Introduction to Mathematical Analysis

Introduction
to Mathematical
Analysis

MTH 311

MTH 311
Introduction to Mathematical Analysis

Lafferriere, Lafferriere, Nam

Beatriz Lafferriere
Gerardo Lafferriere
Nguyen Mau Nam

2015 Portland State University

ISBN: 978-1-312-74284-0
This work is licensed under a
Creative Commons Attribution-NonCommercial 4.0 International License
Published by Portland State University Library
Portland, OR 97207-1151
Cover design by Matt Eich

Contents

TOOLS FOR ANALYSIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.1

BASIC CONCEPTS OF SET THEORY

1.2

FUNCTIONS

10

1.3

THE NATURAL NUMBERS AND MATHEMATICAL INDUCTION

13

1.4

ORDERED FIELD AXIOMS

16

1.5

THE COMPLETENESS AXIOM FOR THE REAL NUMBERS

19

1.6

APPLICATIONS OF THE COMPLETENESS AXIOM

22

SEQUENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.1

CONVERGENCE

27

2.2

LIMIT THEOREMS

32

2.3

MONOTONE SEQUENCES

35

2.4

THE BOLZANO-WEIERSTRASS THEOREM

40

2.5

LIMIT SUPERIOR AND LIMIT INFERIOR

43

2.6

OPEN SETS, CLOSED SETS, AND LIMIT POINTS

47

LIMITS AND CONTINUITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1

LIMITS OF FUNCTIONS

53

3.2

LIMIT THEOREMS

56

3.3

CONTINUITY

65

3.4

PROPERTIES OF CONTINUOUS FUNCTIONS

70

3.5

UNIFORM CONTINUITY

75

3.6

LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

78

DIFFERENTIATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.1

DEFINITION AND BASIC PROPERTIES OF THE DERIVATIVE

85

4.2

THE MEAN VALUE THEOREM

90

4.3

SOME APPLICATIONS OF THE MEAN VALUE THEOREM

96

4.4

LHOSPITALS RULE

98

4.5

TAYLORS THEOREM

101

4.6

CONVEX FUNCTIONS AND DERIVATIVES

104

4.7

NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

109

Solutions and Hints for Selected Exercises . . . . . . . . . . . . . . . . . . . . . . 119

Preface

Our goal in this set of lecture notes is to provide students with a strong foundation in mathematical
analysis. Such a foundation is crucial for future study of deeper topics of analysis. Students should be
familiar with most of the concepts presented here after completing the calculus sequence. However,
these concepts will be reinforced through rigorous proofs.
The lecture notes contain topics of real analysis usually covered in a 10-week course: the
completeness axiom, sequences and convergence, continuity, and differentiation. The lecture notes
also contain many well-selected exercises of various levels. Although these topics are written in a
more abstract way compared with those available in some textbooks, teachers can choose to simplify
them depending on the background of the students. For instance, rather than introducing the topology
of the real line to students, related topological concepts can be replaced by more familiar concepts
such as open and closed intervals. Some other topics such as lower and upper semicontinuity,
differentiation of convex functions, and generalized differentiation of non-differentiable convex
functions can be used as optional mathematical projects. In this way, the lecture notes are suitable
for teaching students of different backgrounds.
Hints and solutions to selected exercises are collected in Chapter 5. For each section, there is at
least one exercise fully solved. For those exercises, in addition to the solutions, there are explanations
about the process itself and examples of more general problems where the same technique may be
used. Exercises with solutions are indicated by a I and those with hints are indicated by a B.
Finally, to make it easier for students to navigate the text, the electronic version of these notes
contains many hyperlinks that students can click on to go to a definition, theorem, example, or
exercise at a different place in the notes. These hyperlinks can be easily recognized because the text
or number is on a different color and the mouse pointer changes shape when going over them.

BASIC CONCEPTS OF SET THEORY

FUNCTIONS
THE NATURAL NUMBERS AND MATHEMATICAL INDUCTION
ORDERED FIELD AXIOMS
THE COMPLETENESS AXIOM FOR THE REAL NUMBERS
APPLICATIONS OF THE COMPLETENESS AXIOM

1. TOOLS FOR ANALYSIS

This chapter discusses various mathematical concepts and constructions which are central to the
study of the many fundamental results in analysis. Generalities are kept to a minimum in order to
move quickly to the heart of analysis: the structure of the real number system and the notion of limit.
The reader should consult the bibliographical references for more details.

1.1

BASIC CONCEPTS OF SET THEORY

Intuitively, a set is a collection of objects with certain properties. The objects in a set are called
the elements or members of the set. We usually use uppercase letters to denote sets and lowercase
letters to denote elements of sets. If a is an element of a set A, we write a A. If a is not an element
of a set A, we write a 6 A. To specify a set, we can list all of its elements, if possible, or we can use
a defining rule. For instance, to specify the fact that a set A contains four elements a, b, c, d, we write
A = {a, b, c, d}.
To describe the set E containing all even integers, we write
E = {x : x = 2k for some integer k}.
We say that a set A is a subset of a set B if every element of A is also an element of B, and write
A B.
Two sets are equal if they contain the same elements. If A and B are equal, we write A = B. The
following result is straightforward and very convenient for proving equality between sets.
Theorem 1.1.1 Two sets A and B are equal if and only if A B and B A.
If A B and A does not equal B, we say that A is a proper subset of B, and write
A ( B.

1.1 BASIC CONCEPTS OF SET THEORY

The set 0/ = {x : x 6= x} is called the empty set. This set clearly has no elements. Using
Theorem 1.1.1, it is easy to show that all sets with no elements are equal. Thus, we refer to the empty
set.
Throughout this book, we will discuss several sets of numbers which should be familiar to the
reader:
N = {1, 2, 3, . . .}, the set of natural numbers or positive integers.
Z = {0, 1, 1, 2, 2, . . .}, the set of integers (that is, the natural numbers together with zero
and the negative of each natural number).
Q = {m/n : m, n Z, n 6= 0}, the set of rational numbers.
R, the set of real numbers.
Intervals. For a, b R, we have
[a, b] = {x R : a x b},
(a, b] = {x R : a < x b},
[a, ) = {x R : a x},
(a, ) = {x R : a < x},
and similar definitions for (a, b), [a, b), (, b], and (, b). We will say more about the
symbols and in Section 1.5.
Since the real numbers are central to the study of analysis, we will discuss them in great detail in
Sections 1.4, 1.5, and 1.6.
For two sets A and B, the union, intersection, difference, and symmetric difference of A and B are
given respectively by
A B = {x : x A or x B},
A B = {x : x A and x B},
A \ B = {x : x A and x
/ B}, and
AB = (A \ B) (B \ A).
If A B = 0,
/ we say that A and B are disjoint.
The difference of A and B is also called the complement of B in A. If X is a universal set, that is,
a set containing all the objects under consideration, then the complement of A in X is denoted simply
by Ac .
Theorem 1.1.2 Let A, B, and C be subsets of a universal set X. Then the following hold:

(1) A Ac = X;
(2) A Ac = 0;
/
(3) (Ac )c = A;
(4) (Distributive law) A (B C) = (A B) (A C);
(5) (Distributive law) A (B C) = (A B) (A C);
(6) (DeMorgans law) A \ (B C) = (A \ B) (A \C);
(7) (DeMorgans law) A \ (B C) = (A \ B) (A \C);
(8) A \ B = A Bc .
Proof: We prove some of the results and leave the rest for the exercises.
(1) Clearly, A Ac X since both A and Ac are subsets of X. Now let x X. Then either x is an
element of A or it is not an element of A. In the first case, x A and, so, x A Ac . In the second
case, x Ac and, so, x A Ac . Thus, X A Ac .

9
(2) No element of x can be simultaneously in A and not in A. Thus, A Ac = 0.
/
(4) Let x A (B C). Then x A and x B C. Therefore, x B or x C. In the first
case, since x is also in A we get x A B and, hence, x (A B) (A C). In the second case,
x A C and, hence, x (A B) (A C). Thus, in all cases, x (A B) (A C). This shows
A (B C) (A B) (A C).
Now we prove the other inclusion. Let x (A B) (A C). Then x A B or x A C.
In either case, x A. In the first case, x B and, hence, x B C. It follows in this case that
x A (B C). In the second case, x C and, hence, x B C. Again, we conclude x A (B C).
Therefore, (A B) (A C) A (B C) as desired. 
A set whose elements are sets is often called a collection/family of sets and is often denoted by
script letters such as A or B.
Let I be a nonempty set such that to each i I corresponds a set Ai . Then the family of all sets
Ai as i ranges over I is denoted by
{Ai : i I}.
Such a family of sets is called an indexed family and the set I is called the index set. Consider the
indexed family of sets {Ai : i I}. The union and intersection of this family as i ranges over I is
defined respectively by
[

Ai = {x : x Ai for some i I}

iI

and
\

Ai = {x : x Ai for every i I}.

iI

Theorem 1.1.3 Let {Ai : i I} be an indexed family of subsets of a universal set X and let B be a

subset of X. Then the following hold:

T
T
(1) B iI Ai
= iI B Ai ;
S
S
(2) B iI Ai
= iI B Ai ;
T
S
(3) B \ iI Ai
= iI B \ Ai .
S
T
(4) B \ iI
Ai = iI B \ Ai ;
T
S
c
(5)
A =
Ac ;
SiI i
c TiI c
(6)
= iI A .
iI Ai
We want to consider pairs of objects in which the order matters. Given objects a and b, we will
denote by (a, b) the ordered pair where a is the first element and b is the second element. The main
characteristic of ordered pairs is that (a, b) = (c, d) if and only if a = c and b = d. Thus, the ordered
pair (0, 1) represents a different object than the pair (1, 0) (while the set {0, 1} is the same as the set
{1, 0})1 .
Given two sets A and B, the Cartesian product of A and B is the set defined by
A B := {(a, b) : a A and b B}.
We will make use of cartesian products in the next section when we discuss functions.
1 For

a precise definition of ordered pair in terms of sets see [Lay13]

1.2 FUNCTIONS

10

Exercises
1.1.1 Prove the remaining items in Theorem 1.1.2.
1.1.2 I Let Y and Z be subsets of X. Prove that

(X \Y ) Z = Z \ (Y Z).
1.1.3 Let A X and B Y . Determine if the following equalities are true and justify your answer:

(a) (X Y ) \ (A B) = (X \ A) (Y \ B).
(b) (X Y ) \ (A B) = [(X \ A) Y ] [X (Y \ B)].

1.2

FUNCTIONS
Definition 1.2.1 Let X and Y be sets. A function from X into Y is a subset f X Y with the
following properties
(a) For all x X there is y Y such that (x, y) f .
(b) If (x, y) f and (x, z) f , then y = z.
The set X is called the domain of f , the set Y is called the codomain of f , and we write f : X Y .
The range of f is the subset of Y defined by {y Y : there is x X such that (x, y) f }.

It follows from the definition that for each x X, there is exactly one element y Y such that
(x, y) f . We will write y = f (x). If x X, the element f (x) is called the value of f at x or the
image of x under f .
Note that, in this definition, a function is a collection of ordered pairs and, thus, corresponds
to the geometric interpretation of the graph of a function given in calculus. In fact, we will refer
indistinctly to the function f or to the graph of f . Both refer to the set {(x, f (x)) : x X}.
Let f : X Y and g : X Y be two functions. Then the two functions are equal if they are
equal as subsets of X Y . It is easy to see that f equals g if and only if
f (x) = g(x) for all x X.
It follows from the definition that two equal functions must have the same domain.
Let f : X Y be a function and let A be a subset of X. The restriction of f on A, denoted by f|A ,
is a new function from A into Y given by
f|A (a) = f (a) for all a A.
Definition 1.2.2 A function f : X Y is called surjective (or is said to map X onto Y ) if for every

element y Y , there exists an element x X such that f (x) = y.

The function f is called injective (or one-to-one) if for each pair of distinct elements of X, their
images under f are also distinct. Thus, f is one-to-one if and only if for all x and x0 in X, the
following implication holds:
[ f (x) = f (x0 )] [x = x0 ].
If f is both surjective and injective, it is called bijective or a one-to-one correspondence. In this case,
for any y Y , there exists a unique element x X such that f (x) = y. This element x is then denoted
by f 1 (y). In this way, we already built a function from Y to X called the inverse of f .

11
Theorem 1.2.1 Let f : X Y . If there are two functions g : Y X and h : Y X such that

g( f (x)) = x for every x X and f (h(y)) = y for every y Y , then f is bijective and g = h = f 1 .
Proof: First we prove that f is surjective. Let y Y and set x = h(y). Then, from the assumption on
h, we have f (x) = f (h(y)) = y. This shows that f is surjective.
Next we prove that f is injective. Let x, x0 X be such that f (x) = f (x0 ). Then x = g( f (x)) =
g( f (x0 )) = x0 . Thus, f is injective.
We have shown that for each y Y , there is a unique x X, which we denote f 1 (y) such that
f (x) = y. Since for such a y, g(y) = g( f (x)) = x, we obtain g(y) = f 1 (y). Since f (h(y)) = y, we
also conclude that h(y) = x = f 1 (y). 
Definition 1.2.3 Let f : X Y be a function and let A be a subset of X. Then the image of A under

f is given by
f (A) = { f (a) : a A}.
It follows from the definition that
f (A) = {b Y : b = f (a) for some a A}.
Moreover, f is surjective if and only if f (X) = Y .
For a subset B of Y , the preimage of B under f is defined by
f 1 (B) = {x X : f (x) B}.
Remark 1.2.2 Note that, despite the notation, the definition of preimage does not require the

function to have an inverse. It does not even require the function to be injective. The examples below
illustrate these concepts.
 Example 1.2.1 Let f : R R be given by f (x) = 3x 1. Let A = [0, 2) and B = {1, 4, 5}.
Then f (A) = [1, 5) and f 1 (B) = { 23 , 1, 2}.

Example 1.2.2 Let f : R R be given by f (x) = x + 7. Let A = [0, 2) and B = (, 3]. Then
f (A) = (5, 7] and f 1 (B) = [4, ).

Example 1.2.3 Let f : R R be given by f (x) = x2 . Let A = (1, 2) and B = [1, 4). Then

f (A) = [0, 4) and f 1 (B) = (2, 1] [1, 2).

Theorem 1.2.3 Let f : X Y be a function, let A be a subset of X, and let B be a subset of Y . The

following hold:
(1) A f 1 ( f (A)).
(2) f ( f 1 (B)) B.
Proof: We prove (1) and leave (2) as an exercise.
(1) Let x A. By the definition of image, f (x) f (A). Now, by the definition of preimage,
x f 1 ( f (A)). 
Theorem 1.2.4 Let f : X Y be a function, let A, B X, and let C, D Y . The following hold:

(1) If C D, then f 1 (C) f 1 (D);

(2) f 1 (D \C) = f 1 (D) \ f 1 (C);
(3) If A B, then f (A) f (B);
(4) f (A \ B) f (A) \ f (B).

1.2 FUNCTIONS

12

Proof: We prove (2) and leave the other parts as an exercise.

(2) We prove first that f 1 (D \ C) f 1 (D) \ f 1 (C). Let x f 1 (D \ C). Then, from the
definition of inverse image, we get f (x) D \C. Thus, f (x) D and f (x) 6 C. Hence x f 1 (D)
and x 6 f 1 (C). We conclude that x f 1 (D) \ f 1 (C).
Next we prove f 1 (D) \ f 1 (C) f 1 (D \C). Let x f 1 (D) \ f 1 (C). Thus, x f 1 (D) and
x 6 f 1 (C). Therefore, f (x) D and f (x) 6 C. This means f (x) D \C and, so, x f 1 (D \C). 
Theorem 1.2.5 Let f : X Y be a function, let {A }I be an indexed family of subsets of X, and

let {B } J be an indexed family of subsets of Y . The following hold:

S

(1) f ( I A ) = I f (A );
T
T
(2) f ( I A ) I f (A );
S
S
(3) f 1 ( J B ) = J f 1 (B );
T
T
(4) f 1 ( J B ) = J f 1 (B ).
Proof: We prove (1) and leave the other parts as an exercise.
S
S
(1) Let y f ( I A ). From the definition of image of a set, there is x I A such that
y = f (x). From the definition of union of a family of sets, there is 0 I such that x A0 . Therefore,
S
y = f (x) f (A0 ) and, so, y I f (A ). .
Given functions f : X Y and g : Y Z, we define the composition function g f of f and g
as the function g f : X Z given by
(g f )(x) = g( f (x)) for all x X.
Theorem 1.2.6 Let f : X Y and g : Y Z be two functions and let B Z. The following hold:

(1) (g f )1 (B) = f 1 (g1 (B));

(2) If f and g are injective, then g f is injective;
(3) If f and g are surjective, then g f is surjective;
(4) If g f is injective, then f is injective;
(5) If g f is surjective, then g is surjective.
Proof: We prove (4) and leave the other parts as an exercise.
(4) Suppose g f is injective and let x, x0 X be such that f (x) = f (x0 ). Then (g f )(x) =
g( f (x)) = g( f (x0 )) = (g f )(x0 ). Since g f is injective, it follows that x = x0 . We conclude that f
is injective. 
Definition 1.2.4 A sequence of elements of a set A is a function with domain N and codomain A.

We discuss sequences in detail in Chapter 2.

Definition 1.2.5 We say that set A is finite if it is empty or if there exists a natural number n and a

one-to-one correspondence f : A {1, 2, . . . , n}. A set is infinite if it is not finite.

We leave it as an exercise to prove that the union of two finite sets is finite. It is also easy to show,
by contradiction, that N is infinite. The following result will be useful when studying sequences and
accumulation points.
Theorem 1.2.7 Suppose A is an infinite set. Then there exists a one-to-one function f : N A.
Proof: Let A be an infinite set. We define f as follows. Choose any element a1 A and set f (1) = a1 .
Now the set A \ {a1 } is again infinite (otherwise A = {a} (A \ {a1 }) would be the union of two
finite sets). So we may choose a2 A with a2 6= a1 and we define f (2) = a2 2 . Having defined
f (1), . . . , f (k), we choose ak+1 A such that ak+1 A \ {a1 , . . . , ak } and define f (k + 1) = ak+1
2 This

fact relies on a basic axiom of set theory called the Axiom of Choice. See [Lay13] for more details.

13
(such an ak+1 exists because A \ {a1 , . . . , ak } is infinite and, so, nonempty). The function f so defined
clearly has the desired properties. 
To paraphrase, the previous theorem says that in every infinite set we can find a sequence made
up of all distinct points.

Exercises
1.2.1 I Let f : X Y be a function. Prove that:

(a) If f is one-to-one, then A = f 1 ( f (A)) for every subset A of X.

(b) If f is onto, then f ( f 1 (B)) = B for every subset B of Y .
1.2.2 Prove that if f : X Y is injective, then the following hold:

(a) f (A B) = f (A) f (B) for A, B X.

(b) f (A \ B) = f (A) \ f (B) for A, B X.
1.2.3 Prove part (2) of Theorem 1.2.3.
1.2.4 Prove parts (1), (3), and (4) of Theorem 1.2.4.
1.2.5 Prove parts (2), (3), and (4) of Theorem 1.2.5.
1.2.6 Prove parts (1), (2), (3), and (5) of Theorem 1.2.6.
1.2.7 Prove that the union of two finite sets is finite. Hint: it is easier to show when the sets are
disjoint.

1.3

THE NATURAL NUMBERS AND MATHEMATICAL INDUCTION

We will assume familiarity with the set N of natural numbers, with the usual arithmetic operations
of addition and multiplication on N, and with the notion of what it means for one natural number to
be less than another.
In addition, we will also assume the following property of the natural numbers.
Well-Ordering Property of the Natural Numbers: If A is a nonempty subset of N, then there
exists an element ` A such that ` x for all x A.
To paraphrase the previous property, every nonempty subset of positive integers has a smallest
element.
The principle of mathematical induction is a useful tool for proving facts about sequences.
Theorem 1.3.1 Principle of Mathematical Induction. For each natural number n N, suppose
that P(n) denotes a proposition which is either true or false. Let A = {n N : P(n) is true}. Suppose
the following conditions hold:
(a) 1 A.
(b) For each k N, if k A, then k + 1 A.

1.3 THE NATURAL NUMBERS AND MATHEMATICAL INDUCTION

14

Then A = N.
Proof: Suppose conditions (a) and (b) hold. Assume, by way of contradiction, that A 6= N. Set
B = N \A, that is, B = {n N : P(n) is false}. Then B is a nonempty subset of N. By the WellOrdering Property of the natural numbers, there exists a smallest element ` B. By condition (a),
1 6 B. Hence, ` 2. It follows that k = ` 1 is a natural number. Since k < `, k 6 B and, hence, we
have that P(k) is true. By condition (b), we obtain that P(k + 1) is true. But k + 1 = `, and P(`) is
false, since ` B. This is a contradiction, so the conclusion follows. 
To paraphrase, the principle says that, given a list of propositions P(n), one for each n N, if
P(1) is true and, moreover, P(k + 1) is true whenever P(k) is true, then all propositions are true.
We will refer to this principle as mathematical induction or simply induction. Condition (a) above
is called the base case and condition (b) the inductive step. When proving (b), the statement P(k) is
called the inductive hypothesis.


Example 1.3.1 Prove using induction that for all n N

1+2++n =

n(n + 1)
.
2

1(1+1)
The statement P(n) is the equality 1+2+ +n = n(n+1)
2 . Now the base case says that 1 =
2 ,
which is clearly true.
Suppose P(k) is true for some k N. That is, suppose that 1 + 2 + + k = k(k+1)
(this is the
2
inductive hypothesis). Now we have

1 + 2 + + k + (k + 1) =

k(k + 1)
k(k + 1) + 2(k + 1) (k + 1)(k + 2)
+ (k + 1) =
=
.
2
2
2

This shows that P(k + 1) is true. We have now proved conditions (a) and (b) of Theorem 1.3.1.
Therefore, by the principle of mathematical induction we conclude that
1+2++n =


n(n + 1)
for all n N .
2

Example 1.3.2 Prove using induction that for all n N, 7n 2n is divisible by 5.

For n = 1, we have 7 2 = 5, which is clearly a multiple of 5.

Suppose that 7k 2k is a multiple of 5 for some k N. That is, there is an integer j such that
7k 2k = 5 j. Let us write 7k = 2k + 5 j. Now, substituting this expression below, we have
7k+1 2k+1 = 77k 22k = 7(2k +5 j)22k = 72k 22k +75 j = 2k (72)+57 j = 5(2k +7 j).
It follows that 7k+1 2k+1 is a multiple of 5. This proves the inductive step.
We conclude by induction that 7n 2n is divisible by 5 for all n N.


Example 1.3.3 Prove using induction that for all n N

n + 1 2n
For n = 1, we have 1 + 1 = 2 = 21 , so the base case is true.
Suppose next that k + 1 2k for some k N. Then k + 1 + 1 2k + 1. Since 2k is a positive
integer, we also have 1 2k . Therefore,
(k + 1) + 1 2k + 1 2k + 2k = 2 2k = 2k+1 .

15
We conclude by the principle of mathematical induction that n + 1 2n for all n N.
The following result is known as the Generalized Principle of Mathematical Induction. It simply
states that we can start the induction process at any integer n0 , and then we obtain the truth of all
statements P(n) for n n0 .
Theorem 1.3.2 Generalized Principle of Mathematical Induction. Let n0 N and for each
natural n n0 , suppose that P(n) denotes a proposition which is either true or false. Let A = {n
N : P(n) is true}. Suppose the following two conditions hold:
(a) n0 A.
(b) For each k N, k n0 , if k A, then k + 1 A.
Then A {k N : k n0 }.
Proof: Suppose conditions (a) and (b) hold. Assume, by way of contradiction, that A 6 {k N :
k n0 }. Set B = {n N : n n0 , P(n) is false}. Then B is a nonempty subset of N. By the WellOrdering Property of the natural numbers, there exists a smallest element ` B. By condition (a),
n0 6 B. Hence, ` n0 + 1. It follows that k = ` 1 n0 . Since k < `, k 6 B and, so, we have that
P(k) is true. By condition (b), we obtain that P(k + 1) is true. But k + 1 = `, and P(`) is false, since
` B. This is a contradiction, so the conclusion follows. 


Example 1.3.4 Prove by induction that 3n < 2n for all n 4.

The statement is true for n = 4 since 12 < 16. Suppose next that 3k < 2k for some k N, k 4.
Now,
3(k + 1) = 3k + 3 < 2k + 3 < 2k + 2k = 2k+1 ,

(1.1)

where the second inequality follows since k 4 and, so, 2k 16 > 3. This shows that P(k + 1) is
true. Thus, by the generalized principle of induction, the inequality holds for all n 4.
Next we present another variant of the induction principle which makes it easier to prove the
inductive step. Despite its name, this principle is equivalent to the standard one.
Theorem 1.3.3 Principle of Strong Induction. For each natural n N, suppose that P(n) denotes
a proposition which is either true or false. Let A = {n N : P(n) is true}. Suppose the following
two conditions hold:
(a) 1 A.
(b) For each k N, if 1, 2, . . . , k A, then k + 1 A.
Then A = N.
Remark 1.3.4 Note that the inductive step above says that, in order to prove P(k + 1) is true, we

may assume not only that P(k) is true, but also that P(1), P(2),. . . ,P(k 1) are true.
There is also a generalized version of this theorem where the base case is for some integer n0 > 1.
Example 1.3.5 Prove by induction that every positive integer greater than 1 is either a prime
number or a product of prime numbers.
Clearly, the statement is true for n = 2. Suppose the statement holds for any positive integer
m {2, . . . , k}, where k N, k 2. If k + 1 is prime, the statement holds for k + 1. Otherwise,
there are positive integers p, q > 1 such that k + 1 = pq. Since p, q k, by the inductive assumption
applied to both p and q we can find prime numbers r1 ,. . . ,r` and s1 ,. . . ,sm such that p = r1 r` and
q = s1 sm (note that ` and m may both equal 1). But then


k + 1 = r1 r` s1 sm .

(1.2)

1.4 ORDERED FIELD AXIOMS

16

Thus, the statement holds true for k + 1. The conclusion now follows by the Principle of Strong
Induction.

Exercises
1.3.1 Prove the following using induction:

(a) 12 + 22 + + n2 =

n(n+1)(2n+1)
6

(b) 13 + 23 + + n3 =

n2 (n+1)2
4

for all n N.

for all n N.

(c) 1 + 3 + + (2n + 1) = (n + 1)2 for all n N.

1.3.2 Given a real number a 6= 1, prove that

1 + a + a2 + + an =

1 an+1
for all n N.
1a

1.3.3 I The Fibonacci sequence is defined by a1 = a2 = 1 and

an+2 = an+1 + an for n 1.

Prove that

1 h 1 + 5 n  1 5 n i

.
an =
2
2
5

1.3.4 Let a 1. Prove by induction that

(1 + a)n 1 + na for all n N .

1.3.5 B Let a, b R and n N. Use Mathematical Induction to prove the binomial theorem
n

 
n k nk
(a + b) =
ab ,
k=0 k
n

where

1.4

 
n
n!
=
.
k
k!(n k)!

ORDERED FIELD AXIOMS

In this book, we will start from an axiomatic presentation of the real numbers. That is, we will
assume that there exists a set, denoted by R, satisfying the ordered field axioms stated below together
with the completeness axiom presented in the next section.
Since there are many examples of ordered fields, we define the concept more generally.
A field F is a set with two binary operations + and that satisfy the following field axioms:
(1a) (x + y) + z = x + (y + z) for all x, y, z F.
(1b) x + y = y + x for all x, y F.
(1c) There exists an element 0 F such that x + 0 = x for all x F.
(1d) For each x F, there exists an element x F such that x + (x) = 0.

17
(2a) (x y) z = x (y z) for all x, y, z F.
(2b) x y = y x for all x, y F.
(2c) There exists an element 1 F such that 1 6= 0 and x 1 = x for all x F.
(2d) For each x F \ {0}, there exists an element x1 F such that x (x1 ) = 1. (We also write 1/x
instead of x1 .)
(2e) x (y + z) = x y + x z for all x, y, z F.
We often write xy instead of x y.
An ordered field F is a field on which, in addition, there is a relation < that satisfies the order
axioms below:
(3a) For all x, y F, exactly one of the three relations holds: x = y, y < x, or x < y.
(3b) For all x, y, z F, if x < y and y < z, then x < z.
(3c) For all x, y, z F, if x < y, then x + z < y + z.
(3d) For all x, y, z F, if x < y and 0 < z, then xz < yz.
We will use the notation x y to mean x < y or x = y. We may also use the notation x > y to
represent y < x and the notation x y to mean x > y or x = y.
An example of an ordered field is the set of rational numbers Q with the familiar operations and
order. The integers Z do not form a field since for an integer m other than 1 or 1, its reciprocal 1/m
is not an integer and, thus, axiom 2(d) above does not hold. In particular, the set of positive integers
N does not form a field either. As mentioned above the real numbers R will be defined as the ordered
field which satisfies one additional property described in the next section: the completeness axiom.
Note that we can assume that the set of all natural numbers is a subset of any ordered field by
identifying the 1 in N with the 1 in axiom (2c) above, the number 2 with 1 + 1, 3 with 1 + 1 + 1, etc.
In a similar way, can include Z and Q as subsets.
We say that a real number x is irrational if x R \ Q, that is, if it is not rational.
Under these axioms, many familiar properties of R can be derived. Some examples are given in
the next theorem. These hold for any ordered field.
Theorem 1.4.1 Let F be an ordered field. For a, b, c F, the following hold:
(1) If a + b = a + c, then b = c;
(2) (a) = a;
(3) If a 6= 0 and ab = ac, then b = c;
(4) If a 6= 0, then 1/(1/a) = a.
Proof: (1) Suppose a + b = a + c. By axiom (1d), we have
(a) + (a + b) = (a) + (a + c).
Then axiom (1a) gives
[(a) + a] + b = [(a) + a] + c.
Thus, 0 + b = 0 + c and, hence, b = c.
(2) Since (a) + a = 0, we have (a) = a.
The proofs of (3) and (4) are similar. 
For any element of an ordered field F, we can define its absolute value.

1.4 ORDERED FIELD AXIOMS

18

Definition 1.4.1 Given an ordered field F and x F, define the absolute value of x by

(
x,
if x 0;
|x| =
x, if x < 0.

Figure 1.1: The absolute value function.

The following properties of absolute value follow directly from the definition.
Proposition 1.4.2 Let x, y, M R and suppose M > 0. The following properties hold:

(a)
(b)
(c)
(d)

|x| 0;
| x| = |x|;
|xy| = |x||y|;
|x| < M if and only if M < x < M. (The same holds if < is replaced with .)

Proof: We prove (d) and leave the other parts as an exercise.

(d) Suppose |x| < M. In particular, this implies M > 0. We consider the two cases separately:
x 0 and x < 0. Suppose first x 0. Then |x| = x and, hence, M < 0 x = |x| < M. Now suppose
x < 0. Then |x| = x. Therefore, x < M and, so x > M. It follows that M < x < 0 < M.
For the converse, suppose M < x < M. Again, we consider different cases. If x 0, then
|x| = x < M as desired. Next suppose x < 0. Now, M < x implies M > x. Then |x| = x < M. 
Note that as a consequence of part (d) above, since |x| |x| we get |x| x |x|.
The next theorem will play an important role in the study of limits.
Theorem 1.4.3 Triangle Inequality. Given x, y R,
|x + y| |x| + |y|.
Proof: From the observation above, we have
|x| x |x|

19
|y| y |y|.
Adding up the inequalities gives
|x| |y| x + y |x| + |y|.
Since |x| |y| = (|x| + |y|), the conclusion follows from Proposition 1.4.2 (d). 
Corollary 1.4.4 For any x, y R,

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

Remark 1.4.5 The absolute value has a geometric interpretation when considering the numbers in

an ordered field as points on a line. The number |a| denotes the distance from the number a to 0.
More generally, the number d(a, b) = |a b| is the distance between the points a and b. It follows
easily from Proposition 1.4.2 that d(x, y) 0, and d(x, y) = 0 if and only if x = y. Moreover, the
triangle inequality implies that
d(x, y) d(x, z) + d(z, y),
for all numbers x, y, z.

Exercises
1.4.1 Prove that n is an even integer if and only if n2 is an even integer. (Hint: prove the if part

by contraposition, that is, prove that if n is odd, then n2 is odd.)

1.4.2 Prove parts (a), (b), and (c) of Proposition 1.4.2.
1.4.3 I Prove Corollary 1.4.4.
1.4.4 Given two real numbers x and y, prove that

max{x, y} =

1.5

x + y + |x y|
x + y |x y|
and min{x, y} =
.
2
2

THE COMPLETENESS AXIOM FOR THE REAL NUMBERS

There are many examples of ordered fields. However, we are interested in the field of real
numbers. There is an additional axiom that will distinguish this ordered field from all others. In
order to introduce our last axiom for the real numbers, we first need some definitions.
Definition 1.5.1 Let A be a subset of R. A number M is called an upper bound of A if

x M for all x A.
If A has an upper bound, then A is said to be bounded above.
Similarly, a number L is a lower bound of A if
L x for all x A,
and A is said to be bounded below if it has a lower bound. We also say that A is bounded if it is both
bounded above and bounded below.

1.5 THE COMPLETENESS AXIOM FOR THE REAL NUMBERS

20

It follows that a set A is bounded if and only if there exist M R such that |x| M for all x A
(see Exercise 1.5.1).
Definition 1.5.2 Let A be a nonempty set that is bounded above. We call a number a least upper
bound or supremum of A, if
(1) x for all x A (that is, is an upper bound of A);
(2) If M is an upper bound of A, then M (this means is smallest among all upper bounds).


Example 1.5.1

(a) sup[0, 3) = sup[0, 3] = 3.

(b) sup{3,
10.
 5, 7,n8, 10} = 
1
(1)
:nN = .
(c) sup
n
2
1
(d) sup{1 : n N} = 1.
n
(e) sup{x2 : 2 < x < 1, x R} = 4.
It is easy to see that if A has a supremum, then it has only one (see Exercise 1.5.2). In this case,
we denote such a number by sup A.
The following proposition is very useful in working with suprema.
Proposition 1.5.1 Let A be a nonempty subset of R that is bounded above. Then = sup A if and

only if
(1) x for all x A;
(2) For any > 0, there exists a A such that < a.
Proof: Suppose first that = sup A. Then clearly (1) holds. Now let > 0. Since < ,
condition (2) in the definition of supremum implies that is not an upper bound of A. Therefore,
there must exist an element a of A such that < a as desired.
Conversely, suppose conditions (1) and (2) in the statement hold. Then all we need to show is
that condition (2) in the definition of supremum holds. Let M be an upper bound of A and assume,
by way of contradiction, that M < . Set = M. By condition (2) in the statement, there is
a A such that a > = M. This contradicts the fact that M is an upper bound. The conclusion
now follows. 
The following is an axiom of the real numbers and is called the completeness axiom.
The Completeness Axiom. Every nonempty subset A of R that is bounded above has a least upper
bound. That is, sup A exists and is a real number.
This axiom distinguishes the real numbers from all other ordered fields and it is crucial in the
proofs of the central theorems of analysis.
There is a corresponding definition for the infimum of a set.
Definition 1.5.3 Let A be a nonempty subset of R that is bounded below. We call a number a
greatest lower bound or infimum of A, denoted by = inf A, if
(1) x for all x A (that is, is a lower bound of A);
(2) If N is a lower bound of A, then N (this means is largest among all lower bounds).
Using the completeness axiom, it is easy to prove that if a nonempty set is bounded below, then

21
its infimum exists.


Example 1.5.2

(a) inf(0, 3] = inf[0, 3] = 0.

(b) inf{3,
3.
 5, 7,n8, 10} = 
(1)
(c) inf
: n N = 1.
n
1
(d) inf{1 + : n N} = 1.
n
(e) inf{x2 : 2 < x < 1, x R} = 0.
The following proposition is useful when dealing with infima and its proof is completely
analogous to that of Proposition 1.5.1.
Proposition 1.5.2 Let A be a nonempty subset of R that is bounded below. Then = inf A if and

only if
(1) x for all x A;
(2) For any > 0, there exists a A such that a < + .
The following is a basic property of suprema. Additional ones are described in the exercises.
Theorem 1.5.3 Let A and B be nonempty sets and A B. Suppose B is bounded above. Then

sup A sup B.
Proof: Let M be an upper bound for B, then for x B, x M. In particular, it is also true that x M
for x A. Thus, A is also bounded above. Now, since sup B is an upper bound for B, it is also an
upper bound for A. Then, by the second condition in the definition of supremum, sup A sup B as
desired. 
It will be convenient for the study of limits of sequences and functions to introduce two additional
symbols.
Definition 1.5.4 The extended real number system consists of the real field R and the two symbols
and . We preserve the original order in R and define
< x <
for every x R
The extended real number system does not form an ordered field, but it is customary to make the
following conventions:
(a) If x is a real number, then
x + = , x + () = .
(b) If x > 0, then x = , x () = .
(c) If x < 0, then x = , x () = .
(d) + = , + () = , = () () = , and () = () = .
We denote the extended real number set by R. The expressions 0 , + (), and () +
are left undefined.
The set R with the above conventions will be convenient to describe results about limits in later
chapters.
Definition 1.5.5 If A 6= 0/ is not bounded above in R, we will write sup A = . If A is not bounded
below in R, we will write inf A = .

1.6 APPLICATIONS OF THE COMPLETENESS AXIOM

22

With this definition, every nonempty subset of R has a supremum and an infimum in R. To
complete the picture we adopt the following conventions for the empty set: sup 0/ = and
inf 0/ = .

Exercises
1.5.1 Prove that a subset A of R is bounded if and only if there is M R such that |x| M for all

x A.
1.5.2 Let A be a nonempty set and suppose and satisfy conditions (1) and (2) in Definition 1.5.2
(that is, both are suprema of A). Prove that = .
1.5.3 For each subset of R below, find its supremum and its infimum.

(a) {1, 5, 17}

(b) [0,
 5)

(1)n
(c) 1 +
:nN
n
(d) (3,
 )

3n
(e)
:nN
n+4
1.5.4 I Suppose A and B are nonempty subsets of R that are bounded above. Define
A + B = {a + b : a A and b B}.
Prove that A + B is bounded above and
sup(A + B) = sup A + sup B.
1.5.5 Suppose A and B are nonempty subsets of R that are bounded below. Prove that A + B is

bounded below and

inf(A + B) = inf A + inf B.
1.5.6 Let A be a nonempty subset of R which is bounded below. Define

A = {a : a A}.
Prove that
inf A = sup(A).
1.5.7 Let A, B be nonempty subsets of R that are bounded below. Prove that if A B, then

inf A inf B.

1.6

APPLICATIONS OF THE COMPLETENESS AXIOM

We prove here several fundamental properties of the real numbers that are direct consequences
of the Completeness Axiom.
Theorem 1.6.1 The Archimedean Property. The set of natural numbers is unbounded above.

23
Proof: Let us assume by contradiction that N is bounded above. Since N is nonempty,
m = sup N
exists and is a real number. By Proposition 1.5.1 (with = 1), there exists n N such that
m 1 < n m.
But then n + 1 > m. This is a contradiction since n + 1 is a natural number. 
The following theorem presents several immediate consequences.
Theorem 1.6.2 The following hold:
(1) For any x R, there exists n N such that x < n;
(2) For any > 0, there exists n N such that 1/n < ;
(3) For any x > 0 and for any y R, there exists n N such that y < nx;
(4) For any x R, there exists m Z such that m 1 x < m.
Proof: (1) Fix any x R. Since N is not bounded above, x cannot be an upper bound of N. Thus,
there exists n N such that x < n.
(2) Fix any > 0. Then 1/ is a real number. By (1), there exists n N such that
1/ < n.
This implies 1/n < .
(3) We only need to apply (1) for the real number y/x.
(4) First we consider the case where x > 0. Define the set
A = {n N : n > x}.
From part (1), A is nonempty. Since A is a subset of N, by the Well-Ordering Property of the natural
numbers, A has a smallest element `. Thus,
` > x and ` n for any n > x, n N .
Then ` 1 cannot be in A since ` 1 < `. Therefore,
` 1 x < `,
and the conclusion follows with m = `.
In the case x 0, by part (1), there exists N N such that
|x| < N.
In this case, N < x < N, so x + N > 0. Then, by the result just obtained for positive numbers, there
exists a natural number k such that k 1 x + N < k. This implies
k N 1 x < k N.
Setting m = k N, the conclusion follows. The proof is now complete. 

1.6 APPLICATIONS OF THE COMPLETENESS AXIOM

24

Theorem 1.6.3 The Density Property of Q. If x and y are two real numbers such that x < y, then

there exists a rational number r such that

x < r < y.
Proof: We are going to prove that there exist an integer m and a positive integer n such that
x < m/n < y,
or, equivalently,
nx < m < ny = nx + n(y x).
Since y x > 0, by Theorem 1.6.2 (3), there exists n N such that 1 < n(y x). Then
ny = nx + n(y x) > nx + 1.
By Theorem 1.6.2 (4), one can choose m Z such that
m 1 nx < m.
Then nx < m nx + 1 < ny. Therefore,
x < m/n < y.
The proof is now complete. 
We will prove in a later section (see Examples 3.4.1 and 4.3.1)
that there exists a (unique)
positive real number x such that x2 = 2. We denote that number by 2. The following result shows,
in particular, that R 6= Q.

Proposition 1.6.4 The number 2 is irrational.

Proof: Suppose, by way of contradiction, that 2 Q. Then there are integers r and s with s 6= 0,
such that

r
2= .
s
By canceling out the common factors of r and s, we may assume that r and s have no common
factors.
Now, by squaring both sides of the equation above, we get
2=

r2
,
s2

and, hence,
2s2 = r2 .

(1.3)

It follows that r2 is an even integer. Therefore, r is an even integer (see Exercise 1.4.1). We can then
write r = 2 j for some integer j. Hence r2 = 4 j2 . Substituting in (1.3), we get s2 = 2 j2 . Therefore,
s2 is even. We conclude as before that s is even. Thus, both r and s have a common factor, which is a
contradiction. 
The next theorem shows that irrational numbers are as ubiquitous as rational numbers.

25
Theorem 1.6.5 Let x and y be two real numbers such that x < y. Then there exists an irrational

number t such that

x < t < y.
Proof: Since x < y, one has

x 2 < y 2
By Theorem 1.6.3, there exists a rational number r such that

x 2 < r < y 2
This implies

x < r + 2 < y.

Since r is rational, the number t = r + 2 is irrational (see Exercise 1.6.3) and x < t < y. 

Exercises
1.6.1 I Let r be a rational number such that 0 < r < 1. Prove that there is n N such that

1
1
<r .
n+1
n
1
.
n
1.6.3 Prove that if x is a rational number and y is an irrational number, then x + y is irrational. What
can you say about xy?
1.6.2 Let x R. Prove that for every n N, there is r Q such that |x r| <

1.6.4 Prove that in between two real numbers a and b with a < b, there are infinitely many rational

numbers.
1.6.5 Prove that in between two real numbers a and b with a < b, there are infinitely many irrational

numbers.

CONVERGENCE
LIMIT THEOREMS
MONOTONE SEQUENCES
THE BOLZANO-WEIERSTRASS THEOREM
LIMIT SUPERIOR AND LIMIT INFERIOR
OPEN SETS, CLOSED SETS, AND LIMIT POINTS

2. SEQUENCES

We introduce the notion of limit first through sequences. As mentioned in Chapter 1, a sequence
is just a function with domain N. More precisely, a sequence of elements of a set A is a function
f : N A. We will denote the image of n under the function with subscripted variables, for example,
an = f (n). We will also denote sequences by {an }
n=1 , {an }n , or even {an }. Each value an is called
a term of the sequence, more precisely, the n-th term of the sequence.


Example 2.0.1 Consider an = n1 for n N. This is a sequence of rational numbers. On occasion,

when the pattern is clear, we may list the terms explicitly as in

1 1 1 1
1, , , , , . . .
2 3 4 5
n
 Example 2.0.2 Let an = (1) for n N. This is a sequence of integers, namely,
1, 1, 1, 1, 1, 1, . . .
Note that the sequence takes on only two values. This should not be confused with the two-element
set {1, 1}.

2.1

CONVERGENCE
Definition 2.1.1 Let {an } be a sequence of real numbers. We say that the sequence {an } converges

to a point a R if for any > 0, there exists a positive integer N such that for any n N with n N,
one has
|an a| < (or equivalently, a < an < a + ).
In this case, we call the point a the limit of the sequence (see Theorem 2.1.3 below). If the sequence
{an } does not converge to any point, we call the sequence divergent.
Remark 2.1.1 It follows directly from the definition, using the Archimedean property, that a

sequence {an } converges to a if and only if for any > 0, there exists a real number N such that for
any n N with n > N, one has
|an a| < .

2.1 CONVERGENCE

28


Example 2.1.1 Let > 0 and consider the sequence given by

an =

1
for n N .
n

Let > 0. Choose an integer N such that N > ( 1 )1/ . For every n N, one has n > ( 1 )1/ and,
hence, n > 1 . This implies


1

 0 = 1 < 1 = .
 n
 n
1/
We conclude that limn an = 0.


Example 2.1.2 Consider the sequence {an } where

an =

3n2 + 4
.
2n2 + n + 5

We will prove directly from the definition that this sequence converges to a = 32 .
Let > 0. We first search for a suitable N. To that end, we simplify and estimate the expression
|an a|. Notice that


 
 


2
2
2


 
 


an 3  =  3n + 4 3  =  2(3n + 4) 3(2n + n + 5)  =  7 3n 
 2n2 + n + 5 2  
  2(2n2 + n + 5) 

2
2(2n2 + n + 5)
3n + 7
10n 10
=
< 2 = .
2
2(2n + n + 5) 4n
4n
10
To guarantee that the last expression is less than , it will suffice to choose N > . Indeed, if n N,
4
we get
|an a|

10
10
10

< 10 = .
4n 4N
4 4

The following result is quite useful in proving certain inequalities between numbers.
Lemma 2.1.2 Let ` 0. If ` < for all > 0, then ` = 0.

Proof: This is easily proved by contraposition. If ` > 0, then there is a positive number, for example
= `/2, such that < `. 
Theorem 2.1.3 A convergent sequence {an } has at most one limit.
Proof: Suppose {an } converges to a and b. Then given > 0, there exist positive integers N1 and
N2 such that
|an a| < /2 for all n N1
and
|an b| < /2 for all n N2 .
Let N = max{N1 , N2 }. Then
|a b| |a aN | + |aN b| < /2 + /2 = .
Since > 0 is arbitrary, by Lemma 2.1.2, |a b| = 0 and, hence, a = b. 
The following comparison theorem shows that (non-strict) inequalities are preserved in the
limit.

29
Theorem 2.1.4 Comparison Theorem. Suppose {an } and {bn } converge to a and b, respec-

tively, and an bn for all n N. Then a b.

Proof: For any > 0, there exist N1 , N2 N such that
a < an < a + ,

for n N1 ,

b < bn < b + ,

for n N2 .

Choose N > max{N1 , N2 }. Then

a < aN bN < b + .
Thus, a < b + 2. This implies a b since is arbitrary. 
Theorem 2.1.5 The Squeeze Theorem. Suppose the sequences {an }, {bn }, and {cn } satisfy

an bn cn for all n N,
and limn an = limn cn = `. Then limn bn = `.
Proof: Fix any > 0. Since limn an = `, there exists N1 N such that
` < an < ` +
for all n N1 . Similarly, since limn cn = `, there exists N2 N such that
` < cn < ` +
for all n N2 . Let N = max{N1 , N2 }. Then, for n N, we have
` < an bn cn < ` + ,
which implies |bn `| < . Therefore, limn bn = `. 
Definition 2.1.2 A sequence {an } is bounded above if the set {an : n N} is bounded above.
Similarly, the sequence {an } is bounded below if the set {an : n N} is bounded below. Finally, we
say that the sequence {an } is bounded if the set {an : n N} is bounded, that is, if it is both bounded
above and bounded below.

It follows from the observation after Definition 1.5.1 that the sequence {an } is bounded if and
only if there is M R such that |an | M for all n N.
Theorem 2.1.6 A convergent sequence is bounded.
Proof: Suppose the sequence {an } converges to a. Then, for = 1, there exists N N such that
|an a| < 1 for all n N.
Since |an | |a| ||an | |a|| |an a|, this implies |an | < 1 + |a| for all n N. Set
M = max{|a1 |, . . . , |aN1 |, |a| + 1}.
Then |an | M for all n N. Therefore, {an } is bounded. 

Definition 2.1.3 Let {an }

n=1 be a sequence of real numbers. The sequence {bn }n=1 is called a

subsequence of {an }n=1 if there exists a sequence of increasing positive integers

n1 < n2 < n3 < ,
such that bk = ank for each k N.

2.1 CONVERGENCE

30


Example 2.1.3 Consider the sequence an = (1)n for n N.

Then {a2k } is a subsequence of {an } and a2k = 1 for all k (here nk = 2k for all k). Similarly,
{a2k+1 } is also a subsequence of {an } and a2k+1 = 1 for all k (here nk = 2k + 1 for all k).
Lemma 2.1.7 Let {nk }k be a sequence of positive integers with

n1 < n2 < n3 <

Then nk k for all k N.
Proof: We use mathematical induction. When k = 1, it is clear that n1 1 since n1 is a positive
integer. Assume nk k for some k. Now nk+1 > nk and, since nk and nk+1 are integers, this implies,
nk+1 nk + 1. Therefore, nk+1 > k + 1 by the inductive hypothesis. The conclusion now follows by
the principle of mathematical induction. 
Theorem 2.1.8 If a sequence {an } converges to a, then any subsequence {ank } of {an } also converges to a.
Proof: Suppose {an } converges to a and let > 0 be given. Then there exists N such that
|an a| < for all n N.
For any k N, since nk k, we also have
|ank a| < .
Thus, {ank } converges to a as k . 


Example 2.1.4 Let an = (1)n for n N. Then the sequence {an } is divergent. Indeed, suppose

by contradiction that
lim an = `.

Then every subsequence of {an } converges to a number ` R. From the previous theorem, it follows,
in particular, that
` = lim a2k = 1 and ` = lim a2k+1 = 1.
k

This contradiction shows that the sequence is divergent.

Since the sequence {an } is bounded but not convergent, this example illustrates the fact that the
converse of theorem 2.1.6 is not true.
Remark 2.1.9 Given a positive integer k0 , it will be convenient to also talk about the sequence

{an }nk0 , that is, a function defined only for the integers greater than or equal to k0 . For simplicity
of notation, we may also denote this sequence by {an } whenever the integer k0 is clear from the
context. For instance, we talk of the sequence {an } given by
an =

n+1
.
(n 1)(n 2)

although a1 and a2 are not defined. In all cases, the sequence must be defined from some integer
onwards.

31

Exercises
2.1.1 Prove the following directly from the definition of limit.

2n2 + 2
= 0.
3n3 + 1
2
n +1
(b) limn 2
= 1.
5n + n + 1 5
2n3 + 1 1
= .
(c) limn 3
4n n 2
(a) limn

2.1.2 Prove that if {an } is a convergent sequence, then {|an |} is a convergent sequence. Is the
converse true?
2.1.3 Let {an } be a sequence. Prove that if the sequence {|an |} converges to 0, then {an } also

converges to 0.
2.1.4 Prove that limn

sin n
=0
n

2.1.5 Let {xn } be a bounded sequence and let {yn } be a sequence that converges to 0. Prove that

the sequence {xn yn } converges to 0.

2.1.6 Prove that if limn an = ` > 0, then there exists N such that an > 0 for all n N.
a

2.1.7 I Prove that if limn an = ` 6= 0, then limn an+1

= 1. Is the conclusion still true if ` = 0?
n
2.1.8 Let an 0 for all n N. Prove that if limn an = `, then limn

an = `.

2.1.9 Prove that the sequence {an } with an = sin(n/2) is divergent.

2.1.10 I Consider a sequence {an }.

(a) Prove that limn an = ` if and only if limk a2k = ` and limk a2k+1 = `.
(b) Prove that limn an = ` if and only if limk a3k = `, limk a3k+1 = `, and
limk a3k+2 = `.
2.1.11 Given a sequence {an }, define a new sequence {bn } by

bn =

a1 + a2 + . . . + an
.
n

(a) Prove that if limn an = `, then limn bn = `.

(b) Find a counterexample to show that the converse does not hold in general.

2.2 LIMIT THEOREMS

32

2.2

LIMIT THEOREMS
We now prove several theorems that facilitate the computation of limits of some sequences in
terms of those of other simpler sequences.
Theorem 2.2.1 Let {an } and {bn } be sequences of real numbers and let k be a real number. Suppose
{an } converges to a and {bn } converges to b. Then the sequences {an + bn }, {kan }, and {an bn }
converge and
(1) limn (an + bn ) = a + b;
(2) limn (kan ) = ka;
(3) limn (an bn ) = ab;



an
an a
(4) If in addition b 6= 0 and bn 6= 0 for n N, then
= .
converges and limn
bn
bn b
Proof: (1) Fix any > 0. Since {an } converges to a, there exists N1 N such that
|an a| <

for all n N1 .
2

Similarly, there exists N2 N such that

|bn b| <

for all n N2 .
2

Let N = max{N1 , N2 }. For any n N, one has

|(an + bn ) (a + b)| |an a| + |bn b| <

+ = .
2 2

Therefore, limn (an + bn ) = a + b. This proves (1).

(2) If k = 0, then ka = 0 and kan = 0 for all n. The conclusion follows immediately. Suppose

next that k 6= 0. Given > 0, let N N be such that |an a| <

for n N. Then for n N,
|k|
|kan ka| = |k||an a| < . It follows that limn (kan ) = ka as desired. This proves (2).
(3) Since {an } is convergent, it follows from Theorem 2.1.6 that it is bounded. Thus, there exists
M > 0 such that
|an | M for all n N .
For every n N, we have the following estimate:
|an bn ab| = |an bn an b + an b ab| |an ||bn b| + |b||an a|.
Let > 0. Since {an } converges to a, we may choose N1 N such that
|an a| <

for all n N1 .
2|b| + 1

Similarly, since {bn } converges to b, we may choose N2 N such that

|bn b| <

for all n N2 .
2M

(2.1)

33
Let N = max{N1 , N2 }. Then, for n N, it follows from (2.1) that
|an bn ab| < M

+ |b|
< for all n N.
2M
2|b| + 1

Therefore, limn an bn = ab. This proves (3).

(4) Let us first show that
1
1
= .
n bn
b
lim

Since {bn } converges to b, there is N1 N such that

|bn b| <

|b|
for n N1 .
2

|b|
|b|
|b|
It follows (using a triangle inequality) that, for such n, < |bn | |b| <
and, hence,
< |bn |.
2
2
2
For each n N1 , we have the following estimate


 1 1  |bn b| 2|bn b|
 =
.
(2.2)

 bn b 
|bn ||b|
b2
Now let > 0. Since limn bn = b, there exists N2 N such that
|bn b| <

b2
for all n N2 .
2

Let N = max{N1 , N2 }. By (2.2), one has



 1 1  2|bn b|
 
< for all n N.
 bn b 
b2
1
1
= .
bn b
Finally, we can apply (3) and have

It follows that limn

lim

an

n bn

= lim an
n

1
a
= .
bn b

The proof is now complete. 



Example 2.2.1 Consider the sequence {an } given by

an =

3n2 2n + 5
.
1 4n + 7n2

(2.3)

Dividing numerator and denominator by n2 , we can write

an =

3 2/n + 5/n2
1/n2 4/n + 7

(2.4)

Therefore, by the limit theorems above,

3 2/n + 5/n2 limn 3 limn 2/n + limn 5/n2 3
=
= .
n 1/n2 4/n + 7
limn 1/n2 limn 4/n + limn 7 7

lim an = lim

(2.5)

2.2 LIMIT THEOREMS

34

n
b, where b > 0. Consider the case where b > 1. In this case, an > 1
for every n. By the binomial theorem,

Example 2.2.2 Let an =

b = ann = (an 1 + 1)n 1 + n(an 1).

This implies
0 < an 1

b1
.
n

b1
. It follows that for n N,

b1 b1

< .
|an 1| = an 1 <
n
N
Thus, limn an = 1.
In the case where b = 1, it is obvious that an = 1 for all n and, hence, limn an = 1.
1
If 0 < b < 1, let c = and define
b

1
xn = n c = .
an
Since c > 1, it has been shown that limn xn = 1. This implies
For each > 0, choose N >

lim an = lim

n xn

= 1.

Exercises
2.2.1 I Find
the following limits if they exist:

(a) limn ( n2 + n n).

(b) limn ( 3 n3 + 3n2 n).

(c) limn ( 3 n3 + 3n2 n2 + n).

(d) limn ( n + 1 n).

(e) limn ( n + 1 n)/n.

2.2.2 Find the following limits.

(a) For |r| < 1 and b R, limn (b + br + brr + + brn ).



2
2
2
+
(b) limn
++ n .
10 102
10
2.2.3 Prove or disprove the following statements:

(a)
(b)
(c)
(d)
(e)
(f)

If {an } and {bn } are convergent sequences, then {an + bn } is a convergent sequence.
If {an } and {bn } are divergent sequences, then {an + bn } is divergent sequence.
If {an } and {bn } are convergent sequences, then {an bn } is a convergent sequence.
If {an } and {bn } are divergent sequences, then {an bn } is a divergent sequence.
If {an } and an + bn are convergent sequences, then {bn } is a convergent sequence.
If {an } and an + bn are divergent sequences, then {bn } is a divergent sequence.

35

2.3

MONOTONE SEQUENCES
Definition 2.3.1 A sequence {an } is called increasing if

an an+1 for all n N.

It is called decreasing if
an an+1 for all n N.
If {an } is increasing or decreasing, then it is called a monotone sequence.
The sequence is called strictly increasing (resp. strictly decreasing) if an < an+1 for all n N
(resp. an > an+1 for all n N).
It is easy to show by induction that if {an } is an increasing sequence, then an am whenever
n m.
Theorem 2.3.1 Monotone Convergence Theorem. Let {an } be a sequence of real numbers.
The following hold:
(1) If {an } is increasing and bounded above, then it is convergent.
(2) If {an } is decreasing and bounded below, then it is convergent.
Proof: (1) Let {an } be an increasing sequence that is bounded above. Define
A = {an : n N}.
Then A is a subset of R that is nonempty and bounded above and, hence, sup A exists. Let ` = sup A
and let > 0. By Proposition 1.5.1, there exists N N such that
` < aN `.
Since {an } is increasing,
` < aN an for all n N.
On the other hand, since ` is an upper bound for A, we have an ` for all n. Thus,
` < an < ` + for all n N.
Therefore, limn an = `.
(2) Let {an } be a decreasing sequence that is bounded below. Define
bn = an .
Then {bn } is increasing and bounded above (if M is a lower bound for {an }, then M is an upper
bound for {bn }). Let
` = lim bn = lim (an ).
n

Then {an } converges to ` by Theorem 2.2.1. 

Remark 2.3.2 It follows from the proof of Theorem 2.3.1 that if {an } is increasing and bounded

above, then
lim an = sup{an : n N}.

Similarly, if {an } is decreasing and bounded below, then

lim an = inf{an : n N}.

2.3 MONOTONE SEQUENCES

36

Corollary 2.3.3 Let {an } be a sequence of real numbers that is bounded. Define

sn = sup{ak : k n}

(2.6)

tn = inf{ak : k n}.

(2.7)

and

Then {sn } and {tn } are convergent.

Proof: If n m, then {ak : k m} {ak : k n}. Therefore, it follows from Theorem 1.5.3
that sn sm and, so, the sequence {sn } is decreasing. Since {an } is bounded, then so is {sn }. In
particular, {sn } is bounded below. Similarly, {tn } is increasing and bounded above. Therefore, both
sequences are convergent by Theorem 2.3.1. 


Example 2.3.1 Given r R with |r| < 1, define an = rn for n N. Then

lim an = 0.

This is clear if r = 0. Let us first consider the case where 0 < r < 1. Then 0 an+1 = ran an for
all n. Therefore, {an } is decreasing and bounded below. By Theorem 2.3.1, the sequence converges.
Let
` = lim an .
n

Since an+1 = ran for all n, taking limits on both sides gives ` = r`. Thus, (1 r)` = 0 and, hence,
` = 0. In the general case, we only need to consider the sequence defined by bn = |an | for n N; see
Exercise 2.1.3.


Example 2.3.2 Consider the sequence {an } given by



1 n
an = 1 +
, n N.
n
By the binomial theorem,
n    k
n
1
an =
n
k=0 k
n(n 1) 1 n(n 1)(n 2) 1
n(n 1) (n (n 1)) 1
+
++
2
3
n
2! n

 3!   n 
 n!  
n 

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

= 1+1+

The corresponding expression for an+1 has one more term and each factor (1 nk ) is replaced by the
k
larger factor (1 n+1
). It is then clear that an < an+1 for all n N. Thus, the sequence is increasing.
Moreover,
1
1
1
+ ++
2! 3!
n!
1
1
1
< 2+
+
++
1.2 2.3
(n 1) n

n1 
1
1
1
= 2+

= 3 < 3.
k+1
n
k=1 k

an 1 + 1 +

37
Hence the sequence is bounded above.
By the Monotone Convergence Theorem, limn an exists and is denoted by e. In fact, e is an
irrational number and e 2.71828.


Example 2.3.3 Consider the sequence {an } defined as follows:

a1 = 2
an+1 =

an + 5
for n 1.
3

First we will show that the sequence is increasing. We prove by induction that for all n N,
an < an+1 . Since a2 = a13+5 = 73 > 2 = a1 , the statement is true for n = 1. Next, suppose ak < ak+1
for some k N. Then ak + 5 < ak+1 + 5 and (ak + 5)/3 < (ak+1 + 5)/3. Therefore,
ak+1 =

ak + 5 ak+1 + 5
<
= ak+2 .
3
3

It follows by induction that the sequence is increasing.

Next we prove that the sequence is bounded by 3. Again, we proceed by induction. The statement
is clearly true for n = 1. Suppose that ak 3 for some k N. Then
ak+1 =

ak + 5 3 + 5 8

= 3.
3
3
3

It follows that an 3 for all n N.

From the Monotone Convergence Theorem, we get that there is ` R such that limn an = `.
Since the subsequence {ak+1 }
k=1 also converges to `, we obtain
`=

`+5
.
3

Therefore, 3` = ` + 5 and, hence, ` = 5/2.

The following result is an application of the Monotone Convergence Theorem and will be used
in the next section to prove the Bolzano-Weierstrass Theorem.
Theorem 2.3.4 Nested Intervals Theorem. Let {In }
n=1 be a sequence of nonempty closed
bounded intervals satisfying In+1 In for all n N. Then the following hold:
T
(1)
/
n=1 In 6= 0.
T
(2) If, in addition, the lengths of the intervals In converge to zero, then
n=1 In consists of a single
point.
Proof: Let {In } be as in the statement with In = [an , bn ]. In particular, an bn for all n N. Given
that In+1 In , we have an an+1 and bn+1 bn for all n N. This shows that {an } is an increasing
sequence bounded above by b1 and {bn } is a decreasing sequence bounded below by a1 . By the
Monotone Convergence Theorem (Theorem 2.3.1), there exist a, b R such that limn an = a
and limn bn = b. Since an bn for all n, by Theorem 2.1.4, we get a b. Now, we also have
an a and b bn for all n N (since {an } is increasing and {bn } is decreasing). This shows that if
T
T
a x b, then x In for all n N. Thus, [a, b]
/ This proves
n=1 In . It follows that n=1 In 6= 0.
part (1).
T
T
Now note also that
n=1 In [a, b]. Indeed, if x n=1 In , then x In for all n. Therefore,
an x bn for all n. Using Theorem 2.1.4, we conclude a x b. Thus, x [a, b]. This proves the
T
desired inclusion and, hence,
n=1 In = [a, b].

2.3 MONOTONE SEQUENCES

38

We now prove part (b). Suppose the lengths of the intervals In converge to zero. This means
T
bn an 0 as n . Then b = limn bn = limn [(bn an ) + an ] = a. It follows that
n=1 In =
{a} as desired. 
When a monotone sequence is not bounded, it does not converge. However, the behavior follows
a clear pattern. To make this precise we provide the following definition.
Definition 2.3.2 A sequence {an } is said to diverge to if for every M R, there exists N N such
that
an > M for all n N.
In this case, we write limn an = . Similarly, we say that {an } diverges to and write
limn an = if for every M R, there exists N N such that
an < M for all n N.
Remark 2.3.5 We should not confuse a sequence that diverges to (that is, one that satisfies the

previous definition), with a divergent sequence (that is, one that does not converge).
Theorem 2.3.6 If a sequence {an } is increasing and not bounded above, then

lim an = .

Similarly, if {an } is decreasing and not bounded below, then

lim an = .

Proof: Fix any real number M. Since {an } is not bounded above, there exists N N such that
aN M. Then
an aN M for all n N
because {an } is increasing. Therefore, limn an = . The proof for the second case is similar. 
Theorem 2.3.7 Let {an }, {bn }, and {cn } be sequences of real numbers and let k be a constant.

Suppose
lim an = , lim bn = , and lim cn =

Then
limn (an + bn ) = ;
limn (an bn ) = ;
limn (an cn ) = ;
limn kan = if k > 0, and limn kan = if k < 0;
1
(5) limn
= 0. (Here we assume an 6= 0 for all n.)
an
Proof: We provide proofs for (1) and (5) since all other proofs are similar.
(1) Fix any M R. Since limn an = , there exists N1 N such that

(1)
(2)
(3)
(4)

an

M
for all n N1 .
2

39
Similarly, there exists N2 N such that
bn

M
for all n N1 .
2

Let N = max{N1 , N2 }. Then it is clear that

an + bn M for all n N.
This implies (1).
1
(5) For any > 0, let M = . Since limn an = , there exists N N such that

an >

1
for all n N.

This implies that for n N,




1
 0 = 1 < .
 an
 an
Thus, (5) holds. 
The proof of the comparison theorem below follows directly from Definition 2.3.2 (see also
Theorem 2.1.4).
Theorem 2.3.8 Suppose an bn for all n N.
(1) If limn an = , then limn bn = .
(2) If limn bn = , then limn an = .
Theorem 2.3.9 If {an } is not bounded above, then
lim sn = ,

where {sn } is defined in (2.6).

Similarly, if {an } is not bounded below, then
lim tn = ,

where {tn } is defined in (2.7).

Proof: Suppose {an } is not bounded above. Then for any k N, the set {ai : i k} is also not
bounded above. Thus, sk = sup{ai : i k} = for all k. Therefore, limk sk = . The proof for
the second case is similar. 

Exercises
2.3.1 I Let a1 =

an+1 =

2. Define

an + 2 for n 1.

(a) Prove that an < 2 for all n N.

(b) Prove that {an } is an increasing sequence.

2.4 THE BOLZANO-WEIERSTRASS THEOREM

40
(c) Prove that limn an = 2.

2.3.2 B Prove that each of the following sequences is convergent and find its limit.

an + 3
for n 1.
(a) a1 = 1 and an+1 =
2

(b) a1 = 6 and
 an+1 = an + 6 for n 1.
1
1
(c) an+1 =
2an + 2 , n 1, a1 > 0.
3
an
1
b
(d) an+1 =
an +
, b > 0.
2
an

2.3.3 B Prove thatqeach of the following sequences is convergent and find its limit.

p
p
2; 2 2; 2 2 2;
1
1
(b) 1/2;
;
;
1
2 + 1/2
2+
2 + 1/2
(a)

2.3.4 Prove that the following sequence is convergent:

an = 1 +

1
1
1
+ ++ , n N.
2! 3!
n!

2.3.5 B Let a and b be two positive real numbers with a < b. Define a1 = a, b1 = b, and

an+1 =

an bn and bn+1 =

an + bn
for n 1.
2

Show that {an } and {bn } are convergent to the same limit.

2.4

THE BOLZANO-WEIERSTRASS THEOREM

The Bolzano-Weierstrass Theorem is at the foundation of many results in analysis. It is, in fact,
equivalent to the completeness axiom of the real numbers.
Theorem 2.4.1 Bolzano-Weierstrass. Every bounded sequence {an } of real numbers has a
convergent subsequence.
Proof: Suppose {an } is a bounded sequence. Define A = {an : n N} (the set of values of the
sequence {an }). If A is finite, then at least one of the elements of A, say x, must be equal to an for
infinitely many choices of n. More precisely, Bx = {n N : an = x} is infinite. We can then define
a convergent subsequence as follows. Pick n1 such that an1 = x. Now, since Bx is infinite, we can
choose n2 > n1 such that an2 = x. Continuing in this way, we can define a subsequence {ank } which
is constant, equal to x and, thus, converges to x.
Suppose now that A is infinite. First observe there exist c, d R such that c an d for all
n N, that is, A [c, d].
We define a sequence of nonempty nested closed bounded intervals as follows. Set I1 = [c, d].
c+d
c+d
Next consider the two subintervals [c, c+d
2 ] and [ 2 , d]. Since A is infinite, at least one of A [c, 2 ]
c+d
c+d
c+d
or A [ c+d
2 , d] is infinite. Let I2 = [c, 2 ] if A [c, 2 ] is infinite and I2 = [ 2 , d] otherwise.

41
Continuing in this way, we construct a nested sequence of nonempty closed bounded intervals {In }
such that In A is infinite and the length of In tends to 0 as n .
We now construct the desired subsequence of {an } as follows. Let n1 = 1. Choose n2 > n1 such
that an2 I2 . This is possible since I2 A is infinite. Next choose n3 > n2 such that an3 I3 . In this
way, we obtain a subsequence {ank } such that ank Ik for all k N.
Set In = [cn , dn ]. Then limn (dn cn ) = 0. We also know from the proof of the Monotone
Convergence Theorem (Theorem 2.3.1), that {cn } converges. Say ` = limn cn . Thus, limn dn =
limn [(dn cn ) + cn ] = ` as well. Since ck ank dk for all k N, it follows from Theorem 2.1.4
that limk ank = `. This completes the proof. 
Definition 2.4.1 (Cauchy sequence). A sequence {an } of real numbers is called a Cauchy sequence

if for any > 0, there exists a positive integer N such that for any m, n N, one has
|am an | < .
Theorem 2.4.2 A convergent sequence is a Cauchy sequence.

Proof: Let {an } be a convergent sequence and let

lim an = a.

Then for any > 0, there exists a positive integer N such that
|an a| < /2 for all n N.
For any m, n N, one has
|am an | |am a| + |an a| < /2 + /2 = .
Thus, {an } is a Cauchy sequence. 
Theorem 2.4.3 A Cauchy sequence is bounded.
Proof: Let {an } be a Cauchy sequence. Then for = 1, there exists a positive integer N such that
|am an | < 1 for all m, n N.
In particular,
|an aN | < 1 for all n N.
Let M = max{|a1 |, . . . , |aN1 |, 1 + |aN |}. Then, for n = 1, . . . , N 1, we clearly have |an | M.
Moreover, for n N,
|an | = |an aN + aN | |an aN | + |aN | 1 + |aN | M.
Therefore, |an | M for all n N and, thus, {an } is bounded. 
Lemma 2.4.4 A Cauchy sequence that has a convergent subsequence is convergent.

Proof: Let {an } be a Cauchy sequence that has a convergent subsequence. For any > 0, there
exists a positive integer N such that
|am an | /2 for all m, n N.

2.4 THE BOLZANO-WEIERSTRASS THEOREM

42

Let {ank } be a subsequence of {an } that converges to some point a. For the above , there exists a
positive number K such that
|ank a| < /2 for all k K.
Thus, we can find a positive integer n` > N such that
|an` a| < /2.
Then for any n N, we have
|an a| |an an` | + |an` a| < .
Therefore, {an } converges to a. 
Theorem 2.4.5 Any Cauchy sequence of real numbers is convergent.
Proof: Let {an } be a Cauchy sequence. Then it is bounded by Theorem 2.4.3. By the BolzanoWeierstrass theorem, {an } has a convergent subsequence. Therefore, it is convergent by Lemma
2.4.4. 
Remark 2.4.6 It follows from Definition 2.4.1 that {an } is a Cauchy sequence if and only if for

every > 0, there exists N N such that

|an+p an | < for all n N and for all p.
Definition 2.4.2 A sequence {an } is called contractive if there exists k [0, 1) such that

|an+2 an+1 | k|an+1 an | for all n N .

Theorem 2.4.7 Every contractive sequence is convergent.

Proof: By induction, one has

|an+1 an | kn1 |a2 a1 | for all n N .
Thus,
|an+p an | |an+1 an | + |an+2 an+1 | + + |an+p an+p1 |
(kn1 + kn + + kn+p2 )|a2 a1 |
kn1 (1 + k + k2 + + k p1 )|a2 a1 |

kn1
|a2 a1 |.
1k

for all n, p N. Since kn1 0 as n (independently of p), this implies {an } is a Cauchy
sequence and, hence, it is convergent. 

Exercises
2.4.1 I Determine which of the following are Cauchy sequences.

(a) an = (1)n .
(b) an = (1)n /n.

43
(c) an = n/(n + 1).
(d) an = (cos n)/n.
2.4.2 Prove that the sequence

an =

n cos(3n2 + 2n + 1)
n+1

has a convergent subsequence.

2.4.3 Let f : [0, ) R be such that f (x) > 0 for all x. Define

an =

f (n)
.
f (n) + 1

Prove that the sequence an has a convergent subsequence.

2.4.4 Define

an =

1 + 2n
for n N .
2n

Prove that the sequence an is contractive.

2.5

LIMIT SUPERIOR AND LIMIT INFERIOR

Definition 2.5.1 Let {an } be a sequence. Then the limit superior of {an }, denoted by lim supn an ,

is defined by
lim sup an = lim sup{ak : k n}.
n

Note that lim supn an = limn sn , where sn is defined as in (2.6).

Similarly, the limit inferior of {an }, denoted by lim infn an , is defined by
lim inf an = lim inf{ak : k n}.
n

Note that lim infn an = limn tn , where tn is defined as in (2.7).

Remark 2.5.1 By Theorem 2.3.9, we see that if {an } is not bounded above, then

lim sup an = .
n

Similarly, if {an } is not bounded below, then

lim inf an = .
n

Theorem 2.5.2 Let {an } be a sequence and ` R. The following are equivalent:

(1) lim supn an = `.

(2) For any > 0, there exists N N such that
an < ` + for all n N,

2.5 LIMIT SUPERIOR AND LIMIT INFERIOR

44

and there exists a subsequence of {ank } of {an } such that

lim ank = `.

Proof: Suppose lim supn an = `. Then limn sn = `, where sn is defined as in (2.6). For any
> 0, there exists N N such that
` < sn < ` + for all n N.
This implies sN = sup{an : n N} < ` + . Thus,
an < ` + for all n N.
Moreover, for = 1, there exists N1 N such that
` 1 < sN1 = sup{an : n N1 } < ` + 1.
Thus, there exists n1 N such that
` 1 < an1 < ` + 1.
1
For = , there exists N2 N and N2 > n1 such that
2
`

1
1
< sN2 = sup{an : n N2 } < ` + .
2
2

Thus, there exists n2 > n1 such that

`

1
1
< an2 < ` + .
2
2

In this way, we can construct a strictly increasing sequence {nk } of positive integers such that
`

1
1
< ank < ` + .
k
k

Therefore, limk ank = `.

We now prove the converse. Given any > 0, there exists N N such that
an < ` + and ` < ank < ` +
for all n N and k N. Let any m N, we have
sm = sup{ak : k m} ` + .
By Lemma 2.1.7, nm m, so we also have
sm = sup{ak : k m} anm > ` .
Therefore, limm sm = lim supm an = `. 
The following result is proved in a similar way.

45
Theorem 2.5.3 Let {an } be a sequence and ` R. The following are equivalent:

(1) lim infn an = `.

(2) For any > 0, there exists N N such that
an > ` for all n N,
and there exists a subsequence of {ank } of {an } such that
lim ank = `.

The following corollary follows directly from Theorems 2.5.2 and 2.5.3.
Corollary 2.5.4 Let {an } be a sequence. Then

lim an = ` if and only if lim sup an = lim inf an = `.

Corollary 2.5.5 Let {an } be a sequence.

(1) Suppose lim supn an = ` and {ank } is a subsequence of {an } with

lim ank = `0 .

Then `0 `.
(2) Suppose lim infn an = ` and {ank } is a subsequence of {an } with
lim ank = `0 .

Then `0 `.
Proof: We prove only (1) because the proof of (2) is similar. By Theorem 2.5.2 and the definition
of limits, for any > 0, there exists N N such that
an < ` + and `0 < ank < `0 +
for all n N and k N. Since nN N, this implies
`0 < anN < ` + .
Thus, `0 < ` + 2 and, hence, `0 ` because is arbitrary. 
Remark 2.5.6 Let {an } be a bounded sequence. Define

A = {x R : there exists a subsequence {ank } with lim ank = x}.

It follows from Theorem 2.5.2, Theorem 2.5.3, and Corollary 2.5.5 that A 6= 0/ and
lim sup an = max A and lim inf an = min A.
n

Theorem 2.5.7 Suppose {an } is a sequence such that an > 0 for every n N and

lim sup
n

an+1
= ` < 1.
an

Then limn an = 0.

2.5 LIMIT SUPERIOR AND LIMIT INFERIOR

46

Proof: Choose > 0 such that ` + < 1. Then there exists N N such that
an+1
< ` + for all n N.
an
Let q = ` + . Then 0 < q < 1. By induction,
0 < an qnN aN for all n N.
Since limn qnN aN = 0, one has limn an = 0. 
By a similar method, we obtain the theorem below.
Theorem 2.5.8 Suppose {an } is a sequence such that an > 0 for every n N and

lim inf
n

an+1
= ` > 1.
an

Then limn an = .


Example 2.5.1 Given a real number , define

an =

n
, n N.
n!

When = 0, it is obvious that limn an = 0. Suppose > 0. Then

lim sup
n

an+1

= lim
= 0 < 1.
n n + 1
an

Thus, limn an = 0. In the general case, we can also show that limn an = 0 by considering
limn |an | and using Exercise 2.1.3.

Exercises
2.5.1 Find lim supn an and lim infn an for each sequence.
n.
(a) an = (1)
 n

(b) an = sin
.
2 n
1 + (1)
(c) an =
.
n 
n
(d) an = n sin
.
2
2.5.2 For a sequence {an }, prove that:
(a) lim infn an = if and only if limn an = .
(b) lim supn an = if and only if limn an = .
2.5.3 Let {an } and {bn } be bounded sequences. Prove that:

(a) supkn (an + bn ) supkn ak + supkn bk .

(b) infkn (an + bn ) infkn ak + infkn bk .

47
2.5.4 I Let {an } and {bn } be bounded sequences.

(a) Prove that lim supn (an + bn ) lim supn an + lim supn bn .
(b) Prove that lim infn (an + bn ) lim infn an + lim infn bn .
(c) Find two counterexamples to show that the equalities may not hold in part (a) and part (b).
Is the conclusion still true in each of parts (a) and (b) if the sequences involved are not necessarily
bounded?
2.5.5 Let {an } be a convergent sequence and let {bn } be an arbitrary sequence. Prove that

(a) lim supn (an + bn ) = lim supn an + lim supn bn = limn an + lim supn bn .
(b) lim infn (an + bn ) = lim infn an + lim infn bn = limn an + lim infn bn .

2.6

OPEN SETS, CLOSED SETS, AND LIMIT POINTS

The open ball in R with center a R and radius > 0 is the set
B(a; ) = (a , a + ).
Definition 2.6.1 A subset A of R is said to be open if for each a A, there exists > 0 such that

B(a; ) A.


Example 2.6.1 (1) Any open interval A = (c, d) is open. Indeed, for each a A, one has c < a < d.

Let
= min{a c, d a}.
Then
B(a; ) = (a , a + ) A.
Therefore, A is open.
(2) The sets A = (, c) and B = (c, ) are open, but the set C = [c, ) is not open. The reader can
easily verify that A and B are open. Let us show that C is not open. Assume by contradiction that C
is open. Then, for the element c C, there exists > 0 such that
B(c; ) = (c , c + ) C.
However, this is a contradiction because c /2 B(c; ), but c /2
/ C.
Theorem 2.6.1 The following hold:

(1) The subsets 0/ and R are open.

(2) The union of any collection of open subsets of R is open.
(3) The intersection of a finite number of open subsets of R is open.
Proof: The proof of (1) is straightforward.
(2) Suppose {G : I} is an arbitrary collection of open subsets of R. That means G is open for
every I. Let us show that the set
[
G=
G
I

2.6 OPEN SETS, CLOSED SETS, AND LIMIT POINTS

48

is open. Take any a G. Then there exists 0 I such that

a G0 .
Since G0 is open, there exists > 0 such that
B(a; ) G0 .
This implies
B(a; ) G
because G0 G. Thus, G is open.
(3) Suppose Gi , i = 1, . . . , n, are open subsets of R. Let us show that the set
G=

n
\

Gi

i=1

is also open. Take any a G. Then a Gi for i = 1, . . . , n. Since each Gi is open, there exists i > 0
such that
B(a; i ) Gi .
Let = min{i : i = 1, . . . , n}. Then > 0 and
B(a; ) G.
Thus, G is open. 
Definition 2.6.2 A subset S of R is called closed if its complement Sc = R \S is open.


Example 2.6.2 The sets [a, b], (, a], and [a, ) are closed. Indeed, (, a]c = (a, ) and

[a, )c = (, a) which are open by Example 2.6.1. Since [a, b]c = (, a) (b, ), [a, b]c is open
by Theorem 2.6.1.
Theorem 2.6.2 The following hold:

(1) The sets 0/ and R are closed.

(2) The intersection of any collection of closed subsets of R is closed.
(3) The union of a finite number of closed subsets of R is closed.
Proof: The proofs for these are simple using the De Morgans law. Let us prove, for instance, (2).
Let {S : I} be a collection of closed sets. We will prove that the set
S=

is also closed. We have

!c
Sc =

\
I

Thus,

Sc

Sc .

is open because it is a union of opens sets in R (Theorem 2.6.1(b)). Therefore, S is closed. 

Theorem 2.6.3 If A is a nonempty subset of R that is closed and bounded above, then max A exists.

Similarly, if A is a nonempty subset of R that is closed and bounded below, then min A exists

49
Proof: Let A be a nonempty closed set that is bounded above. Then sup A exists. Let m = sup A. To
complete the proof, we will show that m A. Assume by contradiction that m
/ A. Then m Ac ,
which is an open set. So there exists > 0 such that
(m , m + ) Ac .
This means there exists no a A with
m < a m.
This contradicts to the fact that m is the least upper bound of A (see Proposition 1.5.1). Therefore,
max A exists. 
Theorem 2.6.4 A subset A of R is closed if and only if for any sequence {an } in A that converges
to a point a R, it follows that a A.
Proof: Suppose A is a closed subset of R and {an } is a sequence in A that converges to a. Suppose by
contradiction that a 6 A. Since A is closed, there exists > 0 such that B(a; ) = (a , a + ) Ac .
Since {an } converges to a, there exists N N such that
a < aN < a + .
This implies aN Ac , a contradiction.
Let us now prove the converse. Suppose by contradiction that A is not closed. Then Ac is not
open. Since Ac is not open, there exists a Ac such that for any > 0, one has B(a; ) A 6= 0.
/ In
particular, for such an a and for each n N, there exists an B(a; 1n ) A. It is clear that the sequence
{an } is in A and it is convergent to a (because |an a| < n1 , for all n N). This is a contradiction
since a
/ A. Therefore, A is closed. .
Definition 2.6.3 (cluster/limit/accumulation point). Let A be a subset of R. A point a R (not
necessarily in A) is called a limit point of A if for any > 0, the open ball B(a; ) contains an infinite
number of points of A.
A point a A which is not an accumulation point of A is called an isolated point of A.
 Example 2.6.3 (1) Let A = [0, 1). Then a = 0 is a limit point of A and b = 1 is also a limit point
of A. In fact, any point of the interval [0, 1] is a limit point of A. The set [0, 1) has no isolated points.

(2) Let A = Z. Then A does not have any limit points. Every element of Z is an isolated point of Z.
(3) Let A = {1/n : n N}. Then a = 0 is the only limit point of A. All elements of A are isolated
points.
The following theorem is a variation of the Bolzano-Weierstrass theorem.
Theorem 2.6.5 Any infinite bounded subset of R has at least one limit point.

Proof: Let A be an infinite subset of R and let {an } be a sequence of A such that
am 6= an for m 6= n
(see Theorem 1.2.7). Since {an } is bounded, by the Bolzano-Weierstrass theorem (Theorem 2.4.1),
it has a convergent subsequence {ank }. Set b = limk ank . Given > 0, there exists K N such
that ank B(b; ) for k K. Since the set {ank : k K} is infinite, it follows that b is a limit point of
A. 
The following definitions and results provide the framework for discussing convergence within
subsets of R.

2.6 OPEN SETS, CLOSED SETS, AND LIMIT POINTS

50

Definition 2.6.4 Let D be a subset of R. We say that a subset V of D is open in D if for every a V ,

there exists > 0 such that

B(a; ) D V.
Theorem 2.6.6 Let D be a subset of R. A subset V of D is open in D if and only if there exists an

open subset G of R such that

V = D G.
Proof: Suppose V is open in D. By definition, for every a V , there exists a > 0 such that
B(a; a ) D V.
Define
G = aV B(a; a )
Then G is a union of open subsets of R, so G is open. Moreover,
V G D = aV [B(a; a ) D] V.
Therefore, V = G D.
Let us now prove the converse. Suppose V = G D, where G is an open set. For any a V , we
have a G, so there exists > 0 such that
B(a; ) G.
It follows that
B(a; ) D G D = V.
The proof is now complete. 
Definition 2.6.5 Let D be a subset of R. We say that a subset A of D is closed in D if D \ A is open
in D.
Theorem 2.6.7 Let D be a subset of R. A subset K of D is closed in D if and only if there exists a
closed subset F of R such that
K = D F.
Proof: Suppose K is a closed set in D. Then D \ K is open in D. By Theorem 2.6.6, there exists an
open set G such that
D \ K = D G.
It follows that
K = D \ (D \ K) = D \ (D G) = D \ G = D Gc .
Let F = Gc . Then F is a closed subset of R and K = D F.
Conversely, suppose that there exists a closed subset F of R such that K = D F. Then
D \ K = D \ (D F) = D \ F = D F c .
Since F c is an open subset of R, applying Theorem 2.6.6 again, one has that D \ K is open in D.
Therefore, K is closed in D by definition. .

51
Corollary 2.6.8 Let D be a subset of R. A subset K of D is closed in D if and only if for every

sequence {xk } in K that converges to a point x D it follows that x K.

Proof: Let D be a subset of R. Suppose K is closed in D. By Theorem 2.6.7, there exists a closed
subset F of R such that
K = D F.
Let {xk } be a sequence in K that converges to a point x D. Since {xk } is also a sequence in F and
F is a closed subset of R, x F. Thus, x D F = K.
Let us prove the converse. Suppose by contradiction that K is not closed in D or D \ K is not
open in D. Then there exists x D \ K such that for every > 0, one has
B(x;
) D * D \ K.
In particular, for every k N,


1
B x;

D * D \ K.
k
For each k N, choose xk B(x;
1k ) D such that xk
/ D \ K. Then {xk } is a sequence in K and,
moreover, {xk } converges to x D. Then x K. This is a contradiction. We conclude that K is
closed in D. 
The following theorem is now a direct consequence of Theorems 2.6.6 and 2.6.1.
Theorem 2.6.9 Let D be a subset of R. The following hold:

(1) The subsets 0/ and D are open in D.

(2) The union of any collection of open sets in D is open in D.
(3) The intersection of a finite number of open sets in D is open in D.

Exercises
2.6.1 Prove that a subset A of R is open if and only if for any x A, there exists n N such that
(x 1/n, x + 1/n) A.
2.6.2 Prove that the interval [0, 1) is neither open nor closed.
2.6.3 Find all limit points and all isolated points of each of the following sets:

(a)
(b)
(c)
(d)

A = (0, 1).
B = [0, 1).
C = Q.
D = {m + 1/n : m, n N}.

2.6.4 I A subset A of R is called compact if for every sequence {an } in A, there exists a subsequence
{ank } that converges to a point a A. Prove that a subset A of R is compact if and only if it is closed
and bounded.

LIMITS OF FUNCTIONS
LIMIT THEOREMS
CONTINUITY
PROPERTIES OF CONTINUOUS FUNCTIONS
UNIFORM CONTINUITY
LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

3. LIMITS AND CONTINUITY

In this chapter, we extend our analysis of limit processes to functions and give the precise
definition of continuous function. We derive rigorously two fundamental theorems about continuous
functions: the extreme value theorem and the intermediate value theorem.

3.1

LIMITS OF FUNCTIONS
Definition 3.1.1 Let f : D R and let x be a limit point of D. We say that f has a limit at x if there

exists a real number ` such that for every > 0, there exists > 0 with
| f (x) `| <
for all x D for which 0 < |x x|
< . In this case, we write
lim f (x) = `.

xx


Example 3.1.1 Let f : [0, 1) R be given by f (x) = x2 + x. Let x = 1 and ` = 2. Given > 0,

choose = /4. Then, if |x 1| < , we have

3
| f (x) `| = |x2 + x 2| |x 1||x + 2| 3 = < .
4
This shows that limx1 f (x) = 2.
Theorem 3.1.1 Sequential Characterization of Limits. Let f : D R and let x be a limit point

of D. Then
lim f (x) = `

xx

(3.1)

if and only if
lim f (xn ) = `

for every sequence {xn } in D such that xn 6= x for every n and {xn } converges to x.

(3.2)

3.1 LIMITS OF FUNCTIONS

54

Proof: Suppose (3.1) holds. Let {xn } be a sequence in D with xn 6= x for every n and such that {xn }
converges to x.
Given any > 0, there exists > 0 such that | f (x) `| < whenever x D and
0 < |x x|
< . Then there exists N N with 0 < |xn x|
< for all n N. For such n, we have
| f (xn ) `| < .
This implies (3.2).
Conversely, suppose (3.1) is false. Then there exists 0 > 0 such that for every > 0, there exists
x D with 0 < |x x|
< and | f (x) f (x)|
0 . Thus, for every n N, there exists xn D with
1
0 < |xn x|
< and | f (xn ) `| 0 . By the squeeze theorem (Theorem 2.1.5), the sequence {xn }
n
converges to x.
Moreover, xn 6= x for every n. This shows that (3.2) is false. It follows that (3.2)
implies (3.1) and the proof is complete. 
then this limit is
Corollary 3.1.2 Let f : D R and let x be a limit point of D. If f has a limit at x,
unique.
Proof: Suppose by contradiction that f has two different limits `1 and `2 . Let {xn } be a sequence
in D \ {x}
that converges to x.
By Theorem 3.1.1, the sequence { f (xn )} converges to two different
limits `1 and `2 . This is a contradiction to Theorem 2.1.3. 
The following corollary follows directly from Theorem 3.1.1.
Corollary 3.1.3 Let f : D R and let x be a limit point of D. Then f does not have a limit at x if
and only if there exists a sequence {xn } in D such that xn 6= x for every n, {xn } converges to x,
and
{ f (xn )} does not converge.


Example 3.1.2 Consider the Dirichlet function f : R R given by

(
1, if x Q;
f (x) =
0, if x Qc .
Then limxx f (x) does not exist for any x R. Indeed, fix x R and choose two sequences {rn },
{sn } converging to x such that rn Q and sn 6 Q for all n N. Define a new sequence {xn } by
(
rk , if n = 2k;
xn =
sk , if n = 2k 1.
It is clear that {xn } converges to x.
Moreover, since { f (rn )} converges to 1 and { f (sn )} converges
to 0, Theorem 2.1.8 implies that the sequence { f (xn )} does not converge. It follows from the
sequential characterization of limits that limxx f (x) does not exist.
Theorem 3.1.4 Let f , g : D R and let x be a limit point of D. Suppose that

lim f (x) = `1 , lim g(x) = `2 ,

xx

xx

and that there exists > 0 such that

f (x) g(x) for all x B(x;
) D, x 6= x.

Then `1 `2 .

55
Proof: Let {xn } be a sequence in B(x;
) D = (x , x + ) D that converges to x and xn 6= x for
all n. By Theorem 3.1.1,
lim f (xn ) = `1 and lim g(xn ) = `2 .

Since f (xn ) g(xn ) for all n N, applying Theorem 2.1.4, we obtain `1 `2 . 

Theorem 3.1.5 Let f , g : D R and let x be a limit point of D. Suppose
lim f (x) = `1 , lim g(x) = `2 ,

xx

xx

and `1 < `2 . Then there exists > 0 such that

f (x) < g(x) for all x B(x;
) D, x 6= x.

Proof: Choose > 0 such that `1 + < `2 (equivalently, such that <
> 0 such that

`2 `1
2 ).

Then there exists

`1 < f (x) < `1 + and `2 < g(x) < `2 +

for all x B(x;
) D, x 6= x.
Thus,
f (x) < `1 + < `2 < g(x) for all x B(x;
) D, x 6= x.

The proof is now complete. 

Theorem 3.1.6 Let f , g, h : D R and let x be a limit point of D. Suppose there exists > 0
such that f (x) g(x) h(x) for all x B(x;
) D, x 6= x.
If limxx f (x) = limxx h(x) = `, then
limxx g(x) = `.
Proof: The proof is straightforward using Theorem 2.1.5 and Theorem 3.1.1. 

Exercises
3.1.1 Use the definition of limit to prove that

(a) limx3 (x2 + 1) = 10.

x+3
(b) limx1
= 2.
x+1

(c) limx0 x = 0.
3.1.2 Prove that the following limits do not exist.

x
.
|x|
(b) limx0 cos(1/x).
(a) limx0

3.1.3 Let f : D R and let x be a limit point of D. Prove that if limxx f (x) = `, then

lim | f (x)| = |`|.

xx

Give an example to show that the converse is not true in general.

3.2 LIMIT THEOREMS

56

3.1.4 Let f : D R and let x be a limit point of D. Suppose f (x) 0 for all x D. Prove that if

limxx f (x) = `, then

p

lim f (x) = `.
xx

3.1.5 Find limx0 x sin(1/x).

3.1.6 I Let f be the function given by

(
x,
if x Q [0, 1];
f (x) =
1 x, if x Qc [0, 1].
Determine which of the following limits exist. For those that exist find their values.
(a) limx1/2 f (x).
(b) limx0 f (x).
(c) limx1 f (x).

3.2

LIMIT THEOREMS
Here we state and prove various theorems that facilitate the computation of general limits.
Definition 3.2.1 Let f , g : D R and let c be a constant. The functions f + g, f g, and c f are

respectively defined as functions from D to R by

( f + g)(x) = f (x) + g(x),
( f g)(x) = f (x)g(x),
(c f )(x) = c f (x)
f
for x D. Let D = {x D : g(x) 6= 0}. The function is defined as a function from D to R by
g
 
f (x)
f
(x) =
g
g(x)

for x D.
Theorem 3.2.1 Let f , g : D R and let c R. Suppose x is a limit point of D and

lim f (x) = `, lim g(x) = m.

xx

xx

Then
(1) limxx ( f + g)(x) = ` + m,
(2) limxx ( f g)(x) = `m,
(3) limxx (c f )(x) = c`,
 
f
`
(4) limxx
(x) = provided that m 6= 0.
g
m

57
Proof: Let us first prove (1). Let {xn } be a sequence in D that converges to x and xn 6= x for every n.
By Theorem 3.1.1,
lim f (xn ) = ` and lim g(xn ) = m.

It follows from Theorem 2.2.1 that

lim ( f (xn ) + g(xn )) = ` + m.

Applying Theorem 3.1.1 again, we get limxx ( f + g)(x) = ` + m. The proofs of (2) and (3) are
similar.
Since x is a limit point of D, there is
Let us now show that if m 6= 0, then x is a limit point of D.
a sequence {uk } in D converging to x such that uk 6= x for every k. Since m 6= 0, it follows from an
easy application of Theorem 3.1.5 that there exists > 0 with
g(x) 6= 0 whenever 0 < |x x|
< , x D.
This implies
x D whenever 0 < |x x|
< , x D.
The rest of the proof of (4)
Then uk D for all k sufficiently large, and hence x is a limit point of D.
can be completed easily following the proof of (1). 
Theorem 3.2.2 (Cauchys criterion) Let f : D R and let x be a limit point of D. Then f has a

limit at x if and only if for any > 0, there exists > 0 such that
| f (r) f (s)| < whenever r, s D and 0 < |r x|
< , 0 < |s x|
< .

(3.3)

Proof: Suppose limxx f (x) = `. Given > 0, there exists > 0 such that
| f (x) `| <

whenever x D and 0 < |x x|

< .
2

Thus, for r, s D with 0 < |r x|

< and 0 < |s x|
< , we have
| f (r) f (s)| | f (r) `| + |` f (s)| < .
Let us prove the converse. Fix a sequence {un } in D such with limn un = x and un 6= x for every n.
Given > 0, there exists > 0 such that
| f (r) f (s)| < whenever r, s D and 0 < |r x|
< , 0 < |s x|
< .
Then there exists N N satisfying
0 < |un x|
< for all n N.
This implies
| f (un ) f (um )| < for all m, n N.

3.2 LIMIT THEOREMS

58

Thus, { f (un )} is a Cauchy sequence, and hence there exists ` R such that
lim f (un ) = `.

We now prove that f has limit ` at x using Theorem 3.1.1. Let {xn } be a sequence in D such that
limn xn = x and xn 6= x for every n. By the previous argument, there exists `0 R such that
lim f (xn ) = `0 .

Fix any > 0 and let > 0 satisfy (3.3). There exists K N such that
|un x|
< and |xn x|
<
for all n K. Then | f (un ) f (xn )| < for such n. Letting n , we have |` `0 | . Thus, ` = `0
since is arbitrary. It now follows from Theorem 3.1.1 that limxx f (x) = `. 
Definition 3.2.2 Let a R and > 0. Define
B (a; ) = (a , a) and B+ (a; ) = (a, a + ).
Given a subset A of R, we say that a is a left limit point of A if for any > 0, B (a; ) contains
an infinite number of elements of A. Similarly, a is called a right limit point of A if for any > 0,
B+ (a; ) contains an infinite number of elements of A.
It follows from the definition that a is a limit point of A if and only if it is a left limit point of A
or it is a right limit point of A.
Definition 3.2.3 (One-sided limits) Let f : D R and let x be a left limit point of D. We write
lim f (x) = `

xx

if for every > 0, there exists > 0 such that

| f (x) `| < for all x B (x;
).
We say that ` is the left-hand limit of f at x.
The right-hand limit of f at x can be defined in a similar
way and is denoted limxx+ f (x).
The following theorem follows directly from the definition of one-sided limits.
Theorem 3.2.3 Let f : D R and let x be both a left limit point of D and a right limit point of D.
Then
lim f (x) = `

xx

if and only if
lim f (x) = ` and lim f (x) = `.

xx+

xx

Definition 3.2.4 (monotonicity) Let f : (a, b) R.

(1) We say that f is increasing on (a, b) if, for all x1 , x2 (a, b), x1 < x2 implies f (x1 ) f (x2 ).
(2) We say that f is decreasing on (a, b) if, for all x1 , x2 (a, b), x1 < x2 implies f (x1 ) f (x2 ).
If f is increasing or decreasing on (a, b), we say that f is monotone on this interval. Strict monotonicity can be defined similarly using strict inequalities: f (x1 ) < f (x2 ) in (1) and f (x1 ) > f (x2 )
in (2).

59
Theorem 3.2.4 Suppose f : (a, b) R is increasing on (a, b) and x (a, b). Then limxx f (x)

and limxx+ f (x) exist. Moreover,

sup f (x) = lim f (x) f (x)
lim+ f (x) = inf f (x).

a<x<x

xx

xx

x<x<b

Proof: Since f (x) f (x)

for all x (a, x),
the set
{ f (x) : x (a, x)}

is nonempty and bounded above. Thus,

` = sup f (x)
a<x<x

is a real number. We will show that limxx f (x) = `. For any > 0, by the definition of the least
upper bound, there exists a < x1 < x such that
` < f (x1 ).
Let = x x1 > 0. Using the increasing monotonicity, we get
` < f (x1 ) f (x) ` < ` + for all x (x1 , x)
= B (x;
).
Therefore, limxx f (x) = `. The rest of the proof of the theorem is similar. 
Let
B0 (x;
) = B (x;
) B+ (x;
).
Definition 3.2.5 (infinite limits) Let f : D R and let x be a limit point of D. We write

lim f (x) =

xx

if for every M R, there exists > 0 such that

f (x) > M for all x B0 (x;
) D.
Similarly, we write
lim f (x) =

xx

if for every M R, there exists > 0 such that

f (x) < M for all x B0 (x;
) D.
Infinite limits of functions have similar properties to those of sequences from Chapter 2 (see
Definition 2.3.2 and Theorem 2.3.7).

3.2 LIMIT THEOREMS

60

Definition 3.2.6 (limits at infinity) Let f : D R, where D is not bounded above. We write

lim f (x) = `

if for every > 0, there exists c R such that

| f (x) `| < for all x > c, x D.
Let f : D R, where D is not bounded below. We write
lim f (x) = `

if for every > 0, there exists c R such that

| f (x) `| < for all x < c, x D.
We can also define
lim f (x) = and lim f (x) =

in a similar way.
Next, we extend to functions the concepts of limit superior and limit inferior.
Definition 3.2.7 Let f : E R and let x be a limit point of D. Recall that
B0 (x;
) = B (x;
) B+ (x;
) = (x , x)
(x,
x + ).
The limit superior of the function f at x is defined by
lim sup f (x) = inf

sup

f (x).

>0 xB0 (x;

)D

xx

Similarly, the limit inferior of the function f at x is defined by

lim inf f (x) = sup
xx

inf

f (x).

)D
>0 xB0 (x;

Consider the extended real-valued function g : (0, ) (, ] defined by

g( ) =

f (x).

sup

(3.4)

xB0 (x;
)D

It is clear that g is increasing and

lim sup f (x) = inf g( ).
xx

>0

We say that the function f is locally bounded above around x if there exists > 0 and M > 0 such
that
f (x) M for all x B(x;
) D.
Clearly, if f is locally bounded above around x,
then lim supxx f (x) is a real number, while
lim supxx f (x) = in the other case. Similar discussion applies for the limit inferior.

61
Theorem 3.2.5 Let f : D R and let x be a limit point of D. Then ` = lim supxx f (x) if and only

if the following two conditions hold:

(1) For every > 0, there exists > 0 such that
f (x) < ` + for all x B0 (x;
) D;
(2) For every > 0 and for every > 0, there exists x B0 (x;
) D such that
` < f (x ).
Proof: Suppose ` = lim supxx f (x). Then
` = inf g( ),
>0

where g is defined in (3.4). For any > 0, there exists > 0 such that
` g( ) =

f (x) < ` + .

sup
xB0 (x;
)D

Thus,
f (x) < ` + for all x B0 (x;
) D.
For any > 0 and > 0, we have
` < ` g( ) =

f (x).

sup
xB0 (x;
)D

Thus, there exists x B0 (x;

) D with
` < f (x ).
Let us prove the converse. Suppose (1) and (2) are satisfied. Fix any > 0 and let > 0 satisfy (1).
Then
g( ) =

f (x) ` + .

sup
xB0 (x;
)D

This implies
lim sup f (x) = inf g( ) ` + .
>0

xx

Since is arbitrary, we get

lim sup f (x) `.
xx

Again, let > 0. Given > 0, let x be as in (2). Therefore,

` < f (x )

sup

f (x) = g( ).

xB0 (x;
)D

This implies
` inf g( ) = lim sup f (x).
>0

xx

It follows that ` lim supxx f (x). Therefore, ` = lim supxx f (x). 

3.2 LIMIT THEOREMS

62

Corollary 3.2.6 Suppose ` = lim supxx f (x). Then there exists a sequence {xk } in D such that

{xk } converges to x,
xk 6= x for every k, and
lim f (xk ) = `.

Moreover, if {yk } is a sequence in D that converges to x,

yk 6= x for every k, and limk f (yk ) = `0 ,
0
then ` `.
1
Proof: For each k N, take k = . By (1) of Theorem 3.2.5, there exists k > 0 such that
k
f (x) < ` + k for all x B0 (x;
k ) D.

(3.5)

Let k0 = min{k , 1k }. Then k0 k and limk k0 = 0. From (2) of Theorem 3.2.5, there exists
xk B0 (x;
k0 ) D such that
` k < f (xk ).
Moreover, f (xk ) < ` + k by (3.5). Therefore, {xk } is a sequence that satisfies the conclusion of the
corollary.
Now let {yk } be a sequence in D that converges to x,
yk 6= x for every k, and limk f (yk ) = `0 .
For any > 0, let > 0 be as in (1) of Theorem 3.2.5. Since yk B0 (x;
) D when k is sufficiently
large, we have
f (yk ) < ` +
for such k. This implies `0 ` + . It follows that `0 `. 
Remark 3.2.7 Let f : D R and let x be a limit point of D. Suppose lim supxx f (x) is a real

number. Define
A = {` R : {xk } D, xk 6= x for every k, xk x,
f (xk ) `}.
Then the previous corollary shows that A 6= 0/ and lim supxx f (x) = max A.
Theorem 3.2.8 Let f : D R and let x be a limit point of D. Then

lim sup f (x) =

xx

if and only if there exists a sequence {xk } in D such that {xk } converges to x,
xk 6= x for every k, and
limk f (xk ) = .
Proof: Suppose lim supxx f (x) = . Then
inf g( ) = ,

>0

where g is the extended real-valued function defined in (3.4). Thus, g( ) = for every > 0. Given
1
k N, for k = , since
k
g(k ) =

sup
xB0 (x;
k )D

f (x) = ,

63
there exists xk B0 (x;
k ) D such that f (xk ) > k. Therefore, limk f (xk ) = .
Let us prove the converse. Since limk f (xk ) = , for every M R, there exists K N such
that
f (xk ) M for every k K.
For any > 0, we have
xk B0 (x;
)D
whenever k is sufficiently large. Thus,
g( ) =

sup

f (x) M.

xB0 (x;
)D

This implies g( ) = , and hence lim supxx f (x) = . 

Theorem 3.2.9 Let f : D R and let x be a limit point of D. Then
lim sup f (x) =
xx

if and only if for any sequence {xk } in D such that {xk } converges to x,
xk 6= x for every k, it follows
that limk f (xk ) = . The latter is equivalent to limxx f (x) = .
Following the same arguments, we can prove similar results for inferior limits of functions.
Theorem 3.2.10 Let f : D R and let x be a limit point of D. Then ` = lim infxx f (x) if and only
if the following two conditions hold:
(1) For every > 0, there exists > 0 such that
` < f (x) for all x B0 (x;
) D;
(2) For every > 0 and for every > 0, there exists x B0 (x;
) D such that
f (x) < ` + .
Corollary 3.2.11 Suppose ` = lim infxx f (x). Then there exists a sequence {xk } in D such that xk

converges to x,
xk 6= x for every k, and
lim f (xk ) = `.

Moreover, if {yk } is a sequence in D that converges to x,

yk 6= x for every k, and limk f (yk ) = `0 ,
0
then ` `.
Remark 3.2.12 Let f : D R and let x be a limit point of D. Suppose lim infxx f (x) is a real

number. Define
B = {` R : {xk } D, xk 6= x for every k, xk x,
f (xk ) `}.
Then B 6= 0/ and lim infxx f (x) = min B.

3.2 LIMIT THEOREMS

64
Theorem 3.2.13 Let f : D R and let x be a limit point of D. Then

lim inf f (x) =

xx

if and only if there exists a sequence {xk } in D such that {xk } converges to x,
xk 6= x for every k, and
limk f (xk ) = .
Theorem 3.2.14 Let f : D R and let x be a limit point of D. Then
lim inf f (x) =
xx

if and only if for any sequence {xk } in D such that {xk } converges to x,
xk 6= x for every k, it follows
that limk f (xk ) = . The latter is equivalent to limxx f (x) = .
Theorem 3.2.15 Let f : D R and let x be a limit point of D. Then
lim f (x) = `

xx

if and only if
lim sup f (x) = lim inf f (x) = `.
xx

xx

Proof: Suppose
lim f (x) = `.

xx

Then for every > 0, there exists > 0 such that

` < f (x) < ` + for all x B0 (x;
) D.
Since this also holds for every 0 < 0 < , we get
` < g( 0 ) ` + .
It follows that
` inf
g( 0 ) ` + .
0
>0

Therefore, lim supxx f (x) = ` since is arbitrary. The proof for the limit inferior is similar. The
converse follows directly from (1) of Theorem 3.2.5 and Theorem 3.2.10. 

Exercises
3.2.1 Let f : D R and let x is a limit point of D. Prove that if limxx f (x) exists, then

lim[ f (x)]n = [lim f (x)]n , for any n N .

xx

3.2.2 Find the

following limits:

(a) limx1
(b) limx1

x1

x2 1
xm 1
xn 1

, where m, n N.

xx

65

n
x1
(c) limx1
, where m, n N, m, n 2.
m
1
x
x 3 x
(d) limx1
.
x1
3.2.3 Find thefollowing limits:

x3 + 3x2
x2 + 1).
(a) limx ( 3
3 3
2
(b) limx ( x + 3x x2 + 1).
3.2.4 I Let f : D R and let x be a limit point of D. Suppose that
| f (x) f (y)| k|x y| for all x, y D \ {x},

where k 0 is a constant. Prove that limxx f (x) exists.

3.2.5 For a R, let f be the function given by

(
x2 ,
if x > 1;
f (x) =
ax 1, if x 1.
Find the value of a such that limx1 f (x) exists.
3.2.6 Determine all values of x such that the limit limxx (1 + x [x]) exits, where [x] denotes the
greatest integer that is less than or equal to x.
3.2.7 Find each of the following limits if they exist:

x+1
.
1
x

3
(b) limx0+ x sin(1/x).
(c) limx1 (x [x]).
(a) limx1+

3.2.8 Find each of the

limits:

 following

1
.
x
 
1
(b) lim infx0 sin
.
x
cos x
(c) lim supx0
.
x
cos x
(d) lim infx0
.
x
(a) lim supx0 sin

3.3

CONTINUITY
Definition 3.3.1 Let D be a nonempty subset of R and let f : D R be a function. The function f

is said to be continuous at x0 D if for any real number > 0, there exists > 0 such that if x D

3.3 CONTINUITY

66
and |x x0 | < , then
| f (x) f (x0 )| < .

If f is continuous at every point x D, we say that f is continuous on D (or just continuous if no

confusion occurs).
Example 3.3.1 Let f : R R be given by f (x) = 3x + 7. Let x0 R and let > 0. Choose
= /3. Then if |x x0 | < , we have


| f (x) f (x0 )| = |3x + 7 (3x0 + 7)| = |3(x x0 )| = 3|x x0 | < 3 = .

This shows that f is continuous at x0 .

Figure 3.1: Definition of continuity.

Remark 3.3.1 Note that the above definition of continuity does not mention limits. This allows

to include in the definition, points x0 D which are not limit points of D. If x0 is an isolated
point of D, then there is > 0 such that B(x0 ; ) D = {x0 }. It follows that for x B(x0 ; ) D,
| f (x) f (x0 )| = 0 < for any epsilon. Therefore, every function is continuous at an isolated point
of its domain.
To study continuity at limit points of D, we have the following theorem which follows directly
from the definitions of continuity and limit.
Theorem 3.3.2 Let f : D R and let x0 D be a limit point of D. Then f is continuous at x0 if
and only if
lim f (x) = f (x0 ).

xx0


Example 3.3.2 Let f : R R be given by f (x) = 3x2 2x + 1. Fix x0 R. Since, from the

results of the previous theorem, we have

lim f (x) = lim 3x2 2x + 1 = 3x02 2x0 + 1 = f (x0 ),

xx0

xx0

it follows that f is continuous at x0 .

Definition 3.3.2 (Hlder continuity). Let D be a nonempty subset of R. A function f : D R is

said to be Hlder continuous if there are constants ` 0 and > 0 such that
| f (u) f (v)| `|u v| for every u, v D.

67
The number is called the Holder exponent of the function. If = 1, then the function f is called
Lipschitz continuous.
Theorem 3.3.3 If a function f : D R is Holder continuous, then it is continuous.
Proof: Fix any x0 D. We will prove that f is continuous at x0 . Since f is Holder continuous, there
are constants ` 0 and > 0 such that
| f (u) f (v)| `|u v| for every u, v D.
This implies
| f (x) f (x0 )| `|x x0 | for every x D.
If ` = 0, then f is constant and, thus, continuous. Suppose next that ` > 0. For any > 0, let

1/
. Then
= `
| f (x) f (x0 )| < whenever x D and |x x0 | < .
The proof is now complete. 

Figure 3.2: The square root function.

x is Lipschitz
continuous on D and, hence, continuous on this set. Indeed, for any u, v D, we have

Example 3.3.3 (1) Let D = [a, ), where a > 0. Then the function f (x) =

|u v|
1
|u v|.
| f (u) f (v)| = | u v| =
u+ v 2 a
(2) Let D = [0, ). Then f is not Lipschitz continuous on D, but it is Holder continuous on D and,
hence, f is also continuous on this set.
Indeed, suppose by contradiction that f is Lipschitz continuous on D. Then there exists a
constant ` > 0 such that

| f (u) f (v)| = | u v| `|u v| for every u, v D.

Thus, for every n N, we have

This implies








 1
 0 `  1 0 .
 n

n


n ` or n `2 for every n N .

3.3 CONTINUITY

68
This is a contradiction. Therefore, f is not Lipschitz continuous on D.
Let us show that f is Holder continuous on D. We are going to prove that
| f (u) f (v)| |u v|1/2 for every u, v D.

(3.6)

The inequality in (3.6) holds obviously for u = v = 0. For u > 0 or v > 0, we have

| f (u) f (v)| = | u v|


 uv 

= 
u+ v
p
p
|u v|

= |u v|
u+ v
p
|u| + |v| p

|u v|
u+ v
p
= |u v|.
Note that one can justify the inequality
p
|u| + |v|

1
u+ v
by squaring both sides since they are both positive. Thus, (3.6) is satisfied.
The proofs of the next two theorems are straightforward using Theorem 3.4.1.
Theorem 3.3.4 Let f , g : D R and let x0 D. Suppose f and g are continuous at x0 . Then
(1) f + g and f g are continuous at x0 .
(2) c f is continuous at x0 for any constant c.
f
(3) If g(x0 ) 6= 0, then (defined on D = {x D : g(x) 6= 0}) is continuous at x0 .
g
Theorem 3.3.5 Let f : D R and let g : E R with f (D) E. If f is continuous at x0 and g is
continuous at f (x0 ), then g f is continuous at x0 .

Exercises
3.3.1 Prove that each of the following functions is continuous on the given domain:

(a) f (x) = ax + b, a, b R, on R.
(b) f (x) = x2 3 on R.

(c) f (x) = x on [0, ).

3.3.2 Determine the values of x at which each function is continuous. The domain of all the

functions is R.

 sin x  , if x 6= 0;
(a) f (x) =  x 

1,
if x = 0.

sin x , if x 6= 0;
|x|
(b) f (x) =

1,
if x = 0.

69

x sin 1 ,
(c) f (x) =
x
0,

cos x ,
2
(d) f (x) =
|x 1|,
(e) f (x) = lim sin
n

if x 6= 0;
if x = 0.
if |x| 1;
if |x| > 1.

,
2(1 + x2n )

x R.

3.3.3 Let f : R R be the function given by

(
x2 + a, if x > 2;
f (x) =
ax 1, if x 2.
Find the value of a such that f is continuous.
3.3.4 Let f : D R and let x0 D. Prove that if f is continuous at x0 , then | f | is continuous at

this point. Is the converse true in general?

3.3.5 I

(a) Let g, h : [0, 1] R be continuous functions. Define

(
g(x), if x Q [0, 1];
f (x) =
h(x), if x Qc [0, 1].
Prove that if g(a) = h(a), a [0, 1], then f is continuous at a.
(b) Let f : [0, 1] R be the function given by
(
x,
if x Q [0, 1];
f (x) =
1 x, if x Qc [0, 1].
Find all the points on [0, 1] at which the function is continuous.
3.3.6 B Consider the Thomae function defined on (0, 1] by

1 , if x = p , p, q N, where p and q have no common factors;

q
f (x) = q

0, if x is irrational.
(a) Prove that for every > 0, the set
A = {x (0, 1] : f (x) }
is finite.
(b) Prove that f is continuous at every irrational point, and discontinuous at every rational point.
3.3.7 B Consider k distinct points x1 , x2 , . . . , xk R, k 1. Find a function defined on R that is
continuous at each xi , i = 1, . . . , k, and discontinuous at all other points.

3.4 PROPERTIES OF CONTINUOUS FUNCTIONS

70

3.4

PROPERTIES OF CONTINUOUS FUNCTIONS

The following theorem follows directly from the definition of continuity, Theorem 3.1.1 and
Theorem 3.3.2.
Theorem 3.4.1 Let f : D R and let x0 D. Then f is continuous at x0 if and only if for any
sequence {xk } in D that converges to x0 , the sequence { f (xk )} converges to f (x0 ).
Recall from Exercise 2.6.4 that a subset A of R is compact if and only if every sequence {an } in
A has a subsequence {ank } that converges to a point a A.
Theorem 3.4.2 Let D be a nonempty compact subset of R and let f : D R be a continuous
function. Then f (D) is a compact subset of R. In particular, f (D) is closed and bounded.
Proof: Take any sequence {yn } in f (D). Then for each n, there exists an D such that yn = f (an ).
Since D is compact, there exists a subsequence {ank } of {an } and a point a D such that
lim ank = a D.

This implies
lim ynk = lim f (ank ) = f (a) f (D).

Therefore, f (D) is compact. 

Definition 3.4.1 We say that the function f : D R has an absolute minimum at x D if
f (x) f (x)
for every x D.
Similarly, we say that f has an absolute maximum at x if
f (x) f (x)
for every x D.

Figure 3.3: Absolute maximum and absolute minimum of f on [a, b].

Theorem 3.4.3 Extreme Value Theorem. Suppose f : D R is continuous and D is a compact

set. Then f has an absolute minimum and an absolute maximum on D.

Proof: Since D is compact, A = f (D) is closed and bounded (see Exercise 2.6.4). Let
m = inf A = inf f (x).
xD

In particular, m R. For every n N, there exists an A such that

m an < m + 1/n.

71
For each n, since an A = f (D), there exists xn D such that an = f (xn ) and, hence,
m f (xn ) < m + 1/n.
By the compactness of D, there exists an element x D and a subsequence {xnk } that converges to
x D as k . Because
1
m f (xnk ) < m + for every k,
nk
by the squeeze theorem (Theorem 2.1.5) we conclude limk f (xnk ) = m. On the other hand, by
continuity we have limk f (xnk ) = f (x).
We conclude that f (x)
= m f (x) for every x D. Thus,
f has an absolute minimum at x.
The proof is similar for the case of absolute maximum. 
Remark 3.4.4 The proof of Theorem 3.4.3 can be shortened by applying Theorem 2.6.3. However,
we have provided a direct proof instead.
Corollary 3.4.5 If f : [a, b] R is continuous, then it has an absolute minimum and an absolute

maximum on [a, b].

Corollary 3.4.5 is sometimes referred to as the Extreme Value Theorem. While it follows
immediately from Theorem 3.4.3, we provide below a direct proof using the Bolzano-Weierstrass
Theorem (Theorem 2.4.1).
Direct Proof of Corollary 3.4.5: We prove first that A = f ([a, b]) is bounded. Suppose, by way of
contradiction, that A is not bounded. Then, for each n N, there is xn [a, b] such that | f (xn )| > n. In
particular, a xn b for all n N. Since the sequence {xn } is bounded, by the Bolzano-Weierstrass
Theorem (Theorem 2.4.1), it has a convergent subsequence {xnk }. Say, xnk c. By the Comparison
Theorem (Theorem 2.1.4), we obtain c [a, b]. Then, by the continuity of f , we get f (xnk ) f (c).
Therefore the sequence { f (xnk )} is bounded. On the other hand, since | f (xnk )| > nk for each k and
nk , we arrive at a contradiction. We have thus proved that A is bounded.
Set M = sup A. We prove next that there is xM [a, b] such that f (xM ) = M. Suppose, by
contradiction, that f (x) < M for all x [a, b]. Define g : [a, b] R by
g(x) =

1
.
M f (x)

Then g is continuous. By the result proved so far, g([a, b]) is bounded above. So, there is K > 0 such
that g(x) K for all x [a, b]. Therefore, for x [a, b], we get
1
K
M f (x)
and, thus,
f (x) M

1
.
K

It follows that M 1/K is an upper bound of f ([a, b]) contradicting the fact that M is the least upper
bound of A = f ([a, b]). This contradiction shows that there is xM [a, b] such that f (xM ) = M and,
hence, f attains its absolute maximum on [a, b]. A similar argument proves the result for absolute
minimum. .
The following result is a basic property of continuous functions that is used in a variety of
situations.

3.4 PROPERTIES OF CONTINUOUS FUNCTIONS

72

Lemma 3.4.6 Let f : D R be continuous at c D. Suppose f (c) > 0. Then there exists > 0

such that
f (x) > 0 for every x B(c; ) D.
Proof: Let = f (c) > 0. By the continuity of f at c, there exists > 0 such that if x D and
|x c| < , then
| f (x) f (c)| < .
This implies, in particular, that f (x) > f (c) = 0 for every x B(c; ) D. The proof is now
complete. 
Remark 3.4.7 An analogous result holds if f (c) < 0.
Theorem 3.4.8 Let f : [a, b] R be a continuous function. Suppose f (a) f (b) < 0. Then there

exists c (a, b) such that f (c) = 0.

Proof: We prove only the case f (a) < 0 < f (b) (the case f (a) > 0 > f (b) is completely analogous).
Define
A = {x [a, b] : f (x) 0}.
This set is nonempty since a A. This set is also bounded since A [a, b]. Therefore, c = sup A
exists and a c b. We are going to prove that f (c) = 0 by showing that f (c) < 0 and f (c) > 0
lead to contradictions.
Suppose f (c) < 0. Then there exists > 0 such that
f (x) < 0 for all x B(c; ) [a, b].
Because c < b (since f (b) > 0), we can find s (c, b) such that f (s) < 0 (indeed s = min{c +
/2, (c + b)/2} will do). This is a contradiction because s A and s > c.
Suppose f (c) > 0. Then there exists > 0 such that
f (x) > 0 for all x B(c; ) [a, b].
Since a < c (because f (a) < 0), there exists t (a, c) such that f (x) > 0 for all x (t, c) (in fact,
t = max{c /2, (a + c)/2} will do). On the other hand, since t < c = sup A, there exists t 0 A
with t < t 0 c. But then t < t 0 and f (t 0 ) 0. This is a contradiction. We conclude that f (c) = 0. 

Figure 3.4: Illustration of the Intermediate Value Theorem.

73
Theorem 3.4.9 Intermediate Value Theorem. Let f : [a, b] R be a continuous function.

Suppose f (a) < < f (b). Then there exists a number c (a, b) such that f (c) = .
The same conclusion follows if f (a) > > f (b).
Proof: Define
(x) = f (x) , x [a, b].
Then is continuous on [a, b]. Moreover,
(a)(b) = [ f (a) ][ f (b) ] < 0.
By Theorem 3.4.8, there exists c (a, b) such that (c) = 0. This is equivalent to f (c) = . The
proof is now complete. 
We present below a second proof that depends directly on the Nested Intervals Theorem (Theorem 2.3.4).
Second Proof of Theorem 3.4.9: We construct a sequence of nested intervals as follows. Set
a1 = a, b1 = b, and let I1 = [a, b]. Let c1 = (a + b)/2. If f (c1 ) = , we are done. Otherwise, either
f (c1 ) >

or

f (c1 ) < .
In the first case, set a2 = a1 and b1 = c1 . In the second case, set a2 = c1 and b2 = b1 . Now set
I2 = [a2 , b2 ]. Note that in either case,
f (a2 ) < < f (b2 ).
Set c2 = (a2 + b2 )/2. If f (c2 ) = , again we are done. Otherwise, either
f (c2 ) >

or

f (c2 ) < .
In the first case, set a3 = a2 and b3 = c2 . In the second case, set a3 = c2 and b3 = b2 . Now set
I3 = [a3 , b3 ]. Note that in either case,
f (a3 ) < < f (b3 ).
Proceeding in this way, either we find some cn0 such that f (cn0 ) = and, hence, the proof is complete,
or we construct a sequence of closed bounded intervals {In } with In = [an , bn ] such that for all n,
(i) In In+1 ,
(ii) bn an = (b a)/2n1 , and
(iii) f (an ) < < f (bn ).
In this case, we proceed as follows. Condition (ii) implies that limn (bn an ) = 0. By the Nested
T
Intervals Theorem (Theorem 2.3.4, part (b)), there exists c [a, b] such that
n=1 In = {c}. Moreover,
as we see from the proof of that theorem, an c and bn c as n .
By the continuity of f , we get
lim f (an ) = f (c)

lim f (bn ) = f (c).

and

3.4 PROPERTIES OF CONTINUOUS FUNCTIONS

74

Since f (an ) < < f (bn ) for all n, condition (iii) above and Theorem 2.1.4 give
f (c)

and

f (c) .
It follows that f (c) = . Note that, since f (a) < < f (b), then c (a, b). The proof is now
complete. 
Corollary 3.4.10 Let f : [a, b] R be a continuous function. Let

m = min{ f (x) : x [a, b]} and M = max{ f (x) : x [a, b]}.

Then for every [m, M], there exists c [a, b] such that f (c) = .


Example 3.4.1 We show now that, given n N, every positive real number has a positive n-th root.

Let n N and let a R with a > 0. First observe that (1 + a)n 1 + na > a (see Exercise 1.3.4).
Now consider the function f : [0, ) R given by f (x) = xn . Since f (0) = 0 and f (1 + a) > a, it
follows from the Intermediate Value Theorem that there is x (0, 1 + a) such that f (x) = a. That is,
xn = a, as desired. (We show in Example 4.3.1 that such an x is unique.)
Now we are going to discuss the continuity of the inverse function. For a function f : D E,
where E is a subset of R, we can define the new function f : D R by the same function notation.
The function f : D E is said to be continuous at a point x D if the corresponding function
f : D R is continuous at x.

Theorem 3.4.11 Let f : [a, b] R be strictly increasing and continuous on [a, b]. Let c = f (a) and
d = f (b). Then f is one-to-one, f ([a, b]) = [c, d], and the inverse function f 1 defined on [c, d] by
f 1 ( f (x)) = x where x [a, b],
is a continuous function from [c, d] onto [a, b].
Proof: The first two assertions follow from the monotonicity of f and the Intermediate Value
Theorem (see also Corollary 3.4.10). We will prove that f 1 is continuous on [c, d]. Fix any y [c, d]
and fix any sequence {yk } in [c, d] that converges to y.
Let x [a, b] and xk [a, b] be such that
f (x)
= y and f (xk ) = yk for every k.
Then f 1 (y)
= x and f 1 (yk ) = xk for every k. Suppose by contradiction that {xk } does not converge
to x.
Then there exist 0 > 0 and a subsequence {xk` } of {xk } such that
|xk` x|
0 for every `.

(3.7)

Since the sequence {xk` } is bounded, it has a further subsequence that converges to x0 [a, b]. To
simplify the notation, we will again call the new subsequence {xk` }. By the continuity of f , { f (xk` )}
converges to f (x0 ). Since f (xk` ) = yk` y as ` , it follows that f (x0 ) = y = f (x).
This implies
x0 = x,
which contradicts (3.7). 
A similar theorem can be written for strictly decreasing functions.

Exercises
3.4.1 Let f , g be continuous functions on [a, b]. Suppose f (a) < g(a) and f (b) > g(b). Prove that
there exists x0 (a, b) such that f (x0 ) = g(x0 ).

75
3.4.2 Suppose that f , g are continuous functions on R and f (x) = g(x) for all x Q. Prove that

f (x) = g(x) for all x R.

3.4.3 Let f : [a, b] [a, b] be a continuous function.

(a) Prove that the equation f (x) = x has a solution on [a, b].
(b) Suppose further that
| f (x) f (y)| < |x y| for all x, y [a, b], x 6= y.
Prove that the equation f (x) = x has a unique solution on [a, b].
3.4.4 B Let f be a continuous function on [a, b] and x1 , x2 , . . . , xn [a, b]. Prove that there exists

c [a, b] with
f (c) =

f (x1 ) + f (x2 ) + f (xn )

.
n

3.4.5 B Suppose f is a continuous function on R such that

| f (x)| < |x| for all x 6= 0.

(a) Prove that f (0) = 0.
(b) Given two positive numbers a and b with a < b, prove that there exists ` [0, 1) such that
| f (x)| `|x| for all x [a, b].
3.4.6 I Let f , g : [0, 1] [0, 1] be continuous functions such that

f (g(x)) = g( f (x)) for all x [0, 1].

Suppose further that f is monotone. Prove that there exists x0 [0, 1] such that
f (x0 ) = g(x0 ) = x0 .

3.5

UNIFORM CONTINUITY
We discuss here a stronger notion of continuity.
Definition 3.5.1 Let D be a nonempty subset of R. A function f : D R is called uniformly
continuous on D if for any > 0, there exists > 0 such that if u, v D and |u v| < , then
| f (u) f (v)| < .
It follows immediately from the definition that if f is Hlder continuous on D (see Definition 3.3.2), then f is uniformly continuous on this set.

It follows from Example 3.3.3 that the function f (x) = x is uniformly continuous on D = [0, ).
It is also straightforward that if a function is uniformly continuous on D, then it is continuous at
every point of D. The following example shows that the converse is not true in general.

3.5 UNIFORM CONTINUITY

76

Figure 3.5: Continuous but not uniformly continuous on (0, ).

Example 3.5.1 Let f : (0, 1) R be given by

1
f (x) = .
x
We already know that this function is continuous at every x (0, 1). We will show that f is not
uniformly continuous on (0, 1). Let = 2 and > 0. Set 0 = min{ /2, 1/4}, x = 0 , and y = 20 .
Then x, y (0, 1) and |x y| = 0 < , but

 
 
 

 1 1   y x   0   1 
=
=
 2 = .
| f (x) f (y)| =   = 
x y
xy   202   20 
This shows f is not uniformly continuous on (0, 1).
The following theorem offers a sequential characterization of uniform continuity analogous to
that in Theorem 3.4.1.
Theorem 3.5.1 Let D be a nonempty subset of R and f : D R. Then f is uniformly continuous
on D if and only if the following condition holds
(C) for every two sequences {un }, {vn } in D such that limn (un vn ) = 0, it follows that
limn ( f (un ) f (vn )) = 0.
Proof: Suppose first that f is uniformly continuous and let {un }, {vn } be sequences in D
such that limn (un vn ) = 0. Let > 0. Choose > 0 such that | f (u) f (v)| < whenever
u, v D and |u v| < . Let N N be such that |un vn | < for n N. For such n, we have
| f (un ) f (vn )| < . This shows limn ( f (un ) f (vn )) = 0.
To prove the converse, assume condition (C) holds and suppose, by way of contradiction, that f
is not uniformly continuous. Then there exists 0 > 0 such that for any > 0, there exist u, v D

77
with
|u v| < and | f (u) f (v)| 0 .
Thus, for every n N, there exist un , vn D with
|un vn | 1/n and | f (un ) f (vn )| 0 .
It follows that for such sequences, limn (un vn ) = 0, but { f (un ) f (vn )} does not converge to
zero, which contradicts the assumption. 
Example 3.5.2 Using this theorem, we can give an easier proof that the function in Example 3.5.1
is not uniformly continuous. Consider the two sequences un = 1/(n + 1) and vn = 1/n for all n 2.
Then clearly, limn (un vn ) = 0, but



lim ( f (un ) f (vn )) = lim

1
1

1/(n + 1) 1/n


= lim (n + 1 n) = 1.
n

The following theorem shows one important case in which continuity implies uniform continuity.
Theorem 3.5.2 Let f : D R be a continuous function. Suppose D is compact. Then f is uniformly
continuous on D.
Proof: Suppose by contradiction that f is not uniformly continuous on D. Then there exists 0 > 0
such that for any > 0, there exist u, v D with
|u v| < and | f (u) f (v)| 0 .
Thus, for every n N, there exist un , vn D with
|un vn | 1/n and | f (un ) f (vn )| 0 .
Since D is compact, there exist u0 D and a subsequence {unk } of {un } such that
unk u0 as k .
Then
|unk vnk |

1
,
nk

for all k and, hence, we also have

vnk u0 as k .
By the continuity of f ,
f (unk ) f (u0 ) and f (vnk ) f (u0 ).
Therefore, { f (unk ) f (vnk )} converges to zero, which is a contradiction. The proof is now complete. 

3.6 LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

78

Exercises
3.5.1 Prove that each of the following functions is uniformly continuous on the given domain:

(a) f (x) = ax + b, a, b R, on R.
(b) f (x) = 1/x on [a, ), where a > 0.
3.5.2 I Prove that each of the following functions is not uniformly continuous on the given domain:

(a) f (x) = x2 on R.
1
(b) f (x) =
on (0, 1).
sin x
(c) f (x) = ln(x) on (0, ).
3.5.3 B Suppose that f is uniformly continuous on (a, b). Prove that limxa+ f (x) and limxb f (x)

exist.
3.5.4 Let a, b R and let f : (a, b) R.

(a) Prove that if f is uniformly continuous, then f is bounded.

(b) Prove that if f is continuous, bounded, and monotone, then it is uniformly continuous.
3.5.5 Let D be a nonempty subset of R and let f : D R.

Suppose that f is uniformly continuous on D. Prove that if {xn } is a Cauchy sequence with xn D
for every n N, then { f (xn )} is also a Cauchy sequence.
3.5.6 B Let f be a continuous function on [a, ). Suppose

lim f (x) = c.

(a) Prove that f is bounded on [a, ).

(b) Prove that f is uniformly continuous on [a, ).
(c) Suppose further that c > f (a). Prove that there exists x0 [a, ) such that
f (x0 ) = inf{ f (x) : x [a, )}.

3.6

LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

The concept of semicontinuity is convenient for the study of maxima and minima of some
discontinuous functions.
Definition 3.6.1 Let f : D R and let x D. We say that f is lower semicontinuous (l.s.c.) at x if
for every > 0, there exists > 0 such that
f (x)
< f (x) for all x B(x;
) D.

(3.8)

Similarly, we say that f is upper semicontinuous (u.s.c.) at x if for every > 0, there exists > 0
such that
f (x) < f (x)
+ for all x B(x;
) D.

79

Figure 3.6: Lower semicontinuity.

Figure 3.7: Upper semicontinuity.

It is clear that f is continuous at x if and only if f is lower semicontinuous and upper semicontinuous at this point.
Theorem 3.6.1 Let f : D R and let x D be a limit point of D. Then f is lower semicontinuous
at x if and only if
lim inf f (x) f (x).

xx

Similarly, f is upper semicontinuous at x if and only if

lim sup f (x) f (x).

xx

Proof: Suppose f is lower semicontinuous at x.

Let > 0. Then there exists 0 > 0 such that
f (x)
< f (x) for all x B(x;
0 ) D.
This implies
f (x)
h(0 ),
where
h( ) =

inf

f (x).

xB0 (x;
)D

Thus,
lim inf f (x) = sup h( ) h(0 ) f (x)
.
xx

>0

3.6 LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

80

Since is arbitrary, we obtain lim infxx f (x) f (x).

We now prove the converse. Suppose

lim inf f (x) = sup h( ) f (x)

xx

>0

and let > 0. Since

sup h( ) > f (x)
,
>0

there exists > 0 such that h( ) > f (x)

. This implies
f (x) > f (x)
for all x B0 (x;
) D.
Since this is also true for x = x,
the function f is lower semicontinuous at x.

The proof for the upper semicontinuous case is similar. 

Theorem 3.6.2 Let f : D R and let x D. Then f is l.s.c. at x if and only if for every sequence

{xk } in D that converges to x,

lim inf f (xk ) f (x).

Similarly, f is u.s.c. at x if and only if for every sequence {xk } in D that converges to x,

lim sup f (xk ) f (x).

Then for any > 0, there exists > 0 such that (3.8) holds. Since
Proof: Suppose f is l.s.c. at x.
{xk } converges to x,
we have xk B(x;
) when k is sufficiently large. Thus,
f (x)
< f (xk )
for such k. It follows that f (x)
lim infk f (xk ). Since is arbitrary, it follows that f (x)

lim infk f (xk ).
We now prove the converse. Suppose lim infk f (xk ) f (x)
and assume, by way of contradiction, that f is not l.s.c. at x.
Then there exists > 0 such that for every > 0, there exists
x B(x;
) D with
f (x)
f (x ).
1
Applying this for k = , we obtain a sequence {xk } in D that converges to x with
k
f (x)
f (xk ) for every k.
This implies
f (x)
lim inf f (xk ).
k

This is a contradiction. 

81
Definition 3.6.2 Let f : D R. We say that f is lower semicontinuous on D (or lower semicontinu-

ous if no confusion occurs) if it is lower semicontinuous at every point of D.

Theorem 3.6.3 Suppose D is a compact set of R and f : D R is lower semicontinuous. Then f

has an absolute minimum on D. That means there exists x D such that

f (x) f (x)
for all x D.
Proof: We first prove that f is bounded below. Suppose by contradiction that for every k N, there
exists xk D such that
f (xk ) < k.
Since D is compact, there exists a subsequence {xk` } of {xk } that converges to x0 D. Since f is
l.s.c., by Theorem 3.6.2
lim inf f (xk` ) f (x0 ).
`

This is a contraction because lim inf` f (xk` ) = . This shows f is bounded below. Define
= inf{ f (x) : x D}.
Since the set { f (x) : x D} is nonempty and bounded below, R.
Let {uk } be a sequence in D such that { f (uk )} converges to . By the compactness of D, the
sequence {uk } has a convergent subsequence {uk` } that converges to some x D. Then
= lim f (uk` ) = lim inf f (uk` ) f (x)
.
`

This implies = f (x)

and, hence,
f (x) f (x)
for all x D.
The proof is now complete. 
The following theorem is proved similarly.
Theorem 3.6.4 Suppose D is a compact subset of R and f : D R is upper semicontinuous. Then

f has an absolute maximum on D. That is, there exists x D such that

f (x) f (x)
for all x D.
For every a R, define
La ( f ) = {x D : f (x) a}
and
Ua ( f ) = {x D : f (x) a}.
Theorem 3.6.5 Let f : D R. Then f is lower semicontinuous if and only if La ( f ) is closed in D

for every a R. Similarly, f is upper semicontinuous if and only if Ua ( f ) is closed in D for every
a R.

3.6 LOWER SEMICONTINUITY AND UPPER SEMICONTINUITY

82

Proof: Suppose f is lower semicontinuous. Using Corollary 2.6.8, we will prove that for every
sequence {xk } in La ( f ) that converges to a point x D, we get x La ( f ). For every k, since
xk La ( f ), f (xk ) a.
Since f is lower semicontinuous at x,

f (x)
lim inf f (xk ) a.
k

Thus, x La ( f ). It follows that La ( f ) is closed.

We now prove the converse. Fix any x D and > 0. Then the set
G = {x D : f (x) > f (x)
} = D \ L f (x)(

f)
is open in D and x G. Thus, there exists > 0 such that
B(x;
) D G.
It follows that
f (x)
< f (x) for all x B(x;
) D.
Therefore, f is lower semicontinuous. The proof for the upper semicontinuous case is similar. 
For every a R, we also define
La ( f ) = {x D : f (x) < a}
and
Ua ( f ) = {x D : f (x) > a}.
Corollary 3.6.6 Let f : D R. Then f is lower semicontinuous if and only if Ua ( f ) is open in D

for every a R. Similarly, f is upper semicontinuous if and only if La ( f ) is open in D for every
a R.
Theorem 3.6.7 Let f : D R. Then f is continuous if and only if for every a, b R with a < b,
the set
Oa,b = {x D : a < f (x) < b} = f 1 ((a, b))
is an open set in D.
Proof: Suppose f is continuous. Then f is lower semicontinuous and upper semicontinuos. Fix
a, b R with a < b. Then
Oa,b = Lb Ua .
By Theorem 3.6.6, the set Oa,b is open since it is the intersection of two open sets La and Ub .
Let us prove the converse. We will only show that f is lower semicontinuous since the proof of
upper semicontinuity is similar. For every a R, we have
Ua ( f ) = {x D : f (x) > a} = nN f 1 ((a, a + n))
Thus, Ua ( f ) is open in D as it is a union of open sets in D. Therefore, f is lower semicontinuous by
Corollary 3.6.6. 

83

Exercises
3.6.1 Let f be the function given by

(
x2 , if x 6= 0;
f (x) =
1, if x = 0.
Prove that f is lower semicontinuous.
3.6.2 Let f be the function given by

(
x2 , if x 6= 0;
f (x) =
1, if x = 0.
Prove that f is upper semicontinuous.
3.6.3 Let f , g : D R be lower semicontinuous functions and let k > 0 be a constant. Prove that
f + g and k f are lower semicontinous functions on D.
3.6.4 I Let f : R R be a lower semicontinuous function such that

lim f (x) = lim f (x) = .

Prove that f has an absolute minimum at some x0 R.

DEFINITION AND BASIC PROPERTIES OF THE DERIVATIVE

THE MEAN VALUE THEOREM
SOME APPLICATIONS OF THE MEAN VALUE THEOREM
LHOSPITALS RULE
TAYLORS THEOREM
CONVEX FUNCTIONS AND DERIVATIVES
NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

4. DIFFERENTIATION

In this chapter, we discuss basic properties of the derivative of a function and several major
theorems, including the Mean Value Theorem and lHpitals Rule.

4.1

DEFINITION AND BASIC PROPERTIES OF THE DERIVATIVE

Let G be an open subset of R and consider a function f : G R. For every a G, the function
a (x) =

f (x) f (a)
xa

is defined on G \ {a}. Since G is an open set, a is a limit point of G \ {a}. Thus, it is possible to
discuss the limit
lim a (x) = lim

xa

xa

f (x) f (a)
.
xa

Definition 4.1.1 Let G be an open subset of R and let a G. We say that the function f defined on

G is differentiable at a if the limit

lim

xa

f (x) f (a)
xa

exists (as a real number). In this case, the limit is called the derivative of f at a denoted by f 0 (a),
and f is said to be differentiable at a. Thus, if f is differentiable at a, then
f 0 (a) = lim

xa

f (x) f (a)
.
xa

We say that f is differentiable on G if f is differentiable at every point a G. In this case, the

function f 0 : G R is called the derivative of f on G.

4.1 DEFINITION AND BASIC PROPERTIES OF THE DERIVATIVE

86


(a) Let f : R R be given by f (x) = x2 and let a R. Then

Example 4.1.1

lim

x a

f (x) f (a)
x 2 a2
(x a)(x + a)
= lim
= lim
= lim (x + a) = 2a.
x a x a
x a
x a
xa
xa

Thus, f is differentiable at every a R and f 0 (a) = 2a.

(b) Let f : R R be given by f (x) = |x| and let a = 0. Then
lim

x 0+

and
lim

x 0

Therefore,

f (0)
limx 0 f (x)
x0

|x|
x
f (x) f (0)
= lim+
= lim+ = 1,
x0
x 0 x
x 0 x

f (x) f (0)
|x|
x
= lim
= lim
= 1.
x0
x 0 x
x 0 x
does not exist and, hence, f is not differentiable at 0.

Theorem 4.1.1 Let G be an open subset of R and let f be defined on G. If f is differentiable at

a G, then f is continuous at this point.

Proof: We have the following identity for x G \ {a}:
f (x) = f (x) f (a) + f (a)
f (x) f (a)
=
(x a) + f (a).
xa
Thus,

f (x) f (a)
(x a) + f (a) = f 0 (a) 0 + f (a) = f (a).
lim f (x) = lim
xa
xa
xa


Therefore, f is continuous at a by Theorem 3.3.2. 

Remark 4.1.2 The converse of Theorem 4.1.1 is not true. For instance, the absolute value function

f (x) = |x| is continuous at 0, but it is not differentiable at this point (as shown in the example above).
Theorem 4.1.3 Let G be an open subset of R and let f , g : G R. Suppose both f and g are

differentiable at a G. Then
(1) The function f + g is differentiable at a and
( f + g)0 (a) = f 0 (a) + g0 (a).
(2) For a constant c, the function c f is differentiable at a and
(c f )0 (a) = c f 0 (a).
(3) The function f g is differentiable at a and
( f g)0 (a) = f 0 (a)g(a) + f (a)g0 (a).
(4) Suppose additionally that g(a) 6= 0. Then the function
 0
f
f 0 (a)g(a) f (a)g0 (a)
(a) =
.
g
(g(a))2

f
is differentiable at a and
g

87
Proof: The proofs of (1) and (2) are straightforward and we leave them as exercises. Let us prove
(3). For every x G \ {a}, we can write
f (x)g(x) f (a)g(x) + f (a)g(x) f (a)g(a)
( f g)(x) ( f g)(a)
=
xa
xa
( f (x) f (a))g(x) f (a)(g(x) g(a))
=
+
.
xa
xa
By Theorem 4.1.1, the function g is continuous at a and, hence,
lim g(x) = g(a).

(4.1)

xa

Thus,
lim

xa

( f g)(x) ( f g)(a)
= f 0 (a)g(a) + f (a)g0 (a).
xa

This implies (3).

Next we show (4). Since g(a) 6= 0, by (4.1), there exists an open interval I containing a such
f
that g(x) 6= 0 for all x I. Let h = . Then h is defined on I. Moreover,
g
h(x) h(a)
xa

=
=
=

f (x)
g(x)

f (a)
f (a)
f (a)
g(x)
+ g(x)
g(a)

xa
f (a)
f (a)) + g(x)g(a)
(g(a) g(x))

1
g(x) ( f (x)

xa


1
f (x) f (a)
g(x) g(a)
g(a)
f (a)
.
g(x)g(a)
xa
xa

Taking the limit as x a, we obtain (4). The proof is now complete. 

The following lemma is very convenient for studying the differentiability of the composition of
functions.
Lemma 4.1.4 Let G be an open subset of R and let f : G R. Suppose f is differentiable at a.
Then there exists a function u : G R satisfying

f (x) f (a) = [ f 0 (a) + u(x)](x a) for all x G

and limxa u(x) = 0.
Proof: Define

f (x) f (a) f 0 (a), x G \ {a}

u(x) =
xa
0,
x = a.
Since f is differentiable at a, we have
lim u(x) = lim

xa

xa

f (x) f (a)
f 0 (a) = f 0 (a) f 0 (a) = 0.
xa

Therefore, the function u satisfies the conditions of the lemma. 

4.1 DEFINITION AND BASIC PROPERTIES OF THE DERIVATIVE

88

Theorem 4.1.5 Chain rule. Let f : G1 R and let g : G2 R, where G1 and G2 are two open

subsets of R with f (G1 ) G2 . Suppose f is differentiable at a and g is differentiable at f (a). Then

the function g f is differentiable at a and
(g f )0 (a) = g0 ( f (a)) f 0 (a).
Proof: Since f is differentiable at a, by Lemma 4.1.4, there exists a function u defined on G1 with
f (x) f (a) = [ f 0 (a) + u(x)](x a) for all x G1 ,
and limxa u(x) = 0.
Similarly, since g is differentiable at f (a), there exists a function v defined on G2 with
g(t) g( f (a)) = [g0 ( f (a)) + v(t)][t f (a)] for all t G2 ,

(4.2)

and limt f (a) v(t) = 0.

Applying (4.2) for t = f (x), we have
g( f (x)) g( f (a)) = [g0 ( f (a)) + v( f (x))][ f (x) f (a)].
Thus,
g( f (x)) g( f (a)) = [g0 ( f (a)) + v( f (x))][ f 0 (a) + u(x)](x a) for all x G1 .
This implies
g( f (x)) g( f (a))
= [g0 ( f (a)) + v( f (x))][ f 0 (a) + u(x)] for all x G1 \ {a}.
xa
By the continuity of f at a and the property of v, we have limxa v( f (x)) = 0 and, hence,
g( f (x)) g( f (a))
= g0 ( f (a)) f 0 (a).
xa
xa
lim

The proof is now complete. 

Definition 4.1.2 Let G be an open set and let f : G R be a differentiable function. If the function

f 0 : G R is also differentiable, we say that f is twice differentiable (on G). The second derivative
of f is denoted by f 00 or f (2) . Thus, f 00 = ( f 0 )0 . Similarly, we say that f is three times differentiable
if f (2) is differentiable, and ( f (2) )0 is called the third derivative of f and is denoted by f 000 or f (3) .
We can define in this way n times differentiability and the nth derivative of f for any positive integer
n. As a convention, f (0) = f .
Definition 4.1.3 Let I be a subset of R and let f : I R. The function f is said to be continuously
differentiable if f is differentiable on I and f 0 is continuous on I. We denote by C1 (I) the set of all
continuously differentiable functions on I. If f is n times differentiable on I and the nth derivative is
continuous, then f is called n times continuously differentiable. We denote by Cn (I) the set of all n
times continuously differentiable functions on I.

89

Exercises
4.1.1 Prove parts (1) and (2) of Theorem 4.1.3.
4.1.2 Let

(
x , if x > 0;
f (x) =
0,
if x 0.
(a) Determine the values of for which f is continuous on R.
(b) Determine the values of for which f is differentiable on R. In this case, find f 0 .
4.1.3 Determine the values of x at which each function is differentiable.

x sin 1 , if x 6= 0;
(a) f (x) =
x
0,
if x = 0.

x2 sin 1 , if x =
6 0;
(b) f (x) =
x
0,
if x = 0.
4.1.4 Determine if each of the following functions is differentiable at 0. Justify your answer.

(
x2 ,
(a) f (x) = 3
x ,

if x Q;
if x
/ Q.

(b) f (x) = [x] sin2 (x).

p
(c) f (x) = cos( |x|).
4.1.5 Let f , g be differentiable at a. Find the following limits:

x f (a) a f (x)
.
xa
f (x)g(a) f (a)g(x)
(b) limxa
.
xa
(a) limxa

4.1.6 B Let f be differentiable at a and f (a) > 0. Find the following limit:

lim

4.1.7 I Consider the function

x2 sin 1 + cx, if x 6= 0;
f (x) =
x
0,
if x = 0,
where 0 < c < 1.

f (a + 1n )
f (a)

!1/n
.

4.2 THE MEAN VALUE THEOREM

90

(a) Prove that the function is differentiable on R.

(b) Prove that for every > 0, the function f 0 changes its sign on (, ).
4.1.8 Let f be differentiable at x0 (a, b) and let c be a constant. Prove that



(a) limn n f (x0 + n1 ) f (x0 ) = f 0 (x0 ).
(b) limh0

f (x0 + ch) f (x0 )

= c f 0 (x0 ).
h

4.1.9 Let f be differentiable at x0 (a, b) and let c be a constant. Find the limit

lim

h0

4.2

f (x0 + ch) f (x0 ch)

.
h

THE MEAN VALUE THEOREM

In this section, we focus on the Mean Value Theorem, one of the most important tools of calculus
and one of the most beautiful results of mathematical analysis. The Mean Value Theorem we study
in this section was stated by the French mathematician Augustin Louis Cauchy (17891857), which
follows from a simpler version called Rolles Theorem.
An important application of differentiation is solving optimization problems. A simple method
for identifying local extrema of a function was found by the French mathematician Pierre de Fermat
(16011665). Fermats method can also be used to prove Rolles Theorem.
We start with some basic definitions of minima and maxima. Recall that for a R and > 0,
the sets B(a; ), B+ (a; ), and B (a; ) denote the intervals (a , a + ), (a, a + ) and (a , a),
respectively.
Definition 4.2.1 Let D be a nonempty subset of R and let f : D R. We say that f has a local (or
relative) minimum at a D if there exists > 0 such that
f (x) f (a) for all x B(a; ) D.
Similarly, we say that f has a local (or relative) maximum at a D if there exists > 0 such that
f (x) f (a) for all x B(a; ) D.
In January 1638, Pierre de Fermat described his method for finding maxima and minima in a
letter written to Marin Mersenne (15881648) who was considered as the center of the world of
science and mathematics during the first half of the 1600s. His method presented in the theorem
below is now known as Fermats Rule.
Theorem 4.2.1 Fermats Rule. Let I be an open interval and f : I R. If f has a local minimum
or maximum at a I and f is differentiable at a, then f 0 (a) = 0.
Proof: Suppose f has a local minimum at a. Then there exists > 0 sufficiently small such that
f (x) f (a) for all x B(a; ).
Since B+ (a; ) is a subset of B(a; ), we have
f (x) f (a)
0 for all x B+ (a; ).
xa

91

Figure 4.1: Illustration of Fermats Rule.

Taking into account the differentiability of f at a yields

f 0 (a) = lim

xa

f (x) f (a)
f (x) f (a)
= lim+
0.
xa
xa
xa

Similarly,
f (x) f (a)
0 for all x B (a; ).
xa
It follows that
f 0 (a) = lim

xa

f (x) f (a)
f (x) f (a)
= lim
0.
xa
xa
xa

Therefore, f 0 (a) = 0. The proof is similar for the case where f has a local maximum at a. 
Theorem 4.2.2 Rolles Theorem. Let a, b R with a < b and f : [a, b] R. Suppose f is

continuous on [a, b] and differentiable on (a, b) with f (a) = f (b). Then there exists c (a, b) such
that
f 0 (c) = 0.

(4.3)

Proof: Since f is continuous on the compact set [a, b], there exist x1 [a, b] and x2 [a, b] such that
f (x1 ) = min{ f (x) : x [a, b]} and f (x2 ) = max{ f (x) : x [a, b]}.
Then
f (x1 ) f (x) f (x2 ) for all x [a, b].

(4.4)

If x1 (a, b) or x2 (a, b), then f has a local minimum at x1 or f has a local maximum at x2 . By
Theorem 4.2.1, f 0 (x1 ) = 0 or f 0 (x2 ) = 0, and (4.3) holds with c = x1 or c = x2 .
If both x1 and x2 are the endpoints of [a, b], then f (x1 ) = f (x2 ) because f (a) = f (b). By (4.4),
f is a constant function, so f 0 (c) = 0 for any c (a, b). 
We are now ready to use Rolles Theorem to prove the Mean Value Theorem presented below.
Theorem 4.2.3 Mean Value Theorem. Let a, b R with a < b and f : [a, b] R. Suppose f
is continuous on [a, b] and differentiable on (a, b). Then there exists c (a, b) such that
f 0 (c) =

f (b) f (a)
.
ba

(4.5)

4.2 THE MEAN VALUE THEOREM

92

Figure 4.2: Illustration of Rolles Theorem.

Figure 4.3: Illustration of the Mean Value Theorem.

Proof: The linear function whose graph goes through (a, f (a)) and (b, f (b)) is
g(x) =

f (b) f (a)
(x a) + f (a).
ba

Define



f (b) f (a)
h(x) = f (x) g(x) = f (x)
(x a) + f (a) for x [a, b].
ba
Then h(a) = h(b), and h satisfies the assumptions of Theorem 4.2.2. Thus, there exists c (a, b)
such that h0 (c) = 0. Since
h0 (x) = f 0 (x)

f (b) f (a)
,
ba

it follows that
f 0 (c)

f (b) f (a)
= 0.
ba

93
Thus, (4.5) holds. 
Example 4.2.1 We show that | sin x| |x| for all x R. Let f (x) = sin x for all x R. Then
f 0 (x) = cos x. Now, fix x R, x > 0. By the mean value theorem applied to f on the interval [0, x],
there exists c (0, x) such that

sin x sin 0
= cos c.
x0
| sin x|
= | cos c|. Since | cos c| 1 we conclude | sin x| |x| for all x > 0. Next suppose
|x|
x < 0. Another application of the mean value theorem shows there exists c (x, 0) such that

Therefore,

sin 0 sin x
= cos c.
0x
| sin x|
= | cos c| 1. It follows that | sin x| |x| for x < 0. Since equality holds for
|x|
x = 0, we conclude that | sin x| |x| for all x R.

Then, again,

A more general result which follows directly from the Mean Value Theorem is known as Cauchys
Theorem.
Theorem 4.2.4 Cauchys Theorem. Let a, b R with a < b. Suppose f and g are continuous
on [a, b] and differentiable on (a, b). Then there exists c (a, b) such that
[ f (b) f (a)]g0 (c) = [g(b) g(a)] f 0 (c).

(4.6)

Proof: Define
h(x) = [ f (b) f (a)]g(x) [g(b) g(a)] f (x) for x [a, b].
Then h(a) = f (b)g(a) f (a)g(b) = h(b), and h satisfies the assumptions of Theorem 4.2.2. Thus,
there exists c (a, b) such that h0 (c) = 0. Since
h0 (x) = [ f (b) f (a)]g0 (x) [g(b) g(a)] f 0 (x),
this implies (4.6). 
The following theorem shows that the derivative of a differentiable function on [a, b] satisfies the
intermediate value property although the derivative function is not assumed to be continuous. To
give the theorem in its greatest generality, we introduce a couple of definitions.
Definition 4.2.2 Let a, b R, a < b, and f : [a, b] R. If the limit
lim+

xa

f (x) f (a)
xa

exists, we say that f has a right derivative at a and write

f+0 (a) = lim+
xa

f (x) f (a)
.
xa

If the limit
lim

xb

f (x) f (b)
xb

4.2 THE MEAN VALUE THEOREM

94
exists, we say that f has a left derivative at b and write
f0 (b) = lim
xb

f (x) f (b)
.
xb

We will say that f is differentiable on [a, b] if f 0 (x) exists for each x (a, b) and, in addition, both
f+0 (a) and f0 (b) exist.

Figure 4.4: Right derivative.

Theorem 4.2.5 Intermediate Value Theorem for Derivatives. Let a, b R with a < b. Sup-

pose f is differentiable on [a, b] and

f+0 (a) < < f0 (b).
Then there exists c (a, b) such that
f 0 (c) = .
Proof: Define the function g : [a, b] R by
g(x) = f (x) x.
Then g is differentiable on [a, b] and
g0+ (a) < 0 < g0 (b).

95
Thus,
lim+

xa

g(x) g(a)
< 0.
xa

It follows that there exists 1 > 0 such that

g(x) < g(a) for all x (a, a + 1 ) [a, b].
Similarly, there exists 2 > 0 such that
g(x) < g(b) for all x (b 2 , b) [a, b].
Since g is continuous on [a, b], it attains its minimum at a point c [a, b]. From the observations
above, it follows that c (a, b). This implies g0 (c) = 0 or, equivalently, that f 0 (c) = . 
Remark 4.2.6 The same conclusion follows if f+0 (a) > > f0 (b).

Exercises
4.2.1 B Let f and g be differentiable at x0 . Suppose f (x0 ) = g(x0 ) and

f (x) g(x) for all x R .

Prove that f 0 (x0 ) = g0 (x0 ).
4.2.2 I Prove the following:

(a) | sin(x) sin(y)| |x y| for all x, y R.

(b) If f is a differentiable function on R with | f 0 (x)| ` for all x R, then f is a Lipschitz function.
4.2.3 B Let n be a positive integer and let ak , bk R for k = 1, . . . , n. Prove that the equation
n

x + (ak sin kx + bk cos kx) = 0

k=1

has a solution on (, ).
4.2.4 B Let f and g be differentiable functions on [a, b]. Suppose g(x) 6= 0 and g0 (x) 6= 0 for all

x [a, b]. Prove that there exists c (a, b) such that




 f (a) f (b) 

1

 = 1  f (c) g(c)


g(b) g(a) g(a) g(b)
g0 (c)  f 0 (c) g0 (c)

where the bars denote determinants of the two-by-two matrices.

4.2.5 BLet n be a fixed positive integer.
(a) Suppose a1 , a2 , . . . , an satisfy

a1 +

a2
an
+ + = 0.
2
n

4.3 SOME APPLICATIONS OF THE MEAN VALUE THEOREM

96
Prove that the equation

a1 + a2 x + a3 x2 + + an xn1 = 0
has a solution in (0, 1).
(b) Suppose a0 , a1 , . . . , an satisfy
n

ak

2k + 1 = 0.

k=0

Prove that the equation

n

ak cos(2k + 1)x = 0
k=0

has a solution on (0, 2 ).

4.2.6 B Let f : [0, ) R be a differentiable function.

f (x)
= a.
x
f (x)
= .
(b) Show that if limx f 0 (x) = , then limx
x
(c) Are the converses in part (a) and part (b) true?
(a) Show that if limx f 0 (x) = a, then limx

4.3

SOME APPLICATIONS OF THE MEAN VALUE THEOREM

In this section, we assume that a, b R and a < b. In the proposition below, we show that it is
possible to use the derivative to determine whether a function is constant. The proof is based on the
Mean Value Theorem.
Proposition 4.3.1 Let f be continuous on [a, b] and differentiable on (a, b). If f 0 (x) = 0 for all

x (a, b), then f is constant on [a, b].

Proof: Suppose by contradiction that f is not constant on [a, b]. Then there exist a1 and b1 such
that a a1 < b1 b and f (a1 ) 6= f (b1 ). By Theorem 4.2.3, there exists c (a1 , b1 ) such that
f 0 (c) =

f (b1 ) f (a1 )
6= 0,
b1 a1

which is a contradiction. 
The next application of the Mean Value Theorem concerns developing simple criteria for
monotonicity of real-valued functions based on the derivative.
Proposition 4.3.2 Let f be differentiable on (a, b).

(i) If f 0 (x) > 0 for all x (a, b), then f is strictly increasing on (a, b).
(ii) If f 0 (x) < 0 for all x (a, b), then f is strictly decreasing on (a, b).

97

Figure 4.5: Strictly Increasing Function.

Proof: Let us prove (i). Fix any x1 , x2 (a, b) with x1 < x2 . By Theorem 4.2.3, there exists
c (x1 , x2 ) such that
f (x2 ) f (x1 )
= f 0 (c) > 0.
x2 x1
This implies f (x1 ) < f (x2 ). Therefore, f is strictly increasing on (a, b). The proof of (ii) is similar. 


Example 4.3.1 Let n N and f : [0, ) R be given by f (x) = xn . Then f 0 (x) = nxn1 . There-

fore, f 0 (x) > 0 for all x > 0 and, so, f is strictly increasing. In particular, this shows that every
positive real number has exactly one n-th root (refer to Example 3.4.1).
Theorem 4.3.3 Inverse Function Theorem. Suppose f is differentiable on I = (a, b) and f 0 (x) 6=

0 for all x (a, b). Then f is one-to-one and the inverse function f 1 : f (I) I is differentiable.
Moreover,
( f 1 )0 (y) =

1
f 0 (x)

(4.7)

where f (x) = y.
Proof: It follows from Theorem 4.2.5 that
f 0 (x) > 0 for all x (a, b), or f 0 (x) < 0 for all x (a, b).
Suppose f 0 (x) > 0 for all x (a, b). Then f is strictly increasing on this interval and, hence, it is
one-to-one. It follows from Theorem 3.4.11 that f 1 is continuous on f (I).
It remains to prove the differentiability of the inverse function f 1 and the representation of its
derivative (4.7). Fix any y f (I) with y = f (x).
Let g = f 1 . We will show that
g(y) g(y)

1
= 0 .
yy
y y
f (x)

lim

Fix any sequence {yk } in f (I) that converges to y and yk 6= y for every k. For each yk , there exists
xk I such that f (xk ) = yk . That is, g(yk ) = xk for all k. It follows from the continuity of g that {xk }

4.4 LHOSPITALS RULE

98
converges to x.
Then
lim

xk x
g(yk ) g(y)

= lim
k f (xk ) f (x)
yk y

1
1
= lim
= 0 .

k f (xk ) f (x)
f (x)

xk x

The proof is now complete. 

Exercises
4.3.1 I Let f : R R. Suppose there exist ` 0 and > 0 such that

| f (u) f (v)| `|u v| for all u, v R .

(4.8)

(a) Prove that f is uniformly continuous on R.

(b) Prove that if > 1, then f is a constant function.
(c) Find a nondifferentiable function that satisfies the condition above for = 1.
4.3.2 B Let f and g be differentiable functions on R such that f (x0 ) = g(x0 ) and

f 0 (x) g0 (x) for all x x0 .

Prove that
f (x) g(x) for all x x0 .
4.3.3 B Let f be twice differentiable on an open interval I. Suppose that there exist a, b, c I
with a < b < c such that f (a) < f (b) and f (b) > f (c). Prove that there exists d (a, b) such that
f 00 (d) < 0.
4.3.4 B Prove that f defined in Exercise 4.1.7 is not monotone on any open interval containing 0.

4.4

LHOSPITALS RULE
We now prove a result that allows us to compute various limits by calculating a related limit
involving derivatives. All three theorems in this section are known as lHospitals Rule.
For this section, we assume a, b R with a < b.
Theorem 4.4.1 Suppose f and g are continuous on [a, b] and differentiable on (a, b). Suppose
f (x)
= g(x)
= 0, where x [a, b]. Suppose further that there exists > 0 such that g0 (x) 6= 0 for all
x B(x;
) [a, b], x 6= x.

If
f 0 (x)
= `,
xx g0 (x)
lim

99
then
lim

xx

f (x)
= `.
g(x)

(4.9)

Proof: Let {xk } be a sequence in [a, b] that converges to x and such that xk 6= x for every k. By
Theorem 4.2.4, for each k, there exists a sequence {ck }, ck is in between xk and x such that
[ f (xk ) f (x)]g
0 (ck ) = [g(xk ) g(x)]
f 0 (ck ).
Since f (x)
= g(x)
= 0, and g0 (ck ) 6= 0 for sufficiently large k, we have
f (xk )
f 0 (ck )
= 0
.
g(xk )
g (ck )
Under the assumptions that g0 (x) 6= 0 for x near x and g(x)
= 0, we also have g(xk ) 6= 0 for sufficiently
large k. By the squeeze theorem (Theorem 2.1.5), {ck } converges to x.
Thus,
f (xk )
f 0 (ck )
= lim 0
= `.
k g(xk )
k g (ck )
lim

Therefore, (4.9) follows from Theorem 3.1.1. 

Theorem 4.4.2 Suppose f and g are continuous on [a, b] and differentiable on (a, b). Suppose

limxx f (x) = limxx g(x) = , where x [a, b]. Suppose further that there exists > 0 such that
g0 (x) 6= 0 for all x B(x;
) [a, b], x 6= x.

If ` R and
lim

f 0 (x)
= `,
xx g0 (x)

(4.10)

f (x)
= `.
xx g(x)

(4.11)

then
lim

Proof: Since limxx f (x) = limxx g(x) = , without loss of generality, we can assume that f (x) 6= 0
and g(x) 6= 0 for all x B(x;
) [a, b]. Fix any > 0. From (4.10), one can choose K > 0 and
1 < such that
 0 
 0

 f (x) 
 f (x)





 g0 (x)  K and  g0 (x) ` < 2
whenever x B(x;
1 ) [a, b], x 6= x.

Fix B(x;
1 ) [a, b], > x.
Then
g()
f (x)
f (x) f ()
g(x)
=
, for all x B(x;
1 ) [a, b], x > x,
x 6= .
f ()
g(x)
g(x) g()
1
f (x)
1

Rolles theorem (Theorem 4.2.2) and the assumptions made guarantee that the fractions are well
defined.

4.4 LHOSPITALS RULE

100
Define
g()
g(x)
for x B(x;
1 ) [a, b], x > x,
x 6= .
H (x) =
f ()
1
f (x)
1

Now, limxx+ H (x) = 1. Thus, there exists < 1 such that

|H (x) 1| <

whenever x B(x;
) [a, b], x > x.

2K

For any x B(x;

) [a, b], x > x,
applying Theorem 4.2.4, we can write
f (x)
f 0 (c)
= 0 H (x),
g(x)
g (c)
where c is in between and x (so c is in B(x;
1 ) and c 6= x).
Therefore,

  0

 f (x)
  f (c)





 g(x) ` =  g0 (c) H (x) `

 0
0 (c)

 f (c)
f
`
=  0 (H (x) 1) + 0
g (c)
g (c)
 0

 0 
 f (c)

 f (c) 



`
 0  |H (x) 1| +  0
g (c)
g (c)

+ = .
<K
2K 2
It follows that limxx+

f (x)
= `.
g(x)

Similarly, we also get limxx

f (x)
= `. The proof is complete. 
g(x)

Remark 4.4.3 The proof of Theorem 4.4.1 and Theorem 4.4.2 show that the results in these theorems
can be applied for left-hand and right-hand limits.

The following theorem can be proved following the method in the proof of Theorem 4.4.2.
Theorem 4.4.4 Let f and g be differentiable on (a, ). Suppose g0 (x) 6= 0 for all x (a, ) and

lim f (x) = lim g(x) = .

If ` R and
f 0 (x)
= `,
x g0 (x)
lim

then
f (x)
= `.
x g(x)
lim

101

Exercises
4.4.1 Use LHospitals Rule to find the following limits:

x arctan x
.
x3
x sin x
limx0
.
x tan x
ex ex
limx0
.
sin x cos x
ex ex
limx0
.
ln(1 + x)
1 cos 2x
limx0
.
x sin x

(a) limx0
(b)
(c)
(d)
(e)

4.4.2 Prove that the following functions are differentiable at 1 and -1.

x2 ex2 , if |x| 1;
(a) f (x) = 1
,
if |x| > 1.
e

arctan x,
if |x| 1;
(b) f (x) =
x1
sign x +
, if |x| > 1.
4
2
4.4.3 B Let P(x) be a polynomial. Prove that

lim P(x)ex = 0.

4.4.4 I Consider the function

( 1
e x2 , if x 6= 0;
f (x) =
0,
if x = 0.
Prove that f Cn (R) for every n N.

4.5

TAYLORS THEOREM
In this section, we prove a result that lets us approximate differentiable functions by polynomials.
Theorem 4.5.1 Taylors Theorem. Let n be a positive integer. Suppose f : [a, b] R is a

function such that f (n) is continuous on [a, b], and f (n+1) (x) exists for all x (a, b). Let [a, b].
Then for any [a, b] with 6= , there exists a number c in between and such that
f ( ) = Pn ( ) +

f (n+1) (c)
( )n+1 ,
(n + 1)!

where
n

Pn (x) =

k=0

f (k) ()
(x )k , x [a, b].
k!

4.5 TAYLORS THEOREM

102

Proof: Let be as in the statement and let 6= . Since 6= 0, there exists a number R
such that
f ( ) = Pn ( ) +

( )n+1 .
(n + 1)!

We will now show that

= f (n+1) (c),
for some c in between and .
Consider the function
n

g(x) = f ( )
k=0

f (k) (x)

( x)k
( x)n+1 .
k!
(n + 1)!

Then
n

g() = f ( )
k=0

f (k) ()

( )k
( )n+1 = f ( )Pn ( )
( )n+1 = 0.
k!
(n + 1)!
(n + 1)!

and
n

g( ) = f ( )
k=0

f (k) ( )

( )k
( )n+1 = f ( ) f ( ) = 0.
k!
(n + 1)!

By Rolles theorem, there exists c in between and such that g0 (c) = 0. Using the product rule
for derivatives, we have
!
n
(k+1) (c)
(k) (c)
f
f

g0 (c) = f 0 (c) +
( c)k +
( c)k1 + ( c)n
k!
(k 1)!
n!
k=1
1

( c)n f (n+1) (c)( c)n

n!
n!
= 0.
=

This implies = f (n+1) (c). The proof is now complete. 

The polynomial Pn (x) given in the theorem is called the n-th Taylor polynomial of f at .
Remark 4.5.2 The conclusion of Taylors theorem still holds true if = . In this case, c = = .
Theorem 4.5.3 Let n be an even positive integer. Suppose f (n) exists and continuous on (a, b). Let

x (a, b) satisfy
f 0 (x)
= . . . = f (n1) (x)
= 0 and f (n) (x)
6= 0.
The following hold:
(1) f (n) (x)
> 0 if and only if f has a local minimum at x.

(n)
(2) f (x)
< 0 if and only if f has a local maximum at x.

103
Proof: We will prove (1). Suppose f (n) (x)
> 0. Since f (n) (x)
> 0 and f (n) is continuous at x,
there
exists > 0 such that
f (n) (t) > 0 for all t B(x;
) (a, b).
Fix any x B(x;
). By Taylors theorem and the given assumption, there exists c in between x and x
such that
f (x) = f (x)
+

f (n) (c)
(x x)
n.
n!

Since n is even and c B(x;

), we have f (x) f (x).
Thus, f has a local minimum at x.
Now
suppose that f has a local minimum at x.
Then there exists > 0 such that
f (x) f (x)
for all x B(x;
) (a, b).
Fix a sequence {xk } that converges to x with xk 6= x for every k. By Taylors theorem, there exists a
sequence {ck } that also converges to x with
f (xk ) = f (x)
+

f (n) (ck )
(xk x)
n.
n!

Since xk B(x;
) for sufficiently large k, we have
f (xk ) f (x)

for such k. It follows that

f (xk ) f (x)
=

f (n) (ck )
(xk x)
n 0.
n!

This implies f (n) (ck ) 0 and, hence, f (n) (x)

= limk f (n) (ck ) 0.
The proof of (2) is similar. 

Exercises
4.5.1 B Use Taylors theorem to prove that
m

ex >

xk

k!

k=0

for all x > 0 and m N.

4.5.2 Suppose f is twice differentiable on (a, b). Show that for every x (a, b),
lim

h0

f (x + h) + f (x h) 2 f (x)
= f 00 (x).
h2

4.5.3 I (a) Suppose f is three times differentiable on (a, b) and x (a, b). Prove that
2

f (x + h) f (x)
f 0 (x)
1!h f 00 (x)
h2!
f 000 (x)

lim
=
.
3
h0
h
3!
(b) State and prove a more general result for the case where f is n times differentiable on (a, b).

4.6 CONVEX FUNCTIONS AND DERIVATIVES

104

4.5.4 Suppose f is n times differentiable on (a, b) and x (a, b). Define

n

Pn (h) =

f (n) (x)

k=0

hn
for h R
n!

Prove that
lim

h0

f (x + h) Pn (h)
= 0.
hn

(Thus, we have
f (x + h) = Pn (h) + g(h),
wher g is a function that satisfies limh0 g(h)
hn = 0. This is called the Taylor expansion with Peanos
remainder.)

4.6

CONVEX FUNCTIONS AND DERIVATIVES

We discuss in this section a class of functions that plays an important role in optimization
problems.
Definition 4.6.1 Let I be an interval of R and let f : I R. We say that f is convex on I if
f ( u + (1 )v) f (u) + (1 ) f (v)
for all u, v I and for all (0, 1).

Figure 4.6: A Convex Function.

Example 4.6.1 The following functions are convex as a direct calculation shows

(a) f : R R, f (x) = x.
(b) f : R R, f (x) = x2 .
(c) f : R R, f (x) = |x|.
Theorem 4.6.1 Let I be an interval of R. A function f : I R is convex if and only if for every

i 0, i = 1, . . . , n, with ni=1 i = 1 (n 2) and for every xi I, i = 1, . . . , n,

!
n

i xi

i=1

i f (xi ).
i=1

(4.12)

105
Proof: Since the converse holds trivially, we only need to prove the implication by induction. The
conclusion holds for n = 2 by the definition of convexity. Suppose the conclusion holds for any
2 n k. We will show that it also holds for n = k + 1. Fix i 0, i = 1, . . . , k + 1, with k+1
i=1 i = 1
and fix every xi I, i = 1, . . . , k + 1. Then
k

i = 1 k+1 .

i=1

If k+1 = 1, then i = 0 for all i = 1, . . . , k, and (4.12) holds. Suppose 0 k+1 < 1. Then
k

1 k+1 = 1.

i=1

It follows that
k+1

i xi

i=1



ki=1 i xi
= f (1 k+1 )
+ k+1 xk+1
1 k+1
 k

i=1 i xi
(1 k+1 ) f
+ k+1 f (xk+1 )
1 k+1
!
k
i
= (1 k+1 ) f
xi + k+1 f (xk+1 )
i=1 1 k+1
k

i
f (xi ) + k+1 f (xk+1 )
1

k+1
i=1

(1 k+1 )
k+1

i f (xi ),

i=1

where the first inequality follows from the definition of convexity (or is trivial if k+1 = 0) and the
last inequality follows from the inductive assumption. The proof is now complete. 
Theorem 4.6.2 Let I be an interval and let f : I R be a convex function. Then f has a local

minimum at x if and only if f has an absolute minimum at x.

Proof: We only need to prove the implication since the converse is trivial. Suppose f has a local
minimum at x.
Then there exists > 0 such that
f (u) f (x)
for all u B(x;
) I.
For any x I, we have xn = (1 n1 )x + 1n x x.
Thus, xn B(x;
) I when n is sufficiently large.
Thus, for such n,
1
1
f (x)
f (xn ) (1 ) f (x)
+ f (x).
n
n
This implies that for a sufficient large n, we have
1
1
f (x)
f (x)
n
n
and, hence, f (x)
f (x). Since x was arbitrary, this shows f has an absolute minimum at x.


4.6 CONVEX FUNCTIONS AND DERIVATIVES

106

Theorem 4.6.3 Let I be an open interval and let f : I R be a convex function. Suppose f is

differentiable at x.
Then
f 0 (x)(x
x)
f (x) f (x)
for all x I.

(4.13)

Proof: For any x R and t (0, 1), we have

f (tx + (1 t)x)
f (x)

f (x + t(x x))
f (x)

=
t
t
t f (x) + (1 t) f (x)
f (x)

t
= f (x) f (x).

Since f is differentiable at x,

f 0 (x)(x
x)
= lim+
t0

f (x + t(x x))
f (x)

f (x) f (x),

which completes the proof. 

Corollary 4.6.4 Let I be an open interval and let f : I R be a convex function. Suppose f is
differentiable at x.
Then f has an absolute minimum at x if and only if f 0 (x)
= 0.
0
Proof: Suppose f has an absolute minimum at x.
By Theorem 4.2.1, f (x)
= 0. Let us prove the
converse. Suppose f 0 (x)
= 0. It follows from Theorem 4.6.3 that
0 = f 0 (x)(x
x)
f (x) f (x)
for all x I.
This implies
f (x) f (x)
for all x I.
Thus, f has an absolute minimum at x.

Lemma 4.6.5 Let I be an open interval and suppose f : I R is a convex function. Fix a, b, x I
with a < x < b. Then

f (b) f (a)
f (b) f (x)
f (x) f (a)

.
xa
ba
bx
Proof: Let
t=

xa
.
ba

Then t (0, 1) and



xa
f (x) = f (a + (x a)) = f a +
(b a) = f (a + t(b a)) = f (tb + (1 t)a).
ba
By convexity of f , we obtain
f (x) t f (b) + (1 t) f (a).

107
Thus,
f (x) f (a) t f (b) + (1 t) f (a) f (a) = t[ f (b) f (a)] =

xa
( f (b) f (a)).
ba

Equivalently,
f (x) f (a)
f (b) f (a)

.
xa
ba
Similarly,
f (x) f (b) t f (b) + (1 t) f (a) f (b) = (1 t)[ f (a) f (b)] =

xb
[ f (b) f (a)].
ba

It follows that
f (b) f (a)
f (b) f (x)

.
ba
bx
The proof is now complete. 
Theorem 4.6.6 Let I be an open interval and let f : I R be a differentiable function. Then f is
convex if and only if f 0 is increasing on I.
Proof: Suppose f is convex. Fix a < b with a, b I. By Lemma 4.6.5, for any x (a, b), we have
f (x) f (a)
f (b) f (a)

.
xa
ba
This implies, taking limits, that
f 0 (a)

f (b) f (a)
.
ba

Similarly,
f (b) f (a)
f 0 (b).
ba
Therefore, f 0 (a) f 0 (b), and f 0 is an increasing function.
Let us prove the converse. Suppose f 0 is increasing. Fix x1 < x2 and t (0, 1). Then
x1 < xt < x2 ,
where xt = tx1 + (1 t)x2 . By the Mean Value Theorem (Theorem 4.2.3), there exist c1 and c2 such
that
x1 < c1 < xt < c2 < x2
with
f (xt ) f (x1 ) = f 0 (c1 )(xt x1 ) = f 0 (c1 )(1 t)(x2 x1 );
f (xt ) f (x2 ) = f 0 (c2 )(xt x2 ) = f 0 (c2 )t(x1 x2 ).

4.6 CONVEX FUNCTIONS AND DERIVATIVES

108
This implies
t f (xt ) t f (x1 ) = f 0 (c1 )t(1 t)(x2 x1 );

(1 t) f (xt ) (1 t) f (x2 ) = f 0 (c2 )t(1 t)(x1 x2 ).

Since f 0 (c1 ) f 0 (c2 ), we have
t f (xt ) t f (x1 ) = f 0 (c1 )t(1 t)(x2 x1 ) f 0 (c2 )t(1 t)(x2 x1 ) = (1 t) f (x2 ) (1 t) f (xt ).
Rearranging terms, we get
f (xt ) t f (x1 ) + (1 t) f (x2 ).
Therefore, f is convex. The proof is now complete. 
Corollary 4.6.7 Let I be an open interval and let f : I R be a function. Suppose f is twice

differentiable on R. Then f is convex if and only if f 00 (x) 0 for all x I.

Proof: It follows from Proposition 4.3.2 that f 00 (x) 0 for all x I if and only if the derivative
function f 0 is increasing on I. The conclusion then follows directly from Theorem 4.6.6. 
Theorem 4.6.8 Let I be an open interval and let f : I R be a convex function. Then it is locally
Lipschitz continuous in the sense that for any x I, there exist ` 0 and > 0 such that
| f (u) f (v)| `|u v| for all u, v B(x;
).

(4.14)

In particular, f is continuous.
Proof: Fix any x I. Choose four numbers a, b, c, d satisfying
a < b < x < c < d with a, d I.
Choose > 0 such that B(x;
) (b, c). Let u, v B(x;
) with v < u. Then by Lemma 4.6.5, we
see that
f (b) f (a)
f (u) f (a)
f (u) f (v)
f (d) f (v)
f (d) f (c)

.
ba
ua
uv
d v
d c
Using a similar approach for the case u < v, we get
f (b) f (a)
f (u) f (v)
f (d) f (c)

for all u, v B(x;

).
ba
uv
d c
Choose ` 0 sufficiently large so that
`

f (b) f (a)
f (u) f (v)
f (d) f (c)

` for all u, v B(x;

).
ba
uv
d c

Then (4.14) holds. The proof is now complete. 

109

Exercises
4.6.1 (a) Let I be an interval and let f , g : I R be convex functions. Prove that c f , f + g, and
max{ f , g} are convex functions on I, where c 0 is a constant.
(b) Find two convex functions f and g on an interval I such that f g is not convex.
4.6.2 Let f : R R be a convex function. Given a, b R, prove that the function defined by

g(x) = f (ax + b), for x R

is also a convex function on R.
4.6.3 I Let I be an interval and let f : I R be a convex function. Suppose that is a convex,
increasing function on an interval J that contains f (I). Prove that f is convex on I.
4.6.4 B Prove that each of the following functions is convex on the given domain:

(a) f (x) = ebx , x R, where b is a constant.

(b) f (x) = xk , x [0, ) and k 1 is a constant.
(c) f (x) = ln(1
x (, 1).
 x),
ex
, x R.
(d) f (x) = ln
1 + ex
4.6.5 B Prove the following:

(a) If a, b are nonnegative real numbers, then

a+b
ab.
2
(b) If a1 , a2 , . . . , an , where n 2, are nonnegative real numbers, then
a1 + a2 + + an
(a1 a2 an )1/n .
n

4.7

NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

In this section, we introduce a new concept that is helpful in the study of optimization problems
in which the objective function may fail to be differentiable.
Definition 4.7.1 Let f : R R be a convex function. A number u R is called a subderivative of
the function f at x if
u (x x)
f (x) f (x)
for all x R .

(4.15)

The set of all subderivatives of f at x is called the subdifferential of f at x and is denoted by f (x).

Example 4.7.1 Let f (x) = |x|. Then

f (0) = [1, 1].

Indeed, for any u f (0), we have
u x = u(x 0) f (x) f (0) = |x| for all x R .

4.7 NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

110

Figure 4.7: A nondifferential convex function.

In particular, u 1 |1| = 1 and u (1) = u | 1| = 1. Thus, u [1, 1]. It follows that

f (0) [1, 1].
For any u [1, 1], we have |u| 1. Then
u x |u x| = |u||x| |x| for all x R .
This implies u f (0). Therefore, f (0) = [1, 1].

Figure 4.8: Definition of subderivative.

Lemma 4.7.1 Let f : R R be a convex function. Fix a R. Define the slope function a by

a (x) =

f (x) f (a)
xa

for x (, a) (a, ). Then, for x1 , x2 (, a) (a, ) with x1 < x2 , we have

a (x1 ) a (x2 ).

(4.16)

111
Proof: This lemma follows directly from Lemma 4.6.5. 
Theorem 4.7.2 Let f : R R be a convex function and let x R. Then f has left derivative and
right derivative at x.
Moreover,
sup x (x) = f0 (x)
f+0 (x)
= inf x (x),
x>x

x<x

where x is defined in (4.16).

Proof: By Lemma 4.7.1, the slope function x defined by (4.16) is increasing on the interval (x,
)
and bounded below by x (x 1). By Theorem 3.2.4, the limit
lim+ x (x) = lim+

xx

xx

f (x) f (x)

x x

exists and is finite. Moreover,

lim x (x) = inf x (x).

xx+

x>x

Thus, f+0 (x)

exists and
= inf x (x).
f+0 (x)
x>x

Similarly, f0 (x)
exists and
f0 (x)
= sup x (x).
x<x

Applying Lemma 4.7.1 again, we see that

x (x) x (y) whenever x < x < y.
This implies f0 (x)
f+0 (x).
The proof is complete. 
Theorem 4.7.3 Let f : R R be a convex function and let x R. Then
f (x)
= [ f0 (x),
f+0 (x)].

Proof: Suppose u f (x).

By the definition (4.15), we have
u (x x)
f (x) f (x)
for all x > x.

This implies
u

f (x) f (x)

for all x > x.

x x

Thus,
u lim+
xx

f (x) f (x)

= f+0 (x).

x x

Similarly, we have
u (x x)
f (x) f (x)
for all x < x.

(4.17)

4.7 NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

112
Thus,
u

f (x) f (x)

for all x < x.

x x

This implies u f0 (x).

So
f (x)
[ f0 (x),
f+0 (x)].

To prove the opposite inclusion, take u [ f0 (x),

f+0 (x)].
By Theorem 4.7.2
sup x (x) = f0 (x)
u f+0 (x)
= inf x (x).
x>x

x<x

Using the upper estimate by f+0 (x)

for u, one has
u x (x) =

f (x) f (x)

for all x > x.

x x

It follows that
u (x x)
f (x) f (x)
for all x x.

Similarly, one also has

u (x x)
f (x) f (x)
for all x < x.

Thus, (4.15) holds and, hence, u f (x).

Therefore, (4.17) holds. 
Corollary 4.7.4 Let f : R R be a convex function and x R. Then f is differentiable at x if and
only if f (x)
is a singleton. In this case,
f (x)
= { f 0 (x)}.

Proof: Suppose f is differentiable at x.

Then
f0 (x)
= f+0 (x)
= f 0 (x).

By Theorem 4.7.3,
f (x)
= [ f0 (x),
f+0 (x)]
= { f 0 (x)}.

Thus, f (x)
is a singleton.
Conversely, if f (x)
is a singleton, we must have f0 (x)
= f+0 (x).
Thus, f is differentiable at x.




Example 4.7.2 Let f (x) = a|x b| + c, where a > 0. Then f is a convex function and

f0 (b) = a, f+0 (b) = a.

Thus,
f (b) = [a, a].
Since f is differentiable on (, b) and (b, ), we have

{a}, if x < b;
f (x) = [a, a], if x = b;

{a},
if x > b.

113

Figure 4.9: Set addition.

Definition 4.7.2 Let A and B be two nonempty subsets of R and let R. Define

A + B = {a + b : a A, b B} and A = {a : a A}.
Theorem 4.7.5 Let f , g : R R be convex functions and let > 0. Then f + g and f are convex

functions and
( f + g)(x)
= f (x)
+ g(x)

( f )(x)
= f (x).

Proof: It is not hard to see that f + g is a convex function and

( f + g)0+ (x)
= f+0 (x)
+ g0+ (x)

( f + g)0 (x)
= f0 (x)
+ g0 (x).

By Theorem 4.7.3,
( f + g)(x)
= [( f + g)0 (x),
( f + g)0+ (x)]

= [ f0 (x)
+ g0 (x),
f+0 (x)
+ g0+ (x)]

= [ f0 (x),
f+0 (x)]
+ [g0 (x),
g0+ (x)]

= f (x)
+ g(x).

The proof for the second formula is similar. 



Example 4.7.3 Let a1 < a2 < < an and let i > 0 for i = 1, . . . , n. Define
n

f (x) = i |x ai |.
i=1

Then f is a convex function. By Theorem 4.7.5, we get

(
if x
/ {a1 , . . . , an }
ai <x i ai >x i ,
f (x)
=
ai <x i ai >x i + [i0 , i0 ], if x = ai0 .
Theorem 4.7.6 Let fi : R R, i = 1, . . . , n, be convex functions. Define

f (x) = max{ fi (x) : i = 1, . . . , n} and I(u) = {i = 1, . . . , n : fi (u) = f (u)}.

Then f is a convex function. Moreover,
f (x)
= [m, M],
where
0
0
m = min fi
(x)
and M = max fi+
(x).

iI(x)

iI(x)

4.7 NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

114

Proof: Fix u, v R and (0, 1). For any i = 1, . . . , n, we have

fi ( u + (1 )v) fi (u) + (1 ) fi (v) f (u) + (1 ) f (v).
This implies
f ( u + (1 )v) = max fi ( u + (1 )v) f (u) + (1 ) f (v).
1in

Thus, f is a convex function. Similarly we verify that f+0 (x)

= M and f0 (x)
= m. By Theorem 4.7.3,
f (x)
= [m, M].
The proof is now complete. 
Remark 4.7.7 The product of two convex functions is not a convex function in general. For instance,

f (x) = x and g(x) = x2 are convex functions, but h(x) = x3 is not a convex function.
The following result may be considered as a version of the first derivative test for extrema in the
case of non differentiable functions.
Theorem 4.7.8 Let f : R R be a convex function. Then f has an absolute minimum at x if and
only if
0 f (x)
= [ f0 (x),
f+0 (x)].

Proof: Suppose f has an absolute minimum at x.

Then
f (x)
f (x) for all x R .
This implies
0 (x x)
= 0 f (x) f (x)
for all x R .
It follows from (4.15) that 0 f (x).

Conversely, if 0 f (x),
again, by (4.15),
0 (x x)
= 0 f (x) f (x)
for all x R .
Thus, f has an absolute minimum at x.



Example 4.7.4 Let k be a positive integer and a1 < a2 < < a2k1 . Define
2k1

f (x) =

|x ai |,

i=1

for x R. It follows from the subdifferential formula in Example 4.7.3 that 0 f (x)
if and only if
x = ak . Thus, f has a unique absolute minimum at ak .
Similarly, if a1 < a2 < < a2k and
2k

g(x) = |x ai |.
i=1

Then 0 g(x)
if and only if x [ak , ak+1 ]. Thus, g has an absolute minimum at any point of
[ak , ak+1 ].

115

Figure 4.10: Subdifferential of f (x) = 2k1

i=1 |x ai |.

Figure 4.11: Subdifferential of g(x) = 2k

i=1 |x ai |.

The following theorem is a version of the Mean Value Theorem (Theorem 4.2.3) for nondifferentiable functions.
Theorem 4.7.9 Let f : R R be a convex function and let a < b. Then there exists c (a, b) such
that
f (b) f (a)
f (c).
ba

(4.18)

Proof: Define



f (b) f (a)
g(x) = f (x)
(x a) + f (a) .
ba
Then g is a convex function and g(a) = g(b). Thus, g has a local minimum at some c (a, b) and,
hence, g also has an absolute minimum at c. Observe that the function


f (b) f (a)
h(x) =
(x a) + f (a)
ba
is differentiable at c and, hence,


f (b) f (a)
0
h(c) = {h (c)} =
.
ba

116

4.7 NONDIFFERENTIABLE CONVEX FUNCTIONS AND SUBDIFFERENTIALS

Figure 4.12: Subdifferential mean value theorem.

By Theorem 4.7.8 and the subdifferential sum rule,



f (b) f (a)
0 g(c) = f (c)
.
ba
This implies (4.18). The proof is now complete. 
Corollary 4.7.10 Let f : R R be a convex function. Then f is Lipschitz continuous if and only
if there exists ` 0 such that
f (x) [`, `] for all x R .
Proof: Suppose f is Lipschitz continuous on R. Then there exists ` 0 such that
| f (u) f (v)| `|u v| for all u, v R .
Then for any x R,
f+0 (x) = lim+
h0

`|h|
f (x + h) f (x)
lim+
= `.
h
h
h0

Similarly, f0 (x) `. Thus,

f (x) = [ f0 (x), f+0 (x)] [`, `].

117
Conversely, fix any u, v R with u 6= v. Applying Theorem 4.7.9, we get
f (v) f (u)
f (c) [`, `],
vu
for some c in between u and v. This implies
| f (u) f (v)| `|u v|.
This inequality obviously holds for u = v. Therefore, f is Lipschitz continuous. 

Exercises
4.7.1 B Find subdifferentials of the following functions:

(a) f (x) = a|x|, a > 0.

(b) f (x) = |x 1| + |x + 1|.
4.7.2 Find the subdifferential of the function

f (x) = max{2x + 1, x, 2x 1}.

4.7.3 I Let f (x) = nk=1 |x k|. Find all absolute minimizers of the function.
4.7.4 Let f : R R be a convex function. Fix a, b R and define the function g by

g(x) = f (ax + b), for x R

Prove that g(x)
= a f (ax + b).
4.7.5 B Let f : R R be a convex function. Suppose that f (x) [0, ) for all x R. Prove that
f is monotone increasing on R.

5. Solutions and Hints for Selected Exercises

SECTION 1.1
Exercise 1.1.2. Applying basic rules of operations on sets yields
(X \Y ) Z = Y c Z = Z \Y.
and
Z \ (Y Z) = (Z \Y ) (Z \ Z) = (Z \Y ) 0/ = Z \Y.
Therefore, (X \Y ) Z = Z \ (Y Z).

SECTION 1.2
Exercise 1.2.1. (a) For any a A, we have f (a) f (A) and, so, a f 1 ( f (A)). This implies A
f 1 ( f (A)). Note that this inclusion does not require the injectivity of f . Now fix any a f 1 ( f (A)).
Then f (a) f (A), so there exists a0 A such that f (a) = f (a0 ). Since f is one-to-one, a = a0 A.
Therefore, f 1 ( f (A)) A and the equality holds.
(b) Fix any b f ( f 1 (B)). Then b = f (x) for some x f 1 (B). Thus, b = f (x) B and, hence,
f ( f 1 (B)) B. This inclusion does not require the surjectivity of f . Now fix b B. Since f is onto,
there exists x X such that f (x) = b B. Thus, x f 1 (B) and, hence, b f ( f 1 (B)). We have
shown that B f ( f 1 (B)) and the equality holds.
Without the injectivity of f , the equality in part (a) is no longer valid. Consider f (x) = x2 , x R,
and A = [1, 2]. Then f (A) = [0, 4] and, hence, f 1 ( f (A)) = [2, 2], which strictly contains A. It is
also not hard to find an example of a function f and a set B for which the equality in part (b) does
not hold true.

120

Solutions and Hints for Selected Exercises

SECTION 1.3
Exercise 1.3.3. For n = 1,

1 h1+ 5 1 5i
1 2 5

=
= 1.
2
2
5
5 2
Thus, the conclusion holds for n = 1. It is also easy to verify that the conclusion holds for n = 2.
Suppose that

1 h 1 + 5 k  1 5 k i
ak =

2
2
5
for all k n, where n 2. Let us show that

1 h 1 + 5 n+1  1 5 n+1 i
an+1 =

.
2
2
5

(5.1)

By the definition of the sequence and the induction hypothesis,

an+1 = an + an1

1 h 1 + 5 n1  1 5 n1 i
1 h 1 + 5 n  1 5 n i
+
=

2
2
2
2
5
5





h


i
1
1 + 5 n1 1 + 5
1 5 n1 1 5
=
+1
1 .
2
2
2
2
5
Observe that

3 + 5  1 + 5 2
1 5
3 5  1 5 2
1+ 5
+1 =
=
and
+1 =
=
.
2
2
2
2
2
2
Therefore, (5.1) follows easily.

1+ 5
1 5
In this exercise, observe that the two numbers
and
are the roots of the quadratic
2
2
equation
x2 = x + 1.

A more general result can be formulated as follows. Consider the sequence {an } defined by
a1 = a;
a2 = b;
an+2 = an+1 + an for n N .
Suppose that the equation x2 = x + has two solutions x1 and x2 . Let c1 and c2 be two constants
such that
c1 x1 + c2 x2 = a;
c1 (x1 )2 + c2 (x2 )2 = b.
Then we can prove by induction that
xn = c1 (x1 )n + c2 (x2 )n for all n N .

121
This is a very useful method to find a general formula for a sequence defined recursively as above.
For example, consider the sequence
a1 = 1;
a2 = 1;
an+2 = an+1 + 2an for n N .
Solving the equation x2 = x + 2 yields two solutions x1 = 2 and x2 = (1). Thus,
xn = c1 2n + c2 (1)n ,
where c1 and c2 are constants such as
c1 (2) + c2 (1) = 1;
c1 (2)2 + c2 (1)2 = 1.
It is not hard to see that c1 = 1/3 and c2 = 1/3. Therefore,
1
1
an = 2n (1)n for all n N .
3
3
Exercise 1.3.5.. Hint: Prove first that, for k = 1, 2, . . . , n, we have
  
 

n
n
n+1
.
+
=
k
k1
k

SECTION 1.4
Exercise 1.4.3. In general, to prove that |a| m, where m 0, we only need to show that a m and
a m.
For any x, y R,
|x| = |x y + y| |x y| + |y|,
This implies
|x| |y| |x y|.
Similarly,
|y| = |y x + x| |x y| + |x|,
This implies
(|x| |y|) |x y|.
Therefore,
||x| |y|| |x y|.

122

Solutions and Hints for Selected Exercises

SECTION 1.5
Exercise 1.5.4. Let us first show that A + B is bounded above given that A and B are bounded above.
Since A and B are bounded above, there exist R and R such that
a for all a A, b for all b B.
For any x A + B, there exist a A and b B such that x = a + b. Thus, x = a + b + , which
shows that A + B is bounded above. Moreover, since sup A is an upper bound of A and sup B is an
upper bound of B, we have
x = a + b sup A + sup B.
This implies that sup A + sup B is an upper bound of A + B and, hence,
sup(A + B) sup A + sup B.
Fix any > 0. By Proposition 1.5.1, there exits a A and b B such that
sup A < a and sup B < b.
It follows that
sup A + sup B 2 < a + b sup(A + B).
Since > 0 is arbitrary,
sup A + sup B sup(A + B).
Therefore, the given equality has been justified.

SECTION 1.6
1
Exercise 1.6.1. Let x = . By Theorem 1.6.2 (4), there exists n N such that
r
n

1
< n + 1,
r

This implies
1
1
<r .
n+1
n

SECTION 2.1
Exercise 2.1.10. (a) Suppose that limn an = `. Then by Theorem 2.1.8,
lim a2n = ` and lim a2n+1 = `.

Now suppose that (5.2) is satisfied. Fix any > 0. Choose N1 N such that
|a2n `| < whenever n N1 ,

(5.2)

123
and choose N2 N such that
|a2n+1 `| < whenever n N2 .
Let N = max{2N1 , 2N2 + 1}. Then
|an `| < whenever n N.
Therefore, limn an = `.
This problem is sometimes very helpful to show that a limit exists. For example, consider the
sequence defined by
x1 = 1/2,
1
xn+1 =
for n N .
2 + xn
We will see later
that {x2n+1 } and {x2n } both converge to
converges to 2 1.

2 1, so we can conclude that {xn }

(b) Use a similar method to the solution of part (a).

Exercise 2.1.7. Consider the case where ` > 0. By the definition of limit, we can find n1 N such
that
|an | > `/2 for all n n1 .
Given any > 0, we can find n2 N such that
|an `| <

`
for all n n2 .
4

Choose n0 = max{n1 , n2 }. For any n n0 , one has



 an+1
 |an an+1 | |an `| + |an+1 `|

=

<
 an

|an |
|an |

`
4

+ `4
`
2

= .

Therefore, limn aan+1

= 1. If ` < 0, consider the sequence {an }.
n
The conclusion is no longer true if ` = 0. A counterexample is an = n where (0, 1).

SECTION 2.2
Exercise 2.2.1. (a) The limit is calculated as follows:

 

p

n2 + n n
n2 + n + n

lim
n2 + n n = lim
n
n
n2 + n + n
n
= lim
n n2 + n + n
n
= lim p
n
n2 (1 + 1/n) + n
1
= lim p
= 1/2.
n
1 + 1/n + 1

124

Solutions and Hints for Selected Exercises

(b) The limit is calculated as follows:


 p


3 3
3
2 n
3 + 3n2 )2 + n 3 n3 + 3n2 + n2

p
n
+
3n
(n
3
p

lim
n3 + 3n2 n = lim
3
n
n
(n3 + 3n2 )2 + n 3 n3 + 3n2 + n2 )
3n2

= lim p
n 3 (n3 + 3n2 )2 + n 3 n3 + 3n2 + n2
3n2
p
= lim p
n 3 n6 (1 + 3/n)2 + n 3 n3 (1 + 3/n) + n2
3n2
p

p
n 2 3
n
(1 + 3/n)2 + 3 (1 + 3/n) + 1

= lim

3
 = 1.
= lim p
p
n 3
3
2
(1 + 3/n) + (1 + 3/n) + 1
(c) We use the result in par (a) and part (b) to obtain
p

p
p
p
3
3
n3 + 3n2 n + n n2 + 1
lim ( n3 + 3n2 n2 + 1) = lim
n
n
p



p
3
= lim
n3 + 3n2 n + lim n n2 + 1 = 1 1/2 = 1/2.
n

Using a similar technique, we can find the following limit:

lim

p

p
3
an3 + bn2 + cn + d n2 + n + ,

where a > 0 and > 0.

SECTION 2.3
Exercise 2.3.1. (a) Clearly, a1 < 2. Suppose that ak < 2 for k N. Then
p

ak+1 = 2 + ak < 2 + 2 = 2.
By induction, an < 2 for all n N.
p

(b) Clearly, a1 = 2 < 2 + 2 = a2 . Suppose that ak < ak+1 for k N. Then

ak + 2 < ak+1 + 2,
which implies
p
p
ak + 2 < ak+1 + 2.
Thus, ak+1 < ak+2 . By induction, an < an+1 for all n N. Therefore, {an } is an increasing sequence.
(c)
By the monotone convergence theorem, limn an exists. Let ` = limn an . Since an+1 =
2 + an and limn an+1 = `, we have

` = 2 + ` or `2 = 2 + `.

125
Solving this quadratic equation yields ` = 1 or ` = 2. Therefore, limn an = 2.
Define a more general sequence as follows:
a1 = c > 0,

an+1 = c + an

for n N .

1 + 1 + 4c
We can prove that {an } is monotone increasing and bounded above by
. In fact, {an }
2

1 + 1 + 4c
converges to this limit. The number
is obtained by solving the equation ` = c + `,
2
where ` > 0.
Exercise 2.3.2. (a) The limit is 3.
(b) The limit is 3.
(c) We use the well-known inequality
a+b+c
3
abc for a, b, c 0.
3
By induction, we see that an > 0 for all n N. Moreover,
1
1
1
1
1
an+1 = (2an + 2 ) = (an + an + 2 )
3
an
3
an
3

s
3

an an

1
= 1.
a2n

We also have


1
2a3 + a2 3 (an 1)(2a2n + 2an + 3)
1
an+1 an =
=
< 0.
2an + 2 an = n 2n
3
an
3an
3a2n
Thus, {an } is monotone deceasing
and bounded below. We can showthat limn an = 1.
(d) Use the inequality a+b

ab for a, b 0 to show that an+1 b for all n N. Then follow

2

part 3 to show that {an } is monotone decreasing. The limit is b.

Exercise 2.3.3. (a) Let {an } be the given sequence. Observe that an+1 = 2an . Then show that
{an } is monotone increasing and bounded above. The limit is 2.
(b) Let {an } be the given sequence. Then
an+1 =

1
.
2 + an

Show that {a2n+1 } is monotone decreasing and bounded below; {a2n }

is monotone increasing and
bounded above. Thus, {an } converges by Exercise 2.1.10. The limit is 2 1.
Exercise 2.3.5. Observe that
bn+1 =

an + bn p
an bn = an+1 for all n N .
2

Thus,
an+1 =
bn+1 =

an bn an an = an for all n N,

an + bn bn + bn

= bn for all n N .
2
2

126

Solutions and Hints for Selected Exercises

It follows that {an } is monotone increasing and bounded above by b1 , and {bn } is decreasing and
bounded below by a1 . Let x = limn an and y = limn bn . Then
x=

x+y

.
xy and y =
2

Therefore, x = y.

SECTION 2.4
Exercise 2.4.1. Here we use the fact that in R a sequence is a Cauchy sequence if and only if it is
convergent.
(a) Not a Cauchy sequence. See Example 2.1.4.
(b) A Cauchy sequence. This sequence converges to 0.
(c) A Cauchy sequence. This sequence converges to 1.
(d) A Cauchy sequence. This sequence converges to 0 (see Exercise 2.1.5).

SECTION 2.5
Exercise 2.5.4. (a) Define
n = sup(an + bn ), n = sup ak , n = sup bk .
kn

kn

kn

By the definition,
lim sup(an + bn ) = lim n , lim sup an = lim n , lim sup bn = lim n .
n

By Exercise 2.5.3,
n n + n for all n N .
This implies
lim n lim n + lim n for all n N .

Therefore,
lim sup(an + bn ) lim sup an + lim sup bn .
n

This conclusion remains valid for unbounded sequences provided that the right-hand side is welldefined. Note that the right-hand side is not well-defined, for example, when lim supn an = and
lim supn bn = .
(b) Define
n = inf (an + bn ), n = inf ak , n = inf bk .
kn

kn

Proceed as in part (a), but use part (b) of Exercise 2.5.3.

(c) Consider an = (1)n and bn = (1)n+1 .

kn

127

SECTION 2.6
Exercise 2.6.4. Suppose A is a compact subset of R. Let us first show that A is bounded. Suppose
by contradiction it is not bounded. Then for every n N, there exists an A such that
|an | n.
Since A is compact, there exists a subsequence {ank } that converges to some a A. Then
|ank | nk k

for all k.

Therefore, limk |ank | = . This is a contradiction because {|ank |} converges to |a|. Thus A is
bounded.
Let us now show that A is closed. Let {an } be a sequence in A that converges to a point a R.
By the definition of compactness, {an } has a subsequence {ank } that converges to b A. Then
a = b A and, hence, A is closed by Theorem 2.6.4. The converse follows directly from the
Bolzano-Weierstrass theorem and Theorem 2.6.4.

SECTION 3.1
Exercise 3.1.6. (a) Observe that when x is near 1/2, f (x) is near 1/2 no matter whether x is rational
or irrational. We have
(
|x 1/2|,
if x Q;
| f (x) 1/2| =
|1 x 1/2|, if x 6 Q.
Thus, | f (x) 1/2| = |x 1/2| for all x R.
Given any > 0, choose = . Then
| f (x) 1/2| < whenever |x 1/2| < .
Therefore, limx1/2 f (x) = 1/2.
(b) Observe that when x is near 0 and x is rational, f (x) is near 0. However, when f is near 0 and x
is irrational, f (x) is near 1. Thus, the given limit does not exists. We justify this using the sequential
criterion for limits (Theorem 3.1.1). By contradiction, assume that
lim f (x) = `,

x0

where ` is a real number. Choose a sequence {rn } of rational numbers that converges to 0, and choose
also a sequence {sn } of irrational numbers that converges to 0. Then f (rn ) = rn and f (sn ) = 1 sn
and, hence,
` = lim f (rn ) = 0
n

and
` = lim f (sn ) = lim (1 sn ) = 1.
n

This is a contradiction.
(c) By a similar method to part (b), we can show that limx1 f (x) does not exists.

128

Solutions and Hints for Selected Exercises

Solving this problem suggests a more general problem as follows. Given two polynomials P and
Q, define the function
(
P(x), if x Q;
f (x) =
Q(x), if x 6 Q.
If a is a solution of the equation P(x) = Q(x), i.e., P(a) = Q(a), then the limit limxa f (x) exists
and the limit is this common value. For all other points the limit does not exist.
Similar problems:
1. Determine all a R at which limxa f (x) exists, where
(
x2 ,
if x Q;
f (x) =
x + 2, if x x 6 Q.
2. Consider the function
(
x2 + 1, if x Q;
f (x) =
x,
if x
6 Q.
Prove that f does not have a limit at any a R.

SECTION 3.2
Exercise 3.2.4. The given condition implies that if both x1 and x2 are close to x,
then they are close
to each other and, hence, f (x1 ) and f (x2 ) are close to each other. This suggests the use of the Cauchy

criterion for limit to solve the problem. Given any > 0, choose =
. If x1 , x2 D \ {x}

2(k + 1)
with |x1 x|
< and |x2 x|
< , then
| f (x1 ) f (x2 )| k|x1 x2 | k(|x1 x|
+ |x2 x|)
< k( + ) = 2k

< .
2(k + 1)

Therefore, limxx f (x) exists.

SECTION 3.3
Exercise 3.3.5. (a) Observe that f (a) = g(a) = h(a) and, hence,
(
|g(x) g(a)|, if x Q [0, 1];
| f (x) f (a)| =
|h(x) h(a)|, if x Qc [0, 1].
It follows that
| f (x) f (a)| |g(x) g(a)| + |h(x) h(a)| for all x [0, 1].
Therefore, limxa f (x) = f (a) and, so, f is continuous at a.
(b) Apply part (a).
Exercise 3.3.6. At any irrational number a (0, 1], we have f (a) = 0. If x is near a and x is irrational,

129
it is obvious that f (x) = 0 is near f (a). In the case when x is near a and x is rational, f (x) = 1/q
where p, q N. We will see in part (a) that for any > 0, there is only a finite number of x (0, 1]
such that f (x) . So f (x) is close to f (a) for all x (0, 1] except for a finite number of x Q.
Since a is irrational, we can choose a sufficiently small neighborhood of a to void such x.
(a) For any > 0,
 


p
1
p
1
A = {x (0, 1] : f (x) } = x = Q : f (x) = = x = Q : q
.
q
q
q

Clearly, the number of q N such that q

A is finite.

is finite. Since 0 <

p
q

1, we have p q. Therefore,

(b) Fix any irrational number a (0, 1]. Then f (a) = 0. Given any > 0, by part (a), the set A is
finite, so we can write
A = {x (0, 1] : f (x) } = {x1 , x2 , . . . , xn },
for some n N, where xi Q for all i = 1, . . . , n. Since a is irrational, we can choose > 0 such that
xi
/ (a , a + ) for all i = 1, . . . , n (more precisely, we can choose = min{|a xi | : i = 1, . . . , n}).
Then
| f (x) f (a)| = f (x) < whenever |x a| < .
Therefore, f is continuous at a.
Now fix any rational number b = qp (0, 1]. Then f (b) = 1q . Choose a sequence of irrational
numbers {sn } that converges to b. Since f (sn ) = 0 for all n N, the sequence { f (sn )} does not
converge to f (b). Therefore, f is not continuous at b.
In this problem, we consider the domain of f to be the interval (0, 1], but the conclusion remain valid
for other intervals. In particular, we can show that the function defined on R by
1
p

q , if x = q , p, q N, where p and q have no common factors;

f (x) = 1, if x = 0;

0, if x is irrational,
is continuous at every irrational point, and discontinuous at every rational point.
Exercise 3.3.7. Consider
(
(x a1 )(x a2 ) (x ak ), if x Q;
f (x) =
0,
if x Qc .

SECTION 3.4
Exercise 3.4.4. Let = min { f (x) : x [a, b]} and = max { f (x) : x [a, b]}. Then
f (x1 ) + f (x2 ) + + f (xn ) n

= .
n
n
Similarly,

f (x1 ) + f (x2 ) + + f (xn )

.
n

130

Solutions and Hints for Selected Exercises

Then the conclusion follows from the Intermediate Value Theorem.

Exercise 3.4.5. (a) Observe that
| f (1/n)| 1/n for all n N .




(b) Apply the Extreme Value Theorem for the function g(x) =  f (x)
x  on the interval [a, b].
Exercise 3.4.6. First consider the case where f is monotone decreasing on [0, 1]. By Exercise 3.4.3,
f has a fixed point in [0, 1], which means that there exists x0 [0, 1] such that
f (x0 ) = x0 .
Since f is monotone decreasing, f has a unique fixed point. Indeed, suppose that there exists
x1 [0, 1] such that f (x1 ) = x1 . If x1 < x0 , then x1 = f (x1 ) f (x0 ) = x0 , which yields a contradiction.
It is similar for the case where x1 > x0 . Therefore, x0 is the unique point in [0, 1] such that f (x0 ) = x0 .
Since f (g(x)) = g( f (x)) for all x [0, 1], we have
f (g(x0 )) = g( f (x0 )) = g(x0 ).
Thus, g(x0 ) is also a fixed point of f and, hence, g(x0 ) = x0 = f (x0 ). The proof is complete in this
case.
Consider the case where f is monotone increasing. In this case, f could have several fixed points on
[0, 1], so the previous argument does not work. However, by Exercise 3.4.3, there exists c [0, 1]
such that g(c) = c. Define the sequence {xn } as follows:
x1 = c,
xn+1 = f (xn ) for all n 1.
Since f is monotone increasing, {xn } is a monotone sequence. In fact, if x1 x2 , then {xn } is
monotone increasing; if x1 x2 , then {xn } is monotone decreasing. Since f is bounded, by the
monotone convergence theorem (Theorem 2.3.1), there exists x0 [0, 1] such that
lim xn = x0 .

Since f is continuous and xn+1 = f (xn ) for all n N, taking limits we have f (x0 ) = x0 .
We can prove by induction that g(xn ) = xn for all n N. Then
g(x0 ) = lim g(xn ) = lim xn = x0 .
n

Therefore, f (x0 ) = g(x0 ) = x0 .

SECTION 3.5
Exercise 3.5.2. (a) Let f : D R. From Theorem 3.5.1 we see that if there exist two sequences {xn }
and {yn } in D such that |xn yn | 0 as n , but {| f (xn ) f (yn )|} does not converge to 0, then
f is not uniformly continuous on D. Roughly speaking, in order for f to be uniformly continuous on
D, if x and y are close to each other, then f (x) and f (y) must be close to each other. The behavior

131
of the graph of the squaring function suggests the argument below to show that f (x) = x2 is not
uniformly continuous on R.
1
Define two sequences {xn } and {yn } as follows: xn = n and yn = n + for n N. Then
n
1
|xn yn | = 0 as n . However,
n


1 2
1
n2 = 2 + 2 2 for all n N .
| f (xn ) f (yn )| = n +
n
n
Therefore, {| f (xn ) f (yn )|} doesr
not converge to 0 and, hence, f is not uniformly continuous on R.

1
In this solution, we can use xn = n + and yn = n for n N instead.
n
1
1
(b) Use xn =
and yn =
, n N.
/2 + 2n
2n
(c) Use xn = 1/n and yn = 1/(2n).
3
It is natural to ask whether the function
r f (x) = x is uniformly continuous on R. Following the

1
solution for part (a), we can use xn = 3 n + and yn = 3 n for n N to prove that f is not uniformly
n
continuous on R. By a similar method, we can show that the function f (x) = xn , n N, n 2, is not
uniformly continuous on R. A more challenging question is to determine whether a polynomial of
degree greater than or equal to two is uniformly continuous on R.

Exercise 3.5.3. Hint: Use Theorem 3.2.2. Note that the one-sided limits in this exercise coincide
with the corresponding limits in the setting of Theorem 3.2.2.
Exercise 3.5.6. (a) Applying the definition of limit, we find b > a such that
c a < f (x) < c + 1 whenever x > b.
Since f is continuous on [a, b], it is bounded on this interval. Therefore, f is bounded on [a, ).
(b) Fix any > 0, by the definition of limit, we find b > a such that
| f (x) c| <

whenever x > c.
2

Since f is continuous on [a, b + 1], it is uniformly continuous on this interval. Thus, there exists
0 < < 1 such that
| f (u) f (v)| <

whenever |u v| < , u, v [a, c + 1].

2

Then we can show that | f (u) f (v)| < whenever |u v| < , u, v [a, ).
(c) Since limx f (x) = c > f (a), there exists b > a such that
f (x) > f (a) whenever x > b.
Thus,
inf{ f (x) : x [a, ) = inf{ f (x) : x [a, b]}.
The conclusion follows from the Extreme Value Theorem for the function f on [a, b].

132

Solutions and Hints for Selected Exercises

SECTION 3.6
Exercise 3.6.4. Since limx f (x) = limx f (x) = , there exists a > 0 such that
f (x) f (0) whenever |x| > a.
Since f is lower semicontinuous, by Theorem 3.6.3, it has an absolute minimum on [a, a] at some
point x [a, a]. Obviously,
f (x) f (x)
for all x [a, a].
In particular, f (0) f (x).
If |x| > a, then
f (x) f (0) f (x).

Therefore, f has an absolute minimum at x.

Observe that in this solution, we can use any number in the range of f instead of f (0). Since
any continuous function is also lower semicontinuous, the result from this problem is applicable for
continuous functions. For example, we can use this theorem to prove that any polynomial with even
degree has an absolute minimum on R. Since R is a not a compact set, we cannot use the extreme
value theorem directly.

SECTION 4.1
Exercise 4.1.6. Use the identity
limn

f (a + 1n )
f (a)

!1/n
= limn exp( n1 [ln( f (a + 1n )) ln( f (a)]).

Exercise 4.1.7. (a) Using the differentiability of sin x and Theorem 4.1.3, we conclude the function
is differentiable at any a 6= 0. So, we only need to show the differentiability of the function at a = 0.
By the definition of the derivative, consider the limit
lim

xa

f (x) f (a)
x2 sin(1/x) + cx
= lim
= lim [x sin(1/x) + c].
x0
x0
xa
x

For any x 6= 0, we have

|x sin(1/x)| = |x| | sin(1/x)| |x|,
which implies
|x| x sin(1/x) |x|.
Since limx0 (|x|) = limx0 |x| = 0, applying the squeeze theorem yields
lim x sin(1/x) = 0.

x0

It now follows that

f 0 (0) = lim

xa

f (x) f (a)
= lim [x sin(1/x) + c] = c.
x0
xa

133
Using Theorem 4.1.3 and the fact that cos x is the derivative of sin x, the derivative of f can be written
explicitly as

2x sin 1 cos(1/x) + c, if x 6= 0;
0
f (x) =
x
c,
if x = 0.
From the solution, it is important to see that the conclusion remains valid if we replace the
function f by

xn sin 1 , if x 6= 0;
g(x) =
x
0,
if x = 0,
where n 2, n N. Note that the function h(x) = cx does not play any role in the differentiability
of f .
We can generalize this problem as follows. Let be a bounded function on R, i.e., there is
M > 0 such that
|(x)| M for all x R .
Define the function
(
xn (1/x), if x 6= 0;
f (x) =
0,
if x = 0,
where n 2, n N. Then f is differentiable at a = 0.
Similar problems:
1. Show that the functions below are differentiable on R:
(
x3/2 cos(1/x), if x 0;
f (x) =
0,
if x < 0
and
(
2
x2 e1/x , if x 6= 0;
f (x) =
0,
if x = 0.
2. Suppose that is bounded and differentiable on R. Define the function
(
xn (1/x), if x 6= 0;
f (x) =
0,
if x = 0.
Show that if n 2, the function is differentiable on R and find its derivative. Show that if n = 1 and
limx (x) does not exists, then f is not differentiable at 0.
(b) Hint: Observe that
f

1
2n


= 1 + c < 0 and f

1
(2n + 1)


= 1 + c > 0.

134

Solutions and Hints for Selected Exercises

SECTION 4.2
Exercise 4.2.1. Define the function
h(x) = f (x) g(x).
Then h has an absolute maximum at x0 . Thus,
h0 (x0 ) = f 0 (x0 ) g0 (x0 ) = 0,
which implies f 0 (x0 ) = g0 (x0 ).
Exercise 4.2.2. (a) The inequality holds obviously if a = b. In the case where a 6= b, the equality
can be rewritten as


 sin(b) sin(a) 
 1.



ba


 sin(b) sin(a) 

 is the slope of the line connecting (a, f (a)) and (b, f (b)). We need
The quotient 

ba
to show that the absolute value of the slope is always bounded by 1, which can also be seen from
the figure. The quotient also reminds us of applying the Mean Value Theorem for the function
f (x) = sin(x).

Figure 5.1: The function f (x) = sin(x).

Consider the case where a < b and define the function f : [a, b] R by f (x) = sin(x). Clearly,
the function satisfies all assumptions of the Mean Value Theorem on this interval with f 0 (x) = cos(x)
for all x (a, b).
By the Mean Value Theorem, there exists c (a, b) such that
f (b) f (a)
= f 0 (c) = cos(c),
ba
which implies


 f (b) f (a) 


 b a  = | cos(c)| 1.

135
It follows that | f (a) f (b)| |a b|. The solution is similar for the case where a > b.
It is essential to realize that the most important property required in solving this problem is the
boundedness of the derivative of the function. Thus, it is possible to solve the following problems
with a similar strategy.
1. Prove that | cos(a) cos(b)| |a b| for all a, b R.
2. Prove that | ln(1 + e2a ) ln(1 + e2b )| 2|a b| for all a, b R.
Part (b) of the problem is a more general form of part (a) which can be solved similarly.
Exercise 4.2.3. Let us define f : [, ] R by
n

f (x) = x + (ak sin kx + bk cos kx).

k=1

We want to find c (, ) such that f (c) = 0.

Now, consider the function

n 
x2
cos(kx)
sin(kx)
g(x) = + ak
+ bk
.
2 k=1
k
k
Observe that g() = g() and g0 = f . The conclusion follows from Rolles Theorem.
Exercise 4.2.4. Use the identity

 f (a) f (b)
1

g(b) g(a)  g(a) g(b)


 f (a)g(b) f (b)g(a)
=
=

g(b) g(a)

f (a)
g(a)
1
g(a)

f (b)
g(b)
1
g(b)

Then apply the Cauchy mean value theorem for two functions (x) =
interval [a, b].

f (x)
g(x)

and (x) =

1
g(x)

on the

Exercise 4.2.5. 1. Apply Rolles theorem to the function

f (x) = a1 x + a2

x2
xn
+ + an
2
n

on the interval [0, 1].

2. Apply Rolles theorem to the function
n

f (x) =

sin(2k + 1)x
2k + 1
k=0

on the interval [0, /2].

Exercise 4.2.6. (a) Given > 0, first find x0 large enough so that a /2 < f 0 (x) < a + /2 for
x > x0 . Then use the identity
f (x)
f (x) f (x0 ) + f (x0 )
=
=
x
x x0 + x0

f (x) f (x0 )
xx0

f (x0 )
+ xx
0

x0
1 + xx
0

136

Solutions and Hints for Selected Exercises

and the mean value theorem to show that, for x large,

a <

f (x)
< a + .
x

(b) Use the method in part (a).

(c) Consider f (x) = sin(x).

SECTION 4.3
Exercise 4.3.1. (a) We can prove that f is uniformly continuous on R by definition. Given any > 0,

 1

and get
choose =
`+1
| f (u) f (v)| `|u v| < ` = `

<
`+1

whenever |u v| < . Note that we use ` + 1 here instead of ` to avoid the case where ` = 0.
(b) We will prove that f is a constant function by showing that it is differentiable on R and f 0 (a) = 0
for all a R. Fix any a R. Then, for x 6= a,


 f (x) f (a)  `|x a|
1


 x a  |x a| = `|x a| .
Since > 1, by the squeeze theorem,
lim

xa

f (x) f (a)
= 0.
xa

This implies that f is differentiable at a and f 0 (a) = 0.

(c) We can verify that the function f (x) = |x| satisfies the requirement.
From this problem, we see that it is only interesting to consider the class of functions that satisfy
(4.8) when 1. It is an exercise to show that the function f (x) = |x|1/2 satisfies this condition
with ` = 1 and = 1/2.
Exercise 4.3.2. Define the function
h(x) = g(x) f (x).
Then h0 (x) = g0 (x) f 0 (x) 0 for all x [x0 , ). Thus, h is monotone increasing on this interval. It
follows that
h(x) h(x0 ) = g(x0 ) f (x0 ) = 0 for all x x0 .
Therefore, g(x) f (x) for all x x0 .
Exercise 4.3.3. Apply the mean value theorem twice.
Exercise 4.3.4. Use proof by contradiction.

137

SECTION 4.4
Exercise 4.4.3 Suppose that
P(x) = a0 + a1 x + + an xn .
Then apply LHospitals rule repeatedly.
Exercise 4.4.4. We first consider the case where n = 1 to get ideas for solving this problem in the
general case. From the standard derivative theorems we get that the function is differentiable at any
x 6= 0 with
1
1
2
f 0 (x) = 2x3 e x2 = 3 e x2 .
x
Consider the limit
1

f (x) f (0)
e x 2
= lim
.
lim
x0 x
x0
x0
Letting t = 1/x and applying LHospital rule yields
1

t
e x 2
1
lim+
= lim t 2 = lim
= 0.
t e
t 2tet 2
x
x0
Similarly,
1

e x2
= 0.
lim
x
x0
It follows that f is differentiable on R with

2 e x12 , if x 6= 0;
0
f (x) = x3
0,
if x = 0.
In a similar way, we can show that f is twice differentiable on R with


6 + 2 e x12 , if x 6= 0;
x4 x6
f 00 (x) =

0,
if x = 0.
Based on these calculations, we predict that f is n times differentiable for every n N with
 
P 1 e x12 , if x 6= 0;
(n)
x
f (x) =

0,
if x = 0,
where P is a polynomial. Now we proceed to prove this conclusion by induction. The conclusion is
true for n = 1 as shown above. Given that the conclusion is true for some n N, for x 6= 0 we have
 
 
 
2
1 12
1 12
(n+1)
2 0 1
x
f
(x) = x P
+ 3P
e
=Q
e x ,
x
x
x
x

138

Solutions and Hints for Selected Exercises

where Q is also a polynomial. It is an easy exercise to write the explicit formula of Q based on P.
Moreover, successive applications of lHpitals rule give
 
1 12
1
tP(t)
f (n) (x) f (n) (0)
= lim+ P
lim+
e x = lim t 2 = 0.
t e
x0
x
x0 x
x0
In a similar way, we can show that
lim

x0

f (n) (x) f (n) (0)

= 0.
x0

Therefore, f (n+1) (0) = 0. We have proved that for every n N, f is n times differentiable and, so,
f Cn (R). Here we do not need to prove the continuity of f (n) because the differentiability of f (n)
implies its continuity.
In a similar way, we can also show that the function
( 1
e x , if x > 0;
f (x) =
0,
if x 0
is n times differentiable for every n N.

SECTION 4.5
Exercise 4.5.1. Let f (x) = ex . By Taylors theorem, for any x > 0, there exists c (0, x) such that
m

f (x) = ex =
=

k=0
m k
x

f (k) (0) k f (m+1) (c) m+1

x +
c
k!
(m + 1)!
ec

xk

k! + (m + 1)! cm+1 > k! .

k=0

k=0

Exercise 4.5.3. (a) Observe that a simpler version of this problem can be stated as follows: If f is
differentiable on (a, b) and x (a, b), then
lim

h0

f 0 (x)

f (x + h) f (x)

=
.
h
1!

This conclusion follows directly from the definition of derivative.

Similarly, if f is twice differentiable on (a, b) and x (a, b), then
lim

h0

f (x + h) f (x)
f 0 (x)
1!h
f 00 (x)

=
.
2
h
2!

We can prove this by applying the LHospital rule to get

f (x + h) f (x)
f 0 (x)
1!h
f 0 (x + h) f 0 (x)

f 00 (x)

lim
=
lim
=
.
h0
h0
h2
2h
2!
It is now clear that we can solve part (a) by using the LHospital rule as follows:
2

h2!
f (x + h) f (x)
f 0 (x)
1!h f 00 (x)
f 0 (x + h) f 0 (x)
f 00 (x)
1!h
f 000 (x)

lim
=
lim
=
.
3
2
h0
h0
h
3h
3!

139
Note that the last equality follows from the previous proof applied to the function f 0 .
(b) With the analysis from part (a), we see that if f is n times differentiable on (a, b) and x (a, b),
then

lim

f (x + h) n1
k=0

h0

hn+1

f (k) (x)h
k
f (n) (x)

k!
=
.
n!

This conclusion can be proved by induction. This general result can be applied to obtain the Taylor
expansion with Peanos remainder in Exercise 4.5.4.

SECTION 4.6
Exercise 4.6.3. We apply the definition to solve this problem. Given any u, v I and (0, 1), we
have
f ( u + (1 )v) f (u) + (1 ) f (v)
by the convexity of f .
Since f (u), f (v) J and J is an interval, f (u) + (1 ) f (v) J. By the nondecreasing
property and the convexity of ,
( f ( u + (1 )v)) ( f (u) + (1 ) f (v)) ( f (u)) + (1 ) ( f (v)).
Therefore, f is convex on I.
The result from this problem allows us to generate convex functions. For example, consider
f (x) = |x| and (x) = x p , p > 1. We have seen that f is convex on R. The function is convex
and increasing on [0, ) which contains the range of the function f . Therefore, the composition
2
g(x) = |x| p , p > 1, is convex on R. Similarly, h(x) = ex is also a convex function on R.
Observe that in this problem, we require the nondecreasing property of . A natural question is
whether the composition of two convex functions is convex. The answer is negative. Observe that
f (x) = x2 and (x) = |x 1| are convex, but ( f )(x) = |x2 1| is nonconvex.
Exercise 4.6.4. Use Theorem 4.6.6 or Corollary 4.6.7.
Exercise 4.6.5. (a) Use the obvious inequality

( a b)2 0.
Alternatively, consider the function f (x) = ln(x), x (0, ). We can show that f is convex on
(0, ). For a, b (0, ), one has


a+b
f (a) + f (b)
f

.
2
2
This implies
ln(
Therefore,

ln(a) ln(b)
a+b
)
= ln( ab).
2
2

a+b
ab.
2
This inequality holds obviously when a = 0 or b = 0.
(b) Use Theorem 4.6.3 for the function f (x) = ln(x) on (0, ).

140

Solutions and Hints for Selected Exercises

SECTION 4.7
Exercise 4.7.1. (a) By Theorem 4.7.5,

{a}, if x < 0;
f (x) = [a, a], if x = 0;

{a},
if x > 0.
(b) By Theorem 4.7.5,

{2},

[2, 0],
f (x) = {0},

[0, 2],

{2},

if
if
if
if
if

x < 1;
x = 1;
x (1, 1);
x = 1;
x > 1.

Exercise 4.7.3. To better understand the problem, we consider some special cases. If n = 1, then
f (x) = |x 1|. Obviously, f has an absolute minimum at x = 1. If n = 2, then f (x) = |x 1| + |x 2|.
The graphing of the function suggests that f has an absolute minimum at any x [1, 2]. In the case
where n = 3, we can see that f has an absolute minimum at x = 2. We then conjecture that if n is
odd with n = 2m 1, then f has an absolute minimum at x = m. If n is even with n = 2m, then f
has an absolute minimum at any point x [m, m + 1].
Let us prove the first conclusion. In this case,
2m1

f (x) =

i=1

2m1

|x i| =

fi (x),

i=1

where fi (x) = |x i|. Consider x = m. Then

fm (x)
= [1, 1], fi (x)
= {1} if i < m, fi (x)
= {1} if i > m.
The subdifferential sum rule yields f (x)
= [1, 1] which contains 0. Thus, f has an absolute
minimum at x.
If x > m, we can see that f (x)
(0, ), which does not contain 0. Similarly, if
x < m, then f (x)
(, 0). Therefore, f has an absolute minimum at the only point x = m.
The case where n is even can be treated similarly.
Exercise 4.7.5. Fix a, b R with a < b. By Theorem 4.7.9, there exists c (a, b) such that
f (b) f (a)
f (c) [0, ).
ba
This implies f (b) f (a) 0 and, hence, f (b) f (a). Therefore, f is monotone increasing on R,

Bibliography

[Die69]

J. Dieudonn. Foundation of Modern Analysis. Second Edition. Academic Press, Inc.,

1969.

[Fit09]

P. M. Fitzpatrick. Advanced Calculus. Second Edition. American Mathematical Society,

2009.

[FJ96]

J. Franchini and J.-C. Jacquens. Analyse I. Ellipses, Paris, 1996.

[HUL01] J.-B. Hiriart-Urruty and C. Lemarchal. Fundamentals of Convex Analysis. Springer,

2001.
[Lay13]

S. R. Lay. Analysis with an Introduction to Proof. Fifth Edition. Pearson, 2013.

[Mun00]

J. Munkres. Topology. Prentice Hall, 2000.

[Ros68]

M. Rosenlicht. Introduction to Analysis. Dover Publications, 1968.

[Rud76]

W. Rudin. Principles of Mathematical Analysis. Third Edition. McGraw-Hill, 1976.

[Sto00]

M. Stoll. Introduction to Real Analysis. Second Edition. Pearson, 2000.

[Wad09]

W. R. Wade. An Introduction to Analysis. Fourth Edition. Prentice Hall, 2009.