Real Analysis

Module 2RA
Associate Professor Elena Berdysheva
Version: 12 August 2023
Contents

0 Preliminaries 1
0.1 Sets 1
0.2 Number systems 3

1 The real numbers 5

1.1 The well-ordering principle and induction 5
1.2 The completeness axiom 8
1.3 Consequences of completeness 12
1.4 Cardinality 16

2 Sequences and series 21

2.1 Limits of sequences 21
2.2 Rules of working with limits 25
2.3 The Monotone Convergence Theorem 29
2.4 A first look at series 32
2.5 The comparison test for series 37
2.6 Subsequences and the Bolzano-Weierstrass Theorem 41
2.7 Cauchy sequences 44
2.8 Absolute and conditional convergence of series 47
2.9 The ratio and the root tests 51
2.10 Rearrangements 54

3 The topology of ℝ 58
3.1 Open sets 58
3.2 Closed sets 60
3.3 Compact sets 64

4 Limits of functions and continuity 66

4.1 Limits of functions 66
4.2 Continuity 73
4.3 Uniform Continuity 76

3
5 Derivatives 80

6 Sequences of functions 87
6.1 Pointwise and uniform convergence 87
6.2 Series of functions 92
6.3 Power series 95
6.4 Taylor series 98
Chapter 0
Preliminaries

In this chapter we give a short overview of basic concepts on sets and number
systems. We expect that you are familiar with these concepts. We will not
discuss them in class. Please make sure that you read this chapter and that you
are comfortable with the concepts.

0.1 Sets
Intuitively, a set is a collection of objects. Given a set 𝐴 and an object 𝑥, we
write 𝑥 ∈ 𝐴 if 𝑥 is in the set 𝐴 and say that 𝑥 is an element of 𝐴. Otherwise, we
write 𝑥 ∉ 𝐴 and say that 𝑥 is not an element of 𝐴. The empty set ∅ is the set that
contains no elements.
We say that two sets 𝐴 and 𝐵 are equal and write 𝐴 = 𝐵 if every element of 𝐴
is an element of 𝐵 and every element of 𝐵 is an element of 𝐴, i.e.

∀ 𝑥 ∶ 𝑥 ∈ 𝐴 ⟺ 𝑥 ∈ 𝐵.

We say that a set 𝐴 is a subset of a set 𝐵 and write 𝐴 ⊆ 𝐵 or 𝐵 ⊇ 𝐴 if every

element of 𝐴 is also an element of 𝐵, i.e.

∀ 𝑥 ∶ 𝑥 ∈ 𝐴 ⟹ 𝑥 ∈ 𝐵.

Clearly, 𝐴 = 𝐵 if and only if 𝐴 ⊆ 𝐵 and 𝐵 ⊆ 𝐴.

For every set 𝐴 we have ∅ ⊆ 𝐴 and 𝐴 ⊆ 𝐴. If 𝐵 ⊆ 𝐴 and 𝐵 ≠ 𝐴, then 𝐵 is
called a proper subset of 𝐴. The union of two sets 𝐴 and 𝐵 is the set

𝐴 ∪ 𝐵 = {𝑥 ∶ 𝑥 ∈ 𝐴 or 𝑥 ∈ 𝐵}.

The intersection of two sets 𝐴 and 𝐵 is the set

𝐴 ∩ 𝐵 = {𝑥 ∶ 𝑥 ∈ 𝐴 and 𝑥 ∈ 𝐵}.

1
 Chapter 0 Preliminaries

𝐴∪𝐵 𝐴∩𝐵 𝐴⧵𝐵

Figure 0.1: 𝐴 ∪ 𝐵, 𝐴 ∩ 𝐵, 𝐴 ⧵ 𝐵

The difference of two sets 𝐴 and 𝐵 is the set

𝐴 ⧵ 𝐵 = {𝑥 ∶ 𝑥 ∈ 𝐴 and 𝑥 ∉ 𝐵}.

Two sets 𝐴 and 𝐵 are called disjoint if 𝐴 ∩ 𝐵 = ∅.

Notice that 𝐴 ∪ 𝐵 = (𝐴 ⧵ 𝐵) ∪ (𝐴 ∩ 𝐵) ∪ (𝐵 ⧵ 𝐴) and that the sets 𝐴 ⧵ 𝐵, 𝐴 ∩ 𝐵
and 𝐵 ⧵ 𝐴 are pairwise disjoint.
The notions of unions and intersections can be extended to more than two
sets: If 𝐴1 , … , 𝐴𝑛 is a collection of sets (or family of sets), we define
𝑛
� 𝐴𝑘 = {𝑥 ∶ 𝑥 ∈ 𝐴𝑘 for some 𝑘 ∈ {1, … , 𝑛}},
𝑘=1
𝑛
� 𝐴𝑘 = {𝑥 ∶ 𝑥 ∈ 𝐴𝑘 for all 𝑘 ∈ {1, … , 𝑛}}.
𝑘=1

Similarly, we will consider unions and intersections of countably many sets.

Intervals of real numbers are defined as follows. If 𝑎, 𝑏 ∈ ℝ with 𝑎 ≤ 𝑏, we
define

[𝑎, 𝑏] = {𝑥 ∈ ℝ ∶ 𝑎 ≤ 𝑥 ≤ 𝑏},
(𝑎, 𝑏) = {𝑥 ∈ ℝ ∶ 𝑎 < 𝑥 < 𝑏},
(𝑎, 𝑏] = {𝑥 ∈ ℝ ∶ 𝑎 < 𝑥 ≤ 𝑏},
[𝑎, 𝑏) = {𝑥 ∈ ℝ ∶ 𝑎 ≤ 𝑥 < 𝑏}.

There are also infinite intervals

[𝑎, ∞) = {𝑥 ∈ ℝ ∶ 𝑥 ≥ 𝑎},
(𝑎, ∞) = {𝑥 ∈ ℝ ∶ 𝑥 > 𝑎},
(−∞, 𝑏] = {𝑥 ∈ ℝ ∶ 𝑥 ≤ 𝑏},
(−∞, 𝑏) = {𝑥 ∈ ℝ ∶ 𝑥 < 𝑏}.

2
 Chapter 0 Preliminaries

Note that “−∞” and “∞” are not real numbers, and the notation “(−∞” and
“∞)” means that the corresponding intervals are infinite.

0.2 Number systems

In this module we will work with sets of numbers. The most important sets of
numbers are:

• The natural numbers ℕ = {1, 2, 3, 4, 5, …}.

Notice that we do not include 0 to be a natural number. Some authors do,
so you must be careful when you work with literature.

• The integers ℤ = {… , −3, −2, −1, 0, 1, 2, 3, …}.

𝑝
• The rational numbers ℚ = � 𝑞 ∶ 𝑝, 𝑞 ∈ ℤ, 𝑞 ≠ 0�.

• The real numbers ℝ.

We have ℕ ⊆ ℤ ⊆ ℚ ⊆ ℝ; each number system is a proper subset of a larger

one.
The natural numbers ℕ are the numbers that we use for counting. They can
be defined axiomatically (e.g., via the Peano Axioms). We do not go into details
in this module.
Within the number system ℕ, the operation of addition works well. However,
a difference of two natural numbers is not always a natural number, for example,
2 − 3 = −1 is not a natural number. This motivates us to extend the number
system ℕ to a larger number system ℤ where also subtraction can be performed.
This is done by adding the number 0 and a copy of ℕ with a negative sign in
front of each number.
The operations of addition, subtraction and multiplication can be performed
within the number system ℤ, but a quotient of two integers is not always an
2
integer, for example, 2÷3 = 3 is not an integer. We enlarge ℤ to ℚ by considering
fractions of integers. Some care is needed at this step because each rational
number has infinitely many representations as a fraction due to the fact that
𝑝 𝑝𝑘
𝑞
= 𝑞𝑘
for 𝑝 ∈ ℤ, 𝑞, 𝑘 ∈ ℤ ⧵ {0}. This issue can be solved by considering an
equivalence relation for pairs (𝑝, 𝑞) ∈ ℤ × ℤ, namely, (𝑝, 𝑞) is equivalent to (𝑚, 𝑛)
if 𝑝𝑛 = 𝑞𝑚. We do not go into details.

3
 Chapter 0 Preliminaries

Now all four arithmetic operations +, −, ×, ÷ can be performed within the

number system ℚ. Moreover, ℚ builds an algebraic structure called a field. This
means that the operations of addition and multiplication satisfy the known
properties such as commutativity, associativity, the existence of an identity, the
existence of an inverse element, and obey the distribution property 𝑎(𝑏 + 𝑐) =
𝑎𝑏 + 𝑎𝑐. Furthermore, there is a natural order defined on ℚ: for any two numbers
𝑥, 𝑦 ∈ ℚ, exactly one of the following relations is true

𝑥 < 𝑦, 𝑥=𝑦 or 𝑥 > 𝑦.

Moreover, the order is consistent with the arithmetic operations. We call such a
system an ordered field. We do not go into further details.
However, ℚ lacks an important property of completeness. We will discuss it
in Sections 1.1 and 1.2. For example, there is no rational number 𝑥 with 𝑥2 = 2.
We extend ℚ to a larger number system ℝ which is a complete ordered field
containing ℚ as a subfield.
It is highly nontrivial to construct ℝ from ℚ. We will not discuss this in our
module; if you are interested, please look at Section 8.6 in Abbott’s textbook
“Understanding Analysis”. On a practical side, there are several models of real
numbers. For example, you can think of real numbers as decimal fractions, or
as points on the real line.
You are probably familiar with a larger number system ℂ of complex numbers.
We will not discuss it in this module.

4
Chapter 1
The real numbers

1.1 The well-ordering principle and induction

The set ℕ has a very important property that distinguishes it from the sets ℤ,
ℚ and ℝ:
Axiom 1.1 (The Well-Ordering Principle) Every non-empty subset of ℕ has a small-
est element.
△
Obviously, this property does not hold for the systems ℤ, ℚ and ℝ. For
example, the set {𝑥 ∈ ℤ (or ℚ, or ℝ) ∶ 𝑥 < 0} is non-empty, but has no smallest
element.
The Well-Ordering Principle has many important consequences. One of them
is the Principle of Mathematical Induction that is well familiar to you.
Theorem 1.2 (The Principle of Mathematical Induction) For each 𝑛 ∈ ℕ, let 𝑃(𝑛)
be a logical proposition depending on 𝑛 (i.e., 𝑃(𝑛) is a statement that is either true or
false, depending on the value of 𝑛). If 𝑃(1) is true and for each 𝑘 ∈ ℕ we have that
𝑃(𝑘) ⟹ 𝑃(𝑘 + 1), then 𝑃(𝑛) is true for all 𝑛 ∈ ℕ.
△

Proof. Consider the set 𝑆 = {𝑛 ∈ ℕ ∶ 𝑃(𝑛) is false}. We have to show that 𝑆 = ∅.

Suppose by contradiction that 𝑆 ≠ ∅. By the Well-Ordering Principle (Ax-
iom 1.1), 𝑆 has a smallest element, say 𝑚. Observe that 𝑚 ≠ 1, since 𝑃(1) is true.
But then 𝑚 = 𝑘 + 1 with some 𝑘 ∈ ℕ. Since 𝑚 is a smallest element in ℕ and
𝑘 < 𝑚, we have that 𝑘 ∉ 𝑆. In other words, 𝑃(𝑘) is true. But this implies for
𝑚 = 𝑘 + 1 that also 𝑃(𝑚) is true, which is a contradiction. The contradiction
shows that 𝑆 = ∅ as desired.
□

5
 Chapter 1 The real numbers

In fact, the Well-Ordering Principle and the Principle of Mathematical Induc-

tion are logically equivalent. We will not discuss this in detail.
Let us consider an example that shows a possible application of the Principle
of Mathematical Induction.
Example 1.3 Consider the sequence (𝑎𝑛 )∞
𝑛=1 defined recursively by

1
𝑎1 = ,
2
1
𝑎𝑛+1 = 𝑎2𝑛 + , 𝑛 ∈ ℕ.
5
We want to show that
1
0 ≤ 𝑎𝑛+1 ≤ 𝑎𝑛 ≤ for all 𝑛 ∈ ℕ.
2
We use the Principle of Mathematical Induction. Let 𝑃(𝑛) be the statement
that the above inequalities hold true for a particular 𝑛 ∈ ℕ. For the induction
2
1 1 9
base, we have to show that 𝑃(1) is true. Noting that 𝑎2 = � 2 � + 5
= 20
, we
observe that
9 1 1
0≤ ≤ ≤
20 ⏟
⏟ 2 2
=𝑎2 =𝑎1

as desired.
1
For the induction step, we assume that 0 ≤ 𝑎𝑘+1 ≤ 𝑎𝑘 ≤ 2
for a certain 𝑘 ∈ ℕ.
1
We have to show that 0 ≤ 𝑎𝑘+2 ≤ 𝑎𝑘+1 ≤ 2 . We know that the function 𝑓(𝑥) = 𝑥2
is increasing for 𝑥 ≥ 0. Thus, the inequality

1
0 ≤ 𝑎𝑘+1 ≤ 𝑎𝑘 ≤
2
implies, by taking a square of each term, that

1
0 ≤ 𝑎2𝑘+1 ≤ 𝑎2𝑘 ≤ .
4
1
Adding 5
to each term now gives

1 1 1 1 1
≤ 𝑎2𝑘+1 + ≤ 𝑎2𝑘 + ≤ + ,
5 5 5 4 5

6
 Chapter 1 The real numbers

1 1 1 1 1 9 1
and since 0 ≤ 5 , 𝑎2𝑘+1 + 5
= 𝑎𝑘+2 , 𝑎2𝑘 + 5
= 𝑎𝑘+1 and 4
+ 5
= 20
≤ 2 , we arrive at

1
0 ≤ 𝑎𝑘+2 ≤ 𝑎𝑘+1 ≤ .
2
By the Principle of Mathematical Induction we conclude that
1
0 ≤ 𝑎𝑛+1 ≤ 𝑎𝑛 ≤ for all 𝑛 ∈ ℕ.
2
△
We will now use the Well-Ordering Principle to prove that the set of rational
numbers ℚ is not “complete”. We will show that there exist lengths along the
real line that are not rational.
Theorem 1.4 There is no rational number 𝑟 with 𝑟2 = 2. In other words, √2 ∉ ℚ.
△

Proof.1 Suppose, by contradiction, that there is a rational number 𝑟 with 𝑟2 = 2.

Since 𝑟 and −𝑟 are simultaneously solutions of 𝑟2 = 2, we assume that 𝑟 > 0 and
𝑝
denote 𝑟 = √2 ∈ ℚ. Then there exist 𝑝, 𝑞 ∈ ℤ, 𝑞 ≠ 0, such that √2 = 𝑞 . We may
assume that 𝑞 > 0 (if this is not the case, we change the signs of both 𝑝 and 𝑞).
Thus, 𝑞 ∈ ℕ and 𝑝 = √2𝑞 ∈ ℤ.
Consider the set 𝑆 = �𝑞 ∈ ℕ ∶ √2𝑞 ∈ ℤ�. By our assumption, 𝑆 ≠ ∅. Therefore,
by the Well-Ordering Principle (Axiom 1.1), 𝑆 has a smallest element, say 𝑞0 ∈ ℕ.
The obvious inequality 1 < 2 < 4 implies, by taking square roots, 1 < √2 < 2.
Subtracting 1, we obtain 0 < √2 − 1 < 1. Now put

𝑞1 = (√2 − 1)𝑞0 .

We have 0 < 𝑞1 < 𝑞0 . Since 𝑞0 ∈ 𝑆, we have that √2𝑞0 ∈ ℤ, and thus 𝑞1 =

√2𝑞0 − 𝑞0 ∈ ℤ as a difference of two integers. Since 𝑞1 > 0, we conclude that
𝑞1 ∈ ℕ. Moreover,

√2𝑞1 = √2(√2 − 1)𝑞0 = 2𝑞0 − √2𝑞0 ∈ ℤ.

It follows that 𝑞1 ∈ 𝑆. But 𝑞1 < 𝑞0 and 𝑞0 is a smallest element in 𝑆, which is a

contradiction. The contradiction shows that our assumption that √2 ∈ ℚ was
wrong. We conclude that √2 ∉ ℚ.
□
1
Notice that a different proof of this theorem is given in Section 1.1 of Abbott’s textbook
“Understanding Analysis”. Read it!

7
 Chapter 1 The real numbers

Theorem 1.4 says that there are “holes” in the set of rational numbers ℚ. The
set of real numbers ℝ is an extension of ℚ that does not have such “holes”. This
will be discussed in the next section.

1.2 The completeness axiom

As we mentioned in Section 0.2, ℝ is an ordered field containing ℚ as a subfield.
Technically, there are many different ordered fields containing ℚ (for example,
there is ℚ[√2], the smallest field containing ℚ and √2, and further similar fields
— we are not going into details). Among all fields containing ℚ, the set of real
numbers ℝ is characterized by an important property that distinguishes ℝ from
all other possible extensions of ℚ. This property is the Axiom of Completeness
(see Axiom 1.8) and says that ℝ has no “holes”. Before we can formulate it, we
need a couple of definitions.
Definition 1.5 Let 𝐴 ⊆ ℝ be a nonempty set.

(1) We say that 𝐴 is bounded above if there exists 𝑀 ∈ ℝ such that 𝑎 ≤ 𝑀 for
all 𝑎 ∈ 𝐴. A number 𝑀 with this property is called an upper bound for 𝐴.

(2 ) Similarly, we say that 𝐴 is bounded below if there exists 𝑚 ∈ ℝ such that

𝑚 ≤ 𝑎 for all 𝑎 ∈ 𝐴. A number 𝑚 with this property is called a lower bound
for 𝐴.

(3 ) We say that 𝐴 is bounded if 𝐴 is bounded below and above. In other words,

there exist 𝑚, 𝑀 ∈ ℝ such that 𝑚 ≤ 𝑎 ≤ 𝑀 for all 𝑎 ∈ 𝐴.

△
Upper and lower bounds are not unique, and they need not belong to 𝐴.
Example 1.6 Consider the set

𝐴 = {𝑥 ∈ ℝ ∶ 𝑥2 < 2}.

Since 0 ∈ 𝐴, 𝐴 is nonempty. The set 𝐴 is bounded above. For example, 3 is an

upper bound for 𝐴, i.e. 𝑥 ≤ 3 for all 𝑥 ∈ 𝐴. Indeed, if 𝑥 > 3, then 𝑥2 > 9 > 2, and
therefore 𝑥 ∉ 𝐴. It follows that for each 𝑥 ∈ 𝐴 we have 𝑥 ≤ 3. Similarly, 2 is an
upper bound for 𝐴: if 𝑥 > 2, then 𝑥2 > 4 > 2, and so 𝑥 ∉ 𝐴. In fact, any 𝑀 ≥ 0
with 𝑀2 ≥ 2 is an upper bound for 𝐴.
△

8
 Chapter 1 The real numbers

Definition 1.7 Let 𝐴 ⊆ ℝ be a nonempty set.

(1) We say that 𝑀 ∈ ℝ is a least upper bound (a supremum) of 𝐴 if

( i ) 𝑀 is an upper bound for 𝐴,
( i i ) if 𝑁 is any upper bound for 𝐴, then 𝑀 ≤ 𝑁.
The supremum of 𝐴 is denoted by 𝑀 = sup 𝐴.

(2 ) Similarly, we say that 𝑚 ∈ ℝ is a greatest lower bound (a infimum) of 𝐴 if

( i ) 𝑚 is a lower bound for 𝐴,
( i i ) if 𝑛 is any lower bound for 𝐴, then 𝑚 ≥ 𝑛.
The infimum of 𝐴 is denoted by 𝑚 = inf 𝐴.

△
The least upper bound of a set, if it exists, is unique. Indeed, suppose that
𝑀1 and 𝑀2 are two least upper bounds for 𝐴. Then by the second property
in (1) in Definition 1.7 we have 𝑀1 ≤ 𝑀2 and, on the other hand, 𝑀2 ≤ 𝑀1 . It
follows that 𝑀1 = 𝑀2 . By a similar argument, the greatest lower bound of a set,
if exists, is unique.
Now we are in position to formulate the Completeness Axiom that is a
defining property of the set ℝ.
Axiom 1.8 (The Completeness Axiom) Every nonempty subset of ℝ that is bounded
above has a least upper bound in ℝ.
△
Theorem 1.9 There exists a real number 𝑥 with 𝑥2 = 2. In other words, √2 ∈ ℝ.
△

Proof. Consider the set

𝐴 = {𝑡 ∈ ℝ ∶ 𝑡2 < 2}.
We discussed in Example 1.6 that 𝐴 is nonempty and bounded above. By the
Completeness Axiom (Axiom 1.8) 𝐴 has a least upper bound. Put 𝑥 = sup 𝐴.
Clearly, 𝑥 ≥ 0, since 0 ∈ 𝐴. We will show that 𝑥2 = 2 by proving that both
possibilities 𝑥2 < 2 and 𝑥2 > 2 cannot occur.
Suppose first that 𝑥2 < 2. Since 2 is an upper bound for 𝐴 and 𝑥 is the least
upper bound, we have 𝑥 ≤ 2. Take ℎ ∈ (0, 1), then ℎ2 < ℎ and
(𝑥 + ℎ)2 = 𝑥2 + 2 ⏟ ℎ2 < 𝑥2 + 4ℎ + ℎ = 𝑥2 + 5ℎ.
𝑥 ℎ +⏟
≤2 <ℎ

9
 Chapter 1 The real numbers

Choose ℎ ∈ (0, 1) so small that 5ℎ < 2 − 𝑥2 (this is possible because, by our

assumption, 𝑥2 < 2). Then we have
(𝑥 + ℎ)2 < 𝑥2 + 5ℎ < 𝑥2 + 2 − 𝑥2 = 2,
i.e. (𝑥 + ℎ)2 < 2. This implies that 𝑥 + ℎ ∈ 𝐴. But 𝑥 + ℎ > 𝑥, and this contradicts
the fact that 𝑥 is an upper bound for 𝐴. The contradiction shows that it is
impossible that 𝑥2 < 2.
Now suppose that 𝑥2 > 2. For ℎ > 0 we have
(𝑥 − ℎ)2 = 𝑥2 − 2 ⏟ ℎ2 > 𝑥2 − 4ℎ.
𝑥 ℎ +⏟
≤2 >0

Choose ℎ > 0 so small that 4ℎ < 𝑥2 − 2 and ℎ < 𝑥. Then

(𝑥 − ℎ)2 > 𝑥2 − 4ℎ > 𝑥2 − (𝑥2 − 2) = 2.
It follows that 𝑥 − ℎ is an upper bound for 𝐴 because for any 𝑎 ≥ 𝑥 − ℎ > 0 we
would have 𝑎2 ≥ (𝑥 − ℎ)2 > 2, so that 𝑎 ∉ 𝐴. But 𝑥 − ℎ < 𝑥 and this contradicts
the fact that 𝑥 is the least upper bound for 𝐴. The contradiction shows that it is
impossible that 𝑥2 > 2.
The above implies that 𝑥2 = 2, i.e. 𝑥 = √2 ∈ ℝ.
□

Remark The Completeness Axiom does not hold for ℚ. For example, the set
𝐵 = {𝑟 ∈ ℚ ∶ 𝑟2 < 2}
is nonempty and bounded above, but it does not have a least upper bound in
ℚ. Indeed, the same argument as above shows that a least upper bound 𝑀 ∈ ℚ,
if it exists, cannot satisfy 𝑀2 < 2 and 𝑀2 > 2. But we showed in Theorem 1.4
that there is no rational number 𝑀 ∈ ℚ with 𝑀2 = 2. Thus, 𝐵 does not have a
least upper bound in ℚ.
△
We continue this section by discussing some properties of least upper bounds
and greatest lower bounds.
Let us consider intervals on the real line. If 𝑎, 𝑏 ∈ ℝ, 𝑎 < 𝑏, then
sup [𝑎, 𝑏] = sup (𝑎, 𝑏) = sup (−∞, 𝑏] = sup (−∞, 𝑏) = 𝑏,
inf [𝑎, 𝑏] = inf (𝑎, 𝑏) = inf [𝑎, ∞) = inf (𝑎, ∞) = 𝑎.
We see that sup 𝐴 and inf 𝐴 may or may not belong to 𝐴. This leads us to the
following definition.

10
 Chapter 1 The real numbers

Definition 1.10 Let 𝐴 ⊆ ℝ be a nonempty set. We say that 𝑀 is a maximum of 𝐴

if 𝑀 ∈ 𝐴 and 𝑎 ≤ 𝑀 for all 𝑎 ∈ 𝐴. We denote the maximum of 𝐴 by 𝑀 = max 𝐴.
Similarly, 𝑚 is a minimum of 𝐴 if 𝑚 ∈ 𝐴 and 𝑎 ≥ 𝑚 for all 𝑎 ∈ 𝐴. We denote the
minimum of 𝐴 by 𝑚 = min 𝐴.
△
If 𝐴 is nonempty and bounded above, then we know by the Completeness
Axiom that sup 𝐴 exists. However, max 𝐴 may or may not exist. For example,
max [𝑎, 𝑏] = 𝑏, but max (𝑎, 𝑏) does not exist. If max 𝐴 exists, then max 𝐴 = sup 𝐴.
If sup 𝐴 ∈ 𝐴, then sup 𝐴 = max 𝐴. Similar statements hold, of course, for inf 𝐴
and min 𝐴.
A very useful reformulation of the second property in Definition 1.7 (1) is
given in the following statement.
Proposition 1.11( 1 ) Let 𝑀 ∈ ℝ be an upper bound of a nonempty set 𝐴 ⊆ ℝ.
Then 𝑀 = sup 𝐴 if and only if for any 𝜀 > 0 there is an element 𝑎 ∈ 𝐴 with
𝑎 > 𝑀 − 𝜀.

(2 ) Let 𝑚 ∈ ℝ be a lower bound of a nonempty set 𝐴 ⊆ ℝ. Then 𝑚 = inf 𝐴 if and

only if for any 𝜀 > 0 there is an element 𝑎 ∈ 𝐴 with 𝑎 < 𝑚 + 𝜀.
△

Proof. We prove (1), the proof of (2) is left as an exercise.

a. For the forward direction, assume that 𝑀 = sup 𝐴. Take any 𝜀 > 0, then
𝑀 − 𝜀 is not an upper bound for 𝐴, since 𝑀 − 𝜀 < 𝑀 and 𝑀 is the smallest upper
bound. Thus, there must exist an element 𝑎 ∈ 𝐴 with 𝑎 > 𝑀 − 𝜀.

b. Now assume that 𝑀 is an upper bound for 𝐴 and that for any 𝜀 > 0 there is an
element 𝑎 ∈ 𝐴 with 𝑎 > 𝑀 − 𝜀. Let 𝑁 < 𝑀. Then 𝑁 = 𝑀 − 𝜀 with 𝜀 = 𝑀 − 𝑁 > 0.
By our assumption, there exists 𝑎 ∈ 𝐴 with 𝑎 > 𝑁. It follows that 𝑁 is not an
upper bound for 𝐴. We conclude that for any upper bound 𝑁 of 𝐴 we must
have 𝑁 ≥ 𝑀. But this is exactly the second property in Definition 1.7 (1). This
proves that 𝑀 = sup 𝐴.

Suprema and infima obey a number of rules. A typical example is the follow-
ing rule: If 𝐴 ⊆ ℝ is a nonempty set that it bounded above and 𝐵 = {𝑥+𝑎 ∶ 𝑎 ∈ 𝐴},
where 𝑥 ∈ ℝ, then
sup 𝐵 = 𝑥 + sup 𝐴.

11
 Chapter 1 The real numbers

𝑎1 𝑎2 𝑎3 𝑎4 ⋯ 𝑏4 𝑏3 𝑏2 𝑏1

Figure 1.1: Nested intervals

We leave the proof as an exercise.

If 𝐴 ⊆ ℝ is a nonempty set that is bounded above, then

−𝐴 = {−𝑎 ∶ 𝑎 ∈ 𝐴}

is nonempty and bounded below. It is not difficult to show using the definition
that
inf (−𝐴) = − sup 𝐴.
Using this property, properties of smallest upper bounds can be reformulated
as properties of greatest lower bounds. In particular, we could formulate the
Completeness Axiom in terms of greatest lower bounds: Every nonempty
subset of ℝ that is bounded below has a greatest lower bound in ℝ.

1.3 Consequences of completeness

The following statement shows again that there are no “holes” in ℝ.
∞
Theorem 1.12 (The Nested Interval Property) Let �[𝑎𝑛 , 𝑏𝑛 ]� be a sequence of
𝑛=1
nonempty closed bounded intervals in ℝ such that

[𝑎1 , 𝑏1 ] ⊇ [𝑎2 , 𝑏2 ] ⊇ [𝑎3 , 𝑏3 ] ⊇ ⋯ .

∞
Then ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ] ≠ ∅.
△
The name of the theorem comes from the fact that each interval is contained
in all its predecessors (see Figure 1.1).
Before proving the theorem, we consider an example.
Example 1.13 Consider the intervals

1 1
[𝑎𝑛 , 𝑏𝑛 ] = �√2 − , √2 + � , 𝑛 ∈ ℕ.
𝑛 𝑛

12
 Chapter 1 The real numbers

∞ 1 1
By the Nested Interval Property, ⋂𝑛=1 �√2 − 𝑛 , √2 + 𝑛 � ≠ ∅. It is easy to see
1 1
this directly. Indeed, √2 ∈ �√2 − 𝑛 , √2 + 𝑛 � for all 𝑛 ∈ ℕ, and therefore √2 ∈

⋂∞ �√2 − 1 , √2 + 1 �. In fact, ⋂∞ �√2 − 1 , √2 + 1 � = �√2�, we will prove this

𝑛=1 𝑛 𝑛 𝑛=1 𝑛 𝑛
shortly in Example 1.16.
△

Proof of Theorem 1.12. Put 𝐴 = {𝑎𝑛 ∶ 𝑛 ∈ ℕ}. The set 𝐴 is bounded above by 𝑏1 .
By the Completeness Axiom (Axiom 1.8), 𝐴 has a least upper bound 𝑀. We
∞
claim that 𝑀 ∈ ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ].
Indeed, since 𝑀 is an upper bound for 𝐴, we have 𝑎𝑛 ≤ 𝑀 for all 𝑛 ∈ ℕ. But
each 𝑏𝑛 , 𝑛 ∈ ℕ, is also an upper bound for 𝐴. Since 𝑀 is the least upper bound,
∞
𝑀 ≤ 𝑏𝑛 , 𝑛 ∈ ℕ. Thus, 𝑀 ∈ [𝑎𝑛 , 𝑏𝑛 ] for all 𝑛 ∈ ℕ, and therefore 𝑀 ∈ ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ].
□

Notice that the set of natural numbers ℕ does not have a maximum element.
Indeed, for each 𝑛 ∈ ℕ we can take 𝑛 + 1 ∈ ℕ which is larger than 𝑛. It is less
obvious that ℕ is not bounded above as a subset of ℝ, that is, there is no 𝑥 ∈ ℝ
that is larger than all natural numbers. This result is established in the next
statement.
Theorem 1.14 (The Archimedean Property) Given any 𝑥 ∈ ℝ, there exists a natural
number 𝑛 ∈ ℕ with 𝑛 > 𝑥.
△

Proof. Suppose, by contradiction, that there exists a number 𝑥 ∈ ℝ such that

𝑥 ≥ 𝑛 for all 𝑛 ∈ ℕ. This means that ℕ is bounded above. By the Completeness
Axiom (Axiom 1.8), ℕ has a least upper bound 𝑀.
The number 𝑀 − 1 is not an upper bound for ℕ, because 𝑀 − 1 < 𝑀 and 𝑀 is
the least upper bound. This means that there is a number 𝑛 ∈ ℕ with 𝑛 > 𝑀 − 1.
But then 𝑛 + 1 > 𝑀 and 𝑛 + 1 ∈ ℕ, so that 𝑀 is not an upper bound for ℕ. The
contradiction shows that our assumption was wrong and ℕ is not bounded
above in ℝ.
□
1
Corollary 1.15 For each 𝜀 > 0 there exists 𝑛 ∈ ℕ with 0 < < 𝜀.
𝑛
△

13
 Chapter 1 The real numbers

1 1
Proof. Take 𝑥 = 𝜀
∈ ℝ in Theorem 1.14. So there is 𝑛 ∈ ℕ with 𝑛 > 𝜀
> 0. This
1
is equivalent to 0 < 𝑛
< 𝜀.
□

Example 1.16 This is a continuation of Example 1.13. We are now in a position

to prove that
∞
1 1
� �√2 − , √2 + � = �√2� .
𝑛=1
𝑛 𝑛
∞ 1 1 1
Let 𝑥 ∈ ⋂𝑛=1 �√2 − 𝑛 , √2 + 𝑛 �. Then �𝑥 − √2� ≤ 𝑛
for each 𝑛 ∈ ℕ. By Corol-
lary 1.15 this is only possible if �𝑥 − √2� = 0, i.e., 𝑥 = √2.
△
Let us turn our attention to the Nested Interval Property (Theorem 1.12)
again.
Remark In Theorem 1.12 we require that the nested intervals [𝑎𝑛 , 𝑏𝑛 ] are closed
and bounded. These assumptions are needed for the theorem to hold. Taking
onto account Theorem 1.14 and Corollary 1.15, we can consider the following
examples.
The intervals [𝑛, ∞), 𝑛 ∈ ℕ, are nested and closed, but unbounded. We have
⋂∞ [𝑛, ∞) = ∅.
𝑛=1
1
On the other hand, the intervals �0, 𝑛 �, 𝑛 ∈ ℕ, are nested and bounded, but
∞ 1
they are not closed. Again we have ⋂𝑛=1 �0, 𝑛 � = ∅.
Of course we also cannot remove the assumptions that the intervals are nested.
For example, the intervals [𝑛, 𝑛 + 1], 𝑛 ∈ ℕ, are closed and bounded, but not
∞
nested, and ⋂𝑛=1 [𝑛, 𝑛 + 1] = ∅.
△
Exactly as the Completeness Axiom (Axiom 1.8), the Nested Interval Property
does not hold if we replace ℝ by ℚ. This can be demonstrated by the following
example.
Example 1.17 Consider the intervals in ℚ

1 1 1 1
𝐼𝑛 = ℚ ∩ �√2 − , √2 + � = �𝑟 ∈ ℚ ∶ √2 − ≤ 𝑟 ≤ √2 + � , 𝑛 ∈ ℕ.
𝑛 𝑛 𝑛 𝑛

14
 Chapter 1 The real numbers

∞
Their intersection is empty: ⋂𝑛=1 𝐼𝑛 = ∅. For,
∞ ∞
1 1
� 𝐼𝑛 ⊆ � �√2 − , √2 + � = �√2� ,
𝑛=1 𝑛=1
𝑛 𝑛

but √2 ∉ ℚ. Thus, the Nested Interval Property in Theorem 1.12 does not hold
if we replace ℝ by ℚ.
△
We will now use the Archimedean Property to show that ℚ is dense in ℝ.
This means that any interval (𝑎, 𝑏) ⊆ ℝ contains rational numbers.
Theorem 1.18 Given any two real numbers 𝑎, 𝑏 ∈ ℝ with 𝑎 < 𝑏, there is a rational
number 𝑟 ∈ ℚ satisfying 𝑎 < 𝑟 < 𝑏.
△

Proof. It suffices to prove the statement for 0 < 𝑎 < 𝑏. Indeed, if 𝑎 ≤ 0, then
by the Archimedean Property (Theorem 1.14) there exists 𝑘 ∈ ℕ with 𝑘 > −𝑎.
Then the numbers 𝑎 + 𝑘 and 𝑏 + 𝑘 satisfy 0 < 𝑎 + 𝑘 < 𝑏 + 𝑘. If we can find 𝑟 ̃ ∈ ℚ
such that 𝑎 + 𝑘 < 𝑟 ̃ < 𝑏 + 𝑘, then for the number 𝑟 = 𝑟 ̃ − 𝑘 we have 𝑟 ∈ ℚ and
𝑎 < 𝑟 < 𝑏 as desired.
Therefore, assume that 0 < 𝑎 < 𝑏. By Corollary 1.15, there exists 𝑛 ∈ ℕ with
1
0 < 𝑛 < 𝑏 − 𝑎. Put 𝑆 = {𝑚 ∈ ℕ ∶ 𝑎𝑛 < 𝑚}. By the Archimedean Property
(Theorem 1.14) 𝑆 ≠ ∅, and by the Well-Ordering Principle (Axiom 1.1) 𝑆 has
𝑚
a smallest element 𝑚0 ∈ ℕ. Put 𝑟 = 𝑛0 . Since 𝑚0 ∈ 𝑆 and 𝑚0 − 1 ∉ 𝑆, we
1
have 𝑚0 − 1 ≤ 𝑎𝑛 < 𝑚0 . This implies 𝑟 − 𝑛
≤ 𝑎 < 𝑟. The first inequality gives
1
𝑟≤𝑎+ 𝑛
< 𝑎 + (𝑏 − 𝑎) = 𝑏, and altogether we obtain 𝑎 < 𝑟 < 𝑏.
□

Given two real numbers 𝑎, 𝑏 with 𝑎 < 𝑏, we can find a number 𝑟 ∈ ℚ such
that 𝑎 + √2 < 𝑟 < 𝑏 + √2. Then 𝑥 = 𝑟 − √2 is an irrational number satisfying
𝑎 < 𝑥 < 𝑏. This proves the following statement.
Corollary 1.19 Given any two real numbers 𝑎, 𝑏 ∈ ℝ with 𝑎 < 𝑏, there is an irrational
number 𝑥 satisfying 𝑎 < 𝑥 < 𝑏.
△
Thus, the set of irrational numbers ℝ ⧵ ℚ is also dense in ℝ.

15
 Chapter 1 The real numbers

1.4 Cardinality
We have seen in the previous section that both the rational and the irrational
numbers are dense in ℝ. In another sense, there are far more irrational numbers
than rational numbers. We will discuss this in this section.
Recall that a function is a rule that assigns to every element 𝑥 of a given set
𝑋 a unique element 𝑦 in a given set 𝑌. We write 𝑓 ∶ 𝑋 → 𝑌. We denote the
element 𝑦 ∈ 𝑌 that is assigned to 𝑥 ∈ 𝑋 by 𝑓(𝑥) and call 𝑦 = 𝑓(𝑥) the value of 𝑓 at
𝑥. The set 𝑋 is called the domain of 𝑓, and 𝑌 is called the codomain.
Definition 1.20 Let 𝑓 ∶ 𝑋 → 𝑌 be a function.

(1) 𝑓 is called injective (a one-to-one function), if

𝑓(𝑥1 ) = 𝑓(𝑥2 ) ⟹ 𝑥1 = 𝑥2

for all 𝑥1 , 𝑥2 ∈ 𝑋.

(2 ) 𝑓 is called surjective (a function onto), if for each 𝑦 ∈ 𝑌 there is an element

𝑥 ∈ 𝑋 with 𝑓(𝑥) = 𝑦.

(3 ) 𝑓 is called a bijection (a one-to-one correspondence), if 𝑓 is both injective and

surjective.

△
Whether or not a function 𝑓 ∶ 𝑋 → 𝑌 is surjective depends on its codomain:
𝑓 is surjective if and only if 𝑌 = {𝑓(𝑥) ∶ 𝑥 ∈ 𝑋}. On the other hand, whether
or not 𝑓 ∶ 𝑋 → 𝑌 is injective depends on its domain: 𝑓 is injective if and only
if for each 𝑦 ∈ 𝑌 there is at most one 𝑥 ∈ 𝑋 with 𝑓(𝑥) = 𝑦. Combining these
two properties we state that 𝑓 is a bijection if and only if for each 𝑦 ∈ 𝑌 there is
exactly one 𝑥 ∈ 𝑋 with 𝑓(𝑥) = 𝑦.
You know that if 𝑓 ∶ 𝑋 → 𝑌 is a bijection, then there exists the inverse function
𝑓 ∶ 𝑌 → 𝑋 such that 𝑓−1 ∘ 𝑓(𝑥) = 𝑥 for all 𝑥 ∈ 𝑋, 𝑓 ∘ 𝑓−1 (𝑦) = 𝑦 for all 𝑦 ∈ 𝑌. If 𝑓
−1

is a bijection, then 𝑓−1 is also a bijection.

Definition 1.21 We say that two sets 𝐴 and 𝐵 have the same cardinality if there
exists a bijection 𝑓 ∶ 𝐴 → 𝐵. In this case we write 𝐴 ∼ 𝐵.
△
Since 𝑓 ∶ 𝐴 → 𝐵 is a bijection if and only if 𝑓−1 ∶ 𝐵 → 𝐴 is a bijection, it does
not matter whether we construct a bijection from 𝐴 to 𝐵 or from 𝐵 to 𝐴. It also
follows that 𝐴 ∼ 𝐵 ⟺ 𝐵 ∼ 𝐴. Since the identity function 𝐴 → 𝐴 is a bijection,

16
 Chapter 1 The real numbers

we always have 𝐴 ∼ 𝐴. Finally, since a composition of two bijections is again a

bijection, we have 𝐴 ∼ 𝐵 and 𝐵 ∼ 𝐶 ⟹ 𝐴 ∼ 𝐶. (These three properties say, in
fact, that ∼ is an equivalence relation — we do not go into details here.)
Definition 1.22 A nonempty set 𝐴 is called finite if 𝐴 ∼ {1, 2, … , 𝑛} with some
𝑛 ∈ ℕ. In this case we call 𝑛 the cardinality of 𝐴. The empty set ∅ is regarded as
being finite with cardinality zero.
△
It is quite easy to show that two finite sets have the same cardinality if and
only if they have the same number of elements. The concept of cardinality was
introduced as a way to describe sizes of infinite sets.
Definition 1.23 ( 1 ) A set 𝐴 is called countable if 𝐴 ∼ ℕ.

(2 ) A set that is neither finite nor countable is called uncountable.

△
Notice that according to our terminology a countable set is infinite, i.e. finite
sets are not regarded as countable. Sets that are either finite or countable are
called at most countable. The terminology here is not consistent, so please be
careful when using other books.
We will now spend some time discussing countable sets. Definition 1.23 says
that 𝐴 is countable if there exists a bijection 𝑓 ∶ ℕ → 𝐴. This means that the
elements of 𝐴 can be listed as a sequence 𝑓(1), 𝑓(2), 𝑓(3), ….
Example 1.24 ℤ is countable because we can list its elements as a sequence:

0, 1, −1, 2, −2, 3, −3, … .

△
Example 1.25 The set of even natural numbers is countable because we can
list its elements as a sequence 2, 4, 6, 8, …. Similarly, the set of prime numbers is
countable: 2, 3, 5, 7, ….
△
In fact, any infinite subset of ℕ is countable. This follows from a more general
statement.
Proposition 1.26 Any subset of a countable set is either finite or countable.
△

Proof. Exercise. □

17
 Chapter 1 The real numbers

The intersection of two countable sets is a subset of both sets, so it must be

either finite or countable. (It can indeed be finite: for example, the intersection
of the countable set of even natural numbers with the countable set of prime
numbers consists of one element 2.)
Proposition 1.27 The union of two countable sets is countable.
△

Proof. Let 𝐴 and 𝐵 be two countable sets. We can list their elements as se-
quences:
𝐴 ∶ 𝑎1 , 𝑎 2 , 𝑎 3 , 𝑎 4 , … ,
𝐵 ∶ 𝑏1 , 𝑏2 , 𝑏3 , 𝑏4 , … .
Now we can list the elements of 𝐴 ∪ 𝐵 as follows: first write
𝐴 ∪ 𝐵 ∶ 𝑎 1 , 𝑏1 , 𝑎2 , 𝑏2 , 𝑎3 , 𝑏3 , … ,
and then remove the repeats. This shows that 𝐴 ∪ 𝐵 is countable.
□

Using induction, we can show that the union of finitely many countable sets
is countable.
Let us now discuss the sets ℚ and ℝ. The following result is perhaps surpris-
ing.
Theorem 1.28 The set ℚ is countable.
△

Proof.2 First we show that we can list positive rational numbers as a sequence.
𝑚
In the first step we arrange numbers of the form 𝑛 , 𝑚, 𝑛 ∈ ℕ, as an infinite
square table:
1 1 1 1
1 2 3 4
⋯
↙ ↙ ↙
2 2 2 2
1 2 3 4
⋯
↙ ↙ ↙
3 3 3 3
1 2 3 4
⋯
↙ ↙ ↙
4 4 4 4
1 2 3 4
⋯
⋮ ⋮ ⋮ ⋮ ⋱
2
Notice that a somewhat different proof of this theorem is given in Section 1.5 of Abbott’s
textbook “Understanding Analysis”.

18
 Chapter 1 The real numbers

Now we list these numbers as a sequence along south-west diagonals:

1 1 2 1 2 3 1 2 3 4 1 2 3 4 5
, , , , , , , , , , , , , , ,….
1 2 1 3 2 1 4 3 2 1 5 4 3 2 1
Note that this sequence contains all positive real numbers. However, each
number comes more than once. We delete the repeats and obtain the desired
sequence
1 1 1 2 3 1
1, , 2, , 3, , , , 4, , 5, … .
2 3 4 3 2 5
In this way, we have listed all positive rational numbers as a sequence:

{𝑞 ∈ ℚ ∶ 𝑞 > 0} = {𝑞1 , 𝑞2 , 𝑞3 , 𝑞4 , …}.

To finish the proof, we list ℚ as the sequence

ℚ = {0, 𝑞1 , −𝑞1 , 𝑞2 , −𝑞2 , 𝑞3 , −𝑞3 , …}.

Combining ideas from the proof of Proposition 1.27 and Theorem 1.28, one
can prove that the union of countable many countable sets is countable. We
will not discuss this in the module, but try to prove it.
Now we turn our attention to the set of real numbers ℝ. We will show that it
is much larger than ℚ.
Theorem 1.29 The set ℝ is uncountable.
△

Proof. Suppose, by contradiction, that ℝ is countable. If this is the case, we are

able to list its elements as a sequence:

ℝ = {𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , …}.

Consider 𝑥1 . Take a closed and bounded interval [𝑎1 , 𝑏1 ] with 𝑎1 < 𝑏1 such
that 𝑥1 ∉ [𝑎1 , 𝑏1 ]. Now we split the interval [𝑎1 , 𝑏1 ] into three thirds [𝑎1 , 𝑐1 ],
[𝑐1 , 𝑑1 ], [𝑑1 , 𝑏1 ]. At least one of these three intervals does not contain 𝑥2 .3 We
denote this interval as [𝑎2 , 𝑏2 ].

3
Notice that it is not enough to divide [𝑎1 , 𝑏1 ] into two halves: if 𝑥2 happens to be exactly the
midpoint of [𝑎1 , 𝑏1 ], it will belong to both halves!

19
 Chapter 1 The real numbers

Now we have 𝑎1 ≤ 𝑎2 < 𝑏2 ≤ 𝑏1 , 𝑥1 ∉ [𝑎1 , 𝑏1 ], 𝑥2 ∉ [𝑎2 , 𝑏2 ]. We continue

inductively. At the 𝑛-th step we divide the interval [𝑎𝑛 , 𝑏𝑛 ] into three thirds and
choose [𝑎𝑛+1 , 𝑏𝑛+1 ] to be a third that does not contain 𝑥𝑛+1 .
In this way we obtain a sequence of nested intervals

[𝑎1 , 𝑏1 ] ⊇ [𝑎2 , 𝑏2 ] ⊇ [𝑎3 , 𝑏3 ] ⊇ ⋯

with the property 𝑥𝑛 ∉ [𝑎𝑛 , 𝑏𝑛 ], 𝑛 ∈ ℕ.

∞
By the Nested Interval Property (Theorem 1.12) we have ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ] ≠ ∅. In
∞
other words, there exists a real number 𝑥 with 𝑥 ∈ ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ]. Since 𝑥 ∈ ℝ, 𝑥
is counted in our list, i.e. 𝑥 = 𝑥𝑚 with some 𝑚 ∈ ℕ. But then, by construction,
∞
𝑥 = 𝑥𝑚 ∉ [𝑎𝑚 , 𝑏𝑚 ]. It follows that 𝑥 ∉ ⋂𝑛=1 [𝑎𝑛 , 𝑏𝑛 ] which is a contradiction. The
contradiction shows that it is not possible to list the elements of ℝ as a sequence,
i.e., ℝ is uncountable.
□

Since the union of two countable sets is countable, we arrive at the following
conclusion.
Corollary 1.30 The set of irrational numbers ℝ ⧵ ℚ is uncountable.
△

20
Chapter 2
Sequences and series

2.1 Limits of sequences

Definition 2.1 A sequence in a set 𝐴 is a function 𝑓 ∶ ℕ → 𝐴.
△
It is common to use the notation 𝑎𝑛 = 𝑓(𝑛), 𝑛 ∈ ℕ. The elements 𝑎𝑛 are called
terms of the sequence. We will frequently write for a sequence (𝑎𝑛 )∞
𝑛=1 , (𝑎𝑛 )𝑛∈ℕ
or simply (𝑎𝑛 ).
In this module we will be mainly concerned with real sequences 𝑓 ∶ ℕ → ℝ.
Examples of sequences are

(𝑎𝑛 )∞
𝑛=1 = (1, −1, 1, −1, 1, −1, …),
∞
(𝑏𝑛 )𝑛=1 = (−1, 1, −1, 1, −1, 1, …).

Notice that (𝑎𝑛 )∞ ∞

𝑛=1 and (𝑏𝑛 )𝑛=1 are two different sequences. For example, 𝑎1 = 1
while 𝑏1 = −1. Both sequences (𝑎𝑛 )∞ ∞
𝑛=1 and (𝑏𝑛 )𝑛=1 have the same set of values
{−1, 1}.
In some cases, it is convenient to start a sequence with 𝑛 = 0, or with any
other 𝑛 = 𝑛0 ∈ ℤ. As an example, consider the sequence
∞
1 1 1 1 1
� 𝑛� = �1, , , , , ⋯� .
2 2 4 8 16
𝑛=0

It is possible to draw a graph of a function 𝑓 ∶ ℕ → ℝ. Such a graph consists

of isolated points. We will see examples below.
One of the most important concepts of analysis is that of convergence and of
a limit. Intuitively, a sequence (𝑎𝑛 )∞
𝑛=1 converges to a limit 𝑎 ∈ ℝ if its terms get
closer and closer to 𝑎 as 𝑛 goes to infinity. A precise definition is given below;
this is one of the most important definitions in this module — learn it.

21
 Chapter 2 Sequences and series

𝑎+𝜀
𝑎
𝑎−𝜀

1 2 3 4 𝑁 𝑛

Figure 2.1: A graph of a converging sequence

Definition 2.2 We say that a sequence (𝑎𝑛 )∞

𝑛=1 converges to a limit 𝑎 ∈ ℝ if

∀ 𝜀 > 0 ∃𝑁 ∈ ℕ ∀ 𝑛 ≥ 𝑁 ∶ |𝑎𝑛 − 𝑎| < 𝜀,

i.e. for any positive number 𝜀 there exists 𝑁 ∈ ℕ such that |𝑎𝑛 − 𝑎| < 𝜀 for all
𝑛 ∈ ℕ with 𝑛 ≥ 𝑁. In this case we write 𝑎 = lim𝑛→∞ 𝑎𝑛 , or 𝑎𝑛 → 𝑎 as 𝑛 → ∞.
△
The definition means that for any 𝜀 > 0 all terms of the sequence but finitely
many (namely, all terms 𝑎𝑛 with 𝑛 ≥ 𝑁) lie between 𝑎 − 𝜀 and 𝑎 + 𝜀. We can
visualize this behavior on the graph of a sequence — see Figure 2.1. For a
smaller value of 𝜀 > 0, in general a larger value of 𝑁 ∈ ℕ is required. But no
matter how small 𝜀 > 0 is, we can always find an 𝑁 ∈ ℕ with this property.
∞
𝑛+1
Example 2.3 Consider the sequence � � .
𝑛
𝑛=1
3 4 5 6
The first few terms of the sequence are �2, 2 , 3 , 4 , 5 , ⋯�. We conjecture that
the sequence converges to the limit 1. For the proof we consider
𝑛+1 1 1
|𝑎𝑛 − 1| = � − 1� = �1 + − 1� = .
𝑛 𝑛 𝑛
Given 𝜀 > 0, by the Archimedean Property (Theorem 1.14) we can find a
1 1
number 𝑁 ∈ ℕ such that 𝜀 < 𝑁. Then 𝑁 < 𝜀, and consequently for all 𝑛 ≥ 𝑁
we have
1 1
|𝑎𝑛 − 1| = ≤ < 𝜀.
𝑛 𝑁
𝑛+1
We conclude that lim𝑛→∞ 𝑛
= 1.

22
 Chapter 2 Sequences and series

Let us look at a couple of concrete values of 𝜀. For example, for 𝜀 = 0.1 we

can take 𝑁 = 11, for 𝜀 = 0.01 we can take 𝑁 = 101, for 𝜀 = 0.005 we can take
𝑁 = 201, and so on.
△
Negating the statement in Definition 2.2, we can formulate a property that a
sequence (𝑎𝑛 )∞
𝑛=1 does not converge to 𝑏 ∈ ℝ:

∃𝜀 > 0 ∀ 𝑁 ∈ ℕ ∃𝑛 ∈ ℕ ∶ 𝑛 ≥ 𝑁 and |𝑎𝑛 − 𝑏| ≥ 𝜀.

∞
𝑛+1
As an example, let us show that the sequence � 𝑛
� does not converge to 0.
𝑛=1
We have
𝑛+1 𝑛+1
|𝑎𝑛 − 0| = � − 0� = >1
𝑛 𝑛
for all 𝑛 ∈ ℕ. We can choose 𝜀 = 1. Then for any 𝑁 ∈ ℕ we can find 𝑛 ≥ 𝑁 —
∞
𝑛+1
for example, 𝑛 = 𝑁 — with |𝑎𝑛 − 0| > 𝜀 = 1. Thus, � 𝑛
� does not converge
𝑛=1
to 0.
𝑛+1
In fact, this follows already from lim𝑛→∞ 𝑛 = 1, since a sequence cannot
have two different limits. Namely, we have the following statement.
Proposition 2.4 The limit of a sequence, if it exists, is unique.
△

Proof. Suppose that the numbers 𝑎, 𝑏 ∈ ℝ are both limits of a sequence (𝑎𝑛 )∞
𝑛=1 .
𝜀
Given 𝜀 > 0, we can find 𝑁1 , 𝑁2 ∈ ℕ such that |𝑎𝑛 − 𝑎| < 2 for all 𝑛 ≥ 𝑁1 and
𝜀
|𝑎𝑛 − 𝑏| < 2 for all 𝑛 ≥ 𝑁2 . Now take a number 𝑛0 ≥ max{𝑁1 , 𝑁2 }. By the triangle
inequality we obtain
𝜀 𝜀
|𝑎 − 𝑏| ≤ |𝑎 − 𝑎𝑛0 | + |𝑎𝑛0 − 𝑏| < + = 𝜀.
2 2
Since this is true for any 𝜀 > 0, we must have |𝑎 − 𝑏| = 0, i.e. 𝑎 = 𝑏.
□

Definition 2.5 We say that a sequence (𝑎𝑛 ) is divergent if it does not converge
to any 𝑎 ∈ ℝ.
△
In other words, a sequence (𝑎𝑛 ) is divergent if

∀ 𝑎 ∈ ℝ ∃𝜀 > 0 ∀ 𝑁 ∈ ℕ ∃𝑛 ∈ ℕ ∶ 𝑛 ≥ 𝑁 and |𝑎𝑛 − 𝑎| ≥ 𝜀.

23
 Chapter 2 Sequences and series

∞
𝑛2
Example 2.6 Consider the sequence � 𝑛+1 � .
𝑛=1
We want to show that this sequence is divergent. Notice that for any 𝑛 ∈ ℕ
1
we have 𝑛 ≥ 2 (𝑛 + 1). It follows that
1
𝑛2 𝑛 ⋅ 2 (𝑛 + 1) 1
≥ = 𝑛.
𝑛+1 𝑛+1 2
Now take an arbitrary 𝑎 ∈ ℝ and 𝜀 = 1. By the Archimedean Property (The-
orem 1.14) there exists 𝑁 ∈ ℕ with 𝑁 > 2(𝑎 + 𝜀). Then for all 𝑛 ≥ 𝑁 we
have
𝑛2 1 1
≥ 𝑛 ≥ 𝑁 > 𝑎 + 𝜀,
𝑛+1 2 2
so that
𝑛2
� − 𝑎� > 𝜀 for all 𝑛 ≥ 𝑁.
𝑛+1
𝑛2
It follows that lim𝑛→∞ 𝑛+1
≠ 𝑎. Since this is true for any 𝑎 ∈ ℝ, we conclude that
∞
𝑛2
the sequence � 𝑛+1 � diverges.
𝑛=1
△
Example 2.7 Another example of a divergent sequence is

((−1)𝑛 )∞
𝑛=0 = (1, −1, 1, −1, 1, −1, …).
1
Take 𝜀 = 2 . It is quite easy to see that, for any 𝑎 ∈ ℝ, at least one half of the
1 1
terms of the sequence lies outside the interval �𝑎 − 2 , 𝑎 + 2 �. (Make sure that
you understand why this is the case.) Thus, 𝑎 cannot be a limit of this sequence.
Since this is true for any 𝑎 ∈ ℝ, we see that the sequence diverges.
△
Definition 2.8 Let (𝑎𝑛 )∞
𝑛=1 be a sequence in ℝ.

(1) We say that (𝑎𝑛 )∞

𝑛=1 is bounded above if there exists a number 𝑀 ∈ ℝ such
that 𝑎𝑛 ≤ 𝑀 for all 𝑛 ∈ ℕ.

(2 ) We say that (𝑎𝑛 )∞

𝑛=1 is bounded below if there exists a number 𝑚 ∈ ℝ such
that 𝑎𝑛 ≥ 𝑚 for all 𝑛 ∈ ℕ.

(3 ) We say that (𝑎𝑛 )∞

𝑛=1 is bounded if it is bounded above and below, i.e. if there
exist numbers 𝑚, 𝑀 ∈ ℝ such that 𝑚 ≤ 𝑎𝑛 ≤ 𝑀 for all 𝑛 ∈ ℕ.

24
 Chapter 2 Sequences and series

Notice that a sequence (𝑎𝑛 )∞ 𝑛=1 is bounded (bounded above, bounded below)
if and only if its set of values {𝑎𝑛 ∶ 𝑛 ∈ ℕ} is bounded (bounded above, bounded
below).
△
Proposition 2.9 Any convergent sequence is bounded.
△

Proof. Consider a sequence (𝑎𝑛 )∞

𝑛=1 with lim𝑛→∞ 𝑎𝑛 = 𝑎. Take 𝜀 = 1. There exists
𝑁 ∈ ℕ such that |𝑎𝑛 − 𝑎| < 1 for all 𝑛 ≥ 𝑁. This is equivalent to saying that
𝑎 − 1 < 𝑎𝑛 < 𝑎 + 1 for all 𝑛 ≥ 𝑁.
Now put 𝑚 = min{𝑎1 , … , 𝑎𝑁−1 , 𝑎 − 1}, 𝑀 = max{𝑎1 , … , 𝑎𝑁−1 , 𝑎 + 1}. By above we
have
𝑚 ≤ 𝑎𝑛 ≤ 𝑀 for all 𝑛 ∈ ℕ,
and this means that (𝑎𝑛 )∞
𝑛=1 is bounded.
□
∞
𝑛2
Notice that the sequence � 𝑛+1 � in Example 2.6 is unbounded. Thus, it
𝑛=1
follows immediately from Proposition 2.9 that this sequence is divergent.
Remark The converse of Proposition 2.9 is not true: A bounded sequence
need not converge. An example of a sequence that is bounded and divergent is
((−1)𝑛 )∞
𝑛=0 .
△

2.2 Rules of working with limits

In the practice, only basic limits are calculated using the definition, and further
limits are determined from the basic limits using rules for working with limits.
We consider some of those rules in this section.
Theorem 2.10 (The Squeeze Theorem) Let (𝑎𝑛 ), (𝑏𝑛 ) be two sequences with
lim 𝑎𝑛 = lim 𝑏𝑛 = 𝐿.
𝑛→∞ 𝑛→∞
Suppose that (𝑥𝑛 ) is a sequence such that
𝑎𝑛 ≤ 𝑥 𝑛 ≤ 𝑏 𝑛 for all large values of 𝑛.
Then the sequence (𝑥𝑛 ) converges and lim𝑛→∞ 𝑥𝑛 = 𝐿.
△

25
 Chapter 2 Sequences and series

Proof. Suppose that 𝑎𝑛 ≤ 𝑥𝑛 ≤ 𝑏𝑛 for all 𝑛 ≥ 𝑁0 , where 𝑁0 ∈ ℕ. Given 𝜀 > 0, we

can find 𝑁1 , 𝑁2 ∈ ℕ such that |𝑎𝑛 − 𝐿| < 𝜀 for all 𝑛 ≥ 𝑁1 and |𝑏𝑛 − 𝐿| < 𝜀 for all
𝑛 ≥ 𝑁2 .
Put 𝑁 = max{𝑁0 , 𝑁1 , 𝑁2 }. For all 𝑛 ≥ 𝑁 we have

𝐿 − 𝜀 < 𝑎𝑛 ≤ 𝑥𝑛 ≤ 𝑏𝑛 ≤ 𝐿 + 𝜀,

which implies |𝑥𝑛 − 𝐿| < 𝜀 for all 𝑛 ≥ 𝑁. This proves that lim𝑛→∞ 𝑥𝑛 = 𝐿.
□
sin 𝑛 ∞
Example 2.11 Consider the sequence � � .
𝑛 𝑛=1
1 sin 𝑛 1
Since −1 ≤ sin 𝑛 ≤ 1, we have − 𝑛 ≤ 𝑛
≤ 𝑛
for all 𝑛 ∈ ℕ. We know that
1 1
lim𝑛→∞ 𝑛
= lim𝑛→∞ �− 𝑛 � = 0. The Squeeze Theorem (Theorem 2.10) yields
sin 𝑛
lim𝑛→∞ 𝑛
= 0.
△
Definition 2.12A sequence (𝑎𝑛 ) is called eventually constant if there are 𝑎 ∈ ℝ
and 𝑁 ∈ ℕ such that 𝑎𝑛 = 𝑎 for all 𝑛 ≥ 𝑁.
△
Proposition 2.13 If a sequence is eventually constant, then it is convergent.
△

Proof. Suppose that (𝑎𝑛 ) is eventually constant with 𝑎 and 𝑁 as in Definition 2.12.
Then for any 𝜀 > 0 we have

|𝑎𝑛 − 𝑎| = 0 < 𝜀 for all 𝑛 ≥ 𝑁.

Thus, lim𝑛→∞ 𝑎𝑛 = 𝑎.
□

Theorem 2.14 (The Algebraic Limit Theorem) Suppose that lim𝑛→∞ 𝑎𝑛 = 𝑎 and
lim𝑛→∞ 𝑏𝑛 = 𝑏. Then

(1) lim𝑛→∞ (𝑐𝑎𝑛 ) = 𝑐𝑎 for all 𝑐 ∈ ℝ,

(2 ) lim𝑛→∞ (𝑎𝑛 + 𝑏𝑛 ) = 𝑎 + 𝑏,

(3 ) lim𝑛→∞ (𝑎𝑛 𝑏𝑛 ) = 𝑎𝑏,

1 1
(4) lim𝑛→∞ 𝑏𝑛
= 𝑏
provided 𝑏 ≠ 0,

26
 Chapter 2 Sequences and series

𝑎𝑛 𝑎
(5) lim𝑛→∞ 𝑏𝑛
= 𝑏
provided 𝑏 ≠ 0.

Proof. ( 1 ) If 𝑐 = 0, then 𝑐𝑎𝑛 = 0 for all 𝑛 ∈ ℕ, and lim𝑛→∞ (𝑐𝑎𝑛 ) = 0 by Propo-

sition 2.13.
Therefore assume that 𝑐 ≠ 0. For any 𝜀 > 0 there is 𝑁 ∈ ℕ such that
𝜀
|𝑎𝑛 − 𝑎| < |𝑐| for all 𝑛 ≥ 𝑁. Then we obtain for all such 𝑛

𝜀
|𝑐𝑎𝑛 − 𝑐𝑎| = |𝑐||𝑎𝑛 − 𝑎| < |𝑐| = 𝜀,
|𝑐|
and the statement follows.
𝜀
(2 ) Given 𝜀 > 0, there are 𝑁1 , 𝑁2 ∈ ℕ such that |𝑎𝑛 − 𝑎| < 2
for all 𝑛 ≥ 𝑁1 and
𝜀
|𝑏𝑛 − 𝑏| < 2 for all 𝑛 ≥ 𝑁2 . Put 𝑁 = max{𝑁1 , 𝑁2 }. For all 𝑛 ≥ 𝑁 we have by
the triangle inequality
𝜀 𝜀
|(𝑎𝑛 + 𝑏𝑛 ) − (𝑎 + 𝑏)| = |(𝑎𝑛 − 𝑎) + (𝑏𝑛 − 𝑏)| ≤ |𝑎𝑛 − 𝑎| + |𝑏𝑛 − 𝑏| < + = 𝜀,
2 2
as desired.

(3 ) There is 𝑁1 ∈ ℕ such that |𝑏𝑛 − 𝑏| < 1 for all 𝑛 ≥ 𝑁1 . For such 𝑛 we have

−(|𝑏| + 1) ≤ 𝑏 − 1 < 𝑏𝑛 < 𝑏 + 1 ≤ |𝑏| + 1,

so that
|𝑏𝑛 | ≤ |𝑏| + 1, 𝑛 ≥ 𝑁1 .

Given 𝜀 > 0, there are 𝑁2 , 𝑁3 ∈ ℕ such that

𝜀
|𝑎𝑛 − 𝑎| < for all 𝑛 ≥ 𝑁2 ,
2(|𝑏| + 1)
𝜀
|𝑏𝑛 − 𝑏| < for all 𝑛 ≥ 𝑁3 .
2(|𝑎| + 1)
Put 𝑁 = max{𝑁1 , 𝑁2 , 𝑁3 }. For all 𝑛 ≥ 𝑁 we have

|𝑎𝑛 𝑏𝑛 − 𝑎𝑏| = |𝑎𝑛 𝑏𝑛 − 𝑎𝑏𝑛 + 𝑎𝑏𝑛 − 𝑎𝑏| ≤ |𝑎𝑛 − 𝑎||𝑏𝑛 | + |𝑎||𝑏𝑛 − 𝑏|

𝜀 𝜀 𝜀 𝜀
< (|𝑏| + 1) + |𝑎| < + = 𝜀,
2(|𝑏| + 1) 2(|𝑎| + 1) 2 2
and this proves the statement.

27
 Chapter 2 Sequences and series

|𝑏|
(4) There is 𝑁1 ∈ ℕ such that |𝑏𝑛 − 𝑏| < 2
for all 𝑛 ≥ 𝑁1 . For such 𝑛 we have

|𝑏|
|𝑏| = |𝑏 − 𝑏𝑛 + 𝑏𝑛 | ≤ |𝑏 − 𝑏𝑛 | + |𝑏𝑛 | < + |𝑏𝑛 |.
2
|𝑏|
This implies |𝑏| − 2
< |𝑏𝑛 |. Finally, we obtain

|𝑏|
|𝑏𝑛 | > , 𝑛 ≥ 𝑁1 .
2
In particular, 𝑏𝑛 ≠ 0 for 𝑛 ≥ 𝑁1 . Moreover, we have

1 1 𝑏 − 𝑏𝑛 |𝑏 − 𝑏𝑛 | 2|𝑏 − 𝑏𝑛 |
� − �=� �< 1
= , 𝑛 ≥ 𝑁1 .
𝑏𝑛 𝑏 𝑏𝑏𝑛 |𝑏| ⋅ 2 |𝑏| |𝑏|2

|𝑏|2
Given 𝜀 > 0, there is 𝑁2 ∈ ℕ such that |𝑏𝑛 − 𝑏| < 2
𝜀 for all 𝑛 ≥ 𝑁2 .
Put 𝑁 = max{𝑁1 , 𝑁2 }. For all 𝑛 ≥ 𝑁 we have

1 1 2|𝑏 − 𝑏𝑛 | 2|𝑏|2 𝜀
� − �< < = 𝜀,
𝑏𝑛 𝑏 |𝑏|2 2|𝑏|2

which yields the statement.

𝑎𝑛 1
(5) We write 𝑏𝑛
= 𝑎𝑛 𝑏 and apply (4) and (3).
𝑛
□

Remark The statements (1) and (2) of Theorem 2.14 imply the following: If
(𝑎𝑛 ), (𝑏𝑛 ) are convergent sequences and 𝛼, 𝛽 ∈ ℝ, then the sequence (𝛼𝑎𝑛 + 𝛽𝑏𝑛 )
is convergent and

lim (𝛼𝑎𝑛 + 𝛽𝑏𝑛 ) = 𝛼 lim 𝑎𝑛 + 𝛽 lim 𝑏𝑛 .

𝑛→∞ 𝑛→∞ 𝑛→∞

Thus, the operation of taking the limits of convergent sequences is a linear

operation.
△
In our next statement we show that the operation of taking the limits preserves
the order on ℝ.
Theorem 2.15 (The Order Limit Theorem) Suppose that lim𝑛→∞ 𝑎𝑛 = 𝑎 and
lim𝑛→∞ 𝑏𝑛 = 𝑏. If 𝑎𝑛 ≤ 𝑏𝑛 for all 𝑛 ∈ ℕ, then 𝑎 ≤ 𝑏.
△

28
 Chapter 2 Sequences and series

𝑎−𝑏
Proof. Suppose, by contradiction, that 𝑎 > 𝑏. Take 𝜀 = 2 > 0 and choose
𝑁 ∈ ℕ such that |𝑎𝑛 − 𝑎| < 𝜀 and |𝑏 − 𝑏𝑛 | < 𝜀 for all 𝑛 ∈ ℕ. But then

𝑎−𝑏 𝑎+𝑏 𝑎−𝑏

𝑏𝑛 < 𝑏 + 𝜀 = 𝑏 + = =𝑎− = 𝑎 − 𝜀 < 𝑎𝑛 , 𝑛 ≥ 𝑁.
2 2 2
This contradicts the assumption 𝑎𝑛 ≤ 𝑏𝑛 for all 𝑛 ∈ ℕ. Thus, we must have
𝑎 ≤ 𝑏.
□

Remark It is enough to assume in Theorem 2.15 that 𝑎𝑛 ≤ 𝑏𝑛 for all 𝑛 ≥ 𝑁0 ,

where 𝑁0 ∈ ℕ.
△
Remark Notice that 𝑎𝑛 < 𝑏𝑛 for all 𝑛 ∈ ℕ implies 𝑎 ≤ 𝑏, but does not imply
∞
1
𝑎 < 𝑏. As an example consider (𝑏𝑛 )∞
𝑛=1 = � 𝑛 � and the constant sequence
𝑛=1
1 1
(𝑎𝑛 )∞ ∞
𝑛=1 = (0)𝑛=1 . We have 𝑛
> 0 for all 𝑛 ∈ ℕ, but lim𝑛→∞ 𝑛
= 0 = lim𝑛→∞ 0.
△

2.3 The Monotone Convergence Theorem

We have seen at the end of Section 2.1 that a bounded sequence need not
converge. However, if a bounded sequence is also monotone, it does converge.
This statement is the content of this section.
Definition 2.16 We say that a sequence (𝑎𝑛 )∞
𝑛=1 is

(1) increasing, if 𝑎𝑛 ≤ 𝑎𝑛+1 for all 𝑛 ∈ ℕ,

(2 ) decreasing, if 𝑎𝑛 ≥ 𝑎𝑛+1 for all 𝑛 ∈ ℕ,

(3 ) monotone, if it is increasing or decreasing.

△
Notice that in some literature increasing sequences of Definition 2.16 (1)
are called non-decreasing, and decreasing sequences of Definition 2.16 (2) are
called non-increasing. If one wishes to emphasize that the inequalities ≤, ≥ are in
fact strict, one uses the terminology of strict monotonicity: A sequence (𝑎𝑛 )∞ 𝑛=1
is called strictly increasing if 𝑎𝑛 < 𝑎𝑛+1 for all 𝑛 ∈ ℕ, and strictly decreasing if
𝑎𝑛 > 𝑎𝑛+1 for all 𝑛 ∈ ℕ.

29
 Chapter 2 Sequences and series

𝑎+𝜀
𝑎
𝑎−𝜀

Figure 2.2: An increasing bounded sequence

Notice that a sequence (𝑎𝑛 )∞

𝑛=1 is increasing if and only if 𝑎𝑛 ≤ 𝑎𝑚 for all
𝑛, 𝑚 ∈ ℕ with 𝑛 ≤ 𝑚. It is decreasing if and only if 𝑎𝑛 ≥ 𝑎𝑚 for all 𝑛, 𝑚 ∈ ℕ with
𝑛 ≤ 𝑚.
Assume that a sequence (𝑎𝑛 )∞ 𝑛=1 is bounded above. Then its set of values
{𝑎𝑛 ∶ 𝑛 ∈ ℕ} is bounded above. By The Completeness Axiom (Axiom 1.8), it has
a supremum
sup 𝑎𝑛 = sup{𝑎𝑛 ∶ 𝑛 ∈ ℕ} ∈ ℝ.
𝑛∈ℕ
Similarly, if a sequence (𝑎𝑛 )∞
𝑛=1 is bounded below, it has an infimum

inf 𝑎𝑛 = inf{𝑎𝑛 ∶ 𝑛 ∈ ℕ} ∈ ℝ.
𝑛∈ℕ

Theorem 2.17 (The Monotone Convergence Theorem) If a sequence is increasing

and bounded above, then it converges, and its limit is its supremum.
△
For an illustration of this behavior, see Figure 2.2.

Proof of Theorem 2.17. Let (𝑎𝑛 )∞

𝑛=1 be an increasing bounded sequence, and put
𝑠 = sup 𝑎𝑛 . Let 𝜀 > 0. By Proposition 1.11, we can find 𝑁 ∈ ℕ such that
𝑛∈ℕ
𝑎𝑁 > 𝑠 − 𝜀. Since the sequence (𝑎𝑛 )∞𝑛=1 is monotone and 𝑠 is its supremum, we
have
𝑠 − 𝜀 < 𝑎𝑁 ≤ 𝑎𝑛 ≤ 𝑠 < 𝑠 + 𝜀 for all 𝑛 ≥ 𝑁,
which implies
|𝑠 − 𝑎𝑛 | < 𝜀 for all 𝑛 ≥ 𝑁.
It follows that lim𝑛→∞ 𝑎𝑛 = 𝑠.

30
 Chapter 2 Sequences and series

Corollary 2.18 If a sequence is decreasing and bounded below, it converges, and its
limit is its infimum.
△
Remark As usual in such statements regarding limits, it is sufficient to suppose
that a sequence is eventually monotone. A sequence (𝑎𝑛 )∞ 𝑛=1 is called eventually
increasing (eventually decreasing) if 𝑎𝑛 ≤ 𝑎𝑛+1 (𝑎𝑛 ≥ 𝑎𝑛+1 ) for all 𝑛 ≥ 𝑁0 with some
𝑁0 ∈ ℕ. A sequence that is eventually increasing and bounded above converges
to its supremum. A sequence that is eventually decreasing and bounded below
converges to its infimum.
△
Theorem 2.17 and Corollary 2.18 allow us to prove that a limit of a sequence
exists without having to find its value. This can be very useful.
Example 2.19 Consider the sequence (𝑎𝑛 )∞
𝑛=1 defined recursively by

1
𝑎0 = 0, 𝑎𝑛 = (𝑎2𝑛−1 + 1), 𝑛 ∈ ℕ.
4
We are going to show that this sequence is increasing and bounded above, and
then Theorem 2.17 says that it is convergent.
Using induction, we show that 𝑎𝑛 < 1 for all 𝑛 = 0, 1, 2, …. For the induction
base, notice that 𝑎0 = 0 < 1. Now suppose that 𝑎𝑘 < 1 for some 𝑘 ∈ ℕ ∪ {0}.
Then also 𝑎2𝑘 < 1 and

1 1 1
𝑎𝑘+1 = (𝑎2𝑘 + 1) < (1 + 1) = < 1.
4 4 2
By the Principle of Mathematical Induction we conclude that 𝑎𝑛 < 1 for all
𝑛 ∈ ℕ ∪ {0}.
To show that (𝑎𝑛 )∞
𝑛=1 is increasing, we will also use induction. For the induc-
1
tion base, notice that 𝑎0 = 0, 𝑎1 = 4 and this 𝑎0 < 𝑎1 . Now assume that 𝑎𝑘 < 𝑎𝑘+1
for some 𝑘 ∈ ℕ ∪ {0}. Then also 𝑎2𝑘 < 𝑎2𝑘+1 , and thus

1 1
𝑎𝑘+1 = (𝑎2𝑘 + 1) < (𝑎2𝑘+1 + 1) = 𝑎𝑘+2 .
4 4
By the Principle of Mathematical Induction, 𝑎𝑛 < 𝑎𝑛+1 for all 𝑛 ∈ ℕ ∪ {0}. Thus,
the sequence (𝑎𝑛 )∞
𝑛=1 is increasing (actually, we have shown that it is even strictly
increasing).

31
 Chapter 2 Sequences and series

By the Monotone Convergence Theorem (Theorem 2.17), the limit

lim 𝑎𝑛 = 𝑎
𝑛→∞

exists. Notice that we have proven the existence of the limit without knowing
its value.
In this particular example, we are even able to determine the value of the
limit — but only as we know that it exists. Consider the equation

1
𝑎𝑛 = (𝑎2𝑛−1 + 1), 𝑛 ∈ ℕ.
4
Applying the Algebraic Limit Theorem (Theorem 2.14), we can take limits at
both sides to obtain
1
𝑎 = (𝑎2 + 1).
4
Notice that this step is only possible since we know that the sequence (𝑎𝑛 )∞
𝑛=1
converges! The above equation for 𝑎 has two solutions 2 + √3 and 2 − √3. Of
course only one of these numbers is the value of lim𝑛→∞ 𝑎𝑛 — the limit of a
sequence, if exists, is unique (see Proposition 2.4). To decide which of the
numbers 2 + √3, 2 − √3 is the value of the limit, we notice that 𝑎𝑛 < 1 for
all 𝑛, and thus by the Order Limit Theorem (Theorem 2.15) 𝑎 = lim𝑛→∞ 𝑎𝑛 ≤ 1.
However, 2 + √3 > 1, and therefore cannot be the limit. We conclude that

lim 𝑎𝑛 = 2 − √3.
𝑛→∞

2.4 A first look at series

Definition 2.20 Let (𝑎𝑛 )∞
𝑛=1 be a sequence. A formal expression on the form
∞
� 𝑎𝑛 = 𝑎 1 + 𝑎 2 + 𝑎 3 + ⋯
𝑛=1
∞
is called an (infinite) series. The partial sums of the series ∑𝑛=1 𝑎𝑛 are the sums
𝑛
𝑠𝑛 = � 𝑎𝑘 = 𝑎1 + 𝑎2 + ⋯ + 𝑎𝑛 , 𝑛 ∈ ℕ.
𝑘=1

32
 Chapter 2 Sequences and series

∞
We say that the series ∑𝑛=1 𝑎𝑛 converges if the sequence of its partial sums
∞
�∑𝑛𝑘=1 𝑎𝑘 � ∞
converges. If ∑𝑛=1 𝑎𝑛 converges, we call the limit
𝑛=1
𝑛
𝑠 = lim 𝑠𝑛 = lim � 𝑎𝑘
𝑛→∞ 𝑛→∞
𝑘=1
∞
the sum of the series and write ∑𝑛=1 𝑎𝑛 = 𝑠.
𝑛 ∞ ∞
If the sequence �∑𝑘=1 𝑎𝑘 � diverges, we say that the series ∑𝑛=1 𝑎𝑛 diverges.
𝑛=1
△
∞
Note that writing the symbol ∑𝑛=1 𝑎𝑛 does not automatically mean that the
∞
series is supposed to be convergent. If the series ∑𝑛=1 𝑎𝑛 diverges, it can be still
interesting to discuss its terms, its partial sums, and study the divergence of
the sequence of partial sums.
Like with sequences, we can start the summation with any 𝑛0 ∈ ℤ and
∞
consider series ∑𝑛=𝑛 𝑎𝑛 . Frequently we take 𝑛0 = 0 like, for example, in the
0
∞ 1 1 1 1
geometric series ∑𝑛=0 2𝑛 = 1 + 2 + 4 + 8 + ⋯.
The next statement gives a simple but useful necessary condition for the
convergence of a series.
∞
Proposition 2.21 If a series ∑𝑛=1 𝑎𝑛 converges, then lim𝑛→∞ 𝑎𝑛 = 0.
△
𝑛 ∞
Proof. Let 𝑠𝑛 = ∑𝑘=1 𝑎𝑘 and ∑𝑘=1 𝑎𝑘 = 𝑠 = lim𝑛→∞ 𝑠𝑛 . Notice that

𝑛 𝑛−1
𝑎𝑛 = � 𝑎𝑘 − � 𝑎𝑘 = 𝑠𝑛 − 𝑠𝑛−1 .
𝑘=1 𝑘=1

Taking the limits as 𝑛 → ∞, we obatain

lim 𝑎𝑛 = lim 𝑠𝑛 − lim 𝑠𝑛−1 = 𝑠 − 𝑠 = 0.

𝑛→∞ 𝑛→∞ 𝑛→∞

Remark The converse of Proposition 2.21 is not true. We will see in Exam-
ple 2.29 an example of a divergent series with lim𝑛→∞ 𝑎𝑛 = 0.
△
Proposition 2.21 is frequently used in order to show that a series does not
converge.

33
 Chapter 2 Sequences and series

∞
Example 2.22 The series ∑𝑛=0 (−1)𝑛 diverges. Indeed, its terms 𝑎𝑛 = (−1)𝑛 do
not tend to zero as 𝑛 → ∞. The partial sums of this series are 𝑠0 = 1, 𝑠1 = 0, 𝑠2 =
1, 𝑠3 = 0, ….
△
A simple consequence of the definition of convergence is the following useful
statement about the tail of a convergent series.
∞
Proposition 2.23 If a series ∑𝑛=1 𝑎𝑛 converges, then for any 𝜀 > 0 there exists
𝑁 ∈ ℕ such that
∞
� � 𝑎𝑘 � < 𝜀 for all 𝑛 ≥ 𝑁.
𝑘=𝑛+1
△
∞ 𝑛 ∞
Proof. For a convergent series ∑𝑛=1 𝑎𝑛 , put 𝑠𝑛 = ∑𝑘=1 𝑎𝑘 and 𝑠 = ∑𝑘=1 𝑎𝑘 . Given
𝜀 > 0, by the definition of the limit lim𝑛→∞ 𝑠𝑛 = 𝑠, there is 𝑁 ∈ ℕ such that

|𝑠 − 𝑠𝑛 | < 𝜀 for any 𝑛 ≥ 𝑁.

∞
Noting that 𝑠 − 𝑠𝑛 = ∑𝑘=𝑛+1 𝑎𝑘 , we arrive at the desired statement.
□

Applying statements (1) and (2) of the Algebraic Limit Theorem (Theo-
rem 2.14) to the sequences of the partial sums, we obtain the following results.
∞
Proposition 2.24 ( 1 ) If a series ∑𝑛=1 𝑎𝑛 is convergent and 𝑐 ∈ ℝ, then the series
∑∞ (𝑐𝑎𝑛 ) is convergent, and
𝑛=1
∞ ∞
�(𝑐𝑎𝑛 ) = 𝑐 � 𝑎𝑛 .
𝑛=1 𝑛=1

∞ ∞ ∞
(2 ) If series ∑𝑛=1 𝑎𝑛 , ∑𝑛=1 𝑏𝑛 are convergent, then the series ∑𝑛=1 (𝑎𝑛 + 𝑏𝑛 ) is con-
vergent, and
∞ ∞ ∞
�(𝑎𝑛 + 𝑏𝑛 ) = � 𝑎𝑛 + � 𝑏𝑛 .
𝑛=1 𝑛=1 𝑛=1
△
We conclude that if ∑∞ 𝑎 𝑛 , ∑ ∞ 𝑏𝑛
are two convergent series and 𝛼, 𝛽 ∈ ℝ,
𝑛=1 𝑛=1
∞
then the series ∑𝑛=1 (𝛼𝑎𝑛 + 𝛽𝑏𝑛 ) is convergent, and
∞ ∞ ∞
�(𝛼𝑎𝑛 + 𝛽𝑏𝑛 ) = 𝛼 � 𝑎𝑛 + 𝛽 � 𝑏𝑛 .
𝑛=1 𝑛=1 𝑛=1

34
 Chapter 2 Sequences and series

∞ 1
Example 2.25 Consider the series ∑𝑛=1 .
𝑛(𝑛+1)
1 1 1
It holds 𝑛(𝑛+1)
= 𝑛
− 𝑛+1
. Thus, for the partial sums of this series we have
𝑚 𝑚
1 1 1
� = �� − �
𝑛=1 𝑛(𝑛 + 1) 𝑛=1 𝑛 𝑛+1
1 1 1 1 1 1 1
=1− + − +⋯+ − + −
2 2 3 𝑚−1 𝑚 𝑚 𝑚+1
1
=1− → 1, 𝑚 → ∞.
𝑚+1
Thus, the series converges and
∞
1
� = 1.
𝑛=1 𝑛(𝑛 + 1)

Series of this type are called telescopic series.

∞ 1 ∞ 1 ∞ 1
Notice that we could not write ∑𝑛=1 𝑛(𝑛+1) = ∑𝑛=1 𝑛
− ∑𝑛=1 𝑛+1
because the
∞ 1 ∞ 1
series ∑𝑛=1 𝑛
and ∑𝑛=1 are divergent (we will see this soon in Example 2.29).
𝑛+1
△
∞
Definition 2.26 A series of the form ∑𝑛=0 𝑎𝑟 , where 𝑎, 𝑟 ∈ ℝ, is called a
𝑛

geometric series.
△
∞
Examples of geometric series are ∑𝑛=0 (−1)𝑛 = 1 − 1 + 1 − 1 + ⋯, ∑∞ 2𝑛 =
𝑛=0
𝑛 𝑛
∞ 1 ∞ 1 1 1 1 1 1
1 + 2 + 4 + 8 + ⋯, ∑𝑛=1 � 2 � = ∑𝑛=0 2 � 2 � = 2 + 4 + 8 + 16 + ⋯.
∞
Theorem 2.27 (The Geometric Series) Let 𝑎 ≠ 0. A geometric series ∑𝑛=0 𝑎𝑟𝑛
converges if and only if |𝑟| < 1. In this case
∞
𝑎
� 𝑎𝑟𝑛 = .
𝑛=0 1−𝑟
△

Proof. If 𝑎 ≠ 0 and |𝑟| ≥ 1, then |𝑎𝑟𝑛 | ≥ |𝑎| ≠ 0, and thus the sequence (𝑎𝑟𝑛 )∞
𝑛=1 of
the terms of the series does not converge to zero. By Proposition 2.21, the series
∑∞ 𝑎𝑟𝑛 diverges.
𝑛=0
𝑛
Now assume that |𝑟| < 1 and put 𝑠𝑛 = ∑𝑘=0 𝑎𝑟𝑘 , 𝑛 = 0, 1, 2, …. We can write

𝑠𝑛 = 𝑎 + 𝑎𝑟 + 𝑎𝑟2 + ⋯ + 𝑎𝑟𝑛 ,
𝑟𝑠𝑛 = 𝑎𝑟 + 𝑎𝑟2 + ⋯ + 𝑎𝑟𝑛 + 𝑎𝑟𝑛+1 ,

35
 Chapter 2 Sequences and series

so that
(1 − 𝑟)𝑠𝑛 = 𝑎 − 𝑎𝑟𝑛+1 = 𝑎(1 − 𝑟𝑛+1 ),
and consequently
1 − 𝑟𝑛+1
𝑠𝑛 = 𝑎 .
1−𝑟
The property |𝑟| < 1 implies that lim𝑛→∞ 𝑟𝑛+1 = 0, and finally we obtain

1 − 𝑟𝑛+1 𝑎
lim 𝑠𝑛 = lim 𝑎 = .
𝑛→∞ 𝑛→∞ 1−𝑟 1−𝑟
∞ 𝑎
This means that the series ∑𝑛=0 𝑎𝑟𝑛 converges and its sum is 1−𝑟
.
□

Example 2.28 We have

∞ 𝑛∞ 𝑛
1 1 1 1 1 1
�� � = � � � = 1
= ⋅ 2 = 1.
𝑛=1 2 𝑛=0 2 2 21− 2
2

△
Another very important series will be considered in the next example.
∞ 1
Example 2.29 (The Harmonic Series) The series ∑𝑛=1
is called the harmonic
𝑛
series.
∞ 1
The harmonic series ∑𝑛=1 𝑛 diverges.
To prove this, we will show that the sequence of the partial sums (𝑠𝑛 )∞
𝑛=1 ,
𝑛 1
𝑠𝑛 = 𝑘=1 𝑘 , is unbounded. Then it follows by Proposition 2.9 that (𝑠𝑛 )∞
∑ 𝑛=1
diverges.
We consider the partial sums with 𝑛 = 2𝑚 , 𝑚 ∈ ℕ. We can estimate them as

36
 Chapter 2 Sequences and series

follows:
2𝑚
1
𝑠2𝑚 = �
𝑘=1
𝑘
1 1 1 1 1 1 1
=1+ +� + �+� + + + �+⋯
2 3 4 5 6 7 8
1 1 1
+ � 𝑚−1 + 𝑚−1 + ⋯ + 𝑚�
2 +1 2 +2 2
1 1 1 1 1 1 1 1 1 1
> 1 + + � + � + � + + + � + ⋯ + � 𝑚 + 𝑚 + ⋯ + 𝑚�
2 4 4 8 8 8 8 2 2 2
1 1 1 1
= 1 + + 2 ⋅ + 4 ⋅ + ⋯ + 2𝑚−1 ⋅ 𝑚
2 2 8 2
1 1 1 1 1
= 1 + + + + ⋯ + = 1 + 𝑚.
2 2 2 2 2
∞
1
The sequence �1 + 2 𝑚� is unbounded. Thus, also (𝑠2𝑚 )∞
𝑚=1 is unbounded,
𝑚=1
and consequently the sequence (𝑠𝑛 )∞
𝑛=1 is unbounded.
1
We observe that the terms 𝑛 of the harmonic series decrease and tens to zero
as 𝑛 → ∞. However, they decrease to zero not fast enough, and the partial
∞ 1
sums of ∑𝑛=1 𝑛 eventually exceed any real number.
△

2.5 The comparison test for series

In this section we consider a very useful test for convergence and divergence of
series with non-negative terms.
Theorem 2.30 (The Comparison Test for Series) Let (𝑎𝑛 )∞ ∞
𝑛=1 , (𝑏𝑛 )𝑛=1 be two se-
quences satisfying 0 ≤ 𝑎𝑛 ≤ 𝑏𝑛 for all 𝑛 ∈ ℕ.
∞ ∞
(1) If ∑𝑛=1 𝑏𝑛 converges, then ∑𝑛=1 𝑎𝑛 converges.
∞ ∞
(2 ) If ∑𝑛=1 𝑎𝑛 diverges, then ∑𝑛=1 𝑏𝑛 diverges.

△
𝑛 ∞
Proof. ( 1 ) Since 𝑎𝑘 ≥ 0, 𝑏𝑘 ≥ 0 for all 𝑘 ∈ ℕ, the sequences �∑𝑘=1 𝑎𝑘 � ,
𝑛=1
∞
�∑𝑛𝑘=1 𝑏𝑘 � are increasing.
𝑛=1

37
 Chapter 2 Sequences and series

∞
Suppose that the series ∑𝑛=1 𝑏𝑛 converges. This implies that the sequence
∞ ∞
�∑𝑛𝑘=1 𝑏𝑘 � converges. By Proposition 2.9, the sequence �∑𝑘=1 𝑏𝑘 �
𝑛
is
𝑛=1 𝑛=1
𝑛
bounded, i.e. there exists 𝑀 > 0 such that ∑𝑘=1 𝑏𝑘 ≤ 𝑀 for all 𝑛 ∈ ℕ. Now
𝑛 𝑛
the assumption 𝑎𝑘 ≤ 𝑏𝑘 for all 𝑘 ∈ ℕ yields ∑𝑘=1 𝑎𝑘 ≤ ∑𝑘=1 𝑏𝑘 ≤ 𝑀 for all
𝑛 ∈ ℕ.
𝑛 ∞
We have shown that the increasing sequence �∑𝑘=1 𝑎𝑘 � is bounded
𝑛=1
above by 𝑀. By the Monotone Convergence Theorem (Theorem 2.17),
𝑛 ∞ ∞
the sequence �∑𝑘=1 𝑎𝑘 � converges, which means that the series ∑𝑛=1 𝑎𝑛
𝑛=1
converges.
∞
(2 ) To prove the second statement, assume that the series ∑𝑛=1 𝑎𝑛 diverges. If
∑∞ 𝑏𝑛 would converge, then by (1) also ∑∞ 𝑎𝑛 would converge. Thus,
𝑛=1 𝑛=1
∑∞ 𝑏𝑛 must be divergent.
𝑛=1
□

Remark ( 1 ) It is sufficient to assume in Theorem 2.30 that the inequalities

0 ≤ 𝑎𝑛 ≤ 𝑏𝑛 hold for all 𝑛 ≥ 𝑁 with some 𝑁 ∈ ℕ.
∞
(2 ) Under the assumptions of the theorem, it is possible that the series ∑𝑛=1 𝑎𝑛
∞
converges and the series ∑𝑛=1 𝑏𝑛 diverges.

(3 ) The statements of the theorem only hold for series with non-negative
terms.
△
∞ 1
Example 2.31 Consider the series ∑𝑛=1 2 .
𝑛
For all 𝑛 ∈ ℕ we have
1 2 2
0≤ 2
= ≤ .
𝑛 𝑛(𝑛 + 𝑛) 𝑛(𝑛 + 1)
∞ 1
From Example 2.25 we know that the series ∑𝑛=1 𝑛(𝑛+1)
converges. By the first
∞ 2
statement in Proposition 2.24, also the series ∑𝑛=1 𝑛(𝑛+1) converges. Finally, by
∞ 1
the Comparison Test (Theorem 2.30), the series ∑𝑛=1 2 converges. Moreover,
𝑛
∞ ∞
1 1
0≤� 2
≤ 2 � = 2.
𝑛=1 𝑛 𝑛=1 𝑛(𝑛 + 1)

38
 Chapter 2 Sequences and series

It can be shown that

∞
1 𝜋2
� 2
= ,
𝑛=1 𝑛 6
we will not go into details.
△
2
Example 2.32 Consider the series ∑∞ � sin 𝑛 � .
𝑛=1 𝑛
The inequalities 0 ≤ sin2 𝑛 ≤ 1 imply

sin 𝑛 2 1
0≤� � ≤ 2, 𝑛 ∈ ℕ.
𝑛 𝑛
∞ 1 ∞ sin 𝑛 2
Since the series ∑𝑛=1 converges, it follows that also the series ∑𝑛=1 � �
𝑛2 𝑛
converges.
△
2
Example 2.33 Consider the series ∑∞ 𝑛3+3 .
𝑛=1 𝑛 +2
We have
𝑛2 + 3 𝑛2 1
3
≥ 3 3
= , 𝑛 ≥ 2.
𝑛 +2 𝑛 +𝑛 2𝑛
∞ 1
Remember that the harmonic series ∑𝑛=1 𝑛
diverges (Example 2.29). By Propo-
sition 2.24, also the series ∑∞ 1 diverges. With the help of the Comparison
𝑛=1 2𝑛
∞ 𝑛2 +3
Test, we conclude that the series ∑𝑛=1 diverges.
𝑛3 +2
△
1
Example 2.34 Consider the series ∑∞ .
𝑛=0 2+3𝑛
𝑛
1 1 1
For all 𝑛 = 0, 1, 2, … we have 0 ≤ 2+3𝑛
≤ 3𝑛
= � 3 � . Since the geometric
𝑛
∞ 1 ∞ 1
series ∑𝑛=0 � 3 � converges, by the Comparison Test also the series ∑𝑛=0 2+3𝑛
converges.
△
The Comparison Test in its essence allows us to figure out whether the terms
of a series converge to zero fast enough (and then the series converges) or too
slowly (and then the series diverges). The following result gives a convenient
“scale” which we can compare a given series against.
∞ 1
Theorem 2.35 (The 𝑝-series) The series ∑𝑛=1 (called the 𝑝-series) converges if
𝑛𝑝
and only if 𝑝 > 1.
△

39
 Chapter 2 Sequences and series

∞ 1
Proof. a. First suppose that 𝑝 ≤ 1. We have to show that the series ∑𝑛=1 𝑛𝑝
diverges. This follows easily by comparison with the harmonic series: for all
𝑛 ∈ ℕ we have
1 1
0 < ≤ 𝑝.
𝑛 𝑛
∞ 1 ∞ 1
Since the harmonic series ∑𝑛=1 𝑛
diverges, it follows that the series ∑𝑛=1 𝑛𝑝
with
𝑝 ≤ 1 diverges as well.

b. Now let us consider the case when 𝑝 > 1. The sequence of the partial sums
∞
1
�∑𝑛𝑘=1 � is increasing. We are going to show that it is bounded above.
𝑘𝑝
𝑛=1
First we take 𝑛 = 2𝑚 − 1 with some 𝑚 ∈ ℕ. We have
2𝑚 −1
1
𝑠2𝑚−1 = �
𝑘=1
𝑘𝑝
1 1 1 1 1 1
=1+� 𝑝 + 𝑝� + � 𝑝 + 𝑝 + 𝑝 + 𝑝�
2 3 4 5 6 7
1 1 1 1
+ � 𝑝 + ⋯ + 𝑝 � + ⋯ + � 𝑚−1 𝑝 + ⋯ + 𝑚 �
8 15 (2 ) (2 − 1)𝑝

1 1 1 1 1 1
≤1+� 𝑝 + 𝑝� + � 𝑝 + 𝑝 + 𝑝 + 𝑝�
2 2 4 4 4 4
1 1 1 1
+ � 𝑝 + ⋯ + 𝑝 � + ⋯ + � 𝑚−1 𝑝 + ⋯ + 𝑚−1 𝑝 �
8 8 (2 ) (2 )
1 1 1 1
= 1 + 2 ⋅ 𝑝 + 4 ⋅ 𝑝 + 8 ⋅ 𝑝 + ⋯ + 2𝑚−1 ⋅ 𝑚−1 𝑝
2 4 8 (2 )
1 2 3 𝑚−1
2 2 2 2
=1+� 𝑝 � + � 𝑝� + � 𝑝� + ⋯ + � 𝑝�
2 2 2 2
𝑚−1 𝑗
2
= � � 𝑝� .
𝑗=0
2
𝑗
2 ∞ 2
Notice that 2𝑝
< 1, and therefore the geometric series ∑𝑗=0 � 2𝑝 � converges to a
𝑗
∞ 2
finite sum 𝑀 = ∑𝑗=0 � 2𝑝 � ∈ ℝ. We obtain

2𝑚 −1 𝑚−1 ∞ 𝑗 𝑗
1 2 2
𝑠
2𝑚 −1 = � ≤ � � 𝑝� ≤ � � 𝑝 � = 𝑀.
𝑘=1
𝑘𝑝 𝑗=0
2 𝑗=0
2

40
 Chapter 2 Sequences and series

For an arbitrary 𝑛 ∈ ℕ we can find 𝑚 ∈ ℕ such that 𝑛 ≤ 2𝑚 − 1. Consequently,

𝑠𝑛 ≤ 𝑠2𝑚−1 ≤ 𝑀 for all 𝑛 ∈ ℕ.

This shows that the sequence (𝑠𝑛 )∞

𝑛=1 is bounded above. By the Monotone
∞ 1
Convergence Theorem (Theorem 2.17), it converges, i.e. the series ∑𝑛=1 𝑛𝑝 is
convergent.
□

2.6 Subsequences and the Bolzano-Weierstrass Theorem

In this section we come back to studying sequences rather then series.
Definition 2.36 Let (𝑎𝑛 )∞
𝑛=1 be a sequence, and let 𝑛1 < 𝑛2 < 𝑛3 < ⋯ be a
strictly increasing sequence of natural numbers. The sequence (𝑎𝑛𝑘 )∞
𝑘=1 is called
a subsequence of the sequence (𝑎𝑛 )∞ 𝑛=1 .
△
Thus, the terms of the subsequence are terms of the original sequence and
they appear in the same order, but some of the terms may be left out. Any
sequence is a subsequence of itself. Any sequence has infinitely many subse-
quences.
Example 2.37 Consider the sequence
∞
1 1 2 3 4 5
� (−1)𝑛−1 �1 − �� = �0, − , , − , , − , ⋯� .
𝑛 2 3 4 5 6
𝑛=1

As we mentioned above, this sequence has infinitely many subsequences. For

example, the sequences
∞
2 4 1
�0, , , ⋯� = �1 − �
3 5 2𝑘 − 1
𝑘=1

and ∞
1 3 5 1
�− , − , − , ⋯� = �− �1 − ��
2 4 6 2𝑘
𝑘=1
are two subsequences of it. Notice that the first of these subsequences converges
∞
1
to 1 while the second one converges to −1. The sequence �(−1)𝑛−1 �1 − 𝑛 ��
𝑛=1
itself diverges.
△

41
 Chapter 2 Sequences and series

In this section we will discuss convergence of subsequences. Our first result

is intuitively very clear.
Proposition 2.38 Any subsequence of a converging sequence converges to the same
limit.
△

Proof. Let (𝑎𝑛 )∞

𝑛=1 be a convergent sequence with lim𝑛→∞ 𝑎𝑛 = 𝑎. By the defini-
tion of the limit, for any 𝜀 > 0 we can find 𝑁 ∈ ℕ such that

|𝑎𝑛 − 𝑎| < 𝜀 for all 𝑛 ≥ 𝑁.

If (𝑛𝑘 )∞
𝑘=1 is a strictly increasing sequence of natural numbers, then 𝑛𝑘 ≥ 𝑘 for all
𝑘 ∈ ℕ. It follows that if 𝑘 ≥ 𝑁, then 𝑛𝑘 ≥ 𝑘 ≥ 𝑁 and thus

|𝑎𝑛𝑘 − 𝑎| < 𝜀 for all 𝑘 ≥ 𝑁.

By definition, lim𝑘→∞ 𝑎𝑛𝑘 = 𝑎.

□

The following corollaries are useful when we want to show that a given
sequence does not converge.
Corollary 2.39 ( 1 ) If a sequence has a divergent subsequence, then the sequence
itself diverges.

(2 ) If a sequence has two subsequences that converge to different limits, than the
sequence itself diverges.
△
It is easy to see that if a sequence is bounded then all its subsequences are
also bounded. We will now prove a surprising results that in this case there is
a least one subsequence that is convergent.
Theorem 2.40 (The Bolzano-Weierstrass Theorem) Every bounded sequence in ℝ
has a convergent subsequence.
△

Proof. Let (𝑎𝑛 )∞

𝑛=1 be a bounded sequence. We can find a closed and bounded
interval 𝐼0 such that 𝑎𝑛 ∈ 𝐼0 for all 𝑛 ∈ ℕ. Let the length of the interval 𝐼0 be ℓ;
we may assume that ℓ > 0.
Using the midpoint of the interval 𝐼0 , we bisect 𝐼0 into two closed and bounded
1
intervals of length 2 ℓ. At least one of these intervals contains infinitely many
terms of (𝑎𝑛 )∞
𝑛=1 . We choose this interval to be 𝐼1 .

42
 Chapter 2 Sequences and series

We continue in the same manner. In the 𝑘-th step we obtain a closed and
𝑘
1
bounded interval 𝐼𝑘 of length � 2 � ℓ that contains infinitely many terms of (𝑎𝑛 )∞
𝑛=1 .
In the (𝑘 + 1)-st step we use the midpoint of 𝐼𝑘 to bisect it into two closed and
𝑘+1
1
bounded intervals of length � 2 � ℓ. At least one of these intervals contains
infinitely many terms of (𝑎𝑛 )∞and we take this interval to be 𝐼𝑘+1 .
𝑛=1 ,
In this way we construct a sequence of closed bounded intervals (𝐼𝑘 )∞𝑘=1 such
that
𝐼0 ⊇ 𝐼 1 ⊇ 𝐼 2 ⊇ ⋯ ,
𝑘
1
the interval 𝐼𝑘 has length � 2 � ℓ, and each interval 𝐼𝑘 contains infinitely many
terms of (𝑎𝑛 )∞
𝑛=1 .

43
 Chapter 2 Sequences and series

∞
By the Nested Interval Property (Theorem 1.12), ⋂𝑘=1 𝐼𝑘 ≠ ∅. Take 𝑎 ∈
⋂ ∞ 𝐼𝑘 . 1
𝑘=1
We will now construct a subsequence (𝑎𝑛𝑘 )∞ 𝑘=1 that converges to 𝑎. We know
that 𝐼1 contains infinitely many terms of (𝑎𝑛 )∞
𝑛=1 . We choose 𝑛1 ∈ ℕ such that
𝑎𝑛1 ∈ 𝐼1 . Furthermore, 𝐼2 contains infinitely many terms of (𝑎𝑛 )∞𝑛=1 . Thus we
can choose 𝑛2 > 𝑛1 such that 𝑎𝑛2 ∈ 𝐼2 . We continue in the same manner. In the
𝑘-th step we choose 𝑛𝑘 > 𝑛𝑘−1 such that 𝑎𝑛𝑘 ∈ 𝐼𝑘 . In this way we construct a
subsequence (𝑎𝑛𝑘 )∞𝑘=1 with the property that 𝑎𝑛𝑘 ∈ 𝐼𝑘 , 𝑘 ∈ ℕ.
𝑁
1
Given 𝜀 > 0, there is 𝑁 ∈ ℕ such that � 2 � ℓ < 𝜀. Since 𝑎 ∈ 𝐼𝑘 and 𝑎𝑛𝑘 ∈ 𝐼𝑘 , we
have
𝑘 𝑁
1 1
|𝑎𝑛𝑘 − 𝑎| ≤ � � ℓ ≤ � � ℓ < 𝜀 for all 𝑘 ≥ 𝑁.
2 2
This proves that lim𝑘→∞ 𝑎𝑛𝑘 = 𝑎.
□

We finish this section by a useful property whose proof is left as an exercise.

Proposition 2.41 Let (𝑎𝑛 )∞
𝑛=1 be a sequence. If lim𝑘→∞ 𝑎2𝑘 = lim𝑘→∞ 𝑎2𝑘−1 = 𝐿,
then lim𝑛→∞ 𝑎𝑛 = 𝐿.
△

Proof. Exercise.
□

2.7 Cauchy sequences

Definition 2.42 We say that a sequence (𝑎𝑛 )∞
𝑛=1 is a Cauchy sequence if

∀ 𝜀 > 0 ∃𝑁 ∈ ℕ ∀ 𝑛, 𝑚 ≥ 𝑁 ∶ |𝑎𝑛 − 𝑎𝑚 | < 𝜀,

i.e. for any positive number 𝜀 there exists 𝑁 ∈ ℕ such that |𝑎𝑛 − 𝑎𝑚 | < 𝜀 for all
𝑛, 𝑚 ∈ ℕ with 𝑛, 𝑚 ≥ 𝑁.
△

∞ ∞
We will not use this in the proof, but in fact ⋂𝑘=1 𝐼𝑘 = {𝑎}. Indeed, suppose that 𝑏 ∈ ⋂𝑘=1 𝐼𝑘 .
1

1 𝑘 1 𝑘
Then 𝑎, 𝑏 ∈ 𝐼𝑘 for any 𝑘 ∈ ℕ, and since the length of 𝐼𝑘 is � 2 � ℓ, we have that |𝑎 − 𝑏| ≤ � 2 � ℓ for
1 𝑘
all 𝑘 ∈ ℕ. Since � 2 � ℓ → 0 as 𝑘 → ∞, it follows that |𝑎 − 𝑏| = 0, i.e. 𝑎 = 𝑏.

44
 Chapter 2 Sequences and series

The definition of a Cauchy sequence must remind you of the definition of a

convergent sequence (Definition 2.2). Indeed, we will see in this section that
convergent sequences and Cauchy sequences are closely related.
Proposition 2.43 Every Cauchy sequence is bounded.
△

Proof. Let (𝑎𝑛 )∞

𝑛=1 be a Cauchy sequence. Using the definition and taking 𝜀 = 1,
we see that there exists 𝑁 ∈ ℕ such that |𝑎𝑛 −𝑎𝑘 | < 1 for all 𝑛, 𝑘 ≥ 𝑁. In particular,

|𝑎𝑛 − 𝑎𝑁 | < 1 for all 𝑛 ≥ 𝑁.

Put 𝑚 = min{𝑎1 , … , 𝑎𝑁−1 , 𝑎𝑁 − 1}, 𝑀 = max{𝑎1 , … , 𝑎𝑁−1 , 𝑎𝑁 + 1}. By above we

have
𝑚 ≤ 𝑎𝑛 ≤ 𝑀 for all 𝑛 ∈ ℕ.
□

Proposition 2.43 and its proof should remind you of the statement and the
proof of Proposition 2.9. Modifying the proof of Theorem 2.14 in a similar
way, one can show that scalar multiples, sums, products and (under sensible
assumptions) quotients of Cauchy sequences are Cauchy sequences. Try to
establish these properties as an exercise!
Let us now investigate a relation between the properties of a sequence to be
Cauchy and to converge.
Lemma 2.44 Every convergent sequence is a Cauchy sequence.
△

Proof. Let (𝑎𝑛 )∞

𝑛=1 be a convergent sequence with lim𝑛→∞ 𝑎𝑛 = 𝑎.
Given 𝜀 > 0, there is 𝑁 ∈ ℕ such that
𝜀
|𝑎𝑛 − 𝑎| < for all 𝑛 ≥ 𝑁.
2
But then for all 𝑛, 𝑚 ≥ 𝑁 we have
𝜀 𝜀
|𝑎𝑛 − 𝑎𝑚 | = |(𝑎𝑛 − 𝑎) − (𝑎𝑚 − 𝑎)| ≤ |𝑎𝑛 − 𝑎| + |𝑎𝑚 − 𝑎| < + = 𝜀,
2 2
which means that (𝑎𝑛 )∞
𝑛=1 is a Cauchy sequence.
□

45
 Chapter 2 Sequences and series

In the theorem below we will show that the converse statement is true in ℝ,
i.e. every Cauchy sequence in ℝ converges in ℝ. Thus, a sequence is convergent
in ℝ if and only if it is a Cauchy sequence. However, it should be noticed that
this property does not hold in other spaces. While Lemma 2.44 is always true
(for example, also for sequences in ℚ), the fact that every Cauchy sequence
has a limit is a property of ℝ. This statement does not hold, for example, in
ℚ: One can construct a Cauchy sequence in ℚ that does not converge in ℚ —
take, say, a sequence of rational numbers that converge to √2. To some extent,
the statement that every Cauchy sequence converges in ℝ is equivalent to the
Completeness Axiom. 2
Theorem 2.45 (The Cauchy Criterion) In ℝ, a sequence converges if and only if it
is a Cauchy sequence.
△

Proof. We have proven in Lemma 2.44 that every convergent sequence is a

Cauchy sequence. Thus, we need to prove the converse statement.
Let (𝑎𝑛 )∞
𝑛=1 be a Cauchy sequence. By Proposition 2.43, it is bounded. There-
fore, by the Bolzano-Weierstrass Theorem (Theorem 2.40), (𝑎𝑛 )∞ 𝑛=1 has a conver-
∞
gent subsequence (𝑎𝑛𝑘 )𝑘=1 . Put 𝑎 = lim𝑘→∞ 𝑎𝑛𝑘 .
We will show that 𝑎 = lim𝑛→∞ 𝑎𝑛 . Take an arbitrary 𝜀 > 0. There is 𝐾 ∈ ℕ
such that
𝜀
|𝑎𝑛𝑘 − 𝑎| < for all 𝑘 ≥ 𝐾.
2
Since (𝑎𝑛 )∞
𝑛=1 is a Cauchy sequence, there is 𝑁 ∈ ℕ such that

𝜀
|𝑎𝑛 − 𝑎𝑚 | < for all 𝑛, 𝑚 ≥ 𝑁.
2
Now choose 𝑘0 ∈ ℕ such that 𝑛𝑘0 ≥ 𝑁. Then for all 𝑚 ≥ 𝑁 we have
𝜀 𝜀
|𝑎𝑚 − 𝑎| ≤ |𝑎𝑚 − 𝑎𝑛𝑘 | + |𝑎𝑛𝑘 − 𝑎| < + = 𝜀,
0 0 2 2
and this proves that lim𝑛→∞ 𝑎𝑛 = 𝑎.
□

Theorem 2.45 has an implication for series.

2
You can find a discussion on the exact relationship between these properties in Section 2.6 of
Abbott’s textbook “Understanding Analysis”.

46
 Chapter 2 Sequences and series

∞
Corollary 2.46 (The Cauchy Criterion for Series) In ℝ, a series ∑𝑛=1 𝑎𝑛 is conver-
gent if and only if for any 𝜀 > 0 there exists 𝑁 ∈ ℕ such that

|𝑎𝑛+1 + 𝑎𝑛+2 + ⋯ + 𝑎𝑚 | < 𝜀 for any 𝑚 > 𝑛 ≥ 𝑁.

△
∞
Proof. By above, the series ∑𝑛=1 𝑎𝑛 is convergent if and only if the sequence of
𝑛 ∞
the partial sums �∑𝑘=1 𝑎𝑘 � is a Cauchy sequence. If 𝑚 > 𝑛, then
𝑛=1
𝑚 𝑛
� 𝑎𝑘 − � 𝑎𝑘 = 𝑎𝑛+1 + 𝑎𝑛+2 + ⋯ + 𝑎𝑚 .
𝑘=1 𝑘=1

𝑛 ∞
Thus, �∑𝑘=1 𝑎𝑘 � is a Cauchy sequence if and only if for any 𝜀 > 0 there exists
𝑛=1
𝑁 ∈ ℕ such that
𝑚 𝑛
�� 𝑎𝑘 − � 𝑎𝑘 � = |𝑎𝑛+1 + 𝑎𝑛+2 + ⋯ + 𝑎𝑚 | < 𝜀 for any 𝑚 > 𝑛 ≥ 𝑁.
𝑘=1 𝑘=1

2.8 Absolute and conditional convergence of series

We have seen that the Comparison Test (Theorem 2.30) only can be applied to
series with non-negative terms. In the general situation, sometimes it may help
∞ ∞
to consider the series ∑𝑛=1 |𝑎𝑛 | along with ∑𝑛=1 𝑎𝑛 .
∞ ∞
Proposition 2.47 If the series ∑𝑛=1 |𝑎𝑛 | converges, then the series ∑𝑛=1 𝑎𝑛 converges
as well.
△

Proof. We are going to use the Cauchy Criterion (Corollary 2.46). If the series
∑∞ |𝑎𝑛 | converges, then for any 𝜀 > 0 there exists 𝑁 ∈ ℕ such that
𝑛=1

|𝑎𝑛+1 | + |𝑎𝑛+2 | + ⋯ + |𝑎𝑚 | < 𝜀 for all 𝑚 > 𝑛 ≥ 𝑁.

But the triangle inequality gives

|𝑎𝑛+1 + 𝑎𝑛+2 + ⋯ + 𝑎𝑚 | ≤ |𝑎𝑛+1 | + |𝑎𝑛+2 | + ⋯ + |𝑎𝑚 | < 𝜀 for all 𝑚 > 𝑛 ≥ 𝑁,

∞ ∞
and the Cauchy Criterion applied to series ∑𝑛=1 𝑎𝑛 yields that ∑𝑛=1 𝑎𝑛 converges.
□

47
 Chapter 2 Sequences and series

The converse of Proposition 2.47 does not hold, we will see examples below.
This motivates the following definition.
∞
Definition 2.48 ( 1 ) We say that a series ∑𝑛=1 𝑎𝑛 converges absolutely if the
∞
series ∑𝑛=1 |𝑎𝑛 | converges.
∞ ∞ ∞
(2 ) If ∑𝑛=1 𝑎𝑛 converges but ∑𝑛=1 |𝑎𝑛 | diverges, we say that the series ∑𝑛=1 𝑎𝑛
converges conditionally.
△
∞ (−1)𝑛−1
An example of an absolutely convergent series is ∑ 2 . Indeed, we
𝑛=1 𝑛
(−1) 𝑛−1 1
∞ ∞
have seen in Example 2.31 that ∑𝑛=1 � 2 � = ∑𝑛=1 converges.
𝑛 𝑛2
∞ (−1)𝑛−1
An example of a conditionally convergent series is ∑𝑛=1 𝑛
. We already
∞ (−1)𝑛−1 ∞ 1
know that the harmonic series ∑𝑛=1 � 𝑛
� = ∑𝑛=1
𝑛
diverges, so that the series
(−1)𝑛−1
∑∞ cannot be absolutely convergent. Its convergence follows from The-
𝑛=1 𝑛
𝑛−1
∞ (−1)
orem 2.49 below. The series ∑𝑛=1 𝑛 is called the alternating harmonic series.
We consider, more generally, series of the form
∞
�(−1)𝑛 𝑎𝑛 , where 𝑎𝑛 ≥ 0, 𝑛 ∈ ℕ.
𝑛=0

Such series are called alternating series. They have the form
∞
�(−1)𝑛 𝑎𝑛 = 𝑎0 − 𝑎1 + 𝑎2 − 𝑎3 + 𝑎4 − 𝑎5 + ⋯ .
𝑛=0

The name comes from the fact that the signs of the terms alternate.
Theorem 2.49 (The Alternating Series Test, or the Leibniz Test) Let (𝑎𝑛 )∞
𝑛=0 be a
sequence such that 𝑎0 ≥ 𝑎1 ≥ 𝑎2 ≥ ⋯ ≥ 𝑎𝑛 ≥ 𝑎𝑛+1 ≥ ⋯ and lim𝑛→∞ 𝑎𝑛 = 0. Then
∞
(1) the alternating series ∑𝑛=0 (−1)𝑛 𝑎𝑛 converges,
𝑚 ∞
(2 ) the partial sums 𝑠𝑚 = ∑𝑛=0 (−1)𝑛 𝑎𝑛 approximate the sum 𝑠 = ∑𝑛=0 (−1)𝑛 𝑎𝑛
within the error
|𝑠 − 𝑠𝑚 | ≤ 𝑎𝑚+1 .

△
Notice that the assumptions 𝑎0 ≥ 𝑎1 ≥ 𝑎2 ≥ ⋯ ≥ 𝑎𝑛 ≥ 𝑎𝑛+1 ≥ ⋯ and
lim𝑛→∞ 𝑎𝑛 = 0 imply that 𝑎𝑛 ≥ 0, 𝑛 ∈ ℕ.

48
 Chapter 2 Sequences and series

Proof. We will consider the subsequences (𝑠2𝑘 )∞ ∞

𝑘=0 and (𝑠2𝑘+1 )𝑘=0 of the sequence
∞
of partial sums (𝑠𝑚 )𝑚=0 . We have
2𝑘+2 2𝑘
𝑠2𝑘+2 − 𝑠2𝑘 = � (−1)𝑛 𝑎𝑛 − �(−1)𝑛 𝑎𝑛
𝑛=0 𝑛=0
= (−1)2𝑘+2 𝑎 2𝑘+2 + (−1)2𝑘+1 𝑎2𝑘+1 = 𝑎2𝑘+2 − 𝑎2𝑘+1 ≤ 0.

Thus, the sequence (𝑠2𝑘 )∞

𝑘=0 is decreasing. Similarly,

2𝑘+1 2𝑘−1
𝑠2𝑘+1 − 𝑠2𝑘−1 = � (−1)𝑛 𝑎𝑛 − � (−1)𝑛 𝑎𝑛
𝑛=0 𝑛=0
= (−1)2𝑘+1 𝑎 2𝑘+1 + (−1)2𝑘 𝑎2𝑘 = −𝑎2𝑘+1 + 𝑎2𝑘 ≥ 0,

and the sequence (𝑠2𝑘+1 )∞

𝑘=0 is increasing. Furthermore,

2𝑘+1 2𝑘
𝑠2𝑘+1 − 𝑠2𝑘 = � (−1)𝑛 𝑎𝑛 − �(−1)𝑛 𝑎𝑛 = (−1)2𝑘+1 𝑎2𝑘+1 ≤ 0,
𝑛=0 𝑛=0

so that 𝑠2𝑘+1 ≤ 𝑠2𝑘 for all 𝑘. Taking into consideration that (𝑠2𝑘 )∞
𝑘=0 is decreasing
∞
and (𝑠2𝑘+1 )𝑘=0 is increasing, we obtain that

𝑠1 ≤ 𝑠2𝑘+1 ≤ 𝑠2𝑘 ≤ 𝑠0 for all 𝑘 = 0, 1, 2, … .

We have shown that the sequence (𝑠2𝑘 )∞ 𝑘=0 is decreasing and bounded below
by 𝑠1 . At the same time, the sequence (𝑠2𝑘+1 )∞
𝑘=0 is increasing and bounded above
by 𝑠0 . By the Monotone Convergence Theorem (Theorem 2.17), both sequences
are convergent.
Put 𝑠 = lim𝑘→∞ 𝑠2𝑘 . As 𝑠2𝑘+1 = 𝑠2𝑘 + (−1)2𝑘+1 𝑎2𝑘+1 = 𝑠2𝑘 − 𝑎2𝑘+1 , we have

lim 𝑠2𝑘+1 = lim 𝑠2𝑘 − lim 𝑎2𝑘+1 = 𝑠 − 0 = 𝑠.

𝑘→∞ 𝑘→∞ 𝑘→∞

Thus, both sequences (𝑠2𝑘 )∞ ∞

𝑘=0 and (𝑠2𝑘+1 )𝑘=0 converge to the same limit

𝑠 = sup 𝑠2𝑘+1 = inf 𝑠2𝑘 .

𝑘=0,1,… 𝑘=0,1,…

It follows by Proposition 2.41 that lim𝑚→∞ 𝑠𝑚 = 𝑠. In other words, the series

∑∞ (−1)𝑛 𝑎𝑛 converges.
𝑛=0
The above consideration implies that

𝑠1 ≤ 𝑠3 ≤ ⋯ ≤ 𝑠2𝑘+1 ≤ ⋯ ≤ 𝑠 ≤ ⋯ ≤ 𝑠2𝑘 ≤ ⋯ ≤ 𝑠2 ≤ 𝑠0 .

49
 Chapter 2 Sequences and series

From here we see that

0 ≤ 𝑠 − 𝑠2𝑘+1 ≤ 𝑠2𝑘+2 − 𝑠2𝑘+1 = 𝑎2𝑘+2

and
0 ≤ 𝑠2𝑘 − 𝑠 ≤ 𝑠2𝑘 − 𝑠2𝑘+1 = 𝑎2𝑘+1 .
Together this gives
|𝑠 − 𝑠𝑚 | ≤ 𝑎𝑚+1 .
□

Example 2.50 It follows by the Alternating Convergence Test (Theorem 2.49)

∞ (−1)𝑛−1
that the alternating harmonic series ∑𝑛=1 𝑛
converges. We have mentioned
above that it converges conditionally.
∞ (−1)𝑛−1
With the help of a Taylor series, one can show that ∑𝑛=1 𝑛 = ln 2, we do
not go into details.
Using the second statement in Theorem 2.49, one can estimate how many
terms of the alternating series one should take in order to calculate its sum
within a given accuracy. Imagine, for example, that we want to calculate the
value of ln 2 using the alternating harmonic series within the accuracy of 0.001.
We should choose 𝑚 such that 3
1 1 1 (−1)𝑚−1 1
�ln 2 − �1 − + − +⋯+ �� ≤ < 0.001.
2 3 4 𝑚 𝑚+1

We should take 𝑚 = 1000.

In practice, other series are used to calculate ln 2, because — as we just have
seen — the alternating harmonic series converges very slowly.
△
Remark In the Alternating Series Test (Theorem 2.49) we assume that the
sequence (𝑎𝑛 ) is decreasing and that lim𝑛→∞ 𝑎𝑛 = 0. Both assumptions are
necessary for the theorem to hold. We already know (Proposition 2.21) that
the assumption lim𝑛→∞ 𝑎𝑛 = 0 is necessary for any series to converge. The
example below shows that the assumption of the monotonicity of the sequence
(𝑎𝑛 ) cannot be removed.
△
3
The rule that is easy to remember is that for estimating the difference between the sum of a
series and its partial sum, we take the first term of the series that was not used in the partial
sum.

50
 Chapter 2 Sequences and series

∞
Example 2.51 Consider the alternating series ∑𝑛=1 (−1)𝑛−1 𝑎𝑛 , where
⎧4
⎪
⎪
⎨ 𝑛2 , if 𝑛 is even,
𝑎𝑛 = ⎪
⎪
⎩ 2 , if 𝑛 is odd.
𝑛+1

Notice that lim𝑛→∞ 𝑎𝑛 = 0.

We consider the partial sums 𝑠2𝑚 , 𝑚 ∈ ℕ. For an odd 𝑛 we write 𝑛 = 2𝑘 − 1
and for an even 𝑛 we write 𝑛 = 2𝑘 with 𝑘 ∈ ℕ. We have
2𝑚 𝑚 𝑚 𝑚
2 4 1 𝑚 1
𝑠2𝑚 = �(−1)𝑛−1 𝑎𝑛 = � −� = � −� .
𝑛=1 𝑘=1
(2𝑘 − 1) + 1 𝑘=1 (2𝑘)2 𝑘=1 𝑘 𝑘=1 𝑘2
∞
𝑚 1
We know that the sequence �∑𝑘=1 � converges and the sequence
𝑘2
𝑚=1
∞
1
�∑𝑚
𝑘=1
� diverges. If follows that the sequence (𝑠2𝑚 )∞
𝑚=1 diverges. This im-
𝑘
𝑚=1
∞
plies that the series ∑𝑛=1 (−1)𝑛−1 𝑎𝑛 diverges.
△
There are several further convergence tests for conditionally convergent series
such as Abel’s Test and Dirichtlet’s Test. These tests are generalizations of the
Alternating Series Test, but they can be applied also to series that are not
alternating. We will not discuss these tests in our course, you can find them in
Abbott’s textbook “Understanding Analysis”.

2.9 The ratio and the root tests

In this section we discuss two tests for absolute convergence that are frequently
used in the practice. Both these tests are based on the comparison with a
geometric series.
Theorem 2.52 (The Ratio Test) Let (𝑎𝑛 )∞
𝑛=1 be a sequence of real numbers such that
the limit
|𝑎𝑛+1 |
𝐿 = lim
𝑛→∞ |𝑎𝑛 |

exists. Then the following holds.

∞
(1) If 𝐿 < 1, then the series ∑𝑛=1 𝑎𝑛 converges absolutely.
∞
(2 ) If 𝐿 > 1, then the series ∑𝑛=1 𝑎𝑛 diverges.

51
 Chapter 2 Sequences and series

(3 ) If 𝐿 = 1, then the test is inconclusive.

Proof. ( 1 ) Suppose that 𝐿 < 1. Take 𝑟 ∈ ℝ such that 𝐿 < 𝑟 < 1. By the
definition of the limit with 𝜀 = 𝑟 − 𝐿 > 0, there exists 𝑁 ∈ ℕ such that
|𝑎𝑛+1 |
<𝑟 for all 𝑛 ≥ 𝑁.
|𝑎𝑛 |

This implies |𝑎𝑁+1 | < |𝑎𝑁 |𝑟, |𝑎𝑁+2 | < |𝑎𝑁+1 |𝑟 < |𝑎𝑁 |𝑟2 , and so on. In general,

|𝑎𝑁+𝑗 | < |𝑎𝑁 |𝑟𝑗 , 𝑗 ∈ ℕ;

∞
this can be shown by induction. Since ∑𝑗=0 |𝑎𝑁 |𝑟𝑗 is a converging geometric
series, it follows by the Comparison Test (Theorem 2.30) that the series
∑∞ |𝑎𝑛 | converges. Hence, the series ∑∞ 𝑎𝑛 converges absolutely.
𝑛=𝑁 𝑛=1

(2 ) Now assume that 𝐿 > 1. Using again the definition of the limit with
𝜀 = 𝐿 − 1 > 0, we conclude that there exists 𝑁 ∈ ℕ such that
|𝑎𝑛+1 |
>1 for all 𝑛 ≥ 𝑁.
|𝑎𝑛 |

But then |𝑎𝑛+1 | > |𝑎𝑛 | > 0 for all 𝑛 ≥ 𝑁, so that (𝑎𝑛 ) does not converge to 0.
∞
Thus, the series ∑𝑛=1 𝑎𝑛 diverges.
∞ 1
(3 ) If 𝐿 = 1, then the test is inconclusive. For the harmonic series ∑𝑛=1 𝑛
we
have
1/(𝑛 + 1) 𝑛
𝐿 = lim = lim =1
𝑛→∞ 1/𝑛 𝑛→∞ 𝑛 + 1

∞ (−1)𝑛−1
and the series is divergent. For the alternating harmonic series ∑𝑛=1 𝑛
we also have 𝐿 = 1 and the series is conditionally convergent. Finally, for
∞ 1
the series ∑𝑛=1 2 we have
𝑛

1/(𝑛 + 1)2 𝑛2
𝐿 = lim = lim =1
𝑛→∞ 1/𝑛2 𝑛→∞ (𝑛 + 1)2

and the series is absolutely convergent.

□

52
 Chapter 2 Sequences and series

∞ 2𝑛
Example 2.53 Consider the series ∑𝑛=1 𝑛! .
Clearly, all terms of the series are positive, so we do not need to care for the
absolute values. We have
𝑎𝑛+1 2𝑛+1 𝑛! 2
= 𝑛
=
𝑎𝑛 (𝑛 + 1)!2 𝑛+1
and
𝑎𝑛+1 2
𝐿 = lim = lim = 0 < 1.
𝑛→∞ 𝑎𝑛 𝑛→∞ 𝑛 + 1

∞ 2𝑛
It follows that the series ∑𝑛=1 𝑛!
converges.
△
Example 2.54 Consider the series ∑∞ 𝑛(𝑛+1) .
𝑛=1 1+2𝑛
Again, all terms are positive. We have
(𝑛 + 1)(𝑛 + 2) 1 + 2𝑛 𝑛+2 1 + 2𝑛
𝐿 = lim ⋅ = lim ⋅ lim
𝑛→∞ 1 + 2𝑛+1 𝑛(𝑛 + 1) 𝑛→∞ 𝑛 𝑛→∞ 1 + 2𝑛+1
1
𝑛 +1 1
=1⋅ lim 2 = < 1,
𝑛→∞ 1 2
2𝑛
+2

∞ 𝑛(𝑛+1)
and thus the series ∑𝑛=1 1+2𝑛
converges.
△
Theorem 2.55 (The Root Test) Let (𝑎𝑛 )∞
𝑛=1 be a sequence of real numbers such that
the limit
1
𝐿 = lim |𝑎𝑛 | 𝑛
𝑛→∞
exists. Then the following holds.
∞
(1) If 𝐿 < 1, then the series ∑𝑛=1 𝑎𝑛 converges absolutely.
∞
(2 ) If 𝐿 > 1, then the series ∑𝑛=1 𝑎𝑛 diverges.

(3 ) If 𝐿 = 1, then the test is inconclusive.

Proof. ( 1 ) Suppose that 𝐿 < 1. Take 𝑟 ∈ ℝ such that 𝐿 < 𝑟 < 1. By the
definition of the limit with 𝜀 = 𝑟 − 𝐿 > 0, there exists 𝑁 ∈ ℕ such that
1
|𝑎𝑛 | 𝑛 < 𝑟 for all 𝑛 ≥ 𝑁.

53
 Chapter 2 Sequences and series

This implies that

|𝑎𝑛 | < 𝑟𝑛 for all 𝑛 ≥ 𝑁.
∞
Since the geometric series ∑𝑛=1 𝑟𝑛 converges, by the Comparison Test
∞ ∞
(Theorem 2.30), the series ∑𝑛=1 |𝑎𝑛 | converges, i.e. the series ∑𝑛=1 𝑎𝑛
converges absolutely.

(2 ) Now assume that 𝐿 > 1. Using again the definition of the limit with
𝜀 = 𝐿 − 1 > 0, we conclude that there exists 𝑁 ∈ ℕ such that
1
|𝑎𝑛 | 𝑛 > 1 for all 𝑛 ≥ 𝑁.

But then |𝑎𝑛 | > 1 for all 𝑛 ≥ 𝑁. In particular, (𝑎𝑛 ) does not converge to 0.
∞
Thus, the series ∑𝑛=1 𝑎𝑛 diverges.

(3 ) The same three examples as in the proof of Theorem 2.52 show that the
test is inconclusive when 𝐿 = 1.
□
∞ 23𝑛+4
Example 2.56 Consider the series ∑𝑛=1 .
𝑛𝑛
We have 𝑛
23𝑛+4 24 ⋅ (23 )𝑛 8
𝑎𝑛 = = = 16 � � .
𝑛𝑛 𝑛𝑛 𝑛
𝑛
∞ 8
By Proposition 2.24, it suffices to show that the series ∑𝑛=1 � 𝑛 � converges. We
apply the Root Test and see that
1
𝑛
8 𝑛
8
𝐿 = lim �� � � = lim = 0 < 1.
𝑛→∞ 𝑛 𝑛→∞ 𝑛

∞ 23𝑛+4
Thus, the series ∑𝑛=1 𝑛𝑛
converges.
△

2.10 Rearrangements
Due to the commutative law of addition 𝑎 + 𝑏 = 𝑏 + 𝑎, 𝑎, 𝑏 ∈ ℝ, we know that the
sum of finitely many terms does not depend on the order of summation. In this
section we investigate the question whether this is also true for infinite series.
∞
Given a series ∑𝑛=1 𝑎𝑛 , we consider a series whose terms are the same, but
their order is different. The formal definition is the following.

54
 Chapter 2 Sequences and series

∞ ∞
Definition 2.57 A series ∑𝑛=1 𝑏𝑛 is called a rearrangement of the series ∑𝑛=1 𝑎𝑛
if there exists a bijection 𝑓 ∶ ℕ → ℕ such that 𝑏𝑓(𝑛) = 𝑎𝑛 , 𝑛 ∈ ℕ.
△
∞
Clearly, any series is a rearrangement of itself. If ∑𝑛=1 𝑏𝑛 is a rearrangement
∞ ∞ ∞ ∞
of ∑𝑛=1 𝑎𝑛 , then ∑𝑛=1 𝑎𝑛 is a rearrangement of ∑𝑛=1 𝑏𝑛 . If ∑𝑛=1 𝑏𝑛 is a rearrange-
∞ ∞ ∞ ∞
ment of ∑𝑛=1 𝑎𝑛 and ∑𝑛=1 𝑐𝑛 is a rearrangement of ∑𝑛=1 𝑏𝑛 , then ∑𝑛=1 𝑐𝑛 is a
∞
rearrangement of ∑𝑛=1 𝑎𝑛 . These properties follow from the properties of bijec-
tions.
The main result of this section says that the sum of a series does not depend
on the order of summation if the series is absolutely convergent.
∞
Theorem 2.58 If a series ∑𝑛=1 𝑎𝑛 is absolutely convergent, then any rearrangement
∞
of ∑𝑛=1 𝑎𝑛 converges to the same sum.
△
∞ ∞ ∞
Proof. Let ∑𝑛=1 𝑏𝑛 be a rearrangement of ∑𝑛=1 𝑎𝑛 . Denote by 𝑠 = ∑𝑛=1 𝑎𝑛 the
∞
sum of the series ∑𝑛=1 𝑎𝑛 . Put
𝑚 𝑚
𝑠𝑚 = � 𝑎𝑛 , 𝑡 𝑚 = � 𝑏𝑛 , 𝑚 ∈ ℕ.
𝑛=1 𝑛=1

We know that lim𝑚→∞ 𝑠𝑚 = 𝑠. Our aim is to show that lim𝑚→∞ 𝑡𝑚 = 𝑠.

Let 𝜀 > 0. By the definition of convergence, there exists 𝑁1 ∈ ℕ such that
𝜀
|𝑠𝑚 − 𝑠| < for all 𝑚 ≥ 𝑁1 .
2
∞
Since also the series ∑𝑛=1 |𝑎𝑛 | converges, by ?? there is 𝑁2 ∈ ℕ such that
∞
𝜀
� |𝑎𝑛 | < for all 𝑚 ≥ 𝑁2 .
𝑛=𝑚+1 2
∞ ∞
Put 𝑁 = max{𝑁1 , 𝑁2 }. Since ∑𝑛=1 𝑏𝑛 is a rearrangement of ∑𝑛=1 𝑎𝑛 , the terms
∞
𝑎1 , … , 𝑎𝑁 will all appear in ∑𝑛=1 𝑏𝑛 . We want to take 𝑚 large enough so that
𝑎1 , … , 𝑎𝑁 all appear in 𝑡𝑚 . Let 𝑓 ∶ ℕ → ℕ be a bijection such that 𝑎𝑛 = 𝑏𝑓(𝑛) ,
𝑛 ∈ ℕ. Put
𝑀 = max{𝑓(1), 𝑓(2), … , 𝑓(𝑁)}.
Then for any 𝑚 ≥ 𝑀 the partial sum 𝑡𝑚 includes all terms 𝑎1 , … , 𝑎𝑁 . The differ-
ence 𝑡𝑚 − 𝑠𝑁 consists of some terms 𝑎𝑗 whose indices satisfy 𝑗 > 𝑁. Thus,
∞
𝜀
|𝑡𝑚 − 𝑠𝑁 | ≤ � |𝑎𝑗 | < .
𝑗=𝑁+1
2

55
 Chapter 2 Sequences and series

Therefore for all 𝑚 ≥ 𝑀 we have

𝜀 𝜀
|𝑡𝑚 − 𝑠| ≤ |𝑡𝑚 − 𝑠𝑁 | + |𝑠𝑁 − 𝑠| < + = 𝜀.
2 2
∞
This implies that lim𝑚→∞ 𝑡𝑚 = 𝑠, i.e. ∑𝑛=1 𝑏𝑛 = 𝑠.
□

The content of the following remark is not included in the obligatory part
of the course, but it gives you an important information about the behavior of
rearrangements.
Remark We can now ask ourselves whether the property described in Theorem 2.58
is true for any series. It turns out that the statement does not hold when the series is
conditionally convergent. Moreover, there is the following perhaps surprising result.
∞
Theorem 2.59 (The Riemann Rearrangement Theorem) If a series ∑𝑛=1 𝑎𝑛 is conditionally
∞
convergent, then for any 𝑥 ∈ ℝ there is a rearrangement of ∑𝑛=1 𝑎𝑛 whose sum is 𝑥.
△
We will not give a formal proof of this result. However, we will consider an example
that demonstrates the method.
We know that the alternating harmonic series
∞
(−1)𝑛−1 1 1 1 1 1
� =1− + − + − +⋯
𝑛=1 𝑛 2 3 4 5 6

is conditionally convergent, and its sum is ln 2 = 0.693147 …. The series built from the
positive terms
∞
1 1 1
� =1+ + +⋯
𝑘=1
2𝑘 − 1 3 5
as well as the series built from the absolute values of the negative terms
∞
1 1 1 1
� = + + +⋯
𝑘=1
2𝑘 2 4 6

both diverge, their partial sums increase to ∞.

𝑛−1
∞ (−1)
Now let us, for example, take 𝑥 = 1.1. We will construct a rearrangement of ∑𝑛=1 𝑛
whose sum is 1.1. In the first step we add positive terms until the sum exceeds 𝑥 = 1.1:

1
1+ = 1.333333 … > 1.1.
3
In the second step we add negative terms until the resulting sum is smaller than 𝑥 = 1.1:

1 1
1+ − = 0.833333 … < 1.1.
3 2

56
 Chapter 2 Sequences and series

In the next step we again add positive terms until the sum is larger that 𝑥 = 1.1:

1 1 1 1
1+ − + + = 1.176190 … > 1.1.
3 2 5 7
Then we add negative terms again until the sum falls below 𝑥 = 1.1:

1 1 1 1 1
1+ − + + − = 0.926190 … < 1.1.
3 2 5 7 4
The next steps are

1 1 1 1 1 1 1
1+ − + + − + + = 1.128210 … > 1.1,
3 2 5 7 4 9 11
1 1 1 1 1 1 1 1
1+ − + + − + + − = 0.961544 … < 1.1,
3 2 5 7 4 9 11 6
and so on. Since the terms of the series go to 0, it is intuitively clear that the partial
sums of the rearrangement built this way will converge to 𝑥 = 1.1.
△

57
Chapter 3
The topology of ℝ

3.1 Open sets

Definition 3.1 Given 𝑥 ∈ ℝ and 𝜀 > 0, the 𝜀-neighborhood of 𝑥 is the set

𝑉𝜀 (𝑥) = {𝑦 ∈ ℝ ∶ |𝑥 − 𝑦| < 𝜀} = (𝑥 − 𝜀, 𝑥 + 𝜀).

△
Definition 3.2 A set 𝐺 ⊆ ℝ is called open if for each 𝑥 ∈ 𝐺 there is 𝜀 > 0 such
that 𝑉𝜀 (𝑥) = (𝑥 − 𝜀, 𝑥 + 𝜀) ⊆ 𝐺.
△
Notice that we regard the empty set as being open.
Example 3.3 An interval (𝑎, 𝑏) with 𝑎, 𝑏 ∈ ℝ, 𝑎 < 𝑏 is open. Indeed, consider
an arbitrary 𝑥 ∈ (𝑎, 𝑏). Take 𝜀 = min{𝑥 − 𝑎, 𝑏 − 𝑥} > 0. Then (𝑥 − 𝜀, 𝑥 + 𝜀) ⊆ (𝑎, 𝑏).
Indeed, if 𝑦 ∈ (𝑥 − 𝜀, 𝑥 + 𝜀), then

𝑦 > 𝑥 − 𝜀 ≥ 𝑥 − (𝑥 − 𝑎) = 𝑎 and 𝑦 < 𝑥 + 𝜀 ≤ 𝑥 − (𝑏 − 𝑥) = 𝑏,

so that 𝑦 ∈ (𝑎, 𝑏).

△
Similarly, the sets (−∞, 𝑎), (𝑎, ∞) with 𝑎 ∈ ℝ are open. Also the set ℝ is open;
in this case any 𝜀-neighborhood of any point 𝑥 ∈ ℝ is in ℝ.
Example 3.4 An interval [𝑐, 𝑑] with 𝑐, 𝑑 ∈ ℝ, 𝑐 ≤ 𝑑 is not open. Indeed,
the point 𝑐 ∈ [𝑐, 𝑑] does not possess an 𝜀-neighborhood 𝑉𝜀 (𝑐) that lies in [𝑐, 𝑑].
1 1
Namely, for any 𝜀 > 0 the point 𝑐− 2 𝜀 lies in 𝑉𝜀 (𝑐) = (𝑐−𝜀, 𝑐+𝜀), but 𝑐− 2 𝜀 ∉ [𝑐, 𝑑].
Thus, 𝑉𝜀 (𝑐) ⊈ [𝑐, 𝑑].
△
Theorem 3.5 ( 1 ) The union of any collection of open sets is open.

58
 Chapter 3 The topology of ℝ

(2 ) The intersection of a finite collection of open sets is open.

△

Proof. ( 1 ) Let {𝐺𝑖 ∶ 𝑖 ∈ 𝐼} be a collection of open sets, with 𝐼 being the index
set of arbitrary cardinality. If ⋃𝑖∈𝐼 𝐺𝑖 = ∅, then we are done, so assume
that ⋃𝑖∈𝐼 𝐺𝑖 ≠ ∅. Take 𝑥 ∈ ⋃𝑖∈𝐼 𝐺𝑖 . Then 𝑥 ∈ 𝐺𝑗 for some 𝑗 ∈ 𝐼. Since 𝐺𝑗 is
open, there is an 𝜀-neighborhood 𝑉𝜀 (𝑥) ⊆ 𝐺𝑗 . But this implies that also
𝑉𝜀 (𝑥) ⊆ ⋃𝑖∈𝐼 𝐺𝑖 . Thus, ⋃𝑖∈𝐼 𝐺𝑖 is open.
𝑛
(2 ) Let {𝐺1 , 𝐺2 , … , 𝐺𝑛 } be a finite collection of open sets. If ⋂𝑖=1 𝐺𝑖 = ∅, then
𝑛 𝑛
we are done, so assume that ⋂𝑖=1 𝐺𝑖 ≠ ∅. Take 𝑥 ∈ ⋂𝑖=1 𝐺𝑖 . Then 𝑥 ∈ 𝐺𝑗
for all 𝑗 = 1, 2, … , 𝑛. Since each 𝐺𝑗 is open, there exists 𝜀𝑗 > 0 such that
𝑛
𝑉𝜀𝑗 (𝑥) ⊆ 𝐺𝑗 . Put 𝜀 = min{𝜀1 , 𝜀2 , … , 𝜀𝑛 } > 0. We claim that 𝑉𝜀 (𝑥) ⊆ ⋂𝑖=1 𝐺𝑖 .
Indeed, since 𝜀 ≤ 𝜀𝑗 , 𝑉𝜀 (𝑥) ⊆ 𝑉𝜀𝑗 (𝑥) ⊆ 𝐺𝑗 for each 𝑗 = 1, 2, … , 𝑛. Thus,
⋂𝑛 𝐺𝑖 is open.
𝑖=1
□

Remark Statement (2) is in general not true for an intersection of infinitely

1 1
many open sets. As an example, consider 𝐺𝑘 = �− 𝑘 , 1 + 𝑘 �, 𝑘 ∈ ℕ. Each 𝐺𝑘 is
open, but ⋂𝑘∈ℕ 𝐺𝑘 = [0, 1] is not open.
△
The next notion and the following statement will prove to be useful later.
Definition 3.6 Let 𝐴 ⊆ ℝ. We say that a sequence (𝑥𝑛 )∞
𝑛=1 is eventually in 𝐴, if
there is 𝑁 ∈ ℕ such that 𝑥𝑛 ∈ 𝐴 for all 𝑛 ≥ 𝑁.
△
Proposition 3.7 Let 𝐺 be an open set. If a sequence (𝑥𝑛 )∞
𝑛=1 converges to 𝑥 ∈ 𝐺, then
(𝑥𝑛 )∞
𝑛=1 is eventually in 𝐺.
△

Proof. Since 𝐺 is open, there exists 𝜀 > 0 such that 𝑉𝜀 (𝑥) = (𝑥 − 𝜀, 𝑥 + 𝜀) ⊆ 𝐺. In

other words, for 𝑦 ∈ ℝ we have

|𝑦 − 𝑥| < 𝜀 ⟹ 𝑦 ∈ 𝐺.

Since lim𝑛→∞ 𝑥𝑛 = 𝑥, there is 𝑁 ∈ ℕ such that |𝑥𝑛 − 𝑥| < 𝜀 for all 𝑛 ≥ 𝑁. By

above, 𝑥𝑛 ∈ 𝐺 for all 𝑛 ≥ 𝑁. This means that (𝑥𝑛 )∞
𝑛=1 is eventually in 𝐺.
□

59
 Chapter 3 The topology of ℝ

3.2 Closed sets

Definition 3.8 Let 𝐴 ⊆ ℝ.

(1) A point 𝑥 ∈ ℝ is called a limit point of 𝐴, if any 𝜀-neighborhood 𝑉𝜀 (𝑥) of

the point 𝑥 contains a point of 𝐴 different from 𝑥.

(2 ) The set 𝐴′ of all limit points of 𝐴 is called the derived set of 𝐴.

△
Other terms for limit points are cluster points or accumulation points. The name
“limit points” is explained by the following characterization of limit points.
Proposition 3.9 A point 𝑥 is a limit point of a set 𝐴 if and only if there is a sequence
(𝑥𝑛 )∞
𝑛=1 in 𝐴 ⧵ {𝑥} with lim𝑛→∞ 𝑥𝑛 = 𝑥.
△

Proof. a. Suppose that 𝑥 is a limit point of the set 𝐴. For each 𝑛 ∈ ℕ, the
1 1
neighborhood 𝑉1/𝑛 (𝑥) = �𝑥 − 𝑛 , 𝑥 + 𝑛 � contains at least one point of 𝐴 that is
1
different from 𝑥. Let 𝑥𝑛 be such a point. Note that |𝑥𝑛 −𝑥| < 𝑛 . Then the sequence
(𝑥𝑛 )∞
𝑛=1 lies in 𝐴 ⧵ {𝑥} and lim𝑛→∞ 𝑥𝑛 = 𝑥.

b. Now assume that a sequence (𝑥𝑛 )∞ 𝑛=1 lies in 𝐴 ⧵ {𝑥} and lim𝑛→∞ 𝑥𝑛 = 𝑥. We
want to show that 𝑥 is a limit point of 𝐴. Let 𝜀 > 0. By the definition of
the limit, we can find 𝑁 ∈ ℕ such that |𝑥 − 𝑥𝑛 | < 𝜀 for all 𝑛 ≥ 𝑁. But then
𝑥𝑁 ∈ (𝑥 − 𝜀, 𝑥 + 𝜀) = 𝑉𝜀 (𝑥) and 𝑥𝑁 ≠ 𝑥. Since this is true for any 𝜀 > 0, 𝑥 is a limit
point of 𝐴.
□

Example 3.10 Let 𝐴 = (0, 1] ∪ {2}. We claim that 𝐴′ = [0, 1].

If 𝑥 ∈ (−∞, 0), then 𝑥 cannot be a limit point of 𝐴, because (−∞, 0) is an open
set and by Proposition 3.7 any sequence converging to 𝑥 must be eventually
in (−∞, 0). But (−∞, 0) does not contain any points from 𝐴 ⧵ {𝑥}. By exactly the
same argument, 𝑥 ∈ (1, ∞) cannot be a limit point of 𝐴. Notice that also 𝑥 = 2 is
not a limit point of 𝐴 since (1, ∞) does not contain any points from 𝐴 ⧵ {2}.
𝑛 ∞
Every point 𝑥 ∈ (0, 1] is a limit point of 𝐴. For example, the sequence �𝑥 � 𝑛+1 𝑛=1
lies in 𝐴 ⧵ {𝑥} and converges to 𝑥. Finally, 𝑥 = 0 is a limit point of 𝐴. In this case
∞
1
we can take the sequence � 𝑛 � , it lies in 𝐴 ⧵ {0} and converges to 0.
𝑛=1
△

60
 Chapter 3 The topology of ℝ

We have seen in the previous example that limit points of 𝐴 may belong or
not belong to 𝐴. On the other hand, not all points of 𝐴 are limit points of 𝐴.
Definition 3.11 If 𝑥 ∈ 𝐴 and 𝑥 is not a limit point of 𝐴, then 𝑥 is called an isolated
point of 𝐴.
△
Thus, each point of 𝐴 is a limit point of 𝐴 or an isolated point of 𝐴. In the
above example of 𝐴 = (0, 1] ∪ {2} the point 𝑥 = 2 is an isolated point of 𝐴 and
all points 𝑥 ∈ (0, 1] are limit points of 𝐴.
Example 3.12 Consider the set 𝐴 = ℕ. Any convergent sequence in ℕ is
eventually constant. This implies that ℕ has no limit points. Each point in ℕ is
an isolated point.
△
1
Example 3.13 Consider the set 𝐴 = � 𝑛 ∶ 𝑛 ∈ ℕ�. The only limit point of 𝐴 is
1
0 ∉ 𝐴. Each point 𝑛 , 𝑛 ∈ ℕ is an isolated point of ℕ.
△
Definition 3.14 A set 𝐹 ⊆ ℝ is called closed if it contains all its limit points, i.e.
if 𝐹′ ⊆ 𝐹.
△
Example 3.15 The set [𝑎, 𝑏] with 𝑎, 𝑏 ∈ ℝ, 𝑎 ≤ 𝑏 is closed. Indeed, for any
convergent sequence (𝑥𝑛 )∞ 𝑛=1 in [𝑎, 𝑏] we have 𝑎 ≤ 𝑥𝑛 ≤ 𝑏, 𝑛 ∈ ℕ, and thus
by Theorem 2.15 𝑎 ≤ lim𝑛→∞ 𝑥𝑛 ≤ 𝑏. Thus, all limits of all converging sequences
in [𝑎, 𝑏] and, in particular, all limit points are in [𝑎, 𝑏].
△
The sets ∅, ℝ, (−∞, 𝑎] and [𝑎, ∞) with 𝑎 ∈ ℝ are closed. The set (𝑎, 𝑏) is not
closed as it does not contain its limit points 𝑎 and 𝑏.
Notice that “closed” does not mean “not open” and vice versa. Thus, the sets
∅ and ℝ are simultaneously open and closed. On the other hand, the sets (0, 1]
and [0, 1) are neither open nor closed.
Theorem 3.16 A set 𝐴 ⊆ ℝ is open if and only if its complement ℝ ⧵ 𝐴 is closed.
△

Proof. Denote 𝐵 = ℝ ⧵ 𝐴.

a. Assume that 𝐴 is open. We want to show that 𝐵 is closed, i.e. all its limit
points are in 𝐵. If 𝑥 ∈ 𝐴, then 𝑥 cannot be a limit point of 𝐵, because by Proposi-

61
 Chapter 3 The topology of ℝ

tion 3.7 any sequence with limit 𝑥 must be eventually in 𝐴, and this is impossible
for a sequence in 𝐵 ⧵ {𝑥}. Thus, if 𝑥 is a limit point of 𝐵, then 𝑥 ∉ 𝐴, i.e. 𝑥 ∈ 𝐵.

b. Now assume that 𝐵 is closed. We want to show that 𝐴 is open. Take 𝑥 ∈ 𝐴.

Since 𝐵 is closed, it contains all its limit points. Therefore 𝑥 cannot be a limit
point of 𝐵. By the definition of a limit point there must exist a 𝜀-neighborhood
𝑉𝜀 (𝑥) that does not contain points from 𝐵 ⧵ {𝑥}. Since also 𝑥 ∉ 𝐵, we conclude
that 𝑉𝜀 (𝑥) does not contain any points of 𝐵. Hence, 𝑉𝜀 (𝑥) ⊆ 𝐴. This means that
𝐴 is open.

Theorem 3.17 ( 1 ) The intersection of any collection of closed sets is closed.

(2 ) The union of a finite collection of closed sets is closed.

△

Proof. Both statements follow from Theorem 3.5 and De Morgan’s Laws
⎛ ⎞ ⎛ ⎞
⎜⎜ ⎟⎟ ⎜⎜ ⎟
ℝ ⧵ ⎝� 𝐴𝑖 ⎠ = �(ℝ ⧵ 𝐴𝑖 ), ℝ ⧵ ⎝� 𝐴𝑖 ⎟⎟⎠ = �(ℝ ⧵ 𝐴𝑖 ).
⎜ ⎟ ⎜
𝑖∈𝐼 𝑖∈𝐼 𝑖∈𝐼 𝑖∈𝐼

Definition 3.18 The set 𝐴 = 𝐴 ∪ 𝐴′ is called a closure of the set 𝐴.

△
Theorem 3.19 Let 𝐴 ⊆ ℝ. The closure 𝐴 is closed. Moreover, 𝐴 is the smallest
closed set containing 𝐴.
△

Proof. a. We show that 𝐴 = 𝐴 ∪ 𝐴′ is closed. Let 𝑥 be a limit point of 𝐴. We

have to show that 𝑥 ∈ 𝐴.
If 𝑥 ∈ 𝐴, then we are done since 𝐴 ⊆ 𝐴. Therefore suppose that 𝑥 ∉ 𝐴.
Let (𝑥𝑛 )∞
𝑛=1 be a sequence in 𝐴 ⧵ {𝑥} that converges to 𝑥. For each 𝑗 ∈ ℕ, there
is 𝑁𝑗 ∈ ℕ such that
1
|𝑥𝑛 − 𝑥| < for all 𝑛 ≥ 𝑁𝑗 .
2𝑗

62
 Chapter 3 The topology of ℝ

1
It follows that we can choose 𝑛1 ∈ ℕ such that |𝑥𝑛1 − 𝑥| < 2 , then choose 𝑛2 > 𝑛1
1 1
such that |𝑥𝑛2 − 𝑥| < 2⋅2
, then choose 𝑛3 > 𝑛2 such that |𝑥𝑛3 − 𝑥| < 2⋅3
, and so on.
In this way we obtain a subsequence (𝑥𝑛𝑗 )∞
𝑗=1 , lying in 𝐴 ⧵ {𝑥}, with

1
|𝑥𝑛𝑗 − 𝑥| < , 𝑗 ∈ ℕ.
2⋅𝑗

Now we will construct a sequence (𝑎𝑗 )𝑗∈ℕ in 𝐴 that converges to 𝑥. If 𝑥𝑛𝑗 ∈ 𝐴,

we put 𝑎𝑗 = 𝑥𝑛𝑗 . If 𝑥𝑛𝑗 ∉ 𝐴, then 𝑥𝑛𝑗 ∈ 𝐴′ , i.e., 𝑥𝑛𝑗 is a limit point of 𝐴. There is a
sequence in 𝐴 that converges to 𝑥𝑛𝑗 . We choose 𝑎𝑗 to be a term of this sequence
with
1
|𝑥𝑛𝑗 − 𝑎𝑗 | < .
2𝑗
Notice that the above estimate holds also in the case when 𝑎𝑗 = 𝑥𝑛𝑗 .
The constructed sequence (𝑎𝑗 )∞
𝑗=1 lies in 𝐴 and converges to 𝑥. Indeed,

1 1 1
|𝑎𝑗 − 𝑥| ≤ |𝑎𝑗 − 𝑥𝑛𝑗 | + |𝑥𝑛𝑗 − 𝑥| < + = , 𝑗 ∈ ℕ.
2𝑗 2𝑗 𝑗

Thus, 𝑥 ∈ 𝐴′ .
The above shows that any limit point of 𝐴 lies in 𝐴 ∪ 𝐴′ = 𝐴. Thus, 𝐴 is
closed.

b. Let us now prove that 𝐴 is the smallest closed set containing 𝐴. Clearly,
𝐴 ⊆ 𝐴 and 𝐴 is closed.
Let 𝐹 be a closed set with 𝐴 ⊆ 𝐹. We want to show that 𝐴 = 𝐴 ∪ 𝐴′ ⊆ 𝐹.
The inclusion 𝐴 ⊆ 𝐹 is given, so that we have to show that 𝐴′ ⊆ 𝐹. Let
𝑥 ∈ 𝐴′ . Then 𝑥 is a limit point of 𝐴, so that there is a sequence (𝑥𝑛 )∞
𝑛=1 in 𝐴 with
𝑥 = lim𝑛→∞ 𝑥𝑛 . But 𝐴 ⊆ 𝐹 which means that (𝑥𝑛 )∞ 𝑛=1 is a sequence in 𝐹. Thus,
𝑥 is a limit point of 𝐹. Since 𝐹 is closed, it contains all its limit points. Hence,
𝑥 ∈ 𝐹, and we are done.
□

Proposition 3.20 Let 𝐴 be a nonempty set in ℝ.

(1) If 𝐴 is bounded above, then sup 𝐴 ∈ 𝐴.

(2 ) If 𝐴 is bounded below, then inf 𝐴 ∈ 𝐴.

63
 Chapter 3 The topology of ℝ

Proof. We prove (1); a proof of (2) is similar.

Let 𝐴 be a nonempty set in ℝ that is bounded above. By the Completeness
Axiom (Axiom 1.8), 𝐴 has a least upper bound sup 𝐴 = 𝑎. If 𝑎 ∈ 𝐴, then 𝑎 ∈ 𝐴
by the inclusion 𝐴 ⊆ 𝐴. If 𝑎 ∉ 𝐴, then for each 𝑛 ∈ ℕ by Proposition 1.11
1
we can find 𝑎𝑛 ∈ 𝐴 with 𝑎 − 𝑛 < 𝑎𝑛 < 𝑎. We obtain a sequence (𝑎𝑛 )∞ 𝑛=1 in
𝐴 ⧵ {𝑎} that converges to 𝑎. Thus, 𝑎 = sup 𝐴 is a limit point of 𝐴, and therefore
𝑎 = sup 𝐴 ∈ 𝐴.
□

3.3 Compact sets

Definition 3.21 A set 𝐾 ⊆ ℝ is called compact if every sequence in 𝐾 has a
subsequence that converges to a point in 𝐾.
△
Theorem 3.22 A set 𝐾 ⊆ ℝ is compact if and only if 𝐾 is closed and bounded.
△

Proof. a. Suppose that 𝐾 is compact.

We first show that 𝐾 is closed. Let 𝑥 be a limit point of 𝐾. Then 𝑥 is a limit of
a sequence (𝑥𝑛 )∞ ∞
𝑛=1 in 𝐾. Since 𝐾 is compact, (𝑥𝑛 )𝑛=1 contains a subsequence that
converges to an element of 𝐾. But all subsequences of a convergent sequence
converge to the same limit (see Proposition 2.38). Thus, 𝑥 ∈ 𝐾. This shows that
all limit points of 𝐾 belong to 𝐾, i.e., 𝐾 is closed.
Now let us prove that 𝐾 is bounded. Assume, by contradiction, that 𝐾 is
unbounded. Then for each 𝑛 ∈ ℕ we can find a point in 𝐾, say 𝑥𝑛 , with
|𝑥𝑛 | > 𝑛. Consider the sequence (𝑥𝑛 )∞𝑛=1 in 𝐾. Clearly, any subsequence of
(𝑥𝑛 )∞
𝑛=1 is unbounded and therefore divergent. Thus, (𝑥𝑛 )∞
𝑛=1 does not contain a
convergent subsequence which contradicts the definition of a compact set. We
conclude that 𝐾 must be bounded.

b. Now assume that 𝐾 is closed and bounded. We want to show that 𝐾 is

compact.
Let (𝑥𝑛 )∞ ∞
𝑛=1 be an arbitrary sequence in 𝐾. Since 𝐾 is bounded, (𝑥𝑛 )𝑛=1 is
∞
bounded. By the Bolzano-Weierstrass Theorem (Theorem 2.40), (𝑥𝑛 )𝑛=1 has a

64
 Chapter 3 The topology of ℝ

convergent subsequence. Since 𝐾 is closed, its limit belongs to 𝐾. Thus, 𝐾 is

compact.
□

Examples of compact sets are intervals [𝑎, 𝑏] with 𝑎, 𝑏 ∈ ℝ, 𝑎 ≤ 𝑏. Also notice

that any finite set in ℝ is compact.

65
Chapter 4
Limits of functions and continuity

4.1 Limits of functions

Definition 4.1 Let 𝐴 ⊆ ℝ be a nonempty set and 𝑓 ∶ 𝐴 → ℝ. Let 𝑎 be a limit
point of 𝐴. We say that 𝑓 converges to a limit 𝐿 ∈ ℝ as 𝑥 → 𝑎 if

∀ 𝜀 > 0 ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 0 < |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝐿| < 𝜀,

i.e. for any 𝜀 > 0 there exists 𝛿 > 0 such that |𝑓(𝑥) − 𝐿| < 𝜀 for all 𝑥 ∈ 𝐴 with
0 < |𝑥 − 𝑎| < 𝛿. We write lim𝑥→𝑎 𝑓(𝑥) = 𝐿.
△
For an illustration for this definition, please see Figure 4.1.
It is not difficult to prove that if a function 𝑓 has a limit at a point 𝑎, then the
limit is unique (compare with Proposition 2.4).
Notice that we do not require that 𝑎 ∈ 𝐴. In particular, 𝑓(𝑎) need not be
defined. However, 𝑓 must be defined at points in 𝐴 that are arbitrary close to 𝑎.
This is ensured by assuming that 𝑎 is a limit point of 𝐴.
The condition 0 < |𝑥 − 𝑎| in the definition of the limit means that we do not
consider the value of the function 𝑓 at the point 𝑎, even if it exists. In particular,
it can happen that 𝑓(𝑎) ≠ 𝐿.
Typically, the set 𝐴 in the definition is the domain of the function 𝑓. However,
in some situations it can be useful to restrict 𝐴. For example, we can ask what
happens if we approach the point 𝑎 from one side, say, from the left. In this
case we take 𝐴 to be the set of points 𝑥 from the domain of 𝑓 with 𝑥 < 𝑎. If we
approach 𝑎 from the right, we take 𝐴 to be the set of points 𝑥 from the domain
of 𝑓 with 𝑥 > 𝑎. In this case we talk about one-sided limits.
𝑥
For example, consider the function 𝑓(𝑥) = |𝑥| . This function is not defined at

66
 Chapter 4 Limits of functions and continuity

𝑦
𝐿+𝜀
𝐿
𝐿−𝜀

𝑎−𝛿 𝑎 𝑎+𝛿 𝑥

Figure 4.1: A limit of a function

𝑥 = 0. The right-sided limit of 𝑓 at 𝑥 = 0 is

𝑥
lim 𝑓(𝑥) = lim = 1.
𝑥→0 𝑥→0 𝑥
𝑥>0 𝑥>0

We also write lim𝑥→0+ 𝑓(𝑥) = 1. The left-sided limit at 𝑥 = 0 is

𝑥
lim 𝑓(𝑥) = lim = −1.
𝑥→0 𝑥→0 −𝑥
𝑥<0 𝑥<0

We also write lim𝑥→0− 𝑓(𝑥) = −1.

Now let us consider some examples how to use Definition 4.1.
Example 4.2 We have lim𝑥→2 (𝑥2 + 3) = 7.
Here we consider the function 𝑓(𝑥) = 𝑥2 + 3, its domain 𝐴 = ℝ and 𝑎 = 2
which is a limit point of 𝐴. We have

|𝑓(𝑥) − 7| = |𝑥2 + 3 − 7| = |𝑥2 − 4| = |𝑥 − 2| ⋅ |𝑥 + 2|.

Since we are interested in the behavior of 𝑓 for 𝑥 close to 2, let us assume that
|𝑥 − 2| < 1, i.e. 1 < 𝑥 < 3. For such 𝑥 we have 3 < 𝑥 + 2 < 5, so that 3 < |𝑥 + 2| < 5
and
|𝑓(𝑥) − 7| < 5|𝑥 − 2|.
𝜀
Given 𝜀 > 0, chose 𝛿 = min �1, 5 �. Then 0 < |𝑥 − 2| < 𝛿 implies that
𝜀
|𝑓(𝑥) − 7| < 5|𝑥 − 2| < 5 ⋅ = 𝜀.
5
Thus, by definition, lim𝑥→2 (𝑥2 + 3) = 7.
△

67
 Chapter 4 Limits of functions and continuity

Example 4.3 Let 𝑔 ∶ ℝ → ℝ, and suppose that lim𝑥→6 𝑔(𝑥) = √2. We will
prove that
1 1
lim 2
= .
𝑥→6 1 + 𝑔(𝑥) 3
By the definition of the limit, we have to show the following:

1 1
∀ 𝜀 > 0 ∃𝛿 > 0 ∀ 𝑥 ∈ ℝ ∶ 0 < |𝑥 − 6| < 𝛿 ⟹ � 2
− � < 𝜀.
1 + 𝑔(𝑥) 3

We have

1 1 3 − 1 − 𝑔(𝑥)2 |2 − 𝑔(𝑥)2 | �√2 − 𝑔(𝑥)� ⋅ �√2 + 𝑔(𝑥)�

� − � = � � = = .
1 + 𝑔(𝑥)2 3 3(1 + 𝑔(𝑥)2 ) 3(1 + 𝑔(𝑥)2 ) 3(1 + 𝑔(𝑥)2 )
1
Since 𝑔(𝑥)2 ≥ 0, we have 1 + 𝑔(𝑥)2 ≥ 1, and ≤ 1. It follows that
1+𝑔(𝑥)2

1 1 1
� 2
− � ≤ ⋅ �√2 − 𝑔(𝑥)� ⋅ �√2 + 𝑔(𝑥)� .
1 + 𝑔(𝑥) 3 3

We know that lim𝑥→6 𝑔(𝑥) = √2. Let us estimate �√2 + 𝑔(𝑥)�. We take in the
definition of the limit, say, 𝜀 = 1. There exists 𝛿1 > 0 such that

0 < |𝑥 − 6| < 𝛿1 ⟹ �𝑔(𝑥) − √2� < 1.

For such 𝑥 we have

√2 − 1 < 𝑔(𝑥) < √2 + 1,
and therefore
0 < 2√2 − 1 < 𝑔(𝑥) + √2 < 2√2 + 1 < 5.
We conclude that

0 < |𝑥 − 6| < 𝛿1 ⟹ �𝑔(𝑥) + √2� < 5.

Thus, for 𝑥 ∈ ℝ with 0 < |𝑥 − 6| < 𝛿1 we have

1 1 5
� − � < ⋅ �√2 − 𝑔(𝑥)� .
1 + 𝑔(𝑥)2 3 3

Given 𝜀 > 0, there exists 𝛿2 > 0 such that

3
0 < |𝑥 − 6| < 𝛿2 ⟹ �𝑔(𝑥) − √2� < 𝜀.
5

68
 Chapter 4 Limits of functions and continuity

Now put 𝛿 = min{𝛿1 , 𝛿2 } > 0. We have

1 1 5 5 3
0 < |𝑥 − 6| < 𝛿 ⟹ � 2
− � < ⋅ �√2 − 𝑔(𝑥)� < ⋅ 𝜀 = 𝜀
1 + 𝑔(𝑥) 3 3 3 5
1 1
as required. Thus, by definition lim𝑥→6 = 3.
1+𝑔(𝑥)2
△
It is easy to prove following statements immediately from the definition.
Proposition 4.4 If 𝑓(𝑥) = 𝑐 with some 𝑐 ∈ ℝ for all 𝑥 ∈ 𝐴, and if 𝑎 is a limit point
of 𝐴, then lim𝑥→𝑎 𝑓(𝑥) = 𝑐.
△
Proposition 4.5 If 𝑓(𝑥) = 𝑥 for all 𝑥 ∈ 𝐴, and if 𝑎 is a limit point of 𝐴, then
lim𝑥→𝑎 𝑓(𝑥) = 𝑎.
△
The following result says that if a function 𝑓 has a positive (negative) limit at
a point 𝑎, then there is an open neighborhood around 𝑎 such that 𝑓(𝑥) is positive
(negative) for all 𝑥 in this neighborhood, 𝑥 ≠ 𝑎.
Lemma 4.6 Suppose that lim𝑥→𝑎 𝑓(𝑥) = 𝐿.

(1) If 𝐿 > 0, then ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 0 < |𝑥 − 𝑎| < 𝛿 ⟹ 𝑓(𝑥) > 0.

(2 ) If 𝐿 < 0, then ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 0 < |𝑥 − 𝑎| < 𝛿 ⟹ 𝑓(𝑥) < 0.

Proof. We will prove (1); the proof of (2) is similar.

𝐿
Let 𝐿 > 0. In the definition of the limit, take 𝜀 = 2
> 0. There exists 𝛿 > 0
𝐿
such that 0 < |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝐿| < 2 . Then, in particular, for such 𝑥

𝐿 𝐿
𝑓(𝑥) > 𝐿 − = > 0.
2 2
□

We have seen that limits of sequences preserve the arithmetic operations

(Theorem 2.14). The same is true for the functional limits. We omit the proof
as it is similar to the proof of Theorem 2.14. Please prove the statements as an
exercise.

69
 Chapter 4 Limits of functions and continuity

Theorem 4.7 (The Algebraic Limit Theorem for Functional Limits)

If lim𝑥→𝑎 𝑓(𝑥) = 𝐿 and lim𝑥→𝑎 𝑔(𝑥) = 𝑀, then

(1) lim𝑥→𝑎 𝑐𝑓(𝑥) = 𝑐𝐿 for any 𝑐 ∈ ℝ,

(2 ) lim𝑥→𝑎 (𝑓(𝑥) + 𝑔(𝑥)) = 𝐿 + 𝑀,

(3 ) lim𝑥→𝑎 (𝑓(𝑥)𝑔(𝑥)) = 𝐿𝑀,

𝑓(𝑥) 𝐿
(4) lim𝑥→𝑎 𝑔(𝑥)
= 𝑀
, provided 𝑀 ≠ 0.

△
Notice that in (4) the assumption 𝑀 ≠ 0 guarantees by Lemma 4.6 that
𝑔(𝑥) ≠ 0 in a certain neighborhood of 𝑎 except, perhaps, at 𝑥 = 𝑎. Therefore, the
𝑓
quotient 𝑔
is defined in this neighborhood for 𝑥 ≠ 𝑎.
Also the Order Limit Theorem for sequences (Theorem 2.15) has a analogue
for the functional limits.
Theorem 4.8 (The Order Limit Theorem for Functional Limits) Let lim𝑥→𝑎 𝑓(𝑥) = 𝐿
and lim𝑥→𝑎 𝑔(𝑥) = 𝑀. If 𝑓(𝑥) ≤ 𝑔(𝑥) for all 𝑥 ∈ 𝐴, 𝑥 ≠ 𝑎, then 𝐿 ≤ 𝑀.
△
This statement follows easily from Lemma 4.6.
In the next theorem we will give an equivalent definition of a functional limit
in terms of sequences.
Theorem 4.9 Let 𝑓 ∶ 𝐴 → ℝ, and let 𝑎 be a limit point of 𝐴. The following
statements are equivalent:

(1) lim𝑥→𝑎 𝑓(𝑥) = 𝐿,

(2 ) for any sequence (𝑥𝑛 )∞

𝑛=1 in 𝐴 ⧵ {𝑎} with the property lim𝑛→∞ 𝑥𝑛 = 𝑎 we have
lim𝑛→∞ 𝑓(𝑥𝑛 ) = 𝐿.

Proof. a. Suppose that lim𝑥→𝑎 𝑓(𝑥) = 𝐿. Let 𝜀 > 0 be given. By the definition of
the limit, there exists 𝛿 > 0 such that 0 < |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝐿| < 𝜀.
Now let (𝑥𝑛 )∞
𝑛=1 be a sequence in 𝐴 ⧵ {𝑎} such that lim𝑛→∞ 𝑥𝑛 = 𝑎. By the
definition of the limit of a sequence, there exists 𝑁 ∈ ℕ such that |𝑥𝑛 − 𝑎| < 𝛿 for
all 𝑛 ≥ 𝑁. But then by above |𝑓(𝑥𝑛 ) − 𝐿| < 𝜀. This implies that lim𝑛→∞ 𝑓(𝑥𝑛 ) = 𝐿.

70
 Chapter 4 Limits of functions and continuity

1
Figure 4.2: A graph of the function 𝑓(𝑥) = sin � 𝑥 �

b. Now assume that lim𝑛→∞ 𝑓(𝑥𝑛 ) = 𝐿 for any sequence (𝑥𝑛 )∞

𝑛=1 in 𝐴 ⧵ {𝑎} with
lim𝑛→∞ 𝑥𝑛 = 𝑎.
Suppose, by contradiction, that lim𝑥→𝑎 𝑓(𝑥) ≠ 𝐿. Then there is a number 𝜀 > 0
such that for each 𝑛 ∈ ℕ there is a point 𝑥𝑛 ∈ 𝐴 ⧵ {𝑎} with
1
0 < |𝑥𝑛 − 𝑎| < and |𝑓(𝑥𝑛 ) − 𝐿| ≥ 𝜀.
𝑛
Consider the sequence (𝑥𝑛 )∞
𝑛=1 . We have that lim𝑛→∞ 𝑥𝑛 = 𝑎 but lim𝑛→∞ 𝑓(𝑥𝑛 ) ≠
𝐿, a contradiction. The contradiction shows that lim𝑥→𝑎 𝑓(𝑥) = 𝐿.
□

Theorem 4.9 gives us useful criteria for divergence.

Corollary 4.10 (The Sequential Test for Divergence)

(1) If there exists a sequence (𝑥𝑛 )∞ ∞

𝑛=1 in 𝐴 ⧵ {𝑎} such that (𝑥𝑛 )𝑛=1 converges to 𝑎, but
(𝑓(𝑥𝑛 ))∞
𝑛=1 diverges, then lim𝑥→𝑎 𝑓(𝑥) does not exist.

(2 ) If there exist two sequences (𝑥𝑛 )∞ ∞

𝑛=1 and (𝑦𝑛 )𝑛=1 in 𝐴 ⧵ {𝑎} that both converge to 𝑎,
such that lim𝑛→∞ 𝑓(𝑥𝑛 ) ≠ lim𝑛→∞ 𝑓(𝑦𝑛 ), then lim𝑥→𝑎 𝑓(𝑥) does not exist.
△
1
Example 4.11 Consider the function 𝑓(𝑥) = sin � 𝑥 �, 𝑥 ≠ 0. Its graph is shown
in Figure 4.2.
1 1 1
The limit lim𝑥→0 sin � 𝑥 � does not exist. Indeed, take 𝑥𝑛 = 2𝜋𝑛
and 𝑦𝑛 = 𝜋 ,
2𝜋𝑛+ 2
𝑛 ∈ ℕ. Clearly, lim𝑛→∞ 𝑥𝑛 = lim𝑛→∞ 𝑦𝑛 = 0. We have 𝑓(𝑥𝑛 ) = sin(2𝜋𝑛) = 0,
𝜋
𝑛 ∈ ℕ, so that lim𝑛→∞ 𝑓(𝑥𝑛 ) = 0. On the other hand, 𝑓(𝑦𝑛 ) = sin �2𝜋𝑛 + 2 � = 1
for all 𝑛 ∈ ℕ, and lim𝑛→∞ 𝑓(𝑦𝑛 ) = 1. The statement follows by Corollary 4.10.

71
 Chapter 4 Limits of functions and continuity

△
Theorem 4.12 (The Squeeze Theorem for Functional Limits) Let 𝑓, 𝑔, ℎ ∶ 𝐴 → ℝ,
and let 𝑎 be a limit point of 𝐴. Suppose that lim𝑥→𝑎 𝑔(𝑥) = lim𝑥→𝑎 ℎ(𝑥) = 𝐿 and

𝑔(𝑥) ≤ 𝑓(𝑥) ≤ ℎ(𝑥) for all 𝑥 ∈ 𝐴.

Then lim𝑥→𝑎 𝑓(𝑥) = 𝐿.

△
This statement can be proven either directly, or using Theorem 4.9 and the
Squeeze Theorem for sequences (Theorem 2.10).
We finish this section by a brief discussion of infinite limits. First we define
infinite limits at a finite point.
Definition 4.13 Let 𝑓 ∶ 𝐴 → ℝ, and let 𝑎 be a limit point of 𝐴.

(1) We say that lim𝑥→𝑎 𝑓(𝑥) = ∞, if

∀ 𝑀 > 0 ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 0 < |𝑥 − 𝑎| < 𝛿 ⟹ 𝑓(𝑥) > 𝑀.

(2 ) Similarly, we say that lim𝑥→𝑎 𝑓(𝑥) = −∞, if

∀ 𝑀 > 0 ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 0 < |𝑥 − 𝑎| < 𝛿 ⟹ 𝑓(𝑥) < −𝑀.

△
Notice that in both cases the limit lim𝑥→𝑎 𝑓(𝑥) does not exist; the symbols ∞
and −∞ are not real numbers.
Finally, we define limits of a function at ∞ and −∞.
Definition 4.14 Let 𝑓 ∶ 𝐴 → ℝ and 𝐿 ∈ ℝ.

(1) Assume that 𝐴 is not bounded above. We say that lim𝑥→∞ 𝑓(𝑥) = 𝐿, if

∀ 𝜀 > 0 ∃𝑀 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 𝑥 > 𝑀 ⟹ |𝑓(𝑥) − 𝐿| < 𝜀.

(2 ) Assume that 𝐴 is not bounded below. We say that lim𝑥→−∞ 𝑓(𝑥) = 𝐿, if

∀ 𝜀 > 0 ∃𝑀 > 0 ∀ 𝑥 ∈ 𝐴 ∶ 𝑥 < −𝑀 ⟹ |𝑓(𝑥) − 𝐿| < 𝜀.

△
Combining ideas from Definition 4.13 and Definition 4.14, one can give
definitions for infinite limits at ∞ and −∞, for example, for lim𝑥→∞ 𝑓(𝑥) = ∞.
We let the details to the reader.

72
 Chapter 4 Limits of functions and continuity

4.2 Continuity
Definition 4.15 Let 𝑓 ∶ 𝐴 → ℝ, and let 𝑎 ∈ 𝐴. We say that 𝑓 is continuous at 𝑎, if

∀ 𝜀 > 0 ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝑓(𝑎)| < 𝜀.

If 𝑓 is continuous at each point 𝑎 ∈ 𝐴, we say that 𝑓 is continuous on 𝐴.

△
The definition above looks very similar to the definition of a limit. One
difference is that now we consider 𝑎 ∈ 𝐴. If 𝑎 ∈ 𝐴 is a limit point of 𝐴, the
definition above means exactly that

lim 𝑓(𝑥) = 𝑓(𝑎).

𝑥→𝑎

However, if 𝑎 is an isolated point of 𝐴, then lim𝑥→𝑎 𝑓(𝑥) is not defined. However,

it follows from the definition that if 𝑎 is an isolated point of 𝐴, then 𝑓 is always
continuous at 𝑎. This is because we can take 𝛿 > 0 so small that the only point
𝑥 ∈ 𝐴 with |𝑥 − 𝑎| < 𝛿 is the point 𝑥 = 𝑎.
The Algebraic Limit Theorem (Theorem 4.7) yields that sums, products and
quotients (under sensible assumptions) of continuous functions are continuous.
It follows that any polynomial function is continuous on ℝ, and any rational
function is continuous on its domain.
Theorem 4.16 (Composition of Continuous Functions) Let 𝑓 ∶ 𝐴 → ℝ and 𝑔 ∶
𝐵 → ℝ. Assume that 𝑓(𝐴) = {𝑓(𝑥) ∶ 𝑥 ∈ 𝐴} ⊆ 𝐵. If 𝑓 is continuous at 𝑎 ∈ 𝐴 and 𝑔 is
continuous at 𝑓(𝑎) ∈ 𝐵, then the composition 𝑔 ∘ 𝑓 ∶ 𝐴 → ℝ is continuous at 𝑎.
△
Notice that the assumption 𝑓(𝐴) = {𝑓(𝑥) ∶ 𝑥 ∈ 𝐴} ⊆ 𝐵 guarantees that the
composition 𝑔 ∘ 𝑓(𝑥) = 𝑔(𝑓(𝑥)) is defined on 𝐴.

Proof. Let 𝜀 > 0. By the continuity of 𝑔 at the point 𝑏 = 𝑓(𝑎), there exists 𝜂 > 0
such that
∀ 𝑦 ∈ 𝐵 ∶ |𝑦 − 𝑓(𝑎)| < 𝜂 ⟹ |𝑔(𝑦) − 𝑔(𝑓(𝑎))| < 𝜀.
On the other hand, 𝑓 is continuous at 𝑎, and therefore there exists 𝛿 > 0 such
that
∀ 𝑥 ∈ 𝐴 ∶ |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝑓(𝑎)| < 𝜂.
Combining the two formulas with 𝑦 = 𝑓(𝑥), we obtain

∀ 𝑥 ∈ 𝐴 ∶ |𝑥 − 𝑎| < 𝛿 ⟹ |𝑔(𝑓(𝑥)) − 𝑔(𝑓(𝑎))| < 𝜀

73
 Chapter 4 Limits of functions and continuity

as desired.
□

Theorem 4.9 gives the following criterion for continuity.

Theorem 4.17 (The Sequential Criterion for Continuity) Let 𝑓 ∶ 𝐴 → ℝ and 𝑎 ∈ 𝐴.
The following statements are equivalent:

(1) 𝑓 is continuous at 𝑎,

(2 ) for any sequence (𝑥𝑛 )∞

𝑛=1 in 𝐴 with lim𝑛→∞ 𝑥𝑛 = 𝑎 we have lim𝑛→∞ 𝑓(𝑥𝑛 ) =
𝑓(𝑎).

△
Corollary 4.18 If there is a sequence (𝑥𝑛 )∞
𝑛=1 in 𝐴 such that lim𝑛→∞ 𝑥𝑛 = 𝑎, but the
sequence (𝑓(𝑥𝑛 ))∞
𝑛=1 does not converge to 𝑓(𝑎), then 𝑓 is not continuous at 𝑎.
△
In the remaining part of this section we will study images and preimages of
sets under continuous functions. First we recall the definitions.
Definition 4.19 Let 𝑓 ∶ 𝑋 → 𝑌.

(1) The image of a set 𝐴 ⊆ 𝑋 is the set

𝑓(𝐴) = {𝑓(𝑥) ∶ 𝑥 ∈ 𝐴} ⊆ 𝑌.

(2 ) The preimage of a set 𝐵 ⊆ 𝑌 is the set

𝑓−1 (𝐵) = {𝑥 ∈ 𝑋 ∶ 𝑓(𝑥) ∈ 𝐵} ⊆ 𝑋.

△
Let 𝑓 ∶ ℝ → ℝ be a continuous function on ℝ. It is possible to show that
the preimage of any open set is open, and the preimage of any closed set is
closed. We will not go into the details. Notice, however, that images are not
well-behaved.
1
Example 4.20 Consider the function 𝑓(𝑥) = that is continuous on ℝ.
1+𝑥2
Observe that 𝑓(ℝ) = (0, 1], so that ℝ is closed but its image 𝑓(ℝ) = (0, 1] is not
1
closed. We also see that 𝑓((−1, 1)) = � 2 , 1�, where (−1, 1) is open, but its image
1
� , 1� is not open.
2
△

74
 Chapter 4 Limits of functions and continuity

𝑥
Example 4.21 Consider the function 𝑓(𝑥) = that is continuous on (−1, 1).
1−𝑥2
Observe that 𝑓((−1, 1)) = ℝ, so that the set (−1, 1) is bounded but its image
𝑓((−1, 1)) = ℝ is unbounded.
△
In light of these examples it is perhaps surprising that images of compact, i.e.
closed and bounded, sets under continuous functions are compact.
Theorem 4.22 If 𝑓 is continuous on a compact set 𝐾, then 𝑓(𝐾) is compact.
△

Proof. Let (𝑦𝑛 )∞

𝑛=1 be an arbitrary sequence in 𝑓(𝐾). To prove that 𝑓(𝐾) is compact,
we have to show that (𝑦𝑛 )∞ ∞
𝑛=1 has a subsequence (𝑦𝑛𝑘 )𝑘=1 that converges to some
𝑦 ∈ 𝑓(𝐾).
For each 𝑛 ∈ ℕ, let 𝑥𝑛 ∈ 𝐾 be such that 𝑦𝑛 = 𝑓(𝑥𝑛 ). Consider the sequence
(𝑥𝑛 )∞ ∞
𝑛=1 in 𝐾. Since 𝐾 is compact, it has a subsequence (𝑥𝑛𝑘 )𝑘=1 that converges
to some 𝑥 ∈ 𝐾. Since 𝑓 is continuous on 𝐾 and, in particular, at 𝑥, we can
apply Theorem 4.17. We conclude that (𝑓(𝑥𝑛𝑘 ))∞ 𝑘=1 converges to 𝑦 = 𝑓(𝑥) ∈ 𝑓(𝐾).
□

One consequence of this fact is the following well-known statement.

Theorem 4.23 (The Extreme Value Theorem) If 𝑓 is continuous on a compact set 𝐾,
then 𝑓 attains its maximum value and its minimum value on 𝐾. In other words, there
exist 𝑢, 𝑣 ∈ 𝐾 such that 𝑓(𝑢) = min𝑥∈𝐾 𝑓(𝑥) and 𝑓(𝑣) = max𝑥∈𝐾 𝑓(𝑥).
△

Proof. By Theorem 4.22 𝑓(𝐾) is compact, i.e. closed and bounded. Let 𝑚 =
inf 𝑓(𝐾) and 𝑀 = sup 𝑓(𝐾). By Proposition 3.20 we have that 𝑚, 𝑀 ∈ 𝑓(𝐾) =
𝑓(𝐾). Thus, there are 𝑢, 𝑣 ∈ 𝐾 such that 𝑚 = 𝑓(𝑢), 𝑀 = 𝑓(𝑣).
□

Another well-known result is the following statement.

Theorem 4.24 (The Intermediate Value Theorem) Let 𝑓 be continuous on a closed
bounded interval [𝑎, 𝑏]. Let 𝑦 ∈ ℝ be a value between 𝑓(𝑎) and 𝑓(𝑏). Then there exists
𝑐 ∈ [𝑎, 𝑏] such that 𝑦 = 𝑓(𝑐).
△

Proof. If 𝑓(𝑎) = 𝑓(𝑏) then there is nothing to prove. So suppose that 𝑓(𝑎) ≠ 𝑓(𝑏).
Without loss if generality, we may assume that 𝑓(𝑎) < 𝑓(𝑏) (if this is not the
case, consider −𝑓 instead of 𝑓).

75
 Chapter 4 Limits of functions and continuity

𝑦
𝑓(𝑏)
𝑦

𝑓(𝑎)

𝑎 𝐵 𝐴 𝑐𝑏 𝑥

Figure 4.3: The proof of the Intermediate Value Theorem

Take 𝑦 ∈ ℝ with 𝑓(𝑎) < 𝑦 < 𝑓(𝑏). Put

𝐴 = {𝑥 ∈ [𝑎, 𝑏] ∶ 𝑓(𝑥) ≤ 𝑦},
𝐵 = {𝑥 ∈ [𝑎, 𝑏] ∶ 𝑓(𝑥) > 𝑦}.
We have 𝑎 ∈ 𝐴, 𝑏 ∈ 𝐵. See Figure 4.3 for an illustration.
By the Completeness Axiom (Axiom 1.8), the set 𝐴 has a least upper bound 𝑐.
Since [𝑎, 𝑏] is closed, 𝑐 ∈ [𝑎, 𝑏]. We will show that 𝑓(𝑐) = 𝑦. Take a sequence
(𝑥𝑛 )∞
𝑛=1 in 𝐴 with lim𝑛→∞ 𝑥𝑛 = 𝑐. Notice that 𝑓(𝑥𝑛 ) ≤ 𝑦 for all 𝑛 ∈ ℕ. Since 𝑓 is
continuous, 𝑓(𝑐) = lim𝑛→∞ 𝑓(𝑥𝑛 ) ≤ 𝑦.
On the other hand, 𝑐 < 𝑏, because 𝑦 < 𝑓(𝑏). Therefore, for 𝑛 large enough
1
we have 𝑐 + 𝑛 ∈ [𝑎, 𝑏]. Since 𝑐 is the least upper bound of 𝐴, we obtain that
1 1 1
𝑐+ 𝑛
∈ 𝐵, so that 𝑓 �𝑐 + 𝑛 � > 𝑦. But lim𝑛→∞ �𝑐 + 𝑛 � = 𝑐, and the continuity of 𝑓
1
yields 𝑓(𝑐) = lim𝑛→∞ 𝑓 �𝑐 + 𝑛 � ≥ 𝑦. The two inequalities 𝑓(𝑐) ≤ 𝑦 and 𝑓(𝑐) ≥ 𝑦
together give 𝑓(𝑐) = 𝑦 as desired.
□

Essentially, the Intermediate Value Theorem tell us that the image of an

interval under a continuous function is an interval.

4.3 Uniform Continuity

Recall that a function 𝑓 is continuous on a set 𝐴 ⊆ ℝ if
∀ 𝜀 > 0 ∀ 𝑎 ∈ 𝐴 ∃𝛿 > 0 ∀ 𝑥 ∈ 𝐴 ∶ |𝑥 − 𝑎| < 𝛿 ⟹ |𝑓(𝑥) − 𝑓(𝑎)| < 𝜀.

76
 Chapter 4 Limits of functions and continuity

𝑓(𝑎2 ) + 𝜀
𝑓(𝑎2 )
𝑓(𝑎2 ) − 𝜀

𝑓(𝑎1 ) + 𝜀
𝑓(𝑎1 )
𝑓(𝑎1 ) − 𝜀
𝑎1 𝑎2 𝑥
𝑎1 − 𝛿 1 𝑎1 + 𝛿 1 𝑎2 − 𝛿 2 𝑎2 + 𝛿 2

Figure 4.4: The function 𝑓(𝑥) = 𝑥2

In this definition, the number 𝛿 > 0 depends on both 𝜀 > 0 and the point 𝑎 ∈ 𝐴.
In general, a choice of 𝛿 > 0 for one point 𝑎 ∈ 𝐴 might not work for other points.
If it happens, however, that, given 𝜀 > 0, we can take 𝛿 > 0 that works for
all 𝑎 ∈ 𝐴, we say that 𝑓 is uniformly continuous on 𝐴. We will give a formal
definition after considering examples.
Example 4.25 Consider the function 𝑓(𝑥) = 2𝑥 + 3 on ℝ. Let 𝑎 ∈ ℝ. We have

|𝑓(𝑥) − 𝑓(𝑎)| = |(2𝑥 + 3) − (2𝑎 + 3)| = 2 ⋅ |𝑥 − 𝑎|.

𝜀
Given 𝜀 > 0, take 𝛿 = 2 . Then for any 𝑎 ∈ ℝ we have |𝑥−𝑎| < 𝛿 ⟹ |𝑓(𝑥)−𝑓(𝑎)| <
𝜀. The function 𝑓(𝑥) = 2𝑥 + 3 is absolutely continuous on ℝ.
△
Example 4.26 Consider the function 𝑓(𝑥) = 𝑥2 on ℝ. The graph of this function
is shown in Figure 4.4. We see that the graph of this function gets steeper for
larger 𝑥, and the value of 𝛿 > 0 depends on 𝑎. The function 𝑓(𝑥) = 𝑥2 is not
uniformly continuous on ℝ. We will give a formal proof of this fact below.
△
Definition 4.27 Let 𝑓 ∶ 𝐴 ∈ ℝ. We say that 𝑓 is uniformly continuous on 𝐴, if

∀ 𝜀 > 0 ∃𝛿 > 0 ∀ 𝑥, 𝑦 ∈ 𝐴 ∶ |𝑥 − 𝑦| < 𝛿 ⟹ |𝑓(𝑥) − 𝑓(𝑦)| < 𝜀.

77
 Chapter 4 Limits of functions and continuity

△
Of course, if 𝑓 is uniformly continuous on 𝐴, then it is also continuous on 𝐴.
The additional property given by the uniform continuity that now 𝛿 > 0 in the
definition of continuity does not depend on a point in 𝐴.
Let us build a negation of the formula in the definition of uniform continuity.
A function 𝑓 ∶ 𝐴 → ℝ is not uniformly continuous on 𝐴, if

∃𝜀 > 0 ∀ 𝛿 > 0 ∃𝑥, 𝑦 ∈ 𝐴 ∶ |𝑥 − 𝑦| < 𝛿 and |𝑓(𝑥) − 𝑓(𝑦)| ≥ 𝜀.

1
Taking 𝛿𝑛 = 𝑛 , 𝑛 ∈ ℕ in the above formula, we obtain two sequences of points
1
(𝑥𝑛 )∞ ∞
𝑛=1 and (𝑦𝑛 )𝑛=1 in 𝐴 such that |𝑥𝑛 − 𝑦𝑛 | < 𝑛
and |𝑓(𝑥𝑛 ) − 𝑓(𝑦𝑛 )| ≥ 𝜀. We arrive
at the following statement.
Proposition 4.28 A function 𝑓 ∶ 𝐴 → ℝ is not uniformly continuous on 𝐴 if and
only if there exists a number 𝜀 > 0 and a pair of sequences (𝑥𝑛 )∞ ∞
𝑛=1 , (𝑦𝑛 )𝑛=1 in 𝐴 such
that lim𝑛→∞ (𝑥𝑛 − 𝑦𝑛 ) = 0 but |𝑓(𝑥𝑛 ) − 𝑓(𝑦𝑛 )| > 𝜀 for all 𝑛 ∈ ℕ.
△
Example 4.29 This is a continuation of Example 4.26. We will now use Propo-
sition 4.28 to show that the function 𝑓(𝑥) = 𝑥2 is not uniformly continuous
on ℝ.
Consider the sequences 𝑥𝑛 = √𝑛 + 1, 𝑦𝑛 = √𝑛, 𝑛 ∈ ℕ. We have
1
lim (𝑥𝑛 − 𝑦𝑛 ) = lim �√𝑛 + 1 − √𝑛� = lim = 0,
𝑛→∞ 𝑛→∞ 𝑛→∞
√𝑛 + 1 + √𝑛
but
2 2
𝑓(𝑥𝑛 ) − 𝑓(𝑦𝑛 ) = �√𝑛 + 1� − �√𝑛� = (𝑛 + 1) − 𝑛 = 1.

This shows that 𝑓(𝑥) = 𝑥2 is not uniformly continuous on ℝ.

△
The sum of two uniformly continuous functions is uniformly continuous
(Exercise). Interestingly, the product of two uniformly continuous functions
need not be uniformly continuous. For example, it is easy to see that the function
𝑔(𝑥) = 𝑥 is uniformly continuous on ℝ. However, as we have seen above, the
function 𝑓(𝑥) = 𝑥2 = 𝑔(𝑥) ⋅ 𝑔(𝑥) is not uniformly continuous on ℝ.
Again, compact sets play a special role.
Theorem 4.30 If 𝑓 is continuous on a compact set 𝐾, then 𝑓 is uniformly continuous
on 𝐾.
△

78
 Chapter 4 Limits of functions and continuity

Proof. We will prove this statement by contradiction. Suppose that 𝑓 is continu-

ous, but not uniformly continuous on 𝐾. By Proposition 4.28 there exists 𝜀 > 0
and a pair of sequences (𝑥𝑛 )∞ ∞
𝑛=1 , (𝑦𝑛 )𝑛=1 in 𝐾 such that lim𝑛→∞ (𝑥𝑛 − 𝑦𝑛 ) = 0 but
|𝑓(𝑥𝑛 ) − 𝑓(𝑦𝑛 )| > 𝜀 for all 𝑛 ∈ ℕ.
Since 𝐾 is compact, the sequence (𝑥𝑛 )∞ ∞
𝑛=1 has a subsequence (𝑥𝑛𝑘 )𝑘=1 that
converges to some 𝑥 ∈ 𝐾. We have

lim 𝑦𝑛𝑘 = lim (𝑦𝑛𝑘 − 𝑥𝑛𝑘 + 𝑥𝑛𝑘 ) = lim (𝑦𝑛𝑘 − 𝑥𝑛𝑘 ) + lim 𝑥𝑛𝑘 = 0 + 𝑥 = 𝑥.
𝑘→∞ 𝑘→∞ 𝑘→∞ 𝑘→∞

Thus, the sequence (𝑦𝑛𝑘 )∞

𝑘=1 also converges to 𝑥.
Since 𝑓 is continuous at 𝑥, we have by Theorem 4.17

lim (𝑓(𝑥𝑛𝑘 ) − 𝑓(𝑦𝑛𝑘 )) = lim 𝑓(𝑥𝑛𝑘 ) − lim 𝑓(𝑦𝑛𝑘 ) = 𝑓(𝑥) − 𝑓(𝑥) = 0,

𝑘→∞ 𝑘→∞ 𝑘→∞

but this is impossible since by assumption |𝑓(𝑥𝑛𝑘 ) − 𝑓(𝑦𝑛𝑘 )| > 𝜀 for all 𝑘 ∈ ℕ. This
contradiction shows that 𝑓 is uniformly continuous on 𝐾.
□

79
Chapter 5
Derivatives

You are familiar with derivatives from your Calculus course. In this short
chapter we will give a definition of the derivative and present several important
theorems that involve derivatives.
In this chapter we will denote by 𝐼 = (𝑎, 𝑏) an open interval. Here 𝑎, 𝑏 ∈ ℝ
and 𝑎 < 𝑏.
Definition 5.1 Let 𝑓 ∶ 𝐼 → ℝ. We will say that 𝑓 is differentiable at 𝑎 ∈ 𝐼 if the
limit
𝑓(𝑥) − 𝑓(𝑎)
lim
𝑥→𝑎 𝑥−𝑎
exists. If this is the case, we call the value
𝑓(𝑥) − 𝑓(𝑎)
𝑓′ (𝑎) = lim
𝑥→𝑎 𝑥−𝑎
the derivative of the function 𝑓 at the point 𝑎. If 𝑓 is differentiable at every point
𝑎 ∈ 𝐼, we say that 𝑓 is differentiable on 𝐼.
△
Sums, products, quotients (under sensible conditions) and compositions of
differentiable functions are differentiable. For the derivatives we have rules
such as the sum rule, the product rule, the quotient rule, the chain rule, etc. We
will not repeat them here.
Also the following results must be known to you.
Theorem 5.2 Let 𝑓 ∶ 𝐼 → ℝ and 𝑎 ∈ 𝐼. If 𝑓 is differentiable at 𝑎, then 𝑓 is continuous
at 𝑎.
△

Proof. Exercise.
□

80
 Chapter 5 Derivatives

The converse of this statement is not true. For example, the function 𝑓(𝑥) = |𝑥|
is continuous but not differentiable at 𝑥 = 0.
Definition 5.3 Let 𝑓 ∶ 𝐼 → ℝ.

(1) We say that 𝑓 has a local maximum at a point 𝑐 ∈ 𝐼 if

∃𝛿 > 0 ∀ 𝑥 ∈ 𝐼 ∶ |𝑥 − 𝑐| < 𝛿 ⟹ 𝑓(𝑥) ≤ 𝑓(𝑐).

(2 ) Similarly, we say that 𝑓 has a local minimum at a point 𝑐 ∈ 𝐼 if

∃𝛿 > 0 ∀ 𝑥 ∈ 𝐼 ∶ |𝑥 − 𝑐| < 𝛿 ⟹ 𝑓(𝑥) ≥ 𝑓(𝑐).

(3 ) We say that 𝑓 has a local extremum at a point 𝑐 ∈ 𝐼 if 𝑓 has a local maximum

or a local minimum at 𝑐.

△
Theorem 5.4 (Fermat’s Theorem) Let 𝑓 ∶ 𝐼 → ℝ be differentiable on a open interval 𝐼.
If 𝑓 has a local extremum at a point 𝑐 ∈ 𝐼, then 𝑓′ (𝑐) = 0.
△

Proof. Suppose that 𝑓′ (𝑐) < 0. We will show that 𝑓 cannot have a local extremum
at 𝑐.
𝑓(𝑥)−𝑓(𝑐)
By Lemma 4.6, 𝑓′ (𝑐) = lim𝑥→𝑐 𝑥−𝑐 < 0 implies that there exists 𝛿 > 0 such
that
𝑓(𝑥) − 𝑓(𝑐)
<0
𝑥−𝑐
for all 𝑥 ∈ 𝐼 with 0 < |𝑥 − 𝑐| < 𝛿. If 𝑐 < 𝑥 < 𝑐 + 𝛿, then 𝑥 − 𝑐 > 0, and thus
𝑓(𝑥) − 𝑓(𝑐) < 0, i.e. 𝑓(𝑥) < 𝑓(𝑐). This means that 𝑐 cannot be a local minimum.
On the other hand, if 𝑐 − 𝛿 < 𝑥 < 𝑐, then 𝑥 − 𝑐 < 0, and thus 𝑓(𝑥) − 𝑓(𝑐) > 0, i.e.
𝑓(𝑥) > 𝑓(𝑐). Thus, 𝑐 cannot be a local maximum either.
The case when 𝑓′ (𝑐) > 0 can be considered similarly.
□

Notice that the converse of Fermat’s Theorem is not true. For example, for
the function 𝑓(𝑥) = 𝑥3 we have 𝑓′ (0) = 0, but 𝑥 = 0 is neither a local maximum
nor a local minimum of 𝑓.
Theorem 5.5 (Rolle’s Theorem) Let 𝑓 ∶ [𝑎, 𝑏] → ℝ be continuous on [𝑎, 𝑏] and
differentiable on (𝑎, 𝑏). If 𝑓(𝑎) = 𝑓(𝑏), then there exists 𝑐 ∈ (𝑎, 𝑏) such that 𝑓′ (𝑐) = 0.
△

81
 Chapter 5 Derivatives

𝑦 𝑓′ (𝑐) = 0

𝑓(𝑎) = 𝑓(𝑏)

𝑎 𝑐 𝑏 𝑥

Figure 5.1: Rolle’s Theorem

For an illustration, see Figure 5.1.

Proof. If 𝑓 is constant on [𝑎, 𝑏], that 𝑓′ (𝑥) = 0 for all 𝑥 ∈ (𝑎, 𝑏), and we are done.
Therefore assume that 𝑓 is not constant. By the Extreme Value Theorem
(Theorem 4.23), 𝑓 attains its minimum and its maximum on [𝑎, 𝑏]. Notice that
max𝑥∈[𝑎,𝑏] 𝑓(𝑥) > min𝑥∈[𝑎,𝑏] 𝑓(𝑥) as 𝑓 is not constant. This means that 𝑓(𝑎) = 𝑓(𝑏)
is different from at least one of the values max𝑥∈[𝑎,𝑏] 𝑓(𝑥), min𝑥∈[𝑎,𝑏] 𝑓(𝑥). We
conclude that 𝑓 has at least one local extremum 𝑐 ∈ (𝑎, 𝑏). By Fermat’s Theorem
(Theorem 5.4), 𝑓′ (𝑐) = 0 as desired.
□

Theorem 5.6 (The Mean Value Theorem) Let 𝑓 ∶ [𝑎, 𝑏] → ℝ be continuous on [𝑎, 𝑏]
and differentiable on (𝑎, 𝑏). There exists 𝑐 ∈ (𝑎, 𝑏) such that
𝑓(𝑏) − 𝑓(𝑎)
𝑓′ (𝑐) = .
𝑏−𝑎
△
Geometrically this theorem says that there is a point 𝑐 ∈ (𝑎, 𝑏) such that the
tangent line to the graph of 𝑓 at the point (𝑐, 𝑓(𝑐)) is parallel to the secant line
through the points (𝑎, 𝑓(𝑎)) and (𝑏, 𝑓(𝑏)). For an illustration, see Figure 5.2.

Proof. The equation of the secant line through the points (𝑎, 𝑓(𝑎)) and (𝑏, 𝑓(𝑏)) is
𝑓(𝑏) − 𝑓(𝑎)
𝑦 = 𝑓(𝑎) + (𝑥 − 𝑎).
𝑏−𝑎

82
 Chapter 5 Derivatives

𝑎 𝑐 𝑏 𝑥

Figure 5.2: Mean Value Theorem

Consider the function

𝑓(𝑏) − 𝑓(𝑎)
𝐹(𝑥) = 𝑓(𝑥) − �𝑓(𝑎) + (𝑥 − 𝑎)� .
𝑏−𝑎

The function 𝐹 is continuous on [𝑎, 𝑏] and differentiable on (𝑎, 𝑏) with

𝑓(𝑏) − 𝑓(𝑎)
𝐹′ (𝑥) = 𝑓′ (𝑥) − .
𝑏−𝑎
Moreover, 𝐹(𝑎) = 𝐹(𝑏) = 0. By Rolle’s Theorem (Theorem 5.5) there exists a
point 𝑐 ∈ (𝑎, 𝑏) with
𝑓(𝑏) − 𝑓(𝑎)
𝐹′ (𝑐) = 𝑓′ (𝑐) − = 0.
𝑏−𝑎
We obtain that
𝑓(𝑏) − 𝑓(𝑎)
𝑓′ (𝑐) =
𝑏−𝑎
as desired.
□

An important application of the Mean Value Theorem is the statement that if

𝑓′ (𝑥)= 0 for all 𝑥 ∈ (𝑎, 𝑏), then 𝑓 is constant on [𝑎, 𝑏]. This result is frequently
used when integrating (remember the integration constant) or when solving
differential equations. We will not go into detail here.

83
 Chapter 5 Derivatives

Instead, we establish a more general version of the Mean Value Theorem.

Assume that we have two functions 𝑓 and 𝑔 that are continuous on [𝑎, 𝑏] and
differentiable on (𝑎, 𝑏). Applying the Mean Value Theorem to each function
separately, we see that there exist two points 𝑐, 𝑑 ∈ (𝑎, 𝑏) with
𝑓(𝑏) − 𝑓(𝑎) 𝑔(𝑏) − 𝑔(𝑎)
𝑓′ (𝑐) = , 𝑔′ (𝑑) = .
𝑏−𝑎 𝑏−𝑎
If we further assume that 𝑓(𝑏) ≠ 𝑓(𝑎), we obtain that
𝑔′ (𝑑) 𝑔(𝑏) − 𝑔(𝑎)
′
= .
𝑓 (𝑐) 𝑓(𝑏) − 𝑓(𝑎)
The statement we are going to prove says that we can find a single point 𝑐 ∈ (𝑎, 𝑏)
such that
𝑔′ (𝑐) 𝑔(𝑏) − 𝑔(𝑎)
′
= .
𝑓 (𝑐) 𝑓(𝑏) − 𝑓(𝑎)
This statement has also a geometric interpretation. Consider in the plane the
parametric curve with the equation 𝑥 = 𝑓(𝑡), 𝑦 = 𝑔(𝑡), 𝑡 ∈ [𝑎, 𝑏]. The slope of
𝑔(𝑏)−𝑔(𝑎)
the secant line joining the points 𝐴(𝑓(𝑎), 𝑔(𝑎)) and 𝐵(𝑓(𝑏), 𝑔(𝑏)) is 𝑓(𝑏)−𝑓(𝑎)
. The
theorem says that there is a point 𝐶(𝑓(𝑐), 𝑔(𝑐)), 𝑐 ∈ (𝑎, 𝑏), such that the tangent
line to the curve at the point 𝐶 is parallel to the secant line through 𝐴 and 𝐵.
For an illustration, see Figure 5.3.
Theorem 5.7 (The Generalized Mean Value Theorem) Let 𝑓, 𝑔 ∶ [𝑎, 𝑏] → ℝ be
continuous on [𝑎, 𝑏] and differentiable on (𝑎, 𝑏). There exists 𝑐 ∈ (𝑎, 𝑏) such that

(𝑓(𝑏) − 𝑓(𝑎))𝑔′ (𝑐) = 𝑓′ (𝑐)(𝑔(𝑏) − 𝑔(𝑎)).

Proof. Consider the function

𝐺(𝑥) = ((𝑓(𝑏) − 𝑓(𝑎))𝑔(𝑥) − (𝑔(𝑏) − 𝑔(𝑎))𝑓(𝑥).

The function 𝐺 is continuous on [𝑎, 𝑏] and differentiable on (𝑎, 𝑏) with

𝐺′ (𝑥) = ((𝑓(𝑏) − 𝑓(𝑎))𝑔′ (𝑥) − (𝑔(𝑏) − 𝑔(𝑎))𝑓′ (𝑥).

Moreover, 𝐺(𝑎) = 𝐺(𝑏) = 𝑓(𝑏)𝑔(𝑎) − 𝑓(𝑎)𝑔(𝑏). By Rolle’s Theorem (Theorem 5.5)

there exists a point 𝑐 ∈ (𝑎, 𝑏) with

𝐺′ (𝑐) = ((𝑓(𝑏) − 𝑓(𝑎))𝑔′ (𝑐) − (𝑔(𝑏) − 𝑔(𝑎))𝑓′ (𝑐) = 0,

84
 Chapter 5 Derivatives

Figure 5.3: Generalized Mean Value Theorem

i.e.,
(𝑓(𝑏) − 𝑓(𝑎))𝑔′ (𝑐) = 𝑓′ (𝑐)(𝑔(𝑏) − 𝑔(𝑎)).
□

As an application of Theorem 5.7 we will prove the well-known l’Hospital’s

0 ∞
Rule. We will consider the 0 case; other variants such as the ∞ case and the
limits at ∞ and −∞ can be deduced from it.
0
Theorem 5.8 (l’Hospital’s Rule, case) Let 𝑓, 𝑔 ∶ 𝐼 → ℝ be differentiable on 𝐼.
0
Assume that for 𝑐 ∈ 𝐼 we have 𝑓(𝑐) = 𝑔(𝑐) = 0, and that 𝑓′ (𝑥) ≠ 0 for all 𝑥 ∈ 𝐼 ⧵ {𝑐}. If
𝑔′ (𝑥)
lim𝑥→𝑐 𝑓′ (𝑥)
exists, then
𝑔(𝑥) 𝑔′ (𝑥)
lim = lim ′ .
𝑥→𝑐 𝑓(𝑥) 𝑥→𝑐 𝑓 (𝑥)
△

Proof. Let 𝑥 ∈ 𝐼 ⧵ {𝑐}. By applying the Generalized Mean Value Theorem (Theo-
rem 5.7) to the functions 𝑓 and 𝑔 at the points 𝑐 and 𝑥, we conclude that there
is a point 𝑐𝑥 that lies strictly between 𝑥 and 𝑐 such that

𝑔′ (𝑐𝑥 ) 𝑔(𝑥) − 𝑔(𝑐) 𝑔(𝑥)

′
= = .
𝑓 (𝑐𝑥 ) 𝑓(𝑥) − 𝑓(𝑐) 𝑓(𝑥)

85
 Chapter 5 Derivatives

The point 𝑐𝑥 depends on 𝑥. Since it lies between 𝑐 and 𝑥, we have lim𝑥→𝑐 𝑐𝑥 = 𝑐.

It follows that
𝑔′ (𝑐𝑥 ) 𝑔′ (𝑥)
lim ′ = lim ′ .
𝑥→𝑐 𝑓 (𝑐𝑥 ) 𝑥→𝑐 𝑓 (𝑥)

𝑔(𝑥)
Thus, the limit lim𝑥→𝑐 𝑓(𝑥)
exists and

𝑔(𝑥) 𝑔′ (𝑥)
lim = lim ′ .
𝑥→𝑐 𝑓(𝑥) 𝑥→𝑐 𝑓 (𝑥)
□

86
Chapter 6
Sequences of functions

6.1 Pointwise and uniform convergence

In this chapter we will consider sequences and series of functions. This section
is devoted to two types of convergence of sequences of functions.
Definition 6.1 Let 𝐴 ⊆ ℝ be a nonempty set. For each 𝑛 ∈ ℕ, let 𝑓𝑛 ∶ 𝐴 → ℝ.
We say that the sequence (𝑓𝑛 )∞
𝑛=1 converges pointwise on 𝐴 to a function 𝑓 ∶ 𝐴 → ℝ
if
lim 𝑓𝑛 (𝑥) = 𝑓(𝑥) for all 𝑥 ∈ 𝐴.
𝑛→∞
In this case we say that 𝑓 is the pointwise limit of the sequence (𝑓𝑛 )∞
𝑛=1 .
△
Example 6.2 Consider the sequence of functions 𝑓𝑛 (𝑥) = 𝑥𝑛 , 𝑛 ∈ ℕ, defined
on ℝ.
For a fixed 𝑥 ∈ ℝ with |𝑥| > 1, the sequence (𝑥𝑛 )∞ 𝑛=1 is unbounded, and
thus lim𝑛→∞ 𝑥𝑛 does not exist. For 𝑥 = −1, this is the sequence (−1, 1, −1, 1, …),
and again lim𝑛→∞ 𝑥𝑛 does not exist. If −1 < 𝑥 < 1, then lim𝑛→∞ 𝑥𝑛 = 0. Fi-
nally, if 𝑥 = 1, then the sequence is the constant sequence (1, 1, 1, 1, …), and
lim𝑛→∞ 𝑥𝑛 = 1.
Thus, the sequence (𝑓𝑛 )∞𝑛=1 converge pointwise on (−1, 1], and
⎧
⎪
⎪
⎨0, −1 < 𝑥 < 1,
lim 𝑓𝑛 (𝑥) = ⎪
⎪
𝑛→∞ ⎩1, 𝑥 = 1.

Notice that the limit function is discontinuous on (−1, 1].

△

87
 Chapter 6 Sequences of functions

Example 6.3 For 𝑛 ∈ ℕ and 𝑥 ∈ ℝ, let

⎧ 𝑛
𝑛−1⎪
⎪ 1−𝑥
⎨ 1−𝑥 , 𝑥 ≠ 1,
𝑓𝑛 (𝑥) = � 𝑥𝑘 = ⎪
⎪
𝑘=0
⎩𝑛, 𝑥 = 1.
1
The sequence (𝑓𝑛 )∞
𝑛=1 converge pointwise to 𝑓(𝑥) = 1−𝑥
if 𝑥 ∈ (−1, 1) and diverges
for 𝑥 ≤ −1 or 𝑥 ≥ 1.
△
1
Example 6.4 For 𝑛 ∈ ℕ and 𝑥 ≥ 0, let 𝑓𝑛 (𝑥) = 𝑥 . If 𝑥 > 0, then lim𝑛→∞ 𝑓𝑛 (𝑥) =
𝑛

1, and lim𝑛→∞ 𝑓𝑛 (0) = 0. Thus, the pointwise limit of (𝑓𝑛 )∞

𝑛=1 on [0, ∞) is
⎧
⎪
⎪
⎨0, 𝑥 = 0,
lim 𝑓𝑛 (𝑥) = ⎪
⎪
𝑛→∞ ⎩1, 𝑥 > 0.
△
These examples show, in particular, that the pointwise limit of a sequence of
continuous functions need not be continuous.
Notice that a sequence (𝑓𝑛 )∞
𝑛=1 converges pointwise on 𝐴 to a function 𝑓, if

∀ 𝜀 > 0 ∀ 𝑥 ∈ 𝐴 ∃𝑁 ∈ ℕ ∀ 𝑛 ≥ 𝑁 ∶ |𝑓𝑛 (𝑥) − 𝑓(𝑥)| < 𝜀.

The number 𝑁 here depends on 𝜀 > 0 and on 𝑥 ∈ 𝐴.
A stronger condition is obtained if we require that 𝑁 depends only on 𝜀 > 0
and does not depend on 𝑥 ∈ 𝐴. Recall that we have used a similar idea to define
uniform continuity in Section 4.3.
Definition 6.5 Let 𝐴 ⊆ ℝ be a nonempty set. For each 𝑛 ∈ ℕ, let 𝑓𝑛 ∶ 𝐴 → ℝ.
We say that the sequence (𝑓𝑛 )∞
𝑛=1 converges uniformly on 𝐴 to a function 𝑓 ∶ 𝐴 → ℝ
if
∀ 𝜀 > 0 ∃𝑁 ∈ ℕ ∀ 𝑥 ∈ 𝐴 ∀ 𝑛 ≥ 𝑁 ∶ |𝑓𝑛 (𝑥) − 𝑓(𝑥)| < 𝜀.
In this case we say that 𝑓 is the uniform limit of the sequence (𝑓𝑛 )∞
𝑛=1 .
△
Uniform convergence implies pointwise convergence. If the uniform limit
exists, then it is also the pointwise limit. The converse is not true, in general.
1−𝑥𝑛
Example 6.6 This is a continuation of Example 6.3. Consider 𝑓𝑛 (𝑥) = 1−𝑥 ,
𝑥 ∈ (−1, 1), 𝑛 ∈ ℕ. For any 𝑟 ∈ (0, 1), the sequence (𝑓𝑛 )∞𝑛=1 converges uniformly
1
on [−𝑟, 𝑟] to 𝑓(𝑥) = 1−𝑥 . Indeed, for 𝑥 ∈ [−𝑟, 𝑟] we have
1 − 𝑥𝑛 1 |𝑥|𝑛 𝑟𝑛
|𝑓𝑛 (𝑥) − 𝑓(𝑥)| = � − �= ≤ .
1−𝑥 1−𝑥 1−𝑥 1−𝑟

88
 Chapter 6 Sequences of functions

𝑟𝑛
We know that 1−𝑟
→ 0 as 𝑛 → ∞. Thus, given 𝜀 > 0, there exists 𝑁 ∈ ℕ such
𝑟𝑛
that 1−𝑟
< 𝜀 for all 𝑛 ≥ 𝑁. For such 𝑛 and for all 𝑥 ∈ [−𝑟, 𝑟] we obtain

|𝑓𝑛 (𝑥) − 𝑓(𝑥)| < 𝜀.

It follows that the convergence is uniform on [−𝑟, 𝑟].

Notice that 𝑓 is uniformly continuous on [−𝑟, 𝑟].
△
Theorem 6.7 Let (𝑓𝑛 )∞
be a sequence of continuous functions on 𝐴 ⊆ ℝ that
𝑛=1
converges uniformly on 𝐴 to a function 𝑓. Then 𝑓 is continuous.
△

Proof. Let 𝑎 ∈ 𝐴. We will show that 𝑓 is continuous at 𝑎.

Given 𝜀 > 0, by the uniform convergence there exists 𝑁 ∈ ℕ such that
𝜀
|𝑓𝑁 (𝑡) − 𝑓(𝑡)| < for all 𝑡 ∈ 𝐴.
3
Since 𝑓𝑁 is continuous at 𝑎, there is 𝛿 > 0 such that for 𝑥 ∈ 𝐴
𝜀
|𝑥 − 𝑎| < 𝛿 ⟹ |𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑎)| < .
3
Now take 𝑥 ∈ 𝐴 with |𝑥 − 𝑎| < 𝛿. For such 𝑥 we have
𝜀 𝜀 𝜀
|𝑓(𝑥) − 𝑓(𝑎)| ≤ |𝑓(𝑥) − 𝑓𝑁 (𝑥)| + |𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑎)| + |𝑓𝑁 (𝑎) − 𝑓(𝑎)| < + + = 𝜀.
3 3 3
Thus, 𝑓 is continuous at 𝑎.
□

Example 6.8 This is a continuation of Example 6.4. We have seen that the
1 ∞
sequence �𝑥 � 𝑛 converges pointwise on [0, ∞) to the function
𝑛=1
⎧
⎪
⎪
⎨0, 𝑥 = 0,
𝑓(𝑥) = ⎪
⎪
⎩1, 𝑥 > 0.

1
Each of the functions 𝑓𝑛 (𝑥) = 𝑥 𝑛 is continuous on [0, ∞), but the function
𝑓(𝑥) = lim𝑛→∞ 𝑓𝑛 (𝑥) is not. It follows that the convergence is not uniform.
△

89
 Chapter 6 Sequences of functions

𝑦
1

−1/𝑛

1/𝑛 𝑥

−1

Figure 6.1: A graph of the function 𝑓𝑛

Example 6.9 For 𝑛 ∈ ℕ and 𝑥 ∈ ℝ we define

⎧ 1
⎪
⎪−1, 𝑥 ≤ − 𝑛 ,
⎪
⎪
⎨ 1 1
𝑓𝑛 (𝑥) = ⎪
⎪𝑛𝑥, − 𝑛 < 𝑥 < 𝑛 ,
⎪
⎪ 1
⎩1, 𝑥≥ . 𝑛

A graph of this function is shown in Figure 6.1

The pointwise limit of the sequence (𝑓𝑛 )∞
𝑛=1 is
⎧
⎪
⎪−1, 𝑥 < 0,
⎪
⎪
⎨
𝑓(𝑥) = ⎪
⎪0, 𝑥 = 0,
⎪
⎪
⎩1, 𝑥 > 0.

As in the previous example, each function 𝑓𝑛 is continuous on ℝ, but the limit

function 𝑓 is not. We conclude that the convergence is not uniform.
△
Recall the Cauchy Criterion (Theorem 2.45) for the convergence of sequences
in ℝ. There is a corresponding variant for the uniform convergence of sequences
of functions. Prove it as an exercise!
Theorem 6.10 (The Cauchy Criterion for Uniform Convergence) A sequence of
functions (𝑓𝑛 )∞
𝑛=1 converges uniformly on a set 𝐴 ⊆ ℝ if and only if

∀ 𝜀 > 0 ∃𝑁 ∈ ℕ ∀ 𝑛, 𝑚 ≥ 𝑁 ∀ 𝑥 ∈ 𝐴 ∶ |𝑓𝑛 (𝑥) − 𝑓𝑚 (𝑥)| < 𝜀.

△
Now suppose that the sequence (𝑓𝑛 )∞converges pointwise to a function 𝑓
𝑛=1
on a certain set 𝐴, and that each function 𝑓𝑛 is differentiable. In general it is not
true that the limit function 𝑓 is differentiable, or that the derivative of the limit
function 𝑓 is equal to the limit of 𝑓𝑛′ . This is not surprising: recall that even the

90
 Chapter 6 Sequences of functions

property of continuity is not preserved under pointwise limits. However, the

desired properties are valid in some situations. The theorem below states the
conditions that guarantee that one can interchange the differentiation and the
limit.
Theorem 6.11 (The Differential Limit Theorem) Let (𝑓𝑛 )∞
𝑛=1 be a sequence of func-
tions that are defined and differentiable on [𝑎, 𝑏]. Suppose that

(1) (𝑓𝑛 )∞
𝑛=1 converges pointwise on [𝑎, 𝑏] to a function 𝑓 ∶ [𝑎, 𝑏] → ℝ,

(2 ) (𝑓𝑛′ )∞
𝑛=1 converges uniformly on [𝑎, 𝑏] to a function 𝑔 ∶ [𝑎, 𝑏] → ℝ.

Then 𝑓 is differentiable on [𝑎, 𝑏] and 𝑓′ = 𝑔. In particular,

𝑓′ (𝑥) = lim 𝑓𝑛′ (𝑥), 𝑥 ∈ [𝑎, 𝑏].

𝑛→∞

Proof. Let 𝑐 ∈ [𝑎, 𝑏] be an arbitrary point. We want to show that 𝑓′ (𝑐) exists and
that 𝑓′ (𝑐) = 𝑔(𝑐). In other words, we want to show that

𝑓(𝑥) − 𝑓(𝑐)
lim = 𝑔(𝑐).
𝑥→𝑐 𝑥−𝑐
Let 𝜀 > 0. We have to find 𝛿 > 0 such that for any 𝑥 ∈ [𝑎, 𝑏]

𝑓(𝑥) − 𝑓(𝑐)
0 < |𝑥 − 𝑐| < 𝛿 ⟹ � − 𝑔(𝑐)� < 𝜀.
𝑥−𝑐
Since lim𝑛→∞ 𝑓𝑛′ (𝑐) = 𝑔(𝑐), there is 𝑁1 ∈ ℕ such that
𝜀
|𝑓𝑛′ (𝑐) − 𝑔(𝑐)| < for all 𝑛 ≥ 𝑁1 .
3
Since (𝑓𝑛′ ) converges uniformly on [𝑎, 𝑏], by Theorem 6.10 there is 𝑁2 ∈ ℕ such
that
′ (𝑥)| < 𝜀
|𝑓𝑛′ (𝑥) − 𝑓𝑚 for all 𝑛, 𝑚 ≥ 𝑁2 and 𝑥 ∈ [𝑎, 𝑏].
3
Put 𝑁 = max{𝑁1 , 𝑁2 }. We have

′ 𝜀
|𝑓𝑁 (𝑐) − 𝑔(𝑐)| < ,
3
′ ′ (𝑥)| < 𝜀
|𝑓𝑁 (𝑥) − 𝑓𝑚 for all 𝑚 ≥ 𝑁 and 𝑥 ∈ [𝑎, 𝑏].
3

91
 Chapter 6 Sequences of functions

The function 𝑓𝑁 is differentiable at 𝑐, and thus there is 𝛿 > 0 such that for
𝑥 ∈ [𝑎, 𝑏]
𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑐) ′ 𝜀
0 < |𝑥 − 𝑐| < 𝛿 ⟹ � − 𝑓𝑁 (𝑐)� < .
𝑥−𝑐 3
We will now show that this 𝛿 > 0 satisfies our requirements.
Let 𝑥 ∈ [𝑎, 𝑏] with 0 < |𝑥 − 𝑐| < 𝛿, and let 𝑚 ≥ 𝑁. By the Mean Value Theorem
(Theorem 5.6), there is a point 𝑧 lying between 𝑥 and 𝑐 such that

(𝑓𝑚 (𝑥) − 𝑓𝑁 (𝑥)) − (𝑓𝑚 (𝑐) − 𝑓𝑁 (𝑐)) ′ (𝑧) − 𝑓 ′ (𝑧).

= 𝑓𝑚 𝑁
𝑥−𝑐
𝜀
′ (𝑧) − 𝑓 ′ (𝑧)| < , so that
But by the above |𝑓𝑚 𝑁 3

(𝑓𝑚 (𝑥) − 𝑓𝑁 (𝑥)) − (𝑓𝑚 (𝑐) − 𝑓𝑁 (𝑐)) 𝑓𝑚 (𝑥) − 𝑓𝑚 (𝑐) 𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑐) 𝜀
� �=� − �< .
𝑥−𝑐 𝑥−𝑐 𝑥−𝑐 3
Taking the limit as 𝑚 → ∞, we obtain

𝑓(𝑥) − 𝑓(𝑐) 𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑐) 𝜀

� − �≤ .
𝑥−𝑐 𝑥−𝑐 3
Finally, the inequalities above and the triangle inequality give

𝑓(𝑥) − 𝑓(𝑐) 𝑓(𝑥) − 𝑓(𝑐) 𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑐)

� − 𝑔(𝑐)� ≤ � − �
𝑥−𝑐 𝑥−𝑐 𝑥−𝑐
𝑓𝑁 (𝑥) − 𝑓𝑁 (𝑐) ′ ′
+� − 𝑓𝑁 (𝑐)� + |𝑓𝑁 (𝑐) − 𝑔(𝑐)|
𝑥−𝑐
𝜀 𝜀 𝜀
< + + =𝜀
3 3 3
for all 𝑥 ∈ [𝑎, 𝑏] with 0 < |𝑥 − 𝑐| < 𝛿 as desired.
□

6.2 Series of functions

Definition 6.12 Let (𝑓𝑘 )∞
𝑘=1 be a series of functions 𝑓𝑘 ∶ 𝐴 → ℝ, and let 𝑓 ∶ 𝐴 →
∞
ℝ. We say that the series ∑𝑘=1 𝑓𝑘

𝑛 ∞
(1) converges pointwise on 𝐴 to 𝑓 if the sequence of partial sums �∑𝑘=1 𝑓𝑘 (𝑥)�
𝑛=1
converges to 𝑓(𝑥) for each 𝑥 ∈ 𝐴,

92
 Chapter 6 Sequences of functions

𝑛 ∞
(2 ) converges uniformly on 𝐴 to 𝑓 if the sequence of partial sums �∑𝑘=1 𝑓𝑘 �
𝑛=1
converges to 𝑓 uniformly on 𝐴.

△
As in the case of series of real numbers, we say that a series ∑ ∞ 𝑓𝑘 converges
𝑘=1
∞
absolutely if the series ∑𝑘=1 |𝑓𝑘 | converges.
∞ ∞
Example 6.13 Let 𝑓𝑘 (𝑥) = 𝑥𝑘 , 𝑘 ∈ ℕ∪{0}, 𝑥 ∈ ℝ. The series ∑𝑘=0 𝑓𝑘 (𝑥) = ∑𝑘=0 𝑥𝑘
does not converge at the whole of ℝ because there exist values of 𝑥 ∈ ℝ, for
∞
which ∑𝑘=0 𝑥𝑘 diverges.
∞
The series ∑𝑘=0 𝑥𝑘 converges pointwise on the open interval (−1, 1) to 𝑓(𝑥) =
1 ∞
1−𝑥
. Moreover, ∑𝑘=0 𝑥𝑘 converges absolutely and uniformly on any interval
∞
[−𝑟, 𝑟], where 𝑟 ∈ (0, 1). We will show this in a moment. Notice that ∑𝑘=0 𝑥𝑘 does
not converge on [−1, 1] as it does not converge for 𝑥 = −1 and 𝑥 = 1.
△
∞
Note that a series series ∑𝑘=1 𝑓𝑘 converges uniformly on a set 𝐴 to a function
𝑓 if and only if
𝑛
∀ 𝜀 > 0 ∃𝑁 ∈ ℕ ∀ 𝑥 ∈ 𝐴 ∀ 𝑛 ≥ 𝑁 ∶ �𝑓(𝑥) − � 𝑓𝑘 (𝑥)� < 𝜀.
𝑘=1

The next statement contains a very useful test for the absolute and uniform
convergence of a series.
Theorem 6.14 (The Weierstrass M-Test) Suppose that for each 𝑘 ∈ ℕ there is a
number 𝑀𝑘 ≥ 0 such that

|𝑓𝑘 (𝑥)| ≤ 𝑀𝑘 for all 𝑥 ∈ 𝐴.

∞ ∞
If ∑𝑘=1 𝑀𝑘 converges, then ∑𝑘=1 𝑓𝑘 converges absolutely and uniformly on 𝐴.
△
∞
Proof. By the Comparison Test (Theorem 2.30), the series ∑𝑘=1 |𝑓𝑘 (𝑥)| converges
∞
for each 𝑥 ∈ 𝐴. Thus, ∑𝑘=1 𝑓𝑘 (𝑥) converges absolutely for every 𝑥 ∈ 𝐴. It remains
to prove that the convergence is uniform.
∞
Put 𝑓(𝑥) = ∑𝑘=1 𝑓𝑘 (𝑥), 𝑥 ∈ 𝐴; this function is well-defined on 𝐴. Given 𝜀 > 0,
∞
there is 𝑁 ∈ ℕ such that ∑𝑘=𝑛+1 𝑀𝑘 < 𝜀 for all 𝑛 ≥ 𝑁. But then for any 𝑛 ≥ 𝑁
and any 𝑥 ∈ 𝐴 we have
𝑛 ∞ ∞ ∞
�𝑓(𝑥) − � 𝑓𝑘 (𝑥)� = � � 𝑓𝑘 (𝑥)� ≤ � |𝑓𝑘 (𝑥)| ≤ � 𝑀𝑘 < 𝜀.
𝑘=1 𝑘=𝑛+1 𝑘=𝑛+1 𝑘=𝑛+1

93
 Chapter 6 Sequences of functions

This proves the uniform convergence.

□

Example 6.15 This is a continuation of Example 6.13. The absolute and uni-
∞
form convergence of the series ∑𝑘=0 𝑥𝑘 on [−𝑟, 𝑟] with 𝑟 ∈ (0, 1) follows from the
Weierstrass M-Test with 𝑀𝑘 = 𝑟𝑘 , 𝑘 ∈ ℕ ∪ {0}.
△
∞ 𝑥𝑘
Example 6.16 Consider the series ∑𝑘=0 , 𝑥 ∈ ℝ. You may recognize the
𝑘!
Maclaurin series of the function 𝑒𝑥 .
Take 𝑟 > 0 and consider the interval [−𝑟, 𝑟]. We have

𝑥𝑘 𝑟𝑘
� �≤ = 𝑀𝑘 , 𝑥 ∈ [−𝑟, 𝑟], 𝑘 ∈ ℕ ∪ {0}.
𝑘! 𝑘!

∞ 𝑟𝑘
The series ∑𝑘=0 𝑘!
converges by the Ratio Test (Theorem 2.52), indeed,

𝑟𝑘+1 𝑘! 𝑟
𝐿 = lim 𝑘
= lim = 0 < 1.
𝑘→∞ (𝑘 + 1)! 𝑟 𝑘→∞ 𝑘 + 1

∞ 𝑥𝑘
Thus, by the Weierstrass M-Test, the series ∑𝑘=0 𝑘!
converges absolutely and
uniformly on [−𝑟, 𝑟].
𝑛 𝑥𝑘
Notice that the partial sums ∑𝑘=0 𝑘! are polynomials. In particular, they are
continuous. It follows from Theorem 6.7 that 𝑒𝑥 is continuous on [−𝑟, 𝑟] with
any 𝑟 > 0. Consequently, 𝑒𝑥 is continuous on ℝ. Furthermore, each partial sum
is differentiable with
𝑛 ′ 𝑛 𝑛−1 𝑘
𝑥𝑘 𝑥𝑘−1 𝑥
�� � = � = � .
𝑘=0
𝑘! 𝑘=1
(𝑘 − 1)! 𝑘=0 𝑘!
∞
𝑛−1 𝑥𝑘
By above, the sequence of the derivatives �∑𝑘=0 𝑘!
� converges uniformly on
𝑛=1
[−𝑟, 𝑟] to 𝑒𝑥 . By Theorem 6.11, 𝑒𝑥 is differentiable and (𝑒𝑥 )′ = 𝑒𝑥 .
△
Example 6.17 Consider the series ∑∞ sin(𝑘𝑥) . This is an example of a Fourier
𝑘=0 2𝑘
series. We have
sin(𝑘𝑥) 1
� 𝑘
� ≤ 𝑘 = 𝑀𝑘 for all 𝑥 ∈ ℝ and 𝑘 ∈ ℕ ∪ {0}.
2 2

94
 Chapter 6 Sequences of functions

∞ 1
The series ∑𝑘=0 is a convergent geometric series. Hence, by the Weierstrass M-
2𝑘
Test the series ∑∞ sin(𝑘𝑥)
converges absolutely and uniformly on ℝ. Moreover,
𝑘=0 2𝑘
its sum is continuous on ℝ.
△

6.3 Power series

Definition 6.18 Let (𝑎𝑛 )∞
𝑛=0 be a sequence of real numbers and 𝑥0 ∈ ℝ. The
series
∞
� 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 = 𝑎0 + 𝑎1 (𝑥 − 𝑥0 ) + 𝑎2 (𝑥 − 𝑥0 )2 + 𝑎3 (𝑥 − 𝑥0 )3 + ⋯
𝑛=0

is called a power series.

△
In case when 𝑥0 = 0, a power series has the form
∞
� 𝑎𝑛 𝑥𝑛 = 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥2 + 𝑎3 𝑥3 + ⋯ .
𝑛=0

For each 𝑛 ∈ ℕ, the partial sum

𝑛
𝑝𝑛 (𝑥) = � 𝑎𝑘 (𝑥 − 𝑥0 )𝑘 = 𝑎0 + 𝑎1 (𝑥 − 𝑥0 ) + 𝑎2 (𝑥 − 𝑥0 )2 + ⋯ + 𝑎𝑛 (𝑥 − 𝑥0 )𝑛
𝑘=0

is a polynomial of degree at most 𝑛.

For a fixed 𝑥 ∈ ℝ, the sequence of the partial sums (𝑝𝑛 (𝑥))∞𝑛=1 might or might
not converge. It will always converge at 𝑥 = 𝑥0 , because 𝑝𝑛 (𝑥0 ) = 𝑎0 for all 𝑛 ∈ ℕ.
On the other hand, if 𝑎𝑛 = 0 for all 𝑛 ≥ 𝑁 with some 𝑁 ∈ ℕ, then the series
converges for all 𝑥 ∈ ℝ. Another example of a power series that converges for
∞ 𝑥𝑛
all 𝑥 ∈ ℝ is ∑𝑛=0 𝑛! .
We are interested in investigating the domain of convergence of the power series
∑∞ 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 which is the set
𝑛=0
∞
𝑛
�𝑥 ∈ ℝ ∶ � 𝑎𝑛 (𝑥 − 𝑥0 ) converges� .
𝑛=0

We will see that that the domain of convergence of a power series is always
an interval centered at 𝑥0 . It can be the whole of ℝ, the single point {𝑥0 }, or an
interval of the form (𝑥0 −𝑅, 𝑥0 +𝑅), [𝑥0 −𝑅, 𝑥0 +𝑅], [𝑥0 −𝑅, 𝑥0 +𝑅), or (𝑥0 −𝑅, 𝑥0 +𝑅]
with some 𝑅 > 0.

95
 Chapter 6 Sequences of functions

∞
Example 6.19 Consider the series ∑𝑛=0 𝑛!𝑥𝑛 . It converges at 𝑥 = 0. However,
if 𝑥 ≠ 0, then it diverges by the Ratio Test (Theorem 2.52):

(𝑛 + 1)!𝑥𝑛+1
𝐿 = lim = lim (𝑛 + 1)𝑥 = ∞.
𝑛→∞ 𝑛!𝑥𝑛 𝑛→∞

The domain of convergence is {0}.

△
Example 6.20 The power series ∑∞ 𝑥𝑛
converges when |𝑥| < 1 and diverges
𝑛=0
when |𝑥| ≥ 1. Its domain of convergence is (−1, 1).
△
∞ 𝑥𝑛
Example 6.21 Consider the power series ∑𝑛=1 . We know that it diverges if
𝑛
𝑥 = 1 (the harmonic series) and converges if 𝑥 = −1 (the alternating harmonic
∞
series). By the comparison with the series ∑𝑛=1 𝑥𝑛 , it converges absolutely if
|𝑥| < 1. The domain of convergence is [−1, 1).
△
∞
Theorem 6.22 ( 1 ) Suppose that the power series ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 ) converges for
𝑛
∞
a certain 𝑥1 ∈ ℝ with 𝑥1 ≠ 𝑥0 . Let 0 < 𝑟 < |𝑥1 − 𝑥0 |. Then ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛
converges absolutely and uniformly on [𝑥0 − 𝑟, 𝑥0 + 𝑟].
∞
(2 ) If ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 diverges for some 𝑥2 ∈ ℝ, then it diverges for all 𝑥 ∈ ℝ with
|𝑥 − 𝑥0 | > |𝑥2 − 𝑥0 |.
△
This theorem implies that the domain of convergence is centered at 𝑥0 .
∞
Proof. We first prove (1). Since the series ∑𝑛=0 𝑎𝑛 (𝑥1 − 𝑥0 )𝑛 converges, we have
lim𝑛→∞ 𝑎𝑛 (𝑥1 − 𝑥0 )𝑛 = 0. By the definition of the limit, there exists 𝑁 ∈ ℕ such
that |𝑎𝑛 (𝑥1 − 𝑥0 )𝑛 | < 1 for all 𝑛 ≥ 𝑁. It follows that
1
|𝑎𝑛 | < for all 𝑛 ≥ 𝑁.
|𝑥1 − 𝑥0 |𝑛
Now let 0 < 𝑟 < |𝑥1 − 𝑥0 | and 𝑥 ∈ [𝑥0 − 𝑟, 𝑥0 + 𝑟]. Using the inequality |𝑥 − 𝑥0 | ≤ 𝑟,
we obtain
𝑟𝑛
|𝑎𝑛 (𝑥 − 𝑥0 )𝑛 | = |𝑎𝑛 | ⋅ |𝑥 − 𝑥0 |𝑛 ≤ for all 𝑛 ≥ 𝑁.
|𝑥1 − 𝑥0 |𝑛
𝑟 ∞ 𝑟𝑛
Notice that � 𝑥 � < 1, and therefore the geometric series ∑𝑛=0 converges.
1 −𝑥0 |𝑥1 −𝑥0 |𝑛
By the Weierstrass M-Test (Theorem 6.14) we conclude that the power series
∑∞ 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 converges absolutely and uniformly on [𝑥0 − 𝑟, 𝑥0 + 𝑟].
𝑛=0

96
 Chapter 6 Sequences of functions

Statement (2) follows from (1). Indeed, suppose, by contradiction, that the
∞
series ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 converges for some 𝑥 ∈ ℝ with |𝑥 − 𝑥0 | > |𝑥2 − 𝑥0 |. Then
by (1) it would converge also at 𝑥2 , which contradicts the assumption.
□
∞
Definition 6.23 The radius of convergence of a power series ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 is
the number
∞
𝑅 = sup �|𝑥 − 𝑥0 | ∶ � 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 converges� .
𝑛=0

If ∑∞ 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 converges for all 𝑥 ∈ ℝ, we put 𝑅 = ∞.

𝑛=0
△
Notice that
∞
𝑅 = inf �|𝑥 − 𝑥0 | ∶ � 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 diverges� .
𝑛=0
∞
If 𝑅 = 0, the power series ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 converges only at 𝑥 = 𝑥0 . If 𝑅 > 0,
∞
then the power series ∑𝑛=0 𝑎𝑛 (𝑥−𝑥0 )𝑛 converges absolutely and uniformly on any
interval [𝑥0 −𝑟, 𝑥0 +𝑟] with 0 < 𝑟 < 𝑅. It converges absolutely on (𝑥0 −𝑅, 𝑥0 +𝑅), but
might not converge uniformly on this interval. It might or might not converge
at the points 𝑥0 − 𝑅 and 𝑥0 + 𝑅.
𝑛
The partial sums ∑𝑘=0 𝑎𝑘 (𝑥 − 𝑥0 )𝑘 are polynomials and therefore differentiable.
∞
It can be shown (we will not do this in our course) that ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 and the
∞
differentiated series ∑𝑛=1 𝑛𝑎𝑛 (𝑥 − 𝑥0 )𝑛−1 have the same radius of convergence.
∞
The Differential Limit Theorem (Theorem 6.11) says that ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 is
differentiable on (𝑥0 − 𝑅, 𝑥0 + 𝑅), and can be differentiated term-by-term.
We finish this section by presenting two useful formulas for the radius of
convergence.
∞
Theorem 6.24 Let ∑ 𝑎𝑛 (𝑥−𝑥0 )𝑛 be a power series with the radius of convergence 𝑅.
𝑛=0

𝑎𝑛+1
(1) If lim𝑛→∞ � 𝑎𝑛
� exists, then

1 𝑎𝑛+1
= lim � �.
𝑅 𝑛→∞ 𝑎𝑛

1
(2 ) If lim𝑛→∞ |𝑎𝑛 | 𝑛 exists, then
1 1
= lim |𝑎𝑛 | 𝑛 .
𝑅 𝑛→∞

97
 Chapter 6 Sequences of functions

Proof. To prove (1), we will use the Ratio Test (Theorem 2.52). Put

|𝑎𝑛+1 (𝑥 − 𝑥0 )𝑛+1 | 𝑎𝑛+1

𝐿 = lim 𝑛
= lim � � |𝑥 − 𝑥0 |.
𝑛→∞ |𝑎𝑛 (𝑥 − 𝑥0 ) | 𝑛→∞ 𝑎𝑛

We have
1
𝐿 < 1 ⟺ |𝑥 − 𝑥0 | < 𝑎𝑛+1
,
lim𝑛→∞ � 𝑎𝑛
�
1
𝐿 > 1 ⟺ |𝑥 − 𝑥0 | > 𝑎𝑛+1
.
lim𝑛→∞ � 𝑎𝑛
�

This gives
1
𝑅= 𝑎𝑛+1
.
lim𝑛→∞ � 𝑎𝑛
�

The proof of (2) is similar. It uses the Root Test (Theorem 2.55). We have
1 1
𝐿 = lim |𝑎𝑛 (𝑥 − 𝑥0 )𝑛 | 𝑛 = lim |𝑎𝑛 | 𝑛 |𝑥 − 𝑥0 |.
𝑛→∞ 𝑛→∞

By a similar argument we conclude that

1
𝑅= 1 .
lim𝑛→∞ |𝑎𝑛 | 𝑛
□

6.4 Taylor series

∞
Let ∑𝑛=0 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 be a power series with radius of convergence 𝑅 > 0. For
0 < 𝑟 < 𝑅, the series converges absolutely and uniformly on [𝑥0 − 𝑟, 𝑥0 + 𝑟] to
a function 𝑓. This defines 𝑓 on the interval (𝑥0 − 𝑅, 𝑥0 + 𝑅). Let us establish a
connection between the function 𝑓 and the coefficients (𝑎𝑛 )∞𝑛=0 .
We have

𝑓(𝑥) = 𝑎0 + 𝑎1 (𝑥 − 𝑥0 ) + 𝑎2 (𝑥 − 𝑥0 )2 + 𝑎3 (𝑥 − 𝑥0 )3 + ⋯ .

Substituting 𝑥 = 𝑥0 in this formula gives 𝑓(𝑥0 ) = 𝑎0 .

98
 Chapter 6 Sequences of functions

As discussed in the previous section, the function 𝑓 is differentiable on (𝑥0 −

∞
𝑅, 𝑥0 + 𝑅), and the differentiated series ∑𝑛=1 𝑛𝑎𝑛 (𝑥 − 𝑥0 )𝑛−1 converges absolutely
and uniformly on [𝑥0 − 𝑟, 𝑥0 + 𝑟] to 𝑓′ (𝑥). Thus, for 𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅) we have

𝑓′ (𝑥) = 𝑎1 + 𝑎2 (𝑥 − 𝑥0 ) + 𝑎3 (𝑥 − 𝑥0 )2 + 𝑎4 (𝑥 − 𝑥0 )3 + ⋯ .

Substituting 𝑥 = 𝑥0 in this formula gives 𝑓′ (𝑥0 ) = 𝑎1 .

Using induction, one can prove that 𝑓 can be differentiated as many times as
we please, and for 𝑚 ∈ ℕ the differentiated series
∞
� 𝑛(𝑛 − 1) ⋯ (𝑛 − 𝑚 + 1)𝑎𝑛 (𝑥 − 𝑥0 )𝑛−𝑚
𝑛=𝑚

converges absolutely and uniformly on [𝑥0 − 𝑟, 𝑥0 + 𝑟] to 𝑓(𝑚) (𝑥). Thus, for

𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅) we have

𝑓𝑚) (𝑥) = 𝑚(𝑚 − 1) ⋯ 1𝑎𝑚 + (𝑚 + 1)𝑚 ⋯ 2𝑎𝑚+1 (𝑥 − 𝑥0 ) + ⋯ .

Substituting 𝑥 = 𝑥0 in this formula, we see that

𝑓(𝑚) (𝑥0 )
𝑎𝑚 = , 𝑚 ∈ ℕ ∪ {0}.
𝑚!
Thus,
∞ ∞
𝑓(𝑛) (𝑥0 )
𝑓(𝑥) = � 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 = � (𝑥 − 𝑥0 )𝑛 , 𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅).
𝑛=0 𝑛=0 𝑛!

We recognize that the latter series is the Taylor series of 𝑓 at the point 𝑥0 . In
the special case when 𝑥0 = 0 it is also called the Maclaurin series
∞
𝑓(𝑛) (0) 𝑛
𝑓(𝑥) = � 𝑥 .
𝑛=0 𝑛!

Let us summarize what we just did. We started with a power series that
converges on the interval (𝑥0 − 𝑅, 𝑥0 + 𝑅) and defined the function 𝑓 to be its
sum. Then we saw that the power series must be the Taylor series of 𝑓.
Now let us instead suppose that a function 𝑓 ∶ (𝑥0 − 𝑅, 𝑥0 + 𝑅) → ℝ is given
and that 𝑓 has derivatives of all orders on (𝑥0 − 𝑅, 𝑥0 + 𝑅). We can then formally
write down the Taylor series of 𝑓
∞
𝑓(𝑛) (𝑥0 )
� (𝑥 − 𝑥0 )𝑛 .
𝑛=0 𝑛!

99
 Chapter 6 Sequences of functions

Figure 6.2: A graph of the function 𝑔0

Unfortunately, it may happen that this series does not converge to 𝑓 on (𝑥0 −
𝑅, 𝑥0 + 𝑅). For a particular 𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅) it can happen that the series
diverges, or that it converges to a value different from 𝑓(𝑥).
Example 6.25 Consider the function
⎧ 1
⎪
⎪ −
⎨𝑒 𝑥2 , 𝑥 ≠ 0,
𝑔0 (𝑥) = ⎪
⎪
⎩0, 𝑥 = 0.

A graph of this function is shown in Figure 6.2

(𝑛)
It is possible to prove that 𝑔0 (0) = 0 for all 𝑛 ∈ ℕ ∪ {0}. It follows that its
Taylor series at 𝑥0 = 0 is

∞ (𝑛)
𝑔0 (0) 𝑛
� 𝑥 =0 for all 𝑥 ∈ ℝ.
𝑛=0 𝑛!

It does not converge to 𝑔0 (𝑥) if 𝑥 ≠ 0.

This example can be generalized. Assume that a function 𝑓 has a power series
∑∞ 𝑎𝑛 (𝑥 − 𝑥0 )𝑛 that converges to 𝑓 on (𝑥0 − 𝑅, 𝑥0 + 𝑅). Take 𝛼 ∈ ℝ and consider
𝑛=0
the function ⎧ 1
⎪
⎪ −
⎨𝑓(𝑥) + 𝛼𝑒 (𝑥−𝑥0)2 , 𝑥 ≠ 𝑥0 ,
𝑔(𝑥) = ⎪⎪
⎩𝑓(𝑥 ), 𝑥 = 𝑥0 .
0

By above, 𝑓(𝑛) (𝑥0 ) = 𝑔(𝑛) (𝑥0 ) for all 𝑛 ∈ ℕ ∪ {0}. This means that 𝑓 and 𝑔 have
the same Taylor series at 𝑥0 . Since 𝛼 ∈ ℝ can be chosen arbitrarily, we see that
there are infinitely many functions having the same Taylor series at 𝑥0 .
△

100
 Chapter 6 Sequences of functions

Let us now address the following task. Let 𝑓 ∶ (𝑥0 − 𝑅, 𝑥0 + 𝑅) → ℝ be a given

function that has derivatives of all orders on (𝑥0 − 𝑅, 𝑥0 + 𝑅). How can we check
whether its Taylor series converges to 𝑓? And it if converges, how fast?
For 𝑛 ∈ ℕ, the partial sum of the Taylor series
𝑛
𝑓(𝑘) (𝑥0 )
𝑇𝑛 (𝑥) = � (𝑥 − 𝑥0 )𝑘
𝑘=0
𝑘!

is called the Taylor polynomial of order 𝑛. We study the reminder

𝑅𝑛 (𝑥) = 𝑓(𝑥) − 𝑇𝑛 (𝑥).

∞ 𝑓(𝑛) (𝑥 )
For a given 𝑥, the Taylor series ∑𝑛=0 𝑛! 0 (𝑥 − 𝑥0 )𝑛 converges to 𝑓(𝑥) if and only
if
lim 𝑅𝑛 (𝑥) = 0.
𝑛→∞
There are several different formulas for the reminder 𝑅𝑛 (𝑥). We are going to
prove one of them.
Theorem 6.26 (Taylor’s Theorem with Lagrange’s Form of the Reminder) Let 𝑓 ∶
(𝑥0 − 𝑅, 𝑥0 + 𝑅) → ℝ be a function that is 𝑛 + 1 times differentiable on (𝑥0 − 𝑅, 𝑥0 + 𝑅).
For any 𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅), there exists a point 𝑐𝑥 that lies between 𝑥 and 𝑥0 such that

𝑓(𝑛+1) (𝑐𝑥 )
𝑅𝑛 (𝑥) = (𝑥 − 𝑥0 )𝑛+1 .
(𝑛 + 1)!
△
According to Taylor’s Theorem,

𝑓″ (𝑥0 ) 𝑓(3) (𝑥0 )

𝑓(𝑥) =𝑓(𝑥0 ) + 𝑓′ (𝑥0 )(𝑥 − 𝑥0 ) + (𝑥 − 𝑥0 )2 + (𝑥 − 𝑥0 )3 + ⋯
2 3!
𝑓(𝑛) (𝑥0 ) 𝑓(𝑛+1) (𝑐𝑥 )
+ (𝑥 − 𝑥0 )𝑛 + (𝑥 − 𝑥0 )𝑛+1 .
𝑛! (𝑛 + 1)!

Proof. Let 𝑥 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅) be a fixed number. For 𝑡 ∈ (𝑥0 − 𝑅, 𝑥0 + 𝑅), consider

the functions
𝑛
𝑛+1 𝑓(𝑘) (𝑡)
𝐹(𝑡) = (𝑡 − 𝑥) , 𝐺(𝑡) = � (𝑥 − 𝑡)𝑘 .
𝑘=0
𝑘!
We have

𝐹(𝑥0 ) = (𝑥0 − 𝑥)𝑛+1 , 𝐹(𝑥) = 0, 𝐹′ (𝑡) = (𝑛 + 1)(𝑡 − 𝑥)𝑛 .

101
 Chapter 6 Sequences of functions

Moreover,
𝑛
𝑓(𝑘) (𝑥0 )
𝐺(𝑥0 ) = � (𝑥 − 𝑥0 )𝑘 , 𝐺(𝑥) = 𝑓(𝑥)
𝑘=0
𝑘!
and
𝑛
𝑓(𝑘+1) (𝑡) 𝑓(𝑘) (𝑡)
𝐺′ (𝑡) = � � (𝑥 − 𝑡)𝑘 − 𝑘(𝑥 − 𝑡)𝑘−1 �
𝑘=0
𝑘! 𝑘!
𝑛 𝑛−1 (𝑘+1)
𝑓(𝑘+1) (𝑡) 𝑘 𝑓 (𝑡) 𝑓(𝑛+1) (𝑡)
=� (𝑥 − 𝑡) − � (𝑥 − 𝑡)𝑘 = (𝑥 − 𝑡)𝑛 .
𝑘=0
𝑘! 𝑘=0
𝑘! 𝑛!

By the Generalized Mean Value Theorem (Theorem 5.7), there is a point 𝑐𝑥

between 𝑥 and 𝑥0 such that

(𝐹(𝑥) − 𝐹(𝑥0 ))𝐺′ (𝑐𝑥 ) = 𝐹′ (𝑐𝑥 )(𝐺(𝑥) − 𝐺(𝑥0 )).

Substituting the above formulas, we obtain

𝑓(𝑛+1) (𝑐𝑥 )
�0 − (𝑥0 − 𝑥)𝑛+1 � (𝑥 − 𝑐𝑥 )𝑛
𝑛!
𝑛
𝑓(𝑘) (𝑥0 )
= (𝑛 + 1)(𝑐𝑥 − 𝑥)𝑛 �𝑓(𝑥) − � (𝑥 − 𝑥0 )𝑘 �,
𝑘!
���������������������������������������
𝑘=0

=𝑅𝑛 (𝑥)

so that

−(−1)𝑛+1 (𝑥 − 𝑥0 )𝑛+1 𝑓(𝑛+1) (𝑐𝑥 )(𝑥 − 𝑐𝑥 )𝑛 𝑓(𝑛+1) (𝑐𝑥 )

𝑅𝑛 (𝑥) = = (𝑥 − 𝑥0 )𝑛+1
𝑛!(𝑛 + 1)(−1)𝑛 (𝑥 − 𝑐𝑥 )𝑛 (𝑛 + 1)!

as desired.
□

Corollary 6.27 (Taylor’s Inequality) If |𝑓(𝑛+1) (𝑥)| ≤ 𝑀 for all 𝑥 ∈ [𝑥0 − 𝑟, 𝑥0 + 𝑟]

with some 𝑀 > 0, then

𝑀𝑟𝑛+1
|𝑅𝑛 (𝑥)| ≤ , 𝑥 ∈ [𝑥0 − 𝑟, 𝑥0 + 𝑟].
(𝑛 + 1)!
△

102
 Chapter 6 Sequences of functions

Example 6.28 Consider the function 𝑓(𝑥) = sin 𝑥, 𝑥 ∈ ℝ.

We know that its Taylor series at the point 0 is
∞
(−1)𝑘 2𝑘+1
� 𝑥 .
𝑘=0
(2𝑘 + 1)!

We are going to discuss the convergence of this series.

We can use the Ratio Test (Theorem 2.52) to show that this series converges
absolutely for all 𝑥 ∈ ℝ. Indeed,

|𝑥|2𝑘+3 (2𝑘 + 1)! 1

𝐿 = lim 2𝑘+1
= |𝑥|2 lim =0<1
𝑘→∞ (2𝑘 + 3)! |𝑥| 𝑘→∞ (2𝑘 + 3)(2𝑘 + 2)

for all 𝑥 ∈ ℝ.
Our next aim is to show that the sum of the series is indeed sin 𝑥. We are
going to use Corollary 6.27. All derivatives of the function 𝑓(𝑥) = sin 𝑥 have the
form ± sin 𝑥 or ± cos 𝑥. Thus, we definitely have

|𝑓(𝑛) (𝑥)| ≤ 1 for all 𝑥 ∈ ℝ and 𝑛 ∈ ℕ.

This gives us
|𝑥|𝑛+1
|𝑅𝑛 (𝑥)| ≤ for all 𝑥 ∈ ℝ and 𝑛 ∈ ℕ.
(𝑛 + 1)!
We conclude that
lim 𝑅𝑛 (𝑥) = 0 for any 𝑥 ∈ ℝ.
𝑛→∞
This implies that
∞
(−1)𝑘 2𝑘+1
sin 𝑥 = � 𝑥 , 𝑥 ∈ ℝ.
𝑘=0
(2𝑘 + 1)!
The above estimate for |𝑅𝑛 (𝑥)| can be also used to determine the number of
summands needed to calculate sin 𝑥 for a given 𝑥 with a given precision. As an
example, let us calculate sin 1 within 4 digits. For, we need to find 𝑛 such that

1
|𝑅𝑛 (1)| < ⋅ 10−4 .
2
1
Using the estimate |𝑅𝑛 (1)| ≤ (𝑛+1)!
, we see that it suffices to find 𝑛 such that

1 1
< ⋅ 10−4 ,
(𝑛 + 1)! 2

103
 Chapter 6 Sequences of functions

or, equivalently,
(𝑛 + 1)! > 20 000.
Calculating 1!, 2! and so on, we find that 8! = 40 320, so we take 𝑛 = 7. The
desired approximation is

1 1 1
sin 1 ≈ 1 − + − = 0.8415.
3! 5! 7!
△

104
Index

𝑝-series, 39 convergence
absolute, 48, 93
absolute convergence, 48, 93 conditional, 48
Algebraic Limit Theorem, 26, 70 of a sequence, 22
alternating harmonic series, 48 of a series, 33, 92
alternating series, 48 pointwise, 87, 92
Alternating Series Test, 48 uniform, 88, 93
Archimedean Property, 13 countable, 17
at most countable, 17
dense, 15
bijection, 16 derivative, 80
Bolzano-Weierstrass Theorem, 42 derived set, 60
bounded, 8, 24 difference of sets, 2
above, 8, 24 differentiability, 80
below, 8, 24 Differentiable Limit Theorem, 91
divergent
cardinality, 16
sequence, 23
Cauchy Criterion, 46
series, 33
for seires, 47
domain, 16
for uniform convergence, 90
domain of convergence, 95
Cauchy sequence, 44
closed set, 61 element, 1
closure, 62 empty set, 1
codomain, 16 eventually in, 59
compact, 64 Extreme Value Theorem, 75
Comparison Test, 37 extremum
Completeness Axiom, 9 local, 81
conditional convergence, 48
continuity, 73 Fermat’s Theorem, 81
uniform, 77 field, 4

105
 Index

ordered, 4 Mean Value Theorem, 82, 84

finite set, 17 minimum, 11
function, 16 local, 81
Monotone Convergence Theorem,
Generalized Mean Value Theorem, 30
84 monotonicity, 29
geometric series, 35
greatest lower bound, 9 natural numbers, 3
neighborhood, 58
harmonic series, 36 Nested Interval Property, 12
image, 74 one-to-one-correspondence, 16
infimum, 9 open set, 58
injective, 16 order, 4
integers, 3 Order Limit Theorem, 28, 70
Intermediate Value Theorem, 75 ordered field, 4
intersection of sets, 1
interval, 2 partial sum, 32
isolated point, 61 pointwise convergence, 87, 92
pointwise limit, 87
l’Hospital’s Rule, 85 power series, 95
least upper bound, 9 domain of convergence, 95
Leibniz Test, 48 radius of convergence, 97
limit preimage, 74
functional, 66 Principle of Mathematical
of a sequence, 22 Induction, 5
one-sided, 66
pointwise, 87 radius of convergence, 97
uniform, 88 Ratio Test, 51
limit point, 60 rational numbers, 3
local real numbers, 3
extremum, 81 rearrangement, 55
maximum, 81 Riemann Rearrangement Theorem,
minimum, 81 56
lower bound, 8 Rolle’s Theorem, 81
Root Test, 53
Maclaurin series, 99
maximum, 11 sequence, 21
local, 81 bounded, 24

106
 Index

bounded above, 24 compact, 64

bounded below, 24 derived, 60
Cauchy, 44 empty, 1
convergent, 22 open, 58
decreasing, 29 sets
divergent, 23 difference, 2
eventually constant, 26 disjoint, 2
eventually decreasing, 31 equal, 1
eventually increasing, 31 intersection, 1
increasing, 29 union, 1
monotone, 29 Squeeze Theorem, 25, 72
subsequence, 41 subsequence, 41
series, 32 subset, 1
𝑝-series, 39 proper, 1
absolutely convergent, 48 sum of the series, 33
alternating, 48 supremum, 9
alternating harmonic, 48 surjective, 16
Cauchy Criterion, 47
conditionally convergent, 48 Taylor polynomial, 101
convergent, 33 Taylor series, 99
divergent, 33 telescopic series, 35
geometric, 35
harmonic, 36 uncountable, 17
of functions, 92 uniform continuity, 77
partial sum, 32 uniform convergence, 88, 93
power, 95 uniform limit, 88
rearrangement, 55 union of sets, 1
sum, 33 upper bound, 8
telescopic, 35
set, 1 Weierstrass M-Test, 93
closed, 61 Well-Ordering Principle, 5

107