Professional Documents
Culture Documents
Kane
Writing
Proofs in
Analysis
Jonathan M. Kane
123
Jonathan M. Kane
Department of Mathematics
University of Wisconsin - Madison
Madison, WI, USA
ISBN 978-3-319-30965-1
ISBN 978-3-319-30967-5 (eBook)
DOI 10.1007/978-3-319-30967-5
Library of Congress Control Number: 2016936668
Springer International Publishing Switzerland 2016
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature
The registered company is Springer International Publishing AG Switzerland
Acknowledgments
I wish to thank Natalya St. Clair for her excellent work creating the illustrations
appearing in this textbook. She took my crude sketches and vague ideas and turned
them into pleasing artwork and instructive diagrams. I also wish to thank Daniel M.
Kane, Alan Gluchoff, Thomas Drucker, and Walter Stromquist for their insightful
comments about the presentation, content, and correctness of the text.
vii
Preface
After learning to solve many types of problems such as those found in the first
courses in Algebra, Geometry, Trigonometry, and Calculus, mathematics students
are usually exposed to a transition course where they are expected to write proofs
of various theorems. I taught such a course for a dozen years and was never satisfied
with the textbooks available for that course. Although such textbooks often teach
the fundamentals of logic (conditionals, biconditionals, negations, truth tables) and
give some common proof strategies such as mathematical induction, the textbooks
failed to teach what a student needs to be thinking about when trying to construct a
proof. Many of these books present a great number of well-written proofs and then
ask students to write proofs of similar statements in the hope that the students will
be able to mimic what they have seen. Some of these books are also designed to be
used as an introductory textbook in Analysis, Abstract Algebra, Topology, Number
Theory, or Discrete Mathematics, and, as such, they concentrate more on explaining
the fundamentals of those topic areas than on the fundamentals of writing good
proofs.
This Book Is Not Your Traditional Transition Textbook The goal of this book
is to give the student precise training in the writing of proofs by explaining what
elements make up a correct proof, by teaching how to construct an acceptable proof,
by explaining what the student is supposed to be thinking about when trying to write
a proof, and by warning about pitfalls that result in incorrect proofs. In particular,
this book was written with the following directives:
Unlike many transition books which do not give enough instruction about how
to write proofs, most of the proofs presented in this text are preceded by detailed
explanations describing the thought process one goes through when constructing
the proof. Then a good proof is given that incorporates the elements of that
discussion.
For proofs that share the same general structure such as the proof of lim f .x/ D L
x!a
for various functions, proof templates are provided that give a generic approach
to writing that type of proof.
ix
Preface
Jonathan M. Kane
Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Proof Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
What Are Proofs, and Why Do We Write Them? . . . . . . . . . . . . . . . . . . . . . . . .
1.1
What Is a Proof? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2
Why We Write Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
The Basics of Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1
The Language of Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1
Conditional Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.2
Negation of a Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.3
Proofs of Conditional Statements . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.4
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2
Template for Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3
Proofs About Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1
Set Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.3
Proofs About Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.4
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.5
Proofs About Set Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.6
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4
Proofs About Even and Odd Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1
Definitions of Even and Odd Integers . . . . . . . . . . . . . . . . . . . . .
2.4.2
Proofs About Even and Odd Integers . . . . . . . . . . . . . . . . . . . . .
2.4.3
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5
Basic Facts About Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.1
Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.2
The Completeness Axiom and the Real Numbers . . . . . . . .
2.5.3
Absolute Value, the Triangle Inequality, and Intervals . . .
2.5.4
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.1
Function, Domain, Codomain . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.2
Surjection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.3
Injection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vii
ix
xvii
xx
1
1
5
9
9
9
10
11
12
12
16
17
17
18
19
22
22
26
27
27
28
31
31
31
35
38
40
40
40
40
41
xi
xii
CONTENTS
2.6.4
Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.5
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1
The Definition of Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2
Proving lim f .x/ D L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
x!a
3.2.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3
One-Sided Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4
Limits at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5
Limit of a Sequence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.1
Definition of Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.2
Arithmetic with Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.3
Monotone Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.4
Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.5
Limit of a Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.6
Limits of Monotone Sequences and
Mathematical Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.7
Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.8
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6
Proving That a Limit Does Not Exist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.1
Why a Limit Might Not Exist . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.2
Quantifiers and Negations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.3
Proving No Limit Exists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.4
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7
Accumulation Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.8
Infinite Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.8.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9
The Arithmetic of Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.1
Limit of a Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.2
Limit of a Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.3
Limit of a Quotient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.4
Limit of Rational Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.5
Other Types of Limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9.6
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10
Other Limit Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10.1
The Limit of a Positive Function . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10.2
Uniqueness of Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10.3
The Squeezing Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10.4
Limits of Subsequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10.5
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
44
47
47
49
49
54
54
56
57
59
60
60
60
60
61
62
62
66
67
68
68
68
70
73
74
79
79
81
81
82
83
85
87
89
89
89
90
90
91
92
93
CONTENTS
3.11
xiii
93
97
99
99
101
105
105
109
110
110
111
115
117
117
120
121
123
123
123
126
127
130
131
133
133
134
135
136
139
140
143
145
146
150
150
155
155
159
159
159
162
163
166
166
169
xiv
CONTENTS
6.5
169
172
173
177
178
183
183
189
189
194
194
200
201
201
204
205
206
208
209
209
212
214
215
218
219
222
222
224
224
225
230
231
236
239
239
241
241
246
246
252
255
255
256
259
263
CONTENTS
8.5.5
Arithmetic of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.5.6
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.6
Fundamental Question of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
Topology of the Real Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1
Interior, Exterior, and Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2
Open and Closed Sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3
Unions and Intersections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.4
Continuous Functions Applied to Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.4.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.5
Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.5.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.6
Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.6.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.7
Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.7.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.1
Definition of Metric Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2
Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2.1
CauchySchwarz Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2.2
Minkowski Inequality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2.3
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.3
Examples of Metric Spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.3.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.4
Topology of Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.4.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.5
Limits in Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.5.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.6
Continuous Functions on Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.6.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.7
Homeomorphism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.7.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.8
Connected Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.8.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.9
Compact Metric Spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.9.1
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.10 Complete Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.10.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.11 Contraction Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.11.1 Contraction Mapping Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.11.2 Picards Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xv
265
266
267
269
269
274
274
278
278
282
282
285
285
288
288
291
291
293
295
295
297
297
298
298
299
305
306
307
308
311
311
314
315
316
316
317
317
322
323
327
327
327
329
xvi
CONTENTS
List of Figures
Fig. 1.1
Fig.
Fig.
Fig.
Fig.
Fig.
15
26
37
39
43
2.1
2.2
2.3
2.4
2.5
Fig. 3.1
Fig. 3.2
Fig. 3.3
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
3.4
3.5
3.6
3.7
3.8
3.9
3.10
Fig. 4.1
Fig. 4.2
Fig.
Fig.
Fig.
Fig.
4.3
4.4
4.5
4.6
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
4.7
4.8
4.9
4.10
4.11
4.12
4.13
lim f .x/ D L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
x!a
lim f .x/ D L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
x!a
lim f .x/ D L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
x!a
Graph of f .x/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Approaching a limit as x ! 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proving bounded monotone sequences converge . . . . . . . . . . . . . . . . . . . . .
f has no limit at x D 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graph of sin 1x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Set with accumulation point a and isolated point b . . . . . . . . . . . . . . . . . . .
Sequences approaching the lim sup and lim inf . . . . . . . . . . . . . . . . . . . . . . .
56
57
63
71
72
74
94
Continuity of a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A function equal to 2x for rational x and x C 1 for
irrational x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
f .x/ D 1x is not uniformly continuous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HeineBorel Theorem first proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HeineBorel Theorem second proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
y and z straddle one endpoint but remain in an interval
of the open cover. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proving that a continuous function on a; b is bounded . . . . . . . . . . . . . .
The maximum and minimum of a function f .x/ on an interval . . . . . .
f passing through each y between f .c/ and f .d/ . . . . . . . . . . . . . . . . . . . . . .
A function with a jump discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graph of sin 1x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graphs of sgn.x/ and bxc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Graphs of functions with discontinuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100
104
106
112
113
115
125
126
128
130
130
131
132
xvii
xviii
LIST OF FIGURES
5.1
5.2
5.3
5.4
5.5
5.6
5.7
134
134
142
145
147
148
156
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
161
161
164
165
168
170
184
185
191
Fig. 7.1
Fig. 7.2
Fig. 7.3
Comparing the series with the integral in the Integral Test . . . . . . . . . . . 216
Converging to ln 2 with an alternating series . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Rearranging terms to converge to L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
Fig. 8.1
Fig. 8.2
Fig. 8.3
Fig. 8.4
Fig. 8.5
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
9.1
9.2
9.3
9.4
9.5
9.6
9.7
9.8
9.9
240
240
241
242
243
270
273
275
278
279
284
287
292
293
293
LIST OF FIGURES
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
10.2
10.3
10.4
10.5
10.6
10.7
10.8
10.9
Fig.
Fig.
Fig.
Fig.
Fig.
Fig.
10.10
10.11
10.12
10.13
10.14
10.15
Euclidean distance is R2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
N.0; 1/ in the Euclidean, taxicab, and supremum metrics . . . . . . . . . . . .
Some functions in C0; 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proving that the union of open sets is open. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Limit of f W X ! Y as x approaches a is L . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Limit of a sequence in a metric space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A compact set of a metric space is closed and bounded . . . . . . . . . . . . . .
Extrema of a continuous real-valued function on a
compact set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Continuous bijection on a compact metric space. . . . . . . . . . . . . . . . . . . . . .
Enclosing a closed bounded set in a grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Contraction mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Mandelbrot set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Stages in the forming of the Sierpinski triangle . . . . . . . . . . . . . . . . . . . . . . .
Generation of a fractal fern. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xix
300
301
302
307
308
310
318
320
322
326
328
333
333
339
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 12
. 20
. 23
. 41
. 42
. 50
. 65
. 70
. 102
. 106
. 296
xx
Chapter 1
But more likely you are interested in investigating some properties of these
numbers having to do with their order or how they behave when operated on
by addition or multiplication. This, of course, would mean that you will need to
make clear statements about addition and multiplication operations and a less than
relationship, again, so that you do not run into problems later because you were
being ambiguous. So, you might write definitions of addition, multiplication, and
less than, and then make statements about how these operations behave such as a
Commutative Law of Addition (mCn D nCm), a Distributive Law of Multiplication
over Addition (a.b C c/ D ab C ac), and an Order Property of Addition (r < s
implies r C t < s C t). These statements about how these defined quantities work
are called axioms, postulates, or principles. They are statements that you accept
as the guiding rules for how your mathematical objects behave and go beyond the
definitions to describe and make precise just what the definitions are talking about.
Once you have made definitions and laid out your axioms, you should have the
tools necessary to begin an investigation of other properties. Suppose that someone
looks at a few examples and notice that 1 C 9 D 10 and 10 is 2 times another
number, 5. They then notice that 4C12 D 16, 3C147 D 150, and 1002C6 D 1008,
and all of these results are also numbers equal to 2 times another number. This might
lead them to make the statement that if you add two natural numbers together,
the result is always 2 times another number. Such a statement would be called a
conjecture, a statement whose truth has not yet been determined. Of course, you
know that this statement is false and came about because the investigator had not
yet considered enough examples. Once they stumble upon 5 C 8 D 13 and notice
that 13 cannot be represented as 2 times another number, they will know that the
statement does not hold true in every case.
Other conjectures such as for every natural number a, the number a2 3a C 12
is a multiple of 2 hold up to more scrutiny. At some point in your investigation you
might see a convincing argument that this conjecture is, in fact, a true statement.
Such a convincing argument is what is called a proof. Once it is known that a
statement has a proof, it is known as a theorem, lemma, corollary, proposition,
or law. So, a proof of a statement in mathematics is a convincing argument that
establishes the truth of that statement.
Some statements are very easily proved, and certainly mathematicians often set
up axioms in order to make particular statements easy to prove. At first this may
appear to be cheating or, at best, unproductive and uninteresting because it seems
to defeat the purpose of establishing truth by dictating rules that make it trivial to
establish the truth. But this is certainly not the case. It is common for mathematicians
to have an intuitive idea about how a system should work before they feel that they
understand it enough to set down formal definitions and axioms. Perhaps you wanted
addition of all natural numbers a and b to satisfy a C b D b C a. Then it would make
sense to include this rule among your axioms. The axioms are written with the idea
of establishing enough structure so that the statements the mathematicians want to
hold true can easily be proved. The richness of mathematics is that after assuring
that the obvious can be proved from the axioms, there are many more results that
can be proved that are not immediately obvious from the definitions and axioms,
statements which might never have been apparent to those who set up the system in
the first place. For example, Fermats Last Theorem (there are no natural numbers a,
b, c, and n > 2 such that an C bn D cn ) is a statement about natural numbers which
could only be conjectured after investigating a large number of examples, and stood
as a conjecture for hundreds of years before a proof was provided.
Occasionally, it is shown that a conjecture is independent of the axioms; that
is, neither the truth nor the falseness of the statement follows from the axioms. Two
famous examples are the statements about sets known as the Axiom of Choice and
the Continuum Hypothesis which have been shown to be independent of the original
axioms of Zermelo-Fraenkel Set Theory. The independence of such statements
suggests that the axiom system is not rich enough in structure to establish the truth
of these statements, and that if one chose to do so, those statements could be added
to the list of axioms for the system. The Axiom of Choice or something equivalent
to it, for example, is now usually listed along with the Zermelo-Fraenkel axioms.
One certainly hopes that it is not possible to prove two contradictory statements
about objects in a system. Such an occurrence would say that the axioms of the
system were inconsistent, and this would require the axioms to be changed. After
the original ground rules for Set Theory were established by Georg Cantor in the
1870s and 1880s, Bertrand Russell pointed out in 1901 a paradox (contradiction)
that is a consequence of those rules. Now commonly known as Russells paradox,
it stimulated a flurry of activity which resulted in the young field of Set Theory
being put on a firm foundation (we hope) with the creation and adoption of the
ZermeloFraenkel axioms.
The language of a proof can vary depending on who is writing the proof and
who is the intended reader. In other words, what makes a convincing argument may
well depend on who it is that needs to be convinced. For example, if two experts
in Functional Analysis are speaking to each other, one might prove a statement by
saying Oh, thats just a consequence of the Hahn-Banach Theorem. That proof
might be sufficient since it completely describes the reasoning behind the statement
in question due to the shared knowledge of the two experts. On the other hand, if
one of these experts were speaking to a beginning mathematics graduate student,
the proof would need to include far more detail in order for it to be a convincing
argument. If the expert were speaking to a high school student, the proof might
need to be a complete book that both introduces the needed concepts and explains
many results needed to understand the proof.
It is important to understand that there is a difference between knowing why
a statement is true and knowing how to write a good proof of the statement. It is
quite possible to learn a great deal of mathematics, to be able to solve many types
of mathematical problems, and to understand why particular properties must hold
without being able to write coherent proofs of these properties. It is analogous to
a police detective who has gathered enough evidence to be convinced which of the
many suspects has committed a particular crime, but it is quite another thing to have
the criminal successfully prosecuted in a court of law resulting in the criminals
conviction and eventual punishment for the crime. A student in Analysis needs to
learn many strategies that can be brought to bear when writing proofs. Some of these
strategies are methods or tricks that enter a students bag of tricks which can be
employed later when solving problems or writing proofs. A student of proof writing
needs to learn how to take those strategies and turn them into coherent proofs where
the ideas are presented in a logical order, fill in all necessary details, and make clear
to the proof reader exactly why the chosen strategies justify the needed result.
This book talks about how you should go about writing proofs of the kinds
of statements typically found in the branch of Mathematics called Analysis. The
branches of mathematics are not precisely defined. After a new branch arises, some
mathematicians begin to combine ideas from older branches with ideas from the new
branch to form even newer areas of study. For example, there are branches called
Algebra, Geometry, and Topology. During the twentieth century mathematicians
began talking about Algebraic Topology, Algebraic Geometry, and Geometric
Topology. Very roughly speaking, then, some of the branches of mathematics are
Set Theory: the study of sets, set operations, functions between sets, orderings
of sets, and sizes of sets
Algebra: the study of sets upon which there are binary operations defined (such
as addition or multiplication) and includes Group Theory, Ring Theory, Field
Theory, and Linear Algebra
Topology: the study of continuous functions and properties of sets that are
preserved by continuous functions
Analysis: the study of sets for which there is a measure of distance allowing for
the definition of various limiting processes such as those found in the subjects
of Calculus, Differential Equations, Functional Analysis, Complex Variables,
Measure Theory, and many other areas.
Other areas of study such as Applied Mathematics, Combinatorics, Geometry,
Logic, Probability are considered by some mathematicians to be their own branch
of mathematics or just as part of one or more of the above four branches. The
exact designation is important to some mathematicians and not to others. Although
mathematicians learn to write proofs in each of these branches of mathematics, one
has to begin the learning process someplace. Many teachers feel that Analysis is a
good area to start because students who have completed a study of Calculus will
already be familiar with just about all of the theorems discussed in a beginning
course in Analysis, and may already have an intuitive feeling for why these results
hold. That does not mean that those same students can write convincing proofs of
these theorems. It is the goal of this book to provide the training necessary so that
a student can learn to write proofs of these and similar theorems. Undergraduate
courses in Topology, Group Theory, Advanced Calculus, Graph Theory, and so
forth generally present the beginning concepts in each of these fields and try to
give students a feel for why the major results in the fields are true. Sometimes
this involves having the students learn proofs of these results while other times it
only involves a presentation of definitions and known results with the idea that the
students will be able to take the why it is true and turn it into a proof themselves.
This book is much more interested in turning known strategies into proofs than in
introducing a wealth of new strategies.
1 Point, 1 Region
2 Points, 2 Regions
3 Points, 4 Regions
4 Points, 8 Regions
5 Points, 16 Regions
6 Points, 31 Regions
resolution has alluded mathematicians for years (at least at the time of the writing of
this text in January 2016): The Riemann Hypothesis, the Goldbach Conjecture, the
Twin Prime Conjecture, the P versus NP Problem, and the NavierStokes Equations
Existence and Smoothness Problem. During the last 40 years resolutions have been
announced for several long-standing problems including the Four Color Theorem,
The Bieberbach Conjecture (now called de Brangess Theorem), Fermats Last
Theorem, and the Poincar Conjecture.
Why do mathematicians expend so much effort trying to prove statements, some
of which may seem obvious from the start? One reason is that mathematicians
are very skeptical of statements that appear obvious, and rightfully so. There is a
long history that includes mathematical statements which appear to be true which
are eventually shown to be false. Even very clear patterns can be deceptively
seductive. Take, for example, the following problem. Select a set of n points along
the circumference of a circle, draw the chords between each pair of points, and find
out the maximum number of regions into which these segments can divide the disk.
Figure 1.1 shows the results for the first few values of n.
Although from considering n D 1; 2; 3; 4; 5 it appears that the chords can divide
the disk into 2n1 regions, this fails to be true when n D 6. With a bit more thinking
n1
itnis not hard to see that2n could not be the correct answer. With n points there are
chords and at most 4 intersections of two chords. This number of intersections
2
grows as a fourth-degree polynomial in n suggesting that the number of regions will
li.x/ D
0
dt
:
ln t
In fact, for many years it was thought that li.x/ > .x/ for all x > 0 because this
holds for all small values of x which can be practically checked, for example, all x
between 0 and 1024 . It has now been shown that li.x/ .x/ switches sign infinitely
often, although only for extremely large values of x.
It is apparent that sometimes seemingly very obvious patterns do not hold
in every case, so mathematicians rely on proofs to convince themselves that the
patterns do indeed hold in the general case.
Testing Axiom Systems In the next chapter you will read about the writing of
proofs for some very elementary facts in mathematics; so elementary that you may
wonder why anyone would bother with these proofs. Clearly, it makes sense to
begin any training in the writing of proofs with some very simple results that are
easy to understand so that the student can feel confident about all the statements
being made in the proofs. But these proofs are not being presented just because they
are elementary. When one sets up a mathematical system by making definitions
and determining axioms, it is usually with a particular application or example in
mind. The desired result is that the new system will include the already partially
understood application so that any new discoveries will immediately tell something
new about the original application. Suppose someone sets up an axiom system for
the real numbers, for example, but is not able to prove that addition of real numbers
satisfies the commutative property. Since the commutative property is an important
aspect of addition of real numbers, it would appear that the new axiom system does
not have enough power to represent all that one would want to show about the real
numbers. Perhaps the axiom system will need to be expanded to include an axiom
about the commutativity of addition. Thus, if one cannot prove that the expected
simple properties hold, then it says that something is missing from the axioms. So
mathematicians write proofs to confirm that their axiom systems are representative
of the applications they are trying to describe.
Exhibiting Beauty There are no rules about what composers of music need to
write, but many composers try to write in standardly accepted formats such as
string quartets or symphonies because there are already organizations ready to
perform such works and groups of people happy to listen to such works. Scholars of
literature compare literary works by writing literary analysis, a form which holds
a lot of meaning for those who read and write in that field. Although painters
choose to make pictures of every sort of object or scene, real or imagined, most
painters eventually try their hand at painting some of the standard subjects (still life,
nudes, famous religious or historical depictions). Similarly, mathematicians write
proofs partly because that is what mathematicians enjoy doing. Although many
mathematicians make substantial contributions to the sciences, social sciences, and
arts through the application of their mathematical skills, others live in a world
of creating and discussing abstract concepts that have no immediate application
to real world problems, or at least no application apparent to the mathematicians
doing the research. To them, mathematics is studied as part of the humanities and
is appreciated for its beauty. And much of the beauty of mathematics lies in the
proofs of its theorems. One gets a great deal of pleasure reading a clever proof of
a complicated result when the proof can be stated in just a few lines, especially if
previous proofs of the same result were considerably longer and more difficult to
understand. Many mathematicians like reading articles and attending conferences
where they are exposed mainly to proofs of results, partly so that they can learn
about new results, but more importantly so they can appreciate the techniques
brought to bear to construct the proofs.
Testing Students One should not underestimate the need to educate future mathematicians. A good way to test whether a student understands a particular result is to
ask the student to present a proof of the result. The presentation of a proof shows a
deep understanding of why the result is true and shows an ability to discuss many
details about the objects involved. At the graduate school level in mathematics, most
test problems require the student to produce a proof of a particular result.
The student who has completed a study of Calculus is likely to have mastered
basic skills in Algebra, Geometry, Trigonometry, and Elementary Functions. This
is a good point in ones studies to begin writing proofs. It should not be assumed
that one can just begin writing proofs at this stage even if they have had years of
experience watching teachers and authors present proofs to them any more than
someone can be expected to sit down and begin playing the piano just because they
have watched many other people present concerts using the instrument. In this book
the reader will be taken through the construction of many proofs in a step-by-step
manner that presents the thought process used to write the proofs. Some incorrect
proofs are shown and explained so that the student can learn about common pitfalls
to avoid. Some students dread the transition to writing proofs because they feel
that they do not understand how to write proofs, and are leery of the day when
they will be expected to produce what they cannot now do. But the ability to write
good proofs is a skill no different from the ability to factor polynomials or integrate
rational functions. There is no expectation that the beginner can produce a good
proof, but every expectation that the beginner can learn.
Chapter 2
All of these statements assert that if a function f satisfies the hypothesis that it has
a derivative at a point x, then f must also satisfy the conclusion that f is continuous
at x. Note that the truth of a conditional statement, P.x/ ! Q.x/, suggests nothing
about the truth of the statement Q.x/ ! P.x/ which is known as the converse
Springer International Publishing Switzerland 2016
J.M. Kane, Writing Proofs in Analysis, DOI 10.1007/978-3-319-30967-5_2
10
of the conditional statement P.x/ ! Q.x/. Indeed, the converse of this theorem
is the clearly false statement: If the function f is continuous at a point, then f is
differentiable at that point. Certainly, there are functions f both continuous and
differentiable at a point, but knowing that a function is continuous at a point does
not allow one to conclude that it is differentiable at that point. The converse of a
conditional statement is not logically equivalent to the original statement, but since
the two statements are concerned with the same subject matter, mathematicians
are often interested in the converse of a given conditional. If someone succeeds
in proving a new theorem expressed as a conditional statement, you might wonder
whether the converse of the statement could also be true. Sometimes the truth of the
converse statement is a trivial matter because it is well known. But there are many
examples where the converse does not hold in every case; that is, there are many
known values of x where the converse statement Q.x/ ! P.x/ is false. Other
times, the converse statement is something that has been previously established.
But very often, the truth of the converse statement remains an open question, and
the proof of the original conditional statement may generate research interest in its
converse.
One of the equivalent forms of a conditional statement P.x/ ! Q.x/ is the
statement if Q.x/ is false, then P.x/ must be false. This can be written as
:Q.x/ ! :P.x/ using the negation symbol : . This form of the statement
is called the contrapositive of the original conditional statement. For example, the
contrapositive of the statement discussed above is If the function f is not continuous
at a point, then f is not differentiable at that point. Although logically equivalent
to the original conditional statement, the contrapositive often gives you a different
way to think about the statement, and you will often see a proof which is a proof of
the contrapositive statement instead of a proof of the original conditional statement.
11
Q.x/ ! P.x/ are both true, the two statements can be combined into one as
P.x/
! Q.x/. This can also be stated as P.x/ if and only if Q.x/. Such
statements are called biconditional statements. Thus, the Pythagorean Theorem and
its converse could be combined into the single biconditional statement: A triangle
is a right triangle if and only if the triangle has side lengths a, b, and c satisfying
a2 C b2 D c2 .
8 for all
3 such that
! implies
! if an only if
12
2.1.4 Exercises
Perform the follows steps for each of the conditional statements in Exercises 16.
A
B
C
D
E
1. If x D 1 and y D 1, then xy D 1.
2. If x is an integer, then 2x C 1 is also an integer.
3. f .x/ and g.x/ are both continuous at x D 0 only if f .x/ C g.x/ is continuous
at x D 0.
4. xy D 0 if x D 0 or y D 0.
5. If xy 9y D 0 and y > 0, then x D 9.
6. A rectangle has area xy if two adjacent sides of the rectangle have lengths x and y.
7. Write the following without using shorthand symbols.
(a) 9x 2 R 3 x C 4 D 2.
(b) 8x 2 R 9y 2 R 3 x C y D 10.
13
.
polynomial ax2 C bx C c has roots given by x D
2a
SET THE CONTEXT: Let a, b, and c be constants with a 0.
ASSERT THE HYPOTHESIS: Suppose that x satisfies ax2 C bx C c D 0.
LIST IMPLICATIONS: Since a 0, it follows that x2 C ba x C ac D 0.
b2
b2
Then x2 C ba x C ac C 4a
2 D 4a2 .
b 2
b2
C ac D 4a
Factoring shows that x C 2a
2.
b 2
b2
c
b2 4ac
Then x C 2a D 4a2 a D 4a2 .
b
4ac
This means that x C 2a
must be one of the two square roots of b 4a
2 .
s
p
p
2
2
2
b b 4ac
b 4ac
b
b 4ac
D
, and x D
.
So, x C
D
2
2a
4a
2a
2a
STATE THE CONCLUSION: Thus, the
p roots of the quadratic polynomial
b b2 4ac
ax2 C bx C c are given by x D
.
2a
The proof template begins with the suggestion to SET THE CONTEXT which
represents statements designed to tell the reader what is being assumed in the proof.
This is usually a sentence or two telling the reader about the properties of the objects
that will be encountered in the proof. It may also introduce which variables will
appear in the proof and what kinds of objects they represent. So, in the given proof
of the Quadratic Formula, the first line tells that the variables a, b, and c are going
to represent known constants with a not being 0. Clearly, the fact that a is not 0
needs to be stipulated because if a D 0, the polynomial ax2 C bx C c would not
be quadratic and would not have the proposed roots. Generally, you are not looking
for a lengthy narrative here, and, in fact, brevity is a particularly cherished attribute
of a proof. Saying what needs to be said, but only what needs to be said is usually
best. Some authors who state a theorem and immediately follow the statement of the
theorem with its proof will forgo setting the context at the beginning of the proof
because the reader will have just seen the statement of the theorem and may not need
to see a repeat of the context for that proof. For example, in the example proof, the
statement of the theorem does introduce the constants a, b, and c and polynomial
ax2 C bx C c, so some authors might just skip the first line of the proof. On the
other hand, if the first line of the proof instead introduced the constants r, s, and
t, the proof could have proceeded using these variables instead of a, b, and c. The
same result would have been proved. So the SET THE CONTEXT of the proof
makes the proof independent of the statement of the theorem being proved. Thus, for
completeness, it is good to establish the habit of including the setting of the context
at the beginning of each proof, at least until the students experience in proof writing
has matured.
Your choices of variables used to represent particular objects in the proof are
not critically important to the structure or correctness of the proof, but there are
14
certain variables that mathematicians associate with various uses, and sticking to
these conventional choices simplifies the understanding of the proof because those
variable choices bring with them a history of context that the reader will recognize.
There are very few Algebra students
who would recognize the Quadratic Formula
p
s s2 4rt
if you gave them z D
. Proofs about limits usually refer to the
2r
variables and which represent small positive real numbers used in specific ways
in the proof. Using these two variables in their traditional contexts makes the proofs
easier to understand because the reader will expect these variables to play specific
roles, just as they have in many other proofs the reader has seen. Seeing many
examples of proofs will familiarize the novice proof writer with these traditional
uses of variables.
Suppose that the statement being proved indicates that every object satisfying
the properties listed in the hypothesis of the theorem also satisfies some properties
listed in the conclusion of the theorem. One generally structures a proof of such
a statement by first selecting a generic object satisfying the properties listed in
the hypothesis. The ASSERT THE HYPOTHESIS part of the proof is where the
writer selects an arbitrary element satisfying the hypothesized properties. In the
Quadratic Formula proof, it was assumed that x satisfied the quadratic equation
ax2 C bx C c D 0. Other examples would be statements such as
It is possible that there are infinitely many objects which could play the role of
the generically chosen object. But if an argument proves the result is true for this
generic object, then the theorem will have been shown to hold for any object that
could have played the role of the generic object, and, therefore, the theorem will
have been proved for all objects satisfying the hypothesis. The Quadratic Formula
proof addresses the one generic polynomial ax2 C bx C c and in doing so derives
a formula that works for all quadratic polynomials including 5x2 17x C 126 and
rx2 C sx C t. Often the reader of a proof will form a mental picture of the generic
object being chosen. For example, after reading Let n be any natural number bigger
than 3, the reader may think, OK, how about n D 7? As the proof progresses,
the reader may take each statement of the proof and verify that it is valid and makes
sense for their choice of n D 7. This helps the reader follow the logic of the proof
and verifies that they are understanding what the proof is saying.
The proof will be completed when it is shown that the generically chosen
element satisfying the hypothesis of the theorem is, in fact, an element satisfying
the conclusion of the theorem as stated in the STATE THE CONCLUSION part
of the template. There will certainly need to be some statements placed between
15
the original assertion of the hypothesis and the end of the proof that justify the
conclusion of the theorem. Those statements make up the LIST IMPLICATIONS
part of the template. In almost all cases, most of the body of the proof belongs to
this list of statements. Each statement in the list should follow from definitions or
be simple implications following from previous statements in the proof. In a wellwritten complete proof, the reader should easily see why each implication follows
logically from other statements made earlier in the proof (Fig. 2.1). If an implication
is not clear on its own, it will need some justification so the reader can follow the
logic. The justification may just be a reminder of a key point made earlier in the
proof (as shown earlier, f is continuous at point a) or a reminder of a well-known
definition or theorem (Since all continuous functions on the interval 0; 4 are
R4
integrable there, it follows that f .x/dx exists.) The given Quadratic Formula proof
0
contains six lines of implications. Each line follows easily from the line before using
standard rules of Algebra, and any student familiar with the algebraic manipulations
of equations will be able to understand these implications. In the fourth step of the
b2
proof, the quantity 4a
2 is added to both sides of an equation. Although this step
surely follows the rules of Algebra, it may not be clear to the reader of the proof
why the step is important. As it turns out, this completing the square operation
prepares for the factoring performed in the fifth step of the proof and is arguably the
most clever step of the proof. A proof will often require a clever step such as this.
The proof writer may have labored for years looking for the inspiration needed to
find such a step, but the proof itself need only make clear the justification for what
is being done and does not need to refer to the sweat that went into producing it.
Some implications will be easy for the reader to follow without having to justify
the step. Other statements may need some deeper explanation. Here is where the
proof writer will need to consider the expertise of the target audience for the proof
in order to decide how much detail to provide. How to make your proof easy to
follow is only clear when you know for whom it is meant to be easy. For example, it
b2
made sense to follow the line x2 C ba xC ac D 0 with the statement x2 C ba xC ac C 4a
2 D
b2
4a2
because this just used the fact that you can add equal quantities to both sides of
an equation to get a new equation that is equivalent. On the other hand, suppose you
wish to combine a conditional statement on line 8 of a proof with the fact stated
on line 15 of the proof in order to show that the hypothesis of that conditional
is satisfied. This would allow the writer to state the conclusion of the conditional
16
statement to get line 16 of the proof, but the reader may have to be reminded about
which statements are being combined to get that conclusion.
Sometimes the writer of a long or complicated proof will need to make a new
definition or point out some new property that will be important later in the proof.
Depending on the complexity of the new idea, the proof writer may want to include
an example or two of objects satisfying the new definition or property. This will
serve to help the reader understand the new concept or to verify that the reader
is understanding the new concept. It is admirable to include such examples if the
complexity of the proof can be made clearer. But in most other contexts, the proof
should be kept short without the inclusion of unnecessary statements. If the intended
readers are able to easily construct these examples on their own, then the examples
should be left out of the proof.
The remainder of this chapter will discuss proofs that follow this general proof
template in contexts that the student should find easy to follow. It will also give
an opportunity to present some definitions and notation that will be used in later
chapters.
2.2.1 Exercises
1. If you were writing a proof of All prime numbers greater than 2 are odd, which
of the following would be appropriate ways to begin the proof. (There may be
more than one correct answer.)
(a)
(b)
(c)
(d)
(e)
(f)
(g)
17
3. If you were writing a proof of Every cubic polynomial with real coefficients has
at least one real root, which of the following would be appropriate ways to begin
the proof. (There may be more than one correct answer.)
(a) Assume that every cubic polynomial with real coefficients has at least one
real root.
(b) Assume that p.x/ is a polynomial with at least one real root.
(c) Assume that a, b, c, and d are real numbers with a 0, and let p.x/ D
ax3 C bx2 C cx C d.
(c) The polynomial x3 8 has exactly one real root at x D 2.
(e) Let p.x/ be a cubic polynomial with real coefficients and q.x/ be a cubic
polynomial with complex coefficients.
(f) Let p.x/ be a cubic polynomial with real coefficients with real root r.
Write an appropriate first sentence that would begin proofs of each of the following
statements.
4. If m and n are relatively prime integers, then there exist integers x and y such that
mx C ny D 1.
5. The three angle bisectors of any triangle intersect at a common point.
6. If a and b are real numbers with a b, and f is a function continuous on the
closed interval a; b, then there is a real number M such that jf .x/j M for all
x 2 a; b.
u , !
u !
u .!
7. If !
v , and !
w are 3-dimensional vectors, then .!
v / !
w D!
v !
w /:
18
There are many ways to express the contents of a set. One is to list the elements
such as A D fa; b; cg or B D f1; 3; 5; 7; : : : g. Another way is to use set builder
notation which states that the set consists of all elements satisfying a given property
P.x/ and is written fx j P.x/g, or to emphasize that the elements of the set are also in
set A, it is often written fx 2 A j P.x/g. Examples are fx j x > 0g, fy j y2 C3y2 > 7g,
and ff j f is a function differentiable at x D 3g. Note that a set is determined by the
elements that are in the set. Thus, f1; 2; 3g D f3; 2; 1g D f1; 2; 2; 3; 3; 3; 1; 2; 3g
because all three of these sets contain exactly the same three elements. In some
contexts, mathematicians will talk about a multiset which is an object similar to
a set but allows elements of the collection to appear with different multiplicities.
Thus, f1; 2; 3g and f1; 2; 2; 3; 3; 3; 1; 2; 3g would be different multisets even though,
in the notation of sets, they are the same set.
One special set is the empty set written as ; or fg and is the set that has no
elements. In some contexts there is an understanding of a universal set, U, such
that all other sets under consideration are subsets of U. For example, the sets
A D f1; 2; 3g and B D f2; 4; 6; 8; : : : g can be thought of as subsets of the universal
set U D f1; 2; 3; 4; : : : g.
Take care not to confuse elements and subsets. Remember that sets are collections of elements and sets are subsets of other sets. It is possible that a set contains
other sets as elements, but this would need to be explicitly clear from the definition
of that set. It is correct to write 3 2 f1; 2; 3; 4; 5g, f1; 3; 5g f1; 2; 3; 4; 5g,
and f1; 2g 2 f1; f1; 2g; f1; f1; 2ggg, but it is incorrect to write 3 f1; 2; 3; 4; 5g,
f1; 2g 2 f1; 2; 3; 4; 5g, or ; 2 f1; 2; 3; 4; 5g.
The student should be familiar with the following standard set operations. The
union of sets A and B is A [ B D fx j x 2 A or x 2 Bg, and the intersection
of sets A and B is A \ B D fx j x 2 A and x 2 Bg. When there is an
understood universal set, U, it makes sense to refer to the complement of a set
Ac D fx 2 U j x Ag. It does not make sense to discuss the complement of a set if
there is no understood universal set. For example, is f1; 2; 3gc D f4; 5; 6; 7; : : : g,
or is it f: : : ; 3; 2; 1; 0; 4; 5; 6; 7; : : : g? For that matter, is your right shoe an
element of f1; 2; 3gc ? The difference of two sets is AnB D fx 2 A j x Bg,
and, equivalently, if there is an understood universal set, AnB D A \ Bc . For
example, if A D f1; 2; 3; 4; 5g and B D f2; 4; 6; 8g, then A [ B D f1; 2; 3; 4; 5; 6; 8g,
A \ B D f2; 4g, AnB D f1; 3; 5g, and BnA D f6; 8g.
2.3.2 Exercises
1. Which of the following statements are true?
(a)
(b)
(c)
(d)
6 2 f1; 2; 3; : : : ; 10g.
f3; 5g 2 f1; 2; 3; : : : ; 10g.
; 2 f1; 2; 3; : : : ; 10g.
f6; 8g f1; 2; 3; : : : ; 10g.
19
A[A
A\B
.A [ B/ \ C
.B [ C/ \ A
.A \ B/nC
.AnB/ [ C
The first four of these statements propose that one set is a subset of a second set.
From the definition of subset, for A B to be true, it is required that for every
x 2 A, x must also be in B. There is a standard template for proofs of statements of
this form:
20
SET THE CONTEXT: Suppose that A and B are any two sets.
ASSERT THE HYPOTHESIS: Let x 2 A.
LIST IMPLICATIONS: Since x 2 A, it is true that x 2 A or x 2 B.
By the definition of set union x 2 A [ B.
STATE THE CONCLUSION: Therefore, by the definition of subset,
A A [ B.
Do the statements of this proof have to appear in exactly this order using exactly
these words? Of course not. There can be many variations in what makes up a good
proof. But it does not hurt to review why these statements make a good proof. The
first line Suppose that A and B are any two sets just makes it clear to the reader
that the variables A and B can be used to represent any two sets. Here is where the
reader of the proof may well mentally choose two sets so that when reading the
remainder of the proof, the reader can verify that the statements make sense when
applied to those two sets. The second line Let x 2 A is required because by the
definition of subset, one must show that each element of A is also an element of
A [ B, so selecting an arbitrary element of A is the natural way to do this. The next
line Since x 2 A, it is true that x 2 A or x 2 B is just a statement of logic that says
if statement p is true, then statement p or q is also true. Of course, this particular
p or q statement is exactly the definition of x being a member of A [ B, which is
exactly what is needed to complete the proof.
21
Could one have interchanged the third and fourth lines of this proof? Well, yes;
the proof would be complete if that were done, but the fact that the definition of set
union is invoked right after its conditions are verified makes the statements of the
proof flow smoothly. The reader facing the definition of set union in line three might
wonder why that definition is being shown at that point. By placing that statement
as the fourth statement where the proof reader has just seen that x 2 A or x 2 B,
the proof reader will immediately see that the definition of set union applies. Note
that each of the five statements in the proof has been placed on a separate line in the
display box. This has been done merely to facilitate the discussion about that proof.
In practice, there is no requirement that these statements appear on a separate lines.
The second statement about all sets is A \ B A. This can be proved using the
same proof template as the first statement. Since this statement also applies to any
two sets A and B, the first line of this proof will be the same as the first line of the
previous proof. Because the assertion of the statement being proved is that A \ B is
a subset of another set, the ASSERT THE HYPOTHESIS line of the proof would
change to the assertion that x belongs to A \ B. After reading this second line, what
does the proof reader know about x? Only that x belongs to the intersection of two
sets. Thus, that only direction that the proof can proceed is to invoke the definition
of set intersection to make the additional assertion that x 2 A and x 2 B. This is a
statement of the form p and q, so logic allows the assertion that p is true, or, in this
case, that x 2 A. This is the required STATE THE CONCLUSION statement, and
the complete proof would be
PROOF: A \ B A.
SET THE CONTEXT: Suppose that A and B are any two sets.
ASSERT THE HYPOTHESIS: Let x 2 A \ B.
LIST IMPLICATIONS: By the definition of set intersection, x 2 A and
x 2 B.
Thus, x 2 A.
STATE THE CONCLUSION: Therefore, by the definition of subset,
A \ B A.
For a more substantial example, consider the third of the list of statements about
sets An.B [ C/ .A [ B/nC. A proof of this statement will need to refer to the
definition of set difference as well as the definitions of set union and subset. Since
the statement being proved involves three sets, the SET THE CONTEXT part of
the proof will need to refer to all three sets. The ASSERT THE HYPOTHESIS
statement will need to select an arbitrary element from An.B [ C/. To emphasize
that the choice of which variable to use is arbitrary, this time use y rather than x to
represent the arbitrarily chosen element. Once it is known that y 2 An.B [ C/,
the only property of y that can be used is the fact that y is a member of a set
difference. Thus, this would be a good time to invoke the definition of set difference.
That assures that y 2 A and y .B [ C/. At that point one can use the definition of
22
set union to conclude that since y .B [ C/ that y B and y C. Now these facts
can be combined to get the STATE THE CONCLUSION statement required by
the proof template. The complete proof would be
PROOF: An.B [ C/ .A [ B/nC.
SET THE CONTEXT: Suppose that A, B, and C are any three sets.
ASSERT THE HYPOTHESIS: Let y 2 An.B [ C/.
LIST IMPLICATIONS: By the definition of set difference, y 2 A and y
.B [ C/.
By the definition of set union y cannot be an element of either set B or set
C, or it would be in B [ C.
Also by the definition of set union, since y 2 A, y is also a member of A [ B.
Now, y 2 .A [ B/ and y C, so by the definition of set difference, y 2
.A [ B/nC.
STATE THE CONCLUSION: Therefore, by the definition of subset,
An.B [ C/ .A [ B/nC.
2.3.4 Exercises
Write proofs for each of the following statements.
1. For all sets A, B, and C, .A \ B/ \ C A \ C.
2. For all sets A, B, and C, .A \ B/ \ .A \ C/ B \ C.
3. For all sets A, B, and C, .AnB/ \ .AnC/ An.B \ C/.
23
24
PROOF: A [ B D B [ A.
SET THE CONTEXT: Suppose that A and B are any two sets.
PART 1 A [ B B [ A
PART 2 B [ A A [ B
In fact, the first half of the proof is the second half of the proof. The first half of
the proof shows that A[B B[A for any two sets A and B. In particular, that proof
applies when the roles of the two sets are interchanged; just let the variable A in the
25
first part of the proof represent the set B from the second part of the proof, and let
the variable B in the first part of the proof represent the set A from the second part
of the proof.
The Associative Law of Intersection refers to three sets and requires repeated
use of the definition of set intersection. The definition is used to break down the
statement x 2 .A \ B/ \ C into the three simple statements x 2 A, x 2 B, and x 2 C
and then these facts are put back together to form the needed x 2 A\.B\C/. Again,
the proof needs two parts. The result is
PROOF: .A \ B/ \ C D A \ .B \ C/.
Suppose that A, B, and C are any three sets.
PART 1 .A \ B/ \ C A \ .B \ C/
Let x 2 .A \ B/ \ C.
By the definition of set intersection, x 2 .A \ B/ and x 2 C.
Also, by the definition of set intersection, x 2 A and x 2 B.
Thus, x 2 A, x 2 B, and x 2 C.
Since x 2 B and x 2 C, by the definition of set intersection x 2 B \ C.
Since x 2 A and x 2 B \ C, by the definition of set intersection
x 2 A \ .B \ C/.
Hence, from the definition of subset, it follows that .A \ B/ \ C
A\ .B \ C/.
PART 2 A \ .B \ C/ .A \ B/ \ C
Now, let x 2 A \ .B \ C/.
By the definition of set intersection, x 2 A and x 2 B \ C.
Also, by the definition of set intersection, x 2 B and x 2 C.
Thus, x 2 A, x 2 B, and x 2 C.
Since x 2 A and x 2 B, by the definition of set intersection x 2 A \ B.
Since x 2 A \ B and x 2 C, by the definition of set intersection
x 2 .A \ B/ \ C.
Hence, from the definition of subset, it follows that A\.B\C/ .A\B/\C.
Therefore, because .A \ B/ \ C and A \ .B [ C/ are subsets of each other,
by the definition of set equality .A \ B/ \ C D A \ .B \ C/.
The two DeMorgans Laws are useful because they tell how to simplify the
complement of a set formed by a combination of unions and intersections of sets.
Proving these laws can follow the template for showing set equality. The proofs will
need to refer to the definitions of set union, set intersection, and set complement.
The order in which these definitions are invoked follows from what is known at that
point of the proof. For example, if you know that x 2 .A [ B/c , then the only way
to make progress in the proof is to apply the definition of set complement because
the only attribute known about the set is that it is the complement of some other set.
26
(A B)C
AC BC
Yes, that other set is a union of two sets, but there is no way to use that information
at this point of the proof because complementation was performed after the union
was taken (Fig. 2.2).
PROOF: .A [ B/c D Ac \ Bc .
Suppose that A and B are any two sets.
PART 1 .A [ B/c Ac \ Bc
Let x 2 .A [ B/c .
By the definition of set complement, x .A [ B/.
If x 2 A or x 2 B, then x 2 A [ B which is false.
Thus, x A and x B, so by the definition of set complement, x 2 Ac and
x 2 Bc .
By the definition of set intersection x 2 Ac \ Bc .
Hence, from the definition of subset, it follows that .A [ B/c Ac \ Bc .
PART 2 Ac \ Bc .A [ B/c
Now, let x 2 Ac \ Bc .
By the definition of set intersection, x 2 Ac and x 2 Bc .
Thus, by the definition of set complement, x A and x B.
If x 2 A [ B, then by the definition of union, it would follow that x 2 A or
x 2 B which is false.
Thus, x A [ B, and, by the definition of set complement, x 2 .A [ B/c .
Hence, from the definition of subset, it follows that Ac \ Bc .A [ B/c .
Therefore, because Ac \ Bc and .A [ B/c are subsets of each other, by the
definition of set equality .A [ B/c D Ac \ Bc .
2.3.6 Exercises
Give that A, B, and C are sets, write proofs for each of the following statements.
1. A \ B D B \ A.
2. A \ .BnA/ D ;.
3. .AnB/ [ .BnA/ D .A [ B/n.A \ B/.
4.
5.
6.
7.
27
.A [ B/ [ C D A [ .B [ C/.
.A [ B/ \ C D .A \ C/ [ .B \ C/.
.A \ B/ [ C D .A [ C/ \ .B [ C/.
An.B [ C/ D .AnB/ \ .AnC/.
28
definition because it provides a fairly easy way to check whether a given integer is
even, and because knowing that a number n is even immediately gives you a number
k for which n D 2k, and that is a powerful tool for proving facts about even integers.
For this reason, this chosen definition is called the working definition, that is, it
is the definition easiest to apply in the wide variety of contexts. It is the definition
chosen from which all other properties of even numbers can be derived.
A similar discussion could take place about how to define odd integers. The
working definition is that n 2 Z is odd if there is a k 2 Z such that n D 2k C 1.
There is a long list of facts you could prove about even and odd numbers.
Facts About Even and Odd Integers
Together, the first two of these facts say that each integer is either even or odd
but not both. This says that the sets of even and odd integers form a partition of
Z, that is, the sets are disjoint and the union of the sets is all of Z. Some authors
require that all the sets of a partition be nonempty as in the case with even and
odd integers. So why is it that every integer is either even or odd? This depends
on the Division Algorithm that states that if m; n 2 Z with n > 0, then there are
unique q; r 2 Z with 0 r < n such that m D nq C r. In this case q is called
the quotient of the division, and r is called the remainder of the division. Using
the Division Algorithm, any integer m can be divided by 2 giving a quotient and
remainder where the remainder is either 0 or 1. If the remainder is 0, then m D 2q
for integer q implying that m is even, and if the remainder is 1, then m D 2q C 1 for
integer q implying that m is odd.
29
definition of being an even integer or the definition of being an odd integer. The
result would be
PROOF: Ever integer is either even or odd.
Let m be an integer.
By the Division Algorithm there are integers q and r with 0 r < 2 such
that m D 2q C r.
If r D 0, then m D 2q for integer q which means that m satisfies the
definition for being even.
If r D 1, then m D 2q C 1 for integer q which means that m satisfies the
definition for being odd.
Since r must be either 0 or 1, it follows that every integer is either even
or odd.
Next consider the how to prove the statement The sum of any two odd integers is
even. The statement concerns the sum of any two odd integers, so the proof reader
would expect the proof to consider two arbitrarily chosen odd integers. Once two
odd integers are chosen, the definition of odd integer should be invoked because, at
that point, that is the only information that is known about the two integers. Finally,
a little algebra will help to show that the sum of these two odd integers satisfies the
definition of even integer. Here is an attempt to write such a proof that makes several
common proof writing errors.
PROOF ATTEMPT: The sum of any two odd integers is even.
The two integers are odd, so each has the form 2k C 1.
The sum of these two integers is .2k C 1/ C .2k C 1/ D 4k C 2.
k could be even or odd.
The number 2 is even since it is 2 1.
4k is even since it is 2 2k.
The sum of two even numbers is even, so the sum of 4k and 2 is an even
number.
Therefore, the sum of two odd integers is always even.
30
start with an odd integer, say m, and then represent it as 2kC1 rather than starting
with 2k C 1. The subtle point is that one should start with odd integer and use
its definition to move on to 2k C 1 rather than starting with 2k C 1 which jumps
the gun. The reader of the proof could wonder whether 2k C 1 could represent
a generic odd integer. Well, it can, but this takes some thought which can be
avoided by starting with an odd integer m and then using the definition of odd to
select the integer k such that m D 2k C 1.
The definition of odd integer does refer to 2k C 1, but it is more precise. It
does not say has the form. It says that there is an integer k such that the odd
number equals 2k C 1:
It is a major error to allow both odd integers to equal 2k C 1 for the same number
k. The only way this can happen is for the two odd integers themselves to be
equal. Thus, this proof only applies to a small subset of cases where one adds
two identical odd integers together such as 3 C 3 or 117 C 117.
The statement k could be even or odd is certainly correct, but it does not
contribute to the proof. It is a statement about items in the proof that is not part
of the proof. Occasionally, one will make a definition as part of a long proof,
and then give some examples to help the reader understand that definition. But
if a statement is not needed either as a critical step in a proof or an important
illustration to aid the understanding of the proof, then the statement should be
left out of the proof because it distracts from the proof and complicates it.
The statement The sum of two even numbers is even is correct, but it has not
been proved yet, at least in this text, and is equivalent in difficulty to proving the
corresponding statement about the sum of odd integers. Thus, it is not appropriate
to use the result about sums of even integers to prove one about the sum of odd
integers.
31
2.4.3 Exercises
Write proofs for each of the following statements.
1.
2.
3.
4.
5.
32
The binary operations satisfy conditions which make Q into a field. A field F
is a set with operations of addition and multiplication that satisfies the following
axioms.
Axioms for a Field F
A set F together with the binary operations of addition .C/ and multiplication
./ form a field if F contains the two elements 0 and 1 with 0 1 such that for
every r; s; t 2 F
r C s 2 F and
r s 2 F and
r Ds ! rCt DsCt
r Ds ! rt Dst
the Closure
Properties
.r C s/ C t D r C .s C t/
.r s/ t D r .s t/
the Associative
Properties
the Commutative
rCsDsCr
rsDsr
Properties
rC0Dr
r1Dr
the Identity
Properties
There exists r 2 F
If r 0, there exists 1r 2 F
the Inverse
such that r C .r/ D 0
such that r 1r D 1
Properties
r .s C t/ D r s C r t
the Distributive
Law of
Multiplication
Over Addition
Notice that the rational numbers do satisfy the eleven field axioms. One defines the
operation subtraction () by r s D r C .s/ and the operation division ( ) for
s 0 by r s D r 1s D rs . Moreover, the field Q together with the less than order
relation is an ordered field that obeys the following axioms.
Axioms for an Ordered Field F
A field F is an ordered field with order relation < if for every r; s; t 2 F
exactly one of the following holds
r < s, r D s, s < r
r < s and s < t imply r < t
r < s implies r C t < s C t
r < s and 0 < t imply r t < s t
33
Notice that the rational numbers do satisfy the four ordered field axioms. One
defines the other order relations of greater than (>), greater than or equal to
(), and less than or equal to () in the obvious ways, that is, r > s whenever
s < r, r s whenever either r > s or r D s, and r s whenever either r < s or
r D s.
There are many other ordered fields, and it is constructive to consider how to
justify the fifteen ordered
field axioms for a different ordered field. For example,
p
the set T D fr C s 2 j r; s 2 Qg is an ordered field
p using the usual
p addition and
multiplication operations.
For
two
elements
a
C
b
2
and
c
C
d
2 in T, define
p
p
p
addition as
.a
C
b
2/
C
.c
C
d
2/
D
.a
C
c/
C
.b
C
d/
2
and
multiplication
p
p
p
as .a C b 2/ .c C d 2/ D .ac
C
2bd/
C
.ad
C
bc/
2.
To
define
the less than
p
p
p
relation you would want .a C b 2/ < .c C d 2/ whenever
a
c
<
.d b/ 2
p
which can be checked by squaring both a c and .d b/ 2, although you will
need topalso considerpthe signs of a c and d b. Thus, the definition becomes
.a C b 2/ < .c C d 2/ if one of the following holds:
a c < 0 and 0 < d b,
0 < a c, 0 < d b, and .a c/2 < 2.d b/2 , or
a c < 0, d b < 0, and .a c/2 > 2.d b/2 .
It is fairly easy to check that T is an ordered field. The only field axiom which
does not follow immediately from the properties of rational
p numbers is the inverse
axiom for multiplication. You should verify that for a C b 2 0, its multiplicative
inverse is
a
b p
C
2
a2 2b2
a2 2b2
which is in T. The order axioms take more work to verify due to the complicated
definition of less than. For example, to verify the less than relation
p
p works correctly
2, c C d 2, and
with addition,
one
would
begin
with
three
elements
of
T,
a
C
b
p
p
p
e C f p2 where it ispgiven that a C
p b 2 < cpC d 2. One needs to compare
.a C b 2/ C .e C f 2/ with .c C d 2/ C .e C f 2/. To do this, one compares the
values of .a C e/ .c C e/ D a c and .d C f / .b C f / D d b. But this reduces to
comparing
p a c and
p d b which are known to satisfy the correct conditions because
a C b 2 < c C d 2 was given.
Every ordered field satisfies a long list of simple properties that you will
associate with facts learned in Arithmetic and Algebra. Here are some of those
properties.
34
r 0 D 0.
If r C t D s C t, then r D s.
If r t D s t and t 0, then r D s.
.r/ D r.
If r 0, then 11 D r.
r
r D s if and only if r D s.
r D .1/ r.
.r/ C .s/ D .r C s/.
.r/ .s/ D r s.
If r < s and t < 0, then s t < r t.
If r 0, then r2 > 0.
0 < 1.
If 0 < r, then 0 < 1r .
If 0 < r < s, then 0 < 1s < 1r .
If n is any natural number and r > 1, then rn < rnC1 .
The reader may wish to prove some of these properties by applying the axioms.
This book will not dwell on these proofs since the techniques used in proving them
are not essential for writing most proofs in Analysis. Two simple proofs are given
here as examples.
PROOF: If r is any element of the field F, then r 0 D 0.
Let r be an element of field F .
Since 0 is the additive identity of F , 0 D 0 C 0.
Then r 0 D r .0 C 0/.
By the Distributive Law, r 0 D r 0 C r 0.
By adding r 0 to each side of this equality, one gets 0 D r 0 r 0 D
.r 0 C r 0/ r 0 D r 0 C .r 0 r 0/ D r 0 C 0 D r 0.
Therefore, for any r 2 F , r 0 D 0.
The next theorem essentially says that if .1/ r has the same properties as r, it
must equal r.
35
.1/ r C 0
.1/ r C .r C r/
.1/ r C r C r
.1/ r C 1 r C r
.1/ C 1 r C r
0 r C r
0 C r
r
Additive Identity
Additive Inverses
Associative Law of Addition
Multiplicative Identity
Distributive Law
Additive Inverses
r0D0
Additive Identity
36
are all upper bounds of S, but 5 is the least upper bound of S. Also, 2, 0, and 12
are all lower bounds of S, but 1 is the greatest lower bound of S. One often uses the
notation l.u.b..S/ or sup.S/ to represent the least upper bound or supremum of S
and g.l.b..S/ or inf.S/ to represent the greatest lower bound or infimum of S.
Axioms for the Real Numbers
The real numbers, R, is an ordered field that satisfies The Completeness
Axiom:
Every nonempty set S R which is bounded above has a least upper bound
in R.
Note, for example, that the set S D fx 2 Q j x2 < 7g is a nonempty subset of Q
which is bounded above by 4, 3, and 2.7, but there is no element of Q which is a
least upper bound of p
S. The set of real numbers, though, does contain a least upper
bound of S, namely 7. The Completeness Axiom is sometimes called the Least
Upper Bound Principle. The Completeness Axiom comes up frequently in proofs
about the real numbers to show that numbers with particular properties exist. For
example, consider the two theorems, the Archimedian Principle and the Existence
of Square Roots. Both of these theorems are easily understood, but they cannot be
proved without using the Completeness Axiom.
The Archimedian Principle states that for every real number r there is a natural
number greater than r. It can be proved using a proof by contradiction. The proof
makes the assumption that there is a real number greater than every natural number
and uses this to derive a contradiction, a statement that is false. Because one cannot
derive a false statement from a true statement, the assumption most recently made
in the proof must be a false statement, and you can conclude that no real number
exists that is greater than every natural number.
PROOF (Archimedian Principle): If r 2 R, then there exists n 2 N such
that r < n.
Suppose that there is an r 2 R such that r > n for every n 2 N.
Then the set N is a nonempty subset of R with an upper bound, so by the
Completeness Axiom, N has least upper bound M.
Then M 1 < M, so M 1 is not an upper bound for N.
Thus, there is a k 2 N with the property that k > M 1.
But then k C 1 is also in N, yet k C 1 > .M 1/ C 1 D M where M is an
upper bound for N.
This is a contradiction since no element of a set can be greater than an upper
bound for that set.
Therefore, the assumption that r > n for every n 2 N must be false, and for
every r 2 R there must be at least one n 2 N with n > r.
37
You may not have ever doubted that every nonnegative real number has a square
root, but this is a fact that can be proved using the axioms for the real numbers. It is
a nice application of both the Trichotomy Property and the Completeness Axiom.
Given a positive real number, r, the proof constructs the set S D fx 2 R j x2 rg
and then uses the Completeness Axiom to exhibit a value, s, equal to the least upper
bound of S. Then it shows that s2 cannot be greater than r and cannot be less than r,
so by the Trichotomy Property, s2 must equal r.
In particular, the proof first assumes that s2 > r and shows that there is a number
y > 0 such that the square of s y is also greater than r. This shows that s y is an
upper bound for S which contradicts the fact that s is the least upper bound of S. The
2
proof magically suggests that y D s 4sr works. Where did this magical expression
for y come from? It came from considering what property you would want such a y
to have. If you want .s y/2 > r, this suggests that you want s2 2sy C y2 > r. This
inequality is quadratic in y and has an unnecessarily messy solution. But one of the
most important lessons about writing proofs in Analysis is that one can often be a
little sloppy when trying to show that an inequality holds. Here, for example, rather
than finding a y such that s2 2sy C y2 > r, it would be sufficient to find a y such
that s2 2sy > r, because if s2 2sy > r, then certainly the needed s2 2sy C y2 > r
also holds. The advantage of making this change is that the inequality s2 2sy > r
2
2
is very easy to solve for y yielding y < s 2sr . Thus, the value y D s 4sr ought to work
fine, and, hence, the magic is demystified. Of course, there are many other possible
values of y that would also have worked in this proof, but only one value for y is
needed.
After showing that s2 > r cannot be true, the proof assumes that s2 < r and
shows that there is a number y > 0 such that the square of s C y is less than r.
This shows that s C y is in S which contradicts the fact that s is an upper bound of
2
S. Again, the proof just suggests setting y D rs
. Can you figure out where this
4s
expression for y came from? Indeed, the calculation is similar to the one above. You
need .s C y/2 r, so s2 C 2sy C y2 r. It is simpler if y2 could be replaced by
2sy. If you assume y 2s, it allows you to conclude y2 2sy so that .s C y/2 D
s2 C 2sy C y2 s2 C 2sy C 2sy D s2 C 4sy. You then want a y that satisfies
2
gives the needed value of y (Fig. 2.3). Putting
s2 C 4sy r. Thus, the value y D rs
4s
these ideas together gives the following proof.
s2 r s
4s
s2
S s r 4s
p
r
38
If s2 > r, note that y D s 4sr > 0, and .s y/2 D s2 2sy C y2 > s2 2sy D
2
2
s2 2s s 4sr D s 2Cr > r. Because .s y/2 > r, it follows that s y < s is an
upper bound of S. This contradicts the fact that s is the least upper bound of
S. Therefore, s2 > r must be false.
2
39
x+y
x
lengths of two sides of a triangle always exceeds the length of the third side of the
triangle (Fig. 2.4). One simple proof of the triangle inequality is
PROOF (Triangle Inequality): jx C yj jxj C jyj
Let x and y be elements of R.
Then jxj x jxj and jyj y jyj.
Adding these inequalities yields .jxj C jyj/ x C y .jxj C jyj/.
This last inequality is equivalent to jx C yj jxj C jyj.
A subset S contained in R is called connected if it has the property that for any
two numbers in S, all the numbers between those two numbers are also in S. More
precisely, S is connected if for all x; y 2 S with x < y, it follows that z 2 S for all z
with x < z < y. Informally, this means that there are no holes in the set S. Another
word for a connected set of real numbers is an interval. If a < b are real numbers,
all of the following sets are intervals.
Intervals of Real Numbers
; empty set
fag D a; a single point
fx j a < x < bg D .a; b/ open bounded interval
fx j a x bg D a; b closed bounded interval
fx j a x < bg D a; b/ bounded interval open on the right
fx j a < x bg D .a; b bounded interval open on the left
fx j a < xg D .a; 1/ open right infinite interval
fx j x < bg D .1; b/ open left infinite interval
fx j a xg D a; 1/ closed right infinite interval
fx j x ag D .1; b closed left infinite interval
R entire real line
40
2.5.4 Exercises
1. Show that for any real number x it follows that jxj C jx 6j 6.
2. Show that for any real number x, jx 1j C jx 3j 2.
3. Show that for any real numbers x and y it follows that jx2 C3x4yjCjx14yj
.x C 1/2 .
4. Show that for any real numbers x and y, jx C yj C jx y 2j C j2x C 8j 10.
5. Show that for any real numbers x and y it follows that jxjCj3x5yjCj5x4yj
jx C yj.
6. Show that the intersection of any two intervals is always an interval.
7. Under what conditions is the union of two intervals an interval?
2.6 Functions
2.6.1 Function, Domain, Codomain
Intuitively, a function is a mapping that assigns to each point of some domain A a
value that resides in some codomain B. This is usually written f W A ! B. More
precisely, the function f is defined as a set of ordered pairs .x; y/ where each x resides
in the domain A of f and each y resides in the codomain B of f , and for each x 2 A
there is exactly one y 2 B such that .x; y/ 2 f . Since there is a unique ordered pair
.x; y/ 2 f for each x 2 A, f associates or links the value of y to the value of x and
allows one to write f .x/ D y.
2.6.2 Surjection
The domain of the function f is exactly the set of all x that are first coordinates of
the order pairs in f , that is, the domain is A D fx j .x; y/ 2 f g. The range of f is
defined as the image of f , that is, the range is fy j .x; y/ 2 f g. Clearly, the codomain
of f can be any set that contains the range of f . This can lead to some confusion
since the codomain of f is not precisely defined. It is simply a convenience. When
one defines a function f W R ! R, one means that f is defined for every real number,
and that for any x 2 R, the value f .x/ also lies in R. This is the case whether or not
R is the range of f or if the range of f is actually some proper subset of R. It could
be difficult and unnecessary to calculate exactly which subset of R is the range of
f , so it might be easier to just give the codomain as R and avoid the technicalities
of figuring out just what values of R are in the range of f . For example, the function
f .x/ D 3x6 15x4 C 12x3 C 25x2 32x C 14 maps the real numbers into the real
numbers, but to find the range of f , you would need to find the minimum value
of f . This minimum exists, but it may not be possible to give its value explicitly.
2.6 Functions
41
. There is no need for the proof to display the steps of solving the
x D y1
5
equation for x. The goal is to produce a value of x 2 A such that f .x/ D y; how you
arrived at that x is not important. It may be interesting, but it is not an essential part
of the proof, and, therefore, it should not be part of the proof.
PROOF: The function f .x/ D 5x2 C 1 is a surjection from the negative
real numbers onto the interval .1; 1/.
Let f .x/ D 5x2 C 1.
q
.
For any y > 1 let x D y1
5
> 0, so
Because y > 1, y1
5
negative real number.
y1
5
q 2
C 1 D y.
Moreover, f .x/ D 5x C 1 D 5 y1
C 1 D 5 y1
5
5
2
Therefore, f is a surjection.
2.6.3 Injection
The definition of function requires that each value x in the domain of f is found in
exactly one ordered pair .x; y/ 2 f . The same does not have to hold for values in
42
the codomain, that is, one value y in the codomain could appear in many order pairs
.x; y/ 2 f . For example, for the constant function f W R ! R given by f .x/ D 1 for
all x 2 R, the value 1 appears as the second coordinate in all the ordered pairs of
the function. If a function has the property that no value of y appears as the second
coordinate of more than one ordered pair in f , then f is said to be injective or, less
formally, that f is one-to-one. In this case the function f is called an injection. In
such a case, one sees that f .x1 / D f .x2 / only if x1 D x2 . This gives a procedure for
proving that a function is injective.
TEMPLATE for proving a function f is injective
SET THE CONTEXT: Make a statement introducing f , its domain A, and
its codomain B.
Assume that for two values x1 and x2 in A that f .x1 / D f .x2 /.
Show that x1 D x2 .
STATE THE CONCLUSION: Therefore, f is an injection.
p
For example, the function f .x/ D 4x C 7 maps the positive real numbers to the
positive real numbers. It is not a surjection, but it is an injection. The proof would
require that you show that f .x1 / D f .x2 / implies that x1 D x2 . Again, this is just an
algebraic manipulation.
p
PROOF: The function f .x/ D 4x C 7 is an injection from the positive
real numbers to the positive real numbers.
p
Let f .x/ D 4x C 7.
Assume
p that for positive
p real numbers x1 and x2 , f .x1 / D f .x2 /.
Then 4x1 C 7 D 4x2 C 7.
Squaring yields 4x1 C 7 D 4x2 C 7, so 4x1 D 4x2 , and x1 D x2 .
Therefore, f is an injection.
If a function f W A ! B is both surjective and injective, that is, if f is both one-toone and onto, then f is bijective, and f is called a bijection. In this case, f exhibits
a one-to-one correspondence between the set A and the set B.
Two functions f and g whose ranges are in the real numbers can be combined
arithmetically. Specifically, one can define f C g, f g, fg, and gf in natural ways:
.f C g/.x/ D f .x/ C g.x/,.f g/.x/ D f .x/ g.x/, .fg/.x/ D f .x/ g.x/, and,
f .x/
for x such that g.x/ 0, gf .x/ D g.x/
. When functions f and g are combined
in this way, the domain of the sum, difference, product, or quotient is assumed to
be the intersection of the domain of f and the domain of g with the exception that
the domain of gf also excludes values of x for which g.x/ D 0. Thus, the function
p
p
f .x/ D x 4 is defined for all x 4, and
g.x/ D 5 x is defined
p the function
p
for all x 5. It follows that the function x 4 C 5 x is defined only for those
x satisfying 4 x 5. Similarly, the function ff .x/
is only defined for x > 4 even
.x/
though it is identically 1 for those x. That function has a natural extension to all real
numbers.
2.6 Functions
43
g(x)
y
f(x)
B
z
A
fg
2.6.4 Composition
If g is a function assigning values in its domain A to values in its range contained
in the set B, and if f is a function assigning values in its domain B to values in
its range contained
in
the set C, then the composition of f with g is the function
.f g/.x/ D f g.x/ which assigns to values in its domain A values in its range
contained in set C (Fig. 2.5). The main reason for considering compositions is that
it is often easiest to represent complicated functions as compositions of simpler
2x
functions. For example, the function f .x/ D psinxC4
is clearly a quotient where the
numerator is the composition of the function p
x2 with the function sin x, and the
denominator is the composition of the function x with the function x C 4.
It is easily shown that if g W A ! B and f W B ! C are both surjective functions,
then their composition, f g W A ! C, is also surjective. To prove this, you would
follow the template for proving that a function is surjective. That requires that you
select an arbitrary z 2 C and show that there is an x 2 A such that .f g/.x/ D z.
Why might you use the variable z here rather than the variable y? Well, that allows
you to think of g as mapping x to y, and f , in turn, mapping y to z. Faced with the
statement f g.x/ D z, there is little you can do except to apply what you know
about the function f , that is, that f is surjective. Because f is surjective, and z is in
the codomain of f , you know that there is a y in the domain of f such that f .y/ D z.
Can you find an x such that g.x/ D y? Of course y is in the domain of f which is the
codomain of g. The function g is surjective, so there must be an x in the domain of
g that maps onto y. These ideas give the following proof.
44
2.6.5 Exercises
Write a proof for each of the following statements.
1. For each real number r there is a real number x such that x3 D r.
2. For each real number r 0 there is a real number x 0 such that x4 D r.
3. If n is an odd positive integer, then for each real number r there is a real number
x such that xn D r.
4. If n is an even positive integer, then for each real number r 0 there is a real
number x 0 such that xn D r.
5. If h W A ! B, g W B ! C, and f W C ! D are three functions, then .f g/ h D
f .g h/. In other words, function composition is associative. (Hint: Show that
both functions .f g/ h and f .g h/ give the same result when applied to an
x 2 A.)
2.6 Functions
45
Chapter 3
Limits
open interval in R containing the point a. This limit should give you a mental image
similar to Fig. 3.1 where the graph of the function gets close to L as x approaches a.
So, how can you quantify what f .x/ is getting close to L means? Is within 1
1
close? Is within 14 close? Is within 1000
close? Clearly, there needs to be a way to say
arbitrarily close or as close as one likes. Analysts have found that a good way
to express f .x/ getting arbitrarily close to L is to say that for any positive distance,
jf .x/ Lj can be made to be less than that distance. Of course, jf .x/ Lj cannot be
made to be negative, and it is not reasonable to require it to be zero since that would
require f .x/ to actually equal L. Hence, one usually says that for any > 0, one can
achieve jf .x/ Lj < . The use of the Greek letter (epsilon) is arbitrary, but the
tradition of using in this context has been universal since Cauchy introduced its
use in the early 1800s. Figure 3.2 shows a tolerance of a small around the limit
value L. The goal is to show that the function f .x/ stays within that tolerance when
x is close to a.
In the figure you can see that for the values of x near a, the function f .x/ falls
within the prescribed tolerance of L. You could find a small interval centered at a
Springer International Publishing Switzerland 2016
J.M. Kane, Writing Proofs in Analysis, DOI 10.1007/978-3-319-30967-5_3
47
48
3 Limits
y = f(x)
y = f(x)
L+
L
L
y = f(x)
a+
such that jf .x/ Lj < for all x in that interval. That is, there is a value > 0
such that every x satisfying jx aj < also satisfies jf .x/ Lj < as seen in
Fig. 3.3. Again, the choice of the Greek letter (delta) is completely arbitrary, but
the tradition of using in this context is universal.
Note that for the function whose graph appears in Fig. 3.3 the value L is the limit
of the function as x approaches a, and L also happens to be the value of f .x/ at x D a.
You should recall that this sometimes happens (specifically when f is continuous at
x D a), but that this is not a requirement. Indeed, one reason for discussing limits
in the first place is because there is a need to evaluate the behavior of a function
as x approaches a value a when the function fails to be defined at x D a. Thus, in
general, one does not want to require jf .x/ Lj < for all x with jx aj less than
some positive value since this would require jf .x/ Lj < at x D a. Instead,
one excludes the need for the function to satisfy any conditions at all at x D a by
saying that there is a positive such that f is within the desired tolerance of L for all
x with 0 < jx aj < . Clearly, the value of must be chosen to be positive since
no negative value would represent a distance and, D 0 would not result in a region
around the number a satisfying jx aj < .
49
x!a
Combining these ideas results in the following definition. Suppose that the
function f is defined for all x in an open interval containing a 2 R except perhaps
at x D a. Then the limit of f as x approaches a is L, lim f .x/ D L, means that for
x!a
every > 0 there exists a > 0 such that for every x satisfying 0 < jx aj < , it
follows that jf .x/ Lj < . The power of this definition is the fact that the and
are arbitrary positive numbers. For example, what if you knew that for every > 0
there were a > 0 such that whenever 0 < jx aj < , then jf .x/ Lj < 2?
Would this be sufficient for showing lim f .x/ D L? The answer is yes, because the
x!a
is arbitrary. Suppose that for any > 0 you can find a > 0 that will ensure that
jf .x/ Lj < 2. Then since 2 is also a positive number, you can find a 0 > 0, likely
smaller than , that will ensure that jf .x/ Lj < 2 2 D . The point here is that
since was arbitrary, you can replace it with any positive number, including 2 .
3.1.1 Exercises
Which of the following definitions is equivalent to the definition of lim f .x/ D L?
x!a
1. For all 0 there is a 0 such that if 0 < jx aj < , then jf .x/ Lj < .
2. For all > 0 there is a > 0 such that if 0 < jx aj < 4 , then jf .x/ Lj < 7.
3. For all > 0:001 there is a > 0:001 such that if 0 < jx aj < , then
jf .x/ Lj < .
4. For all > 0 there is an > 0 such that if 0 < jx aj < , then jf .x/ Lj < .
5. There exists a > 0 such that for all > 0, if 0 < jxaj < , then jf .x/Lj < .
6. For all > 0 there is an > 0, such that if 0 < jx aj < , then jf .x/ Lj < .
7. For all > 1 there is a > 1, such that if 1 < jx aj C 1 < , then
jf .x/ Lj C 1 < .
The definition of limit provides a formula by which one can construct a proof that
a particular function f has a limit of L at the point a. The definition requires that
for every > 0 there is a > 0 that satisfies certain properties. Thus, a proof of a
limit must show that for every > 0 you can exhibit a > 0 which has the needed
property. As with other proofs that some property holds for all elements of a set, the
proof begins by selecting an arbitrary element of that set. In this case, one would
select an arbitrary > 0. The goal is to present a value for > 0 such that every x
satisfying 0 < jx aj < also satisfies jf .x/ Lj < . That suggests the following
proof template.
50
3 Limits
SET THE CONTEXT: Make statements telling what is known about the
function f and the numbers a and L.
SELECT AN ARBITRARY : Given > 0,
PROPOSE A VALUE FOR : let D
. Here you would insert an
appropriate value for .
SELECT AN ARBITRARY x: Select x such that 0 < jx aj < .
LIST IMPLICATIONS: Derive the result jf .x/ Lj < .
STATE THE CONCLUSION: Therefore, lim f .x/ D L.
x!a
the proof would begin with Let f .x/ D 2x 3. Given > 0, : : :. Your task is to
determine a value of > 0 that ensures the inequality jf .x/ Lj < holds for all
x with 0 < jx aj < . Since the function f is not constant, the choice of will
surely depend on the value of . But how is this value of determined? A common
approach is to work backwards from the final conclusion jf .x/ Lj < to see what
value of is needed.
In this example, f .x/ D 2x 3, a D 5, and L D 7. The value of jf .x/ Lj D
j.2x 3/ 7j D j2x 10j D 2jx 5j. Note that this expression has a factor of
x a, where a D 5. When finding the limit of a polynomial where L D f .a/, this
will always be the case. For more complicated functions f , other properties of f will
often allow you to write f .x/ L as an expression where x a is a factor. This makes
it easier to determine a value of since the choice of restricts the size of jx aj
which, in turn, will make jf .x/ Lj small, the desired result.
So here, jf .x/ Lj D 2jx 5j. To force jf .x/ Lj to be less than some arbitrary
> 0, it is, therefore, sufficient for 2jx 5j to be made less than . This is done by
making jx 5j < 2 , and the needed value of is 2 . Note that it is stipulated that is
positive, so D 2 is also greater than zero, a requirement of the definition of limit.
Now a complete proof can be written by following the template.
PROOF: lim 2x 3 D 7
x!5
Let f .x/ D 2x 3.
Given > 0, let D 2 > 0.
Select x such that 0 < jx 5j < D 2 .
Then > jx5j implies > 2jx5j D j2x10j D j.2x3/7j D jf .x/7j.
Therefore, lim 2x 3 D 7.
x!5
x!4
51
x!a
,
3jxC4j
x!4
52
3 Limits
The only problem with the above proof is in its use of the variable . In the second
line of the proof is set to 1, and in the fourth line it is set to 27
. It does not make
sense to set the value of equal to both of these values because, except in the rare
case that D 27, the value of cannot be equal to both values at the same time.
The solution is to choose one value for that satisfies two separate conditions. For
example, you can first require that < 1. Then a choice of x with 0 < jx 4j <
will guarantee that jx C 4j < 9. Then 3jxC4j
> 39
D 27
. This suggests that you
should select D 27 . But you also need 1. What happens if someone suggests
that be some rather large number such as D 100? Then D 27
would not satisfy
< 1. This is not a problem since one can always get away with selecting a positive
value for that is smaller than needed. Thus, you can select
to be the lesser of 1
and 27
. This choice is usually written as D min 27
; 1 . Now you can put this all
together to get a formal proof that is completely correct.
PROOF: lim 3x2 D 48
x!4
x!4
xC2
2
x!2 x C3xC2
D 1. In this
example f .x/ D
a D 2, and L D 1. Note that f .2/ is not defined even
though the limit as x approaches 2 exists. The proof
of this limit must conclude
xC2
with the inequality > jf .x/Lj D x2 C3xC2 .1/. As in the previous examples,
xC2
,
x2 C3xC2
xC2
it would be convenient if the expression x2 C3xC2
.1/ would contain a factor of
x C 2 so that it could be made small by requiring x .2/ to be less than some .
53
x!a
But this follows with some fairly straightforward algebra. Assuming that x 2,
x2
xC2
xC2
1
1 C .x C 1/
xC2
.1/ D
C1 D
C1 D
D
:
C 3x C 2
.x C 2/.x C 1/
xC1
xC1
xC1
PROOF: lim
D 1
xC2
Let f .x/ D x2 C3xC2
.
Given > 0, let D min 2 ; 12 .
Select x such that 0 < jx .2/j < .
Since 12 , it follows that jx C 2j < 12 and 12 < x C 2 < 12 , so 32 <
x C 1 < 12 . Thus, jx C 1j > 12 .
Since 2 , it follows that jx C 2j < 2 < jx C 1j.
Then > jx .2/j > 0 implies 2 > jx C 2j and
1
1
jx C 2j D jxC1j
j1 C .x C 1/j D
> 2jx C 2j > jxC1j
1
xC2
C 1 D 2
.1/ D jf .x/ .1/j.
xC1
x C3xC2
xC2
2
x!2 x C3xC2
Therefore, lim
D 1.
Clearly, at the point that you stipulate that should be less than 12 , you are making
a rather arbitrary decision. What would have happened if you had chosen some
other reasonable bound on the size of ? For example, what if instead you only
require < 34 ? This would also work, although that decision would affect the final
choice of for now jx C 1j can get as small as 14 , and jxC2j
could be as large as
3jxC1j
4jx C 2j. This suggests that you then select D min 4 ; 4 . This choice is no better
or worse than the chosen earlier. When one makes such arbitrary decisions, it
is good form to make a selection that does not lead to unnecessary arithmetic or
algebraic complications because one does not want to make the proof any harder to
read than necessary. Thus, it p
would perfectly adequate but enormously awkward to
select the bound on to be p5 . As long as the bound is less than 1, it will do the
1C 5
p
5
p
1C 5
54
3 Limits
3.2.1 Exercises
Write a proof of each of the following limits.
1. lim 35 x C 1 D 4
x!5
2. lim 5x 8 D 7
x!3
3. lim 2x2 D 18
x!3
4. lim 9x2 D 4
x! 23
5. lim 3x2 5x 7 D 1
x!1
6. lim x2 C 3x C 1 D 29
x!4
7. lim 2x3 D 16
x!2
6
D2
x!1 2xC5
xC4
lim 2
D
x!8 x 10xC10
8. lim
9.
2
10. lim mx C b D ma C b
x!a
x!a
sided limits except that the value of x is only allowed to approach the real number a
from one side. As a result, the definitions of these one-sided limits are very similar
to the definition of limit with minor alterations that forces x to stay on one side of
a. The definition of limit states that for a function f defined in a neighborhood of a,
but not necessarily at a, the limit lim f .x/ D L means for every > 0 there exists a
x!a
> 0 such that for every x satisfying 0 < jx aj < , it follows that jf .x/ Lj < .
What is it about this definition that allows x to approach a from two sides? It is the
inequality 0 < jx aj < that allows x to be either greater than or less than a since
jx aj is positive in either case. By removing the absolute value function in this
inequality and writing instead 0 < x a < , the choice of x becomes restricted
to being a value greater than a, or writing instead 0 < a x < , the choice of x
becomes restricted to being a value less than a. Thus, if f is a function defined for
all x in an open interval with right end at a, then the limit of f at a from the left is
L, lim f .x/ D L, means that for every > 0 there is a > 0 such that for every x
x!a
55
defined for all x in an open interval with left end at a, then the limit of f at a from
the right is L, lim f .x/ D L, means that for every > 0 there is a > 0 such that
x!aC
a D 3, and L D 3. As with a proof of other limits earlier in the chapter, the proof
needs to give a value for > 0 which will ensure > jf .x/Lj D j.2x2 5x/3j D
j.2x C 1/.x 3/j. This will follow if jx 3j < j2xC1j
for all suitable values of
x. What is needed is the largest possible value of 2x C 1, but 2x C 1 is not bounded
unless x is restricted to be close to 3. Thus, stipulate that be less than 1 which will
ensure that x3 will be less than1, x will not exceed 4, and 2xC1 will not exceed 9.
Then can be chosen to be min 9 ; 1 , and the proof can be written as follows.
PROOF: lim 2x2 5x D 3
x!3C
x!3C
Consider a function where its left limit differs from its right limit such as the
function
5 7x if x < 1
f .x/ D
: Then lim f .x/ D 2 while lim f .x/ D 1. Thus,
x!1
x if x 1
x!1C
while proving lim f .x/ D 2, it is important to use that fact that x < 1 as part
x!1
of the proof since the required inequalities will not hold for x > 1 (Fig. 3.4). The
following shows one possible proof.
56
3 Limits
PROOF: lim
x!1
5 7x if x < 1
x if x 1
D 2
5 7x if x < 1
Let f .x/ D
.
x if x 1
Given > 0, let D 7 .
Select x such that 0 < 1 x < D 7 . Then x < 1, so f .x/ D 5 7x.
It follows that > 7.1 x/ D 5 7x .2/ D jf .x/ .2/j.
Therefore, lim f .x/ D 2.
x!1
In the third line of the proof, 0 < 1 x < ensures that x < 1 which, in turn, is
needed to conclude that f .x/ D 5 7x and not f .x/ D x. The fact that x < 1 is also
used in the fourth line of the proof to conclude that 5 7x .2/ D jf .x/ .2/j
which follows because 5 7x .2/ is positive for all x < 1.
3.3.1 Exercises
Write a proof of each of the following one-sided limits.
1. lim x2 C 4x D 21
x!3C
2. lim 8 3x D 1
x!3
3. lim
x!2
4. lim
x!4C
5. lim
x!2
6. lim
x!2C
x2 4
x2 3xC2
D4
x2 4x
2x2 7x4
8jx2j
x2 4
8jx2j
x2 4
D 2
D2
4
9
57
arbitrarily
x
Again, you can work backwards. Since jf .x/ Lj D x2 C6 , as long as x > 0, it
x
would follow that x2 C6
< xx2 D 1x . Thus, there is an expression, 1x , which is larger
than jf .x/ Lj for all suitably large values of x. This will help because if you can
assure that 1x is less than , it will follow that jf .x/ Lj is also less than . It would
not have been helpful to exhibit an expression that was always less than jf .x/ Lj
because making that expression small would not imply that jf .x/ Lj is small. Now,
if x > 1 , it follows that 1x < suggesting that 1 is a suitable value for N.
x
x!1 x2 C6
PROOF: lim
D0
x
Let f .x/ D x2 C6
.
Given > 0, let N D 1 .
Select x such that x > N > 0.
Then x > 1 implies > 1x D xx2 >
x
Therefore, lim x2 C6
D 0.
x!1
x
x2 C6
D x2 C6
0 D jf .x/ 0j.
58
3 Limits
Note that it is important that the third step of the proof pointed out that N is positive.
It is used in the fourth step when 1x is calculated, and this would not have been
allowed if the value of x could have been zero.
For a second example, consider proving lim 2xC5
D 2. Again, you can work
x!1 x7
.2xC5/2.x7/ 19
2xC5
backwards to get > jf .x/ Lj D x7 2 D
D x7 . From here
x7
there are a number of ways to proceed. You can solve for x in the previous inequality
to get x > 7 C 19
which gives a reasonable value for N. Another way would be to
19
is less than
say that if x > 14, then x 7 < x 2x D 2x . In this case the fraction x7
19
38
38
,
and
it
becomes
clear
that
x
>
is
sufficient.
x D
x 2
x
This is an example demonstrating the enormous flexibility one sometimes has in
writing proofs in analysis where you often need to prove an inequality which can
be done in many ways. It is usually easier to prove an inequality involving a simple
fraction rather than a complicated fraction, so you can use the strategy of replacing
a fraction with a simpler fraction that is clearly larger, or in some cases, clearly
smaller. Keep in mind that a ratio of positive values gets larger if its numerator gets
larger or its denominator gets smaller.
A complete proof can be written as follows.
2xC5
x!1 x7
PROOF: lim
D2
19
and >
19
x7
2xC5
x7
2 D jf .x/ 2j.
x!1
As in the previous proof it is important that x > 7 is pointed out in the third step of
the proof because that fact is needed both to ensure that f .x/ is defined by assuring
x 7 0 and that x 7 is positive allowing the absolute value function to be
introduced in the fifth step of the proof.
With a slight adjustment of the definition of lim f .x/ D L, one gets a definition
x!1
of lim f .x/ D L. This time rather than choosing an N and requiring jf .x/ Lj <
x!1
for all x > N, one instead needs f .x/ to be within of L for those x < N. Thus,
lim f .x/ D L means that for every > 0 there exists an N 2 R such that for all
x!1
x < N it follows that jf .x/ Lj < .
2
D 3, one can identify an N such that x < N implies that
To prove lim 6x2x2C5x
7
x!1
2
2
2
6x C5x
.6x C5x/3.2x2 7/
D
3j
<
by
working
backwards.
That
is,
3
j 6x2x2C5x
7
2x2 7
2x2 7
5xC21
59
and it would be nice to remove it. Simply removing this negative term would make
the absolute value of the fraction smaller when what is needed is to make the fraction
larger. A strategy that does work is to take part of the 2x2 term, which grows very
large as x goes to 1, and pair it with the 7 term. For example, 2x2 7 p
can
be written as x2 C .x2 7/. Because x2 7 is a positive value for all x < 7,
removing it from the denominator makes the absolute value of the fraction greater.
Also note that when x < 21
, the numerator j5x C 21j < 5jxj, and this happens for
10
5xC21
p
5
5
PROOF: lim
D3
6x2 C5x
.
2x2 7
Let f .x/ D
p
Given > 0, let N D min 7; 5 .
p
Select x such that x < N 7.
5xC21
> 2 2 D
Then x < N 5 implies > 5x D 5x
x2
x C.x 7/
2
D
D
jf
.x/
3j.
3
2x2 7
2x2 7
6x2 C5x
2
x!1 2x 7
Therefore, lim
D 3.
3.4.1 Exercises
Find ways to justify each of the following inequalities that hold for large values
of x.
1.
2.
3.
3x5
< 2x
2x2
4xC7
< 5x
2x2 6
2
5x C3xC1
< 10
x
x3 x2 1
4. lim
5.
9x2
D3
2
x!1 3x 10
3
x
lim
D 15
3
2
x!1 5x 2x 4
6. lim
7.
60
3 Limits
61
1; 2; 3; : : :
1; 1; 2; 3; 3; 4; 5; 5; : : :
12 ; 23 ; 34 ; 45 ; : : :
13 ; 23 ; 33 ; 43 ; : : :
whereas the following sequences are monotone decreasing:
3.5.4 Subsequences
Intuitively, a subsequence of a sequence <an > is a sequence whose terms include
some of the terms of the sequence <an > in the same order as they appear in the
original sequence. Formally, if there is a strictly increasing sequence of natural
numbers i W N ! N, then <ain > is a subsequence of the sequence <an >. Thus, the
sequence 1; 1; 2; 2; 3; 3; : : : has the following subsequences
1; 2; 3; : : :
1; 1; 3; 3; 5; 5; : : :
2; 3; 5; 7; 11; : : :
The sequence 1; 2; 2; 3; 3; 3; 4; 4; 4; 4; : : : is not a subsequence of 1; 1; 2; 2;
3; 3; : : : since there are no repeated values in the original sequence, so there can
be no repeated values in any of its subsequences.
62
3 Limits
one can
with
proofs
nC16
nC16n
4n2 CnC2
n!1
4n2 CnC2
2n2 7
D2
Let an D 4n2nCnC2
2 7 .
Given > 0, let N D max 3; 17
.
Select an n > N.
Since N 3, it follows that n2 > 9.
Also, n > N gives n 17
. Thus,
2
4n CnC2
Therefore, lim
17
n
nC16n nC16
> 2 2 > 2 D
D 17n
n2
n C.n 7/
2n 7
D 2.
63
the inequality > jan Lj in order to find an appropriate value of N that allows you
to use the definition of limit to complete the proof. But in this case, you do not have
a general expression for the terms an , and you have not been given a value for L.
Somehow you need to use the only known facts about <an >, that is, the fact that the
sequence is both monotone increasing and bounded, to come up with a candidate to
serve as the limit, L, in the proof.
The definition of a sequence being bounded above holds the key. That definition
says that the sequence <an > is bounded above if the set fan j n 2 Ng is bounded
above, so there is a real number M which is greater than or equal to each term of
the sequence. Will this M be the limit of the sequence? Well, not usually. If M is
an upper bound for the sequence, then so are M C 1, M C 100, and M C 20;000.
They are all upper bounds, but they cannot all be limits of the sequence. You should
recognize that the terms of the sequence must get close to the limit, and the only
upper bound of the set fan j n 2 Ng that the terms could get close to is the least
upper bound of the set. Since fan j n 2 Ng is both nonempty and bounded above,
the Completeness Axiom for the real numbers guarantees that such a least upper
bound exists. This gives you a candidate for L.
The proof will require you to show that for all n greater than some N, the terms
of the sequence, <an >, are within of L. How can this be arranged? Here is where
you can use the fact that the sequence is monotone increasing because once you find
a single term, an , that gets within of L, all the terms that come after this term in the
sequence will necessarily have to be between an and L, so they also will be within
of L. How do you find one term, an , within of L? This follows from the fact that
L is a least upper bound of fan j n 2 Ng. Because L is the least upper bound, L
being less than the least upper bound, L, is not an upper bound, so there must be an
element of the set fan j n 2 Ng greater than L . This gives all the tools needed for
the proof (Fig. 3.6).
a1
a2
a3
a4
a5 aN an
64
3 Limits
So how would you write the proof? Certainly the proof would begin with
selecting a generic sequence and making a statement about the properties the
sequence is assumed to have, that is, its being monotone increasing and bounded
above. Then, the proof would proceed to justify the existence of the least upper
bound for the set of terms of the sequence; that will give you the target value of L.
Then, as with most proofs about limits, it would select a value for > 0. Unlike
the limit proofs earlier in this chapter, one cannot immediately state a value for N.
The existence of N must be proved as discussed in the previous paragraph. Finally,
the properties of the sequence can be brought together to show jan Lj < for all
n > N. Here is one possible proof.
PROOF: A monotone increasing sequence that is bound above converges.
Let <aj > be a monotone increasing sequence of real numbers that is
bounded above.
Since the set of terms A D faj j j 2 Ng contains a1 , it is nonempty, and since
it is bounded above, the Completeness Axiom guarantees that A has a least
upper bound, L.
Given > 0, the number L is less than L. Since L is the least upper
bound of A, L is not an upper bound of A. Thus, there is an N 2 N such
that the term aN is in A and is larger than L .
Select an n > N.
Because <aj > is monotone increasing, an aN . Because L is an upper
bound for A, an L. Therefore, L < aN an L, and jan Lj <
j.L / Lj D .
This proves that the sequence <aj > has limit L and that <aj > converges.
Note that the proof needs to refer to the sequence <an > as well as a particular
element of the sequence an . It could be confusing to the proof reader to use the
variable n in both contexts here, especially since the sequence notation <an > is
used after the choice of a specific value of n is made. That is the reason the proof
changed to using the variable j to refer to a generic term index. Then, it could refer
to a specific term using index n without confusing the two uses.
There is also a theorem stating that a monotone decreasing sequence that is
bounded below converges. The proof of this is left as an exercise.
As an illustration of the usefulness of the above result, consider
a sequence
p
defined recursively by a1 D 2, and for n p
1, anC1 D
an C 12. That is,
p
p
p
a1 D 2, a2 D a1 C 12 D 14, a3 D
14 C 12, and so forth. One can
prove that this sequence converges by showing that the sequence is both monotone
increasing and bounded above. Indeed, both of these facts can be established by
mathematical induction. The reader is likely already familiar with proofs by
mathematical induction, but this is an appropriate opportunity to review the method
and its merits.
65
Suppose the variable n represents any natural number, and there is a statement
S.n/ that includes this variable as part of the statement. For example, the statement
could be lim xn D an . Mathematical induction is a proof technique that uses the
x!a
following proof template to show that S.n/ is true for all n greater than or equal to
some base value b 2 N.
TEMPLATE for using mathematical induction to prove the statement
S.n/ is true for all natural numbers n b.
SET THE CONTEXT: The statement will be proved by mathematical
induction on n for all n b.
PROVE S.b/: Prove that the statement is true when the variable n is equal
to the base value, b.
STATE THE INDUCTION HYPOTHESIS: Assume that S.n/ is true for
some natural number n D k b.
PERFORM THE INDUCTION STEP: Using the fact that S.k/ is true, prove
that S.k C 1/ is true.
STATE THE CONCLUSION: Therefore, by mathematical induction, S.n/
is true for all natural numbers n b.
It is important to understand that the technique of mathematical induction works.
That is, if the statement S.b/ is true, and if the statement S.k/ ! S.k C 1/ is true,
then, in fact, S.n/ must be true for all natural numbers n b. Certainly, S.b/ is
true. Because S.b/ is true, and S.k/ ! S.k C 1/ is true for all k b, it follows that
S.b/ ! S.bC1/, so S.bC1/ is true. Then S.bC1/ ! S.bC2/, S.bC2/ ! S.bC3/,
and so forth, so the fact that S.n/ is true for all n b follows.
The strength of mathematical induction is that it is often much easier to provide
a proof for the one step S.k/ ! S.k C 1/ than it is to prove S.n/ in the general
case. The reader has likely seen many statements proved by mathematical induction
while studying Algebra, Calculus, or just about any other branch of mathematics.
Mathematical induction is an excellent tool for proving that the previously
introduced recursive
sequence
is both monotone increasing and bounded above.
p
p
Clearly, a2 D 14 > 4 D 2 D a1 so a1 < a2 . Supposepthat for some
p k 1 one
has ak < akC1 . Then it follows that ak C12 < akC1 C12 so ak C 12 < akC1 C 12
which shows that akC1 < akC2 . Thus, by mathematical induction it follows that
an < anC1 for all n, and the sequence is monotone increasing. Alsop
clear is that
ap1 D 2 < 4. p
Suppose that for some k 1 that ak < 4. Then akC1 D ak C 12 <
4 C 12 D 16 D 4. Thus, by mathematical induction it follows that an < 4 for
all n, and the sequence is bounded above. The limit of this sequence
can be shown
p
to be 4. In particular,
if
the
limit
is
L,
one
can
conclude
that
a
C
12 should be
n
p
which should equal the limit of an which is also L. Thus, one
converging to L C 12p
would expect that L D L C 12. This equation has only one positive real solution,
L D 4.
66
3 Limits
there is an N such that if natural number n > N, then jan Lj < . Well then,
certainly if m and n are both natural numbers greater than N, then both jam Lj <
and jan Lj < . Adding these two inequalities together shows that jam Lj C
jan Lj < C . The triangle inequality states that for any real numbers x and y,
jxj C jyj jx C yj. Thus, 2 > jam Lj C jan Lj D jam Lj C jL an j
j.am L/ C .L an /j D jam an j. Of course, the definition of Cauchy sequence
requires you to show that jam an j is less than , not 2. But you have an enormous
amount of flexibility when working with these types of inequalities, so you could
have asked instead for an N such that for all natural numbers n greater than N,
you have jan Lj less than 2 rather than less than . Thus, the proof could be as
follows.
PROOF: Every convergent sequence is Cauchy.
Let <aj > be a sequence of real numbers with lim aj D L.
j!1
67
is a finite set, so it is bounded by some number, K. That is, jan j K for all n N.
If m > N, then, since both N and m are greater than or equal to N, it follows that
jam aN j < 1 from which it follows that jam j < jaN j C 1 K C 1. Then the
sequence <an > is necessarily bounded above by K C 1 and below by .K C 1/, and
the sequence is bounded. A complete proof follows.
PROOF: All Cauchy sequences are bounded.
Let <an > be a Cauchy sequence.
Then there is a natural number N such that for all m; n N, jam an j < 1.
The set fa1 ; a2 ; a3 ; : : : ; aN g is a finite set, so there is a K such that the set is
bounded above by K and bounded below by K.
Let m be any natural number. If m N, then jam j K. If m > N, then
jam aN j < 1, so jam j D jam aN C aN j jam aN j C jaN j < 1 C K.
It follows that all terms of the sequence lie between .K C 1/ and K C 1,
and, thus, the sequence is bounded.
One consequence of the last two results is that since all convergent sequences are
Cauchy, all convergent sequences are bounded. The concept of a Cauchy sequence is
not only applied to sequences of numbers but also to much more general sequences
such as sequences of vectors, sequences of functions, and sequences of linear
operators. Of course, one would need a way to discuss distances between the terms
of a sequence in these other contexts, but when that makes sense, the concept of a
Cauchy sequence becomes important.
3.5.8 Exercises
1. Which of the following sequences are monotone? Which of them are bounded
above? Which of them are bounded below? Which of them are bounded?
(a)
(b)
(c)
(d)
(e)
(f)
(g)
an
an
an
an
an
an
an
D .1/n
n
D nC1
D 5n
.1/n
D 5n
n
D 1C.1/
nCn1
D 5 n.1/n
D 1 12 13 1n
(a) lim
(b)
(c) lim
D2
D4
n2 C2nC1
2
n!1 n 2n5
D1
68
3 Limits
p
3. If a1 D 3 and an is defined recursively by anC1 D 3an C 10, show that the
sequence <an > converges.
p
4. If a1 D 7 and an is defined recursively by anC1 D 3an C 4, show that the
sequence <an > converges.
5. Prove that a monotone decreasing sequence that is bounded below converges.
6. Let <an > be any sequence. Prove that <an > has a monotone subsequence.
7. Prove that if <an > is a sequence such that L D lim a2n D lim a2nC1 , then the
n!1
n!1
sequence converges to L.
8. Prove that if <an > is a sequence that converges to L, then the sequence
a1 ; a1 ; a2 ; a2 ; a3 ; a3 ; : : : also converges to L.
9. Prove that if <an > is a sequence that converges to L, then the sequence
a1 ; a2 ; a2 ; a3 ; a3 ; a3 ; a4 ; a4 ; a4 ; a4 ; : : : also converges to L.
x!a
to L. So what does it mean for lim f .x/ not to exist? Intuitively, it could mean that
x!a
in every neighborhood of a there are values of x for which f .x/ is close to one value
L1 and other values of x for which
value L2 . That is what
f .x/ is close to another
4x 5 if x < 2
happens with the function f .x/ D
as x approaches 2. For some
10 2x if x 2
values of x near 2, f .x/ is close to 3, and for some values of x near 2, f .x/ is close
to 6. Thus, the limit does not exist. Another well-known example is f .x/ D sin 1x
which oscillates wildly as x approaches zero, and in every neighborhood of 0, the
function takes on all values in the interval 1; 1 infinitely often. Another way for
the limit not to exist is for the values of f .x/ to grow without bound and approach
xC3
infinity or negative infinity such as what happens to f .x/ D .x5/
2 as x approaches 5.
One can write a proof showing that a particular function has no limit at x D a,
but before discussing how to do this, it is worth taking a close look at the definition
of limit.
69
But this is an open statement, that is, even though the function f and the limit point
a are supposedly known, the statement contains variables x, L, , and , all of which
are unknown. Thus, this open statement does not have any truth value until these
four variables have been stipulated. They are stipulated with four phrases: there is
a real number L, for all > 0, there is a > 0, and for every x. These four
phrases are called quantifications of the variables because they indicate for which
values of the variables the following statement must hold. Two of the phrases use
the existential quantifier there exists. It indicates that there is at least one value
of the variable that will make the following statement true. The other two phrases
use the universal quantifier for all. It indicates that every possible value of that
variable will make the following statement true. So
The statement there exists a real number L such that for all > 0 there is a > 0
such that for every x, 0 < jx aj < implies jf .x/ Lj < begins with the
existential quantifier there exists a real number L, and the entire statement is
true if, in fact, there is a value of the variable L that makes the following statement
true, that is, for all > 0 there is a > 0 such that for every x, 0 < jx aj <
implies jf .x/ Lj < .
The statement for all > 0 there is a > 0 such that for every x, 0 < jx aj <
implies jf .x/ Lj < begins with the universal quantifier for all > 0, and
the entire statement is true if, in fact, every possible positive value of the variable
makes the following statement true, that is, there is a > 0 such that for every
x, 0 < jx aj < implies jf .x/ Lj < .
The statement there is a > 0 such that for every x, 0 < jx aj < implies
jf .x/ Lj < begins with the existential quantifier there is a > 0, and the
entire statement is true if, in fact, there is a positive value of the variable that
makes the following statement true, that is, for every x, 0 < jx aj < implies
jf .x/ Lj < .
The statement for every x, 0 < jx aj < implies jf .x/ Lj < begins with
the universal quantifier for every x, and the entire statement is true if, in fact,
every possible value of the variable x makes the following statement true, that is,
0 < jx aj < implies jf .x/ Lj < .
A proof that no limit exists must prove the negation of the statement that says that
a limit does exist, so it is important that one can generate the negation of a statement
that contains quantifiers such as this one does. The logic of doing this is not hard
to follow. Suppose the P.y/ is a statement that depends on the value of a variable y.
Then the universally quantified statement for every y, P.y/ says that P.y/ is true
for every possible value of y. The negation of for every y, P.y/ must be that it
is false that every value of y makes P.y/ true, so there must be at least one y that
makes P.y/ a false statement. This means that the negation of for every y, P.y/
is the statement there is a y such that :P.y/. To negate a universally quantified
statement, change the universal quantifier to an existential quantifier and negate the
statement that follows.
What if the original statement is an existentially quantified statement such as
there is a y such that P.y/? This statement says that some value of y makes
70
3 Limits
P.y/ true. The negation of this statement must be that no value of y makes P.y/
true which is to say that every value of y makes P.y/ a false statement. This means
that the negation of there is a y such that P.y/ is the statement for all y, :P.y/.
To negate an existentially quantified statement, change the existential quantifier to a
universal quantifier and negate the statement that follows.
The statement that f has a limit at x D a is a statement that has an existential
quantifier followed by a universal quantifier followed by an existential quantifier
followed by a universal quantifier followed by a conditional statement. To prove
that f does not have a limit at x D a requires a proof of the negation of that
statement. From the previous discussion it is now clear that to get the negation of
the statement that f has a limit at a, you must flip the two existential quantifiers to
universal quantifiers, flip the two universal quantifiers to existential quantifiers, and
end with the negation of the conditional statement. The result is for all real numbers
L there is an > 0 such that for all > 0 there is an x such that 0 < jx aj < and
jf .x/ Lj .
SET THE CONTEXT: Make statements about what is known about the
function f and the number a.
SELECT AN ARBITRARY LIMIT L: Given L 2 R,
PROPOSE A VALUE FOR : let D . Here you would insert a value for
.
SELECT AN ARBITRARY > 0: Select > 0.
SELECT VALUES FOR x1 AND x2 : Let x1 D
and x2 D . Note that
0 < jx1 aj < , 0 < jx2 aj < , and jf .x1 /f .x2 /j 2. You would have
selected appropriate x1 and x2 in such a way that jf .x1 / f .x2 /j exceeds 2.
LIST IMPLICATIONS: Assume that jf .x1 / Lj < and jf .x2 / Lj < .
Then 2 D C > jf .x1 / Lj C jf .x2 / Lj D jf .x1 / Lj C jL f .x2 /j
jf .x1 / L C L f .x2 /j D jf .x1 / f .x2 /j.
STATE THE CONTRADICTION: This shows that 2 > jf .x1 / f .x2 /j
which is a contradiction.
STATE THE CONCLUSION: Thus, it cannot hold that both jf .x1 / Lj <
and jf .x2 / Lj < , and the limit does not exist.
71
4x 5 if x < 2
For example, consider the limit of f .x/ D
as x approaches
10 2x if x 2
2. Here the limit from the left is 3, and the limit from the right is 6. Thus, no matter
how close x is supposed to be to 2, there will be values x1 and x2 within that required
tolerance where f .x1 / is close to 3 and f .x2 / is close to 6. If f .x1 / and f .x2 / are both
supposed to be within of some limit L, then it will follow that f .x1 / and f .x2 / will
have to be within 2 of each other. Again, you employ the technique of showing that
two quantities close to the same value must be close to each other. In particular, if x1
is chosen to be less than 2, f .x1 / will be less than 3. If x2 is chosen to be between 2
and 2 12 , f .x2 / will be greater than 5. In this case it would be impossible to have f .x1 /
and f .x2 / within 2 of each other, and, therefore, it would be impossible to have them
both within D 1 of some limit L. This suggests that you will get a contradiction if
you set D 1. Indeed, if a > 0 is chosen, you can let x1 D 2 2 (that is, less than 2
but within of 2), and let x2 D min 2 C 2 ; 2 C 12 (that is, greater than 2 but within
of 2 and not so large that f .x/ is less than 5). The point of all of this is that now,
no matter what value is chosen for L, f .x1 / and f .x2 / are more than 2 apart, so how
could they both be within 1 of L? Specifically, if jf .x1 / Lj < 1 and jf .x2 / Lj < 1,
it follows from the triangle inequality that 2 D 1 C 1 > jf .x1 / Lj C jf .x2 / Lj D
jf .x1 / Lj C jL f .x2 /j jf .x1 / L C L f .x2 /j D jf .x1 / f .x2 /j showing
2 > jf .x1 / f .x2 /j which cannot hold. Here is the complete proof (Fig. 3.7).
Fig. 3.7 f has no limit at
xD2
72
3 Limits
PROOF: The function
4x 5 if x < 2
10 2x if x 2
has no limit as x ! 2.
4x 5 if x < 2
.
10 2x if x 2
Given any value for L, let D1, and let > 0 be given.
Let x1 D 2 2 and x2 D min 2 C 2 ; 2 C 14 .
Note that 0 < jx1 2j < and 0 < jx2 2j < .
Since x1 < 2, it follows that f .x1 / < 3. Since x2 > 2 and x2 < 2 14 , it follows
that f .x2 / > 5. As a consequence jf .x1 /f .x2 /j D f .x2 /f .x1 / > 53 D 2.
If jf .x1 / Lj < D 1 and jf .x2 / Lj < D 1, it would follow that
2 D 1 C 1 > jf .x1 / Lj C jf .x2 / Lj D jf .x1 / Lj C jL f .x2 /j
jf .x1 / L C L f .x2 /j D jf .x1 / f .x2 /j > 2. This shows that 2 > 2 which
is a contradiction.
Thus, it cannot hold that both jf .x1 / Lj < and jf .x2 / Lj < , and the
limit does not exist.
Let f .x/ D
It is even easier to show that the function f .x/ D sin 1x has no limit as x
approaches 0. This is because for every > 0 it is easy to find x1 and x2 between
0 and such that f .x1 / D 1 and f .x2 / D 1. This makes it impossible to find an
L where jf .x1 / Lj < 1 and jf .x2 / Lj < 1. Thus, the proof follows the given
template for proving that a limit does not exist (Fig. 3.8).
Fig. 3.8 Graph of sin
1
x
73
1
3
f .x1 / D sin .2k C 2 / D 1, and f .x2 / D sin .2k C 2 / D 1.
If jf .x1 / Lj < D 1 and jf .x2 / Lj < D 1, it would follow that
2 D 1 C 1 > jf .x1 / Lj C jf .x2 / Lj D jf .x1 / Lj C jL f .x2 /j
jf .x1 / L C L f .x2 /j D jf .x1 / f .x2 /j D 2. This shows that 2 > 2 which
is a contradiction.
Thus, it cannot hold that both jf .x1 / Lj < and jf .x2 / Lj < , and the
limit does not exist.
xC3
>L
.x5/2
1
, so by
jx5j
xC3
.x5/2
>
>
making jx 5j <
you will have the inequality that you
need. Note that the absolute value function was introduced in jLj C 1 to take care of
the embarrassing circumstance that L is negative, and in particular, when L D 1.
The proof is as follows.
PROOF: The function
xC3
.x5/2
has no limit as x ! 5.
xC3
Let f .x/ D .x5/
2.
Given any value for L, let D 1, and let >
0 be given.
1
.
Select a value of x between 5 and 5 C min 1; ; jLjC1
Note that 0 < jx 5j <
xC3
1
1
and f .x/ D .x5/
2 > .x5/2 > x5 > jLj C 1.
It follows that jf .x/ Lj > jLj C 1 L 1.
Thus, it cannot hold that jf .x/ Lj < , and the limit does not exist.
3.6.4 Exercises
Write the negation of each of the following statements.
1. There exists x such that x2 D A.
2. For all x there is a y such that g.x/ D f .y/.
74
3 Limits
3. There is an integer k such that f .x/ f .k/ for all x between k and k C 1.
4. For all x > 0 and all y > 0 there exists a z < 0 such that f .z/ xf .y/.
Prove that the following limits do not exist.
x
5. f .x/ D jxj
as x ! 0
1
as x ! 1
6. f .x/ D x sin x1
5x if x < 3
7. f .x/ D
as x ! 3
4x if x 3
8. f .x/ D x244 as x ! 2
75
76
3 Limits
77
78
3 Limits
79
3.7.1 Exercises
1. Write a definition for lim f .x/ where a is an accumulation point of the domain
x!aC
of f .
Identify the accumulation points, if any, of the following sets.
n
n 2 N
2. nC2
3. x 2 Q x2 < 2
1
3
; 2; 52 ; : : :
4. 2 ; 1;
m 2
5. 2n m; n 2 N
n
o
n C4
n 2 N
6. 2n.1/
3nC5
this limit. The reason that the limit does not exist is that the function grows without
bound and, therefore, does not approach any real number value. This behavior can
be quantified by saying that the limit of the function is infinity. Of course, it does not
make sense to say that the function is getting close to infinity, since no real number
is very close to infinity. In the definition of lim f .x/ where it had to be made clear
x!1
what x approaching infinity meant, it was said that there was a number N such that
jf .x/ Lj was small whenever x > N. Similarly, to say that f .x/ approaches infinity,
one needs to say that for any real number M, f .x/ can be made larger than M. Thus, if
a is an accumulation point of the domain of function f .x/, the following two similar
definitions can be given.
The limit of f at a is infinity or lim f .x/ D 1 means that for every M 2 R there
x!a
is a > 0 such that if x is in the domain of f with 0 < jx aj < , then f .x/ > M.
The limit of f at a is negative infinity or lim f .x/ D 1 means that for every
x!a
M 2 R there is a > 0 such that if x is in the domain of f with 0 < jx aj < ,
then f .x/ < M.
What if f .x/ approaches infinity or negative infinity as x is allowed to approach
either infinity or negative infinity? Each of these ideas can be accommodated
resulting in four similar definitions. Remember that the limit of f as x approaches
infinity makes sense only if the domain of f is unbounded above, and as x approaches
negative infinity only if the domain of f is unbounded below. Here are the four
definitions.
The limit of f as x approaches infinity is infinity or lim f .x/ D 1 means that
x!1
for every M 2 R there is an N 2 R such that if x is in the domain of f with x > N,
then f .x/ > M.
80
3 Limits
xC3
of 5 with x 5. Working backwards, you would start with .x5/
2 > M. This is a
complicated inequality with which to work, so it would be more convenient to work
xC3
with an inequality that is easier to solve. If you want f .x/ D .x5/
2 to be bigger than
M, it would be sufficient to make some fraction smaller than f .x/ bigger than M.
1
For example, for all x > 2 the fraction .x5/
2 is smaller than f .x/. Moreover, for
x within 1 of 5,
1
jx5j
1
jx5j
is smaller than
1
.
.x5/2
PROOF: lim
D1
xC3
Let f .x/ D .x5/
2.
Let M 2 R be given. Without loss of generality, assume that M > 1.
Let D M1 > 0. Note that < 1.
If 0 < jx 5j < , then since jx 5j < 1, it follows that jx 5j > .x 5/2 .
Also, since x > 4, it follows that x C 3 > 7 > 1.
xC3
1
1
1
Then f .x/ D .x5/
2 > .x5/2 > jx5j > D M.
xC3
2
x!5 .x5/
D 1.
81
3.8.1 Exercises
Write a proof of each of the following infinite limits.
x
2 D 1
x!4 .x4/
2
lim x 5x D 1
x!1
lim x2 D 1
x!0 jxj
1. lim
2.
3.
4.
lim 10
x!1
x
x!2C x2
5. lim
4 x D 1
D1
x!a
f .x/
x!a g.x/
4. if H 0, lim
L
.
H
Consider how to prove each part of the above theorem. In each case you will need
to prove the validity of a limit, so the proof can follow the usual proof template for
establishing a limit. These proofs differ from limit proofs found earlier in the chapter
in that you know less about the functions whose limits you are trying to establish.
On the other hand, you do know that the limits of the functions f and g exist, and
that gives you a lot of tools with which to work.
82
3 Limits
Similarly, the proof can produce a 2 > 0 coming from the definition of lim g.x/ D
x!a
H such that if x is in the domain of g and 0 < jx aj < 2 , then jg.x/ Hj will be
less than 2 . The proof then easily follows as described above.
83
A proof that the limit of the difference f .x/ g.x/ equals the difference of the
individual limits, L H, is very similar to the above proof and is left as an exercise.
Its goal is to establish the inequality jf .x/g.x/ LHj < . Again, you can use the
definition of limit to make jf .x/ Lj and jg.x/ Hj as small as you need, but how
small these have to be to ensure that jf .x/g.x/LHj is less than is not immediately
obvious. The problem is that it is difficult to gauge how close f .x/g.x/ is to LH when
you know that f .x/ is close to L, and g.x/ is close to H. The difficulty stems from
having to move from f .x/g.x/ to LH, where f .x/ changes to L and g.x/ changes to H
at the same time. If only one of these two changes were made, then it might be easier
to make the needed estimate. That is, it would be easier to work with an expression
like f .x/g.x/ f .x/H than with f .x/g.x/ LH.
Of course, f .x/g.x/ LH is not the same as f .x/g.x/ f .x/H, so one cannot
just use f .x/g.x/ f .x/H in place of f .x/g.x/ LH. Sometimes, though, it is
worth replacing one expression with another expression that is easier to handle,
84
3 Limits
and then adjusting the second expression to make it equivalent to the first. In this
case, the change can be accomplished by employing one of the oldest tricks used in
mathematical proofs, that of adding and subtracting the same quantity. In particular,
you can rewrite jf .x/g.x/ LHj as jf .x/g.x/ f .x/H C f .x/H LHj. The advantage
of doing this is that now you can see how the difference between f .x/g.x/ and LH
depends on the differences between f .x/ and L and g.x/ and H. Indeed, jf .x/g.x/
LHj D jf .x/g.x/ f .x/H C f .x/H LHj D jf .x/.g.x/ H/ C H.f .x/ L/j
jf .x/j jg.x/ Hj C jHj jf .x/ Lj. If each of the two terms, jf .x/j jg.x/ Hj
and jHj jf .x/ Lj, can be made smaller than 2 , then it will have been shown that
jf .x/g.x/ LHj is less than as needed.
So how small does jf .x/ Lj need to be to ensure that jHj jf .x/ Lj is less
than 2 ? Less than 2jHj
appears to be small enough, although one needs to handle
the embarrassing situation where H D 0. You could handle H D 0 and H 0 as
two separate cases, or you can take care of both cases at once by making jf .x/ Lj
less than since jHj C 1 is larger than jHj and can never be 0. Thus, you can
2 jHjC1
.
2 jHjC1
How small does jg.x/ Hj need to be to ensure that jf .x/j jg.x/ Hj is less than
It would be nice to say that jg.x/ Hj < 2jf.x/j suggesting that you set small
enough to ensure jg.x/ Hj < 2jf.x/j , but there is a problem here. The definition of
limit requires that the choice of come before the choice of x, so you cannot have
the value of depending on x. What is needed is an upper bound for jf .x/j because,
if jf .x/j M, the value of can be found to ensure jg.x/ Hj < 2M
which will
always be small enough to guarantee jf .x/j jg.x/ Hj < 2 . You can find such
an upper bound for jf .x/j because the limit of f .x/ exists as x approaches a, and so
jf .x/j can be restricted to being not much larger than jLj. You could, for example,
find 2 > 0 so that if 0 < jx aj < 2 , then jf .x/ Lj < 1. This would ensure that
f .x/ is a distance of no more than 1 from L so that jf .x/j < jLj C 1. Then you would
only need jg.x/ Hj < to get jf .x/j jg.x/ Hj < 2 . This gives you all the
?
2
2 jLjC1
85
2 jHjC1
f .x/
L
inequality which, in this case, is g.x/ H < . Using the trick of adding and
subtracting the same quantity, the left side of the inequality
can be written
as
.f
.x/L/HCL
Hg.x/
f .x/
f .x/HLg.x/
f .x/HLHCLHLg.x/
g.x/ HL D g.x/H D
D
g.x/H
g.x/H
L.g.x/H/
jf .x/Lj
jf .x/Lj
C g.x/H . Again, the goal will be to make each of the terms jg.x/j
jg.x/j
theorem will make that assumption. How far away from zero can you require jg.x/j
to be? Certainly, this will depend on the value of H. If H is close to zero, then jg.x/j
86
3 Limits
will be close to zero as x approaches a. The best you can do is require that jg.x/j be
so close to H that it will keep a known distance from zero. For example, you could
. That will ensure that jg.x/j is at least jHj
require that jg.x/ Hj be less than jHj
2
2
which keeps it a known distance away from zero. So, select a 1 > 0 such that if x
, and jg.x/j will
is in the domain of g with 0 < jx aj < 1 , then jg.x/ Hj < jHj
2
jHj
be greater than 2 .
.x/Lj
2
< jf .x/ Lj jHj
. Thus, it would
Now for these values of x you will have jf jg.x/j
be sufficient if jf .x/Lj were to be less than
jHj
4
jf .x/Lj
jg.x/j
< 2 .
jHj
This can be done by choosing
2 > 0 small enough so that jf .x/Lj is less than 4 .
2
H 2
H
jLj 4jLj
L.g.x/H/
< H42 D 2 . Well, OK, did you catch that the preceding does
g.x/H < jg.x/jjHj
2
not work if L D 0? To avoid this problem it would be better to make jg.x/ Hj less
2
than H . Putting all of these ideas together gives the following proof.
4 jLjC1
PROOF: Suppose that f and g are functions both defined on a set with
accumulation point a. If lim f .x/ D L and lim g.x/ D H with H 0, then
f .x/
x!a g.x/
lim
x!a
x!a
L
.
H
Let D min.1 ; 2 ; 3 /.
Then if x is in the domain of gf with 0 < jx aj < ,
f .x/HLg.x/ f .x/HLHCLHLg.x/ .f .x/L/HCL Hg.x/
f .x/
L
g.x/ H D g.x/H D
D
g.x/H
g.x/H
2
L.g.x/H/ jHj 2
jf .x/Lj
2jLj
H
2 < 2 C 2 D .
jg.x/j C g.x/H < 4 jHj C
H
4 jLjC1
lim f .x/
x!a g.x/
L
.
H
87
p.x/
x!a q.x/
p.a/
.
q.a/
The first two results are very easy to prove, and are left as exercises. The next
two results can be proved by using mathematical induction which is often the first
technique one considers using when trying to prove statements such as these that
depends on a natural number. Here, mathematical induction will be employed to
prove statements about the limits of polynomials, and the degree of the polynomial
provides a natural number to use as the induction variable.
To begin with, try using mathematical induction to prove that lim xn D an for
x!a
any natural number n. In this mathematical induction argument, the base case is
lim x D a, that is, when n D b D 1. The proofs of statements similar to this base
x!a
case were covered earlier. The induction step in the proof will need to show that if
lim xk D ak for some natural number k, then lim xkC1 D akC1 . But xkC1 is just the
x!a
x!a
product xk x, so this result follows immediately from the theorem about the limits
of products. That leads to the following proof that uses the template for proofs by
mathematical induction.
PROOF: lim xn D an for any natural number n.
x!a
SET THE CONTEXT: The statement will be proved for all natural numbers
n by mathematical induction on n.
PROVE S.b/: When n D 1, the statement says that lim x D a which has
x!a
already been established.
STATE THE INDUCTION HYPOTHESIS: Assume that for some natural
number k, lim xk D ak .
x!a
PERFORM THE INDUCTION STEP: Then since the limit of a product
of two functions is the product of the two individual limits, it follows that
lim xkC1 D lim xk x D .lim xk /.lim x/ D ak a D akC1 . So the statement
x!a
x!a
x!a
x!a
is true for n D k C 1.
STATE THE CONCLUSION: Therefore, by mathematical induction,
lim xn D an is true for all natural numbers n.
x!a
88
3 Limits
Mathematical induction can again be employed to prove that for every polynomial, p.x/, lim p.x/ D p.a/. As a reminder, a polynomial of degree n is a function,
x!a
lim x j D a j , from which one gets that the limit of a monomial is lim cj x j D cj a j .
x!a
x!a
A polynomial is just a sum of such monomials, so mathematical induction is a
convenient tool for showing that this sum of an arbitrary number of monomials
has the desired limit.
PROOF: For any constants c0 ; c1 ; c2 ; : : : ; cn and a 2 R, the polynomial p.x/ D cn xn C cn1 xn1 C cn2 xn2 C C c1 x C c0 satisfies
lim p.x/ D p.a/.
x!a
.lim ckC1 /.lim xkC1 / C lim ck xk C ck1 xk1 C C c1 x C c0 D
x!a
x!a
x!a
Recall that a rational function is just a ratio of polynomials, that is, if p.x/ and
q.x/ are polynomials, then p.x/
is a rational function. It is only a simple step to get
q.x/
the following theorem.
PROOF: For any polynomials p and q and a 2 R where q.a/ 0, it follows
p.x/
that lim q.x/
D p.a/
.
q.a/
x!a
lim p.x/
x!a
lim q.x/
x!a
p.a/
.
q.a/
89
3.9.6 Exercises
Write proofs of each of the following statements.
1. If f and g are defined in an open interval containing a, and if lim f .x/ D L and
x!a
x!a
4. If f and g have a common domain with lim f .x/ D L and lim g.x/ D H, then
x!a
x!a
x!a
5. If f and g have a common domain with lim f .x/ D L and lim g.x/ D H, then
x!aC
x!aC
x!aC
6. If f and g have a common domain with lim f .x/ D L and lim g.x/ D H, then
x!1
x!1
x!1
7. If <an > and <bn > are sequences with lim an D L and lim bn D H 0, then
an
n!1 bn
lim
n!1
L
.
H
x!0C
1
x
n!1
D L.
1
x!a f .x/L
lim f .x/CM D 1.
x!a f .x/CN
9. If f .x/ > L for all x a, then lim f .x/ D L if and only if lim
x!a
D 1.
90
3 Limits
Suppose that lim f .x/ D L and that for all x, f .x/ > 0.
x!a
Assume that L < 0.
By the definition of limit, there is a > 0 such that for all x in the domain
of f satisfying 0 < jx aj < , it follows that jf .x/ Lj < L.
For these values of x it must be that L > jf .x/ Lj D f .x/ L implying
that 0 > f .x/ which contradicts the fact that f .x/ is always positive.
Therefore, it must hold that L 0.
Similar statements can be made about the limits of functions f satisfying f .x/ > b
or f .x/ < b for all x where b is a constant real number. One can also extend this to
limits from the left, limits from the right, and limits to infinity and negative infinity.
Several of these possibilities have been left for the exercises.
x!a
M for some M L. But, in fact, limits are unique, that is, the only way for the limit
to be L and the limit to be M is for L and M to be equal. Intuitively, this should make
sense. If the values of f .x/ are getting close to L, then they should not also be able
to get close to a value distinct from L. So how can you prove this using nothing but
the definition of limit as a tool?
The result can be proved by contradiction, that is, if you assume that the function
f has two distinct limits, L and M, as x approaches a, then this leads to a statement
which must be false. Assuming that both limits exist, the definition of limit will
91
allow you to force both jf .x/ Lj < and jf .x/ Mj < for any positive
that you choose. Why cant this happen? Well, if it did, you could get C >
jf .x/ Lj C jf .x/ Mj D jf .x/ Lj C jM f .x/j jf .x/ L C M f .x/j D jM Lj.
If M L, then jM Lj is a positive number, so if is chosen less than or equal to
jMLj
, it will be impossible to have jM Lj < 2 as guaranteed by the definition of
2
limit. That gives you the following proof.
PROOF: If lim f .x/ D L and lim f .x/ D M, then L D M.
x!a
x!a
x!a
x!a
that for every > 0 there is a 1 > 0 such that when x satisfies 0 < jxaj < 1 , then
jg.x/ Lj < . Similarly, for that same , there is a 2 > 0 such that when x satisfies
0 < jx aj < 2 , then jh.x/ Lj < . Thus, you can show for values of x near a that
g.x/ f .x/ h.x/, < g.x/ L < , and < h.x/ L < . Putting these three
sets of inequalities together shows that < g.x/ L f .x/ L h.x/ L <
from which jf .x/ Lj < follows. This gives the following proof.
92
3 Limits
x!a
x!a
x!a
Finally, assume that g.x/ f .x/ h.x/ for all x in the common domain of
f , g, and h.
Let > 0 be given.
By the definition of limit, there is a 1 > 0 such that for all x in the domain
of g that satisfy 0 < jx aj < 1 , it follows that jg.x/ Lj < .
By the definition of limit, there is a 2 > 0 such that for all x in the domain
of h that satisfy 0 < jx aj < 2 , it follows that jh.x/ Lj < .
Then for all x in the common domain of f , g, and h satisfying 0 < jx aj <
min.1 ; 2 /, jg.x/ Lj < and jh.x/ Lj < .
Thus, for those x, < g.x/ L f .x/ L h.x/ L < from which it
follows that jf .x/ Lj < .
Therefore, lim f .x/ D L.
x!a
93
n!1
j!1
Let <an > be a sequence with lim an D L, and let <anj > be any
n!1
subsequence.
Let > 0 be given.
By the definition of limit, there is an N such that for all n > N, jan Lj < .
By the definition of subsequence, <nj > is a strictly increasing sequence of
natural numbers and, as such, satisfies nj j for all natural numbers j.
Thus, for all j > N, nj j > N implies janj Lj < .
This proves that lim anj D L.
j!1
Of course the converse of this theorem is trivially true. That is, if all subsequences
of a given sequence converge, then the original sequence converges. This is trivial
since the original sequence is one of its subsequences.
3.10.5 Exercises
Write proofs of each of the following statements.
1. If lim f .x/ D L and f .x/ < b for all x, then L b.
x!a
3. If f .x/ > 0 for all x, then lim f .x/ cannot equal negative infinity.
x!a
4. Suppose that sequences <an >, <bn >, and <cn > satisfy an bn cn for every
natural number n. If lim an D lim cn D L, then lim bn D L.
n!1
n!1
n!1
94
3 Limits
values do approach 2, but other values approach 2 and every value in between.
More precisely, for each L 2 2; 2, you can find sequences <xn > where lim xn D
n!1
sin x
. Finally, it
oscillate among values within some bounded range such as .3xC100/
xC10
might oscillate and be unbounded like x j sin xj.
Even when f oscillates so that it does not have a finite or infinite limit, it
is helpful to quantify which values the function f .x/ approaches repeatedly as x
grows. This can be done by considering the range of f .x/ when x is restricted to
an interval .M; 1/, and then watching what happens to that range as M gets large.
sin x
For example, consider the function f .x/ D .3xC100/
whose graph is shown in
xC10
3xC100
70
Fig. 3.10. The function xC10 D 3 C xC10 is a decreasing function of x for x > 0,
so on the interval .M; 1/, the function f oscillates in a range bounded between
3MC100
and 3MC100
. What can be said about the sequence <f .xn /> where <xn > is
MC10
MC10
and 3xxnn C100
,
a sequence with lim xn D 1? The values f .xn / are between 3xxnn C100
C10
C10
n!1
so as xn gets large, f .xn / is forced to be inside or very near the interval 3; 3.
Clearly, for no sequence <xn > can f .xn / approach a limit outside of the interval
3; 3, but there are sequences for which f .xn / approaches 3 and others for which
f .xn / approaches 3 as shown in the figure. Finding the greatest and least values
to which f .xn / could converge is the idea behind the limit superior and limit
inferior often referred to simply as the lim sup and lim inf, respectively. In the
sin x
example of f .x/ D .3xC100/
, the values of 3 and 3 came from looking at the
xC10
greatest lower bound and least upper bound of the set ff .x/ j x > Mg and then
letting M go to infinity. In general, let f be a function whose domain is unbounded
above. For each real number M let AM be the range of f for x > M, that is,
AM D ff .x/ j x is in the domain of f with x > Mg. Then define lim sup f .x/ to be
x!1
lim sup AM . Similarly, define lim inf f .x/ to be lim inf AM . Some books use the
M!1
x!1
M!1
notation lim and lim for lim sup and lim inf, respectively.
Fig. 3.10 Sequences
approaching the lim sup and
.3xC100/ sin x
lim inf of f .x/ D
xC10
95
x!1
so lim sup f .x/ will be 1. Analogously, if f .x/ is unbounded below as x gets
x!1
large, lim inf f .x/ D 1, and if lim f .x/ D 1, then lim inf f .x/ D 1. Note
x!1
x!1
x!1
that since sup AM and inf AM are both monotone function of M, their limits always
exist, although they might be infinite limits. Thus, unlike lim f .x/, the values of
x!1
x!1
If f .x/ remains bounded as x gets large, lim sup f .x/ and lim inf f .x/ are finite
x!1
x!1
values. This means lim sup AM is finite. For each natural number n, there must
M!1
be an xn > n such that f .xn / is within, say 1n of sup An . Then, <xn > is a sequence
that diverges to infinity such that for each n, sup An 1n < f .xn / < sup An . By
the Squeezing Theorem, lim f .xn / D lim sup An D lim sup f .x/. Similarly, there
n!1
n!1
x!1
must be a sequence <xn > diverging to infinity with lim f .xn / D lim inf f .x/. This
n!1
x!1
means that there is a sequence such that f converges to its limit superior on that
sequence and another sequence such that f converges to its limit inferior on it.
Consider the three examples: sin x, x j sin xj, and x sin x. None of these functions
has a limit as x approaches infinity because each function oscillates and does not
approach one particular value. On the other hand, in each case it is easy to see upper
and lower bounds to the oscillations. The values ofthe function
sin x clearly stay in
the interval 1; 1. For each integer n, when x D 2n C 12 , the function sin x D
1 D lim sup sin x and when x D 2n 12 , the function sin x D 1 D lim inf sin x.
x!1
x!1
x!a
x!a
!0C
These definitions of lim sup and lim inf have properties similar to the definitions
of lim sup and lim inf at infinity. That is, sup A and inf A are both monotone in
, so their limits as goes to 0 always exist. Moreover, there is a sequence <xn >
where lim xn D a such that lim f .xn / D lim sup f .x/ and another such sequence
n!1
n!1
x!a
96
3 Limits
such that lim f .xn / D lim inf f .x/. Similar definitions can be given for lim inf f .x/,
n!1
x!aC
x!a
x!a
x!aC
x!1
x!a1
The most important theorem concerning lim inf and lim sup is that lim f .x/ D L
x!a
if and only if lim inf f .x/ D lim sup f .x/ D L. Notice first that this is a biconditional
x!a
x!a
statement; that is, an if and only if statement. This requires that its proof have two
parts; one that assumes lim f .x/ D L and proves lim inf f .x/ D lim sup f .x/ D L
x!a
x!a
x!a
and another that assumes lim inf f .x/ D lim sup f .x/ D L and proves lim f .x/ D L.
x!a
x!a
x!a
So, given lim f .x/ D L, how can you conclude that lim inf f .x/ D lim sup f .x/ D
x!a
x!a
x!a
L? What you know is that given > 0, there is a > 0 such that for all x in the
domain of f for which 0 < jx aj < , you have jf .x/ Lj < . But this means that
for small > 0, the supremum sup A and infimum inf A are both within of L and,
therefore, the limits of sup A and inf A must both approach L as decreases to 0.
Conversely, suppose that lim inf f .x/ D lim sup f .x/ D L. Note that for any x a in
x!a
x!a
x!a
x!a
x!a
x!a
x!a
x!a
lim inf f .x/ D lim sup f .x/ D L, completing the first part of the proof.
x!a
x!a
(continued)
97
PART II: lim inf and lim sup equal L implies that the limit equals L
Assume that lim inf f .x/ D lim sup f .x/ D L.
x!a
x!a
For any x in the domain of f with x a, it follows that inf A2jxaj f .x/
sup A2jxaj .
Because lim inf A2jxaj D lim inf f .x/ D L, and lim sup A2jxaj D
x!a
x!a
x!a
lim sup f .x/ D L, the Squeezing Theorem shows that lim f .x/ D L, which
x!a
x!a
3.11.1 Exercises
1. Write definitions for each of the following.
(a) lim inf f .x/
x!aC
98
3 Limits
3. Prove that if a is any accumulation point of the domain of f , then lim inf f .x/
x!a
x!a
5. Suppose that lim inf f .x/ D L and lim inf g.x/ D M. What can you say about
x!a
x!a
x!a
6. Suppose that f is a positive-valued function with lim sup f .x/ D L > 0. Prove
1
that lim inf f .x/
D L1 .
x!a
x!a
Chapter 4
Continuity
99
100
4 Continuity
only requiring the inequality to hold for those x values satisfying 0 < jx aj <
which excludes x D a. This restriction is not necessary in the definition of continuity
of a function at a point.
Suppose that the point a is in the domain of the function f. Then f is continuous at a
means that for every > 0 there exists a > 0 such that for every x in the domain
of f satisfying jx aj < , it follows that jf .x/ f .a/j < .
Notice that the requirement that the point a be an accumulation point of the
domain of f has been dropped. As a result, if the function f is defined at an isolated
point a, then f is continuous at that point. A function that is not continuous at the
point a is discontinuous at the point a.
A function f is continuous on a set A if it is continuous at each point a 2 A. The
function whose graph appears in Fig. 4.1 is discontinuous at x D b because its limit
at x D b does not exist. Similarly, it is discontinuous at x D c. It is discontinuous
at x D d because it is not defined at that point even though the function has a limit
there. The function is continuous on the intervals a; b/, .b; c/, and .c; d/, and at
the points x D e and x D f . The function is not continuous on the intervals a; b
or c; d.
It is a direct consequence of the definition of continuity that if f is continuous
at a point a, and if a is an accumulation point of the domain of f , then the limit of
f .x/ at a exists and is, in fact, f .a/. To prove this you would just need to show that
if f satisfies the definition of continuity at a, then f also satisfies the definition of
lim f .x/ D f .a/. Writing down the definition of continuity gives you that for every
x!a
> 0 there is a > 0 such that jx aj < implies jf .x/ f .a/j < . But if this is
true, then certainly 0 < jx aj < implies jf .x/ f .a/j < , so the definition of
limit is satisfied.
101
102
4 Continuity
For a more challenging example, consider proving that the function f .x/ D
2x3 4x C 1 is continuous for all real numbers. This proof not only tackles a
more complicated function than the one in the previous example, it is supposed to
demonstrate the continuity of the function at the general real number a rather than
at a specific value such as a D 4. This requires the proof to select an arbitrary a and
prove the continuity of f at the point a. By showing that the function is continuous
at any arbitrarily chosen a, it shows that the function is continuous at every point a.
Again, the proof will select an arbitrary > 0 and needs to produce a > 0 such
that jf .x/ f .a/j < for all x satisfying jx aj < . The proof needs to select an
arbitrary a and an arbitrary > 0. Does it matter which it does first? In this case
where the choice of a does not depend on which is chosen, and the choice of
does not depend on which a is chosen, the order is not critical. It makes sense to
select the a first because you are then challenged to prove that f is continuous at
a for which you should choose an > 0. But since both quantifiers are universal
quantifiers (for all a 2 R and for all > 0), the order does not matter. If it had been
103
a universal quantifier and an existential quantifier such as for all > 0 there exists
a > 0, then the order would matter a great deal.
Working backwards from > jf .x/f .a/j you can see that you need > j.2x3
4xC1/.2a3 4aC1/j D j2.x3 a3 /4.xa/j D j2.xa/.x2 CxaCa2 /4.xa/j D
jxajj2.x2 CxaCa2 /4j. You should not be surprised and, in fact, be quite pleased
to see that this last expression contains a factor of jx aj because this will facilitate
making the expression small when jx aj is made small. One only needs to control
the size of the other factor j2.x2 C xa C a2 / 4j. Of course, if x is allowed to wonder
too far from a, this other factor could get arbitrarily large, so care must be taken to
restrict how far x gets from a. This can be done by requiring that not be larger than
some conveniently selected value such as 1. That means that jx aj < 1 would
imply, for example, that jxj < jaj C 1. Given this, there are many ways to find an
upper bound for the quantity j2.x2 C xa C a2 / 4j where the upper bound does not
depend on x. For example, j2.x2 C xa C a2 / 4j 2x2 C 2jxjjaj C 2a2 C 4
2.jaj C 1/2 C 2.jaj C 1/jaj C 2a2 C 4. One can afford to be sloppy here and get a
simpler looking upper bound by saying 2.jaj C 1/2 C 2.jaj C 1/jaj C 2a2 C 4
2.jaj C 1/2 C 2.jaj C 1/.jaj C 1/ C 2.jaj C 1/2 C 4.jaj C 1/2 D 10.jaj C 1/2 . All you
need is an upper bound that depends only on a. This leads to the following proof.
PROOF: The function f .x/ D 2x3 4x C 1 is continuous on the real
numbers.
Let f .x/ D 2x3 4x C 1, and let a 2 R.
Given > 0,
which is greater than 0 since 1, , and 10.jajC1/2
let D min 1; 10.jajC1/
2
are all positive.
Select x such that jx aj < . Then 1 implies jxj < jaj C 1.
Also, 10.jajC1/
2 implies that
3
jf .x/f .a/jDj.2x 4x C 1/ .2a3 4a C 1/j D j2.x3 a3 / 4.x a/jD
j2.x a/.x2 C xa C a2 / 4.x a/j D jx aj j2.x2 C xa C a2 / 4j
jx aj 2.jaj C 1/2 C 2.jaj C 1/jaj C 2a2 C 4
jx aj 2.jaj C 1/2 C 2.jaj C 1/.jaj C 1/ C 2.jaj C 1/2 C 4.jaj C 1/2 D
2
jx aj 10.jaj C 1/2 < 10.jajC1/
2 10.jaj C 1/ D .
Therefore, the function f is continuous at every real number a.
Not all functions
can be expressed with
nice formulas. Take, for example, the
2x if x is rational
function f .x/ D
which behaves differently on the rational
x C 1 if x is irrational
numbers than it does on the irrational numbers. Such functions that are defined
one way on the rational numbers and another way on the irrational numbers make
interesting examples because both the rational and the irrational numbers are dense
in the real numbers; that is, in every nonempty open interval .a; b/, you can find
both rational and irrational numbers. For the given function, in every nonempty
open interval .a; b/ there are values of x where f .x/ D 2x and other values of x
104
4 Continuity
where f .x/ D x C 1. Indeed, for most real numbers a, lim f .x/ does not exist. Only
x!a
at x D 1, where 2x and x C 1 coincide, does this limit exist, and, in fact, at that point
f .x/ is continuous (Fig. 4.2).
A proof that f is continuous at x D 1 would be similar to the two preceding
proofs, but you need to be careful to handle f .x/ differently depending on whether
x is rational or irrational. As in other continuity proofs, given an > 0 you are
faced with producing a value for > 0 which will ensure that jf .x/ f .1/j <
whenever jx aj < . If the function in the proof were equal to x C 1 for every
value of x, then the value D would work because jx 1j < shows that
jf .x/ f .1/j D j.x C 1/ .1 C 1/j D jx 1j < . If the function in the proof
were equal to 2x for every value of x, then the value D 2 would work because
jx 1j < 2 shows that jf .x/ f .1/j D j.2x/ .2 1/j D 2jx 1j < . In this proof,
then, you can choose D min.; 2 / D 2 . After selecting an x with jx 1j < ,
you merely consider two separate cases, one where x is rational, and one where x is
irrational. These ideas allow you to produce the following proof.
PROOF: The function f .x/ D
x D 1.
Let f .x/ D
2x if x is rational
x C 1 if x is irrational
2x if x is rational
.
x C 1 if x is irrational
is continuous at
Given > 0,
let D 2 which is greater than 0 since > 0.
Select x such that jx 1j < D 2 .
If x is a rational number, then jf .x/ f .1/j D j2x 2j D 2jx 1j < 2 D .
If x is an irrational number, then jf .x/ f .1/j D j.x C 1/ 2j D jx 1j <
< .
In either case, jf .x/ f .1/j < .
Therefore, the function f is continuous at 1.
105
4.2.1 Exercises
Write proofs of each of the following statements.
1.
2.
3.
4.
5.
6.
7.
106
4 Continuity
the point a where the continuity needs to be shown. These functions are special and
satisfy the following definition. A function f is uniformly continuous on the set
A if for every > 0 there is a > 0 such that jf .x/ f .y/j < for every x and y
in A satisfying jx yj < . You should compare this definition to the definition of
continuity at a point. The difference centers on when the value of > 0 needs to
be determined. For continuity at a single point, given > 0, one must specify the
value of > 0 after being given the value of a but before being given a value for x.
Thus, the value of > 0 can depend on the value of a even though it cannot depend
on the value of x. On the other hand, for uniform continuity, given > 0, one must
specify the value of > 0 before learning the values of either x or y, and, therefore,
its value cannot depend on either x or y.
The definition of uniform continuity suggests a template for how to prove that a
given function f is uniformly continuous on a set A. As in the proof for continuity
at a point, you would say that a value for > 0 has been given. Then you would
present a value for > 0. Once these two values have been specified, you would
need to show that any x and y in A that satisfy jxyj < also satisfy jf .x/f .y/j < .
This suggests the following.
TEMPLATE for proving the function f is uniformly continuous on the
set A
SET THE CONTEXT: Make statements about what is known about the
function f .
SELECT AN ARBITRARY : Given > 0,
PROPOSE A VALUE FOR : let D
. Here you would insert an
appropriate value for .
SELECT ARBITRARY x and y in A with jx yj < : Let x and y be in A
such that jx yj < .
(continued)
107
Less clear is how to choose a value for > 0 when proving f .x/ D x2 1C1
is uniformly continuous on the real numbers. To do this, you need to find a
way to show jf .x/ f .y/j < . You would try to find an upper bound for
2
2 C1/j
jxCyj
D .x2 C1/.y
jf .x/f .y/j D x2 1C1 y2 1C1 D j.y.x2C1/.x
2 C1/ jxyj. This expression
C1/.y2 C1/
is complicated, so it is convenient to find ways to simplify it. The nice thing about
working with inequalities rather than equalities is that you are not prevented from
making changes that increase the value of your expression. That is, if you can
simplify an expression by substituting an expression that is a little larger, that might
not be a problem. The numerator in the previous expression is jx C yj which does
not simplify algebraically, but it does suggest a possible application of the triangle
inequality, jx C yj jxj C jyj. Changing jx C yj to jxj C jyj allows the fraction to
be broken into two simpler
fractions. It allows you
to continue
with jf .x/ f .y/j D
jxCyj
jyj
jxj
jx yj.
jx yj .x2 C1/.y2 C1/ C .x2 C1/.y2 C1/ jx yj x2jxj
C y2jyj
.x2 C1/.y2 C1/
C1
C1
When jxj < 1, you can conclude that jxj < 1 x2 C 1. When jxj 1, you can
2
conclude that jxj x2 < x2 C 1. In either case .x2jxj
xx2 C1
D 1. This lets you
C1/
C1
jxCyj
jyj
jxj
jx yj
C
state that jf .x/ f .y/j D .x2 C1/.y
2 C1/ jx yj
2
2
2
2
.x C1/.y C1/
.x C1/.y C1/
2jx yj. This suggests that D 2 will work in the proof.
108
4 Continuity
1
x2 C1
jx 2 yj < 2 D 2 .
C1/j
Then jf .x/ f .y/j D x2 1C1 y2 1C1 D j.y.x2C1/.x
D
C1/.y2 C1/
jxCyj
jyj
jxj
jx yj
jx yj .x2 C1/.y
2
2
2 C1/ C .x2 C1/.y2 C1/
.x C1/.y C1/
jxj
jx yj
C y2jyj
x2 C1
C1
Note that if jxj < 1, then jxj < x2 C 1, and if jxj 1, then jxj x2 < x2 C 1.
In either case, jxj < x2 C 1, so x2jxj
< 1, and similarly, y2jyj
< 1.
C1
C1
jyj
jxj
It follows that jf .x/ f .y/j x2 C1 C y2 C1 jx yj < 2jx yj < 2 D .
Therefore, the function f is uniformly continuous on the real numbers.
One of the most memorable theorems from Calculus is the Mean Value
Theorem which states that if the function f is continuous on the interval a; b
and differentiable on the interval .a; b/, then there is a c 2 .a; b/ such that
.a/
f 0 .c/ D f .b/f
. If the function f has a bounded derivative on the interval
ba
a; b, that is, if there is a positive real number M such that jf 0 .x/j M for all
values of x 2 a; b, then one can easily see that f is uniformly continuous on that
interval. Indeed, if x and y are in a; b, then there is a c between x and y such that
jf .x/ f .y/j D jf 0 .c/j jx yj M jx yj. This implies that given > 0, the value
D M > 0 can be used in a proof that f is uniformly continuous on a; b for then
jx yj < implies jf .x/ f .y/j D jf 0 .c/j jx yj < M jx yj < M D . This
is summarized by saying that a function with a bounded derivative on an interval is
uniformly continuous there.
Whenever you learn of the truth of a conditional statement such as the one at the
end of the previous paragraph (bounded derivative implies uniform continuity), it is
natural to ask whether the converse of the statement is also true (uniform continuity
implies bounded derivative). The answer to this particular question is no, not all
functions uniformly continuous on an interval have bounded derivatives there. In
particular, the function f .x/ D jxj is an example of a function uniformly continuous
on the entire real line, yet it fails to be differentiable at x D 0. The function f .x/ D
p
x is uniformly continuous for x 0, but its derivative is unbounded
near x D 0.
A more complex example is the function defined by f .x/ D x2 sin x12 when x
0 and f .0/ D 0. This function is uniformly continuous on the interval 10; 10
even though its derivative, which exists on the entire real line, is not bounded as x
approaches 0.
p
Because the function f .x/ D x has an increasingly large rate of change as x
approaches 0, proving that the function is uniformly continuous for x 0 provides
an interesting challenge. The proof will need to conclude that > jf .x/ f .y/j D
109
p p p p
p p
j x yj. xC y/
p p
p . As expected, there is a factor of jx yj in
j x yj D
D pjxyj
xC y
xC y
this expression, so that you can try to make the expression small by
the
prestricting
p
size of jx yj. This is easy if the denominator of the expression, x C y, does
not get too small. The problem is if x and y get close to 0, the denominator of the
expression will also get close to 0. At first this seems
p likepa significant roadblock.
But this roadblockppresents its own resolution for if x C y is very small, it must
p
certainly be that j x yj is even smaller
is the conclusion that you want.
p which
p
In other words, there are two
cases:
either
x
C
y
is small which would imply that
p
p
jf .x/ f .y/j is small, or x C y is large which would imply that jf .x/ f .y/j D
jxyj
p p is small. You only need to decide what to use as the dividing line between
xC y
p
p
large and small. A natural choice would be itself because x C y <
p
p
p
p
jxyj
implies j x yj < . If x C y , then jf .x/ f .y/j D pxCpy jxyj
2
which suggests letting D 2 so that jx yj < gives jf .x/ f .y/j < D . The
complete proof follows.
p
PROOF: The function f .x/ D x is uniformly continuous on the interval
x 0.
p
Let f .x/ D x.
Given > 0,
let D 2 which is greater than 0 since 0.
Let x and y be nonnegative
real numbers such that jx yj < .p
p
p
p
In
the
case
that
x
C
y
<
, it follows that jf .x/ f .y/j D j x yj
p
p
x C y < . p
p
p
p
In the case that x C y , it follows that jf .x/ f .y/j D j x yj D
p p p p
j x yj. xC y/
p p
xC y
jxyj
p p
xC y
jxyj
<
2
D .
In either case, jx yj < implies that jf .x/ f .y/j < , so the function f is
uniformly continuous on the interval x 0.
There is an important lesson to be learned from this example. When planning how
to write a proof, you can pursue one line of thinking which may solve the problem
in most but not all cases. Sometimes the special cases where the argument does not
work are enough to cause you to abandon your original line of reasoning altogether.
But often you can just break your argument into two or more cases and find other
techniques to handle the special cases where the original argument does not work.
4.3.1 Exercises
Write proofs of each of the following statements.
1. f .x/ D 3x C 11 is uniformly continuous on the set of real numbers.
2. f .x/ D 14x C 5 is uniformly continuous on the set of real numbers.
3. f .x/ D jxj is uniformly continuous on the set of real numbers.
110
4.
5.
6.
7.
8.
4 Continuity
111
112
4 Continuity
PROOF (HeineBorel Theorem): Let a < b be two real numbers, and let
T be an open cover of a; b. Then T contains a finite subcover of a; b.
Let a < b be two real numbers, and let T be an open cover of a; b.
Define set S D fx 2 a; b j T has a finite subcover that covers the interval
a; xg.
The set T is an open cover of a; b, and a 2 a; b, so T must contain at
least one open interval, .p; q/ which contains the point a, that is, p < a < q.
Since the interval a; a is covered by .p; q/ 2 T, the point a 2 S, and S is
not an empty set.
The set S is bounded above by b.
Since S is nonempty and bounded above, it has a least upper bound r.
Since r must be at least a and cannot be greater than b, r 2 a; b, so there
is an interval .p; q/ in T which contains the point r, that is, p < r < q.
Since p < r and r is the least upper bound of S, p is not an upper bound of
S. Thus, there is a point y 2 S with p < y. This means that there is a finite
set of intervals in T that covers a; y.
Let z D min. rCq
; b/. Since z r and z 2 .p; q/, adding the interval .p; q/
2
to the finite set of intervals of T that covers a; y produces a finite set of
intervals in T that covers a; z, and z 2 S.
But r is the least upper bound for S, implying that z r. Because z D
min. rCq
; b/ and rCq
> r, it must be that z D b.
2
2
Because z 2 S, it follows that b 2 S which completes the proof of the
theorem.
The second proof of the HeineBorel Theorem is a proof by contradiction. It
begins as the first proof by assuming that a < b are real numbers, and that the
interval a; b has an open cover T. Then it makes the additional assumption that no
finite collection of intervals in T can cover a; b. This will lead to a contradiction.
This proof is not one that the beginning student is likely to invent on their own
unless they have seen the technique before.
First, the proof sets a0 D a and b0 D b so that the interval a0 ; b0 D a; b. Let
0
m0 D a0 Cb
be the midpoint of a0 ; b0 . It must be the case that at least one of the
2
intervals a0 ; m0 or m0 ; b0 cannot be covered by a finite number of intervals in T
because, if both can be covered by a finite number of intervals, putting those two
collections together would give a finite collection of intervals that covered the entire
interval a0 ; b0 D a; b contradicting the assumption that this could not be done.
p
[
a
]
b
113
j!1
j!1
j!1
Note that since the sequence of aj s increases to r, and the sequence of bj s decrease
to r, the limit r 2 aj ; bj for each j. Because the limit, r, is in a; b, there is an open
interval .p; q/ 2 T such that r 2 .p; q/. The distance the limit r is from the boundary
of the interval .p; q/ is D min.r p; q r/ > 0. Since lim ba
D 0, you can
2j
j!1
r
Fig. 4.5 HeineBorel Theorem second proof
114
4 Continuity
PROOF (HeineBorel Theorem): Let a < b be two real numbers, and let
T be an open cover of a; b. Then T contains a finite subcover of a; b.
Let a < b be two real numbers, and let T be an open cover of a; b.
Assume that T contains no finite subcover of a; b.
Let a0 D a and b0 D b so that the interval a0 ; b0 D a; b, and note that no
finite collection of intervals in T will cover a0 ; b0 .
Define sequences <aj > and <bj > inductively. For j 0, let aj ; bj a; b
be an interval which cannot be covered by a finite collection of open
intervals in T, and where bj aj D ba
.
2j
aj Cbj
Let mj D 2 be the midpoint of aj ; bj .
It must be the case that at least one of the intervals aj ; mj or mj ; bj
cannot be covered by a finite number of intervals in T because, if both can
be covered by a finite number of intervals, putting those two collections
together would give a finite collection of intervals that covered the entire
interval aj ; bj .
If aj ; mj cannot be covered by a finite collection of intervals, let ajC1 D aj
and bjC1 D mj . Otherwise, let ajC1 D mj and bjC1 D bj . In either case
ajC1 ; bjC1 cannot be covered by a finite collection of intervals from T, and
ba
j
j!1
j!1
ba
/,
2j
j!1
115
(z )
Fig. 4.6 y and z straddle one endpoint but remain in an interval of the open cover
116
4 Continuity
A cleaner way to ensure that any y and z within of each other are in one of
the finite number of intervals in the open cover of a; b is to be more clever about
choosing the original open intervals. Suppose that for all y 2 .x ; x C /, it
follows that jf .y/ f .x/j < . You can be very conservative and use the open
interval .x 2 ; x C 2 / as the interval chosen to cover x in the open cover of a; b.
Then if y and z are very close, and y 2 .x 2 ; x C 2 / for some x, it will follow that,
since y and z will be closer together than 2 , guaranteeing that both y and z will be in
.x ; x C /, and the result will follow. The following proof uses the first strategy.
PROOF: A function continuous on a closed bounded interval is uniformly
continuous on that interval.
Let a < b be two real numbers, and let f be continuous on the interval a; b.
Let > 0 be given.
By the definition of continuity, for each x 2 a; b there is a x > 0 such that
for all y in a; b with jy xj < x it follows that jf .y/ f .x/j < 2 .
Because for each x 2 a; b, the point x 2 .x 2x ; x C 2x /, this collection of
open intervals covers a; b.
By the HeineBorel Theorem, there is a finite set fx1 ; x2 ; x3 ; : : : ; xn g a; b
such that the collection I D f.xj xj ; xj C xj / j j D 1; 2; 3; : : : ; ng forms
an open cover of a; b.
The set of endpoints of these intervals, E D fxj xj j j D 1; 2; 3; : : : ; ng, is
a finite set, so let be the smallest positive difference between two elements
of E.
Let y and z be elements of a; b with jy zj < .
Because the collection of intervals I is an open cover of a; b, there are j
and k such that y 2 .xj xj ; xj C xj / and z 2 .xk xk ; xk C xk /. If j D k,
then y and z are in the same interval of I. If j k, then because jy zj < ,
there is at most one endpoint in E between y and z. Thus, either there is at
most one endpoint of .xj xj ; xj C xj / or .xk xk ; xk C xk / between
y and z. This implies that either y and z are both in .xj xj ; xj C xj /
or both in .xk xk ; xk C xk /. In either case, there is a single interval
.xm xm ; xm C xm / 2 I such that y; z 2 .xm xm ; xm C xm /.
Now it follows that jf .y/ f .z/j D j .f .y/ f .xm // .f .z/ f .xm // j
jf .y/ f .xm /j C jf .z/ f .xm /j < 2 C 2 D .
This shows that for every > 0 there is a > 0 such that if y; z 2 a; b
with jy zj < , then jf .y/ f .z/j < . This completes the proof that f is
uniformly continuous on a; b.
Note that the fact that a; b is both closed and bounded is crucial. The function
f .x/ D x2 is continuous on the unbounded interval 0; 1/, but f is not uniformly
continuous on this interval. Similarly, the function f .x/ D 1x is continuous on the
open interval .0; 1/, but f is not uniformly continuous on this interval.
117
4.4.4 Exercises
1. Determine which of the following sets of real numbers are compact.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
0; 12
.2; 2
1; 4 [ 8; 15
f1; 3; 5; 7; 9g
R
;
2; 6/ [
.6; 11
S
h
i
1
1
1
f0g [
jD1 2jC1 ; 2j
118
4 Continuity
points in the intersection of the domain of f and the domain of g and, in the case
of gf , only where g is not 0. Generally, one is interested in functions that have a
common domain, but sometimes this is not the case. Pathological examples do exist.
It could be, for example, that f is only defined for positive real numbers, and g is
only defined for negative real numbers as with f .x/ D p1x and g.x/ D p1x . Then
f C g has
domain and is the empty function, one that contains no ordered
an empty
pairs, x; f .x/ . Oddly, the definition of continuity says that the empty function is
continuous because it satisfies the definition at each point of its empty domain.
Suppose that functions f and g have a common domain where the point a is an
accumulation point of that domain. Also suppose that lim f .x/ D L and lim g.x/ D
x!a
x!a
H. Recall that when proving that the limit of f C g is L C H, you are given an > 0
and can use the definition of limit to conclude that there are 1 > 0 and 2 > 0
such that if x is in the common domain of f and g with 0 < jx aj < 1 , then
jf .x/ Lj < 2 , and if 0 < jx aj < 2 , then jg.x/ Hj < 2 . Then the triangle
inequality allows you to conclude that for all x with 0 < jx aj < min.1 ; 2 / that
j .f .x/ C g.x// .L C H/j D j .f .x/ L/ C .g.x/ H/ j jf .x/ Lj C jg.x/ Hj <
C 2 D . The same method works for the proof about continuity of f C g at a
2
with minor changes made to match the template for writing proofs about continuity
of a function at a point. Of course, the same logic works for proving the continuity
of f g, so the two results might as well be combined as follows.
PROOF: Suppose that f and g are functions with common domain
containing the point a. If both f and g are continuous at the point a, then
so are the functions f C g and f g.
Let f and g be functions both defined on a set A containing the point a, and
assume that f and g are both continuous at a.
Let > 0 be given.
By the definition of continuity, there is a 1 > 0 such that if x 2 A and
jx aj < 1 , then jf .x/ f .a/j < 2 .
Similarly, there is a 2 > 0 such that if x 2 A and jx aj < 2 , then
jg.x/ g.a/j < 2 .
Let D min.1 ; 2 /.
Then
if x 2 Awith
jx aj < ,
j f .x/ g.x/ f .a/ g.a/ j D j f .x/ f .a/ g.x/ g.a/ j
jf .x/ f .a/j C jg.x/ g.a/j < 2 C 2 D .
This shows that f C g and f g are continuous at a.
Now suppose that f and g are functions as discussed above with lim f .x/ D L
x!a
and lim g.x/ D H. Recall how you prove that lim f .x/g.x/ D LH. Again, as with
x!a
x!a
the proof for the sum of the limits, given > 0 you find > 0 so that both jf .x/ Lj
and jg.x/ Mj are small when 0 < jx aj < . How small do these need to
be? The idea was to write jf .x/g.x/ LHj as jf .x/g.x/ f .x/H C f .x/H LHj
jf .x/j jg.x/ Hj C jHj jf .x/ Lj. Thus, 1 > 0 can be chosen to ensure that
119
jf .x/ Lj is less than 1, 2 > 0 so that jf .x/ Lj is less than 2.jHjC1/
, and 3 so that
jg.x/ Hj is less than 2.jLjC1/ . Then can be set to the least of 1 , 2 , and 3 . The
proof for continuity of fg at the point a follows this same strategy.
f .x/
and lim g.x/ D H and H 0. This time recall how you prove that lim g.x/
D HL .
x!a
x!a
The idea is the same
but the algebra took
the proof for products,
as with
f .x/HLg.x/
f .x/
L
a few more steps. g.x/ H D g.x/H D f .x/HLHCLHLg.x/
D
g.x/H
.f .x/L/HCL Hg.x/
g.x/H
jg.x/j
jg.x/jjHj
H 2
jg.x/ Hj < 4.jLjC1/ . This allowed you to conclude f .x/HLg.x/
< . Again, the
g.x/H
proof for continuity can be constructed by changing the limit L to jf .a/j, the limit
H to jg.a/j, and making some other minor wording changes.
120
4 Continuity
2
g.x/ C g.a/ g.x/ D jg.a/j which implies that jg.x/j > jg.a/j jg.a/j D
2
jg.a/j
.
2
By the definition of continuity, there is a 2 > 0 such that if x 2 A and
jx aj < 2 , then jf .x/ f .a/j < jg.a/j
.
4
By the definition of continuity, there is a 3 > 0 such that if x 2 A and
2
.
jx aj < 3 , then jg.x/ g.a/j < 4.jfg.a/
.a/jC1/
Let D min.1 ; 2 ; 3 /.
Then
if x 2 A with
0 < jx aj < ,
C g.x/g.a/ <
g.x/g.a/
jg.x/j
2
jg.a/j
4
2
jg.a/j
C 4.jfg.a/
2jf .a/j < 2 C
.a/jC1/ jg.a/j2
This shows that fg is continuous at a.
2
D .
4.5.1 Exercises
1. Suppose that f and g are functions that are both uniformly continuous of a set
A. Find an example showing that their product need not be uniformly continuous
on A.
Write proofs for each of the following statements.
5
121
122
4 Continuity
As easy as this proof is, the continuity of jxj can more easily be proved as
follows.
PROOF: The function jxj is continuous.
p
Let g.x/ D x2 , and f .x/ D x.
Let a be any real number.
2
Then g is continuous at a, and,
p since g.a/ D a 0, f is continuous at g.a/.
2
Because the function jxj D x D .f g/.x/, it follows that jxj is continuous
at a.
In turn, this result can be used to show that if f and g are functions with domain A,
and f and g are both continuous at a 2 A, then the functions min.f ; g/ and max.f ; g/
are both continuous at a. This is because the functions min.f ; g/ and max.f ; g/ can
be expressed in terms of absolute value.
PROOF: If f and g are functions with the same domain A, and both functions are continuous at a 2 A, then the function min.f ; g/ is continuous
at a.
Let f and g be functions with the same domain A, and assume that both
functions are continuous at a 2 A.
Note that for any two real numbers y and z, if y > z, then y C z jy zj D
y C z .y z/ D 2z, but if y z, then y C z jy zj D y C z .z y/ D 2y.
In either case, y C z jy zj D 2 min.y; z/.
.x/g.x/j
Thus, for any x, min f .x/; g.x/ D f .x/Cg.x/jf
.
2
Since f and g are continuous at a, so is f g.
Since f g is continuous at a, so is jf gj.
gj
It then follows that the combination min.f ; g/ D f Cgjf
is continuous
2
at a.
123
4.6.1 Exercises
1. Find examples of functions f and g defined on R with lim f .x/ D L and
lim g.y/ D M such that lim g .f .x// M.
y!L
x!a
x!a
124
4 Continuity
quantifier, and the second quantifier is a universal quantifier. Thus, the statement
there is an M such that for all x 2 a; b, jf .x/j M has an existential quantifier
stating that there exists a number M satisfying a property. This is followed by a
universal quantifier stating that all x in the interval a; b satisfy a property. Finally,
the property is given as jf .x/j M.
The rule of thumb for constructing the negation of statements with quantifiers
is to replace each existential quantifier with a corresponding universal quantifier,
replace each universal quantifier with a corresponding existential quantifier, and
replace the property with the negation of that property. In this example, the
existential quantifier there exists a number M would be replaced by the universal
quantifier for all numbers M. Then the universal quantifier for all x 2 a; b
would be replaced by the existential quantifier there exists an x 2 a; b. Finally,
the property jf .x/j M would be replaced by its negation jf .x/j > M. The
resulting negation is for all numbers M there is an x 2 a; b such that jf .x/j > M.
Your proof of the boundedness of f would begin by introducing f and the interval
a; b. Then it would assume negation just discussed. The remainder of the proof
would be to derive a contradiction, and that would show that the assumption made
at the outset of the proof is false, so its negation, the statement you were trying to
prove, must be true.
Thus, the proof would begin with a statement about f being a continuous function
on the closed bounded interval a; b which would be followed by the negation
of the statement you want to prove. So how do you use this negation to reach a
contradiction? Well, just see where this assumption leads you. If for each M you
can find an x 2 a; b where jf .x/j > M, it means that there is an x1 such that
jf .x1 /j > 1. Similarly, there is an x2 such that jf .x2 /j > 2. In this way, you can
assert that there is a sequence x1 ; x2 ; x3 ; : : : such that for each n 1, jf .xn /j > n.
Note that this gives you an infinite sequence of values in the closed bounded interval
a; b. The BolzanoWeierstrass Theorem states that every infinite bounded set has
an accumulation point. Does the sequence x1 ; x2 ; x3 ; : : : produce such an infinite
bounded set? Well, it is certainly bounded because each xn is in the interval a; b.
Is it possible that the sequence does not give an infinite collection of points? For
that to happen, it would have to be the case that infinitely many of the value in
the sequence were equal to each other. Actually, just because you choose x1 so that
jf .x1 /j > 1 does not preclude having jf .x1 /j > 100, so the value x1 could appear
in the sequence many times. This is awkward. It would be easier if you chose a
sequence of distinct values. This is actually not hard to do. Rather than choosing xn
so that jf .xn /j > n, why not choose x1 as above, and for each n 1 choose xnC1 so
that jf .xnC1 /j > jf .xn /j C 1. This would not only imply that for each n, jf .xn /j > n
but also that xn could not equal any of the values that appear earlier in the sequence.
So what can you do with the infinite sequence of xn values with its guaranteed
accumulation point, y? First note that the accumulation point y is also in a; b
because all of the xn values satisfy both a xn and xn b, so the accumulation
point y must also satisfy a y b. Otherwise, there would be an interval around y
125
x1
x3 x4
x6 x5 x2 b
that did not share any points with a; b, so it would not contain any of the xn values.
This means that f is defined and continuous at y. That implies that there is a > 0
such that for all x 2 a; b satisfying jx yj < , it follows that jf .x/ f .y/j < 1.
But that means jf .x/j < jf .y/j C 1. But y is an accumulation point for the sequence
of xn values, so there are infinitely many of the xn within of y, and some of them
will necessarily have the property that jf .xn /j > jf .y/j C 1. This gives the needed
contradiction (Fig. 4.7).
PROOF: A function continuous on a closed bounded interval is bounded.
Let a b be real numbers, and let f be a function continuous on the interval
a; b.
Assume that f on a; b is not bounded. That is, for every real number M,
there is an x 2 a; b such that jf .x/j > M.
Select x1 2 a; b so that jf .x1 /j > 1.
Construct a sequence inductively as follows. Assume that for some n 1
the sequence x1 ; x2 ; x3 ; : : : ; xn has been selected. Choose xnC1 2 a; b such
that jf .xnC1 /j > jf .xn /j C 1.
Note that for each n the xn chosen in this manner must be distinct from all
values of xj chosen before it in the sequence, and that jf .xn /j > n.
The terms of the sequence x1 ; x2 ; x3 ; : : : form an infinite set contained in the
interval a; b, so it is an infinite bounded set. By the BolzanoWeierstrass
Theorem the set has an accumulation point y.
If y lies outside of a; b, then there is an open interval containing y that
contains no points of a; b. Thus, that open interval would not contain any
terms of the sequence which cannot happen if y is an accumulation point of
the sequence. Therefore, y 2 a; b.
(continued)
126
4 Continuity
maximum
y = f(x)
minimum
127
128
Fig. 4.9 f passing through
each y between f .c/ and f .d/
4 Continuity
f(c)
f(d)
f without lifting your pencil from the paper, then this intermediate value property
becomes clear because in going from f .c/ to f .d/, your pencil will necessarily cross
over all the y values between f .c/ and f .d/.
To prove the Intermediate Value Theorem you would begin by setting the context
by introducing a function f continuous on an interval a; b and points c and d with
a c d b. Then you would select an arbitrary y between f .c/ and f .d/.
The proof would have to demonstrate the existence of an x between c and d with
f .x/ D y. How is this to be done? As with many other proofs in Analysis, one shows
the existence of a real number by constructing a set for which that number is a least
upper bound. Consider, for example, the case where f .c/ < y < f .d/. You could
construct the set S D fx 2 c; d j f .x/ yg. This set is not an empty set because
c 2 S, and S is certainly bounded above by d. Thus, the Completeness Axiom says
that the set has a least upper bound, s. Now you can refer to the continuity of f
which will show that if f .s/ < y, then there is a > 0 such that jx sj <
implies that f .x/ < y showing that there are values greater than s for which f .x/ < y
contradicting the fact that s is an upper bound of S. If f .s/ > y, then there is a > 0
such that jx sj < implies that f .x/ > y showing that s < s is an upper bound
for S contradicting the fact that s is the least upper bound of S. The only remaining
conclusion is that f .s/ D y which provides the needed example, x D s, needed to
prove the theorem.
Note that the above argument did not cover the general case where f .c/ and f .d/
can be in any order. The argument so far only covers the specific case where f .c/ <
f .d/. So is there more proof to write? It is easy to see that the case f .c/ > f .d/ can be
proved with an argument virtually identical to the one given above by changing the
sense of some of the inequalities. The case of f .c/ D f .d/ is even easier because the
only possible y between f .c/ and f .d/ is f .c/, so the value x D c gives the needed
f .x/ D y. Thus, giving the argument for f .c/ < f .d/ essentially covers all the
needed cases, and it would be very easy for the reader to add the needed arguments
to complete the proof for the missing cases. In this situation it is common for the
proof to cover only the specific condition f .c/ < f .d/ and introduce it with the
phrase without loss of generality. In this case the phrase means that although the
following assumption looks like it only covers some of the necessary cases, in order
129
to make the argument completely general, the omitted cases are either very easy or
virtually identical to the case being considered. With this in mind, the following is
a proof of the Intermediate Value Theorem.
PROOF (Intermediate Value Theorem): Let the function f be continuous
on the interval a; b containing c and d. If y is any value between f .c/ and
f .d/, then there exists x between c and d such that f .x/ D y.
Let f be a function continuous on a; b, and let c and d be in a; b.
Let y be any value between f .c/ and f .d/.
Without loss of generality, assume that c d and f .c/ y f .d/.
Let set S D fx 2 c; d j f .x/ yg.
S is not empty because f .c/ y implying c 2 S.
S is bounded above by d.
By the Completeness Axiom S has a least upper bound s which will be an
element of a; b.
If f .s/ < y, then by the continuity of f , there is a > 0 such that if x 2 a; b
with jx sj < , then jf .x/ f .s/j < yf2.s/ , and, in particular, f .x/ < y.
This shows that there is an x > s with f .x/ < y, so x 2 S contradicting the
fact that s is an upper bound of S.
If f .s/ > y, then by the continuity of f , there is a > 0 such that if x 2 a; b
with jx sj < , then jf .x/ f .s/j < f .s/y
, and, in particular, f .x/ > y.
2
This shows that for all x between s and s that f .x/ > y, so s is an
upper bound of S contradicting the fact that s is the least upper bound of S.
It follows that f .s/ must equal y which completes the proof of the theorem.
In the above proof the steps which begin If f .x/ < y and If f .x/ > y are written
in exactly the same style using almost identical words. If you were writing a short
story, you would avoid writing in this style because it might sound monotonous to
the reader. In creative writing, you would want to be more creative, and you would
reach for your thesaurus to find alternate words to enhance your writing. But in a
mathematical proof, using such parallel construction of sentences actually makes
the proof easier to read. A reader only needs to parse the first of the two steps in
order to have a good idea of what is going to be done in the second of the two steps.
This gives the reader a head start on processing the second step. What is passed off
as boring in creative writing can be applauded in the writing of proofs because of
the way it simplifies the understanding. In fact, one often begins the second of two
such steps with the word similarly to indicate that the argument to follow looks a lot
like the one just completed, again alerting the reader to the parallel construction.
The Intermediate Value Theorem says that functions continuous on an interval
have the intermediate value property there. But a function need not be continuous
for it to have the intermediate value property. Clearly, if a function has a jump
discontinuity at a point a, that is, if lim f .x/ and lim f .x/ both exist but are
x!a
x!aC
different as shown in Fig. 4.10, then there could well be values of y that the function
misses as it passes from .c; f .c// to .d; f .d//.
130
4 Continuity
f(c)
f(d)
1
x
For a discontinuous function to have the intermediate value property, the function
must necessarily
oscillate wildly (Fig. 4.11). A typical example is the function
sin 1x if x > 0
f .x/ D
.
0
if x 0
4.7.4 Exercises
Write proofs for each of the following statements. Each statement can be proved
using one or more of the theorems in this section.
1. Let A R be a bounded set, and let f be a function defined on A. If f is
unbounded on A, then for every > 0, there exists a and b in R with b a <
such that f is unbounded on A \ .a; b/.
2. If a < b and f is a continuous function on a; b with f .a/ D f .b/, then there is a
c 2 .a; b/ such that f obtains an extreme value (either a minimum or maximum)
at c.
3. Suppose that f is a continuous function defined on R such that lim f .x/ D
x!1
x!1
4.8 Discontinuity
131
4. If p is an odd degree polynomial with real coefficients, then p has at least one
real root.
5. Suppose that a plane contains be a polygon G and a line L. Then there is a line
L0 in the plane parallel to L such that exactly half the area of G lies on each side
of L0 .
r
1
2
6. There is a value of x between 0 and 1 such that x equals
.
1 C x2
4.8 Discontinuity
In Calculus students learn about a great many continuous functions. These include
the elementary functions: polynomials, rational functions, algebraic functions,
exponential functions, logarithmic functions, and circular and hyperbolic trigonometric functions and their inverses. How badly can a function be discontinuous? A
function can
8 be discontinuous
9 at a single point such as the signum or sign function
< 1 if x < 0 =
sgn.x/ D
0
if x D 0 or at a sequence of points such as the floor or greatest
:
;
1
if x > 0
integer function bxc D n if n is the integer satisfying n x < n C 1 (Fig. 4.12).
A function
( can be discontinuous at a sequence of points
) that converge such as with
1
1
1
if
<
x
;
for
positive
integer
n
nC1
n
f .x/ D n
. This function is discontin0 otherwise
uous at each x D 1n for positive integers n, but it is continuous everywhere else
including at x D 0 (Fig. 4.13). A function can be discontinuous at every x such as
0 if x is rational
with f .x/ D
.
1 if x is irrational
But one of the most surprising examples is the following often called Thomaes
function but also known as the popcorn function, the raindrop function,
132
4 Continuity
or the modified
Dirichlet function. It is defined on
the interval .0; 1/ by
1
m
if
x
is
rational
written
in
lowest
terms
as
n
n . Its graph is shown in
f .x/ D
0 if x is irrational
Fig. 4.14. It is not hard to see that this function is discontinuous at each rational
number mn 2 .0; 1/. Indeed if mn is in lowest terms, then f . mn / D 1n . If is set
1
for every >
be irrational numbers x 2 .0; 1/ satisfying
0 there will
at 2n ,mthen
x < for which f .x/ f m D j0 1 j > . On the other hand, at each
n
n
n
irrational number a in .0; 1/, the function is continuous. To see this, given an
> 0, notice that there are only finitely many rational numbers r 2 .0; 1/ such that
f .r/ . If there are such rational numbers, there is one, r0 , closest to a, so choose
D jr0 aj. If there are no such rational numbers, you can choose D 1. In either
case, for all x 2 .0; 1/ with jx aj < , it follows that jf .x/ f .a/j < , showing
that f is continuous at a.
Chapter 5
Derivatives
133
134
5 Derivatives
(a, f(a))
f(x) - f(a)
x -a
(x, f(x))
135
x!a
x!a
x!a
0 f 0 .a/ D 0.
Thus, f .x/ is both defined at x D a, and lim f .x/ f .a/ D 0, or
lim f .x/ D f .a/.
x!a
It follows that f is continuous at x D a.
x!a
136
5 Derivatives
For example, consider the function f .x/ D 3x2 8x. To calculate the derivative
of f at a D 4, one needs to evaluate the limit
3x 8x 3 42 8 4
f .x/ f .4/
3x2 8x 16
lim
D lim
D lim
x!4
x!4
x!4
x4
x4
x4
.3x C 4/.x 4/
D lim
D lim 3x C 4 D 16:
x!4
x!4
x4
Since each step of this derivation follows either from rules of algebra or from
the theorems about calculating the limits of various arithmetic combinations of
functions, the calculation given is a complete proof that the derivate of f at x D 4
is 16.
In a more general setting, consider proving that the derivative of f .x/ D 5x4 at
the point x D a is f 0 .a/ D 20a3 . Here you would calculate
5.x a/ x3 C x2 a C xa2 C a3
f .x/ f .a/
5x4 5a4
lim
D lim
D lim
x!a
x!a
x!a
xa
xa
xa
D lim 5.x3 C x2 a C xa2 C a3 / D 5.a3 C a2 a C aa2 C a3 / D 20a3 :
x!a
Again, finding a factor of x a in the numerator of the difference quotient is the key
to evaluating the needed limit.
to show that .cf /0 .a/ D cf 0 .a/ is just a multiple of a known limit, the needed result
follows immediately from the fact that the limit of a constant times a function is the
constant times the limit of the function.
137
x!a
.a/
D lim c f .x/f
D c lim
xa
x!a
x!a
f .x/f .a/
xa
D cf 0 .a/.
g0 .a/ D lim
g.x/g.a/
.
xa
x!a
138
5 Derivatives
.fg/.x/.fg/.a/
xa
x!a
f .x/g.x/f .x/g.a/Cf .x/g.a/f .a/g.a/
lim
xa
x!a
lim f .x/
x!a
D lim
x!a
f .x/g.x/f .a/g.a/
xa
D lim f .x/
x!a
.a/
lim g.x/g.a/
C lim f .x/f
xa
xa
x!a
x!a
0
0
0
g.x/g.a/
xa
f .x/f .a/
xa
g.a/ D
gf 0 fg0
g2
139
Suppose that f and g are functions defined on a common domain, and that f
and g are both differentiable at a with g.x/ 0.
.a/
From the definition of derivative f 0 .a/
D
lim f .x/f
and
xa
x!a
.
g0 .a/ D lim g.x/g.a/
xa
x!a
Because g is differentiable at a, it is continuous at a. This implies that
lim g.x/ D g.a/.
x!a
f
f
0
f .x/
f .a/
.x/
.a/
g.a/
g
g
f
g.x/
Then
D lim
D
.a/ D lim
x!a
x!a
g
xa
xa
f .x/
f .a/
f .a/
f .a/
g.x/
C g.x/
g.a/
g.x/
lim
D
x!a
xa
!
1
1
g.a/
1
f .x/ f .a/
g.x/
C f .a/
lim
D
x!a
xa
g.x/
xa
1
f .a/
g.a/ g.x/
f .x/ f .a/
C
D
lim
x!a
xa
g.x/
g.x/g.a/
xa
lim
x!a
f .x/ f .a/
1
f .a/
g.a/ g.x/
lim
C lim
lim
D
x!a g.x/
x!a g.x/g.a/ x!a
xa
xa
f .a/
1
f 0 .a/g.a/ f .a/g0 .a/
0
.
2 g .a/ D
2
g.a/
g.a/
g.a/
0
f
f 0 .a/g.a/ f .a/g0 .a/
Thus,
.a/ D
.
2
g
g.a/
f 0 .a/
5.4.1 Exercises
Write proofs for each of the following statements.
For any constant c, the function f .x/ D c has derivative f 0 .x/ D 0.
The function f .x/ D x has derivative f 0 .x/ D 1.
For any positive integer n, the function f .x/ D xn has derivative f 0 .x/ D nxn1 .
Any polynomial function f .x/ D an xn Can1 xn1 Can2 xn2 C Ca1 xCa0 has
derivative f 0 .x/ D nan xn1 C .n 1/an1 xn2 C .n 2/an2 xn3 C C a1 .
n
5. For any positive integer n, the function f .x/ D x1n has derivative f 0 .x/ D xnC1
.
1.
2.
3.
4.
140
5 Derivatives
.
you would need to find the limit of the difference quotient D D
xa
0
which
has
limit
g
.a/,
you
might
try
both
Expecting to see the expression g.x/g.a/
xa
multiplying
and dividing the difference quotient D by the factor g.x/ g.a/ to get
f g.x/ f g.a/
g.x/g.a/
g.x/g.a/
.
xa
g.x/g.a/
xa
x!a
lim
D f 0 .g.a// g0 .a/.
x!a
y!g.a/
141
This proof attempt does include the intuitive reasoning behind why the Chain Rule
works, but the proof is not correct. Can you spot the error? The problem is that even
though g.x/ is approaching g.a/ as x approaches a, there is no guarantee where g.x/
is different from g.a/. In fact, it is quite easy to construct functions g.x/ which are
differentiable at a for which g.x/ is equal to g.a/ for infinitely many values of x
as x approaches a. The simplest example
is when g is a constant function. A more
2
x sin 1x if x 0
complicated example is g.x/ D
which has a derivative of
0
if x D 0
1
0 at x D 0 and is equal to g.0/ D 0 at n
for all nonzero integers n. Clearly,
when g.x/ D g.a/, one cannot both multiply and divide the difference quotient
by g.x/ g.a/ and expect to get anything except nonsense. This problem does not
present anenormous
hurdle because, in the cases where g.x/ D g.a/, the difference
f g.x/ f g.a/
xa
>
: 0
;
f g.a/ if y D g.a/
nice property that it is equal to the desired difference quotient when g.x/ differs
from g.a/, and it is continuous at g.a/. Introducing this function into the proof gets
around the technical difficulties of the previously attempted proof.
quotient
=
< yg.a/ if y g.a/ >
, and note that h is continuous
Define h.y/ D
>
;
: 0
f g.a/ if y D g.a/
at g.a/.
f g.x/ .f g/.a/
g.a//
D lim
Then .f g/0 .a/ D lim .f g/.x/.f
xa
xa
x!a
x!a
g.x/g.a/
g.x/g.a/
D lim h g.x/ lim xa D
lim h g.x/ xa
x!a
x!a
x!a
0
0
g.a/
g
D
f
.a/.
h lim g.x/ lim g.x/g.a/
xa
x!a
x!a
0
0
0
Therefore, .f g/ .a/ D f g.a/ g .a/.
142
5 Derivatives
To obtain
domain of sin x is restricted to the interval 2 ; 2 . On this interval sin x is both
injective and surjective and has the inverse sin1 y, sometimes written as arcsin y
(Fig. 5.3).
Now suppose that f is bijective and has inverse function f 1 . If f has a nonzero
derivative at the point a, the Chain Rule can be used to find the derivative of f 1
at f .a/. Indeed,
one has that .f f 1 /.x/ D x, so the Chain Rule implies that
1
0 1
1 0
. Is this conclusion valid? That
f f .x/ .f / .x/ D 1, or .f 1 /0 .x/ D 0 1
f
.x/
is, can you justify taking the derivative of .f f 1 / using the Chain Rule before you
know that the derivative of f 1 exists? The answer is yes, the use of the Chain
Rule
the limit of
is justified here. The proof of the Chain Rule includes taking
which
is
broken
into
the
product
of
the
limit
of
h
g.x/ and the
h g.x/ g.x/g.a/
xa
143
limit of
g.x/g.a/
.
xa
h g.x/
g.x/g.a/
xa
xa
exists; its limit is just the limit of the quotient xa which
h g.x/
144
5 Derivatives
.a/
0
of f and 0 < jx aj < , then f .x/f
f
.a/
< f 0 .a/ implying that
xa
f .x/f .a/
xa
> 0.
.a/
> 0, so f .x/f .a/ >
If x satisfies a < x < aC, then xa > 0 and f .x/f
xa
0, and f .x/ > f .a/.
.a/
If x satisfies a > x > a, then xa < 0 and f .x/f
> 0, so f .x/f .a/ <
xa
0, and f .x/ < f .a/.
This proves the first part of the theorem.
If instead f 0 .a/ < 0, apply the above argument to the function f to obtain
the analogous result.
145
f(x)
Fig. 5.4 This graph of f .x/ on the interval a; h shows relative maxima at b, d, and g, relative
minima at a; c; e; and h; an absolute maximum at b, and an absolute minimum at h. The derivative
f 0 .x/ does not exist at x D d
Any student of Calculus will see applications of this result where one is asked
to identify relative extrema for a particular function, and applications to what are
fondly called Max/Min problems where one is first asked to construct an appropriate
function to fit the application and then find a particular extremum of that function.
One defines a critical point of f to be a value a where either f 0 .a/ D 0 or f 0 .a/ does
not exist. Not all of these points will end up being relative extrema for some may just
be a saddle point of f where f 0 .a/ D 0, but f has no relative extrema at that point.
For example, the function f .x/ D x3 has a saddle point at x D 0 where f 0 is 0, but f
is a strictly increasing function over the entire real line. A function is said to have an
absolute maximum (sometimes called a global maximum) at a if f is defined at
a, and for all other x in the domain of f , f .x/ f .a/. The term absolute minimum
(sometimes called a global minimum) is defined in the analogous way with f .x/
f .a/, and an absolute extremum (sometimes called a global extremum) is either an
absolute maximum or absolute minimum. The theorem about relative extrema shows
that if f is defined on any interval I, then the only places f can have relative extrema
or absolute extrema are critical points or at endpoints of I. You should be able to
identify example functions where each of these criteria give extrema (Fig. 5.4).
5.6.1 Exercises
Identify the relative extrema and absolute extrema of the given functions on the
given intervals.
1.
2.
3.
4.
146
5 Derivatives
147
f(x)
Fig. 5.5 The proof of Rolles Theorem finds an extreme point c between a and b for which
f 0 .c/ D 0
148
Fig. 5.6 Point c between a
and b where the tangent line
is parallel to the secant line
from a to b
5 Derivatives
f(x)
149
150
5 Derivatives
5.7.1 Exercises
Write proofs for each of the following statements.
1. If f is a function whose derivative is negative for all points in an interval, then f
is a decreasing function on the interval.
2. If f and g are functions differentiable on an interval with f 0 .x/ D g0 .x/ for each
x in the interval, then there is a constant C such that f .x/ D g.x/ C C for all x in
the interval.
3. If f .0/ D g.0/ and f 0 .x/ g0 .x/ for each x 0, then f .x/ g.x/ for each x > 0.
x!a
f .x/
0, then lim gf 0.x/
D L implies that lim g.x/
D L. This is very useful because the
x!a .x/
x!a
theorem stating that the limit of a quotient is the quotient of the limits does not
apply in cases when the denominator has a limit of 0.
How would you prove LHopitals Rule? You might try to prove it by using the
Mean Value Theorem because the quotient you are considering is
f .x/ f .a/
f .x/
D
D
g.x/
g.x/ g.a/
f .x/f .a/
xa
g.x/g.a/
xa
151
This is not exactly correct because, as far as you know, f .a/ and g.a/ might not
even be defined, and if they are, they need not be equal to lim f .x/ and lim g.x/.
x!a
x!a
This is not a big stumbling block, because you can always redefine f and g at a to
be equal to 0 without changing the result of the theorem. You also would need to
know that g.x/ g.a/ for x near a so that the needed quotient can be calculated.
Once the quotient of f and g is rewritten as the quotient of the difference quotient
of f and the difference quotient of g, you can apply the Mean Value Theorem to
replace the difference quotients with derivatives, and then take the limit. It might
look something like the following.
PROOF ATTEMPT: Let both f and g be functions differentiable for
all x a in an interval which contains a. Assume that lim f .x/ D
x!a
f 0 .x/
0
x!a g .x/
lim g.x/ D 0, and g0 .x/ 0 for all x a in the interval. Then lim
x!a
implies
lim f .x/
x!a g.x/
DL
D L.
with x a.
Assume that lim
f 0 .x/
0
x!a g .x/
x!a
D L.
lim f .x/
x!a g.x/
x!a
D lim
x!a
f .x/f .a/
xa
g.x/g.a/
xa
x!a
and cg .x/ D x2 ? Even though it is true that cf .x/ and cg .x/ both approach 0 as
152
5 Derivatives
cf .x/2
x!0 cg .x/
approaching a does not allow you to use both of these expressions in place of x
f .x/
x!0 g.x/
when taking the limit. What the proof attempt does show is that lim
lim f 0 .x/
x!0
lim g0 .x/
x!0
153
Now the Extended Mean Value Theorem can be used to give a correct proof of
LHopitals Rule.
PROOF (LHopitals Rule, Part 1): Let f and g be functions differentiable
for all x a in an open interval which contains a. Assume that lim f .x/ D
x!a
f 0 .x/
0
x!a g .x/
lim g.x/ D 0, and g0 .x/ 0 for all x a in the interval. Then lim
x!a
implies
lim f .x/
x!a g.x/
DL
D L.
x!a
with x a.
Assume that lim
f 0 .x/
0
x!a g .x/
D L.
x!a
is within of L.
Fix x in the given interval with 0 < jx aj < .
Since f and g are continuous on the closed interval from a to x and
differentiable on the open interval from a to x, f and g satisfy the hypothesis
of the Extended Mean Value Theorem on the interval from a to x.
It follows that there is a c between x and a such that f 0 .c/g.x/ g.a/ D
g0 .c/f .x/ f .a/.
By assumption g0 is not 0, so g0 .c/ 0. Also the Mean Value Theorem
shows that g.x/ g.a/ D g0 .t/.x a/ for some t between x and a, and this
shows g.x/ g.a/ 0.
0
f .x/f .a/
It follows that g.x/g.a/
D gf 0.c/
.
.c/
f .x/f .a/
0
f .x/
L
Thus, g.x/ L D g.x/g.a/ L D gf 0.c/
< .
.c/
f .x/
x!a g.x/
LHopitals Rule also holds in cases where lim g.x/ is infinite rather than
x!a
zero.
154
5 Derivatives
positive or negative infinity, and g0 .x/ 0 for all x in the interval with
0
f .x/
x a. Then lim gf 0 .x/
D L implies lim g.x/
D L.
.x/
x!a
x!a
is within 2 of L.
Fix x in the given interval with 0 < jx aj < 0 .
Since f and g are differentiable between x and a and continuous at x, for
any y between x and a it follows from the Extended Mean Value Theorem
0
f .y/f .x/
f .y/f .x/
that there is a c between x and y such that g.y/g.x/
D gf 0.c/
. Thus, g.y/g.x/
is
.c/
within 2 of L.
f .y/
f .x/
g.y/
f .y/ f .x/
g.y/
D
.
Note that
g.y/ g.x/
1 g.x/
g.y/
f .y/ f .x/
g.y/ g.y/
f .y/f .x/
f 0 .c/
Because g.y/g.x/ D g0 .c/ is within 2 of L, it follows that
L
< .
1 g.x/
2
g.y/
f .y/ f .x/
< 2 1 g.x/
Then g.y/
g.y/ L 1 g.x/
g.y/
g.y/ .
Because g.y/ approaches positive or negative infinity as y approaches a,
there is a > 0 with < 0 such that for all y with 0 < jy aj < , the
fraction jf .x/jCjLg.x/jCjg.x/j
< 2 .
jg.y/j
f .y/ f .x/
<
g.y/ L 1 g.x/
Then for y with 0 < jy aj < , g.y/
g.y/
g.x/
1 g.y/ implies
2
f .y/ f .x/
g.x/
jg.x/j
jf .x/jCjLg.x/jCjg.x/j
< 2 C 2 D .
g.y/ L < g.y/ L g.y/ C 2 C 2jg.y/j < 2 C
jg.y/j
155
5.8.1 Exercises
Use LHopitals Rule to calculate the following limits.
sin2 .2x3 /
6
x!0 px
xx
lim pxCx
x!0C p
1. lim
2.
3. lim
x!0C
4. lim
x!1
x ln x
ln x
p
x
5. lim .sin.2x//x
x!0C
tan1 x
1
x!0 tan .3x/
6. lim
x!1
f .x/
x!1 g.x/
D L implies lim
x!1
D L.
8. If f and g are differentiable functions for all x > 0, lim g.x/ D 1, and g0 .x/ > 0
for all x > 0, then
0
lim f 0.x/
x!1 g .x/
D L implies
lim f .x/
x!1 g.x/
x!1
D L.
9. If f and g are functions differentiable for all x > a, lim f .x/ D lim g.x/ D 0,
and g0 .x/ 0 for all x with x > a, then lim
x!aC
f 0 .x/
g0 .x/
x!aC
x!aC
f .x/
g.x/
x!aC
D L implies lim
D L.
156
5 Derivatives
1
x2
0 and x. Moreover, it obtains each of those values infinitely often. In fact, between
0 and x, the function f 0 takes on every real number infinitely often (Fig. 5.7).
Note that the function f .x/ C x has a derivative of 1 at x D 0. This is an example
of a function with a positive derivative at 0 which is not an increasing function over
any open interval containing 0. This can easily be seen by the fact that in every open
interval containing 0 there are intervals where the derivative of f .x/ C x is negative.
So, how can you prove that if a function f has a derivative f 0 on an interval I,
that f 0 has the intermediate value property on I? The hypothesis suggests that you
start by taking a function f differentiable on an interval I and values a; b 2 I. Then
you select a value K between f 0 .a/ and f 0 .b/. Without loss of generality, you can
assume that a < b and f 0 .a/ < K < f 0 .b/. The goal would be to show that there
is a c between a and b such that f 0 .c/ D K. One simplification is to replace f with
the function g.x/ D f .x/ Kx. This function is also differentiable on I, and if
f 0 .c/ D K, then g0 .c/ D 0. Which theorems about derivatives allow you to conclude
that a derivative is 0 at some point in an interval? First there is a theorem that states
that if a differentiable function reaches an extreme value at a point in an interval,
then the point is either a critical point of the function or an endpoint of the interval.
A second theorem is Rolles Theorem which talks about a differentiable function
which takes on the same value at the endpoints a and b. Since you do not have any
information about the values of g at the endpoints of the interval, the theorem about
extreme values may be the more promising choice for this proof.
What is known about the function g? You know that g is differentiable at each
point of the interval from a to b. Additionally, g0 .a/ D f 0 .a/ K < 0 and g0 .b/ D
f 0 .b/ K > 0. Does this mean that the function g is decreasing at a and increasing
at b? Well, it would if you knew that g0 were continuous because then g0 would be
negative in an interval around a and positive in an interval around b. But, as you now
know, g0 need not be continuous. On the other hand, there is a theorem that says that
if g0 .a/ is negative, then there is a > 0 such that if x satisfies a < x < a C ,
then g.x/ < g.a/. This does not show much, but you can use it to conclude that g
does not take on its minimum value on a; b at a. A similar argument uses the fact
that g0 .b/ > 0 to show that g does not take on its minimum value on a; b at b.
157
There are simple examples of functions that have discontinuous derivatives that
do not have the intermediate value property; functions such as f .x/ D jxj. This
functions derivative is the constant 1 for all x > 0 and 1 for all x < 0. This
derivative is not continuous at x D 0 because it is not defined there. Clearly, f 0 does
not have the intermediate value property on any interval containing both positive
and negative numbers, but then f does not satisfy the hypothesis of the previous
theorem on any such interval because f 0 .0/ is not defined. Functions that have
discontinuous derivatives that are defined at all points will have to exhibit wild
oscillations
of those discontinuities similar to the example
in the neighborhoods
2
x sin x12 if x 0
.
0
if x D 0
Suppose f is a function whose derivative is defined at all points of an interval
except perhaps at some point c in the interval. What can be said if lim f 0 .x/ exists?
x!c
Such a derivative does not exhibit wild oscillations near c, and, in fact, it must
have a continuous derivative at c. The proof is a consequence of the Mean Value
Theorem.
158
5 Derivatives
From the definition of limit, given > 0, there is a > 0 such that if
y 2 .a; b/ with 0 < jy cj < , then jf 0 .y/ Lj < .
Let x 2 .a; b/ with 0 < jx cj < .
By the Mean Value Theorem there is a y between x and c such that
f .x/f .c/
D f 0 .y/.
xc
.c/
Then y 2 .a; b/ with 0 < jy cj < , so f .x/f
L D jf 0 .y/ Lj < .
xc
f .x/f .c/
D L, so f 0 .c/ D L.
xc
f 0 .c/ D lim f 0 .x/, it follows that f 0 .x/
x!c
Thus, lim
x!c
Because
This completes the proof.
is continuous at c.
Chapter 6
Riemann Integrals
6.1 Area
The first application one usually sees of the Riemann Integral is that of finding
the area of a region in the plane bounded by the graph of a function and the
lines x D a, x D b, and the x-axis. Thus, before discussing integration, it makes
sense to review what is meant by the area of a region in the plane. Clearly, the
measure of area should be a way to assign a size to a region in a way that is
compatible with the well-established rules from Geometry for assigning areas to
regions such as rectangles, triangles, and circles. But there is a need to go beyond
these simple regions so that area can be calculated for far more complicated regions.
For example, consider the region in the coordinate plane f.x; y/ j 0 x 1; 0
y 1; at least one of x or y is rationalg. Regions such as these are not typically
considered in a Geometry course, but being able to calculate areas for such sets
is important in the more general discussion of integration. This chapter, therefore,
begins by considering two different measures of the sizes of sets which will aid the
understanding of integration.
159
160
6 Riemann Integrals
to the elements of the other set. One says that two sets A and B have the same
cardinality if there is a bijection f W A ! B. The bijection demonstrates a oneto-one correspondence between the elements of set A and the elements of set B, so
one concludes that A and B are the same size. Some sets are finite, meaning that the
set is either empty (has cardinality 0) or, for some positive integer n, is in one-toone correspondence with the set f1; 2; 3; ; ng. A set is called denumerable if it
can be put in one-to-one correspondence with the set of positive integers. Thus, the
set of positive multiples of 2 is denumerable. So is the set of all integers since
the positive
can be mapped
onto the set of all integers using the bijection
integers
x
if
x
is
even
2
f .x/ D
. The verification that this map is a bijection is left
if x is odd
1 xC1
2
as an exercises. It shows that the integers and the positive integers have the same
cardinality. Sets that are either finite or denumerable are called countable because
they can be counted out by listing a first, second, third, and so forth. Thus, a good
way to think about a countable set is a set whose elements can be written down in
a finite or infinite sequence x1 ; x2 ; x3 ; because this listing shows the one-to-one
correspondence between the set and the natural numbers or one of its finite subsets.
The union of two countable sets is also countable. This can be seen by representing one set by the sequence x1 ; x2 ; x3 ; and the other by y1 ; y2 ; y3 ; . Then
the elements of the union of the two sets can be written as x1 ; y1 ; x2 ; y2 ; x3 ; y3 ; .
If there are elements that belong to both sets, then one can just leave the second
copies of those elements out of the listing. Clearly, this can be extended to the
union of any finite collection of countable sets, so the union of a finite number
of countable sets is countable. What might seem surprising is that the union of a
countable number of countable sets is still countable. That is, if A1 ; A2 ; A3 ; is
1
a sequence of countable sets, then the union [ Ak is also countable. To see this,
kD1
suppose that the elements in each Ak can be listed in a sequence ak;1 ; ak;2 ; ak;3 ; .
1
One can now list all the elements of [ Ak by listing the ak;j elements in increasing
kD1
a1;1 ; a2;1 ; a1;2 ; a3;1 ; a2;2 ; a1;3 ; a4;1 ; a3;2 ; a2;3 ; a1;4 ; a5;1 ; .
order of k C j resulting in
As above, duplicate elements occurring because they belong to more than one set
can be left out of this listing. Figure 6.1 shows the order that the elements enter
the list.
Note that this result can be used to show that the set of rational numbers is
countable. Indeed, the rational numbers can be written as the union R1 [R2 [R3 [
where Rk are the rational numbers that can be written as a fraction with an integer
in the numerator and the positive integer k in the denominator. For example,
R2 D f 02 ; 12 ; 12 ; 22 ; 22 ; 32 ; 32 ; g. Thus, the rational numbers is a countable union
of countable sets showing that it is countable. The cardinality of a denumerable set is
often written using the symbol @0 (read Aleph knot or Aleph null). The symbol
represents the size of the natural numbers and the size of any set that can be placed
in one-to-one correspondence with the natural numbers.
A set which is not a countable set is called uncountable. There is a standard
argument that shows that the set of real numbers in the interval .0; 1/ is not a
countable set. The method, known as a diagonalization argument, first assumes
161
a1,1
a1,2
a1,3
a1,4
a1,5
a1,6
a2,1
a2,2
a2,3
a2,4
a2,5
a2,6
a3,1
a3,2
a3,3
a3,4
a3,5
a3,6
a4,1
a4,2
a4,3
a4,4
a4,5
a4,6
a5,1
a5,2
a5,3
a5,4
a5,5
a5,6
x1 = 0. 4 9 0 3 2 5 5 9 0 9 9 0
x2 = 0. 1 7 7 3 8 8 0 0 0 0 0 0
x3 = 0. 7 4 1 1 8 9 1 8 2 5 4 4
x4 = 0. 1 1 8 8 8 3 7 2 9 0 0 1
x5 = 0. 5 5 2 7 7 7 1 0 6 4 2 3
x6 = 0. 0 0 0 0 0 2 1 0 9 3 7 3
x7 = 0. 8 2 1 7 4 9 0 3 2 8 5 5
x8 =
y = 0. 7 3 7 7 3 7 7 7 7 7 7 3
that the real numbers between 0 and 1 can all be written down in a sequence
x1 ; x2 ; x3 ; x4 ; . Then one constructs a real number y between 0 and 1 where the
kth digit to the right of the decimal point in y is chosen as follows. If the kth digit
to the right of the decimal point of xk is 7, then let the kth digit to the right of the
decimal point in y be 3. Otherwise, if the kth digit to the right of the decimal point of
xk is not 7, then let the kth digit to the right of the decimal point in y be 7. Figure 6.2
illustrates the process of determining y.
The point of this construction is that the number y is a real number in the interval
.0; 1/, but it cannot be one of the numbers in the sequence x1 ; x2 ; x3 ; x4 ; . This
is because for each k, y cannot equal xk because y and xk differ in their kth digits.
This is a contradiction to the assumption that the sequence contained all of the real
numbers in .0; 1/ and shows that it is impossible to list all the elements of .0; 1/ in
a sequence. Thus, this interval is an uncountable set. If there is a bijection from the
set .0; 1/ to a set B, then it follows that B will also be uncountable. You may wonder
whether all uncountable sets have the same cardinality. They do not, but that fact
will not be needed for the proofs discussed in this book. Refer to a standard text in
Set Theory for a far more in-depth look at the cardinality of sets.
162
6 Riemann Integrals
6.2.1 Exercises
1. Determine whether each of the following sets is finite, denumberable, or
uncountable.
(a) the set of points in the coordinate plane where both x and y coordinates are
rational numbers
(b) the set of points in the coordinate plane where at least one of its x and y
coordinates is a rational number
(c) the set of polynomials p.x/ with integer coefficients
(d) the set of real numbers whose decimal representation does not contain the
digit 5
(e) the set of functions f W f0; 1; 2; 3; 4; 5g ! f1; 2; 3; : : : ; 100g
(f) the set of functions f W f2; 4; 6; 8; 10; : : : g ! f0; 1g
x
if x is even
2
2. Show that the function
is a bijection from the set of
if x is odd
1 xC1
2
natural numbers to the set of integers.
3. Show that .0; 1/ and .1; 5/ have the same cardinality.
4. Show that .0; 1/ and the entire set of real numbers, R, have the same cardinality.
5. Show that .0; 1/ and the interval 0; 1 have the same cardinality. (Hint: Find a
way to bury the endpoints of 0; 1 inside of .0; 1/ by mapping a sequence
x1 ; x2 ; x3 ; : : : to x3 ; x4 ; x5 ; : : : .)
1
least one n, the set An must be uncountable. This can be thought of as an infinite
form of the Pigeonhole Principle.
7. The interval 0; 1 on the real line and the unit square in the plane have the same
cardinality. (Hint: for a point in 0; 1 split up its decimal digits between the x
and y coordinates of a point in the unit square.)
8. Show that the equality of cardinality is an equivalence relation. That is, if A, B,
and C are any sets, then
A has the same cardinality as A.
If A has the same cardinality as B, then B has the same cardinality as A.
If A has the same cardinality as B, and B has the same cardinality as C, then A
has the same cardinality as C.
9. Suppose that you apply the diagonalization argument to the set of rational
numbers in the interval .0; 1/. That is, suppose you list all of the rational numbers
in a sequence x1 ; x2 ; x3 ; : : : and use the diagonalization argument to construct a
number y in .0; 1/ that differs from each element of the sequence. Why is this
not a proof that the rational numbers are uncountable?
163
the intervals S [ .aj ; bj / and the total length of the intervals is less than , that
jD1
n
P
jD1
zero if you can cover it with a sequence of intervals whose total length is as small
as you want.
In particular, any finite set consisting of n real numbers has measure zero because
for any > 0, each point x in the set can be covered by the interval .x 3n
; x C 3n
/,
2
and the total length of these intervals is 3 . Similarly, any countable set of real
numbers fx1 ; x2 ; x3 ; : : : g can be covered by intervals .xj 32
j ; xj C 32j /, and the total
1
P 2
D 23 . Thus, the set of rational numbers, which
length of these intervals is
32j
jD1
a sequence of sets all of which have measure zero, then the union [ Aj also has
jD1
measure zero. Indeed, given > 0, for each j you can cover Aj with a sequence
of open intervals whose total length is less than 2j . Then the sequences of open
1
intervals can be combined into one sequence of intervals which cover [ Aj and has
total length less than
1
P
jD1
jD1
2j
D .
Since any countable set of real numbers has measure zero, if a set does not
have measure zero, it must be an uncountable set. A natural question is whether
an uncountable set of real numbers can have measure zero. The answer to this
question is yes. The most famous example of this is known as the Cantor set which
is constructed as follows. The construction begins with the closed unit interval
C0 D 0; 1. At the first stage, the open interval of length 13 is removed from the
middle of this set leaving two intervals each with length 13 so that C1 D 0; 13 [ 23 ; 1.
164
6 Riemann Integrals
Stage 0
Stage 1
Stage 2
Stage 3
Stage 4
Stage 5
At the second stage, open intervals of length 19 are removed from the middle of each
of the two remaining intervals leaving four intervals each with length 19 so that
C2 D 0; 19 [ 29 ; 39 [ 69 ; 79 [ 89 ; 99 . This process is repeated so that at stage n,
open intervals of length 31n are removed from each of 2n1 closed intervals of length
1
leaving 2n closed intervals each with length 31n (Fig. 6.3). The Cantor set C
3n1
1
The Cantor set is sometimes called the Cantor middle thirds set, because, at each
stage, the middle thirds of the remaining intervals are removed. Other similar types
of Cantor-like sets can be constructed by removing other portions of each interval.
It is clear that the Cantor set has measure zero because it is contained in Cn which
is made up of 2n closed intervals each with length 31n . The total length of the closed
n
intervals in Cn is 23n , a quantity that goes to 0 as n gets large. Cn can be covered by
n
a finite collection of open intervals whose total length is 10 percent larger than 23n
showing that the Cantor set can be covered by open intervals whose total length is
as small as you want. So how do you show that the Cantor set is uncountable? To
see this, consider writing each number in the unit interval 0; 1 in base three. The
numbers in the interval 0; 13 are the numbers between 0 and 1 whose base-three
representation begins with 0.0, and the numbers in the interval 23 ; 1 are the numbers
between 0 and 1 whose base-three representation begins with 0.2. The numbers in
the middle third of the interval that are removed at the first stage of the construction
process are the numbers between 0 and 1 whose base-three representation begins
with 0.1. Note that numbers at the endpoints of the removed interval, 13 and 23 each
has two different representations. Indeed, in base three 13 D 0:1 D 0:0222 and
2
D 0:2 D 0:1222 . One could say that C1 consists of all the numbers between 0
3
and 1 that can be represented in base three without a 1 in the first place to the right
of the decimal point, the one-third place. Similarly, C2 are the numbers between 0
and 1 that have a base-three representation with no 1 in either of the first two places
to the right of the decimal point. The Cantor set C is the set of numbers between 0
165
and 1 that have a base-three representation that contains no digit equal to 1. Then
consider the map that takes each element of the Cantor set and divides it by 2. This
is an injective map that maps the numbers in the Cantor set to the set of numbers in
the unit interval that have base-three representations that include only the digits of
0 and 1 because it takes numbers with representations that only included the digits
of 0 and 2 and divides each of the digits by 2. Now, the numbers between 0 and
1 with base-three representations that include only the digits of 0 and 1 are clearly
in one-to-one correspondence with base-two representations of numbers in between
0 and 1. But all the real numbers between 0 and 1 have base-two representations
containing only 0 and 1, so the numbers in the Cantor set are as numerous as the
real numbers between 0 and 1. Thus, the Cantor set must be uncountable since the
set of real numbers between 0 and 1 is an uncountable set.
The concept of measure zero can be extended to sets in the plane, although here,
rather than being interested in the length of a set, the interest is in the area of the set.
Thus, rather than trying to cover a set with intervals whose total length is small, in
the plane one would try to cover a set with a sequence of squares whose total area
is small. Just as on the real line, it was taken as given that the length of an interval
a; b was b a, in the plane it will be taken as given that the area of a square with
side length x is x2 . Then, a region in the plane is said to have measure zero (or area
zero) if for each > 0, the set is contained in the union of a sequence of squares
whose total area is less than .
As it was with sets of real numbers, any countable set of points in the plane has
area zero because, for any < 0, you can cover the sequence x1 ; x2 ; x3 ; : : : with a
sequence of squares with total area less than . Moreover, let Y be a line segment
with length y > 0. Then Y has area zero. How would you prove this? Certainly, this
line segment is contained in a square with side length y which has area y2 , so the
squares area could be rather large and, in particular, the area of the square is not
zero. Notice, though, that Y can also be covered by two side-by-side squares each
2
with side length 2y and each with area y4 giving a total area of the two squares equal
2
to y2 . This is the key to covering Y with squares with very small total area. If Y is
covered by a sequence of n adjacent squares each with side length ny , then the total
2
Fig. 6.4 Covering a line segment with smaller and smaller squares
166
6 Riemann Integrals
6.3.1 Exercises
1. Rather than constructing the Cantor set only on the interval 0; 1, perform the
same construction on each interval n; n C 1 for every integer n. Show that the
resulting set has measure zero.
2. Beginning with the interval 0; 1 construct a Cantor-like set, but instead of
removing intervals of length 13 at stage 1, 19 at stage 2, and so forth removing
intervals of length 31n at stage n, you remove an interval of length 14 at stage 1,
1
at stage 2, and so forth removing intervals of length 41n at
intervals of length 16
stage n. Show that the total lengths of the intervals remaining after stage n does
not approach zero as n approaches infinity.
3. Which of the following sets of real numbers have measure zero?
(a)
(b)
(c)
(d)
the integers
the irrational
numbers
p
p
fa C b 2 C c 3 j a; b; c are integers g
the Cantor-like set where instead of removing the middle 13 of each
remaining interval at stage n, you remove the middle 14 of each remaining
interval
4. Show that if the set A has measure zero and B A, then the set B has measure
zero.
5. Show that a line in the plane has area zero.
6. Show that the set in the plane f.x; y/ j x is rationalg has area zero.
7. Suppose that the set A R has measure zero. Show that the set f.x; y/ j x 2
A; y 2 0; 1g has area zero.
8. Suppose that the set A R has measure zero. Show that the set f.x; y/ j x 2 Ag
has area zero.
9. Show that f.x; y/ j 0 x 1; 0 y 1; at least one of x or y is rationalg has
area zero.
10. Show that the interval 0; 1 does not have measure zero. (Hint: Use the Heine
Borel Theorem to reduce any cover to a finite subcover.)
167
set A is contained in the union of the squares [ Sk , and for every natural
kD1
P
number n, nkD1 sk < . Then A has area 0.
5. (Union) If set A has area a, set B has area b, and their intersection A \ B has
area 0, then the union A [ B has area a C b.
6. (Exhaustion) Let B be a set. If for each > 0 there are sets A and C with
A B C such that the area of A is greater than b , and the area of C
is less than b C , then the area of B is b.
Axioms 1, 2, and 3 should agree with what you know about area from Geometry,
and they can be used to prove some simple results. For example, since a 1 1 square
has area 1, Axiom 3 can be used to show that an s s square has area s2 .
The result from the previous section that a line segment has area 0 is particularly
useful because of the way it can be used in conjunction with the Union area axiom.
In particular, suppose A and B are two squares or other polygons set side-by-side so
that they only share an edge. Because the shared edge is a line segment, it has 0 for
its area, and the Union area axiom shows that A [ B has an area equal to the sum of
the area of A and the area of B. By using mathematical induction, this result can be
extended to the union of many polygons that share borders. In particular, consider
finding the area of a rectangle with width x and length y. If xy is a rational number
equal to pq , where p and q are positive integers, then the x y rectangle is the union
of p q squares all with side length px . Indeed, the width of the rectangle which has
length x is spanned by p such squares, and the length of the rectangle which has
length y is spanned by q such squares showing that the entire rectangle can be tiled
2
by a p q array of squares, each with area px . The Union axiom then shows that
2
the area of the x y rectangle is p q px D x qp x D x y. It will require
the last of the area axioms to conclude that the area of any rectangle is equal to its
length times its width even when the length of the rectangle is an irrational multiple
of its width.
The last area axiom is essentially the Method of Exhaustion used some by
Euclid and much more extensively by Archimedes to calculate areas and volumes.
It is an example of a use of Calculus about 1800 years before the foundation of
Calculus was formally established by Newton and Leibniz. This axiom says that if
a region in the plane can be closely approximated by sets whose areas you know,
then you can figure out the area of the region. Take, for example, a rectangle B with
168
6 Riemann Integrals
width x > 0 and length y > 0 where the ratio xy D is irrational. It is certainly
possible to find other rectangles close to the size of B whose length to width ratios
are rational. To prove that B has area xy, the axiom requires that for each > 0 you
find a subset A B whose area is greater than xy and a set C containing B whose
area is less than xy C . Suppose you choose A to be a rectangle with width x and
length just a bit short of y, say rx, where r is a rational number chosen to be less
than but suitably close to xy . How close is suitably close? Well, you would need the
area of A, which is x rx D rx2 , to be within of xy, that is, xy rx2 < . Solving
for r shows that r > xy x2 . Is there such an r which is rational and between xy x2
and xy ? Of course there is. The rational numbers are dense in the real line; there are
rational numbers in every interval of positive length. Thus, you can select a rational
number r between xy x2 and xy and let A be an x rx rectangle. Then A can be
placed inside of B, and the area of A is within of xy. Similarly, you can choose a
rectangle C with width x and length sx, where s is a rational number chosen to be
greater than but suitably close to xy . You need the area of C to be within of xy, so
choose s so that x sx xy < . This happens if xy < s < xy C x2 . Since you have
found a rectangle A contained inside B and a rectangle C containing B with the areas
of A and C within of xy, the Exhaustion area axiom shows that B has area xy.
The familiar formula for the area of a triangle given as one half the base times the
height can be derived geometrically, but to prove this formula using the area axioms
requires more work. To begin, consider a right triangle with legs with lengths x and
y. Place this triangle in a rectangle with side lengths x and y. For any natural number
n, the rectangle can be overlaid with an n n grid of rectangles with side lengths
x
and ny . The hypotenuse of the triangle is the diagonal of the x y rectangle and
n
spans the diagonals of n of the smaller rectangles as shown in Fig. 6.5 exhibiting the
case where n D 8.
Because there are n grid rectangles along the hypotenuse of the triangle, it
2
must be that there are n 2n grid rectangles inside the triangle with a total area
2
of n 2n nxy2 D 1 1n xy2 . Similarly, the triangle is enclosed inside the union of
n2 Cn
grid rectangles with a total area of 1 C 1n xy2 . Clearly, n can be chosen large
2
Fig. 6.5 An 8 8 grid of
rectangles overlaying a
triangle
169
enough to make both the total area of grid rectangles inside the triangle and the total
area of grid rectangles enclosing the triangle within a particular > 0 of xy2 . Thus,
the Exhaustion axiom shows that the area of the triangle is xy2 as expected. Since any
triangle can be partitioned into two right triangles, the well-known area formula for
the area of a triangle follows. Since any polygon can be partitioned into triangles,
the usual formulas from Geometry for the areas of polygons can be derived in the
same way they would be in Geometry.
You may wonder whether these techniques can be used to find the area of any
region in the plane, or at least any bounded region in the plane. This is a really good
question with a very complicated answer. The Area Axioms listed in this section
are designed to give the reader a feel for proofs about areas that will be useful
in the upcoming discussion of proofs about Riemann integrals. The axiom list is
not complete enough to allow the calculation of the area of many of the sets that
one might encounter. The area of Analysis known as Measure Theory provides a
somewhat richer environment for this study, but the complexities of measure theory
go beyond the aim of this text. What can be said is that even with the use of measure
theory, there are sets in the plane complex enough that one cannot assign an area
measure to them.
6.4.1 Exercises
1. Show that a circle with radius r has area r2 .
2. Suppose the polygonal region A in the coordinate plane has area K. Show that
the region f.x; y/ j .x; 3y / 2 Ag has area 3K.
170
6 Riemann Integrals
Fig. 6.6 Approximating the area under a curve with narrowing rectangles
If the function is in some sense well behaved, then as the widths of these rectangles
are chosen to be smaller and smaller, the total area of the rectangles will approach
the area of the region (Fig. 6.6). What is meant by well behaved will be a main focus
of the theorems presented in this chapter.
To make the definition of Riemann Integral precise, there needs to be a way to
talk about the placement of the vertical rectangles used to approximate the area
under a curve. This is done by designating the position of the vertical sides of the
rectangles with a collection of x values in the interval a; b. One defines a partition
of the interval a; b to be a finite sequence of x values P W a D x0 x1
x2 xn D b for some natural number n. These values of x break the interval
a; b into n subintervals xj1 ; xj . Note that the definition of partition does not say
anything about the lengths of the subintervals for the partition. Indeed, it could be
that the jth subinterval length xj xj1 could be 0 or could be as large as b a.
In particular, there is no requirement that all the interval lengths be the same size.
Since the lengths of the subintervals xj xj1 are used frequently in the discussion
of Riemann Integrals, one often uses the shorthand notation xj D xj xj1 .
Given a partition, P W a D x0 x1 x2 xn D b, one defines the
norm of the partition P , jjP jj, to be the maximum length of a subinterval of the
partition, that is, jjP jj D max xj . For example, if a; b D 1; 4 has the partition
jn
171
the length of the rectangle (its height above the x-axis) cannot exceed
inf
xj1 xxj
f .x/.
If the rectangle is going to enclose the part of the region between xj1 and xj , then the
length of the rectangle must be at least sup f .x/. The definition of the Riemann
xj1 xxj
Integral uses a value between these two possible extremes. It requires the choice of
a sequence of x values 1 ; 2 ; 3 ; : : : ; n with xj1 j xj for each j. Then the
rectangle on xj1 ; xj is given the length f .j / so that it has area f .j /xj . Clearly,
the choice of j 2 xj1 ; xj results in the length of the rectangle being f .j / which
is between the two extremes inf f .x/ and sup f .x/, so the rectangles that
xj1 xxj
xj1 xxj
result might neither be contained in the region bounded by the curve nor cover the
region. Instead, the lengths of the rectangles are allowed to be in between these two
extremes. The total area of all the rectangles is then given by the Riemann Sum
n
P
f .j /xj .
jD1
Now, given a function f defined on the interval a; b, one can define the Riemann
Rb
Integral of f on a; b to be I D f .x/dx if for every > 0 there is a > 0
a
n
Rb
follows that f .j /xj I < . If f .x/dx exists, then f is said to be integrable
jD1
a
(or Riemann integrable) on the interval a; b. The function f in the integral is
called the integrand. When the integrand f is a nonnegative function, this definition
results in a value for I that can be considered the area of the region bounded by the
x-axis, the lines x D a and x D b, and the curve y D f .x/. When f is allowed to take
on both positive and negative values, the value of I can be thought of as the area of
the region lying above the x-axis minus the area of the region lying below the x-axis.
The power of the definition of Riemann Integral is that it need not be associated
with area at all. The student may well be familiar with other applications to the
determination of moments, work, force, speed, distances, interest rates, populations,
and many other examples. It is convenient to extend the definition of Riemann
Rb
Rb
Ra
Integral to f .x/dx where b < a with the convention f .x/dx D f .x/dx.
a
Note that the definition of Riemann Integral, similar to the definitions of limit and
derivative, states that the integral of f between the numbers a and b is I if for every
> 0 there is a > 0 such that a particular inequality holds. But unlike previous
kinds of limits, the inequality that must hold for Riemann sums is supposed to be
true for every choice of a partition P and every choice of j s as long as jjP jj < .
Thus, it is not just that a region in the plane is being approximated by a sequence
of rectangles, but that the region must be closely approximated by every possible
172
6 Riemann Integrals
n
P
jD1
noting is that the Riemann Integral is not the only way to define integration. Most
of the other definitions give the same value as the Riemann Integral for functions
where the Riemann Integral exists, but some of the other definitions give values to
integrals in situations where the Riemann Integral does not exist. Some examples of
other integration definitions include the RiemannStieltjes Integral, the Lebesgue
Integral, the Darboux Integral, and the Daniell Integral.
There are some fairly easy to describe functions that
do not have a Riemann
1 if x is rational
integral. One simple example is the function f .x/ D
whose
0 if x is irrational
Riemann integral is not defined on any interval a; b with a < b. To see why this is,
n
P
consider any Riemann sum
f .j /xj . Because both the rational numbers and the
jD1
irrational numbers are dense in the real numbers, in any subinterval of the partition
which has positive length, there are values of j in the subinterval where f .j / D 0,
and other values of j in the subinterval where f .j / D 1. Thus, for any partition,
n
P
there are choices of the j s that make the Riemann sums equal to
0 xj D 0 and
other choices that make the Riemann sum equal to
n
P
jD1
jD1
6.5.1 Exercises
1. Let f .x/ D x. Partition the interval 1; 3 into n subintervals with 1 D x0 and
xj D xj1 C 2n for j D 1; 2; 3; : : : ; n.
(a) Find the minimum and maximum possible values for an associated Riemann
n
P
sum
f .j /xj .
jD1
(b) Show that as n gets large, the Riemann sum must approach 4.
2. Let f .x/ D x2 . Partition the interval 1; 2 into n subintervals with 1 D x0 and
xj D xj1 C 3n for j D 1; 2; 3; : : : ; n.
(a) Find the minimum and maximum possible values for an associated Riemann
n
P
sum
f .j /xj .
jD1
(b) Show that as n gets large, the Riemann sum must approach 3.
173
Rb
c dx D c.b a/.
Rb
If f and g are functions integrable on the interval a; b, then .f C g/.x/dx D
Rb
a
f .x/dx C
Rb
g.x/dx.
If f and g are functions integrable on the interval a; b, and f .x/ g.x/ for all
Rb
Rb
x 2 a; b, then f .x/dx g.x/dx.
a
To prove that
Rb
cdx D c.b a/, one needs to find a > 0 so that if the norm of a
n
P
jD1
c.b a/. But in this case f .j / is always equal to the constant c, so the Riemann sum
is always equal to the desired integral, c.b a/. This makes the proof particularly
easy.
Note that the first four steps of this proof merely set up the assumptions required
by the definition of the Riemann Integral. That is, one needs to have constants
a and b and function f defined on the interval a; b. Then one needs to take an
arbitrary > 0, find an appropriate > 0, and consider an arbitrary Riemann
sum which satisfies the needed condition on the norm of the partition. Although
straightforward, these steps are necessary in order to show that the definition of
Riemann Integral is being satisfied.
174
6 Riemann Integrals
Rb
a
c dx D c.b a/.
Then for any choices of j 2 xj1 ; xj , it follows that f .j /xj c.ba/
jD1
P
P
n
n
Rb
a
c f .x/dx D c
Rb
a
f .x/dx.
Rb
In the proof of this result you will need to use the fact that f .x/dx D I to
a
say something about the size of c f .j /xj cI . But this expression equals
jD1
P
P
n
n
jcj f .j /xj I suggesting that if you can arrange for f .j /xj I to be
jD1
jD1
small, then you can arrange for the product jcj f .j /xj I to be small. You
jD1
. This is fine except for the embarrassing case where c D 0.
f .j /xj I < jcj
jD1
One could handle this problem by breaking the proof into two
cases: c D 0 and
P
c 0. Easier, though, is to simply ask for f .j /xj I to be less than jcjC1
.
jD1
The use of jcj C 1 in the denominator is just a trick that takes care of the case where
175
jcj is large
and the case where jcj is 0 both at the same time. Of course, you can
P
Rb
n
arrange f .j /xj I < jcjC1
because that follows from f .x/dx D I.
jD1
a
PROOF: If f is an integrable function on the interval a; b and c is a
Rb
Rb
constant, then c f .x/dx D c f .x/dx.
a
Rb
f .x/dx D I.
every choice of j 2 xj1 ; xj , f .j /xj I < jcjC1
.
jD1
P
n
n
< .
Then c f .j /xj cI D jcj f .j /xj I jcj jcjC1
jD1
jD1
Rb
Rb
Thus, c f .x/dx D cI D c f .x/dx.
a
The third theorem in this section can be summarized by saying that the integral
of a sum is the sum of the integrals. Its proof is reminiscent of the proof of
the theorem stating that the limit of a sum is the sum of the limits, and of the
theorem stating that the derivative of a sum is the sum of the derivatives. In this
Rb
Rb
case, you are given that f .x/dx D I and g.x/dx D J and are then faced with
a
the distance that the Riemann sum for f C g is from the value of the integral
P
I C J given by .f C g/.j /xj .I C J/. This easily breaks into the two
jD1
!
!
P
n
P
n
differences
f .j /xj I C
g.j /xj J . The existence of the two
jD1
jD1
given integrals then lets you choose a value of > 0 that will ensure that the
two parts to this sum are both small.
176
6 Riemann Integrals
Let D min.1 ; 2 /.
Let P W a D x0 x1 x2 xn D b be a partition of a; b with
jjP jj <
, and let j s be chosen with
j 2 xj1 ; xj .
P
!
!
P
n
P
n
f .j /xj I C
g.j /xj J
jD1
jD1
P
P
n
n
Rb
Rb
Rb
Thus, .f C g/.x/dx D I C J D f .x/dx C g.x/dx.
a
The final theorem in this section states that if f .x/ g.x/ for all x 2 a; b, then if
Rb
Rb
the functions are integrable, f .x/dx g.x/dx. It is sufficient to prove this result
a
Rb
h.x/dx 0, this
Rb
a
.g f /.x/dx D
Rb
a
g.x/dx
Rb
a
Rb
h.x/dx 0.
follows. With the assumption that h.x/ 0 for all x 2 a; b, it is not hard to
Rb
show that h.x/dx 0, because the value of every associated Riemann sum must
a
177
be nonnegative. How do you turn this into a proof? Recall how the proof went
when showing that if f .x/ 0, then lim f .x/ cannot be negative. If you assume that
x!a
Rb
a
Rb
h.x/dx 0.
n
P
h.j /xj I I > .
jD1
This contradicts the assumption that I < 0 which completes the proof.
6.6.1 Exercises
Write proofs for each of the following statements.
1. If functions f1 ; f2 ; f3 ; : : : ; fn are integrable on interval a; b, and c1 ; c2 ; c3 ; : : : ; cn
Rb
.c1 f1 .x/ C c2 f2 .x/ C c3 f3 .x/ C C cn fn .x// dx
are constants, then
D
c1
Rb
a
f1 .x/dx C c2
Rb
a
f2 .x/dx C c3
Rb
a
f3 .x/dx C C cn
Rb
of Linear Algebra, this says that the Riemann integral is a linear operator.)
178
6 Riemann Integrals
Rb
3. If f is a function such that both f and jf j are integrable on a; b, then f .x/dx
a
Rb
jf .x/jdx.
a
least one of the subintervals xj1 ; xj ; otherwise, if there is a bound for f on each of
the n subintervals, one merely needs to select the largest of those n bounds to have
a bound for f on the entire interval a; b. So what happens if f is not bounded on
the kth subinterval xk1 ; xk ? It means that k could be changed to be some other
value in the subinterval, say , to make the term f . /xk as large as you like.
Thus, you can make the entire Riemann
sum as largeas you like. So how large do
P
you want f . /xk to be? You want f .j /xj I to be larger than for some
jD1
preassigned > 0 such as D 1. The proof below does this by selecting a value
to replace k in such a way that the kth term of the Riemann sum, f . /xk ,
is
larger by at least
1 than the sum of the absolute values of all the other terms of
P
f .j /xj I guaranteeing that the resulting expression will be bigger than 1.
jD1
179
Then from the definition of Riemann integral, there is a > 0 such that if
P W a D x0 x1 x2 xn D b is a partition with
jjP
jj < and j s are chosen with j 2 xj1 ; xj , the Riemann sum satisfies
P
Let a particular partition P with jjP jj < and choices for j 2 xj1 ; xj be
given.
Because f is not bounded on a; b, it follows that there is a k between 1
and n such that f is not bounded on the interval xk1 ; xk . Otherwise, f
is bounded on each of the subintervals of the partition implying that it is
bounded on the entire interval a; b. Note that xk > 0 because a function
cannot be unbounded on an interval of length 0.
n
P
Let J D
jf .j /jxj jf .k /jxk C jIj.
jD1
n
n
P
jD1
JC1
xk
xk
J D 1.
natural to consider finding a 1 > 0 so that Riemann sums arising from partitions of
Rb
a; b with norm less than 1 are close to I D f .x/dx and finding a 2 > 0 so that
a
Riemann sums arising from partitions of b; c with norm less than 2 are close to
Rc
J D f .x/dx. You would consider allowing to equal the minimum of the 1 and 2 .
b
180
6 Riemann Integrals
Then you could take a partition of a; c with a norm less than . Unfortunately, this
partition of a; c does not separate into a partition of a; b and a partition of b; c
because there is no guarantee that the given partition of a; c includes the point b as
one of the xj values in the partition. But if you change the Riemann sum by altering
the interval of the partition containing the point b by adding b as an extra point to
the partition, you are not making a large change in the total sum. More precisely,
suppose the partition is P W a D x0 x1 x2 xn D c with the point b in
the interval xk1 ; xk . A resulting Riemann sum has the term f .k /.xk xk1 /. If this
term is replaced by two terms f .b/.b xk1 / C f .b/.xk b/, how much does this
change the Riemann sum? The change is exactly f .b/.b xk1 / C f .b/.xk b/
f .k /.xk xk1 / D .f .b/ f .k //.xk xk1 /. Given that f is integrable on a; b and
on b; c, you know that there is a bound M such that jf .x/j M for all x 2 a; c.
An upper bound for the size of this change is, therefore, 2M.xk xk1 / < 2M. This
says that by choosing small enough, you can control the amount of change made
in the Riemann sum by introducing b as a point in the partition of a; c. If is also
Rb
Rc
chosen small enough so that the resulting Riemann sums for f .x/dx and f .x/dx
a
are close to the corresponding integral, then the total difference between original
Rb
Rc
Riemann sum and the sum of the integrals f .x/dx C f .x/dx is small enough.
a
Without loss of generality assume that a < b < c, and let f be a function
Rb
integrable on the interval a; b and on the interval b; c with I D f .x/dx
and J D
Rc
f .x/dx.
181
.
Let D min 1 ; 2 ; 6MC1
Let P W a D x0 x1 x2 xn D c be a partition of a; c with
jjP jj < .
Let s be chosen such that j 2 xj1 ; xj .
Since b 2 a; c, there is a k such that b 2 xk1 ; xk .
Then
n
f .j /xj .I C J/ D
jD1
!
!
k1
n
P
P
f . /xj Cf .b/.b xk1 / C f .b/.xk b/ C
f .j /xj C
jD1 j
jDkC1
.f .xk / f .b//xk .I C J/
k1
n
P
P
f .j /xj J C
f .j /xj Cf .b/.b xk1 / I C f .b/.xk b/C
jD1
jDkC1
jf .xk / f .b/jxk :
Since the partition a D x0 x1 x2 xk1 b D
b is a partition of a; b with norm less than 1 , it follows that
k1
n
P
jDkC1
< 3 .
Also, jf .xk / f .b/jxk < 2M 6MC1
Rc
This proves that f .x/dx D I C J and completes the proof of the theorem.
a
Note that you can easily show that this theorem also holds if a > b or b > c by
simply rearranging the order of the limits on one or more of the integrals.
The previous section discusses the theorem stating that if integrable functions
Rb
Rb
satisfy f g on a; b, then f .x/dx g.x/dx. Can this statement be made
a
stronger? That is, if f .x/ g.x/ for x 2 a; b, with f .x/ < g.x/ for some x 2 a; b,
Rb
Rb
can you conclude that f .x/dx < g.x/dx? The answer is no. For example, if
a
f and g only differ for a finite number of x values, then f and g will have identical
integrals. To prove this, start with two integrable functions, f and g, that are identical
for all x 2 a; b except for some t 2 a; b. How would you prove that f and g have
182
6 Riemann Integrals
identical integrals? Again, you should consider the Riemann sums associated with f
n
P
and g, that is, consider a Riemann sum
g.j /xj for g with a particular partition
jD1
n
P
jD1
f .x/ D g.x/ at all points except x D t, how many of the corresponding terms in
these two Riemann sum could be different? Well, only those terms for which the
chosen j D t and xj 0. This could happen at most twice (twice in the unusual
case of t D xj D j D jC1 ). Thus, the Riemann sum for g is identical to the
Riemann sum for f plus at most two terms. By controlling the size of xj which
you can do by limiting the norm of the partition, you can control the contribution of
those at most two terms in the Riemann sum, thus ensuring that the sums for f and
g are close. That is the idea behind the following proof.
PROOF: Suppose that f and g are functions integrable on the interval
a; b, and that f .x/ D g.x/ for all x 2 a; b except perhaps at t 2 a; b.
Rb
Rb
Then f .x/dx D g.x/dx.
a
Let f and g be a functions integrable on the interval a; b, and suppose that
f .x/ D g.x/ for all x 2 a; b except perhaps at t 2 a; b.
Rb
Let f .x/dx D I.
a
Let 2 D 8M , and set D min.1 ; 2 /.
Select any partition P W a D x0 x1 x2 xn D b with
norm less than , and select any sequence of j 2 xj1 ; xj . Then the
P
D .
f .j /xj I C jg.t/ f .t/j2 < 2 C 2M 2 8M
jD1
Rb
Rb
Thus, g.x/dx D I D f .x/dx which proves the theorem.
a
It is left as an exercise to extend this theorem to the case where f and g differ at a
finite number of points. In fact, this can be extended to f and g which differ on an
infinite sequence of points in a; b as long as the sequence has a limit.
183
6.7.1 Exercises
Write proofs for each of the following statements.
1. If f and g are functions integrable on the interval a; b, and f .x/ D g.x/ for all
x 2 a; b except perhaps at the finite set of points ft1 ; t2 ; t3 ; : : : ; tk g a; b, then
Rb
Rb
f .x/dx D g.x/dx.
a
2. If f and g are functions integrable on the interval a; b, and f .x/ D g.x/ for all
x 2 a; b except perhaps on a sequence of points ft1 ; t2 ; t3 ; : : : g a; b where
Rb
Rb
lim tj D L, then f .x/dx D g.x/dx.
j!1
R1
r!0C r
f .x/dx D 3.
Rx
f .t/dt
3 0 x < 2>
>
>
>
>
=
< 1 2Dx
of the partition, .xj1 ; xj /. For example, the function s.x/ D
4 2 < x < 4 is
>
>
>
>
0 4Dx
>
;
:
1 4 < x 5
a step function defined on the interval 0; 5 (Fig. 6.7). It follows easily that a step
function on an interval a; b is integrable there. Indeed, suppose that P W a D x0
x1 x2 xn D b, and s.x/ D cj for all x satisfying xj1 < x < xj . Clearly,
the constant function cj is integrable on the interval xj1 ; xj , and the function s.x/
differs from this constant function at at most the two endpoints, xj1 and xj . Thus,
184
6 Riemann Integrals
n
P
Rxj
Rb
xj1
s.x/dx D
n Rxj
P
s.x/dx D
jD1 xj1
cj xj .
jD1
The importance of step functions comes from the fact that a function f is
integrable on a; b if and only if f can be closely approximated by step functions.
Precisely, f has a Riemann integral on the interval a; b if and only if for every
> 0, there exist step functions u.x/ and v.x/ on a; b with the property that for
Rb
Rb
all x 2 a; b, v.x/ f .x/ u.x/, and u.x/dx v.x/dx < . That is, f has
a
an integral precisely when for every > 0 there is a lower step function v that is
always less than or equal to f and an upper step function u that is always greater
than or equal to f with the property that the integrals of v and u are within of each
other. This squeezes f between two step functions whose integrals are as close as
you want. This should remind you of the Exhaustion Area Axiom.
The statement of this theorem is a biconditional statement; that is, it is an if and
only if statement. This means that the proof will have two distinct parts. One proof
must show that if a function is integrable, then it can be approximated by very close
upper and lower step functions. The other proof must show that if a function can
be approximated by very close upper and lower step functions, then it is integrable.
Consider how you would approach the proofs of each of these statements.
For the first part of the proof, you would consider a function, f , integrable on an
interval a; b. Given an > 0, somehow you need to show that there are upper and
lower step functions, u and v, whose integrals are within of each other. Where do
you start? All you know about f is that it has a Riemann integral on a; b, thus, all
you have to go on is the definition of Riemann integration which makes a statement
about the properties of Riemann sums. The key observation here is that a Riemann
n
P
sum
f .j /xj is equal to the integral of a step function defined to be equal to the
jD1
constant f .j / on the interval .xj1 ; xj /. Since the definition of the integral guarantees
185
that you can find Riemann sums that are very close to the value of the integral, this
suggests how you might choose step functions whose integrals are close to each
other. How can you assure that you choose a step function that is less than f .x/
for each x 2 a; b? For each interval of the partition .xj1 ; xj / you could consider
selecting j so that f .j / is the minimum value of f on that interval. Unfortunately,
f might not achieve a minimum value on that interval. Certainly, if f is continuous
on xj1 ; xj , then it obtains its minimum on that interval, but there is nothing here
indicating that f is continuous. On the other hand, you do know that, because f is
integrable, it is bounded. Thus, there is a greatest lower bound Mj D inf f .x/.
x2.xj1 ;xj /
There may not be any x 2 .xj1 ; xj / with the property that f .x/ D Mj , but you know
that there are values of x in the interval such that f .x/ is as close as you like to Mj .
Getting specific, now, your goal is to find upper and lower step functions whose
integrals are within some given > 0 of each other. It makes sense, therefore, to
find upper and lower step functions whose integrals are both within 2 of the value of
the integral of f because then the two step functions will be within of each other.
From the definition of Riemann integral, you can find a partition of a; b such that
all associated Riemann sums are within 4 of the integral of f . Then you can define
a lower step function, v.x/, that is equal to the infimum of f on each interval of the
chosen partition. On each interval of the partition you can find j values so that f .j /
is within 4.ba/
of v.j /. Then the integral of the lower step function will be within
.b a/ D 4 of a Riemann sum for f which in turn is within 4 of the integral
4.ba/
of f . This produces a lower step function with the properties you want. A similar
construction will produce an upper step function whose integral is also within 2 of
the integral of f , and that will complete the first part of the proof (Fig. 6.8).
For the second part of the proof, you consider a function, f , such that for each
> 0 you can find a lower step function, v.x/, and an upper step function, u.x/,
whose integrals are within of each other. You must then show that f has an integral.
The first task is to figure out what value I will serve as the integral of f . Your proof
will need to show that Riemann sums for f approach this value of I, so you first
4(b a)
inf f(x)
j
xj1
Fig. 6.8 Choosing j on .xj1 ; xj /
xj
186
6 Riemann Integrals
need a target I for that purpose. To do this, consider the collection of all possible
lower step functions, v.x/. That is, let L D fv j v is a step function with v.x/
f .x/ for all x 2 a; bg. Each v 2 L has an integral, and each integral should be less
than or equal to the needed value of I. How about taking the least upper bound of
all of those integrals? Does the least upper bound exist? It does if the collection
of integrals of elements of L is bounded above. To get that, all you need is one
upper step function u. For each v 2 L and for each x 2 a; b, you know that
Rb
Rb
v.x/ f .x/ u.x/. This ensures that for each v 2 L, v.x/dx u.x/dx showing
a
that the set of integrals of elements in L is bounded above. That allows you to set
Rb
I D sup v.x/dx. This makes sense because I would then be greater than or equal
v2L a
to the integral of any lower step function. It would also have to be less than or equal
to the integral of any upper step function. Since the assumption is that the integrals
of lower step functions and upper step functions can be found arbitrarily close to
each other, and each integral of an upper step function must be greater than or equal
to any integral of a lower step function, you would expect that the least upper bound
of the lower step function integrals would be equal to the greatest lower bound of
the upper step function integrals, and this value is what you will choose for I.
After determining I, your proof can proceed naturally. You need to show that by
restricting the norm of a partition of a; b, you can force an associated Riemann
sum for f to be close to I. What you have at your disposal is the ability to find
upper and lower step functions whose integrals are close to each other. A helpful
observation is that if you have a lower step function v and an upper step function u,
then for any partition and choice of j in the intervals of the partition, you know that
n
n
n
P
P
P
v.j /xj
f .j /xj
u.j /xj . So you can choose upper and lower step
jD1
jD1
jD1
functions, u and v whose integrals are each within, say 2 , of I. Then you can choose
a norm of a partition so that any Riemann sum for v is within 2 of the integral of
v, and any Riemann sum for u is within 2 of the integral of u. That will force the
corresponding Riemann sum for f to be within of I completing the proof.
PROOF: The function f is integrable on the interval a; b if and only if for
every > 0 there are step functions, u and v, such that for each x 2 a; b,
Rb
Rb
v.x/ f .x/ u.x/ and u.x/dx v.x/dx < .
a
187
Rb
f .x/dx D I.
inf
xj1 <x<xj
f .x/ C
.
4.ba/
For each j, define v.xj / D f .xj / and for x 2 .xj1 ; xj /, define v.x/ D
inf f .x/ f .j / 4.ba/
.
xj1 <x<xj
Then v is a step function with the property that v.x/ f .x/ for all x 2
n
n
Rb
P
P
a; b, and v.x/dx
f .j / 4.ba/
xj D
f .j /xj 4 . Since
jD1
jD1
defined to be
sup
xj1 <x<xj
f .x/ f .j / C
.
4.ba/
Then u is a step function with the property that f .x/ u.x/ for all x 2
n
n
Rb
P
P
a; b, and u.x/dx
f .j / C 4.ba/
xj D
f .j /xj C 4 . Since
jD1
jD1
(continued)
188
6 Riemann Integrals
Let u be any upper step function for f . Every lower step function, v,
Rb
satisfies v.x/ f .x/ u .x/ for every x 2 a; b, implying that v.x/dx
a
(
)
Rb
Rb
u .x/dx and that the set
v.x/dx j v is a lower step function of f is
a
bounded above by
Rb
u .x/dx.
Since the integral of any upper step function is an upper bound for the set
Rb
of all integrals of lower step functions, it follows that I u.x/dx <
Rb
v.x/dx C
Also
Rb
2
< I C 2 .
v.x/dx >
Rb
u.x/dx
2
> I 2 .
Similarly, there is a 2 > 0 such that for every partition of a; b with norm
less than 2 and every choice of j s in the intervals of the partition, the
n
Rb
P
Riemann sum
v.j /xj is within 2 of v.x/dx.
jD1
n
P
n
P
Rb
jD1
f .j /
u.j / < u.x/dx C 2 < I C 2 C 2 D I C .
jD1
a
P
n
Rb
Thus, f .j / I < which shows that f .x/dx D I and completes the
jD1
a
proof of PART II.
jD1
189
6.8.1 Exercises
Write proofs for each of the following statements.
1. If s.x/ and t.x/ are both step functions on the interval a; b, then so are
(a)
(b)
(c)
(d)
s.x/ C t.x/.
s.x/t.x/.
max s.x/; t.x/ .
s2 .x/ C t2 .x/.
2. If f and g are integrable functions on the interval a; b, then so is max.f ; g/.
3. If f is an integrable function on interval a; b, then so is jf j.
x2.xj1 ;xj /
Defining upper and lower step functions to be equal to this supremum and infimum,
respectively, on .xj1 ; xj / gives the step functions with the needed property.
190
6 Riemann Integrals
y2xj1 ;xj
function for f .
Similarly, define step function u.x/ by u.xj / D f .xj / for each j D
0; 1; 2; : : : ; n and u.x/ D max f .y/ for each j D 1; 2; 3; : : : ; n. Thus,
y2xj1 ;xj
y2xj1 ;xj
.
u.x/ v.x/ < ba
Rb
Rb
Rb
Rb
u.x/ v.x/ dx <
Thus, u.x/dx v.x/dx D
a
dx
ba
D .
Therefore, u and v are upper and lower step functions for f whose integrals
on a; b differ by less than , so it follows that f is integrable on a; b which
completes the proof.
One thing nice about knowing that a function is integrable on an interval a; b
is that rather than having to consider all partitions of a; b, you can determine the
value of the functions integral by using any collection of partitions of a; b whose
norms approach zero. Thus, if you know that f is integrable on a; b, then for every
n
P
natural number n you could calculate I.n/ D
f a C .b a/ nj ba
which is the
n
jD1
Riemann sum for f based on the very specific partition where xj D a C .b a/ nj and
with j D xj . This is not the more general Riemann sum required by the definition
of the integral, but if you already know that the integral exists, then it must be equal
to lim I.n/.
n!1
jD1
n!1
jD1
R4
0
191
f(c)
0 x is rational
this with the function f .x/ D
on the interval 0; 1, you obtain
1 x is irrational
n
P
f nj 1n D 1. So lim I.n/ D 1 which is not the integral of f . That
I.n/ D
n!1
jD1
nice visual interpretation showing that the area under a continuous curve is equal to
the area of a rectangle with length b a and width f .c/ for some c 2 a; b as shown
in Fig. 6.9. Another way to think about this is that there is a c 2 a; b such that f .c/
Rb
1
is the mean value of f which could be defined as ba
f .x/dx.
a
The proof of this theorem follows easily from three earlier results: (1) the
Intermediate Value Theorem, (2) a continuous function on a closed interval takes
on its extreme values, and (3) if one integrable function is greater than or equal to a
second integrable function, then the integral of the first is greater than or equal to the
integral of the second. The proof starts with a function f continuous of an interval
a; b. That function achieves its minimum value K and its maximum value M on the
interval. Thus, for all x 2 a; b, it follows that K f .x/ M from which it follows
Rb
that .b a/K f .x/dx .b a/M. Then, by the Intermediate Value Theorem,
a
on the interval a; b the function f achieves every value between K and M including
Rb
1
f .x/dx.
ba
a
192
6 Riemann Integrals
f .x/dx
Rb
M dx D M.b a/.
a
1
ba
Because f .s/ D K
Rb
1
ba
Rb
f .x/dx.
point c 2 .a; b/, then F 0 .c/ D f .c/. The second part states that if f is continuous
on a; b, and if F is any function satisfying F 0 .x/ D f .x/ for all x 2 a; b, then
Rb
f .x/dx D F.b/ F.a/. It is fairly straightforward to prove the second part using
a
the derivative. That is, you would start with the difference quotient F.x/F.c/
D
xc
x
x
c
R
R
R
1
1
f .t/dt f .t/dt . This simplifies to xc
f .t/dt. Now if you knew that f
xc
a
193
were continuous between c and x, you could apply the just completed Mean Value
Theorem for Integration to conclude that this difference quotient is equal to f .y/ for
some y between c and x. Then by forcing x to be close to c, you could force f .y/
to be close to f .c/ to complete the proof. But you do not know that f is continuous
between c and x; only that f is continuous at c. Still this is enough. You can use
the continuity of f at c to say that for a given > 0 there is a > 0 that ensures
that if t satisfies jt cj < , then jf .t/ f .c/j < . This shows that for x within
Rx
Rx
Rx
1
1
1
of c, xc
.f .c/ /dx < xc
f .t/dx < xc
.f .c/ C /dx which simplifies to
c
f .c/ <
1
xc
Rx
Since f .c/ < f .t/ < f .c/ C for all t between c and x, it follows that
Rx
Rx
Rx
1
1
1
.f .c//dx < xc
f .t/dx < xc
.f .c/C/dx D f .c/C.
f .c/ D xc
c
c
c
Rx
1
Thus, F.x/F.c/
f .c/ D xc
f .t/dx f .c/ < .
xc
c
194
6 Riemann Integrals
PROOF (Fundamental Theorem of Calculus: Part II): Assume the function f is continuous on the interval a; b and that F is any antiderivative
Rb
of f . Then f .x/dx D F.b/ F.a/.
a
proof.
The importance of the Fundamental Theorem of Calculus cannot be overstated. It
turns the complex operation of finding limits of difficult to calculate Riemann sums
into the somewhat more routine job of finding antiderivatives of functions.
6.9.1 Exercises
1. If F.x/ D
Rx3
x2
t
dt,
1Ct2
find F 0 .x/.
2. Suppose f has a jump discontinuity at c 2 a; b (that is, lim f .x/ and lim f .x/
x!cC
x!c
both exist and are unequal). If f is integrable on a; b, what is the behavior of
Rx
F.x/ D f .t/dt at c?
a
Rx
f .t/dt exists at
195
behaved near points where it is continuous, and the set where it is not well behaved
is very small (has measure zero). The strategy, then, is to construct step functions,
u and v, so that u.x/ v.x/ is very small near points where f is continuous, and
196
6 Riemann Integrals
to limit the size of the intervals where u.x/ v.x/ is large. Suppose, for example,
that near points where f is continuous, you could limit u.x/ v.x/ to be less than
. Then the total contribution to the integral of u v over those sections of the
2.ba/
.b a/ D 2 . The function f is bounded, so
step functions would be at most 2.ba/
there is an M such that jf .x/j < M for all x 2 a; b. It is possible, therefore, to
define upper and lower step functions that differ by at most 2M at points of Df . If
you can limit the regions where u.x/ v.x/ is large to intervals whose total length
is at most 4M
, then the total contribution to the integral of u v over those sections
of the step functions would be at most 2M 4M
D 2 . Accomplishing both of these
goals would then show that the integral of u v is less than 2 C 2 D . Can this be
accomplished? By the definition of continuity, for each point x where f is continuous
there is a > 0 such that if y is in a; b with jy xj < , then jf .y/ f .x/j < 4.ba/
.
That would ensure that for any two values y1 and y2 in the interval .x ; x C /, the
difference jf .y1 /f .y2 /j jf .y1 /f .x/jCjf .x/f .y2 /j < 4.ba/
C 4.ba/
D 2.ba/
.
By the definition of measure zero, the set of discontinuities of f can be covered by a
collection of open intervals whose total length is less than the needed 4M
. Thus, each
point of a; b can be covered by one of the open intervals covering Df or by one of
these .x ; x C / intervals constructed at each point of continuity. The Heine
Borel Theorem then lets you reduce this covering of a; b with open intervals to a
finite subcovering, and from that subcovering, the appropriate upper and lower step
functions can be constructed. That completes the strategy for the first part of the
proof.
Assume, conversely, that the function f defined on the interval a; b is integrable.
You already know that this implies that jf j is bounded by some constant M, so all
you need to prove is that the set of discontinuities of f , Df , has measure zero. This
can be done with a proof by contradiction. That is, by assuming that Df does not
have measure zero, you can show that for any upper and lower step functions, u and
v, the integral of u v is bounded away from 0. To do this it is helpful to consider
how much f can vary near a particular value x. For a point x 2 a; b and a > 0,
you would like to know how much f can change over the interval .x ; x C /. So
define W .x/ D sup f .y/ inf f .y/ where the supremum and infimum are calculated
for y varying over the interval .x ; x C / \ a; b. Note that if f had upper and
lower step functions that were both constant on the interval .x ; x C /, then
the two step functions would have to differ by at least W .x/ on that interval. Now
define the variation of a function f at a point x to be W.x/ D lim W .x/. Since
!0C
197
jzxj<
sets, that is, Df D [ Dnf . The key observation here is that if for each n the Dnf set
nD1
has measure zero, then the entire set of discontinuities, Df , must have measure zero
because it is just a countable union of sets with measure zero. So, if you assume that
Df does not have measure zero, it requires that there is a natural number n such that
Dnf also does not have measure zero. What does it mean for Dnf not to have measure
zero? It means that there is an > 0 such that no collection of open intervals with
total length less than can cover all of Dnf . This will be the key to showing that
upper and lower step functions for f cannot have integrals that are arbitrarily close
to each other, and thus, f cannot be integrable. The result is known as Lebesgues
Theorem.
198
6 Riemann Integrals
u.x/ v.x/dx D
u.x/ v.x/dx.
a
jD2 xj1
(continued)
199
Over the intervals that were subsets of the Ij intervals, u.x/ v.x/ D 2M.
The total length of such intervals cannot exceed 4M
. As a result, the integral
of u.x/ v.x/ over these intervals cannot exceed 2M 4M
D 2 .
Over the intervals that were subsets of the Jx intervals, u.x/ v.x/ < 2.ba/
.
As a result, the integral of u.x/ v.x/ over these intervals cannot exceed
Rb
D 2 .
2.ba/
a
Thus, f has upper and lower step functions, u and v, with the property that
Rb
u.x/ v.x/dx < 2 C 2 D .
a
than inf f .z/. As a consequence, u.x/ v.x/ sup f .z/ inf f .z/. Thus, if
z2Ij
Dnf
z2Ij
1
n
z2Ij
for all x 2 Ij .
(continued)
200
6 Riemann Integrals
Dnf cannot be covered by open intervals whose total length is less than .
Thus, it follows that the total length of the intervals Ij that contain points of
Df must be at least .
Rb
It follows that u.x/ v.x/dx 1n .
a
Thus, f cannot have upper and lower step functions whose integrals differ
by less than n . This implies that f is not integrable which is a contradiction.
Therefore, the assumption that Df does not have measure zero is false,
which completes the proof.
The last section of Chap. 4 introduced Thomaes function, a function defined
on 0; 1 which is discontinuous at each rational number but is continuous at each
irrational number. Since the rational numbers is a countable set, it has measure zero.
Thus, Thomaes function is bounded, and its set of discontinuities has measure zero,
so Thomaes function is Riemann integrable. Compare this to the function that is
equal to 1 for all rational numbers and equal to 0 for all irrational numbers. That
function is discontinuous everywhere, so its set of discontinuities does not have
measure zero, and it is not integrable as seen earlier.
6.10.1 Exercises
1. Suppose f W 0; 2 ! 5; 9 is integrable and g W 5; 9 ! 0; 2 is continuous.
Show that g f is integrable.
Write proofs for each of the following statements.
2. If f .x/ is a function integrable on the interval 0; 10, then so is the function
f .x/f .10 x/.
3. If f .x/ is a function integrable on the interval a; b, and p.x/ is a polynomial,
then p f .x/ is also integrable on a; b.
4. If f .x/ and g.x/ are integrable functions on the interval a; b, then so is f .x/g.x/.
Chapter 7
Infinite Series
of terms a1 ; a2 ; a3 ; : : : which are written with plus signs or minus signs between
the terms of the sequence. In this chapter, most series will begin with a first term
a1 , although there is no problem with beginning the series at other subscript values
1
P
such as the commonly seen
an . Also in this chapter the terms of the series will
nD0
201
202
7 Infinite Series
nD1
says that the series has limit L or even that the series has value L. If the sequence
of partial sums does not converge, then the series is said to diverge. If the limit of
partial sums converges to infinity or negative infinity, the series is said to diverge to
1
P
infinity or negative infinity, respectively. In that case one could write
an D 1 or
nD1
1
P
an D 1.
nD1
k
P
1
P
an one should
nD1
nD1
limit of the partial sums lim sk . Unfortunately, there are relatively few series that
k!1
admit simple closed-form expressions for their partial sums, and this technique for
finding the value of a series has limited use. Still, it is important to know about
some of the cases when this technique does work. Perhaps the best known examples
of series whose partial sums can be explicitly calculated are the geometric series.
These are the series whose sequence of terms can be written in the form an D arn1 ,
where a and r are given real numbers. Then the first term of the series is a and the
n1
an
common ratio of adjacent terms is an1
D ar
D r, at least in the interesting cases
arn2
when ar 0. When r is not equal to 1, there is a simple algebraic trick that gives
the expression for the partial sums.
sk D
k
X
arn1
nD1
r sk D
k
X
nD1
arn
203
sk r sk D
k
X
arn1
nD1
k
X
arn D a ark
nD1
sk .1 r/ D a.1 r /
k
sk D a
1 rk
:
1r
(Of course, there is an even simpler trick for the case when r D 1.) Thus, except
in the trivial case where a D 0, the limit of the partial sum diverges if jrj 1. On
a
which can easily be
the other hand, when jrj < 1, lim rk D 0 so lim sk D 1r
k!1
k!1
remembered as the first term divided by 1 minus the common ratio. The geometric
series is particularly important because one can often compare other series to a
geometric series to determine if the other series converges. It also gives a nice
example showing that series that make a lot of sense when they converge can lead
you to very strange and very incorrect conclusions when they do not converge. In
1
P
r
particular,
rn D 1r
whenever jrj < 1. But when you take a limit as r approaches
nD1
1
P
1
P
r!1 nD1
rn D lim
r
r!1 1r
1
2
nD1 r!1
Another class of series whose partial sums can be calculated are the telescoping
series. This is a class of series where each term an can be written as a difference of
two terms an D bn bnC1 . Then sk D .b1 b2 / C .b2 b3 / C .b3 b4 / C C
.bk bkC1 / D b1 C.b2 Cb2 /C.b3 Cb3 /C.b4 Cb4 /C .bk C bk / bkC1 D
b1 bkC1 . Hence, if lim bkC1 exists, the series converges. The best known example
k!1
1
1
P
P
1
1
1
1
D 1 lim nC1
D
nC1
D 1.
of this type is the series
2
n
n Cn
nD1
nD1
n!1
Fortunately, even though it is often difficult to determine the exact values for
the partial sums of a series, one can very often determine whether or not the series
converges and sometimes the value to which it converges even without knowing an
explicit formula for its partial sums. There are many tools that can be used to do this.
These tools consist of a large collection of convergence tests which can be applied
to determine if a particular series converges. Calculus students often get a great deal
of practice selecting appropriate convergence tests for series. This chapter will be
more interested in proving the theorems that provide these tests.
The simplest and possibly most important convergence test is the Limit of Terms
Test which says that a series can converge only if its sequence of terms has a limit
1
P
of 0. That is, if
an converges, then lim an D 0. This is a direct consequence of
nD1
1
P
nD1
n!1
204
7 Infinite Series
The point is, if lim sk exists, then <sk > is a Cauchy sequence whose term must get
k!1
1
P
an converges only if
nD1
lim an D 0.
n!1
1
P
nD1
k
P
an converges to L.
nD1
nD1
an D bn for all n > N. Why would the convergence of one of the series imply
the convergence of the other? It must depend on the convergence of their partial
k
k
P
P
sums, so let sk D
an and tk D
bn be the sequences of partial sums for
nD1
nD1
the two series. The agreement of an and bn for all n > N shows that for k > N,
k
k
P
P
tk D t N C
bn D tN C
an D tN C sk sN . Thus, lim sk exists if and only
nDNC1
nDNC1
k!1
k!1
7.1.1 Exercises
Find limits for the following series or show that the limit does not exist.
1.
2.
1
P
nD1
1
P
nD1
5
3n
4
22nC1
3.
4.
5.
1
P
nD1
1
P
nD1
1
P
nD1
23
205
7
n2 C5n
1
n2 C9nC14
1
1 1
1
1
1
1
1
C 2 C 3 3 4 C 4 C
2 3 22
3
2
3
2
3
1
1
1
1
7. C 0 C C 0 C 0 C C 0 C 0 C 0 C
C 0 C 0 C 0 C 0
2
4
8
16
1
1
1
1
1
5
C
C
C
C
8. 11 C 3 C C
9
12
23
34
45
56
6.
1
P
1
P
nD1
nD1
1
P
nD1
by using the fact that a series converges if and only if its sequence of partial sums is
1
1
P
P
Cauchy. The proof can begin with the absolutely convergent series
an , so
jan j
converges. Then, knowing that
sequence of partial sums Sk D
nD1
1
P
nD1
nD1
k
P
nD1
1
P
nD1
must be Cauchy, which means that for each > 0 there is an N such that for any
m > k > N, jSm Sk j < . What do you need to know for <sk > to be Cauchy?
You need to know that for each > 0 there
is an mN such that for any m > k > N,
m
P
P
206
7 Infinite Series
1
P
nD1
1
P
1
P
nD1
jan j converges.
nD1
k
P
jan j and sk D
nD1
k
P
an .
nD1
Then, since the series is absolutely convergent, the sequence <Sk > converges implying that <Sk > is a Cauchy sequence.
Given > 0 there is an N such that for all m and k greater than
N, jSm Sk j < .
m
m
P
P
Let m > k > N. Then jsm sk j D
an
jan j D jSm Sk j < .
nDkC1
nDkC1
Thus, the sequence <sk > is a Cauchy sequence and is, therefore, a
convergent sequence.
1
P
This shows that
an is convergent which proves the theorem.
nD1
If the series
1
P
nD1
1
P
nD1
k
P
nD1
an .
can D c
k
P
nD1
k
P
nD1
nD1
nD1
207
1
P
c an D c
nD1
1
P
nD1
an .
nD1
Assume that
Then
1
P
nD1
1
P
can D lim
k
P
k!1 nD1
nD1
k
P
can D lim c
k!1
an D c lim
k
P
k!1 nD1
nD1
an D c
1
P
an
nD1
.an C bn / D
nD1
1
P
1
P
an C
nD1
1
P
an and
nD1
1
P
nD1
nD1
an C
nD1
1
P
an and
nD1
1
P
1
P
nD1
1
P
.an C bn /D
nD1
bn .
nD1
1
P
nD1
k
P
k!1 nD1
1
P
an and
nD1
.an C bn / D lim
an C lim
k
P
k!1 nD1
k
P
1
P
bn both converge.
nD1
.an C bn / D lim
k!1 nD1
bn D
1
P
nD1
an C
k!1
1
P
k
P
an C
nD1
k
P
bn
nD1
nD1
With these theorems you can often start with a series whose value you know
and derive the values of other series. For example, what is the value of the
1
1
1
1
1
series 1 C 12 14 C 18 C 16
32
C 64
C 128
256
? This series looks
something like the geometric series with first term 1 and common ratio 12 which
is 1 C 12 C 14 C 18 C . That series has limit 11 1 D 2. But the new series
2
is clearly not a geometric series because the terms are not all the same sign,
which would be the case for a geometric series with a positive common ratio,
nor are the terms alternating in sign, which would be the case for a geometric
series
with a negativecommon
ratio. The new series can be written, though,
as
2
2
1 C 12 C 14 C 18 C 0 C 0 C 24 C 0 C 0 C 32
C 0 C 0 C 256
C . This is
the difference of two series: the geometric series with first term 1 and common ratio
1
, and a series whose value is the same as a geometric series with first term 24 D 12
2
1
1
10
2
.
and common ratio 18 . Thus, the new series has value
D
1
1
7
1 2
1 8
208
7 Infinite Series
1
1 1
1
1
1
1 1
1
1 1
C C C C
C D 1 C 0 C C
3 2
5
7 4
9
11 6
3 2
5
1
1
1
1 1
C D
C0C C C0C
7 4
9
11 6
1
1 1
1 1
1 C C C
2
3 4
5 6
1
1
1
1
C 0 C C 0 C 0 C C 0 C 0 C D
2
4
6
8
1
1 1
1 1
1 C C C
2
3 4
5 6
1
1 1
1 1
1
C
1 C C C D
2
2
3 4
5 6
ln 2 C
1
3
ln 2 D ln 2:
2
2
It is not unusual that rearranging the order of terms in a series results in the series
converging to a different quantity. This is, in fact, a characteristic of all conditionally
convergent series as will be shown later in this chapter.
7.3.1 Exercises
1. Prove that if the series
1
P
an and
nD1
1
P
nD1
numbers, then
1
1
1
P
P
P
.c an C d bn / D c
an C d
bn .
nD1
nD1
1
P
nD1
converges to 0.
nD1
1
P
mDn
am
209
nD1
1
P
nD1
nD1
nD1
1
P
an
nD1
must converge. The contrapositive of this statement is then also true and states that
1
1
P
P
if
an diverges, then
bn must also diverge.
nD1
nD1
Consider how you would prove that this test is valid. The proof would assume that
1
P
0 an bn for each n, and assume that
bn converges. Then it must show that
1
P
nD1
an converges. One shows that a series converges by showing that its sequence
nD1
of partial sums converges. You do know that the sequence of partial sums for
1
P
bn
nD1
converges, so how can you use that to make a conclusion about the partial sums of
1
P
an ? One idea is to use the technique from the proof that absolutely convergent
nD1
210
7 Infinite Series
series are convergent; that is, a series converges if and only if its sequence of partial
1
m
P
P
sums is Cauchy. If the partial sums of
bn form a Cauchy sequence, then
bn
nDk
nD1
gets small whenever k m are large. Now, the given fact that an bn lets you
m
m
1
P
P
P
conclude that
an
bn which implies that the partial sums of
an are
nDk
1
P
Cauchy. Thus,
nDk
nD1
an must converge.
nD1
an and
nD1
bn are series
nD1
with nonnegative terms and N is a real number such that for every
1
P
integer n > N, the terms satisfy 0 an bn . Then if
bn converges, so
does
1
P
nD1
an .
nD1
Assume that
P
nD1
an and
nD1
Assume that there is an N such that for every n > N, the terms of the series
satisfy 0 an bn .
1
P
Assume that the series
bn converges.
nD1
k
P
nD1
P
P
>
bn
bn D
bn .
nD1
nD1
nDkC1
m
m
P
P
But then whenever M < k m, it follows that >
bn
an D
m
P
nD1
an
k
P
nDkC1
nDkC1
an .
nD1
1
P
1
P
an is Cauchy.
nD1
nD1
converges.
This proves that the Comparison Test is valid.
The Comparison Test can be used in many cases when you are faced with a series
which is similar to a series that you know converges. For example, you already know
1
P
1
that the series
converges because it forms a telescoping series. Can this fact
n2 Cn
nD1
211
1
P
nD1
1
n2
be applied directly because for each n you have n2 1Cn < n12 which is not what you
need. You need to find a convergent series whose terms are greater than or equal to
1
or a divergent series whose terms are less than or equal to them. You have neither.
n2
2
2
On the other hand for each positive integer n, it is true that n12 D n2 Cn
2 n2 Cn .
1
P 2
is twice a convergent series, so it is also convergent. Thus, the
The series
n2 Cn
nD1
1
P
nD1
1
n2
converges.
In this way the Comparison Test can be used to simplify the task of testing the
convergence of many complicated looking series. As another example, consider the
1
P
2nC7
series
. Note that the first two terms of this series are negative. Because
n3 5nC1
nD1
the convergence of a series does not depend on the value of any finite set of its
terms, it is sufficient to test the series by considering the terms where n 3. In
the terms n32nC7
the degree of the polynomial in the denominator is 3 while the
5nC1
degree of the polynomial in the numerator is 1. This suggests that the terms could be
compared to the terms n12 of a known convergent series. The strategy is to compare
2nC7
to a fraction that is greater but look more like n12 . If the series with greater
n3 5nC1
fractions converges, the Comparison Test shows that the original series converges.
This can be done by attempting to eliminate lower degree terms of the numerator
and denominator polynomials, thus, ending up with a simpler fraction greater than
the original. Clearly, when considering the numerator 2n C 7, the constant term,
7, will be dwarfed by the size of the linear term 2n suggesting that you replace
2n C 7 by the larger quantity 2n C 7n D 9n. This replacement will result in a larger
fraction, but it should not affect whether or not the series converges. Similarly, it
would be good to replace the denominator n3 5n C 1 with a smaller polynomial of
the same degree which will result in obtaining a fraction larger than n32nC7
. One
5nC1
can drop the constant term altogether, but one cannot drop the 5n term without
making the denominator polynomial larger. This can be handled by writing n3 as
1
n3 C 12 n3 . For large enough values of n, the value of 12 n3 will exceed 5n making
2
1
n3 5n a positive quantity which could be removed from the polynomial to make
2
the polynomial smaller. Indeed, you need 12 n3 5n 0 implying n2 10. Thus,
if n 4, you can conclude that n3 5n C 1 > 12 n3 . This shows that for n 4, the
1
1
P
P
1
1
18
< 19n
D
18
.
fraction n32nC7
2 . Since the series
2 converges, so does
3
5nC1
n
n
n2
n
2
1
P
nD1
nD1
2nC7
n3 5nC1
nD1
converges.
212
7 Infinite Series
or equal to the corresponding term of the first series. Thus, by the Comparison Test,
the harmonic series must diverge.
1
P
nD1
with positive terms, and suppose that the sequence of ratios of adjacent terms, nC1
an
has limit L as n approaches infinity. If L < 1, then the series can be compared to
a convergent geometric series with a common ratio between L and 1. If L > 1,
then the terms of the series increase in value and do not approach 0, so the series
diverges. When L D 1, the ratio test fails because there are series for which L D 1
that converge and other series for which L D 1 that diverge.
1
P
To prove that the Ratio Test is valid, you would start by assuming that
an
nD1
given series to the geometric series with nth term arn1 , you would need to know
that for all n greater than some N, the terms an are less than arn1 . If you know that
anC1
is always less than r, then the an terms will grow more slowly than the arn1
an
terms, and the Comparison Test can be used. In general, r cannot be set equal to L
a
because knowing that nC1
approaches L in the limit does not ensure that the ratio
an
is ever actually smaller than L and certainly not that it is always less than L. But if
a
the limit of nC1
is L, then by the definition of limit, there is an N such that for all
an
, which is half
n N, the ratio is less than some value greater than L such as LC1
2
way between L and 1. Then, if the value of a is chosen so that aN a, you will have
aNCk ark for each k 0, and the result follows.
213
1
P
nD1
Assume that
n!1
nD1
anC1
an
D L.
CASE 1: L < 1
a
1
P
an is also convergent.
nD1
CASE 2: L > 1
a
1
P
nD1
1
P
n!1
nD1
1
n2
n!1
anC1
an
anC1
an
n2
2
n!1 .nC1/
D lim
1
n!1 1
D lim
D 1.
D 1.
nD1
n!1
214
7 Infinite Series
anC1
an
n!1
anC1
an
anC1
an
> 1 to
assure that the series diverges. The proofs of these facts are left as exercises, but
they are important refinements of the Ratio Test since the lim inf and lim sup always
exist even if the limit does not. For example, consider the series 1 C 23 C 13 C 322 C
a
1
C 323 C 313 C . For this series, the ratio nC1
oscillates between 23 and 12 , so the
an
32
limit of the ratio does not exist. But the lim sup of the ratio is 23 < 1 implying that
the series converges.
The Ratio Test will play a major role in the discussion of power series in the next
chapter.
Assume that
nD1
n!1
p
n a D L.
n
CASE 1: L < 1
p
If L < 1, there is an integer N such that n an < LC1
for all n N.
2
LC1 n
Then, for n N, each term an <
, the corresponding term of a
2
<
1.
geometric series with common ratio LC1
2
Therefore, since the geometric series converges, the Comparison Test shows
1
P
that
an converges.
nD1
(continued)
215
CASE 2: L > 1
p
If L > 1, there is an integer N such that n an > LC1
> 1 for all n N.
2
LC1 n
Then, for all n N, an > 2
which diverges to infinity.
Therefore, the series diverges by the Limit of Terms Test.
CASE 3: L D 1
Note that the constant series
The series
1
P
nD1
1
P
nD1
1
n2
p
n
n!1
n!1
p
n a D lim
n
n!1
an D lim 1 D 1.
n!1
1
p
n 2.
n
Since
natural
function is continuous at 1, it follows that
logarithmp
thep
ln lim n an D lim ln n an D lim 2 lnn n . Then by LHopitals Rule,
n!1
n!1
n!1
2
p
this limit is lim 1n D 0, from which it follows that lim n an D 1.
n!1
n!1
Therefore, no conclusion can be drawn when L D 1, and the Root Test fails.
p
As with the Ratio Test, it is sufficient to know that lim sup n an < 1 to conclude
n!1
p
that the series converges, and that lim inf n an > 1 to conclude that the series
n!1
diverges. For example, the series 12 C 13 C 212 C 312 C 213 C 313 C has general term
p
p
a2n D 31n and a2n1 D 21n . Thus, lim n an does not exist, but lim sup n an D p1
n!1
n!1
R1
f .x/dx D
1
lim 1x jb1
b!1
D 1.
lim
b!1 a
Rb
lim
a!1 b!1 a
1
R1
1
x2
dx D
a!1 a
Rb
lim x12
b!1 1
dx D
After seeing a definition of the improper Riemann integral of the first kind, the
reader may be curious whether there is also an improper Riemann integral of
the second kind. Although this text will not need to deal with improper Riemann
integrals of the second kind, the definition is given here for completeness. Recall
that Riemann integrals over an interval a; b exist only if the integrand is bounded.
So, an improper integral of the second kind is an integral where the integrand is
unbounded in every neighborhood of a point c 2 a; b. In this case, the Riemann
216
7 Infinite Series
integral on a; b can be calculated on a region that excludes c and then the limit can
R4
be taken as the region expands toward c. For example, one would define p1x dx as
lim
R4
a!0C a
1
p
p
dx D lim 2 xj4a D 4.
x
a!0C
The Integral Test for the convergence of a series of positive terms involves
the comparison of an infinite series with an improper Riemann integral. It applies
to series whose terms are equal to a monotonically decreasing function f defined
on an interval a; 1/ such that for all n a, the nth term of the series an is
equal to the function at the point n, that is, an D f .n/. The following figure
makes this comparison clear. Let k be an integer greater than or equal to a. If f
is a monotonically decreasing function, then whenever n x > k, the function
Rn
Rn
f .x/ f .n/ D an showing that
f .x/dx
f .n/dx D f .n/ D an . Thus, by the
Comparison Test, the series
1
P
n1
n1
1 nC1
R
P
nD1
1 nC1
R
P
f .x/dx D
nDk n
f .x/dx converges.
nDk n
R1
f .x/dx. Alternatively, if
1
P
f .x/dx
1
P
nDkC1
an
R1
R1
1 nC1
R
P
f .x/dx D
nDk
k a,
R1
1
P
f .x/dx
nDk n
an converges
nD1
of the infinite series and a good way to obtain an approximation to the value of
the series. This is helpful because it is often easier to evaluate the integral than the
corresponding infinite series (Fig. 7.1).
Fig. 7.1 Comparing the series with the integral in the Integral Test
217
converges.
to an integer k a.
Because f is monotonically decreasing, for any n > k it follows that f .x/
f .n/ for all x 2 n 1; n.
Rn
Rn
Thus, for any n k it follows that an D f .n/ D
f .n/dx
f .x/dx.
By the Comparison Test the series
1
P
Rn
f .x/dx D
nDkC1 n1
R1
1
P
n1
n1
nD1
f .x/dx converges.
1
P
nC1
R
nDkC1 n
f .x/dx D
R1
f .x/dx converges
kC1
an converges.
nD1
n
1
P
1
P
nD1
1
P
nD1
1
np
where p is some constant greater than 0. For which p does the p-series converge?
You have already seen that it converges when p D 2 and diverges when p D 1,
the harmonic series. All the p-series can be handled at once using the Integral Test.
Indeed, since the function f .x/ D x1p is monotonically decreasing in x for each p > 0,
R1
the p-series converges exactly when the integral x1p dx converges. But the integral
1
218
7 Infinite Series
R1
1
1
dx
xp
1 1
D 1p xp1
j1 . This is infinite when p < 1
nD2
nD2
nD1
1.5 and 2. This is not very precise, but one can apply this technique a few terms
1
1
R1
R1
P
P
1
1
farther down the series to get x12 dx >
> x12 dx which shows that
n2
n2
10
nD11
11
2
6
nD1
1:6449.
7.4.5 Exercises
1. Suppose that
1
P
bn is a convergent series,
nD1
1
P
nD1
1
P
constants N and K such that 0 an Kbn for all n > N. Prove that
an
nD1
converges.
2. Suppose that
1
P
nD1
1
P
an is a series
nD1
an
n!1 bn
1
P
an
nD1
1
P
lim sup
n!1
anC1
an
an converges.
n!1
nD1
nD1
1
P
1
P
anC1
an
an diverges.
nD1
1
C 221 C 212 C 222 C 213 C 223 C 214 C 224
21
a
lim inf nC1
. What can you conclude about the
an
n!1
an D
C calculate
nD1
and
the series?
6. Assume that
1
P
convergence of
nD1
1
P
nD1
n!1
an converges.
p
n a
n
7. Assume that
1
P
219
nD1
1
P
p
n a
n
n!1
an diverges.
nD1
1
P
8. For the series
an D 211 C 312 C 213 C 314 C 215 C 316 C 217 C 318 C calculate
nD1
p
p
lim sup n an and lim inf n an . What can you conclude about the convergence of
n!1
n!1
the series?
9. Use the integral estimate from the Integral Test to estimate the size of the series
1
P
1
.
n3
nD1
10. Determine which of the following series are absolutely convergent by applying
an appropriate convergence test.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
1
P
nD1
1
P
nD1
1
P
nD1
1
P
nD1
1
P
nD1
1
P
nD1
1
P
nD1
n
3
n
.2n/
5n
n
nn
.n/2
220
7 Infinite Series
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
s1 D 1 D 1
1
2
1
1
s3 D 1 C
2
3
1 1
1
s4 D 1 C
2
3 4
1 1
1
1
s5 D 1 C C
2
3 4
5
s2 D 1
1
2
5
D
6
7
D
12
47
D
:
60
D
221
1
P
nD1
lim an D 0, and for each n 1, an and anC1 have opposite signs, and
jan j janC1 j. Then the series converges.
n!1
Assume that
1
P
nD1
Note that if n 1 is odd, then anC1 is negative and anC2 is positive with
janC1 j janC2 j implying that snC2 D sn C .anC1 C anC2 / sn .
Similarly, if n 1 is even, then anC1 is positive and anC2 is negative with
janC1 j janC2 j implying that snC2 D sn C .anC1 C anC2 / sn .
Thus, the subsequence of odd numbered partial sums forms a monotonically
decreasing sequence while the subsequence of even numbered partial sums
forms a monotonically increasing sequence.
Because the subsequence of even numbered partial sums is increasing, when
n is an odd positive integer it follows that sn > snC1 s2 showing the
subsequence of odd numbered partial sums is bounded below by s2 implying
that that sequence converges to a limit L1 .
Similarly, the subsequence of even number partial sums is an increasing
sequence that is bounded above by s1 implying that that sequence converges
to a limit L2 .
Then L1 L2 D lim s2nC1 s2n D lim a2nC1 D 0 showing that L1 D L2
n!1
n!1
and that the odd numbered partial sums and the even numbered partial sums
both converge to the same limit.
Therefore, the sequence of partial sums converges and the series converges.
This proof not only says that the given alternating series converges; it gives a
way to estimate the limit of the series. For any series that satisfies the hypothesis
of the theorem, any two adjacent partial sums, sn and snC1 , are on opposite sides of
the limit L of the series. Thus, the distance that sn is from the limit of the series is
less than the distance sn is from snC1 , and that distance is just janC1 j. Therefore, it
is easy to remember that for these series, the distance that a partial sum is from the
limit of the series is no more than the first term that is not part of the sum, janC1 j.
Note that the Alternating Series Test for convergence and this limit estimate apply to
series without regard to whether the series is absolutely convergent or conditionally
convergent.
For example, the number 1e D 01 11 C 21 31 C . This is an absolutely
convergent series as seen by the ratio test. But it is also a series whose terms alternate
in sign, and the absolute values of the terms decrease monotonically to 0. Thus,
1
the partial sum of the series 01 11 C 21 31 C 41 is already within 100
of 1e because
1
the first neglected term is 51 D 120
. This technique gives an easy proof that the
222
7 Infinite Series
number e is irrational. It goes like this: If e were rational, then it could be expressed
as pq , where p and q are positive integers. Then 1e D qp D 01 11 C 21 31 C .
Multiplying both sides of this equation by p yields q.p 1/ D p p C p2 p3 C
1
1
1 pC1
C .pC1/.pC2/
. Thus, the integer q.p 1/ would be an integer
1
1
.pC1/.pC2/
C . But this infinite series is an alternating series
plus (or minus) pC1
1
where the absolute value of the terms decrease to 0, so its value is between pC1
and
1
1
.pC1/.pC2/ . Thus, there would have to be an integer between those two values,
pC1
something clearly not possible. This is a contradiction, so the assumption that e is
rational must be false.
7.5.1 Exercises
Determine which of the following series are conditionally convergent, absolutely
convergent, or divergent.
1. 1
2. 1
3. 1
4. 1 C
5. 1 C
1
C ln13 ln14 C
ln 2
1
C 3 ln1 3 4 ln1 4 C
2 ln 2
1
C 3.ln13/2 4.ln14/2 C
2.ln 2/2
p1 p1 C p1 C p1 p1 C
3
2
5
7
4
p1 p1 C p1 C p1 p1 C
3
4
5
7
8
p1
9
p1
9
C
C
p1
11
p1
11
p1 C
6
p1 C
12
1
P
nD1
1
np
raises a natural question about whether there is, in some sense, a largest series
that converges, or, perhaps a smallest series that diverges. If there were, that might
provide a good series to use in the Comparison Test because all series smaller would
converge, and all series larger would diverge. This turns out not to be the case. For
1
P
every series of positive terms,
an , that diverges, there is a sequence of positive
nD1
1
P
an bn also diverges. In
nD1
1
sn
1
sn
1
P
nD1
goes to 0.
an . Clearly, if the
223
To prove this result you would begin with a divergent the series with positive
1
P
terms,
an . Because the series is divergent, you know that the sequence of partial
nD1
sums must diverge to infinity. The strategy is to show that the partial sums of the new
1
P
an
series
are not Cauchy. In particular, for every integer m, there is an integer k
sn
nD1
such that
k
P
nDmC1
1
.
2
1
2
>
an
sn
showing that the mth and kth partial sums differ by at least
Suppose you are given a positive integer m. Since the original series diverges,
there is a positive integer k such that sk > 2sm . Then the difference between the kth
and the mth partial sums of the new series is
sk sm
sk
>1
1
2
nDmC1
D 12 .
1
P
PROOF: Let
k
P
sums sk D
nD1
sk D
k
P
an
sn
k
P
>
nDmC1
k
P
an
sk
an
nDmC1
sk
nD1
1
P
nD1
Assume that
k
P
1
P
an
sn
also diverges.
nD1
an .
nD1
is
k
P
nDmC1
an
sn
>
k
P
k
P
an
sk
nDmC1
nDmC1
sk
an
sk sm
sk
>1
1
2
D 12 .
1
P
nD1
an
sn
is not a
R1
2
nD2
1
dx
xln x
D ln.ln x/j1
2 D 1 diverges.
224
1
P
7 Infinite Series
It is interesting to note that even though for positive termed divergent series
1
1
P
P
an
an , the series
also diverges, for positive termed series
an , the series
sn
nD1
1
P
nD1
nD1
nD1
always converges. To see this, note that for n > 1 the term
an
s2n
an
s2n
sn sn1
s2n
<
1
P
an
Thus,
the
terms
of
the
series
are less than the terms of a
sn1
s2
nD1 n
1
P
1
convergent telescoping series
s1n D s11 lim s1n . Whether the original
sn1
sn sn1
sn sn1
1
.
sn
n!1
nD2
parentheses change whether or not it converges or change the limit to which the
1
P
series converges? The answer to this is no. The point is, if
an converges, it means
nD1
that its sequence of partial sums converges. By inserting parentheses into the series,
you are just removing some of the terms in the sequence of partial sums. You end up
with a new series whose sequence of partial sums is a subsequence of the sequence
1
P
of partial sums of
an , and any such subsequence will converge to the same limit
nD1
1
P
nD1
parentheses are inserted in such a way that the number of terms contained within
each set of parentheses is bounded, then the insertion of parentheses cannot affect
whether the series converges or the limit to which the series converges. To see this
1
P
assume that each set of parentheses encloses at most K terms. If
an converges
nD1
k
P
nD1
nD1
an . Because lim an D 0, for each > 0, there is an N such that for all
n!1
225
n > N, the size of the terms jan j must be less than K . Suppose that for some m > N
one term of the series with parentheses added is .amC1 CamC2 CamC3 C CamCk /.
Then sm and smCk are both partial sums for the series with parentheses added. For
any j D 1; 2; 3; : : : ; k, jamC1 CamC2 CamC3 C CamCj j K j , showing that for
any of those j, jsmCj sm j < . The sequence of partial sums for the original series
does not converge either because the sequence is unbounded or because its lim sup
and lim inf approach distinct values. Because the subsequence of partial sums for
the original series remains within of the subsequence corresponding to the series
with parentheses added, the subsequence must also either be unbounded or have
distinct lim sup and lim inf values. Thus, the series with parentheses added cannot
converge.
This observation can be very helpful. Consider again the series 1 C 13 12 C 15 C
1
1
14 C 19 C 11
16 C . This series is not absolutely convergent, and it does not
7
satisfy the hypothesis of the Alternating
Series
Yet, if parentheses
are inserted
Test.
1
to group each set of three terms: 1 C 13 12 C 15 C 17 14 C 19 C 11
16 C ,
1
1
4n3
C 2n1
1n D n.2n1/.2n3/
. The series with
one gets a general term equal to 2n3
4n3
terms n.2n1/.2n3/ converges absolutely as can be seen by comparing it to the p4n3
4n
series with p D 2 since, for n 3, one has n.2n1/.2n3/
< n.2nn/.2nn/
D n42 .
So, the series with parentheses added converges, and since each set of parentheses
contains a maximum of three terms, and the terms of the original series approach 0,
this means that the original series converges.
Of course, if the number of terms enclosed by sets of parentheses is not bounded,
one cannot draw the same type of conclusions. The series .1/ C . 12 C 12 12 12 / C
. 13 C 13 C 13 13 13 13 / C . 41 C 14 C 14 C 14 14 14 14 14 / C converges, but
if parentheses are removed, the series diverges even though its terms do approach 0.
The partial sums oscillate between 1 and 2.
nD1
1
P
nD1
nD1
cn diverge to infinity.
nD1
nD1
nD1
226
7 Infinite Series
Suppose you are given a target limit L. You have isolated the positive terms of
the series, the bn terms, and the negative terms of the series, the cn terms, so you can
play a cute game by taking a few bn terms such that the sum of those terms exceeds
L, and then subtract off a few cn terms until the sum decreases below L. You can then
add on more bn terms to make the sum again exceed L, and subtract of a few cn terms
until the sum decreases below L. Thus, by alternating between adding on bn terms
and subtracting off cn terms, you can arrange for the resulting series to have limit
L. More precisely, construct a new series inductively as follows: select u1 so that
u1
P
bn > L. This is always possible because the series with bn terms diverges to
nD1
u2
P
bn
nD1
v2
P
bn
nD1
nD1
u2
P
u1
P
bn
v1
P
v1
P
cn < L.
nD1
cn > L, and
nD1
nD1
selected uk and vk , select ukC1 and vkC1 to be the least positive integers such that
uP
vP
vk
kC1
kC1
P
bn
cn > L, and
bn
cn < L. It is then the case that the series
uP
kC1
nD1
nD1
nD1
nD1
b1
b2
b3
b4
b5
c3
c2
b6
b7
c1
b8
b9
c6
c5 c4
b8
c8
c7
PROOF: Let
1
P
227
nD1
1
P
nD1
1
P
nD1
1
P
bn and
nD1
1
P
.bn cn / D
nD1
1
P
nD1
nD1
nD1
nD1
infinity.
The Limit of Terms Test shows that lim an D 0 and, thus, lim bn D
n!1
n!1
lim cn D 0.
n!1
1
P
Because
bn is unbounded, there is a least positive integer, u1 , such that
u1
P
nD1
bn > L.
nD1
Because
u1
P
nD1
bn
1
P
nD1
v1
P
nD1
Having selected uk and vk for some k 1, let ukC1 > uk be the least
uP
v1
kC1
P
positive integer such that
bn
cn > L. Then let vkC1 > vk be the
nD1
uP
kC1
nD1
bn
nD1
vP
kC1
nD1
nD1
u1
P
cn > L.
nD1
dn be given by
nD1
c1 ; c2 ; c3 ; : : : ; cv1 ,
terms b1 ; b2 ; b3 ; : : : ; bu1
228
7 Infinite Series
1
P
nD1
nD1
Given > 0 there is an N1 u1 such that if n > N1 , then bn < , and there
is an N2 such that if n > N2 , then cn < .
Then there is a k1 such that uk1 > N1 and a k2 such that vk2 > N2 .
Let k D max.k1 ; k2 /, and let N D uk C vk .
Then for all m > N, either there is an r such that dm D bp for some p with
N1 < ur < p urC1 or
is an s such that dm D cq for some q with N2 <
there
m
P
vs < q vsC1 . Thus,
dn L is bounded by either max cvr ; burC1 <
nD1
or by max .bus ; cvs / < .
1
P
This shows that
dn converges to L implying that a rearrangement of the
series
1
P
nD1
an converges to L as claimed.
nD1
This theorem takes care of the case of conditionally convergent series, but what
happens when terms of an absolutely convergent series are rearranged? The answer
is that nothing happens; that is, every rearrangement of an absolutely convergent
1
P
series converges to the same limit. Suppose, for example, the series
an is
absolutely convergent with rearrangement
1
P
bn . Because
nD1
1
P
nDNC1
1
P
k
P
1
P
nD1
nD1
nD1
bn is a rearrangement of
nD1
1
P
an ,
nD1
there is an integer K such that all the terms a1 ; a2 ; a3 ; : : : ; aN are among the terms
k
k
P
P
b1 ; b2 ; b3 ; : : : ; bK . So, if k K, by how much can
an and
bn differ? Both
nD1
nD1
sums contain the terms a1 ; a2 ; a3 ; : : : ; aN , so the two sums differ only by a finite
1
P
number of the terms aNC1 ; aNC2 ; aNC3 ; : : : which add to at most
jan j < .
nDNC1
This shows that the series and its rearrangement have partial sums within of each
other and completes the argument.
PROOF: Let
1
P
229
nD1
1
P
nD1
1
P
nD1
1
P
1
P
Since
nD1
that if k N,
k
P
nD1
1
P
an .
nD1
1
P
nD1
jan j < .
nDNC1
Since
1
P
bn is a rearrangement of
nD1
1
P
nD1
not in
k
P
nD1
nD1
k
P
nD1
nD1
a1 ; a2 ; a3 ; : : : ; aN ,
k
P
an .
nD1
Thus, given > 0, there is a K such that for all k K, the k partial sum of
1
1
P
P
an and the kth partial sum of
bn are within of each other.
nD1
1
P
nD1
k
P
k N1 ,
an L < 2 .
Because
nD1
(continued)
230
7 Infinite Series
k
k
P
P
Also, there is an N2 such that if k N2 ,
an
bn < 2 .
k nD1
nD1 k
k
P
P
an L < 2 C 2 D .
nD1
1
P
nD1
1
P
nD1
7.7.3 Exercises
1. In which of the following series can the parentheses be removed without affecting
the convergence of the series?
1
1
(a) 1 12
C
C 12 13 13
C 13 C 13 14 14 14
2
3
1
1
1
1
1
1
1
1
C
C
C
C
4
4
4 5 5
5 5
4
1
1
C
12
(b) 12 14 C 16 18 C 10
1 1 1 1 1 1
(c) 2 2 C 2 2 C 2 2 C
1 1
1
1
1
1
C
C
9 C 10
C 11
12
13
(d) 12 13 C 14 C 15 16 17
8
1
1
1
1
1
1
C 16 C 17 18 19 C
14 15
1
C
(e) .1/C 1 12 C 1 12 14 C 1 12 14 18 C 1 12 14 18 16
2. Write a proof to show that if
1
P
nD1
there is a rearrangement of the terms of the series that diverges to infinity and a
rearrangement that diverges to negative infinity.
3. Write a proof to show that if a1 , a2 , and a3 are real numbers, the series a11 C a22 C
a3
C a41 C a52 C a63 C converges if and only if a1 C a2 C a3 D 0.
3
1
1
P
P
an is an absolutely convergent series, and
bn
4. Write a proof to show that if
nD1
1
P
nD1
an bn converges.
nD1
1
P
nD1
an and
1
P
nD1
bn where
1
P
an bn diverges.
nD1
6. Using the method described in this section find the first 20 terms of the
rearrangement of the series 1 1 C 12 12 C 13 13 C 14 14 C that converges
to 1.
231
1
P
an converges to L and
nD1
1
P
1
P
bn converges to M, then
nD1
nD1
nD1
nD1
1 P
1
P
an bp , and some
nD1 pD1
sense can be made out of this expression. The notation suggests that for each n,
1
P
one would calculate a limit of
an bp , and then one would consider the series of
pD1
those limits. This raises interesting questions about whether that limit, if it should
1 P
1
P
exist, has anything to do with the similar looking
an bp . In fact, as seen in
pD1 nD1
the exercises, there are examples where interchanging the order of summation in a
double summation can result in a different limit.
1 1
P
P
A simpler approach is to group the terms of the product
an
bn in a
nD1
nD1
way that might allow you to calculate the sum. One strategy is to group the terms
an bp where n C p is a given constant. For example, when the constant is 2, there
is only one term a1 b1 . When the constant is 3, there are two terms a1 b2 C a2 b1 .
n1
P
ap bnp . This
In general, the grouping of the terms whose subscripts add to n is
pD1
!
1
n1
P
P
gives what is known as the Cauchy product of the two series
ap bnp .
nD2
pD1
1
P
nD1
1
P
1
P
bn and
nD1
1
P
1
P
an
nD1
an .
nD1
For example, what is the Cauchy product for the square of the geometric series
n1
P
1
? Here you have two identical series where an D bn D 21n , so
ap bnp D
2n
nD1
n1
P
pD1
pD1
1
2p
1
2np
n1
P
pD1
1
2n
n1
.
2n
1
P
nD2
1
X
n1
nD2
2n
n1
.
2n
232
7 Infinite Series
2S D
1
X
n1
nD2
2n1
1
X
n
n
2
nD1
1 X 1
2S S D C
D1
2 nD2 2n
SD1
This Cauchy product converges to 1 which is the expected limit since
1
P
nD1
1
2n
D 1.
But Cauchy products do not always behave so nicely. For example, find the Cauchy
1
1
P
P
.1/n
.1/n
p
p
product of the two series
and
. The Alternating Series Test shows
n
nC4
nD1
nD1
that both of these series converge, but the Integral Test shows that neither converges
absolutely. The nth term of the Cauchy product of these two series is
n1
n1
P .1/p .1/np
P
p 1
p p
D .1/n
. For even values of n, this is a sum of n1
p
npC4
p.npC4/
pD1
pD1
when p D
nC4
,
2
1
.
p.npC4/
p2
nC4
2
pnC4
D
4
.
nC4
which approaches 4 as
This means that the sum is greater than or equal to 4.n1/
nC4
n gets large. Thus, the terms of the Cauchy product do not approach 0 as n goes
to infinity, and the Limit of Terms Test shows that the Cauchy product does not
converge.
This last example shows what can go wrong with the Cauchy product of two
conditionally convergent series, but the results are better when at least one of
the series is absolutely convergent. For example, if both series are absolutely
convergent, then the Cauchy product is absolutely convergent to the product of the
series. To see why this is, just consider the difference between a partial sum of
the Cauchy product of the two series and the product of two partial sums of the
individual series. That is, let k1 and k2 be positive integers, and find the difference
!
k1P
Ck2 n1
P
between the .k1 C k2 /th partial sum of the Cauchy product,
ap bnp ,
nD2
pD1
and the product of the k1 th and k2 th partial sums, respectively, of the two series,
k1
k2
P
P
am
bn . These are both just finite sums where the Cauchy product partial
mD1
nD1
sum includes all the terms am bn where the sum of the subscripts of m C n add to
something less than or equal to k1 C k2 and the other sum includes all the terms
am bn where M k1 and n k2 . Thus, the
is the
difference
sum of the remaining
k1 Ck
k1 Ck
k1 Ck
k1 Ck
P2 1
P2 m
P2 1
P2 n
am
bn
terms
bn C
am . So by choosing k1 and
mDk1 C1
nD1
nDk2 C1
k1 Ck
P2 1
mDk1 C1
mD1
k1 Ck
P2 1
am and
nDk2 C1
233
Let
1
P
am and
mD1
nD1
1
P
jam j <
mDN1
2 1C
!.
1
P
jbn j
nD1
1
P
Similarly, because
nD1
such that
1
P
jbn j <
nDN2
!.
1
P
2 1C
jam j
mD1
k1P
Ck2 n1
k1
k2
P
P
P
ap bnp
am
bn D
nD2 pD1
mD1
nD1
k1 Ck
k1 Ck
k1 Ck
k1 Ck
P2 m
P2 1
P2 n
P2 1
a
bn
bn C
am
mDk1 C1 m
nD1
nDk2 C1
mD1
k1 Ck
P2 1
mDk1 C1
1
P
mDk1 C1
1
P
jam j
jam j
k1 Ck
P2 m
nD1
1
P
jbn j C
nD1
jbn j
nD1
jbn j C
2 1C
1
P
!
jbn j
k1 Ck
P2 1
nDk2 C1
1
P
nDk2 C1
1
P
jbn j
1
P
jbn j
k1 Ck
P2 n
jam j
mD1
jam j
mD1
jam j
mD1
2 1C
nD1
1
P
mD1
!
jam j
<
2
2
D .
Therefore, since the .k1 C k2 /th partial sum of the Cauchy product of the
two series and the product of the k1 th and k2 th partial sums of the two series
are within of each other when k1 and k2 are large, both expressions must
converge to the same quantity when k1 and k2 approach infinity, and that
limit is the product of the two series.
n1
1
P
P
nDk2 C1
nD1
mDk1 C1
234
1
P
7 Infinite Series
Another way! to think about this theorem is that the Cauchy product
n1
1
1
P
P
P
ap bnp and the product of the two series
am
bn are rearrangements
nD2
pD1
mD1
nD1
of each other. Thus, if either converges absolutely, both converge absolutely to the
same limit. Of course, to make this rigorous, one would need to find at least one
rearrangement of the terms into a form c1 C c2 C c3 C and then show that that
series converges absolutely.
If one series converges absolutely and the other only converges conditionally,
then the Cauchy product of the two series still converges to the product of the two
series, but absolute convergence is not guaranteed. The proof is similar to the proof
of the previous theorem in that it carefully considers the difference between the
partial sum of the Cauchy product and product of the two series. This difference can
be broken into three differences each of which can be bounded. Specifically, assume
1
1
P
P
that
am is absolutely convergent and
bn is convergent. Then consider the
mD1
nD1
difference between the Nth partial sum of the Cauchy product and the product of the
N n1
1
1
N1
1
Nm
P
P
P
P
P
P
P
ap bnp
am
bn D
am
bn
am
series. That difference is
nD2 pD1
mD1
nD1
mD1
nD1
mD1
Nm
N1
1
1
N1
1
1
P
P
P
P
P
P
P
bn D
am
bn
bn C
am
am
bn . In the second term
nD1
mD1
nD1
nD1
mD1
mD1
nD1
N1
1
1
P
P
P
bn is fixed, and the factor
am
am can be made
of this sum, the factor
nD1
mD1
mD1
Nm
P
nD1
nD1
bn
1
P
nD1
nD1
large, or by keeping m small. Both of these can be done, but not at the same time.
The technique one would use here would be to break the sumfrom m D 1 to m
D
N1
Nm
1
P
P
P
am
bn
bn D
N 1 at some intermediate value K < N 1 writing
mD1
nD1
nD1
K
Nm
1
N1
Nm
1
P
P
P
P
P
P
am
bn
bn C
am
bn
bn . You can now choose K
mD1
nD1
nD1
mDKC1
nD1
nD1
so that for m > K the value of am is small, and when m K, the N m is large so
Nm
1
P
P
that
bn
bn will be small. This gives the following proof.
nD1
nD1
235
1
P
PROOF: If
mD1
1
P
bn is a
nD1
convergent series, then the Cauchy product of the two series converges
to the product of the two series.
1
P
Let
mD1
1
P
bn be a convergent
nD1
series.
Then for integers N and K with 1 < K < N 1, the difference between the
Nth partial sum of the Cauchy product of the two series and the product of
the two series is
N n1
1
1
P
P
P
P
ap bnp
am
bn D
nD2 pD1
N1
Nm
P
P
mD1
1
P
nD1
1
P
bn
am
bn D
nD1
mD1
nD1
N1
1
Nm
1
1
P
P
P
P
P
am
bn
bn C
am
am
bn D
mD1
nD1
nD1
mD1
mD1
nD1
K
Nm
1
N1
Nm
1
P
P
P
P
P
P
am
bn
bn C
am
bn
bn C
mD1
nD1
nD1
nD1
nD1
mDKC1
N1
1
1
P
P
P
am
am
bn :
am
mD1
N1
P
mD1
mD1
Because
nD1
T
P
1
P
P
there is a number M such for all T,
bn
bn < M.
nD1
nD1
Let > 0 be given.
1
P
Because
am converges absolutely, there is an integer K such that
nD1
1
P
mD1
jam j <
mDKC1
.
3M
1
P
bn converges to
bn , there is a positive integer N1 such that
nD1
nD1
N
1
P
P
!.
for all N N1 ,
bn
bn <
1
Because
N
P
nD1
nD1
3 1C
jam j
mD1
1
P
am converges to
am , there is a positive integer N2 such that
mD1
mD1
N
1
P
P
for all N N2 ,
am
am <
1 ! .
P
Because
N1
P
mD1
mD1
3 1C
nD1
bn
(continued)
236
7 Infinite Series
P
N1
Nm
1
1
1
1
P
P
P
P
P
P
P
N n1
Then
ap bnp
am
bn D
am
bn
am
bn D
nD2 pD1
mD1
nD1
mD1
nD1
mD1
nD1
Nm
Nm
N1
1
P
1
N1
1
1
P
P
P
P
P
P
P
P
K
a
b
b C
am
bn
bn C
am
am
bn
nD1
nD1
mD1
mD1
nD1
mDKC1
N1
1
Nm
Nm
K
1
N1
1
1
P
P
P
P
P
P
P
P
P
jam j
bn
bn C
jam j
bn
bn C
am
am
bn <
mD1
nD1
nD1
nD1
nD1
mD1
mD1
nD1
mDKC1
K
1
P
P
! C MC
1 !
jam j
bn < 3 C 3 C 3 D .
1
3M
P
P
mD1
3 1C
3 1C
jam j
mD1
the series
1
P
am
mD1
nD1
bn
nD1
N Nn
P
P
nD2 pD1
bn .
nD1
Cauchy products play a particularly useful role in the study of power series, a
topic covered in the next chapter.
7.8.1 Exercises
1. Let am;n be the nth number in the mth row of the following table where m and n
both range from 1 to infinity.
1 1
0
12 12
0
14 14 14 14
0
1
2
1
2
1
4
1
4
1
4
1
4
1
8
1
8
1
8
1
8
Show that
1 P
1
P
mD1 nD1
1
8
1
8
1
8
1
8
18 18 18 18 18 18 18 18
1 P
1
P
am;n .
nD1 mD1
2. Show that the Cauchy product for the square of the conditionally convergent
1
P
.1/n
series
converges.
n
nD1
3. Show that the Cauchy product for the square of the series
1
P
nD1
.1/n
p
n
diverges.
4. Suppose you have two series whose indices begin with 0 rather than 1 as in
1
1
P
P
an and
bn . Show that the Cauchy product of these two series is then
nD0
1 P
n
P
nD0 pD0
nD0
ap bnp .
237
5. In the next chapter it will be shown that for all real values of x, the exponential
1 n
P
x
function has the series representation ex D
. Use the Cauchy product of
n
series to show that ea eb D eaCb .
nD0
Chapter 8
Sequences of Functions
n!1
lim inf an , and subsequences <anj > of such a sequence. If instead of requiring
n!1
the terms of the sequence an to be constants, the an were allowed to depend on
the value of a variable as in fn .x/, then the sequence is a sequence of functions.
Thus, for each value of x, if all the functions fn .x/ are defined at x, then there is
a sequence of real numbers, <fn .x/>. This sequence changes as x changes, and,
indeed, there is a different sequence of real numbers for each choice of x. The limit
of the sequence, if it exists, could be different for each x, and, therefore, the limit
would also be a function, f .x/. The first question that arises is, what is meant by
the convergence of such a sequence? In fact, there are many different definitions
for the convergence of a sequence of functions, each with its own applications and
properties. The next question is, what can one say about the properties of the limit
of the sequence? For example, under what conditions can you know that the limit
function is continuous, differentiable, or integrable? In particular, if the sequence
of integrable functions <fn .x/> converges to an integrable function f .x/, when can
Rb
Rb
you conclude that lim fn .x/dx D f .x/dx?
n!1 a
converges pointwise on the interval .1; 1. When jxj < 1, the powers xn get small
as n gets large so lim xn D 0. But when x D 1, the powers xn D 1, so the limit
n!1
239
240
8 Sequences of Functions
-1
0 if 1 < x < 1
of the sequence is 1. Thus, the limit function is f .x/ D
: Note
1 if x D 1
that this is a sequence of continuous functions that converges to a function that is not
continuous. The sequence does not converge at x D 1 because the terms oscillate
between 1 and 1 (Fig. 8.1).
Continuity is not the only property not preserved by functions converging
nC1
pointwise. The terms of the sequence fn .x/ D jxj n are differentiable functions
for all real numbers, but the limit of the sequence is the function f .x/ D jxj
which
is not differentiable9at x D 0 (Fig. 8.2). The terms of the sequence f .x/ D
8
2
n
x
if 0 x 1n >
>
>
>
>
=
<
R2
2
1
2
all have integral fn .x/dx D 1, yet the sequence
n2 . n x/ if n < x < n
>
>
0
>
>
>
;
:
2
0
if n x 2
converges pointwise on the interval 0; 1 to the function f .x/ D 0 which has integral
equal to 0 (Fig. 8.3).
241
f2
f1
8.1.1 Exercises
Determine the pointwise limits of the following sequences of functions. For which
sequences is the limit continuous? For which sequences is the limit of the integrals
of the terms equal to the integral of the limit?
p
1. fn .x/ D n x for x 2 0; 16.
2. Let r1 ; r2 ; r3 ; : : : be a sequence
consisting of all the rational numbers in the
1 if x D rk for some k n
interval 0; 1. Let fn .x/ D
for x 2 0; 1.
0 otherwise
n
nx for x 2 .0; 1/.
3. fn .x/ D (
)
n
nC4
n if 2nC4
< x < 2nC4
4. fn .x/ D
for x 2 0; 1.
0 otherwise
8
9
1
.2 C .1/n / n2 x
if 0 < x < 2n
<
=
1
5. fn .x/ D .2 C .1/n / n2 1n x if 2n
x < 1n for x 2 0; 1.
:
;
0
otherwise
242
8 Sequences of Functions
the set A where convergence lags behind. For any region of width > 0 around
the limit function, all of the terms suitably far down the sequence enter that region.
Compare, for example, the uniformly convergent sequence depicted in Fig. 8.4 with
the pointwise convergent sequences depicted in Figs. 8.1 and 8.3. In Fig. 8.4 the
functions of the sequence get close to the limit function for all the values of x,
whereas for each function in Figs. 8.1 and 8.3 there is an x for which the function
is far from its limit. Clearly, if a sequence of functions converges uniformly, then it
also converges pointwise. Thus, to converge uniformly is a stronger condition than
to converge pointwise.
As seen in the previous section, the terms of the sequence <fn > can have many
properties that are not automatically inherited by the limit of the sequence, f , when
the convergence is pointwise. Under uniform convergence, more of the properties
of the terms of the sequence are retained by the limit. This is because under uniform
convergence there are no points x 2 A for which the values of <fn .x/> lag behind
as n gets large. For all values of x 2 A, the sequence of <fn .x/> values get close to
the corresponding f .x/ at a rate at least as fast as some fixed rate.
For example, if <fn > is a sequence of functions continuous on the set A which
converges uniformly to f on A, then the limit is guaranteed to be continuous.
Actually, a stronger statement can be made. If all the terms fn are continuous at
some point a 2 A, then the limit, f , will also be continuous at a. To prove that f is
continuous at a, you will need to show that for each > 0 there is a > 0 such
that if x is in A with jx aj < , then jf .x/ f .a/j < . How can you arrange for
f .x/ to be close to f .a/? What you know is that the functions fn get uniformly close
to f , and that the fn functions are continuous at a. Since, for any particular n, the
term fn is continuous at a, you can arrange for fn .x/ to be close to fn .a/. The uniform
convergence allows you to choose an integer n so that for every x 2 A, fn .x/ is close
243
f(x)
fn(x)
)
a
to f .x/. That is, jf .x/ f .a/j D jf .x/ fn .x/ C fn .x/ fn .a/ C fn .a/ f .a/j
jf .x/ fn .x/j C jfn .x/ fn .a/j C jfn .a/ f .a/j. Each of these three terms can be made
small, say less than 3 , so that the sum is less than . The key point here is that only
one value of n needs to be chosen so that jf .x/ fn .x/j can be made less than 3 no
matter which x is chosen (Fig. 8.5).
PROOF: If the sequence <fn > converges uniformly to the limit f on the set
A and if for each n, fn is continuous at a 2 A, then f is continuous at a 2 A.
In particular, if each fn is continuous on A, then f is continuous on A.
Let <fn > be a sequences of functions that converge uniformly to the
function f on a set A.
Assume that each fn is continuous at point a 2 A.
Let > 0 be given.
Because the sequence converges uniformly, there is an integer N such that
jfn .x/ f .x/j < 3 for all x 2 A and all n N.
Because fN is continuous at a, there is a > 0 such that jfN .x/ fN .a/j < 3
for all x 2 A satisfying jx aj < .
Then, for all x 2 A satisfying jx aj < , it follows that
jf .x/ f .a/j D jf .x/ fN .x/ C fN .x/ fN .a/ C fN .a/ f .a/j
jf .x/ fN .x/j C jfN .x/ fN .a/j C jfN .a/ f .a/j < 3 C 3 C 3 D .
Therefore, the function f is continuous at a 2 A.
Moreover, if each function fn is continuous at each a 2 A, then f is
continuous at each x 2 A, so f is continuous on A.
It is worth considering where this proof breaks down if all you assume is that
the sequence <fn > converges pointwise to f . The problem comes in the fact that
although jfN .a/ f .a/j and jfN .x/ fN .a/j can be made smaller than 3 , there could
244
8 Sequences of Functions
be values of x very close to a for which jfN .x/ f .x/j is no longer small. Thus, the
needed inequality jf .x/ fN .x/j C jfN .x/ fN .a/j C jf .a/ f .a/j < might not
hold. Also consider the function f defined on the interval 0; 2 by f .x/ D x if x 1
and f .1/ D 3. If for each positive integer n you let fn .x/ D f .x/ C 1n , then it is
clear that the sequence <fn > converges uniformly to f . At the points where each fn
is continuous, that is, for x 1, the limit function f is also continuous.
Suppose that <fn > is a sequence of functions Riemann integrable on an interval
a; b and that this sequence converges to a limit f . Examples in the last section show
Rb
that if the convergence is pointwise, the limit lim fn .x/dx does not necessarily
equal
Rb
n!1 a
integrable, and the limit of the integrals of the fn might not exist. On the other
hand, if the convergence is uniform, then the limit function f will be Riemann
integrable and the limit of the integrals of the fn will equal the integral of f . Showing
that the uniform limit of Riemann integrable functions is Riemann integrable is not
difficult and is based on the characterization of Riemann integrable functions given
by Lebesgues Theorem. Recall that a function is Riemann integrable on an interval
if and only if it is bounded and the set of points where the function is discontinuous
has measure zero. If each term of the sequence, fn , has these properties, then the limit
function, f , must also have them. By the definition of uniform convergence, there is
an integer N such that jfN .x/ f .x/j < 1 for all x 2 a; b. So, if the function fN is
bounded by some constant M, then the function f must be bounded by M C 1 since
for all x 2 a; b it follows that M 1 fN .x/ 1 < f .x/ < fN .x/ C 1 M C 1.
As for points of discontinuity of f , for each positive integer n, let Dn be the set of
points in a; b where the function fn is discontinuous. Because each fn is Riemann
1
of sets of measure zero, so it also has measure zero. The sequence <fn > converges
uniformly on the set A D a; bnD, and each term of the sequence is continuous at
each point of A, so, by the preceding theorem, the limit f must be continuous
on A. Thus, the set of discontinuities of f must be contained in D, so the set of
discontinuities of f has measure zero. Therefore, f is Riemann integrable.
Rb
Rb
So why does it follow that lim fn .x/dx D f .x/dx? From the definition of
n!1 a
uniform convergence, for every > 0 there is an integer N such that n N implies
that jfn .x/ f .x/j < for every x 2 a; b. This means that for every n N and
Rb
every x 2 a; b it follows that f .x/ < fn .x/ < f .x/ C , so f .x/dx .b a/ D
Rb
a
.f .x//dx
Rb
a
Rb
Rb
fn .x/dx .f .x/C/dx D
Rb
n!1 a
fn .x/dx D
Rb
a
f .x/dx.
245
PROOF: Assume that <fn > is a sequence of functions that are Riemann
integrable on the interval a; b. If the sequence converges uniformly to f ,
Rb
Rb
then f is also Riemann integrable on a; b, and lim fn .x/dx D f .x/dx.
n!1 a
Rb
f .x/dx C .
Rb
n!1 a
fn .x/dx D
Rb
theorem.
. The
Note the use of b a C 1 rather than just b a in the denominator of baC1
extra C1 avoids the embarrassing case of a D b. The addition of C1 allows the
proof to handle the easy special case without having to provide a separate argument
for it.
246
8 Sequences of Functions
8.2.1 Exercises
1. Show that lim xn converges uniformly to 0 on any interval a; a where 0 <
n!1
a < 1.
1
converges uniformly to 0 on any interval a; 1/ for a > 0 but
2. Show that lim nx
n!1
247
Combining these you can show that for any y in some interval around x, the value
of fN .y/ is close to the value of f .y/. The crucial observation here is that once you
know that fN .y/ and f .y/ are close, the monotonicity of the convergence gives you
that fn .y/ is between fN .y/ and f .y/ for all n N, and, thus, fn .y/ will be close to
f .y/ for all n N. Note, though, that the value of N can vary with the value of x.
Well, this means that for each x 2 a; b there is an interval around x where
fn .y/ is close to f .y/ for all y in the interval and all n N. Now you can use the
compactness of the interval a; b, that is, you can use the HeineBorel Theorem to
show that there is a finite collection of these x values, say x1 ; x2 ; x3 ; : : : ; xk , such that
the entire interval a; b is covered by these intervals you constructed around each
of the xj s. Each of the xj s was associated with an Nj , and now one can select the
maximum of these Nj values to get a single function fN which is uniformly close
to f . Again, because the convergence is monotone, once you know that fN is close to
f , you know that fn is close to f for all n N. This will complete the proof.
PROOF: Assume that <fn > is a sequence of functions continuous on the
interval a; b that converges monotonically to the function f that is also
continuous on a; b. Then the sequence converges uniformly to f on a; b.
Assume that <fn > is a sequence of functions continuous on the interval
a; b that converges monotonically to the function f that is also continuous
on a; b.
Let > 0 be given.
Let x 2 a; b.
Because the function f is continuous at x, there is a 1 > 0 such that if
y 2 a; b with jy xj < 1 , then jf .y/ f .x/j < 3 .
Because lim fn .x/ D f .x/, there is an integer Nx such that if n Nx , then
n!1
248
8 Sequences of Functions
p
n!1 a
here is based on the fact that for any Riemann integrable function, fn , you can find
upper and lower step functions, un .x/ and vn .x/, satisfying vn .x/ fn .x/ un .x/
Rb
for every x 2 a; b so that a .un .x/ vn .x//dx is as small as you like. Suppose
Rb
you select un and vn so that a .un .x/ vn .x//dx < 2n . That is, find upper and
lower step functions for each fn such that they give increasingly better and better
1
This proof is based on ideas from the article Monotone Convergence Theorem for the Riemann
Integral by Brian S. Thomson from the American Mathematical Monthly, JuneJuly 2010.
249
1
P
nD1
n
2
nD1
proof, but later this value can be adjusted when you see just how small the bound
needs to be in order to make the proof work.
For each n these un and vn functions are step functions on the interval a; b, so
there must be a positive integer k and a partition of a; b given by a D x0 < x1 <
x2 < < xk D b such that both the un and the vn functions are constant on each of
the open intervals .xj1 ; xj / for j D 1; 2; 3; : : : ; k. Well, for this proof, there will be a
different k and a different partition for each fn function, so it would be better to name
the positive integer kn and the partition a D xn;0 < xn;1 < xn;2 < < xn;kn D b.
For the purposes of this proof, it is important that the endpoints of the partition
associated with the un and vn , that is, xn;1 ; xn;2 ; xn;3 ; ; xn;kn 1 , do not match any
of the endpoints of the partition associated with the next case unC1 and vnC1 . This
is easy to arrange because an upper or lower step function for fn that is constant
on the two intervals .xj1 ; xj / and .xj ; xjC1 / can be altered to be constant on the
three intervals .xj1 ; xj /, .xj ; xj C /, and .xj C ; xjC1 / for some suitably
small > 0 without significantly changing the value of the integral of the step
function and without destroying whether the step function is an upper or lower step
function of fn . Indeed, if, for example, the upper step function un were constant on
the function un by redefining
un on the
.xj1 ; xj / and .xj ; xjC1 /, you could define
interval .xj ; xj C / to equal max un .xj /; un .xj /; un .xj C / . Then un would
Rb
be slightly larger than un on a small interval so that un .x/ un .x/ dx is less than
a
2un .xj / which can be made arbitrarily small by selecting small. Because un is
greater than or equal to un , it is also an upper step function of fn .
For each y 2 a; b, consider the sequence of numbers f1 .y/; f2 .y/; f3 .y/; : : : which
decreases monotonically to 0. For each y, select a positive integer n.y/ such that
fn.y/ .y/ < . Thus, n.y/ associates a term of the sequence fn.y/ with y. That term is
also associated with un and vn and the partition a D xn;0 < xn;1 < xn;2 < <
xn;kn D b. As stated in the previous paragraph, it can be assumed that y is not equal
to any of the endpoints xn;1 ; xn;2 ; xn;3 ; : : : ; xn;kn 1 , so y can be associated with an open
interval .xn;j1 ; xn;j / containing y unless y is a or b in which case y will be associated
with the open interval .a 1; xn;1 / or .xn;kn 1 ; b C 1/, respectively.
Each point y 2 a; b has been associated with an open interval that contains y.
Thus, these open intervals provide an open cover of the interval a; b. The Heine
Borel Theorem says that there exists a finite subcover of a; b. That is, there is a
sequence of y 2 a; b, say y1 ; y2 ; y3 ; : : : ; ym , such that the intervals associated with
these y values cover a; b. Something stronger can be said. In this finite subcover
of open intervals, you can assume that there are no values of y 2 a; b that belong
to more than two of the open intervals in that subcover. Indeed, suppose y is an
element of the three intervals of the subcover .a1 ; b1 /, .a2 ; b2 /, and .a3 ; b3 /. Suppose
a1 is the least of a1 , a2 ; and a3 , and that b2 is the greatest of b1 ; b2 ; and b3 . Then
a1 < a3 < y < b3 < b2 , so .a3 ; b3 / .a1 ; b1 /[.a2 ; b2 /, and the interval .a3 ; b3 / can
250
8 Sequences of Functions
be dropped from the subcover. Because the subcover contains only a finite number of
open intervals, all of these superfluous intervals can be dropped from the subcover.
Consider the intervals associated with each of the yj values. For simplicity, let the
interval associated with yj be renamed .aj ; bj /. Note that the endpoints a and b will
be among the yj values because for every n each of these endpoints was covered by
only one possible open interval. At this point the left endpoint associated with a can
be set to a and the right endpoint of the interval associated with b can be set to b. It
is important to note that if n.yi / D n.yj / for some distinct i and j, then the intervals
associated with yi and yj do not overlap. This is because the intervals associated with
yi and yj are distinct intervals from .a1; x1 /; .x1 ; x2 /; .x2 ; x3 /; : : : ; .xk1 ; bC1/. Let
N be the maximum of the finitely many n.yj / values for j D 1; 2; 3; : : : ; m. Because
no value of y 2 a; b appears in more than two of the open intervals associated with
the yj , it can be concluded that
Zb
a
m Z
X
bj
fN .x/dx
jD1 a
m Z
X
bj
fN .x/dx
jD1 a
fn.yj / .x/dx D
2
2
3
3
Zbj
Zbj
m
m
X
X
6
6
7
7
fn.yj / .x/ fn.yj / .yj / dx C fn.yj / .yj /.bj aj /5
un.yj / .x/ vn.yj / .yj / dx C .bj aj /5
4
4
jD1
jD1
aj
aj
N Z
N
X
X
2p C 2.b a/ < .2b 2a C 1/:
up .x/ vp .x/ dx C 2.b a/
b
pD1 a
pD1
There were two places in the above argument where quantities were forced to be
less than the given value . It can now be seen that those quantities should have been
Rb
made smaller than 2b2aC1
so that the final inequality would show fN .x/dx < as
a
needed. It is also worth noting that there were two places in the argument that use
the fact that the sequence <fn > converges monotonically. The first was to conclude
that when, for a particular y 2 a; b, the value of fn .y/ is small, then the values of
fm .y/ are also small for all m n. The second important use of monotonicity takes
Rb
Rb
the final result that fN .x/dx < and concludes that fm .x/dx < for all m N.
a
251
Because un and vn are step functions, there exist a positive integer kn and a
partition of a; b given by a D xn;0 < xn;1 < xn;3 < < xn;kn D b such
that for each j D 1; 2; 3; : : : ; kn , the functions un and vn are constant on each
open interval .xn;j1 ; xn;j /.
Because there is flexibility in selecting the upper and lower step functions,
it can be assumed that for each positive integer n, except for a and b, the
endpoints of the partition associated with the upper and lower step functions
for fn are distinct from the endpoints of the partition associated with the
upper and lower step functions for fnC1 .
For each y 2 a; b the sequence f1 .y/; f2 .y/; f3 .y/; : : : decreases monotonically to 0, so for each y there is a positive integer n.y/ such that
fm .y/ < 2b2aC1
for all m n.y/. In particular, it can be assumed that,
unless y is a or b, y is not an endpoint of the partition of a; b associated
with the upper and lower step functions of fn.y/ .
Associate with each y 2 a; b an open interval as follows. If y D a,
then let the open interval be .a 1; xn.a/;1 /. If y D b, then let the open
interval be .xn.b/;kn.b/ 1 ; b C 1/. Otherwise, associate y with the open interval
.xn.y/;j1 ; xn.y/;j / that contains y.
Thus, each y 2 a; b is associated with an open interval that contains y, so
this collection of open intervals provides an open cover of a; b.
By the HeineBorel Theorem, there exists a finite subcovering of a; b
consisting of m open intervals associated with m values y1 ; y2 ; y3 ; : : : ; ym
in a; b. Note that since a and b are each covered by at most one of the
open intervals in the covering of a; b, both a and b appear in the list of y1
through ym .
Let the interval associated with yj be called .aj ; bj /. Reset the interval
associated with a so that its aj value is equal to a rather than a 1, and
reset the interval associated with b so that its bj value is equal to b rather
than b C 1. It can be assumed that no value of y 2 a; b belongs to more
than two of the open intervals of the subcovering.
(continued)
252
8 Sequences of Functions
Let N D max n.y1 /; n.y2 /; n.y3 /; : : : ; n.ym / .
Then
Zb
m Z
X
bj
fN .x/dx
jD1 a
m Z
X
bj
fN .x/dx
jD1 a
fn.yj / .x/dx D
3
2
Zbj
m
X
7
6
fn.yj / .x/ fn.yj / .yj / dx C fn.yj / .yj /.bj aj /5
4
jD1
aj
2
3
Zbj
m
N Zb
X
6
7 X
un.yj / .x/ vn.yj / .yj / dx C .bj aj /5
up .x/ vp .x/ dx C
4
jD1
pD1 a
aj
N
X
pD1
2.b a/
2b 2a C 1
2p C
2.b a/ <
.2b 2a C 1/ D :
2b 2a C 1
2b 2a C 1
2b 2a C 1
Pointwise convergence and uniform convergence are not the only methods of
convergence of sequences of functions. Another method suggested by the above
theorem is called convergence in mean or convergence in L1 . A sequence of
Riemann integrable functions <fn > is said to converge in mean to the Riemann
Rb
integrable function f on the interval a; b if lim jfn .x/ f .x/jdx D 0. For
n!1 a
example, consider the following sequence of functions defined on the interval 0; 1.
Define f .xI a; b/ be the function that is 1 for x in the interval a; b and 0 for all
other x. Then for positive integer n and for integer k with 2n1 k < 2n , let
n1 kC12n1
n1 kC12n1
fk .x/ D f .xI k2
; 2n1 /. The integral of f .xI k2
; 2n1 / from 0 to 1 is
2n1
2n1
1
,
so
the
integrals
of
f
.x/
approach
0
as
k
gets
large.
Thus,
fk converges in mean
k
2n1
to the zero function. Yet this sequence of functions does not converge pointwise for
any single value of x.
1
P
nD1
terms of the series are defined and the series converges, f .x/ is just an infinite series
1
P
1
is defined for each x
of real numbers given by an .x/. For example, f .x/ D
n2 Cx
nD1
that is not the negative of a perfect square. If x is the negative of a perfect square,
then there is a term of the series that is not defined. Otherwise, the series converges
253
1
n2
1
n2 Cx
1
n2 Cx
1
,
n2
2
2n2 2jxj
nD1
converges.
1
P
nD1
x
converges to f .x/ D 1x
for all x satisfying jxj < 1. Note here that the function
x
f .x/ D 1x is defined for all x 1, but the infinite series is only defined for jxj < 1.
This is an example of a power series dealt with in considerably more detail in the
next section.
The results concerning the convergence of sequences of functions discussed
earlier in this chapter apply to the study of infinite series of functions because
an infinite series is just defined to be the sequence of its partial sums. Still other
questions arise such as, can one find the derivative or the integral of an infinite series
by simply differentiating or integrating the terms of the series and then finding the
limit of the resulting partial sums? The answer to this question is that sometimes one
gets a correct answer by differentiating or integrating a series term by term, but other
times this process results in nonsense. For example, consider again the function
1
1
1 R
R
R P
P
P
1
1
1
f .x/ D
.
Here,
the
statement
that
f
.x/
dx
D
dx
D
dx
2
2
n Cx
n Cx
n2 Cx
nD1
1 R
P
nD1
1
n2 Cx
dx D
1
P
nD1
nD1
ln n2 C x C C which does not converge
nD1
for any value of x. Alternatively, for this particular series it is valid to use the
1
1 Ry
Ry
Ry P
P
1
1
definite integral from 0 to y and write f .x/ dx D
dx
D
dx D
2
n Cx
n2 Cx
nD1 0
0
0 nD1
1
P
2
which does converge for each y > 1. The integral and derivative of
ln n nCy
2
nD1
1
P
nD1
1
P
1
P
nD1
nD1
1
P
nD1
in known as the Weierstrass M-Test. Consider how the proof of this result would
1
P
go. First, of course, you would assume that you had a series of functions,
an .x/,
and a convergent series of positive numbers,
1
P
nD1
nD1
integer n, jan .x/j Mn for every x 2 A. You should note that for each x 2 A, the
254
8 Sequences of Functions
series
1
P
nD1
1
P
an .x/ converges
nD1
pointwise. You are to prove that the sequence of function converges uniformly, so
you would need to take an > 0 and show that there is an integer N such that
m
P
whenever m N and x 2 A, the partial sum
an .x/ is within of the limit
nD1
1
P
1
P
an .x/. The difference between the mth partial sum of
an .x/ and its limit
nD1
nD1
1
1
1
P
P
P
is
an .x/
jan .x/j
Mn which can be made less than by
nDmC1
nDmC1
nDmC1
selecting m large. The value of m does not depend on x showing that the convergence
is uniform. This gives the following proof.
PROOF (Weierstrass M-Test): Let
defined on the set A, and let
1
P
1
P
nD1
nD1
numbers. If for each n and each x 2 A it holds that jan .x/j Mn , then
1
P
an .x/ converges uniformly on A.
nD1
Let
1
P
1
P
nD1
Mn be
nD1
1
P
an .x/
nD1
nD1
all m N.
But, then, for each x 2 A and each m N,
the difference
between the mth
1
1
1
P
P
P
partial sum of
an .x/ and its limit is
an .x/
jan .x/j
1
P
nDmC1
Thus,
nD1
nDmC1
nDmC1
Mn < .
1
P
nD1
1
P
nD1
1
n2 Cx
1
n2 Cx
1
,
n2
1
P
nD1
1
n2
converges. Since
all the partial sums of the series are continuous functions, it follows from this
255
uniform convergence that the limit function is continuous on 0; 1/. Similarly, the
1
P
sin.n2 x/
sin.n2 x/
n12
series
converges
uniformly
on
the
entire
real
line
because
n2
n2
nD1
for every positive integer n. Again, you can conclude that the limit function is
continuous because all the partial sums are continuous functions. Notice, though,
1
P
that if you differentiate each term of this series, you get
cos.n2 x/ which does not
nD1
center of the power series. This book will consider such series where the variable,
coefficients, and center are real numbers, although most of what is said here holds
when these quantities are allowed to be complex numbers. In fact, such series play
a central role in Complex Analysis.
absolutely for any x satisfying jx cj < jy cj, that is, for any x closer to c than y.
The proof is based on the Weierstrass M-Test where the power series
at the point x
which is less
is compared to a convergent geometric series with common ratio xc
yc
than 1.
256
8 Sequences of Functions
1
P
nD0
1
P
an .x
nD0
1
P
Since
nD0
Terms Test.
Thus, the terms must be bounded, and there exists a real number M such
that jan .y c/n j M for every nonnegative integer n.
Let x be any real number satisfying jx
n cj < jy cj.
n
n
n xc
Then jan .x c/ j D jan .y c/ j yc M xc
.
yc
n
1
P
The series
M xc
is a convergent geometric series with common ratio
yc
nD0
xc
yc < 1.
1
P
Thus,
jan .y c/n j converges absolutely by the Weierstrass M-Test.
nD0
a power series is R D supfjy cj the series converges at yg, and that the power
series converges absolutely for all x 2 .c R; c C R/. This does not say anything
about how the power series behaves at the end points c R and c C R. There are
examples of power series that converge at both endpoints, that converge at one of the
two endpoints, or converge at neither endpoint. It also follows from the above proof,
that if the power series converges absolutely at y, then it converges uniformly for all
x satisfying jx cj jy cj. In particular, since all the partial sums of the series are
continuous functions, if the power series has radius of convergence R > 0 and is
any positive number less than R, then the series converges absolutely at x D cCR,
so the series converges absolutely and uniformly on c R C ; c C R . As a
1
P
result, the function f .x/ D
an .x c/n is continuous on c R C ; c C R for
nD0
all small > 0, so it is continuous on the open interval .c R; c C R/. If the series
converges absolutely for x D c C R, then f .x/ is continuous on the closed interval
1
P
c R; c C R. What if the series
an .x c/n converges conditionally at x D c C R
nD0
or x D c R? Does this mean that the function is continuous at that endpoint? The
answer is yes, but this takes some proof and is known as Abels Theorem.
257
1
P
an .x c/n has
nD0
1
P
nD0
R < 1.
Assume that the series converges at one of the endpoints of the interval of
convergence, c R or c C R.
Without loss of generality c D 0 and R D 1 because the argument can be
applied to the series where x is replaced by Rx C c. Thus, assume that the
1
P
series is
an xn with radius of convergence 1.
nD0
converges at 1.
Finally, by subtracting a constant from the constant term of the series, a0 , it
1
P
can be assumed that
an D 0.
nD0
k
P
an .
nD0
1
P
sn x n
nD1
s0 C
1
P
nD1
sn x .1 x/ s0 x D .1 x/
n
nD1
1
P
nD1
nD1
sn x .
nD0
1n!1
1
N
1
P
P
P
P
n
n
n
n
Then
an x D .1x/
sn x .1x/
sn x C .1x/
sn x
nD0
nDNC1
nD0 N
nD0 N
1
P
P
P
NC1
.1x/
sn xn C.1x/
xn D .1 x/
sn xn C 2 .1x/ x1x D
2
nD0
nDNC1
nD0
N
NC1
P
n
.1 x/
sn x C 2 x
.
nD0
Because the limit of this quantity as x approaches 1 from the left is 2 , there
exists > 0 such that for all x between 1 and 1, this expression is less
than .
1
P
an xn D 0 which completes the proof.
This shows that lim
x!1
nD0
258
8 Sequences of Functions
1p
.
lim sup n jan j
1p
.
lim sup n jan j
n!1
n!1
n!1
1
P
an .x c/n , you see that the series will converge
If you apply the Ratio Test to
nD0
a .xc/nC1
nj
<
1.
Equivalently,
jx cj < lim jajanC1
. The series diverges if
if lim nC1
n
an .xc/
j
n!1
n!1
nj
nj
showing that R D lim jajanC1
. This expression is fine as long
jx cj > lim jajanC1
j
j
n!1
n!1
as the limit of the ratio of terms exists, but it is less helpful when it does not.
It is worth considering a few examples.
1
P
nxn
nD0
n!1
1
p
n
.1/n .x1/
n
lim nC1
n!1 n
nD1
1
The center is c D 1, and the radius of convergence is R D lim p
D
n 1
n!1
lim n D 1. At the right endpoint x D 0 the series is the harmonic series which
n!1 nC1
diverges to infinity, but at the right endpoint x D 2 the series is the alternating
harmonic series which converges conditionally.
1
P
.xC4/n
n2 5n
nD0
n!1
1
p
n 2 n
n 5
n!1
nD0
.2n/
n!1
lim
n!1
.2nC2/
.2n/
q1
n
1
.2n/
1
P
259
nn xn
nD0
nn
n!1
nC1
n!1 .nC1/
1
22 2
x C 32 x C 213 x3
21
1
p
n n
n
1
p
n a
n
3
2
n!1
1
p
n a
n
n
D R. Note that lim inf anC1 D 0 and lim sup .nC1/
nC1 D 1,
an
n!1
n!1
neither of which shed any light on the value of R. This series diverges at both
endpoints by the Limit of Terms Test.
8.5.3 Differentiability
A function represented by a power series in an interval with positive length is said to
be analytic in that interval. Perhaps the most unusual property of analytic functions
is that they are differentiable, and the derivative of a power series can be found by
differentiating the series term by term to get a new series which converges with the
1
P
same radius of convergence as the original series. That is, if f .x/ D
an .x c/n
0
1
P
nD0
n an .x c/
n1
nD1
series converges for these same values of x. It is easy to check that the radius of
1
P
convergence of the derivative series
n an .x c/n1 is the same as the original
nD1
p
p
p
series. This follows from the fact that lim sup n n an D lim n nlim sup n an D R1 ,
n!1
n!1
n!1
1
P
nD0
whether this new power series is, in fact, the derivative of the original series. That is,
does g.x/ D f 0 .x/ hold for all x in the open interval where the two series converge?
.x/
This needs to be proved. The proof needs to show that lim f .xCh/f
D g.x/ for
h
h!0
.x/
g.x/
of this, you might express the difference f .xCh/f
in terms of power series
h
and see if this simplifies to an expression that has a limit of 0 as h approaches 0. The
calculation is simpler if you assume that c D 0. Then,
260
8 Sequences of Functions
1
P
1
an .x C h/n
an xn X
1
f .x C h/ f .x/
nD0
nD0
n1
D
g.x/
n
a
x
n
h
h
nD1
1
1
n
P
P
P
P
n p np
n
h
a
x
a
x
hn an xn1
n
n
nD0 pD0 p
nD0
nD1
A careful accounting of the terms in the numerator shows that all the terms of
1
1
P
P
an xn and all of the terms of
hn an xn1 cancel leaving
nD0
nD1
n
P
P
n p np
!
h
a
x
1
n
nD2 n pD2 p
X
X
p2 np
an
h x :
D jhj
h
nD2 pD2 p
The factor jhj clearly goes to 0 as h goes to 0, but there is a question about what
happens to the other factor. This infinite sum will not be a problem if it remains
bounded as h gets small. Here is where you can use the fact that power series with
radius of convergence R converge absolutely at points less than a distance R from
the center of the series. Assume that jhj is smaller than some fixed value s > 0. Then
the second factor can be estimated as follows.
!
!
!
1
1
n
1
n
n
X
X
X
X X
n p2 np X
n
n p2 np
p2 np
h
jhj
s
an
x
jan j
jxj
jan j
jxj
p
p
nD2 pD2 p
nD2
pD2
nD2
pD2
!
n
1
1
X
jan j X n p np
1 X
s jxj
D 2
jan j.jxj C s/n :
2
s
s
p
nD2
pD0
nD2
This last expression converges as long as jxjCs is a point where the power series for
f converges absolutely. But if x were chosen so that jxj < R, then for any positive s
with s < R jxj, this will happen. Because you are free to choose any s > 0, you
.x/
can choose one less than R jxj which will ensure that f .xCh/f
g.x/ is small
h
whenever 0 < jhj < s, so the proof can be completed.
261
Let
1
P
nD0
R 1.
The power series for f and its derivative depend on x c and not on c, so
there is no loss of generality to assume
p that c D 0. p
p
Note that lim sup n n an D lim n n lim sup n an , so the two series
1
P
n!1
an .x c/n and
nD0
1
P
n!1
n!1
nD1
1C
1
P
s2
jan j.jxjCs/n
A > 0.
nD2
1
1
P an .x C h/n P an xn
1
1
f .x C h/ f .x/ X
X
nD0
nD0
n1
n1
nan x
n an x
D
h
h
nD0
nD1
1
1
1
1
n
n
P
P
P
P
P
P
n p np
n p np
an x n
hn an xn1
an
p h x
nD0 an pD0 p h x
nD0
nD1
nD2 pD2
D
D
h
h
!
!
1
1
n
n
X
X
X X
n p2 np
n
jhj
jhj
h
jhjp2 jxjnp
an
x
ja
j
n
p
nD2 pD2 p
nD2
pD2
(continued)
262
8 Sequences of Functions
jhj
1
X
nD2
jan j
!
!
n
1
n
1
X
X
jan j X n p np
n p2 np
1 X
s jxj
s jxj
jhj
D jhj 2
jan j.jxj C s/n < :
2
s
s
p
p
pD2
nD2
pD0
nD2
1
P
nan xn1 .
nD1
An immediate consequence of this theorem is that not only can you obtain the
1
P
first derivative of f .x/ D
an .x c/n by differentiating term by term, but you
nD0
can also get all the higher derivatives of f by repeating the process. This follows
by induction because, if the mth derivative of f is equal to the series formed by the
mth derivatives of the terms of the series for f , and if that series has the same radius
of convergence as the series for f , then the theorem says that the mC1st derivative of
f can be obtained by differentiating the terms of the series for the mth derivative of
f , and the radius of convergence of that series will remain the same. Moreover, one
can find an antiderivative for f by integrating each term of the series for f . That is, if
1
1
P
P
an
f .x/ D
an .x c/n for all x with jx cj < R, then the series
.x c/nC1 will
nC1
nD0
nD0
have the same radius of convergence as the series for f , and the theorem says that
the derivative of the new series is equal to f . It is important to note that if a function
is analytic by virtue of having a power series representation in an open interval of
radius R around c, then that function is infinitely differentiable in that interval.
These results make it very simple to derive new series from previously known
1
P
1
series. For example, you already know that 1x
D
xn for all x with jxj < 1.
nD0
1
, 1
1x 1Cx
1
, 1
1Cx 1Cx2
1
, 1
1x .1x/2
1
P
1
P
.1/n xn .
nD0
1
P
.1/n x2n .
nD0
nxn1 .
nD1
1
and noting that ln 1 D 0, ln.1 C x/ D
by integrating the series for 1Cx
1
1
P
P
nC1
n
.1/n xnC1
D
.1/n1 xn . In particular, by Abels theorem,
nD0
nD1
ln 2 D 1 12 C 13 14 C .
1
1
0 D 0, tan1 x D
by integrating the series for 1Cx
2 and noting that tan
1
P
2nC1
.1/n x2nC1 . In particular, by Abels Theorem, 4 D 1 13 C 15 17 C .
nD0
263
1
P
nD0
constant term of the series for f . Finding the mth derivative of the series for f and
evaluating it at the center of the series, c, gives that f .m/ .c/ D mam . So, for all
.m/
integers m 0, am D f m.c/ . This gives a straightforward way to generate the
power series representing any analytic function. Moreover, even if f is not infinitely
differentiable, if it is m times differentiable, one can generate the mth degree Taylor
m .n/
P
f .c/
polynomial for f centered at c given by g.x/ D
.x c/n . Then g is an mth
n
nD0
degree polynomial that is equal to f at c, and all of its derivatives up to order m agree
with the corresponding derivatives of f at c. In particular, the first degree Taylor
polynomial is just the familiar linear approximation to f given by the line tangent
to the graph of f at c. If f is m-times differentiable at c, one can generate the mth
degree Taylor polynomial, g.x/, for f centered at c, but this does not say whether
the value of g.x/ is even remotely related to the value of f .x/ when x is different
from c. This issue is what is addressed by Taylors Theorem which states that
f .x/ D g.x/ C Rm .x/ for some remainder function Rm .x/. Depending on various
characteristics of f , one can show that Rm .x/ is suitably small so that g.x/ is a good
approximation for f .x/.
There are many forms of Taylors Theorem that express the remainder term,
Rm .x/, in different ways. The one discussed here is sometimes called Lagranges
form. It says that if f is m C 1 times differentiable on the interval between c and
x, then the difference between f .x/ and the mth degree Taylor polynomial for f
centered at c can be expressed in terms of f .mC1/ ./ for some strictly between c
and x. Its proof follows easily from the following generalization of Rolles Theorem.
PROOF (Higher Order Rolles Theorem): Let f be an m C 1 times
differentiable function on the open interval from a to b with a b,
let f be continuous on the closed interval from a to b, and suppose
that 0 D f .a/ D f 0 .a/ D f 00 .a/ D D f .m/ .a/ D f .b/. Then there is an x
strictly between a and b where f .mC1/ .x/ D 0.
Let f be an m C 1 times differentiable function on the open interval from a
to b with a b, let f be continuous on the closed interval from a to b, and
suppose that 0 D f .a/ D f 0 .a/ D f 00 .a/ D D f .m/ .a/ D f .b/.
Since f .a/ D f .b/, f is continuous on the closed interval between a and b,
and f is differentiable between a and b, then by Rolles Theorem there is an
x1 strictly between a and b such that f 0 .x1 / D 0.
Assume for some k with 1 k m, that f .k/ .a/ D f .k/ .xk / D 0, f .k/ is
continuous on the closed interval between a and xk , and f .k/ is differentiable
between a and xk . Then by Rolles Theorem there is an xkC1 strictly between
a and xk such that f .kC1/ .xkC1 / D 0.
Thus, by mathematical induction, there is an x D xmC1 strictly between a
and b such that f .mC1/ .x/ D 0 completing the proof.
264
8 Sequences of Functions
This Higher Order Rolles Theorem can now be used to prove Taylors Theorem.
If the function f is m C 1 times differentiable between c and x, then f has an
m .n/
P
f .c/
mth degree Taylor polynomial g.y/ D
.y c/n . Notice that the difference
n
nD0
f .y/ g.y/ has the property that this function and its first m derivatives are all equal
to 0 at c. The remainder term RmC1 .x/ will include a factor of f .mC1/ evaluated at
some between c and x, and that value of will come from an application of Rolles
Theorem. Of course, to apply Rolles Theorem, f .y/g.y/ would need to be 0 at y D
x. One needs to add a term to f .y/ g.y/ which will not affect the function and its
derivatives at c but will make the function equal to 0 at x. The term that accomplishes
.yc/mC1
this is f .x/g.x/ .xc/
mC1 since this term equals f .x/g.x/ at x, and it and its
first m derivatives are equal to 0 at c. But now Rolles Theorem can be applied to the
.yc/mC1
function h.y/ D f .y/g.y/ f .x/g.x/ .xc/
mC1 to find a value of between c and x
.mC1/
such that h.mC1/ ./ D 0, or 0 D f .mC1/ ./g.mC1/ ./.f .x/g.x// .xc/
mC1 . Noting
that the mC1st derivative of g at c is equal to 0 gives f .x/ D g.x/Cf .mC1/ ./ .xc/
.mC1/
as desired.
mC1
265
nD0
which both converge when jx cj < R for some R > 0. Then theorems about
the sum and difference of series of real numbers ensure that the sum and difference,
1
1
P
P
.an Cbn /.xc/n and .f g/.x/ D
.an bn /.xc/n , both converge
.f Cg/.x/ D
nD0
nD0
when jx cj < R. Of course, it is possible that the new series converges in an even
larger interval. For example, the series 1 C x C x2 C x3 C and 2 x x2 x3
both have radius of convergence equal to 1, but the sum of the two series is the
constant function 3, and its power series converges for all x.
The product of two power series can be found by using the Cauchy product of
1
P
the two series. If f .x/ D
an .x c/n has radius of convergence R1 > 0 and
g.x/ D
1
P
nD0
nD0
absolutely when jxcj < min.R1 ; R!2 / implying that their Cauchy
product, .fg/.x/ D
!
1
n
1
n
P
P
P
P
ap .x c/p bnp .x c/np D
ap bnp .x c/n , converges for
nD0
pD0
nD0
pD0
nD0
has radius of convergence R2 > 0, and g.c/ is not zero, then one can find the power
f .x/
series for the quotient h.x/ D g.x/
centered at c by working backwards from the
1
P
Cauchy product of h and g. That is, if you assume that h.x/ D
qn .x c/n , then
nD0
!
1
1
n
P
P
P
f .x/ D
an .x c/n D h.x/g.x/ D
bp qnp .x c/n . Because of the
nD0
nD0
pD0
assumption that g.c/ 0, it follows that b0 0. Then equating like terms in the
product gives the sequence of equations
a0 Db0 q0
a1 Db0 q1 C b1 q0
a2 Db0 q2 C b1 q1 C b2 q0
a3 Db0 q3 C b1 q2 C b3 q1 C b4 q0
266
8 Sequences of Functions
and so forth. The first equation can be solved to give q0 . Then the second equation
can be solved to give q1 , and so forth. The fact that g.c/ 0 says that the coefficient
b0 0 which allows the equation for am to be solved for qm for each m 0. Often
this results in a recursive formula for qn . For example, it is known that ln.1 C x/ D
2
3
4
x x2 C x3 C x4 , so you can find the series centered at 0 for the quotient
1
1
P
P
ln.1Cx/
n
n
D
qn x by writing .1 C x/
qn x giving 0 D 1 q0 so q0 D 0. Then
1Cx
nD0
nD0
.1/n1
n
.1/n1
n
D qn C qn1 , so q1 D 12 , q2 D 56 , q3 D 13
, and so forth
12
qn1 .
8.5.6 Exercises
Determine for which x the following power series converge.
1.
2.
3.
4.
1
P
nD0
1
P
nD0
1
P
nD0
1
P
nD0
4n .xC4/n
3n C5n
n5 .x2/n
8n
nxn
.2n/
nxn
nn
ex
e2x
sin.3x/
sin x
1Cx
ln.cos x/
3
5
2
4
Using the fact that sin x D x x3 C x5 and cos x D 1 x2 C x4 ,
show that the powers series satisfy the identity sin2 x C cos2 x D 1.
11. Find the first four nonzero terms of the series for tan x centered at 0 by finding
the quotient of the series for sin x and the series for cos x. Then check your work
by generating those
terms using
(
) Taylors Theorem.
1
e x2 if x > 0
. Prove that for each positive integer n, the
0 if x 0
derivative f .n/ .0/ D 0. Then show that the mth degree Taylor polynomial for
f centered at 0 is p.x/ D 0, and the remainder term is Rm .x/ D f .x/.
267
a n!1
1
d P
where dx
an .x
nD0
n!1 a
1
P
1
P
nD0
nD0
an xn D
d
a .x
dx n
1
P
nD0 x!R
question of Analysis asks when can you interchange the order of two limiting
processes? It is instructive to watch for other occurrences of this question as your
study of Analysis continues.
Chapter 9
269
270
y
x
S
SC
the empty set because all nonempty open intervals contain irrational numbers, so
no nonempty open interval is contained in Q. One sometimes says that Q has no
interior even though it does have an interior; it is just that its interior is the empty
set.
Then ext.S/, called the exterior of S, is just defined to be the interior of Sc , that
is, s 2 ext.S/ if there is an > 0 such that all x satisfying jx sj < are in
Sc D RnS. The exterior of S is the set of points that are completely surrounded by
points outside of S. You can think of the exterior of S as the collection of points
bounded away from S, that is, the points that are a positive distance from S. The
exterior of the set 0; 4 is the union of two open intervals, .1; 0/ [ .4; 1/. The
exterior of the set Q is the empty set.
If a point s is in neither int.S/ nor ext.S/, then it must be that no open interval
containing s is completely inside of S and no open interval containing s is completely
outside of S. Thus, for every > 0, the interval .s ; s C / contains at least one
element of S and at least one element of Sc . Such points are said to be in @S, called
the boundary of S (Fig. 9.1). Note that the symbol used for boundary is @ which
is the same symbol use for partial derivatives in Calculus. There are connections
between derivatives and boundaries that justify the use of the same symbol for both
concepts. The boundary of 0; 4 is the set f0; 4g. The boundary of Q is the entire
real line, R.
It is important to note that for any set S R, the three sets int.S/, ext.S/, and @S
partition R, that is, each real number x is in exactly one of these three sets. A proof
of this fact must show two things about a set S: that R D int.S/ [ ext.S/ [ @S, and
that no point x belongs to more than one of these sets. To show that R is a union of
the three sets, you would take an arbitrary x 2 R and show that it is in at least one of
these sets. One way to show that a point must be one of three things is to assume that
it not one of the first two, and then prove that it must be the third. In this case, you
can assume that a point x 2 R is not in int.S/ or in ext.S/. If x is not in int.S/, then
271
int.int.S// D int.S/.
int.ext.S// D ext.S/.
int.S/ ext.ext.S//.
ext.S/ ext.int.S//.
@.@.S// @S.
@.int.S// @S.
@.ext.S// @S.
@.S/ D @.Sc /.
int.S/ [ int.T/ int.S [ T/.
ext.S [ T/ ext.S/ \ ext.T/.
int.S \ T/ D int.S/ \ int.T/.
@.S [ T/ @S [ @T.
if S T, then int.S/ int.T/.
if S T, then ext.T/ ext.S/.
272
Each of these results is a statement about either two sets being equal to each other
or one set being a subset of another. Thus, one would prove these results using the
techniques discussed in Chap. 2 for proving subset and set equality statements. For
example, how would you write a proof that for any set S, int.int.S// D int.S/?
This would be a proof that two sets are equal, so the proof would consist of two
parts: showing int.int.S// int.S/ and showing int.S/ int.int.S//. The fact that
int.int.S// int.S/ is just a consequence of the definition of interior. For any set
T, int.T/ T, so certainly int.int.S// int.S/. Showing that int.S/ int.int.S//
is showing that one set is a subset of another. So, you would let x be an element of
int.S/, and then show that x is also an element of int.int.S//. By the definition of
interior, there is an > 0 such that the open interval .x ; x C / S. Thus, you
need to show that .x ; x C / is contained in int.S/. That is, each y 2 .x ; x C /
must be in the interior of S. But it is easy to find an open interval centered at y that
is contained in .x ; x C /. Just let D min.y .x /; x C y/ > 0 because
then .y ; y C / .x ; x C /. This shows each point of .x ; x C / is in
int.S/ which completes the proof.
PROOF: For every set S R, int.int.S// D int.S/.
Let S R.
For any set T, int.T/ T, so int.int.S// int.S/.
So let x 2 int.S/.
By the definition of interior, there is an > 0 such that the open interval
.x ; x C / is contained in S.
Let y 2 .x ; x C /, and let D min.y .x /; x C y/ > 0.
Then .y ; y C / .x ; x C / S.
This shows that .x ; x C / int.S/ implying that x is in int.int.S//.
This proves that int.S/ int.int.S// and completes the proof of the
theorem.
For a more difficult challenge, consider writing a proof that for any set S, @.@S/ @S
which, in words, says that the boundary of the boundary of a set is contained in the
boundary of the set. For example, let S be the set of rational numbers in the interval
0; 4. You should prove to yourself that the boundary of this set is the entire interval
0; 4. The boundary of that interval is just f0; 4g which indeed is contained in
@S D 0; 4. To show that @.@S/ is a subset of @S, you would take an arbitrary
point x 2 @.@S/ and show that it is in @S. So what do you know if x 2 @.@S/? The
only tool you have at your disposal here is the definition of the boundary of a set,
so you would proceed to use that definition. It says that for every > 0 the open
interval .x ; x C / contains elements of @S and elements of the complement
of @S. You want to show that x is in @S, so you would need to show that the open
interval .x ; x C / contains elements of S and elements of Sc . Well, what is the
consequence of saying that the open interval .x; xC/ contains elements of @S? It
must mean that there is a y 2 .x ; x C / such that y 2 @S. What does it mean for y
to be in @S? It means that for every > 0, the interval .y; yC/ contains elements
of S and elements of Sc . But this is sufficient if .y ; y C / .x ; x C / because
273
((S
)
c
y S
that would put elements of both S and Sc in .x ; x C /. This can be arranged by
selecting small enough (Fig. 9.2).
PROOF: For every set S R, @.@S/ @S.
Let S R.
Let x 2 @.@S/.
Then by the definition of boundary, for every > 0, the open interval
.x ; x C / contains points of @S and points of the complement of @S.
Let > 0 be given, and let y 2 .x ; x C / such that y 2 @S.
Let D min.y.x/; xCy/ > 0 so that the open interval .y; yC/
.x ; x C /.
By the definition of boundary, the interval .y ; y C / contains an element
of S and an element of Sc .
But .y ; y C / .x ; x C / shows that .x ; x C / contains an
element of S and an element of Sc , so x is in @S which completes the proof.
As a third example, consider proving that for any two sets S and T, that
int.S/ [ int.T/ int.S [ T/. Again, this is proving that one set is a subset of a
second set, so your proof would start by selecting an arbitrary element of the first
set and then proceed to show that that element belongs to the second set. Here the
first set is int.S/ [ int.T/. If you select an x from this set, all you know about x is that
it is in the union of the two sets int.S/ and int.T/. So, the only tool you can use is
the definition of union to say that x must be either a member of int.S/ or a member
of int.T/. In the case that x 2 int.S/, you can then apply the definition of interior to
say that there is an > 0 such that the interval .x ; x C / S. But this is all you
need since S S [ T showing .x ; x C / S [ T proving that x 2 int.S [ T/.
The case where x 2 int.T/ is analogous, completing the proof.
PROOF: For any sets of real numbers S and T, int.S/ [ int.T/ int.S [ T/.
Let S and T be sets of real numbers.
Let x 2 int.S/ [ int.T/.
Then by the definition of the union of two sets, either x 2 int.S/ or x 2
int.T/.
Without loss of generality, assume that x 2 int.S/.
Then there is an > 0 such that the open interval .x ; x C / S.
But since S S [ T, it follows that .x ; x C / S [ T showing that
x 2 int.S [ T/, completing the proof.
Can it be that int.S/ [ int.T/ is not equal to int.S [ T/? The answer is yes. See if
you can think of an example.
274
9.1.1 Exercises
For each of the following sets, find the interior, exterior, and boundary of the set.
1. 0; 3/ [ .3; 6
1
1
1
2. [ 2n
.
; 2n1
nD1
3. 0; 4 \ Q
Write proofs for each of the following statements. For exercises involving the subset
relation rather than the equality relation, give examples showing that the subset
relation in the statement cannot be replaced by an equality.
If S T, then int.S/ int.T/.
If S T, then ext.T/ ext.S/.
int.ext.S// D ext.S/.
ext.S/ ext.int.S//.
@.int.S// @S.
@.S/ D @.Sc /.
@.ext.S// @S.
int.S \ T/ D int.S/ \ int.T/.
ext.S [ T/ D ext.S/ \ ext.T/.
@.S [ T/ @S [ @T.
int.S/ ext.ext.S//
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
[ .2n; 2n C 1/. The intervals that are called closed intervals are, in fact, closed sets.
nD1
In particular, 5; 3, 4; 1/, and ; are all closed sets.
There are actually many equivalent ways to define open and closed sets, so
one usually begins this discussion by proving that all the different definitions are
equivalent. In particular, if S R, then the following are equivalent:
1.
2.
3.
4.
S is an open set.
S D int.S/.
S \ @S D ;.
Sc is a closed set.
Many theorems in mathematics are statements of the form p , q, and the proof of
these statements is often broken into two steps: p ) q and q ) p. Theorems of
that type state that two conditions are equivalent. But it is not uncommon to have
275
a theorem that states that several statements are equivalent, that is, p1 , p2 ,
p3 , , pk . One way to prove theorems of this form is to show in a sequence
of steps that p1 ) p2 , p2 ) p3 , p3 ) p4 , . . . , pk1 ) pk , and then pk ) p1 . This
is the technique you can use to prove the list of statements about open sets. You
would begin by assuming condition 1, that a set S is open and then prove condition
2, that S D int.S/. This can be done by noting that for any set, elements of the set
are either in the interior of the set or on the boundary of the set. But if the set S
is open, it means that for each x 2 S there is an > 0 such that the open interval
.x ; x C / S. Thus, .x ; x C / contains no elements of Sc showing that x
cannot be in @S, so it must be that x 2 int.S/ which proves that S D int.S/.
Now, assuming condition 2 that S D int.S/ it follows immediately that S \ @S D
int.S/ \ @S D ;, which is condition 3. If you assume condition 3 that S \ @S D ;,
how can you conclude that Sc is closed? Well, if S contains no elements of @S, it
must be that all the elements of @S (if there are any) must belong to Sc . But as seen
in the exercises of the previous section, the boundary of S and the boundary of Sc
are always the same. This follows from the fact that the definition of boundary is
symmetric in its references to S and Sc . Therefore, Sc contains its boundary proving
that Sc is a closed set, which is condition 4.
Finally, assuming condition 4 that Sc is a closed set, you know that Sc contains
its boundary, so Sc contains the boundary of S. You must show that for each x 2 S,
there is an open interval centered at x such that the entire interval is contained in S.
But if for every > 0 the open interval .x ; x C / contains elements in Sc , then
x would be in the boundary of S which is false. Thus, there is an > 0 such that
the open interval .x ; x C / is contained in S. This proves that S is an open set,
which is condition 1 (Fig. 9.3).
Fig. 9.3 An open set S, its
boundary, and its
complement Sc
S
SC
276
S is an open set.
S D int.S/.
S \ @S D ;.
Sc is a closed set.
Let S R.
Condition 1 ) Condition 2
Assume that S is an open set.
If x 2 S, then by the definition of open set, there exists an > 0 such that
.x ; x C / S.
But .x ; x C / S shows that x is not an element of @S.
Since S int.S/ [ @S, it can be concluded that S int.S/.
Because the interior of any set is contained in the set, it follows that int.S/
S implying that S D int.S/, which is condition 2.
Condition 2 ) Condition 3
Assume that S D int.S/.
Then S \ @S D int.S/ \ @S D ; because the interior and the boundary of
any set are disjoint.
Thus, S \ @S D ;, which is condition 3.
Condition 3 ) Condition 4
Assume that S \ @S D ;.
Then @S must be contained in Sc .
Because @S D @.Sc /, it follows that @.Sc / Sc implying that Sc is a closed
set, which is condition 4.
Condition 4 ) Condition 1
Assume that Sc is a closed set, which means that Sc contains @.Sc /.
Let x 2 S.
If for every > 0, the open interval .x ; x C / contains elements of Sc ,
then x would be an element of @S D @.Sc /.
But all elements of @.Sc / are contained in Sc , so there must be an > 0
such that the interval .x ; x C / contains no elements of Sc implying that
.x ; x C / S.
This shows that S is an open set, which is condition 1.
A similar theorem can be proved concerning closed sets.
277
S is a closed set.
S D int.S/ [ @S.
Every accumulation point of S is an element of S.
Sc is an open set.
Let S R.
Condition 1 ) Condition 2
Assume that S is a closed set.
Because no point in the exterior of S is a member of S, it is clear that S
int.S/ [ @S.
Because S is closed, it contains @S, and because all sets contain their interior,
S contains int.S/.
Thus, S contains int.S/ [ @S proving that S D int.S/ [ @S, which is
condition 2.
Condition 2 ) Condition 3
Assume that S D int.S/ [ @S.
Let x be an accumulation point of S.
If x S, then for all > 0, the open interval .x ; x C / contains
elements of S (since x is an accumulation point of S) and elements of Sc
(in particular, x).
Thus, x 2 @S implying that x 2 S, a contradiction.
Therefore, all accumulation points of S must be elements of S, which is
condition 3.
Condition 3 ) Condition 4
Assume that every accumulation point of S is an element of S.
Let x 2 Sc .
Because S contains all of its accumulation points, x is not an accumulation
point of S.
Thus, there is an > 0 such that the open interval .x ; x C / contains no
elements of S, and is, therefore, contained in Sc .
This shows that Sc is an open set, which is condition 4.
Condition 4 ) Condition 1
Assume that Sc is an open set.
Then Sc \ @.Sc / D ; implying that @.Sc / S, so @S S.
Thus, S contains @S, so S is a closed set, which is condition 1.
Of course, a set need not be either open or closed as is seen by the interval .0; 5
which contains one but not both of its boundary points (Fig. 9.4), so it is neither
open (because it contains a boundary point) nor closed (because it does not contain
all of its boundary points).
278
Open
Closed
9.2.1 Exercises
Determine which of the following sets of real numbers are open and which are
closed.
1.
2.
3.
4.
279
previous section, you could prove that the set A is open by using the definition of
open set, by showing that A D int.A/, by showing that A \ @A D ;, or by showing
that Ac is a closed set. In this case, it is simple enough to use the definition of open
set. Thus, for each x 2 A, you would need to show that there is an open interval
centered at x such that this open interval is contained in A. All you know about A
is that it is a union of a collection of open sets, so the first thing you should try is
invoking the definition of union. That is, if x 2 A, then there must be a j 2 I such
that x 2 Aj . What do you know about Aj ? Only that it is an open set. That means that
there is an > 0 such that the open interval .x ; x C / is contained in Aj . But
by the definition of union, Aj A implying that .x ; x C / A which is what
you needed to prove.
280
PROOF: Assume that for each i in the index set I, Ai is an open set. Then
[ Ai is an open set.
i2I
i2I
i2I
Now consider proving the result that the intersection of a finite collection of
open sets is itself an open set. This time there is no need to consider an arbitrarily
large collection of open sets; you can just use the finite collection of open sets
A1 ; A2 ; A3 ; : : : ; Ak . Again you would take an arbitrary x 2 A1 \ A2 \ A3 \ \ Ak .
You know from the definition of intersection that for each j D 1; 2; 3; : : : ; k, this
element x must be in Aj . And you know that since Aj is an open set, there must be
an j > 0 such that the interval .x j ; x C j / Aj . Now you have a collection
of k open intervals each centered at x. By selecting D min.1 ; 2 ; 3 ; : : : ; k /, you
will have the least of these j values which is a positive number. This is crucial. The
fact that you have a finite collection of open sets ensures that you can find a finite
number of open intervals centered at x and can find the shortest of these intervals.
If the collection of open sets were infinite, there would be no guarantee that there
would be a minimum j . The fact that there is a minimum value that is greater than
0 allows you to claim that the interval .x ; x C / is contained in each of the Aj
sets, and thus, .x ; x C / is contained in the intersection of the Aj s.
PROOF: Assume that A1 ; A2 ; A3 ; : : : ; Ak
A1 \ A2 \ A3 \ \ Ak is an open set.
are
open
sets.
Then
281
to use the results from the previous section that show that a set is a closed set if and
only if it is the complement is an open set. For example, to show that the union of
a finite number of closed sets is closed, let A1 ; A2 ; A3 ; : : : ; Ak be a finite collection
of closed sets. Then for each j, Acj is the complement of a closed set, so it is open.
By the previous theorem, the intersection of a finite number of open sets is an open
set, so Ac1 \ Ac2 \ Ac3 \ \ Ack is an open set. But DeMorgans Law says that
Ac1 \ Ac2 \ Ac3 \ \ Ack D .A1 [ A2 [ A3 [ [ Ak /c which is an open set, so its
complement, A1 [ A2 [ A3 [ [ Ak is a closed set as desired.
Although not needed in this textbook about writing proofs in Analysis, for
completeness, it makes sense at this point to introduce the definition of a topological
space to be a set S together with a collection T of subsets of S satisfying the
conditions
Both ; and S are in T .
The union of any collection of sets in T is also a set in T .
The intersection of any finite collection of sets in T is also a set in T .
If these conditions are satisfied, then the set T is said to be a topology for the
topological space S. From the definitions and theorems presented so far in this
chapter it follows that the real numbers R along with its collection of open sets
forms a topological space. The advantage of introducing the more general concept
of a topological space is that many theorems about the real numbers extend to
all topological spaces, so once you justify the fact that you are dealing with a
topological space, you then know many theorems about your new space.
As an example of another topological space consider the set of integers, Z, along
with the collection T of subsets of Z consisting of the empty set, ;, and the sets
A Z with the property that Ac D ZnA is a finite set. It is easy to see that both
; and Z are elements of T . To show that T is closed under unions, suppose you
have a collection of sets in T . There are two cases to consider: (1) all the sets in the
collection are the empty set, and (2) at least one of the sets in the collection is not
empty. In the first case, the union of all the sets in the collection is the empty set
which is in T . In the second case, if the collection includes a set A, then the union
of the sets in the collection contains A, and because the complement of the union
lies inside the complement of A which is finite, the union will have to have a finite
complement and be a set in T . To show that T is closed under finite intersections,
suppose you have a finite collection of sets in T . Again, there are two cases to
consider: (1) at least one set in the collection is the empty set, and (2) none of
the sets in the collection is the empty set. In the first case, the intersection of the
collection of sets is the empty set which is in T . In the second case, the complement
of the intersection of the finite collection of sets is the union of the complements of
the sets. If all the complements are finite, then the union of the finite number of
complements is also finite, so the intersection is in T . This verifies that T is a
topology for Z. This is known as the finite complement topology for Z. It is clearly
not the usual topology associated with the integers which is just the usual topology
of R restricted to Z. Generally, a set can have many different topologies, each giving
rise to a different topological space. Most of these topologies are uninteresting and
have few if any applications.
282
9.3.1 Exercises
1. Find a sequence of open sets whose intersection is neither an open nor a
closed set.
2. Find a sequence of closed sets whose union is neither an open nor a closed set.
3. Prove that the intersection of a collection of closed sets is a closed set.
4. Prove that an open set of real numbers is the union of all the open intervals
contained in the set.
5. Verify that if S is any set, then the power set of S, P .S/, consisting of all the
subsets of S, is a topology for S. This is called the discrete topology for S.
6. Let S be the interval 0; 5, and let T include the empty set, the set S, and any
interval of the form 0; x/ where x 2 .0; 5/. Verify that T is a topology for S.
283
set A to elements of the set B, the function f can be extended to f W P .A/ ! P .B/
which maps subsets of the set A to subsets of the set B. If C A, then define f .C/
to be the set fy 2 B j y D f .a/ for some a 2 Cg. Then f .C/ is called the image of
C under f . The notation f .C/ could be confusing because f was originally defined
for elements of A, not subsets of A. The application of f to subsets of A is really
defining a new function f W P .A/ ! P .A/ whose domain is the power set of A and
codomain is the power set of B. The confusion arises because the same name, f , is
given to both functions. The confusion is cleared up by recognizing the distinction
that if the argument of f is an element a 2 A, then f .a/ refers to an element of the
codomain, B, while if the argument of f is a subset C A, then f .C/ is a subset of
B, f .C/ B.
For example, the function f .x/ D x2 is defined to be a function with domain R
and codomain R. It is then easily understood that f .3/ D 9 and f .2/ D 4. But
taking C to be the interval .3; 2/, the expression f .C/ now refers to the function
f W P .R/ ! P .R/, and f .C/ is the set of all elements of R that are images under f
of elements of C. That is, f .C/ D 0; 9/.
If the function f W A ! B is not a bijection mapping A one-to-one and onto B,
then it is not possible to define the inverse function f 1 W B ! A. One problem is
that if f is not surjective (mapping A onto B), there might be an element of b 2 B for
which there is no corresponding element a satisfying f .a/ D b, so f 1 .b/ cannot be
defined. Another problem is that if f is not injective (mapping A one-to-one to B),
then there will be an element b 2 B such that f .x/ D b is satisfied by more than one
value of x, so f 1 .b/ would not be unique. On the other hand, if D B, it is always
possible to define the function f 1 W P .B/ ! P .A/ mapping the power set of B to
the power set of A. Indeed, one defines f 1 .D/ D fx 2 A j f .x/ 2 Dg. In this case
f 1 .D/ is called the preimage of D under f . For example,
returning
to f .x/ D x2 ,
1
1
.1; 16/ D .4; 4/.
it follows that f .4; 9/ D .3; 2/ [ .2; 3/ and f
2
Note
that
when
the
continuous
function
f
.x/
D
x
was
applied to an open set as in
f .3; 2/ D 0; 9/, the image did not end
up
being
an
open
set. But when f 1 was
1
.1; 16/ D .4; 4/, the preimage was also an open
applied to an open set as in f
set. This is an important distinction. A continuous function need not map open
sets to open sets; functions that do map all open sets to open sets are called open
functions. But all continuous functions have the property that their inverses always
map open sets to open sets. Conversely, a function whose inverse always maps open
sets to open sets must be a continuous function. Of course, these statements require
proof, but the proofs follow directly from the definition of continuity and definition
of open set.
Assume, for example, that f W A ! B is a continuous function and D B is an
open set in B. You are challenged to show that f 1 .D/ is an open set in A. To show
that f 1 .D/ is open, you would need to show for every a 2 f 1 .D/ there is a > 0
such that .a ; a C / \ A f 1 .D/. From the definition of f 1 .D/, you know
1
that if a 2 f .D/, then f .a/ 2 D. Because D is open, there is an > 0 such that
f .a/ ; f .a/ C \ B D. This means that if y 2 B such that jy f .a/j < , then
y 2 D. But now, by the definition of continuity, there is a > 0 such that if x 2 A
284
C
A
a
f-1
f(a) = b
f-1(D) = C
B
with jxaj < , then jf .x/f .a/j < implying that f .x/ is in f .a/; f .a/C \B,
and thus, f .x/ is in D. This shows that x 2 f 1 .D/ proving that .a ; a C /
f 1 .D/, so f 1 .D/ is open.
Conversely, suppose that f has the property that f 1 .D/ is an open set in A
whenever D is an open set in B. Then let a 2 A. This time you are challenged
to show that for every > 0, there is a > 0 such that if x 2 A with jx aj < ,
then jf .x/ f .a/j < . But the set D of all y 2 B satisfying jy f .a/j < is an
open set in B implying that f 1 .D/ is an open set in A containing the point a. This
means that there is a > 0 such that .a ; a C / \ A is contained in f 1 .D/.
In other words, if x 2 A with jx aj < , then x is in f 1 .D/, so f .x/ is in D and
jf .x/ f .a/j < , completing the proof that f is continuous (Fig. 9.6).
PROOF: Let A and B be sets of real numbers, and let f W A ! B be a
function from A to B. Then f is continuous on A if and only if for every
open set D B, its preimage under f , f 1 .D/, is an open set in A.
Let A and B be sets of real numbers, and let f W A ! B be a function from A
to B.
Continuity implies that the preimages of open sets are open
Assume that f W A ! B is a continuous function.
Let D be an open set in B, and let a 2 f 1 .D/.
Because D is open in B, there is an > 0 such that
.f .a/ ; f .a/ C / \ B D.
Thus, if y 2 B with jy f .a/j < , then y 2 D.
Because f is a continuous function, there is a > 0 such that for all x 2 A
with jx aj < it follows that jf .x/ f .a/j < .
Thus, if x 2 .a ; a C / \ A, then jf .x/ f .a/j < , so f .x/ 2 D and
x 2 f 1 .D/.
Therefore, .a ; a C / f 1 .D/ and f 1 .D/ is an open set in A. This
proves that the preimage under f of any open set is open.
(continued)
9.5 Closure
285
9.4.1 Exercises
Write proofs for each of the following statements.
1.
2.
3.
4.
5.
9.5 Closure
Recall that if S is any subset of R, then a is an accumulation point of S if for
every > 0 the open interval .a ; a C / contains at least one point of S
other than a itself. An important property of closed sets is that if a closed set,
S, has an accumulation point, a, then a 2 S. You should be able to construct a
286
short proof of this fact that relies only on the definitions of accumulation point,
closed set, and boundary. Such a proof would start with the assumption that a is
an accumulation point of the closed set S. One way to continue is to construct a
proof by contradiction, that is, to assume that a S and hope that this will lead to a
contradiction. Interestingly, you can proceed in more than one way. You could use
the fact that S is a closed set which implies that, since a S, then a 2 ext.S/.
This means that there is an > 0 such that the open interval .a ; a C /
is contained in Sc . But the definition of accumulation point says that every open
interval containing a also contains points of S, so this contradicts the fact that
a is an accumulation point of S. Alternatively, you could use the fact that a is
an accumulation point of S. This means that for every > 0, the open interval
.a ; a C / contains points in S. All of these open intervals also contain a S
implying that each of these open intervals contains points in S and points in Sc . Thus,
a satisfies the definition of being an element of @S. From the definition of closed set,
@S S. Thus, a 2 S.
PROOF: If S is a closed set, then S contains all of its accumulation points.
Let S be a closed set, and let a be an accumulation point of S.
Assume that a S.
From the definition of accumulation point, for every > 0 it follows that
the open interval .a ; a C / contains elements in S.
Because a S, it follows that for every > 0 the open interval .a; aC/
contains elements of S and elements of Sc , so a 2 @S.
From the definition of closed set, @S S, so a 2 S which contradicts the
assumption that a S.
Thus, every accumulation point of S must be contained in S.
The collection of all the accumulation points of S is called the derived set of S
which is written S0 . The previous theorem shows that if S is closed, then S0 S.
The converse is also true, that is, if S0 S, then S must be closed. This follows from
the fact that if a is in the boundary of S but a is not an element of S, then a must be
an accumulation point of S. This should make sense to you. A boundary point is a
point close both to S and to Sc . An accumulation point is close to S, and if it is not
in S, it is close to Sc .
PROOF: If set S contains all of its accumulation points, then S is a closed
set.
Let S be a set that contains all of its accumulation points.
Assume that a 2 @SnS.
Because a 2 @S, for every > 0, the open interval .a ; a C / contains
elements of S and elements of Sc .
Thus, because a itself is not a member of S, .a; aC/ contains an element
of S not equal to a.
It follows that a 2 S0 S which contradicts the assumption that a S.
Therefore, @S S which proves that S is a closed set.
9.5 Closure
287
You can conclude from this result that for any set S, if a 2 @S \ Sc , it is an
accumulation point of S, and, by symmetry, if a 2 @S \ S, then it is an accumulation
point of Sc . The set S is closed if it contains its boundary, @S. But for any set S, the
elements of @S that are not in S are accumulation points of S, so S is closed if and
only if it contains all of its accumulation points. It is important to recognize, though,
that the derived set S0 need not be contained in @S since points in the interior of S are
accumulation points of S, and @S need not be contained in S0 since isolated points
of S are in the boundary of S without being accumulation points of S. On the other
hand, S [ @S D S [ S0 .
For any set S, define the closure of S or cl.S/ to be S [ S0 D S [ @S. Some books
use the notation S or S for the closure of S. Intuitively, the closure of a set S takes
the elements of the boundary of S and adds them to the set so that you now have S
along with its boundary (Fig. 9.7). The closure also has the following properties.
All of these results have short proofs. For example, to get the first result, recall that
if x is in the boundary of the union of two sets, S [ T, then x is either in the boundary
of S or the boundary of T. Thus, if x 2 @ cl.S/, it means that x 2 @.S [ @S/ and,
therefore, x 2 @S or x 2 @.@S/. It was shown in the first section of this chapter that
@.@S/ @S implying that x 2 @S proving that x is in cl.S/. Thus, cl.S/ contains its
boundary, so it is closed.
For the second result, note that if S is closed, it contains its boundary so cl.S/ D
S [ @S D S. Conversely, if S D cl.S/, then S is closed because cl.S/ is always a
closed set.
The third and fourth results follow quickly after noticing that any closed set
containing S must also contain the boundary of S.
](
cl(S)
][
288
9.5.1 Exercises
For each of the following sets S, determine @S, S0 , and cl.S/.
1.
2.
3.
4.
5.
9.6 Compactness
The topics of open cover, finite subcover, compactness, and the HeineBorel
Theorem were introduced in Chap. 4 because of their usefulness in proving that a
function continuous on a closed bounded interval is uniformly continuous on that
interval. Compactness also played an important role in showing that a continuous
function on a closed bounded interval is bounded, a continuous function on a
closed bounded interval obtains its extreme values (maximum and minimum), and a
continuous function on a closed bounded interval has a Riemann integral. Recall
that an open cover of a set S was defined to be a collection open intervals T
where for each x 2 S there is an open interval .p; q/ 2 T such that x 2 .p; q/.
After the introduction of the topological ideas in this chapter, that definition can be
generalized to allow T to be a collection of open sets rather than just open intervals,
that is, a collection of open sets, T, is called an open cover of S if for each x 2 S
there is an open set U 2 T such that x 2 U. Moreover, the HeineBorel Theorem
can now be extended in two ways: the concept of an open cover by intervals can
be generalized to an open cover by open sets, and the concept of closed bounded
interval can be generalized to closed bounded set.
PROOF (HeineBorel Theorem): Let S be any closed bounded set of real
numbers, and let T be a cover of S by open sets. Then T contains a finite
subcover of S.
Let S be a closed bounded set and T be a cover of S by open sets.
Because S is bounded, there are real numbers a and b with a < b such that
S a; b.
(continued)
9.6 Compactness
289
290
9.7 Connectedness
291
9.6.1 Exercises
1. Find an example of a function f and a set C such that f 1 f .C/ is notequal to C.
2. Find an example of a continuous function f and a set D such that f f 1 .D/ is
not equal to D.
3. Find an example of a continuous function f W A ! B and a compact set D B
such that f 1 .D/ is not compact.
4. Suppose that the continuous function f has domain 0; 10 and codomain .4; 4/.
Show that the function is not surjective.
9.7 Connectedness
The intervals on the real line were discussed in Chap. 2. A set of real numbers is an
interval if whenever x and y are elements of the interval, then all the real numbers
between x and y are also elements of the interval. The intervals are the connected
sets on the real line, but the concept of connectedness can be extended to any
topological space. In a general topological space, two nonempty sets A and B are
292
disconnected if there are disjoint open sets U and V with A U and B V. For
example, the sets 0; 1 and .4; 5/ are disconnected because 0; 1 .1; 2/ and
.4; 5/ .4; 5/ where .1; 2/ and .4; 5/ are disjoint open sets (Fig. 9.8). The sets
0; 3 and .3; 5/ are disjoint nonempty sets, but they are not disconnected because
any open set that contains 0; 3 will necessarily share points with any open set
containing .3; 5/, and, in particular, both open sets will contain the element 3. A set
is called connected if it is not the union of two disconnected nonempty sets. Even
though the connected sets of real numbers are just the intervals, the concept of
connectedness gets far more interesting in more general topological spaces.
If f W A ! B is continuous, then it always maps connected sets to connected
sets, that is, if C A is a connected set, then so is f .C/. This is easy to see since,
if f .C/ is disconnected, then there are two disjoint open sets U and V in B and two
nonempty sets S and T in B such that f .C/ D S [ T and S U and T V. But then
C f 1 .U/ [ f 1 .V/ where f 1 .U/ and f 1 .V/ are disjoint open sets in A. Because
S and T are nonempty, C \ f 1 .U/ and C \ f 1 .V/ are nonempty implying that C
is a disconnected set. Thus, if C is connected, f .C/ must also be connected.
PROOF: If f W A ! B is continuous on A, and if C A is a connected set,
then f .C/ is a connected set in B.
Let f W A ! B be a continuous function on A, and assume that C A such
that f .C/ is disconnected.
This means that there are disjoint open sets U and V in B, and nonempty
sets S and T in B with S U and T V such that B D S [ T.
Since f is continuous, f 1 .U/ and f 1 .V/ are open sets in A.
Since S and T are nonempty sets whose union is f .C/, both C \ f 1 .U/ and
C \ f 1 .V/ are nonempty.
This shows that C is a disconnected set.
Therefore, if C is a connected set, f .C/ must also be connected.
When this theorem is applied to functions from the real numbers to the real
numbers, the result is the Intermediate Value Theorem which states that if f is a
real valued function on the interval a; b, then for every c between f .a/ and f .b/
there is an x 2 a; b such that f .x/ D c. This is because f must map the connected
set a; b into a connected set which must include all the elements c between f .a/
and f .b/. Note that f 1 need not bring connected sets to connected sets.
In n-dimensional Euclidean space the concept of connectedness gets considerably richer as the connected sets are not merely the cross products of intervals
(Fig. 9.9). In R2 one introduces what it means for a set to be path-connected which
makes precise the intuitive notion that a set is connected if you can draw a path
9.7 Connectedness
293
N
Fig. 9.9 The set C is a connected set. The set N is not a connected set
Fig. 9.10 Graph of sin
with the y-axis
1
x
between any two of its points where the path stays inside the set. On the real line,
this just means that for any two points in the set, the interval between the two points
stays in the set. But in R2 where paths need not be straight lines, the examples are
far more varied. In fact, in R2 there are examples of connected sets that are not path
connected, a phenomenon that cannot occur on the real line.
A famous example is
the set consisting of the graph of the equation y D sin 1x along with the y-axis.
This is a connected set because
any
open set that contains the y-axis must intersect
parts of the graph of y D sin 1x both to the left and to the right of the y-axis. On
the other hand, this set is not path-connected because there is no way to construct a
path that stays inside the set and connects the points . 1 ; 0/ and . 1 ; 0/ (Fig. 9.10).
9.7.1 Exercises
1. Find an example of a continuous function f W A ! B and connected set D B
such that f 1 .D/ is not connected.
2. Show that in any topological space A, if S and T are connected sets with A D S[T
and S \ T ;, then A is connected.
Chapter 10
Metric Spaces
295
296
10 Metric Spaces
x
d(x,y)
d(x,z)
y
d(y,z)
real numbers. With this in mind a nonempty set X together with distance function
d is defined to be a metric space if, for all x, y, and z in X, this distance function
satisfies the following properties:
The distance function defines a metric for the metric space, and the metric
space is designated as <X; d> (Fig. 10.1). This definition is a generalization of the
distance function defined on the real numbers, d.x; y/ D jx yj. Clearly, for all real
numbers x, y, and z,
d.x; y/ D jx yj 0
0 D d.x; y/ D jx yj if and only if x D y
d.x; y/ D jx yj D jy xj D d.y; x/
d.x; y/ C d.y; z/ D jx yj C jy zj j.x y/ C .y z/j D jx zj D d.x; z/
10.2 Inequalities
297
several different distance functions defined on a particular set X that make X into a
metric space. Each new distance function results in different shaped neighborhoods.
Some give rise to the same topology of X while others may result in quite different
topologies. Many examples of these different distance functions will be explored in
the sections that follow.
10.2 Inequalities
Most proofs in Analysis involve establishing one or more inequalities. Some
inequalities seem to keep reappearing in different guises throughout Analysis, so
they provide great tools for writing proofs. This section presents two very common
inequalities that will be used later in the chapter to justify the triangle inequality for
some examples of metric spaces.
q
a21 C a22 C a33 C C a2n b21 C b22 C b33 C C b2n :
n
P
jD1
n
P
jD1
a2j C 2x
n
P
aj bj C
jD1
real number x. Any quadratic polynomial Ax2 CBxCC with A > 0 is nonnegative for
every x if and only if its discriminant B2 4AC is not positive. But the discriminant
298
10 Metric Spaces
2
of the previous polynomial is 4 4
n
P
!2
aj bj
jD1
n
P
jD1
!
a2j
n
P
jD1
!3
b2j 5. The statement
that this discriminant is less than or equal to 0 is exactly the statement of the
CauchySchwarz Inequality. An even stronger statement can now be made. Equality
occurs in the CauchySchwarz Inequality if and only if the given discriminant is 0
so that the underlying quadratic polynomial has exactly one root, meaning that the
n
P
sum .aj C xbj /2 is 0 for exactly one value of x. This happens if and only if a is
jD1
q
q
a21 C a22 C a33 C C a2n b21 C b22 C b33 C C b2n
doubling it and adding a21 C a22 C a23 C C a2n C b21 C b22 C b23 C C b2n to
both sides yields
.a1 C b1 /2 C .a2 C b2 /2 C .a3 C b3 /2 C C .an C bn /2
q
q
2
a1 C a22 C a23 C C a2n C 2 a21 C a22 C a33 C C a2n b21 C b22 C b33 C C b2n C .b21 C b22 C b23 C C b2n /
which is a special case of the Minkowski Inequality which can be restated jaCbj
jaj C jbj. Again, equality occurs only when one of the points is a scalar multiple of
the other.
10.2.3 Exercises
1. Show that the CauchySchwarz Inequality extends to infinite s
series. That
s is, if
1
1
1
1
1
P 2
P 2
P
P 2 P
an and
bn are both convergent series, then
an bn
an
b2n .
nD1
nD1
nD1
nD1
nD1
299
1
P
2. Show that the Minkowski Inequality extends to infinite series. That is, if
a2n
nD1
s
s
1
1
1
P
P
P
2
2
and
bn are both convergent series, then
.an C bn /
a2n C
nD1
nD1
nD1
s
1
P
b2n .
nD1
3. Show that for any real numbers a1 ; a2 ; a3 ; : : : ; an and positive real numbers
a2
a2
a2
a2
b1 ; b2 ; b3 ; : : : ; bn , the following inequality holds: 1 C 2 C 3 C C n
b1
b2
b3
bn
.a1 C a2 C a3 C C an /2
. This can be shown by mathematical induction on n,
b1 C b2 C b3 C C bn
but can also be shown using the CauchySchwarz Inequality.
300
Fig. 10.2 Euclidean distance
is R2
10 Metric Spaces
(x1, y1)
d(x,y) =
|y2 y1|
|x2 x1|
(x2, y2)
301
Fig. 10.3 N.0; 1/ in the Euclidean, taxicab, and supremum metrics in 2 and 3 dimensions
302
10 Metric Spaces
metrics give rise to the same topology on Rn because each metric gives the same
open sets even though the open neighborhoods are different in shape. But the three
metrics have different algebraic properties, and sometimes it is easier to prove a
particular theorem using one of these metrics rather than the others.
Distance measures in metric spaces need not be complicated. For any set X you
can define d.x; x/ D 0 for all x 2 X and d.x; y/ D 1 for all x and y in X with x y. It
is very easy to see that d.x; y/ is nonnegative, symmetric, and equal to 0 if and only
if x D y. Also, for any x; y; z 2 S, if d.x; z/ D 1, then x z, so at least one of d.x; y/
and d.y; z/ must be 1 which implies the triangle inequality d.x; y/Cd.y; z/ d.x; z/.
Thus, any set X is a metric space with this metric sometimes called the discrete
metric, and <X; d> is called a discrete metric space. Note that for this metric,
each neighborhood, N.a; r/ is either all of X or just the single point fag depending
on whether or not r is greater than 1.
Next, consider a space that looks much different than Euclidean space. Let C0; 1
be the set of all real-valued functions continuous on the interval 0; 1. Certainly,
this set contains all the polynomials with real coefficients, but it also includes the
rational functions that are defined on 0; 1, exponential functions, many elementary
functions, and a much larger class of functions continuous but not differentiable on
0; 1. This set is truly very large as compared, say, to the set of real numbers. There
are many ways you might try to measure the distance between two functions in this
set. For example, you could evaluate the function at one or more points and measure
how much the functions differ at those points.
That is,if f and g are in C0; 1, you
could define d.f ; g/ D jf .0/ g.0/j C f 12 g 12 C jf .1/ g.1/j. The only
problem with this definition is that there are continuous functions f and g which are
equal at 0, 12 , and 1 but not equal at other points such as f .x/ D x.2x 1/.x 1/
and g.x/ D 2x.2x 1/.x 1/. Because the given distance function gives a distance
of 0 between two unequal functions, it cannot serve as a metric for the space of
continuous functions on 0; 1 (Fig. 10.4).
As a result, a distance function that makes C0; 1 into a metric space really
needs to take into account the values of the functions at all the points (or at least
a dense set of points) in 0; 1. One distance measure that does this is called
the supremum metric or sup metric for short. It is defined for all f and g in
C0; 1 as d.f ; g/ D sup jf .x/ g.x/j. It is clear that if f g, then there are
x20;1
303
function has a valid definition, that is, for every f and g in C0; 1 the distance
function gives a nonnegative real number. But if f and g are continuous functions
on 0; 1, then so is jf .x/ g.x/j. Since all functions continuous on 0; 1 are
bounded and jf .x/g.x/j is a continuous function, the needed supremum is defined.
The triangle inequality follows from the fact that the triangle inequality works for
real numbers. Since for any three continuous functions f , g, and h and for each
x 2 0; 1it is true that jf .x/ g.x/j C
jg.x/ h.x/j jf .x/ h.x/j, it follows
that sup jf .x/ g.x/j C jg.x/ h.x/j sup jf .x/ h.x/j. Then, the inequality
x20;1
x20;1
sup.A C B/ sup A C sup B shows that sup jf .x/ g.x/j C sup jg.x/ h.x/j
x20;1
x20;1
sup jf .x/ g.x/j C jg.x/ h.x/j sup jf .x/ h.x/j, and d.f ; g/ C d.g; h/
x20;1
x20;1
d.f ; h/.
PROOF: The set C01 with distance function d.f ; g/ D sup jf .x/ g.x/j
x20;1
is a metric space.
SET THE CONTEXT: Let C0; 1 be the set of real-valued functions
continuous on the interval 0; 1.
METRIC DEFINITION: For any f and g in C0; 1, the function
jf .x/ g.x/j is also in C0; 1. Define d.x; y/ D sup jf .x/ g.x/j which is
x20;1
SYMMETRY: Since for all x 2 0; 1 and all f ; g 2 C0; 1, jf .x/ g.x/j D
jg.x/ f .x/j, it follows that d.f ; g/ D d.g; f /.
TRIANGLE INEQUALITY: Since for all x 2 0; 1 and all f ; g; h 2 C0; 1,
it holds that jf .x/ g.x/j C jg.x/ h.x/j jf .x/ h.x/j, it follows
that sup jf .x/ g.x/j C sup jg.x/ h.x/j sup jf .x/ g.x/j C
x20;1
x20;1
x20;1
jg.x/ h.x/j sup jf .x/ h.x/j, and d.f ; g/ C d.g; h/ d.f ; h/.
x20;1
This shows that C0; 1 with the supremum distance function is a metric
space.
The supremum metric provides only one of many possible distance functions
for the space C0; 1. Another example is called the L1 metric and is defined by
R1
d.f ; g/ D jf .x/ g.x/jdx. Since all functions continuous on a closed interval are
0
integrable there, this distance function is defined. Moreover, since jf .x/ g.x/j 0
for all x 2 0; 1, its integral is also nonnegative. If f g, then there is an a 2 0; 1
where f .a/ g.a/. Because jf .x/ g.x/j is continuous and positive at x D a, there
is a > 0 such that for all x 2 C0; 1 with jxaj < , jf .x/g.x/j > 12 jf .a/g.a/j.
304
10 Metric Spaces
R1
0
aC
R
a
a rigorous proof will take care that the limits of integration in the previous sentence
are chosen in a way that the integral is guaranteed to be defined. The symmetry
of d follows from its definition. For all f ; g; h 2 C0; 1 and each x 2 0; 1,
the triangle inequality gives jf .x/ g.x/j C jg.x/ h.x/j jf .x/ h.x/j. Thus,
R1
R1
R1
jf .x/ g.x/jdx C jg.x/ h.x/jdx D jf .x/ g.x/j C jg.x/ h.x/jdx
0
R1
0
jf .x/ h.x/jdx, so d.f ; g/ C d.g; h/ d.f ; h/, and the needed triangle inequality
holds.
PROOF: The set C01 with distance function d.f ; g/ D
is a metric space.
R1
0
jf .x/ g.x/jdx
SYMMETRY: Since for all x 2 0; 1 and all f ; g 2 C0; 1, jf .x/ g.x/j D
jg.x/ f .x/j, it follows that d.x; y/ D d.y; x/.
TRIANGLE INEQUALITY: Since for all x 2 0; 1 and all f ; g; h 2 C0; 1,
it holds that jf .x/ g.x/j C jg.x/ h.x/j jf .x/ h.x/j, it follows that
R1
R1
R1
jf .x/ g.x/jdx C jg.x/ h.x/jdx D jf .x/ g.x/j C jg.x/ h.x/jdx
0
R1
0
This shows that C0; 1 with the d.f ; g/ distance function is a metric space.
305
It is important to note that the supremum metric and the L1 metric are
distinctly different. In particular,
8
9 consider the sequence of functions fn .x/ D
1
>
0
if
0
>
nC1
>
>
>
<
=
1
1
for all natural numbers n. In the L1 metric,
n.n C 1/x n if nC1 < x n
>
>
>
>
>
:
;
1
1
if n < x 1
these functions converge to the function which is identically 1 on 0; 1. On the
other hand, this sequence is not even a Cauchy sequence in the supremum metric
since d.fn ; fm / D 1 for all n m. All metrics for C0; 1 need to measure the
distance between two continuous functions. The supremum metric measures the
maximum distance between two functions whereas the L1 metric measures a mean
distance between two functions.
10.3.1 Exercises
Write proofs for each of the following statements.
1. Let C be a circle. For x and y in C, define d.x; y/ to be the number in 0;
equal to the measure of the central angle in C of the arc bounded by x and y.
Show that C with this distance function is a metric space.
2. If d is defined for points .x1 ; y1 / and .x2 ; y2 / in R2 by 2jx1 x2 j C 3jy1 y2 j,
then R2 with distance function d is a metric space.
3. Let X be the set consisting of all integers plus one extra point M. For each
1
x 2 X, let d.x; x/ be 0. For integers x y, let d.x; y/ D min.jxj;jyj/C1
, and for
1
each integer x, let d.x; M/ D d.M; x/ D jxjC1 . Then <X; d> is a metric space.
4. Let X be the collection of all sequences of real numbers a1 ; a2 ; a3 ; : : : for which
there exists a natural number n such that aj D ak for all j and k greater than or
equal to n. In other words, X is the collection of all sequences which are constant
from some point on, such as 1; 2; 3; 4; 3; 3; 3; 3; : : : or 12 ; 23 ; 23 ; 23 ; 23 ; : : : . Define
the distance between two sequences <aj > and <bj > to be 0 if the sequences are
identical, and to be the least natural number n for which the difference between
the two sequences <aj bj > is constant for all terms j n. Then X is a metric
space with this metric.
5. Let p be any prime number. Then for any two rational numbersn r and s, define
d.r; s/ D 0 if r D s. Otherwise, if r s, then jr sj D pba where a and
b are relatively prime natural numbers, n is an integer, and neither a nor b is
divisible by p. Define d.r; s/ D pn . Then the rational numbers with distance
function d is a metric space.
6. If <X; d> is a metric space, then for any c > 0, <X; c d> is also a metric
space.
7. If <X; d1 > and <X; d2 > are both metric spaces, then <X; d1 C d2 > is also a
metric space.
306
10 Metric Spaces
8. If <X; dX > and <Y; dY > are both metric spaces, then X Y D f.x; y/ j x 2
X and y 2 Yg with distance function d .x1 ; y1 /; .x2 ; y2 / D dX .x1 ; x2 / C
dY .y1 ; y2 / is a metric space.
s
2
R1
f .x/ g.x/ dx is a metric space.
9. C0; 1 with distance function d.f ; g/ D
0
307
Many of the theorems pertaining to the topology of the real numbers proved in the
preceding chapter can now be reproved in the context of metric spaces by merely
changing references to open intervals .x ; x C / with the new neighborhood
notation, N.x; /. For example, consider the proof that the union of open sets is also
an open set (Fig. 10.5).
PROOF: In metric space <X; d> assume that for each i in the index set I,
Ai is an open set. Then [ Ai is an open set.
i2I
In metric space <X; d> assume that for each i in the index set I, Ai is an
open set.
Let x 2 [ Ai .
i2I
i2I
10.4.1 Exercises
Write proofs for each of the following statements.
1. For every subset, S, of metric space <X; d>, int int.S/ D int.S/.
2. For every subset, S, of metric space <X; d>, @.@S/ @S.
3. For subsets S and T of metric space <X; d>, ext.S [ T/ ext.S/ \ ext.T/.
308
10 Metric Spaces
4. A subset S of a metric space <X; d> is open if and only if its complement
is closed.
5. In any metric space, the intersection of a finite number of open sets is an open set.
6. The subset S of a metric space is closed if and only if S D cl.S/.
7. For any subset S of a metric space, its derived set, S0 , is a closed set.
8. A set U is an open set in Rn using the taxicab metric if and only if U is open in
Rn using the Euclidean metric.
there is a > 0 such that whenever x 2 A with 0 < jx aj < , then jf .x/ Lj < .
This definition can now be extended to general metric spaces. Thus, if A and B are
metric spaces and f W A ! B, then if a 2 A and L 2 B, the limit lim f .x/ D L
x!a
means that for every > 0 there is a > 0 such that whenever x 2 N .a; /, then
f .x/ 2 N.L; / (Fig. 10.6).
For example, consider f W R2 ! R3 given by f .x; y/ D .x y; x C y; x2 C xy/
where R2 and R3 are the metric spaces using their respective Euclidean metrics.
Then lim f .x; y/ D .0; 2; 2/. Consider how you would prove this. The proof
.x;y/!.1;1/
would need to end with showing that the distance from .x y; x C y; x2 C xy/
to .0; 2; 2/ is less
p than some > 0 whenever .x; y/ is within some > 0 of
.1; 1/. That is, .x y 0/2 C .x C y 2/2 C .x2 C xy 2/2 < . At first this
inequality may appear quite intimidating, but keep in mind that when you write
proofs about limits, you have a great deal of flexibility because you do not have
to find an optimal value for > 0, just one that works. In particular, your task
would be done if you could simultaneously make jx yj small, jx C y 2j small,
f
a
X
Fig. 10.6 Limit of f W X ! Y as x approaches a is L
309
and jx2 C xy 2j small. In fact, if these three quantities were each less than 2 ,
q
p
2
then .x y 0/2 C .x C y 2/2 C .x2 C xy 2/2 would be less than 3 4 < .
So how can you arrange for each of p
jx yj, jx C y 2j, and jx2 C y2 2j to
be less than 2 when you know that .x 1/2 C .y 1/2 < ? One thing that
p
.x 1/2 C .y 1/2 < tells you is that each of jx 1j and jy 1j must be
less than because, if either of them exceeded , the square root of the sum of
their squares would also exceed . So look at each inequality separately. To make
jx yj < 2 , it would be enough to have jx 1j and jy 1j both less than 4 because
jxyj D j.x1/.y1/j jx1jCjy1j. Similarly, if jx1j and jy1j were both
less than 4 , then jxCy2j D j.x1/C.y1/j jx1jCjy1j would be less than
. As for jx2 Cxy2j you again want to write the expression using terms that include
2
factors of x1 or y1 so you can make those terms small. One way to do this would
be to write jx2 Cxy2j D j.x2 1/C.xy1/j D j.x1/.xC1/C.x1/yC.y1/j. As
with other limits of quadratic expressions with which you have dealt, it is convenient
to limit the size of so that x and y cannot grow too large. So, if is less than
1, both jxj and jyj will be bounded by 2, and jx C 1j and jy C 1j will each be
bounded by 3. Thus, it would be good enough to know that jx 1j and jy 1j do
not exceed 12
because then, jx2 C y2 2j D j.x 1/.x C 1/ C .x 1/y C .y 1/j
jx 1j jx C 1j C jx 1j jyj C jy 1j 12
3 C 12
2 C 12
D 2 . This results in
the following proof.
PROOF: If f .x; y/D.x y; x C y; x2 C xy/, then
lim
.x;y/!.1;1/
310
10 Metric Spaces
a3
a2
L
a1
(Fig. 10.7).
This raises an interesting question. Suppose <X; d1 > and <X; d2 > are both
metric spaces. Can you have a sequence <an > in X which converges to one limit
in metric d1 and converges to a different limit in metric d2 ? This differs from the
example in the previous paragraph where a sequence converges in one metric and
diverges in the other. The answer to this question is yes, you can have lim an D L
n!1
in the d1 metric, and lim an D M in the d2 metric with L M. For example, let
n!1
311
10.5.1 Exercises
Write proofs for each of the following statements.
1. Let a1 ; a2 ; a3 ; : : : be a sequence in the metric space <X; d>. Then lim an D a
n!1
x!a
x!a
312
10 Metric Spaces
PROOF: Let <A; dA > and <B; dB > be metric spaces, and let f W A ! B be
a function from A to B. Then f is continuous on A if and only if for every
open set D B, its preimage under f , f 1 .D/, is an open set in A.
Let <A; dA > and <B; dB > be metric spaces, and let f W A ! B be a function
from A to B.
Continuity implies that the preimages of open sets are open
Assume that f W A ! B is a continuous function.
Let D be an open set in B, and let a 2 f 1 .D/.
Because D is open in B,
thereis an > 0 such that N.f .a/; / D.
Thus, if y 2 B with dB f .a/; y < , then y 2 D.
Because f is a continuous function, there is a > 0 such that for all x 2 A
with dA .a; x/ < it follows that dB f .a/; f .x/
< .
Thus, if x 2 N.a; /, then f .x/ 2 N f .a/; , so f .x/ 2 D and x 2 f 1 .D/.
Therefore, f 1 .D/ is an open set in A proving that in preimage under f of
any open set is open.
The
set N f .a/; is an open set in B, so its preimage, C D fx 2 A f .x/ 2
N f .a/; g is an open set in A.
Because C is an open set containing a, there is a > 0 such that
N.a; / C.
Thus, if x 2 A with dA.a; x/ <
, then x 2 N.a; / C, so f .x/ 2 f .C/
implying that f .x/ 2 N f .a/; .
Therefore, f is continuous which completes the proof of the theorem.
Another theorem about continuous functions from the real numbers to the real
numbers is that the composition of two continuous functions is a continuous
function. This theorem generalized to metric spaces, but due to the previous result,
it now has a very simple proof which is not only valid for metric spaces, it is valid
for continuous functions between topological spaces.
PROOF: Suppose <X; dX >, <Y; dY >, and <Z; dZ > are metric spaces and
functions f W X ! Y and g W Y ! Z are both continuous functions. Then
g f W X ! Z is a continuous function.
Let <X; dX >, <Y; dY >, and <Z; dZ > be metric spaces, and let functions
f W X ! Y and g W Y ! Z both be continuous functions.
Let U Z be an open set.
Then, because g is a continuous function, the set g1 .U/ is an open set in Y.
(continued)
313
Then, because f is a continuous function, the set f 1 g1 .U/ is an open set
in X.
If x 2 f 1 g1 .U/ , then f .x/ 2 g1 .U/ and g f .x/ 2 U implying that
x 2 .g f /1 .U/.
Conversely, if x 2 .g
f /1 .U/, then .g f /.x/ D g f .x/ 2 U implying
1
that x 2 f 1
g .U/
.
Thus, f 1 g1 .U/ D .g f /1 .U/ is an open set, so g f is a continuous
function.
For some natural number n, consider functions that map a metric space <X; d>
into the metric space Rn using the Euclidean metric. In particular, if f W X ! Rn ,
then for each x 2 X, f .x/ 2 Rn . As a point in Rn , f .x/ has n coordinates
that
each depends on the value of x, that is, f .x/ D f1 .x/; f2 .x/; f3 .x/; : : : ; fn .x/ . It is
important that f is a continuous function if and only if fj is continuous for each
j D 1; 2; 3; : : : ; n. What will be the format of a proof of this result? First of all,
since the statement of the result is the biconditional f is continuous if and only if
fj is continuous for each j, the proof will have two parts. If it is assumed that f is
a continuous function, then you must show that fj is continuous for each j. But this
will not be hard since the Euclidean distance between two points f .x/ and f .a/ will
be greater than or equal to the distance between their jth coordinates, fj .x/ and fj .a/.
Thus, a > 0 that ensures that f .x/ is within > 0 of f .a/ will also insure that
fj .x/ is within of fj .x/. Conversely, if fj is continuous for each j, then a j > 0
can be found that will ensure that fj .x/ is very close to fj .a/. But by making each
jfj .x/ fj .a/j small, say smaller than n , the Euclidean distance from f .x/ to f .a/ will
be less than . This will be what is needed to show that f is continuous.
PROOF: Let n be a natural number, and suppose <X; d> is a metric
space. Let f be a function from X to n-dimensional Euclidean space, Rn ,
using the Euclidean metric. Then f is a continuous function if and only if
for each j D 1; 2 ; 3; : : : ; n the jth coordinate function for f , fj W X ! R, is
continuous.
Let n be a natural number, and suppose <X; d> is a metric space.
Let f be a function from X to n-dimensional Euclidean space, Rn , using the
Euclidean metric.
Continuity of f implies continuity of fj
Assume that f is a continuous function.
Let a 2 X, and let > 0 be given.
Because f is continuous, there is a > 0 such that for all x 2 X with
d.x; a/ < , the distance in Rn from f .x/ to f .a/ is less than .
(continued)
314
10 Metric Spaces
10.6.1 Exercises
Write proofs for each of the following statements.
1. Let <X; d> be a metric space and a 2 X. Then f .x/ D d.a; x/ is a continuous
function from X to R.
2. Let <X; d> be a metric space, and suppose f and g are both continuous functions
from X to R. Then the product f g is a continuous function from X to R.
3. Let d1 be the taxicab metric on R2 and d2 be the supremum metric on R2 .
Then f W <R2 ; d1 > ! <R2 ; d2 > given by f .x/ D x is a continuous function, and
so is its inverse.
4. Let d1 be the supremum metric in C0; 1 and d2 be the L1 metric in C0; 1.
Then f W <C0; 1; d1 > ! <C0; 1; d2 > given by f .x/ D x is a continuous
function.
10.7 Homeomorphism
315
5. Suppose <X; d1 > and <X; d2 > are both metric spaces, and that there are positive
real numbers c and C such that for all x and y in X, the distance function satisfies
cd1 .x; y/ d2 .x; y/ Cd1 .x; y/. Then a function is continuous on X with metric
d1 if and only if it is a function continuous on X with metric d2 .
6. The three metrics for Euclidean n-space: the Euclidean metric, the taxicab metric,
and the supremum metric are related in pairs as described in the previous
problem.
7. If <X; d> is a metric space and D X, then if f W D ! R is uniformly
continuous on D, there is a continuous function g W cl.D/ ! R such that
g.x/ D f .x/ for all x 2 D.
10.7 Homeomorphism
The interval 0; 1 and the interval 5; 30 certainly are different in length and
5
have different arithmetic properties such as 01 30
. On the other hand, the two
intervals have identical topological properties in the sense that there is a one-to-one
correspondence between the points of these two intervals such that a set is open in
the first interval if and only if the corresponding set is open in the second interval.
It can easily be seen that the function f .x/ D 25x C 5 is a bijection from 0; 1 to
5; 30 and a set U is open in 0; 1 if and only if f .U/ is open in 5; 30. This follows
from the fact that both f and its inverse function f 1 .x/ D x5
are continuous
25
and the fact that inverse images of open sets under continuous maps are open.
Two topological spaces X and Y are called homeomorphic if there is a continuous
bijection f W X ! Y whose inverse f 1 W Y ! X is also continuous. In such a case,
the bijection f is called a homeomorphism. If you know that two topological spaces
are homeomorphic, then you know that all of the topological properties of the two
spaces are the same. In particular, if X and Y are homeomorphic spaces, then if you
can prove that X has a particular property, it may follow immediately that Y also has
that property. Since metric spaces are topological spaces, one has that two metric
spaces <X; dX > and <Y; dY > are homeomorphic if there is a continuous bijection
from X to Y whose inverse is continuous. Thus, 0; 1 and 5; 30 are homeomorphic
with homeomorphism f .x/ D 25x C 5.
Note that if f W X ! Y is a continuous bijection, then the function f 1 W Y ! X
will be defined, but it need not be continuous. For example, the function f . / D
.cos ; sin / is a continuous bijection that maps the interval 0; 2/ on the real line
onto the circle radius 1 centered at the origin in R2 . The inverse is not a continuous
function because it maps every neighborhood of the point .1; 0/ 2 R2 to points that
are close to 2 and other points that are close to 0. This argument does not prove
that the interval is not homeomorphic to the circle; just that this particular function
is not a homeomorphism. A proof that the two spaces are not homeomorphic will
be discussed in the next section.
316
10 Metric Spaces
10.7.1 Exercises
Write proofs for each of the following statements.
1. The interval .0; 1/ in R is homeomorphic to the entire real line.
2. The open interval .0; 1/ is not homeomorphic to the closed interval 0; 1.
317
10.8.1 Exercises
Write proofs for each of the following statements.
1. If f is a continuous function from R to R with the property that f .x/ is a rational
number for each irrational x 2 R, then f is a constant function. (Hint: This is a
question of cardinality.)
2. Let Y be a set in R2 shaped like Y, that is, three line segments that share a
common endpoint. Then there is no continuous injective function that maps the
interval a; b in R onto Y.
3. The real line R is not homeomorphic to the plane R2 .
318
10 Metric Spaces
Since metric spaces are topological spaces, the concept of compactness can be
applied to metric spaces. That is, <X; d> is said to be a compact metric space if
every cover of X by open sets contains a finite subcover. It follows easily that if X is
compact, then every close subset of X is also compact. The proof of this is left as an
exercise.
Any set S in a metric space <X; d> is said to be bounded if there is a positive
real number r and a point a 2 X such that S N.a; r/. Equivalently, if S is not an
empty set, one can define the diameter of set S to be sup d.x; y/. Then S is bounded
x;y2S
if its diameter is finite. As it is in the real numbers, any compact set S in a metric
space <X; d> is both closed and bounded. To prove that a compact set S is closed,
suppose instead that a is an accumulation point S that is not contained in S. For each
real number r > 0 let Ur D fx j d.x; a/ > rg. Then S is contained in the union of the
open sets Ur sets, yet, because a is an accumulation point of S, no finite collection of
the Ur sets will cover S. Thus, if S is compact, it must contain all of its accumulation
points, so S is closed. To prove that a compact set S is bounded, let a be any point in
X. Then S is certainly contained in the collection of sets N.a; r/ where r ranges over
the positive real numbers. Thus, if S is compact, it is contained in a finite number
of the N.a; r/. If t is the largest of the r values associated with the finite number of
neighborhoods, then S is contained in N.a; t/ showing that S is bounded (Fig. 10.8).
a
a
S
S
X
319
0
if x nC1
=
<
1
n.n C 1/x n if nC1
< x < 1n for natural numbers n. Let S D ffn j n > 0g.
>
;
:
1
if 1n x
Then if m and n are distinct natural numbers, the distance between fm and fn is 1,
so the set S can have no accumulation point. Thus, S is a closed set. Each fn 2 S is
a distance 1 from all the other functions in S, so S has diameter 1, and, thus, S is a
bounded set. Note that for each natural number n the open neighborhood N.fn ; 12 /
320
10 Metric Spaces
Fig. 10.9 A continuous real-valued function on a compact set in R2 obtains its maximum and its
minimum
321
322
10 Metric Spaces
f
f(U)
U
f-1
X
10.9.1 Exercises
Write proofs for each of the following statements.
1. A closed subset of a compact metric space is compact.
2. A closed bounded subset of Rn using the Euclidean metric is a compact set.
(Hint: Use the fact that each n-dimensional cube is the union of 2n cubes that
each has side lengths half that of the original cube. Then if an open cover of a
cube has no finite subcover, then at least one of the 2n smaller cubes has no finite
subcover.)
3. A nonempty set S in a metric space <X; d> has a finite diameter if and only if
there is an a 2 X and a positive real number r such that S N.a; r/.
4. The BolzanoWeierstrass Theorem holds in a compact metric space, that is, if
<X; d> is a compact metric space and S is an infinite subset of X, then S has an
accumulation point.
323
5. Suppose set S in metric space <X; d> has diameter r. Show that cl.S/ also has
diameter r.
324
10 Metric Spaces
325
PROOF: Let <X; d> be a complete metric space with the property that
every closed bounded subset of X can be written as the union of a finite
number of closed sets each with a diameter that is at most half the
diameter of the original closed bounded subset. Then a subset of X is
compact if and only if it is both closed and bounded.
Let <X; d> be a complete metric space, and assume that every closed and
bounded subset of X can be written as the union of a finite number of closed
sets each with a diameter that is at most half the diameter of the original
closed bounded subset.
It has already been shown that every compact set in X is both closed and
bounded.
Assume that S0 X is both closed and bounded with diameter r.
Let T be a collection of open sets that covers S0 , and assume that no finite
collection of open sets in T covers S0 .
By assumption S0 can be written as a union of a finite number of closed
subsets each diameter at most 2r .
At least one of these finitely many subsets cannot be covered by finitely
many open sets in T. Call one of those sets S1 .
Clearly, S1 is not the empty set, or it could be covered by a single open set
in T.
Assume, inductively, that for some k > 0 the set Sk Sk1 has been chosen
with diameter at most 2rk such that Sk cannot be covered by a finite number
of open sets in T.
Sk can be written as the union of a finite number of closed subsets each with
r
diameter at most 2kC1
.
Because Sk has no finite subcover in T, at least one of these finitely many
subsets of Sk has no finite subcover in T. Call one of these sets SkC1 .
Thus, by mathematical induction, there is a nested sequence of sets S0
S1 S2 where each Sk cannot be covered by a finite number of open
sets in T and each Sk has diameter at most 2rk .
Because none of the Sk are the empty set and the sets are nested, the
intersection of any finite collection of Sk sets is nonempty.
From each Sk set select one element and call it ak .
Because the Sk sets are nested, for each natural number k and each n k,
an 2 Sk .
Let > 0 be given.
Select k such that 2rk < .
Then for all m; n k, am and an are in Sk , so d.am ; an / 2rk < .
This shows that the sequence <an > is Cauchy, and because <X; d> is a
complete metric space, the sequence has a limit L.
Because Sk is closed for each natural number k, and an 2 Sk for each n k,
it follows that the limit of the sequence, L, is also an element of Sk .
(continued)
326
10 Metric Spaces
>
0
if
x
<
=
nC1
1
fn .x/ D n.n C 1/x n if nC1
< x < 1n . This is an example of a closed bounded
>
:
;
1
if 1n x
set with diameter 1, but it is not possible to cover this set with a finite number of sets
with diameter 12 since each such set could contain at most one term of the sequence.
As seen before, C0; 1 is a metric space which contains closed bounded sets that are
not compact. It is left as an exercise to show that C0; 1 with the supremum metric
is complete.
Fig. 10.11 Enclosing a
closed bounded set in a grid
327
10.10.1 Exercises
Write proofs for each of the following statements.
1. The integers, Z, with the usual distance metric, d.x; y/ D jx yj, is a complete
metric space.
2. The space C0; 1 with the supremum metric is a complete metric space.
3. The space C0; 1 with the L1 metric is not a complete metric space. (Hint: Find
a Cauchy sequence of continuous functions that converge in the L1 metric to a
discontinuous function.)
4. In the metric space Rn with the Euclidean metric, a set is compact if and only if
it is both closed and bounded.
5. A closed set in R2 with diameter r > 0 can be enclosed in a square with side
length r.
6. If <an > is a sequence in a complete metric space <X; d> such that
1
P
d.an ; anC1 / is a convergent series, then the sequence <an > converges. Show
nD1
328
10 Metric Spaces
f
that the orbit is a Cauchy sequence. Since X is complete, the sequence must converge
to some point L 2 X. Note that if you begin the sequence at some other point y0 2 X,
its orbit also converges, and since d.x0 ; y0 /; d.x1 ; y1 /; d.x2 ; y2 /; : : : must converge to
0, it follows that the orbit of y0 must also converge to L. In particular, the orbit of
L must converge to L. Thefact that
from
f .L/ D L follows
the triangle inequality
which shows that d f .L/; L d f .L/; f .xn/ Cd f .x
/;
L
n
for any natural
number n,
and, in particular, for n large enough that d f .xn /; L and d f .xn /; f .L/ r d.xn ; L/
are small. The uniqueness of the fixed point
comes from
the simple fact that if L and
M are both fixed points, then d.L; M/ D d f .L/; f .M/ r d.L; M/ which can only
happen if d.L; M/ D 0 and L D M.
PROOF (Contraction Mapping Theorem): Every contraction mapping on
a complete metric space has a unique fixed point.
Let <X; d> be a complete metric space, f be a function mapping
X to X,
and r be a real number in 0; 1/ such that for all x; y 2 X, d f .x/; f .y/
r d.x; y/.
Select any x0 2 X and for each natural number k define xk D f .xk1 /.
Note that for any natural numbers m > n, the triangle inequality shows that
d.xm ; xn / d.xn ; xnC1 / C d.xnC1 ; xnC2 / C C d.xm1 ; xm /
rn
d.x0 ; x1 /rn C rnC1 C rnC2 C C rm1 d.x0 ; x1 / 1r
.
n
r
Since 1r goes to 0 as n gets large, d.xm ; xn / can be made as small as desired
by requiring m and n to be large which shows that the sequence <xn > is a
Cauchy sequence.
Because <X; d> is a complete metric space, the Cauchy sequence <xn >
has a limit L 2 X.
Since lim xn D L, given any > 0, there is a natural number N, such that
n!1
329
The power of contraction mappings is that one can find the fixed point of
f W X ! X by selecting any x 2 X and following its orbit which will converge
to the fixed point at least as fast as a geometric series. The reader may well be
familiar, for example, with a method of calculating the square root of a positive
real number a by starting with a positive number x and iterating the function
2 Ca
f .x/ D x 2x
, a formula which comes from applying Newtons Method to find
2 Ca
a root of the function x2 a.pThe fixed point of f is an x satisfying x D x 2x
which is satisfied by x D
a.pIf a D 2, for example, and you begin with
x D 100 as a first guess for 2, then you generate the orbit of 100 to be
100; 50:01; 25:024996; 12:55245805; 6:355894695; 3:335281609; 1:967465562;
1:49200089; 1:416241332; 1:414215014;
1:414213562; : : : with each successive
p
term being about half as far from 2 as the previous term. Why does this work?
First note that f maps the interval 1; 1/ to itself. On that interval the derivative
2
of f is f 0 .x/ D 2x
which has its maximum value at x D 1 where f 0 .1/ D 12 .
2x2
Thus,
by the Mean Value Theorem, when x > y 1, there is a c > 1 such that
f .x/f .y/
xy D jf 0 .c/j, so jf .x/ f .y/j < 12 jx yj showing that f is a contraction
mapping on 1; 1/.
Another familiar example, known to any student who has played around
with a calculator which can calculate the cosine function at the touch of a
button, is that the orbit of 0 under the cosine function (using radian measure)
is 0; 1; 0:540302306; 0:857553216; 0:65428979; 0:793480359; 0:701368774;
0:763959683; 0:722102425; 0:750417762; 0:731404042; : : : ; 0:739085132; : : : .
Again, this is because the function cos x is a contraction mapping on the interval
12 ; 1 because the derivative of cos x is sin x, and the Mean Value Theorem
guarantees that there is a c 2 12 ; 1 such that j cos x cos yj D jx yj sin c <
jx yj sin 1 where sin 1 < 1.
330
10 Metric Spaces
3
any real number. By differentiating y.x/ D Ce3x , you can verify that it does satisfy
the differential equation. It is typical that the solution of a first order equation will
contain an arbitrary constant such as the C that appears in this solution. This comes
from the fact that knowing a functions derivative only determines the function up to
an additive constant of integration. For this reason, a first order differential equation
is often stated by giving an initial condition y.a/ D b for some constants a and b,
because then, the arbitrary constant in the solution can be determined. For example,
for the equation given above, if it were required that y.1/ D 5, then the solution
3
would be y.x/ D e53 e3x . A differential equation along with an initial condition is
called an initial value problem.
In a first course in Differential Equations you learn a large number of techniques
that can be used to solve various differential equations. For example, the above
equation y0 D 9x2 y can be solved by first dividing both sides of the equation
0
by y to get yy D 9x2 and then integrating both sides to find ln jyj D 3x3 C K.
Exponentiating both sides of this equation and letting C equal either eK or eK gives
the previously stated solution. It comes as a surprise to the student of Differential
Equations that even though the first course in the field contains many techniques
for solving differential equations, subsequent courses contain very little about how
to solve equations and concentrate instead on finding numerical approximations to
solutions and on theorems telling when solutions are expected to exist. One very
powerful theorem which shows there exist unique solutions to a fairly large class of
initial value problems is the Picard Existence Theorem which applies to equations
of the form y0 D F.x; y/ with initial condition y.a/ D b. The theorem requires that
there exists a compact set R R2 surrounding the initial point .a; b/ such that F is
continuous on that compact set R, and there is a constant M such that for any points
.x; y1 / and .x; y2 / in R, the function F satisfies jF.x; y1 / F.x; y2 /j Mjy1 y2 j.
This condition on the second variable of F is known as a Lipschitz condition. The
theorem concludes that there is a > 0 and a function y.x/ defined on the interval
a ; a C such that y.x/ is the unique solution to the initial value problem
y0 D F.x; y/ and y.a/ D b.
The theorem can be proved by noticing that y is a solution to the given initial
Rx
value problem if and only if y satisfies the equation y.x/ D b C F t; y.t/ dt.
a
Rx
But the function G.y/ D b C F t; y.t/ dt turns out to be a contraction mapping
a
331
Rx
jF t; y1 .t/ F t; y2 .t/ j dt Mjy1 .t/ y2 .t/j dt
contraction mapping on C.
Thus, by the Contraction Mapping Theorem, there is a unique y 2 C such
Rx
that y D G.y/ D b C F t; y1 .t/ dt.
a
332
10 Metric Spaces
Of course, this theorem only guarantees that there is a function y.x/ satisfying
the
a ; a C . But if the points
differential equation
in a small neighborhood
a ; y.a / and a C ; y.a C / are in the interior of the compact set R, the
theorem can be applied again to extend the function. As a result, a function y.x/ can
be constructed whose graph extends toward the boundary of R.
Consider the problem of finding a solution to the first order differential equation
p
y0 D xy2 C y with the initial condition y.0/ D 2. Common techniques for solving
differential equations do not yield a solution to this equation. Yet Picards Existence
Theorem shows that the equation has a unique solution. In particular, let R be the
rectangle of points satisfying jxjp 2 and jy 2j 1. On that rectangle the function
p
1
xy2 C y is bounded by 2 32 C 3 < 20 D K. Thus, you can choose D 20
so that
p
p
2
2
2 and K D 1 1.Then jF.x; y1 / F.x; y2 /j D jx.y1 y2 / C . y1 y2 /j D
2 dt D 2 C 2x
0
Zx
y2 .x/ D G.y1 / D 2 C
2 C 2t dt D 2 C 2x C 2
0
Zx
y3 .x/ D G.y2 / D 2 C
2 C 2t C 2
0
Zx
y4 .x/ D G.y3 / D 2 C
2 C 2t C
0
x2
2
x2
x3
x2
dt D 2 C 2x C 2 C 2
2
2
3
x3
x2
x3
x4
x2
C 2 dt D 2 C 2x C 2 C 2 C 2 :
2
3
2
3
4
This shows that the orbit of y0 is just the partial sums of the power series for 2ex
centered at 0, a satisfying result even if it is not surprising.
333
10.11.3 Fractals
There is a beautiful application of contraction mappings in the study of fractals.
It is difficult to define what a fractal is, but you are invited to look at the large
number of pictures of fractals available in books and on the Internet to get a feel
for what they are. Loosely, they are sets that show repeated structure at every level
of magnification so that if you look at a very small piece of the set, it includes
a structure that is either geometrically similar or nearly similar to the entire set.
Famous examples are the Mandelbrot set and the Sierpinski triangle as well as
the Cantor set already discussed in Chap. 6. Benoit Mandelbrot invented the word
fractal in 1975 to describe the phenomenon he observed in the set that now bares
his name (Figs. 10.13 and 10.14).
One method of generating fractals is based on contraction mappings. In this
method one sets up a complete metric space <H; h> whose elements are compact
sets in R2 . Then by selecting an appropriate contraction mapping f W H ! H and
any nonempty compact set A 2 H, the orbit of A under the mapping f will converge
to the unique fixed point of the contraction mapping which is the desired fractal. To
do this let H be the collection of all nonempty compact subsets of R2 . To place a
metric on H, consider first the problem of finding the distance from an individual
point x 2 R2 to a set A 2 H. If R2 has Euclidean metric d, then since for every fixed
y 2 B the distance d.x; y/ is a continuous function of x, it follows that d.x; y/ obtains
Fig. 10.13 The Mandelbrot
set (black)
334
10 Metric Spaces
a minimum value on the compact set A, which is min d.x; y/. Then, if A and B are
x2A
both elements of H, define h .A; B/ D max min d.x; y/ which is the largest distance
y2B x2A
any point of B is from the set A. Again, since the Euclidean distance function is
continuous, and B is compact, this min d.x; y/ obtains a maximum value on B, and
x2A
there are points x 2 A and y 2 B such that d.x; y/ D h.A; B/. Because h.A; B/
need not be the same as h .B; A/, define h.A; B/ D max h .A; B/; h .B; A/ . This
distance function is known as the Hausdorff metric.
For example, let A be the closed disk consisting of all the points in R2 within
1 of the origin, and let B be the closed disk consisting of all the points in R2
within
2 of the origin. Clearly, A B. As a result h .B; A/ D
0. But h .A; B/ D
d .2; 0/; .1; 0/ D 1, so h.A; B/ D max h .A; B/; h .B; A/ D max.1; 0/ D 1.
Intuitively, there are points in B that are a distance 1 from the set A, and this is
the largest distance from points in B to the set A. As a second example, let A be
the square with vertices at .1; 1/; .1; 1/; .1; 1/; and .1; 1/, and let B be the
line segment from .1; 2/ to .1; 2/. Since each point in B is a distance 1 from
A, h .A; B/ D 1. Since the largest distance from a point in A to the set B is
d .0; 1/; .0; 2/ D 3, h .B; A/ D 3. Thus, h.A; B/ D 3.
What does it take to show that <H; h> a metric space? Because H is clearly a
nonempty set, and h.A; B/ is defined for all A and B in H, one merely has to verify
the conditions for h to be a metric. The fact that h is always a nonnegative real
number follows immediately from its definition, as does the fact that h.A; B/ D
h.B; A/. Is it true that for all A 2 H that h.A; A/ D 0? Yes, this follows from the fact
that if y 2 A, then min d.x; y/ D 0, so h.A; A/ D h .A; A/ D max min d.x; y/ D 0.
x2A
y2A x2A
min d.x; y/ > 0, so either h .A; B/ > 0 or h .B; A/ > 0, and h.A; B/ > 0.
y2B
Showing that h satisfies the triangle inequality is a little trickier and relies on a
careful look at the definitions of h and h. Let A, B, and C be elements of H. To
show that h.A; C/ h.A; B/ C h.B; C/, begin by showing that for every a 2 A,
min d.a; c/ h.A; B/ C h.B; C/. So if a 2 A, then it is true for all b 2 B
c2C
that min d.a; c/ min d.a; b/ C d.b; c/ D d.a; b/ C min d.b; c/. In particular,
c2C
c2C
c2C
this is true for b 2 B which minimizes d.a; b/, so min d.a; c/ d.a; b / C
c2C
min d.b ; c/ min d.a; b/ C max min d.b; c/ h .B; A/ C max min d.b; c/
c2C
b2B
b2B c2C
b2B c2C
h.A; B/ C h .C; B/ h.A; B/ C h.B; C/. Since all the distances min d.a; c/ are
c2C
bounded by h.A; B/Ch.B; C/, its maximum also has the same bound, so h .A; C/
h.A; B/ C h.B; C/. The same argument shows that h .C; A/ h.A; B/ C h.B; C/, so
h.A; C/ h.A; B/ C h.B; C/ which is the desired triangle inequality.
335
h .A; A/ D max 0 D 0.
a2A
If A and B are in H with A B, then it cannot be the case that both AnB
and BnA are empty, so there is an x in at least one of the two sets.
If x 2 AnB, then d.x; b/ > 0 for all b 2 B, and since d.x; b/ is a continuous
function of b and B is compact, min d.x; b/ > 0 and h .B; A/ > 0.
b2B
Similarly, if x 2 BnA, then h .A; B/ > 0. In either case, h.A; B/ > 0. This
shows that h separates points of H.
It remains to show that h satisfies the triangle inequality, so let A, B, and C
be elements of H.
Let a 2 A.
It is true for all b 2 B that min d.a; c/ min d.a; b/ C d.b; c/ D
c2C
c2C
c2C
b2B
c2C
b2B c2C
it follows that max min d.a; c/ also has the same bound, so
a2A c2C
336
10 Metric Spaces
Moreover, <H; h> is a complete metric space. A natural way to prove this
completeness is to take a Cauchy sequence of sets in H and show that this sequence
converges in the h metric to some set L 2 H. So let <An > be a Cauchy sequence in
H. The strategy for this proof is to construct another sequence of sets, <Tn >, so that
each Tm is close to one of the terms of the <An > sequence. Then it is shown that
the <Tn > sequence converges to a set L, and since the terms of the <An > sequence
are close to the terms of the <Tn > sequence, the <An > sequence will also converge
to L.
Because <An > is a Cauchy sequence, for each natural number n, there is an Nn
such that for every k and m greater than or equal to Nn , h.Ak ; Am / < 21n . Then define
Tn D fx 2 R2 j min d.a; x/ 21n g. Tn can be thought of as a cloud or halo around
a2An
the set ANn . In particular, Tn contains ANn and all of the points within 21n of An . Since
h.Am ; ANn / < 21n for all m Nn , it follows that Am Tn for all m Nn . It is also
clear that since Tn contains ANnC1 , that Tn contains TnC1 , and <Tn > is a sequence of
nested sets. It is also easy to show that Tn is nonempty, bounded, and closed so that
Tn 2 H.
1
Now you can define L D \ Tn . This set is not empty because if you intersect a
nD1
nested sequence of nonempty compact sets, you get a nonempty set. L is bounded
because it is contained in the bounded set T1 , and it is closed because it is the
intersection of closed sets. Thus, L 2 H. To show that lim Tn D L, you would
n!1
But this is enough to show that L D lim Tn . It is then a simple argument to show
n!1
that lim An D L which proves that all Cauchy sequences in H converge, so H is
n!1
complete.
PROOF: The metric space <H; h> is complete.
Let R2 be the metric space with Euclidean distance function d.x; y/.
2
Let H be the space
of all nonempty
compactsubsets of R with the metric
h.A; B/ D max h .A; b/; h .B; A/ where h .A; B/ D max min d.a; b/.
b2B a2A
337
a2ANn
1
, so
2n
so Tn is closed.
Since Tn is a nonempty, bounded, closed set in R2 , it is a nonempty compact
set and Tn 2 H.
Note that for all natural numbers n, Tn contains Ak for all k Nn , and Tn
contains Tk for all k n.
1
Let L D \ Tn .
nD1
The intersection of any finite collection of Tn sets contains the smallest set
in the collection, so the intersection of any finite collection of Tn sets is
nonempty. Thus, the intersection of all the Tn sets cannot be empty, and L is
nonempty.
L is the intersection of closed sets, so it is closed. L is contained in the
bounded set T1 , so L is bounded. Thus, L 2 H.
Let natural number n be given.
L Tn , so h .Tn ; L/ D 0.
To estimate h .L; Tn /, note that for every natural number m the triangle
inequality shows that h.Tm ; TmC1 / h.Tm ; ANm / C h.ANm ; ANmC1 / C
1
1
< 2m2
.
h.ANmC1 ; TmC1 / 21m C 21m C 2mC1
1
Thus, for any x 2 Tm there is a y 2 TmC1 such that d.x; y/ < 2m2
.
1
Let xn 2 Tn . Then there is an xnC1 2 TnC1 with d.xn ; xnC1 / < 2n2
. By
mathematical induction there is a sequence xn ; xnC1 ; xnC2 ; : : : such that for
1
each m n, xm 2 Tm and d.xm ; xmC1 / < 2m2
.
If k and m are natural numbers with m > k, then d.xk ; xm /
d.xk ; xkC1 / C d.xkC1 ; xkC2 / C d.xkC2 ; xkC3 / C C d.xm1 ; xm / <
1
1
1
1
C 2k1
C 21k C C 2m3
< 2k3
.
2k2
1
Because lim 2k3 D 0, the sequence <xm > is a Cauchy sequence in R2 , so
k!1
it converges to a point x in R2 .
For each k n and each m k, xm 2 Tk , and since Tk is a closed set, it
1
Since lim xm D x, for every > 0 there is an m > n such that d.xm ; x/ < .
m!1
This shows that d.xn ; x/ d.xn ; xnC1 / C d.xnC1 ; xnC2 / C d.xnC2 ; xnC3 / C
1
1
C d.xm1 ; xm / C d.xm ; x/ < 2n3
C . This shows that d.xn ; x/ 2n3
,
1
1
so h .L; Tn / D max min d.a; b/ 2n3 . Thus, h.Tn ; L/ 2n3 .
b2Tn a2L
(continued)
338
10 Metric Spaces
n!1
1
Given > 0, let k be a natural number such that 2k5
< .
Then for every m Nk , h.Am ; L/ h.Am ; ANk / C h.ANk ; Tk / C h.Tk ; L/ <
1
1
1
C 21k C 2k3
< 2k5
< .
2k
Therefore, lim An D L which proves that all Cauchy sequences in H
n!1
converge, so H is complete.
Knowing that <H; h> is a complete metric space shows that every contraction
mapping on H has a unique fixed point. Some fractals can be generated as the result
of being a fixed point of such a contraction mapping f W H ! H. In the words of
the study of fractals, these fixed points are called attractors. What it means is that
if you begin with any nonempty
compact
set A, the orbit of A under the contraction
mapping, A; f .A/; f f .A/ ; f f f .A/ ; : : : , will converge to this attractor.
Where can you find a contraction mapping f W H ! H with an interesting fractal
as its attractor? The first thing to notice is that if f W R2 ! R2 is a contraction
mapping on R2 , then the usual extension of f to subsets of R2 given by f .A/ D
fy 2 R2 j y D f .x/ for some y 2 R2 g gives a function f W H ! H that is a
contraction mapping on H. To see this, let d be the usual Euclidean distance metric
on R2 , let k be a positive real number
less than 1, and let f W R2 ! R2 be a
contraction mapping satisfying d f .x/; f .y/ k d.x;y/ for all x and y in R2 .
Then, if A and B are in H, it follows that h f .A/; f .B/ D max min d.a; b/ D
b2f .B/ a2f .A/
max min d f .a/; f .b/ max min k d.a; b/ D k h .A; B/, and it follows that
b2B
b2B a2A
a2A
h f .A/; f .B/ k h.A; B/.
Next, suppose that f1 ; f2 ; f3 ; : : : ; fs is a finite set of contraction mappings on R2
(therefore on H) with associated contraction constants k1 ; k2 ; k3 ; : : : ; ks , respecs
339
a1
0:5
0:5
0:5
a2
0
0
0
b1
0
0
0
b2
0:5
0:5
0:5
c1
0
0:5
0:25
c2
0
0
p
0:25 3
a1
a2
b1
0
0:85
0:2
0:15
0
0:04
0:26
0:28
0
0:04
0:23
0:26
b2
0:16
0:85
0:22
0:24
c1
0
0
0
0
c2
0
1:6
1:6
0:44
Figure 10.15 shows what happens when this transformation is applied to a simple
equilateral triangle in the plane as shown by the first picture in the figure. Applying
the function once results in the second picture showing that the triangle is mapped
into four separate skewed copies by the four pieces of the transformation. The third
picture shows what happens to this after a second iteration of the transformation.
By the time the transformation has been iterated three times, the image set already
appears to be a recognizable object quite different from the original triangle. The
figure also shows what the set looks like after 5, 10, 15, and 20 iterations. After 20
iterations, the image appears to be quite close to the attractor for this transformation
and does look a great deal like a fern.
Fig. 10.15 Generation of a fractal fern by applying a contraction mapping to a triangle. Shown is
the original triangle and its image after 1, 2, 3, 5, 10, 15, and 20 iterations of the transformation
340
10 Metric Spaces
There are many fractal generating programs available either as web applications
or as stand-alone programs. The interested reader may well find it useful to
investigate the convergence properties of the transformations discussed here and
other similar transformations obtainable by either small or perhaps large changes in
the given affine maps.
10.11.4 Exercises
Write proofs for each of the following statements.
1. If f W X ! X and g W X ! X are both contraction mappings on X, then the
composition g f is also a contraction mapping on X.
2. The function f .x/ D xC2
is a contraction mapping on R with fixed point L D 2.
2
3. Applying Newtons Method to solve the equation ex D x C 2 yields the function
x
f .x/ D x e ex2
x 1 . This function has a unique fixed point on the interval 0; 2.
4. Let f be a differentiable function from a; b into a; b such that for some M < 1
the derivative of f satisfies jf 0 .x/j M for each x 2 a; b. Then f has a unique
fixed point.
5. Show that if the function f W R ! R satisfies the Lipschitz condition
jf .x/ f .y/j 5jx yj for x and y in R, then f is uniformly continuous on
R.
6. Show that if function f W R ! R is differentiable, and its derivative is a bounded
function, then f satisfies a Lipschitz
condition on R.
p
xCy
0
7. The differential equation y D xC2 with initial condition y.1/ D 3 has a unique
solution on the interval 0; 2.
8. If A; B; C, and D are in H, then h.A [ B; C [ D/ max h.A; C/; h.B; D/ .
9. If f1 and f2 are contraction mappings on H with contraction constants k1 and k2 ,
respectively, then F.A/ D f1 .A/ [ f2 .A/ is a contraction mapping on H with
contraction constant k D max.k1 ; k2 /.
341
342
Infinite Series
Bonar, D.D., Khoury, M.J.: Real Infinite Series. Mathematical Association of
America, Washington, DC (2006)
Knopp, K.: Theory and Application of Infinite Series. Dover, Mineola (1990)
Writing Proofs
Lay, S.R.: Analysis with an Introduction to Proof, 5th edn. Pearson, New York
(2013)
Solow, D.: How to Read and do Proofs: An Introduction to Mathematical Thought
Processes, 6th edn. Wiley, New York (2014)
Differential Equations
Birkhoff, G., Rota, G.-C.: Ordinary Differential Equations, 4th edn. Wiley, New
York (1989)
Kelley, W.G., Peterson, A.C.: The Theory of Differential Equations. Springer,
New York (2010)
Fractals
Barnsley, M.F.: Fractals Everywhere. Dover, Mineola (2012)
Devaney, R.L.: Chaos Fractals, and Dynamics: Computer Experiments in Modern
Mathematics. Addison-Wesley, Boston (1991)
Index
A
Abels theorem, 256
absolute extremum, 145
absolute maximum, 145
absolute minimum, 145
absolute value, 38, 121
absolutely convergent series, 205
accumulation point, 74, 286, 306, 308, 318,
319
addition property of less than, 32
affine transformation, 338
Algebra, 4
alternating series, 219
alternating series test, 219
Analysis, 4
analytic function, 259
antecedent, 9
antiderivative, 193
Archimedian principle, 36
area, 159
area axioms, 169
area zero, 165
assert the hypothesis, 14
associative law of intersection, 19
associative law of union, 19
associative laws, 32
attractors, 338
axiom, 2, 7, 3537, 63, 75, 128, 168, 169, 201,
323
axioms for a field, 32
axioms for an ordered field, 32
axioms for area, 166
axioms for the real numbers, 36, 201
B
Barnsley fern, 339
biconditional, 11, 96, 184, 195, 313
bijection, 42, 142, 159, 161, 283, 315, 321, 322
bijective, 42
BolzanoWeierstrass theorem, 75, 76, 78, 124,
125, 322
bound, 35
boundary, 270, 286, 306
bounded, 35
bounded above, 35, 62
bounded below, 35, 62
bounded derivative, 108
bounded function, 62, 123
bounded metric space, 318
bounded sequence, 63
C
C[0,1], 302
Cantor set, 163, 166, 333
cardinality, 159
Cauchy product, 231, 265
Cauchy sequence, 66, 67, 77, 78, 204206,
210, 223, 323, 324, 328, 336
Cauchy sequences converge, 77
CauchySchwarz inequality, 297
chain rule, 140
closed function, 285
closed set, 274, 306
closure of a set, 287, 306
closure properties, 32
codomain, 40, 42
343
344
commutative law of intersection, 19
commutative law of union, 19
commutative laws, 32
compact metric space, 318
compact set, 288
compactness, 288, 318
comparison test, 209, 212, 214, 216
complement of a set, 18
complete metric space, 323
completeness axiom, 3537, 63, 75, 128, 323
composition of continuous functions, 312
composition of functions, 43
conclusion, 9
conditional, 9
conditionally convergent series, 206
conjecture, 2
connected metric space, 316
connected set, 39, 291
connectedness, 291, 316
consequent, 9
continuity of a composition, 121
continuity of a difference, 118
continuity of a product, 119
continuity of a quotient, 119
continuity of a sum, 118
continuous at a point, 100, 134
continuous function on a metric space, 311
continuous function on a topological space,
285
continuous functions are integrable, 189
continuous functions on a set, 302
continuous on a set, 100
contraction mapping, 327, 329331, 333, 338
contradiction, 10, 36, 70, 90, 112, 123, 126,
161, 177, 178, 196, 222, 286, 317, 318,
324
contrapositive, 10
converge pointwise, 239
converge uniformly, 241
convergence in L1 , 252
convergence in mean, 252
convergent sequence, 62
convergent series, 202
converse, 9
corollary, 2
countable, 160, 163, 165, 195, 197, 200
counterexample, 11
cowgirl, 242
critical point, 145
D
decreasing on an interval, 143
decreasing sequence, 60
Index
definition, 1
deleted neighborhood in a metric space,
296
DeMorgans Laws, 19, 25
denumerable, 160
derivative, 133
derivative of a difference, 137
derivative of a product, 138
derivative of a quotient, 138
derivative of a sum, 137
derivative of power series, 259
derived set, 286, 306
diagonalization argument, 160
diameter of a metric space, 318
difference of power series, 265
difference of sets, 18
difference quotient, 133143, 148152,
192
differentiable, 133
differential equation, 329332
disconnected metric space, 316
disconnected set, 291
discontinuities, 197
discontinuous, 100, 131
discrete metric, 302
discrete metric space, 302
discrete topology, 282
distance, 38
distance function, 296
distributive law of intersection over union,
19
distributive law of multiplication over addition,
32
distributive law of union over intersection, 19
divergent sequence, 62
divergent series, 202
division algorithm, 28
domain, 40
dot product, 297
E
element, 17
empty set, 18
Euclidean distance, 299
Euclidean metric, 300
Euclidean plane, 299
Euclidean space, 299
even integers, 27
existence of square roots, 36
existential quantifier, 69
extended mean value theorem, 152
exterior, 270, 306
extreme value, 126
Index
F
fern, 339
field, 32
field axioms, 32
finite complement topology, 281
finite set, 160
finite subcover, 110114, 196, 249, 251, 288,
290, 291, 317, 324, 326
first order differential equation, 329
fixed point, 327
fractal, 333340
function, 40, 41
function applied to sets, 283
fundamental question of Analysis, 267
fundamental theorem of Calculus, 192, 193
G
geometric series, 202, 203, 207, 212, 214, 231,
253, 255, 327, 329
greatest lower bound, 35, 94, 110, 185, 186
H
harmonic series, 212, 223
Hausdorff metric, 334
HeineBorel theorem, 110, 115, 196, 247, 288,
319, 324
higher order Rolles theorem, 263
homeomorphic metric spaces, 315
homeomorphism, 315, 321
hypothesis, 9
I
identity properties, 32
image of a set, 283
improper integral of the first kind, 215
improper integral of the second kind, 215
inconsistent axioms, 3
increasing on an interval, 143
increasing sequence, 60
independence of the axioms, 3
index set, 279
infimum, 36
infinite limit, 79
infinite series, 201237, 252, 255
inherited topology, 282
initial value problem, 330
injection, 42, 44, 142, 283, 316, 317, 321
injective, 42
integers, 27
integrable function, 171200, 239, 244, 248,
252
345
integral, 169, 171
integral of continuous function, 189
integral test, 216
interior, 269, 306
intermediate value property, 127130,
155157
intermediate value property for derivatives,
155157
intermediate value theorem, 128
intersection of closed sets, 280
intersection of open sets, 278
intersection of sets, 18
interval, 39
interval of convergence, 256
inverse function, 136, 142, 143, 283, 315, 317,
321
inverse properties, 32
L
L1 metric, 303, 305, 310
LHopitals rule, 150
Lagranges form of Taylors theorem, 263
law, 2
least upper bound, 3537, 63, 75, 94, 111, 123,
126, 128, 129, 186, 323
least upper bound principle, 36
Lebesgue measure, 248
Lebesgues theorem, 197
lemma, 2
less than, 32
limit at a point, 49
limit at infinity, 57
limit at negative infinity, 58
limit comparison test, 218
limit from the left, 54
limit from the right, 54
limit inferior, 94
limit of a difference, 81, 82
limit of a polynomial, 88
limit of a positive function, 90
limit of a product, 81, 84
limit of a quotient, 81, 86
limit of a rational function, 88
limit of a sequence, 62
limit of a sum, 81, 82
limit of terms test, 203
limit superior, 94
limit, infinite, 79
limits are unique, 91
limits in a metric space, 308
Lipschitz condition, 330
list implications, 15
local property, 105
346
lower bound, 35
lower step function, 184
lowest terms, 31
M
M-test, 253
Mandelbrot set, 333
mathematical induction, 35, 64, 87, 92, 167,
212, 262, 291, 299
maximum, 122
maximum value, 123, 126
mean value theorem, 108, 146, 147, 152
mean value theorem for integration, 191
measure zero, 163, 195, 200, 244
method of exhaustion, 167
metric, 296
metric space, 296340
minimum, 122
minimum value, 123, 126
Minkowski inequality, 298, 299
monotone convergence theorem, 248
monotone decreasing sequence, 60
monotone increasing sequence, 60
monotone sequence, 60
monotonically decreasing sequence of
functions, 246
monotonically increasing sequence of
functions, 246
multiplication property of less than, 32
multiset, 18
N
n-dimensional Euclidean space, 299
natural numbers, 27
negation of a statement, 10
negation of statements with quantifiers,
69
negation symbol, 10
neighborhood in a metric space, 296
norm of a partition, 170
O
odd integers, 27
one-sided limits, 54
one-to-one, 42
onto, 41
open cover, 110, 288, 317
open function, 283
open set, 274, 306
orbit, 327
ordered field, 32
Index
P
p-series, 217
paradox, 3
parentheses in series, 224
partial sums, 202
partition, 28
partition of an interval, 170
path-connected set, 292
Picard existence theorem, 330
pointwise convergence, 239
positive integers, 27
postulate, 2
power series, 201, 215, 253, 255267, 332
power series antiderivative, 262
power series convergence, 255
power series difference, 265
power series differentiability, 259
power series interval of convergence, 256
power series product, 265
power series quotient, 265
power series sum, 265
preimage of a set, 283
principle, 2
product of power series, 265
product rule, 138
proof, 2
proof by contradiction, 10, 36, 70, 90, 112, 123,
126, 161, 177, 178, 196, 222, 286, 317,
318, 324
proof of conditional statements, 11
proof template, 5, 12, 20, 23, 41, 42, 49, 70,
100, 106, 296
proofs about even and odd, 27
proofs about set equality, 22
proofs about subsets, 19
proposition, 2
proving a derivative, 135
proving a limit, 49
proving a metric space, 296
proving continuity, 100
proving no limit exists, 70
proving uniform continuity, 106
Q
quantifier, 69
quotient, 28
quotient of power series, 265
quotient rule, 138
R
radius of convergence, 255
range, 40
Index
ratio test, 212215, 221, 231, 258
rational numbers, 31
rearrangement of terms of a series, 225
relative extremum, 144
relative maximum, 144
relative minimum, 144
remainder, 28
Riemann integral, 169, 171
Riemann sum, 171, 173, 175, 177, 178, 182,
184, 186, 190, 194
Rolles theorem, 146, 263
root test, 214, 215, 258
Russells paradox, 3
S
saddle point, 145
secant line, 133
sequence, 60
sequence of continuous functions, 243
sequence of functions, 239
sequence of integrable functions, 244
sequence of partial sums, 202
sequence of terms, 201
series, 201237, 252, 255
series diverges, 202
series of functions, 252
series rearrangement, 225
series with parentheses, 224
set, 17
set builder notation, 18
set containment, 17
set difference, 18
set equality, 22
set notation, 17
set of discontinuities, 197
set the context, 13
Set Theory, 4
shorthand symbols for proofs, 11
Sierpinski triangle, 333
slope, 133
squeezing theorem, 91
state the conclusion, 14
statement, 1
step function, 183
strictly decreasing on an interval, 143
strictly decreasing sequence, 60
strictly increasing on an interval, 143
strictly increasing sequence, 60
subcover, 110114, 196, 249, 251, 288, 290,
291, 317, 324, 326
subsequence, 61, 92
subset, 17
sum of power series, 265
347
sup metric, 302
supremum, 36
supremum metric, 301, 302
surjection, 4143, 142, 283
surjective, 41
symbols for proofs, 11
T
tangent line, 133
taxicab metric, 300
Taylor polynomial, 263
Taylor series, 263
Taylors theorem, 263
telescoping series, 203
template, 5, 12, 20, 23, 41, 42, 49, 70, 100, 106,
296
term of a sequence, 60
theorem, 2
Thomaes function, 131, 200
topological space, 281
Topology, 4
topology, 269, 281
transitive property, 32
triangle inequality, 38, 39, 66, 71, 82, 107, 115,
118, 295297, 299, 300, 302304, 306,
328, 334, 337
trichotomy property, 32, 37, 38
U
uncountable, 160, 163, 279
uniform convergence, 241
uniformly continuous, 106, 115, 311, 327
union of closes sets, 280
union of open sets, 278, 307
union of sets, 18
universal quantifier, 69
universal set, 18
upper bound, 35
upper step function, 184
W
Weierstrass M-test, 253
without loss of generality, 80, 128, 151, 153,
156, 174, 186, 190, 194, 221, 248, 250,
257, 273
WLOG, 80, 128
working definition, 28
Z
ZermeloFraenkel axioms, 3