Rudin Companion Mathematical Analysis - Silvia

COMPANION NOTES
A Working Excursion to Accompany Baby

Rudin
Evelyn M. Silvia1
April 1, 1999
1
hEvelyn
c M. Silvia 1999
ii
Contents
Preface vii
0.0.1 About the Organization of the Material . . . . . . . . . . . viii
0.0.2 About the Errors . . . . . . . . . . . . . . . . . . . . . . . viii
1 The Field of Reals and Beyond 1

1.1 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Special Subsets of an Ordered Field . . . . . . . . . . . . . 15
1.2.2 Bounding Properties . . . . . . . . . . . . . . . . . . . . . 17
1.3 The Real Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3.1 Density Properties of the Reals . . . . . . . . . . . . . . . . 27
1.3.2 Existence of nth Roots . . . . . . . . . . . . . . . . . . . . 29
1.3.3 The Extended Real Number System . . . . . . . . . . . . . 33
1.4 The Complex Field . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.4.1 Thinking Complex . . . . . . . . . . . . . . . . . . . . . . 39
1.5 Problem Set A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2 From Finite to Uncountable Sets 49

2.1 Some Review of Functions . . . . . . . . . . . . . . . . . . . . . . 49
2.2 A Review of Cardinal Equivalence . . . . . . . . . . . . . . . . . . 54
2.2.1 Denumerable Sets and Sequences . . . . . . . . . . . . . . 61
2.3 Review of Indexed Families of Sets . . . . . . . . . . . . . . . . . 62
2.4 Cardinality of Unions Over Families . . . . . . . . . . . . . . . . . 65
2.5 The Uncountable Reals . . . . . . . . . . . . . . . . . . . . . . . . 69
2.6 Problem Set B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3 METRIC SPACES and SOME BASIC TOPOLOGY 73

3.1 Euclidean n-space . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
iii
iv CONTENTS
3.2 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.3 Point Set Topology on Metric Spaces . . . . . . . . . . . . . . . . . 82
3.3.1 Complements and Families of Subsets of Metric Spaces . . 87
3.3.2 Open Relative to Subsets of Metric Spaces . . . . . . . . . 96
3.3.3 Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3.4 Compactness in Euclidean n-space . . . . . . . . . . . . . . 105
3.3.5 Connected Sets . . . . . . . . . . . . . . . . . . . . . . . . 111
3.3.6 Perfect Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 114
3.4 Problem Set C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4 Sequences and Series–First View 123

4.1 Sequences and Subsequences in Metric Spaces . . . . . . . . . . . 124
4.2 Cauchy Sequences in Metric Spaces . . . . . . . . . . . . . . . . . 135
4.3 Sequences in Euclidean k-space . . . . . . . . . . . . . . . . . . . 139
4.3.1 Upper and Lower Bounds . . . . . . . . . . . . . . . . . . 145
4.4 Some Special Sequences . . . . . . . . . . . . . . . . . . . . . . . 149
4.5 Series of Complex Numbers . . . . . . . . . . . . . . . . . . . . . 152
4.5.1 Some (Absolute) Convergence Tests . . . . . . . . . . . . . 156
4.5.2 Absolute Convergence and Cauchy Products . . . . . . . . 169
4.5.3 Hadamard Products and Series with Positive and Negative
Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.5.4 Discussing Convergence . . . . . . . . . . . . . . . . . . . 176
4.5.5 Rearrangements of Series . . . . . . . . . . . . . . . . . . 177
4.6 Problem Set D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5 Functions on Metric Spaces and Continuity 183

5.1 Limits of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.2 Continuous Functions on Metric Spaces . . . . . . . . . . . . . . . 197
5.2.1 A Characterization of Continuity . . . . . . . . . . . . . . . 200
5.2.2 Continuity and Compactness . . . . . . . . . . . . . . . . . 203
5.2.3 Continuity and Connectedness . . . . . . . . . . . . . . . . 206
5.3 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . 208
5.4 Discontinuities and Monotonic Functions . . . . . . . . . . . . . . 211
5.4.1 Limits of Functions in the Extended Real Number System . 221
5.5 Problem Set E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
CONTENTS v
6 Differentiation: Our First View 229

6.1 The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.1.1 Formulas for Differentiation . . . . . . . . . . . . . . . . . 242
6.1.2 Revisiting A Geometric Interpretation for the Derivative . . 243
6.2 The Derivative and Function Behavior . . . . . . . . . . . . . . . . 244
6.2.1 Continuity (or Discontinuity) of Derivatives . . . . . . . . . 250
6.3 The Derivative and Finding Limits . . . . . . . . . . . . . . . . . . 251
6.4 Inverse Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
6.5 Derivatives of Higher Order . . . . . . . . . . . . . . . . . . . . . 258
6.6 Differentiation of Vector-Valued Functions . . . . . . . . . . . . . . 262
6.7 Problem Set F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
7 Riemann-Stieltjes Integration 275

7.1 Riemann Sums and Integrability . . . . . . . . . . . . . . . . . . . 277
7.1.1 Properties of Riemann-Stieltjes Integrals . . . . . . . . . . 295
7.2 Riemann Integrals and Differentiation . . . . . . . . . . . . . . . . 307
7.2.1 Some Methods of Integration . . . . . . . . . . . . . . . . . 311
7.2.2 The Natural Logarithm Function . . . . . . . . . . . . . . . 314
7.3 Integration of Vector-Valued Functions . . . . . . . . . . . . . . . . 315
7.3.1 Recti¿able Curves . . . . . . . . . . . . . . . . . . . . . . 318
7.4 Problem Set G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
8 Sequences and Series of Functions 325

8.1 Pointwise and Uniform Convergence . . . . . . . . . . . . . . . . . 326
8.1.1 Sequences of Complex-Valued Functions on Metric Spaces . 334
8.2 Conditions for Uniform Convergence . . . . . . . . . . . . . . . . 335
8.3 Property Transmission and Uniform Convergence . . . . . . . . . . 339
8.4 Families of Functions . . . . . . . . . . . . . . . . . . . . . . . . . 349
8.5 The Stone-Weierstrass Theorem . . . . . . . . . . . . . . . . . . . 363
8.6 Problem Set H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
9 Some Special Functions 369

9.1 Power Series Over the Reals . . . . . . . . . . . . . . . . . . . . . 369
9.2 Some General Convergence Properties . . . . . . . . . . . . . . . . 379
9.3 Designer Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
9.3.1 Another Visit With the Logarithm Function . . . . . . . . . 393
9.3.2 A Series Development of Two Trigonometric Functions . . 394
9.4 Series from Taylor’s Theorem . . . . . . . . . . . . . . . . . . . . 397
vi CONTENTS
9.4.1 Some Series To Know & Love . . . . . . . . . . . . . . . . 399

9.4.2 Series From Other Series . . . . . . . . . . . . . . . . . . . 404
9.5 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
9.6 Problem Set I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
Preface
These notes have been prepared to assist students who are learning Advanced Cal-
culus/Real Analysis for the ¿rst time in courses or self-study programs that are
using the text Principles of Mathematical Analysis (3rd Edition) by Walter Rudin.
References to page numbers or general location of results that mention “our text”
are always referring to Rudin’s book. The notes are designed to
encourage or engender an interactive approach to learning the material,
provide more examples at the introductory level,
offer some alternative views of some of the concepts, and
draw a clearer connection to the mathematics that is prerequisite to under-

standing the development of the mathematical analysis.
On our campus, the only prerequisites on the Advanced Calculus course include
an introduction to abstract mathematics (MAT108) course and elementary calculus.
Consequently, the terseness of Rudin can require quite an intellectual leap. One
needs to pause and reÀect on what is being presented stopping to do things like
draw pictures, construct examples or counterexamples for the concepts the are being
discussed, and learn the de¿nitions is an essential part of learning the material.
These Companion Notes explicitly guide the reader/participant to engage in those
activities. With more math experience or maturity such behaviors should become
a natural part of learning mathematics. A math text is not a novel simply reading
it from end to end is unlikely to give you more than a sense for the material. On
the other hand, the level of interaction that is needed to successfully internalize an
understanding of the material varies widely from person to person. For optimal
bene¿t from the combined use of the text (Rudin) and the Companion Notes ¿rst
read the section of interest as offered in Rudin, then work through the relevant
vii
viii PREFACE
section or sections in the Companion Notes, and follow that by a more interactive
review of the section from Rudin with which you started.
One thing that should be quite noticeable is the higher level of detail that is
offered for many of the proofs. This was done largely in response to our campus
prerequisite for the course. Because most students would have had only a brief
exposure to some of the foundational material, a very deliberate attempt has been
made to demonstrate how the prerequisite material that is usually learned in an
introduction to abstract mathematics course is directly applied to the development
of mathematical analysis. You always have the elegant, “no nonsense” approach
available in the text. Learn to pick and choose the level of detail that you need
according to your own personal mathematical needs.
0.0.1 About the Organization of the Material

The chapters and sections of the Companion Notes are not identically matched with
their counterparts in the text. For example, the material related to Rudin’s Chapter
1 can be found in Chapter 1, Chapter 2 and the beginning of Chapter 3 of the
Companion Notes. There are also instances of topic coverage that haven’t made it
into the Companion Notes the exclusions are due to course timing constraints and
not statements concerning importance of the topics.
0.0.2 About the Errors

Of course, there are errors! In spite of my efforts to correct typos and adjust errors
as they have been reported to me by my students, I am sure that there are more errors
to be found and I hope for the assistance of students who ¿nd things that look like
errors as they work through the notes. If you encounter errors or things that look like
errors, please sent me a brief email indicating the nature of the problem. My email
address is [email protected]. Thank you in advance for any comments,
corrections, and/or insights that you decide to share.
Chapter 1
The Field of Reals and Beyond
Our goal with this section is to develop (review) the basic structure that character-
izes the set of real numbers. Much of the material in the ¿rst section is a review
of properties that were studied in MAT108 however, there are a few slight differ-
ences in the de¿nitions for some of the terms. Rather than prove that we can get
from the presentation given by the author of our MAT127A textbook to the previous
set of properties, with one exception, we will base our discussion and derivations
on the new set. As a general rule the de¿nitions offered in this set of Compan-
ion Notes will be stated in symbolic form this is done to reinforce the language
of mathematics and to give the statements in a form that clari¿es how one might
prove satisfaction or lack of satisfaction of the properties. YOUR GLOSSARIES
ALWAYS SHOULD CONTAIN THE (IN SYMBOLIC FORM) DEFINITION AS
GIVEN IN OUR NOTES because that is the form that will be required for suc-
cessful completion of literacy quizzes and exams where such statements may be
requested.
1.1 Fields
Recall the following DEFINITIONS:
The Cartesian product of two sets A and B, denoted by A B, is

a b : a + A F b + B.
1
2 CHAPTER 1. THE FIELD OF REALS AND BEYOND
A function h from A into B is a subset of A B such that

(i) 1a [a + A " 2b b + B F a b + h] i.e., dom h A, and
(ii) 1a 1b 1c [a b + h F a c + h " b c] i.e., h is single-valued.
A binary operation on a set A is a function from A A into A.
A ¿eld is an algebraic structure, denoted by I e f , that includes a

set of objects, I, and two binary operations, addition and multiplication
, that satisfy the Axioms of Addition, Axioms of Multiplication, and the
Distributive Law as described in the following list.
(A) Axioms of Addition (I e is a commutative group under the binary

operation of addition with the additive identity denoted by e)
(A1) : I I I
(A2) 1x 1y x y + I " x y y x (commutative with respect
to addition)
b d ec
(A3) 1x 1y 1z x y z + I " x y z x y z (asso-
ciative with respect to addition)
(A4) 2e [e + I F 1x x + I "x e e x x] (additive identity
property)
(A5) 1x x + I " 2 x [x + I F x x x x e]
(additive inverse property)
(M) Axioms of Multiplication (I f is a commutative group under the
binary operation of multiplication with the multiplicative identity
denoted by f )
(M1) : I I I
(M2) 1x 1y x y + I " x y y x (commutative with respect
to multiplication)
b d ec
(M3) 1x 1y 1z x y z + I " x y z x y z (associative
with respect to multiplication)
d e
(M4) 2 f f + I F f / e F 1x x + I " x f f x x (mul-
tiplicative identity property)
(M5) 1x x + I db
eb " cc b ce
2 x 1 x 1 + I F x x 1 x 1 x f
(multiplicative inverse property)
1.1. FIELDS 3
(D) The Distributive Law
1x 1y 1z x y z + I " [x y z x y x z]
Remark 1.1.1 Properties (A1) and (M1) tell us that I is closed under addition and
closed under multiplication, respectively.
Remark 1.1.2 The additive identity and multiplicative identity properties tell us
that a ¿eld has at least two elements namely, two distinct identities. To see that two
elements is enough, note that, for I
0 1, the algebraic structure I c e 0 1
where c : I I I and e : I I I are de¿ned by the following tables:
c 0 1 e 0 1
0 0 1 0 0 0
1 1 0 1 0 1
is a ¿eld.
Remark 1.1.3 The ¿elds with which you are probably the most comfortable are
the rationals T 0 1 and the reals U 0 1. A ¿eld that we will discuss
shortly is the complex numbers F 0 0 1 0 Since each of these distinctly
different sets satisfy the same list of ¿eld properties, we will expand our list of
properties in search of ones that will give us distinguishing features.
When discussing ¿elds, we should distinguish that which can be claimed as

a basic ¿eld property ((A),(M), and (D)) from properties that can (and must) be
proved from the basic ¿eld properties. For example, given that I is a ¿eld,
we can claim that 1x 1y x y + I " x y + I as an alternative description
of property (A1) while we can not claim that additive inverses are unique. The
latter observation is important because it explains why we can’t claim e * from
I e f being a ¿eld and x * x e x we don’t have anything that
allows us to “subtract from both sides of an equation”. The relatively small number
of properties that are offered in the de¿nition of a ¿eld motivates our search for
additional properties of ¿elds that can be proved using only the basic ¿eld properties
and elementary logic. In general, we don’t claim as axioms that which can be
proved from the “minimal” set of axioms that comprise the de¿nition of a ¿eld. We
will list some properties that require proof and offer some proofs to illustrate an
approach to doing such proofs. A slightly different listing of properties with proofs
of the properties is offered in Rudin.
Proposition 1.1.4 Properties for the Additive Identity of a ¿eld I e f
1. 1x x + I F x x x " x e
2. 1x x + I " x e e x e
d e
3. 1x 1y x y + I F x y e " x e G y e
Proof. (of #1) Suppose that x + I satis¿es x x x. Since x + I, by the

additive inverse property, x + I is such that x x x x e. Now by
substitution and the associativity of addition,
e x x x x x x x x x e x.
(of #3) Suppose that x y + I are such that x y e and x / e. Then, by the
multiplicative inverse property, x 1 + F satis¿es x x 1 x 1 x f . Then
substitution, the associativity of multiplication, and #2 yields that
r s
y f y x 1 x y x 1 x y x 1 e e.
Hence, for x y + I, x y e F x e implies that y e. The claim now follows

immediately upon noting that, for any propositions P, Q, and M, [P " Q G M]
is logically equivalent to [P F Q " M].
Excursion 1.1.5 Use #1 to prove #2.
***The key here was to work from x e x e e.***
Proposition 1.1.6 Uniqueness of Identities and Inverses for a ¿eld I e f
1. The additive identity of a ¿eld is unique.

1.1. FIELDS 5
2. The multiplicative identity of a ¿eld is unique.

3. The additive inverse of any element in I is unique.
4. The multiplicative inverse of any element in I
e is unique.
Proof. (of #1) Suppose that * + I is such that
1x x + I " x * * x x
In particular, since e + I, we have that e e *. Since e is given as an additive

identity and * + I, e * *. From the transitivity of equals, we conclude that
e *. Therefore, the additive identity of a ¿eld is unique.
(of #3) Suppose that a + I is such that there exists * + I and x + I satisfying
a* *a e and a x x a e.
From the additive identity and associative properties

* *e * a x
* a x
ex
x.
Since a was arbitrary, we conclude that the additive inverse of each element in a
¿eld is unique
Excursion 1.1.7 Prove #4.
***Completing this excursion required only appropriate modi¿cation of the proof

that was offered for #3. You needed to remember to take you arbitrary element in F
to not be the additive identity and then simply change the operation to multiplica-
tion. Hopefully, you remembered to start with one of the inverses of your arbitrary
element and work to get it equal to the other one.***
Proposition 1.1.8 Sums and Products Involving Inverses for a ¿eld I e f
1. 1a 1b a b + I " a b a b
2. 1a a + F " a a
3. 1a 1b a b + I " a b a b
4. 1a 1b a b + I " a b a b
5. 1a 1b a b + I " a b a b

r r d e1 b c ss
6. 1a a + I
e " a 1 / e F a 1 a F a 1 a1
b b cb cc
7. 1a 1b a b + I
e " a b1 a 1 b1
Proof. (of #2) Suppose that a + I. By the additive inverse property a + I

and a + I is the additive inverse of a i.e., a a e. Since a
is the additive inverse of a, a a a a e which also justi¿es that a
is an additive inverse of a. From the uniqueness of additive inverses (Proposition
1.1.6), we conclude that a a.
Excursion 1.1.9 Fill in what is missing in order to complete the following proof of
#6.
Proof. Suppose that a + I
e. From the multiplicative inverse property,
a 1 + I satis¿es . If a 1 e, then, by Proposition
1
1.1.4(#2), a 1 a e. Since multiplication is single-valued, this would imply that
which contradicts part of the prop-
2 3
erty. Thus, a 1
/ e.
b c1
Since a 1 + I
e, by the property, a 1 +
4
b c1 1 b c1
I and satis¿es a 1 a a 1 a 1 f but this equation also justi¿es
b 1 c1 1
that a is a multiplicative inverse for a . From Proposition ,
5
b c1
we conclude that a 1 a.
1.1. FIELDS 7
b b cc
From (#5), a 1 a a 1 a f from which we conclude
b c
that a 1 is a for a. Since a1 is a mul-
6
c inverse for a and multiplicative inverses are unique, we have that
tiplicative
b
a 1 a1 as claimed.
***Acceptable responses are: (1) a a 1 f , (2) e f , (3) multiplicative identity,

(4) multiplicative inverse, (5) 1.1.6(#4), and (6) multiplicative inverse.***
Proposition 1.1.10 Solutions to Linear Equations. Given a ¿eld I 0 1,
1. 1a 1b a b + I " 2!x x + I F a x b
2. 1a 1b a b + IFa / 0 " 2!x x + I F a x b
Proof. (of #1) Suppose that a b + I and a / 0. Since a + I

0 there
exists a 1 + I such that a a 1 a 1 a 1. Because a 1 + I and b + I,
x a 1 b + I from (M1). Substitution and the associativity of multiplication
de f
yield that
r s r s
a x a a 1 b a a 1 b 1 b b.
Hence, x satis¿es a x b. Now, suppose that * + I also satis¿es a * b. Then

r s
* 1 * a 1 a * a 1 a * a 1 b x.
Since a and b were arbitrary,
1a 1b a b + I " 2!x x + I F a x b .
Remark 1.1.11 As a consequence of Proposition 1.1.10, we now can claim that, if

x * z + I and x * x z, then * z and if * z + I, x + I
0 and
* x z x, then * z. The justi¿cation is the uniqueness of solutions to linear
equations in a ¿eld. In terms of your previous experience with elementary algebraic
manipulations used to solve equations, the proposition justi¿es what is commonly
referred to as “adding a real number to both sides of an equation” and “dividing
both sides of an equation by a nonzero real number.”
Proposition 1.1.12 Addition and Multiplication Over Fields Containing Three or

More Elements. Suppose that I is a ¿eld and a b c d + I. Then
1. a b c a c b c b a
2. a b c a c b c b a
3. a c b d a b c d
4. a c b d a b c d
Proposition 1.1.13 Multiplicative Inverses in a ¿eld I 0 1

v w
a b c d + I F
b b / 0 cF db / 0 c
1. 1a 1b 1c 1d
" b d / 0 F a b1 c d 1 a c b d1
d b c b c e
2. 1a 1b 1c a b c + I F c / 0 " a c1 b c1 a b c1
b c b c b c
3. 1a 1b [a b + I F b / 0 " a b1 a b1 a b1 ]
4. 1a 1b 1c 1d [a + I F b c d + I

0 " c d 1 / 0
b c b c1 b c b c
F a b1 c d 1 a d b c1 a b1 d c1 ]
5. 1a 1b 1c 1d [a c + I F b d + I
0 " b d / 0F
b c b c
a b1 c d 1 a d b c b d1 ]
Proof. (of #3) Suppose a b + I and b / 0. Since b / 0, the zero of the

¿eld is its own additive inverse, and additive inverses are unique, we have that
b / 0. Since a + I and b + I
0 implies 1
1
b 1
c a + I and b + I
0, by
that
Proposition 1.1.8(#4), a b a b . From Proposition 1.1.8(#6), we
b c
know that b1 b1 . From the distributive law and Proposition 1.1.8(#2),
r s r r s s
a b1 a b1 a b1 b1 a b1 b1 a 0 0
from which we conclude that a b1 is an additive inverse

b for ac b1 . Since
additive inverses are unique, it follows that a b1 a b1 . Combining
our results yields that
r s
a b1 a b1 a b1
as claimed.
1.1. FIELDS 9
Excursion 1.1.14 Fill in what is missing in order to complete the following proof
of #4.
Proof. (of #4) Suppose that a + I and b c d + I
0. Since d + I
0, by
Proposition , d 1 / 0. From the contrapositive of Proposition 1.1.4(#3),
1
c / 0 and d 1 / 0 implies that . In the following, the justi¿cations
2
for the step taken is provided on the line segment to the right of the change that has
been made.
b c b c1 b c r b c1 s
a b1 c d 1 a b1 c1 d 1
b c b c 3
a b1 c1 d
b 1 b 1 cc 4
a b c d
bb c c 5
a b1 c1 d
b b cc 6
a d b1 c1
b c 7
1
a d b c
8
1
a d b c .
9
From Proposition 1.1.8(#7) combined with the associative and commutative prop-
erties of addition we also have that
b c
a d b c1 ba d b1 c c1
ba bd b1cc c1
a d b1 c1

bb c c10
b a b1c bd c1c
a b1 d c1 .
b c b c1 b c b c
Consequently, a b1 c d 1 a d b c1 a b1 d c1 as
claimed.
***Acceptable responses are: (1) 1.1.8(#6), (2) cd 1 / 0, (3) Proposition 1.1.8(#7),
(4) Proposition 1.1.8(#6), (5) associativity of multiplication, (6) associativity of
multiplication, (7) commutativity ofb multiplication,

b 1 cc 1 (8) Proposition 1.1.8(#7), (9)
associativity of multiplication, (10) a b d c .***
The list of properties given in the propositions is, by no means, exhaustive. The
propositions illustrate the kinds of things that can be concluded (proved) from the
core set of basic ¿eld axioms.
Notation 1.1.15 We have listed the properties without making use of some nota-
tional conventions that can make things look simpler. The two that you might ¿nd
particularly helpful are that
the expression a b may be written as a b ab may be written

as a b and
a
the expression a b1 may be written as . (Note that applying this nota-
b
tional convention to the Properties of Multiplicative Inverses stated in the last
proposition can make it easier for you to remember those properties.)
Excursion 1.1.16 On the line segments provided, ¿ll in appropriate justi¿cations

for the steps given in the following outline of a proof that for a b c d in a ¿eld,
a b c d a c b d
Observation Justi¿cation
notational
a b c d a b c d
convention
a b c d a b c d
1
a b c d a b c d
2
a b c d a b c d
3
a b c d a b c d
4
a b c d a c b d
5
a c b d a c b d
6
a c b d a c b d
7
a c b d a c b d
8
1.2. ORDERED FIELDS 11
***Acceptable responses are: (1) Proposition 1.1.8(#1), (2) Proposition 1.1.8(#2),

(3) and (4) associativity of addition, (5) commutativity of addition, (6) and (7) as-
sociativity of addition, and (8) notational convention.***
1.2 Ordered Fields

Our basic ¿eld properties and their consequences tell us how the binary operations
function and interact. The set of basic ¿eld properties doesn’t give us any means of
comparison of elements more structure is needed in order to formalize ideas such
as “positive elements in a ¿eld” or “listing elements in a ¿eld in increasing order.”
To do this we will introduce the concept of an ordered ¿eld.
Recall that, for any set S, a relation on S is any subset of S S
De¿nition 1.2.1 An order, denoted by , on a set S is a relation on S that satis¿es

the following two properties:
1. The Trichotomy Law: If x + S and y + S, then one and only one of
x y or x y or y x
is true.
d e
2. The Transitive Law: 1x 1y 1z x y z + S F x y F y z " x z .
Remark 1.2.2 Satisfaction of the Trichotomy Law requires that
1x 1y x y + S " x y G x y G y x
be true and that each of
1x 1y x y + S " x y " x y F y x ,

1x 1y x y + S " x y " x y F y x , and
1x 1y x y + S " y x " x y F x y
be true. The ¿rst statement, 1x 1y x y + S " x y G x y G y x

is not equivalent to the Trichotomy Law because the disjunction is not mutually
exclusive.
Example 1.2.3 For S

a b c with a, b, and c distinct,
a b b c a c
is an order on S. The notational convention for a b + is a b. The given or-
dering has the minimum and maximum number of ordered pairs that is needed to
meet the de¿nition. This is because, given any two distinct elements of S, x and
y, we must have one and only one of x y + or y x +. After making free
choices of two ordered pairs to go into an acceptable ordering for S, the choice
of the third ordered pair for inclusion will be determined by the need to have the
Transitive Law satis¿ed.
Remark 1.2.4 The de¿nition of a particular order on a set S is, to a point, up to

the de¿ner. You can choose elements of S S almost by preference until you start
having enough elements to force the choice of additional ordered pairs in order
to meet the required properties. In practice, orders are de¿ned by some kind of
formula or equation.
Example 1.2.5 For T, the set of rationals, let t T T be de¿ned by r s +%
s r is a positive rational. Then T is an ordered set.
Remark 1.2.6 The treatment of ordered sets that you saw in MAT108 derived the
Trichotomy Law from a set of properties that de¿ned a linear order on a set. Given
an order on a set, we write x n y for x y G x y. With this notation, the
two linear ordering properties that could have been introduced and used to prove
the Trichotomy Law are the Antisymmetric law,
1x 1y x y + S F x y + n F y x + n " x y
and the Comparability Law,
1x 1y x y + S " x y + n G y x + n
Now, because we have made satisfaction of the Trichotomy Law part of the def-
inition of an order on a set, we can claim that the Antisymmetric Law and the
Comparability Law are satis¿ed for an ordered set.
De¿nition 1.2.7 An ordered ¿eld I 0 1 is an ordered set that satis¿es

the following two properties.
d e
(OF1) 1x 1y 1z x y z + I F x y " x z y z
d e
(OF2) 1x 1y 1z x y z + I F x y F 0 z " x z y z
Remark 1.2.8 In the de¿nition of ordered ¿eld offered here, we have deviated from
one of the statements that is given in our text. The second condition given in the
text is that
d e
1x 1y x y + I F x 0 F y 0 " x y 0
let’s denote this proposition by alt O F2. We will show that satisfaction of O F1
and alt O F2 is, in fact, equivalent to satisfaction of O F1 and O F2. Suppose
that O F1 and O F2 are satis¿ed and let x y + I be such that 0 x and 0 y.
From O F2 and Proposition 1.1.4(#2),
d 0 0 y x y. Since x and y were
e arbi-
trary, we conclude that 1x 1y x y + I F x 0 F y 0 " x y 0 . Hence,
O F2 " alt O F2 from which we have that O F1 F O F2 " O F1 F
alt O F2. Suppose that O F1 and alt O F2 are satis¿ed and let x y z + I be
such that x y and 0 z. From the additive inverse property x + I is such
that [x x x x 0]. From O F1 we have that
0 x x y x .
From alt O F2, the Distributive Law and Proposition 1.1.8 (#4), 0 y x
and 0 z implies that
0 y x z y z x z y z x z .
Because and are binary operations on I, x z + I and y z x z + I.

It now follows from O F1 and the associative property of addition that
0 x z y z x z x z y z x z x z y z 0.
Hence, x z y z. Since x, y, and z were arbitrary, we have shown that
d e
1x 1y 1z x y z + I F x y F 0 z " x z y z
which is O F2. Therefore, O F1 F alt O F2 " O F1 F O F2. Combining
the implications yields that
O F1 F O F2 % O F1 F alt O F2 as claimed.
To get from the requirements for a ¿eld to the requirements for an ordered ¿eld
we added a binary relation (a description of how the elements of the ¿eld are or-
dered or comparable) and four properties that describe how the order and the binary
operations “interact.” The following proposition offers a short list of other order
properties that follow from the basic set.
Proposition 1.2.9 Comparison Properties Over Ordered Fields.

For an ordered ¿eld I 0 1 we have each of the following.
1. 0 1
d e
2. 1x 1y x y + I F x 0Fy 0"xy 0
3. 1x [x + I F x 0 " x 0]

d e
4. 1x 1y x y + I F x y " y x
d e
5. 1x 1y 1z x y z + I F x y F z 0 " x z y z
d e
6. 1x x + I F x / 0 " x x x 2 0
d e
7. 1x 1y x y + I F 0 x y " 0 y 1 x 1
In the Remark 1.2.8, we proved the second claim. We will prove two others.
Proofs for all but two of the statements are given in our text.
Proof. (or #1) By the Trichotomy Law one and only one of 0 1, 0 1, or
1 0 is true in the ¿eld. From the multiplicative identity property, 0 / 1 thus,
we have one and only one of 0 1 or 1 0. Suppose that 1 0. From O F1,
we have that 0 1 1 0 1 1 i.e., 0 1. Hence, O F2
implies that 1 1 0 1 which, by Proposition 1.1.8(#3), is equivalent
to 1 0. But, from the transitivity property, 0 1 F 1 0 " 0 0 which
is a contradiction.
of Proposition 1.2.9(#4).
Proof. Suppose that x y + I are such that x y. In view of the additive
inverse property, x + I and y + I satisfy
x x x x 0 and .
1
From , 0 x x y x i.e.,

2 3
and 0 y y. Repeated use of commutativity and as-

4
sociativity allows us to conclude that y x y x. Hence y x as
claimed.
***Acceptable responses are: (1) y y y y 0, (2) OF1, (3) 0 y x,

(4) y x.***
Remark 1.2.11 From Proposition 1.2.9(#1) we see that the two additional prop-
erties needed to get from an ordered set to an ordered ¿eld led to the requirement
that 0 1 be an element of the ordering (binary relation). From 0 1 and O F1,
we also have that 1 1 1 2, 2 2 1 3 etc. Using the convention
1_ 1 ^]1 1` n, the general statement becomes 0 n n 1.
n of them
1.2.1 Special Subsets of an Ordered Field

There are three special subsets of any ordered ¿eld that are isolated for special con-
sideration. We offer their formal de¿nitions here for completeness and perspective.
De¿nition 1.2.12 Let I 0 1 n be an ordered ¿eld. A subset S of I is said

to be inductive if and only if
1. 1 + S and
2. 1x x + S " x 1 + S.
De¿nition 1.2.13 For I 0 1 n an ordered ¿eld, de¿ne

?
QI S
S+8
where 8
S l I : S is inductive. We will call QI the set of natural numbers of
I.
Note that, T
x + I : x o 1 is inductive because 1 + T and closure of I un-
der addition yields that x 1 + I whenever x + I. Because 1u u 1 " u + T
and T + 8, we immediately have that any n + QI satis¿es n o 1.
De¿nition 1.2.14 Let I 0 1 n be an ordered ¿eld. The set of integers of

I, denoted ]I , is
]I
a + I : a + QI G a + QI G a 0
It can be proved that both the natural numbers of a ¿eld and the integers of a
¿eld are closed under addition and multiplication. That is,
1m 1n n + QI F m + QI " n m + QI F n m + QI
and
1m 1n n + ]I F m + ]I " n m + ]I F n m + ]I .
This claim requires proof because the fact that addition and multiplication are bi-
nary operations on I only places n m and n m in I because QI t I and ]I t I.
Proofs of the closure of QI Q under addition and multiplication that you
might have seen in MAT108 made use of the Principle of Mathematical Induction.
This is a useful tool for proving statements involving the natural numbers.
PRINCIPLE OF MATHEMATICAL INDUCTION (PMI). If S is an inductive

set of natural numbers, then S Q
In MAT108, you should have had lots of practice using the Principle of Mathe-
matical Induction to prove statements involving the natural numbers. Recall that to
do this, you start the proof by de¿ning a set S to be the set of natural numbers for
which a given statement is true. Once we show that 1 + S and
1k k + S " k 1 + S, we observe that S is an inductive set of natural num-
bers. Then we conclude, by the Principle of Mathematical Induction, that S Q
which yields that the given statement is true for all Q
Two other principles that are logically equivalent to the Principle of Mathemat-
ical Induction and still useful for some of the results that we will be proving in this
course are the Well-Ordering Principle and the Principle of Complete Induction:
WELL-ORDERING PRINCIPLE (WOP). Any nonempty set S of natural num-

bers contains a smallest element.
PRINCIPLE OF COMPLETE INDUCTION (PCI). Suppose S is a nonempty

set of natural numbers. If
1m m + Q F
k + Q : k m t S " m + S
then S Q.
De¿nition 1.2.15 Let I 0 1 n be an ordered ¿eld. De¿ne

Q r sR
1
TI r + I : 2m 2n m n + ]I F n / 0 F r mn
The set TI is called the set of rational numbers of I.

Properties #1 and #5 from Proposition 1.1.13 can be used to show the set of
rationals of a ¿eld is also closed under both addition and multiplication.
The set of real numbers U is the ordered ¿eld with which you are most familiar.
Theorem 1.19 in our text asserts that U is an ordered ¿eld the proof is given in
an appendix to the ¿rst chapter. The notation (and numerals) for the corresponding
special subsets of U are:
Q M
1 2 3 4 5 the set of natural numbers
]
m : m + Q G m 0 G m + Q
3 2 1 0 1 2 3
T
p q 1 qp : p q + ] F q / 0.
Remark 1.2.16 The set of natural numbers may also be referred to as the set of
positive integers, while the set of nonnegative integers is M C
0. Another common
term for MC
0 is the set of whole numbers which may be denoted by Z. In MAT108,
the letter Q was used to denote the set of natural numbers, while the author of our
MAT127 text is using the letter J . To make it clearer that we are referring to special
sets of numbers, we will use the “blackboard bold” form of the capital letter. Feel
free to use either (the old) Q or (the new) M for the natural numbers in the ¿eld of
reals.
While Q and ] are not ¿elds, both T and U are ordered ¿elds that have several
distinguishing characteristics we will be discussing shortly. Since T t U and
U T / 3, it is natural to want a notation for the set of elements of U that are not
rational. Towards that end, we let Lrr U T denote the set of irrationals. It
T
de f
T r T s
was shown in MAT108 that 2 is irrational. Because 2 2 0 + Lrr
T T
and 2 2 2 + Lrr , we see that Lrr is not closed under either addition or
multiplication.
1.2.2 Bounding Properties

Because both T and U are ordered ¿elds we note that “satisfaction of the set of
ordered ¿eld axioms” is not enough to characterize the set of reals. This natu-
rally prompts us to look for other properties that will distinguish the two algebraic
systems. The distinction that we will illustrate in this section is that the set of ra-
tionals has “certain gaps.” During this (motivational) part of the discussion, you
might ¿nd it intuitively helpful to visualize the “old numberline” representation
for the reals. Given two rationals r and s such that r s, it can be shown that
m r s 21 + T is such that r m s. Then r1 r m 21 + T and
s1 m s 21 + T are such that r r1 m and m s1 s. Continuing this
process inde¿nitely and “marking the new rationals on an imagined numberline”
might entice us into thinking that we can “¿ll in most of the points on the number
line between r and s.” A rigorous study of the situation will lead us to conclude
that the thought is shockingly inaccurate. We Tcertainly know that not all the reals
can be found this way because, for example, 2 could never be written in the form
of r s 21 for r s + T. The following excursion will motivate the property
that we want to isolate in our formal discussion of bounded sets.
j k
Excursion
j 1.2.17 Let A p +
k T : p 0 F p 2 2 and
B p + T : p 0 F p 2 2 . Now we will expand a bit on the approach used

in our text to show that A has no largest element and B has not smallest element.
For p a positive rational, let
p2 2 2p 2
q p .
p2 p2
Then
b c
2 p2 2
q 22
.
p 22
(a) For p + A, justify that q p and q + A.
(b) For p + B, justify that q p and q + B.

***Hopefully you took a few moments to ¿nd some elements of A and B in order to
get a feel for the nature of the two sets. Finding a q that corresponds to a p + A and
a p + B would pretty much tell byou whyc the claims are true. For (a), you should
have noted that q p because p 2 2 p 21 0 whenever p 2 2 then
b 2 c b b c c
p 2 p 21 0 implies that q p p 2 2 p 21 p0
p. That q is rational follows from the factb that thec rationals are closed under multi-
plication and addition. Finally q 2 2 2 p 2 2 p 22 0 yields that q + A
as claimed. For (b), the same reasons extend tobthe discussion
c needed here the only
change is that, for p + B, p 2 2 implies that p 2 p 21 0 from which
2
b 2 c b b c c
it follows that p 2 p 21 0 and q p p 2 2 p 21
p 0 p.***
Now we formalize the terminology that describes the property that our example
is intended to illustrate. Let S n be an ordered set i.e., is an order on the set
S. A subset A of S is said to be bounded above in S if
2u u + S F 1a a + A " a n u
Any element u + S satisfying this property is called an upper bound of A in S.
De¿nition 1.2.18 Let S n be an ordered set. For A t S, u is a least upper

bound or supremum of A in S if and only if
1. u + S F 1a a + A " a n u and
2. 1b [b + S F 1a a + A " a n b " u n b].
Notation 1.2.19 For S n an ordered set and A t S, the least upper bound of A
is denoted by lub A or sup A.
Since a given set can be a subset of several ordered sets, it is often the case that
we are simply asked to ¿nd the least upper bound of a given set without specifying
the “parent ordered set.” When asked to do this, simply ¿nd, if it exists, the u that
satis¿es
1a a + A " a n u and 1b [1a a + A " a n b " u n b] .
The next few examples illustrate how we can use basic “pre-advanced calculus”
knowledge to ¿nd some least upper bounds of subsets of the reals.
| }
x
Example 1.2.20 Find the lub :x+U .
1 x2
From Proposition 1.2.9(#5), we know that, for x + U, 1 x2 o 0 this is
equivalent to
1 x 2 o 2x
t u
x 1 1
from which we conclude that 1x x + U " n . Thus, is an upper
| } 1x 2 2 2
x 1 1
bound for : x + U . Since , it follows that
1 x2 1 12 2
| }
x 1
lub :x +U .
1x 2 2
The way that this example was done and presented is an excellent illustration
of the difference between scratch work (Phase II) and presentation of an argument
(Phase III) in the mathematical process. From calculus (MAT21A or its equivalent)
x
we can show that f x has a relative minimum at x 1 and a relative
1 x2
maximum at x 1 we also know that yt 0uis a horizontal asymptote for the
1
graph. Armed with the information that 1 is a maximum for f , we know
2
x 1
that all we need to do is use inequalities to show that n . In the scratch
1x 2 2
work phase, we can work backwards from this inequality to try to ¿nd something
that we can claim from what we have done thus far simple algebra gets use from
x 1
n to 1 2x x 2 o 0. Once we see that desire to claim 1 x2 o 0,
1x 2 2
we are home free because that property is given in one of our propositions about
ordered ¿elds.
Excursion 1.2.21 Find the lub A for each of the following. Since your goal is
simply to ¿nd the least upper bound, you can use any pre-advanced calculus infor-
mation that is helpful.
| }
3 1n
1. A :n+M
2n1
2. A
sin x cos x : x + U
3 1n 1
***For (1), let x n n1
then x2 j 2 j1 is a sequence that is strictly
2 2
1 1
decreasing from to 0 while x2 j1 is also decreasing from to 0. Consequently
2 2
1 1
the terms in A are never greater than with the value of being achieved when
2 2
n 1 and the terms get arbitrarily close to 0 as n approaches in¿nity. Hence,
1 1
lub A . For (2), it is helpful to recall that sin x cos x sin 2x. The well
2 2
1
known behavior of the sine function immediately yields that lub A .***
2
j k
Example 1.2.22 Find lub A where A x + U : x 2 x 3 .
What we are looking for here is sup A where A f 1 * 3 for f x
x x. Because
2
t u
1 1 2
y x x % y x
2
,
4 2
t u
1 1
f is a parabola with vertex . Hence,
2 4
T T
1 13 1 13
A f 1 * 3 x + R : x
2 2
T
1 13
from which we conclude that sup A .
2
j k
Note that the set A p + T : p 0 F p2 2 is a subset of T and a subset
of U. We have that T n and U n are ordered T sets where is de¿ned by
r s % s r is positive. Now lub A 2 + T hence, there is no least
upper bound of A in S T, but A t S U has a least upper bound in S U.
This tells us that the “parent set” is important, gives us a distinction between T and
U as ordered ¿elds, and motivates us to name the important distinguishing property.
De¿nition 1.2.23 An ordered set S has the least upper bound property if and
only if
v w
E t S F E / 3 F 2; ; + S F 1a a + E " a n ;
1E
" 2u u lub E F u + S
Remark 1.2.24 As noted above, T n does not satisfy the “lub property”, while
U n does satisfy this property.
The proof of the following lemma is left an exercise.
Lemma 1.2.25 Let X n be an ordered set and A l X. If A has a least upper
bound in X, it is unique.
We have analogous or companion de¿nitions for subsets of an ordered set that

are bounded below. Let S n be an ordered set i.e., is an order on the set S. A
subset A of S is said to be bounded below in S if
2) ) + S F 1a a + A " ) n a
Any element u + S satisfying this property is called a lower bound of A in S.
De¿nition 1.2.26 Let S n be a linearly ordered set. A subset A of S is said to

have a greatest lower bound or in¿mum in S if
1. 2g g + S F 1a a + A " g n a, and

d e
2. 1c c + S F 1a a + A " c n a " c n g .
| t u }
n 1 2
Example 1.2.27 Find the glb A where A 1 :n+Q .
t u 4 n
1 2 2 1
Let x n 1n then, for n odd, xn and, for n even,
4 n n 4
1 2
xn .
4 n
1 1
Suppose that n o 4. By Proposition 1.2.9(#7), it follows that n .
n 4
2 2 1 2 1 1 1 1
Then O F2 and O F1 yield that n and n ,
n 4 2 n 4 2 4 4
2 1 2 1
respectively. From n and Proposition 1.2.9(#4), we have that o .
n 2 n 2
1 2 1 1 1
Thus, o from O F1. Now, it follows from Proposition
4 n 4 2 4
1.2.9(#1) that n 0, for any n + Q . From Proposition 1.2.9(#7) and O F1,
2 2 1 1
n 0 and 2 0 implies that 0 and o . Similarly, from Proposition
n n 4 4
2 2 1 2 1 1
1.2.9(#3) and O F1, 0 implies that 0 and 0 .
n n 4 n 4 4
Combining our observations, we have that
v w
b c 1 1
1n n + Q
1 2 3 F 2 0 n " n xn n
4 4
and
v w
1 1
1n n + Q
1 2 3 F 2 n " n xn n .
4 4
v w
7 3 5 1 1
Finally, x1 , x2 , and x3 , each of which is outside of .
4 4 12 4 4
3
Comparing the values leads to the conclusion that glb A .
4
Excursion 1.2.28 Find glb A for each of the following. Since your goal is simply
to ¿nd the greatest lower bound, you can use any pre-advanced calculus informa-
tion that is helpful.
| }
3 1n
1. A :n+M
2n1
| }
1 1
2. A n m : n m + Q
2 3
***Our earlier discussion in Excursion 1.2.21, the set given in (1) leads to the con-
1 1
clusion that glb A 0. For (2), note that each of n and m are strictly de-
2 3
creasing to 0 as n and m are increasing, respectively. This leads us to conclude that
5
glb A 0 although it was not requested, we note that sup A .***
6
We close this section with a theorem that relates least upper bounds and greatest
lower bounds.
Theorem 1.2.29 Suppose S is an ordered set with the least upper bound prop-
erty and that B is a nonempty subset of S that is bounded below. Let
L
g + S : 1a a + B " g n a .
Then : sup L exists in S, and : inf B.
Proof. Suppose that S is an ordered set with the least upper bound property
and that B is a nonempty subset of S that is bounded below. Then
L
g + S : 1a a + B " g n a .
is not empty. Note that for each b + B we have that g n b for all g + L i.e.,
each element of B is an upper bound for L. Since L t S is bounded above and S
satis¿es the least upper bound property, the least upper bound of L exists and is in
S. Let : sup L.
1.3. THE REAL FIELD 25
Now we want to show that : is the greatest lower bound for B.
De¿nition 1.2.30 An ordered set S has the greatest lower bound property if
and only if
v w
E t S F E / 3 F 2< < + S F 1a a + E " < n a
1E .
" 2* * glb E F * + S
Remark 1.2.31 Theorem 1.2.29 tells us that every ordered set that satis¿es the
least upper bound property also satis¿es the greatest lower bound property.
1.3 The Real Field

The Appendix for Chapter 1 of our text offers a construction of “the reals” from
“the rationals”. In our earlier observation of special subsets of an ordered ¿eld, we
offered formal de¿nitions of the natural numbers of a ¿eld, the integers of a ¿eld,
and the rationals of a ¿eld. Notice that the de¿nitions were not tied to the objects
(symbols) that we already accept as numbers. It is not the form of the objects in the
ordered ¿eld that is important it is the set of properties that must be satis¿ed. Once
we accept the existence of an ordered ¿eld, all ordered ¿elds are alike. While this
identi¿cation of ordered ¿elds and their corresponding special subsets can be made
more formal, we will not seek that formalization.
It is interesting that our mathematics education actually builds up to the formu-
lation of the real number ¿eld. Of course, the presentation is more hands-on and
intuitive. At this point, we accept our knowledge of sums and products involving
real numbers. I want to highlight parts of the building process simply to put the
properties in perspective and to relate the least upper bound property to something
tangible. None of this part of the discussion is rigorous. First, de¿ne the sym-
bols 0 and 1, by
0 and
3 1 and suppose that we have an ordered ¿eld
de f de f
R 0 1 n. Furthermore, picture a representation of a straight horizontal line
z on which we will place elements of this ¿eld in a way that attaches some
geometric meaning to their location. The natural numbers of this ¿eld Q R is the
“smallest” inductive subset it is closed under addition and multiplication. It can be
proved (Some of you saw the proofs in your MAT108 course.) that
1x x + Q R " x o 1
and
1* * + Q R " 2) ) + Q R F * ) * 1 .
This motivates our ¿rst set of markings on the representative line. Let’s indicate the
¿rst mark as a “place for 1.” Then the next natural number of the ¿eld is 11, while
the one after that is 1 1 1, followed by [1 1 1] 1, etc. This naturally
leads us to choose a ¿xed length to represent 1 (or “1 unit”) and place a mark
for each successive natural number 1 away from and to the right of the previous
natural number. It doesn’t take too long to see that our collections of “added 1) s”
is not a pretty or easy to read labelling system this motivates our desire for neater
representations. The symbols that we have come to accept are 1 2 3 4 5 6 7 8
and 9. In the space provided draw a picture that indicates what we have thus far.
The fact that, in an ordered ¿eld, 0 1 tells us to place 0 to the left of 1 on our
representative line then 0 1
C
3
3 1 justi¿es placing 0 “1 unit”
away from the 1. Now the de¿nition of the integers of a ¿eld ] R adjoins the additive
inverses of the natural numbers of a ¿eld our current list of natural numbers leads to
acceptance of 1 2 3 4 5 6 7 8 and 9 as labels of the markings
of the new special elements and their relationship to the natural numbers mandates
their relative locations. Use the space provided to draw a picture that indicates what
we have thus far.
Your picture should show several points with each neighboring pair having the
same distance between them and “lots of space” with no labels or markings, but
we still have the third special subset of the ordered ¿eld namely, the rationals of
the ¿eld T R . We are about to prove an important result concerning the “density
of the rationals” in an ordered ¿eld. But, for this intuitive discussion, our “grade
school knowledge” of fractions will suf¿ce. Picture (or use the last picture that you
drew to illustrate) the following process: Mark the midpoint of the line segment
1
from 0 to 1 and label it 21 or then mark the midpoint of each of the smaller
2
1 1
line segments (the one from 0 to and the one from to 1) and label the two new
2 2
1 3
points and , respectively repeat the process with the four smaller line segments
4 4
1 1 3 1 5 3 7
to get as the marked rationals between 0 and 1. It doesn’t take
8 4 8 2 8 4 8
too many iterations of this process to have it look like you have ¿lled the interval.
Of course, we know that we haven’t because any rational in the from p q 1 where
0 p q and q / 2n for any n has been omitted. What turned out to be a
surprise, at the time of discovery, is that all the rationals r such the 0 n r n 1
will not be “enough to ¿ll the interval [0 1].” At this point we have the set of
elements of the ¿eld that are not in any of the special subsets, R T R , and the
“set of vacancies” on our model line. We don’t know that there is a one-to-one
correspondence between them. That there is a correspondence follows from the
what is proved in the Appendix to Chapter 1 of our text.
Henceforth, we use U 0 1 to denote the ordered ¿eld (of reals) that
satis¿es the least upper bound property and may make free use of the fact that for
any x + U we have that x is either rational or the least upper bound of a set of
rationals. Note that the sub¿eld T 0 1 is an ordered ¿eld that does not
satisfy the least upper bound property.
1.3.1 Density Properties of the Reals

In this section we prove some useful density properties for the reals.
Lemma 1.3.1 If S l U has L as a least upper bound L, then

1 + UF 0 " 2s s + S F L s n L
Proof. Suppose S is a nonempty subset of U such that L sup S and let
+ U be such that 0. By Proposition 1.2.9(#3) and O F1, 0 and
L L. From the de¿nition of least upper bound, each upper bound of S is
greater than or equal to L. Hence, L is not an upper bound for S from which
we conclude that 1s s + S " s n L is satis¿ed i.e.,
2s s + S F L s .
Combining this with L sup S yields that
2s s + S F L s n L .
Since was arbitrary, 1 + UF 0 " 2s s + S F L s n L as
claimed.
Theorem 1.3.2 (The Archimedean Principle for Real Numbers) If : and ; are
positive real numbers, then there is some positive integer n such that n: ;.
Proof. The proof will be by contradiction. Suppose that there exist positive
real numbers : and ; such that n: n ; for every natural number n. Since : 0,
: 2: 3: n: is an increasing sequence of real numbers
that is bounded above by ;. Since U n satis¿es the least upper bound property
n: : n + Q has a least upper bound in U, say L. Choose > 12 : which is positive

because : 0. Since L sup
n: : n + Q, from Lemma 1.3.1, there exists
s +
n: : n + Q such that L s n L. If s N :, then for all natural numbers
m N , we also have that L m: n L. Hence, for m N , 0 n L m: .
In particular,
1
0 n L N 1: :
2
and
1
0 n L N 2: :
2
Thus, L 12 : N 1: and N 2: L L 12 :. But adding : to
both sides of the ¿rst inequality, yields L 12 : N 2: which contradicts
N 2: L 12 :. Hence, contrary to our original assumption, there exists a
natural number n such that n: ;.
Corollary 1.3.3 (Density of the Rational Numbers) If : and ; are real numbers
with : ;, then there is a rational number r such that : r ;.
Proof. Since 1 and ; : are positive real numbers, by the Archimedean Prin-
ciple, there exists a positive integer m such that 1 m; :, or equivalently
m: 1 n m;
Let n be the largest integer such that n n m:. It follows that
n 1 n m: 1 n m;
Since n is the largest integer such that n n m:, we know that m: n 1.

Consequently, m: n 1 m;, which is equivalent to having
n1
: ;
m
Therefore, we have constructed a rational number that is between : and ;.
Corollary 1.3.4 (Density of the Irrational Numbers) If : and ; are real num-
bers with : ;, then there is an irrational number < such that : < ;.
Proof. Suppose that : and ; are real numbers withT: ;. By Corollary 1.3.3,
there is a rational r that is between T: and T; . Since 2 is irrational, we conclude
T 2 2
that < r 2 is an irrational that is between : and ;.
1.3.2 Existence of nth Roots

The primary result in this connection that is offered by the author of our text is the
following
Theorem 1.3.5 For U

x + U : x 0, we have that
b b cc
1x 1n x + U F n + M " 2!y y + U F y n x .
Before we start the proof, we note the following fact that will be used in the
presentation.
d e
Fact 1.3.6 1y 1z 1n y z + U F n + M F 0 y z " y n z n
To see this, for y z + U satisfying are 0 y z, let
j k
S n + M : y n zn .
Our set-up automatically places 1 + S. Suppose that k + S i.e., k + M and y k z k .
Since 0 y, by O F2, y k1 y y k y z k . From 0 z and repeated use
of Proposition 1.2.9(#2), we can justify that 0 z k . Then O F2 with 0 z k and
y z yields that y z k z z k z k1 . As a consequence of the transitive law,
y k1 y z k F y z k z k1 " y k1 z k1
that is, k 1 + S. Since k was arbitrary, we conclude that
1k k + S " k 1 + S.
From 1 + S F 1k k + S " k 1 + S, S is an inductive subset of the
natural numbers. By the Principle of Mathematical Induction (PMI), S M. Since
y and z were arbitrary, this completes the justi¿cation of the claim.
d e
Fact 1.3.7 1* 1n * + U F n + M
1 F 0 * 1 " * n *
Since n o 2, n 1 o 1 and, by Fact 1.3.6, *n1 1n1 1. From O F2,
0 * F * n1 1 implies that * n * n1 * 1 * * i.e., * n * as
claimed.
Fact 1.3.8 1a 1b 1n [a b + U F n + M
1 F 0 a b
" bn a n b a nbn1 ]
From Fact 1.3.6, n o 2 F 0 a b " a n1 bn1 , while O F2 yields
that a b j b b j b j1 for j 1 2 n 2. It can be shown (by repeated
application of Exercise 6(a)) that
bn1 bn2 a ba n2 a n1 bn1 bn1 bn1 nbn1
this, with O F2, implies that
r s
b a b a b
n n n1
b a ba
n2 n2
a n1
b a nbn1
as claimed.
Proof. (of the theorem.) Let U
u + U : u 0. When n 1, there is
nothing to prove so we assume that n o 2. For ¿xed x + U and n + M
1, set
j k
E t + U : t n x .
x
Excursion 1.3.9 Use * to justify that E / 3.
1x
Now let u 1 x and suppose that t u 0. Fact 1.3.6 yields that

1
tn u n . From Proposition 1.2.9(#7), u 1 " 0 1. It follows from
u
1 1
Fact 1.3.7 and Proposition 1.2.9(#7) that 0 n n and u n o u. By transitivity,
u u
t n u n F u n o u implies that t n u. Finally, since u x transitivity leads to the
conclusion that t n x. Hence, t + E. Since t was arbitrary, 1t t u " t + E
which is equivalent to 1t t + E " t n u. Therefore, E t U is bounded above.
From the least upper bound property, lub E exists. Let
y lub E .
Since E t U , we have that y o 0.

By the Trichotomy Law, one and only one of y n x, y n x, or y n x.
In what follows we will that neither of the possibilities y n x, or y n x can hold.
x yn
Case 1: If y n x, then x y n 0. Since y 1 0 and n o 1,
n y 1n1
0 and we can choose h such that 0 h 1 and
x yn
h .
n y 1n1
Taking a y and b y h in Fact 1.3.8 yields that
y hn y n hn y hn1 x y n .
Excursion 1.3.10 Use this to obtain contradict that y sup E.
Case 2: If 0 x y n , then 0 y n x ny n . Hence,
yn x
k
ny n1
is such that 0 k y. For t o y k, Fact 1.3.6 yields that t n o y kn .

From Fact 1.3.8, with b y and a y k, we have that
y n t n n y n y kn kny n1 y n x.
Excursion 1.3.11 Use this to obtain another contradiction.
From case 1 and case 2, we conclude that y n x. this concludes the proof that
there exists a solution to the given equation.
The uniqueness of the solution follows from Fact 1.3.6. To see this, note
that, if y x and * is such that 0 * / y, then * y implies that * n y n
n
x, while y * implies that x y n * n . In either case, * n / x.

***For Excursion 1.3.9, you want to justify that the given * is in E. Because 0
x
x 1x, 0 * 1. In view of fact 1.3.7, * n * for n o 2 or * n n *
1x
1 x
for n o 1. But x 0F1x 1 implies that 1F x 1 x.
1x 1x
From transitivity, * n * F * x " * n x i.e., * + E.
To obtain the desired contradiction for completion of Excursion 1.3.10, hope-
fully you notices that the given inequality implied that y hn x which would
place y h in E since y h y, this would contradict that y sup E from

which we conclude that y x is not true.
n
The work needed to complete Excursion 1.3.11 was a little more involved. In
this case, the given inequality led to t n x or t n x which justi¿es that
t +
E hence, t y k implies that t + E which is logically equivalent to t + E
implies that t y k. This would make y k an upper bound for E which is a
contradiction. Obtaining the contradiction yields that x y n is also not true.***
Remark 1.3.12 For x a positive real number and T n a natural number, the number
y that satis¿es the equation y x is written as x and is read as “the nth root of
n n
x.”
Repeated application of the associativity and commutativity of multiplication

can be used to justify that, for positive real numbers : and ; and n a natural number,
: n ; n :;n .
From this identity and the theorem we have the following identity involving nth
roots of positive real numbers.
Corollary 1.3.13 If a and b are positive real numbers and n is a positive integer,
then
ab1n a 1n b1n .
Proof. For : a 1n and ; b1n , we have that ab : n ; n :;n . Hence

:; is the unique solution to y n ab from which we conclude that ab1n :;
as needed.
1.3.3 The Extended Real Number System

The extended real number system is U C
* * where U 0 1 is the
ordered ¿eld that satis¿es the least upper bound property as discussed above and
the symbols * and * are de¿ned to satisfy * x * for all x + U.
With this convention, any nonempty subset S of the extended real number system is
bounded above by * and below by * if S has no ¿nite upper bound, we write
lub S * and when S has no ¿nite lower limit, we write glb S *.
The * and * are useful symbols they are not numbers. In spite of their
appearance, * is not an additive inverse for *. This means that there is no
* *
meaning attached to any of the expressions * * or or in fact, these
* *
expressions should never appear in things that you write. Because the symbols *
and * do not have additive (or multiplicative) inverses, U C
* * is not a
¿eld. On the other hand, we do have some conventions concerning “interaction” of
the special symbols with elements of the ¿eld U namely,
x x
If x + U, then x * *, x * * and 0.
* *
If x 0, then x * * and x * *.
If x 0, then x * * and x * *.
Notice that nothing is said about the product of zero with either of the special sym-
bols.
1.4 The Complex Field

For F U U, de¿ne addition and multiplication by
x1 y1 x2 y2 x1 x2 y1 y2
and
x1 y1 x2 y2 x1 x2 y1 y2 x1 y2 y1 x2 ,
respectively. That addition and multiplication are binary operations on F is a con-

sequence of the closure of U under addition and multiplication. It follows immedi-
ately that
x y 0 0 x y and x y 1 0 x y .
Hence, 0 0 and 1 0 satisfy the additive identity property and the multiplica-
tive identity ¿eld property, respectively. Since the binary operations are de¿ned as
combinations of sums and products involving reals, direct substitution and appro-
priate manipulation leads to the conclusion that addition and multiplication over
F are commutative and associative under addition and multiplication. (The actual
manipulations are shown in our text on pages 12-13.)
To see that the additive inverse property is satis¿ed, note that x y + F implies
that x + U F y + U. The additive inverse property in the ¿eld U yields that x + U
1.4. THE COMPLEX FIELD 35
and y + U. It follows that x y + F and x y x y 0 0 and
needed.
Suppose x y + F is such that x y / 0 0. tThen x / 0 G y /u 0 from
x y
which we conclude that x 2 y 2 / 0 and a b 2 is well
de f x y x y2
2 2
de¿ned. Now,
t u
x y
x y a b x y
t x 2 y2 x 2 y2 u
x y y x
x 2 y 2 x 2 y 2
t x y x y2 x y 2u x y2
2
x x y y x y y x

t 2 x2 y u x y
2 2 2 2
x y x y yx

x 2 y2 x 2 y2
1 0.
Hence, the multiplicative inverse property is satis¿ed for F .

Checking that the distributive law is satis¿ed is a matter of manipulating the
appropriate combinations over the reals. This is shown in our text on page 13.
Combining our observations justi¿es that F 0 0 1 0 is a ¿eld. It is
known as the complex ¿eld or the ¿eld of complex numbers.
Remark 1.4.1 Identifying each element of F in the form x 0 with x + U leads to
the corresponding identi¿cation of the sums and products, x a x 0a 0
x a 0 and x a x 0 a 0 x a 0. Hence, the real ¿eld is a sub¿eld
of the complex ¿eld.
The following de¿nition will get us to an alternative formulation for the complex
numbers that can make some of their properties easier to remember.
De¿nition 1.4.2 The complex number 0 1 is de¿ned to be i .
With this de¿nition, it can be shown directly that
i 2 1 0 1 and
if a and b are real numbers, then a b a bi.

With these observations we can write

Q R
F a bi : a b + U F i 2 1
with addition and multiplication being carried out using the distributive law, com-
mutativity, and associativity.
We have two useful forms for complex numbers the rectangular and trigono-
metric forms for the complex numbers are freely interchangeable and offer different
geometric advantages.
From Rectangular Coordinates

Complex numbers can be represented geometrically as points in the plane. We
plot them on a rectangular coordinate system that is called an Argand Graph. In
z x i y, x is the real part of z, denoted by Re z, and y is the imaginary part
of z, denoted by Im z. When we think of the complex number x i y as a vector

O P joining the origin O 0 0 to the point P x y, we grasp the natural
geometric interpretation of addition () in F.
De¿nition 1.4.3 The modulus of a complex number z is the magnitude

S of the vector
representation and is denoted by z. If z x i y, then z x y 2 .
2
De¿nition 1.4.4 The argument of a nonzero complex number z, denoted by arg z,

is a measurement of the angle that the vector representation makes with the positive
real axis.
De¿nition 1.4.5 For z x i y, the conjugate of z, denoted by z , is x i y.
Most of the properties that are listed in the following theorems can be shown
fairly directly from the rectangular form.
Theorem 1.4.6 For z and * complex numbers,

1. z o 0 with equality only if z 0,
2. z z,
3. z* z *,
4. Re z n z and Im z n z,
5. z *2 z2 2 Re z* *2 .
The proofs are left as exercises.
Theorem 1.4.7 (The Triangular Inequalities) For complex numbers z 1 and z 2 ,
z 1 z 2 n z 1 z 2 and z 1 z 2 o z 1 z 2 .
Proof. To see the ¿rst one, note that
z 1 z 2 2 z 1 2 2 Re z 1 z 2 z 2 2
.
n z 1 2 2 z 1 z 2 z 2 2 z 1 z 2 2
The proof of the second triangular inequality is left as an exercise.
Theorem 1.4.8 If z and * are complex numbers, then
1. z * z *
2. z* z *
z z z z
3. Re z , Im z ,
2 2i
4. zz is a nonnegative real number.
From Polar Coordinates rys

S
For nonzero z x i y + F, let r x 2 y 2 and A arctan arg z.
x
Then the trigonometric form is
z r cos A i sin A .
In engineering, it is customary to use cis A for cos A i sin A in which case we write
z r cis A.
NOTE: While r A uniquely determines a complex number, the converse is not
true.
Excursion 1.4.9 Use the polar form for complex numbers to develop a geometric
interpretation of the product of two complex numbers.
The following identity can be useful when working with complex numbers in
polar form.
Proposition 1.4.10 (DeMoivre’s Law) For A real and n + ],
[cis A]n cis nA .
Example 1.4.11 Find all the complex numbers that when cubed give the value one.
We are looking for all ? + F such that ? 3 1. DeMoivre’s Law offers us a nice
tool
n 3 for solving
n this equation. Let ? r cis A . Then? 3 1 % r 3 cis 3A 1. Since
nr cis 3A n r , we immediately conclude that we must have r 1. Hence, we need
3
only solve the equation cis 3A 1. Due to the periodicity of the sine and cosine,
we know that the last equation is equivalent to ¿nding all A such | that cis 3A}
2kH
cis 2kH for k + ] which yields that 3A 2kH for k + ]. But :k+]
| } 3
2H 2H
0 . Thus, we have three distinct complex numbers whose cubes are
3 3 t u t u
2H 2H
one namely, cis , cis 0 1, and cis . In rectangular form, the three
3 3 T T
1 3 1 3
complex numbers whose cubes are one are: i , 0, and .
2 2 2 2
Theorem 1.4.12 (Schwarz’s Inequality) If a1 an and b1 bn are complex
numbers, then
n n2
n; n n ;n n n ;n n n
n n na j n 2 nb j n .
2
n a b n n
n j1 j j n j1 j1
Proof. First the statement is certainly true if bk 0 for all k, 1 n k n n. Thus

we assume that not all the bk are zero. Now, for any D + F, note that
;
n n n
na j Db j n2 o 0.
j1
Excursion 1.4.13 Make use of this inequality and the choice of

; n ; n n n 1
D ajbj nb j n2
j1 j1
to complete the proof.
Remark 1.4.14 A special case of Schwarz’s Lemma contains information relating

the modulus of two vectors with the absolute value of their dot product. For ex-
ample, if

) 1 a1 a2 and n

) 2 b1n b2 are vectors in U
nnU,n then
n Schwarz’s

Lemma merely reasserts that ) 1 ) 2 a1 b1 a2 b2 n ) 1 ) 2 n.
n n n n n
1.4.1 Thinking Complex

Complex variables provide a very convenient way of describing shapes and curves.
It is important to gain a facility at representing sets in terms of expressions involving
complex numbers because we will use them for mappings and for applications to
various phenomena happening within “shapes.” Towards this end, let’s do some
work on describing sets of complex numbers given by equations involving complex
variables.
One way to obtain a description is to translate the expressions to equations in-
volving two real variables by substituting z x i y.
Example 1.4.15 Find all complex numbers z that satisfy
2 z 2 Im z 1
Let z x i y. Then
S
2 z 2 Im z 1 % 2 x 2 y 2 2y 1 t u
b b 2 c c 1
% 4 x y 2 4y 2 4y 1 F y o
2
1
% 4x 2 4y 1 F y o
t u 2
1 1
% x y
2 Fyo .
4 2
1 1 1
The last equation implies that y n . Since y n F y o is never satis¿ed, we
4 4 2
conclude that the set of solutions for the given equation is empty.
Excursion 1.4.16 Find all z + F such that z z 1 2i
3
***Your work should have given the 2i as the only solution.***
2
Another way, which can be quite a time saver, is to reason by TRANSLAT-

ING TO THE GEOMETRIC DESCRIPTION. In order to do this, there are some
geometric descriptions that are useful for us to recall:

z : z z 0 r is the locus of all points z equidistant from the ¿xed
point, z 0 , with the distance being r 0. (a circle)

z : z z 1 z z 2 is the locus of all points z equidistant from two
¿xed points, z 1 and z 2 . (the perpendicular bisec-
tor of the line segment joining z 1 and z 2 .)

z : zz 1 zz 2 is the locus of all points for which the sum of the
I for a constant I distances from 2 ¿xed points, z 1 and z 2 , is a con-
z 1 z 2 stant greater than z 1 z 2 . (an ellipse)
Excursion 1.4.17 For each of the following, without substituting x i y for z, sketch
the set of points z that satisfy the given equations. Provide labels, names, and/or
important points for each object.
n n
n z 2i n
1. nn n1
z 3 2i n
2. z 4i z 7i 12
3. 4z 3 i n 3
***The equations described a straight | line, an ellipse, and a disk, respectively.

} In
3 9
set notation, you should have obtained x i y + F : y x ,
4 8
t u
!
! 3 2 !
!
!
2 y !
x 2
x iy + F : t u 1 , and
!
! 23 62 !
!
!
!

4
t u t u t u2
3 2 1 2 3
x iy + F : x y n .***
4 4 4
Remark 1.4.18 In general, if k is a positive real number and a b + F, then
| n n }
nz a n
n
z+F:n n k k / 1
z bn
describes a circle.
Excursion 1.4.19 Use the space below to justify this remark.

1.5. PROBLEM SET A 43
n n
nz an
n
***Simplifying n n k leads to
z bn
r s b _c b _c r s
1 k 2 z2 2 Re a z 2k 2 Re b z a2 k 2 b2
from which the remark follows.***
1.5 Problem Set A

1. For I
p q r, let the binary operations of addition, c, and multiplication,
e, be de¿ned by the following tables.
c r q p e r q p
r r q p r r r r
q q p r q r q p
p p r q p r p q
(a) Is there an additive identity for the algebraic structure I c e? BrieÀy
justify your position.
(b) Is the multiplicative inverse property satis¿ed? If yes, specify a multi-
plicative inverse for each element of I that has one.
c from our ¿eld properties, ¿nd

(c) Assuming bthe notation
1
r c q e p c p .
(d) Is
p p p r q q p q r r a ¿eld ordering on I? BrieÀy
justify your claim.
2. For a ¿eld I e f , prove each of the following parts of Proposition

1.1.6.
(a) The multiplicative identity of a ¿eld is unique.

(b) The multiplicative inverse of any element in I
e is unique.
3. For a ¿eld I e f , prove each of the following parts of Proposition

1.1.8.
(a) 1a 1b a b + I " a b a b

(b) 1a 1b a b + I " a b a b
(c) 1a 1b a b + I " a b a b
(d) 1a 1b a b + I " a b a b
b b cb cc
(e) 1a 1b a b + I
e " a b1 a 1 b1
4. For a ¿eld I 0 1, prove Proposition 1.1.10(#1):

1a 1b a b + I " 2!x x + I F a x b
5. For a ¿eld I 0 1, show that, for a b c + I,
a b c a b c and a b c a b c.
Give reasons for each step of your demonstration.
6. For an ordered ¿eld I 0 1 , prove that
(a) 1a 1b 1c 1d [a b c d + I F a b F c d "

a c b d]
(b) 1a 1b 1c 1d [a b c d + I F 0 a b F 0 c d "
ac bd]
7. For an ordered ¿eld I 0 1 , prove each of the following

d
(a) 1a 1b 1c a b c + I F c / 0 "
b c b c e
a c1 b c1 a b c1
(b) 1a
b 1b
1
c 1c
b 1d
1
c c + I F b d + I
0
[a
1
" b d / 0F
ab cd a d b c b d ]
8. Find the least upper bound and the greatest lower bound for each of the fol-
lowing.
| }
n 1n
(a) :n+Q
n
| t u }
1
(b) 1n H :n+Q
n
| }
1 1
(c) : m n + M
m n
| }
1
(d) :x +U
1 x2
| }
1 1
(e) :n+M
3n 5n1
| }
1
(f) x : x + U
0
x
| }
1 1
(g) x : x 2
x 2
9. Let X n be an ordered set and A l X. Prove that, if A has a least upper
bound in X, it is unique.
10. Suppose that S l U is such that inf S M . Prove that
1 + UF 0 " 2g g + S F M n g M
2 1
11. For f x 2 , ¿nd
x x
(a) sup f 1 * 3

(b) inf f 1 3 *
12. Suppose that P t Q t U and P / 3. If P and Q are bounded above, show

that sup P n sup Q.
j k
13. Let A x + U : x 2 x 31 2 . Find the sup A and the inf A.
14. Use the Principle of Mathematical Induction to prove that, for a o 0 and n a
natural number, 1 an o 1 na.
15. Find all the values of

(a) 2 34 1. (d) 1 i4 .
(b) 1 2i [3 2 i 2 3 6i]. (e) 1 in 1 in .
(c) 1 i3 .
16. Show that the following expressions are both equal to one.
T 3 T 3
1 i 3 1 i 3
(a) (b)
2 2
17. For any integers k and n, show that i n i n4k . How many distinct values
can be assumed by i n ?
18. Use the Principle of Mathematical Induction to prove DeMoivre’s Law.
19. If z 1 3 4i and z 2 2 3i, obtain graphically and analytically

(a) 2z 1 4z 2 . (d) z 1 z 2 .
(b) 3z 1 2z 2 . (e) z 1 z 2 .
(c) z 1 z 2 4. (f) 2z 1 3z 2 1.
20. Prove that there is no ordering on the complex ¿eld that will make it an or-
dered ¿eld.
21. Carefully justify the following parts of Theorem 1.4.6. For z and * complex
numbers,
(a) z o 0 with equality only if z 0,

(b) z z,
(c) z* z *,
(d) Re z n z and Im z n z,
(e) z *2 z2 2 Re z* *2 .
22. Prove the “other” triangular inequality: For complex numbers z 1 and z 2 ,
z 1 z 2 o z 1 z 2 .
23. Carefully justify the following parts of Theorem 1.4.8. If z and * are complex
numbers, then
(a) z * z *
(b) z* z *
z z z z
(c) Re z , Im z ,
2 2i
(d) zz is a nonnegative real number.
24. Find the set of all z + F that satisfy:

(a) 1 z n 3. (d) z 1 z 1 2. (g) z 2 z 2 5.
n n
nz 3n (e) Im z 2 0. (h) z 1 Rez.
(b) nn n 1.
z 2n n
nz 2n
n
(f) nn n 2.
(c) Re z 2 0. z 1n
25. When does az bz c 0 represent a line?
26. Prove that the vector z 1 is parallel to the vector z 2 if and only if Im z 1 z 2 0.
Chapter 2
From Finite to Uncountable Sets
A considerable amount of the material offered in this chapter is a review of termi-

nology and results that were covered in MAT108. Our brief visit allows us to go
beyond some of what we saw and to build a deeper understanding of some of the
material for which a revisit would be bene¿cial.
2.1 Some Review of Functions

We have just seen how the concept of function gives precise meaning for binary
operations that form part of the needed structure for a ¿eld. The other “big” use of
function that was seen in MAT108 was with de¿ning “set size” or cardinality. For
precise meaning of what constitutes set size, we need functions with two additional
properties.
De¿nition 2.1.1 Let A and B be nonempty sets and f : A B. Then

11
1. f is one-to-one, written f : A B, if and only if
1x 1y 1z x z + f F y z + f " x y
2. f is onto, written f : A B, if and only if
1y y + B " 2x x + A F x y + f
11
3. f is a one-to-one correspondence, written f : A B, if and only if f is
one-to-one and onto.
49
50 CHAPTER 2. FROM FINITE TO UNCOUNTABLE SETS
Remark 2.1.2 In terms of our other de¿nitions, f : A B is onto if and only if
rng f
y + B : 2x x + A F x y + f B
de f
which is equivalent to f [A] B.
In the next example, the ¿rst part is shown for completeness and to remind the
reader about how that part of the argument that something is a function can be
proved. As a matter of general practice, as long as we are looking at basic functions
that result in simple algebraic combinations of variables, you can assume that was
is given in that form in a function on either its implied domain or on a domain that
is speci¿ed.
|t u }
x
Example 2.1.3 For f x + U U : 1 x 1 , prove that
1 x
11
f : 1 1 U.
(a) By de¿nition, f l U U i.e., f is a relation from 1 1 to U.

Now suppose that x + 1 1. Then x 1 from which it follows that
1 x / 0. Hence, 1 x1 + U
0 and y x 1 x1 + U
de f
because multiplication is a binary operation on U. Since x was arbitrary, we
have shown that
1x x + 1 1 " 2y y + U F x y + f i.e.,
dom f 1 1.

Suppose that x y + f F x ) + f . Then u x 1 x1 )
because multiplication is single-valued on U U. Since x u, and ) were
arbitrary,
1x 1u 1) x u + f F x ) + f " u )
i.e., f is single-valued.
Because f is a single-valued relation from 1 1 to U whose domain
is 1 1, we conclude that f : 1 1 U.
2.1. SOME REVIEW OF FUNCTIONS 51
x1 x2
(b) Suppose that f x1 f x2 i.e., x1 x2 + 1 1 and .
1 x1 1 x2
Since f x1 f x2 we must have that f x1 0 F f x2 0 or
f x1 o 0 F f x2 o 0 which implies that 1 x1 0 F 1 x2 0
or 1 x1 o 0 F 1 x 2 o 0. Now x1 x2 + 1 0 yields that f x1
x1 x2 x1
f x2 , while x1 x2 + [0 1 leads to f x1
1 x1 1 x2 1 x1
x2
f x2 . In either case, a simple calculation gives that x1 x 2 .
1 x2
Since x1 and x2 were arbitrary, 1x1 1x2 f x1 f x2 " x1 x2 .
Therefore, f is one-to-one.
(c) Finally, ¿ll in what is missing to ¿nish showing that f is onto. Let * + U.
*
Then either * 0 or * o 0. For * 0, let x . Then 1 * 0
1*
and, because 1 0, we have that 1 * * or 1 * *. Hence,
*
1 and we conclude the x + . It follows that
1* 1
x and
2
x
f x .
1 x 3
*
For * o 0, let x . Because 1 0 and * o 0 implies that
1*
*
* 0 which is equivalent to having 1 x 0.
4
1*
Hence, x and
5
f x .
6
Since * + U was arbitrary, we conclude that f maps 1 1 onto U.

*
***Acceptable responses are: (1) 1 0, (2) ,
t ut u1 1*
* * *
(3) 1 *, (4) 1 *, (5) ,
1* 1* 1*
t ut u1
* *
(6) 1 *.***
1* 1*
Given a relation from a set A to a set B, we saw two relations that could be used
to describe or characterize properties of functions.
De¿nition 2.1.4 Given sets A, B, and C, let R + S A B and S + P B C
where S X denotes the power set of X.
1. the inverse of R, denoted by R 1 , is
y x : x y + R
2. the composition of R and S, denoted by S i R, is

x z + A C : 2y x y + R F y z + S
j k
Example 2.1.5 For R x y + Q ] : x 2 y 2 n 4 and
S
x y + U U : y 2x} 1, R 1
0 1 1 1 1 1 0 2, S 1
|
x 1
x y + U U : y , and S i R
1 1 1 3 1 1 2 1.
2
Note that the inverse of a relation from a set A to a set B is always a relation
from B to A this is because a relation is an arbitrary subset of a Cartesian product
that neither restricts nor requires any extent to which elements of A or B must be
used. On the other hand, while the inverse of a function must be a relation, it need
not be a function even if the inverse is a function, it need not be a function with
domain B. The following theorem, from MAT108, gave us necessary and suf¿cient
conditions under which the inverse of a function is a function.
Theorem 2.1.6 Let f : A B. Then f 1 is a function if and only if f is one–to–
one. If f 1 is a function, then f 1 is a function from B into A if and only if f is a
function from A onto B.
We also saw many results that related inverses, compositions and the identity
function. These should have included all or a large subset of the following.
Theorem 2.1.7 Let f : A B and g : B C. Then g i f is a function from A
into C.
Theorem 2.1.8 Suppose that A, B, C, and D are sets, R + S A B, S +
S B C, and T + S C D. Then
T i S i R T i S i R.
and
S i R1 R 1 i S 1 .
2.1. SOME REVIEW OF FUNCTIONS 53
Theorem 2.1.9 Suppose that A and B are sets and that R + S A B. Then
1. R i R 1 + S B B and, whenever R is single-valued, R i R 1 l I B
2. R 1 i R + S A A and, whenever R is one-to-one, R 1 i R l I A
b c1
3. R 1 R
4. I B i R R and R i I A R.
Theorem 2.1.10 For f : A B and g : B C,
1. If f and g are one–to–one, then g i f is one–to–one.
2. If f is onto B and g is onto C, then g i f is onto C.
3. If g i f is one–to–one, then f is one–to–one.
4. If g i f is onto C then g is onto C.
Theorem 2.1.11 Suppose that A, B, C, and D are sets in the universe X .
1. If h is a function having dom h A, g is a function such that dom g B,
and A D B 3, then h C g is a function such that dom h C g A C B.
2. If h : A C, g : B D and A D B 3, then h C g : A C B C C D.
11 11
3. If h : A C, g : B D, A D B 3, and C D D 3, then
11
h C g : A C B C C D
Remark 2.1.12 Theorem 2.1.11 can be used to give a slightly different proof of the
result that was shown in Example 2.1.3. Notice that the relation f that was given
in Example 2.1.3 can be realized as f 1 C f 2 where
|t u }
x
f1 x + U U : 1 x 0
1x
and
|t u }
x
f2 x +UU:0n x 1
1x
11 11
for this set-up, we would show that f 1 : 1 0 * 0 and f 2 : [0 1
11
[0 * and claim f 1 C f 2 : 1 1 U from Theorem 2.1.11, parts #2 and
#3.
2.2 A Review of Cardinal Equivalence

De¿nition 2.2.1 Two sets A and t B are saiduto be cardinally equivalent, denoted
11
by A q B, if and only if 2 f f : A B . If A q B (read “A is equivalent to
B”, then A and B are said to have the same cardinality.
| }
1
Example 2.2.2 Let A
0 and B : n + Q . Make use of the relation
n

x f x : x + [0 1]
where

! 1
!
! , if x + A
!
2
f x x
! , if x + B
!
! 1 2x
!

x , if x + [0 1] A C B
to prove that the closed interval [0 1] is cardinally equivalent to the open interval
0 1.
Proof. Let F
x f x : x + [0 1] where f is de¿ned above. Then F
x g1 : x + A CB C
x g2 : x + [0 1] A C B where
! 1
!
2 , if x + A
g1 x x and g2 f [01]ACB .
!
!
1 2x , if x + B
1
Suppose that x + A C B. Then either x 0 or there exists n + Q such that x .
n
t u 1
1 1 n 1
It follows that g1 x g1 0 or g1 x g1 +
2 n 1 n2
12
n
0 1. Since x was arbitrary, we have that
1x x + A C B " 2y y + 0 1 F g1 x y .

2.2. A REVIEW OF CARDINAL EQUIVALENCE 55
Thus, dom g1 A C B. Furthermore, since 1 2 x / 0 for x + B

x
x 1 2 x1
1 2x
is de¿ned and single-valued because and are the binary operations on the ¿eld
U. Hence, g1 : A C B 0 1.
Since
| }
1
g1 [A C B] : n + Q C,
n2 de f
we have that g1 : A C B C. Now suppose that x1 x2 + A C B are such

1
that g1 x1 g1 x 2 . Then either g1 x1 g1 x2 or g1 x 1 g1 x2 +
| } 2
1
: n + Q . In the ¿rst case, we have that x1 x2 0. In the second case,
n2
we have that g1 x1 g1 x2 "
x1 x2
% x 1 2x1 x2 x2 2x 2 x1 % x1 x2 .
1 2x1 1 2x2
Since x 1 and x2 were arbitrary,
1x1 x2 x1 x2 + A C B F g1 x1 g1 x 2 " x1 x2 i.e.,
g1 is one-to-one. Therefore,
11
g1 : A C B C.
Note that [0 1] A C B 0 1 C. Thus, g2 , as the identity function on

0 1 C, is one-to-one and onto. That is,
11
g2 : 0 1 C 0 1 C .
11 11
From Theorem 2.1.11 #2 and #3, g1 : A C B C, g2 : 0 1 C
0 1 C, [0 1] A C B D A C B 3 and 0 1 C D C 3 implies
that
11
g1 C g2 : A C B C 0 1 C C C 0 1 C . (*)
Substituting 0 1 C [0 1] A C B in addition to noting that

A C B C [0 1] A C B [0 1]
and
C C 0 1 C 0 1
we conclude from (*) that
11
F g1 C g2 : [0 1] 0 1 .
Therefore, [0 1] 0 1.
For the purpose of describing and showing that sets are “¿nite”, we make use of
the following collection of “master sets.” For each k + M, let
Mk
j + M : 1 n j n k
For k + M, the set Mk is de¿ned to have cardinality k. The following de¿nition offers
a classi¿cation that distinguishes set sizes of interest.
De¿nition 2.2.3 Let S be a set in the universe X. Then
1. S is ¿nite % S 3 G 2k k + M F S q Mk .
2. S is in¿nite % S is not ¿nite.
3. S is countably in¿nite or denumerable % S q M.
4. S is at most countable % S is ¿nite G S is denumerable)).
5. S is uncountable % S is neither ¿nite nor countably in¿nite.
Recall that if S 3, then it is said to have cardinal number 0, written S 0.
If S q Mk , then S is said to have cardinal number k i.e., S k.
Remark 2.2.4 Notice that the term countable has been omitted from the list given
in De¿nition 2.2.3 this was done to stress that the de¿nition of countable given by
the author of our textbook is different from the de¿nition that was used in all the
MAT108 sections. The term “at most countable” corresponds to what was de¿ned
as countable in MAT108. In these Companion Notes, we will avoid confusion by
not using the term countable when reading your text, keep in mind that Rudin uses
the term countable for denumerable or countably in¿nite.
We know an in¿nite set is one that is not ¿nite. Now it would be nice to have
some meaningful in¿nite sets. The ¿rst one we think of is Q or M. While this claim
may seem obvious, it needs proving. This leads to the following
Proposition 2.2.5 The set M is in¿nite.
Space for comments.
Proof. Since
3 1 + M, M is not empty. To prove that 2k k + M F Mk q M
de f
rr 11
s s
is suf¿ces to show that 1k 1 f k + MF f : Mk M " f [Mk ] / M . Sup-
11
pose that k + M and f is such that f : Mk M. Let n f 1 f 2 f k1.
For each j, 1 n j n k, we have that f j 0. Hence, n is a natural number that
is greater than each f j. Thus, n / f j for any j + Mk . But then n +
rng f
from which we conclude that f isrrnot onto M i.e., f [Mk ]s/ M. Since k and
s f were
11
arbitrary, we have that 1k 1 f k + MF f : Mk M " f [Mk ] / M which is
equivalent to the claim that 1k k + M " Mk M. Because
M / 3 F 2k k + M F M q Mk
it follows that M is not ¿nite as claimed.
Remark 2.2.6 From the Pigeonhole Principle (various forms of which were visited
in MAT108), we know that, for any set X,
X ¿nite " 1Y Y t X F Y / X " Y X .
The contrapositive tautology yields that
1Y Y t X F Y / X " Y X " X is ¿nite
which is equivalent to
2Y Y t X F Y / X F Y q X " X is in¿nite. (-)

In fact, - could have been used as a alternative de¿nition of in¿nite set. To see
how - can be used to prove that a set is in¿nite, note that
Me
n + M : 2n
is such that Me + M and Me q M where that latter follows because f x 2x :

11
M Me consequently, M is in¿nite.
Recall that the cardinal number assigned to M is 80 which is read as “aleph

naught.” Also shown in MAT108 was that the set SM cannot be (cardinally) equiv-
alent to M this was a special case of
Theorem 2.2.7 (Cantor’s Theorem) For any set S, S SS.
Remark 2.2.8 It can be shown, and in some sections of MAT108 it was shown,
that SM q U. Since M U, the cardinality of U represents a different “level
of in¿nite.” The symbol given for the cardinality of U is c, an abbreviation for
continuum.
Excursion 2.2.9 As a memory refresher concerning proofs of cardinal equivalence,

complete each of the following.
1. Prove 2 4 q 5 20

! n
!
! , n +MF2n
2
2. Use the function f n
!
! n1
!
, n + MF2 0 n
2
to prove that ] is denumerable.
25
***For (1), one of the functions that would have worked is f x x 30 jus-
2
11
tifying that f : 2 4 5 20 involves only simple algebraic manipulations.
Showing that the function given in (2) in one-to-one and onto involves applying
elementary algebra to the several cases that need to be considered for members of
the domain and range.***
We close this section with a proposition that illustrates the general approach that
can be used for drawing conclusions concerning the cardinality of the union of two
sets having known cardinalities
Proposition 2.2.10 The union a denumerable set and a ¿nite set is denumerable
i.e.,
1A 1B A denumerable F B ¿nite " A C B is denumerable .
Proof. Let A and B be sets such that A is denumerable and B is ¿nite. First
we will prove that A C B is denumerable when A D B 3. Since B is ¿nite, we
have that either B 3 or there exists a natural number k and a function f such that
11
f : B
j + Q : j n k.
If B 3, then A C B A is denumerable. If B / 3, then let f be such
11
that f : B Qk where Qk
j + Qk : j n k. Since A is denumerable, there
de f
11
exists a function g such that g : A Q. Now let h
n n k : n + Q.
Because addition is a binary operation on Q and Q is closed under addition, for
each n + Q, n k is a uniquely determined natural number. Hence, we have that
h : Q Q. Since n + Q implies that n o 1, from OF1, nk o 1k consequently,
j + Q : j o 1 k Q Qk is a codomain for h. Thus, h : Q Q Qk .

We will now show that h is one-to-one and onto Q Qk .

(i) Suppose that h n 1 h n 2 i.e., n 1 k n 2 k. Since Q is the set
of natural numbers for the ¿eld of real numbers, there exists an additive inverse
k + U such that k k k k 0. From associativity and substitution,
we have that
n 1 n 1 k k n 1 k k
n 2 k k
n 2 k k
n2
Since n 1 and n 2 were arbitrary, 1n 1 1n 2 h n 1 h n 2 " n 1 n 2 i.e., h is
one-to-one.
(ii) Let * + Q Qk . Then * + Q and * o 1 k. By OF1, associativity of
addition, and the additive inverse property,
* k o 1 k k 1 k k 1.
Hence, x * k + Q dom h. Furthermore,
de f
h x x k * k k * k k *.

Since * was arbitrary, we have shown that
1* * + Q Qk " 2x x + QF x * + h
that is, h is onto.
11
From (i) and (ii), we conclude that h : Q Q Qk . From Theorem 2.1.10, parts
11 11
(1) and (2), g : A Q and h : Q Q Qk implies that
11
h i g : A Q Qk .
Now we consider the new function F f C h i g from B C A into Q
which can also be written as

f x for x + B
F x .

h i g x for x + A
11 11
Since A D B 3, Q D Q Qk 3, f : B Qk and h i g : A Q Qk , by
11
Theorems 2.1.11, part (1) and (2), F : B C A Q C Q Qk Q. Therefore,
B C A or A C B is cardinally equivalent to Q i.e., A C B is denumerable.
If A D B / 3, then we consider the sets A B and B. In this case,

A B D B 3 and A B C B A C B. Now the set B is ¿nite and the set
A B is denumerable. The latter follows from what we showed above because our
proof for the function h was for k arbitrary, which yields that
1k k + Q " Q Qk 80 .
From the argument above, we again conclude that A C B A B C B is denu-
merable.
Since A and B were arbitrary,
1A 1B A denumerable F B ¿nite " A C B is denumerable .
2.2.1 Denumerable Sets and Sequences

An important observation that we will use to prove some results concerning at most
countable sets and families of such sets is the fact that a denumerable set can be
“arranged in an (in¿nite) sequence.” First we will clarify what is meant by arranging
a set as a sequence.
De¿nition 2.2.11 Let A be a nonempty set. A sequence of elements of A is a func-
tion f : M A. Any f : Mk A for a k + M is a ¿nite sequence of elements of
A.
For f : M A, letting an f n leads to the following common notations
for the sequence:
an *n1 ,
an n+M ,
an , or a1 a2 a3 an . It is important to
notice the distinction between
an * n1 and
an : n + M the former is a sequence
where the listed terms need not be distinct, while the latter is a set. For example, if
f : M
1 2 3 is the constant function f n 1, then

an *
n1 1 1 1
while
an : n + M
1.
Now, if A is a denumerable or countably in¿nite set then there exists a function
11
g such that g : M A. In this case, letting g n xn leads to a sequence
xn n+M
of elements of A that exhausts A i.e., every element of A appears someplace in
the sequence. This phenomenon explains our meaning to saying that the “elements
of A can be arranged in an in¿nite sequence.” The proof of the following theorem
illustrates an application of this phenomenon.
Theorem 2.2.12 Every in¿nite subset of a countably in¿nite set is countably in¿-
nite.
Proof. Let A be a denumerable set and E be an in¿nite subset of A. Because A

is denumerable, it can be arranged in an in¿nite sequence, say
an *
n1 . Let
S1
m + M : am + E .
Because S1 is a nonempty set of natural numbers, by the Well-Ordering Principle,

S1 has a least element. Let n 1 denote the least element of S1 and set
S2
m + M : am + E
n 1 .
Since E is in¿nite, S2 is a nonempty set of natural numbers. By the Well-Ordering

Principle, S2 has a least element, say n 2 . In general, for S1 S2 Sk1 and
n 1 n 2 n k1 , we choose
n k min Sk *here Sk
m + M : am + E
n 1 n 2 n k1
Use the space provided

j k*to convince yourself that this choice “arranges E into an
in¿nite sequence an k k1 .”
2.3 Review of Indexed Families of Sets

Recall that if I is an indexed family of subsets of a set S and denotes the indexing
set, then
2.3. REVIEW OF INDEXED FAMILIES OF SETS 63
>
the union of the sets in I
A: : : + , denoted by A: , is
:+
b c

p + S : 2; ; + F p + A;
?
and the intersection of the sets in I , denoted by A: , is
:+
b c

p + S : 1; ; + " p + A; .
Remark 2.3.1 If I is a countably in¿nite or denumerable family of sets (subsets of

a set S), then the indexing set is M or Q in this case, the union and intersection over
>* ?
*
I are commonly written as A j and A j , respectively. If I is a nonempty ¿nite
j1 j1
family of sets, then Mk , for some k + M, can be used as an indexing set in this case,
>
k ?k
the union and intersection over I are written as A j and A j , respectively.
j1 j1
It is important to keep in mind that, in an indexed family I

A: : : + ,
different subscript assignments does not ensure that the sets represented are differ-
ent. An example that you saw in MAT108 was with equivalence classes. For the
relation k3 that was de¿ned over ] by x k3 y % 3 x y, for any : + ], let
A: [:]k3 then A4 A2 A5 , though the subscripts are different. The set
of equivalence classes from an equivalence relation do, however, form a pairwise
disjoint family.
An indexed family I
A: : : + is pairwise disjoint if and only if
b c
1: 1; : ; + F A: D A; / 3 " A: A;
?
it is disjoint if and only if A: 3. Note that being disjoint is a weaker condi-
:+
tion that being pairwise disjoint.
Example 2.3.2 For each j + ], where ] denotes the set of integers, let
A j
x1 x2 + U U : x1 j n 1 F x2 n 1,
> ?
Find A j and Aj.
j+] j+]
Each A j consists of a “2 by 2 square” that is symmetric about the x-

axis. For each j + ], A j and A j1 overlap in the section where j > n x1 n
j 1, while A j and A j3 have nothing in common. Consequently, Aj
j k ? j+]
x1 x2 + U2 : x 2 n 1 and A j 3.
j+]
v u
1 5n 1n
Excursion 2.3.3 For n + Q, let Cn 3 and
> ? 2n n

C
F n : n + Q. Find C j and C j.
j+Q j+Q
***For uthis one, vhopefully u

v you looked atvCn for a ufew n. For example, C1
5 11 1 17 2
4 , C2 5 , and C3 4 . Upon noting that the left
2 4 2 6 3
endpoints of the intervals are decreasing to 3 while the right endpointstare oscil-
u
> 1
lating above and below 5 and closing in on 5, we conclude that C j 3 5
j +Q
2
? v u
5
and C j 4 .***
j+Q
2
j T k
Excursion
j 2.3.4 For
k j + M, let A j x + U : x o j . Justify the claim that
D A j : j + M is disjoint but not pairwise disjoint.
2.4. CARDINALITY OF UNIONS OVER FAMILIES 65
***Hopefully, your discussion? led to your noticing that Ak D Am Amax

km . On
the other hand, to justify that A j 3, you needed to note that given any ¿xed
j+Q
f g
positive real number * there exists p + M such that * +
A p taking p * 2 1 ,
where JK denotes the greatest integer function, works.***
2.4 Cardinality of Unions Over Families

We saw the following result, or a slight variation of it, in MAT108.
Lemma 2.4.1 If A and B are disjoint ¿nite sets, then A C B is ¿nite and
A C B A B .
Excursion 2.4.2 Fill in what is missing to complete the following proof of the
Lemma.
Space for Scratch Work

Proof. Suppose that A and B are ¿nite sets such that
A D B 3. If A 3 or B 3, then A C B
1
or A C B , respectively. In either case A C B
2
is , and 3 0 yields that
3
A B A C B. If A / 3 and B / 3, then there
exists k n + Q such that A
i + Q : i n k and
B
i + Q : i n n. Hence there exist functions f
1–1
and g such that f : A and
4
g: . Now let
5
H
k 1 k 2 k n. Then the function
1–1
h x k x is such that h :
i + Q : i n n H .
Since the composition of one–to–one onto functions is a
one–to–one correspondence,
F hig: .
6
From Theorem 2.1.11, A D B 3,

1–1
i + Q : i n k D H 3, f : A
i + Q : i n k, and
1–1
F : B H implies that
f CF: . Since
7
i + Q : i n k C H , we
8
conclude that A C B is and
9
A C B .
10 11
***Acceptable responses: (1) B, (2) A, (3) ¿nite, (4)

i + Q : i n k or Ak ,
1–1 1–1 1–1
(5) B
i + Q : i n n, (6) B H, (7) A C B
i + Q : i n k C H ,
(8)
i + Q : i n k n or Akn , (9) ¿nite, (10) k n, and (11) A B.***
Lemma 2.4.1 and the Principle of Mathematical Induction can be used to prove
Theorem 2.4.3 The union of a ¿nite family of ¿nite sets is ¿nite.
Proof. The proof is left as an exercise.
Now we want to extend the result of the theorem to a comparable result concern-
ing denumerable sets. The proof should be reminiscent of the proof that T 80 .
Theorem 2.4.4 The union of a denumerable family of denumerable sets is denu-

merable.
Proof. For each n + M, jlet Ekn be a denumerable set. Each E n can be arranged
*
as an in¿nite sequence, say xn j j1 . Then
> j k
Ek xn j : n + M F j + M .
k+M
> >
Because E 1 is denumerable and E 1 t E j , we know that E j is an in¿nite
j+M j +M
set. We can use the sequential arrangement to establish an in¿nite array let the
2.4. CARDINALITY OF UNIONS OVER FAMILIES 67
sequence corresponding to E n form the nt h row.
x11 x12 x13 x14 x15 x16

x21 x22 x23 x24 x25 x26
x31 x32 x33 x34 x35 x36
x41 x42 x43 x44 x45 x46
x51 x52 x53 x54 x55 x56

.The terms in the in¿nite array can be rearranged in an expanding triangular array,
such as
x11
x21 x 12
x31 x 22 x13
x41 x 32 x23 x14
x51 x 42 x33 x24 x15 .
x61 x 52 x43 x34 x25 x16
x71 x 62 x53 x44 x35 x26 x17

This leads us to the following in¿nite sequence:
x11 x21 x21 x31 x22 x13
Because we have not> >in¿nite sequence

speci¿ed that each En is distinct, the
may list elements from E k more than once in this case, E k would corre-
k+M k+M >
spond to an in¿nite subsequence of the given arrangement. Consequently, E k is
k+M
denumerable, as needed.
Corollary 2.4.5 If A is at most countable, and, for each : + A, B: is at most

countable, then
>
T B:
:+ A
is at most countable.
The last theorem in this section determines the cardinality of sets of n tuples
that are formed from a given countably in¿nite set.
Theorem 2.4.6 For A a denumerable set and n + M, let Tn _A A
^] A`
n o f them
An i.e.,
j b ck
Tn a1 a2 an : 1 j j + M F 1 n j n n " a j + A .
Then Tn is denumerable.
Proof. Let S
n + M : Tn q M. Since T1 A and A is denumerable, 1 + S.
Suppose that k + S i.e., k + M and Tk is denumerable. Now Tk1 Tk A where
it is understood that x1 x2 xk a x1 x 2 xk a. For each b + Tk ,
b a : a + A q A. Hence,
1b b + Tk "
b a : a + A q M .
Because Tk is denumerable and
>
Tk1
b a : a + A
b+Tk
it follows from Theorem 2.4.4 that Tk1 is denumerable i.e., k 1 + S. Since k

was arbitrary, we conclude that 1k k + S " k 1 + S.
By the Principle of Mathematical Induction,
1 + S F 1k k + S " k 1 + S
implies that S M.
Corollary 2.4.7 The set of all rational numbers is denumerable.
Proof. This follows immediately upon noting that
| }
p
T : p + ] F q + M F gcd p q 1 q
p q + ] M : gcd p q 1
q
and ] M is an in¿nite subset of ] ] which is denumerable by the theorem.
2.5. THE UNCOUNTABLE REALS 69
2.5 The Uncountable Reals

x 11
In Example 2.1.3, it was shown that f x : 1 1 U. Hence, the
1 x
interval 1 1 is cardinally equivalent to U. The map gx 12 x 1 can be used
to show that 1 1 q 0 1. We noted earlier that M U. For completeness,
we restate the theorem and quickly review the proof.
Theorem 2.5.1 The open interval 0 1 is uncountable. Consequently, U is un-

countable.
Proof. Since
12 13 14 l 0 1 and
12 13 14 q M, we know that 0 1
is not ¿nite.
Suppose that
11
f : M 0 1
Then we can write
f 1 0a11 a12 a13 a14

f 2 0a21 a22 a23 a24
f 3 0a31 a32 a33 a34

f n 0an1 an2 an3 an4

where akm +
0 1 2 3 4 5 6 7 8 9. Because f is one-to-one, we know
that, if 20000 is in the listing, then 199999 isnot.
N O if a j j / N O
Finally, let m 0b1 b2 b3 b4 , where b j (The

[ ] if a j j N O
substitutions for NO.and [] are yours to choose.). Now justify that there is no q + M
such that f q m.
Hence, 2m 1k k + J " f k / m i.e., f is not onto.

Since f was arbitrary, we have shown that
1 f f : M 0 1 F f one-to-one " f is not onto .
Because [P F Q " M] is logically equivalent to [P " Q F M] and

[P " Q] is equivalent to [P F Q] for any propositions P, Q and M, we con-
clude that
d e
1 f f : M 0 1 " f one-to-one F f is onto
% 1 f f : M 0 1 F f one-to-one F f is onto
t u
11
i.e., 2 f f : M 0 1 . Hence, the open interval 0 1 is an in¿nite set that
is not denumerable.
Corollary 2.5.2 The set of sequences whose terms are the digits 0 and 1 is an
uncountable set.
2.6 Problem Set B

1. For each of the following relations, ¿nd R 1 .
(a) R
1 3 1 5 5 7 10 12
j k
(b) R x y + U U : y x 2
(c) R
a b + A B : ab where A M and B
j + ] : j n 6
2. Prove that each of the following is one-to-one on its domain.

2x 5
(a) f x
3x 2
2.6. PROBLEM SET B 71
(b) f x x 3
3. Prove that f x x 2 6x 5 maps U onto [4 *.
4. Prove each of the following parts of theorems that were stated in this chapter.
(a) Suppose that A, B, C, and D are sets, R + S A B, S + S B C,

and T + S C D. Then T i S i R T i S i R
(b) Suppose that A, B, and C are sets, R + S A B and S + S B C.
Then
S i R1 R 1 i S 1 .
(c) Suppose that A and B are sets and that R + S A B. Then
R i R 1 + S B B and, whenever R is single-valued, R i R 1 l I B
(d) Suppose that A and B are sets and that R + S A B. Then
R 1 i R + S A A and, wnenever R is one-to-one, R 1 i R l I A
(e) Suppose that A and B are sets and that R + S A B. Then
r s1
R 1 R I B i R Rand R i I A R
5. For f : A B and g : B C, prove each of the following.
(a) If f and g are one–to–one, then g i f is one–to–one.

(b) If f is onto B and g is onto C, then g i f is onto C.
(c) If g i f is one–to–one, then f is one–to–one.
(d) If g i f is onto C then g is onto C.
6. For A, B, C, and D sets in the universe X, prove each of the following.
(a) If h is a function having dom h A, g is a function such that dom g

B, and A D B 3, then h C g is a function such that dom h C g
A C B.
(b) If h : A C, g : B D and A D B 3, then h C g : A C B C C D.
11 11
(c) If h : A C, g : B D, A D B 3, and C D D 3, then
11
h C g : A C B C C D
7. Prove each of the following cardinal equivalences.
(a) [6 10] q [1 4]

(b) * 3 q 1 *
(c) * 1 q 1 2
(d) Z M C
0 q ]
8. Prove that the set of natural numbers that are primes is in¿nite.
9. Let A be a nonempty ¿nite set and B be a denumerable set. Prove that A B

is denumerable.
10. Find the union and intersection of each of the following families of sets.
(a) D
1 3 5
2 3 4 5 6
0 3 7 9
v u
1 1
(b) D
An : n + M where An 2
n n
t u
1
(c) E
Bn : n + M where Bn n
n
| }
3 2
(d) F
Cn : n + M where Cn x + U : 4 x 6
n 3n
11. Prove that the ¿nite union of ¿nite sets is ¿nite.
12. For Z M C
0, let F : Z Z Z be de¿ned by
k k 1
F i j j
2
where k i j . Prove that F is a one-to-one correspondence.
13. Prove that T T is denumerable.

Chapter 3
METRIC SPACES and SOME
BASIC TOPOLOGY
Thus far, our focus has been on studying, reviewing, and/or developing an under-
standing and ability to make use of properties of U U1 . The next goal is to
generalize our work to Un and, eventually, to study functions on Un .
3.1 Euclidean n-space

The set Un is an extension of the concept of the Cartesian product of two sets that
was studied in MAT108. For completeness, we include the following
De¿nition 3.1.1 Let S and T be sets. The Cartesian product of S and T , denoted
by S T , is

p q : p + S F q + T .
The Cartesian product of any ¿nite number of sets S1 S2 SN , denoted by S1

S2 S N , is
j b ck
p1 p2 p N : 1 j j + M F 1 n j n N " p j + S j .
The object p1 p2 p N is called an N -tuple.
Our primary interest is going to be the case where each set is the set of real
numbers.
73
74 CHAPTER 3. METRIC SPACES AND SOME BASIC TOPOLOGY
De¿nition 3.1.2 Real n-space, denoted Un , is the set all ordered n-tuples of real
numbers i.e.,
Un
x1 x2 xn : x1 x 2 xn + U .
Thus, Un U
_ U
^] U`, the Cartesian product of U with itself n times.
n of them
Remark 3.1.3 From MAT108, recall the de¿nition of an ordered pair:
a b
a
a b.
de f
This de¿nition leads to the more familiar statement that a b c d if and only
if a b and c d. It also follows from the de¿nition that, for sets A, B and
C, A B C is, in general, not equal to A B C i.e., the Cartesian
product is not associative. Hence, some conventions are introduced in order to
give meaning to the extension of the binary operation to more that two sets. If we
de¿ne ordered triples in terms of ordered pairs by setting a b c a b c
this would allow us to claim that a b c x y z if and only if a x, b y,
and c z. With this in mind, we interpret the Cartesian product of sets that are
themselves Cartesian products as “big” Cartesian products with each entry in the
tuple inheriting restrictions from the original sets. The point is to have helpful
descriptions of objects that are described in terms of n-tuple.
Addition and scalar multiplication on n-tuple is de¿ned by
x1 x2 x n y1 y2 yn x1 y1 x2 y2 xn yn
and
: x1 x2 xn :x1 :x 2 :xn , for : + U, respectively.
The geometric meaning of addition and scalar multiplication over U2 and U3 as

well as other properties of these vector spaces was the subject of extensive study in
vector calculus courses (MAT21D on this campus). For each n, n o 2, it can be
shown that Un is a real vector space.
De¿nition 3.1.4 A real vector space Y is a set of elements called vectors, with
given operations of vector addition : Y Y Y and scalar multiplication
: U Y Y that satisfy each of the following:
3.1. EUCLIDEAN N -SPACE 75
1. 1v 1w v w + Y " v w w v commutativity
2. 1u 1v 1w u v w + Y " u v w u v w associativity
3. 20 0 + Y F 1v v + Y " 0 v v 0 v zero vector
4. 1v v + Y " 2 v v + Y F v v v v 0 negatives
5. 1D 1v 1w D + U F v w + Y " D v w D v D w distribu-

tivity
6. 1D 1< 1w D < + U F w + Y " D < w D< w associativity
7. 1D 1< 1w D < + U F w + Y " D < w D w< w distribu-
tivity
8. 1v v + Y " 1 v v 1 v multiplicative identity
Given two vectors, x x1 x2 x n and y y1 y2 yn in Un , the inner
product (also known as the scalar product) is
;
n
xy x j yj
j1
and the Euclidean norm (or magnitude) of x x1 x2 xn + Un is given by

Y
X;
T X n b c2
x x x W xj .
j1
The vector space Un together with the inner product and Euclidean norm is called
Euclidean n-space. The following two theorems pull together the basic properties
that are satis¿ed by the Euclidean norm.
Theorem 3.1.5 Suppose that x y z + Un and : + U. Then
(a) x o 0
(b) x 0 % x 0
(c) :x : x and
(d) x y n x y.
Excursion 3.1.6 Use Schwarz’s Inequality to justify part (d). For x x1 x2 xn
and y y1 y2 yn in Un ,
x y2
Remark 3.1.7 It often helps to take our observations back to the setting that is
“once removed” from U1 . For the case U2 , the statement given in part (d) of the

theorem relates to the dot product of two vectors: For G x1 x2 and @ y1 y2 ,
we have that
G @ x1 y1 x2 y2
which, in vector calculus, was shown to be equivalent to G@cosA where A is the
angle between the vectors G and @.
Theorem 3.1.8 (The Triangular Inequalities) Suppose that x x 1 x 2 x N ,
y y1 y2 y N and z z 1 z 2 z N are elements of U N . Then
(a) x y n x y i.e.,
12 12 12
;N ;
N ;N
x j y j 2 n x 2j y 2j
j1 j1 j1
where 12 denotes the positive square root and equality holds if and only
if either all the x j are zero or there is a nonnegative real number D such that
y j Dx j for each j, 1 n j n N and
(b) x z n x y y z i.e.,
12 12 12
; N ;N ;N
x j z j 2 n x j y j 2 y j z j 2
j1 j1 j1
3.2. METRIC SPACES 77
where 12 denotes the positive square root and equality holds if and only
if there is a real number r , with 0 n r n 1, such that y j r x j 1 r z j
for each j, 1 n j n N .
Remark 3.1.9 Again, it is useful to view the triangular inequalities on “familiar

ground.” Let G x1 x2 and @ y1 y2 . Then the inequalities given in The-
orem 3.1.8 correspond to the statements that were given for the complex numbers
i.e., statements concerning the lengths of the vectors that form the triangles that are
associated with ¿nding G @ and G @.
Observe that, for C

x y : x 2 y 2 1 and I
x : a n x n b where
a b, the Cartesian product of the circle C with I , C I , is the right circular
cylinder,
U
x y z : x 2 y 2 1 F a n z n b
and the Cartesian product of I with C, I C, is the right circular cylinder,
V
x y z : a n x n b, y 2 z 2 1
If graphed on the same U3 -coordinate system, U and V are different objects due to
different orientation on the other hand, U and V have the same height and radius
which yield the same volume, surface area etc. Consequently, distinguishing U
from V depends on perspective and reason for study. In the next section, we lay the
foundation for properties that place U and V in the same category.
3.2 Metric Spaces

In the study of U1 and functions on U1 the length of intervals and intervals to de-
scribe set properties are useful tools. Our starting point for describing properties
for sets in Un is with a formulation of a generalization of distance. It should come
as no surprise that the generalization leads us to multiple interpretations.
De¿nition 3.2.1 Let S be a set and suppose that d : S S U1 . Then d is said

to be a metric (distance function) on S if and only if it satis¿es the following three
properties:
d e
(i) 1x 1y x y + S S " dx y o 0 F dx y 0 % x y ,
d e
(ii) 1x 1y x y + S S " dy x dx y (symmetry), and
d e
(iii) 1x 1y 1z x y z + S " dx z n dx y dy z (triangle inequal-
ity).
De¿nition 3.2.2 A metric space consists of a pair S d–a set, S, and a metric, d,
on S.
Remark 3.2.3 There are three commonly used (studied) metrics for the set U N .
For x x1 x2 x N and y y1 y2 y N , we have:
T3 b c2
N
U N d where dx y j1 x j y j , the Euclidean metric,
3N
U N D where Dx y j1 x j
y j , and
n n
U N d* where d* x y max nx j y j n.
1n jnN
Proving that d, D, and d* are metrics is left as an exercise.
Excursion 3.2.4 Graph each of the following on Cartesian coordinate systems
1. A
x + U2 : d0 x n 1
2. B
x + U2 : D0 x n 1
3. C
x + U2 : d*0 x n 1
***For (1), you should have gotten the closed circle with center at origin and radius
one for (2), your work should have led you to a “diamond” having vertices at 1 0,
0 1, 1 0, and 0 1 the closed shape for (3) is the square with vertices
1 1, 1 1, 1 1, and 1 1.***
Though we haven’t de¿ned continuous and integrable functions yet as a part of
this course, we offer the following observation to make the point that metric spaces
can be over different objects. Let F be the set of all functions that are continuous
real valued functions on the interval I x : 0 x n 1. Then there are two
natural metrics to consider on the set F namely, for f and g in F we have
(1) F d where d f g max f x gx, and
0nxn1
51
(2) F d where d f g 0 f x gxdx
Because metrics on the same set can be distinctly different, we would like to
distinguish those that are related to each other in terms of being able to “travel
between” information given by them. With this in mind, we introduce the notion of
equivalent metrics.
De¿nition 3.2.5 Given a set S and two metric spaces S d1 and S d2 , d1 and
d2 are said to be equivalent metrics if and only if there are positive constants c and
C such that cd1 x y n d2 x y n Cd1 x y for all x y in S.
Excursion 3.2.6 As the result of one of the Exercises in Problem Set

T C, you will
know that the metrics d and d* on U satisfy d*x y n dx y n 2 d* x y.
2
1. Let A
x + U2 : d0 x n 1. Draw a ¿gure showing the boundary of
A and then show the largest circumscribed square that is symmetric about
the origin and the square, symmetric about the origin, that circumscribes the
boundary of A.
2. Let C
x + U2 : d*0 x n 1. Draw a ¿gure showing the boundary of C
and then show the largest circumscribed circle that is centered at the origin
and the circle, centered at the origin, that circumscribes the boundary of C.
***For
Q (1), your outer square shouldR have corresponded to
T
x x1 x 2 + U2 : d*0 x 2 the outer circle that you showed for part of
Q T R
(2) should have corresponded to x x1 x2 + U2 : d0 x 2 .***
Excursion 3.2.7 Let E

cosA sinA : 0 n A 2H and de¿ne d ` p1 p2
A1 A2 where p1 cosA1 si nA1 and p2 cosA2 si nA2 . Show that E d ` is
a metric space.
The author of our textbook refers to an open interval a b

x + U : a x b
as a segment which allows the term interval to be reserved for a closed interval
[a b]
x + U : a n x n b half-open intervals are then in the form of [a b or
a b].
De¿nition 3.2.8 Given real numbers a1 a2 an and b1 b2 bn such that a j
b j for j 1 2 n,
j b ck
x 1 x2 xn + Un : 1 j 1 n j n n " a j n x j n b j
is called an n-cell.
Remark 3.2.9 With this terminology, a 1-cell is an interval and a 2-cell is a rect-
angle.
De¿nition 3.2.10 If x + Un and r is a positive real number, then the open ball with
center x and radius r is given by
j k
B x r y + Un : x y r
and the closed ball with center x and radius r is given by

j k
B x r y + Un : x y n r .
De¿nition 3.2.11 A subset E of Un is convex if and only if

d e
1x 1y 1D x y + E F 0 D 1 " Dx 1 D y + E
Example 3.2.12 For x + Un and r a positive real number, suppose that y and z are
in B x r . If D real is such that 0 D 1, then
Dy 1 D z x D y x 1 D z x
n D y x 1 D z x
Dr 1 D r r .
Hence, Dy 1 D z +B x r . Since y and z were arbitrary,

d e
1y 1z 1D y z + B x r F 0 D 1 " Dy 1 D z + B x r
that is, B x r is a convex subset of Un .
3.3 Point Set Topology on Metric Spaces

Once we have a distance function on a set, we can talk about the proximity of points.
The idea of a segment (interval) in U1 is replaced by the concept of a neighborhood
(closed neighborhood). We have the following
De¿nition 3.3.1 Let p0 be an element of a metric space S whose metric is denoted

by d and r be any positive real number. The neighborhood of the point p0 with
radius r is denoted by N p0 r or Nr p0 and is given by
Nr p0
p + S : d p p0 r
The closed neighborhood with center p0 and radius r is denoted by Nr p0 and

is given by
Nr p0
p + S : d p p0 n r
3.3. POINT SET TOPOLOGY ON METRIC SPACES 83
Remark 3.3.2 The sets A, B and C de¿ned in Excursion 3.2.4 are examples of
closed neighborhoods in U2 that are centered at 0 0 with unit radius.
where
What does the unit neighborhood look like for U2 , d

0 if x y

dx y is known as the discrete metric?

1 if x / y
We want to use the concept of neighborhood to describe the nature of points

that are included in or excluded from sets in relationship to other points that are in
the metric space.
De¿nition 3.3.3 Let A be a set in a metric space S d.
1. Suppose that p0 is an element of A. We say the p0 is an isolated point of A if

and only if
d e
2Nr p0 Nr p0 D A
p0
2. A point p0 is a limit point of the set A if and only if

d e
1Nr p0 2p p / p0 F p + A D Nr p0 .
(N.B. A limit point need not be in the set for which it is a limit point.)
3. The set A is said to be closed if and only if A contains all of its limit points.
4. A point p is an interior point of A if and only if

b cd e
2Nr p p Nr p p t A
5. The set A is open if and only if
b b cd ec
1 p p + A " 2Nr p p Nr p p t A
i.e., every point in A is an interior point of A.
Example 3.3.4 For each of the following subsets of U2 use the space that is pro-
vided to justify the claims that are made for the given set.
(a)
x1 x 2 + U2 : x1 x2 + M F x1 x2 5 is closed because is contains all
none of its limit points.
(b)
x1 x 2 + U2 : 4 x12 F x2 + M is neither open not closed.
(c)
x1 x 2 + U2 : x 2 x1 is open.
Our next result relates neighborhoods to the “open” and “closed” adjectives.
Theorem 3.3.5 (a) Every neighborhood is an open set.

(b) Every closed neighborhood is a closed set.
Use this space to draw some helpful pictures related to proving the results.
Proof. (a) Let Nr p0 be a neighborhood. Suppose that q + Nr p0 and set

r r1 r r1
r1 d p0 q. Let I . If x + NI q, then d x q and the
4 4
triangular inequality yields that
r r1 3r1 r
d p0 x n d p0 q d q x r1 r.
4 4
Hence, x + Nr p0 . Since x was arbitrary, we conclude that
b c
1x x + NI q " x + Nr p0
i.e., NI q t Nr p0 .Therefore, q is an interior point of Nr p0 . Because q was

arbitrary, we have that each element of Nr p0 is an interior point. Thus, Nr p0 is
open, as claimed.
(b)
Let Nr p0 be a closed neighborhood and suppose that q is a limit point of
1
Nr p0 . Then, for each rn , n + M, there exists pn / q such that pn + Nr p0
n
1
and d q pn . Because pn + Nr p0 , d p0 pn n r for each n + M. Hence,
n
by the triangular inequality
d q p0 n d q pn n .
1 2
1
Since q and p0 are ¿xed and goes to 0 as n goes to in¿nity, it follows that
n
d q p0 n r that is, q + . Finally, q and arbitrary limit point of
3
Nr p0 leads to the conclusion that Nr p0 contains .
4
Therefore, Nr p0 is closed.
1
***Acceptable responses are: (1) d pn p0 , (2) r , (3) Nr p0 , (4) all of its
n
limit points.***
The de¿nition of limit point leads us directly to the conclusion that only in¿nite
subsets of metric spaces will have limit points.
Theorem 3.3.7 Suppose that X d is a metric space and A t X. If p is a limit

point of A, then every neighborhood of p contains in¿nitely many points of A.
Space for scratch work.
Proof. For a metric space X d and A t X, suppose that p + X is such

that there exists a neighborhood of p, N p, with the property that N p D A
is a ¿nite set. If N p D A 3 or , N p D A
p, then p is not a limit
point. Otherwise, N p D A being ¿nite implies that it can be realized as a ¿nite
b q1 cq2 q3 qn for some
sequence, say b ¿xed
c n + M. For each j, 1 n j n n,
let r j d x q j . Set I min d x q j . If p +
q1 q2 q3 qn , then
1n jnn
q j / p
NI p D A
p otherwise NI p D A 3. In either case, we conclude that p
is not a limit point of A.
We have shown that if p + X has a neighborhood, N p, with the property
that N p D A is a ¿nite set, then p is not a limit point of A t X. From the
contrapositive tautology it follows immediately that if p is a limit point of A t X,
then every neighborhood of p contains in¿nitely many points of A.
Corollary 3.3.8 Any ¿nite subset of a metric space has no limit point.
From the Corollary, we note that every ¿nite subset of a metric space is closed
because it contains all none of its limit points.
3.3.1 Complements and Families of Subsets of Metric Spaces

Given a family of subsets of a metric space, it is natural to wonder about whether or
not the properties of being open or closed are passed on to the union or intersection.
We have already seen that these properties are not necessarily transmitted when we
look as families of subsets of U.
v w
3n 2 2n 2 n
Example 3.3.9 Let D
An : n + M where An 2
. Note
v w v n w n
3 2 1
that A1 [1 1], A2 2 , and A3 3 2 . More careful
2 3 3
3n 2 2
inspection reveals that 3 is strictly decreasing to 3 and n *,
n n
2n 2 n 1
2 is strictly increasing to 2 as n *, and A1 [1 1] t An
n2 n 6 7
for each n + M. It follows that An 3 2 and An A1 [1 1].
n+M n+M
The example tells us that we may need some special conditions in order to claim
preservation of being open or closed when taking unions and/or intersections over
families of sets.
The other set operation that is commonly studied is complement or relative com-
plement. We know that the complement of a segment in U1 is closed. This moti-
vates us to consider complements of subsets of metric spaces in general. Recall the
following
De¿nition 3.3.10 Suppose that A and B are subsets of a set S. Then the set differ-
ence (or relative complement) A B, read “A not B”, is given by
A B
p + S : p + A F p +
B
the complement of A, denoted by Ac , is S A.
Excursion 3.3.11 Let A

x1 x2 + U2 : x12 x 22 n 1 and
B
x1 x2 + U2 : x1 1 n 1 F x2 1 n 1
On separate copies of Cartesian coordinate systems, show the sets A B and

Ac U2 A.
The following identities, which were proved in MAT108, are helpful when we
are looking at complements of unions and intersections. Namely, we have
Theorem 3.3.12 (deMorgan’s Laws) Suppose that S is any space and I is a fam-
ily of subsets of S. Then
c
> ?
A Ac
A+5 A+5
and
c
? >
A Ac
A+5 A+5
The following theorem pulls together basic statement concerning how unions,
intersections and complements effect the properties of being open or closed. Be-
cause their proofs are straightforward applications of the de¿nitions, most are left
as exercises.
Theorem 3.3.13 Let S be a metric space.
1. The union of any family I of open subsets of S is open.
2. If
7m A1 A2 Am is a ¿nite family of open subsets of S, then the intersection
j1 A j is open.
3. For any subset A of S, A is closed if and only if Ac is open.
4. The intersection of any family I of closed subsets of S is closed.
6mA1 A2 Am is a ¿nite family of closed subsets of S, then the union

5. If
j1 A j is closed.
6. The space S is both open and closed.
7. The null set is both open and closed.
Proof. (of #2)7mSuppose that A1 A7 2 Am is a ¿nite family of open subsets

m
of S, and x + j1 A j . From x + j1 A j , it follows that x + A j for each
j , 1 n j n m. Since each A j is open, for each j , 1 n j n m, there exists
rj 0 such that Nr j x t A j . Let I min r j . Because NI x t A j for
7
1n jnm
each j, 1 n j n m, we conclude that NI x t mj1 A j . Hence, x is an interior
7m
point of j1 A j . Finally, since x was arbitrary, we can claim that each element of
7m 7m
j1 A j is an interior point. Therefore, j1 A j is open.
(or #3) Suppose that A t S is closed and x + Ac . Then x + A and, because

d of its limit points, x eis not a limit point of A. Hence, x +
A contains all AF
1Nr x A D Nr x b
x
/ 3c is true. It follows that x +
A and there exists
a I 0 such that A D NI x
x 3. Thus, A D NI x 3 and we conclude
that NI x t Ac i.e., x is an interior point of Ac . Since x was arbitrary, we have
that each element of Ac is an interior point. Therefore, Ac is open.
To prove the converse, suppose
d that A t S is such ethat Ac is open. If p
is a limit
b c then 1Nr p bA D Nr p c
p / 3 . But, for any
point of A, I 0,
A D NI p
p / 3 implies that NI p
p is not contained in A . Hence,
c
p is not an interior point of Ac and we conclude that p + Ac . Therefore, p + A.

Since p was arbitrary, we have that A contains all of its limit points which yields
that A is closed.
Remark 3.3.14 Take the time to look back at the proof of (#2) to make sure that
you where that fact that the intersection was over a ¿nite family of open subsets of
S was critical to the proof.
Given a subset of a metric space that is neither open nor closed we’d like to have
a way of describing the process of “extracting an open subset” or “building up to a
closed subset.” The following terminology will allow us to classify elements of a
metric space S in terms of their relationship to a subset A t S.
De¿nition 3.3.15 Let A be a subset of a metric space S. Then
1. A point p + S is an exterior point of A if and only if

d e
2Nr p Nr p t Ac ,
where Ac S A.
2. The interior of A, denoted by Int A or A0 , is the set of all interior points
of A.
3. The exterior of A, denoted by Ext A, is the set of all exterior points of A.
4. The derived set of A, denoted by A) , is the set of all limit points of A.
5. The closure of A, denoted by A, is the union of A and its derived set i.e.,
A A C A) .
6. The boundary of A, denoted by " A, is the difference between the closure of

A and the interior of A i.e., " A A A0 .
Remark 3.3.16 Note that, if A is a subset of a metric space S, then Ext A
Int Ac and
d e
x + " A % 1Nr x Nr x D A / 3 F Nr x D Ac / 3 .
The proof of these statements are left as exercises.

Excursion 3.3.17 For A C B where
A
x 1 x 2 + U2 : x 12 x22 1
and
B
x1 x2 + U2 : x1 1 n 1 F x2 1 n 1
1. Sketch a graph of A C B.
2. On separate representations for U2 , show each of the following
Int A C B , Ext A C B , A C B) , and A C B.
***Hopefully, your graph of A C B consisted of the union of the open disc that
is centered at the origin and has radius one with the closed square having vertices
0 0, 1 0, 1 1 and 0 1 the disc and square overlap in the ¿rst quadrant and
the set is not open and not closed. Your sketch of Int A C B should have shown
the disc and square without the boundaries (i.e., with the outline boundaries as not
solid curves), while your sketch of Ext A C B should have shown everything that
is outside the combined disc and square–also with the outlining boundary as not
solid curves. Finally, because A C B has no isolated points, A C B) and A C B
are shown as the same sets–looking like Int A C B with the outlining boundary
now shown as solid curves.***
The following theorem relates the properties of being open or closed to the
concepts described in De¿nition 3.3.15.
Theorem 3.3.18 Let A be any subset of a metric space S.
(a) The derived set of A, A) , is a closed set.
(b) The closure of A, A, is a closed set.
(c) Then A A if and only if A is closed.
(d) The boundary of A, " A, is a closed set.
(e) The interior of A, Int A, is an open set.
(f) If A t B and B is closed, then A t B
(g) If B t A and B is open, B t Int A
(h) Any point (element) of S is a closed set.
The proof of part (a) is problem #6 in WRp43, while (e) and (g) are parts of
problem #9 in WRp43.
Excursion 3.3.19 Fill in what is missing to complete the following proofs of parts
(b), (c), and (f).
Part (b): In view of Theorem 3.3.13(#3), it suf¿ces to show that .
1
b cc )
Suppose that x + S is such that x + A . Because A ACA , it follows that x + A
and . From the latter, there exists a neighborhood of x, N x, such
2

that D A 3 while the former yields that D
3 4
A 3. Hence, N x t Ac . Suppose that y + N x. Since , there
5
exists a neighborhood N ` y such that N ` y t N x. From the transitivity of
subset, from which we conclude that y is not a limit point of A i.e.,
6
b cc
y + A) . Because y was arbitrary,

1y y + N x "
7
i.e., . Combining our containments yields that N x t

8
Ac and . Hence,
8
c
b cc
N x t A D A)
c
.
9
Since x was arbitrary, we have shown that
.
10
b cc
Therefore, A is open.
Part (c): From part (b), if A A, then .

11
Conversely, if , then A) t A. Hence, A C A) that is,
12 13
A A.
Part (f): Suppose that A t B B is closed, and x + A. Then x + A or

. If x + A, then x + B if x + A) , then for every neighborhood
14
of x, N x, there exists * + A such that * / x and . But then
15
* + B and N x
x D B / 3. Since N x was arbitrary, we conclude that
. Because B is closed, . Combining the conclusions
16 17
and noting that x + A was arbitrary, we have that

1x .
18
Thus, A t B
***Acceptable responses are (1) the complement of A closure is open, (2) b A) ,

x c+
(3) N x
x, (4) N x, (5) N x is ropen, (6) N ` y t Ac , r(7) y + A) , (8)
c
b ) cc )
b cc b cc ss
N x t A (9) AC A , and (10) 1x x + A " 2Nr x Nr x t A
)
(11) A is closed, (12) A is closed, (13) A (14) x is a limit point of A (or x + A )
(15) * + N x (16) x is a limit point of B (or x + B ) ) (17) x + B, (18)
x + A " x + B.***
De¿nition 3.3.20 For a metric space X d and E t X, the set E is dense in X if
and only if
b c
1x x + X " x + E G x + E ) .
Remark 3.3.21 Note that for a metric space X d, E t X implies that E t X
because the space X is closed. On the other hand, if E is dense in X, then X t
E C E ) E. Consequently, we see that E is dense in a metric space X if and only
if E X.
Example 3.3.22 We have that the sets of rationals and irrationals are dense in Eu-
clidean 1-space. This was shown in the two Corollaries the Archimedean Principle
for Real Numbers that were appropriately named “Density of the Rational Num-
bers” and “Density of the Irrational Numbers.”
De¿nition 3.3.23 For a metric space X d and E t X, the set E is bounded if
and only if
d e
2M 2q M + U F q + X F E t N M q .
Excursion 3.3.24 Justify that each of the following sets is bounded in Euclidean
space.
j k
1. A x1 x2 + U2 : 1 n x1 2 F x2 3 1
j k
2. B x1 x2 x3 + U3 : x1 o 0 F x2 o 0 F x3 o 0 F 2x1 x2 4x3 2
where
Remark 3.3.25 Note that, for U2 , d,

0 if xy

dx y ,

1 if x / y
the space U2 is bounded. This example stresses that classi¿cation of a set as
bounded is tied to the metric involved and may allow for a set to be bounded
The de¿nitions of least upper bound and greatest lower bound directly lead to
the observation that they are limit points for bounded sets of real numbers.
Theorem 3.3.26 Let E be a nonempty set of real numbers that is bounded, :
sup E, and ; inf E. Then : + E and ; + E.
Space for illustration.
Proof. It suf¿ces to show the result for least upper bounds. Let E be a nonempty
set of real numbers that is bounded above and : sup E. If : + E, then : + E
E C E ) . For : +
E, suppose that h is a positive real number. Because : h : and
: sup E, there exists x + E such that : h x :. Since h was arbitrary,
1h h 0 " 2x : h x :
i.e., : is a limit point for E. Therefore, : + E as needed.
Remark 3.3.27 In view of the theorem we note that any closed nonempty set of
real numbers that is bounded above contains its least upper bound and any closed
nonempty set of real numbers that is bounded below contains its greatest lower
bound.
3.3.2 Open Relative to Subsets of Metric Spaces

Given a metric space X d, for any subset Y of X, d Y is a metric on Y . For
example, given the Euclidean metric de on U2 we have that de U
0 corresponds
to the (absolute value) Euclidean metric, d x y, on the reals. It is natural to
ask about how properties studied in the (parent) metric space transfer to the subset.
De¿nition 3.3.28 Given a metric space X d and Y t X. A subset E of Y is

open relative to Y if and only if
d b d ece
1 p p + E " 2r r 0 F 1q q + Y F d p q r " q + E
d e
1 p p + E " 2r r 0 F Y D Nr p t E .
b c
Example 3.3.29 For Euclidean 2-space, U2 d , consider the subsets
Q R Q R
Y x1 x2 + U2 : x1 o 3 and Z x1 x2 + U2 : x1 0 F 2 n x2 5 .
j k
(a) The set X 1 x1 x 2 + U2 : 3 njx1 5 F 1 x2 4 C
3 1 3 4 is k
not open relative to Y , while X 2 x 1 x2 + U2 : 3 n x1 5 F 1 x2 4
is open relative to Y .
j k
(b) The half open interval x1 x2 + U2 : x1 0 F 2 n x 2 3 is open rela-
tive to Z .
From the example we see that a subset of a metric space can be open relative to
another subset though it is not open in the whole metric space. On the other hand,
the following theorem gives us a characterization of open relative to subsets of a
metric space in terms of sets that are open in the metric space.
Theorem 3.3.30 Suppose that X d is a metric space and Y t X . A subset E of

Y is open relative to Y if and only if there exists an open subset G of X such that
E Y D G.
Proof. Suppose that X d is a metric space, Y t X, and E t Y .

If E is open relative to Y , then corresponding to each p j+ E there exists ak
neighborhood of p, Nr p p, such that Y DNr p p t E. Let D Nr p p : p + E .
By Theorems 3.3.5(a) and 3.3.13(#1), G CD is an open subset of X. Since
de f
p + Nr p p for each p + E, we have that E t G which, with E t Y , implies that
E t G D Y . On the other hand, the neighborhoods Nr p p were chosen such that
Y D Nr p p t E hence,

>b c >
Y D Nr p p Y D Nr p p Y D G t E.
p+E p+E
Therefore, E Y D G, as needed.
Now, suppose that G is an open subset of X such that E Y D G and
p + E. Then p + G and G open in X yields the existence of a neighborhood of p,
N p, such that N p t G. It follows that N p D Y t G D Y E. Since p was
arbitrary, we have that
d d ee
1p p + E " 2N p N p D Y t E
i.e., E is open relative to Y .

3.3.3 Compact Sets
In metric spaces, many of the properties that we study are described in terms of
neighborhoods. The next set characteristic will allow us to extract ¿nite collections
of neighborhoods which can lead to bounds that are useful in proving other results
about subsets of metric spaces or functions on metric spaces.
De¿nition 3.3.31 Given a metric space X d and A t X, the family

G : : : +
6 of X is an open cover for A if and only if G : is open for each : + and
of subsets
At G:.
:+
De¿nition 3.3.32 A subset K of a metric space X d is compact if and only if ev-
ery open cover of K has a ¿nite subcover
j i.e., given any open kcover
G : : : +
of K , there exists an n + M such that G :k : k + M F 1 n k n n is a cover for K .
We have just seen that a subset of a metric space can be open relative to another
subset without being open in the whole metric space. Our ¿rst result on compact
sets is tells us that the situation is different when we look at compactness relative to
subsets.
Theorem 3.3.33 For a metric space X d, suppose that K t Y t X. Then K is
compact relative to X if and only if K is compact relative to Y .
Excursion 3.3.34 Fill in what is missing to complete the following proof of Theo-
rem 3.3.33.
Proof. Let X d be a metric space and K t Y t X.
Suppose that K is compact relative to X and

U: : : +
is a family of sets such that, for each :, U: is open relative to Y
such that
>
K t U: .
:+
By Theorem 3.3.30, corresponding to each : + , there exists a

set G : such that G : is open relative to X and .
1
Since K t Y and
6 6 6
K t U: Y D G : , if
:+ :+ 1 :+
follows that
>
K t G:.
:+
Because K is compact relative to X, there exists a ¿nite number
of elements of , :1 :2 :n , such that
.
2
6
n
Now K t Y and K t G : j yields that
j1
>
n
K tYD G: j .
j1 3 4
Since
U: : : + was arbitrary, we have shown that every
open relative to Y cover of K has a ¿nite subcover. Therefore,
.
5
Conversely, suppose that K is compact relative to Y and

that
W: : : + is a family of sets such that, for each :, W:
is open relative to X and
>
K t W: .
:+
6 : + , let U: Y D W: . Now K t Y and

For each
K t W: implies that
:+
.
6
Consequently,
U: : : + is an open relative to Y
cover for K . Now K compact relative to Y yields that
there exists a ¿nite number of elements of ,
:1 :2 :n , such that . Since
7
>
n >
n b c >
n
U: j Y D W: j Y D W: j
j1 j1 j1
and K t Y , it follows that .

8
Since
W: : : + was arbitrary, we conclude that
every family of sets that form an open relative to X
cover of K has a ¿nite subcover. Therefore,
.
9
***Acceptable ¿ll-ins: (1) U: Y D G : , (2) K t G :1 C G :2 C C G :n (or

6
n n b
6 c 6
n
K t G : j ), (3) Y D G : j , (4) U: j , (5) K is compact relative to Y , (6)
j1 j1 j1
6 6 6 6
n 6
n
K tYD W: Y D W: U: , (7) K t U: j , (8) K t W: j ,
:+ :+ :+ j1 j1
(9) K is compact in X.***
Our next set of results show relationships between the property of being com-
pact and the property of being closed.
Theorem 3.3.35 If A is a compact subset of a metric space S d, then A is closed.
Excursion 3.3.36 Fill-in the steps of the proof as described

Proof. Suppose that A is a compact subset of a metric space S d and

1
p + S is such that p + A. For q + A, let rq d p q. The
j k 4
Nrq q : q + A is an open cover for A. Since A is compact,
there exists a ¿nite number of q, say q1 q2 qn , such that
A t Nrq1 q1 C Nrq2 q2 C C Nrqn qn W .

de f
(a) Justify that the set V Nrq1 p D Nrq2 p D D Nrqn p
is a neighborhood of p such that V D W 3.
(b) Justify that Ac is open.
(c) Justify that the result claimed in the theorem is true.
***For (a), hopefully you noted that taking r min rq j yields that Nrq1 p D
1n j nn
Nrq2 p D D Nrqn p Nr p. To complete (b), you needed to observe
that Nr p t Ac made p an interior point of Ac since p was an arbitrary point
satisfying p + A, it followed that Ac is open. Finally, part (c) followed from
Theorem 3.3.13(#3) which asserts that the complement of an open set is closed
thus, Ac c A is closed.***
Theorem 3.3.37 In any metric space, closed subsets of a compact sets are compact.
Excursion 3.3.38 Fill in the two blanks in order to complete the following proof of
the theorem.
Proof. For a metric space X d, suppose that F t K t X are such that F is
closed (relative to X) and K is compact. Let J
G : : : + be an open cover
for F. Then the family P
V : V + J G V F c is an open cover for K . It
follows from K being compact that there exists a ¿nite number of elements of P,
say V1 V2 Vn , such that
Because F t K , we also have that
If for some j + M, 1 n j n n, F c V j , the family

Vk : 1 n k n n F k / j
would still be a ¿nite open cover for F. Since J was an arbitrary open cover for
F, we conclude that every open cover of F has a ¿nite subcover. Therefore, F is
compact.
Corollary 3.3.39 If F and K are subsets of a metric space such that F is closed
and K is compact, then F D K is compact.
Proof. As a compact subset of a metric space, from Theorem 3.3.35, K is

closed. Then, it follows directly from Theorems 3.3.13(#5) and 3.3.37 that F D K
is compact as a closed subset of the compact set K .
Remark 3.3.40 Noticej that Theorem 3.3.35 and Theorem k 3.3.37 are not converses
of each other. The set x1 x2 + U : x1 o 2 F x 2 0 is an example of a closed
2
set in Euclidean 2-space that is not compact.

De¿nition 3.3.41 Let

Sn *n1 be a sequence of subsets of a metric space X. Then
*
Sn n1 is a nested sequence of sets if and only if 1n n + M " Sn1 t Sn .
De¿nition 3.3.42 A family D

A: : : + of sets in the universe X has the
¿nite intersection property if and only if the intersection over any ¿nite subfamily
of D is nonempty i.e.,

?
1P P t F P ¿nite " A; / 3 .
;+P
The following theorem gives a suf¿cient condition for a family of nonempty

compact sets to be disjoint. The condition is not being offered as something for you
to apply to speci¿c situations it leads us to a useful observation concerning nested
sequences of nonempty compact sets.
Theorem 3.3.43 If
K : : : + is a family of nonempty compact ? subsets of a met-
ric space X that satis¿es the ¿nite intersection property, then K : / 3.
:+
Space for notes.
?
Proof. Suppose that K : 3 and choose K = +
K : : : + . Since
? :+
K : 3,
:+

?
1x x + K = " x +
K: .
:+
Let
J
K : : : + F K : / K = .
?3.3.13(#3), K : is closed and

Because each K : is compact, by Theorems 3.3.35 and
K : is open. For any * + K = , we have that * +
c K : . Hence, there exists a
:+
; + such that * + K ; from which we conclude that * + K ;c and K ; / K = .

Since * was arbitrary, we have that
K r sL
1* * + K = " 2; ; + F K ; / K = F * + K ;c .
6
Thus, K = t G which establishes J as an open cover for K = . Because K = is
G+:
compact there exists a ¿nite number of elements of J, K :c1 K :c2 K :cn , such that
c
>
n ?
n
K= t K :c j K: j
j1 j1
from DeMorgan’s Laws from which it follows that

?n
K= D K : j 3.
j1
Therefore, there exists a ¿nite subfamily of

K : that is disjoint.
?
We have shown that if K : 3, then there exists a ¿nite subfamily of
:+
K : : : + that has empty intersection. From the Contrapositive Tautology, if
K : : : + is a family of nonempty compact subsets of? a metric space such that

the intersection of any ¿nite subfamily is nonempty, then K : / 3.
:+
Corollary
? 3.3.44 If
K n *
n1 is a nested sequence of nonempty compact sets, then
K n / 3.
n+M
7
j : j + 7
Proof. For any ¿nite subset of M, let m max . Because
K n *
n1
is a nested sequence on nonempty sets, K m t K j and K j / 3. Since
j+ j+
?
K n : n + M satis¿es the ¿nite intersection property.
was arbitrary, we conclude that
Hence, by Theorem 3.3.43, K n / 3.
n+M
Corollary 3.3.45 If
Sn *
n1 is a nested
? sequence of nonempty closed subsets of a
compact sets in a metric space, then Sn / 3.
n+M
Theorem 3.3.46 In a metric space, any in¿nite subset of a compact set has a limit
point in the compact set.
Space for notes and/or scratch work.
Proof. Let K be a compact subset of a metric space and E is a nonempty

subset of K . Suppose that no element of K is a limit point for E. Then for each
x in K there exists a neighborhood of x, say N x, such that N x
x D
E 3. Hence, N x contains at most one point from E namely x. The family
N x : x + K forms an open cover for K . Since K is compact, there exists a

¿nite number of elements in
N x : x + K , say N x1 N x2 N xn such
that K t N x 1 C N x2 C C N x n . Because E t K , we also have that
E t N x1 C N x2 C C N xn . From the way that the neighborhoods were
chosen, it follows that E t
x1 x2 xn . Hence, E is ¿nite.
We have shown that for any compact subset K of metric space, every subset
of K that has not limit points in K is ¿nite. Consequently, any in¿nite subset of K
must have at least one limit point that is in K .
3.3.4 Compactness in Euclidean n-space

Thus far our results related to compact subsets of metric spaces described implica-
tions of that property. It would be nice to have some characterizations for compact-
ness. In order to achieve that goal, we need to restrict our consideration to speci¿c
metric spaces. In this section, we consider only real n-space with the Euclidean
metric. Our ¿rst goal is to show that every n-cell is compact in Un . Leading up to
this we will show that every nested sequence of nonempty n-cells is not disjoint.
Theorem 3.3.47 (Nested Intervals Theorem) If

In *
n1 is a nested sequence of
7
*
intervals in U1 , then In / 3.
n1
Proof. For the nested sequence of intervals

In *
n1 , let In [an bn ] and A
*
an : n + M. Because
In n1 is nested, [an bn ] t [a1 b1 ] for each n + M. It
follows that 1n n + M " an n b1 . Hence, A is a nonempty set of real numbers

that is bounded above. By the Least Upper Bound Property, x sup A exists and
de f
is real. From the de¿nition of least upper bound, an n x for each n + M. For any
positive integers k and m, we have that
ak n akm n bkm n bk
from which it follows that x n bn for all n + M. Since an n x n bn for each n + J ,

7* 7
*
we conclude that x + In . Hence, In / 3.
n1 n1
Remark 3.3.48 Note that, for B

bn : n + J appropriate adjustments in the
proof that was given for the Nested Intervals Theorem would allow us to conclude
7
*
that inf B + In . Hence, if lengths of the nested integrals go to 0 as n goes to *,
n1
7
*
then sup A inf B and we conclude that In consists of one real number.
n1
The Nested Intervals Theorem generalizes to d nested e n-cells. The key is to have
the set-up that makes use of the n intervals x j y j , 1 n j n n, that can be
associated with x1 x2 xn and y1 y2 yn in Un .
Theorem 3.3.49 (Nested n-Cells Theorem) Let n be a positive integer. If

Ik *
k1
7
*
is a nested sequence of n-cells, then Ik / 3.
k1
Proof. For the nested sequence of intervals

Ik *
k1 , let
j k
Ik x1 x2 x n + Un : ak j n x j n bk j for j 1 2 n .
d e j k*
For each j , 1 n j n n, let Ik j ak j bk j . Then each Ik j k1 satis¿es the
conditions of the Nested Intervals Theorem. Hence, for each j , 1 n j n n, there
7
* 7
*
exists * j + R such that * j + Ik j . Consequently, *1 *2 *n + Ik as
k1 k1
needed.
Theorem 3.3.50 Every n-cell is compact.

Proof. For real constants a1 a2 an and b1 b2 bn such that a j b j for
each j 1 2 n, let
j b ck
I0 I x1 x2 xn + Un : 1 j + M 1 n j n n " a j n x j n b j
and
Y
X;
X n b c2
=W bj aj .
j1
d e
Then 1x 1y x y + I0 " x y n = . Suppose that I0 is not compact. Then
there exists an open cover J
G : : : + of I0 for which no ¿nite subcollection
covers I0 . Now we will describe the construction of a nested sequence of n-cells
each member of which is not compact. Use the space provided to sketch appropriate
pictures for n 1, n 2, and n 3 that illustrate the described construction.
aj bj
For each j , 1 n j n n, let c j . The sets of intervals
2
jb c k jb c k
aj cj : 1 n j n n and cj bj : 1 n j n n
can be used to determine or generate 2n new n-cells, Ik1 for 1 n k n 2n . For

example, each of
j b ck
x1 x2 xn + Un : j + M 1 n j n n " a j n x j n c j
j b ck
x1 x 2 xn + Un : 1 j + M 1 n j n n " c j n x j n b j
and
j k
x1 x2 xn + Un : a j n x j n c j if 2 j and c j n x j n b j if 2 0 j
Q R
is an element of Ik1 : 1 n k n 2n . For each k + M, 1 n k n 2n , Ik1 is a subset
6
2n
(sub-n-cell) of I0 and Ik1 I0 . Consequently, J
G : : : + is an open
k1
cover for each of the 2n sub-n-cells. Because I0 is such that noQ¿nite subcollection R
from J covers I0 , it follows that at least one of the elements of Ik1 : 1 n k n 2n
Q R
must also satisfy that property. Let I1 denote an element of Ik1 : 1 n k n 2n for
which no ¿nite subcollection from J covers I1 . For x1 x2 xn + I1 we have
that either a j n x j n c j or c j n x j n b j for each j , 1 n j n n. Since
cj aj bj cj bj aj

2 2 2
it follows that, for x x1 x 2 xn y y1 y2 yn + I1
Y Y
X; X n b c
X n b c2 X ; bj aj 2 =
d x y W yj xj n W
2 2 2
j1 j1
=
i.e., the diam I1 is .
2
The process just applied to I0 to obtain I1 can not be applied to obtain a
sub-n-cell of I1 that has the transferred properties. That is, if
Q r sR
I1 x1 x 2 xn + Un : 1 j + M 1 n j n n " a 1
j n x j n b 1
j ,
a 1 1
j bj
letting c1
j generates two set of intervals
2
Qr s R Qr s R
a 1 1
c1 1
j c j : 1 n j n n and j b j : 1 n j n n
that will determine 2n new n-cells, Ik2 for 1 n k n 2n , that are sub-n-cells
of I1 . Now, since J is an open cover for I1 such that no ¿nite subcollection
6
2n
from J covers I1 and Ik2 I1 , it follows that there is at least one element
k1
Q R
of Ik2 : 1 n k n 2n that cannot be covered with a ¿nite subcollection from J
choose one of those elements and denote it by I2 . Now the choice of c1j allows
diam I1 =
us to show that diam I2 2 . Continuing this process generates
2 2
Ik *
k0 that satis¿es each of the following properties:

Ik *
k0 is a nested sequence of n-cells,
for each k + M, no ¿nite subfamily of J covers Ik , and

d e
1x 1y x y + Ik " x y n 2k = .
7
* 7
*
From the Nested n-cells Theorem, Ik / 3. Let ? + Ik . Because
k0 k0
7
*
J
G : : : + is an open cover for I0 and Ik t I0 , there exists G + J such
k0
that ? + G. Since G is open, we there is a positive real number r such that Nr ? t
G. Now diam Nr ? 2r and, for n + M large enough, diam In 2n = 2r .
Now, ? + Ik for all k + M assures that ? + Ik for all k o n. Hence, for all k + M
such that k o n, Ik t Nr ? t G. In particular, each Ik , k o n, can be covered by
one element of J which contradicts the method of choice that is assured if I0 is not
compact. Therefore, I0 is compact.
The next result is a classical result in analysis. It gives us a characterization for
compactness in real n-space that is simple most of the “hard work” for the proof
was done in when we proved Theorem 3.3.50.
Theorem 3.3.51 (The Heine-Borel Theorem) Let A be a subset of Euclidean n-

space. Then A is compact if and only if A is closed and bounded.
Proof. Let A be a subset of Euclidean n-space Un d

Suppose that A is closed and bounded. Then there exists an n-cell I such
that A trI . sFor example, because A is bounded, there exists M 0 such that

A t N M 0 for this case, the n-cell
| }
n n
I x1 x2 xn + R : max n x j n n M 1
n
1n jnn
satis¿es the speci¿ed condition. From Theorem 3.3.50, I is compact. Since A t I

and A is closed, it follows from Theorem 3.3.37 that A is compact.
Suppose that A is a compact subset of Euclidean n-space. From Theorem

3.3.35, we know that A is closed. Assume that A is not bounded and let p1 + A.
Corresponding to each m + M choose a pm in A such that pm / pk for k
1 2 m 1 and d p1 pm m 1. As an in¿nite subset of the compact
set A, by Theorem 3.3.46,
pm : m + M has a limit point in A. Let q + A be a
limit point for
pm : m + M. Then, for each t + M, there exists pm t +
pm : m + M
b c 1
such that d pm t q . From the triangular inequality, it follows that for any
t 1
pm t +
pm : m + M,
b c b c 1
d pm t p1 n d pm t q d q p1 d q p1 1 d q p1 .
1t
b c
But 1d q p1 is a ¿xed real number, while pm t was chosen such that d pm t p1
m t 1 and m t 1 goes to in¿nity as t goes to in¿nity. Thus, we have reached a
contradiction. Therefore, A is bounded.
The next theorem gives us another characterization for compactness. It can be
shown to be valid over arbitrary metric spaces, but we will show it only over real
n-space.
Theorem 3.3.52 Let A be a subset of Euclidean n-space. Then A is compact if and

only if every in¿nite subset of A has a limit point in A.
of Theorem 3.3.52.
Proof. If A is a compact subset of Euclidean n-space, then every in¿nite subset

of A has a limit point in A by Theorem 3.3.46.
Suppose that A is a subset of Euclidean n-space for which every in¿nite
subset of A has a limit point in A. We will show that this assumption implies that A
is closed and bounded. Suppose that * is a limit point of A. Then, for each n + M,
there exists an xn such that
x n + N 1 *
* .
n
Let S
xn : n + M. Then S is an of A. Conse-
1
quently, S has in A. But S has only one limit point
2
namely . Thus, * + A. Since * was arbitrary, we conclude that A contains

3
all of its limit point i.e., .
4
Suppose that A is not bounded. Then, for each n + M, there exists yn such
that yn n. Let S
yn : n + M. Then S is an of
5
A that has no ¿nite limit point in A. Therefore,
b c
A not bounded " 2S S t A F S is in¿nite F S D A) 3
taking the contrapositive and noting that P F Q F M is logically equivalent to

[P F Q " M] for any propositions P, Q and M, we conclude that

1S " S D A) / 3
6
" .
7
***Acceptable completions include: (1) in¿nite subset, (2) a limit point, (3) *, (4)
A is closed, (5) in¿nite subset, (6) S t A F S is in¿nite, and (7) A is bounded.***
As an immediate consequence of Theorems 3.3.50 and 3.3.46, we have the fol-
lowing result that is somewhat of a generalization of the Least Upper Bound Prop-
erty to n-space.
Theorem 3.3.54 (Weierstrass) Every bounded in¿nite subset of Euclidean n-space

has a limit point in Un .
3.3.5 Connected Sets

With this section we take a brief look at one mathematical description for a subset
of a metric space to be “in one piece.” This is one of those situations where “we
recognize it when we see it,” at least with simply described sets in U and U2 . The
concept is more complicated than it seems since it needs to apply to all metric spaces
and, of course, the mathematical description needs to be precise. Connectedness is
de¿ned in terms of the absence of a related property.
De¿nition 3.3.55 Two subsets A and B of a metric space X are separated if and
only if
A D B 3 F A D B 3.
De¿nition 3.3.56 A subset E of a metric space X is connected if and only if E is

not the union of two nonempty separated sets.
Example 3.3.57 To justify that A

x + U : 0 x 2 G 2 x n 3 is not con-
nected, we just have to note that B1
x + U : 0 x 2 and
B2
x + U : 2 x n 3 are separated sets in U such that A B1 C B2 .
Example 3.3.58 In Euclidean 2-space, if C D1 C D2 where

Q R
D1 x1 x2 + U2 : d 1 0 x1 x2 n 1
and
Q R
D2 x1 x2 + U2 : d 1 0 x1 x2 1 ,
then C is a connected subset of U2 .
Remark 3.3.59 The following is a symbolic description for a subset E of a metric

space X to be connected:
1A 1B [A t X F B t X Fb E A C B c

" A D B / 3 G A D B / 3 G A 3 G B 3 ].
The statement is suggestive of the approach that is frequently taken when trying
to prove sets having given properties are connected namely, the direct approach
would take an arbitrary set E and let E A C B. This would be followed by using
other information that is given to show that one of the sets must be empty.
The good news is that connected subsets of U1 can be characterized very nicely.
Theorem 3.3.60 Let E be a subset of U1 . Then E is connected in U1 if and only if

Kr s L
1x 1y 1z x y + E F z + U1 F x z y " z + E .
of the Theorem.
Proof. Suppose that E is a subset of U1 with the property that there exist real
numbers x and y with x y such that x y + E and, for some z + U1 ,
z + x y and z +
E
Let Az E D * z and Bz E D z *. Since z +
E, E A z C Bz . Because
x + Az and y + Bz , both Az and Bz are . Finally, Az t * z
1
and Bz t z * yields that
Az D Bz Az D B z .
2
Hence, E can be written as the union of two sets i.e., E is
3
. Therefore, if E is connected, then x y + E F z +
4
U F x z y implies that .
5
To prove the converse, suppose that E is a subset of U1 that is not con-
nected. Then there exist two nonempty separated subsets of U1 A and B, such that
E AC B.d Choose
e x + A and y + B and assume that the set-up admits that x y.
Since A D x y is a nonempty subset of real numbers, by the least upper bound
b d ec
property, z sup A D x y exists and is real. From Theorem 3.3.26, z + A
de f
then AD B 3 yields that z + B. Now we have two possibilities to consider z +
A
and z + A. If z +
A, then z +
A C B E and x z y. If z + A, then A D B 3
implies that z +
B and we conclude that there exists * such that z * y and
*+ B. From z *, * + A. Hence,
db * + A C B E and x * c y. In either e
case, we have that 1x 1y 1z dbx y + E F z + U1 F x z y c" z + E e.
By the contrapositive 1x 1y 1z x y + E F z + U1 F x z y " z + E
implies that E is connected.
***Acceptable responses are: (1) nonempty, (2) 3, (3) separated, (4) not connected,
and (5) E is connected.***
From the theorem, we know that, for a set of reals to be connected it must be
either empty, all of U, an interval, a segment, or a half open interval.
3.3.6 Perfect Sets

De¿nition 3.3.62 A subset E of a metric space X is perfect if and only if E is
closed and every point of E is a limit point of E.
Alternatively, a subset E of a metric space X is perfect if and only if E is closed

and contains no isolated points.
From Theorem 3.3.7, we know that any neighborhood of a limit point of a subset
E of a metric space contains in¿nitely many points from E. Consequently, any
nonempty perfect subset of a metric space is necessarily in¿nite with the next
theorem it is shown that, in Euclidean n-space, the nonempty perfect subsets are
uncountably in¿nite.
Theorem 3.3.63 If P is a nonempty perfect subset of Euclidean n-space, then P is

uncountable.
Proof. Let P be a nonempty perfect subset of Un . Then P contains at least one

limit point and, by Theorem 3.3.6, P is in¿nite. Suppose that P is denumerable. It
follows that P can be arranged as an in¿nite sequence let
x1 x2 x 3
represent the elements of P. First,

j k* we will justify the existence (or construction) of
a sequence of neighborhoods V j j1 that satis¿es the following conditions:
b c
(i) 1 j j + M " V j1 l V j ,
b c
(ii) 1 j j + M " x j +
V j1 , and
b c
(iii) 1 j j + M " V j D P / 3 .
j kn neighborhood of x1 i.e., let V1 be any neighborhood of

Start with an arbitrary
x1 . Suppose that V j j1 has been constructed satisfying conditions (i)–(iii) for
1 n j n n. Because P is perfect, every x + Vn D P is a limit point of P. Thus there
are an in¿nite number of points of P that are in Vn and we may choose y + Vn D P
such that y / xn . Let Vn1 be a neighborhood of y such that xn + V n1 and
V n1 l Vn . Show that you can do this.
j kn1
Note that Vn1 D P / 3 since y + Vn1 D P. Thus we have a sequence V j j1
satisfying (i)–(iii) for 1 n j n n 1. By the Principle of Complete Induction we
can construct the desired sequence.
j k*
Let K j j1 be the sequence de¿ned by K j V j D P for each j. Since V j and
P are closed, K j is closed. Since V j is bounded, K j is bounded. Thus K j is closed
and bounded and hence compact. Since x j + K j1 , no point of P lies in D*
j 1 K j .
*
Since K j l P, this implies D j1 K j 3. But each K j is nonempty by (iii) and
K j m K j1 by (i). This contradicts the Corollary 3.3.27.
Corollary 3.3.64 For any two real numbers a and b such that a b, the segment
a b is uncountable.
The Cantor Set
The Cantor set is a fascinating example of a perfect subset of U1 that contains no

segments. In Chapter 11 the idea of the measure of a set is studied it generalizes
the idea of length. If you take MAT127C, you will see the Cantor set offered as an
example of a set that has measure zero even though it is uncountable.
The Cantor set is de¿ned to be the intersection of a sequence of closed subsets
of [0 1] the sequence of closed sets is de¿ned recursively. Let E 0 [0 1]. For
E 1 partition the interval E 0 into three subintervals of equal length and remove the
middle segment (the interior of the middle section). Then
v w v w
1 2
E 1 0 C 1 .
3 3
v w v w
1 2
For E 2 partition each of the intervals 0 and 1 into three subintervals of
3 3
equal length and remove the middle segment from each of the partitioned intervals
then
v w v w v w v w
1 2 1 2 7 8
E 2 0 C C C 1
9 9 3 3 9 9
v w v w v w v w
1 2 3 6 7 8
0 C C C 1 .
9 9 9 9 9 9
Continuing the process E n will be the union of 2n intervals. To obtain En1 , we

partition each of the 2n intervals into three subintervals of equal length and remove
the middle segment, then E n1 is the union of the 2n1 intervals that remain.
Excursion 3.3.65 In the space provided sketch pictures of E 0 E 1 E2 and E 3 and

¿nd the sum of the lengths of the intervals that form each set.
By construction
E n *
n1 is a nested sequence of compact subsets of U .
1
Excursion 3.3.66 Find a formula for the sum of the lengths of the intervals that
form each set E n .
?
*
The Cantor set is de¿ned to be P En .
n1
Excursion 3.3.67 Justify each of the following claims.
(a) The Cantor set is compact.
(b) The
E n *
n1 satis¿es the ¿nite intersection property
Remark 3.3.68 It follows from the second assertion that P is nonempty.

Finally we want to justify the claims that were made about the Cantor set before
we described its construction.
The Cantor set contains no segment from E 0 .
To see this, we observe that each segment in the form of
t u
3k 1 3k 2
for k m + M
3m 3m
is disjoint from P. Given any segment : ; for : ;, iftm + M is such thatu
; : 3k 1 3k 2
3m , then : ; contains an interval of the form
6 3m 3m
from which it follows that : ; is not contained in P.
The Cantor set is perfect. For x + P, let S be any segment that contains x.
?
*
Since x + E n , x + E n for each n + M. Corresponding to each n + M, let
n1
In be the interval in E n such that x + In . Now, choose m + M large enough
to get Im t S and let xm be an endpoint of Im such that xm / x. From the
way that P was constructed, xm + P. Since S was arbitrary, we have shown
that every segment containing x also contains at least one element from P.
Hence, x is a limit point of P. That x was arbitrary yields that every element
of P is a limit point of P.
3.4 Problem Set C

1. For x x 1 x2 x N and y y1 y2 y N in U N , let
Y
X N
X; b c2
dx y W x j yj .
j1
Prove that U N d is a metric space.

3.4. PROBLEM SET C 119

;
N
Dx y x j y j .
j1
Prove that U N D is a metric space.

n n
d*x y max nx j y j n .
1n jnN
Prove that U N d* is a metric space.

4. Show that the Euclidean metric d, given in problem #1, is equivalent to the
metric d* , given in problem #3.
b c
5. Suppose that S d is a metric space. Prove that S d ) is a metric space
where
d x y
d ) x y .
1 d x y
G
[Hint: You might ¿nd it helpful to make use of properties of h G
1G
for G o 0.]
6. If a1 a2 an are positive real numbers, is
;
n
d x y ak xk yk
k1
where x x1 x2 xn y y1 y2 yn + Un , a metric on Un ? Does

your response change if the hypothesis is modi¿ed to require that a1 a2 an
are nonnegative real numbers?
7. Is the metric D, given in problem #2, equivalent to the metric d*, given in
problem #3? Carefully justify your position.
b c
8. Are the metric spaces U N d and U N d ) where the metrics d and d ) are
given in problems #1 and #5, respectively, equivalent? Carefully justify the
position taken.
b c
9. For x1 x 2 and x1) x2) in U2 ,
n n n n
b b cc x2 nx2) n nx 1 x1) n , if x1 / x1)
d3 x1 x2 x1) x2
)

nn n
x2 x ) n
2 , if x1 x1)
b c
Show that U2 d3 is a metric space.
10. For x y + U1 , let d x y x 3y. Is U d a metric space? BrieÀy

justify your position.
11. For U1 with d x y x y, give an example of a set which is neither

open nor closed.
12. Show that, in Euclidean n space, a set that is open in Un has no isolated
points.
13. Show that every ¿nite subset of U N is closed.
14. For U1 with the Euclidean metric, let A

x + T : 0 n x n 1. Describe A.
15. Prove each of the following claims that are parts of Theorem 3.3.13. Let S be
a metric space.
(a) The union of any family I of open subsets of S is open.

(b) The intersection of any family I of closed subsets of S is closed.
6mA1 A2 Am is a ¿nite family of closed subsets of S, then the union

(c) If
j1 A j is closed.
(d) The space S is both open and closed.
(e) The null set is both open and closed.
r r T Ls
16. For X [8 4 C
2 0 C T D 1 2 2 as a subset of U1 , identify
(describe or show a picture of) each of the following.
(a) The interior of X, Int X

(b) The exterior of X, Ext X
(c) The closure of X , X
3.4. PROBLEM SET C 121
(d) The boundary of X, " X

(e) The set of isolated points of X
(f) The set of lower bounds for X and the least upper bound of X, sup X
17. As subsets of Euclidean 2-space, let

| }
1
A x1 x2 + U : max
x 1 1 x2 n
2
,
2
j k
B x1 x2 + U2 : max
x1 1 x 2 n 1 and
Q r sR
Y x1 x2 + U2 : x1 x2 + B A G x1 12 x22 1 .
(a) Give a nicely labelled sketch of Y on a representation for the Cartesian

coordinate plane.
(b) Give a nicely labelled sketch of the exterior of Y , Ext Y , on a repre-
sentation for the Cartesian coordinate plane.
(c) Is Y open? BrieÀy justify your response.
(d) Is Y closed? BrieÀy justify your response.
(e) Is Y connected? BrieÀy justify your response.
18. Justify each of the following claims that were made in the Remark following
De¿nition 3.3.15
(a) If A is a subset of a metric space S d, then Ext A Int Ac .
(b) If A is a subset of a metric space S d, then
b c
x + " A % 1Nr x Nr x D A / 3 F Nr x D Ac / 3 .
19. For U2 with the Euclidean metric, show that the set
Q R
S x y + U2 : 0 x 2 y 2 1
is open. Describe each of S 0 S ) " S S and S c .

j k
20. Prove that x1 x2 + U2 : 0 n x1 1 F 0 n x2 n 1 is not compact.
21. Prove that T, the set of rationals in U1 , is not a connected subset of U1 .
22. Let I be any family of connected subsets of a metric 6

space X such that any
two members of I have a common point. Prove that F is connected.
F+5
23. Prove that if S is a connected subset of a metric space, then S is connected.
24. Prove that any interval I t U1 is a connected subset of U1 .
25. Prove that if A is a connected set in a metric space and A t B t A, then B

is connected.
26. Let
Fn *
n1 be a nested sequence of compact sets, each of which is con-
?
*
nected. Prove that Fn is connected.
n1
27. Give an example to show that the compactness of the sets Fk given in problem
#26 is necessary i.e., show that a nested sequence of closed connected sets
would not have been enough to ensure a connected intersection.
Chapter 4
Sequences and Series–First View
Recall that, for any set A, a sequence of elements of A is a function f : M A.

Rather than using the notation f n for the elements that have been selected from
A, since the domain is always the natural numbers, we use the notational convention
an f n and denote sequences in any of the following forms:

an *
n1
an n+M or a1 a2 a3 a4
This is the only time that we use the set bracket notation
in a different con-
text. The distinction is made in the way that the indexing is communicated . For
an :, the
an * n1 is the constant sequence that “lists the term : in¿nitely often,”
: : : : while
an : n + M is the set consisting of one element :. (When you
read the last sentence, you should have come up with some version of “For ‘a sub
n’ equal to :, the sequence of ‘a sub n’ for n going from one to in¿nity is the
constant sequence that “lists the term : in¿nitely often,” : : : while the set
consisting of ‘a sub n’ for n in the set of positive integers is the set consisting of
one element :” i.e., the point is that you should not have skipped over the
an * n1
and
an : n + M.) Most of your previous experience with sequences has been with
sequences of real numbers, like
| }*
3
1 1 2 3 5 8 13 21 34 55 ,...
n 1 n1
| }* | }* | r nH s}*
n 2 3n 5 n3 1 log n
1n and sin .
n 47 n1 n3 1 n1 n 8 n1
In this chapter, most of our sequences will be of elements in Euclidean n-space. In
MAT127B, our second view will focus on sequence of functions.
123
124 CHAPTER 4. SEQUENCES AND SERIES–FIRST VIEW
As children, our ¿rst exposure to sequences was made in an effort to teach us to

look for patterns or to develop an appreciation for patterns that occur naturally.
Excursion 4.0.1 For each of the following, ¿nd a description for the general term
as a function of n + M that ¿ts the terms that are given.
2 4 8 16 32 64
1.
5 7 9 11 13 15
3 7 11
2. 1 9 81 729
5 9 13
***An equation that works for (1) is 2n 2n 31 while (2) needs a different for-
mula for the odd terms and the even terms one pair that works is 2n 1 2n 11
for n even and 3n1 when n is odd.***
As part of the bigger picture, pattern recognition is important in areas of math-
ematics or the mathematical sciences that generate and study models of various
phenomena. Of course, the models are of value when they allow for analysis and/or
making projections. In this chapter, we seek to build a deeper mathematical under-
standing of sequences and series primary attention is on properties associated with
convergence. After preliminary work with sequences in arbitrary metric spaces, we
will restrict our attention to sequences of real and complex numbers.
4.1 Sequences and Subsequences in Metric Spaces

If you recall the de¿nition of convergence from your frosh calculus course, you
might notice that the de¿nition of a limit of a sequence of points in a metric space
merely replaces the role formerly played by absolute value with its generalization,
distance.
4.1. SEQUENCES AND SUBSEQUENCES IN METRIC SPACES 125

pn *n1 denote a sequence of elements of a metric space S d
and p0 be an element of S. The limit of
pn *
n1 is p0 as n tends to (goes to or ap-
proaches) in¿nity if and only if
d e
1 + U F 0 " 2M M M + MF 1n n M " d pn p0
We write either pn p0 or lim pn p0 .

n*
Remark 4.1.2 The description M M indicates that “limit of sequence proofs”
require justi¿cation or speci¿cation of a means of prescribing how to ¿nd an M that
“will work” corresponding to each 0. A function that gives us a nice way to
specify M ’s is de¿ned by
LxM inf
j + ] : x n j
z {
1
and is sometimes referred to as the ceiling function. Note, for example, that
2
1, L22M 2, and L5M 5. Compare this to the greatest integer function, which
is sometimes referred to as the Àoor function.
| }*
2
Example 4.1.3 The sequence has the limit 0 in U. We can take M 1 2,
t u t nu n1 z {
1 3 700
M 200, and M 234. Of course, three examples
100 350 3 z {
2
does not a proof make. In general, for 0, let M . Then n M

implies that
z {
2 2
n o 0

n n
1 2 nn 2 nn
which, by Proposition 1.2.9 (#7) and (#5), implies that and n n .
n 2 n n
Using the de¿nition to prove that the limit of a sequence is some point in the
metric space is an example of where our scratch work towards ¿nding a proof might
be quite different from the proof that is presented. This is because we are allowed
to “work backwards” with our scratch work, but we can not present a proof that
starts with the bound that we want to prove. We illustrate this with the following
excursion.
Excursion 4.1.4 After reading the presented
| }scratch work, ¿ll in what is missing to
*
1 in
complete the proof of the claim that converges to i in F.
n 1 n1
(a) Scratch work towards a proof. Because i + F, it suf¿ces to show that
1 in
lim i. Suppose 0 is given. Then
n* n 1
n n n n n n T T
n 1 in n n 1 in i n 1 n n 1 i n 2 2
n n n nn n
n n 1 in n n1 n n n1 n n1

n

T T
2 2
whenever n. So taking M will work.

(b) A proof. For 0, let M . Then n + M and n M

1
T
2
implies that n which is equivalent to . Because
2
T 4
T 2
n1 n and 2 0, we also know that . Consequently, if
n
3
n M , then
n n n n n n T
n1 in n n 1 i n i n 1 n n 1 i n 2
n nn nn n
nn1 i n n n1 n nn 1n n 1
5
Since 0 was arbitrary, we conclude that

v t t n n uuw
n 1 in n
1 0 " 2M M + MF 1n n M " nn i nn
n1
1 in
i.e., . Finally, i 0 1 + F and lim
n* n 1
6
| }*
1 in
i yields that converges to i in F.
n1 n1
T
T T
2 2 T 2
***Acceptable responses are (1) , (2) , (3) n 1, (4) 2, (5) , (6)
n n
1 in
lim i .***
n* n 1
De¿nition 4.1.5 The sequence

pn * n1 of elements in a metric space S is said to
converge (or be convergent) in S if there is a point p0 + S such that lim pn p0
n*
it is said to diverge in S if it does not converge in S .
Remark 4.1.6 Notice that a sequence in a metric space S will be divergent | in}*
S if
2
its limit is a point that is not in S. In our previous example, we proved that
| }* n n1
2
converges to 0 in U consequently, is convergent in Euclidean 1-space.
| }* n n1
2 b c
On the other hand, is divergent in U
x + U : x 0 d where d
n n1
denotes the Euclidean metric on U, d x y x y.
Our ¿rst result concerning convergent sequences is metric spaces assures us of

the uniqueness of the limits when they exist.
Lemma 4.1.7 Suppose

pn *
n1 is a sequence of elements in a metric space S d.
Then
rK L s
1 p 1q p q + S F lim pn p F lim pn q " q p .
n* n*
Excursion 4.1.8 Fill in what is missing in order to complete the proof of the lemma.
Proof. Let
pn * n1 be a sequence of elements in a metric space S d for
which there exists p and q in S such that lim pn p and lim pn q. Suppose
n* n*
1
the p / q. Then d p q 0 and we let d p q. Because lim pn p and
2 n*
0, there exists a positive integer M1 such that
n M1 " d pn p
similarly, lim pn q yields the existence of a positive integer M2 such that

n*
.
1
Now, let M max

M1 M2 . It follows from the symmetry property and the
triangular inequality for metrics that n M implies that

d p q n d p pn 2 d p q
2 3
which contradicts the trichotomy law. Since we have reached a contradiction, we

conclude that as needed. Therefore, the limit of any convergent sequence
4
in a metric space is unique.
***Acceptable ¿ll-ins are: (1) n M2 " d pn q , (2) d pn q (3)

1
d p q, (4) p q.***
2
De¿nition 4.1.9 The sequence

pn *
n1 of elements in a metric space S d is
bounded if and only if
d e
2M 2x M 0 F x + S F 1n n + M " d x pn M .
Note that if the sequence

pn *
n1 of elements in a metric space S is an not
bounded, then the sequence is divergent in S. On the other hand, our next result
shows that convergence yields boundedness.
Lemma 4.1.10 If the sequence

pn *
n1 of elements in a metric space S d is con-
vergent in S, then it is bounded.
Proof. Suppose that

pn *
n1 is a sequence of elements in a metric space S d
that is convergent to p0 + S. Then, for 1, there exists a positive integer
M M 1 such that
n M " d pn p0 1.
j b c k
Because d p j p0 : j + M F 1 n j n M is a ¿nite set of nonnegative real num-
bers, it has a largest element. Let
j j b c kk
K max 1 max d p j p0 : j + M F 1 n j n M .
Since d pn p0 n K , for each n + M, we conclude that
pn *
n1 is bounded.
Remark 4.1.11 To see that the converse of Lemma 4.1.10 is false, for n + M , let

!
!
1
, if 2 n
!
n2
pn .
!
!
!
1
1
, if 2 0 n
n3
Then, for d the Euclidean metric on U1 , d 0 pn 0 pn 1 for all n + M, but
pn *
n1 is not convergent in U.
Excursion 4.1.12 For each n + M, let an p2n and bn p2n1 where pn is

de¿ned in Remark 4.1.11.
(a) Use the de¿nition to prove that lim an 0.
n*
(b) Use the de¿nition to prove that lim bn 1.

n*
1 1 1
***Note that an and bn 1 1 if you
2n 2 2n 1 3 2 n 1
1 1
used 2 and 1 , respectively, your choices for corresponding M will be
n n3
slightly off. The following are acceptable solutions, which of course are not unique
compare what you did for general sense and content. Make especially certain that
you did not offer a proof that is “working backwards”
z { from what you wanted to
1
show. (a) For 0, let M M T . Then n M implies that
n 2 n
b T c1 1 T n 1 n
n 2 or . If follows that n n 0 n 1 1 1
2n 2n 2 n 2n2 2n 2n
T T
whenever n M. Since 0 was arbitrary,
z we { conclude that
1 1
lim an lim 0. (b) For 0, let M M . Then n M
n* n* 2n2
1
implies that n 1 or . Note that, for n + M, n o 1 0 implies that
n
n 2 0 2 2 0 and 2n 2 n n 2 0 n n. Thus, for n + M
and n M, we have that
nt u n n n
n 1 n n 1 n 1 1
n 1 1nn nn n .
n 2n 2 2n 2 n 2n 2 n
t u
1
Since 0 was arbitrary, we conclude that lim bn lim 1
n* n* 2n 2
1.***
Remark 4.1.13 Hopefully, you spotted that there were some extra steps exhibited
in our solutions to Excursion 4.1.12. I chose to show some of the extra steps that
illustrated where we make explicit use of the ordered ¿eld properties that were dis-
cussed in Chapter 1. In particular, it is unnecessary for you to have explicitly
demonstrated that 2n 2 n from the inequalities that were given in Proposition
1 1 T T
1.2.9 or the step that was shown in part (a). For the former
2n 2n
you can just write things like 2n 2 n for the latter, you could just have written
1 bT c2
.
2n 2
What we just proved about the sequence given in Remark 4.1.11 can be trans-
lated to a statement involving subsequences.
De¿nition 4.1.14 Given a sequence

pn * n1 of elements in a metric space X and
*
a sequencej
n kkk1 of positive integers such that of n k n k1
* *
k*each k + M, the
j for
sequence pn k k1 is called a subsequence of
pn n1 . If pn k k1 converges in
X then its limit is called a subsequential limit of
pn *n1 .
Remark 4.1.15 In our function terminology, a subsequence of f : M X is the

restriction of f to any in¿nite subset of M with the understanding that ordering is
conveyed by the subscripts i.e., n j n j1 for each j + M.
From Excursion 4.1.12, we know that the sequence

pn * n1 given in Remark
4.1.11 has two subsequential limits namely, 0 and 1. The uniqueness of the limit
of a convergent sequence leads us to observe that every subsequence of a conver-
gent sequence must also be convergent to the same limit as the original sequence.
Consequently, the existence of two distinct subsequential limits for
pn * n1 is an
*
alternative means of claiming that
pn n1 is divergent. In fact, it follows from
the de¿nition of the limit of a sequence that in¿nitely many terms outside of any
neighborhood of a point in the metric space from which the sequence is chosen
will eliminate that point as a possible limit. A slight variation of this observation
is offered in the following characterization of convergence of a sequence in metric
space.
Lemma 4.1.16 Let

pn * n1 be a sequence of elements from a metric space X d.
Then
pn *n1 converges to p + X if and only if every neighborhood of p contains
all but ¿nitely many of the terms of
pn *
n1 .
Proof. Let
pn *n1 be a sequence of elements from a metric space X d.
Suppose that
pn * n1 converges to p + X and V is a neighborhood of
p. Then there exists a positive real number r such that V Nr p. From the
de¿nition of a limit, there exists a positive integer M M r such that n M
implies that d p pn r i.e., for all n M, pn + V . Consequently, at most the
pk : k + M F 1 n k n M is excluded from V . Since V was arbitrary, we conclude

that every neighborhood of p contains all but ¿nitely many of the terms of
pn * n1 .
Suppose that every neighborhood of p contains all but ¿nitely many of
the terms of
pn *n1 . For any 0, N p contains all but ¿nitely many of the
*
terms of
pn n1 . Let M max
k + M : pk + N p. Then n M implies that
pn + N p from which it follows that d pn p . Since 0 was arbitrary,
we conclude that, for every 0 there exists a positive integer M M such
that n M implies that d pn p that is,
pn * n1 converges to p + X.
It will come as no surprise that limit point of subsets of metric spaces can be
related to the concept of a limit of a sequence. The approach used in the proof of
the next theorem should look familiar.
Theorem 4.1.17 A point p0 is a limit point of a subset A of a metric space X d

if and only if there is a sequence
pn *
n1 with pn + A and pn / p0 for every n
such that pn p0 as n *
Proof. ! Suppose that there is a sequence

pn * n1 such that pn + A, pn /
p0 for every n, and pn p0 . For r 0, consider the neighborhood Nr p0 . Since
pn p0 , there exists a positive integer M such that d pn p0 r for all n M.
In particular, p M1 + A D Nr p0 and p M1 / p0 . Since r 0 was arbitrary, we
conclude that p0 is a limit point of the set A.
" Suppose that p0 + X is a limit

| }point of A. (Finish this part by ¿rst making
1 *
judicious use of the real sequence to generate a useful sequence
pn * n1
j j1
1
followed by using the fact that 0 as j * to show that
pn *
n1 converges
j
to p0 .)
Remark 4.1.18 Since Theorem 4.1.17 is a characterization for limit points, it gives
us an alternative de¿nition for such. When called upon to prove things related
to limit points, it can be advantageous to think about which description of limit
points would be most fruitful i.e., you can use the de¿nition or the characterization
interchangeably.
We close this section with two results that relate sequences with the metric space
properties of being closed or being compact.
Theorem 4.1.19 If
pn * n1 is a sequence in X and X is a compact subset of a
metric space S d, then there exists a subsequence of
pn *
n1 that is convergent
in X.
Proof. Suppose that

pn *
n1 is a sequence in X and X is a compact subset of
a metric space S d. Let P
pn : n + M. If P is ¿nite, then there is at least
one k such that pk + P and, for in¿nitely many j + M, we have that p j pk .
j k*
Consequently, we can choose a sequence n j j1 such that n j n j1 and pn j k
j k*
pk for each j + M. It follows that pn j j1 is a (constant) subsequence of
pn *n1
that is convergent to pk + X. If P is in¿nite, then P is an in¿nite subset of a
compact set. By Theorem 3.3.46, it follows that P has a limit point p0 in X. From
Theorem 4.1.17, we conclude that there is a sequence
qk * k1 with qk + P and
qk / p0 for every k such that qk p0 as k * that is,
qk *k1 is a subsequence
of
pn *
n1 that is convergent to p 0 + X.
Theorem 4.1.20 If
pn * n1 is a sequence in a metric space S d, then the set of
all subsequential limits of
pn *
n1 is a closed subset of S.
Proof. Let E ` denote the set of all subsequential limits of the sequence
pn * n1
of elements in the metric space S d. If E ` is ¿nite, then it is closed. Thus,
we can assume that E ` is in¿nite. Suppose that * is a limit point of E ` . Then,
corresponding to r 1, there exists x / * such that x + N1 * D E ` . Since
x + E ` , we can ¿nd a subsequence of
pn * b that cconverges to x. Hence,
n1 b we
c
can choose n 1 + M such that pn 1 / * and d pn1 * 1. Let = d pn 1 * .
Because = 0, * is a limit point of E ` , and E ` is in¿nite, there exists y / *
that is in N=4 * D E ` . Again, y + E ` leads to the existence of a subsequence
of
pn *n1 that converges to y. This allows us to choose n 2 + M such that n 2
b c = b c
n 1 and d pn 2 y . From the triangular inequality, d * pn 2 n d * y
4
b c =
d y pn 2 . We can repeat this process. In general, if we have chosen the
2
increasing ¿nite sequence n 1 n 2 n j , then there exists a u such that u / * and
=
u + Nr j * D E ` where r j j1 . Since u + E ` , u is the limit of a subsequence
2 b c
of
pn *n1 . Thus, we can ¿nd n j1 such that d pn j 1 u r j from which it
follows that
b c b c =
d * pn j 1 n d * u d u pn j 1 2r j j .
2
j k*
The method of selection of the subsequence pn j j1 ensures that it converges to
*. Therefore, * + E ` . Because * was arbitrary, we conclude that E ` contains all
of its limit points i.e., E ` is closed.
4.2. CAUCHY SEQUENCES IN METRIC SPACES 135
4.2 Cauchy Sequences in Metric Spaces

The following view of “proximity” of terms in a sequence doesn’t isolate a point to
serve as a limit.

pn * n1 be an in¿nite sequence in a metric space S d. Then
pn *
n1 is said to be a Cauchy sequence if and only if
d e
1 0 " 2M M M + MF 1m 1n n m M " d pn pm .
Another useful property of subsets of a metric space is the diameter. In this sec-
tion, the term leads to a characterization of Cauchy sequences as well as a suf¿cient
condition to ensure that the intersection of a sequence of nested compact sets will
consist of exactly one element.
De¿nition 4.2.2 Let E be a subset of a metric space X d. Then the diameter of
E, denoted by diam E is
sup
d p q : p + E F q + E .
j k
Example 4.2.3 Let A x 1 x2 + U2 : x12 x22 1 and
Q R
B x1 x2 + U2 : max
x1 x2 n 1
T
Then, in Euclidean 2-space, diam A 2 and diam B 2 2.
b c
Note that, for the sets A and B given in Example 4.2.3, diam A 2
b c T
diam A and diam B 2 2 diam B. These illustrate the observation that is
made with the next result.
b c
Lemma 4.2.4 If E is any subset of a metric space X, then diam E diam E .
Excursion 4.2.5 Use the space provided to ¿ll in a proof of the lemma. (If you get
stuck, a proof can be found on page 53 of our text.
The property of being a Cauchy sequence can be characterized nicely in terms

of the diameter of particular subsequence.
Lemma 4.2.6 If
pn *n1 is an in¿nite sequence in a metric space X and E M is the
subsequence p M p M1 p M2 , then
pn *
n1 is a Cauchy sequence if and only
if lim diam E M 0.
M*
Proof. Corresponding to the in¿nite sequence

pn * n1 in a metric space X d
let E M denote the subsequence p M p M1 p M2 .
Suppose that
pn * n1 is a Cauchy sequence. For j + M, there exists a
1
positive integer M `j M `j such that n m M `j implies that d pn pm .
j
1
Let M j M `j 1. Then, for any u ) + E M j , it follows that d u ) .
j
j k 1 b c 1
Hence, sup d u ) : u + E M j F ) + E M j n i.e., diam E M j n . Now
j j
4.2. CAUCHY SEQUENCES IN METRIC SPACES 137
1
given any 0, there exists M ) such that j M ) implies that . For
j k b c j
M max M j M ) and j M, diam E M j . Since 0 was arbitrary, we
conclude that lim diam E M 0.
M*
Suppose that lim diam E M 0 and let 0. Then there exists a
M*
positive integer K such that m K implies that diam E m i.e.,
sup
d u ) : u + E m F ) + Em
In particular, for n j m we can write n m x and j m y for some positive
integers x and y and it follows that
b c
d pn p j n sup
d u ) : u + E m F ) + Em .
Thus, we have shown that, c 0, there exists a*positive integer m such that
b for any
n j m implies that d pn p j . Therefore,
pn n1 is a Cauchy sequence.
With Corollary 3.3.44, we saw that any nested sequence of nonempty com-
pact sets has nonempty intersection. The following slight modi¿cation results from
adding the hypothesis that the diameters of the sets shrink to 0.
Theorem 4.2.7 If
K n *
n1 is a nested sequence of nonempty compact subsets of a
metric space X such that
lim diam K n 0,
n*
?
then K n consists of exactly one point.
n+M
Proof. Suppose that

K n *
n1 is a nested sequence of nonempty compact sub-
sets of a metric space X d such that lim diam K n 0. From Corollary
n*
3.3.44,
?
K n *
n1 being a nested sequence of nonempty compact subsets implies that
K n / 3.
n+M
?
If K n consists of more that one point, then there exists points x and y
n+M ? ?
in X such that x + Kn , y + K n and x / y. But this yields that
n+M n+M
0 d x y n sup
d p q : p + K n F q + K n
for all n + M i.e., diam E M d x y for any M + M. Hence, lim diam E M /
?
n*
0. Because lim diam K n 0, it follows immediately that K n consists of
n*
n+M
exactly one point.
Remark 4.2.8 To see that a Cauchy sequence in an arbitrary metric space need
not converge to a point that is in the space, consider the metric space S d where
S is the set of rational numbers and da b a b.
On the other hand, a sequence that is convergent in a metric space is Cauchy

there.
Theorem 4.2.9 Let

pn * n1 be an in¿nite sequence in a metric space S d. If
pn *
n1 converges in S, then
pn *
n1 is Cauchy.
Proof. Let
pn * n1 be an in¿nite sequence in a metric space S d that con-
verges in S to p0 . Suppose 0 is given. Then, there exists an M + M such that

n M " d pn p0 . From the triangular inequality, if n M and m M,
2
then

d pn pm n d pn p0 d p0 pm .
2 2
pn * n1 is Cauchy.
As noted by Remark 4.2.8, the converse of Theorem 4.2.9 is not true. However,
if we restrict ourselves to sequences of elements from compact subsets of a metric
space, we obtain the following partial converse. Before showing this, we will make
some us
Theorem 4.2.10 Let A be a compact subset of a metric space S d and

pn * n1
be a sequence in A. If
pn *
n1 is Cauchy, then there exists a point p0 + A such that
pn p0 as n *.
4.3. SEQUENCES IN EUCLIDEAN K -SPACE 139
Proof. Let A be a compact subset of a metric space S d and suppose

j k*that
*
pn n1 of elements in A is Cauchy. Let E M be the subsequence p M j j0 .

j k* j k*
Then E M m1 is a nested sequence of closed subsets of A and E M D A m1 is a
b c
nested sequence of compact subsets of S for which lim diam E M D A 0. By
M*
Theorem 4.2.7, there exists a unique p such that p + E M D A for all M.
Now justify that pn p as n *.
4.3 Sequences in Euclidean k-space

When we restrict ourselves to Euclidean space we get several additional results
including the equivalence of sequence convergence with being a Cauchy sequence.
The ¿rst result is the general version of the one for Euclidean n-space that we
discussed in class.
Lemma 4.3.1 On Uk d, where d denotes the Euclidean metric, let
pn x1n x2n x 3n xkn
Then the sequence
pn * n1 converges to P p1 p2 p3 pk if and only if
x jn p j for each j, 1 n j n k as sequences in U1 .
Proof. The result follows from the fact that, for each m, 1 n m n k,
Y
T X k
X; b c2 ; k n n
xmn pm xmn pm n 2 W x jn p j n nx jn p j n .
j1 j1
Suppose that 0 is given. If

pn * n1 converges to P p1 p2 p3 pk , then
there exists a positive real number M M such than n M implies that
Y
X k
X; b c2
d pn P W x jn p j .
j 1
Hence, for each m, 1 n m n k, and for all n M,
xmn pm n d pn P .
Since 0 was arbitrary, we conclude that lim xmn pm . Conversely, suppose

n*
that x jn p j for each j , 1 n j n k as sequences in U1 . Then, for each j,
1 n j n k, there exists a positive integer M j M j such that n M j implies

that xmn pm . Let M max M j . It follows that, for n M,
k 1n jnk
;k n n rs
d pn P n n n
x jn p j k .
j1
k
Because 0 was arbitrary, we have that lim pn P.

n*
Once we are restricted to the real ¿eld we can relate sequence behavior with
algebraic operations involving terms of given sequences. The following result is one
of the ones that allows us to ¿nd limits of given sequences from limits of sequences
that we know or have already proved elsewhere.
Theorem 4.3.2 Suppose that

z n * *
n1 and
?n n1 are sequences of complex num-
bers such that lim z n S and lim ?n T . Then
n* n*
(a) lim z n ?n S T
n*
(b) lim cz n cS, for any constant c

n*
(c) lim z n ?n ST
n*
t u
zn S d e
(d) lim , provided that 1n n + M " ? n / 0 F T / 0.
n* ?n T
Excursion 4.3.3 For each of the following, ¿ll in either the proof in the box on the
left of scratch work (notes) that support the proof that is given. If you get stuck,
proofs can be found on pp 49-50 of our text.
Proof. Suppose that

z n * *
n1 and
?n n1 are sequences of complex numbers
such that lim z n S and lim ?n T .
n* n*

Need look at
z n ?n S T
–Know we can make

(a) z m S for m M1
2
& ?n T for n M2
2
–Go for M max
M1 M2
and use Triangular Ineq.

Need look at
cz n cS c z n S
(b) –Know we can make

z m S for m M1
c
–for c / 0, mention c 0
—–as separate case.
Since z n S, there exists M1 + M such that n M1 Space

implies that z n S 1. Hence, z n S 1 or for
z n 1 S for all n M1 . Suppose that 0 is given. scratch
If T 0, then ?n 0 as n * implies that there exists work.

M ` + M such that ?n whenever n M ` . For
1 S
n max
M1 M ` , it follows that t u

z n ?n ST z n ?n 1 S .
1 S
(c) Thus, lim z n ?n 0.
n*
If T / 0, then ?n T as n * yields that there exists

M2 + M such that ?n whenever n M2 .
2 1 S
From z n S, there exists M3 + M such that n M3 "

z n T . Finally, for any n max
M1 M2 M3 ,
2 T
z n ?n ST
z n ?n z n T z n T ST n z n ?n T ?n z n S

1 S T .
2 1 S 2 T
Space
t u for scratch
t uwork.
zn 1
zn
(d) ?n ?n
—we can just apply
the result from (c).
The following result is a useful tool for proving the limits of given sequences in
U1 .
Lemma 4.3.4 (The Squeeze Principle) Suppose that

xn * *
n1 and
yn n1 are se-
quences of real numbers such that lim xn S and lim yn S. If
u n * n1 is a
n* n*
sequence of real numbers such that, for some positive integer K
xn n u n n yn , for all n K,
then lim u n S.
n*
Excursion 4.3.5 Fill in a proof for The Squeeze Principle.
Theorem 4.3.6 (Bolzano-Weierstrass Theorem) In Uk , every bounded sequence

contains a convergent subsequence.
Proof. Suppose that

pn *
n1 be a bounded sequence in U . Then P
k
de f
pn : n + M is bounded. Since P is a closed and bounded subset of by the Uk ,

*
Heine-Borel Theorem, P is compact. Because
pn n1 is a sequence in P a com-
pact subset of a metric space, by Theorem 4.1.19, there exists a subsequence of
pn *
n1 that is convergent in P.
Theorem 4.3.7 (Uk Completeness Theorem) In Uk , a sequence is convergent if

and only if it is a Cauchy sequence.
the Uk Completeness Theorem.
Proof. Since we are in Euclidean k-space, by Theorem , we
1
know that any sequence that is convergent in Uk is a Cauchy sequence. Conse-
quently, we only need to prove the converse.
Let
pn *
n1 be a Cauchy sequence in U . Then corresponding to 1,
k
there exists M M 1 + M such that m n M implies that

where d denotes the Euclidean metric. In particular,
2
d pn p M1 1 for all n M. Let
| }
b c
B max 1 max d p j d M1 .
1n j nM
b c
Then, for each j + M, d p j d M1 n B and we conclude that
pn *
n1 is a
sequence in U . From the
k Theorem,
3 4
*
pn : n + M is a compact subset of Uk .
Because
pn n1
is a Cauchy sequence
in a compact metric space, by Theorem 4.2.10, there exists a p0 +
pn : n + M
such that pn p0 as n *. Since
pn * n1 was arbitrary, we concluded that
.
5
***Acceptable responses are: (1) 4.2.9, (2) d pn pm 1, (3) bounded, (4) Heine-
Borel, and (5) every Cauchy sequence in Uk is convergent.***
From Theorem 4.3.7, we know that for sequences in Uk , being Cauchy is equiv-
alent to being convergent. Since the equivalence can not be claimed over arbitrary
metric spaces, the presence of that property receives a special designation.
De¿nition 4.3.9 A metric space X is said to be complete if and only if for every
sequence in X, the sequence being Cauchy is equivalent to it being convergent in
X.
Remark 4.3.10 As noted earlier, Uk is complete. In view of Theorem 4.2.10, any
compact metric space is complete. Finally because every closed subset of a metric
space contains all of its limit points and the limit of a sequence is a limit point, we
also have that every closed subset of a complete metric space is complete.
It is always nice to ¿nd other conditions that ensure convergence of a sequence

without actually having the ¿nd its limit. We know that compactness of the metric
space allows us to deduce convergence from being Cauchy. On the other hand, we
know that, in Uk , compactness is equivalent to being closed and bounded. From
the Bolzano-Weierstrass Theorem, boundedness of a sequence gives us a conver-
gent subsequence. The sequence
i n * n1 of elements in F quickly illustrates that
boundedness of a sequence is not enough to give us convergence of the whole se-
quence. The good news is that, in U1 , boundedness coupled with being increasing
or decreasing will do the job.
De¿nition 4.3.11 A sequence of real numbers

xn *
n1 is
(a) monotonically increasing if and only if 1n n + M " x n n xn1 and
(b) monotonically decreasing if and only if 1n n + M " x n o xn1 .
De¿nition 4.3.12 The class of monotonic sequences consists of all the sequences
in U1 that are either monotonically increasing or monotonically decreasing.
t u
n1 n n 1!
Example 4.3.13 For each n + M, o1 . It follows that
n n 1 n!
n! n 1!
o .
n n n 1 n 1n
| }*
n!
Consequently, is monotonically decreasing.
nn n1

xn * *
n1 is monotonic. Then
x n n1 converges if
and only if
xn *
n1 is bounded.
for the case when
xn *
n1 is monotonically decreasing.
Proof. By Lemma 4.1.10, if
xn *
n1 converges, then .
1
Now suppose that
xn * is monotonically decreasing and bounded. Let
n1
P
xn : n + M. If P is ¿nite, then there is at least one k such that xk + P and,
for in¿nitely many j + M, we have that x j xk . On the other hand we have that
xkm o xkm1 for all m + M. It follows that
xn * n1 is eventually a constant
sequence which is convergent to xk . If P is in¿nite and bounded, then from the
greatest lower bound property of the reals, we can let g inf P. Because g is the
greatest lower bound,

1n n + M " .
2
Suppose that 0 is given. Then there exists a positive integer M such that
g n x M g otherwise, .
3
Because
xn *
n1 is , the transitivity of less than or equal to yields
4
that, for all n M, g n x n g . Hence, n M " xn g . Since 0
was arbitrary, we conclude that .
5
***Acceptable responses are: (1) it is bounded, (2) g n xn , (3) g would be a

lower bound that is greater than g, (4) decreasing, and (5) lim xn g.***
n*
4.3.1 Upper and Lower Bounds

Our next de¿nition expands the limit notation to describe sequences that are tending
to in¿nity or negative in¿nity.

xn *
n1 be a sequence of real numbers. Then
(a) xn * as n * if and only if

r s
1K K + U1 " 2M M + MF 1n n M " xn o K
and
(b) xn * as n * if and only if

r s
1K K + U1 " 2M M + MF 1n n M " xn n K
In the ¿rst case, we write lim x n * and in the second case we write
n*
lim xn *.
n*
De¿nition 4.3.17 For

xn * n1 be a sequence of real numbers, let E denote the set
of all subsequential limits in the extended real number system (this means that *
and/or * are included if needed). Then the limit superior of
xn * `
n1 is x
*
sup E and the limit inferior of
xn n1 is x` inf E.
We will use lim supxn to denote the limit superior and lim infxn to denote the
n* n*
limit inferior of
xn *
n1 .
1
Example 4.3.18 For each n + M, let an 1 1n . Then the lim supan 2
2n n*
and lim infan 0.
n*
Excursion 4.3.19 Find the limit superior and the limit inferior for each of the fol-
lowing sequences.
| }
n 1n 2n 1 *
1. sn
n n1
Q H n R*
2. sn 1n1 sin
4 n1
| t u }
1 r Hn s *
3. sn 1 1 sin
n 2 n1
Q n On P R*
4. sn 1n
4 4 n1
***For (1), we have two convergent subsequences to consider s2n 3 while

s2n1 1 and you should have concluded that lim supsn 3 and lim infsn
n* n*
1. In working on (2), you should have gotten 5 subsequential
T limits: s4k
T 1,
2 2
s4k1 and
s4k3 give two subsequential limits, 1 for k even and 1 for
2 2
k odd
s4k2 also gives two subsequential limits, 2 for k odd and 0 for k even.
Comparison of the 5 subsequential limits leads to the conclusion that lim supsn
T n*
2
1 and lim infsn 2. The sequence given in (3) leads to three subsequen-
2 n*
tial limits, namely, 0 1, and 2 which leads to the conclusion that lim supsn 2
n*
and lim infsn 0. Finally, for (4), the subsequences
s4k
s4k1
s4k2 , and
n*
1 3 3 3
s4k3 give limits of 1 , and , respectively hence, lim supsn and

4 2 4 n* 2
3
lim infsn .***
n* 4
Theorem 4.3.20 Let

sn * n1 be a sequence of real numbers and E be the set of
(¿nite) subsequential limits of the sequence plus possibly * and *. Then
(a) lim sup sn + E, and

n*
tt u u
(b) 1x x lim sup sn " 2M n M " sn x .
n*
Moreover, lim sup sn is the only real number that has these two properties.
n*
of the theorem.
Proof. For the sequence of real numbers

sn *
n1 , let E denote the set of sub-
sequential limits of the sequence, adjoining * and/or * if needed, and s `
lim sup sn .
n*
Proof of part (a): If s ` *, then E is unbounded.j Thus
s *
k* n n1 is*not bounded
above and we conclude that there is a subsequence sn k k1 of
sn n1 such that
lim sn k *.
k*
If s ` *, then
sn *n1 has no ¿nite subsequential limits i.e., * is
the only element of E. It follows that lim sn *.
n*
Suppose that s ` + U. Then E is bounded above and contains at least one

element. By CN Theorem 4.1.20, the set E is . It follows from CN
1
Theorem that s` sup E + E E.
2
Proof of part (b): Suppose that there exists x + U such that x s ` and sn o x
for in¿nitely many natural numbers n. Then there exists a subsequence of
sn * n1
that converges to some real number y such that . From the trian-
3
gular inequality, y s ` . But y + E and y s ` contradicts the de¿nition of
`
. It follows that, for any x s there are at most ¿nitely many n + M
4
for which . Hence, for any x s ` there exists a positive integer M
5
such that n M implies that sn x.
Proof of uniqueness. Suppose that p and q are distinct real numbers that satisfy
property (b). Then
1x x p " 2M n M " sn x
and
1x x q " 2K n K " sn x .
Without loss of generality we can assume that p q. Then there exists * + U
such that p * q. Since * p there exists M + M such that n M implies
that sn *. In particular, at most ¿nitely many of the sk satisfy .
6
Therefore, q cannot be the limit of any subsequence of
sn *
n1 from which it follows
that q +
E i.e., q does not satisfy property (a).
4.4. SOME SPECIAL SEQUENCES 149
***Acceptable responses are: (1) closed, (2) 3.3.26, (3) y o x, (4) sup E, (5)
sn o x, and (6) q sk *.***
Remark 4.3.22 Note that, if

sn * n1 is a convergent sequence of real numbers, say
lim sn s0 then the set of subsequential limits is just
s0 and it follows that
lim sup sn lim inf sn .

n* n*
Theorem 4.3.23 If
sn * *
n1 and
tn n1 are sequences of real numbers and there
exists a positive integer M such that n M implies that sn n tn , then
lim inf sn n lim inf tn and lim sup sn n lim sup tn .

n* n* n* n*
Excursion 4.3.24 Offer a well presented justi¿cation for Theorem 4.3.23.
4.4 Some Special Sequences

This section offers some limits for sequences with which you should become famil-
iar. Space is provided so that you can ¿ll in the proofs. If you get stuck, proofs can
be found on page 58 of our text.
1
Lemma 4.4.1 For any ¿xed positive real number, lim 0.
n* n p
t u
1 1 p
Proof. For 0, let M M .

Lemma 4.4.2 For any ¿xed complex number x such that x 1, lim x n 0.
n*
Proof. If x 0, then x n 0 for each n + M and lim x n 0. Suppose that x is

n*
a ¿xed complex number such that 0 x 1. For 0, let

!
! 1 , for o 1

M M z { .
!
! ln
, for 1
ln x
The following theorem makes use of the Squeeze Principle and the Binomial
Theorem. The special case of the latter that we will use is that, for n + M and
4.4. SOME SPECIAL SEQUENCES 151
? + U
1,
n t u
; t u
n n n!
1 ?
n
? , where
k
.
k0
k k n k!k!
bn c
In particular, if ? 0 we have that 1 ? n o 1 n? and 1 ? n k ? k for
each k, 1 n k n n.
T
Theorem 4.4.3 (a) If p 0, then lim n p 1.
n*
T
(b) We have that lim n
n 1.
n*
n:
(c) If p 0 and : + U, then lim 0.
n* 1 pn
Proof of (a). We need prove the statement only for the case of p 1 the result
1
for 0 p 1 will follow by substituting in the proof of the other case. If p 1,
T p
then set xn n p 1. Then xn 0 and from the Binomial Theorem,
1 nxn n 1 xn n p
and
p1
0 xn n .
n
T
Proof of (b). Let xn n
n 1. Then x n o 0 and, from the Binomial Theorem,
n n 1 2
n 1 xn n o xn .
2
Proof of (c). Let k be a positive integer such that k :. For n 2k,

t u
n k n n 1 n 1 n k 1 k n k pk
1 p n
p p
k k! 2k k!
and
n:
0 .
1 pn
4.5 Series of Complex Numbers

For our discussion of series, we will make a slight shift is subscripting namely, it
will turn out to be more convenient for us to have our initial subscript be 0 instead
of 1. Given any sequence of complex numbers
3
ak *
k0 , we can associate (or derive)
* n
a related sequence
Sn n0 where Sn k0 ak called the sequence of nt h partial
sums. The associated sequence allows us to give precise mathematical meaning to
the idea of “¿nding an in¿nite sum.”
3*
De¿nition 4.5.1 Given a sequence of complex numbers
ak * k0 , the symbol k0 ak
is called an in¿nite series or simply a series. The symbol is intended to suggest an
in¿nite summation
a0 a1 a2 a3
3
and each an is called a term in the series. For each n + MC
0, let Sn nk0 ak
*
a30 a1 an . Then
Sn n0 is called the sequence of nth partial sums for
*
k0 ak .
4.5. SERIES OF COMPLEX NUMBERS 153
On the surface, the idea of adding an in¿nite number of numbers has no real
meaning which is why the series has been de¿ned just as a symbol. We use the
associated sequence of nth partial sums to create an interpretation for the symbol
that is tied to a mathematical operation that is well de¿ned.
3
De¿nition 4.5.2 An in¿nite series * k0 ak is said to be convergent to the complex
number S if and only if the sequence
3* of nth partial * sums
Sn *
n0 is convergent to
S when this occurs, we write k0 ak S. If
Sn n0 does not converge, we say
that the series is divergent.
Remark 4.5.3 The way that convergence of series is de¿ned, makes it clear that we
fact, given any sequence
Sn *
really aren’t being given a brand new concept. In3 n0 ,
* n
there exists a sequence
ak k0 such that Sn k1 ak for every k + MC
0: To
see this, simply choose a0 S0 and ak Sk Sk1 for k o 1. We will treat
sequences and series as separate ideas because it is convenient and useful to do so.
The remark leads us immediately to the observation that for a series to converge
it is necessary that the terms go to zero.
3*
Lemma 4.5.4 (kth term test) If the series k0 ak converges, then lim ak 0.
k*
3*
Proof. Suppose that k0 ak S. Then lim Sk S and lim Sk1 S.
k* k*
Hence, by Theorem 4.3.2(a),
lim ak lim Sk Sk1 lim Sk lim Sk1 S S 0.

k* k* k* k*
Remark 4.5.5 To see that the converse is not true, note that the harmonic series
;
*
1
k1
k
is divergent which is a consequence of the following excursion.

Excursion 4.5.6 Use the Principle of Mathematical Induction to prove that, for
3* 1 n
k1 k , S2n 1 .
2
Excursion 4.5.7 Use the de¿nition of convergence (divergence) to discuss the fol-
lowing series.
3* Hk
(a) k1 sin 4
3* 1
(b) k1 kk1
***The ¿rst example can be claimed as divergent by inspection, because the nth
term does not go to zero. The key to proving that the second one converges is
1 1 1
noticing that in fact, the given problem is an example of
kk 1 k k 1
what is known as a telescoping sum.***
The following set of lemmas are just reformulations of results that we proved
for sequences.
Lemma 4.5.8
3*(Cauchy Criteria for Series Convergence) The series (of complex
numbers) k0 ak is convergent if and only if for every 0 there exists a positive
integer M M such that 1m 1n m n M " Sm Sn >.
Proof. The lemma holds because the complex sequence of nth partial sums
Sn *
n0 is convergent if and only if it is Cauchy. This equivalence follows from the
combination of Theorem 4.2.9 and Theorem 4.3.6(b).
Remark 4.5.9 We will frequently make use of the following alternative formulation
for the sequence of nt h partial sums being Cauchy. Namely,
Sn *n0 is Cauchy if
and only if for every > 0,
n3there exists
n a positive integer M such that n M
n n p n
implies that Sn p Sn n kn1 ak n >, for p 1 2 .
3
Lemma 4.5.10 For3 the series (of complex numbers) * ak xk and
3*ak , let Re 3
k0
Im ak yk . Then * a
k0 k is convergent if and only if x
k0 k and *
k0 yk are
convergent (real) sequences.
3*
Proof. For the complex series k0 ak ,

;
n ;
n ;
n ;
n ;
n
Sn ak xk i yk xk yk .
k0 k0 k0 k0 k0
Consequently, the result is simply a statement of Lemma 4.3.1 for the case n 2.
3
Lemma3*4.5.11 Suppose that * k0 ak is a series of nonnegative real numbers.
Then k0 ak is convergent if and only if its sequence of nth partial sums is
bounded.
3
Proof. Suppose that * k0 ak is a series of nonnegative real numbers. Then
*
Sn n0 is a monotonically increasing sequence. Consequently, the result follows

from Theorem 4.3.14.
3 3*
Lemma 4.5.12 Suppose that * k0 u k and k0 ) k are convergent to U and V ,
respectively., and c is a nonzero constant. Then
3*
1. k0 u k ) k U V and
3*
2. k0 cu k cU .
Most of our preliminary discussion of series will be with series for which the
terms are positive real numbers. When not all of the terms are positive reals, we
¿rst check for absolute convergence.
3
De¿nition 4.5.13 The series * j0 a j is said to be absolutely convergent if and
3* 3 * 3*
only if j0 a j converges. If j0 a j converges and j0 a j diverges, then
3
the series *j0 a j is said to be conditionally convergent.
After the discussion of some tests for absolute convergence, we will see that
3* 1n
absolute convergence implies convergence. Also, we will justify that is
n1 n
conditionally convergent.
4.5.1 Some (Absolute) Convergence Tests

While the de¿nition may be fun to use, we would like other means to determine
convergence or divergence of a given series. This leads us to a list of tests, only a
few of which are discussed in this section.
3*
Theorem 4.5.14 (Comparison Test) Suppose that k0 ak is a series (of complex
numbers).
3* k o M " ak n ck for
3 integer M such that 1k
(a) If there exists a positive
real constants ck and * c
k0 k converges, then k0 ak converges absolutely.
k o M " ak o dk o 0
3* M such that 1k3
(b) If there exists a positive integer
for real constants dk and k0 dk diverges, then *k0 ak diverges.
3
Proof of (a). Suppose that * k0 ak is a series (of complex numbers),
3 there
exists a positive integer M such that 1k k o M " ak n ck , and * k0 k con-
c
verges. For ¿xed 0, there exists a positive integer K such that n K and
p + M implies that
n n
n n
; p n ;p
n
n n
n c n c >.
nkn1 k n kn1 k
For n M ` max
M K and any p + M, it follows from the triangular inequality
that
n n
n n
; p n ;p
n ;p
n
n n
n a kn n a k n ck >.
nkn1 n kn1 kn1
3
Since 0 was arbitrary, we conclude that * k0 ak converges.
3*
Proof of (b). Suppose that k0 ak is a series (of real numbers), there 3* ex-
ists a positive integer M such jthat 1k k o M " a o d o 0, and k0 dk
3n k* k k
diverges. From Lemma 4.5.11, k0 dk n0 is an unbounded sequence. Since
;
n ;
n
ak o dk
kM kM
j3n k*
for each n M, it follows that a is an unbounded. Therefore,
3* k0 k n0
k0 ak diverges.
In order for the Comparison Tests to be useful, we need some series about which
convergence or divergence behavior is known. The best known (or most famous)
series is the Geometric Series.
3* k
De¿nition 4.5.15 For a nonzero constant a, the series k0 ar is called a geo-
metric series. The number r is the common ratio.
Theorem 4.5.16 (Convergence Properties of the Geometric Series) For a / 0,

3 a
the geometric series * k
k0 ar converges to the sum whenever 0 r 1
1r
and diverges whenever r o 1.
Proof. The claim will follow upon showing that, for each n + MC
0,
b c
;
n
a 1 r n1
ar k .
k0
1r
The proof of the next result makes use of the “regrouping” process that was
applied to our study of the harmonic series.
j k*
Theorem 4.5.17 If a j j0 is a monotonically decreasing sequence of nonnegative
3 3* j
real numbers, then the series * j0 a j is convergent if and only if j0 2 a2 j
converges.
Excursion 4.5.18 Fill in the two blanks in order to complete the following proof of
Theorem 4.5.17.
j k*
Proof. Suppose that a j j0 is a monotonically decreasing sequence of non-
negative real numbers. For each n k + MC
0, let
;
n ;
k
Sn aj and Tk 2 j a2 j .
j0 j 0
j k*
Note that, because a j j0 is a monotonically decreasing sequence, for any j +
MC
0 and m + M,
m 1 a j o a j a j1 a jm o m 1 a jm ,

j k*
while a j j0 a sequence of nonnegative real numbers yields that
Sn and
Tk are
monotonically decreasing sequences. For n 2k ,
b c
Sn n a0 a1 a2 a3 a4 a5 a6 a7 a2k a2k 1 1
_ ^] ` _ ^] ` _ ^] `
21 ter ms 22 terms 2k ter ms
from which it follows that
1
For n 2k ,
b c
Sn o a0 a1 a2 a3 a4 a5 a6 a7 a8 a2k 1 1 a2k
_ ^] ` _ ^] ` _ ^] `
21 terms 22 ter ms 2k 1 ter ms
2
The result now follows because we have that

Sn and
Tk are simultaneously
bounded or unbounded.
***For (1), the grouping indicated leads to Sn n a1 a0 2a2 4a4 2k a2k
a1 Tk , while the second regrouping yields that Sn o a0 a1 a2 2a4 4a8
1 1
2k1 a2k a0 a1 Tk .***
2 2
As an immediate application of this theorem, we obtain a family of real series
for which convergence and divergence can be claimed by inspection.
3* 1
Theorem 4.5.19 (Convergence Properties of p-series) The series n1 con-
np
verges whenever p 1 and diverges whenever p n 1.
| Proof. }If* p n 0, the pseries diverges by the kth term test. If p 0, then
1
an p is a monotonically decreasing sequence of nonnegative real num-
n n1
bers. Note that
;
* ;
* ;* r sj
1
2 j a2 j 2 j b cp 21 p .
j0 j0 2j j0
Now use your knowledge of the geometric series to ¿nish the discussion.
A similar argument yields the following result with is offered without proof. It
is discussed on page 63 of our text.
3* 1
Lemma 4.5.20 The series converges whenever p 1 and diverges
j2
j ln j p
whenever p n 1.
Excursion 4.5.21 Discuss the convergence (or divergence) of each of the follow-
ing.
;
*
n
(a)
n1
n2 1
;*
1
(b)
n1
n3
;*
n1
(c)
n1
2n 1
;
*
3
(d)
n1
n 2 3n 1
***Notice that all of the series given in this excursion are over the positive reals
thus, checking for absolute convergence is the same as checking for convergence.
At this point, we only the n th term test, Comparison, recognition as a p-series, or
rearrangement in order to identify the given as a geometric series. For (a), noticing
n n 1
that, for each n + M, 2 o 2 allows us to claim divergence
n 1 n n n1
by comparison with the “shifted” harmonic series. The series given in (b) is con-
n1 1
vergent as a p-series for p 3. Because lim / 0 the series given
2n 1 2
in (c) diverges by the n th term test. Finally, since 3n 1 0 for each n + M,
3 3
n 2 which allows us to claim convergence of the series given in
n 2 3n 1 n
; *
3
(d) by comparison with which is convergent as a constant multiple times the
n1
n2
p-series with p 2.***
When trying to make use of the Comparison Test, it is a frequent occurrence
that we know the nature of the series with which to make to comparison almost by
inspection though the exact form of a bene¿cial comparison series requires some
creative algebraic manipulation. In the last excursion, part (a) was a mild exam-
ple of this phenomenon. A quick comparison of the degrees of the rational func-
tions that form the term suggest divergence by association with the harmonic se-
n 1
ries, but when we see that 2 we have to ¿nd some way to manipulate
n 1 n
n
the expression 2 more creatively. I chose to illustrate the “throwing more in
n 1
the denominator” argument as an alternative, note that for any natural number n,
n 1
n 2 o 1 " 2n 2 o n 2 1 " 2 o which would have justi¿ed divergence
n 1 2n
by comparison with a constant multiple of the harmonic series. We have a nice vari-
ation of the comparison test that can enable us to bypass the need for the algebraic
manipulations. We state here and leave its proof as exercise.
Theorem 4.5.22 (Limit Comparison Test) Suppose that
an * *
n0 and
bn n0 are
an
such that an o 0, bn o 0 for each n + MC
0, and lim L 0. Then either
n* bn
;
* ;
*
an and bn both converge or both diverge.
n1 n1
We have two more important and well known tests to consider at this point.
3
Theorem 4.5.23 (Ratio Test) The series * k0 ak
n n
n ak1 n
(i) converges absolutely if lim sup nn n 1
n
k* a k
(ii) diverges
n n if there exists a nonnegative integer M such that k M implies that
n ak1 n
n n
n a n o 1.
k
n n
3* n ak1 n
Proof. Suppose that the series k0 ak is such that lim sup n n n 1. It
a n
k* k
follows that we can ¿nd a positive real number
n n ; such that ; 1 and there exists
n ak1 n
an M + M such that n M implies that nn n ;. It can be shown by induction
n
n n a k
that, for each p + M and n M, nan p n ; an . In particular, for n o M 1 and
p
n n ;*
p + M C
0, nan p n ; p a M1 . Now, the series a M1 ; p is convergent as
p1
;* ;
*
a geometric series with ratio less than one. Hence, aj a M p is abso-
3*p1
jM1
lutely convergent by comparison from which it follows that k0 ak is absolutely
convergent.
3*
Suppose that the series k0 ank is such
n that there exists a nonnegative
n ak1 n
integer M for which k M implies that nn n o 1. BrieÀy justify that this yields
a n
k
divergence as a consequence of the n th term test.
n n
n ak1 n
Remark 4.5.24 Note that lim nn n 1 leads to no conclusive information con-
k* ak n
3
cerning the convergence or divergence of * k0 ak .
Example 4.5.25 Use the Ratio Test to discuss the convergence of each of the fol-
lowing:
;
*
1
1.
n1
n 1!
n n n n
n ak1 n n1 n 1
For an n n n n
n a n n n! n 1!n n 0 as n *. Hence,
1
n1! ,
n n kn n
n ak1 n n ak1 n
lim sup nn n lim n n 1 and we conclude that the series is (abso-
k* ak n k* n ak n
lutely) convergent from the ratio test.
;
* 2
n
2.
n1
2n
n n n n t u2
n2 n ak1 n n n 12 2n n
n n 1 1 1
Let an n . Then nn n n
n n 1 as
2 ak n 2n1 n2 n 2 n 2
n n n n
n ak1 n n ak1 n
n *. Thus, lim sup nn n lim n n 1 1 and we conclude
ak n k* ak n
n 2
k*
that the given series is (absolutely) convergent.
3 T
Theorem 4.5.26 (Root Test) For * k0 ak , let : lim sup ak ,
k
k*
3
(i) if 0 n : 1, then * k0 ak converges absolutely
3
(ii) if : 1, then * k0 ak diverges and
3:* 1, then no information concerning the convergence or divergence of

(iii) if
k0 ak can be claimed.
3 T
Proof. For * k0 a k , let : lim sup k
ak . If : 1, there exists a real number
k*
; such that : ; 1. Because : is a supremum of subsequential limits and
: ; 1, by Theorem 4.3.20, there exists a positive integer M such that n M
T ;
*
implies that n an ; i.e., an ; n for all n M. Since ; j is convergent
j M1
; m1 3
as a geometric series (that sums to ), we conclude that * k0 ak converges
3* 1;
that is, k0 ak converges absolutely. 3
BrieÀy justify that : 1 leads to divergence of * k0 ak as a consequence
th
of the n term test.
T
Finally, since : lim sup k ak 1 for the p-series, we see that no
k*
conclusion can be drawn concerning the convergence of divergence of the given
series.
;*
n
Example 4.5.27 Use the Root Test, to establish the convergence of n1
.
2
T n1
From Theorem 4.4.3(a) and (b), lim n 2n 1. Hence,
n*
U V t u T
k
k k 2k 1
lim sup k1 lim 2 k lim 1
k k
k* 2 k* 2 k* 2 2

from which we claim (absolute) convergence of the given series.
Thus far our examples of applications n ofn the Ration and Root
n test have led us
n ak1 n n ak1 n
to exam sequences for which lim sup nn n lim n n or lim supT k
ak
a n n
k* ak n
T k* k k*
lim k ak . This relates back to the form of the tests that you should have seen with
k*
your ¿rst exposure to series tests, probably in frosh (or AP) calculus. Of course, the
point of offering the more general statements of the tests is to allow us to study the
absolute convergence of series for which appeal to the limit superior is necessary.
The next two excursion are in the vein the parts that are described seek to help you
to develop more comfort with the objects that are examined in order to make use of
the Ratio and Root tests.
t uj
!
! 1 i
!
! , if 2 j
!
2
Excursion 4.5.28 For n + MC
0, let a j .
!
! t uj
!
! 2
!
, if 2 0 j
5
|n n}
n a j1 n *
n
1. Find the ¿rst four terms of n n .
a j n j0
QTn nR*
j n
2. Find the ¿rst four terms of ajn .
j1
|n n}
n a j1 n *
n
3. Find E 1 the set of subsequential limits of n n
a j n j0
QTn nR*
4. Find E 2 the set of subsequential limits of j na j n
j1
5. Find each of the following:

|n n}
n a j1 n *
(a) lim sup nn n
j* a j n j0
|n n}
n a j1 n *
n
(b) lim inf n n
j * a j n j0
QTn nR*
j n
(c) lim sup ajn
j* j1
Q Tn nR*
j n
(d) lim inf ajn
j * j1
;
*
6. Discuss the convergence of aj
j0
| }
2 5 16 125
***For (1), we are looking at while (2) is
T 5 4 125 32
T n n
2 2 2 2 n a j1 n
for (3), if c j nn n, then the possible subsequential
5 2 5 2 aj n
j k j k
limits are given by looking at c2 j and c2 j1 and E 1
0 * if in(4) we let
Tn n T
j k j k 2 2
d j j na j n, then consideration of d2 j and d2 j1 leads to E2
2 5
T
2 2
For (3) and (4), we conclude that the requested values are *, 0, , and , re-
2 5
spectively. For the discussion of (6), note that The Ratio Test yields no information
because neither (a) nor (b) T is satis¿ed in the other hand, from (5c), we see that
Q Tn nR* 2
lim sup j na j n 1, from which we conclude that the given series
j* j1 2
is absolutely convergent. (As an aside, examination of
S2n and
S2n1 corre-
;
*
sponding to a j even allows us to conclude that the sum of the given series is
j 0
4 10 134 168i
.)***
1 2i 21 105
t u
!
! 2 j1
!
! , if 2 j
!
3
Excursion 4.5.29 For n + MC
0, let a j .
!
! t u j1
!
! 2
!
, if 2 0 j
3
|n n}
n a j1 n *
1. Find the ¿rst four terms of nn n .
a j n j0
QTn nR*
j n
2. Find the ¿rst four terms of ajn .
j1
|n n}
n a j1 n *
n
3. Find E 1 the set of subsequential limits of n n
a j n j0
QTn nR*
4. Find E 2 the set of subsequential limits of j na j n
j1
5. Find each of the following:

|n n}
n a j1 n *
n
(a) lim sup n n
a n
j* j j0
|n n}
n a j1 n *
n
(b) lim inf n n
j * a j n j0
QTn nR*
j n
(c) lim sup ajn
j* j1
Q Tn nR*
j n
(d) lim inf ajn
j * j1
;
*
6. Discuss the convergence of aj
j1
t u t u
3 2 3 3 2 3
***Response this time are: (1) ,
2 3 2 3
t u U t u U t u54 U | }
2 32 2 2 2 23 4 2 4 2 3 4
(2) 1 3
4
(3) E 1 ,
3 3 3 3 9 3 9 3 2 9
| } Q Tn nR
2 2j n
(4) E2 where this comes from separate consideration of lim a2 j n
3 j*
Q Tn nR 3 4 2 2
2 j 1 n
and lim a2 j1 n , (5) the values are and , respectively. Finally,
j* 2 9 3 3
the Ratio Test fails to offer information concerning convergence, however, the Root
;
*
Test yields that a j is absolutely convergent. (Again, if we choose to go back to
j1
the de¿nition, examination of the nt h partial sums allows us to conclude that the
series converges to 3.)***
n n
n ak1 n 3
Remark 4.5.30 Note that, if lim sup n n n 1 for a series *
n k0 ak , the ratio
k* ak
test yields no information concerning the convergence of the series.
4.5.2 Absolute Convergence and Cauchy Products

When the terms in the generating sequence for a series are not all nonnegative reals,
we pursue the possibility of different forms of convergence.
The next result tells us that absolute convergence is a stronger condition than
convergence
j k* 3
Lemma 4.5.31 If a j j1 is a sequence of complex numbers and * j1 a j con-
3* n3 n 3
n * n *
verges, then j0 a j converges and n j0 a j n n j0 a j .
Proof. (if we were to restrict ourselves to real series) The following argument
that is a very slight variation of the one offered by the author of our text applies only
to series over the reals it jis followed
k* by a general argument that applies to
3series of
complex terms. Suppose a j j1 is a sequence of real numbers such that * j1 a j
converges and de¿ne
a j a j a j a j
)j and * j
2 2
Then ) j * j a j while ) j * j a j . Furthermore,
a j o 0 implies that ) j a j a j and * j 0
while
a j 0 implies that ) j 0 and * j a j a j
30*n ) n n a3
Consequently, n and 0 n u n n an and, from the Comparison
* 3* b Test, itc
follows that j0 ) j and j0 * j converge. By Lemma 4.5.12, j 0 ) j * j
converges. Finally, since
b c b c b c
) j *j n ) j *j n ) j *j ,
we see that
;
* b c ;
* b c ;
* b c
) j *j n ) j *j n ) j *j
j1 j 1 j1
3* nn nn 3* 3* nn nn n3 n 3 n n
n * n n n
i.e., j1 a j n j1 a j n j1 a j . Hence 0 n n j1 a j n n * j1 a j
The following proof of the lemma in general makes use of the Cauchy Criteria
for Convergence. j k*
Proof. Suppose that a j j1 is a sequence of complex numbers such that
3*
j1 a j converges and 0 is given. Then there exists a positive integer M
3
M such that 1m 1n m n M " Sm Sn > where Sm mj1 a j .
3n p n3 n
n n p n
In particular, for any p + M and n M, jn1 a j n jn1 a j n . From
n3 n 3
n n p n n p
the triangular inequality, it follows that n jn1 a j n n jn1 a j for any
3
p + M and n M. Since 0 was arbitrary, we conclude that * j1 a j converges
by the Cauchy Criteria for Convergence.
Remark 4.5.32 A re-read of the comparison, root and ratio tests reveals that they
are actually tests for absolute convergence.
Absolute convergence offers the advantage of allowing us to treat the absolutely

convergence series much as we do ¿nite sums. We have already discussed the term
by term sums and multiplying by a constant. There are two kinds of product that
come to mind: The ¿rst is the one that generalizes what we do with the distributive
law (multiplying term-by-term and collecting terms), the second just multiplies the
terms with the matching subscripts.
3* 3*
De¿nition 4.5.33 (The Cauchy Product) For j0 a j and j0 b j , set
;
k
Ck a j bk j for each k + MC
0 .
j0
3*
Then k0 C k is called the Cauchy product of the given series.
3 3
De¿nition 4.5.34 (The Hadamard Product) For * a j and * j0 b j , the se-
3* j0
ries j0 a j b j is called the Hadamard product of the given series.
The convergence of two given series does not automatically lead to the conver-
gence of the Cauchy product. The example given in our text (pp 73-74) takes
1 j
aj bj T .
j 1
3
We will see in the next section that * j0 a j converges (conditionally). On the
3 3 1
other hand, Ck kj0 a j bk j 1k kj0 T is such that
k j 1 j 1
;
k
2 2
C k o k 1
j0
k2 k2
which does not go to zero as k goes to in¿nity.

If one of the given series is absolutely convergent and the other is convergent
we have better news to report.
3 3 3
Theorem 4.5.35 (Mertens Theorem) For * a j and * b j , if (i) *j0 a j
3* j03* j0
converges absolutely, (ii) a A, and j0 b j B, then the Cauchy
3* 3* j0 j
product of j0 a j and j0 b j is convergent to AB.
3 3*
Proof. For * j0 a j and j 0 b j , let
An and
Bn be the respective se-
quences of nth partial sums. Then

;n ;k
Cn a j bn j a0 b0 a0 b1 a1 b0 a0 bn a1 bn1 an b0
k0 j0
which can be rearranged–using commutativity, associativity and the distributive

laws–to
a0 b0 b1 bn a1 b0 b1 bn1 an b0 .
Thus,
Cn a0 Bn a1 Bn1 an1 B1 an B0 .
3*
Since j0 b j B, for ;n B Bn we have that lim ;n 0. Substitution in
de f n*
the previous equation yields that
C n a0 B ;n a1 B ;n1 an1 B ;1 an B ;0
which simpli¿es to
Cn An B a0 ;n a1 ;n1 an1 ;1 an ;0 .
Let
<n a0 ;n a1 ;n1 an1 ;1 an ;0
Because lim An A, we will be done if we can show that lim <n 0. In view
n* 3 3* nn n*n
of the absolute convergence of * a j , we can set a j
n :.
j0 j0
Suppose that 0 is given. From the convergence of
;n , there exists a
positive integer M such that n M implies that ;n . For n M, it follows
that
<n a0 ;n a1 ;n1 anM1 ; M1 anM ; M an1 ;1 an ;0
3 n n
From the convergence of
;n and * n n
j0 a j , we have that
a0 ;n a1 ;n1 anM1 ; M1 :
while M being a ¿xed number and ak 0 as k * yields that
anM ; M an1 ;1 an ;0 0 as n *.
Hence, <n
a0 ;n a1 ;n1 anM1 ; M1 anM ; M an1 ;1 an ;0 implies
that lim sup <n n :. Since 0 was arbitrary, it follows that lim <n 0 as
n* n*
needed.
The last theorem in this section asserts that if the Cauchy product of two given
convergent series is known to converge and its limit must be the product of the
limits of the given series.
3 3 3
Theorem 4.5.36 If the series * 0 aj, * b j , and * j0 c j are known to con-
3* 3* j j0 3*
verge, j0 a A, b B, and j0 c j is the Cauchy product of
3* 3j * 3*j
j0
j0 a j and j0 b j , then j0 c j AB.
4.5.3 Hadamard Products and Series with Positive and Negative

Terms
3* 1
Notice that j1 can be realized as several different Hadamard prod-
j j 33
1 1 1 1
ucts letting a j , b j , c j and d j , gives us
j j 33 j j 3 j 32
3* 1 3 3*
j1 as the Hadamard product of * j 1 a j and j1 b j as well as the
j j 3 3
3* 3* 3*
Hadamard product of j1 c j and j1 d j . Note that only j1 a j diverges.
The following theorem offers a useful tool for studying the nt h partial sums for
Hadamard products.
j k*
Theorem 4.5.37 (Summation-by-Parts) Corresponding to the sequences a j j0 ,
let
;
n
An a j for n + MC
0 , and A1 0.
j0
j k*
Then for the sequence b j j0 and nonnegative integers p and q such that 0 n
p n q,
;
q ;
q1
b c
ajbj A j b j b j1 Aq bq A p1 b p
j p j p
Excursion 4.5.38 Fill in a proof for the claim.

As an immediate application of this formula, we can show that the Hadamard

product of a series whose nt h partial sums form a bounded sequence with a series
that is generated from a monotonically decreasing sequence of nonnegative terms
is convergent.
3* 3*
Theorem 4.5.39 Suppose that the series j0 a j and j0 b j are such that
Q3 R*
n
(i) j0 a j is a bounded sequence,
n0
j k*
(ii) b j j0 is a monotonically decreasing sequence of nonnegative reals, and
(iii) lim b j 0.
j*
3*
Then j0 a j b j is convergent.
3
Proof. For each n + M, let An nj0 a j . Then there exists a positive integer
j k*
M such that An n M for all n. Suppose that 0 is given. Because b j j0 is
monotonically decreasing to zero, there exists a positive integer K for which b K

. Using summation-by-parts, for any integers p and q satisfying K n q n p, it
2K
n3that
follows n n3 b c n
n q n n q1 n
n j p j j n
a b n A j b j b j1 A b
q q A p1 p n
b
n3 j p b n
cn n n n n
n q1
n n j p A j b j b j1 n n Aq bq n n A p1 b p n
3q1 nn nn b c n n n n
n A b b n Aq n bq n A p1 n b p
r
j p j
3q1 b
j j1
c s
n M b j b j1 bq b p
bb j p c c
M b p bq bq b p 2Mb p
n 2Mb K
Q3 R*
n
Since 0 was arbitrary, we conclude that j0 a j b j is a Cauchy
n0
sequence of complex numbers. Therefore, it is convergent.
A nice application of this result, gives us an “easy to check” criteria for con-
vergence of series that are generated by sequences with alternating positive and
negative terms.
Theorem 4.5.40 (Alternating Series Test) Suppose that the sequence

u j *
j1 t
U satis¿es the following conditions:
(i) sgn u j sgn u j1 for each j + MC

0, where sgn denotes “the sign
of”
(ii) u j1 n u j for every j and
(iii) lim u j 0.
j*
3
Then * j1 u j is convergent. Furthermore, if the sum is denoted by S, then
Sn n S n Sn1 for each n where
Sn *
n0 is the sequence of nth partial sums.
The result is an immediate

n nconsequence of Theorem 4.5.39 it follows upon set-
ting a j 1 and b j nc j n. As an illustration of how “a regrouping argument”
j
can get us to the conclusion, we offer the following proof for your reading pleasure.
Proof. Without loss in generality, we can take u 0 0. Then u 2k1 0 and
u 2k 0 for k 0 1 2 3 . Note that for each n + MC
0
S2n u 0 u 1 u 2 u 3 u 2n2 u 2n1 u 2n
which can be regrouped as
S2n u 0 u 1 u 2 u 3 u 4 u 2n1 u 2n .
The ¿rst arrangement justi¿es that

S2n *
n0 is monotonically increasing while the
second yields that S2n u 0 for each n. By Theorem 4.3.14, the sequence
S2n *
n0
is convergent. For lim S2n S, we have that S2n n S for each n
n*
Since S2n1 S2n u 2n , S2n1 S2n for each n + M. On the other hand,
S2n1 S2n1 u 2n u 2n1 S2n1
These inequalities combined with S2n S2 u 1 u 2 , yield that the sequence

*
S2n1 n1 is a monotonically decreasing sequence that is bounded below. Again,

by Theorem 4.3.14,
S2n1 * n1 is convergent. From (iii), we deduce that S2n1
S also. We have that S2n1 o S because
S2n1 * n1 is decreasing. Pulling this
together, leads to the conclusion that
Sn converges to S where S n Sk for k odd
and S o Sk when k is even.
Remark 4.5.41 Combining the Alternating Series Test with Remarkn4.5.5 leads to
3*
the quick observation that the alternating harmonic series n1 1
n is condition-
ally convergent.
4.5.4 Discussing Convergence

When asked to discuss the convergence3of a given series, there is a system that we
should keep in mind. Given the series *
n0 u n :
1. Check whether or not lim u n 0. If not, claim divergence by the kth term
n*
test if yes, proceed to the next step.
3 3*
2. Check for absolute convergence by testing * n0 u n . Since n0 u n is a
series having nonnegative terms, we have several tests of convergence at our
disposal–Comparison, Limit Comparison, Ratio, and Root–in addition to the
possibility of recognizing the given series as directly related to a geometric
or a p-series. 3
Practice with the tests leads to a better ability to discern which
test to
3use. If * n0 u n converges, by any of the our tests,3then we conclude
that * u
n0 n converges absolutely and we are done. If *
n0 n 3
u diverges
by either the Ratio Test or the Root Test, then we conclude that * n0 u n
diverges and we are done.
3
3. If * n0 u n diverges
3* by either the Comparison Test or the Limit Comparison
Test , then test n1 u n for conditional convergence–using the Alternating
Series Test if it applies. If the series involves nonreal complex terms, try
checking the corresponding series of real and imaginary parts.
Excursion 4.5.42 Discuss the Convergence of each of the following:
;
* 2n1
3
1.
n1
n2 1
;
*
1n n ln n
2.
n1
en
;
*
1n n
3.
n1
n2
;
*
1
4. ,: 1
n1
1 :n
;
*
cosn:
5.
n1
n2
***The ratio test leads to the divergence of the ¿rst one. The second one is abso-
lutely convergent by the root test. The third one diverges due to failure to pass the
kth term test. The behavior of the fourth one depends on :: it diverges for : 1
and converges for : 1 from the ratio test. Finally the last one converges by
comparison.***
4.5.5 Rearrangements of Series

;
*
11 ;
*
Given any series a j and a function f : MC
0 MC
0, the series a f j
jo jo
;
*
is a rearrangement of the original series. Given a series a j and a rearrange-
jo
;
*
ment a f j , the corresponding sequence of nth partial sums may be completely
jo
different. There is no reason to expect that they would have the same limit. The
commutative law that works so well for ¿nite sums tells us nothing about what may
happen with in¿nite series. It turns out that if the original series is absolutely con-
vergent, then all rearrangements are convergent to the same limit. In the last section
of Chapter 3 in our text, it is shown that the situation is shockingly different for
conditionally convergent real series. We will state the result that is proved on pages
76-77 or our text.
;
*
Theorem 4.5.43 Let a j be a real series that converges conditionally. Then for
j0
any elements in the extended real number system such that * n : n ; n *,
;
*
there exists a rearrangement of the given series a f j such that
j0
;
n ;
n
lim inf a f j : and lim sup a f j ;.
n* n*
j0 j0
;
*
Theorem 4.5.44 Let a j be a series of complex numbers that converges abso-
j0
;
*
lutely. Then every rearrangement of a j converges and each rearrangement con-
j0
verges to the same limit.
4.6 Problem Set D

1. Use the de¿nition to prove each of the following claims. Your arguments
must be well written and make use of appropriate approaches to proof.
n 2 in
(a) lim 1
n* n 2 1
3n 2 i
(b) lim 0
n* 2n 3
3n 2 3
(c) lim
n* 2n 1 2
3n 1 2ni
(d) lim 3 2i
n* n3
1 3n
(e) lim 3i
n* 1 i n
4.6. PROBLEM SET D 179
2. Find the limits, if they exist, of the following sequences in U2 . Show enough
work to justify your conclusions.
|t u}
1n cos n *
(a)
n n n1
|t u}*
3n 1 2n 3 2
(b)
4n 1 n 2 2 n1
|t u}*
1 n 5 1 3n
n 2
(c)
2n 2 1 2n n1
|t u} *
sin nn 1
(d) 2
n n n1
|t u}
cos nH sin nH H2 *
(e)
n n n1
3. Suppose that
xn * n1 converges to x in Euclidean k-space. Show that A

x n : n + M C
x is closed.
t u
Hj
2
n sin 3n
4
4. For j n + M, let f j n . Find the limit of the following
4 j 2 n 2 2 jn 1
sequence in U5 , showing enough work to carefully justify your conclusions:

f 1 n f 2 n f 3 n f 4 n f 5 n*n1 .
5. Find the limit superior and the limit inferior for each of the following se-
quences.
Q nH R*
(a) n cos
2 n1

! nH !*
1 cos
(b) 2
! n 2 !
1 n
n1
| }
1 nH nH *
(c) 1 cos
n
sin
2n 4 2 n1
| t u }*
1 n 1 1 n 1
(d) 2 1 2 3
n n1
6. If
an * *
n0 is a bounded sequence of complex numbers and
bn n0 is a se-
quence of complex numbers that converges to 0, prove that lim an bn 0.
n*
7. If
an *
n0 is a sequence of real numbers with the property that an an1 n
1
for each n + MC
0, prove that
an *
n0 converges.
2n
1
8. If
an *
n0 is a monotonically increasing sequence such that an1 an n
n
for each n + MC
0, must
an *
n0 converge? Carefully justify your response.
9. Discuss the convergence of each of the following. If the given series is con-
vergence and it is possible to ¿nd the sum, do so.
;*
1
(a) T
3
n1 n
;
*
1
(b)
n1
n n 2
;*
1
(c)
n1
2n n
;
*
2n 3
(d)
n1
n3
;
*
n
(e)
n1
en
10. Prove the Limit Comparison Test.

Suppose that
an * *
n0 and
bn n0 are such that an o 0, bn o 0 for each
;* ;*
1
n + MC
0, and lim an bn L 0. Then either an and bn both
n*
n1 n1
converge or both diverge.
1 an 3
[Hint: For suf¿ciently large n, justify that L L.]
2 bn 2
11. Suppose that an o 0 for each n + MC
0.
4.6. PROBLEM SET D 181
;
* ;
* ;
* b
T T c
(a) If an converges and bn ak , prove that bn bn1
n1 kn n1
converges.
;
* ;
n ;
* b
T T c
(b) If an diverges and Sn ak , prove that Sn1 Sn di-
n1 k1 n1
verges.
12. For each of the following use our tests for convergence to check for absolute
convergence and, when needed, conditional convergence.
;
*
2n i 3n
(a) 1n
n1
5 4n
t u
2n 1 H
;* n sin
2
(b)
n1
n 1
2
* rT
; T s
(c) 2n 2 1 2n 1
2
n1
;*
n4
(d) 1n
n1
n 1!
;* t u 2
1 n
(e) cos H n 1
n2
n
;
*
1 in3
(f)
32n1 4n
n2
t ut u t u
;*
1n 1 1 2i n 1n1 1 2 n
(g)
n1
2 5 2 3
;
* t u
1 3 5 2n 1 p
13. Justify that 1 n1
is absolutely convergent for
n1
2 4 6 2n
p 2, conditionally convergent for 0 p n 2, and divergent for p n 0.
;
*
14. Let (2 be the collection of in¿nite sequences
xn *
n1 of reals such that x n2
n1
V
;*
converges and de¿ne d x y xn yn 2 for each x
xn *
n1 y
n1

yn *
n1 + (2 . Show that (2 d is a metric space.
15. A sequence
x n *
n1 of reals is bounded if and only if there is a number m
such that xn n m for each n + M. Let M denote the collection of all
bounded sequences, and de¿ned d x y sup xn yn . Show that M d
n+M
is a metric space.
16. Let B be the collection of all absolutely convergent series and de¿ne d x y
;*
xn yn . Show that B d is a metric space.
n1
Chapter 5
Functions on Metric Spaces and
Continuity
When we studied real-valued functions of a real variable in calculus, the techniques

and theory built on properties of continuity, differentiability, and integrability. All
of these concepts are de¿ned using the precise idea of a limit. In this chapter,
we want to look at functions on metric spaces. In particular, we want to see how
mapping metric spaces to metric spaces relates to properties of subsets of the metric
spaces.
5.1 Limits of Functions

Recall the de¿nitions of limit and continuity of real-valued functions of a real vari-
able.
De¿nition 5.1.1 Suppose that f is a real-valued function of a real variable, p + U,

and there is an interval I containing p which, except possibly for p is in the domain
of f . Then the limit of f as x approaches p is L if and only if
1 0 " 2= = = 0 F 1x 0 x p = " f x L .
In this case, we write lim f x L which is read as “the limit of f of x as x

x p
approaches p is equal to L.”
De¿nition 5.1.2 Suppose that f is a real-valued function of a real variable and

p + dom f . Then f is continuous at p if and only if lim f x f p.
x p
183
184 CHAPTER 5. FUNCTIONS ON METRIC SPACES AND CONTINUITY
These are more or less the way limit of a function and continuity of a function
at a point were de¿ned at the time of your ¿rst encounter with them. With our new
terminology, we can relax some of what goes into the de¿nition of limit. Instead of
going for an interval (with possibly a point missing), we can specify that the point
p be a limit point of the domain of f and then insert that we are only looking at the
real numbers that are both in the domain of the function and in the open interval.
This leads us to the following variation.
De¿nition 5.1.3 Suppose that f is a real-valued function of a real variable,

dom f A, and p + A) (i.e., p is a limit point of the domain of f ). Then the
limit of f as x approaches p is L if and only if
1 0 " 2= =d 0 e
1x x + A F 0 x p = " f x L
b c
Example 5.1.4 Use the de¿nition to prove that lim 2x 2 4x 1 31.
x3
Before we offer a proof, we’ll illustrate some “expanded ”scratch work that
leads to the information needed in order to offer a proof. We want to show that,
corresponding
nb 2 c eachn
to 0 we can ¿nd a = 0 such that 0 x 3 = "
n 2x 4x 1 31n . The easiest way to do this is to come up with a = that
is a function of . Note that
nr s n n n
n n n n
n 2x 2 4x 1 31n n2x 2 4x 30n 2 x 3 x 5 .
The x 3 is good news because it is ours to make as small as we choose. But if

we restrict x 3 there is a corresponding restriction on x 5 to take care of
this part we will put a cap on = which will lead to simpler expressions. Suppose that
we place a 1st restriction on = of requiring that = n 1. If = n 1, then 0 x 3
= n 1 " x 5 x 3 8 n x 3 8 9. Now
nr s n
n n
n 2x 4x 1 31n 2 x 3 x 5 2 = 9 n
2
Q R
whenever = n . To get both bounds to be in effect we will take = max 1 .
18 18
This concludes that “expanded ”scratch
Q work.
R
Proof. For 0, let = max 1 . Then
18
0 x 3 = n 1 " x 5 x 3 8 n x 3 8 9
5.1. LIMITS OF FUNCTIONS 185
and
nr s n
n n
n 2x 4x 1 31n 2 x 3 x 5 2 = 9 n 18
2
.
18
SinceQ 0 was arbitrary, we conclude that, for every 0, there exists a =
R nb 2 c n
min 1 0, such that 0 x 3 = " n 2x 4x 1 31n i.e.,
b 18 c
lim 2x 2 4x 1 31.
x3
b c
Excursion 5.1.5 Use the de¿nition to prove that lim x 2 5x 6.
x1
A Proof.
***For this one, the = that you de¿ne will depend on the nature of the ¿rst restriction
that you placed on = in order to obtain a niceQupperR bound on x 6 if you chose

= n 1 as your ¿rst restriction, then = min 1 would have been what worked
8
in the proof that was offered.***
You want to be careful not to blindly take = n 1 as the ¿rst restriction. For
1
example, if you are looking at the greatest integer function as x , you would
2
1
need to make sure that = never exceeded in order to stay away from the nearest
2
1
“jumps” if you have a rational function for which is a zero of the denominator
2
1 1
and you are looking at the limit as x , then you couldn’t let = be as great at
4 4
1
so you might try taking = n as a ¿rst restriction. Our next example takes such a
6
consideration into account.
x2 3
Example 5.1.6 Use the de¿nition to prove that lim 4.
x1 2x 1
| }
1 2 1
Proof. For 0, let = min . From 0 x 1 = n , we have
4 25 4
that
1 25
x 7 x 1 6 x 1 6 6
4 4
and
n n n n n n t u
n 1 n n 1 nn n 1 nn 1 1
2x 1 2 nnx nn 2 nnx 1 o 2 nx 1 2 .
2 2n n 2n 4 2
Furthermore,
u t
25
nt 2 u n n 2 n =
n x 3 n n x 8x 7 n x 1 x 7 4
n n n n
n 2x 1 4n n 2x 1 n 2x 1

1

2
25 = 25 2
n .
2 2 25
| }
1 2
Since 0 was arbitrary, for every 0 there exists a = min
nt 2 u n 4 25
n x 3 n
0 such that 0 x 1 = implies that nn 4nn that is,
2x 1
x2 3
lim 4.
x1 2x 1
In Euclidean U space, the metric is realized as the absolute value of the differ-
ence. Letting d denote this metric allows us to restate the de¿nition of lim f x
x p
L as
1 0 " 2= = 0

d e
1x x + A F 0 d x p = " d f x L .
Of course, at this point we haven’t gained much this form doesn’t look particu-
larly better than the one with which we started. On the other hand, it gets us nearer
to where we want to go which is to the limit of a function that is from one metric
space to another–neither of which is U1 . As a ¿rst step, let’s look at the de¿nition
when the function is from an arbitrary metric space into U1 . Again we let d denote
the Euclidean 1metric.
De¿nition 5.1.7 Suppose that A is a subset of a metric space S d S and that f is
a function with domain A and range contained in U1 i.e., f : A U1 . then “ f
tends to L as x tends to p through points of A” if and only if
(i) p is a limit point of A, and
(ii) 1 0 2= = 0 1x x + A F 0 d S x p =

" d f x L .
In this case, we write f x L as x p for x + A, or f x L as
x p, or
A
lim f x L
x p
x+A
Example 5.1.8 Let f : F U be given by f z Re z. Prove that
lim f z 3
z3i
z +F

For this one, we will make use of the fact that for any complex number ? ,
Re ? n ? .
Proof. For 0, let = . Then 0 z 3 i = implies that
f z 3 Re z 3 Re z 3 i n z 3 i .
Since 0 was arbitrary, we conclude that lim f z 3.

z3i
z +F
Remark 5.1.9 Notice that, in the de¿nition of lim f x L, there is neither a
x p
x+A
requirement that f be de¿ned at p nor an expectation that p be an element of A.
Also, while it isn’t indicated, the = 0 that is sought may be dependent on p.
Finally we want to make the transition to functions from one arbitrary metric
space to another.
De¿nition 5.1.10 Suppose that A is a subset of a metric space S d S and that f
is a function with domain A and range contained in a metric space X d X i.e.,
f : A X. Then “ f tends to L as x tends to p through points of A” if and only
if
(i) p is a limit point of A, and
(ii) 1 0 2= = 0 1x x + A F 0 d S x p =

" d X f x L .
In this case, we write f x L as x p for x + A, or f x L as
x p, or
A
lim f x L
x p
x+A
Example 5.1.11 For p + U1 , let f p 2 p 1, p 2 . Then f : U1 U2 . Use

the de¿nition of limit to from that lim f p 3 1 with respect to the Euclidean
p1
metrics on each space.
| }

Proof. For 0, let = min 1 T . Then 0 dU p 1 p 1
13
= n 1 implies that
p 1 p 1 2 n p 1 2 3
and
T T T
4 p 12 4 9 13.
Hence, for 0 dU p 1 p 1 =,
T
b c2
dU2 f p 3 1 2 p 1 32 p2 1
T T
p 1 4 p 12 = 13 n .
Since 0 was arbitrary, we conclude that lim 2 p 1, p 2 3 1.

p1
Remark 5.1.12 With few exceptions our limit theorems for functions of real-valued
functions of a real variable that involved basic combinations of functions have di-
rect, straightforward analogs to functions on an arbitrary metric spaces. Things
can get more dif¿cult when we try for generalizations of results that involved com-
paring function values. For the next couple of excursions, you are just being asked
to practice translating results from one setting to our new one.
Excursion 5.1.13 Let A be a subset of a metric space S and suppose that f : A

U1 is given. If
f L as p p0 in A and f M as p p0 in A
prove that L M. After reading the following proof for the case of real-valued
functions of a real variable, use the space provided to write a proof for the new
setting.
Proof. We want to prove that, if f L as x a and f M as x a, then

L M. For L / M, let > 12 L M. By the de¿nition of limit, there exists
positive numbers =1 and =2 such that 0 x a =1 implies f x L >
and 0 x a =2 implies f x M >. Choose x0 + U such that
0 x0 a min
=1 =2 . Then L M n L f x0 M f x0 2>
which contradicts the trichotomy law.
Excursion 5.1.14 Let f and g be real-valued functions with domain A, a subset of

a metric space S d. If lim f p L and lim g p M, then
p p0 p p0
p+ A p+ A
lim f g p L M.
p p0
p+ A
After reading the following proof for the case of real-valued functions of a real
variable, use the space provided to write a proof for the new setting.
Proof. We want to show that, if lim f x L and lim gx M, then
xa xa
lim f gx L M. Let > 0 be given. Then there exists positive numbers =1
xa
and =2 such that 0 x a =1 implies f x L >2 and 0 x a =2
implies gx M >2. For = min
=1 =2 , 0 x a = implies that
f gx L M n f x L gx M >.
Theorem 4.1.17 gave us a characterization of limit points in terms of limits of

sequences. This leads nicely to a characterization of limits of functions in terms of
behavior on convergent sequences.
Theorem 5.1.15 (Sequences Characterization for Limits of Functions) Suppose

that X d X and Y dY are metric spaces, E t X, f : E Y and p is a limit
point of E. Then lim f x q if and only if
x p
x +E
Kr s L
1
pn
pn t E F 1n pn / p F lim pn p " lim f pn q .
n* n*
of the theorem.
Proof. Let X, Y , E, f , and p be as described in the introduction to Theorem

5.1.15. Suppose that lim f x q. Since p is a limit point of E, by Theorem
x p
x+E
, there exists a sequence
pn of elements in E such that pn / p for all
1
n + M, and . For 0, because lim f x q, there
x p
2 x+E
exists = 0 such that 0 d X x p = and x + E implies that .
3
From lim pn p and pn / p, we also know that there exists a positive integer M
n*
such that n M implies that . Thus, it follows that
4
dY f pn q for all n M. Since 0 was arbitrary, we conclude that

lim f pn q. Finally, because
pn t E was arbitrary, we have that
n*

1
pn " lim f pn q .
n*
5
We will give a proof by contrapositive of the converse. Suppose that lim f x / q.
x p
x+E
Then there exists a positive real number such that corresponding to each posi-
tive real number = there is a point x= + E for which 0 d X x= p = and
1
dY f x= q o . In particular, for each n + M, corresponding to there is a point
n
pn + E such that and dY f pn q o . Hence, lim f pn / q.
n*
6
Thus, there exists a sequence
pn t E such that lim pn p and
n*
7
i.e.,
Kr s L
2
pn
pn t E F 1n pn / p F lim pn p F lim f pn / q
n* n*
Kr s L
1
pn
n* n*
Therefore, we have shown that lim f x / q implies that

x p
x +E
Kr s L
1
pn
n* n*
Since the is logically equivalent to

8
the converse, this concludes the proof.
***Acceptable responses are: (1) 4.1.17,(2) lim pn p, (3) dY f x q ,
n*
(4) 0 d X pn p =, (5)
pn t E F 1n pn / p F lim pn p, (6)
n*
1
0 d X pn p , (7) lim f pn / q, (8) contrapositive.***
n n*
The following result is an immediate consequence of the theorem and Lemma
4.1.7.
Corollary 5.1.17 Limits of functions on metric spaces are unique.

Remark 5.1.18 In view of Theorem 5.1.15, functions from metric spaces into sub-
sets of the complex numbers will satisfy the “limits of combinations” properties of
sequences of complex numbers that were given in Theorem 4.3.2. For completeness,
we state it as a separate theorem.
Theorem 5.1.19 Suppose that X d X is a metric space, E t X, p is a limit point
of E, f : E F, g : E F, lim f x A, and lim gx B. Then
x p x p
x +E x+E
(a) lim f g x A B

x p
x +E
(b) lim f g x AB

x p
x +E
f A
(c) lim x whenever B / 0.
x p g B
x +E
While these statements are an immediate consequence of Theorem 4.3.2 and

Theorem 5.1.15 completing the following excursions can help you to learn the ap-
proaches to proof. Each proof offered is independent of Theorems 4.3.2 and Theo-
rem 5.1.15.
Excursion 5.1.20 Fill in what is missing to complete a proof of Theorem 5.1.19(a).
Proof. Suppose 0 is given. Because lim f x A, there exists a positive
x p
x +E

real =1 such that x + E and 0 d X x p =1 implies that f x A .
2
Since , there exists a positive real number =2 such that x + E and
1

0 d X x p =2 implies that g x B . Let = . It
2 2
follows from the triangular inequality that, if x + E and 0 d X x p =, then
n n
n n
n n
f g x A B n f x A n
n 3
n
n .
4 5
Since 0 was arbitrary, we conclude that lim f g x A B as claimed.

x p
x +E
***Acceptable responses are: (1) lim gx B, (2) min

=1 =2 , (3) g x B,
x p
x +E
(4) f x A g x B, (5) ***
Excursion 5.1.21 Fill in what is missing to complete a proof of Theorem 5.1.19(b).
Proof. Because lim f x A, there exist a positive real number =1 such that
x p
x+E
x + E and 0 d X x p =1 implies that f x A 1 i.e., f xA 1.
Hence, f x 1 A for all x + E such that 0 d X x p =1 .
Suppose that 0 is given. If B 0, then lim gx 0 yields the
x p
x+E
existence of a positive real number =2 such that x + E and 0 d X x p =2
implies that

g x .
1 A
Then for = ` , we have that

1
f g x f x g x 1 A .

2
Hence, lim f g x AB 0. Next we suppose that B / 0. Then there exists a

x p
x+E

positive real numbers =3 and =4 for which f x A and g x B
2 B

whenever 0 d X x p =3 and 0 d X x p =4 , respectively, for
2 1 A
x + E. Now let = min
=1 =3 =4 . It follows that if x + E and 0 d X x p
=, then
n n
n n
n n
f g x AB n f x g x AB n
n 3 4
n
n f x g x B
5
n n
n n
n n
1 A g x B B n n
n 6
n
.
7 8
Again, since 0 was arbitrary, we conclude that lim f g x AB as

x p
x +E
needed.
***Acceptable responses include: (1) min
=1 =2 , (2) 1 A1 , (3) f x B,

(4) f x B, (5) B f x A, (6) f x A, (7) 1 A B
2 1 A 2 B
(8) .***
Excursion 5.1.22 Fill in what is missing to complete a proof of Theorem 5.1.19(c).
Proof. In view of Theorem 5.1.19(b), it will suf¿ce to prove that, under the
1 1
given hypotheses, lim . First, we will show that, for B / 0, the modulus
x p g x B
x+E
of g is bounded away from zero. Since B 0 and lim gx B, there exists
x p
x+E
a positive real number =1 0 such that x + E and implies that
1
B
g x B . It follows from the (other) that,
2 2
if x + E and 0 d X x p =1 , then
n n
n n
g x g x B B o ng x B n
B
.
2
B2
Suppose that 0 is given. Then 0 and lim gx B yields the
2 x p
x+E
B2
existence of a positive real number =2 such that g x B whenever
2
x + E and 0 d X x p =2 . Let = min
=1 =2 . Then for x + E and
0 d X x p = we have that
n n
n 1 1 nn g x B
n
n g x B n B g x .
3 4
Since 0 was arbitrary, we conclude that .

5
1
Finally, letting h x , by Theorem ,
g x 6
f
lim x lim f x .
x p g x p
x+E x+E 7 8
***Acceptable responses are: (1) 0 d X x p =1 , (2) triangular inequality (3)

B2 1 1 1
t u , (4) (5) lim , (6) 5.1.19(b), (7) h x, (8) A .***
B x p g x B B
2 B x+E
2
From Lemma 4.3.1, it follows that the limit of the sum and the limit of the prod-
uct parts of Theorem 5.1.19 carry over to the sum and inner product of functions
from metric spaces to Euclidean k space.
Theorem 5.1.23 Suppose that X is a metric space, E t X, p is a limit point of E,

f : E Uk , g : E Uk , lim f x A, and lim gx B. Then
x p x p
x+E x+E
(a) lim f g x A B and

x p
x +E
(b) lim f g x A B

x p
x +E
In the set-up of Theorem 5.1.23, note that fg :E Uk while fg : E U.

5.2. CONTINUOUS FUNCTIONS ON METRIC SPACES 197
5.2 Continuous Functions on Metric Spaces

Recall that in the case of real-valued functions of a real variable getting from the
general idea of a functions having limits to being continuous simply added the prop-
erty that the values approached are actually the values that are achieved. There is
nothing about that transition that was tied to the properties of the reals. Conse-
quently, the de¿nition of continuous functions on arbitrary metric spaces should
come as no surprise. On the other hand, an extra adjustment is needed to allow
for the fact that we can consider functions de¿ned at isolated points of subsets of
metric spaces.
De¿nition 5.2.1 Suppose that X d X and Y dY are metric spaces, E t X, f :
E Y and p + E. Then f is continuous at p if and only if
d e
1 0 2= = 0 1x x + E F d X x p = " dY f x f p .
Theorem 5.2.2 Suppose that X d X and Y dY are metric spaces, E t X, f :
E Y and p + E and p is a limit point of E. Then f is continuous at p if and
only if lim f x f p
x p
x+E
De¿nition 5.2.3 Suppose that X d X and Y dY are metric spaces, E t X and
f : E Y . Then f is continuous on E if and only if f is continuous at each
p + E.
Remark 5.2.4 The property that was added in order to get the characterization
that is given in Theorem 5.2.2 was the need for the point to be a limit point. The
de¿nition of continuity at a point is satis¿ed for isolated points of E because each
isolated point p has the property that there is a neighborhood of p, N=` p, for
which E DN= ` p
p since p + dom f and d X p p dY f p f p
0, we automatically have that 1x x + E F d X x p = " dY f x f p
for any 0 and any positive real number = such that = = ` .
Remark 5.2.5 It follows immediately from our limit theorems concerning the al-
gebraic manipulations of functions for which the limits exist, the all real-valued
polynomials in k real variables are continuous in Uk .
bb cc
Remark 5.2.6 Because f 1 3 1 for the f p 2 p 1 p2 : U1
U2 that was given in Example 5.1.11, our work for the example allows us to claim
that f is continuous at p 1.
Theorem 5.1.15 is not practical for use to show that a speci¿c function is contin-
uous it is a useful tool for proving some general results about continuous functions
on metric spaces and can be a nice way to show that a given function is not contin-
uous.
Example 5.2.7 Prove that the function f : U U U given by f x y

xy
!
3 , for x y / 0 0
x y3
is not continuous at 0 0.
!

0 t, forux y 0
1 1
Let pn . Then
pn *
n1 converges to 0 0, but
n n
t ut u
1 1
n n n
lim f pn lim t u3 t u3 lim * / 0.
n* n* 1 1 n* 2

n n
Hence, by the Sequences Characterization for Limits of Functions, we conclude

that the given f is not continuous at 0 0.
Example 5.2.8 Use2 the de¿nition to prove that f : U U U given by

!
! x y
2 , for x y / 0 0
f x y x y 2
is continuous at 0 0.
!
!

0 , for x y 0
We need to show that lim f x y 0. Because the function is de¿ned
xy00
in two parts, it is necessary to appeal to the de¿nition. For 0, let = . Then
T
0 dUU x y 0 0 x 2 y 2 =
implies that
n 2 n b 2 c T T
n x y n x y 2 y
n
f x y 0 n 2 n n y y n x 2 y 2 .
2
x y2 n x 2 y2
Because 0 was arbitrary, we conclude that

lim f x y 0 f 0 0. Hence, f is continuous at 0 0.
xy00
It follows from the de¿nition and Theorem 5.1.19 that continuity is transmitted
to sums, products, and quotients when the ranges of our functions are subsets of
the complex ¿eld. For completeness, the general result is stated in the following
theorem.
Theorem 5.2.9 If f and g are complex valued functions that are continuous on
f
a metric space X, then f g and f g are continuous on X. Furthermore, is
g
continuous on X
p + X : g p 0.
From Lemma 4.3.1, it follows immediately that functions from arbitrary metric
spaces to Euclidean k-space are continuous if and only if they are continuous by
coordinate. Furthermore, Theorem 5.1.23 tells us that continuity is transmitted to
sums and inner products.
Theorem 5.2.10 (a) Let f 1 f 2 f k be real valued functions on a metric space
X, and F : X Uk be de¿ned by F x f 1 x f 2 x f k x.
Then F is continuous if and only if f j is continuous for each j 1 n j n k.
(b) If f and g are continuous functions from a metric space X into Uk , then f g
and f g are continuous on X.
The other combination of functions that we wish to examine on arbitrary metric
spaces is that of composition. If X, Y , and Z are metric spaces, E t X, f : E
Y , and g : f E Z , then the composition of f and g, denoted by g i f , is
de¿ned by g f x for each x + E. The following theorem tells us that continuity
is transmitted through composition.
Theorem 5.2.11 Suppose that X, Y , and Z are metric spaces, E t X, f : E
Y , and g : f E Z. If f is continuous at p + E and g is continuous at f p,
then the composition g i f is continuous at p + E.
Proof. Suppose that X d X , Y dY , and Z d Z are metric spaces, E t X,

f : E Y , g : f E Z , f is continuous at p + E, and g is contin-
uous at f p. Let 0 be given. Since g is continuous at f p, there exists
a positive real number =1 such that d Z g y g f p for any y + f E
such that dY y f p =1 . From f being continuous at p + E and =1 be-
ing a positive real number, we deduce the existence of another positive real num-
ber = such that x + E and d X x p = implies that dY f x f p =1 .
Substituting f x for y, we have that x + E and d X x p = implies that
dY f x f p =1 which further implies that d Z g f x g f p .
That is, d Z g i f x g i f p for any x + E for which d X x p =.
Therefore, g i f is continuous at p.
Remark 5.2.12 The “with respect to a set” distinction can be an important one to
1 , for x rational
note. For example, the function f x is continuous

0 , for x irrational
with respect to the rationals and it is continuous with respect to the irrationals.
However, it is not continuous on U1 .
5.2.1 A Characterization of Continuity

Because continuity is de¿ned in terms of proximity, it can be helpful to rewrite the
de¿nition in terms of neighborhoods. Recall that, for X d X , p + X, and = 0,
N= p
x + X : d X x p = .
For a metric space Y dY , f : X Y and 0,
N f p
y + Y : dY y f p .
Hence, for metric spaces X d X and Y dY , E t X, f : E Y and p + E, f
is continuous at p if and only if
d e
1 0 2= = 0 f N= p D E t N f p .
Because neighborhoods are used to de¿ne open sets, the neighborhood formu-
lation for the de¿nition of continuity of a function points us in the direction of the
following theorem.
Theorem 5.2.13 (Open Set Characterization of Continuous Functions) Let f be

a mapping on a metric space X d X into a metric space Y dY . Then f is con-
tinuous on X if and only if for every open set V in Y , the set f 1 V is open in
X.
of the theorem.
Proof. Let f be a mapping from a metric space X d X into a metric space
Y dY .
Suppose that f is continuous on X, V is an open set in Y , and p0 +
1
f V . Since V is open and f p0 + V , we can choose 0 such that
N f p0 t V from which it follows that
t f 1 V .
1
Because f is continuous at p0 + X, corresponding to 0, there exists a = 0,

such that f N= p0 t N f p0 which implies that
N= p0 t .
2
From the transitivity of subset, we concluded that N= p0 t f 1 V . Hence,

p0 is an interior point of f 1 V . Since p0 was arbitrary, we conclude that each
p + f 1 V is an interior point. Therefore, f 1 V is open.
To prove the converse, suppose that the inverse image of every open set
in Y is open in X. Let p be an element in X and 0 be given. Now the
neighborhood N f p is open in Y . Consequently, is open in X.
3
Since p is an element of , there exists a positive real number = such

3
that N= p t f 1 N f p i.e., t N f p. Since 0
4
was arbitrary, we conclude that lim f x . Finally, because p was an
x p
5
arbitrary point in X, it follows that f as needed.
6
***Acceptable responses are: (1) f 1 N f p0 , (2) f 1 N f p0 ,
(3) f 1 N f p, (4) f N= p, (5) f p, (6) is continuous on X.***
Excursion 5.2.15 Suppose that f is a mapping on d ac emetric

b 1spacecX d X into a
1 c
metric space Y dY and E t X. Prove that f E f [E] .
The following corollary follows immediately from the Open Set Characteriza-
tion for Continuity, Excursion 5.2.15, and the fact that a set is closed if and only
if its complement is open. Use the space provided after the statement to convince
yourself of the truth of the given statement.
Corollary 5.2.16 A mapping f of a metric space X into a metric space Y is con-

tinuous if and only if f 1 C is closed in X for every closed set C in Y .
Remark 5.2.17 We have stated results in terms of open sets in the full metric space.
We could also discuss functions restricted to subsets of metric spaces and then the
characterization would be in terms of relative openness. Recall that given two sets
X and Y and f : X Y , the corresponding set induced functions satisfy the
following properties for C j t X and D j t Y , j 1 2:
f 1 [D1 D D2 ] f 1 [D1 ] D f 1 [D2 ],
f 1 [D1 C D2 ] f 1 [D1 ] C f 1 [D2 ],
f [C1 D C2 ] t f [C1 ] D f [C2 ], and
f [C1 C C2 ] f [C1 ] C f [C2 ]
Because subsets being open to subsets of metric spaces in characterized by their

realization as intersections with open subsets of the parent metric spaces, our neigh-
borhoods characterization tells us that we loose nothing by looking at restrictions
of given functions to the subsets that we wish to consider rather that stating things
in terms of relative openness.
5.2.2 Continuity and Compactness

Theorem 5.2.18 If f is a continuous function from a compact metric space X to a
metric space Y , then f X is compact.
Excursion 5.2.19 Fill in what is missing to complete the following proof of Theo-
rem 5.2.18.

Proof. Suppose that f is a continuous function from a compact metric space X
to a metric space Y and J
G : : : + is an open cover for f X. Then
G : is open in Y for each : + and .
1
From the Open Set Characterization of
2
Functions, f 1 G : is for each : + .
3 6
Since f : X f X and f X t G : , we have
:+
that

>
X f 1 f X t f 1 G: .
:+ 4
j k
Hence, I f 1 G : : : + is an
for X. Since X is
5
, there is a ¿nite subcollection of I,
j b6 c k
f 1 G : j : j 1 2 n , that covers X i.e.,
>
n b c
Xt f 1 G : j .
j1
It follows that

>n b c >
n >
n
1
f X t f f G: j G: j .
j 1 7
j j1
k j1
Therefore, G : j : j 1 2 n is a ¿nite
subcollection of J that covers f X. Since J was
arbitrary, every
i.e.,
8
f X is .
9
Remark 5.2.20 Just to stress the point, in view of our de¿nition of relative com-
pactness the result just stated is also telling us that the continuous image of any
compact subset of a metric space is a compact subset in the image.
De¿nition 5.2.21 For a set E, a function f : E Uk is said to be bounded if

and only if
2M M + UF 1x x + E " f x n M .
When we add compactness to domain in the metric space, we get some nice
analogs.
Theorem 5.2.22 (Boundedness Theorem) Let A be a compact subset of a metric

space S d and suppose that f : A Uk is continuous. Then f A is closed
and bounded. In particular, f is bounded.
Excursion 5.2.23 Fill in the blanks to complete the following proof of the Bound-
edness Theorem.
Proof. By the , we know that compactness in Uk
1
for any k + M is equivalent to being closed and bounded. Hence, from Theorem
5.2.18, if f : A Uk where A is a compact metric space, then f A is compact.
But f A t and compact yields that f A is .
2 3
In particular, f A is bounded as claimed in the Boundedness Theorem.
***Expected responses are: (1) Heine-Borel Theorem, (2) Uk , and (3) closed and
bounded.***
Theorem 5.2.24 (Extreme Value Theorem) Suppose that f is a continuous func-

tion from a compact subset A of a metric space S into U1 ,
M sup f p and m inf f p .

p+A p+A
Then there exist points u and ) in A such that f u M and f ) m.
Proof. From Theorem 5.2.18 and the Heine-Borel Theorem, f A t U and f
continuous implies that f A is closed and bounded. The Least Upper and Greatest
Lower Bound Properties for the reals yields the existence of ¿nite real numbers M
and m such that M sup f p and m inf f p. Since f A is closed, by
p+A p+A
Theorem 3.3.26, M + f A and m + f A. Hence, there exists u and ) in A such
that f u M and f ) m i.e., f u sup f p and f ) inf f p.
p+A p+A
Theorem 5.2.25 Suppose that f is a continuous one-to-one mapping of a compact

metric space X onto a metric space Y . Then the inverse mapping f 1 which is
de¿ned by f 1 f x x for all x + X is a continuous mapping that is a one-to-
one correspondence from Y to X.
Proof. Suppose that f is a continuous one-to-one mapping of a compact metric
space X onto a metric space Y . Because f is one-to-one, the inverse f 1 is a
function from rng f Y in X. From the Open Set Characterization of Continuous
Functions, we know that f 1 is continuous in Y if f U is open in Y for every U
that is open in X. Suppose that U t X is open. Then, by Theorem 3.3.37, U c is
compact as a closed subset of the compact metric space X. In view of Theorem
5.2.18, f U c is compact. Since every compact subset of a metric space is closed

(Theorem 3.3.35), we conclude that f U c is closed. Because f is one-to-one,
f U c f X f U then f onto yields that f U c Y f U f U c .
Therefore, f U c is closed which is equivalent to f U being open. Since U was
arbitrary, for every U open in X, we have that f U is open in Y . Hence, f 1 is
continuous in Y .
5.2.3 Continuity and Connectedness

Theorem 5.2.26 Suppose that f is a continuous mapping for a metric space X
into a metric space Y and E t X. If E is a connected subset of X, then f E is
connected in Y .
of Theorem 5.2.26.
Proof. Suppose that f is a continuous mapping from a metric space X into a
metric space Y and E t X is such that f E is not connected. Then
we can let f E A C B where A and B are nonempty
subsets of Y i.e., A / 3, B / 3 and
1
A D B A D B 3. Consider G E D f 1 A and
de f
H E D f 1 B. Then neither G nor H is empty and
de f
b c
GCH E D f 1 A C
2
ED
3
E D f 1 A C B .
4
b c
Because A t A, f 1 A t f 1 A . Since
G t f 1 A, the transitivity of containment yields that
. From the Corollary to the Open Set
5
b c
Characterization for Continuous Functions, f 1 A is
.
6
b c b c
It follows that G t f 1 A . From G t f 1 A and
H t f 1 B, we have that

b c
G D H t f 1 A D f 1 B f 1 f 1 .
7 8 9
The same argument yields that G D H 3.

From E G C H, G / 3, H / 3 and
G D H G D H 3, we conclude that E is
. Hence, for f a continuous mapping
10
from a metric space X into a metric space Y and E t X,
if f E is not connected, then E is not connected.
According to the contrapositive, we conclude that, if
, then , as needed.
11 12
***Acceptable responses
b c are: (1) separated (2) E D f 1 B, (3) f 1 AC f 1 B,
(4) E, (5) G t f 1 A , (6) closed, (7) A D B, (8) 3, (9) 3, (10) not connected,
(11) E is connected, and (12) f E is connected.***
Theorem 5.2.28 Suppose that f is a real-valued function on a metric space X d.
If f is continuous on S, a nonempty connected subset of X, then the range of f S ,
denoted by R f S , is either an interval or a point.
Theorem 5.2.29 (The Intermediate Value Theorem) Let f be a continuous real-

valued function on an interval [a b]. If f a f b and if c + f a f b,
then there exist a point x + a b such that f x c.
Proof. Let E f [a b]. Because [a b] is an interval, from Theorem 3.3.60,
we know that [a b] is connected. By Theorem 5.2.26, E is also connected as the
continuous image of a connected set. Since f a and f b are in E, from Theorem
3.3.60, it follows that if c is a real number satisfying f a c f b, then c is in
E. Hence, there exists x in [a b] such that f x c. Since f a is not equal to c
and f b is not equal to c, we conclude that x is in a b. Therefore, there exists x
in a b such that f x c, as claimed.
5.3 Uniform Continuity

Our de¿nition of continuity works from continuity at a point. Consequently, point
dependency is tied to our = proofs of limits. For example, if we carried out a
2x 1
= proof that f x is continuous at x 2, corresponding to 0,
| } x 1
1 2x 1
taking = min will work nicely to show that lim 5 f 2
2 6 x2 x 1
3
however, it would not work for showing continuity at x . On the other hand,
| } 2
1
corresponding to 0, taking = min will work nicely to show that
t u 4 24
2x 1 3
lim 8 f . The point dependence of the work is just buried in the
3 x 1 2
x 2
focus on the local behavior. The next concept demands a “niceness” that is global.
De¿nition 5.3.1 Given metric spaces X d X and Y dY , a function f : X Y

is uniformly continuous on X if and only if
d e
1 0 2= 0 1 p 1q p q + X F d X p q = " dY f p f q .
Example 5.3.2 The function f x x 2 : U U is uniformly continuous on

[1 3].

For 0 let = . For x1 x2 + [1 3], the triangular inequality yields
6
that
x1 x2 n x1 x2 n 6.
Hence, x1 x2 + [1 3] and x 1 x2 = implies that

n n
n n
f x1 f x2 nx12 x22 n x1 x2 x 1 x2 = 6 .
Since 0 and x1 x2 + [1 3] were arbitrary, we conclude that f is uniformly

continuous on [1 3].
Example 5.3.3 The function f x x 2 : U U is not uniformly continuous on

U.
5.3. UNIFORM CONTINUITY 209
We want to show that there exists a positive real number such that corre-
sponding to every positive real number = we have (at least two points) x1 x1 =
and x2 x2 = for which x 1 = x2 = = and f x1 f x2 o . This
statement is an exact translation of the negation of the de¿nition. For the given
function, we want to exploit the fact the as x increases x 2 increase at a rapid (not
really uniform) rate.
= 1
Take 1. For any positive real number =, let x 1 x 1 = and
2 =
1
x2 x2 = . Then x1 and x 2 are real numbers such that
=
nt u t un
n = 1 1 nn =
n
x1 x2 n =
2 = = n 2
while
n n
n 2 2n
f x1 f x2 nx1 x2 n x1 x2 x1 x2
t ut u
= = 2 =2
1 o 1 .
2 2 = 4
Hence, f is not uniformly continuous on U.
Example 5.3.4 For p + U1 , let f p 2 p 1, p 2 . Then f : U1 U2 is

uniformly continuous on the closed interval [0 2].

For 0, let = T . If p1 + [0 2] and p2 + [0 2], then
2 5
4 p1 p2 2 4 p1 p2 2 n 4 p1 p2 2 n 4 2 22 20
and
T
b c2
dU2 f p1 f p2 2 p1 1 2 p2 12 p12 p22
T T T

p1 p2 4 p1 p2 2 = 20 T 20 .
2 5
Since 0 and p1 p2 + [1 2] were arbitrary, we conclude that f is uniformly
continuous on [0 2].
Theorem 5.3.5 (Uniform Continuity Theorem) If f is a continuous mapping from

a compact metric space X to a metric space Y , then f is uniformly continuous on
X.
Excursion 5.3.6 Fill in what is missing in order to complete the proof of the Uni-
form Continuity Theorem.
Proof. Suppose f is a continuous mapping from a compact metric space X d X
to a metric space Y dY and that 0 is given. Since f is continuous, for each
p + X, there exists a positive real
number = p such thatq + X F d X q p = p "

dY f p f q . Let J N 1 p : p + X . Since neighborhoods are
2 =p
2
open sets, we conclude that J is an . Since X is com-
1
pact there exists a ¿nite number of elements of J that covers X, say p1 p2 pn .
Hence,
> n b c
Xt N1 pj .
j1 =p j
2
1 j k
Let = min = p j . Then, = 0 and the minimum of a ¿nite number of
de f 2 1n jnn
positive real numbers.
Suppose that p q + X are such that d p q =. Because p + X and
6
n b c
X t N1 p j , there exists a positive integer k, 1 n k n n, such that
j1 =p j
2
1
. Hence, d p pk = pk . From the triangular in-
2
2
equality
d X q pk n d X q p d X p pk = n = pk .
3
Another application of the triangular inequality and the choices that were made for
= p yield that
dY f p f q n
4 5
1
***Acceptable ¿ll ins are: (1) open cover for X (2) p + N 1 pk , (3) = pk , (4)
=k 2
2

dY f p f pk dY f pk f q, . ***
2 2
5.4. DISCONTINUITIES AND MONOTONIC FUNCTIONS 211
5.4 Discontinuities and Monotonic Functions

Given two metric spaces X d X and Y dY and a function f from a subset A of
X into Y . If p + X and f is not continuous at p, then we can conclude that f is
not de¿ned at p ( p +
A dom f ), lim f x does not exist, and/or p + A and
x p
x+A
lim f x exists but f p / lim f x a point for which any of the three condi-
x p x p
x+A x+A
tions occurs is called a point of discontinuity. In a general discussion of continuity
of given functions, there is no need to discuss behavior at points that are not in the
domain of the function consequently, our consideration of points of discontinuity
is restricted to behavior at points that are in a speci¿ed or implied domain. Fur-
thermore, our discussion will be restricted to points of discontinuity for real-valued
functions of a real-variable. This allows us to talk about one-sided limits, behavior
on both sides of discontinuities and growth behavior.
De¿nition 5.4.1 A function f is discontinuous at a point c + dom f or has a

discontinuity at c if and only if either lim f x doesn’t exist or lim f x exists
xc xc
and is different from f c.
x
Example 5.4.2 The domain of f x is U
0. Consequently, f has no
x
points of discontinuity on its domain.

! x
, for x + U
0
Example 5.4.3 For the function f x x ,
!

1 , for x 0
dom f U and x 0 is a point of discontinuity of f . To see that lim f x
x0
does not exist, note that, for every positive real number =,
t u n t u n
= n = n
f 1 and n f f 0nn 2.
2 n 2
1
Hence, if , then, for every positive real number =, there exists x + dom f
2
such that 0 x = and f x f 0 o . Therefore, f is not continuous at
x 0.
t u
!
! 1
x sin , for x + U
0
x
Example 5.4.4 If g x , then g has no dis-
!
!

0 , for x 0
continuities in U.
Excursion 5.4.5 Graph the following function f and ¿nd
A
x + dom f : f is continuous at x
and B
x + dom f : f is discontinuous at x.

! 4x 1 1
!
! , xn
!
! 1x 2
!
!
!
! 1
!
! x n1
!

2 ,
2
f x 2x 4 , 1x n3
!
!
!
!
!
! JxK 2 , 3x n6
!
!
!
!
!
14 x 10 , 6 x 14 G 14 x
!
x 14
***Hopefully, your graph revealed that A U

1 3 4 5 6 14 and B
3 4 5 6.***
De¿nition 5.4.6 Let f be a function that is de¿ned on the segment a b. Then,
for any point x + [a b, the right-hand limit is denoted by f x and
b c Kr s L
f x q % 1
tn * n1
t n t x b F lim t n x " lim f t n q
n* n*
and, for any x + a b], the left-hand limit is denoted by f x and
b *
c Kr s L
f x q % 1
tn n1
tn t a x F lim tn x " lim f tn q .
n* n*
Remark 5.4.7 From the treatment of one-sided limits in frosh calculus courses,
recall that lim f t q if and only if
tx
d e
1 0 2= = 0 1t t + dom f F x t x = " f t q
and lim f t q if and only if

tx
d e
1 0 2= = 0 1t t + dom f F x = t x " f t q .
The Sequences Characterization for Limits of Functions justi¿es that these de¿ni-
tions are equivalent to the de¿nitions of f x and f x, respectively.
Excursion 5.4.8 Find f x and f x for every x + B where B is de¿ned in
Excursion 5.4.5.
***For this function, we have f 3 2 f 3 5 f 4 5 f 4

6 f 5 6 f 5 7 and f 6 7 f 6 7.***
Remark 5.4.9 For each point x where a function f is continuous, we must have
f x f x f x.
De¿nition 5.4.10 Suppose the function f is de¿ned on the segment a b and dis-
continuous at x + a b. Then f has a discontinuity of the ¿rst kind at x or a
simple discontinuity at x if and only if both f x and f x exist. Otherwise,
the discontinuity is said to be a discontinuity of the second kind.
Excursion 5.4.11 Classify the discontinuities of the function f in Excursion 5.4.5.

f x , for x + R
1 14
Remark 5.4.12 The function F x where

0 , for x 1 G x 14
f is given in Excursion 5.4.5 has discontinuities of the second kind at x 1 and
x 14.
Excursion 5.4.13 Discuss the continuity of each of the following.

2
!
! x x 6
, x 2
1. f x x 2
!
!

2x 1 , x o 2

2 , x rational
2. g x

1 , x irrational
***Your discussion of (1) combines considers some cases. For * x 2,

x2 x 6
is continuous as the quotient of polynomials for which the denominator
x 2
is not going to zero, while continuity of 2x 1 for x 2 follows from the limit
of the sum theorem or because 2x 1 is a polynomial consequently, the only point
in the domain of f that needs to be checked is x 2. Since f 2
x2 x 6
lim lim x 3 5, f 2 lim 2x 1 5,
x2 x 2 x2 x2
and f 2 5, it follows that f is also continuous at x 2. That the function
given in (2) is not continuous anywhere follows from the density of the rationals
and the irrationals each point of discontinuity is a “discontinuity of the second
kind.”***
De¿nition 5.4.14 Let f be a real-valued function on a segment a b. Then f is

said to be monotonically increasing on a b if and only if
d e
1x1 1x2 x1 x2 + a b F x1 x2 " f x1 n f x2
and f is said to be monotonically decreasing on a b if and only if

d e
1x1 1x2 x1 x2 + a b F x1 x2 " f x1 n f x2
The class of monotonic functions is the set consisting of both the functions that are
increasing and the functions that are decreasing.
Excursion 5.4.15 Classify the monotonicity of the function f that was de¿ned in
Excursion 5.4.5
***Based on the graph, we have that f is monotonically increasing in each of

*
t u1, 1 1, and 3 6 the function t
is monotonically
u decreasing in each of
1 1
3 , 6 14, and 14 *. The section 1 is included in both statements
2 2
because the function is constant there. As an alternative, we could have claimed

that
t f uis both monotonically increasing and monotonically decreasing in each of
1
1 , 3 4, 4 5, and 5 6 and distinguished the other segments according to
2
the property of being strictly monotonically increasing and strictly monotonically
decreasing.***
Now we will show that monotonic functions do not have discontinuities of the
second kind.
Theorem 5.4.16 Suppose that the real-valued function f is monotonically increas-

ing on a segment a b. Then, for every x + a b both f x and f x exist,
sup f t f x n f x n f x inf f t

atx xtb
and
1x 1y a x y b " f x n f y .
of the theorem.
Proof. Suppose that f is monotonically increasing on the

segment a b and x + a b. Then, for every
t + a b such that a t x, . Hence,
1
B
f t : a t x is bounded above by f x.
de f
By the , the set B has a
2
least upper bound let u sup B. Now we want to
show that u f x.
Let 0 be given. Then u sup B and
u u yields the existence of a * + B such that
. From the de¿nition of B,
3
* is the image of a point in . Let = 0
4
be such that x = + a x and f x = *. If

t + x = x, then
f x = n f t and .
5
Since u * and f x n u, the transitivity of less
than or equal to yields that
f t and f t n .

6 7
Because t was arbitrary, we conclude that
1t x = t x " u f t n u .
Finally, it follows from 0 being arbitrary that

1 0 i.e.,
8
f x lim f t u.

tx
For every t + a b such that x t b, we also
have that f x n f t from which it follows that
C
f t : x t b is bounded by
de f
9
f x. From the greatest lower bound property of the
reals, C has a greatest lower bound that we will denote
by ).
Use the space provided to prove that
) f x.
10
Next suppose that x y + a b are such that x y.

Because f x lim f t inf
f t : x t b,
tx
x y t x b and f is monotonically increasing, it
follows that
inf
f t : x t y .
11
From our earlier discussion,
f y lim f t .
ty
12
Now, x y t a y yields that
f y sup
f t : x t y .
Therefore, as claimed.
13
***The expected responses are: (1) f t n f x, (2) least upper bound prop-
erty, (3) ud * u, (4) a x, (5) f t n ef x, (6) u , (7) u, (8)
2= 0 1t x = t x " u f t u , (9) below, (10) Let 0
be given. Then ) inf C implies that there exists * + C such that ) * ) .
Since * + C, * is the image of some point in x b. Let = 0 be such that x = +
x b and f x = *. Now suppose t + x x =. Then f x n f t and
f t n f x = *. Since ) n fd x and * ) , it follows that ) n f t
e
and f t ) . Thus, 2= 0 1t x t x = " ) f t n ) .
Because 0 was arbitrary, we conclude that ) lim f t f x., (11)
tx
f x, (12) sup
f t : a t y and (13) f x n f y.***
Corollary 5.4.18 Monotonic functions have no discontinuities of the second kind.
The nature of discontinuities of functions that are monotonic on segments al-

lows us to identify points of discontinuity with rationals in such a way to give us a
limit on the number of them.
Theorem 5.4.19 If f is monotonic on the segment a b, then

x + a b : f is discontinuous at x
is at most countable.
of Theorem 5.4.19.
Proof. Without loss of generality, we assume that f is a function that is mono-
tonically increasing in the segment a b I . If f is continuous in I , then f has
de f
no points of discontinuity there and we are done. Suppose that f is not continuous
on I and let D
* + I : f is not continuous at *.
From our assumption D / 3 and we can suppose that ? + D. Then
? + dom f ,
1x x + I F x ? " f x n f ?
and

1x x + I F ? x " .
1
From Theorem 5.4.16, f ? and f ? exist furthermore,
f ? sup
f x : x ? , f ? .
2
and f ? n f ? . Since ? is a discontinuity for f , f ? f ? .

3
From the Density of the Rationals, it follows that there exists a rational r? such
that f ? r? f ? . Let Ir? f ? f ? . If D
? 3, then
D 1 and we are done. If D
? / 3 then we can choose another G + D
such that G / ? . Without loss of generality suppose that G + D is such that ? G.
Since ? was an arbitrary point in the discussion just completed, we know that there
exists a rational rG , rG / r? , such that and we can
4
let IrG f G f G. Since ? G , it follows from Theorem 5.4.16 that
n f G . Thus, Ir? D IrG .
5 6
j k
Now, let J Ir< : < + D and H : D J be de¿ned by H *
Ir* . Now we claim that H is a one-to-one correspondence. To see that H is
, suppose that *1 *2 + D and H *1 H *2 .
Then
7
To see that H is onto, note that by de¿nition H D t J and suppose that A + J.
Then
8
11
Finally, H : D J yields that D q J. Since r< + T for each Ir< + J, we have
that J n T 80 . Therefore, D n 80 i.e., D is .
9
***Acceptable responses are : (1) f ? n f x, (2) inf

f x : ? x, (3) , (4)
f G rG f G , (5) f ? , (6) 3, (7) Ir*1 Ir*2 . From the Trichotomy
Law, we know that one and only one of *1 *2 , *1 *2 , or *2 *1 holds.
Since either *1 *2 or *2 *1 implies that Ir*1 D Ir*2 3, we conclude that
*1 *2 . Since *1 and *2 were arbitrary,
1*1 1*2 [H *1 H *2 " *1 *2 ] i.e., H is one-to-one., (8) there ex-
ists r + T such that A Ir and r + f D f D for some D + D. It follows
that H D A or A + H D. Since A was an arbitrary element of J, we have that
1A [A + J " A + H D] i.e., J t H D. From H D t J and J t H D,
we conclude that J H D. Hence, H is onto., and (9) at most countable.***
Remark 5.4.21 The level of detail given in Excursion 5.4.20 was more that was
needed in order to offer a well presented argument. Upon establishing the ability to
associate an interval Ir? with each ? + D that is labelled with a rational and jus-
tifying that the set of such intervals is pairwise disjoint, you can simply assert that
you have established a one-to-one correspondence with a subset of the rationals
and the set of discontinuities from which it follows that the set of discontinuities is
at most countable. I chose the higher level of detail–which is also acceptable–in
order to make it clearer where material prerequisite for this course is a part of the
foundation on which we are building. For a really concise presentation of a proof
of Theorem 5.4.19, see pages 96-97 of our text.
Remark 5.4.22 On page 97 of our text, it is noted by the author that the disconti-
nuities of a monotonic function need not be isolated. In fact, given any countable
subset E of a segment a b, he constructs a function f that is monotonic on a b
with E as set of all discontinuities of f in a b. More consideration of the example
is requested in our exercises.
5.4.1 Limits of Functions in the Extended Real Number System

Recall the various forms of de¿nitions for limits of real valued functions in rela-
tionship to in¿nity:
Suppose that f is a real valued function on U, c is a real number, and L real number,
then
lim f x L % 1 0 2K 0 x K " f x L

x*
% 1 0 2K 0 x K " f x + N L
lim f x L % 1 0 2K 0 x K " f x L

x*
% 1 0 2K 0 x K " f x + N L
lim f x * % 1M + U 2= 0 0 x c = " f x M

xc
b c
% 1M + U 2= 0 x + N=d c " f x M
where N=d c denotes the deleted neighborhood of c, N= c
c.
lim f x * % 1M + U 2= 0 0 x c = " f x M

xc
b c
% 1M + U 2= 0 x + N=d c " f x M
Based of the four that are given, complete each of the following.
lim f x * %
x*
lim f x * %
x*
lim f x * %
x*
lim f x * %
x*
Hopefully, the neighborhood formulation and the pattern of the various state-
ments suggests that we could pull things together if we had comparable descriptions
for neighborhoods of * and *.
De¿nition 5.4.23 For any positive real number K ,
N K *
x + UC
* * : x K
and N K *
x + UC
* * : x K are neighborhoods of * and
*, respectively.
With this notation we can consolidate the above de¿nitions.
De¿nition 5.4.24 Let f be a real valued function de¿ned on U. Then for A and c
in the extended real number system, lim f x A if and only if for every neigh-
xc
borhood of A, N A there exists a deleted neighborhood of c, N `d c, such that
x + N `d c implies that f x + N A. When speci¿cation is needed this will be
referred to a the limit of a function in the extended real number system.
Hopefully, the motivation that led us to this de¿nition is enough to justify the
claim that this de¿nition agrees with the de¿nition of lim f x A when c and A
xc
are real. Because the de¿nition is the natural generalization and our proofs for the
properties of limits of function built on information concerning neighborhoods, we
note that we can establish some of the results with only minor modi¿cation in the
proofs that have gone before. We will simply state analogs.
Theorem 5.4.25 Let f be a real-valued function that is de¿ned on a set E t U

and suppose that lim f t A and lim f t C for c, A, and C in the extended
tc tc
real number system. Then A C.
Theorem 5.4.26 Let f and g be real-valued functions that are de¿ned on a set
E t U and suppose that lim f t A and lim g t B for c, A, and B in the
tc tc
extended real number system. Then
1. lim f g t A B,
tc
2. lim f g t AB, and

tc
t u
f A
3. lim t
tc g B
whenever the right hand side of the equation is de¿ned.
Remark 5.4.27 Theorem 5.4.26 is not applicable when the algebraic manipula-
* A
tions lead to the need to consider any of the expressions * *, 0 *, , or
* 0
because none of these symbols are de¿ned.
The theorems in this section have no impact on the process that you use in
order to ¿nd limits of real functions as x goes to in¿nity. At this point in the
coverage of material, given a speci¿c function, we ¿nd the limit as x goes to in¿nity
by using simple algebraic manipulations that allow us to apply our theorems for
algebraic combinations of functions having ¿nite limits. We close this chapter with
two examples that are intended as memory refreshers.
b 2 c b c
x 3x 3 5 i x 3 x sin x
Example 5.4.28 Find lim .
x* 4x 3 7
Since the given function is the quotient of two functions that go to in¿nity as
x goes to in¿nity, we factor in order to transform the given in to the quotient of
functions that will have ¿nite limits. In particular, we want to make use of the fact
1
that, for any p + J , lim p 0. From
x* x
t u t u
b 2 c b c 1 5 sin x
3 3 i 1 2
x 3x 3 5 i x 3 x sin x x x x
lim lim t u ,
x* 4x 7
3 x* 7
4 3
x
The limit of the quotient and limit of the sum theorem yields that
b 2 c b c
x 3x 3 5 i x 3 x sin x 0 3 0 i 1 0 3 i
lim .
x* 4x 3 7 40 4
rT T s
Example 5.4.29 Find lim 2x 2 x 2 2x 2 x 1 .
x*
In its current form, it looks like the function is tending to * * which is
unde¿ned. In this case, we will try “unrationalizing” the expression in order to get
a quotient at will allow some elementary algebraic manipulations. Note that
rT T s
2x x 2 2x x 1
2 2
rT T s rT T s
2x 2 x 2 2x 2 x 1 2x 2 x 2 2x 2 x 1
rT T s
2x 2 x 2 2x 2 x 1
b 2 c b c
2x x 2 2x 2 x 1
rT T s
2x 2 x 2 2x 2 x 1
2x 3
rT T s.
2x x 2 2x x 1
2 2
T
Furthermore, for x 0, x 2 x x. Hence,
rT T s
lim 2x x 2 2x x 1
2 2
x*
2x 3
lim rT T s
x*
2x 2 x 2 2x 2 x 1
2x 3
lim U U
x* T 1 2 1 1
x2 2 2 2 2
x x x x
3
2
lim 1 U x
x*
U
1 2 1 1
2 2 2 2
x x x x
t u T
2 1 2
1 T T T .
2 2 2 2
5.5. PROBLEM SET E 225
5.5 Problem Set E

1. For each of the following real-valued functions of a real variable give a well-
written = proof of the claim.
b c
(a) lim 3x 2 2x 1 9
x2
(b) lim 8x 2 8
x1
T
(c) lim x 4
x16
3
(d) lim 3
x1 x 2
x 4
(e) lim 7.
x3 2x 5
2. For each of the following real-valued functions of a real variable ¿nd the
implicit domain and range.
sin x
(a) f x
x2 1
T
(b) f x 2x 1
x
(c) f x 2
x 5x 6

! 3
!
! , x 0
!
! x 3
!
!

3. Let f x x 2
!
! , 0n x 2Fx 2
!
! x 2
!
!
!

1 , x 2
(a) Sketch a graph for f

(b) Determine where the function f is continuous.
n 2 n n n
nx 5x 6n , for n x 2 n o
7 5
2
4. Let f x n n and
T
36 6x , for n x 72 n 5
x 2 1
2
x1 , for x / 1
gx .

3 , for x 1
(a) Discuss the continuity of f at x 1.

(b) Discuss the continuity of f gx f xgx at x 1.
5. For f : F U given by f z z give a = proof that lim f z

T z1i
2.
6. When it exists, ¿nd

t u
x2 4 T 2
(a) lim 3x 2
x2 x 2
t u
x 1 T x 1
(b) lim x 4 5
x1 x 2 3x 4 x 1
7. Let f : U U and suppose that lim f x L 0. Prove that

xa
S T
lim f x L
xa
8. Using only appropriate de¿nitions and elementary bounding processes, prove

that if g is a real-valued function on U such that lim gx M / 0, then
xa
1 1
lim 2.
xa [gx]2 M
9. Suppose that A is a subset of a metric space S d, f : A U1 , and

g : A U1 . Prove each of the following.
(a) If c is a real number and f p c for all p + A, then, for any limit
point p0 of A, we have that lim f p c.
p p0
p+ A
5.5. PROBLEM SET E 227
(b) If f p g p for all p + A

p0 were p0 is a limit point of A and
lim f p L, then lim g p L.
p p0 p p0
p+ A p+ A
(c) If f p n g p for all p + A, lim f p L and lim g p M,

p p0 p p0
p+ A p+ A
then L n M.
10. For each of the following functions on U2 , determine whether or not the given
function is continuous at 0 0. Use = proofs to justify continuity or show
lack of continuity by justifying that the needed limit does not exist.

!
! x y2
!
b c , for x y / 0 0
x 2 y2 2
(a) f x y
!
!
!

0 , for x y 0

!
! x 3 y3
!
b 2 c , for x y / 0 0
x y 2 2
(b) f x y
!
!
!
0 , for x y 0

!
! x 2 y4
!
b 2 c , for x y / 0 0
x y 4 2
(c) f x y
!
!
!
0 , for x y 0
11. Discuss the uniform continuity of each of the following on the indicated set.
x2 1
(a) f x in the interval [4 9].
2x 3
(b) f x x 3 in [1 *.
12. For a b, let F [a b] denote the set of all real valued functions that are
continuous on the interval [a b]. Prove that d f g max f x g x
anxnb
is a metric on F [a b].
13. Correctly formulate the monotonically decreasing analog for Theorem 5.4.16
and prove it.
14. Suppose e a segment I a b and that

d that f is monotonically increasing on
2M M + U F 1x x + I " f x n M . Prove that there exists a real
number C such that C n M and f b C.
15. A function f de¿ned on an interval I [a b] is called strictly increasing

on I if and only if f x1 f x 2 whenever x1 x2 for x1 x2 + I . Fur-
thermore, a function f is said to have the intermediate value properly in I if
and only if for each c between f a and f b there is an x0 + I such that
f x0 c. Prove that a function f that is strictly increasing and has the
intermediate value property on an interval I [a b] is continuous on a b.
16. Give an example of a real-valued function f that is continuous and bounded

on [0 * while not satisfying the Extreme Value Theorem.
17. Suppose that f is uniformly continuous on the intervals I1 and I2 . Prove that
f is uniformly continuous on S I1 C I2 .
18. Suppose that a real-valued function f is continuous on I i where I [a b].

If f a and f b exist, show that the function

!
! f a , for x a
!

f 0 x f x , for a x b
!
!
!
f b , for x b
is uniformly continuous on I .
19. If a real valued function f is uniformly continuous on the half open interval
0 1], is it true that f is bounded there. Carefully justify the position taken.
Chapter 6
Differentiation: Our First View
We are now ready to reÀect on a particular application of limits of functions namely,

the derivative of a function. This view will focus on the derivative of real-valued
functions on subsets of U1 . Looking at derivatives of functions in Uk requires a dif-
ferent enough perspective to necessitate separate treatment this is done with Chap-
ter 9 of our text. Except for the last section, our discussion is restricted to aspects
of differential calculus of one variable. You should have seen most of the results
in your ¿rst exposure to calculus–MAT21A on this campus. However, some of the
results proved in this chapter were only stated when you ¿rst saw them and some
of the results are more general than the earlier versions that you might have seen.
The good news is that the presentation here isn’t dependent on previous exposure
to the topic on the other hand, reÀecting back on prior work that you did with the
derivative can enhance your understanding and foster a deeper level of appreciation.
6.1 The Derivative

De¿nition 6.1.1 A real-valued function f on a subset P of U is differentiable at a
point ? + P if and only if f is de¿ned in an open interval containing ? and
f * f ?
lim (6.1)
*? *?
exists. The value of the limit is denoted by f ) ? . The function is said to be differ-
entiable on P if and only if it is differentiable at each ? + P.
229
230 CHAPTER 6. DIFFERENTIATION: OUR FIRST VIEW
Remark 6.1.2 For a function f and a ¿xed point ? , the expression

f * f ?
M *
*?
is one form of what is often referred to as a “difference quotient”. Sometimes it is
written as
f
*
where the Greek letter is being offered as a reminder that difference starts with a
df
“d”. It is the latter form that motivates use of the notation for the ¿rst derivative
d*
of f as a function of *. Other commonly used notations are D* and D1 these only
become useful alternatives when we explore functions over several real variables.
There is an alternative form of (6.1) that is often more useful in terms of compu-
tation and formatting of proofs. Namely, if we let * ? h, (6.1) can be written
as
f ? h f ?
lim . (6.2)
h0 h
Remark 6.1.3 With the form given in (6.2), the difference quotient can be abbrevi-
f
ated as .
h
De¿nition 6.1.4 A real-valued function f on a subset P of U is right-hand differ-
entiable at a point ? + P if and only if f is de¿ned in a half open interval in the
form [? ? = for some = 0 and the one-sided derivative from the right, denoted
by D f ? ,
f ? h f ?
lim
h0 h
exists the function f is left-hand differentiable at a point ? + P if and only if
f is de¿ned in a half open interval in the form ? = ? ] for some = 0 and the
one-sided derivative from the left, denoted by D f ? ,
f ? h f ?
lim
h0 h
exists.
6.1. THE DERIVATIVE 231
De¿nition 6.1.5 A real-valued function f is differentiable on a closed interval

[a b] if and only if f is differentiable in a b, right-hand differentiable at x a
and left-hand differentiable at x b.
x 2
Example 6.1.6 Use the de¿nition to prove that f x is differentiable at
x 1
x 2.
Note that f is de¿ned in the open interval 1 3 which contains * 2.
Furthermore,
t u
*2 4

f * f 2 *1 1 3 * 2
lim lim lim lim 3 3.
*2 *2 *2 *2 *2 *2 *2
Hence, f is differentiable at * 2 and f ) 2 3.
Example 6.1.7 Use the de¿nition to prove that gx x 2 is not differentiable
at x 2.
Since dom g U, the function g is de¿ned in any open interval that
contains x 2. Hence, g is differentiable at x 2 if and only if
g 2 h g 2 h
lim lim
h0 h h0 h
h
exists. Let M h for h / 0. Note that
h
h h h h
lim lim 1 and lim lim 1
h0 h h0 h h0 h h0 h
Thus, M 0 / M 0 from which we conclude that lim M h does not exist.
h0
Therefore, g is not differentiable at x 2.
Remark 6.1.8 Because the function g given in Example 6.1.7 is left-hand differen-
tiable at x 2 and right-hand differentiable at x 2, we have that g is differen-
tiable in each of * 2] and [2 *.
Example 6.1.9 Discuss the differentiability of each of the following at x 0.

1
!
x sin , for x / 0
1. G x x
!

0 , for x 0
1
!
x 3 sin , for x / 0
2. F x x
!

0 , for x 0
First of all, notice that, though the directions did not specify appeal to the de¿-
nition, making use of the de¿nition is the only viable option because of the way the
function is de¿ned. Discussing the differentiability of functions that are de¿ned “in
pieces” requires consideration of the pieces. On segments where the functions are
realized as simple algebraic combinations of nice functions, the functions can be
declared as differentiable based on noting the appropriate nice properties. If the
function is de¿ned one way at a point and a different way to the left and/or right,
then appeal to the difference quotient is mandated.
For (1), we note that G is de¿ned for all reals, consequently, it is de¿ned
in every interval that contains 0. Thus, G is differentiable at 0 if and only if
1 t u
G 0 h G 0 h sin 0 1
lim lim h lim sin
h0 h h0 h h0 h
1 2
exists. For h / 0, let M h sin . For each n + M, let pn . Now,
h H 2n
j 1 k
n1 *
pn * *
n1 converges to 0 as n approaches in¿nity but
M pn n1 1 n1
diverges. From the Sequences Characterization for Limits of Functions (Theorem
5.1.15), we conclude that lim M h does not exist. Therefore, G is not differentiable
h0
at x 0.
The function F given in (2) is also de¿ned in every interval that contains
0. Hence, F is differentiable at 0 if and only if
1 t u
F 0 h F 0 h 3 sin 0 1
lim lim h lim h sin
2
h0 h h0 h h0 h
n n
n 1n
exists. Now we know that, for h / 0, nnsin nn n 1 and lim h 2 0 it follows from
h h0
a simple
t modi¿cation
u of what was proved in Exercise #6 of Problem Set D that
1
lim h 2 sin 0. Therefore, F is differentiable at x 0 and F ) 0 0.
h0 h
Excursion 6.1.10 In the space provided, sketch graphs of G and F on two different
representations of the Cartesian coordinate system in intervals containing 0.
***For the sketch of G using the curves y x and y x as guides to stay within
should have helped give a nice sense for the appearance of the graph the guiding
(or bounding) curves for F are y x 3 and y x 3 .***
Remark 6.1.11 The two problems done in the last example illustrate what is some-
times referred to as a smoothing effect. In our text, it is shown that
1
!
x 2 sin , for x / 0
K x x
!

0 , for x0
is also differentiable at x 0. The function

1
!
sin , for x / 0
L x x
!

0 , for x 0
is not continuous at x 0 with the discontinuity being of the second kind. The
“niceness” of the function is improving with the increase in exponent of the “smooth-
ing function” x n .
In the space provided, sketch graphs of K and L on two different representations

of the Cartesian coordinate system in intervals containing 0.
The function L is not continuous at x 0 while G is continuous at x 0

but not differentiable there. Now we know that K and F are both differentiable at
x 0 in fact, it can be shown that F can be de¿ned to be differentiable at x 0
while at most continuity at x 0 can be gained for the derivative of K at x 0.
Our ¿rst theorem in this section will justify the claim that being differentiable is a
stronger condition than being continuous this offers one sense in which we claim
that F is a nicer function in intervals containing 0 than K is there.
Excursion 6.1.12 Fill in Twhat is missing in order to complete the following proof
that the function f x x is differentiable in U 0 *.
T
Proof. Let f x x and suppose that a + U . Then f is
1
ra s
in the segment 2a that contains x a. Hence, f is differentiable at x a if
2
and only if
lim lim
h0 h0
2 3
exists. Now
bT T c bT T c
ah a ah a
lim lim bT T c
h0 h0 h ah a
3

4

5
.
6
Consequently, f is differentiable at x a and f ) a . Since a + U

7
was arbitrary, we conclude that
v w
b T c ) 1
1x x + U F f x x " f x T .
2 x
d b 1 ce
***Acceptable responses are: (1) de¿ned, (2) f a
h f a h , (3)
dbT T cb ce a h a bT T c1
a h a h 1 , (4) lim bT T c , (5) lim ah a ,
h0 h ah a h0
b T c1 1
(6) 2 a , and (7) T .***
2 a
The next result tells us that differentiability of a function at a point is a stronger
condition than continuity at the point.
Theorem 6.1.13 If a function is differentiable at ? + U, then it is continuous there.
Excursion 6.1.14 Make use of the following observations and your understanding
of properties of limits of functions to prove Theorem 6.1.13
Some observations to ponder:
The function f being differentiable at ? assures the existence of a = 0 such

that f is de¿ned in the segment ? = ? =
Given a function G de¿ned in a segment a b, we know that G is continuous

at any point p + a b if and only if lim G x G p which is equivalent
d e x p
to having lim G x G p 0.
x p
Proof.
d e
***Once you think of the possibility of writing G x G p as
d e
G x G p x p1 x p for x / p the limit of the product theorem
does the rest of the work.***
Remark 6.1.15 We have already seen two examples of functions that are continu-
ous at a point without being differentiable at the point namely, g x x 2 at
x 2 and, for x 0,
1
!
x sin , for x / 0
G x x .
!

0 , for x 0
n n
n 1n
To see that G is continuous at x 0, note that nnsin nn n 1 for x / 0 and lim x 0
t t uu x x0
1
implies that lim x sin 0. Alternatively, for 0, let = then
x0 x
0 x 0 = implies that
n n n n
n n n n
nx sin 1 0n x nsin 1 n n x = .
n x n n xn
t u
1
Hence, lim x sin 0 G 0. Either example is suf¿cient to justify that the
x0 x
converse of Theorem 6.1.13 is not true.
Because the derivative is de¿ned as the limit of the difference quotient, it should
come as no surprise that we have a set of properties involving the derivatives of
functions that follow directly and simply from the de¿nition and application of our
limit theorems. The set of basic properties is all that is needed in order to make a
transition from ¿nding derivatives using the de¿nition to ¿nding derivatives using
simple algebraic manipulations.
Theorem 6.1.16 (Properties of Derivatives) (a) If c is a constant function, then

c) x 0.
(b) If f is differentiable at ? and k is a constant, then hx k f x is differen-

tiable at ? and h ) ? k f ) ? .
(c) If f and g are differentiable at ? , then Fx f gx is differentiable

at ? and F ) ? f ) ? g ) ? .
(d) If u and ) are differentiable at ? , then Gx u)x is differentiable at ?

and
G ) ? u? ) ) ? )? u ) ? .
(e) If f is differentiable at ? and f ? / 0, then H x [ f x]1 is differen-

f ) ?
tiable at ? and H ) ? .
[ f ? ]2
(f) If px x n for n an integer, p is differentiable wherever it is de¿ned and
p ) x nx n1 .
The proofs of (a) and (b) are about as easy as it gets while the straightforward
proofs of (c) and (f) are left as exercises. Completing the next two excursions will
provide proofs for (d) and (e).
Excursion 6.1.17 Fill is what is missing in order to complete the following proof
that, if u and ) are differentiable at ? , then Gx u)x is differentiable at ?
and
G ) ? u? ) ) ? )? u ) ? .
Proof. Suppose u, ), and G are as described in the hypothesis. Because u
and ) are differentiable at ? , they are de¿ned in a segment containing ? . Hence,
G x u x ) x is de¿ned in a segment containing ? . Hence, G is differentiable
at ? if and only if lim exists. Note that

h0
1
lim lim
h0 h0
1 2
) ? h [u ? h u ? ] u ? [) ? h ) ? ]
lim
h0
v t h u t uw
u ? h u ? ) ? h ) ?
lim ) ? h u ? .
h0 h h
Since ) is differentiable at ? it is continuous there thus, lim ) ? h .
h0
3
Now the differentiability of u and ) with the limit of the product and limit of the sum
theorems yield that
lim .
h0
4
1
Therefore, G is differentiable at ? .
d e
1 ,
***Acceptable responses are: (1) G ? h G ? h
d e
(2) u ? h ) ? h u ? ) ? h 1 , (3) ) ? , and (4) ) ? u ) ? u ? ) ) ? .***
Excursion 6.1.18 Fill is what is missing in order to complete the following proof
that, if f is differentiable at ? and f ? / 0, then Hx [ f x]1 is differen-
f ) ?
tiable at ? and H ) ? .
[ f ? ]2
Proof. Suppose that the function f is differentiable at ? and f ? / 0. From
Theorem 6.1.13, f is at ? . Hence. lim f x . Since
x?
1 2
f ?
0, it follows that there exists = 0 such that
2 3
f ?
implies that f x f ? . The (other) triangular inequality, yields
2
f ?
that, for , f ? f x from which
3
2
f ?
we conclude that f x in the segment . Therefore, the
2 4
function H x [ f x]1 is de¿ned in a segment that contains ? and it is dif-
de f
H ? h H ?
ferentiable at ? if and only if lim exists. Now simple algebraic
h0 h
manipulations yield that
vt ut uw
H ? h H ? f ? h f ? 1
lim lim .
h0 h h0 h f ? h f ?
From the of f at ? , it follows that lim f ? h .

h0
5 6
In view of the differentiability of f and the limit of the product theorem, we have
that
H ? h H ?
lim .
h0 h 7
***Acceptable responses are: (1) continuous, (2) b )f ?c, (3) x2 ? =, (4)
? = ? =, (5) continuity, (6) f ? , and (7) f ? [ f ? ] .***
The next result offers a different way to think of the difference quotient.
Theorem 6.1.19 (Fundamental Lemma of Differentiation) Suppose that f is dif-

ferentiable at x0 . Then there exists a function @ de¿ned on an open interval con-
taining 0 for which @0 0 and
f x 0 h f x0 [ f ) x0 @h] h (6.3)
and @ is continuous at 0.
Before looking at the proof take a few moments to reÀect on what you can say about
f x0 h f x0
f ) x0
h
for h 0.
Proof. Suppose that = 0 is such that f is de¿ned in x x0 = and let

1d e
!
f x0 h f x0 f ) x0 , if 0 h =
@ h h .
!

0 , if h 0
Because f is differentiable at x0 , it follows from the limit of the sum theorem that
lim @ h 0. Since @ 0 0, we conclude that @ is continuous at 0. Finally,
h0
1d e
solving @ f x0 h f x0 f ) x0 for f x0 h f x0 yields (6.3).
h
Remark 6.1.20 If f is differentiable at x0 , then
f x0 h s f x0 f ) x0 h
for h very small i.e., the function near to x0 is approximated by a linear function
whose slope is f ) x0 .
Next, we will use the Fundamental Lemma of Differentiation to obtain the

derivative of the composition of differentiable functions.
Theorem 6.1.21 (Chain Rule) Suppose that g and u are functions on U and that
f x gux. If u is differentiable at x0 and g is differentiable at ux0 , then f
is differentiable at x0 and
f ) x0 g ) ux0 u ) x0
***************************
Before reviewing the offered proof, look at the following and think about what
prompted the indicated rearrangement What should be put in the boxes to enable
us to relate to the given information?
We want to consider
f x 0 h f x0
lim
h0 h
g ux0 h g ux0
lim
h0 h
% &
% g ux0 h g ux0 &
lim % &
h0 %
# h &
$
***************************
Proof. Let f f x0 h f x 0 , u ux0 h ux0 and u 0 u x0 .
Then
f gux0 h gux0 gu 0 u gu 0 .
Because u is continuous at x0 , we know that lim u 0. By the Fundamental

h0
Lemma of Differentiation, there exists a function @, with @0 0, that is continu-
ous at 0 and is such that f [g ) u 0 @u]u. Hence,
t u
f ) u
lim lim [g u 0 @u] g ) u 0 u ) x0
h0 h h0 h
from the limit of the sum and limit of the product theorems.
6.1.1 Formulas for Differentiation

As a consequence of the results in this section, we can justify the differentiation
of all polynomials and rational functions. From Excursion 6.1.12, we know that
1
the formula given in the Properties of Derivatives Theorem (f) is valid for n .
2
In fact, it is valid for all nonzero real numbers. Prior to the Chain Rule, the only
r b c12 s8
way to ¿nd the derivative of f x x 3 3x 2 7 , other than appeal to
the de¿nition, was to expand the expression and apply the Properties of Derivatives
Theorem, parts (a), (b), (c) and (f) in view of the Chain Rule and the Properties of
Derivatives Theorem, we have
t r s12 u7 v r s11 w
)
f x 8 x 3x 7
3 2
3x 72x 3x 7
2 2
.
What we don’t have yet is the derivatives of functions that are not realized as al-
gebraic combinations of polynomials most notably this includes the trigonometric
functions, the inverse trig functions, : x for any ¿xed positive real number :, and
the logarithm functions.
For any x + U, we know that
sin x h sin x sin h cos x cos h sin x sin x
lim lim
h0 h h0
v t h
u t uw
sin h cos h 1
lim cos x sin x
h0 h h
and
cos x h cos x cos h cos x sin h sin x cos x
lim lim
h0 h h0
v t h u t uw
cos h 1 sin h
lim cos x sin x .
h0 h h
Consequently, in view of the limit of the sum and limit of the product theorems,
¿nding
t the derivatives
u oftthe sine anducosine functions depends on the existence of
sin h cos h 1
lim and lim . Using elementary geometry and trigonom-
h0 h h0 h
etry, it can be shown that the values of these limits are 1 and 0, respectively. An
outline for the proofs of these two limits, which is a review of what is shown in an
elementary calculus course, is given as an exercise. The formulas for the derivatives
of the other trigonometric functions follow as simple applications of the Properties

of Derivatives.
Recall that e lim 1 ? 1? and y ln x % x e y . With these in addition
? 0
to basic properties of logarithms, for x a positive real,
v t uw
ln x h ln x 1 h
lim lim ln 1
h0 h h0 h x
t u1 h
h
lim ln 1 .
h0 x
Keeping in mind that x is a constant, it follows that

t u x h 1x
ln x h ln x h
lim lim ln 1
h0 h h0 x
t u
1 h x h
lim ln 1
x h0 x
t u
h x h
Because 1 e as h 0 and ln e 1, the same argument that
x
was used for the proof of Theorem 5.2.11 allows us to conclude that
ln x h ln x 1
lim .
h0 h x
Formulas for the derivatives of the inverse trigonometric functions and : x , for
any ¿xed positive real number :, will follow from the theorem on the derivative of
the inverses of a function that is proved at the end of this chapter.
6.1.2 Revisiting A Geometric Interpretation for the Derivative

Completing the following ¿gure should serve as a nice reminder of one of the com-
mon interpretations and applications of the derivative of a function at the point.
On the x-axis, label the x-coordinate of the common point of intersection of

the curve, f x, and the three indicated lines as c.
Corresponding to each line–(1 , (2 , and (3 , on the x-axis label the x-coordinate

of the common point of intersection of the curve, f x, with the line as ch 1 ,
c h 2 , and c h 3 in ascending order. Note that h 1 , h 2 and h 3 are negative
in the set-up that is shown. Each of the lines (1 , (2 , and (3 are called secant
lines.
Find the slopes m 1 , m 2 , and m 3 , respectively, of the three lines.
Excursion 6.1.22 Using terminology associated with the derivative, give a brief
description that applies to each of the slopes m j for j 1 2 3.
Excursion 6.1.23 Give a concise well-written description of the geometric inter-

pretation for the derivative of f at x c, if it exists.
6.2 The Derivative and Function Behavior

The difference quotient is the ratio of the change in function values to the change in
arguments. Consequently, it should come as no surprise that the derivative provides
information related to monotonicity of functions.
6.2. THE DERIVATIVE AND FUNCTION BEHAVIOR 245
In the following, continuity on an interval I [a b] is equivalent to having

continuity on a b, right-hand continuity at x a and left-hand continuity at
x b. For right-hand continuity at x a f a f a, while left-hand
continuity at x b requires that f b f b.
De¿nition 6.2.1 A real valued function f on a metric space X d X has a local
maximum at a point p + X if and only if
d e
2= 0 1q q + N= p " f q n f p
the function has a local minimum at a point p + X if and only if
d e
2= 0 1q q + N= p " f p n f q .
De¿nition 6.2.2 A real valued function f on a metric space X d X has a (global)
maximum at a point p + X if and only if
d e
1x x + X " f x n f p
the function has a (global) minimum at a point p + X if and only if
d e
1x q + X " f p n f x
Theorem 6.2.3 (Interior Extrema Theorem) Suppose that f is a function that is

de¿ned on an interval I [a b]. If f has a local maximum or local minimum at a
point x 0 + a b and f is differentiable at x0 , then f ) x0 0.
Space for scratch work or motivational picture.
Proof. Suppose that the function f is de¿ned in the interval I [a b], has a lo-
cal maximum at x0 + a b, and is differentiable at x0 . Because f has a local max-
imum at xd 0 , there exists a positive real number = such
e that x0 = x0 = t a b
and 1t t + x0 = x0 = " f t n f x0 . Thus, for t + x0 = x 0 ,
f t f x0
o0 (6.4)
t x0
while t + x0 x0 = implies that

f t f x0
n 0. (6.5)
t x0
f t f x0
Because f is differential at x0 , lim exists and is equal to f ) x0 .
tx 0 t x0
From (6.4) and (6.5), we know that f ) x0 o 0 and f ) x0 n 0, respectively. The
Trichotomy Law yields that f ) x0 0.
The Generalized Mean-Value Theorem that follows the next two results contains
Rolle’s Theorem and the Mean-Value Theorem as special cases. We offer the results
in this order because it is easier to appreciate the generalized result after reÀecting
upon the geometric perspective that is offered by the two lemmas.
Lemma 6.2.4 (Rolle’s Theorem) Suppose that f is a function that is continuous

on the interval I [a b] and differentiable on the segment I i a b. If f a
f b, then there is a number x0 + I i such that f ) x0 0.
Space for scratch work or building intuition via a typical picture.
Proof. If f is constant, we are done. Thus, we assume that f is not constant

in the interval a b. Since f is continuous on I , by the Extreme Value Theorem,
there exists points ?0 and ?1 in I such that
f ?0 n f x n f ?1 for all x + I .
Because f is not constant, at least one of

x + I : f x f a and

x + I : f x f a is nonempty. If
x + I : f x f a a b, then
f ?0 f a f b and, by the Interior Extrema Theorem, ?1 + a b is such
that f ) ?1 0. If
x + I : f x f a a b, then f ?1 f a f b,
?0 + a b, and the Interior Extrema Theorem implies that f ) ?0 0. Finally, if
x + I : f x f a / a b and

x + I : f x f a / a b, then both ?0
and ?1 are in a b and f ) ?0 f ) ?1 0.
Lemma 6.2.5 (Mean-Value Theorem) Suppose that f is a function that is contin-

uous on the interval I [a b] and differentiable on the segment I i a b. Then
there exists a number G + I i such that
f b f a
f ) G .
ba
Excursion 6.2.6 Use the space provided to complete the proof of the Mean-Value
Theorem.
Proof. Consider the function F de¿ned by
f b f a
Fx f x x a f a
ba
as a candidate for application of Rolle’s Theorem.
Theorem 6.2.7 (Generalized Mean-Value Theorem) Suppose that f and F are

functions that are continuous on the interval I [a b] and differentiable on the
segment I i. If F ) x / 0 on I i, then
(a) Fb Fa / 0, and

t u
f b f a f ) G
(b) 2G G + Ii F ) .
Fb Fa F G
Excursion 6.2.8 Fill in the indicated steps in order to complete the proof of the
Generalized Mean-Value Theorem.
Proof. To complete a proof of (a), apply the Mean-Value Theorem to F.
For (b), for x + I , de¿ne the function by

f b f a
x f x f a [Fx Fa].
Fb Fa
It follows directly that a b 0.
Theorem 6.2.9 (Monotonicity Test) Suppose that a function f is differentiable in

the segment a b.
(a) If f ) x o 0 for all x + a b, then f is monotonically increasing in a b.
(b) If f ) x 0 for all x + a b, then f is constant in a b.
(c) If f ) x n 0 for all x + a b, then f is monotonically decreasing in a b.
of the Monotonicity Test.
Proof. Suppose that f is differentiable in the segment a b and x 1 x2 + a b
are such that x1 x2 . Then f is continuous in [x1 x2 ] and
1
in x1 x2 . From the , there exists G + x1 x2 such

2
that
f x1 f x2
f ) G .
x1 x2
If f ) x o 0 for all x + a b, then f ) G o 0. Since x 1 x2 0, it
follows that i.e., f x1 n f x2 . Since x1 and x2
3
were arbitrary, we have that

1x1 1x2 x1 x2 + a b F " f x1 n f x2 .
4
Hence, f is in a b.

5
If f ) x 0 for all x + a b, then
.
6
Finally, if f ) x n 0 for all x + a b,
7
***Acceptable responses are: (1) differentiable, (2) Mean-Value Theorem,

(3) f x1 f x 2 n 0, (4) x1 x 2 , (5) monotonically increasing,
(6) f x1 f x 2 0 i.e., f x 1 f x2 . Since x1 and x2 were arbitrary, we
have that f is constant throughout a b., (7) then f ) G n 0 and x1 x2 0
implies that f x1 f x 2 o 0 i.e., f x1 o f x2 . Because x1 and x 2 were
arbitrary we conclude that f is monotonically decreasing in a b.***
Example 6.2.11 Discuss the monotonicity of f x 2x 3 3x 2 36x 7.

For x + U, f ) x 6x 2 6x 36 6 x 3 x 2. Since f ) is positive in
* 3 and 2 *, f is monotonically increasing there, while f ) negative in
3 2 yields that f is monotonically decreasing in that segment.
Remark 6.2.12 Actually, in each of open intervals * 3, 2 *, and 3 2
that were found in Example 6.2.11, we have strict monotonicity i.e., for x1 x2 +
* 3 or x1 x2 + 2 *, x1 x2 implies that f x1 f x2 , while x1 x2 +
3 2 and x1 x2 yields that f x1 f x2 .
6.2.1 Continuity (or Discontinuity) of Derivatives

Given a real-valued function f that is differentiable on a subset P of U, the deriva-
tive F f ) is a function with domain P. We have already seen that F need not be
continuous. It is natural to ask if there are any nice properties that can be associated
with the derivative. The next theorem tells us that the derivative of a real function
that is differentiable on an interval satis¿es the intermediate value property there.
Theorem 6.2.13 Suppose that f is a real valued function that is differentiable on

[a b] and f ) a f ) b. Then for any D + U such that f ) a D f ) b, there
exists a point x + a b such the f ) x D.
Proof. Suppose that f is a real valued function that is differentiable on [a b]
and D + U is such that f ) a D f ) b. Let G t f t Dt. From
the Properties of Derivatives, G is differentiable on [a b]. By Theorem 6.1.13, G
is continuous on [a b] from which the Extreme Value Theorem yields that G has
a minimum at some x + [a b]. Since G ) a f ) t D 0 and G ) b
f ) t D 0, there exists a t1 + a b and t2 + a b such that G t1 G a
and G t2 G b. It follows that neither a G a nor b G b is a minimum
of G in [a b]. Thus, a x b. In view of the Interior Extrema Theorem, we have
that G ) x 0 which is equivalent to f ) x D
Remark 6.2.14 With the obvious algebraic modi¿cations, it can be shown that the
same result holds if the real valued function that is differentiable on [a b] satis¿es
f ) a f ) b.
Corollary 6.2.15 If f is a real valued function that is differentiable on [a b], then
f ) cannot have any simple(¿rst kind) discontinuities on [a b].
Remark 6.2.16 The corollary tells us that any discontinuities of real valued func-
tions that are differentiable on an interval will have only discontinuities of the sec-
ond kind.
6.3. THE DERIVATIVE AND FINDING LIMITS 251
6.3 The Derivative and Finding Limits

The next result allows us to make use of derivatives to obtain some limits: It can
be used to ¿nd limits in the situations for which we have been using the Limit of
Almost Equal Functions and to ¿nd some limits that we have not had an easy means
of ¿nding.
Theorem 6.3.1 (L’Hôpital’s Rule I) Suppose that f and F are functions such that
f ) and F ) exist on a segment I a b and F ) / 0 on I .
t )u t u
f f
(a) If f a Fa 0 and )
a L, then a L.
F F
t )u t u
f f
(b) If f a Fa * and )
a L, then a L.
F F
Excursion 6.3.2 Fill in what is missing in order to compete the following proof of
part (a).
Proof. Suppose that f and F are differentiable on a segment I a b, F ) / 0
on I , and f a Fa 0. Setting f a f a and F a Fa
extends f and F to functions that are in [a b. With this, F a 0
1
and F ) x / 0 in I yields that F x .
2
t u
f)
Suppose that 0 is given. Since a L, there exists = 0
F)
such that a * a = implies that
2
From the Generalized Mean-Value Theorem and the fact that F a f a 0, it
follows that
n n nn 4
n n
n n n
n n n ]`_^ n n n
n n n n n
n n n n n n
n f x n n n n n
n L nn n L n nn L nn
n
n n nn F x F a n n
n n_ ^] ` n
n _^]` n n n n
n 3 n n n n 5 n
n n
n f x n
for some G satisfying a G a =. Hence, nn L nn . Since 0 was
F x
arbitrary, we conclude that .
6
n ) n
n f * n
***Acceptable responses are: (1) continuous, (2) / 0, (3) nn ) L nn , (4)
F *
f ) G f x
F x, (5) f x f a, (6) ) , and (7) lim L.***
F G xa F x
Proof. Proof of (b). Suppose that f and F are functions such that f ) and F )
)
existton an
) u open interval I
x : a x b, F / 0 on I, f a Fa *
f
and a L. Then f and F are continuous on I and there exists h 0
F)
such that F ) / 0 in Ih
x : a x a h. For > 0, there exists a = with
0 = h such that
n ) n
n f G n >
n n
n F ) G L n 2 for all G in I=
x : a x a =
Let x and c be such that x c and x c + I= . By the Generalized Mean-Value

f x f c f ) G
Theorem, there exists a G in I= such that ) . Hence,
Fx Fc F G
n n
n f x f c n >
n L n .
n Fx Fc n 2
In particular, for > 1, we have that

n n n n
n f x f c n n f x f c n
n nn L L n L 1 .
n Fx Fc n n Fx Fc n 2
With a certain amount of playing around we claim that
n n n n
n f x f x f c n n f c Fc f x f c n
n n n n
n Fx Fx Fc n n Fx f x Fx Fc n
n n n nt u
n f c n n Fc n 1
n nn nn n L .
Fx n n f x n 2
f c
For c ¿xed, Fx 0 and Fc
f x 0 as x a . Hence, there exists =1 , 0 =1 =,
such that
n n n n
n f c n > n Fc n 1
n n n n
n Fx n 4 and n f x n 4L 12 .
6.3. THE DERIVATIVE AND FINDING LIMITS 253
Combining the inequalities leads to

n n n n n n
n f x n n f x f x f c n n f x f c n
n n n n n n
n Fx L n n n Fx Fx Fc n n Fx Fc L n >
whenever a x a =1 . Since > 0 was arbitrary, we conclude that

t u
f f x
a lim L.
F xa Fx
Remark 6.3.3 The two statements given in L’Hôpital’s Rule are illustrative of the
set of such results. For example, the x a can be replaced with x b ,
x *, x *, and x *, with some appropriate modi¿cations in the
statements. The following statement is the one that is given as Theorem 5.13 in our
text.
Theorem 6.3.4 (L’Hôpital’s Rule II) Suppose f and g are real and differentiable
) f ) x
in a b, where * n a b n *, g x / 0 for all x + a b, and lim )
xa g x
f x
A. If lim f x 0 F lim g x 0 or lim g x *, then lim A.
xa xa xa xa g x
Excursion 6.3.5 Use an appropriate form of L’Hôpital’s Rule to ¿nd
x 2 5x 6 7 sin x 3
1. lim .
x3 2x 6
t u*
1
2. lim 1
** *1
***Hopefully, you got 3 and e, respectively.***
6.4 Inverse Functions

Recall that for a relation, S, on U, the inverse relation of S, denoted by S 1 , is the
set of all ordered pairs y x such that x y + S. While a function is a relation
that is single-valued, its inverse need not be single-valued. Consequently, we cannot
automatically apply the tools of differential calculus to inverses of functions. What
follows if some criteria that enables us to talk about “inverse functions.” The ¿rst
result tells us that where a function is increasing, it has an inverse that is a function.
Remark 6.4.1 If u and ) are monotonic functions with the same monotonicity, then
their composition (if de¿ned) is increasing. If u and ) are monotonic functions with
the opposite monotonicity, then their composition (if de¿ned) is decreasing.
Theorem 6.4.2 (Inverse Function Theorem) Suppose that f is a continuous func-

tion that is strictly monotone on an interval I with f I J . Then
(a) J is an interval
(b) the inverse relation g of f is a function with domain J that is continuous and
strictly monotone on J and
(c) we have g f x x for x + I and f gy y for y + J .
Proof. Because the continuous image of a connected set is connected and f

is strictly monotone, J is an interval. Without loss of generality, we take f to be
6.4. INVERSE FUNCTIONS 255
decreasing in the interval I . Then f x1 / f x2 implies that x1 / x2 and we

conclude that, for each *0 in J , there exists one and only one ?0 + I such that
*0 f ?0 . Hence, the inverse of f is a function and the formulas given in (c)
hold. It follows from the remark above and (c) that g is strictly decreasing.
To see that g is continuous on J , let *0 be an interior point of J and
suppose that g*0 x0 i.e., f x0 *0 . Choose points *1 and *2 in J such
that *1 *0 *2 . Then there exist points x1 and x2 in I , such that x1 x0 x 2 ,
f x1 *2 and f x2 *1 . Hence, x0 is an interior point of I . Now, without loss
of generality, take > 0 small enough that the interval x0 > x0 > is contained
in I and de¿ne *1` f x0 > and *2` f x0 > so *1` *2` . Since g is
decreasing,
x 0 > g*1` o g* o g*2` x0 > for * such that *1` n * n *2`
Hence,
g*0 > o g* o g*0 > for * such that *1` n * n *2`
Now taking = to be the minimum of *2` *0 and *0 *1` leads to
g* g*0 > whenever * *0 =
Remark 6.4.3 While we have stated the Inverse Function Theorem in terms of in-
tervals, please note that the term intervals can be replaced by segments a b where
a can be * and/or b can be *.
In view of the Inverse Function Theorem, when we have strictly monotone con-
tinuous functions, it is natural to think about differentiating their inverses. For a
proof of the general result concerning the derivatives of inverse functions, we will
make use with the following partial converse of the Chain Rule.
Lemma 6.4.4 Suppose the real valued functions F, G, and u are such that F x
G u x, u is continuous at x0 + U, F ) x0 exists, and G ) u x0 exists and
differs from zero. Then u ) x0 is de¿ned and F ) x0 G ) u x0 u ) x0 .
Excursion 6.4.5 Fill in what is missing to complete the following proof of the
Lemma.
Proof. Let F Fx0 h Fx0 , u ux0 hux0 and u 0 u x0 .

Then
F Gu 0 u Gu 0 .

1
Since u is continuous at x0 , we know that lim u 0. By the Fundamental Lemma

h0
of Differentiation, there exists a function @, with , that is continuous
2
at 0 and is such that F . Hence,
3
F
u h
)
.
h [G u 0 @u]
From lim u 0, it follows that @u 0 as h 0. Because G ) u 0 exists
h0
and is nonzero,
F
ux h ux F ) x0
u ) x0 lim h
0 0
lim ) ) .
h0 h h0 [G u 0 @u] G u 0
Therefore, u ) x0 exists and .
4
***Acceptable responses are: (1) Gux0 h Gux0 , (2) @0 0, (3)
[G ) u 0 @u]u, and (4) F ) x0 G ) u 0 u ) x0 .***
Theorem 6.4.6 (Inverse Differentiation Theorem) Suppose that f satis¿es the hy-
potheses of the Inverse Function Theorem. If x0 is a point of J such that f ) gx0
is de¿ned and is different from zero, then g) x0 exists and
1
g ) x0 . (6.6)
f ) g x 0
Proof. From the Inverse Function Theorem, f g x x. Taking u g and

G f in Lemma 6.4.4 yields that g ) x 0 exists and f ) g x g ) x 1. Since
f ) gx 0 / 0, it follows that g ) x0 f ) gx
1
0
as needed.
6.4. INVERSE FUNCTIONS 257
Corollary 6.4.7 For a ¿xed nonnegative real number :, let g x : x . Then
dom g U and, for all x + U, g ) x : x ln :.
Proof. We know that g x : x is the inverse of f x log: x where f is
a strictly increasing function with domain 0 * and range * *. Because
A log: B % : A B % A ln : ln B, it follows that
ln B
log: B .
ln :
Hence
t u)
)
b c) ln x 1
f x log: x .
ln : x ln :
1
From the Inverse Differentiation Theorem, we have that g ) x
f ) g x
g x ln : : x ln :.
Remark 6.4.8 Taking : e in the Corollary yields that e x ) e x .
In practice, ¿nding particular inverses is usually carried out by working directly

with the functions given rather than by making a sequence of substitutions.
Example 6.4.9 Derive a formula, in terms of x, for the derivative of y arctan x,

H H
x .
2 2
We know that the inverse of u tan ) is a relation that is not a function con-
sequently we need to restrict ourselves to a subset ofrthe domain. Because u is
H Hs
strictly increasing and continuous in the segment I the correspond-
2 2
ing segment is * *. We denote the inverse that corresponds to this segment by
y f x arctan x. From y arctan x if and only if x tan y, it follows directly
b c dy dy 1
that sec2 y 1 or . On the other hand, tan2 y 1 sec2 y with
dx dx sec2 y
dy 1
x tan y implies that sec2 y x 2 1. Therefore, f ) x 2 .
dx x 1
x
Excursion 6.4.10 Use f x to verify the Inverse Differentiation Theorem
1x
on the segment 2 4 i.e., show that the theorem applies, ¿nd the inverse g and
its derivative by the usual algebraic manipulations, and then verify the derivative
satis¿es equation (6.6)
***Hopefully, you thought to use the Monotonicity Test that f is strictly increasing
) 2
in I 2 4 follows immediately upon t notinguthat f x 1 x 0 in
4
I . The corresponding segment J 2 is the domain for the inverse g
3
that we seek. The usual algebraic manipulations for ¿nding inverses leads us to
solving x y 1 y1 for y. Then application of the quotient rule should have
led to g ) x 1 x2 . Finally, to verify agreement with what is claimed with
equation (6.6), substitute g into f ) x 1 x2 and simplify.***
6.5 Derivatives of Higher Order

If f is a differentiable function on a set P then corresponding to each x + P, there
is a uniquely determined f ) x. Consequently, f ) is also a function on P. We have
already seen that f ) need not be continuous on P. However, if f ) is differentiable
on a set t P, then its derivative is a function on which can also be considered
for differentiability. When they exist, the subsequent derivatives are called higher
order derivatives. This process can be continued inde¿nitely on the other hand, we
could arrive at a function that is not differentiable or, in the case of polynomials,
we’ll eventually obtain a higher order derivative that is zero everywhere. Note that
we can speak of higher order derivatives only after we have isolated the set on which
the previous derivative exists.
De¿nition 6.5.1 If f is differentiable on a set P and f ) is differentiable on a set
d2 f
P1 t P, then the derivative of f ) is denoted by f )) or and is called the second
dx 2
6.5. DERIVATIVES OF HIGHER ORDER 259
derivative of f if the second derivative of f is differentiable on a set P2 t P1 , then

d3 f
the derivative of f )) , denoted by f ))) or f 3 or , is called the third derivative
dx 3
of f . Continuing bin this manner,
c when it exists, f n denotes the n th derivative of
)
f and is given by f n1 .
Remark 6.5.2 The statement “ f k exists at a point x0 ” asserts that f k1 t is
de¿ned in a segment containing x0 (or in a half-open interval having x0 as the
included endpoint in cases of one-sided differentiability) and differentiable at x 0 . If
k 2, then the same two claims are true for f k2 . In general, “ f k exists at a
point x0 ” implies that each of f j , for j 1 2 k 1, is de¿ned in a segment
containing x0 and is differentiable at x 0 .
| }
3 5
Example 6.5.3 Given f x in U , ¿nd a general formula
5 2x 2 2
for f . n
From f x 3 5 2x2 b , itc follows that f ) x 32 5 2x3 2,
b c
f )) x 32 3 5 2x4 22 , f 3 x 32 3 4 5 2x5 23 ,
b c
and f 4 x 3 2 3 4 5 5 2x6 24 . Basic pattern recognition
suggests that
f n x 1n 3 2n n 1! 5 2xn2 . (6.7)
Remark 6.5.4 Equation (6.7) was not proved to be the case. While it can be proved
by Mathematical Induction, the set-up of the situation is direct enough that claiming
the formula from a suf¿cient number of carefully illustrated cases is suf¿cient for
our purposes.
Theorem 6.5.5 (Taylor’s Approximating Polynomials) Suppose f is a real func-
tion on [a b] such that there exists n + M for which f n1 is continuous on [a b]
and f n exists for every t + a b. For < + [a b], let
n1
; f k <
Pn1 < t t < k .
k0
k!
Then, for : and ; distinct points in [a b], there exists a point x between : and ;
such that
f n x
f ; Pn1 : ; ; :n . (6.8)
n!
Excursion 6.5.6 Fill in what is missing to complete the following proof of Taylor’s
Approximating Polynomials Theorem.
Proof. Since Pn1 : ;, ; :n and f ; are ¿xed, we have that
f ; Pn1 : ; M ; :n
for some M + U. Let
g t f t Pn1 : t M t :n .

de f
Then g is a real function on [a b] for which is continuous and

1
g n exists in a b because . From
2
the Properties of Derivatives, for t + a b, we have that
;
n1 k
f :
) )
g t f t t :k1 n M t :n1 ,
k1
k 1!
and
g )) t .
3
In general, for j such that 1 n j n n 1 and t + a b, it follows that
;
n1 k
f : n!
j j
g t f t t :k j M t :n j .
k j
k j ! n j!
Finally,
g n t (6.9)

4
Direct substitution yields that g : 0. Furthermore, for each j , 1 n

3n1 f k :
j n n 1, t : implies that k j t :k j f k : conse-
k j!
quently,
g : g j : 0 for each j, 1 n j n n 1 .

6.5. DERIVATIVES OF HIGHER ORDER 261
In view of the choice of M, we have that g ; 0. Because g is differentiable in

a b, continuous in [a b], and g : g ; 0, by ,
5
there exists x 1 between : and ; such that g) x
1 0. Assuming that n 1,
from Rolle’s Theorem, g ) differentiable in a b and in [a b] with
6
g ) : g ) x1 0 for : x1 + a b yields the existence of x2 between : and
x1 such that . If n 2, Rolle’s Theorem can be applied to g )) to
7
obtain x3 between such that g 3 x3 0. We can repeat this pro-
8
cess through g n , the last higher order derivative that we are assured exists. After
n steps, we have that there is an xn between : and xn1 such that g n xn 0.
Substituting xn into equation (6.9) yields that
0 gn xn .
9
Hence, there exists a real number x xn that is between : and ; such that
f n x
f n x n!M i.e., M. The de¿nition of M yields equation (6.8).
n!
***Acceptable responses are: (1) g n1 , (2) g is the sum of functions having those
3 f k :
properties, (3) f )) t n1 t :k2 n n 1 M t :n2 , (4)
k2
k 2!
f n t n!M, (5) Rolle’s Theorem or the Mean-Value Theorem, (6) continuous,
(7) g )) x2 0, (8) : and x2 , and (9) f n xn n!M.***
Remark 6.5.7 For n 1, Taylor’s Approximating Polynomials Theorem is the

Mean-Value Theorem. In the general case, the error from using Pn1 : ; instead
f n x
of f ; is ; :n for some x between : and ; consequently, we have
n! n n
an approximation of this error whenever we have bounds on n f n xn.
vw
1 3 7
Example 6.5.8 Let f x 1 x in . Then, for each n + M, f n x
v w 4 8
3 7
n! 1 xn1 is continuous in . Consequently, the hypotheses for Tay-
4 8
lor’s Approximating Polynomial Theorem are met for each n + M. For n 2,

1 1
Pn1 < t P1 < t t < .
1< 1 < 2
t u
1 1 1 1
If : and ; , the Theorem claims the existence of x + such
4 2 2 4
that
t u t u t u
1 1 1 f 2 x 1 1 2
f P1 .
2 4 2 2! 2 4
Since
t u t u
1 1 1 1 1 1
P1 t u2 0
4 2 1 1 2 4
1 1
4 4
t u t u
1 1 2 1 9
we wish to ¿nd x + such that 0 the only real
2 4 3 1 x 16
3
3
solution to the last equation is x 0 1 T which is approximately equal to
234
1 1
055. Because x0 is between : and ; , this veri¿es the Theorem for the
4 2
speci¿ed choices.
6.6 Differentiation of Vector-Valued Functions

In the case of limits and continuity we have already justi¿ed that for functions
from U into Uk , properties are ascribed if and only if the property applies to each
coordinate. Consequently, it will come as no surprise that the same “by co-ordinate
property assignment” carries over to differentiability.
De¿nition 6.6.1 A vector-valued function f from a subset P of U into Uk is differ-
entiable at a point ? + P if and only if f is de¿ned in a segment containing ? and
there exists an element of Uk , denoted by f) ? , such that
n n
n f t f ? n
lim nn f ? nn 0
)
t? t ?
where denotes the Euclidean k-metric.
6.6. DIFFERENTIATION OF VECTOR-VALUED FUNCTIONS 263
Lemma 6.6.2 Suppose that f 1 f 2 f k are real functions on a subset P of U and
f x f 1 x f 2 x f k x for x + P. Then f is differentiable at ? + P with
derivative f) ? b if and only if each of thecfunctions f 1 f 2 f k is differentiable at
? and f) ? f 1) ? f 2) ? f k) ? .
Proof. For t and ? in U, we have that
t u
f t f ? ) f 1 t f 1 ? ) f k t f k ? )
f ? f 1 ? f k ? .
t ? t ? t ?
Consequently, the result follows immediately from Lemma 4.3.1 and the Limit of
Sequences Characterization for the Limits of Functions.
Lemma 6.6.3 If f is a vector-valued function from P t U into Uk that is differen-
tiable at a point ? + P, then f is continuous at ? .
Proof. Suppose that f is a vector-valued function from P t U into Uk that is
differentiable at a point ? + P. Then f is de¿ned in a segment I containing ? and,
for t + I , we have that
t u
f 1 t f 1 ? f k t f k ?
f t f ? t ? t ?
t ? t ?
b c
f 1) ? 0 f 2) ? 0 f k) ? 0 as t ? .
Hence, for each j + M, 1 n j n k, lim f j t f j ? i.e., each f j is continuous
t?
at ? . From Theorem 5.2.10(a), it follows that f is continuous at ? .
We note that an alternative approach to proving Lemma 6.6.3 simply uses Lemma
6.6.2. In particular, from Lemma 6.6.2, f x f 1 x f 2 x f k x differen-
tiable at ? implies that f j is differentiable at ? for each j, 1 n j n k. By Theorem
6.1.13, f j is continuous at ? for each j , 1 n j n k, from which Theorem 5.2.10(a)
allows us to conclude that f x f 1 x f 2 x f k x is continuous at ? .
Lemma 6.6.4 If f and g are vector-valued functions from P t U into Uk that are
differentiable at a point ? + P, then the sum and inner product are also differen-
tiable at ? .
Proof. Suppose that f x f 1 x f 2 x f k x and
g x g1 x g2 x gk x are vector-valued functions from P t U into
Uk that are differentiable at a point ? + P. Then
f g x f 1 g1 x f 2 g2 x f k gk x
and
f g x f 1 g1 x f 2 g2 x f k gk x .
From
b thec Properties
b cof Derivatives (c) and (d), bfor eachc)j + M, 1) n j n )
k,
f j g j and f j g j are differentiable at ? with f j g j ? f j ? g j ?
b c)
and f j g j ? f j) ? g j ? f j ? g )j ? . From Lemma 6.6.2, it follows that
f g is differentiable at ? with
b c bb c b c b c c
f g ? f) g) ? f 1) g1) ? f 2) g2) ? f k) gk) ?
and f g is differentiable at ? with

b c b c
f g) ? f) g ? f g) ? .
The three lemmas might prompt an unwarranted leap to the conclusion that all
of the properties that we have found for real-valued differentiable functions on sub-
sets of U carry over to vector-valued functions on subsets of U. A closer scrutiny
reveals that we have not discussed any results for which the hypotheses or conclu-
sions either made use of or relied on the linear ordering on U. Since we loose the
existence of a linear ordering when we go to U2 , it shouldn’t be a shock that the
Mean-Value Theorem does not extend to the vector-valued functions from subsets
of U to U2 .
Example 6.6.5 For x + U, let f x cos x sin x. Show that there exists an in-
terval [a b] such that f satis¿es the hypotheses of the Mean-Value Theorem without
yielding the conclusion.
From Lemma 6.6.2 and Lemma 6.6.3, we have that f is differentiable in
a b and continuous in [a b] for any a b + U such that a b. Sincen f 0n
f 2H 1 0, f 2H f 0 0 0. Because f) x n) n
b ) sin x cos cx, f x
1 for each x + 0 2H . Inbparticular, 1x + 0 2H f xc / 0 0 from which
we see that 1x + 0 2H f 2H f 0 / 2H 0 f) x i.e.,
d b ce
2x x + 0 2H F f 2H f 0 2H 0 f) x
Remark 6.6.6 Example 5.18 in our text justi¿es that L’Hôpital’s Rule is also not
valid for functions from U into F.
6.6. DIFFERENTIATION OF VECTOR-VALUED FUNCTIONS 265
When we justify that a result known for real-valued differentiable functions on

subsets of U does not carry over to vector-valued functions on subsets of U, it is
natural to seek modi¿cations of the original results in terms of properties that might
carry over to the different situation. In the case of the Mean-Value Theorem, success
in achieved with an inequality that follows directly from the theorem. From the
Mean-Value Theorem, if f is a function that is continuous on the interval I [a b]
and differentiable on the segment I i a b, then there nexists na numbern G + Ini
such that f b f a f ) G b a. Since G + I i , n f ) G n n sup n f ) xn.
n )i n
x+I
This leads to the weaker statement that f b f a n b a sup n f xn. On
x+I i
the other hand, this statement has a natural candidate for generalization because the
absolute value or Euclidean 1-metric can be replaced with the Euclidean k-metric.
We end this section with a proof of a vector-valued adjustment of the Mean-Value
Theorem.
Theorem 6.6.7 Suppose that f is a continuous mapping of [a b] into Uk that is

differentiable in a b. Then there exists x + a b such that
n n
f b f a n b a nf) xn (6.10)
Proof. Suppose that f f 1 f 2 is a continuous mapping of [a b] into Uk that

is differentiable in a b and let z f b f a. Equation 6.10 certainly holds
if z 0 0 consequently, we suppose that z / 0 0. By Theorem 5.2.10(b) and
Lemma 6.6.4, the real-valued function
M t z f t for t + [a b]
is continuous in [a b] and differentiable in a b. Applying the Mean-Value The-
orem to M, we have that there exists x + a b such that
M b M a M ) x b a . (6.11)
Now,
M b M a z f b z f a

f b f a f b f b f a f a
f b f a f b f a
z z z2 .
For z 1 f 1 b f 1 a and z 2 f 2 b f 2 a,

n n n n
M x nz f) xn nz 1 f 1) x z 2 f 2) xn
S Tn n n n n n
n z 1 z 2 n f 1) xn n f 2) xn z nf) xn
by Schwarz’s Inequality. Substituting into equation (6.11) yields

n n n n
z2 b a nz f) xn n b a z nf) xn
n n
which implies z n b a nf) xn because z / 0.
6.7 Problem Set F

1. Use the de¿nition to determine whether or not the given function is differen-
tiable at the speci¿ed point. When it is differentiable, give the value of the
derivative.
(a) f x x 3 x 0
3
x , for 0 n xn1
(b) f x x 1
T
x , for x 1
T 1
!
x sin , for x / 0
(c) f x x x0
!

0 , for x 0
9
(d) f x 2 x2
2x 1
2. Prove that, if f and g are differentiable at ? , then Fx f gx is
differentiable at ? and F ) ? f ) ? g) ? .
3. Use the de¿nition of the derivative to prove that f x x n is differentiable
on U for each n + M.
2
x , for x + T
4. Let f x .

0 , for x + T
Is f differentiable at x 0? Carefully justify your position.
6.7. PROBLEM SET F 267
5. If f is differentiable at ? , prove that

f ? :h f ? ;h
lim : ; f ) ? .
h0 h
6. Discuss the differentiability of the following functions on U.
(a) f x x x 1
(b) f x x x
7. Suppose that f : U U is differentiable at a point c + U. Given any

two sequences
an * *
n1 and
bn n1 such that an / bn for each n + M and
lim an lim bn c, is it true that
n* n*
f bn f an
lim f ) c?
n* bn an
State your position and carefully justify it.
8. Use the Principle of Mathematical Induction to prove the Leibnitz Rule for
the n th derivative of a product:
;n t u
n n nk
f g x f x g k x
k0
k
bn c n!
where and f 0 x f x.
k n k! k!
9. Use derivative formulas to ¿nd f ) x for each of the following. Do only the
obvious simpli¿cations.
4x 6 3x 1
(a) f x r b c7 s
x 4x 5x 7x
5 2 3 4
3
1 2x b 9 c2
(b) f x 4x 2 b c3 4x 3x 2 10
2 x2
15
b 2 c3
% 2x 3x 7 &
5
(c) f x # r T s4 $
14 4 x 3 2
t u10 r
1 T s5 12
(d) f x 3x 5 5 4 7x 4 3 5x 2
x
T S T
(e) f x 3 2 1 x
10. Complete the following steps to prove that

sin A cos A 1
lim 1 and lim 0.
A0 A A0 A
(a) Draw a ¿gure that will serve as an aid towards completion of a proof
sin A
that lim 1.
A0 A
i. On a copy of a Cartesian coordinate system, draw a circle having
radius 1 that is centered at the origin. Then pick an arbitrary point
on the part of the circle that is in the ¿rst quadrant and label it P.
ii. Label the origin, the point 1 0, and the point where the line x
P would intersect the x-axis,
O B and A, respectively.
iii. Suppose that the argument of the point P, in radian measure, is A .

Indicate the coordinates of the point P and show the line segment
joining P to A in your diagram.
iv. If your completed diagram is correctly labelled, it should illustrate
that
n n
sin A n P An
'
A length of P B
n n
where n P An denotes the length of the line segment joining points
'
P and A and P B denotes the arc of the unit circle from the point B
to the point P.
n n
v. Finally, the circle having radius n O An and centered at the origin
will pass through the point A and a point and a point on the ray

O P. Label the point of intersection with O P with the letter C and
'
show the arc C A on your diagram.
(b) Recall that, for a circle of radius r , the area of a sector subtended by A
Ar 2
radians is given by . Prove that
2
A cos2 A cos A sin A A

2 2 2
for A satisfying the set-up from part (a).
sin A
(c) Prove that lim 1.
A 0 A
cos A 1
(d) Recall that sin2 A cos2 A 1. Prove that lim 0.
A 0 A
11. The result of Problem 10 in conjunction with the discussion that was offered
in the section on Formulas for Derivatives justi¿es the claim that, for any
x + U, sin x) cos x and cos x) sin x, where x is interpreted as
radians. Use our Properties of Derivatives and trig identities to prove each of
the following.
(a) tan x) sec2 x

(b) sec x) sec x tan x
(c) csc x) csc x cot x
(d) ln sec x tan x) sec x
(e) ln csc x cot x) csc x
12. Use derivative formulas to ¿nd f ) x for each of the following. Do only the
obvious simpli¿cations.
r r T ss
(a) f x sin5 3x 4 cos2 2x 2 x 4 7
b c
tan3 4x 3x 2
(b) f x b c
1 cos2 4x 5
t u2
b c4 3
(c) f x 1 sec3 3x x3 2 tan x
2x 1
r T s4
(d) f x cos3 x 4 4 1 sec4 x
13. Find each of the following. Use L’Hôpital’s Rule when it applies.
tan x
(a) limH
x 2 x H2
tan5 x tan3 x
(b) lim
x0 1 cos x
x3
(c) lim 2x
x* e
4x 3 2x 2 x
(d) lim 3
x* 5x 3x 2 2x
tan x x
(e) lim
x0 x3
(f) lim x 2 ln x 2
x2
14. For f x x 3 and x0 2 in the Fundamental Lemma of Differentiation,

show that @ h 6h h 2 .
x 1
15. For f x and x0 1 in the Fundamental Lemma of Differentia-
2x 1
tion, ¿nd the corresponding @ h.
16. Suppose that f , g, and h are three real-valued functions on U and c is a ¿xed
real number such that f c g c h c and f ) c g ) c h ) c. If

A1 A2 A3 is a partition of U, and

f x , for x + A1
L x g x , for x + A2 ,

h x , for x + A3
prove that L is differentiable at x c.
17. If the second derivative for a function f exists at x0 + U, show that
f x0 h 2 f x0 f x0 h

lim f )) x0 .
h0 h2
18. For each of the following, ¿nd formulas for f n in terms of n + M.
3
(a) f x
3x 22
(b) f x sin 2x
(c) f x ln 4x 3
(d) f x e5x7
2
ex , for x 0
19. For f x , show that f n 0 exists for each

0 , for x n0
n + M and is equal to 0.
20. Discuss the monotonicity of each of the following.
(a) f x x 4 4x 5
(b) f x 2x 3 3x 5
3x 1
(c) f x
2x 1
(d) f x x 3 ex
(e) f x 1 x ex
ln x
(f) f x 2
x
21. Suppose that f is a real-valued function on U for which both the ¿rst and sec-
ond derivatives exist. Determine conditions on f ) and f )) that will suf¿ce to
justify that the function is increasing at a decreasing rate, increasing at an in-
creasing rate, decreasing at an increasing rate, and decreasing at a decreasing
rate.
22. For a function f from a metric space X to a metric space Y , let F f denote the
inverse relation from Y to X. Prove that F f is a function from rng f into X
if and only if f is one-to-one.
23. For each of the following,
¿nd the segments Ik , k 1 2 , where f is strictly increasing and

strictly decreasing,
¿nd the corresponding segments Jk f Ik on which the correspond-

ing inverses gk of f are de¿ned,
graph f on one Cartesian coordinate system, and each of the corre-
sponding inverses on a separate Cartesian coordinate system, and
whenever possible, with a reasonable amount of algebraic manipula-
tions, ¿nd each gk .
(a) f x x 2 2x 2
2x
(b) f x
x 2
x2
(c) f x 3x 4
2
3H
(d) f x sin x for n x n 2H
2
2x 3
(e) f x x 2 4x 1
3
24. Suppose that f and g are strictly increasing in an interval I and that
f g x 0
for each x + I . Let F and G denote the inverses of f and g, respectively,

and J1 and J2 denote the respective domains for those inverses. Prove that
F x G x for each x + J1 D J2 .
25. For each of the following, the Inverse Function Theorem applies on the indi-
cated subset of U. For each given f ¿nd the corresponding inverse g. Use the
properties of derivatives to ¿nd f ) and g ) . Finally, the formulas for f ) and g )
to verify equation (6.6).
(a) f x x 3 3x for * *

t u
4x 1
(b) f x 2 for *
x 1 2
(c) f x e4x for * *
1
!
x sin , for 0 x n 1
26. For f x x , ¿nd the segments Ik , k1 2 ,
!

0 , for x 0
where f is strictly increasing and strictly decreasing and the corresponding
segments Jk where the Inverse Function Theorem applies.
27. For each of the following, ¿nd the Taylor polynomials P t as described in
Taylor’s Approximating Polynomials Theorem about the indicated point < .
2
(a) f x < 1
5 2x
H
(b) f x sin x <
4
(c) f x e 2x1 < 2
(d) f x ln 4 x < 1
28. For each of the following functions from U into U3 , ¿nd f) .

t u
x 3 sin x
(a) f x x tan 3x e cos 3x 4
2x
x2 1
b b c b cc
(b) f x ln 2x 2 3 sec x sin3 2x cos4 2 3x 2
b c
29. For f x x 2 2x 2 3x 2 in [0 2], verify equation (6.10).
Chapter 7
Riemann-Stieltjes Integration
Calculus provides us with tools to study nicely behaved phenomena using small
discrete increments for information collection. The general idea is to (intelligently)
connect information obtained from examination of a phenomenon over a lot of tiny
discrete increments of some related quantity to “close in on” or approximate some-
thing that behaves in a controlled (i.e., bounded, continuous, etc.) way. The “clos-
ing in on” approach is useful only if we can get back to information concerning the
phenomena that was originally under study. The bene¿t of this approach is most
beautifully illustrated with the elementary theory of integral calculus over U. It en-
ables us to adapt some “limiting” formulas that relate quantities of physical interest
to study more realistic situations involving the quantities.
Consider three formulas that are encountered frequently in most standard phys-
ical science and physics classes at the pre-college level:
A l* d r t m d l.
Use the space that is provided to indicate what you “know” about these formulas.
Our use of these formulas is limited to situations where the quantities on the
right are constant. The minute that we are given a shape that is not rectangular,
a velocity that varies as a function of time, or a density that is determined by our
position in (or on) an object, at ¿rst, we appear to be “out of luck.” However, when
the quantities given are well enough behaved, we can obtain bounds on what we
275
276 CHAPTER 7. RIEMANN-STIELTJES INTEGRATION
wish to study, by making certain assumptions and applying the known formulas
incrementally.
Note that except for the units, the formulas are indistinguishable. Consequently,
illustrating the “closing in on" or approximating process with any one of them car-
ries over to the others, though the physical interpretation (of course) varies.
Let’s get this more down to earth! Suppose that you build a rocket launcher as
part of a physics project. Your launcher ¿res rockets with an initial velocity of 25
ft/min, and, due to various forces, travels at a rate ) t given by
) t 25 t 2 ft/min
where t is the time given in minutes. We want to know how far the rocket travels in
the ¿rst three minutes after launch. The only formula that we have is d r t, but
to use it, we need a constant rate of speed. We can make use of the formula to obtain
bounds or estimates on the distance travelled. To do this, we can take increments in
the time from 0 minutes to 3 minutes and “pick a relevant rate” to compute a bound
on the distance travelled in each section of time. For example, over the entire three
minutes, the velocity of the rocket is never more that 25 f tmi n.
What does this tell us about the product
25 ft/min 3 min
compared to the distance that we seek?
How does the product 16 ft/min 3 min relate to the distance that we seek?
We can improve the estimates by taking smaller increments (subintervals of 0

minutes to 3 minutes) and choosing a different “estimating velocity” on each subin-
terval. For example, using increments of 15 minutes and the maximum velocity
that is achieved in each subinterval as the estimate for a constant rate through each
7.1. RIEMANN SUMS AND INTEGRABILITY 277
subinterval, yields an estimate of

tt u u
9 573
25 ft/min 15 min 25 ft/min 15 min ft.
4 8
Excursion 7.0.1 Find the estimate for the distance travelled taking increments of
one minute (which is not small for the purposes of calculus) and using the minimum
velocity achieved in each subinterval as the “estimating velocity.”
***Hopefully, you obtained 61 feet.***

Notice that none of the work done actually gave us the answer to the original
problem. Using Calculus, we can develop the appropriate tools to solve the problem
as an appropriate limit. This motivates the development of the very important and
useful theory of integration. We start with some formal de¿nitions that enable us to
carry the “closing in on process” to its logical conclusion.
7.1 Riemann Sums and Integrability

De¿nition 7.1.1 Given a closed interval I [a b], a partition of I is any ¿nite
strictly increasing sequence of points S
x0 x1 xn1 xn such that a x0
and b xn . The mesh of the partition
x0 x1 xn1 xn is de¿ned by
b c
mesh S max x j x j1
1n jnn
dEach partition
e of I ,
x0 x1 xn1 xn , decomposes I into n subintervals I j
x j1 x j , j 1 2 n, such that I j D Ik x j if and only if k j 1 and is
empty for k / j or k / j 1. Each such decomposition of I into subintervals is
called a subdivision of I.
Notation 7.1.2 Given a partition Sb c
x0 x1 xn1
b xn of anc interval I
[a b], the two notations x j and ( I j will be used for x j x j1 , the length of
the j th subinterval in the partition. The symbol or I will be used to denote
an arbitrary subdivision of an interval I .
If f is a function whose domain contains the closed interval I and f is bounded

on the interval I , we know that f has both a least upper bound and a greatest lower
bound on I as well as on each interval of any subdivision of I .
De¿nition 7.1.3 Given a function f that is bounded and de¿ned d on thee interval
I and a partition S
x0 x1 xn1 xn of I , let I j x j1 x j , M j
sup f x and m j inf f x for j 1 2 n. Then the upper Riemann sum of
x+I j x+I j
f with respect to the partition S, denoted by U S f , is de¿ned by
;
n
U S f M j x j
j1
and the lower Riemann sum of f with respect to the partition S, denoted by
L S f , is de¿ned by
;
n
L S f m j x j
j1
b c
where x j x j x j1 .
Notation 7.1.4 With the subdivision notation the upper and lower Riemann sums
for f are denoted by U f and L f , respectively.
| }
1 1 3
Example 7.1.5 For f x 2x 1 in I [0 1] and S 0 1 ,
t u t 4 2u 4
1 3 5 9 1 3 5 7
U S f 2 3 and L S f 1 2 .
4 2 2 4 4 2 2 4

0 , for x + TD [0 2]
Example 7.1.6 For g x

1 , for x + TD [0 2]
U I g 2 and L I g 0 for any subdivision of [0 2].
To build on the motivation that constructed some Riemann sums to estimate a

distance travelled, we want to introduce the idea of re¿ning or adding points to
partitions in an attempt to obtain better estimates.
De¿nition 7.1.7 For a partition Sk

x 0 x1 xk1 xk of an interval I
[a b], let k denote to corresponding subdivision of [a b]. If Sn and Sm are
partitions of [a b] having n 1 and m 1 points, respectively, and Sn t Sm , then
Sm is a re¿nement of Sn or m is a re¿nement of n . If the partitions Sn and Sm
are independently chosen, then the partition Sn C Sm is a common re¿nement of
Sn and Sm and the resulting Sn C Sm is called a common re¿nement of n
and m .
| } | }
1 3 ` 1 1 1 5 3
Excursion 7.1.8 Let S 0 1 and S 0 1 .
2 4 4 3 2 8 4
(a) If and ` are the subdivisions
|v that correspond S and S ` ,
w vof I w v[0 1]w}
1 1 3 3
respectively, then 0 1 . Find ` .
2 2 4 4
v w v w v w
1 1 3 3
(b) Set I1 0 , I2 , and I3 1 . For k 1 2 3, let k be
2 2 4 4
the subdivision of Ik that consists of all the elements of ` that are contained
in Ik . Find k for k 1 2 and 3.
(c) For f x x 2 and the notation established in parts (a) and (b), ¿nd each of
the following.
(i) m inf f x

x+I
(ii) m j inf f x for j 1 2 3

x+I j
| }
(iii) m `j inf inf f x : J + j
x+J
(iv) M sup f x

x+I
(v) M j sup f x for j 1 2 3

x+I j
| }
(vi) M `j sup sup f x : J + j
x+J
(d) Note how the values m, m j , m `j , M, M j , and M `j compare. What you ob-
served is a special case of the general situation. Let
S
x0 a x1 xn1 xn b
be a partition of an interval I [a b], be the corresponding subdivision
of [a b] and S ` denote a re¿nement of S with corresponding subdivision de-
noted by ` . For k 1 2 n, let k be the subdivision of Ik consisting
of the elements of ` that are contained in Ik . Justify each of the following
claims for any function that is de¿ned and bounded on I .
(i) If m inf f x and m j inf f x, then, for j 1 2 n, m n m j

x+I x+I j
and m j n inf f x for J + j .
x+J
(ii) If M sup f x and M j sup f x, then, for j 1 2 n, M j n

x+I x+I j
M and M j o sup f x for J + j .
x+J
Our next result relates the Riemann sums taken over various subdivisions of an
interval.
Lemma 7.1.9 Suppose that f is a bounded function with domain I [a b]. Let
be a subdivision of I , M sup f x, and m inf f x. Then
x+I x+I
m b a n L f n U f n M b a (7.1)
and
b c b c
L f n L ` f n U ` f n U f (7.2)
for any re¿nement ` of . Furthermore, if < and D are any two subdivisions
of I , then
b c
L < f n U D f (7.3)
Excursion 7.1.10 Fill in what is missing to complete the following proofs.
Proof. Suppose that f is a bounded function with domain I [a b], M

sup f x, and m inf f x. For
Ik : k 1 2 n an arbitrary subdi-
x+I x+I
vision of I , let M j sup f x and m j inf f x. Then I j t I for each
x+I j x+I j
j 1 2 n, we have that
m n mj n , for each j 1 2 n.

1
b c
Because x j x j x j1 o 0 for each j 1 2 n, it follows immediately
that
;
n b c ;n
m x j x j1 n m j x j L f
2 j1 j1
and
;
n ;
n
m j x j n M j x j U f n M b a .
j1 j1 3
Therefore, m b a n L f n U f n M b a as claimed in equation

(7.1).
Let ` be a re¿nement of and, for each k 1 2 n, let k be the

subdivision of Ik that consists of all the elements of ` that are contained in Ik .
In view of the established conventions for the notation being used, we know that
1J J + ` " 2!k k +
1 2 n F J + k also, for each J + k,
J t Ik " m k inf f x n inf f x and Mk sup f x o sup f x. Thus,
x+Ik x+J x+Ik x+J
m k ( Ik n L k f and Mk ( Ik o U k f

;
n b c ; n b c
L f m j( Ij n L j f L ` f
j1 j1
and
;
n
U f o U j f .
4 j1 5
From equation (7.1), L ` f n U ` f . Finally, combining the inequalities

yields that
b c b c
L f n L ` f n U ` f n U f
which completes the proof of equation (7.2).
Suppose that < and D are two subdivisions of I . Then < C D is

< and D . Because is a re¿nement of < , by the
6 b c
comparison of lower sums given in equation (7.2), L < f n L f . On the
other hand, from being a re¿nement of D , it follows that .
7
Combining the inequalities with equation (7.1) leads to equation (7.3).
3 b c
***Acceptable responses are: (1) M j n M, (2) m b a, (3) M nj1 x j x j1 ,
3 b c
(4) nj 1 M j ( I j , (5) U ` f , (6) the common re¿nement of, and
(7) U f n U f D .***
If f is a bounded function with domain I [a b] and , , [a b] is the

set of all partitions of [a b], then the Lemma assures us that
L f : + , is
bounded above by b a sup f x and
U f : + , is bounded below by
x+I
b a inf f x. Hence, by the least upper bound and greatest lower bound prop-
x+I
erties of the reals both sup
L f : + , and inf
U f : + , exist to
see that they need not be equal, note that–for the bounded function g given in Exam-
ple 7.1–we have that sup
L g : + , 0 while inf
U g : + ,
2.
De¿nition 7.1.11 Suppose that f is a function on U that is de¿ned and bounded on

the interval I [a b] and , , [a b] is the set of all partitions of [a b]. Then
the upper Riemann integral and the lower Riemann integral are de¿ned by
= b = b
f x dx inf U S f and f x dx sup L S f ,
a +, a +,
5b 5b
respectively. If a f x dx
f x dx, then f is Riemann integrable, or just
a
= b
integrable, on I , and the common value of the integral is denoted by f x d x.
a

5x 3 , for x+
T
Excursion 7.1.12 Let f x .

0 , for x + T
For each n + M, let n denote the subdivision of the interval [1 2] that con-
sists of n segments of equal length. Use
n : n + M to ¿nd an upper bound for
52 3n n n 1
1 f x d x. [Hint: Recall that k1 k .]
2
***Corresponding to each n you needed to ¿nd a useful form for U n f . Your
work should have led you to a sequence for which | the limit exists as n }*. For
1 2 n
n + M, the partition that gives the desired n is 1 1 1 1 . Then
v w n n n
j 1 j 5j
n
I1 I2 In with I j 1 1 and M j 8 leads to
n n n
21 5 5 21
U n f . Therefore, you should proved that 12 f x dx n .***
2 2n 2
It is a rather short jump from Lemma 7.1.9 to upper and lower bounds on the
Riemann integrals. They are given by the next theorem.
Theorem 7.1.13 Suppose that f is de¿ned on the interval I [a b] and m n
f x n M for all x + I . Then
= b = b
m b a n f x dx n f x dx n M b a . (7.4)
a a
Furthermore, if f is Riemann integrable on I , then

= b
m b a n f x dx n M b a . (7.5)
a
Proof. Since equation (7.5), is an immediate consequence of the de¿nition of
the Riemann integral, we will prove only equation (7.4). Let G denote the set of all
subdivisions of the interval [a b]. By Lemma 7.1.9, we have that, for ` , + G,
b c
m b a n L f ` n U f n M b a .
Since is arbitrary,
b c
m b a n L f ` n inf U f
+&
and inf U f n M b a i.e.,

+&
=
b `
c b
m b a n L f n f x dx n M b a .
a
Because ` is also arbitrary, m b a n sup L f ` and sup L f ` n

` +& ` +&
5b
a f x dx i.e.,
= b = b
m b a n f x dx n f x dx.
a a
Combining the inequalities leads to equation (7.4).

Before getting into some of the general properties of upper and lower integrals,
we are going to make a slight transfer to a more general set-up. A re-examination
of the proof of Lemma 7.1.9 reveals that it relied only upon independent application
of properties of in¿mums and supremums in conjunction with the fact that, for any
n b
3 c
partition
x0 x1 xn1 xn , x j x j1 0 and x j x j1 xn x0 . Now,
j1
given any function : that is de¿ned and strictly increasing on an interval [a b], for
any partition S
a x0 x1 xn1 xn b of [a b],
: S
: a : x0 : x1 : xn1 : xn : b t : [a b] ,
b c b c n b b c
3 b cc
: x j : x j1 0 and : x j : x j1 : b : a. Consequently,
j1
: S is a partition of
?
[: a : b]
I : I [c d] F : S t I ,
which is the “smallest” interval that contains : [a b]. The case : t t returns
us to the set-up for Riemann sums on the other hand, : [a b] need not be an
interval because : need not be continuous.
Example 7.1.14 Let I [0 3] and : t |t 2 JtK. Then :}I [0 1 C
1 5 8
[2 5 C [6 11 C
12. For the partition S 0 1 2 3 of I , : S
| } 2 4 3
1 41 82
0 2 6 12 is a partition of [0 12] which contains : I .
4 16 9
De¿nition 7.1.15 Given a function f that is bounded and de¿ned on the closed
interval I [a b], a function : that is de¿ned and monotonically increasing on
I , and a partition S
x0 x 1 xn1 xn of I withdcorresponding
e subdivision
, let M j sup f x and m j inf f x, for I j x j1 x j . Then the upper
x+I j x+I j
Riemann-Stieltjes sum of f over : with respect to the partition S, denoted by
U S f : or U f :, is de¿ned by
;
n
U S f : M j : j
j 1
and the lower Riemann-Stieltjes sum of f over : with respect to the partition S,
denoted by L S f : or L f :, is de¿ned by
;
n
L S f : m j : j
j1
b b c b cc
where : j : x j : x j1 .
b c
Replacing x j with : x j in the proof of Lemma 7.1.9 and Theorem 7.1.13
yields the analogous results for Riemann-Stieltjes sums.
Lemma 7.1.16 Suppose that f is a bounded function with domain I [a b] and :
is a function that is de¿ned and monotonically increasing on I . Let S be a partition
of I , M sup f x, and m inf f x. Then
x+I x+I
m : b : a n L S f : n U S f : n M : b : a (7.6)

and
b c b c
L S f : n L S ` f : n U S ` f : n U S f : (7.7)
for any re¿nement S ` of S. Furthermore, if < and D are any two subdivisions
of I , then
b c
L < f : n U D f : (7.8)
The bounds given by Lemma 7.1.16 with the greatest lower and least upper
bound properties of the reals the following de¿nition.
De¿nition 7.1.17 Suppose that f is a function on U that is de¿ned and bounded

on the interval I [a b], , , [a b] is the set of all partitions of [a b], and
: is a function that is de¿ned and monotonically increasing on I . Then the upper
Riemann-Stieltjes integral and the lower Riemann-Stieltjes integral are de¿ned by
= b = b
f x d: x inf U S f : and f x d: x sup L S f : ,
a +, a +,
5b 5b
respectively. If a f x d: x a f x d: x, then f is Riemann-Stieltjes in-
tegrable, or integrable with respect to : in the Riemann sense, on I , and the
= b = b
common value of the integral is denoted by f x d: x or f d:.
a a
De¿nition 7.1.18 Suppose that : is a function that is de¿ned and monotonically in-
creasing on the interval I [a b]. Then the set of all functions that are integrable
with respect to : in the Riemann sense is denoted by 4 :.
Because the proof is essentially the same as what was done for the Riemann
upper and lower integrals, we offer the following theorem without proof.
Theorem 7.1.19 Suppose that f is a bounded function with domain I [a b], : is
a function that is de¿ned and monotonically increasing on I , and m n f x n M
for all x + I . Then
= b = b
m : b : a n f d: n f d: n M : b : a . (7.9)
a a
Furthermore, if f is Riemann-Stieltjes integrable on I , then

= b
m : b : a n f x d: x n M : b : a . (7.10)
a
In elementary Calculus, we restricted our study to Riemann integrals of con-

tinuous functions. Even there we either glossed over the stringent requirement of
needing to check all possible partitions or limited ourselves to functions where some
trick could be used. Depending on how rigorous your course was, some examples of
¿nding the integral from the de¿nition might have been based on taking partitions
of equal length and using some summation formulas (like was done in Excursion
7.1.12) or might have made use of a special bounding lemma that applied to x n for
each n + M.
It is not worth our while to grind out some tedious processes in order to show
that special functions are integrable. Integrability will only be a useful concept if it
is veri¿able with a reasonable amount of effort. Towards this end, we want to seek
some properties of functions that would guarantee integrability.
Theorem 7.1.20 (Integrability Criterion) Suppose that f is a function that is bounded

on an interval I [a b] and : is monotonically increasing on I . Then f + 4 :
on I if and only if for every > 0 there exists a partition S of I such that
U S f : L S f : > (7.11)
Excursion 7.1.21 Fill in what is missing to complete the following proof.
Proof. Let f be a function that is bounded on an interval I [a b] and : be

monotonically increasing on I .
Suppose that for every > 0 there exists a partition S of I such that
U S f : L S f : >. (*)
From the de¿nition of the Riemann-Stieltjes integral and Lemma 7.1.16, we have
that
= b
L S f : n f x d: x n n .
a
1 2
It follows immediately from (*) that

= b
= b
0n f x d: x f x d: x .
a a
Since was arbitrary and the upper and lower Riemann Stieltjes integrals are con-
5 5b
stants, we conclude that ab f x d: x a f x d: x i.e., .
3
Conversely, suppose that f + 4 : and let 0 be given. For ,

5b
, [a b] the set of all partitions of [a b], a f x d: x inf U S f : and
+,
5b
a f x d: x sup L S f :. Thus, 2 0 implies that there exists a S1 +
+,
5b 5b
, [a b] such that a f x d: x U S1 f : a f x d: x and there
5b 2

exists S2 + , [a b] such that a f x d: x .
2 4
Therefore,
= =
b b
U S1 f : f x d: x and f x d: x L S2 f : .
a 2 a 2
(**)
Let S be the common re¿nement of S1 and S2 . Lemma 7.1.16, equation (7.7)

applied to (**) yields that
= =
b b
f x d: x and f x d: x .
5 a 2 a 6
2
Thus
U S f : L S f :

r 5b s
U S f : a f x d: x .
7
5
***Acceptable responses are: (1) ab f x d: x, (2) U S f :, (3) f + 4 :,
5b
(4) L S2 f : a f x d: x, (5) U S f :, (6) L S f :, and (7)
5b
a f x d: x L S f :.***
Theorem 7.1.20 will be useful to us whenever we have a way of closing the gap
between functional values on the same intervals. The corollaries give us two “big”
classes of integrable functions.
Corollary 7.1.22 If f is a function that is continuous on the interval I [a b],

then f is Riemann-Stieltjes integrable on [a b].
Proof. Let : be monotonically increasing on I and f be continuous on I . Sup-

pose that 0 is given. Then there exists an @ 0 such that [: b : a] @ .
By the Uniform Continuity Theorem, f is uniformly continuous in [a b] from
which it follows that there exists a = 0 such that
d e
1u 1) u ) + I F u ) = " f u f ) .
Let S
x 0 a x1 x n1 xn b be a partition of [a b] for which mesh S
= and, for each j, j 1 2 n, set M j sup f x and m j inf f x.
x j 1 nxnx j x j 1 nxnx j
Then M j m j n @ and
;
n b c ;
n
U S f : L S f : M j m j : j n @ : j @ [: b : a] .
j1 j1
Since 0 was arbitrary, we have that
1 0 " 2S S + , [a b] F U S f : L S f : .
In view of the Integrability Criterion, f + 4 :. Because : was arbitrary, we

conclude that f is Riemann-Stieltjes Integrable (with respect to any monotonically
increasing function on [a b]).
Corollary 7.1.23 If f is a function that is monotonic on the interval I [a b] and

: is continuous and monotonically increasing on I , then f + 4 :.
Proof. Suppose that f is a function that is monotonic on the interval I [a b]

and : is continuous and monotonically increasing on I . For 0 given, let n + M,
be such that
: b : a f b f a n.
Because : is continuous and monotonically increasing, we can b bchoose

c a bpartitioncc
S
x0 a x1 xn1 xn b of [a b] such that : j : x j : x j1
: b : a
. If f is monotonically increasing in I , then, for each j +
1 2 n,
n
b c b c
Mj sup f x f x j and m j inf f x f x j1 and
;
n b c
U S f : L S f : M j m j : j
j1
: b : a ;n b b c b cc
f x j f x j 1
n j1
: b : a
f b f a
n
b c b c
while f monotonically decreasing yields that M j f x j1 , m j f x j and
: b : a ;n b b c b cc
U S f : L S f : f x j1 f x j
n j1
: b : a
f a f b .
n
Since 0 was arbitrary, we have that
1 0 " 2S S + , [a b] F U S f : L S f : .
In view of the Integrability Criterion, f + 4 :.
Corollary 7.1.24 Suppose that f is bounded on [a b], f has only ¿nitely many
points of discontinuity in I [a b], and that the monotonically increasing function
: is continuous at each point of discontinuity of f . Then f + 4 :.
Proof. Let 0 be given. Suppose that f is bounded on [a b] and continuous

on [a b] E where E
?1 ?2 ?k is the nonempty ¿nite set of points of
discontinuity of f in [a b]. Suppose further that : is a monotonically increasing
function on [a b] that is continuous at each element of E. Because E is ¿nite
d ande
: is continuous at each ? j + E, we can ¿nd k pairwise disjoint intervals u j ) j ,
j 1 2 k, such that
>
k d e ;
k b b c b cc
Et u j ) j + [a b] and : ) j : u j `
j1 j 1
for any ` 0 furthermore, the intervals can be chosen in such a way that each
point ?m + E D a b is an element of the interior of the corresponding interval,
[u m ) m ]. Let
>
k b c
K [a b] u j ) j .
j1
Then K is compact and f continuous on K implies that f is uniformly continuous

there. Thus, corresponding to ` 0, there exists a = 0 such that
b c
1s 1t s t + K F s t = " f s f t ` .
Now, let S
x0 a x1 xn1 x n b be a partition of [a b] satis-
fying the following conditions:
b c
1 j j +
1 2 k " u j + S F ) j + S ,
b b c c
1 j j +
1 2 k " u j ) j D S 3 , and
db c e
1 p 1 j p +
1 2 n F j +
1 2 k F x p1 / u j " x p = .
Note that under the conditions established, xq1 u j implies that xq ) j . If

M sup f x, M p sup f x and m p inf f x, then for each
x+I x p1 nxnx p x p1 nxnx p
p, M p m p n 2M. Furthermore, M p m p `
as long as x p1 / u j . Using
commutativity to regroup the summation according to the available bounds yields
that
;
n b c
U S f : L S f : M j m j : j n [: b : a] ` 2M `
j1

whenever ` . Since 0 was arbitrary, from the Inte-
2M [: b : a]
grability Criterion we conclude that f + 4 :.
Remark 7.1.25 The three Corollaries correspond to Theorems 6.8, 6.9, and 6.10
in our text.
As a fairly immediate consequence of Lemma 7.1.16 and the Integrability Cri-

terion we have the following Theorem which is Theorem 6.7 in our text.
Theorem 7.1.26 Suppose that f is bounded on [a b] and : is monotonically in-

creasing on [a b].
(a) If there exists an 0 and a partition S ` of [a b] such that equation (7.11)
is satis¿ed, then equation (7.11) is satis¿ed for every re¿nement S of S ` .
(b) If equation (7.11) is satis¿ed for the partition S

x 0 a x1 dx n1 xn
e b
and, for each j, j 1 2 n, s j and t j are arbitrary points in x j1 x j ,
then
;
n n b c b cn
n f s j f t j n : j .
j1
(c) If f + 4 :, equation (7.11) is satis¿ed for the partition
S
x0 a x1 xn1 xn b
d e
and, for each j , j 1 2 n, t j is an arbitrary point in x j1 x j , then
n = b n
n; n b c n
n n
n f t j : j f x d: xn .
n j1 a n
Remark 7.1.27 Recall the following de¿nition of Riemann Integrals that you saw
in elementary calculus: Given a function f that is de¿ned on an interval I
x : a n x n b, the “R” sum for

I1 I2 In a subdivision of I is given by
;
jn
b c b c
f Gj ( Ij
j1
where G j is any element of I j . The point G j is referred to as a sampling point.

To get the “R” integral we want to take the limit over such sums as the mesh
of the partitions associated with goes to 0. In particular, if the function f
is de¿ned on I
x : a n x n b and , [a b] denotes the set of all partitions
x0 a x1 x n1 x n b of the interval I , then f is said to be “R” integrable

over I if and only if
;
jn
b cb c
lim f G j x j x j1
mesh[ab]0
j1
d e
exists for any choices of G j + x j1 x j . The limit is called the “R” integral and is
5b
denoted by a f x dx.
Taking : t t in Theorem 7.1.26 justi¿es that the old concept of an “R”
integrability is equivalent to a Riemann integrability as introduced at the beginning
of this chapter.
The following theorem gives a suf¿cient condition for the composition of a

function with a Riemann-Stieltjes integrable function to be Riemann-Stieltjes inte-
grable.
Theorem 7.1.28 Suppose f + 4 : on [a b], m n f n M on [a b], M is con-

tinuous on [m M], and h x M f x for x + [a b]. Then h + 4 : on
[a b].
Excursion 7.1.29 Fill in what is missing in order to complete the proof.
Proof. For f + 4 : on [a b] such that m n f n M on [a b] and M a

continuous function on [m M], let h x M f x for x + [a b]. Suppose that
0 is given. By the , M is uniformly continuous on
1
[m M]. Hence, there exists a = 0 such that = and
1s 1t s t + [m M] F s t = " M s M t . (!)
Because , there exists a S

x0 a x1 xn b + , [a b]
2
such that
U S f : L S f : = 2 . (!!)
For each j +
1 2 n, let M j sup f x, m j inf f x, M `j
sup h x, and m `j inf h x. From the Trichotomy Law, we know that
j b c k
A j : j +
1 2 n F M j m j =
and
j b c k
B j : j +
1 2 n F M j m j o =
are disjoint.
d e
If j + A, then u ) + x j1 x j " f u f ) =. It follows from
(!) that i.e., h u h ) . Hence, M `j m `j n . Since
3
B t
1 2 j, (!!) implies that
; ;b c
= : j n M j m j : j n =2.
j+B j+B 4
3
Because = by choice, we conclude that : j . Consequently, for
r s j+B
K sup M t, we have that M `j m `j n 2K for each j +
1 2 n and
3r `
mntnM s
M j m `j : j 2K . Combining the bounds yields that
j+B
U S h : L S h :
3 n r s
` `
M j m j : j
j 1 r s
3 ` `
3r ` `
s
M j m j : j M j m j : j
j +A j+B
n 2K .
5
Since 0 was arbitrary, the Integrability Criterion allows us to conclude that

h + 4 :.
***Acceptable responses are: (1) Uniform Continuity Theorem, (2) f + 4 :, (3)
M f u M f ) , (4) U S f :L S f :, (5) [: b : a].***
7.1.1 Properties of Riemann-Stieltjes Integrals

This section offers a list of properties of the various Riemann-Stieltjes integrals.
The ¿rst lemma allows us to draw conclusions concerning the upper and lower
Riemann-Stieltjes sums of a constant times a bounded function in relationship to
the upper and lower Riemann-Stieltjes sums of the function.
Lemma 7.1.30 Suppose that f is a function that is bounded and de¿ned on the
interval I [a b]. For k a nonzero real number and g k f , we have

!
! k inf f x , if k 0 !
! k sup f x , if k 0
x+I x+I
inf g x supg x .
x+I !
! x !
!
k sup f , if k 0 x+I k inf f x , if k 0
x+I x+I
Proof. We will prove two of the four equalities. For f a function that is de¿ned
and bounded on the interval I [a b] and k a nonzero real number, let g x
k f x.
Suppose that k 0 and that M sup f x. Then f x n M for all x + I
x+I
and
g x k f x n k M for all x + I
Hence, k M is an upper bound for g x on the interval I . If k M is not the least upper
bound, then there exists an 0 such that g x n k M for all x + I . (Here,
can be taken to be any positive real that is less than or equal to the distance between
k M and supg x.) By substitution, we have k f x n k M for all x + I . Since
x+I
k is positive, the latter is equivalent to
rs
f x n M for all x + I
k
which contradicts that M is the supremum of f over I . Therefore,
supg x k M ksup f x .
x+I x+I
Next, suppose that k 0 and that M sup f x. Now, f x n M for all
x+I
x + I implies that g x k f x o k M. Hence, k M is a lower bound for g x
on I . If k M is not a greatest lower bound, then there exists an 0, such that
g x o k M for all x + I . But, from k f x o k M and k 0, we conclude
that f x n M k for all x + I . Since k is negative, M k M
which gives us a contradiction to M being the sup f x. Therefore,
x+I
inf g x k M ksup f x .

x+I x+I
Theorem 7.1.31 (Properties of Upper and Lower Riemann-Stieltjes Integrals)

Suppose that the functions f , f 1 , and f 2 are bounded and de¿ned on the closed in-
terval I [a b] and : is a function that is de¿ned and monotonically increasing
in I .
5b
5b k a f x d: x , if k 0
!
(a) If g k f for k + U
0, then a gd:
!
5b
k a f x d: x , if k 0
5
! b
k a f x d: x , if k 0
5b
and a gd: .
k 5 b f x d: x , if k 0
!
a
(b) If h f 1 f 2 , then
5b 5b 5b
(i) a h x d: x o a f 1 x d: x a f 2 x d: x, and
5b 5b 5b
(ii) a h x d: x n a f 1 x d: x a f 2 x d: x.
(c) If f 1 x n f 2 x for all x + I , then

5b 5b
(i) a f 1 x d: x n a f 2 x d: x, and
5b 5b
(ii) a f 1 x d: x n a f 2 x d: x.
(d) If a b c and f is bounded on I `

x : a n x n c and : is monotoni-
cally increasing on I ` , then
5c 5b 5c
(i) a f x d: x a f x d: x b f x d: x, and
5c 5b 5c
(ii) a f x d: x a f x d: x b f x d: x.
of part d(i).
Proof. Suppose that a b c and that the function f isd bounded e in the
`
interval I [a c]. For any ¿nite real < and D, let G < D denote the
d numbers
e
set of all subdivisions of the interval < D . Suppose that > 0 is given. Since
= b = c
f x d: x sup L f : and f x d: x sup L f : ,
a +&[ab] b +&[bc]
there exists partitions Pn and Pm of [a b] and [b c], respectively, with correspond-
ing subdivisions n and m , such that
= b = c
> >
L n f : o f x d: x and L m f : o f x d: x .
a 2 b 2
For P Pn C Pm , let denote the corresponding subdivision of [a c]. Then
5c
a f x d: x
o
1
L n f : L m f :
2
Since 0 was arbitrary, it follows that

= c = b = c
f x d: x o f x d: x f x d: x .
a a b
Now, we want to show that the inequality can be reversed. Suppose that 0 is
given. Since
= c
f x d: x sup L f : ,
a +&[ac]
There exists a ) + G [a c] such that

=
b )
c c
L f : f x d: x .
a
For S ) the partition of [a c] that corresponds to ) , let S )) S ) C

b and ))
denote the re¿nement of ) that corresponds to S )) . From
3
b c b c
L ) f : n L )) f : .
Because )) is the union of a subdivision of [a b] and a subdivision of [b c], it
follows from the de¿nition of the lower Riemann-Stieltjes integrals that
= b = c
b c
f x d: x f x d: x o L )) f : .
a b
Therefore,
= b = c b c b c
f x d: x f x d: x o L )) f : o L ) f :
a b
.
4

= b = c = c
f x d: x f x d: x o f x d: x .
a b a
5c 5b 5c
In view of the Trichotomy Law, a f x d: x o a f x d: x b f x d: x
5b 5c 5c
and a f x d: x b f x d: x o a f x d: x yields the desired result.
***Acceptable responses include: (1) L f :,
5b 5c
(2) a f x d: x b f x d: x, (3) Lemma 7.1.16, (4) same as completion
for (2).***
Given Riemann-Stieltjes integrable functions, the results of Theorem 7.1.31
translate directly to some of the algebraic properties that are listed in the follow-
ing Theorem.
Theorem 7.1.33 (Algebraic Properties of Riemann-Stieltjes Integrals) Suppose

that the functions f , f 1 , f 2 + 4 : on the interval I [a b].
(a) If g x k f x for all x + I , then g + 4 : and

= b = b
g x d: x k f x d: x .
a a
(b) If h f 1 f 2 , then f 1 f 2 + 4 : and

= b = b = b
h x d: x f 1 x d: x f 2 x d: x .
a a a
(c) If f 1 x n f 2 x for all x + I , then

= b = b
f 1 x d: x n f 2 x d: x .
a a
(d) If the function f + 4 : also on I `

x : b n x n c, then f is Riemann-
Stieltjes integrable on I C I ` and
= c = b = c
f x d: x f x d: x f x d: x .
a a b
(e) If f x n M for x + I , then

n= b n
n n
n f x d: x n n M [: b : a] .
n n
a
(f) If f + 4 : ` on I , then f + 4 : : ` and

= = =
b b `
c b b
fd :: f x d: x f x d: ` x .
a a a
(g) If c is any positive real constant, then f + 4 c: and

= b = b
f d c: c f x d: x .
a a
Remark 7.1.34 As long as the integrals exist, the formula given in (d) of the Corol-
lary holds regardless of the location of b i.e., b need not be between a and c.
Remark 7.1.35 Since a point has no dimension (that is, has length 0), we note that
= a
f x d: x 0 for any function f .
a
Remark 7.1.36 If we think of the de¿nition of 5the Riemann-Stieltjes integrals as

b
taking direction into account (for example, with a f x d: x we had a b and
took the sums over subdivisions as we were going from a to b), then it makes sense
to introduce the convention that
= a = b
f x d: x f x d: x
b a
for Riemann-Stieltjes integrable functions f .

The following result follows directly from the observation that corresponding
to each partition of an interval we can derive a partition of any subinterval and vice
versa.
Theorem 7.1.37 (Restrictions of Integrable Functions) If the function f is (Rie-

mann) integrable on an interval I , then f I ` is integrable on I ` for any subinterval
I ` of I .
Choosing different continuous functions for M in Theorem 7.1.28 in combina-

tion with the basic properties of Riemann-Stieltjes integrals allows us to generate
a set of Riemann-Stieltjes integrable functions. For example, because M1 t t 2 ,
M2 t t, and M3 t < t D for any real constants < and D, are continuous
on U, if f + 4 : on an interval [a b], then each of f 2 , f , and < f D will
be Riemann-Stieltjes integrable with respect to : on [a b]. The proof of the next
theorem makes nice use of this observation.
Theorem 7.1.38 If f + 4 : and g + 4 : on [a b], then f g + 4 :.
Proof. Suppose that f + 4 : and g + 4 : on [a b]. From the Algebraic
Properties of the Riemann-Stieltjes Integral, it follows that f g + 4 : on
[a b] and f g + 4 : on [a b]. Taking M t t 2 in Theorem 7.1.28 yields
that f g2 and f g2 are also Riemann-Stieltjes integrable with respect to :
on [a b]. Finally, the difference
4 f g f g2 f g2 + 4 : on [a b]
as claimed.
Theorem 7.1.39 If f + 4 : on [a b], then f + 4 : and

n= b n = b
n n
n f x d: x nn f x d: x .
n n
a a
Proof. Suppose f + 4 : on [a b]. Taking M t 5 t in Theorem 7.1.28

yields that 5 f + 4 : on [a b]. Choose < 1, if f x d: x o 0 and
< 1, if f x d: x n 0. Then
n= n =
n b n b
n f x d: xnn < f x d: x and < f x n f x for x + [a b] .
n
a a
It follows from Algebraic Properties of the Riemann-Stieltjes Integrals (a) and (c)
that
n= b n = b = b = b
n n
n n
f x d: xn < f x d: x < f x d: x n f x d: x .
n
a a a a
One glaring absence from our discussion has been speci¿c examples of ¿nding
the integral for integrable functions using the de¿nition. Think for a moment or
so about what the de¿nition requires us to ¿nd: First, we need to determine the
set of all upper Riemann-Stieltjes sums and the set of all lower Riemann-Stieltjes
sums this is where the subdivisions of the interval over which we are integrating
range over all possibilities. We have no uniformity, no simple interpretation for the
suprema and in¿ma we need, and no systematic way of knowing when we “have
checked enough” subdivisions or sums. On the other hand, whenever we have
general conditions that insure integrability, the uniqueness of the least upper and
greatest lower bounds allows us to ¿nd the value of the integral from considering
wisely selected special subsets of the set of all subdivisions of an interval.
The following result offers a suf¿cient condition under which a Riemann-Stieltjes
integral is obtained as a point evaluation. It makes use of the characteristic function.
Recall that, for a set S and A t S, the characteristic function N A : S
0 1 is
de¿ned by

1 , if x + A
N A x

0 , if x + S A
In the following, N0* denotes the characteristic function with S U and A

0 * i.e.,

1 , if x 0
N0* x .

0 , if x n 0
Lemma 7.1.40 Suppose that f is bounded on [a b] and continuous at s + a b.
If : x N0* x s, then
= b
f x d: x f s .
a
Proof. For each S

x0 a x1 xn1 xn d b be ancarbitrary partition for
[a b], there exists a j +
1 2 n such that s + x j1 x j . From the de¿nition
of :, we have that : xk 0 for each k +
1 2 j 1 and : xk 1 for each
k +
j n. Hence,

1 , if k j
:k : xk : xk1 ,

0 , if k +
1 2 j 1 C
j 1 n
from which we conclude that
U S f : sup f x and L S f : inf f x .

b c
Since f is continuous at s and x j x j1 n mesh S, sup f x s and
x j 1 nxnx j
5b
inf f x s as mesh S 0. Therefore, a f x d: x f s.
x j 1 nxnx j
If the function f is continuous on an interval [a b], then Lemma 7.1.40 can be
extended to a sequence of points in the interval.
Theorem 7.1.41 Suppose the sequence of nonnegative real numbers

cn * n1 is
3
*
such that cn is convergent,
sn *
n1 is a sequence of distinct points in a b,
n1
3
*
and f is a function that is continuous on [a b]. If : x cn N0* x sn ,
n1
then
= b ;
*
f x d: x cn f sn .
a n1
Proof. For u ) + a b such that u ), let Su

n + M : a sn n u and
T)
n + M : a sn n ). Then
; ;
: u cn n cn : )
n+Su n+T)
from which we conclude that : is monotonically increasing. Furthermore, : a

3
*
0 and : b cn .
n1
3
*
Let 0 be given. Since cn is convergent, there exists a positive
n1
integer K such that
;
*

cn
nK 1
M
3
K
where M sup f x. Let :1 x cn N0* x sn and :2 x
x+[ab] n1
3
*
cn N0* x sn . It follows from Lemma 7.1.40 that
nK 1
= b ;
K
f x d:1 x cn f sn
a j1

while :2 b :2 a yields that
M
n= b n
n n
n f x d: x n .
n 2 n
a
Because : :1 :2 , we conclude that

n= n
n b ;K n
n n
n f x d: x cn f s n n .

n a n1
n
5b 3*
Since 0 was arbitrary, a f x d: x n1 cn f sn .
Theorem 7.1.42 Suppose that : is a monotonically increasing function such that

: ) + 4 on [a b] and f is a real function that is bounded on [a b]. Then f + 4 :
if and only if f : ) + 4. Furthermore,
= b = b
f x d: x f x : ) x dx.
a a
of Theorem 7.1.42.
Proof. Suppose that 0 is given. Since : ) + 4 on [a b], by the Integrability

Criterion, there exists a partition S
x0 x1 xn of [a b] such that
b c
U S : ) (7.12)
1
M
where M sup f x. Furthermore, d from e the Mean-Value Theorem, for each
j +
1 2 n there exists a t j + x j1 x j such that
b c
: j : ) t j x j . (7.13)
2
d e
By Theorem 7.1.26(b) and (7.12), for any s j + x j1 x j , j +
1 2 n
;
n n b c b cn
n: ) s j : ) t j n x j . (7.14)
j1
With this set-up, we have that
;
n b c ;
n
f s j : j
j1 j1 3
and
n3 b c 3n b c )b c n
n n n
n j 1 f s j : j j1 f s j : s j x jn
n n
n 3 b c b c n
n n ) n
n j1 f s j : s j x j n
n 4
n
n3 b cd )b c b ce n
n n ) n
n j1 f s j : t j : s j x j n
n3 d b c b ce n
n n
n M n nj1 : ) t j : ) s j x j n .
That is,
n n
n; n b c ;n b c )b c n
n n
n f s j : j f s j : s j x j n (7.15)
n j1 j1
n
d e
for any choice of points s j + x j1 x j , j 1 2 n. Then
;
n b c b c
f s j : j n U S f : )
j1
and
b c
U S f : n U S f : ) .
Equation (7.15) also allows us to conclude that

b c
U S f : ) n U S f : .
Hence,
n b cn
nU S f : U S f : ) n n . (7.16)
Since S was arbitrary, if follows that (7.16) holds for all S + , [a b], the set of all
partitions of [a b]. Therefore,
n= n
n b n
n n
n f x d: x n n .
n a 5
n
Because 0 was arbitrary, we conclude that, for any function f that is bounded
on [a b],
= b = b
a a
Equation (7.15) can be used to draw the same conclusion concerning the compara-
ble lower Riemann and Riemann-Stieltjes integrals in order to claim that
= b = b
a a
The combined equalities leads to theb desired

c conclusion.
b c b c b c b c
***Acceptable responses are: (1) L S : ) , (2) : x j : x j1 (3) f s j : ) t j x j
3 b c b c 5b
(4) nj 1 f s j : ) t j x j , (5) a f x : ) x dx.***
Recall that our original motivation for introducing the concept of the Riemann
integral was adapting formulas such as A l *, d r t and m d l to more
7.2. RIEMANN INTEGRALS AND DIFFERENTIATION 307
general situations the Riemann integral allow us to replace one of the “constant
dimensions” with functions that are at least bounded where being considered. The
Riemann-Stieltjes integral allows us to replace both of the “constant dimensions”
with functions. Remark 6.18 on page 132 of our text describes a speci¿c example
that illustrates to Àexibility that has been obtained.
The last result of this section gives us conditions under which we can transfer
from one Riemann-Stieltjes integral set-up to another one.
Theorem 7.1.44 (Change of Variables) Suppose that M is a strictly increasing con-

tinuous function that maps an interval [A B] onto [a b], : is monotonically in-
creasing on [a b], and f + 4 : on [a b]. For y + [A B], let ; y : M y
and g y f M y. Then g + 4 ; and
= B = b
g y d; y f x d: x .
A a
Proof. Because M is strictly increasing and continuous, each partition S

b xcn + , [a b] if and only if T
y
x0 x 1 yn + bd
bd0 y1 ec , [A B] where
ec
x j M y j for each j +
0 1 n. Since f x j 1 x j g y j1 y j for
each j, it follows that
U T g ; U S f : and L T g ; L S f : .
The result follows immediately from the Integrability Criterion.
7.2 Riemann Integrals and Differentiation

When we restrict ourselves to Riemann integrals, we have some nice results that
allow us to make use of our knowledge of derivatives to compute integrals. The
¿rst result is both of general interest and a useful tool for proving some to the
properties that we seek.
Theorem 7.2.1 (Mean-Value Theorem for Integrals) Suppose that f is continu-

ous on I [a b]. Then there exists a number G in I such that
= b
f x dx f G b a .
a
Proof. This result follows directly from the bounds on integrals given by Theo-
rem 7.1.13 and the Intermediate Value Theorem. Since f is continuous on [a b], it
is integrable there and, by Theorem 7.1.13,
= b
m b a n f x dx n M b a
a
where m inf f x min f x f x0 for some x0 + I and M sup f x
x+I x+I
5b x+I
f x dx
max f x f x1 for some x1 + I . Now, A a is a real number
x+I b a
such that m n A n M. By the Intermediate Value Theorem, f x0 n A n f x1
implies that there exists a G + I such that f G A.
The following two theorems are two of the most celebrated results from integral
calculus. They draw a clear and important connection between integral calculus
and differential calculus. The ¿rst one makes use of the fact that integrability on an
interval allows us to de¿ne a new function in terms of the integral. Namely, if f is
Riemann integrable on the interval [a b], then, by the Theorem on Restrictions of
Integrable Functions, we know that it is integrable on every subinterval of [a b]. In
particular, for each x + [a b], we can consider
= x
f : x f t dt.
a
This function is sometimes referred to as the accumulation of f –probably as

a natural consequence of relating the process of integration to ¿nding the area be-
tween the graph of a positive function and the real axis. The variable t is used as
the dummy variable because x is the argument of the function. The accumulation
function is precisely the object that will allow us to relate the process of integration
back to differentiation at a point.
Theorem 7.2.2 (The First Fundamental Theorem of Calculus) Suppose that f +

4 on I [a b]. Then the function F given by
= x
F x f t dt
a
is uniformly continuous on [a b]. If f is continuous on I , then F is differentiable

in a b and, for each x + a b, F ) x f x.
Proof. Suppose that u ) + [a b]. Without loss of generality we can assume
that u ). Then, from the Algebraic Properties of Riemann-Stieltjes Integrals (d)
and (e),
n= ) n
n n
n
F ) F u n f t dt nn n M u )
u
where M sup f t. Thus,

t+I
r s
1 0 1u 1) u ) + I F u ) = " F ) F u i.e.,
M
F is uniformly continuous on I .
For the second part, suppose f is continuous on [a b] and x + a b. Then
there exists =1 such that
x h : h =1 t a b. Since f is continuous it is in-
5 xh
tegrable on every subinterval of I , for h =1 , we have that each of a f t dt,
5x 5 xh
a f t dt, and x f t dt exists and
= xh = x = xh
f t dt f t dt f t dt
a a x
Consequently, for any h, with h =1 , we have that

= xh
F x h F x f t dt.
x
By the Mean-Value Theorem for Integrals, there exists Gh with x Gh =1 such

that
= xh
f t dt f Gh h.
x
Hence, for h =1 ,
F x h F x
f Gh
h
where x Gh =1 . Now, suppose that > 0 is given. Since f is continuous
at x, there exists a =2 0 such that f * f x > whenever * x =2 .
Choose = min
=1 =2 . Then, for h =, we have
n n
n F x h F x n
n f x n f Gh f x >
n h n
Since > 0 was arbitrary, we conclude that

F x h F x
lim f x i.e.,
h0 h
F ) x f x. Since x + a b was arbitrary, we conclude that F is differentiable
on the open interval a b.
Theorem 7.2.3 (The Second Fundamental Theorem of Calculus) If f + 4 on
I [a b] and there exists a function F that is differentiable on [a b] with F ) f
, then
= b
f t dt F b F a .
a
Excursion 7.2.4 Fill in what is missing in order to complete the following proof.
Proof. Suppose that 0 is given. For f + 4, by the , we can
1
choose a partition S
x0 x1 xn of [a b] such that U S f L S f
d By thee Mean-Value Theorem, for each j +
1 2 n there is a point t j +
.
x j1 x j such that
b c b c b c
F x j F x j1 F ) t j x j .
2
Hence,
;
n b c
f t j x j
j1 3
n n
n3 b c 5b n
n n
n
On the other hand, from Theorem 7.1.26(c), n f t j x j f x dx n .
n j1 a n
Therefore,
n n
n =b n
n n
n f x dx nn .
n
n 3 n
a
5b
Since 0 was arbitrary, a f t dt F b F a. b c
***Acceptable responses are: (1) Integrability Criterion, (2) f t j x j , (3) F b
F a.***
Remark 7.2.5 The statement of the First Fundamental Theorem of Calculus differs
from the one that you had in elementary Calculus. If instead of taking f to be
integrable in I [a b], we take
5 x f to be integrable on an open interval containing
I , we can claim that G x a f t dt is differentiable on [a b] with G ) x
f x on [a b]. This enables us to offer a slightly different proof for the Second
Fundamental Theorem of Calculus. Namely, it follows that if F is any antiderivative
for f then F G c for some constant c and we have that
= b
F b F a [G b c] [G a c] [G b G a] f t dt.
a
Remark 7.2.6 The Fundamental Theorems of Calculus give us a circumstance un-

der which ¿nding the integral of a function is equivalent to ¿nding a primitive or
antiderivative of a function. When f is a continuous5function, we conclude that it
has a primitive and denote the set of all primitives by f x dx to ¿nd the de¿nite
5b
integral a f x dx, we ¿nd any primitive of f , F, and we conclude that
= b n
f x dx F x nab F b F a .
a
7.2.1 Some Methods of Integration

We illustrate with two methods, namely substitution and integration-by-parts. The
theoretical foundation for the method of substitution is given by Theorem 7.1.42
and the Change of Variables Theorem.
Theorem 7.2.7 Suppose that the function f is continuous on a segment I , the func-
du
tions u and are continuous on a segment J , and the range of u is contained in
dx
I . If a b + J , then
= b = ub
)
f u x u x dx f u du
a ua
Proof. By the First Fundamental Theorem of Calculus, for each c + I , the

function
= u
F u f t dt
c
is differentiable with F ) u f u for u + I with c n u. By the Chain

Rule, if G x F u x, then G ) x F ) u x u ) x. Hence, G ) x
f u x u ) x. Also, G ) is continuous from the continuity of f , u, and u ) . It fol-
lows from the Second Fundamental Theorem of Calculus and the de¿nition of G
that
= b = b
)
f u x u x dx G ) x dx G b G a F u b F u a
a a
From the de¿nition of F, we conclude that

= ub = ua = ub
F u b F u a f t dt f t dt f t dt.
c c ua
The theorem follows from the transitivity of equals.
Example 7.2.8 Use of the Substitution Method of Integration to ¿nd

= r
H4 Hs r Hs
cos 2
3t sin 3t dt
0 4 4
r Hs
(a) Take u cos 3t
4
r Hs
(b) Take u cos3 3t .
4
The other common method of integration with which we should all be familiar
is known as Integration by Parts. The IBP identity is given by
= =
ud) u) )du
for u and ) differentiable and follows from observing that d u) ud) )du,
which is the product rule in differential notation. This enables us to tackle in-
tegrands that “are products of functions not related by differentiation” and some
special integrands, such as the inverse trig functions.
Example 7.2.9 Examples of the use of the Integration-by-Parts method of integra-

tion.
5 T
1. Find x3 1 x 2 dx.
5
2. Find arctan x dx
5
3. Find e2x sin 3x dx
7.2.2 The Natural Logarithm Function

The Fundamental Theorems enable us to ¿nd integrals by looking for antideriva-5
tives. The formula x n ) nx n1 for n an integer leads us to conclude that x k dx
x k 1
k1 C for any constant C as long as k 1 / 0. So we can’t use a simple formula
to determine
= b
1
dx,
a x
though we know that it exists for any ¿nite closed interval that does not contain 0
because x 1 is continuous in any such interval. This motivates us to introduce a
notation for a simple form of this integral.
De¿nition 7.2.10 The natural logarithm function, denoted by ln , is de¿ned by the

formula
= x
1
ln x dt, for every x 0.
1 t
As fairly immediate consequences of the de¿nition, we have the following Prop-

erties of the Natural Logarithm Function. Let f x ln x for x 0 and suppose
that a and b are positive real numbers. Then, the following properties hold:
1. ln ab ln a ln b,
2. ln ab ln a ln b,
3. ln 1 0,
7.3. INTEGRATION OF VECTOR-VALUED FUNCTIONS 315
4. ln a r r ln a for every rational number r ,

1
5. f ) x ,
x
6. f is increasing and continuous on I
x : 0 x *,
1
7. n ln 2 n 1,
2
8. ln x * as x *,
9. ln x * as x 0 , and
10. the range of f is all of U.
Remark 7.2.11 Once we have property (6), the Inverse Function Theorem guaran-
tees the existence of an inverse function for ln x. This leads us back to the function
ex .
7.3 Integration of Vector-Valued Functions

Building on the way that limits, continuity, and differentiability from single-valued
functions translated to vector-valued functions, we de¿ne Riemann-Stieltjes inte-
grability of vector-valued functions by assignment of that property to the coordi-
nates.
De¿nition 7.3.1 Given a vector-valued (n-valued) function f f 1 f 2 f n

from [a b] into Un where the real-valued functions f 1 f 2 f n are bounded on
the interval I [a b] and a function : that is de¿ned and monotonically increas-
ing on I , f is Riemann-Stieltjes
b integrable with respect
c to : on I , written f + 4 :,
if and only if 1 j j +
1 2 n " f j + 4 : . In this case,
= b t= b = b = b u
f x d: x f 1 x d: x f 2 x d: x f n x d: x .
a de f a a a
Because of the nature of the de¿nition, any results for Riemann-Stieltjes inte-
grals that involved “simple algebraic evaluations” can be translated to the vector-
valued case.
Theorem 7.3.2 Suppose that the vector-valued functions f and g are Riemann-
Stieltjes integrable with respect to : on the interval I [a b].
(a) If k is a real constant, then kf + 4 : on I and

= b = b
kf x d: x k f x d: x .
a a
(b) If h f g, then h + 4 : and

= b = b = b
h x d: x f x d: x g x d: x .
a a a
(c) If the function f + 4 : also on I `

x : b n x n c, then f is Riemann-
Stieltjes integrable with respect to : on I C I ` and
= c = b = c
f x d: x f x d: x f x d: x .
a a b
(d) If f + 4 : ` on I , then f + 4 : : ` and

= = =
b b `
c b b
fd : : f x d: x f x d: ` x .
a a a
(e) If c is any positive real constant, then f + 4 c: and

= b = b
fd c: c f x d: x .
a a
Theorem 7.3.3 Suppose that : is a monotonically increasing function such that

: ) + 4 on [a b] and f is a vector-valued function that is bounded on [a b]. Then
f + 4 : if and only if f: ) + 4. Furthermore,
= b = b
a a
Theorem 7.3.4 Suppose that f f 1 f 2 f n + 4 on I [a b].

(a) Then the vector-valued function F given by

t= x = x = x u
F x f 1 t dt f 2 t dt f n t dt for x + I
a a a
is continuous on [a b]. Furthermore, if f is continuous on I , then F is differ-

entiable in a b and, for each x + a b, F) x f x
f 1 x f 2 x f n x.
(b) If there exists a vector-valued function G on I that is differentiable there with

G) f , then
= b
f t dt G b G a .
a
On the other hand, any of the results for Riemann-Stieltjes integrals of real-
valued functions that involved inequalities require independent consideration for
formulations that might apply to the vector-valued situation while we will not
pursue the possibilities here, sometimes other geometric conditions can lead to
analogous results. The one place where we do have an almost immediate carry
over is with Theorem 7.1.39 because the inequality involved the absolute value
which generalizes naturally to an inequality in terms of the Euclidean metric. The
generalization–natural as it is–still requires proof.
Theorem 7.3.5 Suppose that f : [a b] Un and f + 4 : on [a b] for some :
that is de¿ned and monotonically increasing on [a b]. Then f + 4 : and
n= b n = b
n n
n f x d: x nn f x d: x . (7.17)
n n
a a
Excursion 7.3.6 Fill in what is missing in order to complete the following proof.
Proof. Suppose that f f 1 f 2 f n + 4 : on I [a b]. Then

T
f x f 12 x f 22 x f n2 x o 0 for x + I (7.18)
and, because f is on I , there exists M 0 such that f I

1 b c
f x : x + I t [0 M]. Since 1 j j +

1 2 n " f j + 4 : and the
function M t t 2 is continuous on U, by ,

r s 2
1 j j +
1 2 n " f j2 + 4 : . From Algebraic Property (b) of the Riemann-
T
Stieltjes integral, f 12 x f 22 x f n2 x + 4 :. Taking M ` t t for
t o 0 in Theorem 7.1.28 yields that f .
3
Since (7.17) is certainly satis¿ed if f, we assume that f / 0. For each
5b 5b
j +
1 2 n, let * j a f j x d: x and set w a f x d: x. Then
;n ;n = b = b ;
n

w2 *2j *j f j t d: t * j f j t d: t .
j1 j1 a a j1
From Schwarz’s inequality,

;
n
* j f j t n f t for t + [a b] . (7.19)
j1 4
3
Now nj1 * j f j t and w f t are real-valued functions on [a b] that are in
4 :. From (7.19) and Algebraic Property (c) of Riemann-Stieltjes integrals, it
follows that
= b ;
n
= b = b
w
2
* j f j t d: t n w f t d: t w f t d: t .
a j1 a a
5b 5b
Because w / 0, w2 n w a f t d: t implies that w n a f t d: t
which is equivalent to equation 7.17).
***Acceptable responses are: (1) bounded, (2) Theorem 7.1.28, (3) + 4 :, (4)
w.***
7.3.1 Recti¿able Curves

As an application of Riemann-Stieltjes integration on vector-valued functions we
can prove a result that you assumed when you took elementary vector calculus.
Recall the following de¿nition.
De¿nition 7.3.7 A continuous function < from an interval [a b] into Un is called
a curve in Un or a curve on [a b] in Un if < is one-to-one, then < is called an arc,
and if < a ;, then < is a closed curve.
Remark 7.3.8 In the de¿nition of curve, we want to think of the curve as the actual
mapping because the associated set of points in Un is not uniquely determined
b 2 c by
a particular mapping. As a simple example, <1 t t t and <2 t t t are
two different mappings that give the same associated subset of U2 .
Given a curve < on [a b], for any partition of [a b],
S
x0 a x 1 xm1 xm b
let
;
m n b c b cn
S < n< x j < x j1 n .
j1
Then S < is the length of a polygonal path having vertices

< x 0 < x1 < xm which, if conditions are right, gives an approximation
for the length of the curve < . For , [a b] the set of all partitions of [a b], it is
reasonable to de¿ne the length of a curve < as
< sup S <

+,[ab]
if < *, then < is said to be recti¿able.

In various applications of mathematics integrating over curves becomes impor-
tant. For this reason, we would like to have conditions under which we can deter-
mine when a given curve is recti¿able. We close this chapter with a theorem that
tells us a condition under which Riemann integration can be used to determine the
length of a recti¿able curve.
Theorem 7.3.9 Suppose that < is a curve on [a b] in Un . If < ) is continuous on

[a b], then < is recti¿able and
= b
n ) n
< n< tn dt.
a
Proof. Suppose that < is a curve on [a b] in Un such that < ) isd continuous.
e From
the Fundamental Theorem of Calculus and Theorem 7.1.39, for x j1 x j t [a b],
n= n =
n b c b cn nn x j ) n
n
xj n n
n< x j < x j1 n n < t dt n n n< ) tn dt.
n x j 1 n x j 1
Hence, for any partition of [a b], S

x0 a x1 xm1 x m b,
= b
n ) n
S < n n< tn dt
a

= b n ) n
< n n< tn dt (7.20)
a
Let 0 be given. By the Uniform Continuity Theorem, < ) is uniformly

continuous on [a b]. Hence, there exists a = 0 such that
n n
s t = " n< ) s < ) tn . (7.21)
2 b a
Choose S
x0 a x1 xm1 xm b + , [a b] be such that mesh S =.
It follows from (7.21) and the (other) triangular inequality that
d e n n n b cn
t + x j1 x j " n< ) tn n n< ) x j n .
2 b a
Thus,
5 xj n ) n n ) b cn
n< tn dt n n< x j n x j x j
x j 1 2 b a
n5 e nn
n x d b c
n x jj1 < ) t < x j < ) t dt n x j
n5 n n5 2 b n a
n x n n x d b c e n
n n x jj1 < ) t dt n n x j j1 < x j < ) t dt n x j
t u 2 b a
n b c b cn
n n< x j < x j1 n 2 x j
2 b a
n b c b cn
n< x j < x j1 n x j .
b a
Summing the inequalities for j 1 2 m yields that
= b
n ) n
n< tn dt n S < . (7.22)
a
5bn n
Since 0 was arbitrary, we conclude that a n< ) tn dt n S < .
Combining the inequalities (7.20) and (7.21) leads to the desired conclu-
sion.
7.4. PROBLEM SET G 321
7.4 Problem Set G

1. Let f x x 2 , g x J2xK, and, for n + M, Sn denote the partition of
[0 2] that subdivides the interval into n subintervals of equal length. Find
each of the following.
(a) U S3 f
(b) U S5 g
(c) L S4 f
(d) L S6 g
2. For f x 2x 2 1, : t t J3tK and the subdivision of [0 1]

consisting of 4 subintervals of equal length, ¿nd U f : and L f :.
v w | }
1 1 1 1
3. For f x 3x in 1 , : t t, and S 0 1 , ¿nd
2 2 4 2
U S f : and L S f :.
4. Suppose that the function f in bounded on the interval [a b] and g k f for
a ¿xed negative real number k. Prove that supg x k inf f x.
x+I x+I
5. Suppose that the function f in bounded on the interval I [a b] and g k f

for a ¿xed negative integer k. Show that
= b = b
g x dx k f x dx.
a a
6. Suppose that the functions f , f 1 , and f 2 are bounded and de¿ned on the
closed interval I [a b] and : is a function that is de¿ned and monotoni-
cally increasing in I . Prove each of the following:
5b 5b 5b
(a) If h f 1 f 2 , then a h x d: x o a f 1 x d: x a f 2 x d: x
5b 5b 5b
(b) If h f 1 f 2 , then h x d: x n a f 1 x d: x a f 2 x d: x
a
5b 5b
(c) If f 1 x n f 2 x for all x + I , then a f 1 x d: x n a f 2 x d: x
5b 5b
(d) If f 1 x n f 2 x for all x + I , then a f 1 x d: x n a f 2 x d: x.
(e) If a b c and f is bounded on I `

x : a n x n c and : is
monotonically increasing on I ` , then
= c = b = c
f x d: x f x d: x f x d: x
a a b
7. Suppose that f is a bounded function on I [a b]. Let M sup f x,

x+I
m inf f x, M ` sup f x, and m ` inf f x.
x+I x+I x+I
(a) Show that M ` m ` n M m.

(b) If f and g are nonnegative bounded functions on I , N supg x, and
x+I
n inf g x, show that
x+I
sup f g x inf f g x n M N mn.

x+I x+I
8. Suppose that f is bounded and Riemann integrable on I [a b].
(a) Prove that f is Riemann integrable on I .

n5 n 5
n b n b
(b) Show that n a f x dx n n a f x dx.
9. Suppose that f and g are nonnegative, bounded and Riemann integrable on

I [a b]. Prove that f g is Riemann integrable on I .
| }
j
10. Let A n : n j + M F j 2 F 2 0 j and
n
2
1
!
n , if x+A
f x 2
!

0 , if x + [0 1] A
Is f Riemann integrable on [0 1]. Carefully state and prove your conclusion.
11. Let f x x 2 and : t J3tK where JK denotes the greatest integer
function.
7.4. PROBLEM SET G 323
| }
2j
(a) For the partition S x j : j 0 1 2 3 with associated sub-
3
division
I1 I2 I3 , ¿nd U S f :.
| } | }
` j 2 2j
(b) If S u j : j 0 1 2 C u j2 : j 1 2 3 4 5 6
3 3 9
and ` denotes the associated subdivision of [0 2], then S ` is a re-
¿nement of S. For each k +
1 2 3, let k be the subdivision
of Ik consisting of the elements of ` that are contained in Ik . Find
L 2 f :.
12. For a b, let F [a b] denote the set of real-valued functions that are con-
tinuous on the interval I [a b]. For f g + F [a b], set
= b
d f g f x g x dx.
de f a
Prove that F [a b] d is a metric space.
13. If f is monotonically increasing on an interval I [a b], prove that f is

Riemann integrable. Hint: Appeal to the Integrability Criterion.
;
n
14. For nonzero real constants c1 c2 cn , let f x c j JxK N[ j j1 x,
j1
where JK denotes the greatest integer function and N denotes the character-
istic function on U. Is f Riemann integrable on U? Carefully justify the
position taken if yes, ¿nd the value of the integral.
15. Prove that if a function f is “R” integrable (see Remark 7.1.27) on the interval
I [a b], then f is Riemann integrable on I .
16. Suppose that f and g are functions that are positive and continuous on an
interval I [a b]. Prove that there is a number ? + I such that
= b = b
f x g x dx f ? g x dx.
a a
17. For a b, let I [a b]. If the function f is continuous on I

c, for a
¿xed c + a b, and bounded on I , prove that f is Riemann integrable on I .
18. Suppose that f in integrable on I [a b] and

2m 2M m 0 F M 0 F 1x x + I " m n f x n M .
1
Prove + 4 on I .
f x
19. For f x x 2 2x, verify the Mean-Value Theorem for integrals in the
interval [1 4].
= x3
sin
b c
t 3 1
20. Find e dt.
1
= ex
2
21. For x 0, let G x T sin9 3t dt. Make use of the First Fundamental
x
Theorem of Calculus and the Chain Rule to ¿nd G ) x. Show your work
carefully.
22. Suppose that f + 4 and g + 4 on I [a b]. Then each of f 2 , g 2 , and f g
are Riemann integrable on I . Prove the Cauchy-Schwarz inequality:
t= b u2 t= b u t= b u
f x g x dx n f x dx
2
g x dx .
2
a a a
5b 5b 5b
[Note that, for : a f 2 x dx, ; a f x g x dx, and < a g 2 x dx,
: 2 x 2;x < is nonnegative for all real numbers x.]
5 x dt
23. For f x ln x 1 for x 0 and a and b positive real numbers, prove
t
each of the following.
(a) ln ab ln a ln b,

(b) ln ab ln a ln b,
(c) ln ar r ln a for every rational number r ,
1
(d) f ) x ,
x
(e) f is increasing and continuous on I
x : 0 x *, and
1
(f) n ln 2 n 1,
2
Chapter 8
Sequences and Series of Functions
Given a set A, a sequence of elements of A is a function F : M A rather than

using the notation F n for the elements that have been selected from A, since the
domain is always the natural numbers, we use the notational convention an F n
and denote sequences in any of the following forms:

an *
n1
an n+M or a1 a2 a3 a4 .
Given any sequence

ck *
k1 of elements of a set A, we have an associated sequence
of nth partial sums
;
n

sn *
n1 where sn ck
k1
3*
the symbol k1 ck is called a series (or in¿nite series). Because the function
11
g x x 1 is a one-to-one correspondence from M into MC
0, i.e., g : M
MC
0, a sequence could have been de¿ned as a function on MC
0. In our dis-
cussion of series, the symbolic descriptions of the sequences of nth partial sums
usually will be generated from a sequence for which the ¿rst subscript is 0. The
notation always makes the indexing clear, when such speci¿city is needed.
Thus far, our discussion has focused on sequences and series of complex (and
real) numbers i.e., we have taken A F and A U. In this chapter, we take A to
be the set of complex (and real) functions on F (and U).
325
326 CHAPTER 8. SEQUENCES AND SERIES OF FUNCTIONS
8.1 Pointwise and Uniform Convergence

The ¿rst important thing to note is that we will have different types of convergence
to consider because we have “more variables.” The ¿rst relates back to numerical
sequences and series. We start with an example for which the work was done in
Chapter 4.
Example 8.1.1 For each n + M, let f n z z n where z + F. We can use results
obtained earlier to draw some conclusions about the convergence of
f n z* n1 . In
Lemma 4.4.2, we showed that, for any ¿xed complex number z 0 such that z 0 1,
lim z n 0. In particular, we showed that for z 0 , 0 z 0 1, if 0, then
n* 0
taking

!
! 1 , for o 1

M M z 0 z { .
!
! ln
, for 1
ln z 0
n n
yields that nz 0n 0n for all n M. When z 0 0, we have the constant
sequence. In offering this version of the statement of what we showed, I made a
“not so subtle” change in format namely, I wrote the former M and M z 0 .
The change was to stress that our discussion was tied to the ¿xed point. In terms
of our sequence
f n z* n1 , we can say that for each ¿xed point z 0 + P
z + F : z 1,
f n z 0 *
n1 is convergent to 0. This gets us to some new termi-
nology: For this example, if f z 0 for all z + F, then we say that
f n z 0 *n1
is pointwise convergent to f on P.
It is very important to keep in mind that our argument for convergence at each
¿xed point made clear and de¿nite use of the fact that we had a point for which a
known modulus was used in ¿nding an M z 0 . It is natural to ask if the pointwise
dependence was necessary. We will see that the answer depends on the nature of
the sequence. For the sequence given in Example 8.1.1, the best that we will be
able to claim over the set P is pointwise convergence. The associated sequence of
nth partial sums for the functions in the previous example give us an example of a
sequence of functions for which the pointwise limit is not a constant.
Example 8.1.2 For a / 0 and each k + MC

0, let f k z az k where z + F.
In Chapter 4, our proof of the Convergence Properties of the Geometric Series
8.1. POINTWISE AND UNIFORM CONVERGENCE 327
Theorem showed that the associated sequence of nth partial sums

sn z*
n0 was
given by
b c
;n ;n
a 1 z n1
sn z f k z az k .
k0 k0
1z
In view of Example 8.1.1, we see that for each ¿xed z 0 + P

z + F : z 1,
a
sn z 0 *
n1 is convergent to . Thus,
sn z*
n0 is pointwise convergent on
1 z0
P. In terminology
3* that is soon to be introduced, we more commonly say that “the
series k0 az is pointwise convergent on P.”
k
Our long term goal is to have an alternative way of looking at functions. In par-
ticular, we want a view that would give promise of transmission of nice properties,
like continuity and differentiability. The following examples show that pointwise
convergence proves to be insuf¿cient.
n2z
Example 8.1.3 For each n + M, let f n z where z + F. For each ¿xed
1 n2z
z we can use our properties of limits to ¿nd the pointwise limit of the sequences
of functions. If z 0, then
f n 0*
n1 converges to 0 as a constant sequence of
zeroes. If z is a ¿xed nonzero complex number, then
n2z z z
lim lim 1.
n* 1 n z2 n* 1 z
2
z
n

1 , for z + F
0
Therefore, f n f where f z .

0 , for z 0
Remark 8.1.4 From Theorem 4.4.3(c) or Theorem 3.20(d) of our text, we know
n: 1
that p 0 and : + U, implies that lim 0. Letting ? for
n* 1 p n
1 p
p 0 leads to the observation that
lim n : ? n 0 (8.1)
n*
whenever 0 n ? 1 and for any : + R. This is the form of the statement that
is used by the author of our text in Example 7.6 where a sequence of functions for
which the integral of the pointwise limit differs from the limit of the integrals of the
functions in the sequence is given.
Example 8.1.5 (7.6 in our text) Consider the sequence

f n * of real-valued func-
b
n1 c
2 n for n + M.
tions on the interval I [0 1] that is given by f n x nx 1 x
b c
For ¿xed x + I
0, taking : 1 and ? 1 x 2 in (8.1) yields that
b cn
n 1 x 2 0 as n *. Hence, f n 0. Because f n 0 0 for all
I
0
n + M, we see that for each x + I ,
r sn
lim f n x lim nx 1 x 2 0.
n* n*
In contrast to having the Riemann integral of the limit function over I being 0, we
have that
= 1
n 1
lim f n x d x lim .
n* 0 n* 2n 2 2
Note that, since : in Equation
b c(8.1) can be any real number, the sequence of real
n
functions gn x n x 1 x
2 2 for n + M converges pointwise to 0 on I with
= 1
n2
gn x dx * as n *.
0 2n 2
This motivates the search for a stronger sense of convergence namely, uniform
convergence of a sequence (and, in turn, of a series) of functions. Remember that
our application of the term “uniform” to continuity required much nicer behavior
of the function than continuity at points. We will make the analogous shift in going
from pointwise convergence to uniform convergence.
De¿nition 8.1.6 A sequence of complex functions

f n *
n1 converges pointwise to
a function f on a subset P of F, written f n f or f n f , if and only if the
z+P
sequence
f n z 0 *
n1 f z 0 for each z 0 + P i.e., for each z 0 + P
1 0 2M M z 0 + M n M > z 0 " f n z 0 f z 0 .

f n converges uniformly to f
on a subset P of F, written f n f , if and only if
1 0 2M M M + M F 1n 1z n M F z + P " f n z f z .
Remark 8.1.8 Uniform convergence implies pointwise convergence. Given a se-

quence of functions, the only candidate for the uniform limit is the pointwise limit.
Example 8.1.9 The sequence considered in Example 8.1.1 exhibits the stronger
sense of convergence if we restrict ourselves to compact subsets of
P
z + F : z 1. For each n + M, let f n z z n where z + F. Then
f n z*
n1 is uniformly convergent to the constant function f z 0 on any com-
pact subset of P.
Suppose K t P is compact. From the Heine-Borel Theorem, we know that
K is closed and bounded. Hence, there exists a positive real number r such that
r 1 and 1z z + K " z n r . Let Pr
z + F : z n r . For 0, let

!
! 1 , for o 1

M M z { .
!
! ln
, for 1
ln r
ln
Then n M "n " n ln r ln because 0 r 1. Consequently,
ln r
r and it follows that
n
n n
f n z 0 nz n n zn n r n .
Since 0 was arbitrary, we conclude that f n f . Because K t Pr , f n f

Pr K
as claimed.
Excursion 8.1.10 When we restrict ourselves to consideration of uniformly con-

vergent sequences of real-valued functions on U, the de¿nition links up nicely to a
graphical representation. Namely, suppose that f n f . Then corresponding to
[ab]
any 0, there exists a positive integer M such that n M " f n x f x
for all x + [a b]. Because we have real-valued functions on the interval, the in-
equality translates to
f x f n x f x for all x + [a b] . (8.2)
Label the following ¿gure to illustrate what is described in (8.2) and illustrate the
implication for any of the functions f n when n M.
Remark 8.1.11 The negation of the de¿nition offers us one way to prove that a
sequence of functions is not uniformly continuous. Given a sequence of functions
f n that are de¿ned on a subset P of F, the convergence of

f n to a function f on
P is not uniform if and only if
2 0 1M [M + M "

b cb n b c b cn c
2n 2z Mn n M F z Mn + P F n f n z Mn f z Mn n o ].
| }
1 *
Example 8.1.12 Use the de¿nition to show that the sequence is point-
nz n1
wise convergent, but not uniformly convergent, to the function f x 0 on P
z + F : 0 z 1.
z {Suppose that z 0 is a ¿xed element of P. For 0, let M M z 0
1 1 1
. Then n M " n " because z 0 0. Hence,
z 0 z 0 n z 0
n n
n 1 n 1
n n
n nz 0n n z .
0 0
| }
1 *
Since 0 was arbitrary, we conclude that is convergent to 0 for each
| }* nz 0 n1
1
z 0 + P. Therefore, is pointwise convergent on P.
nz n1
1 1
On the other hand, let and for each n + M, set z n . Then
2 n1
z n + P and
n n
n n
n n n n
n 1 n n 1 n 1
n n n n
n nz 0n n t 1 u n 1 n o .
n nn n
n n
n1
| }*
1
Hence, is not uniformly convergent on P.
nz n1
| }*
1
Example 8.1.13 Prove that the sequence converges uniformly for
1 nz n1 | }
1
z o 2 and does not converge uniformly in P`
z + F : z n 2 : n + M .
n
1
Let P
z + F : z o 2 and, for each n + M, let f n z . From
1 nz
*
the limit properties of sequences,
f n zn1 is pointwise convergent on F to

0 , for z + F
0
f z .

1 , for z 0
*
t of
fu{
Thus, the pointwisezlimit n zn1 on P is the constant u 0. For 0,
tfunction
1 1 1 1 1
let M M 1 . Then n M " n 1 "
2 2 2n 1
because n 1. Furthermore, z o 2 " n z o 2n " n z 1 o 2n 1 0.
Hence, z o 2 F n M "
n n
n 1 n 1 1 1
n
f n z 0 n nn n n .
1 nz n n z 1 n z 1 2n 1
Because 0 was arbitrary, we conclude that fn 0.
P
1
On the other hand, let and, corresponding to each n + M, set
2
1
z n . Then z n + P` and
n
n n
n n
n n
n 1 n 1
f n z n 0 nn t u nn o .
n1 n 1 n 2
n n
n
Hence,
f n z* `
n1 is not uniformly convergent in P .
Excursion 8.1.14 Use the de¿nition to prove that the sequence

z n is not uni-
formly convergent in z 1.
t u
1
***Hopefully, you thought to make use of the choices ?n 1 that could be
n
related back to e1 .***
Using the de¿nition to show that a sequence of functions is not uniformly con-
vergent, usually, involves exploitation of “bad points.” For Examples 8.1.12 and
8.1.13, the exploitable point was x 0 while, for Example 8.1.14, it was x 1.
Because a series of functions is realized as the sequence of nth partial sums of
a sequence of functions, the de¿nitions of pointwise and uniform convergence of
series simply make statements concerning the nth partial sums. On the other hand,
we add the notion of absolute convergence to our list.
De¿nition 8.1.15 Corresponding to the sequence

ck z*
k0 of complex-valued
functions on a set P t F, let
;
n
Sn z ck z
k0
denote the sequence of nth partial sums. Then

3
(a) the series * k0 ck z is pointwise convergent on P to S if and only if, for
each z 0 + P,
Sn z 0 *
n0 converges to S z 0 and
3
(b) the series *k0 ck z is uniformly convergent on P to S if and only if Sn
P
S.
De¿nition 8.1.16 Corresponding to the 3* sequence

ck z*
k0 of complex-valued
functions on a3set P t F, the series k0 ck z is absolutely convergent on P
if and only if *k0 ck z is convergent for each z + P.
Excursion 8.1.17 For a / 0 and k + M C

0, let ck z az k . In Example 8.1.2,
we saw that
;* ;
*
ck z az k
k0 k0
is pointwise convergent for each z 0 + P

z + F : z 1 to a 1 z 0 1 . Show
that
3*
(i) k0 ck z is absolutely convergent for each z 0 + P
3*
(ii) k0 ck z is uniformly convergent on every compact subset K of P
3*
(iii) k0 ck z is not uniformly convergent on P.
***For part (i), hopefully you noticed that the formula derived for the proof of the
Convergence Properties of the Geometric Series applied to the real b seriesn1
that
c re-
3 a 1 z
sults from replacing az k with a zk . Since nk0 a zk , we
1 z
j3n k
k * a
conclude that k0 a z n0 for each z + C such that z 1
3* 3* 1 z
i.e., k0 ck z k0 az k is absolutely convergent for each z + P. To show
part (ii) it is helpful to make use of the fact that if K is a compact subset of
P then there exists a positive real number r such 3* that r 1 and K t Pr
z + F : z n r . The uniform convergence of k0 ck z on Pr then byields uni- c

3n 3n a 1 z n1
form convergence on K . For Sn z k0 ck z k0 az k
1z
a a r n1
and S z , you should have noted that Sn z S z n for all
1z b c 1r

ln 1 r a1
z + Pr which leads to M max 1 1 as one pos-
ln r
sibility for justifying tthe uniformuconvergence. Finally, with (iii), corresponding to
1
each n + M, let z n 1 then z n + P for each n and Sn z n S z n
n1
t un1
1
n 1 a 1 can be used to justify that we do not have uniform
n1
convergence.***
8.1.1 Sequences of Complex-Valued Functions on Metric Spaces

In much of our discussion thus far and in numerous results to follow, it should
become apparent that the properties claimed are dependent on the properties of the
codomain for the sequence of functions. Indeed our original statement of the de¿ni-
tions of pointwise and uniform convergence require bounded the distance between
images of points from the domain while not requiring any “nice behavior relating
the points of the domain to each other.” To help you keep this in mind, we state the
de¿nitions again for sequences of functions on an arbitrary metric space.

f n * n1 converges pointwise
to a function f on a subset P of a metric space X d, written f n f or
f n f , if and only if the sequence
f n *0 *
n1 f *0 for each *0 + P
*+P
i.e., for each *0 + P
1 0 2M M *0 + M n M > *0 " f n *0 f *0 .

8.2. CONDITIONS FOR UNIFORM CONVERGENCE 335

f n converges uniformly to f
on a subset P of a metric space X d, written f n f on P or f n f , if and
P
only if
1 0 2M M [M + M F 1n 1* n M F * + P

" f n * f * ].
8.2 Conditions for Uniform Convergence

We would like some other criteria that can allow us to make decisions concerning
the uniform convergence of given sequences and series of functions. In addition,
if can be helpful to have a condition for uniform convergence that does not require
knowledge of the limit function.
De¿nition 8.2.1 A sequence

f n *
n1 of complex-valued functions satis¿es the
Cauchy Criterion for Convergence on P t F if and only if
1 0 2M + M [1n 1m 1z n M Fm M Fz +P

" f n z f m z ].
Remark 8.2.2 Alternatively, when a sequence satis¿es the Cauchy Criterion for
Convergence on a subset P t F it may be described as being uniformly Cauchy
on P or simply as being Cauchy.
In Chapter 4, we saw that in Un being convergent was equivalent to being

Cauchy convergent. The same relationship carries over to uniform convergence
of functions.
Theorem 8.2.3 Let

f n *
n1 denote a sequence of complex-valued functions on a
set P t F. Then
f n * *
n1 converges uniformly on P if and only if
f n n1 satis¿es
the Cauchy Criterion for Convergence on P.
Proof. Suppose that

f n *
n1 is a sequence of complex-valued functions on a
set P t F that converges uniformly on P to the function f and let 0 be given.

Then there exists M + M such that n M implies that f n z f z for
2

all z + P. Taking any other m M also yields that f m z f z for all
2
z + P. Hence, for m M F n M,
f m z f n z f m z f z f n z f z

n f m z f z f n z f z
for all z + P. Therefore,

f n *
n1 is uniformly Cauchy on P.
Suppose the sequence

f n *
n1 of complex-valued functions on a set P t
F satis¿es the Cauchy Criterion for Convergence on P and let 0 be given. For
z + P,
f n z*n1 is a Cauchy sequence in F because F is complete, it follows
*
that
f n zn1 is convergent to some ?z + F. Since z + P was arbitrary, we
can de¿ne a function f : P F by 1z z + P " f z ?z . Then, f is the
pointwise limit of
f n * *
n1 . Because
f n n1 is uniformly Cauchy, there exists an
M + M such that m M and n M implies that

f n z f m z for all z + P.
2
Suppose that n M is ¿xed and z + P. Since lim f m G f G for each
m*
G + P, there exists a positive integer M ` M such that m M ` implies that

f m z f z . In particular, we have that f M ` 1 z f z . There-
2 2
fore,
f n z f z f n z f M ` 1 z f M ` 1 z f z

n f n z f M ` 1 z f M ` 1 z f z .
But n M and z + P were both arbitrary. Consequently,

8.2. CONDITIONS FOR UNIFORM CONVERGENCE 337
1n 1z n M F z + P " f n z f z .
Since 0 was arbitrary, we conclude that f n f .

P
Remark 8.2.4 Note that in the proof just given, the positive integer M ` was depen-
dent on the point z and the i.e., M ` M ` z. However, the ¿nal inequality ob-
tained via the intermediate travel through information from M ` , f n z f z
, was independent of the point z. What was illustrated in the proof was a process
that could be used repeatedly for each z + P.
Remark 8.2.5 In the proof of both parts of Theorem 8.2.3, our conclusions relied
on properties of the codomain for the sequence of functions. Namely, we used the
metric on F and the fact that F was complete. Consequently, we could allow P
to be any metric space and claim the same conclusion. The following corollary
formalizes that claim.
Corollary 8.2.6 Let

f n *n1 denote a sequence of complex-valued functions de-
¿ned on a subset P of a metric space X d. Then
f n *n1 converges uniformly on
*
P if and only if
f n n1 satis¿es the Cauchy Criterion for Convergence on P.
Theorem 8.2.7 Let

f n *
n1 denote a sequence of complex-valued functions on a
set P t F that is pointwise convergent on P to the function f i.e.,
lim f n z f z

n*
and, for each n + M, let Mn sup f n z f z. Then f n f if and only if
z+P P
lim Mn 0.
n*
Use this space to ¿ll in a proof for Theorem 8.2.7.

Theorem 8.2.8 (Weierstrass M-Test) For each n + M, let u n * be a complex-

valued function that is de¿ned on a subset P of a metric space X d. Suppose that
there exists a sequence of real constants
M
3n**
n1 such that u n * 3
n Mn for all
* +3 P and for each n + M. If the series n1 Mn converges, then * n1 u n *
*
and n1 u n * converge uniformly on P.
the Weierstrass M-Test.
Proof. Suppose that

u n ** *
n1 , P, and
Mn n1 are as described in the hy-
potheses. For each n + J , let
;
n ;
n
Sn * u k * and Tn * u k *
k1 k1
3
and suppose that 0 is given. Since * *
n1 Mn converges and
Mn n1 t U,
j3n k*
k1 Mk n1 is a convergent sequence
j3n
of real numbers. In view of the com-
k*
pleteness of the reals, we have that k1 Mk n1 is . Hence, there
1
exists a positive integer K such that n K implies that
;p
n
Mk for each p + M.
kn1
Since u k * n Mk for all * + P and for each k + M, we have that

n n
n n nn n;p n
n ;p
n
nTn p * Tn *n n u k *n u * for all * + P.
nkn1 n kn1 k
Therefore,
Tn *
n1 is in P. It follows from the
2 3
that
;p
n ;p
n
n u k * n Mk
4 5 kn1 kn1
for all * + P. Hence,

Sn *
n1 is uniformly Cauchy in P. From Corollary 8.2.6, we
conclude that .
6
8.3. PROPERTY TRANSMISSION AND UNIFORM CONVERGENCE 339
***Acceptable responses include: (1) Cauchy,

n3 (2) uniformly
n Cauchy, (3) triangular
n n n n 3
inequality, (4) n Sn p * Sn *n, (5) n kn1 u k *n, and (6) *
n p
n1 u n * and
3*
n1 u n * converge uniformly on P.***
Excursion 8.2.10 Construct an example to show that the converse of the Weier-
strass M-Test need not hold.
8.3 Property Transmission and Uniform Convergence

We have already seen that pointwise convergence was not suf¿cient to transmit the
property of continuity of each function in a sequence to the limit function. In this
section, we will see that uniform convergence overcomes that drawback and allows
for the transmission of other properties.
Theorem 8.3.1 Let

f n *
n1 denote a sequence of complex-valued functions de¿ned
on a subset P of a metric space X d such that f n f . For * a limit point of P
P
and each n + M, suppose that
lim f n t An .
t*
t +P
Then
An *
n1 converges and lim f t lim An .
t* n*
the Theorem.
Proof. Suppose that the sequence
f n *
n1 of complex-valued functions de¿ned
on a subset P of a metric space X d is such that f n f , * is a limit point of
P
P and, for each n + M, lim f n t An . Let 0 be given. Since f n f ,
t* P
by Corollary 8.2.6,
f n *
n1 is on P. Hence, there exists a positive
1
integer M such that implies that
2

f n t f m t for all .
3 3
Fix m and n such that m M and n M. Since lim f k t Ak for each k + M,
t*
it follows that there exists a = 0 such that 0 d t * = implies that

f m t Am and
3 4
From the triangular inequality,

n n
n n
n n
An Am n An f n t n n f m t Am .
n 5
n
Since m and n were arbitrary, for each 0 there exists a positive integer M
such that 1m 1n n M F m M " An Am i.e.,
An * n1 t F is
Cauchy. From the completeness of the complex numbers, if follows that
An * n1 is
convergent to some complex number let lim An A.
n*
We want to show that A is also equal to lim f t. Again we suppose that
t*
t +P
0 is given. From f n f there exists a positive integer M1 such that n M1
n P n
n n
n n
implies that n n for all t + P, while the convergence of
An *
n 6
n 3 n1

yields a positive integer M2 such that An A whenever n M2 . Fix n such
3
that n max
M1 M2 . Then, for all t + P,

f t f n t and An A .
3 3
Since lim f n t An , there exists a = 0 such that
t*
t +P

f n t An for all t + N= *
* D P.
3
From the triangular inequality, for all t + P such that 0 d t * =,
f t A n .
7
Therefore, .
8
***Acceptable responses are: (1) uniformly Cauchy, (2) n M F m M, (3) t +

P, (4) f n t An , (5) f n t f m t, (6) f t f n t, (7) f t f n t
3
f n t An An A, and (8) lim f t A.***
t*
t +P
Theorem 8.3.3 (The Uniform Limit of Continuous Functions) Let

f n * n1 denote
a sequence of complex-valued functions that are continuous on a subset P of a met-
ric space X d. If f n f , then f is continuous on P.
P
Proof. Suppose that

f n *n1 is a sequence of complex-valued functions that
are continuous on a subset P of a metric space X d. Then for each ? + P,
lim f n t f n ? . Taking An f n ? in Theorem 8.3.1 yields the claim.
t?
Remark 8.3.4 The contrapositive of Theorem 8.3.3 affords us a nice way of show-
ing that we do not have uniform convergence of a given sequence of functions.
Namely, if the limit of a sequence of complex-valued functions that are continuous
on a subset P of a metric space is a function that is not continuous on P, we may
immediately conclude that the convergence in not uniform. Be careful about the
appropriate use of this: The limit function being continuous IS NOT ENOUGH to
conclude that the convergence is uniform.
| }*
1
The converse of Theorem 8.3.3 is false. For example, we know that
nz n1
converges pointwise to the continuous function f z 0 in F
0 and the con-
vergence is not uniform. The following result offers a list of criteria under which
continuity of the limit of a sequence of real-valued continuous functions ensures
that the convergence must be uniform.
Theorem 8.3.5 Suppose that P is a compact subset of a metric space X d and
f n *
n1 satis¿es each of the following:
(i)
f n *
n1 is a sequence of real-valued functions that are continuous on P
(ii) f n f and f is continuous on P and

P
(iii) 1n 1* n + M F * + P " f n * o f n1 *.
Then f n f .
P
Theorem 8.3.5.
Proof. For
f n *
n1 satisfying the hypotheses, set gn f n f . Then, for each
n + M, gn is continuous on P and, for each ? + P, lim gn ? . Since
n*
1
f n * o f n1
* implies that f n * f * o
f n1 * f *, we also have
that 1n 1* n + M F * + P " .

2
To see that gn 0, suppose that 0 is given. For each n + M, let
P
K n
x + P : gn x o .
Because P and U are metric spaces, gn is continuous, and

* + U : * o is a
closed subset of U, by Corollary 5.2.16 to the Open Set Characterization of Con-
tinuous Functions, . As a closed subset of a compact metric space,
3
from Theorem 3.3.37, we conclude that K n is . If x + K n1 , then
4
gn1 x o and gn x o gn1 x it follows from the transitivity of o that
. Hence, x + K n . Since x was arbitrary, 1x x + K n1 " x + K n
5
i.e., . Therefore,
K n *
n1 is a sequence of compact
6 7
subsets
7 of P. From Corollary 3.3.44 to Theorem 3.3.43, 1n + M K n / 3 "
K k / 3.
k+M
Suppose that * + P. Then lim gn * 0 and
gn x decreasing yields
n*
7 that 0 n gn * .
the existence of a positive integer M such that n M implies
In particular, * + K M1 from which it follows that * +
K n . Because * was
n+M

7 7
arbitrary, 1* + P * +
K n i.e., K n 3. We conclude that there exists
n+M n+M
a positive integer P such that K P 3. Hence, K n 3 for all that is, for
8
all n o P,
x + P : gn x o 3. Therefore,
1x 1n x + P F n P " 0 n gn x .
Since 0 was arbitrary, we have that gn 0 which is equivalent to showing that

P
fn f .
P
***Acceptable responses are: (1) 0, (2) gn * o gn1 *, (3) K n is closed, (4)
compact, (5) gn x o , (6) K n1 t K n , (7) nested, and (8) n o P.***
Remark 8.3.7 Since compactness was referred to several times in the proof of The-
orem 8.3.5, it is natural to want to check that the compactness was really
| needed.
}*
1
The example offered by our author in order to illustrate the need is
1 nx n1
in the segment 0 1.
Our results concerning transmission of integrability and differentiability are for

sequences of functions of real-valued functions on subsets of U.
Theorem 8.3.8 (Integration of Uniformly Convergent Sequences) Let : be a func-

tion that is (de¿ned and) monotonically increasing on the interval I [a b]. Sup-
pose that
f n *
n1 is a sequence of real-valued functions such that
1n n + M " f n + 4 : on I
and f n f . Then f + 4 : on I and

[ab]
= b = b
f x d: x lim f n x d: x
a n* a
the Theorem.
Proof. For each n + J , let n sup f n x f x. Then

x+I
f n x n n f x n for a n x n b

1
and if follows that

= b = b = b
f n x n d: x n f x d: x n f x d: x n
a a a
= b
f n x n d: x . (8.3)
a
Properties of linear ordering yield that

= b = b
0n f x d: x f x d: x n
a a
= b
f n x n d: x . (8.4)
a 2
Because the upper bound in equation (8.4) is equivalent to ,

3
that
we conclude
5b 5b
1n + M 0 n a f x d: x a f x d: x n . By The-
4
5b 5b
orem 8.2.7, n 0 as n *. Since a f x d: x a f x d: x is constant,
we conclude that . Hence f + 4 :.
5
Now, from equation 8.3, for each n + J ,

= b = b = b
f n x n d: x n f x d: x n f n x n d: x .
a a a
6 Finish the proof in the space provided.

5b
***Acceptable responses are:(1) f n x n (2) a f n x n d: x,
5b 5b 5b
(3) a en d: x, (4) 2n [: b : a], (5) a f x d: x a f x d: x, (6)
Hopefully, you thought to repeat the process just illustrated. From the modi¿ed
inequality
n5 it follows that n
n b 5b n
n a x
f d: x a n x
f d: xn n n [: b : a] then because n 0 as
n *, given any 0 there exists a positive integer M such that n M implies
that n [: b : a] .***
;
*
Corollary 8.3.10 If f n + 4 : on [a b], for each n + M, and f k x converges
k1
uniformly on [a b] to a function f , then f + 4 : on [a b] and
= b ;* = b
f x d: x f k x d: x .
a k1 a
Having only uniform convergence of a sequence of functions is insuf¿cient to

make claims concerning the sequence of derivatives. There are various results that
offer some additional conditions under which differentiation is transmitted. If we
restrict ourselves to sequences of real-valued functions that are continuous on an
interval [a b] and Riemann integration, then we can use the Fundamental Theorems
of Calculus to draw analogous conclusions. Namely, we have the following two
results.

f n *n1 is a sequence of real-valued functions that
are continuous on the interval [a b] and f n f . For c + [a b] and each n + M,
[ab]
let
= x
Fn x f n t dt.
de f c
Then f is continuous on [a b] and Fn F where

[ab]
= x
F x f t dt.
c
The proof is left as an exercise.


f n *
n1 is such that f n f and, for each n + M,
[ab]
f n) is continuous on an interval [a b]. If f n) g for some function g that is de¿ned
[ab]
on [a b], then g is continuous on [a b] and f ) x g x for all x + [a b].
Proof. Suppose that

f n * )
n1 is such that f n f , f n is continuous on an
[ab]
interval [a b] for each n + M, and f n) g for some function g that is de¿ned on
[ab]
[a b]. From the Uniform Limit of Continuous Functions Theorem, g is continuous.
Because each f n) is continuous and f n) g, by the second Fundamental Theorem
[ab]
of Calculus and Theorem 8.3.11, for [c x] t [a b]
= x = x
d e
g t dt lim f n) t dt lim f n x f n c .
c n* c n*
d e
Now the pointwise convergence of
f n yields that lim f n x f n c f x
n*
f c. Hence, from the properties of derivative and the ¿rst Fundamental Theorem
of Calculus, g x f ) x.
We close with the variation of 8.3.12 that is in our text it is more general in
that it does not require continuity of the derivatives and speci¿es convergence of
the original sequence only at a point.

f n * n1 is a sequence of real-valued functions that
are differentiable on an interval j ) k*b] and that there exists a point x0 + [a*b]
[a
such that lim f n x 0 exists. If f n n1 converges uniformly on [a b] then
f n n1
n*
converges uniformly on [a b] to some function f and
r s
1x x + [a b] " f ) x lim f n) x .
n*
of Theorem 8.3.13.
Proof. Suppose 0 is given. Because

f n x0 *
n1 is convergent sequence
*
of real numbers and U is complete,
f n x0 n1 is . Hence, there
1
exists a positive integer M1 such that n M1 and m M1 implies that

f n x0 f m x 0
2
j k*
Because f n) n1 converges uniformly on [a b], by Theorem , there exists
2
a positive integer M2 such that n M2 and m M2 implies that
n ) n
n f G f ) Gn for .
n m
2 b a 3
For ¿xed m and n, let F f n f m . Since each f k is differentiable on [a b],

F is differentiable on a b and continuous on [a b]. From the
4
Theorem, for any [x t] t a b, there exists a G + x t such that F x F t
F ) G x t. Consequently, if m M2 and n M2 , for any [x t] t a b, there
exists a G + x t, it follows that
n n
f n x f m x f n t f m t n f n) G f m) Gn x t (8.5)

x t n .
2 b a 5
Let M max
M1 M2 . Then m M and n M implies that
f n * f m *
n f n * f m * f n x0 f m x0 f n x0 f m x0
for any * + [a b]. Hence,
f n *
n1 converges uniformly on [a b] to some function.
Let f denote the limit function i.e., f x lim f n x for each x + [a b] and
n*
fn f .
[ab]
Now we want to show that, for each x + [a b], f ) x lim f n) x i.e.,
n*
for ¿xed x + [a b],
f n t f n x f t f x
lim lim lim
n*tx tx tx tx
where the appropriate one-sided limit is assumed when x a or x b. To this
end, for ¿xed x + [a b], let
f n t f n x f t f x
Mn t and M t
de f tx de f tx
for t + [a b]
x and n + M. Then, x + a b implies that limMn t f n) x,
tx
while x a and x b yield that lim Mn t f n) a and lim Mn t f n) b,
ta tb
respectively. Suppose 0 if given. If m M2 , n M2 , and t + [a b]

x,
then

Mn t Mm t
6
2 b a
from equation (8.5). Thus,

Mn *
n1 is uniformly Cauchy and, by Theorem 8.2.3,
uniformly convergent on t + [a b]
x. Since f t lim f n t for t + [a b],
n*
we have that
lim Mn t M t .

n*
Consequently, Mn M on [a b]
x. Finally, applying Theorem 8.3.1 to the
sequence
Mn * )
n1 , where An f n x yields that
f ) x lim M t .

tx
7
***Acceptable responses are: (1) Cauchy, (2) 8.2.3, (3) all G + [a b], (4) Mean-

Value, (5) ,
n 2 n
n f n t f n x f m t f m x nn f n t f m t f n x f m x
n
(6) n
tx t x n t x
,
(7) lim An lim f n) x.***
n* n*
Rudin ends the section of our text that corresponds with these notes by con-
structing an example of a real-valued continuous function that is nowhere differen-
tiable.
Theorem 8.3.15 There exists a real-valued function that is continuous on U and

nowhere differentiable on U.
Proof. First we de¿ne a function M that is continuous on U, periodic with period

2, and not differentiable at each integer. To do this, we de¿ne the function in a
interval that is “2 wide” and extend the de¿nition by reference to the original part.
For x + [1 1], suppose that M x x and, for all x + R, let M x 2 M x.
8.4. FAMILIES OF FUNCTIONS 349
In the space provided sketch a graph of M.
The author shows that the function

;* t un b c
3
f x M 4n x
n0
4
satis¿es the needed conditions. Use the space provided to ¿ll in highlights of the
justi¿cation.
8.4 Families of Functions

Since any sequence of functions is also a set of functions, it is natural to ask ques-
tions about sets of functions that are related by some commonly shared nice behav-
ior. The general idea is to seek additional properties that will shared by such sets
of functions. For example, if I is the set of all real-valued functions from [0 1]
into [0 1] that are continuous, we have seen that an additional shared property is
that 1 f f + I " 2t t + [0 1] F f t t. In the last section, we consid-
ered sets of functions from a metric space into F or U and examined some of the
consequences of uniform convergence of sequences.
Another view of sets of functions is considering the functions as points in a

metric space. Let F [a b] denote the family of real-valued functions that are con-
tinuous on the interval I [a b]. For f and g in F [a b], we have seen that
I* f g max f x g x

anxnb
and
= b
I f g f x g x dx.
a
are metrics on F [a b]. As a homework problem (Problem Set H, #14), you
will show that F [a b] I is not a complete metric space. On the other hand,
F [a b] I* is complete. In fact, the latter generalizes to the set of complex-
valued functions that are continuous and bounded on the same domain.
De¿nition 8.4.1 For a metric space X d, let F X denote the set of all complex-
valued functions that are continuous and bounded on the domain X and, corre-
sponding to each f + F X the supremum norm or sup norm is given by
P f P P f P X sup f x .
x+X
It follows directly that P f P X 0 % f x 0 for all x + X and

b c
1 f 1g f g + F X " P f gP X n P f P X PgP X .
The details of our proof for the corresponding set-up for F [a b] allow us to claim
that I* f g P f gP X is a metric for F X.
Lemma 8.4.2 The convergence of sequences in F x with respect to I* is equiv-

alent to uniform convergence of sequences of continuous functions in subsets of
X.
Use the space below to justify the claim made in the lemma.
***Hopefully, you remembered that the metric replaces the occurrence of the ab-
solute value (or modulus) is the statement of convergence. The immediate trans-
lation is that for every 0, there exists a positive integer M such that n M
implies that I* f n f . Of course, you don’t want to stop there the state-
ment I* f n f translates to sup f n x f x which yields that
x+X
1x x + X " f n x f x . This justi¿es that convergence of
f n *n1
with respect to I* implies that
f n *
n1 converges uniformly to f . Since the con-
verse also follows immediately from the de¿nitions, we can conclude that con-
vergence of sequences in C X with respect to I* is equivalent to uniform
convergence.***
Theorem 8.4.3 For a metric space X, F X I* is a complete metric space.
Theorem 8.4.3.
Proof. Since F X I* is a metric space, from Theorem 4.2.9, we know that
any convergent sequence in F X is Cauchy.
Suppose that
f n *
n1 is a Cauchy sequence in F X I* and that 0
is given. Then there exists a positive integer M such that n M and m M
implies that i.e., for n M and m M,
1
sup f n G f m G .
G +X

Hence, 1x x + X " . Since 0 was arbitrary, we conclude
2
that
f n *
n1 is . As a
3 3
sequence of complex-valued functions on a metric space X, by Theorem ,

4
f n *
n1 is uniformly convergent. Let f : X F denote the uniform limit. Be-
cause f n f , for any 0 there exists a positive integer M such that n M
X
implies that

f n x f x for all x + X.
2

In particular, sup f n G f G n . Since 0 was ar-
5 G +X 2
bitrary, we conclude that I* f n f 0 as n *. Hence,
f n *
n1 is convergent
to f in F X I*.
Now we want to show that f + F X. As the uniform limit of continuous
functions from a metric space X in F, we know that f is . Because
6
f n f , corresponding to 1 there exists a positive integer M such that n M
X
implies that f n x f x 1 for all x + X. In particular, from the (other)
triangular inequality, we have that
1x x + X " f x f M1 x 1 . (8.6)
Since f M1 + F X, f M1 is continuous and on X. From equation

7
(8.6), it follows that f is on X. Because f : X F is contin-
7
uous and bounded on X, . The sequence
f n *
n1 was an arbitrary
8
Cauchy sequence in F X I*. Consequently, we conclude that every Cauchy
sequence in F X I* is convergent in F X I*. This concludes that proof
that convergence in F X I* is equivalent to being Cauchy in F X I*.
***Acceptable responses are: (1) I* f n f m , (2) f n x f m x , (3)
uniformly Cauchy, (4) 8.2.3, (5) I* f n f , (6) continuous, (7) bounded, and (8)
f + F X.***
Remark 8.4.5 At ¿rst, one might suspect that completeness is an intrinsic property
of a set. However, combining our prior discussion of the metric spaces U d and
T d where d denotes the Euclidean metric with our discussion of the two metrics
on F [a b] leads us to the conclusion that completeness depends on two things:
the nature of the underlying set and the way in which distance is measured on the
set.
We have made a signi¿cant transition from concentration on sets whose ele-

ments are points on a plane or number line (or Euclidean n-space) to sets where the
points are functions. Now that we have seen a setting that gives us the notion of
completeness in this new setting, it is natural to ask about generalization or transfer
of other general properties. What might characterizations of compactness look like?
Do we have an analog for the Bolzano-Weierstrass Theorem? In this discussion,
we will concentrate on conditions that allow us to draw conclusions concerning se-
quences of bounded functions and subsequences of convergent sequences. We will
note right away that care must be taken.
De¿nition 8.4.6 Let I denote a family of complex-valued functions de¿ned on a

metric space P d. Then
(a) I is said to be uniformly bounded on P if and only if

2M + U 1 f 1* f + I F * + P " f * n M.
(b) I is said to be locally uniformly bounded on P if and only if

1* * + P " 2N* N* t P F I is uniformly bounded on N* .
(c) any sequence

f n *
n1 t I is said to be pointwise bounded on P if and only
if
b c
1* * + P "
f n ** n1 is bounded i.e., corresponding to each * +
P, there exists a positive real number M* M * such that
de f
f n * M* for all n + M.

| }
x
Example 8.4.7 For x + P U
0, let I f n x 2 : n + M . Then,
n x2
2 *
for * + P, taking M * implies that f n * M * for all n + M.
1 *2
Thus, I is pointwise bounded on P.
Remark 8.4.8 Uniform boundedness of a family implies that each member of the
family is bounded but not conversely.
Excursion 8.4.9 Justify this point with a discussion of I

f n z nz : n + M
on Ur
z + F : z r .
***Hopefully, you observed that each member of I is bounded in Ur but no single

bound works for all of the elements in I.***
Remark 8.4.10 Uniform boundedness of a family implies local uniform bounded-

ness but not conversely.
| }
1
Excursion 8.4.11 To see this, show that : n + M is locally uniformly
1 zn
bounded in U
z : z 1 but not uniformly bounded there.
***Neighborhoods that can justify local uniform boundedness vary the key is to
capitalize on the fact that you can start with an arbitrary ¿xed z + U and make use
of its distance from the origin to de¿ne
t a neighborhood.
u For example, given z 0 + U
1r n n
n n
with z 0 r 1, let Nz0 N z 0 now, N z0 t U and n1 z n 1 n n
4 n n
n n
1 z can be used to justify that, for each n + M, n1 z n 1 n 4 1 r 3.
1
The latter allows us to conclude that the given family is uniformly bounded on
Nz0 . Since z 0 was arbitrary, we can claim local uniform boundedness in U . One
way to justify the lack of uniform boundedness
T is to investigate the behavior of the
1
functions in the family at the points 1 n .***
n
The following theorem gives us a characterization for local uniform bounded-

ness when the metric space is a subset of U or F.
Theorem 8.4.12 A family of complex valued functions I on a subset P of F is

locally uniformly bounded in P if and only if I is uniformly bounded on every
compact subset of P
Proof. (!) This is an immediate consequence of the observation that the clo-
sure of a neighborhood in F or U is compact.
(") Suppose I is locally uniformly bounded on a domain P and K is a compact
subset of P. Then, for each z + K there exists a neighborhood of z, N z >z and a
positive real number, Mz , such that
f ? n Mz , for all ? + N z >z .
j b
N zc >z : z + K covers
Since j exists a ¿nite subcover,
k K , we know that there k say
N z j >z j : j 1 2 n . Then, for M max Mz j : 1 n j n n , f z n
M, for all z + K , and we conclude that I is uniformly bounded on K .
Remark 8.4.13 Note that Theorem 8.4.12 made speci¿c use of the Heine-Borel
Theorem i.e., the fact that we were in a space where compactness is equivalent to
being closed and bounded.
Remark 8.4.14 In our text, an example is given to illustrate that a uniformly bounded
sequence of real-valued continuous functions on a compact metric space need not
yield a subsequence that converges (even) pointwise on the metric space. Because
the veri¿cation of the claim appeals to a theorem given in Chapter 11 of the text, at
this point we accept the example as a reminder to be cautious.
Remark 8.4.15 Again by way of example, the author of our text illustrates that it
is not the case that every convergent sequence of functions contains a uniformly
convergent subsequence. We offer it as our next excursion, providing space for you
to justify the claims.
| 8.4.16 Let P
x
Excursion +U:0nx} n 1 [0 1] and
x2
I f n x :n+M .
x 2 1 nx2
(a) Show that I is uniformly bounded in P.
(b) Find the pointwise limit of

f n *
n1 for x + P.
(c) Justify that no subsequence of

f n *
n1 can converge uniformly on P.
***For (a), observing that x 2 1 nx2 o x 2 0 for x + 0 1] and f n 0 0

for each n + M yields that f n x n 1 for x + P. In (b), since the only occurrence
of n is in the denominator of each f n , for each ¿xed x + P, the corresponding
sequence of real goes to 0 as n *. For (c), in view of the negation of the de¿-
nition of uniform of a sequence, the behavior of the sequence
f n *
| }convergence
*
n1
1
at the points allows us to conclude that no subsequences of
f n * n1 will
n n1
converge uniformly on P.***
Now we know that we don’t have a “straight” analog for the Bolzano-Weierstrass
Theorem when we are in the realm of families of functions in F X. This poses the
challenge of ¿nding an additional property (or set of properties) that will yield such
an analog. Towards that end, we introduce de¿ne a property that requires “local and
global” uniform behavior over a family.
De¿nition 8.4.17 A family I of complex-valued functions de¿ned on a metric space

P d is equicontinuous on P if and only if
1 0 2= 0 1 f 1u 1) f + I F u + P F ) + P F d u ) =

" f u f ) .
Remark 8.4.18 If I is equicontinuous on P, then each f + I is clearly uniformly

continuous in P
Excursion 8.4.19 On the other hand, for U R

z : z n R, show that each func-
tion in I
nz : n + M is uniformly continuous on U R though I is not equicon-
tinuous on U R .
Excursion 8.4.20 Use the Mean-Value Theorem to justify that

Q x R
f n x n sin : n + M
n
is equicontinuous in P [0 *
The next result is particularly useful when we can designate a denumerable

subset of the domains on which our functions are de¿ned. When the domain is
an open connected subset of U or F, then the rationals or points with the real and
imaginary parts as rational work very nicely. In each case the denumerable subset
is dense in the set under consideration.
Lemma 8.4.21 If
f n *
n1 is a pointwise bounded sequence of complex-valued func-
*
j k*
tions on a denumerable set E, then
f n n1 has a subsequence f nk k1 that con-
verges pointwise on E.
Excursion 8.4.22 Finish the following proof.
Proof. Let
f n be sequence of complex-valued functions that is pointwise
bounded on a denumerable set E. Then the set E can be realized as a sequence
*k
of distinct points. This is a natural setting for application of the Cantor diago-
nalization process that we saw earlier in the proof of the denumerability of the
rationals. From the Bolzano–Weierstrass j Theorem, k
f n *1 bounded implies that
j f n1 *k1 . The process can be applied to
jthere existsk a convergent subsequence
f n1 *2 to obtain a subsequence f n2 *2 that is convergent.
f 11 f 21 f 31

f 12 f 22 f 32

j k* j b ck* j k*
In general, f n j n1 is such that f n j * j n1 is convergent and f n j n1 is a
j k* j k*
subsequence of each of f nk n1 for k 1 2 j 1. Now consider f nn n1
j k*
*** For x + E, there exists an M + M such that x * M . Then f nn nM1 is a
j k* j k
subsequence of f nM nM1 from which it follows that f nn x is convergent
at x. ***
The next result tells us that if we restrict ourselves to domains K that are com-
pact metric spaces that any uniformly convergent sequence in F K is also an
equicontinuous family.
Theorem 8.4.23 Suppose that K d is a compact metric space and the sequence
of functions
f n * *
n1 is such that 1n n + M " f n + F K . If
f n n1 converges
uniformly on K , then I
f n : n + M is equicontinuous on K .
Proof. Suppose that K d is a compact metric space, the sequence of functions
f n *
n1 t F K converges uniformly on K and 0 is given. By Theorem 8.2.3,
f n *
n1 is uniformly Cauchy on K . Thus, there exists a positive integer M such

that n o M implies that P f n f m P K . In particular,
3

P f n f M P K for all n M.
3
Because each f n is continuous on a compact set, from the Uniform Continuity
Theorem, for each n + M, f n is uniformly continuous on K . Hence, for each
j +
1 2 M, there exists a = j 0 such that x y + K and d x y = j
n n
n n
implies that f j x f j y . Let = min = j . Then
3 1n jnM

j +
1 2 M F x y + K F d x y = "
n n
1 j 1x 1y n f j x f j yn . (8.7)
3
For n M and x y + K such that d x y =, we also have that
f n x f n y n f n x f M x

f M x f M y f M y f n y . (8.8)
From (8.7) and (8.8) and the fact that 0 was arbitrary, we conclude that
1 0 2= 0 1 f n 1u 1) f n + I F u ) + K F d u ) =

" f u f ) i.e.,
I is equicontinuous on K .
We are now ready to offer conditions on a subfamily of F K that will give us
an analog to the Bolzano-Weierstrass Theorem.
Theorem 8.4.24 Suppose that K d is a compact metric space and the sequence
of functions
f n *
n1 is such that 1n n + M " f n + F K . If
f n : n + M is
pointwise bounded and equicontinuous on K , then
(a)
f n : n + M is uniformly bounded on K and
(b)
f n *
n1 contains a subsequence that is uniformly convergent on K .
of Theorem 8.4.24.
Proof. Suppose that K d is a compact metric space, the sequence of func-

tions
f n *
n1 is such that 1n n + M " f n + F K , the family
f n : n + M is
pointwise bounded and equicontinuous on K .
Proof of part (a):
Let 0 be given. Since
f n : n + M is equicontinuous on K , there exists
a = 0 such that
d e
1n 1x 1y n + M F x y + K F d x y = " f n x f n y .
(8.9)
Because
N= u : u + K forms an for K and K is compact, there
1
exists a ¿nite number of points, say p1 p2 pk , such that K t .
2
On the other hand,
f n : n + M is pointwise bounded consequently, for each p j ,
j +
1 2 k, there exists a positive real number M j such that
b n b cn c
1n n + M " n f n p j n M j
For M , it follows that

3
b n b cn c
1n 1 j n + M F j +
1 2 k " n f n p j n M . (8.10)
6
k b c
Suppose that x + K . Since K t N= p j there exists an m +
1 2 k
j1
such that . Hence, d x pm = and, from (8.9), we conclude that
4
for all n + M. But then f n x f n pm n f n x f n pm
5
yields that f n x f n pm for . From (8.10), we conclude
6
that f n x M for all n + M. Since x was arbitrary, it follows that
d e
1n 1x n + M F x + K " f n x M i.e.,
f n : n + M is .
7
Almost a proof of part (b):

If K were ¿nite, we would be done. For K in¿nite, let E be a denumerable
subset of K that is dense in K . (The reason for the “Almost” in the title of this part
of the proof is that we did not do the Exercise #25 on page 45 for homework. If
K t U or K t F, then the density of the rationals leads immediately to a set E
that satis¿es the desired property in the general case of an arbitrary metric space,
Exercise #25 on page 45 indicates how we can use open coverings with rational
radii to obtain such a set.) Because
f n : n + M is on E, by Lemma
8
j k*
8.4.21, there exists a subsequence of
f n *
n1 , say g j j1 , that is convergent for
each x + E.
Suppose that 0 is given. Since
f n : n + M is equicontinuous on K ,
there exists a = 0 such that
K L
1n 1x 1y n + M F x y + K F d x y = " f n x f n y .
3
Because E is dense in K ,
N= u : u + E forms an open cover for K . Because
K is compact, we conclude that there exists a ¿nite number of elements of E, say
*1 *2 *q , such that
>
q
b c
K t N= * j . (8.11)
j 1
j k j k*
Since *1 *2 *q t E and g j x j1 is a convergent sequence of complex
jnumberskfor each x + E, the completeness of F, yields Cauchy convergence of
*
g j *s j1 for each *s , s +
1 2 q. Hence, for each s +
1 2 q, there
exists a positive integer Ms such that n Ms and m Ms implies that

gn *s gm *s .
3
Suppose that x + K . From (8.11), there exists an s +
1 2 q such that
. Then d x *s = implies that
9

f n x f n *s
3
for all n + M. Let M max

Ms : s +
1 2 q. It follows that, for n M and
m M,
gn x gm x n gn x gn *s

n n
n n
n n
n n gm *s gm x .
n 10
n
Since 0 and x + K were arbitrary, we conclude that

d e
1 0 2M + M n M Fm M " 1x x + K " gn x gm x
j k* j k*
i.e., g j j1 is . By Theorem 8.4.23, g j j1 is uniformly conver-
11
gent on K as needed.
6k b c
***Acceptable responses are: (1) open cover, (2) N= p j ,
j k j1
(3) max M j : j 1 2 k , (4) N= pm , (5) f n x f n pm , (6) all
n + M, (7) uniformly bounded on K , (8) pointwise bounded on K , (9) x + N= *s ,
(10) gn *s gm *s , (11) uniformly Cauchy on K .***
Since we now know that for families of functions it is not the case that every
convergent sequence of functions contains a uniformly convergent subsequence,
families that do have that property warrant a special label.
De¿nition 8.4.26 A family I of complex-valued functions de¿ned on a metric space

j to kbe normal in P if and only if every sequence
f n t I has a subse-
P is said
quence f nk that converges uniformly on compact subsets of P.
Remark 8.4.27 In view of Theorem 8.4.24, any family that is pointwise bounded
and equicontinuous on a compact metric space K is normal in K .
Our last de¿nition takes care of the situation when the limits of the sequences
from a family are in the family.
De¿nition 8.4.28 A normal family of complex-valued functions I is said to be

compact if and only if the uniform limits of all sequences converging in I are
also members of I.
8.5. THE STONE-WEIERSTRASS THEOREM 363
8.5 The Stone-Weierstrass Theorem

In view of our information concerning the transmission of nice properties of func-
tions in sequences (and series), we would like to have results that enable us to
realize a given function as the uniform limit of a sequence of nice functions. The
last result that we will state in this chapter relates a given function to a sequence of
polynomials. Since polynomials are continuously differentiable functions the theo-
rem is particularly good news. We are offering the statement of the theorem without
discussing the proof. Space is provided for you to insert a synopsis or comments
concerning the proof that is offered by the author of our text on pages 159-160.
Theorem 8.5.1 If f + F [a b] for a b, then there exists a sequence of polyno-
mials
Pn *
n1 t F [a b] such that n*
lim Pn x f x where the convergence is
uniform of [a b]. If f is a real-valued function then the polynomials can be taken
as real.
Space for Comments.
8.6 Problem Set H

1. Use properties of limits to ¿nd the pointwise limits for the following se-
quences of complex-valued functions on F.
| }*
nz
(a)
1 nz 2 n1
| }*
nz 2
(b)
z 3n n1
| }*
zn
(c)
1 zn n1
| }*
n2 z
(d)
1 n3 z2 n1
| }*
1 n2 z
n
(e)
1 n 2 z 1 2n n1
j k*
(f) zenz n1
nx
2. For each n + M, let f n x nx . Use the de¿nition to prove that
f n *
n1
e
is pointwise convergent on [0 *, uniformly convergent on [: * for any
¿xed positive real number :, and not uniformly convergent on 0 *.
3. For each of the following sequences of real-valued functions on U, use the

de¿nition to show that
f n x*
n1 converges pointwise to the speci¿ed f x
on the given set I then determine whether or not the convergence is uniform.
Use the de¿nition or its negation to justify your conclusions concerning uni-
form convergence.
| }
2x
(a)
f n x* f x 0 I [0 1]
n1
1 nx
| }
* cos nx
(b)
f n xn1 T f x 0 I [0 1]
n
| 3 }
* n x
(c)
f n xn1 f x 0 I [0 1]
1 n4 x
| }
* n3 x
(d)
f n xn1 f x 0 I [a * where a is a
1 n4x 2
positive ¿xed real number
| } v w
* 1 xn 1 1 1
(e)
f n xn1 f x I
1x 1x 2 2
Q R
(f)
f n x*
n1 nxe
nx 2 f x 0 I [0 1]
8.6. PROBLEM SET H 365
4. For the sequence

f n *n1 of real-valued functions on U given by f n x
n 1 n 2 x n
for n + M and f x 0 for x + I [0 1], show that
1x
f n x f x as n * for each x + I . Is is true that
= 1 = 1
f n x dx f x dx as n *?
0 0
5. Suppose that the sequences of functions

f n * *
n1 and
gn n1 converge uni-
formly to f and g, respectively, on a set A in a metric space S d. Prove
that the sequence
f n gn *
n1 converges uniformly to f g.
;
*
x2
6. Determine all the values of h such that bc T is uniformly conver-
n1 1 nx 2 n
x2
gent in I
x + U : x h. (Hint: Justify that each f n x b c
1 nx 2
is increasing as a function x and make use that the obtain an upper bound on
the summand.)
;
* ;
*
7. Prove that, if an is convergent, then an cos nx converges uniformly
n1 n1
for all x + U.
;
* ;
*
8. Suppose that n bn is convergent and let f x bn sin nx for x + U.
n1 n1
Show that
;
*
f ) x nbn cos nx
n1
;
* ;
*
and that both bn sin nx and nbn cos nx converge uniformly for all x +
n1 n1
U.
9. Prove that if a sequence of complex-valued functions on F converges uni-

formly on a set A and on a set B, then it converges uniformly on A C B.
10. Prove that if the sequence

f n *
n1 of complex-valued functions on F is uni-
formly convergent on a set P to a function f that is bounded on P, then
there exists a positive real number K and a positive integer M such that
1n 1x n M F x + P " f n x K .
11. Suppose that

f n *
n1 is a sequence of real-valued functions each of which is
continuous on an interval I [a b]. If
f n *
n1 is uniformly continuous on
I , prove that there exists a positive real number K such that
1n 1x n + J F x + I " f n x K .
12. Without appeal to Theorem 8.3.8 i.e., using basic properties of integrals,
prove Theorem 8.3.11: Suppose that
f n *n1 is a sequence of real-valued
functions that are continuous on the interval [a b] and f n f . For c +
5x [ab]
[a b] and each n + M, let Fn x c f n t dt.Then f is continuous on
5x
de f
[a b] and Fn F where F x c f t dt.
[ab]
13. Compare the values of the integrals of the nth partials sums over
3*the interval
[0 1] with the integral of their their limit in the case where k1 f k x is
such that
|
x 1 , 1 n x n 0
f 1 x ,
x 1 , 0 x n 1
and, for each n 2 3 4 ,

! 1
!
! 0 , 1 n x
!
! n
!
! 1
!
n2 x n , nx n0
Sn x n .
!
! n 2 x n , 0 x n
1
!
!
!
! n
!
! 1
0 , x n1
n
Does your comparison allow you to conclude anything concerning the uni-
form convergence of the given series n [0 1]? BrieÀy justify your response.
8.6. PROBLEM SET H 367

!
! 1
!
! 0 , if 1 n x n
!
! n
!
!
!

nx 1 1 1
14. For each n + M, let f n x , if x .
!
! 2 n n
!
!
!
!
!
! 1
!
0 , if nx n1
n
Then
f n *
n1 t F [1 1] where F [1 1] is the set of real-valued func-
tions that are continuous on [1 1]. Make use of
f n *
n1 to justify that the
metric space F [1 1] I is not complete, where
= 1
I f g f x g x dx.
1
15. For each of the following families I of real-valued functions on the speci¿ed
sets P, determine whether of not I is pointwise bounded, locally uniformly
bounded, and/or uniformly bounded on P. Justify your conclusions.
| }
1
(a) I 1 : n + M , P 0 1]
nx
| }
sin nx
(b) I T : n + M , P [0 1]
n
| }
nx
(c) I :n+M ,PU
1 n2x 2
| }
x 2n
(d) I :n+M ,PU
1 x 2n
j k
(e) I n 2 x n 1 x : n + M , P [0 1
16. Suppose that I is a family of real-valued k on U that are differen-

j ) functions
)
tiable on the interval [a b] and I f : f + I is uniformly bounded on
[a b]. Prove that I is equicontinuous on a b.
Q R
17. Is I nxe nx 2
: n + M F x + U uniformly bounded on [0 *? State
your position clearly and carefully justify it.
Q x R
18. Is J n cos : n + M F x + U equicontinuous on U? State your posi-
2n
tion clearly and carefully justify it.
*
;n
k 2 x sin kx
19. Is uniformly convergent on [0 *? State your position
k1
1 k4x
n1
clearly and carefully justify it.
Chapter 9
Some Special Functions
Up to this point we have focused on the general properties that are associated with
uniform convergence of sequences and series of functions. In this chapter, most of
our attention will focus on series that are formed from sequences of functions that
are polynomials having one and only one zero of increasing order. In a sense, these
are series of functions that are “about as good as it gets.” It would be even better
if we were doing this discussion in the “Complex World” however, we will restrict
ourselves mostly to power series in the reals.
9.1 Power Series Over the Reals

In
j this section,
k we turn to series that are generated by sequences of functions
k *
ck x : k0 .
De¿nition 9.1.1 A power series in U about the point : + U is a series in the form
;
*
c0 cn x :n
n1
where : and cn , for n + M C

0, are real constants.
Remark 9.1.2 When we discuss power series, we are still interested in the differ-
ent types of convergence that were discussed in the last chapter namely, point-
wise, uniform and absolute. In this context, for example, the power series c0
3 *
n1 cn x : is said to be3 pointwise convergent on a set S t U
n
* 3if*and only if, n
for each x 0 + S, the series c0 n1 cn x0 : converges. If c0 n1 cn x0 :
n
369
370 CHAPTER 9. SOME SPECIAL FUNCTIONS
3*
is divergent, then the power series c0 n1 cn x :n is said to diverge at the
point x 0 .
3
When a given power series c0 * n1 cn x : is known to be pointwise
n
convergent on a set S t U, we de¿ne a function f : S U by f x c0

3 *
n1 cn x : whose range consists of the pointwise limits that are obtained
n
from substituting the elements of S into the given power series.

We’ve already seen an example of a power series
3* about which we know the con-
vergence properties. The geometric series 1 n1 x is a power series about the
n
point 0 with coef¿cients

cn *
n0 satisfying cn 1 for all n. From the Convergence
Properties of the Geometric Series and our work in the last chapter, we know that
3* 1
the series x n is pointwise convergent to in U
x + U : x 1,
n0
1x
3*
the series n0 x n is uniformly convergent in any compact subset of U , and
3*
the series n0 x n is not uniformly convergent in U .
We will see shortly that this list of properties is precisely the one that is associated
with any power series on its segment (usually known as interval) of convergence.
The next result, which follows directly from the Necessary Condition for Conver-
gence, leads us to a characterization of the nature of the sets that serve as domains
for convergence of power series.
3
Lemma 9.1.3 If the series * n0 cn x : converges for x 1 / :, then the series
n
converges absolutely for each x such that x : x1 :. Furthermore, there

is a number M such that
t un
n n
ncn x :n n n M x : for x : n x1 : and for all n. (9.1)
x1 :
3
Proof. Suppose * n0 cn x : converges at x 1 / :. We know that a nec-
n
essary condition for convergence is that the “nth terms” go to zero as n goes to
in¿nity. Consequently, lim cn x1 :n 0 and, corresponding to 1, there
n*
exists a positive integer K such that
n n
n K " ncn x1 :n 0n 1.
9.1. POWER SERIES OVER THE REALS 371
| }
Let M max 1 max c j x1 : j
. Then
0n jnK
n n
ncn x1 :n n n M for all n + M C
0 .
For any ¿xed x + U satisfying x : n x1 :, it follows that

n n
n n n x : nn
ncn x : n cn x : cn x1 : n
n n n n
n x1 : n
n n
n x : nn
n M nn n for all n + M C
0
x1 : n
3* x : nx1 :,
as claimed in equation (9.1). Finally, for ¿xed x + U satisfying
the Comparison Test yields the absolute convergence of n0 cn x : .
The next theorem justi¿es that we have uniform convergence on compact sub-
sets of a segment of convergence.
3
Theorem 9.1.4 Suppose that the series * n0 cn x : converges for x 1 / :.
n
Then the power series converges uniformly on I

x + U : : h n x n : h
for each nonnegative h such that h x1 :. Furthermore, there is a real number
M such that
t un
n n h
ncn x : n n M
n
for x : n h x 1 : and for all n.
x1 :
n n r sn
Proof. The existence of M such that ncn x :n n n M xx:
1 :
was just
shown in our proof of Lemma 9.1.3. For x : n h x1 :, we have that
x : h
n 1.
x1 : x 1 :
The uniform
r sn convergence now follows from the Weierstrass M-Test with Mn
h
x1 : .
3*
Theorem 9.1.5 For the power series c0 n1 cn x :n , either
(i) the series converges only for x : or
(ii) the series converges for all values of x + U or

(iii) there is a positive real number R such that the series converges absolutely for
each x satisfying x : R, converges uniformly in
x + U : x : n R0
for any positive R0 R, and diverges for x + U such that x : R
3
Proof. To see (i) and (ii), note that the power series * n1 n x : diverges
n n
3 x : n
for each x / :, while * n0 is convergent for each x + U. Now, for
n!
(iii), suppose that there is a real number x1 / : for which the series converges
and a real number x2 for which it diverges. By Theorem 9.1.3, it follows that
x1 : n x2 :. Let

;* n n
S I+U: ncn x :n n converges for x : I
n0
and de¿ne
R sup S.
Suppose that x ` is such that x ` : R. Then there exists a I +

S suchn that x ` :
3 n I R. From` the de¿nition of S, we conclude that
* n ` nn
n0 cn x : converges. Since x was arbitrary, the given series is abso-
lutely convergent for each x in
x + U : x : R. The uniform convergence
in
x + U : x : n R0 for any positive R0 R was justi¿ed in Theorem 9.1.4.
n n
Next, suppose that xn + U is such that nx : n I R. From Lemma
3* n b cn n
n
9.1.3, convergence of n0 cn x : would yield absolute convergence of the
given series for all x satisfying x : I and place I in S which would contradict
the de¿nition
3 * n
n of R.n We conclude
nn 3* that for all xn + U, x : R implies that
n0 cn x : as well as n0 cn x : diverge.
The nth Root Test provides us with a formula for ¿nding the radius of conver-
gence, R, that is described in Theorem 9.1.5.
3 T
Lemma 9.1.6 For the power series c0 * n1 cn x : , let I lim sup cn
n n
n*
and

!
! * , if I 0
!
1
R , if 0 I * . (9.2)
! I
!
!

0 , if I *
3
Then c0 * n1 cn x : converges absolutely for each x + : R : R,
n
converges uniformly in
x + U : x : n R0 for any positive R0 R, and di-
verges for x + U such that x : R. The number R is called the radius of
convergence for the given power series and the segment : R : R is called
the “interval of convergence.”
Proof. For any ¿xed x0 , we have that

Tn n r S s
lim sup n ncn x0 :n n lim sup x0 : n cn x0 : I.
n* n*
3*
From the Root Test, the series c0 n1 cn x0 :n converges absolutely when-
ever x0 : I 1 and diverges when x0 : I 1. We conclude that the radius
of convergence justi¿ed in Theorem 9.1.5 is given by equation (9.2).
U
3* 2n1 n 2 2
n
Example 9.1.7 Consider n0 n
x 2n
. Because lim sup n

t u 3 n* 3
2 T 2
2 , from Lemma 9.1.6, it follows that the given power series has
n
lim
n* 3 3
3
radius of convergence . On the other hand, some basic algebraic manipulations
2
yield more information. Namely,
;* ;* v wn
2n1 2 1
x 2 n
2 x 2 2 v w
3 n 3 2
n0 n0 1 x 2
3
n n
n 2 n
as long as nn x 2nn 1, from the Geometric Series Expansion Theorem.
3
3
Therefore, for each x + U such that x 2 , we have that
2
;
*
2n1 6
x 2n .
n0
3n 1 2x
Another useful means of ¿nding the radius of convergence of a power series

follows from the Ratio Test when the limit of the exists.
Lemma 9.1.8 Let : be a realnconstantn and suppose that, for the sequence of nonzero
n n
real constants
cn * , lim n cn1 n L for 0 n L n *.
n0 n* n
cn n
;
*
(i) If L 0, then c0 cn x :n is absolutely convergent for all x + U and
n1
uniformly convergent on compact subsets of U
;* t u
1 1
(ii) If 0 L *, then c0 cn x : is absolutely convergent : :
n
,
L L
n1 t u
1 1
uniformly convergent in any compact subset of : : , and diver-
L L
1
gent for any x + R such that x :
L
;
*
(iii) If L *, then c0 cn x :n is convergent only for x :.
n1
The proof is left as an exercise.
Remark 9.1.9 In view of Lemma n 9.1.8, n whenever the sequence of nonzero real
n c n1 nn
constants
cn * lim nn
n0 satis¿es n* L for 0 n L n * an alternative formula
cn n
;*
for the radius of convergence R of c0 cn x :n is given by
n1

!
! * , if L 0
!
1
R , if 0 L * . (9.3)
!
! L
!

0 , if L *
;
*
1n 2 4 2n
Example 9.1.10 Consider x 2n .
n1
1 4 7 3n 2
1 2 4 2n
n
Let cn . Then
1 4 7 3n 2
n n n n
n cn1 n n 1n 2 4 2n 2 n 1 1 4 7 3n 2 n 2 n 1 2
n nn n
n cn n n 1 4 7 3n 2 3 n 1 2 1n 2 4 2n n 3n 1 3
as n *. Consequently, from Lemma 9.1.8, the radius of convergencet ofuthe

3 8 4
given power series is . Therefore, the “interval of convergence” is .
2 3 3
The simple manipulations illustrated in Example 9.1.7 can also be used to derive
power series expansions for rational functions.
Example 9.1.11 Find a power series about the point : 1 that sums pointwise to
8x 5
and ¿nd its interval of convergence.
1 4x 3 2x
Note that
8x 5 1 2
,
1 4x 3 2x 3 2x 1 4x
1 1 ;
* ;*
1
[2 x 1]n 2n x 1n for x 1
3 2x 1 2 x 1 n0 n0
2
and
2 2 1
vt u w
1 4x 5 4
1 x 1
5
* vt u wn ;
2 ; 4 *
22n1 5
x 1 1n1 n1 x 1n for x 1 .
5 n0 5 n0
5 4
| }
1 5
We have pointwise and absolute convergence of both sums for x 1 min .
2 4
It follows that
;* v w
8x 5 n1 2
2n1 1
1 2 x 1n for x 1 .
n
1 4x 3 2x n0 5 n1 2
The nth partial sums of a power series are polynomials and polynomials are
among the nicest functions that we know. The nature of the convergence of power
series allows for transmission of the nice properties of polynomials to the limit
functions.
3*
Lemma 9.1.12 Suppose that the series f x n0 cn x : converges in
n
x + U : x : R with R 0. Then f is continuous and differentiable in

)
: R : R, f is continuous in : R : R and
;
*
f ) x ncn x :n1 for : R x : R.
n1
Space for comments and scratch work.
Proof. For any x1 such that x1 : R, there exists an h + U with 0 h R

3* that x1 :
such h. Let I
x : x : n h. Then, by Theorem 9.1.4,
n0 cn x : is uniformly convergent on I . From3
n
Theorem 8.3.3, f is con-
tinuous on I as the continuous limit of the polynomials nj0 c j x : j . Conse-
quently, f is continuous at x1 . Since the x1 was arbitrary, we conclude that f is
continuous in x 3 : R.
Note that * j n) kx :
n1 nc
n1
is a power
3n series whose limit, when it is
convergent, is the limit of sn where sn x j0 c j x : . Thus, the second
j
3
part of the theorem will follow from showing that * n1 ncn x :
n1
converges
at least where f is de¿ned i.e., in x : R. Let x0 +
x + U : 0 x : R.
Then there exists an x ` with x 0 : x ` :n R. In the nproof of Lemma ??, it
was shown that there exists an M 0 such that ncn x ` :n n n M for n + MC
0.
Hence,
n n n n
n n1 n n n ` nn n x0 : nn1 M
nncn x 0 : n ` cn nx : n nn ` n
n n ` nr n1
x : x : x :
n n
n x0 : n
for r nn ` n 1. From the ratio test, the series 3* nr n1 converges. Thus,
x :n n1
3* M 3
`
nr n1 is convergent and we conclude that * n1 ncn x :
n1
is
n1
x : 3*
convergent at x0 . Since x0 was arbitrary we conclude that n1 ncn x :n1 is
convergent in x : R. Applying the Theorems 9.1.4 and 8.3.3 as before leads
to the desired conclusion for f ) .
Theorem3 9.1.13 (Differentiation and Integration of Power Series) Suppose f is
given by *n0 cn x : for x + : R : R with R
n
0.
(a) The function f possesses derivatives of all orders. For each positive integer
m, the mth derivative is given by
3 bn c
f m x * nm m cn x :
nm
for x : R
bnc
where m n n 1 n 2 n m 1.
5x
(b) For each x with x : R, de¿ne the function F by F x : f t dt.
3 cn
Then F is also given by * x :n1 which is obtained by term-
n0
n1
by-term integration of the given series for f .
f n :
(c) The constants cn are given by cn .
n!
Excursion 9.1.14 Use the space that is provided to complete the following proof of
the Theorem.
Proof. Since (b) follows directly from Theorem 8.3.3 and (c) follows from
substituting x : in the formula from (a), we need only indicate some of the
details for the proof of (a).
Let

;* t u
n
S m + Q : f m x cn x :nm f or x : R
nm m
b c
where mn n n 1 n 2 n m 1. By Lemma 9.1.12, we know that
1 + S. Now suppose that k + S for some k i.e.,
;
*
f k x n n 1 n 2 n k 1 cn x :nk for x : R.
nk
Remark 9.1.15 Though we have restricted ourselves to power series in U, note

that none of what we have used relied on any properties of U that are not possessed
by F. With that in mind, we state the following theorem and note that the proofs
are the same as the ones given above. However, the region of convergence is a disk
rather than an interval.
3
Theorem 9.1.16 For the complex power series c0 * cn z :n where : and
n1 T
cn , for n + M C
0, are complex constants, let I lim sup n cn and
n*

!
! * , if I 0
!
1
R , if 0 I * .
!
! I
!

0 , if I *
Then the series
(i) converges only for z : when R 0
(ii) converges for all values of z + F when R * and
(iii) converges absolutely for each z + N R :, converges uniformly in

x + U : x : n R0 N R0 :
for any positive R0 R, and diverges for z + F such that z : R

whenever 0 R *. In this case, R is called the radius of convergence for
the series and N R :
z + F : z : R is the corresponding disk of
convergence.
Both Lemma 9.1.12 and Theorem 9.1.13 hold for the complex series in their
disks of convergence.
Remark 9.1.17 Theorem 9.1.13 tells us that every function that is representable
as a power series in some segment : R : R for R 0 has continuous
3* f n :
derivatives of all orders there and has the form f x n0 x :n . It
n!
is natural to ask if the converse is true? The answer to this question is no. Consider
the function
| b c
exp 1x 2 , x / 0
g x .
0 , x 0
9.2. SOME GENERAL CONVERGENCE PROPERTIES 379
It follows from l’Hôpital’s Rule that g is in¿nitely differentiable at x 0 with

g n 0 0 for all n + M C
0. Since the function is clearly not identically equal to
zero in any segment about 0, we can’t write g in the “desired form.” This prompts
us to take a different approach. Namely, we restrict ourselves to a class of functions
that have the desired properties.
De¿nition 9.1.18 A function that has continuous derivatives of all orders in the
neighborhood of a point is said to be in¿nitely differentiable at the point.
De¿nition 9.1.19 Let f be a real-valued function on a segment I . The function f

is said to be analytic at the point : if it is in¿nitely differentiable at : + I and
3 f n :
f x * n0 x :n is valid in a segment : R : R for some
n!
R 0. The function f is called analytic on a set if and only if it is analytic at each
point of the set.
Remark 9.1.20 The example mentioned above tells us that in¿nitely differentiable
at a point is not enough to give analyticity there.
9.2 Some General Convergence Properties

There is a good reason why our discussion has said nothing about what happens at
the points of closure of the segments of convergence. This is because there 3 is no
one conclusion that can be drawn. For example, each of the power series * n
n0 x ,
3* x n 3* x n
n0 , and n0 2 has the same “interval of convergence” 1 1 however,
n n
the ¿rst is divergent at each of the endpoints, the second one is convergent at 1
and divergent at 1, and the last is convergent at both endpoints. The ¿ne point to
keep in mind is that the series when discussed from this viewpoint has nothing to do
with the functions that the series represent if we stay in 1 1. On the other hand,
if a power series that represents a function in its segment is known to converge at
an endpoint, we can say something about the relationship of that limit in relation to
the given function. The precise set-up is given in the following result.
3 3*
Theorem 9.2.1 If 3*n0 cn converges and f x n0 cn x for x + 1 1,
n
*
then lim f x n0 cn .
x1
Theorem 9.2.1.
3
Proof. Let sn nk0 ck and s1 0. It follows that

; m ;m ;
m1
cn x n sn sn1 x n 1 x sn x n sm x m .
n0 n0 n0
3*
Since x 1 and lim sm n0 cn , we have that lim sm x m 0 and we
m* m*
conclude that
;
* ;
*
f x cn x n 1 x sn x n . (9.4)
n0 n0
3* 3*
Let s n0 cn . For each x + 1 1, we know that 1 x n0 x n 1.
Thus,
;
*
s 1 x sx n . (9.5)
n0
Suppose that 0 is given. Because lim sn s there exists a positive integer M

n*

such that implies that sn s . Let
1
2
| }
1 n n
n
K max max s s j n
2 0n jnM
and

!
! 1
!
4 , if o 2K M
= .
!
!
!
, if 2K M
2K M
KM 2K M
Note that, if 2K M n , then n . For 1 = x 1, it
4 8 8 2
follows that
;
M ;
M

1 x sn s x n 1 x
n
xn 1 x M . (9.6)
n0 2 n0 2
2
Use equations (9.4) and (9.5), to show that, if 1 = x 1, then

f x s .
3
***Acceptable responses are: (1) n M, (2) K , 3 (3) Hopefully, you noted that
M
f x s is bounded above by the sum of 1 x n0 sn s xn and
3
1 x * nM1 sn s x . The ¿rst summation is bounded above by
n
as
2
shown in equation (9.6) while the latter summation is bounded above by
b 3 c
1 x * nM1 x with x
n
0 this yields that
2 3* 3* 3
1 x nM1 x 1 x nM1 x n 1 x *
n
n0 x 1.***
n
An application of Theorem 9.2.1 leads to a different proof of the following result

concerning the Cauchy product of convergent numerical series.
3 3* 3*
Corollary 9.2.3 If 3* n0 an , n0 bn , and n0 cn are
3*convergent3to*A, B, and
*
C, respectively, and n0 cn is the Cauchy product of n0 an and n0 bn , then
C AB.
Proof. For 0 n x n 1, let
;* ;
* ;
*
f x an x , g x
n
bn x , and h x
n
cn x n
n0 n0 n0
3n
where cn j0 a j bn j . Because each series converges absolutely for x 1,
for each ¿xed x + [0 1 we have that

;* ;
* ;
*
f x g x an x n
bn x n
cn x n h x .
n0 n0 n0
From Theorem 9.2.1,
lim f x A, lim g x B, and lim h x C.

x1 x1 x1
The result follows from the properties of limits.

One nice argument justifying that a power series is analytic at each point in
its interval of convergence involves rearrangement of the power series. We will
make use of the Binomial Theorem and the following result that justi¿es the needed
rearrangement.
j k 3 n n
Lemma 9.2.4 Given the double sequence ai j i j+J suppose that * n n
j1 ai j bi
3*
and i1 bi converges. Then
;
* ;
* ;
* ;
*
ai j ai j .
i1 j1 j1 i1
Proof. Let E
xn : n + MC
0 be a denumerable set such that lim xn x0
n*
and, for each i n + M let
;
* ;
n
f i x0 ai j and f i xn ai j .
j1 j1
Furthermore, for each x + E, de¿ne the function g on E by

;
*
g x f i x .
i1
From the hypotheses, for each i + M, lim f i xn f i x 0 . Furthermore, the def-
n*
inition of E ensures that for any sequence
*k *
k1 t E such that lim *k x 0 ,
k*
lim f i *k f i x0 . Consequently, from the Limits of Sequences Characteri-
k*
zation for Continuity Theorem, for each i + M, f i is continuous at3x0 . Because
3*
1x 1i i + M F x + E " f i x n bi and i1 bi converges, * i1 f i x is
uniformly convergent in E. From the Uniform Limit of Continuous Functions The-
orem (8.3.3), g is continuous at x0 . Therefore,
;
* ;
* ;
*
ai j f i x0 g x0 lim g xn .
n*
i1 j1 i1
Now
;
* ;
* ;
* ;
* ;
n
ai j lim f i xn lim ai j
n* n*
i1 j1 i1 i1 j1
;n ;* ;;
* *
lim ai j ai j .
n*
j1 i1 ji1 i1
3*
Theorem 9.2.5 Suppose that f x n0 cn x converges in x R. For
n
a + R R, f can be expanded in a power series about the point x a which
3* f n a
converges in
x + U : x a R a and f x n0 x an .
n!
In the following proof, extra space is provided in order to allow more room for
scratch work to check some of the claims.
3*
x n0 cn x in x R, let a + R R. Then f x
Proof. For f n
3* 3*
n0 cn x n0 cn [x a a] and, from the Binomial Theorem,
n n
;
;
*n t u ;* ; n t u
n j n j
f x cn a x a n j
cn a x an j .
n0 j0
j n0 j 0
j
We can think of this form of summation as a “summing by rows.” In this context,

the ¿rst row would
Kb c could be written as c0 x La0 , while the second row could be
b c
written as c1 10 a 0 x a1 11 a 1 x a0 . In general, the ( 1 st row is
given by

( t u
; ( ( j
c( a x a
j
jj0
vt u t u t u w
( 0 ( ( 1 (1 ( (
c( a x a a x a a x a .
0
0 1 (
In the space provided write 4-5 of the rows aligned in such a way as to help you
envision what would happen if we decided to arrange the summation “by columns.”
bn c k
cn k a x ank , if k n n
If *nk , then it follows that
de f
0 , if k n
;
* ;
* ;
* ;
*
f x cn x
n
cn [x a a] n
*nk .
n0 n0 n0 k0
3* b 3* c 3* b3* c
In view of Lemma 9.2.4, n0 k0 *nk k0 n0 *nk whenever
;
* ;
n t u ;*
n
cn a x a
j n j
cn x a an *
n0 j0
j n0
i.e., at least when x a a R. Viewing

d the rearrangement asb “summing
c bye
columns,” yields that ¿rst column asK x a0 c0 a 0 c1 a 1 nn cn a n L
bc bc b n c n1
and the second column as x a1 10 c1 a 0 21 c2 a 1 n1 cn a .
In general, we have that the k 1 st column if given by
v tu t u w
k1 n
x a k
ck a
0
ck1 a
1
cn a nk

1 nk
Use the space that is provided to convince yourself concerning the form of the
general term.
Hence, for any x + U such that x a R a, we have that

;
* ;* t u
n
f x x ak cn a nk
k0 nk
n k

;
* ;
*
n!
x ak cn a nk
k0 nk
k! n k!

;
*
1 ;*
x ak n n 1 n 2 n k 1 a nk cn
k0
k! nk
;
*
f k a
x ak
k0
k!
as needed.
3
Theorem 9.2.6 (Identity Theorem) Suppose that the series * n
n0 an x and
3*
n0 bn x both converge in the segment S R R. If
n

;
* ;
*
E x+S: an x n bn x n
n0 n0
has a limit point in S, then 1n n + M C

0 " an bn and E S.
the Identity Theorem.
3* n
3* n
Proof. Suppose that the series n0 an x and n0 bn x both converge in the
segment S R R and that

;
* ;
*
E x+S: an x n bn x n
n0 n0
has a limit point in S. For each n + M C
0, let cn an bn . Then f x
3* n 0 for each x + E. Let
de f
n0 cn x
j k
A x + S : x + E ) and B S A
x + S : x + A
where E ) denotes the set of limit points of E. Note that S is a connected set such
that S A C B and A D B 3. First we will justify that B is open. If B is empty,
then we are done. If B is not empty and not open, then there exists a * + B such
that 2N= * N= * t B.
(1)
Next we will show that A is open. Suppose that x0 + A. Because x0 + S, by

Theorem 9.2.5,
;
*
f x dn x x0 n for .
n0 2
j k
Suppose that T j + M C
0 : d j
/ 0 / 3. By the , T has a
3
least element,
3* say k. It follows that we can write f x x x0 k g x where
g x m0 dkm x x 0 n for . Because g is continuous at x 0 ,
2
we know that lim g x / 0. Now we will make use of
xx 0
4 5
g x0
the fact that 0 to show that there exists = 0 such that g x / 0 for
2
x x0 =.
(6)
Hence, g x / 0 for x x0 = from which it follows that
f x x x0 k g x / 0
in . But this contradicts the claim that x0 is a limit point of zeroes of

7
f . Therefore, and we conclude that .
8 9
3
Thus, f x * n0 dn x x 0 0 for all x in a neighborhood N x 0 of x 0 .
n
Hence, N x 0 t A. Since x0 was arbitrary, we conclude that

1* * + A " i.e.,
10
.
11
Because S is a connected set for which A and B are open sets such that
S A C B, A / 3, and A D B 3, we conclude that .
12
***Acceptable responses are: (1) Your argument should have generated a sequence
of elements of E that converges to *. This necessitated an intermediate step be-
cause at each step you could only claim to have a point that was in E ) . For example,
if N= * is not contained in B, then there exists a ) + S such that ) + B which
places ) in E ) . While this does not place ) in E, it does insure that any neighbor-
hood of ) contains an element of E. Let u 1 be an element of E such that u 1 / *
and u 1 * =. The process can be continued to generate a sequence of elements
of E,
u n *
n1 , that converges to *. This would place * in A D B which contradicts
the choice of B. (2) x x0 R x0 , (3) Well-Ordering Principle, (4) g x0 ,
g x0
(5) dk , (6) We’ve seen this one a few times before. Corresponding to ,
2
there exists a = 0 such that x x0 = " g x g x0 . The (other)
g x0
triangular inequality, then yields that g x0 g x which implies
2
g x0
that g x whenever x x0 =. (7) 0 x x0 =, (8) T 3,
2
(9) 1n n + M C
0 " dn 0, (10) 2N * N * t A, (11) A is open, (12)
B is empty.***
9.3 Designer Series

With this section, we focus attention on one speci¿c power series expansion that
satis¿es some special function behavior. Thus far we have been using the de¿-
un is developed in most elementary calculus courses, namely, e
nitiontof e that
1
lim 1 . There are alternative approaches that lead us to e. In this section,
n* n
we will obtain e as the value of power series at a point. In Chapter t 3 of Rudin,
u e
3* 1 3* 1 1 n
was de¿ned as n0 and it was shown that n0 lim 1 . We
n! n! n* n
get to this point from work on a specially chosen power series. The series leads to
a de¿nition for the function e x and ln x as well as a “from series perspective” view
of trigonometric functions. b c
For each n + M, if cn n!1 , then lim sup cn1 cn 1 0. Hence, the
3 n*
Ratio Test yields that * n0 cn z n is absolutely convergent for each z + F. Conse-
quently, we can let

;
* n
z
E z for z + F. (9.7)
n0
n!
Complete the following exercises in order to obtain some general properties of

E z. If you get stuck, note that the following is a working excursion version
of a subset of what is done on pages 178-180 of our text.
From the absolute convergence of the power series given in (9.7), for any ¿xed
z * + F, the Cauchy product, as de¿ned in Chapter 4, of E z and E * can be
written as

;* n
z ;*
*n ;* ; n
z k * nk
E z E * .
n0
n! n0
n! n0 k0
k! n k!
bn c n!
From , it follows that
k k! n k!

;* ; ; n t u
n
1 n! *
1 ; n k nk
E z E * z *
k nk
z *
n0 k0
n! k! n k! n0
n! k0 k
;
*
z *n
.
n0
n!
9.3. DESIGNER SERIES 389
Therefore,
E z E * E z * . (9.8)
Suppose there exists a ? + F such that E ? 0. Taking z ? and * ? in

(9.8) yields that
E ? E ? E 0 1 (9.9)
which would contradict our second Property of the Additive Identity of a Field
(Proposition 1.1.4) from which we have to have that E ? E * 0 for all * + F.
Consequently 1z z + F " E z / 0.
1. For x real, use basic bounding arguments and ¿eld properties to justify each
of the following.
(a) 1x x + U " E x 0
(b) lim E x 0

x*
(c) 1x 1y [x y + U F 0 x y

" E x E y F E y E x]
What you have just shown justi¿es that E x over the reals is a strictly increas-
ing function that is positive for each x + U.
2. Use the de¿nition of the derivative to prove that

b c
1z z + F " E ) z E z
Note that when x is real, E ) x E x and 1x x + U " E x 0 with
the Monotonicity Test yields an alternative justi¿cation that E is increasing
in U.
A straight induction argument allows us to claim from (9.8) that

; n <n b c
1n n + M " E zj E zj . (9.10)
j1 j1
3. Complete the justi¿cation that

t un
1
E 1 lim 1
n* n
For each n + M, let
;n t u
1 1 n
sn and tn 1
k0
k! n
(a) Use the Binomial Theorem to justify that,
t u t ut u
1 1 1 1 2
tn 1 1 1 1 1
2! n 3! n n
t ut u t u
1 1 2 n1
1 1 1 .
n! n n n
(b) Use part (a) to justify that lim suptn n E 1.

n*
(c) For n m 2, justify that
t u
1 1
tn o 1 1 1
2! n
t ut u t u
1 1 2 m1
1 1 1 .
m! n n n
(d) Use the inequality you obtain by keeping m ¿xed and letting n *
in the equation from part (c) to obtain a lower bound on lim inftn and an
n*
upper bound on sm for each m.
(e) Finish the argument.
4. Use properties of E to justify each of the following claims.

(a) 1n n + M " E n en .
(b) 1u u + T F u 0 " E u eu
Using ¿eld properties and the density of the rationals can get us to a justi¿cation
that E x e x for x real.
x n1
5. Show that, for x 0, e x and use the inequality to justify that
n 1!
lim x n ex 0 for each n + M.
x*
9.3.1 Another Visit With the Logarithm Function

Because the function E U is strictly increasing and differentiable from U into
U
x + U : x 0, by the Inverse Function Theorem, E U has an inverse
function L : U U, de¿ned by E L y y that is strictly increasing and
differentiable on U . For x + U, we have that L E x x, for x real and the
Inverse Differentiation Theorem yields that
1
L ) y for y 0 (9.11)
y
where y E x. Since E 0 1, L 1 0 and (9.11) implies that

= y
dx
L y
1 x
which gets us back to the natural logarithm as it was de¿ned in Chapter 7 of these
notes. A discussion of some of the properties of the natural logarithm is offered on
pages 180-182 of our text.
9.3.2 A Series Development of Two Trigonometric Functions

The development of the real exponential and logarithm functions followed from re-
stricting consideration of the complex series E z to U. In this section, we consider
E z restricted the subset of F consisting of numbers that are purely imaginary. For
x + U,
;
*
i xn ;
*
in x n
E i x .
n0
n! n0
n!
Since

!
! 1 , if 4n !
! 1 , if 4n

i , if 4 n 1 i , if 4 n 1
in and in ,
!
! 1 , if 4 n 2 !
! 1 , if 4 n 2

i , if 4 n 3 i , if 4 n 3
it follows that each of
1 1
C x [E i x E i x] and S x [E i x E i x] (9.12)
2 2i
have real coef¿cients and are, thus, real valued functions. We also note that
E i x C x i S x (9.13)
from which we conclude that C x and S x are the real and imaginary parts of
E i x, for x + U.
Complete the following exercises in order to obtain some general properties of
C x and S x for x + U. If you get stuck, note that the following is a working
excursion version of a subset of what is done on pages 182-184 of our text. Once
completed, the list of properties justify that C x and S x for x + U correspond to
the cos x and sin x, respectively, though appeal to triangles or the normal geometric
view is never made in the development.
1. Show that E i x 1.
2. By inspection, we see that C 0 1 and S 0 0. Justify that C ) x

S x and S ) x C x.
b c
3. Prove that 2x x + U F C x 0 .
4. Justify that there exists a smallest positive real number x0 such that C x0
0.
5. De¿ne the symbol H by H 2x0 where x0 is the number from #4 and justify
each of the following claims.
rH s
(a) S 1
2
t u
Hi
(b) E i
2
(c) E Hi 1
(d) E 2Hi 1
It follows immediately from equation (9.8) that E is periodic with period 2Hi i.e.,
1z z + F " E z 2Hi E z .
Then the formulas given in equation (9.12) immediately yield that both C and S are
periodic with period 2Hi.
Also shown in Theorem 8.7 of our text is that 1t t + 0 2H " E it / 1
and
1z [z + F F z 1 " 2!t t + [0 2H F E i t z]
The following space is provided for you to enter some helpful notes towards justi-
fying each of these claims.
9.4. SERIES FROM TAYLOR’S THEOREM 397
9.4 Series from Taylor’s Theorem

The following theorem supplies us with a suf¿cient condition for a given function to
be representable as a power series. The statement and proof should be strongly rem-
iniscent of Taylor’s Approximating Polynomials Theorem that we saw in Chapter
6.
Theorem 9.4.1 (Taylor’s Theorem with Remainder) For a b, let I [a b].
Suppose that f and f j are in F I for 1 n j n n and that f n1 is de¿ned for
each x + Int I . Then, for each x + I , there exists a G with a G x such that
;n
f j a
f x x a j Rn x
j0
j !
f n1 G x an1

where Rn x is known as the Lagrange Form of the Re-
n 1!
mainder .
Excursion 9.4.2 Fill in what is missing to complete the following proof.
Proof. It suf¿ces to prove the theorem for the case x b. Since f and f j are
3 f j a
in F I for 1 n j n n, Rn f b nj0 b a j is well de¿ned. In
j!
order to ¿nd a different form of Rn , we introduce a function . For x + I , let
;n
f j x b xn1
x f b b x j Rn .
j0
j! b an1
From the hypotheses and the properties of continuous and functions,

1
we know that is and differentiable for each x + I . Furthermore,
2
a
3 4
and b 0. By , there exists a G + I such that

5
) G 0. Now
n v
; w
) f j x
x b x j1

j1
j 1! 6
n 1 b x n
Rn .
b an1
Because
;n
f j1 x ;n
f j1 x
b x j f ) x b x j
j0
j! j1
j !
;
n1 j
f x
)
f x b x j1
j2
j 1!
it follows that
) x n1bx
n
Rn
ban1 t u
3n
f x
j 3n1 f j x j 1
b x j1
b x
j2
j 1! j2
j 1!
.
7
f n1 G n 1 b Gn
If ) G 0, then b G n Rn . Therefore,
n! b an1
.
8
***Acceptable responses are: (1) differentiable, (2) continuous,

3
(3) f b nj0 f j!x b a j Rn , (4) 0, (5) Rolle’s or the Mean-Value The-
j
3 f j1 x f n1 x

orem, (6) nj0 b x j , (7) b xn ,
j! n!
f n1 G b an1
(8) Rn .***
n 1!
Remark 9.4.3 Notice that the inequality a b was only a convenience for framing
the argument i.e., if we have the conditions holding in a neighborhood of a point
: we have the Taylor’s Series expansion to the left of : and to the right of :. In
this case, we refer to the expansion as a Taylor’s Series with Lagrange Form of the
Remainder about :.
Corollary 9.4.4 For : + U and R 0, suppose that f and f j are in

F : R : R for 1 n j n n and that f n1 is de¿ned for each x +
: R : R. Then, for each x + : R : R, there exists a
G + : R : R such that
; n
f j :
f x x : j Rn
j 0
j !
f n1 G x :n1

where Rn .
n 1!
9.4.1 Some Series To Know & Love

When all of the derivatives of a given function are continuous in a neighborhood
of a point :, the Taylor series expansion about : simply takes the form f x
3* f j :
j0 x : j with its radius of convergence being determined by the
j!
behavior of the coef¿cients. Alternatively, we can justify the series expansion by
proving that the remainder goes to 0 as n *. There are several series expansions
that we should just know and/or be able to use.
Theorem 9.4.5
(a) For all real : and x, we have

;*
x :n
:
e e
x
(9.14)
n0
n!
(b) For all real : and x, we have

b c
;*
sin : nH
sin x 2
x :n (9.15)
n0
n!
and
b c
;*
cos : nH
cos x 2
x :n . (9.16)
n0
n!
(c) For x 1, we have
;
*
1n1 x n
ln 1 x (9.17)
n1
n
and
;
*
1n1 x 2n1
arctan x . (9.18)
n1
2n 1
(d) The Binomial Series Theorem. For each m + U1 and for x 1, we have
;
*
m m 1 m 2 m n 1
1 x 1
m
xn. (9.19)
n1
n!
We will offer proofs for (a), and the ¿rst parts of (b) and (c). A fairly complete
sketch of a proof for the Binomial Series Theorem is given after discussion of a
different form of Taylor’s Theorem.
Proof. Let f x e x . Then f is continuously differentiable on all of U and

f n x e x for each n + M. For : + U, from Taylor’s Theorem with Remainder,
we have that
;n
1 eG x :n1
f x e x e: x : j Rn : x where Rn
j0
j! n 1!
where G is between : and x. Note that

n n
n ; *
x :n n
n : n
0 n ne f xn Rn .
n n0 n! n
Furthermore, because x : with : G x implies that eG e x , while x a

yields that x G : and eG e: ,

x x :
n1
!
!
!
! e , if x o :
eG x :n1 n 1!
Rn n .
n 1! !
!
!
! x : n1
e: , if x :
n 1!
kn
Since lim 0 for any ¿xed k + U, we conclude that Rn 0 as n *.
n* n!
;*
x :n
From the Ratio Test, is convergent for all x + U. We conclude that
n0
n!
the series given in (9.14) converges to f for each x and :.
The expansion claimed in (9.17) follows from the Integrability of Series

because
= x ;*
dt 1
ln 1 x and 1n t n for t 1.
0 1t 1t n0
There are many forms of the remainder for “Taylor expansions” that appear in
the literature. Alternatives can offer different estimates for the error entailed when
a Taylor polynomial is used to replace a function in some mathematical problem.
The integral form is given with the following
Theorem 9.4.6 (Taylor’s Theorem with Integral Form of the Remainder)
Suppose that f and its derivatives of order up to n 1 are continuous on a segment
3n f j :x: j
I containing :. Then, for each x + I , f x j0 j! Rn : x
where
= x
x tn n1
Rn : x f t dt
: n!
Proof. Since f ) is continuous on the interval I , we can integrate the derivative

to obtain
= x
f x f : f ) t dt.
:
As an application of Integration-by-Parts, for ¿xed x, corresponding to u f ) t

and d) dt, du f )) t dt and we can choose ) x t. Then
= x = x
) )
f x f : f t dt f : f t x t t:
tx
x t f )) t dt
: :
= x
f : f ) : x : x t f )) t dt.
:
Next suppose that
; = x
k
f j : x : j x tk k1
f x f t dt
n0
j! : k!
and f k1 is differentiable on I . Then Integration-by-Parts can be applied to

5 x xtk k1 k1 t and d) x t dt leads to u
k
: k! f t dt taking u f
k!
x t k1
f k2 t dt and ) . Substitution and simpli¿cation justi¿es the
k 1!
claim.
As an application of Taylor’s Theorem with Integral Form of Remainder, com-

plete the following proof of the The Binomial Series Theorem.
Proof. For ¿xed m + U1 and x + U such that x 1, from Taylor’s Theorem

with Integral Form of Remainder, we have
;
k
m m 1 m 2 m n 1
1 x 1
m
x n Rk 0 x .
n1
n!
where
= x
x tk k1
Rk 0 x f t dt
0 k!
We want to show that

= x
x tk
Rk 0 x m m 1 m k 1 tmk1 dt 0 as k *
0 k!
for all x such that x 1. Having two expressions in the integrand that involve a
power k suggests a rearrangement of the integrand i.e.,
= x t u
m m 1 m k x t k
Rk 0 x 1 tm1 dt.
0 k! 1t
We discuss the behavior of 1 tm1 , when t is between 0 and x, and

t u
5x x t k
dt separately.
0 1t
On one hand, we have that
1 tm1 n 1 whenever m o 1 F 1 t n 0 G m n 1 F 1 t o 0 .
On the other hand, because t is between 0 and x, if m o 1Fx o 0 or m n 1Fx n 0,

then

0 for m 1
)
g t 1 tm1
implies that g t m 1 1 t m2
.

0 for m 1
Consequently, if m o 1 F x o 0, then 0 t x and g increasing yields the
g t n g x while m n 1 F x n 0, 0 t x and g decreasing, implies that
g x o g t. With this in mind, de¿ne Cm x, for x 1 by
|
1 xm1 , m o 1, x o 0 OR m n 1, x n 0
Cm x .
1 , m o 1, x 0 OR m n 1, x o 0
We have shown that
1 tm1 Cm t n Cm x , for t between 0 and x (9.20)
t u
5x x t k
Next, we turn to 0 dt. Since we want to bound the behavior in
1t
terms of x or a constant, we want to get the x out of the limits of integration. The
standard way to do this is to effect a change of variable. Let t xs. Then dt xds
and
= xt u = 1 t u
x t k 1s k
dt x k1
ds.
0 1t 0 1 xs
t uk
1s
Since s 1 x o 0, we immediately conclude that n 1. Hence, it
1 xs
follows that
n= t n
n x x t uk n
n n
n dt n n xk1 . (9.21)
n 0 1t n
From (9.20) and (9.21), if follows that
0 n Rk 0 x
= 1
m m 1 m k k1
n x C m x dt
0 k!
m m 1 m k k1
x Cm x .
k!
m m 1 m k k1 3
For u k x x Cm x consider *
n1 u n x. Be-
cause k!
n n n n
n u n1 x n n m n
n nn n
n u x n n n 1 1n x x as n *,
n
3*
n1 u n x is convergent for x 1. From the nth term test, it follows that
u k x 0 as k * for all x such that x 1. Finally, from the Squeeze
Principle, we conclude that Rk 0 x 0 as k * for all x with x 1.
9.4.2 Series From Other Series

There are some simple substitutions into power series that can facilitate the deriva-
tion of series expansions from some functions for which series expansions are
“known.” The proof of the following two examples are left as an exercise.
3*
Theorem 9.4.7 Suppose that f u n0 cn u bn for u b R with
R 0.
3*
(a) If b kc d with k / 0, then f kx d n0 cn k
n x cn for
R
x c .
k
d e 3
(b) For every ¿xed positive integer k, f x ck b *n0 cn x c for
kn
x c R 1k .
The proofs are left as an exercise.
We close this section with a set of examples.
1
Example 9.4.8 Find the power series expansion for f x about the
1 x2
1
point : and give the radius of convergence.
2
Note that
v w
1 1 1 1
f x
1 x 1 x 2 1 x 1 x

1t 1 1

t uu t t uu
2 1 1 3 1
x x
2 2 2 2
1 1 1
t t uu t t uu.
1 3 2 1
12 x 1 x
2 3 2
t u n t un n n
1 ;*
1 n n 1 nn n 1 nn
Since t t uu 2 x
n n
for n2 x n
1 or nx n
1 2 2 n 2
12 x n0
2 n t
;* t un t u un
1 1 2 1 n n2 1 nn
and t t uu 1 n
x n
for n x 1 or
2 2 1 3 2 3 2 n
1 x n0
n n 3 2 n n
n n 3 n n
nx n . Because both series expansions are valid in nx n 1 , it follows
1 1
n 2n 2 n 2n 2
that
;* t ut u n n
2n 1 n n 1 n 1
f x 2 n1
n
x for nnx nn .
n0
3 2 2 2
Example 9.4.9 Find the power series expansion for g x arcsin x about the
point : 0. = x
dt
We know that, for x 1, arcsin x T . From the Binomial
0 1 t2
1
Series Theorem, for m , we have that
t 2u t u t u
1 1 1
;* 1 n 1
12 2 2 2
1 u 1 u n for u 1. Since
n!
n1n n
u 1 if and only if nu 2 n 1, it follows that
t ut u t u
1 1 1
r s12 *
; 1 n 1
2 2 2
1 t2 1 1n t 2n for t 1.
n1
n!
Note that
t ut u t u t ut u t u
1 1 1 1 1 1
1 n 1 1
n
1 n 1
2 2 2 2 2 2
_ ^] `
n terms
1 3 2n 1
.
2n
Consequently,
r s12 ;
*
1 3 2n 1
1t 2
1 t 2n for t 1
n1
2n n!
with the convergence being uniform in each t n h for any h such that 0 h 1.
Applying the Integration of Power Series Theorem (Theorem 9.1.13), it follows that
= ;*
x dt 1 3 2n 1 2n1
arcsin x T x x , for x 1
0 1 t2 n1
2n 1 2n n!
where arcsin 0 0.
Excursion 9.4.10 Find the power series expansion about : 0 for f x
e x ex
cosh x and give the radius of convergence.
2
***Upon noting that f 0 1, f ) 0 0, f 2n x f x and f 2n1 x
;*
x 2n
)
f x, it follows that we can write f as for all x + U.***
n0
2n!
Example 9.4.11 Suppose that we want the power series expansion for f x
ln cos x about the point : 0. Find the Taylor Remainder R3 in both the
Lagrange and Integral forms.
f 4 G 4
Since the Lagrange form for R3 is given by x for 0 G x, we
4!
have that
b c
4 sec2 G tan2 G 2 sec4 G x 4
R3 for 0 G x.
24
= x
x tn n1
In general, the integral form is given by Rn : x f t dt. For
: n!
this problem, : 0 and n 3, which gives
= x
x t3 r s
R3 : x 4 sec2 t tan2 t 2 sec4 t dt
: 6
Excursion 9.4.12 Fill in what is missing in the following application of the geo-
metric series expansion and the theorem on the differentiation of power series to
3 3n 1
¿nd * n1 .
4n
;
*
x
Because xn for x 1, it follows that
n1
1 x
;* t un
1

n1
4 1
From the theorem on differentiation of power series,

;* t u)
x
nx x
n

n1
1x 2
in x 1. Hence,
;
*
n
3 .
n1
4n 3
Combining the results yields that

; * t u
*
3n 1 ; n 1
n
3 n n
n1
4 n1
4 4 4
1 4
***Expected responses are: (1) , (2) x 1 x2 , (3) , and (4) 1.***
3 3
9.5 Fourier Series

Our power series expansions are only useful in terms of representing functions that
are nice enough to be continuously differentiable, in¿nitely often. We would like
to be able to have series expansions that represent functions that are not so nicely
behaved. In order to obtain series expansions of functions for which we may have
only a ¿nite number of derivatives at some points and/or discontinuities at other
points, we have to abandon the power series form and seek other “generators.” The
set of generating functions that lead to what is known as Fourier series is
1 C
cos nx : n + M C
sin nx : n + M.
De¿nition 9.5.1 A trigonometric series is de¿ned to be a series that can be written
in the form
1 ;*
a0 an cos nx bn sin nx (9.22)
2 n1
where
an * *
n0 and
bn n1 are sequences of constants.
9.5. FOURIER SERIES 409
De¿nition 9.5.2 A trigonometric polynomial is a ¿nite sum in the form

;
N
ck eikx , x + U (9.23)
kN
where ck , k N N 1 N 1 N , is a ¿nite sequence of constants.
Remark 9.5.3 The trigonometric polynomial given in (9.23) is real if and only if
cn cn for n 0 1 N .
Remark 9.5.4 It follows from equation (9.12) that the Nth partial sum of the trigono-
metric series given in (9.22) can be written in the form given in (9.23). Conse-
1 ;N
quently, a sum in the form a0 ak cos kx bk sin kx is also called a trigono-
2 k0
metric polynomial. The form used is often a matter of convenience.
The following “orthogonality relations” are sometimes proved in elementary

calculus courses as applications of some methods of integration:

= H = H H , if m n
cos mx cos nxdx sin mx sin nxdx
H H
0 , if m / n
and
= H
cos mx sin nxdx 0 for all m n + M.
H
We will make use of these relations in order to ¿nd useful expressions for the coef-
¿cients of trigonometric series that are associated with speci¿c functions.
Theorem 9.5.5 If f is a continuous function on I [H H] and the trigonometric

1 3
series a0 * n1 an cos nx bn sin nx converges uniformly to f on I , then
2
=
1 H
an f t cos nt dt for n + M C
0 (9.24)
H H
and
= H
1
bn f t sin nt dt. (9.25)
H H
1 3
Proof. For each k + M, let sk x a0 km1 am cos mx bm sin mx and
2
suppose that 0 is given. Because sk f there exists a positive integer M such
I
that k M implies that sk x f x for all x + I . It follows that, for each
¿xed n + M,
sk x cos nx f x cos nx sk x f x cos nx n sk x f x
and
sk x sin nx f x sin nx sk x f x sin nx n sk x f x
for all x + I and all k M. Therefore, sk x cos nx f x cos nx and
I
sk x sin nx f x sin nx for each ¿xed n. Then for ¿xed n + M,
I
1 ;*
f x cos nx a0 cos nx am cos mx cos nx bm sin mx cos nx
2 m1
and
1 ;*
f x sin nx a0 sin nx am cos mx sin nx bm sin mx sin nx
2 m1
the uniform convergence allows for term-by-term integration over the interval [H H]
which, from the orthogonality relations yields that
= H = H
f x cos nx d x Han and f x sin nx dx Hbn .
H H
De¿nition 9.5.6 If f is a continuous function on I [H H ] and the trigonomet-

1 3
ric series a0 * n1 an cos nx bn sin nx converges uniformly to f on I , then
2
the trigonometric series
1 ;*
a0 an cos nx bn sin nx
2 n1
is called the Fourier series for the function f and the numbers an and bn are called
the Fourier coef¿cients of f .
Given any Riemann integrable function on an interval [H H], we can use the
formulas given by (9.24) and (9.25) to calculate Fourier coef¿cients that could be
associated with the function. However, the Fourier series formed using those co-
ef¿cients may not converge to f . Consequently, a major concern in the study of
Fourier series is isolating or describing families of functions for which the associ-
ated Fourier series can be identi¿ed with the “generating functions” i.e., we would
like to ¿nd classes of functions for which each Fourier series generated by a func-
tion in the class converges to the generating function.
The discussion of Fourier series in our text highlights some of the convergence
properties of Fourier series and the estimating properties of trigonometric polyno-
mials. The following is a theorem that offers a condition under which we have
pointwise convergence of the associated Fourier polynomials to the function. The
proof can be found on pages 189-190 of our text.
Theorem 9.5.7 For f a periodic function with period 2H that is Riemann inte-
grable on [H H], let
;
N = H
1
s N f x cm e imx
where cm f t eimt dt.
mN
2H H
If, for some x, there are constants = 0 and M * such that
f x t f x n M t
for all t + = =, then lim s N f x f x.

N *
The following theorem that is offered on page 190 of our text can be thought of
as a trigonometric polynomial analog to Taylor’s Theorem with Remainder.
Theorem 9.5.8 If f is a continuous function that is periodic with period 2H and

0, then there exists a trigonometric polynomial P such that P x f x
for all x + U.
For the remainder of this section, we will focus brieÀy on the process of ¿nding
Fourier series for a speci¿c type of functions.
De¿nition 9.5.9 A function f de¿ned on an interval I [a b] is piecewise contin-

uous on I if and only if there exists a partition of I ,
a x0 x1 xn1 xn b
such that (i) f is continuous on each segment xk1 xk and (ii) f a, f b
and, for each k +
1 2 n 1 both f xk and f xk exist.
De¿nition 9.5.10 If f is piecewise continuous on an interval I and xk + I is a

point of discontinuity, then f x k f x k is called the jump at xk . A piecewise
continuous function on an interval I is said to be standardized if the values at points
1d e
of discontinuity are given by f xk f xk f xk .
2
Note that two piecewise continuous functions that differ only at a ¿nite number
of points will generate the same associated Fourier coef¿cients. The following
¿gure illustrates a standardized piecewise continuous function.
y
x
0 x1 x2 ... xn-2 xn=b
a=x0 xn-1
De¿nition 9.5.11 A function f is piecewise smooth on an interval I [a b] if and

only if (i) f is piecewise continuous on I , and (ii) f ) both exists and is piecewise
continuous on the segments corresponding to where f is continuous. The function
f is smooth on I if and only if f and f ) are continuous on I .
De¿nition 9.5.12 Let f be a piecewise continuous function on I [H H]. Then

the periodic extension f of f is de¿ned by

!
! f x , if H n x H
!
!

f x f H f H
! , if x H G x H ,
!
! 2
!
f x 2H , if x + U
f x f x
where f is continuous and by f x an each point of discon-
2
tinuity of f in H H .
It can be shown that, if f is periodic with period 2H and piecewise smooth

on [H H ], then the Fourier series of f converges for every real number x to the
f x f x
limit . In particular, the series converges to the value of the given
2
function f at every point of continuity and to the standardized value at each point
of discontinuity.
Example 9.5.13 Let f x x on I [H H]. Then, for each j + ], the
periodic extension f satis¿es f jH 0 and the graph in each segment of the
form j H j 1 H is identical to the graph in H H. Use the space provided
to sketch a graph for f .
The associated Fourier coef¿cients for f are given by (9.24) and (9.25) from
Theorem 9.5.5. Because t cos nt is an odd function,
=
1 H
an t cos nt dt 0 for n + M C
0 .
H H
According to the formula for integration-by parts, if n + M, then

= =
t cos nt 1
t sin nt dt cos nt dt C
n n
for any constant C. Hence, cos nH 1n for n + M yields that

!
!
2
, if 2 0 n
= H = H !

1 2 n
bn t sin nt dt t sin nt dt .
H H H 0 !
!
!

2
, if 2 n
n
Thus, the Fourier series for f is given by
;
*
sin nx
2 1n1 .
n1
n
The following ¿gure shows the graphs of f , s1 x 2 sin x, and s3 x 2 sin x
2
sin 2x sin 3x in 3 3.
3
3
-3 -2 -1 0 1 2 3
x
-1
-2
-3
37 sin nx
while the following shows the graphs of f and s7 x 2 n1 1
n1
n
in 3 3.
-3 -2 -1 0 1 2 3
x
-1
-2
-3
Example 9.5.14 Find the Fourier series for f x x in H n x n H. Note

that, because f is an even function, f t sin nt is odd.
***Hopefully, you noticed that bn 0 for each n + M and an 0 for each even
natural number n. Furthermore, a0 H while, integration-by-parts yielded that
an 4n 2 H 1 for n odd.***
v shows f x wx and the corresponding Fourier polyno-

The following ¿gure
H 4 1
mial s3 x cos x cos 3x in 3 3.
2 H 9
2.5
1.5
0.5
-3 -2 -1 0 1 2 3
x
We close with a ¿gure that shows f x x and the corresponding Fourier
H 4; 7
1
polynomial s7 x cos 2n 1 x in 3 3. Note how the
2 H n1 2n 12
difference is almost invisible to the naked eye.
2.5
1.5
0.5
-3 -2 -1 0 1 2 3
x
9.6 Problem Set I

1. Apply the Geometric Series Expansion Theorem to ¿nd the power series ex-
3
pansion of f x about : 2 and justify where the expansion is
4 5x
valid. Then verify that the coef¿cients obtained satisfy the equation given in
part (c) of Theorem 9.1.13.
2. Let
| b c
exp 1x 2 , x / 0
g x
0 , x 0
where exp * e* .
c for each n +
(a) Use the Principle of Mathematical Induction tob prove that,
M and x + U
0, g n x x 3n Pn x exp 1x 2 where Pn x is
a polynomial.
(b) Use l’Hôpital’s Rule to justify that, for each n + MC
0, g n 0 0.
3. Use the Ratio Test, as stated in these Companion Notes, to prove Lemma
9.1.8.
4. For each of the following use either the Root Test or the Ratio Test to ¿nd the
“interval of convergence.”
;
*
7xn
(a)
n0
n!
;*
(b) 3n x 1n
n0
;*
x 2n
(c) T
n
n0
n
;*
n!2 x 3
n
(d)
n0
2n!
;*
ln n 3n x 1n
(e) T
nn n
n0
5
9.6. PROBLEM SET I 417
;* t u
ln n 1 2n x 1n 3 1
5. Show that is convergent in .
n0
n 1 2 2
6. For each of the following, derive the power series expansion about the point
: and indicate where it is valid. Remember to brieÀy justify your work.
3x 1
(a) g x :1
3x 2
(b) h x ln x : 2
7. For each of the following, ¿nd the power series expansion about : 0.
b c12
(a) f x 1 x 2
(b) f x 1 x2
(c) f x 1 x3
b c
(d) f x arctan x 2
r T s
8. Find the power series expansion for h x ln x 1 x 2 about : 0
and its interval of convergence. (Hint: Consider h ) .)
3
9. Prove that if f u * n0 cn u b for u b R with R
n
0 and b
3* R
kc d with k / 0, then f kx d n0 cn k n x cn for x c .
k
3*
d that kif f u
10. Prove e bn for u b R with R
3* n0 cn u kn 0, then
f x c b n0 cn x c in x c R 1k for any ¿xed posi-
tive integer k.
11. Find the power series expansions for each of the following about the speci¿ed
point :.
(a) f x 3x 52 : 1

H
(b) g x sin x cos x :
4
t u
x
(c) h x ln : 2
1 x2
;
*
12. Starting from the geometric series x n x 1 x1 for x 1, derive
n1
closed form expressions for each of the following.
;
*
(a) n 1 x n
n1
;
*
(b) n 1 x 2n
n1
;
*
(c) n 1 x n2
n1
;
*
n1
(d) x n3
n1
n3
13. Find each of the following, justifying your work carefully.
;
* 2
n 2n 1
(a)
n1
3n
;
*
n 3n 2n
(b)
n1
6n
14. Verify the orthogonality relations that were stated in the last section.

5H 5H H , if m n
(a) H cos mx cos nxd x H sin mx sin nxdx .

0 , if m / n
5H
(b) H cos mx sin nxd x 0 for all m n + M.
15. For each of the following, verify that the given Fourier series is the one asso-
ciated with the function f according to Theorem 9.5.5.

0 , if H n x 0
1 2;*
sin 2k 1 x
(a) f x
2 H k0 2k 1
1 , if 0 n x n H
9.6. PROBLEM SET I 419
H2 ;*
cos kx
(b) f x x 2 for x + [H H] 4 1k
3 k1
k2
1 cos 2x
(c) f x sin2 x for x + [H H]
2 2
Index
a b, 74 arc

Un , 74 in Un , 318
A) , 92 Archimedean principle
[a b], 81 for real numbers, 28
Int A, 92 argument, 36
A, 92
" A, 92 binary operation, 2
F, 3 Bolzano-Weierstrass theorem, 142
F X, 350 boundary
D x y, 78 of a set, 90
F, 34 bounded
d* x y, 78 a set is, 94
M, 17 above, 19
S d, 78 below, 22
Q, 16 sequence, 128
S X, 52 set, 204
T, 3 Boundedness Theorem, 204
U , 29
Cantor set, 115
I f g, 350
Cantor’s theorem, 58
I* f g, 350
* Cartesian product, 1, 73
a
3n*n1 , 123 Cauchy criteria for series convergence,
k0 ak , 152
155
4 :, 287
Cauchy Criterion for Convergence, 335
absolutely convergent Cauchy product, 170
sequence of functions, 333 Cauchy sequence, 135
absolutely convergent series, 156 Chain Rule, 241
alternating series test, 174 Change of Variables, 307
analytic characteristic function, 302
at a point, 379 class of monotonic functions, 215
on a set, 379 closed ball, 82
421
422 INDEX
closed curve of rational numbers, 29

in Un , 318 of real numbers, 29
closed set, 83 denumerable, 56
closure derived set, 90
of a set, 90 diameter
common ratio, 157 of a set, 135
compact, 98 difference
compact of two sets, 88
normal family of functions, 362 differentiable
comparison test, 156 at a point, 262
complete left-hand, 230
metric space, 143 on a closed interval, 231
composition right-hand, 230
of relations, 52 differentiable at a point, 229
conditionally convergent series, 156 Differentiation and Integration of Power
conjugate, 36 Series, 376
connected discontinuity
set, 112 at a point, 211
continuous of the ¿rst kind, 214
function, 197 of the second kind, 214
function at a point, 183, 197 simple, 214
in a neighborhood of a point, 399 discontinuous
uniformly, 208 at a point, 211
Convergence properties
of p-series, 159 equicontinuous family of functions, 357
convergence properties equivalent
of geometric series, 157 cardinally, 54
convex set, 82 metrics, 79
countable Euclidean metric, 78
at most, 56 Euclidean n-space, 75
countably in¿nite, 56 Euclidean norm, 75
curve exterior
in Un , 318 of a set, 90
exterior point, 90
deMorgan’s Laws, 88 Extreme Value Theorem, 205
dense
a set is, 94 family of functions
density equicontinuous, 357
INDEX 423
locally uniformly bounded, 353 ordered ¿eld, 15

normal, 362 in¿mum, 22
uniformly bounded, 353 in¿nite series, 152
¿eld, 2 in¿nite set, 56
ordered, 12 in¿nitely differentiable
¿nite intersection property, 103 at a point, 379
¿nite sequence, 61 inner product, 75
¿nite set, 56 integers
Fourier coef¿cients, 410 of an ordered ¿eld, 15
Fourier series, 410 Integrability Criterion, 288
function, 2 integrable, 283
differentiable at a point, 229 Integration of Uniformly Convergent
global maximum, 245 Sequences, 343
global minimum, 245 interior
local maximum, 245 of a set, 90
local minimum, 245 Interior Extrema Theorem, 245
monotonically decreasing, 215 interior point, 83
monotonically increasing, 215 Intermediate Value Theorem, 207
natural logarithm, 314 intersection
periodic extension, 412 of a family of sets, 63
piecewise continuous, 411 inverse
piecewise smooth, 412 of a relation, 52
smooth, 412 Inverse Differentiation Theorem, 256
Fundamental Lemma of Differentia- Inverse Function Theorem, 254
tion, 240 isolated point, 83
Fundamental Theorem of Calculus
First, 308 jump
Second, 310 of a piecewise continuous func-
tion, 412
Generalized Mean-Value Theorem, 247
geometric series, 157 kth term test, 153
greatest lower bound, 22 L’Hôpital’s Rule I, 251
greatest lower bound property, 25 L’Hôpital’s Rule II, 253
Hadamard product, 170 Lagrange Form of the Remainder, 397
Heine-Borel theorem, 109 least upper bound, 19
least upper bound property, 22
Identity Theorem, 385 left-hand limit, 213
inductive limit
424 INDEX
function n-tuple, 73, 74

from metric space to metric space, natural logarithm function, 314
188 natural numbers
from metric space to U1 , 187 of an ordered ¿eld, 15
in the extended reals, 222 neighborhood
inferior, 146 of a point, 82
left-hand, 213 nested interval theorem, 105
of a function at a point, 183, 184 nested sequence of sets, 103
of a sequence, 125 norm
of sequence of real numbers, 145 sup, 350
right-hand, 213 supremum, 350
subsequential, 131 normal family, 362
superior, 146
Limit Comparison Test, 176, 180 one-to-one, 49
limit comparison test, 162 one-to-one correspondence, 49
limit point, 83 onto, 49
local maximum, 245 open
local minimum, 245 relative to, 96
locally uniformly bounded, 353 open ball, 82
locus of points, 41 open cover, 98
lower bound, 22 open set, 84
lower Riemann sum, 278 Open Set Characterization
of Continuous Functions, 201
Mean-Value Theorem, 247
order, 11
Mean-Value Theorem for Integrals, 307
ordered ¿eld, 12
Mertens theorem, 171
ordered pair, 74
metric, 77
discrete, 83 pairwise disjoint, 63
metric space, 78
partition, 277
modulus, 36
common re¿nement of, 279
monotonic sequences
mesh of, 277
class of, 144
re¿nement of, 279
monotonically decreasing, 215
subdivision of, 277
sequence of real numbers, 144
PCI, 16
monotonically increasing, 215
perfect
sequence of real numbers, 144
set, 114
Monotonicity Test, 248
periodic extension, 412
n-cell, 81 piecewise continuous, 411
INDEX 425
piecewise continuous function Riemann-Stieltjes sum

jump at a point, 412 lower, 286
standardized, 412 upper, 286
piecewise smooth function, 412 right-hand limit, 213
PMI, 16 Rolle’s Theorem, 246
pointwise convergence root test, 163
sequence of functions, 328
pointwise convergent, 326 scalar multiplication, 74
power series, 370 Schwarz inequality, 38
sequence of functions, 332, 334 secont derivative, 259
power series separated
divergent, 370 two sets are, 112
in U, 369 sequence, 61
pointwise convergent, 370 Cauchy, 135
Principle of Complete Induction, 16 of nth partial sums, 152
Principle of Mathematical Induction, subsequence, 131
16 sequence of functions
converges point wise, 328
Ratio Test, 162 converges uniformly, 328
rational numbers pointwise convergent, 326, 334
of an ordered ¿eld, 17 uniformly convergent, 335
real n-space, 74 Sequences Characterization for Lim-
real vector space, 74 its of Functions, 191
recti¿able, 319 series
Restrictions of Integrable Functions, Fourier, 410
301 in¿nite, 152
Riemann integrable, 283 power, 369
Riemann integral, 283 term in a, 152
upper, 283 set descriptions, 39
Riemann-Stieltjes integrable, 287 smooth function, 412
vector-valued function, 315 squeeze principle lemma, 142
Riemann-Stieltjes integral standardized
lower, 287 piecewise continuous function, 412
upper, 287 subsequence, 131
Riemann-Stieltjes Integrals subsequential limit, 131
Algebraic Properties of, 299 summation-by-parts, 173
Properties of Upper and Lower, sup norm, 350
297 supremum, 19
426 INDEX
supremum norm, 350 Weierstrass theorem, 111

Well-Ordering Principle, 16
Taylor’s Approximating Polynomials, WOP, 16
259
Taylor’s Theorem with Integral Form
of the Remainder, 401
Taylor’s Theorem with Remainder, 397
third derivative, 259
Triangular inequalities, 76
triangular inequalities, 37
trigonometric polynomial, 409
trigonometric series, 408
uncountable, 56
uniform continuity theorem, 209
uniform convergence
uniform convergent
Uniform Limit of Continuous Func-
tions, 341
uniformly bounded
family of functions, 353
uniformly Cauchy, 335
uniformly continuous, 208
uniformly convergent
union
of a family of sets, 63
upper bound, 19
upper Riemann integral, 283
upper Riemann sum, 278
vector addition, 74
vector-valued function
differentiable at a point, 262
Riemann-Stieltjes integrable, 315
Weierstrass M-Test, 338

Rudin Companion Mathematical Analysis - Silvia

Uploaded by

Copyright:

Available Formats

Rudin Companion Mathematical Analysis - Silvia

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rudin Companion Mathematical Analysis - Silvia

Uploaded by

Copyright:

Available Formats

COMPANION NOTES

A Working Excursion to Accompany Baby

1 The Field of Reals and Beyond 1

2 From Finite to Uncountable Sets 49

3 METRIC SPACES and SOME BASIC TOPOLOGY 73

3.2 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4 Sequences and Series–First View 123

5 Functions on Metric Spaces and Continuity 183

6 Differentiation: Our First View 229

7 Riemann-Stieltjes Integration 275

8 Sequences and Series of Functions 325

9 Some Special Functions 369

9.4.1 Some Series To Know & Love . . . . . . . . . . . . . . . . 399

 encourage or engender an interactive approach to learning the material,

 provide more examples at the introductory level,

 offer some alternative views of some of the concepts, and

 draw a clearer connection to the mathematics that is prerequisite to under-

0.0.1 About the Organization of the Material

0.0.2 About the Errors

The Field of Reals and Beyond

 The Cartesian product of two sets A and B, denoted by A  B, is

 A function h from A into B is a subset of A  B such that

 A binary operation on a set A is a function from A  A into A.

 A ¿eld is an algebraic structure, denoted by I   e f , that includes a

(A) Axioms of Addition (I  e is a commutative group under the binary

(D) The Distributive Law

1x 1y 1z x y z + I " [x  y  z  x  y  x  z]

When discussing ¿elds, we should distinguish that which can be claimed as

Proposition 1.1.4 Properties for the Additive Identity of a ¿eld I   e f 

Proof. (of #1) Suppose that x + I satis¿es x  x  x. Since x + I, by the

e  x  x  x  x  x  x  x  x  x  e  x.

Hence, for x y + I, x  y  e F x  e implies that y  e. The claim now follows

Excursion 1.1.5 Use #1 to prove #2.

***The key here was to work from x  e  x e  e.***

Proposition 1.1.6 Uniqueness of Identities and Inverses for a ¿eld I   e f 

1. The additive identity of a ¿eld is unique.

2. The multiplicative identity of a ¿eld is unique.

Proof. (of #1) Suppose that * + I is such that

In particular, since e + I, we have that e  e  *. Since e is given as an additive

a* *a e and a  x  x  a  e.

From the additive identity and associative properties

Excursion 1.1.7 Prove #4.

***Completing this excursion required only appropriate modi¿cation of the proof

1. 1a 1b a b + I " a  b  a  b

2. 1a a + F " a  a

3. 1a 1b a b + I " a  b  a  b

4. 1a 1b a b + I " a  b  a  b

5. 1a 1b a b + I " a  b  a  b

Proof. (of #2) Suppose that a + I. By the additive inverse property a + I

***Acceptable responses are: (1) a a 1  f , (2) e  f , (3) multiplicative identity,

Proposition 1.1.10 Solutions to Linear Equations. Given a ¿eld I   0 1,

1. 1a 1b a b + I " 2!x x + I F a  x  b

2. 1a 1b a b + IFa / 0 " 2!x x + I F a  x  b

Proof. (of #1) Suppose that a b + I and a / 0. Since a + I 

Hence, x satis¿es a  x  b. Now, suppose that * + I also satis¿es a  *  b. Then

Since a and b were arbitrary,

1a 1b a b + I " 2!x x + I F a  x  b .

Remark 1.1.11 As a consequence of Proposition 1.1.10, we now can claim that, if

Proposition 1.1.12 Addition and Multiplication Over Fields Containing Three or

encourage or engender an interactive approach to learning the material,

provide more examples at the introductory level,

offer some alternative views of some of the concepts, and

draw a clearer connection to the mathematics that is prerequisite to under-

The Cartesian product of two sets A and B, denoted by A B, is

A function h from A into B is a subset of A B such that

A binary operation on a set A is a function from A A into A.

A ¿eld is an algebraic structure, denoted by I e f , that includes a

(A) Axioms of Addition (I e is a commutative group under the binary

1x 1y 1z x y z + I " [x y z x y x z]

Proposition 1.1.4 Properties for the Additive Identity of a ¿eld I e f

Proof. (of #1) Suppose that x + I satis¿es x x x. Since x + I, by the

e x x x x x x x x x e x.

Hence, for x y + I, x y e F x e implies that y e. The claim now follows

The key here was to work from x e x e e.

Proposition 1.1.6 Uniqueness of Identities and Inverses for a ¿eld I e f

In particular, since e + I, we have that e e *. Since e is given as an additive

a* *a e and a x x a e.

1. 1a 1b a b + I " a b a b

2. 1a a + F " a a

3. 1a 1b a b + I " a b a b

4. 1a 1b a b + I " a b a b

5. 1a 1b a b + I " a b a b

Proof. (of #2) Suppose that a + I. By the additive inverse property a + I

***Acceptable responses are: (1) a a 1 f , (2) e f , (3) multiplicative identity,

Proposition 1.1.10 Solutions to Linear Equations. Given a ¿eld I 0 1,

1. 1a 1b a b + I " 2!x x + I F a x b

2. 1a 1b a b + IFa / 0 " 2!x x + I F a x b

Proof. (of #1) Suppose that a b + I and a / 0. Since a + I

Hence, x satis¿es a x b. Now, suppose that * + I also satis¿es a * b. Then

1a 1b a b + I " 2!x x + I F a x b .

Proposition 1.1.13 Multiplicative Inverses in a ¿eld I 0 1

4. 1a 1b 1c 1d [a + I F b c d + I

Proof. (of #3) Suppose a b + I and b / 0. Since b / 0, the zero of the

from which we conclude that a b1 is an additive inverse

the expression a b may be written as a b ab may be written

De¿nition 1.2.1 An order, denoted by , on a set S is a relation on S that satis¿es

1x 1y x y + S " x y G x y G y x

1x 1y x y + S " x y " x y F y x ,

be true. The ¿rst statement, 1x 1y x y + S " x y G x y G y x

Example 1.2.3 For S

1x 1y x y + S F x y + n F y x + n " x y

1x 1y x y + S " x y + n G y x + n

De¿nition 1.2.7 An ordered ¿eld I 0 1 is an ordered set that satis¿es

0 y x z y z x z y z x z .

Because and are binary operations on I, x z + I and y z x z + I.

O F1 F O F2 % O F1 F alt O F2 as claimed.

3. 1x [x + I F x 0 " x 0]

From , 0 x x y x i.e.,

and 0 y y. Repeated use of commutativity and as-

***Acceptable responses are: (1) y y y y 0, (2) OF1, (3) 0 y x,

De¿nition 1.2.12 Let I 0 1 n be an ordered ¿eld. A subset S of I is said

2. 1x x + S " x 1 + S.

De¿nition 1.2.13 For I 0 1 n an ordered ¿eld, de¿ne

De¿nition 1.2.14 Let I 0 1 n be an ordered ¿eld. The set of integers of

1m 1n n + QI F m + QI " n m + QI F n m + QI

1m 1n n + ]I F m + ]I " n m + ]I F n m + ]I .

De¿nition 1.2.15 Let I 0 1 n be an ordered ¿eld. De¿ne

B p + T : p 0 F p 2 2 . Now we will expand a bit on the approach used

(b) For p + B, justify that q p and q + B.

2u u + S F 1a a + A " a n u

De¿nition 1.2.18 Let S n be an ordered set. For A t S, u is a least upper

1. u + S F 1a a + A " a n u and

2. 1b [b + S F 1a a + A " a n b " u n b].

1a a + A " a n u and 1b [1a a + A " a n b " u n b] .

2) ) + S F 1a a + A " ) n a

De¿nition 1.2.26 Let S n be a linearly ordered set. A subset A of S is said to

1. 2g g + S F 1a a + A " g n a, and

Then : sup L exists in S, and : inf B.

1 + Q R " 2) ) + Q R F * ) * 1 .

n: : n + Q has a least upper bound in U, say L. Choose > 12 : which is positive

Since n is the largest integer such that n n m:, we know that m: n 1.

Theorem 1.3.5 For U

Now let u 1 x and suppose that t u 0. Fact 1.3.6 yields that