xr if and

See Exercise
still

(\317\207*)7,

x*r.

= l/xq.

\317\207
>

xq >

and

xqxT

(d) lfq>0,
(e) If

real

a positive

is

xq

\321\203
if

xq >
only

and

only

xr if and

if xq
only

if

> yq.
q >

r.

If

\317\207
<

1; then

ifq<r.

5.6.2.

to do real exponentiation; in other


xy where \317\207
number
is a real
> 0 and
\321\203

have

to define

>

words, we still
- but we will

The

5.

144

Section 6.7, oncewe

defer that until


of limit.

In the restof the


to

numbers

obey

all the

the

formalized

have

now just

shall

we

text

real number*

concept

assume the real

usual laws of

algebra, order, and

Lemma 5.6.6.

(Hints: review

exponentiation.

5.6.1.

Exercise
Proposition

5.5.12.

especially

Prove

when

combined

the

with

xl/n

> 1,

and if \317\207
< 1

Exercise5.6.2.Prove
Lemma

5.6.6 and

then

Lemma

xl'n

5.6.9.

parts
part

of

proof

of order in

trichotomy

5.4.7and Proposition5.4.12.The earlier


prove later parts of the lemma. With

to

then

the

proof by contradiction a

you will find

Also,

useful

of the lemma can be used


> \317\207
(e), first show that if \317\207

< 1.)
(Hint:

you should

rely mainly

on algebra.)

Exercise5.6.3.If a:is a realnumber,

tool

Proposition

show

that

\\x\\

(\320\2662)1/2.

on

Chapter

of sequences

Limits

6.1 Convergence and


In

the

limits
various

in

constructing

(Cauchy)
the real

the

integers

eventually

replaced

really

finished
never

actual

actual

with

differences

we

on

operations

got

limits

formal
the

job

around

laws

the real numbers as formal


sequences, and we then defined
numbers. However, unlike our work
(where we eventually replacedformal

we defined

chapter,

previous
rational

of

limit

differences)
quotients

and rationale (where


with actual quotients),

we
we never

the real numbers,because


of constructing
to replacing formal limits LIMn_^ooan
with

In fact, we haven't defined


at all
limits
be rectified.
much of the machinery of \316\265-close
by repeating
- but this
of
time, we do it for sequences
again
not rational numbers. Thus this discussion
will
we did in the previous chapter. First, we
define

limn_>oo an.

yet. This will now


We
sequences,

begin
etc.

real numbers,
supercede

distance

what

for real

numbers:

Definition 6.1.1 (Distancebetweentwo real numbers).


two real numbers \317\207
and
y, we define their distance d(x, y)
d(x,y)

Given
to

be

:= \\x-y\\.

Clearly this definition is consistentwith


works just as well
4.3.3
Further,
Proposition
does for rationals, becausethe real numbers
algebra that the rationalsdo.

4.3.2.

Definition
for

real
obey

numbers
all

the

as it

rules

of

6.

146
6.1.2

Definition

number.

We

<

d(y,

\317\207)

say

Let

numbers).

real numbers

two

of

be

>

\316\265/

are
\320\266,
\321\203

sequences

a real

we

iff

\316\265-close

have

\316\265.

definition of \316\265-close

that this

is clear

it

Again,

is

consistent

4.3.4.

Definition

with

real

(\316\265-close

that

Limits

Now let (an)^=mbea sequence


a real number an for every integer

real

of

i.e., we

numbers;

assign

starting indexrn

m. The

n>

this will be 1, but in some caseswe will


other than 1. (The choiceof labelused
this sequence
is unimportant; we coulduse for instance
index
to
this
and
would
as
represent exactly the samesequence
(ak)^Lm
a
We
can
define
the
notion
of
in
the
sequence
Cauchy
(an)\302\243Lm.)
same manner as before:
is

some

start

usually
integer;
from some index

Definition 6.1.3
real number.
some

integer

close

for

sequence

j,

every

integer

N >

m suchthat

Cauchy

sequence

such

that

consistent

with

\342\200\224

an'\\
the

iff

\316\265-steady.

\316\265-steady

say
for

way, a sequence(an)^=m
for

<

real

every

\316\265
for

\316\265
>

corresponding

that

Let

be a

(an)^Lm

starting

(an)\342\204\242=m

sense of Definition 6.1.3; then


real

\316\265
>

0.

In particular,

it is

it

is

Cauchy

is eventually

eventually

\316\265-steady

for

of rational

sequence

is

numbers

consistency

at some

that

\320\260

an N

for rational

is

in the

Proof. Suppose first

is

0.

>m
N. These definitions are

integer index m. Then(an)^=m


sense
sequence
of Definition5.1.8 if and
6.1.3.
Cauchy sequence in the sense of Definition
bers

an

numbers

exists

verifying
(Definitions 5.1.3,5.1.6,5.1.8),although
Cauchy sequences takes a little bit of care:

Proposition6.1.4.

\316\265
>

real

of

some

(an)\302\243Lm

every

0, there

n>nf >
definitions

all

at

\316\265-

exists

there

iff

at

are
\316\261&

starting

(an)\302\243Lm

We

starting

and

aj

0 be

\316\265
>

numbers

\316\265-steady

is eventually

sequence if,
\\an

ls

(an)\302\243Lw

iff it

it another

To put

A sequence

N.

>
\320\272

\316\265-steady

is said to be eventually

index m

a Cauchy

(an)^=N

Let

reals).

of real

said to be

N is

index

of

sequences

(Cauchy

a Cauchy

only

it is

if

for

every

in the

sequence
\316\265-steady

\320\277\320\270\321\202

for

every

rational

> 0,
\316\265

which

Cauchy sequence in the senseof Definition

it a

makes

147

laws

and limit

Convergence

J.
\316\262

5.1.8.

in the sense
sequence
(an)^=m is a Cauchy
then
it
for
is
rational
5.1.8;
Definition
eventually
\316\265-steady
every
of
then
\316\265
is
a
real
existsa
If
0
there
rational
>
number,
> 0.
\316\265
is smaller than \316\265,
5.4.14.
Since \316\265'
by
Proposition
ef > 0 which
is

that

we know

rational,

ef <

that

suppose

Now

this
\316\265,

that

implies

is

(an)\302\243Lm

is
(an)\342\204\242=m

eventually

eventually

real number, we thus seethat


the
in
sense of Definition6.1.3.
Cauchy sequence
of

Because
distinction

the

of a

concept

we talk

Now
numbers

to

this

converge

Cauchy sequence as a singleunified


about what it means for a sequence
to some limit L.

have

a sequence
N >m suchthat

\\an

is

(an)\302\243Lm

(an)\302\243Lm

real

\316\265
>

One

5.1.8

Definition

of
Definition 6.1.5 (Convergence
and
L
be
a
real
let
number.
number,
is said to be \316\265-close to L
numbers
> \316\233\316\223,
we
\316\267
i.e.,

\342\200\224<

L\\

to

converges

is

iff

\316\265
>

A sequence
iff an

is

to

L.

to

to

for every

N. We
that a
to

\316\265-close

say that
an

exists

iff there

We say

is eventually

real

{\316\261\316\267)^=\316\235

\316\265-close

every

a real

0 be

of

n>

\316\265-close

\316\265-close

it

concept.

of real

Let

sequences).

\316\265
for

eventually

(an)\342\204\242=N

is a

(an)^=m

we will no longer care about the


and Definition 6.1.3, and view

proposition,

between

\316\265
is

Since

\316\265-steady.

positive

an arbitrary

since

\316\265/-steady;

sequence

for every

the

concept

0.

all the definitions here and write


more directly; see Exercise6.1.2.

can unwrap

of convergence

Examples6.1.6.

The

sequence

0.9,0.99,0.999,0.9999,...

is 0.1-close

to 1,

but is not 0.01-close


to

1,

element of the sequence. However,it is eventually


In fact, for every real \316\265
> 0, this
sequence is
1, hence

is convergent

to 1.

because

of
0.01-close

eventually

the

first
to

\316\265-close

1.
to

6.

148

at some integer indexm,

starting

sequence

of

sequence^

be a
(Uniqueness of limits). Let (\316\274\316\267)\342\204\242=\317\204\316\267
real

6.1.7

Proposition

Limits

let

and

\320\246
\317\206
be

real numbers. Then it is not possiblefor (an)\302\243Lm


to L while also convergingto U.

distinct

two

converge

Proof.Suppose

for

converging

since

positive

is

(an)\302\243lm

that

d(an,L)

such

that

:=
\316\267

L') <

d(an,

then
\316\234),

max(iV,

to

>
\316\267

all

for
\316\265

=
\316\265

\\L

all

we

have

was

that

\316\265
is

L, we know that
is an N > m such
there is an \316\234> \321\211
if we set
particular,
to

there

Similarly,

>
\316\267

\316\265
for

note

I/|/3;

thus

L;

N.

(an)\342\204\242=m

\342\200\224

converges

(an)\302\243Lm

\316\265-close

eventually

that

contradiction

Z/. Let

Since

L'.
\317\206

<

of

sake

L and

both

to

In

M.

<

d(an,L)

\316\265
and

hence by the triangle inequality


d(L,
V)
2|L
\342\200\224
<
we have \\L \342\200\224
which
contradicts
I/|/3,
V\\
2\\L
that \\L \342\200\224
> 0. Thus it is not possible to converge
U\\
and V.

\342\200\224

I/|/3.

the fact

then

to

we know

that

L
\316\240

notation

them:

specify

Definition
to

converges
convergent and

both

to

unique, we can set up

limits are

< e,
But

d(an,Lf)

=
2\316\265

<

Now

iQ

that

6.1.8 (Limits of sequences).If


some real number L, we say that
its limit is L; we write

sequence

(an)\302\243Lm

is

(an)\302\243Lm

L = lim an
\320\277-\321\216\320\276

to

any real
and

fact.

this

denote

If a

number L, we

we leave

sequence (an)\302\243Lm
the

that

say

is

sequence

not

to

converging
is

(an)\302\243Lm

divergent

limn_,ooan undefined.

Note that Proposition6.1.7


at most one limit. Thus, the
number,otherwiseit is
if

limit

can have

a sequence

that

ensures

exists,

it is

a single real

undefined.

about

indication

index

the

is irrelevant

discussion we

as we

The notation

6.1.9.

Remark

shall

not

shall be

starting

limn_>oo

index m

(Exercise 6.1.3). Thus


be too careful as to

mostly

does

an

in

the

where

focused

on

not

of the sequence,but

their

limits.

these

rest

give any
the

starting

of this
sequences

start,

the

use

sometimes

We

149

laws

and limit

Convergence

J.
\316\262

phrase

as

\342\200\224>
\316\267
oo\"

\342\200\224>
\317\207
as

\"an

an

to \320\266\".
way of writing the statement\"(an)\302\243Lm converges
\342\200\224>
that
the
individual
statements
\320\266
and
in mind,
an
though,
Bear
this phrase is just a
not have any rigourous
meaning;
n -> oo do
a
of
course
very
though
suggestive one.
convention,

alternate

Remark 6.1.10. The


index (in this case is
exactlythe same

as

meaning

of notation;

conflicts

is
\316\267

because
\320\272

an

As

proposition 6.1.11.
Proof.

have

We

an :=

0, where

limit, we
have

We

that

show

to

1/n. In other

to change \316\267
to
some other purpose,

See Exercise 6.1.4.

confusion.

of a

example

present

limn_>oo

= 0.

1/n

the sequence
for

words,

|On

\342\200\224
<

0|

\316\265
for

\\an
if

Thus,

we

every

0| =

N >

pick

(on)\302\243Li

ls

eventually
to

11/n so

and
\316\265,

\316\265-close

to

\316\265
>

every

0, we need

\316\265-close

to

find

to

So, let

0.

to

an N

such

> \316\235,then
\316\267

if

0| = 1/n< 1/N.
we

l/\316\265(which

principle),then 1/iV <


converges

N. But

n>

to

converged
(\316\261\316\267)\342\204\242=1

show that the sequence(an)\342\204\242=1is eventually


real number. We have
\316\265
> 0 be an arbitrary
that

avoid

to

index

the

want

might

used for

being

Sometimes

instance.

labelof

we

instance,

simultaneously

to reduce

want

we

and

for

phrase

for
\316\261&,

lim^oo

to change the

be convenient

it will

the

irrelevant:

n)

to denote the
limn_>ooan has

used

letter

of

choice

exact

0.

do by

can

\316\265-close

\316\265
was

arbitrary,

{an)\342\204\242=N

Since

the Archimedean

is

to

0.

Thus

(an)\342\204\242=1

0.

6.1.12 (Convergent sequencesare Cauchy).


is a convergent
sequence of real numbers. Then
(an)\342\204\242=m

Proposition
Suppose
(an)^=m

that

is

also

a Cauchy

sequence.

Proof. See Exercise6.1.5.


Example

6.1.13.

The

sequence

\342\200\2241,1,
\342\200\2241,... is
1, \342\200\2241,1,

Cauchy sequence (because it is not


hence not a convergent sequence,by

eventually

Proposition

1-steady),

6.1.12.

not

and

is

6.

150

Remark 6.1.14.
Theorem

see

to Proposition46.1.12,

converse

sequences

below.

6.4.18

that formal limits can be superceded

we show

Now

when

subtraction

by

superceded

division

actual

actual

formal division
the rational

and
integers,
when constructing

the

constructing
actual

by

subtraction was supercededby

as formal

just

limits,

For

Limits of

numbers.

Proposition

6.1.15

that

is

(an)\302\243Li

are genuinelimits).Suppose
of rational numbers. Then

sequence
Cauchy
i-eto LIMn_,ooan;

converges

(an)%Li

(Formal limits

lim an.
LIMn_>ooan= 71\342\200\224\320\256\320\236
6.1.6.

Exercise

See

Proof

A
Definition 6.1.16 (Boundedsequences).

>
\316\267

all

for

by

is bounded

numbers

real

\321\202.We

\316\234>

is consistent

definition

This

(an)^=m

number

real

some

\316\234for

by

that

say

number

a real

sequence

\316\234
iff

(an)\302\243Lm

have

we

is bounded iff

\\an\\

<

of
\316\234

is bounded

it

0.

with Definition 5.1.12;seeExercise

6.1.7.

from

Recall

Lemma

shows

every

Cauchy

from

every Cauchy sequenceof


of the proof of that
the
same
that
argument works for real numbers;
sequence of real numbersis bounded.In particular,
5.1.15 that

Lemma
is

numbers

rational

we see

6.1.12

Proposition

An inspection

bounded.

Corollary 6.1.17.
bounded.

Every

have
sequence

convergent

Example 6.1.18. The


and hence is not convergent.
can

now

prove

the usual limit

Theorem 6.1.19 (LimitLaws).


convergent

:=
\317\207

sequences

limn_,oo

of real

an and

numbers,

:= limn_>oo
\321\203

is not

1,2,3,4,5,...

sequence

We

of real

Let

bounded,

laws.
(an)^=Tn

and

(bn)^=Tn be

and letx^y be the


bn.

numbers is

real numbers

\316\262.1.

and

Convergence

(a) The sequence(an+bn)\302\243Lm


lim (an +
(6)

T/ie

sequence

bn)

any real
ex; in other
For

lim

sequence

an)(

xy;

in other

lim

bn).

number c, the sequence(can)^=m


words,

lim (an

\342\200\224

bn)

words,

to

lim

an

\342\200\224

n\342\200\224>oo

other

in
x\342\200\224y;

lim
n->oo

the sequence

> m.
and that bn \317\2060 /or all \316\267
to
in
other
words,
(b~1)r^Lmconverges y~l;

b\"1

lim

(lim

words,

bn.

that j/^0,

Suppose

to

converges

lim
\321\201
an.
\320\277-\321\216\320\276

converges

{an\342\200\224bn)n%zm

n->oo

(e)

to

lim

lim (can) =

T/ie

an + lim bn.

\317\200\342\200\224\321\216\320\276
n->oo

\320\277-\321\216\320\276

(d)

words,

71\342\200\224\320\256\320\236
\320\237-\320\256\320\236

converges

(anbn)n\302\260=m

in other

to x+y;

converges

lim (anbn) =
(c)

151

laws

limit

Then

bn)-1.

\320\237-\320\256\320\236 71\342\200\224\320\256\320\236

(/)

that

Suppose

0,
\321\203
\317\206

and

iftai

bn

(an/bn)\302\243Lm converges

\316\257/ie
sequence

0
\317\206

for

to x/y;

all n

> m.

Tften

in other words,

On _ limn-\321\216\320\276
an
\320\243
\317\200-\321\216\320\276
bn
bn
limn-,\316\277\316\277

(5)

T/ie

sequence

(max(an,bn))r^=m

converges to

in

\321\202\320\260\321\205(\320\266,\321\203);

\316\277\316\257/ier
words,

lim

max(an,bn)

71\342\200\224\320\256\320\236

(/&)

The

sequence

other

words,

= max(

lim an, lim bn).

71\342\200\224\320\256\320\236
71\342\200\224KX)

(min(an,bn))\302\243Lmconverges

lim min(an,bn)

Ti\342\200\224\302\27300

to

in
\321\202\321\202(\320\266,\321\203);

= min( lim an, lim bn).


71\342\200\224\320\256\320\236
71\342\200\224\302\27300

Limits

6.

152

of sequences

Proof. See Exercise6.1.8.

of real numbers, such that


Let (an)\302\243L0 be a sequence
\316\267
that
whenever
for each natural number n. Prove
and
\321\211
such that m> n, then we have am >an.
numbers
(We refer

6.1.1.

Exercise

> an
\316\261\316\267+\316\271
are

natural

to these sequencesas increasingsequences.)

6.1.2.
Exercise

Let

a real

any

(an)\302\243lm

be

numbers, and let L be

of real

a sequence

to L if and only
number.
Show that (an)\302\243Lm converges
>
N
such
can
find
an
m
that \\an \342\200\224
real \316\265
> 0, one
L\\ <

if,

given

\316\265
for

all

n>iV.

Exercise 6.1.3.
number,
to

Exercise 6.1.4.
to

or

>
\320\272

if and
\321\201

Prove

6.1.5.

Exercise
inequality,

numbers,

Show that

let

(an)\302\243JLm

of real numbers,
let
a non-negative integer. Show that
only if (an+fc)$jjlm converges to \321\201

Proposition

\321\201
be a

real

converges

c.

0 be

6.1.12. (Hint:

use the

\321\201
be a

real

(an)\302\243Lm

triangle

4.3.7.)

Proposition

Exercise 6.1.6.

to

converges

(an)\302\243im/

of real

an integer.

m be

Let (an)^Lm be a sequence

and let

converges

>

ml

only if

\321\201
if and

number,

Let (an)^=mbe a sequence

and let

Prove Proposition6.1.15,using

the following
outline.
Let
and write L := \320\253\320\234\320\277^^\320\271\320
We
have to show that (an)\302\243JLm converges
to L. Let \316\265
> 0. Assume
for
sake of contradiction that
an is not eventually \316\265-close to L.
sequence
to show that there is an
that
Use this, and the fact
(an)\302\243Lm is Cauchy,
> iV, or an < L \342\200\224
N > m such that either an > L + \316\265/2
for all \316\267
\316\265/2
>
for all \316\267 N.
Then use Exercise 5.4.8.

(an)^=m be a Cauchy

sequence

Exercise6.1.7.Show

that

5.1.12

(i.e.,

instead of

Definition

prove an analogue of
Cauchy

of rationals,

6.1.16

Proposition

is consistent with Definition


6.1.4 for bounded sequences

sequences).

Exercise
6.1.8. Prove Theorem 6.1.19.(Hint: you can use some parts of
the theorem to prove others,
can
e.g., (b) can be used to prove
(c);
(a),(c)
can
be used to prove
and
be
usedto
The
prove
proofs
(b), (e)
(d);
(f).
are similar to those of Lemma
5.3.10, and Lemma
5.3.6,
Proposition
that
5.3.15. For (e), you may need to first prove the auxiliary
result
a
are
and
which
to
whose
elements
non-zero,
any sequence
converges
non-zero limit, is bounded
from
away
zero.)
of
Theorem
fails when the limit
Exercise 6.1.9. Explain
why
6.1.19(f)
the denominator
is 0. (To repair that
rule^
requires
problem
L'HopitaUs
see

Section

10.5.)

\316\262tg.

Exercise6.1.10.Show
as

\320\260\320\263\320\265
seQuences
(&\320\273)\321\202\302\243=\320\276

for

g-close

every

rational
0.

\316\265
>

is
\316\265

and

(an)\302\243i0

if and only if

\316\265
> 0

modify the

(Hint:

Cauchy sequence,

to be
required
precisely, if (an)$JL0 and

change if
More

show that

of reals,

every

real

of equivalent

concept
rational.

positive

153

system

5.2.6, does not

positive real instead of


\316\265-close for
eventually

the

that

Definition

in

defined

number

real

extended

The

are

(6n)^Lo

they are

proof of

eventually

Proposition

6.1.4.)

Thereare
number,

For

some

seem
it seems

instance,

system
converge to any

do not

which

sequences

instead

but

real number

extended

The

6.2

to be wanting to
intuitive that the

real

convergeto +ooor
sequence

\342\200\224oo.

1,2,3,4,5,...
be

should

to +oo,

converging

while

-1,-2,-3,-4,-5,...
should be converging

to

the

\342\200\224oo.
Meanwhile,

sequence

1,-1,1,-1,1,-1,...

does not seem to be converging

later that it does


+1
Similarlythe sequence
have

to

and

(although

anything

\342\200\2241
as

\"limit

points\"

we shall see
- see
below).

1,-2,3,-4,5,-6,...
does

not

converge

to any

real number,

and alsodoesnot appear

make
to +oo or convergingto \342\200\224oo.
To
to talk about somethingcalledthe extended

be converging

this

we need

real

to

precise
number

system.

Definition
6.2.1 (Extended real number
real number system R* is the realline
elements

attached,

from each

called

and
+\317\214\316\277

system).

with

\342\200\224oo.
These

other and alsodistinctfrom

every

The

extended

two additional
are distinct
elements
An
real
number.

6. Limits of sequences

154
number

real

extended

infinite

is

it

iff

directly

it is

though

3.6,

since

meaning,
than

Definition 6.2.2
of negation
\342\200\224(+00)

:=

Section

spirit.)

extended

of

:=

The

reals).

now extend

we

R,

operations

system.

(Negation

\342\200\22400
and

to R*

operation
by

defining

+00.

\342\200\224(\342\200\22400)

has
extended real number\317\207

a negation,

and

\342\200\224(-\317\207)

to x.

equal

always

and infinite sets in

^). Now we placea few

inequality

\342\200\224
\316\271-\302\273
\317\207
\317\207
on

Thus every
is

real number,
and
definition
is not

is a

it

\342\200\22400.
(This

in

real number

extended

the

on

or

iff

of finite

notion

of course similar

= and

equality

finite

at present
do not have much
symbols, +00 and \342\200\22400,
we have no operations to manipulatethem (other

new

These

to +00

equal
the

to

related

called

\317\207
is

6.2.3 (Orderingof extendedreals).Let \317\207


and
be
\321\203
< \321\203,
extended
\317\207
is less
real numbers.
We say that \317\207
than or
i.e.,
equal to y, iff one of the following three statements is true:
Definition

and

(b)

=
\321\203 +\320\276\320\276.

(c)

= \342\200\22400.
\317\207

We

write

are
\321\203

that

say

\317\207
as
< \321\203

Some

basic

number

numbers,

\317\207
if we
< \321\203

and
> \320\266,
\321\203

6.2.4.

Examples

real

real

(a)

3 <

have

and

< \321\203
\317\207
and

< \321\203
as
\317\207

of order

properties

numbers.

We
sometimes
\321\205\321\204\321\203.

>
\321\203 \317\207.

but 3

\342\200\22400.

and negation on the extended

system:

6.2.5. Let x, y, \316\266


be
the following statementsaretrue:
(Reflexivity)

real

< +00,
5, 3 < +00, and \342\200\22400

Proposition

(a)

< \321\203
\317\207
as

We

have

<
\317\207

x.

extended

real

numbers.

Then

2.
\316\262

The

real number

extended

(b) (Trichotomy)Exactly
(c) (Transitivity)
Id) (Negation

the

statements

and

If\317\207<y

<

\321\203

z, then

reverses order) If \317\207


<y,

\317\207
<

=
\317\207

y,

y,

<
\317\207

z.

<
\342\200\224y\342\200\224x.

then

6.2.1.

Exercise

See

of

true.

orx>yis

Proof

one

155

system

on the
extended
One could also introduceother
operations
as
such
etc.
number
addition,
system,
multiplication,
real
as these operations will almost
is somewhat
dangerous
However, this
fail to obey the familiar rules of algebra. For instance,
certainly
addition
it seems reasonable (given one's intuitive
notion
define
to
=
=
to set +oo + 5
+oo and +oo + 3 +oo, but then
of infinity)
=
that
5
So things like
+oo
+
+oo
+ 3, while 5^3.
this
implies
law
to
break
down
once
we
cancellation
the
try to operate
begin
To avoid these issues we shall simply
not
define
infinity.
involving
on
the
extended
real
number
arithmetic
operations
system
any
and order.
than negation
other
the
of supremum or least
notion
Remember that we defined
of
a
of
bound
set
\316\225
this
reals;
gave an extended real number
upper
was
finite
which
either
or
infinite. We now extend
this
sup(i\302\243),

notion slightly.

6.2.6 (Supremumof setsof extended


of R*. Then we define the supremum

Definition
a subset

be

of

upper bound

(a) If

\316\225
is

of E),
(b) If
(c)

If

\316\225
by

following

(i.e., +oo and


then we let sup(E) be as defined

\316\225
does

set sup(E)
falls

+oo,
not

contain

:= sup(E
under

case

we set

then

+oo

sup(E)

\316\225

least

\342\200\224oo
are

in

not

elements
5.5.10.

Definition

sup(E) := +oo.

but does

\342\200\224
{\342\200\224oo}) (which

(a)).

or

rule.

in R

contained

\316\225
contains

thus

the

Let

reals).

contain
is a

\342\200\224

oo,

subset

then

we

of R

and

6. Limits of

156

the

define

also

We

infimum

by
greatest lower boundof \316\225

m\302\243(E)

of

sequences

known

\316\225

(also

as the

formula

the

:=-sup(-S)

inf(S)

:= {\342\200\224\317\207
: \317\207
where -i? is the set \342\200\224\316\225
\342\202\254
\302\243?}.

Let

6.2.7.

Example

the

be
\316\225

together

integers,

negative

with

\342\200\224oo:

=
\302\243

Then sup(u') =

{-1,-2,-3,-4,...}U{-oo}.

inf(E)

does

not

from

the

-oo.

right.

6.2.9. The set

supremum +oo.

Example 6.2.10.

{1, 2, 3,

be
\316\225

Let

+oo
(why?).
\316\212\317\204\316\271\316\220(\316\225)

supremum
One

-(+c\302\273)

The set {0.9,0.99,0.999,0.9999,...}has


1. Note that in this casethe supremum
supremum
belong to the set, but it is in somesense
actually

Example

and

-sup(-S)

and

0.9

it\"
\"touching

while
\342\200\2241,

6.2.8.

Example
infimum

\342\200\224
=
{\342\200\224oo})

sup(\302\243'

can be less

than

can intuitively

the

4,

empty

This is
the

Then sup(E)

set.

the

infimum

only

case

in

which

\342\200\224oo

the

(why?).

think of the supremumof \316\225as

the real line with +oo somehow


Imagine
on
the
far left. Imagine a piston at

1 and

infimum

has

5...}

on

+oo

the

moving

far right,
leftward

follows.

and
until

\342\200\224oo

it

where
it stops
by the presenceof a set E\\ the location
is the supremumof E. Similarly
a
if one
piston at \342\200\224oo
imagines
until
it is stopped by the presenceof E, the
rightward
moving
is
location
where
In the case when \316\225
it stops is the infimum
of E.
the empty
the
the
each
set,
other,
pistons pass through
supremum
at \342\200\224oo
the infimum
and
landing
landing at +oo.
The following
theorem
the terminology
\"least upper
justifies

is stopped

bound\" and \"greatestlower


Theorem

6.2.11.

statements

are

true.

Let

\316\225
be

bound\":

a subset

o/R*.

Then the

following

and

\316\262.3. Suprema

(a) For every

\317\207
\316\225
we
\342\202\254

that

Suppose

(b)

all

\317\207
e

that

Suppose

(c)

\317\207
e

\316\234is

upper

and

sup(E)

>
\317\207

E, i.e.,

bound for

sup(E) < M.

E, i.e.,

bound for

\316\212\317\204\316\271\316\257(\316\225

<
\317\207

\316\234for

> \316\234
\317\207
for

all

'mi(E) > M.

we have

Proof.See

6.2.2.

Exercise

6.2.1. Prove

Exercise

an

a lower

\316\234is

Then

E.

<
\317\207

have

we have

Then

E.

157

of sequences

infima

Proposition 6.2.5. (Hint: you

need

may

Proposition
5.4.7.)

Prove Proposition 6.2.11. (Hint: you


on whether +oo or \342\200\224oo
belongs
depending

6.2.2.

Exercise
into

cases

of course use Definition


numbers.)

6.3

of a

notion

the

defined

reals, we can

to

E.

\316\225
consists

and infima of sequences

Suprema

Having

that

provided

5.5.10,

to break
You can
only of real

need

may

now also talk about

supremum and
the

sets

of

infimum

of

of

infimum

and

supremum

sequences.

Definition 6.3.1
sequence

of real

inf

and

(Sup

numbers.

>
supremum of the set {an : \316\267
>
of the same set {an : \316\267
m}.

Remark

sometimes
Example

6.3.2. The
written as
6.3.3.

-1,1,-1,1,

element set

m},

and

Let an := (\342\200\224l)n;thus
Then the set {an :

inf(an)^=1is equalto

hence
\342\200\2241.

sup(an)^=1

to

inf(an)\302\243Lm

to

the

be

\320\260

the

infimum

inf(an)\302\243Lm are

and
an

be

(an)\302\243Lm

define sup(an)\302\243Lm

quantities sup(an)\302\243Lm
supn>man and infn>m

and
{\342\200\2241,1},

Let

of sequences).

Then we

respectively.

is

the

sequence

{\316\274\316\267)\342\204\242=\316\271

>
\316\267

1}

the twoto 1. Similarly

is just

is equal

Limits

6.

158

6.3.4. Let

Example

:=

an

1,1/2,1/3,

1/n;

thus

Thus sup(an)~=1
{1,1/2,1/3,1/4,...}.
(Exercise 6.3.1). Noticehere that the
not actually a memberof the sequence,
close to the sequenceeventually.
(So
of the

think

and

supremum

:=

Let an

6.3.5.

thus
\321\211

it becomes

though

the

is

of a

(an)^=m is

sequence to be +ooor

bounded,

bounded

say

an of the sequencelie between


has

m}

inf

an

\316\234as

upper

by

then

M,

\342\200\224
\316\234and

bound

and

supremum

(i.e., not +oo

of

infimum

and

a bounded

sequence

the

M,

and

so

1.
or

supremum
if a

sequence

all the elements


that the set {an :

\342\200\224
\316\234as

sequence

1, 2,

integers

(an)\342\204\242=1

\342\200\224oo.
However,

set is clearly non-empty, we can thus

Since this

a lower

bound.
that

conclude

are real

the

numbers

\342\200\224oo).

(Least upper bound property). Let (an)\302\243Lm


real
real
be the
extended
numbers, and let \317\207
of
all n>m.
Then
we have an < \317\207
for
Also,
sup(an)\302\243im.
6.3.6

Proposition
be

the

the positive

last example shows, it is possiblefor

As the

>
\316\267

respectively.)

(\316\261\316\267)^=1

\316\267

and

infimum

very

little inaccurate to
\"largest element of the

is a

it

>
:
Then the set
3, 4,
1} is just
=
Then
+oo
{1,2,3,4,...}.
sup^n)^
{an

set

inf(an)~=1 = 0
of the sequence is

sequence\" and \"smallestelementof the sequence\"


Example

sequence

countable

1 and

infimum

as

infimum

is the

1}
=

the

is

(an)\302\243Li

set {an : \316\267>

Then the

of sequences

a sequence

:=
number \317\207

whenever
>
\316\267

\316\234G

we have

m),

number \321\203
for

which

is an

R*

<
\317\207

<
\321\203

upper bound for

\316\234.

there
\317\207,

Finally,

exists

an (i.e., an

<

\316\234
for

all

for
every extended real
> m for
at least one \316\267
which

<an<x.
\321\203

See Exercise

Proof

Remark 6.3.7. Thereis a corresponding


Proposition
but with all the referencesto order
all
reversed,
e.g.,
should now be lower bounds,etc. The proofis exactly
Now
and

for

upper
the

infima,

bounds
same.

an application of these conceptsof supremum


In the previous section we saw that all convergent
bounded.
It is natural to ask whetherthe converse

we give

infimum.

sequences

6.3.2.

are

is

the

instance,
and

convergent? The answer is no; for


is bounded, but not Cauchy
1,-1,1,-1,...
if
we make the sequence
both
convergent.
However,
or decreasing), then it is
monotone
increasing
(i.e.,

bounded

all

are

true:

bounds

and

it must

that

true

converge:

6.3.8 (Monotone boundedsequences


Let
converge).
has
some
real
numbers
which
of
finiteupper
> an for
and which is also increasing (i.e., an+\\

proposition

\320\260
^\320\265
sequence

lan)r^=m

MgR,

bound

sequences

sequence

not

hence

Then

is

(an)\302\243Lm

alln>\342\204\242>)\302\267

lim an

<

One can similarly


below and

that

if a

sequence

(i.e.,
to the infimum.
to be monotone

if

decreasing. Prom Proposition6.3.8and


that

a monotone

and

increasing,

it is

bounded

is either

increasing
6.1.17

Corollary

only

and

convergent,

if

it

or

we see

is bounded.

The sequence 3,3.1,3.14,3.141,3.1415,...is


above by 4. Hence by Proposition
6.3.8

is bounded

have a

it must

it

if and

converges

sequence

6.3.9.

Example

is

(\316\261\316\267)\342\204\242=\317\204\316\267

< \316\261\316\267),
then
\316\261\316\267+\316\271

is equal
is said

sequence

prove

decreasing

limit

the

that

M.

6.3.3.

Exercise

See

in fact

and

convergent,

sup(an)\302\243lm
\316\267\342\200\224>\316\277\316\277

Proof

159

of sequences

and infima

Suprema

3.

limit, which is a realnumber

less

than

or equal

to

4.

Proposition 6.3.8assertsthat
exists,but does not directly
say
with
given

that

the

what

limit

that

monotone sequence
limit is. Nevertheless,
the
limit
once one is

of a

one can often find


does exist. For instance:

work

extra

little

limit

the

Proposition 6.3.10.
0.

Let

\320\241
< \317\207
<

1.

Then we

have limn_,ooa:n =

can show that the sequence


is
(xn)^=i
other
the
has
a
hand,
decreasing
sequence(\317\207\316\267)\342\204\242=1
(why?).
lower bound of 0. Thus by Proposition
6.3.8
(for infima instead of
to
the
some limit L. Since
sequence(xn)\342\204\242=i converges
suprema)
Proof

Since

0 <

\317\207
<

1, one

On the

6. Limitsof

160
_

sequences

foe li^t law;s (Theorem


6.1.\321\211
to xL.
that
But the sequence
(xn+1)\342\204\242=1 converges
is
(\321\217\342\204\242-1)^
the sequence
shifted
one, and so they must have
just
(\317\207\316\267)\342\204\242=2 by
the
same
limits
1, we can solve for
\317\206
(why?). So xL = L. Since\317\207
to 0.
L = 0. Thus (zn)\302\243Li converges
L to obtain
q
\317\207
\317\207
we
iyxus
\317\207\316\267+\316\271
\317\207\316\267^

s6e grom

this proof does not

Note that

when

work

\317\207
>

(Exercise

6.3.4).

Exercise 6.3.1.

Exercise 6.3.2. Prove

6.3.6.

Proposition

Exercise6.3.3.Prove Proposition
together with the assumption
to

converges

6.3.4.

in Example

claim

the

Verify

6.3.8.

6.2.11.)
6.3.6,

use Proposition

(Hint:

increasing, to

an is

that

use Theorem

(Hint:

show

that

q^

sup(an)\302\243Lm.)

\317\207
6.3.4.
6.3.10 fails when
> 1. In
Explain
why Proposition
1.
the
\317\207
that
when
>
show
sequence
fact,
diverges
(Hint: prove
(\317\207\316\267)^=\316\271
= 1 and the limit
laws
by contradiction and use the identity
(l/x)nxn
in
with
the
this
in Theorem
Example
argument
6.1.19.) Compare
1.2.3;
can you now explain the flaws in the reasoning in that example?

Exercise

6.4

Considerthe

plots this sequence,then

is getting closeto
to

-1.0001,1.00001,....

-1.01,1.001,

that this sequencedoesnot


close

points

sequence

1.1,

If one

limit

and

liminf,

Limsup,

-1, but

it is

1,

the

sees
half

converge;
half

and

one

time

of course)

(informally,
the time

the

the sequence
sequence is getting

not convergingto eitherofthem;

for

instance,

eventually
gets eventually 1/2-close to 1, and never
gets
not quite
-1.
-1
and
are
to
even
+1
However,
though
1/2-close
of this sequence,
limits
it does seem that in somevague
they
way
the
we
introduce
\"want\" to be limits. To make this notionprecise
notion
of a limit point
it never

Definition 6.4.1 (Limit points).Let


real

numbers,

let

\320\266
be

real

number,

be

a sequence

(\320\260\320\277)\342\204\242=\321\202

and let

\316\265
>

be

of

a real

and limit

liminf,

Limsup,

L
\316\262

that

\317\207
We
is
say
\320\264\321\211\320\277\320\254\320\265\320\263.

n > m such
tf>m-

is continually

(an)\342\204\242=m

{o>n)%Lm

such that

n>

tfj

\\an

for

much

that

see

is this

L) and L being\316\265-adherent

a single elementof the


Also,

L).

for L

to be

\316\265-adherent

to be

eventually

to

the

to be continually \316\265-adherent
to

for

all

to

every

0.

as

the same

a limit

point

m, there exists

an

\316\265-close

within

definition?)
to

(which

a distance

to

(an)\302\243Lm,

needs
\316\265
of

it

has

m, whereas for (an)\302\243lm


need (an)\342\204\242=N to be \316\265-close

N >

(\316\261\316\267)\302\2431\316\235

\316\265-close

for

sequence
(which only
a distance
within
stay

to

sequence

\320\266
is

the same

being
Note the differencebetweena sequence
all
the
elementsofthe
that
sequencestay
means

e of

\316\265
>

every

every N >

\316\265.
(Why

x\\

(an)\342\204\242=N

means

we

0 and

\316\265
>

\342\200\224
<

continually

\"adhesive\".

Unwrapping all the definitions,


of

\317\207
is

adherent point of (an)\302\243Lm

adhere\"

\"to

to

\316\265-adherent

to

the term

^\316\237\316\223
every

say that

point or

limit

\316\265-adherent

hence

to\";

stick

\320\266
is a

The verb

6.4.2.

Remark
\"to

that

an

exists

there

iff

(\320\260\320\277)\342\204\242=\321\202

We

x.

to

is

it

iff

(an)\302\243Lm

We say

iff it

\316\265-adherent to

that an is \316\265-close

to

^adherent

161

points

L, we only

> m. Thus there are some subtle differences


in
between
and
limits
limit
points.
quantifiers
that
limit
numbers.
are only defined for finite
real
Note
points
the concept
define
of +oo or \342\200\224oo
It is also possible to rigourously
see
a
limit
Exercise
6.4.8.
point;
being
to

L for

some N

Example 6.4.3.

Let

(an)^=1

denote

the

sequence

0.9,0.99,0.999,0.9999,0.99999,....

The number 0.8is 0.1-adherent


to thissequence,
since
0.8 is 0.1close to 0.9, which is a memberof this sequence.
it
However,
is not continually 0.1-adherent to this sequence,
sinceonce one
of
discards
the first element of this sequencethereis no member
the sequence
to be 0.1-close to. In particular,0.8is nota limit
1 is 0.1of this sequence.
On the other hand, the number
point
adherent to this sequence, and in fact is continually
0.1-adherent
to this sequence, sinceno matterhow many
members
of the
initial
for
to
be
0.1-close
one
1
thereis
still
sequence
discards,
something

6. Limits of sequences

162
In

to.

it is

fact,

limit point of

continually

for

\316\265-adherent

the

consider

Now

sequence

-1.0001,1.00001,....

1.1, -1.01,1.001,

tothissequence;
The number1is 0.1-adherent
this

to

continually 0.1-adherent

sequence,

matter

sequence one discards, there are some


that 1 is 0.1-closeto. (As discussed
earlier,

sequence

to
all the elementsto be0.1-close
1,

not need

adherent is weakerthan
a different notion from
\316\265
>

it is

fact

in

because no

of the

elements

and

0.1-close,

how

many

elements

of

one does

some;

just

thus 0.1-

0.1-adherent

continually
In

0.1-close.)

eventually

1 is

number

the

0,

this sequence.

Example 6.4.4.

the

is hence

and
\316\265,

every

continually \316\265-adherent

fact,
this

to

is

for every
sequence,

-1 is a limit
thus a limit point of this sequence.Similarly
not
a
limit
is
0 (say)
point of this
point of this sequence;however
it is not continually 0.1-adherent to it.
since
sequence,

and is

Proposition 6.4.5

case of limit points:

a special

of course

are

Limits

limit

are

(Limits

Let

points).

(an)\342\204\242=m

sequence
converges to a real numberc. Then
and
in fact it is the only limit point
point \320\276/(\320\260\320\277)\342\204\242=\321\202,

which

See

Proof.

(lim

sup)

by

(an)\302\243Lm

the

is

D
types of limit points: the limit
inferior (lim inf).

and limit

Definition 6.4.6
that

We define

a sequence.

limit

inferior).

Suppose

a new sequence(ajv)^=m

formula

informally,

sequence
the

and

superior

(Limit

a+ :=
More

of(an)\342\204\242=m.

two special

at

look

be

a limit

6.4.1.

Exercise

Now we will
superior

\321\201
is

sequence

from

a^

is the

onwards.
\316\261\317\207
(on)\302\243Lm,

denoted

sup(an)\302\243lN.

supremum of all the


We

then

define

limsupn_>00

in

the

the limit superior


an, by the formula

lim sup an := mf(a^)f=m.


\320\277-\321\216\320\276

elements

of

\316\262

the limit inferior of the


by the formula

and define

sequence

(an)\302\243im,

denoted

an,

liminfn-oo

lim
Let

6.4.7.

Example

:=
inf
an
n\342\200\224\321\216\320\276

sup(a^)^=m.

aj,

a]1\",

is the

aj,...

the

denote
\320\260\320\267,...
\316\261\317\212,
\316\2612,

sequence

-1.0001,1.00001,....

1.1, -1.01,1.001,

Then

163

points

define

can

we

Similarly,

and limit

liminf,

Limsupj

sequence

1.1,1.001,1.001,1.00001,1.00001,...
and

(why?),

its

sequenceis 1.

is 1.

infimum

Similarly,

aj^,

a^\",

Hence the limit superior of this


is the sequence
a^\",...

-1.01, -1.01, -1.0001,-1.0001,


-1.000001,...
and

(why?),

the

of this

supremum

1. Hence
sequence is \342\200\224

this sequenceis \342\200\2241.


One
should
of the
sequence,
supremumand infimum

inferior of

compare

are 1.1

which

the

limit

this with

and

the

\342\200\2241.01

respectively.

6.4.8.

Example

Let

denote
\316\261\317\212,
a^ \320\260\320\267,...

the

sequence

1,-2,3,-4,5,-6,7,-8,...

Then a*,

aj,... is the

sequence

+oo,

+oo,

+oo,

+oo,...

(why?) and so the limit superioris +oo. Similarly,


the

sequence
\342\200\224
\342\200\224oo,oo,
\342\200\224oo,
\342\200\224oo,...

and

so

the

limit

inferior is

\342\200\224oo.

\320\260]\",\320\260^,...

is

Limits

6.

164

denote
6.4.9. Let \316\261\317\212,
\316\2612,
\320\260\320\267,...

Example

the

of sequences

sequence

1,-1/2,1/3,-1/4,1/5,-1/6,...

Then

\320\260*,
\320\260^\",...

the

is

sequence

1,1/3,1/3,1/5,1/5,1/7,...

which has an
Similarly,

is the

a^\",...

aj^,

of

infimum

so the limit

(why?),

superior is 0.

sequence

-1/2,-1/2,-1/4,-1/4,-1/6,-1/6

which has a supremum

0.

of

Example6.4.10.

limit inferior

So the

the

denote
\316\261\317\212,
a^ \320\260\320\267,...

Let

is also 0.

sequence

1,2,3,4,5,6,...

Then a*, a^,... is the sequence

+oo,+oo,+oo,...

so

limit

the

Similarly, aj^, a^\",... is

is +oo.

superior

the sequence

1,2,3,4,5,...
of +oo.

a supremum

has

which

Remark 6.4.11.
of

an,

limsupn_>00

and

notation
instead

limn_>ooan
of liminfn_,oo

\320\250\320\237\321\203^^\320\260\321\202\302\273

the starting indexm

Note that

the

use

authors

Some

instead

So the limit inferior is also +oo.

of

the

sequence

is irrelevant

an.

(see

Exercise 6.4.2).
to

Returning
moving

which

element
leftward,

in

our

from
\316\261\317\212
to

piston

analogy,

imagine a

piston at +oo

stopped by the presenceof the sequence


\316\2612,
\320\260\320\267,...,
place
stop is the supremumof \316\261\317\212,
new notation is af. Now
let
the first
us remove

until

leftward

\342\200\224
The
\316\261\317\212,
\316\2612,

the

a new

it is

it will

cause our piston to slip


in
point aj (though
many cases the pistonwill

the

sequence;

this may

\316\271
\316\262 Lirnsup,

165

limit points

and

liminf,

be the same as af). Then


remove
we
If
a2, causing the piston to slipa littlemore.
the piston will keep slipping,
but there will be
it cannot
go any further, and this is the limit
somepoint where
A
similar
of the
sequence.
analogy can describethe limit
superior
and

move

not

will just

a\\

the second element


we keep doing this

the

of

inferior

sequence.

describe some basic propertiesof limit

We now
limit inferior.

and

superior

6.4.12. Let (an)^=mbe a sequence


of real numbers,
Proposition
let
L~
limit
and
the
be the
be
this
L+
superior of
sequence,
let
L+
and L~~ are extended
both
limit inferior of this sequence(thus
real numbers).

(a) For

\317\207
>

every

exists

there

L+,

an N

>m such that

an

N.
>
L+,
(In other words, for every \317\207
less
of the sequence (an)^=mare eventually
< L~ there exists an N >m
Similarly, for every \321\203
N.
all n>
an>
for
\321\203
all

for

<

\317\207

the

>
\316\267

elements

than

x.)

such

that

(b) For

\317\207
<

every

L+,

and every

N >m, there existsann>

an > x.
L+,
(In other words, for every
\317\207
elements of the sequence (an)\302\243Lm exceed
infinitely
often.)
>
L~
N
and
there
exists
>
m,
Similarly,
for every \321\203
every
such

ann> N

(c) We have
(d)

If

is
\321\201

the

\317\207
<

that

that

such

an
<

mf(an)\342\204\242=m

any

limit

point

< y.

< L+

L~

of

<

(an)\342\204\242=m,

sup(an)\302\243Lm.

then

we

have

L~ <

<
\321\201

L+.

(e)

IfL+
if L~

(/)

Let

is finite, then it is a limit point of(an)^==m.


Similarly,
is finite, then it is a limit point of (an)\342\204\242=m.

\321\201
be

must

have

(an)\302\243Lm

a real
L+
converges

number.

If

(an)\342\204\242=m

\342\200\224
\342\200\224
L~
\321\201
Conversely,

to

\321\201

converges

if L+

to

c, then

we

\342\200\224
\342\200\224
L~
then

c,

Proof.
to the

shall

We

exist

and (b),

(a)

prove

Suppose first that

exercises.

>
L+, we have \317\207
then

Limits

6,

166

inf(a^)^=m.

N >

an integer

By

and leave tl^e


\317\207
>

\317\207
>

parts

by definition of

6.3.6, there

Proposition

m suchthat

remaining

Then

L+.

of sequences

By

cl^.

must

definition

of

Thus
> s\\xp(an)\342\204\242=N.
by Proposition
6.3.6
a]^, this means that \317\207
>
\316\267
as
desired.
This proves the
> an for all
iV,
again, we have \317\207
first part of (a); the secondpart of (a) is proven similarly.
\317\207
Then we have
Now we prove (b). Supposethat
< L+.
\317\207
If we fix any N > m, then
< inf(a^)^=m.
by
Proposition
definition
this
we
thus
have
\317\207
of
<
means that
6.3.6,
a]y,
\316\261^. By
\317\207
6.3.6 again, there must
< sup(an)^)=N.
thus
By Proposition
> N
This proves the first
exist
\316\267
such that \316\237\316\271\316\263\316\271
^* X) as desired.
of (b), the second part of (b) is proven similarly.
part
The proofsof (c), (d), (e), (f) are left to Exercise 6.4.3.
Q

6.4.12 say, in particular,


Parts (c) and (d) of Proposition
that
L+ is the largestlimit point of (an)\302\243Lm, and L~ is the smallest
limit point (providing that L+ and L~ are finite.
Proposition
L~
L+
there
6.4.12
then
that
if
and
coincide
is only
says
(so
(f)
limit
then the sequence in fact converges. This gives
one
point),
a way
to test if a sequence converges: computeits limit
superior
and

limit

inferior,

and see

if

are

they

equal.

give a basiccomparison
property
limit inferior.
We now

of

Lemma 6.4.13 (Comparison


principle).Suppose
and

for

(bn)\302\243lm

alln>m.

two

are

of real numbers

sequences

Then we

have

sup(an)\302\243lm

inf(an)-=m

lim sup

the

limit

superior

that

such that

and

(an)\302\243Lm

an

< bn

inequalities

<
<

sup(bn)\302\243lm

inf(bn)-=m

an < lim sup

bn
n\342\200\224\321\216\320\276 \320\277-\321\216\320\276

lim

Proof

See

Exercise

6.4.4.

inf

an <

lim inf

bn

g.j. Limsup, liminf,

be sequences

(cn)n^m

for

>
\316\267

all

that

also

Suppose

limit L.

the same

to

converge

\316\234.

167

points

(Squeeze test). Let (an)\342\204\242=m,


of real numbers such that

6.4.14

Corollary

limit

and

and

(bn)\302\243Lm;

and {cn)^m

(an)^=m

Then (bn)^=m

also

is

both

to

convergent

L.

Proof. See Exercise6.4.5.

that
We already know (see Proposition
the
limit laws (Theorem 6.1.19), this
\342\200\224 =
= 0 and limn_>oo
0. The squeeze

6.1.11)

6.4.15.

Example

Vn

limn-Kx>

impliesthat
test then

0\302\267
By

2/n

limn_,oo

that

shows

to 0.

sequence
to zero. Noteone
for

all

>
\316\267

(bn)\342\204\242=1

2/n for

<bn<

For instance,we

>
all \316\267

use

this

to show

that the

2~n < 1/n

induction

use

can

can

which

to zero, or that 2\"n converges


to show that 0 <

converges

1/n2

(\342\200\224l)n/n+

1.

with
limit
The squeeze t$st, combined
the
monotonebounded
that
sequencesalways
principle
allows to compute a large numberof limits.We
limits,
give
in
next
the
chapter.
examples

6.4.16.

Remark

and the

One

used

commonly

for

(Zero

to zero if

equal to

zero.

and

only

if

the

squeeze test

limit

limit

close
6.1.12.

this

section

some

be a

and
exists
\320\260\320\277
\320\230\321\202\320\277_,\320\276\320\276

limn_,oo

\\an\\

exists

Proof. See Exercise6.4.7.


We

laws
have

is

Let (an)^LM

sequences).
the

Then

is equal

of the

consequence

test
Corollary 6.4.17
real
numbers.
sequenceof

Proposition

also

2/n

for

sequence

any

-2/n
is convergent

with the following improvementto

and

is

6.

168

ofthe reals).
Theorem 6.4.18 (Completeness
is
real
a
and only
numbers
sequenceif
of
Cauchy

Limits

of sequences

A sequence

(an)\302\243L,

is convergent.

if it

similar in spirit
this
is very
6.4.19. Note that while
Remark
to Proposition 6.1.15,it is a bit more
since
general,
Proposition
of real
6.1.15 refers to Cauchy sequencesof rationaleinstead
numbers.

6.1.12

Proposition

Proof

so it suffices

is Cauchy,

sequence
is

sequence

Let

every

convergent

to show that

every

Cauchy

convergent.
a Cauchy

be

(an)\342\204\242=1

the sequence

that

6.1.17

that

tells us

already

We know

sequence.

is
(an)\342\204\242=1

from

Corollary

Lemma

by

bounded;

6.4.13

:=
L~
(or Proposition 6.4.12(c))this implies
L+ := limsupn_>00
and
To show
an of the sequenceareboth
that the sequence converges, it will suffice by Proposition6.4.12(f)
that

liminf\316\267_>\316\277\316\277
\316\261\316\267

finite.

to

Now let

\316\265
>

N > 1 such

exists an

\342\200\224
<
\316\265
\316\261\317\207

6.3.6 (or

Proposition

aN

<
\316\265

by the

hence

the

that,

have

we

particular,

and

any raal number. Since (an)^=1is a


be eventually \316\265-steady, so in particular

0 be

must

it

sequence,

= L+.

L~

that

show

sequence
an

<

Cauchy

there

is

In

\316\265-steady.
(\316\261\316\267)\342\204\242=\316\235

for
+ \316\265
\316\261\317\207

all

>
\316\267

N.

By

Lemma 6.4.13)this impliesthat

mf(an)\342\204\242=N

definition of

<

sup(an)\302\243lN

L~ and L+ (and

< aN

Proposition

\316\265

6.3.6

again)
&N
we

Thus

\342\200\224
<
\316\265

But

set

this

:=
\316\265

L+ -

L~ <

2\316\265.

L+ and L~ do not dependon\316\265;


for all \316\265
> 0, and
have L+ = L~. (If L+ > L~~ then we could
\342\200\224
and obtain a contradiction.) By Proposition
L~)/3

therefore

(L+

6.4.12(f)

we

Remark
12),

+ \316\265.
\316\261\317\207

is true

must

we

L+ <

have

0 <

so

L~~ <

Theorem

thus

see

6.4.20.
6.4.18

the sequence

that

In the

converges.

language of metric spaces (seeChapter


that the real numbersarea complete

asserts

\316\240

and limit

liminf,

Limsup,

g.j.

metric space-

that

contain

not

do

they

169

points
same

the

\"holes\"

way

do. (Certainly the rationalehave lots of Cauchy


which
do not converge to other rationale; take for instance
sequences
which converges to the
the sequence
1,1.4,1.41,1.414,1.4142,...
This
is
to the least upper
related
irrationaly/2.)
property
closely
of
and
is
one
the
bound
property
principal
(Theorem
5.5.9),
which
make
the real numbers superior to the rational
characteristics
for
the purposes
of doing analysis (taking limits, taking
numbers
and
of
derivatives
integrals,
finding zeroes of functions, that kind
we
shall
as
see
in
later
chapters.
thing),
the rationale

Exercise6.4.1.

Prove

for

and

6.4.2. State

Exercise
limit

6.4.3. Prove parts


use
earlier parts of
can
you

(bn)^=1

why

Explain

sup(6n)5iLi\302\267

an <

that

such

Exercise6.4.7.Prove
zero in

+oo as a limit
as a

us

iff

it has

that

is the

superior

the limit inferior

which

(an)^=1

6.4\320\2242.

later ones.)

(Hint:

6.4.14.

Corollary

bounded

>
\316\267

sequences

1, but

(\316\261\316\267)\342\204\24

that sup(a^)^Lx

<\302\243

Lemma 6.4.13.
6.4.17.
Is the corollary still true if we
this
Corollary
by some other number?

does not contradict


of

has
of real numbers
that a sequence {\320\276>\320\277)\342\204\242=\320\274
it
has
\342\200\224oo
has no finite
and
that
bound,
upper
show
this
no finite lower bound. With
definition,
of

point
limit

(an)J\302\260=M,

of

points

largest limit point

is the

use Proposition6.4.12in
Exercise 6.4.9. Using
the
sequence

Proposition

to prove

say
iff it

that limsupn_>00 an is a limit


is larger than all the other
limit

all

for

bn

Corollary

point

limit point

this

the statement

Exercise6.4.8.Let

of

6.4.13.

Lemma

Lemma 6.4.13to prove


6.4.6. Give an exampleof two

Exercise

replace

proposition

6.4.5. Use

Exercise

and

the

6.1.4

6.1.3 and

inferior.

limit

(c),(d),(e),(f)

Exercise

Exercise6.4.4.Prove

of Exercises

analogues

prove

superior, and

limit

points,

6.4.5.

Proposition

the

has

smallestlimit
course

definition

and

that

furthermore

*n other
(\316\261\316\267)^=\316\234\316\257

words,

of a sequence.Similarly,

of the

point

sequence.

show

(One can

proof.)

in Exercise

exactly three

of a

it

the

limit

6.4.8, construct
points,

at

\342\200\224oo,
0,

and

+00.

Exercise 6.4.10.
(bm)i\302\243=M

be

Let (an)^=N be a sequenceof real numbers,


of real numbers such that
each
sequence

another

and

let

bm is a

6. Limits of sequences

170

points

be a limit point of
of {an)\342\204\242=N. Let \321\201
(bm/)~=M.Prove
a limit point of (\316\261\316\267)^=\316\235.
words, limit points
(In other
are themselves
limit points of the original
sequence.)

6.5

Some

limit

point

is also
\321\201

limits

standard

Armed now with the limit laws and


a large
number of limits.
compute
A

the

is that

limit

simple

particularly

that

of limit

squeeze

test,

we can

now

of the constant sequence

have

we
\321\201,\321\201,\321\201,\321\201,...;
clearly

lim c

\321\201
\316\267\342\200\224\321\216\320\276

for

constant

any

\321\201

(why?).

Also,

This now

Proof.

Prom

6.3.8

(for
know

have

We

limn_>oo

= 0.

= 0

l/nl/k

for every integer

that l/nllk is a decreasing


below
n,
by 0. By Proposition
being
sequences instead of increasing sequences)we
decreasing
that this sequence converges to somelimit L > 0:
we know

5.6.6

Lemma

bounded

while

of

thus

1/n

limn_,oo

implies

Corollary 6.5.1.
k> 1.
function

proved that

6.1.11, we

in Proposition

lim

l/n1/k.

\320\277-\321\216\320\276

Raising

this

to the

precisely,

Theorem

kth

and

power

and

6.1.19(b)

the limit

using

laws (or more

we obtain

induction),

Lk = lim 1/n.
71\342\200\224\320\256\320\236
By

Proposition

L cannot

6.1.11

we thus

be positive (else Lk

we are done.
Some

other

have Lk =
would

be

0;

but

positive),

this

means

so L

that

= 0, and
D

basic

limits:

171

\316\262.6. Subsequences

Let \317\207
be a real number.
\321\205\320\277
Then the limit \320\235\321\202^-\321\216\320\276
equal to zero when \\x\\ < 1, exists and is equal to 1
= \342\200\224
and diverges when \317\207
1 or when
1.
\\x\\ >

6.5.2.

Lemma

is

and

gusts

x = 1;
\321\204\320\265\320\270

Lemma 6.5.3. For


See

Proof.

develop the root and

use

(Hint:
that

using

Show

6.5.1.

Exercise

the

6.3.4,

\317\207
>

we have

0,

= 1.

limn_>oo xlln

D
more

standard

tests

that limn-^o
6.5.1 and the limit

nq does not

and for

l/n9 = 0
laws,

later

limits

for series

for

sequences.
rational

any

Theorem

on, once we

6.1.19.)

exist. (Hint: argue by

q > 0.
Conclude

contradiction

6.1.19(e).)

Exercise 6.5.2.
Exercise

few

ratio

Corollary
limit
limn_>oo

Theorem

any

6.5.3.

Exercise

We will derive

6.5.2.

See Exercise

Proof.

Prove

and

Lemma

the squeeze

6.5.2.

(Hint:

use Proposition

6.3.10,

test.)

Lemma
6.5.3.
Exercise6.5.3. Prove
>
1
1
and
\317\207
<
cases\317\207
separately.

(Hint: you may need to treat the


You might wish to first use Lemma
that
for every \316\265
result
> 0 and every real
6.5.2 to prove the preliminary
<
\316\234
there
an
that
M1/71
exists
\316\267
such
>
number
1+ \316\265.)
0,

6.6

Subsequences

This chapter has been devoted to the study


real numbers, and their limits.Some
sequences
a single limit, while othershad multiplelimit

of

(an)^)=1

sequences
were
points.

convergent
For

of
to

instance,

the sequence

1.1,0.1,1.01,0.01,1.001,0.001,1.0001,...
has two limit points at 0 and 1 (which
also the
are
incidentally
Hm inf and lim sup respectively), but is not actually
convergent
are
not
the
lim
lim
inf
and
(since
However, while
sup
equal).

6. Limits of sequences

172
this

is not

sequence

seems

it

components;

convergent

it does appear to
to be a mixture of two

convergent,

contain
convergent

namely

subsequences,

1.1,1.01,1.001,...

and

0.1,0.01,0.001,....

To makethis notion

more

of

a notion

need

we

precise,

subsequence.

6.6.1

Definition

sequences

of (an^o

increasing

of real

exists

there

iff

6.6.2.

sequenceof
f(n)

:= 2n

If

(an)\302\243i:0'

Q>f{n)

(an)\302\243L0

Since

for

all

f\302\260r
all

\316\267
N.
\342\202\254

is

a sequence,

then

(a2n)^=o IS a

function

/ : N

\342\200\224\342\226\272
N
defined

ao>G2j

from
although

sequence

\316\261\316\271,\316\2616>...

of

ao>aba2,a3,a4,

Example

6.6.3. The two

be

that

\316\267
\342\202\254
\316\235)such

is a strictly increasingfunction

a subsequence

(M\302\243L0

the

that we do not assume/ tobebijective,


the
injective (why?). More informally,

is

and

(an)\302\243Lo

We say that (bn)\302\243Lo IS a subsequence


function / : N \342\200\224\342\226\272
N
is strictly
which

(i.e., f(n + 1) > f(n)


bn =

Example

Let

(Subsequences).

numbers.

sequences

1.1,1.01,1.001,...

and

0.1,0.01,0.001,...

mentioned earlier are both subsequences


1.1,0.1,1.01,0.01,1.001,1.0001,...

of

to N.

it is

subby

Note

necessarily

173

\316\262.6. Subsequences

The

of being

property
not

though

symmetric:

Let fan)^,
Then (an)\302\243Lo

6.6.4.
Lemma
numbers.
real
of
Furthermore,

a subsequenceis reflexiveand transitive,

is

if{bn)\342\204\242=b

be

sequences
(\302\260\316\267)^=\316\277

a subsequence

(cn)%Lo

of

{cn)%L0is a
of (an)%L0.
D

We now relate the conceptof subsequences


limits and limit points.

Proposition 6.6.5

related

(Subsequences

be a sequence

of real numbers,

following two

statements

the

(an)\342\204\242=0.

and
o/(an)^0,
is a subsequence

6.6.1.

Exercise

See

Proof.

**

a subsequence

subsequenceof(bn)^L0,then

and

(M\302\243Lo>

are

and letL be
logically

to

the

of

concept

Let (an)\302\243Lo
limits).
Then the
number.

to

a real

(each one

equivalent

implies

other):

The

(a)

(b)

See

of

subsequence

Every

Proof.

{a,n)%Lo converges to

sequence

L.
to L.

converges

(an)^L0

6.6.4.

Exercise

6.6.6 (Subsequences related to limit


Let
points).
let
L
a
real
and
be
real
number.
numbers,
of
are logically
Then the following two statements
equivalent.

Proposition
be
(an)\302\243J-0

(a) L
(b)

a sequence

is a limit point of (an)%L0.


exists

There

a subsequence

Proof.See

6.6.5.

Exercise

6.6.7. The above two


between the notion of a limit,
a sequence has a limit
then
L,
Remark

L. But

when

subsequences

of(an)^=:0 which convergesto L.

sequence

converge

has

to L.

a sharp contrast
limit point. When
also converge to
subsequences

propositions
and

all

L as

that

give

of a

a limit point, then only

some

6. Limits of sequences

174

to

now

can

We
due

Bernard

1897):

prove
Bolzano

bounded

every

an important theorem in real analysis


(1781-1848) and Karl Weierstrass(1815^
has a convergent subsequence.
sequence

Theorem 6.6.8 (Bolzano-Weierstrass


theorem).
bounded
sequence (i.e., there exists a real number
< \316\234
\\\316\270\"\317\200\\ for
\302\260f
(an)\302\243Lo

all

\316\267
\342\202\254
\316\235).

which

is at

(an)\302\243Lo

0 such

\316\234
>

least one

be a

that

subsequence

converges.

limit superiorof the sequence(\316\261\316\267)\342\204\242=$Since

Let L be the
Proof.
\316\234<
have \342\200\224
an <

we
the

there

Then

Let

\316\234
for

natural

(Lemma

principle

comparison

all

n, it follows from

numbers

6.4.13) that

<
\342\200\224M

L <

M. In

or \342\200\224oo).By Proposition
+oo
particular, L is a real number
(not
L
is
thus
limit
of
Thus
a
by Proposition
point (an)^Lo6.4.12(e),
6.6.6, there exists a subsequenceof (an)\302\243Lo which
converges
(in
it
to
0
fact,
converges
L).

Note

could

we

that

of the limit superiorin

as well
the

the limit inferior instead

have used

above

argument.

6.6.9. The Bolzano-Weierstrass theorem says


that
if
a sequence
is bounded, then eventually
no choice but to
it has
converge in someplaces;it has \"no room\" to spread out and stop
itself from acquiring
limit
It is not true for unbounded
points.
for
the
instance,
sequences;
sequence 1,2,3,... has no convergent
In the language of topology,
whatsoever
subsequences
(why?).
Remark

this means

the

that

whereas

interval

an unbounded

R :
\342\202\254
{\320\266

\342\200\224
\316\234<

<
\317\207

set such as the realline

\316\234}

The distinction betweencompactsetsand non-compact


be very important in later chapters- of similar
importance
distinction
between finite sets and infinite sets.

Exercise
6.6.1.

Prove

Lemma

other?

compact,

compact.
will

sets
to

the

6.6.4.

Exercise
6.6.2. Can you find two
not the same sequence, but

are

is

is not

sequences

such that

(an)\302\243L0

each

is a

anc^

(&n)5\302\243=o

subsequence

w^cb

of the

Real

7.
\316\262

is
exists and
the

(an)\302\243Lo

zero. (Hint: for

equal to

rij
quantity
N : \\an\\ >
\342\202\254

{\316\267

:= min{n

j} is

implications

(b),

:=

N
\302\243

min{n

\321\211

<

L\\

each

for
\321\211

: \\an \342\200\224
L\\ <

:=

anj.)

of the two

one

that

(Hint: to show
natural
numbers j

explaining

1/j},

that (a) implies


by the formula
set

the

why

Then consider the

is non-empty.

1/j}

j, introduce
why the set

explaining

6.6.6.

Proposition

numbers

the

first

Proposition 6.6.5. (Note

Exercise6.6.5. Prove
define

> j},

: \\an\\

short proof.)

a very

has

N
\302\243

\320\230\321\202\320\277_\321\216\320\2
\\/bn

number

natural

each

tnat

such

(an)\302\243Lo

non-empty. Then set bj

6.6.4. Prove

Exercise

(6n)^=o OI*

bounded. Show

is not

which

De \320\260
sequence

a subsequence

exists

there

that

Let

6.6.3.

fi$ereise

175

II

part

exponentiation,

N
\302\243

{n

sequenceanj.)

6.7 Real exponentiation,part II


We

finally

that

we

in

started

topic of exponentiation of real numbers


xq for all
Section 5.6. In that sectionwe defined

to the

return

but
we have not yet defined
rational q and positive real numbers
\320\266,
now
is
this
a
real. We
situationusing limits (in a
x<* when
rectify
all the other standard operations
similar way as to how we defined
real
the
on
First, we need a lemma:
numbers).

Lemma6.7.1
a be a
numbers

of

(Continuity

real number.Let (<7n)\302\243Li


to a.

converging

Furthermore, if
to

converging

a,

(<7n)n^=i

then

Then
is

be

other

(xQn)^=:1 has

lim

n->oo

xQn

do the

just

case

\317\207
>

1, and

0,

let

and

also

numbers

of rational

sequence

the same limit as (xqn)^=i-'


lim

xQn.

n->oo

< 1,
Proof There are threecases:\317\207
=
\317\207 1 is rather
easy (because then xq
shall

\317\207
>

sequence
of rational
a convergent
sequence.

any

is
{xqn)\342\204\242=i

any

Let

exponentiation).

=
\317\207
=

1, and

1 for

leave the

\317\207
>

1.

The

all rational

case

\317\207
<

case

q).

1 (which

We

is

very similar) to the reader.

Let

us

first

prove

that

(xqn)\342\204\242=1

enough to show that


To do this, we need to estimate
x4m] let us say for the time being
6.4.18 it is

is

(xqn)^=i

the

that

By

converges.

a Cauchy

distance

qn>

gm,

Proposition

sequence.
xQn and

between
so

that

> \321\205\320\257\321\2
\321\205\320\257\320\277

Limits

6.

176

(since

\317\207
>

We have

1).

d(xqn,xqrn) =

xqm =

xQn

it
(qn)^=1 is a convergentsequence,
<
xM.
Thus
> 1, we have \321\205\320\257\321\202
M; since \317\207
=

d{xqn,xqm)
let

is

(x1/k)(\302\243=1

>
\320\232

xqm\\ <

some

upper

xM(xqn~qm

- 1).

has

bound

by Lemma 6.5.3 that the sequence


sa:~M-close
to 1. Thus there exists some

know

We

0.

\316\265
>

\\xqn

1).

\321\205\320\257\321\202(\321\205\320\257\320\277~\320\257\321\202

Since

Now

of sequence

eventually

that

1 such

\\\321\205\320\263/\320\272
-1\\<\316\265\317\207~\316\234.

is convergent, it is a Cauchy
sequence,
(qn)^Li
that
is an N > 1 such
qn and qm are 1/AT-close for
N. Thus we have

since

Now

so there
n,m>

d(xqn,xqm)=
for

1) <

such that

> N

n,m

every

xM(xqn~q\342\204\242

have this bound when

>

qn

n,m>

we also
symmetry
the sequence

By

qm.
<

qn

all

- 1) < \317\207\316\234\316\265\317\207~\316\234
= \316\265

xM(xl'K

N and

and

Thus

qm.

(xqn)^=1 is eventually
a
Cauchy sequenceas desired.
steady
every
of {xqn)^LiThis
the convergence
proves
Now we prove the secondclaim.
It will
suffice to show that
(xqn)\342\204\242=N

is

for

the

Thus

\316\265-steady.

\316\265
>

0, and

sequence

lim

xqn~q\"
n-\302\273oo

claim

the

since

then follow

would

\316\265-

is thus

1,
laws

limit

from

xQn =

(since

\320\245\320\257\320\277-\320\257\320\277\320\245\320\257\320\277),

Write
converges

rn
0.

to

is eventually

(xrn)^=i
know

:= qn \342\200\224
laws
q'n] by limit
We have to show that for

the

that

sequence

is also
\320\230\321\210^\320\276\320\276\320\226\"1^

is also

{x~l/k)^=l
\320\232
such

that

is
{rn)\342\204\242=l

xllK

convergent

\316\265-close

to

is

(\320\2661/\320\220:)^11

equal

to 1

eventually
and

x~l/K

to 0, it

every

\316\265
>

eventually

0,

the

\316\265-close

by Lemma 6.5.3, we

\316\265-close

to

are both
is eventually

1.

sequence

Lemma 6.5.3

from

But

1.

that (rn)^=1

we know

1.

know

Since
that

we can find

Thus

\316\265-close

to

we

to

1.

But

since

1/AT-closeto 0, so

that

\316\262/f.

Real

eventually

<

x~llK <

and thus

l/K,

also eventually

xTn is

particular

In

<

rn

-1/\316\233\316\223

177

II

part

exponentiation,

\316\265-close

< \317\207\316\273'\316\272
\321\205\320\223\320\277

1 (see

to

Proposition

4.3.7(f)), as desired.
We

definition.

the following

make

now

may

toa real
Definition 6.7.2 (Exponentiation
real, and let

be

of

xa =

formula

by the

numbers

rational

limn-\321\216\320\276
xQn,

where

Let

exponent).

We define

number.

a real

be
\316\261

the

xa

quantity

is any

(qn)%Li

\317\207
>

sequence

to a.

converging

is well-defined.
First
of all,
Let us checkthat this definition
real
number
a
we
have
at
least
one
sequence
any
always
given
numbers
to a, by the definitionof
\302\260^
rational
converging
(9n)^=i
numbers
real
Proposition
(and
6.1.15). Secondly, given any such

sequence

6.7.1.

Lemma

limit

the

(qn)^=i,

exists
\320\235\321\202\320\277_>\320\276\320\276\320\2269\320\277
by

choices
for the sequence
though there can be multiple
the
limit
all
same
Lemma
6.7.1.
Thus
this
give
^ey
by
(<?\316\267)\317\204\302\243=\316\271>

Finally, even

is well-defined.

definition

is not
If \316\261
just real but rational,
could in principle
this
definition
3, then

i.e., a =
be

rational

some

for

inconsistent

with our
But in this

6.7.
earlier definition of exponentiationin Section
case a is clearlythe limit of the sequence(?)\302\243Li, so by definition
xq = xq. Thus
the new definition of exponentiation
xa = limn_>oo
is

6.7.3.

Proposition

All

of Lemma

results

the

5.6.9, which

held

real

= xqxr

(i.e., the

continue to hold for

q and r,

numbers

rational

for

old one.

the

with

consistent

numbers

andr.

We demonstrate

Proof.
first

left

of Lemma

part

to Exercise

for

the

identity

xq+r

the other parts are similarand


idea is to start with Lemma
5.6.9

5.6.9(b));

6.7.1. The

and then

rationale

this

take limits to obtain

the

are

for

results

corresponding

for reals.

Let q and
afld
rationale,

r =

r be realnumbers.
some

\320\270\321\202\320\277-\321\216\320\276
rn for

by

the

definition

Then

sequences

of real

we can

write q =

(qn)%Li and

\320\230\321\202\320\277_\321\216\32
qn

{rn)^Li of

numbers (and Proposition6.1.15).

6.

178

Limits

by the limit laws, q + r is the limitof (qn


inition of real exponentiation,we have
Then

'

xq+r

lim xqn+rn;

But

by

xqn+rn

Lemma
5.6.9(b)
_ xQnxrn
Thus

+ rn)^Lv

xq = lim

\316\267\342\200\224\321\216\320\276

\320\276

By def.

=
xr
\321\205\320\223\320\277.
xqn;
\317\200\342\200\224\321\216\320\276 \317\200\342\200\224\321\216\320\276

(applied to

lim

rational exponents)we

by limit laws we have

xq+r

desired.

Exercise6.7.1.Prove

have

aflxr

the

remaining

components

of Proposition

6.7.3.

Chapter

Series

have

we

that

Now

will

we

sequences,

series

a reasonable theory of
developed
use that theory to develop a theory

limitsof
of

infinite

oo

/ ,

0>n

+ am+2

+ Gm+1

<t>m

71=771

we

develop

of finite

series.

before

But

theory

7.1

infinite series,

first

the

develop

Finite series

Definition7.1.1
a

finite

to
\321\211

each

be

(ai)-Lm

number

<
\320\263

the
\320\260\320\263
by
\320\242\320\273=\321\202

of real

sequence

we define

be integers,

m,n

and let

numbers, assigninga
m

\320\263
between

integer

Then

n).

Let

series).

(Finite

m <

we must

and

\316\267
inclusive

the finite sum (or finite

\316\267

Y^

di

whenever n<m\\

:= 0

i=m

n+1
Z^
i=m

Thus

ai :=

\316\267

\316\243
\316\271

\\\320\263=771

for instance

\\

ai

22\320\260\320\263
\320\263=771

an+1

>
\316\267

whenever

we have the identities


771\342\200\2241

771\342\200\2242

~^~

\302\260;

22

ai = 0;

\320\263=771

771

\316\243

ai

\320\263=771

am>

(i.e.,
series)

formula

recursive

real

\342\200\224

1.

180

7\302\267

Series

771+2

771+1

/ J

a>i

+ \316\261\317\204\317\200+2
\"\316\231\"\"\316\2617\316\267+1

am

i=m

i=77i

we sometimesexpress\316\243\342\204\242=\317\204\316\267\316\261>\31
less

of this,

Because

(why?).

2^f ai

+ flm+li

Q>m

as

formally

\316\267

di = am

/]

am+\\

... +

an.

i=.m

the

of

form
For

\"series\"

and

\"sum\"

one. Strictly speaking,a seriesis an

Y^l=7nQ>u

is mathematically
which
is then

series

this

mantically)equalto a
series.

between

The difference

7.1.2.

Remark

subtle linguistic

real

number,

+ 4+

1+ 2+ 3
instance,
were
to be very picky

is

a series,

is a

expression

(but not

se-

the sum of that


whose sum is

about semantics,onewould
not
l + 2 + 3 + 4 + 5a
not
consider
consider
15 a series and one would
the two expressions
sum,
having the same value. However,
despite
as it is purely
we
will
not be very careful about this distinction
has
no
the
and
expressions
linguistic
bearing on the mathematics;
1+ 2 + 3 + 4 + 5 and 15arethe same number, and thus
in the sense of the axiomof substitution(see
mathematically
interchangeable,
SectionA.7), even if they are not semantically interchangeable.
15; if

one

Remark 7.1.3.
index

of

the

variable);
any

is a

summation)

quantity

expression
i.
named

i with

summation

the

that

Note

variable

called

\320\263

(sometimes

the

(sometimes calleda

bound variable
dummy
not
does
on
actually
depend
X^Lmai
In particular, one can replacethe index
of

any other

symbol, and obtainthe

71

same

sum:

71

\316\243\316\261\316\257
\316\243 \316\261\316\220\302\267
\321\212=7\320\277 j=,m

some basic

list

We

properties of summation below.

Lemma 7.1.4.
(a)

Let

<

\316\267
<

be
\317\201

let ai

and

integers,

assignedto eachintegerm<i<p.
\316\267

ai

we

\317\201

\317\201

\316\2662\316\261\316\257+
\316\243

Then

be a real number

2Zai'

\320\263=\321\202 \320\263=\321\202\320\263+1
\320\263=\321\202

have

7.\316\257.

181

series

Finite

m <n

(b) Let

be

and
integer,
m < i < n.

another

\320\272
be

integers,

real numberassignedto eachinteger

let ai be a

Then we

have
\316\267

n+k

i=m

m <n

(c) Let

be

j=m+k

integer

let a*, bi

and

integers,

to each

\316\267

be real

Then we

<n.

<i

\316\267

\\

Let

< \316\267
\317\204\316\267
be

to each

m <

integer

Then we

and

integers,

\320\263
<n,

(e) (Triangle inequality

n.

a{

Then

\\

be a real numberassigned
\321\201
be

another

real

number.

have

^2(cai)

and let

let a*
and let

\316\267

\316\2434

\316\243(<\316\267+\316\271\316\267)=
\316\243\316\2610+

(d)

numbers assigned

have

be
we

for

a real

finite

number

c[J2ai

Let m <n be integers,


to
each integerm <i<
assigned
series)

have

\316\271\316\243>\316\271<\316\243>\316\271\302\267

(/)

test for finite series) Let m < \316\267


be
be real numbers assignedto eachinteger
< n.
< n. Suppose that ai < bi for all m < \320\263

(Comparison

and
i

integers,

let a^ bi

Then

<

we

have
\316\243\316\261*^\316\2436*\302\267

Proof

See

Exercise

7.1.1.

7. Series

182

Remark 7.1.5.In
series

in
parentheses

simply

make

we may write

+ bA
\316\243\342\204\242=\317\204\316\267(\316\

safe from
the alternative interpretation (X^Lm
in b* is meaningless
sense (the index \320\263

because

not

does

of the

is reasonably

This
a>i + bi.
\316\243\316\257=\317\200\316\271

interpreted,

omit some

may

for instance

expressions,

as

we

future

the

any

is only
a
summation, since\320\263
to also
One can use finite
series

of the

mis-

being
a*)

&.

outside

variable).

dummy

over

define summations

finite

sets:

7.1.6

Definition
with

set

over finite

(Summations

\316\267
elements

(where

function from

X to the realnumbers

f(x) to each

element

\317\207
of

sum

f(x)
\316\243\317\207\316\262\317\207

<

assumedto

have

:
N
\342\202\254
{\320\263

(i.e.,

Then

X).

select
any bijection g
a bijection exists since
then define

such

X\\

We

\316\267
elements.

: X \342\200\224\342\226\272
R be a
/ assigns a real number
we can define the finite
let

first

foU\302\260ws\302\267
We

<
to
\320\263
\316\267}

sets). Let X bea finite

and
\316\235),

\316\267
6

from

X is

\316\243/(*)==\316\243/\320\276\320\274)\302\267
xex

where

Let X

7.1.7.

Example

\316\220=\316\271

are
a, b, \321\201

distinct

be the three-elementset X := {a,b,c},


and let / : X \342\200\224\342\226\272
R be the
objects,

2, /(b) := 5, /(c)
function/(a)
we
select \320\260
bijection
f(x)i
\316\243\317\207\316\265\317\207
:=

5(2) := b,

:=

5(3)

\316\243 /(*)

We then

:=

\342\200\2241.
In

the sum
5(1) := a,

to compute

order

g : {1,2,3}

\342\200\224\342\226\272

X,

e.g.,

have
=

\316\243/(\302\273\302\253)

/\320\274+m+m

=6-

\320\266\320\261\320\245 \320\263=1

could

One

c,

h(2)

bijection from {1,2,3} to X, e.g.,h(l):=


but the end resultis stillthe same:

another
= c,
/i(3)

pick

:=

b,

3
\316\243 /(*)

To

that

verify

to

value

tions g
other words,
from

^c)+/(*)+/\320\274=

6\302\267

\320\263=1

x6X

defined

\316\243/wo)

/(\320\266), one
\316\243\317\207\316\262\317\207

:
N
\342\202\254
{\320\263

we

actually does give a single,wellhas to check that different bijec<


\320\263
to
X give the same sum. In
\316\267}

this definition

must

1 <

prove

183

series

Finite

J.
\316\263

7.1.8 (Finite summationsaxewell-defined).

proposition
set

be cl finite

\316\267
elements

with

a function,

<
\320\263

Then we

be bijections.

1 <

1 <

\342\200\224>
<
X
\320\263
\316\267}

f : X
and h :

be

\342\200\224>
R

let

\316\267
\342\202\254
N);

{where

N
:
and let g : {\320\263
\342\202\254

Let

\342\200\224>
X
\316\267}

:
G \316\235
{\320\263

have

Proof.
/ :X
to

any

R,

X,

when

n; more precisely, we let


set
X of \316\267
any
elements,
any
: 1
N
two bijections 5, h from
\342\202\254
{\320\263

\342\200\224\342\226\272
and

we have

complicated

on

\"For

that

assertion

the

is somewhat more
see
Section 8.2.
sets;

issue

induction

use

We

\320\263=1

infinite

over

summing

\316\243/(\316\234*))\302\267

\320\263=1

The

7.1.9.

Remark

\316\243/(5\302\253)

/fo(0)
\316\243\316\223=\316\271

function
<

<
\320\263

\316\267}

informally,

(More

/(MO)\"\302\267
\316\243\316\223=\316\271

assertion that Proposition7.1.8is true for


for
all natural
of n.) We want to prove that P{n) is true
is the

P(n)

P{n) be

that

value

numbers

n.

We first check the


and

f(h(i))

X)i=1

basecaseP(0). In this
to 0, by

both equal

case

\316\243<=1

of finite

definition

f(g(i))

series,

so

we are done.
Now

P(n + 1)istrue.
/ :X

that

inductively

suppose

Thus,

be

let

a function, and let g


< \316\267
< \320\263
+ 1} to X. We have
be

\342\200\224\302\273
R

N
:
\342\202\254
{\320\263

finite

series,

+ 1);
we

can

thus

\317\207
is

expand

\320\263=1

let us

element

an

the left-hand

\316\243/(*\302\253)=

Now

(7\316\233)

\316\267

of X.

By definition of

side of (7.1) as
\\

\316\243/(*(*))+*.
\\\320\263=1

look at the right-handsideof (7.1).Ideally

have h{n

from

bijections

\320\263=1

n+l

like to

be

\316\243/wo)\302\267

\320\263=1

0(n

to prove that

\302\243/<*(*))

\320\266
:=

and

n+1

n+l

Let

P(n) is true; we now prove that


a set with \316\267
let
+ 1 elements,

+ 1) alsoequal

to

- this
\317\207

would

we

allow us

would

to use

7. Series

184
inductive

the

assume this.
is some

there

much

P(n)

hypothesis

\316\243/wo)

\316\243

\\\320\263=1

that

= x.

h(j)

We

write

/wo)

\\i=j+l

f\302\243/(M0))

+ /Wi))

\\i=l

know

n+i

\\

we cannot

series to

of finite

\316\243/wo)I +

\320\263=1

which

for

definition

the

/ j
=

< n +1,

with 1 < j

now use Lemma7.1.4and

n+i

is

since

However,

index j,

more easily - but


a bijection, we do

/(MO)

\316\243

\\i=j+l

=
(\316\243/(\316\2340))+*+[\316\255/(\316\234<+\316\220))
\\\320\263=1
/

\\i=i

We now
setting
can

We

: 1 <i
X \342\200\224
define the functionh : {i \342\202\254N
<n} \342\200\224>
{x}
by
:=
< j and
+ 1) when i > j.
h(i) := h{i) when \320\263
\320\233(\320\263)h(i
thus
side of (7.1) as
write the right-hand

+*+

where
proof

we

used

have

of (7.1)

7.1.4 once

Lemma

we have

+*
(\317\202/wo))

[\316\243/wo)

(\317\202/wo))

again. Thus to

to show that
=

the

But
is

function

bijection

\316\243/(5(0)

\316\243/(\316\2340)\302\267

\320\263=1

\320\263=1

g (when

from

the

finish

restricted to

:
N
\342\202\254
{\320\263

i <

<

(7-2)

:
N
\342\202\254
{\320\263

n}

<

\342\200\224
\342\200\224>
X

{x}

<
\320\263

\316\267})

(why?).

< i < n} \342\200\224*


also a bijectionfrom
{i e N : 1
\342\200\224
X
Since
has
\316\267
elements
{x}
(why?
3.6.9).
{x}
Lemma
from
the
directly
(by
3.6.9), the claim 7.2 then follows
induction
D
hypothesis P(n).
The
X

function

h is

\342\200\224

cf. Lemma

Remark 7.1.10. Suppose


of
pertainingto element

that

an

\317\207

X,

is a

and

set, that

/ : {y

P(x) is a property

X :
\342\202\254

P(y)

is true}

\342\200\224*

is
\320\233

185

series

Finite

J,
\316\263

we will often

Then

a function.

abbreviate
/(*)

\316\243

xe{yeX:P(y)

is true}

as \316\243\317\207\316\262\316\247:\316\241(\317\207)
/(*) or even \317\211
f(x) when there
is true
is true
\316\243\317\201(*)
0\316\223
is no chance of confusion. For instance,
\316\243\316\2676\316\235\302\2672<\316\267<4^(:\316\225)
is
^2<\320\277<4/(\320\226)

for

Short-hand

/(\320\226)
\316\243\317\2006{2,3,4}

\320\2572)

/(3)

\342\204\226\342\226\240\"

The

following

do require

but

obvious,

on finite sets are fairly

of summation

properties

a rigourous

proof:

Proposition 7.1.11 (Basicpropertiesof

over

summation

finite

sets).

(a)

If X
empty

is

empty,

function),

and

: X

is a

\342\200\224>
R

function

(i.e., f is

the

we have

\316\243 \321\217*)=\302\260\302\267
\321\202\320\265\321\205

(b)

If X
is

consists

a function,

of a single element,X
we have
=
\316\243/(*)

{xo},

and f

:X

\342\200\224>
R

/(*<>).

xex

(c) (Substitution,
function, and

part I) If X is
is a
X
g :Y \342\200\224\302\273
\316\243 /(*)

bisection,

set, f

:X

is a

\342\200\224>
R

then

\316\243 /(\302\273(\302\273))\342\226\240

xex

(d)

a finite

yeY

< m
and let X be
be integers,
part II) Let \316\267
<
<
: \316\267 \320\263 \317\204\316\267}.
is a real
\316\226
number
the
set X := {i \342\202\254
If \316\261\316\271
i
to
we
have
then
each
\342\202\254
X,
assigned
integer

(Substitution,

771

=
\316\243\316\261* \316\243\316\261*\302\267

i=n

ieX

7.

186

(e) Let

be
\316\247,\316\245

XU7->Risa

function.

\316\243

(\316\243/(*))

(Linearity,
and g : X

Let X

part I)

and let

a function,

\321\201
be

number.

a real

\342\200\224\342\226\272
R
be

Then

xex

R
finite set, and let f : X \342\200\224>
and
such that f(x) < g(x) for all
R be functions
g : X \342\200\224>
\317\207
X.
Then we have
\342\202\254

Let

(Monotonicity)

be a

\316\243 /(*)

(Triangle
inequality) Let
be a function,
then

\316\243 \320\266*).

xex

xex

(i)

\342\200\224>
R

*(*)\302\267

be a finite set, let f

xex

(h)

f :X

and let

set,

\320\266\320\261\320\245 \321\205\320\265\321\205

Let X

II)

part

\\yeY

\316\243 /(*)+\316\243

\321\205\320\261\320\245

(\320\264)(Linearity,

(\316\243/\317\211)\302\267

(/(*)+*(*))

Then

be functions.

\342\200\224\342\226\272
R

\316\243

and f

a finite

be

0),

have

we

Then

\\xex

=
\316\247
\316\245
\316\240

(so

/\317\211

zexuY

(/)

sets

finite

disjoint

Series

be

a finite

set,

and let

f :X

\342\200\224>

\316\271
\316\243 /(*)i

xex

Proof.

See Exercise

Remark

7.1.12.

canbe

thought

\316\243 !/(*)!\302\267

xex

7.1.2.
substitution rule in
the substitution
making

The
of

as

\316\240

Proposition 7.1.11(c)
:=
\317\207

g(y)

(hence

name). Note that the assumptionthat g is a bijectionis essential;

the

187

series

Finite

rj\\%

can y\302\260usee
onto? Prom

will fail

*be rule

w^y

when g is not

Proposition 7.1.11(c) and (d)

not

or

one-to-one

that

see

we

771

771

i=n

i=n

< \320\263
<
: \316\267
the set {i \342\202\254
\316\226
itself.
to
\317\200\316\271)
this means that we can rearrangethe elements
of
Informally,
will
and
obtain
the
same
value.
at
still
finite sequence
.

/ from

bijection

any

for

finite series - finite

at double

look

we

Now

. and how they connectwith


be a

Let

7.1.13.

Lemma

Cartesian

be
\316\247,\316\245

finite

and let

sets,

\342\200\224*
\316\245
R
f : \316\247\317\207

of elements

number

7.1.8);

P(n)

all

function

any

numbers

natural

in X.

i.e., we let

We

use

will

P{n) be the

for any set X with \316\267


elements,

is true

7.1.13
\316\245
and

/(*.\302\273)\302\267

(x,y)eXxY

Proposition

finite set
for

yeY
the

\316\267
be

\316\267
(cf.

Lemma

that

series

products.

\316\243

\316\243(\316\243/(*'\302\273))=

xex

inductionon

of finite

Then

function.

Proof. Let

series

/ :

\342\200\224>
\317\207
\316\245
\316\247
R.

We

assertion

and

wish

any

to prove

n.

from
case P(0) is easy, following
Proposition
7.1.11(a)
Now
that
is
suppose
P(n) true; we now show that P(n+1)
(why?).
\316\267
In particular, by
is true. Let X be a set with
+ 1 elements.
Lemma 3.6.9, we canwriteX = \320\245;\320\270
where
xo is an element
{\320\266\320\276}?
\342\200\224
Then
of X and X1 := X
has \316\267
elements.
by Proposition
{xo}

The base

7.1.11(e) we have

\316\243

xex

by

the

]=
J

(\316\243/(*\342\226\240\302\273)

\\yeY
induction

(\316\243

xex'

hypothesis

\316\243

(x,y)ex'xY

+
(\316\243/(*\342\226\240\302\273))
]

\\yeY

this is

[ \316\243
\\yeY

equal to

+
\316\257(*>\316\275)
(\316\243,\316\257(*\316\277,\316\275)).
yeY

\342\226\240\342\2
\316\271

\316\272***)

7.

188

Senes

By Proposition 7.1.11(c)this is equalto


\316\243

/(*>*)+(

\316\243

/\320\276*.*))\302\267

{\321\205,\321\203)\320\265\320\245'\321\205\320\243
(x,y)e{xo}xY

equal to

this is

7.1.11(e)

Proposition

By

(x,y)exxY

as

(why?)

desired.

Corollary
be

sets,

finite

7.1.14 (Fubini's theoremfor finite series).


Let X7
\342\200\224\342\226\272
and let f : \316\247\317\207
\316\245
R be a function.
Then
\316\243(\316\243/(*\302\267\302\273))=
\317\207\316\262\316\247
y\342\202\254Y

\316\243

\316\263

f^y)

(\320\267\320\264)\320\261\320\245\321\205\320\243

=
\316\243

\321\217*,\321\203)

(y,x)\342\202\254YxX

=
\316\243(\316\243/(*\302\267\302\273))\342\226\240

In

Proof.

light

7.1.13, it sufficesto show

of Lemma

\316\243

f(x>y)=

\316\243

that

/(*>*>)\302\267

(\321\205,\321\203)\342\202\254\320\245\321\205\320\243
\320\234\320\261\320\243\321\205\320\245

follows

But

this

tion

h: XxY

a bijection,

from Proposition
\342\200\224>YxX

defined

and why does

7.1.11(c) by applying
by h(x, y) := (y, x).

Proposition 7.1.11(c)give

the

us

bijecis

(Why

what

this

we

want?)

Remark 7.1.15. This should


be
contrasted
with Example
1.2.5;
thus we anticipatesomethinginteresting
to happen
when
we move
from finite sums to infinite sums. However,
see
Theorem
8.2.2.

Exercise
induction,

but

7.1.1. Prove Lemma 7.1.4. (Hint: you will need


the base case might not necessarily be at 0.)

to use

first

it

may

to

turn

as
Proposition 7.1.11. (Hint: this is not as lengthy
a
the
It
is
matter
of
appear.
largely
choosing
right bijections
sums over sets into finite series, and then applying Lemma
Prove

7.1.2.

Exercise

these

189

series

Infinite

7.jg-

7.1-4.)
of the

/(\320\266)* Wnkdi

\316\240
eX

(Note that

products?

for finite

for the finite


anc^
products
\316\240\316\223=\316\271\316\261*
above results for finite series have
analogues

a definition

Form

7.1.3.

Exercise

is

it

dangerous

to apply

logarithms

f(x) could be zero or negative.


because
defined
logarithms
yet.)
haven't
n\\ for
natural
the
function
factorial
Exercise7.1.4. Define
\317\207 +
:=
1
and
:=
\316\267!
definition
0!
recursive
+
the
(n
1)!
(n
by
the
binomial
real numbers,
are
prove
formula
\321\203
a* or

of the

some

Besides,

we

\316\267
numbers
1). If a: and

\342\200\236n-3

for all

e X
the

that

Let

7.1.5.

Exercise
x

numbers

natural

let (an(x))^Lm
sequence

n. (Hint: induct on n.)


and for each
be a finite set, let m be an integer,
of real numbers.
be a convergent
Show
sequence

an(x))^Lm
(\316\243\317\207\316\262\316\247

lim

an{x)
\320\2432

however

lim

an{x).

Thus
cardinality of X, and useTheorem6.1.19(a).)
finite
sums
limits.
with
always interchange
Things
convergent
trickier
with infinite sums; see Exercise19.2.11.
get

induct

(Hint:

\320\243)

and

sex

sex

we may

is convergent,

on the

7.2 Infinite series


We

are

now

ready

Definition 7.2.1
series

is

any

to sum

infinite series.
infinite

(Formal

of the

expression

series).

A (formal)

infinite

form
oo

\316\243

\302\253\316\267.

71=771

where

n>m.

is an integer, and
We sometimes write

a>m

an is a real number

this seriesas

+ ^m+l

+ am+2

for

any

integer

190

'\302\267

Series

At

this

present,

only defined formally]

series is

we

have

n0t

sum equal to any real number;


the notation
am + am+1+
+ ... is of course designedto lookvery
Uke \320\260
suggestively
but
is not actually a finite sum becauseof the \"...\"
symbol.
define
what the series actually sums to, we need
rigourously

set this
am+2

sum,
To

definition.

another

series.

infinite

formal

be a
(Convergence of series). Let \316\243\342\204\242=\317\200\316\271\316\261\316\2
For any integer N > m, we define
the
Nth

7.2.2

Definition

Sn :=

this series to be
partial
is
a
real
number. If the sequence(Sw)5JLm
Sm
limit L as N \342\200\224*
then
we say that the infinite
oo,
sum Sn of

convergent^ and convergesto L\\


L is the sum of the infinite
say that
sums
is

also

we

and

divergent,

series

L =

course
to

converges

write

some
an is

Y^Lm

and
\316\243\342\204\242=\317\200\316\271\316\2

If the
\316\277>\316\267\302\267
\316\243\342\204\242=\317\200\316\271

series

then we say that


we do not assign any

Sn diverge,

of

^2n=man\\

partial

series

infinite

the
real

number

value

]C\302\243Lman

to

that

series.

then

converges,

sum L

Proposition 6.1.7shows that if a series


a unique sum, soit is safe to talk about the

Note that

7.2.3.

Remark

it has

of a

series.

convergent

cbn
\316\243\342\204\242=\317\204\316\267

7.2.4. Considerthe formalinfinite

Examples

oo

2\"n =

^2

2\"1 + 2-2 + 2-3+

series

....

n=l

The

sums

partial

verified to equal

can be

sN

j22'n

= l-2~N

n=l
by

an

sequence

induction

easy

\342\200\224

2~N

argument
converges

to

the
(or by Lemma 7.3.3 below);
\342\200\224>
1 as N
and
hence we have
oo,

=
\316\2432_\316\267
\317\200=1

1\302\267

191

series

Infinite

1/2-

I particular, this
the series

series is convergent.On the otherhand,

we

if

consider

OO

= 21

J^2n

+ 22 +

23

...

n=l
the

then

are

sums

partial

n=l

is easily

and this

shown to be an unbounded

Thus the

divergent.

Now we

question of when a seriesconverges.


shows that a series convergesiff

address the

proposition
is eventually
sequence

following
of

the

Then

less than

Let X^Lman

7.2.5.

Proposition
numbers.

Q>n converges
\316\243\342\204\242=\317\204\316\267

there

> 0,
number\316\265

and

sequence,

hence

^\316\267
series \316\243\316\267*=\316\271
is divergent.

exists

be

if and

\316\265
for

any

series

a formal

only if

N >m

an integer

such

\316\265
>

for

every

The
the

\"tail\"

0:

of real
real

that

\321\217

flnl
\316\231\316\243,

a^
\342\200\224\302\243
for

PiQ

N\342\200\224

n=p

Proof

See Exercise

This Proposition, by
so

not

to

easy

it has

However,

7.2.6

Corollary
of

real

mother

numbers.
way,

T!\302\243Lrnan

Proof.

7.2.2.
itself,

is

not

very

handy, because it

is

the partial sums \316\243\316\267=\317\201\316\261\316\267


*\316\267
practice.
a number of useful corollaries.For instance:

compute

(Zero
Then

i/limn_,oo

test). Let Y^=mQ>n be a convergentseries


= 0. To
we must have limn_+ooan
put this
an is non-zero or divergent, then the series

\342\204\242
divergent.

See Exeioise

7.2.3.

7. Series

192

Example 7.2.7.

1,1,1,1,... isa

sequence an :=

ilarly, the

an

sequence

different notion

is a

series

we

so

\342\200\224*
\316\267
oo,

as

howeverthat

not convergeto 0
1 IS a divergent
know that \316\243\316\267\316\275=\316\271
series. (Note
The

not convergeto zero;thus

:=

1 does

of

convergence

and

(\342\200\224l)ndiverges,

series

the

convergence

sequence]

convergent

from

of

Sim.

sequences.)
in particular
is

also

does

divergent.

\316\243\342\204\242=\316\271(\342\200\2241)\316\267

then the series


to
zero,
sequence (an)^Lm doesconverge
it
or
not
be
convergent; depends on the series.
may
Y^=:m Q>n may
we will soon see that the series\316\243\316\267^\316\271
For
instance,
1/n is divergent
\342\200\224>
oo.
\316\267
to
as
the
fact
0
that
despite
1/n converges

If a

7.2.8

Definition

iff the

convergent

In order
we sometimes

convergent,
case

this

oo

<*\302\273!

53

7.2.10.

Remark
convergent series

See

subclass.

conditionally

7.2.11.
to

We

include

Thus

when

convergent\",

ls n\302\260tabsolutely
an
\316\243\342\204\242=\317\200\316\271
series

The converse to this propositionis


which are conditionally convergentbut

convergent.

absolutely

series

\316\243 1\302\260\302\273!\302\267

71=971

7.2.4.

See Exercise

exist

convergence,
convergence.

convergence test). Let 5Z^=man


numbers.
real
of
If this series is absolutely
then it is also conditionally convergentFurthermore,
we have the triangle inequality

71=971

there

absolute

(Absolute

Remark

absolutely

*s

convergent.
\316\231\316\261\342\204\2421
\316\243\342\204\242=\317\200\316\271

oo

Proof

series

a formal

in

be

\316\243\342\204\242=\317\204\316\267\316\261\316\2

this seriesis

to distinguish convergencefrom
refer to the formeras conditional
7.2.9

Proposition
be

series

Let

convergence).
We say that

(Absolute
numbers.

real

of

series

formal

is conditionally

class

the

consider

we say
this

true;
not

7.2.13.

Example

the class

not

of conditionally

of absolutely convergentseriesas

a statement such
does not

convergent.

convergent

as

*s
\"\316\243\316\267^\317\200\316\271fln

automatically mean that


If we wish to say that

but not absolutely convergent,

will

we

then

use a

instead

phrasing such as

or

convergent\",

conditionally
n0t

193

series

Infinite

7#.

is only
\316\261\316\267
\"\316\243\342\204\242=\317\204\316\267

absolutely\".

7.2.12
(Alternating
real
numbers
which
of
sequence
>
>
and
0
an
for
an+\\
thus an

proposition

o\302\260^m(\342\200\224l)nOn

to

converges

0 as

Prom

Proof

convergent

is

series test). Let (an)\302\243Lm


are non-negativeand decreasing,
every

if and

convergent

>
\316\267

only if the

sequencean

zero test, we know


that
if \316\243\342\204\242=\317\200\316\271(\342\200\2241)\316
is a
to 0, which
series, then the sequence (\342\200\224l)nan converges
the

Sw+2

Sn

+ (-l)N+1aN+i
+

SN +

N is

when

\342\200\224

Q>n+2)

have

each

Job

the

iV,

is to

(-l)N+2aN+2

- aN+2).

(-l)N+1(aN+1

(a/v+i

hypothesis,
>

\342\200\224>
oo.
\316\267

Sn+2 = Sn

by

be

the series

Then

m.

and
an
(\342\200\224l)nan
implies that an also convergesto 0, since
0.
same distance from
that an converges to 0. For
Now
conversely
suppose
the
sum
our
be
Sn := Y,n=mi~lTan\\
let Sn
partial
Observe that
that
Sn
show
converges.

But

but

conditionally,

an
converges
\"\316\243\342\204\242=\317\204\316\267

Thus

is non-negative.

odd and Sn+2 < Sn if

we have

is even.

Now suppose that N is even. Promthe above


discussion
and
< Sn
for all natural numbers \320\272
induction we see that Sjv+2fc
>
Also
we
have
SN+i = Sn aN+i
SN+2k+i
(why?).
(why?).

we have SN+2k+i = SN+2kFinally,


Thus we have

Sn
for

k.

all

\342\200\224

o>n+i

SN+2k+i

a>N+2k+i

< Sn+21c

SN+2k

< Sn

\342\200\224

a>N+i

<

Sn for

Sn<

all

>
\316\267

(why?).
the

In particular, the sequence Sn is eventually


thus
to 0 as N \342\200\224>
sequence \316\277,\316\275
converges
00,

that

is eventually

But

Sn

(why?).

we have

In particular,

Sn

<

<

\316\265-steady

for

every

\316\265
>

a#_|_i-steady.

0 (why?).

this

implies
Thus Sn

is convergent.
D
converges,and so the series\316\243\342\204\242=\317\204\316\267(\342\200\2241)\316\267\316\261\316\267

194

'\302\267

Series

The

7.2.13.

Example
and

decreasing,

converges

sequence (l/n)%Li is- non-negative,


IS
to zero. Thus \316\243\302\243\317\213\316\271(~\316\212)\316\267/\316\267
convergent

not absolutely convergent,because\316\243\316\267^\316\271


(but
Vn
diverges
see Corollary 7.3.7). Thus absolute divergencedoesnot
imply
conditional
even though absolute convergenceimplies
divergence,
it is

conditional

convergence.

Some

concerning convergentseriesarecollected

basic identities

below.

7.2.14

Proposition

laws).

(Series

to x,
converging
(a) IfY^=mQ>nis a seriesof real numbers
a
is
numbers
series
real
to
bn
of
converging y,
\316\243\342\204\242=\317\204\316\267
+ bn) is also a convergentseries, and
converges
Y^Lm(0\"n
x + y. In particular, we have

+ bn)=

^2(an

(b)

IfY^=mCbn
a real
\321\201
is
and

series,

to

oo

oo

oo

and

then

+
\316\261\316\267
\316\243
\316\243

bn'

series of real numbers convergingto x, and


then X^lm(can) is also a convergent
number,
to ex. In particular, we have
converges
is a

= \321\201 an.
\320\243^(\321\201\320\260\320\277)
\320\243^

(c)

Let

be a

integer.
are

If one

of the
then

convergent,

identity

two

oo

Y^=z7n\342\202\254Ln

and

to x.

let

\320\272
be

an

be

22
n=m

a series

integer.

be

an

and

\316\243\342\204\242=\317\200\316\271\316\261\316\267
5^1\342\204\242+*\302\260\302\273

one is also, and

m+k\342\200\2241

z2an

Let

numbers, and letk>0

series

the other

n=m

(d)

of real

series

an
\316\243\342\204\242=\317\200\316\271

an+

we

have

the

oo

22

an.

n=m+k

of real numbers convergingto x,


Then \316\243\342\204\242=\317\200\316\271+^
olso
converges
an-fc

7.5. Sums of non-negativenumbers

7.2.5.

See Exercise

Proof

195

we
From Proposition7.2.14(c)

not

does

series

3S

that

see

elements

few

one

is

There

which

the

value

of a
(though

series

we will usually not pay muchattention


index m of the seriesis.

initial

the

the convergence
of the series

of this,

Because

to).

what

to

depend

first

those elementsdo influence

of course
converges

on the

called telescopingseries,which

of series,

type

are

to sum:

easy

be a sequence
Lemma 7.2.15 (Telescoping series). Let (an)^=0
= 0.
Then
of real numbers which converge to 0, i.e., \320\235\321\202\320\277_\321\216\320\276\320\260\320\277
~
to
series
\316\261\316\277\302\267
the
converges
an+i)
\316\243\302\243\317\213\316\277(\316\261\342\204\242

Exercise7.2.1. Is the
Can

answer.

your

7.2.6.

Exercise

See

Proof

series

Exercise 7.2.2. Prove


and Theorem 6.4.18.)

and

difficulty

7.2.5

in
use

Example

1.2.2?

Proposition

Corollary 7.2.6.
use Proposition
(Hint:

6.1.12

to prove
7.2.9.

7.2.5

7.1.4(e).)

Proposition

Exercise

7.2.5. Prove

Exercise

7.2.6.

Proposition 7.2.14.(Hint: use Theorem6.1.19.)


the
Lemma
7.2.15. (Hint: First work out what

Prove

\342\200\224
should
be,
\316\261\316\267+\316\271)
\316\243,\316\267=0{\316\261\316\267

sums

partial

the

Proposition7.2.5. (Hint:

Exercise7.2.3.UseProposition
Exercise 7.2.4. Prove
Proposition

or divergent? Justify

convergent

J^Li(~l)n

you now resolve

and prove your

assertion using

induction.)

Sums of non-negative numbers

7.3
Now

we

the

specialize

comes

situation

up,

test, since

absolute

non-negative.

Note

the

negative,
and

discussion in order to consider


an are non-negative. This
from
the absolute convergence
instance,
value
\\an\\ of a real number an is always
when all the terms in a seriesare non-

preceding

all the terms


\316\243\342\204\242=\317\204\316\267\316\277>\316\267
where

sums

absolute

there

is no

for
that

distinction between conditionalconvergence

convergence.

196

'\302\267

Serie$

is a

Suppose
the

series

dn
\316\243\342\204\242=\317\200\316\271

Sn :=

sums

partial

numbers.

of non-negative
are

Then

SW+i >

i.e.,

increasing,
\316\243\316\267=\317\204\316\267\316\261\316\267

Shi

(why?). Prom Proposition6.3.8and Corollary


we thus see that the sequence(S#)\302\243Lm
is convergent
6.1.17,
if
M.
and only if it has an upper bound
In other
words, we have
just shown
> m

all

for

Proposition 7.3.1.
real

negative

is a

if there

be

Let

series

a formal

\316\243\342\204\242=\317\200\316\271\316\261\316\267

Then this series

numbers.

real number \316\234such

of

is convergentif and

\320\277\320\276\320\27

only

that

N
<

an

\320\243^

simple

of this

corollary

all

\316\234
for

integers

N >m.

is

7.3.2 (Comparisontest). LetY^=rnan


suppose
formal series of real numbers,and

Corollary
be

two

Then

for alln>m.

ifJ2\342\204\242=mbn

convergent,

oo
\316\231
\316\243 ^l

oo
-

can

convergent,
this follow
A

useful

series

the comparison

run

for all

oo

\316\243*\316\267>
\317\200=0

where

\317\207
is

oo
71=771

test in the contrapositive:if


n> m, and \316\243\342\204\242=\317\204\316\267\316\261\316\267
is not
absolutely
does
then Y^=rnbn is conditionallydivergent.
(Why
immediately from Corollary 7.3.2?)
to use in the comparison test is the geometric
series
also

\\an\\ < bn

have

is
\316\243\342\204\242=\317\200\316\271\

7.3.1.

See Exercise

We

we

\\an\\ < bn

\316\243 I\320\265\302\273!
\316\243bn-

71=771
\316\244\316\231=\316\212\316\267

Proof

then

and in fact

convergent,

absolutely

is

andY%Lmbn

that

some

real

number:

2,

197

numbers

of non-negative

Sums

\320\274

fgjnma 7.3.3 (Geometricseries).Let


>
\\x\\

series

the
^\320\265\320\277

1;

then the

series is

Y^LqX71

If however \\x\\

If
<

1,

and

convergent

absolutely

number.

real

\317\207
be

is divergent

oo
=
J>\302\273

l/(l-z).

\317\200=0

See

Proof

We now give
to test whether
convergent.

the

if

only

the

as
but

>
\316\267

the series

Then

1).

criterion,

Cauchy

terms

decreasing

is

be cl

(an)^Li

(so an>0

real numbers

of non-negative

sequence
decreasing

all

a useful criterion,known
a seriesof non-negative

7.3.4 (Cauchycriterion).Let

Proposition
On for

7.3.2.

Exercise

and

<

\316\261\316\267+\316\271

if and

is convergent
\316\261\316\267
\316\243\316\267^\316\271

series

oo
2a2 +

+
=\320\260\320\263
\320\243^2\320\272\320\2602\320\272

4a4 + 8a8 +

...

k=0

is convergent

7.3.5.

Remark
only

whose

elements

determine
Proof

:=

whole seriesis convergent

be the
\316\261\316\267
\316\243\316\267=1

:= \316\243)\316\261;=\316\277
^ \320\2602\320\272
be
the
\316\244\316\272

Proposition
is bounded

do this

partial

partial

or

sums of

sums of

(namely,

order

to

and

let

not.

ani
\316\243\342\204\242=\316\271
In

of

light
\320\225\320\272=\320\2762\320\272\320\2602\320\272\302\267

that
the sequence
7.3.1, our task is to show
(\302\243/\316\275)^=1
To
if
the
is
bounded.
sequence (\316\244#)\302\260\302\243=0
only

if and

we need the following

Lemma

is that it

criterion

of this

of elements of the sequencean


=
in
index \316\267
is a power
of 2, \316\267
2\320\272)

whether the

Let Spj

feature

interesting

number

a small

uses

those

An

7.3.6.

For any

claim:

natural number K,

we

have

S2k+i_i

<

Tk<2S2k.

Proof We use
if = 0, i.e.

induction on K. First we

Si <T0<2Si.

prove

the

claim

when

198

7.

becomes

This

a\\

Now

try to

a\\ <

<

2\316\261\316\271

is non-negative.
since \316\261\317\212
the claim has been proven

is clearly

which

true,

suppose

prove it

for

\320\232
+

<

<

we
the

an

+
\316\2612\316\272+\316\271.

and (f), and the hypothesis

Lemma 7.1.4(a)
are decreasing)

have

(using

2\316\272+\316\271

2\316\272+\316\271

S2k+i

2S2k+i.

\316\244\316\272+\316\271

Tk+i =Tk + 2
that

now we

we have

Clearly

Also,

AT, and

for

1:

S2k+2_i

S2k +

an ^

2_^

&2K

a2K+1

2-*t

\342\200\224

&2K

+ 2

CL2K+\\

n=2K+l

n=2K+l

and

Series

hence

2S2k+i >
Similarly

we

2S2k + 2

+
\316\2612\316\272+\316\271.

have

S2k+2_i

= S2k+i_i

2^ ^

n=2*+i

2^+2-l

<

+
52\316\272+\316\271_\316\271

a2K+1

2_^

n=2^+i

= ^K+i-i + 2
Combining

these

inequalities

with the

S2k+i_i <Tk <


we

\316\2612\316\272+\316\271.

induction hypothesis
252\316\261\316\263

obtain

S<2*+2-i
as

desired.

This

proves

< T/c+i

the claim.

<

252\316\272+\316\271
\342\226\241

\316\263.3.

Sums

199

numbers

of non-negative

From this claim we

(52k)^=0is bounded,

is bounded.

hence

and

is bounded,

if (Sn)^=1

that

see

then

Conversely,

(7\321\217-)\302\260\302\243=0

ls bounded, then the claimimplies


that
S2k+i-i
<
\316\234for
that
all
bounded, i.e., there is an \316\234such
S2k+i_i
K.
But
one
can
show
that
numbers
easily
natural
(usinginduction)
_ \316\271
<
>
if
\316\234
and
hence
that
for
all
natural
+ 1?
Sk+i
2#+i
if (Tk)?=o

numbers

hence

AT,

7.3.7. Let q

Corollary

>

sequence

0 be
when

\342\204\242
conver9ent
^/nQ
\316\243\316\267^\316\271

Proof.The

is bounded.

(Sjv)n=i

(l/nq)\342\204\242=1

a rational
Q > 1 and
is

number. Then the series


divergent when q < 1.
and

non-negative

Lemma 5.6.9(d)), and so the Cauchy


series is convergentif and only if
oo

is

criterion

decreasing
Thus

applies.

(by
this

V2fc\342\200\224
L\302\267

is

can

by the
as the

But

convergent.

we

this

rewrite

(*>\302\267

laws of exponentiation (Lemma5.6.9)


geometric series
oo

k=0

As

mentioned

and

the geometric

earlier,

series Y^LqX1*

converges

if

if |x| < 1. Thus the series\316\243\316\267^\316\271


w^\302\267converge
Vn9
if
which
if
and
if q > 1
<
1,
only
only
happens
|21_9|
it
without
Lemma
and
5.6.9,
using
Try proving just using

only

if and

(why?
logarithms).

In particular,

series

the

harmonic

is divergent,

series)

known

(also

Vn
\316\243\316\267*=\316\271

as claimed

as the

earlier. However,the series

*s convergent.
Vn2
\316\243\316\257\317\213\316\271

Remark

called C(?)>
important

Riemann-zeta

the

in

number

ofthe primes;

it converges, is
This
function
is very
function
of q.
and in particular in the distribution
famous unsolved problem regarding

The quantity

7.3.8.

there

theory,
is

a very

when

l/nq,
\316\243\342\204\242=\316\271

200

7\302\267

Series

further

called
function,
is fax beyond

that

there

this

the Riemann hypothesis, but to


the scope of this text. I will mention

fame
US$ 1 millionprize- and instant
- attached to the solutionto this problem.

is a

mathematicians

7.3.1 to

Use Proposition

Exercise

7.3.1.

Exercise

7.3.2. Prove

Lemma 7.3.3. (Hint: for

test. For the secondpart, first


senes formula

induction

use

it

however

among

all

7.3.2.

Corollary

prove

discuss

use the zero


the first part,
to establish the geometric

= (l-xN+1)/(l-x)

^2xn

n=0

and then apply Lemma 6.5.2.)

Exercise7.3.3.

Let

such

numbers

^^L0

^\302\267
Show
\316\231\316\261\342\204\2421

that

of series

One featureof finite

the terms

in

A more

no matter

is that

sums

the

sequence,

+ a>2 +
\316\261\317\207

If

convergent series of real


an = 0 for every natural

absolutely

Rearrangement

already

an

^e

n.

number

7.4

X^Loa\342\204\242

that

+ \320\2604
+
\320\260\320\267

sum is

total
as =

g&4

how one rearranges

the same. Forinstance,


a$ +

+
\320\260\320\267

rigourous statement of this, involving


earlier,

appeared

a\\

bijections,

7.4.1.

Let

be

is

\316\243\342\204\242=\316\277
af(m)

a convergent

\316\243\342\204\242=0\316\261\316\267

negative real numbers,and let f


al>so convergent,
oo

71=0

has

7.1.12.

see Remark

One can ask whether


the
same
thing is true for
all the terms are non-negative, the answer
is yes:

Proposition

+ a^

:N

\342\200\224*
N
be

and has the


oo

771=0

infinite

series

a bijection.

same sum:

series.

of nonThen

201

of series

Rearrangement

\316\263^.

the partial sums Sn := \316\243\316\267=\316\277


anc^
\316\261\316\267
^Af :=
know that the sequences (Sn)^Lo an(^
^e
\342\204\226^)\316\257\302\243=\3
\302\243^-0 a/(\342\204\242)\302\267
L := sup(SN)%L0 and V := sup(TM)^=0\302\267
Write
increasing.
are
6.3.8 we know that L is finite, and in fact L =
Proposition
By
be
6.3.8 again we see that we will thus
by
Proposition
S**LqQ*i\\
=
V
L.
that
done as soon as we can show
:= {\320\263\320\260
: m < M}. Note
be
the
set \316\245
e N
Fix M, and let \316\245
proof We introduce

is a

that

bijection between

\316\245
and

7.1.11,

By Proposition

f(Y).

we have
\316\234

\317\204\316\274
\317\202 \320\260/\320\275 \316\243 \320\260/\320\274

m=0

The sequence

an N

<N

the

<
\316\235
: \316\267

that

assumption

iV},

there

i.e.,

bounded,

exists

f(Y) is

In particular

< N.

all

for

\316\261\316\267\302\267

\316\243

nef(Y)

*s ^\320\277^\320\265>
hence

(/(ra))\302\243f=o

such that f(m)

a subset of {n e
(and

meY

Proposition 7.1.11again
are non-negative)

and so by

all the an

=
\320\242\320\274

an

\316\243

nef(Y)

But since(SW)w=o
hence

and

a supremum

<

\320\242\320\274L

for

^n'

\316\243an

n=o

ne{neN:n<N}
^ias

that

an

\316\243

we thus seethat Sn < \302\243>


Since I/ is the least upper

of L,

all M.

Combining

that L' < L.


similar
argument
(using the inverse /-1 instead of /)
is
bounded
above by Z/, and henceL < L'.
Sn
every
D
these two inequalities we obtainL = Z/, as desired.

Example

7.4.2.

of

bound
A
shows

very
that

this implies

(7m)^=0,

Prom Corollary 7.3.7we

know

that

the

1/36

series

oo
=

\316\243 !/\342\204\2422

1 +

1/4 +

1/9 + 1/16+ 1/25+

...

n=l

is convergent. Thus,

if

we

interchange

obtain

1/4 + 1

1/16

+ 1/9

+ 1/36

every

pair

of terms,

+ 1/25 + ...

to

7\302\267

202

we

Series

is alsoconvergent,

this series

that

know

of this
sum is
(It turns out that the value
we shall prove in Exercise 16.5.2.)

as

Then

what

happens

as the

series is

ask

we

Now

long

when the

\316\266(2)

\317\2002/6,

a fact

which

seriesis not non-negative.


we

convergent,

absolutely

same sum.

the

has

and

can

still do

rearrangements:

7.4.3 (Rearrangementof series).


convergent series of real numbers,

Proposition
absolutely
a bijection.

Then ^m=o

which

\\an\\>
\316\243\342\204\242=0

numbers.

negative

Now write U
find an

Since
and find

:=

that

\316\234
such

71=0

771=0

&e an

\342\200\224>
N

be

and

convergent,

Proposition 7.4.1 to the infinite


is a convergent seriesof nonhypothesis
write L := \316\243\342\204\242=0
then
by
|\316\270\317\200|>

\316\243\342\204\242=\316\277
la/(m)l

af{m)IS

\316\265-close

such

iVi

and so

that

<

\316\265
>

to Z/,

the partial

have to

M1 >

M.
7.2.5

use Proposition
all

for

\316\265/2
\\\316\261\316\267\\
\316\243\342\200\236=\317\201

there exists \316\235>

\316\243\342\204\242=\316\277
af(m)

0, we

for every

to

we can
converSent,
\316\243\317\204\316\257=\316\277\316\231\316\261\342\204\242\316\231

an

that

In other words, given any

X)m=0

to L.

converges

to show

We
have
\316\261\316\267.

\302\243^0

also

IS

an
converges
\316\243)\317\204\316\271=\316\277

to Z/,

oo

by
If we

to Z/.

converges

oo

that

know

we

Proposition7.4.1

also

: N

We apply

(Optional)

Proof.

let

\320\260^\302\260
\342\204\242
absolutely

af(m)

X^L0a\342\204\242

same sum:

has the

series

Let

and

Since

p,q>N\\.

sums \316\243\316\267=0
also
\316\261\316\267
converge

such
\316\235\317\207

that

ls \316\265/2-close
\316\243^=0 \316\261\316\267

toL'.

Now

the

sequence

there exists an

for any

particular,

contains {n e
any

\316\234'>

\316\234such

IS

that

<
: \316\267
\316\235

iV}(why?).

set

all

\316\234for

{/(m)

So by

: m

bounded,
< N.
0 < \316\267
G N;m

Proposition

so
In

< M'}

7.1.11,

for

\316\234

\316\234'

\316\243\316\261/\316\234=

m=0

<

f~l{n)

M1 > M, the

hence

finite,

(/-1(\316\267))\317\200=\316\277

\316\235

\316\243

ne{f(m):meN;m<M'}

=
\316\261\317\200

\316\243\316\261\316\267+\316\243\316\261\316\267

n=0

\316\267\302\243*

is the set

#here X

:m

{/(m)

\342\202\254
N; m

< M'}\\{n

<
\316\235
: \316\267
\342\202\254
\316\235}.

is finite, and is thereforebounded by


we
must therefore have
q;
X

set

The

203

of series

\320\275
\316\271
Rearrangement

number

some

natural

+ l<n<q}

XC{neN:N
Thus
(why?).

\320\253^\320\244

\\\316\243\316\261\316\267\\<\316\243\\\316\261\316\267\\<
\316\243

\320\277\320\265\320\245
\320\277\320\265\321\205\316\267=\316\235+\316\271

by

\316\265-close

before is
all M' > M,

mentioned

as

which

to

L for

are

rearrangements

very

to

\316\265/2-close

is not

badly

to

\316\265/2-close

Thus

Z/.

as desired.

series

the

when

Surprisingly,
the

Thus X)^f=0^/(m) is

of N.

choice

our

\316\243^=0\316\261\316\267,

J2m=0 Of(m)

is

absolutely convergent, then

behaved.

Example 7.4.4. Considerthe series

1/3- 1/4
+

This

series

is not

be

seen

1/6 +

1/7 -

1/8+ ....

convergent (why?), but


series test, and in
alternating
converge to a positive number (in
absolutely

by the

conditionally
convergent
can

1/5

to

is
fact the sum
fact,

it

converges
1/2 = 0.193147..., see Example15.5.7).
ln(2)
reason
Basically, the
why the sum is positive is becausethe quantities
(1/3 1/4),(1/5 1/6), (1/7 - 1/8) are all positive,which can
then
be used to show that every
sum
is positive.
partial
(Why?
on whether
there are
you have to breakinto two cases,
depending
an even or odd number
of terms
in the partial sum.)
terms
If, however,we rearrangethe seriesto have two negative
to each positive term, thus

to

1/3 -

1/4- 1/6
+

1/5

1/8

- 1/10 + 1/7-

1/12-

1/14

...

'

204

then the partial

Sertes

(this is because
generally (l/(2n +
are
all negative),
and so this series
1)\342\200\224l/4n\342\200\224l/(4n+2))
converges
in fact, it converges to
to a negative
quantity;
sums

become

quickly

1/6),(1/5-

(1/3 - 1/4-

1/8

negative

and more

1/9),

(ln(2)-l)/2 = -.153426....
is

There

a series
convergent

rearranged

result of Riemann, which shows


that
which is conditionally convergent but not absolutely
in fact
be rearranged to converge to any
can
value
(or
see Exercise8.2.6);seeTheorem
to diverge, in fact
a surprising

fact

in

8.2.8.

To

is safe when the seriesis


dangerous otherwise. (This
an absolutely
rearranging
divergent series
the wrong answer - for instance,
in theoretical
series
rearranging
is somewhat
but

summarize,
convergent,

absolutely

is not to say

that

you

gives

necessarily

and
one still
performs similarmaneuvres,
answer
at
the
end
but
so is risky,
doing
(usually)
result such as Proposition 7.4.3.)
unless it is backed by a rigourous

one often

physics

a correct

obtains

Exercise7.4.1.

Let

Let /

numbers.
all

for

\316\267
e

N).

(slightly different)

7.5

Now

root

The
we

to

partial sum of
ratio

and

and prove

state

can

f(n)

\342\200\224>
N be

Show

series. (Hint: try

^e an absolutely
convergent series of real
an increasing function (i.e., /(n+1) >
that X^L0a/M
*s a^80
^ absolutely convergent
each
of X)\302\243L0Q>f(n) with a
sum
compare
partial

X^L0an

:N

\316\261\316\267.)
\316\243\342\204\242=0

tests
the famous root and ratio testsfor

convergence.

7.5.1

Theorem

(Root

numbers, and let a


(a)

:=

test).

limsupn_>00

Let

be

series

of real

\316\243\342\204\242=\317\204\316\267\316\261\316\267
(\320\260^1/71.

is absolutely
1, then the series \316\243\342\204\242=\317\200\316\271\316\277>\316\267
convergent
hence
conditionally convergent).
(and

If

a <

lb) If ot> 1, then the


(and hencecannot
= l,

(c) Ifa

< 1.

But from the geometric serieswe


ye

test,

comparison

the

convergent,

thus

and

is irrelevant

from

start

Y^=m

must

have a

> 0,

is

\316\265
<

X)^L#(a

the fact

that

1 (note

is
\316\265)\316\267

that

by Proposition 7.2.14(c)).

Thus

we see
Q-a

that

have

a +

0 <

since

convergent,

absolutely

we

that

Note

n. Then we can find an \316\265


that
> 0 such
:= (1\342\200\224\316\261)/2).
we can set \316\265
By
< \316\261+\316\265
there
exists an N > m suchthat \316\231\316\261^1/71
for
for
N.
other words, we have \\an\\ < (\316\261
all n>
+ \316\255)\316\267

In

N.

>
\316\267

conclusion.

any

for every
(for instance,

< 1
\316\261+\316\265

Proposition
6.4.12(a),

all

assert

divergent

convergent).

absolutely

> 0

Ian]1/71

0<

is conditionally

Y^=ma>n

be

suppose that a

proof. First
since

series

cannot

we

205

tests

ratio

and

root

The

r/j).

that

by

is

absolutely
\316\243\342\204\242=\316\275\316\237>\317\200

by Proposition

convergent,

absolutely

7.2.14(c) again.
Now
we

for

!>

\\an\\l^n

to 0

1-close

not

and

1,

>

that

suppose

that

see

every N >
hence that

for any

1. Then by Proposition 6.4.12(b),


m there existsan \316\267> N such that
hence

and

iV,

In particular,

1.

>

\\an\\

(an)^Lm

not
does
1-closeto 0. In particular,
(an)\302\243Lm
is conditionally
Thus by the zerotest,\316\243\342\204\242=\317\200\316\271
\316\261\316\267
For a = 1, see Exercise
7.5.3.

superior.

of the

limit superior, but

The root

replace roots
we

ratios

by

^\302\261i

<

\320\237-\320\256\320\236

Cn

There

Proof.
inequality
last

to zero.

divergent.

the

when

using
(cn)^)=m

the

limit
to

following

be

sequence

the

limit

D
of course
the

limit

instead

exists.
however

we can

of positive

numbers.

use;

lemma.

have

inf

lim

only

test is sometimesdifficult

Lemma 7.5.2. Let


Then

eventually

l^n I1/71 converges


Thus one can phrase the roottest using

limn_Kx>

is

converge

using the limit superior, but


then the limit is the sameas

is phrased

test

root

The
if

(an)\342\204\242=N

is not

inequality,

inf

lim

and

sup clJn < lim

\320\237-\320\256\320\236

sup

^\302\261i.

71-\320\256\320\236
Cn

inequalities to prove here. The middle


from Proposition
6.4.12(c). We shall prove the
leave the first one to Exercise7.5.1.

are three

follows

clJn < lim

\320\237-\320\256\320\236

'

206

Series

L :=

Write

to prove
equal
L>0.

\316\265
>

> m

Cn+i <

0.

for
Cn(L+ \316\265)

If we

write

cN(L

:=

cn(L

all

\316\265
for

all

for

that

exists

implies

that

this implies that


N

have

we

then

\316\265)~\316\2359

This

N.

>
\316\267

\316\265)\316\267~\316\235

Cn<A{L

\302\243)n

thus

and

+ s)

cl/n<Al/n(L
>
for all \316\267

we have

But

N.

lim A1/n(L

n\342\200\224\321\216\320\276

by

>
\316\267

By induction

N.

>
\316\267

all

<

Cn

(why?).

^^ < L +

such that

we

cannot

there

6.4.12(a), we know that

By Proposition

nothing

numberx), so

a finite real number. (Notethat


L
is always positive, we know
Since
^^

\342\200\22400;
why?).

Let

is

there

L is

that

assume

may

an

(since

=
lim supn_>00
^^. If L +00then
< +oo for every extended real
\317\207

limit

the

the

L +

\302\243)

(Theorem 6.1.19) and Lemma6.5.3. Thus


principle
(Lemma 6.4.13) we have

laws

comparison

lim sup

<

L +

\321\201\320\246\320\277
n\342\200\224\321\216\320\276

this

But

\302\243

is true

for all

\316\265
>

0, so

\316\265.

this must

lim sup clJn <

by

imply that

n\342\200\224\321\216\320\276

prove

(why?

by contradiction),

Prom Theorem7.5.1and

as desired.
Lemma

7.5.2

(and

7.5.3) we

Exercise

have

Corollary 7.5.3 (Ratio test). Let


nonzero

ratios

numbers.

|an+i|/|^n|

(The

non-zero

appearing

be

a series

of

\316\243\342\204\242=\317\200\316\271\316\261\316\267

hypothesis

below are

is required

well-defined.)

so

that

the

5.
\321\203

limsupn_QO

\342\200\242
//

liminfn_,oo

> 1; then the

'^V

series

is

\316\243\342\204\242=\317\200\316\271\316\277>\316

be absolutely

cannot

thus

{and

divergent

conditionally

is

\316\243\342\204\242=\317\200\316\271\316\261>\3

convergent).

conditionally

{hence

\342\200\242
//

series

< 1; then the

n^V

convergent

absolutely

207

tests

and ratio

root

The

convergent).

the

In
\317\206
Another

lim sup n1/71<

lim sup (n

by

and limit

6.1.11

Proposition

l)/n

1.
1/n = 1

sup 1 +

= lim

71\342\200\224\320\256\320\236

71\342\200\224\320\256\320\236

limit:

we have

7.5.2

Lemma

conclusion.

7.5.2 is the following

limn_,oonl/n =

We have

7.5.4.

By

cannot assert any

of Lemma

consequence

proposition
Proof.

cases, we

remaining

71\342\200\224\320\256\320\236

laws (Theorem 6.1.19). Similarly

we have

n1/n > lim

inf

lim

71\342\200\224\320\256\320\236

claim

The

I) In

inf
(n +
71\342\200\224\320\256\320\236

71\342\200\224KX)

1/n = 1.

Proposition 6.4.12(c) and (f).

follows from

then

= lim inf 1 +

useful
convergence
11.6.4.
in Proposition

test

ratio and root tests,another


is the integral test, which
we will cover

Exercise7.5.1. Prove the


Exercise7.5.2. Let \320\266
be

first

inequality

In addition

7.5.5.

Remark

very

Show

number.
that

Hindoo

that
nqxn

an

example

of a

a real

the series
= 0.

Exercise 7.5.3. Give an


numbers an such that

to the

example

number

Y^=x
of a

nqxn

in Lemma
with

\\x\\

<

is absolutely

7.5.2.

divergent series J^Lj.

an

of positive

convergent

bn of positive
\316\243\317\204^=\316\271

and

numbers

= lim^oo blln = 1. (Hint: use Corollary


limn^oobn+i/bn
This shows that the ratio and root tests can
inconclusive
the summands are positive and all the limits converge.
that

be

and

convergent,

= linin\342\200\224^oo
\316\261\316\231/\316\267
1,
1\321\2101\320\277_\321\216\320\276\320\260\320\277+1/\320\260\320\277

series

a real

q be

1, and

bn

give
such

7.3.7.)
even

when

Chapter

Infinite

sets

of

study

study of

to the

return

now

We

of sets

cardinality

have

not

set theory, and specificallyto the

which are infinite (i.e., sets which


do
number n), a topic which
natural
any

\316\267
for
cardinality
in Section 3.6.
initiated

was

8.1

Countability

Prom Proposition3.6.14(c)
we

a proper

\320\243
is

with
cardinality

X.

subset

from Theorem

instance,

the case for

is not

this

However,

have

not

\316\245
does

3.6.12we know

set

However,

same

the

\342\200\224

{0},

as N,

cardinality

is

This
We

because the function /

and

the

of N.

N, still has
\342\200\224>
N \342\200\224

{0}

\342\200\224

sets.

uncountable

8.1.1 (Countablesets).A
just

countable)

infinite

(or

natural

numbers N.

countable
infinite
but not

is either

For

is a bijectionfrom N to N {0}. (Why?)


by f(n)
one characteristic of infinite sets; see Exercise
8.1.1.
now
two types of infinite sets: the countable
distinguish

Definition

is

infinite,

\"smaller\" than

being

despite

sets.

of natural
thanks to

set N

subset

:= n+1,

defined

sets

set, \320\260\320\273\320\

equal

infinite

the

that

is also
numbers is infinite. The set N \342\200\224
{0}
Proposition 3.6.14(a) (why?), and is a proper
the

is a finite

if X

that

know

then

of X,

set

it

iff

or finite.

countable.

has

is said
We

say

set

is said

equal

cardinality

to be at
that

to be

a set

countably

with

the

most countableiff

it

is uncountable

it

if

\320\276j,

209

Countability

Countably infinite sets are alsocalled

8.1.2.

jlemark
able sets.

Examples
is countable,

denumer-

N
we
see
that
8.1.3. Prom the precedingdiscussion
and so is N \342\200\224
of
Another
a countable
example
{0}.

set is the even natural


:= 2n provides a
f(n)

numbers

{2n

\316\267
since
\342\202\254
\316\235},

bijection

between

and

the

the

function

even natural

numbers (why?).

Let X be a countable
exists

there

thus

we

Informally,

tt.

X =
a

Thus,

Then,

by definition,

X.
/ : N \342\200\224>
Thus,
the form f(n) for exactly

in

written

be

can

set.

a bijection

set

countable

every
one

we know
element
natural

that
of X

number

have

{/(0),/(l),/(2),/(3),...}.
can be

arranged in a sequence,so that

we

a zeroth element /(0), followedby a first element


/(1), then
a second element/(2), and so forth, in such a way that all these
are all distinct, and togetherthey fill
elements
/(0),
/(1), /(2),...
sets
are called countable; because
these
out all of X. (This is why
count
them
one
can
we
by one, starting from /(0), then
literally
/(1), and so forth.)
in this
Viewed
way, it is clear why the natural numbers
have

N = {0,1,2,3,...},
the

integers

positive

= {1,2,3,...},

N-{0}
the

and

even

natural

numbers

{0,2,4,6,8,...}

are countable. However,it is not as obvious


Z =
or

the

{...,-3,-2,-1,0,1,2,3,...}

rationals

Q =

{0,1/4,-2/3,...}

whether

the

integers

8. Infinite sets

210
reals

the

or

R=

or not;

countable

are
can

arrange

these

answer

will

We

for

instance,

real numbers

the

2.5,...}

{0,^2,-\317\200,

it is not yet clear whether we


in a sequence/(0),/(l),/(2),..t

shortly.

questions

Prom Proposition 3.6.4 and Theorem3.6.12,we


sets are infinite; however it is not so clear
countable
those
infinite
sets are countable. Again, we will answer
first
We
the
need
following
important
principle.
shortly.
8.1.4

Proposition

of the

subset

nonempty

such

that

<
\316\267

non-empty set of

words, every
element

that

whether

all

questions

(Well ordering principle). Let X be a


natural numbers N. Then there existsexactly

\316\267
X
\342\202\254

element

one

know

m for

all m

X.
\302\243

In

other

a minimum

has

numbers

natural

Proof See Exercise8.1.2.

principle

as

is

minimum

the

instance

minimum

the

same

the

clearly

element \316\267
given
by the well-ordering
of X, and write it as min(X). Thus
for
of the set {2,4,6,8,...}
is 2. This minimum

to the

refer

will

We

\316\240

as the

of

infimum

X,

as defined

in Definition

5.5.10 (why?).

Proposition 8.1.5.
is increasing,
particular,

Proof
gaps

Then

N.

numbers

there exists

give

We now
recursively

and

is hence

countable.

some
sketch of the proof, with
mark (?); thesegaps will be filled in

an incomplete

by a question
8.1.3.

marked

Exercise

infinite subset of the natural


X which
a unique bisectionf : N \342\200\224\342\226\272
N.
all
\316\267
In
\342\202\254
>
+
f(n
1)
f(n) for

be an

in the sense that


X has equal cardinality with
will

We

Let

define a sequence\316\261\316\277,\316\261\316\271,\316\2612>\302\267\302\267\302\267
numbers
of
natural

by the formula
X : \317\207
an := min{a:\342\202\254
am
\316\246

for

all m

Intuitively speaking,ao is the smallestelement


second smallest element of X, i.e.,the smallest

< n}.
of

element

X\\

a\\ is the
of X once

211

Countability

J.
\316\262

Observe

that

in

am

for

values of

order
all

is infinite,

since

element of X\\ and so forth.


one only needs to know the
ra < n, so this definition is recursive. Also,
the set {x \342\202\254
X : \317\207
for all m < n} is
am
\316\246

is the
\316\261<\316\271

is

removed;
\316\261\316\277

infinite(?),hence

third smallest
to define an,

Thus

non-empty.

min{a:

the minimum,

:
X
\342\202\254

the

by

for

\317\207
am
\317\206

well-ordering
principle,
all ra < n} is always

well-defined.

can

One

is an increasing
sequence,

that an

show(?)

a.2 <

a\\ <

<
\316\261\316\277

and

in

On

G X

there
Let
for

Suppose

of the

set {x

of an,

\342\200\242.^However,

Also,

we have(?)

an
for sake

an

>
\317\207
\320\260\320\273\320\260
hence

for every

\317\207
\342\202\254
-X\",

that an \317\206\317\207
this niij/iiesi?) that \317\207
is an

X : \317\207
\342\202\254
am
\316\246

is an

\316\267
for

that

of contradiction

n. Then

this impliesthat

since

claim

we
\342\200\224
x.

number

natural

definition

that

\316\267
such

\317\207
X.
\342\202\254

every

element

(?7,

we

an

exists

\316\267 ra.
\317\206

number n.
:= an. Prom the
X by f(n)
/ : N \342\200\224\342\226\272
we
know that / is one-to-one. Now
show

In other words,

/ is onto.

all

for

am
\317\206

...

function

the

paragraph

previous

that

natural

for each
define

Now

an

that(?)

particular

i.e.

for

>
\317\207

an

increasing

every

natural

all m

< n} for all n. By


natural number

for every

sequence, we have an > \316\267


number n. In particular

contradiction. Thus we must have


=
\317\207
for
number n, and hence / is onto.
some
natural
an
\342\200\224>
:
N
is
both
one-to-one and onto, it is a bijection.
X
Since
/
at least one increasing bijection / from
N
thus
We
have
found
to X.
at
Now suppose for sake of contradictionthat there was
N to X which
least
one other increasing bijection g from
was not
: g(n)
set {n \342\202\254
N
is non-empty,
the
equal to /. Then
\317\206
f(n)}
and define m := min{n \342\202\254
N
: g(n)
thus
in particular
\317\206
/(\316\267)},
=
=
=
and
for
all
\316\267
\321\202.But
we
<
am,
g(m) \317\206
f(n) an
f(m)
g(n)
we

then

have

> \317\207
\317\207
+

must

1, which

is a

have(?)

X : \317\207
at
\316\246
g(m) = min{a:\342\202\254

a contradiction.Thus
N to X otherthan

there

/.

is no

for

other

all t

< m}

increasing

= am,
bijection from
D

8. Infinite sets

212

Since finite sets are at

countable

most

we

by definition,

thus

have

of the natural numbersare at

All subsets

8.1.6.

Corollary

most

countable.

Corollary
subset
of

8.1.7. If X is an at mostcountable
X, then \316\245is at most countable.

this
follows
then
Proof If X is finite
so assumeX is countable.Then
there

between

\316\245
and

a bijection

is

we restrict /

of X,

to F,

and

\316\245
is

-\302\273
N

and / is a bijection
we

obtain

a bijection

a bijection?) Thus f(Y)

is this

has

f(Y) is a subsetof N, and


by Corollary 8.1.6. Hence\316\245is also at

hence

most

countable.

\316\240

Let

8.1.8.

Proposition

Proof.See
be

8.1.9.

Proof.See

Exercise

Proposition
countable

See Exercise

Proof.

We

can

countable

is at

f(X)

set,

and let

\342\200\224\342\226\272
\316\245
be a

f :X

-+Y

most countable.
D

be

countable

set,

and let

\316\245
be

8.1.7.
subset

or image

of a

countable set is at

any finite union of countable setsis stillcountable.


establish
count ability of the integers.

Corollary 8.1.11.

The

set.

countable

and

now

:N

8.1.5.

To summarize,any
countable,

be

8.1.10. Let X
Then X UY is

set.

let f

and

Let X

Then

a function.

set,

8.1.4.

Exercise

Corollary

\316\245
be

is at most countable.

Then /(N)

function.

3.6.14(c),

/ :

Y. But

with

countable

most

from Proposition

a subset

\316\245
is

(Why

f(Y).

cardinality

equal

at

N, then when

and

Since

and N.

between
from

set,

integers

\316\226
are

countable.

most

j,
\316\262

213

Countability

proof. We already know


countable.

are

numbers

-\316\235

:=

{-\316\267

the

set N

The

set

\342\200\224N
defined

=
: \316\267
\342\202\254
\316\235}

isalsocountable,since
N and this set. Since

the

the

countability

set

the

that

\317\207
\316\235
\316\235
is

union

of N

and

\342\200\224

N,

we need to relate
In particular, we need to
We first need a preliminary

the

of

Cartesian

with

between

bijection

8.1.10

from Proposition

follows

-3,...}

\342\200\224n
is a

the

are

integers

To establish countability
show

:=

of natural

by

-1, -2,

{0,

f(n)

map

the

claim

= {0,1,2,3,...}

that

rationale,

products.
countable.

lemma:

Lemma8.1.12.

The

A :=

set

{(n, ra)\342\202\254NxN:0<ra<n}

is countable.
ai,
a2,...
Proof. Definethe sequence\316\261\316\277,

:= an +
ao := 0, and \316\261\316\267+\316\271
=

0 +

one

can

0;

\316\261\316\277\316\261\317\212

By

induction

whenever

n>

\316\267
+

=
1; \316\2612

(n7, m!)

that

numbers n.

=
2; \316\2613

1+ 2+

3;....

increasing, i.e., that an

: A

>

am

\342\200\224\342\226\272
N

by

:= an

is one-to-one.

+ m.

In other

are any two distinct elementsof A,

f{n,m)^f(n',m').

0 +

Thus

m (why?).

f(n,m)
claim

1+

0 +

by setting

recursively

all natural

show that On is

Now define the function

We

1 for

words,

then

if

we

(n,ra)

claim

and

that

To prove this claim,let (n,ra) and (n\\m')


be two distinct
elements of A. There are three
cases:
n! = \321\211\320\277!
and
\316\267!
< \316\267.
> \316\267,
=
First
n.
we
m
otherwise
that
nf
Then
must
have
m7,
suppose
\317\206
Thus
an + m \317\206
an + ra7,
(n, m) and (n7, ra') would not be distinct.
and hence /(n,ra) \317\206
as
desired.
f{nf,mf),

8. Infinite

214

Now

/(n7,

But since(n,

thus

\316\267
and

>

ra7)

m7)

/(n7,

nf <

case

The

+
+ \316\267
\316\261\316\267

>
\316\267'

1 >

Thus

\316\267
< \316\267
+

an + m

Proof.

\316\267
wox^
\342\202\254
\316\235}

bijection from

The set

8.1.13.

We already
A

is countable.

hence

= /(n, m),

is

to

/(A),

so A has

and

of N,

a subset

and hence

most countable. ThereforeA is at


not finite. (Why? Hint: if A was
clearly
b^ finite,
and in particular
of A would
be finite, but this is clearly countably
D
Thus, A must be countable.

a contradiction.)

Corollary

1.

is at

f(A)

countable.
But, A is
then
finite,
every subset
infinite,

n +

by switching the roles of


Thus
we have shown that /
argument.

most

{(n, 0) :

1, and

hence

similarly,

proven

is a

8.1.6

On +

\316\261\316\267+\316\271

equal cardinality with f(A). But f(A)


Corollary

1, and

\317\206
/(\316\267,m).
\316\267
is

is one-to-one.
by

\316\267
+

>

ani

ra <

have

the previous

n7 in

m! >

ani +

\342\202\254
A, we

ra)

/(n7,

and

ra7)

n. Then

n' >

that

suppose

sets

countable.

\317\207
\316\235
\316\235
is

know that the set

:=

{(n,

m)\342\202\254NxN:0<m<n}

This impliesthat the set


:=
\316\222

{(\316\267,m)

\317\207
:
\316\235
\316\235

0 <

<
\316\267

\317\204\316\267}

\320\222
:=
since the map / : A \342\200\224\342\226\272
by /(\316\267,\317\204\316\267)
given
\317\207
is
from A to \320\222
But
\316\235
is the
since \316\235
\316\267) a bijection
(\317\204\316\267,
(why?).
union of A and \320\222
from
the
claim
then
follows
Proposition
(why?),

is

also

countable,

8.1.10.
Corollary

8.1.14.

// X

and

\316\245
are

countable,

then

\317\207
is
\316\245
\316\247

countable.

Proof.

See Exercise

Corollary 8.1.15.

8.1.8.
The

rationals

are

countable.

J.
\316\262

215

Countability

\316\226
that
the integers
are
proof. We already know
\342\200\224
the
\316\226
that
non-zero
are
integers
{0}
implies

{0}) =

\317\207
\316\226

(Z

If one

countable.

/(a, b) :=

(why?).

8.1.14, the set

By Corollary

is thus

which

countable,

countable

that

a/\320\254(note

{(a, b) : a, b

\342\202\254
Z, b \317\206
0}

\342\200\224\342\226\272
\317\207
lets / : \316\226
Q be the function
(\316\226
\342\200\224{0})
/ is well-defined since we prohibit b from

that /(\316\226\317\207
we see from Corollary
8.1.9
equal
being
(\316\226- {0}))
\342\200\224
= Q
\317\207
most
But
we
countable.
have
is at
(why?
/(\316\226
(\316\226 {0}))
of the
Thus
is at
rationale
This is basically the definition
Q
Q).
cannot
be
since
it
contains
countable.
most
Q
finite,
However,

to 0),

the infinite set N.

Thus

is countable.

8.1.16. Because the rationaleare


Remark
the
in principle that it is possibleto arrange
a sequence:

Q=
such

that

as

numbers

rational

{\316\2610,\316\261\316\271,\316\2612,\316\2613,...}

sequence is

of the

element

every

know

we

countable,

from

different

every

that the elementsof the sequenceexhaust


Q
turns
as
oneof
the
elements
number
an
up
(i.e.,
not
of the sequence).
However, it is quite difficult
(though
to
actually
try and come up with an explicitsequence
impossible)
which
does
ao,
ai,...
this; see Exercise 8.1.10.
other

element, and
rational
every

Exercise
8.1.1.

Let

there exists

asX.

Exercise8.1.2.Prove
induction,

or

be

a set.

a propersubset

use

least upper bound

of infinite

(or

greatest

lower

principle

work

is infinite
has

if

the same

and

only

(Hint:

you can

Exercise

What if we replacethe natural

numbers

by the

positive

rationals? Explain.

Exercise8.1.3.Fillin

the

gaps

marked

if

cardinality

either use
descent,
4.4.2, or use the
Theorem
bound) principle,
5.5.9.)
numbers
if we replace the natural

8.1.4.

Proposition

the principle

Doesthe well-ordering

by the integers?

Show that X

\320\241
\316\245
\316\247
\316\247
which
\316\277\316\271

(?) in Proposition

8.1.5.

8.

216
Exercise

8.1.4.

is that

/ is not

{n G

N : f{m)

all 0

for

\317\206
f{n)

< m<

\316\267
A is the set of natural
numbers
for
in
Prove
the
sequence
appear
/(0), /(1),... f{n).
a bijection from A to /(N).
/ is restricted to A, it becomes
Proposition 8.1.5.)

which

does not

Exercise8.1.5.UseProposition
Exercise 8.1.6. Let A be a
only if there

exists an

/ :N

natural

number

and

8.1.9,

Jt,

n,

that X

show

Exercise 8.1.8.

8.1.10.

g :

:=
and
/\316\271(2\316\267)/(\317\200)

that /i(N)

cannot
U \316\245

Use Corollary

is at

(Hint:

a bijection

and

show

and

Then

A to

if and

N.

by hypothesis,

we have

\342\200\224>
\320\243. Now

define

:= g(n) for

/i(2n+1)

use

every

= X UY. Then use Corollary


be finite.)

possibly
to prove

8.1.13

when

countable

most

/(n)

that

8.1.9.

Corollary

\342\200\224>
N from

map

Proposition
\342\200\224>

setting

/i:N->XU7by

Show that
/ :A

set.

injective

Exercise 8.1.7. Prove


a bijection

to prove

8.1.8

here

n};

speaking,

informally

sets

Prove Proposition 8.1.8. (Hint: the basic problem


A to be the set
Define
assumed to be one-to-one.

:=

Infinite

Corollary

8.1.14.

Exercise 8.1.9. Supposethat / is an at most countable set, and for each


that
the set
a \302\243
/, let Aa be an at most countable set. Show
\\JaeI Aa
unions
of countable
is also at most countable.In particular,
countable
sets are countable.
the natural numbers
Exercise8.1.10.Find a bijection / : N \342\200\224>
Q from
to the rationals. (Warning:
rather tricky to do explicitly;
this
is actually
to get / to be simultaneously
it is difficult
and surjective.)
injective

We

the

introduce

now
will

which

be

sets

on infinite

Summation

8.2

of summation
that the
provided

concept

well-defined

on countablesets,
sum is absolutely

convergent.

Definition8.2.1
let
set,
/ : X

on

(Series

and

be

IS absolutely

f(x)
\316\243\317\207\316\262\316\247
X,

the

the

sum

sum of

countable

sets).

a function.

\342\200\224\342\226\272
R

Y2n=o f(9(n))

convergent

iff

for

Let X be a countable
We say that the series
some
g : N \342\200\224\342\2
bijection

is absolutely convergent.We

by the forrmila
f(x)
\316\243\317\207\316\265\317\207

oo
=
\316\243/(*)

\316\243w\302\273)).

\317\200=0
\317\207\342\202\254\316\247

then

define

2.
\316\262.

217

sets

on infinite

Summation

From Proposition7.4.3(and Proposition


3.6.4),
definitions
do
not
onthe
choice
these
depend
that
well defined.
can

We

now

of

5,

can show
and

so are

theorem about double

an important

give

one

summations.

8.2.2

Theorem
N

be a

theorem for infinite

(Fubini's

function such

\342\200\224\342\226\272
R

\302\243\\

sums). Let / : \316\235\317\207


\342\204\242
absolutely

/(n>m)

m)eNxN

we have

Then

convergent

that

00

00

\321\217*.\342\204\242)

\316\243(\316\243/(\302\273\342\226\240\"\302\273)) \316\243
\317\200=0
m=0

(n,m)6NxN
/(n>m)

\316\243

(m,n)eNxN

00

00

rri=0
other

In

the

that

provided

we can switch the


words,
sum is absolutely
entire

back and compare


(A

Proof.
than

the

equalityfollows
We

shall

(basically

Let

\317\200=0

order of

sums

infinite

convergent

You

should

go

this with Example1.2.5.

considerably more complex


other
proofs,
reading.) The second
from Proposition
7.4.3 (and Proposition 3.6.4).
easily
similar
just
prove the first equality, as the third is very
sketch

one

only;

this proof is
and is optional

the role of \316\267


and
consider the case when

switches

us first

negative (we will

with

deal

the

general

L:=
\316\243

m).
f(n,m)

case later).

is always

non-

Write

/(\342\204\242>m);

(n,m)eNxN

our

task

is to

show that the series\316\243\302\243\317\213\316\277(\316\243\342\204\242=\316\277


converges
^(n'm))

toL.
L ^\320\276\320\263
One can easily show that \316\243(\316\267,\317\200\316\271)\316\265\317\207
^ finite
\342\200\224
/(n>m)
sets X \321\201
\317\207
\316\235
\317\207
N.
\316\235
a
between
\316\235
and
Use
N,
bijection g
(Why?

8.

218

and then usethe


ticular, every
for

fact

that

\316\267
G

g(X)

and

Infinite sets

In par-

hence bounded.)

is finite,

have

\316\234
N we
\342\202\254

< L
f{p>,
\317\204\316\267)
\316\243%\316\257=\316\277

implies by Proposition 6.3.8 that \316\243\316\267=\316\277


/(\316\267>m) IS
for any N G N and \316\234G N we
m. Similarly,
each
convergent for
(by Corollary 7.1.14)
which

have

\316\235 \316\234

f(^m)<L

\316\243
\316\243\316\243\316\257(\316\267,\317\204\316\267)<

n=0 m=0

where X is the

(n,m)eX

set {(n,ra)eNxN:n<iV,ra<

M}

which

is

3.6.14. Taking supremaof this as \316\234\342\200\224\342\226\272


oo
by Proposition
have (by limit laws, and an induction
on
N)

finite
we

oo

71=0771=0
By

this implies

6.3.8,

Proposition

that

/(\316\267>m)
\316\243\342\204\242=\316\277
\316\243\342\204\242=\316\277

con\"

and

verges,

oo

oo

J2J2f(n,m)<L.
\316\240=0
771=0

To

finish

the proof, it

will suffice to show that


oo

oo

\316\243 \316\243 ^\316\267>m)

\"\316\265

\316\240=0
971=0

be enough? Prove by
let \316\265
of L, we can then find a finite
> 0. By definition
contradiction.) So,
>
setlCNxN
L - \316\265.(Why?)
such that
\317\204\316\267)
/(n>
\316\243(\316\267,\317\200\316\271)\316\265\317\207
This
set of the form
set, being finite, must be containedin some
for

every

:=
\316\245

Thus

\316\265
>

0.

(Why

will this

{(\316\267,m)eNxN:n<JV;m<

\316\234}.

(Why?

use

induction.)

by Corollary 7.1.14

\316\235 \316\234

=
\316\243\316\243^\316\267'\342\204\242)

n=0

771=0

\316\243

(n,m)eY

f(n>\342\204\242>)>

\316\243

(n,m)eX

f(n,\342\204\242)>L-e

219

sets

on infinite

Summation

2.
\316\262

hence

and

N00

oo

oo

\316\235 \316\234

\302\243\302\243/(\320\277,\321\202)>\302\243\302\243/(\320\277,\321\202)>\302\243\302\243/(\320\277,\321\202)>\302\243-\320\265
\317\200=0
m=0
n=0
ra=0

ra=0

n=0

as desired.

This proves the

claim

the

when

are all

m)

/(n,

non-negative.

argument
m) are all
one
can
the
result
fact,
simply apply
just obtainedto
(in
For
function
/(n,
ra), and then use limit laws to removethe
note
that
function
can
be
the general
case,
any
/(n, m)
is the
as /+(n,
where
+
ra)
ra),
f+(n,m)
/_(n,
(why?)
it
of /(n,
equals /(n, ra)
m) (i.e.,
ra) is positive,
/(n,
part
and
is
the
of
and 0 otherwise),
/_
part
/(n, m) (it equals
when
is negative,
and
It is easy to
0
otherwise).
/(n,
/(n,ra)
if
*s
that
absolutely convergent, then
/(n>m)
Z}(nm)GNxN
and
soare
/-(\", m). So now
/+(n>m)
Ern,m)eNxN
to
the
results
to
obtainea
and adds
one applies
/_
just
/+
A

the /(n,

when

works

similar

non-positive

the

\342\200\224

\342\200\224.

written

positive

when

negative

ra)

show

E(n,m)eNxN

and

using limit

them together

laws

to

the

obtain

for a

result

general
\317\200

/\302\267

is another

There

of absolutely

characterization

convergent

series.

and
let f :
be an at most countable
set,
is
Then the series \316\243\317\2076\317\207
absolutely
f(x)
if and only if

Let X

8.2.3.

Lemma

be a

function.

\342\200\224\342\226\272
R

convergent

sup < ^2 1/0*01


[xeA

Proof See

Exercise

\302\267
AC

X,A

next

section.)

(We

> <

00.

J
D

8.2.1.

Inspired
by this lemma, we may
an absolutely convergentserieseven

uncountable.

finite

give

some

examples

now
when

define
the

the
set

of uncountable

concept
X could

of

be

sets in the

8.

220

8.2.4. Let X

Definition

:X

and let /

\342\200\224>
R

be

be

IS absolutely

iff

KxeA

we have

that

Note

Let X

8.2.5.

Lemma

let f :

be

set

series

the
the

be a

is
\316\243\317\2076\317\207
f(x)

lemma.

following

could be

(which

function such that


Then the set {x G
convergent.

\342\200\224>
R

absolutely

not yet said what

shall be accomplishedby

to. This

equal

< oo.

finite I

'-ACX,A

\316\231
\316\243 1/0*01

sup

sets

a set (which could be


uncountable)
a function.
We say that the series

convergent

f(x)
\316\243\317\207\316\265\317\207

Infinite

uncountable), and

series

the

: f(x)

is
f(x)
\316\243\317\207\302\243\317\207
is at

\316\246
0}

most

countable.

Proof. See Exercise8.2.2.


Because

absolutely

\320\236

define the value of \316\243\317\2076\317\207


for
f(x)
set
X by the
convergent series on an uncountable
of this,

we can

any

formula

:=

\316\243 /(*)

since we have

xeX:f(x)^0

replaced a sum on an uncountable

on the countable set {x e X


former sum is absolutelyconvergent,

Note alsothat this definition


already have for series on
We give some laws for
sets.

Proposition 8.2.6
be an
and

set

arbitrary

g : X

9(\317\207)
\316\243\317\207\316\265\317\207

(a)

(possibly

both

absolutely

The series J2xex(f(x)


\316\243(/(*)+<?(*))
\321\202\320\265\321\205

(Note

latter

by a sum
that
if the

one is

also.)
we

the definitions

sets.
convergent

absolutely

be functions

the

with

consistent

countable

(Absolutely

set

\317\206
0}.

f(x)

then

is

\342\200\224\342\226\272
R

are

/w.

\316\243

xGX

series on

series

convergent

uncountable),

such that the

arbitrary

laws).

Let X

and let f : X \342\200\224>


R
and
series \316\243\317\207\302\243\317\207
f(x)

convergent

+g(x)) is
=

absolutely

convergent,

\316\243/(*)+\302\243<?(*).
\321\205\320\265\321\205 xex

and

\316\2622.

on

Summation

is a
(b) If \321\201

real

221

sets

infinite

then

number,

cf(x)
\316\243\317\2076\317\207

is

absolutely

and
convergent,

J2cf(x)=.cj2m-

xex

xex

(c) If X

X\\

sets X\\
disjoint
are absolutely

some
\316\247<\316\271
for

and

f(x)
f(x)
\316\243\317\207\342\202\254\316\247\316\271
\316\243\317\207\316\265\317\2072

/(*)

\316\243

\316\243 \321\217*)

xGXi

x\342\202\254.X\\\\JX2

if

Conversely,

: X

\342\200\224>
R

are
absolutely
^(\321\205)
\316\243\317\207\316\265\317\2072

also

is

is such

(d)

another

is
\316\245

If

set,

and

\316\243 /(*)\302\267

that

and

h{x)
\316\243\316\266\316\276-\316\247\316\212

then

\320\234\320\266)
\316\243\317\207\316\266\317\2071\317\214\317\2

\316\243 *(*)\302\267

\342\200\224>
: \316\245
X
\317\206

\342\204\242
J*(\320\244(\321\203)) absolutely
\316\243\316\275\316\265\316\245

and

XGX2

xGXi

x\342\202\254X\\\\JX2

convergent,

then

and

\320\273\320\270 \316\243>\317\211

\316\243

X2,

\321\205^\320\245\321\207

convergent,

convergent,

absolutely

and

convergent,

is

a bijection,

then

and

=
\316\243^\317\211)

\316\243^)\302\267

xex

yeY

Proof

See Exercise

Recall

in

convergent,

phenomenon

not

but

respectto

7.4.4

Example

8.2.3.

absolutely

series was conditionally


then its behaviour with
convergent,
if a

that

bad.

was

rearrangements

We

now analyze

this

further.

numbers which
absolutely convergent
Define
convergent,
the
:= {n G N : an > 0} and
A:= {n e N : On < 0};
sets
A+
= 0. Then both of the series
thus A+ U A_ = N and A+ \316\240
A-

Lemma

8.2.7.

Let

be a
\316\261\316\267
\316\243\342\204\242=0

is conditionally

but

series

and
n\302\260tconditionally
are
\316\261\316\267
\302\260\"\317\200
\316\243\316\267\316\265\316\221\316\243\316\267\342\202\254\316\233+

thus not

absolutely convergent).

Proof See

Exercise

8.2.4.

of real

not

convergent

{and

8.

222

Infinite

sets

theorem
now ready to presenta remarkably
of Georg
which
which
that
a
series
asserts
(1826-1866),
converges
but not absolutely can be rearrangedto converge
to

We axe
Riemann
conditionally

one pleases!

value

any

8.2.8. Let \316\243\342\204\242=\316\277


be a series
which is conditionally
an
but
not
and let L be any real
convergent,
absolutely
convergent,
N
a
number.
Then
exists
there
such
bisection f : N \342\200\224\342\226\272
that

Theorem

Proof. (Optional)
in Lemma

We

8.2.7; from

are
both
\316\261\316\267
\316\243\316\267\316\262\316\221-

A- are infinite

(why?).

gent

/+

The

that

we know

be the

A-

In particular A+ and

and Em=oa/-(m) are both absolutely


the
from
plan shall be to selectterms
and

series

sets

and
\316\261\316\267
\316\243\316\2676^

8.1.5 we can then find


By Proposition
: N \342\200\224\342\226\272
A-.
and
Thus the
/_ : N \342\200\224\342\226\272
A+

(why?).

sums Em=oa/+(m)

and

A+

divergent.

absolutely

bijections

increasing

8.2.5. Let

lemma

that

proof, leaving the

of the

a sketch

give

in in Exercise

filled

be

to

details

to L.

conditionally

converges
\316\243\342\204\242=\316\277
af(m)

diver-

divergent

order in
^2^=oaMm)
\316\243\342\204\242=\316\277\316\261\316\257-(\342\204\242)
L.
order to keep their difference
towards
converging
We define the sequence no,ni, \320\2372,... of natural
numbers
as

recursively

has
\321\211
j

follows.

already

(I) If

that

Suppose

define

by
\321\211

the

If instead

Note

a natural number,and

\320\263
<

j (this

following

j4+

is vacuously

that

true if

rule:

for
all
\\\320\277\321\204\321\211

^> then

^
\320\260\321\211
\316\2430<\316\220<7

nj := min{nG
that this

A-

are infinite,

and nj

:= min{n

and

well-chosen

< L, then we set


l2o<i<j \302\260\"\320\237\321\212

nj := min{nG
(II)

j is

for all

defined

been

0). We then

\320\263
<

j}.

\320\263
<

j}.

we set

\320\220-\\\320\277\321\204\321\211
for
all

recursive definition is well-definedbecauseA+


and so the sets {n G A+ :\320\277\321\204\321\211
for
< j}
all \320\263
G

A-

all

\320\263
<

number

to

\316\267
for
\321\211
\317\206

(Intuitively, we add a non-negative


the partial sum is toolow, and
sum is too high.)
One
can then

j} are never empty.


the series whenever

number when the


the
following claims:
verify

add

a negative

8.2.

on

Summation

The
\342\200\242

j i->

map

number

infinite

an

occurs

(Why?)

injective.

\321\211

an infinite

I occurs

\342\200\242
Case

is

223

sets

infinite

number of times, and Case II also


of times. (Why? prove by

contradiction.)

\342\200\242
The

\342\200\242
We

have

that
\342\200\242
We

map

\\->

Uj

is surjective.
=

limj\342\200\224ooanj

(Why?)

Corollary 7.2.6

Note from

0. (Why?

=
\320\260\320\277
\320\235\320\277^-\321\216\320\276

0.)

have

liny-,\302\273

then

The claim

Exercise 8.2.1.
to be useful.)

L.

(Why?)

\320\236\321\211
\316\243\316\277<\316\257<\316\257

follows

f(i)

setting

by

:=

for
\321\211

Prove

Lemma

8.2.3.

(Hint:

you may

Exercise 8.2.2. Prove

Lemma

8.2.5.

(Hint:

first show if

find

i.

all

3.6.3

Exercise

the

\316\234
is

quantity
\316\234
:=

\\f{x)\\ :

sup{^

\320\241
\320\220
X,

A finite}

xeA
sets

the

then

> 1/n}

X : \\f(x)\\
\342\202\254

{x

Mn for every positive integer n.


Prove

8.2.3.

Exercise

all the

results

from

8.2.5.

8.2.7.

Lemma

Explain the

with

finite

8.2.6.
Proposition
do
7
to
Chapter
this.)

Exercise 8.2.4. Prove


use limit laws.)
Exercise

are

at most

cardinality

Then use Exercise8.1.9.)


(Hint: you

may

of

prove by contradiction,

(Hint:

gaps marked (why?)

in the

use

course

proof of

and

Theorem

8.2.8.

Exercise8.2.6.Let Y^LqCLu

/ :

not

be

that

oo

liminf

\320\2432af(m)
N^\302\260\302\260

m=N

(Of course,

which is conditionally
Show that there existsa bijection
diverges to +oo, or more precisely

a series

absolutely
convergent.
\342\200\224>
N such
that X^=0a/(m)

convergent,but

a similar

oo
= limsup

N^\302\260\302\260

statement holdswith

a/(m) = +oo.

m=N

+oo

replaced

by

\342\200\224oo.)

8.

224

have

We

just

as the

a
to hope that other
also countableto arrange as

limits

(formal)
are

rationale

rationale,

of the

countable,

such

infinite

one may

examples,
as the

such

sets,

obvious how

is not

it

which

for

After

sequence.

after

sets are countable*

a lot of infinite

that

shown

sets

such

even

sets

sets

Uncountable

8.3

Infinite

begin

real numbers, are

more than
rationale, and we've already shown the
so it seems plausible that the realsare also

all,

real

the

are nothing

numbers

countable.

It

showed

was thus
in 1873

a great shock when


that certain sets -

Cantor

Georg

(1845-1918)
numbers
R

real

the

including

uncountable - no matter how hard you try, you


the
real numbers
R as a sequence\316\261\316\277,\316\261\317\212,\316\261<\316\271
cannot arrange
(Of
the
real
R
numbers
can
containmany infinite
course,
sequences,
is
However,what Cantor proved
e.g., the sequence 0,1,2,3,4,
no such sequence can ever exhaustthe realnumbers;
no
that
matter what
of real numbers you choose, there will
sequence
always
be some real numbers that are not coveredby that
sequence.)
from Remark 3.4.10 that if X is a set, then the power
Recall
\320\241
of X, denoted
2X := {A : \320\220
set
is the set of all subsets of
X},
= {0,{1},{2},{1,2}}.
Thus
for instance
2^
The
X.
reason
for
2X is given in Exercise 8.3.1.
the notation
are in fact

Theorem 8.3.1 (Cantor's


the

Then

and 2X

sets

be an

Let

theorem).

(finite or infinite).

set

arbitrary

cannot

have

equal

cardinality.

Suppose for sake of contradiction


Proof.
had equal cardinality.
Then there

that the sets

exists a bijection/

X and

between

the power
A :=

Note

that this

and is

set of X.

{x e X

\317\207
#

set is well-definedsince f(x) is an

hence a subsetof X.

Clearly

\342\200\224>
2X

set

f(x)}.

is a

elementof 2X. Since/ is a bijection,


X such that
f(x) = A. There are
there

\317\207
\302\243

the

consider

Now

2X

and

now

subset
must
two

element

of X,

therefore
cases,

of

2X

hence is an
exist

depending

0n whether

\317\207
G

#^

/0*0\302\273

A or

hence

the

with

theorem

that

the

to

close

we

A, then
\320\266
a
contradiction.
\342\202\254
A,
\317\207
g

should compare the proof of


of Russell's paradox (Section
statement
X and 2X would come
between
bijection
itself\"
concept of a set X \"containing

is uncountable.

2^

8.3.3.

Corollary

But if

reader

The

3.2).The pointis
dangerously

contradiction.
of A we have

of

by definition

\320\266
then
\342\202\254
A,

case we have a contradiction.

8.3.2.

Cantor's

If

Ay a

\317\207
&

definition

by

Thus in either
Remark

\317\207
A.
\302\243

hence

\317\207
& /(\321\217),

have

225

sets

Uncountable

\316\262\316\2623.

have
with
equal
cardinality
Proof. By Theorem8.3.1,2^ cannot
2W
is
either
or
uncountable
finite.
contains
hence
However,
N,
: \316\267
which
is clearly
\342\202\254
as a subset the set of singletons
{{n}
\316\235},
N
2^
to
and
infinite.
Thus
hence
cannot
be
countably
bijective
D
finite (by Proposition 3.6.14),and is hence uncountable.

theorem has the

Cantor's

(and

important

following

unintuitive) consequence.

R is

8.3.4.

Corollary

uncountable.

:2^

Proof. Let us define the map /


f(A) :=

formula

by the

\342\200\224\342\226\272
R

\316\231\316\237\"\".
\316\243

neA

since

that

Observe

is

(by

convergent

We

defined.

f(A)

Since

n0

:=

8.1.4),

min(A\\B)

we

can

then

U (B\\A).

By

\316\221
\320\257,the
\317\206

the

also

set

series

absolutely

the map

Thus

/ is injective.
were two distinct

non-empty subset of N.
(Proposition

is

that

there

= f(B).

convergent

absolutely
10~n

\316\243\316\267^

8.2.6(c)).

claim

that

contradiction

that

series

Proposition
now

an

\316\243\342\204\242=010~\316\267

(by Lemma 7.3.3), the

/ is well

Supposefor sake

sets

2^
\320\222
\342\202\254
\320\224

(A\\B)

well-ordering

such

U (B\\A)

is a

principle

define the minimum of

Thus n0 either liesin

of

A\\B

this set,
or

B\\A.

say

By

8. Infinite sets

226
we

symmetry

for

and

all

0 =

it lies in

assume

may

\316\267
<

A\\B. TJien no

\320\222
or
\316\267
\342\202\254
\320\220,

no

A,

\320\222
$\302\267

\320\221.Thus
\316\267
^ \320\220,

- /(B)

/(A)

=
\316\243

i<rn

io-n

^
neB

neA

have

we either

no

10~\316\267+10~no

\317\202

10_\316\267)

\316\243

\317\200<\317\200\316\277:\317\200\342\202\254-\316\221\316\267>\316\267\316\277:\317\2006-\316\233

-(

10_\316\267+

\316\243

\316\243

10_\316\267)

\320\277>\320\277\320\276:\320\277\320\265\320\222
\320\277<\320\277\320\276:\320\277\302\243\320\222

10-n0+

\320\272\320\263\320\277-

j^

5Z

>

io_n\302\260

\320\276

10~n

n>no:neB

n>no:n\342\202\254A
10_\316\267
\316\243

n>no

>

ilO~no

\321\216-\320\277\320\276

>0,

a contradiction, where we

(Lemma 7.3.3)to

have

used

the

10~n

OO

= V^

=
\321\216-^0\"\302\267\"1\"1\"971)

\321\216-710-1

V^

10~m

-10~n\302\260.

m=0

771=0

n>no

series lemma

sum

OO

V^

geometric

contradict

injective, which meansthat /(2^) has the same


2W and is thus uncountable. Since /(2^*) is a subset
forces R to be uncountablealso(otherwise
this
would
D
and
we
are
done.
Corollary
8.1.7),

Remark

8.3.5.

Thus

/ is
as

cardinality

of

R,

measure

this

theory

We will give
in Exercise

another proof of this

result

using

18.2.6.

Remark 8.3.6. Corollary8.3.4shows


that
the reals have strictly
than
the
natural
numbers
the sense of
larger cardinality
(in
Exercise 3.6.7).
One
have
which
could ask whether there exist any
sets
than
the
natural
but
strictly larger cardinality
strictly
numbers,

84\302\267

227

of choice

axiom

The

Hypothesis
smaller cardinality than the reals. The Continuum
no
that
such
sets
exist.
was
shown
i
t
in
Interestingly,
asserts
of
Kurt
works
Godel
and
Paul
Cohen
(1906-1978)
(1934-)
septate
is independent
of the other axiomsof set
this
hypothesis
that
neither
be proved nor disproved in that set of axioms
it can
theory;
those
axioms
are inconsistent, which is highly
unlikely).
(unless

Exercise 8.3.2.
that there is a

if

: A

map

(i.e., Dn

have

the

anc^

5 and
same

\316\240
Dm

function

the

\342\200\224>
\320\222
is

to

Do :=
that

Prove

n.

\342\202\254
\317\207
l\302\243?=oDn,

use induction

on

such that

:= x

d(x)

is a

is

n.)

\320\241
\320\241 and
\320\220
\320\222
\320\241,

A.
Define
bijection/ : \320\241\342\200\224>

numbers

each other

sets

be
\320\241
\320\220,
\320\222,

setting

by

recursively

Let

n. Show that 2X

of cardinality

set

finite

2n. (Hint:

of cardinality

set

natural

be a

Let X

8.3.1.

Exercise

a finite

suppose

\302\267
sets \320\224\321\214\302\243*\321\212\320\2202
and
then
-Dn+i := f(Dn) for all
B\\A,
all disjoint
the sets \320\224\321\214
from
\302\243>i,... are
that
show
Also
0 whenever \316\267
\317\206
\317\200\316\271).
defined by setting g(x) := f(x) when
when x & l\302\243?=oDn, then g does indeed

bijectionbetween

the

the

In particular,

two.

and

\320\222

cardinality.

3.6.7 that a set A is said to have


Exercise 8.3.3. Recall from Exercise
or
\320\222
a
iff there
than
set
is an injective map
lesser
equal cardinality
\320\222
A
to
that
if A, B are sets
from
B.
A \342\200\224>
show
:
Exercise
8.3.2,
Using
/
that
A has lesser or equal cardinality
\320\222
or equal
to \320\222
and
has lesser
such
to
then
A
and
\316\262
is
known
as
have
A,
equal
cardinality
cardinality.
(This
and
after
Ernst
Schroder
the Schroder-Bernstein
theorem,
(1841-1902)
Felix

Bernstein

(1878-1956).)

Exercise 8.3.4.
a set

\320\222
if

Let us say

A has

Exercise 3.6.7) but A


for any set X, that X
that

has strictly

if A

than C,

cardinality

that

The

We
Choice

now

axiom

discuss
system

the
of

not

does

A has strictly

equal cardinality
have

has strictly

lesser

then

has

no

strictly

power

lesser cardinality
to

\320\222
(in

B,

and

\320\222
has

lesser

cardinality

set (i.e.,

a set of

sense

of

that

Show

2X. Also, show

than

cardinality

than

the

equal cardinality to B.

lesser cardinality than

Exercise8.3.5.Show that
some set X) can be countably
8.4

a set

lesser than or

lesser

strictly

than C.
the

form

2X for

infinite.

of choice
final axiom
set

theory,

of the standard Zermelo-Praenkelnamely the axiom of choice. We

5.

228

have delayed

introducing this axiomfor

Infinite

to demon*

now,

wljiile

sets

of analysis
strate that a largeportionof the foundations
can be
in
constructed without appealingto this axiom.
However,
many
convenient
further developments of the theory, it is very
in
(and
axiom.
this
some cases even essential) to employ
On
powerful
the
can
lead
to a number of
other hand, the axiomof choice
unintuitive
the Banach-Tarski
instance
consequences
(for
paradox, a
we
in
versionof
which
will
encounter
Section
simplified
18.3), and

can lead to proofs


are philosophically
that
axiomis
the
almostuniversally
Nevertheless,
logician

great

the

axiom

the

axiom

themselves

theory,

of

nor

proved

questions)
of choice,

axiom

can
this

practice,

(more

analysis

\"decidable\"
the

one

axioms

inconsistent

of

disproved

axioms are

true and false.)In

highly unlikely). Moreprecisely,


of choice is undecidable;it can
from the other axioms of set

axiom

the

that

as those

long

application

which is

inconsistent,

demonstrated

so

themselvesconsistent.(Froma set
prove that every statement is both

precisely,
which can

any \"real-life\"

that

means

any application involving


only
be rigourously supportedusing

can also be rigourously supported without


take
a much
though in many cases it would

axiom
of choice,
more complicated and lengthier argumentto
allowed to use the axiomof choice.Thus
one
the

doso

of choiceas a convenient
and
safe labour-saving
In other disciplinesof mathematics,
in
notably
of
the questionsare not decidable,
the
many

accept the axiomof choice

some

ones.

we

However,

products

from

begin

by

Definition

will

not

generalizing
3.5.7

Definition 8.4.1
(possibly infinite),and

more

is

concerns

philosophical

We

by

a theoremdue to the

Kurt
Godel, who showed that a resultproven
using
of choice will never contradict a resultproven
without
of choice (unless all the other axiomsofset theory
are

Godel

neither be

unsatisfying.

accepted

confidence is

for this

reason

One

mathematicians.

somewhat

for

the axiom

device in analysis.
in which
theory

set

of whether

issue

and

to debate,

to

involves

as mathematical and logical


discuss these issues in this text.
Cartesian
the notion of finite
to infinite Cartesian products.
as well

Cartesian

(Infinite

open

view

not

were

one

if

can

each

products).

let

J
\316\261
\342\202\254

Let

Xa be a

J be

set.

We

a set
then

Cartesian

the

define

Xa

\320\224

where we

set
1\321\215\320\265
*he

to

product

< (xa)aei

\320\260\320\261/

229

of choice

axiom

The

\316\262
j.

%<*
\320\237\320\260\320\265/

\342\202\254

e Xa

: xa

X/?)7

(|J

for

/
\320\260
\342\202\254

all

/?6J

recall (from Axiom 3.10)that

(Uae/

all

set of

is the

^<*)7

functions (xa)aei which assignan elementxa G U/?e/ ^ to each


Thus
*s a subset of that set of functions,
I^
a \342\202\254
\320\237\320\260\320\265/
of
instead
those
functions
consisting
(xa)aei which assign an element
xa

\302\243
X<*

Example
(why?).

t\302\260
each

G L

8.4.2. For any sets


If I is a set of the

*he
\316\240\316\261\342\202\254/^<*
ls

then

Definition

and

form

set as

same

X, we
:=

G
{\320\263

the set

1 <

<
\320\263

X1

\316\267},

3.5.7 (why?).

was
a^so
\316\247*
\316\240\316\271<\316\257<\316\267

non-empty.
is also true for

statement

Axiom 8.1 (Choice).


Let

The Axiom

be a

set, and

infinite)
to choose a

(possibly

The intuition
collection of

any finite

were

Cartesian

product

of choice asserts that

for eacha

\342\204\242
a^so
^<*
non-empty set Then \320\237\320\260\320\265/
there
a
exists
which
words,
function (xa)aei
an element xa G Xa-

8.4.3.

,Xn

infinite Cartesian products:

be a

Remark

\320\237\320\260\320\265/^

in
defined
\316\240\316\271<\316\257<\317\204\316\275^

Lemma
Recall from
3.5.12
that
if X\302\261y...
collection of non-empty sets, then the finite

this

have

let Xa

J,

In

non-emptyassigns

to

each

other
a E I

behind this axiomis that given


non-empty sets Xa, oneshould

a
be

single element xa from each one, and then form


the possibly infinite tuple (xa)aei fr\302\260m aU the choices one has
in
made. On one hand, this is a very intuitively
axiom;
appealing
some sense one is just applying
over and over again.
Lemma
3.1.6
On the other hand,
the
fact
that
one is making an infinite
number of
with
no
choices,
arbitrary
explicit rule as to how to make
these
are
a
is
little
choices,
many
disconcerting. Indeed,there
the
theorems
assert
proven using the axiom of choicewhich
able

abstract

existence

of some

object

with
\317\207

certain

properties,

without

8. Infinite sets

230
all

at

saying

that

what

object is,

or how

to

it.

construct

Thus the

which
are
axiom of choice can lead to proofs
\320\277\320\276\320\277-constructive of
an
without
existence
object
actually
demonstrating
This problem is not uniqueto the axiom
the
object
explicitly.
constructing
- it already appears for instance
in Lemma
of choice
3.1.6 - but the
axiom
the
of
choice
to be rather
to
exist
tend
shown
using
objects
extremein their level of non-constructiveness.
However, as long
between a non-constructive
as oneis aware
of the
distinction
a constructive
and
existence statement (with the
existence
statement,

latter being preferable,


there is no
here,

not

but

difficulty

strictly

necessary

on a

perhaps

except

Remark 8.4.4. Thereare many equivalent


axiom of choice; we give someof thesein
In analysis

sup(\302\243?)
exists

oo

<

a^sequence

an =

lim^oo

(i.e.,

\316\225
is

(an)i^=zl

bounded

from

elements

let

n,

can

is the

least upper

bound

for

Xn

denote

the set

sup(\302\243\302\267)}.

E,

then

sup(E)

\342\200\224

1/n

bound for E, and so Xn is non-empty for each


of choice (or the axiomofcountable
we
choice),
> 1.
a sequence (an)^=1 suchthat an G Xn for all \316\267

be an upper
Using the axiom

cannot
n.

We

sup(E).

Xn := {xeE: sup(E)-l/n<x<
sup(-E)

choice

axiom

of countable
but with the
give a typical

of the real line with


Then there
above).
that
an all lie in E, such

Proof For each positivenatural number

Since

axiom

subset

a non-empty

whose

below.
of the

power

the
needs
Instead, one often
only
the
axiom
of
which
the
sameas
is
choice,
index set / restrictedto be at most countable.
example of this below.
\316\225
be

of the

formulations

of choice.

Let

cases),
philosophical level.

exercises

the

one often does not needthe full

Lemma 8.4.5.

in many

then

find

In particular
for
n. But
all

for
all
an G \316\225
then we have

test (Corollary6.4.14).

n, and

sup(E)
=

\342\200\224

1/n

sup(E)

\320\230\321\210\320\277-\321\216\320\276
\320\260\320\277

< sup(E)
by the squeeze

< an

231

of choice

axiom

The

\316\262
j.

jlemark 8.4.6. In many


if

instance,

and

Let

8.4.7.

proposition

pertaining

to an

See Exercise

8.4.1.

of choice

axiom

of the

formulation

Another

axiom of choice. For

set (Definition 12.2.12) then onecan


choice
by the formula an := inf(Xn); the extra
\316\225
is closed
will ensure that an lies in E.

that

hypothesis

the

can obtain

a closed

\316\225
is

without

dn

define

using the

without

lemma

this

of

conclusion

one

cases,

special

\316\245
be

is as follows.
let

and

sets,

be

\320\240(\320\266,
\321\203)

and an object \321\203


such
G X
eY
object \317\207
property
\317\207
X
at
that
there
least
one
such
is
G
is
eY
that for every
\321\203
\320\240(\320\266,
\321\203)
that P(x,
\316\245
Then there exists a function f : X \342\200\224>
such
true.
/(#))
all
\317\207
X.
G
true
is
for
Proof

8.4.1.

Exercise
(Hint:

Show
the

consider

8.4.7.

that the axiom of choice implies Proposition


: P(x,y)
is true} for each \317\207
sets Yx := {y G \321\203
G X.)

show that
Conversely,
is also true.
choice

Exercise 8.4.2. Let


set.

nonempty

J be a set, and
that

Suppose

all the

exists

there

that

(i.e.,\316\245intersects
that

Xa

G
\316\262
\316\261,

disjoint

non-empty

in exactly
sets

Xa,

in Axiom
(Hint: the problem is that
to be
But
this
can
be fixed
disjoint.
=
\317\207
:
G Xa}
{\316\261} \316\247\316\261
\317\207)\317\207
{(\316\261,

/.

that

\316\245
such

from

are
\316\247\316\261
disjoint

above statement was true

if the

/ and

each

a set

let Xa

a G /

each

for

sets

= 0 for all distinct


i.e., Xa \316\240
\316\247\316\262
show

8.4.7 is true, then the axiom of

if Proposition

Using

#(Y

the axiom of choice,


= 1 for all a G /

\316\240
Xa)

one element).
an

for

be a
each other,

show

Conversely,

arbitrary

choice

then the axiom of choice


8.1 the sets Xa are not
by

trick

the

by

of sets
is true.

assumed
at
the sets
looking

instead.)

A and
\320\222
be sets
such that there exists a surjection
there
then
exists an
the axiom of choice, show that
\342\200\224\342\226\272
:
A
A
has
or
in
other
lesser
words
injection /
equal cardinality
B\\
to \320\222
in the sense of Exercise 3.6.7. (Hint: considerthe inverse
images
for each \316\261
3.6.8.
this with Exercise
G A.)
Conversely,
Compare
9~l{{a})

Exercise 8.4.3.

Let

\342\200\224>
A.
g : \320\222
Using

show that

if the

surjections g :
Exercise

8.4.2.)

above statement is true


the

\342\200\224\342\226\272
\320\222
then
A,

for

axiom of

arbitrary

sets

choice is true.

A, B
(Hint:

and
use

8.

232

Infinite sets

Ordered sets

8.5

connected to the theory


of
of
ordered
we
sets; will
actually many types
concern
ourselves
with three such types, the partially
ordered
sets
sets.
the totally ordered sets, and the well-ordered
is intimately

are

There

sets.

ordered

choice

of

axiom

The

Definition8.5.1

ordered

(Partially

(or poset) is a set

together1

X,

for any two objects


true statement or a
is

For

\342\200\242

(Reflexivity)

\317\207
G

If

G
\320\266,
\321\203

(Anti-symmetry)

(Transitivity)

to

<\317\207as

the

if

short)

a partially

arguments

the

(using

so

in

\321\203
<\317\207

In most
and

context,
<\317\207.

write

We

situations it

in those

is

cases we

\317\207
<\317\207\321\203
(or

\317\207
<

\321\203

\321\205\321\204\321\203.

numbers N togetherwith

rationale

partially

speaking, a

But

we shall

technically

and

\317\207
<\317\207y

the

usual

Q,

ordered

and the extended


sets. Meanwhile, if X is any
reals R,

the

\320\241
and oneusesthe relationof is-a-subset-of
(as
X
Definition
then
the
relation
for
<\317\207,
ordering
3.1.15)

Strictly

(X, <x).

such that

of sets,

collection
defined

are

\317\207
and
<\317\207 \321\203

< (as defined in Definition2.2.11)


ordered
set, by Proposition 2.2.12. Similar
and propositions) show
definitions
appropriate

also

are

R*

of

The natural

that the integersZ, the

reals

relation

less-than-or-equal-to
forms

that

such

are

is from

instead

\317\207
and
<\317\207\321\203

8.5.2.

Examples

(thus

either

\317\207
<\317\207\317\207.

relation.

ordering

set X

the

shall simply write <


for

\317\207
is
<\317\207\321\203

\317\207
<\317\207\316\266.

what

understood

we have

X,

\316\266
G X
\320\266,
\321\203,

If

\342\200\242

refer

<\317\207on

=
\317\207
\321\203.

then
<\321\205\321\217,
\320\243

We

statement

the

X,

any

\342\200\242

then
\316\266,

a relation

this relation
false statement). Furthermore,
the following three properties:

to obey

assumed

G
\320\266,
\321\203

ordered set

A partially

sets).
with

in

many

partially

ordered

cases the ordering

refer to X itself as the


incorrect.

partially

set is

not a

will
<\317\207
ordered

set X, but
set

rather

a pair

from context, and


even though this is

be clear

233

sets

Ordered

\316\262.5\302\267

partially ordered (Proposition3.1.18).Note


to give these sets a different
partial
possible
certainly
see
for
instance
Exercise
8.5.3.
standard
one;
fye
is also

some

with

set

ordered

(Totally ordered set).

8.5.3

pefinition

order

totally ordered if,

to be

\320\276\320\263
^\320\245\320\243
\320\243'
\320\243'
<\321\205
(or
\321\203

that

a totally

X is

relation
any

given

If X

both).

Let

<\317\207. A

be

subset

it is

that

than

ordering

a partially
\316\245
of

is said

two y, y' G \320\243,we either


itself is totally ordered,we

ordered set (or chain)

with

order

relation

have
say
<\317\207.

the integersZ, the


reals
reals
R, and the extended
R*, all with the usual
rationale
Q,
relation
2.2.13,
<, are totally ordered (by Proposition
ordering
and
Lemma
4.1.11, Proposition 4.2.9, Proposition 5.4.7,
6.2.5
Proposition
Also, any subset of a totally ordered set
respectively).
On the other hand, a collection
ordered
is again
totally
(why?).
\320\241
with
the
is
not totally ordered. For
relation
sets
of
usually
8.5.4.

Examples

The

numbers N,

natural

instance,if X is the set {{1,2},


{2,3,4}, ordered
{2},{2,3},
inclusion
then
the
and {2,3}
set
relation
the
C,
{1,2}
by
other
of X are not comparableto
% {2,3} and
(i.e.,
{1,2}
{5}},

elements

each

{2,3}

{1,2}).

(Maximal and minimal elements).Let X be a


be
a subset
of X. We say that \321\203
set, and let \320\243
\316\245
and
there
is no element y1 eY
element of
\316\212\316\220
eY
\321\203
is a maximal
element of \316\245if
We say that \321\203
< y.
y1
is no element y' G \316\245
there
that
such
\321\203
<y'.
8.5.5

Definition

ordered

partially
is a minimal
such

that

and
G \316\245
\321\203

8.5.6. Using the set X from


the
example,
previous
{2}
minimal element, {1,2} and {2,3,4}are maximal
elements,
is
and
element,
{2,3}
{5} is both a minimal and a maximal
that
neither
a minimal
nor a maximal element. This exampleshows
a partially
ordered set can have multiple maximaand minima;
a totally
ordered
set cannot (Exercise 8.5.7).
however,
Example

is a

Example 8.5.7. The

numbers

natural

minimal element, namely 0, but


has
integers \316\226

no

maximal

no

and no

N
maximal

by <)
(ordered
The
element.

minimal element.

has a
set of

8.

234

ordered

and

set,

element

sets).

(Well-ordered

let

well-ordered

\316\245
is

sets

Let X be a partially
\320\243
ordered subset of X. We say that
be a totally
has
if every non-empty subset of \316\245
a
minimal

8.5.8

Definition

Infinite

min(Y\.

N
are
8.5.9. The natural numbers
well-ordered
by
8.1.4. However, the integersZ, the rationaleQ, and
Proposition
the
real
numbers
R are not (see Exercise8.1.2). Every
finite
ordered
set
is well-ordered (Exercise 8.5.8). Every
totally
subset
of a well-ordered
set is again well-ordered(why?).

Examples

One

Proposition8.5.10

(Principle

well-orderedset

an

with

that

they

automatically

induction (cf. Proposition2.2.14):

of strong

a principle

obey

sets is

of well-ordered

advantage

ordering

of

strong

relation

Let X be

induction).

<,

and let P(n) be a


e X,
for each \316\267

to an element \316\267
G X
(i.e.,
a
a
true
statement or false statement).
Suppose
P[n)
that for every \316\267
we have the following implication: if P(m)
\316\225
X,
is true for allm G X with m <\317\207\316\267,then P(n) is also true. Prove
that
is true for all \316\267
G X.
P(n)
pertaining

property

is either

in

case

in

strong
2.5.

Axiom

the

It may seem strange that there is no \"base\"


induction,
corresponding to the hypothesis P(0) in
such
a base case is automaticallyincluded
However,
induction
Indeed, if 0 is the minimal
hypothesis.

8.5.11.

Remark

strong

element of X, then by
for all m G X with m
case, we automatically
See

Proof.

So

far

will come
a strict

Exercise

specializing

the

then
<\317\207\316\267,

\316\241
is
{\316\267)

hypothesis
also

\"if

true\"

P(m)

to the

upper

=
\316\267

obtain that P(0) is true. (Why?)

8.5.10.

not seen the axiomof choice


role.
play
any
in once we introducethe notionof an upperbound

we have

is true

\316\240

This
and

bound.

Definition 8.5.12 (Upperbounds


Let
and
strict
upper
bounds).
X be a partially orderedset with ordering
relation
<, and let \316\245
of X. If \317\207
a subset
be
we say that \320\266
is an
bound for
G X,
upper

8.5. Ordered
iff
\316\263

for all

< x

\321\203

\316\245.If
\342\202\254
\321\203

for
< \317\207
\321\203

iff
\316\245

F.

bound for

sirzci upper
for

235

sets

all

eY.
\321\203

Example 8.5.13. Let us


usual

the

B,: 1 <
on the

2} but

<
\317\207

work

is not

hand, is

other

we
\317\207
g \320\243,
\320\266
is

(Why

Then

<.

ordering

addition

in

Equivalently,

is this

in the

2 is

that

say

a strict

is
\320\266

bound

upper

equivalent?)

real number

an upper bound

system R with
{x G

set

the

for

a strict upperbound.The number

a strict upperbound

for

3,

set.

this

Let X be a partially
set with
ordered
ordering
a wellbe
an
X.
is
element
Then
there
xo
<,
relation
of
ordered subset \316\245of X which has xo as its minimalelement,and
has no strict
which
upper bound.
8.5.14.

Lemma

and let

to
one
is trying
Proof. The intuition behind this lemmais that
:= {xo}.
If \316\245
has
perform the following algorithm: we initalize\316\245
then
are
we
we
a
choose
no strict
bound,
otherwise,
done;
upper
\316\245.
and
add
to
Then
\316\245
bound
it
we
look
to
if
see
strict upper
again
we
a strict
has
upper bound or not. If not, we are done; otherwise
and
add
it to \320\243.We continue
choose another strict upper bound
until
we exhaust
often\"
all the strict
this algorithm
\"infinitely

comes
upper bounds; the axiomof choice
choices
are
involved.
is
however
This
many
to precisely
because it is quite difficult
pin
an

perform

is that

we

which

\"infinitely

algorithm

we will isolate

good sets

shall

often\".

because

in

down what it

a collectionof \"partially
sets, and then takethe
Yqq

object

\"completed\"

proof

means to

what we will

Instead,

do

sets

completed\"

call good

to obtain a

infinitely

a rigourous

not

of all

union

\320\243,

these

will indeed

which

have no strict upperbound.


We

now

begin

the

rigourous

proof.

Suppose

for

sake

of

has xo as
bound.
Using
upper
the axiomofchoice (in the form of Proposition 8.4.7), we can thus
a strict
assign
upper bound s(Y) G X to each well-ordered subset
\316\245
of X which
has xo as its minimalelement.
Let
us
We say that
define
a special class of subsets \316\245
of X.
a subset \316\245of X is good iff it is well-ordered,
contains
xq as its
contradiction

that

every

well-ordered

its minimalelementhas

at

least

subset

one

\316\245
of

strict

which

8.

236
minimal

x = s({yeY :y <x})

all

for

Note
of

Let
of

sets

the property that

and obeys

element,

Infinite

\317\207
G

Y\\{xo}\302\267

{y e ^ : y < x} is a subset
and contains xo as itsminimal
element.
X
: \320\243
is good}
\316\251
:= {\316\245\320\241
be the collection of all goodsubsets
X.
This collection
is not empty, sincethe subset{xo}ofX

if

that

the following

We make

if

observation:

important

subsets of X, then every

two good

is

(why?).

good

clearly

^\320\265set
*^\316\262\316\267
*\320\233{\320\266\320\276}

\317\207
G

is well-ordered

which

Y'

\316\245
and

are

strict upper

is a

of Y'\\Y

element

a strict upperbound
for
Y'.
(Exercise
8.5.13). In particular, given any two good sets
\316\245
at
be empty
and
least
one of Y'\\Y and Y\\Yf
must
Y7,
(since
of each
In other words,
they are both strict upperbounds
other).
\316\251
ordered
is totally
by set inclusion: given any two good sets \316\245

bound

and

for

and

Y,

\320\223',either

of Y\\Y' is

element

every

\316\245.
\320\243\321\201\320\223\320\276\320\263\320\223\321\201

Yqo is the set of all elementsof X which


i.e.,
\317\205\316\251,
least
one
xo G Yoo\302\267Also,
belong
good subsetof X. Clearly
since
each good subset of X has xo as its minimal
the
element,
set Yqq also has xo as its minimalelement(why?).
Let

:=

Yqo

to at

that

show

we

Next,

elements

of

good set

\316\245
and

Yqq.

By

definition

in some

xf lies

Let

totally

ordered.

of Yqo,

we know

is

Yqq

good set Y7.

that

be
two
\317\207,\317\2071

in some

\317\207
lies

since

But

\316\251
is

totally

these good sets containsthe other. Thus


are
x^x
in a single good set (either\316\245
or
since good sets
Y7);

one of

ordered,
contained

are totally ordered, we

either

that

see

thus

<
\317\207

xf or

<
\317\2071

\317\207
as

desired.

we show

Next,
nonempty

lies in Y^.

then

Then
the
that

\316\221
\316\245
is
\316\240

set

bound

of

is

Yqo

Let

well-ordered.

A be

any

pick an elementa A,
Therefore there is a goodset such that
we can

has

other good
for

Y,

\316\245'
for
\316\221\316\240

and

of

a minimal

set

Y\\

any

of

good

Y\\

since

\316\245
is

element, call it

every

element

a eY.

well-ordered,

b.

of. Y'\\Y

Now

is a

recall

strict

is larger than b. Sinceb


this implies that b is also a minimal
set Y' (why?). Since every
element

in particular

\316\221
\316\240
F,

which

\316\245

subset

a non-empty

element

a minimal

that

Yqq. Then

\316\221
\316\245
\316\240
thus

for any

upper
element

of

subset

is

of

to

belongs

we

\316\2631

237

sets

Ordered

S.5-

and
hence
\320\243\320\276\320\276

strict

with

bound

upper

see Exercise8.5.11)and has


element (why?). Thus this set is good,and

ordered (why?
in

contained

But

Yqq.

strict upper bound


no

strict

8.5.15

ordered set,
an upper
of X has

the

have

be

therefore

must

consequence:

important

Let X

that

property

its minimal

since s(Yoo) is a
constructed
a set with
D

as desired.

(Zorn's lemma).

with

as

xo

contradiction

Thus
we
\320\243\320\276\320\276.

for

bound,

upper

is a

this

The above lemmahas the following


Lemma

is

Yqq

xo as its minimal element, it


then Yqq U {s(Yoo)} is wellBut
s(YQO).

is well-ordered
Since \320\243\320\276\320\276
has

Thus

as claimed.

well-ordered
a

a minimal elementof A.

b is

that

see

thus

least one good set

to at

belongs

every

be

partially

non-empty

subset \316\245

totally ordered

bound. Then X containsat

leastone

maximal

element

Proof.
We

some

give

principleof

set

Exercise

partially
8.5.2.

of Zorn's

applications

the

with

lemma (also calledthe


below.

exercises

the

order

empty

relation

no elements).Is
empty
ordered?
ordered?
well-ordered?
totally
Explain.
< such that
of a set X and a relation
Give examples

(b) The relation< is reflexive


(c) The relation < is anti-symmetric
Exercise 8.5.3. Given
m,

the

set 0

empty

two

and write

set has

the

because

(a) The relation < is reflexive

\316\267
divides

in

induction)

transfinite

Exercise 8.5.1. Consider


<0 (this relation is vacuous
this

8.5.14.

See Exercise

and

anti-symmetric,

and

transitive,
and

positive

integers

n|m, if there

but not
but not

anti-symmetric;
but not

transitive,

\342\202\254
\321\211\321\202
N\\{0},

existsa positive

transitive;

integer

reflexive.

we say
a such

m = na. Show
the set N\\{0} with
that
the
relation
ordering
partially ordered set but not a totally ordered one. Note that this
< ordering of N\\{0}.
different ordering relation from
the
usual

Exercise8.5.4.Show
0} have

no minimal

that

the

element.

set of

positive reals R+

:= {x

that
that
| is a
is a

R : \317\207
>
\342\202\254

8.

238

Infinite

sets

from one set X to another


\316\245
Let / : X \342\200\224>
be a function
Y. Suppose
is partially
ordered with some ordering relation
set
that \316\245
if and
Define
a relation <\317\207on X by defining \317\207
<\317\207\317\207'
<y.
only if
X into a
that this relation <\317\207turns
Show
<\316\263f(x').
f(x)
partially
If we know in addition that the relation <\316\263makes
set.
ordered
\316\245
X
does this mean that the relation
<\317\207makes
ordered,
totally
totally
needs to be made on
additional
ordered also? If not, what
assumption
X totally
ordered?
<\317\207makes
/ in order to ensure that
8.5.5.

Exercise

Let X be a partially

Exercise 8.5.6.

ideal (x)

the order
(X) :=
be the

map f(x)

only if f(x)

f(y).

by set

that

Exercise 8.5.7. Let X be a partially

Exercise 8.5.8. Show


set

Conclude

in

\317\207
to

/ :X
its order
that

\302\243
X,
\320\266,
\321\203

define

X,

<

\321\203 \317\207}.Let
\342\200\224\342\226\272

(JjQ

ideal.

\317\207
<x

if
\321\203

shows that any partially ordered


of sets whose ordering relation
is

inclusion.

ordered subset of X. Show


at most one minimum.
ordered

any

given

exercise
a collection

by

of

element

every

This

\320\241

set can be represented


given

sends

is a bijection,and

Show that /
and

:= (x) that

in
\317\207

:= {y \302\243
:
X
of all order ideals,and let

set

be the

:
\302\243
{(\317\207) \317\207
X}

For any

set.

ordered

set (x)

be the

to

X
\320\241

has

that

particular

every

finite

and a

maximum.

that

a minimum

every

set, and let


at most one

ordered
\316\245
can

that

finite

have

non-empty

a totally
and

subset of a totally

(Hint:

ordered

totally

\320\243
be
maximum

use

induction.)
set is well-ordered.

set such that every non-empty


Exercise 8.5.9.LetX be a totally
ordered
subset of X has both a minimum and a maximum.
that X is finite.
Show
X is infinite.
assume
for sake of contradiction that
Start with
(Hint:
the
minimal
element xo of X and then construct an increasing sequence

xo < x\\

...

<

in X.)

Exercise 8.5.10.
choice.

(Hint:
\316\245
:=

and

show

Prove

is false for

: P(m)
\316\247
\302\243
{\316\267

that

\316\245
being

Exercise8.5.11.

Let

be

subsets of X.
totally ordered.

is

Exercise 8.5.12. Let


relations
Cartesian

product

and
<\317\207

<y

and

using the

axiom of

X with
\342\202\254

<\317\207
\316\267},

a contradiction.)
Y'
and
partially ordered set, and let \316\245

Show

defining

lead to

\316\245
U Y'

that

\316\245
be

respectively.

\317\207
\316\247
\316\245
by

some m

would

non-empty

well-ordered
if it

without

8.5.10,

Proposition

the set

consider

partially

Define a
(x,y)

if and

is well-ordered

ordered

sets

with

be

only

ordering

relation <\317\207\317\207\316\263
on the

<\317\207\317\207\316\263
(x\\y')

if x

<x

x\\

or

if

2=

and
\317\207'

to the

similar

the lexicographical

is called

<\316\263\321\203'.(This
\321\203

is
and
\321\205\320\243,

jf

239

sets

Ordered

S.5.

alphabetical ordering of

ordering on
a word

words;

in a dictionary
than
another
word w' if the first
letter
appears earlier
than
in
the
the
first
of
or
if
earlier
letter
the
first
\320\263\321\203',
alphabet
0f w is
of w is earlier
the
second
letter
match
and
than
the
second
letter
letters
so forth.)
Show that <\317\207\317\207\316\263
defines
a partial
and
ordering on
of \302\253/,
if
\316\245
show
that
X
and
are
then so
ordered,
totally
X*Y. Furthermore,
\316\245
\317\207
and
if
X
and
are
then
so
is
\316\247
\316\245.
\317\207
well-ordered,
is \316\247 F,
Prove the claim in the proof of Lemma
8.5.13.
8.5.14,
namely
Exercise
is an upper bound for \316\245
and
vice versa.
that every element of Y'\\Y
(Hint: Show using Proposition 8.5.10that

all

for

that

with

similarly

that

\316\245
\316\245'.Conclude
\316\240
\342\202\254

Show

exists.

= {yeYnY':y<a}

= {yeY':y<a}

{yeY:y<a}

s{Y

=
\316\240
\316\245')

\320\243
\320\243
\320\237
is

if

Since

Y' interchanged.

\316\245
and

good,

and hence s{Y \316\240


\316\245')
is non-empty, and
and
are disjoint,
Y\\Y'

Y'\\Y
\321\202\321\210(\320\243'\\\320\243)
Y'\\Y

then concludethat one of these sets is empty, at which


point
the claim becomes easy to establish.)
8.5.14 to prove Lemma 8.5.15. (Hint: first
8.5.14.
Use Lemma
Exercise
if X had no maximal
then
that
show
elements,
any subset of X which
has an upper
bound, also has a strict upper bound.)
one can

8.5.15.

lesser or

nite
for

an injective

\320\222
be

equal cardinality
that

prove

induction,
subset

every

A and

Let

Exercise

not have

two

non-empty
B. Using
or equal
lesser
to

\320\222
has

sets such that A does


the principle of transfi-

cardinality to A.

denotethe property
map from X to A.) This exercise(combined
let P(X)

XC5,

8.3.3) shows that the cardinality of


of choice.
as one assumesthe axiom

Let X

two

any

sets is

there

that

with

(Hint:
exists

Exercise

comparable, as long

be a set, and

\316\241
let
be
the set of all partial
:=
the usual
then
both
partial
orderings
N\\{0},
in
of
and
Exercise
the
8.5.3,are elements
partial ordering
ordering <,
\316\241
is
coarser
than
another
We
that
one
<G
partial ordering
P.)
say

Exercise

8.5.16.

(For instance, if X

of X.

partial ordering

if
<\316\210 \316\241

<r
[\317\207
[x < y) =>
y).
than
Exercise8.5.3is coarser

is coarser

than

set; thus the

set of

if

<

is exactly
maximal
elements
There

we have the
P,
any x,y \342\202\254
Thus for instance the partial

for

the

usual

<'. Show that


partial

one
of

minimal
\316\241
are

orderings
element

precisely

ordering

\302\267<
turns

of

\316\241
into

a partially

ordered

is itself partially
ordered.
the
what
is
it?
Show
that
P;
total orderings of P. Using

on X
the

implication

ordering in
<. Let us write <^<'

8.

240

Zorn's lemma, show

that

given

ordering <' such that <


Exercise 8.5.17.UseZorn'slemma

a total

to

give

< 1
#(Y \316\240
Xa)
most one element.UseZorn'slemma
Zorn's lemma and the
that
Deduce

equivalent (i.e., they

be

can

Exercise8.5.18.Using

to

locate
of

axiom

deduced

of

\320\241
\316\245

\\JaeI
intersect

there

proof of

another

set of all
be the
(Hint: let \316\251
for all a \302\243
all
sets which
/, i.e.,

8.4.2.

Exercise

<

any partial ordering


is coarser than <'.

Infinite

the

exists

claim

in

such that
each Xa in at
Xa

a maximal element
are in fact

choice

from each

sets

of \316\251.)

logically

other).

lemma,
prove Hausdorff's maximality
ordered
partially ordered set, then there existsa totally
\316\245
of X which
to set inclusion (i.e. there
is maximal with
subset
respect
Y' of X which contains Y. Conversely,
is no other totally
ordered
subset
if Hausdorff's maximality principle is true, then Zorn'slemma
show
that
is true.
are logically
Tthus
by Exercise 8.5.17, these two statements
to
of
the
axiom
choice.
equivalent
if X

principle:

{Y,

<),

is a

Let X

8.5.19.

Exercise

Zorn's

where

\316\245
is

be a set, and

a subset

(Y', <') are

of X

let

\316\251
be

the

and < is a well-ordering

space
of

of all
Y.

pairs

If (Y, <)

of \316\251,
we say that
and
elements
segment
(Y, <) is an initial
of (Y', <') if there exists \320\260\320\273\320\266\320\265\320\243
\316\245
:=
\316\245'
: \321\203
that
such
\302\243
< \317\207}
{\321\203
<
\320\241
in
\316\245
if
and
for
and
\320\225
particular
any \321\203,
\321\203' \316\245,
\321\203 \321\203'
(so
\316\245'),
only if
\342\226\240<
<'
a
\316\251
if
Define
relation
on
either
^
V
\321\203'\302\267
by defining
(Y, <)
(Y7, <')
Show
that
{Yr, <'), or if (Y, <) is an initial segment of (Y',
of
\316\251.
is
minimal
There
one
element
of
partial ordering
exactly
\316\251;
are precisely
what
is it? Show that the maximal elements of \316\251
the wellorderings (X, <) of X. Using Zorn's lemma, conclude the well ordering
{Y,

<)

<').

\302\267<
is a

has at leastonewell-ordering.
use the
Conversely,
the
of
Axiom
to
axiom
8.1.
choice,
well-ordering principle
prove
(Hint:
< on Uae/^a> anc^ tnen
a well-ordering
consider
the minimal
place
elements of each XQ.) We thus see that the axiom of choice,Zorn's
the well-ordering
and
to
lemma,
equivalent
principle are all logically
principle:

every

set X

each other.

of
Exercise 8.5.20. Let X be a set, and let \316\251\321\201
be
2\321\205
a collection
subsets of X. Using
Zorn's
there
is
show
that
a
subcollection
lemma,
\320\241
\316\251'
\316\251
such
that
all the elements of \316\251'
are
from each other
disjoint
= 0 whenever
\316\221
\316\222
of \316\251'),but that
\316\240
\320\222
are
elements
distinct
\320\220,
(i.e.,
all the elements of \316\251
at least one element of \316\251'
for all
intersect
(i.e.,
A \302\243
\320\241
there
exists
\316\251'
G \316\251
that
A ^ 0). (Hint: consider all the
such
\320\241
\320\237
of \316\251
whose
subsets
elements
are all disjoint from each other, and locate
a maximal
element of this collection.) Conversely,if the above claim is
and thus this is
true, show that it implies the claim in Exercise
8.4.2,

S.5.

yet

Ordered sets

another

let
(Hint:
I and
a \302\243

claim
\316\251
be

xa

the

241

which is logically equivalent to the axiom


of choice.
set of all pair sets of the form {(0, a), (1, xQ)}, where

\316\225
\316\247\316\261-)

Chapter

In

previous

sequence

(an)\342\204\242=0

number n.
N

to

as take

or

convergent),

did

then

We

such

R,

be

can

which assigns a

an object

i.e.,

been focusing
as a
viewed

we have

chapters

form

the

(if

infinity

the

function

was

the sum
the series was

assuming

(again,

natural

functions from

or computed

etc.,

infima,

each

to

an

with these

things

their limit at
sequence

primarily on sequences.
N
function from
to R,

number

real

various

suprema,

of all the elementsin

on

functions

Continuous

convergent).

are

which

such

continuum1

{x

look at functions not on the natural numbers


\"discrete\", but instead look at functions on a
as the real line R, or perhaps on an interval
such

we will

Now

R :
\342\202\254

a <
on

operations

<
\317\207

these

b}.

a number of
limits,
including taking
computing
In

integrals.

evaluating

as

we will perform

Eventually

functions,

derivatives,and
primarily on limitsof
of a continuous

N,

and

functions,

on

this
the

we will focus
closely related concept
chapter

function.

Before

We

in this

we

discuss

functions,

for subsets of

notation

will

not

text, but

rigourously
roughly

though, we must first

out

set

some

the real line.


define the

speaking

a set

notion

of a

discrete

is discrete if

each

set or a
element

continuum

is separated

from the rest of the set by some non-zero distance, whereasa set isa continuum
if it is connected and contains no \"holes\".

of the

Subsets

gj.

real line

243

Subsets of the realline

g.l

we do not work on the whole


real
subsets of the realline,such as the positive
with
the
work
R : x > 0}. Also, we occasionally
\302\243
[\317\207
of that
real line R* defined in Section6.2,or in subsets
in

often

Very

line

analysis

but on certain

R,

real axis
extended

extended

real line.

There are of courseinfinitely


indeed, Cantor's theorem(Theorem
shows that there are even more
However,

numbers.

extended

the

(and

intervals.

numbers.

closed interval

the

define

[a, b] :=
the

intervals

half-open

[a, b) :=
the

and

{x
open

:
R*
\342\202\254

a <

intervals

call

a the

\317\207
<

a <

and (a, b]
\320\254};

(a,

real

extended

R*
be
\342\202\254

[a, b]

:
R*
\342\202\254

{x

[a, b)

by

<
\317\207

\320\254},

by

b] :=

{x

:
R*
\342\202\254

a <

<
\317\207

\320\254},

(a, b) by

(a, b)
We

there are

than

sets

such

Definition 9.1.1 (Intervals).Leta, b


We

real line;
8.3.4)

Exercise

also

see

8.3.1;

real
are certain
there
special subsets of the real line
real line) which arise quite often. One such

are the

of sets

family

of the

subsets

many

:=

R*
\320\262

{x

left endpoint of

<

: a

\317\207
<

\320\254}.

these intervals,and b

the

right

end-

point

Remark
notation;

9.1.2. Once again, we are


for instance, we are now

an open interval
Cartesian

plane

from

R2

ambiguity, but the

3, as

to

:= R

\317\207
R.

reader

should

This

to

(2,3)

using

well as
still

the

overloading

an ordered pair in

can
be

parenthesis
denote both
the

some genuine
to resolve which

cause
able

from
context.
In some
meaning of the parenthesesis intended
issue
is
resolved
reversed
brackets
instead
of
this
texts,
using
by
now be [a, b[, (a, b] would
parenthesis, thus for instance
[a, b) would
be ]a,b],
and (a, b) would be ]a,b[.

9.

244

9.1.3. If a and

Examples
or

+oo

line,

instance

for

real axis

{x

and

(\342\200\224oo,
0),

\317\207
>

0}

axis

{x

functions

numbers

(i.e., not

above intervals

of the
=

on ft

equal to
are subsetsof the real

R : 2 < \317\207
The positive
<
{x \342\202\254
3}.
is the open interval (0, +oo), while
the
R :
\342\202\254

>
\317\207

is the

half-open
R : \317\207
<
negative real axis {x \342\202\254
0} is
<
R : \317\207
real axis {x \342\202\254
0} is (\342\200\224oo,
0].

the
Similarly,
the non-positive

+oo).

[0,

[2,3)

real

non-negative
interval

all

then
\342\200\224oo)

real

b are

Continuous

0}

R itself is the openinterval


(\342\200\224oo,
+oo),
the extended
real line R* is the closedinterval
[\342\200\224oo,+oo].
sometimes
refer to an interval in which one endpointis infinite
the

Finally,

(either

real line

or

+oo

as

\342\200\224oo)

both endpoints are


intervals are bounded
the

and

positive

Thus

real axes

negative

all

then

set

empty

and

(a,b],

[a,b),

half-infinite

(a,b)

of the

four

If

(why?).

are the

empty set, while

set {a} (why?). Becauseof this, we


most (but not all) of our analysis
degenerate]
to non-degenerate
intervals.

restricted

and

intervals [a, b], [a, b),


a = b, then the three

the singleton

just
intervals

interval,

intervals,

intervals.

Example 9.1.4. Ifa >


(a, b], and (a, b) are the
intervals

[2,3)

are

bounded

is a

We

in which
all other

intervals;

doubly-infinite

intervals.

R* are infinite

R and

as

infinite

and intervals

intervals,

half-infinite

while

is

[a,b]
call

these
be

will

Of courseintervals
are
not the only interesting
subsets of the
real line. Otherimportant
the natural
include
numbers
examples
the
the
form
and
can
rationals
One
additional
N,
Q.
integersZ,
sets using such operationsas union
and
intersection
(see Section
union of two
3.1), for instance one couldhave a disconnected
intervals such
as (1,2) U [3,4], or one couldconsider
the
set
nQ
[\342\200\2241,1]
1 inclusive.
numbers between \342\200\2241
and
Clearly there
are infinitely many possibilities of sets onecouldcreate
such
by

of rational

operations.

Just as sequences of real numbers have


real numbers have adherent points,which
we

9.1.5

Definition

let
there

\316\265
>

0,

exists

and

let

\317\207
R.
\342\202\254

\320\260
G X
\321\203

which

Let

points).

(\316\265-adherent

say

that

\316\265-close

to

We

is

limit

now

\317\207
is
\317\207

(i.e.,

points,

sets

be a

subset of R,

\316\265-adherent to
\\x

of

define.

\342\200\224
<

y\\

\316\265).

iff

flemark 9.1.6. The


adherent

an

use

standard

is not

\"\316\265-adherent\"

terminology

However, we shallshortly
which is standard.
point,

the literature.
0f

245

real line

of the

Subsets

gj.

define the

it to

in

notion

tothe open interval


Example 9.1.7. Thepoint1.1is0.5-adherent
not
to
The
but
is
0.1-adherent
this
point
interval\302\267^why?).
(0,1))
to
the
finite
The
set
0.5-adherent
point 1 is0.5{1,2,3}.
1.1 is
to

adherent

\317\207
G

let

(Adherent points). Let X

9.1.8

Definition
and

(why?).

{1,2,3}

R.

to X

5-adherent

We

say that

\317\207
is

an

adherent

be a subsetof R,
of X

point

iff

is

it

for every \316\265


> 0.

to
The number 1 is \316\265-adherent
the
open
for every \316\265
and is thus an adherent point
> 0 (why?),
interval (0,1)
The
0.5 is similarly an adherent point of (0,1).
of
point
(0,1).
2 is not 0.5-adherent (for instance) to (0,1),
the
number
However,
not an adherent point to (0,1).
is thus
and

9.1.9.

Example

Definition9.1.10(Closure).
Let

of

adherent

points

X is

denoted

sometimes

X,

be a

The closure
the set ofall

subset of R.

defined to be

the

of X.

9.1.11 (Elementarypropertiesof closures).


Let
XUY
subsets
ThenX
o/R.
=IuF,
arbitrary
<^X,

Lemma
be
\316\245

XPlYCXnY.

We

now

Exercise 9.1.2.
some

compute

and
and

thenXCY.

IfXQY,

Proof. See

closures.

Lemma 9.1.12 (Closures of intervals).Let a < b be real numbers,


and let I be any one of the four intervals
(a, 6], [a, b), or
(a, b),
I
is
the closureof
the
closure
Then
of
Similarly,
[a,b].
[a,b].

(a,

or

oo)

is

(\342\200\224oo,a].

of
use

is [a,

will just
(a, b) is [a, b]\\
Exercise

oo); while the closure of(\342\200\224oo, a)

the

Finally,

We

Proof
closure
can

[a, oo)

closure

of

show one of
the

9.1.1).

other

(\342\200\22400,00)

is

(\342\200\224oo,
a]

(\342\200\22400,00).

these facts, namely

results

or

are proven

that

the

similarly (or one

Continuous functions on fi

9.

246

every element of [\320\271,


b] is adherent
to
then
it
is
definitely
adherent to
6)
= b then
\317\207
to (a, b) (why?).
If \320\266
is also
adherent
(a, b).
Similarly
=
to (a, b).
when \317\207 a. Thus every point in [a, b] is adherent
\317\207
that
is adherent
to (a, b) lies
Now we show that every
point
in [a,b]. Supposefor sake of contradiction
that \317\207
does
not lie \321\211
If
\320\266
\320\266
is
then
\320\266
or
\320\266
\316\261.
then
not
either
b
b
>
>
<
[\316\261,
b],
(x
b)not an adherent point to
is hence
adherent to (a,b) (why?),
and
(a, b). Let

(a,

us

let

First

that

shows

\317\207
is

closure

T/ie
the

empty

\320\266
\316\225
(a,

\317\207
then
< \316\261,

\320\266
is

not

\342\200\224

(a

to

rr)-adherent

adherent point to (a,b). This


fact in [a,b] as claimed.

in

The closure

9.1.13.

Lemma

that

If

not an

is hence

and

if

Similarly,

b).

show

\317\207
\342\202\254
[a, b].

o/N is N.

is R; and \316\257/ieclosure
0 is 0.

o/Q
set

6)?

is R.

ofR

\342\200\224

contradiction

closure

TTie

(a

of

\316\226
is

Proo/. See Exercise9.1.3.


The

following

obtained

be

can

an

adherent

Let

be a

of

there

if

only

exists

Then

a sequence

which converges to

in X,

elements

points of a set X

R.
and let \317\207
\342\202\254

subset o/R,

point of X if and

(an)%Lo>consisting entirely

of

lemma
shows that adherent
as the limit of elements in X:

Lemma 9.1.14.
\317\207
is

Z.

T/ie closure

x.

See Exercise

Proof

9.1.5.

Definition 9.1.15. A subset \316\225\320\241


R
or in other words that \316\225
all
contains
Examples 9.1.16. FromLemma
real numbers, then
[a,b],
[a,+oo),
closed,

Prom

while

Lemma

(a,b),

9.1.13

=
to be closedif \316\225
its adherent points.

is said
of

9.1.12

we

see that

if a

< b are

and
(\342\200\224oo,+oo)
(\342\200\224\316\277\316\277,\316\261],

(a,b], [a,b), (a,+oo), and


we see that
0 are closed,
R,

are

not.
are
(\342\200\224\316\277\316\277,\316\261)

\316\235,\316\226,

while Q is

not.

Prom Lemma9.1.14

we

sequences:

E)

can

define

closure

in terms

of

real line

of the

Subsets

gJ.

Corollary 9.1.17.
*5 a
(on)\302\243Lo

limn-\302\253x>an
convergent

as well,

in X,

9.1.18

is an

adherent

of X

if

for

adherent

Let X be a subsetof the


point (or a cluster point) of X

point of X\\{x}.

We

that

say

exists some

there

and

chapter, we shall
point by the closely

next

the

points).

a limit

\320\266
is

X
\342\202\254

every
in X

point

(Limit

that

We say

all

limit

an

of

concept

relatednotionofa
Definition

in

differentiation

need to replacethe

then

closed.

is necessarily

then

If X is closed,and

subset o/R.

sequence consisting of elements


convergent
a^so
ties in \316\247- Conversely,
if it is true that
in X has its limit
dements
sequence
(an)\342\204\242=0 \302\260f

When we study

line.

be a

Let

247

\316\265
>

\317\207
is

isolated

an

that

0 such

real
it

iff

point

\342\200\224

|rr

y|

>

\316\265

\342\202\254
\321\203
X\\{x}.

be the set X = (1,2) U {3}. Then 3 is


of X, but it is not a limit point of X, since3
adherent
an
point
\342\200\224 =
adherent
to
X
isnot
instead, 3 is an isolatedpoint
{3}
(1,2);
On the other
of X, since 2 is
of X.
hand, 2 is still a limit
point
= X; but it is not isolated
adherent to X \342\200\224
(why?).
{2}
Let X

9.1.19.

Example

Remark 9.1.20.
of

to

converges
into

set

the

are distinct

X that

in

elements

a sequence

exists

there

iff

out

turns

\320\266.
It

that

points and

of limit

see that

we

9.1.14

Lemma

Prom

point of X

\320\266
is

(an)^=0i consisting

the set of

that

such

and
\320\266,

from

limit

entirely
(an)\302\243Lo

adherent points splits

the set of isolatedpoints(Exercise

9.1.9).

9.1.21.
I
Let I be an interval (possibly
i.e.,
infinite),
set of the form (a,b), (a,b], [a,b),[a,b],(a,+oo);
[a,+oo);
or
element
Then
every
of I is a limit point of
(\342\200\224oo,a);
(\342\200\224\316\277\316\277,\316\261].

Lemma

is a

I.

Proof.

show

We

similar and
\317\207
is

a limit

and

x = b.

sequence

this

for

the case

the

reader.

are left

to

point

of /.

If

=
\317\207

converges

Let

consider

and
\320\266,

will

the

sequence

lie inside

cases

are

to show

that

other

the

[a,b]\\

\317\207
\342\202\254
/; we

have

=
There are three cases:\317\207

then
\316\261,
to

/ =

\316\261
< \317\207
< \320\254,
\316\261,

(x +

\342\200\224

{a}

~)^=^. This
(a, b] if N

is

9. Continuousfunctions

248
chosen

a <

a limit

\316\261
is
\317\207
<

instead

point

we

Thus by Remdrk 9.1.20we see that


of [a,b]. A similar
works when
argument
b one has to use the sequence(x \342\200\224
^{JL^

argument is otherwisethe same.

but the

(why?)

Next,

=
\317\207

When

b.

have

the real line is

X of

A subset

sets).

(Bounded

if we

bounded

of a boundedset.

the concept

define

Definition9.1.22
said to be

\320\246,

(why?).

enough

large

=
\317\207

on

real

some

for
\320\241
[-\320\234, \316\234]

number

M>0.

9.1.23. For any

Example
is

bounded,

max(|a|,

|\320\254|).

However,

(why?).

In

unbounded

unbounded
A

(a)
(b)

basic

property

9.1.24

fact,

no half-infinite

bounded.

X is

of closed

(Heine-Borel

Then the

of R.

The sets

where

[\342\200\224M,M],

interval

[0,+oo)

interval or
\316\235,\316\226,
Q,

\316\234
:=

is

doubly

and

are all

and bounded
theorem for

following

two

sets is the following.


the line). LetX

statements

are

be

equivalent:

closed and bounded.

Given
values

half-infinite

the

the interval [a, b]

a, 6,

(why?).

Theorem
subset

be

can

interval

infinite

numbers

real

contained inside

it is

because

which
takes
sequence (an)\302\243Lo \302\260freal numbers
a
X
all
exists
there
\342\202\254
for
subsequence
(i.e., an
n),

any
X

in

(a7ij)jio

^\320\265original
\302\260f

sequence,

which

converges

to some

number L in X.

Proof.See

Exercise

9.1.13.

Remark
9.1.25. This theorem shallplay a key role in subsequent
sections of this chapter. In the language
of metric
space topology,
it asserts that every subset of the real line
and
which
is closed
version
bounded, is also compact;seeSection12.5.A more general
of this theorem, due to EduardHeine(1821-1881)
Borel
and
Emile
can be found in Theorem 12.5.7.
(1871-1956),

gj.

Subsets

249

line

real

the

of

Exercise 9.1.1. Let X be any subset


Show that
such that ICFCI
Exercise

9.1.2. Prove

Exercise

9.1.3.

0f

Q,

you

real line,

and let

\320\243
be

a set

X.

Lemma 9.1.11.
Lemma

Prove

will need

of the
=
\316\245

(Hint: for computing the

9.1.13.

closure

Proposition 5.4.14.)

Exercise9.1.4.Give

an

of two

example

subsets

\316\245
of
\316\247,

the

real line

such

thatXrTF^xnF.

Exercise9.1.5.Prove
two implications

Lemma

here you

Exercise

9.1.6. Let

Furthermore,

X
then

X).

\316\245
also

set which

Exercise 9.1.7.
closed subsets

9.1.14.
need

X be a subset
X.

contains
contains

will

of the
one
(Hint: in order to prove
axiom of choice, as in Lemma
8.4.5.)
of

R.

Show

that

X is

closed(i.e.,

show that if \316\245


is any closed set that contains X,
Thus the closure X of X is the smallest closed

X.

> 1 be a positive integer, and let -\316\247\316\212,...


Let \316\267
,Xn
of R. Show that X\\ U X2 U ... U Xn is also closed.

be

and for each a \302\243


/
Exercise 9.1.8. Let J be a set (possibly
infinite),
a
R.
be
subset
Show
the
closed
that
of
intersection \320\237\320\260\320\265/
let Xa
^\320\260
in (3.3)) is also closed.
(defined

Exercise9.1.9.LetX

be a subset
of the real line, and \320\266
be a real number.
adherent point of X is either a limit point or an isolated
that
show
of X, but cannot be both. Conversely,
every limit point
point
and every isolated point of X is an adherent point of X.
Exercise9.1.10.If X is a non-empty subset of R, show that X is
if and
bounded
only if inf (X) and sup(X) are finite.
Exercise
9.1.11.
Show that if X is a bounded
subset
of R, then the
closure X is alsobounded.

Show that every

of bounded
Exercise9.1.12.Show that the union of any finite collection
set. Is this conclusion still true if one
subsets of R is still a bounded
of R?
takes
an infinite collection of boundedsubsets

9.1.13. Prove Theorem 9.1.24. (Hint: to show


implies
(b),
(a)
Bolzano-Weierstrass theorem (Theorem 6.6.8) and Corollary
9.1.17.
To show (b) implies (a), argue by contradiction,
using
to establish
that X is closed. You will need the axiom of
9.1.17
Corollary
choiceto show that X is bounded, as in Lemma
8.4.5.)
Exercise

use the

Exercise

9.1.14. Show that any

finite

subset

of R

is closed

and bounded.

9.

250

Continuous

on

functions

be a bounded
subset of R, and let S := sup(\302\243?) be
Let \316\225
E.
the least upper bound
from
the least upper bound of
(Note
principle,
real
a
that
5
is
Theorem 5.5.9,
number.) Show that S is an adherent
point of E, and is also an adherent point of H\\E.

Exercise 9.1.15.

The algebra of real-valued

9.2

functions

line to the

f(x) :=

real line.
+

2x/(x2

/ :R

many functions

with

familiar

are

You

Some

1); f(x)

are:

examples

from

\342\200\224\342\226\272
R

:= x2

f(x)

:= sin(rr) exp(x)

(we will

the real

+ 3x + 5;
sin

define

from
R
exp formally in Chapter 15). These are functions
\317\207
to
number
a
real
number
since
they
every real
assign
single
more exotic functions, e.g.
We
can
also
consider

/w\"

x>

is not

function

This

of

\317\207
purely

by

e^c-5
\320\220
\320\273/'

still a
f(x)

each

f(x).

ifz\302\243Q.

algebraic (i.e., it

in
cannot be expressed

terms

algebraic operations of +, \342\200\224,


notion
in this text), but it is
R, because it still assigns a realnumber

the standard
using
this
we wiH \316\267\316\277^
need

\317\207
G

to

from

function

to

\\0

and
to R

R.

functions
R
/ : R \342\200\224>
any one of the previous
X \320\241
defined
set
R, and restrictthe domainto a smaller
X to R.
a new function, sometimes called /|\317\207, from
R,
creating
This is the same function
as the
function
/, but is only
original
\317\207
on a smaller
domain. (Thus f\\x(x) := f(x) when
defined
G X,
\317\207
and f\\x(x)
For instance,
is undefined when
we can
\302\243
X.)
:=
which
is
from
R
restrict the function
defined
x2,
initially
f(x)
:
to R, to the interval [1,2], thus creatinga new
function
/|[i,2]

We

can take
on all of

\342\200\224\321\203
which

[1,2]

R,

is undefined

is defined

elsewhere.

One couldalsorestrict
\316\245
of

R,

provided

instance, the
could also be
Y. For

thought

of a

function

from

the

of course

function

of

to

R.

: R

a function

\321\2172
when

R to

from

range

that all the


/

as

as /|[i,2](^)

\317\207
G

[1,2]

but

some smaller subset

values

of

defined

\342\200\224>
R

from R

f(x)

lie inside

by f(x)

:= x2

to [0,oo), instead

Formally, these two functions are

The

2.
\320\264.

the distinction
be careless about

that

often

shall

we

251

functions

but

functions,

different

is so minor

between them

the range of a

in

function

discussion.

our

distinction
value f(x) at a point x. f is a
variable
(which depends on some

and its

number
is rather
: R

free

subtle and we
is the

\342\200\224>
R

operationof

but

function
interval
i.e.,

squaring,

be careless,

often

#2+2\320\266+3\"

when

/ : R

defined

really

-\342\226\272
R

distinction

but there are

between the two. Forinstance,

f(x) :=

and
\320\2662,

then /
= x2 and

[1,2],
f(x)

/,
is a

f(x)

This

x).

it too much,

:=

*s the

/|[i,2]

both performthe
=
x2, but the two
g(x)
and g

consideredthe same
f
g,
domains. Despite this distinction,
and say things like
the
function
we should be saying \"considerthe

and g are not


they have different
/

because
shall

stress

not

will

to distinguish

has

one

restriction of / to the
functions

a function

between

function;

when

times

there is a

speaking,

Strictly

if

of real-valued

algebra

function,

\316\246
we

\"consider

function

by f(x) :=

\320\2662+2\320\266+3\".

(This

distinction

like
more of a differencewhen we start doing
things
R is the function
For instance, if / : R \342\200\224>
f(x)
of / at 3 is
course/(3) = but the derivative
of 9 is of course 0, so we cannot simply
derivative

9,

makes

differentiation.

= x2, then of

6, whereasthe
\"differentiate

= 9 and concludethat 6 = 0.)


R is a function,
we
If X is a subset of R, and / : X \342\200\224\342\226\272
:
function
this
x
of
the
the
can form
G
/;
graph
{(\320\266,f(x))
X}
and
hence a subset of the Euclideanplane
is a subsetof \316\247\317\207
R,
= R \317\207
R.
can certainly
R2
One
study a function through its
such
graph,by using the geometry of the plane R2 (e.g.,employing
so
will
as tangent
and
We
however
concepts
area,
lines,
forth).
on
a
more
in
which
we
instead
pursue
rely
\"analytic\"
approach,
both

sides\"

of /(3)

The
the propertiesofthe realnumbers
to analyze
these functions.
are complementary;the geometric
offers
approach
and
more
visual intuition,
while the analytic approachoffers
rigour
formalism
Both the geometric intuition and the analytic
precision.
become
useful when extending analysis of functions of onevariable

two approaches

to

functions

of

many

variables

(or possibly

even

infinitely

many

variables).

Just

as numbers

can be manipulatedarithmetically,so can

is a

functions

two

9.2.1

Definition

/ :X

functions

/ + g :X

operations

(Arithmetic
\342\200\224>
R and

g :X

\342\200\224>we

R,

on

functions

of two functions is a ftmction,


function, and so forth.

sum

the

functions:

two

Continuous

9.

252

the

product

\320\246

of

on functions). Given
can define their sum

formula

by the

\342\200\224>
R

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

their

difference

\342\200\224X

g/.

formula

by the

\342\200\224\342\226\272
R

(f-9)(x)--=\342\204\226-g(*),

their maximummax(/,g)

: X

\342\200\224\342\226\272
R

by

\321\202\320\260\321\205(/,\320\267)(\320\266)
:=\321\202\320\260\321\205(/(\320\266),\320\267(\320\266)),

minimum

their

g) :

min(/,

\342\200\224\342\226\272
R

by

:=

\320\2501\320\277(/,\320\267)(\320\266)
\321\202\321\202(/(\320\266),\320\267(\320\266)),

their

/3 :

product

\302\267
\342\200\224\342\226\272
:
R

(or /

(fg)(x)
that
g(x)
the
formula
by

and

(provided

\342\200\224>

R)

by

the

formula

\342\226\240\302\267=/\320\250\321\205),

0
\317\206

all

for

\317\207
G

X)

the

quotient

//3

\342\200\224>
R

Finally,
R

(or

if

real

\321\201
is a

\302\267
\321\201
/ :

number,

\342\200\224>

the

by

R)

we can

define the function

cf

: X

\342\200\224>

formula

(c/)(i):=cx/(i).

Example 9.2.2. If /
5 : R

is the

\342\200\224>
R

function

/3(2;) =

:R

is the

\342\200\224\342\226\272
R

function 3(2;) :=

(/ + g)(x) := x2
2\320\2663.Similarly

while
2\320\266,

: R
/\342\200\2243

fg

f(x) := \320\2662,and
R is the
+ 3 : R \342\200\224\342\226\272

function

then
2\320\266,

/
: R

is the

\342\200\224\342\226\272
R

is the

\342\200\224\342\226\272
R

function (f

function

\342\200\224

g)(x)

:=

is the function (Qf)(x) = 6\320\2662. Observe


while
2\320\266,
6/ : R
\316\271\342\200\224>
\320\266
that
4\320\2662,
fg is not the same function as / \320\276
maps
#, which
\316\271\342\200\224>
nor
is it the same as g \316\277
which
\320\266
2\320\2662
Thus
/,
maps
(why?).
of functions and composition of functionsare two
multiplication
\320\2662

different

\342\200\224\342\226\272
R

operations.

g.3.

identities

following

/ :

R -> R, g

are

true,

and

: R

-> R,

which

ones

latter case,give

in the

a proof;

give

case,

Let

9.2.1.

Exercise

253

values of functions

Limiting

(f +

former

In the

false?

are

a counterexample.

g)\302\260h=(foh)

= {fog)
h)

f\302\260{9 +

Which of the

-> R.

: R
/\316\271

(goh)

+ {foh)

(f + 9)-h=(f-h) + (g.h)

f'{9 +

In

limit L.

to a

converge

used the notionsof


local

Definition 9.3.1
X

be a

\342\200\224\302\273
R

real number.
\316\265-close

We

every

eventual

and

limits

with

that
\317\207
G

Let

the

/ is

function

let

and

number,

\316\265-close

\316\265
>

/ :
a
is

0 be

L iff f(x)

to

X.

:=

x2

\\f(x)

then

3.61 <

is

to

restricted

when

\317\207
G

[1,3]

instead it
is 0.41-close

When

5.
then

interval

\\f(x)

and

of R, let

a subset

be

When the function f(x)

[1.9,2.1],

deal

of functions..

since

\317\207
G

as we
to

\316\265-closeness

\342\200\224

subset

we shall needa notionof \316\265-closeness

the interval [1,3], then it is 5-closeto 4,


<
then 1 < f(x) < 9, and hence
4|
is restricted to the smaller
[1.9,2.1],

to 4, since if

what it

Just

a point.

at

value

(\316\265-closeness).

say

9.3.2.

Example

some

function, let L be a real

L for

to

deal

to

\316\265-closeness

(\320\260\320\277)\342\204\242=0

notion for

on the real line,or on some

\316\265-closeness

of sequences,

with limits

to

a sequence

for

a similar

define

now

to converge to

real line,

of the

what it means

We

/ defined

a function

for

means

defined

6 we

Chapter

+ {f.h)

{f.g)

values of functions

Limiting

9.3

h)

it

f(x) < 4.41, and hence

4| < 0.41.

9.3.3 (Local\316\265-closeness).
Let
X be a subset of R,
L
a
be
real
be
a
let
xo be an
/
function,
number,
that
adherent point of X, and \316\265
> 0 be a real number. We
/
say
Definition
let

: X

is \316\265-close
\316\265-close

to

\342\200\224\321\203
R

to

near

L when

xq iff there
to
restricted

exists

\316\264
>

the set {x G

0 such
X

that

/ becomes

: \\x \342\200\224
xq\\ <

\316\264}.

Continuous

0.

254

: [1,3]

9.3.4. Let /

functions

on fi

f(x) := x2
This
function
is
not
0.1-close
[1,3].
to
to
not
4.
since
is
0.1-close
4,
However,
/ is 0.1to the set {x e
restricted
close to 4 near 2, since
when
[1,3].
\342\200\224
t
o
is
indeed
0.1-close
the
function
4.
<
This
/
\\x
2|
0.01},
is
\342\200\224
when
we have 1.99 < \317\207
because
< 2.01,
and hence
\\x
2| < 0.01,
is 0.1-close to 4.
3.9601 < f(x) < 4.0401,and in particular
f(x)

Example

interval
for instance
/(1)
the

to

restricted

Example 9.3.5.
instance

0.1-closeto 9,

that / is not
0.1-close to 9. However,

we observe

is not

/(1)

/ used in the

function

same

the

with

Continuing

example,

previous
for

function

be4he

\342\200\224\342\226\272
R

/ is 0.1-close

near 3, sincewhen restricted


set
3| <
is the sameas the half-open
interval
(2.99,3]
(why?),
<
if 2.99
function / becomes 0.1-closeto 9 (since
< \317\207
3,
<
8.9401 < f(x)
9, and hence f(x) is 0.1-close to 9).
{x G [1,3] : \\x

to the

since

to 9

\342\200\224

- which

0.01}

the
then

Definition 9.3.6 (Convergence


of functions
at a point).
Let X be
\342\200\224>
a subsetof R, let / : X
R be a function, let \316\225
be a subset
of X,
and
let L be a real number. We
xo be an adherentpointof \302\2437,
say
that
/ convergesjo L at xo in E, and write
f(x) = L,
limx_>xo;x6#
iff

to

is

number

any

In other
0, there

such

that

\\x

at

every

zo, we say

0.

\316\265
>

that /

If /

does not converge


at

diverges

xo,

and

leave

undefined.

limx->XQ.xeEf(x)

\316\265
>

xo for

L near

to

\316\265-close

have

we

words,

exists

\342\200\224

<

xo\\

\316\264
>

that

\\f(x)

(Why

\342\200\224<

L\\

definition

is this

\316\264.

/(#)

limx_>Xo;x6\302\243

0 such

iff

for

for
\316\265

all

every
\317\207
\316\225
\342\202\254

to the

equivalent

one given above?)

Remark 9.3.7.In
that

cases

xo

defined
then

one

To give

not.

or

an example, if

by setting f(x) = 1
has

undefined.
does
\316\225

say

lima,_>Q

Some

authors

not

contain

when

= 0
\317\207

the set \316\225from


converges to L at

omit

will
that

although
makes a difference

L),

it sometimes

instance,

we

this is

f(x)
\320\230\321\202\320\266_,\320\245\320\276

contains

when

many

(i.e., we will just

notation

above

the
\320\276\321\202
\320\266\320\276>

slightly dangerous.For
whether \316\225
actually
: R

is the

\342\200\224>
R

function

and f(x) = 0 when \317\207


\316\237,
\317\206

f(x)
only define the limit limx_>Xo;xe\302\243
a limit
xq (so that xq is now
point

\\B

f(x)

of

\316\225

values

Limiting
\302\243

chosen

would

we

what

denote

more

a slightly
contains
\316\225

that

before

/ is

that

shows

argument

/ :
that

Let

seen

have

yffe

value of
pick a smaller
definition

0.01-closeto 4 near2 (one

to

has

just

\316\264).

unwieldy. However, we can rewrite


of
more familiar one, involving
limits

of a

terms

in

R be the function f(x) := x2.


[1,3] \342\200\224\342\226\272
/ is 0.1-close to 4 near 2. A similar

is rather

9.3.6

Definition
this

which allows the possibility

notation,

general

\320\266\320\276\302\267

9.3.8.

gxainple

to

/(#)
point), or would use limx->xo;xeE
we
have
call 1^6\316\261.\316\277;\317\2076\302\243\\{:\316\225()}
but
/(\320\266),

adherent

an

than

rather

255

of functions

sequences.

let

a function,

E, and let L

a subset

\316\225
be

number.

real

be

be

For

sequence

every

elements

of

(an)^=0

which
\316\225f

In view of
as
af(x) -* \316\257
of

a; -^

converges

corollary:

limn_>oo

f(an)

Remark 9.3.11.
when

adherent point
limit.

be

point

of

statements

(Can

you

\302\243\"or
in
\320\266\320\276

to L at

With the

9.3.10.

Remark

case

\342\200\224>
R

which consists entirely of


to xq, the sequence (f(an))^=0

the above proposition,we

\"/ converges

the following

the

:X

9.3.1.

Exercise

See

then
\320\2660,

let f

of X, let xo be an adherent
Then the following two

convergesto f(xo).

instead

of R;

convergesto L at xo in E.

(a) /

Proof

subset

equivalent:

are logically

(b)

Let X

9.3.9.

proposition

We

\"/

has

x$\\ or

will

a limit

L at xo in

limits

E\"

L\".
f(x)
\"1\316\257\317\200\316\271\317\207_>30

notation of Proposition9.3.9,we
f(%) = L, and limn_,oo

consider

write

if limx^XQ.xeE
= L.
only

sometimes

of a

have
an

function / at xo in

an adherent point of E. When


is not an
xo
then it is not worth
it to define the concept of a
see why there will be problems?)

xq is

9.

256
The variable

9.3.12.

Remark

dummy variable;

could

we

obtain

then

Proposition9.3.9

is a

variable and

f(x)

limx^Xo;xeE

\302\243

(why?).
For instance
one limit at each

at most

have

can

function

a limit

denote

to

on R

corollaries.

immediate

some

has

we now know that

functions

any other

For instance, if
and conversely

= L,

limy^Xo.yeEf(y)

\317\207
usfed

it by

replace

limit.

same

the

exactly

Continuous

point:

Let X

9.3.13.

Corollary

X, let xq

be

adherent

an

Then f

function.

can have

Proof Suppose
and such that

V such

and

/ also

we know

of E,

adherent

point

sequence
xo. Since

(an)\342\204\242=0consisting

of

/ has a limit L at

9.3.9,that

Using the limit laws

o/R, let
let

limit

at

by Lemma 9.1.14that
in

elements

xo

in

E,

to

L.

E)

\316\225
which

there

converges

is a

to

see by Proposition
But since / also has a limit
also
to U. But
converges

we thus

(f(an))\342\204\242=0

of

xq in

E. Since xq is an

limits

of sequences

(Proposition

D
one

sequences,

deduce

can now

the

for functions:

laws

\316\225
be

: X

for functions).
Let X be a subset
let
be
an
adherent
xo
of X,
point of E,
R be functions.
g : X \342\200\224>
Suppose that f

9.3.14 (Limit laws

Proposition
and

for

at xo in

converges

(f(o>n))%Lo

V at xo in E, we see that
this contradicts the uniqueness
6.1.7).

limit

that / has

limit

has

there are two

that

contradiction

of

sake

for

numbers

distinct

be a subset ofR, let \316\225be a subset of


R be a
point
of E, and let f : X \342\200\224>
at most
one limit at xo in E.

a subset

and

\342\200\224>
R

has a limit L at xo in E, and g has a limit \316\234at xo in E. Then


\316\234at
a limit L \342\200\224
f + g has a limitL + \316\234at xo in E7 f \342\200\224
g has
at xq in E, min(/, g)
xo in E, max(/, g) has a limit max(L,M)
has a limit min(L, M) at xq in \316\225and fg has a limit LM at xo
in E.
is a real
If \321\201
number, then cf has a limit cL at xo in E.
Finally,
\316\234is

if

non-zero,

is non-zero
then

f/g

on \316\225(i.e.,
has a limit

g(x)

0
\317\206

for

all

\317\207
G

L/M at xq in E.

E)

and

g3.

values

Limiting

prove the first

proof We just
are

others

the

claim

adherent

point of E, we know

a sequence

of

is an

to

consisting
(\316\261,\316\267)\342\204\242^

/ has a

Since

xo-

that

9.3.9,
Proposition

we

9.3.9

#0 in

again,
desired

as
\316\225

to

converging

in E,

E,

we thus

to

L.

which converges
see by

laws

g)(an))^=o converges to L + M.
implies that f + g has a limit

(since

xo

there is

that

9.1.14

Lemma

Similarly
(g(an))^=0
for sequences (Theorem 6.1.19)

converges

limit

the

((f +
this

that

conclude

Proposition

By

L+M)\\

to Exercise9.3.2.Since

elements

limit L at xo in

(/(an))\302\243L0

convergesto M.

by

has a limit

f+g

(that

and are left

similar

very

257

of functions

was

(an)\342\204\242=0

an

By

sequence

arbitrary

in

at
\316\225

x0).

9.3.14
more
9.3.15. One can phraseProposition

Remark
as
informally

that

saying

lim (/

\302\261g)(x)
x\342\200\224*xq

lim

f,g)(x)

max(

X\342\200\224>XQ

lim

min(f,g)(x)

x\342\200\224>xq

lim

(fg)(x)

X\342\200\224*XQ

lim

{f/g){x)

lim

f(x)

\302\261
lim

x\342\200\224>xq

g(x)

x\342\200\224*xq

= max
lim /(\320\266), lim g(x)
(\\ X\342\200\224>XQ
X\342\200\224*XQ

= min ( lim

lim

g{x)
/(\320\266),
x\342\200\224*xq
\\x\342\200\224*xq

= lim f(x) lim


X\342\200\224*XQ

J
J

g(x)

X\342\200\224*XQ

=
;;m\342\200\224\320\276/\320\270

the restriction \317\207


for
but
we have
G \316\225
dropped
brevity)
(where
true
when
the rightbear in mind that these identities are only
we
hand side makes sense, and furthermorefor the final identity
need g to be non-zero,
and
also
1.2.4
for
some
Example
examples
are

several

of what

be

non-zero.

(See

goes wrong when limits

carelessly.)

manipulated

Using
deduce

to

\320\264(\321\205)
\320\230\321\202\321\205_,\320\226\320\276

the limit
limits.

laws in Proposition 9.3.14we can


First of all, it is easy to checkthe
lim

c =

\321\201

x\342\200\224*XQ'yX\342\202\254b\\,

already
basic

limits

Continuous

0.

258

functions

on ft

and

=
\317\207

lim

xo

x\342\200\224*XQ'yX\342\202\254b\\,

for

any

xo and

numbers

real

laws we can thus

the limit

c.

use

(Why?

Proposition

9.3.9.)

By

that

conclude

x2 =

lim

x0

x\342\200\224*XQ\\x\342\202\254b\\,

ex

lim

= exo

x\342\200\224>xo;x\302\243R,

x2 +

lim

d = x$ + exo

ex +

+ d

\316\247\342\200\224*\316\247\316\270',\316\247\302\243\302\261\316\233,

where

etc.,

If /

still an

(why?).

a smaller

Thus

\320\243
is

any

the

subset

/ will

on a

convergence

set. The

Example 9.3.16. Consider


defined

and

xo in X,

adherent point of \320\243,then

\320\243

on

convergence

L at

to

to L at xo in

real numbers.

are arbitrary

converges

xo is

that

c, d

of

A\"

such

also converge

large set

implies

converse,however, is not true:


function

signum

sgn

: R

\342\200\224>

R,

by

sgn(rz) :=

\342\200\2241

(why?)

sgn(rz)

\317\207
>

if

=
\317\207

if:z<0

{1 -1
Then limx_0;a.6(0)Oo)

if

=
whereas lima._0;xG(_OO)0)
is undefined (why?). Thus it

1 (why?),

and

lima._>0.a,6j^sgn(a:)
to drop
dangerous

of
the set X from
the
notation
limit.
However, in many cases it is safe to do so; for instance,
= \320\266\320\276>
we know that limx_>l
since
we
know
in fact that
.^R^2
x2 = Xq for any set X with xq as an adherent point
limx6xo;x6x
it is safe to write limx_>Xo
Thus
x2 = x%.
(why?).
is

sometimes

Example

9.3.17.

Let f(x) be the function


nX)-

\\

ifar^O.

Then

such

it
to

f(x)

liniz-^o

writing

or

singularities,

in the

instance,

but lim^0;iceR/(x)

0 (why?),

we say

this

(When

(why).
singularity\"

\"removable

is
a
that
has
/
happens,
\"removable
at 0. Because
discontinuity\"
is sometimes the convention when
=

f(x)

Hmx^0.xeK_{0}

undefined

of

259

Limiting values of functions

gj.

exclude xo

automatically

textbook,

is

f(x)

limx_>Xo

the

from

for

set;

for

as shorthand

used

f(x)\302\267)
]irax^Xo;xex-{xo}

other

the

On

hand,

values of the function

relevant. The

on

depend

the

xq are not
intuition:
from

away

this

reflects

proposition

following

values

9.3.18 (Limits are local).Let X be a subset ofH, let


ofX, let xo be an adherentpoint ofE,letf:X-^fl
and let L be a real number. Let \316\264
Then we
> 0.

Proposition
be a subset
\316\225
a function,

be

xo shouldonly

the limit at

near xq; the

have
Um

/(*)

\302\243

X\342\200\224>Xq\\x\302\243.L\302\267

if

and

only

if

f(x) = L.

lim

x\342\200\224>xo\\x
\316\262\316\225\316\267(\317\207\316\277-\316\264,\317\2070+\316\264)

9.3.3.

Exercise

See

Proof

Informally, the above propositionassertsthat


lim
x^XQ\\xeE

the

Thus

values of
the

limit

of a

/ near

f(x)

xo\\

lim

f(x)\302\267
\317\207^>\317\207\316\277;\317\207\342\202\254\316\225\316\223\\(\317\207\316\277-\316\264,\31

on the
function at xo, if it exists,only depends
the
far away do not actually influence
values

limit.

We

now

give a

few more examplesof limits.

Example9.3.19.

the

Consider

\342\200\224>
R

defined

by f(x)

:=

functions

\317\207
+

: R

g(x) :=

2 and

g :
Then

and

\342\200\224\342\226\272
R
\317\207
+

1.

= 4 and
Um^2;x6R/(:E)
limx^2;xe^(X)= 3' We WOUld =Uke t0
use the limit laws to concludethat
J (x)/9(x) 4/3,
limx_,2:reR
or

in

other

words

that

lim^^-^eR

=
f+\317\212

|-

Strictly

speaking,

9.

260

Continuous

functions

on fi

use Proposition 9.3.14to ensure


because
\317\207
this,
+ \317\207
^
\342\200\224
at
is not defined. However,this is
zero
so f(x)/g(x)
1, and
R to a
solved,
g from
by restricting the domain of / and
easily
\342\200\224
smaller domain, such as R {1}. Then Proposition
9.3.14 does
=
\302\267
and
have
we
apply,
\302\247.
f\302\261f
\320\230\321\202\321\205^2;\321\205\320\261\320\232_{1}

we cannot

=
\317\207

fined

:= (x2

by f(x)

/ :R-

the function

Consider

9.3.20.

Example

\342\200\224

\342\200\224

l)/(x

1).

function

This

for every real number except1, so/(1)is undefined.


1 is still an adherent point of R \342\200\224
{1}
(why?),

{1}

\342\200\224>
R

de-

is well-defined
However

the

and

limx_>1.x6j^_ rjj f(x) is still defined. This is becauseon the


R

domain

\342\200\224

l)/(x - 1)

\317\207
+

\342\200\224
=
1)
(x2 \342\200\224
l)/(x
x + 1 = 2.

the identity

have

we

{1}

1, and lim^1;x6R_{1}
Let
/ : R -* R be

Example9.3.21.

I(X)\302\267

\\

ifz\302\243Q.

show that

have

would

= L
\320\230\321\202\320\277_\321\216\320\276
f(an)

whenever

numbers converging to
we
would have
sequence,

of non-zero

lim

n-\302\273oo

On
nonzero

other

the

hand,

numbers
instead

irrational

since

have

0,
\317\206

a limit

we

(y/2/n)\342\204\242=0

is

R. Then

(1/\316\267)^0

another

converging

- but now

of rational

- we

have

is

9.3.2. Prove

Exercise 9.3.3.

Prove

we

such

1.

lim

sequence

of

these numbers are


0 =

0.

n-\302\273oo

n\342\200\224\321\216\320\276

a contradiction.

Thus this function doesnot

at 0.

Exercise 9.3.1. Prove


Exercise

0 in

is a sequence
(\320\260\320\277)\342\204\242=0

Since

to 0

have

for sake

Suppose

at

\320\277-\321\216\320\276

L = lim f(y/2/n)
Since

0.

R.

= lim 1 =

/(1/n)

l)(x

the function

f(x) has no limit at 0 in


L
of contradiction that f(x) had somelimit
We will

(x +

limit

9.3.9.

Proposition

the

remaining

Lemma

9.3.18.

claims

in Proposition

9.3.14.

Continuous

functions

Continuous

g#4

now
\317\210\316\262

one of the most fundamental


- that of continuity.

introduce

of functions

theory

Definition 9.4.1
f

: X

that

261

functions

Let

(Continuity).

be a function.
Let
is
continuous
iff
we
at
xq
/
\342\200\224>
R

lim

notions

subset of

be a

in

the

R, and let

xo be an elementof X.

We

say

have

= f(xo)\\

f(x)

x\342\200\224*xq\\x\302\243X

of f(x) as \317\207
to xo in X exists
converges
and is equalto f(xo)\302\267 We say that / is continuous on X (or simply
iff / is continuous at xo for
X.
We say
xq \342\202\254
every
continuous)
iff
it
is
discontinuous
at
is
not
at
continuous
that /
xo
xq.
in

other

the limit

words,

be
Example 9.4.2. Let \321\201

be

the

xq

we
\342\202\254
R,

real

and let

number,

function f(x) := c. Then for

constant

every

/ :R
real

\342\200\224>
R

number

have

f(x) =

lim

lim

=
\321\201

c=

/(^o)?

x\342\200\224*xo\\xG\302\243\\,
x\342\200\224>xq;x\342\202\254\302\243\\,

thus

/ is

continuous

continuous on R.

or
at every point xo \342\202\254
R,

Example9.4.3.
for
x.
real
every
Let

be the

: R

\342\200\224>
R

number

Then

f(x) =

lim

/ is

continuous

continuouson R.

Example9.4.4.

Let

identity

we
\342\202\254
R,

is

function f(x) :=

have

x = xo =

lim

words /

f(xo),

xq\342\202\254x;x

X\342\200\224*XQ\\X

thus

xq

in other

at every point xq \342\202\254


or
R,

sgn

in Example9.3.16.Then

: R

be the

\342\200\224>
R

sgn(rr)

in other

words /

is

signum function defined


at every non-zero

is continuous

9.

262
of x\\

value

at 1, we

for instance,

have

(using

sgn(rr) =

lim

Continuous

lim

z-a;a:eR

on

functions

Proposition

\320\246,

9.3.18)

sgn(rr)

\320\266\342\200\224>1;\321\217\321\201\342\202\254(0.9\320\224.1)

lim
x-^l;x\342\202\254(0.9tl.l)

= sgn(l).

On the other

does

sgn(rr)

lima,_>0.a,6j^

:R

9.4.5. Let /

Example

\\

Then

by the
xq

ifrc^Q.

discussionin the previous

at 0. In fact, it
number

not continuous

/ is

section,

/ is

that

out

turns

not continuous at

Let

: R

\\

Then / is continuousat
is not continuous at 0.

function

be the

\342\200\224>
R

\320\237\320\245)'

every
However,

ifz<0.
real

non-zero
if

we

number

restrict

There

be a

element
equivalent:

becomes

now
Thus

function

again.

continuous
Proposition

but

(why?),

/ to the right-

hand line [0,oo), then the resultingfunction


/|[o|0o)
continuous
everywhere in its domain, including0.
domain
of a function can make a discontinuous
restrictingthe

real

any

(can you see why?).

Example 9.4.6.

continuous

limit

function

be the

\342\200\224>
R

\320\237\320\245)'

since the

at 0,

is not
continuous
not exist.

sgn

hand,

are several

at

ways to phrasethe

statement

that

\"/ is

\317\207$\"\\

9.4.7
formulations of continuity). Let
(Equivalent
subset ofR, let f : X \342\200\224>
R be a function,
and let xo be an
Then the following three statements are logically
of X.

Continuous

g.j.

(a) f is

at

continuous

For every

(b)

For

(c)

<

\342\200\224\342\226\272
then
\316\267
oo,

\317\207
e

useful

particularly
if

in

are

at

continuous

and g are both

is non-zero

imply

on X)

Let

X be

be functions.

\342\200\224>
R

at

continuous

xq,

then the

#) and fg are also

min(/,

#);

max(/,

g,

g :X

and

\342\200\224>
R

\342\200\224

If g

xq.

if f

of course).
limits.

continuity).

preserves

: X

xq

all the

with the

combined

9.3.14,

\342\200\224>

that

9.4.1, immediately

(Arithmetic

and

functions f + g, f

of Proposition

and
an
\320\266\320\276>

domain of /,
in computing

useful

very

Proposition

Proposition 9.4.9
let
a subset of R,
Then

at

^n *be

^e

in Definition

continuity

X.
\342\202\254

\342\200\224

\\f(x)

\316\264.

\342\200\224\342\226\272
\342\200\224>
as \316\267
oo
/(\320\266\320\276)
(provided

f(an)

The limit laws

Let xq

<

consequence

is continuous

Thus continuousfunctions
of

xq\\

\\x

that

such

\316\261
\316\264
>

elements of the sequence(an)^Lo

definition

f(xo)\302\267

\342\200\224

with

f(an)

with

9.4.1.

Remark 9.4.8. A
9.4.7 is the following:
as

exists

there

0,
all

\316\265
for

Exercise

See

Proof

\316\265
>

limn_>oo

of X

of elements
=

consisting

(an)\302\243L0

we have

= xq,

every

f(xo)\\

xq.

sequence

an

limn_>oo

263

functions

then

continuous

is also

f/g

at xq.

In particular,

the

sum,

denominator

functions

f(x)

\317\207
is

function

g(x)
is

instance,

=
\317\207

+2,

use Proposition 9.4.9to


functions
continuous
:=

the

lot

quotient

show

that

of

the
fact
instance, just by starting from
are continuous, and the identity
function
(Exercise
9.4.2), one can show that the

4x2 +

+
\321\202\320\260\321\205(\320\2663

continuous

=
\317\207

as

For

continuous.

are

constant

that

the

and

zero.

become

not

does

One can

and

minimum,

maximum,

difference,

product of continuous functions are continuous;


of two continuous functions is continuousas long

\342\200\224

2 where

at

\317\207
+

5,

\342\200\224
\342\200\224
\317\2074
for
\321\205\320\263)/(\321\2052
4),

every point of R except the


the denominator
vanishes.

Some other examplesof continuous

functions

are

two

given

points

below.

9. Continuous functions on

264
9.4.10

Proposition

:=

by f(x)

9.4.11

real

number.
Then the
xp is continuous.

:=

f(x)

\342\200\224\302\273
R

defined

9.4.3.

Proposition
a

:R

continuous.

is

&x

See Exercise

Proof.

Then the function /

real number.

a positive

be

I). Let a > \320\276

is continuous,

(Exponentiation

R,

\342\226\241

be
II). Let \317\201

is continuous,

(Exponentiation

function f :

(0,oo)

\342\200\224\342\226\272
R

defined

by

Proof. See Exercise9.4.4.


is a

There
that

namely
exponent

base,

: R

\342\200\224>
R

defined

This

Proof.

\342\200\224
are
\320\266, \317\207

is harder to show;

Exercise

see

15.5.10.

(Absolute value is continuous).The


:=

by f(x)

follows since

\\x\\

|rr| =

function

continuous.

is

the

and

functions

\321\202\320\260\321\205(\320\266,\342\200\224
\321\205)

continuous.

already

The class of

continuous functions is not only

subtraction,

addition,

is

but this

9.4.12

Proposition

than Propositions 9.4.10,9.4.11,


jointly continuous in both the

statement

stronger

exponentiation

the

and

\316\240

multiplication,

and

under

closed

but is

division,

also closed

under composition:

Proposition9.4.13

preserves

(Composition

and

\316\245
be

functions.

subsets

g is continuous at
continuous
at xq.

and let f : X
a point in X. If f

of R;

Let xq be

continuity).

Let X

\342\200\224>
\316\245
R be

\342\200\224>
\316\245
and

is

continuous

f(xo), then the compositiong

\320\276 :

Proof. See Exercise9.4.5.


Example

9.4.14.

Since the

on all of R, and the function

and

at

xo,

\342\200\224>
R

is

function f(x) :=

l is

\320\227\320\266
+

continuous

bx is continuous on all of R,
g(x)
= 53x+l is continuous on all of R. By several
the function
gof(x)
of
the
above
applications
propositions, one can show that far more
8x + 7\\^/(x2
+ 1), are
complicated
functions, e.g., h(x) := \\x2 \342\200\224
also continuous.(Why is this function continuous?)
There are still
:=

g.j.

Continuous

a few

functions though

such as k(x) := xx\\

once

have

Section

15.5.

we

the

Exercise 9.4.1.
by

that axe not yet

(a),(b), and (c) are

which we will

of logarithms,

machinery

Prove

to test for continuity,


easy
be dealt with more easily

can

function

this

9.4.7.

Proposition

do not

you

equivalent,

this can

(Hint:

propositions and

the previous

applying

265

functions

lemmas. Note

have to

see

in

largely be done
to

that

prove

prove all six

but
that
you do have to prove at least three;for instance,
showing
and
will
implies
implies
suffice,
implies
although
(c),
(b), (b)
(c)
(a)
(a)
to do this question.)
this is not necessarily the shortestor simplest
way
equivalences,

Exercise 9.4.2. Let

X be a subset of R, and let ceR.


that
the
Show
R defined
is continuous,
and
by f(x) := \321\201
/ : X \342\200\224>
R defined
is also
show that the identity function g : X \342\200\224>
by g(x) := \317\207

function

constant

continuous.
can
Exercise 9.4.3. Prove Proposition9.4.10. (Hint:
you
6.5.3, combined with the squeeze test (Corollary 6.4.14) and

6.7.3.)
Exercise9.4.4.

Prove

one

(Proposition 9.3.14)
this
for

the

and

real

all

9.4.11.

Proposition

can show that

Proposition

from limit laws

(Hint:

limx_>i xn =

use Lemma

1 for

all

integers

limx_>i
squeeze test (Corollary 6.4.14) deducethat
numbers
p. Finally, apply Proposition 6.7.3.)

Exercise9.4.5.Prove

n. Prom
xp = 1

9.4.13.

Proposition

of R,
and let / : X \342\200\224>
R
be a
Let X be a subset
of X, show that the restriction
function. If \316\245
is a subset
\342\200\224>
\316\245
R of / to \316\245
a continuous function. (Hint: this is a
is also
but it requires you to follow the definitions
result,
carefully.)

9.4.6.

Exercise

continuous
f\\y

simple

a real

Let

9.4.7.

Exercise

>
\316\267

0 be

: R
number. Let \316\241

an integer, and for


the function

each

0 < i

<

\316\267
let

\342\200\224>
R be

\316\267

P(*):=5>**;

i=0

such a function
example

is

P(x)

is known
-

= 6x4

as
3x2

a polynomial
4. Show that

of one
\316\241
is

variable; a typical

continuous.

Ci be

Continuous

9.

266

and

Left

9.5

be

of left

notion

the

two seperate

of as

thought

limits

right

introduce

now

We

functions on R,

and right limits,which


of

\"halves\"

the

can

limit

complete

limx^XQ.xex f(x).

Definition 9.5.1
R,

is

an

: X

and

(Left

function, and let xq be a real


then
we define
of
\316\247
\316\240
point
(\320\266\320\276>
\302\260\320\276),
be a

\342\200\224>
R

adherent

f(xo+) of / at xo by

that

of course

adherent

ID

of

point

of / at xq by

be a subset of
number.

If

\317\2070

the right limit

formula

the

Hm

f(xo+):=

provided

Let X

limits).

right

/(\317\207),
x\342\200\224\342\226\272\320\266\320\276;\320\266\342\202\254\320\220\320\237(:\320\263\320\2

this limit exists.

then
(\342\200\224oo,
\320\266\320\276)>

Similarly, if

define

we

is an

xo

the left limit

f(xo~)

formula

the

lim

f(x0-):=

Hx),

x\342\200\224>xo]xGXn(\342\200\224\316\277\316\277,\316\266\316\277)

and

the limit

that

provided

again

f(xo\342\200\224)

will

we use the shorthand

Sometimes

lim

lim

f{x)\\

lim

f(x)

\321\205\342\200\224^\321\205\320\276;\321\205\320\265\320\245\320\223\\(-\320\276\320\276,\3

/ is clear from

context.

the signum function


9.3.16. We have

Consider

9.5.2.

Example
defined

lim

:=

f(x)

X of

domain

in Example

f(xo+)

\317\207\342\200\224\321\216:\320\276;\320\266\302\243\320\245\320\237(:\320\263\320\276,\32

x\342\200\224>xo-

the

cases

notations

:=

f(x)

x\342\200\224>xo+

when

exists. (Thus in many

be defined.)

not

sgn(0+) =

lim

sgn(rr)

x\342\200\224\342\226\272a:o;a:\302\243Rn(0,oo)

lim

sgn : R
1=

\342\200\224>
R

x\342\200\224\342\226\272a:o;z\342\202\254Rn(0,oo)

and

sgn(0\342\200\224)

lim

sgn(rr)

lim

= \342\200\2241,
\342\200\2241

\316\277\316\277,\316\237)
x\342\200\224\342\226\272:ro;a:eRn(\342\200\224\316\277\316\277,\316\237)
\317\207\342\200\224no;a;GRn(-

while

sgn(O)

= 0

by definition.

9.5.

and

Left

have
/ does not necessarily

Note that

order

for

\320\276\321\202
f(xo\342\200\224)

f(xo+)

R- {0}
=

267

limits

right

\342\200\224^
R

function f(x)

\342\200\224
1

even

is the

(why?),

/(0\342\200\224)

though

Prom Proposition9.4.7
IS a sequence
exists,and

see

we

(an)\302\243Lo

an >

(i.e.,

right

if

Similarly?

all

for

Let xo be

\316\267
e

point

at xo, it is

is continuous
(-\316\277\316\277,^\316\277). /
If

that

9.4.7

(Can you see why?)

from

are

\316\247
\316\240

Proposition

to f(xo).

equal

this

true (compare

is also

converse

xo,

number

and

suppose

of R

subset

be

with

f(xo+) and f(xo~)


at

continuous

Let us

both

of

\316\210\316\231
be a

both equal

are

and

exist

both

a real

containing

is an adherentpoint

that xo

\316\247
\316\223\316\252
^e*
\316\247^
f '\302\267
(\342\200\224\320\276\320\276,\320\266\320\276).

and

(#o>\302\260o)

Let X

9.5.3.

Proposition

Proof

left

6.4.12(f)):

Proposition

is

f(xo+)\302\267

from

clear

both exist and

and f(xo~)

f(xo+)

limn_>oo/(\316\261\316\267)

the
converging
=
limn_>oo f(an)
f(xo~).
\316\240
and
of both \316\247
(xo,
oo)

then

N)

adherent

an

For instance,

to xo

a sequence

IS

if / :
:= sc/|sc|,then /(0+) = 1 and
/(0) is undefined.
that
if the right limit f(xo+)
the
in X converging to xo from

then
\316\267
\342\202\254
\316\235),

all

for

(bn)\302\243Lo

< xo

(i.e., an

x0

at xq in

defined

be

to

defined.

be

to

\316\247
\316\240

If

function.

to f(xo), then f

xq-

write L :=

f(xo).

Then

we have

hypothesis

by

f(x) = L

lim

(9.1)

x\342\200\224>xq
;x\302\243
\316\233\316\240(\316\271\316\277
,\316\277\316\277)

and

lim

= L.

f(x)

(9.2)

\320\266\342\200\224*\321\205\320\276]\321\205\302\243\320\245\320\223\\{\342\200\224\320\276\320\276,\320\266\320

Let

given. Prom
exists ai+
> 0

0 be

\316\265
>

there

that

\316\247
\316\240
for
which
(\320\266\320\236)\320\236\320\276)

all

exists

there

that

know

<

\316\264->

0 such

xq\\

i+.

\316\247
\316\240
for
which
(\342\200\224\320\276\320\276,\320\266\320\276)

\317\207
e

\\x

min(5_,5+);

then

that

<

\342\200\224
\\x

such that

\342\200\224

\\x

(9.1) and Proposition9.4.7,we

xq\\

\316\264
>

\316\264.
Then

0 (why?),
there

\342\200\224

L\\

\\f(x)

Prom
that

\342\200\224

xq\\

<

cases:

know

all

for
\316\265

\317\207
e

(9.2) we similarly
\\f(x)

and suppose

are three

<

\342\200\224

L\\

<

\316\265
for

5_.

Now let

that

\317\207
G

=
\321\205
\317\207
> \320\266\320\276?

:=
\316\264

is such
xo, and

\342\200\224
three cases we know
that
L\\ < \316\265.
\\f(x)
(Why?
the reason
is different in each of the three cases.)By Proposition
as desired.
9.4.7 we thus have that / is continuousat xq,
\317\207
<

xo,

but

in all

9.

268

Continuous functions on \320\246

the signum function in Example 9.3.16,it is


for the left and right limits f(xo~), /(^o+) of a function
possible
at
a
/
point xo to both exist,but not be equal to each other; when
at xq. Thus
this
we say that / has a, jump
discontinuity
happens,
we

As

with

saw

for instance, the signum

existand be

each

equal

and
but

other,

has a removable
at x$. For instance,if

singularity)

function
/w
then

and

/(0+)

equals

discontinuity at zero.
limits
right
f(xo~), /(^0+) to
not be equal to f(xo)\\when
this

that /

we say

happens

left

the

1; thus

a jump

has

function

Also, it is possiblefor

/(0-)

(or

discontinuity
we

/ :

take

removable
to be the

\342\200\224>
R

if:z = 0

{\\0

if

\317\207
0,
\317\206

both exist and equal 0

/ has a removable
discontinuity

(why?),

at

but

/(0)

0.

Jump discontinuities and removable


not
the only way a function can be discontinuous.
discontinuities
are
Another
for a function to go to infinity
at
the
is
way
discontinuity:
\342\200\224>
R
defined
for instance, the function / : R \342\200\224
by f(x) := 1/x
{0}
has a discontinuity
at
0 which
is neither a jump discontinuity
or
to
a removable
when
+oo
singularity;
informally, f(x) converges
oo when
\317\207
\317\207
0 from the right, and converges to \342\200\224
approaches
the
These
of
are
0
from
left.
approaches
types
singularities
as asymptotic
There are also
sometimes
known
discontinuities.
remains bounded but
where
the function
discontinuities,
oscillatory
still does not have a limit near xo. For instance, the function
/ : R -> R defined by
Remark

9.5.4.

/W =has

if:z\342\202\254Q

{J

ifrz^Q

discontinuity at 0 (and in fact at any


This is because the function
does
not

an oscillatory

number

also).

right limits at 0, despitethe

The

of

study

continues further,

but

singularities

play

discontinuities

is beyond
a key role

fact

that

the

function

called

real

other

have

left or

is bounded.

(also
singularities)
the scope of this text. For instance,
in complex analysis.

maximum

g \316\262.The

269

principle

\342\200\224>
be a subset
R be a function,
of R, let / : \316\225
Exercise 9.5.1. Let \316\225
an
adherent
be
of E.
a
definition
of what
down
Write
let
xq
point
and
the
limit
mean
and
for
to
exist
+00 or
would
equal
f(x)
limx_>Xo;xG\302\243
it
\342\200\224^
R is the function
-00. If / : R\\{0}
f(x) := 1/rr, use your definition
state
and prove some
Also,
/(0+) = +00 and /(0\342\200\224)= \342\200\22400.
to conclude
= +00 or L = \342\200\22400.
when
9.3.9
L
of
Proposition
analogue

maximum

9.6

The

In the

previous two

functions were

of

is

It

interval.

continuous

functions enjoy a number

theorem

Definition 9.6.1.

(Theorem 9.1.24).

for
f(x) > \342\200\224M
a real number

9.6.2.

both from
and

there

\317\207
X.
We
\342\202\254

that

\316\234
such

f{x)

function

and

for all

exists

a real

say that

\\f{x)\\ <
is bounded

below.

if

set in the senseof Definition

9.1.22

X.
\317\207
We
\342\202\254

/ is

\316\234for

number
\317\207
e

be

\342\200\224>
R

exists a

say that

\316\234such

boundedif
all

if there

above

from

<M

and let /

subset of R,

that / is bounded

(Why?
is slightly trickier than
\342\200\224\342\226\272
if and only if
R is bounded

only

/ :X

above

all

be a

Let

We say

real number \316\234such that


is bounded from below if

Remark

that

saw

properties,
especially if their domain is a closed
here that we shall begin exploitingthe full power
of

the Heine-Borel

a function.

certainly

that

show

now

useful

other

though

a large number of
not all functions were

we

sections

continuous,
We

continuous.

principle

there

that
exists

X.

if and only if it is bounded


Note that one part of the \"if

the other.)

its

image

Also,

f(X)

is a

function

bounded

(why?).

functions are bounded. For instance,


the
domain
R is continuous but unbounded
it is bounded
on some smaller domains, such
although
(why?),
as [1,2].
The function
f(x) := 1/x is continuousbut unbounded
on (0,1)
(why?), though it is continuous and boundedon [1,2]
if the domain
of the continuous function is a
However,
(why?).
closedand bounded
then
we do have boundedness:
interval,
Not

function

all continuous
:= \317\207
on
f(x)

the

9.

270

< b be real numbers,and


on [a, b]. Then f is a
continuous

a function

be

for every
that

such

: [a,

bounded

b]

\320\246,

-\302\273
\321\206

function.
bounded.
\317\207
e

[\320\260,
\320\251

\\f(x)\\>M.

In particular,

|f(x) | > n) is

for every natural number


We

non-empty.

in

let

for sake of contradiction that / is not


an element
exists
real number \316\234there

Suppose

Proof
Thus

on

functions

Let a

9.6.3.

Lemma

Continuous

thus

can

set {x e [a,
b]:
a sequence (rrn)\302\243\302\260
This sequence lies in [a, 6]?
n,

the

choose2

> \316\267
for
all n.
that |/(\320\266\320\277)|
so by Theorem
9.1.24 there exists a subsequence
(xn^Jtn
to some limit L \342\202\254
< ...
no < n\\ < \316\267<\316\271
converges
[a, b], where

[a, b] such

and
which

we
increasing sequence of natural numbers. In particular,
>
N
use
for
all
\342\202\254
j
j
induction).
(why?
Uj
at L, and in
it is continuous
Since / is continuouson [a^b],

is an
see

that

see that

we

particular

lim f(xnj)

= f(L).

J\342\200\224\320\256\320\236

the

Thus
On

the

sequence
{f{xnj))JLo
other hand, we know

&>\316\223
all

!> j
a contradiction.

n>j

Remark

jj and

hence the

IS convergent,
from

the

and hence it is bounded.


that

construction

sequence {f{xnj))jio

9.6.4. There are two

IS

to

bounded,

proof that are


the Heine-Borel
useful
theorem
an indirect proof; it doesn't say

things

about

this

worth noting. Firstly,it shows how


it is
(Theorem 9.1.24) is. Secondly,
how to find the bound for /, but it shows that having /
leads

>

\\f{xnj)\\

not

unbounded

a contradiction.

We now

improve Lemma9.6.3to say

something

more.

9.6.5
R be a
(Maxima and minima). Let / : X \342\200\224\342\226\272
and
let
its
maximumat
xo G X. We say that / attains
function,
if we have f(xo) > f{x) for
X (i.e., the value of / at
all
\317\207
\342\202\254
xo

Definition

this requires the axiom of choice,as in Lemma


speaking,
one can also proceedwithout
of choice, by defining
the axiom
> n},
of / to show
and using the continuity
G [a,b] : \\f(x)\\
sup{a;
>
\320\277.
We
leave
the
the
details
to
reader.
|/(\321\217\320\277)|
2Strictly

However,

8.4.5.
xn

:=
that

271

gt6. The maximum principle

point in X).
<

f(xo)

then

We

say

/(*)\302\267

If a

9.6.6.

Remark

be bounded

must

it

than or equal to the value


of / at any other
that
/ attains its minimum at xo if we have

is larger

xo

point

the

function attains its maximumsomewhere,


from above (why?). Similarly if it attains

be bounded
are
local

somewhere, then it must


of maxima and minima
be defined in Definition 10.2.1.

its minimum
notions
These
will

Proposition 9.6.7
numbers,

: [a,b]

9.6.8.

principle

a moreprecise

on [a,b].

be a function
continuous
at some point xmax
G [a,b],

either a

to denote

used

is

\"extremum\"

\"extremum

been

have

would

name

word

be real
also

and

speaking, \"maximum principle\"


also concerns the minimum.

Strictly

the

since

misnomer,

a <b

minimum at somepoint \321\205\321\210%\320\277


\342\202\254
[\316\261,
b].

attains its
Remark

Let

\342\200\224>
R

its maximum

attains

Then

let

and

versions

global]

principle).

(Maximum

below.

from

is a

Perhaps

principle\"; the

maximum or a

minimum.

shall

We

Proof.

the

that

proof

somewhere;
just show that / attains its maximum
it attains its minimum also is similarbut
is left

to

the reader.

Prom Lemma 9.6.3 we

exists
let

f(x)

is bounded,

<

\316\234for

each

thus there
Now

\317\207
\342\202\254
[\316\261,
b].

set

the

\316\225
denote

that

know

\342\200\224
\316\234<

that

\316\234such

an

E:={f(x):xG[a,b]}.
:=
(In other words, \316\225
a subsetof [\342\200\224M,M].
the

instance

it has

a supremum

that

f(x)
maximum

<

\321\203

< m

for

sup(\302\243?)

all

\317\207
G

that f(xmax)

= m.

[a,b].
it

(Why

will
will

definition

Thus

principle,

real number.

is a

of supremum,
we
of \302\243?,
this
means

to show

suffice to find
this

said, this set is


since it contains for

the least upperbound

definition

\342\202\254
\316\225]by
\321\203

somewhere,

non-empty,

which
By

s\\xp(E).

all

for

also

is

Hence by

f(a).

point

Write m :=

It

we just

what

/([\316\261, \320\254]).) By

suffice?)

an

know
that

that / attains its


xmax

\302\243
[a,

b] such

Continuous functions on \320\246

9.

272

1 be

>
\316\267

Let

least upper

is the

sup(i?)

any integer.

Then m

bound

for

\342\200\224
i

< m

sup(\302\243\.")

be

m\342\200\224 cannot
\302\243

E,

\320\224\320\264

an upper

i
m \342\200\224
that
such
bound for E, thus there exists\320\260
G \316\225
\321\203
<y.
By
there exists an \320\266
that
this
\342\202\254
definition of \302\2437,
implies
[\316\261,
such
\320\254]

that

this

<

a sequence

of [a,b] suchthat
the axiom of choice; however

requires

for

choosing,

(xn)\342\204\242=1 by

an element

be

to

/(\320\266).

choose

now

We
xn

\342\200\224
i

\342\200\224
<
\302\243

is possible

it

each

\316\267

(Again

/(rrn)\302\267

to prove

this principle without the axiomofchoice.For instance,


you will
the
notion
of
see a better proof of this proposition
using
This is a sequence in [a,b]] by the
compactnessin
Proposition
13.3.2.)
Heine-Borel
theorem
9.1.24), we can thus find a
(Theorem
which converges to some
where
<
< ...,
n\\
\316\267<\316\271
subsequence
{xn^JLv

limit

Since

\342\202\254
[a,b].

Xmax

at

continuous

to

converges
(xn^JLi
as before that

/ is

and
\321\205\321\210\320\260\321\205,

we
have
\321\205\321\210\320\260\321\205,

lim f{xnj)
J

f{Xmax)\302\267

J\342\200\224>oo

On

the

other

construction we know

hand, by

>m

f{xnj)

and so by

of both

limits

taking

Uj

f{xmax)=

lim

>m-T,

sides we
>

f(xni)

lim

that

see that
- = m.

m-

j-\321\216\320\276

j->oo

hand, we know that f(x)<m for


in particular
f(xmax) < m. Combiningthesetwo
see that f(xmax) = \317\204\316\267
as desired.
On

the

other

Note
from
For

its

that

the

maximum

Let us
and
[\316\261,
\320\254]},

so

\317\207
\342\202\254
[a,b],

we

inequalities

\316\240

principle does not prevent a function


or minimum at more than onepoint.
attains
f(x) := x2 on the interval
[\342\200\2242,2]

its maximum

attaining
instance,

maximum

all

the

function

at two

write
similarly

different points, at

supx6[a&]

define

f(x)

as

\342\200\2242
and

short-hand

at

2.

for sup{/(rr)

The
maximum
f(x)\302\267
\321\210\302\243\321\205\320\265[\320\260\320\264

\342\202\254
\317\207

principle

The intermediate

p#7.

in

is

[a, b], i.e., thereis


=

for which f(xmax)


[a? b]
or equal to m.
less than

value of

is a

f(x)

supx6[abi

/ on

of

value

maximum

:=

that m

asserts

thus

273

theorem

value

for

and

wi,

real number
at

every

Similarly infxG[ab]

one

least

other

and is the
point

\317\207
\342\202\254
[\316\261,
b],

/(\320\266)

is

the

xmax
/(\320\266)

minimum

/ on [a,b].

closed interval, every


continuous
its maximum at least once and
and attains
function is bounded
The
same
is not true for open or infinite
at
least
once.
minimum
intervals; see Exercise 9.6.1.
We

on a

that

know

now

Remark 9.6.9.

You

a rather

encounter

may

different \"maximum

or
differential
partial
equations,
principle\" in complex analysis
respectively,
involving analytic functions and harmonicfunctions
functions. Those maximum principlesare
of continuous
instead
to this one (though they are also concerned
related
not
directly
whether
maxima
with
exist, and where the maximaare located).

Give

9.6.1.

Exercise

(a) a

of

R which
is
/ : (1,2) \342\200\224\342\226\272
minimum
but
somewhere,

function

its

attains

examples

continuous and bounded,


does not attain its maximum

anywhere;
R which
/ : [0,oo) \342\200\224\302\273

a function

(b)

its maximum

somewhere,

is continuous,
does not attain

but

bounded, attains

its

minimum

anywhere;

/ :

a function

(c)

its minimum
(d) a function
lower

1,1]
[\342\200\224

anywhere

is bounded but does not


maximum anywhere.

\342\200\224\302\273
R which

or its

\342\200\224\302\273
R which
[\342\200\2241,1]

why none of the examplesyou


principle. (Note: read the assumptions

We

intermediate

The
have

maximum

no upper

bound and

no

bound.

Explain

9.7

has

attain

just
value

shown
and

that

value

construct

violate

theorem

a continuous function
value. We now

its minimum

the maximum

carefully])

attains both its


show that / also

Continuous

9.

274

attains

intuitive

To do this, we

in between.

value

every

on

functions

first

\320\246

a very

prove

theorem:

Let a < b, and


value
Theorem 9.7.1 (Intermediate
theorem).
R be a continuous function on [a,b]. Let \321\203
let f : [a,b] \342\200\224>
be a
<
<
real number between /(a) and /(b), i.e., either
f (a)
\321\203 /(b)
or

f{p) >

\320\243 /(b)\302\267

Proof.

We

>

exists

there

Then

\321\201
\342\202\254
[a, b]

such

/(\321\201)

\321\203.

>
f{a)<y<
f(b) or /(a) > \321\203
f(b).
< /(b);
the latter is
the former, that f(a) < \321\203

two cases:

have

We will assume

proven similarly and is left to the

reader.

then the claim is


\320\243 /(b)
=
=
set \321\201 a or \321\201 b, so we will assume that

\342\200\224
\320\232
\320\243 f{\302\260)

simply
Let \302\2437
denote

that

the

or

easy,

as

one

can

< /(b).
/(a) < \321\203

set

E:={x\342\202\254[a,b]:f(x)<y}.
a subset
of [a,b], and is hence bounded.Also,
that
a
we
see
is an elementof\302\2437,so \316\225is non-empty.
< y,
f(a)
the least upper bound principle,the supremum

Clearly

\316\225
is

:=
\321\201

is thus
\316\225
contains

complete

finite. Since \316\225


is
a,

we know

the proof

By

sup(E')
since
by b, we know that
we have \321\201
Thus
\342\202\254
To
[\316\261,
b].

c<b;

bounded

that

since

>
\321\201

a.

we now show that

/(\321\201)

The
\321\203.

idea

is to

< \321\203,
to work
work from the left of \321\201
to
show
and
that
from
/(\321\201)
the right of \321\201
to show
that /(\321\201) > \321\203.
\342\200\224
>
Let
\316\267
1 be an integer.
The number \321\201
is less than
^
=
E.
Thus
\321\201
and
hence
cannot
bound
for
be
an
upper
sup(\302\2437)
there exists a point, callit xn, which
in \316\225
and
which
lies
is greater
\342\200\224
i.
than \321\201
\321\201
Also xn < \321\201
since
is an upper
bound for E. Thus

1
\321\201

< xn

<

\321\201

\316\267

=
xn
(Corollary 6.4.14) we thus have limn_>oo
=
\321\201
Since
\320\230\321\210\320\277-\321\216\320\276
/ is continuous at c, this impliesthat
f(xn)
f\302\260r
But since xn lies in \316\225
for
n, we have f{xn) < \320\243
every
/(c).

By

the

squeeze

test

g 7.

every

The

intermediate

n.

By

an N > 0 such

there is
A

c+

to

converges

\321\201
+

and
A

c+

converges

already

> /(c),
that

>

\302\243

\321\201
+

\321\201
is
\302\243
for

b.

In particular

\321\201
<b.

all

b for

\302\243

Since

oo).

have

c, we thus

\321\201
^

have
<

\321\201
+

\342\200\224>
\316\267

\321\201
as

conclude

we

must

we

\321\201
and
\342\202\254
\321\201\321\204\320\254
[\316\261,
b],

Since

(Lemma 6.4.13) we thus

principle

comparison

Since /(b)

< y.

/(c)

have

the

275

value theorem

the

(since

of

supremum

all

> iV.
have /(c + A) > \321\203
for
all \316\267
to c, and /is
continuous at c, thus /(\321\201) >
= y, as desired.
knew that /(c) < y, thus
/(c)

\302\243\"

N.

Since

But

\321\201
+

\316\267
>

we thus

[a, b],

\316\267
>

\320\220

we

But
\321\203.

The intermediate

that
if / takes the
value
theorem
says
it
must
take
all
the
then
also
values
in
values /(a)
/(b),
if / is not assumed to be continuous,
then
Note
that
between.
no
value
theorem
For
intermediate
instance,
longer applies.
\342\200\224>
R is the function
:
[-1,1]
/
and

-,

ifz>0

/(*):=\\l

then
which
past

and
\342\200\2241,

/(\342\200\2241)

= 0.

/(c)

Thus

intermediate

/(1)

1, but

however

values;

there is no \321\201
for
\342\202\254
[\342\200\2241,1J

is discontinuous,

a function

if

if

if:z<0

-1
\320\223

\321\207

the

it can

\"jump\"

functions cannot do

continuous

so.

Remark 9.7.2.
value

fact,

/(-l)
Remark

this

instance,

:= x3 \342\200\224
then
x,

function f(x)
we know that
in

For

times.

multiple,

function

continuous

/(\342\200\2242)

there exists

\320\260
\321\201
for
\342\202\254
[\342\200\2242,2]

case

= /(0)
9.7.3.

exists three

there

= /(l) =
The

take an

may

if /

intermediate

\342\200\224>
R
[\342\200\2242,2]

\342\200\2246
and

which

/(2)

/(c)

such values of c:

is the

6, so
= 0. In
we

have

0.

intermediate

value theorem

gives another

of a number.
For instance,
to show
that one can take nth roots
to construct the squarerootof2,consider
the
function
/ : [0, 2] \342\200\224>
=
R defined
x2. This function is continuous,with /(0) =
by f(x)
0 and /(2) = 4. Thus
there
exists
\320\260
such
that
\321\201
\342\202\254
[0,2]
/(c) = 2,
=
2.
not
there
c2
does
show
that
is
i.e.,
just one
(Thisargument
squarerootof 2,but it does prove that there is at least one square
way

rootof 2.)

Continuous

9.

276

functions

on fi

9.7.4 (Images of continuousfunctions).


Let
a < 5
: [a,b]\342\200\224\342\226\272
R
be a continuous function on [a,b].
\302\243e^
\316\234:=
maximum
value of f, and let m :==:
be the
supxG[ab] /(\320\266)
be a real
be the minimum value. Let \321\203
number
f(x)
infx6[tt)b]
< M). Then there exists\320\260
m and \316\234(i.e.,
m <y
between
\321\201
6
[a,6]
=
such that /(c) = y. Furthermore,we have
/([a,b])
[m,M].
Corollary
and let f

Exercise 9.7.1.

Proof. See

to the

addition

in

Corollary 9.7.4. (Hint: you may


intermediate value theorem.)

Prove

9.7.1.

Exercise

9.4.6

Exercise9.7.2.Let /

\342\226\241

need

Exercise

: [0,1]

\342\200\224\342\226\272
be a continuous
function. Show
[0,1]
in [0,1]
exists a real number \317\207
such that f(x) = x. (Hint:
\342\200\224
the intermediate
value theorem to the function
This
apply
f(x)
x.)
is known
as a fixed point of /, and this result is a basicexample
point \317\207
role in certain types
of a fixed point theorem, which play an
important

there

that

of analysis.

Monotonic functions

9.8
We

a class

discuss

now

of functions

which is distinctfrom

class

the

the
functions, but has somewhatsimilarproperties:
monotone
(or monotonic) functions.

of continuous
of

class

Definition 9.8.1
R,

and

/ :

let

increasing

be a

We say

function.

\342\200\224\342\226\272
R

iff f(y)

> f(x)

Let

functions).

(Monotonic

whenever x,y

be a

subset

of

that / is monotone

X
and
\342\202\254

We
> \317\207.
\321\203

say

whenever
> f(x)
strictly monotone increasingiff f(y)
iff
X
and
x.
we
is
monotone
\342\202\254
>
\321\217,
decreasing
\321\203
\321\203
Similarly,
say /
/ is

that

<

f(y)

decreasing

iff

f(y)

decreasing,
strictly

and

strictly

monotone

G X
\320\266,
\321\203

<

/ is monotone

that

or

whenever

/(^)

f(x)
if

it

and

whenever

and
> \317\207,
\321\203

x,y

is monotone

monotone

if it is

X
and
\342\202\254

increasing

monotone

We
> \317\207.
\321\203

say

or monotone

strictly monotone increasing

decreasing.

9.8.2. The function f{x) := \320\2662,


when
domain [0, 00), is strictly monotone
increasing
Examples

strictly

to the

restricted
(why?),

but

when

g 8.

restricted
(-oo,0]

and

line

monotoneon a
0n

increasing
(unless

for

are not

/:

decreasing,

necessarily continuous;

for

earlier

defined

\342\200\224*
R
[\342\200\2241,1]

functions

monotone

function

the

consider

instance,

by
if:z<0

-1

\342\200\236
,

strictly monotone

monotone (consider

not necessarily
=
x2 on R), and
f{x)

function

the

is not

are

functions

Continuous
instance

R,

monotone

both

is

\320\241

but

most one point - why?).

of at

consists

it is automatically
monotone as well
The constant function f{x) :\342\200\224
6, when

monotone

and

monotone

strictly

X,

X.

arbitrary domain X

to an

restricted

(\342\200\22400,00).

domain

domain

same

the

is

is strictly monotone on both


strictly monotone (or monotone)
Note
that
if a function is strictly

is not

but

[0,oo),

real

full

the

(\342\200\22400,0],

function

the

Thus

(why?).

decreasing

domain

the

to

instead

0n

277

Monotonia functions

ifz>0.

/(*):=\\l

Monotonefunctionsobeythe maximum
principle
(Exercise
9.8.1),
but not the intermediate value principle (Exercise9.8.2).On the
it is possible for a monotonefunction
to
have
other
hand,
many,
discontinuities
many
(Exercise 9.8.5).
is both
If a function
monotone
and continuous, then it
strictly
it
nice
In
is invertible:
has many
properties.
particular,
Proposition 9.8.3. Let a <b
R

be

real

be

which is both continuous


Then f is a bisectionfrom [a,b]

a function

increasing.

inverse /_1 :

[/(a), /(b)]

is also

\342\200\224\342\226\272

[a,

b]

and
to

: [a, b]

let f

and

numbers,

monotone

strictly
/(b)],

[/(a),

continuous

\342\200\224>

and the

and strictly

monotoneincreasing.

9.8.4.

Exercise

See

Proof.

There is a similarProposition
for
functions
monotone decreasing; see Exercise9.8.4.

Example9.8.4.

Let

function

see

from

\316\267
be

a positive

integer

which

and R

are

strictly

> 0. Sincethe

we
is strictly increasing on the interval
[0, \320\224],
a
from
that
is
this
function
9.8.3
Proposition
bijection

f(x)

:= xn

Continuous

9.

278

hence
and
[0, R] to [0, \320\224\320\277],
This can be usedto give

number

of a

xl/n

root

is an

there
an

inverse
means

alternate

what

than

\317\207
\342\202\254
[0, R]

from

on R

functions

[0,

Rn]

to [0, R]

to construct

the

was done in

Lemma

5.6.5.

Exercise
9.8.1.

the maximum
why
principle remains true if
/ being monotone,
/ is continuous is replacedwith
strictly monotone. (You can usethe same explanation
Explain

that

hypothesis

being

with

both

cases.)

nth

the

or
for

the intermediate
Exercise 9.8.2. Give an exampleto show
that
value
is replaced
theorem becomes false if the hypothesis that / is continuous
with / being monotone, or with
/ being strictly monotone. (You can

usethe same counterexample

for

both

cases.)

b be real numbers,
and let / : [a,6] \342\200\224\302\273
R be \320\260
continuous and one-to-one.Show that / is strictly
monotone. (Hint: divide into the three cases f(a) < /(b), f(a) = /(b),
The second case leads directly
to a contradiction.
In the
> /(&).
f(a)
use contradiction
and the intermediate value
theorem
first
to show
case,
in the third case, argue similarly
that / is strictly monotone
increasing;
to show
/ is strictly monotone
decreasing.)

9.8.3. Let

Exercise

\316\261
<

is both

which

function

Prove Proposition 9.8.3. (Hint: to show


that
is
/_1
of
it is easiest to use the \"epsilon-delta\"
definition
Is the proposition still true if the
Proposition
continuity,
9.4.7(c).)
is dropped,
or if strict monotonicity is replacedjust
continuity assumption
to deal with
How should one modify
the
by monotonicity?
proposition
monotone
functions instead of strictly monotone
strictly
decreasing
9.8.4.

Exercise

continuous,

functions?

increasing

In this exercise we give an example


of a function which
at every rational point, but is continuous
at every
as
Since the rationals are countable, we can write
them

9.8.5.

Exercise
has

a discontinuity

irrational.
Q

= {<7(0),g(l),</(2),...},

to Q.
for
Since

Now

each

natural

number

2~\316\267
is
\316\243\342\204\242=0

absolutely

where q : N \342\200\224>
from N
Q is a bijection
\342\200\224\302\273
:
R
by setting
g Q
g(q(n)) := 2~n
to
n; thus
1, q(l) to 2_1, etc.
g maps
q(0)

a function

define

absolutely

/(*):=

we see that

convergent,

convergent. Now definethe

function

\302\243

g(r).

: R

is also

g(r)
\316\243\316\223\342\202\254\316\267
\342\200\224\302\273
R

by

Since ]CreQ#(r)
for

defined

every

(a) Show that

*s

strictly monotone
5.4.14.)

for

every

since r

is

increasing. (Hint: you

number

rational
r

rational,

q{n)

f{x) >

f(r) + 2\"n for all


Show that for every irrational number
that

Show

(c)

(Hint:

that

demonstrate

first

f(x) is

well-

x.

/ is

(b) Show that

we know that

convergent,

absolutely

real number

Proposition

(Hint:

279

continuity

Uniform

p. P.

r, /

will

is discontinuous
natural

for some

need

at

r.
n.

number

\317\207
> r.)

is continuous

\320\266,
/

at x.

the functions

:=
/\342\200\236(*)

g{r)

\302\243

reQ:r<x,3(r)>2-n

at

are continuous

Uniform

9.9

and
\320\266,

that

\342\200\224

\\f{x)

/n(s)|

^ 2~n.)

continuity

function on a closed interval


[a,b]
attains its maximumand minimum,
the
maximum
However, if we replace the closed
by
principle).
an open
interval
interval, then continuous functions need not
by
\342\200\224>
An example is the function
more.
R
be bounded
/ : (0,2)
any
defined
point
by f(x) := 1/x. This function is continuousat every
and is hence continuous at (0,2), but
is not
bounded.
in (0,2),
the
function
Informally
speaking, the problem here is that while
is indeed continuous at every point in the openinterval
it
(0,2),
\"less and less\" continuous as one approaches
the
becomes
endpoint
We

a continuous

that

know

(and in fact

bounded

remains

0.
the
us analyze this phenomenon further, using
\"epsilonof continuity - Proposition9.4.7(c).We know
definition

Let
delta\"
that

:X

if /

exists

there

is

\317\207
X
\342\202\254
\316\265-close
way

of

to

is continuous

\342\200\224>
R

\316\264
such

\316\264-close

f(xo)

thinking

an \"island of
doesn't stray

to

if we
about

that

f(x)
In other

at a

then
for
point \320\266\320\276>

will be

\316\265-close

to

> 0
every \316\265
whenever
f(xo)

words, we can force f(x) to be


ensure that \317\207
is sufficiently
close to xq. One
this is that around every
is
xo there
point

xo.

xo
\316\264,
stability\" (xo \342\200\224
than
\316\265
from
more
by

where

\316\264),

f{xo)\302\267

the

function

f(x)

Example 9.9.1. Takethe function


to
at the point xo = 1. In order
it

to take

suffices

/(\320\266\320\276)>

close
and

to
so

\317\207
to

be

functions

on R

l/fc mentioned

above

Continuous

9.

280

f{x)

:=

is 0.1-close to

that

ensure

f(x)
to so, since

1/11-close

and so 11/12 <


< 12/11,
10/11 < \317\207
is 0.1-close to f(xo)- Thus the \"5\" one

xo then
f(x)

to f(xo) is about 1/11or so,at

f(x) 0.1-close

the

if

\320\266
is

f(x) <
needs
point

1/\320\246-

11/\320\256,

to make
xq =

1.

Now let us lookinstead


at
the
xo = 0.1. The function
point
=
we
is
still
shall
see the continuity
continuous
but
here,
f(x)
1/x
that
is 0.1-close to /(xq),
is much worse.In orderto ensure
f(x)
to be 1/1010-close
to xq. Indeed, if \320\266
is 1/1010
we need \317\207
close to
and
so
then
\317\207
9.901
<
<
< 102/1010,
< f{x) 10.1,so
\320\2660,
10/101
a
is
one
needs
much
smaller \"5\" for
to
Thus
0.1-close
f{xo)\302\267
f(x)
is
the same
value of \316\265
much
more
\"unstable\"
near 0.1
i.e.,
f(x)
smaller
than
it is near
\"island
1, in the sense that there is a much
1 (if one
is interested
of stability\"
around 0.1 as there is around
in keeping f(x) 0.1-stable).

On the other

are

there

hand,

functions which

continuous

other

Consider
the function
do not exhibitthis behavior.
R
g : (0,2) \342\200\224>
= 0.1 as before, and
:=
2x.
fix
Let
us
defined
\316\265
by g(x)
investigate
is 0.05the islandof stability
around
xo = 1. It is clear that if \317\207
in
this
case we can
close to xo, then g(x) is 0.1-closeto g(xo)\\
take \316\264
to be 0.05
if we move
at xo = 1. And
xo around, say if we
does
even when xo is set
set xo to 0.1 instead, the \316\264
not change
to 0.1instead
will
of 1, we see that
stay 0.1-closeto g{xo)
g(x)
whenever
\317\207
is 0.05-close
works
for every
to xo- Indeed, the same \316\264
the
is
function
xo- When this happens, we say that
g
uniformly
continuous. More precisely:

Definition 9.9.2
and let / :
R
continuous

if,

for every

and f(xo) are

a function.

be

\316\265
>

whenever

\316\265-close

0, there

X be

Let

continuity).

(Uniform

\342\200\224>

We say that
exists a \316\264
> 0

xo
\320\266,

are

X
\342\202\254

which are 5-close.

Remark 9.9.3.
notion
function

of

continuity.
/

This

definition

Prom

is continuous

should

Proposition
if for every

9.4.7(c),
0,

and

/ is

uniformly

such

that

two points

be compared
\316\265
>

a subset of

R,

f(x)
in X

with the

we know that a
every

xq

\342\202\254
X,'

281

continuity

\320\264.Uniform
\320\264.

and f(xo) are

that f(x)
>
0 such
there is a \316\264
x e X is 5-close to xo. The difference

is that in uniform continuity

and continuity
fj

for all

works

which

might

xo

a different

use

\342\202\254
-X\"; for

\316\264.
Thus

\316\265-close

uniform

between

can

one

whenever

continuity
take a single

each xo G X
continuity,
continuous
function
uniformly

ordinary

every

is continuous, but not conversely.


9.9.4.
The function / : (0,2) \342\200\224>
R defined
(Informal)
:=
is
continuous
on
not
but
uniformly
1/x
(0,2),
by f(x)
the continuity
continuous,because
(or more precisely, the dependenceof
Example

worse

and

worse

becomes
\316\265)

on
\316\264

as

\342\200\224>
\317\207
0.

make

will

(We

this

more precise in Example9.9.10.)


point and of continuous
both had \"epsilonformulations;
equivalent
the notion of \316\265-closeness),
formulations
and
(involving
of
\"sequential\" formulations (involving the convergence
see Lemma
9.1.14 and Proposition 9.3.9. The concept
can similarly
be phrased in a sequential
continuity
this
time
the
using
concept of equivalent sequences (cf.
we now generalize to sequences of real
but
5.2.6,
and no longer require the sequences
of rationale,
to

had

function
delta\"

type

had

both

sequences);
of

uniform

formulation,
Definition

instead

numbers
be

9.9.5

(an)\302\243Lm

\316\265
> 0

and

sequences).
(Equivalent
be
two
sequences
(bn)^=Tn

be given. We say that


to bn for each n>m.

close to

(bn)^=m iff
and

(unj^N

there

(bn)\342\204\242=N

(bn)^=m are equivalent


(bn)^=m

Remark

to be

are
9.9.6.

(an)^>=m

is

Let m be an integer,
of real numbers,and let

\316\265-close

are
iff

\316\265-close.

each

for

Two
\316\265
>

iff

(bn)\342\204\242=m

sequences
0, the

eventually

\316\265-close.

One could

debate whether

Tational or real,

6.1.4 shows that this


definitions.

to

an

We say that (an)\302\243Lm is eventually


exists
an N > m such that the sequences

\316\265-close

and

of adherent

notions

several

Cauchy):

Definition
let

the

that

Recall

but

minor

doesnot

make

sequences

\316\265
should

modification
any

(an)%Lm

difference

be

is
e-

and

(an)\302\243Lm

assumed

of Proposition
to the above

9.

282
The notion
our

language

necessarily

(not

are

{bn)%Li

the

Meanwhile,

\320\264\320\277\320\276\320\263\320\265
succinctly

using

be

(bn)\302\243Li

of real

sequences

Then (an)^=l

or convergent).

=
only i/limn_,oo(an \342\200\224
bn)

and

0.

9.9.1.

\342\226\241

continuity can be phrased

of uniform

notion

sequences:

equivalent

using

if and

equivalent

See Exercise

Proof

9.9.8. Let X be a subset


Then the following two

Proposition
a function.

be

bounded

on R

functions

of equivalencecan bephrased
of limits:

Lemma 9.9.7. Let (an)\342\204\242=1and


numbers

Continuous

and let

o/R,

f :X
are

statements

\342\200\224>
R

logically

equivalent:

(a) f

is

uniformly

Whenever

(b)

consisting
(f(yn))^Lo

Proof See

Exercise

9.9.9.

Remark
9.3.9.

Proposition

continuous

(xn)\342\204\242=0

and

on

X.
two

are

(yn)\302\243L0

of elements of X, the
are also equivalent.

The reader should comparethis with


Proposition
9.3.9 asserted that if / was continuous,
then
asserts

9.9.8

Proposition

sequences

(/(zn))\302\243Lo an^

9.9.2.

to convergent
/ maps convergentsequences
contrast,

equivalent

sequences

that

if /

sequences.

is

uniformly

In
continuous,

pairs
pairs of sequencesto equivalent
are
the two Propositions
connected,
Lemma
if
9.9.7 that {xn)%Lo wl^ converge to \320\266*
are
sequences {xn)%Loan(i (x*)%Lo
equivalent.

/ maps equivalent
To see how
sequences.
then

observe

from

only

if the

Example

9.9.10.

Consider the function

of

and

R defined
/ : (0,2) \342\200\224>

Lemma
we see that
9.9.7
by f(x) := 1/x consideredearlier.Prom
the sequence (l/n)\342\204\242=l and (1/271)^
are equivalent
sequences in
are
and
(0,2). However,the sequences
(f(l/n))\342\204\242=1
(/(1/271))^
not equivalent (why? Use Lemma9.9.7again).
So by Proposition
is
start at
not
continuous.
9.9.8, /
sequences
uniformly
(These
1 instead of 0, but the reader
can easily see that this makesno
difference

to

the

above

discussion.)

283

g.9. Uniform continuity

Example 9.9.11. Consider

function

the

/(#) :=

quantify

and

(n+

But

the

equivalent,

sequences

+ A) =
2-close

f(n

Another

and

(/(n

n2 + 2 +

=
\320\224^

equivalent.

not

^))\302\243Li

f(n)

+ 2

are

does

+ ^

By Proposition 9.9.8 we

to /(n).

not

sequences are

of uniformly continuous functions


to Cauchy sequences.
sequences

Cauchy

Proposition 9.9.12. LetX

be

subset

o/R,

and let

function. Let (xn)^Lobe


entirely of elements in X. Then

can

consisting
sequence
is also a Cauchy sequence.

is that

f :X

continuous

be a uniformly

not

continuous.

uniformly

property

map

they

/ is

that

conclude

these

9.9.7,

(f(n))^L1

eventually

become

thus

Lemma

By
^)\302\243Li\302\267

since

out

continuous;
gets
approaches infinity. One way to
9.9.8. Consider the sequences(n)\342\204\242=1
Proposition

is via

this

by

turns

as one

worse\"

and

\302\253worse

defined

\342\200\224\342\226\272
R

function on R, but it
in some sense the continuity

continuous

be uniformly

to

not

is a

\317\2072.This

: R

\342\200\224\302\273
R

Cauchy

(/(#n))\302\243L0

Proof. See Exercise9.9.3.

we demonstrate that the function


:=
continuous.
by f(x)
1/x is not uniformly
is a Cauchy
in (0,2), but the
The
sequence
sequence
(l/n)\342\204\242=l
a
is
not
sequence
Cauchy
sequence(f(l/n))\342\204\242=1
(why?). Thus by
is
not
continuous.
9.9.12,
Proposition
/
uniformly
Once

9.9.13.

Example

again,

defined

\342\200\224\342\226\272
R

/ : (0,2)

be a subset o/R, let f : X \342\200\224\342\226\272


R
be a
be
an
adherent
continuous
and
let
xq
function,
point of
uniformly
X. Then the limit \320\230\321\210^^^\320\265\321\205
exists
it is a
(in particular,
f(x)
real

Let X

9.9.14.

Corollary

number).

Proof.
We

now

show

that

bounded sets to bounded


Proposition
be a

uniformly
subset of X.

9.9.4.

See Exercise

a uniformly

continuous function will map

sets.

be a subset o/R, and


continuous function. Suppose that

9.9.15.

Let X

Then f(E) is alsobounded.

let
\316\225
is

: X

\342\200\224>
R

a bounded

Continuous

9.

284

on R

functions

Proof. See Exercise9.9.5.


are
is

have

we

As

uniformly
a closed

seen repeatedly, not all\302\267continuous


functions
continuous.
However, if the domain of the function
just

then continuous

interval,

functions are in

fact

uniformly

continuous:

9.9.16.
Theorem
R
be a function

and let f : [a,b] \342\200\224>


Let a < b be real numbers,
which is continuouson [a,b]. Then
is also
f

continuous.

uniformly

Proof. Suppose for


continuous.

sequences

9.9.8, there must

we

can

an

find

not eventually

are

this

Fix

value

:=
\316\225

have

must

We

and

(x^^Lq

sequences (f(xn))^>=0and
particular,

\316\265
>

(f(yn))%Lo

that

0 such

we

\316\225
is

that

see

*n

(yn)\302\243Lo

not

are

uniformly

therefore exist two


suc'1 ^bat
In
equivalent.

the

ta> 4

(f(xn))\342\204\242=0

and

(/(l/n))\302\243i0

\316\265-close.

of

G
{\316\267

let

and
\316\265,

\316\225
be

since

\316\225
infinite,

if

\316\225
were

set

are not
then

finite

\316\265-close

(why?).

\316\265-close}.

and

(f(xn))%Lo
By

Proposition

proof of that proposition


an infinite sequence

in fact

countable;

the

and f(yn)

: f(xn)

(f(yn))^=o would be eventually


8.1.5,

/ is not

that

contradiction

Proposition

By

equivalent

of

sake

we can find

from the

< ...
< \316\267>2
no < \317\204\316\220\316\271

consisting

entirely

of elements in

\\f(xnj)

the other

On

and so by
be

E. In particular,we

f(ynj)\\ >

for
\316\265

all

hand, the sequence(xn^JLo

IS

N.
\342\202\254

a sequence

(9.3)

in [a, 6],

(Theorem. 9.1.24) there must


which
converges to somelimit L
(xnj )j^0
In particular,
is
continuousat
/
L, and so by Proposition
the

Heine-Borel

theorem

a subsequence

[a,b].

have

in

9.4.7,

limf(xnJk)Jk

k\342\200\224>oo

= f(L).

(9.4)

jjote that

of (j/n^Q,

lim (xn

yn)
\320\277-\321\216\320\276

By

Sincexni,

is

(2/nifc)fci0

the other

On

hand,

to

converges

L as

0.

have

we thus

6.6.5,

Proposition

and

Ozn)\302\243Lo>

6.6.4.

Lemma

by

we have

9.9.7

Lemma

of

a subsequence

is
0&nifc)jfeLo

a subsequence
from

285

continuity

Uniform

gg.

\342\200\224>
\320\272
oo,

lim

laws

have by limit

thus

we

2/n.
\342\200\242/fc
\320\272\342\200\224\321\216\320\276

and

hence

by continuity

of / at

= f(L).

]imf(ynJ
from

this

Subtracting

we obtain

limit laws,

using

(9.4)

l^(f(xnjk)-f(ynjk))=0.
But

contradicts

this

that

conclude

One

Lemma

Exercise

9.9.1. Prove

Exercise

9.9.2.

9.9.7,

and

but

third,

Prove

instead

they are

9.6.3, Proposition
No two of these

other.

with

the

we

continuous.

should
Remark 9.9.17.
compare
and
Theorem
9.9.16
each
9.9.15,

resultsimply

contradiction

Prom this

(why?).

(9.3)

/ is in fact uniformly

all consistent with

other.

each

Lemma 9.9.7.
Proposition

go back to the

9.9.8. (Hint: you


of

definition

should avoid

equivalent

Lemma

sequences

in

Definition 9.9.5.)

Exercise 9.9.3. Prove

Proposition

9.9.12.

(Hint:

use Definition

9.9.2

directly.)

Exercise

9.9.4.

this corollary
9.9.10.
Example

Use Proposition

to give

an

alternate

9.9.12 to

prove

demonstration

Corollary

of the

9.9.14.

results in

Use

Continuous

9.

286

Prove

9.9.5.

Exercise

9.6.3.

Lemma

proof

point

of

9.9.12or

you will need

9.9.14.)

Corollary

be
subsets
9.9.6. Let \316\247,\316\245,\316\226
continuous
which is uniformly
continuous
which is uniformly

Exercise
function
function

\316\226
: X \342\200\224\342\226\272
is uniformly

go f

9.9.15. (Hint: mimic


the
either Proposition

Proposition

some

At

on \320\246

functions

Let / : X \342\200\224\342\226\272
\316\245
be a
\342\200\224\342\226\272
and let g : \316\245
\316\226
be a

of R.
on

X,

on

\320\243.Show

the function

that

on X.

continuous

Limits at infinity

9.10

discussed, what it means for a function


/ \302\267.
\342\200\224\342\226\272
as
as
is
a
a limit as \317\207
real
number.
xo
xo
long
now
discuss what it would mean to take limits
We
when
briefly
is
of
a
more
xo is equal to +oo or \342\200\224oo.
part
general
(This
theory
of continuous
on a topological space; see Section13.5.)
functions
of what it means for +oo or \342\200\224
we
a notion
oo to be
need
to a set.
adherent
Until

have

we

now,

to have

\342\200\224\342\226\272
R

First,

Definition 9.10.1
of R.

We

exists

an

say
\317\207
G

for

iff

has

iff

sup(X)

+oo.

bound, or

no lower

if and

bounded

only if

an

and

+oo

iff

inf

We

say

that

adherent

f(x)

limx->+00]xexf(x) =
that
\317\207
G

/ is
X

\316\265-close

such

that

to

L,

L on

\317\207
>

iff

for

we

here.

a subset

\\f(x)

function.

an

\342\200\224<

f(x)

of R

and write

X,

exists

0 there

say that

from

structure

topological
not discuss

in

is

points.
different

be

to

a set

\342\200\224oo.
Thus

: X -4 R bea

(i.e.,

bound,

adherent

adherent

to

\317\207
< \316\234.

no upper

rather

will

\316\265
>

every

Similarly

the

let /

that

\342\200\224oo
is

\342\200\224>
L as \317\207
+oo

\316\247
\316\240
(\316\233/,+oo)

\316\234).

has

Let

and

point,

seem

infinity).

converges to

such

not

there

is adherent

oo

(X)

\342\200\224oo
are

9.10.3 (Limitsat

with +oo as an

iff

\342\200\224

that

\317\207
G

a subset

\316\234
G

every

Similarly

Remark
9.10.2. This definition may
Definition 9.1.8, but can be unified
using
of the extended real line R*, which
we

Definition

for

iff

say

adherent to X

+oo is

words,

exists

there

X be

Let

points).

is adherent to
that \317\207
> \316\234;we

\316\234G

or equivalently iff
X

adherent

(Infinite

+oo
such

every

other

In

that

L\\

\316\234
such

for
\316\265

all

converges to

287

at infinity

Limits

gJO.

oo iff for every \316\265


> 0
I as x \342\200\224>
L
on
\316\247
to
\316\240
(\342\200\224oo,
M).
5-close
\342\200\224

9.10.4.

Example

we

Then

have

\317\207/\317\207.

from

the

there

exists

an

\316\234
such

that

/ is

: (0, oo) \342\200\224>


R be the function f(x) :=
=
Ume_>+oo;a.e(0|oo) l/\321\217 0. (Can you see why,

Let /

definition?)

do many of the same things


with
these
limits
at
with
been
limits
at
other
points xq\\ for
doing
infinity
it turns out that all of the limit laws
continue
to hold.
instance,
in
we
will
be
as
not
these
limits
this
much
text,
using
However,
not devote
much attention to these matters. We will note
we will
the
notion
of a limit
though that this definition is consistentwith
One can
as

we

have

\316\277\316\257
a sequence
0\"\320\277
\320\246\321\210\320\277-\321\216\320\276

(Exercise

9.10.1).

Exercise 9.10.1. Let (an)\302\243L0 be a sequence of real numbers, then an can


N to R, which
from
takes
each natural
thought of as a function
to a real number an. Show that
number \316\267
also be

lim

an

n\342\200\224+oo;neN

where

the

left-hand
is defined

limit is

defined by

lim an

\342\204\242\342\200\224\302\260o

Definition

9.10.3

and

the right-

that if one
show
hand
by Definition 6.1.8. Moreprecisely,
limits
exists
then
so
the
both
the
above
two
does
and
then
of
other,
they
the
same
value.
Thus
the
two
limit
here
are
notions
of
have
compatible.
limit

10

Chapter

of functions

Differentiation

10.1 Basic

definitions

with

starting
derivatives
definition

begin the rigourous treatment of calculusin earnest,


the notion of a derivative.We can now define
in contrast to the geometric
limits,
using

now

can

We

analytically,

of

is that

analytically
geometry,

and

these

(b)

of

several

variables,

of

scalar.

Furthermore,

to

rely

on once

(Conversely,

one's
earlier,

use

two

intuition becomes difficult

more than threedimensions

one's

intuition

geometric
the

one's geometric

one has

can

one

uses tangents. The advantage of working


do not needto know the axioms of
we
(a)
can be modified to handlefunctions
definitions
or functions whose values are vectorsinstead

which

derivatives,

in

play.

rigour to extend
settings; as mentioned
rather
than opposeeach
complement
in analytic

experience
to such abstract

viewpoints

other.)

subset

of

R,

and

of X.

point

at a point). Let X bea


(Differentiability
let x$ e X be an element
of X which is also a limit
\342\200\224>
Let / : X
R be a function.
If the limit
10.1.1

Definition

/\317\211-**.)

lim
x\342\200\224>xo;xeX-{xo}

entiable
the

limit

some real
xq on X with

to

converges
at

does

not exist,

X \342\200\224
Xo

number L, then
derivative

or

if

xq

L,
is not

we

and

say

that

write

an element

/ is

f'(xo)

differ\302\267

:= L.

of X or not

If

Basic

at

differentiable

for

and

undefined,

/ is

that

say

not

on X.

xq

10.1.2. Note that we


to be adherent to X

jtemark
order

we leavef(xo)

of X,

limit point

289

definitions

\316\231\316\270\316\233.

need

be a limit

to

xo

\342\200\224
otherwise
{\320\266\320\276},

xo

point in
limit

the

*WW

lim

X \342\200\224
Xo
x\342\200\224>\321\205\320\276;\321\205\320\265\320\245-{\321\205\320\276}

the

at an isolated point; for

function
the function /

one restricts

In particular, we do not define

be undefined.

automatically
of a
derivative

would

:R

if

instance,

by f(x) := x2 to the

defined

\342\200\224>
R

ceases
[1,2] U{3}, then the restrictionofthe function
at 3. (See however Exercise10.1.1below.)
In
to be differentiable
the
X
will
an
almost
so
domain
be
and
interval,
always
practice,
9.1.21
all elements
be limit
Lemma
xo of X will automatically
by

domain

points and

:=

we will not have

Example 10.1.3. Let /


let

and

at xo

be

these issues.

about

much

\342\200\224>
R

the function

To see whether

real number.

xo be any

f(x) := x2,

is differentiable

on R, we computethe limit
\320\271*\320\253\320\253=

lim

x\342\200\224Kro;a:eR\342\200\224{xq}

We can

factor the

R \342\200\224
\317\207
\342\202\254
{^o},

write

and

care

to

we

lim

=
numerator as (x2\342\200\224
Xq)

Since

(x\342\200\224xo)(x+xo).

the

cancel

\321\202\320\260\320\243
legitimately

the above

l^A.
x

^0\"
x\342\200\224>xo;z\302\243R\342\200\224{xo}

^o

of

factors

\342\200\224
\317\207 xo

limit as

lim

\317\207
+

xo

\317\207\342\200\224>\320\266\320\276\302\273

laws is equal to
by limit
differentiable at xo and its derivative
Remark
if

/
xo

: X

is differentiable

(i-e., g(x) = f(x)


and

g'{xo)

for

f(xo)

all

\317\207
e

(why?).

at

it

and
\320\266\320\276>
g

X),

then

f(x)

is

is 2xq.

there

10.1.4. This point is trivial, but


\342\200\224>
R

function

the
Thus
2\320\266\320\276\302\267

which

g is

However,

if

is worth
: X

mentioning:
is equal

\342\200\224>
R

to

also differentiable at
two

functions

/ and

10. Differentiationoffunction

290
g

not

that g'(xo)

imply

value at

the same

have

merely

Thus there is a big difference


on their whole domain,and

see a

you

(Can

f(xo)\302\267

\320\266\320\276>
i-e.,
g(xo)

being equal
at one point.

equal

being

merely

does

counterexample?\\

functions

two

between

this
/(\320\266\320\276)>

writes
instead
Remark 10.1.5. One sometimes
of /'. This
^
is of course
and
familiar
notation
but
convenient, one has
very
to use as long as
to be a little careful, becauseit is only
safe
\317\207
the
for
used
to
is the
variable
input
represent
/; otherwise
only
one can get into all sortsof trouble.For instance,
the
function
=
R defined
but
/ : R \342\200\224>
by f(x) := x2 has derivative
2\320\266,
the
jg

g :

function

defined

\342\200\224>
R

by g(y)

\317\207
derivative g| = 0 if \321\203
and
are
the fact that g and / are exactly
this possible sourceof confusion,

notation

whenever

\302\267\302\243;

introducing

Example 10.1.6.

Let

let

and

xo

= 0.

the

compute

resolve these ambiguities, most notably


of differentiation along vectorfields,

To see

but

this text.)

: R

f(x) := |x|,

function

be the

\342\200\224\342\226\272
R

whether / is differentiable

0 on

at

R,

we

limit

\320\270\321\202 /(\302\253)-/\342\204\226_
\316\266\342\200\2240;a:eR-{0}

Now

refrain

Because of
from using the

could possibly
lead to confusion. (This
worse in the calculus of several variables
notation
can
lead
to some serious
of g\302\243

the scope of

is beyond

this

will

we

despite

function.

same

it

ways to
the notion

are

There

ambiguities.

by

standard

the

and

variables,

independent
the

to have

seem

would

y2

even

becomes

confusion

:=

we take

left limits

Um

lim

\320\253.

\316\266\342\200\2240;zeR-{0}

and right limits.The right

\342\200\224
=

lim

x~\302\260

\342\200\224
=

lim

limit

1=

is

1,

\316\247 \316\266->0;\316\261:\302\243(0,\316\277\316\277)
\316\247 \316\266->0;\316\261:\316\276(0,\316\277\316\277)
\316\266->0;\316\261:\342\202\254(0,\316\277\316\277)

while

the

left

lim

limit

is

\342\200\224
\342\200\224
=
=
= \342\200\2241,
\342\200\2241
X
X
\316\266->0;\316\261:\342\202\254(-\316\277\316\277,0)
\320\266-^0;\320\266\320\261(0,\320\276\320\276)
\316\266->0;\316\261:\342\202\254(0,\316\277\316\277)

lim

lim

Basic

}Qj.

^d these

limits do not match.Thus

exist, and / is not


to

restricts

hma,_>0.a;6j^_

R.

0 on

at

differentiable

does not
rQ| ^
However, if one

then the restrictedfunction


[0, oo), with derivative 1:

[0, oo),

0 on

at

differentiate

291

definitions

lim

/(\302\273)-/\342\204\226)X \342\200\224
0
\320\266-^0;\320\266\320\261[0,\320\276\320\276)-{0}

\342\204\242

/|[o,oo)

,im

= 1.

\316\247
\316\266->0;\316\261:\342\202\254(0,\316\277\316\277)

one restricts
restricted
function
/ to (\342\200\224oo,
0], the
\342\200\224
1. Thus
at
0
on
with
derivative
differentiable
0],
(\342\200\224oo,
/I(-00,0]ls
it is sometimes
even when a function is not differentiable,
possible
the*
the
domain
of the
restore
to
by restricting
differentiability
when

Similarly,

function.

is differentiable

a function

If

at

is approximately

it

then
\320\266\320\276>

linear near xq:

Proposition 10.1.7
subset

let xo

ofR,

and let L

be

be a limit
number.

a real

R be a function,
point ofX, let f : X \342\200\224>
Then the following statements are

equivalent:

logically

(b)

For

every

\316\265
>

\342\200\224
\316\264-close

to

i.e.,

xo,

\\f(x)
whenever

the

\317\207
G

there

scientist

exists

- (f(x0) + L(x

\316\261
\316\264
>
\342\200\224

and

and

\\x

xo)

\342\200\224 <

xo\\

the

Remark

informal
approximation

We can

10.1.9.

way: if /
f(x)

\302\253

f(xo)

x0))\\

such

whenever

<

e\\x

that

is

\316\247is
\317\207
\316\225

x0\\

is of course named
Isaac Newton (1642and integral calculus.
D

phrase Proposition 10.1.7in

is differentiableat xo, then onehas


+ f'(xo)(x

f(x)

\316\264.

approximation
mathematician

of differential
founders
1727),oneof
Proof. See Exercise10.1.2.

L.

derivative

with

+ L(x
f(xo)
we have

Newton's

10.1.8.
great

0,
to

xo\\-close

e\\x

Remark

at xq on X

is differentiable

(a)

after

be a

Let

approximation).

(Newton's

\342\200\224
and
\302\243o)>

conversely.

the

more

10.

292

\\x\\ shows,

at

differentiable

R defined
function / : R \342\200\224>
by /(\317\207) :==:
be continuous at a point without
being
that point. However, the converseis true:

Proposition 10.1.10

implies

(Differentiability

continuity).

jre^

subset o/R, let xq be a limit point of X, and let f : X


Iff is differentiable at xq, then f is alsocontinuous
function.

be a

be

at

xq.

Definition 10.1.11 (Differentiability


a subset of R, and let / : X
is differentiable
at

differentiable

xq

\342\200\224\302\273\32

10.1.3.

See Exercise

Proof.

of functions

of the
example
a function
can

the

As

Differentiation

on X if,
on X.

on

for

xq

every

Prom Proposition 10.1.10and the above


immediate corollary:

Corollary 10.1.12. LetX


a function which is differentiable

a subset

be

on

Let X

domain).

be

We say that
X, the function /

a function.

be

\342\200\224>
R

we

definition

have

is

an

\342\200\224>
and let f :
R 6e
o/R,
X.
Then f is also continuous

on X.

all

state

we

Now

of derivativeswhich you

basic properties

are

with.

familiar

10.1.13 (Differential calculus). Let X be a subset of


a limit
R and g : X \342\200\224>
and let f : X \342\200\224>
R
point ofX,

Theorem
R; let

the

xq be

be functions.

(a)

is a constantfunction,

If f

that

such
xo

and

(b) If f
then

(c)

(Sum

is

= \321\201
for
f(x)
=
0.
f'(xo)
the

f is
rule)

identity

differentiable at xq
If f

and g

X,

then

i.e.,

f(x)

\317\207
G

function,

also differentiable at

there

i.e.,

all

a real number \321\201


is
f
differentiable at

exists

\317\207
for

all

\317\207
G

and f'(xo) = 1.

are differentiableat xq, then

xq, and (f + g)f(xo)

X,

+ g

is

+ \320\264'(\321\205\321\212)\
/7(\320\266\320\276)

Basic

10.1.

293

definitions

rule) If f

(d) {Product

is also

and

are

at

differentiable

at xo, then
= f'(xo)9(xo)

differentiable

and

xq,

(fg)'(xo)

fg
+

f{xo)g'{xo)\302\267

also

at

differentiable

if) (Differencerule)

\342\200\224

g is also

If

(?)

xo, and

for

and

If

differentiable

at

all

and

then

\321\213

Leibnitz

is non-zero on X (i.e.,
1/g is also differentiable at

then

are

is also

f/g

founder

of

differentiable

and g is

at xo, and

\302\267

rule is

product
Leibnitz

Gottfried

differential

at xq,

differentiable

^\321\2133

The

10.1.14.
after

rule,

aad

also known

the

as

(1646-1716), who was the other


integral calculus besides Newton.

Proof. See Exercise10.1.4.


As

you

are

well aware,

many derivatives

is

and g

\320\243

Remark

then cf

number,

= cf'(xo).

-fgf

rule) If f

non-zeroon X,

(cf)'(xo)

real

g are differentiable at xo, then


at xq, and (f-g)'(xo) = /'(^o)\342\200\224

xq,

\317\207
\316\225
X),

(\302\261)'(*o)

(h) (Quotient

and

xo,

\342\204\242
differentiable

0
\317\206

g(x)

at xq and \321\201
is a

is differentiable

If f

(e)

the above rules allow

easily. For instance,if

: R

one

to

compute
is the

\342\200\224 \342\200\224>
R

{1}

the above rules to


=
show
for
that
all x0 e R- {1}. (Why?
Note
that
f(x0)
(\320\266\320\27611)\320\260
\342\200\224
\342\200\224
is a limit point of R
every point xq in R
{1}
{1}.)
Another
of
fundamental
differentiable
functionsis the
property
function f(x)

:=

|5\316\263,

then

it

is easy

to use

following:

10.1.15

Theorem

xo

be a

X
\342\202\254

(Chain

limit point

rule).

Let

be
\316\247,\316\245

\320\243
be
\342\202\254
\320\276
ofX, and let \321\203

subsets
a limit

ofH,

point

let

ofY.

Let f : X \342\200\224>
\316\245
such that f(xo) = 2/0; and such that
be a function
\342\200\224\342\226\272
R is a function
f is differentiable at xq. Supposethat
g : \316\245

10. Differentiationof functions

294
is

which

at

differentiable

at xo,

differentiable

the

yo\302\267Then

and

|5|,

)2, and

(frf

instance

especially

variable,

independent

function

f(x)

:==

\316\277
==
/(\317\207)

then

for
\316\266

5(2/),

2(^f)^.

this
However,
the distinction

5^g|.
it blurs

js

Remark 10.1.17. If onewrites


in
the chain rule can be written
\320\267|

\342\200\224\302\273
R

\317\201

: R- {1}-> R is the
: R \342\200\224>
R is the function g(y) := y2,
the chain rule gives
<\302\273\302\260')'(*o)

manner

9'(yo)f'(xo)\302\267

If /

10.1.16.

Example

10.1.7.

Exercise

See

\320\276 :

and

(gofy(xo) =
Proof

function

for
\321\203

the

/(\320\266),

more

and

visually

then

appealing

notation can be misleading


(for
between dependent variableand
for y), and leads one to believe

dx can be manipulatedlikereal
numbers.
are not real numbers (in fact,
these
However,
quantities
not assigned
we
have
any meaning to them at all), and treating
For
them
as such can lead to problems in the future.
if
instance,
on x\\ and X2, which depend on t, then chainrule for
/depends
=
several
that \302\267\302\243
but this rule
variables
asserts
+ g^^>
-g^^
if one treated d/, di, etc. as real
seem
numbers.
It
suspect
might
is possible to think of dy, dx, etc. as \"infinitesimal
numbers\"
real
if one knows what one is doing,but
for
out in
those
just starting
I would
not recommend this approach, especiallyif one
analysis,
to work rigourously.
wishes
(There is a way to make all of this
of several variables, but it requires
rigourous, even for the calculus
the notion of a tangent
and
the derivative
vector,
map, both of
which
are beyond
the scope of this text.)
that

the

quantities

dz, dy,

Exercise
10.1.1.

that
of X,
X is a subset of R, xo isa limit
point
a function which
X be
at x$. Let \316\245
is differentiable
\320\241
also limit point of Y. Prove
function
the restricted
that
Suppose

and /

:X

such that

\342\200\224\342\226\272
R is
Xq

is

Local maxima, localminima,and

10.2.
f\\Y :

at \320\226\320\276>
and
has the same derivative as
does not contradict the discussionin Remark

also differentiable

\342\200\224\302\273
R is
\316\245

why this

Explain

xo.

at

295

derivatives

10.1.2.

10.1.2. Prove Proposition 10.1.7.(Hint: the


to be treated separately.)

Exercise

cases

=
\317\207

x$ and

have
\317\2070
\317\206

Exercise 10.1.3.Prove
laws

9.3.14),

(Proposition

Exercise10.1.4.Prove
Proposition
the

For

f{*)(9(x)

Let

10.1.5.

function f(x)
nxn~l for all

function

the
=

Exercise

\316\267
be

:= xn.
\317\207
R.
\302\243

Exercise10.1.6.Let

10.1.13.

use the

(Hint:
theorem

this

to

f(x)g(*o) + f(*Mxo) - /(*\316\277)0(*\316\277)


~ /(so)Mso);
~
fl(*o)) + (f(x)

Show

(Hint:
\316\267
be

f(x) :=

R be the
number, and let / : R \342\200\224>
=
that
/ is differentiable on R and
f'(x)
use Theorem 10.1.13 and induction.)

a natural

a negative

xn.

that

Show

\320\277\321\205\320\277~\320\263
all \317\207
R
for
\342\202\254

integer,

{0}.

\342\200\224\342\226\27
and let / : R \342\200\224
R
{0}
is differentiable on R and

10.1.5.)

has to treat the case f'(xo) =


subtleties can occur in that
case.)
Local

10.2

you

application

now

manner.

10.1.13and

(Hint: use Theorem

10.1.7. Prove Theorem 10.1.15. (Hint: one way


via Newton's approximation, Proposition10.1.7.Another
this
to convert
Proposition 9.3.9 and Proposition10.1.10
with
the latter
one involving limits of sequences,however

We

in

laws

limit

prove the latter.

Exercise

As

limit

and subtracting an intermediate term is sometimes


\"middle-man trick\"
and
is very useful in analysis.)

as the

f'(x)

the

of adding

trick

Exercise

be

fi^Mx)

either use

use the identity

f{xo)g{xo)

(Hint:

Proposition 10.1.7.)

Theorem

rule,

f{x)g{x) -

this

or use

9.3.14. Use earlier parts of

product

known

10.1.10.

Proposition

maxima,

local

separately,

minima,

as

some

to
way

of using

present

division-by-zero

and derivatives

again, but this timein

common
minima.

rigourous

is
use

into

strategy

basic calculus courses, one very


derivatives is to locate maximaand

this material

this

is to

problem

in your

learnt

do

one

10.

296

The notionof a
minimum at a point xo
function

10.2.1 (Local maximaand


X.
We
and let \320\266
\342\202\254
function,

Definition
a

be

maximumat xo

a
-

exists

there

iff

\316\264
>

10.2.2.

Remark

/ attains

from the

We

\342\200\224\302\273
R,

a maximum

attains
+ \316\264)
\316\264,\317\207\316\277

at

minimum

local

/ :X

attains a local
the restriction

say that /
0 such that

If / attains a maximumat xo,we


a globalmaximumat xo, in order to
if /

that

Note

here.

also
certainly
minima.

it

then

xq

iff

there

attains

sometimes

to

say

it

distinguish

attains a global

a local maximum

for

similarly

10.2.3.

Example

x2

+ 5)

scq,

and

or

the restriction f\\xn(x0-S,x0+S)


\302\260f
/
attains a minimum at \320\266\320\276\302\267

local maximadefined

maximumat
at this xo,

Let

that

0 such

\342\200\224
\316\247
\316\240
xo
\316\264,
(\320\266\320\276

that

9.6.5.

in Definition

minima).

\316\264
>

(x0
f\\xn(xo-s,x0+S)of / to \316\247\316\240
at xo. We say that / attainsa

exists

a maximum

attaining

defined

was

\342\200\224>
R

definition:

this

localize

now

: X

Differentiation of functions

Let / : R

the function

denote

\342\200\224>
R

f(x) :=

attain a global minimum


at
0,
it
however
does
attain
/(0),
:=
1 and restrict / to the
a local minimum,
for
if we choose
\316\264
for all \317\207
x4 < x2 and thus
interval
then
G (\342\200\2241,1)we have
(\342\200\2241,1),
\342\200\224>
=
=
x2
xA
so
a local minimum
0
and
has
f{x)
/(0),
/|(_1\320\264)
at 0.
\342\200\224

This

xA.

does not

function

for example

since

/(2) =

\342\200\22412
<

Example 10.2.4. Let /


on

defined

the

minimum

global

minimum

local

point xq in X,
the

then

xo

(why?),
at every
If /

(why?).

and

restriction
Similarly

The connection
is

the

following.

be

\342\200\224\302\273
\316\226
R

Then /
but attains

only.

integers

10.2.5.

Remark

0 =

integer
:X

\316\245
X
\320\241

/|y
for

the

function

f(x) = x,

has no globalmaximum
both a localmaximum

or
and

\316\267

(why?).

attains

\342\200\224>
R

is a

a local

maximum at a

X which containsxo,
attains a local maximum at

subset of
also

\342\200\224>
\316\245
R

minima.

between local maxima,minimaand derivatives

297

derivatives

(Local extrema are stationary). Leta <b be


: (a, b) \342\200\224>
R be a function.
If xo \342\202\254
(a,b),
at xo, and f attains eithera local maximum
or

10.2.6

Proposition

and

numbers,

real

local minima, and

Local maxima,

10.2.

is differentiable

let f

at xq, then f{xo)

local minimum

ProofSee

= 0.

10.2.1.

Exercise

to work;
be differentiablefor this proposition
doesnot work if the
see Exercise 10.2.2. Also, this proposition
interval
is
a
interval
closed
replaced by
(a, b)
[a,b]. For
open

/ must

that

Note

function

the

instance,

at xq

maximum

local

: [1,2]

defined

\342\200\224\302\273
R

these local extrema are globalextrema),


derivative is f(xo) = 1? not f(xo)

an interval can be localmaxima


the
is not zero there. Finally,

or

converse

xo

0. Thus

minima

(Exercise10.2.3).
10.2.6

Proposition

combining

By

with the

onecanobtain

Theorem10.2.7
differentiable

anx

on

(a,

Suppose
that

See Exercise

Remark 10.2.8.
the open
(a,
we

/ is

assume

is larger

this

g'{x)

derivative

proposition

is false

maximum principle,

0.

10.2.4.
that

Note

interval

if

if the

be a

\342\200\224>
R

b) such

G (a,

Proof

6]

b).

fact,
points the

Let a < b be real numbers,


theorem).
continuous function which is
also that g(a) = g{b). Then
there
exists

(Rolle's

g : [a,

let

and

\317\207
has

1 (in

the endpoints of

even

of this

both

at

but

:=

by f(x)

= 2 and a localminimum

b),

we

assume /
of course the

only

though

is differentiable on
theorem alsoholds

differentiable on the closedinterval

[a,b],

since

than (a, b).

Rolle's

has

theorem

an

important

corollary.

numbers,

and

[a,b]

and

10.2.9 (Mean value theorem). Let a <b be real


R be a function
let f : [a, b] \342\200\224>
which is continuouson
on
Then
there
existsan \317\207
\342\202\254
differentiable
(a,b)
(a,b).

such

that

f{x)

Corollary

mbZfa{a)

\342\226\240

10.

298

of functions

Differentiation

Proof See Exercise10.2.5.


Prove

10.2.1.

Exercise

an

0.

example

Explain

/ :

of a function

and attains a global


at

differentiable

10.2.6.

Proposition

Exercise 10.2.2. Give


is continuous

\342\226\241

not

does

this

why

at

maximum

0,

\342\200\224\302\273
R which
(\342\200\2241,1)

which is

but

contradict

not

Proposition

10.2.6.

Exercise10.2.3.Give

an

neither a local minimum


not contradict

the

boundary

hypothesis

Explain

this

why

0 is
does

Proposition 10.2.6.

Exercise10.2.4.Prove
and the maximum
10.2.6. Note
the maximum

maximum.

\342\200\224\302\273
R which
(\342\200\2241,1)

0, but such that

0 at

equals

a local

nor

/ :

of a function

example

and whose derivative

is differentiable,

10.2.7.

Theorem

(Hint:

use Corollary

10.1.12

9.6.7, followed by Proposition


Proposition
maximum
does not tell you whether
principle

principle,
the

that

or

is in the

minimum

points a, 6, so you have


= g(b) somehow.)
g(a)

open interval (a, b) or is one of


to divide into cases and use the

10.2.9.
Exercise 10.2.5.UseTheorem10.2.7to prove Corollary
(Hint:
for some
of the form f(x) \342\200\224ex
carefully chosen real
number c.)
R be a function which
Exercise10.2.6.Let \316\234> 0, and let / : [a, b] \342\200\224>
is
<
on [a, b] and differentiable
continuous
on (a, 6), and such that
|/'(s)|
\316\234
the derivative of / is bounded). Show
for all \317\207
that
\342\202\254
for
b) (i.e.,
(\316\261,

consider a function

we have
\342\202\254
\321\205,\320\243
[\316\261>\320\254]

any

the inequality

\\f(x)

\342\200\224

f{y)\\

< M\\x

\342\200\224

y\\.

(Hint:

the mean value theorem (Corollary 10.2.9)to a suitable


restriction
\342\200\224
< M\\x \342\200\224
of /.)
Functions which obey the bound \\f{x)
f{y)\\
y\\ are
as Lipschitz continuous functions with
known
constant
thus
\316\234;
Lipschitz
are Lipschitz
this exercise shows that functions with
bounded
derivative
apply

continuous.

Exercise 10.2.7.Let
/' is bounded. Show
preceding exercise.)

that

elementary

your

the assertion

that

function,

: R

\342\200\224>
R be

is

a differentiable

and

a negative

calculus

courses,

positive

derivative

derivative

continuous.

uniformly

Monotone functions and

10.3
In

function
(Hint:

such that
use the

derivatives
you may have come across
meant an increasing

meant a

decreasing function.This

Monotone

10.3.

accurate,
statement is not completely

the

give

now

proposition 10.3.1.
and let f :
point of X,

f is

and

<

decreasing

Remark 10.3.2.

we

Exercise
cannot

One

be a

be a

\342\200\224>
R

might
and

and of

which are

course if

conclude

naively

guess

that

f(xo)

that

/ was differentiable

if /

at

/ is differentiableat

that

assume

to

functions

10.3.2),
possibly

increasing,

have

We

monotone

exist

There

10.3.1.

See Exercise

Proof

statements below.

o.

/'\320\253

(see

is

monotone

close; we

is pretty

subset o/R, let xo be a limit


function. If f is monotone
at xo, then f(xo) > 0. If
differentiable
is
and
f
differentiable at xo, then

Let

increasing

it

but

of these

version

precise

299

and derivatives

functions

xo>

not always

is not

x$\\

differentiable

differentiable

> 0 or f(xo)

< 0.

at xq

were strictly monotone


the

then

derivative

f'(xo)

strictly positive insteadof merelynon-negative.


this
is not always the case (Exercise 10.3.3).
Unfortunately,
other
we do have a converseresult: if function
the
On
hand,
has strictly
positive derivative, then it must be strictly monotone

would be

increasing:

10.3.3.

Proposition

Let a

<

b,

and

let

: [a,b]

\342\200\224>
R

be a

then
\342\202\254
f is
differentiable function. If f'{x) > 0 for all \317\207
[a,b],
then
monotone
all
\317\207
0
<
\342\202\254
strictly
increasing. If f(x)
for
[a,b],
\342\202\254
f is strictly monotone decreasing. If f(x) = 0 for all \317\207
[a,b],
then f is a constant function.

Proof. See Exercise10.3.4.

Exercise

10.3.1.

Prove

Proposition

10.3.1.

R which
Exercise 10.3.2.Give an example of a function / : (\342\200\2241,1)\342\200\224\342\226\272
and monotone increasing,but which is not differentiable
at 0.
Explain
why this does not contradict Proposition10.3.1.

is continuous

10.

300

zero.

10.3.3.

Exercise10.3.4.

one can

However,

which

derivative

is

at

not contradict Proposition 10.3.1or


Exercise 10.2.3.)

Proposition

fundamental

the

\342\200\224\342\226\272
R

whose

but

functions

10.3.3.
(Hint: you do not have
theorem of calculus yet, sothesetoolscannot

Prove

integrals or

look at

(Hint:

/ :

function

differentiable,

does

this

why

Explain

Proposition

used.

exampleof

monotone increasing and

strictly

0 is

Give an

10.3.3.

Exercise

of

Differentiation

proceed via

mean-value

the

theorem,

be

Corollary

10.2.9.)

Exercise 10.3.5.Give an example of a subset X


R which
that
X \342\200\224>
is differentiable on X, is such
but /

is not strictly

subtly different
and how can

the

following

What

that
question: if we know
it has an inverse /_1
of

with

\342\200\224\342\226\272
\316\245
be an

:
that xq
f~l
Suppose
=
also
yo
impliesthat
f(xo) (which
at xq, and f~l is differentiable
differentiable
\342\200\224>
\316\245
X.

inverse

invertible

(f-1ofY(xo)

But /_1
10.1.13(b)

is the

\320\276

(/-1

function,

and

X
\342\202\254
xq

=
at

yo

f)'(x0)

such

yo,

is

then

have

(r1y(yo)f'(xo).

identity function

\320\276

\316\245
are
\342\202\254

with

If f
/_1(\317\212/\316\277))\302\267

chain rule (Theorem10.1.15)


we

Prom the

what

\342\200\224\342\226\272
\316\245
X,

This will be useful


want to differentiate the

that

Proof

/ :

a function

result.

a preliminary

10.4.1. Let f : X

Lemma

difference,

/_1?

if we

for instance

applications,

begin

is the

derivatives

function f(x) := xlln.


We

> 0 for

the example?)

to obtain

difference

say about the differentiability

many

the

/ :
all \317\207
e Jf,
conditions here are
a function

and

and

\342\200\224\342\226\272
\316\245
is differentiable,

can we
for

one exploitthat

ask

now

(Hint:

10.3.3.

increasing.

in Proposition

those

Inverse functions and

10.4
We

monotone

from

R
\320\241

f'(x)

1. The

on X,

\320\260\320\273\320\260
hence
by

claim follows.

Theorem
\342\226\241

Inverse

and

functions

\316\231\316\2704-

301

derivatives

that
if / is
we
see
particular corollaryof Lemma10.4.1,
=
with
at
cannot
then
differbe
xo
/_1
0,
f(xo)
jifferentiable
since l/f(xo) is undefined in that case.
at yo = f(xo),
entiable
the function
for
defined
instance,
g : [0, oo) \342\200\224\342\226\272
by
[0, oo)
Thus
cannot
be
differentiable
at
since
this
is
function
:=
0,
y1/3
g(y}
=
\342\200\224>
of
function
:
the
defined
inverse
/_1
/ [0,oo) [0, oo)
g
by
the
=
function
has
of
at
and
this
a
derivative
0
:=
0.
\320\2663,
/~1(0)
f(x)

As a

If one
the

write

=
dx/dy

\321\203

so that

f(x),

conclusion

of Lemma

l/(dy/dx).

However,

=
\317\207

then

f~l{y),

one can

10.4.1 in the more appealingform


of
as mentioned before, this way

while very convenient and easy to remember,


can
and cause errors if applied too carelessly(especially

things,

writing
be

writes

misleading

calculus of several variables).


Lemma10.4.1seems to answer the question of how to
the
of a function, however it has one significant
inverse
differentiate
the
lemma
drawback:
only works if one assumesa priorithat
Z\"\"1
that
is differentiable.
Thus, if one does not already know
/_1
one cannot use Lemma 10.4.1to compute
the
is differentiable,
one

when

derivative

to work

begins

in the

of/-1.

will
the following improved versionof Lemma10.4.1
this fact, by relaxing the requirementon f~l from

However,

for

compensate

to

differentiability

continuity.

(Inverse function theorem). Let f : X -+Y be


\342\200\224>
X.
that
with inverse f~l : \316\245
an invertible
Suppose
function,
X
\316\245
are
such
that f(xo)=yoand yo \302\243
\342\202\254
xq
If f \320\2635
is
at
and
differentiable at
continuous
xo, f\"1
yo,
f(xo) \317\2060, then f~l is
differentiable at yo and
Theorem

Proof

10.4.2

We have

to show

that
1

,im

/-'\320\234-\320\223'\320\253.

2/

y->yo;yeY-{yo}

By Proposition

9.3.9, it sufficesto

2/0
show

that

lim

fco)

\316\271
\320\223\320\247\321\203\320\277)
\320\223\320\247\321\203\320\276)

n-*\302\260\302\260

\320\243\320\277
2/0

f'(xo)

10.

302

for

sequence

any

to

Differentiation of functions
in

of elements

(yn)^=i

\342\200\224
\316\245

which

{yo}

To

elements

of

xn := /_1(2/n). Then

we set
in X

this,

prove

sequence

\342\200\224

lim f{Xn)

since

rrn

non-zero.

Xn-xp

,.

since

xn

since
Thus,
9.3.9 again)

f(x0).

lim

'

f'(x0)

thus

we
\320\266\320\276
/-1(\317\212/\316\277)>

and

/_1(\321\203\320\276)

have

_i_

\320\223\320\247\321\203\320\277)-\320\223\320\247\321\203\320\276)

n-\302\260\302\260

2/n

2/0

/\320\247\320\266\320\276)

desired.

some applications of the


the exercises below.
We give

>
Exercise10.4.1.Let \316\267
(0,

we
oo.

/ is

n-oo f(Xn) - f(x0)


But

/_1 is a
know that

a bijection, the fraction \320\220\320\266\~\320\233\320\266\320\2


js
\342\200\242\316\225\317\204\316\27
xq
non-zero.So
is
limit
laws
by
by hypothesis f(xo)

and
\317\207\316\277
\317\206

Also,

that

/(\320\226\320\276)

\342\200\224
\316\267\342\200\224>\316\277\316\277
\317\207\316\267
Xq

But

(\320\266\320\277)^is

Note

(Why?

{xo}-

assumption,
by
bijection) Since /_1 is continuous
->
xn = f~l{yn) convergesto f^iyo) = xo as \316\267
/ is difFerentiableat xq, we have (by Proposition

as

converge

\321\203\321\201

be
ex\302\273)

(a)

that

Show

a natural number, and let g


:=
x1/71.
g(x)

function

the

g is

1 be

continuous

on (0, ex\302\273).(Hint:

(b) Show that g is difFerentiable


for all \317\207
\302\243
(0,oo).
(Hint:

: (0,00)-*\342\226\240

use Proposition

and

on (0,00),
use the

that

in

theorem

function

inverse

g'(x)

9.8.3.)
=

^x*~l

inverse function theorem

and

(a)\302\267)

10.4.2.

Exercise

the function

(a)

Show

Let q

f(x) =
that

be a

rational

/ is

differentiable on

(Hint: use Exercise10.4.1and


Theorem

and

number,

let /

xq.

10.1.13

and Theorem

the

(0,00) and
laws

10.1.15.)

that

: (0,00) \342\200\224\302\273
R be
f'{x)

of differential

= qxq~l

calculus

in

10.5. L'Hopital's rule


=

that

Show

(b)

303

lima._>1;a.\342\202\254(0)OO) ^\316\236\317\204Q

(Hint: use part

(a)

rational number q.

for every

10.1.1.

Definition

and

apply L'Hopital's rule from the next section.)


be a real number,
and let / : (0,oo)
Exercise 10.4.3. Let \316\261
=
\320\266\320\260.
function f{x)
that

Show

(a)

limx_i.x\342\202\254(0)OO)

limits

left

and

right

f^*lz{^

the comparisonprinciple;

10.4.2 and

separately.

<*.

the

\342\200\224\342\226\272
R be

may

Exercise

use

(Hint:

you

is to

route

alternate

An

to consider

need

Proposition 5.4.14 may

also

be

helpful.)

/ is differentiable

(b) Show that

(Hint: use (a), exponent

on
laws

(0, oo)

(Proposition

and that f'(x) = axct~1.


6.7.3), and Definition

10.1.1.)

L'HopitaPs rule

10.5

we

Finally,

a version

present

of a

rule you

are

all

with.

familiar

10.5.1 (L'Hopital's rule I). Let X be a subset o/R,


and let xo be a limit
R be functions,
and g : X \342\200\224>
let f
=
=
0; ^hat f and g are
<?(#o)
point of X. Suppose that f{xo)
there exists\316\261
Then
\316\264
at
but
> 0
both
A0xo,
differentiable
g'(xo)
\342\200\224
\342\200\224
all \317\207
and
0 for
\316\240
\342\202\254
such that g(x) \317\206
xo + \316\264)) {xo},
\316\264,
(\316\247
(\317\207\316\267
Proposition
: X

\342\200\224>
R

/w

lim

\320\263\321\213

\317\207^\317\207\316\277\\\317\207<\316\236(\317\207\316\267(\317\207\316\277-\316\264,\317\207\316\277+\316\264))-{\317\2070}
g(x)
g'{xo)'

The presenceof

\316\264
here

the

neededbecauseg(x) might
would

which
all

in

points

A more

that

imply

10.5.1.

See Exercise

Proof

may
vanish

quotient

strange, but is
other
than xo,
points
is not
defined at
\320\246\320\275\320\263
necessarily
seem
at

somewhat

some

\342\200\224

{xo}.

version of

sophisticated

L'Hopital's rule is the

following.

10.5.2

Proposition
numbers, let

: [a,

b]

(L'Hopital's
and g

\342\200\224>
R

are differentiable on

[a,6].

Suppose

rule II).

: [a, b]
that

Let a <

\342\200\224>
R

f(a)

be

real

be functions
=
=

g(a)

which

0, that

10.

304
g'

on [a, b]

is non-zero

0
\317\206

g'(x)

(i.e.,

Differentiation

\317\207
\342\202\254
(a, 6],

and

Remark

10.5.3.

all

for

\317\207
\342\202\254

0
\317\206

for

\320\260\320\246

L.

and
\320\246& eziste
\320\230\321\202\320\266_>\320\260;\321\217.\320\265(\320\260|5]

equals

This propositiononly

ani

[a,b])7

Then g(x)

equals L.
\320\235\321\210\321\217._\320\260;\321\217.\320\265(\320\260|\321\206
^fjand

eszste

offunction

considers

limits

to the

but one caneasily state and prove a similar proposition


to the left of a, or aroundboth sidesof a.
Speaking
very

right of a,
for

limits

the

informally,

states that

proposition

lim -j-t =

x~*a 9\\x)

to ensure all
that f(a)
particular,
has

one

though
hold

(in

limit

exists), before

Proof.
Suppose

of the conditionsof the proposition


= g(a) = 0, and that the right-hand

one can apply

rule.

L'HopitaTs

We first show that g(x) \317\206


0 for
all
(Optional)
=
for sake of contradiction that g(x)
0 for some
g(a) is also

since

But

hm

x~*a -^-(-,
9 \\x)

zero, we can apply

but
< \317\207,
9l{y) = 0 f\302\260rsome a < \321\203
that g' is non-zeroon [a,b].
we

Now

9.3.9, it

will

show

that

suffice

to

this

\317\207
\342\202\254
(\316\261,
6].

to obtain
the hypothesis

theorem

Rolle's

contradicts

\317\207
\342\202\254
(\316\261,
6].

L.

By Proposition

\320\230\321\202\321\205^\320\260.\321\205\320\265(\320\260\320\264
^j

show

that

\317\200-\316\277\316\277

g(xn)

for

sequence

any

values

(xn)\342\204\242=l taking

to x.

Considera
R defined

single

and

xn,

a and

that
rrn,

(a, b] which

the function

consider

hn

converges
: [a,

xn] ->

by

K{x) :=
Observe

in

and

hn

f{x)g{xn)

is continuous

is difFerentiable

f'{x)g{xn)\342\200\224g'(x)f(xn)\302\267

(Note

g{x)f(xn)\302\267

on

and

[\320\260,\320\266\320\277]equals

xn) with derivative h'n(x)=


and g[xn) are constants
f(xn)

on (a,
that

0 at both

10.5. L'Hopital's

305

rule

to

respect
\302\253rith

thus

find

Since

\302\243
such
\316\267
\321\203
(\316\261,
\316\226\317\200)

2/n ^

theorem (Theorem 10.2.7),we can


= 0, which implies that
h'n(yn)

By Rolle's

x.)

for

that

/(an) =

/'\320\253

5^n)

\320\267\320\247\320\243\320\277)\"

n, and

all

(\320\260,\320\266\320\277)

rrn

Prove

10.5.1.

Exercise

0 near

\320\266\320\276>
you
may

For

10.1.7).

Proposition
wish to use

the rest

10.5.1.

Newton's

a as

to

converges

that
see from the squeeze test (Corollary6.4.14)
\342\200\224\342\226\272
oo.
Thus
converges to L,
to a as \316\267
^rn(
to L, as desired.
converges

yn

\342\200\224\342\226\272
we
\316\267
oo,

also

converges

and thus

(Hint: to

also

show

that

g(x)

\317\206

(Proposition

approximation

of the proposition,use limit

\316\2464

laws,

Proposition

9.3.14.)

Exercise 10.5.2.
of

the

propositions

Explain why
in this

Exercise

section.

1.2.12

does not

contradict either

11

Chapter

The Riemann

integral

differentiation - oneof the two


of single
variable calculus. The other pillar is, ofcourse,
mpillars
which is the focus of the currentchapter.More
tegration,
precisely,
we will turn to the definite integral,the integralof a function
on
as
to
the
a fixed
otherwise
interval,
indefinite
integral,
opposed
of course
are
known
as the antiderivative. These two
linked by
which
will
of
more
the
Fundamental
theorem
be said
of calculus,

In

the

we reviewed

chapter

previous

later.

For us,

the study of the

definite

will

integral

start

with

an

orhalf-open,
and
a function
interval / which couldbe open,closed,
\342\200\224> and
:
J
a
can
will
lead
us
to
number
we
write
this
R,
/
Jj /;
we could
replace \317\207
by
any other
integral as Jj f(x) dx (of course,
J
a
and
shall
also
or
if
has
we
write
b,
endpoints
dummy variable),
this

as f

integral

To
geometric

if

one

does

notions

Ja f(x) dx.
this

define

actually

(especially

f or

such as

tegrable. It turns out that

this integral: the Riemann


(1826-1866), whichwe will
applications,

and

the

(1875-1941),

which

for

a much

constructed

Lebesgue
supercedes

larger class
in Chapter

Jj f is somewhat delicate
to assume any axioms
concerning

integral

not want

area), and not

there

are

at

all

least

functions

two

/ are in-

ways to

define

integral,
Georg Riemann
do here
and which suffices for most
integral, named after Henri Lebesgue
Riemann
the
integral and works

named

after

of functions. The Lebesgueintegralwill


19. There is alsothe Riemann-Steiltjes

be
in-

307

Partitions

\320\246.1.

f(x) da(x), a generalizationof the Riemann


Thomas Stieltjes (1856-1894), which we will discuss

legral JT
to

11.8.

Our strategy in

begin by

first

definingthe Riemann

defining

Section

in

is as

integral

on a very

of integration

a notion

due

integral

We

follows.

class

simple

piecewise constant functions. Thesefunctions


but their advantage is that integrationis very
primitive,
axe quite
as is verifying
all
these
usual
the
for
functions,
properties.
easy
them
Then, we handle more generalfunctions
by
approximating
constant
functions.
by piecewise
- the

of functions

Partitions

11.1

one

how

describe
In

this

chapter,

to

the

more

can
all

general

Let

is connected

the

iff

in X

be

\317\207
<

the
\321\203,

11.1.2. Later on, in

which

in

lie

[1,2],

and
between\317\207
the

set

(1,2)

then 1

also
\321\203

lies

<

to any

2,

and

are
\320\266,
\321\203

[x,y] is
in X).

define a
metric

because

connected,

<
\321\203

\317\207
<

we will

13.4
applies

11.1.3. Theset [1,2]is

Examples
both

whenever
is also
\321\203

so every

that

say

interval

\317\207
and

Section

generalnotionof connectedness,

We

is true:
bounded

between

number

of R.

a subset

property

following

such that

subset of X (i.e., every


Remark

to

need

a large interval into smaller intervals.


partition
intervals will be bounded intervals (as opposed
intervals defined in Definition 9.1.1).

Definition 11.1.1.
elements

of an integral,we

the concept

introduce

can

we

Before

more

space.

if

\317\207
< \321\203

element

argument shows that


the
set
However,
[1,2]U [3,4] is not

in [1,2].

is connected.

A similar

The
empty
(why?). The real line is connected
(why?).
well as singletonsets such as {3}, are connected, but for
rather trivial reasons (these setsdonot contain
two elements
\320\266,
\321\203

connected

set, as
for

which

\317\207
<

y).

Lemma

11.1.4.

following

two

Let X

statements

be

are

subset

logically

of the

real line.

equivalent:

Then the

11. The Riemannintegral

308
and connected.

is bounded

(a)

(b) X is a

interval.

bounded

11.1.1.

See Exercise

Proof.

Remark 11.1.5.
points (e.g., the

q
interval

degenerate

Corollary

11.1.6.

intersection

\316\231\316\240
J

// J and J are bounded


also a bounded interval.

is

Proof. See

[4,6] is {4},

now

each

give

Definition 11.1.8
define

we

the

(Length

of

length

of J,

set,

we define

Example 11.1.9. For


length

The

interval.

of

(3,5);

interval a

bounded

of the intervals [a,b], (a,


a < b, then we define \\I\\
empty

of the boundedintervals

which is alsoa bounded

[2,4]

intersection

and (4,6) is 0.

of (2,4)
We

the

\342\226\241

The intersection

11.1.7.

Example

then

intervals,

11.1.2.

Exercise

interval,

[2,2]

set.

empty

and

to be singleton
{2}), or even the

are allowed
=

intervals

that

Recall

If J

intervals).

denoted

[a,

b),

bounded

\\I\\

follows.

b]

for

some

b), or (a,
a.

is a

as

if /

\342\200\224

:=b

length.

Otherwise,

If J
real

is a

is one

numbers

point or the

|/| =0.
the

instance,

meanwhile,

length

the length of

of [3,5]

is 2, as is the

{5} or the

empty

set

is

0.
A
11.1.10 (Partitions). Let J bea bounded
interval.
in
is
a
\316\241
of
bounded
intervals
contained
finite
set
/,
partition
\317\207
such that every
in J lies in exactly one of the boundedintervals

Definition

of J

in P.

Remark

11.1.11. Notethat a partition

each interval is itself a set of real

set consistingof othersets.

numbers.

is

a set
Thus

of intervals, while
a partition

is a

llJ.

309

Partitions

The set

11.1.12.

=
\316\241

{{1}, (1,3), [3,5), {5}, (5,8],0}


partition of [1,8], becauseall the intervals
one
lie m [15 8], and each element of [1,8] liesin exactly
in \316\241
\316\241
P.
Note that one could have removed the empty
from
set
intervalin
Examples

still

and

is a

intervals

0f bounded

However, the set

a partition.

obtain

{[1,4],[3,5]}is

because some elements of [1,5]are included


a partition
than one interval in the set. The set {(1,3),
not
more
of (1,5) because some elements of (1,5) arenot included
partition
in the set. The set {(0,3),[3,5)}
interval
is not a partition
in any
because some intervals in the set are not
contained
of (1,5)
of [1,5]

(3,5)}is

not

in
a

in

(1,5).

Now we
Theorem
interval,

come to a basicproperty

11.1.13

(Length is finitely

a natural

\316\267
be

about

Let

additive).

and let

number,

length:

\316\241
be

I be

a partition

a bounded
of I of

cardinality n. Then
\\i\\

=
\316\243

\320\270\302\267

JeP

Proof

P(n) be the
whenever

The

this by inductionon n.

We prove

property that

\316\241
is

base

partition

case P(0)

partitioned into an empty

whenever

of /

J is

More

we

precisely,

a bounded

with cardinality n, that

is trivial; the

only

way

let

interval, and

that

\\I\\

J can

be

is if J is itself empty (why?),


partition
at which point the claimis easy. The case P(l) is also very easy;
the only way that J can be partitioned into a singletonset {J} is
if J = /
is again
very easy.
(why?), at which point the^claim
>
Now suppose inductively that P(n) is true
\316\267
for
some
1,
and now we prove P(n +1). Let J bea bounded
and
let
interval,
\316\241
be
a partition
of J of cardinality \316\267
+ 1.
If I is the empty
or a point,
then all the intervals in \316\241
set
must
so every
also be either the empty set or a point(why?),
and
we will
interval has length zero and the claimis trivial.
Thus
form
assume that J is an interval
of
the
(a,b), (a, 6], [a, 6), or

11.

310

Riemann

The

integral

suppose that b e J, i.e., / is either (a,b] or [a,6].


\320\232
in \316\241
b G J, we know that one of the intervals
Since
contains
6.
in J, it must therefore be of the form
Since \320\232is contained
(c,6],
< b
a < \321\201
c, with
(in the latter
[c, b], or {b} for some real number
this means that the
:= b). In particular,
case of \320\232= {b}, we set \321\201
\342\200\224
J
\320\232
form
is
also
an
of
the
interval
set
(a,c),
[a,c],
(a,c],
[a,c)
or
a point
or empty set when a = c. Either
when \321\201
> \316\261,
we
way,
us first

Let

see that

easily

|/| =

On the other hand,


\342\200\224
\316\241
{\320\232}

hypothesis,

forms

since

a partition

|/-*|.

+
|\317\213\316\223|

a partition

\316\241
forms

of J

\342\200\224
\320\232

of J,
By

(why?).

we see that

the

induction

we thus have

\\I-K\\=

\\J\\.

JeP-{K}

Combining

finite sets,

these two identities (and using the laws


see Proposition 7.1.11)we obtain

of

addition

for

J\\
\316\271/\316\271
\316\243 \316\271

JeP

'

as desired.

Now suppose that b # J, i.e., J is either (a, b) or [a, b). Then


one of the intervals \320\232also is of the form (c, b) or [c,b) (see Exercise
\320\232
In particular,
this means that the set I \342\200\224
is also
an
11.1.3).
\321\201
or
of
the
form
a
interval
> a,
point
[a, c], (a, c), (a, c], [a, c) when
or empty set when
a = c. The rest of the argumentthen proceeds

as

above.

are two more things we need to dowith


when
one partition is finer than another,
say

There

is to

is to talk about the commonrefinement

of

two

partitions.

and

One

the other

partitions.

11.1.14 (Finer and coarser partitions).Let J be a


interval, and let \316\241and P7 be two partitions of J. We
P7 is finer than \316\241
that
than
is coarser
that \316\241
say
equivalently,
(or
\320\241
for
if
in
J
\316\241
\320\232.
\320\232
such
that
\320\260
in
J
thereexists
every
P7,
P7)

Definition

bounded

311

Partitions

1.
\320\246.

11.1.15. The partition {[1,2),{2},(2,3),[3,4]}


is finer
Both
are
finer
than
partitions
{[1,2], (2,4]} (why?).
{[1,4]},
that
there
which is the coarsest possible partition of [1,4].Note
of [1,4].
is no such thing as a \"finest\"
partition
(Why? recall all
Example

than

partitions are assumedto befinite.)


of different
P7

intervals,

of [2,5]

a partition

is

We

if

instance

for

not

do

\316\241
is

then we would

not

common

P#P'

refinement

p#p7 :=

of

\316\241
:=

of J.

and

coarser

Then

[3,4]}

P#P7

than

\316\241
and

than

finer

a bounded
the

define

We

the set
\342\202\254
P'}.

:= {[1,2],

and P'

is the set

I be a boundedinterval, and
I.
Then
P#P' is alsoa partition
of

partitions

finer

{[1,2],

\316\241
and

let

of

P7

I, and is

P7.

See Exercise 11.1.4.

Proof

Exercise11.1.1.

Prove

(b) in

implies

(Hint:

is non-empty,

in order to show that


(a)
consider the supremum

X.)

11.1.2.

explain

D
11.1.4.

Lemma

the case when

of

infimum

Exercise
and

\316\241
is

J be

Let

11.1.18.

Lemma

{[1,3),

partitions

Let

to be

\316\241
\316\232
and
\316\262

of [1,4].
(2,4]}be
(2,3),[3,4],0}(why?).
two

P7

\316\241
and

{KC\\

Let

11.1.17.

Example

and

that

say

Definition 11.1.16 (Common


refinement).
interval, and let \316\241and P7 be two partitions

both

of [1,4]

than P7.

or finer

be two

partitions

compare

a partition

Prove Corollary 11.1.6. (Hint: use Lemma


11.1.4,
intersection of two bounded
sets is automatically
of two connected sets is automatically
the intersection

why the

bounded, and why


connected.)

of the form / = (a, b) or


of I.
Let
/\317\207,..., In be a partition
[a, b)
Prove
that
one of the intervals Ij in this partition is of the form Ij = (c, b)
< b. (Hint:
or Ij = [c, b) for some a < \321\201
prove by contradiction. First
Exercise

11.1.3.

/ =

for

some

Let /

be a

bounded

real numbers

interval

a <b.

if Ij is not of the form


(c, b) or [c, b) for any
is
less
than
strictly
6.)
supij
11.1.4. Prove Lemma 11.1.18.
Exercise
show

that

< 6,
< \321\201
\316\261

then

11.

312
11.2

constant

describe

the class

now

can

We

Piece wise

integrate

is

a subset

I\\e

\302\260f
/

to

value of

f(x)

constant

\316\241
if

for

However,
value for f on

of J.
every

if

\316\225
is

possible

simultaneously

always

number

real

every

empty,

\321\201
is

\316\225

(why?).

We say that / is piecewise


\342\202\254
P, / is constant on J.

Let J be

let

with

constant

[1,6] ->

/ :

function

if 1 <

if

if 3

= 6
if \317\207

\316\241
be

to

respect

=
\317\207

<

constant with respect to

\317\207
<

R defined by
3

3
\317\207
<

the

Definition11.2.5
/ is piecewise

constant

on

functions

constant

(Piecewise

bounded interval,and let /

:J

be a

it is

{2},

II).

function.

exists a

respect

to

partition
P.

with

constant

\342\200\224\342\226\272
R

I if there

/ is piecewiseconstant with

{[1,2),

partition

{3},

{[1,3),

partition

{6}} of [1,6]. Note that it is alsopiecewise


to some other partitions as well;
for
instance,

wise constant with respectto the


(3,5), [5,6), {6},0}.

that

\316\225
while

which

/(*)

is piecewise

value;

3 on

is not

it

let

The

(3,6),

constant

a function /

set, then

(Piecewise constant functions I).


R be a function,
and
/ : J \342\200\224>

11.2.3

Example 11.2.4.

respect

the

\321\201
as

only oneconstant

equal

4.

interval,

partition

\316\225

restriction

exists a realnumber
to

refer

iff

\317\207
X.
\342\202\254
If

the

\316\225
if

words there

\317\207
\316\225.We
\342\202\254

all

\321\201
for

on

constant

a non-empty

\316\225
is

can have

on /

Definition
bounded

all

\321\201
for

function to always

equalling

is

f(x)

/ on E.

constant

for a

that

in other

constant,

Remark 11.2.2.If
is

\342\200\224>
R

\316\225
is

that

\321\201
such

of \"simple\"functions which we can

a real number \321\201


such
of X, we say that /

exists

there

functions

(Constant functions). Let X bea subsetof


We say that / is constant
be a function.

11.2.1
/ : X

let

and

integral

easily.

very

Definition

Riemann

The

(2,3),

piece{3},

Let J be a
We say that
\316\241
of

J such

constant

Piecewise

\320\246.2.

313

functions

11.2.6. The functionusedin the previous


constant
every
piecewise constant on [1,6]. Also,
J
interval
is
constant
bounded
automaticallypiecewise
Example

and let f

of I,
P.

:J

is alsopiecewise
constant

Then f

The spaceof
11.2.8.

Lemma

g : J

to P'.

D
functions

constant

piecewise

under

is closed

f+g,

functions

on

I.

be piecewise

:= max(f(x),g(x)). If g doesnot

max(/,g)(x)
on I

R
be a bounded interval, and let f : I \342\200\224>
constant functions on I. Then
the
f\342\200\224g,
g) and fg are also piecewiseconstant
max(/,
R is the function
Here of course max(/, g) : J \342\200\224>

Let I

\342\200\224>
R

functions

(i.e., g(x)

constant

function

0
\317\206

now

are

We

begin

with

all

for

\317\207
\342\202\254
I)

then

f/g

vanish

anywhere

is also a

piecewise

I.

on

See Exercise

Proof.

to

respect

operations:

algebraic

We

with

See Exercise 11.2.1.

proof

and

be a boundedinterval, let \316\241be a partition


be a function which is piecewiseconstant
than
is finer
Let P' be a partition of I which

\342\200\224>
R

to P.

respect

with

(why?).

Let I

11.2.7.

Lemma

is
example
on a
function

11.2.2.

ready

piecewise constant functions.


definition of an integralwith
respect

to integrate

a temporary

a partition.

Definition
11.2.9 (Piecewise constant integral I). Let J be a
R be a
of J. Let / : J \342\200\224>
bounded interval, let \316\241be a partition
to P.
Then we
function which is piecewiseconstant with
respect

define the piecewiseconstant


to the partition \316\241
the
by
P-c

where

for

each

J in

integral

f with

p.\321\201\316\223\317\201\316\271
f of

respect

formula

[ f-^cjlJl,
J[n
JeP

P, we let cj

bethe

constant

value

of

/ on

J.

11.

314

integral

11.2.10. This definition seemslikeit couldbeill-defined


number
J is empty then every
cj can be the constant

Remark
if

because

Riemann

The

in such cases \\J\\ is zero


value of / on J, but fortunately
and so
is
the choice of cj is irrelevant.The notation
rather
p.c.
/
\316\223\317\201\316\271
we shall
but
artificial,
only need it temporarily, en route to a more

is

well-defined

always

to

intuitively

if

be

to

/ is

\316\241
is

the

finite,

sum J2jep

cj\\J\\

divergent or infinite).

The piecewise constant integral corresponds


of area, given that the areaofa rectangle
notion
one's
the product of the lengths of the sides. (Of course,

11.2.11.

Remark

ought

since
is
never
(it
that

Note

definition.

useful

negative somewhere, then the

\"area\"

cj\\J\\

would

also

be

negative.)

11.2.12. Let /

Example

: [1,4]
4

{2 6
and

let

\316\241
:=

P'C'Ip

Alternatively,

{3}, (3,4]}.

{[1,3),

f=

10.

if we
f =

P'C'/p'

if

1 <

\317\207
<

if

=
\317\207

<

if 3

<
\317\207

function

Then
-\320\232\320\276(3,4]1\320\241\320\227,4]|
\320\230-\321\201\321\207:\320\267>|{3}|

c[^^)|[l53)|

be the

\342\200\224\342\226\272
R

\317\207
2 +

\317\207
0 +

\320\261\321\2051

let P7 := {[1,2),[2,3),{3},(3,4],
0}then
CIW)|[1,2)I +c[2,3)|[2,3)|

+ c{3}|{3}|

+ c(3,4]l(3,4]|
+

qz)|0|

+ 6xl + C0xO

+ 2xl+4xO

2xl

= 10.
This

on

what

example

you

partition

constant

with

that

suggests

respect

this integral

pick, so long as
that partition.

to

does not really depend

your function is piecewise


That is indeed true:

11.2.

constant

Piecewise

0f

(Piecewise constant integralis independent


a bounded interval, and let f : J \342\200\224>
R be a
\316\241
and
P' are partitions
I
that
such
f is
of

11.2.13

proposition

I be

Let

partition).

Suppose

function.

that

both

constant

piecewise

with

to

respect

Then p.c. Jjpj / = p.c.

with

\316\241
and

respect

to P'.

/.

Jjp/j

11.2.3.

See Exercise

Proof

315

functions

Becauseof this proposition,

make

now

can

we

the

following

definition:
constant
Let / be
Definition 11.2.14 (Piecewise
integral
II).
\342\200\224>
R be a piecewise constant
a bounded interval, and let / : J
J. We define the piecewise constant integralp.c. f
on
function
Jj

formula

the

by

p.c.

Ji/

/ :=

/,

p.c. /

J[P]

of J with respect to which


/
that
11.2.13
tellsus
that
constant. (Note
Proposition
choice of this partition is irrelevant.)
\316\241
is

where

partition

any

Example11.2.15.If / is

function

the

then p.c.

J[1A]

is piecewise
the

in Example

given

precise

11.2.12,

10.

constant
give some basic propertiesof the piecewise
laws will eventually be supercededby the
laws
for the Riemann
integral (Theorem 11.4.1).

We now

These

integral.
corresponding

Theorem 11.2.16
interval,

let

and

functions on I.
(a)

We

have

p.c.

: I

(Laws

of

\342\200\224>
R

/,(/

and

+ g) =

(b) For

any real numberc, we

(c)

have

(d)

We

If f(x)

p.c.

> 0

/7(/

g :

\342\200\224>
R

p.c./, /
have

then

p.c.

p.c.

- g) = p.c.
fjf-

el,
for all \317\207

be a

Let I

integration).

be piecewise

constant

Jj g.

Jj(cf)

p.c.

p.c. /7

bounded

JT

= c(p.c.
g.

/ > 0.

Jj /).

11.

316
(e)

// f(x)

(/)

f
\320\246

J be
F : J

Let
let

all

constant

\342\204\242
the

=
fj f

pc.
(g)

> g(x) for

in
for all \317\207

I,

(i.e.,I

J)f

a boundedinterval containingI
be the

\\ 0

J, and

p:c. JjF = p.c. L

J
K} is a partition of I into two intervals
\342\200\224>
:
R
and
:
J
~>
the
Then
\320\232
R
functions f\\j
f\\jc
and we have
piecewise constant on J and \320\232respectively,

f=p.c.
Ji

/ /|j + p.c.Jk/

f\\K.

Jj

Proof. See Exercise11.2.4.


our

concludes

now

/.

that {J,

Suppose
and K.

This

\320\260\320\277(\321

ifx^I
on

constant

piecewise

function

\342\200\224>
R

p.c.

We

then

c\\I\\.

ThenF is

are

= c

integral

p.c. Jjg.

p.c. Jjf>

f(x)

function

*(X)'

(h)

then

\317\207
el,

Riemann

The

turn

to the

of piecewise constant functions.


to integrate bounded
of
how
question
integration

functions.

11.2.1.

Exercise

Prove

Exercise 11.2.2.Prove
to

11.2.7

make

Lemma 11.2.7.
11.2.8.

Lemma

/ and g

(Hint:

piecewise constant

use Lemmas
with

respect

11.1.18and
to

the

same

partition of /.)

Exercise11.2.3.

Prove

11.1.13to show
Exercise

11.2.4.

Prove

of the theorem to
the

hint

prove

to Exercise

11.2.13.

Proposition

both

that

integrals

Theorem
some

11.2.2.)

of

are equal

first use Theorem


(Hint:
to p.c.Jjp^p'i /\302\267)

11.2.16. (Hint: you can use earlier parts


the later parts of the theorem. Seealso

317

integrals

Riemann

lower

and

Upper
\320\246.\320\227

Riemann

lower

and

Upper
\320\246.\320\227-

integrals

bounded function defined on a bounded


to
define
the Riemannintegral JT f. To do
interval
to
the
first
need
define
notion
of upper and lower Riemann
this we
let

flow

: I

be a

\342\200\224*
R

to the
Riemann
J f. These notionsare related
the same way that the lim sup and lim inf of a

and

fjf

integrals

want

We

J.

integral in much

sequenceare relatedto the limit of that sequence.


11.3.1 (Majorization of functions). Let / : J \342\200\224\342\226\272
R
Definition
\342\200\224>
I
/
R.
that
on
if we
have
:
We
f
say
g
majorizes
and 5
g(x) >
<
I
\320\266
on
if
for
all
and
that
minorizes
\342\202\254
for
J,
g
f
g(x)
f(x)
f(x)
e J.

all

of the Riemannintegralis to try to integrate


a
or minorizing that function by a piecewise
first
majorizing
we already know how to integrate).
function
(which

The idea
function

by

constant

Definition11.3.2
/

Riemann

lower

and

(Upper

be

integrals).

We define

:= inf{p.c.

/ /
and

the

/ :=

lower

/ g : g isa p.\321\201 function

Riemann

a crude but

useful

J which

J.

majorizes

/}

minorizes

/}

by the formula

J f

/ g : g is a p.c.function

sup{p.c.

We give

integral

on

Let /

function defined_on a bounded interval


the upper Riemann integral JTf by the formula
a bounded

\342\200\224\342\226\272
R

J which

on

on

bound

the

lower and

upper

integral:

Lemma 11.3.3.
I

interval

f(x)

<

I.
\317\207
\321\221

all

numbers

both
(i.e.,

they

\342\200\224>
R
by

Then

-M\\I\\ <
In particular,

is bounded

which

\316\234
for

Let

the lower
are not

some

be a
real

function
number

on a bounded
M,

i.e.,

<
\342\200\224M

we have

f <J f <

M\\I\\.

and upper Riemannintegralsare real


infinite).

11.

318
The

stant,

hence

p-c.Jjg =
similar

f <

of

definition

by

\342\200\224

gives

Jjf.

defined

\342\200\224\342\226\272
R

constant,

piecewise
M\\I\\

argument

that

g : I

function

Proof.

M\\I\\

Let g be any

jorizing /, and let h be any


ing /. Then g majorizesh}
h, we

in

suprema

know

now

We

<

as the
then we can

as large

least

match,

11.3.4

Definition

p.c. Jjh

J f < p.c.

JT

that

\316\234is

\321\201\320\276\321\2

function

<
\320\224

ma-

function minoriz< p.c. Jj g.


Taking

infima

Taking

g.

in g)
\342\226\241

Riemann integral is always

lower Riemannintegral.If the


define the Riemann
integral:
(Riemann

desired.

and

at

integrals

\342\200\224>
R

be

= /7/, then we
/ /

If

J.

two

:J

integral). Let /

function on a boundedinterval
/ is Riemann integrable on I

bounded
say

hence

the upper

that

constant

piecewise

as

integral

and majorizes /; thus f.f


the upper Riemann integral.
J f. Finally, we have to show
piecewise

and

/7<7,

by g(x)

constant

obtain that

obtain J f

we thus

<

Riemann

The

define

/\"-//-7/

If the

upper and lower


Riemann

/ is not

Riemann

are unequal,

integrals

lim

the

or equal

than

lim inf, and limit

sup,

to the lim

of the

the

definition

inf,

but

they

sequence.

relationship

greater

are both equal to


above may

differ

calculus
you may have encountered in your
on Riemann sums. However, the two
definitions

based

equivalent; this is the purposeof the next


Remark
11.3.6. Note that we do not consider
to be

functions

to

functions

is

evaluate

such

Riemann

be
known

integrable;

was

that

only equal when

are

in this case they


The definition given

and

converges,

sequence
limit

the

to

of a sequencean

6.4.12(f); the lim sup is always

in Proposition

established

that

integrable.

Remark 11.3.5. Comparethis definition


between

we say

an integral

the
the

from

courses,
turn

out

section.
unbounded

involving

such

improper integral. It is possibleto still


using more sophisticated
integration methods

as an

integrals

(such as the Lebesgue

integral);

we

shall

do this

in Chapter

19.

Upper
\320\246.\320\227-

lower

and

Riemann

319

integrals

The Riemannintegralis consistentwith


constant
piecewise

kemma 11.3.7. Let f


on

interval

a bounded

(and

the

supercedes)

integral:
I.

:J

be a

\342\200\224>
R

Then f

piecewise constant

is Riemannintegrable,

and

function
=

Jj f

p-c-fif-

Proof. See Exercise11.3.3.


Because

11.3.8.

Remark

of this

lemma, we will

refer

not

to the

and
use the Riemann
just
piecewise constant integral p.c. Jj again,
this
is
itself
integral
superceded
by
integral Jj throughout (until
one special
case
integral in Chapter 19). We observe
the Lebesgue
= 0
11.3.7:
if J is a point or the empty
then
of Lemma
set,
Jj f
\342\200\224>
:
functions
J
R.
all
such
that
functions
are
all
for
/
(Note
constant.)

automatically

We

have

is Riemann
and

general,
this

just

shortly.

defined

have seen in

shown that every piecewise constant function


However, the Riemann integral is more
integrable.
can integrate
a wider class of functions; we shall see
For now,
we connect the Riemann integral we have
to the concept of a Riemann sum,
which
you
may
just

other treatments of the Riemann

Definition

11.3.9

function on

a bounded interval

We
sum

define

L(/,

the

P)

(Riemann

upper

Riemann

R be
/ : J \342\200\224>

sums). Let
J,

let

and

sum [/(/,

integral.

\316\241
be

a bounded
of J.

a partition

P) and the lowerRiemann

by

C/(/,P):=

\302\243

JeP:J^0

(sup/(z))|J|

xeJ

and

L(/,P):=

\316\243

(mf/(*))|J|.

JePiJ^Q

Remark

11.3.10. The restrictionJ

quantities mixej f(x) and


infinite) if J is empty.

0
\317\206

supxeJf(x)

are

is required
infinite

because
(or

the

negative

11.

320
We

Riemann

these Riemann sums to the

now connect

Riemann

The

integral

and

upper

lower

integral.

11.3.11. Let f : J

Lemma

and let

be

\342\200\224\342\226\272
\316\261
R
bounded

g be a function

bounded

interval

which is

piecewise constant with

I,

to

respect

which

some

function

on

majorizes

partition

\316\241

\320\260

and
ofIt

Then

P-cJg>U(f,P).

Similarly,

if h is a function

constant with

to P;

respect

minorizes

which

and is

piecewise

then

p.c.Jh<L{f,Y).

Proof. See Exercise11.3.4.


bounded

Let f
I. Then

11.3.12.

Proposition

interval

I f=

\342\226\241

:I

be a

\342\200\224>
R

bounded function

: \316\241
is

a partition

of 1}

f = sup{L(/, \316\241): \316\241is

a partition

of 1}

inf

{{/(/,

\316\241)

on

and

Proof. See Exercise11.3.5.

11.3.1.
R be functions.
Exercise
Let / : / \342\200\224>
and h : I \342\200\224>
R,
R, g : I \342\200\224>
then
Show that if / majorizes g and g majorizes \316\233,
/ majorizes h. Show
if / and g majorize each other, then they
that
must
be equal.
Exercise 11.3.2. Let / : / \342\200\224>
and h : I \342\200\224>
R be functions.
R,
R, g : I \342\200\224>
If / majorizes g, is it true that f + h majorizes g + h? Is it true that
\302\267
h majorizes
If \321\201
h?
is a real number,
is it true that cf majorizes
/
g \302\267

eg?

Exercise

11.3.3. Prove Lemma 11.3.7.

Exercise11.3.4.

Prove

Exercise

11.3.5. Prove

Lemma

11.3.11.

Proposition 11.3.12.(Hint: you

11.3.11, even though this Lemma will

only

do half

will

need

of the

job.)

Lemma

4}^. Basic propertiesof the


Basic

\321\206.4

just

as

did

we

321

integral

Riemann

of the

properties

the basic laws


^ill eventually

Riemann

integral

with limits, series, and derivatives,we now give


for manipulating the Riemann
These
laws
integral.
be superceded by the corresponding
laws for the

Lebesgue integral (Proposition19.3.3).

Theorem 11.4.1

Riemann

f :J

let

and

interval,

bounded

of

(Laws

Let I

integration).

g :

and

\342\200\224\342\226\272
R

be a

be Riemann

\342\200\224\342\226\272
R

integrable functions on J.
(a)

(b)

The

+ JJ9-

For

real

any

(c)

The

9)

(d) If

Jj(cf)

Riemann

f(x) > 0 for allx\342\202\254l, then

(/) If f

integrable, and we

function cf is
c(fj /).

is
=

\320\233/

for

all

constant

the

/7

then

\317\207
el,

f(x)

function

/ >

have

//(/\342\200\224

0.
Jjg.

fjf>
=

\321\201
for

all

\317\207
in

I,

then

J),

and

\321\201|/|.

J be a boundedinterval containingI
R be the function
let F : J \342\200\224\342\226\272

Let

Then F is

Riemann

integrable

(i.e.,I

ifxeI
on

J,

and JjF

fjf\302\267

two
intervals
J
that {J, K} is a partition of I into
\342\200\224*
R and f\\x : \320\232
R
Then the functionsf\\j : J \342\200\224\342\226\272
and
Riemann
we
integrable on J and \320\232respectively,

Suppose

and K.
are

Jj(f+

Riemann

and we

integrable,

F(x)-=(f{x)

(h)

have

Jif-!i9.

(e) Iff(x) > g(x)

(g)

have

is
f\342\200\224g

function

c, the
=

number
we

and

integrable,

is Riemann

f+g

function

=
9)
IJf

have

[f=

Jl

ff\\j+

JJ

[
JK

f\\K-

11. The Riemannintegral

322

11.4.2.

Remark
/

11.4.1

Theorem
Riemann

give

often

We

defined on

is really

scalar

11.4.1.

Exercise

See

Proof.

integrable
cf

multiple

further

some

abbreviate

Jj

f\\

a larger domainthan

as

Jj /,

even

though

J.

just

the sum or differenceof any two


functions
is Riemann integrable, as is any
of a Riemann integrable function /. We now
ways to create Riemannintegrable
functions.
asserts that

(Max and

11.4.3

Theorem

min preserveintegrability).

bounded interval, and let f : J

and

\342\200\224\302\273
R

g :J

Let I be a
a
be
Riemann
\342\200\224\342\226\272
:
I
R
g)
and

\342\200\224>
R

integrable function. Then the functions


max(/,
: I -> R defined
:= max(f(x),g(x))
by max(f,g)(x)
min(/,flf)
:= mm(f(x),g(x)) are alsoRiemann
integrable.
mm(f,g)(x)

and

the claim for \321\202\320\260\321\205(/,\320\267),


the
case
of
g) being similar. First notethat since/ and g are bounded,
min(/,
then max(/, g) is also bounded.
= J /, thereexistsa piecewise
Let
\316\265
> 0. Since
constant
Jj f
\342\200\224>
:
/
R
which
on
I
suchthat
function
minorizes
f
/
Proof.

we

Similarly
minorizes

shall

We

on

and we can
respectively

and

can

just prove

find

J such

find

a piecewise

constant

g_

which

\342\200\224>
R

that

piecewise

on J suchthat

functions

/,

~g

which

majorize

/,

Basic

of the

properties

1\316\2204\302\267

In particular,if

Riemann

323

integral

the function

denotes

\342\200\224*
R

+ &-2)

h:=G-f)
have

ye

4\316\265.

S,hi
On

other

the

hand,

is a piecewise
constant

max(/, g)

/ (why?) which minorizes max(/,g) (why?),


a piecewise constant functionon
is similarly

function

while

on

max(/,^)

which

majorizes

max(/,3)\302\267 Thus

and

< / max(/,j)<

< / max(/,s)

/ max(/,g)

max(/,ff),

so

0 <

max(/,g) - /

But we

max(/,g) < / max(/,

#)

max(/,g).

have

1{x) = f{x) + (J -

/)(*)

<

fix)

+ hix)

<

g(x)

+ h(x)

and similarly

g(x) = g(x) + (g-

g)(x)

and thus
+ h(x).
max(f(x),g(x)) < \321\202\320\260\321\205(/(\320\266),\320\267(\320\266))

Inserting

this into
0 <

To summarize,

the previousinequality,

max(/, g) -

max(/,

we

g)

obtain

<

h<4e.

we have shown that

0< /

max(/,

g)

max(/, g)

<

4\316\265

for every

\316\265.
Since

g) does

\342\200\224

max(/,

integral

not depend on

see that

thus

we
\316\265,

g)

max(/,

Jj

Riemann

The

11.

324

that max(/, g) is Riemann

and hence

\321\202\320\260\321\205(/,\320\267)\321\202\320\260\321\205(/,\320\267)

integrable.

11.4.4 (Absolute values preserveRiemann


integrabil\342\200\224>
R is a Riemann
I be a bounded interval If f

Corollary
Let
ity).

:J

the
integrable function, then the positive part /+ := max(/,
0) and
on I.
integrable
negative part /_ := min(/, 0) are also Riemann
is also Riemann
Also, the absolutevalue \\f\\ = /+ \342\200\224
/_
integrable
on I.

Theorem 11.4.5 (Products


a boundedinterval

Riemann

preserve

I be

///:/\342\200\224>

then

integrable,

This one

Proof.
g

the

g+

+\320\267_

into

functions

also

fg:I\342\200\224>'R.is

and

/_,

g_
\320\267+,

f9 = f+9+ +

We

negative

Let

are Riemann

\342\200\224>
R

integrable.

split

parts; by

Riemann

are

integrability).

g :J

and

Riemann

is a little trickier.
positive

/+,

= /+

+ /_ and
Corollary 11.4.4,
Since

integrable.

f+9- + /-5+

f-9-

to show that the functions


/-3+,
f~g~
/+5+,/+5_,
are individually Riemann integrable. We will just show this for
/+5+; the other three are similar.
Since
M*i >
M\\,
/+ and g+ are bounded and positive,thereare
0 such that
it suffices

then

0 < f+(x)
for

all

let

Now

\317\207
I.
\342\202\254

Theorem
norizing

/+ on J,

11.4.3,
/+

on J,

<

\316\265
>

0 be

we can find

and a

such that

and

Mi

0 <

g+(x)

arbitrary.

a piecewise

< M2

Then, as in the proofof


constant function/+ mi-

piecewiseconstantfunction

/+

majorizing

Basic

\321\206.]^

of the

properties

Riemann

325

integral

and

/\320\260\320\247^-'-

/+ may be negative at places, but

that

Note

replacing /+
has

still

and

loss of

without
e I\317\207

Similarly

or equalto fjf+\342\200\224e

greater than

integral

still minorizes

this

since

max(/+,0),

by

can

we

that
generality we may assume
we may assume that /+(\320\266) <

f+(x)

fix this

/+

So

(why?).

> 0

for
all
\316\234\316\271

by

(why?)

for all
\317\207
G

J;

thus

0<f\302\261(x)<U(x)<f+(x)<M1
for all

\317\207
e

I.

Similar

norizing

allows us

reasoning
and

<7+,

find

constant

piecewise

#+, such

majorizing

5+

to

5+ mi-

that

5++\316\265
\320\233
/*\"/'

\320\233

and
9+-\316\265,

9\302\261>

and

0 <
all

for

\317\207
e

while

constant

is piecewise

f+g+

we

f+g+

<Mi<3+\317\207
e

I,

0<Jf+9+-J

f+g+ <

j7^gT-f+g+.

have

1+ (x)gT(x)-f\302\261
(x)g\302\261(x)

for all

piecewise constant and minorizesf+g+,


and majorizes f+g+- Thus

f+g+ is

0<
However,

g+(x) < 9T(x) < M2

I.

that

Notice

9+(x) <

and

7+(x)(9+-g\302\261)(x)+9\302\261(x)(7+-f\302\261(x))

g+)(x) +

\320\2342(\320\271

f+(x))

thus

f+9+ < Mi

+
J(g^-g\302\261)

M2J(u-f\302\261)

The Riemann

11.

326

integral

+ \316\2342(2\316\265).
<\316\234\316\271(2\316\265)
Again,

since

Riemann

integrable,

/_5+,

f~g~

\316\265
was

before.

as

Riemann

are

that fg is

arbitrary,

Riemann

we can

conclude that

Similar

argument

/+5+ is

show that

f+g_

combining them we obtain

integrable;

integrable.

\342\226\241

Prove Theorem 11.4.1. (Hint: you may find Theorem


be useful. For part (b): First do the casec> 0. Then do the
= \342\200\224
= 0
1 and
\321\201
case \321\201
Using these cases, deducethe caseof
separately.
to prove later ones.)
\321\201
theorem
< 0. You can use earlier parts of the
6 be real numbers, and let / : [a,6] \342\200\224\342\226\27
Exercise 11.4.2.Let \316\261
<
R
be a continuous,
\342\202\254
non-negative function (so f(x) > 0 for all \317\207
[\316\261,
b]).
= 0 for all \317\207
L 6i / = 0. Show that
that
\342\202\254
Suppose
[\316\261,
f(x)
b].
(Hint:
11.4.1.

Exercise

11.2.16 to

by contradiction.)

argue

Exercise 11.4.3. Let I be a bounded


be
integrable function, and let \316\241

11.4.4.

proofs, give

and

11.4.3

Theorem

Without

a short

the

all

repeating

as to

explanation
Theorem
\320\2241.4.5

Riemann
have

We

so

far,

other

already

why

follow

but

than

we have

a lot

We

begin

with

Theorem 11.5.1.
function
integrable.

which

remaining

automatically
11.4.1
we

Let

constant ones.

the uniformly
I

is uniformly

above

cases of
from the

cases

know that if / is

functions

about Riemann integrablefunctions

showing that a large classof useful


integrable.

in the

computations
the

not yet actually producedany

the piecewise

a Riemann

Show that

of continuous

integrability
said

\342\200\224>
R be

JePJj

presented in the text. (Hint: from Theorem


Riemann integrable, then so is \342\200\224/.)
11.5

/ :/

of /.

a partition

Ji

Exercise

let

interval,

be a

functions

Now

functions

such

this

we

rectify

are

Riemann

by

continuous functions.

bounded interval,

and let f

continuous on I. Thenf

is

Riemann

be

11.5.

Riemann

proof

From

of continuous

integrability

that

/ is

then

the

theorem

we have to show that J f = JjfJ is

If

or the

a point

set

empty

let us assumethat / isoneof the

Let

\\x

\342\200\224
<

y\\

that
N>0
Notethat
such

N intervals Ji,..., J/y, each


to treat each of the cases[a,b],

has

One

By Proposition 11.3.12,we

differently.)

slightly

[a?b)

(a>b]>

(a>b)>

is trivial, so
b], (a, b), (a, b],

J into

partition
(How?

(b\342\200\224a)/N.

Now

< 5.

a)/N

can

we

of length

(b

[a,

327

uniform continuity, there exists a


\316\265
whenever
e I are such that
x,y
there
exists an integer
principle,

Archimedean

the

\316\264.
By

intervals

four

a <b.

or [a, b) for some real numbers


\316\265
> 0 be arbitrary.
By
that \\f(x) \342\200\224
<
> 0 such
\316\264
f(y)\\

bounded.

see

we

9.9\320\2335

Proposition

functions

have

thus

~f

/<Dsup'(*))!J*l
J/ \316\271
xeJk
k=l
and

/
in

so

particular

I f< f>up f(x)

f-

we have \\f(x)

However,
\342\200\224
(b

/>\316\243(\317\211/(*\316\270)\316\220\316\233\316\220

a)/N

<

In
\316\264.

suprema

in
sup

f(y)\\

for
\316\265

for
f{y) + \316\265

all

/(*))|\316\233|.

i,j/G

all

G
\320\266,
\321\203

Jfc.

obtain

we
\320\266,

f(x)

<

mf

we have

particular

/0*0 <
Taking

\342\200\224

< f(y)

for
\316\265

all

\320\265
Jk,
\321\203

xeJk

and

then

taking infima in
sup
xeJk

we
\321\203

f(x)

obtain

< inf

f(y) + \316\265.

y\302\243Jk

Jfc, since

|Jfc.|=

11. The Riemann integral

328

Insertingthis

our

into

bound

J//-//<
\316\271

but by

we obtain

inequality,

previous

f>|Jfc|,

Li

t^i

Theorem 11.1.13we thus have

f<e(b-a).

J f-J
But

\316\265
>

J f

cannot

Riemann

was

be

while (b

arbitrary,

we thus

integrability

a)

fixed.

is

the

11.3.3 and

Lemma

By

positive.

\342\200\224

have that / is

Riemann

Thus

JTf

definition

of

integrable.

11.5.1 with Theorem

Theorem

Combining

9.9.16, we thus

obtain

Let

11.5.2.

Corollary

be continuous.

\342\200\224>
R

[a, b]

Note

[a,b] be a
Then f is

closed interval,
let
Riemannintegrable.
and

is not true if [a,b] is replaced by


since
it is not even
then
any
guaranteed
that
continuous
functions are bounded. For instance,the
\342\200\224>
function /
: (0,1)
R defined
by f(x) := 1/x is continuousbut not
Riemann
is both
However, if we assume that a function
integrable.
and bounded, we can recoverRiemann
continuous
integrability:
this

that

other

sort

Corollary

of interval,

boundedinterval,and
continuous and bounded. Thenf is Riemann
11.5.3.

Proposition
R

be both

on

I.

Let I be a

set
Proof. If J is a point or an empty
J is a closed interval the claim follows

let us

assume that J is of the

form

a<b.
We

have

I.

Now

\317\207
e

a bound

let 0

<

\316\234for
\316\265
<

(b-a)/2

/,

then

(a,b],

from

the claim
Corollary

(a,b),

let

: I

\342\200\224\342\2

integrable

is trivial;
11.5.2.

or [a, b)

if

So

for some

< \316\234for
\316\234<
so that \342\200\224
f(x)
be a small number. The function

all

Riemann

\320\246.5.

of continuous

integrability

f when restricted to the

[a +

interval

329

functions

is
b \342\200\224
\316\265,
\316\265]

and

continuous,

we
hence Riemann integrable by Corollary11.5.2.In particular,
\342\200\224
\342\200\224>
a
constant
:
find
function
h
b
R
which
+
can
piecewise
[a \316\265, \316\265]

[a +

/ on

majorizes

b \342\200\224
such
\316\265,
\316\265]

that

h<

/
h :

Define

\342\200\224>
R

by

h(

^(\320\266)
\342\200\224
f

\\

nW\342\200\224\\

piecewise
we have

11.2.16

lh

= eM+

Jl

if

\320\266
G

on J and majorizes/;

+ eM<

\316\231

(2\316\234

Theorem

1)\316\265.

we have

A similar

by

J[a+e,b-e]

J[a+eJ>-e]

In particular

+ \316\265,
[\316\261
\320\254-\316\265]

+ \320\265,\320\254-\320\265]
]\302\243\321\205\320\265\320\223\\[\320\260

\320\274

constant

h is

Clearly

f + e.

[
J[a+ejb-e]

J[a+e,b-e]

//</
I

/+

(2\316\234

1)\316\265.

J[a+s,b-e]

argument gives

J'[o+e,b-e]
[\316\270-

/ -

(2M
+

l)e

hence

and

f
But

f-

This
already;

class
functions.

gives
the

2)\316\265.

and so we can argue as in


conclude Riemannintegrability.

\316\265
is arbitrary,

11.5.1 to

/<(4\316\234

proof

of Theorem

class of Riemann integrablefunctions


functions.
But we can expandthis
continuous
more, to include the bounded piecewise
a large

bounded

a little

the

continuous

The Riemann

11.

330

integral

and
let f : I ~+
Definition 11.5.4. Let J bea bounded
interval,
R.
We say that
exists
/ is piecewise continuouson I iff there
a
on
J
is
for
\316\241
of
that
J
all
such
continuous
J
\316\225
P.
partition
/| j
R defined
/ : [1,3]\342\200\224>

The function

11.5.5.

Example

F(x)

( x2
:= { 7

continuous on [1,3],but it
it
(since is continuouswhen restricted
intervals

three

those

be both

<

<
\317\207

is piecewise
to

[1,2)

on [1,3]

continuous
or {2} or

(2,3], and

[1,3]).

partition

11.5.6. Let I

Proposition

\316\232 \317\207
<

=
\317\207

if 2

x3
is not

if
if

by

be

bounded

interval,

and let

piecewise continuous and bounded. Thenf

is

f :/

\342\200\224>

Riemann

integrable.

See Exercise

Proof.

Exercise
and

Prove

11.5.1.

11.5.1.

Proposition

\320\236

11.5.6. (Hint:

useTheorem11.4.1(a)

(h).)

11.6

Riemann

In additionto

integrability
piecewise

continuous

functions

of monotone
another

functions,

of functions is Riemann integrable,namely


of this:
functions.
We
two instances
give

the

wide class

monotone

Proposition 11.6.1. Let [a,b] be a closed and bounded interval


and let f : [a,b] \342\200\224>
R be a monotone function. Then f is Riemann
integrable

on

[a, b].

11.6.2. Prom Exercise9.8.5we know that


Remark
monotone functions which are not piecewise
continuous,
is not subsumed by Proposition11.5.6.
proposition

exist

there
so

this

Riemann

\320\246.6.

proof

331

functions

take / to be monotone
(instead
decreasing). Prom Exercise 9.8.1
is
bounded.
Now
that
let N > 0 be an integer,
and
/
:
[a, b] into N half-open intervals {[\316\261+^jf,
a+^(j+l))
we may

of generality

loss

Without

of monotone

increasing
we

of monotone

integrability

know

partition

\342\200\224

Q<j<N

(b

length

l}oi

\342\200\224

sup

(the point {b}clearly


monotoneincreasing,

contribution).

is

Since /

have

thus

N~l

\"7

/(*))^r.
a zero

only

giving

we

{b}.

have

we

7/<\316\243(

the point

with

together

a)/iV,

Then by Proposition11.3.12

\320\270

\321\214

J l
//*!>+\342\200\224\321\214+1\302\273\342\200\224\302\267

j=Q

Similarly we have

Thus

N~l
\\~^

* ^

\320\273/

\320\270
\342\200\224
\342\200\224 \320\270
\320\276 \320\260
..\320\276 \320\260

have

we

+
\316\257
\316\243\342\204\226

//

J/

Using telescoping

~
/

jj

^\317\213

+1\302\273

series (Lemma 7.2.15)we

(/(\302\253 +

/\320\241+

^\320\233)^\302\267
have

thus

/(\316\261+
-\302\253\320\263(JV))\"

^\316\275-0))\316\212\\\316\223

(\342\204\226-f(a))b-j^.
as in the proof of
was arbitrary, so we can conclude
Theorem 11.5.1
that
/ is Riemann integrable.

But N

Corollary 11.6.3.
be both

I.

monotone

Let

be a

bounded interval,

and bounded.

Then f is Riemann

and let f : J \342\200\224>


R
integrable

on

The Riemann

11.

332
11.6.1.

Exercise

See

Proof.

We now give the famous integraltest for


monotone

convergence of

monotone
all
only if

>
\317\207

decreasing

Proof. See

Let

Proof.

See Exercise

a real

be
\317\201

and

Exercise11.6.2.Formulate
then

Riemann

the

11.6.4.
break

test

11.6.5.

Exercise

Corollary

11.6.3.

(Hint:

11.5.3.)

a reasonable

of a piecewisemonotone
functions
piecewisemonotone

notion

11.6.4.

Proposition

sum

^\320\265
sum
/(n)>
\316\243\316\267=\316\271

what is

the

/(\316\267)>anc^
\316\243\316\267=\316\277

the

(Hint:

Give examples to show that


down if / is not assumed to

Use Proposition

A non-Riemann

11.7

functions

to prove

integral

\302\253^?)

Exercise

We

\342\204\242?

integrable.

between
relationship

integral

Then \302\243)nLi
< 1.
when \317\201

show that all bounded

Exercise 11.6.3. Prove


J[otN]

diverges

11.6.1

proof of Proposition

and

number.

11.6.5.

Exercise11.6.1.UseProposition

are

if and

> 1
\317\201

when

absolutely

function,

\342\204\242
convergent

Exercise 11.6.3.

converges

adapt

X)^Lo/(n)

is finite.

11.6.5.

Corollary

the sum

Then

0).

sup N>0 jT[0N]

the

determining

series.

11.6.4 (Integral test). Let f : [0,oo)\342\200\224>


R
be a
decreasing function which is non-negative(i.e., /(\317\207) >

Proposition
0 for

integral

shown

have

are

which

bounded

functions

that
Riemann
which

11.6.4to

integrable

both

directions

be monotone

prove

Corollary

of the
decreasing.

11.6.5.

function

are large classes of bounded


integrable.
Unfortunately, there do exist
are not Riemann
integrable:

there

jj.7.

non-Riemann

11.7.1. Let f :

proposition

333

function

integrable

[0,1]

be the

\342\200\224\342\226\272
R

discontinuous

function
\320\237\320\245)\302\267

\\

in Example

considered

ifx^Q

9.3.21. Thenf

is

but

bounded

not

Riemann

that

it is

integrable.

is clearthat / is bounded,solet us
Riemann integrable.
Proof. It

Let
if

J is

\316\241
be

not a

point or the empty

xeJ

that

= 1
we have

In particular

5.4.14).

Proposition

observe

\342\202\254
P,

not

then

set,

sup f(x)

(by

For any J

of [0,1].

partition

any

show

(eup/(*))|J|

\\J\\.

xeJ

(Note

is also

this

true when J is a point,sincebothsides


are
that

zero.)

we see

In particular

C/(/,P)=

|J|

\316\243

[0,1] =

JeP:J^0
by

Theorem

anything

note

11.1.13;
to

the

total

that

length.

the empty
In particular

Proposition 11.3.12.
A

similar

argument

gives

that

inf/(z)
xGJ
for

all

J (other

than points
L(/fP)=

set doesnot contribute


we have J\\Q ^f = 1, by

= 0

or the empty

set),
0 =

\316\243

and

so

0.

JeP:J^0

we have J
f = 0, by Proposition
the upper and lower Riemannintegralsdonot
function is not Riemannintegrable.
In particular

11.3.12.
match,

and

Thus
so this
D

The

11.

334

Riemann

integral

11.7.2. As you can see, it is only


rather
\"artificial\"
which are not Riemann integrable.Because
of this,
the Riemann
integral is good enoughfor a large majority
or improve
of cases. There are ways
to
this integral
generalize
Remark

functions

bounded

which we will define


though. Oneof theseis the Lebesgue
integral,
in Chapter 19. Another is the Riemann-Stieltjes
L
integral
fda
: J \342\200\224*
R is a monotone
where
\316\261
increasing function, which
we
in the

define

next section.

The Riemann-Stieltjes integral

11.8

: J \342\200\224\342\226\272
R be a monotone
interval, let \316\261
\342\200\224>
:
J
let
R
be
a
function.
Then there is
function,
increasing
/
a generalization
of the Riemann integral, known as the Riemannis defined just like the Riemann
This integral
Stieltjes
integral
with
but
one twist: instead of taking the length
integral,
\\J\\ of
intervals
J, we take the \316\261-length a[J], defined as follows. If J is
a point
or the empty set, then ct[J] := 0. If J is an interval
of the
\342\200\224
:= a(b)
form [a,b], (a,6), (a,b], or [a,b),then
Note
a[J]
a(a).
that in the special case where\316\261
:=
is the
function
identity
a(x)
\317\207,
is just the same as \\J\\.
then
for more
However,
general
a[J]
monotone functions a, the \316\261-length a [J] is a different quantity
from
of the
it turns out one can still do much
Nevertheless,
| J|.
above theory, but replacing \\J\\ by a[J]
throughout.
Let

be

a bounded

and

11.8.1

J be a bounded interval, and


Let
defined on some domainX which
J. Then we define the \316\261-length a[I]
contains
of J as follows. If /
=
of
is a point or the empty
0. If J is an interval
we set a[I]
set,
form [a,b], [a,b), (a,b]> or (a,b) for some
the
b > a, then we set

Definition

let

: X
\316\261

a[J] =
Example

(\316\261-length).

a function

be

\342\200\224>
R

\342\200\224
a(b)

a(a).

Let

11.8.2.

Then a[[2,3]]= a(3)Meanwhile

a[{2}]

Example 11.8.3.

Let

: R
\316\261

a(2)

0 and
a

be

= 5,

9-4

a[0] = 0.

: R

the function

\342\200\224\342\226\272
R

be the

\342\200\224\342\226\272
R

a(x) := x2.

while a[(-3, -2)] =


identity function

-5.

a(x) :=

The

0. Then

a[I] =

\\I\\

length

is a

of

notion

of

One
integral

in

and

intervals / (why?) Thus the


the notionof \316\261-length.

a(sc)|*fja instead of \320\260[[\320\260,\320\254]].


for the theory of the Riemann
which concerned length and partitions,
was
that |/| =
\316\241
a
|J| whenever

afa or

11.1.13,
showed

particular

of /. We

partition

special case of

key theorems

the

Theorem

was

bounded

all

for

write

sometimes

We

335

integral

Riemann-Stieltjes

\321\206\302\243.

X)jGp

now generalize this slightly.

Let I be a boundedinterval, let a


somedomainX which contains
on
junction defined
11.8.4.

lemma

be

I. Then we

of

a partition

I,

be a

\342\200\224>
R

let

and

\316\241

have

a[I]

:X

\302\243

a[J].

JeP

Proof
We

11.8.1.

See Exercise
can

of Definition 11.2.9.

a generalization

define

now

Definition11.8.5(P.c.

Let

integral).

Riemann-Stieltjes

J be

of J. Let a : X \342\200\224>
R
bounded interval, and let \316\241be a partition
which
defined
on
some
domain
X
and
let
contains
be a function
J,
\342\200\224\342\226\272
R
be
a
which
is
function
with
J
:
constant
piecewise
respect
/
to P. Then we define

f
cJalJ]
'PI da:=J2

p.c. /

jeP

where

cj

is the

constant

/ :

Let

11.8.6.

Example

t(

value of / on J.

\\

R be the function
[1,3] \342\200\224\342\226\272

\320\237\320\245)~\\2

let

: R

partition

when

\320\266\320\265
[1,2)

when

\320\266
\342\202\254[2,3],

function a(x)
[2,3]}. Then

the

be

\342\200\224\342\226\272
R

:=
\316\241

{[1,2),

p.c.
= 4(e*(2)

fda =

- a(l)) + 2(e*(3)

c[1)2)\302\243*[[l,

2)]

e*(2))

:=

\320\2662,and

let

\316\241
be

+ c[2j3]a[[2,3]]

=4x3

+ 2x5 =

22.

the

11.

336

integral

Let a : R \342\200\224\342\226\272
R be the identity function
a(x) :^
interval J, any partition \316\241
of J, and
bounded
any

11.8.7.

Example

x.

Riemann

The

for

Then

any

/ that is piecewiseconstant with

function

p.c. /[P] / da = p.c./[P]/

to

respect

we have

P,

(why?).

analogue of Proposition 11.2.13by


all the
p.c. \316\223\317\201\316\271
/ in the proposition with p.c. Jfp
integrals
replacing
f
We can thus definep.c.JT f da for any piecewise
(Exercise
11.8.2).
R and any a : X \342\200\224>
R defined
constant function / : / \342\200\224>
on a
to before,
domain containing J, in analogy
by the formula
an exact

obtain

can

We

p.c.
for

\316\241
on

partition

any

da:= p.c. / f da

Ji f

J[P]

with

respect

to which / is

piecewise

constant.

Up

until

now we

assume

that

a is

all

monotone increasing,i.e.,a(y)
that

such

are

X
\342\202\254

x,y

have made no assumptionon a.


>

\321\203

\317\207.This

>

that

implies

integrals p.c.

JT

are

by the

replaced

are

replaced

by

\316\261-lengths

a(I)\\

> 0

a(I)

in X (why?).
Prom this one can easily
to
hold
results from Theorem 11.2.16continue

now

whenever

a(x)

intervals

all the

us

Let

for

that

verify

when

the

p.c. JT f da, and the lengths


see

Exercise

|/|

11.8.3.

then define upper and lower Riemann-Stieltjes


R is bounded
da and J f da whenever/ : J \342\200\224>
on a domain containing J, by the usual
formulae

We^ can
integrals

is

fjf

defined

/ da

:= inf {p.c. I g da : g

is

p.c.

on J

and

and majorizes

/}

minorizes

/}.

\316\261

and

I f da :=

sup{p.c.

We

then

respect

say that
to a if the

g da

:g

is p.c. on /

and

/ is Riemann-Stieltjesintegrable
and

upper

match, in which case we

lower

Riemann-Stieltjes

set

[fda:=jfda=f

f da.

on

with

integrals

The

As before,

is
when \316\261

integral
Riemann-Stieltjes
thus the Riemann-Stieltjes
(We

integral.

Riemann

a little

integrals

write

sometimes

shall

later,

11.4.1(g),

necessarily

and
is

are
\316\261

unlikely

ct(x) :=

the

\317\207
then

Riemann

integral;

integral is a generalizationof the


see another comparison between the two

11.10.3.) Becauseof this, we


dx.
remaining
theory of the Riemann
or JT f(x)

the

without

over

replacing

difficulty,

integrals

Riemann-Stieltjes
are
a couple results

with

and lengths

11.8.4).

(Exercise

Exercise
Theorem

to the

which break down;


and
11.5.3,
Proposition 11.5.6 are not
Proposition
true
when
\316\261
is discontinuous
at key places (e.g., if /
at the same point, then fjf da
both
discontinuous
to be defined.
However, Theorem 11.5.1 is still true
There

\316\261-lengths.
Theorem

function

identity

is identical

fjfasfjfdx

with

integrals

the

in Corollary

Most (but not all) of


integralthen canbe carried
Riemann

337

integral

Riemann-Stieltjes

\321\206\321\2028.

11.8.1.

Prove Lemma

11.8.1. (Hint: modify

the

of

proof

11.1.13.)

of Proposition
11.8.2. State and prove a version
Riemann-Stieltjesintegral.
and prove a version of Theorem
11.8.3.
State
Exercise
Riemann-Stieltjes integral.

Exercise

Exercise11.8.4.State and
the problem

a version

prove

(Hint: one

integral.

Riemann-Stieltjes

here is that

of

which

| Jk\\

appear

for the

11.2.16

11.5.1 for
with

the

the

proof;

references to the length of \\Jk\\


other referencesto the length of \\Jk\\
basically, all of the occurrences
a( Jk)
inside a summation should be replacedwith
a( Jk),

should

but the

for the

of the

some

remain
unchanged, and
be changed to the \316\261-length

should

of Theorem

has to be careful

11.2.13

rest should be unchanged.)

Exercise11.8.5.

Let

sgn

: R

the signum function

\342\200\224\342\226\272
R be

{1

\342\200\224\342\226\272
: [\342\200\224
R be a continuous
1,1]
to
Stieltjes integrable with respect

./[-1,1]

\317\207
> 0

when

=
\317\207

\342\200\2241
when

\317\207
< 0.

function. Show that

Let /

when

sgn, and that

dsgn

/(0).

is Riemann-

\316\265
> 0, find piecewise constant functions
majorizing
is \316\265-close to
whose
/
Riemann-Stieltjes integral
/(0).)

minorizing

now

have

differentiation

via

and

to connect integration
enough
machinery
the familiar
theorem of calculus.
fundamental
such theorems,

two

are

there

Actually,

and

of calculus

theorems

fundamental

two

The

11.9

the

integral

for every

(Hint:

We

The Riemann

11.

338

other involving

and the

integral,

one

the

involving

the

derivative

of the

integral

of

derivative.

11.9.1 (First FundamentalTheoremof Calculus).Let


R
be a Riemann
numbers, and let f : [a,b] \342\200\224\342\226\272
R be the function
integrable function. Let F : [a,b] \342\200\224>

Theorem
a

<

be real

F{x) :=

/.

J/

[\316\261',\317\207]

Then

is continuous.
Furthermore,
F
then
is
xq,
differentiable

at

continuous

Since

Proof.

Thus

11.3.4).
f(x)

<

\316\234for

Now let

if xq \342\202\254
f is
[a, b] and
at xq, and F'(xo) = /(xo)\302\267

/ is Riemann integrable, it is bounded(by Definition


\342\200\224
we have some real number \316\234such
that
\316\234<
all

\317\207
e

\317\207
be
< \321\203

[a, 6].
two

F(y)-F(x)=

elements

of [a, b].

! f-f

J[a,y]

J[a,x]

Then noticethat

f=f

we
11.4.1(h). By Theorem11.4.1(e)

by Theorem

f <

J[x,y]

J[x,y]

M = p.c.J

J[x,y]

M(y

have

thus

x)

J[x,y)

and

f>f
J[x,y]

-M = p.c. I

and

J[x,y]

J[x,y]

thus

\\F{y)-F{x)\\<M{y-x).

-M

-M(y

x)

two

The

\320\246.9.

This is for

> \317\207.
\321\203
By

see that

thus

we
\321\203

\317\207
and

interchanging

339

of calculus

theorems

fundamental

\\F(y)-F(x)\\<M(x-y)

when

\317\207
>

in all

three

we have

Also,

y.

= 0

\342\200\224

F(y)

F(x)

when

=
\317\207
Thus
\321\203.

cases we have

\\F{y)-F{x)\\<M\\x-y\\.
let

Now

6],

-x\\< F(a:n)-

-M|xn

\342\200\224

But

n.

each

for

M|rrn

\342\200\224>
\316\267
oo,

as

\342\200\224\342\226\272
and
\316\267
oo,

all

\342\200\224

and

x\\

so

thus
xn

sequences

\\f(x)

\342\200\224

/(#o)|

+ \316\264]\316\240
\316\264,\317\207\316\277
[\316\261,
6], or

now

show

differentiable
Now

F(xq)

M\\xn

x\\

\342\200\224
both
\320\266|

converge

to 0

to

we
\320\266,

thus

element

arbitrary

see that F
of [a,b], we

is continuous at x$.
can find a \316\264
we
0
>
by continuity,
\342\200\224
J := [xq
\302\243
for
in the
interval
all \320\266
and

words

/(\320\266)

<

/(xo)

for
\316\265

all

\320\266\320\265/.

that

\\F(y)
for all

<

an

\342\202\254
[a, b],

in other

\342\200\224
<
\316\265

f(xo)
We

xo
Then

0.

\316\265
>

any

that

M|rrn

converging

\342\202\254
[a, b]

that

suppose

Choose
such

<

F(a?)

to 0
test F(xn) \342\200\224
converges
F(x)
=
this
is
true
Since
limn-^00F(xn)
F{x).

was
is continuous at x. Since\320\266
thus see that F is continuous.
Now

in [a,b]

sequence

any

by the squeeze

as
for

be

(#n)\302\243L0

we have

Then

x.

to

converging

let

and

\317\207
\342\202\254
[a,

- F(x0) -

a?o)| <

f(x0)(y

e\\y

xo\\

that F is
desired.

10.1.7 will then imply


Proposition
with
derivative
xo
F'(xo) = f(xo) as
= \320\266\320\276>
J.
There
are three cases. If \321\203
then
fix
\342\202\254
\321\203
F(y)
\342\200\224
\342\200\224 =
and
the
claim
0
is
so
>
\321\203
xo)
f(xo)(y
since
\342\202\254
J,
\321\203
at

obvious.If

then

F(y)-F(xo)=

f.

\342\200\224

\320\266\320\276>

11. The Riemann

340

of

\342\202\254
I,
\320\266\320\276>2/

we have

thus

and

J,

f(x0) -

<
\316\265

/(\320\266)

<

f(xo)

for
\316\265

<

all

\317\207
\342\202\254
[\320\266\320\276,\321\203],

thus

and

- e){y -

(f(x0)

and so in

case

The

desired.

as

z0) < /

\342\204\226o) +

e)(y

x0)

particular

\\F(y)

11.9.2.

Example

a connected set, then [xo,y]is a subset

J is

and

Since

integral

xq is

<
\321\203

Recall

function

monotone

- f(xo)(y -

F(x0)

<

xo)\\

similar and

e\\y

x0\\

is left to the reader.

in Exercise 9.8.5 that we constructed


: R \342\200\224\342\226\272
R which
was discontinuous at

else.
By Proposition
every rational and continuous
everywhere
on [0,1].If
11.6.1, this monotone function is Riemannintegrable
a
F : [0,1] \342\200\224\342\226\272
:= L\302\267\316\271
then
is
we define
R by F(x)
F
continuous
/,
is
number.
function which difFerentiable
at every
irrational
On the

other hand, F is non-difFerentiable

at

every

rational

number;

see

Exercise 11.9.1.
the

Informally,

first

fundamental

theorem

of calculus

asserts

that

(/

/)'(*) = /(*)

J[a,x]

a certain
number of assumptions on /. Roughly,
the derivative of an integral recovers
the
original
Now we show the reverse, that the integralof a derivative

given

that

the

for

let /
all

recovers

11.9.3 (Antiderivatives). Let J bea bounded


interval,
\342\200\224\342\226\272
R
R be a function.
F : / \342\200\224\342\226\27
We say that a function
=
antiderivative
of / if F is differentiable
on J and F'(x)
f(x)

Definition
is an

means

function.

function.

original

and

this

:J

\317\207
e

I.

two

The

\321\206.9.

Theorem 11.9.4

of calculus

theorems

fundamental

f : [a,b]

and let

: [a,

Riemann
antiderivative of f,

\342\200\224>
R

is an

\342\200\224\342\226\272
R

6]

of Calculus).

Theorem

Fundamental

(Second

let a < b be real numbers,


irttegrable function. If F

341

be a

then

/ =

F(b)-F(a).

J[a,b]

sums. The idea is to show

use Riemann

will

We

Proof

that

U(f,P)>F(b)-F(a)>L(f,-p)

partition

for every

f(b)

F(a)

\316\241
of

[a,b]}> while the right inequality


:
upper bound for {L(/,P)

Proposition11.3.12,this

since

/ is

for

{{/(/,
asserts

\316\241
is

: \316\241
is
\316\241)

that

partition

assumed to

F(b)

asserts that
of

a partition

is an

\342\200\224

F(a)

of [a,b]}.

But by

that

means

f>F(b)-F(a)>

but

left inequality

The

[a,b].

lower bound

is a

\342\200\224

/,

be Riemannintegrable,both the

upper

The claim follows.


integral equal \316\223
^ /.
>
to show the bound C/(/, P)
We
have
F(b) shall
show the first inequality {/(/, P) > F(b)\342\200\224
We
just
F(a);
is similar.
other inequality
and

lower

Riemann

F(a)> L(/,P).

Let \316\241be

a partition

F(b)

of [a,

F(a) =

b]. Prom Lemma11.8.4we

\316\243

F[J)

JeP

while from

=
\316\243

jgP:J#0

definition we have

u(f,p)=

\316\243

8\320\270\320\240/(*)\320\230\302\267

JeP:J?9xeJ

Thus

FtJb

it will suffice

to show that

F[J]<8up/(aO|J|
xeJ

have

the

Riemann

The

11.

342

integral

than
for all J \342\202\254
\316\241
the empty set).
(other
since
both
When J is a point then the claimis clear,
sides are
zero. Now suppose that J = [c,d],(c,d],[c,d),or (c,d)
for some
\321\201
the left-hand side is F[J] = F(d) - F(c).By the
< d. Then
for some e e J.
this is equal to (d \342\200\224
mean-value
theorem,
c)F'(e)

But sinceF'(e) =

/(e),

= (d

F[J]

we

thus

- c)f(e) =

have
<

f(e)\\J\\

sup

xeJ

f(x)\\J\\

as desired.
Of

course,
theorem

fundamental

\342\226\241

one can

as you are all aware,


to compute
of calculus

use

second

the

integrals relatively easily

an anti-derivative of the integrand


provided
you
Note
that
the
theorem
of calculus ensures
first
fundamental
/.
function
has an antithat every continuous
Riemann
integrable
the
derivative.
For
situation
discontinuous
is more
functions,
can

that

complicated,and is a

find

real

graduate-level

analysis

topic

which

will

Also, not every function with an antiis Riemann


derivative
as an example, considerthe
integrable;
-> R defined by F(x) := x2 sin(l/:z3)
: [-1,1]
when
function F
\317\207
\317\206
\316\237,
:=
and
Then
F
is
differentiable
0.
so
F'
everywhere
F(0)
(why?),
has an antiderivative, but F' is unbounded
and
so is not
(why?),
Riemann integrable.
We
now
to mention
the infamous \"+C\" ambiguity
in
pause
not

be

here.

discussed

anti-derivatives:

Lemma
a function.

11.9.5. Let I be a bounded


and let f : J \342\200\224>
R be
interval,
R and G : J \342\200\224>
R be two antiderivatives
Let F : J \342\200\224>

off. Then there existsa


e I.
for all \317\207
Proof.

Let

that for every

not differentiable
10.2.9.)

number

\320\241
such

that

= G(x)+C

F(x)

11.9.2.

See Exercise

Exercise11.9.1.
Show

real

: [0,1]

rational

\342\200\224>
R be
number

the

function

\302\243
Q \316\240
[0,1],

at q. (Hint: use the mean-value

in

9.8.5.

Exercise

the
theorem,

function

/ is

Corollary

Exercise11.9.2.

the second

using

but one has

to be

(Hint:

to the function

10.2.9,

Corollary
lemma

this

11.9.5.

Lemma

Prove

theorem,

343

of the fundamental theorems

Consequences

11.10.

apply

\342\200\224One

G.

the mean-value
can also prove

Fundamental theorem of calculus


we do not assume / to be

since

careful

(how?),
Riemann

integrable.)

Let a < b be real numbers,


and
let / : [a, b] \342\200\224>
R be
R be the function
increasing function. Let F : [a, b] \342\200\224>
Let
F is differenxo be an element of [a,6]. Show that
F{x) \302\267=
Jja x] /\302\267
if
and
is
at
if
continuous
one
direction
is
at
tiable
xq
xq. (Hint:
/
only
one
the
of
of
of
fundamental
theorems
calculus.
For
the
care
taken
by
other, considerleft and right limits of / and argue by contradiction.)
11.9.3.

Exercise

a monotone

Consequences of the fundamental

11.10

theorems

now

can

We

fundamental

number of useful consequencesof the


(beyond the obvious application, that
for which an anti-derivative is
integral
any
the
is
familiar integration by parts
application

give a

of calculus

theorems

one can
compute
The
known).
now

first

formula.

Proposition 11.10.1
[a, b], and let F : b]
functions

I.

Then

on [a, b]
we

by

(Integration

and

G :

[a,

\342\200\224\342\226\272
R

such

that

and G

Ff

parts

Let I
formula).
\342\200\224\342\226\272
R be differentiable

[a, b]
are Riemann integrableon

have

FG' =

F(b)G(b) -

F(a)G(a)- f F'G.
J[a,b]

J[a,b]

Proof

See Exercise

11.10.1.

certain
Next, we show that under
a Riemann-Stieltjes integralasa Riemann
piecewise constant functions.

Theorem 11.10.2.
function, and suppose
Let

that

a!

being

Riemann

integrable.

circumstances,
integral.

can write

one
We

begin

R be a monotone
b] \342\200\224\342\226\272
is also differentiable on

: [a,
a

Let f

: [a,b]

\342\200\224>
R

be a

with

increasing

[a,b],

with

piecewise

11. The Riemannintegral

344
constant

on

function

[a,b],

J[atb]

Suppose

that / is piecewiseconstant with


respect
loss
of
without
generality we may
[a^b}\\
the empty set. Then we have
not
contain

\316\241
does

J[*A
cj

Theorem
is

why

is the

/ da = p.c./

/\302\267>'

the

\316\243

Corollary

\316\243

11.10.3.

the

hand,

from

on

integrable

[a,

that

since /

\302\253//\302\267

monotone

increasing
on
differentiable
[a, b], with a'

[o,f>]^R
respect

to

which

a function

be

on [a,b].

Then

then

must

b], and

f da=

[ fa'.
J[a,b]

J[a,b]

Note

\316\243

be a

is also

integrable. Let /:
is Riemann-Stieltjes
integrable with
Riemann

/V

\342\200\224\342\226\272
R

b]

Riemann

and a' are bounded,

bounded. Also, since a


a' is non-negative.
ferentable,
also be

other

the

follows.

claim

Let a : [a,

function, and supposethat

Proof.

cja[J]

theorem of calculus (Theorem

fundamental

second

and

fa' is

on J. On

/^

11.9.4),fja' = a[J],

being

da

assume

j\302\243

constant value of /

by

J[P]

some

11.2.16(h) (generalized to partitions of arbitrary length


this generalization true?) we have

/
But

to

\316\241
of

partition

where

integrable

11.4.5.

Theorem

that

then fa! is Riemann

Riemann integrable,

is also

a'

and
piecewise constant, it is Riemannintegrable,

/ is

Since

Proof.
since

on

fa'.

fda=[

J[afi]

by

is Riemann integrable

Then fa'

[a,b].

and

is

monotone

increasing

fa'
and

dif-

11.10. Consequences
of
Let

we can find

Then,

/ on

f da<

f da-e<

f da-e< I

/
/a'. Thus

<

J/

claim

the

fda<[

makes

We
We

first

advantage

sense

even

now

build

need

fa'<f

f_a! minorizes

f da + e.

fot.

J[afi]

[afi]

for

any

fa' < Jf

J_[a,b]

Ml

e >

0, we

fa'

<

must have
f da

J[*M

to

of the
when

11.10.3

Corollary

Informally,

is essentiallyequivalent
However, the

then

follows.

11.10.4.

Remark

/,

Thus

f da-e<

are true

statements

J[*M
and

minorizes

have

we

these

J[afi]

(why?).

l[ab]fa'

$_[ab]f_a'

f da + e.

< f

J[afi]

is non-negative and /

Similarly

+ e.

J[afi]

7 ol

<

fot

J[afi]

J[ajb]

Since a!

f da

<

da

mi-

11.10.2, we obtain

Theorem

Applying

J
J[afi]

J[afi]

J[ajb]

Since

345

theorems

fundamental

a piecewise constant function


and
function
a
constant
piecewise
[a, b],
_/
on [a, b], such that
0.

\316\265
>

J majorizing

norizing

the

fj^dx,

when

up to the
a preliminary

that

/ da

differentiable.
integral is that it still

a is

Riemann-Stieltjes

a is

asserts

not differentiable.
familiar change of variablesformula.
lemma.

11. The Riemannintegral

346
11.10.5

(Change

interval,

and let

Lemma
a closed

of variables formula
: [a,
\317\206

6]

0(6)]

[0(a),

monotone increasing function. Let f : [0(a),


0(b)]
wise constant function on [0(a),0(b)].Then
f
ateo piecewise
constant on [a,b], and
\316\257 \316\257\316\277\317\206\316\254\317\206=
I

I). Let [a, 6]


be a

\342\200\224\342\226\272

be

continuous
be a piece-

\342\200\224>
R
\316\277 :
\317\206[a,

6]

\342\200\224>
R

/.

the
to be filled
Proof. We give a sketchof the proof,leaving
gaps
in Exercise 11.10.2. Let \316\241
of [0(a), 0(b)] such that
be
a partition

/ is piecewise

with

constant

does

not

constant

to

respect

contain

the empty

value of

/ on J,

we
\316\241;

set. For each J

may

assume

let

P,

that
be
\321\201j

\316\241

the

thus

f =

\316\243 c'i

Ji\302\267

0_1(J) be the set 0_1(J):= {\320\266\342\202\254


[\316\261,
6] :
and
is
connected
is
thus
an
(why?),
J}.
0_1(J)
\316\277
on
interval.
cj is the constant value of / \317\206 0~1(J)
Furthermore,
we
if
define
Q := {0_1(<\320\233 : J G \316\241}(ignoring
Thus,
(why?).
the
the fact that Q has been used to represent
rational
numbers),
is piecewise
then Q partitions [a,b] (why?), and / \316\277
constant
\317\206
with respect to Q (why?).Thus
For

J, let

interval

each

Then

0(\320\266) G

/o0d0=^cJ0[0-1(J)].

/O0d0=/
J[Q]

J[*M

But

0[0~1(jr)]

(why?),

\\J\\

J-p

the

and

claim follows.

of variables
formula
Proposition 11.10.6 (Change
II). Let [a,b]
\342\200\224\342\226\272
: [\316\261,
a continuous
be a closedinterval, and let \317\206
be
b]
0(b)]
[0(a),
\342\200\224>
R be a
monotone increasing function. Let f : [0(a),
0(6)]
\316\277 :
R
Riemann
on
Then
f \317\206[a, b] \342\200\224\342\226\272
integrable
function
[0(a), 0(b)].

is Riemann-Stieltjes

integrable

J[a,b]

with

respect

}\316\277\317\206\316\254\317\206=

J\\fKa)<Xb))

to

f.

on
\317\206

[a, b],

and

of the

Consequences

11.10.

proof This
manner to

how

then

and

bounded,

Let

\316\265
>

0.

[0(a), 0(b)], and a piecewise

Lemma

Applying

Since

\316\277 is
\317\206

\316\231
/\316\277\317\206\316\254\317\206<7\316\277\317\206\316\254\317\206<
f

J[*M
constant

piecewise

and minorizes

\316\265.

f+\316\265.

\316\221\317\206\316\234,\316\246\316\246)}

\316\277 we
\317\206,

have

\316\257 \316\257\316\277\317\206\316\254\317\206
\302\243\316\277\317\206\316\254\317\206<

J[\"M

while

J<[

JM

W*),\342\204\226]

we obtain

11.10.5,

f-e<

function

constant

that

/</

f-e<[

(why?).

a piecewiseconstant function

on [0(a),0(b)],such

minorizing /

Riemann integrable,it is

be bounded

also

\316\277must
\317\206

we can find

Then,

/ on

majorizing

/ is

since

that

observe

First

\320\246.10.2.

obtained

was

11.10.3

Corollary

11.10.5 in a similar
from Theorem

Lemma

from

obtained

be

will

347

theorems

fundamental

J-[a,b]

we have

similarly

\316\231 /\316\277\317\206\316\254\317\206.
\316\257\316\277\317\206\316\254\317\206>

J[a,b]

[a,b]

Thus

f-e<

\316\231

\321\202\320\260),\321\204{\320\254)]

\316\265
>

Since

0 was

/
the

arbitrary,

f<[
claim

J-[a,b]

\320\220\320\244\320\250(\320\252)}
L[at>)

and

\316\257 /\316\277\317\206\316\254\317\206<
\316\257
\317\212\316\277\317\206\316\254\317\206<

WM

this implies that

/o0d0</ J

fo<j>d<i><[
\316\234

f
-/[*(\316\261),*(41

follows.

one
Combining this formula with Corollary11.10.3,
immediately obtains

the

following

f+\316\265.

\316\221\316\246\316\234,\316\246\316\237\302\2

familiar

formula:

11. The Riemann

348
11.10.7

Proposition
be

a closed

(Change

and let

interval,

of variables formula

: [a,b]
\317\206

\342\200\224\342\226\272

0(b)]

[0(a),

such
that
tiable monotone increasingfunction
\342\200\224\342\226\272
R be a Riemann
grable. Let f : [0(a),0(6)]
on

0(b)].

[0(a),

Then

integral

III). Let [a,6]


be a

integrable

is Riemann

\342\200\224\342\226\272
\316\277
: [\316\261,
R
\317\206)\317\206'
6]

(/

11.5.2and

Theorem

integrable.

Then

J[4>(a),<

Proposition

Exercise11.10.2.
Fill

in

11.10.1.

(Hint:

first use

Corollary

F'G are Riemann


product rule (Theorem 10.1.13(d)).)

11.4.5

use the

function

I
\316\275\302\260\316\246)\316\246'=

J[a,b]

Exercise 11.10.1. Prove

inte-

integrable

on [a,b], and

differen-

is Riemann
\317\206'

the

FG' and

to show that

gaps marked

(why?) in the proof of

Lemma

11.10.5.

Exercise
11.10.3. Let \316\261
< 6 be real numbers, and let /
a Riemann integrable function. Let g [\342\200\2246,
->Rbe
\342\200\224a]

be

:=

g(x)

flat]

Show
/(\342\200\224\320\266).

also Riemann integrable,and

-\302\273
R,

defined
Jj_b

by

, g

is the analogue of Proposition


11.10.7
when
monotone
of
instead
increasing? (When
decreasing
\317\206
monotone
increasing or monotone decreasing,the situation
more complicated.)
significantly
11.10.4.

is monotone
\317\206

neither

becomes

g is

[a,i>]

/\302\267

Exercise

is

that

What

Chapter

the basics of

Appendix:

The

this

of

purpose

mathematical

logic,

worksis alsovery
of

helpful

once

which

thinking,

mathematical

is to give a quick introductionto


language one uses to conduct
how mathematical
proofs.
Knowing
logic
for understanding
the mathematical
way
mastered
allows you to approach
in a clear and confident way
problems

appendix
which

mathematical

rigourous

and

concepts

mathematical logic

is the

proof-type questions in this text.


is a very useful skill. It is somewhatrelated
Writing
logically
not
or
the same
as, writing clearly, or efficiently,
to, but
one would want to do all ofthese
ideally
informatively;
convincingly, or
but
one
has
to make compromises,though
sometimes
at
once,
be able to achieve more of your
with
practice
you'll
writing
Thus
a
concurrently.
logical argument may sometimes
objectives
many

including

look unwieldy,

big

that one can be absolutely


correct,

as

long

as all

or otherwise

complicated,

excessively
The

unconvincing.

of the

of writing

advantage
sure

that

your hypotheses

your

were correct and

be
logical; using other styles of writing one can
convinced
that something is true, but thereis a difference
convinced
and being sure.
being

were

Being logicalis not the only


fact sometimes it gets in
often resort to short informal
rigourous when they want

the

desirable

way;
arguments

to

convince

appear

logically, however, is
will be
conclusion
steps

reasonably
between

in writing,

trait

mathematicians

and in

for instance

are not logically


of a
mathematicians

which

other

your

350

A.

without

statement

is true

same

that a

not

reasons

there

to

are

of mathematical

is

as

often

\"not

be emphatic

logical\"

situations

many

l0Qi

the long details,and

non-mathematicians

for

statement or argument

a bad thing;

all of

through

going

of course

basics

the

Appendix:

well.

the

So saying

is not

necessarily
when one has good

about being logical.

one

However,

distinction betweenlogicalreasoning
and
to
more
informal
means of argument, and not try
off an
pass
if an
illogical argument as being logicallyrigourous.In particular,
is asking for a proof, then it is expecting
exercise
to be
you
logical
be aware

should

in your

of the

answer.

like
but this
Logic is a skillthat needsto belearnt
any other,
also innate to all of you - indeed,
use the
you probably
in your
laws of logicunconsciously
and
in your
speech
everyday
own internal (non-mathematical)
it does
take
reasoning.However,
this
innate
a bit of training and practiceto recognize
skill and
as
those
to apply it to abstract situations such
encountered
in
mathematical
of logic
proofs. Because logic is innate, the laws
that
you learn should make sense - if you find yourself
having
to memorize one of the principles
or laws of logic here, without
or
comprehending
why that law should
feeling a mental \"click\"
not
to
be able
use that law of logic
work, then you will probably
and effectively in practice. So, pleasedon't study
this
correctly
- that
cram
is
the
before
a
final
appendix
way you might
going to
be useless. Instead, put away
pen, and read
your
highlighter
and understandthis appendixrather
than
it!
merely
studying

skill is

statements

Mathematical

A.l

mathematical

Any

mathematical
various

statements.

mathematical

argument
These
objects

proceeds
are precise

in a sequence of
statements concerning
vectors,

(numbers,

functions,

relations betweenthem (addition,equality,


differentiation,
These
either
be
more
can
constants
or
variables;
objects
later. Statements1 are either true or false.
\321\205\320\234\320\276\320\263\320\265
statements
precisely,

We shall discuss free variables

with no
later

free variables are either


this appendix.

on in

etc.) and

true

etc.).
on

this

or false.

yjj. Mathematical statements


2 +

A.l.l.

Example

2 =

351

4 is a true statement;

2 +

of mathematical

symbols is a

2 =

5 is

statement.

false

Not
statement.

every
For

combination

instance,

= 2

+ +4

= - =2

call it ill-formed or ill-defined.


in the previous
statements
example are well-formedor wellThe
Thus
well-formed
statements
can be either true or false;
defined.
are considered
to be neither true nor false
statements
ill-formed
not
considered
statements at all). A
are
fact,
usually
they
(in
of
ill-formed
is
an
statement
subtle
more
example
is

not

a statement;

we sometimes

0/0 = 1;
the above statement is illA
should not contain any ill-formed
formed.
argument
logical
an argument uses a statementsuch
for
instance
if
t
hus
statements,
=
is not
to zero.
as \317\207/y
equal
2, it needs to first ensure that \321\203
of
o
r
other
\"0=1\"
false
statements
Many purported proofs
rely on
this
\"statementsmust
be
well-formed\"
criterion.
overlooking
of you have probably written ill-formedor otherwise
Many
in your
mathematical
statements
inaccurate
work, while intending
well-formed
and
accurate
statement.
To a
to mean someother,
- it is similar to misspelling
certain extentthis is permissible
some
words in a sentence, or usinga slightly
inaccurate
or ungrammatical word in place of a correctone(\"She
ran
of \"She
instead
good\"
ran well\.") In many
the
can
reader
detect
this
cases,
(or grader)
mis-stepand correct for it. However, it looks unprofessional and
suggeststhat you may not know what you are talking about. And
if indeed
do not know what you are talking
about,
you actually
and are applying mathematical or logicalrules
then
blindly,
statement
can quickly confuse you into writing
ill-formed
writing an
no
more
and
more nonsense
receives
usually of the sort which
credit in grading. So it is important, especially
when
learning
just
division

by

zero

is undefined,

and so

A,

352

Appendix:

the basics ofmathematicalIon*

care in keeping statements well-formedand pre.


more skill and confidence, of course you
can
will
afford
once again to speak loosely, becauseyou
know
what
of veering off
danger
you are doing and won't be in as much
int0
to take

a subject,

cise.

Once

have

you

nonsense.

basic axioms of mathematical logic is that every


is either true or false, but not both.
statement
(Though

of the

One
well-formed
if there

are free

variables, the

truth of a statementmay

depend

on

variables. Moreon this later.) Furthermore


the
truth
or falsity of a statement is intrinsic to the
statement
does not depend on the opinion of the personviewing
and
the
statement
(as long as all the definitions and notations are agreed
it suffices
of course).
So to prove that a statementis true,
upon,
to
show that it is not false,while
to show
that a statement is false, it
to show that it is not true; this is the principle
suffices
underlying
of these

the

values

the

powerful

technique

later. This

concepts,

of proof

by

axiom is viable as long

for

which

the

truth

contradiction,
as

or falsity

one

which

is working

we discuss

with precise

can be determined(at least

and consistent manner. However,


if
non-mathematical situations, then this
becomes
much
more dubious,
and so it can be a mistake
axiom
to
mathematical
apply
logic to non-mathematical situations. (For
such
as \"this rock weighs 52 pounds\"is
instance,a statement
and
and so it is fairly
to use
safe
reasonably
objective,
precise
mathematical
to manipulate
it, whereas vague statements
reasoning
such as \"this rock is heavy\", \"this piece of music is beautiful\"
or
exists\" are much more problematic. So while
\"God
mathematical
is
and
it
have
a
still
some
useful
does
tool,
logic
very
powerful
of
can
to
limitations
still attempt
One
logic
apply
(or
applicability.)
similar to logic) in these cases(for instance,
by creating
principles
this
but
is now
a mathematical model of a real-lifephenomenon),
and
we will
not discuss it
science or philosophy, not mathematics,
in

in an

principle)

one

is working

further

objective

in very

here.

There are other modelsof logicwhich


attempt
to deal with statements
that are not definitely
or definitely
true
or fuzzy
false, such as modal logic, intuitionist
logic,
logic, but
Remark

A.1.2.

well beyond the scopeofthis text.

these are

from

is different

true

Being

being useful

2 =

is true

or efficient For

statement

the

instance,

353

statements

Mathematical

\302\243j.

but

to

unlikely

be

2
The statement

useful.

very

4<4
is

also

true,

but

It

may

precise).

useful, for

not very efficient (the statement 4


also be that a statement may
be false

= 4 is

more

yet still be

instance

=
\317\200

but is still useful

22/7

In
approximation.
ourselves
with
truth
rather
mathematical
reasoning,
only
or efficiency;
usefulness
the reason is that truth is objective
than
on
can
and
we can deducetrue statements
agree
it)
(everybody
whereas usefulness and efficiency are to some
from
rules,
precise
of opinion,
and do not follow
matters
rules.
extent
precise
Also,
not
seem
even if some of the individual steps in an argument
may
is false,

very useful
the

for

true)

or efficient,

and

first

concern

it

conclusion

final

as

we

is still

to be

possible

(indeed, quite

quite non-trivial (i.e., not

common)
obviously

useful.

are true
Statements are different
from
Statements
expressions.
which
or false; expressionsarea sequence
of mathematical
symbols
produces some mathematical object (a number,matrix,function,
as its value. For instance
set,
etc.)

2 + 3*5

is

an

value.

not

expression,

a statement;

it produces

a number as

its

Meanwhile,

2 +

is a statement,not an
2
sense to askwhether
expressions

can be

3*5 =

expression.

17
Thus

it does

not make any

+ 3 * 5 is true or false.As with


statements,
well-defined or ill-defined;2+ 3/0,for instance,

A.

354

is

examples of ill-definedexpressions
arise
to add a vector to a matrix
instance,
attempting
0r
outside of its domain,e.g.,sin-1
a function
statements
out of expressions by usingrelations
make

for

(2).

evaluating
can

One

as

such

<, >,

=,

prime\",

\"is

can

One

in

\"is invertible\",

by using

if-and-only-if,

is

\"
\320\243is

and

primitive

logical connectives such as and,


or,
not,
and so forth. We give some examplesbelow

If X is a statement

Conjunction.
\"X

words.

if-

of intuitiveness.

order

decreasing

statement

as

statement,

statements are allowed to contain English


more
make a compound statement from

mathematical
statements

using properties (such as %


etc.) For instance, \"30+5
\"30 + 5 < 42 \342\200\224
7\".
Note
that

or by

etc.

\342\202\254,
C,

continuous\",

is prime\" is a

then,

mathematical 1\320\276\321\213

subtle

More

ill-defined.

when,

the basics of

Appendix:

true

\320\243
is

and

both
and \320\243
are
= 4 and 3 + 3 =

if X

a statement,

true,

and is

the

false

is true,
6\"
otherwise. For instance,\"2 + 2
while
=
=
\"2 + 2
4 and 3+ 3 5\" is not. Another example: \"2 + 2 = 4 and
2 + 2 = 4\" is true, even if it is a bit redundant;logicis concerned
with

not

truth,

efficiency.

Due to the expressivenessof the English


reword the statement \"X and \320\243\"in many
\320\243\",or

\"X, but

\320\243\"
is

have

and

\320\243
are

that

in

\320\243
are

logically

etc.

same

ways,

Interestingly,

statement

e.g.,

\"X

can

and

also

the statement

as \"X and \320\243\",but

connotations (both statements affirm


that
X
the
first
but
version
that
X
and
true,
suggests
to each other, while the secondversion
suggests

both

contrast

and

\320\243
support

not about

Disjunction.
\"X

each
or

connotations

statement

the

true\",

one

different

they

\320\243
are

X and

\"Both

language,

other).

Again,

logic is

about truth,

suggestions.

is a statementand \320\243is a statement,


the
or \320\243\"
is true
if either X or \320\243
is true,
or both. For
\"2 + 2 = 5 or
2 = 4 or 3 + 3 = 5\" is true,
but
not.
Also \"2 + 2 = 4 or 3 + 3 = 6\" is true
(even
If X

instance, \"2 +
3 + 3 = 5\" is
if it is a bit inefficient; it would be a stronger
to say
statement
in
\"2 + 2 = 4 and 3 + 3 = 6\.")Thus
\"or\"
the
word
default,
by
this
mathematical
defaults
to
do
inclusive
or.
The
reason
we
logic
is that with inclusive or, to verify
or \320\243\",it suffices
to verify
\"X
that just one of X or \320\243is true; we don't need to show that the

355

statements

Mathematical

J.

for
other one is false. So we know,
=
5931
7284\"
is
true
without
+
2353
in

As

equation.

0r 2 + 2 =

is

4\"

the

previous

true,

even

If one really does want


X or \316\245
is
such as \"Either
\316\245
is
true\".
Exclusive
or
X
as often

near

nowhere

Negation. The
only

if

having

2 =

\"2

to look at

as inclusive
\"X

or.
is not

true\" or

X\", is calledthe
and is false if

is false,

is

\"X

of

negation

and

is true.
5\"

is

to

2^5\".

Negations convert
of

is

and

X,

if X

only

or

false\",

For instance, the statement\"It is not the case that 2 + 2 =


Of course we could abbreviatethis statement
true statement.
\"2 +

4 or

the second

discussion, the statement \"2 + 2 = 4


if it is highly inefficient.
to
use
exclusive
or, use a statement
but not both\" or \"Exactly one of
true,
or does come up in mathematics,but

statement

case that

the

not

is

\"It

true if and

that

instance,

\"Jane

Doe

has

the negation

instance,

and Jane Doehas blueeyes\"


hair or doesn'thave blue eyes\",

black hair

have black

Doe doesn't

For

\"or\".

into

\"and\"

is

not

\"Jane

\"Jane

blue
Doe doesn't have black hair and doesn'thave
eyes\"
(can you
is an integer,
the negation of \"\320\266
is even
see why?). Similarly, if \320\266
is \"x is odd or negative\", not \"x is odd and
and non-negative\"

negative\".
(Note how
rather than exclusive.)
(i.e.,

\"2

<

<
\317\207

6\") is

\"x

it is important here that

Or the negation of

<

2 or

\317\207
>

6\", not

>
\"\320\266

\"x

<

is

or

2 and

2 and

inclusive
<
\317\207
\317\207
>

6\"

6\" or

\"2<\320\266>6.\".

Similarly, negations convert \"or\" into \"and\". The negation of


Doe does not
hair\"
is \"John
\"John Doe has brown hair or black
brown
hair
have
and does not have black hair\",or equivalently
is a real number,
Doe has neither
brown nor black hair\".If \320\266
\"John
< \342\200\2241\"
1\" (i.e.,
the negation of \"x > 1 or \317\207
is \"\320\266
\317\207
< 1 and
> \342\200\224
-1<\320\266<1).

It is

quite possible that

negation

of

a statement

will produce

is
if \317\207
a statement which couldnot possiblybetrue. For instance,
an integer,
the negation of \"x is either
even or odd\" is \"x is neither
even
nor
which cannot possibly be true. Remember,though,
odd\",
and
it is
that
if a statement
even
is false, it is stilla statement,
an
to
arrive
at
a
true
statement
argument
definitely possible
using

356

which at timesinvolves
for instance, fall into

of mathematical

loqic

by contradiction
example is proof by

(Proofs

Another

category.

divides into three mutually

If one

cases.

into

dividing

statements.

false

this

basics

the

\320\233. Appendix:

exclusive

at
2, and Case 3, then
any
given time two of
the cases will be false and only one will be true, however
this
does
asa
whole
is
not necessarily
mean that the proof
incorrect
or that
the conclusion is false.)

1, Case

Case

cases,

to work
unintuitive
with, especially
Negationsare
such
as \"It is not the
are multiplenegations;a
case that either is not odd, or is not larger than or equal to
is not
pleasant to use. Fortunately
3, but not
particularly
to
with
one rarely has
more
than one or two negations at a
For instance, the
cancel
each other.
time, since
negations
sometimes

if there

statement

\317\207

\317\207

both\"

work

often

negation of \"X is not true\" is just \"X is true\", or


careful
when
be
just \"X\". Of course one should
the
becauseof
complicated expressions
switching
\"or\", and similar issues.

If and only if
we say

true,

(iff).

that \"X is true


to

\316\245
has

(i.e.,X and

be also,

If

if

and

is a

and

statement,

only

if

and whenever

\316\245
is
\316\245
is

more succinctly
more
negating
of

\320\243
is

true\",
true,

\"and\"

and

a statement,

whenever

X is

has to

be also

ways of saying the same


and
\316\245
or \"X
are
\"X
are
equivalent
statements\",
thing
logically
iff \316\245
or \"X <-> Yn. Thus for instance, if \320\266
is true
is true\",
is a
=
=
then the statement \"\320\266 3 if and only if 2x
real number,
6\"
= 3 is
=
is true:
this means that whenever \317\207
then
2x
6
is
true,
-=
=
2x
6 is true, then \317\207 3 is true. On the
true, and whenever
other hand, the statement
\"x = 3 if and only if x2 = 9\" is false;
= 3 is true, x2 = 9 is also true,
while it is true that whenever\317\207
= 3 is also
it is not the casethat whenever
x2 = 9 is true, that \317\207
\316\245
are

true\.")

\"equally

automatically true (think of what


and

\316\245
are

false also

true).
be

(because

only

if 4

logically

+ 4

equally

equivalent,

logically

Conversely,

also

are

that

Statements

true, are
and X is

\316\245
were

true,

any

two

statements

= 10.

when

happens

if

equivalent.

Other

then

Thus for

=
\317\207
\342\200\2243).

false:
also equally
false, then \316\245has

if X
to

be

also have to be
which are equally false will
X would

instance 2 + 2 =

if

and

357

Implication

yl.;g.

are

statements

statements

all of the

are

statements

one might want to

true;

all

\316\226
are

logically

statements is true, then


means that if one of the

one of the
and it also

whenever

means

This

and
\320\243,

X,

two

than

more

that

show

for instance,

equivalent;

logically

assert that three


equivalent.

of interest to

it is

Sometimes

then all of the statementsare

is false,

false.
This
statements
may
but
in practice,
seem like a lot of logicalimplications prove,
between
once one demonstrates enough logicalimplications
X, \320\243,
conclude
that
and
Z, one can often conclude all the othersand

are all

they

to

logicallly equivalent. Seefor

A. 1.5,

Exercises

instance

A.1.6.

Exercise A.1.1. What


or

is
\316\245

but

true,

is the

of the

negation

statement

is true,

\"either

not both\"?

of the statement \"X is true if and


be
multiple ways to phrasethis negation).
(There may
that
X is true,
A. 1.3. Suppose that you have shown
whenever
Exercise
is true,
Have you now
and whenever X is false, then \316\245
is false.
then \316\245
are
demonstrated that X and \316\245
equivalent?
Explain.
logically

Exercise A.1.2. Vfaiat

only if

\316\245
is

Exercise A. 1.4.
then

\316\245
is

negation

Supposethat

that X

is true

A.1.5. Suppose you

Exercise

and you know that

to show that

X,

Exercise A.1.6. Supposeyou


true;

that

whenever

then

X is

true. Is

equivalent?

Explain.

A. 2
we

come

true,

that

know

then

enough

to

\316\226
is

show

true if and
\316\226
is

true.

is
\316\245

enough

Explain.

X is true,
whenever
and
whenever
true;
that

if

only

Is this

now

\316\245
is

then
\316\226
is

are
all
\316\247,\316\245,\316\226

true,

logically

least intuitive of the commonly used logical


is a
If X is a statement, and \320\243
implication.
from X to Y\\ it is
\"if X, then \320\243\"is the
implication
to the

connectives

also

this

know

true

you

Implication

Now

statement,

\316\245
is

and

X is true,

false,

X is
that
if and only if
all logically
equivalent?

\316\245
is

\316\226
are
\320\243,

\316\245
is
if

shown that whenever


then X is false. Have
is true?
Explain.
only if \316\245

have

you

and whenever

true,

demonstrated

true,

is the

true\"?

then

written

\"when

is true,

\316\245
is

true\",

or

\"X implies

Yn

or

A.

358

then

X,

\320\243\"
means
\"if

then

X,

true.

But

when

false, then

X is
\320\243
is

\"if

\320\243
is

X,

(this

last
\"if

true or false.If \317\207[s

true,

then

true\"

this statement

What

see).

mathematical loqic

and

false when

\320\243\"
is

always

\316\245

true

of whether \316\245is true or false! To put it another way


the statement \"if X, then \320\243\"implies
that
\316\245
is
when X is false, the statement\"if X, then \320\243\"offers
no
or not; the statement is true
is true
about whether \316\245

regardless
when

to

true

\320\243\"
is

X is

only if

is true

on whether

depends

If however

false.

\"X

or

is\"

a bit of mental effort

then

true,
is

when

true

\"\320\243
is

one takes

the basics of

Appendix:

is true,

information

but vacuous(i.e.,doesnot convey


is false).
the fact that the hypothesis

any

new

information

beyond

=
A.2.1.
If \317\207
is an integer,
then the statement \"If \320\266
2,
=
4\"
is actually
then x2
of whether \317\207
is true,
regardless
equal to
2 or not (though this statementis only
to be useful when
likely
\317\207
to 2). This statement does not assertthat
\317\207
is equal
is equal
to 4, but
it does assert
to 2, and does not assert that x2is equal
that when and if \317\207
to 2, then x2 is equal to 4. If \317\207
is equal
is not
no conclusion
offers
on
equal to 2, the statement is still true but

Examples

or
\317\207

x2.

cases of the above implication:


the
implication
= 4\" is true
The
implication
(true implies true).
= 2, then 32 = 4\" is true
(false implies false). The implication
= 2, then
-2
(-2)2 = 4\" is true (false implies true). The latter
are
considered
vacuous - they do not offer
implications
any
Some

\"If 2
\"If3
\"If
two

new

special

= 2,

then

22

since their

information

is still possible
to
a proof - a

employ

vacously

such

example

true

hypothesis is false. (Nevertheless,it

vacuous

statement

implications

is still

to good

true. We

effect in

shall seeone

shortly.)

the falsity of the hypothesis doesnot destroy the


in fact it is just the opposite! (When
implication,
a hypothesis
is false, the implication is automaticallytrue.) The
an implication is to show
that
the hypothesis
only
way to disprove
is true while the conclusionis false. Thus
\"If 2 + 2 = 4, then
4 + 4 = 2\" is a false implication.
(True does not imply
false.)
can also think of the statement \"if X, then \320\243\"as \"\320\243
is
One
As

truth

we see,

of

an

359

Implication

yl.2.

at least as true as X\" but if X is false, \316\245could


X and

that

asserts

fly\".

choice

for

of
and

used
both

are

regardless.This

statement,

know

conclude

that

pigs

would

is not

here

be

here by

that

believe

if you

contradiction:

\"John

that

wings,

over\" is

5pm, then he would

work at

left

had

then

be
here
This kind
by now.\"
situation in which the conclusion
false; but the implication is still true
the
by
way, can be used to illustrate

by

you also

speech,

in a

the techniqueof proof


John

\320\243\",which

also a popular
more serious one is \"If John
had

then he

is often

hypothesis

if

only

were

wishes

\"If

\"hell freezes

hypothesis.)

at 5pm,

statement

is

example

a false

and

if

\"X

true,

true.

equally

statement

(The

work

left

\316\245
are

frivolous

somewhat

with

to be

has

it could alsobe

implications are often used in ordinary


knowing that the implicationis vacuous.

Vacuously true
without

sometimes

would

be

be compared

should

This

true.

is true, then \316\245


also
as false as X, but

if X

by now\",

now\", then

you

\"If

and
can

not leave work at 5pm\",becauseJohn


work
at 5pm would lead to a contradiction.
Note
how
a
leaving
can
be
used
to
derive
a
useful
truth.
vacuous
implication
To summarize,
are sometimes vacuous, but this
implications
in logic, since these implications are
a problem
is not actually
can still be useful in logical
still true, and vacuous implications
can
I
n
one
like \"If
safely use statements
arguments. particular,
to worry about whether the
\320\243\"
then
without
X,
necessarily
having
or
not
the
X
is
true
implication
actually
hypothesis
(i.e.,whether
is vacuous or not).
Implications can alsobe true even when there is no causal
\"John did

link betweenthe
1 =

States\"is

be

may

one

change,

day

it is true, at

acausal

true

similarly

statement

implications

(false

unstable

while

least

for

in

(the

implies

capital

1+

1 will

the

moment.

a logical

The

conclusion.

and

hypothesis

2, then Washington D.C. is


States\" is true (true impliestrue), although
York
is
statement
\"If 2 + 2 = 3, then New
1 +

the

statement
of

capital

rather

odd;

the

\"If

United

the

United
Of course, such a
false).
of the United States may
the

capital

remain

always

While

argument,

equal

to 2)

but

possible to use
not recommended

it is
it is

of the

360

A.

basics

the

Appendix:

of mathematical

as it can causeunneeded
confusion.
(Thus,
be
used
it is true that a false statementcan
so
true
or
false,
doing
arbitrarily
statement,
be helpful to the reader.)

To

\"If X,

then

for

instance,

logic
while

to imply

would

any other
probably not

way to do
true, and use this (togetherwith
whatever
other
facts and hypotheses you have) to deduceY. This
is still
even if X later turns out to be false\302\267
a valid procedure
the
does not guarantee
of
implication
anything about the truth
of
\316\245
on
and
the
truth
X
first
X,
only guarantees
conditionally
is a valid
proof of a true
being true. For instance, the following
and conclusion
both
of the
hypothesis
proposition, even though
an

prove

this is to

implication

X is

that

assume

first

usual

\320\243\",the

proposition are false:

PropositionA.2.2.//2

2 =

5, then

4=

10 -

4.

2 + 2 = 5. Multiplying both sidesby


Assume
4 = 10. Subtracting4 from both sides, we obtain

Proof.
4 +
as

desired.

by

first

other hand, a common


the

assuming
For

the

instance,

following

that

Suppose

(Incorrect)

distinguish

10 \342\200\224
4

doing
the

=
\317\207

2x

+ 3

= 7.

important that

from

the conclusion;

getting hopelessly
Here is a short proof
confused

if

which

Show that

=
\317\207

4; so 2x+ 3 = 7.

it is

proofs,

hypothesis

2x =

2; so

prove an

implication
the
at
arriving
Proposition is correct, but the
is to

error

PropositionA.2.3.

When

obtain

and then

conclusion

proofis not:

Proof

we

4=

On the
hypothesis.

2,

this

distinction

uses

implications

you

2.
D

are

able

to

there is a danger of
is not clear.
which

are possibly

vacuous.

Theorem A.2.4. Supposethat

an

even

integer.

\316\267
is

an

integer.

Then n(n+

1) is

\316\221.2.

361

Implication

Since

Proof.

\316\267
is

is also

n(n+1)

is odd,
If \316\267

an

even, sinceany

then

1 is

\316\267
+

\316\267
is

again impliesthat

even, which

is even.
n(n

1)

done.

are

we

and

then

even,

even number

of an

multiple

Thus in either casen(n +1) is even,

is even.

If

or odd.

even

\316\267
is

integer,

\"if \316\267
is even,
Note that this proofreliedon two implications:
is odd,
then n(n+l)
is even\". Since
is even\", and \"if \316\267
then n(n+l)
be both
odd and even, at least oneof theseimplications
cannot
\316\267
and is therefore vacuous.Nevertheless,
both
a false hypothesis
has
and one needs both
in order
are true,
of them
these
implications
know
in advance whether
to prove the theorem,becausewe don't
or odd. And even if we did, it might not be worth the
is even
\316\267
troubleto check it. For instance, as a special case of this theorem
we

know

immediately

538

Let

A.2.5.

Corollary
\342\200\224

213.

Then

=
\316\267

+ 1)

n(n

(253

+ 142) *

123 - (423+

+
198)342

is an even integer.

work
In this particularcase,one
out exactly
which parity
- even or odd - and then use
of the two implications
one
vacuous
one.
This
in the above Theorem, discarding
may
it is a false economy, because
but
seem like it is more
what
and
this requires
one then has to
a
is,
parity
can

is
\316\267

only

the

efficient,

determine

\316\267

effort than it would take


more
bit of effort
both implications, including the vacuous
one,
of
So, somewhat paradoxically, the inclusion
otherwise \"useless\"statements in an argument
you effort in the long run! (I'mnot suggesting,
-

ought to

pack your proofs


statements; all I'm saying
that

concerned
correct,

correct

as

long

conclusion

trueor false.)
The
statement

for instance,
= 2\" can
then \317\207
X\";

with

lots
is

here

had

we

if

the

in

argument.

of

of time-wasting
that
you need

or

false,

vacuous,
can

left

just

save

actually

that

course,

you

and irrelevant
not be unduly

argument might not be


is
still
structured to give the
argument
of
those
whether
hypotheseswere
regardless

some

in your

hypotheses

as your

\"If

X,

while \"If
be false if

then
=
\317\207
is
\317\207

\320\243\"
is

2,

not

then x2 =

equal

to

same

the
4\"

\342\200\2242.
These

is

as
true,
two

\"If

\"If

then
\320\243,

=
\320\2662

statements

4,

another
necessarily
\"X if and only if \320\243\"to

not

is

implication

statement

the

then

X,

=
\317\207

that

while if

if

and
\320\243;

4, because
and

true

use

that

for instance, we can say


= 2 then 2x =
if \317\207
4}

thinking about an

way of

2. One

view \"X if

is to

statement

only-if

=
\317\207

4 then

2x

if

statement

the

denote

Thus

X\".

then
\320\243,

and only

2 if

2x

other; thus the converseof a


true implication. We

of each

converses

called

are

\"If

basics of mathematicallogic

A. Appendix: the

362

if

only

\320\243\"
as

if-and-

that

saying

just as true as \316\245;if one is true then so is the other,and if


the
is false, then so is the other. For
statement
\"If
instance,
since both hypothesis and conclusion
2, then 6 = 4\" is true,

X is
one
3

are false.

this

(Under

statement that
and

\"X

\320\243
are

statement

the

Similarly,

of

instead

true\"

equally

if

\"X

and

then

is true,

\"If

be

can
\320\243\"

as a

viewed

as X.) Thus onecouldsay

as true

least

at

\320\243
is

then

\"If X,

view,

if

only
\320\243
is

\320\243\".

is not

true\"

=
the same as \"If X is false, then \320\243
is false\".
that \"if \317\207
Saying
2,
not
\"if
then
\317\2072
then x2 = 4\" does
that
and
imply
4\",
\320\266/2,
\317\206
\342\200\224
\342\200\2242
indeed
we have \317\207
as a counterexample
in this case. If-then

statements are not


knew that
know that

\"X

is

\"X

is

is false,

then

is true,

then

is not

If you

false

if

false\"

is sometimes

true\";

thus

\320\243
is

true

know that
\"If

is
only if \320\243
and only if \320\243
is

\320\243
is

is true,

then

X is

be true, sincethat would


For instance, if we knew that
know

had

left

work

\"If

know

at 5pm\".
as

the

that

John

\"If

he

of

\"If
\"If

then

\320\243
is

2\".
\317\207
\316\246

then

true\",

\"If

by

then

false,

then

\320\243\"
and

\320\243
is

it is

also
then

false,

a contradiction).

2, then x2
Or if we

here

be

true,

= 4\", then we

knew \"If John


now\", then we also

then he couldnot

\320\243
is

X,

\320\243
is

=
\317\207

\"If

would

here now,

The statement

contrapositive

false\".)

false\" (because if
imply

then

\317\2072
4,
\316\246

at 5pm,
isn't

true\",

called the inverse of \"If


the inverse of a true implication

X can't
also

also

we would
The statement

then

implication.

\"If

false,

(If we

statements.

if-and-only-if

if and

\320\243
is

necessarily

true that

as

same

the

true

X is
both

have

work

left

false\" is

known

statements

are

equally true.

In particular, if
known

behind

to

be

proof

false,

you

then

know

that

X implies

something which is

X itself must be false.Thisis the


or reductio ad absurdum: to

by contradiction

idea
show

363

Implication

\316\221.2.

be false,

must

something

assume first

that this impliessomethingwhich

a statementis simultaneously

Proposition A.2.6. Supposethat


>
\317\207
that sin(a:) = 1. Then
\317\200/2.
positive,

sake of

contradiction that

have 0

<

< \317\207< \317\200/2, and sin(O)


have 0 < sin(a:) < 1. But this
>
\317\200/2.
sin(rr) = 1. Hence\317\207

out to

hypothesis
remains

statements

of thing.

But the

sort

is

statements

of blurry.

\"X

is not

written

be

\320\243\"
can

its

about

that

X\",

or

\"!X\",

so

\"\320\245&\320\243\",and

\320\243\"or
\320\233

and

and

\"((\320\266

3)

\320\273 =
(\321\203

5))

then

=
\317\207
+ \321\203

8\".

these

symbols

(except

symbol).

it's not

statements;

=>

or
So

2?)

don't

to

+
{\317\207

8)\" as

\320\243\",

and

SX\302\273,

\"X

forth.

But
often

take

blur

up

for

used;
much

the line

as easy to understand
\"If

=
\317\207

3 and

\321\203

5,

I would not recommendusing


for =\316\246, which is a very intuitive

So in general
possibly

\321\203

==>

not

are

symbols

expressions

\317\207
<

\"X

written

be

symbols

that

symbols to denote logical

general-purpose mathematics, these


English words are often more readable,
tends
more space. Also, using these
between

on

instead

relies

it

negation,

\320\243\"can

be written \"~

\"X

does

rule.

\"X implies

true\" can

that

and

valid,

is particularly useful for showing


X is false, that a is not equal to b,

use special

often
instance

connectives; for

this

line betweenpositiveand negative


>
2 a positive
(Is the statement \317\207

negative statement? What


this is not a hard and fast
Logicians

that

that

conclusion

ultimate

false!).

\"negative\"

at some

is that

(in this case,

be false. Note however

on that hypothesis beingtrue (indeed,

kind

hypothesis

contradiction

it

- that

thus
that

1, we

not rely

by contradiction

sin(a:)

the

not alter the fact that the argument


the conclusion is true; this is becausethe

Proof

\317\207

is increasing
=

does

being

Since

\317\200/2.

8\316\257\316\267(\317\200/2)

contradicts

assume

turns

later

< \317\200/2) which


\317\207

0 and

.=

of proof by

one feature

that

point in the proof you

\317\207
<

\317\200/2. Since

\317\207
<

such

number

a positive

\317\207
be

for

for 0

Note

true).

we thus

Suppose

Proof.
is

to be

not

and

true

and show
false (e.g., that
For instance:

is true,

it

that

know

you

A. Appendix:the

364

To

a statement,

prove

and

one's

working
to

approach

the

like

proving

starts by assuming the hypothesis


a
toward
conclusion; this is the direct
way
a statement.
Such a proof might looksomething
one often

following:

A.3.1.

Proposition

A.3.2.

Proposition

Proof. Let

+ 1 =

In the

\320\241
is

as desired.

true,

is

sin(x/2)

we
\317\200,

have

+ 1

= 2.

\317\207/2

Since sin(x/2) =

= 1.

sin(a:/2)

Since

true.

\317\200/2.

1,

we

Since
have

2.

hypothesis and moved


It
from
is alsopossibleto work
toward a conclusion.
from
the conclusion
and seeing what it would
take
to
A.3.1ofthis
it. For instance, a typical proofof Proposition
above

we

proof,

started

at the

there

steadily
backwards

imply

look

might

Proof To show
we

=
\317\207

\317\200. Since

have

we
\317\200/2,

sin(a:/2)

sort

=
\317\207

=
then
\317\207
\317\200,

If

is
\320\241

true,

such a directapproach

An example of

\317\207/2

B.

implies

true. Since A is
is true. Since D is true, \320\222
is

A is

Assume

Proof.
true,

logic

of proofs

The structure

A.3

of mathematical

basics

just

need

As an

like the
B,

following:

it would

to show

suffice to

C. But

from

follows
\320\241

example of this, we

give

show D. Since\320\241
implies

another

\320\224

A.
proof

of Proposition

A.3.2:

Proof. To show sin(a:/2)+ 1 = 2, it would


imply
sin(a:/2) = 1. Since\317\207/2= \317\200/2would

need to

show

that

\317\207/2

\317\200/2.

But

this

suffice

sin(a:/2)
follows since

to

show

= 1,

that

we just

= \317\200.
\317\207

two proofs of PropositionA.3.2


same,
just arranged
differently. Note how this proof style
from the (incorrect) approach of starting with the
is different
conclusion
and
what it would imply (as in PropositionA.2.3);
seeing
Logically

are

the

speaking,

the above

structure

The
\320\224.\320\227.

365

of proofs

instead, we start

and

conclusion

the

with

see what

would imply

it.
Another

the

of a

example

in this

written

proof

Let 0 <

A.3.3.

series ]CnLi n7\"n

r <1

be

this series

show

To

test to

show that the ratio

is convergent,it

rnn

to

But since

=
\316\217^

clear since
One

be

will

1+

the

by

ratio

\316\267

less than

something

less than 1, it

suffices

n+1

rn+1(n+ l) _
converges

the

Then

number.

real

\342\204\242
convergent

Proof

1 as

\342\200\224*
oo.
\316\267

to show

enough

i, it suffices to

that

Since

\316\217^

show that

r is

already
to

converges

this

\342\200\224>
0. But
\302\243

could

hypothesis

the

following

1.

is
D

\342\200\224>
\316\267
oo.

the

also do any combination of moving forwards


from
and backwards from the conclusion.For instance,
be a valid proof of PropositionA.3.1:
would

Proof To

show

D,

is

following:

Proposition

D.

backwards style

5,

Since we have
have D

we thus

it would
A

by

suffice to

show D. Sonow
we have

hypothesis,

let

C. Since

us

show

\320\241
implies

as desired.

this is exactly the same


the
as
same
before.
Thus
proof
ways to write
certain
sois
to
but
of
how
do
ways
proof down;
up
writing
you
you,
are
readable
and natural than others, and different
more
proofs
to emphasize
different parts of the argument.
tend
arrangements
when
are
course,
you
just starting out doing mathematical
(Of
to get some proof of a result,and
proofs,
happy
you're
generally
don't
of that
so much about getting the \"best\"
care
arrangement
different
proof; but the point here is that a proofcan take
many
Again,

from

a logical

point of view
there are many

forms.)

The above

proofs were pretty

simple

one hypothesis and oneconclusion.


When
hypotheses

and

conclusions,

and

the

because
there

there
are

was just

multiple

proof splits into cases, then

366

A.

Appendix:

proofs can get more complicated.


as tortuous as this:

PropositionA.3.4.

is true.
Proof. Since A is true, \316\225
is true. Also, in light of A, to show

are

cases:

two

now

If

J.

\316\227
and

then

both

instance

and

Prom

a proof might

true.

\320\222
are

look

Then

suffices to

\320\241

G.

/
G,

If instead

G we
and hence

and

that

show G. There

from F

then

true,

obtain

know

\320\222
we

\316\225
and

D it

\316\227
is

\316\227
we
G, and from A and
J
we
from
from
have G, and
we obtain both \320\241
cases
and

obtain

For

logic

true.

are

and

basics

that

Suppose

of mathematical

the

and \320\257we
J is true,

obtain C. Thus in
and
\320\241

D.

into
the above proof couldbe rearranged
a much
but you at least get the ideaof how complicated
a
there
are
several
proof could become. To show an implication
ways
from
the hypothesis;
to proceed: you can work
forward
you can
work backwardfrom the conclusion;
or you can divide into cases

Incidentally,
tidier manner,

in the hopeto splitthe


Another is to argue by
an argumentof the form
A.3.5.

Proposition

imply that

would
D

which

is true;

As you

a proof. With

are

to

likely

much

effort,

there

cases

there may
if you see
try

whichever

another

for sake of

Suppose

Proof.

is
\320\241

easily,
which

is really

several

\320\222
must

things

become

ones are

only one

obvious way
ways

approach

looks

be

false.

to try when attempting


clearer which approaches

probably work but require


probably going to fail. In many

than

if it

This

true.

which ones will

to
definitely be multiple
more
one
to
way
begin

one

\320\222
is

false.

true, this impliesthat

since A is

C. Thus

it will

experience,

and

But

true.

true. Then \320\222is

contradiction that

contradicts

can see,thereare
work

that A is

sub-problems.
you can have

instance

for

contradiction,

Suppose

easier

several

into

problem

to

proceed.

approach

Of course,
a problem,

a problem, you

but be preparedto switch


look
to
hopeless.
begins
the easiest,

so

can just
to

\320\224.^.

and

Variables

Also, it helps

367

quantifiers

when doinga proofto keep

of which

track

or deduced from the


statements
or coming
and results), and which
hypotheses,
are desired
statements
(either the conclusion, or somethingwhich
the
or some intermediate claimor lemma
would
conclusion,
imply
will
be
useful in eventually obtaining the conclusion).
which
the
two
is almost always a bad idea,and
to one
can
lead
up
Mixing
in
a
lost
proof.
getting hopelessly
as hypotheses,
(either
from other theorems

known

are

A.4 Variablesand
One

can

statements

far in logic just

get quite
as

(such

quantifiers

\"2+2

compound statements
logical
various laws of logicto pass

has

this

conclusions;

is known

to lista
ispossible
which

or

dozen

learn how

to

some

other

curious

as

to

propositional

and

such

then forming
then

using

to one's

hypotheses

logic or Booleanlogic.(It
laws of propositional logic,
one wants to do, but I have

do so here,because

to memorize that

tempted

so

everything

not to

chosen

deliberately

one's

as propositional

to do

sufficient

are

primitive

black hair\,")

connectives,

using

from

with

starting

by

\"John

or

4\"

you

might

then

be

list,

and
that
is not
how one should
one
to
be
or
a computer
happens
logic,
device.
However, if you really are
non-thinking
what
the formal laws of logic are, lookup \"laws
of
or
in
the
or
on
the
similar
something
logic\"
library

unless

do

internet.)
of logic
is insufficient,
However, to do mathematics,this level
because it does not incorporate the fundamental
of
concept
such
as
\320\266
or
which
variables - those
familiar
\316\267
denote
various
symbols
or
set
or
which
are
to
some
assumed
value,
quantities
unknown,
to obey some property. Indeed we have
sneaked
in some
already
of these variablesin order
to illustrate
some of the concepts in
because
it gets boring
after a while to
propositionallogic(mainly
talk endlesslyabout variable-free
statements
such
as 2 + 2 = 4 or
\"Jane

has

black

propositional
logic

but

added.

Mathematical

hair\.")
with

the

additional

logic is

thus the sameas

ingredient

of variables

368

A.

a symbol,

is

variable

A
certain

of mathematical

type

the

Appendix:

basics

of mathematical

logic

such as \316\267
or
denotes a
x, which
an
a
a
integer,
vector,
object

matrix

all circumstances,the type of object


that
the variable represents should be declared,otherwise
it will
it.
be difficult to make well-formedstatements using
(There
are
about
variables
very few true statements that onecan make
without knowing
the type
of variables involved. For instance,given
a
=
it is true that \317\207
of any
variable
\317\207
and
if we
\317\207,
type whatsoever,
In almost

of thing.

kind

that

also know

that

= \321\203,
then
\317\207

cannot

say, for instance,

type of

objects

operation

if

of

\317\207
is

do some

useful

that

are
\321\203

conclude

that

=
until
\317\207
+ \321\203
+ \317\207,
\321\203

and

whether

they

=
But
\321\203 \317\207.
we

support

know

one

what

the

the above statement is ill-formed


for instance,
is
a
vector.
Thus if one actually wants to
\321\203

addition;

a matrix

explicit

\317\207
and

can

we

and

then

mathematics,

every

variable

should have

an

type.)

expressions and statementsinvolving


variables,
for instance,
if \320\266
is a real
variable (i.e., a variable which is a real
+ 3 = 5 is a statement.
+ 3 is an expression, and \317\207
number), \317\207
But now the truth of a statement may depend on the value of the
variables
+ 3 = 5 is true if
involved; for instance the statement \317\207
\317\207
2.
is equal
to
to 2, but is false if \317\207
is not
Thus the truth
equal
of a statement
of
involving a variable may depend on the context
- in this case, it depends on what
the statement
\317\207
is supposed
to
be. (This is a modificationof the rule
for
propositional
logic, in
which all statements have a definite truth value.)
One

can form

Sometimes

we

do not

set a

variable to be anything

(other

than

=
the
statement
x+3
Thus, we couldconsider
5 where \317\207
is an unspecified
real number. In such a casewe call
this variable
a, free variable; thus we are considering
x+3 = 5with
\317\207
a free
with free variables might not have
Statements
variable.
a definite
truth
as
value,
they depend on an unspecifiedvariable.
For
we have already remarked that \317\207
+ 3 = 5 does not
instance,
truth value if \320\266
have
a definite
is a free
real variable, though of
for each
course
\317\207
value
of
the
statement
is either true or
given
= x2 + 2x+1 is
false. On the otherhand,
the
statement
(x + l)2
for every
real number x, and so we can regard
true
this as a true
specifying

its type).

and

Variables

\320\233.4-

369

quantifiers

is a free variable.
statement even when \320\266
to equal
At other times, we set a variable

a statement such as
case, the variableis

\"Set

2\" or

bound

as

known

a fixed
\317\207
equal

variable,

value, by using
to 2\". In this
and statements

variables do

no free

and

variables

bound

only

involving

=
\317\207

\"Let

have

the

For instance,
set
342, then
now has a definitetruth
whereas
value,
=
be
true
variable
then
477\"
could
either
a
free
real
is
\"\320\266+135
\320\266
if
\317\207
is.
as we have said before,
or false,dependingon what
Thus,
of a statement
such as \"x + 135 = 477\" depends on the
truth
the
\317\207
or bound, and if it is bound, what it is
is free
context- whether
value.

truth

definite

statement

=
\317\207

we

= 477\"

+135

ux

if

to.

bound

One

using the

a free variable into

also turn

can

quantifiers

or

all\"

\"for

variable

bound

For instance,

\"for some\".

by

the

statement

(x + l)2
is a statement with onefree

x2

and
\320\266,

variable

truth

value,

(x +
is a

l)2 = x2+ 2x

all real

1 for

definite

numbers \317\207

bound variable x, and now


this
value (in
case, the statementis true).
with one

statement

truth

have a

not

need

statement

the

but

+ 1

+ 2x

a definite

has
Similarly,

the

statement

x +

3= 5

has one freevariable,and doesnot have

truth

a definite

value, but

the statement
\317\207
+

is true,

since

it is

3 =

true
\317\207
+

is

false,

numbers

because
\317\207
for

for

3 =

there
which

5 for
=
\317\207

some real number


2. On

5 for

the other

all real numbers

\317\207

hand, the statement


\317\207

are some (indeed, there


3 is not equal to 5.

\317\207
+

are

many)

real

370

A.

Universal

the

Appendix:

Let P(x)

quantifiers.

basics

of mathematical

loq%c

be some statement

x. The statement \"P(x)is true for all \317\207


0f
of type
that
T, the statement \316\241(\317\207)
given any \317\207
type
}s
is.
In other
of what the exact value of \317\207
true
regardless
words, the
\"if \317\207
is of type
statement is the sameas saying
T, then \320\240(\320\266)
}s
is to let \317\207
Thus the usual way to prove such a statement
true\".
be
like \"Let \317\207
a free variable
of type \316\244
be
something
saying
(by
any
variable

a free

on
depending

T\" means

T\,") and then proving

of type

object

statement becomesfalse if
i.e.,

counterexample,

is

false.

positive
example,

element

an

one

P(x)

can

\317\207
which

produce
lies in

for that

even
\316\244
but

object.
a single

for

which

The
\316\241(\317\207)

the statement \"x2 is greater


than \317\207
for all
instance,
to be false by producinga single
\320\266\"
be shown
can
=
= 1 or \317\207
such
as \317\207
1/2, where x2 is not greater than
For

x.

a singleexamplewhere
%s
P{x)
is
true
for
all
x.
For
P(x)
instance, just
= 2 does
\317\207
because the equation \317\207
+ 3 = 5 has a solutionwhen
not imply that \317\207
+ 3 = 5 for all real numbers x\\ it only
shows
=
\317\207
some
x.
3
true
for
real
number
is
that
5 is
the source
+
(This
somewhat
of the often-quoted,
\"One
inaccurate,
though
slogan
an example\".
cannot prove a statement just by giving
The more
a \"for all\"
statement
prove
precise statement is that onecannot
by
one
\"for
some\"
can
statements
examples,
certainly prove
though
this way, and one can also disprove
\"for all\" statements
by a single
counterexample.)
are in fact no variables \317\207
It occasionally
there
that
of
happens
T. In that case the statement uP(x) is true for all \317\207
of type
type
T\" is vacuously true - it is true but
no content,
has
similar to a
the
vacuous implication.Forinstance,
statement
On

will

true

the

other

not

hand, producing

show that

6 <

2x <

4 for

all

3 <

\317\207
<

and easily proven, but is vacuous.(Such


a
statement
can still be useful in an argument,though
is true,

happen
\"For

this

true
doesn't

very often.)

One can use phrasessuch


of

vacuously

all\",

e.g.,

one

as

\"For

can rephrase

every\"

or \"For each\" instead


= x2 + 2x + 1 for

+ l)2
\"(\320\266

and

Variables

j4.^.

all real numbers x\"


to x2 + 2x + 1\".For

+ I)2 is equal
\320\266,
(\317\207
these
purposes
rephrasings are
be used instead of \"For
V can
thus
all\",
is true\" or \"P(x) is true \\/x G X\" is
P(x)

of logic

the

for instance

\"Vrz

synonymous

with \"P(x)

type

G X

Existential

is true for

quantifiers
there

that

means

T\"

real number

each

\"For

as

equivalent. The symbol

x of

371

quantifiers

all

\317\207
G

X\".

The statement \"\320\240(\320\266)


is true
for some
\316\244
exists at least one \317\207
of type
for

than
although it may be that thereis more
would use a quantifier
such
as \"for exactly
one
of \"for some \320\266\"
if one
wanted
both existence and
instead
one \321\217\"
such
it suffices to
a statement
uniquenessof suchan x.) Toprove
o
f
x.
For
to
show that
such
an
a
instance,
example
single
provide

which

is true,
P(x)
such x.
(One

x2 + 2x

\342\200\224
=

to do is
one needs
8 = 0, for instance
2x \342\200\224
all

but

one
to

freedom

select

\317\207
to

statement;

be

=
\317\207

anything

contrast

is in

this

find

some real
real

single

2 will

to use

need

doesn't

0 for

number \317\207

number

do. (One

\317\207
for

which

x2 +

=
could alsouse\317\207
\342\200\2244,

both.) Note that one has the


one wants when proving a for-some
to proving a for-all statement,where

(One can compare the two


between you and an opponent.
games
by
the
\317\207
and
then
the
In
first
is,
opponent gets to pick what
game,
you have to prove P(x)\\ if you can always win this game, then
have
that P(x) is true for all x. In the second game,
proven
you
one

has

let

to

statements

\317\207
be

thinking

what

then you prove P{x)\\ if you can


and
for
some
game, you have proven that P(x) is true
x.)
is
for
\317\207
is
true
all
much
stronger
Usually, saying something
x.
than just saying it is true for some
There
is one exception
if
\317\207
is
to
the
condition
on
impossible
though,
satisfy, then the
the for-some statement
is
for-all statement is vacuously
but
true,
you

get

to

choose

arbitrary.

of two

\317\207
is,

win this

false. For instance

6<
is true,

2x

<

4 for

all 3

<

\317\207
<

but
6 < 2x < 4 for

some

3 <

\317\207
<

is

basics of mathematicalloqic

A. Appendix: the

372

false.

use phrases

can

One

...

exists

such as

at

\"For

or \"There

one\"

least

some\". For

of \"For

instead

that\"

such

instance, onecan

some real number x\"as \"Thereexists


x2 + 2x \342\200\224
\317\207
8 = 0\". The symbol 3
a real number
such
that
can
thus for instance
... such that\",
be used insteadof \"There
exists
\"3x \342\202\254
X : P(x)
is true\" is synonymous with \"P(x) is true for
\342\200\224
=

\"x2+2x

rephrase

some

\317\207
X\".
\342\202\254

A.5

Nested

quantifiers

nest

One can

or

two

more

For

number x, there existsa

every positive
number

positive

There exists
is

each

positive

To

the

continue

where

your

always win
have

you

proven

that

such
\321\203

the

y2

gaming
first

opponent

a positive

pick

such
\321\203

suppose you
a positive number x,

metaphor,
picks

The

of \"All swans

negation

white\", but rather \"There


white\". Similarly, the negation of
not

cos(x)
\"For

>
every

root of
that

\317\207

every

play a game

and

then

you

x. If you can
number y. You win the gameif y2
of
what
regardless
your opponent does,then
game
that for every positive x, thereexistsa positive
= x.

are not

have

\317\207

Negating a universalstatementproduces
statement.

y2 =

that

number

number

positive

a positivenumber

words, one can find a positive


square
x. So the statementis saying
has a positive squareroot.

In other

true.

for

for each

that

means

It

x.

statement

x, the

number

y2 =

that

such
\321\203

What does this statementmean?


positive

For instance,

together.

quantifiers

statement

the

consider

0 for

is

\"For

0 <

\317\207
<

0\"

some
we
\317\200/2,

0 <
have

existential

an

are

white\"

is

is

some

swan

\"For
\317\207
<

every
\317\200/2,

cos(x)

we

not

\"All swans

is not

which
<

have

< 0\".

\317\207
<

\317\200/2,

cos(x)

we

< 0,

Nested

\316\221.5.

statement producesa universal


exists a black swan\"
is not

an existential

Negating
The

statement.

373

quantifiers

of \"There

negation

rather
exists a swan which is non-black\",but
non-black\". Similarly, the negationof \"There

that

such
number \317\207

+
\317\2072
+ \317\207
\320\2662+\320\266
+

\"or\"

and

set
of

value

be
this

x\\

1 =

\"For

is

0\"

a real

exists

\"There

number

real

every

number

\317\207
such

you want, and P(x) will be true for

anything
is what

\"for

for instance

Thus

all\"

means.

+ 1

for all real numbers

are

a real

exists

similar
to how
(The situation here is very
with respect to negations.)
that
\316\241
a statement
is true
for all \320\266,
then
(\317\207)

know
\317\207
to

\317\207
+

not

0\",
\317\206

1^0\".
behave

If you
can

\317\2072
+

\"There

swans

\"All

\321\217,

that
\"and\"

you
that

if you

know that

(x + l)2 =
then

can

you

+ 2x

x2

+
(\317\200

(cos(y)+ l)2 =
(because

if

universal

statements

is
\321\203

P(x)

get

by

statements,

just

gets to
Remark
studied
totlean

pick

\342\200\224
=

that

numbers

\321\203

can

you

you,

in any

\342\200\224
\317\2002
8 =
+ 2\317\200

x2 +

2x \342\200\224
8 = 0

make

you don't

P(x)

Existential

wish.

if you

some real

get to pick which

you don't
\317\207
for

0 for

substitute

that

gaming metaphor,

real

all

for

\317\207
you

more limited;

simply

conclude

that

+ 1

cos(y)

for whatever

are

conclude

still

course
it's

1,

Thus
cos(y) is also real), and so forth.
are very versatile in their applicability - you

to hold

contrast,

and
\317\200,

e.g.,

\317\2002
+
+ 2\317\200

then

real,

cannot

you

+ 2

cos(y)2

x2 + 2x
then

l)2

instance that

or for

can

that

for instance

conclude

\320\266,

know that

number \317\207
real number you
0. However,

for some
\317\207
it

hold,

is.

of

real number \321\217,

(To continue

but your

get to choose for

wish,

you can

the

opponent

yourself.)

history of logic,quantifiers wereformally


thousands
of years before Boolean logic was. Indeed,Arisof course by Aristotle (384BC- 322BC)
logic,
developed
A.5.1.

In the

A.

374

and his school,dealswith objects,


such as \"for all\" and \"for some\".
Aristotlean

in

syllogism)

their

A typical

goes like this:

logic

man. Hence,Socratesis mortal\".


of mathematical logic,but is not

lacks the conceptof logicalconnectives

(although

is

\"not\"

as =

such

and

allowed),

real

we

equivalent

logically

(a+b)2

two

The
=

\"there

mortal.
logic

is

as expressive
as

because it
or, or if-then

and,

the concept of a binary

also lacks

numbers

have

(a +

not make

Swapping two

\"for

all\"

such as
real numbers b,

for all

and

a,

or may

may

quantifiers

statement.

b)2 = a2 +

2ab + b2

statement

to the

For all realnumbers


b, and
we have (a + b)2 =
(why?

are

men

Aristotlean

such

a statement

is harmless:

Forall
is

line of reasoning (or


\"All

or <.

Swapping the orderof two


a differenceto the truth of a
quantifiers

and quantifiers

properties,

is a

Socrates
a subset

relation

of mathematical logic

basics

the

Appendix:

real numbers

for all

2ab +

a2 +

a,

b2

to do with whetherthe identity


true or not). Similarly,
a2+2ab+b2
swapping
exists\" quantifiers has no effect:
reason

Thereexistsa

real

has nothing
is actually

number

a,

such

that

exists a

there

and

a2 + b2

real numberb,

= 0

is logicallyequivalent to

Thereexistsa

real

b, and

number
such

that

a2 +

b2 =

However,swapping a \"for all\" with


lot of difference.Considerthe following
(a)

For every

than n.

exists a

there

integer n, there existsan

real numbera,

0.
exists\" makes a

a \"there
two

integer

statements:
m

which

is larger

6.
\316\221.

(b)

Some

There

exists

an

every

integer

n.

(a) is

Statement

an integer n,

can

you

that m is largerthan

m such

if your

true:

obviously

opponent

an integer

find

always

(b) is false:

Statement

But

n.

than

integer

375

and quantifiers

of proofs

examples

if

you

\316\267
for

hands you

m which is larger
then

m first,

choose

than
n; your
every
integer
you cannot ensure that m is larger
\316\267
than
m to defeat
opponent can easily pick a number
bigger
in
is that
that. The crucialdifferencebetweenthe two statements
the
\316\267
was
m
chosen
and
could
Statement (a),
integer
first,
integer
in a manner
then be chosen
on n; but in Statement (b),
depending
was forced
to choose m first, without knowingin advancewhat
one
is going
to be. In short, the reasonwhy
order
the
of quantifiers
\316\267
on
is important is that the inner variablesmay
possibly
depend
the outer variables, but not vice versa.

What

A.5.1.

Exercise

which of them

of the

does each

are true? Can you

following

find

gaming

statements

and

mean,

for each

metaphors

of

thesestatements?

(a) For

every

have y2 =

(b) Thereexistsa positive


number

y,

we have

y, such

(d) For every


that

y2 =

that

A.6

for every

number

\317\207
such

that

number

x, and

there exists

number

positive

and

we

y,

positive

a positive

?/, there

exists a positive number

\317\207
such

x.
have

y2 =

number

such
\321\203

that

for every

positive

x.

Some examples of proofs and quantifiers

we give
some simple examples of proofs involving
\"there exists\" quantifiers. The resultsthemselves
the
but you should pay attention instead to how
and
how
the
are
structured.
proofs
arranged

Here

number

y2 = x.

(e) Thereexistsa positive


we
number \320\266,

every positive

y2 = x.

(c) Thereexistsa positive


number

x, and

number

positive

x.

the
are

quantifiers

\"for

all\"

simple,

are

A.

376

A.6.1. For

Proposition
that

exists

there

\316\265
>

every

of mathematical logic

basics

the

Appendix:

0 such

\316\261
\316\264
>

2\316\264
< \316\265.

Let

Proof.

\316\265
>

that

\316\264
>

0 such

:=
\316\264

\316\265/3

will

something

every

because you
what

has
\316\265

to

on
\316\265;

the

be arbitrary,
other hand,

need

to show

you want. Note

also that

only

inside

is nested

quantifier

before

\316\264
first

being given
it
because

statement,

e.g.,

proceeds

selecting

by

that

for

there

that

\"Prove

\316\265.
However,

the

=
2\316\264

be

\316\264
can

\316\264
can

\316\265.

\316\240

wish,

\316\264
which

does
the

because
\316\265,

the

would

you

it is

\316\264-

were

quantifiers

\"There exists a

to prove
case

as you

chosen
on

If

\316\265-quantifier.

this

<

exists a

depend

2\316\264
then
< \316\265\",

In
\316\265.

2\316\265/3

because we are proving

that there

asked

\316\264
>

have to

select

impossible to

prove

is false

(why?).
\"Thereexists...\"statement,
an \316\265
exists
that X is true\", one
> 0 such
and
then
\316\265
carefully,
showing that X is true
this
sometimes
requires a lot of foresight,
has to

one

when

Normally,

has

one then

reversed, i.e., if you were


such that for every \316\265
> 0,
the

need

only

since

work,

to show that thereexists


to pick one such 5; choosing

We have

arbitrary.

We
2\316\264
< \316\265.

how

Notice
for

0 be

prove a

of \316\265
until
later
in the
and it is legitimateto defer the selection
it
becomes
clearer
what
when
\316\265
needs
to
argument,
properties
The only thing to watch out for is to make sure that \316\265
satisfy.
not
does
on any of the bound variablesnestedinside
X.
depend
instance:

For

A.6.2.

Proposition

for all0 <


We

Proof

There exists an \316\265


> 0

> 0 to be
pick \316\265
of sin(rr)
derivative

Since

the

value

theorem

sinfrr)

it

0 <

some

suffice

to

<

\321\203

x.

Thus

to ensure
ensure
that 0

suffice

would

chosen later, and let 0 <


is cos(rr), we see from
the

\317\207
<

\316\265.

mean-

we have

\342\200\224z~
\317\207

for

> x/2

that sin(rr)

such

\317\207
< \316\265.

sinfrr)

\342\200\224

sin(0)

\342\200\224
\342\200\224:~n\342\200\224
\317\207 U

in order

\320\247

cos(y)

to ensure that

sin(rr)

>

\320\266/2,

that cos(y) > 1/2. Todothis,it would


< \321\203< \317\200/3 (since
the
cosine
function

377

\320\220\320\233.
Equality

0, takesthe

the value of 1 at
in between).
decreasing

takes

0 <

that

see

0 <

< \317\200/3 as
\321\203

0 <

all

for

at

1/2

and

\317\200/\320\227,

<

\317\207
and

:=
\316\265

pick

\317\200/3,

\317\207
<

we

then

is

we
\316\265,

have

can ensure that sin(x) >

so we

and

desired,

<

\321\203

x/2

\317\207
< \316\265.

of

the value

that

Note

if we

\316\265.Thus

<
\321\203

of

value

Since 0 <

at the end did not


picked
This
makes the above
y.

we

\316\265
that

depend on the nested


argument legitimate.Indeed,
have to postpone

\317\207
and

variables

can

we

it so

rearrange

that we don't

anything:

We

Proof.

that
be

all

for

choose
0 <

\316\265
:=

\317\207
<

we
\317\200/3,

sm(x) =
0 <

some
0 <

have

<

\321\203

<
\321\203

\317\200/3.

Thus

\342\200\224

sin(0)

<

inside

A.7

Equality

some

to show
\317\207
<

\317\200/3

\321\207

cos^)
0

\317\207
and

<

\317\207
<

1/2,

\316\260\316\270\316\262(\317\200/3)

we

have

on

\317\207
and

the

outer

\317\200/3,

since

sin(a:)/a:

as desired.

As mentioned before,onecan
3 +
expressions (such as 2 \317\207
by

So let 0 <

then
\321\203

variable

the

we

cos is
> 1/2
D

argument

are
and x, \321\203

it.

nested

obeys

<

Thus
[\316\237,\317\200/\320\227].

If we had chosen\316\265
to
depend
is
not be valid, because\316\265

expression

\321\203

>

cos(y)

would

related

we have

Now

theorem we have

\316\267
\342\200\224
\317\207 U

\317\207.Since

decreasing on the interval


and hencesin(rr)
> x/2

0.

> x/2.

sin(rr)

sin(rr)
\342\200\224'

\342\200\224\320\2231
\317\207

for

\316\265
>

clearly

have

mean-value

By the

arbitrary.

\317\200/3;

with
statements
by starting
an
then
whether
and
asking
5)
a certain
or whether two expressions are
property,
are many
sort
of relation
There
etc.).
(=, <, \342\202\254,
it
is worth
the most importantoneis equality,
and
create

relations, but
this
concept.
spending a little time reviewing
two
of the
same type
is
a
relation
\320\266,
\321\203
Equality
objects
linking
\316\244
Given
two integers,
or two matrices, or two vectors,
etc.).
(e.g.,
=
or
\317\207
two such objects \317\207
and
the
statement
\321\203
may not
y,
may

A. Appendix: the basicsof mathematical


logic

378
be

it

true;

on the value of

depends

equality is defined for

\317\207
and

on

also

and
\321\203

how

under
of objects
the
two numbers

consideration.
0.9999... and
with
modulus
10 (in which
modulo 10), the
their
remainders
12 = 2, even though this
equal,

class

the

For instance, as real numbers,


1 are equal. In modular
arithmetic
to
numbers are considered
equal
numbers 12and 2 are considered
is not the casein ordinary
arithmetic.
of objects
How equality is defined depends onthe class\316\244
under
of definition.
and to some extent is just a matter
consideration,
for the purposes of logic we requirethat equality
However,
obeys
the following
four axioms of equality:
\342\200\242

\342\200\242

type,

axiom).
=
if \317\207

type,

axiom).
=
if \317\207

(Symmetry

same
\342\200\242

(Transitive

same
\342\200\242

same

if

typej

operations/.
=
if \317\207
\320\266,

we

some

give

= 2y,

real

number

know

\317\207
and

x.

of
\321\203

the

\316\266
of
\320\266,
\321\203,

the

\321\203 \317\207.

any three

Given

=
\321\203

and
\321\203

then
\316\266,

then
\321\203,

f(x)
for

P(x) and

objects
=
\317\207

\316\266.

any two

Given

any

and
of the
\321\203
objects \317\207
f(y) for all functions or
property
P(x) depending on

P(y) are equivalent

statements.

axioms are

that

examples.

z.

A.7.2. Let \316\267


and

Example

then m

objects

=
\317\207

=
A.7.1. Let \317\207
and
If \317\207
then
be
real
numbers.
\321\203
\321\203,
=
=
\317\207
\316\266
for
and sin(rr)
+ \316\266 any
\321\203
sin(y). Furthermore, +

Example
2x

any two

have

clear, together, they assert


To illustratethe substitution
relation.
equivalence

is an

equality

we
\320\266,

then
\321\203,

Similarly,

y, then

three

first

any object

Given

axiom).
=
\317\207

(Substitution

The

Given

axiom).

(Reflexive

that

be

also

must

\316\267
and
> \320\272

odd.
n =

Example

A.7.3. Let

=
\317\207

and

sin(y)

\321\203

m be

integers.

If

is
\316\267

have a third
then
we also know
m,
If we

and

integer

fc,

that

n =

m,

and

we

> k.

If we know that
real
numbers.
the
substitution
axiom)we have
(by

\316\266
be
\320\266,
\321\203,

z2, then

odd

379

\320\220\320\233.
Equality

= sin(z2),

sin(y)
=
\317\207

and hence

(by

transitive

the

axiom)

we have

sin(22).

Thus, from the point of view


we
on a classof objectshowever

of

logic,

we can
so long

please,

define equality
as it obeys the

and
transitive
and is consistent with
axioms,
reflexive,
class
of objects
under discussion in
all other operationson
was
true
for all of those
the sense that the substitution
one day to modify the
operations. For instance, we decided
12 was
now equal to 2, one could
do so if one
integersso
now
to
and
that
also made sure that 2
equal
12,
/(2) = /(12)
we
operation
/ on these modified integers. For
any
2 + 5 to be equal to 12+ 5.
now need
this
this
case,
pursuing
symmetry,

the

axiom

if

that

only

was

for

instance,

(In

line of

reasoning will
modulus 10.)

Exercise
A.7.1.

Suppose

know

a+

a =

that

d= b

c.

and

lead

eventually

you

=
\321\201

have

d. Use

to modular

arithmetic

with

four real numbers a,b,c,d and you


the above four axioms
to deduce
that

\320\222

Chapter

decimal

the

Appendix:

system

the basic number


In Chapters2,4, 5 we painstakingly
constructed
the
of
mathematics:
natural
numbers,
integers,
systems
rationale,
to exist,
and reals. Natural numbers were simply
and
postulated
the integers
then came via (formal) differences
to obey five axioms;
came
from
of the natural numbers; the rationale
then
(formal)
came
from
quotients of the integers,and the realsthen
(formal)
limits of the rationale.
This is all very well and good, but it does seem somewhat

alien to one's
very little use

experience

prior

0,1,2,3,4,5,6,7,8,9
Indeed, except for

to
the

main

the

reason

and

years

of

decimal

and the latter two

to
we

use,

have

but

system itself is
convenient
for
very
grown accustomed to it thanks to a thousand
in the history of mathematics it is actually a
this

for

mathematics.

is that

early

numbers

civilizations

only

relied

dates

the decimal

It is

comparatively recent invention.


about ten thousand years
(starting
walls), but the modernHindu-Arabic
representing

as

rewritten

be

can

(0++)++.

essential

computations,

number

and 2,

0,1,

The basic
not

are

construction,

numbers

0++ and

In particular,
numbers.
system, in which the digits
to represent
these numbers.
combined
of examples
which were not essential
the only decimals we really used were
these

with

of the

made

was

from

on other

have

Numbers

from

scratch

base

the 11th

bases;

for

10

been

marks
system

on cave
for

century or so.
instance

the

for

around

Some

Babyloni-

The
\316\222\316\233.

decimal

of natural

representation

ans used a base60

381

in our

time system

survives

still

(which

system

of hours,minutes,and seconds,and

numbers

in

our

of

system

angular

Greeks were able


minutes,
degrees,
seconds).
the fact that the most
to doquite advanced
mathematics,
despite
number
available
to them was the
advanced
representation
system
numeral
which
was
horrendous
Roman
system
J, //, III, IV,...,
of even two-digit numbers. And
of course
for computations
on binary, hexadecimal,
relies
or byte-based (base
modern computing
And

and

256) arithmetic
do
as the slide
In
at
all.
system

of

instead

the ancient

while

decimal,

such

computers

analog

number representation
really
now
that
can do the menial
fact,
computers
is very little use for decimals
of number-crunching,
in
work
there
mathematics.
we
use
numbers
other
modern
Indeed,
any
rarely
numbers or one-digit fractions(as well as e, \317\200,\320\263)
than
one-digit
in modern
mathematical
work; any more complicated
explicitly
numbers
called
more
usually
get
generic names such as n.
of decimals
does deserve an
the
Nevertheless,
subject
so
is
to
because
it
the
we
use mathematics
in our
appendix,
integral
way
everyday life, and alsobecausewe do want to use such notation as
3.14159... to refer to real numbers, as opposed to the far clunkier
rule

rely on any

not

:= 3.14, \320\260\320\267
:= 3.141,...\".
how
decimal
the
begin by reviewing
systemworks
that
and then turn to the reals. Note
integers,
positive
from
earlier
discussion we shall freely use all the results
where

\"LIMn_>ooan,

a\\

3.1, a2

We

The decimal

B.l
In

this

we will

section

a x b as ab, since
be misconstrued

Definition B.l.l (Digits). digit

this

decimals

is any

such

one of

as

the ten

34 might

symbols

We equate these digits with natural numbers by


=
the
0
0, 1 = 0++, 2 =
way
up to
also
the number
ten by the formula ten :=
define

9.

0,1,2,3,...,
the

in

chapters.

usual conventionof abbreviating

this would meanthat


as 3 x 4.
A

the

of natural numbers

representation
avoid the

for

1-H-,etc.all

formulae

9 = 8-H-.

We

9++.

cannot

(We

use

the

decimal

notation

10 to denoteten yet,

382

the

\320\222.Appendix:

decimal

presumes knowledgeofthe decimalsystem

because that

system

and

would

be circular.)

B.1.2 (Positive integerdecimals).A positive


integer
>
of
\316\267
where
0 is a
\316\261\316\267\316\261\316\267_\316\271...
\316\261\316\277
string
digits,
is
natural number, and the first digit
non-zero.
an
Thus, for
3049
is a positive
instance,
integer decimal, but 0493or 0 is not.
each
We
equate
positive
integer decimal with a positive integerby
Definition

is any

decimal

formula

the

\316\267

=
\317\207
ten1.
\316\261\316\267\316\261\316\267-\316\271...
\316\261\316\271
\316\261\316\277
\320\243,

i=0

B.1.3.

Remark

Note in particular that this

definition

implies

that

10 =

and thus

\317\207
ten0

we can write ten as the

digit integerdecimalis exactly

decimal3

the

by

above

definition

familiar

more
equal

\317\207
ten1

that

to

is equal

3 = 3 x ten0

ten
10.

digit

Also, a

we

Now

integers.

positive
positive

show

decimal

consists

that
It is

digit,
not

of natural

by definition.

Theorem

B.1.4 (Uniqueness and existence


of decimal
is
to
m
Every
positive
equal
integer
exactly one positive

integer

one

gives a positive integer, since the sum


numbers, and the last term antenn

non-zero

representations).

and

this decimal system indeedrepresents


the
clear from the definition that every

representation
entirely

the

to

is no possibility of confusionbetweena single


and
a single digit decimal. (This is a subtledistinction,
which is worth losing much sleepover.)
so there

single

itself, e.g.,

is

decimal.

use the principleof strong induction


(Proposition
with mo := 1). For any
2.2.14,
m, let P(m) denote
positive
integer
the statement \"m is equal to exactly one positive integer decimal\".
Proof.

We shall

B.I.

decimal

The

Suppose we already

m' <

know

First observethat eitherm >


(This is easily proved by ordinary
m

consisting
which
decimal

of

consisting

such a

two

or m G

ten

{1,2,3,4,5,6,7,8,9}.
first that

Suppose

induction.)

a
of a single digit, and thereis
is equal to m. Furthermore,
no
can
since
On...
equal m,
digits

decimal

single-digit

positive integers

P(m).
is equal to

m clearly

Then

{1,2,3,4,5,6,7,8,9}.

integer

for all

true

is

P{m!)

to prove

wish

now

we

m;

383

numbers

of natural

representation

positive

one

only
decimal

or more

if

decimal (with \316\267


> 0)

ao is

we have

\316\267

ao = Y^

an...

>

\317\207
ten1
\316\261\317\212

Now suppose that m > ten.Then


(Proposition 2.3.9) we can write

m= s
where

5 is

> m.

Euclidean

the

by

algorithm

+ r

\317\207
ten

a positive integer,and

> ten

\317\207
ten1

an

{0,1,2,3,4,5,6,7,8,9}.

Since

s <

s x ten < s \317\207


ten

we can usethe stronginduction


is true. In particular,s has

+ r

and

hypothesis
a

decimal

= m
conclude

that

representation
\317\201

s =

bo =

bp...

2J ^ x ten1.
i=0

ten,

by

Multiplying

that

we see

\317\201

\317\207
ten

x tenl+1

^2h

bp...\320\254\320\276\320\236,

i=0

and

then

adding

r we

see that
\317\201

= s

\317\207
ten

+ r

Y^bi

i=0

\317\207
ten1\"1\"1

+r

bp... for.

P(s)

384

has at least one decimal


representation.
that m has at mostone decimal

Thus m
to show
Suppose

the

\320\222.Appendix:

system

Now

we need

representation.

that we have at

of contradiction

sake

for

decimal

least

two

different

representations

m =

an

...

by the previous

First observe

a0.

an,...

that

computation

ao =

an...

clq

(an... a\\) x ten + ao

and
\302\267
\302\267 =
\302\267

anf

ao

and so after somealgebrawe


CLq

\342\200\224=
CiO

The right-hand side


lies strictly between

equal to 0.
But

by

integer

than an...

and
\342\200\224ten

has

X ^en*

\302\267
\302\267
\302\267

al)

Thus

+ten.

the

both

left-hand

the
one

\316\267
and

strong
decimal
must

a\302\243

side

sides must

a!0 and an.. .a\\


we know that an... a\\

only

equal

be

= a'n,...a'v
is

a smaller

induction

hypothesis,
which
representation,
for
all
i =
ai
equal

an ... ao and o!n,...a!0


the assumption that they were

decimals

contradicting

identical,

anf

ao =

that

Thus
\316\261\316\277\302\267
by

the

Thus

+ ao

is a multipleof ten, while

the number an... ao


means that nf must

1,..., n.

x ten

obtain

arguments,

previous

ai)

~~
\302\267
\302\267
\302\267
al
{\316\270\"\317\200

means

This

\302\267
\302\267
\302\267

(anf

are

in

fact

different.

to the decimalgiven

We refer

decimal representationof m. Once


representation,

long

can

one

\317\207 to
\317\207
\320\276\321\202
\321\203

to

that

of

one

the
has

above
this

theorem

as the

decimal

the usual laws of long addition and


connect the decimal representation of \317\207
+ \321\203

derive

then

multiplication

by

\320\266
or

\321\203
(Exercise

\320\222.

1.1).

representation of positive integers,one


canofcourse represent
negative
decimally as well by using
integers
the \342\200\224
we
a
as well. This gives
let
0
be
decimal
sign.
Finally,
the
decimal
of
all
is then
rational
representations
integers. Every
ratio of two decimals, e.g.,335/113
or \342\200\2241/2(with the denominator
to
be
of
be more
there
required
non-zero, course), though
may
Once

one

has decimal

decimal

The

\320\222.2.

than one

to

way

3/2.
as

ten

Since

10 insteadof ten throughout,

now use

10, we will

a ratio, e.g.,6/4 =

as such

a rational

represent

385

numbers

of real

representation

is customary.

The purposeof

Exercise B.l.l.

procedure

of

addition

long

when

\320\263
>

a4

\320\260\320\267
0,

recursively

ai +

bm...

bo

a* =

that

that

+ Si

< 10, we

+ bi

has
\316\265*

> 10,
\316\265\316\271

number

(The

to the

if A =

demonstrate that the


school

elementary
be

positive

0 when

is actually

decimals.

integer

6* =

\320\263
and
>
\316\267,

372, then

'\302\267=
0.
\316\265\316\277

suppose
bi

if ai

=
\320\222

in

ao
m;
2, ax = 7, a<i = 3,
= 0, and so forth.
the numbers
Define
Co, C\\,... and \316\265\316\277,
\316\265\\,...
addition
the
by
algorithm.
following long

for instance,

\342\200\242
set
We

\342\200\242
Now

to you

taught

valid. Let A = an... ao and


Let us adopt the convention

is to

exercise

this

set

c*

we set

is the
\316\265*+\316\271

decimal

+ l)th
(\320\263

:=

defined for some

been

already

6*+

a* +
:=
c\302\273

and

z%

+ 6*
a\302\273

\342\200\224
\320\256
anc^
\316\265\302\273

from the

\"carry digit\"

:= 0;

Ei+\\
ith

> 0.
\320\263

If

otherwise,
= 1\302\267
\316\265\316\220+\316\271

decimal

place

place.)

Prove that the numbers co,Ci,... are all digits,


and
there
that
I such that q \317\206
an
0 and
show that
c* = 0 for all % > l. Then
cjq_i ... CiCq is the decimal representation of A + B.
this algorithm
Note that one could in fact
use
to define addition,
and
to
but it would look extremely
complicated,
prove even such simple
be rather
This is one of
facts as (a + b) + c = a+(b + c) would
difficult.
we
decimal
in
construction
the reasons
have
avoidedthe
our
system
why
of the natural numbers. The procedurefor long
multiplication
(or long
is even
worse to lay out rigourously; we
subtraction, or long division)
will
not do so here.
exists

We

of real numbers

The decimal representation

B.2
need

a new

symbol:

DefinitionB.2.1
of digits, and a

(Real

decimal

the decimal
decimals).
point,

\302\261an..

point \".\".

A real
arranged

.ao-a-iQ>-2

decimal

as
\302\267
- \302\267

is any

sequence

386

which is finite to the


but infinite
number),
is

+ or

either
a

either

to

to the right

and
\342\200\224,

of

(so

point
decimal

the

\316\267
is

point,

system
a natural

where

\302\261

natural number decimal(i.e.


or 0). This decimalis equated

ao is a

an...

integer

positive

decimal,

number

real

the

decimal

the

of

left

decimal

the

\320\222.Appendix:

\316\267

ai x

... = \302\2611\317\207
\302\261an... \316\261\316\277.\316\261_\316\271\316\261_2

^2

'
\320\256*.

i=\342\200\224oo

is always

series

The

that

show

(Exercise B.2.1). Next, we


has at least one decimal

convergent

real number

every

representation:

B.2.2

Theorem
number

real

(Existence
at

has
\317\207

least

of decimal representations).
one decimal representation

Every

= \302\261\316\261\316\267
...
\317\207

ao.a-ia-2

0.000

Also,

sign

get

numbers

Let

\302\261.
Thus
\317\207

0 be

(Corollary

property

that

such
number \316\234

we

find

0 has

the decimal

representation

a decimal

5.4.4).

Proposition

(by

>
\316\267

=
\317\207

representation for \320\266,we


a decimal representation for \342\200\224x
by
changing
it suffices
to prove the theorem for positive
real

once

automatically

the

note that

We first

Proof.

any natural number.


we know that
5.4.13)
\316\234\317\207
10~\316\267 >

Prom the Archimedean


there is a natural

0 x

\320\266.
Since

10~n <

we
\320\266,

thus

see

< \317\207
that sn \317\207
10~n
exist a natural number
sn such
x
and
one
x.
no
such
natural
10~n
number
>
sn-Hexisted,
(If
< \317\207
s \317\207
could use induction to conclude
that
10~n
for
all natural
numbers
the
Archimedean
s, contradicting
property.)

that

there must

consider

Now

we

thus

the

sequence

sn x

10\"n <

so,

Since

\302\253i,
\302\2532?

\317\207
<

(sn

\317\207
<

(10 x

+1)

we

have

\317\207
10\"n

have

(10 x

sn)

\317\207
10\"(n++)

<

sn +

10) x

10_(n++).

decimal

The

B.2.

On the

other

sn+1 x

10-(n+1) <

10 \317\207
sn
these two

Prom

\317\207
<

+ 1)

(sn+1

\317\207
l(T(n+1)

we have

hence

and

have

we

hand,

387

numbers

of real

representation

<

inequalities we seethat we
10 x sn < sn+i < 10 \317\207
sn

we can

and hence

10

and sn+i <

+ 1

sn+i

find

+ 10.

have

+ 9

such
that
\316\261\316\267+\316\271

digit

= 10

sn+i

sn

+ an

\317\207
sn

and hence

=
sn+1 x 10\"(n+1)
and

\320\221\302\245\320\276\321\202
this
identity

\317\207
10\"(n+1).
\316\261\316\267+1

\317\207
10\"\316\267+

sn

we can

induction,

obtain the formula

\316\267

sn

x 10\"n

= s0

+ J^ai

\317\207
10_i.

2=0

limits of both sides (using

we take

Now

Exercise

B.2.1)

to obtain

oo

lim sn

\317\207
10\"n

so +

y^a,iX

10\"\320\263.

2=0

On

hand, we have

other

the

\317\207

for

all

n, so

by the

10_n

< sn

x 10_n <

squeeze test (Corollary


6.4.14)we
=

\317\207
lim
10~n
\320\277-\321\216\320\276

sn

Thus

we

\317\207

have

have

x.

oo

x =

so + ^^\320\263

\320\256~\320\263.

2=0

Since
Theorem

so already
B.1.4,

as desired.

has a positive integerdecimalrepresentation


has
a decimal
we thus see that \317\207
representation

by

388
There is however

possible

one slightflaw

Proposition B.2.3
representations).

1.000...
compute

1 has two

decimal

system

system:

it is

two decimal representations.


of decimal

uniqueness

different decimal

representations:

0.999....

and

The

Proof.

of

(Failure

number

The

the

with

to have

number

real

one

for

decimal

the

\320\222.Appendix:

representation

0.999

definition,

By

1 = 1.000... is clear. Now


this is the limit of the

let's
Cauchy

sequence

0.9,0.99,0.999,0.9999,....

has 1 as a formal

this sequence

But

It turns out that theseare


of 1 (ExerciseB.2.2). In fact,

have either oneor two

decimal

a terminating

and

decimal,

limit

the

only
as

it

two decimal
turns

out,
-

representations
one

otherwise

Exercise B.2.1. If an...

...
\316\261\316\277.\316\261_\316\271\316\261_2
absolutely

Exercise B.2.2.

is a

Show that
1=

of

1 are

1=

the

only

\302\261an...

representations
all real numbers
two if the real is

(Exercise

series

\317\207
10\320\263
is
\316\261\316\271
\316\243\342\204\242=_00

5.2.8.

Proposition

by

B.2.3).

real decimal, show that

the

convergent.
decimal
ao.a-ia-2

representations
\302\267
\302\267
\302\267

1.000... and 1 = 0.999....

\317\207
is said to be a terminating decimalif we
integers n, m. Show that if a: is a terminating
has
is not
two decimal representations, while if \317\207
decimal, then \317\207
exactly
at terminating
has
one decimal representation.
decimal, then \317\207
exactly
Exercise B.2.4. Rewrite
of Corollary 8.3.4 using the decimal
the
proof

B.2.3.

Exercise

have

=
\317\207
n/10~m

system.

A real

number

for some

Index

56

18,

-\316\227-(increment),

absorption laws, 52

87

on integers,

abstraction,24-25,390

+C, 342
334

\316\261-length,

addition

244

\316\265-adherent, 161,

contually

385

long,

\316\265-adherent,

of functions, 252

of complex

115

eventual,

of integers,

of natural

local, 253

147

115,

eventually
146

576,

\316\265-steady,

111,

point
infinite, 286

of sequences:seelimit
of sets:

a posteriori,

a priori, 20

ambient

Abel's

484

theorem,

convergence

test, 192
value

for rationals,
reals,

analysis,

and: see

conjunction

466,

numbers,
129

98

406

to
approximation

absolute
for

space,

antiderivative,340

192, 220

complex

245, 402, 435

alternatingseriestest,193

20

for series,

point

of sequences

595

576,

\317\203-algebra,

measure,

592

adherent

508

\317\200,
506,

additive

(countably)

146

110,

numbers, 27

ofreals,119

reals,146

\316\265-steady,

86

of rationals,93

rationals, 99
sequences,

496

numbers,

253

functions,

for

82

of cardinals,

161

\316\265-close

absolute

617

integrable,

absolutely

498

Archimedian

469,

the

identity,

523

property, 132

arctangent: see
trigonometric
functions

\317\200

INDEX

Aristotlean

of

associativity

of addition in C, 496

of addition

in

of the

of multiplication in N, 34

of scalar

377

52

534

of

67

union,

algebra

268

discontinuity,

asymptotic

Axiom(s)
in mathematics, 24-25

of choice,40,
Axiom

of

Banach-Tarski

seee

377

174

see

Boolean

mathematical

47,

algebra,

591

576,

logic, 367

Boolean

induction

Borel property, 575,596

of

infinity,

50

of

natural

numbers:

see

Peano axioms

of pairwiseunion,
of reflexivity,

377

of regularity,

54

of replacement,

Borel-Cantelli

615

lemma,

bound

180, 369,

variable,

376

boundary (point), 401,

42

435

set, 66

bounded

from

and below,

above

269

function, 269,451
49

244

interval,

of separation,45

sequence,

38,40-42,45,49-

50,54,66

of singleton

formula, 189

binomial

regularity

theory,

62

Bolzano-Weierstrasstheorem,

of induction:

of power

535

vectors,

bijection,

of foundation:seeAxiom

of row

basis

standard

230

choice,

principle of

575,

basis

specification

of

paradox,

590

see

universal

countable

of equality,

400

ball,

baseofthe natural logarithm:

229

73,

of comprehension:

of set

40

of universal specification,

see also: ring, field, laws

of

57, 377

set,

empty

of transitivity,

538

multiplication,

of

377

of substitution,

59-60

addition,

45

specification,

of symmetry,

29

N,

of composition,

of vector

sets, 41

373-374

logic,

sequence away
123,

sets and

pair

113,

set, 248,

127

413

150
from

zero,

III

INDEX

law

245, 403,

closure,

of multiplicationin

limit

N,

in Z,

of multiplication

of multiplication in R,

Cantor's

cocountable

91

coefficient, 476

126

cofinite topology, 437


534

common

of, 81
80

of,

uniqueness

Cartesian

70-71

product,

of addition in

C, 496

of additionin

N,

of convolution,467, 522

criterion,

197

Cauchy

sequence,

111, 146, 409

Cauchy-Schwarz

399,

inequality,

of

of

ordered set

chain rule,293
dimensions,

ring, field,

variables formula,

413, 436

comparisonprinciple

test)

(or

finite

181

series,

for infinite series, 196


for

character,518

166

sequences,

completeness
602

function,

of the

choice

space of

continuousfunctions,

of metric

finite, 73

of the reals, 168

countable,230

of

completion

454

spaces, 410

single, 40

a metric

space,

412

229

arbitrary,

laws

algebra

compactness,

for

346-358
characteristic

34

N,

compact support, 465


552,

555

change of

in

multiplication

see also:

516

higher

spaces,

534

Cauchy

in

29

in vector

of addition

infinite, 228-229

see totally

311

refinement,

commutativity

of finite sets, 80

chain:

topology,

columnvector,

cardinality

arithmetic

point

438

35

224

theorem,

435

cluster point: see

29

in N,

of addition

derivatives

with

556
\320\241,
\320\2410,\320\241\\\320\2412,\320\241\320\272,

cancellation

complex numbers C, 495

closed

box,

complex

580

conjugation,

498

of functions,

59

interval, 243

composition

set, 403, 435

conjunction (and), 354

Clairaut's

theorem:

interchanging

see
derivatives

connectedness,307,430
connected

component,

433

IV

INDEX

constant

de

function, 58, 312

sequence,170
420,

261,

continuity,

Moivre

436

255,

convergence,

point,

degree,

464

559

theorem,

362

dense,

directional,

544

a point,

at

in

matrix,551

total,542,

see pointwise

544

convergence

differenceset, 47

convergence

362

466, 487,

522

differentiability

at a

587

directional,

544

see

in

trigonometric

208
of the integers,212

countability,

rationale,

542

dimensions,

higher

infinite, 478

functions

214

478,

fc-fold,

digit,

556

381

dilation,

536

diophantine, 616

578

see abo:

point, 288

continuous, 556

trigonometric

functions

cover,

matrix

derivative

coset,

ofthe

see

matrix:

differential

corollary, 28

cotangent:

542

rule, 293

difference

uniform

see

of,

uniqueness

uniform:see

cosine:

542,

partial, 546

190

convolution,

dimensions,

higher

544, 546, 551

of sequences, 148,394,434

converse,

countable

465

254, 441

pointwise:

see

denumerable:

derivative, 288

function

series,

integer, 384

real, 385-386

convergence
in L2, 517

of

385

positive

contraction,558

of a

388

representation,

421

hypothesis, 227

contrapositive,

of

non-uniqueness

242

mapping

384

integer,

negative

427

and connectedness,431
continuum,

laws, 47

decimal

and compactness,
and

507

identities,

de Morgan

open cover

critical point,

572

Diracdelta
direct

sum

function,

466

INDEX

function, 58

75, 425

of functions,

discontinuity:see

sequence,
73

singularity

393

discrete

metric,

disjoint

sets, 47

disjunction (or), 354


vs.

inclusive

distance
in C,

354

exclusive,

499

39

equivalence

391

of

laws

of

error-correcting codes,392

497

numbers,

complex

algebra

3, 190

algorithm,

Euclidean

metric,

Euler's

of sequences, 4

Euler's

see also:

35

391
506

formula,

503,

number:

see e

exponential function,490,501

convergence

237

divisibility,

Euclidean

Euclidean space, 391

divergence

of series,

281

116,

sequences,

relation, 378

for natural numbers,34


see also:

58

functions,

of cardinality, 77

law

distributive
for

for

for sets,

145,

R,

579

40, 576,

equality, 377

in Q, 98
in

185

series,

set,

exponentiation

division

by zero,

of

of

exponent

domain, 55

majorize

dominated

exponent

see

convergence:

expression,

variable

101,102

R and
Z,

140

in

Q,

in

R,

and
142

and
177

353

extended real numbersystem

R*, 137,153

e, 491
Egoroff's

exponent

see bound

and

base in R+

with

244

variable:

in

exponent

convergence theorem
dummy

Z,

base in R+

with

dominated

Lebesgue
infinite,

in

with base in

dominate: see

doubly

in Q

base

with

96

of rationale,

exponent

in N, 36

252

functions,

base and

with

formal (//), 93

81

cardinals,

theorem,

617

extremum:

see

maximum,

minimum

empty

Cartesian

product,

73

exterior

(point),

401,

435

VI

INDEX

greatest lowerbound: seeleast

factorial, 189
67

family,

Fatou's

lemma,

524

Fejer kernel,

half-infinite, 244

half-open,243

ordered,

97

intersection

finite set,

property,

80

419

half-space,591
harmonic

fixed point theorem,276,559


forward

see image

image:

Hausdorff

Heine-Borel

inversion

520

formula,

520

periods,

531

536

theorem,

518

Fubini's

theorem,
finite

515

516,

hypersurface,

568

if:

see

iff

(if and

see also:

30

only if),

image

188

of

series, 217

interchanging

with

integrals/sums

implication

ill-defined,351,353

624

series,

for infinite

535

homogeneity,

368

frequency,

for

Hermitian

536

fractional part, 512


variable,

414

theorem,

real line, 248

for the

identity map (or operator),63,

520

transform,

free

240

form,

for arbitrary

series

437, 438

space,

Hausdorff maximality
principle,

coefficients, 520

199

series,

Fourier

series,

559

ball theorem,

hairy

field, 95
finite

bound

upper

614

64

sets,

inverse

image, 65

imaginary, 498

implication

357

(if),

integrals/sums

definition,

implicit

57

fundamental theorems of

implicit function

338,

theorem, 568

improperintegral,318
inclusion

calculus,

568

differentiation,

implicit

55

function,

63

map,

341

227, 228,

inconsistent,

502

geometric

index of summation:seedummy

geodesic,394

index

series, 190, 196


formula, 197, 200,460
550

gradient,

graph,

58,

75, 251, 568

variable

set,

indicator

67

function: see
characteristic

function

VII

INDEX

induced

with

limits

metric, 391,

407

with

limits

of

Principle

limits with sums, 617

infinite

244

interval,

with

sums

sums with
with

sums

function

intermediate

133, 512

part, 103,

integers

interval, 243

94

integral

test,

intrinsic,413

with

interspersing
rationale,

inverse

103

function

332

piecewise

562

in logic, 362

by parts, 343-345,484
315,

301,

theorem,

image, 65

integration
laws,

theorem,

pairwise, 46

with

identification
rationale,

value

intersection

85

definition,

401, 435

432

274,

\316\226

6, 217

sums,

interior (point),

514

product,

integrals, 459,

476, 613,614,616

injection:seeone-to-one
integer

463,

derivatives,

476

set, 80
infimum: see supremum

inner

9,

8,

limits,

450

induction

mathematical

619

limits with length, 12

topology, 407,435

induction:see

9,

integrals,

462, 610,

of

321

constant,

313,

63

functions,

see bijec-

function:

invertible

tion

315

local, 562

Riemann:see

Riemann

498

involution,
integral

108

irrationality,

interchanging

derivatives

with

derivatives,

10, 556

finite sums with finite sums,

187,188
with

integrals

integrals,

7,

of
isolated

\\/2,

104,

137

247

point,

isometry, 406

jump

268

discontinuity,

614, 624

limits with derivatives,9,

463

l\\

i2,

L1,
/\302\260\302\260,

393,

L2,

391L\302\260\302\260,

516, 617

VIII

INDEX

theorem,610

equivalenceof in finite
396

dimensions,

see

also:

Leibnitz

inte-

absolutely

grable

see also:supremumas norm


11, 303

rule,

L'Hopital's

lemma,

limit
at

laws of algebra
497

limiting

of

122

101,

102,141,143,177,490

least upperbound,134

least

135,

397, 442

limit inferior,seelimit
limit

also supremum
convergence

functions,

619

theorem,

integrable

160, 409

of sets, 247
limit superior,162
linear

535

combination,

linearity

618

of nonnegative

approximate,

541

of convolution, 471,522

functions,

of

608

finite

series,

186

of simple functions,604

of limits,

upper

of infinite series, 194

and

vs.

620

lower,

the Riemann

of

590
Lebesguemeasurable,
measure,

motivation

of, 575-577

315,

321,

of transformations,

535

integration,

606,

577

Lebesgue monotone convergence

151

of inner product,515

integral,

622
Lebesgue

point

of sequences,

integral

of absolutely

256,

148,

of,

limit

superior

158

Lebesgue dominated
Lebesgue

see uniform

uniqueness

bound

upper

property,

444

uniform,

of exponentiation,

see

148

sequences,

pointwise,

algebra

265

right,

values of functions,

5, 254,441

95

laws of arithmetic: see laws


laws

and

left

89

rationale,

of

150,

laws, 150,256,500

496,

numbers,

for reals,

118,

(LIM),

412

for integers,
for

286

infinity,

formal

complex

554

length of interval, 308

label, 67
for

293,

rule,

28

612

Lipschitzconstant,298
Lipschitz

continuous,

298

IX

INDEX

(natural), 492

logarithm

monotone

logical

354

connective,

bound

convergence

159

sequence,

536

identification with linear

transformations,
537-

580

576,

measure,

see function

matrix,

monotone

function, 276,336

572

map:

see Lebesgue

theorem

majorize, 317, 608


manifold,

decreasing)

convergence:

see upper

bound:

lower

or

(increasing

powerseriesof, 460,492

see function

morphism:

moving bump example,446,

614

540
multiplication

296

233,

maximum,

of

local, 296

of functions, 252,271
427

271,

principle,

297

value theorem,

mean

measurability
for

597,

functions,

for sets,

598

590

motivation of, 574

see also:

Lebesgue

measure

outer

measure,

140

meta-proof,

ball: see ball

on C, 499

onR,

391

space,

390

see

of

86

integers,

536, 540

of matrices,

of natural numbers,33

of

94

93,

rationale,

of reals,

120

Natural numbers N

are

80

infinite,

see Peano

axioms

identification with integers,

87

17

definition,

in set

theory:

see Axiom

of infinity

233, 296

local, 296
of a set of
numbers,

numbers, 497

of functions,252

informal

also: distance

minimum,

of complex

axioms:

metric, 390

81

cardinals,

natural

of functions,

252, 271

minorize: see majorize

518
monomial,

negation
in logic,

76

355

of extended reals, 154

of

210

of,

uniqueness

complex

of integers,

numbers,

88

of rationals, 93
of reals,121

497

\317\207

INDEX

of

see negation,

negative:

291,

approximation,

544

differentiable

outer

pair set, 41

function,

ordered

partially

partial sum, 190

primitive, 53
61

function,

correspondence:

Parseval

see

partition,

open

path-connected,

box,

511

periodic,

piecewise

constant, 312

437

topology,

constantRiemann-Stieltjes

70

integral,335

construction of, 74

orderedn-tuple,71

239

lexicographical,

reals,

520,

528

444

convergence,

of series, 456

topologyof, 455
154

integers, 91
of the natural numbers,

31

formula (or

pointwise

of sets,233
of the

principle,

Plancherel

theorem),

of partitions, 310

83

pigeonhole

227

of orderings, 240

extended

330

continuous,

ordering

of cardinals,

512

extension,

orderideal,238

the

see

bijection

or: see disjunction

of

18-21, 23

matching:

243

pair,

432

perfect

414

set, 403

ordered

308

Peano axioms,

578

order

531

Plancherel

formula

onto, 61

interval,

set, 45, 232

identity,

also:

see

bijection

cover,

69

function,

partial

objects, 38

one-to-one

579

measure,

non-additivity of, 587, 589

508

one-to-one

268

discontinuity,

oscillatory

non-constructive, 229

464,

reals, 129

orthonormal,519

516
non-degenerate,
nowhere

97

rationale,

orthogonality, 516

434

neighbourhood,

Newton's

the

of the

positive

polar

507

representation,

polynomial,

265, 464

and convolution,

467

approximation

465,

by,

470

XI

INDEX

nested, 372

positive

complex number, 498,502


integer,

88

inner

product,

power

range, 55

ratio test, 206

128

series, 476
487

of,

121

interspersing

with

inverse

see

of infinite

image

real

21

induction,

strong induction,

transfinite,237

33

product

topology, 455

countability of the

rule

352,

363

examples, 364-367,

375-377

part,

rearrangement
of

series, 203,

222

354

of

28

finite

185

series,

series, 200

of non-negative

logic, 367

Pythagoras'theorem,516

reciprocal
of complexnumbers,499

of

369

negation of, 372

202

series,

44

(for some),

convergent

absolutely

of divergent

proper subset,

existential

498

real-valued, 455

proof

propositional

117

definition,

real

projection,536
abstract

see un-

uncountable:
reals

see Leibnitz

rule,

contradiction,

numbers

are

32, 234

product

quantifier,

reals,

real analytic, 478

mathematical

backwards

proposition,

with

interspersing

132

induction,

property,

103

rationale,

descent,

106

by

with reals,

identification

of, 481

power set, 66

principleof

93

definition,

uniqueness

principle

numbers

rational

474

multiplication

pre-image:

475

of convergence,

radius

96

formal,

see division

Quotient rule, 293,555

natural number,30
real,

370

all),

(for

Quotient:
515

measure, 576, 580

rational,

universal

371

95

rationale,

of reals,

125

recursive definitions,

26,76

XII

INDEX

reductio

see

absurdum:

ad

Schroder-Bernstein

relative topology: seeinduced


-topology

removable

singularity

sequence,109

of

Riemann

functions,

on

328

on

of monotonefunctions,

functions,

Riemann

simple function,

continuous

of uniformly

trigonometric

functions

318

(upper

and

singularity,
lower),

321

Riemann zeta function,199


integral,

Riemann-Stieltjes

ring,

89

commutative,

89, 497

Rolle'stheorem,297

root,140
mean

square:

test,

scalar

multiplication,

space, 390

statement, 350
sub-additive

336

subset,

576,

measure,

44

subsequence, 172, 408

substitution: see
subtraction

see

L2
formal

(\342\200\224),

of functions,

row vector, 533


paradox,

268

rearrangement

204

Russell's

41

set,

singleton

upper and lower, 317


sums

602

sine: see

326

integral,

Riemann

258
^\320\277\321\201\321\206\320\276\320\277^

boundedsignum

329

functions,

of set

theory
informal definition,38

330

continuous

see axioms

axioms:

compacta, 328
piecewise

180

sum,

set

functions

of

sets, 220

arbitrary

vs.

continuous

of continuous

459

functions,

on countable sets, 216

332

of bounded

179, 182

of

closureproperties,
321-326
of,

finite,

laws, 194,220

200

Riemann integrability, 318


failure

74

formal infinite, 189

250

functions,

hypothesis,

finite,

series

removable singularity, 259,268


restriction

theorem,

227

see

discontinuity:

removable

252

of functions,

proof

by contradiction

of integers,

52

86

252

91

sum rule,292
533

summation

by

parts,

484

580

INDEX

XIII

sup norm: see supremumas

in inner

516

norm

465

support,

in

as metric, 393

as

of a

for

set of extended reals,

156,157
ofa set of reals,137,139
of

of extended

518

natural

for

167

for sequences,

509

522

Fourier

and

235

bound,

518
polynomials,

trivial topology, 437

see trigonometric

tangent:

504, 508

series,

power
392

metric,

530

series,

trigonometric

surjection: see onto

taxi-cab

97

129

trigonometric functions, 503,

472,

upper

31

numbers,

rationale,

for reals,

Stone-Weierstrass
theorem,
strict

155

reals,

for

test

Squeeze

618

for integers, 91

512,

wave,

186

181,

series,

trichotomy of order

square root, 56

square

finite

for integrals,

158

of reals,

sequences

390

spaces,

499

in R, 99

457

393,

norm,

metric

in C,

(and infimum)

supremum

product spaces,

two-to-one

61

function,

function

480

series,

Taylor

Taylor's

series

series,195
telescoping

ten,381

of the

continuity,

convergence, 447

433

and

totally bounded, 418


totally orderedset,45,233
see

transformation:

translation

and

280,

577, 580,

integrals,

476

as

a metric,

453, 514

of series, 457

transpose,534

triangle

uniform

inequality

Euclidean

459

andradiusofconvergence,

function

invariance,

451

derivatives,

and limits, 450

591

in

428

uniform

uniform

and anti-derivatives,462
space,

topological

reals, 225

undecidable,228

28

theorem,

208

uncountability,

see Taylor

formula:

space,

399

limit,

of bounded

447

functions, 451

INDEX

XIV

of

continuous

450

functions,

and Riemann integration,

458

67

union,

42

pairwise,

set, 53

universal

upper bound,

of a setofreals,133
of

ordered

a partially

set,

234

see also: least upperbound


variable,

368

vector

space,

534

vertical line test,

55,

76,

567

578

volume,

Weierstrass

approximation
470-471,

465,

theorem,

521

Weierstrass

see nowhere

example:

differentiablefunction

WeierstrassM-test,

457

well-defined,

351

well-ordered

sets, 234

well ordering principle


for

natural

numbers,

for arbitrary

210

sets, 241

Zermelo-Fraenkel(-Choice)
axioms,

see

69

also axioms

of set

theory

zero

test

for sequences,

for series, 191

167

Zorn's

lemma,

237

Вам также может понравиться