Contents

1 Metric Spaces 1
1.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Connected Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Sequences and Series 21

2.1 Convergent Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Subsequences and Sequential Compactness . . . . . . . . . . . . . . . 25
2.3 Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Complete Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Series in Normed Vector Spaces . . . . . . . . . . . . . . . . . . . . . 34
2.6 Absolute Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.7 Series of Non-Negative Real Numbers . . . . . . . . . . . . . . . . . . 38
2.8 Two Fundamental Tests and Power Series . . . . . . . . . . . . . . . . 42
2.9 The Cauchy Product of Series . . . . . . . . . . . . . . . . . . . . . . 46
2.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3 Functional Limits and Continuity 53

3.1 Functional Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Limits on the Extended Real Line . . . . . . . . . . . . . . . . . . . . 56
3.3 Asymptotic Comparison . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4 The Derivative 67
4.1 The Derivative of a Real Function . . . . . . . . . . . . . . . . . . . . 67
4.2 Mean Value Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 L’Hôpital’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.4 Fréchet Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

i
ii CONTENTS

4.5 Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.6 The Implicit / Inverse Function Theorem . . . . . . . . . . . . . . . . . 69
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

A Logic and Set Theory 71

A.1 Equivalence Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.2 Countability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
A.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

B The Construction Of The Real Field 81

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
B.2 Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
B.3 The Real and Complex Fields . . . . . . . . . . . . . . . . . . . . . . . 84

C Linear Algebra 89

Solutions 91

Bibliography 105
Preface

to say that there is a huge number of excellent books

I T IS AN UNDERSTATEMENT
for a first course in Analysis. Some of them marked generations of mathematicians
and some of them are cited in more books than I can count. However, I think most are
either too dense or doesn’t cover properly some of the joys of this beautiful area of
mathematics.
It is common to give Analysis books the task of developing mathematical maturity
in the students. While this is one of the goals of this book, I don’t think the best way to
do it is by reading a terse and dense book. To become mature, one has to think a lot to
be sure the theory is concise and complete, and my intent is to make this less painful.
Pierre Deligne once wrote the following in the Notices of the AMS[6]: “From
[Grothendieck], I have learned not to take glory in the difficulty of a proof: difficulty
means we have not understood. The ideal is to be able to paint a landscape in which the
proof is obvious.” This beautifully summarizes my approach with this book. It is the
book I wish I had in the beginning of my studies in Analysis.
Whenever possible and not too troublesome, we’ll generalize our results as much
as possible. This usually makes clear what should be used to prove the theorems. For
example, suppose we want to prove something related to a closed interval. A priori,
we don’t know what property of closed intervals makes our result true. It could be
compactness, connectedness or almost anything. However if we state the theorem using
only the fundamental property, it usually becomes clear what should be used in the proof.
It is expected that the reader knows the basics of set theory and linear algebra.
While you could read this book without having studied calculus beforehand, it is highly
recommended that the reader is familiar with the basic machinery of differential and
integral calculus.
Some passages are denoted with a dangerous bend symbol to warn the reader of
things that may not have been considered. This differs substantially from the original
use of N. Bourbaki, in which it is used to forewarn the reader against serious errors.
By no means one should expect that all this work is original. It is basically my view
on the things I learned with great books and great teachers. During the writing of this
book, a couple of people helped me spot grammatical, theoretical and pedagogical errors.
To them I owe my sincere gratitude. Last, but not least I would like to thank Umberto C.
C. Malanga for showing me the beauty in mathematics.

Gabriel Ribeiro

iii
 CHAPTER
Metric Spaces
1
F RÉCHET introduced in his 1906’s work Sur Quelques Points Du Calcul
M AURICE
Fonctionnel the concept of a metric space, which is basically a set endowed with
a notion of distance between elements. As we’ll see, that notion is enough to study many
concepts in analysis such as sequences and continuity. Afterwards we’ll understand
normed spaces which, as the natural setting to series and derivatives, are metric spaces
whose set of points is a vector space.

1.1 Metric Spaces

Definition 1.1.1 — Metric Space. Let M be a non-empty set whose elements we
shall call points. A metric in M is a function d : M × M → R which satisfies the
following properties for all x, y, z ∈ M.
• (Identity of Indiscernibles) d(x, y) = 0 if and only if x = y.
• (Triangle Inequality) d(x, y) + d(x, z) ≥ d(y, z).
• (Symmetry) d(x, y) = d(y, x).

• (Positivity) d(x, y) ≥ 0.
We call the pair (M, d) a metric space.

Notice that the first two axioms of a metric imply the other two. In fact, taking x = z in
 the triangle inequality implies symmetry and taking y = z implies positivity. You should
convince yourself why this is true if it is not clear.

Whenever the context makes it clear, we’ll talk about the metric space M without
making reference to the metric. One should notice that every subset of a metric space is
a metric space in its own right, with the same metric restricted to the subset.
 2 Chapter 1. Metric Spaces

Example 1.1 The prototypical example of metric spaces, from our standpoint, is the
euclidean space Rn . The metric in Rn is defined by
s
n
d(x, y) = ∑ (xi − yi )2 .
i=1

The Cauchy-Schwarz inequality assures that this is, in fact, a metric. 

 Example 1.2 Another important example (particularly for counter-examples) of metric

space is the discrete metric space. The discrete metric in any non-empty set is defined by
(
1 if x 6= y
d(x, y) = .
0 if x = y


1.2 Open Sets

Definition 1.2.1 — Open Sets. Let M be a metric space and E be a subset of M. All
points mentioned below are understood to be elements of M.
• An open ball centered at x with radius r is the set B(x, r) consisting of all y
such that d(x, y) < r.
• A point x is an interior point of E if there is an open ball centered at x contained
in E. The set of all interior points of E is called int E.
• E is open if every point of E is an interior point of E.

Since we defined an open ball, we ought to check that this is always an open set.

Theorem 1.2.1 In every metric space, an open ball is an open set.

Proof. Let a ∈ B(x, r). Then d(x, a) < r and consequently s = r − d(x, a) is a
positive number. We affirm that B(a, s) ⊂ B(x, r).

r
b s
x a

In fact, if b ∈ B(a, s), then d(a, b) < s and so d(x, b) ≤ d(x, a) + d(a, b) < d(x, a) +
s = r. This implies b ∈ B(x, r).
1.2 Open Sets 3

Theorem 1.2.2 Consider a family of open subsets of a metric space. Then the
following properties hold.
[
• Given a (not necessarily countable) family of indices Λ, Eλ is open.
λ ∈Λ
m
\
• For every positive integer m, Ei is open.
i=1

Proof. Let E = λ Eλ . If x ∈ E, then x ∈ Eλ for some λ ∈ Λ. Since Eλ is open,

S

then x is an interior point of Eλ and, consequently, of E. This proves the first part of the
theorem.
Next, put F = m i=1 Ei . For any x ∈ F, there exist open balls B(x, ri ), such that B(x, ri ) ⊂
T

Ei for all i = 1, . . . , m. Define r to be the minimum of the set {r1 , r2 , . . . , rm }. Then

B(x, r) ⊂ Ei for i = 1, . . . , m, so that B(x, r) ⊂ F, and F is open.

It should be noted that the intersection of a infinite number of open sets need not be
open. Can you think of an example?
Of all the open sets, the open ball is probably the most important since, in some
sense, it provides a "basis" for all the open sets as is shown in the next corollary.

Corollary 1.2.3 A subset E ⊂ M is open if, and only if, it is a union of open balls.

Proof. The fact that if E is a union of open balls, then E is open follows readily
from the preceding theorem. We shall then prove the converse.
If E is open, then for all x ∈ E we have that {x} ⊂ B(x, rx ) ⊂ E for some rx > 0. Taking
unions, we see that [ [
E= {x} ⊂ B(x, rx ) ⊂ E.
x∈E x∈E

It follows that [
E= B(x, rx ).
x∈E

We already know how open sets behave under unions and intersections. The last set
operation to be studied is the Cartesian product.

Theorem 1.2.4 Let E1 be open in (M1 , d1 ) and E2 be open in (M2 , d2 ). Then E1 × E2

is open as a subset of M1 × M2 endowed with the maximum metric.

d((x1 , x2 ), (y1 , y2 )) = max{d1 (x1 , y1 ), d2 (x2 , y2 )}.

Proof. Let x = (x1 , x2 ) be a point of E1 × E2 . Since E1 is open in M1 , there is

r1 > 0 such that B(x1 , r1 ) is contained in E1 . As the same is valid for E2 , we have
that B(x1 , r1 ) × B(x2 , r2 ) is contained in E1 × E2 . Let r = min{r1 , r2 }. Then B(x, r) ⊂
B(x1 , r1 ) × B(x2 , r2 ) ⊂ E1 × E2 and the result follows.
 4 Chapter 1. Metric Spaces

Corollary 1.2.5 Let Ei be open in (Mi , di ) for every i in {1, 2, . . . , n}. Then E1 ×
. . . × En is open as a subset of M1 × . . . × Mn endowed with the maximum metric.

d((x1 , . . . , xn ), (y1 , . . . , yn )) = max{d1 (x1 , y1 ), . . . , dn (xn , yn )}.

Proof. Since we have a finite number of sets in the Cartesian product, we can
choose a smallest radius like we did before. The proof then follows similarly.

At first it seems like we proved a very particular case since the preceding corollary
is only valid for one metric in the product space. However, we will see soon that almost
every metric generates the same open sets, so that corollary is a pretty general one.
Definition 1.2.2 — Equivalence of Metrics. Let M be a set and d1 , d2 be metrics on
M. We say that d1 and d2 are equivalent (or d1 ∼ d2 ) if there exist positive constants
c,C ∈ R such that for all x, y ∈ M:

c d2 (x, y) ≤ d1 (x, y) ≤ C d2 (x, y).

You should check that this is, in fact, a equivalence relation. The motivation for
defining this equivalence is the following theorem.

Theorem 1.2.6 Let (M, d1 ) and (M, d2 ) be metric spaces such that d1 ∼ d2 . A set
E ⊂ M is open in (M, d1 ) if, and only if it is open in (M, d2 ).

Proof. If E is open in (M, d1 ), then for all x ∈ E there is some r1 > 0 such that
{y ∈ M | d1 (x, y) < r1 } ⊂ E. Since d1 (x, y) ≤ C d2 (x, y), taking r2 = r1 /C then we have
that d2 (x, y) < r2 implies d1 (x, y) < r1 and hence E is open in (M, d2 ). The converse is
analogous.

To make Corollary 1.2.5 general we just need to prove that various metrics are
equivalent to the maximum metric. This will be done on the section of normed spaces.

1.3 Closed Sets

Definition 1.3.1 — Closed Sets. All points mentioned below shall be understood as
elements of M.
• A point x is a limit point of the set E if every open ball centered at x contains a
point y 6= x such that y ∈ E.
• We denote by E 0 the set of all limit points of E and by E the union of E and
E 0 . This latter set E is called the closure of E. The set ∂ E = E \ int E is called
the boundary of E.

• E is closed if every limit point of E is a point of E.

Notice that in order for E to be closed it is not necessary that every point of E is a
limit point. A singleton is closed but it’s sole point is not a limit point.
1.3 Closed Sets 5

Theorem 1.3.1 A point x ∈ E is a limit point if and only if every open ball centered
at x contains infinitely many points from E.

Proof. Suppose x is a limit point of E and B(x, r) is a ball that contains finitely
many points of E. Since B(x, r) ∩ E is finite, we can list its elements as {p1 , p2 , . . . , pn }.
Let s be the minimum of all d(x, pi ) for i ∈ {1, 2, . . . , n}.
Then B(x, s) does not contain any points of E besides x, so that x is not a limit point of
E. Absurd!
The converse is trivial.

Based on the way we named some properties of subspaces, one can think that, in
some way, closed sets are the "opposite" of open sets. This is not true, since there exist
sets that are both open and closed. The empty set is a simple example. If that wasn’t
enough, there are sets that are neither closed nor open. The set (0, 1) considered as a
subset of the real plane is neither open nor closed. However, there is a relation between
these kinds of sets.
Theorem 1.3.2 A set E is open if, and only if its complement is closed.

Proof. First suppose E c is closed. Choose x ∈ E. Then x ∈ / E c , and x is not a limit

point of E . Hence there exists a ball B centered at x such that E c ∩ B is empty, that is,
c

B ⊂ E. Thus x is an interior point of E, and E is open.

Next, suppose E is open. Let x be a limit point of E c . Then every open ball centered
at x contains a point of E c , so that x is not an interior point of E. Since E is open, this
means that x ∈ E c . It follows that E c is closed.

Analogously to Theorem 1.2.2, we have the following result.

Theorem 1.3.3 Consider a family of closed subsets of a metric space. Then the
following properties hold.
\
• Given a arbitrary family of indices Λ, Eλ is closed.
λ ∈Λ
m
[
• For every positive integer m, Ei is closed.
i=1

Proof. Taking complements of the equations in Theorem 1.2.2 and using De Mor-
gan’s relations the result is obvious.

Theorem 1.3.4 For any set E, its closure E is closed. Moreover, E is closed if, and
only if E = E.
c
Proof. Note that if x ∈ E , then x is neither a point of E not a limit point of E.
c
Hence there exists an open ball centered at x which does not intersect E so that E is
 6 Chapter 1. Metric Spaces

open. It follows that E is closed.

If E is closed, then E 0 ⊂ E and hence E = E. The converse is obvious.

Theorem 1.3.5 Let E be a nonempty set of real numbers which is bounded above.
Let y = sup E. Then y ∈ E. Hence y ∈ E if E is closed.

Proof. If y ∈ E, then y ∈ E. Assume y ∈ / E. For every r > 0 there exists then a

point x ∈ E such that y − r < x < y, for otherwise y − r would be an upper bound of E.
Thus y is a limit point of E and then y ∈ E.

Corollary 1.3.6 Let E be a nonempty set of real numbers which is bounded below.
Let y = inf E. Then y ∈ E. Hence y ∈ E if E is closed.

1.4 Compact Sets

Since the beginning of the subject, it was noted that bounded closed sets had a special
"something" that made them incredibly useful in analysis. Initially, people thought that
it was the property that every infinite subset has a limit point. It took a long time for
mathematicians to realize that the fundamental property was the one that we describe
here as compactness.
Definition 1.4.1 Let {Gλ : λ ∈ Λ} be a collection of open subset of M. We say that
such collection is an open cover of E ⊂ M if
[
E⊂ Gλ .
λ ∈Λ

If Ω is a subset of Λ such that {Gλ : λ ∈ Ω} is still an open cover of E, we say that

{Gλ : λ ∈ Ω} is a subcover of {Gλ : λ ∈ Λ}.

Definition 1.4.2 — Compact Sets. A subset E of a metric space M is called compact

if every open cover of E contains a finite subcover.
More explicitly, the requirement is that if {Gλ : λ ∈ Λ} is an open cover of E, then
there are finitely many indices λ1 , . . . , λn such that

E ⊂ Gλ1 ∪ . . . ∪ Gλn .

This property of a set is specially important because it allows the passage of some
local properties to global properties.
As was thought before, the property that every infinite subset has a limit point is
common to every set with this special "something". However, it is not the heart of the
matter.
1.4 Compact Sets 7

Theorem 1.4.1 — Bolzano-Weierstrass. If E is an infinite subset of a compact set

K, then E has a limit point in K.

Proof. If no point of K were a limit point of E, then each x ∈ K would have a

neighborhood Nx which contains at most one point of E (namely x, if x ∈ E). It is clear
that no finite subcollection of {Nx } can cover E; and the same is true of K, since E ⊂ K.
This contradicts the compactness of K.

We’ll see now that, in fact compact sets are quite similar to closed bounded sets.

Theorem 1.4.2 If E is a compact set, then E is closed and bounded.

Proof. Let x0 be a point in E and let {Nn (x0 ) | n ∈ N} be an open covering of E.

Since E is compact, for some k ∈ N the set {Nn1 (x0 ), Nn2 (x0 ), . . . , Nnk (x0 )} is a finite
subcover of E. Taking m = max{n1 , n2 , . . . , nk } and using the triangular inequality we
see that for every x, y ∈ E
d(x, y) ≤ d(x, x0 ) + d(x0 , y) < m + m = 2m,
so that E is bounded.
Suppose now x ∈ E c . If y ∈ E, let Vy and Wy be neighborhoods of x and y, respec-
tively, of radius less than 12 d(x, y). Since E is compact, there are finitely many points
y1 , y2 , . . . , yn ∈ E such that
E ⊂ Wy1 ∪ . . . ∪Wyn = W.
If we define V as Vy1 ∩ . . . ∩Vyn , then V is a neighborhood of x which does not intersect
W . Hence V ⊂ E c , so that E c is open. It follows that E is closed.

It should be noted that, in general, the converse is not true! The set of all integers is
closed and bounded when endowed with the discrete metric, however {{x} ⊂ Z | x ∈ Z}
is an open cover which has no finite subcover.

Theorem 1.4.3 Closed subsets of compact sets are compact.

Proof. Suppose E ⊂ K ⊂ M, E is closed (relative to M) and K is compact. Let

{Gλ } be an open cover of E. If E c is adjoined to {Gλ } then this new set becomes an
open cover of K. Since K is compact, there is a finite subcollection of {Gλ } ∪ E c that
covers K and hence E. If E c is in this subcover we may remove it and obtain a finite
subcollection of {Gλ } that covers E.

Theorem 1.4.4 Let Ω be a collection of closed subsets of a compact set E such that
the intersection of every finite subcollection of Ω is non-empty. Then the intersection
of all the elements of Ω is not empty.

Proof. Before we start, one should notice that the contrapositive of compactness is
"Given any collection Φ of open sets, if no finite subcollection of Φ covers E, then Φ
8 Chapter 1. Metric Spaces

does not cover E."

Let Φ = {E \ X | X ∈ Ω}. Of course Φ is a collection of open sets. If a finite subcollection
of Φ covered E, say {φ1 , . . . , φn } ⊂ Φ, we would have that
n
\ n
[
(E \ φi ) = E \ φi = ∅.
i=1 i=1

Since E \ φi ∈ Ω this is absurd! So we have that no finite subcollection of Φ covers E.

Since E is compact, this means that Φ does not cover E. Then,
\ [
X =E\ (Y ) 6= ∅,
X∈Ω Y ∈Φ
and the result follows.

Corollary 1.4.5 If {Kn } is a sequence of nonempty closed subsets of a compact

T∞
metric space such that Kn+1 ⊂ Kn for all n ∈ {1, 2, . . . }, then n=1 Ki is not empty.

It is a fact that the preceding corollary is valid for closed subsets of the real line.
Once we prove that the closed interval [a, b] is compact this becomes trivial. Since we
cannot do this as of yet, we’ll have to prove this result separately.

Theorem 1.4.6 — Nested Interval Theorem. If {In } is a sequence of nonempty

bounded
T∞
closed intervals of the real line such that In+1 ⊂ In for all n ∈ {1, 2, . . . }, then
n=1 In is not empty.

Proof. Suppose In = [an , bn ] and let E be the set of all an . Then E is not empty
and is bounded above by b1 . Let s be the supremum of E. Since s is the supremum of
E, it is clear that an ≤ s for all n. Since every bn is an upper bound of E, it follows that
s ∈ [an , bn ] = In for every n ∈ N. The result follows.

Although the following result is very important, we’ll call it a lemma since it will be
used to prove a much more general result.

Lemma 1.4.7 The closed interval [a, b] ⊂ R is compact.

Proof. Suppose there is an open cover {Gλ } of [a, b] which contains no finite
subcover. Let c = (a + b)/2 and consider the intervals [a, c] and [c, b]. It follows that at
least one of these intervals cannot be covered by a finite subcover of {Gλ }, otherwise
[a, b] itself would be covered. We then call this interval I1 , divide it in half and call
the piece that cannot be covered I2 . Continuing this process we get a sequence {In } of
subsets of [a, b] such that for all n ∈ N: In+1 ⊂ In , In is not covered by any finite subcover
of {Gλ } and diam In = 2−n (b − a).
I1 I3 I2

a x c b
1.4 Compact Sets 9

By the nested interval theorem, there is a number x ∈ [a, b] such that x ∈ In for all n.
Since {Gλ } covers [a, b], there is a set Gλ such that x ∈ Gλ . Since Gλ is open, there is a
neighborhood B(x, r) of x that is entirely contained in Gλ .

In Gλ

x
B(x, r)

If we take n large enough such that 2−n (b − a) < r (which is possible since R is
archimedean) then we have that In ⊂ B(x, r) ⊂ Gλ . This is absurd since no finite subset
of {Gλ } covers In . This establishes the proof.

Theorem 1.4.8 If E and F are compact sets, then E × F is compact.

Proof. Let {Gλ } be an open cover of E × F. For each (a, b) ∈ E × F, we can

choose some λ such that (a, b) ∈ Gλ . Since Gλ is open, the point (a, b) is contained in
some open box U(a,b) ×V(a,b) ⊂ Gλ , where U(a,b) ⊂ E and V(a,b) ⊂ F.
Suppose we fix a and vary b. Then for every point (a, b) we find that the point is
contained in an open box in the product E × F, and that box is then itself the product of a
subset of E with a subset of F. Proceeding in this manner, we observe that the collection
of sets {V(a,b) }b∈F is an open cover of F. Since by assumption F is compact, we can find
a finite cover {V(a,b j (a)) } of F that consists of finitely many open sets containing points
{(a, b j (a))}.
T
Now let Ua = j U(a,b j (a)) . Since Ua is the intersection of finitely many open sets, it
is itself open. Since E is compact, there are finitely many ai such that {Uai } forms an
open cover of E. Then it follows that the collection of sets {Uai ×V(ai ,b j (ai )) } (for all i
and j) is a finite subcover of E × F, hence E × F is compact.

Corollary 1.4.9 If E1 , E2 , . . . , En are compact sets, then E1 × . . . × En is compact.

In fact, a result much stronger than this is true: the arbitrary product of compact sets
is compact. That is called Tychonoff’s theorem and it is way out of our scope.
As a quick corollary we get our most powerful result yet: the Heine-Borel Theorem.

Theorem 1.4.10 — Heine-Borel. A subset of Rn is compact if and only if it is closed

and bounded.

Proof. If E ⊂ Rn is bounded, then it is contained in a box of the form [a1 , b1 ] ×

. . . × [an , bn ] for some suitable ai ’s and bi ’s. Since this box is compact, if E is closed
then E is compact. The converse was already proved in a previous theorem.
 10 Chapter 1. Metric Spaces

1.5 Connected Sets

As we’ll see, unlike compactness, connectedness is a very intuitive notion for metric
spaces.
Definition 1.5.1 — Connected Set. Let E be a metric space. A separation of E is
a pair U, V of disjoint nonempty open subsets of E whose union is E. The set E is
said to be connected if there does not exist a separation of it.

In every metric space M, we have that M and the empty set are both open and
closed (independent of the metric chosen). Connected metric spaces have the remarkable
property that the only sets that are both open and closed are M and ∅.

Theorem 1.5.1 Let M be a connected metric space. Then the only sets that are both
open and closed are M and ∅.

Proof. Suppose A is a proper non-empty subset of M that is both open and closed.
Since A is closed, M \ A is open and since A is a proper subset of M, M \ A 6= ∅. Clearly
(M \ A) ∪ A = M and (M \ A) ∩ A = ∅, so M is disconnected. We conclude that the only
closed and open sets on a connected metric space are the empty set and itself.

We now prove that every interval of the real line and the real line itself are connected.

Theorem 1.5.2 A subset E of the real line R is connected if and only if it is an

interval (bounded or not).

Proof. If E contains a single point, then it is connected. Suppose E contains

two distinct points a < b. We prove that every x such that a < x < b belongs to E.
Otherwise, E would be the union of the nonempty disjoint sets U = E ∩ (−∞, x) and
V = E ∩ (x, +∞), both of which are open in E. From this property, we deduce that E is
necessarily an interval. Indeed, let c ∈ E, p = inf E and q = sup E. If p = −∞, then for
every x < c, there is y < x belonging to E and hence x ∈ E so that (−∞, c) is contained
in E. If p is finite and p < c, for every x such that p < x < c there is y ∈ E such that
p < y < x, hence again x ∈ E, so that E contains the interval (p, c]. Similarly, one shows
that E contains [c, q) if q > c. In any case, it follows that E contains the interval (p, q).
Conversely, suppose E is a nonempty interval of center a and radius b, both elements
of R. Suppose U,V constitute a separation of E. Without loss of generality, suppose
x ∈ U, y ∈ V and x < y. Let z = supU ∩ [x, y]. If z ∈ U, then z < y and there is an interval
[z, z + h) contained in [x, y] and in U, which contradicts the definition of z. On the other
hand, if z ∈ V then x < z, and there is an interval (z − h, z] ⊂ V ∩ [x, y], which again
contradicts the definition of z. Hence z cannot belong to U nor to V , which is absurd
since the closed set [x, y] is contained in E. Hence E is connected.
 1.6 Continuous Functions 11

1.6 Continuous Functions

The concept of a continuous function is essential to almost every every branch of
mathematics. It is truly remarkable that this property of functions generalizes so well to
topological and metric spaces.
Definition 1.6.1 — Continuous Function. Let A and B be metric spaces. A function
f : A → B is said to be continuous if for every open subset E of B, the set f −1 (E) is
an open subset of A. It should be noted that this property does not depend only on f
but too on the metrics of A and B.

Alternatively, we could also define continuity as a property related to closed sets.

If f is continuous and E is closed, A \ f −1 (E) = f −1 (B \ E) is open, so that f −1 (E) is
closed. The converse is just as easy.
Definition 1.6.2 — Continuity at a Point. Let x be a point of A. We say that f :
A → B is continuous at the point x if for each neighborhood V of f (x) there is a
neighborhood U of x such that U ⊂ f −1 (V ).

The definition given above is equivalent to the following:

"A function f : A → B is continuous at p ∈ A if for every ε > 0, there exists a real
number δ > 0 such that for all x ∈ A:

dA (x, p) < δ =⇒ dB ( f (x), f (p)) < ε."

I used indices on the metrics to indicate the sets where they are defined.

Theorem 1.6.1 A function is continuous if it is continuous at every point of its

domain.

Proof. Suppose f is continuous. Let x ∈ A and let V be a neighborhood of f (x).

Then the set f −1 (V ) is open and so contains a neighborhood U of x such that f (U) ⊂
f ( f −1 (V )) ⊂ V .
Suppose now f is continuous at every point. Let E be a open subset of B. Is x is a point
in f −1 (E) then there is y ∈ E such that y = f (x). Since f is continuous in x, then there
is a neighborhood U of x such that f (U) ⊂ E. This implies U ⊂ f −1 (E) so that f −1 (E)
is open. The result follows.

Theorem 1.6.2 Let f : A → B and g : B → C be continuous. Then the composition

g ◦ f : A → C is continuous.

Proof. Let E be an open set in C. Since g is continuous, g−1 (E) is open in

B. Since f is continuous, f −1 (g−1 (E)) = (g ◦ f )−1 (E) is open in A. Hence g ◦ f is
continuous.

We turn now to properties related to continuous functions with compact domain.

12 Chapter 1. Metric Spaces

Theorem 1.6.3 The image of a compact set under a continuous function is compact.

Proof. Let E be compact and f be continuous. Suppose {Gλ } is an open cover of

f (E). Of course the set { f −1 (Gλ )} is an cover of E. Since f is continuous this is an
open cover. But because E is compact we have that

E ⊂ f −1 (Gλ1 ) ∪ . . . ∪ f −1 (Gλn ).

It’s now clear that the set {Gλ1 , . . . , Gλn } is a finite subcover of f (E).

Theorem 1.6.4 — Extreme Value Theorem. Let A be a compact set and f : A → R

be a continuous function. Then there exist points p, q ∈ A such that

f (p) ≤ f (x) ≤ f (q)

for all x ∈ A.

Proof. Since A is compact and f is continuous, f (A) is compact, hence closed and
bounded. Theorem 2.3.9 then implies the result.

Last, but not least we have theorems that relates connected sets with continuous
functions.
Theorem 1.6.5 The image of a connected set under a continuous function is con-
nected.

Proof. Let E be a connected set in the domain of f . Suppose f (E) is not connected.
Then there are sets U,V that make a separation of f (E). Since f is continuous, f −1 (U)
and f −1 (V ) are open, hence so are E ∩ f −1 (U) and E ∩ f −1 (V ). A possible element
of f −1 (U) ∩ f −1 (V ) would its image would be a element of both U and V , which
is impossible. The sets f −1 (U) and f −1 (V ) are non-empty too since U and V are
non-empty. Lastly, its clear that E ⊂ f −1 (U) ∪ f −1 (V ) since if x ∈ E, then f (x) ∈ U
or f (x) ∈ V and that is the exact definition of x ∈ f −1 (U) ∪ f −1 (V ). This implies
E = (E ∩ f −1 (U)) ∪ (E ∩ f −1 (V )), hence E ∩ f −1 (U) and E ∩ f −1 (V ) constitute a
separation of E, which is absurd!

Theorem 1.6.6 — Intermediate Value Theorem. Let A be a connected set and

f : A → R be a continuous function. If p and q are two points of A and if a is a point
of R such that f (p) < a < f (q), then there exists a point r of A such that f (r) = a.

Proof. Since f (A) is connected, Theorem 2.5.2 implies the result.

 1.7 Normed Spaces 13

1.7 Normed Spaces

In general, there are no algebraic operations defined on a metric space, only a distance
function. Most of the spaces that arise in analysis are vector spaces, and the metrics on
them are usually derived from a norm, which gives the “length” of a vector.
Definition 1.7.1 — Normed Vector Spaces. A normed vector space is a (real or
complex) vector space E together with a function k · k : E → R, called a norm on E,
such that for all x, y ∈ E and k ∈ R:
• (Positivity) kxk > 0 if x 6= 0 and k0k = 0.
• (Linearity) kkxk = |k| kxk.
• (Triangle Inequality) kx + yk ≤ kxk + kyk.

If (E, k · k) is a normed vector space, then d : E × E → R defined by d(x, y) = kx − yk

is clearly a metric on E. Note that the metric associated with a norm has the additional
properties that for all x, y, z ∈ E and k ∈ C:
d(x + z, y + z) = d(x, y), d(kx, ky) = |k| d(x, y),
which are called translation invariance and homogeneity, respectively. These properties
do not even make sense in a general metric space since we cannot add points or multiply
them by scalars.

Theorem 1.7.1 Let f : E → F be a linear function between normed spaces and N be

the unit neighborhood of 0. Then the following conditions are equivalent:

i) f is continuous;
ii) f is continuous at 0 ∈ E;
iii) The restriction f |N of f to N is bounded. That is, supx∈N k f (x)k < ∞.

Proof. That i) implies ii) is clear.

Supposing f is continuous at 0,

Theorem 1.7.2 Let f : X → Y , where X, Y are normed vector spaces and X is finite-
dimensional, be a linear function. Then f is continuous.

Actually, we’ll prove a particular case of this result for now. We’ll assume that the
norm defined on X is the 1-norm k·k1 . (The metric without indices will be the one
defined on Y .)
Proof. Let e1 , . . . , en be a basis of X. Then, for x = x1 e1 + . . . + xn en ∈ X and
y = y1 e1 + . . . + yn en ∈ X we have:
n n
k f (x) − f (y)k = ∑ (xi − yi ) f (ei ) ≤ ∑ |xi − yi | k f (ei )k .
i=1 i=1
14 Chapter 1. Metric Spaces

Let ε > 0 be given and let M = max1≤i≤n k f (ei )k. If we define δ = ε/M, then for all x
and y with kx − yk1 < δ :
n n
k f (x) − f (y)k ≤ ∑ |xi − yi | k f (ei )k ≤ M ∑ |xi − yi | = M kx − yk1 < ε.
i=1 i=1

Hence, f is continuous.

Definition 1.7.2 Let X be a vector space and k · ka , k · kb be any norms. We say that
k · ka and k · kb are equivalent the metrics induced by them are equivalent. That is, if
there exist positive constants c, C such that for all x ∈ X:

c kxka ≤ kxkb ≤ C kxka .

We now prove that, in a finite dimensional vector space, the topology generated by
every norm is unique. For that we’ll need the following lemma.

Lemma 1.7.3 Let f : X → R, f (x) = kxk is continuous under the metric induced by
k·k1 on X if X is finite-dimensional.

Proof. Let M = max1≤i≤n kei k and δ = ε/M. Then:

|kxk − kyk| ≤ kx − yk
n
≤ ∑ |xi − yi | kei k
i=1
n
≤ M ∑ |xi − yi |
i=1
≤ M kx − yk1
< ε,

if kx − yk1 < δ .

Theorem 1.7.4 — Equivalence of Norms. Let X be a finite-dimensional vector

space. Then all norms are equivalent.

Proof. Since norm equivalence is transitive, it is sufficient to show that every norm
is equivalent to some fixed norm. Let k · k1 = ∑ni=1 |xi | be that fixed norm.
If x = 0 the result is trivial, so let’s assume x 6= 0 and divide the inequality by kxk1 .
We only need to prove that
c ≤ kuk ≤ C
is true for all u ∈ X such that kuk1 = 1.
The unit sphere
S = {x ∈ X | kxk1 = 1}
 1.8 Exercises 15

is closed and bounded, so it is compact. we have then that f (x) = kxk must attain it’s
bounds on S.
Let
c = min kuk and C = max kuk .
u∈S u∈S

The result follows.

Actually, it isn’t obvious that S is compact since Heine-Borel was only proved on
Rn . However, note that the function f : Rn → X, f (x1 , . . . , xn ) = x1 e1 + . . . + xn en is a
 continuous (since it is linear) bijection. The set S0 = {x ∈ Rn | kxk1 = 1} is compact
(Heine-Borel) and S is the image of S0 by f . Since the continuous image of a compact set
is compact, the result follows.

Exercise 1.1 Use the equivalence of norms to prove Theorem 2.7.2 for any norm
defined on X.

Theorem 1.7.5 — Heine-Borel for Normed Spaces. Let X be a finite dimensional

normed vector space and let E be a subset of X. Then E is compact if, and only if it is
closed and bounded.

Proof. We’ll denote by k·k an arbitrary norm of either X or Rn .

Let E be a closed and bounded set. Consider the function f : Rn → X given by
f (x1 , . . . , xn ) = x1 e1 + . . . + xn en . Since f is continuous (as it is linear), f −1 (E) is
closed. If x, y ∈ f −1 (E), then kx − yk ≤ c1 kx − yk1 = c1 k f (x) − f (y)k1 for some c1 > 0.
Since E is bounded and f (x), f (y) ∈ E, k f (x) − f (y)k1 ≤ c2 k f (x) − f (y)k ≤ c2 M for
some M > 0 and hence f −1 (E) is bounded. As Heine-Borel holds in Rn , f −1 (E) is
compact. Because f is bijective, E = f ( f −1 (E)) is the continuous image of a compact
set and hence compact.
The converse was already proved in an earlier result.

1.8 Exercises
p If d is a metric, prove that so are ρ1 (x, y) = d(x, y)/(1 + d(x, y)) and
Exercise 1.2
ρ2 (x, y) = d(x, y).

Exercise 1.3 — Rn is Second-Countable. A collection B of open subsets of a

metric space M is said to be a basis if every open subset of M is the union of a
subcollection of B. We say that a metric space is second-countable if it has a
countable basis.
Show that Rn is second-countable.
Hint: Consider the neighborhoods centered at points with rational coordinates and
with rational radius.
16 Chapter 1. Metric Spaces

Exercise 1.4 — Urysohn’s Lemma. Let A and B be non-empty disjoint closed

subsets of a metric space M. Show that the function f : M → R,

d(x, B)
f (x) =
d(x, A) + d(x, B)

is continuous, has value 1 for all x ∈ A, has value 0 for all x ∈ B and satisfies
0 ≤ f (x) ≤ 1 for all x ∈ M.
As metric spaces behave much better than topological spaces, it is useful to know
when a topological space is also a metric space. A famous result, called Urysohn
Metrization Theorem states that every normal, second-countable topological space is
in fact a metric space. Urysohn’s lemma is used in the proof of this result.

Exercise 1.5 — Metric Spaces are Normal. Let A and B be non-empty disjoint
closed subsets of a metric space. Prove that there exist open sets E, F such that
A ⊂ E, B ⊂ F and E ∩ F = ∅.

Exercise 1.6 Check the proof of Theorem 2.3.9 again and then read the following
statement.
"Let E be a nonempty subset of R. If y = sup E is finite and δ > 0, then there exists
some x in E such that y − δ < x < y."
I affirm that it is wrong. Can you tell why?

Exercise 1.7 Let f and g be continuous functions from M1 to M2 , where M1 and M2

are metric spaces. If E is dense in M1 , prove that f (E) is dense in M2 . Moreover,
show that if f (x) = g(x) for all x ∈ E, then f (x) = g(x) for all x ∈ M1 .

Exercise 1.8 Let M and N be a metric spaces and f : M → N be a function. Then f

is continuous if, and only if f (E) ⊂ f (E) for all subsets E of M.

Exercise 1.9 — Lebesgue Number. Let M be a compact metric space and let {Gλ }
be an open cover of M. Prove that there is a number δ > 0, called Lebesgue number
of M relative to {Gλ }, such that every subset of M with diameter less than δ is
contained in some element of {Gλ }.

Exercise 1.10 — Lindelöf Covering Theorem. Let E be a subset of Rn and let

{Gλ } be an open cover of E. Prove that there exists a countable subcover {Gλi } of
A.
1.8 Exercises 17

Exercise 1.11 Prove that the finite union of compact sets and the arbitrary intersec-
tion of compact sets is compact.

Exercise 1.12 — The Real Projective Space is Compact. Define an equivalence

relation of Rn+1 \ {0} by

x ∼ y ⇐⇒ x = ty, for some non-zero real number t.

The real projective space RPn is the set of all equivalence classes of this equivalence
relation.
Geometrically, two points in Rn+1 \ {0} are equivalent if they lie on the same line
through the origin, so RPn can be interpreted as the set of all lines through the origin
in Rn+1 . As each line crosses the sphere Sn = {x ∈ Rn+1 | kxk = 1} exactly twice,
this suggests that we define the following equivalence relation on Sn :

x ∼ y ⇐⇒ x = ±y.

Let Sn / ∼ be the set of all equivalence classes of this equivalence relation. As

suggested, there is a bijection from RPn to Sn / ∼. Show that in fact there is a
continuous bijection f from RPn to Sn / ∼ such that f −1 is also continuous. Use this
result to prove that RPn is compact.

Exercise 1.13 Let A and B be subsets of the real line and define the set

A + B = {a + b ∈ R | a ∈ A, b ∈ B}.

If A and B are open, A + B is open too? What if only A is open?

If A and B are closed, A + B is closed too?
If A and B are compact, A + B is compact too?

Exercise 1.14 — Every Non-empty Perfect Set is Uncountable. A subset of a

metric space that is closed and has no isolated points is denominated perfect. Show
that every non-empty perfect set is uncountable. Conclude that R is uncountable.

Exercise 1.15 — R \ Q is Uncountable. Prove that the set of all irrational numbers
R \ Q is uncountable.

Exercise 1.16 — There Exist Transcendental Numbers. A complex number which

is not algebraic is said to be transcendental. Show that there exist transcendental
numbers. Moreover, show that the set of all transcendental numbers is uncountable.
18 Chapter 1. Metric Spaces

It is curious that "almost" every complex number is transcendental and yet the only
transcendental numbers that most people recognize are π and e. Even these numbers
are hard to show that they are not algebraic!

Exercise 1.17 — Cantor Set. Let C0 be the closed interval [0, 1].

C0 :
0 1

C1 will be the set that results when the open middle third is removed. That is,
C1 = [0, 1/3] ∪ [2/3, 1].

C1 :
0 1/3 2/3 1

Now, construct C2 in a similar way, removing the open middle third of each compo-
nent of C1 . Then we have that C2 = [0, 1/9] ∪ [2/9, 1/3] ∪ [2/3, 7/9] ∪ [8/9, 1].

C2 :
0 1/9 2/9 1/3 2/3 7/9 8/9 1

If we continue this process inductively, then for each n = 0, 1, 2, . . . we get a set Cn

consisting of 2n closed intervals each having length 1/3n . Finally, we define the
Cantor set C to be the intersection
∞
\
C= Cn .
n=0

Since C is an intersection of closed sets, C is closed and hence compact. Prove that
C does not contain intervals and does not contain isolated points. Lastly, show that
the Cantor set is uncountable.

Exercise 1.18 Let X ⊂ Rn . Show that there exists a countable set Y ⊂ Rn such that
Y = X.

Exercise 1.19 — Banach Fixed Point Theorem. Let X be a compact metric space
and φ : X → X be a continuous function such that d(φ (x), φ (y)) < d(x, y) for all
x 6= y. Then φ has a unique fixed point. That is, a point x0 such that φ (x0 ) = x0 .

Exercise 1.20 — Every Convex Subset is Connected. A subset E of Rn is said

to be convex if for all p, q ∈ E, the point (1 − t)p + tq exists in E for all t ∈ (0, 1).
1.8 Exercises 19

(That is, every point in the segment that connects p to q is in E.)

Let A and B be separated subset of some Rn , suppose a ∈ A, b ∈ B, and define

p(t) = (1 − t)a + tb

for t ∈ R. Prove that there exists t0 ∈ (0, 1) such that p(t0 ) ∈

/ A ∪ B. Conclude that
every convex set is connected.

Exercise 1.21 — Operator Norm. Let f : X → Y be a linear function, where X and

Y are normed vector spaces. Consider the set

F = {c ≥ 0 | k f (x)k ≤ c kxk for all x ∈ X}.

Show that F is closed and non-empty. The infimum of this set is called the operator
norm k f kop of f . Prove that this is, in fact, a norm.

Exercise 1.22 — Riesz’s Lemma. Let E be a closed proper subspace of a normed

vector space X. Given 0 < r < 1, show that there is x ∈ X with kxk = 1 such that

kx − yk ≥ r

for all y ∈ E.
 CHAPTER
Sequences and Series
2
S WE ’ LL SEE , most of the important concepts in analysis are well described in
A terms of sequences. As before, whenever it is possible and enlightening, we’ll
study sequences in it’s more general setting possible. This will usually be metric or
normed spaces.
In the first chapter, we saw (quite informally) that a sequence is a function x : N → X,
where X is an arbitrary set. We denote the element x(n) as xn and write as (xn ) the
sequence itself.

2.1 Convergent Sequences

Definition 2.1.1 — Convergent Sequence. A sequence (xn ) defined on a metric
space M is said to converge if there exists a point x ∈ M with the following property:
for every neighborhood Nx of x, there is an integer n0 such that n > n0 implies xn ∈ Nx .
If such point does not exist, then (xn ) is said to diverge.

If (xn ) is convergent, we way that (xn ) converges to x or that x is the limit of (xn )
and write this as x = lim xn or xn → x.
n→∞
The preceding definition can be restated in two very useful ways:

• (xn ) converges to x if every neighborhood of x contains every xn with the possible

exception of a finite number of points.
• (xn ) converges to x if for every ε > 0 there is an integer n0 such that n > n0 implies
d(x, xn ) < ε.

A priori, nothing prevents a sequence to converging to more than one point. The
following theorem shows that this is not the case.
22 Chapter 2. Sequences and Series

Theorem 2.1.1 A sequence (xn ) can converge to at most one point.

Proof. Suppose xn → a and xn → b. Then, for every ε > 0, there are integers n0
and m0 such that, if n > n0 we have that d(a, xn ) < ε/2 and if n > m0 we have that
d(b, xn ) < ε/2. Then, if n > max{n0 , m0 }, the triangular inequality implies
d(a, b) ≤ d(a, xn ) + d(b, xn ) < ε.
It follows that a = b.

Theorem 2.1.2 A point x ∈ M is a limit point of M if, and only if there is a sequence
(xn ) of points in M \ {x} such that xn → x.

Proof. Suppose x is a limit point of M and let Nk be a neighborhood of x with

radius 1/k. Since x is a limit point, N1 \ {x} is not empty. Let x1 be one of its elements.
Similarly, for all n ∈ N, let xn be a element of Nn \ {x}. It follows that (xn ) converges to
x. The converse is trivial.

We say that the set of all points xn is called the range of (xn ) and it is denoted {xn }.
When {xn } is bounded we’ll typically abuse language and say that the sequence itself
is bounded. You should notice the similarity between the range of a sequence and the
range of a function.

Theorem 2.1.3 If (xn ) converges, then {xn } is bounded. Moreover, the converse does
not hold.

Proof. If xn → x, then there is an integer n0 such that n ≥ n0 implies d(xn , x) < 1.

Let
r = max{1, d(x1 , x), d(x2 , x), . . . , d(xn0 , x)}.
Then d(xn , x) ≤ r for all n ∈ N.
The sequence xn = (−1)n is a counter-example to the converse since it is bounded
and divergent.

Theorem 2.1.4 A sequence of points zn = (xn , yn ) defined on the Cartesian product

M1 × M2 endowed with the maximum metric converges to z = (x, y) ∈ M1 × M2 if,
and only if xn → x and yn → y.

Proof. Let d1 be the metric defined on M1 , d2 be the metric defined on M2 and

d be the maximum metric on M1 × M2 . Let ε > 0 and assume zn → z = (x, y). By our
assumption, there is a natural number n0 such that n > n0 implies
d(zn , z) < ε.
Since d1 (xn , x) ≤ d(zn , z) and d2 (yn , y) ≤ d(zn , z) it follows that
d1 (xn , x) ≤ d(zn , z) < ε and d2 (yn , y) ≤ d(zn , z) < ε.
2.1 Convergent Sequences 23

Hence xn → x and yn → y.
Conversely, assume xn → x and yn → y. For every ε > 0 there are integers n1 and
n2 such that n > n1 implies d1 (xn , x) < ε and n > n2 implies d2 (yn , y) < ε. Taking
n0 = max{n1 , n2 } we have that n > n0 implies d(zn , z) < ε and hence zn → z.

Corollary 2.1.5 A sequence of points zn = (xn , yn ) defined on the Cartesian product

X1 × X2 of normed vector spaces of finite dimension converges to z = (x, y) ∈ X1 × X2
if, and only if xn → x and yn → y.

Proof. Since all norms are equivalent on finite dimensional normed spaces, the
particular result of Theorem 3.1.4 becomes a general result in finite dimensional normed
spaces.

The two preceding results clearly generalize to products of any finite number of
spaces.
Besides understanding the topological aspects of sequences, we can also study some
algebraic properties.

Theorem 2.1.6 Let (xn ) and (yn ) be sequences defined on a normed vector space
X and let (pn ) be a sequence in X’s field. If xn → x, yn → y and pn → p, then the
following holds.
1 1
lim (xn + yn ) = x + y, lim (pn xn ) = px and lim = .
n→∞ n→∞ n→∞ pn p

The last property is true provided that pn 6= 0 for all n ∈ N.

Proof. For every ε > 0 there are integers n1 and n2 such that

n > n1 =⇒ kxn − xk < ε/2 and n > n2 =⇒ kyn − yk < ε/2.

The triangular inequality then implies

k(xn + yn ) − (x + y)k ≤ kxn − xk + kyn − yk < ε

for all n > max{n1 , n2 }.

For the second result note that

kpn xn − pxk = kpn (xn − x) + (pn − p)xk ≤ |pn | kxn − xk + |pn − p| kxk .

Since xn → x and pn → p, for every ε > 0 there are integers n1 , n2 such that

n > n1 =⇒ kxn − xk < ε/2(max |pn |) and n > n2 =⇒ |pn − p| < ε/2 kxk .

Then, if n > max{n1 , n2 }:

ε|pn | ε
kpn xn − pxk < + < ε.
2(max |pn |) 2
24 Chapter 2. Sequences and Series

Lastly, choosing n0 such that |pn − p| < |p|/2 if n > n0 , we see that∗

1
|pn | > |p| for all n > n0 .
2
Given ε > 0, there is an integer n1 such that n > n1 implies
1
|pn − p| < |p|2 ε.
2
Hence, for n > max{n0 , n1 },

1 1 pn − p 2
− = < 2 |pn − p| < ε.
pn p pn p |p|

Theorem 2.1.7 Let (pn ) be a sequence of real numbers such that pn ≥ 0 for all n > n0 ,
where n0 is a fixed integer. If (pn ) converges to p, then p ≥ 0.

Proof. If p < 0, then there is an integer m0 such that

p
|pn − p| < −
2
for all n > max{n0 , m0 }. Hence pn − p < −p/2 and then pn < p/2 < 0, which is absurd!
The result follows.

Corollary 2.1.8 Let (pn ) and (qn ) be sequences of real numbers such that pn ≥ qn
for all n > n0 , where n0 is a fixed integer. If (pn ) converges to p and (qn ) converges
to q, then p ≥ q.

Proof. Apply Theorem 3.1.7 to the sequence (pn − qn ).

Definition 2.1.2 — Monotonicity. A sequence (pn ) of real numbers is said to be

increasing if pn+1 ≥ pn for all n ∈ N. Similarly, we say that (pn ) is decreasing if
pn+1 ≤ pn . If the strict inequality holds we say that (pn ) is strictly increasing or
decreasing. In every case we say that (pn ) is monotonic.

Suppose (pn ) is an increasing sequence. Then for every n ∈ N we have that pn ≥ n.

The proof follows readily by induction. This fact is a useful property of increasing
sequences. Another useful property is the following.

∗ Combining |pn − p| ≥ |p| − |pn | and 12 |p| > |pn − p| we get the desired result.
 2.2 Subsequences and Sequential Compactness 25

Theorem 2.1.9 — Monotone Convergence Theorem. Let (pn ) be a monotonic

sequence of real numbers. Then (pn ) is convergent if, and only if it is bounded.

Proof. I’ll suppose pn+1 ≥ pn since all the other cases follow readily or are
analogous to it. If (pn ) is bounded, then {pn } has a least-upper-bound. Let p = sup{pn }.
For all ε > 0 there is an integer n0 such that

p − ε < pn0 ≤ p.

Otherwise the leftmost inequality would imply that p − ε is an upper bound of {pn }
smaller than p. The rightmost inequality follows from the fact that p is an upper bound
of {pn }. Since (pn ) is monotonic, for all n > n0 the inequality pn ≥ pn0 holds. Hence,
for all n > n0
p − ε < pn ≤ p < p + ε.
It follows that (pn ) converges (to p). The converse was proved in Theorem 3.1.3.

2.2 Subsequences and Sequential Compactness

Definition 2.2.1 — Subsequence. Let (nk ) be a strictly increasing sequence of
positive integers. Given a arbitrary sequence (xn ), the sequence (xnk ) is said to be a
subsequence of (xn ). If (xnk ) converges, its limit is called a subsequential limit of
(xn ).

Theorem 2.2.1 A sequence (xn ) converges to x if, and only if every subsequence
(xnk ) of (xn ) converges to x.

Proof. If there was a subsequence that didn’t converge to x, then there will be
a neighborhood of x that has infinite terms of the subsequence outside of it. So this
neighborhood has infinite terms of the sequence outside of it, and then the sequence does
not converge to x.
Since every sequence is a subsequence of itself, the other direction is trivial.

Some books say that a topological space is sequentially compact if every sequence
has a convergent subsequence. The following theorem shows that every compact metric
space is sequentially compact.

Theorem 2.2.2 Let (xn ) be a sequence in a compact metric space M. Then some
subsequence (xnk ) converges.

Proof. If {xn } is finite, there is a point x ∈ {xn } and a strictly increasing sequence
of positive integers (nk ) such that

xn1 = xn2 = xn3 = . . . = x.

26 Chapter 2. Sequences and Series

The sequence (xnk ) then converges to x.

If {xn } is infinite, then Bolzano-Weierstrass implies {xn } has a limit point x. Theorem
3.1.2 then shows that there is a sequence of points of {xn } that converges to x. Let n1
be the smallest index among the elements of this sequence. Having chosen n1 , . . . , nk−1
let nk be the smallest index among the elements of this sequence that is bigger than
n1 , . . . , nk−1 . The sequence (xnk ) then converges to x.

Corollary 2.2.3 Let (xn ) be a bounded sequence in Rn . Then some subsequence

(xnk ) converges.a
a Here the n in (xn ) is an index while the n in Rn is a fixed number.

Proof. Since {xn } is a closed and bounded subset of Rn , {xn } is compact. The
previous theorem implies the result.

Actually, in metric spaces the notions of compactness and sequential compactness

are equivalent. We just need to show that every sequentially compact metric space is
also compact. The following lemmata will be of great aid in this task. You should notice
the resemblance of the first lemma to Exercise 2.10.

Lemma 2.2.4 Let M be a sequentially compact metric space and let {G p } be an open
cover of M. Then, there exists a real number δ > 0 such that every neighborhood
with radius less than δ is contained in some element of {G p }.

Proof. Let Nn (x) be a neighborhood of x with radius 1/n. If the theorem is not true,
for every n ∈ N some neighborhood Nn (xn ) is not contained in any element of the cover.
Since M is sequentially compact, we can find a subsequence (xnk ) which converges to
x ∈ M.
Now, x is contained in some G p0 . Since each G p is open, there is an positive integer
m such that Nm (x) ⊂ G p0 . Since xnk → x, for all k > k0 d(x, xnk ) < 1/2m. Since nk ≥ k,
taking k > max{k0 , 2m} We have Nnk (xnk ) ⊂ Nm (x) ⊂ G p0 .

G p0
Nm (x)

m−1
xnk
x

This contradicts our choice of (xn ). The result follows.

2.2 Subsequences and Sequential Compactness 27

Lemma 2.2.5 Let M be a sequentially compact metric space. For all ε > 0 there is a
finite open cover of M which consists only of neighborhoods with radii ε.

Proof. Consider N(x) to be a neighborhood of x with radius ε. If M is finite, then

the result is obvious. Suppose M is infinite and the result does not hold. Construct a
sequence (xn ) as follows. Let x1 be any element of M. Having chosen x1 , . . . , xn , pick
any member of M \ nk=1 N(xk ) to be xn+1 . If for some n the set M \ nk=1 N(xk ) is empty,
S S

them the result is proved. Otherwise, note that d(xn , xm ) ≥ ε for every n 6= m. Hence
(xn ) has no convergent subsequence which is absurd!

 Why does d(xn , xm ) ≥ ε imply (xn ) has no convergent subsequence? Try to figure this
out by yourself. In every case, this should be clear to you by the end of the next section.

Theorem 2.2.6 Let M be a sequentially compact metric space. Then M is compact.

Proof. Let {G p } be an open cover of M. Let δ be the number of Lemma 3.2.4.

Lemma 3.2.5 then implies that M can be covered by a finite number of neighborhoods
with radii ε = δ . Since each neighborhood Ni is contained in some G pi , the set of all
G pi is a finite subcover of M.

Let (pn ) be a sequence of real numbers. If for all real numbers M there is an integer
n0 such that pn > M whenever n > n0 we say that pn → +∞. Analogously if pn < M
whenever n > n0 we say that pn → −∞.
We’re actually abusing a little of the notation here since the symbol → was used
before to denote converging sequences and the kinds of sequences we just defined clearly
diverge. This is in no way a change to the concept of convergence.
Definition 2.2.2 — Limit Superior/Inferior. Let (pn ) be a sequence of real numbers
and E ⊂ R be the set of all subsequential limits of (pn ). We define the limit superior
and the limit inferior of (pn ) to be:

lim sup pn = sup E and lim inf pn = inf E.

n→∞ n→∞

Of course there are sequences (pn ) for which lim supn→∞ pn 6= lim infn→∞ pn . A sim-
ple example is the sequence pn = (−1)n . In this case we have that lim supn→∞ (−1)n = 1
and lim infn→∞ (−1)n = −1.
Another way to state Theorem 3.2.1 for sequences of real numbers is as follows:
(pn ) converges if, and only if

lim sup pn = lim inf pn = p,

n→∞ n→∞

where p is finite. In this case we have that pn → p.

28 Chapter 2. Sequences and Series

Exercise 2.1 Prove that the set E that was just defined is closed in R. That is, there
are subsequences (pnk ) and (pmk ) such that

lim sup pn = lim pnk and lim inf pn = lim pmk .

n→∞ k→∞ n→∞ k→∞

We now prove an analogous result to Corollary 3.1.8 to the limit superior and inferior.

Theorem 2.2.7 Let (pn ) and (qn ) be sequences of real numbers such that pn ≥ qn for
all n > n0 where n0 is a fixed positive integer. Then

lim sup pn ≥ lim sup qn and lim inf pn ≥ lim inf qn .

n→∞ n→∞ n→∞ n→∞

Proof. Assume the theorem is false. As you proved in Exercise 2.1 (if you didn’t it
may be helpful to check the solution in the end of the book), there is a subsequence (qnk )
such that
lim qnk = lim sup qn .
k→∞ n→∞
Hence,
lim sup pn < lim qnk .
n→∞ k→∞

Since lim supn→∞ pn is the biggest subsequential limit of (pn ), we have that:

lim pnk < lim qnk .

k→∞ k→∞

Taking ε = 12 (limk→∞ qnk − limk→∞ pnk ) we see that there is an positive integer k0 for
which
pnk < qnk
whenever k > k0 . This contradiction estabilishes the theorem. The proof for the limit
inferior is analogous.

The following theorem provides a very useful method of calculating limits.

Theorem 2.2.8 Let (pn ) and (qn ) be sequences of real numbers. If 0 ≤ pn ≤ qn for
all n > n0 , where n0 is an positive integer, and qn → 0 then pn → 0.

Proof. This result is a quick corollary to the preceding theorem since

0 ≤ lim inf pn ≤ lim sup pn ≤ lim qn = 0.

n→∞ n→∞ n→∞

We conclude that lim infn→∞ pn = lim supn→∞ pn = 0 and hence pn → 0.

The following lemma will be useful in proving later results.

2.2 Subsequences and Sequential Compactness 29

Lemma 2.2.9 Let (xn ) be a sequence of real numbers. If a is a real number such that
a > lim supn→∞ xn , then there is an integer n0 such that xn < a whenever n > n0 .

Proof. If we had that xn ≥ a for infinitely many values of n, then the subsequence
(xnk ) of all xn such that xn ≥ a converges to a value x ≥ a > lim supn→∞ xn . This
contradicts the definition of the limit superior.

We shall now present some examples of convergent sequences. Unless explicitly

stated, it will be assumed that the metric space is the real field.
 Example 2.1 If p > 0, then lim 1/n p = 0. In fact, for all ε > 0 we can take any integer
n0 that is bigger than 1/ε 1/p (which exists since R is archimedean). Then, if n > n0 :
1 1 1
< p< = ε.
np n0 1/ε

It follows that 1/n p → 0. 

√ √
 Example 2.2 The sequence (
n
n) converges to 1. Take xn = n n − 1. The binomial
theorem then implies
n    
n n k n 2 n(n − 1) 2
n = (1 + xn ) = ∑ xn ≥ x = xn ,
k=0 k 2 n 2

since xn ≥ 0. Hence, for all n ≥ 2:

r
2
0 ≤ xn ≤ .
n−1
√
Since 1/ n − 1 → 0 (from the previous example), we conclude that xn → 0. 

 Example 2.3 If p > 0 and q ∈ R, then lim nq /(1 + p)n

= 0. Let k be an integer such
that k > q and
k > 0. As one can verify, the following inequality holds for binomial
coefficients: nk ≥ nk /kk . Hence,

n k nk k
 
(1 + p)n > p ≥ kp .
k k
And then,
nq kk q−k
0< < n .
(1 + p)n pk
Since q − k < 0, the result follows from Example 3.1. 

 Example 2.4 Taking q = 0 in the previous example we get that

lim xn = 0
n→∞

whenever |x| < 1. 

 30 Chapter 2. Sequences and Series

2.3 Cauchy Sequences

Unfortunately, it is not always easy to find the point x in the definition of a convergent
sequence. So we must find better methods to tell if a sequence converges or not.
Augustin-Louis Cauchy found an ingenious way.
Definition 2.3.1 — Cauchy Sequence. A sequence (xn ) in a metric space X is
said to be a Cauchy sequence if for every ε > 0 there is an integer n0 such that
d(xn , xm ) < ε whenever n, m > n0 .

Note that, although the limit point is explicitly involved in the definition of a conver-
gent sequence, it is not in the definition of a Cauchy sequence. As of now, convergent
sequences and Cauchy sequences are two completely different things, but it would be
wonderful if they shared properties. Fortunately, they do.

Theorem 2.3.1 Every convergent sequence in a metric space is Cauchy.

Proof. Let (xn ) be a sequence convergent to x. Then, for all ε > 0 there is an
integer n0 such that d(xn , x) < ε/2 whenever n > n0 . Hence,

d(xn , xm ) ≤ d(xn , x) + d(xm , x) < ε,

whenever n, m > n0 .

As one can easily verify, the sequence (1/n) defined on (0, 1] is Cauchy but diverges.
However, there exist some important cases where the converse of the preceding theorem
holds.
Theorem 2.3.2 Let M be a compact metric space. Then every Cauchy sequence in
M converges.

Proof. Let (xn ) be a Cauchy sequence in M. For any ε > 0 let n0 be an positive
integer such that
ε
d(xn , xm ) < for all n, m > n0 .
2
Theorem 3.2.2 implies that there is a subsequence (xnk ) that converges to some x ∈ M.
That is, for every ε > 0 there is an positive integer k0 such that
ε
d(xnk , x) < whenever k > k0 .
2
Since nk ≥ k, d(xnk , x) < ε/2 whenever nk > k0 . Let j be an integer such that j > n0 ,
j > k0 and j = nk for some k. We have then that

d(xn , x) ≤ d(xn , x j ) + d(x j , x) < ε.

Hence (xn ) converges (to x).

 2.4 Complete Metric Spaces 31

Lemma 2.3.3 Every Cauchy sequence is bounded.

Proof. Let (xn ) be a Cauchy sequence in a metric space M. For ε = 1 there is a

positive integer n0 such that d(xn , xm ) < 1 whenever n, m > n0 . That is, the set

{xn0 +1 , xn0 +2 , xn0 +3 , . . . }

is bounded, since its diameter is less than 1. It follows that the set

{xn } = {x1 , . . . , xn0 } ∪ {xn0 +1 , xn0 +2 , xn0 +3 , . . . }

is bounded.

Theorem 2.3.4 In Rn every Cauchy sequence converges.

Proof. Let (xn ) be a Cauchy sequence in Rn . It follows from the previous lemma
that {xn } is a bounded closed subset of Rn . Heine-Borel implies compacticity and hence
this theorem follows from the previous one.

We noted before that the sequence (1/n) defined on (0, 1) is Cauchy but diverges.
Our intuition may tell that the problem here is that (0, 1) lacks the point to which (1/n)
"should" converge. The next theorem shows roughly that this is always the case.

Theorem 2.3.5 Let (xn ) be a Cauchy sequence in a metric space M. If there is a

subsequence (xnk ) which converges, then (xn ) converges.

Proof. Let x = limk→∞ xnk . We’ll prove that xn → x. In fact, let ε > 0 be an
arbitrary real number. Since (xnk ) converges to x, there exists a positive integer k0
such that d(xnk , x) < ε/2 whenever k > k0 . Similarly, since (xn ) is a Cauchy sequence,
there exists a positive integer m0 such that d(xn , xm ) < ε/2 whenever n, m > m0 . Let
n0 = {k0 , m0 }. Taking k big enough such that nk > n0 we have that
ε ε
d(xn , x) ≤ d(xn , xnk ) + d(xnk , x) < + = ε,
2 2
whenever n > n0 . The result follows.

2.4 Complete Metric Spaces

The metric spaces in which every Cauchy sequence converges are so special that they
are given a particular name.
Definition 2.4.1 — Complete Metric Spaces. A metric space in which every
Cauchy sequence converges is said to be complete. A complete normed vector
space is said to be a Banach space.
32 Chapter 2. Sequences and Series

Thus, Theorem 3.3.2 and Theorem 3.3.3 say that all compact metric spaces and all
Euclidean spaces are complete. It also implies that every closed subset E of a complete
metric space M is complete.† A example of a metric space which is not complete is the
space of all rational numbers, with d(x, y) = |x − y| as a metric.

Theorem 2.4.1 Let M be a complete metric space. Let {Fn } be a collection of

non-empty closed subsets of M such that Fn+1 ⊂ Fn for all n ∈ N. If diam Fn → 0,
then
∞
\
F= Fn
n=1

consists of a single point.

Proof. For all n ∈ N pick a point xn ∈ Fn . The sequence (xn ) so defined has the
property that xn , xm ∈ Fn0 whenever n, m > n0 . Since diam Fn → 0, for all ε > 0 there is
a positive integer n0 such that diam Fn0 < ε and hence d(xn , xn ) < ε whenever n, m > n0 .
That is, (xn ) is a Cauchy sequence. Since M is complete, let x be the point to which (xn )
converges.
For any positive integer k we have that xn ∈ Fk whenever n ≥ k. That implies x ∈ Fk
for all k ∈ N. We conclude that F is not empty.
If F had more than one point, then it would follow that diam F > 0. For each n ∈ N,
F ⊂ Fn means that diam F ≤ diam Fn and hence diam Fn does not converge to 0.

Note the resemblance of Theorem 3.4.1 and Corollary 2.4.5. The former is about
complete spaces and the latter is about compact spaces. One may ask if the condition
diam Fn → 0 is really necessary. I affirm it is. Let Fn = [n, +∞) be closed subsets of
the real line (which is a complete metric space as we just proved). You may note that
T∞
n=1 Fn = ∅.

Theorem 2.4.2 — Baire’s Theorem. Let M be a complete metric space and {En } be
a collection of sets where each En is open and dense in M. Then the set
∞
\
E= En
n=1

is a dense subset of M.

Proof. Let N be any neighborhood. We’ll prove that E ∩ N is not empty. Let
N1 = N. Since E1 is open and dense, E1 ∩ N1 is open and not empty. Hence it contains a
neighborhood N2 which we can suppose so little that N2 ⊂ E1 ∩ N1 and diam N2 < 1/2.
Similarly, since E2 is open and dense there is a neighborhood N3 such that N3 ⊂ E2 ∩ N2
and diam N3 < 1/3. We then obtain a sequence of neighborhoods Nk such that Nk+1 ⊂ Nk ,
T∞
Nk+1 ⊂ Ek ∩ Nk and diam Nk → 0. Theorem 2.4.1 then implies k=1 Nk consists of a
† Every Cauchy sequence in E is a Cauchy sequence in M, hence it converges to some x ∈ M, and actually

x ∈ E since E is closed.
2.4 Complete Metric Spaces 33

single point x. Since Nk+1 ⊂ Ek ∩ Nk , x ∈ En for all n ∈ N and x ∈ N1 = N. Hence

x ∈ E ∩ N.

We could have proved the following corollary before but, since we didn’t needed it
before and this proof is so simple, this is a better place for it.

Corollary 2.4.3 The set of all real numbers R is uncountable.

Proof. If R was countable we could list its elements as {r1 , r2 , r3 , . . . }. Since each
set with only one element is closed, for each rn the set En = R \ {rn } is open and dense
in R. However note that
∞
\ ∞
\ ∞
[
En = R \ {rn } = R \ {rn } = R \ R = ∅.
n=1 n=1 n=1

This contradicts Baire’s Theorem.

In the first chapter we defined the real field as a ordered field with the least-upper-
bound property. However, what if such a specific field does not exists? It would be a
shame as I’ve spent months writing more than fifty pages of numerous results about the
real field that are now vacuously‡ true since there is no real field. Fortunately I now
possess the machinery to present an example of a ordered field with the least-upper-bound
property!
Definition 2.4.2 — Equivalent Sequences. Two sequences (xn ) and (yn ) defined
on a same metric space M are said to be equivalent if

lim d(xn , yn ) = 0.
n→∞

If (xn ) is equivalent to (yn ) we denote this fact as (xn ) ∼ (yn ).

It should be clear that this is an equivalence relation.

Theorem 2.4.4 Let Q be the set of all Cauchy sequences of rational numbers. The
set Q/ ∼ of all equivalence classes of such sequences is an ordered field with the
least-upper-bound property.

Denoting by x and y the arbitrary elements [(xn )] and [(yn )] of Q/ ∼ we’ll define the
sum of two elements of Q/ ∼ as x + y = x + y = [(xn + yn )] and the product as x · y =
xy = [(xn yn )]. The elements 0 = [(0)] and 1 = [(1)] are the additive and multiplicative
identities and the element −x = [(−xn )] is the additive inverse of x = [(xn )]. Last, but
not least, we say that x > y if x − y = [(xn − yn )] for some Cauchy sequences (xn ) and
(yn ) such that xn > yn for all n > n0 , where n0 is an positive integer.
‡ A logic statement is vacuously true if it asserts that all members of the empty set have a certain property.

It’s like a child saying to his or her parent: "I ate every vegetable in the plate!" when there was not a single
vegetable there to begin with.
 34 Chapter 2. Sequences and Series

We can also embed Q into Q/ ∼ using the function x 7→ [(x)]. This function is
injective since [(x)] = [(y)] imply limn→∞ |x − y| = |x − y| = 0 and hence x = y.
Since the proof of this theorem is not really useful for the rest of the book but only
acts as a motivation for you to believe that this book is not entirely useless, you can work
it by yourself. (I’m not even sorry.)

Exercise 2.2 Prove Theorem 3.4.4.

2.5 Series in Normed Vector Spaces

Definition 2.5.1 — Series. Let X be a normed vector space. Given a sequence (xn )
in X we construct another sequence (sn ) as follows. Let s1 be x1 . Having chosen
s1 , . . . , sn , pick sn+1 to be sn + xn+1 . It is said that the sequence (sn ) is an series. If
(sn ) converges to s we denote this fact as
∞
s= ∑ xn
n=1

and say that s is the sum of the series. It will be common to abuse language and
denote by ∑ xn the series itself. It is said that sn is a partial sum of the series.

It should be noticed that, while we use the term "sum" and the symbol ∑, a series is
no ordinary sum at all. A priori it is not even clear that series share properties with sums.
Usually it is too hard to calculate the limit of a series. Until we develop a sufficiently
robust machinery to deal with this task, we shall focus on convergence and divergence.
Example 2.5 — Geometric Series. Let x be any complex number such that |x| < 1.
Using induction it is simple to verify that
n
1 − xn+1
∑ xk = 1−x
k=0

and hence
∞
1
∑ xn = 1 − x .
n=0


The next few theorems are simple consequences of important facts about sequences.

Theorem 2.5.1 — Cauchy Criterion. Let ∑ xn be a series in a normed space X. If

∑ xn converges, then for all ε > 0 there is an positive integer n0 such that
m
∑ xk <ε
k=n

whenever m ≥ n > n0 . Moreover, in Banach spaces the converse holds.

 2.6 Absolute Convergence 35

Proof. This is merely a restatement of Theorem 2.3.1 applied to the sequence (sn )
of partial sums. The converse is basically the definition of a complete space.

Corollary 2.5.2 If ∑ xn converges, then xn → 0.

Proof. Take m = n in the preceding theorem.

The converse of this theorem does not hold even in complete spaces.
 Example 2.6 — Harmonic Series. Take xn = 1/n. The corresponding series is said
to be the harmonic series. Clearly xn → 0. However, consider the sequence (yn ) such
that 2n
1
yn = ∑ .
k=n+1 k

Since yn+1 − yn = 1/(2k + 1) − 1/(2k + 2) > 0, the sequence (yn ) is strictly increasing
and hence yn > y1 = 1/2. It follows that the Cauchy criterion does not hold for xn = 1/n
and the harmonic series diverges. 

Theorem 2.5.3 If ∑ xn and ∑ yn are series which converge to s1 and s2 respectively

and p is an element of X’s field, the following holds.
∞ ∞
∑ (xn + yn ) = s1 + s2 and ∑ pxn = ps1 .
n=1 n=1

Proof. Follows readily from Theorem 2.1.6.

Theorem 2.5.4 If ∑ xn and ∑ yn are converging series of real numbers such that
xn ≥ yn for all n ∈ N then
∞ ∞
∑ xn ≥ ∑ yn .
n=1 n=1

Proof. Follows from Corollary 2.1.8.

2.6 Absolute Convergence

We shall meet now a particular family of series which behave much like ordinary sums.
Definition 2.6.1 — Absolute Convergence. A series ∑ xn is said to be absolutely
convergent if the series ∑ kxn k converges.
If ∑ xn converges but not absolutely, it is said to be conditionally convergent.
36 Chapter 2. Sequences and Series

Theorem 2.6.1 In a Banach space X, an absolutely convergent series converges.

Moreover we have that
∞ ∞
∑ xn ≤ ∑ kxn k .
n=1 n=1

Proof. The Cauchy criterion readily implies convergence. For m ≥ 1:

m m ∞
∑ xn ≤ ∑ kxn k ≤ ∑ kxn k .
n=1 n=1 n=1

Then Corollary 3.1.8 implies the inequality.

A permutation of a set S is a function σ : S → S which is bijective. Of course, since

the sum of elements in a field or vector space is commutative, taking S = {1, 2, . . . , n}
we see that
n n
∑ xk = ∑ xσ (k)
k=1 k=1

for any sequence (xk ), any n ∈ N and any permutation σ . After all, we just rearranged
the elements of the sum. One would expect that series share this property. As we’ll see
soon, Bernhard Riemann showed that we couldn’t be more wrong!
 Example 2.7 — Alternating Harmonic Series. For now, you’ll trust me that the
alternating harmonic series
∞
(−1)n+1
∑ n
n=1

converges to a value s 6= 0. Consider the following permutation in the set of all positive
integers:

4n/3
 if n is divisible by 3
σ (n) = (2n + 1)/3 if n − 1 is divisible by 3 .

(4n − 2)/3 if n − 2 is divisible by 3


If xn = (−1)n+1 /n, the series ∑ xσ (n) can be decomposed as

∞ ∞
1 ∞ (−1)n+1
 
1 1 1 s
∑ xσ (n) = ∑ − − = ∑ n = 2.
n=1 n=1 2n − 1 2(2n − 1) 4n 2 n=1

Hence, a rearrangement of the terms can modify the value to which a series converges. 

Fortunately, rearranging series that converge absolutely does not change its value.

Theorem 2.6.2 Let ∑ xn be a series in a Banach space X which converges absolutely.

Then every rearrangement ∑ xσ (n) converges, and they all converge to the same value.
2.6 Absolute Convergence 37

Proof. Let (s0n ) be the sequence of the partial sums of ∑ xσ (n) . Since ∑ xn is
absolutely convergent, given ε > 0 there is an integer n0 such that
m
∑ kxk k < ε
k=n0

for all m ≥ n0 . Let

p = max σ −1 (i).
1≤i<n0

If n > p, we have that {1, 2, . . . , n0 − 1} is a subset of {σ (1), σ (2), . . . , σ (n)}. Hence all
the xi for i = 1, 2, . . . , n0 − 1 are cancelled in sn − s0n . So,
m
sn − s0n ≤ ∑ kxk k < ε.
k=n0

We conclude that (s0n ) converges to the same value as (sn ).

The following theorem is Riemann’s way of saying: "Thou shall be careful when
dealing with such outlandish objects as series!"

Theorem 2.6.3 — Riemann Rearrangement Theorem. Let ∑ xn be a conditionally

convergent series of real numbers. Suppose

−∞ ≤ α ≤ β ≤ +∞.

Then there exists a rearrangement ∑ xσ (n) with partial sums s0n such that

lim inf s0n = α, and lim sup s0n = β .

n→∞ n→∞

Proof. For all n ∈ N, define pn = (|xn | + xn )/2 and qn = (|xn | − xn )/2. If both ∑ pn
and ∑ qn were convergent, then so would be ∑ |xn | = ∑(pn + qn ). Since
n n n n
∑ xk = ∑ (pk − qk ) = ∑ pk − ∑ qk ,
k=1 k=1 k=1 k=1

the divergence of just one of ∑ pn or ∑ qn implies in the divergence of ∑ xn . We conclude

that the series ∑ pn and ∑ qn must both diverge.
Now, let P1 , P2 , P3 , . . . denote the non-negative terms of ∑ xn , in the order in which
they occur, and let Q1 , Q2 , Q3 , . . . be the absolute values of the negative terms of ∑ xn ,
also in their original order.
The series ∑ Pn and ∑ Qn differ from ∑ pn and ∑ qn only by zero terms, and are
therefore divergent.
We shall construct sequences (mn ), (kn ), such that the series

P1 + · · · + Pm1 − Q1 − · · · − Qk1 + Pm1 +1 + · · · + Pm2 − Qk1 +1 − · · · − Qk2 + · · · ,

 38 Chapter 2. Sequences and Series

which clearly is a rearrangement of ∑ xn , satisfies the statement of the theorem.

Choose real-valued sequences (αn ), and (βn ) such that αn → α, βn → β , αn < βn
and β1 > 0.
Let m1 , k1 be the smallest integers such that

P1 + · · · + Pm1 > β1 ,
P1 + · · · + Pm1 − Q1 − · · · − Qk1 < α1 .

Let m2 , k2 be the smallest integers such that

P1 + · · · + Pm1 − Q1 − · · · − Qk1 + Pm1 +1 + · · · + Pm2 > β2 ,

P1 + · · · + Pm1 − Q1 − · · · − Qk1 + Pm1 +1 + · · · + Pm2 − Qk1 +1 − · · · − Qk2 < α2 ,

and continue in this way. This is possible since ∑ Pn and ∑ Qn diverge.

If bn and an are the partial sums whose last terms are Pmn and −Qkn , then

|bn − βn | ≤ Pmn , |an − αn | ≤ Qkn .

Since Pn → 0 and Qn → 0, we see that bn → β and an → α.

Finally, it is clear that no number less than α or greater than β can be a subsequen-
tial limit of the partial sums of our rearrangement.

The preceding proof and slightly generalization of the original result by Bernhard
Riemann appears to be due to W. Rudin[13].

2.7 Series of Non-Negative Real Numbers

Series of non-negative real numbers share a particularly simple behaviour, so we’ll focus
on them for a while.
Theorem 2.7.1 Let ∑ xn be a series of non-negative real numbers. Then ∑ xn con-
verges if, and only if its sequence of partial sums (sn ) is bounded.

Proof. Since xn ≥ 0 for all n ∈ N, (sn ) is increasing. Hence the Monotone Conver-
gence Theorem applies.

Theorem 2.7.2 — Comparison Test. Let ∑ xn and ∑ yn be series of real numbers such
that ∑ yn converges and |xn | ≤ yn for all n > n0 . Then ∑ xn is absolutely convergent,
and hence convergent (since R is Banach).

Proof. Since ∑ yn converges, for all ε > 0 there exists n1 ≥ n0 such that

m
∑ yk < ε,
k=n
2.7 Series of Non-Negative Real Numbers 39

whenever m ≥ n > n1 . So,

m m
∑ |xk | ≤ ∑ yk < ε.
k=n k=n

The result follows from the Cauchy Criterion.

Corollary 2.7.3 Let ∑ xn and ∑ yn be series of non-negative real numbers such that
xn ≥ yn for all n > n0 . If ∑ yn diverges, then so do ∑ xn .

Proof. If ∑ xn converges, then the previous theorem implies that ∑ yn converges

too. Which is an absurd!

Corollary 2.7.4 — Limit Comparison Test. Let ∑ xn and ∑ yn be series of non-

negative real numbers. If
xn
0 < lim < +∞,
n→∞ yn

then either both series converge or both series diverge.

Proof. If xn /yn → c, let ε = c/2. The limit then implies that for all n > n0 :

c 3c
yn < xn < yn .
2 2

If ∑ xn converges, then so does ∑ cyn /2 and hence ∑ yn converges. Similarly, if ∑ yn

converges, then so does ∑ 2xn /3c and hence ∑ xn converges.

Although the next example is wonderful, it assumes the knowledge of the fundamen-
tal theorem of arithmetic. So, don’t be afraid to skip it. The following proof is due to J.
A. Clarkson[3].
 Example 2.8 — The Series of Primes Reciprocals. Let (pn ) be the sequence of all
the prime numbers. I affirm that the series of prime reciprocals,
∞
1
∑ pn
n=1

diverges. If it converged, then there would exist an integer k such that

∞ k ∞
1 1 1 1
∑ −∑ = ∑ < .
n=1 pn n=1 pn n=k+1 pn 2

Let Q = p1 · · · pk , and consider the numbers 1 + nQ for n ∈ {1, . . . , r}. Since these num-
bers are not divisible by any of the p1 , . . . , pk , the prime factors of all these numbers are
members of a finite set {pk+1 , pk+2 , . . . , pm(r) }. If τ(n) is the number of (not necessarily
40 Chapter 2. Sequences and Series

distinct) prime factors in 1 + nQ we have that

r ∞ r
1 1
∑ = ∑ ∑
n=1 1 + nQ t=1 n=1 1 + nQ
τ(n)=t
!t
∞ m(r)
1
<∑ ∑
t=1 n=k+1 pn
!t
∞ ∞
1
<∑ ∑
t=1 n=k+1 pn
∞  t
1
<∑ = 1.
t=1 2

The inequality
!t
r m(r)
1 1
∑ < ∑
n=1 1 + nQ n=k+1 pn
τ(n)=t

is justified by the fact that every term in the leftmost sum appears at least once in the
sum on the right. We conclude that ∑ 1/(1 + nQ) converges, since its sequence of partial
sums is bounded. However the limit comparison test says otherwise (when compared to
the harmonic series). 

Leonhard Euler was the first one to prove this result and to notice that it implies
Euclid’s theorem on the infinitude of prime numbers.

Theorem 2.7.5 — Cauchy Condensation Test. Let (xn ) be a decreasing sequence

of non-negative real numbers. Then the series ∑ xn converges if, and only if the series
∞
∑ 2k x2k = x1 + 2x2 + 4x4 + 8x8 + · · ·
k=0

converges.

Proof. Let
n
sn = ∑ xn , tk = ∑ 2i x2i .
i=1 i=0

For n < 2k ,
k
sn ≤ x1 + (x2 + x3 ) + · · · + (x2 + · · · + x2k+1 −1 )
≤ x1 + 2x2 + · · · + 2k x2k
= tk .
2.7 Series of Non-Negative Real Numbers 41

Similarly, for n > 2k ,

sn ≥ x1 + x2 + (x3 + x4 ) = · · · + (x2k−1 +1 + · · · + x2k )

1
≥ x1 + x2 + 2x4 + · · · + 2k−1 x2k
2
1
= tk .
2
We conclude that (sn ) and (tk ) are both bounded or both unbounded. Theorem 3.7.1
implies the result.

 Example 2.9 The series ∑ 1/n p converges if, and only p > 1. In fact, if p ≤ 0 then
1/n p does not tend to 0 as n → ∞ and hence ∑ 1/n p diverges. For p > 0, the preceding
theorem says that ∑ 1/n p converges or diverges together with
∞ ∞
1
∑ 2k · 2kp = ∑ 2(1−p)k .
k=0 k=0

We conclude that ∑ 1/n p converges if, and only if p > 1. 

The following theorem shows that an even stronger result than Corollary 2.5.2 holds
for series of decreasing non-negative real numbers.

Theorem 2.7.6 Let (xn ) be a decreasing sequence of non-negative real numbers.

Then the series ∑ xn converges only if

lim nxn = 0.
n→∞

Proof. If ∑ xn converges, then for all ε > 0 we have that

n
ε
∑ xn < 2
k=m

whenever n ≥ m > n0 . Since (xn ) is decreasing,

n
n ε
xn ≤ (n − m + 1)xn ≤ ∑ xn <
2 k=m 2

whenever n > 2m. It follows that nxn → 0.

The next theorem allow us to determine the convergence of a different family of

series. The series to which Leibniz test applies are called alternating series. A quick
corollary of this test is that the alternating harmonic series converges.
 42 Chapter 2. Sequences and Series

Theorem 2.7.7 — Leibniz Test. Let (xn ) be a decreasing sequence of non-negative

real numbers such that xn → 0. Then, the series
∞
∑ (−1)n+1 xn
n=1
converges.

Proof. Let sn be the partial sums of the series and define In = [s2n , s2n−1 ].
Note that s2n+2 − s2n = −x2n+2 + x2n+1 ≥ 0. Similarly, s2n+1 − s2n−1 ≤ 0 and hence
In+1 ⊂ In for all n ∈ N. Since xn → 0 it follows that diam In → 0 and then
∞
\
In = {s}.
n=1

Pick n0 such that diam In < ε for n > n0 . Then |sn − s| < ε and the result follows.

Exercise 2.3 Give another proof of the Leibniz Test using only the Cauchy criterion.

2.8 Two Fundamental Tests and Power Series

The following two theorems are consequences of the comparison test and constitute our
most powerful machinery.

Theorem 2.8.1 — Root Test. Let ∑ xn be a series in a normed vector space X. If

p
n
lim sup kxn k < 1,
n→∞

∑ xn is absolutely convergent. Hence ∑ xn converges if X is Banach.

pn
Proof. Let k be a real number such that lim supp n→∞ kxn k < k < 1. Lemma 3.2.9
n
then implies that there is an integer n0 such that kxn k < k whenever n > n0 . Hence,
kxn k < kn . Using the comparison test with the geometric series we conclude that ∑ kxn k
converges.

 Note that while ∑ xn is a series in a normed vector space, ∑ kxn k is a series of real
numbers and hence the comparison test applies.

Theorem 2.8.2 — Ratio Test. Let ∑ xn be a series in a normed vector space X. If

kxn+1 k
lim sup < 1,
n→∞ kxn k

then ∑ xn is absolutely convergent. Hence ∑ xn converges if X is Banach.

2.8 Two Fundamental Tests and Power Series 43

p
Proof. Let k be a real number such that lim supn→∞ n kxn k < k < 1. Lemma 2.2.9
then implies that there is an integer n0 such that kxn+1 k < k kxn k whenever n > n0 .
Hence,
xn0 +p < k xn0 +p−1 < k2 xn0 +p−2 < · · · < k p−1 xn0 +1

and then
xn0 +1 n
kxn k < k
kn0 +1
whenever n > n0 . The result now follows from the comparison test.

Whenever the ratio test implies convergence, the root test does too but the converse
does not hold. However it is usually easier to apply the ratio test. It should be noted
that we only have two criteria for determining divergence. Namely Corollary 2.5.2 and
Theorem 2.7.6.
Definition 2.8.1 — Power Series. Given a sequence (an ) of complex numbers, the
series
∞
∑ an zn ,
n=0

defined for complex z, is said to be a power series. The numbers an are the coefficients
of the series.

The convergence of a power series depends on the value of z, so we’ll study which
values of z makes a given power series convergent or not. In general, there is a number
R ∈ R, called the radius of convergence such that ∑ an zn converges for |z| < R. This is
the soul of the following theorem.

Theorem 2.8.3 Given a power series ∑ an zn , let

1 p
= lim sup n |an |.
R n→∞

Then, ∑ an zn converges for |z| < R.

Proof. If |z| < R, then

p
n
p
n |z|
lim sup |an zn | = |z| lim sup |an | = < 1.
n→∞ n→∞ R

The root test then implies convergence.

In this proof we used the fact that C is complete, a result that we did not explicitly proved.
 However, the only difference between C and R2 is the definition of an multiplication of
elements. As normed vector spaces they are exactly the same thing.

The following series defines the most important function in mathematics.

44 Chapter 2. Sequences and Series

 Example 2.10 The series

∞
zn
∑ n!
n=0

converges (absolutely) everywhere. As

|zn+1 |n! |z|
= ,
|zn |(n + 1)! n+1

the ratio test implies convergence for every z ∈ C. (Except for z = 0, but in this case the
result is trivial.) 

Definition 2.8.2 — Exponential Function. For all z ∈ C we define exp(z) to be

∞
zn
exp(z) = ∑ n! .
n=0

This function is said to be the exponential function. The number exp(1) is so

important that it will be denoted e and called Euler’s number, in honor of Leonhard
Euler.
Example 2.10 shows that this is a well defined function. Unfortunately, we cannot
prove some familiar properties of this function yet. However we can prove two known
facts about e.
Theorem 2.8.4 Euler’s Number e is irrational.

Proof. Let sn be the partial sums of ∑ 1/n!. Note that

1 1 1
e − sn = + + +···
(n + 1)! (n + 2)! (n + 3)!
 
1 1 1 1
< 1+ + +··· = .
(n + 1)! n + 1 (n + 1)2 n!n

Hence,
1
0 < n!(e − sn ) < ,
n
for all n ∈ N. The number e is clearly a positive real number. If e was rational, then we
would have that e = m/n for some m, n ∈ N. Since ne is an integer, so is n!(e − sn ). That
is, we found an integer between 0 and 1. Absurd!

Theorem 2.8.5 The sequences

1 n n
 
1
xn = 1 + and yn = ∑ k!
n k=0

converge to the same limit. Namely, Euler’s number e.

2.8 Two Fundamental Tests and Power Series 45

Proof. For n ≥ 2, the binomial theorem implies

1 n n   n
1 n−1 n
   
n 1 j 1
xn = 1 + =∑ k
= ∑ ∏ 1 − ≤ ∑ = yn .
n k=0 k n k=0 k! j=1 n k=0 k!

Hence, lim supn→∞ xn ≤ lim supn→∞ yn = e. Now, if n ≥ m we have that

n m
1 n−1 1 m−1
   
j j
xn = ∑ ∏ 1 − ≥ ∑ ∏ 1 − .
k=0 k! j=1 n k=0 k! j=1 n

Let n → ∞ keeping m fixed. We get

" #
m m m
1 m−1 1 m−1

j ∗ 1
lim inf xn ≥ lim inf ∑ ∏ 1 − = ∑ ∏ (1 − 0) = ∑ = ym .
k=0 k! j=1 n k=0 k! j=1 k=0 k!
n→∞ n→∞

Hence, lim infn→∞ xn ≥ lim infn→∞ yn = e. The result follows.

In this proof we assumed that the properties of Theorem 3.1.6 were also valid for the
 limit inferior. However this is not true as suggested by Exercises 3.13 and 3.14. Show
that the proof of Theorem 3.8.5 is still valid provided that we switch the starred equality
sign by an ≥ sign.
We now define the trigonometric functions without any aid of geometrical interpreta-
tions.
Definition 2.8.3 — Trigonometric Functions. If z ∈ C, we define the trigonometric
functions sine and cosine as
∞ ∞
(−1)n 2n+1 (−1)n 2n
sin(z) = ∑ z , cos(z) = ∑ z .
n=0 (2n + 1)! n=0 (2n)!

The trigonometric function tangent is defined in the usual way as

sin(z)
tan(z) = .
cos(z)

For trigonometric functions it is usual to let sinn (z) mean (sin(z))n .

In the present state we aren’t even able to prove that sin2 (z) + cos2 (z) = 1. However,
Euler’s Identity follows readily from the definition.

Theorem 2.8.6 — Euler’s Identity. For all t ∈ R, the following equation holds:

exp(it) = cos(t) + i sin(t).

Proof. This is merely a restatement of the definition.

 46 Chapter 2. Sequences and Series

2.9 The Cauchy Product of Series

Given two finite sums ∑nk=1 ak and ∑nk=1 bk , it is clear that their product is given by the
sum of all ai b j , such that 1 ≤ i, j ≤ n. However, when dealing with series, the order in
which we sum the ai b j may change the limit. History has shown that, for the purposes
of analysis, there is one best way of dealing with this problem. Is this section we’ll deal
exclusively with series of real or complex numbers.
Definition 2.9.1 — Cauchy Product. Given two series ∑∞ ∞
n=0 an and ∑n=0 bn , we
define a sequence (cn ) as
n
cn = ∑ ak bn−k
k=0

and call ∑∞
n=0 cn the Cauchy product of the two series.

This definition can be motivated by observing that when we multiply two polynomials
of the same degree and collect the terms which contain the same power of z we get the
sequence (cn ):
! !
n n 2n
∑ ak zk ∑ bk zk = ∑ ck zk ,
k=0 k=0 k=0

provided that we consider ak = bk = 0 for k ≥ n.

If ∑∞ ∞ ∞
n=0 an = A and ∑n=0 bn = B, it is natural to ponder whether ∑n=0 cn converges
to AB or not. Mertens proved that if at least one of the series converges absolutely, then
the result holds.
Theorem 2.9.1 Let A = ∑∞
n=0 an be an absolutely convergent series and let B =
∑∞ ∞
n=0 bn be a convergent series. Then ∑n=0 cn converges to AB.

Proof.

2.10 Exercises
Exercise 2.4 Let (xn ) be an sequence of real numbers such that

lim (2xn+1 − xn ) = x.
n→∞

Prove that xn → x.

Exercise 2.5 Let (xn ) and (yn ) be sequences of real numbers. Assume that yn → 0
and that there is a real number k ∈ (0, 1) such that xn+1 ≤ kxn + yn for every n ∈ N.
Prove that xn → 0.
2.10 Exercises 47

Exercise 2.6 If p > 0, prove that

√
lim n p = 1.
n→∞

Exercise 2.7 — The Arithmetic-Geometric Mean. Let x, y be two positive real

√
numbers. Define the sequences (xn ) and (yn ) by x1 = xy, y1 = (x + y)/2, and

√ xn + yn
xn+1 = xn yn , yn+1 = ,
2
for all n ∈ N. Prove that the two sequences are convergent and have the same limit.

Exercise 2.8 — Cesàro-Stolz Theorem. Let (xn ) and (yn ) be two sequences of real
numbers with (yn ) strictly positive, increasing and unbounded. If
xn+1 − xn
lim = L,
n→∞ yn+1 − xn

then the limit

xn
lim
n→∞ yn

exists and is equal to L

Exercise 2.9 — Cesàro Means. Let (xn ) be an convergent sequence of real numbers.
Prove that the sequence
x1 + x2 + · · · + xn
yn =
n
also converges to the same limit.

Exercise 2.10 Let (xn ) and (yn ) be Cauchy sequences in a metric space X. Show
that the sequence (d(xn , yn )) converges.

Exercise 2.11 Let (xn ) be a sequence of real numbers. Decide whether each of the
following propositions are true or false. In each case, prove it or show a counter-
example.

a) If every subsequence of (xn ) that is not itself converges, then (xn ) converges
as well.
b) If (xn ) contains a divergent subsequence, then (xn ) diverges.
48 Chapter 2. Sequences and Series

c) If (xn ) is bounded and diverges, then there exist two subsequences of (xn ) that
converge to different limits.
d) If (xn ) is monotone and contains a convergent subsequence, then (xn ) con-
verges.

Exercise 2.12 Let (xn ) be a sequence of real numbers. Show that

lim sup xn = lim sup{xm | m ≥ n} = lim sup xm ,

n→∞ n→∞ n→∞ m≥n

where the rightmost equality is a common abuse of notation. Moreover, show that
this is also valid for the limit inferior.

Exercise 2.13 For any two real sequences (xn ) and (yn ), prove that

lim inf(xn + yn ) ≥ lim inf xn + lim inf yn ,

n→∞ n→∞ n→∞

provided the sum on the right is not of the form ∞ − ∞. Moreover show that this is
also valid for the limit superior, provided that we switch the side of the inequality.

Exercise 2.14 For any two real sequences (xn ) and (yn ) such that xn and yn non-
negative, prove that
  
lim inf(xn yn ) ≥ lim inf xn lim inf yn ,
n→∞ n→∞ n→∞

provided the sum on the right is not of the form 0 · ∞. Moreover show that this is
also valid for the limit superior, provided that we switch the side of the inequality.

Exercise 2.15 Let (xn ) be a sequence of positive real numbers. Then

xn+1 √ √ xn+1
lim inf ≤ lim inf n xn ≤ lim sup n xn ≤ lim sup .
n→∞ xn n→∞ n→∞ n→∞ xn

In particular this result implies that if limn→∞ xn+1 /xn = L, with 0 ≤ L ≤ ∞, then
√
limn→∞ n xn = L.

Exercise 2.16 For which real numbers p does the series

∞
√ √
n+1− n
∑ np
n=1
2.10 Exercises 49

converge?

Exercise 2.17 Let S = {x1 , x2 , . . . } be the set of all integers that do not contain the
digit 9 in their decimal representation. Prove that
∞
1
∑ xn
n=1

converges.

Exercise 2.18 Find the radius of convergence of each of the following power series:

∞ ∞
2n n
a) ∑ n3 zn , c) ∑ 2z ,
n=0 n=1 n

∞ ∞
2n n3
b) ∑ n! zn , d) ∑ 3n zn .
n=0 n=0

Exercise 2.19 Let (xn ) be a sequence of non-negative real numbers such that ∑ xn
diverges. Consider the following series:
∞ ∞ ∞ ∞
xn xn xn xn
∑ 1 + xn , ∑ 1 + nxn , ∑ 1 + n2 xn , ∑ 1 + xn2 .
n=1 n=1 n=1 n=1

What can be said about their convergence?

Exercise 2.20 Suppose that the coefficients of the power series ∑ an zn are integers,
infinitely many of which are distinct from zero. Prove that the radius of convergence
is at most 1.

Exercise 2.21 Associate to each sequence a = (αn ), in which αn is 0 or 2, the real

number
∞
αn
x(a) = ∑ n
.
n=1 3

Prove that the set of all x(a) is exactly the Cantor set described in Exercise 2.19.
Since there is an uncountable number of such sequences (Theorem 1.5.6), this gives
another proof of the fact that the Cantor set is uncountable.
50 Chapter 2. Sequences and Series

Exercise 2.22 Let (nk ) be a strictly increasing sequence of positive integers such
that
nk
lim = +∞.
k→∞ n1 n2 n3 · · · nk−1

Prove that the series ∑ 1/nk is convergent and its limit is an irrational number.

Exercise 2.23 — Summation by Parts. Let (xn ) and (yn ) be sequences in a normed
vector space X. Let sn = x1 + x2 + · · · + xn and set s0 = 0. Use the observation that
xn = sn − sn−1 to verify the formula
m m
∑ xi yi = sm ym+1 − sn−1 yn + ∑ si (yi − yi+1 ).
i=n i=n

Exercise 2.24 — Dirichlet’s Test. Let (xn ) and (yn ) be sequences of real numbers
such that ∑ xn has bounded partial sums, yn → 0 and (yn ) satisfies

y1 ≥ y2 ≥ y3 ≥ · · · ≥ 0.

Prove that the series

∞
∑ xn yn
n=1

converges. Use this result to give one more proof of the Leibniz test.

Exercise 2.25 — Abel’s Test. In the view of the preceding exercise, show that if
∑ xn converges, then we can drop the restriction yn → 0 while keeping the result.

Exercise 2.26 Let (xnm ) be a doubly indexed sequence of points in a normed space
X. Suppose that:
∞
• For each m ∈ N, the series ∑ xnm is convergent. Denote by ym its limit.
n=1
∞ ∞
• For each n ∈ N, the series ∑ ∑ xi j is convergent. Denote by tn its limit.
j=1 i=n

∞
Show that, for each n ∈ N, the series ∑ xnm is convergent. Denote by zn its limit.
m=1
∞ ∞
Moreover, prove that ∑ ym = ∑ zn if, and only if tn → 0.
m=1 n=1
2.10 Exercises 51

Exercise 2.27 Let (xnm ) be a doubly indexed sequence of real numbers such that
xnm ≥ 0 for all n, m ∈ N. Prove that
∞ ∞ ∞ ∞
∑∑ xnm = ∑ ∑ xnm ,
n=1 m=1 m=1 n=1

whenever one of the double sums converges. Moreover, if one diverges then so does
the other.
 CHAPTER
Functional Limits
and Continuity
3
Since most of this book is about the things we can do to functions, its about time we
generalize the notion of a limit to a function. This will be the basis of our future studies.
As usual, we’ll stick to metric spaces whenever possible to clarify the concepts and
proofs.

3.1 Functional Limits

Unlike in the limit of a sequence, the definition of a limit becomes clearer when we step
a little away from the topological aspect.
Definition 3.1.1 — Limit of a Function. Let M and N be metric spaces, f : M → N
be a function and p be a limit point of M. We say that

lim f (x) = L
x→p

if for all ε > 0 there is a real number δ > 0 such that

d( f (x), L) < ε whenever 0 < d(x, p) < δ .

In this case we say that L is the limit of f (x) as x approaches p or, more concisely,
f (x) → L as x → p.

In this definition and in the entirety of this book I’ll be sloppy and use the same letter
d to denote two distinct metric functions defined on M and on N whenever the context
makes it clear. We should note that p need not be a point of M but only a point of M 0 .
Similarly, L need not be a point of f (M).
If f is a function from E ⊂ M to N, then the x which satisfies 0 < d(x, p) < δ has to be
 an element of E. Otherwise f (x) would not be defined. The point p has also to be a limit
point of E.
54 Chapter 3. Functional Limits and Continuity

Since the statement d( f (x), L) < ε is equivalent to f (x) ∈ U, where U is a neighbor-

hood of L in N, and the statement 0 < d(x, p) < δ is equivalent to x ∈ V \ {p},∗ where
V is a neighborhood of p in M, we can define the limit of a function "topologically" as
follows.
Definition 3.1.2 — Limit of a Function. Let M and N be metric spaces, f : M → N
be a function and p be a limit point of M. We say that

lim f (x) = L,
x→p

if for every neighborhood U of L there is a neighborhood V of p such that f (x) ∈ U

for all x ∈ V \ {p}.

As suggested by the dangerous bend sign in the previous page, if f is a function from
 E ⊂ M to N, then we have that V ∩ E has to be non-empty and f (x) ∈ U for all
x ∈ (V ∩ E) \ {p}.
It is remarkable how similar are the definitions of the limit of a sequence and that of
a function. The definition of (xn ) → p conveys the idea that xn can be made arbitrarily
close to p if n is big enough. Analogously, f (x) → L conveys the idea that f (x) can
be made arbitrarily close to L if x is close enough to p. In fact, the following theorem
shows that the two concepts are closely related.

Theorem 3.1.1 Let M, N, E ⊂ M be metric spaces, f : E → N be a function and p

be an limit point of E. Then,
lim f (x) = L
x→p

if, and only if,

lim f (xn ) = L
n→∞

for every sequence (xn ) of points in E \ {p} which converges to p.

Proof. Let (xn ) be any sequence in M \ {p} which converges to p.

Since f (x) → L, for all ε > 0 there is a δ > 0 such that

d( f (x), L) < ε whenever 0 < d(x, p) < δ .

For this δ , there is a positive integer n0 such that

d(xn , p) < δ whenever n > n0 .

Hence d( f (xn ), L) < ε for n > n0 which implies f (xn ) → L.

We’ll now suppose that f (xn ) → L for every sequence (xn ) and that f (x) does not
converge to L. Since f (x) 9 L, there is a ε > 0 such that for all δ > 0 there is a point
x ∈ M which satisfies

0 < d(x, p) < δ but d( f (x), L) ≥ ε.

∗ It is usual to say that V \ {p} is a punctured neighborhood of p.
3.1 Functional Limits 55

Taking δ = 1/n we obtain a sequence of points (xn ) in M \ {p} such that

0 < d(xn , p) < 1/n but d( f (xn ), L) ≥ ε.

This sequence clearly converges to p but the sequence ( f (xn )) does not converge to L,
contradicting our original assumption.

This theorem makes an simple way to show that a limit does not exist. If we
find sequences (xn ) and (yn ) in E \ {p} such that lim xn = lim yn = p and lim f (xn ) 6=
lim f (yn ) then we can conclude that lim f (x) does not exist.
Now, almost every important result about limits of functions follows from the
preceding theorem and the fact that a point p ∈ M is a limit point if, and only if there is
a sequence (pn ) of points in M \ {p} which converges to p (Theorem 3.1.2).

Corollary 3.1.2 — A function can converge to at most one point. Let M and N
be metric spaces and p be a limit point of M. If f (x) → L1 as x → p and f (x) → L2
as x → p, then L1 = L2 .

Proof. Follows from Theorems 4.1.1, 3.1.2 and 3.1.1.

Corollary 3.1.3 Suppose X1 , X2 are normed vector spaces of finite dimension and M
is a metric space. If f = ( f1 , f2 ) is a function from M to X1 × X2 , then

lim f (x) = (L1 , L2 )

x→p

if, and only if f1 (x) → L1 and f2 (x) → L2 .

Proof. Follows from Theorems 4.1.1, 3.1.2, 3.1.4 and 2.7.4.

Corollary 3.1.4 Suppose X is a normed space, M is a metric space and f , g : M → X

are functions. If f (x) → L1 and g(x) → L2 as x → p, then

lim ( f (x) + g(x)) = L1 + L2 .

x→p

If the codomain of f is R or C, then

lim f (x)g(x) = L1 L2 .
x→p

Moreover, if L1 6= 0, then
1 1
lim = .
x→p f (x) L1

Proof. Follows from Theorems 4.1.1, 3.1.2 and 3.1.6.

 56 Chapter 3. Functional Limits and Continuity

Corollary 3.1.5 Let M be a metric space and f : M → R, g : M → R be functions.

If f (x) ≥ g(x) for all x ∈ M, then

lim f (x) ≥ lim g(x).

x→p x→p

Proof. Follows from Theorems 4.1.1, 3.1.2 and Corollary 3.1.8.

Corollary 3.1.6 Let M be a metric space and f : M → R, g : M → R be functions.

If 0 ≤ f (x) ≤ g(x) for all x ∈ M, and if g(x) → 0 as x → p, then

lim f (x) = 0.
x→p

Proof. Follows from Theorems 4.1.1, 3.1.2 and 3.2.8.

 Example 3.1 Consider the function f : E → R such that f (x) = x/|x|. If E = (0, 1),
then
lim f (x) = 1.
x→0

And if E = (−1, 0), then

lim f (x) = −1.
x→0

This shows that it is important to consider the domain of a function when evaluating its
limit. 

 Example 3.2 Consider the function f : R → R such that

(
1 if x ∈ Q
f (x) = .
0 if x ∈ R \ Q

We’ll show that f (x) has no limit when x → 0. If f (x) converged to L, then we would
have that f (xn ) → L for every sequence of non-zero real numbers. However, if xn = 1/n:

lim f (xn ) = 1.
x→0
√
Hence L = 1. But taking xn = 2/n we see that

lim f (xn ) = 0.
x→0

And then L = 0. This contradiction implies the result. 

3.2 Limits on the Extended Real Line

As it is, the extended real line is not a metric space as there is no obvious way to assign
a real number to the expression d(x, ∞), where x is a real number. However, it is almost
 3.3 Asymptotic Comparison 57

tempting to write things like

1 1
lim =∞ and lim = 0.
x→0 x2 x→∞ x
While we shall not define a metric on R, we can quite naturally define the neighborhoods
of ±∞.
Definition 3.2.1 If c is a real number, the sets (c, +∞) and (−∞, c) are defined to
be neighborhoods of +∞ and −∞, respectively.

We now define limits in the extended real line using the topological definition given
before.
Definition 3.2.2 — Limit of a Function. Let E be a subset of the real line and
f : E → R be a function. We say that

lim f (x) = L,
x→p

where p, L ∈ R, if for every neighborhood U of L there is a neighborhood V of p

such that V ∩ E is not empty, and such that f (x) ∈ U for all x ∈ (V ∩ E) \ {p}.

We can now prove that the results in the beginning of this section are valid. For all
ε > 0, we can take c = 1/ε so that 1/x ∈ (−ε, ε) whenever x ∈ (c, +∞). This implies
1
lim = 0.
x→∞ x
The other limit is analogous. We now define the limit superior and inferior for
functions on the real line.
Definition 3.2.3 — Limit Superior/Inferior of Functions. Let E be a subset of R, p
be a limit point of E and f : E → R be a function. Consider the set

F = {y ∈ R | y = lim f (xn ), {xn } ∈ E \ {p}}.

n→∞

We define the limit superior and the limit inferior of f as

lim sup f (x) = sup F, lim inf f (x) = inf F.

x→p x→p

Since Theorem 4.1.1 holds, the limit superior and inferior of functions satisfies most
of the properties of the limit superior and inferior of sequences. We’ll leave the reader
with the task of checking these properties.

3.3 Asymptotic Comparison

We’re often interested in the behaviour of a function in a normed space when it is
restricted to some neighborhood, as it is the case with the limits. Based on this fact, the
German mathematician Edmund Landau invented the following notation.
58 Chapter 3. Functional Limits and Continuity

Definition 3.3.1 — Little-o Notation. Let M be a metric space, X be a normed space

and p be a limit point of M. Suppose g : M → X is a function. We then define
o p (g(x)) to be the set of all functions f : M → X with the following property.
For every ε > 0, there is a neighborhood V of p such that

k f (x)k < ε kg(x)k

whenever x ∈ V \ {p}.

Whenever the context makes it clear, we’ll omit the index p.

It is usual in literature to abuse notation† and denote f (x) ∈ o(g(x)) as f (x) = o(g(x)).
In this case we define the expression f (x) = h(x) + o(g(x)) as f (x) − h(x) = o(g(x)).
If g(x) is not 0 in a punctured neighborhood of p, then f (x) = o(g(x)) is equivalent
to
f (x)
lim = 0.
x→p g(x)

Theorem 3.3.1 — Properties of Little-o. Let f , f1 , f2 , g, g1 , g2 : M → X be functions.

The following properties hold.
i) If f1 (x) = o(g(x)) and f2 (x) = o(g(x)), then f1 (x) + f2 (x) = o(g(x)).
ii) If f (x) = o(g(x)), then c f (x) = o(g(x)) for every non-zero real number c.

iii) If X = R or X = C, f1 (x) = o(g1 (x)) and f2 (x) = o(g2 (x)), then f1 (x) f2 (x) =
o(g1 (x)g2 (x)).
iv) If f (x) = o(g1 (x)) and g1 (x) = o(g2 (x)), then f (x) = o(g2 (x)).

Proof. In every item, ε > 0 will be a arbitrary positive real number.

i) There is a neighborhood V of p such that k f1 (x)k < ε kg(x)k /2 and a neighbor-
hood U of p such that k f2 (x)k < ε kg(x)k /2 whenever x ∈ V \{p} and x ∈ U \{p}
respectively. Hence,

k f1 (x) + f2 (x)k ≤ k f1 (x)k + k f2 (x)k < ε kg(x)k

whenever x ∈ (V ∩U) \ {p}.

ii) If c is positive, there is a neighborhood V of p such that k f (x)k < ε kg(x)k /c
whenever x ∈ V \ {p}. Hence,

kc f (x)k = c k f (x)k < ε kg(x)k

whenever x ∈ V \ {p}. If c is negative just take a neighborhood U of p such that

k f (x)k < ε kg(x)k /(−c) whenever x ∈ U \ {p}.
† As Donald Knuth pointed out: "mathematicians customarily use the = sign as they use the word ‘is’ in

English: Aristotle is a man, but a man isn’t necessarily Aristotle."

3.3 Asymptotic Comparison 59

√
p such that | f1 (x)| < ε|g1 (x)| and a neighborhood
iii) There is a neighborhood V of√
U of p such that | f2 (x)| < ε|g2 (x)| whenever x ∈ V \ {p} and x ∈ U \ {p}
respectively. Hence,

| f1 (x) f2 (x)| = | f1 (x)|| f2 (x)| < ε|g1 (x)||g2 (x)| = ε|g1 (x)g2 (x)|

whenever x ∈ (V ∩U) \ {p}.

√
iv) There is a neighborhood V of p√ such that k f (x)k < ε kg1 (x)k and a neighbor-
hood U of p such that kg1 (x)k < ε kg2 (x)k whenever x ∈ V \{p} and x ∈ U \{p}
respectively. Hence,
√ √ √
k f (x)k < ε kg1 (x)k < ε( ε kg2 (x)k) = ε kg2 (x)k

whenever x ∈ (V ∩U) \ {p}.

This completes the proof.

Definition 3.3.2 Suppose f , g : M → X are functions. We say that f (x) ∼ p g(x) if

f (x) − g(x) = o p (g(x)).

As you should check, this is an equivalence relation. If g(x) is not 0 in a punctured

neighborhood of p, then f (x) ∼ p g(x) is equivalent to

f (x)
lim = 1.
x→p g(x)
As before, we’ll usually omit the subscript when the context makes it clear.
 Example 3.3 — Fundamental Trigonometric Limit. I affirm that sin(x) ∼ x (as x → 0)
holds. In fact, since
∞ ∞
sin(x) − x (−1)n 2n (−1)n 2n−1
=∑ x =x∑ x ,
x n=1 (2n + 1)! n=1 (2n + 1)!

we have that
∞
| sin(x) − x| (−1)n
≤ |x| ∑ |x|2n−1 .
|x| n=1 (2n + 1)!

As x → 0, |x| → 0 and ∑(−1)n |x|2n−1 /(2n + 1)! → 0.‡ Hence, for all ε > 0,

| sin(x) − x| < ε|x|

if x is close enough to 0. A quick corollary of this result is the following limit:

sin(x)
lim = 1.
x→0 x


‡ Are you sure you can prove this rigorously?

 60 Chapter 3. Functional Limits and Continuity

3.4 Continuity
As we saw in chapter 2, if M and N are metric spaces, a function f : M → N is said to
be continuous at p ∈ M if for all ε > 0 there is a real number δ > 0 such that
d( f (x), f (p)) < ε whenever d(x, p) < δ .
Obviously this definition is very similar to Definition 4.1.1. However it has a few
notable differences. Firstly, for a function to be continuous at p, p has to be an element
of M, while in the limit it can just be a limit point. If p is an isolated point of M, f is
surely continuous at p since there is a neighborhood N of p such that N = {p} and hence
d( f (x), f (p)) = 0 for all x ∈ N.
If p is a limit point of M, then f is continuous at p if, and only if
lim f (x) = f (p).
x→p

We can also characterize continuity with sequences.

Theorem 3.4.1 Let M, N be metric spaces, f : M → N be a function and p be a point

of M. Then f is continuous at p if and only if
 
lim f (xn ) = f lim xn = f (p)
n→∞ n→∞

for every sequence (xn ) of points in M which converges to p.

Proof. The only change from this proof to that of Theorem 4.1.1 is the lack of
requirement that xn 6= p for all n ∈ N.

Analogously to Corollary 4.1.4, the following properties hold.

Theorem 3.4.2 Suppose X is a normed space, M is a metric space and f , g : M → X

are functions.
If f , g are continuous at p ∈ M, then so is f (x) + g(x) and α f (x) for all complex α.
Moreover, if the codomain of f is R or C and f (p) 6= 0, then 1/ f (x) and f (x)g(x)
are also continuous at p.

Proof. If p is an isolated point of M, then the result is trivial. Otherwise the result
follows from Corollary 4.1.4.

 Example 3.4 We’ll now consider rational functions of the form f : C → C such that
an xn + an−1 xn−1 + . . . + a1 x + a0
f (x) =
bm xm + bm−1 xm−1 + . . . + b1 x + b0
for some suitable complex constants an , . . . , a0 , bm , · · · , b0 and positive integers n, m.
Of course the identity function f (x) = x is continuous everywhere (just take δ = ε).
Theorem 4.4.2 implies that every polynomial function of the form
f (x) = an xn + an−1 xn−1 + . . . + a1 x + a0
3.4 Continuity 61

is continuous everywhere.
If bm pm + bm−1 pm−1 + . . . + b1 p + b0 6= 0 then

an xn + an−1 xn−1 + . . . + a1 x + a0
f (x) =
bm xm + bm−1 xm−1 + . . . + b1 x + b0

is also continuous at p. 

From this example it follows that f : R → R such that f (x) = xn − a, where a is

a positive real number and n is a positive integer, is continuous. We construct a new
function f˜ : R → R such that
(
f (x) if x ∈ R
f˜(x) = .
lim f (t) if x = ±∞
t→x

Our new function f˜ is continuous at ±∞ by the definition of limit, hence it is continuous

everywhere. Since f (0) = −a and f (+∞) = +∞, the Intermediate Value Theorem
implies the existence of a positive real number y such that yn = a. If there were two such
y, say y1 and y2 with y1 > y2 , we would have that a = yn1 > yn2 = a, which is absurd.
√ We
√
say that this y is the n-th root of a and denote it by n a. For a = 0 we define n 0 = 0 for
every n ∈ N.
We can now define rational powers of√real numbers as follows. If b = p/q is a
rational number (q >√0), we √ define ab as q a p . You should check that if m/n = p/q,
with n, q > 0, then = a = q a p . This implies that ab is uniquely determined.
n m

For real exponents b, we define ab as the supremum of

{at ∈ R | t ≤ b, t ∈ Q}.

The following exercise shows that this is well defined.

Exercise 3.1 If b = p/q is a rational number and a is a non-negative real number,

show that √
sup{at ∈ R | t ≤ b, t ∈ Q} = q p
a .
Moreover, show that if r, s are two real numbers, then ar+s = ar as .

A continuous function f : E ⊂ R → R is sometimes described, intuitively, as one

whose graph can be drawn without lifting your pencil from the paper. The following
example shows that this is not quite true.
 Example 3.5 Consider the function f : N → R such that f (n) = n.

If p is a fixed positive integer, taking δ = 1/2 we see that

|x − p| < δ =⇒ | f (x) − f (p)| = 0 < ε

for any ε > 0. Hence f is continuous. This also follows readily from the fact that every
point of N is isolated. 
 62 Chapter 3. Functional Limits and Continuity

3.5 Uniform Continuity

Lets prove explicitly (that is, not using Theorem 4.4.2) that two simple functions are
continuous and contrast their proofs.
 Example 3.6 Consider the identity function on R. That is, f : R → R such that
f (x) = x. Suppose p is any point of R and ε > 0 is an arbitrary real number. Then, since
| f (x) − f (p)| = |x − p|, we can take δ = ε and
| f (x) − f (p)| < ε whenever |x − p| < δ
holds. It follows that f is continuous. 

 Example 3.7 Consider the function f : R → R such that f (x) = x2 .

Suppose p is any
point of R and ε > 0 is an arbitrary real number. Note that | f (x) − f (p)| = |x − p||x + p|.
To prove that f is continuous, we would like to bound the term |x + p|. We can use the
triangular inequality in the following way:
|x + p| = |x − p + 2p| ≤ |x − p| + 2|p|.
p
Hence, if δ < min{ ε/2, ε/4|p|}, it follows that
ε ε
| f (x) − f (p)| ≤ |x − p|2 + 2|p||x − p| < + =ε
2 2
whenever |x − p| < δ . 

Note that in Example 4.6, we choose a δ that works for whatever p we picked before.
That was not possible in Example 4.7. This property possessed by the identity function
is called uniform continuity.
Definition 3.5.1 — Uniform Continuity. Let M, N be metric spaces. A function
f : M → N is said to be uniformly continuous if for every ε > 0, there is a δ > 0 such
that for all x, y ∈ M:

d( f (x), f (y)) < ε whenever d(x, y) < δ .

Compare the definitions of continuity and uniform continuity. A function f : M → N

is continuous if:
(∀y ∈ M)(∀ε > 0)(∃δ > 0)(∀x ∈ M)(d(x, y) < δ =⇒ d( f (x), f (y)) < ε).
And it is uniformly continuous if:
(∀ε > 0)(∃δ > 0)(∀x ∈ M)(∀y ∈ M)(d(x, y) < δ =⇒ d( f (x), f (y)) < ε).
The only difference between these definitions is the placement of the quantifier
(∀y ∈ M). This means that for a function to be continuous, the δ chosen can depend
on y. However, for it to be uniformly continuous we have to find a δ that works for
whichever y is available in the domain.
It should be clear that every uniformly continuous function is also continuous but
not the other way around. The next theorem shows when the converse is valid.
3.5 Uniform Continuity 63

Theorem 3.5.1 Let M, N be metric spaces such that M is compact. If f : M → N is

continuous, then it is uniformly continuous.

Proof. If f is not uniformly continuous, there exists a ε > 0 such that for any δ > 0
there exists two points x, y ∈ M such that d(x, y) < δ but d( f (x), f (y)) ≥ ε. Taking δ =
1/n we get two sequences (xn ), (yn ) such that d(xn , yn ) < 1/n but d( f (xn ), f (yn )) ≥ ε
for each n ∈ N.
Since M is compact, (xn ) has a convergent subsequence (xnk ). Similarly, (ynk ) has a
convergent subsequence (ynk j ). As (xnk ) converges, so does (xnk j ). Using the fact that
d(xnk j , ynk j ) < 1/nk j we see that (xnk j ) and (ynk j ) converge to the same value, namely
a ∈ M. But f is continuous, so it follows that

lim f (xnk j ) = f (a) = lim f (ynk j ).

j→∞ j→∞

Hence, there is a positive integer j0 such that for all j > j0 :

d( f (xnk j ), f (a)) < ε/2 and d( f (xnk j ), f (a)) < ε/2.

So,
d( f (xnk j ), f (ynk j )) ≤ d( f (xnk j ), f (a)) + d( f (ynk j ), f (a)) < ε.

This absurd establishes the result.

 Example 3.8 In exercise 4.15 you’ll show that if f is uniformly continuous and
lim(xn − yn ) = 0 then
lim f (xn ) − f (yn ) = 0.
n→∞

Take f (x) = x2 , xn = n + 1/n and yn = n. Clearly lim(xn − yn ) = 0 but

1
lim f (xn ) − f (yn ) = lim n2 + 2 + − n2 = 2.
n→∞ n→∞ n2
Hence f is not uniformly continuous in any domain that contains {xn } and {yn }. Since
f is continuous, this implies R is not compact.§ 

 Example 3.9 While f (x) = x2 is not uniformly continuous in R, it is uniformly

continuous in any bounded set. Suppose E ⊂ R is bounded. That is, there are constants
a, b such that a < x < b for all x ∈ E. We conclude that 2a < x + y < 2b and then
|x + y| < max |2a|, |2b| for all x, y ∈ E. Hence,

| f (x) − f (y)| = |x + y||x − y| < (max{|2a|, |2b|})|x − y| < ε

whenever |x − y| < ε/(max{|2a|, |2b|}). Notice that δ = ε/(max{|2a|, |2b|}) works for
whichever points x, y in E we happen to choose. Hence f is uniformly continuous in E. 
§ As if we didn’t already knew that...
 64 Chapter 3. Functional Limits and Continuity

3.6 Exercises

Exercise 3.2 Show that

r
√ √ 1
q
lim x+ x+ x− x = .
x→∞ 2

Exercise 3.3 Let f : (0, ∞) → (0, ∞) be an increasing function with

f (2t)
lim = 1.
t→∞ f (t)

Prove that
f (mt)
lim =1
t→∞ f (t)
for any m > 0.

Exercise 3.4 Suppose f : R → R function which satisfies

lim [ f (x + h) − f (x − h)] = 0
h→0

for every x ∈ R. Does this imply that f is continuous?

Exercise 3.5 — Thomae’s Function. Consider the function f : R → R such that


1
 if x = 0
f (x) = 1/q if x = p/q ∈ Q \ {0} with q > 0 and gcd(p, q) = 1 .

0 if x ∈ R \ Q


Prove that f is continuous at all x ∈ R \ Q and discontinuous at all x ∈ Q.

Exercise 3.6 Prove that there is not a function g : R → R that is continuous at all
x ∈ Q and discontinuous at all x ∈ R \ Q.
Hint: for each n ∈ N, consider the set

An = {x ∈ R|∃δ > 0 such that |x−a| < δ and |x−b| < δ =⇒ |g(a)−g(b)| < 1/n}.

Prove
T∞
that An is open for each n and that g is continuous at x if, and only if x ∈
n=1 n . Suppose that Q can be written as a countable intersection of open sets and
A
show that this contradicts Baire’s theorem.
3.6 Exercises 65

Exercise 3.7 — Norms are Continuous. If X is a normed space, show that the
function f : X → R such that f (x) = kxk is continuous.

Exercise 3.8 Let M, N be metric spaces such that M is compact. If f : M → N is a

continuous bijection, show that f −1 is also continuous.

Exercise 3.9 Let f , g : [0, 1] → [0, ∞) be continuous functions satisfying

sup f (x) = sup g(x).

0≤x≤1 0≤x≤1

Prove that there exists t ∈ [0, 1] with f (t)2 + 3 f (t) = g(t)2 + 3g(t).

Exercise 3.10 Prove that a continuous function from R to R which maps open sets
to open sets must be monotonic.

Exercise 3.11 Let f be a continuous function from R to R such that | f (x) − f (y)| ≥
|x − y| for all x and y. Prove that the range of f is all of R.

Exercise 3.12 — Croft’s Lemma. Let f : (0, ∞) → R be a continuous function with

the property that for any x > 0, limn→∞ f (nx) = 0. Prove that limx→∞ f (x) = 0.

Exercise 3.13 — Lipschitz Functions. Let E be a subset of R. A function f : E → R

is said to be Lipschitz if there exists a real number M > 0 such that

| f (x) − f (x)| ≤ M|x − y|

for all x, y ∈ E. Prove that if a function is Lipschitz, then it is uniformly continuous.

Show that the converse does not hold.

Exercise 3.14 Prove that a function f : (a, b) → R is uniformly continuous if, and
only if there exists a continuous function f˜ : [a, b] → R such that f˜(x) = f (x) for all
x ∈ (a, b).

Exercise 3.15 Recall the definition of equivalent sequences given in Chapter 3 (Def-
inition 3.4.2). Let M, N be metric spaces and (xn ), (yn ) be two arbitrary equivalent
sequences in M. Prove that a function f : M → N is uniformly continuous if and only
if ( f (xn )) is equivalent to ( f (yn )).
66 Chapter 3. Functional Limits and Continuity

Exercise 3.16 Let M, N be metric spaces. If f : M → N is uniformly continuous and

(xn ) is a Cauchy sequence in M, show that ( f (xn )) is a Cauchy sequence in N.

Exercise 3.17 Let M, N be metric spaces and p be a limit point of M. If f : M → N

is uniformly continuous, show that the limit limx→p f (x) exists.

Exercise 3.18 Let A, B,C be metric spaces. If f : A → B and g : B → C are uniformly

continuous, show that g ◦ f : A → C is uniformly continuous.

Exercise 3.19 Let f : [0, ∞) → R be a continuous function such that

lim f (x)
x→∞

exists and is finite. Prove that f is uniformly continuous.

Exercise 3.20 Let M, N, E ⊂ M be metric spaces such that E is bounded. If f : M →

N is uniformly continuous, show that f (E) is bounded.

Exercise 3.21 Show that the requirement in the definition of uniform continuity can
be rephrased as follows, in terms of diameters of sets: To every ε > 0 there exists a
δ > 0 such that diam f (E) < ε for all E ⊂ X with diam E < δ .
 CHAPTER
The Derivative
4
The motivation for the derivative comes from geometrical investigations. That perspec-
tive, together with countless applications, are the bulk of any calculus book. In analysis
our outlook will be focused on the properties and generalizations of such object.
Unlike the previous chapters, we present firstly the usual derivative of real functions
and then consider generalizations. This is due the fact that there is not a single possible
generalization and it is not obviously clear how one should generalize the derivative to
broader spaces than the real line.

4.1 The Derivative of a Real Function

Definition 4.1.1 — Derivative of a Real Function. Let f : E ⊂ R → R be a function
and p be a point of E. It is said that f is differentiable at p if the limit

f (x) − f (p)
lim
x→p x− p
exists. In this case its value is said to be the derivative of f at p and it is denoted
f 0 (p) or dd xf (p). We then associate with f a function f 0 , defined in the points where
f is differentiable, such that f 0 (x) is the derivative of f at x. This function is, with a
slightly abuse of language, also called the derivative of f .

 Leibniz’s notation dd xf makes the derivative appear to be the quotient of two quantities.
Although this interpretation is possible (at least in the case of a real function), it is not
simple at all. The reader is safer thinking about dd xf as a single expression.
It may be possible that the derivative of f is also differentiable. In this case its
2
derivative is the second derivative of f and it is denoted as f 00 , f (2) or even dd x2f . In
dn f
general, we denote the n-th derivative of f as f (n) or d xn . If f has derivatives of all
orders, then it is said to be smooth.
68 Chapter 4. The Derivative

Theorem 4.1.1 If f : E ⊂ R → R is differentiable at p, then it is continuous at p.

Proof. Since x − p → 0 and ( f (x) − f (p))/(x − p) → f 0 (p) as x → p,

f (x) − f (p)
lim f (x) − f (p) = lim · (x − p) = f 0 (p) · 0 = 0.
x→p x→p x− p
Hence, f is continuous at p.

The converse does not hold, as it can be seen in the following example.
 Example 4.1 Consider f : R → R such that f (x) = |x|. If p > 0, then

f (x) − f (p)
f 0 (p) = lim = 1.
x→p x− p
Similarly, if p < 0 we have that f 0 (p) = −1. However, f is not differentiable at 0 as
|x|
lim
x→0 x
does not exist. Hence f is continuous but not differentiable at 0. 

As we’ll see in the seventh chapter, there are continuous functions on the real line
that are nowhere differentiable.∗
Theorem 4.1.2 Suppose f and g are differentiable at p. Then so are f + g, f g and
f /g (provided g(p) 6= 0 in the last case). Moreover,

( f + g)0 (p) = f 0 (p) + g0 (p), ( f g)0 (p) = f 0 (p)g(p) + f (p)g0 (p)

and
f 0 (p)g(p) − f (p)g0 (p)
( f /g)0 (p) = ,
g(p)2
if g(p) 6= 0.

Proof. The proof of this theorem is based on two algebraic identities. Letting
x → p in
f (x)g(x) − f (p)g(p) g(x) − g(p) f (x) − f (p)
= f (x) · + g(p) ·
x− p x− p x− p
results in ( f g)0 (p) = f 0 (p)g(p) + f (p)g0 (p). Similarly,
 
f (x)/g(x) − f (p)/g(p) 1 f (x) − f (p) g(x) − g(p)
= g(p) · − f (p) ·
x− p g(x)g(p) x− p x− p
results in the corresponding identity for the derivative of the quotient. The derivative of
the sum is trivial.

∗ Actually, in some sense, most continuous functions are nowhere differentiable.

 4.2 Mean Value Theorems 69

 Example 4.2 Consider f : R → R such that f (x) = 1/x. Since

f (x) − f (p) 1
=−
x− p xp
and f is continuous at every p 6= 0 it follows that f 0 (p) = −1/p2 . From this example
and the previous theorem it follows that if f (x) = xn then f 0 (p) = npn−1 . (If n ≤ 0 it is
needed that p 6= 0.) 

Arguably, the following theorem is the most important result about derivatives of
real functions. It is usually denoted by the name of "chain rule" in calculus.

Theorem 4.1.3 — Chain Rule. Suppose f , g : R → R are functions such that f is

differentiable at p ∈ R and g is differentiable at f (p). Then g ◦ f is differentiable at p
and its derivative is
(g ◦ f )0 (p) = g0 ( f (p)) f 0 (p).

Proof. The naive approach would be to write

g( f (x)) − g( f (p)) g( f (x)) − g( f (p)) f (x) − f (p)
= ·
x− p f (x) − f (p) x− p
and let x → p in both sides. This does not work since it may happen that f (x) = f (p)
for every x in a neighborhood of p. To avoid this detail we define a function h : R → R
such that
 g(y) − g( f (p)) if y 6= f (p)


h(y) = y − f (p) .
 0
g ( f (p)) if y = f (p)
Now, if f (x) 6= f (p) we have that
g( f (x)) − g( f (p)) f (x) − f (p)
= h( f (x)) · .
x− p x− p
If f (x) = f (p) while x 6= p, then both sides of the previous equation equal zero and
hence this equation holds for all x ∈ R \ {p}.
Our function h is continuous at f (p) since g is differentiable at f (p). Hence, letting
x → p we get the desired result.

4.2 Mean Value Theorems

4.3 L’Hôpital’s Rule
4.4 Fréchet Derivative
4.5 Partial Derivatives
4.6 The Implicit / Inverse Function Theorem
4.7 Exercises
 APPENDIX
Logic and Set Theory
A
David Hilbert predicted in the beginning of the 90’s, “no one will drive us from
A S
the paradise which Cantor created for us”. Set theory is the basis of almost all
areas in mathematics and analysis is no exception.
We will adopt the naive point of view to set theory, which assumes that the concept
of a set of objects is intuitively clear. The only remark that will be made is that we will
only allow sets to be defined by a property of its elements if it is a subset of a known
set. Of course this is not formal, but I will leave the task of studying sets for a set theory
book.
The only reason for that remark is to avoid contradictions such as Russell’s paradox∗ .
An formal axiomatization is well beyond our scope. Luckily, naive set theory works
wonderfully for all the sets we need to use in analysis.

A.1 Equivalence Relations

Definition A.1.1 An relation between two sets A and B is a subset R of the Cartesian
product A × B.

Since most of the relations we encounter on analysis are denoted by x = y, x > y,

x ∼ y or other similar notations, we denote a general relation by xRy instead of (x, y) ∈ R.
A function f : A → B is a pretty particular case of relation: specifically, one which
every x ∈ A appears exactly once as the first element of (x, f (x)) ∈ A × B.

Definition A.1.2 An equivalence relation in a set A is a relation ∼ in A × A (There

is no special reason to denote a relation by a letter. The symbol ∼ works just fine.)

∗ Consider the set R of all the sets X that are elements of itself. That is, R = {X | X ∈
/ X}. What would
happen if R ∈ R? And if R ∈/ R?
72 Chapter A. Logic and Set Theory

that satisfies the following properties:

• (Reflexitivity) For all x ∈ A, x ∼ x.
• (Symmetry) If x ∼ y, then y ∼ x.

• (Transitivity) If x ∼ y and y ∼ z, then x ∼ z.

If A and B are sets, it will be implicit that A ∼ B if, and only if, there is a bijection
between the elements of A and B. You should verify that this is, in fact, an equivalence
relation. This allows us to define the cardinality of a set. If A is a non-empty set, we say
that |A| = n if A ∼ {1, 2, . . . , n} for some n ∈ N. By definition, |∅| = 0.
We can then extend the concept of cardinality to infinite sets. In this case, we say
that |A| ≤ |B| if there is an injection from A to B. The equality occurs only when A ∼ B.
In 1878 Georg Cantor conjectured that there was not a set S such that |N| < |S| < |R|.
This became known by the name of “Continuum Hypothesis”. In 1940 Kurt Gödel
showed it was impossible to disprove the continuum hypothesis using ZFC. Then Paul
Cohen showed in 1963 that it cannot be proved either (this means that we should take
the continuum hypothesis or it’s negation as an axiom). That was all assuming that ZFC
is consistent, which we still don’t know for sure. This is still an active area of research.
Definition A.1.3 Let ∼ be an equivalence relation of a set A and x be an element of
A. An equivalence class is a subset [x] ⊂ A defined by

[x] = {p ∈ A | p ∼ x}.

Obviously, equivalence classes are non-empty, since equivalence relations are reflex-
ive.

Theorem A.1.1 Two equivalence classes [x] and [y] are either disjoint or equal.

Proof. Let [x] be the equivalence class determined by x, and let [y] be the equiva-
lence class determined by y. Suppose that [x] ∩ [y] is not empty. So, let z ∈ [x] ∩ [y].

A
[x] [y]
w z y
x

By definition, we have z ∼ x and z ∼ y. Symmetry allows us to conclude that x ∼ z

and z ∼ y. From transitivity, it follows that x ∼ y. If w is any point of [x], we have w ∼ x
by definition, and w ∼ y by transitivity. We conclude that [x] ⊂ [y].
The symmetry of the situation allows us to conclude that [y] ⊂ [x] as well, so that
[x] = [y].
 A.2 Countability 73

A partition P on a set A is a collection of subsets of A with the following property:

every element of A belongs to exactly one of the sets in P. Notice that the set of all
equivalence classes for a given equivalence relation satisfies exactly this! We have then
that every partition induces an equivalence relation in A and every equivalence relation
induces a partition. We denote the set of all equivalence classes of a equivalence relation
∼ defined of a set A by A/ ∼.

A.2 Countability
Definition A.2.1 For any set A, we say:

• A is finite if |A| = n for some n in {0, 1, 2, . . . }.

• A is infinite if A is not finite.
• A is countable if A ∼ N.
• A is uncountable if A is neither finite nor countable.

• A is at most countable if A is finite or countable.

It should be clear now that, if C is countable, U is uncountable and M is at most
countable, then |M| ≤ |C| ≤ |U| and |C| < |U|. Countable sets are sometimes called
enumerable, or denumerable. While finite sets cannot be equivalent to one of its proper
subsets, this is possible for infinite sets, like the example shown below:
 Example A.1 Let Z be the set of all integers. Consider the following mapping:
n
 if n is even
f : N → Z, f (n) = 2 n − 1
− if n is odd
2
f is a bijective map from N to Z. So, Z is countable. 

This suggests that we might characterize infinite sets as the ones which are equivalent
to some proper subset of itself. It is true and you should try to prove it for countable sets.
The proof that every infinite (not necessarily countable) set is equivalent to some
proper subset of itself is not as easy and will be omitted.

Theorem A.2.1 Every infinite subset of a countable set A is countable.

Proof. Suppose E ⊂ A, and E is infinite. Since A is countable, we can write it as

{a1 , a2 , . . . }, where all the an are distinct. Construct then a sequence (nk ) as follows:
Let n1 be the smallest positive integer such that an1 ∈ E. Having chosen n1 , . . . , nk−1 ,
let nk be the smallest integer greater than nk−1 such that ank ∈ E.
Putting f (k) = ank , we have a bijection between E and N.

This theorem shows that, roughly speaking, countable sets represent the “smallest”
infinity. That is, no uncountable set can be a subset of a countable set.
74 Chapter A. Logic and Set Theory

The following lemma will be very useful to prove some important theorems.

Lemma A.2.2 Let (En ), n = 1, 2, . . . , be a sequencea of countable sets and put

∞
[
S= En .
n=1

Then S is countable.
a Formally, a sequence is a function from N to any set. It is usual to denote the element f (i) as x and to
i
denote the whole function as (xn ).

Proof. Let every set En be arranged in a sequence (enk ), k = 1, 2, . . . , and consider

the infinite array

e11 e12 e13 e14 ···

e21 e22 e23 e24 ···
e31 e32 e33 e34 ···
e41 e42 e43 e44 ···
.. .. .. .. ..
. . . . .
in which the elements of En form the nth row. The array contains all elements of S. As
indicated by the arrows, these elements can be arranged in a sequence
e11 , e21 , e12 , e31 , e22 , e13 , . . . .
If any two of the sets En have elements in common, these will appear more than once
in our sequence. Hence there is a subset T of the set of all positive integers such that
S ∼ T , which shows that S is at most countable. Since E1 ⊂ S, and E1 is infinite, S is
infinite, and thus countable.

Theorem A.2.3 Let A and B be countable sets. Then the cartesian product A × B is
countable.

Proof. Name the elements of A and B as follows:

A = {a1 , a2 , a3 , . . . }, B = {b1 , b2 , b3 , . . . }.
We can then use the array of Lemma 1.5.2

(a1 , b1 ) (a1 , b2 ) (a1 , b3 ) (a1 , b4 ) · · ·

(a2 , b1 ) (a2 , b2 ) (a2 , b3 ) (a2 , b4 ) · · ·
(a3 , b1 ) (a3 , b2 ) (a3 , b3 ) (a3 , b4 ) · · ·
(a4 , b1 ) (a4 , b2 ) (a4 , b3 ) (a4 , b4 ) · · ·
.. .. .. .. ..
. . . . .
A.2 Countability 75

to create the sequence

(a1 , b1 ), (a2 , b1 ), (a1 , b2 ), (a3 , b1 ), (a2 , b2 ), (a1 , b3 ), . . . .

The result follows with the same argument we used in the end of Lemma 1.5.2.

Corollary A.2.4 Let A be a countable set, and let An be the set of all n-tuples of
elements of A. Then, An is countable.

Proof. We approach by induction.

A1 is obviously countable, since A1 = A. Suppose Ak−1 is countable. The set Ak is
equivalent to A × Ak−1 . Theorem 1.5.3 then implies that Ak is countable. The result
follows by induction.

Cartesian product isn’t associative. An element of (A × B) ×C is of the form ((a, b), c),
 while an element of A × B × C is of the form (a, b, c). However these two sets are
clearly equivalent as there is an obvious bijection between them. In the preceding proof,
A × Ak−1 6= Ak but A × Ak−1 ∼ Ak .

Corollary A.2.5 The set of all rational numbers is countable.

Proof. There is a bijection from the set of all fractions b/a (while −1 1
−2 and 2 are
certainly the same number, we will consider them to be different fractions) to the set Z2
of ordered pairs (b, a). Of course there is an injection from Q to the set of all fractions,
so Q is at most countable. But since N ⊂ Q and N is infinite, Q is countable.

We shall now present our first uncountable set.

Theorem A.2.6 Let A be the set of all sequences whose elements are the digits 0 and
1. Then, A is uncountable.

Proof. Of course A is infinite, so let E be a countable subset of A. Then arrange

the elements of E = {e1 , e2 , . . . } in a array similar to that of lemma 1.5.2. (Here we
used indices to denote the elements of E and arguments to denote the elements of each
sequence.)

e1 : e1 (1) e1 (2) e1 (3) e1 (4) · · ·

e2 : e2 (1) e2 (2) e2 (3) e2 (4) · · ·
e3 : e3 (1) e3 (2) e3 (3) e3 (4) · · ·
e4 : e4 (1) e4 (2) e4 (3) e4 (4) · · ·
.. .. .. .. .. ..
. . . . . .
 76 Chapter A. Logic and Set Theory

Define a new sequence b such that b(n) = 1 − en (n). Notice that b is an element of
A and that b is not a element of E, since it is different from each one of its elements in at
least one place.
So, it’s been shown that every countable subset of A is a proper subset of A. It follows
that A is uncountable (for otherwise A would be a proper subset of itself).

The idea of the above proof was first used by Georg Cantor, and is called Cantor’s
diagonal argument.

Theorem A.2.7 A complex number z is said to be algebraic if there are integers

a0 , . . . , an , not all zero, such that

a0 zn + a1 zn−1 + . . . + an−1 z + an = 0.

Then, the set of all algebraic numbers is countable.

Proof. Let

Am = {z ∈ C | a0 zn + . . . + an = 0 and n + |a0 | + . . . + |an | = m}.

The set Am is finite, since each equation has only a finite set of solutions and there are
only finitely many equations satisfying n + |a0 | + . . .S+ |an | = m for each m.
∞
The set of all algebraic numbers, i.e. the union m=2 Am , is then at most countable.
Since every rational number is algebraic and there are infinite rationals, the set of all
algebraic numbers is exactly countable.

Notice how fortunate the preceding result is: at the present stage we can’t even prove
that the sum of two algebraic numbers is algebraic, let alone that the set of all algebraic
numbers is a field (it is). Yet we can prove that it is countable. To most people, the only
two non-algebraic (i.e. transcendental) numbers they’ll find in their lives are π and e
even though almost all complex numbers are transcendental.

A.3 Exercises
Exercise A.1 Let A be a finite set. Prove that cardinality is a well defined function.
That is, if |A| = n and |A| = m, then n = m.

Exercise A.2 — Inclusion-Exclusion Principle. If A and B are finite sets, prove that
|A ∪ B| = |A| + |B| − |A ∩ B|.
Prove a similar formula for 3 sets.

Exercise A.3 — Dirichlet’s Box Principle. Let A and B be finite sets such that
|A| > |B|. Prove that there is no injection from A to B.
It is usual to write this principle in the following way:
A.3 Exercises 77

"If you have m objects to put in n < m boxes, then at least one box will have more
than one object."
Hint: It may be useful to prove the result of Exercise 9 for finite sets.

Exercise A.4 — Cantor-Schröder-Bernstein. If you are feeling really brave today,

try to prove the following theorem:
Suppose there exist functions f : A → B and g : B → A such that f and g are injective.
Prove that there is a bijection from A to B.
It may be useful to partition the sets A and B in four parts A1 , A2 , B1 , B2 such that

• A1 ∪ A2 = A • B1 ∪ B2 = B • f (A1 ) = B1

• A1 ∩ A2 = ∅ • B1 ∩ B2 = ∅ • g(B2 ) = B2 .

Exercise A.5 Prove that no order can be defined in the complex field that turns it
into an ordered field.
Hint: −1 is a square.

Exercise A.6 Prove that every ordered field contains a copy of the rational field.
(Actually it may contain a set that is isomorphic to the rational field. That is, a set
that is equivalent to Q and preserves all the ordered field operations.)

Exercise A.7 Let p be a fixed prime number. If x, y ∈ Z define the equivalence

relation x ∼ y to hold if, and only if p divides x − y. Check that this is, in fact, a
equivalence relation.
We denote by Z/pZ the set of all equivalence classes of ∼. In this set we’ll define
two operations + and ·. If a is the equivalence class determined by α and b is the
equivalence class determined by β , then a + b = [α + β ] and a · b = [αβ ].
Prove that Z/pZ is a field.

Exercise A.8 In the proof of Lemma 1.5.2 we ordered the elements enk in a sequence
e11 , e21 , e12 , e31 , e22 , e13 , . . . . That is, we implied that there is a bijection f : N → N2
such that f (1) = (1, 1), f (2) = (2, 1), f (3) = (1, 2), f (4) = (3, 1) and so on.
Find this function explicitly and prove that it is a bijection.

Exercise A.9 — The Real Field is Unique up to a Isomorphism. Let F be an

ordered field with the least-upper-bound property endowed with sum ⊕, product ⊗
and order ≺. Denote by 0 and 1 the additive and multiplicative identities of F and by
n = 1 ⊕ 1 ⊕ . . . ⊕ 1 (n times). For negative numbers we define −n = −n.
78 Chapter A. Logic and Set Theory

We’ll define a function f : R → F as

f (p/q) = p/q for all p/q ∈ Q

and, for irrational x,

f (x) = sup{p/q ∈ F | p/q < x}.
Prove that f is an isomorphism. That is, f is an bijection and for all x, y ∈ R it holds
that f (x + y) = f (x) ⊕ f (y) and f (xy) = f (x) ⊗ f (y). It also holds that x < y implies
f (x) ≺ f (y). (This is another hard exercise.)

Exercise A.10 Let α be a irrational number. We denote by {x} the fractional part
of x. That is, the only real number that satisfies 0 ≤ {x} < 1 and x = m + {x} for
some integer m.
Prove that the set A = { {nα} | n ∈ N} is dense in [0, 1). That is, for all non-empty
subsets (a, b) ⊂ [0, 1) there is a real number x ∈ A such that a < x < b.
You may want to partition the interval [0, 1) in the following way
     
1 1 2 n−1
[0, 1) = 0, ∪ , ∪...∪ ,1
n n n n

and use Dirichlet’s box principle.

Exercise A.11 — Dirichlet’s Approximation Theorem. Use the preceding result to

show that, given α ∈ R \ Q and n0 ∈ N, there exists p ∈ Z and q ∈ N such that

p 1
α− < .
q qn0

Exercise A.12 — Pythagorean Theorem. Prove the following theorem:

If x · y = 0, then kx + yk2 = kxk2 + kyk2 .
Show that the converse does not necessarily hold.

Exercise A.13 — Parallelogram Identity. Prove the following identity:

kx + yk2 + kx − yk2 = 2 kxk2 + 2 kyk2 .
What does this identity mean geometrically for vectors of R3 ?

Exercise A.14 Let X be an infinite set. Prove that X contains a countable subset.
This exercise is meant to be easy. However, your solution is surely wrong if we do
not assume the Axiom of Choice. Now would be a good time to delight yourself with
A.3 Exercises 79

a bit of math.
Take a look at the precise statements of the Axiom of Choice and the Well Ordering
Theorem. It should be almost clear that the Axiom of Choice is "obviously" right
and the Well Ordering Theorem is "obviously" wrong. Yet they are equivalent...

Exercise A.15 Let B be the set of all bijections from N to itself. Prove that B is
uncountable.
Hint: use a variation of Cantor’s diagonal argument here.

Exercise A.16 — Cantor’s Theorem. Let X be a set and P(X) be the set of its
subsets. We will prove that there is no bijection from X to P(X).
Suppose there is a bijection f : X → P(X) and consider the set A = {x ∈ X | x ∈/
f (x)}. Since A is a subset of X, there is an element a ∈ X such that A = f (a). What
can we conclude from here?

Exercise A.17 At first it seems like we could use Cantor’s diagonal argument to
prove that (0, 1) is uncountable in the following way.
Suppose (0, 1) is countable, then list the decimal expansions of all its elements. We
create a new number that is in (0, 1) but was not counted by switching the i-th digit
of the i-th number in our sequence from k to k + 1 if k 6= 9 and to 0 if k = 9.
This is not a valid proof. Why?

Exercise A.18 Let B be a family of disjoint intervals. Prove that B is at most

countable. Notice the paramount importance of the word "disjoint" here. If the
intervals were not disjoint, then B could be the set of all intervals, which is equivalent
to R2 . (Why?)
 APPENDIX
The Construction Of The
Real Field
B
B.1 Introduction
The real line has exceptional importance in analysis basically because it is continuous.
In some sense, it means that it has no "gaps". We will see that, while the rational line
has lots of numbers, it has even more gaps than numbers.

Theorem B.1.1 There is no rational x such that x2 = 2. Moreover, the set A = {x ∈

Q+ | x2 < 2} has no largest element and the set B = {x ∈ Q+ | x2 > 2} has no
smallest element.

Proof. Suppose there was a rational x such that

x2 = 2.

Since x is rational, we can write x as p/q, where p and q are integers not both even. We
get then that
p2 = 2q2 ,
which implies that p2 and hence p is even. Since p is even, we can write p as 2k, for
some integer k, and conclude that q is even too. This contradicts our assumption that p
and q are not both even.
Suppose now that B has a smallest element b and consider a = 2/b. Of course
a 6= b, as that would imply a rational solution to x2 = 2. Since a2 = 4/b2 < 4/2 = 2, we
conclude that a ∈ A and hence a < b. However (a − b)2 > 0 implies ( a+b 2
2 ) > 2, which
contradicts the minimality of b as (a + b)/2 ∈ B and (a + b)/2 < b.
Suppose A had a largest element a and consider the numbers b = 2/a and c =
(a + b)/2. Similarly to what we just did, it follows that a < c < b. Writing b and c in
terms of a, we get that ( a+c 2 2 2 2
2 ) − 2 = 9(a − 2)(a − 2/9)/16a < 0 for 2/9 < a < 2.
2

This is a contradiction to the maximality of a.

 82 Chapter B. The Construction Of The Real Field

As the set of all real numbers is so important in analysis, now would be a good time
to construct such set. However, I find it’s hard for most students to appreciate such
construction and to notice why it is important at this point. My approach will be to show
some standard set-theoretic definitions and we’ll define the set of all real numbers as
the set that satisfies some desirable properties. As soon as we have the machinery of
Cauchy sequences I shall present one of the possible constructions. I also recommend
the reading of Rudin’s[13] first chapter for an algebraic approach to the real number set.

B.2 Ordered Fields

To characterize the real number system, we will define a set of axioms for an algebraic
structure called field and then say that the real set is a "special" kind of field.
Definition B.2.1 A field is a set F with two operations, + and · (both functions from
F × F to F) that satisfies the following axioms.
• (Closure) For all x, y in F, both x + y and x · y are elements of F.
• (Associativity) For all x, y, z in F the following holds: (x + y) + z = x + (y + z)
and (x · y) · z = x · (y · z).

• (Commutativity) For all x, y in F the following holds: x + y = y + x and

x · y = y · x.
• (Existence of identity) There exists two distinct elements of F denoted 0 and 1
such that for all x in F, 0 + x = x and 1 · x = x.

• (Existence of inverses) For all x in F there exists an element denoted −x such

that x + (−x) = 0. Likewise, for all x 6= 0 in F there exists an element denoted
x−1 such that x · x−1 = 1. The elements a + (−b) and a · b−1 are also denoted
a − b and a/b respectively.
• (Distributivity) For all x, y, z in F, the following holds: x · (y + z) = x · y + x · z.

It is usual to drop the dot and write products as xy instead of x · y.

It is not hard to memorize the axioms for the frequently used algebraic structures
because they are essential to some important sets. For example, both R and Q are fields.
However, R and Q are not the only fields that exists. We need further characterization to
know what really is R.
Definition B.2.2 — Ordered Sets. An ordered set is a set S, together with a relation
> such that:
• (Trichotomy) For any x, y ∈ S, exactly one of x > y, x = y, or y > x holds.

• (Transitivity) If x > y and y > z, then x > z.

We say that x < y if and only if y > x.
B.2 Ordered Fields 83

The set of all rational numbers becomes an ordered set when we define x > y if
x − y = p/q where p, q ∈ N. Similarly, Z and N are ordered sets with the usual order
relation.
It is customary to define x ≥ y to mean x = y or x > y, without specifying which one
applies. The notation x ≤ y is defined in the obvious way.
It is not obvious at all that the order we defined earlier for the cardinality of sets is in
 fact a order relation. Trichotomy holds if, and only if |A| ≥ |B| and |B| ≥ |A| implies
|A| = |B|, which is a result called Cantor-Schröder-Bernstein Theorem.

Definition B.2.3 Let S be an ordered set and E be a subset of S.

• If there exists an element u of S such that x ≤ u for all x ∈ E, we say that E is

bounded above and u is an upper bound of E. The least upper bound of a set
(if it exists) is called the supremum of E and is denoted sup E.
• If there exists an element l of S such that x ≥ l for all x ∈ E, we say that E is
bounded below and l is an lower bound of E. The greatest lower bound of a
set (if it exists) is called the infimum of E and is denoted inf E.

Notice that a supremum or infimum of E need not be in E. For example, the set
E = {x ∈ Q | x > 0} has infimum 0 and 0 is not an element of E. The set Z has no
infimum and no supremum since it is not bounded above or below.
Definition B.2.4 An ordered set S is said to possess the least-upper-bound property
if every nonempty subset E ⊂ S that is bounded above has a least upper bound in S.
That is, for all non-empty set E ⊂ S that is bounded above, sup E exists in S.

This property is sometimes called the Dedekind completeness property. Theorem

1.1.1 showed that the set of all rational numbers does not possess the least-upper-bound
property.
Definition B.2.5 An ordered field is a field F which is also an ordered set and satisfies
the following axioms.

• For all x, y, z in F such that x > y we have that x + z > y + z.

• For all x, y in F such that x > 0 and y > 0 we have that xy > 0.

If x > 0 we say that x is positive and if x < 0 we say that is it negative.

You should prove that some familiar properties of Q follow directly from the ordered
field axioms. Once we know that every field satisfies these properties we won’t need to
prove them again for R and for C.

Exercise B.1 Prove the following statements using only the field axioms.
All letters are meant to denote elements from a field.

a) If x + y = x + z then y = z.
 84 Chapter B. The Construction Of The Real Field

b) If x 6= 0 and xy = xz then y = z.
c) 0x = 0.
d) If x 6= 0 and y 6= 0 then xy 6= 0.

The following properties hold for ordered fields. Notice that, while the complex
number set is a field, it is not an ordered field.

Exercise B.2 Prove the following statements using only the ordered field axioms.a
(That means you can only use definitions 1.3.1, 1.3.2 and 1.3.5.)
All letters are meant to denote elements from an ordered field.
a) If x is positive then −x is negative.
b) If x is positive and y > z then xy > xz.

c) If x is negative and y > z then xy < xz.

d) If x 6= 0 then x2 > 0.
a Even if you managed to do all the items in Exercise 1.1 and 1.2 you should check the solutions in the

end of the book.

B.3 The Real and Complex Fields

We are now ready to define the set of all real numbers.
Definition B.3.1 — The Real Field. The real field is the only ordered field with the
least-upper-bound property. Moreover, Q is a subfield of R. That is, Q ⊂ R and Q
inherits its order relation and its field operations from R.

Well, we can always rename the elements of R to create another set that satisfies
 definition 1.4.1. Technically we say that the real number set is the only ordered field
with the least-upper-bound property up to a isomorphism. And that is a theorem that I
will not prove.
One of the most important consequences of the least-upper-bound property in the
real line is the Archimedean property. It roughly says that the set {x ∈ R | x > 0} has no
smallest element.
Theorem B.3.1 — Archimedean Property. For all ε > 0 in R there exists a natural
n such that 1/n < ε.

Proof. Suppose the theorem is false. Then there exists x0 > 0 in R such that no
natural n makes 1/n < x0 true. That is, for all n ∈ N we have that nx0 ≤ 1. Let

S = {y ∈ R | y = nx0 , n ∈ N}.

Since S is bounded above by 1 and S is a subset of R, S has a least upper bound. We

B.3 The Real and Complex Fields 85

shall call it k.
Of course we have that k − x0 < k since x0 is positive. But then k − x0 is not an upper
bound of S, so mx0 > k − x0 for some m ∈ N. That means k < (m + 1)x0 , which is absurd
since k is an upper bound of S.

For now, we’ll say that a set E ⊂ R is dense in R if, and only if for all a, b in R such
that b > a there exists c in E such that a < c < b. (We will generalize this definition to
metric spaces in the following chapter.)

Theorem B.3.2 The set of all rational numbers is dense in the set of all real numbers.

Proof. I will assume a > 0 without loss of generality.

By the Archimedean property, there exists an integer n0 such that 1/n0 < b − a. Let

k
M = {m ∈ Q | m = , k ∈ N and m ≤ a}.
n0

Let s = max M. Of course s ≤ a. Since s is the maximum of M, s + 1/n0 ∈

/ M. That is,
s + 1/n0 > a. Since 1/n0 < b − a, we have that

a < s + 1/n0 < b.

| {z }
c∈Q

The result follows.

Corollary B.3.3 The set of all irrational numbers is dense in the set of all real
numbers.
Proof. Substitute M in the proof of Theorem 1.4.2 by
√ k
M 0 = {m ∈ R \ Q | m =
2 , k ∈ N and m ≤ a}
n0
√
and c = max M + 1/n0 by c0 = max M 0 + 2/n0 .

We now define another set called the extended real line.

Definition B.3.2 — Extended Real Line. The extended real line is the set R =
R ∪ {−∞, +∞} which inherits R order, operations and satisfies

−∞ < x < +∞

for all x ∈ R.
We also define the following operations, which are valid for all x ∈ R:
86 Chapter B. The Construction Of The Real Field

• x + ∞ = +∞ if x 6= −∞ • x/(+∞) = 0 if x 6= ±∞
• x − ∞ = −∞ if x 6= +∞ • x/(−∞) = 0 if x 6= ±∞
• x · (+∞) = +∞ if x > 0 • +∞/x = +∞ if x > 0 and x 6= ±∞
• x · (+∞) = −∞ if x < 0 • +∞/x = −∞ if x < 0 and x 6= ±∞
• x · (−∞) = −∞ if x > 0 • −∞/x = −∞ if x > 0 and x 6= ±∞

• x · (−∞) = +∞ if x < 0 • −∞/x = +∞ if x < 0 and x 6= ±∞

The operations +∞ − ∞, 0 · (±∞) and ±∞/ ± ∞ are left undefined.
Notice that x + ∞ means both x + (+∞) and x − (−∞), while x − ∞ means both
x + (−∞) and x − (+∞). If the context makes it clear, we’ll usually write ∞ instead of
+∞. It should be clear that R is not a field.
A very important property of R, which R lacks, is that every subset has a supremum
and a infimum. In particular, the extended real line has the least-upper-bound property.
To define the set of all complex numbers rigorously, we’ll characterize them as
ordered pairs of real numbers and then see how the real line "sits" inside this set.
Definition B.3.3 — The Complex Field. The complex field is the set of all ordered
pairs (a, b) of real numbers such that

• For all a, b, c, d in R: (a, b) + (c, d) = (a + c, b + d).

• For all a, b, c, d in R: (a, b) · (c, d) = (ac − bd, ad + bc).
• (a, b) = (c, d) implies a = c and b = d.

You should prove that this is, in fact, a field. For every complex number z = (a, b),
we define a to be the real part Re(z) and b to be the imaginary part Im(z) of z.

Exercise B.3 Prove that the complex field, with the operations defined above and
with (0, 0) and (1, 0) as the sum and product identities respectively, satisfies all the

−b
field axioms. You should define −(a, b) as (−a, −b) and (a, b)−1 as a2 +b a
2 , a2 +b2 .

Since (a, 0) + (b, 0) = (a + b, 0) and (a, 0) · (b, 0) = (ab, 0) we will write (a, 0) as
simply a from now on. Provided that we define i as (0, 1) we can write every complex
number as a + bi since a + bi = (a, b).
We shall then define the conjugate of the number a + bi as a − bi and√denote it by
a + bi. We define the absolute value of a complex number a + bi to be a2 + b2 and
denote it by |a + bi|.∗ While square roots are still undefined, this will soon be fixed.
∗ We had a few notations popping in different places here. They usually aren’t a coincidence. If A is a set,

|A| measures roughly how "big" a set is and if z is a complex number, |z| measures roughly how "big" it is.
We’ll see in the next chapter that E denotes the closure of a set E. If (a, b) is a subset of R, its closure is [a, b].
It is based on this fact that we denote the extended real line as R. However, I don’t know a good reason for the
conjugate of a complex number to be defined with the same symbol as the closure of a set. Neither do I know
B.3 The Real and Complex Fields 87

You should prove now the usual properties of these operations.

Exercise B.4 Prove the following properties about complex conjugation and absolute
value. All letters are meant to be complex numbers.
a) z + w = z + w e) |zw| = |z||w|

b) zw = z · w f) z + z = 2 Re(z)

c) zz = |z|2 g) Re(z) ≤ |z|

d) |z| = |z| h) |z + w| ≤ |z| + |w|

Definition B.3.4 — Real and Complex Vector Spaces. If n is a positive integer,

we define Cn to be the set of all n-tuples of complex numbers. That is, the elements
of Cn are numbers of the form

x = (x1 , x2 , . . . , xn ),

where xi ∈ C for all 1 ≤ i ≤ n.

The elements of Cn are called vectors.
We define the following operations on Cn :
• (x1 , x2 , . . . , xn ) + (y1 , y2 , . . . , yn ) = (x1 + y1 , x2 + y2 , . . . , xn + yn ).
• If a is a complex number, a(x1 , x2 , . . . , xn ) = (ax1 , ax2 , . . . , axn ).

• (x1 , x2 , . . . , xn ) · (y1 , y2 , . . . , yn ) = ∑ni=1 xi yi

p
• k(x1 , x2 , . . . , xn )k = ∑ni=1 |xi |2 .
The set Rn is defined in the same way.
It will be usual to call the identity element of the sum (0, 0, . . . , 0) as simply 0. We’ll
call the set Rn an euclidean space. As usual, you should prove the following properties.

Exercise B.5 Prove the following properties about norms and inner products (the
product of vectors we just defined). All letters are meant to be elements of a vector
space (Rn or Cn ).
a) x · y = y · x. d) (x + y) · z = x · z + y · z.

b) If a ∈ C, (ax) · y = a(x · y). e) kxk2 = x · x.

c) If a ∈ C, x · (ay) = a(x · y). f) kxk = 0 implies x = 0.

One of the most important properties of this vector space (or any vector space in
general) is the Cauchy-Schwarz inequality.

why we denote ordered pairs the same way as we denote open intervals of the real line.
88 Chapter B. The Construction Of The Real Field

Theorem B.3.4 — Cauchy-Schwarz Inequality. Let x and y be vectors of Cn . Then

the following inequality holds.

|x · y| ≤ kxk kyk .

Proof. If x · y = 0, then the theorem holds trivially. Otherwise, let λ be the complex
number x · y/ kyk2 . Since the square of every real number is positive (or zero),

0 ≤ kx − λ yk2
= kxk2 − λ (x · y) − λ (y · x) + |λ |2 kyk2
|x · y|
= kxk2 − .
kyk2

This implies then that, |x · y| ≤ kxk kyk.

A quick corollary of the Cauchy-Schwarz inequality is a generalization of the

property proved in exercise 1.4.h.

Corollary B.3.5 — Triangle Inequality. Let x and y be vectors of Cn . Then the

following inequality holds.

kx + yk ≤ kxk + kyk .

Proof.

kx + yk2 = kxk2 + x · y + y · x + kyk2

= kxk2 + 2 Re(x · y) + kyk2
≤ kxk2 + 2|x · y| + kyk2
≤ kxk2 + 2 kxk kyk + kyk2
= (kxk + kyk)2 .

The Cauchy-Schwarz inequality was used to justify the fourth line.

 APPENDIX
Linear Algebra
C
Solutions

1- Naive Set Theory

Solution 1 Although this might seem obvious, these propositions need to be proved and
it is important to understand each step in order to obtain a solid knowledge about fields

a) Since x belongs to a field, by the property of existence of inverses, there exists -x

such that

x + y = x + z =⇒ (−x) + (x + y) = (−x) + (x + z)
=⇒ ((−x) + x) + y = ((−x) + x) + z (By associativity)
=⇒ 0 + y = 0 + z
=⇒ y = z.

b) By the existence of inverses, for all x 6= 0

xy = xz =⇒ (x−1 ) · (xy) = (x−1 ) · (xz)

=⇒ (x−1 x) · y = (x−1 x) · z (By associativity)
=⇒ 1 · y = 1 · z
=⇒ y = z.

c) Distributivity implies

0x = (a + (−a)) · x = ax + (−ax) = 0.

d) Suppose xy = 0 then, since y 6= 0, there exists y−1 such that yy−1 = 1. From (c),

xy(y−1 ) = 0(y−1 ) = 0
=⇒ x(yy−1 ) = 0 (By associativity)
=⇒ x · 1 = x = 0.

Which is a contradiction.

Remark: You might ask why we can add or multiply a number in both sides of an
equation. In fact, this is basically the formal definition of equality. Given two objects x
and y, we say that x = y if and only if, P(x) = P(y) for any predicate P.

91
92 Chapter J. Solutions

Solution 2

a) From definition 1.3.5, 0 < x implies (−x) < x + (−x) = 0.

b) By the associativity and the second part of definition 1.3.5 it follows that
y > z =⇒ y − z > 0 =⇒ x(y − z) > 0 =⇒ xy − xz > 0 =⇒ xy > xz.
c) From (a), (−x)(y − z) > 0. Hence, xy < xz.
d) By trichotomy, x > 0, x = 0 or x < 0. If x > 0, then by second part of definition
1.3.5 x2 > 0. If x < 0 however, from (b), x < 0 implies x2 > 0 · x, then by the result
from exercise 1.1(c), x2 > 0.
Solution 3 The closure property is assured by the definition given. Let x, y, z be
(a, b),(c, d) and (e, f ), respectively.
Associativity: (x + y) + z = (a + c, b + d) + (e, f ) = (a + c + e, b + d + f ) = (a, b) +
(c + e, d + f ) = x + (y + z).
Commutativity: x + y = (a + c, b + d) = (c + a, d + b) = y + x.
Existence of identity: 0 + x = (0 + a, 0 + b) = (a, b) = x; 1 · x = (a, b) · (1, 0) =
(a · 1 − b · 0, a · 0 + b · 1) = (a, b) = x.
Existence of  inverses: x + (−x)  =(a,2 b) +2 (−a, −b) = (a − a, b − b) = (0, 0) = 0;
a −b a + b a(−b) + ab
xx−1 = (a, b) · 2 , = , = (1, 0).
a + b2 a2 + b2 a2 + b2 a2 + b2
Distributivity: x · (y + z) = (a, b) · (c + e, d + f ) = (a(c + e) − b(d + f ), a(d + f ) +
b(c + e)) = (ac + ae − bd − b f , ad + a f + bc + be) = ((ac − bd) + (ae − b f ), (ad + bc) +
(a f + be)) = xy + xz.
Solution 4 Let z and w be (a, b) and (c, d), respectively.

a) z + w = (a + c, −(b + d)) = (a, −b) + (c, −d) = z + w

b) zw = (ac−bd, −(ad +bc)) = (ac−(−b)(−d), a(−d)+(−b)c) = (a, −b)·(c, −d) =
z · w.
c) zz = (a2 − b(−b), a(−b) + ba) = (a2 + b2 , 0) = |z|2 .
p √
d) |z| = a2 + (−b)2 = a2 + b2 = |z|.
p
= (ac − bd)2 + (ad + bc)2 =
|zw| = |(a + bi)(c + di)| = |ac + bd + (ad + bc)i| p
e) p
2 2 2 2 2 2 2 2 2 2
pa (c + dp) + b (c + d ) − 2abcd + 2abcd = (a + b )(c + d ) =
(a2 + b2 ) (c2 + d 2 ) = |z||w|.
f) z + z = a + bi + (a − bi) = 2a = 2Re(z).
√ √
g) Re(z) = a ≤ a2 ≤ a2 + b2 = |z|.
h) First we will prove that x2 ≤ y2 =⇒ x ≤ y. If x, y ≥ 0. x2 ≤ y2 =⇒ x2 −
y2 ≤ 0 =⇒ (x + y)(x − y) ≤ 0 =⇒ x − y ≤ 0 =⇒ x ≤ y. So, |z + w|2 = (z +
w)(z + w) = zz + ww + zw + zw = |z|2 + |w|2 + zw + zw = |z|2 |w|2 + 2Re(zw) ≤
|z|2 + |z|2 + 2|zw| = (|z| + |w|)2 . The result follows.
 93

Solution 5 a) x · y = ∑ni=1 xi yi = ∑ni=1 xi yi = ∑ni=1 yi xi = y · x.

b) (ax) · y = ∑ni=1 (axi )yi = ∑ni=1 a(xi yi ) = a(x · y).

c) x · (ay) = ∑ni=1 xi ayi = ∑ni=1 a(xi yi ) = a(x · y).

d) (x + y) · z = ∑ni=1 (xi + yi )zi = ∑ni=1 xi zi + yi zi = ∑ni=1 xi zi + ∑ni=1 yi zi = x · z + y · z.

e) ||x||2 = ∑ni=1 |xi |2 = ∑ni=1 xi xi = x · x.

f) ||x|| = 0 =⇒ ∑ni=1 |xi |2 = 0. But |xi |2 > 0 if xi 6= 0, so xi = 0 for i = 1, 2, ...n and

hence x = 0.
Solution 6 Suppose n > m. It follows that A ∼ {1, . . . , n} and A ∼ {1, . . . , m} implies
{1, . . . , n} ∼ {1, . . . , m}, by the transitivity of the equivalence. But it is an absurd since
no set can be equivalent to its own proper subset.
Solution 7 Let’s prove by induction on n = |A| + |B|.
If n = 0, then |A| = |B| = 0 and hence the result is trivial.
Now, suppose the proposition is valid for n = k. If we add a element x to A or B,
either x belongs to A ∩ B or not, so let’s study both cases.
If x is added to A and belongs to B, it follows that

|A ∪ {x}| + |B| − |(A ∪ {x}) ∩ B| = |A| + 1 + |B| − |(A ∩ B) ∪ ({x} ∩ B)|

= |A| + 1 + |B| − (|A ∩ B| + 1)
= |A ∪ B|
= |A ∪ (B ∪ {x})|
= |(A ∪ {x}) ∪ B|.

If x do not belong to B,

|A ∪ {x}| + |B| − |(A ∪ {x}) ∩ B| = |A| + 1 + |B| + |(A ∩ B) ∪ ({x} ∩ B)|

= |A| + 1 + |B| + |(A ∩ B)|
= |A ∪ B| + 1
= |(A ∪ {x}) ∪ B|.

Hence, the proposition holds for any non-negative integer n.

Now, in possession of this result, for three sets:

|A ∪ B ∪C| = |(A ∪ B) ∪C|

= |A ∪ B| + |C| − |(A ∪ B) ∩C|
= |A ∪ B| + |C| − |(A ∩C) ∪ (B ∩C)|
= |A ∪ B| + |C| − (|A ∩C| + |B ∩C| − |(A ∩C) ∩ (B ∩C)|)
= |A| + |B| + |C| − (|A ∩ B| + |A ∩C| + |B ∩C|) + |A ∩ B ∩C|.
94 Chapter J. Solutions

Solution 8 Firstly, I’ll prove the result of Exercise 1.9 for finite sets. That is, if there is
an injection from A to B and an injection from B to A then there is a bijection from A to
B.
Let f : A → B and g : B → A be the aforementioned functions. Since f is an injection,
|A| ≤ |B|. Since g is an injection, |A| ≥ |B|. Hence |A| = |B| and there is an bijection
from A to B. (Note that this proof does not involves the fact that | · | is an order relation.
It just involves the fact that |A| is an integer if A is a finite set.)
To prove our exercise we just need to know that |A| > |B| means that there is an
injection from B to A but there is not a bijection. If there was an injection from A to B,
the result we just proved would imply in such bijection, which is absurd!

Solution 9 First lets see that, if the partition described exists, then there is a bijection
from A to B. In fact, since f (A1 ) = B1 and g(B2 ) = A2 , the restrictions of f and g to A1
and B2 respectively are bijections.

A B
f
A1 B1

g
A2 B2

Then the function

(
f (x) if x ∈ A1
h(x) =
g−1 (x) if x ∈ A2

is a bijection. (With a little of notational abuse, since the functions f and g above are the
restrictions of f and g.) We now show that there exist such partition.
Let ρ : P(A) → P(A) be such that ρ(X) = A \ g(B \ f (X)). Note that, if ρ has
a fixed point, that is, a set F ⊂ A such that ρ(F) = F, then we could take A1 = F,
A2 = A \ F, B1 = f (F) and B2 = B \ f (F). It is the existence of such fixed point that I
will prove.
Consider the set E = {X ⊂ A | X ⊂ ρ(X)}. Since ∅ ∈ E, E is not empty. Let

[
F= X.
X∈E

I affirm that F is a fixed point of ρ.

 95

Note that
 [ 
ρ(F) = A \ g B \ f X
 [ 
= A \ g B \ f (X)
\ 
= A\g (B \ f (X)) (De Morgan)
\
= A\ g(B \ f (X)) (g is injective)
[
= A \ g(B \ f (X)) (De Morgan)
[
= ρ(X).

Then, F = X ⊂ ρ(X) = ρ(F). We only have to prove now that ρ(F) ⊂ F.

S S

Indeed, if x ∈ ρ(F) = A \ g(B \ f (F)), then x ∈

/ g(B \ f (F)).
Since F ⊂ A \ g(B \ f (F)), we have that g(B \ f (F)) ⊂ A \ F and hence g(B \ f (F)) ⊂
A \ (F ∪ {x}).
As f (F) ⊂ f (F ∪ {x}), it follows that

g(B \ f (F ∪ {x})) ⊂ g(B \ f (F)) ⊂ A \ (F ∪ {x})

and then F ∪ {x} ⊂ ρ(F ∪ {x}). That is, F ∪ {x} ∈ E.

We conclude that F ∪ {x} ⊂ F and hence x ∈ F. The result follows.
(I told you it was hard.)
Solution 10 — From [13]. From Exercise 1.2d we know that if C is an ordered field,
then x2 > 0 for all x 6= 0. However note that i2 = (0, 1) · (0, 1) = (−1, 0) = −1. Taking
x = 1 in the inequality x2 > 0 we see that 1 > 0. From Exercise 1.2a if follows that −1
is negative from which it follows that C cannot be an ordered field.
Solution 11 We say that the characteristic of a field is the least positive integer n such
that
1 + 1 + 1 + · · · + 1 = 0.
| {z }
n ones

If such a number does not exist, we say that the characteristic of the field is 0.
I affirm that every ordered field has characteristic 0.
In fact, if the field had characteristic n, then we would have that

0 < 1 < 1 + 1 < · · · < 1 + 1 + 1 + · · · + 1 = 0,

| {z }
n ones

which is absurd! This automatically implies that every field contains a subfield isomor-
phic to N.
Since fields have additive inverses, if x is in the field, so is −x. This means that every
field contains a subfield isomorphic to Z. The existence of inverses and the closure by
multiplication then implies the result.
96 Chapter J. Solutions

Solution 12 We’ll prove some results about Z/pZ.

First result: ∼ is an equivalence relation.
(Reflexitivity) Since every integer divides 0, x ∼ x for all x ∈ Z.
(Simmetry) If x ∼ y, then x − y = np for some integer n. That is, y − x = (−n)p and
hence y ∼ x.
(Transitivity) If x ∼ y and y ∼ z, then x − y = np and y − z = mp. Summing these
equations we get that x − z = (n + m)p and hence x ∼ z.
Second result: Z/pZ has exactly p elements.
I affirm that if x = y + kp for some integer k, then [x] = [y]. In fact, if z ∈ [x], then
z − x = np for some integer n and hence z − y = (k + n)p. That is, [x] ⊂ [y]. The other
inclusion is analogous.
Third result: If p does not divide neither x nor y, then p does not divide xy.
The converse of this result "If p divides xy then either p divides x or p divides y is a
basic fact in number theory. So, I’ll not prove it. In our context it says that if [x] and [y]
are both different from [0], then [x] · [y] 6= [0].
The only relevant part of the proof that Z/pZ is a field is the existence of multiplica-
tive inverses. If [a] ∈ Z/pZ is not equal to [0], then the set

{[0] · [a], [1] · [a], [2] · [a], . . . .[p − 1] · [a]}

has exactly p distinct elements and hence it is equal to Z/pZ. Since [1] ∈ Z/pZ, there
is an integer b ∈ {0, 1, . . . , p − 1} such that [b] · [a] = [1].
Solution 13 Let n be a positive integer. The fundamental theorem of arithmetic implies
that every positive integer is the product of an power of 2 and an odd integer. That is,

n = 2un −1 (2vn − 1),

for some positive integers un and vn .

The function f : N → N2 such that

f (n) = (un , vn )

is clearly onto. If we had that f (n) = f (m), then it would follow un = um and vn = vm ,
hence n = m.
Solution 14 — From [14]. Let Ax be the set {p/q ∈ F | p/q < x}.
It is easy to check that

f (m + n) = f (m) ⊕ f (n), f (mn) = f (m) ⊗ f (n),

for all integers m and n. Note that the definition of f for rational numbers makes sense
because if m/n = k/l, then ml = nk, so m · l = k · n. Hence, m/n = k/l.
It is also easy to check that

f (r1 + r2 ) = f (r1 ) ⊕ f (r2 ), f (r1 r2 ) = f (r1 ) ⊗ f (r2 ),

 97

for all rational numbers r1 and r2 , and that f (r1 ) ≺ f (r2 ) if r1 < r2 .
The set Ax is certainly not empty, and it is also bounded above, for if r0 is a rational
number with r0 > x, then f (r0 ) > f (r) for all f (r) ∈ Ax . Since F has the least-upper-
bound property, the set Ax has a least upper bound so that sup Ax is well defined. We now
shall show that the definition of f for irrational x is actually a general definition. In other
words, if x is a rational number, we want to show that sup Ax = f (x), where f (x) here
denotes m/n, for x = m/n. This is not automatic, but depends on the least-upper-bound
property of F; a slight digression is thus required.
Since F has the least-upper-bound property, F is archimedean. The consequences
of this fact for R have exact analogues in F: in particular, if a and b are elements of F
with a ≺ b, then there is a rational number r such that a ≺ f (r) ≺ b. Having made this
observation, we return to the proof that the two definitions of f (x) agree for rational x.
If y is a rational number with y < x, then we have already seen that f (y) ≺ f (x). Thus
every element of Ax is ≺ f (x). Consequently,

sup Ax  f (x).

On the other hand, suppose that we had sup Ax ≺ f (x). Then there would be a rational
number r such that
sup Ax ≺ f (r) ≺ f (x).
But the condition f (r) ≺ f (x) means that r < x, which means that f (r) is in the set
Ax ; this clearly contradicts the condition sup Ax ≺ f (r). This shows that the original
assumption is false, so sup Ax = f (x). It follows that f (x) = sup Ax holds for all x ∈ R.
We’ll now prove that f is an isomorphism of fields.

1. If x < y, then f (x) ≺ f (y).

If x and y are real numbers with x < y, then clearly Ax is contained in Ay . Thus

f (x) = sup Ax  sup Ay = f (y).

To rule out the possibility of equality, notice that there are rational numbers r and
s with x < r < s < y. We know that f (r) ≺ f (s). It follows that

f (x)  f (r) < f (s)  f (y).

This proves this item.

2. f is injective.
If x 6= y, then either x < y or y < x; in the first case f (x) ≺ f (y), and in the second
case f (y) ≺ f (x); in either case f (x) 6= f (y).

3. f is onto.
Let a be an element of F, and let B be the set of all rational numbers r with
f (r) ≺ a. The set B is not empty, and it is also bounded above, because there is
a rational number s with a ≺ f (s), so that f (r) ≺ f (s) for r in B, which implies
98 Chapter J. Solutions

that r < s. Let x be the least upper bound of B; we claim that f (x) = a. In order to
prove this it suffices to eliminate the alternatives f (x) ≺ a and a ≺ f (x).
In the first case there would be a rational number r with f (x) ≺ f (r) ≺ a. But this
means that x < r and that r is in B, which contradicts the fact that x = sup B. In
the second case there would be a rational number r with a ≺ f (r) ≺ f (x). This
implies that r < x. Since x = sup B, this means that r < s for some s in B. Hence
f (r) ≺ f (s) ≺ a, again a contradiction. Thus f (x) = a.
4. f (x + y) = f (x) ⊕ f (y).
Suppose that f (x + y) 6= f (x) ⊕ f (y). Then either f (x + y) ≺ f (x) ⊕ f (y) or
f (x) ⊕ f (y) ≺ f (x + y). In the first case there would be a rational number r such
that f (x + y) ≺ f (r) ≺ f (x) ⊕ f (y). But this would mean that x + y < r. Therefore
r could be written as the sum of two rational numbers r = r1 + r2 , where x < r1
and y < r2 . Then, using the facts checked about f for rational numbers, it would
follow that f (r) = f (r1 + r2 ) = f (r1 ) ⊕ f (r2 )  f (x) + f (y), a contradiction. The
other case is handled similarly.
5. f (xy) = f (x) ⊗ f (y).
The same reasoning of the item 4 proves this for positive real numbers. The
general case is then a simple consequence.

(Solution from [14])

Solution 15 The Dirichlet’s box principle implies that there exists two distinct integers
k and j such that {kα} and { jα} lie in the same element of the partition. Without loss
of generality, suppose that k > j. It follows that
   
1 n−1
{(k − j)α} ∈ 0, or {(k − j)α} ∈ ,1 ,
n n
since (
{kα} − { jα} if {kα} ≥ { jα}
{(k − j)α} = .
1 + {kα} − { jα} if {kα} < { jα}
Then we have that there is an element of the set

{{m(k − j)α} | m ∈ N}

in each of the n elements of the partition. The result follows since for every subset
(a, b) ⊂ [0, 1) there are integers m, n such that
 
m m+1
, ⊂ (a, b).
n n
Solution 16 The result is equivalent to

1
|qα − p| < .
n0
 99

Taking p = bqαc, we just need to find some q ∈ N such that

1
{qα} < .
n0

But this follows readily from the previous exercise.

Solution 17 ||x + y||2 = ||x||2 + x · y + y · x + ||y||2 = ||x||2 + 2 Re(x · y) + ||y||2 =
||x||2 + ||y||2 .
Notice that ||x + y||2 = ||x||2 + ||y||2 does not imply x · y = 0, since Re(x · y) can be 0
while x · y 6= 0. Take x = (1, 1 + i) and y = (1, 1 − 2i), for instance.
Solution 18 — From [13]. ||x + y||2 + ||x − y||2 = (||x||2 + 2 Re(x · y) + ||y||2 )+
(||x||2 − 2 Re(x · y) + ||y||2 ) = 2||x||2 + 2||y||2 .
You may interpret ||x||, ||y|| as the lengths of the edges and ||x + y||, ||x − y|| as the
lengths of the diagonals.

x−
y y
x+
x

Solution 19 — Adapted from [2]. Let x1 be a element of X. Having chosen x1 , . . . , xn−1 ,

define xn to be any element of X that was not already chosen. Then the set {x1 , x2 , x3 , . . . }
is a countable subset of X.
However, the axiom of choice basically means that given a set we can always choose
a arbitrary element of it. Even if this seems to be obvious, it is not possible to prove it
from the usual axioms of set theory.
Solution 20 Let E = { f1 , f2 , f3 , · · · } be a countable subset of B (of course B is infinite
since we can permute the elements of N in an infinitely many ways). Let P be the set of
all prime numbers (P could be any infinite subset of N such that N \ P is infinite too).
We’ll construct a bijection g : N → N in the following way:

• g(2) is the smallest element of P which is not an element of { f1 (2)}.

• After defining the values of g(2), g(4), · · · , g(2k − 2), define g(2k) to be the small-
est element of P which is not an element of {g(2), g(4), · · · , g(2k − 2), fk (2k)}.

List now all the elements of A = N \ P = {a1 , a2 , · · · } and define g(2k − 1) as ak for all
k ∈ N.
We have then that g is an element of B which is not an element of E, since g(2k) 6= fk (2k)
for all k ∈ N. We conclude that every countable subset of B is proper. It follows that B is
uncountable. (Otherwise B would be a proper subset of B.)
100 Chapter J. Solutions

Solution 21 If a ∈ A, then a ∈ f (a) which is a contradiction. But if a ∈

/ A, then a ∈
/ f (a),
which is a contradiction too. We conclude that no such function exists.
Solution 22 There are two problems in this proof. Firstly, real numbers have more than
one decimal representation. For example,
1
= 0, 5 = 0, 49999 . . . .
2
Another flaw lies in the fact that is possible that the process described generates the
number 1 = 0, 9999 . . . , which is not an element of (0, 1).
Solution 23 Suppose B is uncountable. As Q is dense in R, each element x of B contains
a distinct rational number q(x). The function q : B → q(B) ⊂ Q is bijective, meaning
that q(B) is an uncountable subset of Q. Since Q is countable, this is absurd! The result
follows.
To write the bijection from the set of all (open) intervals to R2 just map the interval
(a, b) to the element (a, b) of R2 . (Horrible notation issues...)

2- Elements of Topology
Solution 1
Solution 2
Solution 3
Solution 4
Solution 5
Solution 6
Solution 7
Solution 8
Solution 9
Solution 10
Solution 11
Solution 12
Solution 13
Solution 14
Solution 15
Solution 16
Solution 17
Solution 18
 101

Solution 19
Solution 20
Solution 21 Let f (x) = d(x, φ (x)). Note that, by the triangular inequality

d(x, φ (x)) ≤ d(x, y) + d(y, φ (y)) + d(φ (y), φ (x)),

for all x, y ∈ X. That is,

d(x, φ (x)) − d(y, φ (y)) ≤ d(x, y) + d(φ (y), φ (x)).

Reversing the roles of x and y we get

|d(x, φ (x)) − d(y, φ (y))| ≤ d(x, y) + d(φ (x), φ (y)) < 2δ

whenever d(x, y) < δ . This means that f is continuous.

Let α = infx∈X f (x). By the extreme value theorem, we know that there is some
x0 ∈ X such that f (x0 ) = α. If α > 0, then

f (φ (x0 )) < f (x0 ) = α,

which is absurd since α is the least value f can take. We conclude that α = 0 and
φ (x0 ) = x0 .
Suppose now that there is some other fixed point x00 . Then d(φ (x0 ), φ (x00 )) =
d(x0 , x00 ), which implies x0 = x00 .
Solution 22 — From [13].
Solution 23
Solution 24 Let x1 ∈ X \ E. Since E is a closed set, the distance from x1 to the elements
of E possesses a positive minimum d. Since d/r > d, there exists p ∈ E for which
kx1 − pk ≤ d/r. Define now x = (x1 − p)/ kx1 − pk. Then,
kx1 − (p + y kx1 − pk)k r
kx − yk = ≥ d = r.
kx1 − pk d
Since p + y kx1 − pk is an element of E, the result follows.

3- Sequences and Series

Solution 1
Solution 2
Solution 3 Let sn be the partial sums of the series. If m is even and m < n we have that

sn − sm = xm+1 − xm+2 + xm+3 − xm+4 + xm+5 − · · · + (−1)n+1 xn

= xm+1 − (xm+2 − xm+3 ) − (xm+4 − xm+5 ) − · · · + (−1)n+1 xn
≤ xm+1 ≤ xm .
102 Chapter J. Solutions

Similarly,

sm − sn = −xm+1 + xm+2 − xm+3 + xm+4 − xm+5 + · · · − (−1)n+1 xn

= −xm+1 + xm+2 − (xm+3 − xm+4 ) − xm+5 + · · · − (−1)n+1 xn
≤ −xm+2 + xm+2 ≤ xm+2 ≤ xm+1 ≤ xm .

Hence |sn − sm | < xm . Since xm → 0, the Cauchy criterion implies that (sn ) converges.
The proof for m odd is analogous.
Solution 4 — From [4].

Solution 5 — From [4].

Solution 6

Solution 7

Solution 8

Solution 9

Solution 10 — From [13].

Solution 11 — From [1].

Solution 12

Solution 13

Solution 14

Solution 15

Solution 16 — From [4].

Solution 17 — From [4].

Solution 18 — From [13].

Solution 19 — From [13].

Solution 20 — From [13].

Solution 21 — From [13].

Solution 22 — From P. Erdös in the American Mathematical Monthly.

Solution 23

Solution 24

Solution 25

Solution 26 — From [7].

Solution 27
 103

4- Functional Limits and Continuity

Solution 1
Solution 2 — From [8].
Solution 3 — From [8].
Solution 4 — From [13].
Solution 5
Solution 6
Solution 7
Solution 8
Solution 9 — From [4].
Solution 10 — From [4].
Solution 11 — From [4].
Solution 12 — From [8].
Solution 13
Solution 14
Solution 15
Solution 16
Solution 17
Solution 18
Solution 19
Solution 20
Solution 21
Bibliography

[1] S. Abbott. Understanding Analysis. Undergraduate Texts in Mathematics. Springer,

2016.
[2] J. Casasayas and M. D. C. Cascante. Problemas de Análisis Matemático de una
Variable Real. Edunsa, 1990.
[3] J. A. Clarkson. On the series of prime reciprocals. Proc. Amer. Math. Soc., 17:541,
1966.
[4] P. N. de Souza and J. N. Silva. Berkeley Problems in Mathematics. Springer, 2004.

[5] F. del Castillo. Analisis Matematico II. Alhambra Universidad, 1987.

[6] P. Deligne. Alexandre Grothendieck 1928–2014, Part 1. Notices of the AMS,
63(03):249–250, 2016.
[7] J. Dieudonné. Foundations of Modern Analysis. Academic Press, 1960.

[8] R. Gelca and T. Andreescu. Putnam and Beyond. Springer, 2007.

[9] K. Knopp. Theory and Application of Infinite Series. Dover Publications, 1990.
[10] E. L. Lima. Curso de Análise: Volume 1. IMPA, 2014.
[11] E. L. Lima. Espaços Métricos. Projeto Euclides. IMPA, 2015.

[12] J. Munkres. Topology. Pearson, 2014.

[13] W. Rudin. Principles of Mathematical Analysis. International Series in Pure &
Applied Mathematics. McGraw-Hill Education, 1976.
[14] M. Spivak. Calculus. Publish or Perish, 2008.

105