Professional Documents
Culture Documents
(MA4613&4623, MA6213&6223) Introduction To Real Analysis 2e PDF
(MA4613&4623, MA6213&6223) Introduction To Real Analysis 2e PDF
09 Cx) 1r
Ifs 01-2004-2085
'M!FAVF PearsonEducation( * &*L$ SIM)&* Fjjjj#-No 3E *;Ia
22 -Ea WBittffili81100037)
WfIANIA: Y$ 3R
Iii #: I7 194,1:! £C
PREFACE x
TO THE STUDENT xv
vi
Contents vii
5 j Differentiation 165
5.1 The Derivative 166
5.2 The Mean Value Theorem 176
5.3 L'Hospital's Rule 190
5.4 Newton's Method 197
Notes 203
Miscellaneous Exercises 204
Supplemental Reading 205
j
..
q
r Orthogonal Functions and
Fourier Series
9.1 Orthogonal Functions 380
9.2 Completeness and Parseval's Equality 390
379
Bibliography 522
Hints and Solutions to Selected Exercises 523
Notation Index 543
Index 545
Preface
The subject of real analysis is one of the fundamental areas of mathematics, and is the
foundation for the study of many advanced topics, not only in mathematics, but also in
engineering and the physical sciences. A thorough understanding of the concepts of real
analysis has also become increasingly important for the study of advanced topics in
economics and the social sciences. Topics such as Fourier series, measure theory, and
integration are fundamental in mathematics and physics as well as engineering, eco-
nomics, and many other areas.
Due to the increased importance of real analysis in many diverse subject areas, the
typical first semester course on this subject has a varied student enrollment in terms of
both ability and motivation. From my own experience, the audience typically includes
mathematics majors, for whom this course represents the only rigorous treatment of
analysis in their collegiate career, and students who plan to pursue graduate study in
mathematics. In addition, there are mathematics education majors who need a strong
background in analysis in preparation for teaching high school calculus. Occasionally,
the enrollment includes graduate students in economics, engineering, physics, and other
areas, who need a thorough treatment of analysis in preparation for additional graduate
study either in mathematics or their own subject area. In an ideal situation, it would be
desirable to offer separate courses for each of these categories of students. Unfortu-
nately, staffing and enrollment usually make such choices impossible.
In the preparation of the text there were several goals I had in mind. The first was
to write a text suitable for a one-year sequence in real analysis at the junior or senior
level, providing a rigorous and comprehensive treatment of the theoretical concepts of
analysis. The topics chosen for inclusion are based on my experience in teaching grad-
uate courses in mathematics, and reflect what I feel are minimal requirements for suc-
cessful graduate study. I get to the least upper bound property as quickly as possible,
and emphasize this important property in the text. For this reason, the algebraic prop-
erties of the rational and real number systems are treated very informally, and the con-
struction of the real number system from the rational numbers is included only as a mis-
cellaneous exercise. I have attempted to keep the proofs as concise as possible, and to
x
Preface xi
let the subject matter progress in a natural manner. Topics or sections that are not specif-
ically required in subsequent chapters are indicated by a footnote.
My second goal was to make the text understandable to the typical student enrolled
in the course, taking into consideration the variations in abilities, background. and mo-
tivation. For this reason, Chapters 1 through 6 have been written with the intent to be
accessible to the average student, while at the same time challenging the more talented
student through the exercises. The basic topological concepts of open, closed, and com-
pact sets, as well as limits of sequences and functions, are introduced for the real num-
bers only. However, the proofs of many of the theorems, especially those involving
topological concepts, are presented in a manner that permits easy extensions to more
abstract settings. These chapters also include a large number of examples and more rou-
tine and computational exercises. Chapters 7 through 10 assume that the students have
achieved some level of expertise in the subject. In these chapters, function spaces are
introduced and studied in greater detail. The theorems, examples, and exercises require
greater sophistication and mathematical maturity for full understanding. From my own
experiences, these are not unrealistic expectations.
The book contains most of the standard topics one would expect to find in an
introductory text on real analysis-limits of sequences, limits of functions, conti-
nuity, differentiation, integration, series, sequences and series of functions, and
power series. These topics are basic to the study of real analysis and are included
in most texts at this level. In addition, I have included a number of topics that are
not always included in comparable texts. For instance, Chapter 6 contains a section
on the Riemann-Stieltjes integral, and a section on numerical methods. Chapter 7
also includes a section on square summable sequences and a brief introduction to
nonmed linear spaces. Both of these concepts appear again in later chapters of the
text.
In Chapter 8, to prove the Weierstrass approximation theorem, I use the method of
approximate identities. This exposes the student to a very important technique in analy-
sis that is used again in the chapter on Fourier series. The study of Fourier series, and
the representation of functions in terms of series of orthogonal functions, has become
increasingly important in many diverse areas. The inclusion of Fourier series in the text
allows the student to gain some exposure to this important subject, without the neces-
sity of taking a full semester course on partial differential equations. In the final chap-
ter I have also included a detailed treatment of Lebesgue measure and the Lebesgue in-
tegral. The approach to measure theory follows the original method of Lebesgue, using
inner and outer measure. This provides an intuitive and leisurely approach to this very
important topic.
The exercises at the end of each section are intended to reinforce the concepts of
the section and to help the students gain experience in developing their own proofs. Al-
though the text contains some routine and computational problems, many of the exer-
cises are designed to make the students think about the basic concepts of analysis, and
to challenge their creativity and logical thinking. Solutions and hints to selected exer-
cises are included at the end of the text. These. problems are marked by an asterisk (*).
At the end of each chapter I have also included a section of notes on the chapter,
miscellaneous exercises, and a supplemental reading list. The notes in many cases pro-
xii Preface
vide historical comments on the development of the subject, or discuss topics not in-
cluded in the chapter. The miscellaneous exercises are intended to extend the subject
matter of the text or to cover topics that, although important, are not covered in the
chapter itself. The supplemental reading list provides references to topics that relate to
the subject under discussion. Some of the references provide historical information;
others provide alternative solutions of results or interesting related problems. Most of
the articles appear ih the American Mathematical Monthly or Mathematics Magazine.
and should be easily accessible for students' reference.
To cover all the chapters in a one-year sequence is perhaps overly ambitious. How-
ever, from my own experience in teaching the course, with a judicious choice of topics
it is possible to cover most of the text in two semesters. A one-semester course should
at a minimum include all or most of the first five chapters, and part or all of Chapter 6
or Chapter 7. The latter chapter can be taught independently of Chapter 6; the only de-
pendence on Chapter 6 is the integral test, and this can be covered without a theoretical
treatment of Riemann integration. The remaining topics should be more than sufficient
for a full second semester. The only formal prerequisite for reading the text is a stan-
dard three- or four-semester sequence in calculus. Even though an occasional talented
student has completed one semester of this course during their sophomore year. some
mathematical maturity is expected, and the average student might be advised to take the
course during their junior or senior year.
cal logic. For this reason I have included a brief appendix on logic and proofs. The ap-
pendix is not intended to replace a formal course in logic; it is only intended to intro-
duce the rules of logic that students need to know in order to better understand proofs.
These rules are also crucial in helping students develop the ability to write their own
proofs. The various methods of proof are discussed in detail, and examples of each
method are included and analyzed. The appendix also includes a section on the use of
quantifiers. with special emphasis on the proper negation of quantified sentences. The
appendix itself is independent of the text: however, references to it are included
throughout the first several chapters of the text. The appendix can be included as part
of the course, or assigned as independent reading.
LL Acknowledgments
I would like to thank the students at South Carolina who have learned this material from
me, or my colleagues, from preliminary versions of this text. Your criticisms, com-
ments, and suggestions were appreciated. I am also indebted to those colleagues, espe-
cially the late Jeong Yang, who agreed to use the manuscript in their courses.
Special thanks are also due to the reviewers who examined the manuscript for the
first edition and provided constructive criticisms and suggestions for its improvement:
Joel Anderson, Pennsylvania State University; Bogdam Baishanski, Ohio State Univer-
sity; Robert Brown, University of Kansas; Donald Edmondson, University of Texas at
Austin; Kevin Grasse, University of Oklahoma; Harvey Greenwald, California Poly-
technic State University; Adam Helfer, University of Missouri, Columbia: Jan Kucera.
Washington State University; Thomas Reidel, University of Louisville; Joel Robin,
University of Wisconsin, Madison; Stuart Robinson, Cleveland State University; Dan
Shea, University of Wisconsin, Madison; Richard B. Sher, University of North Car-
olina: Thomas Smith, Manhattan College. Your careful reading of the manuscript
helped to turn the preliminary drafts into a polished text.
1 would also like to thank Carolyn Lee-Davis and the staff at Addison Wesley
Longman for their assistance in the preparation of the second edition, and the re-
viewers for this edition for their comments and recommendations: William Barnier,
Sonoma State University; Rene Barrientos, Miami Dade Community College; Denine
Burkett, Lock Haven University; Steve Deckelman, University of Wisconsin; Lyn
Geisler, Randolph-Macon College; Constant Goutziers, State University of New
York; Christopher Heil, Georgia Institute of Technology; William Stout. Salve Regina
University. Special thanks go to my colleague, George McNulty, for his careful read-
ing of the appendix. His constructive criticisms and suggestions were appreciated. I
am also grateful to the readers who informed me of errors in the first edition, and to
the instructors who conveyed to me some of the difficulties encountered while using
the book as a text. Hopefully all of the errors and shortcomings of the first edition
have been corrected.
Finally, I would especially like to thank my wife, Mary Lee, without whose en-
couragement this project might never have been completed.
Manfred Stoll
To the Student
The difference between a course on calculus and a course on real analysis is analogous
to the difference in the approach to the subject prior to the nineteenth century and since
that time. Most of the topics in calculus were developed in the late seventeenth and
eighteenth centuries by such prominent mathematicians as Newton, Leibniz, Bernoulli,
Euler, and many others. Newton and Leibniz developed the differential and integral cal-
culus; their successors extended and applied the theory to many problems in mathe-
matics and the physical sciences. They had phenomenal insight into the problems, and
were extremely proficient and ingenious in deriving complex formulas. What they
lacked, however, were the tools to place the subject on a rigorous mathematical foun-
dation. This did not occur until the nineteenth century with the contributions of Cauchy,
Bolzano, Weierstrass, Cantor, and many others.
In calculus, the emphasis is primarily on developing expertise in computational
techniques and applications. In real analysis, you will be expected to understand the
concepts and to develop the ability to prove results using the definitions and previ-
ous theorems. Understanding the concept of a limit, and proving results about limits,
will be significantly more important than computing limits. To accomplish this, it is
essential that all definitions and statements of theorems be learned precisely. Most of
the proofs of the theorems and solutions of the problems are logical consequences of
the definitions and previous results; some, however, do require ingenuity and cre-
ativity.
The text contains numerous examples and counterexamples to illustrate the partic-
ular topics under discussion. These are included to show why certain hypotheses are re-
quired, and to help develop a more thorough understanding of the subject. It is crucial
that you not only learn what is true, but that you also have sufficient counterexamples
at your disposal. I have included hints and answers to selected exercises at the end of
the text; these are indicated by an asterisk (*). For some of the problems I have provided
complete details; for others I have provided only brief hints, leaving the details to you.
As always, you are encouraged to first attempt the exercises, and to look at the hints or
solutions only after repeated attempts have been unsuccessful.
xc
To the Student xV
At the end of each chapter I have included a supplemental reading list. The journal
articles or books are all related to the topics in the chapter. Some provide historical in-
formation or extensions of the topics to more general settings; others provide alterna-
tive solutions of results in the text. or solutions of interesting related problems. All of
the articles should be accessible in your library. They are included to encourage you to
develop the habit of looking into the mathematical literature.
An excellent source for additional historical information and biographies of fa-
mous mathematicians is the MacTutor History of Mathematics archive at the University
of St. Andrews, Scotland. The URL of their webpage is http://www-history.mcs.st-
andrews.as.uk/
On reading the text you will inevitably encounter topics, formulas, or examples that
may appear too technical and difficult to comprehend. Skip them for the moment; there
will be plenty for you to understand in what follows. Upon later reading the section, you
may be surprised that it is not nearly as difficult as previously imagined. Concepts that
initially appear difficult become clearer once you develop a greater understanding of the
subject. It is important to keep in mind that many of the examples and topics that ap-
pear difficult to you were most likely just as difficult to the mathematicians of the era
in which they first appeared.
The material in the text is self-contained and independent of calculus. I do not use
any results from calculus in the definitions and development of the subject matter. Oc-
casionally, however, in the examples and exercises I do assume knowledge of the ele-
mentary functions and of notation and concepts that should have been encountered else-
where. These concepts will be defined carefully at the appropriate place in the text.
Manfred Stoll
The Real Number System
1.1 Sets and Operations on Sets
1.2 Functions
1.3 Mathematical Induction
1.4 The Least Upper Bound Property
1.5 Consequences of the Least Upper Bound Property
1.6 Binary and Ternary Expansions
1.7 Countable and Uncountable Sets
The key to understanding many of the fundamental concepts of calculus, such as limits,
continuity, and the integral, is the least upper bound property of the real number system
It As we all know, the rational number system contains gaps. For example. there does
not exist a rational number r such that r2 = 2, i.e., f is irrational. The fact that the
rational numbers do contain gaps makes them inadequate for any meaningful discus-
sion of the above concepts.
The standard argument used in proving that the equation r2 = 2 does not have a
solution in the rational numbers goes as follows: Suppose that there exists a rational
number r such that r2 = 2. Write r = m/n where in, it are integers that are not both
even. Thus m' = 2n2. Therefore in2 is even, and hence in itself must be even. But m2,
and hence also 2n2, are both divisible by 4. Therefore n2 is even, and as a consequence
n is also even. This, however, contradicts our assumption that not both in and n are even.
The method of proof used in this example is proof by contradiction; namely, we assume
the negation of the conclusion and arrive at a logical contradiction. A discussion of the
various methods of proof is included in Section A.3 of the Appendix.
The above argument shows that there does not exist a rational number r such that
r2 = 2. This argument was known to Pythagoras (around 500 B.C.), and even the Greek
mathematicians of this era noted that the straight line contains many more points than
the rational numbers. It was not until the nineteenth century, however, when mathe-
maticians became concerned with putting calculus on a firm mathematical footing, that
the development of the real number system was accomplished. The construction of the
real number system is attributed to Richard Dedekind (1831-1916) and Georg Cantor
0
2 Chapter I The Real Number System
Set Operations
There are a number of elementary operations that may be performed on sets. If A and
B are sets, the union of A and B, denoted A U B, is the set of all elements that belong
either to A or to B or to both A and B. Symbolically,
AUB={x:xEA or xEB}.
The intersection of A and B, denoted A fl B, is the set of elements that belong to both
A and B; that is,
Af1B={x:xEA and xEB}.
Two sets A and B are disjoint if A fl B = ¢. The relative complement of A in B, de-
noted B \A, is the set of all elements that are in B but not in A. In set notation,
B\ A= {x : x E B and x q-A}.
If the set A is a subset of some fixed set X. then X \ A is usually referred to as the com-
plement of A and is denoted by A`. These basic set operations are illustrated in
Figure 1.1 with the shaded areas representing A U B, A fl B, and B \ A, respectively.
Figure 1.1
4 Chapter 1 The Real Number System
There are several elementary set theoretic identities that will be encountered
throughout the text. We state some of these in the following theorem; others are given
in the exercises.
We now prove that this is indeed the case. We form subsets B of A by deciding for
each element of A whether to include it in B or to leave it out. Thus for each element of
A there are exactly two possible choices. Since A has n elements, there are exactly 2"
possible decisions, each decision corresponding to a subset of A.
Finally, if A and B are two sets, the Cartesian product of A and B, denoted A X B.
is defined as the set of all ordered pairs' (a. b), where the first component a is from A
and the second component b is from B, i.e.,
A X B= {(a, b) : a E A, b E B}.
For example, if A = (1, 2} and B = {-1, 2, 4}, then
A X B = ((l, -1), (1, 2), (1, 4), (2, -l),(2,2),(2,4)).
The Cartesian product of R with R is usually denoted by R2 and is referred to as the eu-
clidean plane. If A and B are subsets of R, then A X B is a subset of R2. The case where
A and B are intervals is illustrated in Figure 1.2.
a
A
EXERCISES 1.1
1. LetA = {-1,0,1,2},B = {-2,3}, and C = (-2,0,1,5).
a. Find each of the following: (A U B), (B U C), (A n B), (B n C), (A n C). A n (B u C), A \ B. C \ B.
A\ (BUC).
1. A set theoretic definition of ordered pair can be given as follows: (a, b) - ((a), {a, b}}. With this
definition two ordered pairs (a, b) and (c, d) are equal if and only if a = c and b = d (Miscellaneous
Exercise 1).
6 Chapter 1 The Real Number System
1.21 Functions
We begin this section with the fundamental concept of a function. In many texts, a func-
tion or a mapping f from a set A to a set B is described as a rule that assigns to each
element x E A a unique element y E B. This is generally expressed by writing y = f(x)
to denote the value of the function f at x. The difficulty with this "definition" is that the
terms "rule" and "assigns" are vague and difficult to define. Consequently we will de-
fine "function" strictly in terms of sets, using the notation and concepts introduced in
the preceding section.
The motivation for the following definition is to think of the graph of a function;
namely the set of ordered pairs (x, v) where .v is given by the "rule" that defines the func-
tion.
1.2 Functions 7
1.2.1 DEFINITION Let A and B be any two sets. A function f from A into B is a subset of
A X B with the property that each x E A is the first component of precisely one ordered
pair (x, y) E f ; that is, for every x E A there exists Y E B such that (x, y) E f, and if
(x, y) and (x, y') are elements of f, then y = y'. The set A is called the domain off, de-
noted Dom f The range of f, denoted Range f, is defined by
Range f= (yEB:(x,y)Ef forsomnexEA).
If Range f = B, then the function f is said to be onto B. (See Figure 1.3.)
AxB
1- - - -f- - (x.y) I
If f is a function from A to B and (x, y) E f, then the element y is called the value
of the function f at x and we write
y = f (x) or f : x -+ y.
We also use the notation f : A -± B to indicate that f is a function from (or on) A into
(or to) B, or that f maps A to B or is a mapping of A to B. If we think of a function
f : A -> B as mapping an element x E A to an element y = f (x) in B, then this is often
represented by a diagram as in Figure 1.4. If f : A -* Q8, then f is said to be a real-
valued function on A.
A function f from A to B is not just any subset of A X B. The key phrase in Defini-
tion 1.2.1 is that "each x E A is the first component of precisely one ordered pair
(x, y) E f." To better understand the notion of function we consider several examples.
8 Chapter I The Real Number System
1.2.2 EXAMPLES
(a) Let A = 1-3, -2, -1, 0, 11 and B = Z. Consider the subset f of A X B given by
f = {(-3. 2). (-2. -2). (-1.4). (0. -6), (1.4)}.
Since each x E A belongs to precisely one ordered pair (x. y) E f. f is a function from
A into B with Range f = (-6. -2. 2.4). Figure 1.5 indicates what f does to each el-
ement of A. Even though the element 4 in B is the second component of the two dis-
tinct ordered pairs (- 1, 4) and (1, 4). this does not contradict the definition of function.
-3 2
-2 --2
0'--6
-1 4
1 4
Figure 1.5
2. Since x' z 0 for all x E R. the range of It is a subset of (v E R : ? 2). To obtain equality. we re-
y
quire that for every e > 2 there exists an .r E R such that x' + 2 = v. The existence of such a y will fol-
low as a consequence of Example 1.4.8.
1.2 Functions 9
and
f-'(H)=(xEA:f(x)EH).
If H = {y}, we will write f -'(y) instead of f ''({y}). Thus for Y E B.
1.2.4 EXAMPLES
(a) As in Example 1.2.2, let A = {-3, -2,-1.0, 1},B = 71, and f : A -* 2 the func-
tion given by
f = ((-3,2),(-2. -2),(-l,4).(0. -6),(1.4)1.
Consider the subset E _ {-1, 0, 1) of A. Then
f(E) {f(-1), f(0). f(l)} {-6,4}.
If H = (0, 1, 2, 3, 4}, then
f-'(H)= {xEA: f(x)EH}= {-3. -1,1}.
Since both f(- 1) and f (l) are equal to 4, f '(4) -1, 1). On the other hand, since
(x, 0) 0 f for any x E A, f -'(0) _ 46.
(b) Consider the function g : 7 -+ Z given by g (x) = x 2, and let E -1, -2,
-3,...). Then
g(E) = {(-n)2 : n E NI} _ {1,4,9,. ..}.
On the other hand,
g-'(g(E)) = 71\ {0}.
For this example, E C# g-'(g(E)).
(c) Let h be the function defined by h(x) = 2x + 3, Dom h = R. If E _ {x E R :
-1 < x s 2}, then
h(E) = {2x+3:-1 5 xs2} _ {yER: l y:5 7).
For the set E we also have
1.2.5 THEOREM Let f be a function from A into B. If A, and A2 are subsets of A. then
(a) f(Ai U A2) = f(A,) Uf(A2)
(b) f(A, n A2) C f(A,) nf(A2)
1.2 Functions 11
Proof. To prove (a), let y be an element of f(A, U A,). Then Y = f(x) for some x in
A, U A2. Thus x E A, or x E A,. Suppose x E A,. Then y = f(x) E f(A,). Similarly, if
x E A2, Y E f (A2). Therefore y E f (A,) U f (A2). Thus
f(A, U A2) C f(A,) U f(A,).
Since it is clear that f(A,) and f(A2) are subsets of f(A, U A,), the reverse inclusion
also holds, thereby proving equality.
Since f(A, n A2) is a subset of both f(Al) and f(A2), the relation stated in (b) is
also true.
To. see that equality need not hold in (b), consider the function g(x) = x2
Dom g = 7L, of Example 1.2.4(b). If A, = (0, -1, -2, -3, ...), and A2 = (0. I, 2,
3, . . . } , then f(Al) = f(A2) _ {0, 1, 4, 9, . . .}, but A, n A, = {0}. Thus
f(A, fl A2) = f({0}) = {0} # f(A,) n f(A2) _ (0, 1, 4, 9, ...).
Inverse Function
1.2.7 DEFINITION A function f from A into B is said to be one-to-one if whenever
x, * x2, then f(x,) * f(x2).
Alternatively, a function f is one-to-one if whenever (x,, y) and (x2, y) are elements
off then x, = x2. From the definition it follows that f is one-to-one if and only if
f `(y) consists of at most one element of A for every y E B. If f is onto B. then
f -'(y) # rb for every y E B. Thus if f is one-to-one and onto B, then f `(y) consists
of exactly one element x E A and
g={(y,x)EBXA: f(x)=y}
defines a function from B to A. This leads to the following definition.
12 Chapter 1 The Real Number System
1.2.9 EXAMPLES
(a) Let h be the function of Example 1.2.4(c); that is, h(x) = 2x + 3, Dom h = R. The
function h is clearly one-to-one and onto R with
x=h-'(v)=2(y-3). Domh-' - R.
(b) Consider the function f defined by the equation y = 2. If we take for the domain
of f all of R, then f is not a one-to-one function. However, if we let
Domf=A={xER:xa0},
then f becomes a one-to-one mapping of A into A. To see that f is one-to-one, let
x,, x2 E A with x, # x2. Suppose x, < x2. Then x1 < x2; that is, f(x1) # f(x2). Tbero-
fore f is one-to-one. To show that f is onto A. we need to show that for each y E A.
y > 0, there exists a positive real number x such that
x2=y.
1.2 Functions 13
Intuitively we know that such an x exists; namely, the square root of y. However.
a rigorous proof of the existence of such an x will require the least upper bound
property of the real numbers. In Example 1.4.8 we will prove that for each y > 0
there exists a unique positive real number x such that X' = y. The number x is called
the square root of y and is denoted by \. Thus the inverse function of f is
given by
f-'(y) _ Domf ' _ {y E R:y :0}.
Composition of Functions
Suppose f is a function from A to B and g is a function from B to C. If a E A. then f(a)
is an element of B, the domain of g. Consequently we can apply the function g to f(a)
to obtain the element g(f(a)) in C. This process, illustrated in Figure 1.7. gives a new
function h which maps a E A to g(f(a)) in C.
h(a) =g(f(a)) C
1.2.11 EXAMPLES
(a) If f (x) = v -+x with Dom f = {x E R : x z - I } and g(x) = x', Dom g = R.
then
Even though the equation (g of)(x) = 1 + x is defined for all real numbers x, the do-
main of the composite function g o f is still only the set {x E R : x,2: -1 }. For this ex-
ample, since Range g C Dom f, we can also find f o g; namely,
(fof-')(Y) =f(f+I(Y)) = ( = Y.
R EXERCISES 1.2
1. Let A = {-1, 0, 1, 2) and B = N. Which of the following subsets of A X B is a function from A into B? Explain
your answer.
a. f= {(-l.2).(0.3).(2.5)} b. g = {(-1,2),(0,7).(1,3),(2,7)}
c. h = -1)} d. k = {(x, y) : y = 2r + 3, x E A}
2. *a. Let A = {(x, y) E R X R : x2 + y2 = 11. Is A a function? Explain your answer.
b. Let B = {(x, y) E A : y z 0}. Is B a function? Explain your answer.
3. Let f : N -* N be the function defined by f(n) = 2n - 1. Find f(E) and f -'(E) for each of the following sub-
sets E of N.
*a. {1,2,3,4} b. {1,3,5.7} c. N
4 . Let f= {(x,y):xE R,y =x3 + 1}.
*a. Let A = {x : -1 <- x 5 2}. Find f(A) and f '(A).
b. Show that f is a one-to-one function of R onto R.
*c. Find the inverse function f - 1.
S. Let f, g mapping Z into Z be given by f(x) = x + 3 and g(x) = 2x. Find (f o g)(N) and (g of)(N).
6. For each of the following real-valued functions, find the range of the function f and determine whether f is one-
to-one. If f is one-to-one, find the inverse function and specify the domain of f-'.
*a. f(x)3x-2, Domf=R b. f(x)=5x+4. Domf=R
*c. 1(x) = 1 1
s Dom f= {x E R : O s x < l} d. Ax) = I X X Dom f= {x E R : 0 < x < 1)
7. Let A = {t E R : 0:5 t < 21r} and B = R2. and let f : A -+ B be defined by J(t) = (co. r. sin t).
*a. What is the range of f?
*b. Find f ' of each of the following points: (1, 0), (0, -1), ;-, ), (0, 1).
c. Is the function f one-to-one?
8. Prove parts (b) and (c) of Theorem 1.2.6.
9. Letf:A -+ BandletFCA.
a. Prove that f(A) \f(F) C f(A \ F). b. Give an example for which f(A) \f(F) * f(A \ F).
10. Let f be a one-to-one function from A into B. Show that' o(f f)(x) = x for all x E A and that (f of
for ally E Range f.
11. *Let f : A -* B and g : B -> A be functions satisfying (g o f)(x) = x for all x E A. Show that f is a one-to-one
function. Must f be onto B?
12. If f : A -s B and g : B -+ C are one-to-one functions, show that (g of) f -' o g' on Range (g of).
true, then Q is true. In a proof by contradiction one assumes that P is true and Q is false.
and then shows that these two assumptions lead to a logical contradiction; namely, show
that some statement R is both true and false. Further details on the method of proof by
contradiction are provided in Section A.3 of the Appendix.
Proof of Theorem 1.3.1. Assume that the hypotheses of Theorem 1.3.1 are true,
but that the conclusion is false; that is, there exists a positive integer n such that the
statement P(n) is false. Let
A = {k E N : P(k) is false}.
By our assumption, the set A is nonempty. Thus by the well-ordering principle A has a
smallest element k,,. Since P(1) is true, k > 1. Also, since ko is the smallest element of
A. P(k,, - 1) is true. But then by hypothesis (b), P(k,,) is also true. Thus k, a A. This,
however, is a contradiction. Consequently, P(n) must be true for all n E N. Q
1.3.3 EXAMPLES We now provide two examples to illustrate the method of proof by math-
ematical induction. The first example provides a proof of the identity in the introduc-
tion to the section. An alternative method of proof will be requested in the exercises
(Exercise 7).
(a) To use mathematical induction, we let our statement P(n), n E N, be as follows:
r+ +r"= r - r"+i . r # 1.
I -r
When n = I we have
r(1 -r) _ r-r2
r = (1 - r) I - r , provided r * 1.
Thus the identity is valid for n = 1. Assume P(k) is true for k i' I; i.e.,
r - rA+I
r+ +r I-r r # 1.
We must now show that the statement P(k + 1) is true; that is.
r+ r- r(A+ p+ i
r # 1.
I -r '
But
r+ + rk+i = r + + rk + rk+1,
I-r I - r
r - rk+'
1-r
1.3 Mathematical induction 17
Thus the identity is valid for k + 1, and hence by the principle of mathematical induc-
tion for all n E N.
(b) For our second example, we use mathematical induction to prove Bernoulli's in-
equality. If h > -1, then
(I +h)">1 + nh for alln EN.
When n = 1, (1 + h)' = I + h. Thus since equality holds, the inequality is certainly
valid. Assume that the inequality is true when n = k, k -- 1. Then for is = k + 1,
(I + h)'-I = (I + h)'"(1 + h),
which by the induction hypothesis and the fact that (1 + h) > 0
?(1 +kh)(1 +h)= I +(k+ 1)h+kh=
1 + (k + 1)h.
Therefore the inequality holds for n = k + 1, and thus by the principle of mathemati-
cal induction for all n E N.
. Although the statement of Theorem 1.3.1 starts with n = 1, the result is still true if
we start with any integer n,, E Z. The modified principle of mathematical induction is
as follows: If for each n E 77, n ? n,,, P(n) is a statement about the integer n satisfying
(a) is true, and
(b') P(k + 1) is true whenever P(k) is true. k ? no,
then P(n) is true for all n E 77, n ? n,.
The proof of this follows from Theorem 1.3.1 by simply setting
Q(n)=P(n+n,- I), n E N,
which is now a statement about the positive integer n.
Remark. In the principle of mathematical induction, the hypothesis that P(I) be true
is essential. For example, consider the statement P(n):
n+i=n, nEN.
This is clearly false! However, if we assume that P(k) is true, then we also obtain that
P(k + 1) is true. Thus it is absolutely essential that P(n,) be true for at least one fixed
value of n,.
Proof. Exercise 3. Q
Mathematical induction is also used in the recursive definition of functions defined
for the positive integers. In this procedure, we give an initial value of the function fat
n = 1. then assuming that f has been defined for all integers k = 1. .... n, the value
off at n + I is given in terms of the values off at k, k 5 n. This is illustrated by the
following examples.
1.3.5 EXAMPLES
(a) Suppose f : N -' N is defined by f(1) = t and f(n + 1) = nf(n). n E N. The
values off for n = 1, 2, 3.4 are given as follows:
3.2.1.
f(1) = 1, f(2) = lf(1) = 1, f(3) = 2f(2) = 2.1, f(4) = 3f(3) =
Thus we conjecture that f(n) (n - 1)!, where 0. is defined to be equal to one, and
for n E N, n! (read n factorial) is defined as
The conjecture is certainly true when n = 1. Thus assume that it is true for n = k,
k 1; that is, f(k) = (k - 1)!. Then for n = k + 1,
f(k + 1) = kf(k),
which by the induction hypothesis
k(k - 1)! = k!
Therefore the identity holds for it = (k + 1), and thus by the principle of mathematical
induction, for all it E N.
(b) For our second example, consider the function f : N - R defined by f (l) = 0,
f(2) = 3, and for n ? 2 by f (n) = (1, +i) f (n - 2). Computing the values of f for
n = 3, 4, 5, and 6, we have
EXERCISES 1.3
1. Use mathematical induction to prove that each of the following identities are valid for all n E N.
n(n + 1)
a. 1+2+3+ +n= 2 *b. 1 +3+5+ +(2n- 1)=n2
n(n + 1)(2n + 1)
. 12+22+ +n2= d. 13+23+ +n3=[Zn(n+l)]2
6
e. 2+22+23+ +2"=2(2"- 1)
*f For x, y E Il,
x"+1 -y",i =(x-y)(x"+x"-iy+...+y")
a. f(1)=2,andforn> 1,f(n)=(n-1)f(n-1)-n+1.
*b. f(l)=1,1(2)=4, andforn>2,f(n)=2f(n-1)-f(n-2)+2.
(n
c. f (l) = 1, and for n > 1, f (n) = 3n 1) f (n - 1).
r+r2+ +r"= r - r 1
, r# nEN
without using mathematical induction.
20 Chapter) The Real Number System
8. Use mathematical induction to prove the arithmetic geometric mean inequality. If a,, a,.. . . . a,,, n E N. are
nonnegative real numbers, then
a, + a2 + ' - - +a.
ala, , .
-a. S (
with equality if and only if a, = a, = = a".
8. The operation of multiplication is distributive over addition; that is, for all
a, b, c E F,
The element 0 is called the zero of F and the element I is called the unit of F. For
a # 0, the element a-' is customarily written as or 1/a. Similarly, we write a - b in-
stead of a + (-b), ab instead of a b, and alb or instead of a b-.
The real numbers R contain a subset P known as the positive real numbers satis-
fying the following:
1.4 The Least Upper Bound Property 21
The concepts bounded below and lower bound are defined similarly. A set E is
bounded if E is bounded both above and below. We now consider several examples to
illustrate these concepts.
1A.2 EXAMPLES
(a) Let A = {0, 12, 3, s, ...} = { l - n = 1, 2, 3, ...} (see Figure l.8). Clearly A
is bounded below by any real number r 0 and above by any real numbers ? 1.
22 Chapter 1 The Real Number System
o 1 2 .1 a
2 a 4s
Figure 1.8
(b) N = {1, 2, 3, ...}. This set is bounded below; e.g., I is a lower bound. Our intu-
ition tells us that N is not bounded above. It is obvious that there is no positive integer
n such that j n for all j E N. However, what is not so obvious is that there is no real
number .6 such that j s /3 for all j E N. In fact, given /3 E R, the proof of the exis-
tence of a positive integer n > /3 will require the least upper bound property of R (The-
orem 1.5.1).
(c) B = {r E 0 : r > 0 and r2 < 2}. Again it is clear that 0 is a lower bound for B.
and that B is bounded above; e.g., 2 is an upper bound for B. What is not so obvious,
however, is that B has no maximum. By the maximum or largest element of B we mean
an element a E B such that p a for all p E B. Suppose p E B. Define the rational
number q by
_ 2-p2 2p+2
q-p+(p+2) p+2
With q as defined, a simple computation gives
z_
q2-=
2 (p+2)2
Since p2 < 2, q > p and q2 < 2. Thus B has no largest element. Similarly, the set
has no minimum or smallest element. Intuitively, the largest element of B would satisfy
p2 = 2. However, as was shown in the introduction, there is no rational number p for
which p2 = 2.
There is one important fact about the supremum of a set that will be used repeat-
edly throughout the text. Due to its importance we state it as a theorem.
1.4.5 EXAMPLES In the following examples, let's consider again the three sets of the pre-
vious examples.
(a) As in Example 1.4.2(a), let A = {0, Z, 3, ;, ...}. Since 0 is a lower bound of A and
0 E A, infA = 0. We now prove that sup A = 1. Since I - I'-1 < I for all n = 1,
2, ... , 1 is an upper bound. To show that I = sup A we need to show that if 13 E R
with $ < 1, then f3 is not an upper bound of A. Clearly if 0 0, then $ is not an up-
per bound of A. Suppose, as in Figure 1.9, 0 < 13 < 1. Then our intuition tells us that
there exists an integer no such that
1
n1
element, there exists q E B such that q > a. This contradicts the fact that a is an up-
per bound of B. Similarly, if a2 > 2, then there exists a q < a such that q2 > 2. But
then q is an upper bound of R, which is a contradiction of property (ii) of Definition
1.4.3. The least upper bound of B in N is V2 (Section 1.5, Exercise 9), which we know
is not rational.
1.4.8 EXAMPLE In this example we show that for every positive real number y > 0, there
exists a unique positive real number a such that a2 = y; i.e., a = v. The uniqueness
of a was established in Example 1.2.9(b).
We only prove the result for y > 1, leaving the case 0 < y s I to the exercises (Ex-
ercise 6). Let
C={xEH:x>0 and x2<y}.
With y = 2, this set is similar to the set B of Example 1.4.2(c), except that here we con-
sider all positive real numbers x for which x2 < Y. Since y > 1, 1 E C and thus C is
nonempty. Also since y > 1, y2 > y, and thus y is an upper bound of C. Hence by the
least upper bound property, C has a supremum in P. Let a = sup C. We now prove that
a2 = y. To accomplish this we show that the assumptions a2 < Y and a2 > v lead to
contradictions. Thus a2 = y.
1.4 The Least Upper Bound Property 25
Then
y(! - I)(a2 )
P'- - y = (2)
(a + V)2
If a2 < y, then by (1) $3 > a, and by (2) $32 < v. This contradicts that a is an upper
bound for C. On the other hand, if Cr' > y, then by (1),6 < a and by (2), j32 > Y. Thus
if x E R with x ? $3, then x2 > y. Therefore $3 is an upper bound of C. This contradicts
that a is the least upper bound of C. Since S defined by (I) may not be rational, the
same proof will not work for the set B of Example 1.4.2(c). However, using Theorem
1.5.2 of the following section, it is possible to also prove that sup B = 12.
For convenience, we extend the definition of supremum and infimum of a subset E
of N to include the case where E is not necessarily bounded above or below.
Intervals
Using the order properties of R, we can define certain subsets of R known as intervals.
with analogous definitions for (-cc, b) and (-oo, b). The intervals (a. 00), (-oo, b)
and (-oo, oo) = R are also referred to as open intervals, whereas the intervals [a, oo)
and (-oo, b) are called closed intervals. (See Figure 1.10.)
(a. b)
a b
[a, b]
a b
(a, b)
In this definition, when b = a, (a, a) = 0 and [a, a] = {a}. Although the empty
set 0 and the singleton {a} do not fit our intuitive definition of an interval, we will in-
clude them as the degenerate case of open and closed intervals, respectively. It should
be noted that the intervals of the form (a, b), (a, b], [a, b), and [a, b] with a, b E R.
a s b, are all bounded subsets of R.
An alternative way of defining intervals without use of the adjectives open and
closed is as follows.
EXERCISES 1.4
1. Use the axioms for addition and multiplication to prove the following: If a E R, then
a. 0. b. -a.
c. -(-a) = a.
2. Let a, b E It Prove the following.
a. If a # 0, then I/a # 0. b. If ab - 0, then either a = 0 or b = 0.
3. Let a, b, c E R. Prove the following.
a. If a > b, then a + c > b + c. b. If a > b and c < 0, then ac < bc.
c. If a * 0, then a2 > 0. d. If a > 0 then I/a > 0, and if a < 0 then 1/a < 0.
4. *If a, b E IR, prove that ab ](a2 + b2).
1.4 The Least Upper Bound Property 27
*a.A1,ll1,...}=
248 1l 1J ll
nEll}
'):E N b.Bsinn 22 :nEN}
ll J
"e.E{2nn:nEt`11 f. F {(-lr-!:nGN}
1(
"g. G{(1 nEN) h. H +(-lY')":nEN}
i.l={xER:xz<4} j.J={x2:-2<x<2}
6. If 0 < y < 1, prove that there exists a unique positive real number x such that x2 = y.
7. Prove that there exists a positive real number x such that x3 = 2.
8. Let A be a nonempty subset of R. If a = sup A is finite, show that for each e > 0, there is an a E A such that
a - e < a < a.
9. Let E be a nonempty subset of R that is bounded above, and set U = jig E R is an upper bound of E}. Prove
that sup E = inf U.
10. Let A be a nonempty subset of R and let -A = (-x: x E A). Prove that infA = -sup(-A).
11. Use the least upper bound property of R to prove that every nonempty subset of R that is bounded below has an
infimum.
12. If A and B are nonempty subsets of R with A C B, prove that
inf B s infA s sup A <_ sup B.
13. Suppose that A and B are bounded subsets of R. Prove that A U B is bounded and that
sup(A U B) = sup{sup A, sup B}.
14. For A. B, subsets of R, define
A+B=(a+b:a6A,bEB) and
A = (-1, 2, 4, 7) and B = {-2, -1, 1), find A + B and A B.
'b. If A and B are nonempty and bounded above, prove that
sup(A + B) = sup A + sup B.
c. If A and B are nonempty subsets of the positive real numbers that are bounded above, prove that
sup(A B) = (sup A)(sup B).
d. Give an example of two nonempty bounded sets A and B for which sup(A B) # (sup A)(sup B).
15. Let f, g be real-valued functions defined on a nonempty set X satisfying Range f and Range g are bounded sub-
sets of R. Prove each of the following.
*a. sup{f(x) + g(x) . X E X) s sup{f(x) : x e X} + sup{g(x) : x E X}
b. inf{f(x) : x E X} + inf{g(x) : x E X} s inf{f(x) + g(x) : x r= X}
c. If f(x) : g(x) for all x E X. then sup{f(x) : x E X} s sup{f(x) : x e X}.
16. Let X = Y = [0, 1 ] and let f : X X Y -+ R be defined by f (x, y) = 3x + 2y.
a. For each x E X, find F(x) = sup{f(x, y) y E Y}; then find sup{F(x) : x E X}.
b. For each y E Y, find G(y) = sup{ f(x, y) : x E X}; then find sup(G(y) : y E Y).
c. Find sup{f(x, y) : (x, y) E X X Y}. Compare your answer with your answers in parts (a) and (b).
28 Chapter 1 The Real Number System
17. Perform the computations of Exercise 16 with X = [-1, 11. Y = [0, 2], and f(x, v) = 3x - 2y.
18. Let X, Y be nonempty sets, and let f be a nonnegative real-valued function defined on X X Y. For each x E X and
y E Y, define
F(x) = sup{f(x, y) : y E Y}, G(y) = sup{f(x. v) : x E X}.
Prove that
sup{F(x) : x E X} = sup{G(v) : y E Y} = sup{f(x, y): (x, v) E X X Y}.
19. Let X, Y be nonempty sets and let f : X X Y -+ R be a function with bounded range. Let
F(x) = sup{f(x, y) : y E Y} and H(y) = inf{f(x, y) : x E X}. Prove that
sup{H(y) : y E Y} inf{F(x) : x E X}.
20. Let X = Y = [0, I ]. Perform the computations of Exercise 19 for each of the following functions f(x, y).
1, x=y,
a. f(x, v) = 3x + 2r b. f(x, y) =
x*v
21. Let J be a subset of R that has the following property: if x, y E J with x < y, then t E J for every t satisfying
x < t < Y. Prove that J is an interval as defined in Definition 1.4.10.
1.5.1 THEOREM (Archimedian Property) If x, y E Ll and x > 0, then there exists a posi-
tive integer n such that
nx > Y.
Proof. If y <- 0, then the result is true for all n. Thus assume that y > 0. We will
again use the method of proof by contradiction. Let
A={nx:nEN}.
If the result is false, that is, there does not exist an n E N such that nx > y. then nx y
for all n E N. Thus y is an upper bound for A. Thus since A * 46, A has a least upper
bound in R. Let a = sup A. Since x > 0, a - x < a. Therefore a - x is not an upper
bound and thus there exists an element of A, say mx, such that
a - x<mx.
But then a < (m + 1)x, which contradicts the fact that a is an upper bound of A.
Therefore, there exists a positive integer n such that nx > y.
Remark. One way in which the previous result is often used is as follows: given
e > 0, there exists a positive integer n such that noe > 1. As a consequence,
1
- < E
n
x< n <y.
If x < 0 and y > 0, then the result is obvious. Finally, if x < y < 0, then by the pre-
ceding calculations there exists r E 0 such that -y < r < -x; i.e., x < -r < y.
The conclusion of Theorem 1.5.2 is often expressed by the statement that the ra-
tional numbers are dense in the real numbers; that is, between any two real numbers
there exists a rational number. A precise definition of "dense" is given in Definition
3.1.12.
Another consequence of the least upper bound property is the following theorem
concerning the existence of nth roots.
1.5.3 THEOREM For every real number x > 0 and every positive integer n, there exists a
unique positive real number y so that y" = x.
The number y is written as lcx or x'1" and is called the nth root of x. The unique-
ness of y is obvious. Since the existence of y will be an immediate consequence of the
intermediate value theorem (Theorem 4.2.11), we omit the details of the proof in the
text. A sketch of the proof of Theorem 1.5.3, using the least upper bound property, is
included in the miscellaneous exercises. It should be emphasized that the proof of The-
orem 4.2.11 also depends on the least upper bound property.
1.5.4 COROLLARY If a, b are positive real numbers, and n is a positive integer, then
(ab)Un = al/nb'l".
0 EXERCISES 1.5
1. *If r and s are positive rational numbers, prove directly (without using the supremum property) that there exists
an n E N such that nr > s.
2. Given any x E R, prove that there exists a unique n E Z such that n - 1 5 x < n.
3. If r * 0 is a rational number and x is an irrational number, prove that r + x and rt are irrational.
4. Prove directly (without using Theorem 1.5.1) that between any two rational numbers there exists a rational
number.
5. If x. y E R with x < y, show that x < ty + (I - t)x < y for all t, 0 < t < 1.
6. *a. Prove that between any two rational numbers there exists an irrational number.
*b. Prove that between any two real numbers there exists an irrational number.
7. If x > 0, show that there exists n E N such that 1/2" < x.
8. Let x, y E R with x < y. If u E P with u > 0. show that there exists a rational number r such that x < ru < Y.
9. Let B = {r E 0 : r > 0 and r2 < 2} and a = sup B. Prove that a2 = 2.
-
given by
0 2
-+1
10 107+70+70-
1
3. This section can be omined on first reading. The ternary expansion of a real number is only required in
Section 3.3 and can be covered at that point.
1.6 Binary and Ternary Expansions 31
Having chosen n1, ... , nk, let nk+ 1 E {0, 1, 2} be the largest integer such that
n,
+n2+. . .+nk+nk+1 <X.
3 32 3k 3k+'
The binary expansion of x, or the expansion of x to any other base, is defined simi-
larly. For the binary expansion, the integer nk at each step is chosen as the largest inte-
ger in {0, 1).
1.6.2 EXAMPLES
(a) We now use Definition 1.6.1 to obtain the ternary expansion of 3. As the first step.
we must choose n, as the largest integer in {0, 1, 2} such that
3 < 3'
This inequality fails for n, = 1, 2. Thus n, = 0. To find n2, we choose the largest inte-
ger n2 E {0, 1, 2) such that
3+32<3'
which is satisfied by n2 = 2. To find n3 we must have
2 1 3.
0+ + n3 <
3 32 33
To see that this indeed is the case, we use the fact that for the geometric series.
rk=(11
k=o
r), O<r<1.
Thus
.0222...=
k=, 3k
- _-
2 2
9 kap 3
I k 2
9 1
1
i
1
In (c) we will illustrate how mathematical induction may also be used to prove such a
result. The above ternary expansion is not unique. The number ;also has the finite ex-
pansion
I = .1000.
(base 3).
3
This expansion can be obtained using the definition and induction (Exercise 3). Alter-
natively, using the geometric series we have
X
r_I 2
1 I" I I I
. 4
-U 4 = 3.
(c) The ternary expansion of 1 is given by
We now show in detail how this expansion is derived. We will use the second principle
of mathematical induction to prove that nk = I for all k E K. Since < ; and
i > n, = 1. Thus, the result is true for k = 1. Let k > I and assume that n; = I for
3k-1
+nk<1
3` 2
(3)
r - r"+i
r+r2+ +r"=
1-r'
r # 1, (Example 1.3.3(a))
with n = k - 1, we obtain
3+
+ 3kli = ' 2
` 1 - 3k1_ 1
1.6 Binary and Ternary Expansion 33
1.6.3 THEOREM Let x E U8 with 0 < x <- 1. and with {n} as defined in Definition
1.6.1, let
(n1 nk
3 3'
Then sup E = x.
Proof. Let a = sup E. Since x is an upper bound, a s x. Suppose a < r. Let k be
the smallest positive integer such that
1 <.r - a or a+ .<x.
3k 3k
Since a = sup E,
But
n1
+...+3k+;k=
I
3 +... i
irk n1 nj.._ 1
3k i+ nk + < x.
3 3k
If any n;, 1 s j <_ k - l is 0 or 1, we have a contradiction to the choice of n;. If all the
ni = 2, then we obtain
2
3+3<xsl, 1
ber in a given base is not always unique; when x has a finite expansion, it has an infi-
nite expansion as well. For the ternary expansion, when
x= + +3m, a,"E{1,2}.
or an infinite expansion of the form
3
a, 0 2
x= +- +3m+ k when a,"=1,
3 k=m+i 3*
or
x=3+ + + k, when am = 2.
I k=m +1 3
© EXERCISES 1.6
1. Find the ternary expansion of each of the following.
*a. i b. i3
2. Find the real number determined by each of the following finite or infinite ternary expansions. In the case of an
infinite expansion, use the geometric series as in Example 1.6.2 (c) to determine your answer.
*a. .0022 b. .00222 c..1010
*d..101010 e..001001001 f. .121212
3. *Find the binary expansion of 3. Use induction to prove the result.
4. Find both the finite and infinite binary expansions of 6.
5. If 0 < x s 1, prove that the infinite ternary expansion of x is unique.
6. If x = a/2m. with a E N odd, and 0 < a < 2m, show that x has a finite binary expansion of the form
al ni 0
x=
2
+ + a'",
2'"
a.m = 1, and an infinite expansion x=
2
+ +
2m
+7
k.m+i
2k.
1
this by counting the number of elements in each of the two sets. Alternatively, we can
reach the same conclusion without counting, by simply pairing up the elements:
7+2
3+7
19 + 11
32 + 21
For infinite sets, the concept of two sets being of the same size or having the same
number of elements is vague. For example, let S denote the squares of the positive in-
tegers; namely,
S = {12, 22, 32, . . J.
Then on one hand, S is a proper subset of the positive integers N, yet as Galileo
(1564-1642) observed, the sets N and S can be placed into a one-to-one correspondence
as follows:
1+12
2+22
3+32
This example caused Galileo, and many subsequent mathematicians, to conclude that
the standard notion of "size of a set" did not apply to infinite sets. Cantor, on the other
hand, realized that the concept of one-to-one correspondence raised many interesting
questions about the theory of infinite sets. In this section we take a closer look at infi-
nite sets and what it means for an infinite set to be countable. We begin by defining the
concept of equivalence of sets.
1.7.1 DEFINITION Two sets A and B are said to be equivalent (or to have the same car-
dinaiity), denoted A -- B, if there exists a one-to-one function of A onto B.
The notion of equivalence of sets satisfies the following:
(I) A -A. (reflexive)
(ii) If A B, then B --A. (symmetric)
(iii) If A - B and B -- C, then A ^- C. (transitive)
1.7.2 DEFINITION For each positive integer n, let N. = {1, 2, ... , n}. As in Section 1.1,
N denotes the set of all positive integers. If A is a set, we say:
(a) A is finite if A N. for some n, or if A = t¢.
(b) A is infinite if A is not finite.
(c) A is countable if A N.
36 Chapter 1 The Real Number System
1.7.3 EXAMPLES
112 ,
22,32, J. Then the function g(n) = n2 is a one-to-one
(a) As before, let S =
mapping of N onto S. Thus S - N, and S is countable.
(b) For our second example, we show that 7L -- N. To see this, consider the function
f: N -+ Z defined by
n
(n even),
.f(n) (n 2
2 1), (n odd).
Figure 1.11 illustrates what the mapping f does to the first few integers. It is left as an
exercise (Exercise 1) to show that this function is a one-to-one mapping of N onto Z.
Thus 7L -- N, and the set Z is also countable.
Figure 1.11
Sequences
1.7.5 DEFINITION If A is a set, by a sequence in A we nean a function f from N into A.
For each n E N, let x = f(n). Then x is called the nth term of the sequence f.
For notational convenience, sequences are denoted by or just rather
than the function f Note, however, the distinction between which denotes the
sequence, and {x : n = 1, 2, . . .}, which denotes the range of the sequence. For ex-
ample, {1 - (-1)"} denotes the sequence f where
f(n) = X. = 1 - (-1)".
On the other hand, {x : n = 1, 2, ...} = {0, 2}.
By definition, if A is a countable set, then there exists a one-to-one function f from
N onto A. Thus
A = Rangef = {x,,:n = 1,2,...}.
The sequence f is called an enumeration of the set A; i.e., A = {x : n = 1, 2, .. .}
with x * x,R whenever n * m. This ability to enumerate the elements of a countable
set plays a key role in the proof of the following theorem.
1.7.8 DEFINITION Let A and X be nonempty sets. An indexed family of subsets of X with
index set A is a function from A into 9(X).
If f : A -+ 13(A), then for each a E A, we let Ea = f(a). As for sequences, we
denote this function by {Ea}aE,%. If A = N, then {En}nEN is called a sequence of sub-
sets of X. In this instance, we adopt the more conventional notation {En} 1 to denote
{En}nEN
1.7.9 EXAMPLES The following are all examples of indexed families of sets.
(a) The sequence {N.} 1, where N. = {1, 2.... , n}, is a sequence of subsets of N.
(b) For each n E N, set 1n = {x (E R : 0 < x < Then {1n} 1 is a sequence of sub-
sets of N.
(c) For each x, 0 < x < 1, let
Ex= {rE0:0t=- r<x}.
Then {E;},,E(0.1)is an indexed family of subsets of 0. In this example, the open interval
(0, 1) is our index set.
1.7.10 DEFINITION Suppose {Ea}aEA is an indexed family of subsets of X. The union of the
family of sets (Ea}aEA is defined to be
with an analogous definition for the intersection. Occasionally, when the index set A is
fixed in the discussion, we will use the shorthand notation U. E. or !1a E. rather than
aU E. or f A Ea.
1.7.11 EXAMPLES We now consider the union and intersection of the families of sets given
in the previous example.
1.7 Countable and Uncountable Sets 39
Since I is the only element that is in N. for all n, fl,, Ni,, = {1}. For the union, since
N. C Ni for all n, U. Nn C Ni. On the other hand, if n E N, then n E NI,,, and as a
consequence, N C U,, N,,, which proves equality.
(b) As in the previous example, for n E NI, let In = {x E R:0 < x < .1}. We first
show that 00
n in = (Q.
Suppose the conclusion is false. Then there exists x E H such that x E In for all n; i.e.,
R=1
={xE08:0<x< 1}.
(c) We leave it as an exercise (Exercise 9) to show that if E, is defined as in Example
1.7.9(c); i.e., E, = {r E O : 0 5 r < x}, then
As for a finite number of sets, we also have analogues of the distributive laws and
De Morgan's laws for arbitrary unions and intersections.
1.7.12 THEOREM (Distributive Laws) If Ea, a E A. and E are subsets of a set X, then
(a)
aE
UA EQ = aEA
fl Ea,
c
(b) *GA
n Ea = aEA
u E:-
The proofs of both of these theorems are left to the exercises. The following theo-
rem is an analogue of Theorems 1.25 and 1.2.6.
40 Chapter 1 The Real Number System
P B,)
(U = U f -'(B.).
Proof. (a) To illustrate the method of proof, we will prove that f (n. E,) C fl, f (E,).
The proof that f (U, E,) = U f(E,) is left to the exercises.
Suppose y E f(fl, E,.). Then by definition of the image of a set, Y = f (.Y) for some
x E fl, E,. But then x E E, for all a E A, and thus f (x) E f (E.) for all a E A. There-
fore y E fl, f(E,).
(b) Again we will only prove one of the two identities. The other is left to the exercises.
In this instance we will show that f -' (U, B,) = U. f Let B = U, B,. Since
B. C B for all a e A, we have f `(B,) C f `(B) for all a e A. Thus U. f "' (B,)
C f -' ( U,, B.). In the other direction, suppose x E f -'(U,, B,). Then by the definition
of the inverse image of a set, f (x) E U. B,. Thus f (x) E B,,, for some index a,, E A.
But then x E f -' (B,), and hence x E U, f '(B.). This proves equality. U
The Countability of 0
The countability of the set of rational numbers Q will. follow as a corollary of the fol-
lowing theorem.
Em= {iE}.
Ua_, Eby
Theorem
Then E. is countable, and since 0 = 1.7.15 the set 0 is count-
able.
The UncountabUity of 08
In November 1873, in a letter to Dedekind, Cantor asked whether the set R itself was
countable. A month later he answered his own question by proving that 01 was not count-
able. We now prove, using Cantor's elegant "diagonal" argument, that the closed interval
[0, 11 is uncountable,. and thus 08 itself is uncountable. For the proof we will use the fact
that as in Section 1.6, every x E [0, 11 has a decimal expansion of the form x = n ,n,
with n; E {0, 1, 2, . .. , 9}. As for the binary and ternary expansions, the decimal expan-
sion is not necessarily unique. Certain numbers such as ,'-o have two expansions; namely.
This, however, will not be crucial in the proof of the following theorem.
1.7.18 THEOREM If A is the set of all sequences whose elements are 0 or 1, then A is
uncountable.
Remark. The set A is the set of all functions f from N into {0, I }. Thus a sequence
f E A if and only if f(n) = O or I for all n.
Proof. As in the previous theorem, we will also prove that every countable subset of
A is a proper subset of A, and thus A cannot be countable.
Let E be a countable subset of A and let {s : n = 1 , 2, .. .} be an enumeration of
the set E. For each n, s. is a sequence of 0's and l's. We construct a new sequence s as
follows: for each k E N, let
s(k) = I - sk(k).
Thus if sk(k) = 0, s(k) = 1, and if sk(k) = 1, s(k) = 0. Thus s e A. Since for all k E N
s(k) # sk(k), we haves* s for any n E N. Therefore s a E; i.e., Ec A. 0
In the previous two theorems we proved that the closed interval [0, 1 ] and the set
of all sequences of 0's and I's are both uncountable sets. One may ask, are these two
sets equivalent? Considering the fact that every real number x E [0,1 ] has a binary ex-
pansion, which is really a sequence of 0's and I's, one would expect that the answer is
yes. This indeed is the case (Miscellaneous Exercise 7).
EXERCISES 1.7
1. a. Prove that the function f of Example 1.7.3(b) is a one-to-one function of N onto Z.
b. Find a one-to-one function of Z onto N.
2. *Let 0 denote the set of odd positive integers. Prove that 0 -- N.
3. Prove that the function f of Theorem 1.7.4 is a one-to-one function of N X N onto N.
4. *a. If a. b e R with a < b, prove that (a, b) - (0, 1).
b. Prove that (0, 1) -- (0, oo).
5. Suppose X. Y, Z are sets. If X - Y and Y- Z, prove that X Z.
6. *a. If A ^- X and B Y, prove that (A X B) (X X Y).
b. If A and B are countable sets, prove that A X B is countable.
7. If X - Y, prove that 9(X) - 9(Y).
& Find U , A. and fl°, A. for each of the following sequences of sets
nEN b.A.={xCE R:-1,<x<1}. nEN
nEN x:5 l-,-,), nEN
e.A ={xER: x51nEN,n; 2 nEN
9. For each x E (0, 1), let Ex = {r E Q : 0 s r < x}. Prove that
n E,={0}
xE(0,I)
and U E,={r E Q:0sr< 1).
10. Prove Theorem 1.7.12.
11. Prove Theorem 1.7.13.
Notes 43
NOTES
The most important concept of this chapter is the least up- pend on previous theorems that require the least upper
per bound property of the real numbers. This property bound property in their proofs.
will be fundamental in the development of the underlying The emphasis on the least upper bound property is
theory of calculus. In the present chapter we have already not meant to overshadow the importance of the concepts
seen its application in proving the Archimedian property of countable and uncountable sets. The fact that the ratio-
(Theorem 1.5.1). For the rational number system this nal numbers are countable, and thus can be enumerated,
property can be proved directly. For the real number sys- will be used on several occasions in the construction of
tem it was originally assumed as an axiom by Archimedes examples. In all the examples and exercises. every infinite
(287-212 B.c.). Cantor, however, proved that the Archi- subset of R turns out to be either countable or equivalent
median property was no axiom, but a proposition deriv- to [0, 1 ]. Cantor also made this observation, and it led
able from the least upper bound property. him to ask whether this result was true for every infinite
In subsequent chapters the least upper bound prop- subset of R. Cantor was never able to answer this ques-
erty will occur in proofs of theorems, either directly or in- tion; nor has anyone else. The assertion that every infinite
directly, with regular frequency. It will play a crucial role subset of R is either countable or equivalent to [0, 11 is
in the characterization of the compact subsets of R and in known as the continuum hypothesis. In 1938 Kurt Godel
the study of sequences of real numbers. It will also be re- proved that the continuum hypothesis is consistent with
quired in the proof of the intermediate value theorem for the standard axioms of set theory; that is. Godel showed
continuous functions. One of the corollaries of this theo- that the continuum hypothesis cannot be disproved on the
rem is Theorem 1.5.3 on the existence of nth roots of pos- basis of the standard axioms of set theory. On the other
itive real numbers. Many other results in the text will de- hand, in 1963 Paul Cohen showed that the continuum hy-
44 Chapter 1 The Real Number System
pothesis is undecidable on the basis of the current axioms is equivalent to some subset of B. We also write A < 8
of set theory. provided A s B and A # B. C)early, N < [0, 1 ]. Also.
Cantor's creation of the theory of infinite sets was Exercise 22 of the previous section proves that for any set
motivated to a great extent by problems arising in the
A, A < Using equivalence of sets, it follows that if
study of convergence of Fourier series. We will discuss
some of these problems in greater detail in Chapter 9. The A s B and B C, then A s C. This naturally leads to
notion of one-to-one. correspondence allowed Cantor to the following question: If A and 8 are sets such that
define the "power" or what is now called the "cardinality" A c B and B < A. is A = B? The answer is yes, and the
of a set. For a given set A, the cardinal number or power result is known as the Schroder-Bernstein theorem.
of A is denoted by A. Two sets A and B are said to have Cantor did more than just introduce cardinal num-
bers; he also defined addition, multiplication, and expo-
the same cardinality, denoted A = B, provided that A is
nentiation of cardinal numbers. Cantor's original work
equivalent to B. Some commonly encountered sets have
on cardinal numbers and the theory of infinite sets can
special symbols attached to their cardinal number. For ex-
be found in his monograph listed in the Supplemental
ample 0 = 0, N, = n, N = K. and [0. 1) = c. Readings. A more modern treatment. including a proof of
An ordering can be defined on cardinal numbers as the Schrdder-Bemstein theorem, is available in the text by
follows: If A and B are sets, then A B provided that A Halmos.
MISCELLANEOUS EXERCISES
1. Let A and B be nonempty sets. For a E A, b E B. define the ordered pair (a. b) by
(a, b) _ {{a}, {a, b}}..
Prove that two ordered pairs (a, b) and (c, d) are equal if and only if a = c and b = d.
The following-two exercises are detailed and lengthy. The first is a sketch of the proof of Theorem 1.5.3. The second
shows how the least upper bound property may be used to define the exponential function b', b > 1.
2. LetE= {t E R:t> 0 and t' <x).
a. Show that E # ¢ by showing that x/(x + 1) E E.
b. Show that I. + x is an upper bound of E.
Let y = sup E. The remaining parts of the exercise are to show that y' = x. This will be accomplished by show-
ing that y" < x and y" > x lead to contradictions, leaving y" = x. To accomplish this, the following inequality
will prove useful. Suppose 0 < a < b, then
b"-a"=(b-a)(b"-'+ab" +... '+ a`)<n(b-a)b"''.
c. Show that the assumption y" < x contradicts y = sup E as follows: Choose 0 < h < I such that
x-y
n(v + I)
Use (*) to show that. y + h E E.
d. Show that the assumption y" > x also leads to 'a contradiction of the definition of y as follows: Set
k=yy"_x
Show that if tat y-k,then t44;E.
3. Fix b > 1.
a. Suppose m, n, p, q are integers with n > 0 and q > 0. If m/n = p/q, prove that
(hl" =
-Thus if r is rational, b' is well defined. '
Miscellaneous Exercises 45
The following two exercises provide a detailed development of the field of complex numbers.
4. Definition. A complex number is an ordered pair (a. b) of real numbers. If ;, = (a, b) and w _ (c, d), we write z = w
if and only if a = c and b = d. For complex number z and w we define addition and multiplication as follows:
z+w=(a+ c, b + d)
zw (or - bd, ad + bc).
The set of ordered pairs (a, b) of real numbers with the above operations of addition and multiplication is denoted
by C.
a. Find elements 0 and 1 in C such that 0 + z = z and lz = z for all z E C.
b. Show that if z = (a, b), then -z - (-a,-b) is the additive inverse of z.
c. For z E C with z * 0, find the multiplicative inverse z-'.
d. Prove that the set of complex numbers C with addition and multiplication as defined is a field.
e. Set i = (0, 1). Show that i2 = -1.
f. Show that every complex number z can be written as z = a + bi where a. b E R. The real numbers a and b are
called the real part and the imaginary part of z, respectively. We write a = Re(z) and b = Im(z).
g. Prove that C is not an ordered field.
5. Definition. If z = a + bi E C. then the complex number i = a - bi is called the conjugate of z. The absolute
value of z, denoted IzI, is defined by IzI = a2 +
a. Prove each of the following.
(1) i z+
(ii) zw = z N
(iii) z + i = 2 Re(z), z - z = 2i Im(z)
(iv) zz = 1z12
b. Prove each of the following.
(1) IZI = IzI
(ii) Izwl = IzIIwI
(iii) IRe(z)I IzI, IIm(z)I s IzI
(iv) Iz + w12 = Izl2 + Iwl2 + 2 Re(zw)
(v) Iz - w12 = 1z12 + Iwl2 - 2 Re(zw)
(vi) Iz + wl IZI + Iw'I
The following result, known as the Schroder-Bernstein theorem, is nontrivial, but very- important. It is included as an
exercise to motivate further thought and additional studies. A proof of the result can be found in the text b) Halmos
listed in the Supplemental Reading.
6. Let X and Y be infinite sets. If X is equivalent to a subset of Y. and Y is equivalent to a subset of X. prove that X is
equivalent to Y.
7. As in Theorem 1.7.18, let A denote the set of all sequences of 0's and I's. Use the previous result to prove that
A-[0,1].
46 Chapter i The Real Number System
SUPPLEMENTAL READING
Buck, R. C., "Mathematical induction and recursive defi- GOdel, Kurt, "What is Cantor's continuum problem?"
nition:' Amer. Math. Monthly 70 (1963),128-135. Amer. Math. Monthly 54 (1947), 515-525.
Burrill, Claude W., Foundations of Real Numbers. Halmos, Paul. Naive Set Theory, Springer-Verlag. New
McGraw-Hill, Inc., New York, 1967. York. Heidelberg. Berlin, 1974.
Cantor, Georg, Contributions to the Founding of the The- Richman, F. "Is 0.999... = 0- Math. Mag. 72
ory of Transfrnite Numbers (translated by Philip E. (1999), 396-400.
B. Jourdain), Open Court Publ. Co., Chicago and Shrader-Frechette. M., "Complementary rational num-
London, 1915. bers," Math. Mag. 51(1978), 90-98.
Dauben, Joseph W., Georg Cantor: His Mathematics Spooner, George and Mentzer, Richard, Introduction to
and Philosophy of the Infinite, Princeton University Number Systems, Prentice-Hall, Inc., Englewood
Press, Princeton, N.J., 1979. Cliffs, N.J., 1968.
Sequences of Real Numbers
2.1 Convergent Sequences
2.2 Limit Theorems
2.3 Monotone Sequences
2.4 Subsequences and the Bolzano-Weierstrass Theorem
2.5 Limit Superior and Inferior of a Sequence
2.6 Cauchy Sequences
2.7 Series of Real Numbers
In our study of sequences of real numbers we encounter our first serious introduction to
the limit process. The notion of convergence of a sequence dates back to the early nine-
teenth century and the work of Bolzano (1817) and Cauchy (1821). Some of the con-
cepts and results included in this chapter have undoubtedly been encountered previ-
ously in the study of calculus. Our presentation, however, will be considerably more
rigorous-emphasizing proofs rather than computations.
We begin the chapter by introducing the notion of convergence of a sequence of real
numbers and by proving the standard limit theorems for sequences normally encountered
in calculus. In Section 2.3 we will use the least upper bound property of 1q to prove that
every bounded monotone sequence of real numbers converges in R. The study of subse-
quences and subsequential limits will be the topic of Section 2.4. In this section we also
prove the well-known result of Bolzano and Weierstrass that every bounded sequence of
real numbers has a convergent subsequence. This result will then be used to provide a short
proof of the fact that every Cauchy sequence of real numbers converges. Although the
study of series of real numbers is the main topic of Chapter 7, some knowledge of series
will be required in the construction of certain examples in Chapters 4 and 6. For this rea-
son, we include a brief introduction to series as the last section of this chapter.
Even though our emphasis in this chapter is on sequences of real numbers, in sub-
sequent chapters we will also encounter sequences of functions and convergence of se-
quences in normed linear spaces. A good understanding of sequences of real numbers
will prove very helpful in providing insight into properties of sequences in more gen-
eral settings.
47
48 Chaprer2 Sequences of Real Numbers
2.1.1 DEFINITION For a real number x. the absolute value of x, denoted ix . is de-
fined by
V. ifx > 0,
1x1 = -.r, ifx - 0.
For example, 141 = 4 and 1-5I = 5. From the definition, Ixl ? 0 for all x E R.
and 1x1 = 0 if and only if x = 0. This last statement follows from the fact that if x * 0.
then -x * 0 and thus 1x1 > 0. The following theorem, the proof of which is left to the
exercises (Exercise 1), summarizes several well-known properties of absolute value.
2.1.2 THEOREM
(a) I-xI = Ix1 for all x E R.
(b) lxvI = I.xI IyI for all x. Y E R.
(c) IxI = \ for all x E R.
(d) If r > 0. then lxl < r if and only if -r < x < r.
(e) - Ixl s x < Ixl for all x E R.
The following inequality is very important and will be used frequently throughout
the text.
Proof. We provide the proof of (a), leaving the proof of (b) as an exercise (Exercise
2). If x, y, z E 08, then by the triangle inequality,
2.1.5 EXAMPLE Determine the set of all real numbers x that satisfy the inequality
12x + 41 < 8. By Theorem 2.1.2(d), 12x + 41 < 8 if and only if -8 < 2x + 4 < 8.
or equivalently, -12 < 2x < 4. Thus the given inequality is satisfied by a real number
x if and only if -6 < x < 2.
Geometrically, 1x1 represents the distance from x to the origin 0. More generally,
for x, y E 1i8, the euclidean distance d(x, y) between x and y is defined by
d(x,y)= Ix - yl
For example, d(-1, 3) = I-1 - (3)1 = 1-41 = 4, and d(5, -2) = 15 - (-2)1 = 7.
The distance .d, may be regarded as a function on R X R which satisfies the following
properties: d(x, y) _,- 0, d(x, y) = 0 if and only if x = y, d(x, y) = d(y, x), and
d(x, y) s d(x, z) + d(z, y)
for all x, y, z E R. This last inequality, also referred to as the triangle inequality, fol-
lows from Corollary 2.1.4(a).
Neighborhood of a Point
The study of the convergence of a sequence or the limit of a function requires the no-
tion of one real number being "close to" another. Since the euclidean distance between
two points a and x is given by d(a, x) = j a - xl, saying that x is "close to" a is equiv-
alent to saying that the distance lx - aI between them is "small" A convenient method
for expressing this idea is with the concept of an e-neighborhood of a point. This con-
cept will prove useful not only in the study of the limit of a sequence but also in our
study of the limit of a function and the structure of point sets in R.
p-E p p+E
Convergence of a Sequence
Recall from Definition 1.7.5 that a sequence in R is a function f :N -i R. For each
n E N, p = f(n) is called the nth term of the sequence f, and for convenience, the se-
quence f is denoted by I or simply
2.1.8 EXAMPLES
(a) For our first example we show that the sequence t converges to 0 in R. The
proof of this is the remark following Theorem 1.5.1; namely, given e > 0, there exists
a positive integer n, such that n,e > 1. Thus for all n _> no,
H_o <e.
Therefore 1im,, = 0. In this example, the integer n, must be chosen so that
n "o
n, > lie.
(b) If p E R, the sequence defined by p = p for all n E N is called the constant
sequence p. Since Jp - pl = 0 for all n E 101, we have lim p = p.
lim 3n+2=3.
1. The negation of the definition of convergence of a sequence is included in Section A.4 of the Appendix.
2.1 Convergent Sequences 51
Since
2n+ l 2 l 1
n+1+Vn-
1
2Vn-
Given e > 0, we want to choose no E NI such that 1/(2\) < e for all n y n,. This
is easily verified to be the case if n, E N is chosen so that n, a 1/4E2. With this choice
of n, we now have I n + 1 - VnI < e for all it at n,.
2.1.10 THEOREM
(a) If a sequence in ir'8 converges, then its limit is unique.
(b) Every convergent sequence in R is bounded.
Proof. (a) To prove (a) we use the method of proof by contradiction. Thus we as-
sume that the sequence converges but that its limit is not unique. So suppose the se-
52 Chapter2 Sequences of Real Numbers
quence {p"} converges to two distinct points p,q E R. Let e = 11p - q1. Since
p" - * p, there exists an integer n, such that I p" - p I < e for all n ? n1. Also, since
p" - q, there exists an integer n2 such that I p. - qI < e for all is ? n2. Thus if n
max{ni, n2}, by the triangle inequality (Corollary 2.1.4(a)),
2
Ip - qI s Ip" - pi + Ipn - qI < 2e=-glp-qI
The inequality Ip - qI <'Ip - qI however gives a contradiction. Thus the limit must
be unique.
(b) Let {p"} be a convergent sequence in 11 that converges top E R. Take e = 1.
For this e, there exists an integer na such that I p" - pI < 1 for all n > no. Thus by
Corollary 2.1.4(b), I p" I < I p I + 1 for all n > no. Let
M = max{Ip,l. .Ip".I. Ipl + 1}.
Then Ip"I s M for all n E N. Therefore the sequence is bounded. O
2.1.11 EXAMPLES
(a) According to the previous theorem, every convergent sequence is bounded. The
converse, however, is false. The sequence (I - (-1)"}"= I is bounded, but by Exam-
ple 2.1.8(d), the sequence does not converge. The sequence is bounded since Ip"I =
II -(-1)"1:5 2forallnEN.
(b) The sequence {n(- I r} is not bounded in R, and thus cannot converge.
EXERCISES 2.1
1. Prove Theorem 2.1.2.
2. *Prove Corollary 2.1.4(b).
3. Prove that for xi...._x. E R, Ix, + + X.1 Sri I + + Ix" I
4. If x, y E R, show that (x + yl = Ixl + lyl if and only if xy ? 0.
5. If a, b e R, prove that labs s 21(a2 + b2).
6. Determine all x E R that satisfy each of the following inequalities.
"a.13x-21<-I1 b. Ix2-4I<5 'c.lxl+Ix-11<3 d. Ix-li<Ix+l1
7. Determine and sketch the set of ordered pairs (x, y) in R X R that satisfies the following.
a. 4x1=II b.IxI:IyI C.Ixyl!5 2
8. For each of the following sequences, prove, using an e, no argument that the sequence converges to the given limit
p; that is, given e > 0, determine n such that I p" - pI < e for all n ;' no.
13n ++5 _ 3 b2n+5 _ 1 n2+ I
'a. P 2\
6n-3 p 3 c 2n2
2n 7 ' 2 'P
j1-(- rJ,p=1
/1
( 1
M'p=0 f. In 1 +R 1 J},p=2.
l n
2.2 Limit Theorems 53
2.2.1 THEOREM If {a.} and are convergent sequences of real numbers with
lima.=a
n- oo
and limb,=b,
n- 00
then
lira (a, + b,) = a + b, and
(a) n_00
(b) ltm
.-000
a,b, = ab.
L. = b.
(c) Furthermore, if a * 0, and a. * O for all n, then lint
n--an a
Proof. The proof of (a) is left to the exercises (Exercise 1). To prove (b), we add and
subtract the term a,b to obtain
la,b. - abl = I(a,b, - a,b) + (a,b - ab)l
c la,llb. - bl + Iblla. - al.
54 Chapter2 Sequences of Real Numbers
Since {an} converges, by Theorem 2.1.10(b), {an} is bounded. Thus there exists a con-
stant M > 0 such that Ia.I <- M for all n. Therefore,
lanbn - abl s Mlbn - bl + IbI Ian - at.
Let e > 0 be given. Since an -+a, there exists a positive integer n, such that
E
In - al <
n 2(lbl + 1)
for all n a n1. Also, since bn -+ b, there exists a positive integer n_ such that
Ibn-bt<2mf
for all n at n2. Thus if n a max{n1, n2},
Ianl z Ilai
Ian - al < E 2-
a_aI<E
I
2.2 Limit Theorems 55
lim
I
n-+oo a, - a'
--
2.2.2 COROLLARY If {a,,} is a convergent sequence of real numbers with lim an = a, then
for any c E R, n"°`
(a) lim (an + c) = a + c, and
(b) Inn
n-ow
can = ca.
Proof. If we define the sequence {c,} by c, = c for all n E N, then the conclusions
follow by (a) and (b) of the previous theorem.
2.2.3 THEOREM Let {a,,} and {b,,} be sequences of real numbers. If {b,,} is bounded and
lima,, = 0, then
lim anbn = 0.
n-.oo
Proof. Exercise 3.
Remark. Since the sequence {b,} may not converge, Theorem 2.2.1(c) does not ap-
ply. The fact that the sequence {b,,} is bounded is crucial. For example, consider the se-
quences {,1,} and {3n}.
2.2.4 THEOREM Suppose {a,,}, {b,,}, and {c,,} are sequences of real numbers for which
there exists n, E N such that
an<bn5c, for all nEN,n?n
and that lint an = lim c. = L. Then the sequence {b,} converges and
n-roo n-+oo
limb,,=L.
n--.W
Proof. Exercise 4.
This result, commonly called the "squeeze theorem," is very useful in applications.
Quite often to show that a given sequence {a,,} in R converges to a, we will first prove
that
la, - al S Mb,
for some positive constant M and a nonnegative sequence {b,,} with lim b, = 0. (See,
for example, the proof of Theorem 2.2.1(c).) At this point we can uselheorem 2.2.4
to conclude that the sequence {(a,, - a1} converges to zero, or equivalently, that
lim an = a.
now
56 Chapter 2 Sequences of Real Numbers
This can also be proved directly. SinceM-OGlim b" = 0, given e > 0. there exists
n,, E IN such that 0 S b" < e/(M + 1) for all /n ? no. Thus for n ? n,,,
Ia"-aI5Mb"<M`M+ 1)<E.
Therefore lim a" = a.
k k!(W:: k)!
is the binomial coefficient.
As in Section 1.3, for n E N, n! (read n factorial) is defined by
2-1,
with the usual convention that 0! = 1. Since the binomial theorem is ancillary to our
main topic of discussion, we leave the proof, using mathematical induction, to the ex-
ercises (Exercise 13). An alternative proof using Taylor ser ies will be provided in Sec-
tion 8.7.
2.2.6 THEOREM
Proof. The proofs of (a) and (b) are left to the exercises (Exercise 5). The proof
of (a) is straightforward and the proof of (b) (for p > 1) is similar to the proof
2.2 Limit Theorems 57
of (c). For the proof of (c), let x" _ '"In-n - 1. Since x" is positive, by the binomial
theorem
.(nl n(n2- 1)
n = (1 + x")" /f xn =
Thus by (a) and Theorem 2.2.4, lim x" = 0, from which the result follows.
"
(d) Let k be a positive integer so that k > a. Since p > 1, write p = (I + q), with
q > 0. By the binomial theorem, for n > 2k,
kt 2kk!'
and as a consequence,
n°
05 c MnF"
k-"
The result now follows by part (a) and Theorem 2.2.4.
(e) Write p as p = ±1/q, where q > 1. Then,
IP"I = IPA" =
4"
which by (d) (with a = 0) converges to 0 as n -- oo.
(f) Fix k E N such that k > dpi. For n > k,
IPr < klk-t) IPI "
n! - n! (k - 1)! ( k }
Since ip(/k < 1, the result follows by (e).
2.2.7 , EXAMPLES We now provide several examples to illustrate the previous theorems.
f 2n + 1
(a) As in Example 2.1.8(c) consider the sequence' } We write
in + 2
2n+1_n(2+'-) 2+-',
3n + 2 3+
58 Chapter 2 Sequences of Real Numbers
1
Since lim = lam 2 = 0, by Corollary 2.2.2(a),
n-"o n n-.oo n
(2+n?)-_1
3+n
oo\2+nJ3+?3
/
)=2-1=2.
1 1
On
0.
"-Cc 2 n+
n 3l
(c) For our next example we consider the sequence +n} As in (a), we first want
3'
to factor out the dominant power in both the numerator and the denominator. By
Theorem 2.2.6(d), lim n°/p" = 0 for any a E R and p > 1. This simply states that
p" (p > 1) grows taster than any power of n. Therefore the dominant terms in the nu-
merator and denominator are 2" and 3", respectively. Thus
"' 1+
11+F
3
1.
n-. 2n +n2=0.
3"
(d) As our final example we consider the sequence {n((1 + n)'2 - 1)}. Before we can
evaluate the limit of this sequence we must first simplify the nth term of the sequence.
This is accomplished as follows:
Now we can factor out an n2 from both the numerator and the denominator. This gives
-2 -
X, (1 + i)2.
"
Using the limit theorems we now conclude that lim x" = -2.
EXERCISES 2.2
1. Prove Theorem 2.2.1(a).
L{ n2+n-n}.,
{\ 11+" - i)E
g. +a - h. {(2" + 3)""r..,
7. For each of the following sequences, determine whether the given sequence converges or diverges. If the sequence
converges, find its limit; if it diverges, explain why.
I + (-1)"1°0 1
-sin -}
-
(( n "-1 n 2
9. Let {x"} be a sequence in R with x" - 0. and x" * 0 for all n. Prove that lim x" sin = 0.
z
60 Chapter2 Sequences of Real Numbers
lim
.-. a" =L.
*a. If L < 1, prove that the sequence {a"} converges and that lim a = 0.
RtiOG
b. If L > 1, prove that the sequence {a"} is unbounded.
c. Give an example of a convergent sequence {a"} of positive real numbers for which L = 1.
d. Give an example of a divergent sequence {a"} of positive real numbers for which L = 1.
11. Use the previous exercise to determine convergence or divergence of each of the following sequences.
e. {-}.aER.p> I
(n
k l)*b.
Use mathematical induction to prove the binomial theorem (Theorem 2.2.5).
14. Let {ak}k i be a sequence in R. For each n E N, define
a,+ +a"
s" n
*a. Prove that if li-m ak = a, then lim s" = a.
b. Give an example of a sequence {ak} which diverges, but for which {s") converges.
As a general rule, bounded sequences need not converge; e.g., {I - (-1)"}. For
monotone sequences, however, we have the following convergence result.
Let e > 0 be given. Since a - E is not an upper bound of E, there exists a positive in-
teger n,, such that
a - E <an 5a.
This is illustrated in Figure 2.2. Since {an} is monotone increasing,
a - e < an<- a forallnatn0.
Thus an E NE (a) for all n ? no and therefore lim a" = a.
a-t a
Figure 2.2
a, a2 an a,.«, b,,., b b2 b,
Figure 2.3
nI ,,*o.
62 Chapter2 Sequences of Real Numbers
Proof. Suppose In = [an, bn), a", b" E R, an 5 bn. Since 1n 3 /".," for all m ? 0,
an!5 an+m:s bnt,,,<_b,,,
for all n, m E N. Thus the sequence {a"} is monotone increasing and bounded above
by every b" m E N. Thus by the previous theorem, a = lim an exists with a s b", for
all m E N. Therefore a E In, for all m E N and thus "~'°
00
a E nm-l/,,,,
which proves the result. Q
Remark. Similarly, if b = lim bthen b E In for all n, and thus
[a, b] C 1 1n.
2.3A EXAMPLES
(a) Our first example shows that the conclusion of Corollary 2.3.3 is false if the inter-
vals 1" are not closed. As in Example 1.7.11(b), for each n E N set 1" = (0, ). Then
1,, 3 1n+, for all n, but
00
n 1" = o.
The conclusion of Corollary 2.3.3 may also be false if the intervals 1" are unbounded
(Exercise 2).
(b) Consider the sequence {j?'}. , with 0 < p < 1. Even though Theorem 2.2.6(e)
applies, we use the results of this section to prove thatn-WO
lim p" = 0. For n E N set
s" = p". Since p > 0, s" > 0 for all n E N. Thus {sn} is bounded below. Also,
"+'
The last inequality follows since s,, > 0 and 0 < p < 1. Thus the sequence {s"} is
monotone decreasing, bounded below, and hence by Theorem 2.3.2, is convergent. Let
s = lim
n-O oW
s". Then
s= lim
n-"O
sn+, = n-+oo
limp"+' =p lims"=ps.
Therefore s = ps or s(1 - p) = 0. But p # I and thus we must have s = 0.
(c) Let a, = 1 and for n ? I set an+i = 6(2a,, + 5). The first three terms of the se-
quence {a"} are a, = 1, a2 = 6, and a3 = y. Thus we suspect that the sequence {a"} is
monotone increasing and bounded above by 2. Since a, < 2, if we assume that a" <_ 2,
then
Thus by mathematical induction a, s 2 for all n E N. Likewise, since a, < a2, if our
induction hypothesis is an < then the result is true for n = 1. and
Therefore the sequence {an} is monotone increasing, bounded above, and thus is con-
vergent. Let a = lim an. Then
nr00
a lima..,,
R-KO
6 (tan + 5) = (2a + 5).
aim 6
Solving the equation a = '-6(2a + 5) for a gives a = ;.
(d) Let a, = 1, and for n ? 1, set To investigate the convergence of the
sequence {an}. we will establish by induction that
1 a.<a,, 1<2,
for all n r= N. When n = 1, we have
1=a,<N/2 =a2<2.
Thus the statement is true for n = 1. Assume that it is true for n = k. Then
I S ak+t = V Lak < V 2ak+, = ak+2,
and
a = lim
n-"0
an+, = lim
w_00
= \1521
Euler's Number e
2.3.5 EXAMPLE In this example we consider in detail the very important sequence
{tn}= 1, where for each n E N,
tn='1+nJn.
We will show that the sequence {t,} is monotone increasing and bounded above, and thus
has a limit. The standard notation for this limit is e (in honor of Leonhard Euler); i.e.,
11n
e= lim I+
n- n
64 Chapter 2 Sequences of Real Numbers
t"=C1+n)"
I n(n - 1) 1 n(n - 1) 1 1
+n n2+
+ (1)
n+ 1.2 1.2 n n"
1.2...k nt'
which is equal to
1)...(I
1 2 1 . .k (1 n+ 1)( n+ - n + 1),
which is greater than the corresponding term in (2). Thus t < r,, for all n. From (1)
we also obtain
4 + + 1+
1.2 1.2.3
+- 1 + 1
2 1+1+2+22+
+r---
which by the identity I + r + + r'-' _ (1 - r")/(1 - r), r # 1,
=1+ I <I+ .. !- . = 3 ..
1 -g i -g
Thus {t"} is bounded above by 3, and we can apply Theorem 2.3.2. Since t" s 3 for all
n e N we also have that e s 3. To five decimal places, e = 2.71828 . The num-
ber e is the base of the natural logarithm function which will be defined in Example
6.3.5 as a definite integral.
Infinite Limits
If a monotone increasing sequence {a"} is bounded above, then by Theorem 2.3.2 the
.sequence converges. If the sequence {a'} is not bounded above, then for each posi-
tive real number M there exists n E N such that a" ? M for all n ? n,,. Since the
real number M can be taken to be arbitrarily large, this is usually expressed by say-
ing that the sequence {a"} diverges to oo. We make this concept precise, not only for
monotone sequences, but for any sequence of real numbers, with the following def-
inition.
23 Monotone Sequences 65
2.3.6 DEFINITION Let {a,j be a sequence of real numbers. We say that {an} approaches
infinity, or that {an} diverges to oc, denoted a,, -+ oo, if for every positive real number
M, there exists an integer n,, E N such that
a > M for all n ? n,,.
We will also use the notation lim an = oo to denote that a - oc as n -+ oc. The
n"X
concept of a,, - -oo is defined similarly.
2.3.7 THEOREM If {a"} is monotone increasing and not bounded above. then a,, -> oc as
n--)I oo.
As a consequence of Theorems 2.3.2 and 2.3.7, every monotone increasing se-
quence {an} either converges to a real number (if the sequence is bounded above) or di-
verges to oo. In either case,
lim a,, = sup{an : n E N}.
2.3.8 EXAMPLE Consider the sequence {n(2 + (- I )")}. If n is even, then n(2 + (- I Y')
= 3n; if n is odd, then n(2 + (-1)") = n. In either case,
n(2+(-1)")ten,
and thus the sequence diverges to oo. The sequence, however, is clearly not monotone.
EXERCISES 2.3
1. Let 1n = [an, n E N, be closed and bounded intervals satisfying I,, D I,,. i for all n. Prove that
x
n 1n = [a, b],
"=1
where a = sup{an : n E N} and b = inf{b : n e NJ.
2. 'Show by example that the conclusion of Corollary 2.3.3 is false if the intervals 1,, with 1. D 1_ are not
bounded.
3. Show that each of the following sequences are monotone. Find a lower or upper bound if it exists; find the limit if
you can.
{ a+If)(a- J-)1,wherea>
t It
b. 1
2
66 Chapter2 Sequences of Real Numbers
xn+1 = 21X,+SJ.
\
*a. Show that the sequence {x"} is monotone and bounded.
b. Prove that lim x" = %/a-.
nb.
c. Prove that 0 c x" - NF. s (x.
10. In Exercise 9, let a = 3 and x, = 2. Use part (c) to find x" such that Ix" - \/31 < 10-5.
11. Let A be a nonempty subset of R that is bounded above and let a = sup A. Show that there exists a monotone in-
creasing sequence {a"} in A such that a = I m a". Can the sequence (a"} be chosen to be strictly increasing?
12. Use Example 2.3.5 to find the limit of each of the following sequences.
- n) '
1 j(1
*a. lI + +
d.
*c. 1 jl l
+ 2n/3n}
n
Sn= +2+ 1 1
+n
Show that {s,} is monotone increasing but not bounded above.
14. For each n E N, let
I I
s"= + 72+ +
n
Show that the sequence {s"} is monotone increasing but not bounded above.
15. *For each n E N. let
S"= i2+22+.+n2.
1I 1
Show that the sequence {s"} is monotone increasing and bounded above by 2.
16. Let 0 < b < 1. For each n E N, let s" = I + b + b2 + + b". Prove that the sequence {sue} is monotone in-
creasing and bounded above. Find lim s".
2.4 Subsequences and the Bolzano-Weierstrass Theorem 67
n n
18. *Which of the sequences in the previous exercise are monotone? Explain your answer.
19. If a -+oc and (b^} converges in R, prove that (a, + diverges to oc.
20. If a, > 0 for all n E h and lim a^ = 0, prove that I la, -+ oc.
21. Suppose a, > a, > 0. For n ? 2, set a^ _ i = ; (a, + a^ 1). Prove that
a. {au, ;} is monotone decreasing. b. {a2k) is monotone increasing, and
c. {a^} converges.
22. Let {s,} be a bounded sequence of real numbers. For each n E N. let a^ and b^ be defined as follows:
a, = inf{sk : k ? n}, b" = sup{sk : k ? n}.
a. Prove that the sequences {a^} and (b,,) are monotone and bounded.
b. Prove that lim a^ = lim b" if and only if the sequence {s^} converges.
nix nix
23. *In Theorem 2.3.2 we used the supremum property of R to prove that every bounded monotone sequence con-
verges. Prove that the converse is also true; namely, if every bounded monotone sequence in l8 converges. then
every nonempty subset of R that is bounded above has a supremum in R.
24. *Use the nested intervals property to prove that [0, 1 ] is uncountable.
2.4.1 DEFINITION Given a sequence { p^} in OB, consider a sequence {nk}k 1 of positive
integers such that n, < n, < n3 < . Then the sequence { p^,}k , is called a sub-
sequence of the sequence { p"}.
If the sequence {p,,} converges, its limit is called a subsequential limit of the se-
quence { p,}. Specifically, a point p E 18 is a subsequential limit of the sequence { p"}
if there exists a subsequence {p,,,} of {p,,} that converges to p. Also, we say that oo is
a subsequential limit of { p"} if there exists a subsequence { p,, } so that p^, -+oo as
k -+ oo. Similarly for -no.
2.4.2 EXAMPLES
(a) Consider the sequence {(I - (-1)")}. If n is even, then a, = 0, and if n is odd, then
a = 2. Thus 0 and 2 are subsequential limits of the given sequence. That these are the
only two subsequential limits is left to the exercises (Exercise 1).
68 Chapter 2 Sequences of Real Numbers
(b) As our second example, consider the sequence {(-1 )" + }. Both I and - I are
subsequential limits. If n is even, i.e., n = 2k, then
an = au = l +
2k'
which converges to t. On the other hand, if n is odd, i.e., n = 2k + 1, then
1
a,, = a2k+ i = -1 +
2k + 1'
which converges to -1. This shows that -1 and I are subsequential limits. Suppose
{a",} is any subsequence of {a"}. If the sequence Ink) contains an infinite number of
both odd and even integers, then the subsequence {a",} cannot converge. (Why?) On the
other hand, if all but a finite number of the nk are even, then {a",} converges to 1. Sim-
ilarly, if all but a finite number of the nk are odd, then {a",} converges to -1. Thus -1
and I are the only subsequential limits of {a"}.
(c) Consider the sequence {n(1 + If n is even, then n(1 + (-I )") = 2n,
whereas if n is odd, n(1 + (-1)") = 0. Thus 0 and oo are two subsequential limits of
the sequence. The same argument as in (b) proves that these are the only two subse-
quential limits.
Our first result assures us that for convergent sequences, every subsequence also
converges to the same limit.
2.4.3 THEOREM Let (p,,) be a sequence in R. If {p.1 converges to p, then every subse-
quence of { p"} also converges to p.
Proof. Let {p",} be a subsequence of {p"}, and let e > 0 be given. Since p"-+p,
there exists a positive integer n such that 1p. - pl < e for all n ? n0. Since {nk} is
strictly increasing, nk ? no for all k ? n". Therefore,
Ipn, - pI < e
forallk? n0;i.e.,p",-+p.
2.4A EXAMPLES
(a) In our first example we illustrate how Theorem 2.4.3 may be used to compute the
limit of a sequence. Consider the sequence {p"} where 0 < p < 1. By Example
2.3.4(b), the sequence {p"} is monotone decreasing, bounded below, and hence con-
verges. Let
a = lim p".
By Theorem 2.4.3 the subsequence {pen} also converges to a. But p2" = (p")2, and thus
a = lim p2" = lim (p")2 = a2.
4 (b) In our second example we show how the previous theorem may be used to pros e
divergence of a given sequence. Consider the sequence { sin n07r}, where 8 is a ra-
tional number with 0 < 8 < 1. Write 0 = alb, with a, b E N and b ? 2. When
n = kb, k E N, then sin n8ir = sin kair = 0. Therefore 0 is a subsequential limit of
the sequence. On the other hand, if n = 2kb + 1, k E N, then
Since 0 < alb < 1, sin bar * 0. Thus sin bar is another distinct subsequential limit
of {sin nBar}. Hence as a consequence of Theorem 2.4.3, the sequence { sin n07r}
diverges. The result is still true if 0 is irrational. The proof however is much more dif-
ficult.
2A.6 EXAMPLES
(a) E _ (a, b), a < b. Every point p, a < p < b, is a limit point of E. This fol-
lows from the fact that for any e > 0, there exists a point x E (a, b) satisfying
p < x < p + e. These, however, are not the only limit points. Both a and b are also
limit points of E, but they do not belong to E.
(b) E _ {,'-, : n = 1, 2, ...}. Each n is an isolated point of E. If a is chosen so that
0<E< I _ I -I
n(n + 1) n+I n'
then N,(,-,) n E _ Hence no point of E is a limit point of E. However, 0 is a
limit point of E which does not belong to E. To see that 0 is a limit point, given
70 Chapter2 Sequences of Real Numbers
1 EN,(0)flE,
n
and thus 0 is a limit point of E.
(c) Let E =o fl [0, 1 ]. If p (4 [0, 1 ], then p is not a limit point of E. For if p > 1.
then for e = 1(p - 1) we have k(p) fl E = 46. Likewise when p < 0. On the other
hand, every p E [0, 1] is a limit point of E. Let e > 0 be given. Suppose first that
0 s p < 1. Then by Theorem 1.5.2 there exists r E Q such that
p<r<min{p+e.l}.
When p = 1, Theorem 1.5.2 also guarantees the existence of an r E Q fl [0, ii with
p - to < r < p. Thus for every e > 0, N,(p) contains a point r E E with r * p. The
same argument also proves that every point of R is a limit point of Q.
Proof.2 (a) Suppose there exists a neighborhood Na of p which contains only finitely
many points of E, say q1, ... , q, # p. Let
emin{lp-q1J... >lp-q.11.
Then N,(p) fl E contains at most p. Thus p is not a limit point of E.
(b) We construct the sequence in E as follows: Since p is a limit point of E,
for each positive integer n, by the definition of limit point, there exists p E E with
p # p such that I p - p I < ,1-,. The sequence {p.) clearly satisfies p --.>p. L)
2.4.9 EXAMPLE In this example we illustrate part (b) of the previous theorem. Since
is a limit point of Q (Example 2.4.6(c)), by Theorem 2.4.7(b) there exists a sequence
of rational numbers with r -> N/-2. Such a sequence, however, need not be unique.
If r V, then the same is also true for the sequence {r + ,1-,}.
2. The proof of (a) is an example of a proof by contraposition. In this method of proof we assume the nepa-
tion of the conclusion and prove the negation of the hypothesis.
2.4 Subsequences and the Bolzano-Weierstrass Theorem 71
Bolzano-Weierstrass Theorem
We are now ready to state and prove the Bolzano-Weierstrass theorem. This impor-
tant result was originally proved by Bernhard Bolzano (1781-1848) and modified
slightly in the 1860s by Karl Weierstrass (1815-1897). The theorem is one of the
fundamental results of real analysis. Its importance will become evident in subse-
quent sections. Although not immediately obvious from the proof, the least upper
bound property plays a crucial role in the proof of the theorem. The least upper
bound property was used in proving the nested intervals property (2.3.3), and this re-
sult is crucial in the proof.
each of length (b - a)/2. At least one of these, call it 12, must contain infinitely many
points of S. If not, then S would be the union of two finite sets and thus be a finite set
itself. Repeating this process, we obtain a sequence {1"} of closed and bounded inter-
vals satisfying
(a) [a, b]=11DIZD.. .D1"D...,
(b) length of I" = (b - a)/2"-, and
(c) for each n, in fl s is infinite.
By Corollary 2.3.3, lono- 11, * 0. Let x e fly 1 In. It remains to be shown that x is a
limit point of S. If e > 0 is given, choose n E N so that (b - a)/2"-1 < e. If y E I",
then
ly - xj S length I" < e.
Therefore 1" C N.(x). Since In fl S is infinite, there exists y E S fl N1(x) with y * x.
Since e > 0 was arbitrary, x is a limit point of S.
The conclusion of the Bolzano-Weierstrass theorem may fail if either hypothesis is
removed. By Corollary 2.4.8 a finite set has no limit point. On the other hand, the set N
is an infinite unbounded subset of II with no limit points. The following corollary is of-
ten called the sequential version of the Bolzano-Weierstrass theorem.
IP - Pn,I < kI
Such an integer nk exists since every neighborhood of p contains infinitely many points
of S. The sequence {p",}k , is a subsequence of (pn) that by construction converges
to p.
An argument similar to the one used in the previous Corollary may be used to prove
the following.
2.4.12 THEOREM Let { pn} be a sequence in R. If p is a limit point of { p" : n E K), then
there exists a subsequence {p.,) of { such that p,,, -+p as k -+ co.
Proof. Exercise It.
As an application of Theorem 2.4.12 we consider the following example.
2.4.13 EXAMPLE Since the set E = a fl [0, 1 ] is countable, there exists an enumeration
{rn : n E NJ of E. The sequence {rn}W , is called an enumeration of the rational numbers
in (0,1 ]. By Example 2.4.6(c), every p E [0, 1 ] is a limit point of {r" : n = 1, 2, ...}.
Thus if p E [0, 1 ], there exists a subsequence {rn } of {r,} such that rn, -+p. The se-
quence {rn} has the property that every p e [0, 1) is a subsequential limit of the se-
quence. This sequence also provides an example of a sequence for which the set of sub-
sequential limits of the sequence is uncountable.
EXERCISES 2.4
1. a. Prove that 0 and 2 .are the only subsequential limits of the sequence (1 - (-1n
b. Prove that 0 and oo are the only subsequenual limits of the sequence {n(1 +
2. a. Construct a sequence {sn} for which the subsequential limits are {-oo, -2. 1}.
b. Construct a sequence {s"} for which the set of subsequential limits of the sequence is countable.
3. The proof of Corollary 2.4.11 is an example of a proof by cases (see Appendix A.3). If S = { p, : n E101}.
then there are two cases; namely. S is finite or S is infinite. For each case we prove the existence of a con-
vergem subsequence of {p"}.
2.5 Limit Superior and Inferior of a Sequence 73
b. s cos
Slln
-
3. Find all the subsequential limits of each of the following sequences.
*a. s sin 2
2
l
c. Sll-I)" +12sin
l -I"
2 1
d.
l
sin?
*e. I--
n
f {(1.5
4. Use Example 2.3.5 to find the limit of each of the following sequences. Justify your answer.
*a.
I \ + 3n1J b. .CI + 2nI
j c_ 1(I + n/nj
5. Suppose p > 1. Use the method of Example 2.4.4 to show that lim = 1.
6. For n E Nl, set p" = n'I".
a. Prove that I < p, , < p" for all n ? 3.
b. Let p = limp". Use the fact that the subsequence {p,,,} of {p,,} also converges to p to conclude that p = 1.
7. Determine the limit points and the isolated points of each of the following sets.
"
((
*a. S1-
n
:nEN! b. -1)"+!:nE!i } 1
ll
c. (0. l) U {2} d. N
e. Q8\Q f. Qfl(0,1)
8. Let A be a nonempty subset of U8 that is bounded above and let a = sup A. If a E A. prove that a is a limit point
of A.
9. Let { p"} be a bounded sequence of real numbers and let p E R be such that every convergent subsequence of
converges to p. Prove that the sequence {p") converges to p.
10. a. Construct a subset of 68 with exactly two limit points.
b. Find an infinite subset of R with no limit points.
c. Construct a countable subset of U8 with countably many limit points.
d. Find a countable subset of R with uncountably many limit points.
11. Prove Theorem 2.4.12.
12. Prove that every sequence in R has a monotone subsequence.
13. Use the Bolzano-Weierstrass theorem to prove the nested intervals property (Corollary 2.3.3).
14. *Use the nested intervals property to prove the least upper bound property of R.
15. Prove that every uncountable subset of R has a limit point in R.
2.5.1 DEFINITION Let {s"} be a sequence in R. The limit superior of {s"}, denoted lim s"
"~"
or lim s", is defined as
ligoo s" k
urn bk A sup{S" : n > k}.
The limit inferior of {s"}, denoted lim s" or lim s", is defined as
"-00
lim s" = lim
k-oo
ak = sup inf{s" : n at k}.
nToo kEN
We now give several examples for which we will compute the limit inferior and
limit superior. As will be evident, these computations are very tedious. An easier
method will be given in Theorem 2.5.7.
2.5.2 EXAMPLES
(a) {1 + (-1. Let s,, = 1 + (-1)". Then s" = 2 if n is even, 0 otherwise. Thus
ak = 0 for all k and bk = 2 for all k. Therefore
lim s" = 2 and lims"=0.
(b) {n(1 + (-,. In this example,
ifnisodd.
s"=n(1 +(-1)")={0
2. if n is even.
1+ n even.
n
The first few terms of the sequence {sn} are illustrated in Figure 2.4.
$5 S3 $3 S( 'i4 52
0 I 75 3
5 3 64 2
Figure 2.4
To compute the limit superior and inferior of the sequence {sn}, we set Ek =
{sn : n ? k}. If k is even, then
Ek1+k,-1+k+1' I+kI . 1.
Therefore, for k even,
1 + k, k even,
a. _ -1 for all k. and b. _
1
1+ k odd.
k 1
Thus
Our first theorem provides an (e, n0) characterization of the limit superior. An anal-
ogous characterization for the limit inferior is given in Theorem 2.5.4.
76 Chapter 2 Sequences of Real Numbers
Remark. The statement "s,, < (3 + e for all n a means that sn < /3 + e for all
but finitely many n. On the other hand, the statement "given n, there exists k E N with
k ? n such that sk > /3 - e" means that sn > /3 - e for infinitely many indices n.
Proof of Theorem 2.5.3. We will only prove (a). The proofs of (b) and (c) are left
to the exercises (Exercise 5).
(a) Suppose /3 = lim S. = rlim bk where
Since the sequence {bn} is monotone decreasing. bn <- 6 + E for all n ? n,,. Thus
2.5.5 COROLLARY Jim sn = lim s" if and only if Jim s" exists in R U {-oo, oo}.
n-+00 nix
n-M
Proof.. Suppose lim s" = lim sn = a E R. Let e > 0 be given. By (a) of the previ-
ous two theorems, there exist positive integers ni and n2, such that
s" < a + e for all n ? n,, and
sn > a - E for all n ? n,.
Thus if n,, = max{nt, n2},
a - e < sn<a+E
for all n ? no; i.e., n-.oo
lim sn = a. The proofs of the cases a = oo or a = -oo are similar.
If lim sn = a, then it easily follows that both Jim s,, = a and lim s, = a. O
<- nyOG
lim (an + bn) S ntiR
lim an + lim b,,.
Proof. Exercise 6. U
The following theorem relates the limit superior and inferior of a sequence to the
subsequential limits of the sequence, and is in fact very useful for finding Tim- S. and
lin s of a sequence {sn}.
n:
Then lim s, and lim s" are in E and
"x-100
Proof. Let s = lim s". Suppose s E R. To show that s E E, we show the existence of
a subsequence {s"j of {s"} which converges to s. Take e = 1. Let n, be the smallest in-
teger such that
s - I <S". < S + 1.
Such an integer exists by (i) and (ii) of Theorem 2.5.3(a). Suppose the integers
nl < n2 < .. < nt have been chosen. Take e = 1/(k + 1). Let n1 be the smallest
integer greater than nk such that
$ -k+1<S",. <S+k+1
1 1
Again, such an integer exists by (i) and (ii) of Theorem 2.5.3(a). Then (sn j is a subse-
quence of {s"} which clearly converges to s. Therefore s E E. The case s = no is
treated similarly. Ifs = -oo, then by (c) of Theorem 2.5.3, sn -i -oo as n -* no.
Since s E E, s < sup E. It remains to be shown that sup E = s. If s = no we are
done. Otherwise, suppose sup E = S > s. Suppose S # no. Then there exists a E E
such that
sass.
Since a E R, we can choose e > 0 such that s + e < a - e. For this e, there exists
n0 E N such that s" < s + e for all n z n0. Hence there can exist only finitely many k
such that
(sk - al < e.
Consequently no subsequence of {s"} can converge to a. This contradiction shows that
sup E = s. The case S = no is treated similarly. Q
2.5.8 EXAMPLES In the following examples we use Theorem 2.5.7 to compute lim s" and
lim s" for each of the given sequences {s"}.
(a) Let s,, = (- I)" + 1. By Example 2.4.2(b), the set of subsequential limits of {sn} is
{-1, I}. Thus by the previous theorem,
lim sn = -1 and lim s" = 1.
(b) Let sn = n(1 + (-1)"). By Example 2.4.2(c) the subsequential limits of {sn} are 0
and no. Therefore,
lim s" = 0 and lim s" = no.
2.5 Limit Superior and Inferior of a Sequence 79
(c) Let s" = sin" If n is even, i.e., n = 2k, then s,t = sin kir = 0. On the other
.
EXERCISES 2.5
1. Find the limit inferior and limit superior of each of the following sequences.
Remark. In Definition 2.6.1, the criterion I p - p,,,l < e for all integers n. m ? it,,
is equivalent to
PnI = 0 (4)
for every k E N. The converse, however, is false: namely, if { p,,} is a sequence in 6a that
satisfies lim 0 for every k E N. this does not imply that'the sequence
n-.x
is a Cauchy sequence (Exercise 4). The hypothesis (4) only implies that for each
k E N, given E > 0, there exists a positive integer n such that Ip,,.a - E for all
n at n,,,
2.6.2 THEOREM
(a) Every convergent sequence in R is a Cauchy sequence.
(b) Every Cauchy sequence is bounded.
Proof. (a) Suppose that {p,,} converges to p E R. Let e > 0 be given. Then for the
given e, there exists a positive integer n,, such that
l
In. - PI < e
that I p,, - p0I < I for all n, m ? no. Therefore, with m = nby Corollary 2.1.4(b).
I PnI { 1 + IP I for all n ? n,,. Let
M = max{1 + Ipnj, Ipil,. . ., Ipn _,I}.
Then for all n, I pnI :SM. Thus {pn} is bounded. 0
Proof. Suppose { p,, } is a convergent subsequence of { pn} with lim pn, = p. Let
e > 0 be given. Since {pn} is Cauchy, there exists an integer ni such that
Remark The statement that "every Cauchy sequence in R converges" is often ex-
pressed by saying that R is complete. Since the proof-of Theorem 2.6.4 used the
Bolzano-Weierstrass theorem, the completeness of R ultimately depends on the least
upper bound property of R. Conversely, if we assume completeness of K, then we can
prove that R satisfies the least upper bound property (Exercise 12). For this reason, the
least upper bound or supremum property of R is often called the completeness prop-
erty of R.
2.6.5 EXAMPLES
(a) For our first example we consider the sequence {sn} where for n E N,
1+
2+. .+'2.
.
2 n
82 Chapter2 Sequences of Real Numbers
For k E N,
(n+l)z+...+(n+k)z
k E N. Thus IS. +k - S. I < e for all n ? no and all k E N. Therefore the sequence {s"}
is a Cauchy sequence.
(b) In this example we illustrate how the concept of a Cauchy sequence may be used
to prove convergence of a given sequence. Additional applications will be given in the
exercises. Let a,, a2 be arbitrary real numbers with a1 # a2. For n > 3, define an in-
ductively by
Our first goal is to show that the sequence {a"} is Cauchy. We first note that
an+1-an=-2(an-an-1)
As a consequence, for n ? 2,
1 2"
an+1 - an (a2 - a1). (6)
This last statement is most easily verified by induction (Exercise 5). For m ? 1, con-
Ian+'n - anI
m:Ian+k+I
sider Ian+," - a,, 1. By the triangle inequality,
- an+k) S I
7,(an+k+I -
-1
-0
which by (6)
I 1 I m
1
Ia2 - a]I 7, 2n+k-f 2"-21a2 - all 7, 2k.
By Example 1.3.3(a)
rk r-r'* r # 1. (7)
kl 1-r
2.6 Cauchy Sequences 83
Thus with r = 2,
M 1
2 - 2)m+I l
= 1 - - < 1.
k=12k 1-Z 2m
Therefore,
1
Ian+m - and :S- 22 Ia2 - all (S)
2
for all n >- 2 and m E N. Let e > 0 be given. Choose no such that lag - a, I/2"-2 < e
for all n ? no. Then by (8),
Ian+m - and < E
for all m E N, n ? nn. This, however, is just another way of stating that
Ian - am < e for all m, n ano.
Therefore the sequence {an) is a Cauchy sequence in R, and thus by Theorem 2.6.4.
a = lima"
n- 00
exists in R.
Can we find the limit a here? If we follow the approach in Example 2.3.4(c), by
taking the limit of both sides of equation (5), we only get a = a. To find the value of a,
let us observe that
an+1 - a, = (an+, - an) + (an - a"-,) + ... + (a2 - a1)
n
Y, (ak+ I - ak),
k=1
The last equality follows from formula (7). Since an+, and (-3f -+0, upon tak-
ing the limit of both sides we obtain
2
a -a, = (a2 - a,) or a=a,+3(a2-a,).
Contractive Sequences
One of the key properties of the sequence {an} of the previous example was that
for all n ? 2. This property was used to show that the sequence {a"} was a Cauchy se-
quence and thus converged. Such sequences are commonly referred to as contractive
sequences. We make this precise in the following definition.
forallnEN,na2.
If {p.) is a contractive sequence, then an argument similar to the one used in the
previous example shows that
IPn+I - P"I `- b"-'IP2 - P1I
for all n >_ 1, and that
IP"+m-P.1
Cb"-'IP2-PII(l+b+. +b'"-I)< 1-bIP,-PII
b"
I(b)
quence {p.1 is contractive and p = lim p", then
P-P"::=:':r'
where 0 < b < 1 is the constant in Definition 2.6.6.
Proof. We leave the details of the proof to the exercises (Exercise 7). 0
2.6.8 EXAMPLE Suppose we are given that the polynomial p(x) = x2 - 3x + I has ex-
actly one zero in the open interval (0, 1). If c r= (0, 1) is such that p(c) = 0, then
c = (c2 + 1). We start with cl E (0, 1) arbitrary, and for n a- I we set
i
1 ,
c"+1=3(c;,+I).
Since cl E (0, 1) we have c2 E (0, 1), and by induction, c" E (0, 1) for all n E N. To
prove that the sequence {c"} converges we prove that it is contractive. For n > 2 we
have
Ic"+I-c"I=I3(c+1)-3(c,,-1+1)I
2.6 Cauchy Sequences 85
2
:5 3ICn - Cn-1I
< 10-1.
3 12 2
Solving the above inequality for n - I gives
(n - 1) >
3 In 10 - In 4
n3 - In 2 = 13.62 (to two decimal places).
Thus the choice of n = 15 is sufficient to ensure accuracy to within 10-3. This estimate
on n however is very crude. In fact, after six steps, part (a) of Theorem 2.6.7 only gives
.Ic - c6I < 3(3)SIc2 - c1I = .032922. The actual approximation however is much bet-
ter than this. By computation we find that c3 = .391204, c4 = .384347, cs = .382574.
and c6 = .382121. By part (b) of Theorem 2.6.7, Ic - c6I 21c6 - csl = 000906,
and this is less than 10-3.
0 EXERCISES 2.6
1. If and are Cauchy sequences in R, prove (without using Theorem 2.6.4) that {a + and are
also Cauchy sequences.
2. For each of the following, determine whether the given sequence is a Cauchy sequence.
+l} b. {(-I)"} c.
(n+(-I}"}
-a. In in ll(( n
adI
. 1(1)'n {.1 + (-l)nn2}
l
1n2+3 2n2 +3
3. For n E 4 let
S. +21+31++n1.
Tlak=ap+ap,,+ +av.
k=p
2.7.1 DEFINITION Let {an}'°_, be a sequence of real numbers. Let {s,,}., be the sequence
obtained from where for each n E N, sn = 7-k=, ak. The sequence is called
an infinite series, or series, and is denoted either as
2.7 Series of Real Numbers 87
ak or as a1 + a2 + + an +
k-I
For each n E N, s" is called the nth partial sum of the series and a" is called the nth
term of the series.
The series 7,I ak converges if and only if the sequence (s"} of nth partial sums
converges in R. if im s" = s, then s is called the sum of the series, and we write
00
s = I a,
k=1
If the sequence {s"} diverges, then the series joko. I ak is said to diverge.
2.7.2 EXAMPLES
(a) For (rl < 1, consider the geometric series
00
lrrk
k=1
s"=
k-I
rk=r+r2+ +r
Thus
(1 -r)s"=s"-rs"=r-r"+',
and as a consequence,
S" _ r - r"+
I-r
Since Iri < 1, by Theorem 2.2.6(e),"-.00
lim r" = 0. Therefore lira s" = r/(1 - r), and
thus '~°°
°O _ r
A rk n!<I.
1)
Then
n
s" _ j ak
k-I
2.7.3 THEOREM (Cauchy Criterion) The series Y,k I ak converges if and only if given
e > 0, there exists a positive integer n,,. such that
< e
k_n+I akI
Remark The previous theorem simply states that the series I ak converges if and
only if the sequence {s,,} of nth partial sums is a Cauchy sequence.
2.7.4 EXAMPLE In this example we show that the series 7" 1 k diverges. We accomplish
this by showing that the sequence (sn) of partial sums is not a Cauchy sequence. Con-
sider
stn-sn=n+t.+.
nEN.
+2n'
There are exactly n terms in the sum on the right, and each term is greater than or equal
to 1/2n. Therefore,
)= 1'
(s2. -sn ? n J
2n 2
The sequence {sn} therefore fails to be a Cauchy sequence and thus the series diverges.
The divergence of this series appears to have been first established by Nicole Oresme
(1323?-1382) using a method of proof similar to that suggested in the solution of Ex-
ercise 13 of Section 2.3.
2.7 Series of Real Numbers 89
Remark. The condition lim at = 0 is not sufficient for the convergence of I ak. For
example, the series 1 diverges, yet slim k = 0.
k -oc
2.7.6 THEOREM S u p p o s e ak at 0 f o r all k E N. T h e n E , at converges if and only if
is bounded above.
Proof. Since ak a 0 for all k, the sequence is monotone increasing. Thus by The-
orem 2.3.2, the sequence converges if and only if it is bounded above.
EXERCISES 2.7
1. *'Using the inequality
1 I 1 1
k2 k(k - l) k-I k
k
converges.
NOTES
This chapter provided our first serious introduction to the consequences of the least upper bound property of R. The
limit process. In subsequent chapters we will encounter least upper bound property was used to prove that every
limits of functions, the derivative, and the integral, all of bounded monotone sequence converges. This result was
which are further examples of the limit process. Of the subsequently used to prove the nested intervals property.
many results proved in this chapter, it is difficult to select which in turn was used to prove the Bolzano-Weierstrass
one or two for special emphasis. They are all important! theorem. By Exercise 23 of Section 2.3 and Exercises 13
Many of them will be encountered again-either directly and 14 of Section 2.4. each of these imply the least upper
or indirectly-throughout the text. bound property of R. Another property of the real num-
Some of the concepts and results of this chapter have bers that is equivalent to the least upper bound property is
certainly been encountered previously; others undoubtedly the completeness property of R; namely, every Cauchy
are new. Two concepts that may not have been previously sequence of real numbers converges. Other consequences
encountered are limit point of a set and the limit superior of the least upper bound property will be encountered in
(inferior) of a sequence of real numbers. The notion of subsequent chapters.
limit point of a set is one of the fundamental concepts of Cauchy sequences were originally studied by Cantor
analysis. We will encounter limit points again when we in the middle of the nineteenth century. He referred to
characterize the closed subsets of R. The notion of limit them as fundamental sequences and used them in his con-
point will also be crucial in the definition of the limit of a struction of the real number system R (see Miscellaneous
function. The results of Theorem 2.4.7, although elemen- Exercises 4-11). The main reason that these sequences are
tary, are very useful. The fact that every limit point of a set attributed to Cauchy, rather than Cantor, is because his
is the limit of a sequence of distinct points in the set will 1821 criterion for convergence of a series (Theorem 2.7.3)
be exploited in several instances in subsequent chapters. is equivalent to the statement that the sequence of partial
The primary importance of the limit superior and in- sums is a Cauchy sequence. The fact that Cauchy was a
ferior of a sequence is that these two limit operations al- more prominent mathematician than Cantor may also have
ways exist in R U {-oo, oo}. As we will see in Chapter been a factor. In later chapters we will encounter examples
7, this will allow us to present the correct statements of of spaces of functions that have defined on them a func-
the root and ratio test for convergence of a series. The tion, called a norm, having properties analogous to those
limit superior will also be required to define the radius of of the absolute value function. For such spaces it will also
convergence of a power series. There will be other in- be possible to define both convergence of a sequence and
stances in the text where these two limit operations will the notion of Cauchy sequence. Many of these spaces will
be encountered. also have the property that they are complete; that is, every
In this chapter we have proved several important Cauchy sequence in the space converges.
MISCELLANEOUS EXERCISES
The first three exercises involve the concept of an infinite product. Let {ak} be a sequence of
nonzero real numbers. F o r e a c h is = 1, 2, ... , define
pn = flak = a, . a2 . . . an.
k=1
If p = lim pn exists, then p is the infinite product of the sequence {ak}k 1, and we write
TM1
p=1
k-1
ak.
If the limit does not exist, then the infinite product is said to diverge. Some authors require that
p * 0. We will not make this requirement; rather we will specify p # 0 if this hypothesis is re-
quired in a result.
Miscellaneous Exercises 91
1. Determine whether each of the following infinite products converge. If it converges, find the infinite product.
Let Q denote the set of rational numbers. A sequence in 0 is Cauchy if for every r f-= Q, r > 0, there exists
a positive integer n° such that Ia - a,,' < r for all n, m z n°. A sequence in 0 is called a null sequence if
for every r (=- 0, r > 0, there exists a positive integer n° such that Ia I < r for all n z n°. Two Cauchy sequences
and in Q are said to be equivalent, denoted provided (a° - is a null sequence.
4. Let and be Cauchy sequences in Q. Prove the following.
a. {a°) -- b. If {b.), then {b°} -
One needs to show that these operations are well defined; that is, independent of the representative of the equiva-
lence class. For example, to prove that -a is well defined, we suppose that and are two representatives
of a; i.e., {a°} -- {b°}. But by 4(d), Therefore, [{-a.)) = [(-b°)]. This shows that -a is well
defined.
6. Prove that the operations + and are well defined on 9t.
For each p E Q, let { p} denote the constant sequence p. and set ap = [{ p}]. Also, we set
0 = [{0}], t, = [{l}].
As we will see, the element 0 will be the zero of 9t and k will be the unit of R. A Cauchy sequence in 0 be-
longs to 0 if and only if b -> 0. Similarly, E i if and only if (a - 1) -+ 0. The following problem provides
us with the multiplicative inverse of a # 0.
92 Chapter2 Sequences of Real Numbers
7. If a # 0, prove that there exists E a such that a, * 0 for all n E N, and that { is a Cauchy sequence.
Define a-' _
S. Prove that 9t with operations + and is a field.
We now proceed to define an order relation on 9t. A Cauchy sequence in Q is positive if there exists
r E Q, r > 0. and no E N such that a > r for all n >_ no. Let 91 be defined by
91 _ is a positive Cauchy sequence}.
9. Prove that the set 'satisfies the order properties (01) and (02) of Section 1.4.
10. Show that the mapping p -> a,, is a one-to-one mapping of Q into % which satisfies
ap+a, =ap,q
The above exercises prove that 9t is an ordered field that satisfies the least upper bound property
One can show that any two complete ordered fields are in fact isomorphic, that is, there exists a
one-to-one mapping of one onto the other that preserves the operations of addition, multiplication,
and the order properties. Thus 9t is isomorphic to the real numbers R.
SUPPLEMENTAL READING
Aguirre, J. A. F., "A note on Cauchy sequences," Math. subsequences;' Amer. Math. Monthly 95 (1988),
Mag. 68 (1995), 296-297. 44-45.
Bell, H. B., "Proof of a fundamental theorem on se- Staib, J. H. and Demos, M. S., "On the limit points of
quences;' Amer. Math. Monthly 71 (1964), 665-666. the sequence (sin n 1:' Math. Mag. 40 (1967),
Goffman. C., 'Completeness of the real numbers;' Math. 210-213.
Mag. 47 (1974), I-8. Wenner, B. R., 'The uncountability of the reals;' Amer
Newman, Donald J. and Parsons, T. D., "On monotone Math. Monthly 76 (1969), 679-680.
3 Structure of Point Sets
3.1 Open and Closed Sets
In this chapter we introduce some of the basic concepts fundamental to the study of limits
and continuity, and study the structure of point sets in R. The branch of mathematics con-
ceded with the study of these topics-not only for the real numbers but also for more gen-
eral sets-is known as topology. Modern point set topology dates back to the early part of
this century; its roots, however, date back to the 1850s and 1860s and the studies of Bolzano,
Cantor, and Weierstrass on sets of real numbers. Many important mathematical concepts de-
pend on the concept of a limit point of a set and the limit process, and one of the primary
goals of topology is to provide an appropriate setting for the study of these concepts.
Although we restrict our study to the topology of the real line, all of the concepts
encountered in this chapter can be defined in the more general setting of metric spaces.
A thorough understanding of these topics on the real line will prove invaluable when
they are encountered again in more abstract settings. On first reading, the concepts in-
troduced in this chapter may seem difficult and challenging. With perseverance, how-
ever, understanding will follow.
NN(P) N,(b)
a P b
Figure 3.1
(b) Let E denote the set of irrational real numbers, i.e., E = R \ Q. If p E E, then by
Theorem 1.5.2, for every e > 0 there exists r E Q fl N,(p). Thus N,(p) always con-
tains a point of R not in E. Therefore no point of E is an interior point of E; i.e.,
Int(E) = 0. Using the fact that between any two real numbers there exists an irrational
number (Exercise 6, Section 1.5), a similar argument also proves that Int(Q) _ ¢.
3.1.3 DEFINITION
(a) A subset 0 of R is open if every point of 0 is an interior point of 0.
(b) A subset F of P is dosed if F` = R \ F is open.
Remark. From the definition of an interior point it should be clear that a set 0 C R
is open if and only if for every p E 0 there exists an e > 0 (depending on p) so that
N,(p) C 0. In Theorem 3.1.9 we will provide a characterization of closed sets in terms
of limit points.
3.1.4 EXAMPLES
(a) The entire set N is open. For any p E N and e > 0, N,(p) C R. Since R is open,
by definition the empty set 0 is closed. However, the empty set is also open. Since 4)
3.1 Open and Closed Sets 95
Ne(p)
-N6(q)
p-E p q p+E
Figure 3.2
(c) Let E = (a, b], a < b, be as in Example 3.1.2(a). Since the point b E E is not an
interior point of E. the set E is not open. The complement of E is given by
E`=(-oo,a]U(b,oo).
An argument similar to the one given in Example 3.1.2(a) shows that a is not an inte-
rior point of E`. Thus E` is not open and hence by definition E is not closed. Hence E
is neither open nor closed.
(d) Let F = [a, b], a < b. Then
F` _ (-oo, a) U (b, oo),
and this set is open. This can be proved directly, but also follows as a consequence of
Theorem 3.1.6(a) below. Thus F is closed.
(e) Consider the set 0. Since no point of 0 is an interior point of Q (Example
3.1.2(b)), the set Q is not open. Also, C is not closed.
The use of the adjective open in describing the intervals (a, b), (a, on), (-cc, b)
and (-oo, oo) is justified by the following theorem.
3.1.6 THEOREM
(a) For any collection {Oa},,EA of open subsets of R, n o. is open.
&FA
(b) F o r any f i n i t e collection {O,, ... , of open subsets of R. fl O; is open.
96 Chapter 3 Structure of Point Sets
pE0=.n 0;.
Then p E 0, for all i = 1, . .. , n. Since 0; is open, there exists an e; > 0 such that
NE,(p) C 0..
Let e = min{e,, ... , e,}. Then e > 0 and NE(p) C 0, for all i. Therefore
NE(p) C 0; i.e., p is an interior point of 0. Since p E 0 was arbitrary, 0 is open. C)
As a consequence of the previous theorem every closed interval [a. b], a, b E R with
a s b, is a closed subset of R. Since
H \ [a, b] = (-oo, a) U (b, oo)
is the union of two open intervals, by the previous theorem P \ [a, b] is an open subset
of R. Thus [a, b] is a closed set.
For closed subsets of R we have the following analogue of the previous result.
3.1.7 THEOREM
(a) For any collection {Fa}aEA of closed subsets of P. n F. is closed.
(b) For any finite collection IF,, ... , of closed subsets of R. U F is closed
j°
Proof. The proofs of (a) and (b) follow from the previous theorem and De Morgan's
laws:
A EA (")c
j=1 j .1
Remark. The fact that the intersection of a finite number of open sets is open is due
to the fact that the minimum of a finite number of positive numbers is positive. This
guarantees the existence of an e > 0 such that the e-neighborhood of p is contained in
the intersection. For an infinite number of open sets, the choice of a positive e may no
longer be possible. This is illustrated by the following two examples.
3.1.8 EXAMPLES We now provide two examples to show that part (b) of Theorem 3.1.6
is, in general, false for a countable collection of open sets. Likewise, part (b) of Theo-
rem 3.1.7 is, in general, also false for an arbitrary union of closed sets (Exercise 6b).
(a) For each n = 1, 2, .. . , let 0 = (-,1-,,,-,). Then each O is open, but
00
n1 O = {0},
3.1.9 THEOREM A subset F of R is closed if and only if F contains all its limit points.
Proof. Suppose F is closed. Then by definition F` is open and thus for every p E F'
there exists e > 0 such that N, (p) C F'; that is, Ne(p) fl F = 0. Consequently. no
point of F` is a limit point of F. Therefore F must contain all its limit points.
Conversely, let F be a subset of R that contains all its limit points. To show F is
closed we must show F'' is open. Let p E V. Since F contains all its limit points. p is
not a limit point of F. Thus there exists an e > 0 such that N,(p) n F = 44. Hence
N,(p) C F' and p is an interior point of V. Since p E F` was arbitrary, F` is open and
therefore F is closed.
Closure of a Set
3.1.10 DEFINITION If E is a subset of R, let E' denote the set of limit points of E. The
closure of E, denoted E, is defined as
E=EUE'.
3.1.11 THEOREM If E is a subset of R, then
(a) E is closed.
(b) E = E if and only if E is closed.
(c) E C F for every closed set F C R such that E C F.
Proof. (a) To show that E is closed, we must show that k is open. Let p E E`. Then
p 44 E.and p is not a limit point of E. Thus there exists an e > 0 such that
N,(p)flE_¢.
We complete the proof by showing that NE(p) fl E' is also empty and thus N,(p) fl E =
Therefore N,(p) C E`; i.e., p is an interior point of E`.
Suppose N,(p) fl E' * 4,. Let q E NE(p) n E', and choose S > 0 such that
Na(q) C N,(p). Since q E E, q is a limit point of E and thus N8(q) fl E * 46. But this
implies that N. (p) n E * tb, which is a contradiction. Therefore N,(p) n E' = 46,
which proves the result.
_ (b) If E = E, then E is closed. Conversely, if E is closed, then E' C E and thus
E. _
(c) If E C F and F is closed, then E' C F. Thus E C F.
98 Chapter3 Structure of Point Sets
3.1.13 THEOREM If U is an open subset of R, then there exists a finite or countable col-
lection {1n} ofpairwise disjoint open intervals such that
U=UI..
Proof. Let x E U. Since U is open, there exists an e > 0 such that
(X-E,x+E)C U.
In particular (s, x) and [x, t) are subsets of U for some s < x and some t > x. Define
r, and lx as follows:
r, = sup{t : t > x and [x, t) C U}, and
lx = inf{s : s < x and (s, x] C U}.
Then x < r, < oo and -oo s 1, < x. For each x E U, let 1, = (t r,). Then
(a) 1, C U, and
(b) if x, y E U, then either Ix = I, or 1, fl /,. = 4).
The proofs of (a) and (b) are left as exercises (Exercise 16).
To complete the proof, we let 2 = {I, : x E U}. For each interval I E 9, choose
rt E Q such that rt E I. If !, J E 2 are distinct intervals, then rt * rj. Therefore the
mapping I - rt is a one-to-one mapping of T into O. Thus the collection Y is at most
countable and therefore can be enumerated as {I;}iEA, where A is either a finite subset
of N, or A = N. Clearly
U=iE Ii,
1. This topic can be omitted upon first reading of the text. The structure of open sets will only be required in
Chapter 10 when defining the measure of an open subset of R.
3.1 Open and Closed Sets 99
and by (b), if n * j, then 1 f11, = 4). Thus the collection {1%EA is pairwise dis-
joint.
3.1.15 EXAMPLE Let X = [0, oo) and let U = (0, 1). Then U is not open in R but is open
in X. (Why?)
The following theorem, the proof of which is left as an exercise (Exercise 21), pro-
vides a simple characterization of what it means for a set to be open or closed in X.
Connected Sets2
Our final topic in this section involves the notion of a "connected set." The idea of con-
nectedness is just one more of the many mathematical concepts that have their roots in
Cantor's studies on the structure of subsets of R. When we use the term "connected sub-
set' of R, intuitively we are inclined to think of an interval as opposed to sets such as the
positive integers N or (0, 1) U {2}. We make this precise with the following definition.
3.1.17 DEFINITION A subset A of R is connected if there do not exist two disjoint open
sets U and V such that
(a) AlU*4)andAflV*e(,,and
(b) (An u)U(Af1V)=A.
2. this concept, though important and used implicitly in several instances in the text, will not be required
specifically in subsequent chapters except in a few exercises. Thus the topic of connectedne ss can be omitted
upon first reading of the text.
100 Chapter 3 Structure of Point Sets
The definition for a connected set differs from most definitions in that it defines
connectedness by negation; i.e., defining what it means for a set not to be connected.
According to the definition, a set A is not connected if there exist disjoint open sets U
and V satisfying both (a) and (b). As an example of a subset of R that is not connected,
consider the set of positive integers N. If we let U = (1, Z) and V = (, co). then U and
V are disjoint open subsets of 68 with
u n Nl l} and v n N = {2, 3, ...}
that also satisfy (U n N) U (V n N) = N. That the interval (a, b) is connected is a con-
sequence of the following theoreth, the proof of which is left to the exercises (Exercise 25).
EXERCISES 3.1
1. Prove Theorem 3.1.5.
2. *Show that every finite subset of I8 is closed.
3. Show that the intervals (-oo, a) and [a, oo) are closed subsets of R.
4. For the following subsets E of R, fill in the chart.
[o, 1]na
5. a. Let F be a closed subset of R and let { p,} be a sequence in F which converges to p E R. Prove that p E F.
b. Show by example that the conclusion is false if F is not closed.
6. *a. Prove Theorem 3.1.6(a).
b. Give an example of a countable collection of closed subsets of R such that U.=, F. is not closed.
7. Let A, B be subsets of R.
a. If A C B, show that Int(A) C Int(B). b. Show that Int(A n B) = Int(A) n Int(B).
c. Is Int(A U B) = Int(A) U Int(B)?
8. Let E be a subset of R.
*a. Prove that Int(E) is open. b. Prove that E is open if and only if E = Int(E).
c. If G C E and G is open, prove that G C Int(E).
3.2 Compact Sets 101
9. Let A. B be subsets of R.
*a. Show that (A U B) = A U B. b. Show that (A fl B) C A fl B.
c. Give an example for which the containment in part (b) is proper.
10. Prove that the set of limit points of a set is closed.
11. Let E C R. A point p E 18 is a boundary point of E if for every e > 0, N,(p) contains both points of E and
points of E. Find the boundary points of each of the following sets.
*a. (a, b) b. E = {,'-, : n E N} c. N d. Q
12. a. Prove that a set E C B8 is open if and only if E does not contain any of its boundary points.
b. Prove that a set E C 18 is closed if and only if E contains all its boundary points.
13. *Prove that the set of irrational numbers is dense in R.
14. *If D is dense in 18, prove that for every p E IB there exists a sequence in D with lim p = p.
15. Let Do = {0, 11, and for each n E N, let D" = {a/2" : a E N, u is odd, 0 < a < 2"}. Let D = U `-0 D". Prose
that D is a countable dense subset of [0, I ].
16. Prove statements (a) and (b) of Theorem 3.1.13.
17. *Prove that there exists a countable collection 9 of open intervals such that if U is an open subset of E8 and
p E U, there exists / E I with p E /" C U.
18. Let X = (0, oo). For each of the following subsets of X determine whether the given set is open in X. closed in X.
or neither.
*a. (0, 1 ] b . (0, 1 ) *c. (0, 1 ] U (2, 3) d. (0, 1 ] U (21 e. (1, : n E NJ
19. For each of the following subsets of 0, determine whether the set is open in 0, closed in Q. both open and
closed in 0, or neither.
a.A={pE0:I<p2<2} b.B=(pE0:2<p2<3) c.N
20. *If U is an open subset of 68, prove that E C U is open in U if and only if E is an open subset of R.
21. Prove Theorem 3.1.16.
22. Let X be a subset of R. Prove each of the following.
a. If {Ua},EA is any collection of subsets of X with U. open in X for all a E A. then Ua is open in X.
b. If { U,, ... , Ua} is any finite collection of subsets of X with each U; open in X. then f;'= , U, is open in X.
23. For each of the following, use the definition to prove that the given set is not connected.
*a. (0, 1) U {2} b. n= 1.2,...} c. (pE0:p>0andI <p2<3}
24. *a. If A is connected, prove that A is connected.
b. Give an example of a set A such that A is not connected, but A is connected.
25. *Prove Theorem 3.1.18.
Compact Sets
In this section we introduce the concept of a "compact set" and provide a characteriza-
tion of the compact subsets of R. The notion of a compact set is very important in the
study of analysis, and many significant results in the text will depend on the fact that
every closed and bounded interval in F8 is compact. The modern definition of a compact
102 Chapter3 Structure of Point Sets
set given in Definition 3.2.3 dates back to the second half of the nineteenth century and
the studies of Heine and Borel on compact subsets of R.
3.2.2 EXAMPLES
(a) Let E = (0, 1) and 0n = (0, 1 - ), n = 2,3..... Then 2 is an open
cover of E. To see this, suppose x E E. Then since x < 1, there exists an integer n such
that x < 1 - R. Thus x E 0,,, and as a consequence,
all
E C n=2
U 0,,.
which proves the assertion. In fact, since 0 C E for each n. we have E = U. 't2 0n.
(b) Let F = ;0, oo) and for each n E I,1, let U,, _ (-1, n). Then {Un}nENis an open
cover of F.
(c) If E is any nonempty subset of R, then for every e > 0. the collection
{N,,(x): x E E} is an open cover of E.
K C U0,.
Remark To prove (using the definition) that a given set K is compact, we must prove
that f o r m= open cover {O,} of K, there exists a finite subcollection 0,,, ... , 0..
whose union covers K. This is illustrated in (a) of the following examples. It is W suf-
ficient to find an open cover of K for which a finite subcollection also covers K. Such
an open cover always exists. (Why?) On the other hand, to show that a set E is not com-
pact, it suffices to find (construct) an open cover {O,} of E for which the union of no
finite subcollection contains E. This is illustrated in Examples 3.2.4 (b) and (c).
3.2A EXAMPLES
(a) Every finite set is compact. Suppose E = {p,, ... , pJ is a finite subset of N and
{0,},EA is an open cover of E. Then for each j, j = 1, ,.. , n, there exists a; E A such
that pi E 0,; But then {O,}7. is a finite subcollection which covers E.
3.2 Compact Sets 103
(b) The open interval (0, 1) is not compact. For the open cover {O"} 2 of (0, 1) in Ex-
ample 3.2.2(a), no finite subcollection can cover (0, 1). Suppose on the contrary that a
finite number, say O . . . , O",, covers (0, 1). Let N = max{n . . . , Pik}. Then
k N
(0.1)000"CUO"= 0,1-N,
which is a contradiction.
(c) The closed set F = [0, oo) is not compact. For the open cover CU.
of F, no finite subcollection can cover F. If there exist a finite number of sets in al1
which cover F, then there exists N E N such that F C (-1, N). (Why?) This, however,
is a contradiction.
3.2.5 THEOREM
(a) Every compact subset of IR is closed and bounded.
(b) Every closed subset of a compact set is compact.
Proof. (a) Let K be a compact subset of R. To show that K is bounded, consider the
open cover {(-k, k)}5EN of 6$, and hence also of K. Since K is compact, there exist
k,, ... , k" E IN such that
KCU
K C U N. (qj).
j=1 qj
(b) Let F be a closed subset of the compact set K and let {4".},,E.a be an open cover
of F. Then
{Oa}aEA U {F`}
is an open cover of K. Since K is compact. a finite number of these will cover K. and
hence also F.
As a consequence of the previous theorem, the open interval (0. 1) is not compact
since it is not closed. Likewise, the set A = {n : n E N} is not compact since again A
is not closed. By Example 2.4.6(b), the point 0 is a limit point of A but 0 0 A. Also, the
closed set [0, oo) is not compact since it is not bounded. In Theorem 3 .2.9 we will prove
the converse of Theorem 3.2.5(a); namely, every closed and bounded subset of l is
compact.
Remark. In proving that the compact set K was closed, compactness allowed us to
select a finite subcover from the constructed open cover of K. Finiteness then assured
that the e as defined was positive. This method of first constructing an open cover pos-
sessing certain properties and then using compactness to assure the existence of a finite
subcover will be used on other occasions in the text.
3.2.6 THEOREM If S is an infinite subset of a compact set K. then S has a limit point in K.
Proof.3 If no point of K is a limit point of S, then for each q E K, there exists a
neighborhood Nq of q so that Nq contains at most one point of S, namely q if q E S.
Since S is infinite, no finite subcollection of {Nq}qEK can cover S, and consequently no
finite subcollection of {Nq}qEK can cover K. This is a contradiction.
Remark. Since every compact subset of R is closed and bounded, the previous the-
orem can also be proved using the Bolzano-Weicrstrass theorem (2.4.10) (see Exercise
6). Moreover, as we will see in Theorem 3.2.9, if K is a subset of R having the prop-
erty that every infinite subset of K has a limit point in K, then K is compact. The prop-
erty that "every infinite subset of K has a limit point in K" is usually referred to as the
Bolzano-Welerstrass property.
Another useful consequence of compactness is the following analogue of the
nested intervals property.
K= fIK,,
3. The proof of Theorem 3.2.6 is again a proof by contradiction. In the proof we assume that the hypothesis
"S is an infinite subset of K and K is compact" is true, and that the conclusion "S has a limit point in K" is
false.
3.2 Compact Sets 105
n K. = 45 if and only if Ul O = R.
Thus if fl. , K. then is an open cover of R, and thus also of K1. But K,
is compact. Therefore there exists n, < < nk such that
k
K, C j_
U 0,,J..
But then K, fl K,,, fl fl K,,, _ ¢. This, however, is a contradiction, since the in-
tersection is equal to K,,,, which by hypothesis is nonempty. Thus K= fl K # 40. By
Theorem 3.1.7, K is closed, and hence by Theorem 3.2.5(b), K is compact. Q
3.2.8 THEOREM (Heine-Borel) Every closed and bounded interval [a, b) is compact.
Proof. Let °1L = {U,},,, be an open cover of [a, b] and let
E = {r e [a, b] : [a, r] is covered by a finite number of the sets U,}.
The set E is bounded above by b, and since a E U. for some a E A, E is nonempty.
Thus by the least upper bound property, the supremum of E exists in R. Let c = sup E.
Since b is an upper bound of E, c < b.
We first show that c E E; i.e., [a, c] is covered by a finite number of sets in 91.
Since c E [a, b], c E Up for some (3 E A. Since Up is open, there exists e > 0 such
that (c - e, c + e) C Up. Furthermore, since c - e is not an upper bound of E, there
exists r E E such that c - e < r c. But then [a, r] is covered by a finite number,
s a y U.,,, ... , U , , of sets in °U. But then the finite collection (U.,, ... , U,, , Up} cov-
ers [a, c]. Therefore, c E E.
To conclude the proof, we show that c = b. Suppose c < b. If we choose s < b
such that c < s < c 4- e, then the collection {U,,, ... , U,,, Up} also covers [a. s].
Thus s E E, which contradicts that c = sup E. Hence c must equal b. Q
An alternative proof of the Heine-Borel theorem using the nested interval property
(2.3.3) is outlined in the exercises (Exercise 11). The statement of the Heine-Borel theo-
rem was initially due to Heine, a student of Weierstrass, who used the result implicitly in
106 Chapter3 Structure of Point Sets
the 1870s in his studies on continuous functions. The theorem was proved by Borel in
1894 for the case where the open cover was countable. For an arbitrary open cover the re-
sult was finally proved in 1904 by Henri Lebesgue (1875-1941). Using the Heine-Borel
theorem, we now prove the following characterization of the compact subsets of R.
3.2.10 THEOREM Let K be a nonempty compact subset of R. Then every sequence in K has
a convergent subsequence that converges to a point in K.
Proof. Exercise 7.
3.3 The Cantor Set 107
D EXERCISES 3.2
1. Let A = {;,:n = 1, 2,. . . }.
a. Show that the set A is not compact by constructing an open cover of A that does not have a finite subcoser
of A.
b. Prove directly (using the definition) that K = A U {0} is compact.
2. Suppose is a convergent sequence in R with lim p = p. Prove, using the definition. that the set
A = {p} U {p" n E N} is a compact subset of W.
:
3. Show that (0, 1 ) is not compact by constructing an open cover of (0. 1' that does not have a finite subcover.
4. Suppose A and B are compact subsets of R.
'a. Prove (using only the definition) that A U B is compact. b. Prove that A (1 B is compact.
5. 'Let K be a nonempty compact subset of R. Prove that sup K and inf K exist and are in K.
6. Let K be a compact subset of R. Use the Bolzano-Weierstrass theorem (2.4.10) to prove that every infinite subset
of K has a limit point in K.
7. Prove Theorem 3.2.10.
8. Prove that a compact set has only a finite number of isolated points.
9. Let K be a compact subset of R and p E R \ K.
a. Prove that there exist points a, h E K such that
Ja - p, = inf{lx - pI : .r E K} and jb - p = sup{jx - pI : x E K}.
b. Are either of the above still true if K is only closed'?
10. a. Prove that the intersection of an arbitrary collection of compact sets is compact.
b. Find a countable collection {K,} . , of compact subsets of R such that U -, K, is not compact.
11. This exercise outlines an alternative proof of the Heine-Borel theorem. Suppose [a. h: is not compact. Then there
exists an open cover 1t = of [a, h] such that no finite subcollection of 1l covers 'a. b'.We now proceed
to show that this leads to a contradiction. Divide [a, b] into two closed subintervals [a. "- * n] and [" t. b'. each
of length (b - a)/2. At least one of these, call it h, cannot be covered by a finite number of the U,. Repeating
this process, obtain a sequence of closed and bounded intervals satisfying
(a) (a, b] D /, D 1, D D/D (b) length of 1 = (b - a)/2", and
(e) for each n, 1" is not covered by a finite number of the U,,.
Now use Corollary 2.3.3 to obtain a contradiction.
J 1.2
f1rY
9
VF1}
9 3
2
3
`iu}
JZ ,
r
7
9
8
9
zs
Figure 3.3
Since each P, is a finite union of closed intervals, P. is closed and bounded, hence com-
pact. Furthermore, since P0 D P, D P2 D , by Theorem 3.2.7,
a
P= t5 P"
is is a nonempty compact subset of (0, 1 ]. The set P is called the Cantor ternary set. We
now consider some properties of the set P.
PROPERTY 2 P contains all the end points of the closed intervals {J",r}, n = 1,
2, . . , k = 1,2,. . ,2".
Proof. Let p E P and let e > 0 be given. Choose in E I 1 such that 1/3< e. Since
p E P P E J,".k for some k, 1 <_ k s 2"'. But
xk L L 1
Jm. k = 3,,,
3m
Since the length of = 1/3'" < E, Jm.k C N,(p). Thus both end points of Jm., are in
P fl Nf(p), and at least one of these is distinct from p. 0
'
=3+232+...2"-13"+..
x 2'?- 1x 2
3" 3 3
t
31
O
3
As a consequence of Property 4,
where 3 does not divide a. In this case, x has two expansions: a finite expansion
x=-+ a1
3
++3m
0
co 71
k.m+l
2
3k
Proof. Exercise 2.
As a consequence of Property 6 and Theorem 1.7.18,
PROPERTY 7 P is uncountable.
For each n, the set P. has only a finite number of end points. As a consequence, the
set of points of P that are end points of some open interval removed in the construction
is countable. Since P is uncountable, P contains points other than end points. By Exer-
cise 1(a) of Section 1.6, the ternary expansion of; is
1
= .020202
4
Thus E P. but 1 is not an end point of any closed interval J,",k.
;
Remark. By Property 4, the sum of the lengths of the intervals removed is 1. This
seems to imply that P is in some sense very "small." On the other hand, by Property 7.
P is uncountable, which seems to imply that P is "large:'
EXERCISES 3.3
1. Determine whether ; is in the Cantor set.
2. Prove Property 6 of the Cantor set.
3. Let 0 < a < 1. Construct a closed subset F of [0, 1 ] in a manner similar to the construction of the Cantor set
such that the sum of all the open intervals removed is equal to a.
4. *Prove that the Cantor set P is equivalent to [0, I ].
NOTES
Without a doubt, the most important concept of this chap- set of K has a limit point in K. The converse of this result
ter is compactness. The fact that every open cover of a is also true, not only in R (Theorem 3.2.9), but also in the
compact set has a finite subcover will be crucial in the more general setting of metric spaces. A proof of this im-
study of continuous functions, especially uniform conti- portant result is outlined in the miscellaneous exercises.
nuity. As we will see in many instances, the applications The notion of a metric and a metric space is moti-
of compactness depend on the ability to choose a finite vated by geometrical considerations in R and n-dimen-
subcover from a particular open cover. A good example of sional euclidean space R". Many of the results involving
this is the proof of Theorem 3.2.5. Other instances will the structure of point sets in R depend on the fact that the
occur later in the text. euclidean distance d(x, y) = Ix - yI between the points x
Since compactness is the most important concept, and y in R satisfy the following properties: d(x. y) z 0,
Theorems 3.2.8 and 3.2.9 are the two most important re- d(x, y) = 0 if and only if x = y, d(x, y) = d(y. x), and
sults. In the Heine-Borel theorem we proved that every
d(x, y) s d(x. z) + d(z, y)
closed and bounded interval is compact, whereas in the
Heine-Borel-Bolzano-Weierstrass theorem, we character- for all x, y, z E R. Using these properties as a motivation.
ized the compact subsets of R. In Theorem 3.2.6 we the notion of distance can be extended to sets other than
proved that if K is a compact set, then every infinite sub- P as follows.
Miscellaneous Exercises 111
DEFINITION Let X be a nonempty set. A real-valued of a metric. It allows one to prove that every e-neighbor-
function d defined on X X X satisfying hood of a point is an open set. Also, combined with the
property that d(p, q) = 0 if and only if p = q, it guaran-
(a) d(x, y) ? O for all x, y E X.
tees that the limit of a convergent sequence in X is unique.
(b) d(x, y) = 0 if and only if x = y, With the exception of Theorems 3.1.13 and 3.1.18, all of
(c) d(x, y) = d(y, x), the results of Section 3.1 are still valid for any metric
space (X, d). In fact, the proofs follow verbatim the proofs
(d) d(x, y) s d(x, z) + d(z, y) for all x, y, z E X,
given in the text. The proofs of Theorems 3.1.13 and
is called a metric or distance function on X. The set X 3.1.18, however, require properties of the real number
with metric d is called a metric space, and is denoted by system and thus cannot be expected to be valid in an ar-
(X, d). bitrary metric space.
As for the real line, if (X, d) is a metric space, a sub-
Inequality (d) is called the triangle inequality. If
set K of X is compact if every open cover of K has a finite
d is a metric on a set X, then as in Definition 2.1.6, if
subcover of K. Again, since the proofs of Theorems 3.2.5.
p E X and e > 0, the e-neighborhood N,(p) of p is de-
3.2.6, and 3.2.7 only require the definition of compactness.
fined as
these results are again valid for an arbitrary metric space. In
N,(p) = {x E X : d(p,x) < e). particular, if K is a compact subset of (X, d) then K is closed
Since the notions of limit point of a set, limit of a se- and bounded. In this setting, a subset E of X is bounded if
quence, and interior point of a set are all defined in terms there exists a positive constant M and p e X such that
of e-neighborhoods, all of these concepts can be defined d(p, x) M for all x E E. The converse, however, is false.
verbatim for an arbitrary metric space. As in R, if (X, d) For example, if we set p(x, y) _ Ix - yI/(1 + Ix - v ).
is a metric space, a subset 0 of X is open in (X, d) if and then p is a metric on R satisfying p(x, y) < I for all
only if every point of 0 is an interior point of O. It is im- x, y E R (see Miscellaneous Exercise 4). In particular.
portant to note, however, that different metrics on a given [0. oo) is a closed and bounded subset of (R, p) that is not
set result in different c-neighborhoods, and thus a set that compact. Thus the Heine-Borel-Bolzano-Weierstrass theo-
is open with respect to one metric may not be open with rem is not valid in an arbitrary metric space. This is to be ex-
respect to a different metric. If X is any nonempty subset pected since the least upper bound property of R was used
of lR, then d defined on X X X by d(x, y) = Ix - yI is a in part of the proof. However, there is one important part of
metric on X. In this setting, the concept of a subset U of this theorem that is still valid; namely, if a subset K of a met-
X being relatively open in X is consistent with U being an ric space (X, d) satisfies the Bolzano-Weierstrass property,
open subset of X. then K is compact. The proof of this very important result is
The triangle inequality (d) is crucial in the definition outlined in Miscellaneous Exercise 7.
MISCELLANEOUS EXERCISES
The first two exercises involve the geometric and euclidean metric structure of IB". For n > 2,
R" _ {(xt, .. , x") xi E R, i = 1, .. , n). For p = (pt, , p"), q = (qi, . q") in I2" and
c E R. define p + q = (pt + qt, .. , p" + q"), and cp = (cpt..... cp.).
Also. let 0 = (0, ... , 0). For p, q E R", the inner product of p and q, denoted (p, q), is definedas
(p,q)=ptq.+"+p"q"
1. Prove each of the following. For p, q. r E P8",
a. (p, p) a 0 with equality if and only if p = 0.
b. (p, q) _ (q, p).
112 Chapter3 Structure of Point Sets
d, (p,q)_ 1p:-q,1.
a. Prove that d, is a metric on R".
b. In R2 sketch the 1-neighborhood of 0.
c. Suppose that {pk} is a sequence in R", where for each k E N. pk = (pk,,, ... , pk."). Prove that the sequence
{pk} converges top = (p ... , p") if and only if lim pk.; = p; for all i = 1, ... , n.
d Prove that a sequence {pk} in R" converges to p E R' with respect to the metric d, if and only if it converges to
p with respect to the metric d2. Specifically, prove that lim d,(pk, p) = 0 if and only if Alimd2(pk, p) = 0.
4. a. If (X, d) is a metric space, prove that p(x, y) = d(x, v)/(I + d(x, y)) is also a metric on X.
b. Prove that a subset U of X is open in (X, d) if and only if it is open in (X, p).
5. If E is an uncountable subset of R. prove that some point of E is a limit point of E. (Hint: Use Exercise 17 of Sec-
tion 3. 1).
6. Let {D"} be a countable collection of dense open subsets of R. Prove that (1 , O" is dense in R.
The following exercise is designed to prove the converse of Theorem 3.2.6; namely, if K is a subset
of a metric space (X, d) having the properly that every infinite subset of K has a limit point in K.
then K is compact.
7. Let K be a subset of a metric space (X. d) that has the property that every infinite subset of K has a limit point in K.
a. Prove that there exists a countable subset D of K which is dense in K. (Hint: Fix n E N. Let p, E K be arbi-
trary. Choose p2 E K, if possible, such that d(p,, P2) ? Suppose pl.. . . , pi have been chosen. Choose
pj+,, if possible, such that d(pi, pj ) >_ in for all i = 1, ... , j. Use the assumption about K to prove that this
process must terminate after a finite number of steps. Let 91 denote this finite collection of points, and let
D = U "eN P". Prove that D is countable and dense in K.)
b. Let D be as in (a), and let U be an open subset of X such that u n K * 0. Prove that there exists p E D and
n E N such that N,t"(p) C U.
c. Using the result of (b), prove that for every open cover qt. of K, there exists a finite or countable collection
{ U")" C °U, such that K C U Un.
d. Prove that every countable open cover of K has a finite subcover. (Hint: If is a countable open cover of
K, for each n E N. let W. = U, U,. Prove that K C W. for some n E N. Assume that the result is false, and
obtain an infinite subset of K with no limit point in K, which is contradiction.)
Supplemental Reading 113
SUPPLEMENTAL READING
Asic, M. D. and Adamovic, D. D., "Limit points of se- Geissinger, Ladner, "Pythagoras and the Cauchy-
quences in metric spaces," Amer. Math. Monthly 77 Schwarz inequality," Amer. Math. Monthly 83
(1970), 613-616. (1976), 40-41.
Corazza, P., "Introduction to metric-preserving func- Kaplansky, Irving, Set Theory and Metric Spaces.
tions:' Amer. Math. Monthly 106 (1999). 309-323. Chelsea Publ. Co., New York, 1977.
Dubeau, Francis, "Cauchy-Bunyakowski-Schwarz in- Kraft, R. L., "A golden Cantor set." Amer. Math.
equality revisited," Amer. Math. Monthly 99 (1990), Monthly 105 (1998), 718-725.
419-421. Labarre, Jr., A. E., "Structure theorem for open sets of
Espelie, M. S. and Joseph, J. I-, "Compact subsets of the real numbers;' Amer. Math. Monthly 72 (1965),
Sorgenfrey line," Math. Mag. 49 (1976), 250-251. 1114.
Fleron, Julian F., "A note on the history of the Cantor set Nathanson, M. B., "Round metric spaces," Amer. Math.
and Cantor function;' Math. Mag. 67 (1994),136-140. Monthly 82 (1975), 738-741.
Limits and Continuity
4.1 Limit of a Function
4.2 Continuous Functions
4.3 Uniform Continuity
4.4 Monotone Functions and Discontinuities
The concept of limit dates back to the late seventeenth century and the work of Isaac
Newton (1642-1727) and Gottfried Leibniz (1646-1716). Both of these mathemati-
cians are given historical credit for inventing the differential and integral calculus. Al-
though the idea of limit occurs in Newton's work Philosophia Naturalis Principia
Mathematica of 1687, he never expressed the concept algebraically; rather, he used the
phrase "ultimate ratios of evanescent quantities" to describe the limit process involved
in computing the derivatives of functions.
The subject of limits lacked mathematical rigor until 1821 when Augustin-Louis
Cauchy (1789-1857) published his Cours d'Analyse in which he offered the following
definition of limit: "If the successive values attributed to the same variable approach
indefinitely a fixed value, such that finally they dijf°er from it by as little as desired, this
latter is called the limit of all the others." Even this statement does not resemble the
modem delta-epsilon version of limit given in Section 4.1. Although Cauchy gave a
strictly verbal definition of limit, he did use epsilons, deltas, and inequalities in his
proofs. For this reason, Cauchy is credited for putting calculus on the rigorous basis we
are familiar with today.
Based on the previous study of calculus, the student should have an intuitive notion
of what it means for a function to be continuous. This most likely compares to how
mathematicians of the eighteenth century perceived a continuous function; namely, one
that can be expressed by a single formula or equation involving a variable x. Mathe-
maticians of this period certainly accepted functions that failed to be continuous at a fi-
nite number of points. However, even they might have difficulty envisaging a function
115
116 Chapter 4 Limits and Continuity
that is continuous at every irrational number and discontinuous at every rational num-
ber in its domain. Such a function is given in Example 4.2.2(g). An example of an in-
creasing function having the same properties will also be given in Section 4 of this
chapter.
for all x * 0, our intuition again should tell us that f(x) approaches 0 as x approaches
0. This is indeed the case as we will see in Example 4.1.10(c).
We now make this idea of f(x) approaching a value L as x approaches a point p
precise. To make the definition meaningful, we must require that the point p be a limit
point of the domain of the function f.
The definition of the limit of a function can also be stated in terms of e and 6 neigh-
borhoods as follows: If E C R, f : E --- R, and p is a limit point of E. then
lim f (x) = L
XP
4.1 Limit of a Function 117
L+E
Lt
Remarks
(a) In the definition of limit, the choice of S for a given e may depend not only on e
and the function, but also on the point p. This will be illustrated in Example 4.1.2(g).
(b) If p is not a limit point of E, then for S sufficiently small, there do not exist any
x E E so that 0 < Ix - pI < S. Thus if p is an isolated point of E, the concept of the
limit of a function at p has no meaning.
(c) In the definition of limit, it is not required that p E E, only that p is a limit point of
E. Even if p E E, and f has a limit at p, we may very well have that
lf(X) *f(p).
This will be the case in Example 4.1.2(c).
(d) Let E C R and p a limit point of E. To show that a given function f does tlt have
a limit at p, we must show that for every L E R, there exists an e > 0, such that for
every S > 0, there exists an x E E with 0 < Ix - pI < S, for which
Lf(x)-Llae.
We will illustrate this in Example 4.1.2(e).
118 Chapter 4 Limits and Continuity
4.1.2 EXAMPLES
These limits are also expressed as limc = c, limx = p, and limx'' = p2.
Even though we may feel that these limits are obvious, theypstill have to be proved.
We illustrate the method of proof by using the definition to prove that lim h(x) = p2.
The proofs of the other two limits are left to the exercises (Exercise 2). Fpor x E E,
Ih(x) - p2I = Ix2 - p2I = Ix - pIIx + pi (IkI + Ipj)Ix - pI
If pl < 1,then kI < IpI + 1. Hence forallxEEwithIx - pl < 1,
Ih(x) - p2I < (2lpl + l)Ix - pI.
This last term will be less than e provided Ix - pl < e/(21 pl + 1). Thus given e > 0,
we choose 8 = min{ 1, E/(2 1 pI + 1)}. With this choice of 8, if x E E with 0 < Ix - pl
< S,we first of all have Ix - pl < 1, and therefore also
h(x)=
x+1-1
x
4.1 Limit of a Function 119
g(x), x s 2
x( x+1+1) x+I+1
From this last term we now conjecture that h(x) -+ as x --+0. By the above,
h(x) - 12 (
+1 121 - I2( I x +X+
l + 11)
x+1)(1+Vx- -+I) _ -x
2(V, +1+1)2 I -I2(V+
IxI
2( x+1+1)2
For x E Ewe have ( x + 1 + 1)2 > 1, and thus
Ih(x)I < 2I
Given e > 0, let S = e. ThenI for all x e E with 0 < jxI < S,
f(x) - 10, x E O,
x le a.
120 Chapter 4 Limits and Continuity
We will show that for this function, lim f(x) fails to exist for every p E R. Fix p E R.
Let L E R and let
e = max( IL - 11, ILI).
Suppose e = IL - 11. By Theorem 1.5.2. for any S > 0, there exists an x E 0 such
that 0 < Ip - xI < S. For such an x,
If(x)-LI=11-LI=e-
lf e = ILI, then by Exercise 6, Section 1.5, for any 8 > 0, there exists an irrational
number x with 0 < Ix - p l < S. Again, for such an x, V (x) - LI = e. Thus with
e as defined, for any 8 > 0, there exists an x with 0 < Ix - pI < S such that
V (x) - LI ? e. Since this works for every L E 6L lim f(x) does not exist.
(f) Let f : 68 -* l be defined by
0' x E Q,
AX) - IX, x Q.
Then lim f (x) = 0. Since L f (x)1 jx i for all x, given e > 0. any S. 0 < 8 e, will
work in the definition of the limit. A modification of the argument given in (e) shows
that for any p 0 0, iim f (x) does not exist. An alternative proof will be provided in
Example 4.1.5(b).
(g) Our final example shows dramatically how the choice of S will generally depend
not only on e, but also on the point p. Let E _ (0, oo) and let f : E - R be defined by
f(x) I
x -P < p2Sse.
The S as defined depends on both p and e. This suggests that any S that works for
a given p and a must depend on both p and e. Suppose on the contrary that for a given
e > 0, the choice of 8 is independent of p E (0, oo). Then with e = 1, there exists a
S > 0 such that
z- p1<1
I
4.1 Limit of a Function 121
for all x, p E (0, oo) with 0 < Ix - p I < S. Since any smaller S must also work, we can
assume that 0 < S < 1. But now if we take p = 18 and x = 8, then 0 < Ix - pI < S,
and thus I f(x) - f(p) I < 1. However,
This contradiction proves that the choice of S must depend on both p and e. 0
4.1.3 THEOREM Let E be a subset of l1, p a limit point of E, and f a real-valued function
defined on E. Then
lim f (x) = L if and only if lint f (pn) = L
X-+p n-+oc
V (x)
(x) - L I < e for all'x E E, 0 < Ix - p I < S. (1)
Since lint pn = p, for the above S, there exists a positive integer no such that
m
0 < Ipn - P I < S for all n ? no.
Thus if n ? n by (1), l f (pn) - LI < e. Therefore, lint f (pn) = L.
R-00
Conversely, suppose f(pn) -4 L for every sequence { pn} in E with pn * p for all
a and pn -+ p. Suppose lim f (x) . # L. Then there exists an e > 0 such that for every
5 > 0, there exists an
x-Px E Ewith 0<Ix-p1 <8andlf(x)-LIae.Foreach
n r= N, take S = ,-,. Then for each n, there exists p,, E E such that
4.1.5 EXAMPLES
(a) Let E = (0, oo) and f(x) = sin (11x), x E E. We use the previous theorem to show
that
lim sin
x-.0
-
1
Thus ii f (p,) does not exist, and consequently by Theorem 4.1.3. liim f (x) also does
"7
not exist. The graph of f(x) = sin(1/x) is given in Figure 4.3.
2rr
Limit Theorems
4.1.6 THEOREM Suppose E C R, f. g : E -+ P. and p is a limit point of E. If
limf(x) = A and x-4p
lim g(x) = B,
then
Proof. For (a) and (b) apply Theorem 4.1.3 and Theorem 2.2.1. We leave the details
to the exercises (Exercise 11).
Proof of (c): By (b) it suffices to show that
1 1
lim g(x)
=B
We first show that since B # 0, g(x) * 0 for all x sufficiently close to p, x * p. Take
e = IBI/2. Then by the definition of limit, there exists a St > 0 such that
11
2
Ig(x) - BI <
for all x E E, 0 < Ix - pI < SI. By Corollary 2.1.4, Ig(x) - BI > Ilg(x)I - IBII.
Thus
the above S1, there exists an no E Rl such that 0 < jp" - pl < S1 for all n ? no. Thus
g(p") * 0 for all n ? n0. Therefore by Theorem 2.2.1(c),
lim
"°° g(p")
1
= -
1
B.
Since this holds for every sequence p" -> p, by Theorem 4.1.3,
lmg(x) = 1
B.
L1
The proofs of the following two theorems are easy consequences of Theorem 4.1.3
and the corresponding theorems for sequences (2.2.3 and 2.2.4). First, however, we give
the following definition.
lim f(x)g(x) = 0.
4.1.10 EXAMPLES
(a) By Example 4.1.2(a), lim x = c. Thus, using mathematical induction and Theorem
4.1.6(b), x.+c
lim x" = c" for all n E N. If p(x) is a polynomial !unction of degree n, that is,
.+a,x+ao,
where n is a n o n n e g a t i v e i n t e g e r and a 0 ,
. .. , a" E R with a" * 0, then a repeated ap-
plication of Theorem 4.1.6(a) gives lim p(x) = p(c).
(b) Consider
lim
x'+2x2-2x-4
x-.-2 x2 - 4
4A Limit of a Function 125
By part (a), xlim2 (x' + 2x2 - 2x - 4) = 0 and dint (x2 - 4) = 0. Since the denomi-
nator has limit zero, Theorem 4.1.6(c) does not apply. In this example, however, for
x # -2,
x3+2x2-2x-4 (x+2)(x2-2) x' -2
x2-4 (x+2)(x-2) x-2
Since slim (x - 2) = -4, which is nonzero, we can now apply Theorem 4.1.6(c) to
conclude tat
lim
x3+2x2-2x-4= lim x2-21
x2-4 x--2x-2 2
(c) Let E = R \ {0}, and let f : E -r R be defined by
f (x) = x sin x.
Since I sin (l /x) I <- I for all x E i8, x # 0, and Ii m x = 0, by Theorem 4.1.8
lim x sin 1 = 0.
x-w x
The graph of f(x) = x sin(1/x) is given in Figure 4.4.
-0.2
lim
r-0 -smt= 1.
1
126 Chapter 4 Limits and Continuity
As we will see in the next chapter, this limit will be crucial in computing the derivative
of the sine function. From Figure 4.5, we have
Figure 4.5
2sintcost<2t<2tan:.
Therefore,
sin t < 1
cost <
t cost
Using the fact that lim cos t = 1 (Exercise 5), by Theorem 4.1.9 we obtain
sin t
li m
1
= I.
Limits at Infinity
Up to this point we have only considered limits at points p E N. We now extend the def-
inition to include limits at oo or -oo. The definition of the limit at oo is very similar to
lim f(n) where f : N , R; that is, f is a sequence in R.
4.1.11 DEFINITION Let f be a real-valued function such that Domf tl (a, oo) * ¢ for
every a E P. The function f has a limit at oo if there exists a number L E R such that
given e > 0, there exists a real number M for which
lf(x) - LI < e
4.1 Limit of a Function 127
for all x E Dom f fl (M, oo). (See Figure 4.6.) If this is the case, we write
lim (x) = L.
X-*00f
L+E
L
L-E
4.1.12 EXAMPLES
(a) As our first example, consider the function f(x) = (sin x)/x defined on (0, oo).
Since Isin xl 1,
U(x)I x
for all x E (0, oo). Let e > 0 be given. Then with M = 1/e,
I f (x)I < e for all x > M.
Therefore, 1 (sin x)/x = 0.
128 Chapter 4 Limits and Continuity
(b) For our second example consider f(x) = x sin irx. If we set p" = (n + ;). n E N.
then
f(p") = (n + !)sin(n + ;)ir = (-1)"(n + 2).
Thus the sequence {f(p")}n°_ I is unbounded, and as a consequence, lim x sin zrx does
not exist. x
EXERCISES 4.1
1. Use the definition to establish each of the following limits.
*a. lim (2x - 7) = -3 b6 fim2 (3x + 5) = -1
x _ I
*c. I m d. lim 2x2-3x-4=
l+ x 2 x-+-i
i x3-2x-4 = 5
e..-- I x + 1 - 3 L lim
x2-4 2
2. Use the definition to establish each of the following limits.
a. lim c=C b. lim x = p
x-v x-+v
C. lim x3 = p3 d. lim
x-0 x" = p", n E RI
-v
p>0 x +o x
I
, p>0
v 2Vp
3. For each of the following, determine whether the indicated limit exists in R. Justify your answer!
limWx-
*a x-.° limx2x +- I
b.x-.i I
4. *Define f : (-1, 1) R by
f(x)=x2-x-2
x+l
Determine the limit L of f at -1 and prove, using e and S, that f has limit L at -1.
5. *a Using Figure 4.5, prove that I sin hl -- jhI for all It E R.
b. Using the trigonometric identity I - cos It = 2 sin 2 2, prove that
(i) limo cos h = I.
1 - cosh
00lim h 0.
6. Let E C R, p a limit point of E, and f : E -* R. Suppose there exist a constant M > 0 and L E R such that
[f(x) - LI c Mix - pl for all x E E. Prove that lim
X-P
f(x) = L.
4.1 Limit of a Function 129
Ix+21
'1
1
s
9. 'Suppose f : (a, b) -+ R, p E [a, b], and + m f (x) > 0. Prove that there exists a 8 > 0 such that f (x) > 0 for all
x E (a, b) with 0 < Ix - pi < S.
10. Suppose E C R, p is a limit point of E, and f : E -+ R. Prove that if f has a limit at p, then there exists a posi-
tive constant M and a S > 0, such that [f(x)I s M for all x E E, 0 < Ix - pi < 6.
11. a. Prove Theorem 4.1.6(a).
b. Prove Theorem 4.1.6(b).
12. 'Prove Theorem.4.1.8.
13. Prove Theorem 4.1.9.
14. Let f, g be real-valued functions defined on E C R and let p be a limit point of E.
a. If limf(x)
xrp
and lim
x-p
(f(x) + g(x)) exist, prove that lim
x-p
g(x) exists.
b. If lim f(x) and lim (f(x)g(x)) exist, does it follow that lim g(x) exists?
x-ap X-P X-P
15. Let E be a nonempty subset of R and let p be a limit point of E. Suppose f is a bounded real-valued function on
E having the property that iiin f(x) does not exist. Prove that there exist sequences (p.) and {q"} in E with
lim p" = lim q" = p such that lim f(p") and lim f(q") exist, but are not equal.
.-CO "-W
16. *Let f be a real-valued function defined on (a, oo) for some a > 0. Define g on by g(t) Prove that
lim f (x) = L if and only if lint g(t) = L.
17. Investigate the limits at oo of each of the following functions defined on (0, oo).
3x2
a. f(x) =
2+ +x ! 1
b. f (x) =
1 + x2
V-47+ 1 d. f(x) - 2x + 3
c. f(x) = X x+l
e. f( x) = AX) = V x - 2x
2NA + 3x
g. f (x) = x cos x h. f(x) = x sin
18. Let f : (a, oo) R be such that lim x f (x) = L where L E R. Prove that lim f (x) = 0.
130 Chapter4 Limits and Continuity
19. Let f : R -> R satisfy f(x + y) = f(x) + f(y) for all.r. v E R. If lim f(.r) exists, prove that
a. lim f(x) = 0, and b. lim f(x) exists for every p E R.
=-v
Remarks
(a) If p E E is a limit point of E, then f is continuous at p if and only if
Lmf(x) =f(p).
X-P
Figure 4.7
4.2.2 EXAMPLES
x#2,
g(x)=
2, x=2.
At the point p = 2, lim g(x) = 4 # g(2). Thus g is not continuous at p = 2. However,
if we redefine g at p = 2 so that g(2) = 4, then this function is now continuous at
p=2.
{0. x E O,
(b1 Let f be as defined in Example 4 . 1 . 2(f ): i . e., f (x) =
x x Q
132 Chapter4 Lindta and Continuity
Since
limf(x) = 0 = f(0),
f is continuous at p = 0. On the other hand, since limf(x)
-P fails to exist for every
p * 0, f is discontinuous at every p E Il, p * 0.
(e) The function f defined by
f(x)
- 11, xE0,
0, x VE Q,
0, if x is irrational,
AX) m
= in1, if x is rational with x =
n
in lowest terms.
The graph of f, at least for a few rational numbers, is given in Figure 4.8.
To establish our claim we will show that
lira f(x) = 0
X-P
4.2 Continuous Functions 133
for every p E (0, 1). As a consequence, since f(p) = 0 for every irrational number
p E (0, 1),f is continuous at every irrational number. Also, since f(p) # 0 when
p r= 0 fl (0, 1), f is discontinuous at every rational number in (0, 1).
8
16
- 1 I 3 I 5 3 7 1 9 S 11 3 13 '7 U
16 8 16 4 16 8 16 2 16 8 16 4 16 8 16
Figure 4.8
Fix p E (0, 1) and let e > 0 be given. To prove that lim f (x) = 0 we need to show
X __*P
that there exists a 8> 0 such that
lf(x)I < E
for all x E Na(p) 11(0, 1), x * p. This is certainly the case for any irrational number x.
On the other hand, if x is rational with x = m/n (in lowest terms), then f (x) = I/n.
Choose n0 E 1\J such that 1/n,, < E. There exist only a finite number of rational num-
bers m/n (in lowest terms) in (0, 1) with denominator less than no. Denote these by
rl, ... , rk, and let
8=min{Ir1-pI:i= 1,.. ,k,r,*p}.
(Note: Since p may be a rational number and thus possibly equal to r, for some
i = 1, ... , k, we take the minimum of fir, - pI) only for those i for which r, # p).
Thus 8 > 0, and if r E 0 fl Na(p) fl (0, 1), r * p, with r = m/n in lowest terms, then
n ? n0. Therefore,
if(r)I < E.
\g/(x) g(x)*
More generally, if f and g are real-valued functions defined on a set E. the quotient f/g
can always be defined on E, = {x E E : g(x) * 0}.
As an application of Theorem 4.1.6 we prove that continuity is preserved under the
algebraic operations defined above. The proof that Ifl is continuous whenever f is con-
tinuous is left as an exercise (Exercise 6).
4.2.5 EXAMPLES
(a) If p is a polynomial function on R, then by Example 4.1.10(a), lim p(x) = p(c) for
every c E R. Thus every polynomial function p is continuous on i$.`~`
4.2 Continuous Functions 135
(b) Suppose p and q are polynomials on P and E _ {x E R : q(x) = 01. Then by The-
orem 4.2.3 the rational function r defined on H \ E by
q(x)
r(x)=p(-,
xER\E,
is continuous on H \ E.
(c) By Example 4.2.2(f), f(x) = sin x is continuous on P. Hence if p is a polynomial
function on R, by Theorem 4.2.4, (f o p)(x) = sin(p(x)) is also continuous on R.
4.2.7 EXAMPLES
(a) We illustrate the previous theorem for the function f(x) = \, Dom f = [0, oo).
Suppose first that V is an open interval (a, b) with a < b. Then
1
¢, b < 0,
f-,(V)= [O,b2), as0<b,
(a2, b2), 0 < a.
Clearly 46 and (a2, b2) are open subsets of H and hence also of [0, oo). Although [0, b2)
is not open in R,
[0, b2) = (-b2, b2) fl [0, oo).
Thus by Theorem 3.1.16, [0, b2) is open in [0, oo). If V is an arbitrary open subset of
R, then by Theorem 3.1.13 (or Exercise 17, Section 3.1) V = U 1,,, where is a fi-
136 Chapter 4 Limits and Continuity
X2 xx> I.
Figure 4.9 Graph of f(x) _
13-2x, >I
Proof. Let { V,},EA bean open cover of f(K). Since f is continuous on Kj -'(V.) is
open in K for every a E A. By Theorem 3.1.16, for each a there exists an open subset
U. of l such that
f -'(V,) = K n U,.
We claim that {U,}0EA is an open cover of K. If p E K, then f(p) Ef(K) and thus
f(p) E V. for some a E A. But then p is in j-'(V.) and hence also in U,. Since each
U. is also open, the collection { U«}.eA is an open cover of K. Since K is compact, there
exists a,,. . . , an E A such that
K C ,U U..
I
Therefore,
n n
K =U
j=t
(u.fl K) = Utf
j- -'(V,),
and by Theorem 1.7.14(a),
n
f(K) = Uf(r'(V0,))
Since f (f -'(V.,)) C V,, f (K) C U; _ IV.,. Thus f (K) is compact.
As a corollary of the previous theorem we obtain the following generalization of
the usual maximum-minimum theorem encountered in calculus.
4.2.10 EXAMPLES
(a) Suppose E is a subset of R. If E is unbounded, consider
f(x) = x2
I +x"
Then f is continuous on E, sup { f (x) : x E E} = 1, but f(x) < I for all x e E. To see
that the supremum is 1, we note that since E is unbounded, there exists a sequence {xn}
in E such that x2. --* oo as n -+ oo. But then
Thus if 0 < (3 < 1, there exists an integer n such that f (x.) > p. Hence sup{ f (x) :
x E E} = 1. The graph of f with E = (0, oo) is given in Figure 4.10.
4
1 2 3 4 5 6
E=(0.-)
x2
Figure 4.10 Graph of f(x) = x E [0,oo)
+z2
0.5
0 E=[0, I) I
4.2.13 COROLLARY For every real number y > 0 and every positive integer n, there exists
a unique positive real number y so that y° = y.
4.2 Continuous Functions 141
Proof. That y is unique is clear. Let f(x) = x", which by Example 4.2.5(a) is contin-
uous on R. Let a = 0 and b = y + 1. Since (y + 1)" > y, f satisfies the hypothesis of
Theorem 4.2.11. Thus there exists y, 0 < y < y + 1, such that
f(y) = y" = y
4.2.14 COROLLARY If f : [0, 1 ] -* [0, 1 ] is continuous, then there exists y E [0, 1 ] such
that Ay) = y.
Proof. Let g(x) = f(x) - x. Then g(0) = f(0) ? 0 and g(1) = f(1) - I 0. Thus
there exists y E [0, 11 such that g(y) = 0; i.e., f(y) = y. Q
4.2.15 EXAMPLES
(a) In the proof of Theorem 4.2.11 continuity of the function f was required. The fol-
lowing example shows that the converse of Theorem 4.2.11 is false; that is, if a func-
tion f satisfies the intermediate value property on an interval [a, b], this does not im-
ply that f is continuous on [a, b]. Let f be defined on [0, 2] as follows-
f(x) =
1
x
, 0<xs?, or,
- -1, x = 0.
Then f(0) _ -1, f(r) = 1, and for every y, -1 < y < 1, there exists an x E (0,
such that f(x) = y. However, the function f is not continuous at x = 0 (see Figure 4.3).
(b) In this example we show that the conclusion of the intermediate value theorem is
false if the interval [a, b] of real numbers is replaced by an interval of rational numbers.
Let E _ {x E 0 : 0 5 x < 2}, and let f (x) = x2. Then f is continuous on E with
f(O) < 2 < f(2). However, there does not exist r E E such that f(r) = 2. 0
EXERCISES 4.2
1. For each of the following, determine whether the given function is continuous at the indicated point x".
2x2-5x-3 x03, _ Vc-2 x#4,
f(x) h(x)
at x" = 3 x 4 x at x" = 4
o x 6, 3 x= 3, 4 4
1- cos x
`c. g(x) =
x# 0, d. k(x)
- J x2, x 5 2,
at x, =.2
x at x° = 0 1 4 - x, x>2.
0, x = 0,
2. Let f : R -' R be defined by
f(x) _ 8x, when x is rational,
2x2 + 8, when x is irrational.
a. Prove, using e and 8, that f is continuous at 2.
b. Is f continuous at 1? Justify your answer.
142 Chapter 4 Limits and Continuity
3. Let f : R -r R be defined by
f(x) __ x2, x E Q,
x+2, x0Q.
Finds all points (if any) where f is continuous.
4. *Prove (without using Example 4.2.7) that f(x) = V is continuous on [0, oo).
5. Define f : (0, 1 ] - R by
f x+l
x
a. Justify that f is continuous on (0, 1].
b, Can one define f(0) so that f is continuous on [0, 1]?
6. Let E C R, and suppose f : E --+ R is continuous at p e E.
*a. Prove that Ifl is continuous at p. Is the converse true?
b. Set g(x) =, x E E. Prove that g is continuous at p.
7. *Let E C R and suppose f : E-*R is continuous on E. Prove that f" defined by f'(x) _ (f(x))" is continuous on
E for each n E N.
8. a. Prove that f (x) = cos x is continuous on R.
b. If E C R and f : E -+ R is continuous on E, prove that g(x) = cos (fix)) is continuous on E.
9. For each of the following equations, determine the largest subset E of R such that the given equation defines a
continuous function on E. In each case state which theorems or examples are used to show that the function is
continuous on E.
*a. AX) =
x 4x
b. g(x) = sin-
4)5
z
*c. d. k(x) =cos
h(x) x2 + sin x
10. Prove that
x2, x s 1,
f(x) _ { 32x, x> 1,
is continuous on R and that Range f = R.
11. As in Example 4.2.7(a), use Theorem 4.2.6 to prove that each of the following functions is continuous on the
given domain.
1,
a. f(x) = Dom f = (0, oo) b, g(x) = x2, Dom g = R
X
12. Suppose E is a subset of R and f, g : E -+ R are continuous at p E E. Prove that each of the functions defined
below is continuous at p.
*a. max{f, g}(x) = max{f(x), g(x)}, x E E
b. min{f, g}(x) = min{f(x), g(x)}, x E E
c. f'(x) = max{f(x), 0), x E E
13. Suppose f : [a, b] -* R and g : [b, c] - R are continuous on [a, b] and [b. c], respectively. Leth : [a. c] -+R be
defined by
h(x) _ {f
(x a s x s b,
g(x). b < x s c.
Prove that h is continuous on [a, c] if and only if f(b) = g(b).
4.2 Continuous Functions 143
14. Use the intermediate value theorem to prove that every polynomial of odd degree has at least one real zero.
15. Prove that there exists x E (0, z) such that cos x = x.
16. *Suppose f : [ -1, 1 ] -+ R is continuous and satisfies f(- I) = f(l). Prove that there exists y E [0, 17 such that
f(Y) = f(Y - 1)
17. Suppose f : [0, 1 ] --? R is continuous and satisfies f(0) = f(l). Prove that there exists y E (0, ;] such that
ICY) = f(y + D.
18. *Let E C R and let f : E -a R be continuous. Let F = {x E E : f(x) = 0}. Prove that F is closed in E. Is F neces-
sarily closed in R?
19. Suppose f : (0, 1) -- R is continuous and satisfies f(r) = 0 for each rational number r E (0, 1). Prove that
f(x) = 0 for all x E (0, 1).
20. Let f be a real-valued function on R satisfying f(x + y) f(x) + f(y) for all x, y E R.
a. If f is continuous at some p E R, prove that f is continuous at every point of R.
*b. If f is continuous on R, prove that f(x) = cx for all x E R, where c = f(1).
21. Let E C R and let f be a real-valued function on E that is continuous at p E E. If f(p) > 0, prove that there ex-
ists an a > 0 and a S > 0 such that f(x) 2- a for all x e N$(p) fl E.
22. *Let f : E.- R be continuous at p E E. Prove that there exists a positive constant M and S > 0 such that
lf(x)I sMforallxEEflN6(p).
23. Let f : (0, 1) --> R be defined by
f(x) _ 0, if x is irrational,
- In, if x is rational with x=m/n in lowest terms.
a. Prove that f is unbounded on every open interval I C (0, 1).
b. Use part (a) and the previous exercise to conclude that f is discontinuous at every point of (0, 1).
24. Suppose E is a subset of R and f, g : E -- R are continuous on E. Show that {x E E : f(x) > g(x)} is open in E.
25. *Let K be a compact subset of R and let f : K -a R be continuous on K. Prove that f(K) is compact by showing
that f(K) is closed and bounded.
26. Let E C R and let f be a real-valued function on E. Prove that f is continuous on E if and only iff-'(F) is
closed in E for every closed subset F of R.
27. Let A, B C R and let f : A -- ). R and g : B -a R be functions such that Range f C B.
a. If V C R, prove that (g -f)-'(V) = f-'(g-'(V)).
b. If f and g are continuous on A and B respectively, use Theorem 4.2.6 to prove that g of is continuous on A.
28. Suppose 1 is a connected subset of R and f : I--> R is continuous on I. Prove, using only the properties of continu-
ity and the definition of connected set, that f(1) is connected.
29. *Let K C R be compact and let f be a real-valued function on K. Suppose that for each x E K there exists
e, > 0 such that f is bounded on NE,(x) n K. Prove that f is bounded on K.
30. Let A C R. For p E R, the distance from p to the set A, denoted d(p, A), is defined by
d(p, A) = inf{Ip - xJ : x E A}.
a. Prove that d(p, A) = 0 if and only if p E A.
b. For x, y E R, prove that Id(x, A) - d(y, A) Jx - y1.
e. Prove that the function x -+ d(x, A) is continuous on R.
d. If A, B are disjoint closed subsets of R, prove that
d(x, A)
AX)
(x)
= d(x, A) + d(x, B)
144 Chapter4 Limits and Continuity
Uniform Continuity
In the previous section we discussed continuity of a function at a point and on a set. By
Definition 4.2.1, a function f : E -i R is continuous on E if for each p E E, given any
e > 0, there exists a S > 0 such that Lf(x) - f(p)I < e for all x E E fl N5(p). In gen-
eral, for a given e > 0, the choice of S that works depends not only on a and the func-
tion f, but also on the point p. This was illustrated in Example 4.1.2(g) for the function
f(x) = 1/x, x E (0, oo). Functions for which a choice of S independent of p is possible
are given a special name.
4.3.2 EXAMPLES
(a) If E is a bounded subset of R, then f(x) = .r` is uniformly continuous on E. Since
E is bounded, there exists a positive constant C so that 1x1 C for all x E E. If
x, y E E, then
Lf(x) - f(v)I = 1x2 - y2I = Ix + yI Ix - v1
(Ix1 + yI)Ix - yI 2CIx - v1.
Let e > 0 be given. Take S = e/2C. If x, y E E with Ix - yI < S, then by the above
Lf(x) -f(y)1 s 2CIx - yI < 2C8 =
Therefore, f is uniformly continuous on E. In this example, the choice of S depends
both on e and the set E. In the exercises you will be asked to show that this result is
false if the set E is an unbounded interval.
(b) Let f(x) = sinx. As in Example 4.2.2(f),
Lf(y) -f(x)I < Iy - xI
for all x, y E R. Consequently, f is uniformly continuous on R.
4.3 Uniform Continuity 145
(c) In this example we show that the function f(x) = I /.r, x E (0. oo) is r= uniformly
continuous on (0, oo). Suppose, on the contrary, that f is uniformly continuous on
(0, oo). Then, as in Example 4.1.2(g), if we take e = 1, there exists a S > 0 such that
X. - X.+1 =I
n-n+1 < n <
but
Lipschitz Functions
Both of the functions in Example 4.3.2(a) and (b), and the function f(x) = 1/x with
Dom f = [a, oo), a > 0, are examples of an extensive class of functions. If E C R, a
function f : E -+ R satisfies a Lipschitz condition on E if there exists a positive con-
stant M such that
[1(x) - f(Y) I < Mix - Y I
for all x, y E E. Functions satisfying the above inequality are usually referred to as Lip-
schitz functions. As we will see in the next chapter, functions for which the derivative
is bounded are Lipschitz functions.. As a consequence of the following theorem, every
Lipschitz function is uniformly continuous. However, not every uniformly continuous
function is a Lipschitz function. For example, the function f(x) = V is uniformly
continuous on (0, oo), but f does not satisfy a Lipschitz condition on [0, oo) (see Ex-
ercise 5).
Proof. Exercise 1. O
K C U Ns,(pi).
i=[
Let
S = min{SP, : i = 1, , n).
Then S > 0. Suppose x, y E K with Ix - yi < S. Since x E K, x E N,&(pi) for some
i. Furthermore, since k - yI < S s Sp,
x, y E Nzs,.(Pi)
4.3.6 EXAMPLE In this example, we show that the properties closed and bounded are both
required in Corollary 4.3.5. The interval [0, oo) is closed, but not bounded. The func-
tion f(x) = x2 is continuous on (0, oo), but not uniformly continuous on [0, co) (Exer-
cise 2). On the other hand, the interval (0, 1) is bounded, but not closed. The function
f(x) = 1/x is continuous on (0, 1), but is not uniformly continuous on (0, 1).
4.3 Uniform Continuity 147
EXERCISES 4.3
1. Prove Theorem 4.3.3.
2. Show that the following functions are not uniformly continuous on the given domain.
I. c. h(x) = sin x, Dom h = (0, co)
*a. AX) = x2, Dom f = [0, oo) b, g(x) = Dom g = (0, oo)
3. Prove that each of the following functions is uniformly continuous on the indicated set.
*a. f(x) = , Dom f = [a, oo), a > 0 b. g(x) = x2 + 1, Dom g = (0. oo)
c. h(x) = sin x, Dom h = [a, oo), a > 0 d. p(x) a polynomial, Dom p = [ -a, a], a > 0
5. *a. Show that f(x) _ \ satisfies a Lipschitz condition on [a, oo), a > 0.
b. Prove that V is uniformly continuous on (0, oo).
c. Show that f does not satisfy a Lipschitz condition on (0, oo).
6. Suppose E C R and f, g are Lipschitz functions on E.
a. Prove that f + g is a Lipschitz function on E.
b. If in addition f and g are bounded on E, or the set E is compact, prove that fg is a Lipschitz function on E.
7. Suppose E C R and f. g are uniformly continuous real-valued functions on E.
a. Prove that f + g is uniformly continuous on E.
*b. If, in addition, f and g are bounded, prove that fg is uniformly continuous on E.
c. Is part (b) still true if only one of the two functions is bounded?
8. Suppose E C R and f : E - R is uniformly continuous. If is a Cauchy sequence in E, prove that { f(x )} is a
Cauchy sequence.
9. Let f : (a, b) -+ R be uniformly continuous on (a, b). Use the previous exercise to show that f can be defined at a
and b such that f is continuous on [a, b].
10. Suppose that E is a bounded subset of R and f : E - R is uniformly continuous on E. Prove that f is bounded
on E.
11. Suppose -oo s a < c < b <_ oo, and suppose f : (a, b) - R is continuous on (a. b).
a. If f is uniformly continuous on (a, c) and also on (c, b), prove that f is uniformly continuous on (a, b).
b. Show by example that the conclusion in part (a) may be false if f is not continuous on (a, b).
12. Let a E R. Suppose f is a continuous real-valued function on [a, oo) satisfying lim f(x) = L. where L E R.
Prove that
a, f is bounded on [a, oo), and
b. f is uniformly continuous on (a, oo).
148 Chapter 4 Limits and Continuity
13. Let E C R. A function f : E E is contractive if there exists a constant b, 0 < b < I. such that
If(x) - f(y)I b Ix - yI.
*a. If E is closed, and f : E -* E is contractive, prove that there exists a unique point x E E such that x,,.
(Such a point xo is called a fixed point of f.)
b. Let E _ (0,;]. Show that f(x) = r' is contractive on E. but that I 'does not have a fixed point in E.
14. A function f : Il -> R is periodic if there exists p E R such that f(x + p) = f(x) for all x E R. Prove that a con-
tinuous periodic function on R is bounded and uniformly continuous on R.
Similarly, if p is a limit point of En( - oo, p), the left limit off at p, if it exists, is de-
noted by f( p- ), and we write
f(p-) = lim.f(x) = limf(x).
x<p
The hypothesis that p is a limit point of E fl (p, oo) guarantees that for every
S > 0, E n (p, p + S) * 4i. If E is an open interval (a, b), -oo < a < b s oo, then
any p satisfying a s p < b is a limit point of E fl (p, oo). Similarly, if -oo <- a
< b < oo, then any p satisfying a < p s b is a limit point of (-oo, p) fl E. If 1 is any
interval with Int(!) # 0, and f : I -> R, then f has a limit at p E lnt(I) if and only if
(a) f (p+) and f (p-) both exist, and
(b) f(p+) =f(p-).
The hypothesis that p E Int(!) guarantees that p is a limit point of both (-oo, p) fl 1
and I fl (p, oo). If p is the left endpoint of the interval 1, then the right limit of f at p
4.4 Monotone Functions and Discontinuities 149
coincides with the limit off at p. The analogous statement is also true if p is the right
endpoint of I.
We also define right and left continuity of a function at a point p as follows.
The following theorem, the proof of which is left to the exercises, is an immediate
consequence of the definitions.
4A.3 THEOREM A function f : (a, b) -+R is right continuous at p E (a, b) if and only if
f (p+) exists and equals f (p). Similarly, f is left continuous at p if and only if f (p- )
exists and equals f (p).
Proof. Exercise 1. Q
Types of Discontinuities
By the previous theorem a function f is continuous at p E (a, b) if and only if
(a) f (p+) and f (p-) both exist, and
(b) f(p+) =f(p-) =f(p).
A real-valued function f defined on an interval I can fail to be continuous at a point
p E 1 (the closure of I) for several reasons. One possibility is that lim f(x) exists but
xZ
either does not equal f (p), or f is not defined at p. Such a function can easily be made
continuous at p by either defining or redefining f at p as follows:
lim f(x).
f(p) = X-P
For this reason, such a discontinuity is called a removable discontinuity. For example,
the function
-4 x #.2,
g(x) x-2'
2, x = 2,
of Example 4.2.2(a) is not continuous at 2 since
1 m g(x) = 4 * g(2).
150 Chapter 4 Limits and Continuity
By redefining g such that g(2) = 4. the resulting function is then continuous at 2. An-
other example is given by f (x) = x sin(] /x), x E (0, oo), which is not defined at 0. If
we define f on [0, oo) by
0, x=0
f(x) i , x > 0,
x sin
x
then by Example 4.2.2(e), f is now continuous at 0.
Another possibility is that f(p+) and f(p-) both exist, but are not equal. This
type of discontinuity is called a jump discontinuity. (See Figure 4.13.)
Ap+)
AP-)
In case (a) f has a jump discontinuity at p, whereas in case (b) the discontinuity is re-
movable. All discontinuities for which f (p+) or f (p-) does not exist are discontinu-
ities of the second kind.
4.4.5 EXAMPLES
(a) Let f be defined by
x, 0<x-1,
f(x)= 3-x2, x> 1.
The graph off is given in Figure 4.14. If x < 1, then f(x) = x. Therefore,
f(I-)= lint f(x) = limx=
X +F X-1
1 =f(1).
Likewise, the right limit off at 1 is
Figure 4.14
f(1+)=Xlimf(x)=lim(3-x2)=2.
Therefore, f (I -) = f (l) = 1, and f (I +) = 2. Thus f is left continuous at 1, but not
continuous. Since both right and left limits exist at 1, but are not equal, the function f
has a jump discontinuity at 1.
(b) Let [x] denote the greatest Integer function; that is, for each x, [x] = largest
integer n that is less than or equal to x. For example, [2.9] = 2,[3.1) = 3, and
[ 1.5] = -2. The graph of y = [x] is given in Figure 4.15. It is clear that for each n E Z,
lim [x] = n - 1 and lim. [x] = n.
Thus f has a jump discontinuity at each n E Z. Also, since f (n) = [n] = n, f (x) = [x]
is right continuous at each integer. Finally, since f is constant on each interval
(n - 1, n), n E Z, f is continuous at every x E R\7L.
152 Chapter4 Limits and Continuity
34-
24-
I +
-1 1 2 3
00 -1
Figure 4.15 Graph of [x]
Then f(0-) = 0, but f(0+) does not exist. Thus the discontinuity is of second kind.
(d) Consider the function g : R -+ R defined by
g(x) = sin (21rx[x]).
For x E (n, n + 1), n E Z, x[x] = nx, and thus g(x) is continuous on every interval
(n, n + 1), n E Z. On the other hand, for n E 71,
lim sin(2ax[x]) = sin(21rn2) = 0, and
sn+
lim sin(2irx[x]) = sin(2an(n - 1)) = 0.
Monotone Functions
4.4.6 DEFINITION Let f be a real-valued function defined on an interval I.
(a) f is monotone increasing (increasing, nondecreasing) on I if f(x) < f(y)
for all x, y E I with x < y.
4.4 Monotone Functions and Discontinuities 153
A A
AAIA
AA
I IV TV-11- A V
The proof of this is similar to the proof of Theorem 2.3.2. Let e > 0 be given. Since A
is the least upper bound of {f (x) : x < p}, there exists xo < p such that
A-e<f(x,,)5A.
Thus if xo < x < p, A - e < f (x4,) <- f (x) s A. Therefore,
Lf(x) - AI < e
x+p
f(x) = A. Similarly,
f(p) f(p+) = inf{f (x) : p < x, x E l}.
154 Chapter 4 Limits and Continuity
4A.10 THEOREM Let a, b e R and let be a countable subset of (a, b). Let
be any sequence of positive real numbers such that 7,0,0. , c converges. Then there ex-
ists a monotone increasing function f on [a, b) such that
(a) f(a) = 0 and f(b) _ c,,,
Thus for each x E [a, b], the sequence {s (x)} of partial sums is monotone increasing
and bounded above, and hence by Theorem 2.7.6 converges. Since
n= 1,2,.. .
156 Chapter 4 Limits and Continuity
for all x, y with x < y, f is monotone increasing on (a, b). Furthermore, since x > a for
all n, 1(a - 0 for all n. Therefore f(a) = 0. Also, since 1(b - I for all n,
f(b) = 7, C.
k-l
Ck < E.
k-N+1
021(x-xk)-1(p-xk): 1 forallk6N.
Therefore, for p < x < p + S,
00
I00
kN+l
Ck < E.
4.4 Monotone Functions and Discontinuities 157
As in (b), there exists a S > 0 such that if xk. E (x,,, x,, + S) fl E, then k > N. Thus
a
0 s f(y) - 7, ck < e for all y E (x,, x + S).
k=N+l
I(xn-xn)=1(0)= 1.
Therefore,
Ax) _ 71
n-1
In this example, the sequence {xn} satisfies 0 < xt < x2 < < I. If 0 < x <
xt = 1, then I(x - 0 for all n. Thus
f(x)=0, xE [0'2).
If x1 x < x2 = 3, then !(x - x1) = 1 and !(x - xk) = 0 for all k =- 2. Therefore,
f(x)=c1=2, X
If x2 5 x < x3 = 4, then I(x - xk) = 1 for k = 1,2 and I(x - xk) = 0 fork a 3.
Therefore,
1 34, J23. 34
AX) +c2= xE ,
2+
and so forth. The graph off is depicted in Figure 4.18.
158 Chapter 4 Limits and Continuity
7
8
0-0
4
.-o
2
S 0
0 I 2 3 4
2 3 4 S
Figure 4.18
(b) Let c" = 2-", and let {x"} be an enumeration of the rationals in (0, 1). Theorem
4.4.10 guarantees the existence of a nondecreasing function on [0. lj, which is discon-
tinuous at each rational number in (0, 1). and continuous at every irrational number in
(0, 1).
(c) If in the proof of Theorem 4.4.10, we take to be a countable subset of R
and choose {c"}"EN (c" > 0) such that Z'n'o- , c" = 1, then we obtain a nondecreasing,
real-valued function f on R satisfying lie f(x) = 0 and lim f(x) = 1. (See Exercise
20.) Such a function is called a distribution function on d. Snch functions arise natu-
rally in probability theory.
Inverse Functions
Suppose f is a strictly increasing real-valued function on an interval I. Let x, y E I with
x 0 y. If x < y, then since f is strictly increasing. f(x) < f(y). Similarly, if x > y then
f(x) > f(y). Thus f(x) # f(y) for any jr, y E I with x # y. Therefore, f is one-to-one,
and consequently has an inverse function f -' defined on f(1). In the following theorem
we prove that if f is continuous on 1, then f -' is also continuous on f(1).
4.4.12 THEOREM Let I C R be an interval and let f : I-+ R be strictly monotone and
continuous on !. Then f -' is strictly monotone and continuous on J = f(1).
Proof. Without loss of generality, we consider the case where f is strictly increasing
on I. Since f is continuous, by Corollary 4.2.12, f(1) = J is an interval. Furthermore,
since f is strictly increasing on I, f is a one-to-one function from I onto J. Hence f
is a one-to-one function from J onto I.
Suppose y,, y2 E J with y, < y2. Then there exist distinct points x,, x2 E I such
that f(x;) = y;, i = 1, 2. Since f is strictly increasing, we have x, < x2. Thus f -'(y,) <
f -' (y2); i.e., f ` is strictly increasing.
4.4 Monotone Functions and Discontinuities 159
Figure 4.19
Therefore, f "' is left continuous at y,. A similar argument also proves that f -' is right
continuous at each y, E J that is not the right endpoint of J. Thus f -' is continuous at
each y, E J.
For a strictly increasing function f on an open interval I, an alternative proof of the
continuity of the inverse function f `' is suggested in Exercise 14.
4.4.13 EXAMPLE The function f(x) = x2 is continuous and strictly increasing on I = [0, oo)
with J = f(I) = [0, oo). Thus the inverse function f `'(y) = N /y is continuous on
(0, oo). As a consequence, the function g(x) = N/x is continuous on (0, oo). Applying
the same argument to f (x) = x" shows that the function g(x) = 'Clx is strictly increas-
ing and continuous on (0, oo). 0
160 Chapter 4 Limits and Continuity
Remark. In the statement of Theorem 4.4.12 we assumed that f was strictly mono-
tone and continuous on the interval 1. The fact that f is either strictly increasing or
strictly decreasing on I implies that f is one-to-one on the interval 1. Conversely, if f is
one-to-one and continuous on an interval 1. then as a consequence of the intermediate
value theorem the function f is strictly monotone on I (Exercise 15). This, however, is
false if either f is not continuous on the interval 1, or if Dom f is not an interval. (See
Exercises 16, 17, and 18.)
EXERCISES 4.4
1. Prove Theorem 4.4.3.
2. For each of the following functions f defined on R \ {0}, find ,4 m_ f(x) and -Ii m, f(x), provided the limits exist.
Ix[x]. X < 0,
a. f(x) = Ixl *b. f(x) _ [X] X>0
X
13. a. If f and g are monotone increasing functions on an interval I. prove that f + g is monotone increasing on I.
b. If in addition f and g are positive on 1, prove that fg is monotone increasing on I.
c. Show by example that the conclusion in part (b) may be false if f and g are not both positive on I.
14. Let I C R be an open interval and let f : I -+ R be strictly increasing and continuous on I.
a. If U C I is open, prove that f(U) is open.
b. Use (a) and Theorem 4.2.6 to prove that f'' is continuous on f(I).
13. Let I C R be an interval and let f be a one-to-one continuous real-valued function on 1. Prove that f is strictly
monotone on I.
16. Let f : [0,1 ) -' R be defined by
2x, Osx<2.
AX)
3-2x, ?5xsl.
a. Sketch the graph of f.
b. Show that f is one-to-one on [0, 1 ], but not strictly monotone on [0, 1 ].
c. Show that f([0.. I ]) = [0, 2).
d. Find f-' (y), y E [0, 2]. and show that f-' is not continuous at y, = 1.
17. Let E = [0, 11 U (2, 3), and for x E E set
f(x) _ C. !(x -
NOTES
The limit of a function at a point is one of the fundamen- a unique positive real number y such that y" = x. Even
tal tools of analysis. Not only is it crucial to continuity, though the existence of nth roots can be proved without
but also to many subsequent topics in the text. The limit the intermediate value theorem, any such proof is simply
process will occur over and over again. We will encounter the statement that the function f(x) = x" satisfies the in-
it in the next chapter in the definition of the derivative and termediate value property on [0, a] for every a > 0. Other
again in the chapters on integration, series, etc. applications of the intermediate value theorem will occur
Another very important concept that will be encoun- elsewhere in the text.
tered often in the text is uniform continuity. Uniform con- The proof of the intermediate value theorem de-
tinuity is important in that given e > 0, it guarantees the pended on the fact that the connected subsets of R are the
existence of a S > 0 such that If(x) - f(y) I < e for all intervals (Theorem 3.1.18) and that the continuous image
x, y E Dom f with Ix - yI < 6. In Chapter 6 we will use of a connected set is connected (Exercise 28 of Section
this to prove that every continuous real-valued function 4.2). Assuming these two results, the intermediate value
on [a, b], a, b E R, is Riemann integrable on [a, b]. theorem is an immediate consequence as follows: Sup-
Other applications of uniform continuity will occur in pose f is continuous on [a, b]. Let I = f([a, b]). Then I
many other theorems and in the exercises. is connected and thus must be an interval. Thus if y satis-
One of the most important results of this chapter is fies f(a) < -y < f(b), then y E I, and hence there exists
the intermediate value theorem (Theorem 4.2.11). The in- c E [a, b] such that f(c) = y. That the continuous image
termediate value theorem has already been used in Corol- of a connected set is connected follows from the defini-
lary 4.2.13 to prove the existence of nth roots; namely, for tion. However, the proof that the connected subsets of R
every positive real number x and n E 101, there exists are the intervals requires the least upper bound property.
MISCELLANEOUS EXERCISES
1. Let f be a continuous real-valued function on [a, b] with f(a) < 0 < f(b). Let c, = 2(a + b). If f(c,) > 0, let
c, = } (a + c,). If f(cl) < 0, let c2 = 1(c, + b). Continue this process inductively to obtain a sequence {c"} in
(a, b) which converges to a point c E (a, b) for which f(c) = 0.
2. Let E C R, p a limit point of E, and f a real-valued function defined on E. The limit superior of f at p, denoted
1 m f(x), is defined by
1 m f(x) = anfo sup{ f(x) : x E (N8( p) \ { p}) fl E}.
Similarly, the limit inferior off at p, denoted lim f(x), is defined by
=FP
!,f f (x) = ssu$ inf{}(x) : x e (N8(p) \ { p}) fl }.
x-P
Prove each of the following:
a. lim f (x) <_ L if and only if given e > 0, there exists a S > 0 such that f(x) < L + e for all x E E,
0<Ix-pI<S.
b. lim f(x) z L if and only if given e > 0 and 8 > 0, there exists x e E with 0 < I x - p I < S such that
J-P
f(x)>L-e.
c. If I m f(x) = L, then for any sequence {xn} in E with x" # p for all n E N. and limx" = P, lim f(x") 5 L.
d. There exists a sequence {x"} in E with x" # p for all n E N, such that and limx" = P,
lim f(xn) = lim f(x).
xp
3. Let X C R and f a real-valued function on X. For p E X, the oscillation of fat p, denoted w(f; p), is defined as
w(f; p) = inf sup{ If(x) - f(y) I : x, y e Na(p) fl X}.
8>0
Supplemental Reading 163
f(x)= _12
I
The Cantor ternary function f on [0, 1 ] is defined as follows: f(0) = 0, and if x E (0. 1 ] with ternary expan-
sion x = .aja2a3 .
b
, set
SUPPLEMENTAL READING
Bryant, J., Kuzmanovich, J. and Pavlichenkov, A., Grabinger, Judith V., "Who gave you the epsilon?
"Functions with compact preimages of compact Cauchy and the origins of rigorous calculus;' Amer.
sets:' Math. Mag. 70 (1997), 362-364. Math. Monthly 90 (1983), 185-194.
Bumcrot, R. and Sheingorn, M., "Variations on continuity: Martelli, M., Dang, M. and Seph, T.. "Defining chaos;'
Sets of infinite limits,' Math. Mag. 47 (1974), 41-43. Math. Mag. 71 (1999), 112-122.
Cauchy, A. L., Cours d'Analyse, Paris, 1821, in Oeuvres Snipes, Ray F., "Is every continuous function uniformly
compldtes d'Augustin Cauchy, series 2, vol. 3, continuous?" Math. Mag. 57 (1994),169-173.
Gauthier-Villars, Paris, 1899. Straffin, Jr., Philip. D., "Periodic points of continuous
Fleron, Julian F., "A note on the history of the Cantor set functions;' Math. Mag. 51 (1978), 99-105.
and Cantor function;' Math. Mag. 67 (1994), Velleman, D. J., "Characterizing continuity," Amer.
136-140. Math. Monthly 104 (1997), 318-322.
Differentiation
5.1 The Derivative
52 The Mean Value Theorem
5.3 L'Hospital's Rule
5.4 Newton's Method
d (for difference) and f (for sum) notation of Leibniz to denote differentiation and in-
tegration. Many of the basic rules and formulas of the differential calculus were devel-
oped by these two remarkable mathematicians. In the paper A New Method for Maxima
and Minima, and also for Tangents, which is not Obstructed by Irrational Quantities,
published in 1684, Leibniz gave correct rules for differentiation of sums, products. quo-
tients, powers, and roots. In addition to his many contributions to the subject. Leibniz
also disseminated his results in publications and correspondence with colleagues
throughout Europe.
Newton and Leibniz, with their invention of the calculus, had created a tool of such
novel subtlety that its utility was proved for over 150 years before its limitations forced
mathematicians to clarify its foundations. The rigorous formulation of the derivative did
not occur until 1821 when Cauchy provided a formal definition of limit. This helped to
place the theory on a firm mathematical footing. Cauchy's contributions to the rigorous
development of calculus will be evident in both this and subsequent chapters.
In this chapter we develop the theory of differentiation based on the definition of
Cauchy, with special emphasis on the mean value theorem and consequences thereof.
The first section presents the standard results concerning derivatives of functions ob-
tained by means of algebraic operations and composition. In the examples and exercises
we will derive the derivatives of some of the basic algebraic and trigonometric func-
tions. However, throughout the chapter we will assume that the reader is already famil-
iar with standard techniques of differentiation and some of its applications. Therefore,
we will concentrate on the mathematical concepts of the derivative. emphasizing many
of its more subtle properties.
Figure 5.1
The limit of this quantity as t approaches to, again provided that the limit exists, is taken
as the definition of the velocity of the object at time ta.
Both of the previous examples involve identical limits; namely,
s(t) - s(t°)
limf (x) - A p) and lira
x-p 1-.1, t - t,
These limits,. if they exist, are called the derivatives of the functions f and s at p and t
respectively. The term derivative comes from the French fonction derivee.
5.1.1 DEFINITION Let I C R be an interval and let f be a real-valued function with do-
main I. For fixed p E 1, the derivative off at p. denoted f'(p), is defined to be
Pp) = limf(x)
x-,P x-p
-f(p)
provided the limit exists. If f(p) is defined at a point p E 1, we say that f is differen-
tiable at p. If the derivative f is defined at every point of a set E C 1, we say that f is
differentiable on E.
If p is an interior point of I, then p + h E I for all h sufficiently small. If we set
x = p + h, h # 0, then the definition of the derivative of f at p can be expressed as
+ hh' - f(P)
f'(P) = lt_mf(P
provided the limit exists. This formulation of the derivative is sometimes easier to use.
In the definition of the derivative we do not exclude the possibility that p is an
endpoint of I. If p E I is the left endpoint of 1, then
provided, of course, that the limit exists. The analogous formula also holds if p E I is
the right endpoint of L In analogy with the right and left limit of a function, we also de-
fine the right and left derivative of a function.
168 Chapter5 Differentiation
5.1.2 DEFINITION Let I C I be an interval and let f be a real-valued function with do-
main I. If p E I is such that I fl (p, oo) # 0, then the right derivative off at p, de-
noted f' (p), is defined as
f(P + h) - f(P)
f+(P) = lim
h-4O h
provided the limit exists. Similarly, if p E I satisfies (-oo, p) fl I # 4,, then the left de-
rivative off at p, denoted f'_(p), is given by
f(P + h) - f(P)
f'(P)=hli3t- h
Remarks
(a) If p E Int(/), then f'(p) exists if and only if both f"+(p) and f_(p) exist and are
equal. On the other hand, if p E I is the left (right) endpoint of I, then f'(p) exists if
and only if f+(p) (f_(p)) exists. In this case, f'(p) = f+ (P) (.f' (P))
The reader should note the distinction between f+(p) and f'(p+). The first de-
notes the right derivative off at p, whereas the latter is the right limit of the deriva-
tive; i.e.,
f'(P+) = lim f'(-).
Here, of course, we are assuming that f' is defined for all x E (p, p + 8) for some
8>0.
(b) If f is a differentiable function on an interval 1, we will also occasionally use Leib-
niz's notation
d
f(x), "& , or '*,
dx
In a similar fashion we can define the third derivative off at p, denoted f'"(p) or f(')(p).
In general, for n E N, ft"t(p) denotes the nth derivative off at p. In order to discuss
the existence of the nth derivative off at p, we require the existence of the (n - 1)st de-
rivative off on an interval containing p.
5.1 The Derivative 169
5.1.3 EXAMPLES
(a) In the exercises (Exercise 2) you will be asked to prove that if f(x) = x", n E Z,
then f'(x) = nx"- ' for all x E IB (x * 0 if n is negative). For the function f(x) = xz,
the result is obtained as follows:
z -
,y9 (x +
fi(x) = h = h) = 2x.
= ( x+h-V)( x+h+V)
h ( x+h+Vx)
x+h+;7x
Since h x -+h = ix, we have
f'(x)=h+ 1
=
1
x + h + vx 2Vx
(c) Consider f(x) = sin x. From the identity
sin(x + h) = sin x cos h + cos x sin h,
we obtain
sin(x + h) - sin x r cos h-I sin h
h = sin x l h I+ cos x f h
lmi-
sin h
h
=l and lim
cos h - 1 = 0.
Therefore,
sin(x + h) - sin x
f'(x).
LIn h
cos h - 11 [ sin h
= sin x lim i + cos x l
= cos X. J J
d (cos x) = -sin x.
170 Chapters Differentiation
Thus f (0) and f'_(0) both exist, but are unequal. Therefore f'(0) does not exist.
(e) In this example let g(x) = x312 with Dom g = [0, oo). Then for p = 0,
x3/2
8'(0) = 8+(0) =I-,Q
limX- = Jim
.Ho
V = 0.
Thus g is differentiable at 0 with g'(0) = 0.
(f) Let f be defined by
xsin,
x x#0,
x
f(x) =
g(x) =
II
(g) Consider the following variation of (f). Let
x2 sin x
0,
, x # 0,
x0.
This example is very important! In Exercise 9 you will be asked to show that g'(x) ex-
ists for all x E R with g'(0) = 0, but that the derivative g' is not continuous at 0.
When Cauchy gave the rigorous definition of the derivative, he assumed that the
given function was continuous on its domain. As a consequence of the following theo-
rem this requirement is not necessary.
Since
limi(t) -f(p)
'-+p t-p
exists and equals f'(p), by Theorem 4.1.6(b),
(f(t) - f(P)1Jlim(r - P) = Pp) 0 = 0.
lim(f(t) - f(p)) = lim
r-+p _P t-p r-+p
Remark In both Examples 5.1.3(d) and (f), the given function is continuous at 0, but
n o t d i ff e r e n t i a b l e a t 0. Given a f i n it e number of points, say p i, ... , pn, it is easy to con-
struct a function f which is continuous but not differentiable at pI, .. . , p,,. For example,
n
f(x)= k-I
I Ix-Pk1
has the desired properties. In 1861, Weierstrass constructed a function f which is con-
tinuous at every point of 11 but nowhere differentiable. When published in 1874, this
example astounded the mathematical community. Prior to this time mathematicians
generally believed that continuous functions were differentiable (except perhaps at a
finite number of points). In Example 8.5.3 we will consider the function of Weierstrass
in detail.
+ (f(x + h) - f(x)lAX)
h J
172 Chapters Differentiation
= f(x)g'(x) + g(x)f'(x).
To prove (c), we first prove that (1/g)'(x) = -g'(x)/[g(x)]2, provided g(x) * 0.
The result then follows by writing f/g as f (1/g) and applying the product formula (b).
Suppose g(x) * 0. By continuity of g at x, there exists a S > 0 such that g(x + h) * 0
for all h, I h i < S. Thus for h sufficiently small and nonzero,
1 1
1 1
1ll' __ lim
g(x + h) g(x)
C-/
8 (x) h-0 h
(g(x + h) - g(x)1
_ -lim
11-0 \
) lim
n-o g(x)g(x + h)
h
-g'(x)
= g2(x) ,
Then by Definition 5.1.1, Q(t) --+f(x) as tax. If we let u(t) = Q(t) - f'(x). then
u(t) -- 0 as t -+ x. Therefore, if f is differentiable at x, fort * x,
f(t) - f(x) = (t - x)[f'(x) + u(t)] where u(t) -+0 as (1)
By setting u(x) = 0, the above identity is valid for all r E I.
5.1 The Derivative 173
5.1.7 EXAMPLES
(a) By Example 5.1.3(c) the function f (x) = sin x is differentiable on R. Hence if
g :1 * R is differentiable on the interval 1, h(x) = If o g)(x) = sin g(x) is differentiable
on I with
h'(x) = f'(g(x)) g'(x) = g'(x)cos g(x).
In particular, if g(x) = 1/x2, Dom g = (0, oo), then by Theorem 5.1.5(c),
- 2x 2
g'(x) _ (x2)2 = x3, x E (0, oo).
Therefore,
d
dx
sm-
.
x2
-=-cos-
2
x2
1 1
2.
(b) By Exercise 2,
dx"=nx"-' forallnEfol.
174 Chapters Differentiation
Thus if f : I -+ R is differentiable on the interval 1, then by the chain rule g(x) = [f(x) ]1.
n E N, is differentiable on I with g'(x) = n[f(x)]"-'f'(x). This formula can also be ob-
tained from Theorem 5.1.5(b) using mathematical induction.
EXERCISES 5.1
1. Use the definition to find the derivative of each of the following functions.
*a. f(x)=x;, x E R b. g(x) = x+2, x> -2
X, x # 0
*c. h(x) = 1 d. k(x) = l
'
x > -2
*e. f (x)
x+1
x* -1 f. g(x) = xZ + 1' x E R
2. *Prove that for all integers n,
d (cos x) = -sin x.
6. Let f(x) = lxI3. Compute f'(x), f"(x), and show that f"'(0) does not exist.
7. Determine where each of the following functions from R to R is differentiable and find the derivative.
*a. 1(x) = x(x]
b.g(x)=Ix-21+Ix+tI
*c. h(x) = I sin x I
{x2[].
0, :::"
8. Use the product rule, quotient rule, and chain rule to find the derivative of each of the following functions.
a. f (z) = x sin x*0 b. f(x) = (cos (sin x) ), n, m E h!
c. f(x)= x+1Y d. f(x)=x°(2+sink), x * 0
5.1 The Derivative 175
9. Let g be defined by
g(x) =
x2 sin1,
x
x0,
15. If f : (a, b) -+ R is differentiable at p E (a, b), prove that f'(p) = lim n[f(p + ,l,) - f(p)]. Show by example that
the existence of the limit of the sequence {n[ f(p + f(p)j) does not imply the existence of f'(p).
16. Leibniz's Rule: Suppose f and g have nth order derivatives on (a, b). Prove that
The function f, illustrated in Figure 5.2, has a local maximum at a, p,, and p, and
a local minimum at and b. The points (p4,f(p4)) and (p,.f(p,)) are absolute
maxima and absolute minima, respectively.
Figure 5.2
5.2 The Mean Value Theorem 177
The following theorem gives the relationship between local maxima or minima of
a function defined on an interval and the values of its derivative.
5.2.2 THEOREM Let fbe a real-valued function defined on an interval 1, and supposef has
either a local minimum or local maximum at p E Int(l). 1f f is differentiable at p, then
f(P) = 0.
Proof. If f is differentiable at p E Int(1), then f'- (p) and f; (p) both exist and are
equal. Suppose f has a local maximum at p. Then there exists a S > 0 such that
f (t) f (p) for all t E I with I t - p I < S. In particular, if p < t < p + S, t E 1, then
f(t) - f(p) s 0.
t-P
Thus f+(P) 0. Similarly, if p - S < t < p,
f(t) - f(P) 0,
t-p
and therefore f'_(p)' 0. Finally, since f+ (p) = f'_ (p) = f'(p), we have f(p) = 0.
The proof of the case wheref has a local minimum at p is similar. O
As a consequence of the previous theorem we have the following corollary.
5.2.3 COROLLARY Lei f be a continuous real-valued function on [a, b]. 1f f has a local
maximum or minimum at p E (a, b), then either the derivative off at p does not exist,
or f(p) = 0.
Remark. The conclusion of Theorem 5.2.2 is not valid if p E I is an endpoint of the
interval. For example, if f :[a, b] -- R has a local maximum at a, and if f is differen-
tiole at a, then we can only conclude that f'(n) = f+ (a) 5 0. This is illustrated in the
following examples.
5.2.4 EXAMPLES
(a) The function
f(x)= x- lZ , 05x2,
has a local maximum at p = 0 and p = 2, and an absolute minimum at q = ;. By com-
putation, we have f'(0) _ -1, f' (2) = 3, and f (2) = 0. The graph of f is given in
Figure 5.3.
(b) The function f (x) _ x 1, x E [ -1, 11, has an absolute minimum at p = 0. How-
ever, by Example 5.1.3(d) the derivative does not exist at p = 0.
Rolle's Theorem
Prior to stating and proving the mean value theorem, we first state and prove the fol-
lowing theorem credited to Michel Rolle (1652-1719).
178 Chapter 5 Differentiation
24J
21
Il
4
Since the derivative off at c gives the slope of the tangent line at (c, f (c)), a geomet-
ric interpretation of Rolle's theorem is that if f satisfies the hypothesis of the theorem, then
there exists at least one value of c E (a, b) for which the tangent line to the graph off is
horizontal. For the function f depicted in Figure 5.4, there are exactly two such points.
Proof. If f is constant on [a, b], then f'(x) = 0 for all x E [a, b]. Thus, we assume
that f is not constant. Since the closed interval [a, b] is compact, by Corollary 4.2.9,f
has a maximum and a minimum on [a. b]. If f(t) > f(a) for some t, then f has a max-
imum at some c E (a, b). Thus by Theorem 5.2.2, f'(c) = 0. If f(t) < f(a) for some t,
then f has a minimum at some c E (a, b), and thus again f'(c) = 0.
Remarks
(a) Continuity off on [a, b] is required in the proof of Rolle's theorem. The function
Ix, 05x< 1,
f(x)= 0, x=1
is differentiable on (0, 1) and satisfies f(0) = f(1) = 0; yet f' (x) * 0 for all x E (0, 1).
The function f fails to be continuous at 1.
(b) For Rolle's theorem, differentiability of f at a and b is not required. For example,
the function f(x) = - xZ, x e [-2, 2], satisfies the hypothesis of Rolle's theo-
rem, yet the derivative does not exist at -2 and 2. For x E (-2, 2),
fi(x) _
V4 - X2
and the conclusion of Rolle's theorem is satisfied with c = 0.
5.2 The Mean Value Theorem 179
f(b) - f(a)
g'(x) =f'(x) - b - a
f (b)
for all x E (a, b). Taking x = c gives f(c) =
now follows.
- a(a), from which the conclusion
180 Chapters Differentiation
The mean value theorem is one of the fundamental results of differential calcu-
lus. Its importance lies in the fact that,it enables us to obtain information about a
function f from its derivative f'. In Example 5.2.7 we will illustrate how the mean
value theorem can be used to derive inequalities. Other applications will be given
later in this section and in the exercises. It will also be used in many other instances
in the text.
5.2.7 EXAMPLE In this example we illustrate how the mean value theorem may be used
in proving elementary inequalities. We will use it to prove that
x
s ln(1 + x) < x for all x > - 1,
1 + x
where In x denotes the natural logarithm function on (0, oo). This function is defined
and considered in detail in Example 6.3.5 of the next chapter. There it is proved that the
derivative of In x is 1/x. Let f(x) = ln(1 + x), x E (-1, oo). Then f(0) = 0. If x > 0,
then by the mean value theorem, there exists c e (0, x) such that
In(1 + x) = f(x) - f(0) = f(c)x.
But f'(c) _ (1 + c)-' and (I + x)-' < (1 + c)-t < 1 for all c E (0, x). Therefore,
x
1 +x<f'(c)x<x
and as a consequence,
x
1 +x`ln(1 +x)<x for all xz0.
5.2 The Mean Value Theorem 181
Now suppose -1 < x < 0. Then again by the mean value theorem there exists c E (x. 0)
such that
But since x < c < 0, 1 < (1 + c)-' < (1 + x)-', and since x is negative
Hence the desired inequality holds for all x > -1, with equality if and only if x = 0.
U
The following theorem, attributed to Cauchy, is a useful generalization of the mean
value theorem.
5.2.8 THEOREM (Cauchy Mean Value Theorem) If f, g are continuous real-valued func-
tions on [a, b] that are differentiable on (a, b), then there exists c E (a, b) such that
. Proof. Let
Thus by Rolle's theorem, there exists c E (a, b) such that h'(c) = 0, which gives the
result. U
The geometric interpretation of the Cauchy mean value theorem is very similar to
that of the mean value theorem. If g'(x) * 0 for all'x E (a, b), then g(a) 0 g(b) and the
conclusion of Theorem 5.2.8 can be written as
Q=(g(b). f(b))
P=(g(a), f(a))
Figure 5.6
Theorem 5.2.9 is often used to determine maxima and minima of functions as fol-
lows: Suppose f is a real-valued continuous function on (a, b), and c E (a, b) is such
that f'(c) = 0 or f(c) does not exist. Suppose f is differentiable on (a, c) and (c, b). If
f'(x) < 0 for all x E (a, c) and f'(x) > 0 for all x E (c, b), then by Theorem 5.2.9. f
is decreasing on (a, c) and increasing on (c, b). As a consequence, one concludes that f
has a local minimum at c. This method is usually referred to as the first derivative test
for local maxima or minima. The natural inclination is to think that the converse is also
true; namely, if f has a local minimum at c, then f is decreasing to the left of c and in-
creasing to the right of c. As the following example shows, however, this is false!
f(x) = {x4(2 x ,
0, x = 0.
The function f has an absolute minimum at x = 0; however, f'(x) has both negative
and positive values in every neighborhood of 0. The details are left as an exercise
(Exercise 19). The graph of f'(x) = 4x3(2 + sin 1/x) - x2 cos l/x, x * 0, for x in a
neighborhood of zero is given in Figure 5.7.
5.2.11 THEOREM Suppose f :[a, b) -+ R is continuous on [a, b) and differentiable on (a, b).
If lim f'(x) exists, then f+ (a) exists and
x-+a
f, (a) = 1 m, f'(x).
Xa
184 Chapter5 Differentiation
5.2.12 EXAMPLES
(a) To illustrate the previous theorem, consider the function
Jx2 + 1, x < 1,
f(x) ° 3-x, x> 1.
For x < 1, f'(x) 2x, which has a left limit of 2 at x = I. Thus by the theorem,
f'_(1) = lim 2x = 2.
Similarly,
f+'(1) = lim (-2x) = -2.
Ix2 sin 1 x # 0,
$(x) = x
0, x = 0,
of Example 5.1.3(g) has the property that g'(0) exists but lidg'(x) does not.
Suppose f'(a) < A < f'(b). Then g'(a) < 0 and g'(b) > 0. As in the remark fol-
lowing Theorem 5.2.9. since g'(a) < 0, there exists an x, > a such that g(.r,) < g(a)-
Also, since g'(b) > 0, there exists an x, < b such that g(x.) < g(b). As a conse-
quence, g has an absolute minimum at some point c E (a, b). But then
g'(c) = f'(c) - A = 0,
i.e., f'(c) = A. Q
The previous theorem is often used in calculus to determine where a function is in-
creasing or decreasing. Suppose it has been determined that the derivative f is zero at
c, and c2 with c, < c2, and that f'(x) * 0 for all x E (c,, c2). Then by the previous the-
orem, it suffices to check the sign of the derivative at a single point in the interval
(cr, c2) to determine whether f is positive or negative on the whole interval (c c,).
Theorem 5.2.9 then allows us to determine whether f is increasing or decreasing on
(cj, c2).
Since this holds for any sequence with y, -r y,,, y * y,,, by Theorem 4.1.3 and the
definition of the derivative,
(f TO = rl(xo)' v
186 Chapter 5 Differentiation
Remark, The hypothesis that f'(x) # 0 for all x E I is crucial. For example, the
function f(x) = x3 is strictly increasing on [ -1, 1 ] with f'(0) = 0. The inverse func.
tion f -'(y) = y'13, however, is not differentiable at y = 0.
5.2.15 EXAMPLES
(a) As an application of the previous theorem, we show that f (x) = x'1", x E (0, oo),
n E N, is differentiable on (0, oo) with
n
for all x E (0, oo). Consider the function g(x) = x", n E N, Dom g = (0, oo). Then
g'(x) = nx"-' and g'(x) > 0 for all x E (0, oo). By the previous theorem, g-' is dif-
ferentiable on J = g((0, oo)) _ (0, oo) with
(g-')'(g(x)) = = 1-1
g'(x)
If we set y = g(x) = x", then x = y"" and
If we set E = L', then E'(L(x)) = x, or E'(y) = E(y) where y = L(x). The function
E(x), x e 08, is called the natural exponential function on } and is usually denoted
by e', where e is Euler's number of Example 2.3.5. The exponential function E(x) is
considered in greater detail in Example 8.7.20.
(c) In this example we consider the inverse function of g(x) = cos x, x E [0, ir]. Since
g'(x) = -sin x is strictly negative for x E (0, ir), the function g is strictly decreasing
on [0, a] with g([0, a]) _ [- I, 1 ]. Therefore its inverse function g- 1, which we de-
note by Arccos, exists on [ -1, 1 ]. Thus for y E [ -1, 1 ], x = Arccos y if and only if
y = cos x. Finally, since g'(x) 0 for x E (0, ir), by the inverse function theorem,
-1
(8-,)'(g(x)) = g'(x) -sinx = N/l z
'
or since y = cos x,
-l
Arccos y =
dy
The graphs of both cos x, x E [0, ir], and Arccos x, x E [ -1, 13, are given in Figure 5.8.
5.2 The Mean Value Theorem 187
EXERCISES 5.2
1. For each of the following functions, determine the interval(s) where the function is increasing or decreasing, and
find all local maxima and minima.
*a f(x)=x3+Gx-5, xER b.g(x)=4x-x°, xER
c. h(x) = 1+x2x2, x e R d. k(x) _ V- x, xz0
2
e.l(x)=x+7, x#0 f f(x)x-b, a*b,x*b
2. Let f (x) _ (x - a,)2, where a,, a2, ... , a are constants. Find the value of x where f is a minimum.
3. As in Example 5.2.7, use the mean value theorem to establish each of the following inequalities.
a. 1 +xs1+Ix, x> -I b.exZI+x, xER
C., (I +x)a? I + ax, x> -1,a> I
1. For a E N, this inequality was proved by mathematical induction in Example 1.3.3(b). In this exercise,
and in Exercise 4(b), you may assume that for a E R.
d x° = ax°-
188 Chapter 5 Differentiation
14. A differentiable function f : [a, b] - R is uniformly differentiable on [a. b] if for every E > 0 there exists a
S > 0 such that
f(t) - f(x) <E
t-x
5.2 The Mean Value Theorem 189
for all r, x E (a, b] with 0 < It - xl < S. Show that f is uniformly differentiable on (a. b] if and only if f is
continuous on [a, b].
15. *Suppose f : [a, b] -> R with f. (a) > 0. Prove that there exists a S > 0 such that f (x) > f(a) for all
x, a<x<a+S.
16. Prove that the equation x3 - 3x + b = 0 has at most one root in the interval [ -1. 1 ].
M. 'Suppose gas differentiable on (a. b) with I g'(x) I s M for all x E (a. b). Prove that there exists an e > 0 such
that the function f(x) = x + eg(x) is one-to-one on (a, b).
18. Let
x'(2r.,+ all) X :A 0, ..
f(x) =
0, x=0.
a. Show that f has an absolute minimum at x = 0.
b. Show that f'(x) assumes both negative and positive values in every neighborhood of 0.
20. Let f(x)=x2,g(x)=x3,xE[-1,l].
a. Find c E (-1, 1) such that the conclusion of Theorem 5.2.8 holds.
b. Show that there does not exist any c E (-1, 1) for which
Al) -f(-1) f(c)
g(1) - g(-1) g'(c)'
21. For r E 0 and x > 0, let f(x) = x'. Prove that f'(x) = rx'-'.
22. Suppose L : (0, oo) R is a differentiable function satisfying L'(x) = 1/x with L(1) = 0. Prove each of the fol-
lowing.
a. L(ab) = L(a) + L(b) for all a, b E (0, oo)
b, L(I/b) = -L(b), b > 0
c. L(b') = rL(b), b > 0, r E R
d. L(e) = 1, where e is Euler's number of Example 2.3.5
e. Range L = R
23. Let g(x) - tan x, -f < x < .
a. Show that g is one-to-one on (-f, i) with Range g = R.
'b. Let Arctan x, x E R, denote the inverse function of g. Use Theorem 5.2.14 to prove that
d
An:tan x
+ x2 .
24. a. , Show that f(x) = sin x is one-to-one on [ -i, f] with f([ - z, ]) _ [ - I,.1 J.
z
b. For x E (- I, 1]. let Arcsin x denote the inverse function off. Show that Arcsin x is differentiable on (-1, 1).
and find the derivative of Arcsin x.
190 Chapter 5 Differentiation
25. Let f : (0, oo) -> R be differentiable on (0, oo) and suppose that lim f'(x) = L.
f (x + h f(x)
a. Show that for any h > 0, lim
h
- = L.
x
b. Show that lim f )) = L.
Infinite Limits
Since l'Hospital's rule allows for infinite limits, we provide the following definitions.
Remark. Since we now allow the possibility of a function having infinite limits, it
needs to be emphasized that when we say that a function f has a limit at p e I3
(or at ± oo), we mean a finite limit.
L'Hospital's Rule
L'Hospital's rule is useful for evaluating limits of the form
gx(x)
lim
where either (a) lim f(x) = lim g(x) = 0 or (b) f and g tend to ± oo as x -+p. If (a)
holds, then 1 m W 419W) is usually referred to as indeterminate of form 0/0, whereas
in (b) the limit is referred to as indeterminate of form oo/oo. The reason that (a) and (b)
are indeterminate is that previous methods may no longer apply.
In (a), if either lim f (x) or lim g(x) is nonzero, then previous methods discussed in
x-ip
Section 4.1 apply. For example, if both f and g have limits at p and lim g(x) * 0, then
by Theorem 4.1.6(c), p
lim f(x)
f(x)
lim -_
x-*p 8(x) lim g(x)
On the other hand, if lim f (x) = A * 0 and g(x) > 0 with lim g(x) = 0, then as
X-P
x-ip, f(x)/g(x) tends to `Z
oo if A > 0, and to -oo if A < 0 (Exercise 5). However, if
lim f(x) = lim g(x) = 0, then unless the quotient f(x)/g(x) can somehow be simplified,
previous methods may no longer be applicable.
then
lim f ((x() = L.
x-.a' g(x)
Remark. The analogous result where x b- is obviously also true. A more elemen-
tary version of 1'Hospital's rule, which relies only on the definition of the derivative, is
192 ChaprerS Differentiation
given in Exercise 2. Also, Exercise 7 provides examples of two functions f and g satis-
fying (a) for which lim (f(x)/g(x)) exists but lim (f'(x)/g'(x)) does not exist.
x-a x-a
Proof. (a) Suppose (a) holds. We first prove the case where a is finite. Let be a
sequence in (a, b) with x -+a and x * a for all n. Since we want to apply the gener-
alized mean value theorem to f and g on the interval [a, x,), we need both f and g con-
tinuous at a. This is accomplished by setting
f(a) = g(a) = 0.
Then by hypothesis (a), f and g are continuous at a. Thus by the generalized mean value
theorem, for each n E NI there exists c between a and /x such that
p
[f(xa) - J g(a))f (cn),
or
f(xn) _ f'(Ca)
g(xa) g'(Cn)
Note, since g'(x) # 0 for all x E (a, b), g(x,) # g(a) for all n. As n --I- oc, c -> a'.
Thus by Theorem 4.1.3 and the hypothesis,
'_°° AX-) =
l- f' (x)
m g'(x) -
= L.
Since the above holds for every sequence with x tea', the result follows.
Suppose a = -oo. To handle this case, we make the substitution x = -1/t. Then
as t -> 0+, x -> -oo. Define the functions (p(t) and *(t) on (0, c) for some c > 0 by
t-0 +G r)
x lim
- g(X)L,
and that
1m P(t) = lim iy(t) = 0.
Thus by the above,
t
= L.
xli.m- g(x) = 1-'0 00)
(b) Suppose I m, g(x) = oo. The case in which g(x) -+ -oo is treated similarly.
Rather than treating the finite case and infinite case separately, we provide a proof that
works for both.
5.3 CHospital's Rule 193
Suppose first that -oo s L < oo, and /3 E R satisfies /3 > L. Choose r such that
L < r < /3. Since
or
hm. = 0.
xm g-) =
Therefore,
8X)
1'm_IgX)+rl I
L<
--ii =r</3.
Thus there exists c3, a < c3 c2, such that /
- +r(l
for all x, a < x < c3. Thus by inequality (5),
f(x)
< /3 for all x, a < x < c3. (6)
x
AX)
If L = -oo, then for any /3 E Q8, there exists c3 such that inequality (6) holds for
all x, a < x < c3. Thus by definition,
f(x) -oo.
'tea' g(x )
194 Chapter5 Differentiation
x
<L+e for all x, a < x < c3. (7)
Suppose -oo < L <_ oo. Let a E R, a < L be arbitrary. Then an argument sim-
ilar to the above gives the existence of c3 (=- (a, b) such that
f(x) > a
for all x, a <x<c3 .
g(X)
If L = oo, then the above implies that
limf(x)
,/ = oo.
x-+a' g(x)
On the other hand, if L is finite, taking a = L - e gives the existence of c3 such that
f(x)
>L-e forallx , a<x<c3 .
8()
x
Combining this with inequality (7) proves that
f(x)
lim_ = L. Q
x-*a 8(x)
Remarks
(a) The proof of case (a) could have been done similarly to that of (b), treating the case
where a is finite and -oo simultaneously. I chose not to do so since making the substi-
tution x = -I It is a useful technique, reducing problems involving limits at -oo to
right limits at 0. Conversely, limits at 0 can be transformed to limits at too with the
substitution x = 1/t. These new limits are in many instances easier to evaluate than the
original. This is illustrated in Example 5.3.4(c).
(b) In hypothesis (b), we only required that lim g(x) = too. If lira f (x) is finite, then
x-a x-,a'
it immediately follows that
lint-=0,
f (X)
x+a 8(x)
and l'Hospital's rule is not required (Exercise 4). Thus in practice, hypothesis (b) of
1' Hospital's rule is used only if both f and g have infinite limits.
For convenience we stated and proved l'Hospital's rule in terms of right limits.
Since the analogous results for left limits are also true, combining the two results gives
the following corollary.
If
f(x) = lim g(x) = 0, or
(a) limf(x)
(b) lim g(x) 00,
then
f(x)
limg(x) = L.
5.3.4 EXAMPLES
(a) Consider
ln(1+x
lira
x-.0 X
where In is the natural logarithm function on (0, oo). This limit is indeterminate of form
0/0. With f(x) = In(1 + x) and g(x) = x,
f'x 10 1+x=1.
x .O g'(x)_
Thus by 1'Hospital's rule,
ln(1 + x)
A
lira
X
= 1.
Although l'Hospital's rule provides an easy method for evaluating this limit, the re-
sult can also be obtained by using previous techniques. In Example 5.2.7 we proved that
+xsln(1+x)5x
11>o
ln(1 + x)
x
- 1.
which is again indeterminate of form 0/0. However, applying ('Hospital's rule one more
time gives
cosx
lim
x-O 2
= -
I
Therefore,
lim
1-cosx I
2.
xti0 x2
(e) Consider
e-six
lim
x-.0- X
Since lim a-11x = 0, the above limit is indeterminate of form 0/0. If we apply I'Hospi-
x0'
tal's rule, we obtain
0X e- tix
and this limit is more complicated than the original limit. However, if we let t = 1/x,
then
e-11X t
-,0 X = lim e .
EXERCISES 5.3
1. Provide definitions for each of the following limits.
a. A-P
limf(x)=oo b. limf(x)=oo
x-4ao
2. *Suppose f g are differentiable on (a, b), xa E (a, b) and g'(xa) * 0. If f (xa) = g(xa) = 0, prove that
f(x) f'(xa)
lim
X _X, g(x) g,(x0)
(Hint: Apply the definition of the derivative.)
3. Let h(x) be defined on (-oo, b). Show that there exists a c > 0 such that rp(t) = h(-;) is defined on (0, c), and
that l-.-aa
lim h(x) = lim op(t).
1H04
4. 'Let f, g be real-valued functions defined on (a, b). If lim f(x) exists in R andx-a'
lim g(x) = oo, prove that
lint
xa' f(x)/g(x) = 0.
5.4 Newton's Method 197
5. Suppose f, g are real-valued functions on (a, b) satisfying lim, f(x) = A * 0, lim* g(x) = 0. and g(x) > 0 for all
1
x E (a, b). If A > 0, prove that lim f(x)/g(x) = oc. If A <0, prove that lim. f(x)/g(x) _ -oc.
6. Use I'Hospital's rule and any of the differentiation formulas from calculus to find each of the following limits. In
the following, In x, x > 0, denotes the natural logarithm function.
x5 +2x-3 x`+2x-3
a. lim b. ,Imi
2x3 - x2 - I 2x' - x' - I
I - cos 2A-
c. lim -x In x
d. lira
_.0 sin x
(i. lim
=-o' .r sin .rJ
f( x) fg'(-x(x)
7. Let f(x) = x2 sin(1/x) and g(x) = sin x. Show that lim exists but that lim ) does not exist.
box)
p(X) p(x),
8. Investigate lim
. q(x)
and lim where p and q are polynomials of degree it and in. respectively.
x_'-« q(x)
9. Let f (x) = (sin x)/x for x * 0, and f (0) = I .
a. Show that f'(0) exists, and determine its value.
b. Show that f"(0) exists, and determine its value.
e_
10. Let f be defined on R by, f(x) = for x # 0. and f(0) = 0. Prove that ft"'(0) = 0 for all n = I. 2. .
c,=2(a+b).
If f(c,) = 0, we are done. If f(c1) * 0, then c belongs to one of the two intervals
(a, c1) or (c1, b), and thus Ic1 - ci < I' (b - a). Suppose (as in Figure 5.9) f(c1) > 0.
Then c E (a, c1), and in this case we set co = a, and
C2 = 2 (CO + CO-
If f(c2) = 0, we are done. If not, then suppose (as in Figure 5.9) f(c2) < 0. Then
c E (c2, c1), and as above, we set
C3 = I (Cl + C2).
I//
c1,C2,...,cn,n?2 C. = lCn-1 + C)
Figure 5.9
exact value. If f(cn_ 1)f(cn) < 0, then c lies between cn_ 1 and c,, and we define
On the other hand, if 0, then c lies between c and c,, and in this case.
we define
:2I(b-a).
Thus lim c = c and by continuity, f(c) = 0.
Although this method provides a sequence of numbers that converges to the zero off.
it has the disadvantage that the convergence is rather slow. An alternative method, due to
Raphson, uses tangent lines to the curve to find successive points c approximating the zero
off. As we will see, this method will often converge much more rapidly to the solution.
As above, assume that f is differentiable on [a, b] with f (a) f (b) < 0 and
f'(x) # 0 for all x E [a, b]. Let ct be an initial guess to the value c. The line tangent
to the graph off at (c,, f (c,)) has an equation given by
Y =f(ct) +f(ct)(x - CO.
Since f (c,) # 0, the line crosses the x-axis at a point that we denote by c2 (Figure 5.10).
Thus
0 =f(ci) +f'(ct)(c2 - Cl),
that, upon solving for c2, gives
f(CI)
C2 = C, -
f'(CI)
We now replace the point c, by the second estimate c2 to obtain c3, and so forth. In-
ductively, we obtain a sequence (cn) given by the formula
c
----- , n = 1,2... (8)
c"
in which c, is an initial guess to the solution f(c) = 0. As we will see, under suitable
hypothesis, the sequence {c"} will converge very rapidly to a solution of the equation
f(x) = 0. Before we prove the main result, we illustrate the above with an example.
T(-
J (Cn)
c.), nEN,
Cn+l=cn-f,(c),
is in 1, and n-co
lim c" _. c. Furthermore,
Prior to proving Theorem 5.4.2 we first state and prove the following lemma. The
result is in fact a special case of Taylor's theorem (8.7.16), which will be discussed in
Chapter 8.
5.4.3 LEMMA Suppose f : [a, b] -a R is such that f and f' are continuous on [a, b] and
f"(x) exists for all x E (a, b). Let x" E [a, b). Then for any x E [a, b], there exists a
real number between x, and x such that
or 2
If x, is defined by
f(x.)
X1 = X. - f,(xo),
then by equation (10).
Therefore,
"
IX, - cl = 2 Ic - x,12 s Ic - x,12. (11)
x,
Choose s>0sothat 8<2m/M andI=(c-S,c+S] C [a,b].Ifc"E1,then
lc - c"I < S. If c"+, is defined by formula (8), then by formula (11),
M
c"+,-c1<- 82<S.
Therefore, c"+, E I. Thus if the initial choice c, E I, then c" E I f o r all n = 2, 3, ... .
202 Chaprer5 Differentiation
ICn+I - CI <
2 - I.
Remarks
(a) For a given function f satisfying the hypothesis of the theorem, the constants M and
m, and thus S can be determined. To determine the interval I, one can use the method of
bisection to find an approximation xn to c satisfying Ix,, - x - I I < S. If cl is taken to
be xn, then Icl - ci < S. In practice, however, one usually makes a judicious guess for
c1 and proceeds with the computations.
(b) Let en = c - cn be the error in approximating c, and let K = M/2m. Then in-
equality (9) can be expressed as
ien+1i C Kleni2.
Consequently, if I e,,I < 10', then I en + I I < K 10-2. Thus, except for the constant
factor K, the accuracy actually doubles at each step. For this reason. Newton's method
is usually referred to as a second order or quadratic method.
(c) Even though Newton's method is very efficient, there are a number of things that
can go wrong if c1 is poorly chosen. For example, in Figure 5.11, the initial choice of
c1 gives a c2 outside the interval, and the subsequent cn tend to -cc. Such a function is
given by f (x) = x/(x2 + 1). In Figure 5.12, the initial choice of c1 causes the subse-
quent values to oscillate between c1 and c2. A function having this property is given by
g(x) = x - J Taking c1 = 1 gives c2 = -1, c3 = 1, etc. For this reason, the initial
choice of c1 for many functions has to be sufficiently close to c in order to be sure that
the method works.
(cl.ftc1))
(c2. f(c2))
Figure 5.11
Notes 203
Figure 5.12
EXERCISES 5.4
1. *For a > 0, apply Newton's method to f(x) = x3 - a to obtain a sequence that converges to the cube root
of a.
2. Use Newton's method to find approximations to the roots, accurate to six decimal places, of the given functions on
the interval [0, 1 ].
*a. f(x) = x3 - 3x + 1 b. f(x) = 3x3 - 5x2 + I
c. f(x)=x`-3x3+1
3. Use Newton's method to approximate the real zeros of f(x) = x' - 4x - 3 accurate to four decimal places.
4. Show that f(x) = In x - x + 3, x E (0, oo), has two real zeros. Use Newton's method to approximate them accu-
rate to four decimal places.
5. Suppose g : [a, b] -i, R is differentiable on an interval [a, b] and suppose that there exists a constant c, 0 < c < I,
such that Ig'(x)I s c for all x E [a, b]. Prove, that there exists a unique point x, E [a, b] such that g(xo) = x,. (See
Exercise 13, Section 4.3.)
6. Let f : [a, b] -* R be differentiable on [a, b] with f (a) < 0 < f(b). Suppose there exist constants m and M such
that 0 < m <_ f'(x) s M for all x E [a, b]. Let c, E [a, b] be arbitrary, and define
C-1 = f(c,,)
M
Prove that the sequence converges to the unique zero off on [a, b]. (Hint: Consider the function
g(x) = x - f(x)/M.)
NOTES
Without question the most significant result of this chap- was illustrated by the subsequent theorems and exercises.
ter is the mean value theorem. The simplicity of its proof Additional applications will be encountered throughout
disguises the importance and usefulness of the result. The the text.
theorem allows us to obtain information about the func- Although the mean value theorem is attributed to La-
don from its derivative. This has many applications, as grange, his proof, which appeared about 1772, was based
M Chapter 5 Differentiation
on the false assumption that every function could be The Bernoulli brothers, Jakob (1654-1705) and Jo-
expanded in a power series. Cauchy, in his 1823 text Rf- hann (1667-1748), were among the first mathematicians
sumf des Lecons donnees a L'Ecole Royale Polytechnique in Europe to use the new techniques of Newton and Leib-
sur le Calcul Infinitesimal, used the modern definition of niz in the study of curves and related physical problems.
the derivative to provide a proof of the mean value theo- Among these were finding the equations of the catenary
rem. His statement and proof of the theorem, however, and isochrone.` Both brothers also contributed to the
differ from the version detailed here in that he assumed study of differential equations by solving the Bernoulli
continuity of the derivative. What Cauchy actually proved equation: y' + P(x)v = Q(x)y". Through their numerous
was that if f is continuous on [a, b], then the quantity publications and correspondence with other mathemati-
{f(b) - f(a)}/{b - a) lies between the minimum and cians, the Bernoulli brothers helped to establish the utility
maximum values of f on [a, b]. (See Miscellaneous Ex- of the new calculus. The first text on differential calculus
ercise 4.) Then by the intermediate value theorem (Theo- by ('Hospital also contributed significantly to populariz-
rem 4.2.1 1) applied to the continuous function f, there ing the subject.
exists c E (a, b) such that Leonhard Euler (1707-1783), one of the most pro-
lific mathematicians in history, contributed significantly
f(b) - f(a) = f'(c)(b - a). to establishing calculus as an independent science. Even
It is worth noting that our proof of the mean value theo- though the calculus of exponential and logarithmic func-
rem depends ultimately on the completeness property of tions was basically developed by Johann Bernoulli, it was
R (through Rolle's theorem and Corollary 4.2.9). Euler's expositions in the eighteenth century that brought
The mean value theorem can justifiably be called the these topics into the mainstream of mathematics. Much of
fundamental theorem of differential calculus. It allowed what we know today about the exponential, logarithmic.
the development of rigorous proofs of many results that and trigonometric functions is due to Euler. He was also
were previously taken as fact or "proved" from geometric among the first mathematicians to define the concept of a
constructions. Although the modern proof of l'Hospital's function. However, to Euler-as to the other mathemati-
rule uses the mean value theorem, it should be remem- cians of that period-a function was one that had a power
bered that the original version for calculating the limit of series expansion. It is important to note that most mathe-
a quotient where both the numerator and denominator be- maticians of the eighteenth century. including Euler, were
come zero first appeared in 1696, over 70 years before primarily concerned with computations needed for the ap,
Lagrange's proof of the mean value theorem. The original plications of calculus; proofs did not gain prominence un-
version was stated and "proved" in a purely geometric til the nineteenth century. For this reason, numerous re-
manner without reference to limits. For further details, in- sults of that era that were assumed to be true were
cluding a history of calculus, the reader is referred to the subsequently proved to be true only under more restric-
text by Katz listed in the Bibliography. tive conditions.
MISCELLANEOUS EXERCISES
1. Let f : (a, b) -> R and suppose f" exists at x E (a, b). Prove that
limf(x,+h)+f(x0-h) -2f(x")
f"(z,) hn0 h2
Give an example where this limit exists at x,, but f" does not exist at x,.
3. The catenary problem involves finding the equation of a freely hanging cable, whereas the isochrone
problem involves finding the equation of a curve along which an object would fall with uniform vertical
velocity.
Supplemental Reading 205
5. (T. M. Flett) If f is differentiable on [a, b] and f'(a) = f'(b), prove that there exists ( E (a, b) such that
f(C) - f(a) =
C - a
6. Suppose f: (a, b) R is differentiable at c E (a, b). If (s") and {t"} are sequences in (a, b) with s,, < c < t" and
limoo (t" - s") = 0, prove that
f(s") =
lim
t"-s"
SUPPLEMENTAL READING
Baxley, J. V. and Hayashi, E. K., "Indeterminate forms f (a) to f'(a+)," Math. Mag. 39 (1966),
of exponential type," Amer. Math. Monthly 85 112-120.
(1978), 484-486. Katznelson, Y. and Stromberg, K., "Everywhere differ-
Cajori, Florian, "Historical note on the Newton-Raphson entiable, nowhere monotone function;' Amer. Math.
method of approximation;' Amer. Math. Monthly 18 Monthly 81(1974). 349-354.
(1911), 29-33. Miller, A. D. and Vyborny, R., "Some remarks on func-
Corless, R. M., "Variations on a theme of Newton;' tions with one-sided derivatives;' Amer. Math.
Math. Mag. 71 (1998), 34-41. Monthly 93 (1986),471-475.
Flett, T. M., "A mean value theorem," Math. Gaz. 42 Rosenholtz, "Mere is no differentiable metric for W."
(1958), 38-39. Amer. Math. Monthly 86 (1979), 585-586.
Hall, W. S. and Newell, M. L., 'The mean value theorem Rotando, L. M. and Korn, H., "The indeterminate form
for vector valued functions;' Math. Mag. 52 (1979), 0°," Math. Mag. 50 (1977),41-42.
157-158. Thurston, H. A., "On the definition of the tangent line;'
Hartig, Donald, "L'Hospital's rule via integration;' Amer. Math. Monthly 71 (1964), 1099-1103.
Amer. Math. Monthly 98 (1991). 156-157. Tong, J. and Braza, P. A., "A converse of the mean-value
Langlois, W. E. and Holder, L. 1., "The relation of theorem;' Amer. Math. Monthly 104 (1997). 939-942.
The Riemann and
Riemann-Stieltjes Integral
6.1 The Riemann integral
6.2 Properties of the Riemann Integral
6.3 Fundamental Theorem of Calculus
6.4 Improper Riemann Integrals
6.5 The Riemann-Stieltjes integral
6.6 Numerical Methods
6.7 Proof of Lebesgue's Theorem
When Newton and Leibniz developed the calculus, both considered integration as the
inverse operation of differentiation. For example, in the De analysi. t Newton proved
that the area under the curve y = a''(m/n # -1) is given by
t
an
m+n
by using his differential calculus to prove that if A(x) represents the area from 0 to x
then A'(x) = ax'"/". Even though Leibniz arrived at the concept of the integral by using
sums to compute the area, integration itself was always the inverse operation of differ-
entiation. Throughout the eighteenth century, the definite integral of a function f(x) on
[a, b], denoted fa f(x)dx, was defined as F(b) - F(a) where F was any function whose
derivative was f (x). This remained the definition of the definite integral until the 1820s.
The modem approach to integration is again due to Cauchy, who was the first
mathematician to construct a theory of integration based on approximating the area un-
der the curve. Euler had previously used sums of the form J'=, f(xk _ ,)(xk - x,-,) to
approximate the integral of a function f(x) in situations where the function F(x) could
not be computed. Cauchy, however, used limits of such sums to develop a theory of in-
tegration that was independent of the differential calculus. One of the difficulties with
1. "The Mathematical Works of Isaac Newton;' edited by D. T. Whiteside, Johnson Reprint Corporation. Ness
York, 1964.
208 Chapter 6 The Riemann and Riemann-Stieltjes Integral
Cauchy's definition of the integral was that it was very restrictive: only functions that
were continuous, or continuous except at a finite number of points, were proved to be
integrable. However, one of Cauchy's key achievements was that using his definition,
he was able to prove the fundamental theorem of calculus; specifically. if f is continu-
ous on [a, bi, then there exists a function F on [a, hl such that F'(x) = f(x) for all
x E [a, b].
The modern definition of integration was developed in 1853 by Georg Bernhard
Riemann (1826-1866). Riemann was led to the development of the integral by trying
to characterize which functions were integrable according to Cauchy's definition. In the
process, he modified Cauchy's definition and developed the theory of integration that
bears his name. One of his achievements was providing necessary and sufficient condi-
tions for a real-valued bounded function to be integrable. In Section 6.1 we develop the
theory of the Riemann integral using the approach of Jean Gaston Darboux (1842-1917).
In this section we also include the statement of Lebesgue's theorem, which provides
necessary and sufficient conditions that a bounded real-valued function defined on
a closed and bounded interval be Riemann integrable. A self-contained proof of
Lebesgue's theorem will be given in Section 6.7. The equivalence of the Riemann and
Darboux approach will be proved in Section 6.2.
In Section 6.5 we will consider the more general Riemann-Stieltjes integral that
will give meaning to the following types of integrals:
+
f1
f(x) dx', f J '(x) d[xor f(x) da(x),
J0 ,
11
where a is a monotone increasing function on [a, b]. These types of integrals were de-
veloped by Thomas-Jean Stieltjes (1856-1894) and arise in many applications in both
mathematics and physics. The theory itself involves only minor modifications in the
definition of the Riemann integral; the consequences, however, are far-reaching. The
Riemann-Stieltjes integral permits the expression of many seemingly diverse results as
a single formula.
There is no requirement that the points x, be equally spaced. For each i = 1. 2.... , n.
set
Ax, = x, - x;_,,
which is equal to the length of the interval [x,x,].
Suppose f is a bounded real-valued function on [a, b]. Given a partition
9 = {xo, x...... of [a, b], for each i = 1.2.. ... n. let
in, = inf{f(t):x,_, <_ t <_ x,}.
M, = sup{f(t):x, _, 5 t s x,}.
Since f is bounded, by the least upper bound property, the quantities m, and M, exist in
R. If f is a continuous function on [a, b], then by Corollary 4.2.9. for each i there exist
points t s; E [x,_,.x,] such that M, = f(t,) and m, = f(sj.
The upper sum °U(9, f) for the partition 9 and function f is defined by
1110"D = Y M;ox,.
20"D = I,8=1MAX,
Since m, <_ M, for all i = 1, . . . , n, we always have
s
for any partition 91 of [a, b]. The upper sum for a nonnegative continuous function f is
illustrated in Figure 6.1. In this case °lt(2P, f) represents the circumscribed rectangular
approximation to the area under the graph off. Similarly, the lower sum represents the
inscribed rectangular approximation to the area under the graph off.
for any partition 9 of [a, b]. To see that inequality (1) holds, let iI' = {x°, ... , be
any partition of [a, b]. Since M; <- M for all i = 1, ... , n,
6.1.1 DEFINITION Let f be abounded real-valued function on the closed and bounded
interval [a, b]. The upper and lower Integrals off, denoted fQ f and fa f respectively,
are defined by
b
f = inf{°U,(9, f) : 91 is a partition of [a, b]),
rb
f = sup{.T(91, f) : 9' is a partition of [a, b]}.
Ja
Since the sets {° L (91j )j )} and {2(9,f)} are nonempty and bounded, the lower and
upper integrals of a bounded function f : [a, b] -+ R always exist. Our first goal is to
prove that f; f s f. f for any bounded real-valued function f on [a, b]. To this end, we
make the f!1owing definition.
But
Mk(X* - Xk-1) + Mi(xk - X*) MkAXA.
Therefore,
The proof for the lower sum is similar. If P* contains k more points than 91, we need
only repeat the above argument k times to obtain the result.
t
jaf :S: f
Proof. Given any two partitions 9, 9 of [a, b].
2(91,f) S 2(9 U 9.,f) S 1(9 U 9.,f) t
Thus 2(91,f) < °ll.(a f) for any partitions 9, 9.. Hence
6.1.5 DEFINITION Let f be a bounded real-valued function on the closed and bounded
interval [a, b]. If
then f is said to be Riemann integrable or integrable on [a. b]. The common value is
denoted by
Jb
or Jf(x)dx,
b
a f a
and is called the Riemann Integral or integral off over [a, b]. We denote by 9t[a, b]
the set of Riemann integrable functions on [a, b]. In addition, if f E 9t[a, b], we define
Jbaf -Jabs
212 Chapter6 The Riemann and Riemann-Stieltjes Integral
If f : [a, b) -> R satisfies m s f(t) M for all t e [a, b], then by inequality (1),
m(b-a)<- Jf
b sM(b-a).
a
In particular, if f(x) ? 0 for all x E [a, b], then fa f >_ 0. If f E 9t[a, b] is nonnega-
tive, then the quantity fa f represents the area of the region bounded above by the graph
y = f(x), below by the x-axis, and by the lines x = a and x = b.
6.1.6 EXAMPLES
(a) The following function, attributed to Dirichlet, is the canonical example of a func-
tion that is not Riemann integrable on a closed interval. Let f be defined by
11, x E Q,
f(x) = 0, x0Q.
Suppose a < b. If 9 = {xo, ... , x} is any partition of [a, b], then m; = 0 and
M; = 1 for all i = 1, ... , n. Thus
2(9,.f) = 0 and °1L(9,f) = (b - a)
for any partition 9 of [a, b]. Therefore,
fb
1a
f=0 and J f=(b-a),
a
0:5 x < 1
I Xk xn=1
2
Figure 6.2
Thus
n
x(91, f) = Axi = (1 - xk), and
i-k+t
ft
f f - ifk49-1f)s°uP1,f)
r1
s 2011f) + e c sup2(`9-,f)+e51 f+e,
0
where the infimum and supremum are taken over all partitions a of [0, 1). Since E > 0
was arbitrary, we have j' If = fo f. Thereforef is integrable on [0, 11 with f, f = 2.
214 Chapter 6 The Riemann and Riemann-Stieltjes Integral
(c) We now provide another example to illustrate how tedious even a trivial integral can
be if one relies only on the definition of the integral. Luckily, the fundamental theorem
of calculus (Theorem 6.3.2) will allow us to avoid such tedious computations. Let
f(x) = x, x E [a, b]. For the purpose of illustration we take a > 0 (Figure 6.3). Inter-
preting the integral as the area under the curve, we intuitively see that
tbx = -a)(b+a)=(b2-a22).
This is obtained from the formula for the area of a parallelogram. Let 91
{x0, x,, ... , be any partition of [a, b]. Since f(x) = x is increasing on [a, b],
m;=f(x;_,)=x;_, and M;=f(x;)=x;.
Figure 6.3
Therefore,
But
;_, 2
(x?-xi_i)= 2 (xn-4o)= 2 (b2-a2).
6.1 The Riemann Integral 21S
Finally, if we take the supremum and infimum over all partitions 9' of [a, b], we have
J6
xdxs2(b2-a2):5 f xdx.
a
This in itself does not prove that f(x) = x is integrable on [a, b] with
Jb
xdx = 2 (b2 - a2).
a
To prove that f is integrable on [a, b], we note that for any partition 9' of [a, b],
fb
rb
0c xdx - J x dx <_ alt(9i, f) - T(91,f) = I (xi - xi -1)Axi
a a i-1
(d) Consider the function f(x) = x2, x E [0, 1]. For n E Ni, let 9' be the partition
{0, ,'- 2, ... , 1}. Since f is increasing on [0, 1 ], its infimum and supremum on each
interval [Y,11 are attained at the left and right endpoint respectively, with
m, = (i - 1)2/n2 and Mi = i2/n2. Since Axi = 1/n for all i,
n
Using the identity 12 + 22 + + m2 = 6m(m + lX2m + 1) (Exercise 1, Section
1.3), we have
C1
i
216 Chapter6 The Rietnann and Riemann-Stieltles Integral
0C I f- Jf_<oL(P.f)_(f)<E.
b
a a
Therefore, °U.(9, f) - %(9,f) < e, which proves inequality (2). If D. is any refinement
of 9, then by Lemma 6.1.3,
0 <_ qt(L,f) - 2(a,f) :9 6u(P,f) - e(91,f) < E.
Thus inequality (2) is also valid for any refinement 2 of 9. Q
for all x, t E [a, b] with Ix - tl < S. Choose a partition 91 of [a, b] such that Ox; < S
for all i = 1, 2, ... , n. Then by inequality (3),
Mi - misrl
for all i = 1, 2, ... , n. Therefore,
=h (f (xi) - f (xi- 01
(b - a)
n [f(b) - f(a)].
Given e > 0, choose n E N such that
(b - a)
n
[f(b) - f(a)] < E.
For this n and corresponding partition 9, OIL.(91, f) - 2(91j) < e. Thus f is integrable
on [a, b].
6.1.9 THEOREM Let f be a bounded Riemann integrable function on [a, b] with Range
f C [c, d]. If ip is continuous on [c, d], then tp -f is Riemann integrable on [a, b].
Proof. Since 9 is continuous on the closed and bounded interval [c, d], ip is bounded
and uniformly continuous on [c, d]. Let K = sup{IQp(t)I: t E [c, d]}, and let e > 0 be
given. Set e' = e/(b - a + 2K).
Since ip is uniformly continuous on [c, d]. there exists 8, 0 < S < e', such that
19(s) - (P(01 < e' (4)
for all s, t E [c, d] with Is - tI < S. Furthermore, since f E 9%[a, b], by Theorem
6.1.7 there exists a partition 9 ' = {xo, . . . . . x}} of [a, b] such that
%(9',f) - _y(9', f) < S2.
218 Chapter6 The Rlemann and Riemann-Stieltjes Integral
6.1.10 COROLLARY If f E 9t[a, b], then If ! and f 2 are Riemann integrable on [a, b].
A natural question to ask is whether the composition of two Riemann integrable
functions is Riemann integrable. In Example 6.1.14(b) we will show that the answer is
an emphatic no!
6.1 The Riemann integral 219
Lebesgue's Theorem
In Theorem 6.1.8(a) we proved that every continuous function on [a, b] is Riemann in-
tegrable on [a, b]. By Exercise 16, this is also true for every bounded function on [a, b]
that is continuous except at a finite number of points. On the other hand, as a conse-
quence of Theorem 6.1.8(b), every monotone function on [a, bJ is Riemann integrable.
Hence, for example, if 1
is an enumeration of the rational numbers in [0, 1 ] and
c > 0 are such that T, c, converges, then by Theorem 4.4.10,
00
f(x) = 71 C. 1(x -
n=1
is monotone increasing on [0, 1 ], and thus is Riemann integrable on [0, 11. By Theo-
rem 4.4.10, the function f is continuous at every irrational number and discontinuous at
every, rational number in [0, 1].
We now state the beautiful result of Lebesgue that provides necessary and suffi-
cient conditions that a bounded real-valued function on [a, b) be Riemann integrable.
To properly state Lebesgue's result we need to introduce the idea of a set of measure
zero. The concept of measure of a set will be treated in detail in Chapter 10. The basic
idea is that the measure of an interval is its length. This is then used to define what we
mean by measurable set and the measure of a measurable set. At this point we only need
to know what it means for a set to have measure zero.
6 1.11 DEFINITION A subset E of P has measure zero if given any e > 0, there exists a
finite or countable collection {1n)n of open intervals such that
ECUJ and e,
n
6.1.12 EXAMPLES
(a) Every finite set E has measure zero. Suppose E = {x1, . . . , xN} is a finite subset of
R. For each n = 1, 2, ... , N, as in Figure 6.4 (with N = 6), let
/ E
Nxn+2N, .
Then
N N
ECU / and > l(!n) = e.
n=1 n=1
Therefore, E has measure zero.
it 14 12 13 16 l5
Xt
(0) (0)
X4 X2 X3 X6 XS
Figure 6.4
220 Chapter 6 The Riemann and Riemann-Stleltjea Integral
(b) Every countable subset of R has measure zero. Suppose E = {x,}00 I is a countable
subset of R. Let f > 0 be given. For each n E N. let
E E l
j l(In) = E 2n = E.
w=1 n=1
Remark. If f is continuous on [a, b], then clearly f satisfies the hypothesis of Theo-
rem 6.1.13 and thus is Riemann integrable. If f is a bounded function that is continuous
except at a finite number of points, then by Example 6.1.12(a) the set of discontinuities
off has measure zero. Hence f E 9t[a, b]. If f is monotone on [a, b], then by Corollary
4.4.8, the set of discontinuities off is at most countable, and thus by Example 6.1.12(b).
has measure zero. Hence again f E 9t[a, b].
6.1.14 EXAMPLES
(a) As in Example 4.2.2(g), let f be defined on [0, 1] by
11, x=0,
0, if x is irrational,
f(x) =
1n, if x = n in lowest terms, x * 0.
6.1 The Riemann Integral 221
Since f is continuous except at the rational numbers, which have measure zero, f is Rie-
mann integrable on [0, 1 ]. Furthermore, since $(P, f) = 0 for all partitions 9 of [0, 1 ],
Jf(x)dx = 0.
(b) Let f be the Riemann integrable function on [0, 1 ] given in (a), and let g : [0, 1 ] - R
be defined by
0, x0,
g(x) __ 11, x E (0, 1].
Since g is continuous except at 0, g E %[0, 1 ]. But for x E [0, 1 ],
(g of)(x) = J 1, if x is rational,
0, if x is irrational.
By Example 6.1.6(a), g of 6E Jt[O, 1).
EXERCISES 6.1
1. Let f (x) = 1 - x2, x E [ -1, 2]. Find Sf(9, f) and qt(9J) for each of the following partitions of [ -1, 2].
a. 9 = {-1,0, 1,2)
b.9={-1,-12,12,1,2,2}
2. Show that each of the following functions is Riemann integrable on [0, 2], and use the definition to find f02 I
1, 0, sx < 1
*a f(x) = 2, 15 x 5 2 ' b f (x) 3, 25 x<2
-2, sx 52
3. Show that each of the following functions are Riemann integrable
2 on [a, b] and find f; f.
0, a s x < c,
*a. f(x) = c (c a constant) b. f(x) = x = c.
2,
1, c<xsb
4. Use one of the methods of Examples 6.1.6 to find fo f for each of the following monotone functions f.
*a. f(x) = [3x], where (x) is the greatest integer function b. f(x) = x[2x]
c. f(x) - 2x + I d. f(x) = 1 - x2
5. Prove that f. *f = 0 for any real-valued function f on [a. a].
6 . *If f , g E 3t(a, b) with f ( x ) 5 g ( x ) f o r all x E [a, b], p r o v e that f f s f g.
7. a. Suppose f is continuous on (a. b] with f(x) z 0 for all x E [a, b]. If f. f = 0. prove that f(x) = 0 for all
x r= [a. b].
b. Show by example that the conclusion may be false if f is not continuous.
222 Chapter 6 The Rlemann and Riemann-Stieltjes integral
xdx= 3(b3-a3)(0:a<b).
Ja
11. Suppose f is monotone increasing on [a, b]. For n E N, set It = (b - a)/n. L e t 9n = {xo, x,, ... , xn} where for
each k = 0,. . ., n,xk = a + kh.
a. Prove that
(b a)
0 s Wt(9n,f) - f [f(b) -f(a)]
13. Let f be a bounded function on [a, b]. Suppose there exists a sequence {9n} of partitions of [a, b) such that
lim SC(9n, f) = lim QU (9n, f) = L (L E 18).
ntiPO n-wo
Prove that f is Riemann integrable on [a, b] with fa f = L.
14. a. If f e 9t[a, b], prove directly (without using Theorems 6.1.9 or 6.1.13) that If I E 9t[a, b].
b. If Ill E 9t[a, b], is f e 9t[a, b]?
15. *a. If f e 91[a, b], prove directly that f2 E 9t[a, b].
b. Give an example of a bounded function f on [a. b] for which f2 E 9t[a, b], but f e 9t[a, b].
16. *Suppose f is a bounded real-valued function on [a, b] that has only a finite number of discontinuities. Prove di-
rectly that f is Riemann integrable on [a, b].
17. a. If E has measure zero, prove that every subset of E has measure zero.
b. If E,, E2 have measure zero, prove that E, U E2 has measure zero.
ao
c. If each En, n = 1, 2, ... , has measure zero, prove that U E. has measure zero.
n=I
18. Use Theorem 6.1.13 to prove that if f, g r= 9t[a, b], then f + g E 9t[a, b].
19. Prove directly that the function f of Example 6.1.14(a) is Riemann integrable on (a, b).
20. Let f : [a, b] -+ R be continuous. Suppose that for every Riemann integrable function g : (a, b] -+ R the product
fg is Riemann integrable and f; fg = 0. Prove that f(x) = 0 for all x E [a, b].
21. Prove that the Cantor set P in [0. 1) has measure zero.
6.2 Properties of the Riemann Integral 223
Therefore,
b b fb
(f+g) < f+ ag+e.
f
Since the above holds for all e > 0,
b
jb(f+ g) _- jaf+ f bg
224 Chapter6 The Riemann and Riemann-Stieltjes Integral
jn(f + g) 2: J f + f g.
Thus the lower integral of (f + g) is equal to the upper integral of (f + g), and as a con-
sequence, (f + g) E 9t[a, b] with
fb(f+g)=
ff+ fhg.
(b) The proof of (b) is left as an exercise (Exercise 1).
(c) To prove (c), we first note that
fg = 4[(f+g)2- (f-g)2].
By part (a) the functions (f + g) and (f - g) are integrable on [a, b]. and by Corollary
6.1.10, (f + g)'- and (f - g)2 are integrable. Hence by parts (a) and (b),fg is integrable
on [a, b].
6.2.3 THEOREM Let f be a bounded real-valued function on [a, b], and suppose
a < c < b. Then f E 9t[a, b] if and only if f e 9t[a, c] and f E 9t[c, b]. if this is the
case, then
Jf=faf+ff (6)
Proof. Let c E R satisfy a < c < b. We first prove that if f is a bounded real-valued
function on [a, b], then
Suppose 9, and 92 are partitions of [a, c] and [c, b], respectively. Then 9 = 91 U 92
6.2 Properties of the ltiemann Integral 225
f
f-jaf+If
To prove the reverse inequality, let e > 0 be given. Then there exist partitions
91, and 91, of [a, c] and [c, b], respectively such that
t f:5 fa'+ Jf
c
which when combined with the previous inequality proves equation (7). A similar ar-
gument also proves
f f=a a
+ f
'ff f"c
. (8)
If f is integrable on [a, c] and [c, b], then by equations (7) and (8),
b
J af+jf=lfslf=Jf+Jf
Therefore, f E Jt[a, b] and identity (6) holds. Conversely, if f E 9t[a, b], then by
equations (7) and (8),
Jf+if=lf+If
226 Chapter 6 The Riemann and Riemann-Stleltjes Integral
Since the lower integral off is always less than or equal to the upper integral off, the
ff
above holds if and only if
f-ff
11 b
and
a u
f(p.f) = ,if(oaxi
I
is called a Riemann sum off cith respect to the partition 91 and the points {t,}.
In the Riemann approach to integration, one defines the integral of a bounded real-
valued function f as the limit of the Riemann sums off. Since 9(91,f) depends not only
on the partition °,D = {x0, xi,... , but also on the points t; E [x;_ 1, xi], we first
6.2 Properties of the Riemann Integral 227
need to clarify what we mean by the limit of the Riemann sums &'(9, f ). For a partition
_ {x0, x1.... , xn} of [a, b]. set
I1911 max{oxi: i = 1,2,...,n}.
The quantity Il9 11 is called the norm or the mesh of the partition 9.
for all partitions 9 of [a, bl with 11911 < S, and all choices of ti E [x, _ I, xi].
then f E 9t[a, b] and fa f = 1. Conversely, if f E 9t[a, b], then 111im0 &(91,f) exists and
el-
rb
lim x(9,.1) = J f.
11911-.0
a
Proof. Suppose , la tprt0 P(9, f) = I. Let e > 0 be given, and let S > 0 be such that
inequality ( 9 ) holds f o r all partitions 9 = {x0, xl, ... , xn} of [a, b) with 11211 < S,
and all ti E [xi_ 1, xi]. By the definition of Mi, for each i = 1, ... , n, there exists
i E [xi_ 1, xi] such that f M, - e. Thus
n
OU(911f) = M4xi
i=1
< f(OAXi + E
//
AX,
i=1 i.l
< 1+ E + e(b -a]= 1+ E[ 1 + b -a].
Similarly T(21,f) > I - E[ 1 + b - a]. Therefore,
°U.(9, f) - 2(91,f) < 2E[ 1 + b - a].
Thus as a consequence of Theorem 6.1.7, f E 9t[a, b) with fa f = I.
Conversely, suppose f r= Jt[a, b]. Let M > 0 be such that I f(x)I s M for all
x E [a, b]. Let e > 0 be given. Since f E 9t[a, b], by Theorem 6.1.7 there exists a
partition . of [a, b] such that
Jb
Suppose % = {xa, . . . , x,,,}. Let S = e/NM, and let 9 = ( ) , 0 ,...,y . ) be any parti-
tion of [a, b] with 11 9 II < S. As in the definition of the integral, let
M;=sup{f(x):xE[x;_1,x;]}, i= I,. . .. N.
Consider any interval [yk_ 1, yk], k = 1, . .. , n. This interval may or may not contain
points x. E a. Since 9 . contains N + I points, there are at most N - I intervals
[y,-,, yk] which contain an x, E a, i # 0, N. Suppose, as in Figure 6.6, {.x t, ... , xj+m}
C [Yk- 1, Yk] [f xx * yk_ 1, set M k = sup{ f (x) : x E [yk_,, xj]}. Similarly, if xl+m * yk,
set Mk = sup{f(x) : x E [x, +m, y'k]}.
Yk - 1 Y4
X1 xj+1 x142 X1
Figure 6.6
Let tk E [yk_ 1, Yk] be arbitrary. Since 1(t) - f(s)I s 2M for all t, s E [yk_,, yk],
f(tk) 2M + M1 s, s = 1, . .. , m, and
f(tk) 2M + Mk, i = 1, 2.
Therefore,
m
f(9j) _ k=1
f(tk)DYk < 2M(N - 1)8 +
k-1
6u(9k,f)
b
<2e+04(9',f)<2e+Ru(a,f)<3e+ f. f.
a
herefore,
Since this holds for any partition QP _ {yo, ... , y,,} of [a, b] with 11 1,1P11 < a and any
choice of tk e k = 1, . . . , n,
rb
6.2.7 EXAMPLE In this example we will use the method of Riemann sums to evaluate
f; x dx. Since f(x) = x is Riemann integrable on [a, b],
rb
x dx = lim 9(91, f).
Ja
Since the limit exists for any t, E [x1_,, x,], we can take t; = (x;_, + x;)/2. With this
choice of t,,
1, R T R 2 2
fyfy /!11
"(.7,f) 7,(x,_, + xj)(x,
2j_j
1 - x,_,) 2 (x, - x?2 (b - Q ),
1
i -11
EXERCISES 6.2
1. Prove Theorem 6.2.1(b).
2. a. Use the method of Riemann sums to evaluate f; x2 d,
b. Use the method of Riemann sums to evaluate fa x" dx, n E N, n >_ 3.
3. a. Let fbe a real,valued function on [a, b] such that f(x) = 0 for all x # c, , ... , cR. Prove that f E A[a, b]
with f; f = 0.
*b. Let f, g E.9t[a, b] be such that f(x) = g(x) for all but a finite number of points in ra, b]. Prove that
faf=fog.
c. Is the result of part (a) still true if f (x) = 0 for all but countably many points in [a, b]?
4. Let f E 9t[-a, a], a > 0. Prove each of the following.
a. If f is even (i.e., f (-x) = f(x) for all x (E [ -a, a]), then f f = 25f.
b. Iff is odd (i.e.,f(-x) _ -f(x) for all x E [-a, a]), then f f = 0.
5. *Let f be a bounded real-valued function on [a, b] such that f E k[c, b] for every c, a < c < b. Prove that
fE f; f= lira, f b f
6. Let f be continuo6s on [0, 1]. Prove that
lim f(jf(x)dx.
7. Use the previous exercise to evaluate each of the following limits. (You may use any applicable methods from cal-
culus to evaluate the definite integrals.)
R R R
1 k n
a. lim
R-"OnYk k' b. lim Y c. lim
i n2 k2 R-rook=, n2 + k2
230 Chapter 6 The Riemann and Riemann-Stieltjes Integral
x = n/(n + 1), n E N. Find fo f(x) dx. Leave your answer in the form of an infinite series.
9. Suppose f E 9t[a, b] and c E R. Define f, on [a - c. b - c] by f.(x) = f(x + c). Prove that
ffEdt[a-c,b-c]with
Jf(x)ds = I f(x) dx.
6
10. Let f. g, h be bounded real-valued functions on [a. b] satisfying f(x) s g(x) S h(x) for all x E [a. b]. If
f. h E Il[a. b] with f. f = fu h = 1, prove that g E 'lt[a. b] with ff g = 1.
6.
I Fundamental Theorem of Calculus
In this section we will prove two well-known and very important results. Collectively
they are commonly referred to as the fundamental theorem of calculus. To Newton and
Leibniz, integration was the inverse operation of differentiation. In Leibniz's notation,
this result would simply be expressed as f f'(x) dx = f (x), where the symbol f (denot-
ing sum) was Leibniz's notation for the integral. This notation is still used in most cal-
culus texts to denote the indefinite integral of a function. Since the modern definition of
the integral is based on either Riemann or Darboux sums, we now need to prove that in-
tegration is indeed the inverse operation of differentiation. Both versions of the funda-
mental theorem of calculus presented here are essentially due to Cauchy, who proved
the results for continuous functions.
As was illustrated in Examples 6.1.6 and 6.2.7, the computation of the integral of
a function, using either Darboux's or Riemann's definition, can be extremely tedious.
For nontrivial functions, these computations are in most instances impossible. The first
version of the fundamental theorem of calculus provides a major tool for the evaluation
of Riemann integrals. We begin with the following definition.
Since
a
i=1
we have
Ibf(x)dx=F(b)-F(a). U
a
6.3.3 EXAMPLES
(a) If f (x) = x", n E N, then F(x) _ + 1
X"11 is an antiderivative off. Thus for any
a,bER,
fb
xZ sin X, x # 0,
F(x) =
f=F(1)-F(0)=sin1.
J
Thus
x, 0sxs1,
F(x) 2x2-x+2, 1 <x<-2.
The graphs off and F are given in Figure 6.7. Even though f is not continuous at x = 1,
the function F is continuous everywhere on [0, 2]. This in fact is always the case, as
will be proved in Theorem 6.3.4.
Proof. The proof that F is continuous is left as an exercise (Exercise 1). Suppose f is
continuous at c E [a, b). We will show that F' (c) = f (c). Let h > 0. Then by Theo-
rem 6.2.3,
c+h c c+h
F(c + h) - F(c) J f(t) dt - f(t) dt = { f(t) A
f o c
Therefore,
h
f" [f(t) - f(c)) dt.
Let E > 0 be given. Since f is continuous at c, there exists a S > 0 such that
NO - f(c) I < E
f o r all t, I t - c I < S. Therefore, if 0 < h < S,
F(c + h) - F(c) - f(c) I
Jc+h
h
If(t) - f(c)I dt
c
h Jc +h
<h Edt=E.
c
Thus
i.e., F+' (c) = f (c). Similarly, if f is continuous at c E (a, b], F'_ (c) = f (c), which
proves the result. Q
Remarks
(a) If f is continuous on [a, b], then an antiderivative of f always exists; namely, the
function F given by
Since f is continuous, F'(x) = f(x) for all x E [a, b]. As a consequence, we obtain
the following more elementary version of Theorem 6.3.2 normally encountered in the
study of calculus: If f is continuous on [a, b] and G is any antiderivative of f, then
f f = G(b) - G(a).
234 Chapter6 The Riemann and Rlemann-Stieltjes Integral
(b) Integrability of a function f on [a, b] does not imply the existence of an antideriv-
ative off. For example, if f is monotone increasing on [a, b] and F(x) = f` f, then for
any c E (a, b), F' (c) = f (c+) and F_(c) = f(c-) (Exercise 14). Thus if f is not con-
tinuous at c, the derivative of F does not exist at c.
L(z) = dt.
i
Since f (t) = 1/t is continuous on (0, oo), by Theorem 6.3.4, L(x) satisfies L'(x) = l/x
for all x > 0. Furthermore, since L'(x) > 0 for all x E (0, oo), L is strictly increasing
on (0, oo).
We now prove that the function L(x) satisfies the usual properties of a logarithm
function; namely,
(a) L(ab) = L(a) + L(b) for all a, b > 0,
(b) L(e) _ -L(b), b > 0, and
(c) L(b') = rL(b), b > 0. r E R.
To prove (a), consider the function L(ax), x > 0. By the chain rule (Theorem 5.1.6),
d L(ax) =
dX
.a=X=L'(x).
ax
Thus by Theorem 5.2.9, L(ax) = L(x) + C for some constant C. From the definition of
L we have L(l) = 0. Therefore,
L(a) = L(1) + C = C.
Hence L(ax) = L(a) + L(x) for all x > 0, which proves (a). The proof of (b) proceeds
analogously. It is worth noting that for the proof of (a) and (b) we only used the fact
that L'(x) = l/x and L(1) = 0.
To prove (c), if n E N, then by (a) L(b") = nL(b). Also by (b),
Our final step will be to prove that L(e) = 1, where e is Euler's number of Exam-
ple 2.3.5. To accomplish this we use the definition to compute the derivative of L at 1.
Since L'(1) exists,
L(l + L(1)
1 =L'(1)= ]im
n--,ao
i
= 1mnL(l+nl
= lim L(t I + !T) = L(e).
n
The last equality follows by the continuity of L and the definition of e. Therefore,
L(e) = 1 and the function L(x) is the logarithm function to the base e. This function is
usually denoted by log, x or In x, and is called the natural logarithm function.
6.3.6 THEOREM (Mean Value Theorem for Integrals) Let f be a continuous real-valued
function on [a, b]. Then there exists c e [a, b] such that
b
f =f(c)(b - a).
fa
Proof. Let F(x) = fa f. Since f is continuous on [a, b], F'(x) = f(x) for all x E [a, b].
Thus by the mean value theorem (Theorem 5.2.6), there exists c E [a, b] such that
An alternative proof of the above can also be based on the intermediate value the-
orem using the continuity of f. This alternative method will be used in the proof of the
analogous result for the Riemann-Stieltjes integral.
'Proof. Since f, g are differentiable on [a, b], they are continuous and thus also inte-
grable on [a, b]. Therefore by Theorem 6.2.1(c), fg' and gf' are integrable on [a, b].
Since
(fgY = gf' + fg',
236 Chapter 6 The Riemann and Riemann-Stieltjes Integral
the function (fg)' E 9t[a, b]. By the fundamental theorem of calculus (Theorem 6.3.2),
b Vg), b b
Proof. Since tP is continuous, I = rp([a, b]) is a closed and bounded interval. Also,
since f a rp is continuous and tp' E 9t[a. b], by Theorem 6.2.1(c), (f o W)tp' E 9t[a, b].
If I = r([a, b]) is a single point, then tp is constant on [a. b]. In this case rp'(t) = 0 for
all t and both integrals above are zero. Otherwise, for x E I define
Remark Another version of the change of variable theorem is given in Exercise 10.
6.3.9 EXAMPLE To illustrate the change of variable theorem, consider fo t/(1 + t2) di.
If we let (,p(t) = 1 + t2 and f(x) = 1/x, then
Joe
dt = 2 102f(`p(t))(p'(t) dt
1 + t2
EXERCISES 6.3
1. *Let f E 9t[a, b]. For x E [a. b], set F(x) = f' f. Prove that F is continuous on [a, b].
6.3 Fundamental Theorem of Calculus 237
2. For x E [0, 1 ], find F(x) = fo f(t) dt for each of the following functions f defined on [0, 11. In each case verify
that F is continuous on [0, 11. and that F'(x) = f(x) at all points where f is continuous.
1, 0 s x <2
a. f(x)=xz-3x+5 *b. f(x) =
-2, 1 5 x <-1
c. f(x) = x - [3x] d. f(x) = x[3x]
Find H'(x).
& Let f : R -+ R be continuous. For a > 0 define g on R by
10. Prove the following change of variables formula: Let rp be a real-valued differentiable function on [a, b] with
(x) # 0 fof all x E [a, b]. Let 0 be the inverse function of rp on I = tp([a. b]). If f : 1-+ R is continuous on 1, then
(o(b) f(t)4x'(t)dt.
. f(rp(x))dx = -
a )v(a)
238 Chapter6 The Riemann and RIemann-Stieltjes Integral
a I
2 In x
X
dx
+b (4
iI vx
dx
(I x
c. J x In x d x d. J t at + b dr. a, b> 0
0 0
I
1/X
f j4
e, 10 dx dx
1+V I x
13. Suppose f : (a. b] -> R is continuous. Let M = max{jf(x)j: x E [a, b]}. Show that
/b
tim I If(x)r dx = M.
14. Let f be a monotone increasing function on [a, b] and let F(x) = fo f(t) dt. Prove that F' (c) = f(c+) and
F_(c) = f(c-) for every c E (a, b).
15. Use Theorem 4.4.10 and the previous exercise to construct a continuous, increasing function F on [0, 1) that is
differentiable at every irrational number in [0, 11, and not differentiable at any rational number in (0, 1 ].
16. Cauchy-Schwarc Inequality for Integrals: Let f, g E &(a, b]. Prove that
(jb
J,bf(x)g(x) dx I 2 5 (Lb g2(x) dl/
f 2(x) dx/
L(x) f!df.
I t
f of
- hf I = IJ f I < Ja` U1 < M(c - a).
As a consequence, if f E lt[a, b], then f E k[c, b] for every c E (a, b) and
e n
limf f=Jf.
C a
lim
C a- J
o f= f.
f a
6.4.1 DEFINITION Let f be a real-valued function on (a, b) such that f E dt[c, b] for even
c E (a, b). The improper Riemann integral off on (a, b], denoted f,' f, is defined to be
b b
limf f,
Ca'
Ja f
provided the limit exists. If the limit exists. then the improper integral is said to be con-
vergent. Otherwise, the improper integral is said to be divergent.
240 Chapter6 TheRlemann and Riemann-Stleltjes Integral
To evaluate lim In c we consider In r", where 0 < r < I and n E N. Since In r" _
C- Oo
n In r and In r < 0,
lim In r" = limn In r = -oo.
Therefore, since r" _+0 as n -+ oo and In x is monotone increasing, lirm In c = -oo.
Thus the improper integral of 1/x on (0, 1) diverges.
(b) For our second example we consider the improper integral of L(x) = In x on (0, 1 ].
Since L(x) is continuous on (0, 1), L E 9t[c, 1 ] for every c E (0, 1). Consider ff' In x dx.
If we take g(x) = x, then by the integration by parts formula,
j In x dx = L(x)g'(x) dx
J
=-cInc-1+c.
By the substitution c = 1/t, t > 0, and l'Hospital's rule,
-1n t
lira c In c = lim
1-.0o t
= -lim 1 = 0.
t-00 t
Therefore,
lim I In x dx = -1.
c-+O'
C
I lnxdx = -1.
0
6.4 Improper Riemann Integrals 241
Jldx
x ,l° z
Thus the improper integral off converges on (0, 1]. However, f2(x) = t/x, and the im-
proper integral of 1/x on (0, 1 ] diverges. Also, if f E 9t[a, b], then [f I E 9t[a, b]. This
.again is false for improper Riemann integrals. In Exercise 4, we provide an example of
a function f for which the improper integral off converges, but the improper integral of
L f I diverges.
Infinite intervals
We now turn our attention to functions defined on infinite intervals.
6A.3 DEFINITION Lei f be a real-valued function on [a, co) that is Riemann integrable on
[a, c) for every c > a. The Improper Rlemann integral off on [a. oo), denoted f, f.
is defined to be
j= lim
C-400
J f,
n
provided the limit exists. If the limit exists, then the improper integral is said to be con-
vergent. Otherwise, the improper integral is said to be divergent.
f f bf,
provided the limit exists. If f is defined on (-oo, oo), then the improper integral off is
defined as
xf+I 30f,
242 Chapter 6 The Riemann and Riemann-Stieltjes Integral
for some fixed p E R, provided that the improper integrals off on (-oo, p] and [ p, oo)
are both convergent. For a function f defined on (-oo, oo), care must be exercised in
computing the improper integral off. It is incorrect to compute
Remark. If f is nonnegative on (a, oo) with f E Jt[a, c] for every c > a, then f.f
is a monotone increasing function of c on [a, oo). Thus lim f. f exists either as a real
number or diverges to no. For this reason, if f (x) 0 for C
all x E [a, oc), we use the notation
rb
Jf(x)dx < oo or J f(x) dx = 00
to denote that the improper integral off on [a, oo) converges or diverges to no respec-
tively.
6.4.4 EXAMPLES
(a) Let f (x) = 1/x2, x E [ 1, oo). Since f is continuous on [ 1, co), f e Jt[ 1, c) for
everyc > 1. Therefore,
Ji=l c-+ao Ix
imf'Idx=liml-1+1)=1.
c-.oo\ C
Isin xl
IfI = r =oo.
6.4 Improper Riemann Integrals 243
For n E N, consider
(k +3/4h. Isin
(k+I)" {sin xf
>
x J(k+1/4hr x
(k+3/4)n Isin x / ll ll
x Ia 22(k+1)rr
k+ 1'
and as a consequence,
"+
sin xl 1,/2
f x
dx 4 k_ik+l
By Example 2.7.4, the series 7,k 1 k diverges. Therefore,
As the previous example shows, the convergence of the improper integral off does
imply the convergence of the improper integral of If 1. If f is a real-valued function
on [a, oo) such that f e ,[a, c] for every c > a and the improper integral of If I con-
verges on (a, oo), then f is said to be absolutely integrable on [a, oo). An analogous
definition can be given for unbounded functions on a finite interval. We leave it as an
exercise to prove that if f is absolutely integrable on [a, oo), then the improper integral
off also converges on [a, oo) (Exercise 5).
We conclude this section with the following useful comparison test for improper
'integrals.
244 Chapter 6 The Riemann and Rlemann-Stleltjes Integral
(x) dx :5 J g(x)dx.
J
N EXERCISES 6.4
1. For each of the following functions f defined on (0, 1), determine whether the improper integral off converges. If it
converges, find fo f.
Inxx
'a f(x) = Xp> 0 < p < I 1> f(x) _ x c. f(x) =
°O In.r °D dx °G dx
d. X ax *e. E x l n x P> 1
2 2
x(l n x)P"
f'xZ+,dx f IV
g. h. f(x2+ 1)pdx. P> I I. 1)dx
(x2+ 1)(x+
3. For each of the following, determine the values of p and q for which the improper integral converges.
Show that the improper Riemann integral off converges on (0, I ], but that the improper integral of VI diverges on
(0,1 1.
5. *If f is absolutely integrable on [a. oo) and integrable on [a, c] for every c > a, prove that the improper Riemann
integral off on [a, oo) converges.
6. Prove Theorem 6.4.5.
6.5 The Riemann-Stieltjes Integral 245
6.5.1 EXAMPLE Consider n-masses, each of mass m;, i = 1, ... , n, located along the
x-axis at distances r, from the origin with 0 < rt < < r (Figure 6.8). The mo-
ment of inertia 1, about an axis through the origin at right angles to the system of
masses, is given by
1 = ; r?m;.
0 HIM
Figure 6.8
3. Since the results of this section are not specifically required in subsequent chapters, this topic can be omit-
ted on first reading of the text.
246 Chapters The Riemann and Riemann-Stieltjes Integral
On the other hand, if we have a wire of length I along the x-axis with one end at the ori-
gin, then the moment of inertia 1 is given by
! = jxp(x)dx.
where for each x E [0, 1). p(x) denotes the cross-sectional density at x.
Although these two problems are totally different, the first being discrete and the
second continuous, the Riemann-Stieltjes integral will allow us to express both of these
formulas as a single integral. In the definition of the Riemann integral, we used the
length Axi of the ith interval to define the upper and lower Riemann sums of a bounded
function f. The only difference between the Riemann and Riemann-Stieltjes integral is
that we replace Axi by
Aai = a(xi) - a(x; _ 1),
where a is a nondecreasing function on [a, b]. Taking a(x) = x will give the usual
Riemann integral. Although the modification in the definition is only minor, the
consequences are far-reaching. Not only will we obtain a more extensive theory of
integration, but also an integral that has broad applications in the mathematical
sciences.
kt(9, f, a) _ M;Aa,.
r=
Similarly, the lower Riemann-Stleltjes sum off with respect to a and the partition 9,
denoted 2(9, f, a), is defined by
-Y(9, f, a) _ miAai.
i=1
Since m; s M, and Aai ? 0, we always have 2(91,f a) < all(9, f, a). Further-
more, if m s f(x) s M for all x E [a, b], then
m[a(b) - a(a)] <_ Z(9, f, a) 5 Rt(9, f, a) < M[a(b) - a(a)) (10)
6.5 The Riemann-Stieltjes Integral 247
for all partitions 9 of [a, b]. Let 96 be any partition of (a, b). Since Mi s M for all i and
Aai ? 0,
n
Thus °U,(9), f, a) <_ M[a(b) - a(a)]. The other inequality follows similarly. In the above
we have used the fact that
n
By inequality (10) the set {qL(91, f, a) : 9' is a partition of [a, b]} is bounded below, and
thus the upper integral off with respect to a exists as a real number. Similarly, the lower
sums are bounded above, and thus the supremum defining the lower integral is also
finite. As for the Riemann integral, our first step is to prove that the lower integral is less
than or equal to the upper integral.
6.5.2 THEOREM Let f be abounded real-valued function on [a, b], and a a monotone
increasing function on [a, b]. Then
Jfda.
e
Jabfda:5
Proof. As in the proof of Lemma 6.1.3, if 9'* is a refinement of the partition 9, then
X(91,f a) <_ X(91*,f, a) :5 gL(96*,f a) c ah(y,f, a).
Thus if 91, 2, are any two partitions of [a, b],
f, a) s 2(9 U a j a) :S %(91 U ll, f, a) < 61L(,, f, a).
Therefore T(91, f, a) <_ 6tL(9;, f, a) for any partitions 9, 9.. Hence
Jfda = sp96,f,a) a)
a
for any partition 9.. Taking the infimum over a gives the result. O
248 Chapters The Riemann and Riemann-Stieltjes Integral
6.5.3 DEFINITION Let f be a bounded real-valued function on [a, bl, and a a monotone
increasing function on [a, b). If
h h
fda = Jfda.
f
As was indicated previously, the special case a(x) = x gives the usual Riemann in-
tegral on [a, b).
6.5.4 EXAMPLES
(a) Fix a < c 5 b. As in Definition 4.4.9, let 1,.(x) = I (x - c) be the unit jump func-
tion at c defined by
= f0, x<c,
1`(x) t 1, x?c.
We now prove the following: If f is a bounded real-valued function on [a, b] that is con-
tinuous at c, a < c < b, then f is integrable with respect to 1, and
h 'f(x)
1 f dl,. = dl (x - c) = f (c).
a Ja
For convenience, we set a(x) = 1,.(x), which is clearly monotone increasing on [a, b).
Let 91 = {xo, x1..... x,} be any partition of [a, b]. Since a < c b, there exists an
index k, 1 5 k 5 n, such that
Xk- I < C 5 xk.
Then
Da* = a(xk) - a(xk_1) = I - 0 = 1, and
Aa;=0, foralli#k.
Therefore,
f o r all t E [a, b] with I t - cI < S. If 91 is any partition of (a, b] with x, - xx_, < S
for all j, then
AC) - E S Mk c Mk :5 AC) + E.
Therefore, f (c) - E S Y(91, f , a) : 5U(91, f, a) <_ f(c) + e. As a consequence.
Since e > 0 was arbitrary, the upper and lower integrals off are equal, and thus f is in-
tegrable with respect to a on [a, b] with
Jb
f da = f (c).
Remark. It should be noted that in Example 6.5.4(a) only left continuity off at c is
required (Exercise 2(a)).
The following theorem, which is the analogue of Theorem 6.1.7 for the Riemann
integral, provides necessary and sufficient conditions for the existence of the Riemann-
Stieltjes integral. The proof of the theorem follows the proof of Theorem 6.1.7 verba-
tim and thus is omitted.
6.5.5 THEOREM Let a 'be a monotone increasing function on [a, b]. A bounded real-
valuedfunction f is Riemann-Stieltjes integrable with respect to a on [a, b] if and only
if for every e > 0; there exists a partition 9 of [a, b] such that
a) - a) < E.
Furthermore, if 9 is a partition of [a, b] for which the above holds, then the inequality
also holds for all refinements of P.
We now use the previous theorem to prove the following analogue of Theorem
6.1.8. Except for some minor differences, the two proofs are very similar.
250 Chapter 6 The Riemann and Riemann-Stieltjes Integral
Proof. (a) The proof of (a) is identical to the proof of Theorem 6.1.8(a) except that
given e > 0, we choose 71 > 0 such that
[a(b) - a(a)),q < e.
The remainder of the proof now follows verbatim the proof of Theorem 6.1.8(a).
(b) For any positive integer n, choose a partition 91 = {x°, x,.... , of [a, b]
such that
[a(b) - a(a))
n U(b) - f(a)).
For this n and corresponding partition 91, °ll(?, f, a) - f, a) < e, which proves
the result. Q
Remark. In part (b) above, the result may be false if a is not continuous. For exam-
ple, if a < c < b, then the monotone function 1, is not integrable with respect to 1, on
[a, b] (Exercise 2b). The article by Kenneth Ross, listed in the Supplemental Read-
ing section, suggests an alternative approach to the Riemann-Stieltjes integral that
eliminates this difficulty.
6.5.7 DEFINITION For a given monotone increasing function a on [a, bJ, 9R(a) denotes
the set of bounded real-valued functions f on [a. b] that are Riemann-Stieltjes inte-
grable with respect to a.
6.5.8 THEOREM
(a) If f, g E gi(a), then f + g and cf are in %(a) for every c E R and
Jb b b
b b
cf da = c fda.
a I
(b) If f E 9t(a;), i =Jfd(ai
1, 2, then f E 9R(a1 + a2) and
+ a 2) = Jfdai + Jfda2.
(c) If f E °Jt(a) and a < c < b, then f is integrable with respect to a on [a, c]
and [c, b] with
b b
(d) If f, g E t(a) with f (x) <_ g(x) for all x E [a, b), then
rb rb
Ja f da Ja g da.
Proof. We provide the proofs of (b) and (e). The proofs of (a), (c), and (d), along with
other properties of the Riemann-Stieltjes integral. are left to the exercises.
(b) Since f E %(a;), given e > 0, there exists a partition 9D,, i = 1, 2, such that
Therefore by Theorem 6.5.5, f E 9t(al + a2). Furthermore, for any partition 91 of [a, b),
2(91j, a, + a2) _ ZP,f al) + £(9j a2)
t((b rb
Jfda,+Jfda2
a a
f f d(a, + a2) =
I f da, + f f dal.
(e) Suppose f E 9t(a) and 91 _ {x°, x,. ... . xa} is a partition of [a, b]. For each
i= 1,. ..,n,let
M; = sup{f(t): t E [xi_,,xi]}, M* = sup{I f(t)I:t e [xi_,,xi]},
mi = inf{f(t): t e [xi_,,x.]}, m* = inf{I f(t)I : t E [xi_,,xi]}.
If t, x E [xi_,, xi], then
II f(t) - If(x)II 1f(t) -f(x)I <_Mi - mi.
Thus M*i - m * 5 Mi - mi, for all i = 1, . .. , n, and as a consequence,
°u(91, If 1 , a) - ,x(91, I f I, a) s tlL(91, f, a) - X(Ol, f, a).
Therefore by Theorem 6.5.5, If I E 9t(a). Choose c = t l such that c f f da -e 0. Then
= M [a(b) - a(a)].
As for the Riemann integral, we also have the following mean value theorem and
integration by parts formula for the Riemann-Stieltjes integral.
Proof. Let m and M denote the minimum and maximum off on [a. b} respectively.
Then by Theorem 6.5.8(d),
b
m[a(b) - a(a)] s
f fda <_ M[a(b) - a(a)].
a
If a(b) - a(a) = 0, then any c E [a, b] will work. If a(b) - a(a) # 0, then by the in-
termediate value theorem there exists c E [a, b) such that
b
6.5.10 THEOREM (Integration by Parts Formula) Suppose a and Pare monotone increasing
functions on [a, b]. Then a E 91.(/3) if and only if fl E 9t(a). If this is the case,
Therefore,
11[(91, a, /3) - 201, a, /3) = 1401, /3, a) - 49. is, a).
From this identity it immediately follows by Theorem 6.5.5 that a E 9t(t) if and only
if /3 E M(a). Furthermore, if /3 E 9t(a), then given e > 0, there exists a partition 9' of
[a. b] such that
Hence,
Jb/3
A similar argument using the lower sum proves the reverse inequality.
254 chapter 6 The Riemann and Riemann-Stieltjes Integral
We conclude this section with two results that represent the extremes encountered
in Riemann-Stieltjes integration. As in Example 6.5.4(a), let I(x - c) be the unit jump
function at c E R. Suppose {s"},N,- I is a finite subset of (a, b] and are nonneg-
ative real numbers. Define the monotone increasing function a on [a, b] by
N
a(X) _ I cn I (X - Sn).
n-1
fa
f da =
N 1bN
If f is continuous on [a, b], then by Example 6.5.4(a) and Theorem 6.5.8(b),
c"
1
f(x) dl (x - sn) = "'1
I Cnf (sn) (12)
n=1 a
Since 0 : I (x - sn) <_ 1 for all n, the series in equation (13) converges for every
x E [a, b], and a is a monotone increasing function on [a, b]. For such a function awe
have the following theorem.
6.5.11 THEOREM Let f be a continuous real-valued function on [a, b], and let a be the
monotone function defined by equation (13). Then f E Jt(a) and
rb oo+
( f da = G c"f(s")
fa n=1
Proof. Since f is continuous on [a, b], f E lt(a) (Theorem 6.5.6(a)). Let e > 0 be
given. Choose a positive integer N such that
00
I C. < e.
n-N+1
Let M = max{I f(x)I : x (=- [a, b]). Then by Theorem 6.5.8(b) and (e),
W
:5 Mlc,,<Me,
n=x+i
which proves the result.
At the other extreme, if the monotone function a is also differentiable, then we
have the following result.
J f da = J f(x)a'(x) dx.
a a
Proof. Since both f and a' are Riemann integrable on [a, b], by Theorem 6.2.1(c),
fa' E 9t[a, b]. Let e > 0 be given. Since a' E 01[a, b], by Theorem 6.1.7 there ex-
ists a partition 91 of (a, b) such that
°U.(9, a') - Y(), a') < e. (14)
Let °. 2. _ {xo, ... , xn} be any refinement of 91. As in Theorem 6.2.6, for each i =
1, ... , n, we can choose s, E [xi_,, x,] such that
By the mean value theorem, for each i = 1, ... , n, there exists t; E [xi _,, x,] such that
Dai = a(x;) - a(x,_1) = a'(ti)Axi.
Therefore,
n a
Let M = sup{ f(x)I :x E [a, b]}, and for i = 1, ... , n, let in, and M; denote the in-
fimtun and supremum respectively of a' over the interval [x_ ,, x,). Then
la'(si)-a'(t;)l:!-- Mi - m,
f o r all i = 1, ... , n. Therefore,
f(si)a'(ti),xi - In As;)a'(sAXi`
i 1 i-l i- 1f(si)I1a'(1i) - a'(sMAxi
`M(Mi - mi)ixi
= M(%(a, a') - -W(9, a')) < Me.
The last inequality follows by inequality (14) since .. is a refinement of 91. Therefore,
n
f(s;)a'(ti)Ax; S
i-i i- f(si)a'(s;)Ax, + Me <_ °U.(9, fa') + Me.
256 Chapter 6 The Riemann and Rlemann-Stieltjes Integral
Thus
b
rh
f da < fa' + (M + 2)e.
t
Since this holds for any E > 0, fQ f da <_ f' fa'. A similar argument using lower sums
proves the reverse inequality. Thus f E J1(a) and
h b
J fda = I fa'.
a 1a
6.5.13 EXAMPLES
(a) First, we illustrate the finite version of Theorem 6.5.11; namely, identity (12). Consider
j exd(x].
[x]=1(x- 1)+1(x-2).
Therefore since ex is continuous, it is integrable with respect to [x], and by identity (12),
J exd[x] = + e2.
(b) To illustrate Theorem 6.5.12, if a(x) = x2 on [0, 1), then for any Riemann inte-
grable function f on [0, 1 ],
I
_ -- ITcos V + sin Vx U
?r
13[x] 2 J ;[x]e2x dx
0 0
6.5.14 EXAMPLE In this example we will show that both formulas of Example 6.5.1 can be
expressed as
I
I= x dm(x),
0
where m(x) denotes the mass of the wire or system from 0 to x. Clearly the function
m(x) is nondecreasing, and thus since x2 is continuous, the above integral exists.
In the first case, since 0 < rt < r2 < < r,,, and the masses m; are located at
r, (see Figure 6.8), m(x) is given by
m(x) m; l (x - r;).
x2dm(x) _ r?m;.
Io
On the other hand, if m(x) is differentiable and m'(x) is Riemann integrable, then by
Theorem 6.5.12,
It only remains to be shown that m'(x) is the density. The density p(x) of a wire (mass
per unit length) is defined as the limit of the average density. In the interval [x, x + Ax,
the average density is
m(x + Ax) - m(x)
Ax
Thus if m(x) is differentiable, p(x) = m'(x).
Rlemann-Stieltjes Sums
We conclude this section with a few remarks concerning Riemann-Stieltjes sums. As in
Definition 6.2.4, let f be a bounded real-valued function on [a, b], a a monotone in-
creasing function on [a, b], and 9' _ {xo, ... , a partition of [a, b]. For each
i = 1, 2, . , n, choosey E [x,_,, x;]. Then
F(`.'P, f a) = Ef(t,)Aa,
The result where f is continuous and a is monotone increasing is given in Exercise 10.
EXERCISES 6.5
1. *Evaluate f ' f (x) da(x) wheref is bounded on (-1. 11 and continuous at 0, and a is given by
-1, x<0,
a(x) = 0, x = 0,
1, x > 0.
2. a. In Example 6.5.4(a), prove that if the function f is left continuous at c, then f is integrable with respect to 1, on
[a, b].
b. Show that 1, is not integrable with respect to 1,
3. Let a be nondecreasing on (a, b). Suppose f is bounded on [a. b] and integrable with respect to a on [a. b]. For
X E [a, b] set F(x) = f. f da.
*a. Prove that IF(x) - F(y)l s Mla(x) - a(y)l for some positive constant M and all x. y E [a, b].
b. Prove that if a is continuous at x E [a, b), then F is also continuous at x0.
4. a. Prove Theorem 6.5.8(a).
b. Prove Theorem 6.5.8(c).
c. Prove Theorem 6.5.8(d).
5. Use the theorems from the text to compute each of the following integrals.
W/2
d. 1([x] + x) da(x)
3 where a(x) - x2 + e'
I
4 I
6. Verify the integration by parts formula with parts (b) and (c) of the previous exercise.
7. Find fu fda, where f is continuous on [0, 1 ] and
CO)
9. Suppose a and j6 are monotone increasing on [a, b], and 91 is a partition of [a, b]. Prove that
11L(91, a, P) = a(b)$(b) - a(a)3(a) - (3, a).
10. Prove that if f is a continuous real-valued function on [a, bland a is monotone increasing on [a, b], then
imoY(P, f, a) exists and ii it,o pp, f, a) = f; f da.
a
260 Chapter 6 The Riemann and Riemann-Stleltjes Integral
11. Let f be a bounded real-valued function on [a, h] and a a monotone increasing function on (a. b]. Prove the fol-
lowing: If Arn0Y(91. f, a) = 1, then f E 9t(a) and f ."f da = 1.
12. *a. Let a be a monotone increasing function on [a, b]. If f E R(a), prove that j2 E Jt(a).
b. If f, g E llt(a), prove that f g E Jt(a).
9t(a) on [a, b].
13. If f E Jt(a) on [a, b] with Range f C [c, d], and rp is continuous on [c, d], prove that V of E
14. Suppose f is a nonnegative continuous function on [a, b], and a is nondecreasing on [a, b]. Define the function R
on [a, b] by 6(x) = fa f da. If g is continuous on [a, b], prove that
h b
1 g df = J fg da.
4,
Numerical Methods4
In this section, we will take a brief look at some elementary numerical methods that are
useful in obtaining approximations to Riemann integrals. Even though the fundamental
theorem of calculus provides an easy method for evaluating definite integrals, it is use-
ful only if we can find an antiderivative of the function being integrated. To illustrate
this, in Example 6.3.9 we showed that
z
dr = 2 In 5.
f 1 + 1--
JO
This, however, is not particularly useful if we do not know the value of In 5. By Exam-
ple 6.3.5,
ln5 = dr.
J
t
To obtain an approximation to In 5, we can choose from several available methods for
obtaining numerical approximations to the definite integral.
I J(x
k-1
.. t)(Xk - Xk-1)
as an approximation to the integral. This is nothing but the Riemann sum for the parti-
tion 91 = {x0, x,, . . . , of [a, b] with tk = xk_, for all k.
In using Riemann sums to approximate the integral of f, it is convenient to take
equally spaced partitions. Let n E N, and set h = (b - a)/n. Define
xo=a, x,=a+h, (17)
Thus if 9 = {x0, x,, ... , with xi as defined, we always have .x; = h for all
i = 1, , n, and
n
Y,f(ti)Axi = h Yf(t,),
i-t i_,
where for each i = 1, ... , n, t; E [xi_,,xi). If we take t; = xi_ 1 for all i, then we ob-
tain Euler's formula above. Similarly, we could take t; to be the right endpoint xi of the
interval [xi_ 1, xi]. Another choice of t; would be the midpoint; i.e., t; = (xi_, + x;)/2.
For monotone functions, it is intuitively obvious that the midpoint gives a better ap-
proximation than either the right or left endpoint to the integral of f over the interval
[xi_ 1, xi](see Figure 6.9). With xi as defined by equation (17) and t; = (xi _, + x,)/2,
the above formula becomes
M (f) is called the nth midpoint \àpproximation to the integral off over
[a, b].
Regardless of the method used, it is important to be able to estimate the error be-
tween the true value and the approximate value. If f E 9t[a, b], then for any partition
for any choice of tk E [xk _ 1. xk ]. Inequality (19) follows from the fact that both the Rie-
mann sum of f and the integral of f lie between the lower and upper sum of f. If f is
monotone increasing on [a. b], then by the proof of Theorem 6.1.8(b),
6.6.1 EXAMPLE The function f(x) = 1/x is decreasing on [ 1, 5]. Thus by inequality (20),
for n E Id and h = (b - a)/n,
for any choice of t; E [xi _ 1, x;]. Thus with n = 8, the error is less than 0.4;
n = 160 only guarantees an error of less than slo = 0.02. We would be required to take
very large values of n to be guaranteed a sufficiently small error.
With n=8,h=Z andxi=I+i,i=0,1,...,8,andt,=xi_j=(1+i)/2
(the left endpoint),
8
1
9(Ps'f) = 1 +i
=2+3+4+5+6+7
I
8+9
= 1.8290 (to four decimal places).
To four decimal places, In 5 = 1.6094. Thus the error is 0.2196-less than the pre-
dicted error of 0.4.
If we use the midpoint approximation (18) (again with n = 8), then t; = 1 + (i - i) i'
which upon simplification equals (3 + 2i)/4. Thus
1 $ 4
M8(f)=2j3+2i
6.6 Numerical Methods 263
6.6.2 THEOREM Lei f be a real-valued differentiable function on [a, b] for which f'(x) is
bounded on [a, b]. Then for any partition 01 of [a, b],
for any choice of ti E [x,_ 1, xi]. A slight improvement of this inequality is given in the
exercises (Exercise 6). If denotes the error between the true value and the ap-
proximate value, that is,
En() = f (x) dx - f ),
Ja
264 Chapter 6 The Riemann and Riemann-Stleltjes Integral
then the above inequality can simply be written as IE"(f)I s Ch, where C is a fixed
constant depending only on f and the interval [a, b]. Since the term "h" occurs to the
first power, this method is commonly referred to as a first order method.
In Example 6.6.1 we saw that the midpoint rule provided much greater accuracy
than using the left endpoint. This is not a coincidence as the following theorem proves.
6.6.3 THEOREM Iff is twice differentiable on [a, b] and f"(x) is bounded on [a, b], then
f (X)
I Ja
- M"(f) I <- M( 24 a)
b2'
Remark Since the error between the true value and the approximate value involves
the term h2, the midpoint approximation is a second order method.
Proof. To prove the result, we first prove that for each i = I, ... , n,
f(x)dx - hf
(a +(i-
gi(2)= h)).
Since f is continuous, by Theorem 6.3.4
g;(t) = f(ci + t) + f(ci - t) - 2 f'(c), and
g1(t) = f'(ci + t) - f'(ci - t).
By the mean value theorem,
f'(c, + t) - f'(c; - t) = f"(C)2t
for some C E (ci - t, ci + t). Since If"(l;)1 M, we obtain
Ig;(t)I <- 2Mt
for all t E [0, J. Since g;(0) = 0, by the fundamental theorem of calculus,
W(t)I = [fg7(x)dx I S
jIg7(x)Idr2Mfxdr1.rMt2.
0 0 0
6.6 Numerical Methods 265
f(x)dx-M"(t)I = I j f(x)dx-hf1a+li-2)h)
j8i(2)I
jt,
82 24Mh3n = M(b - a)
h
24 h
2.
2[J(xi-I) +f(xi)] x.
xi
Figure 6.10
as an approximation to the integral off on [a. b]. If, as previously, we set h = (b - a)In
and x; = a + ih, i = 0, ... , n, then the above sum becomes
h" h -I
2 7, [f(a + (i - 1)h) + f(a + ih)J = 2 {f(a) + 2 f(a + A) + f(b),.
where h = (b - a)/n, is called the nth trapezoidal approximation to the integral off
on [a, b). If we set yi = f(a + ih), then T"(f) can be expressed as
h l
TV) =2 Yo+22yi+Y.J.
The following theorem, under suitable hypothesis on f, provides an estimate of the
error of the trapezoidal approximation.
6.6A THEOREM 1f f is twice differentiable on [a, bJ and f"(x) is bounded on [a, b], then
b
M(b - a) h2
l f(x) dx - TT(f)
12
Proof. The proof of this theorem is very similar to the proof of Theorem 6.6.3. As a
consequence, we leave most of the details to the exercises (Exercise 5). The first step is
to prove that for each i = 1,...,n,
Simpson's Rule
The trapezoidal approximation T (f) amounts to approximating the function f with a
piecewise linear function which passes through the points {(x;, f(x;))}"=0 (see
Figure 6.11). Our intuition should convince us that one way to obtain a better approxi-
mation to the integral off over [a, b] is to use smoother curves. This is exactly what is
done in Simpson's rule, which uses parabolas to approximate the integral. To use qua-
dratic approximations we will need to use three successive points of the partition of
[a, b].This is because three points are required to uniquely determine a parabola.
( 2k h
J p(x) dx = (24)
0
s [Yo + 4Yi + Y2]
One way to derive this formula would be to first determine the coefficients A, B, C so
that p(O) = yo, p(h) = yi, p(2h) = y2, and then integrate p(x). This, however, is unnec-
essary. By integrating first,
268 Chapter 6 The Riemann and Riemann-Stieltjes Integral
Let f E 9t[a, b] and let n E N be even. Set h = (b - a)/n. On each of the inter-
vals [a + 2(i - 1)h, a + 2ih], i = 1, ... we approximate the integral off by the
integral of the quadratic function that agrees with fat the points
Yo = f(a + 2(i - 1)h), y' = f(a + (2i - 1)h), y2 = f (a + 2ih).
By identity (24) this gives
3[f(a)+4f(a+h)+2f(a+2h)
+4f(a+3h)+ +4f(a+(n-1)h)+f(b)]
as an approximation to the integral off over (a, b]. The quantity S (f) is called the nth
Simpson approximation to the integral off over [a, b]. If we set y; = f(a + ih), then
is given by
(25)
The following theorem, again under suitable restrictions on f, provides an error es-
timate for Simpson's rule.
6.6.5 THEOREM If f is four times differentiable on [a, b] and f t4i(x) is bounded on [a, b],
.
Remarks. Since the error term involves h4, Simpson's rule is a fourth order method.
Also, if f(x) is a polynomial of degree less than or equal to 3, then f(4i(x) = 0 and thus
f f (x) dx = S&).
0
6.6 Numerical Methods 269
If f satisfies the hypothesis of the theorem and n E RI is even, then the above error es-
timate can be expressed as
rb M(b - a)5 I
1, f (x) dx - S"(f) I `- (26)
180 n4
Proof. The proof of the theorem proceeds in an analogous manner to the proofs of
the previous two results. We first prove that for i = 1.... , ;,
To prove inequality (27), we are required to show that Ig,(h)I <- Mh5/90. Upon com-
puting the successive derivatives of g, we obtain
gi(t)=3[f'(c;-t)-f'(c,+t)]+3[f(ci-t)+f(cr+t)]-4f(ci).
g7(t)=3[ f"(c,--f"(c;+t)]+3[ t)+f'(c;+t)],
and
I8,""(t) l s Mt g.
3
As in the previous two theorems, since g;(0) = g;(0) = g"(0) = 0, upon three integra-
tion we have Ig;(h)l s Mh590. This proves inequality (27). Finally,
n/2 nr_
1a f(x)dx - S"(f)I = Igi(x)I lgr(x)l
Mh5 n M(b - a)
h
90(2) 180
.6.6.6 EXAMPLE In this example we will use the trapezoidal rule and Simpson's rule with
n = 8 to obtain approximations to In 5. For f(x) = x-', f"(x) = 2x'3, and f(4kx) _
24x-'. Therefore,
sup{If"(x)I: x E [1, 5]} = 2 and sup{Ift4I(x)I: x E [1, 5]) = 24.
270 Chapter 6 The Riemann and Riemann-Stieltjes Integral
6
IEa(f)I 212(-21 Y = = 0.16666
On the other hand, the error Es in using Simpson's rule is guaranteed to satisfy
a
IEsl)1 0.03333
2180 2 30 =
which is considerably better.
With x, = 1 + ih and y; = f(x;), i = 0, 1, .. . , 8,
2 1 2 1 2 1 2 1
Yo=1,Yi=3, Y6=4,Y7=9,Ye5
Therefore,
Ts(f)=4 Yo+2ly,+Ys
i=1 I
4+3+1+5+3+7+2+9+5
= 1.6290 (to four decimal places).
Since In 5 = 1.6094 to four decimal places, the error is less than 0.02, well within the
tolerance predicted by the theory. With Simpson's rule,
1
=I[I+3+1+8+3+g+2+9+I
= 1.6108 (to four decimal places).
With Simpson's rule, the actual error is less than 0.0014.
We can also use the results of the theory to determine how large n must be chosen
to guarantee predetermined accuracy. For the trapezoidal rule, since M = 2 and
(b - a) = 4, by inequality (23)
3
IEn(f)It212 n2'
Thus to obtain accuracy to within 0.001, we need
3
which is accomplished with n ? 104. On the other hand, using Simpson's rule,
24.45 1
1< n4.
180
6.6 Numerical Methods 271
24.4'
nS 103.
> 180
EXERCISES 6.6
1. a. Use the midpoint rule, trapezoidal rule, and Simpson's rule to approximate In 2 = f, 2(1/x) dx with n = 4. For each
method, determine the estimated error and compare your answer to In 2 = 0.69315 (to five decimal places).
b. Repeat part (a) with n = 8.
c. For each of the three methods (midpoint rule, trapezoidal rule, Simpson's rule) determine how large a must be
chosen to assure accuracy in the approximation of In 2 to within 0.0001.
2. a. Using the fact that
' dt it
f01+t 4
use the midpoint rule, trapezoidal rule, and Simpson's rule to approximate ,r/4 with n = 4. For each method,
determine the estimated error.
b. Repeat part (a) with n = 8.
c. For each of the three methods determine how large n must be chosen to obtain an approximation of n accurate
to within 0.0001.
3. Use Simpson's rule to obtain approximations of each of the following integrals accurate to at least four decimal
places.
2 i
a. dx "b. 10 1 + 2 dx c. to sin(x2) dx
1a 1 + aP
4. Determine how large a must be chosen so that the trapezoidal approximation T. approximates 5 a-'2dx with an er-
ror less than 10-6.
5. Fill in the details of the proof of Theorem 6.6.4.
6. Let f be a differentiable function on [a, b] with f(x) bounded on [a, b]. Let n E N. Prove that
e M(b - a)
I J f(x) dx - h f(a + ih) I S 2 h,
9. a. Use the previous exercise to show that ,O,Z < TA - In 2 < 384
b. Using part (a) and Exercise 1(b) show that .6915 < In 2 < .6938.
6.7.1 LEMMA A finite or countable union of sets of measure zero has measure zero.
Proof. We will prove the lemma for the case of a countable union of sets of measure
zero. The result for a finite union is an immediate consequence. Suppose {En}FEN is a
countable collection of sets that measure zero. Set E = U E and let e > 0 be given.
Since each set E. is a set of measure zero, for each n E N there exists a finite or count-
able collection {1n.k}k of open intervals such that E.C and kl(I,,.k) < e/2".
Since we can always take In, k to be the empty set, there is no loss of generality in as-
suming that the collection {I , k}k is countable. Then {In, k}n. k is again a countable col-
lection of open intervals with E C U,,, k In, k.
Since N X N is countable, there exists a one-to-one function f of N onto N X N.
For each m E N, set J. = Then {J,n}T=, is a countable collection of open
intervals with EC Un, J,n. Since f is one-to-one, for each N E N, the set
F N = f({ 1, ... , N)) is a finite subset of N X N. Hence there exist positive integers N,
and KI such that for all (n, k) E FN we have I S n < N, and 1 <_ k <_ K1. Hence
N
t nsN,
I (In, k) = G I I (In. k) c 7.1 I (In. k) < Y
n=1k=1 n=Ik=I n-12
n< e.
15ksK,
Proof. We first prove that for each c > 0, the set EE _ {x E [a, b]: f(x) ? c} has
measure zero. Let e > 0 be given. Since fo f = 0, there exists a partition 9 =
{x0, x1,.., x,,} of [a, b) with °lt(9, f) < ce, where
6.7 Proof of Lebesgue's Theorem 273
and Mi = sup{ f (x) :x E [xi _ 1, xi] }. Let I= {i : E,0 (xi _ 1, x, ] # ¢}. If i E 1, then
Mi at c. Hence
CE > V(P,f) 2t I MiAxi CY1AX,
idl tEl
Thus J:1Ei1([x,_,, xi]) _ 7,iEi Axi < e. Finally, since EC Uiel [xi_ 1, xi], it follows
that Ec has measure zero.
To conclude the proof we note that {x E [a. b] : f(x) > 0} = U' 1 E. where for
each n E (\J,
En={xe[a.b]:f(x)!}.
By the above, each E. has measure zero. The conclusion now follows by Lemma 6.7. 1.
J
6.7.3 DEFINITION If E is a subset of IL the characteristic function of E, denoted Xe, is
the function defined by
XE(x) _ 1, x E E,
0, x 44 E.
As in Section 6.2, if 9 = {xo, xl, ... , xn} is a partition of [a, b], the norm 119111 of the
partition 9 is defined by 119I1 = max Axi. If f is a bounded real-valued function on [a, b],
15i5n
the infimum and supremum off on [xi_ 1, xi] are denoted by mi and Mi respectively; i.e..
Lf(x) and
i
i=1
The graphs of the lower function L f and upper function Uf off are depicted in Figure
6.12. Since mi s f(x) c Mi for all x E [xi_ 1, xi),
L1(x) s f(x) s Uf(x) for all x E [a, b).
Also, since L f and Uf are continuous except at a finite number of points, they are Rie-
mann integrable on [a, b] (Exercise 1) with
J Lf=T(9P,f) and
Ja Uf=a13,(2,f)
a
274 Chapter 6 The Riemann and Riemann-Stieltjes Integral
1b1
Assume first that f is Riemann integrable on [a, b]. Then for each n E N,
there exists a partition 91, of [a, b] with
0:S
a
1/n, such that
Since 2(91.,f) t 2(91,f) and V(91, f) s 61L(91, f) for any refinement 91 of a,,, the
partitions 9P,,, n = 1, 2, .. . , can be chosen so that SPn+I is a refinement of 91..
For each n E N, let L,, and U respectively be the lower and upper functions off
corresponding to the partition P,,. Then L,(x) s f(x) <_ U,(x) for all x E [a, b) and
1a
bLn = 49., f) and
f bUn = (29)
a
Since 91n+ l is a refinement of p,,, the functions L. and U. satisfy La(x) _< L,,, i(x) and
Un+ 1(x) s U,(x) for all x E [a. b]. Define L and U on [a, b] by
L(x) = lim L,(x) and U(x) = lim U,(x).
Then
I L = JbU= fbf
Since U(x) - L(x) ? 0 for all x E [a, b], by Lemma 6.7.2, {x E [a, b] : U(x) # L(x)}
has measure zero. Furthermore, since each 91 has measure zero, by Lemma 6.7.1, the set
E = {x E [a, b] : U(x) # L(x)} U (U 9'n)
also has measure zero. We conclude by showing that f is continuous on [a, b]\E.
Fix x, E [a, b]\ E, and let e > 0 be given. Since L(x,,) = U(x,), there exists an in-
teger k E N such that Uk (x,) - Lk(x,) < e. Also, since xa E 9'k, the functions Uk and
Lk are constant in a neighborhood of x0. Hence there exists a S > 0 such that
0 < Uk (x,,) - Lk (x) = Lk (x) - Lk (xo) < E
for all x E [a, b] with Ix - S. Finally, since Lk(x) <- f(x) Uk(x) for all
X E [a, b),
- E < Lk (x) - Uk (xo) C f (x) - f (x,,) < Uk (x) - Lk (xo) < E
for all x with Ix - x0I < S. Therefore f is continuous at x,,.
Conversely, suppose f is continuous on [a, b]\E, where E is a set of measure zero
and a < b. Let M > 0 be such that I f(x)I M for all x E [a, b], and let e > 0 be
given. Since the set E has measure zero, there exists a finite or countable collection
of open intervals such that E C U I and E 1(1,,) < E/4M. Also, since f is continuous
on [a, b]\E, for each x E [a, b] \ E there exists an open interval J, such that
1f (z) - f (y) I < E/2(b - a) for all y, z e J, fl [a, b]. The collection U {J, :
x E [a, b]\ E} is an open cover of [a, b]. Thus by compactness, a finite number, say
{Ik}x=, and {JX}; , also cover [a, b]. Let
9' = {a = to, t,, .. . , tN = b}
be the partition of [a, b] determined by those endpoints of Ik, k = 1. ... , n, and
J,,, j = 1, ... , m, that are in [a, b]. For each j, 1 s j c N, the interval (t,_,, t,) is
contained in some Ik or J,,. Let J = {j: (t, _ 1, t,) C Ik for some k}. Also, for each
j E {1, ...,NJ, let m, and M; denote the infimum and supremum off on [t,_,. t;] re-
spectively.Then M, - m, <- 2M for all j E J, and M, - m, < E/2(b - a) for all j 44 J.
Thus
52M:01 + E
2(b - a)1EJ
c1t!
,EJ f
276 Chapter 6 The Rlemann and Riemann-Stieltjes Integral
N
E
:5 2M7, 1 (1k)+2(b-a 4r;
E E
52M\4M)+2(b-a)(b-a)=E.
Hence by Theorem 6.1.7 the function f is Riemann integrable on [a, b]. 1]
W EXERCISES 6.7
1. Let f be a bounded real-valued function on [a, b] and 91 a partition of [a, b]. Prove that the lower function Lt and
upper function Uf off for the partition ' are Riemann integrable on [a, b] with
rb
jb Lf = Y(9,f) and = W495,f)
0 1a
2. Let P be the Cantor set in [0, 1 ]. Show that Xr E %[(), 1) and find f.' kAX)dx.
3. Let f E R[a, b], and for x E [a, b], set F(x) = fo f(t) dt. Prove that there exists a subset E of [a, b] of measure
zero such that F'(x) = f(x) for all x E [a, b] \ E.
4. *Suppose f E %[a, b) and g is a bounded real-valued function on [a. b). If {x E [a, b) : g(x) # f(x)} has mea-
sure zero, is g Riemann integrable on [a, b)?
NOTES
The fundamental theorem of calculus is without question the Lebesgue's theorem is one of the most beautiful
key theorem of calculus; it relates the Cauchy-Riemann the- results of analysis. It provides very concise necessary
ory of integration to differentiation. For Newton, Leibniz, and sufficient conditions for Riemann integrability of a
and their successors, integration was the inverse operation of bounded real-valued function f. Although Riemann was
differentiation. When Cauchy, however, defined integration the first to provide such conditions, his result lacked the
independent of differentiation, the fundamental theorem of simplicity and elegance of Lebesgue's theorem. Unfortu-
calculus became a necessity. It was crucial in proving that nately for Riemann, the concept of measure had not yet
Cauchy's integral was the inverse of differentiation, thereby been developed when he stated and proved his result. In
providing both a convenient tool for the evaluation of defi- Chapter 10 we will develop the theory of measure and
nite integrals and proving that every continuous real-valued Lebesgue's extension of the Riemann integral.
function defined on an interval I has an antiderivative on /. The need for numerical methods for the evaluation of
Although we stated the fundamental theorem of calcu- definite integrals was recognized as early as the eighteenth
lus as two separate theorems, for continuous functions they century. Euler and Thomas Simpson (1710-1761), among
are the same. Specifically, if f is a continuous real-valued others, used numerical techniques to approximate the defi-
function on [a, b] and F is an antiderivative off, then for nite integral in problems where an antiderivative could not
any x E [a, b], be found. Even the error estimates developed in Section 6.6
date back to that era. With the availability of efficient cal-
F(x) = Fa) + 11f(t) dt. culators and high-speed computers, numerical methods
have increased in importance in the past few decades. This
Conversely, if f is continuous and F is defined as above. has led to the development of very sophisticated numerical
then F(x) = f(x). algorithms for the evaluation of definite integrals.
Miscellaneous Exercises 277
Although Newton and Leibniz are credited with differential calculus for finding tangents to a curve in
inventing the differential and integral calculus, many his 1670 treatise Lectiones geometriae. However, all these
mathematicians prior to their time knew formulas for early methods were developed using geometric arguments.
computing tangents and areas in particular instances. Newton and Leibniz developed the concepts, the notation.
Archimedes (around 200 a.c.), in his treatise Quadrature and the algorithms for making these computations for ar-
of the Parabola, used the method of exhaustion by in- bitrary functions. Most importantly, however, both men
scribed triangles to derive a formula for the area under a realized the inverse nature of the problems of tangents
parabolic segment. By the mid-1640s, Pierre de Fermat and areas. For these reasons they are credited with the de-
(1601-1665) had determined the formulas for the area velopment of the differential and integral calculus. Fur-
under any curve of the form y = x' (k * -1), and for ther information on the historical development of calculus
finding tangents to such curves. Isaac Barrow (1630- may be found in the text by Katz listed in the Bibliogra-
1677), a professor of gegmetry at Cambridge, developed phy and the article by Rosenthal listed at the end of the
an algebraic procedure that is virtually identical to the chapter.
MISCELLANEOUS EXERCISES
A real-valued function f on [a, b] is a step function if there exist a finite number of disjoint intervals
(Ii}n- 1 with U Il = [a, b) such that f is constant on each of the intervals lt, j = 1, ... , n.
1. a. If f is a step function on [a, b), prove that f E 9t[a, b) with
b a
- f=7,
'-1
2. Let f be defined on (0, oo). Prove that there exists a e R U {-oo, oo) such that f(f)(s) is defined for all
s e (a, oo), and that the integral defining Z(f)(s) diverges for all s E (-oo, a).
(Hint: First show that if `d;(f)(sa) exists for some sa, then T(f)(s) exists for all s > se.)
3. Suppose f E 9t[0, c] for every c > 0, and that there exists a positive constant C, and a E R. such that
I f(r) 1 s Ce- for all t ? 0. Prove that T(f)(s) exists for all s > a.
278 Chapter6 The Riemann and Rlemann-Stieltjes Integral
4. Compute the Laplace transform of each of the following functions. In each case, specify the interval on which
`af(f)(s) is defined
a. At) = 1
b. f (r)
c. f(t) = cos wt
d. f(t) = sin tot
e. f(t) = t", n E N (Use induction.)
I. f(t) = I (t - c), where J(t - c) is the unit jump function at t = c
g. f(r) = t°, a > - I (See Exercise 9, Section 6.4.)
5. Suppose f is differentiable on [0, oo) and a E R is such that `.f(f)(s) exists for all s > a. If lim e-"f(t) = 0 for all
s > a, prove that Y(f')(s) = s .(f)(s) - f(0).
® SUPPLEMENTAL READING
Bagby, R. J., "A convergence of limits," Math. Mag. 71 Jones, W. R. and Landau, M. D., "One-sided limits and
(1998),270-277. integrability," Math. Mag. 45 (1972), 19-21.
Bao-lin, Z., "A note on the mean-value theorem for inte- Klippert. J.. "On the right-hand derivative of a certain
grals." Amer. Math. Monthly 104 (1997), 561-562. integral function;' Amer. Math. Monthly 98 (1991).
Bartle, R. G., "Return to the Riemann integral;' Amer. 751-752.
Math. Monthly 103 (1996), 625-632. Kristensen, E., Poulsen, E. T. and Reich, E., "A charac-
Botsko, M. W., "An elementary proof that a bounded a.e. terization of Riemann integrabiliry"Amer. Math.
continuous function is Riemann integrable," Amer. Monthly 69 (1962), 498-505.
Math. Monthly 95 (1988), 249-252. Quadling, Douglas, "a in my eye," Math. Gaz. 77
Bullock. G. L., "A geometric interpretation of the Rie- (1993),314-321.
mann-Stieltjes integral:' Amer. Math. Monthly 95 Rickey, V. F. and Tuchinsky, P. M., "An application of
(1988), 448-455. geography to mathematics: History of the integral
Goel, S. K. and Rodriguez, D. M., "A note on evaluating of the secant," Math. Mag. 53 (1980), 162-166.
limits using Riemann sums:' Math. Mag. 60 Rosenthal, Arthur, "The history of calculus:' Amer.
(1987), 225-228. Math. Monthly 58 (1951), 75-86.
Grabinger, J. V., "Was Newton's calculus a dead end? Ross, K., "Another approach to Riemann-Stieltjes inte-
The Continental influence of Maclaurin's Treatise grals:' Amer. Math. Monthly 87 (1980).660-662.
of Fluxions," Amer. Math. Monthly 104 (1997), Stein, S. K., "The error of the trapezoidal method for a
393-410. concave curve:' Amer. Math. Monthly 83 (1976)
Hartman, F. and Sprows, D., "Oscillating sawtooth func- 643-645.
tions:' Math. Mag. 68 (1995), 211-213. Williams, K. S., "Note on 5 (sin xlx)dx," Math. Mag.
Jacobson, B., "On the mean-value theorem for inte- 44 (1971),9-11.
grals:' Amer. Math. Monthly 89 (1982), 300-301.
71 Series of Real Numbers
7.1 Convergence Tests
7.2 The Dirichlet Test
7.3 Absolute and Conditional Convergence
7.4 Square Summable Sequences
Although the study of series has a long history in mathematics,' the modem definition
of convergence dates back only to the beginning of the nineteenth century. In 1821,
Cauchy, in his text Cours d'Analyse, used his definition of limit to provide the first for-
mal definition of convergence of a series in terms of convergence of the sequence of
partial sums. The Cauchy criterion (Theorem 2.7.3) was the first significant result to
provide necessary and sufficient conditions for convergence of a series. Cauchy not
only stated and proved the result, he also applied his result to prove convergence and
divergence of given series. Many of the early convergence tests, such as the root and ra-
tio test, are due to him. Cauchy, with his formal development of series, placed the sub-
ject matter on a rigorous mathematical foundation.
In this chapter we will continue our study of series of real numbers. Our primary
emphasis in Section 7.1 will be on deriving several tests that are useful in determining
the convergence or divergence of a given series. In Section 7.3, we will study the con-
cepts of absolute convergence, conditional convergence, and rearrangements of series.
One of the key results of this section is that every rearrangement of an absolutely con-
vergent series not only converges, but converges to the same sum. As we will also see,
this fails dramatically if the series converges but fails to converge absolutely.
In Section 7.4, we give a brief introduction to the topic of square summable se-
quences. These play an important role in the study of Fourier series. One of the main
results of this section will be the Cauchy-Schwarz inequality for series. The section also
contains a brief introduction to normed linear spaces.
279
280 Chapter 7 Series of Real Numbers
Convergence Tests
In Section 2.7 we provided a very brief introduction to the subject of infinite series. In
the study of infinite series it is very useful to have tests available to help determine
whether a given series converges or diverges. For example, Corollary 2.7.5 is very use-
ful in determining divergence of a series. If the sequence {ak} does not converge to zero,
then the series 7- ak diverges. On the other hand, if lim ak = 0, then nothing can be as-
certained concerning convergence or divergence of the series I ak. In this section we
will state and prove several useful results that can be used to establish convergence or
divergence of a given series. Additional tests for convergence will be given in the exer-
cises and subsequent sections. With the exception of Theorem 7.1.1, all of our results
in this section will be stated for series of nonnegative terms.
As in Definition 2.7.1, given an infinite series Ja I ak of real numbers, I will
denote the associated sequence of partial sums defined by
n
sn _ Y, ak.
k=1
The series 7, ak converges if and only if the sequence of nth partial sums con-
verges in R. Furthermore, iffl-00
lim s = s, then s is called the sum of the series, and we
write
M
I ak = S.
k=1
Remarks
(a) Although we generally index a series by the positive integers N, it is sometimes
more convenient to start with k = 0 or with k = ka for some integer ko. In this case, the
resulting series are denoted as
I ak,
k=0
T, ak
k=k
7.1 Convergence Tests 281
Also, from the Cauchy criterion (Theorem 2.7.3) it is clear that I ak converges if
and only if 7, a k., ak converges for some, and hence every, k,, E N.
(b) Given any sequence {sn}.'= 1 of real numbers, we can always find a series I ak
whose nth partial sum is sn. If we set a1 = s1 and ak = sk - sk_ 1, k > 1, then
n
T. ak = Si + (SZ - S1) + . .
+ (Sn - S;_,) = Sn.
k=1
a
(a) Y, cak = ca, for any c E R, and
k-1
DO
(b) F; (ak+bk)=a+$3.
k-1
Proof. The proof of (a) is similar to (b) and thus is omitted. To prove (b), for each
n E N, let
n n
Sn = ak and to = bk.
k-1 k-1
Since the series converge to a and $3 respectively, tim s = a and lint t = $3. Therefore,
by Theorem 2.2.1, lim(sn + tn) = a + $3. But
n n
S. + to = ak + T, bk = 7, (ak + bk).
k=1 k-1 k-1
Therefore, sn + to is the nth partial sum of the - series Z(ak + bk). Since the sequence
{sn + t,,} converges to a + (3,
OD
I (ak+bk)=a+A
k-1
Comparison Test
One of the most important convergence tests is the comparison test. Though very ele-
mentary, it provides one of the most useful tools in determining convergence or diver-
gence of a series. It is useful both in applications and theory. Several of the proofs of
subsequent theorems rely on this test. In applications, by comparing the terms of a
given series with the terms of a series for which convergence or divergence is known,
we are then able to determine whether the given series converges or diverges.
282 Chapter 7 Series of Real Numbers
7.1.2 THEOREM (Comparison Test) Suppose 7, ak and 7, bk are two given series of
nonnegative real numbers satisfying
0 < at Mbk
for some positive constant M and all integers k a kn, for some fixed k E N.
00
(a) If bk < oo, then Y, ak < oo
k=1 k-1
W x
(b)IfY, at = oo, then 7, bk = oo.
k=1 k-1
Proof. Suppose the terms {ak} and {bk} satisfy at 5 Mbk for all k > ko and some pos-
itive constant M. Then for n > m ? ko,
n n
05 1 akSM 7, bk.
k=m+I k=m+I
Suppose I bk converges. Then given e > 0, by the Cauchy criterion (2.7.3) there ex-
ists an integer n,, ? k such that
km+1 bk < M
for all n > m ? n,. Thus 0 <_ L+k=m+I at < e for all n > mat n0. Hence by the
Cauchy criterion E at converges. On the other hand, if I at diverges, then 1 bk must
also diverge.
As a corollary of the previous theorem we also have the following version of the
comparison test.
7.1.3 COROLLARY (Limit Comparison Test) Suppose E at and E bk are two given series
of positive real numbers.
(a) If liimW bn = L with 0 < L < oo, then 7. at converges if and only if E bk
converges.
Proof. The proof, the details of which are left to the exercises (Exercise 6), follows
immediately from the definition of the limit and the comparison test.
Remark. If Jim an/bn = 0 and I at converges, then nothing can be concluded about
the convergence of the series 7, bk. In 7.1.4(d) we provide an example of a divergent
series 7, bk and a convergent series E at for which lim 0. On the other hand,
n-+oo
7.1 Convergence Tests 283
in Exercise 23, given a convergent series E ak with ak > 0, you will be asked to con-
struct a convergent series I bk with bk > 0 such that lim a"Jb, = 0.
7.1.4 EXAMPLES
(a) As an application of the comparison test, consider the series
°O k
ksi 3
We will compare the given series with the convergent geometric series 7, Thus we
wish to show that there exists ko E Fk1 such that k/3k s 1/2k for all k a k0. By Theorem
2.2.6(d),
k
k k 3 = 0.
Thus by taking e = 1, there exists an integer ko such that k(3)k <_ 1 for all k ? k,. As a
consequence,
we will compare the given series with the series Y, 1/k. This series is known to diverge
(Example 2.7.4). If we take e = 4%52, then we can conclude that there exists ko E N
such that
1+k 11vF2
2+k:2- e
4
for all k at k0. Thus there exists a positive constant M and k, E N such that
k+1 I
2k3+1aMk
284 Chapter 7 Series of Real Numbers
for all k ? ko. Since the series E 1/k diverges, by the comparison test the given series
also diverges.
(c) The divergence of the series
co
k+ 1
1 2k'+l
can also be obtained by the limit comparison test. Comparing the terms of the given se-
ries to the terms of the series Y, 1/k we have
?kT __ 1+k
=
lint
-"o 1
k
k-r+moo 2+k -2
Thus since the series 11/k diverges, by the limit comparison test the series
°O k+1
I 'V 2k' + 1
also diverges.
2_k
(d) Let ak = and bk = 1/k. Then 7, ak converges, 7, bk diverges, and
Thus if rim " = 0, convergence of the series 7, ak does not imply convergence of the
b
series bk.
Integral Test
Our second major convergence test is the integral test. Recall from Section 6.4, if f is a
real-valued function defined on [a, oo) with f E %[a, c] for every c > a, then the im-
proper Riemann integral off is defined by
Therefore,
ak = X(9j) .f(x)dx C ak
k-2 I k-1
7.1.6 EXAMPLES
(a) As our first application of the integral test, we consider the p-series
When p = 1 this series is referred to as the harmonic series. If p 0, then {k-°} does
not converge to zero, and thus by Corollary 2.7.5, the series diverges. Suppose
p > 0, p 1. Let f(x) = xo, which is decreasing on [ 1, oo). Then
`
Jx-Pdr=---___[i
1
p-1 --Jc
1
1
Therefore,
rco 1 p > 1,
x-odx=lim x-°dx= p- 1
JI `~00
JI 00, p < 1.
286 Chapter 7 Series of Real Numbers
Thus by the integral test, the series diverges for p < I and converges for p > 1. When
p = 1, then by Example 6.3.5,
f = Inc.
i
`Xdx
Since lim
X-W
In x lim+
1-.0
In t = oo (Example 6.4.2(a)), by the integral test the series
also diverges for p = I. Summarizing our results, we find
°O+ f converges, if p > 1,
G kr diverges, if p 1.
(b) As our second application of the integral test, we consider the series
I
00 1
k..,- kInk*
f'(x) < 0 for all x > 1. Thus f is monotone decreasing on [2, oo). But
'° Ink
7 kp. pEIR.
We first consider the case p = 1. Since f(x) = (In x)/x is decreasing on [e, oo) and
`Inx
lira - dx = lim0100
(in c)2 - 1 = oo,
C X
Ink _ 1 Ink 1
kr kq k' kq
7.1 Convergence Tests 287
for all k ? k0. Since q > I the series I 1/kq converges. Hence by the comparison test,
the series I (In k)/kP also converges.
Finally, if p < 1, then again by the comparison test, the divergence of I 1/k' im-
plies that 7, (In k)/kP also diverges. This follows from the fact that (In k)/k" ? (In 2)/k"
for all k E N, k ? 2. Summarizing our results we find
ink f converges, if p > 1,
k__Z kP diverges, if p 5 1.
lim s, kE
sup{s : n k} = slim sup(s : n !- k).
By Theorem 2.5.7, if E denotes the set of subsequential limits of {sn} in OI U {-oo, oo},
then
In particular, if the sequence {sn} converges (or diverges to either - oo or oo), then
Urn S. = lira S. = lim S.
7.1.7 THEOREM (Ratio Test) Let I ak be a series of positive terms, and let
and r = lim
a"-+'.
R = lim ak'+
k-+oo ak k-+oo ak
00
(a) If R < 1, then I, ak < 00-
k-1
00
7.1.8 THEOREM (Root Test) Let 7, ak be a series of nonnegative terms, and let
k
a = lim
k-0oo
ak
7.1.9 EXAMPLES
(a) Consider the p-series
°O 1
A kp, p E R.
k- I
limakk1=lim(1+k1 =
kroo a 1.
kyoo \\
p 4+I
+l
lim ak+1
k-+oo ak
= lim
k-boo
pk 1
(k + 1)! .
= p lim I
k-+= k + 1
= 0.
7.1 Convergence Tests 289
By the ratio test the series converges for all p, 0 < p < oo. As a consequence of Corol-
lary 2.7.5, we also obtain
k
limp =0
k-.oo k!
for all p E R. In this example, the presence of k! makes the root test difficult to use.
(c) Consider I a where
Zk, if n = 2k,
_
an l
3k. ifn=2k+1.
Here,
00
1+2+3+22+32+
-i
By computation,
2
a.+, (3) 1
a2k+I lk
21
lim
k- oo a2k
= lim
k-,oo 3
= 0.
1
n= 2k,
1 M )1/(2k+ t), n=2k+1.
By Theorem 2.2.6(b), urno (V)v(2k+ q = 1. Thus the sequence { } has two sub-
sequential limits; namely, 1/V and 1/V/3. Therefore, a = 1/N/2, and the series con-
verges by the root test.
290 Chapter 7 Series of Real Numbers
lim
a= = R < 1.
n-.m a,,
Choose c such that R < c < 1. Then by Theorem 2.5.3(a), there exists a positive inte-
ger n, such that
an
±an1 s c for all n ? no.
In particular, ane, 1 s can . By induction on m, a,, ,,,, < C'"a,, Therefore, writing
n=no+m,mz0,
an <_ Mc" for all n at no,
where M = a,. /c"°. Thus since 0 < c < 1, by Example 2.7.2(a), the series E c" con-
verges. Therefore, E an also converges by the comparison test.
Suppose
lima"}=r> I.
nToo an
Again choose c so that r > c > 1. As above, there exists a positive integer n, such that
a,, z Mc" for some constant M and all n _> n,. But since c > 1,1 c" = oo, and thus by
the comparison test, B an = oo. Q
Proof of Root Test. Let
a = lim
n-oo
<a..
Suppose a < 1. Choose c so that a < c < 1. Again by Theorem 2.5.3, there exists a
positive integer n, such that
" an < c for all n a n,.
But then a,, s c" for all n ? n and E an < oo by the comparison test.
If a > 1, then > I for infinitely many n. Thus a" > 1 for infinitely many
n, and as a consequence, {a,} does not converge to zero. Hence by Corollary 2.7.5, the
series B a,, diverges. Q
Example 7.1.9(c) provides an example of a series for which the ratio test is incon-
clusive, but the root test worked. Thus it appears that the root test is a stronger test; a
fact which is confirmed by the following theorem.
lim aa,
n-rm an
lim lim s lim a-`
n-+oo an
=oo
7.1 Convergence Tests 291
R
n-+oo a"
If R = oo, then there is nothing to prove. Thus assume that R < oo, and let P > R be
arbitrary. Then there exists a positive integer n, such that
an
an
s0 for all n ? n,.
As in the proof of the ratio test, this gives an s MI" for all n ? n where M = a,.19"
Hence
llm "a:s p.
n-,00
Since J3 > R was arbitrary, Tim s R, which proves the result. The inequality for
the limit inferior is proved similarly. Q
7.1.11 EXAMPLE If ak = k!, then lim(ak+I/ak) = oo. Thus as a consequence of the pre-
vious theorem,
EXERCISES 7.1
1. If a and b are positive real numbers, prove that
W I
Al (ak + b)o
converges if p > I and diverges if p c 1.
2. Test each of the following series for convergence:
sA, 2
k' +I
k b.
M
kt,
_
k2+2k-1
01+
C. G2k
k2
292 Chapter 7 Series of Real Numbers
00
k'e_k 3k
d. F, *e.
k1
k-1
(kl)2
*L
k+ 1 - f
k.
*9. (2k)!
*m.
k-1
sin
(1k)
p>0 p>0 0. ykln(l+k
k=1
3. For each of the following, determine all values of p E R for which the given geometric series converges, and find
the sum of the series.
*a 7, (sin 0 b. c. (l ± p) (p * 1)
3 k-11
4. Suppose ak a 0 for all k E N and I a,, < oo. For each of the following, either prove that the given series con-
verges, or provide an example for which the series diverges.
00 00 00
a,2, c.
4; 1 + ak k-1 k-I
M
*d. 7, kak e. V k ak f.
t-1 k-1
I 1
k4(ln k)P'
(Hint: Consider the three cases q > 1, q = 1, q < 1.)
6. *Prove Corollary 7.1.3.
7. If I ak converges and B bk = oo, prove that E (ak + bk) = oo.
8. Suppose is a sequence in R with a > 0 for all n E N. For each k E 101 set
k
1
bk = - ± a,,.
k n=I
Prove that Jk , bk diverges.
9. Suppose that the series E ak converges and {n;} is a strictly increasing sequence of positive integers. Define the se-
quence {bk} as follows:
b,=a,+ +a ,
bk=a.,_,+,+...+a,,:
Prove that E bk converges and that $k , bk = k I ak. (This exercise proves that if the series E ak converges,
then any series obtained from Eat by inserting parentheses also converges to the same sum. The following exer-
cise shows that removing parentheses may lead to difficulties.)
7.1 Convergence Tests 293
10. Give an example of a series F. ak such that Ek , (a2k _, + a2k) converges, but E ak diverges.
11. Prove that
p(k)
k-1
converges for any polynomial p and a > 1.
12. *Suppose that the series E ak of positive real numbers converges by virtue of the root or ratio test. Show that the
series Yt , k"ak converges for all is E N.
13. *Show that the series
1 1
+23+32++...
1 I
converges, but that both the ratio and root tests are inconclusive.
14. Apply the root and ratio tests to the series I ak where
15. Suppose ak 0 for all k E N. Prove that the series Ik , ak converges if and only if some subsequence {sj of the
sequence of partial sums converges.
16. *Cauchy Condensation Test: Suppose that a, z a2 z a3 a ? 0. Use the previous exercise to prove that
7, R , ak converges if and only if P 2kag converges.
17. Use the Cauchy condensation test to show that $4 , 1/n^ converges for all p > 1, and diverges for all p,
0<ps1.
18. Use the Cauchy condensation test to determine the convergence or divergence of each of the following series.
00+
00 1 O0 1
b.
*a. k_'
2 k Ink k_'2 k(In k)" (P > I) e' L k(ln kiln Ink)
19. *For k E N, let ck be defined by
Prove that {ck} is a monotone decreasing sequence of positive numbers that is bounded below. The limit c of the se-
quence is called Euler's constant. Show that c is approximately 0.577.
20. Raabe's Test: Let ak > 0 for all k r= N. Prove the following.
a. If ak+,/ak -- I - r/k for some r > 1 and all k a k k, E Id, then T. ak < oo. (Hint: Show that
(k - 1)ak - kak1 a (r - I)ak for all k ? k,.)
b. If ak+,/ak z I - 1/k for all k k k0 E N, then I ak = oo. (Hint: Show that {kak,1} is monotone increasing
fork z k,.)
21. *lf p, q > 0, show that the series
1)(p+2)... (p+k)
(p+
1)(q+2)...(q+k)
(q+
converges for q > p + I and diverges for q 5 p + 1.
294 Chapter 7 Series of Real Numbers
7.2.1 THEOREM (Abel Partial Summation Formula) Let {ak} and {bk} be sequences of
real numbers. Set
n
Ao=0 and A'= Iak
ifn?1.
k-I
Then if15p5q,
q 4-1
Y, akbk = E Akbk - bk+ 1) + Agbq - Ap-1b .
k=p k=p
4 q-1
Akbk Akbk+1
k-p k=p-I
7.2.2 THEOREM (Dirichlet Test) Suppose {ak} and {bk} are sequences of real numbers
satisfying the following:
(a) the partial sums A. _ Jk=, ak form a bounded sequence.
(b) b, a b2 a b3 a ? O, and
(c) k-occ
liimbk=0.
Then Jko' , akbk converges.
Proof. Since {A,} is a bounded sequence, we can choose M > 0 such that IA. 1 s M
for all n. Also, since b -+ 0, given e > 0, there exists a positive integer n, such that
b < e/2M for all n z n,. Thus, if n, s p s q, by the partial summation formula,
q- I
k.p akbkl =
L , Ak(bk - bk+1) + Aqbq - Ap_Ibp
k-p
1
Cq.P
-1
M7, (bk-bk+1)+bq+bp
2Mbp<E.
Hence by the Cauchy criterion (Theorem 2.7.3), Jk , akbk converges. 0
Alternating Series
Our first application of the Dirichlet test is to alternating series. An alternating series
is a series of the form Y, (-1)k bk or I(-1)k+ 1bk, with bk > 0 for all k
7.2.3 THEOREM (Alternating Series Test) If {bk} is a sequence of real numbers satis-
fying
(a) b1?b2z 0, and
(b) k bk = 0,
OD
then I (-1)k+lbkconverges.
k-1
Proof. Let ak = (-1)k+ 1. Then IA. I < 1 for all n, and the Dirichlet test applies. O
Remark The hypothesis (a) that {bk} is decreasing is required. If we only assume
that bk a 0 and Jim bk = 0, then the conclusion may be false (Exercise 2).
For an alternating series satisfying the hypothesis of Theorem 7.2.3, we can actu-
ally do better than just prove convergence. The following theorem provides an estimate
on the rate of convergence of the partial sums to the sum of the series.
296 Chapter? Series of Real Numbers
7.2.4 THEOREM Consider the series 7,(-1}k+Ibk, where the sequence {bk} satisfies the
hypothesis of Theorem 7.2.3. Let
00
Sn = (-1)k+Ibk and s = 7, (-1)k+1bk
k=I k=1
7.2.5 EXAMPLE Since the sequence fl/(2k - l)} decreases to zero, by Theorem 7.2.3, the
series
00 (-1)k+I
k- I 2k - l
converges. As we will see in Chapter 9 (Example 9.5.7), this series converges to 7r/4.
Thus by Theorem 7.2.4, if s is the nth partial sum of the series,
4 - s" 12n + 1
for all n E N. Although this can be used to obtain an approximation to ar, the conver-
gence is very slow. We would have to take n = 50 to be guaranteed accuracy to two
decimal places.
Trigonometric Series
Our next application of the Dirichlet test is to the convergence of trigonometric series.
These types of series will be studied in much greater detail in Chapter 9.
7.2.6 THEOREM (Trigonometric Series) Suppose {bk} is a sequence of real numbers sat-
isfying bI ? b2 ? ? 0 and kim bk = 0. Then
Proof. To prove the result, we require the following two identities: For t * 2pir,
pE1,
cos Zt - cos(n + 2)t
71si n kt =
k=1 2 sin i tt
(1)
n
sin(n + !)t - sin i t
Ylcos kt =
k=,
2
2 sin gt
2 (2)
We will prove identity (1), leaving identity (2) for the exercises (Exercise 4). Set
A. = Ink= I sin kt. Using the trigonometric identity
we obtain
(sm)An=
2 k=t
sin i t sin kt
2
=2kCosk2t-cosk+2 ItJ
=2fcos2t-cosn+21IJ.
Thus for t # 2pir, p E 7L,
which is finite provided t * 2pir, p e Z. Consequently by the Dirichlet test, the series
in (a) converges for all t # 2 pir, p E Z. If t = 2prr, p E Z. then sin kt = 0 for all k.
Thus the series in (a) converges for all t E R.
The proof of the convergence of the series in (b) is similar. However in this case, when
t = 2pir, cos kt = 1 for all k E N, and the given series may or may not converge. Q
Coskt
k-i k
298 Chapter 7 Series of Real Numbers
converges for all t * 2pir, p E Z. When t = 2prr, p e Z, then cos kt = I for all k and
the series E k diverges. On the other hand, the series
00
1
- zcos ki
k-I k
EXERCISES 7.2
1. *Abel's Test: Prove that if ak converges, and {bk} is monotone and bounded, then E akbk converges.
2. *Show by example that the hypothesis of Theorem 7.2.3 cannot be replaced by bk z 0 and ,, bk = 0.
3. If I ak converges, does E ak always converge?
4. *Prove that
a, (-1)k+I' p>0
00 (-1)klnk (-Ir
kal k' b. k-2
7 k
*G
Mklnk
00 kk kk
sink
* d.
k-1 (k + 1) k
!. k-1D -1) k+1
(k + I)k+1
*f
k_Z In k
determine how large n E N must be chosen so that I nrz/12 - s^ I < 10-', where s^ is the nth partial sum of the
series.
7. If p and q are positive real numbers, show that
00 (ln k)a
converges.
& *Suppose that E ak converges. Prove that
lim lkak=0.
0-00 n k.1
7.3 Absolute and Conditional Convergence 299
b = I - l2 + 1 - ... _ 1 = (-1)k+1
3 2n k=1 k
7.3.2 EXAMPLES
(a) Since the sequence {l/k} decreases to zero, by Theorem 7.2.3 the series
(-1)k+ 1/k converges. However,
° (-I)k+I xI
II
k=1
k I =
k-1
Ik=oo.
akl
k=1 k-1 lakl.
Proof. Suppose Zak converges absolutely; i.e., `jak1 < oo. By the triangle in-
equality, for 1 < p s q,
Thus by the Cauchy criterion (Theorem 2.7.3) the series 2 , ak converges. Finally.
with p = I in the above,
kI ak l = 1im
l t a, I
:!s: l t lakl :9 kY" IakI. Q
Remark, To test a series I ak for absolute convergence we can apply any of the ap-
propriate convergence tests of Section 7.1 to the series 7, Iakl. There is, however, one
important fact which needs to be emphasized. If the series Iak I diverges by virtue of
the ratio or root test, i.e.,
7.3A THEOREM (Ratio and Root Test) Let I ak be a series of real numbers, and let
aslim .
7.3.5 EXAMPLE To illustrate the previous theorem, we consider the series Z. -I P(k)o.
where p is a polynomial and b E R with I b I < 1. In this example, ak = p(k)b` and
thus
ak+, p(k + 1)
ak Ib1 I p(k) (
Since kim p(k + 1)/p(k) = 1 (verify this), Iak+,/a,I = IbI < 1. Thus by the ra-
,im
tio test the series is absolutely convergent.
7.3 Absolute and Conditional Convergence 301
Rearrangements of Series.
We next take up the topic of rearrangements of series. Loosely speaking, a series I a,.
is a "rearrangement" of the series 7- ak if all the terms in the original series 7, a4 ap-
pear exactly once in the series 7, ak, but not necessarily in the same order. For exam-
ple, the series
l 1z+
3-
+ 22+
4-
+
l2+
5 7-
k= k
k-l
which converges, but not absolutely. Consider also the following series, which is a re-
arrangement of the above:
s= k
k=1
s<s3 =1-23
+
6'
Let s,, denote the nth partial sum of the series in (3). Then
1-3+4k-1 1 _ 1)
3'3- R 71
k -I 4k 2k)
302 Chapter 7 Series of Real Numbers
Since
I + 1 1 - 8k-3
4k - 3 4k - 1 2k 2k(4k - 1)(4k - 3)'
we have
1 I _ 1 1
0<4k-3+4k-1
for some constant M. Thus the sequence {sin} is strictly increasing, and by the com-
parison test converges. Let s' = lim sin. Since
n-,00
A. +I -'S3n+ 4n + I and
1 1
S3n+2=S3,,+4n+ 1 +4n+3'
the sequences {sin+I} and {s',+2} also converge to s'. Therefore, lim sn = s'. Thus the
series (3) also converges. However, since
6=Si<Sg<Sy<
s' = lim s,, > 6. Thus the series (3) does not converge to the same sum as the original
n-i00
series. This, as we will see, is because the given series does not converge absolutely.
Sn - sn = ak - L./ a1(kN
k-! k-I
7.3 Absolute and Conditional Convergence 303
By the choice of p, the numbers a,, ... , aN appear in both sums and consequently will
cancel. Thus the only terms remaining will have index k or j(k) greater than or equal to
N. Thus, by inequality (4),
Is.-S;,I 52E
for all n a p. Therefore, lim s;, = lim sn, and thus the rearrangement converges to the
n-o0 n- 00
same sum as the original series. p
For conditionally convergent series we have the following very interesting result of
Riemann.
k-I
This series converges, but fails to converge absolutely. The positive terms of this series
are Pk = 1/(2k - 1), k E N; and the absolute values of the negative terms are
Qk = 1/2k, k E N. Since 7, 1 = oo, we also have 7, Pk = 7, Qk = oo. Suppose for
purposes of illustration a = 1.5. Our first step is to add "just enough" positive terms to
exceed a. More precisely, we let m, be the smallest integer such that P, + + P,,, > a.
For a = 1.5. m1 = 3; i.e., 1 + 3 + 3 > 1.5, whereas 1 + 3 < 1.5. Our next step is to
go in the other direction; namely, we let n 1 be the smallest integer such that
PI+...+pm,-Q,-. . .-Qn <a.
Again, for a = 1.5, n1 = 1, and 1 + 3 + s - ? < 1.5. We are able to do this since the
series 7, Pk and I Qk both diverge to oo. We continue this process inductively. Choos-
ing the smallest integers mk and nk at each stage of the construction will be the key in
proving that the resulting series converges to a.
Proof. Without loss of generality, we assume that ak * 0 for all k. Let pk and qk be
defined by
I ak = 7,
k-I
Pk - 7, qk
k=I k-I
304 Chapter 7 Series of Real Numbers
the convergence of one implies the convergence of the other. Thus they both must di-
verge.
Let P1, P;, P3,. .. denote the positive terms of 7, ak in the original order, and let
Q1, Q2, Q3, ... be the absolute values of the negative terms, also in the original order.
The series Pk and 7, Qk differ from I pk and I qk only by zero terms, and thus are
also divergent.
We will inductively construct sequences {mk} and {n*} such that the series
p1 + ... + Pm, - QI - . . . - Qn, + P.'+1
+ + Pm, - Qn..1 - - Q,, + . . . (S)
has the desired property. Clearly series (5) is a rearrangement of the original series.
Let m1 be the smallest integer such that
XI=PI+. ..+P,n'>a.
Such an m1 exists since 7, Pk = oo. Similarly, let n1 be the smallest integer such that
Y1
=X1-Q1-. . .-Qn, <a.
Suppose {m1, ... , mk} and {n1, . . . , nk} have been chosen. Let mk+ 1 and nk+ I be the
smallest integers greater than mk and nk respectively, such that
Xk+I = Yk + Pm,+l + . .. + Pm,., > a, and
Yk+I -Xk+1 -Qn,+I -- . - Qn,., <a.
Such integers always exist due to the divergence of the series I Pk and I Qk. Since
mk+1 was chosen to be the smallest integer for which the above holds,
Xk+I - Pm, 5a.
Therefore,
0<Xk+I -a-15 P.,.,
Similarly,
0<a-Yk+1:Qn,.,-
Since 7, an converges, we havek-.oc
lim Pk = lim Qk = 0. Therefore,
k-*oo
lim Xk = k-.oo
kroo
lim Yk = a.
Let S. be the nth partial sum of the series (5). If the last term of Sn is a Pn, then there
exists a k such that
Yk<Sn:!:-: Xi+I
If the last term of Sn is -Qn, then there exists a k such that
Yk+1 E S. < Xk+1.
In either case, we obtain 1 m Sn = a, which proves the result.
7.3 Absolute and Conditional Convergence 305
Remark. By a variation of the above proof one can show that if the series 7. ak is
conditionally convergent, then given a, 6 with
-oo<a 5 0<_00,
there exists a rearrangement 7, ak of 7, ak such that
EXERCISES 7.3
1. Prove the following:
a. If lim k°ak = A for some p > I, then F, ak converges absolutely.
b. If iimn k ak = A * 0, then Y, ak diverges.
k
c. If kim k ak = 0, then the test is inbonclusive.
2. 'Suppose I ak < oo and Mk < oo. Prove that the series V, ak bk converges absolutely.
3. a. Prove that if I ak converges and I bk converges absolutely, then 7, ak bk converges.
b. Show by example that the conclusion may be false if one of the two series does not converge absolutely.
4. Suppose that the sequence is monotone decreasing with lim b = 0. If {a,} is a sequence in R satisfying
I b- for all n E N, prove that ak converges absolutely.
5. a. If 7, ak converges absolutely, prove that ekak converges for every choice of ek E { -1. I }.
b. If Y_ ek ak converges for every choice of ek E { -1. 1 }, prove that 7, ak converges absolutely.
6. Test each of the following series for absolute and conditional convergence.
(_I)k+l
= 71
k-1 Nrk
b. I (-1)k
G N/k In(In k) sG
k-I kv
= (-I)k+Ikk
d.k-a-I(-I)kln(lnk)
In k
'e.
k
{-I)k
z k (In
k)P
P>0 k-1 (k + 1)
9. k-1
1 (k +
(-1)k+lkk
h.
00
,(-1+1 sin
() I I.
= (-1)k
,2kt+(_l)k
1)1` k-1 k
7. Test the series 1;k 1 pk/k°, p E R, for absolute and conditional convergence.
8. 'Show that the series I + - 3 + a + s - 1 + diverges.
9. Determine whether the series I - z - i + + s a i +s+s o ill + converges or diverges.
10. 'Prove that the series Ik 1 (sin k)/k is conditionally convergent.
11. Prove that if 7, ak is conditionally convergent, then there exists a rearrangement 7, ak of ak which diverges to c.
12. If ak a 0 for all k E N, and B ak = co, prove that 7, ak = oo for any rearrangement ak of I ak.
13. Suppose the series I ak is conditionally convergent. Given a,)9 with -oo s a <- rA s oo, show that there exists
II {ak} II 2 = Xak.
The set 12 is called the space of square summable sequences, and the quantity II{ak}II2
is called the norm of the sequence {ak}.
II a Ill =
F 1
I a(k)I2 < C.O.
7.4.2 EXAMPLES
(a) For the sequence { l/k}x 1,
kk2.
H{'}1122
Since this is ap-series with p = 2. the series I 1/k2 converges and thus { 1/k} E 12. On
the other hand, since
2. The topic of square summable sequences, although important in the study of Fourier series, is not specif-
ically required in Chapter 9 and thus can be omitted on first reading. The concept of a nonmed linear space
occurs on several occasions in the discussion of subsequent topics in the text. At that point, the reader can
study this topic more carefully.
7.4 Square Summable Sequences 307
(b) For fixed q, 0 < q < oc, consider the sequence { I /k v }. Then
{ klq 22 k'
By Example 7.1.9(a) this series converges for all q > ; and diverges for all q s ;-Thus
{ 1/k°} E 12 if and only if q > ;
Cauchy-Schwarz Inequality
Our main goal in this section is to prove the Cauchy-Schwarz inequality for sequences
in l2. First, however, we prove the finite version of this inequality.
Iakbkl `- , bk
k=1 Fk= k=1
0 ` 7,(Iakl - AIbkl)2 = 1 ak - 2A
n
` Iakbkl + A2 , bk.
n
OVA -
or C2 <_ AB; that is,
k lak 5
()(n)
Taking the square root of both sides gives the desired result. 0
As a consequence of the previous result we have the following corollary.
7A.4 COROLLARY (Cauchy-Schwarz Inequality) If {ak}, {bk} E 12, then 7,i 1 akbk is
absolutely convergent and
lakbkl s II{ak}112II{bk}II2.
k-1
308 Chapter 7 Series of Real Numbers
Iakbkl ak bk II{ak}II2II{bk}II2-
k-1 k-l k-I
Minkowski's Inequality
Our next result shows that the norm 11 112 satisfies the "triangle inequality" on 12.
7A.5 THEOREM (Minkowski's Inequality) If {ak} and {bk} are in 12. then {ak + bk} E 12
and
II{ak + bk}II2 c II{ak}1I2 + II{bk}II2-
Proof. By hypothesis, each of the series 7, a2 and I bi' converge. Also, by Corol-
lary 7.4.4, the series E ak bk converges absolutely. Since
(ak + bk)2 = ak + 2akbk + bk <_ ak + 2lakbkl + Mk,
we have
00 -0
II{ak + bk}IIi = 2 (ak + bk)2 5 II{ak}IIz + 2 7, Iakbkl + II{bk}IIi
k-1 k=1
S = (11{ak}II2 + 110J12)2-
Taking the square root of both sides gives the desired result.
Although not specifically stated, Minkowski's inequality is also true for finite
sums. In particular, if n E N and a,, . .. , a,,,'bi, . , ..
are real numbers, then
(d) If {ak}, {bk} E l2, then {ak + bk} E 12 and II{ak + bk}IL II{ak)II2 + II{bk}II2.
Proof. The results (a) and (b) are obvious from the definition, and (d) is just a restatement
of Minkowski's inequality. The verification of (c) is left as an exercise (Exercise 5). 0
For two vectors a and b in IR", the Inner product of a and b, denoted (a. b). is de-
fined by
(a, b) akbk.
kai
Also, for a E l8" we define the norm or length of the vector a by
7.4.7 DEFINITION A set X with two operations " + ", vector addition, and " ", scalar
multiplication, satisfying
x+ y E X for all x, y E X, and
c'xEX forallxEX,cER
is a vector space over R if the following are satisfied:
(a) x + y = y + x. (commutative law)
(b) x + (y + z) = (x + y) + z. (associative law)
(c) There is a unique element in X called the zero element, denoted 0, such that
x+0=x forallxEX.
(d) For each x E X, there exists a unique element -x E X such that
x+(-x)=0.
(e)
(f) +a'y.
(g) (a+b)'x=a'x+b'x.
(h) I'x=x.
It is clear that R" with + and ' defined by
a+b=(a,+b,,. . .,an+bn),
c' a = (cab. . .,Can)
is a vector space over R. Similarly, if addition and scalar multiplication of sequences
{ak}, {bk} in 12 are defined by
{ak} + {bk} = {ak + bk},
c {ak) = {cak}, c E R,
then it is easily shown that 12 is a vector space over R. The fact that {ak + bk} and {cak}
are in 12 is part of the statement of Theorems 7.4.5 and 7.4.6. The zero element 0 in l2
is the sequence {ak} where ak = 0 for all k E N.
In addition to being vector spaces, the spaces R" and 12 are also examples of
normed linear spaces. The concepts of a "normed linear space" and a "norm" are de-
fined as follows:
Inequality (d) is called the triangle inequality for the norm 11 11. By Theorem 7.4.6,
is a nonmed linear space. Additional examples of normed linear spaces will
(12, II Il2)
occur elsewhere in the text and the exercises.
7.4 Square Summable Sequences 311
If (X, 11 11) is a normed linear space, the distance d(x. y) between two points
x, y E X is defined as d(x, y) = II x - y II. From the definition of the norm II 1. it im-
mediately follows that
(a) d(x, y) ? 0 for all x, y E X.
(b) d(x,y) = 0 if and only if x = y,
(c) d(x, y) = d(y, x), and
(d) d(x, y) s d(x, z) + d(z. y) for all x, y, z e X.
Thus d is a metric on the vector space X. As in Definition 2.1.6, fore > 0. p E X.
the e-neighborhood Nf(p) is defined as
For example, in 1I2. with norm II II2 an e-neighborhood of a = (a,, a2) is the set of
points x = (x,, x,) satisfying (x, - a,)2 + (x, - a2)2 < e-, that is. an open disc with
center a and radius a (See Figure 7.2). In 1183, an e-neighborhood of a is simply an open
ball with center a and radius e.
X2
a,
a,
Since the notions of "convergence of a sequence" (2.1.7) and "limit point of a set"
(2.4.5) are defined in terms of e-neighborhoods, both of these concepts have analogous
definitions in a normed linear space (X, 11 11). Thus for example, a sequence in X
converges to x E X, if given e > 0, there exists n,, E IO' such that IIx - xfl < e for
all n E N. n ? no. If this is the case, we say that x,, converges in the norm II II to x.
This type of convergence, referred to as norm convergence, will be encountered again
later in the text.
As a general rule, many of the results involving sequences of real numbers are still
valid in the setting of a normed linear space. This is especially true for those theorems
whose proofs relied only on properties of the absolute value. Great care, however, must
312 Chapter 7 Series of Real Numbers
be exercised with theorems that rely on the supremum property of R. For example. the
Bolzano-Weierstrass theorem fails for (12, 11 112) (Exercise 7). On the other hand, the
Bolzano-Weierstrass theorem still holds in (R". Ii IIz) (Exercise 6).
6 EXERCISES 7.4
1. Determine which of the following sequences are in 12.
1 ( ] " c I In k
d.
J !,in kj"
sa. b.
Ink k=z t Vk Ink I k- 2 Vk k-z k k-1
2. Determine all values of p E R such that the given sequence is in 12.
10. *For a, b E R", use the law of cosines to prove that (a. b) = Ilal2llbf 2cos 0, where B is the angle between the
vectors a and b.
11. Let (X, II V) be a normed linear space. Prove that I lIxII - Ilyll l 5 Ox - yll for all x. y E X.
12. Let 1' denote the set of all sequences {ak} satisfying H{ak}ll, _ Ik , lakl < oo.
a. Prove that (1', II II,) is a normed linear space.
b. Prove that 1' C 12.
13. Determine all values of p E R for which each of the sequences in Exercise 2 is in 1'.
14. Let X be a nonempty set, and let S(X) denote the set of all bounded real-valued functions on X. For f E 31(X), set
Ilf L = sup{I f(x) I : x r= X}. Prove that (11(X), 11 IL) is a nonmed linear space.
Notes 313
NOTES
The geometric series is perhaps one of the most important gence. An excellent illustration of these eighteenth-century
series in analysis. It forms the basis for the proof of the techniques is Euler's derivation of the sum of the series
ratio and root test and thus also for the study of conver-
gence of power series. In the seventeenth and eighteenth 1I I
+3-+..
I
MISCELLANEOUS EXERCISES
1. Given two series Ek oak and Boko.o bk, set c, = I , _ o ak b _ k, n = 0, 1.2.. . ..The series !.O c is called the
Cauchy product of E ak and I bk.
a. If I ak andd I bk converge absolutely, prove that E c converges absolutely and that
-oc"-\koak)(kobk).
b. Let ak = bk = (- I)k/, k = 0, 1, 2. ... Prove that the Cauchy product of F. ak and B bk diverges.
c. Prove that the result of part (a) is still true if only one of the two series converge absolutely: the other series
must still converge.
2. Let X be a nonempty set. If f is a real-valued function on X, define
IIf112= I I
a. Prove that for f, g E R[a, b], f. I jr (x)g(x) I dx s 11 f 11211 g I1r
b. Forf, g E JL[a, b], prove that 11f + g 112 s 11f112 + 118112
c. Prove that 11 16 defines a norm on %[a, b], the space of continuous real-valued functions on [a, b).
d. is11 I6 anorm on Jt(a,b]?
4. As in the previous exercise, let b] denote the vector space of continuous real-valued functions on (a, b). For
f E 19[a, b], set IIf1I. = max{ If(x) I : x r= [a, b]}.
a. Prove that 11II. is a norm on %(a, b].
b. Prove that a sequence {f.) in 4[a, b] converges to f e 14[a, b] if and only if given e > 0. there exists
n. E N such that I f (x) - f (x) I < e for all x E [a, b] and all n E N, n a n,.
DEFINITION
(a) Let (X,11 11) be a normed linear space, and let E C X. A function f : E -+ R is continuous at p E E If given
e > 0, there exists a S > 0 such that If(x) - f(p) I < e for all x E E with IIx - pIi < S.
(b) Let X be a vector space over It A function f : X -+ R is linear if f(ar + by) = af(x) + bf(y) for all x, y E X
and a, b E R.
5. Let (X, II II)be a normed linear space and let f : X - R be a linear function. Prove that the following are equivalent.
a. f is continuous at some p E X.
b. f is continuous at 0.
Supplemental Reading 315
c. There exists a positive constant M such that If(x)1 <- MIIxiifor all x E X.
d. There exists a positive constant M such that I f(x) - f(y)I <_ Mix - ylifor all x, y E X.
e. f is uniformly continuous on X.
X
6. For fixed b E 12, define 1' : 12 -+ R by r(a) = (a, b) = I a(k)b(k). Prove that r is a continuous linear function on l
k=1
7. As in Exercise 3, let b] denote the vector space of continuous real-valued functions on [a. b] with norm l N2.
For fixed g E 9t[a, b], prove that r defined by r(f) = fa f(x)g(x) dx is a continuous linear function on `g[a, b].
SUPPLEMENTAL READING
Behforooz, G. H.. "Thinning out the harmonic series;' Jungck, G., "An alternative to the integral test;' Math.
Math. Mag. 68 (1995), 289-293. Mag. 56 (1983), 232-235.
Boas, R. P., "Estimating remainders:' Math. Mag. 51 Katz, V. J "Ideas of Calculus in Islam and India," Math.
(1978), 83-89. Mag. 68 (1995),163-174.
Boas, R. R, "Partial sums of infinite series, and how they =I
grow," Amer. Math. Monthly 84 (1977), 237-258.
Kortam, R. A., "Simple proofs for 7 2 and
k_I k 6
Cowen, C. C., Davidson. K. R., and Kaufman, R. P., 00 x2
"Rearranging the alternating harmonic series," sin x =x l - a f ; 'Marh. Mag. 69 (1996),
k.1 kir'
Amer. Math. Monthly 87 (1980), 817-819. 122-125.
Efthimiou, C. J., "Finding exact values for infinite se-
ries:' Math. Mag. 72 (1999), 45-51. Maher, P., "Jensen's inequality by differentiation:' Math.
Goar, M., "Olivier and Abel on series convergence: An Gaz. 73 (1989). 139-140.
episode from early 19th century analysis:' Math. Tolsted, E., "An elementary derivation of the Cauchy,
Mag. 72 (1999), 347-355. Holder, and Minkowski inequalities from Young's
Grabinger, Judith V., The Origins of Cauchy's Rigorous inequality," Math. Mag. 37 (1964), 2-12.
Calculus, The MIT Press, Cambridge, Massachu- Young, R. M., "Euler's constant:' Math. Gazette 75
setts, 1981. (1991), 187-190.
pl Sequences and Series
J of Functions
8.1 Pointwise Convergence and Interchange of Limits
8.2 Uniform Convergence
8.3 Uniform Convergence and Continuity
8.4 Uniform Convergence and Integration
8.5 Uniform Convergence and Differentiation
8.6 The Weierstrass Approximation Theorem
8.7 Power Series Expansions
8.8 The Gamma Function
In this chapter we will study convergence properties of sequences and series of real-
valued functions defined on a set E. In most instances E will be a subset of R. Since we
are dealing with sequences and series of functions, there naturally arise questions in-
volving preservation of continuity, differentiability, and integrability. Specifically. is the
limit function of a convergent sequence of continuous, differentiable, or integrable func-
tions again continuous, differentiable, or integrable? We will discuss these questions in
detail in Section 8.1 and show by examples that the answer to all of these questions is
generally no. Convergence by itself is not sufficient for preservation of either continu-
ity, differentiability, or integrability. Additional hypotheses will be required.
In the 1850s Weierstrass made a careful distinction between convergence of a se-
quence or series of numbers and that of a sequence or series of functions. His concept
of uniform convergence is the additional hypothesis required for the preservation of
continuity and integrability. It was also Weierstrass who constructed a continuous but
nowhere differentiable function, and who proved that every continuous real-valued
function on a closed and bounded interval can be uniformly approximated by a poly-
nomial. As prominently as Cauchy is associated with the study of sequences and series
of numbers, Weierstrass is likewise associated with the study of sequences and series of
functions. For his many contributions to the subject area, Weierstrass is often referred
to as the father of modern analysis.
The study of sequences and series of functions has its origins in the study of power
series representation of functions. The power series of In (I + x) was known to Nicolaus
Mercator (1620-1687) by 1668, and the power series for many of the transcendental
317
318 Chapter 8 Sequences and Series of Functions
functions such as Arctan x, Arcsin x, among others, were discovered around 1670 by
James Gregory (1625-1683). All of these series were obtained without any reference to
calculus. Newton's first discoveries, dating back to the early months of 1665. resulted
from his ability to express functions in terms of power series. His treatise on calculus,
published posthumously in 1737, was appropriately entitled A treatise of the method of
fiuxions and infinite series. Among his many accomplishments. Newton derived the
power series expansion of (1 + x)" using algebraic techniques. This series and the
geometric series were crucial in many of his computations. Newton also displayed the
power of his calculus by deriving the power series expansion of ln(I + x) using term-
by-term integration of the expansion of 1/(1 + x). The mathematicians Colin Maclau-
rin (1698-1746) and Brooks Taylor (1685-1731) are prominent for being the first math-
ematicians to use the new calculus in determining the coefficients in the power series
expansion of a function.
defines a function on F. The fiaiction f is called the limit of the sequence If,).
In terms of e and n, the sequence {f,) converges pointwise (of if for each x E E,
given e > 0, there exists a positive integer n,, = e) such that
If,,(x) - f(x)I < e
8.1 Pointwise Convergence and Interchange of Limits 319
for all n ? na. The expression no = na(x, e) indicates that the positive integer n, may
depend both on e and x E E.
If as above, {fa}°a° is a sequence of real-valued functions on a nonempty set E,
1
then with the sequence {fa} we can associate the sequence {S0} of nth partial sums,
where for each n E N, S. is the real-valued function on E defined by
Suppose fa : [a, b] -+ R for all n e Ni, and f(x) each x E [a, b].
Among the questions we want to consider are the following:
(c) If for each n E N the function fa is Riemann integrable on [a, b], is f Riemann
integrable? If so, does
rb jb
f= a--.f f.?
a a
We now provide a number of examples to show that the answer to all of these ques-
tions is generally no.
320 Chapter 8 Sequences and Series of Functions
8.12 EXAMPLES
(a) Let E = [0, 1 ], and for each x E E, n E N. let f"(x) = x". Clearly each f is con-
tinuous on E. The graphs of f, f,-, f4, f6, and f,i, are given in Figure 8.1. Since fn(1) = I
for all n, Jim 1. If 0 :5 x < 1, then by Theorem 2.2.6(e), lim fn(x) = 0. There-
n-r x
fore.
fk(x)= x (1
L=O k=0 + x2
8.1 Pointwise Convergence and interchange of limits 321
We now show that this series converges for all x E I and also find its sum f. If x = 0.
then fk(O) = 0 for all k, and thus f(0) = 0. If x * 0, then 1/(1 + x2) < 1, and hence
by Example 2.7.2(a),
X2
71
(- = 1 + xz = f(x).
k=O
Therefore,
0, x = 0,
f(x) 1 +xz, x#0,
which again, is not continuous on R.
(c) Let {xk} be an enumeration of the rational numbers in [0, 1]. For each n E N,
define fn as follows:
J 0, if x = xk, 1 <_ k 5 n,
fn(x) = 1, otherwise.
Since each fn is continuous except at xi, ... , xn, fn is Riemann integrable on [0, 1 ]
with fn fn(x) dx = 1. On the other hand,
if x is rational,
lim f(x) = s10,1, if x is irrational,
which by Example 6.1.6(a) is not Riemann integrable on (0, 11.
(d) For x E [0, 1], n E N, let fn(x) = nx (I - x2)n. Since each fn is continuous, fn is
Riemann integrable on [0, 1 J. If 0 < x < 1, then 0 < 1 - x'' < 1, and thus by
Theorem 2.2.6,
lim nx(1 - X'-)n = 0.
nyoc
1 n
dx = n x (1 - xZ)r dx =
10 0
1
2n+1
Therefore.
lim
n-ioc
f' frt(x) dx = 2 # 0 = I f (x) dx.
0 J11
(e) As our final example, consider fn(x) = (sin nx)/n, x E R. Since I sin nx -5 1 for all
x E=- R and n E N,
f (x) = lim
n-+oo
fn(x) = 0 for all x E R.
322 Chapcer8 Sequences and Series of Functions
EXERCISES 8.1
1. Find the pointwise limits of each of the following sequences of functions on the given set.
sin nx
*a. 11 _, }, xE(0,oo) b.
11+nx
. X E f 0. oo)
n
<xs 2n
0, ?<x<_1.
it
a. Sketch the graph of f, for it = 2, 3, and 4.
b. Prove that lim f,(x) = 0 for each x E [0, 1 ].
c. Show that fo fn(x) dx = I for all n = 2, 3.... .
4. Let g (x) = e-n`/n, x E 10, oo), it E N. Find lim
n-+oo
g,(x) and lim
ny0o
g;,(x).
5. Let f,(x) _ (x/n)e- `1n, X E [0.00).
a. Show that tim fn(x) = 0 for all x E [0, oo).
b. Given e > 0, does there exist an integer it, E N such that If,(x) I < e for all x E (0, oo) and all it z it,,.
(Hint: Determine the maximum of fn on [0, oo).)
c. Answer the same question as in pan (b) for x E [0, a], a > 0.
6. *If a,, m ? 0, it, m E N, prove that
cc cc - cc
71 71 an.,, = L an.,,,
n=Im=1 n=1
with the convention that if one of the sums is finite, so is the other, and equality holds. Conversely, if one is infi-
nite, so is the other.
8.2 Uniform Convergence 323
Nx) - L WI < E
for all n ? n0. The key here is that the choice of the integer n,, may depend not only on e.
but also on x E E. If this dependence on x can be removed, then we have the following.
f(x)- E <f(x)
for all x E E and n ? n,,. If E is a subset of R, then the geometric interpretation of the
above inequality is that for n ? n,,, the graph of y = f"(x) lies between the graphs of
y = f(x) - e and y = f(x) + E. This is illustrated in Figure 8.2 with E = [a, b].
8.2.2 EXAMPLES
(a) For x E [0, 1 ], n E t\i, let f"(x) = x". By Example 8.1.2(a), the sequence {f.1 con-
verges pointwise to the function
0sx< I,
f(x) _ {°, x= 1.
We now show that the convergence is not uniform. If the convergence were uniform,
then given E > 0, there would exist a positive integer n, such that I "(j) - f(x) 1 < e
for all n >_ n" and all x E [0, 1 ]. In particular,
a b
Figure 8.2
This, however, is a contradiction if e < 1. Even though the convergence is not uniform
on (0, 1 ], the sequence does converge uniformly to 0 on [0, a] for every a, 0 < a < 1.
This follows immediately from the fact that for x E [0, a], LAW I = I x' I -S a".
(b) Consider the series
00
k71
{ixe-k'' - (k - 0:9 x < 00.
Since the series is a telescoping series, the nth partial sum S"(x) is given by
S"(x) = nxe"";.
It is easily shown (Exercise Id, Section 8.1) that
S(x) = lim
M- 00
S"(x) = 0 for all x E (0, oo).
We now show that the convergence is not uniform. Suppose that the sequence {S"} con-
verges uniformly to 0 on [0, oo). Then if we take e = 1, there exists a positive integer
n0 such that
IS.(x) - S(x) l = S"(x) = nxe -"'_ < I
for all n a n, and x E [0, oo). However, for each n E N, by the first derivative test S.
has a maximum at x, = with
M. = max S"(x)
F2e_
This term, however, is greater than 1 for n =' 6. Thus the convergence is not uniform.
The graphs of S4, Ss, and S16 are given in Figure 8.3. Since the maximum of each S.
moves along the x-axis as n -* oo, such functions are often referred to as "sliding-
hump" functions. For this example, M. -+ oo as n -+ oo.
8.2 Uniform Convergence 325
for all x E E. Since the above holds for all m ? n the sequence {f.} converges uni-
formly to f on E. Q
The analogous result for series is as follows.
326 Chap{er 8 Sequences and Series of Functions
8.2.5 THEOREM Suppose the sequence (f.} of real-valued functions on the set E converges
pointwise to f on E. For each n E N, set
p
M. sup
I fn(x) - ! (x) I I.
8.2.6 EXAMPLES
(a) To illustrate the previous theorem we consider the sequence
S"(x) = nxe-"XI, n = 1, 2, .. .
of Example 8.2.2(b). For this sequence, lim S"(x) = 0 for all x, 0 s x < oo. However,
n
M. = sup S"(x)
xE[0,oo)
which diverges to oo. Thus the convergence is not uniform on (0, oo). However, the se-
quence {S"} does converge uniformly to the zero function on [a, oo) for every a > 0
(Exercise 6).
(b) Consider the sequence { f} of Example 8.2-2(a) given by f"(x) = x", x E [0,1 ].
This sequence converges pointwise to the function f (x) = 0, 0 < x < 1, and f (l) = 1.
Since
Jx", 0x :S x < i,
I.f"(x) -f(x)I = l0, x= = 1,
we have
Mn = SU Ifnx) -f(x)I = 1.
:E[0, Ij
Thus since {M"} does not converge to zero, the sequence { f"} does not converge uni-
formly to f on [0, 1 ]. On the other hand, if 0 < a < I is fixed, then
Mn I fn(x)I = a".
xE[ Dal
8.2 Uniform Convergence 327
Since lim an = 0, by Theorem 8.2.5 the sequence { fn} converges to zero uniformly on
n-400
[0, a] for every fixed a, 0 < a < 1.
n n n
8.2.8 EXAMPLES
.(a) If Zak converges absolutely, then since Iak cos kxl s IakI for all x E R, by the
Weierstrass M-test, the series I ak cos kx converges uniformly on R. Similarly for the
series I ak sin kx. In particular, the series
00
cos kx sin kx
7
1
p>
k-1 kr k=1 k°
converge uniformly on R.
(b) Consider the series 7,I (x12)k. This is a geometric series that converges for all
x E R satisfying I x I< 2. If 0< a < 2 and I x s a, then
(2)k.
\2)k I
Since a/2 < 1 the series 7, (a/2)k converges. Thus by the Weierstrass M-test, the se-
ries I', (x/2)k converges uniformly on [-a, a] for any a, 0 < a < 2. The series,
however, does not converge uniformly on (-2, 2) (Exercise 11).
Although the Weierstrass M-test automatically implies absolute convergence, the
following example shows that uniform convergence as a general rule does not imply ab-
solute convergence.
328 Chapter 8 Sequences and Series of Functions
Thus by Theorem 7.2.3, the series I (-I)k "ak(x) converges for all x E [0, 1].
Let
S(x) _ (-1Y+'ak(x).
k=1
If S (x) is the nth partial sum of the series, then by Theorem 7.2.4
Thus converges uniformly to S on [0, 1 ]. However, the given series does not con-
verge absolutely when x = 1.
The converse is also false; absolute convergence need not imply uniform conver-
gence! As an example, consider the series Y,1 x2(1 + x2)-k of Example 8.1.2(b).
Since all the terms are nonnegative, the series converges absolutely to
1 0, x = 0,
f (x) = l + x2, x # 0,
on R. However, as a consequence of Corollary 8.3.2 of the next section, since f is not
continuous at 0, the convergence cannot be uniform on any interval containing 0.
The series I (- 1)t+ txk/k, x E [0, 1]. also provides an example of a series that
converges uniformly on [0, 1 ] but for which the Weierstrass M-Test fails.
EXERCISES 8.2
1. Prove Theorem 8.2.5.
2. a. If (f.) and converge uniformly on a set E, prove that {f. + converges uniformly on E.
+b. If (f.1 and converge uniformly on a set E, and there exist constants M and N such that I f.(x)I <- M and
1g.(x)l s N for all n E N and all x r= E, prove that { converges uniformly on E.
c. Find examples of sequences f f.) and that converge uniformly on a set E, but for which { f does not
converge uniformly on E.
3. Show that if (f.1 converges uniformly on (a, b) and (f. (a)} and { ff (b)} converge, then (f.) converges uniformly
on (a, b].
4. Let f .(x) = nx(1 - x2r, 0 5 x 5 1. Show that does not converge uniformly to 0 on (0, 1
5. Let 0 s x 5 I.
x"'
a. Show that (f.} converges uniformly to 0 on [0, a] for any a, 0 < a < 1.
b. Does I f.1 converge uniformly on [0, fl?
8.2 Uniform Convergence 329
6. Show that the sequence {nxe converges uniformly to 0 on [a. no) for every a > 0.
7. For each n E N. set
x
x + - sin nx, x r= R.
n
Show that the sequence { converges uniformly to f(x) = x for all x E (-a, a], a > 0. Does (f.1 converge uni-
formly to f on R?
8. Show that each of the following series converge uniformly on the indicated interval.
Ca. ; k-
00
k-1
,
+ xk-1
0 5 x < oc b, i e-k`xt, 0 5 x< oo
00 00 (-l)k*1
"c. Y, 1 5 x < on d. , 0 5 x< on
k-1 k-, k+x
9. Test each.of the following series for uniform convergence on the indicated interval.
sin 2kx 00 xk
'a.
00
x E G8 b F, x5I
(2k + 1 ?/2 ' k(In k)2 '
*c' k
1
(kx+2(k+1)x+2
1
05x51
00 (-1)k+lx2k+l
(xi<_1 e.
00
//\
Xsinikol, p> 1.JxIs2
d. k=O 2k+1
10. Show that each of the following series converge uniformly on (a, oo) for any a > 0, but do not converge uni-
formly on (0, oo).
M I I
'a. 2 I + k2x
00
b. I ii-+;
11. Show that the series E, , (x/2)k does not converge uniformly on (-2, 2).
12. If Jk o ak converges absolutely, prove that I-k o akxk converges uniformly on [ -1. I ].
13. If I .oak converges, prove that Iku akxk converges uniformly on [0, 1].
14. Let {ck) be a sequence of real numbers satisfying 7, Ickl < on, and let {xk} be a countable subset of (a, b]. Prove
that the' series lk , ck !(x - xk) converges uniformly on [a, b]. Here ! is the unit jump function defined in 4.4.9.
15. Dirichlet'17est for Uniform Convergence: Suppose {)k) and {gk} are sequences of functions on a set E satisfying
(a) the partial sums Ik., gk(x) are uniformly bounded on E; i.e., there exists M > 0 such that
<Mforalln E NlandxEE,
(b) k(x) >- fk I(x) > 0 for all k E hi and x E E, and
(c) jim fk(x) = 0 uniformly on E.
Prove that I fk(x)gk(x) converges uniformly on E.
16. Prove that
. sin kx . t os kz (p > 0)
k=1 kP k-1
converge uniformly on any closed interval that does not contain an integer multiple of 21r.
330 Chapter 8 Sequences and Series of Functions
if ii < x `
f(x)= 2" 2"
0, elsewhere.
Prove that 1 1 fn(x) converges uniformly on [0, 1 ], but that the Weierstrass M-test fails.
18. 'Let F0 be a bounded Riemann integrable function on [0. I ]. For n E 101, define F .(x) on [0, 1 ] by
F (x) = fo F.- 1(t) dt. Prove that 7,; o Fk(x) converges uniformly on (0. 1 ].
lim I lim
X_p \n-iW /I = lim I lim
n-.00 \X- p
for all n, m _a n and all x E E. Since inequality (2) holds for all x E E. letting x -+p
gives
IAn - Amt s e for all n, m ? no.
It remains to be shown that lim f(x) = A. Again, let e > 0 be given. First, by the
uniform convergence of the sequence { fn(x)} and the convergence of the sequence
there exists a positive integer m such that
IA - Aml < 3.
< e + I fm(x) - A, I
S(x) _ 71 fn(x)
M=1
is continuous on E.
Proof. (a) If p E E is an isolated point, then f is automatically continuous at p. If
p E E is a limit point of E, then since f is continuous for each n E N,
Jim MX) = MP).
Thus by the previous theorem,
m f(x) = urn fn(P) = f(p).
s n-.oo
Therefore, f is continuous at p.
332 Chapter 8 Sequences and Series of Functions
SS(x) = fk(X)
krl
8.3.3 EXAMPLE The sequence {xn},.,, x E [0, 1 ], of Example 8.1.2(a) does not converge
uniformly on [0, 1 ] since the limit function
o x < 1,
}.(x) =
J0,
1, x= 1,
is not continuous on [0, 11. Likewise, the series
00
( 1 __ 0, x = 0.
k ,1 + x2 Jk 1 + x2, X # 0,
of Example 8.1.2(b) cannot converge uniformly on any interval containing 0, since the
sum of the series is not continuous at 0. 0
Dini's Theorem'
A natural question to ask is whether the converse of Corollary 8.3.2 is true. Namely, if
f and f. are continuous for all it and f -+f pointwise, is the convergence necessarily
uniform? The following example shows that this need not be the case. However, in The-
orem 8.3.5 we will prove that with the additional assumption that the sequence { f (x)}
is monotone for all x, the convergence is indeed uniform.
max
Osssl S (x) = n ,
by Theorem 8,2.5 the convergence cannot be uniform.
8.3.5. THEOREM (Dins) Suppose K is a compact subset of R and (f.} is a sequence of con-
tinuous real-valued functions on K satisfying the following:
(a) {f,,) converges pointwise on K to a continuous function f, and
(b) f (x) ? f,,. 1(x) for all x E K and n E N.
Then converges uniformly to f on K.
1. This topic is not required in subsequent sections and thus can be omitted on first reading.
8.3 Uniform Convergence and Continuity 333
.Therefore, p E K,,. Thus K,, is closed, and as a consequence of Theorem 3.2.5, also
compact. Furthermore, since g (x) >_ g..#) for all x E K,
K,,. i C K,, foralln.
Finally, since g,,(x) -+ 0 for each x E K,
00
n-i
However, by Theorem 3.2.7 this can only be the case if K,,, = 4 for some n E N. Thus
for all n ? no,
for allxEK.
Therefore the sequence {g,} converges uniformly to 0 on K. Q
8.3.6 EXAMPLE We now provide an example to show that compactness is required. For
each n E N, set
I
nx+ V
0<x<l.
Then { f,(x)} monotonically decreases to f (x) = 0 foreach x E (0, 1). However, since
linof,,(x) = 1, by Theorem 8.3.1 the convergence cannot be uniform.
2. This topic can be omitted on first reading. The concept of a complete normed linear space (Definition
8.3.10) is only required in Section 10.8.
334 Chapter 8 Sequences and Series of Functions
That II 11. is indeed a norm on b] is left to the exercises (Exercise 12). We now
take a closer look at uniform convergence of a sequence of continuous real-valued func-
tions and introduce the concept of convergence in norm. We first note that a sequence
of continuous real-valued functions on [a, b] is nothing but a sequence in the set
%[a, b]. Suppose that the sequence { in `P[a, b] converges uniformly to f on (a. b].
Then by Corollary 8.3.2, the function f E'F[a, b]. By the definition of uniform con-
vergence, given c > 0, there exists a positive integer n,, such that
I fn(x) - Ax) I < E
for all x E [a, b] and all n ? n,,. This is illustrated in Figure 8.2. Since f - f is con-
tinuous, if n z no, by Corollary 4.2.9
IIf,, - fllu = max{ I f0(x) - f(x)l : x E [a, b]} < e.
Therefore, 11f. - f Il u < e for all n ? no.
Conversely, suppose f, f E 'P[a, b] satisfy the following: For each e > 0 there ex-
ists a positive integer n,, such that N f - f j. < e for all n ? n0. But then If(x) - f (x)I
< E for all x E [a, b] and all n ? n0; i.e., { f on [a, b]. This
proves the following theorem.
8.3.9 DEFINITION Let (X, 11 11) be a normed linear space. A sequence in X converges
in norm if there exists x E X such that for every e > 0, there exists a positive integer
ne such that IIx - x.11 < e for all n ? no. If this is the case, we say that converges
in norm to x, and denote this by x x as n -+ oa.
i
From the definition it is clear that a sequence in X converges in norm to x E X
if and only if lim 11x - x11 = 0. Also, as in the proof of Theorem 2.1.10, if con-
verges in norm "&n its limit is unique. Using the norm it is also possible to define what
we mean by a Cauchy sequence in a normed linear space.
8.3.10 DEFINITION
(a) A sequence {x} in a normed linear space (X, ll h) is a Cauchy sequence if for
every e > 0, there exists a positive integer n, such that
Ilxn - xmll < E
for all integers n, m ? n,,.
(b) A normed linear space (X,11 II) is complete if every Cauchy sequence in X
converges in norm to an element of X.
8.3 Uniform Convergence and Continuity 335
As for sequences of real numbers, every sequence {x,,} in X that converges in norm
to x E X is a Cauchy sequence. In Theorem 2.6.4 we proved that the normed linear space
(R,I I) is complete. The following theorem proves that (%[a, b], II ll.) is also complete.
8.3.11 THEOREM The normed linear space (`P[a, b], II II.) is complete.
Proof. Let {fn} be a Cauchy sequence in %[a, b]; i.e., given e > 0. there exists a
positive integer no such that Ilfn - fmll. < E for all n,m >- no. But then
Ifn(x) - fm(x)I Il fn - fmllu < E
for all x E [a, b] and all n, m a n,. Thus by Theorem 8.2.3 and Corollary 8.3.2, the se-
quence { fn} converges uniformly to a continuous function f on [a, b]. Finally. since the
convergence is uniform, given e > 0, there exists an integer no such that
8.3.12 DEFINITION Let (X, N II) be a nonmed linear space. A mapping (function) T : X -+X
is called a contraction mapping (function) if there exists a constant c, 0 < c < 1.
such that
8.3.13 THEOREM Let (X, II II) be a complete normed linear space and let T : X -+ X be a
contraction mapping. Then there exists a unique point x E X such that T(x) = x.
Proof. Suppose T : X - X satisfies 11T (x) - T(y)p <- cll x - yq for fixed c,
0 < c < 1. and all x, y e X. We now define a sequence {xn} in X as follows: Let x, E X
be arbitrary. For n E N set xn = T(xn_,). That is, x, = T(x,), x2 = T(xi), etc. Since
IIxn+1 - xnll = lIT(xn) - T(xn-1)II cllxn - xn-111,
the sequence {xn} is a contractive sequence in X (see Definition 2.6.6). An argument
similar to the one used for a contractive sequence in Section 2.6 shows that
C.
Ilxn+m - xnll C 1 C Ilx, - xoll
336 Chapter 8 Sequences and Series of Functions
for all n, m E N. The details are left as an exercise (Exercise 10). Since c" -* 0. the se-
quence {x"} is a Cauchy sequence in X. By completeness, the sequence {x"} converges
(in the norm) to some x E X. But by continuity of the mapping T,
x = lim x"+t = lim T(x") = T(x);
EXERCISES 8.3
1. *Show that the series I'U x (1 - x)0 cannot converge uniformly for 0 s x 5 1.
2. For n E N. let f .(x) = x"/(I + x"), x (5 [0, 1 ]. Prove that the sequence f f.) does not converge uniformly on
[0, 1].
3. Give an example of a sequence f f.) of functions on [0, 1]. such that each f" is not continuous at any point of
[0, 11, but for which the sequence {f.) converges uniformly to a continuous function f on [0. 11.
4. *Suppose that f is uniformly continuous on R. For each n E N. set f"(x) = f(x + Prove that the sequence ff.)
converges uniformly to f on R.
S. Let (f.1 be a sequence of continuous real-valued functions that converges uniformly to a function f on a set
E C R. Prove that liim f"(x") = f(x) for every sequence {x"} C E such that x" -*x E E.
6. *Let E C R and let D be a dense subset of E. If (f.} is a sequence of continuous real-valued functions on E. and
if if.) converges uniformly on D. prove that (f.) converges uniformly on E. (Recall that D is dense in E if every
point of E is either a point of D or a limit point of D.)
7. Find a sequence {f"} in IC[0, I ] with II f.11. = I such that no subsequence of (f,) converges (in norm) in %[0. C.
8. Suppose {f.1 is a sequence of continuous functions on [a, h] that converges uniformly on [a, b]. For each
x E [a, b], set g(x) = sup{f" (x)}.
a. Prove that g is continuous on [a, b].
b. Show by example that the conclusion may be false if the sequence {f"} converges only pointwise on [a. b].
9. For each n E N and x E It, set f,(x) = (1 + )". Use Dini's theorem to prove that the sequence {f"} converges
uniformly to e' on [a, b] for any fixed a, b E R.
10. Let (X, II II) be a normed linear space and let T : X -* X be a contraction mapping with constant c, 0 < c < 1. If
{x,} is the sequence in X as defined in the proof of Theorem 8.3.13, prove that
11x... - x,, lI S (c"/(1 - c)) II x, - x.11 for all n, m E N.
11. Define T : [0, 1 ] -a %[0, 1 ] by (Tto)(x) = fo cp(t) dt, 0 S x S 1, v E %[0, 1 ], and set T2 = ToT.
a. Prove that I (T2V)(x)I s z X211011..
b. Show that T2 is a contraction mapping on `P[0, I ] and thus has a fixed point in `P[0. 1 ].
c. Prove that T has a fixed point in `4[0, 1 ].
12. Prove that b], f Q is a normed linear space.
13. Prove that (12, Ii 02) is a complete normed linear space.
8.4 Uniform Convergence and Integration 337
8.4.1 THEOREM Suppose fn E 9t[a, b] for all n E N. and suppose that the sequence {fn}
converges uniformly to f on [a, b]. Then f E gt[a, b] and
b Jb
f(x) dx = li m
f
a
n-M
a
fn(x)dx.
Therefore,
0-
fa
b
f- rb
Ja
f<2En[b-a].
rh rh
f(x) dx - Ja fn(x)dxI G En[b - Q],
Ja
and thus
b b
hm
fa
fn(x) dx = j f(x) dx.
a
where the series converges uniformly on [a, b], then f E Jt[a, b] and
b m b
Ja
f(x) dx = 7, f A (x) dx.
k.1 a
Proof. Apply the previous theorem to S"(x) = 1,"t. I fk(x), which by Theorem 6.2.1
is integrable for each n E N. O
Although uniform convergence is sufficient for the conclusion of Theorem 8.4.1, it
is not necessary. For example, if f"(x) = x", x E [0, 1 ], then {f"} converges pointwise,
but not uniformly, to the function
J0, 0 : 5x<1 ,
f(x) = 1, x = 1.
The function f E 9t[0, 1 ] and
°n+ 1
=0= f f(x) dx.
In Section 10.6, using results from the Lebesgue theory of integration, we will be able
to prove a stronger convergence result that does not require uniform convergence of the
sequence If.}. However, it does require that the limit function f is Riemann integrable.
For completeness we include a statement of that result at this point.
8.4.3 THEOREM (Bounded Convergence Theorem) Suppose f and fn, n E N, are Rie-
mann integrable functions on (a, b] with lim f"(x) = f(x) for all x E [a, b]. Suppose
n-.oo
also that there exists a positive constant M such that [fn(x) I < M for all x E [a, b] and
allnE N. Then
It is easily checked that the sequence If.} on (0, 1], where for each n E N
fn(x) = x", satisfies the hypothesis of the prwtvious theorem. Also, since the limit func-
tion f is continuous except at x = 1, f E J.[0, 1].
EXERCISES 8.4
f (akx*)dx=
0 7 I , k+1
ak
8.5 Uniform Convergence and Differentiation 339
2. Let a be a monotone increasing function on [a, b]. Suppose f, E A(a) on [a, b) for all n E R;. and suppose that
the sequence converges uniformly to f on :a. b,. Then f E .R(o) and
(h (
f f da = lim f. f da.
3. For each n E N. let nx/(l + nx), x E [0, I ] . Show that the sequence { f } converges pointwise. but
not uniformly, to an integrable function f on i 0, 1 ]. and that lim f,' f;,(x) dx = fol f(.r) dr.
4. *If f is Riemann integrable on [0, 11. use the bounded convergence theorem to prove that Iim f r f(x) dr = 0.
5. Let (f,} be a sequence in :7i[a, b] that converges uniformly to f E .R[a, b'. For n E ) set F,(x) = f' f , and let
F(x) = f f, r E [a, b]. Prove that converges uniformly to F on [a, bl.
6. Suppose f : [0, I ] R is continuous. Prove that Iim f f(r") dr = f (0).
7. *Let {r"} be an enumeration of the rational numbers in [0, 11, and let f : [0, 11 -- R be defined by
Y I
I.X
for all x e [a, b]. The result can now be proved using Corollary 8.3.2 and Theorem
8.4.1. The details are left to the exercises (Exercise 2).
Proof. Let e > 0 be given. Since { f (x,)} converges and If.') converges uniformly,
time exists no E RI such that
and
a
I f"(t) - f",(t) I < 2(b a) for all t r= [a, b] and all n, m ? n,. (5)
Apply the mean value theorem to the functions f - fm with n, m ? n, fixed. Then for
x, y E [a, b], there exists t between x and y such that
Take y = x, in inequality (6). Then by inequalities (4) and (6). for all x E [a, b] and
n, ma n
If,(x) - fm(X) I I MW - fm(x)) - (ff(x0) - fm(X0))I + I f.(xo) - fm(X0) I
Hence by Theorem 8.2.3, the sequence (f,) converges uniformly on [a. b). Let
f(x) = .limff(x)
It remains to be shown that f is differentiable, and that
f'(x) = hm f;,(x)
for all x E [a, b]. Fix p E [a, b], and fort # p, t E [a, b], define
f'(P) _r-P
lim g(t) =f.'(p) O
8.5.2 EXAMPLE To illustrate the previous theorem, consider the series
k-i .
Since 12-k sin kxl <_ 2-'for all x E R, by the Weierstrass M-test this series converges
uniformly to a function S on R. For n E N, let
sin kx
S, (x) =k,`
7 2
Then
S.'(X) k cos kx
k=1 2k
Since E k2 -k converges, by the Weierstrass M-test the sequence {S,,} converges uni-
formly on R. Thus by Theorem 8.5.1,
k cos kx
S'(x) = lim S; ,(x) _ 2k
cos ak9rx 1
2k 2k,
the series (7) converges uniformly on R, and hence f is continuous. The graphs of the
partial sums (with a = 13) Si(x) = cos(wx), S2(x) = Si(x) + 1 cos(13ax), and S3(x) _
S2(x) + 2, cos(1321rx) are illustrated in Figure 8.4.
2 2
S2
1
r
-1
Figure 8.4
We prove that f is nowhere differentiable by showing that for each x E R, there ex-
ists a sequence h -* 0 such that
f(x + hn) - f(x)
= 00.
1im I
11i00 hn
For n E N, set
Cos ak?rx
A-0 2'
cos aklrx
RR(x) = 0
k=n 2k
h" -
1-B"
am
Since -'-z s B" < '-z, we have 2 >- 1 - B" > '-i. Therefore,
Fork? n,
akir(X + h") = a-"a"ir(x + h")
ak_"ir(a"x
= + ak "Tr(k" + 1).
Since ak-" is odd and k" is an integer,
Since -Z s S" < Z, cos 8"1r 2t 0. Therefore, by inequality (9) and the above,
R"(x+h,,)-R"(x) " 1 3(21". (10)
Remark. The above proof is based on the proof of a more general result given in the
text by E. Hewitt and K. Stromberg. There it is proved (Theorem 17.7) that
has the desired property if a is an odd positive integer, and b is any real number with
b > I satisfying
6> 1+3ir.
The above function was carefully examined by G. H. Hardy [Trans. Amer. Math. Soc.,
17, 301-325 (1916)] who proved that the above f has the stated properties provided
1 :<.b s a.
These are by no means the only examples of such functions. A slightly easier con-
struction of a continuous function that is.nowhere differentiable is given in Exercise 7.
EXERCISES 8.5
1. For n E N, set f(x) = [0, 1 that the sequence converges on [0,
that f on that converge to
a on [a, b] for [a, b] n
converges on and for the of
that f on and that f'(x) = li m for all
xE
o a of real numbers satisfying I kjak I < oo. Show that the series Yk o akxk converges uni-
formly to a function f on jx] 5 1 and that f'(x) = joko-, kakxk- ' for all x, ] x 1 s 1.
4. *Let If.) be a sequence of differentiable real-valued functions on (a, b) that converges pointwise to a function f on
(a, b). Suppose the sequence {f;,} converges uniformly on every compact subset of (a, b). Prove that f is differen-
tiable on (a, b) and that f'(x) = lim f;,(x) for all x E (a, b).
S. State and prove an analogue of Theorem 8.5.1 for a series of functions 7, fk(x).
6. Show that each of the following series converge on the indicated interval and that the derivative of the sum can be
obtained by term-by-term differentiation of the series.
M 0o xk
C. 2xk, ]x]<l d. k, x E (-oo, oo)
k=0
7. This exercise provides another construction of a continuous function f on R which is nowhere differentiable. Set
g(x) = x {, = 15 x is 1, and extend g to R to be periodic of period 2 by setting g(x + 2) = g(x). Define f on
R by
f(x) _ k=o(31
4
g(4kx).
346 Chapter 8 Sequences and Series of Functions
Since 8," -+ 0 as m - oo, it now follows that f'(.t") does not exist.
Before we prove Theorem 8.6.1, we state and prove a more fundamental result that
will also have applications later. Prior to doing so, we need the following definitions.
The canonical examples of periodic functions are the functions sin x and cos x, both
of which are periodic of period 2ir. The graph of a periodic function of period p is il-
lustrated in Figure 8.5. The graphs of a periodic function of period p on any two suc-
cessive intervals of length p are identical. It is clear that if f is periodic of period p, then
f(x+kp) =f(x) forallkEZ.
Another useful property of periodic functions follows.
8.6.3 THEOREM 1f f is periodic of period p and Riemann integrable on [0, p), then f is Rie-
mann integrable on [a, a + p] for every a E R. and
+P
f° f(x) dx = J r f (x) dx.
c J0
Proof. Exercise 2.
8.6 The Weierstrass Approximation Theorem 347
Approximate Identities
8.6A DEFINITION A sequence {Qn} of nonnegative Riemann iruegrable functions on
[-a, a] satisfying
(b) lim
n-roo
J dt = 0 for every S > 0,
{dsiro
Remark. In (b), by the integral over the set {8 s I t 1} we mean the integral over the
two intervals [-a,-8] and [S, a].
n 1 1
n
Sxs n'
2'
QQ (t) =
0,
n
It is easily shown that the sequence {Q,,} is an approximate identity on [ -1, 1 ] (Exer-
cise 3). Other examples will be encountered in the proof of Theorem 8.6.1 and in the
exercises, and still others when we study Fourier series.
As a general rule, the Q. are usually taken to be even functions: i.e.,
QQ(-x) = Qn(x). T h e f a c t that the integrals over the set {t : f t I S} become small as
348 Chapter 8 - Sequences and Series of Functions
n --), oo seems to suggest that in some sense the functions themselves become small as
n becomes large. On the other hand, since the integrals over [-a, a] are always 1, by
property (b) of Definition 8.6.4
(8
lim J dt = 1
a
for every 8 > 0. This seems to indicate that the functions are concentrated near 0 and
must become very large near 0 (see Exercise 6). The graphs of the first few functions
Q1, Q2, and Q3 of a typical approximate identity are given in Figure 8.6.
-t -0.5 0 0.5
8.6.5 THEOREM- Let bean approximate identity on [-I, I], and let f be abounded
real-valued periodic function on R of period 2 with f E 9t[ -1, 1 ]. For n e N, x E R,
define
1f f is continuous at x E R, then
lima Sn(x) = f(x).
Proof. We first note that since f is periodic and integrable on [-1. 1). f is integrable
on every finite subinterval of R. Thus the integral in equation (11) is defined for all
x E R. Also, since f is bounded, there exists a constant M > 0 such that L f (x)I s M
for all x E R.
8.6 The Weierstrass Approximation Theorem 349
Therefore,
lf(X+t)-f(X)I < 2
for all t, I t I < S. Therefore,
5 J Q0(t) dt = 2. (13)
2
On the other hand,
Since {Qn) is an approximate identity , by property (b) there exists no E N such that
Qn(t) dt <
J{8s111} 4M
for all x E R and all t, I t l < 6. As a consequence, inequality (13) holds for all x E R.
Therefore, as above, there exists n E N such that
I f(x) I < E
for all x E R and all n ? no. This proves that the sequence converges uniformly to
f on R. Q
Proof of the Weierstrass Approximation Theorem. We now use Theorem
8.6.5 to prove the Weierstrass approximation theorem (Theorem 8.6.1). Suppose f is a
continuous real-valued function on [a, b]. By making a change of variable, i.e.,
g(x) = f((b - a)x + a), x E [0, 1 ],
we can assume that f is continuous on [0, 1 ]. Also, if we let
g(x)=f(x)-f(0)-x[f(l)-f(0)], xE[0,11,
then g(0) = g(l) = 0 and g(x) - f(x) is a polynomial. If we can approximate g by a
polynomial Q and set
P(x) = Q(x) + [f(x) - 8(x)]
then P is also a polynomial with I f(x) - P(x) I = Ig(x) - Q(x)I. Therefore, without
loss of generality, we can assume that f is defined on [0,1 ] satisfying
f(0) = f(1) = 0.
Extend f to [ -1, 1 ] by defining f (x) = 0 for all x E [ -1, 0). Then f is continuous
on [ -1, 1]. Finally, we extend f to all of R by defining
f (x) = f (x - 2k), k E Z,
where k E Z is chosen so that x - 2k E (-1, 1 ] (see Figure 8.7). Then f is continuous
and periodic of period 2 on all of R, and thus satisfies the hypothesis of Theorem 8.6.5.
f
-2 -1 o 2
Figure 8.7
8.6 The Weierstrass Approximation Theorem 351
Our next step is to find an approximate identity {Q"} on [ -1, 1 ] such that the cor-
responding functions S. of Theorem 8.6.5 defined by equation (11) are polynomials. To
accomplish this we let
Q,(t) = C"(1 - t2)",
Q"(t) tit = 1.
12,
Thus the sequence {Q"} satisfies property (a) of Definition 8.6.4. To show that it also
satisfies (b) we need an estimate on the magnitude of c". Since
I I
1=c"J(l -t2rdt=2c"J(l -t2rdt
i 0
?2c"J (1-t2)"dt
o
2c,, (1-nt2)dt=2c" (
1
- 1
o 37 /
4c"
3\fn
we obtain
In the above we have used the inequality (1 - t2)" z 1 - nt2 valid for all t E [0, 1]
(Example 1.3.3(b)). Finally, for any 5, 0 < S < 1,
This is the function S"(x) of Theorem 8.6.5, restricted to x E [0, 1]. Let x E [0, 1
Since f(t) = 0 for t E [ -1.0] U [ 1, 2],
f(x
Jt f(x + dr + t)QQ(t) dt,
c
Therefore, P"(x), for x E [0, 1 ], is a polynomial of degree less than or equal to 2n. As
a consequence of Theorem 8.6.5,
lim P"(x) = f(x)
n-M
uniformly on [0, 1).
EXERCISES 8.6
1. If f : lR - R is periodic of period 2 and continuous on [-1, 11, prove that f is uniformly continuous on R.
2. *Prove Theorem 8.6.3.
3. Forn E N, define Q. on [ -1, 1 ] as follows:
n -ln cx ln,
2
Q"(x) _
0, 1 <1xI sl.
Show that {Q"} is an approximate identity on [ -1, 1 ].
4. FornEN,setQ"(x)=c"(1 - IxI)",x E[-1,1].
a. Determine c" > 0 so that f', Q"(r) dr = 1.
b. Prove that with the above choice of c", the sequence {Q"} is an approximate identity on [ -1. 1 ].
c. Sketch the graph of Q"(x) for n = 2, 4, 8.
S. Forn E N, set Q"(x) = c,, x l e -', x E [ -1, 1 ].
a. Determine c" > 0 so that f', Q"(r) dr = 1.
b. Prove that with the above choice of c", the sequence {Q"} is an approximate identity on [ 1, 1 ].
6. *If {Q"} is an approximate identity on [ -1, 1 ], prove that Fim- x E [ -8. S)) = oo for every S > 0.
7. Let f be a continuous real-valued function on [0, 1). Prove that given e > 0, there exists a polynomial P with ratio-
nal coefficients such that 1f(x) - P(x) I < e for all x E [0, 1 ].
8.7 Power Series Expansions 353
8.7.1 DEFINITION Let {ak}k_o be a sequence of real numbers, and let c E R. A series of
the form
00
is called a power series in (x - c). When c = 0, the series is called a power series in
x. The numbers at are called the coefficients of the power series.
Even though the study of representation of functions by means of power series dates
back to the mid-seventeenth century, the rigorous study of convergence is much more re-
cent. Certainly Newton and his successors were concerned with questions involving the
convergence of a power series to its defining function. It was Cauchy, however, who,
with his formal development of series, brought mathematical rigor to the subject. As an
application of his root and ratio test, Cauchy was among the first to use these tests to de-
termine the interval of convergence of a power series. This is accomplished as follows:
Consider a power series 7, ak(x - c)k. Applying the root test to this series gives
km Iakl Ix - cl A = Ix - cla.
m
k
where a = lim
k-.oo
I ak [ . Thus by Theorem 7.3.4, the series converges absolutely if
alx - cl < 1, and divergesifalx - cl > 1.Ifa=0,then alx-cl < I for all
x E R. If 0 < a < oo, then
aix - cl < I if and only if Ix - cl <
8.7.2 DEFINITION Given a power series 7- ak(x - c)k, the radius of convergence R is
defined by
R = litn.
1
When R = 0, the power series E ak(x - c)k converges only for x = c. On the
other hand, if R = oo, then the power series converges for all x E R.
Remark If ak # 0 for all k and lim I ak+, I / I ak I exists, then by Theorem 7.1.10 the
radius of convergence of E ak(x - c)k is also given by
I lak+1I
= lim Iaj
R k-,oo
8.7.3 THEOREM Given a power series 5:k'. O ak(x - c)k with radius of convergence R,
0 < R 5 oo, then the series
(a) converges absolutely for all x with Ix - cI < R, and
(b) diverges for all x with I x - c I > R.
(c) Furthermore, if 0 < p < R, then the series converges uniformly for all x with
Ix - ci :S P.
Proof. Statements (a) and (b) were proved in the discussion preceding the statement
of the theorem. Suppose 0 < p < R. Choose /3 such that p < 6 < R. Since
k
lim
k-roo = RI < I,
But (p//3) < I and thus 7,(p/j3)k < oo. Therefore, by the Weierstrass M-test, the se-
ries converges uniformly on Ix - c) s p. 0
The previous theorem provides no suggestion as to what happens when }x - c l = R.
As the following examples (with c = 0) illustrate, the series may either converge or di-
verge when I x I = R.
8.7.4 EXAMPLES
(a) The series
00
xk
71
k-1
(b) The series J , xk/k also has radius of convergence R = 1. In this case, when
x = 1 the series diverges; whereas when x = -1, the series is an alternating series
which converges by Theorem 7.2.3.
(c) Consider the series 7,k'-, xk/k2. Again the radius of convergence is R = 1. In this
example the series converges at both x = I and - 1.
(d) Consider the series
where
__ 3k, if k is even,
ak
2' if k is odd.
Hence Tim V1 akl = 3, and therefore, R = 3. The series diverges at both x = 3 and
X = -3.
(e) Finally, consider the series 7, k!xk. Here ak = k!, and
a541
lim- = kim(k + 1) = oo.
k--.oo ak
Abel's Theorem
Suppose we are given a power series Y, ak(x - c)k with radius of convergence R > 0.
By setting
we obtain a function that is defined for all x, f x - cI < R. Functions that are defined
in terms of a power series (as in series (14)) are usually referred to as ral.aoalytic
functions. Fix p with 0 < p < R. Since the series converges uniformly to f on
Ix - cI p, by Corollary 8.3.2, f is continuous on Ix - cI <- p. Since this holds for
all p < R. the function f is continuous on Ix - c I < R. If the series (14) also con-
verges at an endpoint, say at x = c + R, then f is continuous not only in (c - R, c + R)
but also at x = c + R. This follows from the theorem of Abel, which is shown next. For
convenience, we take c = 0 and R = 1.
8.7.5 THEOREM (Abel's Theorem) Suppose f(x) = Y-k o akxk has radius of conver-
gence R = 1, and that 7_' o ak converges. Then
00
f(x) = Y, ak.
k=0
356 Chapter 8 Sequences and Series of Functions
Proof. Set s_I = 0, and f o r n = 0, 1, 2, ... let s = 7,1.0 ak. Then by the partial
summation formula (7.2.1),
akXk =
Sk(Xk - xk+I) + Sin
k-0 k=0 /
n-I
(1 - X) I Skxk +
k-0
Since the sequence {sn} converges, if we let n -a oo, then for all x. Ix I < 1,
x+
f (X) _ (1 - X) L, SkXk
k-0
Lets = lim s,,, and let e > 0 be given. Choose n,, E N such that I s - sn I < 1 e for
all n Z no. Since
00
(1 -x)7, xk= 1, IXI <
A-0
(I-X)Sk-SI
k=0
+E(1-x),xk
2 k=n,+I
s(1-x)M+2
where M = Yk,olsk - sI. If we now choose 8 > 0 such that I - S < x < I im-
plies that (1 - x) M < e, then if(x) - s l < e for all x, 1 - 8 < x < I. Thus
lim f (x) = s. Q i
8.7.6 EXAMPLE To illustrate Abel's theorem, consider the series Icko-o (-1)ktk. This series
has radius of convergence R = 1. Furthermore, the series converges to f (t) = 1/(1 + t)
for all t, I t I < 1. Since the convergence is uniform on I t I s I x I where I x I < 1. by
Corollary 8.4.2,
dt °°
ln(1 + x) =
+
7(-1)J
= k.0 tk d
0 0
00
(-1)kxk+I = (-1)t+l xk
0
k.ok + 1 k.I k
has radius of convergence R = 1, and also converges when x = 1. Thus by Abel's the-
orem,
001
n2=
k
1-2+3-4
1
The obvious question is what is the radius of convergence of the differentiated series
(15)? Furthermore, if f is defined by f(x) = I' oak(x - c)k, lx - cI < R, does the
series (15) converge to f'(x)? The answers to both of these questions are provided by
the following theorem.
8.7.7 THEOREM Suppose 7,- o ak(x - c)k has radius of convergence R > 0, and
Proof For convenience we take c = 0. Consider the differentiated series 7, kalxk -1.
By Theorem 2.2.6, lim \ = 1, and for x * 0,
lxI _
k
k
I k-1 = W. . By Exercise 10 of Section 2.5, for x * 0,
fi-m k IakIkIXIk-1
k
=lxliklakl.
Therefore, if R = oo, the differentiated series (15) converges for all x, and if 0 < R < oo,
the differentiated series converges for all x, IxI < R, and diverges for all x, Ixl > R.
Thus the radius of convergence of Joko- 1 kakxk-1 is also R.
358 Chapter 8 Sequences and Series of Functions
Furthermore, for any p, 0 < p < R, by Theorem 8.7.3 the series 7, kakxk-' con-
verges uniformly for all x, I x 1 s p. Thus by Theorem 8.5.1, the series (15), obtained
by term-by-term differentiation, converges to f'(x), i.e.,
OQ
8.7.8 COROLLARY Suppose Ioko- o ak(x - c)k has radius of convergence R > 0, and
x
f (x) = Y, ak(x - c)k, I x - c I < R.
k=o
Then f has derivatives of all orders in Ix - c I < R, and for each n E IN,
x
ft")(x) _ ',k(k - 1) (k - n + 1)ak(x - c)k-". (16)
k-n
In particular,
ft")(c) = n! a,,. (17)
(1 - x)2 = kYa
kz k-1 = 1( k + l)x k ,
00 00
2
(1 - x)3
= k k( k - 1)x k-2 = k ',( k + 2)( k + 1)xk ,
-2 -0
and for arbitrary n E N,
(n-1)!-
(1 -xY' J(k+n-1 )...(k+ 1)xk.
8.7 Power Series Expansions 359
8.7.11 COROLLARY Suppose ak(x - c)k and 7, bk(x - c)k are two power series which
converge for all x, Ix - c < R, for some R > 0. Then
00 00
If f (x) = g(x) for all x, x - c I < R, then r)(x) = gt") (x) for all n = 0, 1, 2.. . .
and all x, I x - c I < R. In particular, f")(c) = gO)(c) for all n = 0. 1, 2, .... Thus by
equation (17), a" = b" for all n.
0, x=0.
Since X
m
e" = liim a-`' = 0, f is continuous at 0. For x * 0,
x3
360 Chapter8 Sequences and Series of Functions
When x = 0, we have
f'(0) = lim
h-*0
f (h)
-h A o) = lim
h-0
e
h
= lim
' '
t.
e'
= 0.
.f'(x) _ X, e '/'', x 0 0,
0, x = 0.
8.7.13 DEFINITION Let f be a real-valued function defined on an open interval 1, and let
c E I and n E N. Suppose f W(x) exists for all x E I. The polynomial
(k) (c)
r.( C)(x) _
kmo
f kf (x - c)k
is called the Taylor polynomial of order n off at the point c. If f is infinitely differen-
tiable on 1, the series
= fk)(c)
(x - c)k
k.0 k!
For the special case c = 0, the Taylor series of a function f is often referred to as
the Maclaurin series. The first three Taylor polynomials, To, T1, T2, are given specifi-
cally by
T0(f, c)(x) = f(c),
T1(, c)(x) = f(c) + f'(c)(x - c),
f 2r) (x - c)2.
T2(f, c)(x) = f(c) + f'(c)(x - c) +
The Taylor polynomial T,(f, c) is the linear approximation to f at c; that is, the equa-
tion of the straight line passing through (c, f(c)) with slope f'(c).
In general, the Taylor polynomial T. of f is a polynomial of degree less than or
equal to n that satisfies
T (k)(f c)(c) = f tk)(c),
for all k = 0, 1, ... , n. Since ft">(c) might possibly be zero, T. (as the next example
shows) could very well be a polynomial of degree strictly less than n.
8.7.14 EXAMPLES In the following examples we compute the Taylor series of several func-
tions. At this stage nothing is implied about the convergence of the series to the function.
(a) Let f (x) = sin x and take c = Z. Then
f(2) = sin 2 = 1,
f'(2)=cos2=0,
f.("(2) = -sin 2 = -1,
J 3)(2) = -Cos Z = 0.
Thus
2
T3(f2)(x)=1-2t x-2/,
which is a polynomial of degree 2. In general, if n is odd, 0, and if n = 2k is
even, P 20(f) = (-1)k. Therefore, if n is even,
T ( f , 2)(x) = T. 1(J,
n12 (-I r
2)(x) = 0 (2k) \x - -2\zk
The Taylor expansion of f(x) = sin x about c = 2 is given by
00 (-1)k 2k
ir
()! IX
2)
(b) For the function f(x) = e-"X', by Example 8.7.12
0)(x) = 0 for all n E N.
362 Chapter 8 Sequences and Series of Functions
Thus the Taylor series of f at c = 0 converges for all x E l ; namely, to the zero func-
tion. It, however, does not converge to f.
(c) In many instances, the Taylor expansion of a given function can be computed from
a known series. As an example, we find the Taylor series expansion of f(x) = 1/x about
c = 2. This could be done by computing the derivatives off and evaluating them at
c = 2. However, it would still remain to be shown that the given series converges to
f (x). An easier method is as follows: We first write
1 1 _l 1
x 2-(2-x) (2 2
1-w k_o
for all x, I x - 21 < 2. By uniqueness, the given series must be the Taylor series of
f(x) = 1/x. In this instance, the power series also converges to the function f(x) for all
x satisfying Ix - 21 < 2. U
11
Remainder Estimates
To investigate when the Taylor series of a function f converges to f(x), we consider
R (x) = c)(x) = f (x) - c)(x). (18)
The function R. is called the remainder or error function between jr and c).
Clearly,
f (x) = Jim TT(f, c)(x) if and only if lim c)(x) = 0.
n
Since the Taylor polynomial T. is the nth partial sum of the Taylor series of f, the
Taylor series converges to f at a point x if and only if nliini c)(x) = 0. To emphasize
this fact, we state it as a theorem.
AX) = 00
fck)(c) (x - c) , k
k-O k.E
The formula
theorem once more to the function g(n) on the interval [c, xn], we obtain the existence of
a t E (c, xn) such that
0 = 8(n)(Xn) - g(n)(c) = Jn+ l)( )(X - C).
Thus 0; i.e., f(n+1)(C) - (n + 1)! M = 0, for some C between x and c. 0
In Example 8.7.20 we will give several examples to show how the remainder esti-
mates may be used to prove convergence of the Taylor series to its defining function. In
the following example we show how the previous theorem may be used to derive sim-
ple estimates and inequalities.
8.7.17 EXAMPLES
(a) In this example we use Theorem 8.7.16 with n = 2 to approximate f (x) = 1 + x,
x > -1. With c = 0 we find that
2
Therefore, T2(f, 0)(x) = 1 + 2 x - e x, and thus
1
1 +x = 1+2x-8x2+ RZV, 0x
1
) ( ).
By formula (19),
R2(f, 0)(x) = f (3)
L) X3
= 16
I (1 + )-SR x3
3! -
for some C between 0 and x. If x > 0, then > 0, and thus (1 + C)"s2 < 1. There-
fore, we have
for any x > 0. If we let x = 0.4, then TZ(f, 0)(.4) = 1.18, and by the above,
I 1.4 - 1.181 < 0.004, so that two-decimal-place accuracy is assured. In fact, to five
decimal places 1.4 = 1.18322.
(b) The error estimates can also be used to derive inequalities. As in the previous
example,
1
-1/1+1= 1 + 2 x - 8x2 + R2(f, 0) (x).
I+2x-8x2< VT +x<1+2x-gx2+ I x3
16
Is (x - t) f"(i) dt.
To complete the proof, we assume that the result holds for n = k, and prove that
this implies the result for n = k + 1. Thus assume Rk(x) is given by equation (20). Then
1 k (k+I) (k+')
366 Chapter a Sequences and Series of Functions
Rk+i(x) = 1 J s (x - t)k''f(k+'>(t)dt.
(k + 1)! J.
Proof Since f("+')(t)(x - t)" is continuous on the interval from c to x, by the mean
value theorem for integrals (Theorem 6.3.6), there exists a C between c and x such that
f fn+I)(:)(x - t/ dt = (x - c) f("+'kC)(x
l - C)".
C
8.7.20 EXAMPLES
(a) As our first example, we prove the binomial theorem (Theorem 2.2.5). For n E N
let f(x) = (1 + x)", x E R. Since f is a polynomial of degree n, if k > n then
f (k)(x) = 0 for all x E R. Therefore, by Theorem 8.7.16,
f(x) _
The series expansion of (1 + x)° for a E R with a < 0 is given in Theorem 8.8.4,
whereas the expansion for a > 0 is given in Exercise 7 of the next section. For ratio-
nal numbers a, the expansion of (1 + x)° was known to Newton as early as 1664.
(b) Let f (x) = sin x with c = 0. Then
The sine function, as well as the cosine function, can be defined strictly in terms of
power series. For further details, see Miscellaneous Exercise 3.
(c) As our third example, we derive the Taylor series for f (x) = ln(1 + x), where, as
in Example 6.3.5,
denotes the natural logarithm function on (0, oo). Then f(0) = In (1) = 0, and by the
fundamental theorem of calculus, f '(x) = 1/(1 + x). Thus for n = 1, 2, ... ,
f("i(x) _ (-1)n+i (n - 1)!
0+x)"
In particular, f W(0) _ (-1)"+ 1(n - 1)!, and the Taylor series off at 0 becomes
. (-1)n+I
7'
n., n
x
368 Chapter 8 Sequences and Series of Functions
Although we have already proved that this series converges to In (1 + x) for all
x, -1 < x 1 (Example 8.7.6), we will prove this again to illustrate the use of the
remainder formulas.
Suppose first that 0 < x <- 1. By Theorem 8.7.16,
(_ 1)"+zx"+l
R (x) = R If 0)(x1 =
(n + 1)(l + C)"''
for some C, 0 < < x. In this case, (1 + C) > 1, and thus
IR"(x) I n+ l x
"+l<
n+ 1
for all x, 0 < x s 1. Therefore,
1im R"(x) = 0 for all x e [0, 1 ].
n-Oc
We next consider the more difficult case -1 < x < 0. By the Cauchy form of the
remainder, if - I < x < 0, there exists ', x s s 0, such that
IR"(x)I
(II+I0 [(l + )]"'
Consider the function ap(t) defined on [x, 0] by
x
(P(t) = 1 + t
Then cp'(t) = (I + x)/(1 + t)2, which is positive on [x, 0] provided x > -1. There-
fore,
(P(t) q,(0) = -x = IxI
for all t, x s t <_ 0. Thus
I+I0lxln.
IR"(x)I `- (1
Since IxI < 1, lim R"(x) = 0. Therefore, the Taylor series converges to In (1 + x) for
all x,-1 <x -_ 1;i.e.,
00 (-I)n+l
ln(l+x)=
n=l n
x", -l<xsl.
8.7 Power Series Expansions 369
The Taylor expansion of In(l + x) was first obtained in 1668 by Nicolaus Mercator.
Newton shortly afterward obtained the same expansion by term-by-term integration of
the series expansion of 1/(1 + x) (see Example 8.7.6).
(d) As our final example, we consider the natural exponential function E(x) = exp x
that is defined as the inverse function of the natural logarithm function L(x) = In x. The
domain of L is (0, oo) with range (-oo, oo). Since L'(x) = 1/x is strictly positive
on (0, oo), L is a strictly increasing function on (0, oo). The inverse function E(x) is de-
fined by
y = E(x) if and only if x = In y.
By the inverse function theorem (Theorem 5.2.14), E is differentiable on R with
'X1
k-0 k
It is left as an exercise (Exercise 11) to show that this series converges to e' for all
x E R.
There is a more subtle question involving power series representation of functions
that we have not touched upon. The question concerns the following: How is the radius
of convergence R of the Taylor series off related to the function f? The full answer to
this question requires a knowledge of complex analysis and thus is beyond the scope of
this text. However, we will illustrate the question and provide a hint of the answer with
the following examples. If f(x) = 1/(1 + x), then the Taylor expansion off about
x = 0 is given by
1. (- I )kxk,
k=0
370 Chapter 8 Sequences and Series of Functions
which has radius of convergence R = 1. This is expected since the given function f is
not defined at x = -1, and thus the series could not have radius of convergence R > 1.
If it did, this would imply that f would then have a finite limit at x = -1.
On the other hand, the function g(x) = 1 /(1 + x2) is infinitely differentiable on all
of R. However, the Taylor series expansion of g about c = 0 is given by
00
,(-I)kX2k,
k=0
which again has only radius of convergence R = 1. The reason for this is that even
though g is well-behaved on R, if we extend g to the complex plane C by
1
g(z) = 1 + z2
then g is not defined when z2 = -1; i.e., z = ±i, where i is the complex number that
satisfies i2 = -1.
EXERCISES 8.7
1. Find the radius of convergence of each of the following power series.
00 3k
a. D
Xk *b. x+12*
k-1 k k=0 4
00
1 - 1 xk
*d.
1 71 k
1 when k is even, 00 / kt
C. I,, akxk where ak - 2" . f. j k1 2)
k'0 when k is odd k1
2k +2
2. For each of the following, determine all values of x for which the given series converges.
l+x k
a.
k-0 X
k, (X * 0) tb G
k- 1 2(1
3kxk
x)"k
(X # 1) C. I 1-x
(x #1)
3. Using the power series expansion of 1/(1 - x) and its derivatives, And
00 00
4. a. Use Theorem 8.7.16 to show that 1i - (I + 1 x - 9 x2) I < 81x3 for all x < 0.
b. Use the above inequality to approximate 'Mi, and provide an estimate of the error.
5. Determine how large n must be chosen so that I sin x - T (sin, 0)(x)l < .001 for all x, IXI s 1.
6. Use the Taylor series and remainder estimate of Example 8.7.20(c) to compute In 1.2 accurate to four decimal
places.
8.7 Power Series Expansions 371
7. Suppose f(x) = Ik o ak(x - c)t has radius of convergence R > 0. For Ix - cl < R, set F(x) = f' f(r) dr.
Prove that
00 a,
k+1(x-c)k+i, Ix - cl < R.
F(x) = k
Arctan x =
1 + x2
to obtain the Taylor series expansion of Arctan x about c - 0.
b Use pact (a) to obtain a series expansion for ir.
*c. How large must n be chosen so that the nth partial sum of the series in part (b) provides an approximation of
it correct to four decimal places?
9. Use Exercise 8 and Abel's theorem to prove that
(-1 a
k=o2k+1 4'
10. Find constants ao, a,, a2, a3, a4 such that
x4+3x2-2x+5=a4(x- 1)4+a3(x- 1)3+a2(x- 1)2+a,(x- 1)+ao.
11. Prove that the Taylor series of e` (with c = 0) converges toe` for all x E R.
12. Using any applicable method, find the Taylor series of each of the following functions at the indicated point, and
specify the interval on which the series converges to the function.
a. &) = cos x, c = 0 *b. f (x) = In x, c = 1
± \),
c. f(x) = In1 l x c=0 *d. f(x) = (I - x)-". c = 0
x
*e. f(x) = Aresinx, c = 0 L f(x) = 1 x2, c=0
P()e-V' x * 0,
Mx) 0, x = 0,
where P is a polynomial of degree 3n.
16. Suppose b > 1. For x E R define b(x) = E(x In b), where E is the natural exponential function.
a. Prove that b(r) _- b' for all r E Q.
b. For x E R, prove that b(x) = sup(b' : r E 0, r < x}.
372 Chapter 8 Sequences and Series of Functions
When 0 < x < 1, the integral in equation (22) is an improper integral not only at
oo, but also at 0. The convergence of the improper integral defining f(x), x > 0, was
given as an exercise (Exercise 9) in Section 6.4. The graph of r(x) for 0 < x < 5 is
given in Figure 8.8. The following properties of the Gamma function show that it is
closely related to factorials.
6 I 2 3
8.8.2 THEOREM
(a) For each x, 0 < x < oo, r(x + 1) = x r(x).
(b) For n E ICU, r(n + I) = n!.
Proof. Let 0 < c < R < oo. We apply integration by parts to
1 R t'e-' dt.
8.8 The Gamma Function 373
R
Rx
e
Since im; cxe-` = 0 and m R*e-R = 0, taking the appropriate limits in the above
R
yields
This proves (a). For the proof of (b) we first note that
I'(1) = e-' dt = 1.
J000
8.8.3 EXAMPLE Since the value of I'(!) occurs frequently, we now show that I'('-2) = .
By definition,
To complete the result, we need to evaluate the so-called probability integral f0 e-" ds.
This can be accomplished by the following trick using the change of variables theorem
from multivariable calculus. Consider the double integral
e-xl-''dxdy.
J= Joo f
0 0
WRe `rdrd9
J= 100
0 10
a (O0
= ?J e_,: rdr4. IT
0
374 Chapter 8 Sequences and Series of Functions
Therefore,
f0c
e-'2 dx = 2
0
(1-x)-n=
(n - 1)! k.0 k!
1 _ 1 r(k+n) k
x
(1 - xr 1'(n) k=o k!
We will now prove that this formula is still valid for all a E R with a > 0.
Proof. We first show that the radius of convergence of the series is R = 1. Set
an = 17(n + a)/n!. Then
an+,r(n+1+a) n!
a (n + 1)! T(n + a)*
But by Theorem 8.8.2, I'(n + 1 + a) = (n + a)I'(n + a). Therefore,
a^+, =limn + a = 1.
lim
n-"o a -oo n + 1
and as a consequence of Theorem 7.1.10, we have R = 1.
8.8 The Gamma Function 375
fv(x) = r(a)
1
I- nr(n + a) n-1
n!
x X.
Multiplying by (I - x) gives
' nr(n + a)
(1 - x)x rt-'
1
(l - x)fa(x) = r( a) Y' n!
r(x)F(y)
J' t'`-'(i - ty-' dt =
o F(x + y).
376 Chapter 8 Sequences and Series of Functions
The function
r(x)r(y)
B(x' A T(x + y)' x, y > 0'
is called the Beta function.
EXERCISES 8.8
1. 'a. Compute r(z), r('2222).
b. Prove that for n E N,
CIn+2)= (2n)! n .
2. By making a change of variable, prove that
*a I edt
0 t
b. f
dt, n E N.
0(
4. By making the change of variable t = sin 2 u in Theorem 8.8.5, prove that
(Ar 1 r(n)r(m)
Jo (sin u)2"`'(cos u)z `' du = n, m > 0.
2 r(n + m)'
5. Evaluate each of the following integrals.
a f+.R
J0
(sin x)2" dx, nEN b. f
0
n!1
(sin x)'' *' dx, n E N
6. Use the binomial series and term-by-term integration to find the power series expansion of
(a a(a-1)(a-2) (a-k+l)
\k/ k!
Note, if m e N,
i
k k! (m - k)!'
k 5 m, and (k) = 0 fork > m.
a. Prove that the series I:-k-0(k) x' converges uniformly and absolutely for x E H, 11.
b. Prove that xt = (1 + x)°, x E [-1, 1 ].
Miscellaneous Exercises 377
NOTES
Without question the most important concept of this chap- entiable. Although this construction is much easier, the
ter is that of uniform convergence of a sequence or series partial sums of the series defining the function f are them-
of functions. It is the additional hypothesis required in selves not differentiable everywhere. Thus it is not so sur-
proving that the limit function of a sequence of continu- prising that f itself is not differentiable anywhere on It
ous or integrable functions is again continuous or inte- The proof of the Weierstrass approximation theorem
grable. As was shown by numerous examples, pointwise presented in the text is only one of the many proofs avail-
convergence is not sufficient. For differentiation, uniform able. A constructive proof by S. N. Bernstein using the so-
convergence of {f"} is not sufficient; uniform conver- called Bernstein polynomials can be found on page 107 of
gence of the sequence of derivatives (f' ) is also required. the text by Natanson listed in the Bibliography. The proof
The example of Weierstrass (Example 8.5.3) is inter- in the text, using approximate identities, was chosen be-
esting for several reasons. First, it provides an example of cause the technique involved is very important in analysis
a continuous function which is nowhere differentiable on and will be encountered later in the text. In Theorem
R. Furthermore, it provides an example of a sequence of 9.4.5 we will prove a variation of the Weierstrass ap-
infinitely differentiable functions that converges uniformly proximation theorem. At that point we will show that
on R, but for which the limit function is nowhere differ- every continuous real-valued function on [-vr, w] with
entiable. Exercise 7 of Section 8.5 provides another con- f(-ir) = f(r) can be uniformly approximated to within
struction of a continuous function f that is nowhere differ- a given e > 0 by a finite sum of a trigonometric series.
MISCELLANEOUS EXERCISES
1. Using Miscellaneous Exercise I of Chapter 6 and the Weierstrass approximation theorem, prove the following: If
f E 9t.[a, b] and e > 0 is given, then there exists a polynomial P such that
tk PI < e.
£1_
2. DefinefonRby
d. Show that if f : R --)- R satisfies f "(x) = -f(x) with f(O) = 0, f'(0) = 1, then f (x) = S(x) for all x E R.
e. If f: R - R satisfies f"(x) = 1(x), prove that there exist constants c,, c2, such that f(x) = c,S(x) + c,C(x).
L Show that (S(x))2 + (C(x))2 = 1. (Hint: Consider the function f(x) _ (S(x))2 + (C(x))2.)
g. Show that C(x + y) = C(x) C(y) - S(x) S(y) and S(x + y) = S(x) C(y) + C(x) S(y) for all x, y E R.
SUPPLEMENTAL READING
Andrushkiw, J. W., "A note on multiple series of positive Mathf, P., "Approximation of Holder continuous func-
terms," Amer. Math. Monthly 68 (1961), 253-258. tions by Bernstein polynomials;' Amer. Math.
Billingsley, P., "Van der Waerden's continuous nowhere Monthly 106 (1999), 568-725.
differentiable function," Amer. Math. Monthly 89 Miller, K. S., "Derivatives of non-integer order," Math.
(1982), 691. Mag. 68 (1995), 183-192.
Blank, A. A., "A simple example of a Weierstrass func- Minassian, D. P. and Gaisser, J. W.. A simple Weier-
tion;' Amer. Math. Monthly 73 (1966), 515-519. strass function," Amer. Math. Monthly 91 (1984).
Boas, Jr., R. P., "Partial sums of infinite series and how 254-256.
they grow," Amer. Math. Monthly 84 (1977), Patin, J. M., "A very short proof of Stirling's formula,"
237-258. Amer. Math. Monthly % (1989),41-42.
Boas, Jr., R. P. and Pollard, H., "Continuous analogues Roy, Ranjan, "The discovery of the series formula for rr
of series;' Amer. Mark Monthly 80 (1973),18-25. by Leibniz, Gregory and Nilakantha," Math. Mag.
Cunningham, Jr., F., "faking limits under the integral 63 (1990),291-306.
sign;' Math. Mag. 40 (1967), 179-186. Sagan, H., "An elementary proof that Schoenberg's
Davis, P. J., "Leonhard Euler's integral: A historical pro- space filling curve is nowhere differentiable" Math.
file of the Gamma function;' Amer. Math. Monthly Mag. 65 (1992), 125-128.
66 (1959), 849-869. Schenkman, Eugene, "The Weierstrass approximation
French, A. P., " he integral definition of the logarithm theorem;' Amer. Math. Monthly 79 (1972),
and the logarithmic series;' Amer. Math. Monthly 65-66.
85 (1978), 580-582. Weinstock, Robert, "Elementary evaluations of
Kestleman, H., "Riemann integration of limit functions;' Ja a dr. Jo eos x2 dx, and Jo sin x2 dx; 'Amer.
Amer. Math. Monthly 77 (1970),182-187. Math. Monthly 97 (1990), 39-42.
Lewin, J. W., "Some applications of the bounded conver-
gence theorem for an introductory course in analy-
sis," Amer. Math. Monthly 94 (1987), 988-993.
Orthogonal Functions
y J and Fourier Series
9.1 Orthogonal Functions
9.2 Completeness and Parseval's Equality
9.3 Trigonometric and Fourier Series
9.4 Convergence in the Mean of Fourier Series
9.5 Pointwise Convergence of Fourier Series
379
380 Chapter 9 Orthogonal Functions and Fourier Series
in view of the eighteenth-century concept of a function this was not an unrealistic ex-
pectation. Fourier's claim had an immediate impact on nineteenth-century mathemat-
ics. It caused mathematicians to reconsider the definition of "function." The question of
what type of function has a Fourier series expansion also led Riemann to the develop-
ment of the theory of the integral and the notion of an integrable function. The first sub-
stantial progress on the convergence of a Fourier series to its defining function is due to
Dirichlet in 1829. Instead of trying to prove, like Fourier, that the Fourier series always
converges to its defining function, Dirichlet considered the more restrictive problem of
finding sufficient conditions on the function f for which the Fourier series converges
pointwise to the function.
In the first section, we provide a brief introduction to the theory of orthogonal func-
tions and to the concept of approximation in the mean. In Section 9.2 we also introduce
the notion of a complete sequence of orthogonal functions and show that this is equiv-
alent to convergence in the mean of the sequence of partial sums of the Fourier series
to its defining function. The proof of the completeness of the trigonometric system
{ 1, sin nx, cos nx}' , will be presented in Section 9.4. In this section we also prove
Fejdr's theorem on the uniform approximation of a continuous function by the nth par-
tial sum of a trigonometric series. In the final section, we present Dirichlet's contribu-
tions to the pointwise convergence problem.
(a, b) = I, aab1
i=1
f o r a = (a,, ... , a") and b = (b,, ... , b") in R". If () is an inner product on X, then
two nonzero vectors x, y E X are orthogonal if (x, y) = 0. The term "orthogonal" is syn-
1. For a detailed treatment of this subject see the texts by Berg and McGregor or Weinberger listed in the
Bibliography.
9.1 Orthogonal Functions 381
onymous with "perpendicular;" and comes from geometric considerations in R". Two
nonzero vectors a and b in IB" are orthogonal if and only if they are mutually perpendic-
ular; that is, the angle 9 between the two vectors a and b is z or 90° (see Exercise 10, Sec-
tion 7.4).
In the study of analysis we typically encounter vector spaces whose elements are
functions. For example, in previous sections we have shown that the space 12 of square
summable sequences and the space 11[a, b] of continuous real-valued functions on
[a, b] are vector spaces over R. With the usual rules of addition and scalar multiplica-
tion, 9t[a, b], the set of Riemann integrable functions on (a, b), is also a vector space
over R. If for f, g E 9t[a, b] we define
U, g) = f f(x)g(x)
b dx,
I
then it is easily shown that (,) satisfies (a), (c), and (d) of the definition of an inner prod-
uct. It does not, however, satisfy (b). If a < b and ci, . . . , c" are a finite number of
points in [a, b], then the function
0, x # c,,
f(x) = 11 , x=ci,
is in 9t[a, b] satisfying (f, f) = 0, but f is not the zero function. Thus technically () is
not an inner product on 9t[a, b]; a minor difficulty which can easily be overcome by
defining two Riemann integrable functions f and g to be equal if f(x) = g(x) for all
x E [a, b] except on a set of measure zero. This will be explored in greater detail in
Chapter 10. Alternatively, if we restrict ourselves to the subset b] of 9t[a, b], then
(f, g) as defined above is an inner product on b] (Exercise 11).
Orthogonal Functions
We now define orthogonality with respect to the above inner product on 9t[a, b].
9.1.2 EXAMPLES
(a) For our first. example, we consider the two, functions ¢(x) = I and 4i(x) = x,
x E [ -1, 1]. Since
f f O(x)*(x) dx = f t X dx = 0,
(b) In this example we show that the sequence of functions (sin nx}. , is orthogonal
on [-ir, ar]. By the trigonometric identity
for n # m,
sin nXlI'
X- = Jr.
-p
2\
(c) As our final example, we consider the collection 11, sin ¶, cos T )"., on the in-
terval [ -L, L] where L > 0. As in (b), if n * m, then
1L
sinnlrx
L sin m7rx
L dx = 0.
L
Thus the collection (sin ¶ ) is orthogonal on [ -L, U. Also, by the trigonometric iden-
tities
we have for n # m,
(L narx mirx (L nirx m7rx dx
cos L dx = 1 cos L
f
1 L cos
L
L sin
L
= 0.
Thus the functions in the collection {cos z} are all orthogonal on [-L, L] as are the
functions sin T and cos L" for all n, m E N with n * m. For m = n,
This last identity shows that the functions sin Of and cos "LF are also orthogonal on
[-L. L] for all n E N. Finally, since
Jsindxr fcos
for all n E N, the constant function 1 is orthogonal to sin "lr` and cos " for all n E N.
In this example we also have
L
2 nWX L 2 ri?lX
dx = Co. .L dx = L.
f Lsin
L J
If in Example 9.1.2(b) we define ¢"(x) _ j sin nx, then the sequence {4"(x)}R ,
satisfies
{0, when n # in.
4)"(x)¢m(x) dx =
fA 1, when n = m.
Given a collection {¢"} of orthogonal functions on (a, b], we can always construct
a family {fir"} of orthonormal functions on [a, b] by setting
c. = J ¢^(x) dx.
a
A natural question is, given a real-valued function f on [a, b], how must the coefficients
c" be chosen so that SN gives the best approximation to f on [a, b]? In the Weierstrass
384 Chapter 9 Orthogonal Functions and Fourier Series
IIxII= ,
left to the exercises (Exercise 12). The crucial step in proving the triangle inequality for
II II is the following version of the Cauchy-Schwarz inequality: For all x, y E X.
The proof of this inequality follows verbatim the proof of Theorem 7.4.3. For the vec-
tor space R[a. b] with inner product (f, g) = fa f(x)g(x) dx, the norm of a function f,
denoted I I f 112, is given by
111112 = If
f E &[a, b], the problem to consider is, how must the constants c be chosen
in order to minimize the quantity
9.1A THEOREM Let f E gt[a, b] and let be a finite or countable collection of or-
thogonal functions on [a, b]. For N E N, let SN be defined by equation (1). Then the
quantity
b
[f(x) - SN(x)]2 dx
Ja
fa f(x)-OR(x) dx
cn n = I , 2, ... N. (2)
fa 02(x) dx
[.f(x) - SN(x)]2 dx =
jf2(x) dx - c2 I
b
Prior to proving the result, we give the following alternative statement of the pre-
vious theorem.
9.1.5 COROLLARY Let f E Ii(a, b] and let SN(x) = I:- I caaya(x) where the ca are de-
fined by equation (2). If TN(x) = 7.N_ 1 an E R, then
b
0 5 J [f(x) - S,,(x)]2 dx
a
Also,
But
b N
which by orthogonality,
b
Cn 0(x) dX.
I
Therefore,
b
N= c .
jS(x)dx
N
n=1
2 +(
l Wn(X) dx.
JJJ
0 G i [f(x) - SN(X)]2 dx
b
N C.
bf2(X) dX - 2 Cn bf(X)fn(X) dx + j- 6 'On(x) dx.
Ja nil 1a nil 1a
386 Chapter 9 Orthogonal Functions and Fourier Series
N
f 2(X) dx + b n(x) Ic b 2J L./ f bo-12
6 2
Ia -I aJ L
f. b n=1 fa n
The coefficients cn occur only in the middle term. Since this term is nonnegative, the
right side is a minimum if and only if
cn =
a n
With this choice of cn, we also obtain formula (3) upon substitution. U
9.1.6 EXAMPLE As was previously shown, the functions 01(x) = 1 and ¢2(x) = x are
orthogonal on [ -1, 11. Let f (x) = x3 + 1. Then
c f',f(x)01(x)
= l J (x3 + 1) dx = 1.
' f!14i(x)dx 2 I
and
f_',f(x)d2(x)dx (' 3
C2 = =- I3
(x,
+ x)dx = .
0z(x) 2 -, 5
-1 -0.5 0 0.5 1
is called the Fourier coefficient off with respect to the system The series
I
00
=1
is called the Fourier series off. This is denoted by
00
Remark. The notation "--" in formula (6) means only that the coefficients in the
series are given by formula (5). Nothing is implied about convergence of the series'.
9.1.8 EXAMPLE In Example 9.1.2(b) it was shown that the sequence of functions
{sin nx},1 is orthogonal on [ -Tr, ir]. Since
n = 1, 2, .. ,
if f E Jt[ -ar, Tr], the Fourier coefficients c, n = 1. 2_ ... , off with respect to the or-
thogonal system {sin nx} are given by
In this case, the series converges for all x. but clearly not to f (x) = 1.
Bessel's Inequality
For each N E N, let SN(x) denote the Nth partial sum of the Fourier series off, i.e..
N
SN(x) = 71 CIAn(x),
n=1
388 Chapter 9 Orthogonal Functions and Fourier Series
where the cn are the Fourier coefficients off with respect to the sequence of or-
thogonal functions on [a, b]. Then by identity (3) of Theorem 9.1.4.
b
f
N r0.
f2(x)dx- cnJ
Therefore,
fb
cn 0.2(x)dx < bf 2(X)dx.
J
a a
Since this holds for every N E J. by letting N -+ oo we obtain the following inequal-
ity.
9.1.9 THEOREM (Bessel's Inequality) If f E 9t[a, b] and {cn}R , are the Fourier coeffi-
cients off with respect to the sequence of orthogonal functions {tbn}". ,, then
ao ,!, rb
cn fa V'n(x) dx :5 f 2(x) dx.
n=I J
a
n-oo
lim f f (x)fn(x) dx = 0.
a
Ia (6.2(x) dx
Proof. Since f E 9t[a, b]. fa f2(x) dx is finite. Thus by Bessel's inequality, the series
c. f° d, converges. As a consequence,
Jim b
cn 46.(x) dx = 0,
n-+oo
a
and thus,
fboa f(x)On(x) dx = 0.
(x) -n faV
b 2(
n(x) dx
a
EXERCISES 9.1
1. *Let f(x) = sin rrx, ¢,(x) = 1, and 42(x) = x. Find c, and c2 so that S2(x) = c,4,(x) + c2O2(x) gives the best ap-
proximation in the mean to f on [ -1, 1 ].
9.1 Orthogonal Functions 389
2. a. Show that the polynomials Po(x) = 1, PI(x) = x. and P2(x) = 12x2 - 1 are orthogonal on [ -1, l ].
b. Let
(0, -1 <_ x < 0,
AX) 1, 0<-xs 1.
Find the constants co, c,, c2, such that S2(x) = cnPo(x) + c^(x) + c2P2(x) gives the best approximation in the
mean to f on [-1, 1].
3. +a. Let 00(x) = 1, ¢t,(x) = x - a,, 42(x) = x2 - a2x - a3. Determine the constants a,, a2, and a3, so that
{40, m,, 42} are orthogonal on [0, I ].
b. Find the polynomial of degree less than or equal to 2 that best approximates f(x) = sin rrx in the mean on
[0,1].
4. Let {4nr.- t be a sequence of orthogonal functions on [a, b]. For f, g E J1[a, b] with f - Man4n and g Ybn4n.
show that for a, jS E R,
9. Let {an} be a sequence in (0, 1) satisfying 0 < an+, < an < I for all n E N. Define 4" on [0, 1 ] by
0. 05x <an+i,
an+I x an+, an),
4,(x) =
-2(x - an)
, 1(an + an)
an - an+,
0, an<x:5 1.
Show that {4n} is orthogonal in 9t[0, l] and compute II4nII2 for each n E N.
.
14. Let X be a vector space over IIi with inner product Q. If {y,, ... , yn} are nonzero orthogonal vectors in X and
x E X. prove that the quantity II x - (c,y, + + cyn) II is a minimum if and only if
_ (x. Y;)
`' IIY;Ii-
for all i = 1.... , n. (Hint: Imitate the proof of Theorem 9.1.4.)
f rh
111112 = f(x)dx
LJ
then convergence in the mean is nothing but convergence in norm as defined in Defi-
nition 8.3.9. Thus a sequence in 91t[a, b] converges to f E &[a, b] in the mean if
and only if lim 11f - f 112 = 0. Convergence in the mean is sometimes also referred to
as mean-square convergence.
It is natural to ask how convergence in the mean is related to pointwise or uniform
convergence. Our first theorem proves that uniform convergence implies convergence
in the mean. As should be expected, pointwise convergence is not sufficient (Exer-
9.2 Completeness and Parseval's Equality 391
cise 2). In the other direction, we will show in Example 9.2.3 that convergence in
the mean does not imply pointwise convergence, and thus certainly not uniform con-
vergence. There we construct a sequence (f.1 of Riemann integrable functions on
[0, 1 ] such that II f 11 2 -+ 0, but for which { f (x)} fails to converge for any
xE[0,1).
9.2.2 THEOREM I f f , f , , , n = 1, 2, . .. , are Riemann integrable on [a, b], and (f.) con-
verges uniformly to f on [a, b], then converges in the mean to f on [a, b].
Woof. Since the proof of this result is similar to the proof of Theorem 8.4.1, we leave
it as an exercise (Exercise 1).
9.2.3 EXAMPLE In this example we construct a sequence I f.1 on [0, 1] that converges to
zero in the mean, but for which { f (x)} does not converge for any x E [0, 1 ]. This se-
quence is constructed as follows: For each n E N, write n = 2k + j where k =
0, 1, 2, ... , and 0 s j < 2k For example, 1 = 21+0,2 = 2' +0,3 = 2' + 1, etc.
Define ff on [0, 1 ] by
1,
jksxsj+ k
1
2 2
0, otherwise.
of
The first four functions ff, f2, f3, and f4 are given as follows: fi(x) = 1, and
f 1, 0:5 z:S 2,
0<x<2,
fi(x)- {0,I, isx<
.fa(x) _ 1, 0 :5 :5 41
0, 1<xsl.
For each n e K f. E 3R[0, 1 ] with
J fn(X) dx = 1 dr = 2k
0 1
J/2'
Thus lim Jt; f n(x) dx = 0. On the other hand, if x E [0, 1 ], then the sequence { f (x)}
contains
n +00an infinite number of 0's and l's, and thus does not converge.
In the following theorem we prove that convergence in the mean of the partial sums
of the Fourier series is equivalent to equality in Bessel's inequality.
392 Chapter 9 Orthogonal Functions and Fourier Series
9.2.4 THEOREM Let {Wn}: I be a sequence of orthogonal functions on [a, b',. Then the
following are equivalent:
(a) For every f E 9t[a, b],
jb
- SN(X)]2 dx = 0,
limo
x b b
From this it follows immediately that {SN} converges in the mean to f if and only if Par-
seval's equality holds. 0
9.2.6 THEOREM If the sequence {¢n}f j of orthogonal functions on (a, b] is complete, and
if f is a continuous real-valued function on [a, b] satisfying
b
J f(x)4n(x) dx = 0 for all n = 1, 2, .. ,
a
Proof. The hypothesis implies that the Fourier coefficients c off are zero for all
n E N. Thus by Parseval's equality,
fb
f2(x) dx = 0.
Ja
Since f2 is continuous and nonnegative, by Exercise 7, Section 6.1, this holds if and
only if f 2(x) = 0 for all x E [a, b]. Thus f (x) = 0 for all x E [a, b]. E)
There is a converse to Theorem 9.2.6. Since the proof of the converse requires a
knowledge of the Lebesgue integral, we only state the result. A sketch of the proof is
provided in the miscellaneous exercises (Exercise 3) of Chapter 10.
9.2.7 THEOREM If {4n}' , is a sequence of orthogonal functions on [a, b] for which the
only real-valued continuous function f on [a, b] satisfying
b
J f (x)4n(x) dx = 0 for all n = 1, 2, .
a
Jbf(x)On(x) dx = Jg(x)4a(x) dx
a a
for all n = 1, 2, ... , then f(x) = g(x) for all x E [a, b]. The above assumption sim-
ply means that f and g have the same Fourier coefficients. To prove the result, apply
Theorem 9.2.6 to h(x) = f(x) - g(x).
Proof. Exercise 5. 0
394 Chapter 9 Orthogonal Functions and Fourier Series
El EXERCISES 9.2
1. Prove Theorem 9.2.2.
2. For n E f0i, define the function f on [0, I ] by
f() =
v 0<x< !, n
0, elsewhere.
f(x)
I
2 ao + a cosnrx
L+b nirx
*b. Use Parseval's equality and the indicated function to find the sum of the given series.
I
(1) 7, (2k 1)2'
f(x) = I (ii) kI ji, f(x) = x
1, cos
nlrx,
L
--
sin
nTT x
L _,'
9.3 Trigonometric and Fourier Series 395
which by Example 9.1.2(c) is orthogonal on [-L, L]. For convenience we will take
L = Tr.
Any series of the form
where the A. and B,, are real numbers, is called a trigonometric series. For example,
the series
00 sin nx
..2 In n
and I
MMI
cos nx
and
are nonnegative and decrease to zero, by Theorem 7.2.6 the first series converges for all
x E R, whereas the second converges for all x E R except x = 2pir, p E Z.
Fourier Series
For the orthogonal system {1, cos nx, sin nx} si on [-1r, 7r] we have
ff 4 n
-v 1 tr
Thus by Definition 9.1.7, the Fourier coefficients of a function f E 9t[ -a, ir] with re-
spect to the orthogonal system are defined as follows.
9.3.1 DEFINITION Let f E R[-7r, 1r]. The Fourier coefficients off with respect to the
orthogonal system (1, cos nx, sin nx} are defined by
ao = Jf(x)
-ir
dr,
1 "
an ,f (x) cos nx dx, n = 1, 2- ,
1 "
b.
b f (x) sin nx dx, n = 1, 2, ... .
J n
1 °O
AX) - 2ao + (a cos nx + b sin nx).
3% Chapter 9 Orthogonal Functions and Fourier Series
Remark. For the constant function 0o = 1, since f T, 020 = 21r, the term ao should be
defined as za f', f (x) dx, according to Definition 9.1.7. However, for notational conve-
nience it is easier to define ao as J°a f(x) dx and to include the constant 1 in the defi-
nition of the Fourier series.
For the orthogonal system ( 1, cos nx, sin nx), by Exercise 3(b) of the previous sec-
tion, Bessel's inequality 9.1.9 becomes
a
I ao +
00
(a + b2) <-L - -a
V(x))2 dx (Bessel's Inequality)
n-1
Thus for f E &[a, b] the sequences {a"} and {b,,} of Fourier coefficients off are square
summable sequences. In Theorem 10.8.7 we will prove that if {an} and {b"} are square
summable sequences, then there exists a Lebesgue integrable function f such that {a"}
and {b"} are the Fourier coefficients off.
9.3.2 DEFINITION For a real-valued function f defined on( - ir, vr), the periodic exten-
sion (of period 21r) off to R is obtained by defining f (x) = f(x - 2kir). where k E 1
is such that x - 2kvr E [-vr, vr).
9.3.3 EXAMPLES
ao=-faf(x)dx=1aldx= 1,
11
a 0
and for n = 1, 2, .. ,
fir
a" _
1
I
0
cos nx dx = n1
sin nx
Iff
0
= 0,
I
-I
IT
In the above we have used the fact that cos nvr = (-1)". Thus the Fourier series off is
given by
00
1
f(x) ^- + na[1 - (-1)"]sinnx = 2 + yr T 2k + 1 sin(2k - I)x.
2 n -l k=0
9.3 Trigonometric and Fourier Series 397
If SN(x) denotes the Nth partial sum of the Fourier series, then S1 and S., are given by
2
S1(x) = + 2 sin x and SAX) = 2 + 2 [sin x + 3 sin 3x
7r ir
The graphs of f, S1. S3, S5, and S15 are given in Figure 9.2.
S1 S3 S5 S15
(b) Let f (x) = x. Before we compute the Fourier coefficients, we will make several
observations that simplify this task. Recall that a function g(x) is even on [-a, a] if
g(-x) = g(x) for all x, and g(x) is odd if g(-x) = -g(x) for all x. By Exercise 4 of
Section 6.2, if g(x) is even on (-a, a], then
fa
g(x)dx = 21 ag(x)dx,
a 0
a g(x) dx = 0.
! -a
The functions sin nx are all odd, whereas cos nx are even for all n. Therefore, since
f (x) = x is odd, x cos nx is odd and x sin nx is even. Thus. a" = 0 for all n = 0,
1,2,...,and
2
f
b= x sin nx dx,
a
=-ncosnar=n(-1)"+i
398 Chapter 9 Orthogonal Functions and Fourier Series
Therefore,
x, (-1)n+1
x -- 2 sin nx.
n=1 It
Riemann-Lebesgue Lemma
There is one additional result from the general theory that will be needed later. For the
orthogonal system { 1, cos nx, sin nx}, Corollary 9.1.10 is as follows.
lim
n- 00
jf(x) cos nxdx = n-x
lim Jf(x) sin nxdx = 0.
Hence,
0 0
2. The value of a/2 for the improper integral is most easily obtained by contour integration and the theory
of residues of complex analysis. A real variables approach that computes the value of this integral is given in
the article by K. S. Williams listed in the supplemental readings.
9.3 Trigonometric and Fourier Series 399
with and converging to zero, does there exist a function f on [ -nr, a; such
that the coefficients A,, and B,, are given by Definition 9.3.1? As we will see, the answer
is no! First, however, in the positive direction, we prove the following.
9.3.6 THEOREM If the trigonometric series ;Ao + 7-(A cos nx + B. sin nx) converges
uniformly on [ -ir, a], then it is the Fourier series of a continuous real-valued function
on [ -?r, Tr].
Proof. For n E Nl, let
S (x) = 1 Ao + (Ak cos kx + Bk sin kx).
2 k-l
Since the series converges uniformly on [ -ir,1r], and S. is continuous for each n,
f (X) = urn S (x)
is a continuous function on [-ir, vr]. Form E N, consider
Since for each m, the sequence {S (x) cos mx} converges uniformly to f(x) cos mx on
[-ir, 1r], the above interchange of limits and integration is valid by Theorem 8.4.1. If
n > m, then
Thus by orthogonality,
A. = r
f (X) cos mx dx.
The analogous formula also holds for Bm, and thus the given series is the Fourier series
of f.
Remarks
(a) If 2 I Ak I and E I Bk I both converge, then by the Weierstrass M-test, the series
00
2 Ao + kY, (AA cos kx + Bk sin kx)
400 Chapter9 Orthogonal Functions and Fourier Series
1 sinn Innxn
°`
= 00.
=2n In n
(b) In 1903, Lebesgue proved the following stronger version of Theorem 9.3.6: If
AX) = 2Ao + l ° t(Ak cos kx + Bk sin kx) for all x E (-1r, a), and if f is continuous
(in fact, measurable), then Ak and Bk are the Fourier coefficients of the function f.3 (See
also Miscellaneous Exercise 1.)
We now turn to the negative results. Consider the series
0 sin nx
a2 Inn (n
which by Theorem 7.2.6, converges for all x E R. However, there is no Riemann inte-
grable function f on [ -7r, or] such that
Inn =
b = If * f(x) sin nxdx.
n
n=2
hi
°,
=2 (Inn)'
' n
f2(x)dx.
3. "Sur les series trigonometric:' Annales Scienriftques de I'Ecole Normale Supfrieure. (3) 20 (1903).
453-485.
9.3 Trigonometric and Fourier Series 401
9.3.7 DEFINITION For f E 9t[0, a], the Fourier sine series off is given by
00
f(x) 7, b sin nx,
n-I
where
rA
bn = ir J f (x) sin nxdx, n = 1, 2, ..
0
are the Fourier sine coefficients of f Similarly, the Fourier cosine series of
f E 9t[0, 7r] is given by
1
AX) - a0 + 00 a, cos nx,
n=
where
A
2 2
110 = f(x) dx and a=a f(x) cos nxdx, n = 1, 2, ..
0 0
00
1
AX) a0 + I a cos nx, where a _ jf(x)cosnxdx.
2 n-l 0
00 A
2
f (x) -r b,, sin nx, where b = f (x) sin nxdx.
n=1 77 I
0
402 Chapter 9 Orthogonal Functions and Fourier Series
Thus the coefficients and depend only on the values of the function f on
[0, ar]. Conversely, given a function f on [0, ir), the following definition extends f both
as an even and an odd function to the interval (-i, ir).
9.3.8 DEFINITION Let f be a real-valued function defined on [0, ir). The even extension of
f to (-it, it), denoted f is the function defined by
f(-x). -a < x < 0,
fe(x) =
f (x), 0 s x < V.
Similarly, the odd extension off to (-a, a), denoted f0, is the function defined by
1-f(-X). -T < x < 0,
f,(x) = 0, x = 0,
P X), 0 < x < V.
It is easily seen that the functions f,(x) and fo(x) are even and odd respectively on
(-1r, it), and agree with the given function f on (0, ar). For the function f given by the
graph in Figure 9.3, the graph of the even and odd extensions off are given in Figures
9.4(a) and (b) respectively. From the above discussion it is easily seen that if
f E 9t[0, ir], then the Fourier sine series off is equal to the Fourier series of f and the
Fourier cosine series off is equal to the Fourier series of f, (Exercise 2).
Figure 9.3
A -A
Figure 9.4
(a) Even Extension off (b) Odd Extension off
9.3 Trigonometric and Fourier Series 403
Remark, In the definition of the even and odd extension we only assumed that f was
defined on [0, a), and then defined f, and f, on (-a, Ir). If
Air-) = xlim_ f(x)
exists, then for the even extension we define f,(- v) = f(zr-). Thus f, is now defined
on [-ir, ir) and hence can be extended to all of TlR as a 2v-periodic function. For the
odd extension f we set f0(-w) = f(ir-), thereby defining f on [-1r, a).
EXERCISES 9.3
1. Let f E 9t(-a, ar). Prove the following.
'a. If f is even on [ -a, in, then the Fourier series off is given by
fv f (x) cos nxdx, it = 0, 1, 2, .
Ax) ^- 1 as + Y , a,, cos nx, where a
2 n-I o
1, 0 x< 2,
0, 2 x< in
4. On the interval [-2A, 2Tr) sketch the graph of the 2w-periodic extension of each of the functions in Exercise 3.
S. Find the Fourier sine and cosine series on [0, ar) of each of the following functions.
0 is
'a. f(x) = I b. f(x) = x c. f(x) = x2 'd. f(x) _ {0' x <a
2
cx, OsxsT2-
h(X) = 0 -4 2<x<a,
where c > 0 is a constant.
404 Chapter 9 Orthogonal Functions and Fourier Series
sinnx
n=i n
converges for all x E IR but is not the Fourier series of a Riemann integrable function on [ -ir, zr].
10. If f is absolutely integrable on [ -it, zr] (see Section 6.4), prove that
n
n
lint J
n
f(x) cos nxdx = lim
J
f(x) sin nxdx = 0.
11. Using the previous exercise, show that each of the following hold.
12. Suppose an ? 0 for all n E N, and {nan} is monotone decreasing with lim nan = 0. Prove that J. ,an sin nx
converges uniformly. (Hint: Write I an sin nx as I (nan) (" n') and use the fact that the partial sums of ";"'
are uniformly bounded. See also Exercise 15 of Section 8.2.)
Our goal in this section will be to prove that if f E Jt[-a, irJ, then the sequence
converges in the mean to the function f on [ - ir, or]. By Theorem 9.2.4, this is
equivalent to completeness of the trigonometric system (1, cos nx, sin nx}. To investi-
gate mean-square convergence, and also pointwise convergence in the next section, it
will be useful to obtain an integral expression for S..
sin (n + j)t
2sin Z
t#2pir,pEZ,
D (t) = 1 + 7, Cos kt =
2 k-t
n+2, 1
t=2pir.
'r j
1 1 "
f (t) - + cos k(x - t) J dt.
A
2 k-1
J t) dt.
A
406 Chapter 9 Orthogonal Functions and Fourier Series
Y'I=n+1.
2 ka, 2
I JDn(t) dt = 1,
A
(8)
a fact which will prove useful later. To see that equation (8) holds, it suffices to inte-
grate the sum defining term by term. The only nonzero term will be the integral
involving Z, and thus
Ltt=Lt= .
4T
o"(x) = n+1
However, it is possible that the sequence {S"(x)} diverges for a particularx, whereas the
sequence {o"(x)} may converge.
9.4.2 LEMMA For n E N, let S"(x) = iao + Y,k.1(ak cos kx + bk sin kx). Then
and
A
F"(t) = I r pk(t) -
n+lk,.o
2(n + 1)[
n+1
sin 2
t#
t - 2pzr.
2 '
408 Chapter 9 Orthogonal Functions and Fourier Series
fA s) ds,
f(s)Fn(x - s) ds,
where
1 7, Dk(t).
n+ 1 k=0
F 2or) =
"(P n+I2) =(n+1)
1 A0( k +
t
2
If t * 2p7r, p E Z, then
IT 1
k+ t
FF(t)
= 2(n + 1) sin 71 sin 2
r r
sin -sin k +
2
i2t
t=-
1
2k-0
(cos kt - cos(k + 1)t)
k-0
9.4.4 THEOREM
(a) F. is periodic of period 21r with
F (t) i? 0 for all t.
(c) IV
dt = 1.
1
(d) For 0 < 8 < ir, 1im 0 uniformly for all t, 8 s 1 tt c ir.
To prove (d), we first note that sine 2 --- sine I for all t, 0 < 8 Itt ir. Also,
since Isin(n + 1);I s 1 for all t,
on(X) _ - I t)dt,
J n
Since the function s -+ f(s + x)F (s) is periodic of period 21r, by Theorem 8.6.3
limo,(x) = f(x)
9A.6 COROLLARY If f is a continuous real-valued function on[-1r, or] with f(-v) = f(or),
then
n
lim [f(x) - SS(x)]Z dx = 0.
_n
Remark. There is a similarity between Theorem 9.4.5 and the Weierstrass approxi-
mation theorem. An alternative way of expressing. Theorem 9.4.5 is as follows: Let! be
9.4 Convergence in the Mean of Fourier Series 411
continuous on [ -vr, vr] with f(-v) = f(ir). Given e > 0, there exists a trigonomet-
ric polynomial
n
T (x) _ Ao + (Ak cos kx + Bk sin kr)
2 k-I
such that
I1(x) - E
for all x E vr, vr]. The function is called a trigonometric polynomial since in
complex form it can be rewritten as
cke'kx,
T.(x) = L;
k--
where by De Moivre's formula e' = cos kx + i sin kx.
9A.8 LEMMA, Let f E 3t[a, b] with I f(x)I s M for all x E [a, b). Then given e > 0,
there exists a continuous function g on [a, b] with g(a) = g(b) and I g(x)I <- M for all
x E [a, b], such that
b
f
a
If(x) g(x)I dx < e.
Before proving the lemma, we will first use the result to prove Theorem 9.4.7, then
consider some consequences of this. theorem.
Proof of Theorem 9.4.7. Suppose I f(x)I <_ M with M > 0, and let e > 0 be
given. By the lemma, there exists a continuous function g on [-n. vr] with
g(-vr) = g(vr) and Ig(x)I 5 M for all x E [-vr, vr] such that
A
where D. is the Dirichlet kernel. Then since g is continuous, by Corollary 9.4.6 the se-
quence {S (g)} converges in the mean to g on [ - ir, it ]. Thus there exists n,, E N such that
11(x)-g(x)I2dx52MIf(x)-g(x)Idx<4
Therefore, for all n ? n,,,
However, for each n E N, S (g) is the nth partial sum of a trigonometric series. Thus if
S is the nth partial sum of the Fourier series off, by Corollary 9.1.5
[f(x) - E
for all n > n,; i.e., {S.) converges in the mean to f on [ -rr,1r). 0
As a consequence of Theorem 9.2.4 and Theorem 9.4.7 we have the following
corollary.
2 ao +
00
(a,2, + b.) fr f 2(X) dx.
I, I A
9.4 Convergence in the Mean of Fourier Series 413
00
4 2
_n X2dx=-a2,
n2 3
1
which gives
= 1r2
., _n2 6
0
Proof of Lemma 9.4.8. Let f E Jt[a, b] with if (x)I s M for all x E [a, b]. Since
f,(x) = f(x) + M is nonnegative, if we can prove the result for the function f,, the re-
sult also follows for the function f. Thus we can assume that f satisfies 0 <- f(x) s M
for all xE[a,b].
Let e > 0 be given. Since f is Riemann integrable on [a, b], there exists a partition
_ {xo, x,, ... , of [a, b] such that
0s jf(x)
b dx - f(9,f) < 2'
where
m;Ox;,
and m; = inf { f(t) : t E [x;_,, x;]}, i = 1 , 2, ... , n. For each i = 1, 2... , n, define
the functions h; on [a, b] by
mi, x..1 <_ x < X"
hi(x) =
0, elsewhere,
and let h(x) = "_, h,{x) (see Figure 9.7). The function h is called a step function
on [a, b]. Since his continuous except at a finite number of points, h is Riemann integ-
rable and
rb
h(x) dx n m;Ox; = Y(J', f).
J
Therefore,
0s I[f(x)-h(x)]dx<2.
b
a
414 Chapter9 Orthogonal Functions and Fourier Series
m3 +
M5 t
M4 -4-
M2 t
mi t
mb + 0
I I I I i I I
xp=a X1 X2 X3 X4. X5 x6=b
Also, since m; = inf{f(t) : t E [x;_ 1, x;]}, 0:5 h(x) s f(x) for all x E [a, b]. By tatt-
ing slightly shorter intervals and connecting the endpoints with straight line segments.
we leave it as an exercise (Exercise 9) to show that there exists a continuous function g
on [a, b] with g(a) = g(b) = 0, 0 s g(x) h(x), such that
Jb
<ZE+je=E. O
EXERCISES 9.4
1. Prove Lemma 9.4.2(a).
2. *Using the Fourier series of f(x) = x2 and Parseval's equality, find 1/n4.
3. *Prove that the orthogonal systems {sin nx} -I and {1, cos nx}; , are both complete on [0, a].
4. Show that the set {sin sin x, sin i, ...} is complete on [0, ir].
5. Let f E 9t[ -ar, ir], and let S. denote the nth partial sum of the Fourier series of f.
a. Using the Cauchy-Schwarz inequality for integrals (Exercise 13, Section 9.1), prove that
JIR
(Sa(x) - f(x))dxl 29rt Jp IS.(x) - f(x)I3dx
9.5 Pointwise Convergence of Fourier Series 415
b. Prove that
A (!I
6. *Use the previous exercise to prove the following: Suppose f E 1t[ -7r, a] with
AX) - ao + (a cos nx + b sin nx).
For x E [-7r, ar], set F(x) = f af (t) dt - 1 Then F is continuous on [ v. ar] with F(-7r) = F(n). and
for all x E [-ir, ir],
v[f(x + t) f(x - t)
A=2J 2
- 1
d4
0
where D. is the Dirichlet kernel and S. is the nth partial sum of the Fourier series off.
Therefore,
S (x) _ f (x - s)D(s) ds
J 0
1
Jo f(x - ds = -- J of(x + u)DD(u) du
A A
fn
= 1IT f(x + s)D.(s) ds.
0
Therefore,
Z (1f(x+s) f(x- s)
Jo 2
which is the desired identity. U
Proof. Extend g to [-or, or) by defining g(x) = 0 for all x, -ors x < 0. Since
sin(n + J)t = cos i sin nt + sin i cos nt, we have
where g,(t) = g(t) cos 1 and g2(t) = g(t) sin 2. Since g is Riemann integrable on [-ar, ir;,
so are both g, and $2. Thus by Theorem 9.3.4,
A
(
lim 1 g,(i) sin nt dt = n-
lim g2(t) cos nt dt = 0,
nom= 00
A -A
Dirichlet's Theorem
Before we state and prove Dirichlet's theorem, we briefly review some notation introduced
in Chapter 4. For a real-valued function f defined in a neighborhood of a given point x the
right and left limits off at x0, denoted f (x,+) and f (x,-) respectively, are defined by
f(x,+) = lim f(x) and f(x,-) = lim_ f(x),
2IV[f(x°+t) 2
f(x°-t)
lsin(n + ;)t
+ (f(x. - t) - f(xo-))1 2 sin i
dt
(A A
where
gi(t)
=f(x + t) -f(xo+)
and
=f(x,, - t) -f(x.-)
2sin1 g2(t)
2sin f
To prove the result, it suffices to show that the functions g, and g2 are Riemann inte-
grable on [0, Tr]. If this is the case, then by Lemma 9.5.2.
lim1gi(t)sinln+2ftdt=0, i = 1,2.
U
\
Therefore,
A = 2[f(xo+) +f(xo-)]-
To finish the proof we still have to show that g, E R[0, ar], i = 1, 2. We will prove
the result for the function g,, the proof for g2 being similar. Set h(t) = f(x, + t) -
f (x,+ ). Since f E Jt[ -vr, 1r] and is periodic of period 2ir, h is Riemann integrable on
[0, a]. Also, (sin Z)-' is continuous on (0, v] and thus Riemann integrable on [c. 1r] for
any c, 0 < c < ir. Therefore,
gilt) = 2 h(t)
sin
is Riemann integrable on [c, 17] for any c, 0 < c < a. Let S > 0 be as in hypothesis
(b). Then for 0 < t < S, using inequality (10),
If(; + t) - f(x,+)I
I g i (t)I = 12 stn 21
t
If(xo + t) -f(x,+)I
Itl Si SMlsinll.
Since
lirax-
X-,0' sin x
there exists a constant C such that
tt2
Z sC
sin 2t
for all t, 0 < t s ir. Therefore, g, is bounded on (0, S), and hence also on (0, w]. Thus
since g, is Riemann integrable on [c, ir] for any c, 0 < c < r, by Exercise 5 of Sec-
tion 6.2, g, is Riemann integrable on (0, w].
or jump discontinuity at x0. Functions that have only simple discontinuities, and at most
a finite number on an interval [a, b], are said to be piecewise continuous on [a, b . We
make this precise with the following definition.
xf b
Figure 9.8
9.5.5 EXAMPLES
(a) Consider the function
x, 0<x<1,
f(x)- xZ+2, I <x<2.
Since f is continuous on (0, 1) and (1, 2) and f(0+), f(1-), f(1 +), and f(2-) all ex-
ist, f is piecewise continuous on [0, 2]. Also, since
f,(x)-{1, 0<x<1,
2x, 1 <x<2,
and f'(0+), f'(I-),f'(1 + ), and f'(2 -)all exist, f' is also piecewise continuous on [0, 2].
420 Chapter 9 Orthogonal Functions and Fourier Series
0, x=0,
f(x) xI sin 1, O < x <- 1.
X
0, x=0,
'x
f 2x sin 1
x
cos 1,
x
0 < x < 1.
However, since f'(0+) does not exist, f' is not piecewise continuous on [0, 1].
Returning to Dirichlet's theorem, suppose f is such that both f'(x,-) and f'(x,+)
exist at the point x0. If f' is piecewise continuous on [ -Tr, in, then there exists x, > x
such that f and f' are continuous on (x x,). If f is not continuous at x redefine f at x,
as f (x,+ ). Then f (redefined if necessary) is continuous on [x, x,), and thus by Theo-
rem 5.2.11,
Ax. + Ax" +)
li o t) I
As a consequence, there exists a constant M and a S > 0 such that
If(xo + t) - f(x,+)I <- Mr
for all t, 0 < t < S. Similarly, the existence of f'(x,-) implies the existence of a con-
stant M and a S > 0 such that
If(x - t) -f(x,,-)I <_ Mr
for all t, 0 < t < S. Thus f satisfies the hypothesis of Dirichiet's theorem. On combin-
ing the above discussion with Theorem 9.5.3 we obtain the following corollary.
9.5.7 EXAMPLES
(a) Consider the function f defined on [-ir, ar) by
-ITsx<- 2'
-2. . 2,
2 <x<ar.
9.5 Pointwise Convergence of Fourier Series 421
Extend f to all of U8 as a periodic function of period 2a. The function f is then continu-
ous except at x = 2 it + krr, k e Z. At each discontinuity x,,,
I[f(x"+) + f(xo-)] = 2
The graph of 2[f(x+) + f(x-)] on the interval [-27r, 21r] is given in Figure 9.9.
-O O-3 0 O-
To discuss convergence of the Fourier series off we first note that f is piecewise
continuous on [-1r, ar] and differentiable on (-1r, -2), (--, 1), and (Z, zr) with
f'(x) = 0 in the respective intervals. Thus f' is also piecewise continuous on [ -ir, ir].
As a consequence of Corollary 9.5.6, the Fourier series of f converges to
2[f(x+) +f(x-)] for all x E 08.
The Fourier series off is obtained as follows:
1
a2
ao=- J 3dx=3,
°/2
(2
6
a° 3cosnxdx=sin
n 2,n=1,2, ..
n/2
3
3 6 `
°° (-1)t
0 2k+l'
422 Chapter 9 Orthogonal Functions and Fourier Series
(b) Dirichlet's theorem can also be applied to the Fourier sine and cosine series of a
real-valued function f defined on [0,1r). As an example, consider the cosine series of
f(x) = x on (0, 7r). By Exercise 5(b) of Section 9.3, the cosine series of f(x) = x is
given by
x^ 2 4 00
I
k-I (2k 1
2 cos(2k - I)x.
1)'
Since the even extension f, off to [-a, Tr] is given by f,(x) x 1, the cosine series
of x is the Fourier series of I x I on [ -ir, 7r]. Since both f, and f; are piecewise con-
tinuous on [-ir, ir], the Fourier series converges to the 21r-periodic extension of I x
for all x E R. This function is illustrated in Figure 9.10.
-2n _p IT 2n
By Dirichlet's theorem, the Fourier series converges to I x I for all x E [-ir, rr].
Taking x = 0 gives
ir 4 00 1
0
2 k-l ( 2k- 1)2'
)
00
1 _ 7r -
.
.
or I (2k - 1)2 8
Remarks. Although the inequalities (10) of Theorem 9.5.3 are sufficient, they are by
no means necessary. There are variations of Dirichlet's theorem that provide sufficient
conditions on f to guarantee convergence of the series to 1 [ f (xo+) + f (xo-) ]. For ex-
ample, the inequalities of Theorem 9.5.3 can be replaced by the following:
for all t, 0 < t < 8, and some a, 0 < a 5 1.1f f satisfies the above at x0, then the con-
clusion of Dirichlet's theorem is still valid. An even more general condition comes from
Ulisse Dini (1845-1918). who proved that if f E k[ -zr, zr] satisfies
o8lf(xo + t) - f(x"+)I
dt < oo and
J t
ff&If(x"-t)-.f(x.-)Idt<oo
to t
for some S > 0, then the Fourier series off converges to [f(x"+) + at x".
2
Both of the above hold if f satisfies inequalities (11) at x0.
does not converge since its nth term fails to approach zero as n -> oo. The 21r-periodic
extension off has discontinuities at x = ±(2k - 1) it, k E Ni. It turns out that continu-
ity of the periodic function is important for differentiability of the Fourier series. Suffi-
cient conditions are given by the following theorem.
9.5.8 THEOREM Let f be a continuous function on [-ir, zr] with f(-ir) = f(v), and let
f be piecewise continuous on [-ir,,r]. If
is the Fourier series off then at each x E (-1r, ar) where f"(x) exists,
00
Remark. At a point x where f"(x) does not exist but f"(x-) and f" (x+) both exist,
the series (12).converges to 2[f'(x-) + f'(x+)].
424 Chapter9 Orthogonal Functions and Fourier Series
Proof. Suppose x E (-vr, ar) is such that f"(x) exists. Since fis continuous at x. by
Dirichlet's theorem
00
1
f' (x) = as + I (a" cos nx + P. sin nx).
2
1
a" = - f'(t) cos nt dt
2ar
= nb".
Similarly, /3" = -na,,, which proves the result. O
9.5.9 EXAMPLE Consider the function f(x) = x2. Since f is even on (-ar, ar),
00
i
f (x) = 2 ao + a" cos nx.
Therefore, a" = 4(-1)"/n2 for n = 1, 2, 3, . ... To find ao we use the definition. This
gives ao = 2ar2/3, and thus
.2f 2)"
x2 = + 4 001 ( cos nx, x E [-ar, 71.
3 "_1 n
Remarks. In closing this section, it should be mentioned that there exist continuous
functions f for which the Fourier series of f fails to converge at a given point. This was
first shown by P. du Bois Reymond. Other examples were subsequently constructed by
L. Fej6r and Lebesgue. The example of Fej6r can be found on p. 416 of the text by
9.5 Pointwise Convergence of Fourier Series 425
EXERCISES 9.5
1. For each of the functions of Exercise 3, Section 9.3, sketch the graph of the function on the interval [ - 21r. 2v to
which the Fourier series converges.
2. ;Use the Fourier series of f(x) = x2 to find
CO (-1)M+1
Y,
n-t
n22 and I00
n=1
1
n2.
3. Using the Fourier cosine series of f(x) = sin x on [0, ar) (Exercise 7, Section 9.3), find each of the following
sums:
;a 1
sb (-1)"
4rt2 I Ls+t 4n2 - 1
G 12 , a
e' =
eII
2
e-'[, - 2(cos;
(-1 nx - n sin nx)
'
2a nz+
b. Using part (a) find
W 1-11.-I W 1
G z and z
NOTES
There is no doubt about the significance of Fourier's con- The quest for an understanding of what types of functions
tributions to the areas of mathematical physics and ap- possessed Fourier series is partly responsible for the devel-
plied mathematics; one only needs to consult a text on opment of Lebesgue's theory of integration. The need for a
partial differential equations. The methods he developed more extensive theory of integration was illustrated by
relating to the theory of heat conduction are applicable to Lebesgue in 1903 in the paper Sur les series trigonometric.°
a large class of physical phenomena, including acoustics, In it he constructed an example of a function that is not Rie-
elasticity, optics, and the theory of electrical networks, mann integrable but that is representable everywhere by
among others. Fourier's work, however, is even more sig- its Fourier series. Such a function is f(x) _ -1a12 sin j1,
nificant in that it inaugurated a new area of mathematics. whose Fourier series is I. ,(cos kx)/k. This series con-
The study of Fourier series led to the development of verges everywere to the function f on [ - 7r, 1r), but since f is
the fundamental concepts and methods of what is now unbounded, it is not Riemann integrable on [ -ar, zr]. It is,
called real analysis. The study of the concept of a function however, integrable in the sense of Lebesgue. The article by
by Dirichlet and others was directly linked to their inter- Alan Gluchoff provides an excellent exposition on the influ-
est in Fourier series. The study of Fourier series by ence of trigonometric series to the theories of integration of
Riemann led to his development of the Riemann integral. Cauchy, Riemann, and Lebesgue.
He was concerned with the question of finding sufficient There are several important topics that we have not
conditions for the existence of the integrals that gave the touched upon in this chapter. One of these is the Gibbs
Fourier coefficients of a function f, that is, phenomenon, named after Josiah Gibbs (1839-1903).
To explain this phenomenon we consider the Fourier se-
a.= - f(x) cos nx dx and ries of f(x) = 0, x E [ -a, 0), and f(x) = 1, x E [0, or)
-II of Example 9.3.3 By Dirichlet's theorem we have
II
b= 1
f(x) sin nx dx. 8. Annales Scient s de PEcole Normale Supdrieure, (3) 20
-II (1903), 453-485.
Votes 427
+
3
+ (2k 1) sin(2k - 1)x
x
sin x x2 x4 x6 xt 13t+st7t+9!-..
1 (" sin x
dt= Ir-3,31+S. St
1 Tr3
Sy,_,(x) = 2[cosx + cos3x +
ir
+ cos(2k - 1)x].
0 x
-7.T++9 91-
sinxcosjx = 2 [sin(j + I)x - sin(j - I)x], ,R2 ,T4 ,F6
=1 3.3!+5.5! 7.7!
we obtain
- 1 - .54831 + .16235
From this it now follows that S2k_ t(x) has relative maxima
- .02725 + .00291
and minima at the points
.59 (to two decimal places).
2kx = ± 9r, ± 21r, .. , ± 2(k - 1)1r.
The notation'*-" denotes approximately equal to. There-
These points are equally spaced in (-,it, a). Consider the fore, 1.09. Even though ultimo S. (x) = 1
points xk = ar/(2k), at which each S3k_, has an (absolute) for all x E (0, Ir), if we approach zero along the points xt,
maximum with then S7k_, (xk) overshoots the value I by approximately
.09; i.e., kim LSy,_,(xk) - f(xt)l - 0.09. If the Fourier
series converges uniformly, this cannot happen.
Su_,(xk) = 2+ I sin 2k+ sin 2k
The above behavior, known as the Gibbs phenome-
3 non, is due to the fact that f has a jump discontinuity at
I (2k - 1)7r x0 = 0, and is typical of the behavior of the Fourier series
+ + Isin
2k of a piecewise continuous function at a jump discontinu-
ity. Furthermore, if f and f are piecewise continuous on
To find kiimaoS2k_,(xk), we write the above sum as a [ -a, 1r], then the amount of overshoot at a discontinuity
Riemann sum of the function g(x) = (sin x)/x. This func- x0, due to the Gibbs phenomenon, is approximately equal
tion is Riemann integrable on [0, a]. With the partition ' to 0.09[f(x0+) - The article by Shelupsky in
428 Chapter 9 Orthogonal Functions and Fourier Series
the supplemental readings provides a very nice discussion for all x E rr, rr]. where S. is the nth partial sum of the
of this phenomenon. series.
Another important question involves the uniqueness In 1870 Eduard Heine proved that if the sequence (S.}
of the representation of a function by a trigonometric se- converges uniformly to 0 on [ -ir, rr] then AA =Bk = 0 for
ries. Specifically, if all k. This is Theorem 9.3.6. The general uniqueness prob-
lem was solved by Cantor in the early 1870's. Furthermore,
1(x) _ 7, (Ak cos kx + Bk sin kx) he also proved uniqueness if equation (13) holds for all but
k-0 a finite number of x in [-rr, rr]. This then led Cantor to
(Ck cos kx + Dk sin kr) consider the uniqueness problem for infinite subsets of
k-0 [ - rr, rr]; specifically, if E C [ - ir, rr] is infinite and
for all x E [ -a, 'r], must Ak = Ck and Bk = Dk for all equation (13) holds for all x E [ -rr, a) \ E, must
Ak = Bk = 0 for all k? Since point set theory was undevel-
k = 0, 1, 2, . .? Alternatively, if
oped at this time, Cantor devoted much of his time and ef-
t
- Ao + T, (Ak cos kx + Bk sin kx) = 0 (13) fort to studying point subsets of R. For a thorough discus-
2 k-l sion of the uniqueness problem, refer to the article by
for all x E [ -rr, rr], must Ak = Bk = 0 for all k = 0, Marshall Ash listed in the supplemental readings, or to the
I.2, ... ? By equation (13) we mean that N-.OU 0 text by A. Zygmund listed in the Bibliography.
MISCELLANEOUS EXERCISES
1. Suppose f is continuous on [ -rr, rr) and f(x) = i A0 + 7,k i (At cos kr + Bk sin kr). Let S. be the nth partial sum
of the series. If there exists a positive constant M such that IS.(x)I < M for alI x E [ -rr. rr] and n e N. prove
that Ak and Bk are the Fourier coefficients off.
2. Prove that lim j ID.(t)I dt = oo, where D, is the Dirichlet kernel. (See Example 6.4.4(b).)
3. Let f(x) = -In12 sin J. Show that the Fourier series of f is given by 1(cos kx)lk (Note: Since f is unbounded
at x = 0, the integrals defining ak and bk have to be interpreted as improper integrals.)
SUPPLEMENTAL READING
Ash, Marshall J., "Uniqueness of representation by Lanczos, Cornelius, Discourse on Fourier Series. Hafner
trigonometric series:' Amer. Math. Monthly % Pubi. Co., New York, 1966.
(1989),873-885. Mulcahy, C., "Plotting and scheming with wavelets;'
Askey, R. and Haimo. D. T., "Similarities between Math. Mag. 69 (1996), 323-343.
Fourier series and power series,' Amer. Math. Shelupsky, D., "Derivation of the Gibbs phenomenon;'
Monthly 103 (19%), 297-304. Amer. Math. Monthly 87 (1980). 210-212.
Gluchoff, Alan D., "Trigonometric series and theories of Shepp, L. A. and Kruskal, J. B., "Computerized tomog-
integration;' Math. Mag. 67 (1994). 3-20. raphy: The new medical X-ray technology;' Amer
Gonzalez-Velasco. Enrique A., "Connections in mathe- Math. Monthly 85 (1978), 420-439.
matical analysis: The case of Fourier series;' Amer. Simon, Barry, "Uniform convergence of Fourier series;'
Math. Monthly 99 (1992), 427-441. Amer. Math. Monthly 67 (1969), 55-56.
Halmos, R. "Fourier series;' Amer. Math. Monthly 85 Williams, K. S., "Note on f o (sin x/x) dx," Math. Mag.
(1978), 33-34. 44 (1971),9-11.
1 O Lebesgue Measure
and Integration
10.1 Introduction to Measure
10.2 Measure of Open Sets; Compact Sets
10.3 Inner and Outer Measure; Measurable Sets
10.4 Properties of Measurable Sets
10.5 Measurable Functions
10.6 The Lebesgue Integral of a Bounded Function
10.7 The General Lebesgue Integral
10.8 Square Integrable Functions
The concept of measure plays a very important role in the theory of real analysis. On
the real line, the idea of measure generalizes the length of an interval, in the plane, the
area of a rectangle, and so forth. It allows us to talk about the measure of a set in the
same way that we talk about the length of an interval. The development of the Riemann
integral of a bounded function on a closed and bounded interval [a, b depended very
much on the fact that we partitioned [a, b] into intervals. The notion of measure and
measurable set will play a prominent role in the development of the Lebesgue integral
in that we will partition [a, b] not into intervals, but instead into pairwise disjoint mea-
surable sets.
The theory of measure is credited to Henri Lebesgue (1875-1941) who, in his
famous 1902 thesis, defined measure of subsets of the line and the plane, as well as the
Lebesgue integral of a nonnegative function. Like Riemann, Lebesgue was led to the
development of his theory of integration while searching for sufficient conditions on
a function f for which the integrals defining the Fourier coefficients off exist. In this
chapter we will develop the theory of Lebesgue measure of subsets of I following
the original approach of Lebesgue using inner and outer measure. Although this ap-
proach is somewhat more tedious than the modern approach developed by Constantin
Carathdodory (1873-1950), it has the advantage of being more intuitive and easier to
visualize.
In the first section we will illustrate the need for the concept of measure of a set
and measurable function by considering an alternative approach to integration devel-
oped by Lebesgue in 1928. Although this ultimately will not be how we define the
429
430 Chapter 10 Lebesgue Measure and Integration
Lebesgue integral, the approach is instructive in emphasizing the concepts required for
the development of the Lebesgue theory of integration. In Section 10.2 we use the fact
that every open subset of R can be expressed as a finite or countable union of disjoint
open intervals to define the measure of open sets, and then of compact sets. These are
then used to define the inner and outer measure of subsets of R. A bounded subset of R
is then said to be measurable if these two quantities are the same.
In Section 10.6 we develop the theory of the Lebesgue integral of a bounded real-
valued function using upper and lower sums as in the development of the Riemann
integral. However, rather than using point partitions of the interval, we will use mea-
surable partitions. The key result of the section is that a bounded real-valued function
on [a, b] is Lebesgue integrable if and only if it is measurable. As we will see, the class
of Lebesgue integrable functions contains the class of Riemann integrable functions,
and for Riemann integrable functions, the two integrals coincide. One of the advantages
of the Lebesgue theory of integration involves the interchange of limits of sequences of
functions and integration. We will prove several very important and useful convergence
theorems, including the well-known bounded convergence theorem and Lebesgue's
dominated convergence theorem.
1U-1)nn j = 1, 2, , n.
Y
4
1$
11
6
{
1 1 I
{
x
a x1 ;2 -3 -'14 '1 b
Figure 10.1
For each j = 1, 2, ... , n, we let
E4= {xE [a, b] : 3 8 s f(x) < 4 8 } = [a, x1) U (xz, x3) U (xs, xs]
For a set such as E4, a finite union of disjoint intervals, it is reasonable to define the
measure of E,, denoted m(E,), as the sum of the length of the intervals, i.e.,
m(E4) = (x1 - a) + (x3 - x2) + (x5 - X,).
Assuming that we can do this for each of the sets Ep a lower approximation to the area
under the graph off is given by
f,
and an upper approximation is given by
i- j8m(Ei)
Taking limits as n -), oo, assuming that they exist and are equal, would in fact, provide
another approach to integration. That this -indeed is the case for a large class of func-
tions, including the Riemann integrable functions, will be proved in Section 10.6.
For nice functions, namely those for which the sets E, are finite unions of intervals,
the above is perfectly reasonable. However, suppose our function f is defined on [0, 1: by
f(x) = J0, x E a fl [0, 1],
x, elsewhere.
432 Chapter 10 Lebesgue Measure and Integration
E= {xE[O.l]:-1f(x)<i-}
n n
I
E, = x irrational : < x < i-
n n
Here the sets Ej are no longer unions of intervals, so what is meant by the measure of
the set is by no means obvious.
Our goal in the next two sections is to define a function A on a family .44 of subsets
of I8, called the measurable sets. The collection M will contain all the intervals and A
will satisfy the following:
Thus if J is (a, b), (a, b], [a. b), or [a, b], a, b E I8, then
m(J)=b-a.
If J is R, (a, oo), [a, oo), (-oo, b) or (-oo, b], we set m(J) = oo. In dealing with the
symbols oo and -oo, it is customary to adopt the following conventions:
(a) If x is real, then x + oo = oo, x - oo = -oo.
(b) If x > 0 then x oo = oo, x (-oo) -oo.
(c) If x < 0 then x - oo = - oo, x (- oo) = oo.
(d) Also, oo + co = oo, -oo -oo = =oo, oo (too) = ±oo,
-00*(±oo)=zoo.
The symbols oo - oo and -oo + oo are undefined, but we shall adopt the arbitrary
convention that 0 oo = 0.
10.2 Measure of Open Sets; Compact Sets 433
U=U In. R
Recall, the family {1"} is pairwise disjoint if and only if I" f l 1,n = 40 whenever n * m.
m(U) m(I,).
Remarks
(a) For the empty set 0, we set m(¢) = 0.
(b) The sum defining m(U) may be either finite or infinite. If any of the intervals are of
infinite length, then m(U) = no. On the other hand, if
U=UI,,,
where the 1n are pairwise disjoint bounded open intervals, we may still have
00
m(U) = I m(1") = 00,
n-1
due to the divergence of the series to no. Since m(I,) ? 0 for all n, the sequence of par-
tial sums is monotone increasing and thus will either converge to a real number or di-
verge to no.
10.2.3 EXAMPLES
(a) Foreach n= 1,2,. ..,set
In=(n - 2",n+Zn
1 °D 1 1
m(U) = G m(I4) _ 2-2 = 2n
2.
n-1 n 11 n-0 12
434 Chapter 10 Lebesgue Measure and Integration
(c) As in Section 3.3, let U denote the complement of the Cantor set in [0, 1]. Since
U is the union of the open intervals that have been removed, by Property 4 of the Can-
tor set,
m(U)= 1.
We now state and prove several results concerning the measure of open sets.
Proof. The statement of the theorem appears to be so obvious that no proof seems to
be required. However, it is important to keep in mind how the measure of an open set
is defined. Suppose
U=UI, n
and V=UJm,
where {In}n and {Jm}m are finite or countable collections of pairwise disjoint open in-
tervals. Since U C V, each interval 4 C J. for some m. For each m, let
Since the collection {Jm}m is pairwise disjoint, so is the collection {Nm}m, and
U=U1.=UnU I,,.
Therefore,
m(U) _mnEN,
. m(I.).
But by Exercise 1,
I m(I4) c m(Jm),
FE N.
for all k E N. By Theorem 10.2.4, the sequence {m(Uk)} is monotone increasing with
m(Uk) : m(U) for all k. Therefore,
rlim m(Uk) 5 m(U). (1)
oo
If U is bounded, then there exists ko E N such that
Ufl(-k,k)=U
for all k a ko. Hence m(Uk) = m(U) for all k k and thus equality holds in equa-
tion (1).
Suppose that U is an unbounded open subset of R with
U=UIi,n
Therefore equality holds in equation (1). The other two cases follow similarly.
Suppose m(In) < oo for all n. Since U is unbounded, the collection {In} must be
infinite. If the collection were finite, then since each interval has finite length, each in-
terval is bounded, and as a consequence U must also be bounded. Let a E R with
a < m(U). Since
0
G m(I4) = m(U) > a,
n-1
Let V = UN 11,,. Then V is a bounded open set, and thus by the above.
m(V) =slim m(V fl (-k, k)).
Remark. In proving results about open sets, the previous theorem allows us to first
prove the result for the case where U is a bounded open set, and then to use the limit
process to extend the result to the unbounded case.
Our next goal is to prove the following.
tn(U U.) t n
m(U.).
m( U In I M(In)-
n=1 n=1
It should be noted that the collection {In} is not assumed to be pairwise disjoint.
The lemma is most easily proved by resorting to the theory of the Riemann integral.
Suppose 1 is a bounded open interval. Choose a, b E R such that I C [a, b]. Since
Xf is continuous on [a, b] except at the two endpoints of 1, x, E k[a, b] with
Jb XI(x) dx = m(1).
U= UJk,
k=1
Xu(x) = I Xj,(x)
k-I
and thus
n n
Xu(x) t 7, X4(x).
n=1
Proof of Theorem 10.2.6. Since the result for a finite collection follows obviously
from that of a countable collection, we suppose {Un}'° 1 is a countable collection of
open sets and
00
U= UIU,,.
U. = U In, k.
where for each n, {I,, k}k is a finite or countable collection of pairwise disjoint open in-
tervals.
438 Chapter 10 Lebesgue Measure and Integration
We assume first that U is bounded with m(U) < oo. Let e > 0 be given. If the col-
lection {Jm}m is infinite, then since F'm m(Jm) < oo, there exists a positive integer N
such that
00
I m(J,) < e.
m-N+1
Thus
N
m(U) < Y, m(J,) + E.
M-1
Let A = U v=1 Km. Since each k. is closed and bounded, so is the set A. Thus A is com-
pact, and since
AC U=nU1(
the collection {I,,, k} is an open cover of A. Hence by compactness there exists a finite
number k,,, i = 1, ... , J. j = 1, . . . , mi, such that
ACUIn,.k
i.!
Y, m(K.) = m(U
m=I
Km) mU
aJ
m =1
`- m(U,,) s
!=1 n=I
Since e > 0 was arbitrary, the result follows for the case where U is bounded.
If U is unbounded, then for each k E N,
00 x
m(Ufl(-k,k))<_y m(u,, n(-k. k))52 m(U,).
n=1 n=1
Proof. (a) If both U and V are unions of a finite number of bounded open intervals.
then so are u n V and U U V. Thus the functions
XU, Xv, Xuuv, Xunv
are all Riemann integrable on some interval [a, b] with
Xu(X) + Xv(x) = Xuuv(x) + Xunv(x)
for all x E [a, b]. Therefore by identity (2),
m(U) + m(V) = m(u U V) + m(u n v).
(b) For the general case, suppose
00
U= U
n-I
In and V = U Jn,
n=1
where the collections {In} and {J.} consist of pairwise disjoint open intervals respec-
tively. If one of m(U) or m(V) is oo, then by Theorem 10.2.4, m(U U V) = oo and the
conclusion holds. Thus we can assume that both m(U) and m(V) are finite.
Let e > 0 be given. Choose N E N such that
00 00
Y, m(I,) < e and m(JJ) < C.
n=N+I n-N+1
U \ K = 0(x,-6,x;)U(xx,x;+S).
10.2.13 THEOREM
Since my. fl [a, b)) = m(1 fl (a, b)) for all n, we have
m(U fl [a, b]) = m(U fl (a, b)).
Remark. What the above definition really defines is the measure of relatively open
subsets of [a, b) (see Definition 3.1.14). By Theorem 3.1.16, a subset G of [a, b] is
open in [a, b] if and only if there exists an open subset U of R such that G = U n [a, b].
Since the set U is not unique, we leave it as an exercise (Exercise 4) to show that if U.
V are open subsets of ltd with
u fl [a, b] = V fl [a, b],
then m(U fl [a, b]) = m(V fl [a, b]).
m(Un[a,b])+m(U`n[a,b])<--b-a.
To prove the reverse inequality,
let 1, = [a + e, b - e], where 0 < e <
2(b - a). Then
m(U n [a, b])+ m(u° n [a, b]) >- m(U n (a, b)) + m(U` n 4).
Since (a, b) is an open set containing U` n /E,
EXERCISES 10.2
1. If is a finite or countable collection of disjoint open intervals with U 1 C (a, b), prove that
10.3.2 THEOREM
(a) For any subset E of R, 0 < A*(E) t A*(E).
(b) If EI and E2 are subsets of a8 with El C E,, then
A*(EI) A*(E,) and A*(EI) < A*(E,).
Taking the supremum over all compact subsets K of E proves (a). The proof of (b) is
similar and is left as an exercise (Exercise 7). 0
10.3.3 EXAMPLES
(a) If E is any countable subset of R. then
A*(E) = A*(E) = 0.
Suppose E = {xn}" I. Let E > 0 be arbitrary. For each n, let
e
In = Xn - 2n, Xn + 2n
Therefore, A*(E) < 2e. Since e > 0 was arbitrary, A*(E) = 0. As a consequence, we
also have A*(E) = 0.
10.3 Inner and Outer Measure; Measurable Sets 445
b -a - e=ml [ a+2,b-2Jj_A*(1).
Therefore,
b-a-ESA*(1)SA*(1)Sb-a.
Since e > 0 was arbitrary, equality holds. A similar argument proves that if 1 is any
closed and bounded interval, then
As a consequence of Theorem 10.3.2(b), the result holds for any bounded interval 1.
(c) For any open set U,
A*(U) = A*(U) = m(U).
By definition, A*(U) = m(U). But m(U) _ where {I,} is a pairwise disjoint
collection of open intervals with U = Suppose a E R satisfies a < A*(U).
Since m(U) > a, there exist a finite number of intervals 11, ... ,1N such that
$N t m(1) > a. For each j, choose a closed and bounded interval Ji C I, such that
j tm(J,) > a. Let K = U t t J. Then K is a compact subset of U and thus
A*(U) > m(K). Finally, since the intervals {J% I are pairwise disjoint, by Exercise 6
of Section 10.2,
Therefore A*(U) > a. If A*(U) = oo, then by the above A*(U) > a for every a E R;
that is, A*(U) = oo. On the other hand, if A*(U) is finite, take a = A*(U) - E, where
e > 0 is arbitrary. But then A*(U) ? A* (U) > A*(U) - e for every e > 0. From this
it now follows that A*(U) = A*(U) = m(U).
Measurable Sets
In both of the previous examples, the inner and outer measures of the sets are equal. As
we shall see, all subsets of R built out of open sets or closed sets by countable unions,
intersections, and complementation will have this property. This includes most sets en-
countered in practice. In fact, the explicit construction of a set whose inner and outer
measures are different requires use of an axiom from set theory, the Axiom of Choice,
446 Chapter 1 0 Lebesgue Measure and Integration
which we have not discussed. The construction of such a set is outlined in the miscel-
laneous exercises.
10.3A DEFINITION
(a) A bounded subset E of ff is said to be LAbesgue measurable or measurable if
A*(E) = A*(E).
If this is the case, then the measure of E, denoted A(E), is defined as
A(E) = A*(E) = k*(E).
(b) An unbounded set E is measurable if E fl [a, b] is measurable for every
closed and bounded interval [a, b]. If this is the case, we define
A(E) = klint A(E fl [-k, kJ).
Remarks
(a) If E is unbounded and E fl l is measurable for every closed and bounded interval
1, then by Theorem 10.3.2 the sequence {A(E rl [-k, k])} t is nondecreasing, and as
a consequence A(E) = lim A(E fl [ -k, k]) exists.
(b) There is no discrepancy between the two parts of the definition. We will shortly
prove in Theorem 10.4.1 that if E is a bounded measurable set, then E fl 1 is measur-
able for every interval 1. Conversely, if E is a bounded set for which
10.3.5 THEOREM Every set E of outer measure zero is measurable with A(E) = 0.
Proof. Suppose E C R with A*(E) = 0. Then for any closed and bounded interval 1.
A*(E fl I) A*(E) = 0.
Thus A* (En I) = A*(E fl I) = 0, and hence E fl I is measurable for every closed and
bounded interval I. Since A(E fl [ -k, k]) = 0 for every k E N, A(E) = 0. 0
As a consequence of the previous theorem and Example 10.3.3(a), every countable
set E is measurable with A(E) = 0. In particular, Q is measurable with A(Q) = 0. An-
other consequence of Theorem 10.3.5 is that every subset of a set of measure zero is
measurable.
10.3 Inner and Outer Measure; Measurable Sets 447
Proof. Since the proof is similar to that of Theorem 10.2.5 we leave it as an exercise
(Exercise 8).
Proof. (a) If A*(E;) = oo for some i, i = 1, 2, then inequality (a) certainly holds.
Thus suppose A*(E;) < oo for i = 1, 2. Let e > 0 be given. By the definition of outer
measure, for each i, we can choose an open set U. containing E, such that
Therefore,
e + A*(E,) + A*(E,) > m(U,) + m(U,),
which by Theorem 10.2.9
=m(U,UU2)+m(U,nU2)
? A*(E, U E2) + A*(E, n E2).
The last inequality follows from the definition of outer measure. Since e > 0 was ar-
bitrary, inequality (a) follows.
(b) Let a, b E Il8 be arbitrary. By (a) applied to [a, b] n E;, we have
A*([a, b] n E,) + A*([a, b] n El)
? A*([a, b] n (E, U El)) + A*([a, b] n (E, n E;))
= A*([a, b] n (E, n E2)`) + A*([a, b] n (E, U E,)`).
But by Theorem 10.3.7, for any E C R,
A*([a, b] n E`) = (b - a) - A*(E n [a, b)).
Therefore,
A*(E, n [a, b]) + A*(E2 n [a, b])
<_ A*([a, b] n (E, n E2)) + A*([a, b] n (E, U E,)).
For each k E N. set 4 = [ -k, k]. By the above,
A*(E, n Ik) + A* (E2 n Ik) 5 A*((E, n E,) n 1k)
+A*((E, U E2) n 1k) s A*(E, n E2) + A*(E, U E2).
The result now follows by Lemma 10.3.8.
EXERCISES 10.3
1. a. If E C R. a, b E IR, prove that A*(E n (a. b)) = A*(E n [a, b]) and A*(E n (a, b)) = A*(E n [a. b]).
*b. If E C R, prove that A*(E + x) = A*(E) and A*(E + x) = A*(E) for every x E R, where
E+x=(a+ x: aEE).
2. Prove that every subset of a set of measure zero is measurable.
3. *Let E, C R with A*(E,) = 0. If E2 is a measurable subset of R, prove that E, n E. and E, U E2 are measurable.
4. Let E _ [0, 1 ] \ Q. Prove that E is measurable and A(E) = 1.
5. Let P denote the Cantor set in [0, 11. Prove that A*(P) = 0.
6. *If E C R, prove that there exists a sequence of open sets with E C U. for all n E N such that
-A*(E) = A*( n U,).
7. Prove Theorem 10.3.2(b).
& *Prove Lemma 10.3.8.
9. a. Prove that every open set U is measurable with A(U) = m(U).
b. Prove that every compact set K is measurable with A(K) = m(K).
c. Prove that every closed set is measurable.
10.4 Properties of Measurable Sets 449
l Ek and U Ek
are measurable.
Since A*(E n [a, b]) c A*(E n [a, b]), the above can hold if and only if
A*(E Cl [a, b]) = A*(E n [a, b]).
Thus E n [a, b] is measurable for every a, b E R. Hence E is measurable. Q
Proof. This is an immediate consequence of the fact that E satisfies Theorem 10.4.2
if and only if E' satisfies Theorem 10.4.2. Q
Our next goal is to show that the union and intersection of a countable collection
of measurable sets is again measurable. For the proof of this result we require the fol-
lowing theorem.
10.4 Properties of Measurable Sets 451
10.4A THEOREM
(a) If is a sequence of subsets of 18, then
k( U'
n=1
E- ?I
R-I
Proof. (a) If :, A*(En) = oo, then the conclusion in (a) is certainly true. Thus we
assume that Y,, A*(EE) < coo. Let e > 0 be given. For each n E N. there exists an
open set U. with En C Un such that
U En and nE
are measurable with
Proof. Let E = Un°_ I E. Without loss of generality we can assume that E (and
hence each of the sets ER) is bounded. If not, we consider
En[a,b] = RU (Enn(a,b])
for a, b E R.
Set A, = El, and for each n E N, n ? 2, set
An = ER \ (U
k=I
Ek) = (n EE)c
k-1
Since finite unions, intersections, and complements of measurable sets are again mea-
surable, An is measurable for each n E N. Furthermore, the sets An, n E N, are pair-
wise disjoint with
«;
U An = E.
10.4.6 THEOREM
(a) If {En}' 1 is a sequence of measurable sets with E, C E2 C , then
AI U En J = lim A(E0).
N=1 noc
10.4 Properties of Measurable Sets 453
A(E,) = A(E) +
n-I
N
= A(E) + lim T, [A(E.) - A(En+1)]
n=1
countable union of pairwise disjoint open intervals, by Theorem 10.4.5 every open set
U is measurable with
A(U) = m(U).
Since the complement of every measurable set is measurable, At also contains all the
closed subsets of R. In particular, every compact set K is measurable with A(K) = m(K),
where m(K) is as defined in Definition 10.2.10. These by no means exhaust the mea-
surable sets. Any set obtained from a countable union or intersection of open sets or
closed sets, or of sets obtained in this manner, is again measurable.
The collection .,il. is very large. To illustrate just how large, we use the fact that the
Cantor set P has measure zero. Thus any subset of the Cantor set has outer measure
zero, and as a consequence of Theorem 10.3.5, is measurable. By Property 6 of the Can-
tor set (Section 3.3), P has the same cardinality (Definition 1.7.1) as the set of all
sequences of 0's and l's. By Miscellaneous Exercise 7 of Chapter 1, the set of all se-
quences of 0's and l's is equivalent to [0, 1 ], and this set has the same cardinality as all
of R. Thus the set of all subsets of P is equivalent (not equal) to the set of all subsets of
R. As a consequence, in the terminology of equivalence of sets, .At has the same cardi-
nality as the set of all subsets of P. However, nonmeasurable subsets of P do exist. The
construction of such a set will be outlined in the miscellaneous exercises.
We conclude this section by summarizing some of the properties of M.
10.4.7 THEOREM
(a) IfEEA,then E`E.M..
(b) 0, H E A.
( e ) I f E E .M., n = 1 , 2, . , t h en
en
00 00
UEnEAM, and
M-1 n-1
Proof. The result (a) is Corollary 10.4.3, whereas (b) follows from Theorem 10.3.6
and Corollary 10.4.3. The statement (c) is Theorem 10.4.5, and (d) is Theorem 10.3.6.
The proof of (e) follows from Exercise lb of the previous section.
Any collection sd of subsets of a set X satisfying (a), (b), and (c) of the previous
theorem is called a sigma-algebra of sets. The o denotes that the collec-
tion s4 is closed under countable unions.
D EXERCISES 10.4
1. Find a sequence of measurable sets with E, D E, D such that
OAI fl E I + lim
n-'
2. *If E is a measurable subset of I8, prove that given e > 0, there exists an open set U D E and a closed set F C E
such that A(U \ E) < e and A(E \ F) < e.
3. Let E be a bounded subset of R. Prove that E is measurable if and only if given e > 0. there exists an open set U
and a compact set K with K C E C U such that A(U \ K) < e.
4. *If E E, are measurable subsets of [0. I], and if A(E1) = 1. prove that A(E, n E,) = A(E2).
5. Suppose E, is a nonmeasurable subset of [0. I ]. Set E = E, U (l, oo). Prove that E is nonmeasurable but that
A*(E) = A*(E) = oo.
6. *Prove that a subset E of R is measurable if and only if A*(E fl T) + A*(E` fl T) = A*(T) for every subset T of R.
10.5.1 DEFINITION Let f be a real-valued function defined on (a, b]. The function f is said
to be measurable if for every s E 18, the set
{xE[a,b]:f(x)>s}
is measurable. More generally, if E is a measurable subset of I8, a function f : E - R
is measurable if
{xEE:f(x)>s}
is measurable for every s e R.
Since f -'((s, oo)) = {x : f(x) > s}, f is measurable if and only if f -'((s, oo)) is a
measurable set for every s E R. We illustrate the idea of a measurable function with the
following examples.
10.5.2 EXAMPLES
(a) Let A be a measurable subset of I8 and let XA denote the characteristic function of
A. Then
R, s < 0,
{X: XA(x) > s} = A, 0:S s<1,
0, s ? 1.
Since each of the sets ¢, A, and l are measurable. XA is a measurable function on R.
(b) Let f : [0, 1) -+ I8 be defined by
1
[0, 1], ifs < 0,
{x E [0, 1] : f(x) > s} = Q ` fl (s, l), if 0 <_ s < 1,
¢, if s?1.
Again, since each of the sets is a measurable subset of f8, f is measurable.
{x : AX) a s} = 1 ; Ea,
{x : f(x) + g(x) > s} = U ({x : f(x) > ra} fl {x: ra > s - g(x)}).
But each of these two sets are measurable. Thus their union is measurable. Since
10.5.6 DEFINITION A property P is said to hold almost everywhere (abbreviated a.e.) if the
set of points where P does not hold has measure zero, i.e.,
A({x : P does not hold}) = 0.
Remark The assertion that a set is of measure zero includes the assertion that it
is measurable. This, however, is not necessary. If instead we only require that
A*({x : P does not hold}) = 0, then by Theorem 10.3.5 the set {x: P does not hold) is,
in fact, measurable.
We will illustrate the concept of a property holding almost everywhere by means
of the following examples.
10.5.7 EXAMPLES
(a) Suppose f and g are real-valued functions defined on [a, b]. The functions f and g
are said to be equal almost everywhere, denoted f = g a.e., if
{x E [a, b] :f (x) g(x))
has measure zero. For example, if g(x) = I for all x E [0, 11 and
0, if x is irrational,
1(x) =
1 m.
if x = - in lowest terms, x # 0.
n n
As was shown in Example 4.2.2(g), the function f is continuous at every irrational num-
ber in [0, 1], and discontinuous at every rational number in [0, 1 ]. Therefore,
A({x E [0, 1 ] : f is not continuous at x}) = Ap fl [0, 11) = 0.
Thus f is continuous a.e. on [0, 1].
(d) Let f and f, n = 1, 2, ... , be real-valued functions defined on [a, b]. The se-
quence {f.) is said to converge almost everywhere to f, denoted f --+f a.e., if
{x E [a, b] : {f. (x)} does not converge to f (x)}
has measure zero. To illustrate this, consider the sequence (f,,} defined in Example
8.1.2(c) as follows: Let {xk} be an enumeration of 0 n [0, 1 ]. For each n E N, define
f, on [0, 11 by
10.5.8 THEOREM Suppose f and g are real-valued functions defined on a measurable set A.
!f f is measurable and g = f a.e., then g is measurable on A.
Pr oof. Let E = {x E A:g(x) # f(x)}. Then A(E) = 0. Let B = A\ E. Since E is
measurable, so is B. Also on B, g(x) = f(x). Fix s E R. Then
{xEA:g(x)>s}={xEB:g(x)>s}UIx EE:g(x)>s}
= {x E B:f(x) > s} U {x E E: g(x) > s}.
460 Chapter 10 Lebesgue Measure and Integration
Proof. The result follows by Theorem 10.4.5, and the fact that for every s E R.
a
{x:(p(x) > s) = U00 {x: f"(x) > s} and (x: 4,(x) < s} = n-i
U {x: f"(x) < s}.
n-i
Proof. Let E = {x: { f"(x)} does not converge tof(x)}. By hypothesis A(E) = 0. Set
{f"ax),
xEA\E,
gn(x) =
x E E.
Then g" = fn a.e., and thus is measurable. Also, lim gn(x) = g(x) exists for all x E A.
But
g(x) = lim g" (x) = lim gn (x) = inf sup{ fk(x) : k at n}.
n-oo n-"o n
Suppose { fn} is a sequence of measurable functions on [a, b] such that fn --).f a.e.
Then by definition there exists a subset E of [a, b] such that A([a, b]\E) = 0, and
slim fn (x) = f (x) for all x E E. Exercise 15 provides a significant strengthening of
this result. There you will be asked to prove that given e > 0, there exists a measurable
set E C [a, b], such that A([a, b] \ E) < e, and {fn} converges uniformly to f on E. This
result is known as Egorov's theorem.
10.5 Measurable Functions 461
EXERCISES 10.5
1. *Let f be defined on [0, 1) by
0, x=0,
f(x)= x,
:.<'Prove
Show that f is measurable for each n E N. and that (f.1 converges to 0 almost everywhere on [0, I ].
14. *If f is differentiable on [a, b], prove that f' is measurable on [a, b].
462 Chapter 10 Lebesgue Measure and Integration
15. Egorov's Theorem: Let {f} be a sequence of measurable functions on [a. b] such that f. --+f a.e. on [a. b].
Given e > 0. prove that there exists a measurable subset E of [a, h] such that A([a. b] \E) < e and {f} con-
verges uniformly to f on E.
16. Construct a sequence (J,} of measurable functions on [0. 1 ] such that { f (x)} converges for each x E [0, 1 j but
that (f.) does not converge uniformly on any measurable set E C [0. I ] with A([0, I ] \E) = 0.
IDEA.=E.
k=1
Suppose 91 = {x,,, x1, ... , xn} is a point partition of [a, b] as considered in Chap-
ter 6. Set E1 = [x,,, x,], and f o r k = 2, .. . , n, set Ek = (xk_ 1, xk]. Then the collec-
tion 91 = {E,, . . . , En} is a measurable partition of [a, b]. A measurable partition of
[a, b], however, need not consist of intervals. For example, if E, = [a, b] n0 and
E2 = [a, b] \E1, then {E1, E,} is a measurable partition of [a, b].
As for the Riemann integral, if f is a bounded real-valued function on [a, b], and
01 = {E1, . .. , En} is a measurable partition of [a, b], we define the lower and upper
Lebesgue sums off with respect to 91, denoted L(9), f) and ql.L(91, f) respectively, by
n n
where
mk = inf{ f (x) : x E Ek} and Mk = sup { f (x) : x E Ek}.
Clearly, 510, f) : OUL(O, f) for every measurable partition P of [a, b]. As for the
Riemann integral, we could now define the upper and lower Lebesgue integrals off, and
then define a function to be Lebesgue integrable if and only if these two quantities are
equal. The following theorem shows, however, that this is unnecessary.
where the inftmum and supremum are taken over all measurable partitions 0 and of
[a, b], if and only if f is measurable on [a, b].
10.6 The Lebesgue Integral of a Bounded Function 463
Remark. Although the previous theorem is stated for a closed interval [a, b], the re-
sult is also true for f defined on any measurable subset A of l with A (A) < cc.
As a consequence of the previous theorem, which we will shortly prove, we make
the following definition.
10.6.3 DEFINITION If f is a bounded real-valued measurable function on [a, b], the Le-
besgue Integral off over [a, b], denoted f(a bj f dA (or f. f dA), is defined by
t b)
f dA = sup `.EL(9, f),
where the supreemum is taken over all measurable partitions of [a, b]. If A is a measur-
able subset of [a, b], the Lebesgue integral off over A, denoted fA f dA, is defined by
f dA = fXA dA,
A J (a.bj
Remarks
(a) In defining fA f dA, it was implicitly assumed that f is defined on all of [a, b]. If f
is only defined on the measurable set A, A C [a, b], then fXA can still be defined on all
of [a, b] in the obvious manner; namely,
{f(x)'
(fxA)(x) _ x A.
dx.
f bf(x)
a
464 Chapter to Lebesgue Measure and Integration
10.6.5 LEMMA If 91,.q are measurable partitions of [a, b] such that 2 is a refinement of
91, then
-YL(a,f) c qLL(a.f)
As a consequence,
scup `.TL(91,f)
Proof. The proof of the lemma is almost verbatim the proof of Lemma 6.1.3 and The-
orem 6.1.4, and thus is left to the exercises (Exercise 2).
10.6.6 EXAMPLE In this example, we calculate the Lebesgue integral of what is commonly
called a simple function on [a, b]. A simple function on [a, b] is a measurable real-
valued function on [a, b] that assumes only a finite number of values.
Suppose s is a simple function on [a, b] with Range s = {a,, .. . , where
a, * of whenever i 0 j. For each i = I, . . . , n, set
E; = {x E [a, b] : s(x) = a;} = s-'({ail).
Since s is measurable, each E; is a measurable set, and
n++
We will now show that every simple function s is Lebesgue integrable on [a, b] and
compute the Lebesgue integral of s.
10.6 The Lebesgue Integral of a Bounded Function 465
s = 7,, arXE,
where {E;}"= I are pairwise disjoint measurable subsets of [a, b] with U"= I E, = [a, b],
then s is Lebesgue integrable on [a. b] with
s dA aA(E,).
J(u, hl j=1
Proof. To show that s is Lebesgue integrable on [a, b], we will prove that
sup s) = inf 61L.L(9., s),
a .4
where the supremum and infmum are taken over all measurable partitions 9. of [a, b].
Since 9' = {E1..... ER} is a measurable partition of [a, b] and s(x) = a; for all
x E E.
But then
n n
I ai A(Ej) !-__ sup $L(9., s) < inf altl(9 s) :5:- 1 ai A(Ei).
-1 a a ;-1
Therefore, s is Lebesgue integrable on [a, b] with f[, bis dA = I°=1 a;A(E;). 0
Remark Suppose f is a bounded real-valued measurable function on [a, b]. If
91 = {E ,, ... , En} is a measurable partition of [a, b], set
where Mk = inf {f(x) : x E Ek} and Mk = sup{ f (x) : x E Ek}. Then tD and alt are simple
functions on [a, b] with O(x) f(x) s *(x) for all x E [a, b]. Furthermore, by Lemma
10.6.7
J[a. b] OdA = Y,
k-1
MkA(EE) = °LL(9,.f)-
where the supremum is taken over all simple functions ap on [a, b] satisfying V(x) f(x)
for all x E [a, b].
466 Chapter 10 Lebesgue Measure and Integration
Therefore,
L (Mk - mk)A(Ek)
k-1
n \\
[(m A(Ek)
= - J A(Ek) = (b - a).
n
-TL(O'n,f) 5P
Define (p and 11 on [a, b] by v(x) = sup,, (p,,(x) and 11(x) = infra By Theorem
10.5.9, the functions (p and 11 are measurable functions on [a, b], with
(p(x) s AX) < (G(x)
for all x E [a, b].
10.6 The Lebesgue Integral of a Bounded Function 467
To complete the proof we will show that V = 41 a.e. on [a, b]. Then as a conse-
quence of Theorem 10.5.8, the function f will be measurable on [a, b]. Let
E _ {x E [a, b] : (p(x) < i/i(x)},
and for each k E Nl, let
1=I
Combining the above two inequalities gives A(A,,. k) < k/n for all k, n E N. Since
Ek C A,,, k for all n; for each k,
A(Ek) < n
we have A(E) = 0. Thus V = 4 a.e. on [a. b), which proves the result.
468 Chapter io Lebesgue Measure and Integration
If f is Riemann integrable on [a, b], then the upper and lower Riemann integrals off are
equal, and thus
where the supremum and infimum are taken over all measurable partitions a of [a, b].
As a consequence of Theorem 10.6.2, this proves the following result.
f Ax) dx.
u
f dA = A(E) = 1.
Jlo. l i
10.6 The Lebesgue Integral of a Bounded Function 469
10.6.10 THEOREM Suppose f, g are bounded real-valued measurable functions on [a. b). Then
(a) for all a,,8 E R,
Proof. Since the proof of (a) is similar to the proof of the corresponding result for the
Riemann integral we leave it as an exercise (Exercise 4). For the proof of (b), by defi-
nition
f fdA = f f%A,UA, A.
A,UA2 ia.b]
Since A, fl A2 = fXA,UA, = fXA, + fXA2, and the result now follows by (a).
(c) Consider the function h(x) = f(x) - g(x). By hypothesis h ? 0 a.e. on [a, b i.
Let
E, = {x:h(x) >_ 0} and E2 = [a, b]\ E,.
Consider the measurable partition 9' = {Et, E2} of [a, b]. Then
Since h(x) z 0 for all x E E,, m, = inf{h(x):x E E,} > 0. On the other hand, since
h ? 0 a.e., A(E2) = 0. Therefore, fa h dA ? 0. The result now follows by (a).
The result (d) is an immediate consequence of (c), and (e) is left for the exercises.
The measurability of f I follows from Exercise 8 or 10 of the previous section. Q
1 f dA = lim fn dA.
n-4ooJ
[a. bj la. bj
Remark. Although we state and prove the bounded convergence theorem for a closed
and bounded interval [a, b], the conclusion is still valid if the sequence {fn) is defined
on a bounded measurable set A. The necessary modifications to the proof are left to the
exercises.
Proof. Since fn --> f a.e., f is measurable by Corollary 10.5.10, and thus Lebesgue
integrable. Let
E = {x e (a, b): fn(x) does not converge to f(x)).
Define the functions g and g,,, n E N, on [a, b] as follows:
x E (a, b) \ E, g(x) _ f(x), x E [a, b] \ E.
gn(x) = {f(x), and
0, xEE, 0, xEE.
Since A(E) = 0, gn = fn a.e. and g = f a.e. Therefore,
r6 b b b
"dA = fndA and Ja gdA= I fdA.
Ja 1a
Furthermore, gn(x) -+g(x) for all x E [a, b]. Let e > 0 be given. Form E N. set
E. _ {x E [a, b]:Ig(x) - gn(x)I < e forall n >_ m}.
10.6 The Lebesgue Integral of a Bounded Function 471
Here Em = [a, b] \ Em. Thus by Theorem 10.4.6, lim A(E,) = 0. Choose m E N such
that A(Em) < e. Then I g(x) - e for all7n*w m and all x E Em. Therefore,
= jt - g d + t1& - gdA
< e A(E,) + 2M A(Ec) < [b - a + 2M].
Since e > 0 was arbitrary, we have lire fla b]
f dA = fla b, f dA. Q
Combining the bounded convergence theorem with Corollary 10.6.8, we obtain the
bounded convergence theorem for Riemann integrable functions previously stated in
Chapter 8. The theorem does require the additional hypothesis that the limit function f
is Riemann integrable.
fl-
lim
I
a a
f (x) dx.
10.6.12 EXAMPLES
(a) In the first example, we show that the conclusion of the bounded convergence the-
orem is false if the sequence f f,,) is not bounded; that is, there does not exist a finite
constant M such that I f (x)l <_ M for all n E N, and all x E [a, b]. For each n E N,
define f on [0, 11 by
f n, 0<x51n
(x) =
Mx) Sl
0, otherwise.
lim f dA = 1 vk 0= fdA.
n-fM J [0.11 110,1:
(b) As our second example, we consider the sequence { of Example 9.2.3. For each
n E N, write n = 2k +j. where k = 0, 1, 2, .. . , and 0:!:-: _ j < 2k. Define f on [0, 1 ] by
j j+1
2k
`- x `- 2k
.f "(x)
0, otherwise.
The first few of these are as follows: f, = %,o_ :, f,- _ X;0.1/: .f3 - X.1/2. l: f4
X[o,1/41 .. For each n E N, f. E Jt[0, 1 ] with
1 1
dx = 2k
f
J0
Thus lim fo f (x) dx = 0. On the other hand, if x E [0, 1], then the sequence { f (x)}
contains an infinite number of 0's and 1's, and thus does not converge.
EXERCISES 10.6
1. *Let f be a bounded real-valued measurable function on [a, b) with m < f(x) < M for all x E [a, b]. Set
S=M-m.FornENandj=1....,n,letE1={xE[a,b]:m+(j-1)9,<_f(x)<m+jj}.The
Lebesgue sums for f are defined by mJ , (m + (j -1))A(E,). Prove that
lint fdA.
8. Fundamental Theorem of Calculus for the Lebesgue Integral: if f is differentiable on ia. bj and f' is bounded
on [a, h], then f' is Lebesgue integrable, and f,,.,, f' dA = f(b) - f (a).
9. Let.f be a bounded measurable function on [0. 11. Show that x"f(x) is measurable on [0. 11 for n = I. 2.... .
and find lint
f be a bounded measurable function on [a, b] such that f [ < I a.e. on [a, b]. Prove that lim f f " dA = 0.
11. Let be a sequence of nonnegative measurable functions on [a. h' satisfying /;,(.r) - M for all x E :a. b' and
n E N. If f f,) converges to f a.e. on [a. b], prove that
lint f
f e f- A = f fe-f dA.
" b:
12. *If f is a bounded real-valued measurable function on [a. bl, prove that there exists a sequence {s"} of simple
functions on [a, b] such that lint s"(x) = f(r) uniformly on [a, b].
13. Modify the proof of the bounded convergence theorem when the interval [a. b] is replaced by a bounded measur-
able set A.
14. Use Egorov's theorem (Exercise 15. Section 10.51 to provide an alternative proof of the bounded convergence the-
orem.
15. *If p is a simple function on (a. h] and e > 0 is given, prove that there is a step function b on :a. h_ such that
c(x) = h(x) except on a set of measure less than e.
16. Let S"(x) = A0 + Yi-i (Ax cos kx + BA sin kx). If , S"(x) I s M for all .r E [ - ar. a] and n E N. and
f(x) = lint S"(x) exists a.e. on [-a, a], prove that f is measurable and that the A4 and B5 are the Fourier coeffi-
cients off.
jf(t:), iff(x) n ,
f(x) n} =
I. n. if f(x) > n.
474 Chapter 10 Lebesgue Measure and Integration
JAfn
dAK
is monotone increasing, and converges either to a real number or diverges to oo. This
leads us to the following definition.
10.7.1 DEFINITION
(a) Let f be a nonnegative measurable function defined on a bounded measurable
subset A of R. The Lebesgue integral off over A, denoted fA f dt, is defused by
A A A
Jfr,J
A
-+oc
Afl[-n, n]
fdA.
f dAl }nEN
fArj -n, n] J
is also monotone increasing, and thus converges either to a nonnegative real number or
diverges to oo. In the definition we do not exclude the possible value of oo for the inte-
gral off. If the integral is finite, however, we make the following definition.
10.7.3 EXAMPLES
(a) For our first example, we consider the function f(x) = I// defined on (0,1).
Then for each n E IN,
n, 0<x< 1
n-
fn(x) = min{f(x), n} =
1
Therefore,
f fndA= 1
0 0
ndx+
J
1 dx= l
n n
1
n
As a consequence,
I f dA = lim 1 f dA = liml 2 - 1) = 2.
The answer in this example corresponds to the improper Riemann integral of the func
tion f. This will always be the case for nonnegative functions for which the improper
Riemann integral exists (Exercise 18).
(b) Let g(x) = 1/x, 0 < x K 1. For the function g,
n, 0<x< 1,n
min{g(x), n} = 1
x'
-<x<
n
1
1.
Therefore,
1 I/n t 1
J min{g, n} dA = I n dx + J - dx = 1 + Inn.
0 0 1/fix
Thus
Jgdlt=lim(1
+Op
+ In n) = oo.
Therefore,
I` fdA=lim1-n=1.
I
JA, UA.
fdA =
1A,
fdl + j fdA.
A,
Proof. We will indicate the method of proof by proving part of (a). The remaining
proofs are left to the exercises (Exercise 2). Suppose first that the set A is bounded. Let
h = f + g. Since
min{ f (x) + g(x), n} 5 min{ f (x), n) + min{g(x), n} <_ min{ f (z) + g(x), 2n},
!g,, dA _ Ih2.dA.
fAh- dAs
A A A
mHf,, dA + fgndA=
A
) 1m
A
jf dA + Aliimm f g dA = JfdA+ JIdA.
A A A A
Therefore, since
f
lim h dA = lim J h2i dA = j(f + g) dA,
n-.Wf
A A A
10.7 The General Lebesgue Integral 477
the result follows. If one or both of the sequences { fA fn dA}, { f, gn dA} diverges to x,
then so does their sum. In this case, we obtain fA(f + g)dA = oo. If A is unbounded.
then by the above for each n E N,
j (f + g)dA = J
An;-n.._
fdA +
JAnl-n.r,
gdA,
Ant-n.n)
Fatou's Lemma
Our first major convergence theorem for integrals of nonnegative measurable functions
is the following result of Fatou.
Proof. Suppose first that the set A is bounded. For each k E N, let
hn(x) = min{ fn(x), k} and h(x) = min{ f (x), k}.
Then for each k E N, the sequence {hn} converges a.e. to It on A. Since I hn(x) 1 s k for
all x E A, by the bounded convergence theorem,
10.7.6 EXAMPLE In this example we show that equality need not hold in Fatou's lemma.
Consider the sequence ( fn} on [0, 1 ] of Example 10.6.12(a). For each n E N, fn(x) = n
if 0 < x s 1, and fn(x) = 0 elsewhere. This sequence satisfies
0= f (lim
nix fn) dA < I =n-.oo
lim J f dA.
f 0, I j 0, l1
Remark. Fatou's lemma is often used to prove that the limit function f of a conver-
gent sequence of nonnegative Lebesgue integrable functions is Lebesgue integrable.
For if lim fA fn dA < oo and if fn -+f a.e. on A with fn ? 0 a.e. for all n, then by Fa-
tou's lemma, fA f dA < oo. Thus f is integrable on A.
jfdA = - jrdA.
A A A
The only problem with this definition is that it is possible that fAf' dA = fA f- dA = co,
giving the undefined oo - oo in the above. However, if we assume that both f' and f-
are integrable on A, then the above definition makes sense. Furthermore, if f is measur-
able, and f' and f- are both integrable on A, then I f j is also integrable on A. Con-
versely, if f is measurable and f I is integrable on A, then since f' s If I and f - s if 1,
by Theorem 10.7.4(c) both f' and f - are integrable on A. Therefore, we make the fol-
lowing definition.
f f dA = f f+ dA - jf- dA.
A A A
10.7 The General Lebesgue Integral 479
Remark. The set Z(A) of Lebesgue integrable functions on A is often also denoted
by -T'(A).
The definition of the general Lebesgue integral is consistent with our definition of
the Lebesgue integral of a bounded function on [a, b'. If f is a bounded real-valued
measurable function on [a, b] then so are the functions f' and f-. Let E, =
{x: f(x) a 0} and E2 = {x: f(x) < 0}. Then E, and E2 are disjoint measurable subsets
of [a, b] with E, U E2 = [a, b]. Furthermore, AE, = f" and fXE, = -f -. By Theo-
rem 10.6.10(b)
f
jfdA = j fdA + jfdA
fb f bb b (b
fXE, dA + dA = j f + dA - 1 f - dA.
a a a a
Remark For a nonnegative measurable function, the Lebesgue integral and the improper
Riemann integral are the same, provided of course that the latter exists (Exercise 18). This,
however, is false for functions that are not nonnegative. For example, consider the function
f(x) = (sinx)/x, x E [7r,oo), of Example 6.4.4(b). By Exercise 7 of Section 6.4, the im-
proper integral off exists on [-rr, oo). However, as was shown in Example 6.4.4,
j
tn
sinxl
IfIdA= lintj "I x dx=00.
it
Thus f is not Lebesgue integrable on [ir,oo). Another such example for a finite interval
is given in Exercise 23. The crucial fact to remember is that a measurable function f is
Lebesgue integrable on a measurable set A if and only if IA If I dA < oo.
Our first result is the following extension of Theorem 10.7.4 to the class of inte-
grable functions.
10.7.8 THEOREM Suppose f and g are Lebesgue integrable functions on the measurable
set A. Then
(a) f + g a nd cf, c E R, are integrable on A with
= J fdA+ 1 fdA.
JAU! A, A;
Proof. (a) The proof that cf is integrable and that fA cf dA = cfA fdA follows im-
mediately from the definition. Before proving the result about the sum, we first note that
the definition of the integral off on A is independent of the decomposition f = f - - f -.
Suppose that f = f, - f2 where f, and f, are nonnegative integrable functions on A.
Then
If f and g are integrable on A, then by definition so are the functions f' + g` and
f + g-. Since f + g = (f+ + g+) - (f- + g-), by the above
= ji dA - ji- dA + jg dA - f g' dA
A A A A
= f fdA + jgdA.
A A
from which the result follows. The proof of (c) also follows from (a) and the fact that
since A, n At = b, fXA,UA, -- f XA, +.fXA;
10.7 The General Lebesgue Integral 481
jfdA I'mJAfndA.
=
Proof. Since g is integrable on A, and I fn 15 g a.e. on A, by Theorem 10.7.4 each fn
is also integrable on A. Also, by Corollary 10.5.10, the function f is measurable on A.
Furthermore, by Fatou's lemma,
AA
jIfId!jIfnIdA s
Thus f is also integrable on A.
jgdA < oo.
A
.By redefining all the f,,, n E N. on a set of measure zero if necessary, we can with-
out loss of generality assume that If. (x) I <_ g(x) for all x E A. Consider the sequence
(g + fn}nEN of nonnegative measurable functions on A. By Fatou's lemma,
J (g + /) dA j(g+fn)dA
A = I A
= f gdA + lim
nix A
f fn A.
A
Therefore,
f f dA <_ lim f
n~x A
A.
A
lim f fn dA.
f f dA > n-.x
A A
Remark. The hypothesis that there exists an integrable function g satisfying I fn I !f- g
a.e. is required in the proof in order to subtract f g dA in the above inequalities. This is
not possible if f gdA = oo. As the following example shows, if such a function g does
not exist, then the conclusion may be false.
10.7.10 EXAMPLE As in Example 10.6.12(a) consider the sequence { fn} on [0, 1] defined by
In, 0 <x< 1
fn(x) = n'
10, elsewhere.
For the sequence { fn} we have lim fn (x) = 0 for all x E [0,1 ] but f o If,, dA = 1 for
all n. We now show that any measurable function g satisfying g(i) ? fn (x) for all
X F: [0,I] and n E N satisfies f o 11gdA = oo. Since g(x) _' fn(x) f o r all x E (0,I]
we have g(x) ? n for all x E (n:11 ,1-, ]. Since the collection {(kk + 11, k]}k=t of intervals is
pairwise disjoint, by Theorem 10.7.4,
n
EXERCISES 10.7
1. Let f be a nonnegative measurable function on a measurable set A. If f4 f dA = 0. prove that f = 0 a.e. on A.
2. *a. Prove Theorem 10.7.4(b).
b. Prove Theorem 10.7.4(c).
3. Let A be a measurable subset of R.
a. If f is integrable on A and g is bounded and measurable on A, prove thatfg is integrable on A.
b. If f and g are integrable on A, is the function fg integrable on A?
4. *Let f,(x) = x-°, x E (0, 1). Prove that f, is integrable on (0, 1) for all p, 0 < p < 1, and that
1
(Q,,
J(-' 1)
1 --p'
1 p.
5. Define f on [ 1, oo) by
f(x) _ { V, if x E [n, n + l ln'"), n = 1, 2, .. ,
0, otherwise.
Show that f E X([ 1, oo)) but f2 `.P([ 1, oo)).
6. Let P denote the Cantor set of Section 3.3. Define f on [0, I ] as follows: f(x) = 0 for every x E P. and f(x) = k
for each x in the open interval of length 1/3k on [0, 1] 1 P. Prove that f is integrable on [0, 1 ] and that fo fdA = 3.
7. *Let f be a nonnegative integrable function on [a, b]. For each n E N, let E. = {x: n s f(x) < it + 1 }. Prove that
7, , nA(E,) < oo.
8. Evaluate( each of the following limits:
a, lim J (I - dx b. lim
n-.oo
(1 - x/n)nex"2 dx
1
(0. 1) L. W
10.7 The General Lebesgue Integral 483
9. Let f be an integrable function on (a, b]. Given e > 0, prove that there exists a bounded measurable function g
on [a, b] such that fla bl If - g I dA < e.
10. *Suppose f is integrable on a measurable set A with A (A) = oo. Given e > 0, prove that there exists a bounded
measurable set E C A such that LIE I f I dA < E.
11. Show by example that Fatou's lemma is false if the functions f,,, n E N, are not nonnegative.
12. Show by example that the bounded convergence theorem is false for a measurable set A with A (A) = oo.
13. *Let f be a Lebesgue integrable function on a measurable set A. Prove that given e > 0, there exists a 8 > 0 such
that fE If I dA < e for all measurable subsets E of A with A (E) < S.
14. Let f be amr integrable function on (a, b), where -oo <_ a < b s oo. Define F on (a, b) by
= 7, jfdA.
00
n"I A,
b, Prove that the conclusion of part (a) is still valid for arbitrary f e T(A).
22. Let { f,} be a sequence of measurable functions on (a, b) satisfying I f,(x)I s g(x) a.e., where g is integrable on
[a, b]. If liimf,(x) = f(x) exist a.e. on [a, b], and h is any bounded measurable function on [a, b], prove that
(b
fh dA = lim f f,h dA.
1
111
a
23. *As in Exercise 4, Section 6.4, let f be defined on (0, 1) by
10.8.1 DEFINITION Let A be a measurable subset of R. We denote by T2(A) the set of all
measurable functions f on A for which I f I2 is integrable on A. For f E f2(A), set
IIfII2=\JaIfI2dA/
The quantity IIf II2 is called the 2-norm or norm off. Clearly. IIf II2 >_ 0, and from
the definition it follows that if f E _T22(A) and c E 18, then cf E Y2(A) with II cfII2 =
IcIIIf1I2. We will shortly prove that if f, g E `.P2(A), then f + g E Y2(A) with
Ill + 8 112 < IVII2 + 119 112- Thus .T2(A) is a vector space over R.
If f E 22(A) satisfies IIf II2 = 0, then by Exercise 1, Section 10.7, 1 f 122 = 0 a.e., and
thus f = 0 a.e. on A. This does not mean that f (x) = 0 for all x E A; only that f = 0
except on a set of measure zero. Thus II 112 satisfies all the properties of a norm except
for IIf II2 = 0.if and only if f = 0. To get around this difficulty we will consider any two
functions f and g in X2 (A) for which f = g a.e. as representing the same function. For-
mally, we define two measurable functions f and g on A to be equivalent if f = g a.e. In
this way it is possible to define 22(A) as the set of equivalence classes of square inte-
grable functions on A. Rather than proceeding in this formal fashion, we will take the
customary approach of simply saying that two functions in f22 are equal if and only if
they are equal almost everywhere. With this definition Y2(A) is a normed linear space.
10.8.2 EXAMPLES
(a) For our first example, let f (x) = 1/Vc, x E (0, 1). By Exercise 4 of the previous
section, f is integrable on (0, 1) with 1(0 1)f dA = 2. Since f 2(x) = 1/x, by Example
10.7.3(b), f2 is not integrable on (0, 1) and thus f `12((0, 1)). On the other hand, if
g(x) = x- 'I, then g2(x) = x-21-', which by Exercise 4 of the previous section is inte-
grable. Thus g E `12((0, 1)) with
IIBII2 J x 2/'dx=3.
0
(b) Consider the function f (x) = 1/x for x E [ 1, oo). For any n E N, n ? 2,
J IfI2dA = J x2dx = rI
L i
- 1J.
n
Thus
Therefore, f E 12([ 1, oo)) with 11f112 = 1. It is easily shown that f E1([ 1, x)).
10.8 Square Integrable Functions 485
Cauchy-Schwarz Inequality
Our first result will be the analogue of the Cauchy-Schwarz inequality for 12. The fol-
lowing inequality is sometimes also referred to as Holder's inequality.
j
A
1181 dA s 111 1I211g112
Proof. By Theorem 10.5.4, the productfg is measurable, and for any x E A, we have
I f(x)g(x) 1:S
1 (I f(x) 12 + Ig(x)12).
2
Since by hypothesis f, g E `.E2(A), the function I fgI is integrable on A, and thusfg is in-
tegrable on A. As in the proof of Theorem 7.4.3, for y E R,
0s f (IfI -yIBI)2dA=IIf
A
j lfgI dA.
A
(9)
(fAIfgI dA)2
0 111112 -
IIg112
and thus (fA I fg I dA)2 11f 112118112, which proves the result. Q
Our next result is Minkowski's inequality for the space X22. Since the proof of this
is identical to the proof of Theorem 7.4.5, we leave the details to the exercises.
111+g112s111112+118112
Proof. Exercise 4. Q
Properties (a) and (c) follow from the definition, and (d) is Minkowski's inequality.
With the convention that two functions f and g are equal if and only if f = g a.e.. T2(A)
is a normed linear space.
By Definition 8.3.9, a sequence {/n} in ce2 converges in norm to a function f in `.EZ
if and only if Jim Of - fn112 = 0. Example 10.6.12(b) shows that convergence in Y2 does
not imply pointwise convergence of the sequence. As in Chapter 9, convergence in S2
is usually called convergence in the mean. We now prove that T22([a. b]) is a complete
normed linear space.
10.8.5 THEOREM The normed linear space (.T2([a, b]), 11 A2) is complete.
Remark. Although we state and prove the result for a closed and bounded interval.
the same method of proof will work for T2(A) where A is any measurable subset of R.
Before proving the theorem, we first state and prove the following lemma.
and for some finite constant C. Then f (x) = lim is finite a.e. on A.
Proof. Since { f (x)} is monotone increasing for each x E A, the sequence either con-
verges to a real number or diverges to oo. Let f(x) = liymeOfn(x). and let
E_ {xEA:f(x) = oo}.
We will prove that A(E) = 0. For each k E N, let
EE= {xEA:f(x)>k}.
Then Ek D Ek+, for all k E tN with lkE Ek = E. For fixed k E IN, set
An, k= {x E A: fn(x) > k}, n E N.
Then A..k C An+I.k With U. EN A..k = Ek. Thus by Theorem 10.4.6(a),
A(E)= limA(EE)=0.
Therefore, f is finite a.e. on A. 0
10.8 Square Integrable Functions 487
Proof of Theorem 10.8.5. Let {fn} be a Cauchy sequence in 22. Then given
e > 0, there exists no E Ni such that Ilfn - fn II2 < e for all n, m ? no. For each k E NI,
let nk be the smallest integer such that Ilfn - fn 112 < 1/2k for all m, n ? nk. Then
n1cn2<...
CnkG...,and
IIJn,., ll2 <
2k.
Thus the sequence {gk} satisfies the hypothesis of Lemma 10.8.6. Therefore, urn k is
finite a.e. on [a, b]. But then lim gk is also finite a.e. on [a, b]. As a consequence, the
k-+oo
series
00
But the kth partial sum of this series is Therefore, the sequence {fn,}kEh con-
verges a.e. on [a, b]. Let E denote the set of x E [a, b] for which this sequence con-
verges. Then A([a, b) \ E) = 0. Define
f(),x=- lim fn,(x)' x E E,
J 0, otherwise.
Then {f,} converges to f ae. on. [a, b].
It remains to be shown that f E 22, and that { fn} converges to f in 2 2. Since
IfnIf ..I
+
k
j=1
p
Ifnf., -fn,l = 8k
by Fatou's lemma
iab]
(f n'.' - f., 11 < \2
(a, b]
Therefore, II f - fn,112 < 1/2k for all k E N. Thus the subsequence { f,,,}kEN converges to
f in the norm of _y2. Finally, by the triangle inequality,
From this it now follows that the original sequence { fn} also converges to f in the norm
of 22.
Fourier Series
We close this chapter by making a few observations about Fourier series and the space
22([-ir, ir]). As in Definition 9.3.1, if f is Lebesgue integrable on [-a, ir]. the
Fourier coefficients of f with respect to the orthogonal system {1, cos nx, sin nx}.,
are given by
A
Since f is Lebesgue integrable, the functions f(x) sin nx and f(x) cos nx are measurable
on [-zr, -rr], and in absolute value less than or equal to I f(x)I. Thus the functions
f(x) cos nx and f(x) sin nx are all integrable on [ -ir, a].
The same method used in proving Bessel's inequality for Riemann integrable func-
tions proves the following (Exercise 8): If f E :E2([-ar, ir]) and {ak} and {bk} are the
Fourier coefficients off, then
00 i 7r
Thus the sequences {ak}k o and {bk}' , are square summable. We now use complete-
ness of the space 22([-ir, it)) to prove the converse.
10.8.7 THEOREM If {ak}k=o and {bk}k , are any sequences of real numbers satisfying
1 00
2 ap + (ak + bk) < oo,
k=I
then there exists f E `.E2([-tr, rr]) whose Fourier coefficients are precisely {ak} and
{bk}.
10.8 Square Integrable Functions 489
which by orthogonality
= jr j, (a + b2E).
Since the series converges, the sequence {Sn] is a Cauchy sequence in `.C2([-Tr, a i).
Thus by Theorem 10.8.5, there exists a function f E X2([-zr, ar]) such that Sn con-
verges to f in 22; i.e., lim 11f - S,,92 = 0. If n > m, then
A A
1
am - jf(x)cosmxdr.
-A
A similar argument proves that the b,n are the sine coefficients off. O
classical result is very useful in answering this question. Since the proof of the theorem
is beyond the level of this text, we state the result without proof.2
10.8.8 THEOREM
EXERCISES 10.8
1. *For x E (0, 1) let fo (x) = x'°, p > 0. Determine all values of p such that f, E Z2((0, I )).
2. For each n E N, let 1, = (n, n + =,). For a given sequence {c,} of real numbers, define f on 11. oo) by
f(x) = 7,W , c, Xtn (x). Show that f E 22([ 1. oo)) if and only if
00c2
a < oo.
n.l n
3. Find an example of a real-valued function f on (0, oo) such that f2 is integrable on (0, oo). but If V° is not in-
tegrable on (0, oo) for any p, 0 < p < oo, p * 2.
6. Let { fa} be a sequence in 22([a. b]). Suppose {fa} converges in 22 to f E T2 and {fa} converges a.e. to some
measurable function g. Prove that f = g a.e. on [a, b].
7. 'Let If.) be a sequence in 22(A) that converges in cy2 to a function f r= If g E 222(A). prove that
lim
a-0D J
fag dA = jfdA.
A A
& If f E T2([-ir, ar]) and {ak} and {bk} are the Fourier coefficients of f, prove that
I
ao+ I(at+b})s-ar-° f2dA.
O° 1
2 k°1
NOTES
Lebesgue's development of the theory of measure and in- convergence of the sequence (f.) is required. Otherwise,
tegration was one of the great mathematical achievements the limit function may not be Riemann integrable. On the
of the twentieth century. His proof that every bounded other hand, if { fa} is a sequence of measurable functions
measurable function is Lebesgue integrable was based on on [a, b], then its pointwise limit f is also measurable.
the new idea of partitioning the range of a function, rather Hence if f is also bounded, then f is integrable. The
than its domain. Lebesgue's theory of integration also bounded convergence theorem is notable for its simplicity
permitted him to provide necessary and sufficient condi- of hypotheses and proof. It only requires that (f.) be uni-
tions for Riemann integrability of a bounded function f. formly bounded and converge a.e. on [a, b]. This is suffi-
In addition to the fact that the Lebesgue integral en- cient to ensure that
larged the family of integrable functions, its power results
from the ease with which it handles the interchange of f, dA = (lim
limits and integration. For the Riemann integral, uniform n
Urn
-.m
a, b] o. bJ
.x f.) A.
492 Chapter 10 Lebesgue Measure and Integration
With the additional hypothesis that the pointwise limit f is function is differentiable a.e. on [a, b).3 A slight general-
Riemann integrable, the bounded convergence theorem is ization of Theorem A then gives that
also applicable to a sequence if,) of Riemann integrable
functions.
F'dA = F(x) - F(a) = J fdA,
The bounded convergence theorem and the dominated
convergence theorem are the tools required to prove the
fundamental theorem of calculus for the Lebesgue integral. or that f.[ F' - f] dA = 0 for all x E [a. b]. As a conse-
quence of Miscellaneous Exercise 5 we have F' = f a.e.
THEOREM A 1ff is differentiable and f is bounded Newton and Leibniz realized the inverse relationship of
on [a, b], then f is Lebesgue integrable, and differentiation and integration. The above two versions of
fa f'dA = f(b) - f(a). the fundamental theorem of calculus provide a rigorous
formulation of this inverse relationship for a large class of
The proof of this result was requested in Exercise 8 of functions.
Section 10.6. It follows simply by applying the bounded The Lebesgue theory of integration also provides the
convergence theorem to the sequence defined by proper setting for the study of Fourier series. The bounded
g (x) = n[f(x + '-,) - f(x)]. Since f is differentiable, the convergence theorem was used by Lebesgue to prove uni-
sequence converges pointwise to f. Also, by the mean queness of a Fourier series. If
value theorem the sequence is uniformly bounded on
[a, b]. This then establishes the analogue of Theorem 6.3.2
for the Lebesgue integral. If instead of bounded, one as- f(x)=2Ao+ k-l (Ak cos kx + 8k sin kx).
sumes that f is Lebesgue integrable, then the result also
follows by Lebesgue's dominated convergence theorem.
for all x E [ -it. zr], then f, being the pointwise limit of a
The Riemann theory of integration allows us to
sequence of continuous functions, is automatically mea-
prove that if f E gt[a, b], and F is defined by F(x)
surable on [-rr, a]. If f is also bounded, then f is inte-
f.'f(t) dt, then F'(x) = f(x) at any x e [a, h] at which f
grable. If the sequence of partial sums of the trigono-
is continuous. Since f E 9l[a, b] if and only if f is con-
metric series is uniformly bounded, then by the bounded
tinuous a.e. on [a, b], we have that F'(x) = f(x) a.e. on
convergence theorem, the trigonometric series is the Fourier
[a, b]. Although not proved in the text, this result is still
series off. In his 1903 paper Sur les series trigonometric,'
valid for the Lebesgue integral.
Lebesgue showed that uniform boundedness of the partial
sums may be removed; that boundedness of the function f
THEOREM B Let f be Lebesgue integrable on (a, b),
itself was sufficient. This result was extended in 1912 by
and define F(x) = f, f dA. Then F'(x) = f(x) a. e. (a, b). de la VallEe-Poussins to the case where the function f is in-
By writing f = f * - f -, it suffices to assume that tegrable on (- rr, a ). See the article by Alan Gluchoff for
f 0. and thus F is monotone increasing on [a, b). It is an overview of how trigonometric series has influenced the
a fact independent of integration that every monotone various theories of integration.
MISCELLANEOUS EXERCISES
1. Let A be a measurable subset of R. For f E 2(A) set lI f 11, = f,, I f I dA. Prove that (.T(A ), (I (I,) is a complete
normed linear space.
2. Let A be a measurable subset of R, and let f, f, n = 1, 2, ... be measurable functions on A. The sequence If.) is
said to converge in measure to f if for every 6 > 0.
lim A({x E f(x)l 6}) = 0.
a-m
a. If is a sequence in :E(A), and converges in the norm of .Y(A) to f E `.E(A). prove that converges in
measure to f.
b. Find a sequence ff.) of measurable functions on a measurable set A that converges to a function fin measure.
but does not converge to f in norm.
c. If A(A) is finite and (f.} is a sequence of measurable functions that converges in measure to a measurable func-
tion f, prove that there exists a subsequence { fk} of {f} such that fk -+f a.e. on A.
d. Show that the sequence { f } of Example 10.6.12(b) converges to 0 in measure.
e. Find a sequence {f.) of measurable functions on a measurable set A such that f -a 0 everywhere on A but {
does not converge to 0 in measure.
3. Let be a sequence of orthogonal functions on [a, b] having the property that the zero function is the only con-
tinuous real-valued function f satisfying fa fkp, dA = 0 for all n E N. Prove that the system is complete.
(Hint: First use the hypothesis to prove that if f 2([a, b]) satisfies Ja' dA = 0 for all n E N. then f = 0
a.e. Next use completeness of `. b2 to prove that Parseval's equality holds for every f E `f ([a, b]).)
4. Construction of a Noumeasurable Set: For each x E [-12, Z], define the set K(x) by
K(x)={yE[-12,Z]:y-xE0}.
a. Prove that for any x, y E [ -s, ;], either K(x) n K(v) = 4 or K(.r) = KO,). (Note: K(x) = K(y) does not imply
that x = y; it only implies that x - y is rational.)
Consider the family & = {K(x):x E [ -12,121} of disjoint subsets of J. Choose one point from each
distinct set in this family and let A denote the set of points selected. The ability to choose such a point from
each of the disjoint sets requires an axiom from set theory known as the axiom of choice. Further information
about this very important axiom can be found in the text by Halmos.
Let rk, k = 0, 1, 2, ... be an enumeration of the rational numbers in [ -1. I ], with r1 = 0. For each
k=0,1,2..., setAkA+rk.
b. Show that the collection {Ak} is pairwise disjoint with [- i, ] CU u Ak C [- _, ;
c. Use the above to show that A*(A) = 0 and A*(A) > 0, thus proving that A is nonmeasurable.
S. Suppose f is Lebesgue integrable on [a, b]. If f. f dA = 0 for every x E [a, b], prove that f = 0 a.e.
SUPPLEMENTAL READING
Botts, Truman, "Probability theory and the Lebesgue in- Katznelson, Y. and Stromberg, K., "Everywhere differ-
tegral," Math. Mag. 42 (1969). 105-1 I1. entiable, nowhere monotone function;' Amer. Math.
Burkill, H., `"The periods of a periodic function;' Math Monthly 81(1974), 349-354.
Mag. 47 (1974),206-210. Kraft, R. L., "What's the difference between Cantor
Darst, R. B., "Some Cantor sets and Cantor functions;' sets?" Amer. Math. Monthly 101 (1994). 640-650.
Math. Mag. 45 (1972). 2-7. Ladder, G., "A simple introduction to integral equa-
Dressler, R. E. and Stromberg, K. R., '"The Tonelli inte- tions," Math. Mag. 69 (1996), 172-181.
gral," Amer. Math. Monthly 81 (1974), 67-68. Maligranda, L., "A simple proof of the Holder and
Gluchoff, Alan D.. 'Trigonometric series and theories of Minkowski inequality," Amer. Math. Monthly 102
integration;' Math. Mag. 67 (1994), 3-20. (1995), 256-259.
494 Chapter 10 Lebesgue Measure and Integration
McShane, E. J., "A unified theory of integration;' Amer. Varberg, Dale E., "On absolutely continuous functions.
Math. Monthly 80 (1973), 349-359. Amer. Math. Monrhl' 72 (1965). 831-841.
Priestly, W. M., "Sets thick and thin;' Amer. Math. Wade. W. R., 'The bounded convergence theorem;'
Monthly 83 (1976), 648-650. Amer. Math. Monthly 81 (1974). 387-389.
Appendix:
Logic and Proofs
A.1 Propositions and Connectives
A.2 Rules of Inference
A.3 Mathematical Proofs
A.4 Use of Quantifiers
In searching for new mathematics, mathematicians use their imagination and ingenuity
to make a conjecture about new theories based on observations of particular cases or
phenomena. This approach is called inductive reasoning. Natural or social scientists
may test the conjecture by making further observations. If the results are incompatible
with the conjecture, then scientists must usually reject or modify the theory. Mathe-
maticians, on the other hand, attempt to change the conjecture into a theorem by prov-
ing (or disproving) that the conjecture follows logically from the accepted axioms, the
hypothesis, and proved theorems of other mathematicians. In constructing a proof
mathematicians are restricted to the use of deductive reasoning-the use of logic to
draw conclusions based on hypothesis and statements accepted as true.
In developing new mathematics, mathematicians have the advantage of not being
restricted to the study of observable phenomena or even axioms that have been accepted
as true. Euclid's fifth postulate (the parallel axiom) had been accepted as true for over
2200 years until the early nineteenth century when Janbs Bolyai (1802-1860), Carl
Friederich Gauss (1777-1855), and Nicolai Lobatchewsky (1793-1856) challenged
this axiom. The acceptance of alternatives to the parallel postulate led to the develop-
ment of new and interesting non-euclidean geometries, many of which are still very im-
portant today.
Deductive reasoning provides mathematics with an efficient means of both exposi-
tion and organization of its subject matter. In what follows we will attempt to explain
this formal, logical side of mathematics. Formal logic will not indicate the steps neces-
sary to prove a conjecture; it only provides a system of rules and procedures used to
495
496 Appendix Logic and Proofs
decide whether a given statement follows from some given set of statements. Intuition.
insight, and knowledge are still required to formulate the necessary steps in a proof.
P Q -P PAQ PVQ
T T F T T
T F F F T
F T T F T
F F T F F
The connectives "and," "or;" and "not" are illustrated by the following examples.
Using (a), (b), and (c) above we have
-- (b) "V is not rational."
(a) A (b) "4 is an even number and V is rational,"
A.1 Propositions and Connectives 497
(a) V (c) "4 is an even number or the author of this text was born in Frankfurt, Ger-
many."
Since the statement "V is rational" is false, the proposition (a) A (b) is also false. On
the other hand, since (a) is true. (a) V (c) is true regardless of where the author was
born.
In logic the use of the word "or" is the inclusive or, meaning "one or the other, or
both" as opposed to the exclusive or, meaning "one or the other, but not both" For
propositions P and Q the exclusive or can be expressed in terms of A, V, and by
(P V Q) A -- (P A Q) (see Exercise 2). In addition to and, other common connectives
used in English are but, while, and although. Each of these has the same meaning as
and, and is represented symbolically by A. For example, if P is "It is cold" and Q is "It
is snowing;" then the statement "It is cold, but it is not snowing" would be expressed
symbolically by P A - Q.
In addition to the connectives A, V, and -, there are two other connectives that are
commonly used in logic; namely, the conditional and biconditional.
P Q P=*Q PaQ
T T T T
T F F F
F T T F
F . F T T
In the truth table for P = Q, the only line where P Q is true and P is true is the
first line, where Q is also true. Thus the conditional statement P Q is often also ex-
pressed by saying that P is a sufficient condition for Q (if P is true then Q follows), or
that Q is a necessary condition for P (P cannot be true unless Q is true). That Q is a
necessary condition for P is sometimes also expressed by the phrase "P. only if Q"
Since the biconditional P,4* Q is true exactly when P and Q have the same truth val-
ues, this is often also verbally expressed by "P is necessary and sufficient for Q:'
To illustrate the truth values assigned in the conditional sentence, let us consider
the following example. Your professor agrees
"If you earn an A on the final, I will assign you an A for the course."
498 Appendix Logic and Proofs
Here the antecedent P is "You earn an A on the final" and the consequent Q is "He as-
signs you an A for the course" The only case in which you have reason to be angry (the
sentence is false) is when P is true and Q is false. If both P and Q are false, you may
not be happy, but you have no cause to be angry with your professor. On the other hand,
if P is false and Q is true, you certainly will not be angry.
Closely related to the conditional sentence are the converse and the contrapositive
of P =*. Q.
A.l.4 DEFINITION For propositions P and Q, the converse of P . Q is Q =o. P, and the
contrapositive of P . Q is --Q' --P.
The truth values for each of these is given in the following table.
P Q P=Q -PvQ
T T T T
T F F F
F T T T
F F T T
Also, from the table for the contrapositive we see that -Q -P is equivalent to
P Q. To emphasize this we state it as a theorem.
have the value T for any assignment of T or F to P and Q. Such propositions are called
tautologies. A tautology is a propositional formula that is true for every assignment of
truth value to its components. A contradiction is the negation of a tautology. Thus
(P V -P) is a contradiction. By means of a truth table it is easily verified that
-- (P V -P) is equivalent to -P A P, which simply states that "P" and "not P" can-
not both be true simultaneously. This is sometimes referred to as the law of the excluded
middle. Exercise 3 contains several basic tautologies that will be very useful in our dis-
cussion on rules of inference.
From the definitions of equivalence of propositional formulas and the biconditional
it should be clear that two propositional formulas P and Q are equivalent if and only if
P q Q is a tautology. To emphasize its importance we state it as a theorem.
A.1.6 THEOREM Two propositional formulas P and Q are equivalent if and only if P q Q
is a tautology.
To illustrate this we consider the equivalent propositional formulas P Q and
-P V Q. The truth table for (P Q) a (-P V Q) is as follows:
P Q (P * Q) q (--P V Q)
T T T T T
T F F T F
F T T T T
F F T T T
EXERCISES Al
,
1. For propositions P. Q, and R, verify by means of a truth table that each of the following pairs of propositional for-
mulas are equivalent.
a. ^(-P) and P b. P A Q and Q A P; P V Q and Q V P
c. PA (QVR)and(PA Q)V (PAR) d. P V (Q A R)and(P V Q) A (Q V R)
e. ^-(P A Q) and -P V --Q. f. ^-(P V Q) and ^-P A --Q
S. P4* Qand(P-r Q) A (Q=P)
Parts (c) and (d) above are referred to as the distributive laws for A and V, whereas parts (e)
and (f) are De Morgan's laws for A and V.
2. By means of a truth table show that (P V Q) A -- (P A Q) is true if either P or Q is true, and false otherwise.
3. Prove that each of the following is a tautology
a. P v -P (Excluded Middle)
b -(-P) q P (Double Negative)
c. P a [-P (R A --R)] (Contradiction)
d. (P A Q) - P (Conjunctive Simplification)
e. P (P V Q) (Disjunction)
500 Appendix Logic and Proofs
P1....,P. :. Q, or as
P.
Q,
Here the formulas P,, ... , P are the premises and Q is the conclusion. (The symbol
:. is used in mathematics to denote therefore.) We even allow the case when no
premises are present. In that case, Q can be regarded as an axiom. The most important
thing about a rule of inference is that it should be logically valid; that is,
P, A . . A P. = Q (or just Q if no premises are given)
should be a tautology. For example, to show that P, P Q :. Q is logically valid, it
suffices to show that (P A (P Q)) Q is a. tautology. This is easily verified by
means of a truth table as follows.
P Q P=* Q P A (P=Q)=:;- Q
T T T T T
T F F F T
F T T F T
F F T F T
P=*Qa-Q=*--P (Contrapositive)
Pa[--P==O-(R A-R)] (Contradiction)
One of the most fundamental rules of inference, in fact usually taken as an axiom,
is the following.
This rule simply states that the propositional formula P V --P can be inferred, for any
proposition P. For example, if an argument involves a real number x, one can assert at
any time that "Either x = 0 or x # 0."
Rule of Conjunction:
From P and Q, one can infer P A Q.
Rule of Disjunction:
From P, one can infer P v Q.
Modus Ponens Rule or Rule of Detachment:
From P and P Q, one can infer Q.
Modus Tollens Rule or Rule of Contrapositive Inference:
From -Q and P = Q, one can infer -P.
A.2 Rules of Inference 503
The rule of conjunction, although not listed as a tautology, simply asserts that
P, Q :. P A Q is logically valid; that is, P A Q P A Q is a tautology. Likewise,
the rule of disjunction follows from the fact that P = P V Q is also a tautology. The
modus ponens rule is again a verbal statement of the modus ponens tautology
[P A (P ' Q)] Q. The implication P Q by itself, even if known to be true, infers
nothing about Q. The implication P Q is also true when both P and Q are false. How-
ever, if both P and P =*- Q are true, then Q must also be true.
The rule of conjunctive simplification is a restatement of the corresponding tautol-
ogy. However, it also follows as a consequence of the modus ponens rule. Since
(P A Q) P (or Q) is a tautology, by the rules of conjunction and modus ponens
(P A Q) A [(P A Q) P] P (or Q).
The modus tollens rule follows from the modus tollens tautology. It is also a simple
consequence of the contrapositive law and modus ponens. Since P Q is equivalent
to -Q -P, it is easily verified by a truth table that the formula -Q A (P Q) is
equivalent to -Q A (-Q -P). Thus by the modus ponens rule --P can be
inferred.
The above argument illustrates the use of the replacement rule. In the formula
-Q A (P = Q).
the expressions -Q and P =:;- Q are subformulas of the original formula. The new for-
mula -Q A (-Q =*- --P) was obtained from the original formula by replacing the sub-
formula P Q by its equivalent formula -Q = -P. The resulting formula is then
equivalent to the original. As a rule of inference, this is stated as follows.
Simple Replacement:
If P is a subformula of a formula P and P q Q', then from P one can infer
any formula Q that results from replacing an occurence of P in P with Q'.
There are several additional rules that are worthy of mention. The justification of
these rules is left to the exercises.
504 Appendix Logic and Proofs
Disjunctive Syllogism:
From P V Q and --P, one can infer Q.
Rule of Inference by Cases:
From P Q and R Q. one can infer (P V R) Q.
Rules of Biconditional Inference:
From P Q and Q P, one can infer P q Q.
From P Q, one can infer P . Q and Q . P.
A more detailed discussion of the propositional proof system would have taken us
too far astray from our main goal; namely, to provide basic rules of inference with
which to construct valid arguments. A formal discussion of the propositional proof sys-
tem can be found in the text by Bums listed in the Supplemental Reading section. In
that text the author proves that every tautology, and only tautologies, can be derived
from the listed rules of inference.
T o p r o v e the validity of an inference PL, ... , P. .. Q. we simply have to verify
that using the rules of inference, we can infer the conclusion Q from the given premises
Pr, ... , P. Each statement in the proof should be either a premise or an axiom, or
should follow from previous statements by one of the accepted rules of inference. De-
vising a proof is a feat of mathematical ingenuity. However, once a purported proof is
in hand, it can be easily checked step by step for validity. We illustrate the use of the
rules of inference with the following examples.
All EXAMPLES
(a) As our first example we consider the following verbal argument.
If John is a Democrat, he associates with Democrats.
But John does not associate with Democrats.
Therefore, John is not a Democrat.
This argument can be written in symbolic form as
P=:0. Q
-P.
By the rule of contrapositive inference the argument is valid, whatever the truth or fal-
sity of the statements in it may be.
(b) As our second example we consider the following.
(a) P
(b) P = Q
A.2 Rules of Inference 505
(c) P=(QTR)
By the modus ponens rule, from (a) and (b) we can infer Q. Likewise from (a) and (c)
we can also infer Q R. But now from Q and Q R, we can infer R. Thus the infer-
ence is logically valid.
In symbolic form, the above proof can be written as follows:
1. P (premise)
2. P Q (premise)
3. P (Q R) (premise)
4. Q (modus ponens I & 2)
5. Q 'R (modus ponens l & 3)
.. R (modus ponens 4 & 5)
(c) For our final example we consider the following argument concerning Cauchy se-
quences. For the purposes of illustration we need to know nothing about sequences,
with the exception that they are usually denoted as and that divergence is the nega-
tion of convergence.
The sequence diverges.
A bounded sequence has a convergent subsequence.
Every Cauchy sequence that has a convergent subsequence converges.
Therefore, the sequence is not Cauchy.
In this particular example the conclusion happens to be true, but the argument is not
valid. To see this we will write the argument in symbolic form. Let P, Q, R, and S de-
note the following statements
P: "The sequence {p,} is Cauchy:'
Q: 'The sequence {p} converges:'
R: "The sequence is bounded:'
S: 'The sequence has a convergent subsequence."
In symbolic form the above argument is expressed as follows:
(a) - Q
(b) R=* S
(c) (P A S) Q
By (a) and (c) and the modus tollens rule we can infer -(P A S). But by De Morgan's
law, -(P A S) is equivalent to the statement -Pv - S. However, from -P V -- S we
can infer neither --P nor -S. If we knew that S (i.e., -(--5)) was true, then by the rule
of disjunctive syllogism we could infer -P. Unfortunately, from the given premises
nothing can be inferred about S. If S has the truth value F, then -P V -S is true re-
gardless of the truth value of P. This allows us to obtain an assignment of truth values
that make the premises true and the conclusion false. Thus the argument is not valid.
506 Appendix Logic and Proofs
If the statement R had been included as a premise, then the resulting inference
would be valid. As is often the case in proofs written by the beginning student, state-
ment (b) of the proof is totally extraneous.
EXERCISES A.2
1. Construct a truth table for [(P' Q)A -.QJ -P to verify the validity of the argument in Example A.2.1(a).
2. a. Justify the rule of disjunctive syllogism.
b. Justify the rule of inference by cases.
3. Justify the rules of biconditional inference.
In Exercises 4-12 use the rules of inference or a truth table to test the validity of each of the following.
Direct Proof
The most straightforward type of proof of "P Q" is the direct proof; namely, we as-
sume the hypothesis P and use the axioms, computations, or other theorems and the
rules of logic to infer Q.
Direct Proof of P= Q
Proof. Assume P. Therefore Q. Thus P Q. t
We illustrate the method of direct proof and the use of the rules of inference in jus-
tifying the validity of an argument with the following examples.
A.3.1 EXAMPLES
(a) In Section 2.6 of the text the following theorem about Cauchy sequences is proved.
Theorem 2.6.4. Every Cauchy sequence of real numbers converges.
I In the text the symbol is used to mark the end of a proof. Some authors prefer to use QED. which is an
abbreviation of the Latin "quod eras demonstrandum." meaning "which was to be proved."
508 Appendix Logic and Proofs
From Th 1 and Th2, by the transitive rule we can infer P ' S. Thus from our assump-
tion P. by modus ponens we can infer S. Hence by the rule of conjunctive inference we
can infer P A S. But now by Th3 and the modus ponens rule we have
{(P A S) A [(P A S)*Q]}z*Q.
It is important to again emphasize that the fact that (P A S) Q is true does not allow
us to infer anything about Q. It is also required that P A S must be true.
In symbolic form, the above proof can be written as follows:
1. P (hypothesis)
2. PAR (Thl)
3. RCS (Th2)
4. P .S (transitive rule)
5. S (modus ponens)
6. PAS (conjunctive inference)
7. (PAS) = Q (Th3)
. Q (modus ponens)
The above provides a very,methodical argument illustrating the validity of the im-
plication P Q. For a short proof this works very nicely. However, there is not a sin-
gle proof in this text that is written in such detail using symbolic logic to proceed from
P to Q. Mathematical proofs should be written in complete sentences, including justifi-
cations. The truth of any statement in the proof must follow from the initial hypothesis,
the axioms, or previously proved theorems. A typical proof of the theorem, using the
A.3 Mathematical Proofs 509
facts introduced above, would proceed as follows. The comments in parenthesis are not
part of the proof; they are included as explanations of the statements.
Proof. Let be a Cauchy sequence of real numbers. (This asserts the truth of the hy-
pothesis P.) By Theorem 2.6.2 the sequence is bounded. (This asserts the implication
P R.) Thus by Corollary 2.4.1 the sequence 1p.1 has a convergent subsequence. (This is
the implication R'' S, which by the transitive rule gives P * S. In the next step we invoke
Theorem 2.6.3; namely, that P A S =:O- Q.) The result now follows by Theorem 2.6.3. Q
A better and more careful way, especially for the novice, to express the last sen-
tence would be as follows: "Since the Cauchy sequence has a convergent subse-
quence, by Theorem 2.6.3 the sequence converges."
(b) For our second example we prove the following statement.
"If m is an even integer, then m2 is divisible by 4:'
Let P be the statement "The integer in is an even integer," and Q the statement "The in-
teger m2 is divisible by 4." Thus we wish to prove that P Q.
Note: If m and n are integers, we say that m divides n, or n is divisible by in, if n = km
for some integer k.
Proof. (Assume P.) Suppose m is an even integer. Then m = 2k for some integer k.
(Here we use the definition of an even integer.) Then m2 = 4k2. Thus m2 is divisible by
4. (The conclusion Q.) U
Proof by Contraposition
Since the implication P Q is equivalent to its contrapositive -Q P, we can
prove the implication P Q by assuming --Q and deriving P. Such a proof is called
a contrapositive proof or proof by contraposition.
10,
Contrapositive Proof of P .Q
(In the above we have used the rules concerning algebraic operations on the integers.
i.e., proved theorems.) Since m = 2k2 + 2k is an integer, we have
n2=2m+1
for some integer m. Thus n2 is odd (-P). U
Proof of P by Contradiction
Proof. Assume -P. Therefore R. Therefore -R. Thus P is true.
A.3.3 EXAMPLE To illustrate the method of proof by contradiction we consider the fol-
lowing classical proof that is irrational. This proof is due to Pythagoras, and is the
first known proof using contradiction. The statement P to be proved is as follows:
P "V is irrational."
Proof. (We assume - P; that is, is not irrational, i.e., rational.) Suppose that
is rational. Then V2 = m/n, where m and n are integers, with m and n not both
even. (The sentence "m and n not both even" will be our statement R.) But then
m2 = 2n2.
Therefore m2 is even. Thus by Example A.3.2 the integer m is even. Since m is even, by
Example A.3.l(b) the integer m2 is divisible by 4. Since m2 = 2n2, 2n2 is divisible by 4.
Thus n2 is even, and by Example A.3.2 the integer n is also even. Thus m and n are both
even. (This is our statement -- R. The negation of "not both are even" is "both are
even.") This is a contradiction. Thus V is irrational. U
The method of proof by contradiction can also be used to prove the implication
P Q. Using the law of contradiction we have that
(P = Q) . [- (P =* Q) = (R A - R) ].
A.3 Mathematical Proofs 511
Thus to prove "P Q" by contradiction we must prove that -(P = Q) implies a con-
tradiction R A - R. Since P =:: Q a -P V Q, by De Morgan's law and double nega-
tion
-(P=* Q)q(PA-Q).
Thus in the proof of P Q by contradiction we must show that the assumption
PA ^ Q logically implies a contradiction RA R, for some appropriate statement R.
Proof of P= Q by Contradiction
Proof. Suppose PA Q. Therefore R. Therefore -R. Thus P Q
O
A.3.4 EXAMPLES
(a) To illustrate the method of proof by contradiction we will prove the following the-
orem about the integers.
Theorem. If a, b, c are integers satisfying a2 + b2 = c2, then a or b is an even
integer.
Let P denote the statement "a, b, c are integers satisfying a2 + b2 = c2" and Q denote
the statement "a or b is an even integer."
In the proof by contradiction we assume P A -- Q. Since a and b are assumed to be
integers, from the assumption PA -- Q we conclude that a and b are odd integers. Thus
we can write a = 2m + 1 and b = 2n + 1, where m and n are integers. But then
c2=a2+b2=(2m+ 1)2+(2n+ 1)2,
which upon simplification gives c2 = 4k + 2 for some integer k. Thus c2 is an even in-
teger. But then by Example A.3.2 the integer c itself is even. Hence c = 2p for some in-
teger p. This gives
c2=4p2=4k+2 or (p2-k)=1.
Since p and k are integers, p2 - k is an integer. (This is our statement R.) On the
other hand, 1 is not an integer. (This is our statement -R.) From our assumption
PA _ Q we derived a contradiction R A - R.
(b) For our second example we prove the theorem of Example A.3.2 by contradiction.
For an integer n, let P and Q denote "n2 is even" and "n is even" respectively.
To prove P Q we assume P and -Q; that is, n2 is even, and n is odd. Since n is
odd, n = 2k + I for some integer k. Therefore,
n2 = (2k + 1)2 = 2(2k2 + 2k) + 1.
Therefore n2 is odd (-P). This is a contradiction (PA P).
512 Appendix Logic and Proofs
Proof by Cases
The method of proof by cases is based on the rule of inference by cases. Thus to prove
that (P V Q) R it suffices to prove that P R and that Q - R. We illustrate this
with the following examples.
A.3.5 EXAMPLES
(a) For our first example we prove the following: "If n is a positive integer, then
n2 + n + l is odd:'
If n is a positive integer, then n can be either even or odd. Thus if P and Q repre-
sent the statements "n is an even positive integer" and "n is an odd positive integer" re-
spectively, then the statement we wish to prove is (P V Q) R, where R represents the
statement "n2 + n + 1 is odd:' By the rule of inference by cases it suffices to prove
P R and Q =* R. The details are left to the exercises (Exercise 6).
'(b) For our second illustration of the method of proof by cases we consider the fol-
lowing theorem.
Theorem. There are two irrational numbers a and b such that ab is rational.
Proof. First case: is rational. If this is the case we take a = b = V. Second
case: N/r2`r' is irrational. In this case we take a = \/'2-"2 and b = V2-. So
2(`/-2)'=
ab=(V "i)V2 _ 2=2,
which is rational. Therefore, there must be irrational numbers a and b so that ab is
rational.
A.3 Mathematical Proofs 513
In the proof, we let R denote the statement of the theorem and P the statement " V-2 % 2
is rational." In Case 1 we have P R, and in Case 2 we have -P = R. Thus by the rule
of inference by cases we have (P V -- P) R. But P V -- P is true regardless of P. Hence
R follows by the modus ponens rule. This example not only illustrates the method of proof
by cases, it also illustrates how P V -- P can always be asserted in a proof.
Counterexamples
Some conjectures in mathematics are simply statements that are either true or false. For
example, "V is irrational:' Other conjectures are general in that they assert something
about a whole class of objects. For example,
"Every Cauchy sequence of real numbers converges:'
If n is an even integer, then n2 is even."
The first makes an assertion about all Cauchy sequences, whereas the second makes an
assertion about all even integers. Both of these statements are true. If we are presented
with a statement about a class of objects, then such a statement is true if and only if it is
true for every object in the class. Thus to conclude that such a statement is false it suf-
fices to exhibit one object in the class for which the statement is not true. Such an object
is called a counterexample. To illustrate this we consider the following conjecture:
"If n is a positive integer, then n2 - n + 5 is prime."
With a little bit of thought, most students will immediately conclude that this conjec-
ture is false. If we check for n = 1, 2, 3, 4, then n2 - n + 5 becomes 5, 7, 11, and 17,
which are indeed all prime. However, when n = 5, n2 - n + 5 is equal to 25, which is
certainly not prime. Since we have exhibited an object (n = 5) for which the hypothe-
sis P is true but the conclusion Q is false, the implication P Q is false.
The previous example was very elementary, and to find a counterexample was not
very difficult. This, however, is not always the case. As an example, consider the fol-
lowing conjecture.
"Every continuous function on a closed and bounded interval is differentiable
except perhaps at a finite number of points:'
Most students who have completed a basic calculus sequence might be inclined to be-
lieve that this conjecture is true. Certainly, on the basis of most examples encountered
in calculus such a conjecture seems reasonable. In fact, many mathematicians through
the mid-nineteenth century accepted this, or a variation of it, as true. It was not until
1874, when Weierstrass constructed an example of a continuous function that was
nowhere differentiable (see Section 8.5), that the above conjecture was proved false.
Helpful Hints
The most common complaint heard from students when asked to prove a theorem is "I
don't know where to start" Unfortunately, there are no easy rules that can be used to
tell someone how to prove a theorem. Many of the problems and theorems in the exer-
514 Appendix Logic and Proofs
cises follow from the definitions or from previous theorems. Some, however, require in-
sight and creativity, and for these the students must devise their own arguments. Some
helpful hints in constructing proofs follow.
(1) Make a list of all hypotheses and of what you want to prove. Do not ignore any
of the hypotheses. As a general rule they are all required. If you have not used all the
hypotheses then most likely your proof is incorrect.
(2) Refresh your memory with the pertinent definitions. If necessary, write them
out. This will help you to memorize them and also to understand them. It is very im-
portant that you know and understand all the definitions. They are the foundations upon
which the theory is built.
(3) Search for theorems that have similar hypotheses or similar conclusions.
Proved theorems are not just results; they are also tools that enable you to develop the
theory further. Suppose you are given P and R as hypotheses and are asked to prove Q.
If you have a theorem that P S and you can prove that (R A S) Q, then you have
your desired proof. An alternative approach is to work backward; that is. to find a the-
orem that has the same conclusion Q with given hypothesis S, and see if you can prove
S from the given hypotheses P and R.
(4) Learn the statements and the proofs of theorems. When I teach real analysis. I
always require students to memorize the statements of all the theorems and the proofs
of selected theorems. Contrary to student beliefs, this is not done to torture them. The
statements of the theorems are the tools used in proving other theorems; the proofs pro-
vide useful techniques that may be used elsewhere. They also provide a good model of
how to write a correct proof.
Some common errors committed by students include the following.
(1) Using theorems for which all the hypotheses are not satisfied.
(2) Making extra assumptions beyond those given in the statement of the problem
or theorem.
(3) When asked to prove something, for example all continuous functions f, the the-
orem is proved for a particular function such as f (x) = x-. Even though this is incorrect,
by attempting to prove the theorem for a special case, the student may in fact gain insight
on how to prove the result for the general case.
R EXERCISES A.3
1. As in Example A.3.l(a), write the proof of Example A.3.I(b) in symbolic form.
2. Same as the previous exercise for Example A.3.2.
3. Construct proofs of each of the following statements about the positive integers. Suppose k. m, and n are positive
integers.
a. If m and n are odd, then mn is odd. b. If m is odd, then m, is odd.
c. If m2 is odd, then m is odd. d. If k divides m and m divides n, then k divides n.
e. If n is odd, then n2 + 1 is even.
A.4 Use of Quantifiers 515
4. Consider the following statement. "If n is a positive integer, then > Prove the statement by pros iding
a. a direct proof, b. a proof by contradiction.
5. Prove that is irrational by contradiction.
6. Complete the details of the proof in Example A.3.5.
7. Prove that the equation a2 = 4b + 3 has no integer solutions.
S. Provide a counterexample to each of the following statements.
a. If n is a positive integer, then n2 - n + 41 is prime.
b. If n is a positive integer, then n! < 2'.
c. If n is a positive integer, then n' < (n + I)".
d. Every continuous function is differentiable.
Quantified Sentences
The sentence "x'- = 4" is not a proposition as it is neither true nor false. If we replace
x by specific values then the statement "x 2 = 4" becomes a proposition. For example,
the sentence is true for x = 2 and false for x = 3. Likewise, the sentence "x is a ratio-
nal number" is neither true nor false until x is replaced by a specific quantity. In the
sentences "x 2 = 4" and "x is a rational number," the 'x' is called a variable and the sen-
tences themselves are called formulas or open sentences in the variable x. Specifically,
a formula (in logic) is a statement containing one or more variables which becomes a
proposition when the variables are replaced by particular objects. W e will use the nota-
tion P(x) to denote that P is a formula in the variable x. Likewise, a formula in the vari-
ables x 1 ,. xk will be denoted by P(xt, .. . , xk). For example, "xi = x2 + x3' is a
formula in the variables xi, x2, x3.
Before the truth of a formula P(x) can be determined we must specify what objects
are available for discussion. This is called the universe U for P(x). For example. for the
formula x2 = 4 an appropriate choice for the universe U may be either the set of posi-
tive integers N, the set of all integers 71, or even the set of real numbers R. It is not
enough, however, to just specify the universe: For example, in the formula x - y = x.
a meaning must also be given to " " and If we are discussing 2 X 2 matrices.
then in addition to specifying the universe C' as the set of 2 X 2 matrices, we must also
2 Since this section requires some basic knowledge of the terminology of sets it is best postponed until Sec-
tion 1.1 has been read.
516 Appendix Logic and Proofs
AA.2 EXAMPLES
(a) As our first example we express the following theorem of the text as a quantified
sentence.
Theorem 2.6.4. Every Cauchy sequence of real numbers converges.
Clearly, the quantifier to be used is V. For our universe we take U to be the set of
all sequences of real numbers, and let C(x) and Q(x) denote the open sentences "x
is a Cauchy sequence" and "x converges" respectively. Consider first the sentence
3 The symbol 3! is often used to denote the existence of a unique x for which P(x) is true.
A.4 Use of Quantifiers 517
(b'x)[C(x) A Q(x)]. This sentence would be translated as follows: "For all sequences
x, x is a Cauchy sequence and x converges:' This sentence, however, is the same as
"every sequence is a convergent Cauchy sequence:' and this is clearly not the intent of
the original statement.
In other texts, Theorem 2.6.4 is sometimes expressed as "if is a Cauchy se-
quence of real numbers, then {an} converges" This statement can also be rewritten as
"For all sequences {an} of real numbers, if {an} is a Cauchy sequence, then {an} con-
verges:' This version, although somewhat awkward grammatically, is now easily writ-
ten in symbolic form as (Vx E U)[C(x) ' Q(x)l. In general, a sentence of the form
"All P(x) are Q(x)" is expressed in symbolic form as (Vx)[P(x) Q(x)].
(b) As another example consider the statement
"Some bounded sequences converge."
As in (a) let Q(x) be the sentence "x converges" and B(x) be the sentence "x is a
bounded sequence." Since "some" is taken to mean at least one, the proper quantifier is
3. However, should the statement be expressed symbolically as (3x E U)[B(x) Q(x)]
or as (3x E U)[B(x) A Q(x)]? The first would be interpreted as "There exists a se-
quence x, such that if x is bounded, then x converges." This clearly is not the intent of
the sentence. It does not ensure the existence of a bounded sequence that converges. The
second, (3x)[B(x) A Q(x)l, reads "There exists a sequence x, such that x is bounded
and x converges," and this is the correct interpretation. In general, the statement "Some
P(x) are Q(x)" is expressed symbolically as (3x)[P(x) A Q(x)].
(c) Most expressions in mathematics require the use of many quantifiers. To illustrate
this, consider the definition of convergence of a sequence of real numbers as given in
Section 2.1 of the text.
DefWtion 2.1.7. A sequence {pn} in Il8 is said to converge if there exists a point
p E R such that for every e > 0, there exists a positive integer n such that
Ipn - pI < efor all n?no.
This definition uses the quantifiers V and 3, not only once, but several times. We have
"3p E R," "3no E N," "(Ve)(e > 0)," and "(Vn)(n Z n.)." To write this statement in
symbolic form, we begin with (3p E R)[' ]. Consider the sentence "for every e > 0,
there exists ..:' This phrase, properly stated, should be expressed as "for all e E 13, if
e > 0, then ...,"which in symbolic form would be written as (Ve)[(e > 0) . (. )].
This leaves us with the final phrase "there exists a positive integer n, such that
Ipn - pI < e for all n n,,. This would be written as (3no E N) ("Ipn - p] < e for
all n ? no"). The statement "Ipn - pi < e for all n ? no," again if properly stated,
would read as "for all n E N, if n z n,,, then Ipn - pI < e," or in symbolic form,
(`tin E N)((n a I p, - p I < e). Combining all of the above finally gives "The se-
quence { pn} in R is said to converge" if
(3p e R)[(`de){(e > 0) ' [(3no E NI){(Vn E N)((n a no) Ipn - pI < e)}]}].
4 In mathematics the term "some" is taken to mean at least one. This differs from the colloquial interpreta-
tion where "some' is occasionally taken to mean two or more.
518 Appendix Logic and Proofs
A.4.4 EXAMPLES
(a) As in Example A.4.2(a) we consider the statement "Every Cauchy sequence con-
verges.' Using the same notation, in symbolic form this sentence was expressed as
(Vx)[C(x) .Q(x)]. The negation of the statement becomes (3x) - [C(x) Q(x)J.
Now ( C(x) Q(x)] is equivalent to (- C(x) V Q(x)), and by De Morgan's law the
latter is equivalent to C(x) A -- Q(x). Thus the negation of (Vx E U)[C(x) Q(x)] is
(3x E U)[C(x) A -- Q(x)]. This last statement would be read as "Mere exists a Cauchy
sequence in R that diverges." (Note: This .statement, however, is false in the real num-
ber system R, but true in the rational number system 0.)
(b) For our next example consider the negation of the statement "Some bounded se-
quences converge" In symbolic form this was expressed as (3x)[ B(x) A Q(x)] (see Ex-
ample A.4.2(b)). The negation of this statement becomes (Vx) -- [B(x) A Q(x)j. which
by De Morgan's law is equivalent to (Vx)[- B(x) V Q(x)]. But - B(x) V -- Q(x) is
equivalent to B(x) -- Q(x). Hence the negation of (3x)[B(x) A Q(x)] is equivalent to
(dx)[B(x) - Q(x)]. This last statement would read as "For all sequences of real
numbers, if is bounded, then diverges," or more simply as 'All bounded se-
quences in R diverge:'
(c) As our final example we undertake the negation of the definition of convergence
of a sequence of real numbers. In symbolic form, a sequence {pn} in R is said to con-
verge if
(3p (=- l)[(`de){(e > 0)' [(3n E N){(`dn E N)((n >_ Ip,, - pi <
We proceed to negate this sentence step by step. First, the negation of (3p E R)[ . }
becomes (Vp E 111) -- [ ]. Now inside the bracket [. ] we have (`de){P(e)w
Q(e, no, n)}, where P(e) denotes "e > 0" and Q(e, no, n) denotes "(3n0 E N)
{(V n E N)((n ? nn) a 1p. pi < e)}." But (Ve){P(e) Q(e, n,,, n)} is equiva-
lent to (3e){P(e) A -- Q(e, no, n)}. It is left as an exercise (Exercise 7) to show that the
negation of Q(e, no, n) becomes
(Vn,, E N){(3n E N)[(n ? n,) A (IPn - pi e)]}.
Combining all the above gives us the following. "A sequence in R is said to di-
verge" if
(Vp E R)(3e){(e > 0) A (Vn E N)[(3n E N)(n =' no) A (Ip - pi ? e)]},
or
0 EXERCISES A.4
1. Express each of the following sentences in symbolic form. Specify an appropriate universe for each.
a. All men are mortal.
b. Not all mortals are men.
C. Some isoceles triangles are equilateral triangles.
d. Some triangles are isoceles triangles and some are equilateral triangles.
e. Some triangles are isoceles and equilateral triangles.
520 Appendix Logic and Proofs
SUPPLEMENTAL READING
Burris, S. N., Logic for Mathematics and Computer Smith, D., Eggen, M., and St. Andre, R.. A Transition
Science, Prentice Hall, New Jersey, 1998. to Advanced Mathematics, Brooks/Cole. Pacific
Lemmon, E. J., Beginning Logic, Chapman & Grove, CA, 1997.
Hall/CRC, Boca Raton, Florida, 1998.
Bibliography
522
Hints and Solutions
to Selected Exercises
S CHAPTER 1
EXERCISES 1.1 page 5
2.(a)AfB=B,Af17L={-1,0,1,2,3,4,5},Bf1C={x:2:5x!53} (b)AXB={(x,y):-l<x<_5,
O!5 y<_3} 4. (a) Suppose xEAfl(Bf1C).ThenxEAandxEBflC.SincexEBf1C. xEBandxEC.
Thus x E A fl B and x E C. Therefore, x E (A f7 B) fl C. This proves An (an c) c (A fl B) fl C. The reverse
containment is proved similarly. 7. (a) Let x E A U (B fl C). Then x E A or x E s fl C. If x E A. then x E A U B
EXERCISES 13 page 19
1. (b) For n = 1, 1 = l2. Assume the result is true for n = k. Then for n = k + 1,
I +3+ . +(2k- 1)+(2(k+ 1) - 1)=k2+(2k+ 1)=(k+ 1)2. (f) When
n = 1, .r' - y2 = (x - y)(x + y) and equality holds. For n = k + 1, write
xk+2 xk-2 - Xyk-' + XY`-' - Yk+ = X(xk'' - k-')+ (x - y)yk+', and now apply the induction hypothesis.
2. (a) The result is true for n = 1. Assume that fork E N. 2k > k. Then by the induction hypothesis,
2k-' = 2k 2 > k 2 = k + k ? k + 1. (c) For n = 4,4! = 24 > 16 = 24. Thus the inequality is true when
n = 4. Assume that n! > 2" for some n z 4. Then (n + I)! = (n + I )n! > (n + 1)2" > 2 2" = 2"-'. Thus by the
modified principle of mathematical induction the inequality holds for all n E N. n ? 4. 4. For n E N let P(n) be
the statement f(n) = 3 2" + (-1)". Then P(n) is true for n = 1, 2. Fork ? 3, assume that P(j) is true for all
523
524 Hints and Solutions to Selected Exercises
j E N1, j < k. Use the fact that f(k) = 2f(k - 2) + f(k - 1) and the induction hypothesis to show that P(k) is true.
Thus by the second principle of mathematical induction, the result holds for all n E N. 5. (b) f(n) = n2
(d) f(n) = 0 if n is even, and f(n) _ (- 1)t"-'n/n! if n is odd. 7. For each n E N let S" = r + r2 + + r".
Then S. - rS" = r - r"+', from which the result follows. 8. Hint: Let A = '(a, + + a") and write a" xA
for some x a 0. Use the induction hypothesis to prove that (a, a" '') s x'1t"+')A.
Now use Bernoulli's inequality to prove that xj'j"'') s (n + x)/(n + 1). From this it now follows that
xii("+')A 5 n + x A =
n + 1
1
n + 1
(a,+.+a"+a"+,).
EXERCISES 1A page 26
4. Consider (a - b)2. S. (a) inf A = 0, sup A = I (e) inf C = -oo, sup C = 00 (e) inf E = 1, sup E = 3
(g) inf C = 0. sup C = no (i) inf I = - 2, sup ! = 2 14. (b) Since A and B are nonempty and bounded above,
a = supA and is = sup B both exist in R. Since a = supA we have a s a for all a E A. Similarly b :5 0 for all
b E B. Therefore, a + b 5 a + $ for all a E A, b E B. Thus a + /3 is an upper bound for A + B, and thus
y=sup(A+B)5a+J3.
To prove the reverse inequality, we first note that since y is an upper bound for A + B, a + b s y forall a E A.
b r= B. Let b E B be arbitrary, but fixed. Then a c y - b for all a E A. Thus y - b is an upper bound for A and
hence a <_ y - b. Since this holds for all b E B, we also have that b s y - a for all b E B. Thus S s a; i.e.,
a + / <- y. An alternative way to prove that y = a + /3 is by contradiction. Suppose y < a + P. Set
e = a + /3 - y. By assumption e > 0. Now consider a - Ie and 0 - ?e, and use Theorem 1.4.4to obtain a
contradiction. 15. (a) Let a = sup{f(x) : x E X}, )3 = sup{g(x) : x E X}. Since the range off and g am bounded,
a and P are finite with f(x) + g(x) a + 6 for every x E X. Therefore, a + 0 is an upper bound for
{f(x) + g(x) : x E X}. Thus sup{f(x) + g(x) : x E X} s a + /3. 16. (a) F(x) = 3x + 2. sup{F(x):x E [0, 1)} = 5
20. (a) F(x) = 3x + 2 and H(y) = 2y. Thus sup{H(y) : y E [0, 111 = 2 and inf{F(x) : x E [0, 1 ]} = 2.
EXERCISES 15 page 30
1. First prove that if p and q are positive integers, then there exists n E N such that np > q. If p > q then n = 1
works. If p s q, consider (q + I )p. 6. (a) Use the fact V2/2 is irrational. (b) Use Theorem 1.5.2 and (a).
CHAPTER 2
EXERCISES 2.1 page 52
2. Ix I = Ix - y + y 1 5 Ix - y I + I y 1. Therefore, I x I- I y 15 I x - y 1. Interchanging x and y gives I y i- Ix 1 5
I y- x I= I x- y I Now use the definition of II x I- I y I I. 6. (a) -3:5,x:5 13/3 (c) -1 <x < 2 & (a) Let
a" = (3n + 5)/(2n + 7). Then I a. - 21 = 11/2(2n + 7) < 3/n. Given e > 0, choose n E N such that n, a 3/e.
Then for all n a n I a. - 2 I < e. Therefore, lim a" = 2(c) Set a" = (n2 + l )/2n2. Then I a. - III = < -.
Given e > 0, choose n, E N such that no a 1/2e. Then for all n a n", I a. - 121 < e.Thus lim a" = 2.
9. (a) If it is even, then n(1 + (-1)") = 2n. Thus the sequence is unbounded and diverges in R.
(c) When n is even, i.e., n = 2k, then sin i = sinker = 0. On the other hand, when it is odd, i.e., n = 2k + 1, then
sin If = sin(2k + 1) = (-1)k. Thus sin 9 assumes the values -1, 0, and 1 for infinitely many values of n. Therefore
the sequence diverges. 11. (a) Write b = 1 + a where a > 0. By Example 1.3.3(b), b" ? I + no. Now use the
previous problem. 12. First show that I a2. - a21 5 (I a" I + I a I) I a" - a I. Now use the fact that since {a"}
converges, there exists a positive constant M such that I a" 15 M for all n e N. 14. Consider a = 0 and a > 0
separately. For a > 0, V - V = (a" - a)/(V + Va-).
EXERCISES 2.2 page 59
S. (b) If p > 1, let x" = -Vp - 1. Apply the inequality of Example 1.3.3(b) to (I + x")". 6. (a) Is (c) -1 (e) 2
7. (a) Converges to 1. (c) Since
1 n2 + 1 2 (1 + 1/n2)2
n2 \2n + 3 (2 + 3/n)2'
by use of the limit theorems the sequence converges to 4.
8. Use the fact that I cos x 1 5 1 for all x E R. 10. (a) Suppose lim a"+ ,/a" = L < 1. Choose e > 0 such that
L + e < 1. For this a there exists n. E 101 such that a"+ ,/a" < L + e for all n a n,. From this one obtains that for
i > no, 0 < a" <_ (L + M(L + e)", where M = a",(L + Since (L + e) < 1, by Theorem 2.2.6(e).
slim (L + e)" = 0. The result now follows by Theorem 2.2.4. 11. (a) With a" = n2a", 0 < a <
L = Iimoa". i/a" = a lim (I + n)2 = a. Thus since L = a < 1, the sequence converges. 12. Set
x" (a" - 1)/(a" + 1), and solve for a". 13. (b) Verify the result for it = 1. Assume the result holds for it = k. For
n = k + 1, (1 + a)(k+1) _ (1 + a)(1 + a)k, which by the induction hypothesis
=(1 +a) Li
+ k j- k
k k+1
k
).v.
a ±i a'+ 71
(- _ 1
I=a i=o i-t
Ck +
Using (a) show that the above is equal to I {a' 14. (a) By considering the sequence {ak - a} show first
that one can assume a = 0.
I
i j
EXERCISES 2.3 page 65
2. Take I, _ [n oo). 3. (a) Set x" _ ( n2 + I)/n. Then x.2= 1 + , > 1 + " ; = x.'+,. Thus
x2. > x,2,+, > 1. From this it now follows that x" > x"+1 > 1 for all n E N. and that lim x" = 1. (c) Since
a > 1, a"+' = a a" > a". Thus {a"} is monotone increasing. Let or = sup{a" : n E N!}. If a < oo, then
a = lim a"+ = a lim a" = as > a. This, however, is a contradiction. Thus a" -4 oo. S. Use mathematical
induction to show that a" > 1 for all n (=- N. From the inequality 2ab s a2 + 62, a, b >_ 0, we have 2a" 5 a.2 + I or
a"+, = 2 - 1/a" s a". Therefore, {a"} is monotone decreasing. Finally, if a = Jim a", then
a = woo
lint a"+,="moo 1=2- 1.a
limit- a")
Therefore, a = 1. 7. (c) By induction, the sequence {a"} is monotone increasing and bounded above by 3. If
a = Jim a", then a2 = lim a"+, = lim(2a" + 3) = 2a + 3. Thus a is a solution of a2 - 2a - 3 = 0. The two solutions
526 Hints and Solutions to Selected Exercises
are - I and 3. Since a must be positive, we have a = 3. (e) lim a" = 2. 9. (a) Use the inequality ab !< i (a2 + b2)
to prove that x"., > \ for all n. To show that {x"} is monotone decreasing, consider x",, - x" and simplify.
12. (a) e2 (e) e312 13. To show that {s"} is unbounded, show that s2. > I + "I. Hint: First show it for n = 1. 2. and
3; then use mathematical induction to prove the result for all n E N. 15. Hint: Fork = 2. t < k,k ! l) = k i - ;.
17. (b) n + n- (n - 1) (d) Hint: Write a = (1 + b) with b > 0. Now use the binomial theorem to
show that a"In° ? cn for some positive constant c and all n sufficiently large. 18. (c) The sequence is not monotone
increasing: If x" = n + (-1)'Vn-, then xu,, < x2,,. 23. This problem is somewhat tricky. It is not sufficient to just
choose a monotone increasing sequence in the set; one also has to guarantee that the sequence converges to the least
upper bound of the set. Let E be a nonempty subset of R that is bounded above. Let aqt denote the set of upper bounds of
E. Since E * 0, we can choose an element x, E E. Also, since E is bounded above, OIL * 46. Choose /3, E V. Let
a, = 12 (x, + 6, ), and consider the two intervals [x,, a, ] and (a,,19, ). Since x, E E, one or both of these intervals have
nonempty intersection with E. If (a,, /3,] fl E * 0. choose x2 E E such that a, < x2 s S,. In this case, set $2 = /3,. If
(a,, /3, ] fl E _ 0, choose x2 E E such that x, < x2 a,, and set /32 = a,. In this case, $2 E OU.. Proceeding
inductively, construct two monotone sequences {)"} and {/3"} such that (a) {x"} C E with x" 5 x"+, for all n, (b)
{/3"} C OIL with ,6. ? /3"+, for all n, and (c) 0 0" - x" <- 2-"+ 1(b, - x1). Assuming that every bounded monotone
sequence converges, let A = Jim /i". By (c) we also have /3 = lim x". It only remains to be shown that /3 = sup E.
24. Suppose A = {x" : n E N} is a countable subset of [0, 1 ]. To show that A C [0, l ] proceed as follows:
At least one of the three closed intervals [0, 13], [13.23). [j, 1 ] does not contain x,. Call it!,. Divide 1, into three closed
intervals of length 1/32. At least one of these, say 12, does not contain x2.
subsequence is bounded. Consider the subsequence {s,,) where nk = 2k - I. and prove that s,,. < I + a + +
where a = 2-tP-1l. For example, when k = 2, nk = 3 and
s3=1+2P+
3P
=1+( +]1<1+P=1+2v -,-'
P
0 CHAPTER 3
EXERCISES 3.1 page 100
2. By Corollary 2.4.8 a finite set has no limit points. Now apply Theorem 3.1.9. This can also be proved directly by
showing that the complement of a finite set is the finite union of open intervals. 6. (a) Let p E 0 = U,O0 be
arbitrary. Then p E 0, for some a E A. Since 0, is open, there exists e > 0 such that N,(p) C 0,: But then
N,(p) C 0; i.e., p is an interior point of 0. 8. (a) Let p E Int(E) be arbitrary. Since p is an interior point of E, there
exists an e > 0 such that N,(p) C E. To show that N,(p) C Int(E) it remains to be shown that every q E N,(p) is an
interior point of E. 9. (a) Since A U B is a subset of A U B (which is closed), by Theorem 3.1.11(c),
A U B C A U B. The reverse containment follows analogously. 11. (a) (a, b) 13. Use Exercise 6(b) of Section
1.5. 14. If p E D, set p, = p for all n E N. If p is a limit point of D, use Theorem 2.4.7. 17. Let {r,}'_, bean
enumeration of 0, and {t:,}- , an enumeration of the positive rational numbers. Take ' = {N,lr,) : j, n E N).
18. (a) Closed in X. (c) Neither open nor closed in X. 20. Suppose U C R is open, and suppose E C U is open in
U. Let p E E be arbitrary. Use the fact that E is open in U and that U is open to show that there exists an e > 0 such
that N,(p) C E. Thus p is an interior point of E. Since p E E was arbitrary, E is open in R. The converse is obvious.
23. (a) Let U = (0, 1) and V = (;, ;). 24. (a) Use the fact that A = A U A', and if U open satisfies u n A' # 0
then U fl A # 4,. 25. Let A be a nonempty subset of R. If A contains at least two points and is not an interval, then
there exist r, s E A with r < s and r E P with r < r < s. but r E A. The open sets U = (-no, r) and V = (r, oo) will
prove that A is not connected. Therefore, every connected set is an interval. Conversely, suppose A is an interval and A
is not connected. Then there exist disjoint open sets U and V with A fl U 0 ¢.. A fl V * 4,, and A C U U V. Suppose
a E A fl U and b E A fl V. By Theorem 3.1.13 applied to U and V, there exist disjoint open intervals f and J such
that a E I and b E J. Suppose a < b and J = (r, s). Show that t 14 U U V butt E A. This contradiction proves that
A is connected.
CHAPTER 4
EXERCISES 4.1 page 128
1. (a) If f(x) = 2x - 7, L = -3, then 1f(x) - L 1 = 21x - 21. Given e > 0 take S = e/2. Then for all x with
Ix - 21 < 8, 1 f (x) - L 1 = 2 ; x - 21 < 2S = e. (c) If f(x) = k/(1 + x), then
j(x) 21 21x+1111
< 2 lx- 1
for all x > 0. Hence given e > 0, choose S = min {2e, I ). With this choice of 8, x > 0 and thus
11(x) - I < i S s e. 3. (a) The limit does not exist. For x > 0, x/I x 1 = 1, whereas for x < 0, x/I x 1
(c) The limit does not exist. Consider the sequence { I /nzr), which has limit 0 as n -+ no. 4. lim1 f (x) 3 5. (a) By
Figure 4.5, for 0 < t < -rr/2, sin t = length of PQ < length of arc PR = t. 7. (a) Use the inequality
528 Hints and Solutions to Selected Exercises
IIf(x)I - ILII s If(x) - LI (see 2.1.4). (c) Use induction on n. S. (a) 0 (c) (e) J (g) 2. Note:
2( ). 9. Let L = lim f(x). By hypothesis L > 0. Take e = 1 L and use the definition of limit. 12. Since
x-r
g is bounded on E, there exists a positive constant M such that Ig(x) I s M for all x E E. Thus 1f(x) g(x) 15 MI f(x)'.
for all x E E. Now use the fact that lim f(x) = 0.
'v
16. Suppose lim f(x) = L. Let e > 0 be given. By definition there exists M > 0 such that I1(x) - L I < e for all
X-M
x E (a, oo) with x > M. Let S = 1/M. Then for all t E (0, 1/a) with t < S, 1 It E (a, oc) and I It > M. Therefore.
IS(r) - LI = If(',) - LI < e. The proof that limo g(r) = L implies iim f(x) = L is similar. 17. (a) 22 (c) 2
-M
(e) 2 (g) Limit does not exist. For all x > ;, cos ; > 1. Thus x cos x > 1 x and x cos ; - oo as x - oo.
EXERCISES 4.2 page 141
1. (c) Since I - cos x = 2 sin2(x/2), for x # 0, g(x) sin2(x/2). Now use the fact that I sin t I s 11 1.
4.Ifp>0.
If(x) - f(P)I = If - VI = Ix - PI/(V + V P - ) < LIx - PI
Let e > 0 be given. Set S = m i n { p , V p e } . Then I x - p I < S implies that 11(x) - f(p)I < e. Therefore, f is
continuous at p. If p = 0, set S = e'-. Alternatively use Theorem 4.1.3 and Exercise 14 of Section 2.1.
6. (a) See Exercise 7a of Section 4.1. 7. Use Theorem 4.2.4, and the fact that x" is continuous on R for all
n E N. 9. (a) R \ (-2.0,2). (c) R.
12. (a) Use the fact that max {f(x), g(x)} =12(1(x) + g(x) + I f(x) - g(x) 1).
16. Consider g(x) = f(x) - f(x - 1), xE(O, I J. 18. Let p E E be a limit point of F. Use Theorem 2.4.7 and
continuity off to show that p E F. 20. (b) By induction, f(nx) = nf(x) for all n E h) and x E R. In particular,
f (n) = en where c = Al). Also, c = Al) = An -.) = nf(.1). Therefore, Al) = cln. Since f is continuous, letting
n -ioo gives f(0) = 0. From this it now follows that f(-x) = f(x) for all x E R. Thus f(n) = cn for all n E Z and
f(r) = cr for all r E 0 (write r = m/n, m, n E Z). Finally, by continuity Ax) = cx for all x E R. 22. Take
e = 1. Then for this choice of a there exists a 8 > 0 such that 11(x) - f(P)I s 1 for all x E Na(p) fl E. Show that
this implies that If(x)I c (If(P)I + 1) for all x E Na(p) () E. 25. Theorem 2.4.7 and Theorem 3.2.10 should
prove helpful. 29. By hypothesis, for each x E K there exists e, > 0 and M, > 0 such that I Ay) 1 5 M, for all
y E N1(x) fl K. The collection {N,,(x)},Ex is an open cover of K. Now use compactness of K to show that there exists
a positive constant M such that I f(y)I S M for all y E K.
EXERCISES 4.3 page 147
2. (a) Suppose f(x) = x2 is uniformly continuous on [0, oo). Then with e = 1, there exists a S > 0 such that
J Ax) - f C Y ) I < I f o r all x, y E [0, oo) satisfying I x - y I < 6. Set x" = n and y" = it + 1. If n E IOl is such that
no6 > 1, then Iy" - x" I = I_11 < S for all n a n,,. But I f&.) -1(x")1 = 2+3, z 2 for all n. This is a contradiction!
3. (a) For all x, y E [0, oo),
Thus given e > 0, the choice 6 = e will work. (f) Set g(x) = sin x/x, x E (0. 1 ] and g(0) = 1. Then g is continuous
on [0, 1 ], and thus by Theorem 4.3.4 uniformly continuous on [0, 1 ]. From this it now follows that f is uniformly con
tinuous on (0, 1). 4. (a) Show that If(x) - f(y)I 5 1 x - yI for all x, y E [a, oo).
a3
Now use the uniform continuity off and g. 13. (a) Let x, E E be arbitrary. For it = 1 set x,,. = f(x.). Show that
the sequence {x"} is contractive.
EXERCISES 4.4 page 160
2. (b) m. f(x) = ,lito. f(x) = 0 (f) lin f(x) = I. Hint: For <x '[ ] = n. 4. Use the fact that [.r; is
6
bounded near x,, = 2. 6. b = 14 7. For n E N set x = n and y = it + and show that I f (x.) - f(;)1 = sin 2.
12. If it E N, g(x) = x" is continuous and strictly increasing on (0, cc) with Range g = (0, oo). Therefore by
Theorem 4.4.12, its inverse function g-'(x) = x"" is also continuous on (0, oo). From this it now follows that f(x) is
continuous on (0, oo). 14. (a) Suppose first that U = (a, b) C 1. Then since f is strictly increasing and continuous
on 1. f((a, b)) = (f(a), f(b)), which is open. For an arbitrary open set U C 1, write U = where is a finite or
countable collection of open intervals, and use Theorems 1.7.14 and 3.1.6.
CHAPTER 5
EXERCISES S.1 page 174
1. (a) For f(x) = x3.
f(x + h) - f(x) - (x + h)3 - x' _
f '(X) = lim lim (3x2 + 3xh + h2) = 3x.
h ilm h =
(c) For h(x) = 1/x, x # 0,
h ,=
0
h '(.x) = .1im
.,
Dividing by h and taking the limit as h -- 0 proves the result for n E N. If n E 71 is negative, write x" = I /xf,
m E N. and use Theorem 5.1.5(c). 3. (a) Since cos x = sin(x + ), by the chain rule cos x = cos(x +
i (c) No. 7. (a) f'(x) exists for all x E 12\71.
-sin x. Alternatively, use the definition of the derivative. S. (a) Yes
Forx E (k, k + 1), k E Z,f(x) = x[x] = kr. Thus f'(x) = k for all x E (k, k+1), k E Z. (c) The function h is
differentiable at all x where sin x # 0. For x E (2kw, (2k + I ytr), k E 71, h'(x) = cos x. For x E ((2k - I )w, 2kwr),
k e Z, h'(x) -cos x. 9. (b) For x f 0, g'(x) = 2x sin - cos Since lynx cos does not exist. ii g'(x) does
0
not exist. 12. (a) f'(x) = 2/(2x + 1) (c) h'(x) = 3[L(x)]2/x 13. (b) f;(0) = hliw, ht''-'t sin ,which exists and
equals 0 if and only if (b - 1) > 0; i.e, b > I.
now follows that L(b') = rL(b) for all r E Q. Now use the continuity of L to prove that L(b") = xL(b) for all x E R.
where If = sup{b' : r E Q. r : x}. 23. (b) Since tan(Arctan x) = x. by Theorem 5.2.14 and the chain rule,
tan(Arctan x) = (sec(Arctan x))(' Arctan x) = I. The result now follows from the identity
sec2(Arctan x) = 1 + x2. To prove this, consider the right triangle with sides of length 1. 1 x (, 1 + x2 respectively.
4. Use the fact that since ,ilitf(x) exists, f(x) is bounded on (a, a + 8) for some S > 0.
6. (a) lim
x3+2x-3 = lint 5x4+2 = 7 (e) By L'Hospital's rule. lim
Inx
- = =-
1
lim= -x = 0.
!- 2x3 - x2 - 1 =-+I 6x2 - 2x 4" +-'= x
(e) Make the substitution x = 1/t.
CHAPTER 6
EXERCISES 6.1 page 221
(a) f(x)
_ 1, 0Sx< 1, Let 91= xa
{ x j,
,, ... , x"} be any partition of [0, 2] and let k E {l.... , n} be such
2, 1 s x <- 2.
that x4 _, < I s x4. Then W(9D, f) = 4 - 3x4 and gt(9P, f) = 4 - 3x4_ ,. Thus 4k(9', f) - W(9', f) = 3(x4 - x4-,). By
Theorem 6.1.7 it now follows that f is Riemann integrable on (0, 2]. Also, since x,,- I < I S x4, 4914) 5 1 < 14911f)
for any partition P. Hence f f = 1. Alternatively, consider the partition 91 = {0, c, 1, 2} where 0 < c < I is
arbitrary. For this partition, 2(9j) = 1 and 411.(9', f) = 4 - 3c. The result now follows as above. 3. (a) If
91 = {xo, x,, .... x"} is a partition of [a, b], then inf{ f(t) : t E [x;_ x;]} = sup{f(t) : t E [x;_,, xj]} = c. Therefore,
20,J) = %4911f) = jt c(x; - x,-,) - c(b - a). 4. (a) Since f(x) _ [3x] is monotone increasing on (0, I ], f is
Riemann integrable on (0, 1]. For n 2:4 consider the partition P. _ {0, 1- t 1 }. For this partition
2(91, f) = I and 9t(9>", f) = I + I. From this it now follows that Jo [ 3x] dx = 1. 6. Let 91 = {xj,, x,, .... ..x.) be a
partition of [a, b). Since f(x) <- g(x) for all x E [a, b), sup{f(t) : t E [x,_,, x,]) :s sup{g(t) : t E [xj_, x,]) for all
i_= 1.... , n. As a consequence, alt(9P, f) s '9t(9', g) for all partitions 9 of (a, b). Ta ag the infimum over 9' gives
f f s " j g. The result now follows from the fact that f, g r= 9t[a, b]. 8. f f - 0, fo f = 3. 10. Since a z 0,
f is increasing on [a, b]. Thus if 91 = {xo, x,, ... , x"} is a partition of [a, b], and M, = x}. Therefore.
JI "
.T(9,e) = j zj-. £1 and `L(',f) _ x Axj.
ice!
Now show that x2_ i Axj :53(x3 - x3_,) s x,2Ax4. From this it will now follow that 2(9, f) s 3(b3 - a3) s
Since f is continuous, f E 9t(a, b) with f; f = 3(b3 - a3). 12. (a) With 91" =.-O.: k = 0, 1, ... , n} and f(x) = x,
By Exercise la, Section 1.3, Zk_, k = 3n(n + 1). Therefore, fo x dx = him "V = 3. 15. (a) Since f is
bounded on [a, b], If(x)1 s M for all x E [a, b]. Let 9 - {xo, x...... x"} be any partition of [a, b). For each i let
Hints and Solutions to Selected Exercises 531
M(f2) and m,(f2) denote the supremum and infimum of f'- respectively over [x,_,, x,] with an analogous definition for
M,(f) and m,(f). Let e > 0 be given. Then for each i there exists s t, E [x, _ x,] such that M,(f2) <f2(s,) + ?e and
m,(f2) >f2(t,) -/ `Z e. Therefore,
Use this to show that f E i [a, bJ. 7. (a) lim k2 = x2 dx = -. 8. For each fixed c, 0 < c < 1, the
-n3 Ju 3
series defining f becomes a finite sum on [0, c]. Evaluate and use Exercise 5.
IF(y)-F(x)I = JH
Thus F satisfies a Lipschitz condition on [a, b] and hence is uniformly continuous. 2. (b) F(x) = x for 0 5 x :s
F(x) = i - 2x for I < x 5 1. 2:
S. (b) F'(x) = cosx2 6. By the chain rule, -dO-.L(() = -1 = j[-L(x)]-Therefore, L(F) = -L(x) + C for some
constant C. Taking x = I shows that C = 0. 9. Let m and M denote the minimum and maximum off on [a. b],
respectively. Since g(x) z 0 for all x, mg(x) s f(x)g(x) 5 Mg(x). If fQ g > 0, then from the previous inequality one
obtains
Jb1g/Jbg
m SM.
a a
Nowapply the intermediate value theorem to f. If f, g = 0, use Theorem 6.2.2 to prove that fa fg = 0 and thus
the conclusion holds for any c E [a, b]. 11. (a) With rp(x) = In x. by Theorem 6.3.8 f '; dx = fo" 2 x dx = 2(In 2)2.
i
(b) Use integration by parts. (e) 2 In 2 - 1. Use Exercise 10 with (p (x) _ V. and f(t) = t/(1 + t).
S32 Hints and Solutions to Selected Exercises
Jo (x2+1)(x+1)
x
+2 _1
c+11+2Arctanc-2.
Thus 1dx ar 3. (a) For p > -1, the improper integral converges for all q E R. and
Jo ( + l)(
xz+ ) 4 2.
for p < -1, the improper integral diverges for all q E R. When p = -1. the improper integral converges for all
q < -1, and diverges for all q ;2! -1. 5. Use the fact that 0 s I f(x) I - f(x) s 21 f(x) I
7 . (a) I;{ Idx 5 j;i dx = ; - Thus Jim J,,l fI < oo. (b) By integration by parts.
!; dx = -
s - j: r ZZ dx. By (a) and Exercise 5. Wn j*'X2 dx exists. Also, Win = 0.7herefore.
C-00
Jim f
C-W
dx exists. 9. (a) To show convergence of the improper integral, consider the integrals of t"' a-' over
the two intervals (0, 1 ) and [ 1, oo). If jr ? 1, then t'-'e-'is continuous on [0, 1). and thus the integral over [0, 1 ]
clearly exists. If 0 < x < 1, then since a-' s 1 fort E (0, 1 ],
Os
f t'-'e'dts I
C e
Thus 1 to + f'tx-'e-'dt < on. For tit 1, use l'Hospital's rule to show that there exists a t, ? 1 such that
tx'' 07 for all t 2t t,. Thus tx' a-' s e'"2 for all t Z t, and as a consequence, Pm. IR tx-'' tit < on. Thus the
improper integral defining r (x) converges for all x > 0.
EXERCISES 6.5 page 259
1. 2f(0) 3. (a) See the hint to Exercise 1, Section 6.3. 5. (a) f - 1. Use Theorem 6.5.10. (C) 12 + 22 + 32.
For x r= [0, 3], [x] = 1(x - 1) + 1(x - 2) + I(x - 3), now use formula (12). & (a) jx , ii2. 12. (a) As in the
solution of Exercise 15a of Section 6.1, if 9 = {xo, x1, .... x"} is a partition of (a, b), Mi(f2) - m,(j2) s
2M[M,(f) - m,(f)] where M > 0 is such that I f(x)I s M for all x r= [a, b]. From this it now follows that
0 s ft (9,f2, a) - 2(9, f2, a) s 2M[olt(9, f, a) - T(9,f, a)].
Now apply Theorem 6.5.5.
EXERCISES 6.6 page 271
.
2. (a) With n = 4, h = .25. Set x, _ .25i, y, = f(x,), i = 0, 1.2.3.4. Then yo = 1.00000, y, = 0.94118,
Y2 = 0.80000, y3 = 0.64000, y4 = 0.50000. Therefore,
By computation f"(t) - 2(3t2 - 1)/(1 + t)3. Using the first derivative test, f(t) has a local maximum of at t = 1.
Therefore, I f (t) I s((I for all t E [0, 1]. Thus by equation (23) with n = 4,
I4_T4(f)1s12\2)42=0.0026042.
Hints and Solutions to Selected Exercises 533
Since 4 = 0.7853982 (to seven decimal places), I , - T4(f) I = 0.0026032. 3. (b) By computation, f (41(X)
3(4x2 - 1)(1 + X2)-712. By the first derivative test, the function f (") has a maximum on [0, 2] at x = v5/2. Thus
ft4>(x) I s 6/ (1 -+41Y < 1. If we choose n (even) so that fo f - S (f)j < 10-3, then we will be guaranteed
accuracy to four decimal places. By inequality (26) with M = 1, we need to choose n so that Ik'o -L < 10-3, or
n4 > 17,778. The value n = 12 will work. This value of n will guarantee that E12(f) < 0.0000086. Compare your
answer with the exact answer of V + 2ln(2 + V5).
CHAPTER 7
EXERCISES 7.1 page 291
2. (a) Diverges (c) Converges (e) Diverges (g) Converges by the ratio test (I) Converges (m) Converges for
p > 1; diverges for 0 < p s 1 3. (a) Converges to 1 /(1 - sin p) for all p E R for which sin p I < 1; that is, for
all p * (2k + 1)j, k E Z 4. (b) Since 7, ak converges, lim ak = 0. Thus there exists k, E N such that 0 s ak s 1
for all k -- k,. But then 0 5 ak s ak for all k a k and 7, ak converges by the comparison test. (d) Take ak = I/k2.
5. The series diverges for all q < 1, p E R, and converges for all q > 1, p E R. If q = 1, the series diverges for
p s 1 and converges for p > 1. 6. Suppose L, where 0 < L < oo. Take e = IL. For this e, there
exist k, E N such that #L 5 ak/bk s }L for all k z k,. The result now follows by the comparison test. 12. The proof
u++ses the fact that lim(LPY' Vk- = 1 for all n r= Z. 13. The given series is the sum of the two series
`k 1 Ijt and Mk t ?Jp, each of which converges. 16. let s, = a, + a2 + + a,,, and tk = a1 + 2a1 + +
2ka2-. By writing s, = ak + (a2 + a3) + (a4 - as + a6 + a7) + , show that if n < 22, then s s tk, and if is > 2k,
then s Z I'tk. From these two inequalities it now follows that F ak < oo if and only if I 2ka2, < oo.
1& (a) Diverges. If ak - 1/(k In k), then 2ka2. = 1/(k In 2). 19. Use Example 5.2.7 to show that ck - ck+ 1 a 0 for
all k. Thus {ck} is monotone decreasing. Use the definition of In k and the method of proof of the integral test to show
that ck =' 0 for all k. 21. Write ak+I/ak - I - xk/k where xk = (q - p)(k/(q + k + 1)). 22. (c) When p = 2,
k-fI.3...(2k- 1)12 Tk r =11`1-2,J
112
2.4... 2k ) zl-2k
( 1)a
Now use the fact that M9(1 + h)I"k = e. 23. (a) Set s, = L4., ak and let s = lim s,,. Consider the series E bk
where b, = (1< - W-771 ) and fork z 2, bk = (Vs_, - V).
EXERCISES 7.2 page 298
1. If {b,} is monotone increasing to b. consider $00- , (b - bk)ak. 2. Take bk = 1/k fork odd, and bk = i/k2 for k
even. 4. If D. = E[SiO(k
1 coos kt, then
1 r \ 1
1 sin 2 I sin ?tJ.
2 2
5. (a) Converges (c) Converges (d) Diverges; kim , , = * 0 (f) Converges (h) Converges for all
t * 2nsr, n E Z. If t - 2nar, then the series converges for p > 1 and diverges for 0 < p s 1. & Use the partial
summation formula to prove that
n n-1 n
CHAPTER 8
EXERCISES 8.1 page 322
nx _ 0, x = 0, 0, x * k,r, k E Z,
1. (a) lim 1, x>0 (c) lira (cos x)2n __
"-.o, 1, x=kar,kEZ
3. (c)
rf n = j iM
n2x d x + f
2/
(2n - n2x) dx = i + = 1 S. (a) If x = 0, fn(0) = 0 for all n e N. If x > 0,
10 0 1/n
then 0 < fn(x) < x/n, f r o m which the result follows. (b) F o r each n E N& (x) has a maximum of e ' at x = n.
6. Use the fact that forN,MEN,
N(M M N M 00 00 00
"-I M-1
an. m
M-1 R-t
an. m
M-1 R=t
a, m ( 11-1
an. m
The above inequalities hold since a0, m > 0 for all n, m E N. Now first let M - oo, and then N -b oo, to obtain
an. m an. m
n=1 mil m=1 nil
The same argument also proves the reverse inequality.
EXERCISES 8.2 page 328
2. (b) Suppose {f,} and {g,} converge uniformly to f and g respectively on E. Then I f"(x)gn(x) - f(x)g(x)1
I g"(x) I I f"(x) - f(x) I+ I f"(x) I I g" (x) - g(x) 1. By hypothesis, I S. (x) I s N for all x E E. n E N. Also, since
I f"(x) I s M for all /x E E, n E! /N, I f(x)1 fm M for all x E E. Therefore,
Now use the definition of uniform convergence of { fn} and {g,} to show that given e > 0, there exists no E N such
that I fn (x) g" (x) - f(x)g (x) ( < e for all x E E and n a n0. 4. Find Mn = max{ fn (x) : x E [0, 111, and show that
M. -+ oo. S. (a) For x E [0, a], I f"(x)1 s a". If 0 < a < 1, then lim a" = 0. Thus given e > 0, there exists
n, E N so that d` < e for all n at n,; that is I f"(x) I < e for all x e[ -w
O, a], n z n,. Therefore, f f.) converges
uniformly to 0 on [0, a] whenever a < 1.
Hints and Solutions to Selected Exercises 535
If f-Jf.1 5Jlf-fa[+2Js.
Use the uniform convergence of to f on [0, c] to finish the proof. 11. (b) To show that (19[a, b), n I6) is not
complete, it suffices to find a sequence If.) of continuous functions that converges in the norm II 1i to a Riemann
integrable function f that is not continuous.
kx)_2
6. (a) Use the comparison test to show that the given series converges for all x > 0. Let S(x) _ 7,k , (1 + and
S (x) = Mk=, (1 + kx)-2. Then S.(x) = -228k= I k(1 + kx) -3. Use the Weierstrass M-test and the comparison test to
show that the sequences {S (x)} and IS' (x)) converge uniformly on [a. oo) for every a > 0. Thus by Theorem 8.5.1.
00
kx)_3
S'(x) = limS., (x) _ -2 ;k(1 +
k-I
for all x E (a, oo). Since this holds for every a > 0, the result holds for all x E (0, oo).
EXERCISES 8.6 page 352
2. Let 91 _ {xo, x ... , be a partition of [a. a + p]. Set y1 = x, - a. Then 91* = {y y ... , is a partition
of [0, p]. If t E [x;_,, x;], then r = s + p for some s E [y; _,, y;]. Since f is periodic of period p,
f(t) = f(s + p) = f(s). Therefore, sup{f(t) : t E [x,x,]} = sup{f(s) : s E and as a consequence,
xlL(91, f) = q.L(9", f). From this it now follows that:
J.+P v
a
f=f'f
The proof for the lower integral is similar. Thus f E 9L[0,p] if and only if f E 9t[o,a + p].
4. (a) c = }(n + 1) 6. Set A,(8) = sup{Q (x) : x E [-S, S]). Then 0 < 8, < S2 implies A (S2).
Suppose tim A (S,) < oo for some 81 > 0. Then there exists a finite constant C. and n, E N such that A (S) s C for
all n ? n 0 < S 5 81. Use this fact to obtain a contradiction to the hypothesis that is an approximate identity.
(n + 1)!
Use convergence of the series
13 (n + )
IxIA.I. IxI < 1,
7 (n + l)!
to conclude that lim R0(x) = 0, - i < x 5 0. If 0 < x < 1, use Corollary 8.7.19 to show that
1.3...(n+ x fx-
n! (1-x)341-0)
Hints and Solutions to Selected Exercises 537
for some {', 0 < C < x. Now use the method of Example 8.7.20(c) to show that lim R"(x) = 0 for all x, 0 < x < 1.
Thus the series converges to (I - x)-'R for all x, I x I < I. (e) Use the fact that '
CHAPTER 9
EXERCISES 9.1 page 388
1. j' 0i = j- 1 = 2 and j 022 = j x2 = ;. Therefore, c, _ If 1, sin wx dx = 0 and c2 = 2 j' 1 x sin ax dx
Thus by Theorem 9.1.4. S2(x) _ ;x gives the best approximation in the mean to sinir x on [ -1, 1 ].
3. (a) a, 1, a2 = 1. a3 = -6 S. (c) b" = : fo x sin nx dx - - w cos mr = 2,(-1)"''. Therefore,
f(x) ^' 2 k)k+
sin kx.
k=1
6. (c) x - 2 - a E00 ( + l)2 cos(2k + 1)x 12. (a) As in the proof of Theorem 7.4.3, for A E R,
k-O
O :s II x - kyll2 = Kxil2 - 2A(x,y) + A=IIy112. If y * 0, take k = (x,y)11IYII2 to derive the inequality.
M-1 fIr
0
2 2
(b) (1) g , (11)S. Use Parseval's equality and the fact that fg = i((f +'g)2 - f2 - g2] 6. Any function
6
that is identically zero except at a finite number of points will satisfy ja f dx = 0.
IT
(c) I xI ^-
- ,rr F, (2k + 1)2
cos(2k + 1)x. (e) 1 + x -- I + 2 1 (- k sin kx
k=0
4 ao 00 1 k x
5. (a) 1,
1
a k<0 2k + 1
sin(2k + 1)x (d)
2 a k.o 2k + I
cos(2k + 1)x, -?
a"=o n LL
1 I (-1)" - cos I2 sin nx
538 Hints and Solutions to Selected Exercises
6. (c) Since h, is even, the Fourier series of h, is the cosine series of h. Therefore.
(aR r*
a0 =
o x dx + I (ir - x) dx = itc, and
l
i
iran
(a/2 a
a" = x cos nx dx + -J (in - x)cos nx dx = Z l 2cos Z + (- I)"- ' - I
0 a/2 L
Thus h,(x) ^-
c-
4
c
a k-, k2
cos 2kx. & By integration by parts, bk = -
Since f" E fit[-ir, ir], it is bounded on [-ir, ir]; i.e.,I f" (x) I <- M. Therefore, I bk I S 2M/k2. Similarly for ak.
! Z1
irk
a
, f'(.x)sin kx dx.
Thus by the Weierstrass M-test, the Fourier series off converges uniformly on [-ir, ir).
2. 90 3. To show that {sin nx}.° , is complete on [0,ir] it suffices to show that Parseval's equality holds for every
f E 9t[0,ir]; i.e.,
0°
b,2, f2(x)dz.
^' 1 o
To accomplish this, let f^ denote the odd extension off to [ -ir,ir]. Since f" is odd, a^ = 0 for all n = 0, 1, 2. .
and b^ = A fo f(x)sin nxdx. Since the orthogonal system {I,cos nx,sin is complete on [-a, ir],
pp a a
b2, =7fI J fo(x)dx = 7r
? f2(x)dx.
0
^-1 a
6. Let S^(t) = lao + 1k-1(ak cos kt + bk sin kt). If x E [-ir, irthen
1 J s a a
k-i\1
Thus by Exercise 5a and Theorem 9.4.7, lim f'. S^(t) dt = f', f(t) dt, with the convergence being uniform on
[-ir,ir]. But
S^(t) dt = 2ao(x + ir) + ± I sin kx - kl (cos kx - cos kir) .
JJa k
from which the result follows. 10. Use Lemma 9.4.8 and the Weierstrass approximation theorem.
CHAPTER 10
EXERCISES 10.2 page 443
2. Since U is open and nonempty, there exists x E U and r > 0 such that (x - r, x + r) C U. Thus by Theorem
10.2.4, m(U) z m((x - r, x + r)) = 2r > 0. 3. First show that m(P^) = 2m(P^_1) for all n E N. From this it now
follows that m(P) s (3)' for all n E N.
Hints and Solutions to Selected Exercises 539
6. First show that there exist disjoint bounded open sets U1, U2 with U, 1) K, and U2 J K2. Then
m(K,UK2)=m(U,U(.2)-m((U,UU2)\(K,UK2)).But(U,UU2)\(K,UK2)=(U,\K,)U(U2\K2).Now
use Theorem 10.2.9.
the bounded convergence theorem. S. By Theorem 10.6.10(b), fF f fE f dA + fE. f dA >- fE f dA. 7. For
each n E N, let E. = {x : f(x) > 11. Then U E, = {x : f(x) > 0). Use the previous exercise to show A(E) = 0. Now
use Theorem 10.4.5. 12. The function rp, defined in the solution to Exercise 1 satisfies I f(x) - (p,(x) I < $/n for all
x E [a, b]. Thus .{9.1 converges uniformly to f on (a, b]. 15. Suppose first that rp = XA where A is a measurable
subset of [a, b]. By Exercise 2, Section 10.4, there exists an open set U ) A such that A(U \ A) < e/2. Use the set U
to show that there exists a finite number of disjoint closed intervals {J,}.N=, such that V = U J. C U and
A(U \ V) < e/2. Let h = -X ,.
j.. Then his a step function on [a, b] and {x : h(x) * rp(x)} C (U 1 V) U (U \ A). If
rp = JJ.,a;XAJ, where the Al are disjoint measurable subsets of [a, b), approximate each XA by a step function hi that
agrees with XA, except on a set of measure less than a/n.
Since each of the sequences { fA f,} ,, i = 1, 2 are monotone increasing, they converge either to a finite number, or
diverge to oo. In either case,
JAUA
= +4-+O
J
/
j of
dA = i(
A,
f,dA + J f,dA)
A,
JfdA
A.
j.fdA.
A
If either A, or A2 is unbounded, consider the integral off over (A, U A2) n[ - n, n], and use the above. 4. For
vv
f(x) = x -P, x E (0,1), f,(x) = min{f(x), n} = x'v n- o <_ x < I Therefore,
I
I
1
p ,1vvvJ
0 o n,, n
.4
n< I rs<1PPYP
10. Let A. = An[ -n, n). By definition, fA f dA = lim fA. f dA. Since f is integrable, given e > 0, there exists is E N
such that O :C- fA fdA - fA, fdA < e. By Theorem 10.7.4(b), fA f dA - fA f dA = fA%A. f dA. Thus E = A. is the
desired set. 13. Let f, = min{ I f I. n}. Then 0 < f, s f I. Since f is integrable, by Lebesgue's dominated
convergence theorem,
lim f, dA
1A
jIfIdA.
A
Therefore, given e > 0, there exists n E N such that fA(I f I - f,)dA < e/2. In particular, if E is any measurable
subset of A.
But if A(E) < oo, fEf dA : nA(E). Hence choose S > 0 so that nS < e/2.
19. Since {f"} is monotone increasing on A, f(x) = lim f"(x) exists, either as a finite number or as oc, for every
x E A. By Fatou's lemma, fA f dA s lim fA f dA. On the other hand, since f s f for all n E N.
lim fAf" dA t5 fA f dA. Combining the two inequalities proves the result.
21. (a) Use the monotone convergence theorem. 23. Hint: I f(x)I ? (I/x)Icot I/r=1 - 2x z x ' - 2x on each of
the intervals ((2n + 3)1r)-? s x <- ((2n - 3)rr)-'.
EXERCISES 10.8 page 490
1. 0 < p < i 4. Since If + g 12 s 2(1 f 12 + I g I').f + g E 22(A). Assume II f + g 112 ;E 0. Then
which by the Cauchy-Schwarz inequality <- 11 f+ g Jz lI f lL + 11f+ g 16IIg II,. The result follows upon simplification.
7. Use the Cauchy-Schwarz inequality. 12. Let f(x) = I/ln x, x E (1, oo). Since f"(x) > 0 for all x E (1, ac),
f is convex on (1, oo) (see Miscellaneous Exercise 3, Chapter 5). Therefore f(lx + ,'-y) 5 1/(x) + ;-f(y) for all
x, y E (1, oo). Since n = 1(n - 1) + 11(n + I) the inequality follows.
Notation Index
functions. 251
I al series of real numbers, 87
12 square summable sequences, 306 1-1
[(at) 2 l2 norm, 306 sequence of functions. 318
norm. 310
euclidean norm, 309 series of functions, 319
inner product, 309
norm convergence. 334
II f II uniform norm. 334
Ii f I) 2' norm. 492 GREEK ALPHABET
..,f g. inner product in JtIa. bl, 381
alpha
E(A). Y'(A) Lebesgue integrable functions. 478 beta
12(A) square integrable functions. 484
chi
II f II2 f2 norm. 484 delta
epsilon
MEASURE AND INTEGRATION eta
91 partition of (a, 6), 208 gamma
IIphI norm of a partition, 227 iota
lower, upper sum. 209 kappa
f(p.f) Riemann sum, 226 lambda
f°ff,f lower, upper integral. 210 mu
no
fo1 f,f(x)dx Riemann integral, 211 omega
fa f(x)dv improper Riemann integral. 241 phi
'f(3,f,a), lower, upper Riemann-Stieltjes Pi
psi
3[(.9.f. a) sums. 246
sigma
f, "f da. f,, f da lower, upper. Riemann-Stieltjes tau
integral, 247 theta
f; f da Riemann-Stieltjes integral, 248 upsilon
midpoint approximation. 261 xi
T,(f) trapezoidal approximation. 266 zeta
Index
545
546 Index
Lebesgue integral, 463. 474, 478 inner, 444 Norm of partition. 227
of nonnegative function. 474 of interval. 432 Normed linear space. 310
properties of. 469.476, 479 of measurable set. 446 Nowhere differentiable function. 342
Lebesgue measure. 446 of open set. 433 Null sequence. 91
sums. 472 outer. 444 Number(s):
Lebesgue's dominated convergence zero. 219 algebraic, 43
theorem. 481 Measurable function, 456 complex, 45
Lebesguc's theorem, 220, 274 partition, 462 irrational. 3
Left continuous, 149 set. 446 natural. 2
derivative. 168 Mengoli. Pietro, 313 rational. 3
limit, 148 Mercator. Nicolaus. 317 real. 3
Legendre polynomials, 389 Mesh, of partition, 227
Leibniz, Gottfried. 115, 165. 207. 276 Metric. 111. 311 0
Leibniz's rule. 175 space, I I I Odd extension. 402
:Hospital, Marquis de. 190 Midpoint approximation, 261 function. 229. 397
Minimum: One-to-one function, II
L' l lospital's rule. 191, 194
absolute. 176 Onto function. 7
Limit:
Open:
of a function. 116 local, 176
cover. 102
inferior, 74, 162 Minkowski's inequality, 308, 485
in, 99
infinite. 65. 190 Modified principle of mathematical
interval, 25
at infinity. 126, 190 induction, 17
Open set. 94, 111
left (right). 148 Monotone convergence theorem, 483
measure of. 433
of a sequence, 50 Monotone function. 152
Order properties of S2, 21
of a sequence of functions, 318 integrability of, 216. 250
Ordered field, 21
subsequential, 67 Monotone sequence. 60
Ordered pair, 5
superior. 74. 162 convergence of. 61
equality of. 5
Limit comparison test, 282 Oresme. Nicole. 88, 313
Limit of sum, product. quotient: N Orthogonal. 380
of functions. 123 nth derivative, 168
functions. 381
of sequences, 53 nth partial sum, 87. 280, 319
Orthonormal functions. 383
Limit point, 69, 97 nth root, 29
Oscillation of a function. 162
Linear approximation. 361 nth term. 37, 50. 87 Outer measure. 444
Lipschitz condition. 145 Natural exponential function. 186, 238
function, 145 logarithm function, 234 P
Local maximum (minimum), 176 Natural numbers, 2 p-series. 285
Logarithm, natural, 234 Negation, 496 Pairwise disjoint, 98, 433
Logically valid, 501 of quantified sentence. 518 Parseval's equality. 392, 412. 491
Lower bound, 21 Neighborhood. 49. 311 Partial sum. 87, 280. 319
integral. 210, 247 Nested intervals property. 61 Partition. 208
sum, 209, 246,462 Newton, lssac, 115. 165. 197, 207, 276. measurable. 462
313. 318 norm (mesh) of. 227
M Newton's method. 197 refinement of, 210, 464
Maclaurin. Colin. 318 Nondecreasing (increasing): Periodic extension, 396
Maclaurin series, 361 function, 152 function, 148. 346
Mapping. See Function sequence.60 Piecewise continuous function, 419
Mathematical induction, 15, 17 Nonmeasurdble set. 493 Point:
Maximum: Nonnegative integers. 3 interior, 94
absolute, 176 Nonterminating expansion, 33 isolated. 69
local, 176 Norm. 310 limit, 69
Maximum element, 22 of integrable function. 492 Pointwise convergence. 318. 319
Mean-square convergence. 390 of square integrable function, 484 of Fourier series. 415. 417. 420
Mean value theorem, 179, 181 of square summable sequence, 306 Polynomial. 124
for integrals, 235.252 uniform. 334 degree of, 124
Measure: of vector, 112, 309 Positive integers, 2
of compact set, 440 Norm convergence. 311, 334. 390 rational numbers. 21
Index 549