MATH/MTHE 281 Notes

Introduction to Real analysis

Department of Mathematics and Statistics

Queen’s University

Ole A. Nielsen R. Daniel Norman

This version: 2017/04/06

 i

Preface for 1997 version

Most of Mathematics 281 deals with some fundamental ideas about and proper-
ties of the real numbers and continuous functions. These ideas and properties include
sequences and series (which are studied in some first-year calculus), the least upper
bound property, elementary topology of R and Rd , convergence and uniform conver-
gence of series of functions, and term-by-term differentiation and integration of series
of functions. These notes will serve as a text for this part of the course.
For approximately twenty-five years ending with the 1996/97 academic year the
last seven or so weeks of the second-year honours calculus course (Mathematics 220)
dealt with the topics just described. In the 1970’s there was no suitable text dealing
with these topics and the first version of these notes was written by R. D. Norman; this
version was really an outline and omitted some proofs and contained few examples or
exercises. Over the years various other instructors added examples, comments, and
other additions, with the result that the already terse notes became even harder to
read. In 1992 I began revising these notes with the idea of adding exercises, an index,
a table of contents, and more examples and exercises. This has proceeded since then
with the assistance of Professor Norman and with students Peter Attia and Christina
MacRae in 1994, Peter Chamberlain in 1995, and Greg Baker in 1996. Versions of
these revised notes have been used since 1994/95.

Ole A. Nielsen
August 1997

Preface for ongoing revisions

In the Winter term of 2016, I taught MATH/MTHE 281 for the first time, and
thought it a good idea to take the LATEX for the original version of these notes, and
make minor revisions. Also, the intention is to arrive at a “living” version of these
course notes that can evolve, while still maintaining the suitability of the original
form as a course text for MATH/MTHE 281.

Andrew D. Lewis
2017/04/06
 Contents
0 Mathematical notation and logic 1
0.1 Mathematical notation . . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.1.1 Sets and elementary set notation . . . . . . . . . . . . . . . . 1
0.1.2 Notation for number systems . . . . . . . . . . . . . . . . . . 2
0.2 Mathematical logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.2.1 Propositions and symbolic logic . . . . . . . . . . . . . . . . . 3
0.2.2 Predicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.2.3 Quantifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.2.4 Assertions and proofs . . . . . . . . . . . . . . . . . . . . . . 5
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1 The real numbers 12

1.1 Construction of the nonnegative integers . . . . . . . . . . . . . . . . 12
1.2 Construction of the integers . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Construction of the rational numbers . . . . . . . . . . . . . . . . . . 14
1.4 The (lack of the) least upper bound property for Q . . . . . . . . . . 15
1.5 Cauchy sequences in Q . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6 The definition of the real numbers . . . . . . . . . . . . . . . . . . . 17
1.6.1 Using Dedekind cuts . . . . . . . . . . . . . . . . . . . . . . . 17
1.6.2 Using Cauchy sequences . . . . . . . . . . . . . . . . . . . . . 18
1.6.3 The punchline . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 The least upper bound property for R . . . . . . . . . . . . . . . . . 20
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Sequences of real numbers 25

2.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Subsequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 Cauchy sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3 Sequences in Rd 45
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Some topology in Rd 50
4.1 Open and closed sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2 Cluster points, limit points, and boundary points . . . . . . . . . . . 54
4.3 Properties of closed sets . . . . . . . . . . . . . . . . . . . . . . . . . 56
 iii

4.4 Continuous functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.5 Compact sets and open coverings . . . . . . . . . . . . . . . . . . . . 59
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5 Infinite series 69
5.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 Convergence of nonnegative series . . . . . . . . . . . . . . . . . . . 72
5.3 Convergence of general series . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Estimating remainders . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6 Sequences of functions 91
6.1 Motivating example . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Pointwise convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.3 Uniform convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.4 Properties of uniform convergence . . . . . . . . . . . . . . . . . . . 98
6.5 The sup-norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7 Power series and Taylor series 114

7.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . 114
7.2 Convergence of power series . . . . . . . . . . . . . . . . . . . . . . . 115
7.3 Properties of power series . . . . . . . . . . . . . . . . . . . . . . . . 119
7.4 Taylor series and Taylor polynomials . . . . . . . . . . . . . . . . . . 123
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

Index 139
 Chapter 0
Mathematical notation and logic

In this chapter, we provide the briefest of overviews of the basic mathematical no-
tation we shall use, and of mathematical logic, as this course is intended to introduce,
not just mathematical ideas, but also techniques of proof.

0.1 Mathematical notation

Part of the power of mathematics is that it provides clear and compact notation
for expression important ideas. In this section we give a rapid overview of the notation
we use.

0.1.1 Sets and elementary set notation

To properly define what one means by a “set” is not an easy task. Fortunately,
the work required to do this does not seem to be essential to function in mathematics.
For our purposes, a set is a well-defined collection of objects. The idea is that one
should be able to determine, at least in principle, whether something is a member of
a set or not. For example, one might have the “set of beagles with no eyes who live in
my house,” which is a set whose members consist of a single beagle named “Rudy.”
More mathematically, one might have “the set of even integers,” whose membership
is pretty easily recognized, or “the set of prime numbers” whose membership is less
easily recognized. A set is comprised of its elements or members, i.e., all those
things in the set.
Associated with constructions involving sets is some notation, which we list.
1. The empty set is the set with no elements, and is denoted by ∅.
2. Membership in a set is designated with the symbol “∈.” Thus x ∈ S means that
x is a member of the set S. Nonmembership is denoted by “6∈.”
3. A subset of a set A is another set, all whose elements are elements of S. If A is
a subset of S, we write A ⊆ S. If we wish to exclude the possibility that A = S,
we write A ⊂ S. If A is not a subset of S, then we write A 6⊆ S. Note that does
not mean that some elements of A are not in S, only that not all elements of A
are in S.
It is not uncommon to use ( in place of our ⊂.
2 0 Mathematical notation and logic

4. If S and T are sets, we denote by S × T the product of S and T . Thus S × T is

the set of elements we write as (x, y) for x ∈ S and y ∈ T . This is generalized in
an obvious to finite products.
5. If S and T are sets, then S ∪ T denotes the union of S and T , by which we mean
elements of S or T . That is, x ∈ S ∪ T if x ∈ S or x ∈ T .
6. If S and T are sets, then S ∩ T denotes the intersection of S and T , by which
we mean elements of S and T . That is, x ∈ S ∩ T if x ∈ S and x ∈ T .
7. The notions of union and intersection can be generalised to, not just two sets,
but arbitrary numbers of sets. Thus we suppose that we have a family of sets Sλ ,
λ ∈ Λ, where Λ is an index set. That is, for each λ ∈ Λ, Sλ is a set. Now we
denote ∪λ∈Λ Sλ to be union of the sets Sλ and ∩λ∈Λ Sλ to be the intersection of the
sets Sλ . Thus x ∈ ∪λ∈Λ Sλ if x ∈ Sλ for some λ ∈ Λ, and x ∈ ∩λ∈Λ Sλ if x ∈ Sλ
for all λ ∈ Λ.
8. For a set S and a subset A ⊆ S, the complement of A in S is the set S \ A of
elements of S that are not in A. We will also use the notation A{ for complement.
9. A map f from a set S to a set T is an assignment of a single element f (x) ∈ T
to each element x ∈ S. We write f : S → T to denote a map from S to T . The
set S is the domain of f and the set T is the codomain of f . We also denote

image(f ) = {f (x) | x ∈ S} ⊆ T.

Note that, generally, we will not have image(f ) = T .

0.1.2 Notation for number systems

We shall sketch the construction of the real numbers in Chapter 1. Here we assume
that the reader knows about real numbers, integers, rational numbers, etc., and we
merely provide the notation we use for these.
By Z we denote the set of integers. Thus Z consists of the numbers
. . . , −2, −1, 0, 1, 2, . . .. By Z>0 we denote the positive integers, i.e., the numbers
1, 2, . . .. By Z≥0 we denote the nonnegative integers, i.e., 0, 1, 2, . . .. By Q we denote
the set of rational numbers, i.e., fractions of integers. That is, q ∈ Q if q = nd for
some n ∈ Z and d ∈ Z>0 . By R we denote the set of real numbers, i.e., the points on
the number line that we learn about in our school days.
It is convenient to have special notation for certain subsets of real numbers, those
called intervals. There are nine types of intervals. In the above list, a and b are real
numbers with a ≤ b.

1. [a, b] = {x ∈ R | a ≤ x ≤ b} 5. [a, ∞) = {x ∈ R | a ≤ x}
2. [a, b) = {x ∈ R | a ≤ x < b} 6. (a, ∞) = {x ∈ R | a < x}
3. (a, b] = {x ∈ R | a < x ≤ b} 7. (−∞, b] = {x ∈ R | x ≤ b}
4. (a, b) = {x ∈ R | a < x < b} 8. (−∞, b) = {x ∈ R | x < b}
9. (−∞, ∞) = R
 0.2 Mathematical logic 3

0.2 Mathematical logic

In this course we will learn how to construct mathematical proofs. There are
a number of elementary logical tools needed to accomplish this. In this section we
overview these.

0.2.1 Propositions and symbolic logic

The basic object is a “proposition”: a statement that has the truth value of either
true or false. Here are some examples of propositions:
1. p = “The year is 2017.”
2. q = “Pierre Elliott Trudeau is dead.”
3. r = “Pierre Elliott Trudeau died in 2001.”
4. s = “every even number larger than 2 is a sum of two primes.”
We will probably agree that p is true if the year is 2017 (otherwise is is false),
and q are true, and r is false (Trudeau died in 2000). Such propositions are easily
proved, and do not require exotic notation to give a proof. The proposition s is one
whose truth value is not known (it is the Goldbach Conjecture). Proofs of more
difficult propositions such as this often require complicated logical arguments, and
our intention in this section is to sketch these.
First of all, given some propositions, we can create new ones by the four basic
logical operations that we now explain.
1. Negation: For a proposition p, the proposition ¬p has the opposite truth value of
p: if p is true then ¬p is false and vice-versa. Also ¬(¬p) has the same truth
value as p.
2. Conjunction: For propositions p and q, the proposition p ∧ q is true only if both
p and q are true, otherwise p ∧ q is false.
3. Disjunction: p ∨ q is true if either p or q (or both) is true.
4. Implication: p =⇒ q is true only if either p is false or q is true, i.e., p =⇒ q
is logically equivalent to ¬p ∨ q.
The kind of implications we will dealing with are often called “material implica-
tion” by logicians because it is not assumed that there is a causal connection between
p and q. In our examples above, we would agree that (p ∧ r) =⇒ q is true although
we admit that r is false.
Using the four basic operations, some additional constructions immediately follow
and will be useful.
1. Logical Equivalence: We have just seen that p =⇒ q means ¬p ∨ q, i.e., these
assertions are logically equivalent which we write (p =⇒ q) ⇐⇒ (¬p ∨ q). This
means that p =⇒ q and ¬p∨q have identical truth values. There are also relations
between ∨ and ∧, namely (p ∧ q) ⇐⇒ ¬(¬p ∨ ¬q), and (p ∨ q) ⇐⇒ ¬(¬p ∧ ¬q).
If we were being careful, we would have defined =⇒ and ∨ in terms of ¬ and ∧.
4 0 Mathematical notation and logic

However, we leave these details for a proper course in mathematical logic. Note
that using these rules ¬(p =⇒ q) ⇐⇒ (p ∧ ¬q).
2. Negation of an implication: An important application of the rules above occurs
when we wish to negate an implication. ¬(p =⇒ q) ⇐⇒ ¬(¬p∨q) ⇐⇒ p∧¬q.

0.2.2 Predicates
Let X be a set. A predicate p in X assigns to each x ∈ X the value true or
false. Note that, for each x ∈ X, p(x) is a proposition since it is either true or
false. Thus a predicate can be called a “propositional function.” The variable x is
the subject of the predicate. Often p(x) is thought of as being a “property” of x, as
will be clear from the examples below.
Predicates are often used to form conditional propositions, i.e., “if/then” state-
ments. Thus, if p and q are predicates in X, one has a proposition p(x) =⇒ q(x).
This means that, if x ∈ X is such that p(x) is true, then so too is q(x) true. As a
proposition itself, p(x) =⇒ q(x) may be true or false.
Here are some examples.
0.1 Examples: 1. Take X = Z>0 and consider the predicates p and q defined
by p(n)=“n is odd” and q(n)=“n is prime”. Thus p(1)=true, p(3)=true,
q(1)=false, and q(3)=true. Note that both of the propositions p(n) =⇒ q(n)
and q(n) =⇒ p(n) are false.
2. Let X be the set of maps f : [a, b] → R. Consider the two predicates p(f )=“f is
continuous” and q(f )=“there exists c ∈ (a, b) such that f (c) = 0”. The interme-
diate value theorem says that p(f ) =⇒ q(f ) is true.
3. Let p(x) be the predicate in R given by “x is such that 1 − x2 = 0.” This is true
if x = ±1 and false otherwise. •

0.2.3 Quantifiers
Associated with predicates are some useful logical constructions called “quanti-
fiers.”
1. Existential quantifier: ∃ x p(x) means that there is x such that p(x) is true. For
example, ∃ x x2 − 1 = 0 means that there is an x such that x2 − 1 = 0. We might
find it necessary to be more specific about where x is as in ∃ x (x ∈ R)∧(x2 −1 = 0),
which we shall write as ∃ x ∈ R, x2 − 1 = 0. We shall also write @x ∈ R, x2 + 1 = 0
for ¬(∃ x ∈ R, x2 + 1 = 0). Finally, we shall write ∃! x to mean there exists a
unique x, as in ∃! x ∈ R, x3 − 1 = 0.
2. Universal quantifier: ∀ x p(x) means that, for all x, p(x) is true. For example,
∀ x ∈ R x2 ≥ 0 means that the square of every real number is nonnegative.
In constructions such as p(x), x is a free variable; whereas in ∀ x p(x), x is a bound
variable. A bound variable is like a dummy variable in calculus: the proposition
∀ x p(x) is logically equivalent to ∀ y p(y).
 0.2 Mathematical logic 5

The quantifiers ∀ and ∃ are related by the rule ¬(∃ x p(x)) ⇐⇒ ∀ x ¬p(x) and
¬(∀ x p(x)) ⇐⇒ ∃ x ¬p(x). Indeed, one can define ∃ in terms of ∀ and ¬.
Let us give an example that illustrates how a well-known notion, that of continuity
of a function (see Section 4.4), can be expressed using our above symbolic logic
terminology.
0.2 Example: Let f be a function and p(x) be the proposition
∀  > 0 ∃ δ > 0 ∀ y |x − y| < δ =⇒ |f (x) − f (y)| < 
This is the way of saying that f is continuous at x.
Let us use our rules to find ¬p(x):
¬p(x) ⇐⇒ ¬(∀  > 0 ∃ δ > 0 ∀ y |x − y| < δ =⇒ |f (x) − f (y)| < )
⇐⇒ ∃  > 0 ¬(∃ δ > 0 ∀ y |x − y| < δ =⇒ |f (x) − f (y)| < )
⇐⇒ ∃  > 0 ∀ δ > 0 ¬(∀ y |x − y| < δ =⇒ |f (x) − f (y)| < )
⇐⇒ ∃  > 0 ∀ δ > 0 ∃ y¬(|x − y| < δ =⇒ |f (x) − f (y)| < )
⇐⇒ ∃  > 0 ∀ δ > 0 ∃ y |x − y| < δ ∧ |f (x) − f (y)| ≥ 
Thus ¬p(x) is how we write the proposition that f is not continuous at x. •

0.2.4 Assertions and proofs

In this section we consider propositions as assertions we wish to prove. While each
proof of a proposition tends to be its own entity, with its own particular personality,
there are some broad strategies for proving propositions that are common enough
that they are worth describing. Here we mention some of these, and give examples
of how they are used.

Types of logical assertions one might prove

There are, broadly, four type of assertions we will prove.
1. Prove that a proposition p is true.
2. Prove that a proposition p is false.
3. Given propositions p and q, prove that p =⇒ q. This is, prove that, if p is true,
then q is true.
4. Given propositions p and q, prove that p ⇐⇒ q. That is, prove that, if p is
true, then q is true and that, if q is true, then p is true.
5. Suppose that we have the proposition p =⇒ q. The converse proposition is
q =⇒ p. It is very frequent in informal speech to confuse q =⇒ p with p =⇒ q;
in mathematics this is a fatal error.
To be clear, 2 3, and 4 are special cases of 1 applied to the propositions p0 = ¬p,
p0 = (p =⇒ q), and p00 = (p ⇐⇒ q), respectively. Sometimes, however, it is
convenient to make use of the special structure of the proposition in these cases.
Next we describe some of the sorts of proofs one may employ.
6 0 Mathematical notation and logic

Direct proof
Here we wish to prove that, given propositions p and q, p =⇒ q. The method of
direct proof is that one uses p the fact that p is true, with no other assumptions, to
prove that q is true.
We illustrate this with a few simple examples.
0.3 Examples: 1. Consider the proposition p given by “4 is an even number.” To
prove that p is true, we note that 4 = 2 · 2, and so 4 is even, being an integer
multiple of 2.
2. Consider the proposition p given by “4 is a prime number.” To prove that p is
false, we note that 4 = 2 · 2, and so 4 has a factor other than 1 and 4.
3. We take p(n) to be the predicate with integer subject “n is a square of odd
numbers” and q(n) to be the predicate with integer subject “n is an odd number.”
We will show that p(n) =⇒ q(n). Suppose that n is such that p(n) is true.
Then n = m2 for an odd number m. Thus m = 2k + 1 for an integer k and

n = m2 = (2k + 1)2 = 2(2k 2 + 2k) + 1

and so q(n) is true. •

Proof by contradiction
For any proposition p, the proposition p ∧ ¬p is always false. If, in a proof, one shows
that a certain hypothesis implies the proposition p ∧ ¬p for some p, then we say we
have reached a contradiction and our hypothesis must be false. Sometimes, when
we wish to prove p is true, we assume that p is false and then reach a contradiction,
say q∧¬q. That is, we assume ¬p and then reach a contradiction; thus we have proved
¬(¬p)), i.e., p is true.
We give an example of proof by contradiction.
√ √
0.4 Example: We prove that 2 is irrational. Suppose that 2 = p/q for some
integers p and q with p and q relatively prime, i.e., having no common factor. Then
p2 = 2q 2 . So p2 is even and hence p is even and thus p is divisible by 2. Hence
p2 = (p0 )2 · 22 for p0 = p/2. We can now cancel a 2 to obtain 2(p0 )2 = q 2 and conclude
that q is also even. This contradicts our assumption that p and q are relatively prime.
Hence we have reached a contradiction. √ Thus our assumption that √ there were integers
p and q, relatively prime, such that 2 = p/q was false, i.e., 2 is irrational. •

Proof of contrapositive
We saw that p =⇒ q was logically equivalent to ¬p ∨ q, which is the same as
¬(¬q) ∨ ¬p, which finally is the same as ¬q =⇒ ¬p. This last expression is called
the contrapositive of p =⇒ q and we have just seen that (p =⇒ q) ⇐⇒
(¬q =⇒ ¬p).
Let us give an example of proof by contrapositive.
 0.2 Mathematical logic 7

0.5 Example: Let X = Z>0 ×Z>0 ×Z>0 , and consider the predicates p(n1 , n2 , n3 ) and
q(n1 , n2 , n3 ) in X given by “the product of n1 , n2 , and n3 is greater than 1000” and “at
least one of n1 , n2 , and n3 is greater than 10.” We will prove that p(n1 , n2 , n3 ) implies
q(n1 , n2 , n3 ) by proving the contrapositive. Then we assume that ¬q(n1 , n2 , n3 ) is
true, meaning that we assume that n1 , n2 , n3 ≤ 10. Then

n1 · n2 · n3 ≤ 10 · 10 · 10 = 1000.

Thus ¬p(n1 , n2 , n3 ) is true. •

Here is a more complicated example where the contrapositive notion is easier to
prove.
0.6 Example: Suppose we wish to prove that for two real numbers x and y we have
x = y. Quite often it will happen that we will have obtained one of the numbers,
x say, as the reult of some kind of limiting process. This makes it difficult to show
equality; so instead we show that x ≤ y and y ≤ x. However, sometimes we find even
showing x ≤ y too difficult, so we cut ourselves some further slack and prove that,
for every  > 0, we have x ≤ y + . Let us now show that this implies that x ≤ y.
Let p be the statement ∀  > 0 x ≤ y +  and q the statement x ≤ y. We have to
show p =⇒ q. This is logically equivalent to the contrapositive ¬q =⇒ ¬p. Let us
expand this out:

(¬q =⇒ ¬p) ⇐⇒ (¬(x ≤ y) =⇒ ¬(∀  > 0 x ≤ y + ))

⇐⇒ (x > y) =⇒ (∃  > 0 x > y + )

Now, for the last statement, we can give a direct proof: given x > y, let  = (x−y)/2,
then  > 0 and y +  = (x + y)/2 < x as required. •

Proof by induction
It is common to have situations where one has an assertion that is said to hold for
every n ∈ Z>0 . That is, one has a predicate p in Z>0 , and one wishes to prove the
proposition ∀ n ∈ Z>0 p(n). Frequently such propositions are proved by induction.
The way this works is as follows. The assertion p(1) is proved, somehow, someway.
Then the proposition p(n) is assumed; this is called the induction hypothesis.
Using the fact that p(n) is true, p(n + 1) is proved to be true, somehow, someway.
In this way we prove the proposition ∀ n ∈ Z>0 p(n). Another strategy that is rather
similar is strong induction, where the induction hypothesis is replaced with the
strong induction hypothesis, which is that p(k) is true for k ∈ {1, . . . , n}, and
then it is proved that p(n + 1) is true.
Let’s look at a couple of examples.
0.7 Examples: 1. Consider the predicate p in Z>0 given by

p(n)=“1 + · · · + n = 21 n(n + 1)”.

8 0 Mathematical notation and logic

We prove this by induction. For n = 1 we are asked to prove that p(1) = “1 =

1
2
· 1 · 2 = 1”, which is pretty clearly true. The induction hypothesis is now that,
for n ∈ Z>0 , we assume that
n
X
p(n) = “ k = 12 n(n + 1)”
k=1

is true, and we are to show that

n+1
X
p(n + 1) = “ k = 12 (n + 1)(n + 2)”
k=1

is true. We prove this as follows:

n+1
X n
X 
k= k + (n + 1) = 12 n(n + 1) + (n + 1) = 21 (n + 1)(n + 2).
k=1 k=1

2. As an example of a proof by strong induction, we offer the following. We X =

{n ∈ Z>0 | n ≥ 2} let p be the predicate
p(n) = “n is a product of prime numbers”.
In this case, although the index for p(n) is not Z>0 , we can clearly still use an
inductive proof. For n = 2 we have p(2) = “2 = 2”, giving 2 as a product of
primes. Now the strong induction hypothesis is that, for k ∈ {1, . . . , n},
p(k) = “k is a product of primes”
is true, and we are to show that
p(n + 1) = “n + 1 is a product of primes”
is true. To show that this is the case, there are two possibilities.
(a) n + 1 is prime: In this case, p(n + 1) is true immediately.
(b) n + 1 is not prime: In this case, the definition of “not prime” means that
n + 1 = a · b for a, b ∈ X. We claim that a, b ≤ n. Suppose otherwise and
that one of a or b, say a, exceeds n. Then
ab ≥ 2a > 2n = n + n > n + 1,
which contradicts the fact that ab = n + 1. Thus we can suppose that
a, b ≤ n. In this case, the strong induction hypothesis gives
a = p1 · · · pk , b = q1 · · · ql
for primes p1 , . . . , pk , q1 , . . . , ql . Thus
n + 1 = p1 · · · p k q1 · · · q l ,
showing that p(n + 1) is true. •
 0.2 Mathematical logic 9

Nonconstructive proof
In a nonconstructive proof, one proves a statement about the existence of something,
but without giving a specific instance of it, or an algorithm for finding it.
Here are two examples.
0.8 Examples: 1. If there are 367 people in a room, then at least two of the people
share the same birthday. A nonconstructive proof is immediate: if the mapping
from “person” to “birth date” is one-to-one, this would mean that there are at
least 367 days in a year, which cannot be. However, in this case a constructive
proof is also possible by interviewing all people in the room and establishing which
have the same birthday.
2. The intermediate value theorem asserts that, if f : [a, b] → R is a continuous
function for which f (a) < 0 and f (b) > 0, then there exists c ∈ (a, b) such that
f (c) = 0. While this theorem is true, a proof does not give any indication of what
c is or how to find it. •

Proof by counterexample
A counterexample is typically used to show that a proposition is not true. Normally
a counterexample is most useful in a situation where the proposition is of the form
p(x) for a predicate in a set X. To show that ∀ x ∈ X p(x) = true is true, one
must do exactly that: show that it is true for all x. To show that ∀ x p(x) is false,
one need only show that ∃ x ∈ X p(x) = false, i.e., a single x ∈ X is all that is
required, and this is said to be a counterexample.
Here are some examples.
0.9 Examples: We show that the converses to some propositions proved above are
false, by finding counterexamples.
1. We consider the predicates in Z>0 given by

p(n) = “n is a square of odd numbers”,

q(n) = “n is an odd number”,

and recall that in Example 0.3–3 we showed that p(n) =⇒ q(n) is true. We
now show by counterexample that q(n) =⇒ p(n) is false. Indeed, q(7) is true,
while p(7) is false, and so q(n) =⇒ p(n) is false.
2. We take X to be the set of mappings f : [a, b] → R and consider the two predicates
p(f )=“f is continuous” and q(f )=“there exists c ∈ (a, b) such that f (c) = 0”.
As remarked in Example 0.1–2, the proposition p(f ) =⇒ q(f ) is true. We will
show by counterexample that the proposition q(f ) =⇒ p(f ) is false. Indeed,
the mapping f ∈ X defined by
(
1, x = 21 (a + b),
f (x) =
0, otherwise
10 0 Mathematical notation and logic

satisfies q(f ) = true and p(f ) = false, and so the proposition q(f ) =⇒ p(f ) is
false. (Note: We should assume that a < b for this counterexample to be valid.
Indeed, if a > b, then X = ∅, and so q(f ) =⇒ p(f ) is true, since q(f ) is never
true. Also, the proposition is actually true if a = b.) •
 Exercises for Chapter 0 11

Exercises
E0.1 Let A be a set of real numbers. What does the following proposition say?

∀ x ∈ A ∃ y ∈ A (x < y) ∧ (∀ z ∈ A (x < z) =⇒ (y ≤ z))

Is there a set A for which this proposition is true? Is there a set A for which
it is false?
E0.2 Let A = [0, 1]. Let p(x) be the proposition

(∀ a ∈ A a ≤ x) ∧ (∀ b ∈ R (∀ a ∈ A a ≤ b) =⇒ x ≤ b)

For which x is p(x) true?

E0.3 (a) For each of the following, give (if possible) an example of a function
f : R → R that makes the proposition true.
1. ∀ x ∈ R ∃ y ∈ R s.t. f (x) = y
2. ∀ y ∈ R ∃ x ∈ R s.t. f (x) = y
3. ∃ y ∈ R s.t.∀ x ∈ R f (x) = y
(b) Give the negations of the above propositions.
E0.4 For some A ⊆ R, consider the following proposition:

∀ x ∈ A ∃ y ∈ A s.t.(x < y) ∧ (∀z ∈ A (x < z =⇒ y ≤ z))

(a) Describe simply what the proposition is saying.

(b) Find, if possible, a set A ⊆ R for which the above proposition is
1. true
2. false
 Chapter 1
The real numbers

The basic ingredient in “real analysis” is the set of real numbers. While we may
well have some understanding of what these are—we are taught about the “number
line” in school—a full and complete understanding of the crucial attributes of the
real numbers may not be entirely familiar. While we understand that the algebraic
properties of addition, subtraction, multiplication, and division are important, these
fall into the category of “algebra.” What concerns us here is “analysis,” and this
requires the property of the real numbers known of as “completeness.” This property
can be arrived at in a variety of ways, and we shall sketch two of these, the “least upper
bound property” and “convergence of Cauchy sequences.” Ultimately we arrive at
the former of these as the starting point for the properties of the real numbers that
allows one to do analysis. We do this by giving an overview, without all details,
of the manner in which one builds the real numbers “from nothing.” To do this
properly takes more space and effort than we can give here, but to not mention this
at all is rather unsatisfying. Thus we attempt to tread the line between making the
construction clear, but without making it excessively burdensome.

1.1 Construction of the nonnegative integers

The first construction we make is of the nonnegative integers. We do this recur-
sively, defining first 0, then 1, then 2, etc., as sets. To be precise, we define

0=∅
1 = 0 ∪ {0} = ∅ ∪ {∅} = {∅} = {0}
2 = 1 ∪ {1} = {0} ∪ {1} = {0, 1}
3 = 2 ∪ {2} = {0, 1} ∪ {2} = {0, 1, 2}
..
.

Note that 0 is a set with zero elements, 1 is a set with one elements, and so on. By
Z≥0 we denote the set whose elements are all the results of the preceding recursive
construction. We call this the set of nonnegative integers. Note that the lan-
guage here is confusing, since it seems to presuppose that we know what the integers
are. However, keep in mind that it is the “nonnegative integers” (all one notion) we
 1.2 Construction of the integers 13

are defining. The integers will come shortly.

We now define the familiar operations on Z≥0 . First we define “less than” and
“less than or equal to.” We write j < k if j ⊂ k and j ≤ k if j ⊆ k. One can
check that these operations have the expected interpretation and properties. We also
denote Z>0 = Z≥0 \ {0}, i.e., the set of numbers 1, 2, 3, . . ..
Now we need to define addition and multiplication in Z≥0 . Note that our definition
of Z≥0 essentially defines the operation of “adding 1” by k + 1 = k ∪ {k}. Given
k ∈ Z≥0 we define two maps ak : Z≥0 → Z≥0 and mk : Z≥0 → Z≥0 to be “addition of
k” and “multiplication by k.” The definition we give for these operations is recursive.
We define ak (0) = k and mk (0) = 0. Assume now that we have defined ak (j) and
mk (j) for j ≤ m. We then define
ak (m + 1) = ak (m) + 1, mk (m + 1) = mk (m) + k.
One can tediously verify that these definitions of addition and multiplication have
the expected commutativity, associativity, and distributivity properties.

1.2 Construction of the integers

Now we describe how one builds the integers from the set Z≥0 of nonnegative
integers. Here is the idea. Given m ∈ Z≥0 , one would like “−m” to have the property
that, if j ≥ m, then k = j + (−m) is that nonnegative integer which, when added to
m, gives j. There are many possible j’s and k’s for which this holds, and we define
−m to be the set of all of these. Here’s how we do this precisely.
We say that two pairs (j1 , k1 ) and (j2 , k2 ) of nonnegative integers are equivalent
if j1 + k2 = j2 + k1 . Given a pair (j, k) of nonnegative integers, we denote by [(j, k)]
the set of pairs of nonnegative integers that are equivalent to (j, k). We define the
set of integers to be the set {[(j, k)] | j, k ∈ Z≥0 } of equivalent pairs. One can then
easily prove the following.
1.1 Lemma: If (j, k) is a pair of nonnegative integers, then exactly one of the fol-
lowing statements holds:
(i) (j, k) is equivalent to (m, 0) for a unique m ∈ Z>0 ;
(ii) (j, k) is equivalent to (0, m) for a unique m ∈ Z>0 ;
(iii) (j, k) is equivalent to (0, 0).
With this in mind, we denote by m the set of pairs of nonnegative integers equiv-
alent to (m, 0), by −m the set of pairs of nonnegative integers equivalent to (0, m),
and by 0 the set of pairs of nonnegative integers equivalent to (0, 0). This now makes
it clear how we assign to our abstract notion of the set of integers to the one that
we think about in practice. We should also define the operations of addition and
multiplication in Z. This we do as follows:
[(j1 , k1 )] + [(j2 , k2 )] = [(j1 + j2 , k1 + k2 )],
[(j1 , k1 )] · [(j2 , k2 )] = [(j1 · j2 + k1 · k2 , j1 · k2 + j2 · k1 )].
14 1 The real numbers

Thus we define addition and multiplication only knowing addition and multiplication
in Z≥0 . Note that we can now define subtraction by
[(j1 , k1 )] − [(j2 , k2 )] = [(j1 , k1 )] + (−1) · [(j2 , k2 )].
One can verify, first of all, that this corresponds to the usual notions we have in mind
for addition and multiplication. Then one can tediously verify that all commutativity,
associativity, and distributivity properties we know about addition and multiplication
are valid.
One can also order the integers in the usual way. This is done with our abstract
definition by defining
[(j1 , k1 )] < [(j2 , k2 )] ⇐⇒ j1 + k2 < k1 + j2 ,
[(j1 , k1 )] ≤ [(j2 , k2 )] ⇐⇒ j1 + k2 ≤ k1 + j2 .
Again, one can work through the definitions to see that this corresponds to the
expected notion of “less than” and “less than or equal to,” and that it satisfies the
usual properties we know about these relations.
Finally, we define the absolute value function on Z by

k,
 0 < k,
|k| = 0, k = 0,

−k, k < 0.


This is the expected thing, and has all the expected properties, although, again, one
must tediously prove them.

1.3 Construction of the rational numbers

We next turn to constructing the rational numbers, i.e., fractions of the form kj .
Since we do not know what “division” means, we need to make this definition using
what we have at hand, which is the set of integers and all of the associated operations.
We again work with equivalence of pairs, but now pairs of integers.
Given pairs (j1 , k1 ), (j2 , k2 ) ∈ Z×Z>0 , we say they are equivalent if j1 ·k2 = j2 ·k1 .
Given a pair (j, k) ∈ Z × Z>0 , we denote by [(j, k)] the set of such pairs equivalent
to (j, k). The set of rational numbers is then the set {[(j, k)] | j ∈ Z, k ∈ Z>0 }
of equivalent pairs. We denote [(j, k)] = kj .
Now one defines the expected operations on Q. Thus we define
[(j1 , k2 )] + [(j2 , k2 )] = [(j1 · k2 + j2 · k1 , k1 · k2 )],
[(j1 , k1 )] · [(j2 , k2 )] = [(j1 · j2 , k1 · k2 )].
We can now also define division. If [(j1 , k1 )], [(j2 , k2 )] ∈ Q with j2 6= 0, we define
[(j1 , k1 )]
= [(j1 · k2 , j2 · k1 )].
[(j2 , k2 )]
 1.4 The (lack of the) least upper bound property for Q 15

One can verify easily that these are the usual operations for rational numbers, and
that they satisfy the expected properties of commutativity, associativity, and dis-
tributivity. We also define subtraction in the expected way:

[(j1 , k1 )] − [(j2 , k2 )] = [(j1 , k1 )] + [(−1, 1)] · [(j2 , k2 )].

We also have the order relations on Q given by

[(j1 , k1 )] < [(j2 , k2 )] ⇐⇒ j1 · k2 < j2 · k1 ,

[(j1 , k1 )] ≤ [(j2 , k2 )] ⇐⇒ j1 · k2 ≤ j2 · k1 .

These are easily verified to be the usual “less than” and “less than or equal to,” and
they have the expected properties. We denote by Q>0 the set of rational numbers
greater than 0.
Finally, we define the absolute value function on Q by

[(j, k)],
 [(0, 1)] < [(j, k)],
|[(j, k)]| = [(0, 1)], [(j, k)] = [(0, 1)],

−[(j, k)], [(j, k)] < [(0, 1)].


Again, one can verify that the usual properties of absolute value are valid.
Now we have at hand the set Q of rational numbers, with its expected operations
of addition, subtraction, multiplication, and division, with its expected relations <
and ≤, and with its absolute value function. Our next step is to build the real
numbers. There are two common ways of building the real numbers, and we shall
sketch both. They turn out to be equivalent, and this is something we will outline in
Chapter 2.

1.4 The (lack of the) least upper bound property for Q

The set of rational numbers has (at least) two defects, one of which we address in
this section, and the other of which we address in the next section.
We begin with some easily understood definitions.
1.2 Definition: Let A ⊆ Q.
(i) The set A is bounded if there exists M ∈ Q>0 such that |q| ≤ M for all q ∈ A.
(ii) A lower bound for A is a number ` ∈ Q such that ` ≤ q for all q ∈ A.
(iii) An upper bound for A is a number u ∈ Q such that q ≤ u for all q ∈ A.
(iv) A number ` ∈ Q is a greatest lower bound for A if `0 ≤ ` for every lower
bound `0 for A.
(v) A number u ∈ Q is a least upper bound for A if u ≤ u0 for every upper bound
u0 for A. •
16 1 The real numbers

One can easily see that A is bounded if and only if it has both an upper and lower
bound (Exercise E1.3).
The subjects of least upper bound and greatest lower bound are dominated by
two important questions:
1. If a set A is bounded, does it always possess a greatest lower bound and a least
upper bound?
2. If a set A possesses a least upper bound or greatest lower bound, is it unique?
The answer to these questions is, “No,” and, “Yes.” Just by using properties of Q
as enumerated so far, and not presupposing the existence of the real numbers, it is
not easy to prove that a bounded set might not possess, say, a least upper bound.
However, it one presupposes the existence of the real numbers, this is easy to see. So
let us consider this for the moment.
√ √
1.3 Example: Take A = {x ∈ Q | x ≤ 2}. We know that 2 is irrational by
Example 0.4. Now√suppose √ for√A. Thus x ≤ u for every
√ that u is any upper bound
x ∈ A. Since u > 2, u − 2 > 0. Thus u − 12 (u − 2) > 2 and so is also an upper
bound. Moreover, it is smaller than u, and so there can no no least upper bound. •
It is possible to make the preceding example make sense without presupposing
the existence of the real numbers, but that would take too much effort. And, in any
case, it is something that is at least intuitively clear.
The main point, however, is that the quite intuitive property of existence of least
upper bounds is not true in Q. It is true, however, for the set of real numbers.

1.5 Cauchy sequences in Q

Next we consider another way of characterising an important defect of the rational
numbers. It will not be too difficult to imagine that this defect is rather related to
the lack of existence of a least upper bound.
We consider sequences of rational numbers. In Definition 2.1 we carefully consider
what we mean by a sequence of real numbers, and this definition carries over to
sequences of real numbers. We will use this notion of sequences of rational numbers
in the following definition.
1.4 Definition: Let (qn ) be a sequence in Q.
(i) The sequence (qn ) is a Cauchy sequence if, for every  ∈ Q>0 , there exists
N ∈ Z>0 such that |qn − qm | <  for m, n ≥ N .
(ii) The sequence (qn ) converges to L ∈ Q if, for every  ∈ Q>0 , there exists
N ∈ Z>0 such that |qn − L| <  for n ≥ N . •
Now, sequences that converge are Cauchy; we will prove this for Cauchy sequences
of real numbers as one half of Theorem 2.23, and the proof for rational numbers is
exactly the same. However, it is not the case that Cauchy sequences of rational
numbers converge to some rational number. As with the least upper bound property
 1.6 The definition of the real numbers 17

in the preceding section, this is easy to imagine if one presupposes the real numbers,
but is not so easy to prove otherwise. We thus consider an example of a nonconvergent
Cauchy sequence in Q.
1.5 Example: For n ∈ Z>0 , consider the rational number qn = (1 + n1 )n . Perhaps
the reader has run into the sequence (qn ), and indeed it converges to the real number
e ≈ 2.718281828 . . ., i.e., to Euler’s constant. Since it is a convergent sequence of
real numbers, it is also a convergent sequence of rational numbers (convergence for
real numbers is defined in Definition 2.4, but, anyway, is exactly like for rational
numbers). Therefore, it is a Cauchy sequence of rational numbers. However, since
it’s limit in the set of real numbers is e, which is not rational, it cannot converge in
Q. •
It is possible to come up with nonconvergent Cauchy sequences without making
reference a priori to the set of real numbers, but it is simply difficult to do this.
This lack of necessity of Cauchy sequences converging is problematic, since Cauchy
sequences “seem like” they want to converge., and it would be ever so convenient if
they did.

1.6 The definition of the real numbers

In the previous two sections we gave two properties not possessed by the rational
numbers, but which we would like to be possessed, and in fact are possessed, by
the real numbers. Indeed, one can take these properties as the definition of the real
numbers in some sense. In this section we sketch how to do this in both cases, but
without going through the details.

1.6.1 Using Dedekind cuts

The method we outline here is to add least upper bounds to sets of rational
numbers that may not have one. The idea is made somewhat precise in the following.
A Dedekind cut in Q is a partition of Q into two sets A and B with the following
properties:
1. A and B are both nonempty;
2. A ∪ B = Q;
3. A is closed downwards, i.e., for all q, r ∈ A with q < r, if r ∈ A then q ∈ A;
4. B is closed upwards, i.e., for all q, r ∈ B with q < r, if q ∈ B then r ∈ B;
5. A contains no greatest element, i.e., there is no q ∈ A such that r ≤ q for all
r ∈ A.
The set R of real numbers is then the set of Dedekind cuts. Let us consider a few
examples of this to see how the definitions work. We do not do the work to show
that these are indeed Dedekind cuts.
18 1 The real numbers
√
1.6 Examples: 1. The real number we know as 2 is defined by the Dedekind cut

A = {q ∈ Q | q · q < 2}, B = Q \ A.

2. For n ∈ Z>0 , let ( )

n
X 1
An = q∈Q| q<
j=0
j!
and A = ∪∞ n=0 An . The number e can be defined to be the Dedekind cut with
defined by A and B = Q \ A.
3. Rational numbers are defined by Dedekind cuts as follows. Let q ∈ Q, and
associate to q the Dedekind cut

A = {r ∈ Q | r < q}, B = Q \ A. •

Note that a Dedekind cut is effectively defined by one of the sets A and B, and
in what follows we will work with the set A.
We need to define the standard constructions on R using this definition. First we
define the order relations “less than” and “less than or equal to” as follows:

A1 < A2 ⇐⇒ A1 ⊂ A2 , A1 ≤ A2 ⇐⇒ A1 ⊆ A2 .

Next we define addition and multiplication:

A1 + A2 = {a1 + a2 | a1 ∈ A1 , a2 ∈ A2 },


 {a1 · a2 | a1 ∈ A1 , a2 ∈ A2 , a1 , a2 ≥ 0} A1 , A2 ≥ 0,

∪{q ∈ Q | q ≤ 0},




A1 · A2 = −(A1 · (−A2 )), A1 ≥ 0, A2 < 0,

−((−A1 ) · A2 ), A1 < 0, A2 ≥ 0,





(−A ) · (−A ),
1 2 A1 , A2 < 0.

Finally, we define the absolute value function by


A,
 A > 0,
|A| = 0, A = 0,

−A, A < 0.


One can show that all these operations satisfy the expected properties. We leave
these rather tedious exercises to the reader.

1.6.2 Using Cauchy sequences

The construction we sketch here “fills in the gaps” between the convergent Cauchy
sequences with the nonconvergent Cauchy sequences. We will be slightly precise about
this.
 1.6 The definition of the real numbers 19

Two Cauchy sequences (qn ) and (rn ) of rational numbers are equivalent if, for
each  ∈ Q>0 , there exists N ∈ Z>0 such that |qn − rn | <  for every n ≥ N . The idea
is that equivalent Cauchy sequences have tails that get close to one another. Given a
Cauchy sequence (qn ), we denote by [(qn )] the set of all Cauchy sequences equivalent
to (qn ). We then define R, the set of real numbers, to be the set

{[(qn )] | [(qn )] is a Cauchy sequence in Q}.

This construction, then, includes in R all Cauchy sequences, even those that do not
converge. In this way, it is a natural sort of definition.
One now has to show that R has all of the attributes one is familiar with. First,
we indicate how we add and multiply in R:

[(qn )] + [(rn )] = [(qn + rn )],

[(qn )] · [(rn )] = [(qn · rn )].

One has many things to verify here. First of all, the definitions need to be shown to be
independent of the choice of equivalent Cauchy sequences. Second, one should show
that the righthand sides of the preceding definitions are, in fact, Cauchy sequences.
After this, one should show that addition and multiplication have the expected prop-
erties.
We can define the usual order relations on R as well. Specifically, we write [(qn )] <
[(rn )] if there exists N ∈ Z>0 such that qn < rn for n ≥ N . We also write [(qn )] ≤ [(rn )]
if either [(qn )] < [(rn )] or [(qn )] = [(rn )]. Of course, these order relations have all of
the expected properties.
Finally, we define the absolute value function on R. For this, we note that, if (qn )
is a Cauchy sequence, then (|qn |) is a Cauchy sequence. We then define

|[(qn )]| = [(|qn |)].

The absolute value function has all of the properties we expect. One of the most
important of these is the following:

|[(qn )] + [(rn )]| ≤ |[(qn )]| + |[(rn )]|,

which is called the triangle inequality . If we eliminate the tedious equivalence

class notation, the triangle inequality reads |a + b| ≤ |a| + |b| for a, b ∈ R.
What we arrive at is the set R with its familiar algebraic, order properties, and
valuation properties; but these properties are possessed by Q. What R has that Q
does not is the property that Cauchy sequences converge, an assertion we will prove
as Theorem 2.23. This important property has a name: completeness.

1.6.3 The punchline

In this section we have constructed the real numbers in two different ways. It
turns out that the two constructions give equivalent outcomes. That is, the sets of
20 1 The real numbers

real numbers we construct in each case amount to the same thing. The way one
normally proceeds is by selecting one of the constructions as the starting point, and
proves everything with this as a starting point. That is to say, one works with
either (1) all bounded sets in R have a least upper bound or (2) all Cauchy sequences
in R converge. In this text we will be assuming that all bounded sets have a least
upper bound, and proving facts with this as a starting point.

1.7 The least upper bound property for R

Based on our discussion above, we will assume that all bounded subsets of R have a
least upper bound. From the point of view of analysis, the least upper bound property
is one of the most fundamental properties of the real numbers. This property (or
something equivalent to it) is usually stated as an assumption in a first-year calculus
course, and the purpose of this section is to discuss and explain this property in more
detail than we managed above in our rather hurried construction of the real numbers.
We begin with definitions we have already seen for rational numbers, but now in
the setting of real numbers.
1.7 Definition: A nonempty subset A of R is said to be bounded below if there is a
number s such that s ≤ a for all a ∈ A and in this case any such number s is called a
lower bound for A. Similarly, A is said to be bounded above if there is a number
t such that a ≤ t for all a ∈ A, and any such number t is called an upper bound for
A. Finally, A is said to be bounded if it is both bounded below and bounded above
and to be unbounded if it is not bounded. •
In general, a set need not be bounded above, but if it is then it will necessarily
have an infinite number of upper bounds. Indeed, if a set is bounded above then any
number larger than an upper bound is also an upper bound.
1.8 Example: Consider the sets A = [0, 1] and B = (−∞, 1). Here A is bounded
below and bounded above, and hence bounded, while B is bounded above but is not
bounded below. Moreover, the lower bounds for A are the numbers in the interval
(−∞, 0], the upper bounds for A are the numbers in the interval [1, ∞), and the sets
A and B have the same upper bounds. Notice that the smallest of the upper bounds
for A belongs to A but that the smallest of the upper bounds for B does not belong
to B. •
Suppose that A is a nonempty subset of R that is bounded above. Then the set U
of all upper bounds of A is an infinite set which is bounded below; in fact, U is bounded
below since any number in A is a lower bound for U . One can now ask whether U
contains a number which is smaller then every other number in U , i.e., whether there
is an upper bound for A which is smaller than every other upper bound for A. It is
clear that there can be at most one such smallest upper bound, i.e., if it exists then
it is unique. But, as the reader may believe from our discussion in the first part of
this chapter, it is not at all clear whether a smallest upper bound must always exist.
 1.7 The least upper bound property for R 21

However, our construction of the reals by Dedekind cuts above is precisely designed
so that the following theorem holds.
1.9 Theorem: Every nonempty subset of R that is bounded above has a least upper
bound, i.e., an upper bound that is smaller than any other upper bound. Similarly,
every nonempty subset of R that is bounded below has a greatest lower bound, i.e., a
lower bound that is greater than any other lower bound.
This theorem justifies the following definition.
1.10 Definition: Suppose that A is a nonempty subset of R.
(i) If A is bounded above then the smallest upper bound for A is called the least
upper bound or supremum of A and is denoted by sup A.
(ii) If A is bounded below then the largest lower bound for A is called the greatest
lower bound or infimum of A and is denoted by inf A. •
The proof of the following result is both easy and instructive.
1.11 Proposition: If A and B are two nonempty subsets of R that are bounded
above and if A ⊆ B then sup A ≤ sup B.
Proof: If a is any element of A then a ∈ B since A ⊆ B, and thus a ≤ sup B since
sup B is an upper bound for B. But this means that sup B is an upper bound for A,
and therefore sup A ≤ sup B by the definition of sup A as the smallest upper bound
of A. 
The next result gives a useful characterization of the supremum of those sets that
are bounded above.
1.12 Theorem: If A is a nonempty subset of R that is bounded above and if s ∈ R
is an upper bound for A, then s = sup A if and only if
(i) a ≤ s for all a ∈ A and
(ii) A ∩ (s − , s] 6= ∅ for all  > 0.
Proof: Suppose first that s = sup A. Then s is an upper bound for A and thus (i)
holds. If (ii) were false there would be an  > 0 such that A ∩ (s − , s] = ∅. Then,
since s is an upper bound for A and since A ∩ (s − , s] = ∅, s −  would also be an
upper bound for A. But this contradicts the fact that s is the least upper bound for
A, and therefore (ii) must be true.
Now suppose that, conversely, (i) and (ii) are true. Then (i) implies that s is an
upper bound for A. If s were not the least upper bound for A then there would be an
upper bound t for A such that t < s. Thus a ≤ t for all a ∈ A and so if  = (s − t)/2
then t +  = s −  and a < t +  for all a ∈ A, and thus A ∩ (s − , s] = ∅. This
contradicts (ii) and shows that s = sup A. 
Theorem 1.12 implies the following assertion that is sometimes useful.
1.13 Corollary: If A is a nonempty subset of R that is bounded above and if s =
sup A then either s ∈ A or else A ∩ (s − , s) 6= ∅ for all  > 0.
22 1 The real numbers

Exercises
E1.1 For j1 , j2 , k ∈ Z, prove the distributive rule (j1 + j2 ) · k = j1 · k + j2 · k.
E1.2 Show that the relations < and ≤ on Z have the following properties:
1. [(0, j)] < [(0, 0)] for all j ∈ Z>0 ;
2. [(0, j)] < [(k, 0)] for all j, k ∈ Z>0 ;
3. [(0, j)] < [(0, k)], j, k, ∈ Z≥0 , if and only if k < j;
4. [(0, 0)] < [(j, 0)] for all j ∈ Z>0 ;
5. [(j, 0)] < [(k, 0)], j, k ∈ Z≥0 , if and only if j < k;
6. [(0, j)] ≤ [(0, 0)] for all j ∈ Z≥0 ;
7. [(0, j)] ≤ [(k, 0)] for all j, k ∈ Z≥0 ;
8. [(0, j)] ≤ [(0, k)], j, k, ∈ Z≥0 , if and only if k ≤ j;
9. [(0, 0)] ≤ [(j, 0)] for all j ∈ Z≥0 ;
10. [(j, 0)] ≤ [(k, 0)], j, k ∈ Z≥0 , if and only if j ≤ k.
E1.3 Show that a subset A ⊆ Q is bounded if and only it is has a lower bound and
an upper bound.
E1.4 For each of the following subsets of R, find the least upper bound and greatest
lower bound if they exist. n n  o
(d) n ∈

Z >0
n+1

(a) {1, 3} n 
n (−1)  o
(b) [0, 4] (e) n+  n ∈ Z>0
n1  o n
∞
(c)  ∈ Z>0
  
\ 1 1
n (f) 1 − ,1 +
n=1
n n
E1.5 Find the least upper bound and the greatest lower bound of the sets below.
For any given  > 0 find a number in the set that is greater than sup A −  and
a number in the set that is smaller than inf A + .
n4 + x  o
(a) A =  x≥1

x
n √x − 1  o
(b) A = x ≥ 2

x

2
(c) A = {x | x − x < 6}
E1.6 If S = {x ∈ R | x2 + x < 3}, find sup S and inf S.
Hint: Draw a graph.
n n  o
E1.7 Let A =  n ∈ Z>0 and notice that sup A = 1. For  = 0.01 and for

n+1
 = 0.0001 find an element of A larger that 1 − .
E1.8 Suppose A is a nonempty bounded subset of R and let −A denote the set
{−x | x ∈ A}. Show that sup(−A) = − inf A.
E1.9 For each nonempty subset S of R and each real number k, let kS = {ks | s ∈
S}.
 Exercises for Chapter 1 23

(a) Suppose that S = {−1, 2, 4} (so that 3S = {−3, 6, 12} and −2S =
{2, −4, −8}). Find sup S, sup 3S, sup −2S, inf S, inf 3S, and inf −2S.
Do you notice a pattern?
(b) If S is any nonempty bounded subset of R show that:
1. If k ≥ 0 then sup kS = k · sup S and inf kS = k · inf S.
2. If k < 0 then sup kS = k · inf S and inf kS = k · sup S.
E1.10 Prove that, if a subset S of R has a maximal element s (that is, there exists
an s ∈ S such that x ≤ s for all x in S), then s = sup S.
E1.11 Prove that x = sup{q ∈ Q | q < x} for each x ∈ R.
E1.12 Suppose that A and B are two nonempty bounded subsets of R and let

A + B = {x + y | x ∈ A and y ∈ B}.

Show that
sup(A + B) = sup A + sup B.
E1.13 Let A and B be two nonempty sets of real numbers that are bounded above
and put a = sup A, b = sup B, and

AB = {xy | x ∈ A and y ∈ B}.

Answer the following questions.

(a) Give an example to show that, in general, sup(AB) 6= ab.
(b) Show that, if a < 0 and b < 0, then inf AB = ab.
(c) Show that, if inf A > 0 and inf B0, then sup AB = ab.
E1.14 For nonempty sets A, B ⊆ R determine which of the following statements are
true and which are false. Prove those that are true and give counterexamples
for those that are false.
(a) sup(A ∩ B) ≤ min{sup A, sup B}
(b) sup(A ∩ B) = min{sup A, sup B}
(c) sup(A ∪ B) ≥ max{sup A, sup B}
(d) sup(A ∪ B) = max{sup A, sup B}
E1.15 Let A be a nonempty subset of R that is bounded above and put s = sup A.
(a) Show that if s 6∈ A then the set A ∩ (s − , s) is infinite for any  > 0.
(b) Give an example to show that, if s ∈ A, then the set A ∩ (s − , s) may
be finite for each  > 0.
E1.16 Let a and b be two numbers satisfying a < b.
(a) Show that the set {x | a < x < b} contains neither its least upper bound
not its greatest lower bound.
(b) Show that the set {x | a ≤ x ≤ b} contains both its least upper bound
and its greatest lower bound.
24 1 The real numbers

E1.17 Let S denote the set consisting of all those numbers with decimal expansions
of the form x = 0.a1 a2 a3 . . ., where all but finitely many of the digits a1 , a2 , . . .
are 5 or 6. Find sup S and inf S.
E1.18 Suppose that P is a subset of [0, ∞) with the property that, for any integer k,
there is an xk ∈ P such that kxk ≤ 1. Prove that 0 = inf P .
E1.19 If P and Q are two subsets of R such that sup P = sup Q and inf P = inf Q
does it follow that P = Q? Support your answer with either a proof or a
counterexample.
E1.20 For any two nonempty subsets of R let us write P ≤ Q if, for each x ∈ P ,
there is a y ∈ Q satisfying x ≤ y.
(a) Show that, if P ≤ Q, then sup P ≤ sup Q.
(b) Give an example to show that, if P ≤ Q, then it does not follow that
inf P ≤ inf Q.
(c) Give an example to show that, if P ≤ Q and if Q ≤ P , then it does not
follow that P = Q.
E1.21 Let S be a nonempty subset of R that is bounded below. Show that inf S =
sup{x ∈ R | x is a lower bound for S}.
E1.22 Prove that every nonnegative real number has a nonnegative square
root, i.e., prove that, if x is any nonnegative real number, then there is non-
negative number y such that y 2 = x.
Hint: Consider the set sup{u ∈ R | u2 ≤ x}.
E1.23 One of the important consequences of Theorem 1.9 is the Archimedean prop-
erty of R: The set Z>0 of positive integers is not bounded above in R. Although
this property may seem obvious, its proof actually depends on Theorem 1.9.
Prove the Archimedean property of R.
E1.24 Suppose that D is a nonempty subset of R and that f and g are two functions
from D to R such that f (x) ≤ g(y) for all x, y ∈ D. Show that f (D) is
bounded above, that g(D) is bounded below, and that sup f (D) ≤ inf g(D).
(Recall that f (D) = {f (x) | x ∈ D}.)
E1.25 Let D be a nonempty subset of R and suppose that f and g are two real-valued
functions defined on D. Let f + g denote the function on D defined by the
formula (f + g)(x) = f (x) + g(x) for all x ∈ D.
(a) If f (D) and g(D) are bounded above prove that (f + g)(D) is bounded
above and that sup(f + g)(D) ≤ sup f (D) + sup g(D).
(b) Show by means of an example that the inequality in (a) may be strict.
(c) State and prove the analog of (a) for infima.
 Chapter 2
Sequences of real numbers

Sequences arise in a variety of ways in real analysis. They show up, for example,
when considering convergence of infinite series, such as we shall discuss in Chapter 5.
What is less clear to a newcomer, however, is that sequences in some way characterize
the very fabric of the set R of real numbers, and also its multi-dimensional version
Rd .

2.1 Definitions and examples

We get the ball rolling with the basic definitions and some elementary examples.
2.1 Definition: Let Z>0 denote the natural numbers, i.e., the positive integers. A
sequence (of real numbers) is a function from Z>0 to R. If a is a sequence, it is usual
to write an in place of a(n), to call an the nth term of the sequence, and to denote
the sequence by (an ), or by (an )n∈Z>0 if we wish for the index set to be explicit. •
It is customary to think of terms of a sequence as being ordered, so that a5 “occurs
before” a7 in the sequence, and so on. It is sometimes convenient to “start” a sequence
with a0 (or a2 or even a4 ) instead of a1 .
Sequences may be defined explicitly, e.g., by writing
n
an = ,
n+1
or by recursion, e.g., by the formulae
an
a1 = 1, an+1 =
n+1
or
a0 = a1 = 1, an+2 = an+1 + an .
It is common (but not strictly correct) to describe a sequence by its first few
terms, e.g.,
2, 4, 8, 16, . . . .
The difficulty is that the terms are not really determined this way—there is an in-
teresting sequence whose first five terms are 2, 4, 8, 16, 31 (the number of regions
26 2 Sequences of real numbers

obtained by joining every pair of points among n points in general position on the
circumference of a circle for n = 2, 3, 4, 5, . . .).
Beware of the difference between a sequence and the range of that sequence. The
range of a sequence is analogous to the range of a function, i.e., the set of values
attained by the sequence (or function). The following two examples illustrate this.
2.2 Examples: 1. The sequence 1, 2, 3, 1, 2, 3, 1, 2, 3, . . . has an infinite number of
terms, but its range contains only the three points 1, 2, 3.
2. The range of the sequence (sin nπ
2
) is the set {−1, 0, 1}. •

2.3 Definition: A sequence (an ) is said to be bounded if there exist an M > 0 such
that |an | ≤ M for all n. •
Notice that, according to this definition, a sequence is bounded if and only if its
range is a bounded set in the sense of Definition 1.7.
2.4 Definition: A sequence (an ) is said to have a limit L if, for any  > 0, there
exists a number N (depending on ) such that

|an − L| < 

for every integer n ≥ N . If (an ) has a limit L, then we say that the sequence is
convergent and that it converges to L and we write limn→∞ an = L. If (an ) is not
convergent, it is said to diverge or to be divergent. If, for every c > 0, there exists
a number N such that an > c for all n ≥ N , we will write limn→∞ an = ∞ and say
that (an ) diverges to ∞. Similarly, if, for every c > 0, there exists a number N
such that an < −c for all n ≥ N , we will write limn→∞ an = −∞ and say that (an )
diverges to −∞. •

2.5 Examples: 1. If an = n, then limn→∞ an = ∞ or (an ) diverges to ∞.

2. If an = (−1)n then limn→∞ an does not exist, (an ) is divergent, but it does not
diverge to ±∞.
n
3. If an = n+1 then limn→∞ an = 1 or (an ) converges to 1. •
In simple cases, such as Example 2.5–3, we can establish that the limit L exists
by determining L (and then, if necessary, applying Definition 2.4 to prove that it is
the limit). However, in many important cases it is not easy to find L or to apply
the definition directly. For such cases we need general theorems to help us establish
whether a limit exists. The two most important such results are Theorem 2.9 and
Theorem 2.23. The next theorem is a first negative step in this direction (negative
because it says “an unbounded sequence cannot converge”).
2.6 Proposition: Every convergent sequence of real numbers is bounded.
Proof: Suppose that (an ) is a convergent sequence that converges to L. Then, by
definition, for any  > 0 there exists an N > 0 such that |an − L| <  for all n ≥ N .
In particular, choosing  = 1 gives an N > 0 such that L − 1 < an < L + 1 for all
n ≥ N . The set {a1 , a2 , . . . , aN −1 } is bounded because it is finite and hence has a
 2.1 Definitions and examples 27

minimum and a maximum; let A and B be the minimum and maximum, respectively.
So if m = min{A, L − 1} and if M = max{B, L + 1}, then clearly m ≤ an ≤ M for
all n and so (an ) is bounded. 
Example 2.5–2 shows that the converse of Proposition 2.6 is false and so bound-
edness is necessary but not sufficient for convergence. To obtain a sufficient condition
for convergence, we introduce the following definition.
2.7 Definition: A sequence (an ) is said to be nondecreasing if am ≤ an whenever
m < n and to be strictly increasing if am < an whenever m < n. Similarly, the
sequence is said to be nonincreasing if am ≥ an whenever m < n and to be strictly
decreasing if am > an whenever m < n. Finally, (an ) is said to be monotone if it
is either nondecreasing or nonincreasing. •
Note that a strictly increasing sequence is nondecreasing and a strictly decreasing
sequence is nonincreasing. The point of this definition is to have a formal way to
avoid the ambiguous term “increasing,” which sometimes means strictly increasing
and, at other times, nondecreasing.
If (an ) is a nondecreasing sequence and if A = {an | n ∈ Z>0 }, then clearly
a1 = inf A and (an ) is bounded if and only if A is bounded above.
The following example shows that a strictly increasing sequence may be bounded
or unbounded. The terms in part 1 of this example are the partial sums of the
exponential series while those in part 2 are the partial sums of the harmonic series,
cf. Example 5.2–4.
2.8 Examples: 1. If
1 1 1
an = 1 + 1 + + + ··· +
2! 3! n!
n−1
then an < an+1 for all n and, since n! > 2 if n ≥ 2,
1 1 1 1
an < 1 + 1 + + + + · · · + n−1
2 4 8 2
1−n
=3−2 <3
by the familiar formula for the sum of a geometric series. The sequence (an ) is
therefore strictly increasing and bounded.
2. If
1 1 1
an = 1 + + + · · · +
2 3 n
then an+1 > an for all n and
     
1 1 1 1 1 1 1
a2n = 1 + + + + + ··· + + ··· + + ··· + n
2 3 4 5 8 2n−1 + 1 2
     
1 1 1 1 1 1 1 1 1
>1+ + + + + + + +··· + n
+ ··· + n
2 4 4 8 8 8 8 2 2
| {z } | {z } | {z }
2 terms 4 terms 2n−1 textrmterms
1 1 n
>1+ + ··· + = 1 + .
2 2 2
28 2 Sequences of real numbers

The sequence (an ) is thus strictly increasing and unbounded. •

The two series in this example illustrate the next theorem.
2.9 Theorem: Every bounded monotone sequence is convergent. More precisely, if
(an ) is a bounded nondecreasing sequence then limn→∞ an = sup{an | n ∈ Z>0 }.
Proof: Suppose that (an ) is a bounded nondecreasing sequence. Then {an | n ∈ Z>0 }
is a nonempty bounded set; let L = sup{an | n ∈ Z>0 }. Then, by Theorem 1.12,
given any  > 0, there is an m such that L −  < am ≤ L. Now, for any n ≥ m, we
have am ≤ an ≤ L by the definition of L and thus |L − an | < . This shows that (an )
converges and that limn→∞ an = L.
The case of a nonincreasing sequence is similar. 
Now looking back at Example 2.8, we see that the sequence in part 1 is conver-
gent by Theorem 2.9, while the sequence in 2 is divergent by Proposition 2.6. Note
that Proposition 2.6 and Theorem 2.9 together imply that a monotone sequence is
convergent if and only if it is bounded.
While Theorem 2.9 is a very useful result, we require some additional ideas in
order to discuss sequences that are not monotone and/or do not converge.

2.2 Subsequences
A subsequence of a sequence is, roughly speaking, a sequence that is obtained by
taking only certain terms from the given sequence in the given order. Thus

2, 4, 6, 8, . . .

is a subsequence of 1, 2, 3, . . . whereas

2, 1, 8, 7, 3, 2, 31, . . .

is not. It is possible for a subsequence of a divergent sequence to converge. For

example, both of the sequences 1, 1, 1, . . . and 2, 2, 2, . . . are convergent subsequences
of the divergent sequence 1, 2, 3, 1, 2, 3, 1, 2, 3, . . ..
Slightly more generally, if (an ) is a sequence, then (a2n ) or a2 , a4 , a6 , . . . is a sub-
sequence and (a3n+1 ) or a1 , a4 , a7 , . . . is another. Here is a formal definition.
2.10 Definition: If (an ) is a sequence and if (nk ) is a strictly increasing sequence of
positive integers, then the sequence (ank ) is said to be a subsequence of (an ). •

2.11 Example: Suppose that (an ) is a sequence that is bounded above and put t =
sup{an | n ∈ Z>0 }. We consider two cases.
1. If an = t for some n, then t may or may not be the limit of a convergent subse-
quence whereas, if an < t for all n, then t is the limit of a convergent subsequence.
To see this, first notice that, if (an ) is the sequence 1, 0, 0, . . ., then t = 1 and no
subsequence of (an ) converges to 1.
 2.2 Subsequences 29

2. Now suppose that an < t for all n. We will show that, in this case, there is a
subsequence (ank ) that converges to t.
First we make an observation. For any number  > 0 and any positive integer m,
consider the number

0 = min{/m, t − a1 , t − a2 , . . . , t − am }.

Then 0 > 0 and so, by Theorem 1.12, there will be an integer n such that
t − 0m < an < t, and n > m by the definition of 0 . We shall use this observation
repeatedly to construct the subsequence (nk ).
Let n1 = 1 and let 1 = t − a1 . Then 1 > 0 and there is an n2 such that
t − 1 < an2 < t and n2 > n1 . Now, if

2 = min{1 /2, t − a1 , t − a2 , . . . , t − an2 },

then there is an n3 such that t − 2 < an3 < t and n3 > n2 . Next, if

3 = min{2 /2, t − a1 , t − a2 , . . . , t − an3 },

then there is an n4 such that t − 3 < an3 < t and n4 > n3 . Continuing in this
manner will give a subsequence (ank ) that converges to t. •
The following lemma (whose proof is easy and will be omitted) asserts that every
subsequence of a convergent sequence converges to the same limit. What turns out to
be much more interesting is that, as several of the above examples show, a subsequence
of a divergent sequence may well converge.
2.12 Lemma: If (ank ) is a subsequence of a sequence (an ) and if limn→∞ an = L,
then limk→∞ ank = L.
The next theorem turns out to be crucial for the subsequent development of the
theory and is even equivalent to the least upper bound property.
2.13 Theorem: (Bolzano–Weierstrass theorem) Every bounded sequence has a
convergent subsequence.
Proof: Suppose that (an ) is a bounded sequence and let m and M be numbers sat-
isfying m ≤ an ≤ M for all n. Put c = (m + M )/2, the midpoint of [m, M ]. If the
inequality m ≤ an ≤ c holds for infinitely many n, let m1 = m and M1 = c. On the
other hand, if the inequality m ≤ an ≤ c holds for only finitely many n, then the
inequality c ≤ an ≤ M must hold for infinitely many n, and in this case put m1 = c
and M1 = M . So, in either case [m1 , M1 ] ⊆ [m, M ], M1 − m1 = 2−1 (M − m), and
the inequality m1 ≤ an ≤ M1 holds for infinitely many n. Now let n1 be any integer
for which m1 ≤ an1 ≤ M1 .
The argument just described can be repeated. Let c1 be the midpoint of [m1 , M1 ].
Since the inequality m1 ≤ an ≤ M1 holds for infinitely many n, it follows that
m1 ≤ an ≤ c1 for infinitely many n or c1 ≤ an ≤ M1 for infinitely many n or both.
30 2 Sequences of real numbers

So by letting either m2 = m1 and M2 = c1 or by letting m2 = c1 and M2 = M1 we

can be sure that [m2 , M2 ] ⊆ [m1 , M1 ], that
M2 − m2 = 2−1 (M1 − m1 ) = 2−2 (M − m),
and that m2 ≤ an ≤ M2 for infinitely many n. Now let n2 be any integer satisfying
n2 > n1 and m2 ≤ an2 ≤ M2 .
We can evidently continue this construction (or more formally, proceed by recur-
sion) and doing so will yield two sequences (mk ) and (Mk ) of numbers and a sequence
(nk ) of positive integers with the following properties:
m1 ≤ m2 ≤ · · · ;
M1 ≥ M2 ≥ · · · ;
n1 < n2 < · · · ;
Mk − mk = 2−k (M − m);
mk ≤ ank ≤ Mk ;
for each k.
Now (mk ) is a bounded nondecreasing sequence and so, by Theorem 2.9, it con-
verges to L = sup{mk | k ∈ Z>0 }. It is easy to see that, if j and k are any two
positive integers, then mj < Mk . Indeed, if j ≤ k then mj ≤ mk < Mk , whereas, if
j ≥ k, then mj < Mj ≤ Mk . This means that, for each k ∈ Z>0 , Mk is an upper
bound for the set {mj | j ∈ Z>0 } and hence L ≤ Mk . It now follows that both ank
and L belong to the interval [mk , Mk ] and hence that
|L − ank | ≤ Mk − mk = 2−k (M − m)
for each k, and this clearly implies that limk→∞ ank = L. This means that (ank ) is
the required convergent subsequence. 
Convergent sequences are in some sense easy to deal with, but we need to gain
a better understanding of divergent sequences. Recall from Lemma 2.12 and Theo-
rem 2.13 that, if a sequence converges to a limit L, then every subsequence converges
to L. On the other hand, it is clear from Example 2.14 (below) that a subsequence of
a divergent sequence may converge or diverge, and that the convergent subsequences
need not all converge to the same limit. This suggests that a suitable “replacement”
for the limit of a convergent sequence in the context of arbitrary sequences might be
either
1. the set of all limits of convergent subsequences or
2. the largest and smallest numbers that are the limits of convergent subsequences.
Now 2 seems to be simpler than 2, but there is a potential problem with 2. Namely,
even if the set of all limits of convergent subsequences is bounded, it is not clear that
this set contains its least upper bound or, equivalently, that it has a largest element.
The next lemma will show that this is, in fact, not a problem and that we really can
use 2. But before giving this lemma it will be useful to first present some examples
of sequences and, for each one, the set all limits of convergent subsequences.
 2.2 Subsequences 31

2.14 Example: Here are six sequences and, for each one, the corresponding set of all
limits of convergent subsequences:
Sequence Set of limits of all convergent subsequences
1, 12 , 13 , . . . {0}
1, −1, 1, −1, . . . {−1, 1}
1, 2, 3, . . . ∅
1, 23 , 13 , 54 , 15 , 76 , . . . {0, 1}
0, 1, 21 , 41 , 34 , 18 , . . . , 78 , 16
1
,... [0, 1]
(an ) is an enumeration of Q R
Let us characterize the convergent subsequences of a sequence.
2.15 Lemma: Suppose that (an ) is a sequence and let T denote the set of all limits
of convergent subsequences.
(i) Suppose that T 6= ∅.
(a) If sup T < ∞, then sup T ∈ T and, if inf T > −∞, then inf T ∈ T .
(b) If sup T = ∞, there is a subsequence of (an ) diverging to ∞ and, if inf T =
−∞, there is a subsequence of (an ) diverging to −∞.
(ii) If T = ∅ then no subsequence of (an ) converges and there is a subsequence
diverging to ∞ or to −∞ (it is possible that some subsequences of (an ) diverge
to ∞ and others diverge to −∞).
Proof: (i) Suppose first that T is not empty and that sup T < ∞. Let t = sup T and
let  > 0. Theorem 1.12 implies that T ∩ (t − , t] 6= ∅ and hence there is a convergent
subsequence of (an ) whose limit lies in the interval (t − , t]. Now if (ank ) is such a
subsequence then t −  < ank < t +  for infinitely many values of k and, therefore,
an must lie in the interval (t − , t + ) for infinitely many values of n.
Since this is true for each  > 0 and it follows that, for each positive integer
k, the inequality t − 1/k < an < t + 1/k must hold for infinitely many values of
n. We can use this to build a subsequence whose limit is t as follows. Let n1 be
any positive integer satisfying t − 1 < an1 < t + 1, let n2 be any integer satisfying
n2 > n1 and t − 1/2 < an2 < t + 1/2, let n3 be any integer satisfying n3 > n2
and t − 1/3 < an3 < t + 1/3, etc. This construction gives a subsequence (ank ) that
converges to t and therefore t ∈ T .
This proves the assertion in (i a) dealing with the case sup T < ∞, and the proof
of the one dealing with the case inf T > −∞ is similar and will be omitted.
Now suppose that T is not empty and that sup T = ∞. Then, for any positive
integer p, there is a convergent subsequence of (an ) whose limit lies in the interval
(p, ∞); if q is any other integer, there will be an integer n > q such that an > p. This
means that, given any two integers p and q, there is an integer n such that n > q and
an > p. We can use this observation to build a subsequence of (an ) that diverges to
∞ as follows. Let n1 be any positive integer satisfying 1 < an1 , let n2 be any integer
32 2 Sequences of real numbers

satisfying n2 > n1 and 2 < an2 , let n3 be any integer satisfying n3 > n2 and 3 < an3 ,
etc.
This proves the assertion in (i b) dealing with the case sup T = ∞, and the proof
of the one dealing with the case inf T = −∞ is similar and will be omitted.
(ii) Now suppose that T = ∅. The definition of T then implies that no subsequence
of (an ) can converge to a real number. Consider a positive integer k and the inequality
−k < an < k. If this inequality holds for infinitely many n, then (an ) would have
a bounded subsequence and this subsequence would, by the Bolzano–Weierstrass
theorem, have a convergent subsequence. But, since a subsequence of a subsequence
is a subsequence, this is impossible by the assumption that T = ∅. This shows that,
for each integer k, the inequality −k < an < k holds for only finitely many integers n.
So, for each integer k, the inequality |an | ≥ k must hold for infinitely many integers
n, and hence at least one of the inequalities an ≥ k and an ≤ −k must hold for
infinitely many values of n. We can use this observation to build a subsequence of
(an ) that diverges to ∞ or to −∞ as follows.
First suppose that, for each positive integer k, the inequality an ≥ k holds for
infinitely many values of n. Let n1 be any integer satisfying an1 ≥ 1, let n2 be any
integer satisfying n2 > n1 and an2 ≥ 2, let n3 be any integer satisfying n3 > n2 and
an3 ≥ 3, etc. Then (ank ) is clearly a subsequence diverging to ∞.
Now suppose that there is a positive integer k such that the inequality an ≥ k holds
for only finitely many values of n. Then, for each positive integer k, the inequality
an ≤ −k must hold for infinitely many values of n, and a similar argument will lead
to a subsequence that diverges to −∞. 
Consider a sequence (an ) and let T be as in the preceding lemma. We want to
associate with this sequence two elements of the set R ∪ {−∞, ∞} to be denoted
by lim inf n→∞ an and lim supn→∞ an with the following desiderata in mind. If the se-
quence converges to a real number, say L, then both lim inf n→∞ an and lim supn→∞ an
are to be equal to L. More generally, lim supn→∞ an is to be the largest element of
the set R ∪ {−∞, ∞} to which a subsequence of (an ) converges or diverges, and
lim inf n→∞ an is to be the smallest element to which a subsequence converges or di-
verges (where, of course, we regard −∞ < x < ∞ for all x ∈ R). The point of the
above lemma is that this really does make sense. For example, if, say, T 6= ∅ and
sup T < ∞, then sup T belongs to T and so is the limit of a convergent subsequence.
In this example, then, lim supn→∞ an will be ∞ if a subsequence of (an ) diverges to
∞ and will be sup T otherwise. As a second example, if T 6= ∅ and sup T = ∞, then
there is a subsequence diverging to ∞ and lim supn→∞ an will be ∞. And as a final
example, if T = ∅, then no subsequence converges to a real number and there are
subsequences diverging to at least one of ±∞; so lim supn→∞ an will be ∞ if there
is a subsequence diverging to ∞ and will be −∞ otherwise. The next definition
makes this intuitive discussion of the definitions of lim supn→∞ an and lim inf n→∞ an
precise.
2.16 Definition: Suppose that (an ) is a sequence and let T be the set of all limits of
 2.2 Subsequences 33

convergent subsequences.
(i) The limit superior of the sequence (an ) is the element lim supn→∞ an of the
set R ∪ {−∞, ∞} defined as follows:
(a) If (an ) is not bounded above, then lim supn→∞ an = ∞.
(b) If (an ) is bounded above and if T 6= ∅, then lim supn→∞ an is the largest
number that is the limit of a convergent subsequence of (an ).
(c) If (an ) is bounded above and if T = ∅, then lim supn→∞ an = −∞.
(ii) The limit inferior of the sequence (an ) is the element lim inf n→∞ an of the
set R ∪ {−∞, ∞} defined as follows:
(a) If (an ) is not bounded below, then lim inf n→∞ an = −∞.
(b) If (an ) is bounded below and if T 6= ∅, then lim inf n→∞ an is the smallest
number that is the limit of a convergent subsequences of (an ).
(c) If (an ) is bounded below and if T = ∅, then lim inf n→∞ an = ∞. •
The idea that the limit superior and limit inferior are in some sense replacements
for the limit is borne out by the following result.
2.17 Proposition: Suppose that (an ) is a sequence.
(i) If the sequence (an ) converges to a number or diverges to ±∞ then
lim inf an = lim an = lim sup an . (2.1 )
n→∞ n→∞ n→∞

(ii) If lim inf n→∞ an = limn→∞ an then either the sequence (an ) converges or di-
verges to ±∞; in either case (2.1) holds.
Proof: It is possible to prove this directly from the above definitions, but it is more
easily deduced from the properties of the limit superior and inferior contained in
Propositions 2.19 and 2.20 (cf. Exercise E2.44). 
2.18 Example: This example is a continuation of Example 2.14.
Sequence lim inf n→∞ an lim supn→∞ an
1, 21 , 13 , . . . 0 0
1, −1, 1, −1, . . . -1 1
1, 2, 3, . . . ∞ ∞
1, 23 , 31 , 54 , 15 , 76 , . . . 0 1
0, 1, 21 , 41 , 34 , 18 , . . . , 78 , 16
1
,... 0 1
(an ) is an enumeration of Q −∞ ∞
The next two propositions contain important properties of the limit superior; the
corresponding properties of the limit inferior can be deduced from these propositions
and the easily verified formula
lim inf an = − lim sup(−an ) (2.2)
n→∞ n→∞

(cf. Exercise E2.35).

34 2 Sequences of real numbers

2.19 Proposition: Let (an ) be a sequence of numbers and let L be a number.

(i) If L = lim supn→∞ an then
(a) for each  > 0 the inequality an > L +  holds for only finitely many n and
(b) for each  > 0 the inequality an > L −  holds for infinitely many n.
(ii) If conditions (i a) and (i b) hold, then L = lim supn→∞ an .
Proof: (i) Suppose that lim supn→∞ an = L. Since the limit superior of (an ) is a real
number, it follows from Definition 2.16 that this sequence is bounded above (by M ,
say), that L is a limit of a convergent subsequence, and that no number larger than
L is the limit of a convergent subsequence. If (i a) did not hold, then, for some  > 0,
we would have
an > L +  for infinitely many n.
But then there would be a subsequence of (an ) all of whose terms belong to the interval
[L + , M ]. By the Bolzano–Weierstrass theorem, the subsequence would itself have a
subsequence converging to a point in [L + , M ]. But then this limit would also be the
limit of a convergent subsequence of (an ), contradicting the definition of L. If (i b)
did not hold, then, for some  > 0, an > L −  for only finitely many n, contradicting
the fact that L is the limit of a convergent subsequence of (an ).
(ii) Now suppose, conversely, that L satisfies (i a) and (i b) and let L0 =
lim supn→∞ an . Consider first the case in which L > L0 , and put  = (L − L0 )/2.
Then the definition of L0 and part (i) implies that the inequality

an > L0 +  = L − 

holds for only finitely many integers n, contradicting the assumption that L satis-
fies (i b). Now consider the case in which L0 > L and put  = (L0 − L)/2. Then (just
as in the first case) the definition of L0 and part (i) implies that the inequality

an > L0 −  = L + 

holds for only infinitely many integers n, contradicting the assumption that L satis-
fies (i b).
This means that both of the assumptions L > L0 and L0 > L lead to contradictions,
and thus L = L0 . 
Definition 2.16 together with Proposition 2.19 give us two different ways of char-
acterizing the limit superior and limit inferior of a sequence. We get a third charac-
terization by considering the limit of the sequence formed from (an ) by taking the
suprema and infima of its tails.
 2.2 Subsequences 35

2.20 Proposition: For any bounded sequence (an ) we have

lim sup an = lim sup{ak | k ≥ n}

n→∞ n→∞

and
lim inf an = lim inf{ak | k ≥ n}.
n→∞ n→∞

Before turning to the proof of this proposition it will be useful to consider an

example to help unravel the notation.
2.21 Example: Let (an ) denote the sequence
1 −4 1 −5 1
−2, , , , , ,...
4 3 16 4 64
By considering subsequences of this sequence, it is clear that lim supn→∞ an = 0 and
that lim inf n→∞ an = −1. Put Mn = sup{ak | k ≥ n} and mn = inf{ak | k ≥ n} for
n ∈ Z>0 . Then
 
1 4 1 1
M1 = sup −2, , − , , . . . =
4 3 16 4
 
1 4 1 1
M2 = sup ,− , ,... =
4 3 16 4
 
4 1 1
M3 = sup − , , . . . =
3 16 16
..
.

and so (Mn ) is the sequence

1 1 1 1
, , , ,...
4 4 16 16
and thus limn→∞ Mn = 0. Similarly, (mn ) is the sequence
4 4 5 5
−2, − , − , − , − . . .
3 3 4 4
and thus limn→∞ mn = −1. •
Proof of Proposition 2.20: Suppose that A is a lower bound for {an | n ∈ Z>0 } and
let Mn = sup{ak | k ≥ n} for each n ∈ Z>0 . Since {ak | k ≥ n} ⊇ {ak | k ≥ n + 1},
it follows from Proposition 1.11 that Mn+1 ≤ Mn . Hence (Mn ) is a nonincreasing
sequence, and, since A ≤ ak ≤ Mk for all k ≥ 1, the sequence (Mn ) is bounded
below by A. It now follows from Theorem 2.9 that limn→∞ Mn exists and that, if
L = limn→∞ Mn , then L = inf{Mn | n ∈ Z>0 }. To show that L = lim supn→∞ an it
is only necessary to show that L satisfies the two condition in Proposition 2.19.
Let  be a positive number. It follows from Theorem 1.12 that there is an n ∈ Z>0
such that Mn ≤ L + . But then ak ≤ L +  for all k ≥ n, and hence the inequality
36 2 Sequences of real numbers

ak > L +  holds for only finitely many k. Now suppose that the inequality ak > L − 
holds for only finitely many k. Then there would exist n ∈ Z>0 such that ak ≤ L− for
all k ≥ n. But then Mn ≤ L − , contradicting the fact that L = inf{Mn | n ∈ Z>0 }.
So L satisfies conditions (i a) and (i b) of Proposition 2.19 and so equals lim supn→∞ an .
The assertion dealing with the limit inferior can be proven in a similar manner or
can be deduced from what was just proven and equation (2.2). 

2.3 Cauchy sequences

The definition of the limit of a sequence can be used to verify whether or not a
given number is the limit of a given sequence. But this definition cannot be used
directly in answering the question of whether or not a given sequence converges. Now
whether a sequence converges clearly depends only on the sequence itself and, at least
in principle, we should be able to decide this solely by examining the sequence itself
(and without looking outside the sequence at candidates for its limit). The purpose
of this section is to identify a property that a given sequence may or may not satisfy
and that is both intrinsic to the sequence and equivalent to its converging.
Consider a convergent sequence (an ) and let L = limn→∞ an and let  be a a
positive number. Then there is an integer N such that |an − L| < /2 for all n > N .
Now
|am − an | ≤ |am − L| + |L − an |
for any two integers m and n by the triangle inequality, and, therefore, |am − an | < 
for any two integers m, n ≥ N . This turns out to be the desired condition.
2.22 Definition: A sequence (an ) is said to be a Cauchy sequence if, for every
 > 0, there is an integer N such that |an − am | <  for all n, m ≥ N . •
The following result now shows the importance of the notion of a Cauchy se-
quence.
2.23 Theorem: A sequence of real numbers is convergent if and only if it is a Cauchy
sequence.
Proof: The discussion preceding the statement of the theorem shows that every con-
vergent sequence is a Cauchy sequence. Now suppose that, conversely, (an ) is a
Cauchy sequence. Taking  = 1 in Definition 2.22 shows that there is an integer N
such that |an − am | < 1 for all m, n ≥ N . This means that all but a finite number
of the terms of the sequence (an ) are contained in the interval (aN − 1, aN + 1) and,
therefore, (an ) itself must be a bounded sequence. The Bolzano–Weierstrass theorem
(Theorem 2.13) now implies that (an ) has a convergent subsequence; let (ank ) be such
a subsequence and put L = limk→∞ ank . The rest of the proof consists of showing
that L = limn→∞ an .
Notice that

|an − L| = |an − ank + ank − L| ≤ |an − ank | + |ank − L|

 2.3 Cauchy sequences 37

for any two positive integers k and n. If  is some positive number, then, since
L = limk→∞ ank , it follows that there will be an integer N1 such that |ank − L| < /2
whenever k ≥ N1 . Next, the assumption that (an ) is a Cauchy sequence implies
that there is an integer N2 such that |an − am | < /2 whenever m, n ≥ N2 . Now let
N = max{N1 , N2 } and recall that nk ≥ k for all k by the definition of a subsequence.
So, if n ≥ N , then
|an − L| ≤ |an − anN | + |anN − L| < ,
and this shows that L = limn→∞ an . 

2.24 Example: We can use the preceding theorem to show that the sequence whose
nth term is
1 1 1 1
an = 1 − + − + · · · + (−1)n−1
2 3 4 n
is convergent. To do this, we must examine the expression |am − an | and, in doing
so, we may as well assume that m > n. The fact that n1 − n+1 1
> 0 for any positive
integer n implies that, if m − n is even, then
 
 n 1 n+1 1 m+1 1 
|am − an | = (−1) + (−1) + · · · + (−1) 
n+1 n+2 m
 
 1 1 1 1 1 1 
=  − + − + ··· + − 
n+1 n+2 n+3 n+4 m−1 m
1 1 1 1 1 1
= − + − + ··· + −
n+1  n + 2 n + 3 n + 4  m−1 m 
1 1 1 1 1 1
= − − − ··· − − −
n+1 n+2 n+3 m−2 m−1 m
1
<
n+1
whereas, if m − n is odd, then
 
 1 1 1 1 1 
|am − an | =  − + ··· − + 
n+1 n+2 m−2 m−1 m
   
1 1 1 1 1
= − − − ··· − −
n+1 n+2 n+3 m−1 m
1
< .
n+1
Now suppose that an  > 0 is given and let N be an integer satisfying N + 1 > 1 .
1
Then |an − am | ≤ n+1 <  whenever m ≥ n ≥ N and, therefore, (an ) is a Cauchy,
and hence a convergent, sequence.
The value of limn→∞ an will be determined in Example 5.13–1 to be ln 2. •
38 2 Sequences of real numbers

2.25 Example: Let (an ) be as in Example 2.24. Recall from this example that the
limit limn→∞ an exists and that, for any integer m ≥ 5, we have |a4 − am | < 1/5. So,
if L = limn→∞ an , then |a4 − L| ≤ 1/5, and since
1 1 1 7
a4 = 1 − + − = ,
2 3 4 12
7
we can conclude that | 12 − L| < 14 . This is consistent with the assertion made in the
previous example that L = ln 2. •
 Exercises for Chapter 2 39

Exercises
E2.1 Determine limn→∞ an if:
n+1
(a) an =
3n
n3 + n
(b) an = 2
n + n

(c) an = 1 − 21 1 − 13 1 − 14 · · · 1 − n+1 1

1 2 n
(d) an = 2 + 2 + · · · 2
n n n
E2.2 Find limn→∞ an in the following cases:
n2
(a) an = , n ≥ 1.
n + 1000
1 1 1 1
(b) an = 1·2 + 2·3 + 3·4 + · · · + n(n+1) , n ≥ 1.
In each case, you must give a proof using the definition of limn→∞ an .
1
Hint: For (b), express k(k+1) as the difference of two fractions.
E2.3 Using the definition of a limit, find limn→∞ an , where
( 2
n +1
sin(n2 +1)
, if n < 21000 ,
an = sin(n 2 +1)

n2 +1
, if n ≥ 21000 .
n
E2.4 It is easy to see that limn→∞ n+1 = 1. For each number  > 0 find an integer
N such that  
 n 
 n + 1 − 1 < 
 

for all n ≥ N .
E2.5 Prove, using the definition of convergence, that
n−1
lim = 1.
n→∞ n + 1
√ √
E2.6 Define a sequence (an ) by the two conditions a1 = 2 and an+1 = 2 + an for
n ≥ 1. Show that this sequence converges and find its limit.
Hint: Show that the sequence is monotone and bounded.

n
E2.7 Assuming that limn→∞ 1 + n1 = e, compute:
1 n


(a) limn→∞ 1 + 3n ;
n 2
(b) limn→∞ 1 − n12 .

E2.8 Show that the sequence (nn / (n!en )) has a limit.

E2.9 Recall that nk = k!(n−k)! . Let an = 2n

n!

−n
n
4 , for all n ≥ 1.
(a) Prove that the sequence (an ) converges to a limit.
(b) Find limn→∞ an .
40 2 Sequences of real numbers

E2.10 Show that, if

1 1 1
an = 1 + 1 + + + ··· + ,
2! 3! n!
then limn→∞ an exists by showing that the sequence is Cauchy.
E2.11 Suppose that 0 ≤ b < 1 and let an = (1 + b)(1 + b2 ) · · · (1 + bn ) for n ≥ 1.
Show that limn→∞ an exists.
Hint: Show that 1 + b < eb .
E2.12 Let (an ) and (bn ) be two nondecreasing sequences with the property that, for
each positive integer n, there are integers p and q such that an ≤ bp and
bn ≤ aq . Show that (an ) and (bn ) either both converge or both diverge to ∞
and that, moreover, if they both converge they have the same limit.
E2.13 Prove the uniqueness of limits: If (xn ) is a sequence in R such that limn→∞ xn =
a and limn→∞ xn = b then a = b. Include the cases where the sequence may
diverge to ±∞.
E2.14 Suppose that S is a nonempty subset of R that is bounded above and put
s = sup S. Show that there is a sequence (xn ) such that xn ∈ S for all n and
limn→∞ xn = s.
E2.15 Show that, if a sequence (xn ) satisfies |xn − xn+1 | < 3−n for all n, then it is
convergent.
E2.16 Let (xn ) be a Cauchy sequence and suppose that, for every  > 0, there is an
integer n satisfying n > 1/ and |xn | < . Prove that limn→∞ xn = 0.
E2.17 Show that any sequence (an ) satisfying |an −an+1 | ≤ |an−1 −an |/2 for all n ≥ 2
is convergent.
E2.18 Suppose that (an ) and (bn ) are two sequences and assume that |am − an | ≤
|bm −bn | for all integers m and n. Show that, if the sequence (bn ) is convergent,
then so is the sequence (an ).
E2.19 Suppose that (an ) is a sequence of numbers and that a is a number.
(a) Show by means of examples that, if limn→∞ |an | = |a|, then the limit
limn→∞ an need not exist and that, if it does exist, it need not equal |a|.
(b) Prove that limn→∞ |an | = 0 if and only if limn→∞ an = 0.
E2.20 Suppose that (an ) and (bn ) are two sequences of numbers.
(a) Prove that, if limn→∞ an = 0 and if (bn ) is bounded, then limn→∞ an bn =
0.
(b) Show by example that, in part (a), it is necessary that the sequence (bn )
be bounded.
(c) Prove that, if limn→∞ an = A and if limn→∞ bn = B, then limn→∞ an bn =
AB.
Hint: Subtract and add the same quantity to an bn − AB and use (a).
E2.21 Let (an ) and (bn ) be convergent sequences of real numbers such that
limn→∞ an = A and limn→∞ bn = B, for some A, B ∈ R. Show (using the
-N definition of a limit) that
 Exercises for Chapter 2 41

(a) limn→∞ an + bn = A + B
(b) limn→∞ an bn = AB
Hint: For part (b), see Exercise E2.20.
E2.22 Show that, if (an ) and (bn ) are two sequences such that an ≤ bn for every
n ∈ Z>0 and limn→∞ an = ∞, then limn→∞ bn = ∞.
E2.23 Show that, if (an ) is a sequence such that limn→∞ an = ∞, then limn→∞ a1n = 0.
E2.24 Suppose that limn→∞ an = a, with a 6= 0. Prove that limn→∞ 1/an = 1/a.
√
E2.25 (a) Show that xy ≤ (x + y)/2 for any two nonnegative numbers x and y.
(The left and right sides of this inequality are the geometric and arithmetic
means of x and y, respectively.)
(b) Let x1 and y1 be two numbers satisfying 0 < x1 < y1 and put xn+1 =
√
xn yn and yn+1 = (xn + yn ) /2 for n ≥ 1. Prove that the sequences (xn )
and (yn ) are convergent and that they have the same limit.
E2.26 Let (an ) be a sequence and define a sequence (bn ) by putting
a1 + a2 + · · · + an
bn =
n
for n ≥ 1.
(a) Show that, if limn→∞ an = 0, then limn→∞ bn = 0.
(b) Show that, if limn→∞ an = L, then limn→∞ bn = L, where L is a real
number. (This result is important in the study of probability theory and
Fourier series.)
(c) Give an example for which (an ) diverges but (bn ) converges.
E2.27 (a) Use induction to prove Bernoulli’s inequality: (1 + h)n > 1 + nh or all
integers n > 1 and all numbers h > 0.
(b) Use this inequality to prove that if r > 1 then limn→∞ rn = ∞.
(c) Prove that if |r| < 1 then limn→∞ rn = 0.
E2.28 In this exercise, we will prove the following statement, known as Fekete’s
Lemma:
Let a1 , a2 , a3 , . . . be a sequence of nonnegative real numbers with the

property that am+n ≤ am + an for all m, n. Then, the sequence ann
converges to inf{ ann | n ≥ 1}.
So, let (an ) be a sequence as in the hypothesis of the above statement. The
set { ann | n ≥ 1} is bounded below by 0, so its infimum exists. Set L =
inf{ ann | n ≥ 1}.
(a) For positive integers k, m, prove that akm ≤ mak .
(b) Show that, for any  > 0, there exists N such that, for all n ≥ N ,
an
n
≤ L + .
Hint: Let  > 0 be arbitrary. Pick a k such that akk < L + /2 (why
can this be done?). Now, any n can be expressed as n = km + r, with
42 2 Sequences of real numbers

0 ≤ r < k (why?). Part (a) shows that an ≤ mak + ar . Deduce from this
that there exists N such that, for all n > N , ann < akk + /2.
(c) Deduce from (b) that limn→∞ ann = L.
E2.29 Suppose that (ank ) is a subsequence of a monotone sequence (an ). Is it true
that, if the limit limk→∞ ank exists, then so does limn→∞ an and that these
limits are equal? Why?
E2.30 Suppose that f is a continuous function whose domain is the real line and that
(an ) is a convergent sequence. Show that limn→∞ f (an ) = f (limn→∞ an ).
E2.31 A sequence (an ) of real numbers is said to be contractive if there exists a
constant β with 0 < β < 1 such that

|an+1 − an | ≤ β, |an − an−1 |

for all n ≥ 1. Show that any contractive sequence is a Cauchy sequence, and,
therefore, is convergent.
Hint: Obtain an estimate for |an − am | in terms of β and |a1 − a0 |.
E2.32 A function f : R → R is said to be a contraction on R if there exists a
constant β with 0 < β < 1 such that

|f (x) − f (y)| ≤ β|x − y|

for all x, y ∈ R. In this problem, we will prove the following contraction

mapping theorem:
A contraction on R is continuous and has a unique fixed point.
(A fixed point of a function f : R → R is a point x ∈ R such that f (x) = x.)
Let f : R → R be a contraction on R.
(a) Show that f is continuous.
(b) Show that f can have at most one fixed point.
Hint: Suppose there were two.
(c) Let x0 be any point in R. Let x1 = f (x0 ), x2 = f (x1 ), and in general, for
n ≥ 1, xn = f (xn−1 ). Show that (xn ) is a convergent sequence.
Hint: Exercise E2.31.
(d) Show that x = limn→∞ xn is a fixed point of f .
E2.33 Determine the limit superior lim supn→∞ an and the limit inferior lim inf n→∞ an
if:
(a) an = 3 + (−1)n (1 + 1/n);
(b) an = 1 + sin(nπ/2);
(c) an = (2 − 1/n)(−1)n ;
n
(d) an = −n/4+[n/4]+(−1)  , where 
[·] denotes the greatest integer function;
n  1 2n + 1
(e) an = cos π + sin π .
2 n 2
 Exercises for Chapter 2 43

E2.34 Suppose that (an ) is a sequence such that L = lim supn→∞ an is a real number.
Then we know that, for any number M > L, there are only finitely many
integers n for which an > M . Show by means of an example that it is possible
to have an > L for infinitely many n.
E2.35 Prove equation (2.2).
E2.36 Suppose that (an ) and (bn ) are two bounded sequences.
(a) Show that, if an ≤ bn for all n then lim supn→∞ an ≤ lim supn→∞ bn .
(b) Show that
lim sup (an + bn ) ≤ lim sup an + lim sup bn .
n→∞ n→∞ n→∞
(c) Give an example for which equality holds in (a) and an example for which
strict inequality holds.
E2.37 Can you find two distinct sequences (an ) and (bn ) such that each is a subse-
quence of the other? If yes, give an example; if no, give a proof.
E2.38 Find the limit superior, the limit inferior, and the set of limits of convergent
subsequences of the sequence 1, 2, 1, 4, 1, 6, 1, 8, . . ..
E2.39 For each of the sequences (an ) below, identify the set of all limits of convergent
subsequences. Also, determine lim inf n→∞ an and lim supn→∞ an .
n!
(a) an = n , for n ≥ 1
n
(−1)n
 
n
(b) an = (−1) 1 + , for n ≥ 1
n
(c) 1, 1, 2, 1, 2, 3, 1, 2, 3, 4, 1, 2, 3, 4, 5, 1, 2, . . . (The first term is 1, the next two
terms are 1, 2, the next three terms are 1, 2, 3, and so on.)
(d) 21 , 31 , 23 , 14 , 42 , 34 , 15 , 25 , 53 , 45 , . . . (The first term is 12 , the next two terms are
1 2
, , the next three terms are 14 , 42 , 34 , and so on.)
3 3
E2.40 For each of the sequences (an ) below, identify the set of all limits of convergent
subsequences. Also, determine lim inf n→∞ an and lim supn→∞ an .
 
1
(a) an = 2 − (−1)n , for all n ≥ 1
n
1 1 1
(b) an = 1 + + + · · · + , for all n ≥ 1
2 3 n
(c) 1, 1, 12 , 1, 12 , 13 , 1, 21 , 13 , 14 , 1, 12 , 13 , 14 , 15 , . . . (The first term is 1, the next two
terms are 1, 12 , the next three terms are 1, 21 , 13 , and so on.)
E2.41 Let 
−n,
 n ≡ 0 mod 3,
an = 1 + 1/n, n ≡ 1 mod 3,

−1 − 1/n, n ≡ 2 mod 3


for n ∈ Z>0 .
(a) Show that (an ) is bounded above, is not bounded below, and that the set
of limits of convergent subsequences is {−1, 1}.
44 2 Sequences of real numbers

(b) Prove that lim supn→∞ an = 1 and lim inf n→∞ an = −∞.
E2.42 (a) Suppose that (an ) and (bn ) are two bounded sequences. Show that
lim supn→∞ (an + bn ) ≤ lim supn→∞ an + lim supn→∞ bn .
(b) Give an example in which the inequality above is strict, and another in
which the inequality holds with equality.
E2.43 Prove that, if a sequence diverges to ∞, then it has no convergent subsequences.
E2.44 Prove that, if L is a real number and that (an ) is a sequence, then limn→∞ an =
L if and only if lim inf n→∞ an = L and lim supn→∞ an = L (cf. Proposi-
tion 2.17).
 Chapter 3
Sequences in Rd

Sequences and their limits can be defined in Rd for d ≥ 2 almost exactly as they
were defined in R. (It will be convenient to use d for the dimension of the Euclidean
space because of the ubiquitous use of n as a subscript.) As is customary in linear
algebra, if x is a vector in Rd then xk denotes the kth-component of x for k = 1, . . . , d,
so that x = (x1 , . . . , xd ).
The norm of a vector x ∈ Rd is the number
q
kxk = x21 + · · · + x2d .

Notice that, if d = 1, then the norm is just the absolute value. The norm has the
following properties:
1. kxk ≥ 0 for all x ∈ Rd with equality if only if x = 0;
2. kaxk = |a|kxk for all a ∈ R and x ∈ Rd ;
3. kx + yk ≤ kxk + kyk for all x, y ∈ Rd .
The first two of these properties are obvious, and the third, called the triangle
inequality , is most easily verified by first proving the Cauchy–Schwartz inequality

|x · y| ≤ kxkkyk,

which is valid for all x, y ∈ Rd . Both this inequality and property 3 above should be
familiar from linear algebra.
If (xn ) is a sequence in Rd the kth-component of xn is denoted by xn,k and so xn
itself is (xn,1 , . . . , xn,d ).
3.1 Definition: A sequence (xn ) in Rd is said to have a limit L, or to converge to
L if, for any  > 0, there exists an integer N such that

kxn − Lk < 

for every n ≥ N , and in this case we will write limn→∞ xn = L. •

The following theorem says that we find limits “component-wise” and means that
many of the results for sequences in R can be carried over effortlessly to Rd .
46 3 Sequences in Rd

3.2 Theorem: If (xn ) is a sequence in Rd and if L is a point in Rd , then

limn→∞ xn = L if and only if limn→∞ xn,i = Li for i ∈ {1, . . . , d}.
Proof: The theorem follows almost immediately from the identity
q
kxn − Lk = (xn,1 − L1 )2 + · · · + (xn,d − Ld )2 .

√ If limn→∞ xn,i = Li for i ∈ {1, . . . , d}, then there

Indeed, suppose that  > 0 is given.
is an N such that |xn,i − Li | < / d for all n ≥ N and for i ∈ {1, . . . , d}. But then
it follows that p
kxn − Lk < 2 /d + · · · + 2 /d = 
for all n ≥ N . Conversely, if limn→∞ xn = L, there is an N such that kxn − Lk < 
for all n ≥ N , and then
q
|xn,i − Li | ≤ (xn,1 − L1 )2 + · · · + (xn,d − Ld )2 < 

for all i ∈ {1, . . . , d} and all n ≥ N . 

3.3 Definition: A sequence (xn ) in Rd is said to be bounded if there exists a number

M > 0 such that kxn k ≤ M for all n ∈ Z>0 . •
Since every ball in Rd lies inside some cube and every cube lies inside some ball,
boundedness can also be defined by requiring that, for each i ∈ {1, . . . , d}, there exist
numbers mi and Mi such that mi ≤ xi ≤ Mi for all n ∈ Z>0 . It is left as an exercise
to prove that the two definitions are equivalent (cf. Exercise E3.1).
If d ≥ 2, there is no useful definition of “greater than” in Rd . (What would it
mean to say that x < y in R2 , let alone Rd ?) Hence, there can be no results for
Rd analogous to those for R which were phrased in terms of the order properties
of R, such as Theorem 2.9. Nevertheless, it is possible to generalize many of the
convergence theorems for R to Rd by working component-wise.
Subsequences of sequences in Rd can be defined just as they were for sequences in
R and a Bolzano–Weierstrass theorem for sequences in Rd can be proven by repeated
applications of the theorem for R.
3.4 Theorem: (Bolzano–Weierstrass theorem) Every bounded sequence in Rd
has a convergent subsequence.
Proof: Suppose that (xn ) is a bounded sequence in Rd . Then it is clear that
(xn,i )n∈Z>0 is a bounded sequence in R for i ∈ {1, . . . , d}. We cannot simply ap-
ply the Bolzano–Weierstrass theorem (Theorem 2.13) to each of the d components
of the sequence (xn ) since the resulting d convergent subsequences may be indexed
on completely different subsets of Z>0 . Instead, we apply the Bolzano–Weierstrass
theorem to the sequence (xn,1 ) to find a convergent subsequence (xnk ,1 ). Next, we
apply the Bolzano–Weierstrass theorem to the sequence (xnk ,2 ) to find a convergent
subsequence of it and hence a convergent subsubsequence of the original sequence.
This idea works but both the notation and the terminology is rapidly getting out of
 3 Sequences in Rd 47

hand, and it is necessary to change notation and replace the original sequence (xn )
by the subsequence (xnk ). Now the sequence (xn,1 ) converges, and we can apply the
Bolzano–Weierstrass theorem to the sequence (xn,2 ) to find a convergent subsequence,
say, (xnk ,2 ). The sequence (xnk ,1 ) still converges by Lemma 2.12, and the sequence
(xnk ) is a subsequence of the original sequence. Let us now change notation once
again, replacing the sequence (xn ) by the subsequence (xnk ); this gives a sequence
which is a subsequence of the original sequence and with the property that both of
the sequences (xn,1 ) and (xn,2 ) converge. We can now repeat the argument, applying
the Bolzano–Weierstrass theorem to the sequence (xn,3 ).
After carrying out this argument d times, we will get a sequence (xn ) which is a
subsequence of the original sequence, and such that the sequence (xn,i ) converges for
i ∈ {1, . . . , d}, and hence, by Theorem 3.2, such that limn→∞ xn exists. 
The notion of Cauchy sequence in Rd can be made entirely analogously to the same
notion in R. Moreover, it is still true that a sequence in Rd converges if and only if
it is a Cauchy sequence (cf. Exercise E3.3). The fact that this sort of equivalence is
not true for infinite-dimensional R-vector spaces with norms is a launching point for
a great deal of interesting and useful analysis.
A contraction on Rd is a function f from Rd to Rd such that there is a number
K < 1 with the property that

|f (u) − f (v)| ≤ Kku − vk

for all u, v ∈ Rd . The important property of contractions is they have unique fixed
points, i.e., if f is a contraction on Rd , there is exactly one point x ∈ Rd with the
property that f (x) = x (cf. the following theorem). This fact (albeit in a more
general context) can be used to prove a number of important theorems, such as the
differentiability of the inverse of a one-to-one differentiable function or the existence
of solutions of ordinary differential equations.
3.5 Theorem: (Contraction mapping theorem) A contraction on Rd is contin-
uous and has a unique fixed point.
Proof: The proof will be left as an exercise (cf. Exercise E3.9). 
48 3 Sequences in Rd

Exercises
E3.1 Prove the equivalence of the two possible definitions of boundedness in Rd .
E3.2 Find the limits, if they exist, of these sequences in R2 :
 
n 1
(a) (−1) , ;
n
 
1
(b) 1, ;
n
     
1 1 π 
(c) (cos(nπ)), sin nπ + ;
n n 2
 
1 −n
(d) ,n .
n
E3.3 (a) Give a definition of a Cauchy sequence in Rd .
(b) Show that a sequence (xn ) in Rd is Cauchy if and only if (xn,k ) is a Cauchy
sequence of real numbers for k = 1, . . . , d.
(c) Prove that every convergent sequence in Rd is Cauchy.
(d) Prove that every Cauchy sequence in Rd converges.
E3.4 Suppose that (xn ) is a Cauchy sequence of points in Rd .
(a) Prove that (xn ) is bounded.
(b) Prove that, if (an ) is a sequence of real numbers such that limn→∞ an = 0
then ((1 + an )xn ) is a Cauchy sequence in Rd .
E3.5 (a) Prove that every subset of R that is not bounded above contains a se-
quence that diverges to ∞.
(b) Prove that every unbounded subset of Rd contains a sequence (xn ) with
the property that limn→∞ kxn k = ∞.
E3.6 Let S = {(x, y) ∈ R2 | xy > 1} and B = {k(x, y)k | (x, y) ∈ S}. Find inf B.
E3.7 For n ≥ 1, let an = (−1)n , and let
(
1, if n < 10100 ,
bn =
1/n otherwise.

Let (xn ) be the sequence in R2 defined by xn = (an , bn ). Determine the set of

all limits of convergent subsequences of (xn ).
E3.8 Suppose that S is a subset of Rd , that y is a point in S, and that f is a
real-valued function with domain S. Show that the following two statements
are equivalent:
(a) limx→y f (x) = f (y);
(b) limn→∞ f (xn ) = f (y) whenever (xn ) is a sequence in S such that
limn→∞ xn = y.
 Exercises for Chapter 3 49

E3.9 Suppose that f is a contraction on Rd and let K denote the associated constant
(so that 0 < K < 1). Prove Theorem 3.5 as follows. Let x1 be any point in
Rd and let x2 = f (x1 ), x3 = f (x2 ), . . ..
(a) Prove that f is continuous.
(b) Prove that f has at most one fixed point.
Hint: Suppose there were two.
(c) Obtain an estimate for kxm+1 − xm k in terms of K and kx2 − x1 k for any
integer m.
(d) Obtain an estimate for kxn − xm k in terms of K and kx2 − x1 k for any
two integers m, n.
(e) Show that (xn ) is a Cauchy sequence in Rd .
(f) Show that the vector x = limn→∞ xn is a fixed point of f .
E3.10 For each number α put fα (x) = αx for all x ∈ R.
(a) For which numbers α is the function fα a contraction on R?
(b) For those numbers α for which fα is a contraction, what is the fixed point
of fα ?
E3.11 For what intervals [0, r], r ≤ 1, is

f : [0, r] → [0, r]
x 7→ x2

contraction?
 Chapter 4
Some topology in Rd

Topology is, roughly speaking, that part of mathematics that studies open and
closed sets and continuous functions, and the purpose of this section is to give an
introduction to these ideas and some related ones in the context of Euclidean space.
In particular, we shall prove that a continuous function defined on a closed and
bounded set has a bounded range and attains both a maximum and a minimum
value. This result should be familiar from first-year calculus, where it is often stated
but rarely proven.

4.1 Open and closed sets

The properties of the norm (from the beginning of Chapter 3) make it reasonable
to regard kx − yk as the distance between two points x, y ∈ Rd . In fact, if d = 1 this
is just the usual distance and if d ∈ {2, 3} it is also the usual “physical” distance by
Pythagoras’s theorem.
For each point x ∈ Rd and each r > 0, put

Br (x) = {y ∈ Rd | ky − xk < r},

B r (x) = {y ∈ Rd | ky − xk ≤ r},
br (x) = {y ∈ Rd | 0 < ky − xk < r}.
B

These are the open ball of radius r and center x, the closed ball of radius r and
center x, and the punctured ball of radius r and center x, respectively.
If d = 1, then Br (x) and B r (x) are just the familiar open and closed intervals
(x − r, x + r) and [x − r, x + r], respectively, and Bbr (x) = (x − r, x) ∪ (x, x + r).
If d = 2, then Br (x) and B r (x) are disks in R2 with center at x and radius r,
with the former not including its rim and the latter including its rim. The set B br (x)
is the “punctured” disk with center x and radius r that includes neither its center
nor its rim.
If d = 3, the three sets Br (x), B r (x), and B
br (x) are balls (in the familiar sense)
with center x and radius r.
Balls will be essential to phrase many of the notions we will introduce to charac-
terize subsets of Rd . We begin this process now.
 4.1 Open and closed sets 51

4.1 Definition: For any subset A of Rd , let A{ denote the complement of A in

Rd , i.e.,
A{ = Rd \ A = {x ∈ Rd | x 6∈ A}. •
Notice that, for any two subsets A and B of Rd , we have

A ⊆ B if and only if B { ⊆ A{ .

4.2 Definition: A subset A ⊆ Rd is said to be open if, for each point x ∈ A, there
is an r > 0 such that Br (x) ⊆ A, and to be closed if A{ is open. •
Open and closed sets are important building blocks for the structure of Rd . It is
obvious that Rd is an open set and hence that the empty set is a closed set. The empty
set is also an open subset of Rd , and to see this it is best to argue by contradiction.
Indeed, if a subset A of Rd is not open then, by definition, it must contain a point x
with the property that Br (x) 6⊆ A for all numbers r > 0. But the empty set does not
have this property since it does not contain any points. The fact that the empty set
is open now implies that Rd is closed. So the empty set and Rd are both open and
closed subsets of Rd , and it is true (but not so easy to prove) that these are the only
subsets of Rd that are both open and closed.
Points in open sets have a useful property.
4.3 Definition: A point x ∈ Rd is said to be an interior point of a subset A ⊆ Rd
if there is an r > 0 such that Br (x) ⊆ A. By int(A) we denote the set of interior
points of A, which is the interior of A. •
Sometimes the notation A◦ is used for the interior.
4.4 Proposition: A set A is open if and only if every point of A is an interior point
of A. Equivalently, A is open if and only if int(A) = A.
Proof: This follows immediately from Definitions 4.2 and 4.3. 
Informally speaking, we can say that a set is finite if it is empty, or contains 1
point, or contains 2 points, or . . . and is infinite if it is not finite. More formally, a
set is usually said to be finite if every one-to-one function from the set into itself is
onto and to be infinite if there is a one-to-one function from the set into itself that is
not onto. For example, the function that multiplies each integer by 2 is a one-to-one
function from Z>0 into itself that is not onto, and so Z>0 is infinite. It is important
to not confuse the adjectives “finite” and “bounded”: Every finite set is bounded but
a bounded set need not be finite.
Let us consider some examples of open and closed sets.
4.5 Examples: 1. Every nonempty finite subset of Rd is closed and not open. Indeed,
suppose that A is a nonempty finite subset of Rd . To show that A is closed it is
necessary to show that A{ is open, so consider a point x ∈ A{ . Then the number

r = min{kx − yk | y ∈ A}
52 4 Some topology in Rd

is positive since A is finite and x ∈/ A, and it is clear that Br (x) ⊆ A{ . Finally, to

see that A is not open, it is only necessary to observe that every nonempty open
set is infinite since it contains a ball and since every ball is infinite.
2. It is evident that the graph of, say, y = x or = x2 is a closed subset of R2 and that
the graph of, say, z = 2x − y or z = x2 + y 2 is a closed subset of R3 . However,
the graph of y = sin(1/x) is not a closed subset of R2 because any point of the
form (0, a) with |a| ≤ 1 is not on the graph and is not an interior point of the
complement of the graph.
3. For any x ∈ Rd and any r > 0, the sets Br (x) and B br (x) are open but not closed
subsets of R , and the set B r (x) is a closed but not an open subset of Rd . To see
d

that Br (x) is an open set, consider a point y ∈ Br (x) and put s = r − ky − xk.
Then s > 0 by the definition of Br (x) and we can show that Bs (y) ⊆ Br (x) as
follows. If z ∈ Bs (y), then

kz − xk ≤ kz − yk + ky − xk < s + ky − xk = r,

and hence z ∈ Br (x). This shows that Br (x) is open.

To see that Br (x) is not a closed set, it is necessary to show that its complement
Br (x){ is not open, and for this it is necessary to find a point that belongs to
Br (x){ but which is not an interior point of Br (x){ . Let y be a point satisfying
kx − yk = r, so that y ∈ Br (x){ (in fact, y is on the “boundary” of Br (x),
something we shall make sense of shortly). Intuitively, it seems clear that the ball
Bs (y) for any s > 0 contains points in Br (x), so that Bs (y) 6⊆ Br (x){ as required.
The following argument verifies this intuition. Let s be any number such that
0 < s < r and let z = y + 2rs (x − y) (so that z is a point on the line segment
joining y and x). Then z ∈ Bs (y) since
s s
kz − yk = kx − yk = ,
2r 2
and z 6∈ Br (x){ since
 s s
kz − xk = 1 − kx − yk = r − < r.
2r 2
This proves the two assertions about Br (x), and those dealing with B r (x) and
B
br (x) will be left as exercises. •
It follows from Definition 4.2 that every nonempty open set is a union of open
balls. (Indeed, let A be a nonempty open set and, for Seach point x ∈ A, let rx be
any positive number such that Brx (x) ⊆ A. Then A = x∈A Brx (x).) The converse
of this—that every union of open balls is an open set—is a consequence of the next
result.
 4.1 Open and closed sets 53

4.6 Proposition: (i) The union of any number of open subsets of Rd is open.
(ii) The intersection of any finite number of open subsets of Rd is open.
(iii) The union of any finite number of closed subsets of Rd is closed.
(iv) The intersection of any number of closed subsets of Rd is closed.
Proof: (i) Suppose that a set A is the union of open sets and let x be a point in A.
Then x must belong to one of the open sets, say B, whose union is A. But then
Br (x) ⊆ B for some r > 0 since B is open, and thus Br (x) ⊆ A since B ⊆ A. This
shows that A is open.
(ii) Suppose that A = m
T
i=1 Bi , where each of the sets B1 , . . . , Bm is open, and
let x be a point in A. Then, for each i ∈ {1, . . . , m}, there is an ri > 0 such that
Bri (x) ⊆ Bi . If r = min{r1 , . . . , rm }, then r > 0 and

Br (x) ⊆ Bri (x) ⊆ Bi

for each i ∈ {1, . . . , m}, and hence

m
\
Br (x) ⊆ Bi = A.
i=1

This shows that A is open.

Finally, parts (iii) and (iv) are consequences of parts (i) and (ii) and the following
two identities (which are known as de Morgan’s laws):
!{
[ \
Ai = (Ai ){ ,
i∈I i∈I
!{
\ [
Ai = (Ai ){ . 
i∈I i∈I

The following example will show that one cannot omit the adjective “finite” in
parts (ii) and (iii) of the above proposition.
k−1
4.7 Example: Consider a point x ∈ Rd . For any k ∈ Z>0 it is clear that k
< 1 and
hence that B (k−1)/k (x) ⊆ B1 (x), and, therefore,
∞
[
B (k−1)/k (x) ⊆ B1 (x).
k=1

On the other hand, if y ∈ B1 (x), then there is a k ∈ Z>0 such that kx−yk ≤ (k−1)/k.
Thus y ∈ B (k−1)/k (x) and this gives
∞
[
B (k−1)/k (x) = B1 (x),
k=1
54 4 Some topology in Rd

and so the union of an infinite number of closed sets may well be open and not closed.
A similar analysis will show that
∞
\
B(k+1)/k (x) = B 1 (x),
k=1

and so the intersection of an infinite number of open sets may be closed and not
open. •

4.2 Cluster points, limit points, and boundary points

It will be useful to use sequences in discussing closed sets and continuous func-
tions, and we shall need the “set version” of the Bolzano–Weierstrass Theorem (the
“sequence version” was Theorems 2.13 and 3.4).
4.8 Theorem: (Bolzano–Weierstrass theorem) If A is a bounded infinite subset
of Rd , then there is a point x ∈ Rd with the property that the set A∩Br (x) is infinite
for each r > 0.
Proof: Since A is infinite, there is a sequence (xk ) in A with xk 6= xj whenever k 6= j.
Then (xk ) is a bounded sequence in Rd since A is a bounded set in Rd and so, by
the Bolzano–Weierstrass theorem for sequences, (xk ) has a convergent subsequence
(y j ) = (xkj ). Denote the limit of the convergent subsequence by x. Then, for any
r > 0, there is some integer N such that y j ∈ Br (x) for all j ≥ N . But y j = xkj ∈ A,
so we have y j ∈ A ∩ Br (x) for j ≥ N . Thus A ∩ Br (x) is an infinite set for any r > 0,
and so x is the required point. 
The following definition will be useful in studying closed sets.
4.9 Definition: A point x ∈ Rd is said to be a cluster point of a subset A ⊆ Rd if
A∩B br (x) 6= ∅ for all r > 0; let der(A) denote the set of all cluster points of A, which
is called the derived set of A. •
Sometimes the notation A0 is used for the derived set.
The examples which follow will make it clear that a cluster point of A may belong
to A or to A{ , and that points in A may or may not be cluster points of A.
4.10 Examples: 1. A finite set does not have any cluster points. To see this, suppose
that A is a finite subset of Rd . It is obvious that, if A is empty, then it has no
cluster points, so suppose that A is nonempty and that x is a point in Rd . If

r = min{kx − yk | y ∈ A and y 6= x},

then r > 0 since A is finite, and A ∩ Br (x) is either ∅ or {x}. In either case,
A∩B br (x) = ∅, and thus x is not a cluster point of A.
2. It is easy to see that, for each x ∈ Rd and any r > 0, each of the sets der(Br (x)),
der(B r (x)), and der(B br (x)) is equal to B r (x). •
 4.2 Cluster points, limit points, and boundary points 55

4.11 Definition: A sequence in a set A is a sequence, all of whose terms belong to

the set A. •

4.12 Proposition: If x ∈ Rd and if A is a subset of Rd , the following three condi-

tions are equivalent:
(i) x is a cluster point of A;
(ii) A ∩ Br (x) is infinite for each r > 0.
(iii) there is a sequence (xn ) in A such that xm 6= xn whenever m 6= n and
limn→∞ xn = x.
Proof: (i) =⇒ (iii) Let us assume that (i) is true. Then there is certainly a point
x1 ∈ A ∩ B b1 (x). Now, if r = min{2−1 , kx − x1 k}, then r > 0 and so A ∩ B br (x) 6= ∅
−2
by (i); let x2 be any point in A ∩ B br (x). Now let r = min{2 , kx − x2 k} and repeat
the argument: r > 0 and A ∩ B br (x) 6= ∅ by (i), and let x3 be any point in A ∩ B
br (x).
Continuing in this manner will lead to a sequence (xn ) such that xm 6= xn whenever
m 6= n and kx − xn k < 2−n for all n ∈ Z>0 . So the sequence (xn ) has all of the
required properties.
(iii) =⇒ (ii) Let (xn ) be a sequence with the properties stated in (iii). Then,
given an r > 0, there is an N such that kxn − xk < r for all n ≥ N . But then
{xn | n ≥ N } is an infinite subset of Br (x) ∩ A, and thus Br (x) ∩ A is infinite.
(ii) =⇒ (i) This is obvious. 
We can combine Theorem 4.8 and the preceding proposition to obtain the following
succinct form of the Bolzano–Weierstrass theorem: Every bounded infinite set has a
cluster point.
Now we turn to an idea related to, but different from, cluster points.
4.13 Definition: A point x ∈ Rd is said to be a limit point of a subset A ⊆ Rd if
A ∩ Br (x) 6= ∅ for all r > 0; let cl(A) denote the set of all limit points of A, which is
called the closure of A.
Sometimes the notation A is used for the closure.
It is clear from the definition that der(A) ⊆ cl(A). It is also clear from the
definition that A ⊆ cl(A), a relation that does not hold for the derived set. Let us
illustrate by examples the difference between the two sets.
4.14 Examples: 1. Every point in a finite set is a limit point, merely because every
point in a set is a limit point. Thus, for finite sets, we have der(A) ⊂ cl(A).
2. As with the derived set, the closures cl(Br (x)), cl(B r (x)), and cl(Bbr (x)) are all
equal to B r (x). •
Let us give an alternative characterisations of limit points.
56 4 Some topology in Rd

4.15 Proposition: If x ∈ Rd and if A is a subset of Rd , the following two conditions

are equivalent:
(i) x is a limit point of A;
(ii) there is a sequence (xn ) in A such that limn→∞ xn = x.
Proof: (i) =⇒ (ii) Assume x is a limit point of A. Let n ∈ Z>0 and let xn ∈
A ∩ B1/n (x), this being possible by definition of limit point. The sequence (xn ) is in
A, and converges to x.
(ii) =⇒ (i) Let (xn ) be a sequence in A converging to x. Let r > 0 and let
N ∈ Z>0 be such that kxn − xk < r for n ≥ N , this being possible by definition of
convergence. Now note that xn ∈ A ∩ Br (x). Thus x is a limit point of A. 
To fully characterize the difference between cluster points and boundary points,
the following notion is useful.
4.16 Definition: A point x ∈ Rd is said to be an isolated point of a subset A ⊆ Rd
if there exists r > 0 such that A ∩ Br (x) = {x}. •
We now have the following result.
4.17 Proposition: For a subset A ⊆ Rd , a limit point x ∈ Rd is a cluster point for
A if and only if it is not an isolated point of A.
Proof: We leave this as Exercise E4.23. 
Next we define boundary points.
4.18 Definition: A point x ∈ Rd is said to be a boundary point of a subset A of
Rd if, for each r > 0, the ball Br (x) contains at least one point in A and at least one
point in A{ ; let bd(A) denote the set of boundary points of the set A. •
Sometimes the notation ∂A is used for the boundary.
Note that bd(A) = bd(A{ ) because of the symmetry of A and A{ in the definition
of boundary point. It is obvious that bd(∅) = bd(Rd ) = ∅.
The following characterisation of boundary is useful.
4.19 Proposition: For a subset A ⊆ Rd we have cl(A) = A ∪ bd(A).
Proof: Let x ∈ cl(A). If x ∈ A, then clearly x ∈ A ∪ bd(A). If x ∈ A{ , then
x ∈ A{ ∩ Br (x) for every r > 0. Also, since x ∈ cl(A), A ∩ Br (x) 6= ∅ for every r > 0.
Thus x ∈ bd(A) and so x ∈ A ∪ bd(A).
Next let x ∈ A ∪ bd(A). If x ∈ A, then x ∈ cl(A). If x ∈ bd(A), then
A ∩ Br (x) 6= ∅ for each r > 0, in which case x ∈ cl(A). 

4.3 Properties of closed sets

It is necessary study closed sets in order to better understand cluster points,
boundary points, and continuous functions.
 4.3 Properties of closed sets 57

4.20 Theorem: The following conditions on a subset A of Rd are equivalent:

(i) A is closed;
(ii) der(A) ⊆ A, i.e., A contains all of its cluster points;
(iii) bd(A) ⊆ A, i.e., A contains all of its boundary points;
(iv) cl(A) ⊆ A, i.e., A contains all of its limit points;
(v) if (xn ) is any convergent sequence in A, then limn→∞ xn ∈ A.
Proof: (i) =⇒ (ii) Suppose that A is closed and consider a point x ∈ A{ . Since
A{ is open, there is an r > 0 such that Br (x) ⊆ A{ , and thus x 6∈ der(A). This
argument shows that A{ ⊆ der(A){ , and, therefore, der(A) ⊆ A (cf. the remark
following Definition 4.1).
(ii) =⇒ (iii) Suppose that der(A) ⊆ A and consider a point x ∈ A{ . Then
x ∈ der(A){ and so there is an r > 0 such that A ∩ B br (x) = ∅. Now, since x 6∈ A,
this actually means that Br (x) ∩ A = ∅, and, therefore, x 6∈ bd(A). This argument
shows that A{ ⊆ bd(A){ , and, therefore, bd(A) ⊆ A.
(iii) =⇒ (iv) If bd(A) ⊆ A, then cl(A) ⊆ A by Proposition 4.19.
(iv) =⇒ (v) Suppose that A is contains all of its limit points and let (xn ) be a
sequence in A that converges to x ∈ Rd . Thus x is a limit point of A by Proposi-
tion 4.15, and thus x ∈ cl(A) and so x ∈ A.
(v) =⇒ (i) We prove the contrapositive. Suppose that A is not closed or, equiv-
alently, that A{ is not open. Then there is an x ∈ A{ with the property that
Br (x) 6⊆ A{ for every r > 0. This means that, for each k ∈ Z>0 , there will be a
point xk ∈ B1/k (x) ∩ A. But then (xk ) is a sequence in A converging to x and x 6∈ A,
and so (v) does not hold. 
The following theorem gives us, for the first time, something interesting about sets
that are both closed and bounded. Perhaps not obviously, such sets are extremely
important in analysis, and we shall revisit these below in Theorem 4.27.
4.21 Theorem: The following two conditions are equivalent for a subset A ⊆ Rd :
(i) A is closed and bounded;
(ii) every sequence in A has a subsequence that converges to a point in A.
Proof: (i) =⇒ (ii) Suppose first that A is closed and bounded and let (xn ) be a
sequence in A. Then (xn ) is a bounded sequence and so has a convergent subsequence
by the Bolzano–Weierstrass theorem and the limit of this subsequence is in A by
Theorem 4.20.
(ii) =⇒ (i) Now suppose that A satisfies (ii). If A were not bounded, there would
be a sequence (xk ) in A such that kxk k > k for each integer k, and such a sequence
clearly cannot have a convergent subsequence. But this contradicts (ii) and so A must
be bounded. On the other hand, if A were not closed, there would be a boundary
point b of A that is not in A by Theorem 4.20. Then, for each positive integer k, the
ball B1/k (b) would contain a point, say xk , of A and so there would be a sequence
(xk ) in A that converges to b. But every subsequence of this sequence also converges
to b by Lemma 2.12. This too contradicts (ii) and so A must be closed. 
58 4 Some topology in Rd

Closed and bounded subsets of Rd are sufficiently interesting to warrant a special

name.
4.22 Definition: A subset K ⊆ Rd is compact if it is closed and bounded. •
For those of you who grow up and do more advanced analysis, you will see that
“compact” and “closed and bounded” do not agree in more general spaces; but they
do agree in Euclidean space. We shall have more to say about this below, specifically
regarding the characterization of Theorem 4.27 for compact sets.

4.4 Continuous functions

We want to prove a theorem about the ranges of continuous functions that says,
among other things, that a continuous function whose domain is a closed and bounded
interval attains a minimum and a maximum value. As a first step to doing this it will
be useful to formally reconcile the two common ways of thinking about continuity of
functions: in terms of sequences and in terms of  and δ.
4.23 Theorem: For any nonempty set A, any point a ∈ A, and any real-valued
function f with domain A, the following are equivalent:
(i) for each point a ∈ A and each  > 0, there is a δ > 0 such that |f (x)−f (a)| < 
for all x ∈ A ∩ Bδ (a);
(ii) if (an ) is a sequence in A that converges to a, then limn→∞ f (an ) = f (a).
Proof: (i) =⇒ (ii) Suppose that (i) holds and let (an ) be a sequence in A that
converges to a and let  be a positive number. Then (by (i)) there is a δ > 0 such
that |f (x) − f (a)| <  for all x ∈ A ∩ Bδ (a). But, since a = limn→∞ an , there will
be an integer N such that ka − an k < δ for all n ≥ N . So, for any n ≥ N , we have
an ∈ A∩Bδ (a) and hence |f (an )−f (a)| < . As  was an arbitrary positive number,
this shows that limn→∞ f (an ) = f (a) and hence (ii) holds.
(ii) =⇒ (i) We will prove the contrapositive, so suppose that (i) does not hold.
Then there must be an  > 0 such that, for each δ > 0, there is a point x ∈ A ∩ Bδ (a)
satisfying |f (x) − f (a)| ≥ . In particular, for each positive integer n, we can take
δ = 1/n and obtain a point an ∈ A ∩ B1/n (a) satisfying |f (an ) − f (a)| ≥ . But then
limn→∞ an = a, and so we see that (ii) does not hold. 
Now consider a nonempty subset A of Rd and a real-valued function f whose
domain is A. As is well-known from first-year calculus, if the two equivalent conditions
of the previous theorem hold for a point a ∈ A, then f is said to be continuous at a
and, if these conditions hold for every point a ∈ A, then f is said to be continuous.
This definition makes every function continuous at every isolated point in its domain.
This discussion of continuous functions has an obvious analogue for Rm -valued
functions, i.e., for functions whose values are points in Rd . In particular, such a func-
tion is continuous if and only if each of its m real-valued components are continuous
(cf. Exercise E4.33 and the discussion of sequences in Rd in Chapter 3).
 4.5 Compact sets and open coverings 59

We are now ready to prove one of the fundamental theorems about continuous
functions, one that has to do with functions with a compact domain.
4.24 Theorem: Suppose that A is a compact subset of Rd and that f is a continuous
Rm -valued function with domain A.
(i) The image image(f ) = {f (x) | x ∈ A} of f is a compact subset of Rm .
(ii) If m = 1, then f attains a minimum value and a maximum value on
A, i.e., there are points u and v in A such that

f (u) ≤ f (x) ≤ f (v)

for all x ∈ A.
Proof: (i) To show that image(f ) is compact it is only necessary to show that it
satisfies condition (ii) of Theorem 4.21. Thus suppose that (y n ) is a sequence in
image(f ). Then the definition of image(f ) implies that there must be a sequence
(xn ) in A such that f (xn ) = y n for all n ∈ Z>0 . Theorem 4.21 now implies that
there is a subsequence (xnk ) of (xn ) that converges to a point, say x, in A. But then
Theorem 4.23 implies that limk→∞ y nk = limk→∞ f (xnk ) = f (x), and so (y n ) does
indeed have a subsequence that converges to a point in image(f ).
(ii) Now suppose that m = 1. Since image(f ) is compact by (i), both inf image(f )
and sup image(f ) belong to image(f ). If u and v are points in A such that f (u) =
inf image(f ) and f (v) = sup image(f ), then it is clear that f attains a minimum
value at u and a maximum value at v. 
The proof of Theorem 4.24 used the Bolzano–Weierstrass theorem and thus de-
pended on the completeness of R, i.e., the fact that all Cauchy sequences in R con-
verge. The intermediate value theorem is another theorem that is familiar from
first-year calculus and that depends on the completeness of R (cf. Exercise E4.29).

4.5 Compact sets and open coverings

In this section we provide some alternative characterizations of compact sets. In
particular, we shall see that compactness can be described more abstractly in terms
of coverings by open sets; it is this more abstract characterization that carries over
to more general settings than compact subsets of Euclidean space, where “compact”
is no longer equivalent to “closed and bounded.”
4.25 Definition: Suppose that A is a subset of Rd . A set U of subsets of Rd is said
S
(i) to be an open covering of A if each set in U is open and if A ⊆ U ∈ U U , and
(ii) to have a finite subcover if there are U1 , . . . , Un in U such that A ⊆ ni=1 Ui . •
S
60 4 Some topology in Rd

4.26 Example: The set {(1/n, 1) | n ∈ Z>0 } of open intervals is an open covering of
the interval (0, 1) that has no finite subcover. •
The point of introducing open coverings and finite subcovers is that they can be
used to give an third condition equivalent to the two conditions in Theorem 4.21, and
to give another proof of Theorem 4.24. Why is it desirable to do this since we already
know these theorems are true? The reason is that, in more advanced analysis, it is
sometimes necessary, when proving properties of closed and bounded sets, to think of
them in terms of open coverings and finite subcovers. There are also many situations
in analysis and topology where the three conditions are no longer equivalent and the
correct (or at least the most useful) one is the one involving open coverings and finite
subcovers.
4.27 Theorem: (Heine–Borel theorem) For a subset A ⊆ Rd , the following three
conditions are equivalent:
(i) A is closed and bounded;
(ii) every sequence in A has a subsequence converging to a point in A;
(iii) every open covering of A has a finite subcover.
Proof: We already know that (i) and (ii) are equivalent.
(i) =⇒ (iii) For this, the more difficult part of the proof, we first prove a couple
of lemmata.
1 Lemma: If K1 ⊆ Rm is compact and if K2 ⊆ Rn is compact then K1 ×K2 ⊆ Rm+n
is compact.
Proof: Let us denote points in Rm+n by (x, y) ∈ Rm × Rn . For x ∈ Rm , denote

K2,x = {(x, y) | y ∈ K2 }.

Let U be an open cover of K1 × K2 . For x ∈ K1 , denote

Ux = {U ∈ U | U ∩ K2,x 6= ∅}.

For U ∈ Ux , define
VU = {y ∈ U | (x, y) ∈ K2,x }.
We claim that VU is open. Indeed, let y ∈ VU so that (x, y) ∈ U . Since U is open,
there exists r > 0 such that Br (x, y) ⊆ U . Therefore, Br (y) ⊆ VU , and so VU is
open as claimed. Therefore, (VU )U ∈ Ux is an open cover of K2 . Thus there exists
Ux,1 , . . . , Ux,kx ∈ Ux such that K2 ⊆ ∪kj=1
x
VUx,j .
Now, for U ∈ U, denote

WU = {x ∈ Rm | (x, y) ∈ U for some y ∈ Rn }.

We claim that WU is open. To see this, let x ∈ WU and let y ∈ Rn be such that
(x, y) ∈ U . Since U is open, there exists r > 0 such that Br (x, y) ⊆ U . Therefore,
Br (x) ⊆ WU , giving WU as open, as desired. Now define Wx = ∩kj=1 x
WUx,j and note
 4.5 Compact sets and open coverings 61

that, by Proposition 4.6, it follows that Wx is open. Thus (Wx )x∈K1 is an open cover
for K1 . By compactness of K1 , there exists x1 , . . . , xm ∈ K1 such that K1 = ∪m
l=1 Wxl .
Therefore,
kx
K1 × K2 = ∪x∈K1 K2,x = ∪x∈K1 ∪kj=1
x
Ux,j = ∪m
l=1 ∪j=1 Uxl ,j ,
l

so giving a finite subcover of K1 × K2 . H

2 Lemma: The set [a, b] ⊆ R satisfies condition (iii).
Proof: Let U be an open cover for [a, b] and let
S[a,b] = {x ∈ R | x ≤ b and [a, x] has a finite subcover in U}.
Note that S[a,b] 6= ∅ since a ∈ S[a,b] . Let c = sup S[a,b] . We claim that c = b. Suppose
that c < b. Since c ∈ [a, b] there is some U ∈ U such that c ∈ U . As U is open,
there is some r > 0 sufficiently small that Br (c) ⊆ U . By definition of c, there
exists some x ∈ (c − r, c) for which x ∈ S[a,b] . By definition of S[a,b] , there is a finite
collection of open sets U1 , . . . , Um from U which cover [a, x]. Therefore, the finite
collection U1 , . . . , Um , U of open sets covers [a, c + ). This then contradicts the fact
that c = sup S[a,b] , so showing that b = sup S[a,b] . The result follows by definition of
S[a,b] . H
3 Lemma: If A ⊆ Rd satisfies condition (iii) and if B ⊆ A is closed, then B
satisfies (iii).
Proof: Let U be an open cover for B and define V = Rd \ B. Since B is closed,
U ∪ (V ) is an open cover for A. Since A satisfies (iii), there exists U1 , . . . , Uk ∈ U
such that A ⊆ ∪kj=1 Uj ∪ V . Therefore, B ⊆ ∪kj=1 Uj , giving a finite subcover of B. H
Now we proceed with the proof of this part of the theorem. Suppose that A is
closed and bounded. Let R > 0 be sufficiently large that A ⊆ [−R, R] × · · · × [−R, R].
By Lemma 2 it follows that [−R, R] satisfies condition (iii). By induction, using
Lemma 1, it follows that [−R, R] × · · · × [−R, R] is compact. By Lemma 3 it follows
that K is compact.
(iii) =⇒ (i) Suppose that A satisfies (iii). Consider the set {Bn (0) | n ∈ Z>0 }
of open balls in Rd ; this set is certainly an open covering of A (it is even an open
d
covering
Sk of all of R ). Then (iii) implies that there are integers nS
1 , . . . , nk such that
A ⊆ i=1 Bni (0). Now, if m = max{n1 , . . . , nk }, then Bm (0) = ki=1 Bni (0) and so
A ⊆ Bm (0) and, therefore, A is bounded.
To show that A is also closed, it is sufficient to show that its complement A{ is
open. Let x ∈ A{ and consider the set
U = {B r (x){ | r > 0}
of open sets. Note that
\ [
{x} = B r (x) =⇒ {x}{ = B r (x){ ,
r>0 r>0
62 4 Some topology in Rd

using de Morgan’s Laws. Then, if a ∈ A then a 6= x and so

[
a ∈ {x}{ =⇒ a∈ B r (x){ ,
r>0

and so U is an open covering of A. Sn But then{ (iii) implies that there are positive
numbers r1 , . . . , rn such that A ⊆ j=1 B rj (x) . Now, if r = min{r1 , . . . , rn }, then
r > 0 and n n
\ [
{
B r (x) ⊆ B rj (x) =⇒ B r (x) = B rj (x){
j=1 j=1

and, therefore, A ⊆ B r (x){ . But this means that B r (x) ⊆ A{ and so Br (x) ⊆ A{ .
Since x was an arbitrary point in A{ , it follows that A{ is open. 
The more difficult part of Theorem 4.27 is the implication (i) =⇒ (iii). One part
of the proof we pull out as the following example, since this may help us to understand
this implication.
4.28 Example: We shall show that any open covering of [0, 1] has a finite subcover.
Indeed, suppose that U is an open covering of [0, 1] and consider the set

C = {x ∈ (0, 1] | there is finite subcover of [0, x]}.

The first step is to show that C is not empty. There is a set U in U with 0 ∈ U and
hence with (−r, r) ⊆ U for some r > 0. But then [0, r/2] ⊆ U and so there is finite
subcover (consisting of just one set, in fact) of [0, r/2] and thus r/2 ∈ C.
To prove that U contains a finite subcover of [0, 1], it is enough to show that
1 ∈ C and we will do this by putting s = sup C and showing that s = 1 and that
s ∈ C.
Since s ∈ [0, 1] there will be a set U ∈ U with s ∈ U and hence with (s−, s+) ⊆
U for some  > 0. Now, by Theorem 1.12, the interval (s − , s] must contain a point
in C, sayS x. Then there are a finite number of sets U1 , . . . , Un in U such that
[0, x] ⊆ ni=1 Ui . But then the sets U, U1 , . . . , Un are a finite subcover of [0, s] and
so s ∈ C. Moreover, if s < 1 and if r = min{1 − s, /2}, then it is even true that
s + r ≤ 1 and these same n + 1 sets cover [0, s + r], and hence s + r ∈ C. But this
contradicts the choice of s and therefore s = 1. •
Let us give a couple of examples of the use of the finite subcover characterisation
of compactness. For the first result, we shall make use of the following definition.
4.29 Definition: Let A ⊆ Rd and let f be a real-valued function on A. The function
f is uniformly continuous if, for each  > 0, there exists δ > 0 such that |f (x) −
f (y)| <  if x, y ∈ A satisfy kx − yk < δ. •
Note carefully the difference between “continuous” and “uniformly continuous.”
The next theorem shows that these notions align for compact domains.
 4.5 Compact sets and open coverings 63

4.30 Theorem: (Heine–Cantor Theorem) If K ⊆ Rd is compact, then a real-

valued function f on K is continuous if and only if it is uniformly continuous.
Proof: It is clear that, if f is uniformly continuous, then it is continuous. To show that
f is uniformly continuous when it is continuous and when its domain K is compact,
let  > 0. Continuity of f implies that, for x ∈ K, there exists δx > 0 such that
|f (y) − f (x)| < /2 for y ∈ Bδx (x) ∩ K. Note that the balls

{Bδx /2 (x) | x ∈ K}

cover K. Compactness of K implies that there exist x1 , . . . , xk ∈ K such that

K ⊆ ∪kj=1 Bδxj /2 (xj ). Let

δ = min{δx1 /2, . . . , δxk /2}.

Now let x, y ∈ K satisfy kx − yk < δ. We must have x ∈ Bδxj /2 (xj ) for some
j ∈ {1, . . . , k}. This, in turn, gives

kxj − yk ≤ kxj − xk + kx − yk < 21 δxj + δ < δxj .

Thus x, y ∈ Bδxj (xj ). Then

 
|f (x) − f (y)| ≤ |f (x) − f (xj )| + |f (xj ) − f (y)| < + = ,
2 2
giving uniform continuity. 
As another application of the open cover characterisation of a compact set, let us
give an alternative proof of the fact, proved as Theorem 4.24(i) above, that the image
of a compact set under a continuous function is compact.
4.31 Theorem: If A ⊆ Rd is compact and if f is a continuous Rm -valued function
on A, then image(f ) is compact.
Proof: Let U be an open cover of image(f ). Let x ∈ A and let Ux ∈ U be such that
f (x) ∈ Ux . Openness of Ux implies that there exists x > 0 such that Bx (f (x)) ⊆
Ux . By continuity of f , there exists δx > 0 such that, if y ∈ Bδx (x) ∩ A, then
f (y) ∈ Bx (f (x)). Note that

{Bδx (x) | x ∈ A}

is an open cover of A. Compactness of A means that there are x1 , . . . , xk ∈ A

such that A ⊆ ∪kj=1 Bδxj (xj ). We claim that image(f ) ⊆ ∪kj=1 Uxj . Indeed, let
f (x) ∈ image(f ). Then x ∈ Bδxj (xj ) for some j ∈ {1, . . . , k}. But then

f (x) ∈ Bxj (xj ) ⊆ Uxj ,

giving the desired conclusion. 

64 4 Some topology in Rd

Exercises
E4.1 Determine the interior points, the cluster points, the limit points, the boundary
points, and the isolated points of each of the following sets (note that the sets
in (a)–(c) are subsets of R2 whereas the set in (d) is a subset of R):
(a) {(q, r) | q and r are rational and 0 ≤ q, r < 1}
(b) {(x, y) | x ∈ R and − 1 < y < 1}
n1 1   o
(c) ,  n, m ∈ Z>0

n m
∞  
[ 1 1
(d) ,
n=1
n + 1 n
E4.2 Determine the interior points, the cluster points, the limit points, the isolated
points, and the boundary points of each of the following sets. Also, classify
each of the sets as open, closed, neither or both.
  
1 
(a) n ∈ Z>0 , n ≥ 1
n
(b) {x ∈ Rd | kxk = 1}
E4.3 Determine the interior points, the boundary points, the cluster points, the
limit points, and the isolated points of each of the following subsets of R2 .
Also, classify each of the sets as open, closed, neither or both.
(a) {(m, n) | m, n ∈ Z}
(b) {(x, y) ∈ R2 | 0 < x ≤ 1}
E4.4 Determine the interior points, the cluster points, the limit points, the boundary
points, and the isolated points for each of the following sets. Also, state
whether the sets are open or closed or neither. (Note that the sets in (a)–
(e) are subsets of R.)
(a) (1, 2)
(b) [2, 3]
(c) (1, 4]
\∞
(d) [−1, 1/n)
n=1
(e) (0, 1) ∩ Q
(f) Rd
(g) A hyperplane in Rd .
(h) {(x, y) ∈ R2 | 0 < x ≤ 1}
(i) {x ∈ Rd | kxk = 1}
E4.5 Let S = {(x, y) ∈ R2 | |x| ≤ 1 and |y| < 1}.
(a) Show that S is not closed by finding a sequence in S that converges to a
point not in S.
 Exercises for Chapter 4 65

(b) Show S is not closed by showing that R2 \ S is not open.

(c) Determine the interior points of S.
E4.6 Consider the following subset S of R2 :

S = {(x, y) ∈ R2 | y > x2 }.

Answer the following questions with complete justifications.

(a) Is S open?
(b) If possible, find a point in S that is not an interior point.
(c) Is S closed?
(d) If possible, find a point not in S that is the limit of a sequence of points
in S.
(e) Is S bounded?
(f) Is S compact?
E4.7 Let (an ) be a sequence of real numbers, and let T be the set of all limits of its
convergent subsequences.
(a) Show that T contains all of its cluster points, i.e., der(T ) ⊆ T .
(b) Give an example of a sequence (an ) for which der(T ) = T , and an example
for which der(T ) 6= T .
E4.8 Give examples in Rd of the following or else explain why there are no such
examples.
(a) A boundary point of a set S that is not a cluster point of S.
(b) A cluster point of a set S that is not a boundary point of S.
(c) An isolated point of a set S that is not a boundary point of S.
E4.9 Suppose that A ⊆ R is such that A 6= ∅ and A{ 6= ∅. The aim of this exercise
is to show that bd(A) 6= ∅.
Pick any x ∈ A and y ∈ A{ . Without loss of generality, we may assume
x < y. Define P = {p ∈ A | p < y}.
(a) Justify the fact that sup P exists.
Set a = sup P , and define Q = {q ∈ A{ | q ≥ a}.
(b) Prove that a = inf Q.
(c) Deduce from part (b) that a ∈ bd(A) and, therefore, bd(A) 6= ∅.
E4.10 Give examples of:
(a) An infinite set in R with no cluster points.
(b) A nonempty subset of R that is contained in its set of cluster points.
(c) A subset of R that has infinitely many cluster points but contains none
of them.
(d) A set S such bd(S) = S ∪ der(S).
E4.11 Suppose that (An ) is a sequence of nonempty subsets of R, and assume that
A1 is bounded and that A1 ⊇ A2 ⊇ A3 . . . ,, and let an = inf An for each n.
66 4 Some topology in Rd

(a) Show that a1 ≤ a2 ≤ a3 ≤ . . ..

(b) Show that, if each An is closed, then
∞
\
sup{a1 , a2 , a3 , . . .} ∈ An .
n=1

(c) Give an example to show that it is possible for ∩∞ n=1 An to be empty.

E4.12 Conditions (i)–(iv) in the statement of Theorem 4.20 are equivalent and in the
proof of this theorem it was shown that (i) implies (ii), that (ii) implies (iii),
that (iii) implies (iv), and that (iv) implies (v).
(a) Give a direct argument showing that (iv) implies (iii).
(b) Give a direct argument showing that (iii) implies (ii).
(c) Give a direct argument showing that (ii) implies (i).
E4.13 (a) Prove that, if A ⊆ Rd is both open and closed, then bd(A) = ∅. (You
may not use the as-yet-unproved fact that ∅ and Rd are the only subsets
of Rd that are both open and closed.)
(b) Prove that, if A ⊆ R is both open and closed, then either A = ∅ or A = R.
Hint: Use Exercise E4.9.
E4.14 Suppose that x ∈ Rd and that S is a nonempty subset of Rd . Show that, if x
is a boundary point but not an isolated point of S, then x is a cluster point
of S.
E4.15 Let A be a subset of Rd . Show that cl(A) = A ∪ der(A).
E4.16 Suppose that (an ) is a sequence in Rd and let R = {an | n ∈ Z>0 } be its
range, let C be the set of cluster points of R, and let L be the set of all limit
points of R.
(a) Show that C ⊆ L.
(b) Give an example of a sequence in R such that C 6= L.
E4.17 Suppose that S is a subset of Rd .
(a) Show that the set of cluster points of S is a closed subset of Rd .
(b) Show that the set of limit points of S is a closed subset of Rd .
(c) Show that the set of boundary points of S is a closed subset of Rd .
(d) Show that the set of isolated points of S is a closed subset of Rd .
E4.18 Suppose that A and B are two nonempty subsets of Rd and assume that A is
open. Is the set
{x + y | x ∈ A and y ∈ B}
open? Why?
E4.19 (a) Show that der(A ∪ B) = der(A) ∪ der(B) for any two nonempty subsets
A and B of Rd .
 Exercises for Chapter 4 67

(b) Is it true that !

n
[ n
[
der Ai = der(Ai )
i=1 i=1

whenever A1 , . . . , An are nonempty subsets of Rd ?

(c) Is it true that
∞
! ∞
[ [
der Ai = der(Ai )
i=1 i=1

for any sequence (Ai ) of nonempty subsets of Rd ?

E4.20 (a) Show that, if A ⊆ Rd , then bd(A ∪ bd(A)) ⊆ bd(A).
(b) Show that, if A ⊆ Rd , then A ∪ bd(A) is a closed set.
(c) Give an example of a set A ⊆ R such that bd(A ∪ bd(A)) 6= bd(A).
E4.21 Prove directly that the intersection of two closed sets is closed.
E4.22 If A is any subset of Rd , then der(der(A)) is the set of all cluster points of
der(A).
(a) Give an example of a subset A of R for which der(der(A)) = ∅ and
der(A) 6= ∅.
(b) Show that, if A is a nonempty subset of Rd , then der(der(A)) ⊆ der(A).
(c) Give an example of a subset A of R for which der(der(A)) = der(A).
(d) Give an example of a subset A of R for which ∅ = 6 der(der(A)) 6= der(A).
E4.23 Prove Proposition 4.17.
E4.24 Suppose S is a compact subset of Rd and that (xn ) is a sequence in S. Show
that every convergent subsequence of (xn ) converges to a point in S.
E4.25 Suppose that f is a real-valued function with domain R that is nondecreas-
ing. Prove that the limit limx→0+ f (x) exists, limx→0+ meaning the limit as x
approaches zero from the right.
Hint: Let A = f ({x ∈ R | x > 0}) and consider inf A.
E4.26 Suppose that S is a closed subset of Rd and that f is a continuous real-valued
function with domain S. Show that, for every real number a, the set

{x ∈ S | f (x) ≤ a}

is a closed subset of Rd .
E4.27 Let f : Rd → Rm be a function. For any B ⊆ Rm , the preimage under f , of B
is defined to be the set

f −1 (B) , {x ∈ Rd | f (x) ∈ B}.

Show that f is a continuous function if and only if, for each open set U ⊆ Rm ,
the preimage f −1 (U ) is an open set in Rd .
E4.28 Let S be a nonempty subset of R.
68 4 Some topology in Rd

(a) Give an example to show that, if S is not closed, then there is a continuous
real-valued function with domain S that does not attain a maximum value.
(b) Give an example to show that, if S is not bounded, then there is a contin-
uous real-valued function with domain S that does not attain a maximum
value.
(c) Give an example of a continuous bounded function with domain R that
attains neither a maximum or minimum value.
E4.29 (Intermediate value theorem) Suppose that f is a continuous real-valued
function whose domain includes a closed interval [a, b]. Show that, if f (a) < 0
and f (b) > 0, then there is a point c ∈ [a, b] such that f (c) = 0.
Hint: Consider the set {x ∈ [a, b] | f (x) < 0}.
E4.30 Show that every cubic polynomial has a root.
Hint: Use Exercise E4.29.
E4.31 Show that every continuous function from the unit interval to itself has a fixed
point.
Hint: Apply the intermediate value theorem to the function x 7→ f (x) − x.
E4.32 Suppose that f is a continuous function whose domain is [1, 2] and that satisfies
0 ≤ f (x) ≤ 3 for all x ∈ [1, 2]. Show that f has a fixed point, i.e., that there
is a point x ∈ [1, 2] satisfying f (x) = x.
E4.33 Let f be a function from Rm to Rn and recall that we can write f (x) =
(f1 (x), . . . , fn (x)) for x ∈ Rm , where f1 , . . . , fn are real-valued functions on
Rm . Show that f is continuous if and only if fi is continuous for i ∈ {1, . . . , n}.
 Chapter 5
Infinite series

In Chapter 2 we considered various attributes of sequences in R. We saw in that

section that many interesting examples of sequences arise as partial sums of infinite
series. In this section we carefully consider infinite series, and their convergence.

5.1 Definitions and examples

The problem which will be considered in this section is that of assigning a meaning
to an infinite sum of real numbers, i.e., to an expression of the form
∞
X
an = a1 + a2 + · · ·
n=1

for a sequence (ak ). The natural way to do this would be to consider the limit (as n
tends to ∞) of the sum of the first n terms, and this is, in fact, what will be done.
5.1 Definition: Consider a sequence (ak ) of real numbers defined for k ≥ 1. The
formal expression
X∞
ak (5.1)
k=1

is called the infinite series corresponding to the sequence (ak ). For n ∈ Z>0 the
sum n
X
Sn = ak
k=1

is called the n partial sum of (5.1) and the sequence (Sn ) is called the sequence
of partial sums corresponding to (5.1).
If the sequence (Sn )P
converges then the series (5.1) is said to converge
P∞and one
∞
says that the value of k=1 ak is the limit limn→∞ Sn , and one writes k=1 ak =
limn→∞ Sn . On the other hand, if the sequence (Sn ) diverges, then the series
P (5.1) is
said to diverge or to not exist and no meaning is assigned to the symbol ∞ k=1 ak . •
The point of this definition is that, for some sequences (ak ), the infinite series (5.1)
is defined to be a real number, while for other sequences (5.1) is not assigned a
meaning. Namely, if the limit limn→∞ Sn exists, then (5.1) is defined to be the value
70 5 Infinite series

of this limit, whereas if this limit does not exist then the expression (5.1) is not
assigned a value. In the latter case, then, we cannot “talk” about and manipulate
the “sum” as though it were a number—it is merely a “formal expression.1 ” Students
sometimes ignore the importance of the sequence (Sn ) of partial sums and thereby
miss the essential feature of infinite series. Partial sums are finite sums, so they can
be dealt with and manipulated according to all the familiar rules for addition; the
corresponding statement for infinite series is not true. All statements about infinite
series must be understood as abbreviations for statements about limits of partial
sums.
Notice that, if two sequences are identical except for a finite number of terms, then
the two corresponding infinite series will either both converge or both diverge. Indeed,
for n sufficiently large the two nth partial sums will have a constant difference, and
hence will either both converge or both diverge. A particularly important case of this
arises when we “neglect the first few terms of aP series:” If (aP
k ) is a sequence defined
for k ∈ Z>0 and if N ∈ Z>0 , then the two series k=1 ak and ∞
∞
k=1 ak will either
P both
converge or both diverge. To see this,Plet Sn denote the nth partial sum nk=1 ak ,
and let Tn denote the nth partial sum nk=N ak for n ≥ N . Then Sn = SN −1 + Tn for
n ≥ N and, since SN −1 is fixed, limn→∞ Sn exists if and only if limn→∞ Tn exists.
It seems somewhat ironic that, while it is generally difficult to calculate the value
of a convergent infinite series, it is often relatively easy to decide whether an infinite
series converges or diverges. What this really means, when rephrased in terms of
partial sums, is that, while it is generally difficult to calculate the limit of a convergent
sequence, it is often relatively easy to decide whether a sequence converges or diverges.
(Of course, if a series is known to converge one can, in principle, calculate its sum as
accurately as desired by simply calculating its nth partial sum for large values of n.
But this is not the same as finding the value of the infinite series.) Before going on to
describe these so-called “convergence tests,” it will be useful to give a few examples.
5.2 Examples: 1. It is easy to see that both of the infinite series
∞
X ∞
X
1 and (−1)k
k=1 k=1

diverge. Indeed, for the first series the nth partial sum is n and so the sequence
of partial sums diverges to ∞. And for the second the sequence of partial sums
is −1, 0 − 1, 0, . . . and hence is divergent.
2. Let r be a real number and consider the infinite series ∞ k
P
k=0 r . This series is
called a geometric series; let (Sn ) be its sequence of partial sums. It is obvious
that, if r = 1, then Sn = n + 1 and hence the geometric series diverges in this
case. Now suppose that r 6= 1. Then

(1 − r)(1 + r + · · · + rn ) = 1 − rn+1 ,
1
Something which can be made precise, but which will definitely not be made precise here.
 5.1 Definitions and examples 71

and hence
1 − rn+1
Sn = .
1−r
This makes it clear that, if |r| < 1, the geometric series converges with
∞
X 1
rk = ,
k=0
1−r

and that, if |r| ≥ 1, the geometric series diverges.

3. Consider the series ∞
X 1
.
k=1
k(k + 1)
Notice that the nth partial sum is
n n  
X 1 X 1 1
= −
k=1
k(k + 1) k=1
k k+1
     
1 1 1 1 1
= 1− + − + ··· + −
2 2 3 n n+1
1
=1−
n+1
and, therefore,
∞  
X 1 1
= lim 1 − = 1.
k=1
k(k + 1) n→∞ n+1
Here the partial sums are said to be “telescoping sums” because of the way they
simplified.
4. The series ∞
X 1
n=1
n
is called the harmonic series. Recall from Example 2.8–2 that the 2n th partial
sum S2n exceeds n/2, and hence the harmonic series diverges to ∞.
We now begin considering theorems that will allow us to decide questions about
convergence without obtaining explicit formulae for the partial sums. The first theo-
rem is very simple.
5.3 Theorem: Let (ak ) be a sequence of numbers.
(i) If the infinite series ∞
P
k=1 ak converges, then limk→∞ ak = 0.
P∞
(ii) If the sequence (ak ) does not converge to 0, then the infinite series k=1 ak
diverges.
72 5 Infinite series

Proof: The two parts of this theorem are, of course, logically equivalent and so it is
only
P∞ necessary to prove (i). Let (Sn ) be the sequence of partial sums of the series
k=1 ak and let  be a given positive number. Then limn→∞ Sn exists and, if L =
limn→∞ Sn , there is an integer N such that |L − Sn | < 2 whenever n ≥ N . But then,
for any integer n ≥ N , we have

|an+1 | = |Sn+1 − Sn |
≤ |Sn+1 − L| + |L − Sn |
 
< + = .
2 2
This shows that limn→∞ an = 0. 
Notice that Example 5.2–4 shows that the converse of this theorem is false. It is
very important to realize that this theorem gives a necessary, but not a sufficient,
condition for convergence of a series. This theorem can sometimes be used to establish
the divergence of a series but cannot be used to establish convergence.
P∞One way to understand the preceding example and theorem is that, if a series
k=1 ak is to converge, then it is necessary that ak → 0 quite quickly as k → ∞, and
that the terms of the harmonic series simply approach zero too slowly.

5.2 Convergence of nonnegative series

It turns out that series with nonnegative terms are easier to discuss than those in
which the terms may be positive or negative. To see why, consider a sequence (ak )
of nonnegative
P∞ numbers and let (Sn ) be the sequence of partial sums of the infinite
series k=1 ak . Then Sn+1 = Sn + an+1 ≥ Sn for each n and, therefore, (Sn ) is a
nondecreasing sequence. So, by Theorem 2.9, the sequence (Sn ) converges if and only
if it is bounded. It will be useful to record this observation as a theorem.
5.4
P∞Theorem: If (ak ) is a sequence of nonnegative terms, then the infinite series
k=1 ak converges if and only if the corresponding sequence of partial sums is bounded.
Next we give four “convergence tests” that are useful in deciding whether a given
series with nonnegative terms is convergent or divergent.
5.5 Theorem: (Comparison test) Suppose that (ak ) and (bk ) are two sequences.
P∞
(i) If 0 ≤Pak ≤ bk for all k and if the series k=1 bk converges, then so does the
∞
series k=1 ak .
(ii) If 0 ≤Pak ≤ bk for all k and if the series ∞
P
k=1 ak diverges, then so does the
∞
series k=1 bk .
P∞
0 for all k, if 0 ≤ limk→∞ ak /bk < ∞, and if
(iii) If bk > P k=1 bk converges, then
∞
so does k=1 ak .
P∞
> 0 for all k, if 0 < limk→∞ ak /bk ≤ ∞, and if
(iv) If bk P k=1 bk diverges, then so
∞
does k=1 ak .
 5.2 Convergence of nonnegative series 73

Proof: (i) Let Sn = nk=1 ak and Tn = nk=1 bk for n ∈ Z>0 . Then the assumption
P P
that P0 ≤ ak ≤ bk for all k implies that 0 ≤ Sn ≤ Tn for all n, and the assumption
that ∞ k=1 bk converges impliesPthat (Tn ) is a bounded sequence. But then (Sn ) too
is a bounded sequence, and so ∞ k=1 ak converges by Theorem 5.4.
(ii) This assertion is logically equivalent to (i).
(iii) The hypotheses implies that, if L = limk→∞ ak /bk , then there exists an integer
N such that ak /bk ≤ L + 1 for all k ≥ N , and hence such that ak ≤ (L + 1)bk for
all k ≥ N . The desired conclusion now follows from (i), together with our earlier
observation that we may “neglect the first few terms” in showing convergence or
divergence (cf. page 70).
(iv) In this case, let 0 < L = limk→∞ ak /bk . Let  > 0 be such that L −  > 0.
Then there exists N ∈ Z>0 such P that L −  < ak /bk for all k ≥ N . Thus, if (Sn ) are
the partial sums for the sum ∞ k=1 an , ak > (L − )bk for k ≥ N and so, for n > N ,

n
X n
X
Sn − SN = an > (L − ) bn .
k=N +1 k=N +1

Therefore n
 X 
lim Sn > lim (L − ) bn + SN = ∞,
n→∞ n→∞
k=N +1

as desired. 
In applying parts (i) and (ii) of this theorem, it is sufficient that there be an n0
such that the inequalities 0 ≤ ak ≤ bk hold for all k ≥ N (and not for all k as in the
statement of the theorem). This follows from our earlier remark (cf. page 70) about
“neglecting the first few terms” in proving convergence or divergence.
5.6 Examples: 1. Since k 1/2 ≤ k for k ≥ 1, it follows that 1/k 1/2 ≥ 1/k for
P∞k ≥ 11.
The comparison test and Example 5.2–4 then imply that the series k=1 k1/2
diverges.
2. Since
1 1
>
2n + 1 2(n + 1)
for n ∈ Z>0 , the comparison test and Example 5.2–4 imply that the series
P ∞ 1
n=1 2n+1 diverges. Notice that this series consists of odd terms of the harmonic
sequence.
3. Since
1 2
2
≤
n n(n + 1)
for n ∈ Z>0 , the comparison test and Example 5.2–3 imply that the series ∞ 1
P
n=1 n2
converges.
74 5 Infinite series
P∞ 3
4. Does the series n=1 2n2 −n1/2 converge? Since the terms are positive and
 . 1 3
3
lim = ,
n→∞ 2n2 − n1/2 n2 2
the series converges by 3 and the comparison test. Note that it would be more
awkward to use part (i) of the comparison test for this particular example. •

5.7 Theorem: (Integral test for positive series) Suppose that (ak ) is a sequence
and that f is a function which is continuous,2 nonnegative-valued, and nonincreasing
half-line [1, ∞), and satisfies ak = f (k) for all k ∈ Z>0 . Then the infinite
on the P
series ∞ k=1 ak and the improper integral
Z ∞
f (x) dx
1

either both converge or both diverge.

Proof: Since f is nonincreasing and nonnegative-valued, it is clear from thinking
about its graph that
Z k+1
ak+1 = f (k + 1) ≤ f (x) dx ≤ f (k) = ak (5.2)
k

for all k ∈ Z>0 and, therefore, that

n
X n Z
X k+1 n
X
ak+1 ≤ f (x) dx ≤ ak
k=1 k=1 k k=1

or
n+1
X Z n+1 n
X
ak ≤ f (x) dx ≤ ak (5.3)
k=2 1 k=1

for each positive integer n. P∞

Suppose
Pn first that the series k=1 ak converges. Then there is a number, say M ,
such that k=1 ak ≤ M for all n ≥ 1 (by Theorem 5.4) and thus
Z n+1
f (x) dx ≤ M
1

for all n ≥ 1 by (5.3). But Z n

n 7→ f (x) dx
1
is a nondecreasing
Rn function of n since f is nonnegative valued and, therefore, the
limit limn→∞ 1 f (x) dx exists by Theorem 2.9.
2
The assumption that f is continuous is made merely to ensure that it is integrable over any
interval of finite length.
 5.2 Convergence of nonnegative series 75
R∞
Now suppose that, conversely, the improper integral 1
f (x) dx converges. Then
n+1
X Z n+1 Z ∞
ak ≤ f (x) dx ≤ f (x) dx < ∞
k=2 1 1

P∞
for all n ∈ Z>0 by (5.3), and so the series k=1 ak converges by 2.9. 
Under the hypothesis of the integral test, it is sufficient to assume that there
is a positive integer N such that ak = f (k) for all k ≥ N , and f is nonincreas-
ing, nonnegative-valued, and continuous on [N, ∞). This remark is similar to one
made following the proof of the comparison test and follows from our earlier remark
(cf. page 70) about “neglecting the first few terms” in proving convergence or diver-
gence.
Let us consider some applications of the integral test.
5.8 Example: The two infinite series
∞ ∞
X 1 X 1
p
,
n=1
n k=2
k(log k)p

both converge for p > 1 and diverge for p ≤ 1. To see this, it is enough to apply the
integral test with the functions f (x) = x1p and f (x) = x(log1 x)p . In carrying out this
integration is is necessary to distinguish between p = 1 and p 6= 1. •
P∞ 1
The infinite series k=1 kp is called the p-series. Notice that, if p = 1, then the
p-series is just the harmonic series.
Let us turn to another convergence test.
5.9 Theorem: (Ratio test for positive series) Suppose that (ak ) is a sequence of
positive terms and assume that limk→∞ ak+1 ak
= L exists.
P∞
(i) If L < 1, then k=1 ak converges.
(ii) If L > 1, then ∞
P
k=1 ak diverges.
(iii) If L = 1, then this test is inconclusive.
Proof: (i) Suppose that L < 1 and let r be any number satisfying L < r < 1. Then
there must be N ∈ Z>0 such that 0 < ak+1ak
< r for all k ≥ N . Thus ak+1 < rak for
all k ≥ N and, therefore,

aN +1 ≤ aN r
aN +2 ≤ aN +1 r ≤ aN r2
aN +3 ≤ aN +2 r ≤ aN r3
..
.
aN +k ≤ aN rk
76 5 Infinite series

for k ≥ 1. Now ∞
P k
P∞
k=1 r is a convergent geometric series and, therefore, k=1 ak
converges by the comparison test (Theorem 5.5(i)).
(ii) Suppose that L > 1. Then there exists an N ∈ Z>0 such that ak+1
ak
≥ 1 for all
k ≥ N . This means that ak+1 ≥ ak for all k ≥ N and, therefore,

aN +1 ≥ aN
aN +2 ≥ aN +1 ≥ aN
aN +3 ≥ aN +2 ≥ aN
..
.

This shows that limk→∞ ak 6= 0 and therefore ∞

P
k=1 ak diverges by Theorem 5.3.
(iii) If ak = k for all k or if ak = k −2 for all k, then limk→∞ ak+1
ak
= 1, yet the first
of these series diverges and the second converges. 
The following test sometimes works when the ratio test does not apply. Note that
it uses the limit superior and so may be applicable even in cases where the terms are
√
quite irregular and limn→∞ n an does not exist.
5.10 Theorem: (Root test for positive series) Suppose that (an ) is a sequence
√
of positive terms and let L = lim supn→∞ n an .
(i) If L < 1 then ∞
P
n=1 an converges.
P∞
(ii) If L > 1 then n=1 an diverges.
(iii) If L = 1 this test is inconclusive.
Proof: (i) Say L < 1 and let r be a number satisfying L < r < 1. Then (by
√
Proposition 2.19) the inequality n an > r holds for only finitely many n and, therefore,
√
there is anP integer N such that n an ≤ r for all n ≥ N . Thus an ≤ rn for all n ≥ N
and, since P ∞ n
n=1 r is a convergent geometric series, the comparison test implies that
∞
the series n=1 an converges.
(ii) Now suppose that L > 1. Then Proposition 2.19 implies that the inequality
√n a > 1 holds for infinitely many n, and thus a > 1 for infinitely many n. But then
n n
it cannot be true that limn→∞ an = 0 and, therefore, the series ∞
P
n=1 an diverges by
Theorem 5.3.
(iii) The same examples which were used in Theorem 5.9(iii) can be used here. 
At this point some general comments on the four convergence tests in this section
are probably in order. The comparison test is often the subtlest and most sensitive
as well as the hardest ofPthe four tests to use. In using the comparison test to decide
∞
whether a given series k=1 ak converges or diverges, we must employ a second series
whose behavior is understood.
P∞ In fact, if we suspect that the given series converges,
we must find a series k=1 bk which we know converges and which satisfies ak ≤ bk
for all sufficiently large values of k. OnP the other hand, if we suspect that the given
series diverges, we must find a series ∞ k=1 bk which we know diverges and which
satisfies bk ≤ ak for all sufficiently large values of k. So, if we have a large store of
 5.3 Convergence of general series 77

series whose behaviors we understand, then it is possible to make good use of the
comparison test.
The other three tests are more straightforward to use. If either the ratio or root
tests predicts the convergence of a series, then that series converges at least as rapidly
as a geometric series (this is clear from the proofs of these two tests). For series that
converge relatively slowly (such as the p-series), the ratio and root tests are necessarily
inconclusive and there is no alternative but to use the comparison or integral test, or
possibly some other device.
In applying the ratio and root tests the following simple lemma is sometimes
useful.
√
5.11 Lemma: limn→∞ n n = 1.
Proof: If f (x) = x1/x then

ln x 1/x
lim ln f (x) = lim = lim =0
x→∞ x→∞ x x→∞ 1
√
by l’Hôspital’s rule. Thus limx→∞ f (x) = e0 = 1 and so limn→∞ n
n = 1. 
Let us consider an example that makes use of this fact.
5.12 Example: Consider the series ∞ n
P
n=0 nr , where r ≥ 0. Here
√ √
lim n nrn = lim r n n = r
n→∞ n→∞

by Lemma 5.11, and so the root test implies that the series converges if r < 1 and
diverges if r > 1. The root test is inconclusive if r = 1 but it is clear that this series
diverges in this case. •

5.3 Convergence of general series

Consider a series (an ) whose terms may bePpositive, negative, or zero, and let (Sn )
∞
be the sequence of partial sums of the series n=1 an . Define two new sequences (pn )
and (qn ) as follows: (
an , an ≥ 0,
pn =
0, an < 0.
and (
0, an ≥ 0,
qn =
−an , an < 0.
Pn
Then pn , qPn ≥ 0, an = pn − qn , and |an | = pn + qn for all n. Now put Un = k=1 pk
n
and Vn = k=1 qk for n ∈ Z>0 , so that Sn = Un − Vn for n ≥ 1 and U1 ≤ U2 ≤ · · · ,
and V1 ≤ V2 ≤ · · · . The sequences (Un ) and (Vn ) thus either converge or diverge
to ∞ by Theorem 2.9; put U = limn→∞ Un and V = limn→∞ Vn . There are four
possibilities to be considered:
78 5 Infinite series

1. U < ∞ and V < ∞, in which case limn→∞ Sn = U − V ;

2. U < ∞ and V = ∞, in which case limn→∞ Sn = −∞;
3. U = ∞ and V < ∞, in which case limn→∞ Sn = ∞;
4. U = ∞ and V = ∞, in which case it is not clear what is limn→∞ Sn .
In case 4, the series ∞
P
n=1 an is very unstable and, by permuting its terms, can be
made to do any of the following: converge to any specified number; diverge to ∞;
diverge to −∞; diverge without diverging to ±∞. Such series should be avoided
since, if they converge, they “just barely” converge and do not obey the usual rules
of arithmetic. In contrast, in cases 1–3 the series is stable and its value is unaffected
by permutations of its terms.
The next two examples deal with the so-called alternating harmonic series:
∞
X (−1)n+1 1 1 1
=1− + − + ··· .
n=1
n 2 3 4

These examples will show that this series belongs to case 4 and that it is very much
affected by permutations of its terms.
5.13 Examples: 1. The key observation (or trick) in studyingR 1 n−1the convergence of
the alternating harmonic series is to notice that 1/n = 0 x dx for all n ∈ Z>0
and thus that
m m Z 1
X (−1)n+1 X
n+1
= (−1) xn−1 dx
n=1
n n=1 0
Z 1X m
= (−1)n+1 xn−1 dx
0 n=1
Z 1Xm
= (−x)n−1 dx
0 n=1
1
1 − (−x)m
Z
= dx
0 1+x
Z 1 Z 1
dx (−x)m
= − dx
0 1+x 0 1+x
Z 1 m
m x
= ln 2 − (−1) dx.
0 1+x

Now, for 0 ≤ x ≤ 1, we have

Z 1 m Z 1
x 1
dx ≤ xm dx =
0 1+x 0 m+1
since 1 + x ≥ 1 and, therefore,
1
xm
Z
lim dx = 0.
m→∞ 0 1+x
 5.3 Convergence of general series 79

This implies that

∞ m
X (−1)n+1 X (−1)n+1
= lim
n=1
n m→∞
n=1
n
1
xm
 Z 
m
= lim ln 2 − (−1) dx
m→∞ 0 1+x
= ln 2.

2. This example continues the analysis of the alternating harmonic series in part 1.
In the notation introduced at the beginning of this section, we have
(
1/n, n odd,
pn =
0, n even

and (
1/n, n even,
qn =
0, n odd
and thus ∞
X 1 1
pn = 1 + 0 + + 0 + + 0 + ···
n=1
3 5
and ∞
X 1 1 1
qn = + 0 + + 0 + + 0 + ··· .
n=1
2 4 6
So each of n=1 pn and ∞
P∞ P
n=1 qn is more or less one half the sum of the harmonic
series, and hence they both diverge.
Now let us consider the series
1 1 1 1 1 1 1 1 1
1+ − + + − + + − + − ··· . (5.4)
3 2 5 7 4 9 11 6 13
Notice that this series is a permutation of the alternating harmonic series and
that, if (Sn ) denotes its sequence of partial sums, then
   
1 1 1 1 1 1
S3n+1 = 1 + − + + ··· + − + >1
3 2 5 4n − 1 2n 4n + 1

since
1 1 1
− + >0
2n − 1 2n 2n + 2
for all n ≥ 1. This inequality means that, if the series (5.4) converges, then its
limit exceeds 1 and hence does not converge to ln 2. The alternating harmonic
series and the series (5.4), therefore, behave very differently. •
80 5 Infinite series

The next theorem is very important because it sometimes allows us to use the
comparison, integral, ratio, or root test to establish convergence for series with pos-
itive and negative terms. Note, however, that the converse of this theorem is false
(cf. Example 5.13–1) so that tests for series with positive terms cannot be used to
prove divergence of series with terms of both signs.
P∞ P∞
5.14 Theorem: If n=1 |an | converges, then so does n=1 an .
Pn Pn
Proof: Put Sn = k=1 ak and Tn = k=1 |ak | for n ≥ 1. The hypothesis is that
the limit limn→∞ Tn exists and hence (Tn ) is a Cauchy sequence by Theorem 2.23.
So, given  > 0, there will be an integer N ∈ Z>0 such that |Tn − Tm | <  for all
m, n ≥ N . But then, for any two integers m and n satisfying n > m > N , we have
|Sn − Sm | = |am+1 + am+2 + · · · + an |
≤ |am+1 | + |am+2 | + · · · + |an |
= Tn − Tm < .
This means that (Sn ) is a Cauchy sequence and then 2.23 implies that it converges. 
5.15 Definition: The series ∞
P
n=1 an is said
(i) to converge absolutely if the series ∞
P
n=1 |an | converges and
(ii) to converge conditionally if n=1 an converges but ∞
P∞ P
n=1 |an | diverges. •
This is, at first sight, a very strange definition. It is, in fact, one of the (few?)
situations in mathematics where terminology is a hindrance rather than an aid to
understanding. The first thing to note is that a conditionally convergent series is
convergent, the second is that (by Theorem 5.14) an absolutely convergent series
is convergent, and the third is that there are series that converge conditionally but
not absolutely (the alternating harmonic series, for example). When Theorem 5.14,
in conjunction with the comparison, integral, ratio or root test is used to establish
the convergence of a series with positive and negative terms, it is actually absolute
convergence which is being established. To emphasise this, we state the corresponding
versions of these theorems for general series.
5.16 Theorem: (Integral test for general series) Suppose that (ak ) is a sequence
and that f is a function which is continuous,3 nonnegative-valued, and nonincreasing
half-line [1, ∞), and satisfies |ak | = f (k) for all k ∈ Z>0 . Then the infinite
on the P
series ∞ k=1 ak converges if the improper integral
Z ∞
f (x) dx
1
converges.
Proof: By applying Theorem 5.7 to the series ∞
P P∞
k=1 |ak |, we see that the series k=1 ak
is absolutely convergent, and so convergent by Theorem 5.14. 
3
The assumption that f is continuous is made merely to ensure that it is integrable over any
interval of finite length.
 5.3 Convergence of general series 81

5.17 Theorem: (Ratio test for general series) Suppose that (ak ) is a sequence
of nonzero terms and assume that limk→∞ | ak+1 ak
| = L exists.
P∞
(i) If L < 1, then k=1 ak converges.
(ii) If L > 1, then ∞
P
k=1 ak diverges.
(iii) If L = 1, then this test is inconclusive.
P∞ P∞
Proof: (i) If L < 1 then the series k=1 |ak | converges. Thus k=1 ak converges
absolutely and so converges by Theorem 5.14.
(ii) Referring to the proof of the corresponding part of Theorem 5.9, we see that, if
L > 1, then there exists N ∈ Z>0 such that |ak | ≥ |aN | for all k ≥ N . This prohibits
limk→∞ ak = 0, and so prohibits convergence of the series by Theorem 5.3.
(iii) This follows from the corresponding part of Theorem 5.9. 

5.18 Theorem: (Root test for general p series) Suppose that (an ) is a sequence of
nonzero terms and let L = lim supn→∞ |an |.
n

(i) If L < 1 then ∞

P
n=1 an converges.
P∞
(ii) If L > 1 then n=1 an diverges.
(iii) If L = 1 this test is inconclusive.
Proof: This follows from Theorem 5.10 in exactly the same way as Theorem 5.17
follows from Theorem 5.9. 
The only test in these notes that is capable of establishing the convergence of a
series which does not converge absolutely is the following one.
5.19 Theorem: (Alternating series test) If (an ) is a sequence such that
(i) limn→∞ an = 0,
(ii) |an+1 | < |an | for all n, and
(iii) an+1 /an < 0 for all n,
then the series ∞
P
n=1 an converges.
Proof: Suppose, to be definite, that a1 > 0. Then a2n + a2n+1 < 0, whereas a2n−1 +
a2n > 0 for all n. Let (Sn ) denote the sequence of partial sums of the series in
question. Then

S2n = a1 + · · · + a2n
= a1 + · · · + a2n−2 + a2n−1 + a2n
= S2n−2 + a2n−1 + a2n
> S2n−2

for all n ∈ Z>0 , and so the sequence (S2n ) is strictly increasing. Moreover,

S2n = a1 + (a2 + a3 ) + · · · + (a2n−2 + a2n−1 ) + a2n

< a1 + (a2 + a3 ) + · · · + (a2n−2 + a2n−1 ) + (a2n + a2n+1 ) < a1 ,
82 5 Infinite series

and so the sequence (S2n ) is bounded above by a1 . But then (S2n ) must be a conver-
gent sequence by Theorem 2.9; let S denote its limit. Now S2n+1 = S2n + a2n+1 for
all n ∈ Z>0 , and, since limn→∞ an = 0, it follows that
lim S2n+1 = lim (S2n + a2n+1 ) = S
n→∞ n→∞
P∞
and, therefore, limn→∞ Sn = S and so the series n=1 an converges. 
Let us point out the significance of the three hypothesis. Condition (iii) says that
the terms are nonzero and alternate in sign and condition (ii) says that, in magnitude,
the terms are strictly decreasing. Of course, (i) is necessary for convergence by
Theorem 5.3.
We have already remarked that a series converges conditionally if the sum of its
positive terms diverges to ∞ and the series of its negative terms diverges to −∞, yet
the sum itself converges and has a finite value. This phenomenon can occur because of
a careful “balancing” or “near-cancellation” of the positive and negative contributions
within any particular partial sum. A special case of such “near-cancellation” occurs in
Example 5.13–2 and in the proof of Theorem 5.19. We might expect that permuting
the terms in a conditionally convergent series (a “rearrangement” of the series) would
alter this balancing act and hence change the sum of the series. This is indeed the
case and is exactly the point of the second part of Example 5.13–2.
5.20 Theorem: By permuting the terms of a conditionally convergent series we can
(i) make it converge to any number we wish or diverge to ±∞ or
(ii) make its partial sums oscillate between any two numbers.
Proof: We will only show how to rearrange the terms of a conditionally convergent
series so as to make it converge to a given positive number S; rearranging the terms
to make the series converge to a negative number or diverge to ±∞ is similar or to
make its partial sums oscillate is similar.
First we can reorder the sequence, which we denote by (an ), so that (|an |)
is nonincreasing, this being possible since conditional convergence ensures that
limn→∞ an = 0. First add together the positive terms from the beginning of the
sequence to make a sum that is larger than S (this is possible since the sum of the
positive terms diverges). Now add enough negative terms from the beginning of the
sequence to make the sum less than S. Next, add more positive terms, until the sum is
again larger than S, and then add negative terms to make the sum less than S. Since
the magnitudes of the terms is decreasing (this was the point of initially reordering
so that (|an |) is nonincreasing), the oscillations about S are becoming smaller and
smaller and so the rearranged series converges to S. 

5.4 Estimating remainders

If a series ∞
P
n=1 an converges to, say, S how well is S approximated by the nth
partial sum? This is clearly an important question if we wish to estimate the sum of
 5.4 Estimating remainders 83

a series numerically. Now the error in the approximation is

   
X∞ Xn   X ∞ 
|S − Sn | =  ak − ak  =  ak  .
   
   
k=1 k=1 k=n+1
P∞
We shall refer to k=n+1 ak as the “remainder” or “tail” after the nth-term. Two of
the tests for convergence already considered can be made to yield estimates for these
tails.
5.21 Theorem: (Alternating series test remainder) Suppose the series (an ) sat-
isfies the hypothesis of Theorem 5.19. Then
 ∞ 
 X 
|an+1 + an+2 | <  ak  < |an+1 |
 
 
k=n+1

for any n ≥ 1.
Proof: First notice that an + an+1 has the same sign as an and the opposite sign as
an−1 by assumptions (ii) and (iii). Thus
 ∞ 
 X 
ak  = |(an+1 + an+2 ) + (an+3 + an+4 ) + · · ·|
 

 
k=n+1

= |an+1 + an+2 | + |an+3 + an+4 | + · · ·

> |an+1 + an+2 |
and
 
∞
 X 
a k  = |an+1 + (an+2 + an+3 ) + (an+4 + an+5 ) + · · ·|
 

 
k=n+1

= |an+1 | − |an+2 + an+3 | − |an+4 + an+5 | − · · ·

< |an+1 |.
The two inequalities in the theorem clearly follow from these two calculations. 
5.22 Example: The alternating harmonic series satisfies the hypothesis of Theo-
rem 5.19 and converges to ln 2 by Example 5.13–1. It, therefore, follows from Theo-
rem 5.21 that the 1000th partial sum of the alternating harmonic series will approx-
imate ln 2 with an error less than 1/1001. (Rather disappointing accuracy!). •
5.23 Theorem: (Integral test remainder) If the sequence (anP ) and the function
f satisfies the hypotheses of the integral test (Theorem 5.7) and if ∞n=1 an converges,
then Z ∞ ∞ Z ∞
X
f (x) dx ≤ ak ≤ f (x) dx
n+1 k=n+1 n

for all n ≥ 1.
84 5 Infinite series

Proof: If m and n are any two integers satisfying m > n ≥ 1, then the inequality (5.2)
implies
Z m m−1
X Z k+1
f (x) dx = f (x) dx
n+1 k=n+1 k
m−1
X m−2
X
≤ ak = ak+1
k=n+1 k=n
m−2
X Z k+1 Z m−1
≤ f (x) dx = f (x) dx.
k=n k n

The desired inequality now follows by taking the limit as m tends to infinity. 

5.24 Example: What can we say about the error when the series ∞ 1
P
P10 1 P∞ n=1 n2 is approx-
imated by the sum n=1 n2 ? The error here is n=11 n12 and Theorem 5.23 implies
that Z ∞ ∞ Z ∞
1 1 X 1 1 1
= 2
dx ≤ 2
≤ 2
dx = .
11 11 x n=11
n 10 x 10
We thus have ! ∞
10 10
!
X 1 1 X 1 X 1 1
+ ≤ ≤ + ,
n=1
n2 11 n=1 n2 n=1
n2 10
and, since
10
X 1
1.5497 < < 1.5498,
n=1
n2
it follows that
∞
X 1
1.6406 ≈ 1.5497 + .0909 < < 1.5498 + 0.1000 ≈ 1.6498.
k=1
n2

We can, therefore, say that ∞ 1

P
= 1.6452 with an error of no more than ±0.0046.
P∞ n=11 n2 π2
In fact, the true value of n=1 n2 is 6 ≈ 1.6449. •
The final example in this chapter will illustrate that the ratio test can sometimes
be used to estimate errors.
P n P10 n
5.25 Example: How great is the error if 3n is approximated by n=1 3n ? Observe
that, if n ≥ 10, then 
n+1 n (n + 1) 1 1.1
n+1 n
= <
3 3 n 3 3
and hence
n+1 1.1 n
n+1
< .
3 3 3n
 5.4 Estimating remainders 85

This implies that

 k
10 + k 1.1 10
<
310+k 3 310
for all k ≥ 1 (cf. the proof of the ratio test) and, therefore,
∞ ∞
X n X 10 + k
n
=
n=11
3 k=1
310+k
 2
1.1 10 1.1 10
< 10
+ + ···
3 3 3 310
 
10 1.1 1
= 10
3 3 1 − 1.1/3
< 10−4 . •

The last two examples

P∞illustrate something which was said earlier, following The-
1
orem 5.10. The series n=1 n2 converges slowly in that the ratio and root tests are
inconclusive and, correspondingly, the 1000th
P partial sum
P∞is not a good approxima-
tion. On the other hand, the two series ∞ 1
n=1 2n and n
n=1 3n converge rapidly in
that the ratio and root tests do predict convergence and, correspondingly, even the
10th partial sum is a good approximation.
86 5 Infinite series

Exercises
E5.1 Determine
P∞ whether the following series converge or diverge:
2
(a) √ 2n +3 P∞ n!
n=1 n2 +3n+2
P∞ 1·2·3·...·n
(e) n=1 nn
(b) n=1 3·5·7·...·(2n+1) P∞ n2 (n+1)n
P∞ (f) n=1 (2n)n
−n2
(c) n=1 ne P∞ e1/n
P∞ (n!)2 1000n (g) n=1 n2
(d) n=1 2n 2

E5.2 Determine whether or not each of the following series is convergent:

P∞ √ √
(a) n=1 ( n + 1 − n)
P∞ n cos(nπ)
(b) n=1 n+1
P∞ −1/2
+ n1 (n6 + ln n)


(c) n=1 n
E5.3 Determine whether each of the following series diverge or converge.
∞
X n
(a) 2
n=1
(n + 1)(n + 2)
∞
X 1
(b)
n=1
n + 6.022 × 1023
∞
X (−1)n+1
(c) √
n=1
n
E5.4 Determine whether or not the series
∞ √ √
X n+1− n
n=1
n

is convergent.
2 −b2
Hint: Use the identity a − b = aa+b .
P∞
E5.5 Determine whether the series n=1 k + k1 (k 5 + ln k)−1/2 converges.

E5.6 Determine whether or not the series ∞

P
n=1 (1 − cos(1/n)) is convergent.
E5.7 Find both the nth partial sum as well as the sum of each of the following
series:
∞
X 1
(a)
n=1
(2n − 1)(2n + 1)
∞
X 1
(b)
n=1
n(n + 1)(n + 2)
E5.8 Determine all the real numbers p for which the series
∞
X 1
k=2
k(ln k)p
converges.
 Exercises for Chapter 5 87

E5.9 In each of the following cases, determine all the values of x ∈ R for which the
series converges.
P∞ n2 n
(a) n=0 n+1 x
P∞
2
x n
(b) n=1 1 + n
E5.10 Determine all the values of x for which the following series converge:
P∞ (n+3)x2n
(a) nn
Pn=1
∞ (n+1)(2x−1)n
(b) n=1 2n
E5.11 Determine all the values of x for which the following series converge:
P∞ (n+3)x2n
(a) nn
Pn=1
∞ (n+1)(2x−1)n
(b) n=1 2n

2
x n
E5.12 Determine all the values of x for which the infinite series ∞
P
n=1 1 + n con-
verges.
E5.13 Recall that the binomial coefficient nk is defined as k!(n−k)!
n!


. Determine all
P∞ 2n n

values of x ≥ 0 for which the series n=0 n x converges. What can you say
about the series when x < 0?
Hint: You may find the following bounds on 2n


n
to be useful. Use the grouping

(2n)! = (1 · 2) × (3 · 4) × · · · × ((2n − 1)(2n))

≤ 22 × 42 × · · · × (2n)2

to deduce that (2n)! ≤ 4n (n!)2 , and so, 2n

n
≤ 4n . Similarly, use the grouping

(2n)! = (2 · 3) × (4 · 5) × · · · × ((2n − 2)(2n − 1)) × (2n)

to deduce that 2n 1


n
≥ ( 2n )4n .
E5.14 If x > y > 0, does the series ∞ 1
P
n=1 xn −y n converge or diverge? Why?
E5.15 (a) PProve that, if (an ) is a sequence of positiveP numbers such that the series
∞ ∞ 2
a
n=1 n converges, then so does the series n=1 an .
(b) Show by example that the converse of (a) is false.
P∞
(c) Give an example of
P∞ 2 a sequence (a n ) such that the series n=1 an converges
but the series n=1 an diverges.
E5.16 Let (ak ) be a sequence of nonnegative real numbers.
P∞ 2
(a) PShow that, if the series k=1 ak converges, then so does the series
∞ ak
k=1 k .
Hint: Apply the Cauchy–Schwartz inequality
  v  v 
Xn  u n
X
u n
X
u u
xk yk  ≤ t x2k  t yk2 
 

 
k=1 k=1 k=1
P∞ ak
to the partial sums of k=1 k .
88 5 Infinite series

(b) Is the converse also true in general?

E5.17 Show that, if 0 < r < 1, then it is not true that n−2 ≤ rn for infinitely many
positive integers n. Explain why, in view of this fact, neither
Pthe ratio test nor
∞
the root test could possibly be used to show that the series n=1 n12 converges.
E5.18 Consider the series
∞
X 1 1 1
ak = + 2 2 + 3 + 24 + 5 + 26 + · · · .
n=1
10 10 10

√
(a) Show that limn→∞ n an does not exist.
√
(b) Find lim supn→∞ n an .
(c) Does the series converge or diverge?
E5.19 (a) Show that, if (an ) is a sequence of positive numbers, then
√ an+1
lim sup n
an ≤ lim sup .
n→∞ n→∞ an
√
Hint: Show that, if L > lim supn→∞ an+1an
, then L ≥ lim supn→∞ n an .
(b) Explain the significance of this inequality with respect to the relative
strengths of the ratio and root tests.
n ∈ Z>0 and if the series ∞
P
E5.20 If an >
P0∞ for aall n=1 an , diverges prove that the
series n=1 1+an diverges.
n

E5.21 Let (ak ) be a sequence of positive

P∞ real k−1 numbers that is nonincreasing.
Show that the alternating series k=1 (−1) ak is convergent if and only if
limk→∞ ak = 0.
Hint: For the ⇐ direction, show that the sequence of partial sums (Sn ) is
Cauchy by generalizing the argument of Example 2.24.
E5.22 The integral test shows that the harmonic series ∞ 1
P
k=1 k is divergent. Consider
the sequence defined by
 
1 1 1
an = 1 + + + · · · + − ln n.
2 3 n

Show that (an ) is a bounded monotone sequence, and hence is convergent.

Hint: It may be useful to draw a figure illustrating the integral test. Express
an in terms of areas in this figure.
(The limit limn→∞ an is called Euler’s constant (or sometimes, the Eu-
ler–Mascheroni constant), and is usually denoted by γ. Its value is approxi-
mately 0.5772 . . .. It is not known whether γ is rational or irrational!)
E5.23 Determine whether the following series converge absolutely, converge condi-
tionally, or diverge:
P∞ n
(a) n=1 (−1) (1 − cos(1/n))
 Exercises for Chapter 5 89
P∞ n

n
(b) n=1 (−1) n1/n −1
P∞ (n+3) cos nπ
(c) n=1 1003n
E5.24 Determine whether the following series converge absolutely, converge condi-
tionally, or diverge.
∞ ∞
X
n
X cos n
(a) (−1) (n sin(1/n) − 1) (c)
n=1 n=1
n2
∞ ∞
X cos n X (−1)n
(b) (d)
n=1
2n n=2
ln n
E5.25 Determine whether each of the following series converges absolutely, converges
conditionally, or diverges.
P∞ k ln k
(a) k=1 (−1) k
(b) 12 − 13 + 14 − 312 + 16 − 313 + 18 − 314 + . . .
Hint: Consider the partial sums S2n
E5.26 Determine whether each of the following series converges absolutely, converges
conditionally, or diverges.
P∞ k 1
(a) k=2 (−1) k(ln k)
(b) 1 − 12 + 13 − 212 + 15 − 213 + 17 − 214 + · · ·
Hint: Consider the partial sums S2n .
P∞
E5.27 Prove
P∞ n+1
that, if the series n=1 an is absolutely convergent, then so is the series
n=1 n
a n .
Suppose that a series ∞
P P∞
E5.28 n=1 an converges absolutely
P∞ and that a series n=1 bn
converges conditionally. Prove that the series n=1 an bn converges. Does this
series converge conditionally or absolutely?
E5.29 Consider two sequences (an ) and P (bn ) and suppose that an 6= 0 and that bn > 0
for all n. Show that if the series ∞ n=1 bn converges and if
 
 an+1  bn+1
 an  ≤ b n
 

for all n then the series ∞

P
n=1 an converges absolutely.
P∞
E5.30 Evaluate n=10 e−n and compare its value with the lower and upper estimates
obtained from the integral test.
E5.31 Verify that each of the following two series converge, and obtain lower and
upper estimates for the sums.
P∞ n
(a) n=1 2n (n+3)
P∞ −n
(b) n=1 ne
E5.32 (a) Give an example of a continuous function f with domain [1, ∞) and with
the property that the limit
Z n
lim f (x) dx
n→∞ 1
90 5 Infinite series

exists, but the improper integral

Z ∞
f (x) dx
1

does not exist.

(b) Prove that, if f is a nonnegative-valued continuous function with domain
[1, ∞) such that the limit
Z n
lim f (x) dx
n→∞ 1

exists, then the improper integral

Z ∞
f (x) dx
1

converges and Z n Z ∞
lim f (x) dx = f (x) dx.
n→∞ 1 1
E5.33 Suppose we have an inexhaustible supply of uniform rectangular cards. Show
that, given any distance d, a finite number of these cards can be stacked on a
table so that the outside edge of the top card projects a distance d beyond the
edge of the table.
 Chapter 6
Sequences of functions

In this chapter we are going to examine what it means for a sequence of functions
to converge to a function. This turns out not to be as straightforward as one might
think but, at the same time, is a very useful and powerful concept in analysis, applied
mathematics, and geometry. We will begin with an example that illustrates both
the use of convergence of functions and points out some of subtleties involved in a
discussion of such convergence.

6.1 Motivating example

In the study of ordinary differential equations, one needs to know that, under
appropriate conditions, the initial value problem

y 0 = f (x, y)
y(x0 ) = y0

has a unique solution y(x) defined for x near x0 .

The first step towards a proof of such an existence and uniqueness theorem is to
observe that a function y is a solution of the initial value problem if and only if it
satisfies the integral equation
Z x
y(x) = y0 + f (t, y(t)) dt. (6.1)
x0

(This equivalence is easy to check using the fundamental theorem of calculus.) The
point of replacing the original initial value problem by the integral equation is simply
that integration is much easier to deal with than differentiation.
We will solve this integral equation by finding a sequence of approximate solutions
that converges to the actual solution. The initial approximation is the constant
function y0 satisfying y0 (x) = y0 for all x, and the successive approximations y1 , y2 , . . .
92 6 Sequences of functions

are then defined as follows:

Z x
y1 (x) = y0 + f (t, y0 (t)) dt,
Zx0x
y2 (x) = y0 + f (t, y1 (t)) dt,
x0
..
.
Z x
yn (x) = y0 + f (t, yn−1 (t)) dt.
x0

The question now, of course, is whether this sequence of functions converge (in some
sense) to a function y and, if so, whether this limit function is a solution of (6.1). If
we believe that the sequence (yn ) converges to a function y, it is tempting to argue
as follows:

y(x) = lim yn (x)

n→∞
 Z x 
= lim y0 + f (t, yn−1 (t)) dt
n→∞ x0
Z x
= y0 + lim f (t, yn−1 (t)) dt
n→∞ x
0
Z x
= y0 + lim f (t, yn−1 (t)) dt
x0 n→∞
Z x
= y0 + f (t, lim yn−1 (t)) dt
x0 n→∞
Z x
= y0 + f (t, y(t)) dt.
x0

There are two questionable steps here, in the fourth and fifth lines. The fourth
line involves interchanging the limit and the integral and is not always valid. The
conditions under which this is valid is one of the topics to be studied in this chapter
(cf. Theorem 6.9). The fifth line is also contentious although it is clearly related to
the continuity of f . But, at any rate, if this calculation can be justified, then y is a
solution of the given initial value problem.
Many of the classical methods of solving ordinary and partial differential equations
involve series of functions (for example, power, Fourier, and Bessel series). To under-
stand and justify the use of such methods, it is necessary to understand convergence
of sequences and series of functions.

6.2 Pointwise convergence

Suppose that I is an interval in R and that f1 , f2 , . . . are real-valued functions
defined on I. (The interval may be open, closed, or neither, and may be of finite or
 6.2 Pointwise convergence 93

infinite length.) Then, for each point x ∈ I, we have a sequence (fn (x)) of numbers.
In fact, we can think of fn (x) as a function of the two variables n ∈ Z>0 and x ∈ I,
and so examining this sequence for a fixed value of x is not unlike examining a
function of two real variables along, say, a vertical line in the plane. For each fixed
value of
P∞ x ∈ I, we can ask whether the limit limn→∞ fn (x) exists or whether the
series n=1 fn (x) converges. The point here is that, since x is fixed, the discussion
of convergence of sequences in Chapter 2 and of series in Chapter 5 can be brought
to bear on these questions. Since this sequence and series are being considered for
a fixed value of x, the adjective “pointwise” is attached to these questions and their
answers to distinguish them from a related situation we will consider later in this
chapter.
6.1 Definition: Let I ⊆ R be an interval and suppose that (fn ) is a sequence of
real-valued functions defined on I. We will say that
(i) the sequence (fn ) converges pointwise on I if the limit limn→∞ fn (x) exists
for each point x ∈ I and that
(ii) the series ∞
P P∞
n=1 fn converges pointwise on I if the series n=1 fn (x) con-
verges for each point x ∈ I. •
Suppose that (fn ) is a sequence of functions defined on an interval I ⊆ R. If
this sequence converges pointwise on I, we can define a function f on I by letting
f (x) = limn→∞ fn (x) for all x ∈ I,P and, in this case, we will write “limn→∞ fn = f
∞
pointwise”. Similarly, if the series P n=1 fn converges pointwise on I, we can define
a function fPon I by letting f (x) = ∞ n=1 fn (x) for all x ∈ I, P
and in this case, we
“ ∞
will write P f
n=1 n = f pointwise”. Notice that, if we let gn = n
k=1 fk for n ∈ Z>0 ,
∞
then f = n=1 fn pointwise if and only if f = limn→∞ gn pointwise.
According to Definition 2.4, this means that the following condition holds.
6.2 Lemma: A sequence (fn ) of functions on an interval I ⊆ R converges pointwise
if and only if the following condition holds:
for each  > 0 and for each x ∈ I, there is an N ∈ Z>0 such that
|fn (x) − f (x)| <  for all n ≥ N .
Note that here N is chosen after  and x are selected, and hence that N depends
on  and x, i.e., N = N (, x).
Let us apply these definitions in an example.
6.3 Example: It is easy to see that
(
1, x = 1,
lim xn =
n→∞ 0, 0 ≤ x < 1.

So, if fn (x) = xn , for x ∈ [0, 1] and if

(
1, x = 1,
f (x) =
0 0 ≤ x < 1,
94 6 Sequences of functions

then limn→∞ fn = f pointwise on [0, 1]. Notice that each fn is continuous and that
the limit function f is discontinuous.
If  and x are two numbers in the open interval (0, 1), then |fn (x) − f (x)| = xn ,
ln 
and thus |fn (x) − f (x)| <  if and only if n > ln x
. So the smallest possible value of
ln 
N satisfying Lemma 6.2 is the smallest integer larger than ln x
. Thus N does indeed
depend on both  and x and, for a fixed  ∈ (0, 1), limx→1− N (, x) = ∞. •
Pointwise convergence is a very weak notion of convergence and consequently not
a very useful one. That is to say, one can draw very few conclusions from the assertion
that limn→∞ fn = f pointwise on I. For example, the preceding example shows that
1. f need not be continuous, even if each of the fn is C∞ ,
and the next few examples will show that
2. f need not be integrable, even if each of the fn is integrable,
3. if f and each of the fn is continuous, then it is not necessarily true that
Z b Z b
lim fn (x) dx = f (x) dx,
n→∞ a a

where I = [a, b], and

4. if f and each of the fn is C 1 , it is not necessarily true that
lim fn0 = f 0 pointwise.
n→∞

6.4 Example: For each n ∈ Z>0 , let an be a positive number and let fn be the
function with domain [0, 1] which is zero on [1/n, 1] and whose graph over [0, 1/n] is
an isosceles triangle with base [0, 1/n] and height an . Formally,

2nan x,
 0 ≤ x ≤ 1/(2n),
fn (x) = 2an − 2nan x, 1/(2n) ≤ x ≤ 1/n,

0, 1/n ≤ x ≤ 1.


The sequence (fn ) converges pointwise to the zero function on [0, 1], regardless of the
choice of the sequence (an ). Indeed, it is certainly true that fn (0) = 0 for all n. And,
if x ∈ (0, 1] and if N ∈ Z>0 such that 1/N < x, then fn (x) = 0 for all n ≥ N , and,
therefore, limn→∞ fn (x) = 0. Notice that, just as in the preceding example, the N of
Lemma 6.2 appears to depend on both x and .
It is easy to see that, by choosing the sequence (an ) appropriately, the limit
Z 1
an
lim fn (x) dx = lim
n→∞ 0 n→∞ 2n

can be made to equal 0, 1, or ∞ by taking, for example, an = 1, an = 2n, or an = n2 ,

respectively. At any rate, it is not necessarily true that
Z 1 Z 1 
lim fn (x) dx = lim fn (x) dx. •
n→∞ 0 0 n→∞
 6.2 Pointwise convergence 95

The example just presented, where pointwise convergence of functions does not
imply the convergence of their integrals, is perhaps artificial in that the functions are
piecewise linear, but not C∞ . The next example exhibits this same phenomenon with
C∞ functions.
n
6.5 Examples: 1. If fn (x) = nx(1 − x2 ) and gn (x) = nfn (x) for x ∈ [0, 1], and
n ∈ Z>0 , then
lim fn (x) = 0, lim gn (x) = 0
n→∞ n→∞

pointwise on [0, 1] and

Z 1 Z 1
1
lim fn (x) dx = lim gn (x) dx = ∞
n→∞ 0 2 n→∞ 0

by l’Hôspital’s rule. The verification of these claims is left to the reader.

2. If r1 , r2 , . . . is an enumeration of Q (that is, a list of all the rational numbers) and
if (
1, x ∈ {r1 , r2 , . . . , rn }
fn (x) =
0, otherwise,
then limn→∞ fn = f pointwise, where
(
1, x ∈ Q,
f (x) =
0, x ∈
/ Q.

Here each of the fn is integrable and satisfies

Z 1
fn (x) dx = 0,
0

but the limit f is not even Riemann integrable.

3. The series ∞ x2
P
n=0 (1+x2 )n obviously converges to 0 for x = 0, and for x 6= 0 is a
multiple of a convergent geometric series whose value can be calculated as follows:
∞ ∞
X x2 X
=x 2
(1 + x2 )−n
n=0
(1 + x2 )n n=0
1
= x2
1 − (1 + x2 )−1
= 1 + x2 .
P∞ x2
So the series n=0 (1+x2 )n converges pointwise for all x ∈ R to the function f
defined by (
0, x = 0,
f (x) = 2
1+x , x=6 0.
96 6 Sequences of functions

4. Notice that, in the preceding example, the partial sums of the series are C∞
functions, but that the pointwise limit of the series is not even continuous. Let’s
explore this a little more deeply.
Let fn (x) = sin√nx
n
for n ∈ Z>0 and x ∈ R. Then limn→∞ fn = 0 pointwise on R
since |sin nx| ≤ 1. In fact, if  > 0 is given and if N is an integer with 1/N < ,
then |fn (x)| <  for all n ≥ N and all x. So, in this example, the N of Lemma
√ 6.2
depends only on , in contrast to several earlier examples. Here fn0 (x) = n cos nx,
and thus limn→∞ fn0 (x) = ∞ for every x, and hence the sequence (fn0 ) does not
converge pointwise. •

6.3 Uniform convergence

Example 6.3 and the Examples 6.5 illustrate the fact that pointwise convergence
of a sequence (fn ) of functions does not guarantee that the limit function inherits
desirable properties from the terms fn . Example 6.4 clearly illustrates the weakness
of the concept of pointwise convergence: if an = 1 for all n, then limn→∞ fn = 0
pointwise on [0, 1], but there is no integer n for which the graph of fn is close to
that of the limit function 0 on the whole interval [0, 1]. The definition of uniform
convergence captures the idea that, for each large value of n, the number fn (x)
should be close to f (x) for all x ∈ [0, 1].
6.6 Definition: Let I be an interval and suppose that f, f1 , f2 , . . . are real-valued
functions defined on I. (The interval may again be open, closed, or neither, and may
be of finite or infinite length.)
(i) We will say that the sequence (fn ) converges uniformly to f on I and write
“limn→∞ fn = f uniformly on I” if, for each  > 0, there is an integer N ∈ Z>0
such that |fn (x) − f (x)| <  for all n ≥ N and all x ∈ I.
P∞
(ii) We willPsay that the series n=1 fn convergesPn uniformly to f on I and
write “ ∞ f
n=1 n = f uniformly on I” if limn→∞ k=1 fk = f uniformly on I (as
defined
Pn in (i)), i.e., if, for each  > 0, there is an integer N ∈ Z>0 such that
| k=1 fk (x) − f (x)| <  for all n ≥ N and all x ∈ I. •
If a sequence of functions converges uniformly on an interval, then it certainly
converges pointwise on that interval.
Note the subtle difference between this definition and that of pointwise conver-
gence. In the definition of pointwise convergence, N was selected after x and  were
specified, and so N can depend on both x and  (cf. Lemma 6.2). In the definition of
uniform convergence, N is selected after only  is specified, and so N can depend on
 but must be independent of x.
There are two helpful ways to think about Definition 6.6. Firstly, the condition
that |fn (x) − f (x)| <  for all x ∈ I can be rewritten as

f (x) −  < fn (x) < f (x) + , x ∈ I,

 6.3 Uniform convergence 97

and interpreted as saying that the graph of fn lies in a “tubular” neighbourhood of

radius  centered along the graph of f . And, secondly, the condition that |fn (x) −
f (x)| <  for all x ∈ I means that the maximum value of fn − f on I is less than 
and that the minimum value of fn − f on I is more than −. It is, therefore, possible
to verify the condition in Definition 6.6 by either graphing functions or solving one-
variable max/min problems.
6.7 Examples: 1. If (an ) and (fn ) are as in Example 6.4, then limn→∞ fn = 0 uni-
formly on [0, 1] if and only if limn→∞ an = 0. Indeed, this follows from the above
discussion and the fact that an is the maximum value of fn = fn − 0 on [0, 1].
2. The sequence considered in Example 6.3 does not converge uniformly on [0, 1].
This can be seen either by examining the graphs of the functions in question or
by using Theorem 6.8 below.
3. Let (fn ) be the sequence considered in Example 6.5–4. Then
 
 sin nx  1
|fn (x) − 0| =  √  ≤ √ < 
n n

for all x ∈ R and all n > −2 and, therefore, limn→∞ sin√nx n
= 0 uniformly on R and
hence on any interval.
4. This example is a continuationP∞ of Example 6.5–3. If f is as in that example and if
x2
I is a closed interval, then n=0 (1+x2 )n = f (x) uniformly on I if and only if 0 6∈ I.
To see that this it is necessary, in view of the definition of uniform convergence
of a series, to estimate the partial sums. First of all, if x 6= 0 we have
n n
!
X x2 X
2 )k
= (1 + x2 )−k x2
k=0
(1 + x
 k=0
1 − (1 + x2 )−n−1

= x2
1 − (1 + x2 )−1
= 1 + x2 − (1 + x2 )−n

and, therefore,
  (
n 2
X x  0, x = 0,
− f (x)=
 
2
(1 + x ) k 2 −n


k=0
 6 0.
(1 + x ) , x =

Suppose first that 0 ∈ I and assume that 0 is not √ the right end point of I. Then
I = [a, b], where a ≤ 0 < b, and the number xn = 21/n − 1 belongs to I at least
for sufficiently large values of n. Now
 n 
X x2n 
−n 1
− f (x )  = 1 + x2n − 1 + x2n − (1 + x2n ) =
 
2 k n
(1 + xn ) 2

 
k=0
98 6 Sequences of functions

for all sufficiently large values of n by the above calculation, and the formula for
f contained in Example 6.5–3. A similar argument will show that the series does
not converge uniformly if 0 is not the left end point of I.
Now suppose that 0 6∈ I and assume, to be definite, that I consists of positive
numbers. Then I = [a, b], where 0 < a < b, and
 
n 2
X x 
− f (x) = (1 + x2 )−n ≤ (1 + a2 )−n
 
2
(1 + x ) k

 
k=0

for all x ∈ I and all n ∈ Z>0 . Now limn→∞ (1 + a2 )−n = 0 and so, given an  > 0,
there will be an integer N ∈ Z>0 such that (1 + a2 )−N < . But this means that
 n 
2
X x 
− f (x)<
 
(1 + x2 )k

 
k=0

for all n ≥ N and all x ∈ I and, therefore, the series converges uniformly on I. A
similar argument will evidently show that the series converges uniformly on I if I
consists of negative numbers.
Finally, note that, if 0 ∈ I, then it also follows from Theorem 6.8 (below), and
the fact that f is discontinuous at 0, that the convergence is not uniform on I. •

6.4 Properties of uniform convergence

The examples of Section 6.2 show that pointwise convergence is a very weak prop-
erty in that there cannot be any theorems relating pointwise convergence to continu-
ity, integration, and differentiation. The point of this section is to show that uniform
convergence is a stronger and much more useful assumption than pointwise conver-
gence; we will do this by proving some nice theorems relating uniform convergence
to continuity, integration, and differentiation.
Examples 6.3 and 6.5–3 show that a pointwise limit of continuous functions need
not be continuous. The first theorem of this section shows that a uniform limit of
continuous functions is necessarily continuous.
6.8 Theorem: Suppose that (fn ) is a sequence of continuous functions defined on an
interval I. If there is a function f on I such that limn→∞ fn = f uniformly on I
then f is continuous on I.
Proof: Recall that, by definition, f is continuous on I if it is continuous at each point
of I. So, to prove the theorem, we must show that, corresponding to each point
x0 ∈ I and each number  > 0, there is a number δ > 0 such that |f (x) − f (x0 )| < 
for all x ∈ I satisfying |x − x0 | < δ. First note that

|f (x) − f (x0 )| = |f (x) − fn (x) + fn (x) − fn (x0 ) + fn (x0 ) − f (x0 )|

≤ |f (x) − fn (x)| + |fn (x) − fn (x0 )| + |fn (x0 ) − f (x0 )|
 6.4 Properties of uniform convergence 99

for all x ∈ I and all n ∈ Z>0 . Now, since limn→∞ fn = f uniformly on I, there is an
integer N ∈ Z>0 such that |f (x) − fn (x)| <  for all x ∈ I and all n ≥ N and then,
since fN is continuous, there is a number δ > 0 such that |fN (x) − fN (x0 )| < /3 for
all x ∈ I satisfying |x − x0 | < δ. But then, if x is any point in I satisfying |x − x0 | < δ
and if n = N , each of the three summands in the last line of the above estimate is
less than /3 and, therefore,
|f (x) − f (x0 )| < 
for all such x. 
While Example 6.5–1 showed that we cannot interchange a pointwise limit and
an integral, the next result shows that we can interchange a uniform limit and an
integral.
6.9 Theorem: Suppose that (fn ) is a sequence of continuous functions defined on an
interval [a, b] and assume that there is a function f on I such that limn→∞ fn = f
uniformly on [a, b]. Then
Z b Z b
lim fn (x) dx = f (x) dx.
n→∞ a a

Proof: Recall from Theorem 6.8 that f is continuous and hence that all of the integrals
in the statement of the theorem exist. It is necessary to show that, given an  > 0,
there is an integer N ∈ Z>0 such that
Z b Z b 
 

 fn (x) dx − f (x) dx < 
a a

for all n ≥ N . Now, since limn→∞ fn = f uniformly on [a, b], it follows that there is

an integer N ∈ Z>0 such that |fn (x) − f (x)| < 2(b−a) for all x ∈ [a, b] and all n ≥ N .
But then
Z b Z b  Z b
 

 fn (x) dx − f (x) dx ≤
 |fn (x) − f (x)| dx
a a a
Z b

≤ dx
a 2(b − a)

= <
2
for all n ≥ N . 
We already know from Example 6.5–4 that, if limn→∞ fn = f pointwise, then it
does not follow that limn→∞ fn0 = f 0 pointwise and, in analogy with the preceding
theorem, we might hope to be able to prove that, if limn→∞ fn = f uniformly, then
limn→∞ fn0 = f 0 uniformly. But Exercise 6.7 and the next example both show that this
is false in general, although Theorem 6.11 shows it is true under certain hypotheses.
The point here is that Z x
g(x) = f (t) dt
a
100 6 Sequences of functions

is a more smooth function than f , but that f 0 is a less smooth function. (For instance,
if f is, say, C4 , then g is C5 , whereas f 0 is C3 .) The effect of this is that, in the
above theorem on integration, it was not necessary to make assumptions beyond
those involving the functions themselves, whereas in the theorem on differentiation
(see below) it is necessary to make assumptions about both the functions and their
derivatives.
x
6.10 Example: Let fn (x) = for n ∈ Z>0 and x ∈ [−1, 1]. It follows easily
1 + n 2 x2
from elementary calculus1 that
1 1
− ≤ fn (x) ≤
2n 2n
for all n ∈ Z>0 and all x ∈ [−1, 1], and, therefore, limn→∞ fn = 0 uniformly on [−1, 1].
On the other hand,
0 1 − n 2 x2
fn (x) =
(1 + n2 x2 )2
and so (
1, x = 0,
lim fn0 (x) =
n→∞ 0, 0 < |x| ≤ 1.
So here limn→∞ fn0 6= (limn→∞ fn )0 and, moreover, since limn→∞ fn0 is not continuous,
the sequence (fn0 ) does not converge uniformly by Theorem 6.8. •
The derivative of a function defined on an interval is generally only considered at
interior points of that interval and, for this reason, a function is usually only described
as being C1 if its domain is an open interval. For the purpose of the next theorem and
corollary, it will be convenient to modify this terminology a little. Namely, suppose
that h is a function defined on a closed interval [a, b]. Then, for this theorem and
corollary only, we will let

the right derivative of h at a, x = a,

0
h (x) = the derivative of h at x, a<x<b

the left derivative of h at b, x = b.


In addition, we will say that h is C1 on [a, b] if the function h0 as just defined is

continuous on [a, b].
6.11 Theorem: Suppose that (fn ) is a sequence of functions defined on an interval
[a, b] and assume that
(i) fn is C1 on [a, b] for each n ∈ Z>0 ,
(ii) there is a function g defined on [a, b] such that limn→∞ fn0 = g uniformly on
[a, b], and
1
One computes (1) that the derivative of fn vanishes at x if and only if x = ±1/n (2) that x =
−1/n is a minimum and 1/n is a maximum of fn , and (3) fn (−1/n) = −1/2n and fn (1/n) = 1/2n.
 6.4 Properties of uniform convergence 101

(iii) there is a point c ∈ [a, b] for which the limit limn→∞ fn (c) exists.
Then there is a function f on [a, b] such that limn→∞ fn = f uniformly on [a, b] and
f 0 = g.
Proof: The function f that is asserted to exist must satisfy the equation
Z x Z x
f (x) = f (a) + g(t) dt = lim fn (a) + g(t) dt
a n→∞ a
by the fundamental theorem of calculus. (Notice that g is continuous by assump-
tions (i) and (ii), and Theorem 6.8 and hence is integrable.) This suggests that we
attempt to carry out the proof by
1. first showing that the limn→∞ fn (a) exists and defining a function f on [a, b] by
the formula Z x
f (x) = lim fn (a) + g(t) dt, (6.2)
n→∞ a
2. showing that limn→∞ fn = f uniformly on [a, b], and
3. showing f 0 = g.
The proof of 1 is easy. We know that
Z c
fn (c) − fn (a) = fn0 (t) dt
a
for all n ∈ Z>0 by the fundamental theorem of calculus and that limn→∞ fn0 = g
uniformly on [a, c], and so
 Z c 
0
lim fn (a) = lim fn (c) − fn (t) dt
n→∞ n→∞ a
Z c
= lim fn (c) − g(t) dt
n→∞ a
by Theorem 6.9.
We can now use (6.2) to define a function f on [a, b].
Since g is continuous, the definition of f and the fundamental theorem of calculus
imply that f satisfies 3. To show that f also satisfies 2, suppose that  > 0 is given.
Then (by (ii)) there will be an integer N ∈ Z>0 such that |fn0 (x) − g(x)| < 2(b−a) 

for all x ∈ [a, b] and all n ≥ N and such that |fm (a) − limn→∞ fn (a)| < /2 for all
m ≥ N . It now follows from the fundamental theorem of calculus and the definition
of f that
 Z x   Z x 
0
 
|fm (x) − f (x)| =  fm (a) +
 fm (t) dt − lim fn (a) + g(t) dt 
a n→∞ a
  Z x 
0

≤ fm (a) − lim fn (a) +  (fm (t) − g(t)) dt
  
n→∞ a
Z x
 0
< + |fm (t) − g(t)| dt
2 a
 
≤ + =
2 2
102 6 Sequences of functions

for all x ∈ [a, b] and all m ≥ N . 

The hypotheses in this theorem seem somewhat technical but they are often easy
to verify and so the theorem is useful. The next corollary is really just a restatement
of a special case of the theorem and so does not require a proof.
6.12 Corollary: Suppose that (fn ) is a sequence of C1 functions defined on an in-
terval [a, b] and assume that there are two functions f and g, also defined on [a, b],
such that limn→∞ fn = f uniformly on [a, b] and limn→∞ fn0 = g uniformly on [a, b].
Then f 0 = g.
The condition (iii) in the above theorem is not superfluous in that there are
sequences of functions which satisfy (i) and (ii), but not the conclusion of the theorem.
The simplest such sequence is the one in which fn (x) = n for all x and all n. This
sequence certainly satisfies (i) and (ii), but not (iii) and does not converge pointwise.
This concludes the discussion of the properties of uniform convergence of sequences
of functions. But what about uniform convergence of series of functions? In principle,
there is nothing new here since a series is just the limit of its sequence of partial
sums. But, in applying the three preceding theorems, it would be desirable to have
a criteria for deciding when a series of functions converges uniformly, and there is a
such a criteria.
6.13 Theorem: (Weierstrass M -test) Suppose that (fn ) is a sequence of functions
defined on an interval I and assume that there is a sequence (Mn ) of numbers such
that
(i) |fn (x)| ≤ Mn for all x ∈ I and all n ∈ Z>0 and
(ii) the series ∞
P
Mn converges.
Pn=1
∞
Then the series n=1 fn converges uniformly on I.
Proof: Put Sn (x) = nk=1 fk (x) and Tn = nk=1 Mk for n ∈ Z>0 and x ∈ I. Then
P P
the sequence (Tn ) is clearly nondecreasing and it converges by assumption (ii), and
hence is bounded by Proposition 2.6. Moreover, if 1 ≤ n < m, then
 
m
 X 
|Sm (x) − Sn (x)| =  fk (x)
 
 
k=n+1
Xm
≤ |fk (x)|
k=n+1
Xm
≤ Mk = Tm − Tn
k=n+1

for all x ∈ I. Now, since (Tn ) is a Cauchy sequence, this calculation implies that
(Sn (x)) is a Cauchy sequence for each x ∈ I and, therefore, that the sequence (Sn )
converges pointwise by Theorem 2.23. So, if we put f (x) = limn→∞ Sn (x) for each
 6.5 The sup-norm 103

x ∈ I, then f = ∞
P
n=1 fn pointwise on I, and it remains to show that this convergence
is actually uniform. Now, if x is any point in I, then
 
 Xn 
f (x) − fn (x) = |f (x) − Sm (x) + Sm (x) − Sn (x)|
 
 
k=1
≤ |f (x) − Sm (x)| + |Sm (x) − Sn (x)|
≤ |f (x) − Sm (x)| + Tm − Tn

for any m ≥ n and, therefore,

 n

 X 
f (x) − fn (x) ≤ lim (|f (x) − Sm (x)| + Tm − Tn )
 
  m→∞
k=1
= lim Tm − Tn .
m→∞

Now, if a number  > 0 is given, there will be an integer N ∈PZ>0 such that
n
limm→∞ Tm − Tn <  whenever n ≥ N , and, therefore, P∞ |f (x) − k=1 fn (x)| < 
for all n ≥ N and all x ∈ I. This makes it clear that n=1 fn = f uniformly on I. 

6.14 Example: Consider the series ∞ 1

P
n3
n=1 P sin(nx) (this is an example of a Fourier
sin nx n
series). Put un (x) = n3 and Sn (x) = k=1 uk (x) for n ∈ Z>0 and x ∈ R. The
series ∞   X ∞
X d 1 1
3
sin(nx) = cos(nx)
n=1
dx n n=1
n2
converges uniformly on R by the Weierstrass M -test since
 
1
 cos(nx) ≤ 1

 n2  n2

the series ∞ 1
P
and sinceP n=1 n2 converges by Example 5.8.
∞ 1
Now n=1 n3 sin(nx) converges to 0 when x = 0. Hence, by Theorem 6.11, we
conclude that the series may be differentiated term-by-term, and
∞ ∞  
d X 1 X d 1
sin(nx) = sin(nx) . •
dx n=1 n3 n=1
dx n3

6.5 The sup-norm

Suppose that I is an interval (which may be open, closed. or neither, and of
finite or infinite length) in R. Recall that a function f with domain I is said to
be bounded if there is a number M such that |f (x)| ≤ M for all x ∈ I. Let B(I)
denote the set consisting of all the bounded functions with domain I and C(I) the
set consisting of all the continuous functions with domain I. In general, neither of
104 6 Sequences of functions

these sets of functions is a subset of the other, although if I is closed and bounded
then C(I) ⊆ B(I) by Theorem 4.24(i). Both of these sets of functions are (infinite-
dimensional) vector spaces in the sense that, if f and g are functions in one of these
sets and if a is a number, then the functions f + g and af defined by the equations

(f + g)(x) = f (x) + g(x)

and
(af )(x) = af (x)
also belong to that set. This is known in the case of C(I) (since linear combinations
of continuous functions are again continuous) and in the case of B(I) can be shown as
follows. If f and g are in B(I) then there are numbers M and N satisfying |f (x)| ≤ M
and |g(x)| ≤ N for all x ∈ I, and consequently

|(f + g)(x)| = |f (x) + g(x)| ≤ |f (x)| + |g(x)| ≤ M + N

and
|(af )(x)| = |af (x)| = |a||f (x)| ≤ |a|M
for all x ∈ I. The origin in both of these vector spaces is the zero-function.
For any function f ∈ B(I), the quantity

kf k = sup{|f (x)| | x ∈ I}

is a finite number called the sup-norm of f on I. (Notice that the sup-norm k·k
depends on I. but that I is not included in the notation.) Thus k·k is a function on
B(I) and it has the three properties:
1. kf k ≥ 0 for all f ∈ B(I) with equality if and only if f = 0;
2. kaf k = |a| kf k for all a ∈ R and all f ∈ B(I);
3. kf + gk ≤ kf k + kgk for all f, g ∈ B(I).
Only the third of these properties is not obvious and it can be proven as follows. If
f and g are two functions in B(I) and if x is a point in I, then

|(f + g)(x)| = |f (x) + g(x)| ≤ |f (x)| + |g(x)| ≤ kf k + kgk

and, therefore,

kf + gk = sup{|(f + g)(x)| | x ∈ I} ≤ kf k + kgk.

The sup-norm on B(I), therefore, has the same basic properties as the usual norm
has on Rd (cf. the beginning of Chapter 4). Now a cursory examination will reveal
that all of the ideas presented in Chapter 4 depended ultimately on the balls Br (x)
and that the definitions of these balls depended only on the norm. This suggests
that it should be possible to carry over to B(I) all of the definitions that were made
for Rd in Chapter 4 and most (but definitely not all) of the results that were proven
 6.5 The sup-norm 105

there. In doing this we must replace Rd by B(I) and regard this as the set we are
studying and we must replace the vectors in Rd by the functions in B(I) and regard
these as the points in the set we are studying. Moreover, the functions that were
studied in Section 4.4 (which all had subsets of Rd as their domains) must now be
replaced by functions defined on subsets of B(I), i.e., functions whose argument is
itself a function. (We already know many examples of such functions: evaluation,
differentiation, and integration are just three such examples).
The definitions of distance between points in B(I) and convergence of sequences
in B(I) are completely analogous to those in Rd .
6.15 Definition: Let I be an interval in R.
(i) The distance between two functions f and g in B(I) is the number |f − g|.
(ii) A sequence (fn ) in B(I) is said to converge to a function f ∈ B(I) if, for each
 > 0, there is an integer N ∈ Z>0 such that kfn − f k <  whenever n ≥ N . •
Consider two functions f and g in B(I) and an  > 0. Now

kf − gk = sup{|f (x) − g(x)| | x ∈ I}

and so, if kf − gk < , then |f (x) − g(x)| <  for all x ∈ I and, conversely, if
|f (x) − g(x)| <  for all x ∈ I, then kf − gk ≤ . In particular, if (fn ) is a sequence in
B(I), this observation takes the following form: If kfn − f k <  for all n ≥ N , then
|fn (x) − f (x)| <  for all x ∈ I and all n ≥ N and, conversely, if |fn (x) − g(x)| < 
for all x ∈ I and all n ∈ N , then kfn − f k ≤  for all n ∈ Z>0 . This just means that
convergence in B(I) in the sense of Definition 6.15 is the same as uniform convergence
on I. This observation can be summed up as follows:
6.16 Proposition: A sequence (fn ) in B(I) converges to a function f ∈ B(I) (in the
sense of Definition 6.15) if and only if limn→∞ fn = f uniformly on I.
Consider again an interval I in R. A sequence (fn ) of functions in B(I) is said to be
pointwise Cauchy if the numerical sequence (fn (x)) is Cauchy for all x ∈ I. With
this definition, it follows from Theorem 2.23 that the sequence (fn ) converges point-
wise on I if and only if it is pointwise Cauchy. The following definition and theorem
capture the corresponding definition and characterization for uniform convergence.
6.17 Definition: Suppose that I is an interval in R. A sequence (fn ) of functions in
B(I) is said to be uniformly Cauchy on I if, for each  > 0, there exists an integer
N such that kfn − fm k <  for all m, n > N . •

6.18 Theorem: Suppose that I is an interval in R. A sequence (fn ) of functions in

B(I) converges uniformly if and only if it is uniformly Cauchy.
Proof: Suppose first that (fn ) converges uniformly on I to a function f . Then, given
 > 0, there will be an integer N such that |fn (x) − f (x)| < /2 for all x ∈ I and all
106 6 Sequences of functions

n ≥ N . But this implies that

|fn (x) − fm (x)| ≤ |fn (x) − f (x)| + |f (x) − fm (x)|

 
< + =
2 2
for all x ∈ I and all m, n ≥ N , hence kfn − fm k ≤  for all m, n ≥ N , and, therefore,
(fn ) is a uniformly Cauchy.
Now suppose, conversely, that (fn ) is a uniformly Cauchy sequence on I. Then
|fm (x) − fn (x)| ≤ kfm − fn k for all x ∈ I and all m, n ∈ Z>0 and, therefore, (fn (x))
is a Cauchy and hence (by Theorem 2.23) a convergent sequence of numbers for each
x ∈ I. This means that the formula f (x) = limn→∞ fn (x) for x ∈ I defines a function
f on I such that limn→∞ fn = f pointwise on I. To complete the proof it is sufficient
to show that this convergence is actually uniform on I.
Let  be a positive number and let x be a point in I. Then, as (fn ) is uniformly
Cauchy, there is an integer N such that kfm − fn k < /2 for all m, n ≥ N . And, as
limn→∞ fn = f pointwise on I, there is an integer M such that |fn (x) − f (x)| < /2
for all n ≥ M . So, if n ≥ N and if m = max{M, N }, then

|fn (x) − f (x)| ≤ |fn (x) − fm (x)| + |fm (x) − f (x)|

≤ kfn − fm k + |fm (x) − f (x)|
 
< + = .
2 2
Since this is true for any point x ∈ I, it follows that kfn − f k ≤  whenever n ≥ N ,
and this just means that limn→∞ fn = f uniformly on I. 
It is possible to use Theorem 6.18 to give a very short proof of the Weierstrass
M -test (Theorem 6.13).
6.19 Example: Let f and fn for n ∈ Z>0 be as in Example 6.3. Then kfn − f k = 1
for all n ∈ Z>0 since, if x is a little less than 1, then fn (x) = xn is almost 1 and
f (x) = 0. More precisely, if we let
 1/n
n
xn =
n+1

for n ∈ Z>0 , then 0 < xn < 1 for each n and

n
kfn − f k ≥ |fn (xn ) − f (xn )| = xnn =
n+1
and so kfn − f k = 1 for each n ∈ Z>0 . This shows once more that limn→∞ fn = f
pointwise, but not uniformly, on [0, 1]. •
 6.5 The sup-norm 107

6.20 Corollary: If I is a closed and bounded interval in R, then every sequence in

C(I) that is uniformly Cauchy converges uniformly to a function in C(I).
Proof: Suppose that (fn ) is a uniformly Cauchy sequence in C(I). Now, since I is
compact, we know from Theorem 4.24pl:cptcont1 that C(I) ⊆ B(I) and hence (fn )
is a uniformly Cauchy sequence in B(I). The preceding theorem now asserts that
there is a function f in B(I) such that limn→∞ fn = f uniformly on I, and then
Theorem 6.8 implies that f is continuous, i.e., belongs to C(I). 
The emphasis on the completeness of the real line is the major difference between
the discussion contained in these notes and that which could or would be given in a
first-year calculus course. As has already been remarked, the following assertions are
equivalent and any one of them can be taken as the statement of the completeness of
the real line:
1. every nonempty subset of R that is bounded above has a least upper bound
(Theorem 1.9);
2. the Bolzano–Weierstrass theorem (Theorem 2.13);
3. every Cauchy sequence in R converges (Theorem 2.23).
In the context of B(I), conditions 1 and 2 no longer make sense, since there is no
suitable order on this set of functions, but the analogue of condition 3 is true by
Theorem 6.18. And, if I is a closed and bounded interval in R, the analogue of 3 is
also true for C(I) by Corollary 6.20.
108 6 Sequences of functions

Exercises
E6.1 Show that each of the following sequences of functions converges pointwise on
[0, 1] to the zero function and determine whether the convergence is uniform
on that interval.
(a) fn (x) = nxe−nx
2
(b) fn (x) = nxe−n x
2
(c) fn (x) = nxe−nx
E6.2 Determine if the following sequences (fn ) of functions on [0, 1] converge point-
wise and/or uniformly.

1
2nx,
 x ∈ [0, 2n ],
1 1
(a) fn (x) = 2 − 2nx, x ∈ ( 2n , n ],

0, x ∈ ( n1 , 1]

sin(1/n2 x)
(
√
n
, x ∈ (0, 1],
(b) fn (x) =
0, x=0
n
X cos(kx)
(c) fn (x) =
k=1
k
E6.3 Let fn (x) = x1 + n1 and let gn (x) = 1 + n1 for n ∈ Z>0 and x ∈ (0, 1).
(a) Show that limn→∞ fn = 1/x uniformly on (0, 1).
(b) Show that limn→∞ gn = 1 uniformly on (0, 1).
(c) Show that limn→∞ fn gn = 1/x pointwise, but not uniformly, on (0, 1).
E6.4 Determine the pointwise limit limn→∞ fn of the given sequence of functions on
the interval [0, 1], and also determine whether the convergence is uniform on
that interval.
(a) fn (x) = 1/(1 + nx)
(b) fn (x) = x/(1 + nx)
E6.5 Compute the pointwise limits of the following sequences of functions on the
indicated interval and explain why the convergence is not uniform.
n 2 xn
(a) Sn (x) = on [0, 2]
1 + n2 xn
1 − n2 x2
(b) Sn (x) = on [−1, 1]
1 + n2 x2
nx2
E6.6 Let fn (x) = 1+nx 2 for x ∈ R and n ∈ Z>0 .

(a) Show that the sequence (fn ) converges pointwise on R.

(b) Show that the sequence (fn ) does not converge uniformly on any closed
interval containing 0.
(c) Show that the sequence (fn ) converges uniformly on any closed interval
not containing 0.
 Exercises for Chapter 6 109

E6.7 Show that ∞ n

P
n=0 x converges uniformly on [−a, a] for every a satisfying 0 <
a < 1, but does not converge uniformly on (−1, 1).
E6.8 Prove that ∞ xn
P
n=1 n! converges uniformly on [−a, a] for each positive real num-
ber a, but does not converge uniformly on R.
E6.9 (a) Show that limn→∞ nxp e−nx = 0 pointwise on [0, 1] for each positive num-
ber p.
(b) Determine all the positive numbers p for which the convergence in (a) is
uniform on [0, 1].
E6.10 Consider the sequence of functions (fn ), where
 πx 
fn (x) = sin
2n
for n ≥ 1 and x ∈ R.
(a) Determine the pointwise limit of this sequence on R.
(b) Show that the sequence converges uniformly on the closed interval [−a, a]
for any a > 0.
(c) Show that the sequence does not converge uniformly on R.
E6.11 Suppose that a sequence of functions (fn ) defined on an interval I satisfies
1. |fn (x) − fn (y)| ≤ |x − y| for any x, y ∈ I and any n ≥ 1 and
2. (fn ) converges pointwise on I to a function f .
Show that f satisfies |f (x) − f (y)| ≤ 2|x − y| for all x, y ∈ I.
Determine all the values of x for which the series ∞ 1
P
E6.12 n=1 (x(2−x))n converges.
Also, determine interval(s) on which this series converges uniformly. On what
set is this series a continuous function?
E6.13 Find a sequence of functions (fn ) defined on [0, 1] such that each fn is discon-
tinuous at each point in [0, 1], but (fn ) converges uniformly to a function f
that is continuous on [0, 1].
Hint: Define f1 : [0, 1] → R by setting f1 (x) to be 1 for x ∈ Q, and 0 for x ∈ /
rational.
E6.14 Prove that, if a sequence (fn ) of functions converges uniformly to a function f
on an interval I, and if a second sequence (gn ) converges uniformly to a second
function g on I then the sequence (fn + gn ) converges uniformly to (f + g)
on I.
E6.15 (a) Prove that, if (fn ) converges uniformly to f on an interval I, if (gn )
converges uniformly to g on I, and if (fn ) and (gn ) are uniformly bounded
on I, then (fn gn ) converges uniformly to f g on I. (Use the following
definition: A sequence (fn ) of functions on I is said to be uniformly
bounded on I if there exists a constant M such that |fn (x)| < M for all
n and for all x ∈ I.)
(b) Give an example to show that, if one of (fn ) or (gn ) is uniformly bounded
and the other is not, then the result of (a) is false.
110 6 Sequences of functions

E6.16 Suppose that f is a real-valued function with domain R and with the property
that |f (x) − f (y)| ≤ K|x − y| for all x, y ∈ R and some K > 0. Suppose also
that (gn ) is a sequence of functions from R to R which converges uniformly
on R to a function g. Prove that the sequence (f ◦ gn ) converges uniformly to
f ◦ g on R.
E6.17 Suppose that (fn ) is a sequence of differentiable functions on [0, 1] that con-
verges uniformly to some function. Must the sequence fn0 converge uniformly?
Hint: Consider the functions fn (x) = (sin(n2 x))/n or gn (x) = xn+1 /(n + 1)
on [0, 1].
E6.18 Recall the series ∞ 2n
P
n
n=0 n x from Exercise E5.13. It was determined there
that the series converges pointwise on (−1/4, 1/4).
P∞ 2n
n
(a) Show that the series n=1 n x converges uniformly on the interval
[−a, a] for any real number a satisfying 0 < a < 1/4.
(b) Set f (x) = ∞ 2n
P
n
n=1 n x forP x ∈ (−1/4,

1/4). Explain why the derivative
0 ∞ 2n n−1
of f (x) is given by f (x) = n=1 n n x for any x ∈ (−1/4, 1/4).
P∞
E6.19 Show that, if the series n=1 un (x) converges uniformly on an interval I, then
limn→∞ un = 0 uniformly on I.
E6.20 (a) Use the Weierstrass M -test to show that the series ∞ −3/2
P
n=1 n cos(x/n)
converges uniformly on R. Let S denote its sum.
(b) Deduce from (a) and Theorem 6.11 that the series ∞ −1/2
P
n=1 n sin(x/n)
converges uniformly on all closed intervals of finite length, and that its
derivative Pis the function S(x) of (a). This theorem does not imply that
the series ∞ n=1 n
−1/2
sin(x/n) converges uniformly on R. Why?
(c) Use the Weierstrass M -test and the inequality P∞ |sin−1/2
u| ≤ |u| (which is valid
for all u ∈ R) to prove that the series n=1 n sin(x/n) converges
uniformly on all closed intervals of finite length.
(d) Does the series ∞ −1/2
P
n=1 n sin(x/n) converges uniformly on R?
E6.21 Let Sn (x) = nx/(nx + 1) for n ≥ 1 and x ∈ [0, 1].
(a) Compute the pointwise limit limn→∞ Sn (x) on [0, 1]. Is the convergence
uniform on this interval?
R1
(b) Compute limn→∞ 0 Sn (x) dx and
Z 1
lim Sn (x) dx.
0 n→∞

E6.22 (a) Consider functions defined on a subset S of R2 . How should the definition
of “sup-norm of a function with respect to S” be defined?
(b) Suppose that (gn ) is a sequence of functions defined a compact subset S
of R2 which converges uniformly on S to a function g. Prove that
ZZ ZZ
lim gn (x, y) dA = g(x, y) dA.
n→∞ S S
 Exercises for Chapter 6 111
P∞ sin nx
E6.23 If f (x) = n=1 n3
, show that
Z π ∞
X 1
f (x) dx = 2 .
0 n=1
(2n − 1)4

Justify each step in your proof by quoting an appropriate theorem.

E6.24 (a) Prove that the series
∞
X 1
n=1
1 + n2 x
converges uniformly on the interval [a, ∞) for any a > 0.
(b) Show that if we put
∞
X 1
f (x) =
n=1
1 + n2 x
for x > 0, then f is a continuous function on (0, ∞).
(c) Explain why the derivative f 0 of f is given by the formula
∞
X n2
f 0 (x) = −
n=1
(1 + n2 x)2

for all x > 0 and deduce that f is strictly decreasing on (0, ∞).
(d) Let fn (x) = nk=1 1+k1 2 x be the nth partial sum of f . Use the fact that
P
limx→∞ fn (x) = 0 for each fixed n ∈ Z>0 to deduce that limx→∞ f (x) = 0.
(e) If fn is as in (d), show that fn (1/n2 ) > n/2 and deduce that
limx→0+ f (x) = ∞.
(f) Use the information in parts (a)–(e) to make a rough sketch of the graph
of f .
E6.25 Show that the series ∞ n−1 1
P
n=1 (−1) x2 +n
is uniformly convergent on R, but is
not absolutely convergent for any x ∈ R.
Hint: Use the Cauchy criterion for uniform convergence.
E6.26 For each n ∈ Z>0 , let fn be the function with domain R defined as follows:
(
x/n, |x| ≤ n
fn (x) =
1, |x| > n.

(a) Determine all the values of x for which the limit limn→∞ fn (x) exists.
(b) Determine the sup-norm kfn k of fn on the interval [−a, a] for each n ∈ Z>0
and each a > 0.
Hint: Consider separately the two cases n > a and n ≤ a.
(c) For which numbers a > 0 does the sequence (fn ) converge uniformly on
the interval [−a, a]?
112 6 Sequences of functions

(d) Does the sequence (fn ) converge uniformly on R?

E6.27 Consider the functions fn (x) = nxn (1 − x) for x ∈ [0, 1] and n ∈ Z>0 .
(a) Find f (x) = limn→∞ fn (x) for x ∈ [0, 1].
(b) Find the sup-norm kfn − f k of fn − f on [0, 1].
(c) Does limn→∞ fn = f uniformly [0, 1]?
Z 1 Z 1
(d) Does lim fn (x) dx = f (x) dx?
n→∞ 0 0
E6.28 Decide whether each of the following statements is true or false and justify
your answers with either a proof or an example.
(a) If limn→∞ fn = f uniformly on an interval I, then limn→∞ fn = f point-
wise on that interval.
(b) If limn→∞ fn = f pointwise on an interval I and if each fn , is of class C2
on I, then limn→∞ fn = f uniformly on I.
(c) If limn→∞ fn = f pointwise on [0, 1] and if
Z 1 Z 1
lim fn (t) dt = f (t) dt,
n→∞ 0 0

then limn→∞ fn = f uniformly on [0, 1].

E6.29 Consider the functions fn defined as follows:
(
1/n, 0 ≤ x ≤ n2 ,
fn (x) =
0, x > n2 .
Show that limn→∞ fn = 0 uniformly on [0, ∞), that the improper integral
Z ∞
fn (x) dx
0
Z ∞
exists for each n ≥ 1, but that lim fn (x) dx = ∞. This seems to contra-
n→∞ 0
dict Theorem 6.9; explain why there is, in fact, no contradiction.
E6.30 When answering the following problems, state clearly either (1) how you use
results from this course or (2) what assumptions that you had to make about
things that do not follow from results in this course.
(a) Show that if (Cn ) is a bounded sequence of real numbers then the series
∞
2
X
u(x, t) = Cn e−n t sin(nx)
n=1

is a solution of the boundary value problem

∂ 2u ∂u
= , u(0, t) = u(π, t) = 0
∂x2 ∂t
for t > 0 and x ∈ [0, π].
 Exercises for Chapter 6 113

(b) Show that if, for t = 0, the series in (a) is uniformly convergent for
x ∈ [0, π], then u(x, t) is continuous on the set

{(x, t) ∈ R2 | 0 ≤ t < ∞ and 0 ≤ x ≤ π}.

Remark: The partial differential equation uxx = ut is the heat equation and
can be derived from conservation of heat and Gauss’s theorem. The function
u is the expansion in a Fourier series of the solution of the heat equation.
 Chapter 7
Power series and Taylor series

An important class of series of functions are the power series, such as arise in the
theory of Taylor series. The theory of Taylor series, then, plays an important role in
complex analysis.1 In this chapter we shall consider the special properties of power
series.

7.1 Definitions and examples

An infinite series of the form
X∞
an (x − c)n = a0 + a1 (x − c) + a2 (x − c)2 + · · · ,
n=0

where x is a real variable, c is a number, and (an ) is a sequence of numbers, is called a

power series. Such series play important roles in numerous areas in analysis: they
are used in approximation theory, they provide an important method of solving dif-
ferential equations, and they are indispensable in the study of functions of a complex
variable.
There are several questions about power series which must be dealt with before
they can be understood. Firstly, for which values of x does a power series converge?
And, secondly, on the set consisting of those x for which the series converges, is the
convergence pointwise or uniform, is the power series a continuous function, and can
we differentiate and integrate the series formally term-by-term? These questions will
be answered systematically in the next two sections; the present section will simply
give some examples to motivate the general discussion in those sections.
P∞ xn
7.1 Examples: 1. Consider the power series n=1 n . Here c = 0, a0 = 0, and
an = 1/n for n ≥ 1. Now
 n+1   n 
x  x 
lim     = |x| lim n = |x|
n→∞ n + 1  n n→∞ n + 1

and so, by the ratio test (Theorem 5.9), the power series converges for |x| < 1
and diverges for x > 1. As we shall see in the proof of Theorem 7.3 below, the
1
While we consider here only real-valued functions of a real variable, the theory for complex-
valued functions of a complex-variable is hardly any different.
 7.2 Convergence of power series 115

series also diverges for x < −1, although this does not follow immediately from
the ratio test as stated in Theorem 5.9 since the series is alternating. In addition,
the series clearly diverges for x = 1 and converges for x = −1, since then it is just
the alternating harmonic series.
P∞ an
For any a ∈ (0, 1), the series n=1 n converges and it follows from this, the
n
Weierstrass M -test, and the inequality |x|n ≤ an , which is valid for |x| ≤ a,
n

that the power series in question converges uniformly on the interval [−a, a]. On
the other hand, it is not hard to show that the power series does not converge
uniformly on the interval (−1, 1). In fact, if it did, then (taking  = 1 in the
definition P
of uniform convergence) there would be an integer N with the prop-
erty that ∞ n=N
xn
n
< 1 for all x ∈ (−1, 1), and hence with the property that
PM xn P∞ 1
n=N n < 1 for all integers M > N . Now the series P n=N n diverges to ∞, and
so there will be an integer M satisfying M > N and M 1
n=N n > 1. Since this is

P sum,xnit follows by continuity that there will be an x ∈ (0, 1) so close to 1

a finite
that M n=N n > 1, and this is a contradiction.
P∞ n2 x2n
2. Consider the power series n=0 2n . This is a power series with c = 0 and
with every second coefficient equal to zero or with every second term missing. We
can use either the ratio or the root test to determine the x for which this series
converges. In fact,
2
(n + 1)2 x2(n+1) n2 x2n n + 1 x2 x2
   
lim = lim =
n→∞ 2n+1 2n n→∞ n 2 2

and 1/n
n2 x2n n2/n x2 x2

lim = lim =
n→∞ 2n n→∞ 2 2
by
P∞Example 6.5–1, and both the √ratio and the root tests assert
√ that the series
n2 x2n
converges for |x| < 2 and diverges for |x| > 2. It is obvious
n=0 2n √
that this power series diverges for x = ± 2. Finally, the argument using the
Weierstrass M -test given in 1 will work here, and leads to the conclusion
√ that this
series converging uniformly on the interval [−a, a] for any a ∈ [0, 2). •

7.2 Convergence of power series

The two
P∞ examples in the preceding section suggest that, for any power series of
n
the form n=0 an x , there should be a number R such that the series converges
pointwise for |x| < R, converges uniformly on the interval [−r, r] for each r ∈ (0, R),
and diverges for |x| > R. There is indeed such a number and (as the next theorem
will show) it is given by the formula
1
R= p ,
lim supn→∞ n
|an |
116 7 Power series and Taylor series

with the understanding that 10 = ∞ and that ∞

1
= 0. Because of these properties, R
is known as the radius of convergence of the power series.
7.2 Example: For the power series considered in Example 7.1–1, we have
p
lim sup n |an | = lim sup n−1/n = 1
n→∞ n→∞

by Lemma 5.11 and so R = 1. For the power series considered in Example 7.1–2, we
have  2 1/(2n)
p
n n 1
lim sup |an | = lim sup n
=√
n→∞ n→∞ 2 2
√
and so R = 2. •

7.3 Theorem: Consider the power series

∞
X
an (x − c)n
n=0
. p 
and the quantity R = 1 lim supn→∞ n
|an | , where 1/0 = ∞ and 1/∞ = 0.
(i) If R = 0, then the power series converges only for x = c.
(ii) If 0 < R < ∞, then the power series converges absolutely for any x satisfying
|x − c| < R, diverges for any x satisfying |x − c| > R, and converges uniformly
on [c − r, c + r] for any r ∈ (0, R).
(iii) If R = ∞, then the power series converges absolutely for any x and uniformly
on any interval of finite length.
p
Proof: Suppose first that Rp = 0. Then lim supn→∞ n |an | = ∞ and so some sub-
sequence of the sequence ( n p |an |) must diverge to ∞. So, if x 6= c, then some
subsequence of the sequence ( n |an ||x − c|) will diverge to ∞, and hence some sub-
sequence of the sequence (|an (x − c)|n ) will also diverge to ∞. But this implies that
limn→∞ an (x − c)n 6= 0, and so it follows from Theorem 5.3 that the power series does
not converge. This proves (i).
Now suppose that R > 0 and consider a fixed number x and the quantity
p
L = lim sup n |an (x − c)n |.
n→∞

Then
p
n |x − c|
L = |x − c| lim sup |an | =
n→∞ R
and so, if |x − c| < R, then L < 1 and the series converges absolutely by the root
test (Theorem 5.10). On the other hand, if |x − c| > R, then L > 1, and we conclude
divergence of the series by the general version of the root test (Theorem 5.18).
 7.2 Convergence of power series 117

Now consider a numberP r ∈ (0, R). Then |an (x−c)n | ≤ |an |rn for all x ∈ [c−r, c+r]
and all n, and the series ∞ n
n=0 |an r | converges by the first part of the proof, and,
therefore, the power series converges uniformly on the interval [c − r, c + r] by the
Weierstrass M -test (Theorem 6.13). This completes the proof of the theorem. 
Let us consider a few interesting facts about radii of convergence of power series.
To do this, let us begin by proving a generalization of Lemma 5.11.
7.4 Lemma: If P is a polynomial function, then
p
lim n |P (n)| = 1.
n→∞

Proof: Let us write

P (n) = pk nk + pk−1 nk−1 + · · · + p1 n + p0

for p0 , p1 , . . . , pk ∈ R with pk 6= 0. We then have

p p
lim n |P (n)| = lim n |pk nk + pn−1 nk−1 + · · · + p1 n + p0 |
n→∞ n→∞
s
√

p
n n
k

n  pn−1 1 p1 1 p0 1 
= lim |pk | n 1 + pk n + · · · + pk nk−1 + pk nk .
n→∞

We have
 
p
n ln(|pk |)
lim |pk | = lim exp
n→∞ n→∞ n
 
ln(|pk |)
= exp lim = 1.
n→∞ n
By Lemma 5.11 we have
√
n √
k  √ k
lim nk = lim n
n = lim n n = 1.
n→∞ n→∞ n→∞

By combining the preceding two calculations with this one,

s 

n  pn−1 1 p1 1 p0 1 
lim
n→∞ 1 + pk n + · · · + pk nk−1 + pk nk 
pn−1 1 p1 1 p0 1
!
ln(1 + pk n
+ ··· + pk nk−1
+ pk nk
)
= lim exp
n→∞ n
pn−1 1 p1 1 p0 1
!
ln(1 + pk n
+ ··· + pk nk−1
+ pk nk
)
= exp lim = 1,
n→∞ n

one arrives at the statement of the lemma. 

With the lemma at hand, one easily proves the following useful result.
118 7 Power series and Taylor series

7.5 Proposition: Let the power series

∞
X
an (x − c)n
n=0

have radius of convergence R. Then the following statements hold:

(i) if P is a polynomial function, then the power series
∞
X
P (n)an (x − c)n
n=0

has radius of convergence R;

(ii) if Q is a polynomial function for which Q(n) 6= 0 for every n ∈ Z≥0 , then the
power series
∞
X an
(x − c)n
n=0
Q(n)
has radius of convergence R;
(iii) if b ∈ R, then the power series
∞
X
an bn (x − c)n
n=0

R
has radius of convergence .
|b|
Proof: (i) and (ii) We shall prove that the power series
∞
X P (n)
(x − c)n
n=0
Q(n)

has radius of convergence R. This, however, is straightforward, given Lemma 7.4.

Indeed, we have
s  p p
 P (n)a n  limn→∞
n
|P (n)| lim supn→∞
n
|an |
lim sup n  = p
n→∞ Q(n)  limn→∞ |Q(n)|
n

p
= lim sup n |an |,
n→∞

from which the conclusion follows immediately.

(iii) In this case, we directly compute
p
n
p |b|
lim sup |an bn | = |b| lim sup n |an | = .
n→∞ n→∞ R
 7.3 Properties of power series 119

Thus the radius of convergence of the power series

∞
X
an bn (x − c)n
n=0

R
is , as stated. 
|b|

7.3 Properties of power series

A power series with a nonzero radius of convergence defines a function on an open
interval or on the whole real line. What are the properties of this function? For
instance, is it continuous or differentiable, and (assuming that they exist) what are
its integral and its derivative?
P∞ n
7.6 Theorem: A power series n=0 an (x − c) whose radius of convergence R is
positive is a continuous function on (c − R, c + R).
Proof: Consider a point a ∈ (c−R, c+R)Pand let r be a number satisfying |a−c| < r <
R. For each integer n, the partial sum nk=0 ak (x − c)k is a polynomial, and hence a
continuous
P∞ function and (by Theorem 7.3(iii)) these partial sums converge uniformly
n
to P n=0 an (x − c) on the interval [c − r, c + r]. It thus follows from Theorem 6.8
that ∞ n
n=0 an (x − c) is a continuous function on the interval [c − r, c + r], and hence
a continuous function at a. Since this is true for each a ∈ (c − R, c + R), it means
that ∞ − c)n is continuous at each point of (c − R, c + R), and hence on
P
a
n=0 n (x
(c − R, c + R). 
The formal term-by-term derivative or integral of a power series are themselves
power series. Are they actually the derivative and integral of the power series itself?
If this were the case it would make the calculus of power series extremely easy. The
first step in investigating this question is to determine the radius of convergence of
these two associated power series.
7.7 Lemma: For any sequence (an ) and any c ∈ R, the three power series
∞ ∞ ∞
X
n
X
n−1
X an
an (x − c) , nan (x − c) , (x − c)n+1
n=0 n=1 n=0
n + 1

have the same radius of convergence.

Proof: For perfect transparency, let us make a change of index for the second and
third of these series:
X∞ ∞
X
n−1
nan (x − c) = (n + 1)an+1 (x − c)n ,
n=1 n=0
∞ ∞
X an X an−1
(x − c)n+1 = (x − c)n .
n=0
n+1 n=1
n
120 7 Power series and Taylor series

We next calculate
 
n/(n−1) n
lim sup|an−1 | = lim sup exp ln(|an−1 |)
n→∞ n→∞ n−1
 
n
= exp lim sup ln(|an−1 |)
n→∞ n + 1
 
= exp lim sup ln(|an−1 |)
n→∞

= lim sup exp(ln(|an−1 |)) = lim sup|an−1 |.

n→∞ n→∞

Therefore,
 1/n
p
lim sup n |an−1 | = lim sup|an−1 |
n→∞ n→∞
 
n/(n−1)
= lim sup|an−1 |
n→∞

= lim sup(|an−1 |n/(n−1) )1/n

n→∞
p
= lim sup|an−1 |1/(n−1) = lim sup n
|an |.
n→∞ n→∞

In similar manner, p p
n n
lim sup |an+1 | = lim sup |an |.
n→∞ n→∞

The result now follows from Proposition 7.5. 

Note that the second and third power series in the preceding lemma are obtained
from the first one by formal term-by-term differentiation and integration, respec-
tively.
P∞ n
7.8 Theorem: If n=0 an (x − c) is a power series whose radius of convergence R
is positive, then
∞
Z x X ! ∞
X an
an (t − c)n dt = (x − c)n+1
c n=0 n=0
n+1

and
∞
! ∞
d X
n
X
an (x − c) = nan (x − c)n−1
dx n=0 n=1

for all x ∈ (c − R, c + R).

Proof: Let f (x) = ∞
P n
Pn k
n=0 a n (x − c) and P n (x) = k=0 ak (x − c) for n ∈ Z>0 and
x ∈ (c − R, c + R). To prove the first formula, consider an x ∈ (c − R, c + R) and
let r be a number satisfying |x − c| < r < R. Then limn→∞ Pn = f uniformly on
 7.3 Properties of power series 121

[c − r, c + r] by Theorem 7.3, and hence uniformly on the interval with endpoints c

and x, regardless of whether x > c or x < c, and, therefore,
Z x Z x
f (t) dt = lim Pn (t) dt
c c n→∞
Z x
= lim Pn (t) dt
n→∞ c
n
X ak
= lim ak (x − c)k+1
n→∞
k=0
k+1
∞
X an
= (x − c)n+1 .
n=0
n + 1
This proves the first formula. P∞ n−1
To prove the second formula, let g(x) = n=1 nan (x − c) and recall from
Lemma 7.7 that this power series has a radius of convergence
P∞ of R. So, if r is a
n−1
number satisfying 0 < r < R, then the power series n=1 nan (x − c) converges
uniformly on [c − r, c + r]. But this just means that limn→∞ Pn0 = g uniformly on
[c − r, c + r] and, therefore, Theorem 6.11 implies that f 0 = g on [c − r, c + r]. Finally,
since r was an arbitrary number in (0, R), it follows that f 0 = g on (c − R, c + R). 
This theorem has two corollaries but it will be useful to consider several examples
before going on to these corollaries.
7.9 Examples: 1. From our knowledge of the geometric series, we know that
∞
1 X
= xn = 1 + x + x2 + x3 + · · ·
1 − x n=0
with a radius of convergence of 1. The above theorem on differentiation then gives
 
1 d 1
=
(1 − x)2 dx 1 − x
d
1 + x + x 2 + x3 + · · ·


=
dx
= 1 + 2x + 3x2 + · · ·
and thus ∞
1 X
= (n + 1)xn . (7.1)
(1 − x)2 n=0
Rx dt
On the other hand, since 0 1−t
= − ln(1 − x) for |x| < 1, the above theorem on
integration gives
Z x
ln(1 − x) = − (1 + t + t2 + · · · ) dt
0
x2 x3

= − x+ + + ··· ,
2 3
122 7 Power series and Taylor series

or ∞
x2 x3 X (−1)n+1
ln(1 + x) = x − + − ··· = xn . (7.2)
2 3 n=0
n
The two power series (7.1) and (7.2) both have radius of convergence of 1.
We can combine the above power series for ln(1 + x) and ln(1 − x) to obtain
 
1+x
ln = ln(1 + x) − ln(1 − x)
1−x
 
1 2 1 3
= x − x + x − ···
2 3
 
1 2 1 3
+ x + x + x + ···
2 3
3 5
 
x x
=2 x+ + + ···
3 5
∞
X x2n+1
=2 .
n=0
2n + 1

Letting, say, x = 1/3 in this last expression gives

   
1 + 1/3 1 1 1
ln 2 = ln =2 + + + ··· .
1 − 1/3 3 81 1215

2. Substituting −x2 in place of x in the geometric series

1
= 1 + x + x2 + x3 + · · ·
1−x
gives the power series
1
= 1 − x 2 + x4 − x6 + · · · ;
1 + x2
this isRvalid for x2 < 1, and so the radius of convergence of this last series is 1.
x dt
Since 0 1+t 2 = arctan x, we can integrate this last power series to obtain

1 1
arctan x = x − x3 + x5 − · · · ,
3 5
and this
√ power series too has a radius of convergence of 1. In particular, letting
x = 1/ 3 in this series gives
 
π 1 1 1 1
= √ 1− + − + ··· . •
6 3 3 · 3 5 · 32 7 · 33
The theorem on differentiation and integration of power series has the following
two consequences.
 7.4 Taylor series and Taylor polynomials 123
P∞ n
7.10 Corollary: A power series n=0 an (x − c) whose radius of convergence R is
∞
positive is a C -function on the interval interval(c − R, c + R) and
∞
!
dk X 
n 
an (x − c)  = k!ak
dxk n=0 
x=c

for all k ∈ Z>0 .

Proof: Applying the previous theorem and lemma k times gives
∞
!
dk X
an (x − c)n = k!ak + k!ak+1 (x − c) + · · ·
dxk n=0

and the corollary is now clear. 

It is obvious that the coefficients of power series determine the power series as a
function, but what about the converse? That is to say, if two power series with the
same centers represent the same function, must their corresponding coefficients be
equal? The next result asserts that this is indeed the case.
P∞ n
P∞ n
7.11 Corollary: If n=0 a n (x − c) and n=0 bn (x − c) are two power series with
positive radii of convergence R and S, respectively, and if there is a number r such
that 0 < r ≤ min{R, S} and
∞
X ∞
X
n
an (x − c) = bn (x − c)n
n=0 n=0

for all x ∈ (c − r, c + r), then an = bn for all n ∈ Z>0 .

Proof: This corollary is an immediate consequence of the preceding one. 

7.4 Taylor series and Taylor polynomials

P∞ n
Consider a power
P∞ series n=0 an (x−c) whose radius of convergence R is positive,
and put f (x) = n=0 an (x − c)n for x ∈ (c − R, c + R). Then, from the preceding
two sections, we know that
1. f is a C∞ -function on the interval (c − R, c + R) (by Corollary 7.10),
(n)
2. an = f n!(c) for n ∈ Z≥0 (also by Corollary 7.10), and
3. if Pn (x) = nk=0 ak (x−c)k for n ∈ Z>0 and x ∈ (c−R, c+R), then limn→∞ Pn = f
P
pointwise on (c − R, c + R) and uniformly on [c − r, c + r] for each r ∈ (0, R) (by
Theorem 7.3).
This section is concerned with what one might think of as a converse of 1, namely, is
every C∞ -function a power series?
124 7 Power series and Taylor series

To describe this converse more precisely, suppose that f is a C∞ -function defined

on some open interval I and that c is a point in I. We can then form the power series
∞
X f (n) (c)
(x − c)n ,
n=0
n!

and ask whether it converges on some interval of positive radius and, if so, whether
it converges to f on that interval. This power series is called the Taylor series of
f about c. Now the question of the convergence of this power series is that of the
convergence of its sequence of partial sums, and this suggests that we consider its
partial sums, i.e., the polynomials
n
X f (k) (c)
Pn (x) = (x − c)k (7.3)
k=0
k!

for n ∈ Z>0 . The polynomial (7.3) is called the nth Taylor polynomial of f at
(k)
c and has the property that f (k) (c) = Pn (c) for k ∈ Z≥0 . This property suggests
that Pn should be a good approximation to f near c since the two functions f and
Pn have the same value, the same slope (and hence the same tangent line), the same
concavity, etc., at c. In fact, if we let Rn = f − Pn , then f = Pn + Rn , and Rn is just
the error when we regard Pn as an approximation to f . The question of how good an
approximation Pn is to f , therefore, becomes the question of how small Rn is and, in
particular, the question of whether limn→∞ Pn (x) = f (x) for a particular value of x
is the question of whether limn→∞ Rn (x) = 0. To answer this question we evidently
need an expression for Rn and the next theorem contains two such expressions. Notice
that this theorem does not require that f be a C∞ -function, but merely of class Cn+1
on I.
7.12 Theorem: Suppose that f is a function defined on an open interval I, that c is
a point in I, and that n is a positive integer. If f (n+1) exists and is continuous on I
and if Rn is as above, then
1 x
Z
Rn (x) = (x − u)n f (n+1) (u) du (7.4 )
n! c
and there is a point ξ ∈ I such that

f (n+1) (ξ)
Rn (x) = (x − c)n+1 . (7.5 )
(n + 1)!

Proof: The best way to prove this theorem is probably to first prove a lemma and
then deduce the theorem from the lemma. The reason for is two-fold: the lemma is of
independent interest and doing it this way will make the proof easier to comprehend.
Part (ii) of the lemma is a special case of the generalized mean value theorem for
integrals (cf. Exercise E7.35).
 7.4 Taylor series and Taylor polynomials 125

1 Lemma: Let g be a continuously differentiable function defined on an open interval

I and let c be a point in I.
(i) If g 0 exists and is continuous on I, then
Z x
g(c)
(x − u)j g(u) du = (x − c)j+1
c j+1
Z x
1
+ (x − u)j+1 g 0 (u) du
j+1 c
for all j ∈ Z≥0 and all x ∈ I.
(ii) Corresponding to each x ∈ I \ {c} and each n ∈ Z>0 , there is a point ξ ∈ I
satisfying |ξ − c| ≤ |x − c| and
Z x
n+1
g(ξ) = (x − u)n g(u) du.
(x − c)n+1 c

Proof: (i) This follows easily from the formula for integration by parts:
Z x Z x 
j −1 d j+1
(x − u) g(u) du = (x − u) g(u) du
c j+1 c dx
−1  x
= (x − u)j+1 g(u)

j+1 c
Z x 
j+1 0
− (x − u) g (u) du
c
Z x
g(c) 1
= j+1
(x − c) + (x − u)j+1 g 0 (u) du.
j+1 j+1 c
(ii) Fix a point x ∈ I \ {c} and assume that x > c (the proof in the case x < c is
similar). If
m = inf{g(u) | c ≤ u ≤ x}
and
M = sup{g(u) | c ≤ u ≤ x},
then it is certainly true that m ≤ g(u) ≤ M for all u ∈ [c, x]. Since x − u ≥ 0 for all
u ∈ [c, x], it follows that

m(x − u)n ≤ (x − u)n g(u) ≤ M (x − u)n

for all u ∈ [c, x]. Integrating this inequality gives

Z x Z x Z x
n n
m (x − u) du ≤ (x − u) g(u) du ≤ M (x − u)n du,
c c c

hence Z x
m M
(x − c)n+1 ≤ (x − u)n g(u) du ≤ (x − c)n+1 ,
n+1 c n+1
126 7 Power series and Taylor series

and, therefore, Z x
n+1
m≤ (x − u)n g(u) du ≤ M. (7.6)
(x − c)n+1 c

Now let R denote the range of the restriction of g to the closed interval [c, x]. Since g
is continuous on [c, x] it follows from Theorem 4.24(i) that m, M ∈ R, and then (7.6)
and the intermediate value theorem (see Exercise E4.29) imply that
Z x
n+1
(x − u)n g(u) du
(x − c)n+1 c

also belongs to R. But, if this number belongs to R, then, by the definition of R, it

must be equal to g(ξ) for some point ξ ∈ [c, x]. H
Now we proceed with the proof of the theorem. The proof of (7.4) consists of
repeated applications of Lemma 1(i). First of all,
Z x
f (x) − f (c) = (x − u)0 f 0 (u) du
c
Z x
0
= (x − c)f (c) + (x − u)1 f 00 (u) du,
c

by the fundamental theorem of calculus and Lemma 1 and, therefore,

Z x
0
f (x) = f (c) + f (c)(x − c) + (x − u)1 f 00 (u) du.
c

Since
f (x) = P1 (x) + R1 (x), P1 (x) = f (c) + f 0 (c)(x − c)
by the definition of P1 and R1 , this verifies (7.4) for n = 1. Next,
Z x
f 00 (c) 1 x
Z
1 00
(x − u) f (u) du = 2
(x − c) + (x − u)2 f 000 (u) du
c 2 2 c

by Lemma 1, and thus

x
f 00 (c)
Z
0
f (x) = f (c) + f (c)(x − c) + (x − c)2 + (x − u)2 f 000 (u) du.
2 c

The first three terms on the right side are just P2 (x), and so the fourth term must
be R2 (x) and this verifies (7.4) for n = 2. Let’s play this game one more time:
Z x
f 000 (c) 1 x
Z
2 000
(x − u) f (x) du = + (x − u)3 f (4) du
c 3 3 c
 7.4 Taylor series and Taylor polynomials 127

and hence
f 00 (c) f 000 (c)
f (x) = f (c) + f 0 (c)(x − c) + (x − c)2 + (x − c)3
2 3!
1 x
Z
+ (x − u)3 f (4) (x) du
3! c
1 x
Z
= P3 (x) + (x − u)3 f (4) (x) du.
3! c
Just as before, this verifies (7.4) for n = 3. Continuing in this manner (or, better
still, starting over and proceeding by induction) will lead to (7.4) for a general value
of n. In doing so, the assumption that f (n+1) is continuous on I is sufficient to justify
the last application of integration by parts in this calculation.
Next we verify (7.5). According to Lemma 1(ii), there is a point ξ ∈ I such that
Z x
(n+1) n+1
f (ξ) = (x − u)n f (n+1) du
(x − c)n+1 c
and, therefore,
1 (x − c)n+1 (n+1) f (n+1) (ξ)
Rn (x) = f (ξ) = (x − c)n+1
n! n + 1 (n + 1)!
by (7.4). 
We can use either (7.4) or (7.5) in trying to decide whether the Taylor series of a
given function converges to that function.
7.13 Examples: 1. The derivatives of the sine function at the origin are easy to
calculate and the Taylor series of the sine function at 0 is
∞
X (−1)n 2n+1 x 3 x5 x7
x =x− + − + ··· .
n=0
(2n + 1)! 3! 5! 7!

Now consider a number x 6= 0 and an integer n. Then, by (7.5), there is a number

ξ such that
sin(n+1) (ξ)
Rn (x) = (x − 0)n+1
(n + 1)!
and, since all of the derivatives of sin are ± sin or ± cos, it follows that
|x|n+1
|Rn (x)| ≤ .
(n + 1)!
|x|n
Now the power series ∞
P
n=0 n! has a radius of convergence equal to ∞ (this is an
n
easy consequence of the ratio test) and, therefore, limn→∞ |x|n! = 0 by Theorem 5.3.
This means that limn→∞ Rn (x) = 0 and thus limn→∞ Pn (x) = sin x for each
number x.
128 7 Power series and Taylor series

The sine function is therefore equal to its Taylor series for all real numbers and
(from our knowledge of power series) it follows that
∞
X (−1)n 2n+1
sin x = x
n=0
(2n + 1)!

uniformly on intervals of finite length.

2. An argument virtually identical to that of the preceding example will show that
the Taylor series of the cosine function at the origin converges pointwise on R and
uniformly on intervals of finite length to the cosine function. Thus
∞
x2 x4 X (−1)n
cos x = 1 − + − ··· = x2n
2! 4! n=0
(2n)!

uniformly on intervals of finite length.

3. Since the derivative of the exponential function ex is just ex , it follows that the
Taylor series of this function is
∞
X xn x2 x3
=1+x+ + + ··· .
n=0
n! 2! 3!

It is easy to verify (using the ratio test) that this power series has a radius of
convergence equal to ∞. To show that it actually converges to ex , it is sufficient
to prove that limn→∞ Rn (x) = 0 for each number x.
Now, for each number x and positive integer n, there is a number ξ such that
dn+1 ex  xn+1 eξ

1 n+1
Rn (x) = x =
(n + 1)! dxn+1 x=ξ (n + 1)!
and, therefore,
|x|n+1 e|x|
0 ≤ Rn (x) ≤ .
(n + 1)!
|x|n
We saw in 1 that limn→∞ n!
= 0, hence limn→∞ Rn (x) = 0, and, therefore,
∞
x
X xn
e =
n=0
n!

for all real numbers x.

4. We want to identify the Taylor series of the function f (x) = (1 + x)r at 0, where
r is a real number, and show that it converges to f on the interval (−1, 1). Now,
if r is a positive integer, then f is just a polynomial (as can be seen by expanding
(1 + x)r by the binomial formula) and there is nothing to prove, so we may as well
assume that r is neither 0 nor a positive integer. It is easy to see that
f (k) (x) = r(r − 1) · · · (r − k + 1)(1 + x)r−k ,
 7.4 Taylor series and Taylor polynomials 129

and hence the coefficient of xk in the Taylor series of f at 0 is

f (k) (0) r(r − 1) · · · (r − k + 1)

= .
k! k!
Since this expression

is reminiscent of a binomial coefficient, it will be convenient
to denote it by k , with the understanding that 0 = 1. Notice that kr 6= 0 for
r r

all k ∈ Z>0 since r is not a nonnegative integer. So the Taylor series of f at 0 is

∞  
X r n r(r − 1) 2 r(r − 1)(r − 2) 3
x = 1 + rx + x + x + ··· .
n=0
n 2! 3!

The radius of convergence of this power series is best calculated by the ratio test:
     
 r   r n
n+1  |r − k|
lim 
 x   x  = |x| lim
n→∞ n+1 n n→∞ (k + 1)!

= |x|,

and so the radius of convergence is 1.

For this particular function f , it is very difficult to estimate the remainder Rn as
was done in the previous examples, so the proof of the convergence of the Taylor
series for f to f will be accomplished by another method. If we let g denote the
Taylor series of f at 0, i.e., if
∞  
X r n
g(x) = x
n=0
n

for |x| < 1, then we know that g is a C∞ -function defined on the interval (−1, 1).
To show that the Taylor series of f at 0 converges to f , it is sufficient to show
that f (x) = g(x) for |x| < 1, and for this it is, in turn, sufficient to show that
 
d g(x)
f (0) = g(0), = 0, x ∈ (−1, 1). (7.7)
dx f (x)
Indeed, the second of these equations implies that f (x)/g(x) is constant, and then
the first equation implies that this constant is 1.
Now, it is obvious from the definitions of f and g that f (0) = 1 = g(0) and the
quotient rule gives
g 0 (x)f (x) − g(x)f 0 (x)
 
d g(x)
=
dx f (x) f 2 (x)
(1 + x)r g 0 (x) − r(1 + x)r−1 g(x)
=
(1 + x)2r
(1 + x)g 0 (x) − rg(x)
= .
(1 + x)r+1
130 7 Power series and Taylor series

To complete this example, it will be sufficient to show that (1+x)g 0 (x) = rg(x) for
|x| < 1. Since g is a power series, its derivative g 0 can be found by term-by-term
differentiation (by Theorem 7.8), and hence
∞  
0
X r n−1
(1 + x)g (x) = (1 + x) n x
n=1
n
∞   ∞  
X r n−1 X r n
= n x + n x
n=1
n n=1
n
  X ∞   ∞  
r r n
X r n
= + (n + 1) x + n x
1 n=1
n+1 n=1
n
∞     
X r r
=r+ (n + 1) +n xn .
n=1
n + 1 n

The expression in the square brackets can be simplified as follows:

   
r r r(r − 1) · · · (r − n)
(n + 1) +n = (n + 1) +
n+1 n (n + 1)!
r(r − 1) · · · (r − n + 1)
n
n!  
r−n n
= r(r − 1) · · · (r − n + 1) +
n! n!
 
r
=r ,
n
and, therefore,
∞  
0
X r n
(1 + x)g (x) = r + r x = rg(x).
n=1
n
5. For two numbers v and c satisfying 0 < v < c, we have
−1/2
v2

1
p = 1− 2
1 − v 2 /c2 c
2
− 12 − 32
  2


1 v v2
=1+ − − 2 + − 2 + ···
2 c 2! c
1 v2 3 v4
= 1 + 2 + 4 + ···
2c 8c
by the previous example. This expression leads to the approximation
1 1 v2
p ≈1+ ,
1 − v 2 /c2 2 c2
v
which is frequently used in relativity theory when c
 1.
 7.4 Taylor series and Taylor polynomials 131

6. Although it is true for all of the common functions, it is not always the case
that the Taylor series of a C∞ -function converges to that function. The simplest
function for which this does not occur is probably the one in the following example.
The function f in this example satisfies f (0) = 0, 0 ≤ f (x) ≤ 1 for all x, is
asymptotic to 1 in both directions, and is incredibly flat at the origin. (Try using
plotting the graph of this function using some software package.)
Consider the function f defined by
( 2
e−1/x , x 6= 0,
f (x) =
0, x = 0.
2
Then f 0 (x) = 2x−3 e−1/x for x 6= 0 by the usual formulae for differentiation,
whereas f 0 (0) must be calculated directly from the definition of the derivative and
l’Hôspital’s rule:
2
0 f (h) − f (0) e−1/h
f (0) = lim = lim
h→0 h h→0 h
1/h k
= lim 1/h2 = lim k2
h→0 e k→±∞ e
1
= lim 2 = 0.
k→±∞ 2kek

A similar calculation will show that

( 2
00 p2 (1/x)e−1/x , x 6= 0
f (x) =
0, x = 0,

where p2 is a polynomial of degree 6, and an inductive argument will even show

that ( 2
(k) pk (1/x)e−1/x , x 6= 0,
f (x) =
0, x = 0.
for k ∈ Z>0 , where pk is a polynomial of degree 3k. So the derivatives f (k) of f
exists on R for all k ∈ Z>0 and f is C∞ , yet the Taylor series of f about 0 is the
zero function and does not converge to or equal f except at 0 itself. •
132 7 Power series and Taylor series

Exercises
E7.1 Consider a power series ∞ n an
P
n=0 an (x − c) . Show that, if the sequence (| an+1 |)
converges or diverges to ∞, then the radius of convergence of the power series
is  a 
 n 
R = lim  .
n→∞ an+1

an xn , where an is as
P
E7.2 Find the radius of convergence of the power series
given below, and determine whether the series converges absolutely, converges
conditionally, or diverges at the endpoints of the interval of convergence.
(a) an = 2−n n4 (n5 + 1)
(n!)2
(b) an =
(2n)!
(c) an = (3n + 4)n 5−n
E7.3 Determine the radius of convergence, R, of the power series ∞ n
P
n=1 an x , where
an is as given below. Also, determine whether the series converges absolutely,
converges conditionally, or diverges at the points x = ±R.
(a) an = 10−n + 10n

(n!)2
(b) an = 1/ 2n n
= (2n)! .
E7.4 Determine the radius of convergence of each of the following power series:
P∞ n10 n
(a) x
Pn=1
∞
n!
n! n
(b) n=1 nn (x − 1)
E7.5 Determine the interval of convergence of each of the following series. (Do not
investigate convergence or divergence at the endpoints.)
∞  n
bn

X a
(a) + 2 (x − 1)2n , where a and b are two positive numbers
n=1
n n
∞
X (2x + 3)n
(b)
n=1
n
∞
X en  x n
(c) − 1
n=1
n3 + 1 2
E7.6 Determine the interval of convergence of the power series
∞
X (2x)2n
.
n=1
n log n

E7.7 Determine the interval of convergence of and an explicit formula for the power
series
f (x) = 1 + 2x + x2 + 2x3 + x4 + · · · .
 Exercises for Chapter 7 133

E7.8 Prove that the power series

∞
X xn
f (x) =
n=1
(n + 1)(n + 2)

converges uniformly on the closed interval [−1, 1].

E7.9 Determine all of the numbers x for which the power series ∞ n
P
n=0 cn (x − 1)
converges, where (
2n /n, n is odd,
cn =
1/n, n even.
E7.10 Consider the following power series: ∞ n
P
n=0 x .
(a) What is the radius of convergence R of this power series?
(b) For x ∈ (−R, R), explain why ∞ n 1
P
n=0 x = 1−x .
(c) For what values of x can we write
∞
1 X
= (−1)n x2n .
1 + x2 n=0

(d) For what values of x can we write

Z x ∞
X (−1)n x2n+1
1
dy = arctan(x) = .
0 1 + y2 n=0
2n + 1

(e) Can you conclude the validity of the formula

∞
X (−1)n
π=4 .
n=0
2n + 1

E7.11 (a) Determine the interval of convergence, I, of the power series ∞ xn

P
n=1 n(n+1) .
(b) Show that ∞ xn
P
n=1 n(n+1) converges uniformly on I.
(c) Deduce from part (b) that the function f (x) = ∞ xn
P
n=1 n(n+1) is continuous
on I.
(d) Starting from the fact that ∞ n 1
P
P∞ n=0 x = 1−x for |x| < 1, derive an expres-
xn
sion for n=1 n(n+1) that is valid for all x ∈ I. You must carefully justify
all the steps in your derivation.
E7.12 Determine the radius of convergence, R, of the power series ∞ n
P
n=1 an x , where
an is as given below. If R ∈ (0, ∞), determine whether the series converges at
the points x = ±R.

n+1 ln(n+1)
(a) an = −1 2 n+1
.
n
(b) an = 2n! .
134 7 Power series and Taylor series

E7.13 Note that (at least formally)

∞ ∞ ∞
!
X X d X n
nxn = x nxn−1 = x x (E7.1)
n=1 n=1
dx n=0

(a) Find the sum of the series ∞ n

P
n=1 nx .
(b) For what values of x is the calculation (E7.1) valid? Justify your answer
by appealing to theorems about power series.
E7.14 Find the sum of the power series ∞ 4n+3
P
n=100 x .
E7.15 (a) PFind a simple expression for the sum of the power series f (x) =
∞ xn
n=1 n(n+1) .
Hint: Multiply by x and take the second derivative.
(b) PUse a procedure similar to that in (a) to find a simple expression for
∞ 2 n
n=0 n x .
E7.16 Let r be a fixed non-negative integer. It is not hard to check that the radius
of convergence of the power series
∞
X (−1)n x2n+r
n=0
22n+r n!(n + r)!

is R = ∞, so that the series converges absolutely for all x ∈ R (you don’t need
to show this). Let Jr (x) denote the function represented by this series. This
function is called the r-th order Bessel function of the first kind.
Verify that Jr (x) is a solution y(x) to Bessel’s differential equation

x2 y 00 + xy 0 + (x2 − r2 )y = 0.

You must justify all the steps in your verification by quoting appropriate the-
orems wherever necessary.
(Remark : Bessel functions have many applications; they arise for instance
when studying the propagation of electromagnetic waves through cylindrical
waveguides.)
E7.17 It is shown in Example 7.13–4 of the course reader (read it!) that the Taylor
series about 0 of f (x) = (1 + x)r , where r is a real number, is given by
P ∞ r

n
n=0 n x . Since r can be any real number, not necessarily a positive integer,
the binomial coefficient
n is to be interpreted as r(r−1)(r−2)...(r−n+1)
r


n!
, with the
r
understanding that 0 = 1. It is also shown in the example that the Taylor
series, in fact, converges to f (x) for all x ∈ (−1, 1).
(a) Use the above fact (and some algebraic manipulations)
P∞ 2n
nto deduce that the
1
Taylor series about 0 of 1−4x is the series n=0 n x .
√
P∞ 2n
n 1
Conclude from this that n=0 n x converges to √1−4x for x ∈
(−1/4, 1/4).
 Exercises for Chapter 7 135

(b) (This is hard, and need not be attempted, but you may wish P∞to give
itn
2n
a shot.) A solution to Exercise E5.13 shows that the series n=0 n x
converges conditionally at x = −1/4. Show that
∞  
X 2n 1 1
(−1/4)n = p =√ .
n=0
n 1 − 4(−1/4) 2
1 2n


E7.18 The nth Catalan number is defined as Cn = n+1 n
, for all integers n ≥ 0.
(a) Use the inequality  
2n −n 1 1
4 ≤ √ e 24n
n πn
P∞
to deduce that n=0 Cn 4−n converges.
(b) Determine the interval of convergence, I, of the power series ∞ n
P
P∞ n=0 Cn x .
(c) Show that n=0 Cn xn converges uniformly on I.
(d) Deduce from part (c) that the function f (x) = ∞ n
P
n=0 Cn x is continuous
on I.
(e) Use the result of Exercise E7.17(a) to derive an expression for ∞ n
P
n=0 Cn x
that is valid for all x ∈ I. You must carefully justify all the steps in your
derivation.
(Catalan numbers occur in various counting problems; the question above
derives a generating function for the Catalan numbers.)
series ∞ n
P
E7.19 If the radius of convergence of the powerP n=0 an x is R, what is the
∞ 2n
radius of convergence of the power series n=0 an x ?
Suppose that the power series ∞ an xn has radius of convergence R ∈ (0, ∞).
P
E7.20 n=0 P∞ 2
Find the radius of convergence of n=0 an xn .
X∞
E7.21 Suppose that the power series f (x) = an xn has a nonzero radius of con-
n=0
vergence.
(a) If f is an odd function, show that

a0 = a2 = a4 = · · · = 0.

(b) If f is an even function, show that

a1 = a3 = a5 = · · · = 0.

E7.22 Prove that if (an ) is a sequence such that the sequence

P∞ (|an |) is bounded
away from both 0 and ∞, then the power series n=0 an xn has a radius of
convergence equal to 1.
E7.23 Suppose that the sequence (an ) is bounded, but that the seriesP ∞
P
n=0 an di-
verges. Prove that the radius of convergence of the power series ∞ n
n=0 an x is
equal to 1.
136 7 Power series and Taylor series

E7.24 Suppose that ∞

P n
P∞ n
n=0 an x and n=0 bn x are two power series whose radii of
convergence are R1 and R 2 ,, respectively, what can you say (prove) about the
radius of convergence of ∞ n
P
n=0 (an + bn )x ??
P∞ xn
E7.25 (a) Determine the radius of convergence of the power series Rx f (x) = n=0 n!
and obtain expressions for the power series f 0 (x) and 0 f (t) dt and cal-
culate their radii of convergence.
x n
(b) Repeat this for the power series f (x) = ∞ en
P

n=1 5 .
P∞ xnn +1 2
E7.26 Let S(x) be the sum of the power series n=0 n+1 . (This is not the power
series of a familiar function. Except for part (c), you must answer the following
questions by using only properties of power series.)
(a) What is the longest open interval I on which this power series converges?
(b) If S continuous on I? Is S differentiable on I? Why?
(c) Show that, if |x| < 1 then
 n   n+1 
 x  x 
n + 1 ≥ n + 2 .
   

(d) Prove that S(x) > 0 for all x ∈ I.

d
(e) Calculate dx (xS(x)) and use this to deduce a simple expression for S(x).
E7.27 The radius of convergence of the power series
∞
X x2n+1
f (x) =
n=1
4n n2

is 2 (it is not necessary to prove this).

(a) Prove that this series converges uniformly on [−2, 2].
(b) Explain how we know that f 0 (x) is defined at least on (−2, 2). Quote
relevant theorems.
(c) Determine all the open intervals on which f is increasing and all the open
intervals on which the graph of f is concave up. (Do not attempt to find
a formula for f .)
(d) What symmetry (if any) does f have? Use the information determined
so far to make as accurate a sketch as you can of the graph of f .
E7.28 Consider the power series
∞
X 1
f (x) = 3 n
(x − 2)2n .
n=1
n 9

(a) Determine the radius of convergence of this series and the interval I con-
sisting of all the numbers x for which the series converges.
(b) Show that it converges uniformly on the set I.
 Exercises for Chapter 7 137

(c) Obtain expressions for f 0 (x) and f 00 (x).

(d) Determine all the open intervals on which f is increasing, is decreasing,
is concave up, and is concave down. Also, does the graph of f have any
symmetry and what are its intercepts? Finally, make a rough sketch of
the graph of f .
E7.29 Determine the Taylor series for each of the following functions about the given
point (use any convenient procedure):
(a) x sin x about x = 0
(b) cos x about x = π/3
√
(c) Z x about x = 1
x
2
(d) e−t dt about x = 0
Z0 x
sin t2 − t2
(e) dt about x = 0
0 t6
E7.30 Use the Taylor series for f (x) = ex about x = 0 and the methods described
in Section 7.4 to obtain the Taylor series for each of the following functions
about x = 0:
(a) e−x
1 x
e + e−x


(b) cosh x =
2
1 x
e − e−x


(c) sinh x =
2
2
(d) ex
2
(e) Zxex
x
2 1  x2 
(f) tet dt = e −1
0 2
E7.31 Find enough of the Taylor series expansion of the function f (x) = x3 ex to
determine f (5) (0).
E7.32 Find the Taylor series for f (x) = sin x about x = π/3 and prove that this
Taylor series really does converge to sin x for all x ∈ R.
E7.33 (a) Find the Taylor polynomial of degree n about x = 0 for the function ex .
(b) Use your answer to (a) to find the Taylor polynomial of degree n about
x = 0 for the function
(
(ex − 1)/x, x 6= 0,
f (x) =
0, x = 0.

(c) Use your answer to (b) to express the integral

Z 0
ex − 1
dx
−1/2 x
138 7 Power series and Taylor series

as an infinite series, and estimate the error in approximating the integral

by the first n terms in the series.
1
E7.34 Determine the Taylor series about x = 0 of the function f (x) = 1+2x 3 . What

is its radius of convergence of this Taylor series?

E7.35 Prove the generalized mean value theorem for integrals: If f and g are two
continuous function on an interval [a, b] and if f is nonnegative-valued and
positive at at least one point, then there is point ξ ∈ [a, b] such that
Z b Z b
f (ξ) g(t) dt = f (t)g(t) dt
a a

(cf. the proof of Lemma 1(ii) from the proof of Theorem 7.12).
E7.36 The binomial theorem states that
p(p − 1) 2 p(p − 1)(p − 2) 3
(1 + x)p = 1 + px + x + x + ···
2! 3!
for |x| ≤ 1. Use the familiar formula
Z x
−1 du
sin x= √
0 1 − u2
to deduce that
π 1 1 1·3
= + 3
+ + ··· .
6 2 2·3·2 2 · 4 · 5 · 25
Justify each of the steps in your work.
E7.37 Use the binomial theorem to obtain simple approximations for the following
expressions valid for arguments indicated:
1
(a) for |u|  1
1−u
−1/2
v2 v2

(b) 1 − 2 for 2  1
c c
2 3/2 2
(c) (1 + x ) for x  1
(d) (1 + 3x)π for |x|  1
E7.38 Compute the first few terms in the power series expansion about x = 0 of the
following functions:
(a) x3 cos x
e−x
(b) (1−x)
(c) ex sin x
(d) tan x
 Index
absolute convergence, 80, 116 construction of the reals, 17, 19
alternating harmonic series, 78, 79, 115 continuity, 5, 58
alternating series test, 81 continuous, 58
remainder, 83 contraction, 47
contradiction, 6
balls, 50 contrapositive, 6
closed, 50 convergence
open, 50 absolute, 80, 116
punctured, 50 conditional, 82
binomial coefficient, 129 of power series, 115
Bolzano-Weierstrass Theorem pointwise, 92
for sequences, 29 uniform, 116
for sets, 54 for sequences, 96
boundary, 56 convergence tests
boundary point, 56 absolute convergence, 80
bounded, 103 alternating series test, 81
above, 20, 21 Cauchy criterion, 105
below, 20, 21 comparison test, 72
sequence, 26, 28, 29, 46 integral test, 74, 80
set, 20, 21 p-series, 75
ratio test, 75, 80, 114
Cauchy root test, 76, 81
pointwise, 105 convergent sequence, 26, 28, 45
Cauchy criterion Cauchy, 36, 47
for sequences, 36, 47 uniform, 96
Cauchy sequence, 36, 47 uniform Cauchy, 105
Cauchy–Schwartz inequality, 45 convergent sequences, 30
closed and bounded set, 57 convergent series, 69, 80
closed ball, 50 absolute, 80, 116
closed set, 51, 56 conditional, 82
cluster point, 54–56 uniform, 116
compact set, 59 convergent subsequence, 29, 57
comparison test, 72 converse, 5
complement, 2 cosine function
complement of a set, 51 Taylor series of, 128
completeness, 19 counterexample, 9
conditional convergence, 82
conjunction, 3 de Morgan’s laws, 53
140 Index

Dedekind cut, 17 integral test remainder, 83

derivative interior point, 51
of sequences of functions, 100 intersection, 2
differentiation intersection of closed sets, 52, 53
term-by-term, 103 intersection of open sets, 52, 53
direct proof, 6 interval, 2
disjunction, 3 isolated point, 56
divergent sequence, 26
divergent series, 69 l’Hôspital’s rule, 95
least upper bound, 21
empty set, 1 least upper bound property, 20
enumeration of Q, 95 lim inf, 33
estimates of remainders, 83, 84 lim sup, 33
existential quantifier, 4 limit
exponential series, 27 of a pointwise convergent sequence,
94
finite set, 51 of a sequence, 26, 45, 71
fixed point, 47 of sequences, 33
function limit inferior, 33, 34
bounded, 103 limit superior, 33, 34
contraction, 47 logical equivalence, 3
sup norm of, 104 lower bound, 20
function(s)
continuous, 5, 58, 59 map, 2
sequence of, 91 monotone sequence, 27, 28
series of, 92
natural numbers, 25
geometric series, 70 negation, 3, 4
greatest lower bound, 21 Nonconstructive proof, 9
nondecreasing sequence, 27
harmonic series, 27, 71, 75 nonincreasing sequence, 27
alternating, 78, 79 nonnegative integers, 2, 12
implication, 3 norm, 45, 104
negation of, 4 open ball, 50
induction, 7 open covering, 59
inf, 21 open set, 51
infimum, 21
infinite set, 51 p-series, 75
initial value problem, 91 partial sums, 27, 69, 82
integers, 2, 13 pointwise Cauchy, 105
nonnegative, 12 pointwise convergence, 92
integral  definition, 93
of a sequence of functions, 99 polynomials, 123
integral test, 74, 80 power series, 114
 Index 141

predicate, 4, 7 of partial sums, 69

subject of, 4 range of, 26
product of sets, 2 strictly decreasing, 27
proof strictly increasing, 27
by contradiction, 6 unbounded, 27
by counterexample, 9 uniformly Cauchy, 105
by induction, 7 uniformly convergent, 96
by strong induction, 7 sequence of functions
direct, 6 derivative of, 100
nonconstructive, 9 sequences
properties of uniform limits, 98 convergent, 30
proposition, 3 series
punctured ball, 50 alternating harmonic, 115
convergent, 69, 80
quantifier, 4 divergent, 69, 72
existential, 4 exponential, 27
universal, 4 geometric, 70
radius of convergence, 116 harmonic, 27, 71, 75
ratio test, 75, 80, 114 alternating, 78, 79
remainder, 84 of functions, 92
rational numbers, 2 p-series, 75
real line, 17, 19 power, 114
real numbers, 2 rearrangement of, 82
rearrangement of a series, 82 remainders, 83, 84
remainder, 83, 84 Taylor series, 123
alternating series test, 83 term-by-term differentiation, 103
integral test, 83 with positive and negative terms,
ratio test estimate, 84 77
root test, 76, 81 set, 1
boundary, 56
sequence, 25 bounded, 20
bounded, 26, 28, 29, 46 bounded above, 20, 21
Cauchy, 36, 47 bounded below, 20, 21
convergent, 26, 28, 45 closed, 51
decreasing, 27 equivalent conditions on, 56
divergent, 26 closed and bounded, 57
increasing, 27 cluster point, 56
limit, 33 compact, 59
limit of, 26 complement of, 51
monotone, 27, 28 finite, 51
nondecreasing, 27 infinite, 51
nonincreasing, 27 interior of, 51
of functions, 91 open, 51
142 Index

product of sets, 2 tent function, 94

unbounded, 20 term-by-term differentiation, 103
sine function triangle inequality, 19
Taylor series of, 127
strictly decreasing sequence, 27 unbounded
strictly increasing sequence, 27 sequence, 27
strong induction, 7 set, 20
subject of a predicate, 4 uniform Cauchy sequence, 105
subsequence, 28–30 uniform convergence, 116
convergent, 29, 57 of sequences, 96
definition of, 28 uniform limits
subset, 1 properties of, 98
sup, 21 union, 2
sup-norm, 104 union of closed sets, 52, 53
supremum, 21 union of open sets, 52, 53
upper bound, 20
Taylor series, 123
telescoping, 71 Weierstrass M -test, 102