Metric Spaces

An Introduction to Metric Spaces
Department of Mathematics, Linköping University
Samuel E. Andersson, David Wiman
LiTH-MAT-EX–2022/20–SE
Credits: 16 hp
Level: G2
Supervisor: Anders Björn,
Examiner: Jana Björn,
Linköping: September 2022
Abstract
In this thesis we start off by ensuring that the reader is up to speed when it
comes to some well known definitions and theorems from real analysis. We
then introduce the reader to metric spaces and provide them with some exam-
ples such as the real numbers with the Euclidean distance, and compact sets
with the Hausdorff distance. Then, we go on to define important concepts such
as inner points, limit points, open sets, boundary and much more. We also
show, whenever we can, how these concepts are connected. With these tools
in place we move on to explain how limits and continuity are defined in metric
spaces as well as providing the reader with several examples. We then intro-
duce the reader to the concepts of compactness and uniform convergence, for
which we show some interesting results such as how uniform convergence and
the supremum norm are related. We finish off by covering curves and connect-
edness (including pathconnectedness) in metric spaces, before we briefly touch
on topological spaces as to give the reader a hint of what further mathematics
studies might hold.
Keywords:
Analysis, compact, convergence, limit, metric space, sequence, set, topol-
ogy
URL for electronic version:

http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-188913
E. Andersson, Wiman, 2022. iii

Sammanfattning
I detta examensarbete börjar vi med att försäkra oss om att läsaren har de
förkunskaper som behövs för att kunna ta del av arbetet. Detta görs genom att
påminna läsaren om viktiga definitioner och satser från reell analys. Därefter
introducerar vi läsaren till metriska rum och ger en mängd olika exempel på
dessa som läsaren förhoppningsvis redan stött på. Detta inkluderar bland an-
nat de reella talen med euklidiskt avstånd och slutna och begränsade mängder
med Hausdorff-avstånd. När vi väl förklarat distanskonceptet introducerar vi
inre punkter, hopningspunkter, öppna mängder, randpunkter och mycket mer.
Vi visar dessutom, närhelst vi kan, hur dessa koncept hänger samman. När alla
dessa grundbegrepp är etablerade kan vi fortsätta med att förklara gränsvärden
och kontinuitet i metriska rum. Vi ger även läsaren flera exempel på detta. I
arbetets andra hälft tar vi upp kompakthet och likformig konvergens, för vil-
ka vi presenterar en del intressanta resultat, såsom hur likformig konvergens
och supremumnormen är relaterade. Vi avslutar examensarbetet genom att gå
igenom kurvor och sammanhängande mängder (inklusive bågvis sammanhäng-
ande mängder) i metriska rum, innan vi kort tar upp topologiska rum för att ge
läsaren en föraning om vad vidare matematikstudier kan innehålla.
Nyckelord:
Analys, följd, gränsvärde, kompakt, konvergens, metriskt rum, mängd,
topologi
URL för elektronisk version:

http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-188913
E. Andersson, Wiman, 2022. v

Acknowledgements
We would first and foremost like to extend our thanks to Anders Björn for pro-
viding us with the material needed for this thesis as well as being our supervisor.
We wrote this thesis during the summer holiday when most people are on va-
cation, but Anders took the time to see us even during this inconvenient period
and for this we are immensely grateful. Had it not been for this non-traditional
writing period, this thesis would probably have never been written.
Secondly, we would like to thank Jana Björn for taking on the role of ex-
aminer and allotting her time at the end of the summer to read our work. We
had a hope of presenting the thesis in September and she provided us with clear
guidelines and easy to understand feedback so that we could finish on time.
Beyond these two, we of course want to thank the entire department of math-
ematics at Linköping University for providing us with a first-rate mathematics
educations and for inspiring us to pursue a mathematics degree even though we
are enrolled in an engineering program.
We would also like to thank our families and friends for helping us with
proofreading and commenting on the pedagogy of this thesis.
E. Andersson, Wiman, 2022. vii

Nomenclature
Symbol Explanation
N The set of all natural numbers, {0, 1, 2, ...}.
Z The set of all integers, {0, ±1, ±2, ...}.
Q The set of all rational numbers.
R The set of all real numbers.
Br (a) The open ball of radius r centered at a.
BrX (a) The open ball of radius r centered at a ∈ X.
Ac The complement of A, X \ A.
f (A) The image of A under f , {f (x) : x ∈ A}.
∂A The boundary of A.
⊂ Subset, not necessarily proper.
X ×Y The set of points (x, y) such that x ∈ X and y ∈ Y .
A The closure of A.
A◦ The interior of A.
E. Andersson, Wiman, 2022. ix

Preface
The study of metric spaces began in the early 20th century when Maurice
Fréchet1 published his PhD dissertation Sur quelques points du calcul fonction-
nel in 1906. In it, he introduced the idea of a space where not much else but
the notion of distance existed. However, Fréchet did not call it a metric space.
The name is due to Felix Hausdorff2 , one of the founding fathers of modern
topology.
In the beginning of the 20th century, functional analysis was still a novelty
and an axiomatic foundation for mathematics had not yet been properly estab-
lished. In the world of mathematics at that time, many people were studying
spaces of functions and since many of them did not communicate with each
other, they ended up with different notions of convergence in different spaces.
The concept of convergence is strongly linked to the concept of distance, since
we usually tie convergence to the notion of approaching certain points within a
space. To make research simpler and more unified, Fréchet came up with the
idea of axiomatizing the notion of distance so that a single definition of conver-
gence could be used in most of the spaces at the time. Proofs based on these
axioms would also hold in all of these spaces, which led to faster development
as not every space had to be treated individually.
In this thesis, we will introduce you to metric spaces and some of the theory
surrounding them. The purpose of this text is to be used as course material
in TATA34 Real analysis, honours course at Linköping University. You will
learn about how we define different distance functions, open sets, compactness
and much more. At the end of the thesis, we will also quickly introduce you
to topological spaces as to give you a hint of what future mathematics studies
might hold.
1 Maurice Fréchet (1878–1973), French mathematician.

2 Felix Hausdorff (1868–1942), German mathematician.
E. Andersson, Wiman, 2022. 1

Contents
1 Background 5
2 Introduction to Metric Spaces 7

2.1 Definition of a Metric Space . . . . . . . . . . . . . . . . . . . . . 7
2.2 Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Closed Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Complement and De Morgan’s Laws . . . . . . . . . . . . . . . . 15
2.5 Connection Between Open and Closed Sets . . . . . . . . . . . . 17
2.6 Perfect, Dense, Bounded and Boundary . . . . . . . . . . . . . . 18
2.7 Relative or Induced Topology . . . . . . . . . . . . . . . . . . . . 19
3 Limits and Sequences 21

3.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . 24
4 Compactness 29
5 Uniform Convergence 35
6 Curves and Connectedness 39

6.1 Curves and Pathconnectedness . . . . . . . . . . . . . . . . . . . 39
6.2 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7 Topological Spaces 45
7.1 Definition of a Topological Space . . . . . . . . . . . . . . . . . . 45
7.2 Unique and Non-unique Limits . . . . . . . . . . . . . . . . . . . 48
7.3 Lp -Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.4 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.5 Further studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Contents
A English–Swedish Glossary 55
B Swedish–English Glossary 57
Chapter 1
Background
Before we introduce you to metric spaces, we want to remind you of a few defini-
tions and theorems for R and Rn that are relevant in understanding this topic.
Most of these should be recognizable from prior single variable and multivari-
able analysis courses, but some results from slightly more advanced analysis will
also be presented.
In the theory of metric spaces, the convergence of sequences will play an im-
portant role and thus we remind ourselves of the definition of limits of sequences
as well as the definition of Cauchy1 sequences.
Definition 1.1. A sequence (xn ) in R converges to the limit L as n → ∞,
written
xn → L or lim xn = L,
n→∞
if for every ε > 0, there exists some N ∈ N such that |xn − L| < ε for all n ≥ N .
This can be expressed with quantifiers as
∀ε > 0 ∃N ∈ N : n ≥ N =⇒ |xn − L| < ε.
Definition 1.2. A sequence (xn ) in R is a Cauchy sequence if for every ε > 0,

there exists some N ∈ N such that |xm − xn | < ε for all m, n ≥ N . Expressed
with quantifiers,
∀ε > 0 ∃N ∈ N : m, n ≥ N =⇒ |xm − xn | < ε.
Although this thesis is mainly concerned with sequences in other spaces

than R, it is still of great interest how these sequences specifically behave in
1 Louis Cauchy (1789–1857), French mathematician.

6 Chapter 1. Background
R. We continue with two theorems regarding convergence of sequences and

subsequences. These theorems are not typically presented in a first course in
single variable analysis, but they can be found on p. 67 and p. 64 respectively
in [1].
Theorem 1.3. (Cauchy’s convergence principle)
The sequence (an ) is convergent in R if and only if (an ) is a Cauchy sequence
in R.
Theorem 1.4. (The Bolzano2 –Weierstrass3 theorem)
If (an ) is bounded, then (an ) has a convergent subsequence.
Next we remind ourselves of the definitions of convergence and continuity of
real functions.
Definition 1.5. A function f : R → R has the limit L as x → a, written
f (x) → L or lim f (x) = L,

x→a
if for every ε > 0, there exists some δ > 0 such that |f (x) − L| < ε whenever
0 < |x − a| < δ. Expressed with quantifiers,
∀ε > 0 ∃δ > 0 : 0 < |x − a| < δ =⇒ |f (x) − L| < ε.
Definition 1.6. A function f : R → R is continuous at some point a ∈ R if
lim f (x) = f (a).

x→a
Remark 1.7. All the facts mentioned above also hold in Rn , and Definitions 1.5
and 1.6 also hold for f : Rn → Rm , where n, m ≥ 1 if | · | is taken to be the
Euclidean4 norm (see Example 2.2).
In your multivariable analysis course, you hopefully came across the notion
of compactness of sets. We remind ourselves of the definition in Rn .
Definition 1.8. A set K ⊂ Rn is compact if it is closed and bounded.

You are now brought up to speed and we are ready to introduce you to metric
spaces. We will look at limits and continuity in metric spaces in Chapter 3, while
compactness is studied in Chapter 4.
2 Bernard Bolzano (1781–1848), Czech mathematician.

3 Karl Weierstrass (1815–1897), German mathematician.
4 Euclid (∼ 325–265 BC), Greek mathematician.
Chapter 2
Introduction to Metric Spaces
2.1 Definition of a Metric Space

Simply put, we want to establish a type of space with an inherent concept of
distance between points within the space. The definition should be applicable to
a wide variety of spaces, going beyond the standard geometric idea of distance.
But there are some key properties we want to adhere to in order for this distance
to make sense.
First, we would like two points to have zero distance between them only if
they are one and the same. Second, the distance should be symmetric, meaning
that the distance between two points should be the same regardless of which
one you start from. Third, we want the triangle inequality to be true, i.e. the
shortest distance between two points is the direct one without any “detours”.
Definition 2.1. A metric space X = (X, d) is a set X together with a distance

function, or metric, d : X × X → [0, ∞) such that for all x, y, z ∈ X,
(i) d(x, y) = 0 if and only if x = y,
(ii) d(x, y) = d(y, x),
(iii) d(x, y) ≤ d(x, z) + d(z, y).
Notice that by setting y = x in constraint (iii) we get 0 ≤ d(x, y) since

d(x, x) = 0 and d(x, y) = d(y, x), so the three constraints would guarantee that
the metric is a non-negative function even if we only required d : X × X → R.
This is of course very important in conveying the notion of distance.

8 Chapter 2. Introduction to Metric Spaces
From the small number of constraints of the definition, we can tell that metric
spaces are quite general spaces. In fact, most spaces you have encountered are
metric spaces.
Example 2.2. Below, we give some examples of metric spaces you have en-
countered before but perhaps not recognized as metric spaces.
(
x − y, if x ≥ y,
• X = (R, d) with d(x, y) = |x − y| =
y − x, if x < y.
• X = (Rn , d) with the Euclidean distance

v
u n
uX
d(x, y) = t (xk − yk )2 ,
k=1
where x = (x1 , ... , xn ) and y = (y1 , ... , yn ).

• X = (E, d), where E ⊂ Rn , with d(x, y) being the same as for Rn .
Example 2.3. There are many other metric spaces that you might not have
encountered yet which are important in other fields, both within and outside
pure mathematics. The following are some examples of such metric spaces.
• The set of all binary numbers of length n with the Hamming1 distance
n
X
d(x, y) = |xk − yk |,
k=1
where x = (x1 ... xn ) and y = (y1 ... yn ), with xk , yk ∈ {0, 1}. This is intu-
itively understood as the smallest number of bit-flips required to change
x into y. This distance function is also referred to as the ℓ1 -distance, see
Example 2.4.
• The set of all compact K ⊂ Rn (see Definition 1.8) with the Hausdorff
distance d(K, L) = inf{ε > 0 : K ⊂ Lε and L ⊂ Kε }, where
[
Aε = {y ∈ Rn : |x − y| < ε}.
x∈A
Intuitively, the Hausdorff distance between K and L is the largest possible

distance between a point in K and the set L, or vice versa.
1 Richard Hamming (1915–1998), American mathematician.
2.1. Definition of a Metric Space 9
• The complex plane C with chordal distance on the Riemann2 sphere,

which you will encounter in a course in complex analysis if you have not
already. The chordal distance between two points z1 , z2 ∈ C is given by
2|z1 − z2 |
d(z1 , z2 ) = p .
(1 + |z1 |2 )(1 + |z2 |2 )
z1 z2
Chordal distance on the Riemann sphere.
• The set C([a, b]) of all real-valued continuous functions on the interval
[a, b], with the L1 -norm,
Z b
d(f, g) = |f (x) − g(x)| dx.
a
• C([a, b]) with the supremum norm,
d(f, g) = ∥f − g∥∞ = sup |f (x) − g(x)|.

a≤x≤b
If you have taken a course in Fourier3 analysis you may recognize the L1 -
norm and the supremum norm as these are norms on some of the most commonly
studied spaces, L1 ([a, b]) and L∞ ([a, b]) respectively. We will discuss these fur-
ther in Chapter 7.
Of course there is not only one distance function per set that qualifies it as
a metric space. In fact, for all sets with at least two points, there is an infinite
number of possible distance functions.
2 Bernhard Riemann (1826–1866), German mathematician.

3 Joseph Fourier (1768–1830), French mathematician.
Example 2.4. The following are some examples of different distance functions
for Rn . Here x = (x1 , ... , xn ) and y = (y1 , ... , yn ).
P 1/p
n
• ℓp -distance: d(x, y) = k=1 |xk − yk |
p
for any 1 ≤ p < ∞.
• ℓ∞ -distance: d(x, y) = maxk |xk − yk |.

(
0, if x = y,
• Discrete distance: d(x, y) =
1, if x ̸= y.
Remark 2.5. The ℓ1 -distance function is often called the taxicab distance. It is
intuitive to see why when regarding the case of R2 if you are familiar with how
taxicabs have to drive in New York City for example. Note that the shortest
path is not unique.
The ℓ2 -distance is the Euclidean distance in Example 2.2. It is also the only
ℓ -distance which has a corresponding
p
p inner product, being the standard dot
product. Meaning that dℓ2 (0, x) = ⟨x|x⟩.
The discrete distance is the simplest distance function, and can be defined
on any set X.
4 4
3 3
2 2
1 1
1 2 3 4 1 2 3 4
2
√
Taxicab distance, 6 units. ℓ -distance, 18 ≈ 4.24 units.
2.2 Open Sets

Now that we have a concept of distance within spaces, we can introduce the
concept of neighborhoods to points. We define the set of all points that are less
than some distance r > 0 away from a given point, which can be referred to as
an r-neighborhood. From now on, all sets X and Y will be metric spaces unless
otherwise stated.
2.2. Open Sets 11
Definition 2.6. An open ball of radius r centered at the point a is defined by

Br (a) = {x ∈ X : d(x, a) < r}.
When working with different metric spaces we sometimes write BrX (a).
Notice that for the metric space X = [0, ∞) with d(x, y) = |x − y|,
B1 (1) = (0, 2), but B1 (0) = [0, 1).
In fact
1

B3/4 4 = [0, 1) = B1 (0),
so the center and radius of a ball are not unique. This may seem strange at first
glance, but is in fact common in metric spaces.
Furthermore, we would like a way to categorize the points in a set based
upon their surroundings. One can imagine a type of point in a set which is
completely surrounded by other points within the set, i.e. has a neighborhood
which is also contained by the set.
Definition 2.7. A point a is an inner point, or interior point, in G ⊂ X if there
exists an ε > 0 such that Bε (a) ⊂ G.
Based on the characteristics of the points within a set, we can now begin to
categorize sets. We introduce a category of sets which is exclusively made up of
inner points.
Definition 2.8. G ⊂ X is open if every point a ∈ G is an inner point in G.
Note that open balls are indeed open sets, hence the name.
Example 2.9. If X = R, then R, ∅ and open intervals of the form (a, b) are
open, just to name a few.
Now that we have defined open sets and seen some examples, we devise
methods to determine whether or not sets are open, depending on how they are
constructed.
S
Theorem 2.10. If Gλ is open for every λ ∈ Λ, then λ∈Λ Gλ is open.
Notice that the theorem does not require any limitation on the index set Λ.
This means that in addition to being valid for finite Λ, it is valid for infinite
index sets, even uncountable ones.
Proof. If x ∈ λ∈Λ Gλ then there is some λ0 ∈ Λ for which x ∈ Gλ0 . As Gλ0 is
S
open there is some ε > 0 such that
[
Bε (x) ⊂ Gλ0 ⊂ Gλ .
λ∈Λ
Now that we have seen that we can construct open sets by taking a union
of open sets, it is natural to consider intersections of open sets.
TN
Theorem 2.11. If G1 , ... , GN are open, then j=1 Gj is open.
TN
Proof. If x ∈ j=1 Gj , then x ∈ Gj for every j = 1, ... , N . Thus there are
ε1 , ... , εN > 0 such that Bεj (x) ⊂ Gj . If we let ε = min{ε1 , ... , εN } > 0 then
TN
Bε (x) ⊂ j=1 Gj . Therefore the intersection is open.
Example 2.12. The set

∞
\ 1 1
A= − , = {0}
n=1
n n
is an intersection of open sets but is in fact not open. Note that this is because
Theorem 2.11 is only valid for finite intersections.
Since not all sets are open and only consist of inner points, we might want
to be able to single out and refer to the inner points of a set in a handy way.
Definition 2.13. The interior of A is the set of all inner points of A. It is
denoted A◦ .
Theorem 2.14. Let A ⊂ X where X is a metric space. Then
(a) A◦ is open.
(b) A◦ is the largest open subset contained in A.
Proof. (a) If x ∈ A◦ then there exists an rx > 0 such that Brx (x) ⊂ A. Since
Brx (x) is open, it only contains inner points of A, thus Brx (x) ⊂ A◦ . Hence A◦
is open per Definition 2.8.
(b) Fix a set A and let G ⊂ A be an open set. Let a ∈ G be arbitrary. Since
G is open, a is an inner point and there exists some ε > 0 such that Bε (a) ⊂ G.
But since G ⊂ A, then Bε (a) ⊂ A and thus a ∈ A◦ , showing that G ⊂ A◦ .
Since A◦ is always open, A◦ is the largest open subset contained in A.
2.3 Closed Sets

A different way of categorizing points in a set based on their surroundings is to
look at if there are other points belonging to the same set in every neighborhood
of the point. We are also interested in categorizing points that have no other
points from the same set that are arbitrarily close.
2.3. Closed Sets 13
Definition 2.15. A point x ∈ X is a limit point, or accumulation point, of A if

Bε (x) ∩ (A \ {x}) ̸= ∅ for every ε > 0. If x ∈ A is not a limit point, it is called
isolated in A.
Intuitively, limit points of a set are points which are arbitrarily close to other
points of the set.
The definition of a limit point is quite similar to that of an inner point, and
the statement that a certain point is a limit point is weaker than it being an
inner point, as indeed all inner points that are not isolated are limit points.
Much like we derived open sets from the definition of inner points, we can
categorize sets based upon their limit points. We introduce a category of sets
which contain all of their limit points.
Definition 2.16. F ⊂ X is closed if it contains all of its limit points.
Just like for Theorems 2.10 and 2.11 for open sets, we will show that in-
tersections and finite unions of closed sets are indeed also closed. The proof
however requires theory presented later in this thesis, so we will be coming back
to this theorem later to prove it in Section 2.5.
Theorem 2.17.
(a) If Fλ is closed for all λ ∈ Λ, then λ∈Λ Fλ is closed.
T
SN
(b) If F1 , ... , FN are closed, then j=1 Fj is closed.
Remark 2.18. We commonly use F for closed sets and G for open ones. This is
a tradition that most likely came from French where the word for closed, fermé,
begins with an f. G was simply the next letter in the alphabet. This thesis will
use this convention where applicable. When we do not know if a set is open or
closed, or it is not important for the statement, we will use some other letter.
Example 2.19. Let A = [0, 1] ∪ {2}. Its limit points are all the points in [0, 1]
and its isolated point is 2. The set contains all of its limit points and is therefore
closed.
Example 2.20. Let A = [0, 1). Its limit points are all the points in [0, 1] and it
has no isolated points. Since the limit point 1 is not in the set, it is not closed.
Example 2.21. Let A = { n1 : n = 1, 2, ...}. Then A consists only of isolated
points. The only limit point to A is 0, which is not in the set, and therefore A
is not closed. However, A is not open either since none of its points are inner
points. This is fairly common and most sets do not fall into any of these two
categories. Sets that fall into both categories are sometimes called clopen, more
on them later on.
In Examples 2.19–2.21 we (implicitly) assumed them to be within the metric

space R with the standard metric. Both openness and closedness depend on
which metric space they are considered in. For instance A = [0, 1) is clopen
within itself.
Whether or not a set does not contain all of its limit points, we can add
them to the set to create a new set.
Definition 2.22. A = A ∪ L, where L is the set of all limit points to A, is called

the closure of A.
Note that A ⊂ A. This is clear by the definition. From this point, it is not
all that hard to realize that
A⊂B =⇒ A ⊂ B. (2.1)
The closure of a set A is always closed, and is the smallest closed set con-
taining A. It is also easy to realize that A is closed if and only if it is equal to
its closure, and thus the closure of an arbitrary closure A is equal to itself.
Theorem 2.23. Let A ⊂ X where X is a metric space. Then
(a) A is closed if and only if A = A.
(b) A = A.
(c) A is always closed.
(d) A is the smallest closed set containing A.
Proof. (a) If A = A then A must contain all of its limit points, per definition,
and is therefore closed. Conversely, if we know that A is closed and therefore
contains all of its limit points, it must necessarily be equal to A.
(b) By Definition 2.22, we conclude that A ⊂ A since
A = A ∪ {the limit points of A} = A ∪ LA .
We now need to show that LA ⊂ A. To this end, let x ∈ LA and xj ∈ A be

such that xj → x. Then there are also yj ∈ A ∩ B1/j (xj ). It follows that also
yj → x, and thus x ∈ LA . Hence LA ⊂ LA ⊂ A and therefore
A ⊂ A,
i.e. A = A.
2.4. Complement and De Morgan’s Laws 15
(c) A is closed if and only if A = A by (a) and since A = A is always true

by (b), A is always closed.
(d) Take any closed set F such that A ⊂ F . Then A ⊂ F = F by (2.1)
and (a). Since the choice of F was arbitrary, all closed sets that contain A will
also contain A. As A is closed by (c), A is the smallest closed set containing
A.
2.4 Complement and De Morgan’s Laws

Now that we have defined both open and closed sets, we are interested in their
relationship. However, first we need to introduce some new tools.
We leave the world of analysis for a brief detour to the realm of discrete
mathematics. We begin by defining the set of all points in a metric space which
are not contained within a certain set.
Definition 2.24. The complement of a set A ⊂ X is defined by
Ac = X \ A = {b ∈ X : b ∈
/ A}.
Notice how applying the complement twice on a set will result in the original
set, i.e.
(Ac )c = A. (2.2)
Note also that we can express the interior in terms of the closure of the comple-
ment by
A◦ = X \ Ac .
To be able to prove Theorem 2.17 we need to introduce you to De Morgan’s4
laws. You may have encountered them in a previous discrete mathematics course
or perhaps in a course on logic gate networks.
Theorem 2.25. (De Morgan’s laws)

\ c [
(a) Aλ = Acλ .
λ∈Λ λ∈Λ
[ c \
(b) Aλ = Acλ .
λ∈Λ λ∈Λ
As before Λ is an index set that may be finite, countable or uncountable.

4 Augustus De Morgan (1806–1871), British mathematician.
Proof. (a) Let

\ c
x∈ Aλ .
λ∈Λ
This means that x is not contained in every set Aλ so there must be at least
one λ0 ∈ Λ for which x is in Acλ0 . Therefore,
[
x ∈ Acλ0 ⊂ Acλ .
λ∈Λ
This shows that \ c [

Aλ ⊂ Acλ .
λ∈Λ λ∈Λ
Conversely, let
[
y∈ Acλ .
λ∈Λ
This means that there is some λ1 such that y ∈ Acλ1 . Therefore

\
y∈
/ Aλ1 ⊃ Aλ .
λ∈Λ
So \ c
y∈ Aλ ,
λ∈Λ
which in turns shows that

[ \ c
Acλ ⊂ Aλ .
λ∈Λ λ∈Λ
These two conclusions together show that

\ c [
Aλ = Acλ .
λ∈Λ λ∈Λ
(b) Using (2.2), and then (a) applied to Aλ , gives
\ \ c c [ c c [ c
Acλ = Acλ = Acλ = Aλ .
λ∈Λ λ∈Λ λ∈Λ λ∈Λ
2.5. Connection Between Open and Closed Sets 17
2.5 Connection Between Open and Closed Sets

We now have the tools necessary to examine the relationship between open and
closed sets.
Theorem 2.26. Let F and G be subsets of the metric space X. Then
(a) F is closed if and only if F c is open,
(b) G is open if and only if Gc is closed.
Remark 2.27. This gives an alternative way of defining closed sets assuming
we have defined open sets. It is in fact very common to define closed sets in this
manner.
Proof. (a) First we prove that F being closed implies that F c is open: Let
x ∈ F c . Since F is closed, x is not a limit point to F . This in turn means there
exists an ε > 0 such that Bε (x) ∩ F = ∅. Therefore Bε (x) ⊂ F c , and hence F c
is open.
Next we prove the converse; that F c being open implies that F is closed:
Let x ∈ F c . Since F c is open, there exists Bε (x) ⊂ F c for some ε > 0. This
means that x is not a limit point to F which in turn means that all limit points
belong to F , i.e. F is closed.
(b) We prove that G is open if and only if Gc is closed by utilizing (a): We
know that Gc is closed if and only if (Gc )c is open. But we also know (Gc )c = G
by (2.2). Thus G is open if and only Gc is closed.
We have now developed all the tools needed to prove Theorem 2.17.
Proof of Theorem 2.17. (a) By use of De Morgan’s laws, we can write

\ c [
Fλ = Fλc .
λ∈Λ λ∈Λ
By Theorem 2.26 (a) each

T Fλ is open, and a union of open sets is open by The-
c
orem 2.10. Therefore ( Tλ∈Λ Fλ )c is also open. By applying Theorem 2.26 (a)
again we conclude that λ∈Λ Fλ is closed.
(b) Using De Morgan’s laws again, we can write
N
[ c N
\
Fj = Fjc .
j=1 j=1
SN
A finite intersection of open sets is open by Theorem 2.11 and therefore ( j=1 Fj )c
SN
is also open. By using Theorem 2.26 (a) we conclude that j=1 Fj is closed.
2.6 Perfect, Dense, Bounded and Boundary

Now that we have laid some ground work for open and closed sets as well as
looked at their relation, we shall dive a bit deeper and look at some finer details.
We are particularly interested in closed sets that lack isolated points. Such
sets contain all of their limit points, and only limit points. We will see that this
distinction from arbitrary closed sets is important for certain statements.
Definition 2.28. F is perfect if it is closed and has no isolated points.
As previously, you have encountered perfect sets before but perhaps not
recognized them as such.
Example 2.29. The following are some examples of perfect sets.
• Any interval on the form [a, b] ⊂ R, and any finite union of such intervals.
• The Cantor5 set C .
In certain contexts a set can be perfect while in other contexts it is not. For
example, Q has no isolated points but is not closed in R and is therefore not
perfect in R. However, Q is closed in itself and therefore perfect in itself. In
fact a metric space that lacks isolated points will always be perfect in relation
to itself.
Another way we can categorize sets is by their relation to the metric space
they are in. We introduce a category of sets where all points in the metric space
are either in the set or limit points of the set.
Definition 2.30. A ⊂ X is dense in X if A = X.
Example 2.31. Let x ∈ R. Then every Bε (x) contains y ∈ Q \ {x}. Therefore
x is a limit point of Q. This means that Q = R, i.e. Q is dense in R.
We can also categorize points in a set based on their surroundings by looking
at whether or not they make up the boundary of a set, i.e. if every neighborhood
around the point is only partially contained in the set.
Definition 2.32. A point x ∈ X is a boundary point of A if both Bε (x) ∩ A
and Bε (x) \ A are non-empty for every ε > 0. The set of all boundary points of
A is called the boundary of A, written as ∂A.
We can conclude that a non-isolated point is a boundary point if and only
if it is a limit point, but not an interior point. Thus, a set is closed if and only
if it contains its boundary points. Moreover A = A ∪ ∂A.
We also define what it means for a set to be bounded. You probably have
an intuitive understanding of this but here we give you the formal definition.
5 Georg Cantor (1845–1918), German mathematician.
2.7. Relative or Induced Topology 19
Definition 2.33. A set A ⊂ X is bounded if there is some x0 ∈ X and r0 > 0

such that A ⊂ Br0 (x0 ).
Note that if A ⊂ Br0 (x0 ), x1 ∈ X and r1 = r0 +d(x0 , x1 ), then A ⊂ Br1 (x1 ).
So Definition 2.33 is independent of the choice of x0 ∈ X.
2.7 Relative or Induced Topology

Within a metric space we can also talk about the topology, i.e. which sets are
open and which are closed, relative to subsets of the space. We call this the
relative or induced topology. If (X, d) is a metric space and Y is a subset of X
then d is automatically a distance function also on Y , and thus (Y, d) is also a
metric space. If E is a subset of Y then we can talk about E being open or closed
with respect to either X or Y . Within Y we often say relatively open/closed.
Example 2.34. Let X = R. If Y = (0, 2), then E = (0, 1] is relatively closed
with respect to Y but it is neither closed nor open with respect to X.
E
Y X
−1 0 1 2 3
If we instead let Y = (−1, 1], then E is relatively open in Y .
E
Y X
−1 0 1 2 3
Theorem 2.35. E is relatively open/closed in Y , where Y ⊂ X, if and only if

there exists a set A that is open/closed in X such that E = A ∩ Y , i.e. E is the
part of A that is in Y .
E
Y X
Proof. We only prove the theorem in the open case, but the proof for the closed
case is very similar. First we prove that if there exists an open set A ⊂ X such
that E = A ∩ Y , then E is open in Y :
A is open in X so for every a ∈ A, there exists an ε > 0 such that
BεX (a) = {x ∈ X : d(x, a) < ε} ⊂ A.
As Y ⊂ X it follows that
BεY (a) = {x ∈ Y : d(x, a) < ε} ⊂ A.
As also BεY (a) ⊂ Y , we see that
BεY (a) ⊂ E,
and thus E is open in Y .

We conclude the proof by proving the converse statement, i.e. that if E is
open in Y , then there exists an open set A ⊂ X such that E = A ∩ Y .
Since E is open in Y we can for every point x ∈ E choose an rx > 0 such
that BrYx (x) ⊂ E. Choose
[
A= BrXx (x),
x∈E
which is open in X, since the union of open balls is also open by Theorem 2.10.
Furthermore, we realize that
[
A∩Y = BrYx (x) ⊂ E,
x∈E
and that since the set BrYx (x) must contain at least all points in E,
S
x∈E
[
E⊂ BrYx (x) = A ∩ Y.
x∈E
Thus E = A ∩ Y .
Chapter 3
Limits and Sequences
3.1 Sequences
We start by looking at sequences in metric spaces. A sequence, much like a set,
is a collection of elements. But unlike a set, the order matters and the same
element can appear more than once. Sequences can be finite or infinite, in which
case we can talk about the limit of the sequence when we go arbitrarily far along
the sequence. Here we will only consider infinite sequences (xn ) where n runs
through 1, 2, 3, ... . Definitions 1.1 and 1.2 easily generalize to metric spaces, as
we only need to replace |x − y| with d(x, y).
Definition 3.1. A sequence (xn ) ⊂ X converges to the limit L as n → ∞ if for

every ε > 0 there exists some N ∈ N such that d(xn , L) < ε for all n ≥ N . We
write
xn → L or lim xn = L.
n→∞
Expressed with quantifiers, xn → L as n → ∞ if
∀ε > 0 ∃N ∈ N : n ≥ N =⇒ d(xn , L) < ε.
The limit of a sequence in a metric space is unique, i.e. if xn → L and

xn → M then L = M . We leave showing this as an exercise for the reader.
Now that we have defined convergence of sequences in metric spaces, we can
present a new way of determining if a point is a limit point to a set or not.
Theorem 3.2. x is a limit point to A ⊂ X if and only if there exists a sequence

(xn ) ⊂ A \ {x} such that xn → x.

22 Chapter 3. Limits and Sequences
Proof. We begin by proving that x being a limit point implies that there exists
a sequence (xn ) ⊂ A \ {x} such that xn → x: For all j, there is an xj such that
xj ∈ B1/j (x) ∩ (A \ {x}).
This means that d(xj , x) < 1/j and thus xj → x.

Next we want to prove the converse, i.e. that the existence of a sequence
(xn ) ⊂ A \ {x} such that xn → x implies that x is a limit point: For every
ε > 0, there is an n such that d(xn , x) < ε, which means that
xn ∈ Bε (x) ∩ (A \ {x})
and thus
Bε (x) ∩ (A \ {x}) ̸= ∅.
A particularly interesting type of sequence that we can talk about when we

have a notion of distance are so-called Cauchy sequences. Intuitively, a Cauchy
sequence is a sequence where, arbitrarily far along the sequence, the distance
between elements becomes arbitrarily small.
Definition 3.3. (an ) ⊂ X is a Cauchy sequence (or (an ) is Cauchy for short)
if for every ε > 0, there exists some N ∈ N such that d(an , am ) < ε for all
n, m ≥ N . Expressed with quantifiers, (an ) is a Cauchy sequence if
∀ε > 0 ∃N ∈ N : n, m ≥ N =⇒ d(an , am ) < ε.
This type of sequence is interesting because the property it has is closely tied
to convergence, and entails other important properties of the sequence. First
we look at the relation between Cauchy sequences and convergent sequences.
Theorem 3.4. If (an ) converges, then (an ) is a Cauchy sequence.
Proof. Let ε > 0. Suppose an → L. Then there exists N ∈ N such that if

n ≥ N then d(xn , L) < 2ε . Then, for all n, m ≥ N , we have d(xn , xm ) ≤
d(xn , L) + d(L, xm ) < 2ε + 2ε = ε.
Next we want to examine the relation between Cauchy sequences and bounded
sequences. Recall that as a sequence is a set (if we ignore the ordering), Defini-
tion 2.33 of bounded sets applies here as well.
Theorem 3.5. If (an ) is a Cauchy sequence, then (an ) is bounded.

3.2. Completeness 23
Proof. Since (an ) is Cauchy, for every ε > 0, there exists N ∈ N such that
d(an , am ) < ε for all n, m ≥ N . Now let ε = 1, that is d(an , am ) < 1 if
n, m ≥ N . Then
d(an , aN ) < 1
if n > N . Thus all an , n ≥ N , are contained within the open ball B1 (aN ),
so (an )∞
n=N is bounded. The sequence (an )n=1 is finite and thus is bounded.
N
Hence (an ) is bounded.
3.2 Completeness
From Theorem 3.4 we know that all convergent sequences in a metric space
are Cauchy sequences. But what about the converse statement; are all Cauchy
sequences convergent?
In the case of R with the standard distance function the answer is yes,
by Cauchy’s convergence principle (Theorem 1.3). However, using the same
metric on Q we can easily see that the sequence (xn ) = (3, 3.1, 3.14, 3.141, ...)
is a Cauchy sequence, and that it approaches π, which is not in Q. Hence the
sequence is Cauchy but not convergent in Q.
So the converse statement is not generally true. But the property of all
Cauchy sequences being convergent is so important that metric spaces which
have it are given their own category.
Definition 3.6. A metric space is complete if every Cauchy sequence is conver-

gent.
As we saw above, when we use the standard distance function, R is complete

but Q is not. Using different distance functions on the same set can change
whether or not that metric space is complete. Looking back at Example 2.3 we
can examine some of the different metric spaces to see if they are complete.
The L1 -norm on C([0, 1]) does not give rise to a complete metric space. For
example, consider the sequence of functions (fn ) where

0,
 if 0 ≤ x ≤ 12 − n1 ,
n 1 1
, if 12 − n1 ≤ x ≤ 21 + n1 ,

fn (x) = x− −
2 2 n
1, if 12 + n1 ≤ x ≤ 1,

on A = [0, 1] with 1 ≤ n < ∞. All of these functions are continuous on the

interval [0, 1] so they are in C([0, 1]). They also get arbitrarily close to each
other using the L1 -norm as n → ∞ so (fn ) is Cauchy. However, the pointwise
limit of this sequence is the function


0, if 0 ≤ x < 2 ,
 1
f∞ (x) = 12 , if x = 12 ,
1, if 12 < x ≤ 1,


which is not continuous on [0, 1] and therefore not in C([0, 1]). Moreover there
is no continuous function f such that f = f∞ almost everywhere. Hence the
sequence does not converge and (X, d) = (C([0, 1]), L1 ) is not a complete space.
However, using the supremum norm, C([0, 1]) gives rise to a complete metric
space.
1
1
fn (x) 2 f∞ (x)
2
1− 1 1+ 1 1
2 n 2 n 2
Theorem 3.7. C([a, b]) with the supremum norm is a complete metric space.
We do not have the tools necessary to prove this yet so we will return to it
later and prove it in Chapter 5.
Next we present a theorem relating completeness and perfect sets to cardi-
nality.
Theorem 3.8. If X is a complete metric space and F ⊂ X is perfect, then F

is uncountable.
The proof for Theorem 3.8 is very tricky, even for simple special cases like
X = R, and is outside the scope of this thesis.
3.3 Continuous Functions

Close to the heart of analysis are limits of functions and continuity, which are
essential for things such as derivatives and integrals, among many other things.
Limits are most often very closely tied to notions of distance, as we talk about
3.3. Continuous Functions 25
functions approaching a value for instance. You are likely familiar with the defi-
nition of limits for real functions (see Definition 1.5), so we start by generalizing
that definition to metric spaces.
Definition 3.9. Let f : X → Y be a mapping, where X and Y are metric spaces
with metrics dX and dY respectively. Assume that a is not isolated in X. Then
f (x) tends to L as x tends to a if for every ε > 0 there exists δ > 0 such that
dY (f (x), L) < ε whenever 0 < dX (x, a) < δ. We write
f (x) → L as x → a or lim f (x) = L.

x→a
Using quantifiers to express the definition, we see that f (x) → L as x → a if
∀ε > 0 ∃δ > 0 : 0 < dX (x, a) < δ =⇒ dY (f (x), L) < ε.
As with limits of sequences also here the limit L is unique if it exists. That is
provided that a is not isolated. On the other hand, if a is isolated the condition
0 < dX (x, a) < δ is void if δ > 0 is small enough and thus any L satisfies the
definition. Thus we do not define limits as x → a if a is isolated.
Example √ We begin with a familiar example in R. We want to show that
√ 3.10.
limx→a x = a when a > 0. We use the standard distance function in R and
we let ε > 0,
√ √ |x − a| |x − a|
| x − a| = √ √ ≤ √ .
x+ a a
√
If |x − a| < ε a then
|x − a|
√ < ε,
a
√
which is what we want. We set δ = ε a so that the inequality holds.
Next we generalize the definition of continuity to functions on general metric
spaces. This definition, much like the definition of limits we have just seen, will
be very recognizable if you are familiar with the definition in the case of real-
valued functions with the standard metric (see Definition 1.6).
Definition 3.11. f : X → Y is continuous at a ∈ X if for every ε > 0, there ex-
ists some δ > 0 such that dY (f (x), f (a)) < ε whenever dX (x, a) < δ. Expressed
with quantifiers, f is continuous at a ∈ X if
∀ε > 0 ∃δ > 0 : dX (x, a) < δ =⇒ dY (f (x), f (a)) < ε.
If a function is continuous at every point of a set then the function is said to be

continuous on that set.
Remark 3.12. An alternative, and completely equivalent, way of defining con-

tinuity is by using open balls. Expressed with quantifiers, f is continuous at
a ∈ X if
∀ε > 0 ∃δ > 0 : x ∈ BδX (a) =⇒ f (x) ∈ BεY (f (a)).
Here BδX (a) is an open ball in X and BεY (f (a)) is an open ball in Y .
Note that if f is defined at a ∈ X and a is isolated, then f is continuous at
a. Thus every function f : Z → Z is automatically continuous. If a however is a
limit point, then f is continuous at a if and only if limx→a f (x) = f (a). Recall
that limits are not defined at isolated points.
Example 3.13. The function

x,
 if 1 < x < 2,
f (x) = 5 − x, if 2 < x ≤ 3,
if x = 4,

2,

is continuous on (1, 2) ∪ (2, 3] ∪ {4}. We see this since all points between 1 and
3 are limit points and 4 is isolated but f is not defined at 1 and 2.
y
4
3
2
1
x
1 2 3 4
Theorem 3.14. f : X → Y is continuous if and only if the preimage

f −1 (G) = {x ∈ X : f (x) ∈ G}
is open for all open G ⊂ Y .
Proof. We prove that if f is continuous then f −1 (G) is open for all open G ⊂ Y :
Let x ∈ f −1 (G). Then f (x) ∈ G. Since G is open, Bε (f (x)) ⊂ G for some
ε > 0. Also, since f is continuous, there exists δ > 0 so that Bδ (x) ⊂ f −1 (G),
see Remark 3.12. Hence f −1 (G) is open.
Next we prove the converse: Let a ∈ X and ε > 0. The set U = f −1 (Bε (f (a)))
is open and contains a, so there exists some Bδ (a) ⊂ U . So if d(x, a) < δ, then
x ∈ Bδ (a) ⊂ U . Therefore f (x) ∈ f (U ) = Bε (f (a)), i.e. d(f (x), f (a)) < ε.
3.3. Continuous Functions 27
Remark 3.15. Theorem 3.14 gives an alternative to Definition 3.11 for defining
continuous function. This is also the standard way of doing it in more general
topological spaces (see Chapter 7). However a pointwise definition (as in Defi-
nition 3.11) is also essential. Note as well that we can substitute open sets with
closed sets in Theorem 3.14, and the statement remains true.
The following two examples illustrate why one has to use preimages in The-
orem 3.14.
Example 3.16. Let f (x) = x2 . Then f ((−1, 1)) = [0, 1). The set [0, 1) is not
open while (−1, 1) obviously is. However, f −1 ((0, 1)) = (−1, 0) ∪ (0, 1). Here,
both (−1, 0) ∪ (0, 1) and (0, 1) are open.
Example 3.17. Let f (x) = 1/(1 + x2 ). Then the image of the closed set [0, ∞)
under f is (0, 1], which is not closed. However the preimage of the closed set
[ 21 , 1] is [−1, 1], which is also closed.
Chapter 4
Compactness
In real analysis, you have likely noticed the importance of compact sets. For
instance, functions defined on compact sets have more predictable properties
than ones defined on other types of sets. The definition of compact sets in Rn
(Definition 1.8) does not carry over to arbitrary metric spaces, and it is simply
a special case of the “real” definition of compactness. For the general definition
we first need a building block that utilizes open sets.
Definition 4.1. An open cover of A is a family of open sets Gλ such that
[
A⊂ Gλ .
λ
Definition 4.2. A set K is compact if every open cover {Gλ } of K has a finite
subcover Gλ1 ∪ ... ∪ GλN that also covers K.
Although this definition is seemingly very different from Definition 1.8, it
is equivalent in the case of Rn . Moreover, we can see that proving that a set
is not compact is easier than proving that it is compact. All we need when
proving non-compactness is a single example of an open cover that lacks a finite
subcover, while to prove compactness, we need to show that every open cover
has a finite subcover.
Example 4.3. The set A = [1, ∞) has an open cover such that
∞
[
A⊂ Gn , where Gn = (n − 1, n + 1).
n=1
This open cover has no finite subcover that contains A, therefore A is not
compact. Note that A is closed but not bounded.

30 Chapter 4. Compactness
G1 G3
A
0 1 2 3 4 5 6 7
G2
Example 4.4. The set A = (0, 1] has an open cover such that
∞
[ 1
A⊂ ,2 .
n=1
n
This time, A is bounded but not closed. However, there is still no finite
Smsubcover.
For any m, the points in the interval (0, m
1
] will not be covered by n=1 n1 , 2 .

A is again not compact.
G3
G2
A G1
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2
Next we define another type of compactness based on the convergence of

subsequences, which we first need to define.
Definition 4.5. A subsequence of a sequence (xn ) is a sequence (yk ) such that

yk = xnk , where n1 < n2 < ... is a sequence of indices.
Intuitively, we can think of a subsequence of (xn ) as a sequence constructed

by removing some of the elements of (xn ) and keeping the relative order of the
remaining elements.
Example 4.6. The sequence of the even positive integers and the sequence of
prime numbers are subsequences of the natural numbers N.
Definition 4.7. A set K is sequentially compact if every sequence in K has a

subsequence that converges to a limit in K.
We will now examine how compactness and sequential compactness relate

to each other.
Theorem 4.8. Let X be a metric space. A set K ⊂ X is compact if and only

if it is sequentially compact.
31
The proof for this theorem is very tedious and is outside the scope of this
thesis. The interested reader can find the proof on pp. 149–150 in [3] under
Theorem 9.14.
For the special case of X = Rn in the previous theorem, we get an additional
equivalent property, as previously mentioned. This is formulated in a theorem
that is very important in real analysis.
Theorem 4.9. (The Heine1 –Borel2 –Lebesgue3 theorem)
For a set K ⊂ Rn , the following are equivalent:
• K is compact,
• K is sequentially compact,
• K is closed and bounded.
The proof for this theorem will also be omitted but can be found in the
same place as the proof for the previous theorem, pp. 149–150 in [3] under
Theorem 9.14.
Remark 4.10. In [1], Abbott refers to sequential compactness simply as com-
pactness. Since he is only concerned with R, these are equivalent but it is still
important to note that they are not the same. Both Definition 4.2 and 4.7 apply
in more general spaces than metric spaces, namely topological spaces, in which
they are not equivalent. More on this in Chapter 7.
Although a set being closed and bounded does not imply compactness in
general metric spaces, compactness does imply closedness and boundedness.
Theorem 4.11. Let X be a metric space. If K ⊂ X is compact then K is
closed.
Proof. Let K be a compact set in a metric space and a ∈ K c . For every x ∈ K,

let rx = 12 d(x, a). The collection {Brx (x)}x∈K forms an open cover of K, and
since K is compact there exists a finite subcover {Brxj (xj )}nj=1 that also covers
K. Let r0 = min1≤j≤n rxj > 0. Then Br0 (a) ∩ K = ∅, and thus a is not
a limit point to K. Hence K contains all of its limit points and is closed by
Definition 2.16.
Theorem 4.12. Let X be a metric space. If K ⊂ X is compact then K is

bounded.
1 Eduard Heine (1821–1881), German mathematician.
2 Émile Borel (1871–1956), French mathematician.
3 Henri Lebesgue (1875–1941), French mathematician.
32 Chapter 4. Compactness
Proof. Let K ⊂ X be compact. Take the family of concentric open balls

{Bn (a)}, where a ∈ X is arbitrary. This is an open cover of the whole space,
hence it is also an open cover of K. Since K is compact, this open cover has a
finite subcover
N
[
Bn (a) = BN (a) ⊃ K.
n=1
Hence K is bounded by Definition 2.33.
Example 4.13. Here, we give an example of a closed and bounded subset of a

metric space that is not compact nor sequentially compact. The space
n ∞
X 1/2 o
2
ℓ = x = (x1 , x2 , ...) : xk ∈ R, ∥x∥2 = x2k <∞
k=1
with d(x, x ) = ∥x − x ∥ is a complete metric space of infinite dimension. Take

′ ′
the set E = {e1 , e2 , ...} ⊂ ℓ2 , where e1 = (1, 0, ...), e2 = (0, 1, ...), etc. Note that
∥en ∥2 = 1 for all n, so E is bounded. E has no limit points, neither in E nor
outside of E, hence E is also closed. However, the sequence (en )∞ n=1 ⊂ E has no
convergent subsequence so E is not sequentially compact. So by Theorem 4.8,
E is not compact either. This can also be seen directly, take for instance the
open cover
∞
[
B1 (en ) ⊃ E,
n=1
which lacks a finite subcover.

Next we shall formulate several theorems relating to properties of compact
sets. These will illustrate how important and useful the property of compactness
is, both in general metric spaces and in the specific case of R.
Theorem 4.14. (Cantor’s encapsulation theorem for compact sets)
If K1 ⊃ K2 ⊃ ... are non-empty compact subsets of a metric space, then
∞
\
Kn ̸= ∅.
n=1
Proof. Choose xn ∈ Kn for each n ∈ N. Because of the nesting, all xn are

in K1 and thus there exists a subsequence (xnk ) that converges to x ∈ K1 .
Since nk > nk−1 we get that nk ≥ k and thus xnk ∈ Km for all k ≥ m. As
Km T is closed, by Theorem 4.11, we get that x ∈ Km for all m. Therefore
∞
x ∈ m=1 Km , so it is non-empty.
33
Theorem 4.15. If K ⊂ X is compact and E ⊂ K is closed, then E is compact.
Proof. Let {Gλ }λ∈Λ be an arbitrary open cover of E. Since E is closed, E c is

open by Theorem 2.26, and is, trivially, an open cover of itself. Thus, {E c } ∪
{Gλ : λ ∈ Λ} is an open cover of K, which has a finite subcover {E c } ∪ {Gj }N
j=1
since K is compact. Then {Gj }N j=1 is a finite subcover of E ⊂ K, and hence E
is compact.
Theorem 4.16. If f : A → R is continuous and K ⊂ A is compact, then f (K)

is compact.
Proof. Let {Gλ } be an open cover of f (K). Then {f −1 (Gλ )} is an open cover
of K by Theorem 3.14 as f is continuous. Since K is compact, there exists
some finite subcover {f −1 (Gj )}N
j=1 of K which means that {Gj }j=1 is a finite
N
subcover of f (K).
The next theorem will be very familiar from single variable analysis. How-
ever, the formulation of the theorem you have seen previously will most likely
have only regarded functions of a real variable. Here we will give the theorem
in its most general form.
Theorem 4.17. (Extreme value theorem)
If a function f : K → R is continuous and K is compact, then there exist
x1 , x2 ∈ K such that f (x1 ) ≤ f (x) ≤ f (x2 ) for all x ∈ K.
Proof. Since f is continuous and K is compact, it follows from Theorem 4.16

that f (K) is compact. As f (K) ⊂ R it is closed and bounded, by Theorem 4.9.
Hence inf f (K) ∈ f (K) and sup f (K) ∈ f (K).
Example 4.18. Let f (x) = 1/x with x ∈ (0, 1). The set (0, 1) is not compact,
and f lacks both maximum and minimum.
Example 4.19. Let (
x, if 0 ≤ x < 1,
f (x) =
0, if x = 1.
f is defined on a compact set but is not continuous, and lacks a maximum.
Theorem 4.20. E is compact in Y ⊂ X if and only if E is compact in X,
where X and Y have the same metric.
Proof. Sequential compactness is an inner property, i.e. it is not dependent

on anything outside of E, so it is automatically the same for X and Y . By
Theorem 4.8 the same is true for compactness.
Chapter 5
Uniform Convergence
In Chapter 3 we discussed the convergence of sequences in metric spaces. We will

now return to this subject, specifically looking at sequences in metric function
spaces. New tools will be introduced which will help us study sequences of
functions. We begin by defining two new types of convergence.
Definition 5.1. Let f, fn : X → R. The sequence fn → f pointwise if for every

x ∈ X and for every ε > 0, there exists some N ∈ N such that |fn (x)−f (x)| < ε
for all n ≥ N . Expressed with quantifiers,
∀x ∈ X ∀ε > 0 ∃N ∈ N : n ≥ N =⇒ |fn (x) − f (x)| < ε. (5.1)
Definition 5.2. Let f, fn : X → R. The sequence fn → f uniformly (some-

times written fn ⇒ f ) if for every ε > 0 there exists some N ∈ N such that
|fn (x) − f (x)| < ε for all x ∈ X and all n ≥ N . Expressed with quantifiers,
∀ε > 0 ∃N ∈ N ∀x ∈ X : n ≥ N =⇒ |fn (x) − f (x)| < ε. (5.2)
Note that the difference between pointwise and uniform convergence is seen
in the positioning of the quantifiers in (5.1) and (5.2).
The function sequence (fn ) converging to f uniformly means that, by (5.2),
for every ε > 0 there exists N such that
∥fn − f ∥∞ = sup |fn (x) − f (x)| ≤ ε if n ≥ N,

x∈X
and thus fn → f uniformly if and only if ∥fn − f ∥∞ → 0. Uniform convergence

is exactly convergence in the supremum norm.

36 Chapter 5. Uniform Convergence
Example 5.3. Let fn (x) = nxe−nx , defined on A = [0, ∞). Since x ≥ 0,

fn (x) → 0 as n → ∞ for every x ∈ A. Therefore fn → 0 pointwise. Notice that
fn′ (x) = ne−nx − n2 e−nx = n(1 − nx)e−nx .
So ∥fn − 0∥∞ = supx≥0 |fn (x)| = 1

e ̸→ 0 and thus fn ̸→ 0 uniformly.
1 1
x n e fn
fn′ + 0 − fn+1
1 fn+2
fn ↗ e ↘
Example 5.4. The same function as in Example 5.3 but defined on Aδ = [δ, ∞),
δ > 0. If n ≥ 1/δ, then fn is decreasing on Aδ and thus
sup |fn (x)| = fn (δ) → 0.

x∈Aδ
So fn → f uniformly on Aδ for all δ > 0.
Example 5.5. Let fn (x) = xn , defined on A = [0, 1]. Then

(
1, if x = 1,
fn (x) → f (x) := ,
0, if 0 ≤ x < 1,
but f is not continuous. We see that ∥fn − f ∥∞ = 1 for all n and therefore
fn ̸→ f uniformly.
f1
fn
As you know, in analysis we are particularly interested in continuous func-

tions. Thus is it of great interest whether or not different types of convergence
of function sequences preserve continuity.
Theorem 5.6. If fn is continuous at a ∈ A and fn → f uniformly then f is

continuous at a.
37
Proof. Let ε > 0. Then for every x ∈ A,

|f (x) − f (a)| ≤ |f (x) − fn (x)| + |fn (x) − fn (a)| + |fn (a) − f (a)| . (5.3)
| {z } | {z } | {z }
< 3ε if n≥N < 3ε if d(x,a)<δ < 3ε if n≥N
Since fn → f uniformly, for every ε > 0 there exists some N ∈ N such that
|fn (x) − f (x)| < 3ε for all n ≥ N and all x ∈ A. Thus, the first and the third
terms in (5.3) are bounded from above by 3ε for all n ≥ N .
Since fn is continuous, for every ε > 0 there exists some δ > 0 such that
|fn (x) − fn (a)| < 3ε whenever d(x, a) < δ. Thus, the second term in (5.3) is also
bounded from above by 3ε when d(x, a) < δ.
If we choose n ≥ N and then δ > 0, we see that |f (x) − f (a)| < ε whenever
d(x, a) < δ, and so f is continuous at a.
Remark 5.7. If fn → f pointwise and f and all fn are continuous, then
lim lim fn (x) = lim lim fn (x)
n→∞ x→a x→a n→∞
since both sides evaluate to f (a). But in Example 5.5,

lim xn = 1 ̸= 0 = lim− lim xn .
x→1− x→1 n→∞
Recall 5.8. Convergence in the supremum norm is equivalent to uniform con-

vergence.
We now return to the proof of Theorem 3.7 as we have developed the neces-
sary tools.
Proof of Theorem 3.7. Let (fn ) be Cauchy in C([a, b]), i.e.
∀ε > 0 ∃N ∈ N : n, m ≥ N =⇒ |fm (x) − fn (x)| < ε (5.4)
is true for all x ∈ [a, b]. Therefore (fn (x)) is also Cauchy in R for every x, and
hence it is also convergent since R is complete. Let
f (x) := lim fn (x).
n→∞
Now we need to show that fn → f uniformly: Let ε > 0. Take N to be the

same as in (5.4), then fix n ≥ N and let m → ∞. This gives, if n ≥ N , that
|fn (x) − fm (x)| < ε for all x and all m ≥ N,
and hence
|fn (x) − f (x)| ≤ ε for all x,
i.e. fn → f uniformly.
38 Chapter 5. Uniform Convergence
Since we did not use the fact that fn ∈ C([a, b]) in the proof of Theorem 3.7,
we can make an even stronger statement.
Theorem 5.9. (Cauchy’s criterion)
The sequence (fn ) converges uniformly if and only if (fn ) is Cauchy with respect
to the supremum norm.
Example 5.10. Let fn (x) = sin(2n πx). Then (fn ) is bounded in C([0, 1]). Fix
m > n ≥ 1 and let x = 2m+1
3
. Then
3π
fm (x) = sin(2m πx) = sin = −1.
2
But 0 < 2n πx ≤ 3π
4 so fn (x) = sin(2n πx) > 0. Hence
∥fn − fm ∥∞ ≥ |fn (x) − fm (x)| > 1, (5.5)
i.e. (fn ) is not Cauchy and therefore not convergent, by Theorem 3.4. Moreover,
it also follows from (5.5) that (fn ) has no convergent subsequence.
Remark 5.11. Let B = {f ∈ C([0, 1]) : ∥f ∥∞ ≤ 1} be the closed unit ball

in C([0, 1]). This is a closed and bounded set which by Example 5.10 is not
sequentially compact, hence it is not compact either.
Next, we present a famous theorem relating two properties of sequences of
functions that guarantee uniform convergence of their subsequences. However,
we first need to give the following definition.
Definition 5.12. A family of functions fn : X → Y is equicontinuous at a if
for every ε > 0 there exists some δ > 0 such that for all n, dY (fn (a), fn (x)) < ε
whenever dX (a, x) < δ.
Theorem 5.13. (The Arzelà1 –Ascoli2 theorem)
If fn : [a, b] → R is an equicontinuous family of function and |fn (x)| ≤ M for all
n (uniformly bounded ) and all x ∈ [a, b], then (fn ) has a uniformly convergent
subsequence.
Conversely, if every subsequence of (fn ), where fn ∈ C([a, b]), has a uni-
formly convergent subsequence, then (fn ) is uniformly bounded and equicontin-
uous.
1 Cesare Arzelà (1847–1912), Italian mathematician.

2 Giulio Ascoli (1843–1896), Italian mathematician.
Chapter 6
Curves and Connectedness
6.1 Curves and Pathconnectedness

In your previous analysis courses, you have probably encountered curves. This
might have been in your multivariable analysis or perhaps in a course in complex
analysis. Now, we will revisit them as we have developed some tools that will
help us analyse them. We begin by defining curves in general metric spaces.
Definition 6.1. A curve γ : I → X, where X is a metric space, is a continuous
function on an interval I ⊂ R.
This allows us to depart from the usual geometric interpretation of curves.
With the inherent distance function of our metric space, we can talk about how
long a curve is.
Definition 6.2. Consider the curve γ : [a, b] → X where X is a metric space
with the distance function d. The length of γ is,
n
X
ℓ(γ) = sup d(γ(xj ), γ(xj−1 ))
j=1
over all partitions a = x0 < x1 < ... < xn = b of [a, b]. We say that γ is
rectifiable if ℓ(γ) < ∞.
A rectifiable curve can be parameterized by arc length, denoted ds, so that
ℓ(γ|[s,t] ) = t − s
for all a ≤ s < t ≤ b. Then for all s < t,
d(γ(t), γ(s)) ≤ ℓ(γ|[s,t] ) = t − s.

40 Chapter 6. Curves and Connectedness
Example 6.3. A curve γ : [0, 1] → R2 .
y
4
γ(1)
3
2
1
x
−4 −3 −2 −1 1 2 3 4
−1
γ(0)
−2
−3
−4
We can get a lower bound on the length of the curve by drawing and mea-
suring the length of straight lines between some of the points it crosses through.
We will use the ℓ2 -distance for R2 .
y
4
γ(1)
3
2
1
x
−4 −3 −2 −1 1 2 3 4
−1
γ(0)
−2
−3
−4
The cumulative length of the gray lines is approximately 13.92 units. So

ℓ(γ) ≥ 13.92.
The image of a curve γ is a set in a metric space X. In mathematics, it is
common that we want to integrate over such a set. We can do this using arc
length parameterization.
6.2. Connectedness 41
Definition 6.4. A curve integral, or path integral, with respect to arc length on
the arc length parameterized curve γ : [a, b] → X is defined by,
Z Z b
f ds := f (γ(t)) dt.
γ a
With curves, we can now express a notion of connectedness within sets in

metric spaces.
Definition 6.5. X is pathconnected, or arcwise connected, if for every pair of

points x, y ∈ X, there exists a curve γ : [0, 1] → X such that γ(0) = x and
γ(1) = y.
γ(1) = y
γ(0) = x
If you have taken a course in complex analysis you may recognize this type
of connectedness. Pathconnected sets are very important in complex analysis
as curves and curve integrals are essential to the subject.
6.2 Connectedness
We have presented one way of describing connectedness in metric spaces. Now
we present another way to express this notion of connectedness, this time using
open sets as opposed to curves.
Definition 6.6. A metric space X is disconnected if there exists two sets, A

and B, which are non-empty, open and disjoint (i.e. A ∩ B = ∅), such that the
union of A and B makes up the entire space, X = A ∪ B. A metric space is
connected if it is not disconnected.
√ √
Example 6.7. Let X = Q, and let A = Q ∩ (−∞, 2) and B = Q ∩ ( 2, ∞).
Both A and B are non-empty and open. They are also disjoint since they share
no points. Finally, A and B together make up the entire space, Q = A ∪ B.
Hence Q is disconnected.
Remark 6.8. A set A ⊂ X is connected in the metric space (X, d) if it is

connected as a metric space (A, d) in its own right.
Next we regard the case of subsets of R. The result presented in the next
theorem will be very intuitive if we think about connectedness in a geometric
sense.
Theorem 6.9. A set E ⊂ R is connected if and only if it is either an interval,

a single point set, ∅ or R.
The implication P ⇒ Q is equivalent to its contrapositive statement ¬Q ⇒

¬P , which will be used in the proof below.
Proof. First we show that a connected set E ⊂ R must be either an interval, a

single point set, ∅ or R. We prove the contrapositive statement: If a set E is
not an interval, a single point set, ∅ or R, then there exist a < b < c such that
a, c ∈ E and b ∈/ E. Now, let A = E ∩ (−∞, b) and B = E ∩ (b, ∞). Then A
and B are non-empty, relatively open and disjoint, and they also make up the
entire set, E = A ∪ B. Hence E is disconnected.
Next we show that E = [0, 1] is connected. The other sets are handled
similarly. We give a proof by contradiction: Assume that E is disconnected.
Then there exist A and B that are non-empty, relatively open and disjoint such
that E = A ∪ B. Now we may assume that 1 ∈ B (if this is false, just change
names between A and B). Furthermore, let x = sup A (which exists by the
supremum axiom since A is non-empty and bounded). Then x is a limit point
of A and since A is relatively closed (as B = E \ A is relatively open), we
conclude that x ∈ A. This forces x < 1. However, since A is also relatively
open, there exists some ε > 0 such that x + ε ∈ A. This directly contradicts the
statement that x = sup A.
A B
0 1
Now we will examine some of the cases where connectedness is preserved.

The next two theorems will explore preservation under two types of operations
on sets.
Theorem 6.10. If E ⊂ Rn is connected, then E is connected.

6.2. Connectedness 43
We leave the proof of Theorem 6.10 as an exercise to the reader.
Theorem 6.11. Let X and Y be metric spaces. If f : X → Y is continuous

and X is connected, then f (X) is connected.
Proof. We prove the contrapositive statement: Assume f (X) is disconnected.

Then there exist non-empty, open and disjoint A and B such that A∪B = f (X).
Let C = f −1 (A) and D = f −1 (B). Then C and D are non-empty, disjoint and
X = C ∪ D. As C and D are also open by Theorem 3.14, X is disconnected.
A theorem you have most likely seen before in your first year single variable
analysis course is the intermediate value theorem. We can now prove it using
connectedness.
Theorem 6.12. (Intermediate value theorem)

If f : [a, b] → R is continuous and f (a) < m < f (b), then there exists some
x ∈ (a, b) such that f (x) = m.
Proof. The set f ([a, b]) is connected by Theorems 6.9 and 6.11, and so it is an
interval I according to Theorem 6.9. As f (a), f (b) ∈ I, we have also m ∈ I and
thus there exists some x ∈ [a, b] such that f (x) = m.
Now that we have defined connectedness and present the necessary theory,
we can return to pathconnectedness to formulate two theorems relating these
two concepts. We look at their relation both in general metric spaces and in
the case of Rn .
Theorem 6.13. If X is pathconnected, then X is also connected.
Proof. We give a proof by contradiction: Assume X is disconnected. Then there

exist A and B such that they are non-empty, open, disjoint and X = A ∪ B.
Now let a ∈ A and b ∈ B. There exists a curve γ : [0, 1] → X such that γ(0) = a
and γ(1) = b. Let E = γ([0, 1]). Then E is connected by Theorems 6.9 and 6.11.
However, the sets A′ = A ∩ E and B ′ = B ∩ E are non-empty, relatively open,
disjoint and E = A′ ∪ B ′ which means that E is disconnected.
Example 6.14. The following is an example of a set that is connected but not
pathconnected. Let E = {(x, sin x1 ) : 0 < x ≤ 1}. Since E is pathconnected
(this is quite clear as it is a curve) it is also connected by Theorem 6.13. Thus
E = E ∪ I is connected by Theorem 6.10, where I = {0} × [−1, 1]. However,
there are no curves between E and I and thus E is not pathconnected.
Theorem 6.15. If G ⊂ Rn is open, then G is connected if and only if G is

pathconnected.
Proof. The implication that an open set G ⊂ Rn is connected if it is pathcon-

nected is true by Theorem 6.13. Proving the converse statement, however, is
very tricky and outside the scope of this thesis.
Chapter 7
Topological Spaces
7.1 Definition of a Topological Space

Much like when we defined metric spaces, we want to consider more general
spaces then we previously had by means of fewer constraints. This time however,
the structure of the space is not defined by the notion of distance but instead
the notion of openness.
First, we want the entire space and the empty set to be open. Second,
we would like arbitrary unions of open sets to remain open and lastly, finite
intersections of open sets should also remain open.
Definition 7.1. A topological space X = (X, τ ) is a set X together with a
topology τ . A topology τ on X is a set of subsets to X such that
(i) ∅ and X are in τ ,
(ii) a union of sets that are in τ , is in τ ,
(iii) a finite intersection of sets that are in τ , is in τ .
Sets that are in τ are called open.
Almost all spaces you have encountered before are topological spaces, you
might have just not recognized them as such. As an example, all metric spaces
with τ = {G : G is open} (using the definition of open sets from Definition 2.8),
are topological spaces, by Theorems 2.10 and 2.11. If a topology can be defined
using a distance function, we call that topology metrizable. We also say that the
distance function induces the topology. Note that different distance functions
can induce the same topology.

46 Chapter 7. Topological Spaces
Example 7.2. Let X = {a, b, c} and τ = {∅, X, {a}, {b}, {b, c}}. Is (X, τ ) a
topological space?
First we conclude that both ∅ and the whole space X, are open. Next, we
can examine different intersections. Both {a} ∩ {b} and {a} ∩ {b, c} are equal
to the empty set, which is open, and {b} ∩ {b, c} = {b} is also open. Lastly, we
examine have the unions. The union {a} ∪ {b} = {a, b} is not open, and hence
(X, τ ) is not a topological space.
Remark 7.3. There are a few topologies that qualify a space as a topological
space but which are not very interesting. The first one of these is the discrete
topology which is the set of all subsets. Another one is the trivial topology which
only contains ∅ and the set itself.
We continue by defining the convergence of sequences in arbitrary topological
spaces. This definition is quite different from the one given for metric spaces in
Definition 3.1 and thus gives rise to new properties which we will examine in
more detail later on. First however, we need a new tool.
Definition 7.4. A neighborhood V ⊂ of a point a ∈ X, where X is a topological
space, is a subset of X that contains an open set G such that,
a ∈ G ⊂ V.
Definition 7.5. Let (X, τ ) be a topological space. If (an ) is a sequence in X and
L ∈ X, then an converges to L, written an → L, if for every open neighborhood
G of L, there exists an N ∈ N such that an ∈ G for all n ≥ N .
Next, we define pointwise continuity of functions in arbitrary topological
spaces.
Definition 7.6. Let (X, τ ) and (X ′ , τ ′ ) be topological spaces. A function
f : X → X ′ is continuous at a ∈ X if for every open neighborhood G′ of
f (a) there exists some open neighborhood G of a such that f (x) ∈ G′ whenever
x ∈ G. Expressed with quantifiers, f is continuous at a ∈ X if
∀ G′ ∈ τ ′ such that f (a) ∈ G′ ∃ G ∈ τ such that a ∈ G,
and
x ∈ G =⇒ f (x) ∈ G′ .
Remark 7.7. This definition is very similar, and completely equivalent, to Re-
mark 3.12 if X is a metric space, as every open neighborhood must contain
an open ball around the point. We can take the set G′ to be the open ball
BεY (f (a)) as in the remark. Continuity implies the existence of the ball BδX (a)
as in Remark 3.12. This ball in turn qualifies as the open neighborhood G.
Similarly, Definition 7.5 is equivalent to Definition 3.1 if X is a metric space.
7.1. Definition of a Topological Space 47
We can also define continuity of a function on an entire space. This is similar

to Theorem 3.14 just like we mentioned in Remark 3.15.
Definition 7.8. Let (X, τ ) and (X ′ , τ ′ ) be topological spaces. A function

f : X → X ′ is continuous if f −1 (G) is open in X for all open G ⊂ X ′ .
Just like we studied the interior and closure of sets in metric spaces, we can
talk about them in topological spaces. However, instead of using open balls, we
only use open and closed sets together with intersections and unions.
As previously stated, open sets in a topological space are the sets that are
in the topology. Closed sets are defined by Theorem 2.26, i.e. a set is closed if
its complement is open. This means that in all topological spaces, the whole
space and the empty set are both open and closed at the same time, they are
so-called clopen sets.
Definition 7.9. Let Y ⊂ X where X is a topological space.
• The closure of Y , written Y , is the intersection of all closed sets that

contain Y .
• The interior of Y , written Y ◦ , is the union of all open sets that are
contained in Y .
Note that Y ◦ ⊂ Y ⊂ Y . Definition 7.9 is completely equivalent to Defini-

tions 2.13 and 2.22 in metric spaces, by Theorems 2.14 (b) and 2.23 (d), even
though they look very different. We can however define interior just like we did
for metric spaces if we first define inner points in topological spaces.
Definition 7.10. Let A ⊂ X, where X is a topological space. Then a ∈ A is

an inner point of A if a ∈ G ⊂ A for some open set G.
Another category of points from earlier in this thesis, whose definition we

would like to extend to topological spaces, is isolated points.
Definition 7.11. Let A ⊂ X, where X is a topological space. A point a ∈ A

is isolated in A if there exists a neighborhood of a which does not contain any
other points of A.
We also revisit the definition of the boundary of a set but now in arbitrary
topological spaces.
Definition 7.12. Let A ⊂ X, where X is a topological space. The boundary

of A is the set of all points in the closure of A that are not in the interior of A,
i.e. ∂A = A \ A◦ .
7.2 Unique and Non-unique Limits

In metric spaces, we know that the limit of a sequence is unique, i.e. a sequence
can at most converge to one point. This is not generally true in topological
spaces. We illustrate this with an example.
Example 7.13. Let X = {a, b} be a topological space with τ = {X, ∅, {a}}.

Since {a} and X = {a, b} are all of the open sets which contain a, we can
conclude that an → a and bn ̸→ a, where an = a and bn = b for all n. X = {a, b}
is the only open set which contains b, so an → b. Therefore (an ) converges to
both a and b which cannot be regarded as equivalent since bn ̸→ a.
If we let K = {a} then K is compact (as all finite sets are compact), but
since K c = {b} is not open we see that K is not closed. Note that in metric
spaces, compact sets are closed and bounded (but these two properties are not
enough to guarantee compactness), see also Theorem 4.9.
To keep sequences in topological spaces from converging to multiple points,

we can put in place a constraint on the relationship between open sets around
the points in question. We define a type of space with this constraint.
Definition 7.14. A topological space (X, τ ) is a Hausdorff space if for every

pair of points a, b ∈ X, where a ̸= b, there exist open sets A and B such that
a ∈ A, b ∈ B and A ∩ B = ∅.
Intuitively, we can see this as ensuring that all points are sufficiently far
away from each other.
Theorem 7.15. If X = (X, τ ) be a Hausdorff space, then every convergent

sequence in X has a unique limit.
Proof. We give a proof by contradiction: Let X = (X, τ ) be a Hausdorff space

and (xn ) be a convergent sequence in X. Now assume that limn→∞ xn = a and
limn→∞ xn = b where a ̸= b. Since X is Hausdorff, there exist open disjoint sets
A and B such that a ∈ A and b ∈ B. From Definition 7.5, we know that there
exist Na and Nb such that xn ∈ A when n ≥ Na and xn ∈ B when n ≥ Nb .
Now let n ≥ max{Na , Nb } so that xn is in both A and B. But since A ∩ B = ∅,
this is impossible.
As previously stated, we know that sequences in metric spaces are unique

but now we will prove it using Hausdorff spaces.
Theorem 7.16. Every metric space is a Hausdorff space.

7.3. Lp -Spaces 49
Proof. Let (X, d) be a metric space and a, b ∈ X where a ̸= b. Then A := Br (a)

and B := Br (b), with r = 12 d(a, b), are open and disjoint. Hence the space is
Hausdorff.
B
A
b
a
7.3 Lp -Spaces
For 1 ≤ p < ∞, consider
Z 1
L̃p ([0, 1]) = f: |f |p dx < ∞ ,
0
where f is measurable (see p. 127 in [2]) and defined on [0, 1], together with the
distance function Z 1 1/p
d˜p (f, g) = |f − g|p dx , (7.1)
0
where f, g ∈ L̃ ([0, 1]). This does not qualify as a metric space since there are
p
many different functions that have the distance 0 between each other. As an
example, consider the characteristic function at the point 1/2,
(
1, if x = 1/2,
δ1/2 =
0, if x ̸= 1/2.
The distance between δ1/2 and the constant function f (x) = 0 will be zero even
though they are not identical.
Instead we call this type of space a semi-metric space. It is, simply put, a
metric space where the constraint (i) in Definition 2.1 is relaxed as we no longer
require the distance between two distinct points to be non-zero. We still require
that d(x, x) = 0 for every x ∈ X.
This semi-metric space does qualify as a topological space where the topology
is induced by the semi-metric (7.1). A set G ⊂ L̃p ([0, 1]) is open if for every
f ∈ G, there exists some ε > 0 such that g ∈ G whenever d˜p (f, g) < ε.
In order to create a metric space out of this set and distance function, we
have to introduce equivalence classes.
Definition 7.17. A function f is equivalent to another function g, written
˜ g) = 0.
f ∼ g, if d(f,
Remark 7.18. Notice that this equivalence relation is reflexive, symmetrical

and transitive, i.e.
(a) f ∼ f ,
(b) if f ∼ g then g ∼ f ,
(c) if f ∼ g and g ∼ h, then f ∼ h.
Definition 7.19. The equivalence class of f is a set such that
[f ] = {g : g ∼ f }.
˜
Intuitively, we group together all functions that have zero d-distance to f ,
and by transitivity to each other.
Remark 7.20. If d˜p (f, g) = 0 and fn → f then fn → g. Thus we do not have
unique limits in L̃p ([0, 1]).
With this equivalence class we can adjust L̃p ([0, 1]) slightly so that it becomes
a proper metric space. The set
Lp ([0, 1]) = {[f ] : f ∈ L̃p ([0, 1])}
together with the Lp -norm
d([f ], [g]) = d˜p (f, g)
qualifies as a metric space since all functions that previously had zero distance
between them are now one single element; an equivalence class. By Theo-
rems 7.15 and 7.16, limits in Lp ([0, 1]) are unique.
Remark 7.21. Note that in Example 7.13 we cannot create “natural” equiva-
lence classes (so that an → a if and only if [an ] → [a]) to get unique limits.
7.4 Compactness
Since the compactness of a set K ⊂ X is immediately dependent on which sets
are open in X, we take particular interest in compactness when studying topo-
logical spaces. We begin with a definition which is identical to the corresponding
Definition 4.2 for metric spaces.
Definition 7.22. A topological space (X, τ ) is compact if every open cover
{Gλ } of X has a finite subcover Gλ1 ∪ ... ∪ GλN that also covers X.
7.5. Further studies 51
Additionally, we can talk about sequential compactness. This is also very

strongly related to which sets are open when we consider the definition of con-
vergent sequences in topological spaces. Again, we give a definition which is
identical to the corresponding Definition 4.7 for metric spaces.
Definition 7.23. A topological space (X, τ ) is sequentially compact if every
sequence in X has a convergent subsequence.
We previously stated in Theorem 4.8 that in all metric spaces, compactness
and sequential compactness are equivalent. This is no longer true in arbitrary
topological spaces. If you would like to see an example of this, we recommend
Example 13.5 in [3] which shows a sequentially compact topological space that
is not compact and Example 105 in [4] that shows a compact topological space
that is not sequentially compact.
7.5 Further studies

What we have shown you so far contains some of the core elements for further
studies in analysis. We will end with a few more examples of topological spaces.
We begin with R = R ∪ {±∞} and Rn = Rn ∪ {∞}.
Example 7.24. We can construct a “natural” topology on R if we let G be
open in R if the following conditions hold:
(a) G ∩ R is open in R.
(b) There exists a point a ∈ R such that (a, ∞] ⊂ G if ∞ ∈ G,
(c) There exists a point a ∈ R such that [−∞, a) ⊂ G if −∞ ∈ G.
This topology is metrizable but there is no “natural” metric.
Example 7.25. Similarly, we can construct a topology on Rn , by letting G be
open in Rn if
(a) G ∩ Rn is open in Rn ,
(b) there exists a compact set K such that Rn \ K ⊂ G if ∞ ∈ G.
This topology is also metrizable, but there is no natural metric.
If you have taken, or will take, a course in Fourier analysis you will most likely
encounter distributions, a generalization of functions. Sequences of distributions
can converge and will give rise to a non-metrizable topology. Thus we can
talk about sets of distributions being compact or not. This is usually not very
rigorously covered in first courses on Fourier analysis but nonetheless interesting

to know.
Another example is weak and weak∗ convergence on a Banach space X which
appears in functional analysis. They give rise to the weak and weak∗ topologies,
which are non-metrizable if X is of infinite dimension. These types of conver-
gence are also important in e.g. probability theory and when solving certain
minimization problems.
Bibliography
[1] Stephen Abbott. Understanding Analysis. Springer, London, 2nd edition,

2015. 6, 31
[2] Frank Burk. Lebesgue Measure and Integration: An Introduction. Wiley,
New York, 1st edition, 1997. 49
[3] Gerard Buskes and Arnoud van Rooij. Topological Spaces. Springer, New
York, 1st edition, 1997. 31, 51
[4] Lynn Arthur Steen and J. Arthur Seebach. Counterexamples in Topology.
Springer, New York, 2nd edition, 1978. 51

Appendix A
English–Swedish Glossary
English Swedish
Ball Klot
Boundary point Randpunkt
Bounded Begränsad
Closed Sluten
Closure Slutet hölje
Complete Fullständig
Connected Sammanhängande
Dense Tät
Disjoint Disjunkt
Interior Inre
Limit point Hopningspunkt
Open cover Öppen övertäckning
Pathconnected Bågvis sammanhängande
Sequence Följd
Uniform convergence Likformig konvergens

Appendix B
Swedish–English Glossary
Swedish English
Begränsad Bounded
Bågvis sammanhängande Pathconnected
Disjunkt Disjoint
Fullständig Complete
Följd Sequence
Hopningspunkt Limit point
Inre Interior
Klot Ball
Likformig konvergens Uniform convergence
Randpunkt Boundary point
Sammanhängande Connected
Sluten Closed
Slutet hölje Closure
Tät Dense
Öppen övertäckning Open cover

Linköping University Electronic Press
Copyright
The publishers will keep this document online on the Internet – or its possible
replacement – from the date of publication barring exceptional circumstances.
The online availability of the document implies permanent permission for
anyone to read, to download, or to print out single copies for his/her own use
and to use it unchanged for non-commercial research and educational purpose.
Subsequent transfers of copyright cannot revoke this permission. All other uses
of the document are conditional upon the consent of the copyright owner. The
publisher has taken technical and administrative measures to assure authentic-
ity, security and accessibility.
According to intellectual property law the author has the right to be men-
tioned when his/her work is accessed as described above and to be protected
against infringement.
For additional information about the Linköping University Electronic Press
and its procedures for publication and for assurance of document integrity,
please refer to its www home page: http://www.ep.liu.se/.
Upphovsrätt
Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare –
från publiceringsdatum under förutsättning att inga extraordinära omständig-
heter uppstår.
Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda
ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för
ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid
en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av
dokumentet kräver upphovsmannens medgivande. För att garantera äktheten,
säkerheten och tillgängligheten finns lösningar av teknisk och administrativ art.
Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman
i den omfattning som god sed kräver vid användning av dokumentet på ovan
beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan
form eller i sådant sammanhang som är kränkande för upphovsmannens litterära
eller konstnärliga anseende eller egenart.
För ytterligare information om Linköping University Electronic Press se för-
lagets hemsida http://www.ep.liu.se/.
© 2022, Samuel E. Andersson, David Wiman

Metric Spaces

Uploaded by

Copyright:

Available Formats

Metric Spaces

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metric Spaces

Uploaded by

Copyright:

Available Formats

An Introduction to Metric Spaces

Department of Mathematics, Linköping University

Samuel E. Andersson, David Wiman

URL for electronic version:

E. Andersson, Wiman, 2022. iii

URL för elektronisk version:

E. Andersson, Wiman, 2022. v

E. Andersson, Wiman, 2022. vii

E. Andersson, Wiman, 2022. ix

1 Maurice Fréchet (1878–1973), French mathematician.

E. Andersson, Wiman, 2022. 1

2 Introduction to Metric Spaces 7

3 Limits and Sequences 21

6 Curves and Connectedness 39

E. Andersson, Wiman, 2022. 3

∀ε > 0 ∃N ∈ N : n ≥ N =⇒ |xn − L| < ε.

Definition 1.2. A sequence (xn ) in R is a Cauchy sequence if for every ε > 0,

∀ε > 0 ∃N ∈ N : m, n ≥ N =⇒ |xm − xn | < ε.

Although this thesis is mainly concerned with sequences in other spaces

E. Andersson, Wiman, 2022. 5

R. We continue with two theorems regarding convergence of sequences and

f (x) → L or lim f (x) = L,

∀ε > 0 ∃δ > 0 : 0 < |x − a| < δ =⇒ |f (x) − L| < ε.

Definition 1.6. A function f : R → R is continuous at some point a ∈ R if

lim f (x) = f (a).

Definition 1.8. A set K ⊂ Rn is compact if it is closed and bounded.

2 Bernard Bolzano (1781–1848), Czech mathematician.

Introduction to Metric Spaces

2.1 Definition of a Metric Space

Definition 2.1. A metric space X = (X, d) is a set X together with a distance

(i) d(x, y) = 0 if and only if x = y,

(ii) d(x, y) = d(y, x),

(iii) d(x, y) ≤ d(x, z) + d(z, y).

Notice that by setting y = x in constraint (iii) we get 0 ≤ d(x, y) since

E. Andersson, Wiman, 2022. 7

• X = (Rn , d) with the Euclidean distance

where x = (x1 , ... , xn ) and y = (y1 , ... , yn ).

Intuitively, the Hausdorff distance between K and L is the largest possible

• The complex plane C with chordal distance on the Riemann2 sphere,

Chordal distance on the Riemann sphere.

• C([a, b]) with the supremum norm,

d(f, g) = ∥f − g∥∞ = sup |f (x) − g(x)|.

2 Bernhard Riemann (1826–1866), German mathematician.

• ℓ∞ -distance: d(x, y) = maxk |xk − yk |.

2.2 Open Sets

Definition 2.6. An open ball of radius r centered at the point a is defined by

Example 2.12. The set

2.3 Closed Sets

Definition 2.15. A point x ∈ X is a limit point, or accumulation point, of A if

In Examples 2.19–2.21 we (implicitly) assumed them to be within the metric

Definition 2.22. A = A ∪ L, where L is the set of all limit points to A, is called

Theorem 2.23. Let A ⊂ X where X is a metric space. Then

(a) A is closed if and only if A = A.

(c) A is always closed.

(d) A is the smallest closed set containing A.

A = A ∪ {the limit points of A} = A ∪ LA .

We now need to show that LA ⊂ A. To this end, let x ∈ LA and xj ∈ A be

(c) A is closed if and only if A = A by (a) and since A = A is always true

2.4 Complement and De Morgan’s Laws

This shows that \ c [