You are on page 1of 351

Cal

ulus of One Variable

Jerry Shurman
Reed College

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1

The Parabola . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1 The Parabola in Eu lidean Geometry and in Algebra . . . . . . . .


1.1.1 The Geometri De ning Property . . . . . . . . . . . . . . . . . . .
1.1.2 The Algebrai De ning Equation . . . . . . . . . . . . . . . . . . . .
1.2 Quadrature of the Parabola . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 The First Ins ribed Triangle and Its Key Property . . . . .
1.2.3 Adding More Triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.4 Ar himedes's Evaluation of a Sum . . . . . . . . . . . . . . . . . . .
1.2.5 Solution of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Tangent Slopes of the Parabola . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Di eren e-Quotient and Se ant Slope . . . . . . . . . . . . . . . .
1.3.2 The Cal ulation Algebrai ally and Geometri ally . . . . . .
1.3.3 The Ins ribed Triangle Again . . . . . . . . . . . . . . . . . . . . . . .
1.3.4 The Re e tion Property of the Parabola . . . . . . . . . . . . .
1.4 The Parabola, Origami, and the Cubi Equation . . . . . . . . . . . .
1.4.1 Origami Folds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.2 Solving the Cubi Equation . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
1
3
6
6
7
8
11
11
13
13
14
17
18
20
20
21
24

The Rational Power Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 Assumptions About the Number System . . . . . . . . . . . . .
2.1.2 The Finite Geometri Sum Formula . . . . . . . . . . . . . . . . .
2.2 The Rational Power Fun tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 De nition of the Rational Power Fun tion . . . . . . . . . . . .
2.2.2 In reasing/De reasing Behavior . . . . . . . . . . . . . . . . . . . . .

26
26
28
29
30
34

vi

Contents

2.3 Integration of a Parti ular Rational Power Fun tion . . . . . . . . . .


2.3.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 Intuitive Vo abulary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3 The Idea to be Demonstrated . . . . . . . . . . . . . . . . . . . . . . .
2.3.4 The Problem Again, and the Pending Cal ulation . . . . .
2.3.5 Tools To Be Used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.6 The Geometri Partition . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.7 The Intervals and Their Widths . . . . . . . . . . . . . . . . . . . . .
2.3.8 The Inner Box-Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.9 The Sum of the Inner Box-Areas . . . . . . . . . . . . . . . . . . . .
2.3.10 The Limiting Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Di erentiation of the Rational Power Fun tion . . . . . . . . . . . . . .
2.4.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.2 The Cal ulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.3 A Fundamental Observation . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Integration of the Rational Power Fun tion . . . . . . . . . . . . . . . . .
2.5.1 The Normalized Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.2 The General Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3

36
36
36
37
39
40
41
43
44
44
46
47
47
48
50
51
51
54
57

Sequence Limits and the Integral . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.1 Sets, Fun tions, and Sequen es . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60


3.1.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.1.2 Fun tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1.3 Sequen es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.1.4 Previous Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2 The Limit of a Real Sequen e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.1 Absolute Value and Distan e . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.2 The Ar himedean Property of the Real Number System 76
3.2.3 De nition of Sequen e Limit . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2.4 Basi Sequen e Limit Rules . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2.5 Irrelevan e of Finite Index-Shifts . . . . . . . . . . . . . . . . . . . . 89
3.2.6 Uniqueness of the Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.2.7 Generative Sequen e Limit Rules . . . . . . . . . . . . . . . . . . . . 91
3.2.8 Geometri Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.2.9 Order Sequen e Limit Rules . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3 Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.3.1 The Previous Examples Revisited . . . . . . . . . . . . . . . . . . . 103
3.3.2 De nition of Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.3 Monotoni ity and Integrability . . . . . . . . . . . . . . . . . . . . . . 112
3.3.4 A Basi Property of the Integral . . . . . . . . . . . . . . . . . . . . . 113

Contents

vii

3.3.5 Pie ewise Monotoni ity and Integrability . . . . . . . . . . . . . 115


3.3.6 Generative Integral Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4

Function Limits and the Derivative . . . . . . . . . . . . . . . . . . . . . . . . 123

The Logarithm Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

4.1 The Limit of a Fun tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124


4.1.1 De nition of Fun tion Limit . . . . . . . . . . . . . . . . . . . . . . . . 124
4.1.2 Basi Fun tion Limit Rules . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.1.3 Generative Fun tion Limit Rules . . . . . . . . . . . . . . . . . . . . 130
4.1.4 Order Fun tion Limit Rules . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2 The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2.1 De nition of the Derivative . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2.2 A Consequen e Worth Noting Immediately . . . . . . . . . . . 134
4.2.3 The Derivative and the Tangent Line . . . . . . . . . . . . . . . . 135
4.2.4 A Basi Derivative: the Power Fun tion Revisited . . . . . 137
4.2.5 Generative Derivative Rules . . . . . . . . . . . . . . . . . . . . . . . . 140
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.1 De nition and Properties of the Logarithm . . . . . . . . . . . . . . . . . 147
5.1.1 Integration With Out-of-Order Endpoints . . . . . . . . . . . . 147
5.1.2 The Fundamental Theorem of Cal ulus . . . . . . . . . . . . . . 148
5.1.3 De nition of the Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.1.4 The Key Property of the Logarithm . . . . . . . . . . . . . . . . . 149
5.1.5 Proof of the Key Property: A Generality . . . . . . . . . . . . . 150
5.1.6 Proof of the Key Property: A Spe i Argument . . . . . . 152
5.1.7 Proof of the Key Property: End of the Proof . . . . . . . . . . 153
5.1.8 Further Properties of the Logarithm . . . . . . . . . . . . . . . . . 154
5.2 Logarithmi Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.3 Di erentiation of the Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.4 Integration of the Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.4.1 An Analyti Expression for the Logarithm . . . . . . . . . . . . 164
5.4.2 Another Summation Formula . . . . . . . . . . . . . . . . . . . . . . . 165
5.4.3 The Normalized Case: Left Endpoint 1 . . . . . . . . . . . . . . . 167
5.4.4 The General Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.4.5 The Fundamental Theorem of Cal ulus Again . . . . . . . . . 172
5.5 Signed Integration in General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.5.1 The Integral Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.5.2 Generative Integral Rules Revisited . . . . . . . . . . . . . . . . . . 175
5.5.3 The Area Between Two Curves . . . . . . . . . . . . . . . . . . . . . 179

viii
6

Contents
The Exponential Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

6.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186


6.1.1 De nition of Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.1.2 Continuity and Integrability . . . . . . . . . . . . . . . . . . . . . . . . 188
6.1.3 The Intermediate Value Theorem . . . . . . . . . . . . . . . . . . . . 189
6.1.4 Appli ations of the Intermediate Value Theorem . . . . . . 191
6.2 De nition and Properties of the Exponential Fun tion . . . . . . . 194
6.2.1 De nition and Basi Properties . . . . . . . . . . . . . . . . . . . . . 194
6.2.2 Raising to Powers Revisited . . . . . . . . . . . . . . . . . . . . . . . . 196
6.3 Exponential Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
6.4 Di erentiation of the Exponential . . . . . . . . . . . . . . . . . . . . . . . . . 199
6.5 Integration of the Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
6.6 The Exponential as a Limit of Powers . . . . . . . . . . . . . . . . . . . . . . 206
6.6.1 The Des ription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.6.2 An Interpretation: Compound Interest . . . . . . . . . . . . . . . 207
The Cosine and Sine Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

7.1 The Cir umferen e of the Unit Cir le . . . . . . . . . . . . . . . . . . . . . . 213


7.2 De nition of the Cosine and the Sine . . . . . . . . . . . . . . . . . . . . . . 214
7.3 Identities for the Cosine and the Sine . . . . . . . . . . . . . . . . . . . . . . 215
7.3.1 Basi Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.3.2 Angle Sum and Di eren e Formulas . . . . . . . . . . . . . . . . . 217
7.3.3 Double and Half Angle Formulas . . . . . . . . . . . . . . . . . . . . 218
7.3.4 Produ t Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.3.5 Di eren e Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
7.4 Di erentiation of the Cosine and the Sine . . . . . . . . . . . . . . . . . . 219
7.5 Integration of the Cosine and the Sine . . . . . . . . . . . . . . . . . . . . . 223
7.6 Other Trigonometri Fun tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.7 Inverse Trigonometri Fun tions . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Polynomial Approximation and Series Representation . . . . . 233

8.1 The Finite Binomial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234


8.2 Preliminaries for the Pending Cal ulations . . . . . . . . . . . . . . . . . . 237
8.2.1 An Alternative Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.2.2 The Power Fun tion Integral With Endpoint 0 . . . . . . . . 238
8.3 The Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
8.4 The Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8.4.1 A Pre al ulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8.4.2 The Cal ulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
8.5 The Cosine and the Sine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
8.6 The Power Fun tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
8.6.1 The Polynomial and the Remainder . . . . . . . . . . . . . . . . . 252

Contents

ix

8.6.2 The In nite Binomial Theorem . . . . . . . . . . . . . . . . . . . . . 255


8.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
9

Theory and Applications of the Derivative . . . . . . . . . . . . . . . . . 261

9.1 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261


9.1.1 The Extreme Value Theorem . . . . . . . . . . . . . . . . . . . . . . . 262
9.1.2 Conditions for Optimization . . . . . . . . . . . . . . . . . . . . . . . . 264
9.1.3 Optimization Story-Problems . . . . . . . . . . . . . . . . . . . . . . . 266
9.2 The Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
9.2.1 Statement of the Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 273
9.2.2 Consequen es of the Mean Value Theorem . . . . . . . . . . . . 276
9.3 Curve Sket hing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
9.4 Related Rates Story-Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

10 Integration via Antidifferentiation . . . . . . . . . . . . . . . . . . . . . . . . . 293

10.1 The Fundamental Theorem of Cal ulus . . . . . . . . . . . . . . . . . . . . . 293


10.1.1 Inde nite Integrals, Antiderivatives . . . . . . . . . . . . . . . . . . 293
10.1.2 The Fundamental Theorem, Part I . . . . . . . . . . . . . . . . . . 296
10.1.3 The Fundamental Theorem, Part II . . . . . . . . . . . . . . . . . . 301
10.2 Basi Antidi erentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
10.3 Antidi erentiation by Forward Substitution . . . . . . . . . . . . . . . . . 307
10.3.1 The Forward Substitution Formula . . . . . . . . . . . . . . . . . . 307
10.3.2 What the Formula Says and Why It Is True . . . . . . . . . . 308
10.3.3 Using the Formula in its Variable-Free Form . . . . . . . . . . 308
10.3.4 Improvement: the Formula With Variables . . . . . . . . . . . . 309
10.3.5 Se ond Improvement: the Pro edure Instead of the
Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
10.3.6 Basi Forward Substitution Formulas . . . . . . . . . . . . . . . . 312
10.3.7 Forward Substitution in Integrals . . . . . . . . . . . . . . . . . . . . 314
10.4 Antidi erentiation by Inverse Substitution . . . . . . . . . . . . . . . . . . 317
10.4.1 The Inverse Substitution Formula and Why It Is True . 317
10.4.2 The Formula With Variables . . . . . . . . . . . . . . . . . . . . . . . . 318
10.4.3 The Pro edure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
10.4.4 Inverse Substitution in Integrals . . . . . . . . . . . . . . . . . . . . . 321
10.5 Antidi erentiation by Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322

Assumptions About the Real Number System . . . . . . . . . . . . . 331

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333


Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

Preface

These are ourse notes for Mathemati s 111 at Reed College. They are written for serious liberal arts students who want to understand al ulus beyond
memorizing formulas and pro edures. The prerequisite is three years of high
s hool mathemati s, in luding algebra, eu lidean geometry, analyti geometry, and (ideally) trigonometry. To pro t from these notes, the student needn't
be a math genius or possess large doses of the omputational fa ilities that
al ulus ourses often sele t for. But the student does need su ient algebra
skills, study habits, energy, and genuine interest to on entrate an investment
in the material.
I have tried to put enough verbal exposition in these notes to make at
least portions of them readable outside of lass. And I have tried to keep the
al ulations short, tidy, and lightly notated, in the hope of rendering them
omprehensible stories that in ur belief, rather than rituals to endure. To the
extent that the notes are readable, I hope to use lasstime dis ussing their
ontents rather than onform to the model of the instru tor trans ribing a
le ture onto the bla kboard from whi h the students trans ribe it into their
notebooks in turn. The goal is that the students leave the ourse not having
taken my word about the results, but feeling truly vis erally that the results
are inevitable.
Exigen ies di tate that Math 111 simultaneously serve students who have
taken a al ulus ourse already and students who haven't. These notes attempt to do so in two ways,
 by rebalan ing the weight of explanation between mathemati al symbols

and natural language,

 and by presenting the omputations of al ulus as little more than end-

produ ts of algebra that one ould imagine naturally working out for oneself with some nudges in the right dire tion.

xii

Prefa e

The presentation is meant to defamiliarize al ulus for those who have seen it
already, by undoing any impression of the subje t as te hnology to use without
understanding, while making al ulus familiar to a wide range of readers, by
whi h I mean omprehensible in its underlying me hanisms. Thus the notes
will pose di erent hallenges to students with prior al ulus experien e and to
students without it. For students in the rst group, the task is to onsider the
subje t anew rather than fall ba k on invoking rote te hniques. For students
with no prior al ulus, the task is to gain fa ility with the te hniques as well
as the ideas.
These notes address three subje ts:


Integration. What is the area under a urve? More pre isely, what is a

pro edure to al ulate the area under a urve?


Di erentiation. What is the tangent line to a urve? And again, whatever
it is, how do we al ulate it?
 Approximation. What is a good polynomial approximation of a fun tion,
how do we al ulate it, and what an we say about the a ura y of the
approximation?


Part of the ompli ation here is that area under a urve and tangent line to
a urve are geometri notions, but we want to al ulate them using analyti
methods. Thus the interfa e between geometry and analysis needs dis ussion.
The basi pedagogy is to let ideas emerge from al ulations. In su ession,
these notes de ne, integrate, and di erentiate
 the rational power fun tion, f(x) = x where is a rational number,

meaning a ratio of whole numbers,

 the logarithm fun tion, f(x) = ln(x),


 the exponential fun tion, f(x) = exp(x) = ex ,
 the osine and sine fun tions, f(x) = os(x) and f(x) = sin(x).

The integrals are omputed without using the Fundamental Theorem of Cal ulus. Integrating the power fun tion leads to the idea that an integral is not
only an area, but more spe i ally an area that is well approximated from
below and from above by suitable sums of box-areas. Although the geometri ally natural idea is to integrate nonnegative-valued fun tions from a left
endpoint to a right endpoint, the logarithm leads to the idea of integrating a
fun tion that ould be negative between endpoints that need not be in order.
The logarithm also illustrates the idea of de ning a fun tion as an integral
and then studying its properties as su h. Similarly, the exponential fun tion
illustrates the idea of de ning and then studying a fun tion as the inverse of
another, and it suggests the idea of hara terizing a fun tion by a di erential
equation.

Prefa e

xiii

With the power fun tion, the logarithm, the exponential, and the osine
and sine integrated and di erentiated, we then nd approximating polynomials for these fun tions and estimate the a ura y of the approximations.
Essentially all of the program just sket hed an be arried out onvin ingly (if not \fully rigorously") based on only one small-but-versatile pie e of
te hnology, the nite geometri sum formula. This formula redu es many area
al ulations, limits of sums of many terms, to limits of quotients of two terms.
In fan ier language, the formula redu es integration to di erentiation. This
phenomenon is perhaps unsurprising sin e the Fundamental Theorem of Cal ulus says that integration and di erentiation are losely related. But whereas
the Fundamental Theorem is often taught as a pro edure that ir umvents
omputing integrals dire tly, a goal of these notes is to see di erentiation
emerge repeatedly from a tual integration. Students who learn to integrate
only by using the Fundamental Theorem risk gaining no real appre iation for
what integration really is, an appre iation worth having if only be ause the
Fundamental Theorem is irrelevant to so mu h integration in the real world.
Cal ulus does at some point require the te hni al ma hinery of limits.
These will be treated lightly after they are used informally. Cau hy's magni ent grammar deserves its due, but rst working informally with spe i
examples is meant to help the reader tangibly appre iate its e onomy and
nesse.
The last two hapters of these notes, on appli ations of the derivative and
on the Fundamental Theorem of Cal ulus, are traditional. In the footsteps
of so many before us, we will move ladders around orners, drain oni al
swimming pools, and generate blizzards of antiderivatives.
These notes are based on a set of notes by Ray Mayer. The motivation for
reating a new set of notes was that when this proje t began, the other set of
notes was not available in ele troni form. That situation has now hanged,
and the reader of these notes is en ouraged to look at Ray Mayer's notes as
well.

August 2007

Jerry Shurman
Reed College
Portland, OR

1
The Parabola

This hapter uses the parabola to illustrate ideas from al ulus qui kly and
informally. Se tion 1.1 hara terizes the parabola geometri ally and algebrai ally. Se tion 1.2 omputes the area of the region between a parabola
and a hord joining two of its points. The omputation pro eeds by systemati ally lling the region with triangles, adding ever smaller triangles at ea h
step. The individual triangle-areas are al ulable, and the sum of the areas
after ea h step takes a form that lets us determine the value to whi h it tends
the more steps we arry out. This value is the desired area. The paraboli
area- al ulation is our rst example of integration, a fundamental pro ess of
al ulus. Se tion 1.3 omputes the slope of a tangent line to the parabola,
rst by algebra and then again by geometry. Here the idea is that although
the tangent line passes through only one point of the parabola, we an approximate it by se ant lines that pass through two parabola points, and the
se ant slopes are easy to understand. As the se ond parabola point nears
the rst, the value tended to by the se ant slopes is the tangent slope. The
paraboli tangent slope al ulation is our rst example of di erentiation,
another fundamental pro ess of al ulus. Se tion 1.4 explains how tangent
slopes of parabolas an be used to solve ubi equations (polynomial equations of third degree). It turns out that the solutions an be onstru ted by a
paper-folding pro ess, i.e., by origami.

1.1 The Parabola in Euclidean Geometry and in Algebra


1.1.1 The Geometric Defining Property

Working in the eu lidean plane and using artesian oordinates, onsider a


horizontal line alled the directrix, set one-quarter unit down from the xaxis,

1 The Parabola
y

PSfrag repla ements

F = (0, 1/4)
x
D : y = 1/4
Figure 1.1.

Dire trix and fo us

D = the points (x, 1/4) for all values of x.

We abbreviate the des ription of the dire trix by writing


D : y = 1/4.

Consider also a point alled the focus, set one-quarter unit up the y-axis,
F = (0, 1/4).

(See gure 1.1.) The parabola is de ned geometri ally as the lo us of all
points P that are equidistant from the dire trix and from the fo us,
PD = PF.

(See gure 1.2.)


y

PSfrag repla ements


F

x
D
Figure 1.2.

Point equidistant from the dire trix and the fo us

1.1 The Parabola in Eu lidean Geometry and in Algebra

1.1.2 The Algebraic Defining Equation

To translate the geometri ondition de ning the parabola into an algebrai


ondition, note that for any point P = (x, y), the square of the distan e from P
to the dire trix D is the square of the di eren e of the y- oordinates,
(1.1)

PD2 = (y + 1/4)2 .

Also, the Pythagorean Theorem (exer ise 1.1.1) says that the square of the
distan e from P to the fo us F is
PF2 = (y 1/4)2 + x2 .

(See gure 1.3.) In the general algebrai identity A2 B2 = (A + B)(A B),


let A = y 1/4 and B = y + 1/4, so that (A + B)(A B) = 2y (1/2) = y,
to get (y 1/4)2 = (y + 1/4)2 y. So the previous display rewrites as
(1.2)

PF2 = (y + 1/4)2 + x2 y.

The left sides of (1.1) and (1.2) are equal by the geometri de nition PD = PF
of the parabola. Therefore the right sides are equal, and so the geometri
ondition de ning the parabola is exa tly the algebrai equation
(1.3)

y = x2 .

That is, the parabola is the graph of the squaring fun tion f(x) = x2 . The
parabola is shown in gure 1.4.
y

PSfrag repla ements

x
y 1/4
(0, 1/4)

(x, y)

PF
x

Figure 1.3.

Point-to-fo us distan e by the Pythagorean Theorem

The algebrai equation y = x2 of the parabola is normalized by the hoi e


to pla e the fo us and the dire trix one-quarter of a unit away from the x-axis.

1 The Parabola
y

PSfrag repla ements

F
x
D
Figure 1.4.

The parabola

For any positive number r, suppose that instead the fo us and the dire trix
are


F : (x, y) =

0,

1
4r

D:y=

1
.
4r

Make the hange of variables

x
~ = rx,

y
~ = ry.

In the (x~, y~ )- oordinate system, the fo us is




1
F : (x
~, y~ ) = r 0, r
= (0, 1/4),
4r

and the dire trix is



1
= 1/4.
D:y
~ =r
4r

That is, in the (x~, y~ )- oordinate system, the fo us and the dire trix are ba k
in their normalized positions where we have already studied them, and so the
equation of the parabola is
~ = x~2 .
y
Returning to the (x, y)- oordinate system, sin e y~ = ry and x~ = rx, the
parabola with fo us F = (0, 1/(4r)) and dire trix D : y = 1/(4r) therefore
has equation ry = (rx)2 , or
y = rx2 .

We an turn this reasoning around as well. And so an equation y = rx2


(where r is positive) des ribes a parabola with fo us F = (0, 1/(4r)) and
dire trix D : y = 1/(4r). If instead r is negative then the parabola opens
down instead of up. Similarly, ex hanging the roles of x and y to obtain an
equation

1.1 The Parabola in Eu lidean Geometry and in Algebra

x = ry2

des ribes a parabola that opens to the right if r is positive, or to the left if r
is negative. More generally, the equations
y c = r(x b)2 ,

x b = r(y c)2

des ribe parabolas that are shifted a horizontal distan e b and a verti al
distan e c. For the rst of these, the fo us and dire trix are


1
b, c +
,
4r

D:y=c

1
,
4r



1
F = b+ ,c ,
4r

D:x=b

1
.
4r

F=

and for the se ond they are

We will use su h parabolas in se tion 1.4.


Exercises
1.1.1. Consider a right triangle with sides a and b and hypotenuse c. The

Pythagorean Theorem states that

a2 + b2 = c2 ,

i.e., the square of the hypotenuse is the sum of the squares of the other
two sides. Explain why the shaded region in the right side of gure 1.5 is
a square. Then explain why the gure proves the theorem. (Your argument
should involve labeling some lengths and angles in the gure.)
1.1.2. What is the equation of the parabola with fo us F = (1, 2) and dire trix
D : x = 3?
1.1.3. Consider the parabola with equation
y = Ax2 + Bx + C,

Where are its fo us and its dire trix?


1.1.4. Explain why the equation
y2 + 2xy + x2

A 6= 0.

2y + 2x = 0

des ribes a parabola. What are its fo us and its dire trix? It may help to
onsider the hange of variables
x+y
x
~= ,
2

(See gure 1.6.)

y
~=

x + y
.
2

1 The Parabola

Figure 1.5.

Proof of the Pythagorean Theorem


y

y
~

x
~

PSfrag repla ements


x

Figure 1.6.

Rotated oordinate system

1.2 Quadrature of the Parabola


The earliest example of integration goes ba k not to Newton (1642{1727) or
Leibniz (1646{1716) or Barrow (1630{1677) or Des artes (1596{1650), but to
Ar himedes ( . 287 BC{ . 212 BC).
1.2.1 The Problem

Let

a and b be numbers with a < b. Find the area between the parabola
y = x2 and its hord from (a, a2 ) to (b, b2 ). (See gure 1.7.) Note that the

word area informally has two di erent meanings that an easily blur together
in this ontext. First, area means the shaded portion in gure 1.7, i.e., a
region. But se ond, area means a number that somehow measures the planar
size of the region on a linear s ale, despite the fa t that the region itself lies
in the plane rather than on the line. For the parabola problem, we are ta itly
assuming that indeed some number is the measure of the shaded region.

1.2 Quadrature of the Parabola

PSfrag repla ements


a
Figure 1.7.

Region between a parabola and its hord

PSfrag repla ements


a

b
Figure 1.8.

Triangle as the di eren e of trapezoids

1.2.2 The First Inscribed Triangle and Its Key Property

The rst approximation to the region is a triangle with its left and right
verti es above a and b and its middle vertex above the midpoint (a + b)/2.
This triangle an be viewed as a large trapezoid with two smaller trapezoids
removed. (See gure 1.8.) In general, the area of a trapezoid of base B and
heights H1 and H2 is the base times the average of the heights,
Atrap = B

H1 + H2
.
2

1 The Parabola

The trapezoid that ontains the triangle has base ba and heights a2 and b2 .
(These are the heights be ause the points have x- oordinates a and b, and
they lie on the parabola y = x2 .) The left trapezoid underneath the triangle
has base (ba)/2 and heights a2 and (a+b)2 /4. (Re all that the x- oordinate
of the third vertex is the average (a + b)/2.) The right trapezoid underneath
the triangle has the same base (b a)/2 as the left trapezoid, but heights
(a + b)2 /4 and b2 . It follows that the area of the triangle is
Atri



a2 + b2 b a a2 + 2(a + b)2 /4 + b2
= (b a)
,

2
2
2

(1.4)

and by some algebra (exer ise 1.2.1) this works out to


Atri =

1
(b a)3 .
8

(1.5)

So the area of the triangle in the left part of gure 1.8 is one-eighth of its
width ubed. Now (1.5) allows us to make a ru ial observation:

The area of the triangle depends only on the width of the triangle.
That is, (1.5) shows that the area depends on the di eren e b a but not on
a and b individually, so long as the x- oordinate of the third vertex is their
average (a + b)/2.
Exercise
1.2.1. Carry out the algebra that leads from (1.4) to (1.5).
1.2.3 Adding More Triangles

This observation that the area of a triangle ins ribed in the parabola depends
only on the triangle's width, provided that the x- oordinate of its middle
vertex is the average of the x- oordinates of the left and right verti es, says
that very di erent-looking triangles ins ribed in the parabola will have the
same area.
In parti ular, if we add two more triangles to ll in some of the missing
spa e in the left part of gure 1.9, as shown in the right part of the gure,
then even though the two new triangles are not ongruent, they have the same
area, one-eighth of their width ubed. Sin e their width is one-half the width
of the rst triangle, their areas are one-eighth the area of the rst triangle,

Atri

1
=
8

ba
2

3

1
1
1
(b a)3 = Atri .
8
8
8

1.2 Quadrature of the Parabola

Adding the two new triangles, ea h having one-eighth the area of the original
triangle, adds to the original area a new fa tor of one-quarter the original
area, making the total area of the three triangles


1

.
S2 = Atri + 2Atri
= Atri 1 +
4

We are alling this quantity S2 sin e it is the triangle-area sum after the
se ond generation of adding triangles. Naturally, S1 is just Atri itself.

Figure 1.9.

Filling in with two more triangles

Next add four more triangles, ea h half as wide as the two just added.
(See gure 1.10. In the gure it is not at all visually suggestive to the author
of these notes that the four new triangles all have the same area, and it is
hard to tell the di eren e between three of the four new triangles and the
paraboli region that they partially ll.) Thus we add twi e as many triangles
as at the previous step, ea h with one-eighth the area of the ones added at
the previous step, so that the new ontribution to the area is one-fourth the
ontribution of the previous step, whi h in turn was one-fourth the area of
the original triangle. So after three generations of adding triangles, the area
of the seven triangles is

S3 = Atri + 2Atri + 4Atri = Atri

"

1
1+ +
4

 2 #
1
.
4

By the same sort of reasoning (exer ise 1.2.2), adding eight more triangles
gives fourth-generation area

10

1 The Parabola

Figure 1.10.

Four more triangles

S4 = Atri + 2Atri + 4Atri + 8Atri = Atri

"

1
1+ +
4

 2  3 #
1
1
,
+
4
4

(1.6)

and so on. After n generations of adding triangles, ea h step has added twi e
as many triangles as the previous step, ea h triangle having one-eighth the
area of those added at the previous step, ontributing in total a quarter of
the previous ontribution. That is, after n generations the area is
Sn = Atri

"

1
1+ +
4

 n1 #
 2
1
1
.
+ +
4
4

The next task is to evaluate the sum in (1.7).

(1.7)

1.2 Quadrature of the Parabola

11

Exercise
1.2.2. Explain arefully why equation (1.6) is orre t.
1.2.4 Archimedess Evaluation of a Sum

As mentioned, the al ulation being shown here is due to Ar himedes, but


his mathemati al environment was purely geometri whereas we have made
heavy use of artesian oordinates and algebra. In parti ular, Ar himedes
used a geometri argument to evaluate the sum in (1.7). For the argument
when n = 4, onsider the unit square shown in gure 1.11. The largest,
lightest L-shaped region has area three-quarters. Ea h su essive, darker Lshaped region has linear dimensions half as big as its prede essor's, and hen e
area one-fourth of its prede essor's. That is, the total area of the four L-shaped
pie es is
"
#
1
3
1+ +
4
4

 2  3
1
1
.
+
4
4

Also, the small, dark square in the upper left orner has sides (1/2)4 and
hen e area (1/4)4 . And the total area of the square is 1. Thus
"
 2  3 #  4
3
1
1
1
1
1+ +
+
+
= 1,
4
4
4
4
4

and the geometri argument has given a tidy form to the sum of the rst four
powers of one-fourth,
1
1+ +
4

 2  3
4
1
1
+
=
4
4
3

 4 !
1
1
.
4

The argument generalizes immediately to the sum of the rst n powers of


one-fourth,
1+

1
+
4


 n 
 n1
 2
4
1
1
1
1
.
+ +
=
4
4
3
4

(1.8)

1.2.5 Solution of the Problem

Formulas (1.7) and (1.8) ombine to show that after n generations of adding
triangles, the total area is
Sn = Atri

4
3

 n 
1
.
4

12

1 The Parabola

Figure 1.11.

Subdivided unit square

As n grows very large, the triangles ll up the region whose area we want,
and so the limiting value of Sn is the region's area. But also, as n grows very
large, (1/4)n gets very lose to 0, and so the limiting value of the area is
Slim = Atri

4
.
3

That is, the area between the parabola and the hord is four-thirds times the
area of the rst ins ribed triangle. This is how Ar himedes formulated the
solution. Sin e Atri = 81 (b a)3 , another formulation is
Slim =

1
(b a)3 .
6

(1.9)

The reasoning just given to obtain the boxed formulas for Slim alls for
s rutiny. To dis uss what happens as n grows very large is not to say that any
nite number of triangles ll up the paraboli region whose area we want, nor
is it to say that the areas of any nite number of triangles sum to 4Atri /3.
And we should be deeply skepti al about treating \in nity" as a number. A
more sophisti ated formulation of what is being said is that by our hoi e of
on guration,

the paraboli region is


lling,

toward

exactly

the region that the triangles tend

1.3 Tangent Slopes of the Parabola

13

and that by our al ulation,

the number 4Atri /3 is exactly the number that the sums of the
triangle-areas tend toward rea hing,
and that

as the triangles tend toward lling the paraboli region, the sums
of the triangle-areas must tend toward rea hing the area of the
paraboli region,
and therefore

the area of the paraboli region must be

4Atri /3.

This dis ussion is not at all satisfa tory. It will be re ned over the ourse of
these notes.
Exercises
1.2.3. As in the se tion, let a and b be numbers with a < b. Find the area
under the parabola and above the x-axis from a to b by using formula (1.9)

and the formula for the area of a trapezoid.

1.2.4. Let n denote a positive integer. The larger n is, the loser the quantity
1 + 1/n is to 0. Does this mean that the quantity tends to 0? Explain.
1.2.5. Raise some riti isms of the \more sophisti ated formulation" given at

the end of the se tion.

1.3 Tangent Slopes of the Parabola


1.3.1 Difference-Quotient and Secant Slope

The squaring fun tion is

f(x) = x2 .

This is the fun tion whose graph is the parabola. For any xed x, the quantity
s2 x 2
,
sx

s 6= x

is a difference-quotient and a secant slope. Di eren e-quotient means


quotient of di eren es, i.e., the numerator of the previous display is the

di eren e f(s) f(x) of output-values of f, while the denominator s x is the


di eren e of the orresponding input-values. Meanwhile, a se ant line of the

14

1 The Parabola

parabola is a line through two parabola points, and a se ant slope is the slope
of a se ant line. For the interpretation of the di eren e-quotient as a se ant
slope, see gure 1.12. The problem is: To what value does the di eren equotient (se ant slope) tend as s tends to x? The limiting value is alled, for
reasons to be explained soon, the tangent slope of the parabola at (x, x2 ).
y

PSfrag repla ements


s2
s2 x 2
x2

sx
x

Figure 1.12.

Se ant lines through (1, 1)

What makes the problem subtle is that the numerator s2 x2 and the
denominator s x of the di eren e-quotient both tend to 0 individually as s
tends to x. If instead they both tended to nonzero values then naturally we
would guess that the limiting value of the quotient exists, sin e it should be
the quotient of the limiting values. But the fa t that the numerator and the
denominator both tend to 0, and the fa t that the quotient 0/0 is unde ned
(this point will be dis ussed in hapter 2) doesn't pre lude the possibility
that the di eren e-quotient tends to some well de ned value. The issue is
that determining the value will require a little analysis.
1.3.2 The Calculation Algebraically and Geometrically

For any real number x, ompute


s2 x 2
=s+x
sx

for s 6= x.

As s tends to x, s + x tends to 2x. It's that simple. And so:


s2 x 2
sx

tends to 2x as s tends to x.

1.3 Tangent Slopes of the Parabola

15

Geometri ally, the result is:


The tangent slope of the parabola y = x2 at the point (x, x2 ) is 2x.
This is understood to mean that as s tends to x, the se ant line of the parabola
through (x, x2 ) and (s, s2 ) tends to the tangent line to the parabola at (x, x2 ),
and therefore the number 2x tended to by the se ant slopes must be the
tangent slope. Like the argument about quadrature, this argument is open
to many riti isms|for example, the phrase tends to apparently now applies
to lines in addition to applying already to regions and to numbers|but we
make do with it for the time being.
We an also obtain the tangent slope of the parabola from a dire t geometri argument. Let x be xed, and onsider the tangent line to the parabola
at (x, x2 ). For any value s, moving along the parabola from (x, x2 ) to (s, s2 )
hanges the verti al oordinate by
s2 x 2 .

(This distan e is the gray portion of the y-axis in gure 1.13.) On the other
hand, moving from (x, x2 ) along the tangent line to the point with rst oordinate s hanges the verti al oordinate by
m(s x),

where m is the slope of the tangent line.

(This is the other gray distan e in gure 1.13.) We want to nd m in terms


of x. Sin e the parabola is onvex (i.e., it bends up everywhere), the tangent
line lies below the parabola everywhere ex ept at the point of tangen y. The
verti al distan e formulas in the previous two displays show that:
Given x, we want m su h that s2 x2 m(s x) for all s.
Equivalently:
Given x, we want m su h that (s x)(s + x m) 0 for all s.
Guided either by hindsight or insight, we see that the orre t hoi e is
m = 2x,

sin e then (s x)(s + x m) = (s m)2 , and indeed


(s x)2 0

for all s.

That is, the tangent slope to the parabola y = x2 at the point (x, x2 ) is 2x,
as we already omputed above.

16

1 The Parabola
y

PSfrag repla ements


s2
s2 x 2
x

m(s x)

x
Figure 1.13.

Height di eren es along the parabola and along a tangent line

Note the di eren es between the two ways of omputing the tangent slope
of the parabola. The al ulus argument is very qui k, and it tells us the
answer, but it relies on the nebulous notion of tends to. By ontrast, the
geometri argument stands on rmer footing, but it requires us to know the
answer somehow and then verify it.
Exercise
1.3.1. (a) Make a rough sket h of the ubi urve y = x3 . Your sket h should
show that the urve is onvex (bends up) for x > 0, is on ave (bends down)
for x < 0, and in e ts at x = 0.

(b) Here is a geometri argument similar to the one just given for the
paraboli urve y = x2 , but for the ubi urve instead. Let x be xed, and

onsider the tangent line to the ubi urve at (x, x3 ). For any value s,
moving along the urve to (s, s3 ) hanges the verti al oordinate by s3 x3 ,
and moving along the tangent line to the point with rst oordinate s
hanges the verti al oordinate by m(s x), where m is the tangent slope.
We want to nd m in terms of x. By a little algebra, the di eren e
between the two height- hanges is
s3 x3 m(s x) = (s x)(s2 + sx + x2 m).

Divine inspiration tells us to onsider


m = 3x2 .

Doing so makes the di eren e between the two height- hanges

1.3 Tangent Slopes of the Parabola


(s x)(s2 + sx + x2 3x2 ) = (s x)2 (s + 2x).

17

(1.10)

By the geometry of the ubi urve, as des ribed in part (a). . .

Complete the argument that m = 3x2 is the orre t hoi e by explaining


why the height di eren e on the right side of (1.10) is behaving appropriately.
Be lear about the respe tive roles and behaviors of x and s. The solution is
not a simple mimi ry of what's in the text|the ubi urve requires a more
areful analysis than the parabola.
( ) What does the height di eren e on the right side of (1.10) tell us about
where the ubi urve and its tangent line at (x, x3 ) meet?
(d) Pro eed similarly to parts (a) through ( ) but with the quarti urve
y = x4 .
1.3.3 The Inscribed Triangle Again

Re all that the quadrature of the parabola began with an ins ribed triangle,
its left vertex at (x, y) = (a, a2 ) and its right vertex at (x, y) = (b, b2 ). (The
triangle is shown in the left parts of gure 1.8 and gure 1.9, and it is the
large triangle in gure 1.10|see pages 7, 9, and 10.) The slope of the paraboli
se ant hord between these two verti es is
slope between the left and right verti es =

b2 a 2
= a + b.
ba

Re all also that the triangle's third vertex has x- oordinate (a + b)/2. By our
formula that the tangent slope to the parabola at any point (x, x2 ) is 2x, we
have in parti ular (sin e 2 (a + b)/2 = a + b),
tangent slope to the parabola at the third vertex = a + b.
That is, the ins ribed triangle now has a geometri hara terization: The

middle vertex is the point where the tangent line to the parabola is parallel
to the line through the left and right verti es. (See gure 1.14.) And in fa t
this hara terization applies to all of the triangles in the quadrature of the
parabola.
Exercise
1.3.2. Let a and b be real numbers with a < b. Consider two points on the
parabola, P = (a, a2) and Q = (b, b2 ). For any number c between a and b,
onsider also a third point on the parabola, R = (c, c2 ). Thus the ins ribed

triangle used by Ar himedes for the quadrature of the parabola o urs when
in parti ular c is the average (a + b)/2. Give a geometri argument that of

18

1 The Parabola

Figure 1.14.

Tangent line parallel to se ant hord

all triangles PQR where P and Q are the xed points just mentioned, and R
is some third point between them on the parabola, Ar himedes hose the
triangle of greatest area, i.e., the triangle that lls as mu h as possible of the
region between the parabola and the hord PQ. (Hint: Triangle-area is onehalf of base times height. View PQ is the ommon base of all the triangles
in question. The tangent line to the parabola at Ar himedes's hoi e of R
lies below the parabola ex ept at R, and the se tion has explained that this
tangent line is parallel to the hord PQ. Your argument should be based on
these ideas and make no referen e to the (x, y)- oordinate system.)
1.3.4 The Reflection Property of the Parabola

Let

P = (x, x2 )

be a point on the parabola, where x is any number. The geometri de nition


of the parabola is PD = PF, and PD is the verti al distan e x2 + 1/4 from the
point to the dire trix, so that also
PF = x2 + 1/4.

Consider the point verti ally above P, also at distan e x2 + 1/4 from P,
Q = (x, 2x2 + 1/4)

(see gure 1.15). The slope from F to Q is the y- oordinate di eren e divided
by the x- oordinate di eren e,

1.3 Tangent Slopes of the Parabola


y

19

PSfrag repla ements


P

Figure 1.15.

m=

Equal angles

2x2 + 1/4 1/4


= 2x.
x0

This is also the tangent slope of the parabola at P. By Eu lidean geometry,


the segments QP and PF therefore form the same angle with the tangent line.
(Again see gure 1.15.) This gives the re e tion property of the parabola:
every verti al ray re e ts in the parabola to a ray through the fo us. (See
gure 1.16.) This property of the parabola is used to onstru t teles opes,
and it is often demonstrated in s ien e museum exhibits.
y

PSfrag repla ements

Figure 1.16.

Re e ted rays meet at the fo us

Exercise
1.3.3. Explain why the four angles in gure 1.15 are equal.

20

1 The Parabola

1.4 The Parabola, Origami, and the Cubic Equation

Eu lidean onstru tions are arried out with the geometri tools of antiquity: straight-edge and ompass. It is known that viewed as algebrai methods,
Eu lidean onstru tions solve linear and quadrati equations but fail at ubi s. By ontrast, origami (paper-folding) has the apa ity to solve ubi
equations. The key idea is to use the ommon tangents to two parabolas. In
fa t, origami onstru ts ommon tangents to parabolas using only the parabolas's fo i and dire tri es, not the parabolas themselves. Thus origami use only
points and lines, not the urved parabolas. This se tion dis usses these ideas
very brie y.
1.4.1 Origami Folds

Folding a given point F onto any point Q of a line D onstru ts a tangent


to the parabola P having F and D for its fo us and dire trix (exer ise 1.4.1).
(See Figure 1.17.)

F
P

PSfrag repla ements


D
Q

Figure 1.17.

Folding a tangent to a parabola

Suppose that two parabolas P1 , P2 , with fo i F1 , F2 and dire tri es D1 , D2 ,


have ommon tangents. A ontinuum of foldings takes F1 to points of D1 ,
onstru ting all tangents to P1 . In parti ular, sliding F1 along D1 until F2
also lies on D2 onstru ts the ommon tangents to P1 and P2 . Thus we have
the axiom that the ommon tangents to two parabolas, ea h spe i ed by its
fo us and dire trix, are onstru tible by origami when they exist. (But they

1.4 The Parabola, Origami, and the Cubi Equation

21

needn't exist at all: onsider the ase when one parabola lies entirely inside
the other.)
Exercises
1.4.1. Explain why folding a given point F onto any point Q of a line D
onstru ts a tangent to the parabola P having F and D as its fo us and

dire trix.

1.4.2. (a) Mark a pie e of paper with a fo us-point F and a dire trix-line D.
Fold F onto various points Q of D. How many folds does it take before you
an see the parabola PD = PF learly?
(b) Mark a pie e of paper with two fo us-points F1 and F2 , and with
two dire trix-lines D1 and D2 . Fold F1 onto D1 enough times that you see
the parabola PD1 = PF1 learly, and then do the same for the parabola
PD2 = PF2 . If possible, fold F1 onto D1 and F2 onto D2 simultaneously to see

a ommon tangent of the two parabolas.


( ) Mark a pie e of paper with fo us-points F1 and F2 and dire trixlines D1 and D2 so that the parabolas PD1 = PF1 and PD2 = PF2 have
as many ommon tangents as you an make them have.
1.4.2 Solving the Cubic Equation

Let b, c, and d be arbitrary numbers ex ept that d is nonzero. Consider two


parabolas, the rst one spe i ed by b, c, and d,
P1 : (y + c)2 = 4d(x b),
P2 :

x2 = 4y.

The rst parabola has equation


P1 : y
~ 2 = x~,

where y~ = y + c and x~ = 4d(x b).

(1.11)

That is, after a hange of variables, P1 has the normalized parabola equation
but with the roles of the two variable reversed. Let p1 = (x1 , y1 ) be a point
on P1 , and let T1 denote the tangent line to P1 at p1 . In (x, y)- oordinates,
the tangent slope of P1 at p1 is the ratio
m1 =

y
x

along T1 , where means hange in.

Consequently, a ording to the hange of variables in (1.11),


m1 =

~ c)
(y
(x
~/(4d) + b)

along T1 .

22

1 The Parabola

Change in y~ c produ es the same hange in y~ sin e c is onstant. Similarly,


hange in x~/(4d) + b is the hange in x~ divided by 4d. (E.g., sin e feet is
in hes/12, also hange in (feet+b) is ( hange in in hes)/12.) So the previous
ratio is in fa t
y
y
~
~
m1 =
along T1 ,
= 4d
x
~/(4d)
x
~
whi h is
 x
~
m1 = 4d
along T1 .
y
~
The ratio in this last display is the slope of T1 in (y~ , x~)- oordinates, the
oordinates in whi h the equation of the parabola P1 is normalized. Hen e
we may quote our di erentiation result: the slope is 2y~ 1 . That is, in (x, y) oordinates the tangent slope of P1 at p1 is
m1 =

4d
2d
= .
2y
~1
y
~1

Returning to the (x, y)- oordinate system, the result is


m1 =

2d
.
y1 + c

(1.12)

Similarly, let p2 = (x2 , y2 ) be a point on P2 . Then the tangent slope to P2


at p2 is (exer ise 1.4.3)
x2
m2 = .
(1.13)
2

Now suppose that a line through the point (x1 , y1 ) on the rst parabola
and the point (x2 , y2 ) on the se ond parabola is tangent to both parabolas.
Let m denote the slope of this ommon tangent. Sin e the points are on the
respe tive parabolas, and sin e the line is tangent to both parabolas, we have
the relations
(y1 + c)2 = 4d(x1 b),
y1 + c = 2d/m
x22

by (1.12),

= 4y2 ,

x2 = 2m

by (1.13).

Substitute the se ond relation into the rst and substitute the fourth relation
into the third to get expressions for x1 , y1 , x2 , and y2 in terms of m,
x1 = d/m2 + b,
y1 = 2d/m c,
y2 = m2 ,
x2 = 2m.

1.4 The Parabola, Origami, and the Cubi Equation

23

But sin e the line passes through the points (x1 , y1 ) and (x2 , y2 ), and its
slope is m, also
m(x1 x2 ) = y1 y2 .

In this last relation, repla e the x's and the y's by their expressions in terms
of m to get


m

d
+ b + 2m
m2

2d
c + m2 .
m

After some arithmeti with the fra tions, this gives

d + bm2 + 2m3 = 2d cm + m3 ,

or, nally,

m3 + bm2 + cm + d = 0.

This dis ussion has proved the following result:

Given the ubi equation


X3 + bX2 + cX + d = 0,

d 6= 0,

the slopes of the ommon tangents to the two parabolas


P1 : (y + c)2 = 4d(x b)

and

are roots of the equation.

P2 : x2 = 4y

(There is no loss in taking d 6= 0, sin e if d = 0 then the equation fa tors


as X(X2 + bX + c) = 0, whi h we already know how to solve by the quadrati
formula.) And furthermore:

Sin e the fo us F1 = (b d, c) and the dire trix D1 : x = b + d of


the rst parabola are known, as are the fo us F2 = (0, 1) and the
dire trix D2 : y = 1 of the se ond parabola, the ommon tangents
an be obtained by origami.
Figure 1.18 shows this method applied to the ubi equation
x3 2x2 x + 2 = 0,

with roots 2, 1, and 1. For more about mathemati al origami, see the text
by Thomas Hull, or see his web site.
Exercises
1.4.3. Establish equality (1.13).
1.4.4. Choose a ubi equation and fold its roots.

24

1 The Parabola
6

slope 2
4

P1

-6

-4

-2

slope 1

slope -1
-2

-4

P2
-6

Figure 1.18.

Solving a ubi equation by slopes of ommon tangents

1.5 Summary
Most of the work with the parabola in this hapter was geometri or algebrai .
What gives the material aspe ts of al ulus as well is that we determined
the pre ise values that approximations tended to as they be ame ever more
a urate. However, the entities involved in the al ulations, and the reasoning
about them, require loser s rutiny. The al ulus that we have done so far is
only provisional.

2
The Rational Power Function

The rational power fun tion is the fun tion


f (x) = x

for x > 0,

where the exponent is a rational number, meaning a ratio of whole numbers,


e.g., = 3 or = 2 or = 3/2 or = 17/3. While x is easy to understand
when is a positive whole number|it is x x x ( times)|the meaning
of x for negative whole , or fra tional , or negative fra tional is less
lear.
Se tion 2.1 lays out some ideas preliminary to studying the rational power
fun tion. Basi assumptions about the real number system are stated informally, and then a ubiquitously useful formula is introdu ed, the nite geometri sum formula. Se tion 2.2 de nes the rational power fun tion f and
shows that for positive values of the fun tion is always limbing, while
for negative values of the fun tion is always falling. Se tion 2.3 nds the
area under the graph of the spe i rational power fun tion f2/3 from x = 1
to x = 8. The pro ess here is integration, and the exposition tries to onvey a
on eptual sense of it along with the details as it unfolds. Se tion 2.4 omputes
the derivative of the rational power fun tion, arrying out the al ulation in
several steps from a normalized spe ial ase to full generality. The derivative al ulation reprodu es some of the ending work of the integration in the
previous se tion, suggesting a onne tion between derivatives and integrals.
Se tion 2.5 exploits this onne tion to al ulate the integral of the general
power fun tion f (ex epting the ase = 1) between general endpoints
x = a and x = b.

26

2 The Rational Power Fun tion

2.1 Preliminaries
2.1.1 Assumptions About the Number System

Among the many ta it assumptions permeating hapter 1 were assumptions


about numbers. We need to pro eed from some onsensus about how numbers
behave. Thus:

We assume that there is a system of real

numbers.

The assumed real number system has properties that should be familiar.
Spe i ally:






The real number system subsumes the rational number system. An

integer is a whole number su h as 0, 1, 1, 2, 2, 3, . A rational


number is a ratio p/q where p and q are integers and q is not 0. But q
an be 1, so that the rational numbers subsume the integers. All rational

numbers are real, but not all real numbers are rational.

Real numbers an be added, subtra ted, multiplied, and divided, all


subje t to the usual rules of algebra. Division by 0 is prohibited.
Every real number is nite. \In nity" is not a real number.
Real numbers an be ompared. Given any two real numbers, either the

rst one given is the lesser, or the two are equal, or the rst one given is
the greater. In parti ular, the positive real numbers are the ones that are
greater than 0, and the negative real numbers are the ones that are less
than 0.
 The real numbers an be interpreted as the points of a line. By onvention, greater numbers are lo ated to the right of lesser ones. Under this
interpretation, the rational numbers are only some of the points of a line,
and the rest of the real numbers somehow ll the holes. Any segment of the
line having positive length ontains both rational and irrational numbers.
Here are some omments about these assumptions. Ea h point re eiving a
omment is rst repeated in itali s.
We assume that there is a system of real numbers. This assumption does
not say what a real number is. Nor does the mere a t of using the word real
ause anything to exist. In fa t, the notion of the real number system has been
understood in di erent ways at di erent times, and the urrent orthodoxy may
well no longer be a epted a generation from now. These matters are beyond
the s ope of this ourse.

A rational number is a ratio p/q where p and q are integers and q is


not 0. But 3/2, 15/10, (60)/(40), and so on are all the same rational
number, so really ea h p/q is only a name of a rational number, and ea h

2.1 Preliminaries

27

rational number has in nitely many names. The usual hoi e of name for a
rational number is the one where p and q are in lowest terms (i.e., they have
no ommon fa tor greater than 1) and q is positive|for example, 5/3 rather
than 15/(9).
Not all real numbers are rational. The standard example of an irrational
number is the square root of 2. The argument is that if the square root of 2
takes the form p/q then p2 /q2 = 2, and so p2 = 2q2 . But p and q ea h
take the form of a power of 2 times an odd number (i.e., p = 2e p~ where p~
is odd, and q = 2f q~ where q~ is odd), so p2 and q2 ea h take the form of an
even power of 2 times an odd number (i.e., p2 = 22e p~2 and p~2 is odd, and
q2 = 22f q
~ 2 and q~ 2 is odd), so p2 is divisible by an even number of 2's while
2
2q is divisible by an odd number of 2's. Therefore p2 an't equal 2q2 , and
the assumption that the square root of 2 takes the form p/q is untenable.
However, this argument relies on a ta it assumption that an integer fa tors
in only one way as a power of 2 times an odd number. The ta it assumption
is true, but its proof takes a nontrivial e ort.
Subje t to the usual rules of algebra. The re ipro al of a sum is not the sum
of the re ipro als:
In general,

1
a+b

is not

1
1
+ .
a b

For example, try a = b = 1: 1/(1+1) = 1/2 is not 1/1+1/1 = 2. But students


whose algebra is rusty sometimes slip up on this point.
Division by 0 is prohibited. This prohibition is not arbitrary.
 One explanation is that to divide by a number b is to multiply by its
re ipro al, the number b su h that bb = 1. But 0 has no re ipro al sin e
0b = 0 6= 1 for all b , and so division by 0 makes no sense.
 A se ond explanation begins by observing that to say that a/b = c is to
say that c is the number su h that a = bc. So for b = 0, to say that
a/0 = c is to say that c is the number su h that a = 0c. If a 6= 0 then
no su h c exists, while if a = 0 then any c will work, i.e., all values are
equally plausible hoi es for 0/0, and so no one value an be preferred.
Nonetheless, the parti ular hoi es of 0 or 1 as the de nition of 0/0 are

often put forward as somehow being natural and not ontradi ting this
se ond explanation.
 But a third explanation shows that any de nition of 0/0 leads qui kly to
nonsense. The general rule for adding fra tions is
ad + bc
a c
+ =
,
b d
bd

so that, if 0/0 is to have meaning, for any number a,

28

2 The Rational Power Fun tion


a0+10
0
a 0
+ =
= .
1 0
10
0

Subtra t 0/0 from both sides to get a/1 = 0, i.e., a = 0. That is, if 0/0 is
de ned then all numbers must be 0.
Truly, division by 0 is a bad idea, even when the numerator is 0 as well.
In nity is not a real number. There exist extensions of the real number
system that ontain the symbols and and rules su h as a + =
and a = for all nite numbers a. But these extensions ompromise
the integrity of the original system's algebra, requiring vigilan e for ases and
leaving new operations unde ned, su h as . In this ontext, note that
the equality a/b + 0/0 = 0/0 from the previous omment suggests that
or rather than 0 or 1 might be a plausible de nition of 0/0. Predi tably,
either of these de nitions leads to its own set of problems.
In fa t, our assumptions so far about the real number system are inadequate for al ulus. The pro ess of working through al ulus examples will
illustrate the various additional assumptions that the subje t requires.
Exercises
2.1.1. Argue similarly to the text that the square root of 3 is irrational. Why
doesn't the argument apply to the square root of 4? In general, for what
positive integers n is the square root of n rational?
2.1.2. Are there numbers a and b for whi h is it true that \by a ident"
1/(a + b) does equal 1/a + 1/b?
2.1.2 The Finite Geometric Sum Formula

For any real number r 6= 1 and any positive integer n, the sum of the rst
onse utive n powers of r (starting at the 0th power r0 = 1) is
1 + r + r2 + + rn1 =

rn 1
.
r1

This formula is the nite geometri sum formula. It redu es a sum of many
terms to a quotient of two terms. The quantity whose powers we are summing
is denoted r be ause it is the ratio of ea h pair of onse utive terms: 1 and r,
r and r2 , and so on. When onvenient (espe ially when 1 < r < 1, i.e., when
1 is larger in magnitude than r and its powers), we make the numerator and
the denominator of the fra tion positive by writing instead

2.2 The Rational Power Fun tion


1 + r + r2 + + rn1 =

29

1 rn
.
1r

The two ways of writing the nite geometri sum formula have the exa t same
ontent. To prove the formula, we may verify instead that the left side times
the right side denominator equals the right side numerator,
(1 + r + r2 + + rn1 )(1 r) = 1 rn ,

(2.1)

and this follows (exer ise 2.1.3) from multiplying out the left side of (2.1).
Exercises
2.1.3. Verify formula (2.1) by multiplying out its left side.
2.1.4. Use an appropriate hoi e of r to show that the nite geometri sum

formula reprodu es Ar himedes's al ulation of the sum in (1.8) on page 11.

2.2 The Rational Power Function


The following notation is onvenient to have at hand:
Z

= the integers,

Z1 = the positive integers,

Z0 = the nonnegative integers,


Z1 = the negative integers,
Q

= the rational numbers,

R>0 = the positive real numbers.

All of the symbols just introdu ed are names of sets. Set means olle tion of
elements. We take the notion of a set as something that will be omprehensible in our ontext. In fa t set theory leads to slippery issues very qui kly
(see exer ise 2.2.1), but what matters to us here is that the paradigm and the
notation of set theory are tremendously helpful for organizing one's thoughts
in the pro ess of doing mathemati s.
Also, the following symbol is ubiquitous in mathemati s:
\" means in or is in or is an element of.
And similarly,
\/ " means not in or is not in or is not an element of.

30

2 The Rational Power Fun tion

Thus, 1/2 Q (read one-half is an element of Q) be ause indeed 1/2 is a


rational number, but 1/2 / Z (read one-half is not in Z ) be ause 1/2 is not
an integer. Note: we do not write \Z Q". Every integer is indeed a rational
number, so that Z is a subset of Q, but the symbol \" denotes element
ontainment, not set ontainment. That is, the symbol \" is understood to
have an individual element to its left and a set ontaining the element to its
right. The language and notation of set theory will be dis ussed further in
hapter 3.
2.2.1 Definition of the Rational Power Function

For any rational number Q, the th power fun tion is denoted f ,


f (x) = x

for positive real numbers x R>0 .

The symbol-string \x " is easy enough to write down, but it is only notation.

Writing x does not address the question of what|if anything|raising a


positive real number x to a rational power a tually means. We approa h

the question systemati ally.


For any positive integer , de ne for any positive real number x,
x = x x x ( times) for Z1 .

Thus for example, exploiting the fa t that 1 is multipli atively neutral,


x3 = 1 x x x,
x2 = 1 x x,

x1 = 1 x,

and this pattern extends naturally to the de nition


x0 = 1.

For a negative integer (so is a positive integer), de ne for any positive


real number x,
x = 1/x for Z1 .

In general for a nonzero real number t, 1/t denotes the multipli ative inverse
of t, i.e., the number whose produ t with t equals 1. Thus the display says
that if is a negative integer, so that is a positive integer and we understand x for any positive real number x, then x is the number whose
produ t with x is 1. For example, for any x R>0 , x3 = 1/x3 is the
number that when multiplied by x3 gives 1; here = 3 and = 3.

2.2 The Rational Power Fun tion

31

Next let be the re ipro al of a positive integer, so that 1/ is itself a


positive integer. De ne for any positive real number x,
x = the unique positive number y su h that y1/ = x

for 1/ Z1 .

That is,
x1/n = the unique positive number y su h that yn = x

for n Z1 .

For the just-displayed de nition of x1/n to make sense, there must be at least
one suitable y, and there must be at most one su h y. For now we assume
that these onditions do hold, so that indeed a unique y exists, making the
de nition sensible. This y is alled the positive nth root of x. For example,
x1/2 is the unique positive number y su h that y2 = x, i.e., x1/2 is the
positive square root of x. Thus 41/2 unambiguously means 2, even though 2
also squares to 4. The de nition of x1/n as the unique positive nth root of x
relies on an assumption about the real number system beyond those that we
have already made:


Every positive real number has a unique positive nth root for any
positive integer n.

To nish de ning the rational power fun tion, let = p/q be any rational
number whatsoever, where p is an integer and q is a positive integer. De ne
for any positive real number x,
x = (x1/q )p

for = p/q Q, p Z , q Z1 .

One an show (see exer ise 2.2.2 for a partial proof) that if also = p /q

where p is an integer and q is a positive integer then (x1/q )p = (x1/q )p ,


and so the de nition of x is independent of how the rational exponent is
represented. This ompletes the de nition of the power fun tion.
The relevant body of algebra in this ontext is the laws of exponents.
These state that for any positive real numbers x and y, and for any rational
numbers and ,
x x = x+ ,

(x ) = x = (x ) ,

x y = (xy) .

One an show that the laws of exponents for rational powers are onsequen es
of our de nition of raising a positive real number to a rational power. However, sin e doing so is an exer ise in Math 112, we omit it here. It is worth
appre iating that the laws of exponents are uniform, i.e., even though the
de nition of the power fun tion pro eeded by ases, the laws of exponents
work the same way regardless of whether ea h of and is a nonnegative

32

2 The Rational Power Fun tion

integer, a negative integer, a positive rational number that is not an integer,


or a negative rational number that is not an integer.
If is a nonnegative integer then the given de nition of x an be extended
to all real numbers x, not only positive ones (e.g., we understand x3 = x x x
for any x, and we onsidered the squaring fun tion f(x) = x2 for any x in
hapter 1). Note that 0n = 0 for n Z1 but 00 = 1. Similarly, if is
a negative integer then the given de nition of x an be extended to all
nonzero real numbers x (e.g., we understand x3 = 1/x3 for any x 6= 0). And
if = 1/n for some positive integer n then the de nition of x = x1/n an
be extended to 0 = 0 always (sin e 0n = 0, i.e., the nth root of 0 is 0)
and also to negative values of x if n is odd (e.g., if y3 = 5 then (y)3 = 5
be ause 3 is odd, so that the ube root of 5 is the negative of the ube root
of 5). Finally, if = p/q where p is an integer and q is a positive integer
and the fra tion p/q is in lowest terms, then the de nition x = (x1/q )p an
be extended to all x if p is nonnegative and q is odd, to all nonzero x if p is
negative and q is odd, to all nonnegative x if p is nonnegative and q is even,
and to all positive x if p is negative and q is even. The multitude of ases is
bewildering, to say the least. To avoid onsidering ases in analyzing the th
power fun tion for general rational , we have simpli ed our lives by insisting
that its inputs be positive, and we will generally restri t our analysis of the
power fun tion to positive inputs. But the reader should be aware that by
standard onvention, the inputs to the th power fun tion are in fa t taken
to be
 all real numbers if is a nonnegative integer (for example, f3 (x) = x3 is
de ned for all x),
 all nonzero real numbers if is a negative integer (for example, f2 (x) =
x2 is de ned for all x 6= 0),
 all nonnegative real numbers if is a nonnegative rational number that is
not an integer (for example, f3/2 (x) = x3/2 is de ned for all x 0),
 all positive real numbers if is a negative rational number that is not an
integer (for example, f2/5 (x) = x2/5 is de ned for all x > 0).

Later in these notes, we will de ne the power fun tion for an arbitrary
real exponent , i.e., the exponent will no longer be restri ted to rational
values.
The last point to be made in this se tion is that ertain parti ular power
fun tions will arise frequently through these notes, and so the reader should
learn to re ognize them:

2.2 The Rational Power Fun tion


f0 is the onstant fun tion 1,

f0 (x) = 1 for all x,

f1 is the identity fun tion,

f1 (x) = x,

f1 is the re ipro al fun tion,

f1 (x) = 1/x.

33

Similarly, the reader should be qui kly able to re ognize f2 as the squaring
fun tion, f1/2 as the square root fun tion, and so on.
Exercises
2.2.1. Let S be the set whose elements are the sets that do not ontain themselves as an element. Does the set S ontain itself as an element?
2.2.2. Suppose that a positive rational number takes the forms = p/q
and = p /q where p, p , q, q Z1 . Let x be a positive real number. We

want to show that

(x1/q )p = (x1/q )p .

(a) An assumption in the se tion says that it su es to show instead that




(x1/q )p

qq




qq
= (x1/q )p
.

Explain.
(b) Without quoting the laws of exponents, explain why the de nition of
raising a real number to a positive integer exponent and then the de nition
of raising a real number to the re ipro al of a positive integer imply that


(x1/q )p

and similarly


(x1/q )p

qq

qq

pq


= (x1/q )pqq = (x1/q )q


= xpq ,

= (x1/q )p

qq




p q
= xp q .
= (x1/q )q

( ) Explain why the quantities on the right sides of the two displays in
part (b) are equal. This ompletes the argument.
2.2.3. Let x be a positive real number, and let and be rational numbers.
The symbol-string
x

has two plausible interpretations. Explain. Show by example that the two
interpretations an give di erent values. Whi h interpretation is the preferred
one? Why?

34

2 The Rational Power Fun tion

2.2.2 Increasing/Decreasing Behavior

A fun tion f is alled strictly increasing if for any two input-values s and t
with t > s, also f(t) > f(s); that is, larger input-values yield larger outputvalues. Equivalently, f is stri tly in reasing if for any distin t input-values s
and t (distin t means that s 6= t), the input-di eren e t s and the outputdi eren e f(t) f(s) have the same sign. Similarly, f is strictly decreasing
if for any distin t input-values s and t, the input-di eren e t s and the
output-di eren e f(t) f(s) have opposite signs. Visually, the idea is that the
graph of a stri tly in reasing fun tion is higher in the y-dire tion over x-values
that are farther to the right, and similarly for stri tly de reasing fun tions.
We now show that

The power fun tion f for any rational number is stri tly in reasing if is positive and stri tly de reasing if is negative.
Having a omputer plot various power fun tions demonstrates the result visually, but showing it symboli ally is a signi ant intelle tual improvement over
taking omputer gures as God-given. To show the fa t, we need to ompare
the signs of an input-di eren e t s and the orresponding output-di eren e
f (t) f (s).
This rst step of the argument is to ompute for any positive real numbers
s and t, and any positive integer n (using the laws of exponents and the nite
geometri sum formula),
tn sn = sn ((t/s)n 1)
= sn ((t/s) 1)(1 + (t/s) + (t/s)2 + + (t/s)n1 )
= (t s)s

n1

n1

(1 + (t/s) + (t/s) + + (t/s)

(2.2)

).

In the last line of (2.2), sn1 and the sum are both positive, and so the
omputation has shown that
tn sn and t s have the same sign for s, t R>0 , n Z1 .

(2.3)

Next let n be a negative integer, so that n is a positive integer. Again


suppose that s and t are positive real numbers. Then 1/s and 1/t are also
positive real numbers. By de nition,
tn sn = (1/t)n (1/s)n .

Also, (2.3) with 1/s in pla e of s, 1/t in pla e of t, and n in pla e of n says
that
(1/t)n (1/s)n and 1/t 1/s have the same sign.

2.2 The Rational Power Fun tion

35

To study how the sign of 1/t 1/s relates to the sign of t s, suppose
rst that 1/t > 1/s. Multiplying the inequality by the positive quantity st
preserves its dire tion, giving s > t. Similarly, if 1/t < 1/s then s < t. That
is,
1/t 1/s and t s have the opposite signs.
And so, putting the last three displays together gives
tn sn and t s have opposite signs for s, t R>0 , n Z1 .

(2.4)

Now let = p/q where p is an integer and q is a positive integer. Suppose


that s and t are positive real numbers. Sin e s = (s1/q )p and t = (t1/q )p ,
(2.3) and (2.4) with n = p give
t s and t

1/q

1/q

have

the same sign if p > 0,


opposite signs if p < 0.

Also, sin e s = (s1/q )q and t = (t1/q )q , (2.3) with n = q gives


t s and t1/q s1/q have the same sign.

Combine the previous two displays to get


t s and t s have

the same sign if > 0,


opposite signs if < 0.

That is:
The rational power fun tion f (x) is

stri tly in reasing if > 0,


stri tly de reasing if < 0.

In the remaining ase = 0, the power fun tion f0 (x) is the onstant fun tion 1.
Exercise
2.2.4. (a) Continue the al ulation (2.2) in the text to establish a slight gen-

eralization of the nite geometri sum formula, the di eren e of powers


formula: For any positive real numbers s and t, and any positive integer n,
tn sn = (t s)(tn1 + tn2 s + tn3 s2 + + sn1 ).

(2.5)

(b) Write out the di eren e of powers formula for n = 1, n = 2, n = 3,


and n = 4.

36

2 The Rational Power Fun tion

2.3 Integration of a Particular Rational Power Function


2.3.1 The Problem

Find the area under the graph of the fun tion


f(x) = x2/3

from x = 1 to x = 8. The situation is shown in gure 2.1. Sin e f is the power


fun tion f where = 2/3 is positive, f is indeed stri tly in reasing as shown
in the gure.

4
f(x) = x2/3

PSfrag repla ements


1

8
Figure 2.1.

Area under a urve

Re all that we believe that raising any real number between 1 and 8 to
the two-thirds power is a meaningful thing to do. Spe i ally, it is understood
to mean taking the ube root and then squaring. Squaring a number is non ontroversial, sin e it is a spe ial ase of multiplying two numbers, something
that we have assumed we an do. Taking ube roots|and taking nth roots
in general|is not part of basi algebra, but it has been appended to the list
of things that we assume we may take for granted.
As with the quadrature of the parabola, a ta it assumption in our new
problem that an easily pass unnoti ed is the assumption that indeed there
is an area-number to be found.
2.3.2 Intuitive Vocabulary

Call a positive real number

2.3 Integration of a Parti ular Rational Power Fun tion

37

 large if it is very far away from 0,


 small if it is very lose to 0,
 medium-sized if it is moderately far away from or lose to 0.

Yes, these terms are too qualitative to be mathemati ally pre ise. So the language will need to be understood from ontext. The point is that being able
to tra k the qualitative sizes of various quantities fa ilitates insight into the
omputations of al ulus. Impre ision is not a eptable mathemati al methodology, but pre ision is guided by insight. Furthermore, omplete pre ision is
virtually never attainable, and it is not the goal in and of itself. Pre ision and
insight omplement ea h other in strengthening our understanding.
2.3.3 The Idea to be Demonstrated

The idea is:

Computing a medium-sized number an require al ulations that


use large and small numbers en route. Cal ulus en odes methods
for doing so.
Indeed, a al ulus is a stone or pebble. The pebbles of mathemati al al ulus

are the intermediate small numbers that generate a nal medium-sized one.
The end-results of al ulus on eal the intermediate steps in whi h the
large and the small are manipulated to obtain a medium-sized answer. Operationally the on ealment is a onvenien e sin e the al ulations are detailed,
but in pra ti e it means that students often learn to apply al ulus me hani ally, substituting values into its formulas, without appre iating its nesse. A
goal of this set of notes is that students do gain some sense of these matters.
In some situations, rules about numbers are plausibly obvious, but for
other s enarios there are no rules. For the following dis ussion, re all that our
informal taxonomy of large, small, and medium-sized applies only to positive
numbers.

 (Situations with plausible rules.) The sum of two large numbers is a large

number. Similarly for two medium-sized numbers and for two small numbers. The sum of a large number and any other positive number is again
large. The sum of a medium-sized number and a small number is again
medium-sized. The produ t of a small number and a medium-sized number is small. The produ t of a medium-sized number and a large number
is large. And so on.
These rules are plausible only at the level of intuition sin e (again) the
terms large, small, and medium-sized are impre ise. To illustrate the
impre ision, if the sum of two small numbers is again small, then the sum
of three numbers should be small too, sin e the sum of the rst two small

38

2 The Rational Power Fun tion

numbers is small and then the threefold sum is the sum of the small twofold
sum and the small third number. But by iterating this reasoning, the sum
of a thousand small numbers is small, or a million, and the intuition is
no longer valid. The intuition of al ulus is fragile be ause the a tual
al ulations an be deli ate.
 (A situation with no rule.) The sum of many small numbers an be large,
medium-sized, or small. For example,
101000 + 101000 + + 101000 (102000 times) = 101000 ,

while
101000 + 101000 + + 101000 (101000 times) = 1,

and
101000 + 101000 + + 101000 (10500 times) = 10500 .

In fa t, numbers su h as 101000 and 10500 are unimaginably large and


small in any sort of physi al terms. There are some 1077 elementary parti les in the universe, and 77 13 = 1001, so 101000 elementary parti les
would make roughly one-tenth of a universe of universes of universes of
universes of universes of universes of universes of universes of universes of
universes of universes of universes of universes. But this is of no onsequen e, sin e we are treating numbers as purely platoni entities, not as
des riptions of physi al quantities.

One master on ept of al ulus that we will study, the integral,


omes|in its simplest form|from sums of ever more, eversmaller numbers.

Thus although this bullet says that we do not know in general how su h
sums behave, the ones that arise in al ulus from reasonable situations
will behave well in the sense of produ ing medium-sized answers as they
should. We have already seen an example of this in hapter 1, where the
nite geometri sum
1
1+ +
4


 n 
 n1
 2
4
1
1
1
1
.
+ +
=
4
4
3
4

visibly tends to 4/3 when we add more and more terms by letting n grow.
situation with no rule.) The quotient of two small numbers
an be large, or medium-sized, or small. Indeed, the al ulations

 (Another

10500
= 10500 ,
101000

provide examples.

101000
= 1,
101000

101000
= 10500
10500

2.3 Integration of a Parti ular Rational Power Fun tion

39

The other master on ept of al ulus that we will study, the


omes from quotients of two ever-smaller numbers.

derivative,

This bullet says that su h quotients an behave wildly, but again the ones
that arise in al ulus from reasonable situations will produ e medium-sized
answers. We have already seen an example of this in hapter 1, where the
di eren e-quotient
s2 x 2
,
sx

s 6= x

is a ratio of terms that both grow small as s tends to x, but sin e the
di eren e-quotient is also s + x, it is medium-sized, and it visibly tends
to 2x as s tends to x.
 (Not-really-another situation with no rule.) The produ t of a small number and a large number ould be small, medium-sized, or large. This is
nothing new be ause the produ t an be interpreted as a quotient of two
small numbers, or as the quotient of two large numbers. Spe i ally, if a is
small and b is large then the re ipro als a1 and b1 are large and small,
and ab = a/b1 = b/a1 .
To repeat, an intuitive understanding of al ulus is an understanding of how
to ompute medium-sized quantities using very large and very small numbers orre tly en route. The intermediate steps will require are sin e their
workings are not immediately transparent to our intuition.
Exercise
2.3.1. Des ribe more situations with plausible rules.
2.3.4 The Problem Again, and the Pending Calculation

Re all the problem: Find the area under the graph of the fun tion
f(x) = x2/3

from x = 1 to x = 8. We are going to approximate the area by al ulating


the areas of many boxes, as shown in gure 2.2. Here are some features to
observe about the gure:
 The region in question is roughly a trapezoid, so our eventual answer

should be roughly the orresponding trapezoid-area, the base times the


average of the heights, (8 1) (1 + 4)/2 = 17.5. But sin e the graph is
on ave (i.e., it bulges up in the middle, at least a ording to the omputer
that drew the gure), the true answer will be a little larger than this.
 Ea h box-height is determined by the value of the fun tion over the left
endpoint of the box-base.

40

2 The Rational Power Fun tion

4
f(x) = x2/3

PSfrag repla ements


1

8
Figure 2.2.

Box-areas under a urve

 The boxes do not all have the same width, but their widths seem to be

regular in some way, and be ause the graph of the height-determining fun tion f(x) = x2/3 is also regular, the box-areas appear regular in turn. The
visual regularity of the box-widths and box-areas will soon be explained
symboli ally.
 Figure 2.2 shows twenty boxes in parti ular, but the idea is to al ulate
for n boxes where n is a general-purpose symbol, and then at the end
of the al ulation, let n grow very large. Although right-more boxes are
wider, if n in reases enough then plausibly even the rightmost box will
grow narrow, and so the boxes will tend toward lling all of the region
under the graph.
We will obtain a formula for the sum of the box-areas. Initially, the formula
will be a sum of many small numbers, and so its nature will be un lear.
But patient al ulation will manipulate the formula into an expression that
involves only medium-sized numbers, making it easy to understand. Only
then will we let the number of boxes grow very large and see to what number
the sum of their areas tends.
2.3.5 Tools To Be Used

The laws of exponents. Again, these state that for any positive real num-

bers x and y, and for any rational numbers and ,


x x = x+ ,

(x ) = x = (x ) ,

x y = (xy) .

2.3 Integration of a Parti ular Rational Power Fun tion




The nite geometri sum formula. Again, the formula is that for any
real number r 6= 1 and any positive integer n,
1 + r + r2 + + rn1 =

rn 1
,
r1

1 + r + r2 + + rn1 =

1 rn
.
1r

or


41

Algebra. As mentioned already, the idea is to al ulate for n boxes where

n is a general-purpose symbol. Consequently, various other quantities in


the al ulation will have to be represented by symbols as well rather than
numbers, be ause they depend on n. Only at the end of the al ulation,
when we let n grow very large, will the symbols that we are working with
nally yield an a tual number as the answer.

Working through the al ulation will also require patien e, attention-span,


persisten e, and study-skills. Sin e the problem being solved is nontrivial, the
solution is larger than bite-sized, perhaps too mu h to pro ess in one reading.
Even for several readings, having a pen and s rat h paper at hand to keep
tra k of the main quantities in play may be helpful.
2.3.6 The Geometric Partition

We return to the problem of nding the area under the graph of the fun tion
f(x) = x2/3 from x = 1 to x = 8. Again see gure 2.2. Throughout the
following al ulation, one fundamental quantity is driving everything else:
The number of boxes is n.
Thus

n is large.

As already explained, the gure shows twenty boxes but the idea is to al ulate
for a generi number of boxes, and then only after the al ulation yields its
result, the number of boxes will then grow very large. Make the following
de nition:
The rst partition point is s = 81/n .
That is, the rst partition point s is a real number|dependent on the number
of boxes|that is greater than 1. In gure 2.2, s is the right endpoint of the
base of the leftmost box. To rephrase the de nition:
The rst partition point is the positive number s su h that sn = 8.

42

2 The Rational Power Fun tion

Note that 1n = 1, while 2n is large when n is large. So s lies between 1 and 2,


and the more boxes there are, the loser s tends to 1 from the right. Thus
(exer ise 2.3.2)
s 1 is small.
Divide the x-axis from x = 1 to x = 8 into n intervals having the partition
points
x0 = s0 = 1,
x1 = s1 = s,
x 2 = s2 ,
x 3 = s3 ,

..
.
xn1 = sn1 ,
xn = sn = 8.

That is, using the symbol i to serve as a ounter:


The partition points are xi = si for i = 0, , n.

This partition of the x-axis from x = 1 to x = 8 is a geometri partition (see


gure 2.3, in whi h n = 10), as ompared to a uniform partition, where all
intervals have the same width. The geometri partition will lead ni ely to a
geometri sum in our pending area- al ulation. It does so for reasons that rely
on the fun tion f(x) = x2/3 of our example being a rational power fun tion.
The hoi e of a geometri partition rather than a uniform partition to solve
PSfrag repla ements
our integration problem is guided by hindsight, an example of the artfulness
of al ulus.

1 = s0 s1s2

si1 si

Figure 2.3.

sn1

sn = 8

A geometri partition

Exercise
2.3.2. This exer ise quanti es the assertion that if s = 81/n then s1 is small
when n is large. More generally, let b be any real number greater than 1, and
let s = b1/n . Here b is xed but the positive integer n varies.

2.3 Integration of a Parti ular Rational Power Fun tion

43

(a) Explain why s 6= 1.


(b) Explain why
s1=

( ) Explain why

b1
.
1 + s + + + sn1
s2

s1<

b1
,
n

and therefore s 1 is small.


(d) If 0 < b < 1 (instead of b > 1) then what is the nature of s 1 when
n is large?
2.3.7 The Intervals and Their Widths

The intervals determined by the geometri partition are


I1 = the x-axis from x0 to x1 ,
I2 = the x-axis from x1 to x2 ,
I3 = the x-axis from x2 to x3 ,

..
.

In = the x-axis from xn1 to xn .

That is:
The intervals are Ii = the x-axis from xi1 to xi for i = 1, , n.
Their widths are
x1 = x1 x0 = s 1,
x2 = x2 x1 = s2 s = (s 1)s,
x3 = x3 x2 = s3 s2 = (s 1)s2 ,

..
.
xn = xn xn1 = sn sn1 = (s 1)sn1 ,

That is:
The interval-widths are xi = (s 1)si1 for i = 1, , n.
Thus the ith interval-width is the produ t of the small number s 1 with
a medium-sized number si1 . This symboli regularity in the formula for

44

2 The Rational Power Fun tion

the interval widths orresponds to the geometri regularity of the widths in


gure 2.2. Be ause s is greater than 1, the formula xi = (s 1)si1 shows
that the intervals are getting wider as i in reases, but even the greatest width,
(s 1)sn1 , is less than (s 1)sn = (81/n 1) 8, and as n gets large this
be omes a produ t of a small number and a medium-sized number, i.e., it
be omes small.
2.3.8 The Inner Box-Areas

The base of the ith box is xi . The height of the ith box is the value of the
fun tion f(x) = x2/3 at the left endpoint of the ith interval,
2/3

f(xi1 ) = xi1 = (si1 )2/3

for i = 1, , n.

Thus the area of the ith box is


xi f(xi1 ) = (s 1)si1 (si1 )2/3

for i = 1, , n.

But by the laws of exponents,


si1 (si1 )2/3 = (si1 )5/3 = (s5/3 )i1 .

And so:
The inner box-areas are (s 1)(s5/3 )i1 for i = 1, , n.
Thus the ith box-area is the produ t of the small number s1 with a mediumsized number (s5/3 )i1 . As with the interval-widths, the symboli regularity
in the formula for the inner box-areas orresponds to the geometri regularity
of the areas in gure 2.2.
2.3.9 The Sum of the Inner Box-Areas

Re all that we have n boxes and that s = 81/n . The sum of the inner box-areas
is
i
h
Sn = (s 1) (s5/3 )0 + (s5/3 )1 + (s5/3 )2 + + (s5/3 )n1
i
h
= (s 1) 1 + (s5/3 ) + (s5/3 )2 + + (s5/3 )n1 .

This is a small number, s 1, times a sum of many medium-sized numbers,


1 + s5/3 + (s5/3 )2 + + (s5/3 )n1 . So it is a small number times a large
number, and as su h, it does not have an obvious size. But, as anti ipated,
the happy hoi e of a geometri partition of the x-axis has redu ed the sum

2.3 Integration of a Parti ular Rational Power Fun tion

45

of inner box-areas to a nite geometri sum. Spe i ally, the sum in square
bra kets is a nite geometri sum with ratio r = s5/3 , and so by the nite
geometri sum formula,
Sn = (s 1)

(s5/3 )n 1
.
s5/3 1

So we have ollapsed the sum, with its many terms, to a quotient of only
two terms. But still the fa tor s 1 out front is small, as is the denominator
s5/3 1 of the fra tion. On the other hand, the numerator is 31, sin e sn = 8
and so (s5/3 )n = (sn )5/3 = 85/3 = 32. After rearranging, the sum of the inner
box-areas is
s1
, s = 81/n .
Sn = 31 5/3
(2.6)
s

The 31 is quintessentially medium-sized, but the numerator and the denominator of the fra tion are both small.
We make a substitution to eliminate the fra tional exponent 5/3 from our
expression for Sn . Let
s~ = s1/3 = 21/n .
So s~ is slightly bigger than 1. That is,
~s 1 is small.
The sum of the inner box-areas is now
Sn = 31

whi h rewrites as

s~3 1
,
~s5 1


s~3 1
s~ 1
,
Sn = 31  5
s~ 1
s~ 1

s~ = 21/n ,

s~ = 21/n .

Rewriting Sn this way may not seem to help matters, sin e s~3 1, ~s 1, and
s~2 1 are all small. But it sets up the nite geometri sum formula twi e
more (sin e s~ 6= 1), expanding sums now rather than ollapsing them:
The inner box-area sum is Sn = 31

1 + s~ + ~
s2
,
1 + s~ + ~
s2 + ~
s3 + ~
s4

s~ = 21/n .

And sin e s~ is lose to 1, the numerator and the denominator of the fra tion
are now medium-sized. Our pres ient hoi e to use the geometri partition,
and then our patient e ort of
 deriving a long geometri sum,

46

2 The Rational Power Fun tion

 ollapsing it to a quotient,
 rearranging the quotient,
 and nally expanding two short geometri sums in the numerator and the

denominator of the quotient

have eliminated all large or small numbers from the formula for the sum of
the box-areas.
2.3.10 The Limiting Value

Finally the al ulation an give a meaningful medium-sized answer. As the


number n of boxes grows very large, the auxiliary quantity s~ = 21/n will
tend to 1, and so the sum of inner box-areas will tend to an easily al ulable
number,
1+1+1
93
Sn tends to 31
=
= 18.6.
1+1+1+1+1

Sin e the boxes are lling up the region under this urve, this number must
be the area. And indeed, it is slightly larger than the original underestimate
of 17.5. Summarizing,
The area under the graph of f(x) = x2/3 from x = 1 to x = 8 is 18.6.
Or, introdu ing some notation,
Z8

f = 18.6

where f(x) = x2/3 .

That is, the integral sign \ " is simply shorthand for the area under the
graph.
This is al ulus.
Exercises
2.3.3. Show that in a al ulation similar to the one in the se tion but using

outer boxes rather than inner boxes gives the following result:

The ith outer box-area is s2/3 times the ith inner box-area, i = 1, , n.
Therefore the sum of the outer box-areas is s2/3 times the sum of the inner
box-areas. To what value does s2/3 tend as the number n of boxes grows? To
what value does the sum of the outer box-areas onsequently tend?

2.4 Di erentiation of the Rational Power Fun tion

47

1
f(x) = x1/4

PSfrag repla ements 1/2

16
Figure 2.4.

Boxes for exer ise 2.3.4

2.3.4. Find the area under the graph of the fun tion f(x) = x1/4 = 1/x1/4
from x = 1 to x = 16, using outer boxes. (See gure 2.4. Here the power
fun tion f = f is stri tly de reasing be ause = 1/4 is negative.) Your

writeup should review the ideas of the se tion.

2.3.5. Find the area under the graph of the fun tion f(x) = x7/4 from
x = 1 to x = 16. (For this fun tion, the pi ture still looks qualitatively

like gure 2.4 be ause of the negative exponent, but the algebra involves a
new wrinkle: your al ulations should lead you to an expression involving the
quantity 1/(s3/4 1), di erent from the example in the se tion and from
the previous exer ise be ause of the negative exponent. However, note that
1
s3/4

s3/4
,
1

s3/4

and now the al ulation an ontinue as before.)


2.3.6. Try to apply the same ideas to nd the area under the graph of the
fun tion f(x) = 1/x = x1 from x = 1 to x = 10. This time the pro ess breaks
down. Where does it do so, and why? For what rational exponents will the
power fun tion f(x) = x lead to this breakdown?

2.4 Differentiation of the Rational Power Function


2.4.1 The Problem

Re all that for any rational number , the th power fun tion is denoted f ,

48

2 The Rational Power Fun tion


f (x) = x

for positive real numbers x R>0 .

The problem is: For any x > 0, nd the limiting value of the di eren e-

quotient
as

f (s) f (x)
sx

tends to x. As with the squaring fun tion in hapter 1, the numerator

in the display is the di eren e of the output-values of the power fun tion f
at an input s 6= x and at x itself, and the denominator is the di eren e of the
orresponding input-values s and x.
2.4.2 The Calculation

Consider the spe ial ase that is a nonnegative integer. Let s R>0 be a
positive number other than 1. Re all the nite geometri sum formula, but
with the r in the formula being the s here,
s 1
= 1 + s + s2 + + s1 ,
s1

s 6= 1.

(The sum on the right side is understood to be 0 if = 0.) As mentioned


above, the left side of the previous display is a di eren e-quotient. On the
other hand, the right side of the display is an -fold sum. If the inputdi eren e s 1 is very small then the summands all tend to 1, and so:
For Z0 , the limiting value of

f (s) f (1)
s1

as s tends to 1 is .

(The argument just given supports the statement in the ase = 0 if we


understand the summands all tend to 1 to be va uous in that ase.) So far
the boxed result holds only if is a nonnegative integer. The goal of this
se tion is to show that the boxed result holds when is any rational number
whatsoever, and then to generalize the \1" in the formula to any positive
number x R>0 .
Suppose next that Z1 is a negative integer. Thus now is a
positive integer. That is, the boxed result holds with in pla e of , and
we want to re-establish the boxed result for itself. Note that
x = (1/x)

for positive real numbers x R>0 .

This formula is useful be ause (1/x) is a positive integer power, the sort of
thing that we analyzed a moment ago. The idea now is to redu e the behavior
of the negative integer power to that of the positive integer power. By the
previous display and a little algebra,

2.4 Di erentiation of the Rational Power Fun tion

49

s 1
1 (1/s) 1
=
s1
s
1/s 1
t 1
= t
where t = 1/s, s 6= 1.
t1

If s tends to 1 then so does t. Thus the t on the right side of the previous
display tends to 1 as s tends to 1. The fra tion on the right side of the
display tends to by the previous al ulation. So the entire right side tends
to (1)() = . That is, the boxed result has been extended to all integers:
For Z , the limiting value of

f (s) f (1)
s1

as s tends to 1 is .

Now let = p/q where p and q are integers with q nonzero. Then
s = (s1/q )p

for positive real numbers s R>0 .

Consequently,
s 1
(s1/q )p 1 s1/q 1

=
s1
s1
s1/q 1
tp 1 t 1

where t = s1/q , s 6= 1.
=
t 1 tq 1

As s tends to 1, so does t. By the results already established, the quotients


on the right side of the previous display tend respe tively to p and 1/q as s
tends to 1. Thus their produ t tends to p/q, i.e., it tends to . Now the boxed
result has been extended to all rational numbers:
For Q, the limiting value of

f (s) f (1)
s1

as s tends to 1 is .

Finally, repla e the normalized value 1 by any positive number x R>0 .


Then
s x
x ((s/x) 1)
=
sx
x((s/x) 1)
t 1
where t = s/x, s 6= x.
= x1
t1

If s tends to x then t tends to 1, and so the quotient on the right side of the
previous display tends to . The boxed result has been extended from x = 1
to any positive real number x R>0 :
For Q, the limiting value of

f (s) f (x)
sx

as s tends to x is x1 .

50

2 The Rational Power Fun tion

This ompletes the argument. Again introdu ing some notation, the on lusion is
For Q, f = f1 .

That is, the prime is simply shorthand for the limiting value of di eren equotients.

Note that when = 2, we re over the formula for the tangent slope of
the parabola: the derivative of the squaring fun tion f2 (x) = x2 for all x > 0
is the fun tion 2f1 (x) = 2x for all x > 0. Similarly, the derivative of the
identity fun tion f1 (x) = x for all x > 0 is the onstant fun tion f0 (x) = 1 for
all x > 0, and the derivative of the onstant fun tion f0 (x) = 1 for all x > 0 is
the onstant fun tion 0f1 (x) = 0 for all x > 0. The reader should understand
these last two fa ts in terms of tangent slopes (exer ise 2.4.3).
Exercises

2.4.1. Is there a rational power fun tion f whose derivative is f1 ? Is there


a rational power fun tion f whose derivative is any onstant multiple of f1 ?
2.4.2. The last boxed result in the se tion took four steps to derive. Rederive
it in three steps instead by using the di eren e of powers formula (2.5) from
exer ise 2.2.4.
2.4.3. (a) Graph the fun tion f1 (x) = x for all x > 0. For any x, what is the
tangent slope to the graph at the point (x, f(x))?
(b) Graph the fun tion f0 (x) = 1 for all x > 0. For any x, what is the
tangent slope to the graph at the point (x, f(x))?
2.4.3 A Fundamental Observation

The area al ulation in se tion 2.3 redu ed the problem of studying an

integral|the limiting value of sums of many small terms|to the problem

of studying the limiting value of quotients of two small terms. Spe i ally,
omputing the area under the graph of the power fun tion
f2/3 (x) = x2/3

from x = 1 to x = 8 led to equation (2.6) on page 45, now slightly rewritten,


. s5/3 1
Sn = 31
,
s1

s = 81/n .

Here Sn is the sum of the inner box-areas for n boxes, and the question was
to what value Sn tends as the number n of boxes grows.

2.5 Integration of the Rational Power Fun tion

51

In fa t, as n grows, s tends to 1, and so the area al ulation is redu ed to


the derivative al ulation of se tion 2.4. That al ulation says that
the limiting value of

f5/3 (s) f5/3 (1)


s1

as s tends to 1 is 5/3,

and so now we an nish the integration more qui kly than we did in se tion 2.3,

the limiting value of Sn is 31 (5/3) = 18.6.

In fa t, a rereading of se tion 2.4 and then se tion 2.3 from equation (2.6)
to the end shows that the general derivative al ulation en odes the end al ulation of the integral as a spe ial ase.
On the fa e of things, the original integration problem is unrelated to any
derivative, and yet the al ulation redu ed to a derivative: not the derivative
of the original power fun tion f2/3 , but of a di erent power fun tion f5/3
instead. Computing the derivative thus enabled us to ompute the integral. A
result alled the Fundamental Theorem of Calculus will tell us that this
was no uke. Derivative-values will give integral-values under a wide range of
ir umstan es.

2.5 Integration of the Rational Power Function


The solution of a slightly more general integration problem than the one
in se tion 2.3 should be digestible now. The only new issue is that the left
endpoint 1, the right endpoint 8, and the power 2/3 will be ome general
symbols a, b, and . Thus the problem is: Let a and b be real numbers with
0 < a < b, and let 6= 1 be a rational number. Find the area under the
graph of the fun tion
f (x) = x

from

x = a to x = b. If > 0 then f is stri tly in reasing, while if < 0


then f is stri tly de reasing, but this will turn out to be irrelevant. The
odd-seeming restri tion that 6= 1 will emerge naturally from the pending
al ulation.
2.5.1 The Normalized Case

First onsider the ase where the left endpoint is still 1 and the right endpoint
is b where b > 1. As before, let n be the number of boxes, and let s = b1/n ,
i.e., s is the positive number su h that sn = b. As shown in exer ise 2.3.2,
s 1 is small. The points of the relevant geometri partition are again

52

2 The Rational Power Fun tion


x i = si

for i = 0, , n,

the intervals determined by the geometri partition are


Ii = the x-axis from xi1 to xi

for i = 1, , n,

and their widths are


xi = (s 1)si1

for i = 1, , n.

The base of the ith box is xi . The height of the ith box is the value of the
fun tion f over the left endpoint of the ith interval,
i1
f (xi1 ) = x
)
i1 = (s

for i = 1, , n.

Thus the area of the ith box is


xi f (xi1 ) = (s 1)si1 (si1 ) = (s 1)(s+1 )i1

for i = 1, , n.

The sum of the box-areas is onsequently




Sn = (s 1) 1 + (s+1 ) + (s+1 )2 + + (s+1 )n1 .

Be ause 6= 1, the ratio s+1 in the geometri sum is not 1, and so the
nite geometri sum formula applies,


Sn = (s 1) 1 + (s+1 ) + (s+1 )2 + + (s+1 )n1
(s+1 )n 1
s+1 1
s1
= ((sn )+1 1) +1
s
1
. s+1 1
, s = b1/n .
= (b+1 1)
s1

= (s 1)

The derivative al ulation in se tion 2.4 at x = 1 shows that therefore:


The limiting value of Sn as n gets large is

b+1 1
.
+1

A similar al ulation using the right endpoints of the intervals gives the
same limiting value. Now the box-heights are
f(xi ) = f(si ) = (si ) = s (si1 ) = s f(si1 ) = s f(xi1 ),

so that ea h box-area is multiplied by s , giving box-area sums


Tn = s Sn .

2.5 Integration of the Rational Power Fun tion

53

And be ause s tends to 1 (exer ise 2.5.1), Tn tends to the same limiting
value as Sn as the number of boxes grows.
If > 0, so that f is stri tly in reasing, then the boxes whose areas sum
to Sn lie beneath the graph of f from 1 to b, and so the values Sn are all less
than the area under the graph; similarly the values Tn are all greater than the
area if > 0. And if < 0, so that f is stri tly de reasing, then onversely.
In either ase, the ommon value tended to by Sn and Tn must be the area
trapped between them. Summarizing, for any rational number 6= 1 and
any real number b > 1,
The area under the graph of f (x) = x from x = 1 to x = b is

b+1 1
.
+1

And in more mathemati al notation,


Zb
1

f =

b+1 1
,
+1

Q, 6= 1, b > 1.

(2.7)

The end of the integration argument, as presented here, has improved


over its prior in arnations in se tions 1.2 (quadrature of the parabola) and 2.3
(integration of f2/3 from 1 to 8). In those se tions, the argument was that as n
grows, the triangles ll the paraboli region, or the boxes ll the region under
the power fun tion's graph. Now the argument is less reliant on geometry and
more on numbers: as n grows, box-area sums too small to be the desired area
and box-area sums too large to be the desired area tend to a ommon value,
and so this value must be the desired area. This point is important. So far
we have been using the terms area and integral roughly as synonyms, but a
better approximation to the right idea is that:

An integral is an area that is the ommon limiting value of boxarea sums that are at most big enough and box-area sums that are
at least big enough.
This language will be made quantitative at the end of the next hapter.
Exercises
2.5.1. Let b > 1 be a real number. Let n be a positive integer and let s = b1/n .
Let = p/q be a rational number, with p an integer and q a positive integer.
This exer ise shows that s tends to 1 as n grows.
(a) Explain why s = b~ 1/n where b~ = b .

(b) Explain why exer ise 2.3.2 (page 42) now ompletes the argument.

2.5.2. What happens in the al ulation of

Rb
1

f when = 0?

54

2 The Rational Power Fun tion

2.5.2 The General Case

The al ulation so far has been normalized in that its left endpoint is 1. To
hange the left endpoint to an arbitrary positive real number a, we rst give
a geometri argument using boxes to establish the following proposition.
Proposition 2.5.1 (Scaling Result for the Power Function). Let a,
b, and c be real numbers with 0 < a b and c > 0. Let be any rational
number, in luding the possibility = 1. Then
Z bc
ac

f = c+1

Zb

f .

See gure 2.5. In the gure, the s aled interval [ac, bc] lies entirely to the
right of the original interval [a, b], but in general this need not be the ase:
the s aled interval an also lie to the right of the original interval but with
overlap, or to the left of the original interval but with overlap, or entirely to
the left of the original interval (exer ise 2.5.3).

y = x2/3

PSfrag repla ements

Figure 2.5.

ac

bc

The right area is c5/3 times the left one

In our present ontext of integrating the power fun tion f we are assuming that 6= 1. The proposition is being presented as a self- ontained entity
be ause we will refer to it again in hapter 5 for the ase = 1, to whi h it
applies as well.
The proof of the proposition pro eeds as follows. First let b 1, let n
be a positive integer, let s = b1/n , and re all our ma hinery from the normalized al ulation|the partition points xi of [1, b], the interval-widths xi ,

2.5 Integration of the Rational Power Fun tion

55

the heights f (xi1 ) over the left endpoints, and the heights f (xi ) over the
right endpoints, ulminating in the box-area sums Sn and Tn = s Sn . The
left part of gure 2.6 illustrates the boxes whose areas sum to Sn in a ase
where > 0. Here Sn is less than the true area under the graph from 1 to b,
while Tn is greater than the true area. If instead < 0 then Sn is greater than
the true area and Tn is less than it. So far this dis ussion has only repeated
ideas from the normalized al ulation.

y = x2/3

PSfrag repla ements

Figure 2.6.

bc

The right box-area sum is c5/3 times the left one

Now let c > 0. S ale the partition of [1, b] by c to get a partition of [c, bc],
x
~i = cxi

for i = 0, , n.

These partition points determine subintervals of width


x
~i = c xi ,

heights over left endpoints


f (x
~i1 ) = f (cxi1 ) = c f (xi1 ),

and heights over right endpoints


f (x
~i ) = f (cxi ) = c f (xi ).

The box-areas are now x~i f (x~i1 ) = c+1 xi f (xi1 ) and x~i f (x~i ) =
c+1 xi f (xi ), and the box-area sums are now
en = c+1 Sn
S

and Ten = c+1 Tn .

(2.8)

56

2 The Rational Power Fun tion

The new boxes whose areas sum to Sen are shown in the right part of gure 2.6.
If > 0 then Sen is too small to be the true area under the graph of f from c
to bc, and so on, just as before.
Let n grow large. Then on the one hand, making no referen e to the
expli it formulas for Sn and Tn :

The fa t that Sn and Tn trap the area under the graph of f from 1
to b between them, and the fa ts that Tn = s Sn and s tends to 1,
ombine to show that Sn and Tn tend to the same limiting value,
that value being the area.
(For the reader who is justi ably uneasy with the argument just displayed in
itali s: itR will be shored up at the end of the next hapter.) Sin e
R Sn and Tn
b
+1 b
e
e
tend to 1 f , (2.8) shows that Sn and Tn both tend to c
f . But on
1
the other hand, the geometry underlying Sen and Ten shows that the ommon
R
value that they tend to must be the area trapped between them, cbc f .
Consequently:
If b 1 and c > 0 then

Z bc

+1

f = c

Zb

f .

(2.9)

And now, more generally, if 0 < a b and c > 0 then


Z bc

f = (ac)+1

ac

and

Z b/a

Zb
a

f = a+1

Z b/a
1

by (2.9) with b/a, ac for b, c

by (2.9) with b/a, a for b, c.

Combining the last two displays gives the desired result,


If 0 < a b and c > 0 then

Z bc
ac

f = c+1

Zb

f .

(2.10)

Thus Proposition 2.5.1 is proved. To review, the basi idea is that for the
power fun tion, s aling a box horizontally by the fa tor c s ales it verti ally
by c , giving an area-s aling fa tor of c+1 . This observation doesn't depend
on the boxes arising from the geometri partition in parti ular, or on our
being able to put the box-area sum into a tidy form. Instead, it depended on
the admittedly hand-waving argument displayed in itali s above. Again: we
will return to that argument in the following hapter.
With (2.10) in hand, we an omplete the integration of the power fun tion. Again let 6= 1 be a rational number, and let 0 < a b. Then

2.6 Summary
Zb

f =

Z (b/a)a
1a

= a+1

by basi algebra

Z b/a

57

by (2.10) with 1, b/a, a for a, b, c

(b/a)+1 1
+1
b+1 a+1
=
+1
= a+1

by the normalized result (2.7)


by algebra.

That is,
Zb
a

f =

b+1 a+1
,
+1

Q, 6= 1, 0 < a b.

(2.11)

This ompletes the integration of the power fun tion, ex luding the spe ial
ase of the re ipro al fun tion f1 . We will return to the integral of the
re ipro al fun tion in hapter 5.
Exercise
2.5.3. Given positive numbers a and b with a < b, give onditions on the
positive number c su h that
(a) [ac, bc] lies entirely to the right of [a, b],
(b) [ac, bc] lies partially to the right of [a, b] but with overlap,
( ) [ac, bc] lies partially to the left of [a, b] but with overlap,
(d) [ac, bc] lies entirely to the left of [a, b].

2.6 Summary
Having dis ussed the parabola very informally in the previous hapter and
then the power fun tion somewhat informally in this hapter, we now have
raised enough questions to make a loser dis ussion of al ulus ne essary,
and we now have worked through enough examples for the dis ussion to be
omprehensible.

3
Sequence Limits and the Integral

A limit is a value that is tended to, whether it is attained or not. The unexplained notion of tends to, whi h has served as the workhorse for nishing
o arguments in the previous two hapters, needs to be made quantitative.
A des ription of how to do so emerged in the nineteenth entury, long after
Newton and Leibniz. It has sin e stimulated generations of al ulus students,
for better or for worse.
The de nition of limit should be enlightening on e the student understands it by parsing it, seeing it used, and learning to use it. But be ause
the de nition involves two diagnosti quantities that intera t in a deli ate
way, and be ause working with the de nition requires skill with symbolmanipulation and with language in on ert with geometri intuition, oming
to terms with it an take some time, a resour e always in short supply during
a al ulus ourse. Hen e:

The student is en ouraged to engage with the limit arguments in


this hapter lightly and to taste.
Said engagement should give some sense of why the de nition of limit aptures
the right idea, and some sense of what form an argument using the de nition
should take. But it is mu h more important to understand the results|and
their uses|than to understand every detail of every argument that the results
hold.
Se tion 3.1 dis usses preliminary matters: sets, fun tions, and sequen es.
A sequen e is a spe ial kind of fun tion, naturally viewed as a list of data,
su h as the lists of su essive area-approximations that we generated during
the ourse of integrating the power fun tion. Se tion 3.2 de nes the limit of a
sequen e, the value to whi h the data are tending. With the limit of a sequen e
de ned, we an prove basi results about sequen e limits, and we an make
inferen es about unknown sequen e limits in terms of known ones, leading to

60

3 Sequen e Limits and the Integral

more results. Se tion 3.3 uses the results of this hapter to redo some of the
limit al ulations from hapters 1 and 2 more satisfyingly. On e we see the
methods work, it be omes lear that their s ope extends beyond the parti ular
instan e of the power fun tion. A pre ise and manipulable de nition of the
integral be omes natural to write down, and results about integration be ome
natural to prove. As a payo on our investment in de nitions, the language
suddenly, unexpe tedly, arries us farther in larity and results, with no more
omputational e ort. The serendipitous e onomy of ideas is pleasing.
Still, the de nition of limit given in this hapter is neither the alpha nor
the omega of the idea. Cal ulus ourished for enturies before this de nition
evolved. We should not be so arrogant as to presume that the great mathemati ians of the seventeenth and eighteenth enturies ouldn't understand
their subje t without the nineteenth entury de nition of limit. Indeed, a
1980 text alled Cal ulus Unlimited by Jerrold Marsden and Alan Weinstein
develops the material with no re ourse to limits at all. Nor should we believe
that the nineteenth entury de nition of limit is the end of the story. It prominently features the phrase there exists, whose meaning is still in ontention.
Does something exist only if we know an algorithm to ompute it, or does
it exist if its nonexisten e seems untenable, i.e., does it exist abstra tly as
ompared to omputationally? Do the two di erent notions of existen e lead
to di erent bodies of mathemati s? Sadly, a traditional rst al ulus ourse
has no time for these questions, but the student should be aware that they
are serious ones. A 2001 text alled Computable Cal ulus by Oliver Aberth
develops al ulus using only omputability.
In keeping with this hapter's attempt to be more te hni al mathemati ally than hapters 1 and 2, the writing onventions here will be di erent.
De nitions and propositions will be numbered, and proofs will be delineated.
The hange in style is not formalism for formalism's sake, but an attempt to
lay the ideas out learly.

3.1 Sets, Functions, and Sequences


3.1.1 Sets

As dis ussed in se tion 2.2, a set is a olle tion of elements. A set is often
des ribed by listing its elements in urly bra es,
S = {elements of S}.

The order in whi h the elements are listed is irrelevant, as are repeat listings
of the same element. Thus

3.1 Sets, Fun tions, and Sequen es

61

{2, 3} = {3, 2} = {2, 2, 3} = the set with elements 2 and 3.

Some ubiquitous sets in mathemati s are


= {integers} = {0, 1, 1, 2, 2, 3, },

Z0 = {natural numbers} = {0, 1, 2, 3, },


Z1 = {positive integers} = {1, 2, 3, },

Z1 = {negative integers} = {1, 2, 3, },


= {rational numbers},

= {real numbers},

R0 = {nonnegative real numbers},

R>0 = {positive real numbers},


R2

= {points in the plane}.

= the empty set = the set ontaining no elements.

The left urly bra e reads the set of or the set, so that, for example, the rst
line in the previous display reads altogether,
Z

is the set of integers, the set 0, 1, 1, 2, 2, 3, .

The empty set is not 0, nor is it {0}. In urly bra es notation,


= {}.

Perhaps the reason that the empty set is often onfused with 0 is that the
number of elements in the empty set is 0. However, a set is not the same thing
as the number of its elements.
Sets are often de ned by onditions. In this ontext, a olon \:" reads su h
that. So, for example, the notation
R2 = {(x, y) : x, y R}.

reads
R2

is the set of ordered pairs (x, y) su h that x and y are real numbers.

Ordered pair means a pair with one of its elements designated as the rst
of the two. Sin e R2 was de ned a moment ago as the set of points in the
plane, the last two displays give the appearan e of a rede nition. However,
the reader is assumed to be familiar with the representation of points in the
plane as ordered pairs of numbers, so that the last two displays only rephrase
the de nition of R2 rather than revise it. From now on, the terms point in
the plane and ordered pair of real numbers will be taken as synonyms.

62

3 Sequen e Limits and the Integral

(An unordered pair of numbers would be, for instan e, the set {2, 3},
whi h is also {3, 2}. That is, viewing the pair of numbers 2 and 3 as a set
does not onnote that one of them is innately the rst. By ontrast, the
notation (2, 3) expressly means the number-pair with 2 in its rst position
and 3 in its se ond. It is in fa t possible to formulate the notion of ordered
pair in terms of set theory rather than as a new primitive. Consider the
sets {{2, 3}, 2} and {{2, 3}, 3}. Ea h of these sets has another set as one of its
elements and a number its other element. Both of them an be understood
to spe ify the unordered pair {2, 3} and then to spe ify in addition whi h of 2
and 3 should be taken as the rst element of the orresponding ordered pair.
Similarly, ordered triples su h as (2, 3, 4), ordered quadruples, and so on an
all be de ned purely in terms of set theory, but on e this is done, ontinuing
to drag the resulting umbersome notation around is silly.)
Sets de ned by onditions also arise from the fa t that analyti geometry
des ribes geometri al obje ts by equalities and inequalities. The reader is
assumed to be familiar with su h representations. So, for example, the set
R = {(x, y) R2 : 1 x 8, 0 y x2/3 }

is the region between the x-axis and the graph of the power fun tion f2/3
from x = 1 to x = 8, depi ted ba k in gure 2.1 on page 36. The last omma
in the previous display is read and, and so the display reads altogether,
R is the set of
0 y x2/3 .

points

(x, y)

in the plane su h that

1x8

and

Here it is understood that \1 x 8" means that 1 x and x 8,


and similarly for any on atenation of inequalities. Inequalities should be
on atenated only if they point in the same dire tion, i.e., something like
\1 x > y" is poor style that an lead to al ulation errors.
Another type of set that is de ned by onditions is the interval. There
are nine types of interval (!):

Let
De ne the following subsets of R:
Definition 3.1.1 (Intervals).

and

be real numbers with

a b.

3.1 Sets, Fun tions, and Sequen es

63

(a, b) = {x R : a < x < b},

(a, b] = {x R : a < x b},

[a, b) = {x R : a x < b},

[a, b] = {x R : a x b},

(a, ) = {x R : a < x},

[a, ) = {x R : a x},

(, b) = {x R : x < b},

(, b] = {x R : x b},

(, ) = R.

A subset of R is alled an interval if it is a set of one of these nine types.

So, for example, (a, b) is the set of real numbers x su h that a < x < b.
Note that [a, a] = {a} and (a, a) = , showing that a set onsisting of just one
point is an interval and so is the empty set.
The use of the symbols \" and \" in the notation for some types of
interval is traditional, and it does uniformize the notations for the nine types.
But it is pedagogi ally regrettable sin e, as already mentioned, and
are not real numbers. However, note that they o ur only on the left sides of
the above equalities; that is, truly they are nothing but shorthand notation
to des ribe the sets on the right sides of the equalities, where they do not
appear. And in the notation, they always o ur adja ent to a parenthesis,
never a square bra ket, so even the shorthand notation does not suggest that
an interval ever ontains or . In any ase, alternative notations for the
fth through eighth types of interval are R>a , Ra , R<b , and Rb , while
the ninth type of interval really needs no notation sin e it is simply R.

If I is a nonempty interval of one of the


rst four types in De nition 3.1.1 then its endpoints are a and b. If I
is an interval of one of the next two types then its one endpoint is a.
If I is an interval of one of the next two types then its one endpoint
is b. The interval (, ) has no endpoints. The empty interval has no
endpoints.
Definition 3.1.2 (Endpoints).

Definition 3.1.3 (Open and Closed Intervals). An interval is closed


if it ontains all of its endpoints. An interval is open if it ontains none

of its endpoints.

Note that the mathemati al usages of open and losed need not be ex lusive or exhaustive. That is, an interval an on eivably be open, losed,
neither, or both. Exer ise 3.1.1 is to determine whi h of the nine types of
interval are open and/or losed.

64

3 Sequen e Limits and the Integral

The wording of the previous de nition (and of many de nitions to ome


in these notes) deserves a omment. The reader who is parsing grammar with
ex eptional are ould raise the following point:

The de nition says that if an interval ontains all of its endpoints


then it is losed.
But:
The de nition does not say that if an interval is losed then it
ontains all of its endpoints.
So, if we are told that an interval is losed, are we being told anything at
all? Yes, we are. Within the de nition, we an not yet talk about an interval

being losed implying anything until we have rst assigned meaning to the
notion of an interval being losed. But it is a onvention of mathemati s that
as soon as the meaning is assigned, the if ta itly evolves into an if and only
if. An interval that ontains all of its endpoints is losed, and a losed interval
ontains all of its endpoints. That is:
On e the de nition is stated, saying that an interval is losed is synonomous with saying that it ontains all of its endpoints.
A similar dis ussion applies to every further de nition in these notes that
takes the form \A if B": on e we are done reading the de nition, to say that
A holds is to say that B holds, and onversely.
Two sets whose elements are not numbers have already gured ta itly
through these notes. The rst is
P = {polygons in the plane}.

Elements of P are not numbers or individual planar points, but instead they
are planar regions. The reader should look ba k at gure 1.9, gure 1.10,
gure 2.2, and gure 2.4 (pages 9, 10, 40, and 47) to see that these gures
show elements of P and that these elements are highly relevant to integration.
As for the se ond parti ular set to be aware of, de ne a subset of the plane
to be bounded if some box ontains it, and de ne
B = {bounded subsets of the plane}.

So elements of B are planar regions too. The reader should look at gure 1.7
and gure 2.1 (pages 7 and 36) to see that these gures show nonpolygonal
elements of B , regions whose areas we wanted to nd. No gure in these notes
an a urately depi t a planar set that is not an element of B be ause the
entire gure will be ontained in a box: the page.

3.1 Sets, Fun tions, and Sequen es

65

Exercise
3.1.1. Let a and b be real numbers.
(a) Assume that a < b. For ea h of the rst four types of interval des ribed
in De nition 3.1.1, is the relevant interval of that type for su h a and b open?

Is it losed?
(b) Now drop the assumption that a < b. For ea h of the next ve types
of interval des ribed in De nition 3.1.1, is the relevant interval of that type
open? Is it losed?
( ) Is the one-point interval [a, a] open? Is it losed?
(d) Is the empty interval open? Is it losed?
3.1.2 Functions

The reader is presumed to have some experien e with fun tions. The basi
idea is that a fun tion re eives inputs and produ es outputs. We will notate
fun tions as follows:
f : A B.

Here A and B are sets, and f is a rule or a pro ess that assigns to ea h element
of A an element of B. The set A is alled the domain of the fun tion, and the
set B is a set alled the codomain of the fun tion. Thus the domain onsists
of all legal inputs to f, and the odomain onsists of all potential outputs
of f. Stri tly speaking, the fun tion onsists of all three data, A, B, and f, but
we often abbreviate it to f. The notation f : A B reads f is a fun tion
from A to B, or f maps A to B, or f from A to B, or various other phrasings
along these lines.
For example, onsider the fun tion
f : R R,

f(x) = x2 .

(3.1)

This is the squaring fun tion that takes all real numbers as its inputs and is
understood to produ e real numbers as its outputs. Note, however, that sin e
the square of any real number is nonnegative, not all points of the odomain R
are a tual outputs of f. This is an example of what was meant a moment ago
in des ribing the odomain as all potential outputs of f: all squares are real
numbers, but in fa t not all real numbers are squares. Sin e a fun tion stri tly
depends on its domain, its odomain, and its rule, the squaring fun tion that
takes all real numbers as its inputs and is understood to produ e nonnegative
real numbers as its outputs,
f : R R0 ,

f(x) = x2 ,

66

3 Sequen e Limits and the Integral

is not quite the same obje t as the squaring fun tion in (3.1). This distin tion
is admittedly pedanti , but on the other hand, the squaring fun tion from the
integers,
f : Z R,

f(x) = x2

is more ompellingly di erent from (3.1) be ause it has a di erent domain,


i.e., a di erent set of legal inputs. Thus although f(2) = 4 and f(2) = 4
as for the squaring fun tion (3.1) with domain R, now f(1/2) is not de ned
be ause 1/2 / Z .
The range of a fun tion is the set of its a tual outputs, a subset of the
odomain,
f(A) = {f(x) : x A} = {outputs of f}.
Any fun tion an have its odomain pruned down to its range and be rewritten
f : A f(A).

Under many ir umstan es we don't bother doing this when spe ifying a
fun tion, sin e the purpose of the odomain is only to give us some sense
of where the outputs of f are to be found. In fa t, when the des ription of
a fun tion's domain and rule make its odomain lear, the odomain an
go unmentioned. For example, saying \Let f(x) = x2 for x R" des ribes
the squaring fun tion on real inputs, whose outputs are understood to be
real numbers or nonnegative real numbers, it doesn't parti ularly matter.
(But saying only \Let f(x) = x2 " is too vague, unless the ontext has learly
established that the intent here is for x to vary through the real numbers.) On
the other hand, one ir umstan e where it is worth tightening the odomain
down to the range is when we want to follow f by a se ond fun tion,
g : B C,

i.e., when we want the outputs of f to serve as inputs to g. Doing so is sensible


only when the range f(A) of f is a subset of the domain B of g. For example, if
we want the squaring fun tion f = f2 and the square root fun tion g = f1/2 to
undo the e e ts of one another, we need to take are to spe ify their domains
and odomains ompatibly,
f : R0 R0

and g : R0 R0 .

A fun tion stri tly depends on its domain, its odomain, and its rule, but
not on its typography. For example, onsider the two fun tions
f : Q Q,

and

f(x) = x2 1

3.1 Sets, Fun tions, and Sequen es


g : Q Q,

67

g(y) = (y + 1)(y 1).

These two fun tions are the same fun tion even though their shared rule has
been given two names (f and g) and des ribed by di erent formulas with
di erent variables (x2 1 and (y + 1)(y 1)).
A fun tion is not the same thing as its graph. Let
f : A B

be a fun tion. The graph of f is a set of ordered pairs,


graph(f) = {(x, y) : x A, y B, y = f(x)} = {(x, f(x)) : x A}.
If the domain and odomain of f are subsets of R, then the graph of f an be
identi ed with a subset of the plane. The language here is entirely onsistent
with the des ription in hapter 1 of the parabola as the graph of the squaring
fun tion f(x) = x2 , other than the fa t that the hapter 1 des ription didn't
bother making expli it mention of the domain and odomain of f. Again, the
graph of a fun tion is not the fun tion itself. Students often refer to a
graph as a fun tion, and this is understandable be ause the graph des ribes
the fun tion visually, but for purposes of reasoning the distin tion is worth
retaining.
An algebraic function is any fun tion that an be built up from a nite
su ession of additions, subtra tions, multipli ations, divisions, and roots.
Thus a typi al algebrai fun tion is
f(x) =

(x2 + 1)1/2 x
x+1

1/3

The ta it understanding here is that the domain of f is the set of real numbers x for whi h the formula is sensible. That is, the domain of f is the set of
real numbers x whi h when substituted into the formula do not lead to any
square roots or ube roots of nonpositive numbers, or to a divide by 0. For
the fun tion f in the display, the domain is all real numbers x > 1.
Not all fun tions are algebrai . A nonalgebrai fun tion whose domain is a
suitable subset of R (the meaning of suitable isn't worth going into in detail
right now) and whose odomain is R is alled transcendental. We have
not yet seen any trans endental fun tions in these notes, but some examples
for the reader who may have seen them elsewhere are the logarithm, the
exponential fun tion, and the trigonometri fun tions.
The notion of a rule as part of a fun tion alls for some explanation. Just
as a fun tion is not a graph, a fun tion is not a formula. The term fun tion
(fun tio ) was introdu ed into mathemati s by Leibniz, and its meaning has
hanged ever sin e. During the seventeenth entury the ideas of fun tion and

68

3 Sequen e Limits and the Integral

urve were usually thought of as being the same, and a urve was often thought
of as the path of a moving point. By the eighteenth entury the idea of fun tion
was asso iated with analyti expression. Leonard Euler (1707{1783) gave the
following de nition:
A fun tion of a variable quantity is an analyti expression omposed in
any way whatsoever of the variable quantity and numbers or onstant
quantities.
Hen e every analyti expression, in whi h all omponent quantities
ex ept the variable z are onstants, will be a fun tion of that z; Thus
a + 3z; az 4z2 ; az + b a2 z2 ; cz ; et . are fun tions of z
The use of the notation \f(x)" to represent the value of f at x was introdu ed
by Euler in 1734. Our ontemporary notion of a fun tion as a rule is di erent
from Euler's notion unless every analyti expression is understood it to produ e output-values from input-values, and every rule or pro ess that produ es
output-values from input-values is understood to have an analyti expression.
If rules or pro esses are not the same thing as analyti expressions, then the
next question is just what rules/pro esses are sensible. Must we be able to
arry them out? What does arry them out mean?
Neither the inputs nor the outputs of a fun tion need even be numeri al.
For examples of nonnumeri al output, let P denote the set of all polygons
in the plane, and let a and b be real numbers with a < b. The pro ess in
se tion 1.2 of starting with one triangle ins ribed in the parabola with its left
and right endpoints over a and b, then adding two more smaller triangles,
then four more smaller-yet triangles, and so on, de nes a fun tion based on
the original endpoints a and b, whose input is the generation-number and
whose output is not a number at all, but rather is the orresponding polygonal
amalgamation of triangles,
pa,b : Z1 P.

That is, pa,b(n) is the nth generation polygonal approximation of the region
whose area we wanted to ompute. Similarly, the pro ess in se tion 2.3 of
omputing the area under the graph of the power fun tion f2/3 from x = 1
to x = 8 de nes a fun tion taking the number of boxes to the polygonal
amalgamation of boxes,
p : Z1 P.

That is, p(n) is the polygon onsisting of n boxes, shown in gure 2.2 for
n = 20 (page 40).
For an example of nonnumeri al input, let B denote the set of bounded
subsets of the plane, introdu ed on page 64. We would like an area-fun tion
Ar : B R0

3.1 Sets, Fun tions, and Sequen es

69

that assigns to ea h bounded subset of the plane a nonnegative real number to


be onsidered its area. Certainly area-fun tions exist (we ould simply assign
areas in some silly way), but the question is whether area-fun tions having
good properties exist. (Going into detail about good properties would take
us too far a eld, but they are very basi : the area of a box is its base times its
height, the area of two nonoverlapping sets is the sum of their areas, and so
on.) Similarly, let B3 denote the set of all bounded subsets of 3-dimensional
spa e. (A subset of spa e is bounded if it sits inside some 3-dimensional box.)
A volume-fun tion
Vol : B3 R0
would assign to ea h bounded subset of spa e a nonnegative real number
to be onsidered its volume, and would have good properties. Remarkably,
area-fun tions exist but volume-fun tions do not.
The invo ation that area fun tions exist will ease our lives onsiderably.
Let a and b be real numbers with a b. Let M be a nonnegative real number.
Consider a fun tion
f : [a, b] [0, M].

The region under the graph of f, a subset of the plane, is


R = {(x, y) R2 : a x b, 0 y f(x)},

and by our invo ation, it has an area,


Arba (f) = Ar(R).
Indeed, there may be more than one plausible area if f is strange enough|too
strange to draw or even to imagine visually, so this point is best not dwelled
on|and if we swit h our hoi e of area fun tion, however we are \ hoosing"
one in the rst pla e. The reader would be thoroughly justi ed in obje ting
that the previous senten e amounts to speaking in tongues rather than mathemati s, but the real point is that we are entering into a so ial ontra t: The

question of integrating f is not whether an area under its graph exists|it


abstra tly does, granting our invo ation of area fun tions|but whether
the area is a suitable limit of box-area sums. The invo ation of area is

an expedient that lets us nesse the existen e question. Yes, the existen e
question is important, but a one-semester al ulus ourse has no time to address it, espe ially sin e, as mentioned at the beginning of the hapter, the
mathemati al meaning of existen e is a live, arguable issue.
Exercise
3.1.2. Sket h the graphs of the following fun tions:

70

3 Sequen e Limits and the Integral

(a) f(x) = (x 1)2 for all x [0, 4],


(b) g(x) = (x 2)2 for all x [1, 3],
( ) h(x) = x2 1 for all x [2, 2],
(d) k(x) = x2 22 for all x [2, 2].
3.1.3 Sequences

A sequen e is a list of data. More spe i ally, a sequen e is one datum per
generation, where there is a starting generation and then an endless su ession
of generations thereafter. Formally, a sequen e is a fun tion whose domain is
the positive integers,
f : Z1 S.

The domain Z1 is the set of generations. The odomain S is often a subset


of the real numbers, but it need not always be. For instan e, we re ently
mentioned the sequen e of polygons arising from Ar himedes's quadrature
of the parabola in hapter 1, and the sequen e of polygons arising from the
integration of the power fun tion in hapter 2, both sequen es of the form
Any sequen e

f : Z1 P.
f : Z1 S

an be des ribed by listing its outputs, onsonantly with the idea of a sequen e
as a list,
(f(1), f(2), f(3), ).

Sequen es are usually written this way, with the domain and odomain ta it.
Furthermore, sequen es tend to have names su h as s or x or a rather
than f. And nally, to streamline the notation, outputs are denoted sn rather
than s(n), or xn , or an . Thus a typi al sequen e is written
More brie y, we write

(s1 , s2 , s3 , ).
(sn )n1

or
(sn )
n=1

even though (yet again) is not a number|here the notation is meant to


onvey that the terms of the sequen e go on and on. On e the ontext is lear,
even the notation
(sn )

will do, so long as we understand what is happening: n is varying through Z1 ,


and the sequen e is a orresponding list of values sn .

3.2 The Limit of a Real Sequen e

71

3.1.4 Previous Examples

In se tion 1.2, Ar himedes's quadrature of the parabola led to the trianglearea sums
S1 = Atri ,
S2 = Atri [1 + 1/4] ,


S3 = Atri 1 + 1/4 + (1/4)2 ,


S4 = Atri 1 + 1/4 + (1/4)2 + (1/4)3 ,

and in general for n Z1 ,



Sn = Atri 1 + 1/4 + (1/4)2 + + (1/4)n1 .

And Ar himedes's evaluation of the nite geometri sum with ratio r = 1/4
gave a losed form (ellipsis-free) expression for the sequen e entries, so that
the sequen e of triangle-area sums was in fa t


(Sn ) = Atri (4/3) 1 (1/4)n

n1

(3.2)

In se tion 2.5, integrating the rational power fun tion f (where 6= 1)


from 1 to b gave rise to the sequen e of box-area sums (see page 52)
(Sn ) =


. s+1 1 
(b+1 1) n
sn 1 n1

where sn = b1/n .

(3.3)

In the previous two hapters, we made assertions about the limiting behaviors
of sequen es (3.2) and (3.3). Later in this hapter we will be able to substantiate the assertions, as well as other matters from the end of hapter 2.

3.2 The Limit of a Real Sequence


3.2.1 Absolute Value and Distance

To des ribe quantitatively the idea of two real numbers being near ea h other,
regardless of whi h is the larger, we rst des ribe the idea of one real number
being near 0, regardless of whether it is positive or negative. The de nition
innately must be asewise:
Definition 3.2.1 (Absolute Value).
| | : R R0 ,

The absolute value fun tion is

|x| =

x
x

if
if

x 0,
x < 0.

72

3 Sequen e Limits and the Integral

So, for example, |5| = 5 and | 1/10| = 1/10. A number is near 0 if its
absolute value is small. It is worth pausing to onvin e oneself that indeed the
asewise formula for the absolute value fun tion always yields a nonnegative
real value, so that designating the odomain to be R0 makes sense.
Cases are a nuisan e to drag around, and so our short-term program is to
use the asewise de nition of the absolute value to establish a olle tion of
absolute value properties that no longer make dire t referen e to ases. On e
that is done, absolute values an be manipulated by using the properties
with no further referen e to the underlying ases, and indeed, with no further
thought of them.
Proposition 3.2.2 (Basic Absolute Value Properties). Let x and y be
real numbers. Then
(1) |x| = 0 if and only if x = 0.
(2) |x| x |x|.
(3) |xy| = |x| |y|. In
parti ular, | x| = |x| sin e x = x (1).
1
x |x|
1
(4) If y 6= 0 then = , and so by (3), also = .
y

|y|

|y|

Proof. (Sket h.) For instan e, to verify (2), note that

if x 0 then x = |x|, and so |x| = x 0 x = |x|,


and

if x < 0 then x = |x|, and so |x| = x < 0 < x = |x|.

Verifying the rst statement in (3) requires he king four ases, sin e x and y
an ea h be nonnegative or negative independently of the other. Four ases
amount to one small nuisan e, but as explained a moment ago, the point is
that after they are he ked on e and only on e, we never have to think about
them again. The reader is en ouraged to verify enough of Proposition 3.2.2
to onvin e himself or herself that the entire proposition an be veri ed in a

similar fashion.
(The symbol \
" at the end of the previous line denotes the end of a proof.)
Theorem 3.2.3 (Triangle Inequality).

For all real numbers

|x + y| |x| + |y|,

|x y| |x| + |y|,


|x| |y| |x + y|,


|x| |y| |x y|.

and y,
(3.4)
(3.5)
(3.6)
(3.7)

3.2 The Limit of a Real Sequen e

73

The rst inequality (3.4) of Theorem 3.2.3 is the Basi Triangle Inequality. The four inequalities an be gathered together as the statement that for
all real numbers x and y,



|x| |y| |x y| |x| + |y|.

(3.8)

The reader should beware that (3.8) does not say that |x y| |x| |y| in
general, and the reader should further beware that even after one hears this
and understands it in the abstra t, a frequent al ulation error is to write
some spe i version of the false inequality nonetheless.

Proof. For all x and y in R we have by Proposition 3.2.2 (2),


|x| x |x|

and

|y| y |y|,

and so adding the inequalities gives |x| |y| x + y |x| + |y|, or


(|x| + |y|) x + y |x| + |y|.

If x + y 0 then |x + y| = x + y, and so the right inequality in the previous


display be omes |x + y| |x| + |y|. If x + y < 0 then |x + y| = (x + y), i.e.,
x + y = |x + y|, and hen e the left inequality gives (|x| + |y|) |x + y|. In
either ase we have the Basi Triangle Inequality (3.4),
|x + y| |x| + |y|.

The other inequalities (3.5) through (3.7) are onsequen es of (3.4) and are

left as an exer ise.


We introdu e the symbol \ " as shorthand for if and only if. That
is, the symbol \ " between two statements means that the statement to
its left is true exa tly when the statement to its right is true.
In your writing, do not use the symbol to mean
anything other than if and only if. This, and nothing else, is
its meaning.

Again let x R be a real number, and let p R>0 be a positive real


number. Then
and

|x| < p p < x < p

|x| p p x p.

(3.9)

To establish (3.9), argue that if x 0 then sin e |x| = x and sin e the
statement \p x" is true (be ause p < 0 x),

74

3 Sequen e Limits and the Integral


|x| p x p p x p.

If x < 0 then sin e |x| = x and sin e \x p" is true (be ause x < 0 < p),
and sin e multiplying ea h side of an inequality by 1 swit hes its dire tion,
|x| p x p p x p x p.

Thus (3.9) holds regardless of whether x 0 or x < 0.

Proposition 3.2.4 (Relation Between Absolute Values and Intervals). Let a R and let p R>0 . Then for all x R,

and

|x a| < p a p < x < a + p,

|x a| p a p x a + p.

The two statements rephrase as assertions that ertain sets are intervals,
{x R : |x a| < p} = (a p, a + p)

and
{x R : |x a| p} = [a p, a + p].

Proof. For the se ond statement of the proposition, use (3.9) and re all that
adding the same quantity to both sides of an inequality preserves the inequality,
|x a| p p x a p a p x a + p.

The rst statement of the proposition has virtually the same proof. And the
third and fourth statements of the proposition are rephrasings of the rst
two.

The geometri distan e between two real numbers x and y on the number
line is the absolute value of their di eren e, |x y|. So, for example, Proposition 3.2.4 says that the set of numbers whose distan e from a is smaller
than p is the interval entered at a extending distan e p in both dire tions,
(a p, a + p). This is exa tly as our visual intuition tells us that it should
be, and it is easiest to remember by seeing the relevant pi ture in one's mind.
But the fa t that it follows readily from our de nitions by analyti arguments
sends a reassuring message about our methodology.
The following result sometimes provides the pun hline of an argument. Its
point is that to show that two quantities are equal we need only show that
they lie arbitrarily lose to ea h other.

3.2 The Limit of a Real Sequen e


Proposition 3.2.5 (Strong Approximation Lemma).

real numbers. Suppose that


| | <

Let and

75

be

for every positive number .

Then = .
Proof. Either | | is positive or it is zero. But the given ondition implies
that

| | 6= for every positive number .

So | | = 0. Consequently = 0, i.e., = .

Exercises
3.2.1. Prove inequalities (3.5) through (3.7) of Theorem 3.2.3. Prove them

by showing that they are onsequen es of (3.4), not be repeating the e ort of
proving (3.4) three more times.

3.2.2. Let x and y be nonzero. In ea h of (3.4) through (3.7), under what


onditions on the signs of x and y does equality hold?
3.2.3. Des ribe ea h of the four sets below in terms of intervals. A set may

require more than one interval for its des ription. (You may do this problem
by inspe tion.)
(a) A1 = {x R : |x 1/2| < 3/2},
(b) A2 = {x R : |x + 1/2| 3/2},
( ) A3 = {x R : |3/2 x| < 1/2},
(d) A4 = {x R : |3/2 + x| 3/2}.

3.2.4. Sket h the graphs of the following fun tions from R to R de ned by
the following equations (no explanations are needed for this problem):
(a) f1 (x) = |x|,
(b) f2 (x) = |x 2|,
( ) f3 (x) = |x| |x 2|,
(d) f4 (x) = |x| + |x 2|,
(e) f5 (x) = x2 1,
(f) f6 (x) = |x2 1|,
(g) f7 (x) = |x2 1|2 .
3.2.5. Let f1 through f7 be the fun tions des ribed in the previous exer ise.

By looking at their graphs, express ea h of the following six sets in terms of


intervals.
(a) S1 = {x R : f1 (x) < 1},

76

3 Sequen e Limits and the Integral

(b) S2 = {x R : f2 (x) < 1},


( ) S3 = {x R : f3 (x) < 1},
(d) S4 = {x R : f4 (x) < 3},
(e) S5 = {x R : f5 (x) < 3},
(f) S6 = {x R : f6 (x) < 3}.
Also, let S7 = {x R : f7 (x) < 1/2}. Represent S7 graphi ally on a number
line.
3.2.2 The Archimedean Property of the Real Number System

Any positive real number, however large, is ex eeded by some positive integer:
Proposition 3.2.6 (Archimedean Property of the Real Number System). Let x R>0 be any positive real number. There exists a positive
integer N Z1 su h that N > x.

The reader may feel that the Ar himedean Property is self-evident and
hardly deserves its own name. But in fa t there are number systems other
than the real number system (whi h, again, is not innately extant, mu h less
unique or preferred among number systems, just be ause it is named real) in
whi h the property does not hold. Indeed, early attempts at reasoning about
al ulus made referen e to in nitesimals, quantities that we now think of as
positive numbers so small that their re ipro als ex eed all positive integers,
this happening in a hyper-real number system that subsumes the reals. These
ideas of non-standard analysis were made rigorous by Abraham Robinson
only as re ently as 1960. A freshman al ulus text based on Robinson's in nitesimals, written by H. Jerome Keisler, is online at
http://www.math.wisc.edu/~keisler/calc.html

Here is an attempt to prove the Ar himedean Property rather than assume


it: Suppose that some positive real number x ex eeds all the positive integers,
x>N

for all N Z1 .

Then surely there is a least x at least as big as all the positive integers.
Consider the positive real number x 1. Sin e it is less than x, it is less than
some positive integer, i.e.,
x1<N

for some n Z1 .

x<N+1

for some N Z1 .

Consequently,

3.2 The Limit of a Real Sequen e

77

But N + 1 is again a positive integer, so that x is not at least as big as


all positive integers after all. Thus the supposition that some positive real
number ex eeds all the positive integers must be false.
However, rather than prove the Ar himedean Property, this argument
shows only that it follows from any assumption about the real number system that makes valid the Then surely there is a least x. . . statement in the
previous paragraph.
3.2.3 Definition of Sequence Limit

Re all that a sequen e is a fun tion whose domain is Z1 ,


s : Z1 S,

often viewed as a list of data,


(sn ) = (s1 , s2 , s3 , ).

In parti ular, a real sequence is a sequen e whose odomain is R, i.e., a list


of numeri al values. From now on, all sequen es in this hapter will be real
sequen es, and so usually they will simply be alled sequen es sin e bothering
to say real sequen e ea h time would be silly.
A real sequen e onverges to a limit if the terms of the sequen e approa h and stay near , as losely as desired, although they may or may
not a tually rea h , and they may or may not stay at should they rea h it.
Figure 3.1 depi ts onvergen e for a sequen e viewed as a fun tion,
s : Z1 R,

and gure 3.2 depi ts the onvergen e of the same sequen e viewed as a list
of data,
(sn ) = (s1 , s2 , s3 , ).

The following de nition aptures quantitatively and on isely the abovementioned notion of approa h and stay near , as losely as desired,

although they may or may not a tually rea h , and they may or may
not stay at should they rea h it.

Definition 3.2.7 (Limit of a Real Sequence, Convergent Sequence,


Divergent Sequence). Let (sn ) be a real sequen e, and let be a real
number. We say that (sn ) is a convergent sequence with limit if the

following ondition holds.

For every positive real number > 0,


there exists a positive integer N su h that
for all integers n N, |sn | < .

PSfrag repla ements


78

3 Sequen e Limits and the Integral


s(n)

PSfrag repla ements

Figure 3.1.

10

A onvergent sequen e, viewed as a fun tion

s3 s5

s1

Figure 3.2.

s4

s2

s(n)

A onvergent sequen e, viewed as a list of data

When the ondition holds, the fa t that (sn ) has limit is notated
lim(sn ) = .
n

In this ase we also say that (sn ) converges


does not onverge then it is divergent.

to .

If the sequen e

(sn )

The typography \limn " is short for \limn ", but as usual we try to
leave out of our notation as mu h as possible.
The grammar of De nition 3.2.7 is sophisti ated. Again, the idea of the
de nition is that a sequen e (sn ) has limit if the terms of the sequen e
eventually get lose to and stay lose to . The numbers and N, and
the intera tion between them, are the mathemati al ma hinery that together
quantify the idea. To work su essfully with De nition 3.2.7, one needs some
sense of how the quanti ation indeed aptures the idea, and one also needs
some pra ti e with the symbol-based language of the quanti ation. For this
reason, the rst few results that we will prove with the de nition are meant
to be simple and obvious-sounding: their point isn't to be earth-shattering,
but to demonstrate what the de nition says and how it works. The intent is
that as the examples a rue, the student will see that the de nition en odes a

3.2 The Limit of a Real Sequen e

79

natural idea in a way that is sensible and usable. Nonetheless, every al ulus
tea her understands that for the student, oming to grips with one's rst
arguments with the de nition of limit poses the double hallenge of parsing
the de nition's grammar in general and isolating the key parti ular of ea h
situation at hand. One gets better at this with time and experien e.
A few words about mathemati al proof may be useful here. Proofs in
mathemati s are not alienating formalisms, or at least they shouldn't be.
The reader may have heard the maxim that the ex eption proves the rule.
Sin e a mathemati al proof is meant to establish a rule in all ases, with
no ex eptions, the maxim doesn't sound sensible in our ontext. But it is.
The word prove is a variant of probe, and to say that the ex eption probes
the rule is to say that knowing when a prin iple an break down informs us
about the prin iple's s ope|its extent and its limitations. In mathemati s,
we often prove a statement to the e e t that if ertain onditions A hold then
some other ondition B follows. Proving su h a statement tea hes us be ause
the argument will give us insight into how onditions A lead to ondition B.
Condition B may well fail without onditions A in pla e|an ex eption that
probes the rule that ondition B holds.
De nition 3.2.7 is illustrated in gures 3.3 and 3.4. In both gures the
idea is that no matter how narrow the gray zone is, all but nitely many
of the dots lie in it. A more narrow gray zone may ex lude more dots, but
always only nitely many. Both gures are oversimpli ed in that they show a
sequen e with ea h su essive term getting loser to the limit. A onvergent
sequen e an behave more oyly, repeatedly approa hing its limit and then
ba king away, until eventually it approa hes the limit and stays lose.

s(n)

PSfrag repla ements


+

N N+1 N+2
Figure 3.3.

The de nition of limit, viewing a sequen e as a fun tion

80

3 Sequen e Limits and the Integral

PSfrag repla ements

Figure 3.4.

s(n)

The de nition of limit, viewing a sequen e as a list of data

In omplement to geometry, another way to understand De nition 3.2.7


is to interpret it as legislating a sort of adversarial pro ess.
To argue that a sequen e has limit , we rst allow someone who doubts
this to demand how lose the terms of the sequen e must get and stay to .
That is, the skepti provides the error toleran e > 0, whi h an be very
small but must be positive. On e is spe i ed, the skepti has to be quiet as
we onsult with the sequen e. If we an ome up with a starting index N
Z1 su h that the terms of the sequen e from that index onward,
(sN , sN+1 , sN+2 , ),

are all within of , then we have su essfully responded to the skepti . To


say that that the sequen e has limit is to say that we an so respond to
the skepti 's spe i ed positive error toleran e with a orresponding starting
index, no matter how small the error toleran e is.
On the other hand, to argue that a sequen e does not have limit , it is
we who play the role of the skepti . After sizing up the sequen e, we leverly
pres ribe an error toleran e > 0 for whi h there is no starting index N, i.e.,
the terms of the sequen e will never get and stay within distan e of . In
this ase, we bear the onus of demonstrating that no starting index exists in
response to our pres ribed error toleran e.
All of this said, geometri and dramati understanding of De nition 3.2.7
are ultimately developmental stages en route to a symboli understanding of
it.
Exercise
3.2.6. For ea h of the sequen es below, al ulate the rst few terms, and make

a guess as to whether or not the sequen e onverges. In some ases you will
need to use a al ulator. Try to explain the basis for your guess.
(a) (sn ) = (1, 1 + 1/2, 1 + 1/2 + 1/3, 1 + 1/2 + 1/3 + 1/4, ).
(b) (sn ) = (1, 1 1/2, 1 1/2 + 1/3, 1 1/2 + 1/3 1/4, ).
( ) (sn ) = (1, 1 + 1/22 , 1 + 1/22 + 1/32 , 1 + 1/22 + 1/32 + 1/42 , ).
(d) (sn ) = (1, 1 + 1/3, 1 + 1/3 + 1/32 , 1 + 1/3 + 1/32 + 1/33 , ).

3.2 The Limit of a Real Sequen e

81

(e) (sn ) = ((1 + 1/1)1 , (1 + 1/2)2 , (1 + 1/3)3 , (1 + 1/4)4 , ).


(f) (sn ) = (1(2 1), 2(21/2 1), 3(21/3 1), 4(21/4 1), ).
3.2.4 Basic Sequence Limit Rules

Here are some examples of how to use De nition 3.2.7 to prove beginning
results. None of the statements in the following proposition should be the
least bit surprising.
Proposition 3.2.8 (Basic Sequence Limits).

(1) (Constant Sequen e Rule.) Let c be any real number. Consider the

sequen e s whose terms are sn = c for all n Z1 ,


(sn ) = (c, c, c, ) = (c).

This sequen e's limit is c,


lim(c) = c for c R.
n

(2) (1/n Rule.) Consider the sequen e s whose terms are sn = 1/n,
(sn ) = (1, 1/2, 1/3, ) = (1/n).

This sequen e's limit is 0,


lim(1/n) = 0.
n

(3) (1/n Rule.) Let be a positive rational number. Consider the se-

quen e s whose terms are sn = 1/n ,

(sn ) = (1, 1/2 , 1/3 , ) = (1/n ).

This sequen e's limit is 0,


lim(1/n ) = 0 for Q>0 .
n

(4) (nth Root Rule.) Let b be a positive real number. Consider the se-

quen e s whose terms are sn = b1/n ,

(sn ) = (b, b1/2 , b1/3 , ) = (b1/n ).

This sequen e's limit is 1,


lim(b1/n ) = 1 for b R>0 .
n

82

3 Sequen e Limits and the Integral

(5) (nth Power Rule.) Let r be a real number su h that |r| < 1 (1 < r < 1).

Consider the sequen e s whose terms are sn = rn ,


(sn ) = (1, r, r2 , ) = (rn ).

This sequen e's limit is 0,


lim(rn ) = 0 for |r| < 1.
n

The sequen e in the nth Power Rule has domain Z0 rather than Z1 ,
but this is not a serious issue. More generally, the terms of a sequen e an
start at any index n rather than at n = 1, su h as
(sn )n17 = (s17 , s18 , s19 , )

or
(sn )
n=5 = (s5 , s4 , s3 , ).

As before, the idea is that a sequen e is one datum per generation where there
is a starting generation (e.g., 17 or 5 as in the two examples just given) and
then an endless su ession of generations thereafter. We ould insist that the
initial generation always be indexed 1, but this would lead to notational ontortions in situations su h as the nth Power Rule where the initial generation
learly warrants the index 0 instead. Stri tly speaking, the de nition of a
sequen e and the de nition of a sequen e limit should be phrased to take
into a ount the freer indexing s heme, but this is notationally onerous to no
substantive purpose, espe ially in an environment whose grammar is already
so symbol-heavy. In the ase of sequen e limits, we are only about the longterm behavior of the sequen e anyhow, and so fussing about a nite shift in
its indexing, or about aberrant behavior on the part of a small number of
early terms, is patently irrelevant. We will soon quantify this Irrelevan e of
Finite Index-Shifts.

Proof. (1) To argue that limn (sn ) = c when sn = c for all n Z1 , let
any positive error toleran e > 0 whatsoever be given. Then the appropriate
starting index in response to is simply N = 1. Indeed, be ause sn = c for
all n, we have |sn c| = 0 for all n, and so ertainly
for all integers n 1,

|sn c| < .

Thus De nition 3.2.7 is satis ed.


(2) To argue that limn (1/n) = 0, again let an error toleran e > 0 be
given. Note that |1/n 0| = 1/n for n Z1 . So we need to nd a starting
index N Z1 su h that

3.2 The Limit of a Real Sequen e

for all integers n N,

83

1/n < .

By algebra, the previous display is equivalent to


for all integers n N,

n > 1/,

and to show this, it su es to show instead that some suitable starting index N satis es
N > 1/,

be ause then also n > 1/ for all n N. Sin e is positive and presumably
small, 1/ is positive and presumably big. However, no matter how big 1/
is, the Ar himedean Property of the real number system says that there exists some positive integer N > 1/. This ompletes the argument that the
sequen e (1/n) has limit 0.
(3) Let be a positive rational number. To argue that limn (1/n ) = 0,
again let > 0 be given. We want to nd a orresponding N Z1 su h that
for all integers n N,

1/n < .

By a little algebra, the previous display is equivalent to


for all integers n N,

n > 1/1/ ,

and to show this, it su es to show instead that some suitable starting index N satis es
N > 1/1/ ,

be ause then also n > 1/1/ for all n N. If is a small positive real number
and is a small positive rational number then 1/1/ is very big. Nonetheless,
iting the Ar himedean Property of the real number system ompletes the
argument, as in the proof of (2).
(4) Let b be a positive real number. We need to argue that limn (b1/n ) = 1.
First, if b = 1 then the sequen e (b1/n ) is the onstant sequen e (1), and
the result follows from the Constant Sequen e Rule.
Se ond, if b > 1 then also b1/n > 1 for ea h positive integer n, and so
|b1/n 1| = b1/n 1 for ea h positive integer n. As shown in exer ise 2.3.2,
b1/n 1 =

b1
b1
.
<
n
1 + b1/n + + b(n1)/n

By the Ar himedean Property, there is a positive integer N su h that


N>

and hen e that

b1
,

84

3 Sequen e Limits and the Integral

It follows that

b1
.

for all integers n N,

n>

for all integers n N,

b1
< ,
n

and hen e, sin e |b1/n 1| = b1/n 1 < (b 1)/n, that


for all integers n N,

|b1/n 1| < .

This ompletes the argument when b > 1.


Third, if 0 < b < 1 then let = 1/b > 1. Then b = 1/ and
(b1/n ) = (b, b1/2 , b1/3 , ) =

1
1
1
, 1/2 , 1/3 ,

Now, sin e 0 < b < 1 and > 1 (so that 1/n > 1 for all n),
|b1/n 1| = 1 b1/n = 1

1
1/n

1/n 1
< 1/n 1.
1/n

From a moment ago we know that limn (1/n ) = 1 sin e > 1. Let > 0 be
given. Then there exists some starting index N su h that
for all integers n N,

1/n 1 < .

It follows from the previous two displays that


for all integers n N,

|b1/n 1| < .

This ompletes the argument.


(5) Let r be a real number su h that |r| < 1. We need to argue that
limn (rn ) = 0.
If r = 0 then (rn ) = (1, 0, 0, 0, ), and so the result follows from the Constant Sequen e Rule. As dis ussed above, the aberrant rst term is irrelevant
to the limiting behavior.
Now let r 6= 0. Note that |rN 0| = |rN | = |r|N ( f. Proposition 3.2.2 (3)).
Sin e r 6= 0 and |r| < 1, it follows that 1/|r| exists and ex eeds 1,
1
=1+x
|r|

where x > 0.

Therefore, for any n Z1 ,


1
= (1 + x)n = (1 + x)(1 + x) (1 + x).
|r|n

3.2 The Limit of a Real Sequen e

85

But multiplying out the n-fold produ t (1 + x)(1 + x) (1 + x) gives a 1 (the


produ t of n 1's, one from ea h multipli and), and an nx (the sum of the n
produ ts of n 1 1's and one x), and also other terms if n > 1. So in fa t
1
= (1 + x)n 1 + nx > nx,
|r|n

and thus
|r|n <

1
.
nx

With this result in hand, let any error toleran e > 0 be given. By the
Ar himedean Property, there is a positive integer N su h that
N>

It follows that
and hen e that

1
.
x

1
< ,
Nx

for all integers n N,

1
< ,
nx

and hen e, sin e |r|n < 1/(nx), that


for all integers n N,

|r|n < ,

This ompletes the argument.

Although the proof of Proposition 3.2.8 has been written out in onsiderable length, on e the reader digests its ideas, most of them should not seem
di ult. The Constant Sequen e Rule is an instant onsequen e of the de nition of limit, and so are the 1/n Rule and the 1/n Rule on e one is aware
of the Ar himedean property of the real number system. The proof of the
nth Root Rule relies on a al ulation that was arried out in exer ise 2.3.2,
and the proof of the nth Power Rule also involves a bit of algebrai insight,
but easier proofs of the nth Root Rule and the nth Power Rule will be ome
available (in exer ise 5.1.4 to follow) on e we have the logarithm.
Example 3.2.9. For an example of a divergent sequen e, onsider
(n) = (1, 2, 3, ).

To see that (n) is divergent, suppose instead that it has a limit , and pres ribe
the error toleran e = 1/3. For any andidate positive integer N to serve as
a suitable starting index, we would need to have

86

3 Sequen e Limits and the Integral


|n | < 1/3 for all n N,

so that in parti ular, letting n = N and then letting n = N + 1,


and |N + 1 | < 1/3,

|N | < 1/3

(3.10)

It is intuitively impossible that N and N + 1, whi h are distan e 1 away from


ea h other, ould both be within distan e 1/3 of . One way to quantify the
impossibility is by using the Triangle Inequality,
1 = |(N + 1 ) (N )| |N + 1 | + |N | < 1/3 + 1/3 = 2/3,

giving 1 < 2/3, whi h is nonsense. Another way to quantify the impossibility
is by rewriting (3.10),
| (N + 1)| < 1/3 and | N| < 1/3,

and then noting that onsequently by Proposition 3.2.4,


N + 2/3 = N + 1 1/3 < < N + 1/3,

giving 2/3 < 1/3, whi h is again nonsense. Sin e the assumption that
limn (n) = for some has led to a ontradi tion, the sequen e (n) has
no limit.
Example 3.2.10. For another example of a divergent sequen e, onsider
(sn )n0 = ((1)n )n0 = (1, 1, 1, 1, ).

Suppose that (sn ) has limit . Let = 1/2. Then all terms sn for large
enough n lie within 1/2 of . Sin e there are terms sn = 1 and terms sn = 1
for arbitrarily large n, we have
and | (1)| < 1/2,

| 1| < 1/2

and so by Proposition 3.2.4,


and (3/2, 1/2),

(1/2, 3/2)

whi h is nonsense. So the supposition that (sn ) has a limit is unsustainable.


The sequen e (sn ) diverges by os illation.
Example 3.2.11. The equalities
4=2+2
6=3+3
8=3+5
10 = 3 + 7
12 = 5 + 7
14 = 3 + 11

3.2 The Limit of a Real Sequen e

87

show that ea h even integer from 4 to 14 is the sum of two prime numbers. The
Goldba h Conje ture (GC) states that in fa t every even integer at least 4 is
the sum of two primes. The onje ture dates ba k to the 18th entury, and to
this day nobody has shown a proof or a ounterexample. De ne a sequen e
(gn )n2

as follows:
gn =

1 if ea h of 4, 6, , 2n is the sum of two primes,


0 if not.

Thus the sequen e (gn ) begins


(1, 1, 1, 1, 1, 1, ),

and either it says at 1 forever, or at some point it hanges to 0 and then stays
at 0 forever. Is (gn ) onvergent, and if so then what is its limit? The sequen e
is onstru ted so that by de nition its limit is
lim(gn ) =
n

1 if GC is true,
0 if GC is false,

but is this a satisfying answer? Playing the role of the skepti in the framework of De nition 3.2.7, we set = 1/2 and request a orresponding starting
index N. The best response that an advo ate of the onvergen e of (gn ) an
give is to de ne, onditionally on GC being false, N0 as the smallest integer
at least 2 su h that 2N0 is not the sum of two primes, and then to say that
the starting index is

if GC is true,
2
N=
N0 if GC is false.
Again, is this answer satisfa tory, or even meaningful? What if GC is neither
provable nor disprovable from the usual starting assumptions about arithmeti (whatever these may be)? The questions here are questions of logi and
philosophy, not the subje t-matter of these notes, but this example is meant
to show that even the beautifully- rafted grammar of De nition 3.2.7 does
not answer all questions about sequen e limits.
Exercises
3.2.7. (a) Re all the fa torial fun tion, denoted by an ex lamation mark,
1! = 1,

2! = 2 1,

3! = 3 2 1,

88

3 Sequen e Limits and the Integral

Consider the sequen e


(sn ) = (1/n!) = (1/1!, 1/2!, 1/3!, ).

Neither the 1/n Rule nor the 1/n Rule (Proposition 3.2.8 (2) and (3)) applies
to this sequen e, but does either of them suggest anything about it? Explain.
(b) Consider the sequen e
(sn ) = (n1/n ) = lim(1, 21/2 , 31/3 , ).
n

Explain why the nth Root Rule (Proposition 3.2.8 (4)) does not apply dire tly
to this sequen e. Does the rule suggest anything about the sequen e? Using
suitable omputing power, al ulate some terms of (sn ) and then make a
onje ture about its long-term behavior.
( ) Consider the sequen e
(sn ) = ((n!)1/n ) = lim(1, 21/2 , 61/3 , ).
n

Explain why the nth Root Rule (Proposition 3.2.8 (4)) does not apply dire tly
to this sequen e. Does the rule suggest anything about the sequen e? Using
suitable omputing power, al ulate some terms of (sn ) and then make a
onje ture about its long-term behavior.
(d) Consider the sequen e
(sn ) = (n10 /(1.1)n ) = (1/1.1, 210 /(1.1)2 , 310 /(1.1)3 , ).

Using suitable omputing power, al ulate some terms of (sn ) and then make
a onje ture about its long-term behavior.
3.2.8. (a) Explain why the sequen e (rn ) is onvergent for r = 1. What is its

limit?
(b) Explain why the sequen e (rn ) is divergent for r = 1.

3.2.9. The argument given in the text that (n) is divergent showed that the
error toleran e = 1/3 has no orresponding starting index N. Show that
= 1/2 also works as the error toleran e in the given argument. On the other
hand, = 2/3 does not work in the given argument, but it works in a modi ed

argument. Provide the modi ation.

3.2.10. Let (sn ) be a real sequen e, and let be a real number. Suppose
that as n gets ever larger, sn gets ever nearer to . That is, suppose that for
all n, m Z1 ,
if n > m then |sn | < |sm |.

It does not follow that (sn ) onverges to . Provide a ounterexample.

3.2 The Limit of a Real Sequen e

89

3.2.5 Irrelevance of Finite Index-Shifts

A brief dis ussion will quantify the earlier omment that a nite shift in a
sequen e's indexing is irrelevant to its limiting behavior.
Definition 3.2.12 (Index-Translate of a Sequence).

Let

(sn ) = (s1 , s2 , s3 , )

be a real sequen e, and let p Z1 . Then the sequen e


(sp+n ) = (sp+1 , sp+2 , sp+3 , )

is alled an index-translate of (sn ).


For example, the sequen e


1
1 1
,
,
,
9 16 25

1
n2

n3

1
(n + 2)2

n1

is an index-translate of the sequen e


  

1
1 1
.
1, , , =
4 9
n2 n1
Proposition 3.2.13 (Index-Translation Rule for Sequences). Let (sn )
and (tn ) be real sequen es, where (tn ) is an index-translate of (sn ). Then
(sn ) onverges and has limit if and only if (tn ) onverges and has the
same limit . That is, the two sequen es onverge or diverge together,

and if they both onverge then they have the same limit.
Proof. We have
(sn ) = (s1 , s2 , s3 , )

and, for some positive integer p,

(tn ) = (sp+1 , sp+2 , sp+3 , ).

Suppose that (tn ) onverges and has limit . We need to show that also (sn )
onverges and has limit . So, let > 0 be given. We need to nd a suitable
starting index N for (sn ) in response to . On the other hand, we know that
there is a suitable starting index M for (tn ) in response to . That is,
for all n M, |tn | < .
Sin e tn = sp+n for all n, the previous display rewrites as

90

3 Sequen e Limits and the Integral

for all n M, |sp+n | < ,


or

for all n M + p,

|sn | < ,

Thus the appropriate starting index for (sn ) in response to is N = M + p.


Now suppose that (sn ) onverges and has limit . We need to show that

also (tn ) onverges and has limit . Doing so is exer ise 3.2.11
For example,
lim
n

1
(n + 2)2

= lim
n

1
n2

=0

by the 1/n rule.

Exercises
3.2.11. Complete the proof of Theorem 3.2.13.
3.2.12. Let (sn ) and (tn ) be real sequen es. Suppose that limn (sn ) = , and
suppose that
|tn | |sn | for all n Z1 .

Show that onsequently limn (tn ) = .


3.2.6 Uniqueness of the Limit

De nition 3.2.7 (page 77) has a subtle worrisome feature: its wording allows
the possibility of a onvergent sequen e having more than one limit.
Common sense di tates that a sequen e an have at most one limit, but
sin e our notion of limit is en oded as the grammar of De nition 3.2.7, we
an not prove that a sequen e has at most one limit by appealing to ommon
sense. In fa t, what ommon sense really di tates is that if some sequen e
has more than one limit under De nition 3.2.7, then the de nition is foolishly
posed. On the other hand, a light, gra eful argument that a sequen e an have
at most one limit under De nition 3.2.7 would be eviden e that the de nition
has been well formulated to apture the right ideas in the right way. Here is
the argument.
Proposition 3.2.14 (Uniqueness of Limits). Let (sn ) be a real sequen e, and let and be real numbers. Suppose that

lim(sn ) = and
n

Then = .

lim(sn ) = .
n

3.2 The Limit of a Real Sequen e

91

The idea of the proof is that sin e the varying terms of the sequen e (sn )
get arbitrarily lose to and to , ne essarily the xed numbers and must
be arbitrarily lose to ea h other, making them equal.

Proof. First note that for any positive integer


Inequality gives

n whatsoever, the Triangle

| | = |(sn ) (sn )| |sn | + |sn |.

(3.11)

Next let > 0 be given. Then also /2 > 0. (This seemingly pointless observation is a small pie e of artfulness, guided by hindsight, that will pay o
below.) Sin e limn (sn ) = , in response to the error toleran e /2 there is a
starting index M Z1 su h that
|sn | < /2

for all n M.

(3.12)

Similarly, sin e limn (sn ) = there is a starting index M Z1 su h that


|sn | < /2

for all n M .

(3.13)

Let N be the larger of M and M . For any n N, (3.11), (3.12), and (3.13)
ombine to give
| | |sn | + |sn | < /2 + /2 = .

And now the ni e little point is that keeping only the quantities at the two
extreme ends of the previous display gives an inequality that makes no referen e to the indi es n or to any sequen e entries sn that went into establishing
it,
| | < .

Sin e this inequality holds for all > 0, we have = by the Strong Ap

proximation Lemma.
The arti e in the proof, of gaining more insight by simplifying|forgetting
auxiliary matters that were relevant only temporarily, rather than doggedly
insisting that every detail must ontinue to matter|is a small instan e of
mathemati al elegan e.
3.2.7 Generative Sequence Limit Rules

Thanks to Proposition 3.2.8, we have ve spe i sequen e limits in hand. But


in addition to omputing the limits of parti ular sequen es, we an also ompute the limits of ombinations of sequen es, assuming that we already know

92

3 Sequen e Limits and the Integral

the limits of the sequen es individually. That is, in addition to omputing


limits from s rat h, we an ompute limits generatively
The ombinations of real sequen es involved are as follows. Consider two
sequen es
s, t : Z1 R.

Let c R be any number. Then the sequen es


s t, cs, st : Z1 R

are de ned as follows:


for all n Z1 ,
for all n Z1 ,
for all n Z1 .

(s t)n = sn tn
(cs)n = c sn
(st)n = sn tn

These sequen es are the sum/di eren e of s and t, a onstant multiple of s,


and the produ t of s and t. Also, if tn 6= 0 for all n Z1 then the sequen es
1/t, s/t : Z1 R

are de ned to be
(1/t)n = 1/tn

for all n Z1

(s/t)n = sn /tn

for all n Z1 .

and

These sequen es are the re ipro al of t and the quotient of s and t. The
following result gives the limits of these newly-de ned sequen es in terms of
the limits of s and t.
Proposition 3.2.15 (Generative Sequence Limit Rules). Consider two
real sequen es s and t. Let c R be any number. Suppose that limn s =
and limn t = m. Then

(1) (Sum/Di eren e Rule.) limn (s t) exists and is m. That is,


lim(sn tn ) = lim(sn ) lim(tn ),
n

if both limits on the right exist.

(2) (Constant Multiple Rule.) limn cs exists and is c. That is,


lim(csn ) = c lim(sn ),
n

if the limit on the right exists.

(3) (Produ t Rule.) limn st exists and is m. That is,


lim(sn tn ) = lim(sn ) lim(tn ),
n

if both limits on the right exist.

3.2 The Limit of a Real Sequen e

93

(4) (Re ipro al Rule.) If tn 6= 0 for all n Z1 and m 6= 0 then limn 1/t

exists and is
lim
n

1
tn

1/m.

That is,

1
,
limn (tn )

if ea h tn is nonzero and the limit


on the right exists and is nonzero.

(5) (Quotient Rule.) If tn 6= 0 for all n Z1 and m 6= 0 then limn s/t

exists and is
lim
n

sn
tn

/m.

That is,

limn (sn )
,
limn (tn )

if ea h tn is nonzero and both limits


on the right exist and lim
(tn ) is nonzero.
n

The Di eren e Rule is a onsequen e of the Sum Rule and the Constant
Multiple Rule, and so perhaps it doesn't deserve its own name. The fussy
ondition in the Re ipro al Rule and the Quotient Rule that tn 6= 0 for
all n Z1 an handwaved away: if limn (tn ) 6= 0 then ne essarily all tn
are nonzero past some starting index, and nite index-shifts are irrelevant to
limits.

Proof. (Sket h.) (1) To prove the Sum Rule by arguing that limn (sn + tn) =
limn (sn ) + limn (tn ) provided both limits on the right exist, let = limn (sn )
and let m = limn (tn ). First note that for any positive integer n,
|(s + t)n ( + m)| = |sn + tn m|
= |sn + tn m|
|sn | + |tn m|.

Now let any error toleran e > 0 be given for s + t. Pres ribe the error
toleran e /2 for s to get a starting index Ns su h that
for all n Ns ,

|sn | < /2.

Similarly, there is a starting index Nt su h that


for all n Nt ,

|tn m| < /2.

Let N be the larger of Ns and Nt . Then, using the Triangle Inequality and
the previous three displays,
for all n N, |(s + t)n ( + m)| < .
Thus N is a suitable starting index in response to .
(2) The argument for the Constant Multiple Rule is very similar. The key
al ulation is

94

3 Sequen e Limits and the Integral


|(cs)n c| = |c| |sn |.

The details need to handle the ase c = 0 separately to avoid dividing by 0.


(3) For the Produ t Rule, the argument is a bit more elaborate. This time
the key al ulation is not purely me hani al: an auxiliary term is subtra ted
and added ba k before things arrange themselves ni ely,
|(st)n m| = |sn tn m| = |sn tn sn m + sn m m|
|sn tn sn m| + |sn m m|
= |sn | |tn m| + |sn | |m|.

Let > 0 be given. For large enough n, we have simultaneously that sn is so


lose to and tn is so lose to m that plausibly the right side is less than .
The details of the argument work as follows. We an ensure that for all large
enough n,
 |sn | < || + 1 and |tn m| < /(2(|| + 1)), so that |sn | |tn m| < /2;
 |sn | < /(2(|m| + 1)), so that |sn | |m| < /2.

It follows that for all large enough n,


|(st)n m| |sn | |tn m| + |sn | |m| < .

(4) For the Re ipro al Rule, the key al ulation is again me hani al,



1
1 m tn |tn m|

|(1/t)n 1/m| = =
=
.
tn m
mtn
|m| |tn |

For large enough n, simultaneously |tn m| < |m|2 /2 and |tn | > |m|/2, and
so
|m|2 /2
|tn m|
<
= .
|m| |tn |
|m| |m|/2

Con atenating the previous two displays gives the desired result.
(5) The Quotient Rule follows from the Produ t Rule and the Re ipro al
Rule.

For example, onsider the limit




n3 2n2
lim
n
3n3 + 4

The Quotient Rule does not apply immediately, be ause the limits of the
numerator and denominator do not exist. (As always, is not a number.)
However, fa tor the highest power of n out of the numerator and denominator
of ea h term of the sequen e,

3.2 The Limit of a Real Sequen e


n3 (1 2/n)
1 2/n
n3 2n2
= 3
=
3
3n + 4
n (3 + 4/n3 )
3 + 4/n3

95

for all n Z1 .

It is now lear what the limit is.


lim
n

n3 2n2
3n3 + 4

= lim
n

1 2/n
3 + 4/n2

120
1
= .
3+40
3

The se ond equality in the last display omes from applying the Produ t Rule,
the 1/n Rule, and the Sum Rule in the numerator, the Produ t Rule, the
1/n Rule, and the Sum Rule in the denominator, and the Quotient Rule. The
main feature of this example is that the original numerator and denominator
had the same highest power of n, and the limit was the ratio of the relevant
oe ients.
Earlier we ontended with the subtle issue of the uniqueness of limits. The
existen e of limits, or la k thereof, is another subtle point that raises possible
misappli ations of the various sequen e rules.
Example 3.2.16. The Produ t Rule says that a sequen e that is the produ t

of two onvergent sequen es is again onvergent. But the onverse is not true.
That is, the produ t of two sequen es, at least one of whi h diverges, still an
onverge, although it may well diverge.
For a very easy example, the produ t of any divergent sequen e whatsoever of nonzero real numbers with its re ipro al sequen e is the onstant
sequen e (1), as ni ely onvergent as an be. For more interesting examples,
onsider the sequen es
(sn ) = ((1)n (1/2)n )

and (tn ) = ((1)n (1/2)1/n ).

Ea h of these sequen es is a produ t,


(sn ) = ((1)n ) ((1/2)n )

and (tn ) = ((1)n ) ((1/2)1/n ).

And we know that the sequen e ((1)n ) diverges. But as just explained, it
does not follow that the sequen es (sn ) and (tn ) diverge in onsequen e.
Indeed, note that
(sn ) = ((1/2)n ) = (rn )

where r = 1/2,

and so limn (sn ) exists and is 0 by the nth Power Rule. But on the other
hand, we know that limn ((1/2)1/n ) = 1 by the nth Root Rule, and so the
terms of (tn ) tend ever more losely to alternating between 1 and 1. Thus
limn (t) does not exist.

96

3 Sequen e Limits and the Integral

Example 3.2.17. Let r be a real number su h that |r| < 1. Consider the
sequen e s whose terms are sn = rn ,
(sn ) = (1, r, r2 , ),

and onsider a onstant multiple of the sequen e, t = rs,


(tn ) = (r, r2 , r3 , ),

By the Constant Multiple Rule,


lim(tn ) = r lim(sn ).
n

But also, sin e (tn ) is an index-translate of (sn ), the Irrelevan e of Finite


Index-Shifts gives
lim(tn ) = lim(sn ).
n

Therefore

r lim(sn ) = lim(sn ),
n

and sin e r 6= 1, this gives the nth Power Rule,

lim(sn ) = lim(rn ) = 0.
n

So apparently the earlier deli ate proof of the nth Power Rule was unne essary.
However, there must be a aw in the reasoning here. The argument used
only the assumption that r 6= 1, not that |r| < 1. So it purports to show, for
example that the sequen e for r = 1,
(1, 1, 1, 1, )

has limit 0, whi h it does not. Furthermore, the argument purports to show
that the sequen e for r = 2,
(1, 2, 4, 8, )

also has limit 0, whi h it most ertainly does not. The aw in the reasoning is
the assumption that limn (sn ) exists at all. What the argument has orre tly
shown is that if limn (rn ) exists and r 6= 1 then limn (rn ) = 0.
Example 3.2.18. Consider the sequen e (sn ) de ned by the rules

s1 = 1,

s2 = 1,

1 + sn1

sn =

for n > 2.
sn2

(3.14)

3.2 The Limit of a Real Sequen e

Thus

1+1
= 2,
1

97

1+2
= 3,
1
and so on. Note that sn is positive for ea h n Z1 . Let
s3 =

s4 =

= lim(sn ).
n

By the Quotient Rule, the Sum Rule, and the Index-Translation Rule, also
=

1+
1 + limn (sn )
.
=
limn (sn )

Thus 2 = 1 + , so that by the Quadrati Formula

1 5
.
=
2

Sin e ea h sn is positive, must be the positive root. That is, the sequen e (sn )
has limit

1+ 5
.
2

This example has the same aw in its reasoning as the previous one, and its
on lusion is at-out wrong (exer ise 3.2.15).
Exercises
3.2.13. Find the following limits,
or explain why they don't exist.

(a) limn 7 + 6/n+ 8/ n .
4 + 1/n
.
(b) lim
n
5

 +21/n
3n + n + 1
( ) lim
.
2
n
1 + 3n + 4n

(2 + 1/n)2 + 4
.
(d) lim
3
n
 (2 + 1/n)2 + 8
(2 + 1/n) 4
.
(e) lim
3
n
(2
 3+ 1/n)  8
8n + 13n
(f) lim
.
3
n

 17 + 12n3
8(n + 4) + 13(n + 4)
.
(g) lim
3
n
 17 +12(n + 4)
n+1
.
(h) lim 2
n
n +1

98

3 Sequen e Limits and the Integral

3.2.14. (a) Consider the following argument:

The onstant sequen e

(0)

is

(0) = (0, 0, 0, 0, )

= (1 1, 1 + 1, 1 1, 1 + 1, )

= (1, 1, 1, 1, ) + (1, 1, 1, +1, ).

But both of the last two sequen es diverge by os illation, and so the onstant sequen e (0) diverges. The argument must be wrong sin e limn (0) = 0

by the Constant Sequen e Rule. What is the aw in the reasoning?


(b) Consider the following argument: The onstant sequen e (1) has

limit

lim(1) = lim(1, 1, 1, 1, )
n

= lim(1 0, 2 1, 3 2, 4 3, )
n

= lim(1, 2, 3, 4, ) lim(0, 1, 2, 3, )
n

= 0.

But also limn (1) = 1 by the Constant Sequen e Rule, and so 0 = 1. What
is the aw in the reasoning?

3.2.15. List the rst ten terms of the sequen e (3.14). What is the sequen e's

long-term
behavior? Explain the aw in the reasoning that sequen e's limit
is (1 + 5)/2.

3.2.16. Similarly to the sequen e (3.14), onsider the sequen e

s1 = 1,
2

sn+1 = sn + 2 for n 1.

2sn

Assuming that this sequen e has a limit , what is ? Compute some terms of
the sequen e and use them to onje ture its a tual behavior.
3.2.8 Geometric Series
Definition 3.2.19 (Geometric Series).

sequen e

Let

be a real number. The

(sn ) = (1, 1 + r, 1 + r + r2 , ) = (1 + r + + rn1 )n0

is the geometric series with ratio r.

3.2 The Limit of a Real Sequen e

99

It is ru ial here to understand that the terms of the geometri series are
ever-longer sums, spe i ally, ever-longer nite geometri sums.
Proposition 3.2.20 (Geometric Series Formula). Let r be a real number su h that |r| < 1. Then the geometri series with ratio r onverges,

and its limit is

lim(1 + r + r2 + + rn1 ) =
n

1
.
1r

Proof. By the nite geometri sum formula,


1 + r + r2 + + rn1 =

1 rn
1r

for r 6= 1.

Sin e in fa t |r| < 1, various sequen e rules give the result immediately.

It is tempting to write the Geometri Sum Formula as follows:


1 + r + r2 + + rn + =

1
1r

for |r| < 1.

Note the se ond \+ " on the left side of the equality, onnoting that the
sum does not stop after any nite number of terms. That is, the formula is
giving the value of an in nite sum, understood to be the limit of nite sums
having more and more terms.
When r = 1/4, the Geometri Series Formula en odes the end- al ulation
of Ar himedes's quadrature of the parabola from se tion 1.2
Exercises
3.2.17. (a) Let (sn ) = (1 + 9/10 + (9/10)2 + + (9/10)n1 ). Find limn (sn ).
(b) Let (sn ) = (1 9/10 + (9/10)2 + (1)n1 (9/10)n1 ). Find
limn (sn ).
3.2.18. (a) The \in nite de imal"
0.111 = lim(0.1, 0.11, 0.111, )
n

Is naturally viewed as a ertain rational number. What rational number?


Explain.
(b) Similarly, what rational number is 0.123123123 ?

100

3 Sequen e Limits and the Integral

3.2.9 Order Sequence Limit Rules


Proposition 3.2.21 (Inequality Rule for Sequences).

be onvergent sequen es. Suppose that

for all n Z1 .

sn t n

Then

Let (sn ) and (tn )

lim(sn ) lim(tn ).
n

Proof. Introdu e a sequen e

(un ) = (tn sn ). Sin e both (sn ) and (tn )


onverge, limn (un ) exists and equals limn (tn ) limn (sn ). And so it su es
to prove that sin e un 0 for all n Z1 , also limn (un ) 0.
For any > 0 there is a starting index N su h that

That is,

for all n N,

un < lim(un ) + .

for all n N,

un < lim(un ).

But ea h un 0, and so, giving away ground in order to simplify,


< lim(un ).
n

That is, limn (un ) is greater than every negative number, no matter how lose

the negative number is to 0. Therefore limn (un ) 0.


The most ommon use of the Inequality Rule is in situations where
for all n,

0 tn

and we on lude that

0 lim(tn ).
n

Indeed, the proof of the rule pro eeded by redu ing it to this ase.
Proposition 3.2.22 (Squeezing Rule for Sequences).
and (un ) be three real sequen es. Suppose that
sn un tn

Suppose further that (sn ) and


Then (un ) also onverges to .

(tn )

Let

(sn ), (tn ),

for all n Z1 .
both onverge to the same limit .

3.2 The Limit of a Real Sequen e

101

The ni e point here is existen e: the Squeezing Rule says that the middle
sequen e has a limit, and furthermore the limit is the shared limit of the
outer sequen es. If the middle sequen e were already known to have a limit,
then the limit would be by two appli ations of the Inequality Rule. The
Squeezing Rule improves on the Inequality Rule in that it does not require
us to know that the middle sequen e has a limit. But on the other hand, it
requires bounds on a sequen e from both sides.

Proof. Let > 0 be given. For some starting index Ns ,


for all n Ns , < sn .
And for some starting index Nt ,
for all n Nt ,

tn < + .

Let N be the larger of Ns and Nt . Then by the previous two displays and the
hypothesis that sn un tn for all n Z1 ,
for all n N,
That is,

< un < + .

for all n N,

|un | < .

By the Irrelevan e of Finite Reindexing, the Inequality Rule holds if instead:


For some N Z1 , sn tn for all n N.
And the Squeezing Rule holds if instead:
For some N Z1 ,

sn un tn

for all n N.

But the more simply stated versions are tidier to prove.


Exercises
3.2.19. Exer ise 3.2.7 asks about the limits of four sequen es whose limits do

not follow from the ve basi seqen e limit rules. One of those four sequen es
has a limit that now an be found qui kly by using the generative sequen e
limit rules. Whi h sequen e is it, and how do generative seque e limit rules
tell us its limit?

102

3 Sequen e Limits and the Integral

3.2.20. For ea h of the statements to follow: if the statement is true then

justify it by means of limit rules; if the statement is false then give a ounterexample.
(a) Let (sn ) be a onvergent real sequen e. If sn > 0 for all n Z1 then
limn (sn ) > 0.
(b) Let (sn ) and (tn ) be real sequen es. If limn (sn ) = 0 then limn (sn tn) =
0.
( ) Let (sn ) be a real sequen e. If limn (s2n ) = 1 then either limn (sn ) = 1
or limn (sn ) = 1.
(d) Let (sn ) and (tn ) be real sequen es. If limn (sn tn ) = 0 then either
limn (sn ) = 0 or limn (tn ) = 0.
3.2.21. (a) Suppose that we know the 1/n Rule (Proposition 3.2.8 (2)) but do
not know the 1/n Rule (Proposition 3.2.8 (3)). Suppose further that Q
and > 1. Use a result or results from this se tion to establish the 1/n Rule,
i.e., limn (1/n ) exists and is 0. Where does the argument require > 1?
(b) Again suppose that we know the 1/n Rule but do not know the 1/n
Rule. This time suppose further that Q and 0 < < 1. Explain why N >
1 for some N Z1 . Now reason as follows. By N appli ations of the Produ t
Rule, the Nth power of the limit is the limit of the Nth powers,

(lim(1/n ))N = lim (1/n )N ,
n

and the al ulation ontinues,

(lim(1/n ))N = lim (1/n )N


n

= lim(1/nN )
n

as just explained
by algebra
by part (a), sin e N > 1.

=0

Therefore, limn (1/n ) = 0 as well, and so the 1/n Rule holds for 0 < < 1
in onsequen e of the 1/n Rule as well. But there is a aw in the reasoning
here. What is it?
3.2.22. Consider the following variant of the Squeezing Rule:
and (un ) be three real sequen es. Suppose that

and that

sn un tn

Let (sn ), (tn ),

for all n Z1

lim(tn sn ) = 0.
n

Then (un ) also onverges to the ommon limit of (sn ) and (tn ), whi h
exists be ause limn (tn ) limn (sn ) = limn (tn sn ) = 0. Is this variant
orre t? Either prove it, or explain the aw in the reasoning and provide a
ounterexample.

3.3 Integrability

103

3.3 Integrability
3.3.1 The Previous Examples Revisited

Se tion 3.1 re alled two sequen es from hapters 1 and 2. The rst was the
sequen e of triangle-area sums arising from Ar himedes's quadrature of the
parabola,


(Sn ) = Atri (4/3) 1 (1/4)n

n1

Now we an analyze this sequen e quantitatively. By the Constant Sequen e


Rule and the nth Power Rule,
lim(1) = 1 and lim((1/4)n ) = 0,
n

and so by the Di eren e Rule,


lim(1 (1/4)n ) = 1,
n

and then by the Constant Multiple Rule,


4
3

lim (Sn ) = Atri .


n

Of ourse this is the value that we already obtained for the limiting value,
but now it is on a mu h rmer footing.
The se ond sequen e was a sequen e of box-area sums arising from the
integration of the rational power fun tion,
(Sn ) =


. s+1 1 
(b+1 1) n
sn 1 n1

where sn = b1/n .

Here b > 1, and Q but 6= 1. We know by the nth Root Rule and
be ause b > 1 that
 sn R>0 for all n Z1 ,
 limn (sn ) = 1,
 sn =
6 1 for ea h n Z1 .

These properties of (sn ) are all that we need to arry out an analysis that
formalizes the derivative al ulation in se tion 2.4. Sin e the analysis will be
ited again in the next hapter, we isolate it. The general symbol in the
following proposition is not the spe i of the ambient dis ussion that has
been broken o momentarily in order to establish the proposition.

104

3 Sequen e Limits and the Integral

Proposition 3.3.1. Let Q

be any rational number. Let (sn ) be any


sequen e of positive real numbers su h that limn (sn ) = 1 but sn 6= 1 for
ea h n Z1 . For ea h n Z1 , let
un =

Then

s
n1
.
sn 1

lim(un ) = .
n

Proof. As explained in se tion 2.4,

1 + sn + s2n + + sn

tn 1
un = tn t 1 where tn = 1/sn
n

tp 1 tn 1

where tn = s1/q
n
tn 1 tq
n1

if Z0 ,

if Z1 ,
if = p/q, p Z , q Z1 .

In the ase Z0 , sin e limn (1) = 1 by the Constant Sequen e Rule, and
sin e we know that limn (sn ) = 1, many appli ations of the Produ t Rule and
then the Sum Rule give


lim(un ) = lim 1 + sn + s2n + + sn1 = .


n

Here the sum, and therefore the limit, are understood to be 0 if = 0.


Next onsider the ase Z1 , so that Z1 . In this ase,
un = tn

t
n 1
where tn = 1/sn .
tn 1

Sin e ea h sn is nonzero and limn (sn ) = 1, the Re ipro al Rule gives


limn (tn ) = 1/1 = 1, and so by the argument from a moment ago for Z1 ,
by the Produ t Rule, and by the Constant Multiple Rule,
lim(un ) = 1 () = .
n

Thus limn (un ) = for all Z .


For general = p/q Q where p Z and q Z1 , the formula for un
is
tp
n 1 tn 1
q
un =
where tn = s1/q
n .
tn 1 tn 1

By the nite geometri sum formula, and by the fa t that ea h tn R>0


be ause ea h sn R>0 ,




sn 1
|sn 1|


|tn 1| =
|sn 1|.
=
q1
q1
1 + tn + t2n + + tn 1 + tn + t2n + + tn

3.3 Integrability

105

And so sin e limn (sn ) = 1, also limn (tn ) = 1 by exer ise 3.2.12 (page 90).
Now the formula
lim(un ) = p/q =
follows in this ase from the previous two ases, the Re ipro al Rule, and the

Produ t Rule.
With the proposition proved, let the symbol again take on its meaning
from the sequen e (Sn ) and apply the proposition with + 1 as the of the
proposition. That is, now 6= 1 again, and if we let
un =

+1
sn
1
,
sn 1

n Z1 ,

then the sequen e that we want to analyze is


(Sn ) =

b+1 1
un

The proposition, the Re ipro al Rule, and the Constant Multiple Rule give
the desired result,
lim(Sn ) =
n

b+1 1
.
+1

The following onsequen e of the Squeezing Rule was ta itly used twi e
in hapter 2.
Proposition 3.3.2 (The Trapped Quantity is the Common Limit).
Let (sn ) and (tn ) be real sequen es, and let u be a real number. Suppose

that

for all n Z1 ,
and suppose that (sn ) and (tn ) onverge to the same limit,
sn u t n

lim(sn ) = lim(tn ) = .
n

Then u = .
Proof. Consider the onstant sequen e (u), ea h of whose terms un is the
number u. By the Constant Sequen e Rule, limn (u) = u. Also, we are given
that
sn un tn for all n Z1 ,
so that by the Squeezing Rule, limn (u) = . Thus u = .

106

3 Sequen e Limits and the Integral

Proposition 3.3.2 is useful when u is some unknown area that we want to


nd and is the ommon limit of two sequen es of box-area sums, one too
small to be the area and one too big. Our rst ta it use of the proposition was
in se tion 2.5, where in addition to the sequen e (Sn ) from a moment ago, a
se ond sequen e appeared,
(Tn ) = (s
n Sn ),

sn = b1/n .

For this sequen e, the box-heights were determined by the values of the power
fun tion f over the right endpoints of their bases, rather than the left endpoints. And sin e (using the nth Root Rule for the last step in the display to
follow)
lim(sn ) = lim((b1/n ) ) = lim((b )1/n ) = 1,
n

it follows that (assuming 6= 1)


lim(Tn ) = lim(Sn ) =
n

Let

b+1 1
.
+1

Arb1 (f )

denote the area under the graph of f from 1 to b, a onstant. Sin e the power
fun tion is in reasing for > 0 and de reasing for < 0, we have

Sn Arb
1 (f ) Tn
Tn Ar

b
1 (f )

Sn

for all n Z1 , if > 0,


for all n Z1 , if < 0.

In either ase, Proposition 3.3.2 now gives rm footing to the familiar result
that onsequently the normalized power fun tion area is
Arb1 (f ) =

b+1 1
,
+1

b > 1, Q, 6= 1.

And be ause the area is the limit of box-area sums from above and below, it
is in fa t an integral,
Zb
1

f =

b+1 1
,
+1

b > 1, Q, 6= 1.

The ase = 1, where we don't know that limn (Sn ) and limn (Tn ) exist,
mu h less have a ommon value for whi h we have a formula, is more subtle.
We will return to it shortly.
Our se ond ta it use of Proposition 3.3.2 was in omputing the nonnormalized power fun tion integral, also in se tion 2.5. Given b and c with
1 b and c > 0, we de ned the sequen es

3.3 Integrability

107

en ) = c+1 (Sn )
(S

and

(Ten ) = c+1 (Tn ).

By the Constant Multiple Rule, and the fa t that limn (Sn ) = limn (Tn ) =
R
b
1 f ,
Zb
lim(Sen ) = lim(Ten ) = c+1 f .
n

But by the geometry that led to (Sen ) and (Ten ), also




en Arbc (f ) Ten
S
c
en
Ten Arbc (f ) S
c

for all n Z1 , if > 0,


for all n Z1 , if < 0.

And so it follows from Proposition 3.3.2 that


+1
Arbc
c (f ) = c

Zb

f .

Furthermore, as the ommon limit of box area sums from above and below,
the area a quires the status of integral, and the previous equation rewrites,
as in (2.9) (page 56),
Z bc

f = c+1

Zb

f .

This result holds for all Q su h that limn (Sn ) = limn (Tn ). For 6= 1,
the limits were shown to be equal by the simple expedient of evaluating them.
But for = 1 they are not yet established, and it turns out that they don't
evaluate to anything yet in our ken. So in the ase = 1, we need to argue
that the limits exist and are equal even though we an't nd a formula for
them. This is the last point that this se tion will dis uss.
Re all the verbal argument on page 56, whi h was made for all values of :

The fa t that Sn and Tn trap the area under the graph of f from 1
to b between them, and the fa ts that Tn = s Sn and s tends to 1,
ombine to show that Sn and Tn tend to the same limiting value,
that value being the area.
With sequen e limit results in hand, we now an quantify the reasoning. For
onvenien e, assume that > 0. Again, let
Arb1 (f )
denote the area under the graph of f from 1 to b, a onstant. Then we have
the following information:

108

3 Sequen e Limits and the Integral

(1) Sn Arb1 (f ) Tn for all n Z1 .


(2) Tn = (b )1/n Sn for all n Z1 .
From (1), then (2), and then (1) again,
0 Tn Sn = ((b )1/n 1)Sn ((b )1/n 1)Arb
1 (f ) for all n Z1 .

That is,

0 Tn Sn ((b )1/n 1)Arb


1 (f )

for all n Z1 .

By various sequen e limit rules, in luding the Squeezing Rule, it follows (exer ise 3.3.1) that
lim(Tn Sn ) exists and equals 0.
(3.15)
n

But also from (1),


0 Arb
1 (f ) Sn Tn Sn

and so again by the Squeezing Rule,

for all n Z1 ,

lim(Arb1 (f ) Sn ) exists and equals 0.


n

(3.16)

Sin e Arb1 (f ) is onstant, the Constant Sequen e Rule says that also
lim(Arb1 (f )) exists and equals Arb1 (f ).
n

(3.17)

Now, note that sin e (using square bra kets rather than parentheses to group
two numbers without overloading the sequen e notation)
b
Sn = Arb
1 (f ) [Ar1 (f ) Sn ] for all n Z1 ,

the de nition of the di eren e of two sequen es gives

b
(Sn ) = (Arb
1 (f )) (Ar1 (f ) Sn ).

A ording to the Di eren e Rule, now (3.16) and (3.17) give


lim(Sn ) exists and equals Arb1 (f ).
n

And sin e
(Tn ) = (Tn Sn ) + (Sn ),

(3.15) and the previous display ombine in turn to give


lim(Tn ) exists and equals Arb1 (f ).
n

The argument in itali s is now fully quanti ed, for all values of .
Note the nesse of the argument: deftly using the Squeezing Rule twi e
to show that auxiliary limits exist subtly but inexorably ornered our desired
limits until they were for ed to exist as well, and to be equal. And again,
the reasoning was arried out with no re ourse to expli it formulas, meaning
that it should apply in ontexts beyond the power fun tion in parti ular. The
remainder of this hapter will expand its s ope.

3.3 Integrability

109

Exercise
3.3.1. Show that (3.15) follows from the display immediately pre eding it.
3.3.2 Definition of Integrability
Definition 3.3.3 (Lower Sum, Upper Sum). Let a and b be real numbers with a b. Let M be a nonnegative real number. Consider a fun tion
f : [a, b] [0, M].

The region under the graph of f, a subset of the plane, is


R = {(x, y) R2 : a x b, 0 y f(x)}.

and it has an area,

Arba (f) = Ar(R).

Suppose that a number S is a sum of nitely many box-areas, where the


base of ea h box lies on the x-axis, the top of ea h box (at least as high
as the base) lies under the graph of f, the overlap of any two boxes is at
most a verti al line segment, and the bases ombine to over the x-axis
from a to b. Then S is a lower sum for Arba (f). Suppose that a number T
is a sum of nitely many box-areas. where the boxes satisfy the same
onditions ex ept that their tops lie over the graph of f. Then T is an
upper sum for Arb
a (f).
We saw lower sums and upper sums throughout the integration of the
rational power fun tion. The sums Sn were lower sums and the sums Tn
were upper sums only for > 0; unfortunately, the Sn were upper sums and
the Tn were lower sums for < 0, but this is only a notational irritant of no
onsequen e.
Retaining the terminology of the de nition, sin e any lower sum for
Arba (f) = Ar(R) is the area of a polygon that is a subset of R, and any upper
sum for Arba (f) is the area of a polygon that is a superset of R, the following
result is automati from the basi properties of area.
Proposition 3.3.4 (Basic Property of Lower and Upper Sums).
a b, and let M 0. Consider a fun tion

Let

f : [a, b] [0, M].

Let S be any lower sum for Arba (f), and let T be any upper sum for Arba (f).
Then
S Arb
a (f) T.

110

3 Sequen e Limits and the Integral

Now we an generalize the re ent argument that a minimal good property


of lower and upper sums, that the limit of their di eren es is zero, has further
good onsequen es, that the lower and the upper sums themselves have limits,
that the two limits are equal, and that they equal the area.
Proposition 3.3.5 (Bootstrap Result for Lower and Upper Sums).
Let a b, and let M 0. Consider a fun tion
f : [a, b] [0, M].

Suppose that a sequen e (Sn ) of lower sums for Arba (f) and a sequen e (Tn )
of upper sums for Arba (f) satisfy the ondition
lim(Tn Sn ) exists and equals 0.
n

Then limn (Sn ) and limn (Tn ) both exist, and


lim(Sn ) = lim(Tn ) = Arba (f).
n

The proof (exer ise 3.3.2) is similar to the argument re ently given in the
spe ial ase of the power fun tion. It requires the Squeezing Rule on e to
make a limit exist, and then basi and generative results to rea h the desired
on lusions.
Definition 3.3.6 (Integral).

fun tion

Let

a b,

and let

M 0.

Consider a

f : [a, b] [0, M].

If there exist a sequen e (Sn ) of lower sums for Arba (f), and a sequen e (Tn ) of upper sums for Arba (f), su h that
lim(Tn Sn ) exists and equals 0
n

then f is integrable from


area under its graph,

to b. The
Zb
a

Equivalently,

Zb
a

integral

of f from

f = Arb
a (f).

f = lim(Sn ) = lim(Tn ),
n

sin e by Proposition 3.3.5 both limits exist and equal Arba (f).

to

is the

3.3 Integrability

111

Note how neatly the de nitions and propositions of this se tion quantitatively apture our earlier des ription of the integral in natural language,
displayed in itali s on page 53.
Certainly, if there exist a sequen e (Sn ) of lower sums for Arba (f) and a
sequen e (Tn ) of upper sums for Arba (f) su h that
lim(Sn ) and lim(Tn ) both exist, and they are equal
n

then f is integrable from a to b. The point of Proposition 3.3.5 is that these


onditions follow from the seemingly-weaker onditions required in De nition 3.3.6. But it is perfe tly ne to establish these onditions instead.
To review some of the ideas, again let f : [a, b] [0, M] be a fun tion.

Does Arba (f) exist? Yes, always. The reader should be aware that many
al ulus ourses treat all existen e issues as obvious, perhaps not even rais-

ing them, whereas many beginning real analysis ourses derive existen e
results from a property of the real number system alled ompleteness. In
ontrast to both of these approa hes, our method is to invoke the existen e
of area fun tions but then derive further onsequen es of the invo ation
arefully.
 Are there sequen es (Sn ) and (Tn ) of lower and upper sums for Arb
a (f)
both with Arba (f) as their limit? Sometimes. The Bootstrapping Result
shows that there are su h sequen es if there are sequen es of lower and
upper sums su h that lim(Tn Sn ) = 0, and the previous paragraph
(starting Certainly. . . ) observed that the onverse holds as well. Under
these ir umstan es, we view f as integrable. Thus, integrability means
not that the area exists, but that the area is the limit of suitable box-area
sums.


When f is integrable, an we put the ommon limit of (Sn ) and (Tn )


into some onvenient form, su h as an expression in terms of fun tions that we already know? Not always. In the ase of the power fun tion f , we an do so for all 6= 1 but not for = 1.

When f is integrable but the area under its graph does not take a
onvenient form that we already understand, what good does the integrability do us? We an still study the integral as a limit in order to
learn more about its properties. To know a fun tion's properties is to understand it. For example, in hapter 5 we will study the logarithm as an
integral.

Exercise
3.3.2. Prove Proposition 3.3.5.

112

3 Sequen e Limits and the Integral

3.3.3 Monotonicity and Integrability

Let

Definition 3.3.7 (Monotonic Function).

Consider a fun tion

a b,

and let

M 0.

f : [a, b] [0, M].

The fun tion f is increasing if for all x1 , x2 [a, b] with x2 > x1 , also
f(x2 ) f(x1 ). The fun tion f is decreasing if for all x1 , x2 [a, b] with
x2 > x1 , also f(x2 ) f(x1 ). The fun tion f is monotonic if it is in reasing
or it is de reasing.
Thus a fun tion is in reasing if its graph, traversed from left to right, is
everywhere rising or level, never falling. The or level distinguishes between
an in reasing fun tion and a stri tly in reasing fun tion as dis ussed earlier.
And similarly, a fun tion is de reasing if its graph is everywhere falling or
level, never rising.
Theorem 3.3.8 (Monotonic Functions are Integrable).
and let M 0. Let
f : [a, b] [0, M]

Let

a b,

be monotoni . Then f is integrable.

Proof. Now we use a uniform partition rather than a geometri one. The
relevant partition-widths are
n =

ba
n

for ea h n Z1 ,

and the partition points are


x0 = a,

x1 = a + n ,

x2 = a + 2n ,

xn = a + nn = b.

Assume that f is in reasing. Then (exer ise 3.3.3 (a)) the quantity
Sn = n (f(x0 ) + f(x1 ) + f(xn1 ))

is a lower sum for Arba (f), and the quantity


Tn = n (f(x1 ) + f(x2 ) + f(xn ))

is an upper sum. In their di eren e, nearly all the terms an el,


Tn Sn = n (f(xn ) f(x0 )),

and sin e x0 = a and xn = b, their di eren e is in fa t


Tn Sn = n (f(b) f(a)).

3.3 Integrability

113

Consequently, by various sequen e limit rules (exer ise 3.3.3 (b)),


lim(Tn Sn ) exists and equals 0.
n

This shows that f is integrable from a to b, as desired.


The proof when f is de reasing is virtually identi al.

Sin e the power fun tion is monotoni , Theorem 3.3.8 en ompasses it,
even though the theorem's proof used uniform lower and upper sums rather
than the geometri ones that we used earlier to analyze the power fun tion.
The uniform lower and upper sums used to prove the theorem do not readily ompute the integral of the power fun tion, but they do re on rm its
existen e.
Exercise
3.3.3. In the proof of Theorem 3.3.8:
(a) Explain why Sn is a lower sum for Arba (f) and Tn is an upper sum.
(b) Explain why limn (Tn Sn ) exists and equals 0.
3.3.4 A Basic Property of the Integral
Proposition 3.3.9. Let a, b, and c be
M 0 be a positive real number. Let

real numbers with

a b c.

Let

f : [a, c] [0, M]

be a fun tion. Then


Zc
a

exists

Zb

exists and

and when the various integrals exist,


Zc
a

f=

Zb
a

f+

Zc

Zc
b

exists,

f.

Proof. The region under the graph of f from a to c is a bounded subset of

the plane, and so it has an area. Similarly for the regions from a to b and
from b to c. By the fa t that area has basi sensible properties,
Arca (f) = Arba (f) + Arcb (f).
Granting momentarily that the three integrals in the proposition exist, it
follows that

114

3 Sequen e Limits and the Integral


Zc

c
f = Arca (f) = Arb
a (f) + Arb (f) =

Zb

f+

Zc

f.

This givesR the last equality in the


Thus what needs to be proved
Rb proposition.
Rc
is that if ac f exists
f
f
then
so
do
and
,
and
onversely.
a
b
Rc
Suppose that a f exists. This means that there are sequen es of lower
sums Sn for Arca (f), and sequen es of upper sums Tn for Arca (f), su h that
lim(Tn Sn ) = 0.
n

For ea h n, if the boxes whose areas sum to Sn in lude a box whose base
straddles the intermediate point b, then subdivide that box into two by adding
a verti al line segment at b. This has no e e t on Sn sin e the areas of two
subboxes just reated total the area of the box that was subdivided. And
similarly for the upper sums Tn . That is, we may assume that ea h Sn and
ea h Tn is the sum of box-areas for boxes whose bases omprise the x-axis
from a to b and then more boxes whose bases omprise the x-axis from b
to c. The sums de ompose a ordingly,
Sn = Sn + Sn and Tn = Tn + Tn

Here ea h
is a lower sum for Ar
and similarly for Tn and Tn , so that

, ea h

b
a (f)
Sn

Sn

Tn

for n Z1 .

is a lower sum for Arcb (f),


and Sn Tn . Sin e
Sn

Tn Sn = (Tn Sn ) + (Tn Sn ),

if follows that
0 Tn Sn Tn Sn

and 0 Tn Sn Tn Sn ,

and so by the Squeezing Rule for sequen es,

lim(Tn Sn ) = 0. and lim(Tn Sn ) = 0.


n

Rb

Rc

Thus a f and b f exist.


R
R
Now suppose that ab f and bc f exist. This means that there are sequen es
of lower sums Sn for Arba (f), sequen es of lower sums Sn for Arcb (f), sequen es
of upper sums Tn for Arba (f), and sequen es of upper sums Tn for Arcb (f), su h
that
lim(Tn Sn ) = 0 and lim(Tn Sn ) = 0.
n

For ea h n, the sum Sn = Sn + Sn is a lower sum for Arca (f) and the sum
Tn = Tn + Tn is an upper sum for Arca (f). Sin e
Tn Sn = (Tn Sn ) + (Tn Sn ),

if follows that
Thus

Rc

f exists.

lim(Tn Sn ) = 0.

3.3 Integrability

115

3.3.5 Piecewise Monotonicity and Integrability


Definition 3.3.10 (Piecewise Monotonic Function). Let a and b
real numbers with a b, and let M 0 be a real number. A fun tion

be

f : [a, b] [0, M]

is alled piecewise monotonic if there is a partition of [a, b],


a = x0 < x1 < < xn = b,

su h that f is monotoni on ea h interval [xi1 , xi ] for i = 1, , n.


For example, the absolute value fun tion is pie ewise monotoni on [1, 1],
but it is not monotoni there.
Proposition 3.3.11 (Piecewise Monotonic Functions are Integrable).
Let a and b be real numbers with a b, and let M 0 be a real number.

Any pie ewise monotoni fun tion

f : [a, b] [0, M]

is integrable from a to b.
This follows from Theorem 3.3.8 and Proposition 3.3.9.
Example 3.3.12. We give two fun tions, neither of whi h is pie ewise monotoni , but one of whi h is integrable. For any nonnegative integers k and ,
de ne two sets of points, both subsets of the interval [0, 1],


0
1
2
2k 1 2k
Pk =
,
,
, ...,
, k ,
2k 2k 2k
2k
2


3 1 3
0 1 2
,
,
, ...,
, .
Q =
3 3 3
3
3

That is, the points of Pk are spa ed a ross [0, 1] in uniform steps of size 1/2k ,
and similarly for Q with step-size 1/3 . For a point to lie simultaneously in
some Pk and some Q requires
b
a
= ,
k
2
3

or

3 a = 2k b,

0 a 2k ,
0 a 2k ,

0 b 3 ,
0 b 3 .

Be ause positive integers fa tor uniquely into prime powers, the only solutions
are a = b = 0, i.e., the ommon point is the left endpoint 0, and a = 2k , b =

116

3 Sequen e Limits and the Integral

3 , i.e., the ommon point is the right endpoint 1. Ex luding the endpoints,
there is no overlap among the sets Pk and Q .

Now onsider two fun tions,

f : [0, 1] [0, 1],

where
f(x) =

and

1 if x Pk for some k,
0 if x
/ Pk for all k,

g : [0, 1] [0, 1],

where
g(x) =

1/2k
0

if k is the smallest integer su h that x Pk ,


if x / Pk for all k.

An approximation of the graph of f is shown in gure 3.5. Rather than he k


whether a point x [0, 1] lies in Pk for all k Z0 , the gure was generated
by he king only up to k = 6. A similar approximation of the graph of g is
shown in gure 3.6. The gure shows why g is alled the ruler fun tion.
To see that neither f nor g is pie ewise monotoni , note that any subinterval [a, b] of [0, 1] having positive width ontains points i/2k and (i + 1)/2k
onse utive in Pk for some k, and then a point j/3 of Q for some su h
that i/2k < j/3 < (i + 1)/2k . Sin e f(i/2k ) > 0 and f((i + 1)/2k ) > 0 while
f(j/3 ) = 0, f is not monotoni on [a, b]. And similarly for g.
Sin e any subinterval [a, b] of [0, 1] having positive width ontains a point
k
i/2 of Pk for some k, and a point j/3 of Q for some , it follows that every
lower sum S and every upper sum T for Ar10 (f) must satisfy
S = 0,

T 1.

Therefore there are no sequen es (Sn ) of lower sums and (Tn ) of upper sums
satisfying the ondition limn (Tn Sn ) = 0 that is ne essary for f to be
integrable (see De nition 3.3.6). That is,
Z1

f does not exist.

On the other hand, g is integrable. Every lower sum S for Ar10 (g) is 0, and
so by De nition 3.3.6, the question is whether a sequen e (Tn ) of upper sums
has limit 0. It does. The idea is to over nitely many high spikes e iently
with very narrow boxes, so that overing the in nitely many remaining low

3.3 Integrability

Figure 3.5.

117

Approximation of the fun tion that is 1 at values i/2k

spikes ine iently still produ es a small upper sum. Quantitatively, given
any > 0, there is a positive integer k su h that

1
< .
2k+1
2

Let w = /(2(2k + 1)), a positive value. Then


(2k + 1)w +

1
< .
2k+1

Cover the 2k +1 spikes over the points 0/2k , 1/2k , , 2k /2k of Pk with boxes
of width w and height 1. Cover the remainder of the graph of g with boxes
of height 1/2k+1 and total width less than 1. This gives an upper sum T su h
that
T < (2k + 1)w +

2k+1

< .

(Figure 3.7 shows the re tangles for su h an upper sum T that is slightly
bigger than 1/8.) Sin e is arbitrarily, we an reate a sequen e (Tn ) of su h

118

3 Sequen e Limits and the Integral

Figure 3.6.

Approximation of the fun tion that is 1/2k at values i/2k

upper sums with limit 0. Therefore,


Z1

g = 0.

Exercises
3.3.4. Consider two fun tions,
f : [0, 2] [0, 1],

f(x) =

g(x) =

and
g : [0, 2] [0, 1],

if 0 x 1,
x 1 if 1 < x 2,
x

if 0 x < 1,
x 1 if 1 x 2.

3.3 Integrability

Figure 3.7.

119

Small upper sum for the area under the ruler fun tion

(a) Graph f and g side by side, in a way that shows the di eren e between
them.
(b) Is f pie ewise monotoni ? Is g? Is any fun tion
h : [0, 2] R,

x
h(x) = c

x 1

if 0 x < 1,
if x = 1,
if 1 < x 2

pie ewise monotoni ?


( ) Figure 3.8 shows two arrangements of boxes. Explain why the sum of
box-areas arising from one arrangement is a lower sum for Ar20 (f) or for Ar20 (g),
and the sum of box-areas arising from the other arrangement is an upper sum
for Ar20 (f) or for Ar20 (g), but it is not the ase that the gure shows a lower{
upper sum pair for either f or g.
(d) Draw two more arrangements of boxes, similar to the gure, so that
the gure and your pi ture give you a lower{upper sum pair for f and a

120

3 Sequen e Limits and the Integral

Figure 3.8.

Box-arrangements

lower{upper sum pair for g. Using these gures, dis uss qualitatively whether
f and g are integrable, and if so, whether their integrals are equal.
3.3.5. Show by example that if f, g : [a, b] [0, M] are pie ewise monotoni
then their sum f+g : [a, b] [0, 2M] need not again be pie ewise monotoni .

(An explanation that involves some mixture of formulas, pi tures, and words
is ne.)

3.3.6 Generative Integral Rules


Proposition 3.3.13 (Generative Integral Rules).

grable fun tions

f
f~ : [a, b] [0, M].

f : [a, b] [0, M],

Then the fun tion


f
f + f~ : [a, b] [0, M + M],

is integrable, and

Zb

~ =
(f + f)

~
~
(f + f)(x)
= f(x) + f(x)

Zb

f+

Consider two inte-

Zb

~
f.

Let c R0 be a nonnegative real number. Then the fun tion


cf : [a, b] [0, cM],

is integrable, and

Zb
a

(cf) = c

(cf)(x) = c f(x)
Zb

f.

Proof. There exist a sequen e (Sn ) of lower sums for Arba (f) and a sequen e
of (Tn ) of upper sums for Arba (f) su h that

3.3 Integrability

121

lim(Tn Sn ) = 0.
n

And there exist similar sequen es (Sen ) and (Ten ) for Arba (f~) with
lim(Ten Sen ) = 0.
n

Consequently, (Sn + Sen ) is a sequen e of lower sums for Arba (f + f~) (exer ise 3.3.6) and (Tn + Ten ) is a sequen e of upper sums for Arba (f + f~), and by
the Sum Rule for sequen es,
lim((Tn + Ten ) (Sn + Sen )) = lim((Tn Sn ) + (Ten Sen ))
n

en )
= lim(Tn Sn ) + lim(Ten S
n

= 0 + 0 = 0.

Thus

Rb
a

(f + f~) exists, and its value is

Zb
a

en ) = lim(Sn ) + lim(S
en ) =
(f + f~) = lim(Sn + S
n

Zb
a

f+

Zb

f~.

The se ond part of the proposition is proved similarly (exer ise 3.3.7).

In onne tion with Proposition 3.3.13, it deserves note that the formula
~ = Arba (f) + Arba (f)
~
Arba (f + f)

is not geometri ally immediate. The problem is that the region under the
graph of f+ f~ does not naturally de ompose into two pie es with one ongruent
to the area under the graph of f and the other similar but for f~.
Proposition 3.3.14 (Inequality Rule for Integrals).

tegrable fun tions

Consider two in-

f, g : [a, b] [0, M]

su h that

f g,

meaning that f(x) g(x) for all x [a, b]. Then


Zb
a

Zb

g.

Proof. This follows from the fa t that area has sensible properties, sin e
Zb
a

b
f = Arb
a (f) Ara (g) =

Zb

g.

122

3 Sequen e Limits and the Integral

Exercises
3.3.6. The proof of Proposition 3.3.13 ta itly ites the following assertion:

Consider two fun tions

f : [a, b] [0, M],

and onsider their sum,


If

f
f + f~ : [a, b] [0, M + M],

S is a lower sum
e is a lower sum
S+S

f
f~ : [a, b] [0, M],
(f + f~)(x) = f(x) + f~(x).

for Arba (f), and Se is a lower sum for Arba (f~), then
for Arba (f + f~). While this assertion is orre t, it is not

quite automati .
(a) For onvenien e, let a = 0 and b = 1. Draw the graph of a random
fun tion f as above, and then draw three boxes whose areas add up to a lower
sum S for Ar10 (f). Separately, draw the graph of a se ond random fun tion f~,
~.
and then draw four boxes whose areas add up to a lower sum Se for Ar10 (f)
Make the breakpoints that determine the bases of the four boxes be di erent
from those that determine the three boxes from a moment ago (ex ept for the
breakpoints 0 and 1, of ourse).
(b) Draw a graph that shows the fun tions f and f + f~. Why isn't it
immediately obvious geometri ally that we an sta k the boxes from the se ond graph in part (a) on top of the boxes from the rst graph to show that
e is a lower sum for Ar1 (f + f~)? Explain how to x the problem. Your
S+S
0
answer needn't involve mathemati al symbols, but rather should be an easyto-understand des ription, perhaps illustrated by more pi tures, of what the
geometri issue is and how to address it.
3.3.7. Prove the se ond part of Proposition 3.3.13.

3.4 Summary
The notion of a sequen e limit leads to a more pre ise understanding of the
integral than we ould attain in hapters 1 and 2. In the next hapter, the
related notion of a fun tion limit will similarly larify the derivative.

4
Function Limits and the Derivative

A fun tion f has limit at the point x if its output-values f(s) approa h
as its input-values s approa h x ontinuously. To say that s approa hes x
ontinuously is to say that s approa hes x sequentially in any way whatsoever,
ex ept that s should never a tually rea h x. The fun tion limit will thus be a
ommon value of sequen e limits, the limits of the output-sequen es (f(sn ))
orresponding to all suitable input-sequen es (sn ) approa hing x. This hapter
begins by de ning fun tion limits and establishing some of their properties.
After the examples of the rst two hapters, the theory has amassed over
the most re ent hapter and will ontinue to do so over this one. So the
reader should periodi ally step ba k from details in order to appre iate the
umulative arrangement of the ideas. On e the previous hapter's de nition
of sequen e limit was in pla e, it led to basi results and generative results.
This hapter's de nition of fun tion limit will be phrased in terms of the
de nition of sequen e limit, and then it too will lead to basi and generative
results, based on their ounterparts for sequen es. All of this is arried out in
se tion 4.1. With fun tion limits in pla e, the derivative an be be de ned as
a parti ular fun tion limit. Basi derivative results and generative derivative
results thus follow from orresponding fun tion limit results, as shown in
se tion 4.2.
So far, the only spe i derivative that we know is that of the power
fun tion, but we will ompute other spe i derivatives in the hapters to
ome.

124

4 Fun tion Limits and the Derivative

4.1 The Limit of a Function


4.1.1 Definition of Function Limit

As just mentioned, the suitable input-sequen es for the de nition of fun tion limit are those sequen es that tend to a point but never rea h it. For
onvenien e, we name the phenomenon.
Definition 4.1.1 (Approaches, Approachable). Let (sn ) be a
quen e, and let x be a real number. Then (sn ) approaches x if

real se-

lim(sn ) = x but sn 6= x for ea h index n.


n

Let

A be a set of real numbers, and let x be a real number. Then x


approachable from A if some sequen e (sn ) in A approa hes x.

is

Whether a point x is approa hable from a set A is in general independent


of whether x is an element of A. That is, there are situations where x is
approa hable from A and lies in A, where x is approa hable from A but does
not lie in A, where x is not approa hable from A but lies in A, and where
x is not approa hable from A and does not lie in A. Exer ise 4.1.1 asks for
examples.
Definition 4.1.2 (Limit of a Function).

Let A be a subset of R, and let

f : A R

be a fun tion. Let


to x, notated

x R,

and let

R.

Then

f has limit as s goes

lim f(s) = ,

s x

if
(1) The point x is approa hable from A.
(2) For every sequen e (sn ) in A that approa hes x, limn (f(sn )) = .
To be clear about the notation, observe that the symbolstring limn refers to a sequence limit, whereas lims x
refers to a function limit.

Again, De nition 4.1.2 en odes the notion of the input s approa hing x
ontinuously as en ompassing all ways that s an approa h x sequentially,
and the fun tion limit is the ommon value of all the orresponding outputsequen e limits, if onditions are suitable and a ommon value exists. In natural language, the fun tion limit of f at x is the output-value that the

4.1 The Limit of a Fun tion

125

behavior of f near x suggests that f should take at x. Note that this natural language des ription uses the symbols f and x but not s. The reader is
autioned that if a al ulation of lims x f(s) in some parti ular instan e seems
to give an answer involving the symbol s then something has gone wrong.
On the other hand, although De nition 4.1.2 of lims x f(s) aptures the
value that f(s) tends to as s tends to x, the de nition makes no referen e
whatsoever to f(x) itself. Indeed, x need not even be in the domain of f. That
is:

f(x) ould exist and equal lims x f(s),

If lim f(s) exists then f(x) ould exist and not equal lims x f(s),
s x

f(x) ould fail to exist.

The de nition's insisten e on sequen es (sn ) that approa h x but never rea h
it prevents any a idental referen e to f(x).
The de nition is of most interest to us when the domain A of f ex ludes a
point x where we want to know what value f should take. For example, let x
be any real number, let
R6=x = {s R : s 6= x},

and onsider the fun tion


f : R6=x R,

f(s) =

s2 x 2
.
sx

Then also this fun tion is


f : R6=x R,

f(s) = s + x,

but even though the formula s + x is sensible for s = x, the fun tion f is not
de ned there. (Figure 4.1 shows the graph of f.) Nonetheless, x is approa hable
from R6=x , e.g., by the sequen e (sn ) = (x + 1/n), and so the rst ondition of
De nition 4.1.2 is met. Furthermore, for any sequen e (sn ) that approa hes x
we have limn (sn ) = x by De nition 4.1.1, and so the Sum Rule rule for
sequen es and the Constant Sequen e Rule ombine to give
lim(f(sn )) = lim(sn + x) = 2x.
n

Thus the se ond ondition of De nition 4.1.2 is also satis ed (with = 2x),
and so we have established a fun tion limit,
lim f(s) = 2x.

s x

Visually, the idea is that the limit ma hinery has plugged the gap in gure 4.1.
This little argument has essentially repeated the derivation of the tangent

126

4 Fun tion Limits and the Derivative

slope of the parabola in se tion 1.3, but now using the more pre ise language
at hand to buttress the ideas. Note that the al ulated value 2x of lims x f(s)
does not ontain the symbol s, as remarked after De nition 4.1.2.
f(s)

2x

PSfrag repla ements


x

Figure 4.1.

The graph of a di eren e-quotient fun tion

The next result illustrates the idea that universalizing over all sequen eapproa hes aptures ontinuous approa h.
Proposition 4.1.3 (Persistence of Inequality). Let A be a subset of R,
let f : A R be a fun tion, and let x be a point of A that is approa hable
from A. Suppose that

lim f(s) > 0 and f(x) > 0.

s x

Then for all s A lose enough to x, also f(s) > 0.


Proof. If

f(s) > 0 for all s A su h that |s x| < 1 then we are done.


Otherwise there exists some s1 A su h that
|s1 x| < 1 and f(s1 ) 0.

And s1 6= x sin e f(x) > 0. If f(s) > 0 for all s A su h that |sx| < |s1 x|/2
then we are done. Otherwise, sin e |s1 x|/2 < 1/2, there exists some s2 A
su h that
|s2 x| < 1/2 and f(s2 ) 0.

4.1 The Limit of a Fun tion

127

And s2 6= x sin e f(x) > 0. If f(s) > 0 for all s A su h that |sx| < |s2 x|/2
then we are done. Otherwise, sin e |s2 x|/2 < 1/4, there exists some s3 A
su h that
|s3 x| < 1/4 and f(s3 ) 0.

And s3 6= x sin e f(x) > 0. Continue in this fashion. Unless the pro ess proves
the proposition after nitely many steps, it produ es a sequen e (sn ) that
approa hes x but (be ause f(sn ) 0 for ea h n) fails to satisfy the ondition
limn (f(sn )) > 0. This ontradi ts the hypothesis that lims x f(s) > 0, and
so the pro ess must prove the proposition after nitely many steps.

We end this se tion with one more remark. Let f be a fun tion and let
be a real number. Sin e


|f(s) | = |f(s) | 0 ,

it follows immediately that

lim f(s) = lim |f(s) | = 0.

s x

s x

(4.1)

Like the Strong Approximation Lemma (page 75), this prin iple an be handy
to have available in isolated form for the sake of smoothing out the endgames
of arguments. We will use it in the next hapter, for example.
Exercises
4.1.1. (a) Find a subset A of the real numbers and a real number x su h that
x is approa hable from A and lies in A.
(b) Find a subset A of the real numbers and a real number x su h that x
is approa hable from A but does not lie in A.
( ) Find a subset A of the real numbers and a real number x su h that x
is not approa hable from A but lies in A.
(d) Find a subset A of the real numbers and a real number x su h that x
is not approa hable from A and does not lie in A.
4.1.2. Let x be any positive real number. Let R6=x = {s R : s 6= x}. Consider

the fun tion

f : R6=x R,

f(s) =

s3 x 3
.
sx

Does lims x f(s) exist, and if so, what is it? Explain. (The di eren e of powers
formula on page 35 may help.)

128

4 Fun tion Limits and the Derivative

4.1.2 Basic Function Limit Rules

Two examples of fun tion limits are eminently believable but still deserve to
be stated learly.
Proposition 4.1.4 (Basic Function Limits). Let A be a subset
and let x R be approa hable from A. Consider the fun tions

and
Then

f0 : A R,

f0 (s) = 1

f1 : A R,

f1 (s) = s.

of

R,

for all s

lim f0 (s) = 1

s x

and

lim f1 (s) = x.

s x

Less pedanti ally, the limits in the proposition are written


lim 1 = 1

s x

and

lim s = x.

s x

Note that here the power fun tions f0 and f1 have an arbitrary subset of R
(not R>0 ) as their domain. This point was dis ussed on page 32.

Proof. For any sequen e (sn ) in A that approa hes x,

lim(f0 (sn )) = lim(1, 1, 1, ) = 1


n

and

lim(f1 (sn )) = lim(sn ) = x.


n

For another basi fun tion limit, re all (from page 32) that the domain of
the power fun tion f is

R
6=0

R0

R>0

if Z0 ,
if Z1 ,
if Q0 but / Z0 ,
if Q<0 but / Z1 .

Note that in for all Q, 0 is approa hable from the domain of f , even
though 0 a tually lies in the domain only for Q0 .

4.1 The Limit of a Fun tion


Proposition 4.1.5 (Limit of the Power Function at Zero).
Q>0 be any positive rational number. Then

Let

129

lim f (s) = 0.

s 0

Let Q<0 be any negative rational number. Then


lim f (s) does not exist.

s 0

Finally,

lim f0 (s) = 1.

s 0

Proof. Let A denote the domain of f .

Suppose that Q>0 . Consider any sequen e (sn ) in A that approa hes 0. Let an arbitrary > 0 be given. De ne in turn = 1/ . Sin e
limn (sn ) = 0, there is a starting index N su h that:
For n N, |sn | < .
It follows that
For all n N,


|f (sn )| = |s
n | = |sn | < ( ) = .

Here sn an be negative only if Z1 , in whi h ase |(1) | = 1 and so

|s
n | = |(|sn |) | = |(1) |sn | | = | |sn | | = |sn | , giving the se ond equality
in the display. For positive sn , the se ond equality in the display is trivial. The
\<" in the display follows from its ounterpart in the previous one, be ause f
is stri tly in reasing on R>0 |here is where we use the ondition that > 0.
With the display justi ed, note that it says that limn (f (sn )) = 0. And sin e
the argument applies to any sequen e (sn ) in A that approa hes 0, we have
shown that lims 0 f (s) = 0.
Now suppose that Q<0 , so that Q>0 . For any positive integer n,
f (1/n) = (1/n) = 1/(1/n ).

By the 1/n Rule for sequen es, where the in the rule is the here,
limn (1/n ) = 0, and so f (1/n) grows without bound as n grows. Thus f (0)
does not exist.
Finally, the result that lims 0 f0 (s) = 1 is the rst part of the previous
proposition.

Exercise
4.1.3. Dis uss the meaning, the existen e, and the value of lims x s2 .

130

4 Fun tion Limits and the Derivative

4.1.3 Generative Function Limit Rules

Let A be a subset of R, and onsider two fun tions


f, g : A R.

Let c R be any number. Then the fun tions


f g, cf, fg : A R

are de ned as follows:


(f g)(x) = f(x) g(x)
(cf)(x) = c f(x)

(fg)(x) = f(x)g(x)

for all x A,
for all x A,
for all x A.

As with sequen es, these fun tions are the sum/di eren e of f and g, a onstant multiple of f, and the produ t of f and g. Also, onsider the subset of A
where g is nonzero,
Then the fun tions
are de ned to be

A = {x A : g(x) 6= 0}.
1/g, f/g : A R

(1/g)(x) = 1/g(x)

and
(f/g)(x) = f(x)/g(x)

for all x A
for all x A .

These fun tions are the re ipro al of g and the quotient of f and g.
For example, given two rational power fun tions on the positive real numbers,
their produ t is

f , f : R>0 R,

(f f )(x) = f (x)f (x) = x x = x+ ,

x R>0 ,

whi h is to say that their produ t is


f f = f+ : R>0 R.

The following result gives the limits of the newly-de ned fun tions above
in terms of the limits of f and g.

4.1 The Limit of a Fun tion


Theorem 4.1.6 (Generative Function Limit Rules).
of R, and onsider two fun tions

131

Let A be a subset

f, g : A R.

Let c R be any number. Suppose that the point x R is approa hable


from A. Suppose that lims x f(s) and lims x g(s) exist. Then
(1) (Sum/Di eren e Rule.) lims x (f g)(s) exists and is
lim (f g)(s) = lim f(s) lim g(s).

s x

s x

s x

(2) (Constant Multiple Rule.) lims x (cf)(s) exists and is


lim (cf)(s) = c lim f(s).

s x

s x

(3) (Produ t Rule.) lims x (fg)(s) exists and is


lim (fg)(s) = lim f(s) lim g(s).

s x

s x

s x

(4) (Re ipro al Rule.) Let A = {s A : g(s) 6= 0}. If x is approa hable


from A and lims x g(s) 6= 0 then lims x (1/g)(s) exists and is
lim (1/g)(s) = 1/ lim g(s).

s x

s x

(5) (Quotient Rule.) Let A = {s A : g(s) 6= 0}. If x is approa hable


from A and lims x g(s) 6= 0 then lims x (f/g)(s) exists and is
lim (f/g)(s) = lim f(s)/ lim g(s).

s x

s x

s x

The reader is reminded that the boxed formulas apply only when the limits
on their rights sides exist (and are nonzero when ne essary).

Proof. (Sket h.) Ea h of these follows immediately from the orresponding


generative rule for sequen e limits. However, the yoga of the grammar is elaborate enough to all for an example. So, let
= lim f(s),
s x

m = lim g(s).
s x

Let (sn ) be any sequen e in A that approa hes x. Then from De nition 4.1.2,
lim(f(sn )) = ,
n

lim(g(sn )) = m.
n

To prove the Sum Rule for fun tions, note that by the Sum Rule for sequen es,
also

132

4 Fun tion Limits and the Derivative

lim(f(sn ) + g(sn )) = + m,
n

whi h is to say, by de nition of the fun tion f + g,


lim((f + g)(sn )) = + m.
n

Sin e this last display is valid for every sequen e (sn ) in A that approa hes x,
De nition 4.1.2 now gives the Sum Rule for fun tions,
lim (f + g)(s) = + m.

s x

The other parts of the proposition are proved virtually identi ally.

For example, onsider any rational fun tion


f : A R,

f = g/h,

where g is a polynomial and h is a polynomial other than the zero polynomial,


and A = {x R : h(x) 6= 0}. Sin e h has only nitely many roots, any point
x A is approa hable from A. In onsequen e of the proposition,
lim f(s) = f(x).

s x

4.1.4 Order Function Limit Rules


Proposition 4.1.7 (Inequality Rule for Functions).
of R, and onsider two fun tions

Let

be a subset

f, g : A R.

Suppose that the point x R is approa hable from A, and that the limits

lims x f(s) and lims x g(s) exist. Suppose further that


f(s) g(s)

Then

lim f(s) lim g(s).

s x

Proof. Take any sequen e

for all s A.
s x

(sn ) in A approa hing x. Then limn (f(sn )) exists and equals lims x f(s), and limn (g(sn )) exists and equals lims x g(s).
Sin e f(sn ) g(sn ) for ea h n, the Inequality Rule for Sequen es gives
limn (f(sn )) limn (g(sn )). That is, lims x f(s) lims x g(s) as desired.

4.1 The Limit of a Fun tion


Proposition 4.1.8 (Squeezing Rule for Functions).
of R, and onsider three fun tions

Let

133

be a subset

f, g, h : A R.

Suppose that the point x R is approa hable from A, and that the limits

lims x f(s) and lims x g(s) exist and are equal to a ommon value .

Suppose further that

f(s) h(s) g(s)

for all s A.

Then lims x h(s) also exists and equals .


Proof. Take any sequen e (sn ) in A approa hing x. Then limn (f(sn )) exists
and equals lims x f(s) = , and limn (g(sn )) exists and equals lims x g(s) =
. Sin e f(sn ) h(sn ) g(sn ) for ea h n, also limn (h(sn )) = by the
Squeezing Rule for Sequen es. This argument holds for all sequen e (sn ) in A
approa hing x, and so lims x h(s) = as desired.

Analogues of various remarks about the Inequality Rule and the Squeezing
Rule for Sequen es apply to these rules for fun tions as well. First, if g(s) 0
for all s A and lims x g(s) exists then also lims x g(s) 0. Se ond, again
a point of the Squeezing Rule is that the middle limit exists. And third, the
hypothesis that f(s) g(s) (or f(s) h(s) g(s)) for all s A an be
weakened: the inequalities need to hold only for all s A lose enough to x.
But similarly to reindexing sequen es, we an shrink the domains of fun tions,
and in parti ular A an be made small enough so that the additional fussing
required to quantify and tra k the lose enough to x through statements and
proofs is not worth the e ort.
The basi fun tion limit rules and generative fun tion limit rules have
been established by working with De nition 4.1.2 of a fun tion limit as the
ommon value of all relevant sequen e limits. But now that some fun tion
limit rules are in pla e, the idea is to use them whenever we an do so rather
than making unne essary further arguments that refer ba k to the de nition
and universalize over sequen es. The pro ess here is similar to establishing
properties of the absolute value that make no referen e to ases and then
using the properties with no further referen e to the asewise de nition. And
it is similar to reasoning about sequen es by re ourse to the sequen e limit
rules rather than the de nition of a sequen e limit. Often in mathemati s, we
want our de nition to give rise qui kly to desirable, handy onsequen es that
we an then use and build on further, with little need to work dire tly with
the de nition thereafter.

134

4 Fun tion Limits and the Derivative

4.2 The Derivative


4.2.1 Definition of the Derivative
Definition 4.2.1 (Derivative).

Let

f : A R

be a fun tion. Let x be a point of A that is approa hable from A. Let


A6=x = {s A : s 6= x}.

Consider an auxiliary fun tion

g : A6=x R,

g(s) =

f(s) f(x)
.
sx

If the fun tion limit


f(s) f(x)
s x
sx

lim g(s) = lim

s x

exists, then its value is the derivative of f at x, denoted f (x). That is,
f (x) = lim

s x

f(s) f(x)
,
sx

if the limit exists.

If f (x) exists then f is differentiable at x.


Exercise
4.2.1. Let R6=0 = {s R : s 6= 0}. Consider the fun tion
g : R6=0 R,

g(s) =

|s| |0|
.
s0

(a) De ne a sequen e in R6=0 , (sn ) = (1/n). Does limn (sn ) exist? If so,
what is it?
(b) De ne a sequen e in R6=0 , (sn ) = (1/n). Does limn (sn ) exist? If so,
what is it?
( ) Is the absolute value fun tion di erentiable at 0? Explain.
4.2.2 A Consequence Worth Noting Immediately
Proposition 4.2.2. Let A be a subset of
suppose that the fun tion

is di erentiable at x. Then

R,

f : A R

lim f(s) = f(x).

s x

let

be a point of A, and

4.2 The Derivative

135

Exercise
4.2.2. Use generative fun tion limit rules to prove Proposition 4.2.2. The
argument ould start from the fa t that for s 6= x,
f(s) f(x) =

f(s) f(x)
(s x),
sx

and the argument should address the existen e issue.


4.2.3 The Derivative and the Tangent Line

The notion of a tangent slope as a limit of se ant slopes is unsatisfying. Yes,


the formula
f(s) f(x)
f (x) = lim
s x

sx

en odes the idea of the derivative as the limit of the se ant slopes, but alling
the left side of the display the tangent slope nesses the question of what
the tangent slope really is on eptually, of why the limit gives us something
that we already are about rather than something on whi h we bestow a
geometri ally suggestive name to make ourselves are about it. Perhaps the
tangent line is somehow the limit of the se ant lines, and perhaps it follows
that the tangent slope is the limit of the se ant slopes, but any su h argument
is well beyond the s ope of our grammar.
As a better formulation, a re hara terization of the derivative de nition|
really just a small algebrai rearrangement|gives us the quantitative language to show that the derivative has a property that aptures the idea of
tangent slope analyti ally.
Proposition 4.2.3 (Recharacterization of the Derivative). Let A be
a subset of R. Let f : A R be a fun tion. Consider a point x A that
is approa hable from A. Then for any real number ,
f (x)

exists and equals s


lim
x

f(s) f(x) (s x)
= 0.
sx

Proof. Given , ompute for any s A su h that s 6= x,


f(s) f(x) (s x)
f(s) f(x)
=
.
sx
sx

The result follows immediately.

To apply the proposition, set up its environment by taking a subset A


of R, a fun tion f : A R, and a point x A that is approa hable from A.
For any real number , the fun tion

136

4 Fun tion Limits and the Derivative


L : R R,

L (s) = f(x) + (s x)

is the fun tion whose graph is the line through the point (x, f(x)) having
slope . The quantity
f(s) L (s) = f(s) f(x) (s a),

sA

is the verti al distan e between the graph of f and L over points s A.


Assuming that lims x f(s) = f(x), we have
lim (f(s) L (s)) = lim (f(s) f(x)) lim (s a) = 0 0 = 0.

s x

s x

s x

That is, the verti al distan e over s between the graph and the line goes to 0
as s moves toward x. But by the proposition, only when f (x) exists and

is set to f (x) do we also have


lim

s x

f(s) L (s)
= 0.
sx

That is:

Only when f (x) exists is there a line through (x, f(x)) that ts the
graph of f so well that the verti al distan e from the graph to the
line tends to 0 faster than sx as s tends to x. The line is unique,
and its slope is f (x).
In other words, we now have an analyti des ription of the tangent line to
the graph of f at the point (x, f(x)). It is the best- tting line, in the sense
just quanti ed in the displayed text. These ideas lead onward to a geometri
des ription of the tangent line, to be presented in the following exer ise.
Exercise
4.2.3. Let A be a subset of R, let f : A R be a fun tion, and suppose
that f is di erentiable at the point x of A. For any real number , de ne a
fun tion (as in the se tion)
L : R R,

L (s) = f(x) + (s x).

(a) Show that


lim

s x

f(s) f(x) (s x)
sx

= f (x) .

(b) Suppose that < f (x), so that f (x) > 0. What does the limit in (a)
say about the verti al di eren e f(s) L (s) for all s lose enough to x su h

4.2 The Derivative

137

that s > x? What does this say about the graphs of L and f in relation to
one another near (x, f(x))? What does the limit in (a) say about the verti al
di eren e f(s) L (s) for all s lose enough to x su h that s < x? What does
this say about the graphs of L and f in relation to one another near (x, f(x))?
( ) Now suppose that > f (x), so that f (x) < 0. Now what holds for
the graphs of L and f in relation to one another near (x, f(x))?
(d) Let x be a point of A su h that there exist other points s of A as lose
to x as desired with s < x, and there exist other points s of A as lose to x as
desired with s > x. Complete the following senten e by repla ing ea h of X,
Y , Z and W by \above" or \below": This exer ise has shown that f (x) is

plausibly the tangent slope of the graph of f at (x, f(x)) in the geometri
sense that any line through (x, f(x)) with shallower slope uts the graph
there from X to Y moving left to right, while any line through (x, f(x))
with steeper slope uts the graph from Z to W moving left to right.
(e) What needs to be said, similarly to (d), if x is a point of A that
is approa hable from A only from the right? Same question but with left
instead of right.

4.2.4 A Basic Derivative: the Power Function Revisited

To apply the de nition of the derivative to the power fun tion from hapter 2,
f (x) = x

f : R>0 R,

rst let
and onsider the fun tion

(where Q),

R+
6=1 = {s R>0 : s 6= 1},

g : R+
6=1 R,

g(s) =

f (s) f (1)
.
s1

Consider any sequen e (sn ) in R+


6=1 that approa hes 1. This is exa tly the
sort of sequen e required by Proposition 3.3.1 (page 104), whi h says that for
any su h sequen e,
lim(g(sn )) = .
n

That is, exists at x = 1 and is f (1) = .


For general x R>0 rather than x = 1, de ne
f

R+
6=x = {s R>0 : s 6= x},

and onsider the fun tion


h : R+
6=x R,

h(s) =

f (s) f (x)
.
sx

138

4 Fun tion Limits and the Derivative

Then by a little algebra as on page 49 (exer ise 4.2.4(a)),


h(s) = x1 g(t)

where t = s/x.

(4.2)

It follows (exer ise 4.2.4(b)) that, as we found less formally in se tion 2.4,
lim h(s) = x1.

s x

That is, for ea h x R>0 , f (x) exists and is f1(x). And so:

For Q, f = f1 .
Sin e the power fun tion is di erentiable, it follows as a spe ial ase of Proposition 4.2.2 (page 134) that also:
lim f (s) = f (x).
For Q and x R>0 , s
x
In the previous paragraph, if is an integer, then the argument applies to
any nonzero x R, not only to x R>0 . If is a nonnegative integer then
a straightforward argument shows dire tly that f (0) = f1 (0), provided
that we understand 0 f1 to take the value 0 at x = 0. (exer ise 4.2.5). If
Q0 is a nonnegative rational number that isn't an integer, and s R>0 ,
then

s
f (s) f (0)
=
= s1 ,
s0
s

and so, a ording to Proposition 4.1.5, f (0) = 0 if > 1 but f (0) does not
exist if 0 < < 1. In sum:
Proposition 4.2.4 (Derivative of the Power Function).
rational number. Take the domain of f to be

R
if Z0 ,

R
if Z1 ,
6=0

/ Z0 ,
R0 if Q0 but

R>0 if Q<0 but


/ Z1 .

Let

be a

Then the formula

f = f1 ,

holds everywhere on the domain of f unless 0 < 1, in whi h ase it


holds everywhere on the domain of f ex ept at x = 0. For = 0, we have
f0 (0) = 0. For 0 < < 1, f (0) does not exist.

4.2 The Derivative

139

Sin e 0 f1 (x) = 0 for all x 6= 0, sometimes the boxed formula is understood to en ompass the ase = 0 and x = 0, i.e., 0 f1 (0) is understood to
mean 0. This is not stri tly orre t: f1 (0) does not exist. What is orre t is
that lims 0 0 f1 (s) = 0, and so in some sense 0 f1 should take the value 0
at x = 0.
At the purely pro edural level, the boxed formula says that to di eren-

tiate the power fun tion, bring the power down in front and redu e it by
one in the exponent. This is indeed the pro edure, but the reader should be

aware that there even though the pro edure is easy, there is substan e to the
result.

Corollary 4.2.5. Let be


the domain of f . Then

a rational number, and let x be any element of


lim f (s) = f (x).

s x

Proof. This follows from the proposition unless 0 < < 1 and x = 0. These
ex eptional ases are overed by Proposition 4.1.5 (page 129).

We end the se tion with one more derivative result. Exer ise 4.2.7 is to
give a sket h of the proof.
Proposition 4.2.6 (Derivative of the Absolute Value Function Away
From Zero). Let R6=0 = {x R : x 6= 0}. Consider the fun tion
f : R6=0 R,

f(x) = |x|.

This fun tion is di erentiable and its derivative is

f (x) =

1
1

if x > 0,
if x < 0.

Exercises
4.2.4. (a) Establish (4.2).

(b) Explain arefully why onsequently lims x h(s) = x1.

4.2.5. Let be a nonnegative integer. Show that f (0) = f1 (0), provided


that we understand 0 f1 (0) to be 0.

4.2.6. Explain (an informal explanation is ne) why it follows from Proposi-

tion 4.2.4 and the various fun tion limit results in this hapter that for any
algebrai fun tion f : A R (see page 67) and any x A that is approa hable from A,
lim f(s) = f(x).
s x

140

4 Fun tion Limits and the Derivative

4.2.7. Sket h a proof of Proposition 4.2.6. Your argument should not involve
any detail-work, but rather it should explain why for x > 0 the issue redu es to
the derivative of a power fun tion while for x < 0 it redu es to the derivative
of the negative of a power fun tion. For now, ite the fa t (to be proved in
the next se tion) that the derivative of the negative is the negative of the
derivative.
4.2.5 Generative Derivative Rules

We have omputed only one derivative so far, the derivative of the power
fun tion. Soon we will ompute other spe i derivatives. But in addition
to omputing the derivatives of parti ular fun tions, we an also ompute
the derivatives of ombinations of fun tions generatively, assuming that we
already know the derivatives of the fun tions individually.
Theorem 4.2.7 (Generative Derivative Rules).

and onsider two fun tions

Let A be a subset of R,

f, g : A R.

Let

c R be any number. Suppose that f and g exist on A. Also,


A = {x A : g(x) 6= 0} and suppose that f and g exist on A . Then

let

(1) (Sum/Di eren e Rule.) (f g) exists on A and is (f g) = f g .

That is,

for all x A.
(2) (Constant Multiple Rule.) (cf) exists on A and is (cf) = cf . That is,
(f g) (x) = f (x) g (x)
(cf) (x) = c f (x)

for all x A.

(3) (Produ t Rule.) (fg) exists on A, and is (fg) = fg + f g. That is,


(fg) (x) = f(x)g (x) + f (x)g(x)

for all x A.

(4) (Re ipro al Rule.) (1/g) exists on A and is (1/g) = g /g2 . That

is,

(1/g) (x) =

g (x)
g(x)2

for all x A .

(5) (Quotient Rule.) (f/g) exists on A and is (f/g) = (f g fg )/g2 .

That is,

(f/g) (x) =

f (x)g(x) f(x)g (x)


g(x)2

for all x A .

4.2 The Derivative

141

Regarding the hypotheses of the theorem, in fa t f and g are guaranteed


to exist on A , but showing this right now would take us too far a eld. As
with the generative fun tion limit rules, the reader is reminded that the boxed
formulas apply only when the derivatives on their right sides exist.

Proof. (1) To prove the Sum Rule, ompute for any x A that
(f + g)(s) (f + g)(x)
f(s) + g(s) f(x) g(x)
=
sx
sx
f(s) f(x) g(s) g(x)
+
.
=
sx
sx

The quotients on the right side have limits f (x) and g (x) individually as s
tends to x, and so their sum has limit f (x) + g (x).
(2) The proof of the Constant Multiple Rule is very similar to the proof
of the Sum/Di eren e Rule.
(3) To prove the Produ t Rule, we rst need to re all that by Proposition 4.2.2 (page 134), for any x A, sin e f (x) exists, lims x f(s) exists and
is f(x). Now ompute that
f(s)g(s) f(x)g(x)
(fg)(s) (fg)(x)
=
sx
sx
f(s)g(s) f(s)g(x) + f(s)g(x) f(x)g(x)
=
sx
g(s) g(x) f(s) f(x)
+
g(x).
= f(s)
sx
sx

By various fun tion limit rules, the right side has the desired limit as s tends
to x.
(4) For the Re ipro al Rule, suppose that x A . Then
(1/g)(s) (1/g)(x)
1/g(s) 1/g(x)
g(x) g(s)
1
=
=

.
sx
sx
sx
g(s)g(x)

As s tends to s, the rst quotient has limit g (x) and the se ond has
limit 1/g(x)2 , giving the result. Sin e g (x) exists, lims x g(s) = g(x) (again
by Proposition 4.2.2) and so sin e g(x) is nonzero, also g(s) is nonzero for
s-values lose to x. That is, we needn't worry about dividing by 0 somewhere
in this argument.
(5) Finally, the Quotient Rule follows from the Produ t Rule and the

Re ipro al Rule sin e f/g = f 1/g.


The most important generative derivative rule is alled the Chain Rule.
Given two fun tions where the odomain of the rst is the domain of the
se ond,

142

4 Fun tion Limits and the Derivative

their composition is

f : A B

g f : A C,

and g : B C,
(g f)(x) = g(f(x)).

A tually, the omposition is sensible as long as the range f(A) of A is a subset


of the domain B of g, sin e then we may respe ify the odomain of f to be B.
For example, given two rational power fun tions on the positive real numbers,
their omposition is

f , f : R>0 R>0 ,

(f f )(x) = f (f (x)) = (x ) = x ,

x R>0 ,

whi h is to say that their omposition is


f f = f : R>0 R>0 .

The Chain Rule says that the derivative of the omposition at x, i.e.,
(g f) (x), is the produ t of g (f(x)) and f (x), assuming that these both

exist. A se ond re hara terization of the derivative helps to prove the Chain
Rule smoothly.

Proposition 4.2.8 (Second Recharacterization


Consider a fun tion f : A R. Let x A be
Then for any real number ,
f (x)

of the Derivative).

approa hable from A.

exists and equals

if and only if there is a fun tion q : A R su h that



f(s) f(x) = + q(s) (s x)

where s
lim q(s) = 0 and q(0) = 0.
x

Proof. Suppose that f (x) exists and equals . De ne

f(s) f(x) if s 6= x,
sx
q(s) =

0
if s = x.


Then f(s) f(x) = + q(s) (s x) for s 6= x and for s = x. Sin e f (x) = it


follows that lims x q(s) = 0. And q(0) = 0, so q has the desired properties.
Conversely, suppose that su h a fun tion q exists. Then for all s A su h
that s 6= x we have
f(s) f(x)
= + q(s),
sx

4.2 The Derivative

and so
lim

s x

143

f(s) f(x)
= lim ( + q(s)) = .
s x
sx

That is, f (x) exists and equals .

Now the Chain Rule is straightforward to prove.


Theorem 4.2.9 (Chain Rule).

sider two fun tions

f : A B

Let A, B, and
and

be subsets of R. Con-

g : B C.

Let x be a point of A. Suppose that f is di erentiable at x and g is


di erentiable at f(x). Then the omposition g f is di erentiable at x,
and its derivative there is
(g f) (x) = g (f(x)) f (x).

Consequently, if f is di erentiable on A and g is di erentiable on


then g f is di erentiable on A with derivative

f(A),

(g f) = (g f) f : A R.

Proof. We know that for s A,


f(s) f(x) = f (x) + q(s) (s x),

lim q(s) = 0 = q(x),

s x

and that for t B,


g(t) g(f(x)) = g (f(x)) + r(t) (t f(x)),

lim r(t) = 0 = r(f(x)).

t f(x)

And we need to show that for s A,


(g f)(s) (g f)(x) = g (f(x))f (x) + q
~ (s) (s x),

lim q~ (s) = 0 = q~ (x).

s x

Compute for s lose to x,

(g f)(s) (g f)(x) = g(f(s)) g(f(x))

where


= g (f(x)) + r(f(s)) (f(s) f(x))


= g (f(x)) + r(f(s)) f (x) + q(s) (s x)

~ (s) (s x),
= g (f(x))f (x) + q


q
~ (s) = r(f(s))f (x) + g (f(x)) + r(f(s)) q(s).

144

4 Fun tion Limits and the Derivative

By Proposition 4.2.2 (page 134),


lim f(s) = f(x),

s x

and thus it seems lear that as a spe ial ase of limt f(x) r(t) = 0 we have
lim r(f(s)) = 0.

s x

But in fa t, this last display takes some detail-managing to shore up. The
problem is that the limit limt f(x) r(t) is a universal limit over all sequen es
(tn ) in B that approa h f(x), where approa h onnotes does not rea h. But
as s approa hes x, in fa t the quantity t = f(s) ould well equal f(x), and so the
limit lims x r(f(s)) in the display is not guaranteed to exist in onsequen e
of the limit being ited. However, under the ex eptional ir umstan e where
s 6= x but f(s) = f(x), we have r(f(s)) = r(f(x)) = 0, and so the display is in
fa t valid.
With the fa t that limx x r(f(s)) = 0 established, it follows by various
fun tion limit rules that lims x q~ (s) = 0. Sin e also q~ (x) = 0, the argument
is omplete.

A seemingly more natural way to go about proving the Chain Rule is by


writing for s 6= x in A,
g(f(s)) g(f(x)) f(s) f(x)
(g f)(s) (g f)(x)
=

,
sx
f(s) f(x)
sx

or, letting t = f(s) and y = f(x),


(g f)(s) (g f)(x)
g(t) g(y) f(s) f(x)
=

.
sx
ty
sx

Sin e f (x) exists, Proposition 4.2.2 says that lims x t = y. And so





g(t) g(y) f(s) f(x)

ty
sx
f(s) f(x)
g(t) g(y)
= lim
lim
s x
t y
ty
sx

= g (y) f (x) = g (f(x)) f(x).

(g f)(s) (g f)(x)
= lim
s x
s x
sx

lim

The problem with this argument is that it assumes that f(s) 6= f(x). The
ex eptional ase when s 6= x but f(s) = f(x), whi h aused a little trouble
in the argument that we gave, leads to worse lutter in the seemingly more
natural argument here.

4.2 The Derivative

145

Exercises
4.2.8. (a) Prove the Constant Multiple Rule.

(b) Prove the Quotient Rule.

4.2.9. (a) Consider the fun tion


f1 : R R,

f1 (x) = x.

Prove from s rat h that its derivative is


f1 = f0 : R R,

f0 (x) = 1.

(b) Use only the result from (a) and the Produ t Rule for derivatives to
prove that sin e f2 = f1 f1 , it follows that f2 = 2f1 .
( ) Use only the results from (a) and (b) and the Produ t Rule to prove
that f3 = 3f2 . Convin e yourself that this pro ess extends to the rule fn =
nfn1 for all n Z1 .
(d) Use the result from (a) and the Re ipro al Rule for derivatives to prove
that sin e fn = 1/fn for all n Z1 (now restri ting the domain to the

set of nonzero real numbers), it follows that fn


= nfn1 . That is, the
derivative of the power fun tion for integer exponents follows from the result
in (a) and generative derivative rules,
fn = nfn1

for all n Z.

(e) Imagine that we know f1/2 to be di erentiable, but we don't know its
derivative. Di erentiate both sides of the relation
f1 = f1/2 f1/2 ,

applying the Produ t Rule to the right side to obtain an expression involving

f1/2 and the extant-but-unknown derivative f1/2


. Solve the resulting equality

to nd f1/2 . Why isn't this pro edure an honest derivation of f1/2


?
4.2.10. Let p(x) be a polynomial. Explain why di erentiating p(x) some nite

number of times yields the zero fun tion.

4.2.11. Re all the nite geometri sum formula,


1 + x + x2 + + xn1 =

xn 1
,
x1

x 6= 1.

Di erentiate both sides of this equality to obtain a formula for the sum
1 + 2x + 3x2 + + (n 1)xn2 .

What derivative rules does this require?

146

4 Fun tion Limits and the Derivative

4.2.12. Let f : R R be any fun tion. What an be said about the ompositions f0 f, f1 f, f f0 , and f f1 , where f0 : R R and f1 : R R

are the usual power fun tions?

4.2.13. Let a, b, c, and d be real numbers, with c and d positive. Let and

be rational numbers. De ne

f : R>0 R,

f=

a + bf
.
c + df

Compute the derivative f .


4.2.14. (a) Let a and b be positive real numbers. Let and be rational

numbers. De ne

f : R>0 R,

f = (a + bf ) .

Compute the derivative f .


(b) Let a, b, c, and d be positive real numbers. Let , , and be rational
numbers. De ne
f : R>0 R,

f=

a + bf
c + df

Compute the derivative f .

4.3 Summary
The underlying on epts of sequen e limit and fun tion limit provide us with
an environment to study integrals and derivatives. So far we have studied
the al ulus of only one fun tion, the rational power fun tion. The next three
hapters will de ne, integrate, and di erentiate more fun tions: the logarithm,
the exponential, the osine and the sine.

5
The Logarithm Function

The logarithm is de ned as an area, an integral of the re ipro al fun tion f1 .


In se tion 2.5 we integrated the rational power fun tion f for 6= 1, but
exer ise 2.3.6 showed that the methods for doing so fail for the re ipro al
fun tion. That is, the methods of hapter 2 will not al ulate the logarithm.
In fa t, the logarithm is not an algebrai fun tion, and so hoping for it to have
a formula that we an study algebrai ally is futile. Nonetheless, we an analyze
its properties, and we an di erentiate and integrate it. In the pro ess, we will
expand our notion of integration, no longer requiring that the endpoints of
integration be in in reasing order, and no longer requiring that the fun tion
being integrated be nonnegative.
Se tion 5.1 de nes the logarithm and establishes its important properties. Se tion 5.2 shows that although the logarithm grows without bound, it
grows very slowly. Se tion 5.3 omputes the derivative of the logarithm, and
se tion 5.4 integrates the logarithm. Se tion 5.5 generalizes ideas that have
emerged during the ourse of the hapter to extend integration to fun tions
that aren't ne essarily nonnegative and to endpoints that aren't ne essarily
in in reasing order, and then re-establishes some of our earlier integration
results.

5.1 Definition and Properties of the Logarithm


5.1.1 Integration With Out-of-Order Endpoints
Definition 5.1.1 (Integral With Out-of-Order Endpoints). Let a
and b be real numbers with b < a. Let M 0 be any nonnegative real

number. Let

f : [b, a] [0, M]

148

5 The Logarithm Fun tion

be an integrable fun tion. Then the integral of f from


to be
Zb
Za
f=

to

is de ned

f.

That is, if the endpoints a and b are out of order then the integral from a
to b is the negative of the integral from b to a, where b and a are in order.
For example, let Q, 6= 1, and let 0 < b a. Let M be the
maximum of f (a) and f (b), so that we an onsider the fun tion
f : [b, a] [0, M].

Its integral from a to b is, by De nition 5.1.1 and then (2.11) (page 57),
Zb

f =

Za

f =

a+1 b+1
+1

b+1 a+1
.
+1

That is, the formula in (2.11) holds regardless of whether a and b are in order,
Zb

f =

b+1 a+1
+1

Q, 6= 1, a, b R>0 .

(5.1)

The formula is symboli ally robust, even though it no longer represents area
(whi h is always nonnegative) when b < a.
Exercise
5.1.1. Let a and b be real numbers with a b. Consider two integrable

fun tions

f : [a, b] [0, M],

g : [a, b] [0, N],

and let c R0 be a nonnegative real number. A ording to Proposition 3.3.13 (page 120), the fun tions f + g and cf are integrable, and
Zb
a

(f + g) =

Zb
a

f+

Zb
a

g,

Zb

(cf) = c

Zb

f.

Show that these formulas still hold if instead b < a and the fun tions f and g
have domain [b, a].
5.1.2 The Fundamental Theorem of Calculus

As another omment on formula (5.1), again given Q, 6= 1, onsider


the fun tion
F : R>0 R,

F = f+1 /( + 1).

5.1 De nition and Properties of the Logarithm

149

Then F = f , and the formula rewrites as


Zb

Q, 6= 1, a, b R>0 .

f = F(b) F(a),

(5.2)

This is an instan e of the Fundamental Theorem of Cal ulus. The theorem


says that under appropriate ir umstan es, the integral of the derivative is
the di eren e of the derivatand's values at the endpoints.
5.1.3 Definition of the Logarithm

We now use al ulus to de ne a fun tion.


Definition 5.1.2 (Logarithm).

The logarithm function,

ln : R>0 R,

is de ned as follows: For any x R>0 ,


ln(x) =

Zx

f1 .

In words, the logarithm of


from 1 to x.

is the integral of the re ipro al fun tion

De nitions 5.1.1 and 5.1.2 ombine to say that ln(x) is positive for x > 1
but negative for 0 < x < 1. Figure 5.1 shows a portion of the y = 1/x graph
together with the orresponding portion of the y = ln(x) graph. The key to
understanding the gure is to realize that the height of the y = ln(x) graph
is the area under the y = 1/x graph from 1 to x if x 1, and the negative of
the area if 0 < x < 1. Thus the y = ln(x) graph lies below the x-axis to the
left of the verti al line x = 1.
5.1.4 The Key Property of the Logarithm
Theorem 5.1.3 (The Logarithm Converts Multiplication to Addition). For all positive real numbers a and b,

ln(ab) = ln(a) + ln(b).


Proving the theorem spot-on requires some are, and it also requires moving from geometri intuition to algebrai intuition, guided by natural language. We begin with a generality before pro eeding to spe i s. This tidies
the exposition.

150

5 The Logarithm Fun tion

y = 1/x

y = ln(x)

PSfrag repla ements


x=1

Figure 5.1.

The re ipro al fun tion and the logarithm fun tion

5.1.5 Proof of the Key Property: A Generality

Let I be a nonempty interval in R, let M 0 be any nonnegative real number,


and onsider any fun tion
f : I [0, M],

integrable or not. Let a and b be any two points of I. If a and b are in order,
i.e., if a b, then we have a good visual sense of the area under the urve
y = f(x) from a to b. As before, let Arb
a (f) denote this area. Thus
For a b, Arba (f) is the area over [a, b] and under the graph of f.
This notion of Arba (f) is geometri ally intuitive, but it is symboli ally fragile
be ause it relies on the side ondition a b. It would be unsustainable to verify the side ondition every time that we make referen e to a quantity Arba (f).
But for now, we must. (Certainly we may not assume that whenever two
letters represent numbers, the earlier letter in the alphabet is guaranteed to
represent the smaller of the two numbers.) So, to make the expression Arba (f)
symboli ally robust, extend its de nition as follows:
If a > b then Arba (f) = Arab (f).

(5.3)

(For integrable fun tions, this repeats De nition 5.1.1, but the de nition here
is more general sin e f need not be integrable.) This extended de nition
of Arba (f) may not feel entirely lear or natural geometri ally, but it is utterly platoni symboli ally. An immediate onsequen e of the de nition is
that

5.1 De nition and Properties of the Logarithm

Arab (f) = Arba (f) for all a, b I.

151

(5.4)

Indeed, if a b then (5.4) simply repeats (5.3), while if a < b then (5.4)
follows from (5.3) with the roles of a and b reversed (exer ise 5.1.2).
A onsequen e of (5.4) is that
For all a, b, c I,

Arba (f) + Arcb (f) = Arca (f).

(5.5)

The all in (5.5) onnotes that the equality holds regardless of the order of a,
b, and c. Proving this amounts to verifying that all six ases redu e ba k to
the natural ase where a b c, when (5.5) holds for geometri reasons (see

gure 5.2) sin e our area fun tion, whatever it is, has sensible basi properties.
Life is too short to arry out all ve other ases, so we ontent ourselves with
working one of them, hosen at random, to get a sense of all ve arguments.
For example, suppose that b c a. Then
Arba (f)+Arcb (f)
c
= Ara
b (f) + Arb (f)
c
= (Arcb (f) + Ara
c (f)) + Arb (f)
= Ara
c (f)
= Arca (f)

by (5.4)
by (5.5) with b, c, a for a, b, c
sin e two of the terms an el
by (5.4).

To repeat, nothing in this proof of (5.5) is related to integrability.

PSfrag repla ements


a

Figure 5.2.

Arba (f) + Arcb (f) = Arca (f): the obvious ase

152

5 The Logarithm Fun tion

Exercises
5.1.2. Carefully explain the statement in the text that if a < b then (5.4)
follows from (5.3) with the roles of a and b reversed.
5.1.3. Choose another ordering of a, b, and c, and redu e the equality in (5.5)
to the ase that the three points are in order.
5.1.6 Proof of the Key Property: A Specific Argument

Now we return to studying areas under the graph of one parti ular fun tion,
the re ipro al fun tion on the positive real numbers,
f1 : R>0 R,

f(x) = 1/x.

If we restri t f to any losed subinterval I = [A, B] or I = [A, ) of its domain,


then ne essarily A > 0 and we an set M = 1/A to get
f1 : I [0, M].

Thus the restri tion of the re ipro al fun tion to I falls under the previous
and hen e,
general dis ussion. However, the power fun tion f1 is monotoni ,
R
beyond the previous generalities, it is integrable. So we write rather than Ar
in the following dis ussion. As in De nition 5.1.1, for any positive a and b,
Zb

f1 =

And (5.5) implies

For all a, b, c R>0 ,

Za

if a > b.

f1

Zb

f1 +

Zc

(5.6)

f1 =

Zc

f1 .

(5.7)

Re all that Proposition 2.5.1 (page 54) showed that for any Q,
If 0 < a b and c > 0 then

Z bc

f = c+1

ac

Zb

f .

The idea was that s aling a box horizontally by the fa tor c s ales it verti ally
by c , thus giving an area-s aling fa tor of c+1 . In the spe ial ase = 1,
i.e., in the ase of the re ipro al fun tion, we therefore have
If 0 < a b and c > 0 then

Z bc
ac

f1 =

Zb

f1 .

(5.8)

(See gure 5.3. As before, the s aled interval [ac, bc], whi h the gure shows
lying entirely to the right of the original interval [a, b], an also lie to the

5.1 De nition and Properties of the Logarithm

PSfrag repla ements

153

y = 1/x

ac
Figure 5.3.

bc

Equal areas

right of the original interval but with overlap, or to the left of the original
interval but with overlap, or entirely to the left of the original interval.)
If instead, a and b are out of order (i.e., 0 < b < a), while still c > 0,
then symboli reasoning shows that the equality in (5.8) still holds, ourtesy
of (5.6),
Z bc
ac

Z ac

f1
Zbc
a
= f1

by (5.6)

by (5.6) again.

f1 =

Zb

f1

by (5.8), with a and b ex hanged

Sin e the formula holds regardless of whether 0 < a b or 0 < b < a, we


have shown:
Zb
Z bc
f1 .
f1 =
(5.9)
For all a, b, c R>0 ,
ac

Again, we have extended a formula to make it symboli ally robust, even


though the generalized formula is no longer so intuitive geometri ally in all
ases.
5.1.7 Proof of the Key Property: End of the Proof

Using formulas (5.7) and (5.9), ompute that for all a, b R>0 ,

154

5 The Logarithm Fun tion


Z ab

f1 =

=
=

Za

1
Za

1
Za
1

f1 +
f1 +
f1 +

Z ab

by (5.7), with 1, a, ab as a, b, c

f1

a
Z ba

by basi algebra

f1

1a
Zb

by (5.9), with 1, b, a as a, b, c.

f1

In sum, we have established that


For all a, b R>0 ,

Z ab

f1 =

f1 +

But the natural logarithm is by de nition


ln : R>0 R,

Za

ln(x) =

Zb

(5.10)

f1 .

Zx

f1 .

That is, formula (5.10) is pre isely our desired result:


For all a, b R>0 ,

ln(ab) = ln(a) + ln(b).

This ompletes the proof of Theorem 5.1.3.


5.1.8 Further Properties of the Logarithm

The remaining standard properties of the logarithm follow from its de nition
and its key property of onverting multipli ation to addition.
Theorem 5.1.4 (Properties of the Logarithm).

(1) ln(1) = 0.
(2) For all positive real numbers x and x ,
ln(xx ) = ln(x) + ln(x ).
(3) For all positive real numbers x
ln(1/x) = ln(x).
(4) For all positive real numbers x and all rational numbers ,
ln(x ) = ln(x).

Proof. (1) follows from the de nition of the logarithm: ln(1) =

R1

1 f1 is the
area under the graph of the re ipro al fun tion from x = 1 to x = 1, i.e., it
is 0.

5.1 De nition and Properties of the Logarithm

155

(2) is Theorem 5.1.3.


For (3), ompute that
ln(x) + ln(1/x) = ln(x 1/x) by (2)
= ln(1)
= 0.

That is, ln(1/x) is the additive inverse of ln(x), as desired.


For (4), rst let n Z1 be a positive integer. Then
ln(xn ) = ln(x x x)
= ln(x) + ln(x) + + ln(x) by repeated appli ation of (2)
= n ln(x).
Note also that the formula ln(xn ) = n ln(x) for n = 0 redu es to 0 = 0, whi h
ertainly is true. Next let n Z1 be a negative integer. Sin e n Z1 ,
ln(xn ) = ln((1/x)n )
sin e n Z1
= n ln(1/x)
= n ( ln(x)) by (3)
= n ln(x).
Finally, let = p/q where p and q are integers with q > 0. For any x R>0 ,
let x~ = x1/q . Then x = x~p and x = x~q , and so
ln(x ) = ln(x~p ) = p ln(x~),
and onsequently

ln(x ) =

ln(x) = ln(x~q ) = q ln(x~),

p
ln(x) = ln(x).
q

This ompletes the argument.

With the properties of the logarithm established, we an revisit the proofs


of the nth Root Rule and the nth Power Rule for sequen es (exer ise 5.1.4).
Exercise
5.1.4. (a) Let b > 1 be a real number, and let > 0 be a real number. Show
that for any positive integer n,
b1/n 1 < n >

ln(b)
.
ln(1 + )

156

5 The Logarithm Fun tion

Note that this fa t gives another version of the part of the proof of Proposition 3.2.8 (4) that overs the ase b > 1 (see page 83). Using a omputer,
hoose various values b > 0 and > 0 (with presumably small) to ompare
the new starting index N > ln(b)/ ln(1 + ) from this exer ise against the
starting index N > (b 1)/ in the original proof.
(b) Let r be a nonzero real number su h that |r| < 1, and let > 0 be a
real number. Show that for any positive integer n,
|r|n < n >

ln()
.
ln(|r|)

Note that this fa t gives another proof of Proposition 3.2.8 (5) (see page 84).
Using a omputer, hoose various values r (with 0 < |r| < 1) and > 0 to
ompare the new starting index N > ln()/ ln(|r|) from this exer ise against
the starting index N > 1/(x) (where 1/|r| = 1 + x) in the original proof.

5.2 Logarithmic Growth


Sin e 2 is bigger than 1, it follows that ln(2) > 0. And by the properties of
the logarithm,
ln(4) = 2 ln(2),
ln(8) = 3 ln(2),
ln(16) = 4 ln(2),
and so on. It follows that as n grows large, so does ln(2n ). However, the
sequen e
(2n ) = (2, 22 , 23 , 24 , )

is doubling at ea h generation, growing faster and faster, whereas the orresponding sequen e of logarithms,
(ln(2n )) = (ln(2), 2 ln(2), 3 ln(2), 4 ln(2), ) = ln(2) (n)

is growing steadily in proportion to the generation- ount. Both sequen es get


large with n, but apparently at di erent rates.
Moving from a generation- ount n to a ontinuous variable x, any x 1
lies between 2n and 2n+1 for some n. Thus the ratio ln(x)/x is the quotient
of two large numbers when x is large, so that we do not immediately know
how it behaves as x grows, but we suspe t that
ln(x)
x

tends to 0 as x gets large.

5.2 Logarithmi Growth

157

In more suggestive notation, albeit involving a taboo symbol, we suspe t that


lim

ln(x)
x

= 0.

(To make the notation sensible, de ne for any fun tion f : R>0 R,
lim f(x) = lim f(1/s).

s 0

That is, the left limit exists if the right limit does, in whi h ase it takes its
value from the right limit.) To quantify our sense that the boxed statement
holds, rst see gure 5.4. Re all that ln x is the area under the y = 1/x urve
from 1 to x, the darker shaded area in the gure. On the other hand, x itself
is the area of the box in the gure, the remainder of whi h is shaded more
lightly. As the box grows rightward, the darker shaded area be omes negligible
as a portion of the total shaded area.

PSfrag repla ements


1

x
Figure 5.4.

ln(x) as a portion of x

To further quantify the argument, see gure 5.5. Again, ln(x) is the area
from 1 to x. Given > 0 (and also < 1: as usual, is a small positive number), this area is less than the area of the two boxes in the gure, the ex ess
area being shown as lighter gray. Thus, for any x > 2/ (whi h presumably is
large),


2
2
x
ln x < 1 + x
=C+

2
2

It follows that
and so

ln(x)
x

ln(x)
x

<

<

(where C =

2
2 > 0).

C
+ ,
x
2

for all x > 2C/.

But an be as small as we wish, and so the desired on lusion follows.

158

5 The Logarithm Fun tion

PSfrag repla ements /2

2/

Figure 5.5.

ln(x) is less than the two box-areas

Exercises
5.2.1. Let n be a positive integer. Draw a gure that shows learly why boxes
of base 1 on the x-axis and heights 1, 1/2, 1/3, have their tops above the
graph of the re ipro al fun tion f1 . Explain why this shows that
1 > ln(2),
1
1 + > ln(3),
2
1 1
1 + + > ln(4),
2 3

and in general
1+

1
1 1
+ + + > ln(n + 1).
2 3
n

This shows that the sum 1 + 1/2 + 1/3 + + 1/n grows without bound as
n grows.
5.2.2. (a) Use a box to show that ln(2) < 1.
(b) Use a box to show that 1/2 < ln(2). Explain why it follows that
1 < ln(4).

( ) By its de nition, the logarithm fun tion is stri tly in reasing. Therefore
there is one, and only one, number e su h that ln(e) = 1. Parts (a) and (b) of
this exer ise have shown that 2 < e < 4. Use boxes to show that in fa t e < 3.
Explain arefully whether using inner or outer boxes is the orre t hoi e for
the argument, and why. How many boxes does your argument require?

5.3 Di erentiation of the Logarithm

159

5.2.3. (a) How does ln(x10 )/x behave as x grows large?


(b) How does ln(x)/x1/10 behave as x grows large?

5.3 Differentiation of the Logarithm


Theorem 5.3.1 (Derivative of the Logarithm).

The logarithm fun tion

ln : R>0 R,

is di erentiable on its entire domain, and its derivative is the re ipro al


fun tion,
ln = f1 : R>0 R.

That is,
Consequently, also

ln (x) = 1/x for all x R>0 .


lim ln(s) = ln(x) for all x R>0 .

s x

Proof. The bulk of the work is to establish the parti ular ase that ln (1) = 1,
lim

s 1

ln(s) ln(1)
s1

= 1.

To study the limit in the display, onsider rst s-values greater than 1, and
onsider gure 5.6. The relative areas of the small box, the region under the
graph of f1 , from 1 to s, and the large box show that for s > 1,
s1
ln(s) s 1.
s

Therefore, re alling that ln(1) = 0 and dividing through by the positive quantity s 1 gives
1
ln(s) ln(1)
s

s1

1,

and subtra ting 1 all through the inequality gives


1
ln(s) ln(1)
1
10
s
s1

for s > 1.

(5.11)

Next onsider s-values less than 1 (though still positive, of ourse). This
time the numerator ln(s) ln(1) = ln(s) and the denominator s 1 of the
di eren e-quotient are both negative, so that their quotient is again positive.

160

5 The Logarithm Fun tion

1
1/s

PSfrag repla ements

1
Figure 5.6.

Bounds on the logarithm as s tends to 1 from the right

Figure 5.7 shows that now we an list three negative numbers in in reasing
order, from most negative to least negative, again giving
s1
ln(s) s 1.
s

(The fa t that the same quantities bound ln(s) regardless of whether s > 1
or 0 < s < 1 is an instan e of symboli robustness.) But this time, dividing
through by the negative quantity s 1 reverses the inequalities,
1

ln(s) ln(1)
s1

1
,
s

so that again subtra ting 1 through the inequality gives


0

ln(s) ln(1)
s1

1
1 for 0 < s < 1.
s

Inequality (5.11) and the previous display ombine into a ase-free estimate,
via the absolute value fun tion,



ln(s) ln(1)
1

0
1 1
s1
s

for all positive s 6= 1.

(5.12)

By the Re ipro al Rule for fun tions and the se ond basi fun tion limit
rule in Proposition 4.1.4 (page 128),
lim

s 1

1
1
1
= = 1,
=
s
lims 1 s 1

5.3 Di erentiation of the Logarithm

161

1/s
1

PSfrag repla ements

s
Figure 5.7.

Bounds on the logarithm as s tends to 1 from the left

so that, by (4.1) (page 127),





1

lim 1 = 0.
s 1 s

The Squeezing Rule for fun tions, applied to (5.12), now gives
s 1

so that, again by (4.1),


ln(s) ln(1)

lim

s1

lim



1 = 0,

ln(s) ln(1)
s1

s 1

= 1.

That is, we have shown that ln (1) = 1.


To nish the proof, let x be any positive real number. For any positive
real number s 6= x, properties of the logarithm and basi algebra give
ln(s) ln(x)
sx

ln(x s/x) ln(x)

ln(x) + ln(s/x) ln(x)

1
=
x
1
=
x

sx

x(s/x 1)

ln(s/x) ln(1)
s/x 1

ln(~s) ln(1)

where ~s = s/x.
s~ 1

As s tends to x, ~s tends to 1. And so by the previous al ulation,

162

5 The Logarithm Fun tion

lim

For all x R>0 ,

ln(s) ln(x)

s x

sx

1
.
x

This ompletes the proof of Theorem 5.3.1.

The fa t that ln = f1 on R>0 on e again shows us a onne tion between


integration and di erentiation: the logarithm was de ned by integrating the
re ipro al fun tion from startpoint 1 to a variable endpoint x, and the derivative of the logarithm at x is the re ipro al fun tion there. More spe i ally,
ompute for any a, b R>0 that
Zb
a

f1 =

Z1
a

f1 +

Zb

f1 =

Zb

f1

Za
1

f1 = ln(b) ln(a).

Sin e ln = f1 , this shows that formula (5.2) (page 149) extends to the
ase = 1 as well. Given Q, de ne
F : R>0 R,

F=

f+1 /( + 1)

ln

if 6= 1,
if = 1.

Then F = f in all ases, and


Zb

Q, a, b R>0 .

f = F(b) F(a),

Exercises
5.3.1. Let A be a subset of R, let R6=0 = {x R : x 6= 0}, and let
f : A R6=0

be any di erentiable fun tion. Prove that the fun tion


g : A R,

g(x) = ln(|f(x)|)

is di erentiable and its derivative is


g (x) =

f (x)
.
f(x)

You will need to use the Chain Rule (Theorem 4.2.9, page 143) twi e, and
you will need to use Proposition 4.2.6 (page 139).
5.3.2. Find the derivatives of the following fun tions f : R>0 R.
(a) f(x) = ln(cx) where c R>0 .
(b) f(x) = ln(x ) where Q.

5.3 Di erentiation of the Logarithm

163

( ) f(x) = x ln(cx ) where , Q and c R>0 .


(d) f(x) = (ln(x))2 . (Note that (ln(x))2 is not ln(x2 ).)
(e) f(x) = (ln(x))3 .
(f) f(x) = (ln(x))n where n Z1 .
(g) f(x) = (ln(x + 1)) where Q.
(h) f(x) = ln(ln(x + 1)).
5.3.3. (a) The tangent line to the graph of the logarithm at x = 1 passes
through the point (1, 0) and has slope 1/1 = 1, and so it is the graph of the

fun tion

t(x) = x 1,

Explain why
t(x) =

Zx
1

f0

x R>0 .

for x 1.

(b) Explain why f0 (x) f1 (x) for all x 1, with equality only for x = 1.
( ) Explain why parts (a) and (b) argue that that the tangent line to the
graph of the logarithm at x = 1 lies above the graph to the right of x = 1.
(d) Re alling De nition 5.1.1, explain why also
t(x) =

Zx
1

f0

for 0 < x < 1.

(e) Explain why f0 (x) f1 (x) for all x su h that 0 < x 1, with equality
only for x = 1.
(f) Explain why despite the inequality in (e) pointing the other dire tion
from the inequality in (b), parts (d) and (e) argue that that the tangent line
to the graph of the logarithm at x = 1 lies above the graph to the left of x = 1
as well.
(g) Now let x0 R>0 be any positive number, and onsider the tangent
line to the graph of the logarithm at x0 . Explain why it is the graph of the
fun tion
1
t(x) = ln(x0 ) + (x x0 ).
x0

Show that
t(x) ln(x) = ln(x/x0 )


x
1 .
x0

Explain why therefore the earlier portions of this exer ise show that t(x)
ln(x) 0, with equality only for x = x0 . That is, for any x0 R>0 , the
tangent line to the graph of the logarithm at x = x0 lies above the graph.

164

5 The Logarithm Fun tion

5.4 Integration of the Logarithm


The logarithm fun tion is tri kier to integrate than the power fun tion. We
pro eed in steps. First, we express the logarithm as a ertain nonobvious limit
that will arise in the al ulation. Se ond, we pre-emptively ompute a sum
that will arise in the al ulation as well. The sum an be omputed using
al ulus, as will be done in the text, or using only algebra, as in an exer ise to
follow. Third, with the limit and the sum in hand, we integrate the logarithm
from 1 to b where b > 1. Finally we integrate the logarithm between general
endpoints a, b R>0 . This last al ulation requires expanding our notion of
integral, sin e part of the logarithm graph lies below the x-axis.
5.4.1 An Analytic Expression for the Logarithm

Let b be a real number greater than 1. We repeat the pro ess of integrating a power fun tion, as laid out in se tion 2.5, but now for the re ipro al
fun tion f1 . That is, we engage in a pro ess of omputing ln(b).
As in se tion 2.5, let n be a positive integer, denoting the number of
boxes. Let s = b1/n , so that sn = b. Consider a geometri partition of the
interval [1, b],
Then as before,

{1, s, s2 , , sn1 , b}.

the ith interval-width is (s 1)si1 for i = 1, , n.


Take the height of the ith box to be the value of the re ipro al fun tion over
the left endpoint of its base,
the ith box-height is 1/si1 for i = 1, , n.
Thus
and so

the ith box-area is s 1 for i = 1, , n,


the sum of the box-areas is Sn = n(s 1) where s = b1/n .

As dis ussed at the end of hapter 3, Sn is an upper sum for the area under the
graph, i.e., for the logarithm, and a similar analysis with lower sums implies
that lim(Sn ) is the logarithm. That is, we have shown:
For b > 1, ln(b) = lim(b 1, 2(b1/2 1), 3(b1/3 1), ).
n

That is,

5.4 Integration of the Logarithm

165

For b > 1, ln(b) = lim(n(b1/n 1)).


n

If 0 < b < 1 instead, then ln(b) = ln(1/b) where now 1/b > 1, and this is
the limiting value of n((1/b)1/n 1). But by algebra,
n((1/b)1/n 1) = (1/b)1/n n(1 b1/n ) = (1/b)1/n n(b1/n 1).

And limn ((1/b)1/n ) = 1 for 0 < b < 1, by the nth Root Rule for sequen es,
so that the des ription of the logarithm that we derived for b > 1 holds for
all positive b:
Proposition 5.4.1. Let b R>0

be any positive real number. Then

ln(b) = lim(n(b1/n 1)).


n

The formula in the proposition overs the ase b = 1 as well, sin e in this
ase ln(b) = 0 and limn (n(b1/n 1)) = limn (0, 0, 0, ) = 0.
For large n and for b 6= 1, the sequen e-entry n(b1/n 1) is the produ t of a
large number and a number lose to 0 (positive if b > 1, negative if 0 < b < 1),
so the fa t that the sequen e tends to any value as n grows, mu h less to ln(b),
is not at all obvious. Here again we see the subtlety of al ulus.
Exercise
5.4.1. Use a omputer to get the rst digits (at least four digits) of ln(2).
Also, ask the omputer for values n(21/n 1) for large values of n. How do
these values ompare to ln(2)?
5.4.2 Another Summation Formula

To integrate the logarithm, we will need to evaluate the sum


(x) = 1 + 2x + 3x2 + + (n 1)xn2 ,

x 6= 1.

Evaluating this sum was exer ise 4.2.11, but we repeat the work here.
Let n be a positive integer and let
h(x) = 1 + x + x2 + x3 + + xn1 ,

x 6= 1.

Repeatedly applying the Sum Rule for derivatives and the power fun tion
derivative formula gives
h (x) = 1 + 2x + 3x2 + + (n 1)xn2 = (x).

But also, by the nite geometri sum formula,

166

5 The Logarithm Fun tion


h(x) =

xn 1
,
x1

x 6= 1.

Thus h = f/g where f(x) = xn 1 and g(x) = x 1, and so the Quotient


Rule for derivatives and other rules give
(xn 1) (x 1) (xn 1)(x 1)
(x 1)2
nxn1 (x 1) + 1 xn
.
=
(x 1)2

h (x) =

Equate the two expressions for h to evaluate (x),


1 + 2x + 3x2 + + (n 1)xn2 =

xn n(x 1)/x + 1 xn
,
(x 1)2

x 6= 1. (5.13)

The next exer ise shows how to evaluate this sum without using al ulus.
Exercise
5.4.2. A moment ago we used al ulus to evaluate the sum
(x) = 1 + 2x + 3x2 + + (n 1)xn2 ,

x 6= 1,

but in fa t the sum does not require al ulus. Instead, ompute that
(x) = 1 + x + x2 + + xn2
+ x + x2 + + xn2
+ x2 + + xn2

..

..
.

+ xn2 .

That is,
(x) = (1 + x + x2 + + xn2 )
+ x(1 + x + + xn3 )

+ x2 (1 + + xn4 )

+ xn2 .

Re all that x 6= 1, apply the nite geometri sum formula, and do some algebra
to rederive (5.13).

5.4 Integration of the Logarithm

167

5.4.3 The Normalized Case: Left Endpoint 1

Again let b be a real number greater than 1. To integrate the logarithm


fun tion from 1 to b, rst use a geometri partition and left endpoints, thus
reating a lower sum. Again n is the number of boxes, and s = b1/n so that
sn = b, and the ith interval-width is (s 1)si1 for i = 1, , n. However,
now
the ith box-height is ln(si1 ) = (i 1) ln(s) for i = 1, , n,

and so the sum of the box-areas is



Sn = (s 1) ln(s) 0 s0 + 1 s1 + 2s2 + 3s3 + + (n 1)sn1


= s(s 1) ln(s) 1 + 2s + 3s2 + + (n 1)sn2 .

We just found the sum here. By (5.13),


Sn = s

ln(s) n
(s n(s 1)/s + 1 sn ),
s1

and so, sin e s = b1/n ,


Sn = b1/n

ln(b1/n ) ln(1)
b1/n 1

(bn(b1/n 1)/b1/n + 1 b).

By various sequen e limit results, by the derivative al ulation for the logarithm, and by Proposition 5.4.1, it follows that
lim(Sn ) = 1 1 (b ln(b)/1 + 1 b) = b ln(b) + 1 b.
n

The sequen e (Tn ) of upper sums obtained by taking fun tion values at right
endpoints has the same limit (exer ise 5.4.3). Therefore limn (Sn ) is the integral,
Z
b

ln = b ln(b) + 1 b,

b > 1.

This is the area under the logarithm urve from x = 1 to x = b (see gure 5.8).
Exercise

5.4.3. In the al ulation just ompleted, ea h Sn is a sum of inner box-areas

be ause we determined the box-heights by sampling the logarithm fun tion


over the left endpoints of the box-bases. Show that if instead we use the right
endpoints then the relevant sum of n outer box-areas is


Tn = (s 1) ln(s) 1 s0 + 2 s1 + 3s2 + 4s3 + + nsn1
= Sn /s + (s 1) ln(s)nsn1 .

Explain why onsequently limn (Tn ) = limn (Sn ).

168

5 The Logarithm Fun tion

y = ln(x)

PSfrag repla ements


1

Figure 5.8.

Area under the logarithm urve

5.4.4 The General Case

The opening idea of this hapter was to begin moving from geometri intuition
to symboli intuition by de ning under suitable ir umstan es
Zb
a

f=

Za

if a > b.

This de nition worked perfe tly well symboli ally, but geometri ally it alled
on us to expand our notion of the integral, thinking of it as a sort of signed
area, depending on whi h dire tion we traverse the x-axis horizontally. Now,
the fa t that the logarithm fun tion takes positive and negative values (as
ompared to the power fun tion on R>0 whi h is always positive) is in entive
to extend our notion of the integral as signed area verti ally as well. That is,
the losing idea of this hapter is to view area below the x-axis as negative
area when the x-axis is traversed in the positive dire tion, i.e., from left to
right. Sin e a negative times a negative is a positive, area below the x-axis
is positive when the x-axis is traversed from right to left. The methodology
that makes all the ases uniform is to think of the integral analyti ally as a
limit of sums.
Let a and b be positive real numbers in order. That is, 0 < a b. We
want to dis uss the integral
Z
b

ln .

Although the logarithm ould take positive and/or negative values on [a, b],
depending on the values of a and b, its output-values are in any ase trapped
between two numbers, ln(a) and ln(b). As shown in gure 5.9, it is geometri ally natural to interpret the integral is the sum of two possible quantities
a possible negative quantity, the negative of the area between the
graph of the logarithm and the x-axis under the portion of the x-axis
from a to the smaller of b and 1 (this term is present only if 0 < a < 1),

5.4 Integration of the Logarithm

169

and
a possible positive quantity, the area between the graph of the logarithm and the x-axis over the portion of the x-axis from the larger
of a and 1 to b (this term is present only if b > 1).
Figure 5.9 shows a s enario where both terms are present be ause a < 1 < b.

y = ln(x)

PSfrag repla ements


a

Figure 5.9.

Integral of the logarithm as a sum of positive and negative areas

Putting aside the geometri ally natural for a moment, we pro eed in a way
that is symboli ally natural. Sin e 0 < a b, it follows that b/a 1. Let (Sn )
Rb/a
be the sequen e of lower sums for Arb/a
ln in the
1 (ln) used to ompute 1
previous subse tion, where the b there is the b/a here. Thus for ea h n Z1
there are partition points
1 = x0 < x1 < < xn1 < xn = b/a,

and boxes with bases and heights


Bi = xi ,

Hi = ln(xi1 ),

i = 1, , n,

so that the box-areas are


Ai = xi ln(xi1 ),

i = 1, , n,

and the box-area sum is


Sn = A1 + + An .

170

5 The Logarithm Fun tion

Now s ale the partition points by a to get new partition points x~i = axi
for i = 0, , n,
a=x
~0 < x~1 < < x~n1 < x~n = b,

new box-bases and box-heights for i = 1, , n,


and

e i = x
~i = axi = aBi
B

e i = ln(x
H
~i1 ) = ln(axi1 ) = ln(a) + ln(xi1 ) = ln(a) + Hi ,

and hen e new box-areas

e i = a ln(a)Bi + aAi ,
A

for i = 1, , n.

Sin e the original box-bases Bi sum to b/a 1, the new box-areas sum to
en = A
e1 + + A
en
S

= a ln(a)(B1 + + Bn ) + a(A1 + + An )

= a ln(a)(b/a 1) + aSn
= ln(a)(b a) + aSn .

So, nally,
lim(Sen ) = ln(a)(b a) + a

Z b/a
1

ln

= ln(a)(b a) + a((b/a) ln(b/a) + 1 b/a)

= (b ln(b) b) (a ln(a) a).

The question is whether this symboli Rwork is ompatible with the desired
geometri interpretation of the integral ab ln when 0 < a b. We an not
generally interpret Sen as a lower sum sin e the graph of the logarithm, and
the boxes, an lie below the x-axis during some or all of the al ulation.
However, the lower sum idea an be salvaged by hoisting the graph and the
boxes verti ally to pla e the region of interest just above the x-axis. That is,
rather than study the logarithm itself, we an study its verti al translate,
f : [a, b] [0, ln(b) ln(a)],

f(x) = ln(x) ln(a).

Subtra ting ln(a) raises the graph when ln(a) is negative, as shown in ge i ontributing to S
en satisfy
ure 5.10. Also, the heights H
e i ln(a) = ln(x
~i1 ) ln(a) = f(x~i1 ),
H

5.4 Integration of the Logarithm

171

y = ln(x) ln(a)

PSfrag repla ements


a
Figure 5.10.

Hoisting the logarithm graph

e i (H
e i ln(a)) = B
e i f(x
and summing the values B
~i1 ) from i = 1 to n shows
that
en (b a) ln(a) is a lower sum for Arb (f).
S
a

Similarly for upper sums, and so

lim(Sen ) (b a) ln(a) =

or

lim(Sen ) =

Zb

Zb

f,

f + (b a) ln(a).

The right side here is the sum of the dark area and the negative of the light
area in the right side of gure 5.11. But this is also the sum of the dark
area and the negative of the light area in the left side of the gure. Thus the
symboli methods have aptured the desired geometri quantity, and without
further ado we de ree that this is the integral,
Zb

ln =

Zb
a

en ) = (b ln(b) b) (a ln(a) a).


f + (b a) ln(a) = lim(S

If instead, 0 < b < a then, naturally,


Zb
a

ln =

Za

ln = ((a ln(a) a) (b ln(b) b)) by the previous display

so that (robustly),

(b ln(b) b) (a ln(a) a)

again,

172

5 The Logarithm Fun tion


y = ln(x) ln(a)

PSfrag repla ements

y = ln(x)

Figure 5.11.

Zb
a

Signed area as di eren e of positive areas

ln = (b ln(b) b) (a ln(a) a),

a, b R>0 .

Espe ially when a = 1, this formula extends the formula from the end of the
previous subse tion, now giving
Zb

ln = b ln(b) + 1 b,

b R>0 .

Sin e a and b an be in either order, and sin e the logarithm fun tion
takes positive and negative values, we summarize how to tra k the boxed
formula for the integral as a signed area. First, area between the x-axis and
the graph of the logarithm is deemed positive when the graph lies above
the axis, negative when it lies below, and the integral is the net signed area
if 0 < a b. However, if 0 < b < a then the integral is the negative of the net
positive area. All of this is easier to understand visually than verbally. But
sin e the boxed formula is insensitive to the ases, symboli understanding is
truly the easiest of all in this ontext.
5.4.5 The Fundamental Theorem of Calculus Again

To end this se tion, onsider the fun tion


F : R>0 R,

F(x) = x ln(x) x.

We have established the formula


Zb
a

ln = F(b) F(a),

a, b R>0 .

By now the reader anti ipates that the derivative of F is the fun tion whose
integral F is, i.e., F = ln. To see this, note rst that F is built from the
di erentiable fun tions,
f1 , ln : R>0 R.
Various results about derivatives on rm that indeed F = ln (exer ise 5.4.4).

5.5 Signed Integration in General

173

Exercise
5.4.4. Complete the veri ation that the fun tion F just given indeed has
derivative F = ln.

5.5 Signed Integration in General


5.5.1 The Integral Revisited

The following de nition aptures the expanded notion of the integral that
emerged during this hapter in the ourse of integrating the logarithm.
Definition 5.5.1. Let a and b be
be real numbers with L M. Let

real numbers with

a < b.

Let

and

f : [a, b] [L, M]

be a fun tion. If the verti ally translated fun tion with nonnegative outputs
g : [a, b] [0, M L],

g(x) = f(x) L

is integrable from a to b, then the integral of f from a to b is de ned to


be
Zb
Zb
f=

g + (b a)L.

And then, as before, also the integral with out-of-order endpoints is de ned to be
Z
Z
a

f=

f.

The rst part of De nition 5.5.1 raises a problem. Its formula


Zb
a

f=

Zb

g + (b a)L

makes use of the datum L from the odomain of f, but this odomain an be
hosen with onsiderable exibility. If we keep the domain and the rule but
hange the odomain to get
f
f : [a, b] [e
L, M],

then on the fa e of it, the de nition ould pres ribe a di erent value of ab f.
We need to ensure that this doesn't happen.
Geometri ally, the issue is as follows. The rst part of De nition 5.5.1
says to hoist the graph of f above the x-axis, ompute the integral under

174

5 The Logarithm Fun tion

the hoisted graph, and then adjust by an amount orresponding to the area
gained by the hoisting. (Again, see gure 5.11.) What we need to verify is that
hoisting the graph by a di erent amount and then making the orresponding
di erent adjustment leads to the same answer. Phrasing the matter in these
geometri terms makes the desired on lusion essentially ines apable: hoisting
the graph higher adds a re tangular region between the x-axis and the hoisted
graph, but the orre tion fa tor will be modi ed by an amount equal to the
area of the added re tangular region, leading to the same value for the signed
area as before. (See gure 5.12.)

e
L

y = ln(x)

PSfrag repla ements


a
e
L

y = ln(x) e
L

Figure 5.12.

e
L

Same signed area as another di eren e of positive areas

To verify analyti ally that the de nition makes sense, we pro eed as follows. To repeat, for the original hoi e of odomain we have
f : [a, b] [L, M],

and then
and then

g : [a, b] [0, M L],


Zb
a

f=

Zb

g(x) = f(x) L,

g + (b a)L.

For the modi ed hoi e of odomain we have

and then

f (same domain and same rule),


f : [a, b] [e
L, M]
fe
~ : [a, b] [0, M
g
L],

g
~(x) = f(x) eL,

5.5 Signed Integration in General

and then also

Zb

f=

Zb
a

175

g
~ + (b a)eL.
R

We want to show that the two de nitions of ab f are equal. By symmetry we


may assume that eL L. Thus for all x [a, b],
or

g(x) g~(x) = (f(x) L) (f(x) e


L) = e
L L,
g = g~ + (e
L L).

Sin e g and g~ are both nonnegative fun tions, and eL L is a nonnegative


onstant, the fa t that area has sensible properties says that
Zb

g=

Zb
a

g
~ + (b a)(eL L).

Consequently, the rst de nition of


Zb

f=

=
=

Zb

a
Zb

a
Zb
a

Rb
a

f be omes

g + (b a)L
g~ + (b a)(e
L L) + (b a)L
g~ + (b a)e
L.

This is the se ond de nition of f, and we are done showing


Rb that De nition 5.5.1
is well de ned when a b. Sin e the de nition of a f when a > b is given
R
in terms of ab f when a b, it is well de ned too.
Exercise
5.5.1.
(a) Draw a gure illustrating why it is geometri ally obvious that
R1
f
=
0.
1
1
R1
f1 by hoisting.
(b) Carry out the pro ess of omputing 1
5.5.2 Generative Integral Rules Revisited

With integration now de ned for fun tions whose values ould be negative,
and for out-of-order endpoints, we need to revisit the results that (suppressing
the ne print)
 the integral of a sum is the sum of the integrals, and

176

5 The Logarithm Fun tion

 the integral of a onstant multiple is the onstant multiple of the integral.

The proofs of the upgraded results should not repeat the earlier proofs for
nonnegative fun tions, but rather should ite them and arry out only the
new work required now for the upgrades. In other words, the required work is
to push the previous results through the hoisting pro ess that de nes integrals
of fun tions whose values ould be negative.
Proposition 5.5.2 (Generative Integral Rules).

grable fun tions

f
f~ : [a, b] [e
L, M].

f : [a, b] [L, M],

Then the fun tion

f
f + f~ : [a, b] [L + e
L, M + M],

is integrable, and

Zb

(f + f~) =

Zb

(f + f~)(x) = f(x) + f~(x)

f+

Consider two inte-

Zb

f~.

Let c R0 be a nonnegative real number. Then the fun tion


cf : [a, b] [cL, cM],

is integrable, and

Zb

(cf) = c

(cf)(x) = c f(x)
Zb

f.

Let c R<0 be a negative real number. Then the fun tion


cf : [a, b] [cM, cL],

is integrable, and

Zb

(cf) = c

(cf)(x) = c f(x)
Zb

f.

Finally, the three equalities hold if instead a > b.


Proof. De ne
and
Then their sum

g : [a, b] [0, M L],

g(x) = f(x) L

fe
g
L],
~ : [a, b] [0, M

g
~(x) = f~(x) eL.

f L L]
e
g+g
~ : [a, b] [0, M + M

5.5 Signed Integration in General

is

(g + g
~)(x) = (f + f~)(x) (L + eL).

Therefore
Zb

177

(f + f~) =

=
=

Zb

a
Zb

a
Zb

(g + g
~) + (b a)(L + eL)
g + (b a)L +

f+

Zb

Zb
a

f~.

g
~ + (b a)eL

(The integral on the right side of the top line exists and is the sum of the two
integrals on the se ond line by Proposition 3.3.13 on page 120 be ause they
exist and g and g~ are nonnegative, and they exist be ause f and f~ are given
to be integrable.)
For c R0 , de ne
cg : [a, b] [0, c(M L)],

(cg)(x) = (cf)(x) cL.

Then
Zb

(cf) =

Zb

(cg) + (b a)cL

=c

Zb

=c

Zb

g + (b a)L

=c

Zb

by Proposition 3.3.13

g + c(b a)L

f.

To establish the result for c R<0 , we may show that


Zb

(f) =

Zb

f,

be ause any negative number is a positive multiple of 1, and we already


have the result for c R>0 . Re all that
Zb
a

f=

Zb

g + (b a)L where g(x) = f(x) L.

Also de ne
h : [a, b] [0, M L],

h(x) = f(x) + M.

178

5 The Logarithm Fun tion

Then

Zb
a

It follows that

Zb

(f) =

f+

Zb

Zb

h (b a)M.

(f) =

Zb

a
Zb

g + (b a)L +

Zb

h (b a)M

(g + h) + (b a)(L M).

But note that for any x [a, b],


(g + h)(x) = f(x) L f(x) + M = M L,

so that

Zb

and therefore
Zb

(g + h) = (b a)(M L),

f+

Zb

(f) = (b a)(M L) + (b a)(L M) = 0.

This is the desired result,


Zb

Zb

(f) =

f.

Finally, the three results


R for a R> b follow from those for a b be ause of

the basi de nition that ab f = ba f.

Similarly, the Inequality Rule for integrals needn't require that the fun tions involved be nonnegative.

Proposition 5.5.3 (Inequality Rule for Integrals, Second Version).

Consider two integrable fun tions

f, g : [a, b] [L, M]

su h that

f g,

meaning that f(x) g(x) for all x [a, b]. Then


Zb
a

Zb

g.

Proof. We already have the result for f L and g L, and the result follows
immediately sin e
Zb
a

f=

Zb
a

(f L) + (b a)L

Zb
a

(g L) + (b a)L =

Zb

g.

5.5 Signed Integration in General

179

5.5.3 The Area Between Two Curves


Proposition 5.5.4 (Area Between Two Curves As An Integral). Let
a and b be real numbers with a b. Consider two integrable fun tions
f, g : [a, b] R.

The area between the graphs of f and g is


Zb

|g f|.

Proof (Sket h of the proof.). For nonnegative fun tions

[0, M], de ne

f, g : [a, b]

max{f, g}, min{f, g} : [a, b] [0, M]


as follows. For any x [a, b],
max{f, g}(x) =

min{f, g}(x) =

and

g(x) if g(x) f(x),


f(x)

if g(x) < f(x)

if g(x) f(x),
g(x) if g(x) < f(x).

f(x)

Thus for all x [a, b],


max{f, g}(x) min{f, g}(x) =

g(x) f(x) if g(x) f(x),


f(x) g(x) if g(x) < f(x)

= |g(x) f(x)|.

Then the area is


Ar max{f, g}) Ar min{f, g}) =
b
a(

b
a(

=
=

Zb

a
Zb

a
Zb

max{f, g}

Zb

min{f, g}

(max{f, g} min{f, g})


|g f|.

A subtle point here is that if f and g are integrable then so are max{f, g}
and min{f, g}, so that the areas in the argument are integrals as ta itly asserted. This will be obvious in the examples where we apply the proposition,
and so we omit the general argument.
For bounded fun tions f, g : [a, b] [L, M], possibly taking negative
values now, a hoisting argument redu es the problem to the nonnegative ase.

180

5 The Logarithm Fun tion

For example, onsider the fun tions


f, g : [0, 2] R

where

f(x) = x2 /2 ,

g(x) = x3 5x2 /2 + 2x.

Compute that their di eren e is


(g f)(x) = x3 3x2 + 2x = x(x2 3x + 2) = x(x 1)(x 2).

This shows that f = g at x = 0, at x = 1, and at x = 2, while


g > f on (0, 1)

and g < f on (1, 2).

(See gure 5.13.) Therefore the area between the graphs of f and g is
A=

Z1

(g f) +

The rst integral is


Z1

(g f) =

and the se ond is


Z2

(f g) =

14 04
4

24 14
4

Z2

(f g).

+3

13 03
3

23 13
3

Thus the total area between the graphs is


A=

+2

12 02
2

22 12
2

1
,
4

1
.
4

1
.
2

(In the rst half of this al ulation, we have applied the formula
Zb
a

f =

b+1 a+1
,
+1

6= 1

in the ase where the endpoint a is 0. This extension has not yet been justi ed,
and it is valid only for 0, but we take it as granted for now sin e we will
dis uss it arefully in hapter 8.)
Exercises
5.5.2. Find the area en losed by the urve whose equation is
y2 + 2xy + 2x2 = 1.

(See gure 5.14. The urve is an ellipse. Use the quadrati equation to solve
the equation for y in terms of x, and then integrate between the two x-values
for whi h there is one y-value.)

5.5 Signed Integration in General

181

g
g

PSfrag repla ements

f
0

1
Figure 5.13.

Area between two urves

5.5.3. (a) Find the area of the region shown between the urves y = x(x 2)
and y = x3 in gure 5.15.
(b) Find the area of the region shown between the urves y = 8x3 /9
2
2x /9 x and y = 2x/3 in gure 5.16.
( ) Find the area of the region shown between the urves y = (x2)(x3)
and y = (x 1)(x 2)(x 3) in gure 5.17.

182

5 The Logarithm Fun tion

Figure 5.14.

Figure 5.15.

Figure for exer ise 5.5.2

Figure for exer ise 5.5.3 (a)

5.5 Signed Integration in General

Figure 5.16.

Figure for exer ise 5.5.3 (b)

Figure 5.17.

Figure for exer ise 5.5.3 ( )

183

6
The Exponential Function

The exponential fun tion is the most important fun tion in mathemati s. It
an be des ribed in various ways, all ompatible. This hapter de nes the exponential fun tion as the inverse fun tion of the logarithm, i.e., as the fun tion
that undoes the e e t of the logarithm,
y = exp(x) if and only if x = ln(y).

The exponential fun tion is also the unique di erentiable fun tion f : R R
that is its own derivative and takes a normalized value at 0:
If f = f and f(0) = 1 then f = exp.
And the exponential fun tion is a limit of ever higher powers of quantities
ever loser to 1,

x n 
exp(x) = lim 1 +
.
n

Se tion 6.1 introdu es the notion of a ontinuous fun tion. The general
prin iple that di erentiable fun tions are ontinuous gives the ontinuity of
the parti ular fun tions that we have di erentiated in these notes. The Intermediate Value Theorem says that a ontinuous fun tion whose domain is
an interval an not jump over feasible outputs: if the fun tion assumes two
values then it assumes every value between them as well. One onsequen e
of the Intermediate Value Theorem is the existen e of nth roots of positive
real numbers, something that we invoked in hapter 2. Another onsequen e
of the theorem is that every real number is a logarithm, and so the logarithm
fun tion has an inverse. Se tion 6.2 de nes the exponential as this inverse
fun tion. The properties of the logarithm therefore give rise to orresponding
properties of the exponential. These properties lead to a de nition of raising
any positive real number to any real exponent, whereas before we ould raise
a positive real number only to a rational exponent. Se tion 6.3 quanti es the

186

6 The Exponential Fun tion

oft- ited fa t that the exponential fun tion grows very qui kly. Se tion 6.4
shows that the exponential fun tion is its own derivative. Se tion 6.5 integrates the exponential fun tion. Se tion 6.6 shows that the exponential is a
limit of powers as des ribed above, and then gives an interpretation of the
limit in terms of ompound interest.

6.1 Continuity
6.1.1 Definition of Continuity
Definition 6.1.1 (Continuity).

Let A be a subset of R, and let

f : A R

be a fun tion. Let


if

be a point of A. The fun tion f is

continuous at x

lim f(s) = f(x).

s x

The fun tion f is

continuous on A

if it is ontinuous at ea h x A.

A ording to this de nition, in order for f : A R to be ontinuous


at x, ne essarily
 x A,
 x is approa hable from A,
 and for every sequen e (sn ) in A that approa hes x, limn (f(sn )) = f(x).

Espe ially, if x A but x is not approa hable from A then f an not be ontinuous at x. So, for example, under our de nition no fun tion f : Z R
an be ontinuous (exer ise 6.1.1). The reader is alerted that under a di erent
mathemati al onvention, more ommon than ours, the approa hability ondition is not required for ontinuity, and so every su h fun tion is ontinuous.
To show that a fun tion f : A R is dis ontinuous at a point x A,
that is approa hable from A, it su es to nd a sequen e (sn ) in A that approa hes x while the orresponding output sequen e (f(sn )) does not onverge
to f(x). For example, take the fun tion
f : R R,

f(x) =

0 if x 0,
1 if x > 0.

Consider the sequen e (sn ) = (1/n). The sequen e approa hes 0. The orresponding sequen e of outputs,
(f(sn )) = (1, 1, 1, ),

6.1 Continuity

187

has limit 1. And so, re alling that f(0) = 0,


(sn ) approa hes 0

but lim(f(sn )) 6= f(0).


n

Thus De nition 6.1.1 is not satis ed for x = 0, i.e., f is dis ontinuous at 0.


The following result says that most of the fun tions that we have worked
with in these notes are ontinuous.
Proposition 6.1.2 (Differentiability Implies Continuity).

subset of R, and onsider a fun tion

Let

be a

f : A R.

If f is di erentiable on A then f is ontinuous on A.


Proof. This proposition only rephrases Proposition 4.2.2 on page 134 (exer ise 6.1.3).

Although all di erentiable fun tions are ontinuous, not all ontinuous
fun tions are di erentiable. The simplest example is the absolute value fun tion,
f : R R,

f(x) = |x|.

We saw in exer ise 4.2.1 (page 134) that f is not di erentiable at 0. But it
is ontinuous at 0, be ause for any sequen e (sn ) in R6=0 , to say that (sn )
approa hes 0 is to say that |sn 0| grows small as n grows large, i.e., f(sn )
tends to 0, whi h is f(0).
Extending the example of the absolute value fun tion, Weierstrass used an
analyti pro ess of superimposing ever more, ever smaller orners, to reate
a fun tion f : R R that is ontinuous everywhere and di erentiable
nowhere. Its graph is somehow jagged no matter how losely we zoom in.
Proposition 6.1.3 (Continuity of the Power Function). Let be
rational number. The power fun tion f is ontinuous on its domain.

Proof. This proposition rephrases Corollary 4.2.5 on page 139.

However, for another example of ontinuity without di erentiability, let

be any rational number su h that 0 < < 1. Consider the th power


fun tion, whose value at 0 is f (0) = 0 (see the dis ussion on page 32),
f : R0 R,

f (x) = x .

This fun tion is ontinuous at 0. However, Proposition 4.2.4 (page 138) says
that f (0) does not exist.

188

6 The Exponential Fun tion

Proposition 6.1.2 says fairly broadly that the rational power fun tions f
and the logarithm fun tion are ontinuous on their domains. (The ex eption
is that the proposition does not say that f is ontinuous at 0 for 0 < < 1,
but the previous paragraph has taken are of this ase.) In parti ular, let
R6=0 = {x R : x 6= 0} and onsider the re ipro al fun tion
f1 : R6=0 R,

f(x) = 1/x.

As just remarked, f1 is ontinuous on R6=0 . And yet to graph f1 , we must


drasti ally lift the pen il from the page sin e f(x) is very negative for x a
little less than 0, while f(x) is very positive for x a little greater than 0. This
example shows that the ommon idea of a ontinuous fun tion as one that
an be graphed without lifting one's pen il is not entirely orre t. The issue
here is that the domain R6=0 of f has two pie es. The graph of f on ea h pie e
of R6=0 an be drawn in one stroke. The ontinuity of f means that its graph
has no more breaks than its domain.
Exercises
6.1.1. Explain why no fun tion f : Z R is ontinuous under De ni-

tion 6.1.1. A qualitative explanation is ne. For a more quantitative one, the
hoi e = 1/2 ould be helpful.

6.1.2. Consider the fun tion


f : R R,

f(x) =

0 if x < 0,
1 if x 0.

Show that f is not ontinuous at 0.


6.1.3. Explain with some are why Proposition 6.1.2 repeats Proposition 4.2.2.
6.1.2 Continuity and Integrability
Theorem 6.1.4 (Continuity Implies Integrability).

numbers with a b, and onsider a fun tion


f : [a, b] R.

If f is ontinuous on [a, b] then the integral

Rb
a

Let a and b be real

exists.

Unfortunately, the proof of Theorem 6.1.4 is beyond our s ope.


The fun tions that we know to be integrable are bounded pie ewise monotoni fun tions. Su h a fun tion need not be ontinuous, showing that the
onverse of Theorem 6.1.4 does not hold.

6.1 Continuity

189

For an example of a bounded ontinuous fun tion that is not pie ewise
monotoni , onsider
f : [1, 1] [1, 1],

f(x) =

x sin(1/x) if x 6= 0,

if x = 0.

The graph of f is shown in gure 6.1.


0.8
0.6
0.4
0.2
-1

-0.5

0.5
-0.2

Figure 6.1.

A bounded, ontinuous, but not pie ewise monotoni fun tion

6.1.3 The Intermediate Value Theorem

As dis ussed a moment ago, even though the re ipro al fun tion f(x) = 1/x is
ontinuous, its graph has two pie es. However, this is be ause its domain R6=0
has two pie es already. If a fun tion is ontinuous and its domain is an interval,
then be ause the domain has no breaks, the graph has no breaks either. The
following result expresses this idea analyti ally.
Theorem 6.1.5 (Intermediate Value Theorem).
numbers with a < b. Let the fun tion

Let

and

be real

f : [a, b] R

be ontinuous, and suppose that f(a) < 0 and f(b) > 0. Then there exists
some number c (a, b) su h that f(c) = 0.
That is, if a ontinuous fun tion whose domain is a losed interval goes
from taking a negative output-value to taking a positive output-value, then

190

6 The Exponential Fun tion

it must take the output-value 0 somewhere in between. Note that the on lusion of the theorem isn't simply \f(c) = 0," whi h in isolation would be
meaningless sin e the hypotheses make no mention of a point c; rather the
on lusion is that there exists some c su h that f(c) = 0, although the theorem is mute on the whereabouts of c or how to nd it. Again the phrase there
exists, whose meaning is not truly settled, has arisen.
The negative, positive, and zero in Theorem 6.1.5 are normalizations.
More generally we have
Corollary 6.1.6 (Intermediate Value Theorem, Second Version).
a and b be real numbers with a < b. Let the fun tion

Let

f : [a, b] R

be ontinuous. Let v be any value between


some number c (a, b) su h that f(c) = v.

f(a)

and

f(b).

Then there is

To redu e the orollary to the theorem, rst repla e the f in the orollary
by f1 = f v. Then f(c) = v if and only if f1 (c) = 0, and 0 lies between f1 (a)
and f1 (b). Se ond, if f1 (a) > 0 and f1 (b) < 0 then repla e f1 by f2 = f1 .
Now the hypotheses for the original theorem are met, and the on lusion of
the original theorem gives the on lusion of the orollary, as desired. A ta it
point here is that be ause f is ontinuous, so are f1 and f2 . Exer ise 6.1.5
is to draw pi tures illustrating the argument given in this paragraph, and to
explain the ta it point.
The orollary says that if a fun tion is ontinuous on an interval, its output
an not jump over a value: if two numbers are output-values of the fun tion
then all numbers between them are output-values as well. This is the sense
in whi h a ontinuous fun tion is graphed without lifting the pen il from the
page.
Here is an attempt to prove Theorem 6.1.5 rather than assume it. There are

f-inputs in [a, b] su h that the orresponding f-output is negative, su h as a.


The number b ex eeds all su h inputs. So surely there is a least value c that
is at least as big as all su h inputs. If f(c) > 0 then sin e lims c f(s) = f(c),
ne essarily f(s) > 0 for all s lose enough to c (see Proposition 4.1.3 on
page 126), and so some value s < c is also at least as big as all f-inputs
that produ e negative outputs, ontradi ting the fa t that c is the least su h
value. Similarly, if f(c) < 0 then ne essarily f(s) < 0 for some value s > c,
ontradi ting the fa t that c is at least as big as all f-inputs that produ e
negative outputs. The only possibility remaining is that f(c) = 0.

However, rather than prove the Intermediate Value Theorem, this argument shows only that it follows from any assumption about the real number

6.1 Continuity

191

system that makes valid the Then surely there is a least value. . . statement
in the previous paragraph. The issue here is identi al to the one that arose
from the attempt to prove the Ar himedean Property of the real number
system ba k on page 76.
Exercises
6.1.4. Let p(x) = x3 3x + 1. Use the Intermediate Value Theorem to show
that there are at least three di erent numbers a, b, and c su h that p(a) =
p(b) = p(c) = 0.
6.1.5. (a) Illustrate the argument that the se ond version of the Intermediate
Value Theorem follows from the rst.
(b) Let f : [a, b] R be a ontinuous fun tion. Show that for any real
number h, also f + h is ontinuous. Show that for any real number c, also cf
is ontinuous.
6.1.6. Let f : [0, 1] [0, 1] be a ontinuous fun tion su h that f(0) = 1
and f(1) = 0. Draw a pi ture illustrating the situation. Geometri ally, it is
ompelling that the graph of f must ross the 45-degree line y = x at least
on e; that is, f(c) = c for some c (0, 1). Use the Intermediate Value Theorem

to prove this. (Suggestion: use an auxiliary fun tion.)

6.1.4 Applications of the Intermediate Value Theorem

For our rst appli ation of the Intermediate Value Theorem, we return to
the subje t of nth roots of positive real numbers. This topi was dis ussed
starting on page 30, and the reader is en ouraged to review the dis ussion
there before ontinuing here.
Let n 2 be an integer, and let b > 1 be a real number. Consider the
power fun tion
fn : [1, b] R,

fn (x) = x x (n times).

We have argued that fn is stri tly in reasing on R>0 . Also, we have argued
that fn is di erentiable and hen e ontinuous on all of R, so that its restri tion
here to [1, b] is ontinuous as well. Neither of these arguments made any
referen e to the existen e of nth roots. Note that be ause b > 1 and n 2,
f(1) = 1 < b

and

f(b) = bn > b.

By the Intermediate Value Theorem, there exists some number c (1, b) su h


that f(c) = b. And there is only one su h c be ause f is stri tly in reasing.

192

6 The Exponential Fun tion

That is, there is exa tly one c su h that cn = b. In other words, c is the
unique positive nth root of b, ni ely served up to us by the theorem.
If 0 < b < 1, then 1/b > 1 and the argument just given produ es the nth
root c of 1/b, and 1/c is the nth root of b. And if b = 1 then b is its own
nth root. This overs all ases, and now our invo ation of unique nth roots
(page 31) is a onsequen e of whatever assumed property of the real number
system will prove the Intermediate Value Theorem.
For our se ond appli ation of the Intermediate Value Theorem, we know
that the logarithm fun tion is ontinuous on R>0 , and we have shown in
exer ise 5.2.2 (page 158) that
ln(2) < 1 < ln(4).
The exer ise then argued that onsequently there is one and only one number e
between 2 and 4 su h that ln(e) = 1. There is at most one su h number sin e
the logarithm is stri tly in reasing, but the fa t that there is at least one su h
number was supportable for us earlier only at the level of intuition. Now it
follows from the Intermediate Value Theorem. To repeat:
Definition 6.1.7 (The number e). The unique real number x that sat= 1 is denoted e. That is, e is de ned by the

is es the ondition ln(x)


property

ln(e) = 1.

Continuing to work with the logarithm fun tion, let y be any positive real
number. By the Ar himedean property of the real number system, there is
some positive integer n su h that n > y. Thus the logarithm fun tion
ln : [1, en ] R

satis es (sin e ln(1) = 0 and ln(en ) = n ln(e) = n)


ln(1) < y and ln(en ) > y.
By the Intermediate Value Theorem,
ln(x) = y for some x (1, en ).

That is, every positive real number is a logarithm. Similarly, if b < 0 then
sin e b = ln(x) for some x, it follows that b = ln(x) = ln(1/x). And of
ourse, 0 = ln(1). Sin e the logarithm fun tion is stri tly in reasing, we have
proved the following result.
Proposition 6.1.8. Ea h

real number y takes the form


a tly one positive number x. That is, the fun tion
ln : R>0 R

takes ea h value in its odomain exa tly on e.

y = ln(x)

for ex-

6.1 Continuity

193

Exercise
6.1.7. Let b R>0 be a positive real number. Consider the fun tion
g : R R,

g(x) = x2 b.

Thus the unique positive number c su h that g(c) = 0 is the square root of b.
(a) For any positive real number s, the height of the graph of g over s is g(s)
(this ould be negative) and the tangent slope of the graph is g (s) = 2s. Use
these data and analyti geometry to show that the tangent line to the graph
of g at (s, g(s)) meets the x-axis at
s~ =

1
2



b
s+
.
s

Note that sin e b and s are positive, so is ~s.


(b) Choose any real number s1 > b, and then de ne a sequen e re ursively using the formula from (a),
sn+1 =

1
2



b
sn +
,
sn

n 1.

(6.1)

Show that for any n Z1 , if s2n > b then onsequently s2n+1 > b. Sin e
s21 > b, it follows (you need not explain this part, but put some thought into
it) that s2n > b for all n Z1 . And
then it further follows, be ause all
the sn -values are positive, that sn > b for all n Z1 .
( ) Show that sn+1 sn for all n Z1 . So (again, you needn't explain
what follows, but put thought into it) the sequen e(s1 , s2 , s3 , ) onsists
of entries that grow ever smaller, but the number b is at most as big as all
the sn . Then surely there is a greatest number c at most as big as all the sn .
This c is the limit of the sequen e: the sequen e elements sn get ever loser
to c as n grows, and if they don't get within > 0 of c then c + is at most
as big all the sn , ontradi ting the fa t that c is the greatest su h number.
(d) Take the limit of both sides of (6.1), arefully
explaining your use of
various sequen e limit rules, to on lude that c = b. Thus the pro ess in
this exer ise (a spe ial ase of Newton's method) omputes b.
(e) Use some
form of omputing power to investigate how qui kly the
values sn tend to b for various values of b and various starting approximations s1 for ea h b.

194

6 The Exponential Fun tion

6.2 Definition and Properties of the Exponential


Function
6.2.1 Definition and Basic Properties

As just dis ussed, for ea h real number y there is exa tly one positive real
number x su h that y = ln(x).
The names x and y, being mere symbols, an be inter hanged: For ea h
real number x there is exa tly one positive real number y su h that x = ln(y).
The fun tion that takes ea h real x to the orresponding positive real y is the
exponential fun tion. So earlier the logarithm was de ned as an integral,
and now the exponential fun tion is de ned as the inverse fun tion of the
logarithm: it undoes the logarithm, and the logarithm undoes it. All of this
takes us far from the idea of a fun tion as an analyti expression.
Definition 6.2.1 (Exponential Function).

The exponential function,

exp : R R>0 ,

is the inverse fun tion of the logarithm. That is, the fun tions exp and ln
are related by the following property:
For all x R and all y R>0 ,

y = exp(x) x = ln(y).

Sin e the exponential fun tion and the logarithm fun tion ex hange the
roles of x and y, the graph of the exponential fun tion is obtained by re e ting
the graph of the logarithm fun tion through the line y = x (exer ise 6.2.1).
Figure 6.2 shows (portions of) the graphs of the two fun tions.
Immediately in onsequen e of De nition 6.2.1, we have,
for all x R,
and

for all y R>0 ,

ln(exp(x)) = x,
exp(ln(y)) = y.

(6.2)
(6.3)

To establish (6.2), let x R and let y = exp(x). A ording to the de ning


property of the exponential fun tion, x = ln(y), i.e., x = ln(exp(x)) as desired.
That is, (6.2) follows from one dire tion a ross the double-headed arrow \"
in the de ning property of the exponential fun tion. Naturally, (6.3) follows
from the other (exer ise 6.2.2).
Theorem 6.2.2 (Properties of the Exponential Function).

(1) exp(0) = 1.

6.2 De nition and Properties of the Exponential Fun tion

195

y = exp(x)

PSfrag repla ements

Figure 6.2.

y = ln(x)

Graphs of the exponential and logarithm fun tions

(2) For all real numbers x and x~,


exp(x + x~) = exp(x) exp(x~).
(3) For all real numbers x
exp(x) = 1/ exp(x).
(4) For all real numbers x and all rational numbers ,
exp(x) = (exp(x)) .

Proof. These are all onsequen es of the orresponding properties of the logarithm. For example, to prove (2), let
y = exp(x),

Then
so that

x = ln(y),

y
~ = exp(x~).
x
~ = ln(y~ ),

x+x
~ = ln(y) + ln(y~ ) = ln(yy~ ),

and onsequently
exp(x + x~) = yy~ = exp(x) exp(x~).
The remainder of the proof is exer ise 6.2.3.

196

6 The Exponential Fun tion

Exercises
6.2.1. Explain why the geometri operation of re e ting a point p = (x, y)
in the plane through the 45-degree line y = x is the same as the algebrai
operation of ex hanging the point's x- and y- oordinates.
6.2.2. Show that (6.3) follows from the de ning property of the exponential

fun tion.

6.2.3. Prove the rest of Theorem 6.2.2.


6.2.2 Raising to Powers Revisited

Re all that so far we understand raising any positive real number b R>0
to any rational exponent Q. The exponential fun tion provides us a
me hanism to raise any b R>0 to any real exponent x R, and to reestablish the laws of exponents in this ontext. The idea is that for b R>0
and Q,
b = exp(ln(b )) = exp( ln(b)).
But in the display, the right side makes sense with no referen e to the fa t
that is rational. Thus the following de nition is natural.
Definition 6.2.3 (Raising a Positive Real Number to a Real Power).
Let b R>0 be any positive real number, and let x R be any real
number. Then bx is de ned to be
bx = exp(x ln(b)).

Again, this de nition of exponentiation agrees with our previous notion


of it when x is rational.
As a spe ial ase of the de nition, re all the number e su h that
ln(e) = 1.
By the de ning property of the exponential fun tion, we now know that this
hara terization of e rephrases as
e = exp(1).

So De nition 6.2.3 spe ializes to say that


ex = exp(x),

and this explains why the exponential fun tion is often written ex and alled
e to the x.
Returning from ex to bx for any positive real number b, our new notion
of exponentiation satis es the appropriate laws.

6.3 Exponential Growth

197

~ R>0
Proposition 6.2.4 (Laws of Real Exponents). Let b, b
positive real numbers. Let x, x~ R be any real numbers. Then

be any

(1) b0 = 1 and b1 = b.
(2) bx bx~ = bx+x~ .
(3) (bx )x~ = bxx~ .
(4) (bb~ )x = bx b~ x .

The only obsta le to proving the proposition is that its formulas are so
familiar. But on e one realizes that the idea is to use De nition 6.2.3, the
omputations are easy.

Proof. (1) is immediate sin e the exponents 0 and 1 are rational numbers.

We an also obfus ate matters and argue that b0 = exp(0 ln(b)) = exp(0) = 1
by Theorem 6.2.2, and b1 = exp(1 ln(b)) = exp(ln(b)) = b by (6.3), but this
argument is gratuitous. Similarly for (3), ompute that

(bx )x~ = exp x
~ ln(bx )


= exp x
~ ln exp(x ln(b))

~x ln(b))
= exp(x
~ ln(b))
= exp(xx
= bxx~ .

Parts (2) and (4) are exer ise 6.2.4.

Also, Theorem 5.1.4(4) (page 154) no longer requires a rational exponent:


Proposition 6.2.5 (Enhanced Property of the Logarithm).
positive real numbers b and all real numbers x,

ln(bx ) = x ln(b).
Exercises
6.2.4. Prove parts (2) and (4) of Proposition 6.2.4.
6.2.5. Prove Proposition 6.2.5.

6.3 Exponential Growth


The rst few terms of the sequen e
(sn ) =

n100000000
1.00000001n

For all

198

6 The Exponential Fun tion

are roughly, a ording to a omputer,


s1 = 0.999999,
s2 = 3.684665 1030102999 ,

s3 = 2.964601 1047712125 ,

s4 = 1.357676 1060205999 ,

s5 = 2.713950 1069897000 .

These are enormous. On the other hand, the jumps in the powers of 10|
from zero to 30 million to 48 million to 60 million to 70 million|seem to
be slowing down, suggesting that perhaps the sequen e is tending upward to
some huge-but- nite value. In fa t, limn (sn ) = 0. (!)
Theorem 6.3.1. Exponential

sense that

lim

growth dominates polynomial growth in the

xa
=0
bx

for any a > 0 and b > 1.

As on page 157, for any fun tion f : R>0 R, we de ne


lim f(x) = lim f(1/s).

s 0

That is, the left limit exists if the right limit does, in whi h ase it takes its
value from the right limit.

Proof. We have already established that


lim

ln x
x

= 0.

The result follows. For all large enough x we have


0<

i.e.,
i.e.,

ln x
x

<

ln b
,
a+1

0 < (a + 1) ln x < x ln b,
0 < xa+1 < bx ,

i.e.,
0<

1
xa
< .
x
b
x

And sin e limx 1/x = 0, we are done.

6.4 Di erentiation of the Exponential

199

6.4 Differentiation of the Exponential


The main result of this se tion is as follows.
Theorem 6.4.1 (The Exponential Function is Its Own Derivative).

The exponential fun tion is its own derivative,


exp = exp .

In onsequen e of the theorem, the exponential fun tion is ontinuous.


The theorem is lose to self-evident geometri ally in onsequen e of the
derivative of the logarithm being the re ipro al fun tion. For any real number x, let y = exp(x), a positive number. We have:
The tangent slope to the logarithm graph at (y, ln(y)) is 1/y.
Re e ting the logarithm graph through the y = x line gives the exponential
graph. And surely the tangent line of the re e ted graph at the re e ted point
is the re e tion of the tangent line to the original graph at the original point.
(For example, it should be easy to argue this using the geometri hara terization of the tangent line in exer ise 4.2.3 on page 136.) Re e ting the line
inter hanges the roles of rise and run, so that the slope of the re e ted line
is the re ipro al of the original slope. In sum:
The tangent slope to the exponential graph at (ln(y), y) is y.
Re all that y = exp(x), so that ln(y) = x. So:
The tangent slope to the exponential graph at (x, exp(x)) is exp(x).
That is, exp (x) = exp(x), and so this geometri argument strongly supports
the theorem and perhaps already proves it.
The Chain Rule also supports the theorem strongly. For all x R,
ln(exp(x)) = x,
and so taking derivatives gives
ln (exp(x)) exp (x) = 1,
or, sin e the derivative of the logarithm is the re ipro al,
exp (x)
= 1,
exp(x)
whi h is to say (again) that exp (x) = exp(x). However, the problem with this
argument is that it assumes that the exponential fun tion is di erentiable.

200

6 The Exponential Fun tion

The existen e of the derivative of the exponential fun tion is the subtle issue
here, not the value of the derivative on e its existen e is known. The justgiven Chain Rule argument ignored this point, while the pre eding geometri
argument handwaved it. Our only te hnique to show in a satisfa tory way
that that the derivative exists is to work analyti ally and al ulate it, and so
our proof of the theorem will pro eed by doing so.

Proof. (Proof that exp exists and equals exp.) As a preliminary matter, on-

sider a sequen e (sn ) that approa hes 0 in R. Ea h element of the sequen e


takes the form sn = ln(tn ) for some tn R>0 . We show that (tn ) approa hes 1. Let > 0 be given. Sin e our goal is to make values tn lie within
of 1, we may assume that < 1 by shrinking it if ne essary. Consider the negative number 1 = ln(1 ) and the positive number 2 = ln(1 + ). Sin e
(sn ) approa hes 0, there is a starting generation N su h that 1 < sn < 2
for all n N. Sin e tn = exp(sn ) for all n it follows that 1 < tn < 1 +
for all n N; that is, |tn 1| < for all n N, showing that limn (tn ) = 1.
Furthermore, tn 6= 1 for all n sin e sn 6= 0 for all n, and so (tn ) approa hes 1
as laimed. (See gure 6.3.)

y = exp(x)
1+

PSfrag repla ements

1
x
1

Figure 6.3.

Convergen e to 0 for es onvergen e to 1

Next we establish the normalized ase that exp (0) exists and equals 1,
that is,

6.4 Di erentiation of the Exponential

lim

exp(s) exp(0)
s0

s 0

201

= 1.

To establish this fun tion limit, onsider any sequen e (sn ) that approa hes 0
in R. The resulting sequen e


is also

exp(sn ) exp(0)

tn 1
ln(tn ) ln(1)

sn 0


n1

where (sn ) = (ln(tn )).

n1

As just shown, (tn ) approa hes 1 in R>0 . Consequently, the Re ipro al Rule
for sequen e limits says that the sequen e in the previous display has limit
1/ ln (1) = 1. This establishes the desired fun tion limit at the beginning of
this paragraph. That is, exp (0) exists and equals 1 as laimed.
For general x R and for any s 6= x, ompute that
exp(s) exp(x)
sx

= exp(x)

exp(s x) 1
sx

exp(~s) exp(0)
where s~ = s x.
= exp(x)
s~ 0

Let (sn ) be any sequen e in R that approa hes x. Then the sequen e (~sn ) =
(sn x) approa hes 0. Hen e




exp(sn ) exp(x)
exp(~sn ) exp(0)
lim
= exp(x) lim
n
n
sn x
s~n 0
= exp(x) exp (0)
= exp(x).

Sin e {sn } approa hes x arbitrarily in R, we have established the fun tion
limit
exp(s) exp(x)
= exp(x).
lim
s x

sx

That is, exp (x) exists and equals exp(x). This ompletes the proof.

Exercises
6.4.1. Let R be any real number, not ne essarily rational. De ne the
orresponding power fun tion
f : R>0 R,

f (x) = exp( ln(x)).

Observe that if Q is rational then this fun tion is the familiar rational
power fun tion f , other than the issue that its domain may now be smaller.

202

6 The Exponential Fun tion

Show that f is di erentiable and as before,


f = f1 .

(Write f as a omposition and use generative di erentiation rules.)


6.4.2. (a) For any positive real number b, de ne
gb (x) = bx = exp(x ln(b)).

gb : R R,

Thus ge is the exponential fun tion. Show that gb is di erentiable on R, and


gb = ln(b)gb .

That is, slightly abusing notation,


(bx ) = ln(b) bx .

(Write gb as a omposition.)
(b) Again let b be any positive real number. Part (a) says in parti ular
that gb (0) = ln(b), i.e.,
bs b0
= ln(b).
s 0 s 0

lim

(6.4)

On the other hand, Proposition 5.4.1 (page 165) says that


ln(b) = lim(n(b1/n 1)).
n

That is, the proposition says that




b1/n b0
lim
n
1/n 0

= ln(b).

(6.5)

Why does (6.5) only support (6.4), rather than fully prove it?
6.4.3. We know that the exponential fun tion satis es the onditions

exp = exp

and exp(0) = 1.

Suppose that some unknown fun tion f : R R also satis es the onditions
f = f

De ne
g : R R,

and f(0) = 1.
g(x) = exp(x)f(x).

Show that g = 0, i.e., g (x) = 0 for all x R. This fa t suggests powerfully


that that the fun tion g itself must be some onstant c. Granting this, nd c
by evaluating g(0). What does this say about f?

6.5 Integration of the Exponential

203

6.4.4. For ea h of the following fun tions, determine the fun tion's domain,
and then di erentiate the fun tion on its domain.
(a) f(x) = exp(x) ln(x).
(b) f(x) = exp(ln(x) + 1/x)
( ) f(x) = xa bx . (Here a R and b R>0 are onstants.)
6.4.5. The hyperboli osine and the hyperboli sine fun tions are

osh : R R,

osh(x) =

ex + ex
2

sinh : R R,

osh(x) =

ex ex
.
2

and

(Their pronun iations rhyme with gosh and grin h.)


(a) Show that osh = sinh and sinh = osh.
(b) Compute ( osh2 sinh2 ) , putting your answer in as simple a form as
you an. What does your answer suggest about osh2 sinh2 ?
( ) Sket h the graphs of osh and sinh on one set of oordinate axes.

6.5 Integration of the Exponential


Let b be a positive real number. Sin e the exponential fun tion is monotoni ,
its integral
from 0 to b exists. Now see gure 6.4. Its light-shaded region has
R
area 1exp(b) ln. The formula for the integral of the logarithm (see page 172)
says that
Z exp(b)
1

ln = b exp(b) + 1 exp(b).

And the entire shaded box in the gure has area b exp(b). It follows that the
integral of the exponential from 0 to b is
Zb
0

exp = b exp(b) (b exp(b) + 1 exp(b)) = exp(b) 1.

If the method just given to integrate the exponential fun tion seems too
easy, or too sneaky, then we an also ompute with a lower sum arising from
a uniform partition of [0, b] and use our usual bag of tri ks:

204

6 The Exponential Fun tion

eb

PSfrag repla ements

0
Figure 6.4.

The integral of the exponential via the integral of the logarithm

b
[exp(0) + exp(b/n) + exp(2b/n) + + exp((n 1)b/n)]
n

b
1 + exp(b/n) + (exp(b/n))2 + + (exp(b/n))n1
=
n
b (exp(b/n))n 1
=
n
exp(b/n) 1
. exp(b/n) exp(0)
.
= (exp(b) 1)
b/n 0

Sn =

And thus

lim(Sn ) = (exp(b) 1)/ exp (0) = exp(b) 1.


n

The orresponding upper sums Tn satisfy


Tn Sn =

b
[exp(b) 1] ,
n

and so limn (Tn Sn ) = 0. Therefore, again,


Zb

exp = exp(b) 1.

And now the method of the previous paragraph an be used instead to rederive
the integral of the logarithm with no work. Sin e the logarithm took some
e ort to integrate, there is a real gain of e ien y here. But integrating the
logarithm this way would have deferred our learning the answer until now.
For the more general integral of the exponential fun tion,

6.5 Integration of the Exponential


Zb
a

exp,

205

a, b R,

rst assume that a < b and translate the boxes over [0, b a] by a. Note that
exp(x + a) = exp(a) exp(x), and very qui kly it follows that (exer ise 6.5.1)
Zb
a

exp = exp(b) exp(a),

a b.

(6.6)

And then, nally, if instead a > b then


Zb
a

exp =

Za
b

exp = (exp(a) exp(b)) = exp(b) exp(a).

That is,
Theorem 6.5.1. The
Zb

integral of the exponential fun tion is


exp = exp(b) exp(a),

a, b R.

Naturally this is another instan e of the Fundamental Theorem of Cal ulus. Let F = exp, so that F = exp. Then the formula says that
Zb

exp = F(b) F(a),

a, b R.

Exercises
en
6.5.1. Let a and b be real numbers with a b. Explain why lower sums S
b
for Ara (exp) satisfy
en = ea Sn ,
S

where ea h Sn is in turn a lower sum for Ar0ba (exp), and similarly for upper
sums. Show that formula (6.6) follows.
6.5.2. Re over formula (2.11),
Zb
a

f =

b+1 a+1
,
+1

6= 1, 0 < a b

where now is a real number rather than ne essarily a rational number. (See
exer ise 6.4.1 on page 201 for the new de nition of f .)

206

6 The Exponential Fun tion

6.6 The Exponential as a Limit of Powers


6.6.1 The Description
Theorem 6.6.1 (The Exponential Function as a Limit of Powers).
For any real number x,

x n 
.
exp(x) = lim 1 +
n
n

Before the proof, it deserves noti e that the limit in the theorem is subtle.
Given a xed real number x, the question is to what value the quantity


1+

x n
n

tends as n grows large. For that matter, does su h a value even exist?
One argument pro eeds as follows: Sin e x is xed, 1 + x/n tends to 1

as n grows large, and so (1 + x/n)n behaves like


all n, the limiting value of (1 + x/n)n must be 1.

1n .

Sin e

1n = 1

for

A se ond argument rea hes a drasti ally di erent on lusion: Assume that

x is positive.
(1 + x/n)n is

Sin e 1 + x/n is greater than 1 for all n, it follows that


a high power of a quantity greater than 1. But the high
powers of any quantity greater than 1 grow very large, and so the limiting
value of (1 + x/n)n must be in nite. Similarly, if x is negative then for
large n, 1+x/n lies between 0 and 1, and so the limiting value of (1+x/n)n
must be 0. Thus the limit is in nite for positive x and 0 for negative x.
And of ourse it is 1 for x = 0.

These arguments, and their on lusions, are in orre t. The problem with
both of them is that to study the limiting behavior of (1 + x/n)n as n grows
large, we must take are that both o urren es of n in (1 + x/n)n grow
large together. The rst argument let the denominator-n grow rst and only
thereafter let the exponent-n grow as well. The se ond argument made the
omplementary error. The falsity of the resulting on lusions illustrates yet
again that to obtain the orre t medium-sized answers in al ulus, we need
to manipulate small and large quantities arefully.

Proof. During the ourse of di erentiating the logarithm, we established the


bounds

s1
ln(s) s 1,
s

s R>0 .

Repla e s by 1 + s to get
s
ln(1 + s) s,
1+s

s > 1.

6.6 The Exponential as a Limit of Powers

207

Let x R be any real number, let n Z1 be any positive integer su h


that n > |x|, and repla e s by x/n to get

x
x
x/n
,
ln 1 +
1 + x/n
n
n

n > |x|.

By a property of the logarithm, n ln(1 + x/n) = ln((1 + x/n)n ), and so multiplying the inequalities through by n gives

x n 
x
x,
ln 1 +
1 + x/n
n

n > |x|.

Be ause the exponential fun tion is stri tly in reasing, the inequalities are
preserved upon passing the quantities through it,
exp

x
1 + x/n


x n
1+
exp(x),
n

n > |x|.

(6.7)

By various sequen e limit results (exer ise 6.6.1),


lim
n

x
1 + x/n

= x,

and so, sin e the exponential fun tion is ontinuous,




lim exp
n

x
1 + x/n



= exp(x),

To omplete the argument, apply the Squeezing Rule for sequen e limits
to (6.7), showing that indeed
lim
n


x n 
1+
exists and equals exp(x) for all x R.
n

Exercise
6.6.1. Show that for any real number x,


x
lim
= x.
n
1 + x/n
6.6.2 An Interpretation: Compound Interest

The bank promises you an annual interest rate of x. For example, x ould be
some value su h as x = 0.05, i.e., ve per ent, but to avoid being overspe i
we view x as a generi positive real number.

208

6 The Exponential Fun tion

You make a deposit d.


A year later the bank informs you that you now have your original deposit
plus the interest on your deposit, said interest amounting to your deposit
multiplied by the annual interest rate. That is, the bank tells you that the
amount in your a ount is now
d(1 + x).

This is not fair to you. The bank has been free to invest your deposit from
the moment that you made it, and then further to invest any pro ts from the
initial investment, and so on; but the bank has ompounded interest for you
only at the year's end, despite earning money with your deposit all through
the year.
It would be more fair for the bank to ompound your interest for the rst
half of the year halfway through the year, at half of the annual interest rate,
and then to ompound your interest for the se ond half of the year at the end
of the year, again at half of the annual interest rate. Thus at six months you
would have your deposit plus your deposit times half the annual interest rate,
d(1 + x/2),

and at the end of the year you would have your amount at six months plus
your amount at six months times half the annual interest rate,
d(1 + x/2)2 .

But this is not fair to you either. The bank has made money with your deposit
ontinuously, but still it has ompounded your interest only after six months
and then again only after another six months. On the brighter side, the endyear amount now satis es
d(1 + x/2)2 = d(1 + x + x2 /4) > d(1 + x),

i.e., the situation is more fair to you now than it was after a single interest
payment at the end of the year.
Similarly, if the bank ompounds interest monthly then at the end of the
year you have
d(1 + x/12)12 .

Before we ontinue to analyze the situation, here is a small observation: For


any positive quantity s and any integer n > 1, expanding the n-fold produ t
(1 + s)n = (1 + s)(1 + s) (1 + s)

gives the term 1 (from multiplying all of the 1's together), gives the term ns
(from the n di erent ways of multiplying one s and n 1 1's together), and
gives other terms, all of whi h are positive. Thus

6.6 The Exponential as a Limit of Powers


(1 + s)n > 1 + ns

209

for s > 0 and n > 1.

Returning to the ompound interest al ulation we have, in onsequen e of


the observation just made,
(1 + x/12)6 > 1 + 6x/12 = 1 + x/2,

and so

d(1 + x/12)12 = d (1 + x/12)6

2

> d(1 + x/2)2 .

That is, you have more at the end of the year if the bank ompounds interest
monthly than if the bank ompounds interest only twi e a year, just as you
have more if the bank ompounds interest twi e a year than if the bank
ompounds interest only on e a year.
If the bank ompounds interest daily then at the end of the year your
balan e is
d(1 + x/365)365 .

Presumably this is more than you have if the bank ompounds interest
monthly, but the relevant algebra to justify this fa t is a bit hairier than
the two omparisons that we we have made so far. (The problem is that 365
is not an integer multiple of 12, whereas 12 is an integer multiple of 2, and 2
is an integer multiple of 1.) Instead, note that if the bank ompounds interest
12 times daily then sin e 12 365 = 4380, at the end of the year you have
d(1 + x/4380)4380 .

Furthermore, applying our small observation twi e gives


(1 + x/4380)12 > 1 + x/365

and (1 + x/4380)365 > 1 + x/12,

so that
(1 + x/4380)4380 > (1 + x/365)365

and (1 + x/4380)4380 > (1 + x/12)12 .

In other words, if the bank ompounds interest 12 times daily then at the end
of the year then you have more than if the bank ompounds interest daily
or monthly. (Unsurprisingly, the daily ompounding does yield more interest
than the monthly, and this is easy to show using al ulus rather than algebra.
See exer ise 6.6.2.)
Even if the bank ompounds your interest 4380 times, this still isn't quite
fair to you be ause the bank is making money ontinuously with your deposit.
The orre t s enario is for the bank to ompound your interest ontinuously
as well. Thus we are led to the exponential fun tion: at the end of the year,
your a ount value should be

210

6 The Exponential Fun tion


d lim((1 + x/n)n ) = dex .
n

Interpret an interest rate as a proportion fa tor relating the rate of in rease


of a quantity to the amount of the quantity. That is, viewing time t as an
independent variable, the a ruing value of your a ount is an unknown timedependent fun tion f(t) su h that
(initial ondition),
f (t) = xf(t) (di erential equation).
f(0) = d

Similar to exer ise 6.4.3, the only fun tion satisfying these onditions is
f(t) = dext .

Thus we re over our result that the exponential fun tion arises naturally
from ontinuously ompounded interest. And more generally, it des ribes any
quantity that in reases in proportion to its amount.
Exercises
6.6.2. Re all that in the ourse of proving that the exponential is a limit of

powers, we made use of the inequality

s
ln(1 + s),
1+s

s > 1.

Sin e the inequality was derived by omparing a box-area to an area under a


urve, the inequality is stri t (i.e., it is \<") unless s = 0.
(a) Let x R>0 be a xed positive real number. De ne a fun tion of a
variable t that also makes referen e to the onstant x,
Show that

f : R1 R,
f (t) = ln(1 + s)

f(t) = ln((1 + x/t)t ).


s
,
1+s

x
t

where s = .

(b) Part (a) shows that f is always positive by the inequality at the
beginning of this exer ise, sin e s = x/t is positive. The fa t that f is always
positive suggests powerfully that f is stri tly in reasing. Granting this, explain
why the fun tion
g : R1 R,

g(x) = exp(f(x))

is stri tly in reasing as well.


( ) In parti ular,

g(1) < g(2) < g(3) < .

What does this say about your a ount balan e at the end of the year if the
bank ompounds interest daily rather than monthly?

6.6 The Exponential as a Limit of Powers

211

6.6.3. Let d and x be positive real numbers. What is the doubling time of
the quantity
f : R0 ,

f(t) = dext ,

i.e., the t-value su h that f(t) = 2d? How does the doubling time depend on
the initial value d? If instead x is negative, what is the halving time of f?

7
The Cosine and Sine Functions

The basi trigonometri fun tions osine and sine des ribe uniform os illation,
analogously to how the exponential fun tion des ribes natural growth. Sin e
os illation is a more ompli ated phenomenon than natural growth, the properties of the osine and sine are orrespondingly more elaborate than those of
the exponential. After establishing the properties of osine and sine, we arry
out our usual program of di erentiating and integrating these fun tions.
Se tion 7.1 establishes the fa t that the ir umferen e of the unit ir le is
its diameter times its area, a fa t relevant to the properties of osine and the
sine. Se tion 7.2 de nes the osine and the sine, and se tion 7.3 establishes
some of their properties: basi identities, angle sum and di eren e formulas,
double and half angle formulas, produ t and di eren e formulas. Using these
properties, we an di erentiate the osine and the sine in se tion 7.4, and we
an integrate them in se tion 7.5. Se tion 7.6 introdu es other trigonometri
fun tions, the tangent, otangent, se ant, and ose ant, and se tion 7.7 introdu es their inverse fun tions. All of these fun tions are within our power
to di erentiate by using generative derivative rules, but for the most part we
don't yet know how to integrate them.

7.1 The Circumference of the Unit Circle


Let denote the area of the unit ir le, and let c denote its ir umferen e.
We qui kly review the proof that c = 2. Figure 7.1 shows n triangles ir ums ribing the unit ir le (n = 8 in the gure). One triangle is shaded, and
the altitude from its vertex at the enter of the ir le to the enter of its base
is shown. Sin e the altitude is a ir le-radius, ea h triangle has height 1. Also,
the triangle-bases have total length slightly larger than the ir umferen e c.
Meanwhile, gure 7.2 shows the same number of triangles, all having height 1

214

7 The Cosine and Sine Fun tions

and equal bases, the ommon base of the ir ums ribing triangles; thus the
area of the triangle in in gure 7.2 equals the sum of the areas of the ir ums ribing triangles in gure 7.1. As n grows, the area of the ir ums ribing
triangles tends to the area of the unit ir le, while the base of the triangle
in gure 7.2 tends to the ir umferen e c and its height is always 1. Thus in
the limit, the equal areas are and c/2, from whi h
c = 2.

Figure 7.1.

Figure 7.2.

Triangles ir ums ribing the unit ir le

Same area as the ir ums ribing triangles

7.2 Definition of the Cosine and the Sine


Let s R0 be a nonnegative real number. Starting at (1, 0), pro eed ounter lo kwise along the unit ir le through ar length s, and let P(s) denote
the point thus rea hed. The x- and y- oordinates are respe tively the cosine
and the sine of s,
P(s) = ( os(s), sin(s)).
(7.1)
For s R<0 , i.e., for negative s, pro eeding ounter lo kwise along the unit
ir le through ar length s has the obvious interpretation of pro eeding lo kwise instead through ar length s. Again let P(s) denote the point thus

7.3 Identities for the Cosine and the Sine

215

rea hed, and extend formula (7.1) to this ase as well. Thus we have de ned
fun tions
os, sin : R [1, 1].

So, for example, starting at (1, 0) and then going one-quarter, one-half, threequarters, and all the way around the unit ir le ounter lo kwise gives the
values
( os(0), sin(0)) = (1, 0),
( os(/2), sin(/2)) = (0, 1),
( os(), sin()) = (1, 0),
( os(3/2), sin(3/2)) = (0, 1),
( os(2), sin(2)) = (1, 0).

The graphs of the osine and sine fun tions (or rather, portions of the graphs)
are shown in gure 7.3.
PSfrag repla ements

os(s)

sin(s)

Figure 7.3.

Cosine and sine

7.3 Identities for the Cosine and the Sine


7.3.1 Basic Identities

The following properties of sine and osine are onsequen es of the de nition.
For all s R,

216

7 The Cosine and Sine Fun tions

sin(s)2 + os(s)2 = 1,
os(s + 2n) = os(s) for all n Z,
sin(s + 2n) = sin(s) for all n Z,
os(s) = os(s),
sin(s) = sin(s),
os(/2 s) = sin(s),
sin(/2 s) = os(s).

(7.2)
(7.3)
(7.4)
(7.5)
(7.6)
(7.7)
(7.8)

Property (7.2) holds by the Pythagorean Theorem be ause P(s) lies on the
unit ir le. Properties (7.3) and (7.4) hold be ause the ir le has ir umferen e 2. Properties (7.5) and (7.6) hold be ause pro eeding from (1, 0)
lo kwise rather than ounter lo kwise but through the same ar length s
gives the same x- oordinate but the opposite y- oordinate. Properties (7.7)
and (7.8) hold be ause starting at (0, 1) (rather than at (1, 0)) and pro eeding
lo kwise through ar length s along the unit ir le is the re e tion through
the 45-degree line y = x of starting at (1, 0) and pro eeding ounter lo kwise
through ar length s. Thus the x- oordinate produ ed by the rst pro ess,
os(/2 s), must equal the y- oordinate produ ed by the se ond, sin(s),
and similarly sin(/2 s) = os(s).
The reader should visually identify as many as possible of the basi properties of osine and sine in gure 7.3. Also, anti ipating a result to ome, the
reader should see that plausibly the tangent slope of the sine graph is the
height of the osine graph, and the tangent slope of the osine graph is minus
the height of the sine graph. That is, the graphs suggest (as we will show
arefully soon) that sin = os and os = sin.
Exercise
7.3.1. Consider the fun tions f, g, h : [0, /2] R where
f(s) = (sin(4s))5 ,

g(s) = (sin(3s))5 ,

h(s) = ( os(3s))5 ,

and onsider the quantities


A=

Z /2
0

f,

B=

Z /2
0

g,

C=

Z /2

h.

Arrange A, B, and C in in reasing order. The idea here is not to ompute


pre isely but to reason, with explanation, from rough sket hes (whi h you
should show) of the graphs of f, g, and h.

7.3 Identities for the Cosine and the Sine

217

P(t)
P(s t)

P(0)

PSfrag repla ements

P(s)

Figure 7.4.

Geometry for the angle sum and di eren e formulas

7.3.2 Angle Sum and Difference Formulas

Let s, t R be any real numbers. As far as their osines and sines are
on erned, we may assume that in fa t s, t [0, 2), and for the moment
we further assume that s > t. Figure 7.4 shows the points P(0) = (1, 0),
P(t) = ( os(t), sin(t)), P(s) = ( os(s), sin(s)), P(st) = ( os(st), sin(st)).
The ar length distan e s t from P(0) to P(s t) equals the ar length distan e from P(t) to P(s), so the orresponding linear distan es are equal as well,
and hen e so are their squares. By the Pythagorean Theorem, the equality of
squares of distan es is
( os(s t) 1)2 + sin2 (s t) = ( os(s) os(t))2 + (sin(s) sin(t))2 ,

or, after a little algebra that makes use of the rst basi identity (7.2),
os(s t) = os(s) os(t) + sin(s) sin(t).
Sin e this formula is una e ted by ex hanging s and t, our initial assumption
that s > t (after both are translated to lie in [0, 2)) is unne essary, and
the formula holds for all s, t R. Altogether the angle sum and di eren e
formulas are

218

7 The Cosine and Sine Fun tions

os(s + t) = os(s) os(t) sin(s) sin(t),


os(s t) = os(s) os(t) + sin(s) sin(t),
sin(s + t) = sin(s) os(t) + os(s) sin(t),
sin(s t) = sin(s) os(t) os(s) sin(t).

(7.9)
(7.10)
(7.11)
(7.12)

We have established (7.10). The others follow from substitutions in (7.10) and
the basi identities. The slight elaborateness of these formulas orresponds to
the slightly ompli ated way that the os illating x- and y- oordinates of a
point moving around the ir le are related to ea h other.
Exercise
7.3.2. Prove the other three angle sum and di eren e formulas.
7.3.3 Double and Half Angle Formulas

For all s R,
os(2s) = os2 (s) sin2 (s) = 2 os2 (s) 1 = 1 2 sin2 (s),
sin(2s) = 2 sin(s) os(s),
os2 (s/2) = 12 (1 + os(s)),
sin2 (s/2) = 12 (1 os(s)).

(7.13)
(7.14)
(7.15)
(7.16)

Here (7.13) and (7.14) follow from (7.9) and the rst basi identity (7.2)
and (7.11). Then (7.15) and (7.16) follow from (7.13).
Exercise
7.3.3. Prove the double and half angle formulas.
7.3.4 Product Formulas

For all s, t R,
os(s) os(t) = 21 ( os(s t) + os(s + t)),
sin(s) sin(t) = 21 ( os(s t) os(s + t)),
os(s) sin(t) = 12 (sin(s + t) sin(s t)).

(7.17)
(7.18)
(7.19)

Here (7.17) follows from adding (7.9) and (7.10), and (7.18) follows from
subtra ting (7.9) from (7.10), and (7.19) follows from subtra ting (7.12)
from (7.11). The produ t formulas will let us integrate the osine and the
sine.

7.4 Di erentiation of the Cosine and the Sine

219

Exercise
7.3.4. Prove the produ t formulas.
7.3.5 Difference Formulas

For all s, t R,

os(s) os(t) = 2 sin s+t


,
sin st
2
2


s+t
st
sin(s) sin(t) = 2 os 2 sin 2 .

(7.20)
(7.21)

st
Here (7.20) follows from substituting s+t
2 for s and 2 for t in (7.18). And
(7.21) follows from the same substitutions in (7.19). The di eren e formulas
will let us di erentiate the osine and the sine.

Exercises
7.3.5. Prove the di eren e formulas.
7.3.6. Show that for all x R,

os(3x) = 4 os3 (x) 3 os(x).


7.3.7. Complete the table of sines and osines whose rst half is
0 /6 /4 /3 /2 2/3 3/4 5/6

sin
os
and whose se ond half is
7/6 5/4 4/3 3/2 5/3 7/4 11/6 2

sin
os
In lude an explanation for how you found sin(/6) and os(/6) (or sin(/3)
and os(/3)). No explanation is required for the remaining values.

7.4 Differentiation of the Cosine and the Sine


The osine and sine fun tions os illate so regularly that one normalized derivative al ulation | that sin (0) = 1 | qui kly gives the derivatives of osine
and of sine everywhere. To di erentiate the sine at 0, the rst step is to
establish some estimates.

220

7 The Cosine and Sine Fun tions

Lemma 7.4.1. For

all real numbers

s R,

| sin(s)| |s|.

Also, for all numbers s [0, /2),

s sin(s)/ os(s).

Proof. For the rst statement, sin e sin(s)

= sin(s), we may assume


that s 0. If 0 s /2 (see gure 7.5) then sin(s) is the distan e
from ( os(s), 0) to ( os(s), sin(s)), whi h is less than the distan e from (1, 0)
to ( os(s), sin(s)), and this is in turn less than the ar length from (1, 0)
to ( os(s), sin(s)), whi h is s. And if s > /2 then | sin(s)| 1 < /2 < s.
For the se ond statement, again see gure 7.5. The right triangle in the
gure with legs sin(s) and os(s) is similar to the right triangle in the gure
with legs d and 1, so that
d = sin(s)/ os(s).

The gure suggests strongly that also d s. We invoke this as an assumption.


(For Ar himedes, it was a parti ular onsequen e of a more general assumption
about urves that are on ave in the same dire tion.) Thus
s sin(s)/ os(s).

This is the desired result.

PSfrag repla ements

sin(s)
s

os(s)
Figure 7.5.

(1, 0)

Analysis of sine and osine

The next proposition is self-evident geometri ally. It says that a moving


point on the ir le tends to a parti lar xed point, its oordinates tend to

7.4 Di erentiation of the Cosine and the Sine

221

the oordinates of the xed point. We give an analyti proof to redu e our
dependen e on geometry by using the trigonometry identities..
Proposition 7.4.2. The
for any t R,

osine and sine are ontinuous, meaning that

lim os(s) = os(t) and

s t

lim sin(s) = sin(t).

s t

Proof. For all s, t R, the di eren e formula (7.20) is


os(s) os(t) = 2 sin

s+t
2

sin

st
2

Therefore, sin e the absolute value of a produ t is the produ t of the absolute
values,


| | sin st
|.
| os(s) os(t)| = 2 | sin s+t
2
2

But be ause sin


1, we have


| | sin st
| 2 | sin
2 | sin s+t
2
2

s+t
2

st
2

and by the rst statement of the previous lemma,


2 | sin

st
2

So we have established that

|,

| 2| st
2 | = |s t|.

0 | os(s) os(t)| |s t|,

and so by the Squeeze Rule for fun tion limits,


lim | os(s) os(t)| = 0.

s t

It follows, as explained in display (4.1) on page 127, that


lim os(s) = os(t).

s t

The proof for the sine is virtually identi al (exer ise 7.4.1).
Proposition 7.4.3. The

sine is di erentiable at 0, and

sin (0) = 1.

Proof. Sin e sin(s)/(s) = sin(s)/s for nonzero s R, we may onsider


the fun tion
sin(s)
g : (0, /2) R,

g(s) =

and show that lims 0 g(s) = 1. Using both statements of the lemma,

222

7 The Cosine and Sine Fun tions

os(s)

sin(s)
s

for all s (0, ).

Sin e lims 0 os(s) = os(0) = 1 by the previous proposition, the Squeeze


Rule for fun tion limits shows that
lim

sin(s)
s

s 0

That is,
lim

= 1.

sin(s) sin(0)

= 1,

s0

s 0

whi h is to say that sin (0) exists and equals 1.

As mentioned, the general derivatives of osine and sine are onsequen es


of their regular os illatory nature and of the fa t that sin (0) = 1.
Theorem 7.4.4 (Derivatives of Cosine and Sine).
sine fun tions are di erentiable on R, and

os = sin,

The osine and the

sin = os .

Proof. For any s and t, we have by (7.20),


os(s) os(t)
st

2 sin

s+t
2

Now let s tend to t. Then also,


tends to t, so that sin

and
st
2

tends to 0, so that

That is,
lim

s t

sin

st

s+t
2

st
2
st
2

os(s) os(t)

sin

st

s+t sin

= sin

s+t
2

st
2

st
2
st
2

tends to sin(t),

tends to sin (0) = 1.

= sin(t).

In other words, os = sin. The argument that sin = os is virtually iden

ti al (exer ise 7.4.2)..

7.5 Integration of the Cosine and the Sine

223

Exercises
7.4.1. Show that the sine is ontinuous.
7.4.2. Carry out the argument that sin = os.
7.4.3. Find the derivatives of the following fun tions.
(a) f(x) = ln( os(x) + 2).
(b) f(x) = sin(4(x3 + 2)).
( ) f(x) = ln((sin(x) + 1)/ os(x)) for x (/2, /2).

7.5 Integration of the Cosine and the Sine


The osine and sine fun tions are both bounded, sin e their values lie
in [1, 1]. Also, the osine fun tion is
in reasing on [, 0] and de reasing on [0, ],
de reasing on [2, ] and in reasing on [, 2],
in reasing on [3, 2] and de reasing on [2, 3],
and so on, while the sine fun tion is
in reasing on [/2, /2],
de reasing on [3/2, /2] and [/2, 3/2],
in reasing on [5/2, 3/2] and [3/2, 5/2],
and so on. All of this is self-evident if one remembers the interpretation of
the osine and the sine as the oordinates of a point that moves around the
ir le.
Theorem 7.5.1 (Integrals ofRCosine and
Rb Sine).
b

numbers. Then the integrals

os and
a

Zb

os = sin(b) sin(a)

Zb

sin = os(a) os(b).

and

Let a and b be any real

sin exist, and their values are


a

224

7 The Cosine and Sine Fun tions

Proof. As usual, if we an prove the results for a b then they follow as well
for a > b. So we assume that a b, and in fa t we assume that a < b sin e
the ase a = b is trivial.
The sine and the osine are bounded and pie ewise monotoni . Any
bounded monotoni fun tion is integrable by Theorem 3.3.8 on page 112 (for
nonnegative su h fun tions) and the dis ussion in hapter 5 (extending the
integral to bounded fun tions that an also take negative values). The sine
and the osine are therefore integrable by Proposition 3.3.9 on page 113. Also
by the proposition, we may arry out the integral over an interval [a, b] where
the fun tion (sine or osine) is monotoni .
We integrate the sine, leaving the osine as a very similar exer ise. Let
n Z1 be a positive integer. To set up a uniform partition, let
n =

ba
,
n

so that a + nn = b. Let


Sn = n sin(a) + sin(a + n ) + sin(a + 2n ) + + sin(b n )

and let



Tn = n sin(a + n ) + sin(a + 2n ) + sin(a + 3n ) + + sin(b)

Then Sn is a lower sum and Tn is an upper sum, or vi e versa. And


|Tn Sn | = n | sin(b) sin(a)|,

so that

lim(Tn Sn ) = 0.
n

Rb

Thus a sin exists, and limn (Sn ) = limn (Tn ) =


For ea h n Z1 , let
Un = n

Rb
a

sin.







n
3n
5n
+ sin a +
+ sin a +
+
sin a +
2
2
2


(2n 1)n
+ sin a +
.
2

Then Un lies between Sn and Tn , and so also


lim(Un ) =
n

Zb
a

sin .

As written, Un is a small positive number (n ) times a sum of many numbers.


Guided by hindsight, divide the n by a fa tor of omparable magnitude, and
multiply the summands by the same fa tor,

7.5 Integration of the Cosine and the Sine

225


 

n

n /2
Un =
sin n
2 sin a +
sin(n /2)
2
2


 
3n

+ 2 sin a +
sin n
2
2

 

5n
n
+ 2 sin a +
sin
2
2




 
(2n 1)n

sin n .
+ 2 sin a +
2
2

The summands inside the square bra kets are



 
(2i 1)n

2 sin a +
sin n ,
2
2

i = 1, , n.

By the produ t formula (7.18), we have for ea h i,




(2i 1)n
2 sin a +
2

sin n
2

= os(a + (i 1)n ) os(a + in ).

This is the di eren e of the osine fun tion at two input-values distan e n
apart. And so, the sum in square bra kets ollapses down to only two terms,
Un =

n /2 
os(a) os(a + x )
sin(n /2)

+ os(a + x ) os(a + 2x )
+ os(a + 2x )

os(b n )


+ os(b n ) os(b) .

That is,
Un =

But
lim
n

and so we are done,


Zb
a


n /2 
os(a) os(b) .
sin(n /2)

sin(n /2)
n /2

= sin (0) = 1,

sin = lim(Un ) = os(a) os(b).


n

Exercise
7.5.1. Integrate the osine fun tion.

226

7 The Cosine and Sine Fun tions

7.6 Other Trigonometric Functions


Definition 7.6.1 (Tangent, Cotangent, Secant, Cosecant). The tangent, cotangent, secant, and cosecant fun tions have the formulas

tan(s) = sin(s)/ os(s),


ot(s) = os(s)/ sin(s),
se (s) = 1/ os(s),
s (s) = 1/ sin(s).

The domains of these fun tions are the largest subsets of


dividing by 0. For example, the domain of the tangent is

that avoid

dom(tan) = {s R : s 6= /2, 3/2, 5/2, }.


Example 7.6.2. We di erentiate the tangent. Any point s dom(tan) is also
approa hable from dom(tan), and


sin
(s)
tan (s) =
by de nition of the tangent
os


sin os sin os
=
(s) by the Quotient Rule
os2
!
os2 + sin2
(s)
=
sin e sin = os and os = sin
os2

sin e os2 + sin2 = 1

os2 (s)
= se 2 (s).

That is,
tan = se 2 .
Exer ise 7.6.1 is to show that also
ot = s 2
and
and

se = tan se
s = ot s .

7.7 Inverse Trigonometri Fun tions

227

Exercises
7.6.1. What is the domain of the otangent? The se ant? The ose ant? Show

that

ot = s 2 ,

se = tan se ,

s = ot s .

7.6.2. (a) Des ribe the domain of the fun tion given by the formula
f(x) = ln(| tan(x) + se (x)|).

Explain why f is di erentiable and ompute f . The answer should simplify


ni ely.
(b) Similarly for
f(x) = ln(| ot(x) + s (x)|).

7.7 Inverse Trigonometric Functions


Example 7.7.1 (The Inverse Cosine Function and Its Derivative).
Restri t the domain of the osine to [0, ]. The resulting fun tion

os : [0, ] [1, 1]
takes ea h value in its odomain exa tly on e, so it is invertible. The inverse
fun tion is the arc-cosine,
ar os : [1, 1] [0, ].
A graph of the ar - osine is shown in gure 7.6.
Be ause we an integrate the osine, we an integrate the ar - osine by
using the same tri k that let us integrate the exponential via the integral of
the logarithm (or onversely) in se tion 6.5. Consider any value bR [1, 1],
b
let c = ar os(b), and onsider
gure 7.7. The dark-shaded area is 1
ar os.
R
The light-shaded area is c (b os). The entire shaded area is (b + 1).
Therefore,
Zb

ar os = (b + 1)

(b os)
Z
os
= (b + 1) ( c)b +
c

= + bc + sin() sin(c)
= + bc sin(c).

228

7 The Cosine and Sine Fun tions


s

/2

PSfrag repla ements

x
1
Figure 7.6.

Graph of the ar - osine fun tion

But c = ar os(b) and sin(ar os(b)) = 1 b2 (the general fa t that


sin(ar os(x)) =

p
1 x2 ,

x [1, 1]

is demonstrated by gure 7.8), and so


Zb

ar os = + b ar os(b)

p
1 b2 .

It follows that more generally, for all a, b [1, 1],


Zb
a

ar os = (b ar os(b)

p
p
1 b2 ) (a ar os(a) 1 a2 ).

We an also di erentiate the ar - osine. Note that

os(ar os(x)) = x, x [1, 1].


The Chain Rule says that onsequently, for all x [1, 1] su h that the
derivative ar os (x) exists,
sin(ar os(x)) ar os (x) = 1.

7.7 Inverse Trigonometri Fun tions

229

PSfrag repla ements


(b, c)

x
1

Figure 7.7.

PSfrag repla ements

Integrating the ar - osine

y
s
x
Figure 7.8.

Figure to ompute sin(ar os(x))

So for all x [1, 1] su h that the derivative ar os (x) exists and furthermore
sin(ar os(x)) is nonzero,
ar os (x) =
But as already noted,

sin(ar os(x))

230

7 The Cosine and Sine Fun tions

sin(ar os(x)) =

p
1 x2 ,

x [1, 1].

Thus sin(ar os(x)) 6= 0 for all x (1, 1). Only the endpoints x = 1 need
to be ex luded to avoid the divide by zero. But we still have to address the
question of whether ar os (x) exists for x (1, 1).
For any x (1, 1), if s tends to x then also t = ar os(s) tends
to ar os(x), and now we an reason similarly to di erentiating the exponential fun tion,
lim

ar os(s) ar os(x)
sx

s x

t ar os(x)
t ar os(x) os(t) os(x)

lim

= 1/ os (ar os(x))
= 1/ sin(ar os(x))
p
= 1/ 1 x2 .

In sum,
ar os (x) =

1 x2

1 < x < 1.

With the derivative of the ar - osine in hand, we an see the Fundamental


Theorem of Cal ulus in a tion on e again. De ne a fun tion
F : [1, 1] R,

F(x) = x ar os(x)

Then (exer ise 7.7.1)


F (x) = ar os(x),

p
1 x2 .

1 < x < 1.

And so, as usual, for all a, b (1, 1), glossing over the issue of the endpoints
for now,
Z
b

ar os = F(b) F(a),

F = ar os .

Similarly to the inverse osine fun tion, we an de ne the inverse sine,


ar sin : [1, 1] [/2, /2],
the inverse tangent,

ar tan : R (/2, /2),

and the inverse otangent,

ar ot : R (0, ).
Exer ises 7.7.2 and 7.7.3 are to dis uss these fun tions, to al ulate their
derivatives, and to integrate the ar -sine.

7.7 Inverse Trigonometri Fun tions

231

Exercises

7.7.1. As in the se tion, let F(x) = x ar os(x) 1 x2 for x [1, 1]. Show
that F = ar os on (1, 1).
7.7.2. Restri t the domain of the sine to [/2, /2]. The resulting fun tion

sin : [/2, /2] [1, 1]

takes ea h value in its odomain exa tly on e, so it is invertible. The inverse


fun tion is the arc-sine,
ar sin : [1, 1] [/2, /2].

(a) Sket h a graph of the ar -sine.


(b) Explain why for all y [1, 1] su h that the derivative ar sin (y)
exists and os(ar sin(y)) is nonzero,
1

ar sin (y) =

os(ar sin(y))

( ) For generi y [1, 1], draw a right triangle with verti es (0, 0), (0, y),
and (x, y) where the third vertex lies on the right half of the unit ir le.
Explain how this gure shows that for su h y,
os(ar sin(y)) =
(d) Show that
ar sin (y) = p

1 y2

(e) Compute

Rb
a

p
1 y2 .

1 < y < 1.

ar sin for any a, b [1, 1].

7.7.3. (a) Dis uss the inverse tangent fun tion

and show that

ar tan : R (/2, /2)


ar tan (x) =

1
,
1 + x2

x R.

(b) Dis uss the inverse otangent fun tion


and show that

ar ot : R (0, )
ar ot (x) =

1
,
1 + x2

x R.

8
Polynomial Approximation and Series
Representation

The basi operations with real numbers are addition and multipli ation. So the
simplest fun tions are those that are evaluated by arrying out nitely many
su h operations. These fun tions are pre isely the polynomials. It is natural
to approximate the more ompli ated fun tions that we have studied|the
power fun tion for exponents other than nonnegative integers, the logarithm,
the exponential, the trigonometri fun tions|by polynomials. And it is natural to investigate how good the approximations are, and whether the more
ompli ated fun tions, despite not being polynomials, are somehow limits of
polynomials.
This hapter derives approximating polynomials Pn of ea h degree n for
the just-mentioned fun tions. For ea h su h fun tion f, for all x in a ertain
domain that may not be the full domain of f, the values Pn (x) tend to f(x)
as n grows. Thus limits of polynomials provide a uniform des ription of
the fun tions, despite the fun tions all being so di erent from ea h other.
Also, estimates of how well the polynomials approximate their limit-fun tions
provide a nite pro ess to ompute the fun tions to any desired a ura y. This
makes the fun tions more tangible than they were previously.
Most of the arguments in this hapter are not hard to follow, but in a few
pla es the al ulations get detailed. The reader is en ouraged to read those
passages lightly rather than get bogged down.
Se tion 8.1 is somewhat of a warmup, expanding the polynomial (1 + x)n
in powers of x. Se tion 8.2 establishes preliminary results for the al ulations
to follow in the hapter|an alternative notation for the integral, better suited
to omputation, and a dis ussion of the integral of the power fun tion when
the left endpoint of integration is zero. Se tion 8.3 nds approximating polynomials and remainders for the logarithm, and se tion 8.4 does the same for
the exponential, and se tion 8.5 does the same for the osine and the sine.
Finally, se tion 8.6 nds approximating polynomials and remainders for the

234

8 Polynomial Approximation and Series Representation

power fun tion at 1 + x, i.e., for (1 + x) . The results here generalize the
results of se tion 8.1.

8.1 The Finite Binomial Theorem


Let n Z0 be a nonnegative integer. The fun tion
f(x) = (1 + x)n

f : R R,

is a polynomial. To des ribe it in terms of nonnegative integer powers of x,


introdu e the notation
 
n(n 1) (n k + 1)
n
,
=
k!
k

k = 0, , n.

The numerator of the fra tion is the produ t of k terms, starting at n 0


and de rementing to n (k 1). The denominator is also the produ t of k
terms, from 1 to k, the factorial of k. When k = 0 the numerator and the
denominator are both understood to be 1 be ause a produ t of no terms
naturally should be the multipli atively neutral quantity 1, just as a sum of
no terms is the additively neutral quantity 0.
For nonnegative integers n and k with 0 k n, the binomial oe ients
n
k arrange themselves in a pleasing pattern. The rst few are
 
1
0
=
= 1,
0
0!

and
and
and

 
1
1
= 1,
=
0!
0
 
1
2
=
= 1,
0
0!

 
1
3
= 1,
=
0!
0

 
1
1
= 1,
=
1!
1

 
2
2
= 2,
=
1!
1

 
3
3
= 3,
=
1!
1

 
21
2
= 1,
=
2!
2

 
32
3
= 3,
=
2!
2

 
321
3
= 1.
=
3!
3

The famous arrangement is Pas al's Triangle, in whi h ea h internal entry


is the sum of the two entries above it on either side:

8.1 The Finite Binomial Theorem

235

1
1

1
1
1
1
1

1
3

4
5

6
10

15

4
10

20

et .

1
5

15

1
6

That is, the nth row's kth entry is nk , where the top row is onsidered the
0th row and the left entry of ea h row is onsidered the 0th entry.
Theorem 8.1.1 (Finite Binomial Theorem). Let n Z0 be a nonnegative integer, and let x R be any real number. Then
 
 
 
n
n 2
n n
(1 + x) = 1 +
x+
x + +
x ,
1
2
n
n

x R, n Z0 .

A more ompa t formulation uses the Sigma-notation for sums,


n

(1 + x) =

n  
X
n

k=0

xk ,

x R, n Z0 .

The oe ients in the Finite Binomial Theorem form the appropriate row
of Pas al's Triangle. For example,
(1 + x)5 = 1 + 5x + 10x2 + 10x3 + 5x4 + x5 .


Sin e n1 = n, the Binomial Theorem on rms our observation in se tion 6.6.2 that (1 + x)n > 1 + nx for any x > 0 and any n > 1.
The Finite Binomial Theorem is essentially a ombinatorial statement,
and so one an prove it (in the sense that a proof is a onvin ing argument
rather than a formal ritual) by talking rather than manipulating symbols.
Consider for example the x3 term in (1 + x)10 . Expand the tenfold produ t
of two-term sums
(1 + x)10 = (1 + x)(1 + x) (1 + x)

by multiplying out all tenfold produ ts that result from hoosing a 1 or an x


from ea h fa tor 1 + x. (There are 210 = 1024 su h terms.) The oe ient
of x3 in the expansion of (1+x)10 is the number of ways that we an make ten
onse utive hoi es of an x or a 1 that result in three net x's and seven net 1's.
The rst x that we hoose an be drawn from any of the ten fa tors 1 + x,

236

8 Polynomial Approximation and Series Representation

the se ond x an o ur at any of the nine remaining fa tors, and the third x
an o ur at any of the eight fa tors then remaining in turn. Thus there are
10 9 8 ways of hoosing the three x's. But in this ounting s heme, hoosing
the x's from, say, the se ond, fth, and eighth fa tors is viewed as a separate
event from hoosing the x's from the fth, se ond, and eighth fa tors, or from
the se ond, eighth, and fth fa tors, and so on. To eliminate over ounting,
we must divide the number of ways of arranging the labels (2, 5, and 8 in the
example) of the three fa tors where we hose x. There are 3 2 1 = 3! su h
arrangements. So nally, the oe ient of x3 in (1 + x)10 is
10 9 8
=
3!

 
10
.
3

The argument with general nonnegative integers n and k (where k n) in


pla e of 10 and 3 is the same.
This se tion has expanded a ertain polynomial (a nonnegative power
of 1 + x) in powers of x, naturally obtaining an expansion with only nitely
many terms. The rest of the hapter will expand nonpolynomial fun tions (see
exer ise 8.1.3) in powers of x as well, but the expansions will not terminate.
Exercises
8.1.1. Write out the binomial expansions for (1+x)3 and for (1+x)4 , multiply

them together, and on rm that you have obtained the binomial expansion
for (1 + x)7 .
8.1.2. Here is a third false argument about (1 + x/n)n , in the spirit of the

two arguments on page 206. A ording to the Binomial Theorem,




 
  n
  2
n x
x n
n x
n x
=1+
+

+
.
+
n
1 n
n nn
2 n2

n
n
1 = n, it follows that (1 + x/n) equals 1 + x plus

1+

That is, sin e


nitely
many terms, ea h of whi h is a onstant times a negative power of n.
Thus as n gets very large, (1 + x/n)n tends toward the value 1 + x.
This argument is in orre t. Explain at least one of its aws.

8.1.3. (a) Explain why the derivative of any polynomial is another polyno-

mial, and why the only polynomial that an equal its derivative, or equal the
negative of the derivative of its derivative, is the zero polynomial.
(b) Why an't any of the fun tions ln, exp, os, or sin be a polynomial?

8.2 Preliminaries for the Pending Cal ulations

237

8.2 Preliminaries for the Pending Calculations


8.2.1 An Alternative Notation
Definition 8.2.1 (New Notation for the Integral).
integrable from a to b, we write
Zb

If a fun tion f is

f(x)

x=a

as a synonym for

Zb

f.

Sin e ab f is a number that does not depend in any way on the symbol x
that is present in the new notation, that symbol an be
Rbrepla ed by any other
symbol not already in use. Thus other synonyms for a f are, for instan e,
Zb

Zb

f(t),

f(x1 ),

f() .

=a

x1 =a

t=a

Zb

The variable of integration, meaning the x or the t or the x1 or the here,


is alled a dummy variable be ause its name does not a e t the value of the
integral. It omes into existen e temporarily as we al ulate, only to disappear
when the al ulation is omplete. While our new notation is less streamlined
than the old, its advantage for the purposes of this hapter is that having the
variable of integration appear expli itly will let us keep tra k of events as we
integrate fun tions that are themselves integrals of other fun tions, whi h are
integrals in turn, and so on. The new notation fa ilitates omputing.
One parti ular formula will be useful in the new notation, so we establish
it immediately. For any R de ne a fun tion
g(x) = f (1 + x) = (1 + x) .

g : R>1 R,

The graph of g is the graph of the power fun tion f translated one unit
to the left. So for any x 0, Ar1+x
(f ) = Arx0 (g), and for any x su h that
1
1
0
1 < x < 0, Ar1+x (f ) = Arx (g). That is in all ases,
Z 1+x

f =

Zx

g,

x > 1.

Rewrite the right side in our new notation to get the useful formula
Z 1+x
1

f =

Zx

x1 =0

(1 + x1 ) ,

x > 1.

(8.1)

238

8 Polynomial Approximation and Series Representation

One last omment (for now) about notation: The reader with prior ba kground in al ulus has almost ertainly seen an even more adorned form of
writing the integral than the one just introdu ed here, to wit,
Zb
a

f=

Zb

f(x) dx.

x=a

We will bring the dx into our notation later, when it too will help keep tra k
of ertain al ulations, but for this hapter it is unne essary.
8.2.2 The Power Function Integral With Endpoint 0

For any positive real number b R>0 and for any nonnegative exponent
R0 , the region under the graph of the th power fun tion from 0 to b,
R = {(x, y) R2 : 0 x b, 0 y f (x)},

is a bounded subset of the plane, and so it has an area. (The relevant fa t


in play here are that for 0, the power fun tion f extends ontinuously
from R>0 to R0 . Spe i aly, f (0) = 0 for > 0 while f0 (0) = 1.) Furthermore, sin e the power fun tion is monotoni on [0, b], the area is an integral,
Zb

exists and is Arb0 (f ).

So to nd the integral, we need only to nd the area.


Sin e the power fun tion is nonnegative on R0 , we have for any number a
su h that 0 < a b,
Arba (f ) Arb0 (f ).

On the other hand, a box having base [0, a] and height a shows (see gure 8.1) that also.
Arb0 (f ) a+1 + Arba (f ).
That is, remembering the expli it formula for Arba (f ),

b+1 a+1
b+1 a+1
+1
Arb
.
(f
)

a
+

0
+1
+1

Now let a tend to 0. Sin e 0, ertainly +1 > 0, so that lima 0 a+1 = 0.


(Here it is understood that a is tending to 0 from the positive side.) Consequently, by the Squeezing Rule and various other limit rules,
Arb0 (f ) =
That is,

b+1
.
+1

8.2 Preliminaries for the Pending Cal ulations

239

f (x) = x

PSfrag repla ements

b
Figure 8.1.

Zb
0

f =

One more outer box

b+1
,
+1

0, b 0.

Here we have extended the formula to b = 0, when it simply says that 0 = 0.


(Before ontinuing, we make a brief digression. Even though we assumed
that R0 , it is striking that pro eeding from the inequalities
b+1 a+1
b+1 a+1
+1
(f
)

a
+
Arb

0
+1
+1

to the on lusion that


Arb0 (f ) =

b+1
+1

required only that + 1 > 0, i.e., it required only that > 1. This suggests,
for example (letting = 1/2), that even though the graph of the fun tion
f1/2 : R>0 R,

f1/2 (x) = 1/ x

has a verti al asymptote at x = 0, so that the region under the graph over the
interval (0, 1] is unbounded, apparently the very nite number 2 is a redible
value for the region's area. The problem with this reasoning is that our underlying invo ation that regions have areas has been made only for bounded
regions. In our framework, the area of the unbounded region doesn't exist in
the rst pla e, and so the inequalities at the beginning of this paragraph are
meaningless for negative values of su h as = 1/2. But this example en ourages autiously expanding the notion of area to apply to some unbounded
regions. More spe i ally, the idea is that 2 is the area of the region under

240

8 Polynomial Approximation and Series Representation

the graph of f1/2 over (0, 1] in the sense that 2 is the least number that is
at least as big as all areas of nite trun ations of the region. The equation
Z1
0

1
=2
x

gives the value of an improper integral. We will not pursue this subje t
further.)
Spe i examples of the power fun tion integrals with left endpoint 0 are,
using the new notation,
Zx

Zx

1 = x,

x1 =0

x1 =

x1 =0

and in general,

Zx

xn
1 =

x1 =0

xn+1
,
n+1

x2
,
2

Zx

x21 =

x1 =0

x3
,
3

n Z0 , x 0.

The previous display holds only for x 0 so far, but sin e the exponent n
is a nonnegative integer we want to extend it to negative x as well. Sin e
(x1 )n = (1)n xn
1 , for negative x naturally
Zx

n+1
xn
1 = (1)

x1 =0

Z x

x1 =0

xn
1,

n Z0 , x < 0,

the extra power of 1 oming from the reversed dire tion of integration. Thus
Zx

xn+1
(x)n+1
=
, n Z0 , x < 0.
n+1
n+1
x1 =0
R
And so predi tably enough, the formula for 0x fn is symboli ally robust,
n+1
xn
1 = (1)

Zx

xn
1 =

x1 =0

xn+1
,
n+1

n Z0 , x R.

Exercise
8.2.1. Figure 8.1 shows a ase where > 0. How would the
Rb gure hange
for = 0? Does this a e t the argument in the text that 0 f exists and
equals b+1 /( + 1)?

8.3 The Logarithm


By the de nition of the logarithm and by the useful formula (8.1),

8.3 The Logarithm

ln(1 + x) =

Z 1+x

f1 =

Zx

x1 =0

1
,
1 + x1

241

x > 1.

Re all the nite geometri sum formula for any n Z0 ,


1 + r + r2 + + rn1 =

1 rn
,
1r

r 6= 1.

Rearrange the formula to get


rn
1
= 1 + r + r2 + + rn1 +
,
1r
1r

r 6= 1,

and then substitute x1 for r,


xn
1
= 1 x1 + x21 + (1)n1 x1n1 + (1)n 1 ,
1 + x1
1 + x1

x1 6= 1. (8.2)

Now integrate, letting x1 vary from 0 to x, to obtain the logarithm as a


polynomial and a remainder, using Proposition 5.5.2 (page 176),
ln(1 + x) =

Zx

x1 =0

1
= Pn (x) + Rn (x),
1 + x1

x > 1,

(8.3)

where Pn (x) is obtained by integrating the polynomial in (8.2) term-by-term,


arrying out power fun tion integrals with left endpoint 0 (exer ise 8.3.1(a)),
Pn (x) = x

x2 x3
xn
+
+ (1)n1
,
2
3
n

and the remainder is the integral of the rest of the right side of (8.2),
n

Rn (x) = (1)

Zx

x1 =0

xn
1
.
1 + x1

Equation (8.3) and sequen e limit rules ombine to show that for any x su h
that x > 1, the sequen e of polynomials (Pn (x)) onverges to ln(1+x) if and
only if the sequen e of remainders (Rn (x)) onverges to 0 (exer ise 8.3.1(b)).
So the next question is how the remainder Rn (x) behaves as n grows.
We address the question by estimating the remainder. The integral is being
taken from x1 = 0 to x1 = x, where x is xed through this dis ussion and
now we stipulate that 1 < x 1. (The analysis will not work for x > 1.) If
0 x 1 then throughout the integration pro ess,
0<

and so

1
1,
1 + x1

242

8 Polynomial Approximation and Series Representation


|Rn (x)|

Zx

xn
1 =

x1 =0

xn+1
.
n+1

If 1 < x < 0 then throughout the integration,


1
1

,
1 + x1
1+x

0<

and so

Z




|x|n+1
1
1 xn+1
1 x
n

.
=
x1 =
|Rn (x)|



1 + x x1 =0
1+x n+1
1+x n+1

In both ases, letting C be the maximum of 1 and 1/(1 + x),


|Rn (x)| C

|x|n+1
.
n+1

Therefore by sequen e limit rules (exer ise 8.3.1( )),


lim(Rn (x)) = 0,
and we have shown that
ln(1 + x) = lim(Pn (x)),
n

1 < x 1.

Less formally,
ln(1 + x) = x

x2 x3
xn
+
+ (1)n1
+ ,
2
3
n

1 < x 1.

And in Sigma-notation,
ln(1 + x) =

(1)n1

n=1

xn
,
n

1 < x 1.

Note in parti ular the formula when x = 1,


ln(2) = 1

1 1 1 1 1 1 1
+ + + + .
2 3 4 5 6 7 8

(See exer ise 8.3.2 for a geometri derivation of this formula.)


Beyond being pretty, the boxed formulas are shorthand for an algorithm to
ompute ln(1+x) (where 1 < x 1) to any desired a ura y by omputing a
polynomial, i.e., by arrying out nitely many additions and multipli ations..
The idea is that given a desired a ura y, i.e., an error toleran e for our
answer, the analysis that we just arried out lets us nd a degree n so that
|Rn (x)| is smaller than the error toleran e. Thus Pn (x) is as lose to ln(1 + x)
as was desired.

8.3 The Logarithm

243

Theorem 8.3.1 (Taylor Polynomial and Remainder for the Logarithm). For any x (1, 1], and for any n Z0 ,

ln(1 + x) = Pn (x) + Rn (x)

where
Pn (x) =

n
X

(1)k1

k=1

and

x2 x3
xn
xk
=x
+
+ (1)n1
k
2
3
n



1
|x|n+1
max 1,
.
n+1
1+x
x (1, 1],

|Rn (x)|

Consequently, for any

xn
x2 x3
+
+ (1)n1
+ .
n
2
3
n
For example, we use the theorem to estimate ln(1.1) by hand to within
1/500, 000. The nth degree polynomial approximation to ln(1.1) is

ln(1 + x) = lim(Pn (x)) = x

Pn (0.1) = (0.1)

(0.1)2
(0.1)3
(0.1)n
+
+ (1)n1
,
2
3
n

and the remainder satis es


|Rn (0.1)|

(0.1)n+1
.
(n + 1)

The only symboli variable is n, and the goal is to approximate ln(1.1) to


within 1/500, 000. Set n = 4 in the previous display to get
|R4 (0.1)|

1
.
500, 000

That is, the fourth degree Taylor polynomial


1
1
1
1

10 200 3000 40000


= 0.10000000 0.00500000 + 0.00033333 0.00002500

P4 (0.1) =

= 0.09530833

agrees with ln(1.1) to within 0.00000200 , so that

0.09530633 ln(1.1) 0.09531033 .

Ma hine te hnology should on rm this.


The graphs of the natural logarithm and its rst ve Taylor polynomials
are plotted from 0 to 2 in gure 8.2. (Here the fun tions are ln(x) and Pn (x1)
so that the oordinate axes are in their usual position for the logarithm.) A
good he k of your understanding is to see if you an determine whi h graph
is whi h in the gure (exer ise 8.3.3).

244

8 Polynomial Approximation and Series Representation

0.5

1.5

-1

-2

-3

Figure 8.2.

The natural logarithm and its Taylor polynomials

Exercises
8.3.1. (a) Carry out the integration to obtain the polynomial Pn (x) in the

se tion.
(b) Explain how sequen e limit rules show that limn (Pn (x)) = ln(1 + x)
if and only if limn (Rn (x)) = 0.
( ) Use sequen e limit rules to show that limn (Rn (x)) = 0 if 1 < x 1.
8.3.2. Figure 8.3 shows a partial geometri de omposition of ln(2).
(a) Identify the one box of width 1 in the gure and explain why its area
is 1 1/2.
(b) Identify the one box of width 1/2 in the gure and explain why its
area is 1/3 1/4.
( ) Identify the two boxes of width 1/4 in the gure and explain why their
areas are 1/5 1/6 and 1/7 1/8.
(d) Identify the four boxes of width 1/8 in the gure and explain why
their areas are 1/9 1/10, 1/11 1/12, 1/13 1/14, and 1/15 1/16.
(e) Identify the eight boxes of width 1/16 in the gure and explain why
the areas of the rst two su h boxes are 1/17 1/18 and 1/19 1/20.
8.3.3. In gure 8.2, identify the graphs of P1 (x 1) through P5 (x 1) and
the graph of ln(x) near x = 0 and near x = 2.

8.4 The Exponential

y = 1/x

PSfrag repla ements

Partial geometri de omposition of ln(2)

Figure 8.3.

8.4 The Exponential


8.4.1 A Precalculation

De ne
I0 : R R,

I0 (x) = 1,

and then de ne for n = 1, 2, 3,

In : R R,

So, for example, I1 (x) =

Consequently, I2 (x) =

Similarly, I3 =

Rx

f /2
0 2

Rx

Rx

f
0 1

In (x) =

1 = x. That is,
I1 = f1 /1.
= x2 /2, or
I2 = f2 /2.

= x3 /(3 2),
I3 = f3 /3!.

Continuing in this vein shows that

Zx
0

In1 .

245

246

8 Polynomial Approximation and Series Representation


n Z0 .

In = fn /n!,

(Re all that 0! = 1 by onvention, so that the formula for In overs the
ase n = 0.)
In our new notation, the al ulations are written,
I0 (x) = 1,
Zx
I1 (x) =
I2 (x) =
I3 (x) =

x1 =0
Zx

1 = x,
Z x1

1=

x1 =0 x2 =0
Z x1

Zx

Z x2

Zx

x1 =

x1 =0

1=

x1 =0 x2 =0 x3 =0

Zx

x2
,
2
Z x1

x2 =

x1 =0 x2 =0

Zx

x1 =0

x3
x21
=
,
2
32

and in general,
In (x) =

Zx

Z x1

x1 =0 x2 =0

Z xn1

1=

xn =0

xn
,
n!

n Z0 .

8.4.2 The Calculation

The method for expressing the exponential fun tion as ever-higher degree
polynomials plus the orresponding remainders is to use the basi identity
exp(x) = 1 +

Zx

exp,

xR

(a rearrangement of the formula 0x exp = exp(x) exp(0)) over and over.


Fix any real number x for the duration of this dis ussion, and start from
the basi identity, renotated,
x

e =1+

Zx

ex1 .

x1 =0

By the basi identity again, then Proposition 5.5.2, and then the pre al ulation,


Z x1
x2
e
1+
e =1+
x2 =0
x1 =0
Zx
Zx
Z x1
=1+
1+
ex2
x1 =0
x1 =0 x2 =0
Zx
Z x1
=1+x+
ex2 .
x

Zx

x1 =0 x2 =0

8.4 The Exponential

247

On e more by the same pro ess,




Z x2
x3
e
1+
e =1+x+
x3 =0
x1 =0 x2 =0
Zx
Zx
Z x1
Z x1 Z x2
=1+x+
1+
Zx

=1+x+

Z x1

x1 =0 x2 =0
Zx
Z x1
2

x
+
2!

ex3

x1 =0 x2 =0 x3 =0
Z x2
x3

e .

x1 =0 x2 =0 x3 =0

Continuing this pro ess through n iterations shows that


ex = Pn (x) + Rn (x),

x R,

where Pn (x) is an nth degree polynomial,


Pn (x) = 1 + x +

x2
xn
+ +
,
2!
n!

and Rn (x) is the remainder integral,


Rn (x) =

Zx

Z x1

x1 =0 x2 =0

Z xn

exn+1 .

xn+1 =0

The task now is to analyze Rn (x), sin e as with the logarithm, the sequen e
of polynomials (Pn (x)) onverges to ex if and only if limn (Rn (x)) = 0.
If x 0 then the integrand exn+1 of Rn (x) lies between 1 and ex . If
x < 0 then the integrand lies between ex and 1. In either ase there is a
onstant C (spe i ally, C is the maximum of ex and 1) su h that all through
the integration,
0 < exn+1 C.

So if x 0 then integrating this inequality n + 1 times (see Proposition 3.3.14


on page 121) orrespondingly bounds the remainder integral by a multiple of
the pre al ulated integral,
0 Rn (x) C

Zx

Z x1

x1 =0 x2 =0

Z xn

1=C

xn+1 =0

xn+1
.
(n + 1)!

If x < 0 then all of the integrals in Rn(x) have out-of-order endpoints, and
n+1

Rn (x) = (1)

Z0

Z0

x1 =x x2 =x1

and so

Z0

xn+1 =xn

exn+1 ,

248

8 Polynomial Approximation and Series Representation


|Rn (x)| =

Z0

Z0

x1 =x x2 =x1
Z0
Z0

x1 =x x2 =x1
Zx
n+1

= (1)

Z0

exn+1

xn+1 =xn
Z0

x1 =0 x2 =0

=C

(x)n+1
.
(n + 1)!

xn+1 =xn
Z xn
Z x1

xn+1 =0

Combining the ases, we have shown that for all x R,


|Rn (x)| C

|x|n+1
.
(n + 1)!

Sin e |x| an be greater than 1, the quotient |x|n+1 /(n + 1)! an be a ratio of
two large numbers, making its behavior initially un lear. But we an analyze
it. The Ar himedean property of the real number system says that there exists
some positive integer m su h that m > 2|x|. It follows that
1
|x|
< ,
m+1
2

|x|
1
< ,
m+2
2

|x|
1
< ,
m+3
2

and so on. Now onsider the onstant


K=

|x|m
,
m!

and note that (ea h line after the rst in the following display making referen e
to the line before it)
|x|m
= K,
m!
|x|m+1
|x|m
|x|
|x|
1
=

=K
K ,
(m + 1)!
m! m + 1
m+1
2

 2
|x|m+2
|x|m+1
|x|
1 1
1
,
=

K =K
(m + 2)!
(m + 1)! m + 2
2 2
2
 2
 3
1
|x|m+2
|x|
1
1
|x|m+3
=K
,
=

K
(m + 3)!
(m + 2)! m + 3
2
2
2

and in general,
|x|m+
K
(m + )!

The previous inequality rewrites as

 
1
,
2

Z0 .

8.4 The Exponential

And so,

 n+1m
|x|n+1
1
,
K
(n + 1)!
2

n + 1 m.

 n
1
,
2

n m 1.

|Rn (x)| CK2m1

It follows that for any x R,

249

lim(Rn (x)) = 0,
n

and therefore

ex = lim(Pn (x)).

Less formally,
ex = 1 + x +

x2
xn
+ +
+ ,
2!
n!

x R,

or
ex =

X
xn
,
n!

n=0

x R.

Note in parti ular the formula when x = 1,


1
1
1
1
1
1
1
+ + + + + + + .
2! 3! 4! 5! 6! 7! 8!
Theorem 8.4.1 (Taylor Polynomial and Remainder for the Exponential). For any x R, and for any n Z0 ,
e=1+1+

ex = Pn (x) + Rn (x)

where
Pn (x) =

n
X
xk
x2
xn
=1+x+
+ +
k!
2!
n!

k=0

and (letting x denote the smallest integer that is at least x)


|Rn (x)|

Consequently, for any x R,

|x|n+1
max 1, 3x .
(n + 1)!

ex = lim(Pn (x)) = 1 + x +
n

Proof. Here the point is that we know


|Rn (x)|

xn
x2
+ +
+ .
2!
n!

|x|n+1
max {1, ex } ,
(n + 1)!

but sin e we don't know ex , we loosen the bound a little more to get a bound
that we an ompute.

250

8 Polynomial Approximation and Series Representation

Exercise
8.4.1. Estimate the a ura y to whi h P10 (1) approximates e.

8.5 The Cosine and the Sine


This time, start from the identities
os(x) = 1

Zx

x =0

Z x1

sin(x) =

x1 =0

sin(x1 ),
os(x1 ).

Next,
os(x) = 1
sin(x) =

Zx

Zx

x1 =0 x2 =0

Z x1

x1 =0
Zx

=x

Z x1

os(x2 ),

1
x =0
Z x1 2

x1 =0 x2 =0


sin(x2 )

sin(x2 ).

And then,
os(x) = 1

Zx

Z x1


Z x2
1

x1 =0 x2 =0
Zx
Z x1

x2
+
=1
2!

sin(x) = x

x3 =0

Z x2

x1 =0 x2 =0 x3 =0
Zx
Z x1 Z x2
x1 =0 x2 =0 x3 =0


sin(x3 )

sin(x3 ),
os(x3 ).

And
Zx
Z x1 Z x2 Z x3
x2
os(x) = 1 +
os(x3 ),
2!
x1 =0 x2 =0 x3 =0 x4 =0


Z x3
Zx
Z x1 Z x2
1
sin(x) = x
sin(x4 )
=x

x1 =0 x2 =0 x3 =0
Z x1 Z x2
Zx
3

x
+
3!

x4 =0

Z x3

x1 =0 x2 =0 x3 =0 x4 =0

sin(x4 ).

Sin e the integrands of the remainder integrals are always sines and
osines, they are always bounded in absolute value by 1. So the analysis of

8.5 The Cosine and the Sine

251

the exponential fun tion applies here to show that sin(x) and os(x) are the
limits of nite sums,
x2 x4
+
, x R,
2!
4!
x3 x5
sin(x) = x + , x R,
3!
5!

os(x) = 1

or
os(x) =
sin(x) =

(1)n

n=0

(1)n

n=0

x2n
,
(2n)!

x R,

x2n+1
, x R.
(2n + 1)!

Theorem 8.5.1 (Taylor Polynomial and Remainder for the Cosine


and Sine). For any x R, and for any n Z0 ,

os(x) = Pn (x) + Rn (x),

where
Pn (x) =

n
X

(1)k

k=0

and

x2k
x2 x4
x2n
=1
+
+ (1)n
(2k)!
2!
4!
(2n)!

|Rn (x)|

Consequently, for any x R,

os(x) = lim(Pn (x)) = 1


n

|x|2n+2
.
(2n + 2)!

x2 x4
x2n
+
+ (1)n
+ .
2!
4!
(2n)!

For any x R, and for any n Z0 ,

sin(x) = Pn (x) + Rn (x),

where
Pn (x) =

n
X

k=0

and

(1)k

x3 x5
x2n+1
x2k+1
=x
+
+ (1)n
(2k + 1)!
3!
5!
(2n + 1)!

|Rn (x)|

Consequently, for any x R,

sin(x) = lim(Pn (x)) = x


n

|x|2n+3
.
(2n + 3)!

x3 x5
x2n+1
+
+ (1)n
+ .
3!
5!
(2n + 1)!

252

8 Polynomial Approximation and Series Representation

Here the polynomial Pn for the osine has degree 2n rather than n,
and the polynomial Pn for the sine has degree 2n + 1. The reader who
wants polynomial subs ripts to mat h polynomial degrees is wel ome to renotate them P2n and P2n+1 , and then to renotate the remainders orrespondingly R2n and R2n+1 .
Exercises
8.5.1. Without a al ulator, use the rst three terms of the Taylor polynomials for sin(x) at 0 to approximate a de imal representation of sin(0.1). Also
ompute the de imal representation of an upper bound for the error of the
approximation. Bound sin(0.1) between two de imal representations.
8.5.2. For any value s (0, /2), let a(s) be the area of the shaded region in
the left half of gure 8.4, and let b(s) be the area of the shaded region in the
right half of the gure.
(a) Explain why a(s) = (1/2)(1 os(s)) sin(s). Approximate 1 os(s)
by a polynomial in s that in ludes all powers of s no higher than s3 . Do the
same for sin(s). Use these two polynomials to do the same for a(s).
(b) Explain why b(s) = (1/2)s (1/2) os(s) sin(s). Approximate os(s)
and sin(s) by polynomials in s that in lude all powers of s no higher than s3 ,
and then do the same for b(s).
( ) Evaluate
a(s)
.
s 0 b(s)

lim

PSfrag repla ements


s

sin(s)
os(s)

sin(s)
os(s)

Figure 8.4.

Two areas

8.6 The Power Function


8.6.1 The Polynomial and the Remainder

The method for expressing the power fun tion as ever-higher degree polynomials plus the orresponding remainders is essentially the same as for the

8.6 The Power Fun tion

253

other fun tions in this hapter. (And a fairly general theory is little di erent
from our spe i examples.) This time, begin by re alling formula (2.11) from
the end of hapter 2 as extended in exer ise 6.5.2 (page 205),
Zb

f =

b+1 a+1
+1

R, 6= 1, 0 < a b.

The usual veri ation on rms that in fa t the formula holds if a and b are
out of order as well. Spe ialize to a = 1 and b = 1 + x where x > 1, and
repla e by 1,
Z 1+x

f1 =

(1 + x) 1
,

R, 6= 0, x > 1.

Rewrite the formula as follows, swit hing to our new notation and iting the
useful formula (8.1) (page 237),

(1 + x) = 1 +

Zx

(1 + x1 )1 ,

x1 =0

(8.4)

R, x > 1.

(For = 0, this formula doesn't follow from the previous one, but in this ase
it simply says that 1 = 1.)
Fix any real number x > 1. Let R be any real number. By (8.4)
on e,
Z
x

(1 + x) = 1 +

(1 + x1 )1 .

x1 =0

By (8.4) again, and by the results ited in the al ulation for the exponential
fun tion, the integral is
Zx



Z x1
(1 + x2 )2
1 + ( 1)
x =0
x =0
Z x1
Z x 2 Z x1
(1 + x2 )2
=
1 + ( 1)
x1 =0
x1 =0 x2 =0
Zx
Z x1
(1 + x2 )2 ,
= x + ( 1)

(1 + x1 )1 =

x1 =0

Zx

x1 =0 x2 =0

so that

(1 + x) = 1 + x + ( 1)
Z x1

x1 =0 x2 =0

Z x1

(1 + x2 )2 .

x1 =0 x2 =0

Similarly, the double integral is


Zx

Zx

Zx

Z x1

Z x2

(1 + x3 )
1 + ( 2)
(1 + x2 )
=
x3 =0
x1 =0 x2 =0
Zx
Z x1 Z x2
Zx
Z x1
(1 + x3 )3
1 + ( 2)
=
x1 =0 x2 =0
2

x
+ ( 2)
2!

x1 =0 x2 =0 x3 =0
Z x2
(1 + x3 )3 ,
x1 =0 x2 =0 x3 =0

Zx

Z x1

254

8 Polynomial Approximation and Series Representation

so that now
(1+x) = 1+x+

( 1) 2
x +(1)(2)
2!

Zx

Z x1

Z x2

(1+x3 )3 .

x1 =0 x2 =0 x3 =0

Reintrodu e the binomial oe ient notation, but where now an be any


real number,
 
( 1) ( k + 1)

,
=
k!
k

k Z0 .


Unless is a nonnegative integer, the binomial


oe ient k is nonzero for

all k Z0 , and as soon as k ex eeds , k hanges sign with ea h in rement
of k. After n iterations of the pro ess, we get
(1 + x) = Pn (x) + Rn (x),

x > 1,

where similarly to the Finite Binomial Theorem,


 
 
 

2
n
Pn (x) = 1 +
x+
x + +
x
1
2
n

and the remainder is


Rn (x) = ( 1) ( n)

Zx

Z x1

x1 =0 x2 =0

Z xn

(1 + xn+1 )n1 .

xn+1 =0

We have derived the formula up to n = 2, and exer ise 8.6.1 is to obtain it


for n = 3. As usual, the sequen e of polynomials (Pn (x)) onverges to (1+x)
if and only if the sequen e of remainders (Rn (x)) has limit 0. If Z0 then
the formulas for Pn (x) and Rn (x) show that Pn (x) = P (x) and Rn (x) = 0
for all n . This situation is overed by the Finite Binomial Theorem. The
situation where R but / Z0 will be dis ussed next.
Exercises
8.6.1. Carry out one more step of the pro ess in the se tion to get from the
formula (1 + x) = P2 (x) + R2 (x) to the formula (1 + x) = P3 (x) + R3 (x).
8.6.2. Re all that

ln(1 + x) =

Zx

(1 + x)1 ,

x > 1.

x1 =0

Also, spe ializing the work just done to the ase = 1 gives an equality
(1 + x1 )1 = Pn1 (x1 ) + Rn1 (x1 ),

x > 1, n 1.

8.6 The Power Fun tion

255

Integrate the equality to re-obtain the degreeR n polynomial approximation


of ln(1 + x) (i.e., the Pn (x) for ln(1 + x) is xx1 =0 Pn1 (x), integrating the
Pn1 (x1 ) for (1 + x1 )1 ), and to obtain an expression for the remainder that
is valid for x > 1 rather than only for |x| < 1. (To make the notation
work smoothly, write the Rn1 (x1 ) for (1 + x1 )1 with outermost variable of
integration x2 .)
8.6.2 The Infinite Binomial Theorem

Now take R but / Z0 . That is, is a real number other than a


nonnegative integer. To review, we have the formula
(1 + x) = Pn (x) + Rn (x),

x > 1,

where Pn (x) is the nth degree polynomial


Pn (x) = 1 +

and
Rn (x) =

 
 
 

2
n
x+
x + +
x ,
1
2
n


Z xn
Zx
Z x1

(1 + xn+1 )n1 ,

(n + 1)!
n+1
xn+1 =0
x1 =0 x2 =0

so that




Zx
Z xn
Z x1

n1

(1 + xn+1 )
Rn (x) =
(n + 1)!

.
n+1
xn+1 =0
x1 =0 x2 =0

The task is to analyze Rn (x) as n grows large. The analysis here is more
ompli ated than those earlier in this hapter, and so the reader should engage
with it to taste.
Sin e n is growing large, we assume that n > 1 (su h n exists by
the Ar himedean property of the real number system), so that the exponent
n 1 in the remainder integral is negative.
If x 0 then the integrand (1 + xn+1 )n1 lies between (1 + x)n1
and 1. If x < 0 then the integrand lies between 1 and (1 + x)n1 . In either
ase there is a onstant C (the maximum of 1 and (1 + x)n1 ) su h that
all through the integration,
0 < (1 + xn+1 )n1 C.

So if x 0 then integrating this inequality n + 1 times bounds the remainder


integral by a multiple of the pre al ulated integral,

256

8 Polynomial Approximation and Series Representation




Z xn
Zx
Z x1

(n + 1)!
1
|Rn (x)| C

n+1
xn+1 =0
x1 =0 x2 =0



n+1

x
= C
.
n+1

If x < 0 then all of the integrals in Rn (x) have out-of-order endpoints, and so


Z0
Z0
Z0

(n + 1)!
|Rn (x)| =
(1 + xn+1 )n1

n+1
xn+1 =xn
x1 =x x2 =x1


Z0
Z0
Z0

(n + 1)!
C
1

n+1
xn+1 =xn
x1 =x x2 =x1


Zx
Z xn
Z x1

(n + 1)!(1)n+1
= C
1

n+1
xn+1 =0
x1 =0 x2 =0




(x)n+1 .
= C
n+1

Combining the ases, we have shown that for all x > 1,


|Rn (x)| Bn (x),




n+1

|x|
where Bn (x) = C
.
n+1

Now impose the stronger ondition that |x| < 1. Then more spe i ally,
|x| = 1 2,

(8.5)

where = (1 |x|)/2 > 0. Consider the sequen e (whose origin will be explained in a moment)


(sn ) =

n


n+1 .

By sequen e limit rules and the fa t that the absolute value fun tion is ontinuous,
lim(sn ) = 1,
(8.6)
n

and so, sin e 1/(1 ) > 1, there exists a starting index N su h that
0 < sn <

1
1

for all n N.

(8.7)

Now onsider the ratio of the estimates of onse utive-generation remainders,


the generation being omfortably large,
 n+1

|x|
n
C n+1
Bn (x)



=
|x|n = n + 1 |x| = sn |x|.
Bn1 (x)
C
n

Thus the ratio is what gives rise to sn . Let

8.6 The Power Fun tion


r=

257

1 2
,
1

so that 0 r < 1. By (8.5) and (8.7),


Bn (x)
<r
Bn1 (x)

for all n N.

That is (starting at n = N + 1 rather than at n = N just to be tidy, and ea h


line after the rst in the following display making referen e to the line before
it),
BN+1 (x) rBN (x),

BN+2 (x) rBN+1 (x) r2 BN (x),

BN+3 (x) rBN+2 (x) r3 BN (x),

and in general,

BN+ (x) r BN (x) for all Z0 .

By sequen e limit rules it follows that for any x su h that |x| < 1,
lim(Bn (x)) = 0,
n

and therefore that

lim(Rn (x)) = 0.
n

Re all that
(1 + x) = Pn (x) + Rn (x),

|x| < 1, n Z0 .

Sin e limn (Rn (x)) = 0, it follows that lim(Pn (x)) = (1 + x) . Less formally,
 
 
 

2
n
(1 + x) = 1 +
x+
x + +
x + ,
1
2
n

|x| < 1, R.

In Sigma-notation,
 
X
n
(1 + x) =
x ,
n

n=0

|x| < 1, R.

Theorem 8.6.1 (Taylor Polynomial and Remainder for the Power


Function: the Binomial Theorem). For any x (1, 1), for any R,
and for any n Z0 su h that n > 1,
(1 + x) = Pn (x) + Rn (x)

258

8 Polynomial Approximation and Series Representation

where
 
 
 
n  
X
k

n
2
Pn (x) =
x =1+
x+
2x + +
x
k
1
n
n
k=0

and




n+1



|x|
|Rn (x)|
max 1, (1 + x)n1 .

n+1

Consequently, for any x (1, 1) and for any R,


(1 + x) = lim(Pn (x)) = 1 +
n

 
 
 

n
x+
2x2 + +
x + .
1
n
n

The graphs of the square root and its rst six Taylor
polynomials are

plotted from 0 to 4 in gure 8.5. (Here the fun tions are x and Pn (x 1) so
that the oordinate axes are in their usual position.) The gure shows that
the polynomials are approximating the square root well from 0 to 2, but not
for x beyond 2.

2.5
2
1.5
1
0.5
1
Figure 8.5.

The square root and its Taylor polynomials

Exercises
8.6.3. Let be a real number.
(a) Let (~sn ) = (( n)/(n + 1)). Show that limn (~sn ) = 1. Explain

why (8.6) follows.

8.7 Summary

259

(b) Con rm the al ulation in the se tion that Bn (x)/Bn1 (x) = sn |x|.
( ) Use sequen e limit rules to explain why limn (Bn (x)) = 0.
(d) Use sequen e limit rules to explain why limn (Rn (x)) = 0.
8.6.4. Use a se ond degree polynomial to approximate

1.2. Find a guaran-

teed a ura y of the approximation and thus nd upper and lower bounds for

1.2.
8.6.5. (For the symboli ally-in lined.) The identity
(1 + x) = e ln(1+x) ,

|x| < 1, R

now has a formal interpretation as an equality involving three in nite sums:


on the left side of the equality, a sum of oe ients times powers of x; and on
the right side of the equality, a se ond sum of oe ients times powers of a
third sum. Does the formal interpretation seem to gives the same expression
on both sides of the equality?
8.6.6. The remainder analysis for the logarithm and the power fun tion required |x| < 1 (or x = 1 for the logarithm) to show that limn (Rn (x)) = 0, but

the analysis for the exponential fun tion, the osine, and the sine did not.
Why is this? That is, what aspe t of the analysis was di erent enough among
the various fun tions to produ e two pairs of di erent- avored results?

8.7 Summary
The al ulations that we arried out for various fun tions in this hapter have
mu h in ommon. On e we have the Fundamental Theorem of Cal ulus, an
exer ise (exer ise 10.1.8) will show how to des ribe all of the al ululations in
one general en oding.

9
Theory and Applications of the Derivative

The derivative is a lo al onstru t, omputed by zooming in on the behavior


of a fun tion about a single point. This hapter dis usses how despite being
lo al, the derivative an help as answer large-s ale questions.
One su h question, the problem of optimizing a fun tion|making it as
large or as small as possible|is dis ussed in se tion 9.1. Here the basi idea is
intuitive geometri ally: where the fun tion is optimized, its graph should have
a horizontal tangent, i.e., its derivative should be zero. Se tion 9.2 introdu es
the Mean Value Theorem and its onsequen es. This theorem gives a formula
relating a fun tion and its derivative, with no referen e to the fa t that the
derivative is a limit. With the formula in hand, plausible statements about
fun tions (e.g., the fa t that if the derivative is positive, the fun tion is stri tly
in reasing) be ome easy to prove. Se tion 9.3 shows how to use al ulus to
sket h the graphs of fun tions. Computer graphing te hnology is e e tive and
readily available nowadays, but nonetheless sometimes al ulus an inform us
about features of a graph that are not shown learly by the ma hine-generated
gure. Se tion 9.4 dis usses problems of the following form: given two related
quantities, and given the rate of hange of one of them, what is the rate of
hange of the other? Su h problems are known as related rates problems.

9.1 Optimization
To optimize a fun tion is to make it as big as possible or as small as possible by
suitably spe ifying its input-value. The following de nition gives us language
to dis uss this subje t.
Definition 9.1.1 (Minimum, Maximum, Extremum, Local Minimum,
Local Maximum, Local Extremum). Let A be a subset of R, and let
f : A R be a fun tion. Then

262

9 Theory and Appli ations of the Derivative

A minimum of f is a fun tion-value f(x) su h that f(x) f(s) for


all s A. The fun tion f has a minimum at x if f(x) is a minimum
of f.
A maximum of f is a fun tion-value f(x) su h that f(x) f(s) for
all s A. The fun tion f has a maximum at x if f(x) is a maximum
of f.
An extremum of f is a minimum or a maximum of f. The fun tion f
has an extremum at x if f(x) is an extremum of f.
A local minimum of f is a fun tion-value f(x) su h that f(x) f(s)
for all s A within some positive distan e of x. The fun tion f has
a local minimum at x if f(x) is a lo al minimum of f.
A local maximum of f is a fun tion-value f(x) su h that f(x) f(s)
for all s A within some positive distan e of x. The fun tion f has
a local maximum at x if f(x) is a lo al maximum of f.
A local extremum of f is a lo al minimum or a lo al maximum of f.
The fun tion f has a local extremum at x if f(x) is a lo al extremum
of f.

A minimum of a fun tion is a lo al minimum but not ne essarily onversely, and similarly for maximum and extremum. A fun tion an have at
most one minimum and at most one maximum (although the fun tion an
takes its minimum at many di erent inputs, and similarly for the maximum),
but it an have many lo al minima and lo al maxima. Exer ise 9.1.1 asks for
examples of these phenomena. A ording to the de nition, if f is onstant,
then its value is both a minimum and a maximum, and it has a minimum and
a maximum at every input-value. To es ape this ounterintuitive linguisti
ir umstan e, one an further de ne a strict minimum of f by hanging
\f(x) f(s)" to \f(x) < f(s)" in the de nition, and so on. (In general, an
inequality is alled stri t if it pre ludes equality.)
9.1.1 The Extreme Value Theorem

For an arbitrary fun tion f, there is no reason to believe that extrema exist
to be found. But for a ertain lass of fun tions, optimization is guaranteed,
at least in the abstra t.

Let f be a ontinuous fun tion whose domain is a losed, bounded interval,

Theorem 9.1.2 (Extreme Value Theorem).

f : [a, b] R,

ontinuous.

Then f assumes a maximum value and a minimum value.

9.1 Optimization

263

Like the Intermediate Value Theorem (page 189), the Extreme Value Theorem is an abstra t existen e theorem. That is, its on lusion is not that
(for example) \f(x) f(s) for all s [a, b]," whi h in isolation would be
meaningless sin e the hypotheses make no mention of any parti ular point x.
Rather the on lusion is that there exists some x su h that f(x) f(s) for
all s [a, b]. But the theorem says nothing about where x is or how to nd
it.
The Extreme Value Theorem is not a al ulus theorem: it makes no referen e to derivatives, integrals, or in nite sums. Espe ially, the fun tion f
in the theorem is not assumed to be di erentiable. On the other hand,
the theorem does make use of limits, sin e limits are at the heart of ontinuity. Also, the theorem depends on a property of the real number system, a
property equivalent|after some work|to the Completeness Property that we
have invoked throughout these notes. Be ause the issues that arise in proving
Theorem 9.1.2 are foundational, the proof is beyond our s ope.
In our ontext, what is more important anyway than proving the theorem
is that one an gain intuition about its ontent by onvin ing oneself via
examples that its on lusion an fail unless all of the hypotheses are met.
That is,
 A dis ontinuous fun tion whose domain is a losed, bounded interval need

not assume a maximum value or a minimum value.

 A ontinuous fun tion whose domain is a losed but unbounded interval

need not assume a maximum value or a minimum value.

 A ontinuous fun tion whose domain is a bounded but non- losed interval

need not assume a maximum value or a minimum value.

 A ontinuous fun tion whose domain is a non- losed, unbounded interval,

or whose domain is not an interval at all, need not assume a maximum


value or a minimum value.

Exer ise 9.1.2 asks for examples.


Exercises
9.1.1. Illustrate graphi ally examples of the following phenomena:

(a) A lo al minimum of a fun tion need not be a minimum of the fun tion.
(b) A fun tion an take its maximum at many inputs.
( ) A fun tion an have many lo al minima, all distin t.
(d) A fun tion an have many lo al maxima, ea h taken at many inputs.

9.1.2. (a) Find a fun tion f : [0, 1] R that assumes no maximum value
and no minimum value. (Give a lear sket h of the graph of su h an f, or a
formula for su h an f, with some explanation in either ase.)

264

9 Theory and Appli ations of the Derivative

(b) Find a ontinuous fun tion f : [0, ) R that assumes no minimum


value and no maximum value.
( ) Find a ontinuous fun tion f : (0, 1] R that assumes no maximum
value.
(d) Find a ontinuous fun tion f : (0, 1] R that assumes no minimum
value and no maximum value.
9.1.3. (a) Explain why for any ontinuous fun tion f : [a, b] R, we may
take the odomain of f to be some interval [L, M] instead.
(b) May we take the range of f to be [L, M]?
9.1.2 Conditions for Optimization

Under suitable onditions, the onditions under whi h a fun tion is lo ally
optimized are very onstrained.
Theorem 9.1.3 (Critical Point Theorem).

losed, bounded interval,

Consider a fun tion on a

f : [a, b] R.

Suppose that f has a lo al extremum at the point x [a, b], i.e., f(x) is a
lo al extremum of f. Then at least one of the following onditions holds:
(a) f is not di erentiable at x.
(b) x is an endpoint of [a, b], i.e., x = a or x = b.
( ) f is di erentiable at x and f (x) = 0.
For an example of onditions (a) and (b), the absolute value fun tion
| | : [1, 1] R

assumes its minimum at x = 0, where it is not di erentiable, and the fun tion
assumes its maximum at the endpoints x = 1 and x = 1.
Corollary 9.1.4. Let f : [a, b] R be ontinuous, and let f be di erentiable on (a, b). The only possible points where its extrema an o ur
are a, b, and x (a, b) su h that f (x) = 0.

Proof (of the theorem). Suppose that f(x) is a lo al maximum. Suppose also
that f is di erentiable at x, and that x is not an endpoint, i.e., that onditions (a) and (b) do not hold. We want to show that onsequently ondition ( )
holds, i.e., that f (x) = 0.
If f (x) is positive then let = f (x)/2, again a positive number. A ording
to exer ise 4.2.3(d) (page 136), the line through (x, f(x)) of slope uts the

9.1 Optimization

265

graph of f from above to below moving from left to right. That is, for some
values s > x, the point (s, f(s)) on the graph lies above the line, whose height
over s is greater than f(x) sin e its slope is positive. In other words, for some
values s > x we have f(s) > f(x), ontradi ting the fa t that f(x) is a lo al
maximum.
The remaining ases, when f (x) is negative rather than positive, and
when f(x) is a lo al minimum rather than a lo al maximum, an be handled
by arguments that are virtual repeats of the previous paragraph. Alternatively, they an be redu ed to the ase that we have overed. For example, if
f (x) is negative then the fun tion g(s) = f(s) has a lo al maximum at x,
and g (x) = f (x) by the Chain Rule; thus g (x) > 0, whi h we have argued
is impossible. And similarly, if f has a lo al minimum at x then the fun tion
g = f has a lo al maximum at x.

Example 9.1.5. Consider the fun tion


f(x) = x3 3x.

f : [2, 2] R,

This fun tion is di erentiable with derivative


f : [2, 2] R,

f (x) = 3x2 3.

The values x su h that f (x) = 0 are therefore x = 1 and x = 1, and so any


possible lo al extrema of f o ur at these points or at the endpoints x = 2
and x = 2. Compute that
f(2) = 2,

f(1) = 2,

f(1) = 2,

f(2) = 2.

Sin e f assumes a minimum by the Extreme Value Theorem, the minimum is


f(2) = f(1) = 2. Similarly, the maximum of f is f(1) = f(2) = 2.
Example 9.1.6. Consider the fun tion
f : R R,

f(x) =

x2

1
.
+1

Then f is always positive. The denominator of f(x) is smallest at x = 0, so


that the maximum of f is f(0) = 1. The denominator of f(x) grows without
bound as |x| grows without bound, so that the positive value f(x) grows ever
loser to 0 as |x| grows without bound, and f has no minimum.
Exercises
9.1.4. Sket h the graph of a generi fun tion f : [a, b] R. Then sket h
the graph of the related fun tion g(s) = f(s).

266

9 Theory and Appli ations of the Derivative

9.1.5. Find all lo al extrema of the following fun tions. For ea h lo al extremum, state whether it is also global.
(a) f : [2, 2] R, f(x) = x4 x2 .
(b) f : [2, 2] R, f(x) = 4x3 3x4 .
9.1.6. Consider the fun tion

f : [1, ) R,

f(x) = ln(x)/x.

Note that f(1) = 0, that f(x) 0 for all x [1, ), and that limx f(x) = 0
(see page 157).
(a) Does the Extreme Value Theorem guarantee that f has a maximum?
(b) Show that the unique x [1, ) su h that f (x) = 0 is x = e.
( ) Sin e limx f(x) = 0, there exists some value b 1 su h that f(x)
f(e)/2 for all x b. Explain why f has a maximum on [1, b], and why this
maximum is also the maximum of f on [1, ).
(d) Whi h is larger, ln(e)/e or ln()/? Explain. Consequently, whi h is
larger, ln(e) or e ln()? ln(e ) or ln(e )? Whi h is larger, e or e ?
9.1.3 Optimization Story-Problems

To solve an optimization story-problem, pro eed as follows.


 Draw and label a gure.
 Write an equation for the quantity to optimize. If possible, express the

quantity in terms of a single independent variable. Be aware of the domain


of values for the variable.
 Typi ally the domain is an interval. If the interval is losed and bounded,
evaluate the quantity at the endpoints. Otherwise analyze the quantity
near the missing endpoints, or as the variable gets large or small.
 Find the values of the variable for whi h the derivative of the quantity is
zero, and evaluate the quantity at ea h su h value.
 Evaluate the quantity at points where its derivative fails to exist.
In pra ti e, one sometimes gets a little asual with this pro edure.
Example 9.1.7. Optimize

that sum to 1.

the produ t

x1 x2

of two nonnegative numbers

Experimentation suggests strongly that the answer is x1 = x2 = 1/2. For


example,
9
2 8
16
3 7
21
4 6
24
5 5
25
1 9

=
,

=
,

=
,

=
,

=
.
10 10
100 10 10
100 10 10
100 10 10
100 10 10
100

To phrase the problem in terms of one variable, we want to optimize the


fun tion

9.1 Optimization
f(x) = x(1 x),

267

x [0, 1].

Sin e the domain of f is a losed, bounded interval, and sin e f is ontinuous,


f assumes a maximum and a minimum. The endpoint value f(0) = f(1) = 0
is the minimum sin e f is nonnegative. We an nd the maximum in various
ways:
 Complete the square:
f(x) = x2 + x 1/4 + 1/4 = 1/4 (x 1/2)2 ,

so sin e the square is nonnegative, f takes its maximum at x = 1/2, and


the maximum is 1/4.
 Use al ulus: Sin e f(x) = x x2 ,
f (x) = 1 2x,

and so f (x) = 0 if and only if x = 1/2. By Corollary 9.1.4, the maximum


of f therefore o urs at x = 1/2, and it is f(1/2) = 1/4.
Example 9.1.8. Optimize

sphere.

the volume of a ylinder that sits inside a

PSfrag repla ements

Figure 9.1.

Cylinder inside sphere, seen in pro le

The situation is depi ted in pro le in gure 9.1. Let the ylinder have
base-radius r and half-height a. Then its volume is

268

9 Theory and Appli ations of the Derivative


V = r2 2a.

But sin e the ylinder ts in a sphere, whi h may as well be the unit sphere,
we also have
a2 + r2 = 1.

The extreme ases of the geometry are ylinders that degenerate either to a
line segment between the spheri al poles or an equatorial disk, both having
volume zero. And so, as a fun tion of a, the volume is
f(a) = (1 a2 ) 2a = 2(a a3 ),

a [0, 1].

Sin e f is di erentiable and its domain is a losed, bounded interval, it assumes


a minimum and a maximum. Sin e the endpoint value f(0) = f(1) = 0 is the
minimum, the maximum value must be assumed somewhere where f = 0.
Compute that
f (a) = 2(1 3a2 ),

and so the maximum o urs when a2 = 1/3. Re all that a2 + r2 = 1, so


that also r2 = 2/3 for the maximum volume. That is, the proportions for the
maximum volume satisfy 2a2 = r2 , or
r=

2 a.

The ylinder's height is h = 2a, so the answer rewrites as


h=

2 r.

Sin e this answer is phrased in terms of proportions, it does not depend on


normalizing the sphere-radius to 1.
Example 9.1.9. Optimize

the surfa e-area of a ylinder of given volume.


We may normalize the volume and phrase the answer in terms of proportions. Let the ylinder have radius r and height h. Then its surfa e area,
en ompassing the base, the top, and the side, is
A = 2r2 + 2rh = 2(r2 + rh),

while the volume is

V = r2 h.

Normalize the volume to , so that r2 h = 1 and thus rh = 1/r. Then the


surfa e area is
A(r) = 2(r2 + 1/r).

Note that A is very large for small positive r (a very tall, thin ylinder) and
for large positive r (a very wide, squat ylinder). To make A small, set its
derivative to 0,

9.1 Optimization

269

A (r) = 2(2r 1/r2 ) = 2(2r3 1)/r2 .

Thus A (r) = 0 when r3 = 1/2. But sin e r2 h = 1, onsequently r3 h = r, or


r3 = r/h. And so the proportions of the optimal ylinder are
r
1
= ,
h
2

or h = 2r. The an has diameter d = 2r, so in fa t the proportions are


h = d.

That is, the an sits tightly inside a ube.

b s (s)

a se (s)

PSfrag repla ements


s
a
Figure 9.2.

Moving a toothpi k around a orner

Example 9.1.10. An

ant wants to move a toothpi k around a 90-degree


orner between a tunnel of width a and a tunnel of width b. Both tunnels
are horizontal and have negligible height. How long a toothpi k an t
around the orner?
The situation is depi ted in gure 9.2. For any angle s (0, /2), the
longest toothpi k that an t into the orner at angle s has length
f(s) = a se (s) + b s (s).

270

9 Theory and Appli ations of the Derivative

The smallest value of f is length of the longest toothpi k that will t all the
way around the orner.
The domain of f is a bounded interval, but it is missing both of its endpoints. However, note that f(s) is very large for s slightly greater than 0 and
for s slightly less than /2, and this is onsonant with our geometri intuition
that the toothpi k's t is tightest somewhere in the middle of the pro ess
of getting it around the orner. So, sin e f is di erentiable, we onsider its
derivative,
f (s) = a tan(s) se (s) b ot(s) s (s) =

a sin3 (s) b os3 (s)

sin2 (s) os2 (s)

This derivative vanishes for



s = ar tan (b/a)1/3 ,

and for this value of s we have


f(s) = a2/3

p
p
3/2
.
a2/3 + b2/3 + b2/3 a2/3 + b2/3 = a2/3 + b2/3

For example, if a = 8 and b = 27 then a toothpi k of length

(82/3 + 272/3 )3/2 = 13 13 47

will t around the orner.


Example 9.1.11. The bottom of a drive-in movie s reen is h units higher
than the viewer's eye. The top of the s reen is H units higher than the
viewer's eye. How far ba k should the viewer park to maximize the verti al angle that she per eives the s reen to ll?

The situation is depi ted in gure 9.3. Let the horizontal distan e from
the viewer's eye to the s reen be x, a positive number. Then the angle in
question is b a where
tan(a) =

h
,
x

tan(b) =

H
.
x

To maximize b a, we may maximize tan(b a) instead. But for any a and b


whatsoever,
tan(ba) =

sin(b a) sin(b) os(a) os(b) sin(a)


tan(b) tan(a)
=
=
,
os(b a) os(b) os(a) + sin(b) sin(a) 1 + tan(a) tan(b)

so that in parti ular for our a and b,


f(x) = tan(b a) =

x
H/x h/x
= (H h) 2
,
2
1 + hH/x
x + hH

x > 0.

9.1 Optimization

271

PSfrag repla ements

ba
x

Figure 9.3.

At the drive-in

For small positive x, tan(ba) is lose to 0, and for large positive x, tan(ba)
is lose to 0 as well, so we sear h for x-values su h that f (x) = 0. Compute
hH x2
x2 + hH 2x2
= (H h) 2
.
2
2
(x + hH)
(x + hH)2

And so the optimal parking-distan e is x = hH. This distan e is the geometri mean of h and H, meaning their multipli ative average, as ompared
to their arithmeti mean (h + H)/2.
f (x) = (H h)

Example 9.1.12. A

parti le travels through medium 1 at speed v, and


through medium 2 at speed w. If the parti le travels from point A to
point B (see gure 9.4) in the least possible amount of time, what is the
relation between angles and ?

If, for example, v is greater than w, then one argument is that the parti le should travel in medium 1 (where it moves faster) to the point on the
boundary between the media just above point B, and then drop straight down
to B, thus spending as little time as possible traveling slowly. But this strategy entails taking a long path from A to B. A se ond argument is that the
parti le should take the shortest path from A to B, the line segment joining
them, regardless of the fa t that in doing so it will traverse a longer path
in medium 2, where it moves slowly. The orre t answer will lie somewhere
between the answers suggested by these two arguments.

PSfrag repla ements


272

9 Theory and Appli ations of the Derivative


a tan()

A
a se ()

medium 1

medium 2

b se ()

b tan() B
Figure 9.4.

Geometry of Snell's Law

Sin e time is distan e over speed, a little trigonometry shows that this
problem is to minimize the time
t=

a
b
se () + se (),
v
w

where the lateral distan e traveled is


a tan() + b tan() = d.

View as the independent variable and as a fun tion of . Di erentiate


the expression for the lateral distan e to get
a se 2 () + b se 2 () = 0.

Now di erentiate t to get


t =

a
b
tan() se () + tan() se () ,
v
w

whi h rewrites as
t =

a
b
sin() se 2 () + sin() se 2 () .
v
w

But from the derivative of the lateral distan e, b se 2 () = a se 2 (),


and so
a
a
sin() se 2 () sin() se 2 ()
v
w


sin() sin()
2
= a se ()
.

v
w

t =

That is, t = 0 exa tly when sin()/v = sin()/w, or


sin() v
= .
sin() w
This relation is alled Snell's Law.

9.2 The Mean Value Theorem

273

Exercises
9.1.7. Optimize the weighted produ t x1 x22 of two nonnegative numbers that
sum to 1.
9.1.8. Optimize the volume of a one that sits inside a sphere. (Let r be the
radius of the one's ir ular base, and let h be the height from the one's base

to its vertex. Your answer should des ribe the proportions of the one.)

9.1.9. Optimize the volume of a ylinder that sits inside a one.


9.1.10. Rotate a right triangle of a given hypotenuse to form a one of greatest

volume.

9.1.11. Optimize the volume of a box reated by utting four small squares

away from the orners of a large square and then folding up the resulting
aps.

9.1.12. Find the point(s) on the parabola y = x2 that are nearest to the point
(0, 9/2).

9.1.13. Optimize the geometri mean hH where h and H are nonnegative


numbers whose arithmeti mean is 1.

9.2 The Mean Value Theorem


9.2.1 Statement of the Theorem
Theorem 9.2.1 (Rolles Theorem).

Suppose that a fun tion

f : [a, b] R

is ontinuous on [a, b] and di erentiable on (a, b), and suppose further


that f(a) = f(b) = 0. Then there exists some value c (a, b) su h that
f (c) = 0.
Rolle's Theorem is illustrated in gure 9.5.

Proof. If f is identi ally 0 then so is f , and any c (a, b) will do.

Otherwise, note that f has a minimum and a maximum by the Extreme


Value Theorem. Either the minimum or the maximum is nonzero, and so it o urs at a nonendpoint c (a, b), where f is di erentiable. By Corollary 9.1.4,
f (c) = 0.

274

9 Theory and Appli ations of the Derivative


f(x)

PSfrag repla ements

x
a

Figure 9.5.

Rolle's Theorem

The proof of Rolle's Theorem relies on the Extreme Value Theorem, whi h
we have not proved. Like the Extreme Value Theorem, Rolle's Theorem is an
existen e theorem: its on lusion is not that \f (c) = 0," whi h in isolation
would be meaningless sin e the hypotheses make no mention of a point c, but
that there exists some c su h that f (c) = 0.

Suppose that a fun tion

Theorem 9.2.2 (Mean Value Theorem).


f : [a, b] R

is ontinuous on [a, b] and di erentiable on (a, b). Then there exists some
value c (a, b) su h that
f (c) =

f(b) f(a)
.
ba

The Mean Value Theorem is illustrated in gure 9.6. Note also that in
Ar himedes's quadrature of the parabola, ea h ins ribed triangle has its middle vertex over the point c from the Mean Value Theorem ( f. page 17 and
gure 1.14 on page 18). The ondition that f needs to be ontinuous on the
losed interval but need not be di erentiable at the endpoints means, for example, that the Mean Value Theorem applies to a fun tion su h as the square
root on [0, 1] despite the fa t that its graph has a verti al tangent at the origin.

Proof. De ne auxiliary fun tions


g : [a, b] R,

g(x) = f(a) +

f(b) f(a)
(x a)
ba

9.2 The Mean Value Theorem

275

f(x)

PSfrag repla ements

f(b)

f(b) f(a)
ba

f(a)

x
a

Figure 9.6.

The Mean Value Theorem

and
h : [a, b] R,

h(x) = f(x) g(x).

The graph of g is the line segment from (a, f(a)) to (b, f(b)), and the fun tion h measures the verti al distan e from the graph of g to the graph of f.
Sin e h meets the onditions for Rolle's Theorem, there exists some value
c (a, b) su h that h (c) = 0. But in general,
h (x) = f (x) g (x) = f (x)

f(b) f(a)
,
ba

so that the ondition h (c) = 0 is, as desired,


f (c) =

f(b) f(a)
,
ba

Exercise
9.2.1. (a) Sket h the graph of a fun tion f : [a, b] R for whi h there are
exa tly four possible values of c in the Mean Value Theorem.
(b) Sket h the graph of a fun tion f : [a, b] R for whi h there are
in nitely many values of c, but not every c (a, b) is su h a value.

276

9 Theory and Appli ations of the Derivative

9.2.2 Consequences of the Mean Value Theorem

The Mean Value Theorem has a wealth of onsequen es. To rephrase, it says
that if f is ontinuous on [a, b] and di erentiable on (a, b) then
f(b) f(a)
= f (c)
ba

for some c (a, b).

To appre iate why this statement enables us to do things that we an't do


without it, rst note that it involves a sort of tradeo . The drawba k is that:

The statement involves a point


value of c.

c (a, b),

but we don't know the

But on the other hand, the advantage is that:

The statement gives us a onne tion between the fun tion f and
its derivative f with no referen e to a limit.
Sin e limits are elaborate, te hni al, and sometimes unwieldy, the gain outweighs the drawba k on e we learn how to use the theorem despite not knowing c.
Here is an example. Let I be any interval in R, and let
f : I R

be a di erentiable fun tion su h that f = 0 everywhere on I. As mentioned


in exer ise 6.4.3 (page 202), this strongly suggests that f is onstant, but
until now an easy proof was not a essible to us. The easy proof pro eeds
as follows. Let a and b be any two distin t points of I. We may assume
that a < b. Restri t the domain of f to [a, b]. The resulting fun tion satis es
the hypotheses of the Mean Value Theorem. Therefore,
f(b) f(a) = f (c)(b a) for some c (a, b).

We don't know where c is, but this doesn't matter be ause f (c) = 0 for all c.
That is,
f(a) = f(b).

Sin e a and b are arbitrary points of I it follows that f is onstant.


It would be eminently reasonable for the reader to underwhelmed by an
argument to support the patently obvious fa t that if the derivative is always zero then the fun tion is onstant. However, the underlying issue is that
the fa t is patently obvious only be ause of our intuition that the set of real
numbers geometri ally forms an unbroken line, a ontinuum. The set of rational numbers also omes with a linear order, and as a subset of the line
the rationals leave no gaps of positive length|that is, every real interval of

9.3 Curve Sket hing

277

positive length ontains rational numbers. Algebrai ally, the rational numbers and the real numbers an be hara terized indistinguishably: addition,
subtra tion, multipli ation, and division work as they should. Nonetheless, if
we go through the exer ise of de ning the on epts in these notes only in the
restri ted ontext of the rational numbers, then not all of the results ontinue
to hold. In parti ular, the fun tion
f : Q Q,

f(x) =

if x2 < 2,
1 if x2 > 2
0

is di erentiable at ea h point x Q, its derivative is 0 everywhere, and yet it


is not a onstant fun tion. Thus, any argument that if the derivative is zero
the fun tion is onstant must somehow rely on a property of the real number
system that distinguishes it from the rational number system.
The next exer ise is to derive more onsequen es of the Mean Value Theorem.
Exercise
9.2.2. (a) Let f1 , f2 : [a, b] R be di erentiable fun tions su h that f1 = f2
on [a, b]. Show that f2 = f1 + C for some onstant C.
(b) Let f : [a, b] R be a di erentiable fun tion su h that f > 0
on [a, b]. Show that f is stri tly in reasing on [a, b].
( ) Let f : [a, b] R be di erentiable and stri tly in reasing. Must it be
true that f > 0 on [a, b]? Proof or ounterexample.
(d) Let f : [a, b] R be a di erentiable fun tion su h that f 0
on [a, b]. Show that f is in reasing on [a, b].
(e) Let f : [a, b] R be di erentiable and in reasing. Must it be true
that f 0 on [a, b]? Proof or ounterexample.

9.3 Curve Sketching


To sket h the graph of a fun tion f with the help of al ulus, here are some
points to bear in mind.
 The formula for f may make lear where f is positive, negative, and zero,
i.e., where the graph is above, below, or rossing the x-axis.
 If the formula for f has a denominator then f is unde ned at x-values
where the denominator is zero. If the numerator is nonzero for su h x
then f probably has a verti al asymptote at x. Che k the sign of f at
values slightly larger than x and slightly smaller than x to see whether

the graph is rising very high or dropping very low on ea h side of the
asymptote.

278

9 Theory and Appli ations of the Derivative

 Similarly, if the formula has a square root then f is de ned only for x-




values where the quantity under the square root is nonnegative, and so
on.
The graph may also have horizontal asymptotes or diagonal asymptotes.
Horizontal asymptotes arise if f(x) tends to a limit as x + or as x
, and similarly for diagonal asymptotes if f(x)/x tends to a limit.
The formula for f may make lear where f is positive, negative, and zero,
i.e., where the graph is rising, falling, or has a horizontal tangent.
The formula for f may make lear where f is positive, negative, and zero,
i.e., where the graph is onvex (bending up), on ave (bending down), or
in e ting (swit hing bend-dire tions).
At an x-value where f is unde ned it is understood that f and f are
unde ned as well, and similarly at an x-value where f is unde ned it is
understood that f is unde ned as well. At an x-value where f is de ned
but f is not, the graph may have a orner or some other exoti behavior.

Computer graphing te hnology is so e e tive and so readily available that


sket hing urves with the help of al ulus may feel like a pointless endeavor,
espe ially sin e the omputer an plot many points qui kly and produ e a
gure that is a urate in shape and s ale. However, sometimes al ulus an tell
us about key features of the graph that are hard to see in omputer-generated
plots, e.g., the pre ise lo ation of lo al extrema, or points of in e tion.
Example 9.3.1. Consider the fun tion
f(x) =

x2
,
x2 1

x R, x 6= 1.

Note that f is even (i.e., f(x) = f(x)), so we need only study it for x 0.
Also,
 f(0) = 0.
 limx + f(x) = 1.
 limx 1+ f(x) = + and limx 1 f(x) = .

(Here the se ond bullet is shorthand for f tends to 1 as its inputs grow large
and positive, and the third bullet is shorthand for f is large and positive at
inputs slightly greater than 1 and f is large and negative at inputs slightly
less than 1.) Compute that the derivative of f is
f (x) =

2x
2x (x2 1) x2 2x
= 2
.
(x2 1)2
(x 1)2

Thus f (0) = 0 and f (x) < 0 for 0 < x < 1 and for 1 < x. Similarly, a bit of
algebra shows that

9.3 Curve Sket hing


f (x) =

279

2(3x2 + 1)
.
(x2 1)3

Thus f (x) < 0 for 0 x < 1 and f (x) > 0 for 1 < x. We an present many
of our observations in a table. The i ons indi ate whether the graph of f is
rising or falling, and whether it is onvex or on ave.
0<x<1 1<x
f

+
f

A omputer-generated plot of f ( gure 9.7) shows the features that we have


dedu ed analyti ally,

20
10

-2

-1

-10
-20
Figure 9.7.

Graph of f(x) = x2 /(x2 1)

Example 9.3.2. Consider the fun tion


f(x) = x1/3 + x1/3 = x1/3 (x2/3 + 1),

x > 0.

This fun tion should behave like x1/3 for x near 0, and like x1/3 for large x.
More spe i ally,
 f(x) > 0 for all x > 0.

280

9 Theory and Appli ations of the Derivative

 limx 0+ f(x) = + and limx + f(x) = +.

Compute that the derivative of f is


f (x) =

1 2/3 1 4/3 1 4/3 2/3


x
x
= x
(x
1),
3
3
3

x > 0.

Thus f (1) = 0, and f (x) < 0 for 0 < x < 1, and f (x) > 0 for 1 < x. Similarly,
a bit of algebra shows that
f (x) =

2 7/3
x
(2 x2/3 ),
9

x > 0.

Thus f (23/2 ) = 0, and f (x) < 0 for 0 < x < 23/2 , and f (x) > 0 for 23/2 <
x. We an present many of these observations in a table.
0 < x < 1 1 < x < 23/2 23/2 < x
f
+
+
+
f

+
+
f
+
+

A omputer-generated plot of f ( gure 9.8) shows some of the features that


we have dedu ed analyti ally, but the transition from positive to negative
urvature at x = 23/2 2.828 is not really visible, nor is the asymptoti
behavior f(x) x1/3 for large x.
Example 9.3.3. Consider the fun tion
f(x) = 2 sin(x) sin(2x),

x .

Note that f is odd (i.e., f(x) = f(x)), so we may study it on [0, ] instead.
In parti ular, f(0) = f() = 0. The derivative of f is (now suppressing the
domain from the notation)
f (x) = 2 os(x) 2 os(2x).

Re all that os(2x) = 2 os2 (x) 1. Therefore,


f (x) = 2( os(x) 2 os2 (x) + 1) = 2( os(x) 1)(2 os(x) + 1).

Thus f (x) = 0 if x = 0, 2/3. And f(2/3) = 3 + 3/2 = 3 3/2 2.6.


Sin e 2( os(x) 1) 0 for all x, the sign of f is determined by the sign of
2 os(x) + 1, whi h is positive for 0 x < /3 and negative for /3 < x .
The se ond derivative of f is

9.3 Curve Sket hing

281

3.5
3.25
3
2.75
2.5
2.25
1
Figure 9.8.

Graph of f(x) = x1/3 + x1/3

f (x) = 2 sin(x) + 4 sin(2x).

Re all that sin(2x) = 2 sin(x) os(x). Therefore,


f (x) = 2 sin(x)(1 4 os(x)).

Thus f (x) = 0 at x = 0, , ar os(1/4). Sin e os(/3) = 1/2 and os(/2) =


0, it follows that ar os(1/4) lies
between /3 and /2. And a small al ulation
shows that f(ar os(1/4)) = 3 15/8 1.45. Sin e sin(x) is positive for 0 <
x < , while 1 4 os(x) is positive for 0 x < ar os(1/4) and negative
for ar os(1/4) < x , we have the following table.
f
f
f

0 < x < ar os(1/4) ar os(1/4) < x < 2/3 2/3 < x


+
+
+
+
+

A omputer-generated plot of f ( gure 9.9) shows most of the features that we


have dedu ed analyti ally, although the in e tion points over ar tan(1/4),
where the graph hanges from bending up to bending down, were not easy to
pi k out until the gure is enhan ed to emphasize them.
This example arises from the very beginnings of Fourier analysis, loosely
the theory of expressing general fun tions as ombinations of os illations, similarly to how we expressed fun tions as ombinations of powers (polynomials)

282

9 Theory and Appli ations of the Derivative

2
1
-3

-2

-1

-1
-2
Figure 9.9.

Graph of f(x) = 2 sin(x) sin(2x)

in hapter 8. The weighted ombination of the os illations sin(x) and sin(2x)


is approximating the 45-degree line identity fun tion f1 (x) = x on [, ].
For any positive integer n, the fun tion
gn (x) = 2

n
X

k=1

(1)k1

1
sin(kx)
k


1
1
n1 1
= 2 sin(x) sin(2x) + sin(3x) + (1)
sin(nx)
2
3
n

uses more os illations to approximate the the line more losely. The graph
of g6 is shown in gure 9.10.
Example 9.3.4. Consider the fun tion
f(x) = 2(x 1)5/3 + 5(x 1)2/3 = (x 1)2/3 (2x + 3).

Here we take the domain of f to be the set of all real numbers, even though
a ording to our formalism f(x) is sensible only for x 1. The idea is that
for x < 1, we an take (x 1)1/3 as the negative number whose ube is x 1
(this number is (1 x)1/3 ), and then f(x) is its square. Observe that
 f(x) = 0 for x = 0 and x = 3/2.
 f(0) = 3.
 limx + f(x) = + and limx f(x) = .

Next ompute that the derivative of f is

9.3 Curve Sket hing

283

3
2
1
-3

-1

-2

-1
-2
-3
Figure 9.10.

Graph of g6 (x) = 2

P6

k=1 (1)

k1

sin(kx)/k

2
(x 1)1/3 (2x + 3) + 2(x 1)2/3
3
2
= (x 1)1/3 ( (2x + 3) + 2(x 1))
3
10
1/3
(x 1)
x.
=
3

f (x) =

Observe that
 f (1) is unde ned, limx 1 f (x) = and limx 1+ f (x) = .
 f (0) = 0.
 f (x) > 0 for x < 0, f (x) > 0 for 0 < x < 1, and f (x) > 0 for x > 1.

The se ond derivative of f is



10
1
4/3
1/3
f (x) =
(x 1)
x + (x 1)
3
3


1
10
4/3
x+x1
(x 1)
=
3
3
2x 3
10
(x 1)4/3
.
=
3
3

Observe that
 f (x) is unde ned at x = 1 (naturally, sin e f (1) was already unde ned).
 f (3/2) = 0.
 f (x) < 0 for x < 1 and 1 < x < 3/2, and f (x) > 0 for x > 3/2.

284

9 Theory and Appli ations of the Derivative

Along with the value f(0) = 3, note that


f(3/2) = (1/2)2/3 6 = 6/22/3 > 6/2 = 3.

We an present many of our observations in a table.


x < 3/2 3/2 < x < 0 0 < x < 1 1 < x < 3/2 3/2 < x
f

+
+
+
+
f
+
+

+
+
f

A omputer-generated plot of f ( gure 9.11) shows most of the features that


we have dedu ed analyti ally, but it does not learly show the in e tion point
at (3/2, f(3/2)), where the graph in e ts.

6
4
2

-2

-1

-2
Figure 9.11.

Graph of f(x) = (x 1)2/3 (2x + 3)

Example 9.3.5 (Shape of the power function). The following table summarizes many of our observations about the power fun tion f (x) on R>0 ,
extended to x = 0 when possible. In all ases the graph lies in the rst oordinate quadrant and passes through the point (1, 1). The observations ombine
to show that for all < 0, the graph of the fun tion looks qualitatively like
0 and 1, the
the hyperbola-bran h y = 1/x, that for all stri tly between

graph looks qualitatively like the square root urve y = x, and that for all

9.4 Related Rates Story-Problems

285

> 1, the graph looks qualitatively like the parabola y = x2 . Of ourse, the
graph of f0 is the line y = 1 and the graph of f1 is the line y = x.
f
f (1)
lims 0 f (s)
f = f1
f (0)

f = ( 1)f2

<0=00<<1=11<
+
+
+
+
+
1
1
1
1
1

1
0
0
0

0
+
+
+
0

1
0
+
0

0
+

Exercises
9.3.1. Let f = ln. What is f ? What is its sign? How does this relate to

exer ise 5.3.3 (page 163)?

9.3.2. Graph the following fun tions, giving some dis ussing riti al points,
asymptotes, onvexity/ on avity, and so on, as relevant.
(a) f(x) = x3 /(1 x2 ).
(b) f(x) = (1 x2 )2 .
( ) f(x) = x/(1 + x2 )
(d) f(x) = x + sin(x).
(e) f(x) = x ln(x).

9.4 Related Rates Story-Problems


In a typi al related rates problem, some time-dependent pro ess involves two
related quantities. At some moment, we presumably an measure one quantity,
and we know its rate of hange. The idea is to determine the rate of hange
of the se ond quantity at that moment. The te hnique is to di erentiate the
original relation between the quantities with respe t to time, remembering to
use the Chain Rule.
Example 9.4.1. The

bottom end of a ladder of length is being moved


away from the wall at onstant speed. At what speed is the top of the
ladder sliding down the wall?

Let x(t) denote the time-dependent distan e of the base of the ladder from
the wall, and let h(t) denote the height of the top of the ladder up the wall.
The situation is depi ted in gure 9.12. Then we have

286

9 Theory and Appli ations of the Derivative

PSfrag repla ements


x
Figure 9.12.

x(t)2 + h(t)2 = ,

Ladder

x(0) = 0,

x (t) = 1.

(Here we normalize the onstant speed given to us by the problem to 1 for


onvenien e.) Di erentiate with respe t to time,
2x(t)x (t) + 2h(t)h (t) = 0.

That is, sin e x (t) = 1,


h (t) =

Sin e h(t) =

x(t)
.
h(t)

p
2 x(t)2 , we have (now suppressing t from the notation),
x
.
h =
2
x2

Thus, at the beginning moment of the pro ess, when the ladder's base is at
the wall (x = 0), the horizontal motion of the ladder's base is not ausing any
verti al motion of the ladder's top down the wall. On the other hand, at the
end-moment, when x = , the top of the ladder instantaneously has in nite
verti al velo ity down the wall.
Example 9.4.2. A pedestrian of height h walks away
of height H at onstant speed. At what speed is her

reasing?

from a street light


shadow-length in-

Let x denote the pedestrian's horizontal distan e from the base of the
street light, and let s denote the pedestrian's shadow-length. The situation is
depi ted in gure 9.13. Similar triangles show that

9.4 Related Rates Story-Problems

287

PSfrag repla ements

x
Figure 9.13.

Street light, pedestrian, and shadow


s
x+s
=
.
h
H

Di erentiate with respe t to time,


x + s
s
=
,
h
H

or, after a little bit of algebra, and again normalizing to x = 1,


s =

h
1/H
=
.
1/h 1/H
Hh

The shadow-length is in reasing at a onstant rate.


Example 9.4.3. A hild is ying a kite at onstant height. Wind is blowing the kite horizontally at onstant speed. At what speed is string playing
out through the hild's hand?

Let h denote the height of the kite, let x denote the horizontal distan e
from the kite to the hild, and let s denote the length of string from the hild
to the kite. Unrealisti ally idealize the string as a line segment, so that
s2 = x 2 + h 2 .

The situation is depi ted in gure 9.14. Take time-derivatives,


2ss = 2xx .

Normalizing to x = 1, we have

288

9 Theory and Appli ations of the Derivative


x

PSfrag repla ements

Figure 9.14.

s =

Kite

x
x
=
.
2
s
x + h2

Thus s = 0 when the kite is dire tly over the hild's head. Also, in the limit
as ever more string is played out, s tends to 1; this is sensible sin e the
proportions of the triangle degenerate toward s = x in the limit.
Example 9.4.4. A rope is suspended over a pulley at height y. A weight
is atta hed to one end of the pulley, and the other end of the rope is
being pulled horizontally away from beneath the pulley at onstant rate,
lifting the weight. At what rate is the pulley rising?

Let x denote the horizontal distan e from point on the oor beneath the
pulley to the end of the rope that is being pulled. Let h (for hypotenuse)
denote the length of rope from the pulley to the end being pulled. Then
x2 + y2 = h2 .

The situation is depi ted in gure 9.15. The weight is rising at the rate that
rope is passing over the pulley, and rope is passing over the pulley at rate h ,
so we want to nd h . Di erentiate the previous relation,
2xx = 2hh .

And so, normalizing to x = 1,


h =

x
x
.
= p
2
h
x + y2

9.4 Related Rates Story-Problems

289

PSfrag repla ements


x
Figure 9.15.

Weight and pulley

The weight is initially rising at speed 0, and (assuming that the pulled end
of the rope was initially
the pulley when
beneath the pulley) it has rea hed
h = 2y and thus x = 3 y. The weight is rising at speed 3/2 when it rea hes
the pulley.
Example 9.4.5. Water is being drained from
and radius R at onstant rate. At what rate is

the tank hanging?

oni al tank of height H


the height of the water in

Let h denote the height of the water in the tank, and r the radius. At any
instant, the water in the tank forms a one similar to the tank, so that
r
R
= ,
h
H

and therefore
r = (R/H)h.

The situation is depi ted in gure 9.16. The volume of water in the tank is
V=

Therefore,

1
1 2
r h = (R/H)2 h3 .
3
3
V = (R/H)2 h2 h .

But V = 1, and so
h =

1
.
(R/H)2 h2

290

9 Theory and Appli ations of the Derivative


R

r
H

PSfrag repla ements

Figure 9.16.

Half-pro le of oni al tank

At the beginning of the pro ess, when h = H, we have h = 1/(R2 ), and at


the end, when h = 0, apparently h is instantaneously in nite.
Exercises
9.4.1. A spe tator at a tennis mat h is sitting netside. The ourt has length
and width w. A player standing at the middle of the baseline hits the ball
perfe tly horizontally, giving it velo ity v. At what rate is the spe tator's head

swiveling as the ball moves?

9.4.2. A Ferris wheel has radius R, and it rotates at rate . It is nighttime.

A lantern is suspended immediately above the top of the Ferris wheel. As you
ride the Ferris wheel, at what rate is your shadow moving when you are at
angle t from the top?

9.4.3. Point p moves along the x-axis at rate a, and point q moves along the
y-axis at rate b. At what rate does the distan e between them hange?
9.4.4. A sphere of radius r has volume V = 4r3 /3 and area A = 4r2 . If an

evaporating spheri al drop of water is losing volume at a rate proportional to


its area, show that it is losing area at a rate proportional to its radius.
9.4.5. A balloon rises verti ally at onstant rate, and ar travels horizontally

at a di erent onstant rate. How is the distan e between them hanging?

9.4 Related Rates Story-Problems

291

9.4.6. One end of a rope is atta hed to the bow of a boat. The other end of
the rope passes through a ring on a do k, distan e h higher than the bow, at

onstant rate. At what rate does the bow move toward the do k?

10
Integration via Antidifferentiation

Ea h time that we have integrated a fun tion f in these notes, the result was
the di eren e of two values of a se ond fun tion F whose derivative was f. The
Fundamental Theorem of Cal ulus says that this phenomenon is general: If f
is ontinuous on [a, b] and F = f then
Zb

f = F(b) F(a).

Thus the problem of integrating f is solved whenever we an antidi erentiate f, i.e., whenever we an nd a fun tion F whose derivative is f. This hapter
establishes the Fundamental Theorem of Cal ulus in se tion 10.1 and then
lays out some antidi erentiation te hniques. Some basi antiderivatives are
given in se tion 10.2. Se tion 10.3 explains how to nd ertain antiderivatives
by a pro ess alled forward substitution, and se tion 10.4 explains a related
pro ess alled inverse substitution. Se tion 10.5 presents a useful te hnique
alled antidi erentiation by parts.

10.1 The Fundamental Theorem of Calculus


10.1.1 Indefinite Integrals, Antiderivatives

To make the ideas lear, we begin by naming some of the phenomena just
dis ussed in the hapter introdu tion.
Definition 10.1.1 (Indefinite Integral).
in R, and onsider a fun tion
f : I R

Let

be a nonempty interval

294

10 Integration via Antidi erentiation


R

su h that ab f exists for all


se ond fun tion
su h that

a, b I.

An

indefinite integral of f

is a

F : I R
Zb

f = F(b) F(a)

for all a, b I.

Various inde nite integrals that we have found during the ourse of these
notes are shown in gure 10.1.
f

f ( 6= 1) on R>0

f+1 /( + 1)

f1 on R>0

ln

ln

f1 ln f1

exp

exp

os

sin

sin

os

Figure 10.1.

Inde nite integrals

Proposition 10.1.2 (Indefinite Integral Properties). Let I be a nonempty interval in R, let


f, g : I R
Rb
R
be fun tions su h that a f and ab g exist for all a, b I, and let c R

be a onstant.
(a) Suppose that F, G : I R are respe tively inde nite integrals of f
and g. Then F + G is an inde nite integral of f + g and cF is an
inde nite integral of cf.
(b) Suppose that F : I R is an inde nite integral of f. Then eF : I R
is also an inde nite integral of f if and only if eF = F + C for some
onstant C.
Proof. Exer ise 10.1.1(a).

10.1 The Fundamental Theorem of Cal ulus


Definition 10.1.3 (Antiderivative).

and onsider a fun tion

An antiderivative

295

Let I be a nonempty interval in R,

f : I R.

of f

is a se ond fun tion


F : I R

su h that

F = f.

Proposition 10.1.4 (Antiderivative Properties).


interval in R, let
f, g : I R

Let I be a nonempty

be fun tions, and let c R be a onstant.


(a) Suppose that F, G : I R are respe tively antiderivatives of f and g.
Then F + G is an antiderivative of f + g and cF is an antiderivative
of cf.
(b) Suppose that F : I R is an antiderivative of f. Then eF : I R is
also an antiderivative of f if and only if eF = F+C for some onstant C.
Proof. Exer ise 10.1.2(a).

Ea h time that we found an inde nite integral F for a fun tion f, the
inde nite integral was also an antiderivative of f, as shown in gure 10.2. The
Fundamental Theorem of Cal ulus asserts that under reasonable onditions,
this will always be so. That is, the Fundamental Theorem says that:

Under suitable onditions, integration redu es to antidi erentiation.


With this slogan in mind, we are motivated to study antidi erentiation, whi h
is not innately of interest, as a means to integration, whi h is.
Exercises
10.1.1. (a) Prove Proposition 10.1.2.
(b) Suppose that F, G : I R are respe tively inde nite integrals of f
and g. Need the produ t FG be an inde nite integral of the produ t fg?
10.1.2. (a) Prove Proposition 10.1.4.
(b) Suppose that F, G : I R are respe tively antiderivatives of f and g.
Need the produ t FG be an antiderivative of the produ t fg?

296

10 Integration via Antidi erentiation


f

f ( 6= 1) on R>0

f+1 /( + 1)

f1 on R>0

ln

f1

ln

f1 ln f1

ln

exp

exp

exp

os

sin

os

sin

os

sin

Figure 10.2.

Inde nite integrals and their derivatives

10.1.2 The Fundamental Theorem, Part I

The Fundamental Theorem of Cal ulus is really two theorems, ea h of whi h


des ribes a sense in whi h di erentiation and integration are inverse operations. The se ond of the two theorems is the one that relates integration and
antidi erentiation, but we naturally begin with the rst.
Theorem 10.1.5 (Fundamental Theorem of Calculus, Part I). Let I
be a nonempty interval in R, and let a be a point of I. Let the fun tion
f : I R be ontinuous. De ne a fun tion
Zx
f.
F : I R,
F(x) =
a

Then F is di erentiable on I and

F = f.

Thus the rst part of the Fundamental Theorem says loosely that di erentiation inverts integration, in that the derivative of the integral up to a
variable endpoint is the original fun tion. Geometri ally, the idea is that:

The rate at whi h the area of the region under a urve grows as an
endpoint moves is the height of the urve over the moving point.
R

Note that sin e the logarithm is de ned as ln(x) = 1x f1 , our al ulation in


hapter 5 of the derivative ln = f1 amounted to proving a spe ial ase of
Theorem 10.1.5.
A point to observe here is that Theorem 10.1.5 says that every ontinuous
fun tion on an interval has an antiderivative. We may not be able to write

10.1 The Fundamental Theorem of Cal ulus

297

the antiderivative without


an integral sign (i.e., we may not be able to nd a
R
ni e expression for ax f), but nonetheless its derivative is f. For example, the
fun tion
2
2
f(x) = ex

f : R R,

has as an antiderivative the error fun tion,




1 1
2 2

erf : R ,

erf (x) =

Zx

f,

even though we an't simplify the formula for erf. Like the logarithm, the error
fun tion, de ned as an integral, needs its own name. It is part of the area under
a bell-shaped
urve that des ribes many limiting behaviors in probability. The

2/ normalizes the bell urve so that the total area beneath it is 1. (See
gure 10.3.)

y=

2
2 ex

PSfrag repla ements


x
Figure 10.3.

erf (x): area under a bell-shaped urve

The hypothesis in Theorem 10.1.5 that f is ontinuous warrants a qui k


remark. As we have dis ussed, every ontinuous fun tion f : [a, x] R (or
f : [x, a] R if x < a) is integrable, but the proof of this fa t is te hni al and so we omitted it. The fun tions that we know to be integrable over
losed, bounded intervals are the (not ne essarily ontinuous) bounded, pie ewise monotoni fun tions. We will state a weaker version of Theorem 10.1.5,
hypothesizing a ontinuous su h fun tion f rather than any ontinuous f, at
the end of this se tion.

Proof (of Theorem 10.1.5). Let x and s be any points of I with x < s. By
the Extreme Value Theorem, f on [x, s] has a minimum f(xm ) and a maximum

298

10 Integration via Antidi erentiation

f(xM ), although we don't know xm and xM . That is,


f(xm ) f(t) f(xM ) for all t [x, s].

By the Inequality Rule for integrals (Proposition 5.5.3, page 178), it follows
that
Zs
(s x)f(xm )

or, equivalently,
R

f(xm )

f (s x)f(xM ),

Rs

xf
f(xM ).
sx

That is, ( xs f)/(s x) is an intermediate value of f. Consequently, by the


Intermediate Value Theorem,
Rs

xf
= f(c)
sx

for some c between x and s.

(10.1)

If instead s < x then (10.1) still holds by a typi al veri ation of symboli
robustness (exer ise 10.1.4). Sin e f is ontinuous and c lies between x and s,
we have
Rs
f
= lim f(c) = f(x).
s x s x
s x

lim

Re all the fun tion in the statement of the theorem,


F : I R,

For any distin t points x, s I,


F(s) F(x)
=
sx

Rs

F(x) =

Zx

f.

Rx
Rs
f af
f
= x ,
sx
sx

and so we have shown that for all x I,


lim

s x

F(s) F(x)
= f(x).
sx

In other words, F (x) exists and equals f(x) for all x I. This is the desired

result.

To summarize, the argument just given is that integrating the Extreme


Value Theorem and then iting the Intermediate Value Theorem shows that
the derivative of the fun tion given by integration from a xed endpoint to a
variable endpoint is the original fun tion at the variable endpoint.
As mentioned earlier, the fun tions that we know to be integrable are the
bounded pie ewise monotoni fun tions. The variant of Theorem 10.1.5 that
assumes that f is a pie ewise monotoni su h fun tion is

10.1 The Fundamental Theorem of Cal ulus

299

Theorem 10.1.6 (FTC I, Weaker Variant). Let I be a nonempty interval in R, and let a be a point of I. Let the fun tion f : I R be

ontinuous and pie ewise monotoni . De ne a fun tion


F : I R,

F(x) =

Zx

f.

Then F is di erentiable on I and F = f.

This theorem an be proved with no referen e to the Extreme Value Theorem or the Intermediate Value Theorem (exer ise 10.1.5).
Example 10.1.7. Consider the fun tions
f : R R,

f(x) =

1
1 + x2

and
F : R R,

F(x) =

Zx

f.

By Theorem 10.1.6, F = f. But also the fun tion ar tan : R R has


derivative f, so that F(x) = ar tan(x) + C for some onstant C. Substitute
x = 0 to get C = 0. That is, using the notation introdu ed in De nition 8.2.1
(page 237),
Z
ar tan(x) =

x1 =0

1
,
1 + x21

(10.2)

x R.

This formula is similar to the de ning formula for the logarithm on page 241,
ln(1 + x) =

Zx

x1 =0

1
,
1 + x1

x > 1.

Just as the logarithm formula led in se tion 8.3 to an expression for ln(1 + x)
as a limit of polynomials when 1 < x 1, formula (10.2) leads to su h an
expression for ar tan(x) (exer ise 10.1.7),
ar tan(x) = x

x2n+1
x3 x5
+
+ (1)n
+ ,
3
5
2n + 1

1 x 1,

or
ar tan(x) =

n=0

(1)n

x2n+1
,
2n + 1

1 x 1.

In parti ular, it gives the lovely formula


1 1 1 1
1
1
1

=1 + +
+

+ ,
4
3 5 7 9 11 13 15

similar to the formula for ln(2) on page 242.

300

10 Integration via Antidi erentiation

Exercises
10.1.3. De ne a fun tion
f : R [0, 1],

f(x) =

0 if x < 0,
1 if x 0.

This fun tion is bounded and monotoni , and so it is integrable over any
interval [a, b]. De ne a se ond fun tion
F : R R,

Zx

F(x) =

f.

Is F di erentiable? Does this ontradi t Theorem 10.1.5 or Theorem 10.1.6?


10.1.4. Establish (10.1) when s < x.
10.1.5. (a) Explain why to prove Theorem 10.1.6, it su es to prove the
theorem but with the stronger hypothesis that the ontinuous fun tion f is

monotoni rather than only pie ewise monotoni . (To make things simpler,
assume that the pie ewise monotoni ity involves only two pie es.)
(b) Assume that f is in reasing. Show that with this assumption, Theorem 10.1.6 an be proved with no referen e to the Extreme Value Theorem
or the Intermediate Value Theorem.
10.1.6. Let f : R R be ontinuous. Consider the fun tion
F : R R,

F(x) =

Z x2

f.

Theorem 10.1.5 does not immediately apply here, be ause the upper limit of
integration in the integral that de nes F is not x itself, but a fun tion of x.
(a) De ne two fun tions,

and

Zx

e
F : R R,

e
F(x) =

g : R R,

g(x) = x2 .

Explain why F is a omposition (whi h?) of eF and g.


(b) Use the Chain Rule and Theorem 10.1.5 to di erentiate F.
( ) Let g, h : R R be di erentiable fun tions, and onsider the fun tion
Z
h(x)

G : R R,

G(x) =

f.

g(x)

10.1 The Fundamental Theorem of Cal ulus

301

Explain why G is di erentiable, and ompute G . (The new wrinkle here is


that now both endpoints of integration vary. Your solution should redu e the
situation ba k to appli ations of the ase of one variable endpoint, and then
use the Chain Rule for those ases as in part (b).)
10.1.7. Similarly to the analysis of the logarithm in se tion 8.3, obtain the
boxed formula for ar tan(x) given in the se tion.
10.1.3 The Fundamental Theorem, Part II

Whereas Part I of the Fundamental Theorem is a statement about the derivative of the integral, our rst statement of Part II is a statement about the
integral of the derivative.
Theorem 10.1.8 (Fundamental Theorem of Calculus, Part II). Let I
be a nonempty interval in R. Suppose that the fun tion F : I R is
di erentiable and that F is ontinuous. Let f = F . Then for any points
a and b of I,
Zb
f = F(b) F(a).
a

Thus the se ond part of the Fundamental Theorem says loosely that integration inverts di erentiation, in that the integral of the derivative is the
di eren e of the original fun tion's values at the two endpoints. That is:

The integral of the rate of hange of a fun tion is the net hange
in the fun tion.
Proof. De ne

Rx

f. Then e
F = f by Part I of the
Fundamental Theorem, and so, sin e also F = f, there exists a onstant C

su h that

e
F : I R by e
F(x) =

e
F(x) = F(x) + C for all x I.

(10.3)

Substitute x = a in (10.3) to get 0 = F(a)+C, i.e.,


Rb C = F(a). Next substitute
F(b) = F(b) F(a). Sin e e
F(b) = a f by de nition, the proof is
x = b to get e

omplete.

The proof just given that Part II of the Fundamental Theorem follows from
Part I is essentially instant, but also Theorem 10.1.8 an be proved dire tly
without referen e to Theorem 10.1.5. Here is one version of the argument.
Fix points a, b I with a < b. Choose any partition points
a = x0 < x1 < < xn1 < xn = b.

302

10 Integration via Antidi erentiation

For i = 1, , n, restri t the domain of F to the ith subinterval [xi1 , xi ], and


let [xi1 , xi ] and F play the roles of [a, b] and f in the Mean Value Theorem.
Then the theorem says that
F(xi ) F(xi1 ) = (xi xi1 )F (ci ) for some ci (xi1 , xi ).

But we have de ned f = F , so that previous display rewrites as


F(xi ) F(xi1 ) = (xi xi1 )f(ci ) for some ci (xi1 , xi ).

Let Mi be any number at least as big as all f-values on [xi1 , xi ]. Then the
previous equality implies the inequality
F(xi ) F(xi1 ) (xi xi1 )Mi .

Summing the left sides of this inequality for i = 1, , n gives


(F(x1 )F(x0 ))+(F(x2 )F(x1 ))+ +(F(xn1 )F(xn2 )+(F(xn )F(xn1 )),

whi h teles opes to F(b) F(a). On the other hand, summing the right sides
of the inequality for i = 1, , n gives
(x1 x0 )M1 + (x2 x1 )M2 + + (xn1 xn2 )Mn1 + (xn xn1 )Mn .

This is an upper sum for Arba (f), and sin e the number n of partition points
is arbitrary, as are the partition points xi and the values Mi that ex eed f
on [xi1 , xi ], the upper sum is utterly general. So we have shown that
for any upper sum T for Arba (f).

F(b) F(a) T

Some sequen e of upper sums onverges to


F(b) F(a)

Similarly

Rb
a

Rb
a

Zb

, and so it follows that


f.

f F(b) F(a), and so we are done,


Zb

f = F(b) F(a).

So far the argument has assumed that f 0, but extending it to general ontinuous f is just a matter of passing it through the usual hoisting ritual. The
key idea here is that loosely speaking, Part II of the Fundamental Theorem
omes from the Mean Value Theorem. Note that the rst proof that we gave
of Part II ta itly used the Mean Value Theorem as well, along with quoting

10.1 The Fundamental Theorem of Cal ulus

303

Part I. Unlike Part I, Part II an not be proved without re ourse to an abstra t existen e theorem even under simplifying onditions su h as assuming
that f is monotoni . There is a on eptual reason for this: Whereas Part I
zooms in to measure the lo al rate of hange of a quantity arising from larges ale synthesis (the total area of a region), Part II pulls the amera ba k to
make an assertion about a quantity arising from large-s ale synthesis of lo al
information.
Introdu ing a little more notation will larify how Part II of the Fundamental Theorem of Cal ulus sometimes redu es integration to antidi erentiation.
Definition 10.1.9 (Antiderivative Notation). Let I be
terval in R, and let f : I R be a fun tion. Then
Z
f denotes any antiderivative of f.

a nonempty in-

This de nition needs to be parsed arefully. Re all that an antiderivative


of f is a fun tion whose derivative is f, even though the notation just introdu ed for an antiderivative is very similar Rto that for an integral. The di eren e is that in the antiderivative notation f, the integral sign is bare rather
Rb
than adorned
R by limits of integration as it is in the integral notation a f.
Note that f is a fun tion, any of a family of fun tions di ering from ea h
R
other by onstants, whereas ab f is a number.
Introdu e one more pie e of notation:
Definition 10.1.10 (Notation for Difference of Function-Values at
Two Points). For any fun tion F whose domain in ludes the points a
and b,
b

F is short for F(b) F(a).
a

With all of this notation in pla e, Part II of the Fundamental Theorem of


Cal ulus an be rephrased.
Theorem 10.1.11 (Fundamental Theorem of Calculus, Part II, Rein R. Suppose that the fun tion
phrased). Let I be a nonempty interval
R
f : I R is ontinuous, and that f is an antiderivative of f. Then for
any points a and b of I,
Z 
Zb
b
f=
f .
a

That is, any antiderivative of f is an inde nite integral of f.

The last statement in the theorem is why the antiderivative notation was
hosen to resemble an integral. To rephrase the theorem one more time, the
integral of f from a to b is the di eren e of the antiderivative values at the
endpoints. And hen e, to integrate it su es to antidi erentiate.

304

10 Integration via Antidi erentiation

Exercises
10.1.8. Let I be a nonempty interval in R, let n be a nonnegative integer,

and let

f : I R

have n + 1 ontinuous derivatives. (This means that the fun tion f(0) = f,
the derivative f(1) = f , the se ond derivative f(2) = f , and so on up to the
(n + 1)st derivative f(n+1) exist and are ontinuous on I.) Let a and x be
points of I.
(a) Explain why
Z
x

f (x1 ).

f(x) = f(a) +

x1 =a

(b) Assuming that n 1, explain why

f (x1 ) = f (a) +

Z x1

f (x2 ),

x2 =a

and therefore
f(x) = f(a) + f (a)(x a) +

Zx

Z x1

f (x2 ).

x1 =a x2 =a

( ) Continue in this vein to explain why, if n 2,


f (a)
f(x) = f(a) + f (a)(x a) +
(x a)2
2
Zx
Z x1 Z x2
f (x3 ),
+
x1 =a x2 =a x3 =a

(d) Explain why in general,


f(x) = Pn (x) + Rn (x),

where the degree n approximating polynomial is


Pn (x) = f(a) + f (a)(x a) +
=

f(n) (a)
f (a)
(x a)2 + +
(x a)n
2
n!

n
X
f(k) (a)
(x a)k ,
k!

k=0

and the orresponding remainder is an (n + 1)-fold iterated integral,


Rn (x) =

Zx

x1 =a

Z xn

f(n+1) (xn+1 ).

xn+1 =a

Note that this exer ise uniformizes mu h of the work in hapter 8.

10.2 Basi Antidi erentiation

305

10.2 Basic Antidifferentiation


We expand the antiderivative notation introdu ed a moment ago.
Definition 10.2.1 (Antiderivative Notation With a Variable). Let I
be a nonempty interval in R, and let f : I R be a fun tion. Then,
letting x denote a variable,
Z
f(x) dx denotes any antiderivative of f, viewed as a fun tion of x.

Every di erentiation formula gives rise to an antidi erentiation formula.


Some antidi erentiation formulas arising from the derivatives that we have
omputed during the ourse of these notes are shown in gure 10.4, using the
notation just introdu ed. In ea h formula, the right side is one antiderivative
of the left side, and the general antiderivative is the given spe i one plus an
arbitrary onstant. For example, the rst formula in the table perhaps should
say instead,
Z
x dx =

x+1
+ C,
+1

6= 1.

But to keep the notation lean, we omit the \+C" throughout the table, remembering to in orporate it as ne essary when we ompute. You should know
the formulas in the table ba kward and forward.
Example 10.2.2. To al ulate the antiderivative
Z
x(x3 + 1)3 dx,

use the Finite Binomial Theorem to expand (x3 + 1)3 ,


Z

x(x3 + 1)3 dx =


x (x3 )3 + 3(x3 )2 + 3(x3 ) + 1 dx

Z
= (x10 + 3x7 + 3x4 + x) dx
=

x2
x11 3x8 3x5
+
+
+
+ C.
11
8
5
2

Example 10.2.3. The antiderivative


Z

s 2 (3x) dx

isn't quite to beRfound in the gure 10.4 table be ause of the 3x. Nonetheless,
the table entry s 2 (x) dx = ot(x) suggests that a natural starting guess
is ot(3x). By the Chain Rule, the derivative of ot(3x) is 3 s 2 (3x), and
sin e onstants pass through di erentiation it follows that

306

10 Integration via Antidi erentiation


Z
Z
Z
Z
Z

x+1
,
x dx =
+1
Z
1
dx = ln(|x|),
x

6= 1,

ln(x) dx = x ln(x) x,
Z

ex dx = ex ,

os(x) dx = sin(x),
sin(x) dx = os(x),

se 2 (x) dx = tan(x),
s 2 (x) dx = ot(x),

tan(x) se (x) dx = se (x),


ot(x) s (x) dx = s (x),
Z

1
dx = ar tan(x),
1 + x2
Z
1

dx = ar sin(x),
1 x2
Z
se (x) dx = ln(| tan(x) + se (x)|),
Z
s (x) dx = ln(| ot(x) + s (x)|).

Figure 10.4.

Basi antidi erentiation formulas


1
3

s 2 (3x) dx = ot(3x) + C.

Example 10.2.4. The antiderivative


Z
1
dx (where a > 0 is onstant)
2
a + x2
R 1
looks similar to the table entry 1+x
2 dx = ar tan(x). To make the fun tion

whose antiderivative we want look more like the fun tion in the table entry,
rewrite it,

10.3 Antidi erentiation by Forward Substitution


a2

307

1
1
1
= 2
.
2
+x
a 1 + (x/a)2

Thus the natural starting guess for our antiderivative is ar tan(x/a). By the
Chain Rule, the derivative of ar tan(x/a) is 1/(1+(x/a)2 )(1/a), and so sin e
onstants pass through derivatives it follows that
Z

x
1
1
ar tan
+ C.
dx
=
a 2 + x2
a
a

Exercises
10.2.1. Verify all of the formulas in gure 10.4 by lo ating the text or exer ise

in these notes that di erentiates ea h quantity on the right side.

10.2.2.ZFind the following antiderivatives.

(a) x(1 + 3 x) dx.


Z 2x
e + e3x
dx.
(b)
e4x
Z

( ) se (2x + 3) dx.
(d)

1
dx where a > 0 is onstant.
x2

a2

10.3 Antidifferentiation by Forward Substitution


Sin e the emphasis will now be on al ulating, we revert to a less formal
writing style. All fun tions are assumed to be integrable and/or di erentiable
as ne essary.
10.3.1 The Forward Substitution Formula

The forward substitution formula is


Z

(f g) g =

Z 
f g.

(10.4)

The natural rst response to this formula is to have no idea what it says,
mu h less how to use it. We will ta kle these matters one at a time.

308

10 Integration via Antidi erentiation

10.3.2 What the Formula Says and Why It Is True

Formula (10.4) says:


R

If f is an antiderivative of f then the omposition


turn an antiderivative of (f g) g .

That is:

R
f g

is in

Finding an antiderivative of the more ompli ated fun tion (fg)


g redu es to nding an antiderivative of the simpler fun tion f.
R

To establish the formula, it su es to show


R  that given an antiderivative f
of f, the derivative of the omposition
R f g is (f g) g . Compute, using
the Chain Rule and the fa t that ( f) = f by de nition, that the derivative
of the omposition is indeed
Z 

f g =

!
Z 
f g g = (f g) g .

This proof is so simple be ause the forward substitution formula (10.4) uses
the variable-free notation for fun tions, the un luttered notation that is well
suited to arguments.
10.3.3 Using the Formula in its Variable-Free Form

A tual al ulational examples involve fun tions-as-formulas, so that their notation and the variable-free notation are at odds. Here is an example of translating the with-variable notation (as in De nition 10.2.1) of an antiderivative
problem into the variable-free notation of formula (10.4).
Example 10.3.1. To nd the antiderivative
Z
etan(x) se 2 (x) dx (this is a fun tion of the variable x),

introdu e (using variable-free notation)


f = exp and g = tan,

so that g = se 2 .

Then the antiderivative takes the desired form,


Z

Z
etan(x) se 2 (x) dx = (f g) g .

Here the notations are in on i t: the left side arries the information that
the variable of the antiderivative is to be named x, while the right side makes

10.3 Antidi erentiation by Forward Substitution

309

no referen e to the variable name. In using the variable-free notation, we


now must also remember that in the nal answer|a fun tion de ned by a
formula|the variable is to be named x. With this detail led away somewhere
in our memories, note that the forward substitution formula (10.4) says that
the antiderivative is instead
Z

But
so that

etan(x) se 2 (x) dx =
Z

Z 
f g.

Z
f = exp = exp + C,

Z 
f g = (exp + C) tan = exp tan + C.

That is, bringing the variable x ba k into the notation,


Z

etan(x) se 2 (x) dx = etan(x) + C.

10.3.4 Improvement: the Formula With Variables

It would be silly to keep working examples in the fashion of the previous


paragraph. Rather than translate every example into variable-free notation,
we should translate the one forward substitution formula into notation that
in orporates variables. The result is (exer ise 10.3.1)
Z

f(g(x))g (x) dx =

f(u) du

where u = g(x).

(10.5)

Example 10.3.2. Using the boxed formula, the previous example an be reworked more su in tly, again with f = exp and g = tan,
Z
Z
where u = tan(x)
etan(x) se 2 (x) dx = eu du
= eu + C

where u = tan(x)

= etan(x) + C.

The basi mnemoni for forward substitution is:

See something and its derivative.


In the previous example, the something was tan(x) and its derivative was
se 2 (x). That is, the something is the g(x) in the forward substitution formula.

310

10 Integration via Antidi erentiation

Example 10.3.3. Similarly, let a > 0 and onsider the antiderivative


Z p
x a2 x2 dx.

Here the something is a2 x2 , and its derivative is 2x. The problem doesn't
quite give us the derivative, but the imperfe t t is easy to x sin e onstants
pass through antidi erentiation,
Z p
Z
p
1
2
2
x a x dx = (2x) a2 x2 dx.
2

And so by forward substitution,

Z p
Z
p
1
x a2 x2 dx = (2x) a2 x2 dx
2
Z
1
=
g (x)f(g(x)) dx
2
Z
1
f(u) du
=
where u = g(x)
2
Z
1
u du
where u = a2 x2
=
2
1 2
= u3/2 + C
where u = a2 x2
2 3
1
= (a2 x2 )3/2 + C.
3
Exercise
10.3.1. Explain arefully how ea h side of the forward substitution formula

with variables (10.5) arises from its ounterpart in the variable-free forward
substitution formula (10.4).

10.3.5 Second Improvement: the Procedure Instead of the Formula

Forward substitution is easier in pra ti e if it is viewed as a pro edure rather


than a formula, whether the formula has variables or not. The pro edure is
abetted by a pie e of notation due to Leibniz.
Definition 10.3.4 (Leibniz Notation for the Derivative).

di erentiable fun tion of x. Then


du
dx

is a synonym for u (x).

Let

be a

10.3 Antidi erentiation by Forward Substitution

311

Example 10.3.5. Returning to the antiderivative


Z p
x a2 x2 dx,

the pro edure is to make the substitution

(the something).

u = a 2 x2

Then

du
= 2x
dx

(its derivative),

so that
du = 2x dx,

and onsequently

1
x dx = du.
2

Therefore the antiderivative is


Z p
Z
1
x a2 x2 dx =
u du (substituting)
2
1 2
= u3/2 + C
2 3
1
= (a2 x2 )3/2 + C.
3

This pro edure works, and it has been learned by generations of al ulus
students. But it is not as self-evident as it appears. The problem is that the

Leibniz notation for the derivative,

du
,
dx

is a single, indivisible symbol, while the separate notations dx and du


have not been given meanings at all unless they o ur in onjun tion
with the integral sign as part of the antiderivative notation. So the idea

that in general

\if u = g(x) then du = g (x) dx"

is in isolation a meaningless statement, mu h less a valid argument. But the


Leibniz notation relentlessly suggests it as a valid ritual to pra ti e during the
ourse of antidi erentiation, and the notation has been designed so that the
ritual is valid. In pro edural terms, the mnemoni for forward substitution is:

See

and du.

312

10 Integration via Antidi erentiation

Example 10.3.6. Consider the antiderivative


Z
Z
sin x
tan(x) dx =
dx.
os x

Make the substitution u = os x. Then du = sin x dx, and so


Z

du
sin x
dx =
= ln(|u|) + C = ln(| os x|) + C.
os x
u

To summarize, the variable-free formulation is the right environment for


proving the forward substitution formula, and the algorithm is the right environment for applying it.
10.3.6 Basic Forward Substitution Formulas

Every basi antidi erentiation formula ombines with the Chain Rule to give
rise to a forward substitution antidi erentiation formula. These are shown in
gure 10.5. As in gure 10.4, a \+C" is ta it in the right side of ea h formula.
R
Example 10.3.7. We will al ulate x2 se 2 (x3 + 1) dx. Compute that
Z
Z
1
se 2 (g(x))g (x) dx where g(x) = x3 + 1
x2 se 2 (x3 + 1) dx =
3
1
= tan(x3 + 1) + C,
3

by the seventh formula in table 10.5.


R 2
Example 10.3.8. We will al ulate tet dt. The antiderivative is
Z
Z
1
t2
te dt =
g (t)eg(t) dt where g(t) = t2
2
1 2
= et + C.
2
Exercise
10.3.2.ZFind the following antiderivatives.
(a) ex sin(ex ) dx.
Z
sin(x)
dx.
(b)
sin
( os(x))
Z
( ) (3w4 + w)2 (12w3 + 1) dw.
Z
1
dx.
(d) p
x 1 (ln(x))2

10.3 Antidi erentiation by Forward Substitution


Z
Z
Z
Z
Z

(g(x))+1
,
(g(x)) g (x) dx =
+1
Z
g (x)
dx = ln(|g(x)|),
g(x)

ln(g(x))g (x) dx = g(x) ln(g(x)) g(x),


Z

eg(x) g (x) dx = eg(x) ,

os(g(x)) g (x) dx = sin(g(x)),


sin(g(x)) g (x) dx = os(g(x)),

se 2 (g(x)) g (x) dx = tan(g(x)),


s 2 (g(x)) g (x) dx = ot(g(x)),

se (g(x)) tan(g(x)) g (x) dx = se (g(x)),


s (g(x)) ot(g(x)) g (x) dx = s (g(x)),
Z

Z
Z

g (x)
dx = ar tan(g(x)),
1 + g2 (x)
Z
g (x)
p
dx = ar sin(g(x)),
1 g2 (x)

se (g(x)) g (x) dx = ln(| se (g(x)) + tan(g(x))|),


s (g(x)) g (x) dx = ln(| s (g(x)) + ot(g(x))|).

Figure 10.5.

x
dx.
2
Z 1+x
2
dw.
(f)
2
Z1+w

(e)

6= 1,

(g) sin3 (x) dx.


Z

(h) sin4 (x) dx.


Z
sin(ln(x))
dx.
(i)
x

Basi forward substitution formulas

313

314

10 Integration via Antidi erentiation

(j)

os(tan( x)) se 2 ( x)

dx.

10.3.7 Forward Substitution in Integrals

The forward substitution formula for integrals (as ompared to antiderivatives) is


Zb
a

(f g) g =

Z g(b)

(10.6)

f.

g(a)

This formula follows from its ounterpart (10.4) for antiderivatives and Part II
of the Fundamental Theorem of Cal ulus (Theorem 10.1.8),
Zb

Z
b

(f g) g = (f g) g
a
a
Z 
b

=
f g
a
Z 
g(b)
=
f
g(a)

Z g(b)

by Theorem 10.1.8
by (10.4)
by de nition of omposition
by Theorem 10.1.8 again.

g(a)

The following notation is well suited to hange of variable integral al ulations.


Definition 10.3.9 (New Notation for the Integral).
integrable from a to b then
Zb
Zb
f(x) dx is a synonym for
f.
x=a

If a fun tion f is

As explained after De nition 8.2.1 (page 237), the dummy variable x in


this notation an be repla ed by any other symbol not already in use. Using
this notation, formula (10.6) is
Zb

x=a

f(g(x))g (x) dx =

Z g(b)

u=g(a)

Example 10.3.10. To evaluate the integral


Ze
(ln x)2
dx,
x
x=1

f(u) du.

(10.7)

let u = ln(x). Then du = dx/x. Also if x = 1 then u = 0, and if x = e then


u = 1. Thus, by a straightforward appli ation of (10.7),

10.3 Antidi erentiation by Forward Substitution


Ze

x=1

Z1

(ln x)2
dx =
x

u=0

Example 10.3.11. To evaluate

Z 42

sin( x)

u2 du =

315

u3 1 1
= .
3 0 3

dx,

let u = x. Then du = dx/(2 x), and so dx/ x = 2 du. Also, if x = 2 then


u = , and if x = 42 then u = 2. Thus
Z 42

sin( x)

dx = 2

Z 2

sin(u)du

2

= 2 os(u)

= 2( os(2) os())

= 2(1 + 1) = 4.

Example 10.3.12. Let a > 0 be a onstant. We will al ulate


Za
1
dx.
2 + x2
a
0

From example 10.2.4 (page 306), an antiderivative of 1/(a2 + x2 ) is


Z

x
1
1
ar tan
dx
=
.
a 2 + x2
a
a

So by Part II of the Fundamental Theorem of Cal ulus,


Za
0

 x  a
1
1
= 1 ar tan(1) = .
dx = ar tan
a 2 + x2
a
a 0
a
4a

(We did not use (10.7) for this example.)


Example 10.3.13. To evaluate
Z1

x=1

p
1 x2 dx,

Note that the upper half of the unit ir le is the graph of the fun tion f(x) =

1 x2 . Thus the integral is the area above the x-axis and below the upper
half of the unit ir le, i.e., it is /2. (We did not use (10.7) for this example.)
Example 10.3.14. To evaluate
Z /2
t=0

sin2 (t) dt,

316

10 Integration via Antidi erentiation

note that sin e sin(t) = os(/2 t), this integral is also


Z /2

os2 (t) dt.

t=0

(See gure 10.6.) Therefore, twi e the integral is


Z /2
t=0

Thus

(sin (t) + os (t)) dt =

Z /2

sin2 (t) dt =

Z /2

Z /2

1 dt = /2.

t=0

os2 (t) dt = /4.

t=0

t=0

(We did not use (10.7) for this example.)

1
y = os2 (t)

y = sin2 (t)

PSfrag repla ements

/2
Figure 10.6.

Same area under the graphs of os2 and sin2

Example 10.3.15. To evaluate


Z ln(3)

x= ln( 3)

ex

1
dx,
+ ex

note that this integral is


Z ln(3)

x= ln( 3)

1
dx =
x
e + ex

Z ln(3)

x= ln( 3)

Z 3

ex
dx
1 + (ex )2

1
du where u = ex
1 + u2
3


= ar tan(u) = = .
3
6
6
1/ 3
=

u=1/ 3

10.4 Antidi erentiation by Inverse Substitution

317

10.3.3. Find
the following integrals.
3
Z

(a)
x2 sin(x3 ) dx.
x=0
Z ln(e3 1)
ex
(b)
dx.
x
x=ln(e2 1) 1 + e
Z e/3
ln(3x)
( )
dx.
x
x=1/3
Z ln(ln(2))/ ln(2)
(d)
2x dx.
x=0

10.4 Antidifferentiation by Inverse Substitution


10.4.1 The Inverse Substitution Formula and Why It Is True

The forward substitution formula is a bit ontrived in that it works only for
antiderivatives of fun tions of the spe ial form (f g) g . That is, it works
only when u and du are present. The inverse substitution formula is more
general. In its variable-free form it says
Z

fg=

Z

fh

where h inverts g.

As with the initial presentation of the forward substitution formula, the meaning and the use of the inverse substitution formula are probably opaque at
rst glan e.
R

The formula says that


R if h inverts g and f h is an antiderivative of f h
then the omposition f h g is in turn an antiderivative of f g. That is:

Finding an antiderivative of the omposition f g redu es to nding an antiderivative of the produ t f h where h inverts g.

And pure symbol-pushing shows that the inverse formula follows from the
forward formula (10.4). The veri ation will be easy be ause we are using the
variable-free versions of the formulas. Indeed, re all that the forward formula
is
Z 
Z
(f g) g =

f g.

Substitute f g for f, and h for g, and ex hange the two sides of the equality.
This gives
Z

Z
f g h = (f g h) h ,

or, re alling that g h is the identity mapping, whi h does nothing,

318

10 Integration via Antidi erentiation


Z


Z
f g h = f h .

Next prepend g to both sides to get


Z


Z

fg hg =
f h g.

But h g also does nothing, and so the desired formula follows.


As with forward substitution, we now want to rewrite the inverse substitution formula with variables, and then redu e it to a pro edure.
10.4.2 The Formula With Variables

The formula with variables is


Z

f(g(x)) dx =

where h inverts g and u = g(x).

f(u)h (u) du

For example, onsider the antiderivative

This is

Z
Z

Z
dx = f(g(x)) dx

dx.

where f = exp and g = f1/2 .

(Here f1/2 is the square root fun tion as usual.) The inverse fun tion of g is
the squaring fun tion h = f2 , whose derivative is h = 2f1 where f1 is the
identity fun tion. Thus a ording to the inverse substitution formula,
Z

dx = 2 ueu du where u =
R

x.

A plausible rst guess for ueu du is ueu . This guess has derivative ueu +eu ,
so that orre ting it to ueu eu gives the orre t antiderivative. Therefore,
the original antiderivative is
Z


dx = 2 xe x 2e x + C.

10.4.3 The Procedure

The pro edure is as follows. To ompute the antiderivative


Z

dx,

10.4 Antidi erentiation by Inverse Substitution

make the substitution


u=

319

x.

Then the inverse substitution is


x = u2 ,

and so
dx = 2u du.

Therefore the antiderivative is


Z

where u = x.

dx = 2 eu u du

And from here things go as before.


The inverse substitution pro edure di ers from the forward substitution
pro edure in that for inverse substitution, after determining the substitution
u = g(x), we indeed invert it by nding x = h(u) and then express dx in terms
of u and du, i.e., dx = h (u) du. Then we substitute u and dx. Unlike forward
substitution, this doesn't require the problem to ontain both u and du.
Inverse substitution doesn't really have a mnemoni ounterpart to the
see u and du slogan for forward substitution. The idea is to hoose some g
to make the fun tion whose antiderivative we want have the form f(g(x)), and
let u = g(x). Invert g and di erentiate the inverse h. Then the new fun tion
to antidi erentiate is f(u)h (u). If this is easier, then the inverse substitution
has helped. But there are no general rules for hoosing g well. A promisinglooking inverse substitution an lead nowhere, and an outlandish-looking one
an render a problem trivial. The losest thing to a mnemoni for inverse
substitution is:

Choose u and express

dx

in terms of u and du.

Example 10.4.1. For a promising-looking inverse substitution that does no

good, onsider an antiderivative losely related to the error fun tion introdu ed on page 297,
Z
2

ex dx.

A plausible hoi e is u = x2 . Then x = u, and so dx = du/(2 u). Thus the


antiderivative is
Z
1 eu
du where u = x2 ,
2

but this is no better than what we started with. In fa t, no amount of substitution will allow us to express this antiderivative in terms of fun tions that
have been studied in this ourse. It has no su h expression.

320

10 Integration via Antidi erentiation

Example 10.4.2. For an unpromising-looking inverse substitution that does


good, onsider the antiderivative
Z

dx
p
.
1+ x
p

With re kless abandon, let u = 1 + x. Now invert: u2 = 1 + x, and so


x = (u2 1)2 . It follows that
dx = 2(u2 1) 2u du,

and so
Z

Z
(u2 1)u du
= 4 (u2 1) du
u

 3

4
u
u + C = (1 + x)3/2 4(1 + x)1/2 + C.
=4
3
3

dx
p
=4
1+ x

Example 10.4.3. Sometimes a workable inverse substitution takes a little

algebra to nd. For instan e, the antiderivative


Z

an be rewritten as
Z

(1 x)2/5
dx =
x12/5

Z

(1 x)2/5
dx,
x12/5

1x
x

2/5

1
dx =
x2

Z

2/5
1
1
dx.
1
x
x2

Let u = 1/x 1. Then du = (1/x2 ) dx, and so we have


Z


7/5
Z
(1 x)2/5
5 1
5 7/5
2/5
u
=

1
+ C.
dx
=

u
du
=

7
7 x
x12/5

Exercises
10.4.1. (a) Find

x2 + 1
dx.
(2x 3)2

(b) Let p(x) denote a generi polynomial. Let a and b be real numbers, not
both zero, andZ let n be a positive integer. Explain how an inverse substitution
p(x)
dx.
will evaluate
n

(ax + b)
Z
1 x2
10.4.2. Find
dx. (Let x = os(u).)
x2

10.4 Antidi erentiation by Inverse Substitution

321

10.4.4 Inverse Substitution in Integrals

The inverse substitution formula for integrals (as ompared to antiderivatives)


is
Zb
a

fg=

Z g(b)
g(a)

where h inverts g.

f h

This formula follows from its variable-free ounterpart, similarly to the ase
of forward substitution.
Example 10.4.4. To evaluate
Z9
0

dx
p
.
1+ x

re all the substitution u = 1 + x from example 10.4.2. If x = 0 then u = 1


and if x = 1 then u = 2. Thus, by the al ulation in example 10.4.2 and by
the inverse substitution formula for integrals,
Z9
0

dx

p
=4
1+ x

Z2

(u 1) du = 4


1 3
2 16
u u =
.
3
3
1

R0

Example 10.4.5. To nd t=1 t t + 1 dt, let u = t + 1. Then t = u 1,


so dt = du. And if t = 1 then u = 0, while if t = 0 then u = 1. Hen e
Z0

t t + 1 dt =

t=1

Z1

(u 1) u du =

u=0

Z1

(u3/2 u1/2 ) du

u=0

1
2 5/2 2 3/2
4
2 2
= u
u = = .
5
3
5 3
15
0

Example 10.4.6. Let a > 0 be onstant. To evaluate


Za
p
a2 x2 dx,
x=a

let u = x/a, so that x = au. Then a2 x2 = a 1 u2 and dx = a du.


Also, if x = a then u = 1. Thus the integral is, iting example 10.3.13
(page 315) at the last step,
Za

x=a

Z1
p
a2 x2 dx = a2

p
a2
1 u2 du =
.
2
u=1

The answer isunsurprising sin e the upper half of the ir le of radius a has
equation y = a2 x2 .

322

10 Integration via Antidi erentiation

Exercises
10.4.3. Find the following integrals.
Z1
(a) x2 (x3 + 1)3 dx.
Z03/2
1

dx.
(b)
2
Z 01 9 x
( ) x 1 x dx.
0

10.4.4. Find the area of the region bounded by the ellipse x2 /4 + y2 = 1 and
the lines x = 1.
10.4.5. Consider the following argument: To evaluate
Z
os2 (x) dx,

the integral

x=0

let u = sin(x). Then os(x) = 1 u2 and du = os(x) dx.


then u = 0, and if x = then u = 0. Thus the integral is
Z

x=0

os (x) dx =
2

os(x) os(x) dx =

x=0

Z0

u=0

Also, if

x=0

p
1 u2 u du = 0.

And so the inverse substitution pro edure has shown that the integral is
zero.

(a) This argument an not be orre t sin e the integral is visibly positive.
What is wrong with it?
(b) Explain why the integral is also
Z

x=0

sin2 (x) dx.

Use this fa t and the idea of example 10.3.14 (page 315) to nd the integral.

10.5 Antidifferentiation by Parts


The formula for antidi erentiation by parts is
Z

fg = fg gf .

(10.8)

(Here fg means f times the derivative of g, and similarly for gf .) The formula
follows immediately from antidi erentiating the Produ t Rule for derivatives,

10.5 Antidi erentiation by Parts

323

(fg) = fg + gf

to get
fg =

fg + gf .

The orresponding formula for integration by parts is, naturally,


Zb
a

b Z b

fg = fg
gf .
a

(10.9)

To use the formula, the idea is to write the fun tion whose antiderivative
we seek in the form fg , where the antiderivative of gf is easier to nd.
R
Example 10.5.1. We will al ulate x sin(3x) dx. The rst step is to sear h
for an antiderivative of x sin(3x). Let
f(x) = x, g (x) = sin(3x),
1
f (x) = 1, g(x) = os(3x).
3

Then by the formula for antidi erentiation by parts


Z

Hen e

Z
0

x sin(3x) dx =

f(x)g (x) dx
Z
= f(x)g(x) f (x)g(x) dx
Z
1
x
os(3x) dx
= os(3x) +
3
3
1
x
= os(3x) + sin(3x).
3
9



x
1


x sin(3x) dx = os(3x) + sin(3x) = os(3) = .
3
9
3
3
0

If instead we had pro eeded by setting


f(x) = sin(3x),

g (x) = x,
1
f (x) = 3 os(3x), g(x) = x2 ,
2

then by the formula for antidi erentiation by parts


Z

x sin(3x) dx =

f(x)g (x) dx
Z
= f(x)g(x) f (x)g(x) dx
Z
1 2
3
x2 os(3x) dx.
= x sin(3x)
2
2

324

10 Integration via Antidi erentiation


R

In this ase the antiderivative x2 os(3x) dx looks more ompli ated than
the one we started with. When you antidi erentiate by parts, it is not always
lear what you should take for f and for g . If you nd that things are starting
to look more ompli ated rather than less ompli ated, you might try another
hoi e for f and g .
R

Example 10.5.2. To nd sin( x) dx, rst arry out the inverse substitu
tion u = x. Then x = u2 , so that dx = 2u du. Thus
Z
Z
Z

sin( x) dx = sin(u) 2u du = 2 u sin(u) du.


R

Now we an antidi erentiate by parts to nd u sin(u) du. Let


g (u) = sin(u),

f(u) = u,

g(u) = os(u).

f (u) = 1,

Then
Z

u sin(u) du =

f(u)g (u)du
Z
= f(u)g(u) f (u)g(u) du
Z
= u os(u) + os(u) du
= u os(u) + sin(u).

Hen e
Z

sin( x) dx = 2 u sin u du
= 2u os(u) + 2 sin(u)

= 2 x os( x) + 2 sin( x).

Example 10.5.3. Antidi erentiation by parts is used to evaluate antideriva-

tives of the forms


Z

x sin(ax) dx,
n

x os(ax) dx,
n

xn eax dx,

where n is a positive integer. All three antiderivatives an be redu ed to


antiderivatives of the forms
Z

n1

sin(ax) dx,

n1

os(ax) dx,

xn1 ex dx,

10.5 Antidi erentiation by Parts

325

and so by applying the pro ess n times we redu e the power of x down
to xR0 , whi h gives us antiderivatives that we an nd easily. For example,
for xn sin(ax) dx, let
g (x) = sin(ax),
1
f (x) = nxn1 , g(x) = os(ax).
a
f(x) = xn ,

Then
Z

x sin(ax) dx =

f(x)g (x) dx = f(x)g(x) f (x)g(x) dx


Z
xn
n
=
xn1 os(ax) dx.
os(ax) +
a
a
R
Example 10.5.4. We will al ulate sin(ln(x)) dx. To do so, let
n

f(x) = sin(ln(x)),

f (x) =

os(ln(x))
x

g (x) = 1,

g(x) = x.

(10.10)

Then
Z

sin(ln(x)) dx = f(x)g (x) dx = f(x)g(x) f (x)g(x) dx


Z

= x sin(ln(x)) os(ln(x)) dx

Next use the same te hnique to nd an antiderivative of os(ln(x)). Let


f(x) = os(ln(x)),
f (x) =

sin(ln(x))
x

g (x) = 1,
,

g(x) = x.

Then
Z

os(ln(x)) dx = f(x)g (x) dx = f(x)g(x) f (x)g(x) dx

Z
= x os(ln(x)) + sin(ln(x)) dx

It may seem as though we are ba k where we started, but in fa t the two


al ulations ombine to give

Thus

sin(ln(x)) dx = x sin(ln(x)) x os(ln(x)) sin(ln(x)) dx.

326

10 Integration via Antidi erentiation


Z

2 sin(ln(x)) dx = x sin(ln(x)) x os(ln(x)),

and

x
2

sin(ln(x)) dx = (sin(ln(x)) os(ln(x))).

Example 10.5.5. We already know from working a sum in hapter 5 that


Z
ln(x) dx = x ln(x) x.

Now we rederive the result using antidi erentiation by parts. Let


f(x) = ln(x), g (x) = 1,
f (x) = 1/x,

Then

g(x) = x.
Z

ln(x) dx = f(x)g (x) dx = f(x)g(x) f (x)g(x) dx

Z
= x ln(x) 1 dx

= x ln(x) x.

We naturally wonder whether this method of nding ln(x) dx is related


to the summation method of hapter 5. It is, losely, as explained in exer ise 10.5.4.
Theorem 10.5.6 (Antiderivative of the Inverse Function).
be intervals in R. Let the fun tion

Let I and J

g : I J

have inverse fun tion

h : J I.

Suppose that g is di erentiable and that g is ontinuous. Re all that f1


denotes the identity fun tion. Then an antiderivative of h is
Z

h = f1 h

Z 
g h.

Thus the antiderivative of the inverse fun tion an be expressed in terms


of the inverse fun tion and the antiderivative of the original fun tion.
With variables, the previous formula is
Z

h(x) dx = xh(x) g(u) du

where u = h(x).

10.5 Antidi erentiation by Parts

327

It an be shown that in onsequen e of the hypotheses of the theorem,


the inverse fun tion h is ontinuous, making the pending al ulations valid.
We omit the proof, but it deserves omment that the argument requires that
I and J be intervals. (See exer ise 10.5.6.) This is an instan e where the full
des ription of fun tions|in luding domains and odomains|is ne essary to
analyze a situation.

Proof. The inverse substitution formula, but with the roles of g and h ex hanged, is
Z

Z
fh=

f g h.

Spe ialize f to the identity fun tion f1 . Thus f1 h = h, and the formula
be omes
Z

Z
h=

f1 g

h.

Note that f1 = f0 is the onstant fun tion 1. Thus f1 g = g, so that antidifferentiation by parts gives
Z

Z 
h = (f1 g) h
g h.

But sin e f1 is the identity fun tion and sin e g inverts h,


(f1 g) h = (f1 h) (g h) = h f1 = f1 h,

and so we have the desired formula,


Z

Z 
h = f1 h
g h.

The formula in Theorem 10.5.6 says that the antiderivative of the inverse
fun tion equals a produ t minus a omplementary antiderivative of the original fun tion. This fa t about antiderivatives dovetails perfe tly, via the Fundamental Theorem of Cal ulus, with our earlier al ulations of the integrals
of the exponential and the ar - osine (see pages 203 and 227): the integral of
the inverse fun tion equals a box-area minus a omplementary integral of the
original fun tion.
Exercises
10.5.1.ZCal ulate the following antiderivatives.
(a) xex dx.

328

10 Integration via Antidi erentiation


Z

(b) x2 ex dx.
Z

( ) ex sin(x) dx. (Integrate by parts twi e, and then don't give up.)
Z

dx.
4
Z p x2
(e) x 4 x2 dx.
Z
(f) x ln(|x|) dx, where R. (Don't forget the ase where = 1.)
Z
(g) x2 os(2x) dx.
Z
(h) x ln(x) dx.

(d)

10.5.2. Evaluate the following integrals.


Z1
(a)
x ar sin(x) dx.
Zx=0
4

(b)
ar se ( x) dx. (The ar -se ant fun tion has domain [1, ) and
x=1

odomain [0, /2).)

10.5.3. (a) Antidi erentiate by parts to nd an antiderivative of ar os.

(b) Antidi erentiate by parts to nd an antiderivative of ar tan.

10.5.4. This exer ise des ribes summation

by parts, the dis rete analogue of

antidi erentiation by parts, whi h is a ontinuous pro ess.


(a) Consider two fun tions (sequen es, in fa t)
f, g : Z0 R,

and de ne for k Z1
f(k) = f(k + 1) f(k),

g(k) = g(k) g(k 1).

Show that for any n Z1 ,


n1
X
k=1

n n1
X

f(k)g(k) = f(k)g(k 1)
f(k)g(k).
1

k=1

(It may be most onvin ing to write out ea h side of the equality and on rm
that the same terms o ur.) Note the similarity between this formula and
formula (10.9) for integration by parts.
(b) Re all that in hapter 5 we integrated the logarithm by evaluating the
sum
(x) = 1 + 2x + 3x2 + + (n 1)xn2 ,

x 6= 1.

10.5 Antidi erentiation by Parts

329

Using the notation of part (a), let


and

f(k) = k

g(k) = 1 + x + + xk1 ,

in luding the ase g(0) = 0. Show that


(x) =

n1
X

f(k)g(k),

k=1

so that by part (a), in fa t

(x) = f(n)g(n 1)

n1
X

f(k)g(k).

k=1

Evaluate this se ond expression for (x) to rederive the sum as omputed in
exer ise 5.4.2 (page 166).
10.5.5. What is wrong with the following argument? Let
f(x) =

1
,
x

f (x) =

Then

If we subtra t

1
,
x2

g(x) = x.

Z
f(x)g (x) dx = f(x)g(x) f (x)g(x) dx
Z
1
dx.
=1+
x

1
dx =
x

g (x) = 1,

1
x

dx from both sides we obtain


0 = 1.

10.5.6. Consider the set


I = {x R : 0 x < 1 or 2 x 3 or 4 < x 5}.

Note that I is not an interval. Let J = [0, 3], an interval. Consider the fun tion
g : I J,

if 0 x < 1,
g(x) = x 1 if 2 x 3,

x 2 if 4 < x 5.

The graph of g is shown in gure 10.7. Draw the graph of the inverse fun tion
h : J I.

Explain why g is ontinuous but h is not. Is h integrable? What would hange


in this exer ise if the domain of g were extended to in lude x = 1 and x = 4?

330

10 Integration via Antidi erentiation


g(x)

PSfrag repla ements


x
Figure 10.7.

A ontinuous, invertible fun tion with a dis ontinuous inverse

10.5.7. Let
In =

Z /2
0

(a) Evaluate I0 and I1 .


(b) Show that

sinn (x) dx, n Z0 .

n1
In2 ,
n
Use this formula to evaluate I3 and I4 .
( ) Show that for n odd,
In =

In =

Show that for n even,


In =

n 2.

2 4 6 (n 1)
.
3 5 7n

1 3 5 (n 1)
.
2 4 6n
2

10.5.8.ZFind redu tion formula for the following antiderivatives.


(a) xn os(ax) dx.
Z
(b) xn eax dx.
Z
( ) (ln(x))n dx.

A
Assumptions About the Real Number System

We assume that there is a real number system, a set R that ontains two
distin t elements 0 and 1 and is endowed with the algebrai operations of
addition,
and multipli ation,

+ : R R R,
: R R R.

The sum +(a, b) is written a + b, and the produ t (a, b) is written a b or


more brie y as ab.
The assumed algebrai properties of the real number system are as follows.
Theorem A.0.1 (Field Axioms for (R, +, )). The real number system,
with its distin t 0 and 1 and with its addition and multipli ation, is

assumed to satisfy the following set of axioms.


(a1) Addition is asso iative: (x + y) + z = x + (y + z) for all x, y, z R.
(a2) 0 is an additive identity: x + 0 = x for all x R.
(a3) Existen e of additive inverses: For ea h x R there exists y R
su h that x + y = 0.
(a4) Addition is ommutative: x + y = y + x for all x, y R.
(m1) Multipli ation is asso iative: x(yz) = (xy)z for all x, y, z R.
(m2) 1 is a multipli ative identity: 1x = x for all x R.
(m3) Existen e of multipli ative inverses: For ea h nonzero x R there
exists y R su h that xy = 1.
(m4) Multipli ation is ommutative: xy = yx for all x, y R.
(d1) Multipli ation distributes over addition: (x + y)z = xz + yz for all
x, y, z R.

332

A Assumptions About the Real Number System

All of basi algebra follows from the eld axioms. For example, additive
and multipli ative inverses are unique, the an ellation law holds, 0 x = 0
for all real numbers x, and so on.
Subtra ting a real number from another is de ned as adding the additive
inverse. In symbols,
: R R R,

x y = x + (y)

for all x, y R.

We also assume that R is an ordered eld. This means that there is a


subset R+ of R (the positive elements) su h that the following axioms hold.
Theorem A.0.2 (Order Axioms).

(o1) Tri hotomy Axiom: For every real number x, exa tly one of the fol-

lowing onditions holds:

x R+ ,

x R+ ,

x = 0.

(o2) Closure of positive numbers under addition: For all real numbers x

and y, if x R+ and y R+ then also x + y R+ .


(o3) Closure of positive numbers under multipli ation: For all real numbers x and y, if x R+ and y R+ then also xy R+ .

For all real numbers x and y, de ne \x < y" to mean \y x R+ ." The
de nitions of \x y" and \x > y" and \x y" are analogous. The usual
rules for inequalities then follow from the axioms.
Finally, we assume that the real number system is complete. Completeness an be phrased in various ways, all logi ally equivalent. The version of
ompleteness that is urrently in Ray Mayer's notes for Mathemati s 112 is
as follows.

Every
binary sear h sequen e in the real number system onverges to a unique
limit.
Theorem A.0.3 (Completeness as a Binary Search Criterion).

Two other versions of ompleteness are phrased in terms of sequen es and


in terms of set-bounds:
Theorem A.0.4 (Completeness as a Monotonic Sequence Criterion).
Every bounded monotoni sequen e in R onverges to a unique limit.
Theorem A.0.5 (Completeness as a Set-Bound Criterion). Every
nonempty subset of R that is bounded above has a least upper bound.

Convergen e is a on ept of analysis, and therefore so is ompleteness. All


three statements of ompleteness are existen e statements.

List of Symbols

(a, b) (open interval), 63


(a, b] (half-open interval), 63
(a, ) (open half-in nite interval), 63
(, b) (open half-in nite interval), 63
(, b] ( losed half-in nite interval),

63

(, ) (the number line), 63


(sn ) (sequen e), 70
[a, b) (half-open interval), 63
[a, b] ( losed interval), 63
[a, ) ( losed half-in nite interval),

63

Arba (f) (area under a graph), 69


A (area of rst triangle ins ribed in
parabola), 8
|` | (absolute value), 71

(binomial oe ient), 254


k
Ar (area-fun tion), 68
ar os (inverse osine fun tion), 227
ar ot (inverse otangent fun tion), 230
ar sin (inverse sine fun tion), 230
ar tan (inverse tangent fun tion), 230
Ar(R) (area of a region), 69
tri

(the set of bounded subsets of the


plane), 64

os ( osine fun tion), 215


ot ( otangent fun tion), 226
s ( ose ant fun tion), 226
D

(parabola dire trix), 2, 4

(Leibniz notation for derivative),


310

du/dx

(base of the natural logarithm, 192


(empty set), 61
erf (error fun tion), 297
exp (exponential fun tion), 194
e

F (parabola fo us), 2, 4
f0 ( onstant fun tion 1), 33
f1 (identity fun tion), 33
F|ba (di eren e of fun tion-values), 303
f (rational power fun tion), 25, 30
f : A B (f is a fun tion from A
to B), 65
f1 (re ipro al fun tion), 33
f (derivative of f), 134
f (x) (derivative of f at x), 134
f(A) (range of a fun tion), 66

graph(f) (graph of a fun tion), 67


(if and only if), 73

R b(in, is in, is an element of), 29


Ra f (integral of f from a to b), 110
R f (antiderivative of f), 303
f(x) dx (antiderivative of f), 305
Rb
f(x) dx (integral notation with
x=a
Rb

variable and in nitesimal), 314


(integral notation with
variable), 237

f(x) dx
x=a

{}

(set notation), 60

334

List of Symbols

lim (limit of a fun tion), 124


limn (limit of a sequen e), 78
ln (logarithm fun tion), 149

S (lower sum for an area), 109


Sn (sum of areas of n generations

(binomial oe ient), 234


(not in, is not in, is not an element
of), 29

`n
k

P (the set of polygons in the plane),


(area of the unit ir le), 213
Q

64

(the rational numbers), 29, 61

R (the real numbers), 29, 61


R2 (the eu lidean plane), 61
R>0 (the positive real numbers), 29, 61
R0 (the nonnegative real numbers), 61

of
triangles ins ribed in parabola),
10
Sn (sum of inner box-areas), 44
se (se ant fun tion), 226
sin (sine fun tion), 215
(upper sum for an area), 109
tan (tangent fun tion), 226

Z (the integers), 29, 61


Z1 (the negative integers), 29, 61
Z1 (the positive integers), 29, 61
Z0 (the nonnegative integers), 29, 61

Index

absolute value, 71
basi properties, 72
relation with intervals, 74
algebrai fun tion, 67
antiderivative, 295
notation, 303
notation with variable, 305
properties, 295
antidi erentation by parts formula, 322
antidi erentiation formulas, 306
approa h, 124
approa hability of a point from a set,
124
ar - osine fun tion, 227
derivative, 228
integral, 227
ar - otangent fun tion, 230
derivative, 231
ar -sine fun tion, 230
derivative, 231
ar -tangent fun tion, 230
derivative, 231
Ar himedean property of the real
number system, 76
area between two urves as an integral,
179
area-fun tion, 68
basi fun tion limits, 128
basi sequen e limits, 81
1/n Rule, 81
1/n Rule, 81

nth
nth

Power Rule, 82
Root Rule, 81
Constant Sequen e Rule, 81
Binomial Theorem, 257
Finite, 235
bounded
subset of the plane, 64
Broun ker's formula for ln(2), 242
Chain Rule for derivatives, 143
losed interval, 63
odomain of a fun tion, 65
ompleteness of the real number system,
332
as a binary sear h riterion, 332
as a monotoni sequen e riterion,
332
as a set-bound riterion, 332
ompound interest, 207
onstant multiple of a sequen e, 92
Constant Multiple Rule
for derivatives, 140
for fun tion limits, 131
for sequen es, 92
Constant Sequen e Rule, 81
ontinuity
de nition, 186
of the power fun tion, 187
ontinuity implies integrability, 188
onvergent sequen e, 77
osine, 214
osine and sine

336

Index

angle sum and di eren e formulas,


218
basi identities, 216
derivatives, 222
di eren e formulas, 219
double and half angle formulas, 218
integrals, 223
produ t formulas, 218
Taylor polynomial and remainder,
251
Criti al Point Theorem, 264
ubi equation
solving with parabolas, 21
de reasing fun tion, 112
derivative
de nition, 134
Leibniz notation, 310
of the absolute value fun tion away
from zero, 139
of the logarithm, 159
of the power fun tion, 138
re hara terization, 135
se ond re hara terization, 142
di eren e of powers formula, 35
di eren e of two sequen es, 92
di eren e-quotient
for the parabola, 13
di erentiability implies ontinuity, 187
di erentiable fun tion, 134
di erentiation rules
generative, 140
dire trix of a parabola, 1
divergent sequen e, 77, 78
domain of a fun tion, 65
e, 192
empty set, 61
endpoints of an interval, 63
error fun tion, 297
exponential fun tion
as a limit of powers, 206
de nition, 194
integral of, 205
is its own derivative, 199
properties, 194

Taylor polynomial and remainder,


249
exponential growth dominates
polynomial growth, 198
Extreme Value Theorem, 262
extremum of a fun tion, 262
lo al, 262
stri t, 262
stri t, 262
eld axioms, 331
Finite Binomial Theorem, 235
nite geometri sum formula, 28, 41
for r = 1/4, 11
fo us of a parabola, 2
forward substitution antidi erentiation
formulas, 313
forward substitution formula, 307
for integrals, 314
for integrals with variable and dx,
314
with variables, 309
fun tion, 65
algebrai , 67
odomain, 65
ontinuous, 186
de reasing, 112
di erentiable, 134
domain of, 65
graph, 67
in reasing, 112
integrable, 110
limit of, 124
monotoni , 112
pie ewise monotoni , 115
range, 66
stri tly de reasing, 34
stri tly in reasing, 34
trans endental, 67
fun tion limits
basi , 128
generative, 131
Fundamental Theorem of Cal ulus
Part I, 296
Part I, weaker variant, 299
Part II, 301
Part II, rephrased, 303

Index
generative, 92
generative derivative rules, 140
Chain Rule, 143
Constant Multiple Rule, 140
Produ t Rule, 140
Quotient Rule, 140
Re ipro al Rule, 140
Sum/Di eren e Rule, 140
generative fun tion limit rules, 131
Constant Multiple Rule, 131
Inequality Rule, 132
Produ t Rule, 131
Quotient Rule, 131
Re ipro al Rule, 131
Squeezing Rule, 133
Sum/Di eren e Rule, 131
generative integral rules, 120, 176
Inequality Rule, 121
Inequality Rule, se ond version, 178
generative sequen e limits, 92
Constant Multiple Rule, 92
Inequality Rule, 100
Produ t Rule, 92
Quotient Rule, 93
Re ipro al Rule, 93
Squeezing Rule, 100
Sum/Di eren e Rule, 92
geometri partition, 42
geometri series, 98
with ratio r, 98
geometri series formula, 99
Goldba h Conje ture, 87
graph of a fun tion, 67
in reasing fun tion, 112
inde nite integral, 293
properties, 294
index-translate of a sequen e, 89
index-translation rule for sequen es, 89
inequality
stri t, 262
Inequality Rule
for sequen es, 100
Inequality Rule for integrals, 121
integer, 26
integrability of pie ewise monotoni
fun tions, 115

337

integrable fun tion, 110


integral
de nition, 110
generative rules, 120
notation with variable, 237
notation with variable and dx, 314
of the logarithm, 164
with out-of-order endpoints, 147
integration
signed, 173
with out-of-order endpoints, 173
integration by parts formula, 323
Intermediate Value Theorem, 189
and nth roots, 192
se ond version, 190
interval, 62
losed, 63
endpoints of, 63
open, 63
relation with absolute value, 74
inverse osine fun tion, 227
derivativeintegral, 228
integral, 227
inverse otangent fun tion, 230
derivative, 231
inverse fun tion
antiderivative, 326
inverse sine fun tion, 230
derivative, 231
inverse substitution formula, 317
for integrals, 321
with variable, 318
inverse tangent fun tion, 230
derivative, 231
Irrelevan e of Finite Index-shifts, 89
large positive real number, 37
laws of exponents, 31, 40
laws of real exponents, 197
Leibniz notation for the derivative, 310
limit
basi sequen e limits, 81
of a sequen e, 77
of the power fun tion at zero, 129
limit of a fun tion, 124
logarithm
de nition, 149

338

Index

derivative, 159
integral, 164
key property, 149
properties, 154
Taylor polynomial and remainder,
243
logarithmi growth, 156
lower and upper sums
basi property, 109
bootstrap result, 110
lower sum for the area under a graph,
109
maximum of a fun tion, 262
lo al, 262
stri t, 262
stri t, 262
Mean Value Theorem, 274
medium-sized positive real number, 37
minimum of a fun tion, 262
lo al, 262
stri t, 262
stri t, 262
monotoni fun tion, 112
integrability of, 112
nth
nth

Power Rule, 82
Root Rule, 81

1/n Rule, 81
1/n Rule, 81

open interval, 63
order axioms, 332
ordered pair, 61
origami, 20
pair
ordered, 61
unordered, 62
parabola
algebrai de ning equation, 3
and solving the ubi equation, 21
di eren e-quotient, 13
dire trix, 1
fo us, 2
geometri de ning property, 2
more general algebrai equations, 5

quadrature of, 6
se ant slope, 13
tangent slope, 14
partition
geometri , 42
uniform, 42
Persisten e of Inequality, 126
pie ewise monotoni fun tion, 115
integrability, 115
power fun tion
ontinuity of, 187
limit at zero, 129
Taylor polynomial and remainder,
257
Produ t Rule
for derivatives, 140
for fun tion limits, 131
for sequen es, 92
Pythagorean Theorem, 5
quadrature of the parabola, 6
quotient of two sequen es, 92
Quotient Rule
for derivatives, 140
for fun tion limits, 131
for sequen es, 93
raising a positive real number to a real
power, 196
range of a fun tion, 66
rational number, 26
rational power fun tion
de nition, 30
di erentiation of, 47
integration of, 51
real number system, 331
Ar himedean property, 76
real sequen e, 77
re ipro al of a sequen e, 92
Re ipro al Rule
for derivatives, 140
for fun tion limits, 131
for sequen es, 93
Rolle's Theorem, 273
ruler fun tion, 116
s aling result for power fun tions, 54

Index
se ant slope
for the parabola, 13
sequen e, 70, 77
approa hes a point, 124
onvergent, 77
divergent, 77, 78
index-translate of, 89
limit, 77
real, 77
uniqueness of limit, 90
sequen e, index-translation rule for, 89
set, 29, 60
de ned by onditions, 61
empty, 61
sine, 214
small positive real number, 37
Snell's Law, 272
Squeezing Rule
for sequen es, 100
stri t inequality, 262
stri tly de reasing fun tion, 34
stri tly in reasing fun tion, 34
Strong Approximation Lemma, 75
sum of two sequen es, 92
Sum/Di eren e Rule

339

for derivatives, 140


for fun tion limits, 131
for sequen es, 92
summation by parts, 328
tangent line
analyti des ription, 136
geometri des ription, 136
tangent slope
of the parabola, 14
Taylor polynomial and remainder
for osine and sine, 251
for the exponential, 249
for the logarithm, 243
for the power fun tion, 257
trans endental fun tion, 67
triangle inequality, 72
basi , 73
uniform partition, 42
uniqueness of sequen e limits, 90
unordered pair, 62
upper sum for the area under a graph,
109
volume-fun tion, 69