17.06.2015 Views

Lecture Notes Topology (2301631) Phichet Chaoha Department of ...

Lecture Notes Topology (2301631) Phichet Chaoha Department of ...

Lecture Notes Topology (2301631) Phichet Chaoha Department of ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Lecture</strong> <strong>Notes</strong><br />

<strong>Topology</strong> (<strong>2301631</strong>)<br />

<strong>Phichet</strong> <strong>Chaoha</strong><br />

<strong>Department</strong> <strong>of</strong> Mathematics and Computer Science,<br />

Faculty <strong>of</strong> Science, Chulalongkorn University,<br />

Bangkok 10330, Thailand.<br />

[2001-03-05-06-08-09-11-12]


Contents<br />

Preliminaries 5<br />

Sets and Maps 5<br />

Countable and Uncountable Sets 5<br />

Ordered Sets 6<br />

Chapter 1. Basic Concepts in <strong>Topology</strong> 7<br />

1. Spaces and Subspaces 7<br />

2. Basis for a <strong>Topology</strong> 11<br />

3. Limit Points 14<br />

4. Continuous Maps 17<br />

5. Sequences and Nets 21<br />

6. Product <strong>Topology</strong> 23<br />

7. Quotient <strong>Topology</strong> 26<br />

8. Metric <strong>Topology</strong> 29<br />

Chapter 2. Countability and Separation Axioms 33<br />

1. The Countability Axioms 33<br />

2. The Separation Axioms 35<br />

Chapter 3. Connectedness 41<br />

Chapter 4. Compactness 47<br />

Chapter 5. Function Spaces 59<br />

Chapter 6. Complete Metric Spaces 65<br />

Bibliography 73<br />

3


Preliminaries<br />

Sets and Maps<br />

We assume that the reader is familiar with definitions and basic operations <strong>of</strong><br />

sets as well as <strong>of</strong> maps (or functions) between sets. We will use X c and P(X) to<br />

denote the complement (with respect to some universe) and the powerset <strong>of</strong> the<br />

set X, respectively. For a collection A <strong>of</strong> sets, ∪ A and ∩ A are the union and<br />

intersection <strong>of</strong> sets in A, respectively.<br />

Conventionally, the sets <strong>of</strong> natural numbers, integers, rational numbers, real<br />

numbers and complex numbers will be denoted by N, Z, Q, R and C, respectively.<br />

The properties listed below are easy to verify.<br />

(1) Let {A α } α∈Λ be a collection <strong>of</strong> subsets <strong>of</strong> X.<br />

(a) X − ∪ α∈Λ A α = ∩ α∈Λ (X − A α).<br />

(b) X − ∩ α∈Λ A α = ∪ α∈Λ (X − A α).<br />

(2) Let f : X → Y be a map, A ⊆ X and B ⊆ Y .<br />

(a) f(X − A) ⊇ f(X) − f(A). The equality holds when f is 1-1.<br />

(b) f −1 (Y − B) = X − f −1 (B).<br />

(c) f −1 (f(A)) ⊇ A. The equality holds when f is 1-1.<br />

(d) f(f −1 (B)) ⊆ B. The equality holds when f is onto.<br />

(3) Let f : X → Y be a map, {A α } α∈Λ a collection <strong>of</strong> subsets <strong>of</strong> X, and<br />

{B β } β∈∆ a collection <strong>of</strong> subsets <strong>of</strong> Y .<br />

(a) f( ∪ α∈Λ A α) = ∪ α∈Λ f(A α).<br />

(b) f( ∩ α∈Λ A α) ⊆ ∩ α∈Λ f(A α). The equality holds when f is 1-1.<br />

(c) f −1 ( ∪ β∈∆ B β) = ∪ β∈∆ f −1 (B β ).<br />

(d) f −1 ( ∩ β∈∆ B β) = ∩ β∈∆ f −1 (B β ).<br />

Countable and Uncountable Sets<br />

Definition 0.1. A set A is countable if there is a surjection f : N → A, or<br />

equivalently an injection g : A → N. Thus, a countable set is either finite or<br />

countably infinite. A set that is not countable is called an uncountable set.<br />

Example 0.2. N × N is countable because f : N × N → N given by f(m, n) =<br />

2 m 3 n is injective.<br />

Example 0.3. Since g : N × N → Q + given by g(m, n) = m n<br />

is surjective<br />

and, from the previous example, there is a surjection h : N → N × N, then the<br />

composition g ◦ h : N → Q + is clearly surjective; i.e., Q + is countable.<br />

Theorem 0.4. A subset <strong>of</strong> a countable set is countable.<br />

Theorem 0.5. A countable union <strong>of</strong> countable sets is countable.<br />

5


6 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Let {A α } α∈Λ be a countable collection <strong>of</strong> countable sets. Then there<br />

are surjections i : N → Λ and f α : N → A α for each α ∈ Λ. It follows that the<br />

function F : N × N → ∪ α∈Λ A α defined by<br />

is a surjection.<br />

F (m, n) = f i(m) (n)<br />

Example 0.6. Z × Z and Q are countable.<br />

Theorem 0.7. A finite product <strong>of</strong> countable sets is countable.<br />

Example 0.8. If B i is a countable set for each i ∈ N, the set ∪ n∈N<br />

countable.<br />

Ordered Sets<br />

□<br />

n∏<br />

B i is also<br />

Definition 0.9. A relation ≤ on a set A is called a preorder if it is reflexive<br />

and transitive. When ≤ is a preorder on A, we will call (A, ≤) a preordered set.<br />

Definition 0.10. A relation ≤ on a set A is called a partial order if it is<br />

reflexive, antisymmetric and transitive. When ≤ is a partial order on A, we will<br />

call (A, ≤) a partially ordered set or a poset.<br />

It is not difficult to see that a preorder ≤ on a set A induces a partial order on<br />

A/ ∼ , where ∼ is the equivalence relation on A defined by a ∼ b iff a ≤ b and b ≤ a.<br />

Theorem 0.11 (Zorn’s lemma). Let (A, ≤) be a nonempty partially ordered<br />

set. If every chain in A has an upper bound in A, then A has a maximal element.<br />

Definition 0.12. A partially ordered set (A, ≤) is said to be totally ordered if<br />

every a, b ∈ A, we have a ≤ b or b ≤ a.<br />

Definition 0.13. A totally ordered set (A, ≤) is said to be well ordered if every<br />

nonempty subset <strong>of</strong> A has a smallest element.<br />

Definition 0.14. For a well ordered set (A, ≤) and α ∈ A, the section <strong>of</strong> A by<br />

α defined to be the set S α = {x ∈ A | x < α}.<br />

Theorem 0.15. There exists a well ordered set L having the largest element<br />

Ω such that the section S Ω is uncountable but every other section is countable.<br />

Moreover, each countable subset <strong>of</strong> S Ω has an upper bound in S Ω .<br />

i=1


CHAPTER 1<br />

Basic Concepts in <strong>Topology</strong><br />

1. Spaces and Subspaces<br />

Definition 1.1. A topological space (or simply a space) consists <strong>of</strong> a set X<br />

together with a subcollection τ X <strong>of</strong> P(X) satisfying the following properties :<br />

(1) ∅ ∈ τ X ,<br />

(2) X ∈ τ X ,<br />

(3) ∪ A ∈ τ X for an arbitrary subcollection A <strong>of</strong> τ X ,<br />

(4) ∩ A ∈ τ X for a finite subcollection A <strong>of</strong> τ X .<br />

The collection τ X is called a topology on X and an element U ∈ τ X is said to<br />

be open (in X).<br />

We usually write τ instead <strong>of</strong> τ X when there is no ambiguity. Most <strong>of</strong> the time,<br />

we will assume that X ≠ ∅ and write a space X without specifying its topology τ X .<br />

The reader should keep in mind that a space X always comes with a topology on<br />

it.<br />

Although ∅ is trivially a space, we will usually assume that a space in this note<br />

is nonempty.<br />

Example 1.2. The collection <strong>of</strong> all open subsets <strong>of</strong> R forms a topology on R.<br />

Example 1.3. Let X = {a, b} (where a ≠ b). Then {∅, {a, b}}, {∅, {a}, {a, b}},<br />

{∅, {b}, {a, b}} {∅, {a}, {b}, {a, b}} are all possible topologies on X.<br />

Exercise 1.4. Find all possible topologies on the set {a, b, c} (where a, b, c are<br />

all distinct).<br />

Example 1.5. Some interesting topologies on a given set X.<br />

(1) The discrete topology : τ dis = P(X).<br />

(2) The indiscrete topology : τ indis = {∅, X}.<br />

(3) The c<strong>of</strong>inite topology : τ cf = {∅, X} ∪ {A ⊆ X | A c is finite}.<br />

(4) The cocountable topology : τ cc = {∅, X} ∪ {A ⊆ X | A c is countable}.<br />

7


8 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Exercise 1.6. In R with the c<strong>of</strong>inite topology, prove that any two nonempty<br />

open sets always intersect. Is this still true for R equipped with the cocountable<br />

topology?<br />

Exercise 1.7. For a collection {τ α } α∈Λ <strong>of</strong> topologies on a set X, show that<br />

the intersection ∩ α∈Λ τ α is still a topology on X. What about ∪ α∈Λ τ α?<br />

Definition 1.8. A subset A <strong>of</strong> a space X is said to be<br />

• closed if A c = X − A is an open set,<br />

• G δ if A is a countable intersection <strong>of</strong> open sets,<br />

• F σ if A is a countable union <strong>of</strong> closed sets, or equivalently, if A c is a G δ -set<br />

in X.<br />

Example 1.9. In R (with all usual open subsets), Q = ∪ q∈Q {q} is clearly F σ,<br />

and hence Q c is G δ . However, both <strong>of</strong> them are neither open nor closed.<br />

Definition 1.10. Let (X, τ X ) be a space, x ∈ X and A ⊆ X. A subset U <strong>of</strong><br />

X is called a neighborhood <strong>of</strong> x if x ∈ U ∈ τ X . Similarly, a subset V <strong>of</strong> X is called<br />

a neighborhood <strong>of</strong> A if A ⊆ V ∈ τ X .<br />

Definition 1.11. A space X is said to be Hausdorff if any two distinct points<br />

<strong>of</strong> X have disjoint neighborhoods.


<strong>Phichet</strong> <strong>Chaoha</strong> 9<br />

Example 1.12. The space {a, b} (where a ≠ b) equipped with the topology<br />

{∅, {a}, {a, b}} is not Hausdorff.<br />

Definition 1.13. Let d : R n × R n → R be the euclidean metric defined by<br />

∑<br />

d(x, y) = √ n (x i − y i ) 2 ,<br />

i=1<br />

for all x, y ∈ R n . The open ball centered at x ∈ R n <strong>of</strong> radius r > 0 is defined to be<br />

B(x; r) = {y ∈ R n | d(x, y) < r}.<br />

A subset U <strong>of</strong> R n is said to be open if for each x ∈ U, there exists r > 0 such that<br />

B(x; r) ⊆ U.<br />

It is straightforward to verify that the collection <strong>of</strong> all such open sets (notice that<br />

both ∅ and R n are open) forms a topology on R n . This topology is called the usual<br />

topology on R n . Note also that R n with the usual topology is Hausdorff.


10 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Exercise 1.14. Prove that a closed [open] set in R n (with the usual topology)<br />

is a G δ -set [F σ -set].<br />

Definition 1.15. Let τ and τ ′ be two topologies on a set X. We say that τ ′<br />

is finer than τ (or equivalently, τ is coarser than τ ′ ) if τ ⊆ τ ′ .<br />

Example 1.16. τ dis is finer than τ indis . What can we say about τ cf and τ cc ?<br />

Exercise 1.17. Let X be a space and Y ⊆ X.<br />

{U ∩ Y | U ∈ τ X } forms a topology on Y .<br />

Prove that the collection<br />

Definition 1.18. Let X be a space and Y ⊆ X. The subspace topology <strong>of</strong> Y<br />

(in X) is defined to be<br />

τ Y = {U ∩ Y | U ∈ τ X }.<br />

In this case, we say that (Y, τ Y ) is a subspace <strong>of</strong> X.<br />

Example 1.19. The subspace topology <strong>of</strong> Z (in R) is discrete.<br />

Exercise 1.20. Let Y be a subspace <strong>of</strong> X and A ⊆ Y . Prove that A is closed<br />

in Y iff A = C ∩ Y for some closed set C in X.<br />

Theorem 1.21. Let Y be an open [closed] subspace <strong>of</strong> X. If A is open [closed]<br />

in Y , then A is also open [closed] in X.<br />

Pro<strong>of</strong>. Let A be open [closed] in Y . Then A = B ∩ Y for some open [closed]<br />

subset B <strong>of</strong> X. Since Y itself is open [closed] in X and the interesection is finite,<br />

it follows that A is also open [closed] in X.<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 11<br />

2. Basis for a <strong>Topology</strong><br />

Definition 1.22. For a set X and B ⊆ P(X), we call B a basis for a topology<br />

on X if it satisfies the following conditions :<br />

(1) For each x ∈ X, there exists B x ∈ B such that x ∈ B x .<br />

(2) For each B 1 , B 2 ∈ B and x ∈ B 1 ∩ B 2 , there exists B 3 ∈ B such that<br />

x ∈ B 3 ⊆ B 1 ∩ B 2 .<br />

Exercise 1.23. Let B be a basis for a topology on a set X. Prove that X = ∪ B<br />

and {U ⊆ X | ∀x ∈ U∃B x ∈ B, x ∈ B x ⊆ U} forms a topology on X.<br />

Definition 1.24. Let B be a basis for a topology on a set X. We define the<br />

topology generated by B to be<br />

Clearly, B ⊆< B >.<br />

< B >= {U ⊆ X | ∀x ∈ U∃B x ∈ B, x ∈ B x ⊆ U}.<br />

Lemma 1.25. Let B be a basis for a topology on a set X. Every element <strong>of</strong><br />

< B > is a union <strong>of</strong> elements <strong>of</strong> B, and < B > is the coarsest (smallest) topology<br />

on X containing B.<br />

Pro<strong>of</strong>. Let U ∈< B >. For each x ∈ U, let B x ∈ B be chosen from the the<br />

definition above so that x ∈ B x ⊆ U. Then U = ∪ x∈U B x; i.e., U is a union <strong>of</strong><br />

elements <strong>of</strong> B. Moreover, if τ is a topology on X such that B ⊆ τ, then a union <strong>of</strong><br />

elements <strong>of</strong> B is clearly in τ. Therefore, < B >⊆ τ as desired.<br />

□<br />

Remark 1.26. The previous lemma implies that τ =< τ > for any topology τ<br />

on X. Hence, U ∈ τ iff for each x ∈ U, there exists U x ∈ τ such that x ∈ U x ⊆ U.<br />

Also, if B ⊆ B ′ , we clearly have < B >⊆< B ′ >. The converse is not true, but we<br />

have the following theorem instead.<br />

Theorem 1.27. Let B and B ′ be bases for topologies on a set X. We have<br />

< B >⊆< B ′ > iff for each B ∈ B and x ∈ B, there exists B ′ ∈ B ′ such that<br />

x ∈ B ′ ⊆ B.


12 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. (⇒) Trivial.<br />

(⇐) Let U ∈< B > and x ∈ U. Then there is B ∈ B such that x ∈ B ⊆ U. By<br />

assumption, we can find B ′ ∈ B ′ such that x ∈ B ′ ⊆ B. Therefore, x ∈ B ′ ⊆ U;<br />

i.e., U ∈< B ′ >.<br />

□<br />

Corollary 1.28. Let (X, τ) be a space and a basis B such that B ⊆ τ. Then<br />

τ is generated by B iff for each U ∈ τ and x ∈ U, there exists B ∈ B such that<br />

x ∈ B ⊆ U.<br />

Example 1.29. The usual topology on R n is generated by the collection <strong>of</strong><br />

open balls {B(x; r) | x ∈ R n , r > 0}. In particular, the usual topology on R is<br />

generated by {(a, b) | a < b}.<br />

Definition 1.30. The lower limit topology on R is the topology generated by<br />

{[a, b)|a < b}. We simply write R l to denote R equipped with the lower limit<br />

topology.<br />

Exercise 1.31. Prove that the lower limit topology on R is strictly finer than<br />

the usual topology.<br />

Definition 1.32. Let (X, ≤) be a totally ordered set with |X| > 1. An interval<br />

in X is one <strong>of</strong> the following sets :<br />

(a, b) = {x ∈ X | a < x < b}<br />

[a, b) = {x ∈ X | a ≤ x < b}<br />

(a, b] = {x ∈ X | a < x ≤ b}<br />

[a, b] = {x ∈ X | a ≤ x ≤ b}<br />

The order topology on X is the topology generated by the basis consisting <strong>of</strong> all<br />

intervals <strong>of</strong> the following forms :<br />

• (a, b),<br />

• [x min , b) if X has the smallest element x min ,<br />

• (a, x max ] if X has the largest element x max .<br />

Exercise 1.33. Consider the set R 2 = R × R whose element is denoted by<br />

a × b, and the dictionary order (≼) on R 2 defined by<br />

a 1 × b 1 ≼ a 2 × b 2 iff (a 1 × b 1 = a 2 × b 2 ) or (a 1 × b 1 ≺ a 2 × b 2 ),<br />

where a 1 × b 1 ≺ a 2 × b 2 iff a 1 < a 2 or (a 1 = a 2 and b 1 < b 2 ).<br />

Is the dictionary order topology on R × R the same as the usual topology on<br />

R 2 ? If not, compare them.


<strong>Phichet</strong> <strong>Chaoha</strong> 13<br />

Exercise 1.34. For a set X and S ⊆ P(X) with ∪ S = X, prove that the<br />

collection <strong>of</strong> all finite intersections <strong>of</strong> elements <strong>of</strong> S forms a basis for a topology on<br />

X.<br />

Definition 1.35. For a set X and S ⊆ P(X), we will call S a subbasis for a<br />

topology on X if ∪ S = X.<br />

If S is a subbasis for a topology on X, we will define the topology generated<br />

by S (as a subbasis), denoted by < S >, to be the collection <strong>of</strong> all unions <strong>of</strong> finite<br />

intersections <strong>of</strong> elements <strong>of</strong> S.<br />

Exercise 1.36. Let (X, ≤) be a totally ordered set with |X| > 1 and for each<br />

a ∈ X, let S a = {x ∈ X | x < a} and T a = {x ∈ X | x > a}.<br />

Prove that the collection S = {S a | a ∈ X} ∪ {T a | a ∈ X} forms a subbasis for the<br />

order topology on X.


14 <strong>Topology</strong> (<strong>2301631</strong>)<br />

3. Limit Points<br />

Let X be a (nonempty) space and ∅ ̸= A ⊆ X.<br />

Definition 1.37. A point x ∈ X is called a limit point <strong>of</strong> A if, for each<br />

neighborhood U <strong>of</strong> x, we have U ∩ A − {x} ̸= ∅.<br />

The set <strong>of</strong> all limit points <strong>of</strong> A is called the derived set <strong>of</strong> A and denoted by A ′ . If<br />

A = A ′ , we will call A a perfect set.<br />

Example 1.38. In R (with the usual topology), we have (0, 1] ′ = [0, 1], Q ′ =<br />

R, { 1 n | n > 0}′ = {0}, and N ′ = ∅.<br />

Exercise 1.39. Let (L, ≤) be a well ordered set having the largest element Ω<br />

together with the following properties :<br />

(1) the section S Ω is uncountable,<br />

(2) the section S α is countable for any α < Ω,<br />

(3) each countable subset <strong>of</strong> S Ω has an upper bound in S Ω .<br />

In L equipped with the order topology, prove that Ω is a limit point <strong>of</strong> S Ω .<br />

Definition 1.40. The interior <strong>of</strong> A (in X), denoted by Int X (A) or A ◦ , is the<br />

largest open subset <strong>of</strong> X contained in A, or equivalently<br />

Int X (A) = ∪ {U ⊆ X | U ⊆ A and U is open}.<br />

Definition 1.41. The closure <strong>of</strong> A (in X), denoted by Cl X (A) or A, is the<br />

smallest closed subset <strong>of</strong> X containing A, or equivalently<br />

Cl X (A) = ∩ {C ⊆ X | A ⊆ C and C is closed}.<br />

Definition 1.42. The boundary (or frontier) <strong>of</strong> A is defined to be<br />

∂A = A ∩ (X − A).<br />

Remark 1.43. From the definition above, we clearly have<br />

• A is open in X iff A = Int X (A).<br />

• A is closed in X iff A = Cl X (A).<br />

Example 1.44. In R (with the usual topology), we clearly have (0, 1] ◦ = (0, 1),<br />

(0, 1] = [0, 1] and ∂(0, 1] = {0, 1}.


<strong>Phichet</strong> <strong>Chaoha</strong> 15<br />

{ 1 n<br />

Exercise 1.45. What are the interior, the closure and the boundary <strong>of</strong> Q and<br />

| n ∈ N}? Justify your answers.<br />

Exercise 1.46. Let X be a space and A ⊆ X. Are the following pairs <strong>of</strong> sets<br />

comparable? If so, compare them.<br />

(1) (A ◦ ) and (A) ◦ .<br />

(2) (∂A) and ∂(A).<br />

(3) ∂(A ◦ ) and (∂A) ◦ .<br />

Theorem 1.47. For a subspace Y <strong>of</strong> X such that A ⊆ Y , we have<br />

Cl Y (A) = Cl X (A) ∩ Y.<br />

Pro<strong>of</strong>. (⊆) Since Cl X (A) is closed in X, then Cl X (A) ∩ Y is closed in Y .<br />

Since A ⊆ Cl X (A) ∩ Y , it follows that Cl Y (A) ⊆ Cl X (A) ∩ Y .<br />

(⊇) Since Cl Y (A) is closed in Y , we have Cl Y (A) = C ∩ Y for some closed<br />

subset C <strong>of</strong> X containing A. Therefore, Cl X (A) ⊆ C, and hence Cl X (A) ∩ Y ⊆<br />

C ∩ Y = Cl Y (A).<br />

□<br />

Example 1.48. Consider Y = (0, 2] as a subspace <strong>of</strong> R. Then,<br />

Cl Y ((0, 1]) = Cl R ((0, 1]) ∩ Y = [0, 1] ∩ (0, 2] = (0, 1]<br />

which is closed in Y , but not in R.<br />

Theorem 1.49. Let X be a space, A ⊆ X and x ∈ X. Then<br />

x ∈ A iff U ∩ A ≠ ∅ for each neighborhood U <strong>of</strong> x.<br />

Pro<strong>of</strong>. We will show that x /∈ A iff U ∩ A = ∅ for some neighborhood U <strong>of</strong> x.<br />

(⇒) Suppose x /∈ A and let U = X − A. Then, it is easy to see that U is a<br />

neighborhood <strong>of</strong> x such that U ∩ A = ∅ as desired.<br />

(⇐) Suppose there is a neighborhood U <strong>of</strong> x such that U ∩ A = ∅. Then<br />

C = X − U is closed, A ⊆ C and x /∈ C. Since A ⊆ C, it follows that x /∈ A. □<br />

Remark 1.50. If the topology on X is generated by a basis B, we clearly have<br />

the followings :<br />

• x ∈ A ′ iff B ∩ A − {x} ≠ ∅ for each B ∈ B containing x.<br />

• x ∈ A iff B ∩ A ≠ ∅ for each B ∈ B containing x.<br />

Exercise 1.51. Let X be a space and A ⊆ X. Prove that A = A ∪ A ′ .<br />

Remark 1.52. From Exercise 1.39 and the previous exercise, we may represent<br />

L by S Ω .<br />

Corollary 1.53. Let X be a space and A ⊆ X. Then A is closed iff A ′ ⊆ A.<br />

Pro<strong>of</strong>. A is closed ⇔ A = A ⇔ A ′ ⊆ A.<br />

Exercise 1.54. Let A and B be subsets <strong>of</strong> a space X. Prove that<br />

(1) A ◦ ∩ ∂A = ∅.<br />

(2) A ◦ ∪ ∂A = A.<br />

(3) A = A.<br />

(4) A ⊆ B ⇒ A ⊆ B.<br />

(5) A ∪ B = A ∪ B.<br />

(6) A ∩ B ⊆ A ∩ B. When does the other containment fail?<br />


16 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Definition 1.55. Let X be a space and A ⊆ X. We say that A is dense (in<br />

X) if A = X. On the other hand, A is nowhere dense if X − A is dense in X.<br />

Example 1.56. In R, Q is dense while Z is nowhere dense.<br />

Definition 1.57. Subsets A and B <strong>of</strong> a space X are said to be separated if<br />

A ∩ B = A ∩ B = ∅.<br />

Example 1.58. In R, the intervals (0, 1) and [1, 2) are disjoint but not separated,<br />

while (0, 1) and (1, 2) are separated.


<strong>Phichet</strong> <strong>Chaoha</strong> 17<br />

Let X and Y be (nonempty) spaces.<br />

4. Continuous Maps<br />

Definition 1.59. A map f : X → Y is said to be continuous at x ∈ X if, for<br />

any neighborhood V <strong>of</strong> f(x), there is a neighborhood U <strong>of</strong> x such that f(U) ⊆ V .<br />

We simply say that f is continuous if it is continuous at each x ∈ X.<br />

Example 1.60. For a selfmap <strong>of</strong> R n (with the usual topology), the definition<br />

<strong>of</strong> continuity above is exactly the ε − δ definition in analysis.<br />

Theorem 1.61. A map f : X → Y is continuous iff f −1 (V ) is open in X for<br />

each open subset V <strong>of</strong> Y .<br />

Pro<strong>of</strong>. (⇒) Let V ⊆ Y be an open set and x ∈ f −1 (V ). Since f is continuous<br />

at x and V is a neighborhood <strong>of</strong> f(x), there is a neighborhood U <strong>of</strong> x such that<br />

f(U) ⊆ V . Hence, U ⊆ f −1 (f(U)) ⊆ f −1 (V ) which implies that f −1 (V ) is open in<br />

X.<br />

(⇐) Let x ∈ X and V a neighborhood <strong>of</strong> f(x). It follows from the assumption<br />

that f −1 (V ) is a neighborhood <strong>of</strong> x. So, by letting U = f −1 (V ), we clearly have<br />

f(U) = f(f −1 (V )) ⊆ V and this proves the continuity <strong>of</strong> f.<br />


18 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Remark 1.62. Definition 1.59 is generally known as the local definition <strong>of</strong><br />

continuity while Theorem 1.61 is usually taken to be the global definition. Moreover,<br />

using the global definition, one can easily prove that f : X → Y is continuous iff<br />

f −1 (C) is closed in X for each closed subset C <strong>of</strong> Y .<br />

Example 1.63. Let τ and τ ′ be topologies on X. It is easy to verify that the<br />

identity map 1 X : (X, τ) → (X, τ ′ ) is always continuous iff τ is finer than τ ′ .<br />

Remark 1.64. If the topology <strong>of</strong> Y is generated by a basis B, it is easy to see<br />

that a map f : X → Y is continuous iff f −1 (B) is open for each B ∈ B. This is<br />

because f −1 ( ∪ α B α) = ∪ α f −1 (B α ). Note also that f −1 (B) may not be a basis<br />

element <strong>of</strong> X.<br />

Exercise 1.65. If the topology <strong>of</strong> Y is generated by a subbasis S, prove that<br />

a map f : X → Y is continuous iff f −1 (S) is open in X for each S ∈ S.<br />

Theorem 1.66. A map f : X → Y is continuous iff f(A) ⊆ f(A) for any<br />

subset A <strong>of</strong> X.<br />

Pro<strong>of</strong>. (⇒) Let A ⊆ X, y ∈ f(A) and V be a neighborhood <strong>of</strong> y. Then, there<br />

exists x ∈ A such that f(x) = y. Hence, x ∈ f −1 (V ). By the continuity <strong>of</strong> f, it<br />

follows that f −1 (V ) is a neighborhood <strong>of</strong> x. Therefore, f −1 (V ) ∩ A ≠ ∅ and hence<br />

∅ ̸= f(f −1 (V ) ∩ A) ⊆ f(f −1 (V )) ∩ f(A) ⊆ V ∩ f(A).<br />

This implies y ∈ f(A).<br />

(⇐) Let C be a closed subset <strong>of</strong> Y and A = f −1 (C). Since<br />

f(A) ⊆ f(A) = f(f −1 (C)) ⊆ C = C,<br />

then we have A ⊆ f −1 (f(A)) ⊆ f −1 (C) = A.<br />

desired.<br />

Therefore, A is closed in X as<br />

□<br />

Example 1.67. It is easy to verify that a constant map, an inclusion and a<br />

restriction <strong>of</strong> a continuous map are continuous.<br />

Theorem 1.68. If f : X → Y and g : Y → Z are continuous maps, then so is<br />

g ◦ f.<br />

Pro<strong>of</strong>. Let W be an open subset <strong>of</strong> Z.<br />

Then g −1 (W ) is open in Y by the continuity <strong>of</strong> g. It follows that (g ◦ f) −1 (W ) =<br />

f −1 (g −1 (W )) is open in X by the continuity <strong>of</strong> f. This proves the continuity <strong>of</strong><br />

g ◦ f as desired.<br />

□<br />

Exercise 1.69. Let X be a space and A ⊆ X. Prove that the subspace<br />

topology on A is the smallest (coarsest) topology that makes the inclusion i : A ↩→<br />

X continuous.


<strong>Phichet</strong> <strong>Chaoha</strong> 19<br />

Exercise 1.70. Let f : X → Y be a map. If X = ∪ α U α where each U α is<br />

open in X and f| Uα is continuous, prove that f is continuous. Is it still true if we<br />

do not assume that each U α is open?<br />

Now, suppose we have two maps f : A → Y and g : B → Y that agree on the<br />

intersection A ∩ B. We clearly have a well-defined map f ∪ g : A ∪ B → Y given by<br />

{<br />

f(x) if x ∈ A,<br />

f ∪ g(x) =<br />

g(x) if x ∈ B.<br />

Notice that the previous exercise already guarantees the continuity <strong>of</strong> f ∪ g<br />

whenever both A and B are open subspaces <strong>of</strong> X. The next theorem also gives the<br />

similar result when X is a union <strong>of</strong> two closed subspaces.<br />

Theorem 1.71 (Pasting Lemma). Let X = A ∪ B, where A and B are closed<br />

in X. If f : A → Y and g : B → Y are continuous maps that agree on A ∩ B, then<br />

h = f ∪ g : A ∪ B → Y is continuous.<br />

Pro<strong>of</strong>. Let C be a closed subset <strong>of</strong> Y . Since<br />

h −1 (C) = f −1 (C) ∪ g −1 (C),<br />

the continuity <strong>of</strong> f and g implies that the sets f −1 (C) and g −1 (C) are closed in<br />

A and B, respectively. Also, since A and B are closed in X, so are f −1 (C) and<br />

g −1 (C). Therefore, h −1 (C) = f −1 (C) ∪ g −1 (C) is also closed in X.<br />

□<br />

The pasting lemma can also be extended to a finite union <strong>of</strong> closed subspaces<br />

<strong>of</strong> X, but for an infinite union, we have the following counterexample.<br />

Example 1.72. For each r ∈ R, let f r : {r} → R be defined by<br />

{<br />

1 ; r ∈ Q<br />

f r (r) =<br />

0 ; r /∈ Q.<br />

Clearly, each f r is continuous and f = ∪ r∈R f r is well-defined. It is easy to see that<br />

f is the characteristic map on Q. Therefore, f is not continuous.<br />

Definition 1.73. A map f : X → Y is called a homeomorphism if f is a<br />

continuous bijection whose inverse map is also continuous. We will say that X is<br />

homeomorphic to Y , written X ∼ = Y , if there is a homeomorphism from X to Y .<br />

Remark 1.74. It is easy to verify that being homeomorphic is an equivalence<br />

relation.<br />

Definition 1.75. We say that a property P <strong>of</strong> a space is a topological property<br />

or a topological invariant if P is invariant under homeomorphisms.<br />

Example 1.76. For topologies τ and τ ′ on a set X such that τ is strictly finer<br />

than τ ′ , the identity map 1 X : (X, τ) → (X, τ ′ ) is a continuous bijection, but not a<br />

homeomorphism!<br />

Definition 1.77. A continuous injection f : X → Y is called an imbedding if<br />

the induced bijection f : X → f(X) is a homeomorphism, where f(X) is considered<br />

as a subspace <strong>of</strong> Y .


20 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Example 1.78. Let S 1 = {(x, y) | x 2 +y 2 = 1} be a subspace <strong>of</strong> R 2 . The continuous<br />

bijection f : [0, 1) → R 2 given by f(t) = (cos 2πt, sin 2πt) is not an imbedding<br />

because f −1 : S 1 → [0, 1) is not continuous. Notice that (f −1 ) −1 ([0, 1 2 )) = f([0, 1 2 ))<br />

is not open in S 1 . However, f| (0,1) is an imbedding.<br />

Exercise 1.79. Prove that f : N ∪ {0} → R given by f(n) = (−1) n n 2 is an<br />

imbedding. What about g : N ∪ {0} → R given by g(0) = 0 and g(n) = 1 n<br />

for<br />

n > 0?


<strong>Phichet</strong> <strong>Chaoha</strong> 21<br />

5. Sequences and Nets<br />

Definition 1.80. A sequence in a space X is a map (x n ) : N → X. We usually<br />

denote (x n )(i) by x i .<br />

Definition 1.81. The sequence (x n ) in X is said to converge to x ∈ X, abbreviated<br />

by (x n ) → x, if for each neighborhood U <strong>of</strong> x, there exists N ∈ N such<br />

that x n ∈ U for all n ≥ N. When (x n ) → x, we will call x a limit <strong>of</strong> (x n ).<br />

Example 1.82. In the space {a, b}, where a ≠ b, with the indiscrete topology,<br />

the constant sequence (a) converges to both a and b. Hence, a limit <strong>of</strong> a convergent<br />

sequence may not be unique.<br />

Theorem 1.83. A convergent sequence in a Hausdorff space has a unique limit.<br />

Pro<strong>of</strong>. Let (x n ) be a convergent sequence in X. Suppose (x n ) → x and<br />

(x n ) → x ′ . If x ≠ x ′ , there exist disjoint neighborhoods U and V <strong>of</strong> x and x ′ ,<br />

respectively. Hence, there is a large enough N ∈ N such that x n ∈ U ∩ V for all<br />

n ≥ N (can you see why?). This is impossible because U ∩ V = ∅. Hence, we must<br />

have x = x ′ .<br />

□<br />

Definition 1.84. Let (x n ) and (y n ) be sequences in a space X. We say that<br />

(y n ) is a subsequence <strong>of</strong> (x n ) if there is a strictly increasing map f : N → N such<br />

that y i = x f(i) for all i ∈ N.<br />

Exercise 1.85. Let (y n ) be a subsequence <strong>of</strong> a convergent sequence (x n ) in a<br />

Hausdorff space. Prove that if (y n ) converges to y ∞ ∈ X, then so is (x n ). What<br />

happens if (x n ) is not assumed to be convergent?<br />

Theorem 1.86 (Sequence Lemma). Let X be a space, x ∈ X and A ⊆ X. If<br />

there is a sequence in A converging to x, then x ∈ A.<br />

Pro<strong>of</strong>. Let (a n ) be a sequence in A converging to x, and U a neighborhood<br />

<strong>of</strong> x. Then, by the definition <strong>of</strong> convergence, there exists N ∈ N such that a n ∈ U<br />

for all n ≥ N. In particular, a N ∈ U ∩ A. Hence, U ∩ A ≠ ∅; i.e, x ∈ A. □<br />

The converse <strong>of</strong> the above theorem is not true in general as we will see in the<br />

next example.<br />

Example 1.87. The space S Ω does not satisfy the converse <strong>of</strong> the sequence<br />

lemma because there is no sequence in S Ω converging to Ω. To see this, let (x n ) be<br />

a sequence in S Ω . Since C = {x n | n ∈ N} is a countable subset <strong>of</strong> S Ω , it has an<br />

upper bound in S Ω , says α. Now it is easy to see that C ∩ (α, Ω] = ∅ and hence<br />

(x n ) does not converge to Ω.


22 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Definition 1.88. A directed set is a preordered set (I, ≤) such that for any<br />

α, β ∈ I, we have α ≤ γ and β ≤ γ for some γ ∈ I.<br />

Example 1.89. Clearly, (P(S), ⊆) and (N, |) are directed sets. Moreover, if<br />

N (x) denotes the set <strong>of</strong> all neighborhoods <strong>of</strong> x ∈ X, both (N (x), ⊆) and (N (x), ⊇)<br />

are also directed sets.<br />

Definition 1.90. A net in a space X is simply a map (x α ) : I → X where I<br />

is a directed set.<br />

Example 1.91. A sequence in a space X is simply a net in X whose directed<br />

set is (N, ≤).<br />

Definition 1.92. A net (x α ) : I → X is said to converge to x ∈ X, abbreviated<br />

by (x α ) → x, if for each neighborhood U <strong>of</strong> x, there exists γ ∈ I such that x α ∈ U<br />

for all α ≥ γ.<br />

Exercise 1.93. Prove that a convergent net in a Hausdorff space has a unique<br />

limit.<br />

Theorem 1.94. Let X be a space, x ∈ X and A ⊆ X. There is a net in A<br />

converging to x iff x ∈ A.<br />

Pro<strong>of</strong>. (⇒) Let (a α ) : (I, ≤) → A be a net in A converging to x, and U a<br />

neighborhood <strong>of</strong> x. Then, by the definition <strong>of</strong> convergence, there exists γ ∈ I such<br />

that a α ∈ U for all α ≥ γ. In particular, a γ ∈ U ∩ A. Hence, U ∩ A ≠ ∅; i.e, x ∈ A.<br />

(⇐) Suppose x ∈ A. Let I be the poset <strong>of</strong> neighborhoods <strong>of</strong> x partially ordered<br />

by inverse inclusions; i.e., U ≤ V iff U ⊇ V . Then for each U ∈ I, we have U∩A ≠ ∅.<br />

By the axiom <strong>of</strong> choice, we can form a net (a U ) in A such that a U ∈ U ∩ A for each<br />

U ∈ I. Now, it is clear from the construction that (a U ) → x.<br />

□<br />

Exercise 1.95. Let X and Y be spaces, x ∈ X and f : X → Y a map.<br />

(1) Prove that if f is continuous at x, then for each sequence (x n ) in X<br />

converging to x, the sequence (f(x n )) converges to f(x).<br />

(2) Show that the converse <strong>of</strong> the above statement fails in general.<br />

(3) Prove that a map f : X → Y is continuous at x iff for each net (x α ) in X<br />

converging to x, the net (f(x α )) converges to f(x).<br />

Definition 1.96. A subset I ′ <strong>of</strong> a directed set I is said to be c<strong>of</strong>inal in I if<br />

for each α ∈ I, we have α ≤ α ′ for some α ′ ∈ I ′ .<br />

Proposition 1.97. A c<strong>of</strong>inal subset <strong>of</strong> a directed set is also a directed set.<br />

Pro<strong>of</strong>. Suppose I ′ be a c<strong>of</strong>inal subset <strong>of</strong> a directed set I. Let α ′ , β ′ ∈ I ′ .<br />

Since I ′ ⊆ I and I is directed, we clearly have α ′ ≤ γ and β ′ ≤ γ for some γ ∈ I.<br />

Since I ′ is c<strong>of</strong>inal in I, there exists γ ′ ∈ I ′ such that γ ≤ γ ′ . Hence, α ′ ≤ γ ′ and<br />

β ′ ≤ γ ′ as required.<br />

□<br />

Definition 1.98. A net (y β ) : J → X is said to be a subnet <strong>of</strong> a net (x α ) :<br />

I → X if (y β ) = (x α ) ◦ f for some order-preserving map f : J → I where f(J ) is<br />

c<strong>of</strong>inal in I.<br />

Example 1.99. A subsequence is clearly a subnet.<br />

Exercise 1.100. Let X be a Hausdorff space, (x α ) : I → X a convergent net<br />

in X and (y β ) : J → X a subnet <strong>of</strong> (x α ). Prove that if (y β ) converges to a point<br />

z ∈ X, then so does (x α ).


<strong>Phichet</strong> <strong>Chaoha</strong> 23<br />

6. Product <strong>Topology</strong><br />

Let Λ be an index set. In this section, we will introduce some topologies on a<br />

cartesian product ∏ α∈Λ X α, where each X α is a topological space. Conventionally,<br />

an element <strong>of</strong> ∏ α∈Λ X α is written as (x α ) α∈Λ , and for a collection {f α : A →<br />

X α } α∈Λ <strong>of</strong> maps, (f α ) α∈Λ denotes the map f : A → ∏ α∈Λ X α defined by f(a) =<br />

(f α (a)) α∈Λ for each a ∈ A.<br />

Definition 1.101. Let {X α } α∈Λ be a collection <strong>of</strong> spaces. We define the box<br />

topology on the set ∏ α∈Λ X α to be the topology generated by the basis<br />

{ ∏ α∈Λ<br />

U α | U α is open in X α }.<br />

We also define the product topology on the set ∏ α∈Λ X α to be the topology<br />

generated by the basis<br />

{ ∏ α∈Λ<br />

U α | U α is open in X α and U α ≠ X α for finitely many α’s}.<br />

The set ∏ α∈Λ X α together with the product topology will be simply called the<br />

product space.<br />

Example 1.102. In R ω = ∏ n∈N R, the set ∏ n∈N<br />

(−n, n) is open in the box<br />

topology, but not in the product topology.<br />

Remark 1.103. The box topology is generally finer than the product topology.<br />

However, when Λ is finite, the box and the product topologies are the same.<br />

Exercise 1.104. Let A and B be subsets <strong>of</strong> a space X. Prove that A × B =<br />

A × B in the product topology.<br />

The pro<strong>of</strong>s <strong>of</strong> the following two theorems are straightforward, and hence left to<br />

reader.<br />

Theorem 1.105. Suppose for each α ∈ Λ, the topology on X α is generated by<br />

a basis B α . Then the collection { ∏ α∈Λ B α | B α ∈ B α } forms a basis for the box<br />

topology on the set ∏ α∈Λ X α.<br />

Similarly, the collection<br />

{ ∏ α∈Λ<br />

B α | B α ∈ B α ∪ {X α } and B α ≠ X α for finitely many α’s}<br />

forms a basis for the product topology on the set ∏ α∈Λ X α.<br />

Theorem 1.106. If A α is a subspace <strong>of</strong> X α for each α ∈ Λ, then ∏ α∈Λ A α is<br />

also a subspace <strong>of</strong> ∏ α∈Λ X α when both products are given the box or the product<br />

topology.<br />

Remark 1.107. For simplicity, we sometimes write U α1 ×· · ·×U αn × ∏ α≠α 1 ,...,α n<br />

X α<br />

to represent the subset ∏ α∈Λ U α <strong>of</strong> ∏ α∈Λ X α such that U α = X α for all α /∈<br />

{α 1 , . . . , α n }.<br />

Theorem 1.108. For each β ∈ Λ, the projection map<br />

π β : ∏ α∈Λ<br />

X α → X β<br />

is continuous when the product is given the box or the product topology.


24 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. First note that it suffices to prove only the case where the product<br />

topology is given because the box topology is finer. Let β ∈ Λ and U β an open<br />

subset <strong>of</strong> X β . Since π −1<br />

β<br />

(U β) = U β × ∏ α≠β X α which is clearly open in the product<br />

topology, then π β is continuous as desired.<br />

□<br />

Exercise 1.109. Show that the product topology on ∏ α∈Λ X α is the coarsest<br />

topology that makes the projection map π β : ∏ α∈Λ X α → X β continuous for each<br />

β ∈ Λ. In other words, the collection<br />

∪<br />

{π −1<br />

β (U β) | U β is open in X β }<br />

β∈Λ<br />

forms a subbasis for the product topology on the set ∏ α∈Λ X α.<br />

Exercise 1.110. If X α is Hausdorff for each α ∈ Λ, then ∏ α∈Λ X α is also<br />

Hausdorff in the box and the product topologies.<br />

Theorem 1.111. Let ∏ α∈Λ X α be given the product topology. The map f =<br />

(f α ) α∈Λ : A → ∏ α∈Λ X α is continuous iff f α : A → X α is continuous for each<br />

α ∈ Λ.<br />

Pro<strong>of</strong>. (⇒) Trivial by composing f with the projection maps.<br />

(⇐) Let B be a basis element in ∏ α∈Λ X α with respect to the product topology;<br />

i.e., B = U α1 × · · · × U αn × ∏ α≠α 1 ,...,α n<br />

X α for some open sets U αi ⊆ X αi . Clearly,<br />

f −1 (B) = ∩ n<br />

i=1 f α −1<br />

i<br />

(U αi ) and each fα −1<br />

i<br />

(U αi ) is open by assumption. Therefore,<br />

f −1 (B) is open.<br />

□<br />

Example 1.112. The previous theorem is not true if the topology on ∏ α∈Λ X α<br />

is taken to be the box topology. Consider R ω = ∏ n∈N R. Then ∆ : R → Rω defined<br />

by ∆(x) = (x, x, . . . ) is clearly continuous (being continuous in each component)<br />

when R ω is given the product topology. However, ∆ is not continuous when R ω is<br />

given the box topology because the set<br />

(<br />

∆ −1 (−1, 1) × (− 1 2 , 1 2 ) × (−1 3 , 1 ) ∞∩<br />

3 ) × . . . = (− 1 n , 1 n ) = {0}<br />

is not open in R.<br />

Exercise 1.113. For a sequence (x n ) in the product space ∏ α∈Λ X α, prove<br />

that (x n ) converges to x iff the sequence (π α (x n )) converges to π α (x) for all α ∈ Λ.<br />

We end this subsection by stating the universal property <strong>of</strong> product spaces.<br />

∏<br />

Theorem 1.114. Let {X α } α∈Λ be a collection <strong>of</strong> spaces. The product space<br />

α∈Λ X α together with the collection {π β : ∏ α∈Λ X α → X β } β∈Λ <strong>of</strong> projection<br />

maps satisfies the following universal property :<br />

Given a space A with a collection {p β : A → X β } β∈Λ <strong>of</strong> continuous maps, there<br />

exists a unique continuous map f : A → ∏ α∈Λ X α such that p β = π β ◦ f for each<br />

β ∈ Λ; i.e., we have the following commutative diagram for each β ∈ Λ :<br />

A<br />

♣ ♣ ♣ ♣ ♣ ♣<br />

p β<br />

✲<br />

X β<br />

n=1<br />

∃!f<br />

❘<br />

∏<br />

α∈Λ<br />

X α<br />

✒ π β


<strong>Phichet</strong> <strong>Chaoha</strong> 25<br />

Moreover, if P is a space together with a collection {π ′ β : P → X β} β∈Λ satisfying<br />

the above universal property, then P must be homeomorphic to the product<br />

space ∏ α∈Λ X α.<br />

Pro<strong>of</strong>. Let A be a space and {p β : A → X β } β∈Λ a collection <strong>of</strong> continuous<br />

maps. Let f = (p α ) α∈Λ : A → ∏ α∈Λ X α and β ∈ Λ. Clearly, p β = π β ◦ f. Since<br />

each p β is continuous, then f is continuous by the previous theorem. Now, suppose<br />

f ′ : A → ∏ α∈Λ X α has the property that p β = π β ◦ f ′ for each β ∈ Λ. Then,<br />

f = (π β ◦ f) β∈Λ = (p β ) β∈Λ = (π β ◦ f ′ ) β∈Λ = f ′<br />

as desired.<br />

Finally, suppose P together a the collection {π<br />

β ′ : P → X β} β∈Λ satisfies the<br />

above universal property. Then, there exist continuous maps f : P → ∏ α∈Λ X α<br />

and g : ∏ α∈Λ X α → P such that π<br />

β ′ = π β ◦ f and π β = π<br />

β ′ ◦ g. It follows that<br />

π ′ β = π′ β ◦ (g ◦ f) and π β = π β ◦ (f ◦ g). Again, since ∏ α∈Λ X α satisfies the universal<br />

property, the identity map id : ∏ α∈Λ X α → ∏ α∈Λ X α must be the only continuous<br />

map satisfying π β = π β ◦ id. It follows that f ◦ g = id. Similarly, g ◦ f must be the<br />

identity map on P . Therefore, f is a homeomorphism whose inverse is g. □


26 <strong>Topology</strong> (<strong>2301631</strong>)<br />

7. Quotient <strong>Topology</strong><br />

Definition 1.115. Let X and Y be spaces. A surjective map p : X → Y is<br />

called a quotient map (or an identification) if for each V ⊆ Y , V is open in Y iff<br />

p −1 (V ) is open in X.<br />

Remark 1.116. Every quotient map is always continuous, but not vice versa.<br />

Note also that, the condition in the definition above is equivalent to ”C is closed in<br />

Y iff p −1 (C) is closed in X”, but NOT to ”U is open in X iff p(U) is open in Y ”.<br />

Definition 1.117. Let X and Y be spaces. A map f : X → Y is said to be<br />

open if for each open subset U <strong>of</strong> X, the set f(U) is open in Y . A closed map is<br />

defined in the similar manner.<br />

Theorem 1.118. Every continuous surjective map that is either open or closed<br />

is a quotient map.<br />

Pro<strong>of</strong>. Let f : X → Y be a continuous surjective map that is open, and<br />

V ⊆ Y . By the continuity <strong>of</strong> f, if V is open in Y , then f −1 (V ) is open in X.<br />

Conversely, assume that f −1 (V ) is open in X. Since f is open and surjective,<br />

then V = f(f −1 (V )) is open in Y . The case where f is closed can be proved<br />

similarly.<br />

□<br />

Exercise 1.119. Find (and justify your answers) :<br />

(1) a surjective continuous map that is not a quotient map.<br />

(2) a quotient map that is neither open nor closed.<br />

Exercise 1.120. Let {X α } α∈Λ be a collection <strong>of</strong> spaces. Show that for each<br />

β ∈ Λ, the projection map π β : ∏ α∈Λ X α → X β is always open, but it may not be<br />

closed.<br />

Definition 1.121. Let (X, τ X ) be a space, A a set and p : X → A a surjective<br />

map. The quotient topology τ p on A induced by p is the unique topology on A<br />

which makes p a quotient map; i.e.,<br />

τ p = {U ⊆ A | p −1 (U) ∈ τ X }.<br />

Example 1.122. Consider the surjective map p : R → {a, b, c} defined by<br />

⎧<br />

⎪⎨ a ; x < 0<br />

p(x) = b ; x = 0<br />

⎪⎩<br />

c ; x > 0.<br />

Then the quotient topology on {a, b, c} induced by p is {∅, {a, b, c}, {a}, {c}, {a, c}}.<br />

Definition 1.123. Let X be a space, A a partition <strong>of</strong> X into disjoint subsets<br />

and p : X → A the surjective map that sends each point in X to the element <strong>of</strong><br />

A containing it. The set A together with the quotient topology induced by p is<br />

usually called a quotient space (or an identification space) <strong>of</strong> X. Moreover, if ∼ is<br />

the equivalence relation on X induced by A, we will usually denote the quotient<br />

space A by X/ ∼.<br />

Remark 1.124. Since a subset V <strong>of</strong> the quotient space X/ ∼ is just a collection<br />

<strong>of</strong> equivalence classes, then V is open in X/ ∼ iff ∪ V is open in X.<br />

Example 1.125. Let I = [0, 1], and I 2 = I × I.


<strong>Phichet</strong> <strong>Chaoha</strong> 27<br />

(1) The 1-sphere (S 1 ) := I/ ∼, where 0 ∼ 1.<br />

(2) The 2-sphere (S 2 ) := I 2 / ∼, where (x, 0) ∼ (x ′ , 1) ∼ (0, y) ∼ (1, y ′ ), for<br />

each x, x ′ , y, y ′ ∈ I.<br />

(3) The Mobius strip := I 2 / ∼, where (0, y) ∼ (1, 1 − y), for each y ∈ I.<br />

(4) The 2-torus (T 2 ) := I 2 / ∼, where (x, 0) ∼ (x, 1) and (0, y) ∼ (1, y), for<br />

each x, y ∈ I.<br />

(5) The Klien bottle := I 2 / ∼, where (x, 0) ∼ (x, 1) and (0, y) ∼ (1, 1 − y),<br />

for each x, y ∈ I.<br />

(6) The real projective space (RP 2 ) := I 2 / ∼, where (x, 0) ∼ (1 − x, 1) and<br />

(0, y) ∼ (1, 1 − y), for each x, y ∈ I.


28 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Example 1.126. Let L = R × {−1, 1}. Define an equivalence relation ∼ on L<br />

by (x, −1) ∼ (x, 1) for all x ≠ 0. Then, the quotient space L/ ∼ is called the line<br />

with double points. It is easy to see that L is Hausdorff, but L/ ∼ is not!<br />

Definition 1.127. Let p : X → Y be a map and y ∈ Y . We will call the set<br />

p −1 ({y}) the fiber <strong>of</strong> p over y.<br />

Theorem 1.128. Let p : X → Y be a quotient map, Z a space, and g :<br />

X → Z a continuous map. If g is constant on each fiber <strong>of</strong> p, then there exists a<br />

unique continuous map f : Y → Z such that g = f ◦ p; i.e., we have the following<br />

commutative diagram :<br />

X<br />

❅ ❅❘ p<br />

g<br />

✲ Z<br />

♣ ♣ ♣ ♣<br />

✒<br />

∃!f<br />

Y<br />

Pro<strong>of</strong>. Since g is constant on each fiber <strong>of</strong> p, we can simply define f : Y → Z<br />

by f(y) = g(a) for each y ∈ Y and some point a in the fiber <strong>of</strong> p over y. Clearly,<br />

g = f ◦ p. For each open subset V <strong>of</strong> Z, the set g −1 (V ) = p −1 (f −1 (V )) is open by<br />

the continuity <strong>of</strong> g, and hence f −1 (V ) must be open in Y since p is a quotient map.<br />

Therefore, f is continuous. Now suppose f ′ : Y → Z is a map such that g = f ′ ◦ p.<br />

Then, for each y ∈ Y , we have f(y) = g(a) = f ′ ◦ p(a) = f ′ (y) where a is any point<br />

in the fiber over y.<br />

□<br />

Example 1.129. Let p : I 2 → S 2 and q : I 2 → T 2 be the quotient maps<br />

from the previous example. Then, by the previous theorem, there exists a unique<br />

continuous map f : T 2 → S 2 such that p = f ◦ q. Morever, it is easy to see that f<br />

is surjective.<br />

Exercise 1.130. Prove that there is a surjective continuous map from the<br />

Mobius strip onto the Klien bottle.


<strong>Phichet</strong> <strong>Chaoha</strong> 29<br />

8. Metric <strong>Topology</strong><br />

Definition 1.131. A metric on a set X ≠ ∅ is a map d : X ×X → R satisfying<br />

the following properties for all x, y, z ∈ X :<br />

(1) d(x, y) = d(y, x).<br />

(2) d(x, y) = 0 iff x = y.<br />

(3) d(x, z) ≤ d(x, y) + d(y, z).<br />

The pair (X, d) is called a metric space.<br />

Remark 1.132. From the conditions above, we have<br />

0 = d(x, x) ≤ d(x, y) + d(y, x) = d(x, y) + d(x, y) = 2d(x, y).<br />

Therefore, d(x, y) ≥ 0; i.e. d : X × X → R + 0 .<br />

Exercise 1.133. Let d be a metric on the set X. Prove that the map defined<br />

by d(x, y) = min{d(x, y), 1}, for any x, y ∈ X, is also a metric on X. In fact, this<br />

is still true if we replace 1 by any k ∈ R + .<br />

Definition 1.134. For a metric space (X, d), the metric d defined as in the<br />

previous example is called the standard bounded metric.<br />

Definition 1.135. Let (X, d) be a metric space, x ∈ X and r > 0. We define<br />

the open ball centered at x <strong>of</strong> radius r to be<br />

B d (x; r) = {y ∈ X | d(x, y) < r}.<br />

Definition 1.136. For a metric space (X, d), we define the metric topology<br />

induced by d, denoted by τ d , to be the topology on X whose basis is the set <strong>of</strong> all<br />

open balls.<br />

Example 1.137. For any set X ≠ ∅, the map d : X × X → R + 0<br />

{<br />

0 ; x = y<br />

d(x, y) =<br />

1 ; x ≠ y<br />

given by<br />

is a metric on X. This metric induces the discrete topology on X and hence it is<br />

called the discrete metric.<br />

Definition 1.138. A space X is said to be metrizable if its topology is induced<br />

by a metric.<br />

Example 1.139. The euclidean space R n is metrizable because its topology<br />

(the usual topology) is induced by the euclidean metric d.<br />

Exercise 1.140. For a metric space (X, d), prove that the map d : (X, τ d ) ×<br />

(X, τ d ) → R + 0 is continuous.<br />

Theorem 1.141. For a metric d on the set X, the metric topology induced by<br />

d is the coarsest topology on X which makes d continuous.<br />

Pro<strong>of</strong>. The map d : (X, τ d ) × (X, τ d ) → R + 0 is continuous by the previous<br />

exercise. Now, let τ be a topology on the set X which makes the map d : (X, τ) ×<br />

(X, τ) → R + 0 continuous. It suffices to show that τ contains all open balls. For x ∈<br />

X and r > 0, the continuity <strong>of</strong> d implies that the restriction d| {x}×X : {x} × X →<br />

R + 0 is continuous. Let d x : (X, τ) → R + 0 be the following composition :


30 <strong>Topology</strong> (<strong>2301631</strong>)<br />

d x<br />

X ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣✲♣ ♣ R + 0<br />

◗<br />

ι ◗◗ ✑ ✑✸<br />

d {x}×X<br />

{x} × X<br />

where ι(y) = (x, y) for each y ∈ X. Clearly, d x is continuous and hence B d (x; r) =<br />

d −1<br />

x ([0, r)) is open in X; i.e., B d (x; r) ∈ τ. □<br />

Theorem 1.142. The converse <strong>of</strong> the sequence lemma holds in a metric space.<br />

Pro<strong>of</strong>. Let (X, d) be a metric space, A ⊆ X and a ∈ A. Then, we have<br />

B d (a; 1 n )∩A ≠ ∅ for each n ∈ N. By the axiom <strong>of</strong> choice, we obtain a sequence (a n)<br />

in A such that a n ∈ B d (a; 1 n ) ∩ A for each n ∈ N. To show that (a n) converges to a,<br />

we let U be an neighborhood <strong>of</strong> a. Since the metric topology on X is generated by<br />

open balls, there exits ϵ > 0 such that B d (a; ϵ) ⊆ U. By choosing N ∈ N such that<br />

1<br />

N < ϵ, we have a n ∈ B d (a; 1 n ) ∩ A ⊆ B d(a; 1 N ) ∩ A ⊆ B d(a; ϵ) ∩ A ⊆ U ∩ A ⊆ U for<br />

all n ≥ N. This proves that (a n ) → a as desired.<br />

□<br />

Theorem 1.143. Let f : X → Y be a map <strong>of</strong> spaces, where X satisfies the<br />

converse <strong>of</strong> the sequence lemma. Then, f is continuous iff for any x ∈ X and any<br />

sequence (x n ) converging x, the sequence (f(x n )) converges to f(x).<br />

Pro<strong>of</strong>. (⇒) Follows from Exercise 1.95(1).<br />

(⇐) Let A ⊆ X and y ∈ f(A). Then y = f(x) for some x ∈ A. Since X<br />

satisfies the converse <strong>of</strong> the sequence lemma, there exists a sequence (a n ) in A<br />

converging to x. By the assumption, the sequence (f(a n )) (in f(A)) converges<br />

to f(x). Therefore, by the sequence lemma, y = f(x) ∈ f(A). It follows that<br />

f(A) ⊆ f(A) which implies the continuity <strong>of</strong> f.<br />

□<br />

Corollary 1.144. The previous theorem is true when X is metrizable.<br />

Example 1.145. The space S Ω is not metrizable because it does not satisfy<br />

the converse <strong>of</strong> the sequence lemma.<br />

Definition 1.146. Two metrics d 1 and d 2 on a set X are said to be (topologically)<br />

equivalent if d 1 and d 2 induce the same metric topology on X.<br />

Exercise 1.147. Let d and ρ be metrics on a set X. If there exist positive real<br />

numbers a, b and c such that for any x, y ∈ X,<br />

prove that d and ρ are equivalent.<br />

ad(x, y) ≤ bρ(x, y) ≤ cd(x, y),<br />

Exercise 1.148. On R n , we can define the square metric ρ and the taxicab<br />

metric d t by<br />

• ρ(x, y) = max{|x 1 − y 1 |, . . . , |x n − y n |}<br />

• d t (x, y) = ∑ n<br />

i=1 |x i − y i |<br />

for any x, y ∈ R n . Prove that ρ and d t are really metrics on R n , and both are<br />

equivalent to the euclidean metric d. [Hint : use the previous exercise]<br />

Unfortunately, for the metrizability <strong>of</strong> R ω , we do not have obvious generalizations<br />

<strong>of</strong> the d or d t because <strong>of</strong> the convergence issue. However, we can modify ρ to<br />

get the bounded version by letting<br />

ρ(x, y) = sup{d(x i , y i ) | i ∈ N}


<strong>Phichet</strong> <strong>Chaoha</strong> 31<br />

where d(a, b) = min{|a − b|, 1} denotes the standard bounded metric on R.<br />

Definition 1.149. The metric ρ as defined above is called the uniform metric<br />

on R ω , and the metric topology induced by ρ is called the uniform topology.<br />

Exercise 1.150. Prove that the uniform topology on R ω is finer than the<br />

product topology, and coarser than the box topology.<br />

To obtain the product topology on R ω , we need to modify ρ to get a new metric<br />

D as in the following theorem.<br />

Theorem 1.151. The map D : R ω × R ω → R defined by<br />

D(x, y) = sup{ d(x i, y i )<br />

| i ∈ N}<br />

i<br />

is a metric and it induces the product topology on R ω .<br />

Pro<strong>of</strong>. The fact that D is a metric is quite easy to verify.<br />

To show that the metric topology induced by D is finer than the product<br />

topology, we let U be an open set in the product topology and x = (x 1 , x 2 , . . . ) ∈ U.<br />

Then there is a basis element B in the product topology such that x ∈ B ⊆ U;<br />

says B = B 1 × B 2 × · · · × B n × ∏ i>n R for some n, where each B i is open in R.<br />

Then for each i = 1, . . . , n, there exists 0 < ϵ i ≤ 1 such that (x i − ϵ i , x i + ϵ i ) ⊆ B i .<br />

Now, let ϵ = min{ ϵ i<br />

i<br />

|i = 1, 2, . . . , n} and we claim that B D (x, ϵ) ⊆ B. To see this,<br />

let y ∈ B D (x, ϵ). Then for each i = 1, 2, . . . , n, we have |x i − y i | = d(x i , y i ) ≤<br />

iD(x, y) < iϵ ≤ ϵ i ≤ 1. Hence, y i ∈ (x i − ϵ i , x i + ϵ i ) ⊆ B i for each i = 1, 2, . . . , n;<br />

i.e., y ∈ B.<br />

On the other hand, we will show that the product topology on R ω is finer than<br />

the metric topology induced by D. Let U be an open set in the metric topology<br />

induced by D and x = (x 1 , x 2 , . . . ) ∈ U. Then B D (x, ϵ) ⊆ U for some ϵ > 0. Now,<br />

choose N ∈ N such that N > 1 ϵ . We claim that V = ∏ N<br />

i=1 (x i−ϵ, x i +ϵ)× ∏ i>N R ⊆<br />

B D (x, ϵ). Let y ∈ V . Clearly, for 1 ≤ i ≤ N, we have d(x i,y i )<br />

i<br />

≤ |x i−y i |<br />

i<br />

< ϵ i ≤ ϵ.<br />

Also, since d is bounded above by 1, we clearly have d(x i,y i )<br />

i<br />

≤ 1 N<br />

< ϵ for a i ≥ N.<br />

It follows that<br />

D(x, y) ≤ max{ d(x 1, y 1 )<br />

1<br />

That is y ∈ B D (x, ϵ) as desired.<br />

, d(x 2, y 2 )<br />

2<br />

, . . . , d(x N , y N )<br />

, 1 N N } < ϵ.<br />


CHAPTER 2<br />

Countability and Separation Axioms<br />

1. The Countability Axioms<br />

Definition 2.1. Let X be a space and x ∈ X. A collection B x <strong>of</strong> open subsets<br />

<strong>of</strong> X is called a (local) basis at x if for each neighborhood U <strong>of</strong> x, there exists<br />

B ∈ B x such that x ∈ B ⊆ U.<br />

Definition 2.2. A space X is called<br />

(1) first-countable if it has a countable basis at each point.<br />

(2) second-countable if it has a countable basis.<br />

(3) separable if it has a countable dense subset.<br />

Theorem 2.3. If X is second-countable, then it is first-countable and separable.<br />

Pro<strong>of</strong>. Let X be second-countable. Then it is clear from the definition that<br />

X is also first-countable. Now, let B be a countable basis for X. By the axiom <strong>of</strong><br />

choice, we can form a subset D <strong>of</strong> X by choosing one element from each set in B.<br />

It follows that D is the desired countable dense subset <strong>of</strong> X.<br />

□<br />

Example 2.4. R n is clearly second-countable by taking the collection all products<br />

<strong>of</strong> intervals with rational endpoints as a basis. Hence, it is first-countable by<br />

the previous theorem. Moreover, it is separable since Q n is its countable dense<br />

subset. Hence, R n satisfies all countability axioms.<br />

Example 2.5. R l is first-countable and separable, but not second-countable.<br />

It is easy to see that the intervals [x, x + 1 n ) form the basis at x ∈ R l, and clearly<br />

Q is dense in R l . To see that R l is not second-countable, let B be a basis for R l .<br />

Then, for x, x ′ ∈ R l , there exists B x , B x ′ ∈ B such that x ∈ B x ⊆ [x, x + 1) and<br />

x ′ ∈ B x ′ ⊆ [x ′ , x ′ + 1). Clearly, inf(B x ) = x because {x} ⊆ B x ⊆ [x, x + 1), and<br />

similarly, inf(B x) ′ = x ′ . Hence, if x ≠ x ′ , we must have B x ≠ B x ′. So, B contains<br />

an uncountable subcollection {B x | x ∈ R l }, and hence B is uncountable.<br />

Exercise 2.6. Let X be a second-countable space. Prove that<br />

(1) Every collection <strong>of</strong> disjoint open sets in X is countable.<br />

(2) The subspace topology on an uncountable subset <strong>of</strong> X cannot be discrete.<br />

Theorem 2.7. A subspace <strong>of</strong> a first-countable [second-countable] space is firstcountable<br />

[second-countable]. A countable product <strong>of</strong> first-countable [second-countable]<br />

spaces is first-countable [second-countable].<br />

Pro<strong>of</strong>. The first statement is clear since we can construct a basis element<br />

<strong>of</strong> a subspace simply by intersecting a basis element <strong>of</strong> the whole space with the<br />

subspace. For a countable product, we use the fact that a countable union <strong>of</strong><br />

countable sets and a finite product <strong>of</strong> countable sets are countable. For example,<br />

33


34 <strong>Topology</strong> (<strong>2301631</strong>)<br />

let {X n } n∈N be a countable collection <strong>of</strong> second countable spaces and B n a countable<br />

basis <strong>of</strong> X n . Then<br />

B = ∪ (<br />

∏ n<br />

B i × ∏ )<br />

{X i }<br />

n∈N i=1 i>n<br />

can be shown to be a countable basis for the product space ∏ n∈N X n. Notice the<br />

use <strong>of</strong> the product topology here!<br />

□<br />

Exercise 2.8. Prove that a countable product <strong>of</strong> a separable space is also<br />

separable.<br />

Example 2.9. R ω satisfies all countability axioms by the previous theorem and<br />

exercise.<br />

The next example shows that a subspace <strong>of</strong> a separable space may not be<br />

separable.<br />

Example 2.10. Since R l is separable (Q is also dense in R l ), then from the<br />

previous exercise, so is R 2 l<br />

. However, the subspace L = {(x, y) | x + y = 1} <strong>of</strong><br />

R 2 l<br />

is not separable because the topology on L is discrete (i.e., for each (x, y) ∈ L,<br />

we have ([x, x + 1) × [y, y + 1)) ∩ L = {(x, y)} which is open in L). Therefore, the<br />

uncountable space L cannot have a countable dense subset.<br />

As we have seen in Theorem 2.3, the separability is generally weaker than<br />

second-countability. However, these two concepts are equivalent when the space is<br />

metrizable.<br />

Theorem 2.11. A metric space is first-countable, and it is separable iff it is<br />

second-countable.<br />

Pro<strong>of</strong>. Let X be a metric space. Then, for each x ∈ X, the collection<br />

{B(x; 1 n<br />

) | n ∈ N} forms a countable basis at x. Now, if X is separable, we let D be a<br />

countable dense subset <strong>of</strong> X. To see that the collection {B(p; 1 n<br />

) | p ∈ D and n ∈ N}<br />

forms a countable basis for X, let U be an open subset <strong>of</strong> X and x ∈ U. Then there<br />

is n ∈ N such that B(x; 1 1<br />

n<br />

) ⊆ U. Since D is dense in X, there exists p ∈ D∩B(x;<br />

2n ).<br />

Now it is easy to verify that x ∈ B(p; 1<br />

2n ) ⊆ B(x; 1 n<br />

) ⊆ U. The converse follows<br />

directly from Theorem 2.3.<br />

□<br />

Corollary 2.12. R l is not metrizable.<br />

Theorem 2.13. The converse <strong>of</strong> the sequence lemma holds for a first-countable<br />

space (e.g. a metric space).<br />

Pro<strong>of</strong>. Let X be a first-countable space, A ⊆ X and a ∈ A. Let B a =<br />

{B 1 , B 2 , . . . } be a countable basis at a. For each n ∈ N, let B n ′ = B 1 ∩ · · · ∩ B n .<br />

Clearly, each B n ′ is open and B n+1 ′ ⊆ B n ′ ⊆ B n . Since a ∈ A, we can form a<br />

sequence (a n ) in A by choosing a n ∈ B n ′ ∩ A for each n. To see that the sequence<br />

(a n ) converges to a, we let U be a neighborhood <strong>of</strong> a. Since B a is a basis at a, there is<br />

N ∈ N such that a ∈ B N ⊆ U. Then for n ≥ N, we have a n ∈ B n ′ ⊆ B N ′ ⊆ B N ⊆ U<br />

as desired.<br />

□<br />

Corollary 2.14. The space S Ω is not first-countable.<br />

Exercise 2.15. Is S Ω separable? Justify your answer.<br />

Exercise 2.16. Which one <strong>of</strong> our three countability axioms does S Ω satisfy?<br />

Justify your answer.


<strong>Phichet</strong> <strong>Chaoha</strong> 35<br />

Definition 2.17. Let X be a space.<br />

2. The Separation Axioms<br />

(1) X is a T 0 −space if for any two distinct points in X, we can find a neighborhood<br />

<strong>of</strong> one <strong>of</strong> the point not containing the other.<br />

(2) X is a T 1 −space or a quasi-separated space if for any two distinct points<br />

in X, we can find a neighborhood <strong>of</strong> each point not containing the other.<br />

(3) X is a T 2 −space or a separated space or a Hausdorff space if any two<br />

distinct points <strong>of</strong> X have disjoint neighborhoods.<br />

(4) X is a T 3 −space or a regular space if it is T 1 and any point and any closed<br />

subset not containing that point have disjoint neighborhoods.<br />

(5) X is a T 3 1 −space or a completely regular space if it is T 1 and any closed<br />

2<br />

subset A <strong>of</strong> X and any point x /∈ A, there exists a continuous map f :<br />

X → [0, 1] such that f(x) = 0 and A ⊆ f −1 ({1}). (Note that if A = ∅,<br />

then f is the zero map)<br />

(6) X is a T 4 −space or a normal space if it is T 1 and any two disjoint closed<br />

subsets have disjoint neighborhoods.<br />

(7) X is a T 5 −space or a completely normal space if it is T 1 and any two<br />

separated subsets A, B <strong>of</strong> X (i.e. A ∩ B = A ∩ B = ∅) have disjoint<br />

neighborhoods.<br />

Remark 2.18. We need X to be T 1 in the definition <strong>of</strong> T i for i ≥ 3 because<br />

the space {a, b} with indiscrete topology satisfies the other part <strong>of</strong> the definition,<br />

but it is not even Hausdorff!<br />

Example 2.19. X = {a, b} with the topology {∅, {a}, {a, b}} is clearly T 0 , but<br />

not T 1 .<br />

Exercise 2.20. Prove that X is T 0 iff {x} ̸= {y}, for any distinct points x, y<br />

in X.<br />

Theorem 2.21. A space X is T 1 iff every singleton is closed.<br />

Pro<strong>of</strong>. (⇒) Suppose X is T 1 and let x ∈ X. For each y ≠ x, we can find a<br />

neighborhood U y not containing x. Then X − {x} = ∪ y≠x U y is open, and hence<br />

{x} is closed.<br />

(⇐) Suppose that every singleton is closed. Then, for distinct points x, y ∈ X,<br />

the open sets X − {y} and X − {x} are the desired neighborhoods <strong>of</strong> x and y,<br />

respectively.<br />

□<br />

Corollary 2.22. A space X is T 1 iff every finite subset <strong>of</strong> X is closed.<br />

Remark 2.23. Using the previous theorem, it is not difficult to see that a<br />

T i -space is also a T i−1 -space for i = 1, 2, 3, 4, 5. Moreover, it is straightforward to<br />

verify that a T 3 1 -space is a T 3-space. Leter on, we will also see that a T<br />

2 4 -space is<br />

a T 3 1 -space.<br />

2<br />

Theorem 2.24. A space X is Hausdorff iff the diagonal<br />

is closed in X × X.<br />

∆ = {(x, x) | x ∈ X}


36 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Suppose X is T 2 . Let (x, y) ∈ X × X − ∆; i.e., x ≠ y. Then there are<br />

neighborhoods U x and U y <strong>of</strong> x and y, respectively, such that U x ∩ U y = ∅. This<br />

implies that V = U x × U y is open in X × X and (x, y) ∈ V ⊆ X × X − ∆. Thus,<br />

∆ is closed in X × X. We can prove the other direction simply by reversing the<br />

above argument.<br />

□<br />

Exercise 2.25. If f, g : X → Y are continuous maps and Y is Hausdorff, then<br />

the coincidence set<br />

C(f, g) = {x ∈ X | f(x) = g(x)}<br />

is closed in X. In particular, for a continuous self-map f <strong>of</strong> a Hausdorff space X,<br />

the fixed point set<br />

is closed in X.<br />

F (f) = C(f, id X ) = {x ∈ X | f(x) = x}<br />

Theorem 2.26. Let D be a dense subset <strong>of</strong> X. If f : D → Y is a continuous<br />

map and Y is Hausdorff, then there is at most one continuous extension <strong>of</strong> f to X.<br />

Pro<strong>of</strong>. Suppose g, h : X → Y are continuous extensions <strong>of</strong> f; i.e., g| D = h| D .<br />

Let x ∈ X and suppose that g(x) ≠ h(x). Since Y is Hausdorff, there exist disjoint<br />

neighborhoods U ′ and V ′ <strong>of</strong> g(x) and h(x), respectively. By continuity <strong>of</strong> g and h,<br />

the sets U = g −1 (U ′ ), V = h −1 (V ′ ) and W = U ∩ V are all neighborhoods <strong>of</strong> x.<br />

Since D is dense in X, there exists d ∈ D ∩ W . Since both h and g are extensions<br />

<strong>of</strong> f, we have g(d) = f(d) = h(d) ∈ U ′ ∩ V ′ . This contradicts to the fact that<br />

U ′ ∩ V ′ = ∅. Therefore, g(x) = h(x).<br />

□<br />

Remark 2.27. The previous theorem only guarantees the uniqueness, not the<br />

existence, <strong>of</strong> an extension. However, we will see later on that if f is ”uniformly”<br />

continuous, an extension always exists.<br />

Theorem 2.28. A T 1 -space X is regular iff for each point x ∈ X and a neighborhood<br />

U <strong>of</strong> x, there exists a neighborhood V <strong>of</strong> x such that V ⊆ V ⊆ U.<br />

Pro<strong>of</strong>. (⇒) Assume that X is regular. Let x ∈ X and U a neighborhood <strong>of</strong><br />

x. Then, by regularity, there exist disjoint neighborhoods V <strong>of</strong> x and W <strong>of</strong> X − U.<br />

So, V ⊆ X − W and X − W is a closed set contained in U. Therefore, we have<br />

x ∈ V ⊆ V ⊆ X − W = X − W ⊆ U as desired.<br />

(⇐) Assume that, for each x ∈ X and a neighborhood U <strong>of</strong> x, there exists a<br />

neighborhood V <strong>of</strong> x such that x ∈ V ⊆ V ⊆ U. Now, let x ∈ X and A be a closed<br />

subset <strong>of</strong> X not containing x. By setting U = X −A which is clearly a neighborhood<br />

<strong>of</strong> x, there exists a neighborhood V <strong>of</strong> X such that x ∈ V ⊆ V ⊆ U = X − A.<br />

Hence, X − V is a neighborhood <strong>of</strong> A such that V ∩ (X − V ) = ∅.<br />

□<br />

Theorem 2.29. A T 1 -space X is normal iff for each closed set A in X and an<br />

open set U containing A, there exists a neighborhood V <strong>of</strong> A such that V ⊆ V ⊆ U.<br />

Pro<strong>of</strong>. The pro<strong>of</strong> is similar to the pro<strong>of</strong> <strong>of</strong> the above theorem.<br />

Exercise 2.30. Prove that every well-ordered set with the order topology is<br />

normal. In particular, S Ω and S Ω are normal.<br />

Exercise 2.31. Prove that a space X is completely normal iff every subspace<br />

<strong>of</strong> X is normal.<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 37<br />

Lemma 2.32. A metric space is normal.<br />

Pro<strong>of</strong>. Let A and B be disjoint closed subsets <strong>of</strong> metric space X. For each<br />

a ∈ A and each b ∈ B, let ϵ a > 0 and ϵ b > 0 be such that<br />

B(a; ϵ a ) ∩ B = ∅ and B(b; ϵ b ) ∩ A = ∅.<br />

Now, let U = ∪ a∈A B(a; ϵ a<br />

2<br />

), and V = ∪ b∈B B(b; ϵ b<br />

2<br />

). Clearly, U and V are open sets<br />

containing A and B, respectively. To see that U ∩V = ∅, suppose there is z ∈ U ∩V .<br />

Then, z ∈ B(a; ϵ a<br />

2<br />

) ∩ B(b; ϵ b<br />

2<br />

) for some a ∈ A and b ∈ B. WLOG, we may assume<br />

that ϵ b ≤ ϵ a . So, we have d(a, b) ≤ d(a, z) + d(z, b) < ϵa 2 + ϵ b<br />

2<br />

≤ ϵa 2 + ϵa 2 = ϵ a which<br />

is impossible because B(a; ϵ a ) ∩ B = ∅.<br />

□<br />

Corollary 2.33. A metric space is completely normal.<br />

Pro<strong>of</strong>. Since every subspace <strong>of</strong> a metric space X is also a metric space, and<br />

hence normal, then X itself is completely normal.<br />

□<br />

Theorem 2.34. A second-countable regular space is normal.<br />

Pro<strong>of</strong>. Let A and B be disjoint closed subsets <strong>of</strong> a second-countable regular<br />

space X, and B a countable basis <strong>of</strong> X. By regularity, we can find countable<br />

subcollections {U 1 , U 2 , . . . } and {V 1 , V 2 , . . . } <strong>of</strong> B such that A ⊆ ∪ ∞ i=1 U i, B ⊆<br />

∪ ∞ i=1 V i, U n ∩ B = ∅ and V n ∩ A = ∅ for all n ∈ N.<br />

Since ∪ ∞ i=1 U i and ∪ ∞ i=1 V i may not be disjoint, then for each n ∈ N, we let<br />

U n ′ = U n − ∪ n i=1 V i and V n ′ = V n − ∪ n i=1 U i. Clearly, U n’s ′ and V n’s ′ are open and<br />

pairwise disjoint. Now, it is rather straightforward to check that U ′ = ∪ n≥1 U n ′ and<br />

V ′ = ∪ n≥1 V n ′ are disjoint neighborhoods <strong>of</strong> A and B, respectively.<br />

□<br />

Theorem 2.35 (Urysohn Lemma). Let A and B be disjoint closed subsets <strong>of</strong> a<br />

normal space X, and [a, b] a closed in interval in R. Then there exists a continuous<br />

map f : X → [a, b] such that A ⊆ f −1 ({a}) and B ⊆ f −1 ({b}).<br />

Pro<strong>of</strong>. WLOG, we can assume that A ≠ ∅ ̸= B, a = 0 and b = 1.<br />

We will construct a continuous map f : X → [0, 1] with the property that<br />

f(A) = {0} and f(B) = {1}. First let P = Q ∩ [0, 1] and fix a representation<br />

<strong>of</strong> P in such a way that the first two elements are 1 and 0 (for example, P =<br />

{1, 0, 1 2 , 1 3 , 2 3 , . . . }). Let P n be the subset <strong>of</strong> the first n elements <strong>of</strong> P according to<br />

the representation we have just fixed.<br />

First let U 1 = X − B and, by normality, let U 0 be an open set such that<br />

A ⊆ U 0 ⊆ U 0 ⊆ U 1 . Hence, we have defined U i for all i ∈ P 2 . Inductively, we<br />

assume that for some n ≥ 2, U i is defined for all i ∈ P n so that U i ⊆ U j for i < j.<br />

Now, suppose P n+1 = P n ∪ {r}. Since r /∈ {0, 1}, we can find in P n the immediate<br />

predecessor and the immediate successor <strong>of</strong> r (according to the usual ordering <strong>of</strong><br />

R), says p and q, respectively. Again, by normality, there exists an open set U r<br />

such that U p ⊆ U r ⊆ U r ⊆ U q ; i.e., U i is now defined for all i ∈ P n+1 . Hence, we<br />

obtain the collection {U i } i∈P with the desired property. Moreover, we can extend<br />

the definition <strong>of</strong> U i to all i ∈ Q simply by letting<br />

{<br />

∅ for i ∈ Q ∩ (−∞, 0)<br />

U i =<br />

X for i ∈ Q ∩ (1, ∞).<br />

Finally, we define f : X → [0, 1] by<br />

f(x) = inf{i | x ∈ U i }


38 <strong>Topology</strong> (<strong>2301631</strong>)<br />

for all x ∈ X. Clearly, we have f(A) = {0} and f(B) = {1} as desired. For the<br />

continuity <strong>of</strong> f, we leave it as the next exercise.<br />

□<br />

Exercise 2.36. Verify the continuity <strong>of</strong> the map f in the theorem above.<br />

Remark 2.37. From the theorem above, we usually says that A and B can<br />

be separated by a continuous map. The reason we do not write f(A) = {0} and<br />

f(B) = {1} is because A or B may be empty.<br />

Corollary 2.38. A normal space is always completely regular.<br />

Recall : (Weierstrass M-test)<br />

Let (f n : X → R) be a sequence <strong>of</strong> maps. If, for each n ≥ 1, there exists<br />

M n ∈ R such that |f n (x)| ≤ M n for all x ∈ X, and the series ∑ ∞<br />

n=1 M n converges,<br />

then the series <strong>of</strong> maps ∑ ∞<br />

n=1 f n converges uniformly to a map f : X → R.<br />

Theorem 2.39 (Tietze Extension Theorem). Let A be a closed subset <strong>of</strong> a<br />

normal space X. Then any continuous map f : A → [a, b] can be extended to a<br />

continuous map g : X → [a, b].<br />

Pro<strong>of</strong>. WLOG, we may assume that A ≠ ∅, a = −1 and b = 1.<br />

First, we claim that for any r > 0 and a continuous map f : A → [−r, r], there<br />

exists a continuous map g : X → [− r 3 , r 2r<br />

3<br />

] such that |f(a) − g(a)| ≤<br />

3<br />

for all a ∈ A.<br />

To prove this claim, we divide [−r, r] into 3 intervals, namely I 1 = [−r, − r 3 ],<br />

I 2 = [− r 3 , r 3 ] and I 3 = [ r 3 , r]. Let B = f −1 (I 1 ) and C = f −1 (I 3 ). Then B and<br />

C are disjoint closed subsets <strong>of</strong> X because f is continuous and A is closed in X.<br />

By the Urysohn lemma, there exists a continuous map g : X → [− r 3 , r 3<br />

] such that<br />

B ⊆ g −1 ({− r 3 }) and C ⊆ g−1 ({ r 3<br />

}). Also, it is not difficult to check (by considering<br />

3 cases where a ∈ B, a ∈ C and a /∈ B ∪ C) that |f(a) − g(a)| ≤ 2r<br />

3<br />

for all a ∈ A.<br />

Therefore, we have proved the claim.<br />

Now, apply the above claim to f : A → [−1, 1] in the condition <strong>of</strong> this theorem.<br />

We get g 1 : X → [− 1 3 , 1 3 ] such that |f(a) − g 1(a)| ≤ 2 3<br />

for all a ∈ A. Apply the<br />

claim again to (f − g 1 ) : A → [− 2 3 , 2 3 ], we get g 2 : X → [−( 1 3 )( 2 3 ), ( 1 3 )( 2 3<br />

)] such that<br />

|f(a) − g 1 (a) − g 2 (a)| ≤ ( 2 3 )2 for all a ∈ A. Inductively, by applying the claim to<br />

the map (f − g 1 − g 2 − · · · − g n ) : A → [−( 2 3 )n , ( 2 3 )n ], we get a continuous map<br />

g n+1 : X → [−( 1 3 )( 2 3 )n , ( 1 3 )( 2 3 )n ] such that<br />

n+1<br />

∑<br />

|f(a) − g i (a)| = |f(a) − g 1 (a) − g 2 (a) − · · · − g n (a) − g n+1 (a)| ≤ ( 2 3 )n+1 . . . (∗)<br />

i=1<br />

for all a ∈ A.<br />

Since, for each n ≥ 1 and x ∈ X, |g n (x)| ≤ ( 1 3 )( 2 3 )n−1 and ∑ ∞<br />

n=1 ( 1 3 )( 2 3 )n−1 converges,<br />

Weierstrass M-test implies that the sequence <strong>of</strong> maps ( ∑ n<br />

i=1 g i) converges<br />

uniformly on X, says to a g. Clearly, g is continuous (since each ∑ n<br />

i=1 g i is continuous)<br />

and |g(x)| ≤ 1 (by comparing with the geometric series ∑ ∞<br />

n=1 ( 1 3 )( 2 3 )n−1 ).<br />

Moreover, f(a) = g(a) for all a ∈ A (by *). Hence, g : X → [−1, 1] is the desired<br />

continuous extension <strong>of</strong> f.<br />

□<br />

Exercise 2.40. Prove that if A be a closed subset <strong>of</strong> a normal space X, then<br />

any continuous map f : A → R can be extended to a continuous map g : X → R.<br />

Theorem 2.41. For a T 1 space X, the following statements are equivalent :<br />

(1) X is normal.


<strong>Phichet</strong> <strong>Chaoha</strong> 39<br />

(2) For any disjoint closed subsets A and B <strong>of</strong> X, there exists a continuous<br />

map f : X → [0, 1] such that A ⊆ f −1 ({0}) and B ⊆ f −1 ({1}).<br />

(3) For any closed subsets A <strong>of</strong> X, a continuous map f : A → [0, 1] can be<br />

extended to a continuous map g : X → [0, 1].<br />

Pro<strong>of</strong>. (1) ⇒ (2) : Urysohn lemma.<br />

(1) ⇒ (3) : Tietze extension theorem.<br />

(2) ⇒ (1) : Let A, B be disjoint closed subsets <strong>of</strong> X. By (2), there exists<br />

a continuous map f : X → [0, 1] such that A ⊆ f −1 ({0}) and B ⊆ f −1 ({1}).<br />

Then, U = f −1 [0, 1 2 ) and V = f −1 ( 1 2<br />

, 1] for disjoint neighborhoods <strong>of</strong> A and B,<br />

respectively.<br />

(3) ⇒ (2) : Let A, B be disjoint closed subsets <strong>of</strong> X. By the pasting lemma,<br />

we have a continuous map f : A ∪ B → [0, 1] defined by f(x) = 0 if x ∈ A, and<br />

f(x) = 1 if x ∈ B. Then, by (3), f can be extended the desired map.<br />

□<br />

The following theorem summarizes some other properties <strong>of</strong> T i -spaces, the<br />

pro<strong>of</strong>s are straightforward and hence left to the reader.<br />

Theorem 2.42. Let i ≤ 3 1 2 .<br />

(1) A subspace <strong>of</strong> a T i -space is also a T i -space.<br />

(2) A product T i -spaces is a T i -space.<br />

Pro<strong>of</strong>. Exercise.<br />

Example 2.43 (See [4] for details). The product space S Ω ×S Ω is regular since<br />

both S Ω and S Ω are normal (being well-ordered). However, S Ω × S Ω is not normal.<br />

In fact, we will see later on that S Ω ×S Ω is normal (being compact and Hausdorff).<br />

Hence, S Ω × S Ω also serves as an example <strong>of</strong> a subspace <strong>of</strong> a normal space which is<br />

not normal.<br />

Exercise 2.44. Prove that a closed subspace <strong>of</strong> a normal space is always normal.<br />

We end this chapter by stating the Urysohn metrization theorem. The pro<strong>of</strong><br />

can be found in [4].<br />

Theorem 2.45 (Urysohn Metrization Theorem). A second-countable regular<br />

space is metrizable.<br />

Exercise 2.46. Give an example <strong>of</strong> a second-countable Hausdorff space that<br />

is not metrizable.<br />


CHAPTER 3<br />

Connectedness<br />

Definition 3.1. A separation (or a disconnection) <strong>of</strong> a space X is a pair {U, V }<br />

<strong>of</strong> disjoint nonempty open subsets <strong>of</strong> X such that X = U ∪ V .<br />

Remark 3.2. If {U, V } is a separation <strong>of</strong> X, then U and V are both open and<br />

closed in X at the same time.<br />

Definition 3.3. A space X is said to be connected if it has no separation.<br />

Otherwise, it is said to be disconnected. Similary, a subset A <strong>of</strong> X is said to<br />

be connected if it is connected as a subspace <strong>of</strong> X. Otherwise, it is said to be<br />

disconnected.<br />

Example 3.4. A discrete space X with |X| ≥ 2 is always disconnected. An<br />

empty space and a one-point space are always connected.<br />

Exercise 3.5. Prove that, for any −∞ ≤ a < b ≤ ∞, an open interval (a, b)<br />

is connected. In particular, R is connected.<br />

Exercise 3.6. Prove that X is disconnected iff there exists a continuous surjective<br />

map f : X → Y where Y is any discrete space with 2 points.<br />

Theorem 3.7. A space X is connected iff ∅ and X are the only subsets <strong>of</strong> X<br />

that are both open and closed in X.<br />

Pro<strong>of</strong>. Follows directly from the definition.<br />

Theorem 3.8. Let X be a space and Y ⊆ X. The pair {U, V } is a separation<br />

<strong>of</strong> Y iff U and V are nonempty separated sets in X such that U ∪ V = Y .<br />

Pro<strong>of</strong>. (⇒) Let {U, V } be a separation <strong>of</strong> Y . Then, U and V are disjoint<br />

nonempty open subsets <strong>of</strong> Y such that U ∪ V = Y . It remains to show that<br />

U and V are separated in X. Since both U and V are also closed in Y , we have<br />

Cl Y (U) = U and Cl Y (V ) = V . It follows that U ∩V = (U ∩Y )∩V = U ∩(Y ∩V ) =<br />

U ∩ Cl Y (V ) = U ∩ V = ∅, and similarly, U ∩ V = ∅ as desired.<br />

(⇐) Now, let U and V be nonempty separated sets in X such that U ∪ V = Y .<br />

Clearly, U and V are disjoint subsets <strong>of</strong> Y . Since Cl Y (U) ∩ V ⊆ U ∩ V = ∅, it<br />

follows that Cl Y (U) ⊆ Y − V = U; i.e., U is closed in Y . Similarly, V is also closed<br />

in Y . Therefore, both U and V are open, and hence forms a separation <strong>of</strong> Y . □<br />

Example 3.9. In R, the pair {[−1, 0), (0, 1]} forms a separation <strong>of</strong> the subspace<br />

Y = [−1, 0) ∪ (0, 1], while the pair {[−1, 0], (0, 1]} does not form a separation <strong>of</strong> the<br />

subspace Z = [−1, 1].<br />

Lemma 3.10. If {U, V } is a separation <strong>of</strong> X and Y is a nonempty connected<br />

subspace <strong>of</strong> X, then either Y ⊆ U or Y ⊆ V .<br />

41<br />


42 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Since U and V are both open and closed in X, it follows that U ∩ Y<br />

and V ∩ Y are both open and closed in Y . Since Y is connected, either U ∩ Y = Y<br />

or V ∩Y = Y (and the other one will be ∅). Therefore, either Y ⊆ U or Y ⊆ V . □<br />

Theorem 3.11. Let {X α } α∈Λ be a collection <strong>of</strong> connected spaces such that<br />

X α ∩ X β ≠ ∅ for each α, β ∈ Λ. Then ∪ α∈Λ X α is connected.<br />

Pro<strong>of</strong>. Suppose ∪ α∈Λ X α is disconnected with a separation {U, V }. We claim<br />

that either X α ⊆ U for all α ∈ Λ, or X α ⊆ V for all α ∈ Λ. The claim will imply<br />

that either V = ∅ or U = ∅, which is contradiction. Hence, ∪ α∈Λ X α must be<br />

connected. To prove the claim, suppose X β ⊆ U and X γ ⊆ V for some β, γ ∈ Λ.<br />

Since X β ∩ X γ ≠ ∅, it follows that U ∩ V ≠ ∅ which is a contradiction. Therefore,<br />

we have proved the claim.<br />

□<br />

Corollary 3.12. The union <strong>of</strong> a collection <strong>of</strong> connected spaces that have a<br />

point in common is connected.<br />

Theorem 3.13. If A is a connected subset <strong>of</strong> X and A ⊆ B ⊆ A, then B is<br />

also connected.<br />

Pro<strong>of</strong>. Suppose B is disconnected with a separation {U, V }. Then U and V<br />

are nonempty separated sets in X such that U ∪V = B. Since A is connected, then<br />

either A ⊆ U or A ⊆ V . WLOG, assume that A ⊆ U. Then, A ⊆ U and hence<br />

V = B ∩ V ⊆ A ∩ V ⊆ U ∩ V = ∅ which is a contradition. Therefore, B must be<br />

connected.<br />

□<br />

Corollary 3.14. If A is a connected subset <strong>of</strong> X, then so is A.<br />

Example 3.15. The intervals [a, b], [a, b) and (a, b] are connected.<br />

Theorem 3.16. If X is a connected space and f : X → Y is a continuous map,<br />

then f(X) is also connected.<br />

Pro<strong>of</strong>. WLOG, assume that f is surjective; i.e., f(X) = Y . Suppose Y is<br />

disconnected with a separation {U, V }. Then it is easy to verify that the pair<br />

{f −1 (U), f −1 (V )} forms a separation <strong>of</strong> X which contradicts to the fact that X is<br />

connected. Therefore, Y must be connected.<br />

□<br />

Corollary 3.17. Connectedness is a topological property.<br />

Corollary 3.18 (Intermediate Value Theorem). If f : [a, b] → R is a continuous<br />

map and y 0 is a point between f(a) and f(b), then there exists x 0 ∈ (a, b) such<br />

that f(x 0 ) = y 0 .<br />

Pro<strong>of</strong>. Let y 0 be a point between f(a) and f(b). Suppose y 0 /∈ f([a, b]).<br />

Then, f : [a, b] → R − {y 0 } is clearly continuous. Since {(−∞, y 0 ), (y 0 , ∞)} forms<br />

a separation <strong>of</strong> R − {y 0 } and f([a, b]) is connected by the previous theorem, then<br />

either f([a, b]) ⊆ (−∞, y 0 ) or f([a, b]) ⊆ (y 0 , ∞) which contradicts to the fact that<br />

y 0 is between f(a) and f(b). It follows that y 0 = f(x 0 ) for some x 0 ∈ [a, b]−{a, b} =<br />

(a, b).<br />

□<br />

Theorem 3.19. An arbitrary product <strong>of</strong> connected spaces is connected.


<strong>Phichet</strong> <strong>Chaoha</strong> 43<br />

Pro<strong>of</strong>. First, we will prove that X ×Y is connected if X and Y are connected.<br />

Fix (a, b) ∈ X ×Y and consider T x = (X ×{b})∪({x}×Y ). Clearly, T x is connected<br />

for all x ∈ X because X×{b} ∼ = X, {x}×Y ∼ = Y and (X×{b})∩({x}×Y ) = {(x, b)}.<br />

Now, since X × Y = ∪ x∈X T x and (a, b) ∈ T x for all x ∈ X, it follows that X × Y<br />

is also connected, and hence so is a finite product <strong>of</strong> connected spaces.<br />

To prove this theorem for an arbitrary product, let X = ∏ α∈Λ X α be a product<br />

<strong>of</strong> connected spaces X α ’s. WLOG, we may assume that each X α has more than<br />

one element. Fix a point b = (b α ) ∈ X, and let X(α 1 , . . . , α n ) denote the subspace<br />

<strong>of</strong> X containing all points (x α ) such that x α = b α for α /∈ {α 1 , . . . , α n }. It is not<br />

difficult to see that X(α 1 , . . . , α n ) ∼ = X α1 × · · · × X αn and hence it is connected<br />

from the previous paragraph. Now, let Y = ∪ α 1 ,...,α n ∈Λ X(α 1, . . . , α n ) ⊆ X. Since<br />

b ∈ X(α 1 , . . . , α n ) for any α 1 , . . . , α n ∈ Λ, then Y is connected. Note that Y ≠ X<br />

because, for example, the point (x α ) with x α ≠ b α for all α ∈ Λ is not in Y .<br />

However, we can show that Y = X. For a point (x α ) ∈ X and a basis neighborhood<br />

(in the product topology) ∏ α∈Λ U α <strong>of</strong> (x α ), we have U α ≠ X α for finitely many<br />

α’s, says α 1 , . . . , α n . Let (y α ) ∈ X be defined by<br />

{<br />

x α if α ∈ {α 1 , . . . α n },<br />

y α =<br />

b α if α /∈ {α 1 , . . . α n }.<br />

It is easy to see that<br />

(y α ) ∈ X(α 1 , . . . , α n ) ∩ ∏ α∈Λ<br />

U α ⊆ Y ∩ ∏ α∈Λ<br />

U α .<br />

Therefore, Y = X and hence, by Corollary 3.14, X is also connected as desired.<br />

Definition 3.20. Let x and y be points in a space X. A path (in X) from x<br />

to y is a continuous map p : [0, 1] → X such that p(0) = x and p(1) = y. A loop<br />

(in X) at x is simply a path from x to x.<br />

Definition 3.21. A space X is said to be path connected if each pair <strong>of</strong> points<br />

in X can be joined by a path in X.<br />

Example 3.22. Both R n and D n = {x ∈ R n : ||x|| ≤ 1} are path connected<br />

for n ≥ 0, while S n = ∂D n+1 is path connected for n > 0.<br />

Exercise 3.23. Prove that a continuous image <strong>of</strong> a path connected space is<br />

always path connected, and use this fact to show that R cannot be homeomorphic<br />

to R 2 .<br />

Exercise 3.24. Let X and Y be path connected spaces. Prove that X × Y is<br />

path connected, and if X ∩ Y ≠ ∅, then X ∪ Y is also path connected.<br />

Theorem 3.25. A path connected space is always connected.<br />

Pro<strong>of</strong>. Let X be a path connected space. Suppose X is not connected with<br />

a separation {U, V }. Since a path in X is a continuous map, then the image <strong>of</strong><br />

any path in X (which is connected) must lie entirely in either U or V . That is we<br />

cannot join a point in U and a point in V by a path. This contradicts to the fact<br />

that X is path connected. Hence, X must be connected.<br />

□<br />

Example 3.26. In R 2 , let<br />

A = (I × {0}) ∪ ({ 1 | n > 0} × I)<br />

n<br />


44 <strong>Topology</strong> (<strong>2301631</strong>)<br />

and the comb space<br />

C = A ∪ ({0} × I).<br />

Clearly, both A and C are path connected. Moreover, since A is connected and<br />

C = A, then C is also connected. However, the deleted comb space<br />

C ∗ = C − ({0} × (0, 1))<br />

is connected (A ⊆ C ∗ ⊆ A = C) but not path connected.<br />

Exercise 3.27. The topologist’s sine curve (denoted by TSin) is a subspace <strong>of</strong><br />

R 2 defined by<br />

TSin = Cl R 2({(x, sin( 1 )) | 0 < x ≤ 1}).<br />

x<br />

Prove that TSin is connected, but not path connected.<br />

Definition 3.28. For a space X, we can define an equivalence relation (verify!)<br />

on X by x ∼ y if x and y belong to the same connected subspace <strong>of</strong> X. The<br />

equivalence classes <strong>of</strong> this relation are call the (connected) components <strong>of</strong> X.<br />

Theorem 3.29. The components <strong>of</strong> X are connected disjoint subspaces <strong>of</strong> X<br />

whose union is X and each connected subspace <strong>of</strong> X intersects only one <strong>of</strong> them.<br />

Pro<strong>of</strong>. Since ∼ is an equivalence relation, then the components <strong>of</strong> X are<br />

disjoint and their union is X. Next, we will show that each connected subspace A<br />

<strong>of</strong> X intersects only one <strong>of</strong> the components. Suppose A intersects the components<br />

C 1 and C 2 . Then there exist x 1 ∈ A ∩ C 1 and x 2 ∈ A ∩ C 2 , which implies x 1 ∼ x 2<br />

because x 1 and x 2 belong to the same connected subspace A <strong>of</strong> X. Since C 1 and<br />

C 2 are equivalence classes, we must have C 1 = C 2 .<br />

To show that the component C is connected, pick a point x 0 ∈ C. For each<br />

x ∈ C, we have x ∼ x 0 and hence there exists a connected subspace A x <strong>of</strong> X<br />

containing both x and x 0 . It is not difficult to see that C ⊆ ∪ x∈C A x. Also, since<br />

for each x ∈ C, A x is connected and x 0 ∈ A x ∩ C, then A x ⊆ C (by the above<br />

paragraph). Therefore, C = ∪ x∈C A x, and hence C is connected because x 0 ∈ A x<br />

for all x ∈ X.<br />

□<br />

Remark 3.30. The previous theorem clearly implies that if A is a connected<br />

subspace <strong>of</strong> X and C is a component <strong>of</strong> X such that A ∩ C ≠ ∅, the we must have<br />

A ⊆ C.<br />

Definition 3.31. For a space X, we can define an equivalence relation (verify!)<br />

on X by x ∼ p y if there is a path in X from x to y. The equivalence classes <strong>of</strong> this<br />

relation are called the path components <strong>of</strong> X.<br />

Theorem 3.32. The path components <strong>of</strong> X are path connected disjoint subspaces<br />

<strong>of</strong> X whose union is X and each path connected subspace <strong>of</strong> X intersects<br />

only one <strong>of</strong> them.<br />

Pro<strong>of</strong>. Similar to the previous theorem.<br />

Example 3.33. The deleted comb space C ∗ has only one component, but two<br />

path components. Moreover, C ∗ ∪ {(0, q) | q ∈ Q ∩ (0, 1)} also has one component,<br />

but countably many path components.<br />

Exercise 3.34. Prove that a component <strong>of</strong> X is always closed in X. Can we<br />

say the same thing for path component? Why?<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 45<br />

Definition 3.35. A space X is said to be locally connected at x if each neighborhood<br />

<strong>of</strong> x contains a connected neighborhood <strong>of</strong> x. We say that X is locally<br />

connected if it is locally connected at each <strong>of</strong> its points.<br />

Definition 3.36. A space X is said to be locally path connected at x if each<br />

neighborhood <strong>of</strong> x contains a path connected neighborhood <strong>of</strong> x. We say that X is<br />

locally path connected if it is locally path connected at each <strong>of</strong> its points.<br />

Example 3.37. The subspace [−1, 0) ∪ (0, 1] <strong>of</strong> R is locally connected and<br />

locally path connected, but it is neither connected nor path connected. On the<br />

other hand, TSin is neither locally connected nor locally path connected.<br />

Theorem 3.38. A space X is locally connected iff for each open subset U <strong>of</strong><br />

X, each component <strong>of</strong> U is open in X<br />

Pro<strong>of</strong>. Suppose X is locally connected. Let U be an open subset <strong>of</strong> X, C<br />

the component <strong>of</strong> U and x ∈ C. Then by local connectedness (at x), U contains a<br />

connected neighborhood V <strong>of</strong> x. Since C is a component <strong>of</strong> U and V is a connected<br />

subset <strong>of</strong> U, we must have V ⊆ C. Therefore, C is open in X.<br />

Conversely, assume that each component <strong>of</strong> an open subset <strong>of</strong> X is also open<br />

in X. Let x ∈ X, U a neighborhood <strong>of</strong> x and C the component <strong>of</strong> U containing x.<br />

By assumption, C is open in X and hence it is the desired connected neighborhood<br />

<strong>of</strong> x contained in U.<br />

□<br />

Theorem 3.39. A space X is locally path connected iff for each open subset U<br />

<strong>of</strong> X, each path component <strong>of</strong> U is open in X<br />

Pro<strong>of</strong>. Similar to the previous theorem.<br />

Remark 3.40. The previous two theorems immediately imply that a (path)<br />

component <strong>of</strong> a locally (path) connected space is always open.<br />

Theorem 3.41. A connected and locally path connected space is path connected.<br />

Pro<strong>of</strong>. Let X be a connected, locally path connected space. Fix x 0 ∈ X<br />

and let P be the path component <strong>of</strong> X containing x 0 . Then P ≠ ∅. Since X is<br />

locally path connected, the path connected component P must be open in X by the<br />

previous theorem. Moreover, if y ∈ X − P , the path component Q <strong>of</strong> X containing<br />

y must be open in X. Since y ∈ Q ⊆ X − P , the set X − P must be open in X.<br />

Now, if X − P ≠ ∅, then {P, X − P } will clearly form a separation <strong>of</strong> X which<br />

contradicts to the assumption. Therefore, X − P = ∅; i.e., P = X.<br />

□<br />

Exercise 3.42. Prove that an open connected subset <strong>of</strong> R n is always path<br />

connected.<br />

Exercise 3.43. Discuss the connectedness, path connectedness, local connectedness<br />

and local path connectedness <strong>of</strong> Q and R − Q.<br />

Definition 3.44. A space X is said to be totally disconnected if only connected<br />

subspaces <strong>of</strong> X are one-point sets.<br />

Example 3.45. Q is totally disconnected. For if a connected subspaces A <strong>of</strong><br />

Q contains distinct points p < q, we can find an irrational number s such that<br />

p < s < q and hence {A ∩ (−∞, s), A ∩ (s, ∞)} will form a separation <strong>of</strong> A which<br />

is a contradiction.<br />


CHAPTER 4<br />

Compactness<br />

Definition 4.1. Let X be a space and C a collection <strong>of</strong> subsets <strong>of</strong> X. We say<br />

that C is a covering <strong>of</strong> X (or C covers X) if ∪ C = X. A subcollection C that still<br />

covers X is called a subcovering <strong>of</strong> C. If each element <strong>of</strong> C is open in X, we will<br />

call C an open covering. Moreover, a space X is said to be compact if every open<br />

covering has a finite subcovering.<br />

Definition 4.2. When X is a subspace <strong>of</strong> Y , we define a [open] covering <strong>of</strong> X<br />

in Y to be a collection <strong>of</strong> [open] subsets <strong>of</strong> Y whose union contains X. Therefore, a<br />

(open) covering <strong>of</strong> a space X in the previous definition is simply a [open] covering<br />

<strong>of</strong> X in X.<br />

Theorem 4.3. Let X and Y be spaces such that X ⊆ Y . Then X is compact<br />

iff every open covering <strong>of</strong> X in Y has a finite subcovering.<br />

Pro<strong>of</strong>. Suppose X is compact and let C be an open covering <strong>of</strong> X in Y .<br />

It follows that the collection {A ∩ X | A ∈ C} forms an open covering <strong>of</strong> X (in<br />

X). Then, by compactness <strong>of</strong> X, it has a finite subcovering, says {A 1 ∩ X, A 2 ∩<br />

X, . . . , A n ∩ X}. Now, it is clear that {A 1 , A 2 , . . . , A n } is a finite subcovering <strong>of</strong> C.<br />

Conversely, assume that every open covering <strong>of</strong> X in Y has a finite subcovering.<br />

Let D be an open covering <strong>of</strong> X (in X). For each D ∈ D, there is an open subset U D<br />

<strong>of</strong> Y such that D = U D ∩ X. Then, by axiom <strong>of</strong> choice, we have an open covering<br />

{U D |D ∈ D} <strong>of</strong> X in Y and, by assumption, it reduces to a finite subcovering, says<br />

{U D1 , U D2 , . . . , U Dn }. Now, it is easy to verify that {D 1 , D 2 , . . . , D n } is a finite<br />

subcovering <strong>of</strong> D. Therefore, X is compact.<br />

□<br />

Example 4.4. Any finite discrete space is compact, but an infinite discrete<br />

space is not.<br />

Example 4.5. R is not compact because {(n, n+2) | n ∈ Z} is an open covering<br />

<strong>of</strong> R that does not have a finite subcovering.<br />

Example 4.6. A finite union <strong>of</strong> compact spaces is compact.<br />

Exercise 4.7. Show that { 1 n | n ∈ N} is compact, but { 1 n<br />

| n ∈ N} is not.<br />

Remark 4.8. There are some other types <strong>of</strong> compactness : a space X is called<br />

• countably compact if every countable open covering <strong>of</strong> X has a finite subcovering.<br />

• limit point compact if every infinite subset <strong>of</strong> X has a limit point.<br />

• sequentially compact if every sequence in X has a convergent subsequence.<br />

It is immediate from the definitions that compactness implies countable compactness.<br />

In general, one can show that (see [4] for details):<br />

47


48 <strong>Topology</strong> (<strong>2301631</strong>)<br />

compact<br />

⇓<br />

sequentially compact ⇒ countably compact ⇒ limit point compact<br />

When the space is metrizable, we also have the following implications :<br />

limit point compact ⇒ sequentially compact ⇒ compact<br />

Therefore, the notions <strong>of</strong> compactness above are all equivalent for metric spaces,<br />

and hence any closed interval [a, b] in R is compact since it is limit point compact<br />

by the famous Bolzano-Wierestrass theorem.<br />

Exercise 4.9. Show that the space X = N × {1, 2} indiscrete is not compact,<br />

but limit point compact.<br />

Exercise 4.10. Prove that a space X is compact iff every net in X has a<br />

convergent subnet.<br />

Theorem 4.11. A closed subspace <strong>of</strong> a compact space is compact.<br />

Pro<strong>of</strong>. Let C be a closed subspace <strong>of</strong> a compact space X and C a collection<br />

<strong>of</strong> open subsets <strong>of</strong> X such that C ⊆ ∪ C. Then the collection D = C ∪ {X − C} is<br />

clearly an open covering <strong>of</strong> X. Since X is compact, there is a finite subcollection<br />

D ′ <strong>of</strong> D such that X = ∪ D ′ . Let C ′ = D ′ − {X − C}. Clearly, C ′ is finite and<br />

∪<br />

C ′ = ∪ (D ′ − {X − C}) ⊇ ( ∪ D ′ ) − (X − C) = X − (X − C) = C<br />

Hence, C ′ is a finite subcollection <strong>of</strong> C whose union still covers C; i.e., C is compact.<br />

□<br />

Lemma 4.12. If A is a compact subspace <strong>of</strong> a Hausdorff space X and x ∈ X−A,<br />

then there exist neighborhoods U and V <strong>of</strong> x and A, respectively, such that U∩V = ∅.<br />

Pro<strong>of</strong>. Since X is Hausdorff, for each a ∈ A, there exist neighborhoods U a<br />

and V a <strong>of</strong> x and a, respectively, such that U a ∩ V a = ∅. Clearly, {V a } a∈A is an open<br />

covering <strong>of</strong> A, and since A is compact, we obtain a finite subcovering {V ai } n i=1 .<br />

Then, by letting U = ∩ n i=1 U a i<br />

and V = ∪ n i=1 V a i<br />

, it is easy to see that both are<br />

open in X. Moreover, if there is y ∈ U ∩ V , we will have y ∈ U ak ∩ V ak for some<br />

k = 1, . . . , n. This contradicts to the fact that U ak ∩ V ak = ∅. Hence, we must have<br />

U ∩ V = ∅.<br />

□<br />

Theorem 4.13. A compact subspace <strong>of</strong> a Hausdorff space is closed.<br />

Pro<strong>of</strong>. Let A be a compact subspace <strong>of</strong> a Hausdorff space X, and x /∈ A. It<br />

follows directly from the previous lemma that there is a neighborhood U <strong>of</strong> x such<br />

that U ∩ A = ∅. Therefore, A is closed in X.<br />

□<br />

Corollary 4.14. An arbitrary intersection <strong>of</strong> compact subspace <strong>of</strong> a Hausdorff<br />

space is also compact.<br />

Pro<strong>of</strong>. Let K be a collection <strong>of</strong> compact subspaces <strong>of</strong> a Hausdorff space X.<br />

Since X is Hausdorff, each K ∈ K is also closed in X by the previous theorem.<br />

Hence, ∩ K is also closed in each K ∈ K. Since each K ∈ K is compact, by<br />

Theorem 4.11, ∩ K is then compact.<br />

□<br />

Theorem 4.15. A compact Hausdorff space is normal.


<strong>Phichet</strong> <strong>Chaoha</strong> 49<br />

Pro<strong>of</strong>. Let X be a compact Hausdorff space. Let A and B be disjoint<br />

closed subsets <strong>of</strong> X. By Theorem 4.11, the sets A and B are compact. Also,<br />

by Lemma 4.12, for each a ∈ A, there are disjoint neighborhoods U a <strong>of</strong> a and V a <strong>of</strong><br />

B. Since {U a } a∈A forms an open covering <strong>of</strong> A and A is compact, there is a finite<br />

subcollection {U ai } n i=1 that covers A. By letting U = ∪n i=1 U a i<br />

and V = ∩ n i=1 V a i<br />

, it<br />

is easy to check that U and V are disjoint neighborhoods <strong>of</strong> A and B, respectively.<br />

Therefore, X is normal.<br />

□<br />

Theorem 4.16. If X is a compact space and f : X → Y is a continuous map,<br />

then f(X) is also compact.<br />

Pro<strong>of</strong>. Suppose X is a compact space and f : X → Y is a continuous map.<br />

Let {U α } α∈Λ be an open covering <strong>of</strong> f(X). Then, {f −1 (U α )} α∈Λ clearly forms<br />

an open covering <strong>of</strong> X and, by compactness, it has a finite subcovering, says<br />

{f −1 (U αi )} n i=1 . Hence, {U α i<br />

} n i=1 is the desired finite subcollection <strong>of</strong> {U α} α∈Λ<br />

that still covers f(X).<br />

□<br />

Example 4.17. A quotient space <strong>of</strong> a compact space is always compact.<br />

Corollary 4.18. Compactness is a topological property.<br />

Theorem 4.19. A bijective continuous map from a compact space onto a Hausdorff<br />

space is a homeomorphism.<br />

Pro<strong>of</strong>. To show that f −1 is continuous, let C be a closed subset <strong>of</strong> X. Since X<br />

is compact, then so is C (by Theorem 4.11). Since f is continuous, by the previous<br />

theorem, f(C) is compact in Y . Finally, since Y is Hausdorff, it follows directly by<br />

Theorem 4.13 that f(C) is closed in Y .<br />

□<br />

Example 4.20. Let C = {(x, y) : x 2 + y 2 = 1} ⊆ R 2 and f : I → C be given<br />

by f(t) = (cos 2πt, sin 2πt). Clearly, f is continuous. Also, let p : I → I /∼ = S 1 be<br />

the quotient map where ∼ is given by 0 ∼ 1. So, f is constant on each fiber <strong>of</strong> p<br />

and hence it induces a continuous map ˜f : S 1 → C such that ˜f ◦ p = f. It is easy<br />

to see that ˜f is bijective. Since S 1 is compact and C is Hausdorff, ˜f is in fact a<br />

homeomorphism; i.e., S 1 ∼ = C.<br />

Theorem 4.21 (Extreme Value Theorem). If X is a compact space and f :<br />

X → R is a continuous map, then f attains its minimum and maximum on X.<br />

Pro<strong>of</strong>. Suppose that f does not attain its minimum on X; i.e., for each<br />

z ∈ f(X), there exists y ∈ f(X) such that y < z. Then the collection {(y, ∞) | y ∈<br />

f(X)} clearly forms an open covering <strong>of</strong> f(X). Since f(X) is compact, we obtain<br />

a finite subcovering <strong>of</strong> f(X), says {(y 1 , ∞), (y 2 , ∞), . . . , (y n , ∞)}. Hence,<br />

f(X) ⊆ (y min , ∞) where y min = min{y 1 , y 2 , . . . , y n }. This is clearly a contradition<br />

since y min ∈ f(X) but y min /∈ (y min , ∞). Therefore, f must attain its minimum.<br />

Similarly, we can show that f attains its maximum as well.<br />

□<br />

Definition 4.22. Let (X, d) be a metric space, ∅ ̸= A ⊆ X and x ∈ X. We<br />

define the diameter <strong>of</strong> A, denoted by diam(A), and the distance between x and A,<br />

denoted by d(x, A), as follows :<br />

diam(A) = sup{d(a, b) | a, b ∈ A},<br />

d(x, A) = inf{d(x, a) | a ∈ A}.


50 <strong>Topology</strong> (<strong>2301631</strong>)<br />

If {d(a, b) | a, b ∈ A} is not bounded above, we simply write diam(A) = ∞. Otherwise,<br />

we will say that A is bounded.<br />

Exercise 4.23. Let A ≠ ∅ subset <strong>of</strong> a metric space (X, d). Prove that A is<br />

bounded iff A ⊆ B d (x; R) for some x ∈ X and R > 0.<br />

Exercise 4.24. Let C ≠ ∅ be a closed subset <strong>of</strong> a metric space (X, d) and<br />

x /∈ C. Prove that d(x, C) > 0.<br />

Theorem 4.25 (Lebesgue Lemma). If C is an open covering <strong>of</strong> a compact<br />

metric space (X, d), then there is δ > 0 such that, for every subset A <strong>of</strong> X with<br />

diam(A) < δ, we have A ⊆ C for some C ∈ C.<br />

Pro<strong>of</strong>. Let C be an open covering <strong>of</strong> a compact metric space (X, d). If X ∈ C,<br />

we are done. Assume that X /∈ C. Since X is compact, C has a finite subcovering,<br />

says {C 1 , C 2 , . . . , C n }. Now, define f : X → R by<br />

f(x) = 1 n<br />

n∑<br />

d(x, X − C i )<br />

i=1<br />

for each x ∈ X. Since X /∈ C, then each X − C i is nonempty and hence f is<br />

well-defined. It is not difficult to see that f is also continuous. Then f attains<br />

its minimum, says δ. Note also that for each x ∈ X, there exists k = 1, 2, . . . , n<br />

such that x ∈ C k and, since X − C k is closed, the previous exercise implies that<br />

d(x, X − C k ) > 0. It follows that f(x) > 0 for all x ∈ X. Hence, δ > 0.<br />

Finally, let A be any subset <strong>of</strong> X with diam(A) < δ, and a ∈ A. Clearly,<br />

A ⊆ B(a; δ) and hence<br />

f(a) = 1 n<br />

n∑<br />

d(a, X − C i ) ≤ d(a, X − C m ),<br />

i=1<br />

where d(a, X −C m ) = max{d(a, X −C 1 ), . . . , d(a, X −C n )} (depending on a). Since<br />

δ is the minimum <strong>of</strong> f, we have f(a) ≥ δ and hence d(a, X − C m ) ≥ δ. This implies<br />

B(a; δ) ⊆ C m and hence A ⊆ C m ∈ C as desired.<br />

□<br />

Definition 4.26. Let (X, d X ), (Y, d Y ) be metric spaces and f : X → Y a map.<br />

We say that f is uniformly continuous if for each ϵ > 0, there is δ > 0 such that<br />

for each x 1 , x 2 ∈ X, if d X (x 1 , x 2 ) < δ, we have d Y (f(x 1 ), f(x 2 )) < ϵ.<br />

Clearly, uniform continuity implies continuity. However, when X is compact,<br />

the two notions are equivalent.<br />

Theorem 4.27 (Uniform Continuity Theorem). Let (X, d X ) and (Y, d Y ) be<br />

metric spaces where X is compact. Any continuous map f : X → Y is uniformly<br />

continuous.<br />

Pro<strong>of</strong>. Let ϵ > 0. Since {B(y, ϵ 2<br />

) | y ∈ Y } is an open covering <strong>of</strong> Y and<br />

f is continuous, then C = {f −1 (B(y, ϵ 2<br />

)) | y ∈ Y } forms an open covering <strong>of</strong> X.<br />

Since X is compact, there exists a Lebesgue number δ > 0 for C. Then for any<br />

x 1 , x 2 ∈ X such that d X (x 1 , x 2 ) < δ, there exists y 0 ∈ Y such that {x 1 , x 2 } ⊆<br />

f −1 (B(y 0 , ϵ 2 )) because diam{x 1, x 2 } < δ. It follows that f(x 1 ), f(x 2 ) ∈ B(y 0 , ϵ 2 )<br />

and hence d Y (f(x 1 ), f(x 2 )) < ϵ. Therefore, f is uniformly continuous. □


<strong>Phichet</strong> <strong>Chaoha</strong> 51<br />

Lemma 4.28 (Tube Lemma). Let X, Y be spaces such that Y is compact. If N<br />

is an open subset <strong>of</strong> X × Y containing a slice {x 0 } × Y <strong>of</strong> X × Y , then there is a<br />

neighborhood U <strong>of</strong> x 0 such that U × Y ⊆ N.<br />

Pro<strong>of</strong>. Let X, Y be spaces such that Y is compact. Let N be an open subset<br />

<strong>of</strong> X × Y containing {x 0 } × Y <strong>of</strong> X × Y . For each y ∈ Y , let U y × V y be a basis<br />

neighborhood <strong>of</strong> (x 0 , y) such that U y × V y ⊆ N. Then, {U y × V y } y∈Y is an open<br />

covering <strong>of</strong> {x 0 } × Y such that ∪ y∈Y (U y × V y ) ⊆ N. Since Y is compact, then<br />

so is {x 0 } × Y , and hence there is a finite subcollection {U yi × V yi } n i=1 that covers<br />

{x 0 } × Y . Let U = ∩ n<br />

i=1 U y i<br />

, which is clearly a neighborhood <strong>of</strong> x 0 . Therefore, we<br />

have {x 0 } × Y ⊆ U × Y = U × ( ∪ n<br />

i=1 V y i<br />

) = ∪ n<br />

i=1 (U × V y i<br />

) ⊆ ∪ n<br />

∪<br />

i=1 (U y i<br />

× V yi ) ⊆<br />

y∈Y (U y × V y ) ⊆ N.<br />

□<br />

Theorem 4.29. A finite product <strong>of</strong> compact spaces is compact.<br />

Pro<strong>of</strong>. By using the induction on the number <strong>of</strong> the factors, it suffices to<br />

prove for a product <strong>of</strong> two compact spaces X and Y . Let C be an open covering <strong>of</strong><br />

X × Y . Then, for each x ∈ X, since Y is compact, there is a finite subcollection<br />

C<br />

∪x f <strong>of</strong> C that covers {x} × Y . Now, by applying the tube lemma to the open set<br />

C<br />

f<br />

x containing {x} × Y , we get a neighborhood U x <strong>of</strong> x such that U x × Y ⊆ ∪ Cx.<br />

f<br />

Since X is compact, we can reduce the covering {U x } x∈X to a finite subcollection<br />

{U xi } n i=1 that still covers X. Now, ∪ n<br />

i=1 Cf x i<br />

is a finite subcollection <strong>of</strong> C that covers<br />

X × Y since<br />

n∪<br />

n∪<br />

n∪<br />

X × Y ⊆ ( U xi ) × Y = (U xi × Y ) ⊆ ( ∪ Cx f i<br />

) = ∪ n∪<br />

( Cx f i<br />

).<br />

i=1<br />

Therefore, X × Y is compact.<br />

i=1<br />

i=1<br />

Example 4.30. A cube [a, b] n is always compact.<br />

Lemma 4.31. Let X be a subspace <strong>of</strong> a metric space (Y, d). If X is compact,<br />

then X is closed and bounded.<br />

Pro<strong>of</strong>. Let X be a compact subspace <strong>of</strong> (Y, d) and fix y 0 ∈ Y . Then X is<br />

closed by Theorem 4.13. Since {B d (y 0 ; r) | r > 0} is an open covering <strong>of</strong> X, we then<br />

have X ⊆ B(y 0 ; R) for some R > 0 by compactness. Hence, X is also bounded. □<br />

Theorem 4.32. A subspace X <strong>of</strong> R n is compact iff X is closed and bounded<br />

(with respect to the euclidean metric on R n ).<br />

Pro<strong>of</strong>. Let X be a subspace <strong>of</strong> R n .<br />

(⇒) Follows directly from the previous lemma.<br />

(⇐) Suppose X is closed and bounded. Then X ⊆ B(x 0 ; R) for some x 0 ∈ R n<br />

and R > 0. WLOG, we may assume that x 0 = 0 (otherwise, we may replace R<br />

by R + d(x 0 , 0)). Since B(0; R) ⊆ [−R, R] n , we have X ⊆ [−R, R] n . Since the<br />

cube [−R, R] n is compact by the previous theorem and X is closed, it follows from<br />

Theorem 4.11 that X is compact.<br />

□<br />

Example 4.33. A closed ball B(x; R) and a sphere ∂B(x; R) in R n are compact.<br />

In particular D n and S n are compact for each n ≥ 0.<br />

Definition 4.34. Let X be a set and C a collection <strong>of</strong> subsets <strong>of</strong> X. We say<br />

that C has the finite intersection property if every finite subcollection <strong>of</strong> C has a<br />

nonempty intersection.<br />

i=1<br />


52 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Example 4.35. The collection {[−n, n] | n ∈ N} <strong>of</strong> closed subsets <strong>of</strong> R has the<br />

finite intersection property and so does the collection {[n, ∞) | n ∈ N}.<br />

Theorem 4.36. A space X is compact if and only if every collection <strong>of</strong> closed<br />

subsets <strong>of</strong> X having the finite intersection property has a nonempty intersection.<br />

Pro<strong>of</strong>. (⇒) Assume that X is a compact space. Let C be a collection <strong>of</strong><br />

closed subsets <strong>of</strong> X having the finite intersection property. For each C ∈ C, let<br />

A C = X − C which is open in X. Then, by DeMorgan’s law, we have<br />

∩<br />

A C .<br />

C∈C<br />

C = ∩ C∈C(X − A C ) = X − ∪ C∈C<br />

If ∩ C∈C C = ∅, the collection {A C} C∈C will then form an open covering <strong>of</strong> X and it<br />

can be reduced to a finite subcovering {A Ci } n i=1 by compactness <strong>of</strong> X. This implies<br />

(again, by DeMorgan’s law) ∩ n<br />

i=1 C i = ∅, which contradicts to the fact that C has<br />

the finite intersection property.<br />

(⇐) Assume that every collection <strong>of</strong> closed subsets <strong>of</strong> X having the finite intersection<br />

property has a nonempty intersection. Let {U α } α∈Λ be an open covering<br />

<strong>of</strong> X. It follows that the collection C = {X − U α } α∈Λ <strong>of</strong> closed subsets <strong>of</strong> X has<br />

an empty intersection and hence it does not have the finite intersection property.<br />

Then, there exists a finite subcollection {X − U αi } n i=1 (<strong>of</strong> C) whose intersection is<br />

empty. Hence, by DeMorgan’s law, {U αi } n i=1 is the desired finite subcollection <strong>of</strong><br />

{U α } α∈Λ that still covers X. □<br />

Example 4.37. From the previous example, it is easy to see that ∩ {[−n, n] | n ∈<br />

N} ̸= ∅ while ∩ {[n, ∞) | n ∈ N} = ∅. Hence, R is not compact.<br />

Exercise 4.38. Prove that a space X is compact if and only if, for every<br />

collection A <strong>of</strong> subsets <strong>of</strong> X having the finite intersection property, ∩ A∈A A ≠ ∅.<br />

Lemma 4.39. Let X be a set and A a collection <strong>of</strong> subsets <strong>of</strong> X having the<br />

finite intersection property, then there exists a maximal collection D <strong>of</strong> subsets <strong>of</strong><br />

X with respect to the finite intersection property such that A ⊆ D.<br />

Pro<strong>of</strong>. Consider the set<br />

F = {B | A ⊆ B ⊆ P(X) and B has the finite intersection property}<br />

which is partially ordered by set inclusion. Clearly, F ≠ ∅. Moreover, for each<br />

chain C ⊆ F, we clearly have A ⊆ ∪ C ⊆ P(X) and ∪ C has the finite intersection<br />

property. For if {C i } n i=1 is a finite subcollection <strong>of</strong> ∪ C, then for each i, there exists<br />

B i ∈ C so that C i ∈ B i and hence {C i } n i=1 ⊆ ∪ n<br />

i=1 B i. Since ∪ n<br />

i=1 B i = B k ∈ C<br />

for some k ∈ {1, . . . , n} and B k have the finite intersection property, we must have<br />

∩ {Ci } n i=1 = ∩ n<br />

i=1 C i ≠ ∅. Therefore, ∪ C is an upper bound <strong>of</strong> C in F and the<br />

existence <strong>of</strong> D follows directly from Zorn’s lemma.<br />

□<br />

Exercise 4.40. Let X be a set and D a maximal collection <strong>of</strong> subsets <strong>of</strong> X<br />

with respect to the finite intersection property. Prove that any finite intersections<br />

<strong>of</strong> elements <strong>of</strong> D belongs to D and any S ⊆ X that intersects every elements <strong>of</strong> D<br />

belongs to D.<br />

Theorem 4.41 (Tychon<strong>of</strong>f’s Theorem). An arbitrary product <strong>of</strong> compact spaces<br />

is compact.


<strong>Phichet</strong> <strong>Chaoha</strong> 53<br />

Pro<strong>of</strong>. Let X = ∏ α∈Λ X α where X α is compact, and A a collection <strong>of</strong> subsets<br />

∩<br />

<strong>of</strong> X having the finite intersection property. By Exercise 4.38, we need show that<br />

A∈A<br />

A ≠ ∅. According to the previous lemma, let D be a maximal collection <strong>of</strong><br />

subsets <strong>of</strong> X with respect to the finite intersection property such that A ⊆ D. The<br />

result will then follow once we can show that ∩ D∈D D ≠ ∅.<br />

Let us fix α ∈ Λ for a moment and consider {π α (D)|D ∈ D}, where π α :<br />

X → X α is the projection map. This collection clearly has the finite intersection<br />

property since D does. Then, by compactness <strong>of</strong> X α and Exercise 4.38, there exists<br />

x α ∈ ∩ D∈D π α(D). Note also that for each D ∈ D and a neighborhood U α <strong>of</strong> x α ,<br />

since U α ∩ π α (D) ≠ ∅, then there is y ∈ D such that π α (y) ∈ U α ∩ π α (D) and hence<br />

y ∈ πα<br />

−1 (U α ) ∩ D. In other words, πα<br />

−1 (U α ) intersects every element <strong>of</strong> D. Hence,<br />

the previous exercise implies that πα −1 (U α ) ∈ D.<br />

Now, let x = (x α ) α∈Λ and a basis neighborhood U = ∩ n<br />

i=1 π−1 α i<br />

(U αi ) <strong>of</strong> x.<br />

Since U is a finite intersection <strong>of</strong> elements <strong>of</strong> D, we have U ∈ D by the previous<br />

exercise. Since D has the finite intersection property, then U ∩ D ≠ ∅ for each<br />

D ∈ D. It follows that x ∈ D for each D ∈ D, and hence x ∈ ∩ D∈D<br />

D. Therefore,<br />

D ≠ ∅ as desired.<br />

□<br />

∩<br />

D∈D<br />

Definition 4.42. A space X is said to be locally compact at x if there is a<br />

compact subset <strong>of</strong> X containing a neighborhood <strong>of</strong> x. If X is locally compact at<br />

each <strong>of</strong> its points, we simply say that X is locally compact.<br />

Remark 4.43. Local compactness is a topological property, and a compact<br />

space is always locally compact.<br />

Example 4.44. R n is locally compact.<br />

Exercise 4.45. Show that Q is not locally compact.<br />

Definition 4.46. A compactification <strong>of</strong> a space X is a pair ( ̂X, ι) where ̂X is<br />

a compact Hausdorff space and ι is an imbedding <strong>of</strong> X into ̂X as a dense subspace<br />

(i.e., ι(X) = ̂X).<br />

Remark 4.47. When there is no ambiguity, we may identify X with the subspace<br />

ι(X), and simply regard ̂X as a compactifaction <strong>of</strong> X without specifying<br />

ι. Moreover, if a space X has a compactification, it must be at least completely<br />

regular since a subspace <strong>of</strong> a completely regular space is always completely regular<br />

(but not true for a normal space!).<br />

Definition 4.48. A compactification ( ̂X, ι) <strong>of</strong> a space X is said to be a onepoint<br />

compactification if ̂X − ι(X) is a singleton.<br />

Example 4.49. ([0, 1], (0, 1) ↩→ [0, 1]) is a compactification <strong>of</strong> (0, 1) while<br />

(S 1 , f| (0,1) ), as in Example 1.78, is a one-point compactification <strong>of</strong> (0, 1).<br />

Theorem 4.50. Let X be a non-compact space. Then X has a one-point compactification<br />

iff it is locally compact Hausdorff.<br />

Pro<strong>of</strong>. (⇒) Let ( ̂X, ι) be a one-point compactification <strong>of</strong> X, says ̂X − ι(X) =<br />

{∞}. Since ̂X is Hausdorff, then so is the subspace ι(X). Let x ∈ ι(X). Since<br />

̂X is Hausdorff, there exist disjoint neighborhoods U and V <strong>of</strong> x and ∞ in ̂X,<br />

respectively. Then ̂X − V is closed in ̂X, and hence compact. Since ̂X − V ⊆ ι(X),


54 <strong>Topology</strong> (<strong>2301631</strong>)<br />

then ̂X − V is also a compact subset <strong>of</strong> ι(X) containing the neighborhood U <strong>of</strong> x;<br />

i.e., ι(X) is locally compact at x. Therefore, X ∼ = ι(X) is locally compact Hausdorff.<br />

(⇐) Let ̂X = X ∪ {∞}. We will call a subset U <strong>of</strong> ̂X open if U is an open<br />

subset <strong>of</strong> X or U = ̂X − K for some compact subset K <strong>of</strong> X. It is easy to see<br />

that the definition <strong>of</strong> open sets above in fact gives us a topology on ̂X since a finite<br />

union and an arbitrary intersection (in a Hausdorff space) <strong>of</strong> compact sets are also<br />

compact. Moreover, X is dense in ̂X because a neighborhood <strong>of</strong> ∞ must be <strong>of</strong> the<br />

form ̂X − K and hence intersects X (since X itself is not compact!). To show that<br />

̂X is Hausdorff, let x and y be distinct points in ̂X. Since X itself is Hausdorff,<br />

we may assume that x ∈ X and y = ∞. By local compactness at x, there is a<br />

compact subset K <strong>of</strong> X containing a neighborhood U <strong>of</strong> x. Then V = ̂X − K is<br />

clearly a neighborhood <strong>of</strong> ∞ that is disjoint from U. For the compactness <strong>of</strong> ̂X,<br />

we let C be an open covering <strong>of</strong> ̂X. Then there is an element A <strong>of</strong> C containing ∞.<br />

Since A is open in ̂X, it must be <strong>of</strong> the form ̂X − K for some compact subset K <strong>of</strong><br />

X. Since C also covers K, there exists a finite subcollection {C 1 , C 2 , . . . , C n } that<br />

still covers K. Now, it is easy to see that {A, C 1 , C 2 , . . . , C n } is the desired finite<br />

subcollection <strong>of</strong> C that covers ̂X. Therefore, ̂X is compact, and ( ̂X, X ↩→ ̂X) is a<br />

one-point compactification <strong>of</strong> X.<br />

□<br />

Definition 4.51. Two compactifications ( ̂X, ι) and ( ̂X ′ , ι ′ ) are said to be<br />

homeomorphic if there is a homeomorphism h : ̂X → ̂X′ that makes the following<br />

diagram commutative :<br />

̂X<br />

♣ ♣ ♣ ♣ ♣ ♣<br />

h♣ ♣ ♣ ♣ ♣✲♣ ♣ ̂X′<br />

❅■ ι ❅<br />

X.<br />

<br />

✒<br />

ι ′<br />

Theorem 4.52. A one-point compactification <strong>of</strong> a space X is unique up to a<br />

homeomorphism.<br />

Pro<strong>of</strong>. Suppose ( ̂X, ι) and ( ̂X ′ , ι ′ ) are one-point compactifications <strong>of</strong> X, says<br />

̂X − ι(X) = {∞} and ̂X ′ − ι ′ (X) = {ξ}. Define h : ̂X → ̂X′ by h(∞) = ξ and<br />

h(x) = ι ′ ◦ ι −1 (x) for all x ∈ ι(X) . Clearly, h is a bijection and ι ′ = h ◦ ι. To<br />

see that h is a homeomorphism, it suffices to show that h is continuous since a<br />

compactification is always compact Hausdorff. Let U be an open subset <strong>of</strong> ̂X′ .<br />

If ξ /∈ U, then h −1 (U) = ι ◦ (ι ′ ) −1 (U) is also an open subset <strong>of</strong> ι(X). Since<br />

ι(X) = ̂X − {∞} is open in ̂X, h −1 (U) is open in ̂X. Otherwise, if ξ ∈ U, then<br />

K = ̂X ′ − U ⊆ ι ′ (X) is closed in ̂X ′ and hence compact. It follows that h −1 (K) is<br />

compact and hence closed in ̂X. Therefore, h −1 (U) = h −1 ( ̂X ′ − K) = ̂X − h −1 (K)<br />

is open in ̂X, and h is continuous.<br />

□<br />

Remark 4.53. Since a one-point compactification a space X is unique up to a<br />

homeomorphism, we usually call it ”the” one-point compactification <strong>of</strong> X.<br />

Exercise 4.54. Find the one-point compactifications <strong>of</strong> N and R 2 .<br />

your answers.<br />

Justify<br />

Theorem 4.55. Let X be a Hausdorff space. Then X is locally compact iff for<br />

each x ∈ X and a neighborhood U <strong>of</strong> x, there is a neighborhood V <strong>of</strong> x such that V<br />

is compact and V ⊆ U.


<strong>Phichet</strong> <strong>Chaoha</strong> 55<br />

Pro<strong>of</strong>. Let X be a Hausdorff space.<br />

(⇒) Suppose X is locally compact. Let x ∈ X and U a neighborhood <strong>of</strong> x<br />

(in X). If X itself is compact, let ̂X = X. Otherwise, let ̂X be the one-point<br />

compactification <strong>of</strong> X (by identifying X with ι(X)). Notice that X is open in ̂X<br />

for both cases. Thus, U is also open in ̂X. Since ̂X is regular (it is even normal by<br />

Theorem 4.15), there is a neighborhood V <strong>of</strong> x in ̂X such that x ∈ V ⊆ Cl ̂X<br />

V ⊆ U.<br />

Being closed in ̂X, Cl ̂X<br />

V is clearly compact. Moreover, since U ⊆ X, we have<br />

U ∩ X = U, V ∩ X = V and V = Cl ̂X<br />

V ∩ X = Cl ̂X<br />

V . Therefore, x ∈ V ⊆ V ⊆ U<br />

and V is compact as desired.<br />

(⇐) Follows directly from the definition.<br />

□<br />

Theorem 4.56 (Imbedding Theorem). If X is a T 0 -space and Λ is a collection<br />

<strong>of</strong> continuous maps from X to [0, 1] satisfying that, for each x ∈ X and a neighborhood<br />

U <strong>of</strong> x, there exists f ∈ Λ such that f(x) > 0 and X − U ⊆ f −1 ({0}) (i.e.,<br />

f vanishes outside U), then the map F : X → [0, 1] Λ defined by<br />

is an imbedding <strong>of</strong> X into [0, 1] Λ .<br />

F (x) = (f(x)) f∈Λ<br />

Pro<strong>of</strong>. Clearly, F is continuous since each f ∈ Λ is continuous. To show<br />

that F is injective, suppose x ≠ y. Since X is T 0 , we may assume that there is a<br />

neighborhood U <strong>of</strong> x not containing y. By the requirement above, there is f 0 ∈ Λ<br />

such that f 0 (x) > 0 while f 0 (y) = 0. Hence, F (x) ≠ F (y). Now, we need to show<br />

that the bijection F : X → F (X) is a homeomorphism. Since F : X → F (X) is<br />

continuous, we only have show that F is open. Let U be a nonempty open subset<br />

<strong>of</strong> X, y ∈ F (U) and pick x ∈ U such that F (x) = y. Since U is a neighborhood<br />

<strong>of</strong> x, there is f 0 ∈ Λ such that f 0 (x) > 0 and f 0 (X − U) = {0}. By letting<br />

W = F (X) ∩ π −1<br />

f 0<br />

((0, 1]) ⊆ [0, 1] Λ where π f0 : [0, 1] Λ → [0, 1] is the projection<br />

map, W is clearly open in F (X) and y = F (x) ∈ W . To see that W ⊆ F (U),<br />

let w ∈ W . Then w = F (a) for some a ∈ X and π f0 (w) > 0. It follows that<br />

0 < π f0 (w) = π f0 (F (a)) = f 0 (a), and hence, a ∈ U (since f vanishes outside U).<br />

Therefore, w ∈ F (U) as desired.<br />

□<br />

Remark 4.57. The collection Λ having the requirement as above is usually<br />

said to separate a point from a closed set not containing it.<br />

Corollary 4.58. A space X is completely regular iff it can be imbedded into<br />

[0, 1] Λ for some Λ.<br />

Pro<strong>of</strong>. (⇒) Assume that X is completely regular. Let Λ be the collection <strong>of</strong><br />

all continuous maps from X to [0, 1]. Let x ∈ X and U a neighborhood <strong>of</strong> x. Then<br />

by complete regularity <strong>of</strong> X, there exists a continuous map f : X → [0, 1] such that<br />

f(x) = 0 and X − U ⊆ f −1 ({1}). Clearly, 1 − f ∈ Λ, (1 − f)(x) = 1 > 0 and<br />

X − U ⊆ (1 − f) −1 ({0}). Therefore, Λ satisfies the requirement in the imbedding<br />

theorem, and hence, X can be embedded into [0, 1] Λ .<br />

(⇐) Since [0, 1] Λ is compact Hausdorff, then it is normal and hence completely<br />

regular. It follows that X is completely regular because it is homeomorphic to a<br />

subspace <strong>of</strong> [0, 1] Λ .<br />


56 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Exercise 4.59. If X is a completely regular space and F is a collection <strong>of</strong><br />

all bounded continuous maps from X to R, prove that X can be imbedded into<br />

∏<br />

f∈F [a f , b f ], where f(X) ⊆ [a f , b f ] for each f ∈ F.<br />

For an imbedding h : X → K <strong>of</strong> X into a compact Hausdorff space K, the<br />

subspace h(X) <strong>of</strong> K is clearly compact (because it is closed in the compact space<br />

K) and Hausdorff (because it is a subspace <strong>of</strong> a Hausdorff space K). Moreover,<br />

h(X) contains h(X) as a dense subspace. Therefore, (h(X), h) is a compactification<br />

<strong>of</strong> X, and we call it the compactification induced by h. The above corollary together<br />

with the Tychon<strong>of</strong>f’s theorem now imply :<br />

Corollary 4.60. If X is a completely regular space, then X has a compactification.<br />

When a space X has a compactification ( ̂X, ι), it is natural to ask whether a<br />

given continuous map f : X → R can be extended to a continuous map F : ̂X → R<br />

in the sense that the following diagram commutes :<br />

X<br />

❅ ❅❘ ι<br />

f<br />

̂X<br />

✲ R<br />

♣ ♣ ♣ ♣<br />

✒ F<br />

This is certainly not true for an arbitrary map f. In fact, f must be bounded<br />

since the domain <strong>of</strong> F is compact. However, sin( 1 x<br />

) : (0, 1) → R is bounded, but it<br />

can not be extended to either compactification S 1 or [0, 1] <strong>of</strong> (0, 1). The following<br />

examples give us other compactifications <strong>of</strong> (0, 1) such that the map sin( 1 x<br />

) can be<br />

extended.<br />

Example 4.61. The compactification <strong>of</strong> (0, 1) induced by the imbedding<br />

h(x) = (x, sin( 1 )) ∈ [0, 1] × [−1, 1]<br />

x<br />

is clearly TSin. Note also that the map sin( 1 x<br />

) : (0, 1) → R can be extended to<br />

TSin as follows :<br />

sin( 1 x<br />

(0, 1) ) ✲ R<br />

❅<br />

h❅ ❅❘ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣<br />

✒<br />

π 2<br />

TSin<br />

Example 4.62. Let ̂X be a compactification induced by the imbedding<br />

h(x) = (x, sin( 1 x ), cos( 1 x ))<br />

<strong>of</strong> (0, 1) into [0, 1] × [−1, 1] 2 . Then, the maps x, sin( 1 x ) and cos( 1 x<br />

) can all be<br />

extended to ̂X because <strong>of</strong> the following commutative diagrams :<br />

sin( 1 x )<br />

cos( 1 x )<br />

x<br />

(0, 1) ✲ R (0, 1) ✲ R (0, 1) ✲ R<br />

❅ ❅ ❅<br />

h ❅❘ ♣ ♣ ♣ ♣ ♣ ♣<br />

✒<br />

π1<br />

h ❅❘ ♣ ♣ ♣ ♣ ♣ ♣<br />

✒<br />

π2<br />

h ❅❘ ♣ ♣ ♣ ♣ ♣ ♣<br />

✒<br />

π3<br />

̂X ̂X ̂X


<strong>Phichet</strong> <strong>Chaoha</strong> 57<br />

Exercise 4.63. Let f : D → Y be a continuous map, ι : D → X an imbedding<br />

such that ι(D) = X and Y a Hausdorff space. If F is a continuous extension <strong>of</strong> f<br />

in the sense that the following diagram commutes :<br />

f<br />

D ✲ Y<br />

then F must be unique.<br />

❅ι<br />

❅❘<br />

✒ F<br />

X,<br />

Theorem 4.64. Let X be a completely regular space. There exists a compactification<br />

( ̂X, ι) <strong>of</strong> X such that every bounded continuous map f : X → R extends to<br />

a unique continuous map from ̂X to R.<br />

Pro<strong>of</strong>. Let F be the set <strong>of</strong> all bounded continuous maps from X to R.<br />

By Exercise 4.59, we obtain an imbedding <strong>of</strong> X into a compact Hausdorff space<br />

∏<br />

f∈F [a f , b f ], says h : X → ∏ f∈F [a f , b f ]. Let ( ̂X = h(X), ι = h) be the compactification<br />

<strong>of</strong> X induced by h. Then, for each f 0 ∈ F, the projection π f0 : ̂X → [a f0 , b f0 ]<br />

is clearly the unique by the previous exercise (since h(X) is dense in ̂X and R is<br />

Hausdorff) continuous extension <strong>of</strong> f 0 to ̂X.<br />

□<br />

Lemma 4.65. Let X be a completely regular space and let ( ̂X, ι) be a compactification<br />

<strong>of</strong> X satisfying the extension property in Theorem 4.64. Then, any<br />

continuous map f from X to a compact Hausdorff space K extends uniquely and<br />

continuously to ̂X.<br />

Pro<strong>of</strong>. Let K be a compact Hausdorff space and f : X → K a continuous<br />

map. Then we can imbed K (since K is completely regular) into the product<br />

space [0, 1] Λ , for some Λ, via h : K ∼= → h(K) ⊆ [0, 1] Λ . Now, for each α ∈ Λ, the<br />

composition π α ◦h◦f : X → [0, 1] is bounded and can be continuously extended to ̂X<br />

by the extension property, says to g α : ̂X → [0, 1]. It is not hard to check that we get<br />

the continuous map G = (g α ) α∈Λ : ̂X → [0, 1] Λ which extends h ◦ f : X → [0, 1] Λ .<br />

Moreover, by the continuity <strong>of</strong> G, we have<br />

G( ̂X) = G(ι(X)) ⊆ G(ι(X)) = h ◦ f(X) ⊆ h(K) = h(K).<br />

So, G really maps X into the subspace h(K) <strong>of</strong> [0, 1] Λ . Therefore, we can define<br />

F = h −1 ◦ G : ̂X → K which is the desired extension <strong>of</strong> f.<br />

f<br />

X ✲ K h ✲ [0, 1] Λ<br />

ι<br />

π α<br />

❄<br />

❄<br />

̂X ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣✯♣ ♣<br />

G<br />

♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣<br />

✒<br />

F<br />

♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣✲♣ ♣ [0, 1]<br />

g α<br />

As usual, the uniqueness <strong>of</strong> F follows directly from the fact that K is Hausdorff.<br />

□<br />

Theorem 4.66. For a completely regular space X, any two compactifications<br />

<strong>of</strong> X satisfying the extension property in Theorem 4.64 are homeomorphic.


58 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Let ( ̂X, ι) and ( ̂X ′ , ι ′ ) be two such compactifications <strong>of</strong> X. Then, by<br />

the previous lemma, ι : X → ̂X and ι ′ : X → ̂X ′ extend continuously to f : ̂X ′ → ̂X<br />

and g : ̂X → ̂X ′ , respectively, as follows :<br />

ι<br />

X ✲ ̂X X ✲ ̂X′<br />

❅ ❅<br />

ι<br />

′ ❅❘ ♣ ♣ ♣ ♣<br />

✒ f<br />

ι ❅❘ ♣ ♣ ♣ ♣<br />

✒ g<br />

̂X ′<br />

̂X<br />

Then, ι = f ◦g ◦ι and ι ′ = g ◦f ◦ι ′ ; i.e., f ◦g and g ◦f are also continuous extensions<br />

<strong>of</strong> ι and ι ′ , respectively.<br />

ι<br />

X ✲ ̂X X ✲ ̂X′<br />

ι ′<br />

ι ′<br />

❅<br />

ι ❅❘<br />

̂X<br />

✒ f◦g<br />

❅ ❅❘ ι<br />

′<br />

̂X′<br />

✒ g◦f<br />

As usual, since both ̂X and ̂X ′ are Hausdorff, f ◦ g and g ◦ f must be unique.<br />

Therefore, we must have f ◦ g = id ̂X<br />

and g ◦ f = id ̂X′; i.e., f and g are homeomorphisms.<br />

□<br />

Definition 4.67. The Stone- ˘Cech compactification β(X) <strong>of</strong> X is the (up to a<br />

homeomorphism) is the compactification satisfying the extension property in Theorem<br />

4.64.<br />

Exercise 4.68. Let X be a completely regular space. Show that X is connected<br />

iff β(X) is connected.


CHAPTER 5<br />

Function Spaces<br />

For spaces X and Y , let Y X denote the set <strong>of</strong> all maps from X to Y , and<br />

C[X, Y ] the set <strong>of</strong> all continuous maps from X to Y . In this chapter, we will<br />

introduce some topologies on Y X .<br />

Definition 5.1. Let X and Y be spaces. For each x ∈ X and an open subset<br />

U <strong>of</strong> Y , let<br />

S(x, U) = {f ∈ Y X | f(x) ∈ U}.<br />

The point-open topology on Y X is the topology generated by the sets S(x, U) as a<br />

subbasis.<br />

Remark 5.2. Since a basis element is a finite intersection <strong>of</strong> subbasis elements,<br />

then a basis neighborhood (w.r.t. the point-open topology) <strong>of</strong> f ∈ Y X consists <strong>of</strong><br />

g ∈ Y X that is ”close” to f at finitely many points. The point-open topology<br />

is sometimes called the topology <strong>of</strong> pointwise convergence because <strong>of</strong> the following<br />

exercise.<br />

Exercise 5.3. Prove that a sequence (f n ) in Y X converges to f ∈ Y X in the<br />

point-open topology iff for each x ∈ X, (f n (x)) converges to f(x) in Y .<br />

Example 5.4. Let I = [0, 1]. In the topology <strong>of</strong> pointwise convergence, the<br />

subspace C[I, R] <strong>of</strong> R I is not closed. This is because the sequence (f n ) in C[I, R],<br />

where f n (x) = x n , converges pointwise to the map<br />

{<br />

0 for 0 ≤ x < 1,<br />

f(x) =<br />

1 for x = 1.<br />

which is not in C[I, R].<br />

Definition 5.5. Let X and Y be spaces. For a compact subset K <strong>of</strong> X and<br />

an open subset U <strong>of</strong> Y , let<br />

S(K, U) = {f ∈ Y X | f(K) ⊆ U}.<br />

The compact-open topology on Y X is the topology generated by the sets S(K, U)<br />

as a subbasis.<br />

Remark 5.6. It is clear from the definitions above that the compact-open<br />

topology is finer than the point-open topology. Even though the compact-open<br />

topology can be defined on Y X , we usually consider this topology on the subset<br />

C[X, Y ].<br />

Theorem 5.7. Let X and Y be spaces where X is locally compact Hausdorff.<br />

If C[X, Y ] is given the compact-open topology, then the evaluation map ev : X ×<br />

C[X, Y ] → Y , defined by ev(x, f) = f(x), is continuous.<br />

59


60 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Let (x, f) ∈ X × C[X, Y ], we will show that ev is continuous at (x, f).<br />

Let V be a neighborhood <strong>of</strong> f(x). Since f is continuous, f −1 (V ) is a neighborhood<br />

<strong>of</strong> x. Since X is locally compact Hausdorff, there is a neighborhood U <strong>of</strong> x such<br />

that x ∈ U ⊆ U ⊆ f −1 (V ) and U is compact. Clearly, f(U) ⊆ f(f −1 (V )) ⊆ V , and<br />

hence W = U × S(U, V ) is a neighborhood <strong>of</strong> (x, f). Moreover, for (x ′ , f ′ ) ∈ W ,<br />

we have ev(x ′ , f ′ ) = f ′ (x ′ ) ∈ V ; i.e, W ⊆ ev −1 (V ).<br />

□<br />

Exercise 5.8. Let X, Y, Z be spaces where Y is locally compact Hausdorff.<br />

Prove that the composition<br />

C[X, Y ] × C[Y, Z] → C[X, Z]<br />

is continuous when the compact-open topology is used throughout.<br />

as sets.<br />

Lemma 5.9 (Exponential Law). For any sets X, Y and Z, we have<br />

Z X×Y ∼ = (Z Y ) X<br />

Pro<strong>of</strong>. Define ϕ : Z X×Y → (Z Y ) X by<br />

for each f ∈ Z X×Y , x ∈ X and y ∈ Y .<br />

Also, define ψ : (Z Y ) X → Z X×Y by<br />

ϕ(f)(x)(y) = f(x, y)<br />

ψ(g)(x, y) = g(x)(y)<br />

for each g ∈ (Z Y ) X , x ∈ X and y ∈ Y .<br />

It is easy to check that ϕ is bijective with ϕ −1 = ψ.<br />

Exercise 5.10. Let X, Y and Z be spaces, and give C[Y, Z] the compact-open<br />

topology. If f : X × Y → Z is continuous, then ϕ(f) in the above lemma is a<br />

continuous map from X to C[Y, Z].<br />

Theorem 5.11 (Adjunction Formula). Let X, Y, Z be spaces where Y is locally<br />

compact Hausdorff. If C[Y, Z] has the compact-open topology, then the set <strong>of</strong> all<br />

continuous maps from X to C[Y, Z] is isomorphic to the set <strong>of</strong> all continuous maps<br />

from X × Y to Z; i.e.,<br />

as sets.<br />

C[X × Y, Z] ∼ = C[X, C[Y, Z]]<br />

Pro<strong>of</strong>. From the previous exercise, we see that ϕ decends to the function<br />

ϕ : C[X × Y, Z] → C[X, C[Y, Z]].<br />

Then, the result will follow once we can show that ψ also decends to the function<br />

ψ : C[X, C[Y, Z]] → C[X × Y, Z]<br />

which is the inverse <strong>of</strong> ϕ. However, for each g ∈ C[X, C[Y, Z]], it is not difficult to<br />

verify that ψ(g) is the composition<br />

X × Y g×1 Y<br />

→ C[Y, Z] × Y → ev Z.<br />

Since Y is locally compact Hausdorff and C[Y, Z] is given the compact-open topology,<br />

the evaluation map ev is continuous and hence ψ(g) ∈ C[X × Y, Z] as desired.<br />

□<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 61<br />

Exercise 5.12. Let X, Y, Z be spaces where X is Hausdorff. If the compactopen<br />

topology is used throughout, then<br />

C[X, Y × Z] ∼ = C[X, Y ] × C[X, Z]<br />

(as spaces).<br />

When Y is a metric space, we can define the uniform topology on Y X as follows.<br />

Definition 5.13. Let X be a space (or even a set), (Y, d) a metric space and<br />

d be the standard bounded metric (i.e., d = min{d, 1}). Define<br />

ρ(f, g) = sup{d(f(x), g(x)) | x ∈ X}<br />

for f, g ∈ Y X . It is easy to check that ρ is also a metric on Y X , and it is called the<br />

uniform metric corresponding to the metric d. The topology on Y X induced by ρ<br />

is then called the uniform topology.<br />

If F is a subset <strong>of</strong> Y X such that the set {d(f(x), g(x)) | x ∈ X} is bounded for<br />

any f, g ∈ F, we can also define another metric on F by<br />

ρ(f, g) = sup{d(f(x), g(x)) | x ∈ X}.<br />

(For examples, when F = B[X, Y ] - the set <strong>of</strong> all bounded map from X to Y ,<br />

or when F = C[X, Y ] when X is compact) This metric is called the sup metric<br />

corresponding to the metric d. Under this circumstance, one can show that ρ is in<br />

fact the standard bounded metric <strong>of</strong> ρ; i.e., ρ = min{ρ, 1}, and hence both metric<br />

are equivalent.<br />

Definition 5.14. Let X be a space, (Y, d) a metric space and (f n ) a sequence<br />

in Y X . We say that (f n ) converges uniformly to f : X → Y if for each ϵ > 0, there<br />

is N ∈ N such that<br />

d(f n (x), f(x)) < ϵ<br />

for all n ≥ N and x ∈ X.<br />

Exercise 5.15. Let X be a space and (Y, d) a metric space. Prove that a<br />

sequence (f n ) in Y X converges uniformly to f ∈ Y X iff (f n ) converges to f in the<br />

uniform topology.<br />

Theorem 5.16 (Uniform Limit Theorem). Let X be a space and (Y, d) a metric<br />

space. If a sequence (f n ) in C[X, Y ] converges uniformly to a map f, then f ∈<br />

C[X, Y ].<br />

Pro<strong>of</strong>. Suppose (f n ) converges uniformly to f. Let x 0 ∈ X, and we will show<br />

that f is continuous at x 0 . Let ϵ > 0, there exists N ∈ N such that d(f N (x), f(x)) <<br />

ϵ/3 for each x ∈ X. Since f N is continuous at x 0 , there is a neighborhood U <strong>of</strong> x 0<br />

such that d(f N (x), f N (x 0 )) < ϵ/3 for any x ∈ U. Then, for x ∈ U, we have<br />

d(f(x), f(x 0 )) ≤ d(f(x), f N (x)) + d(f N (x), f N (x 0 )) + d(f N (x 0 ), f(x 0 )) < ϵ<br />

which implies the continuity <strong>of</strong> f at x 0 as desired.<br />

Corollary 5.17. Let X be a space and (Y, d) a metric space. The subspace<br />

C[X, Y ] <strong>of</strong> Y X is closed in the uniform topology.<br />

Pro<strong>of</strong>. Let Y X be equipped with the uniform topology and f ∈ Y X a limit<br />

point <strong>of</strong> C[X, Y ]. Since Y X is metrizable in the uniform topology, then f is the<br />

limit <strong>of</strong> a sequence in C[X, Y ]. By the uniform limit theorem, we have f ∈ C[X, Y ].<br />

Therefore, C[X, Y ] is closed in Y X .<br />

□<br />


62 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Exercise 5.18. Let X be a space and (Y, d) a metric space. Let B[X, Y ]<br />

denote the set <strong>of</strong> all bounded maps from X to Y . Prove that if a sequence (f n ) in<br />

B[X, Y ] converges uniformly to a map f, then f ∈ B[X, Y ]. Hence, the subspace<br />

B[X, Y ] <strong>of</strong> Y X is also closed in the uniform topology.<br />

Definition 5.19. Let X be a space and (Y, d) a metric space. For each f ∈ Y X ,<br />

a compact subset K <strong>of</strong> X and ϵ > 0, let<br />

B K (f, ϵ) = { g ∈ Y X | sup{d(f(x), g(x)) | x ∈ K} < ϵ } .<br />

Then B = {B K (f, ϵ) | f ∈ Y X , ϵ > 0, K a compact subset <strong>of</strong> X} forms a basis<br />

for a topology on Y X . We will call the topology generated by B the topology <strong>of</strong><br />

compact convergence or the topology <strong>of</strong> uniform convergence on compact sets.<br />

Exercise 5.20. Let X be a space and (Y, d) a metric space. Prove that a<br />

sequence (f n ) in Y X converges uniformly to f on each compact subset K <strong>of</strong> X iff<br />

(f n ) converges to f ∈ Y X in the topology compact convergence.<br />

Definition 5.21. A space X is said to be compactly generated if it satisfies<br />

the following condition : a set A is open [closed] in X if A ∩ K is open [closed] in<br />

K for each compact subset K <strong>of</strong> X.<br />

Theorem 5.22. Every locally compact space is compactly generated.<br />

Pro<strong>of</strong>. Let X be a locally compact space and A ⊆ X. Suppose A ∩ K is open<br />

in K for any compact subset K <strong>of</strong> X. To show that A is open in X, let x ∈ A. By<br />

local compactness at x, we can find a compact set K and an open set U in X such<br />

that x ∈ U ⊆ K. By assumption, A∩K is open in K and hence A∩U = (A∩K)∩U<br />

is open in U. Since U is open in X, so is U ∩ A and hence U ∩ A is a neighborhood<br />

<strong>of</strong> x such that U ∩ A ⊆ A. Therefore, A is open in X.<br />

□<br />

Theorem 5.23. Every first countable space are compactly generated.<br />

Pro<strong>of</strong>. Let X be a first countable space and B ⊆ X. Suppose B ∩ K is closed<br />

in K for any compact subset K <strong>of</strong> X. To show that B is closed in X, let x ∈ B.<br />

Then there is a sequence (x n ) in B converging to x. Now, K = {x n | n ∈ N} ∪ {x}<br />

is clearly compact and hence B ∩ K is closed in K by assumption. Since (x n ) is<br />

also a sequence in B ∩ K converging to x, we must have x ∈ B ∩ K ⊆ B. Therefore<br />

B is closed in X.<br />

□<br />

Lemma 5.24. Let X be a compactly generated space and Y a space. Then a<br />

map f : X → Y is continuous iff for each compact subset K <strong>of</strong> X, the restriction<br />

f| K is continuous.<br />

Pro<strong>of</strong>. (⇒) Trivial.<br />

(⇐) Suppose f is continuous on each compact subset <strong>of</strong> X. To show that f<br />

is continuous, let A be an open subset <strong>of</strong> Y . Then, for each compact subset K <strong>of</strong><br />

X, the set f −1 (A) ∩ K = (f| K ) −1 (A) is open in K since f| K is continuous. Since<br />

X is compactly generated, it follows that f −1 (A) is open in X. Therefore, f is<br />

continuous as desired.<br />

□<br />

Theorem 5.25. Let X be a compactly generated space and (Y, d) a metric space.<br />

The subspace C[X, Y ] <strong>of</strong> Y X is closed in the topology <strong>of</strong> compact convergence.


<strong>Phichet</strong> <strong>Chaoha</strong> 63<br />

Pro<strong>of</strong>. Let Y X be equipped with the topology <strong>of</strong> compact convergence, f a<br />

limit point to C[X, Y ] and K a compact subset <strong>of</strong> X. Then, for each n ∈ N, we<br />

can pick f n ∈ B K (f, 1 n ) ∩ C[X, Y ] and hence obtain a sequence (f n) in C[X, Y ].<br />

It is straightforward to verify that (f n ) converges uniformly to f on K. Then,<br />

the uniform limit theorem implies that f| K is continuous. Therefore, f itself is<br />

continuous by the previous lemma.<br />

□<br />

Theorem 5.26. Let X be a space and (Y, d) a metric space.<br />

following inclusions <strong>of</strong> topologies on Y X :<br />

We have the<br />

uniform topology ⊇ topology <strong>of</strong> compact convergence ⊇ point-open topology.<br />

The first two coincide when X is compact, and the second two coincide when X is<br />

discrete.<br />

Pro<strong>of</strong>. For a basis neighborhood B K (f, ϵ) <strong>of</strong> f ∈ Y X in the topology <strong>of</strong> compact<br />

convergence, we cleary have B ρ (f, δ) ⊆ B K (f, ϵ), where δ = min{ϵ, 1}. Hence,<br />

the uniform topology is generally finer than the topology <strong>of</strong> compact convergence.<br />

Moreover, When X is compact, for a basis neighborhood B ρ (f, ϵ) <strong>of</strong> f in the uniform<br />

topology, we also have B X (f, δ) ⊆ B ρ (f, ϵ), where δ = min{ϵ, 1}.<br />

Also, for a basis neighborhood ∩ n<br />

i=1 S(x i, U i ) <strong>of</strong> f ∈ Y X in the topology <strong>of</strong><br />

point-open convergence, we clearly have B K (f, ϵ) ⊆ ∩ n<br />

i=1 S(x i, U i ), where ϵ =<br />

min{ϵ 1 , ϵ 2 , . . . , ϵ n }, B d (f(x i ), ϵ i ) ⊆ U i and K = {x 1 , x 2 , . . . , x n }. Hence, the topology<br />

<strong>of</strong> compact convergence is finer than the point-open topology. Moreover, when<br />

X is discrete, for a basis neighborhood B K (f, ϵ) <strong>of</strong> f in the topology <strong>of</strong> compact<br />

convergence, we have K = {x 1 , x 2 , . . . , x n } (since a compact subset <strong>of</strong> a discrete<br />

space must be finite) and hence ∩ n<br />

i=1 S(x i, U i ) ⊆ B K (f, ϵ), where U i = B d (f(x i ), ϵ)<br />

for all i = 1, 2, . . . , n.<br />

□<br />

Exercise 5.27. Let X be a space and (Y, d) a metric space. Prove that,<br />

on C[X, Y ], the topology <strong>of</strong> compact convergence and the compact-open topology<br />

coincide. Hence, the topology <strong>of</strong> compact convergence on C[X, Y ] depends on the<br />

metric topology induced by d instead <strong>of</strong> the metric d.<br />

Corollary 5.28. When X is compact and (Y, d) is a metric space, the uniform<br />

topology on C[X, Y ] depends on the metric topology induced by d instead <strong>of</strong> the<br />

metric d.<br />

Pro<strong>of</strong>. Follows directly from the previous theorem and example.<br />

Example 5.29. When X = {1, 2, . . . , n} and Y = R with the euclidean metric,<br />

we clearly have Y X = C[X, Y ] = R n . Since X is compact and discrete, all four<br />

topologies defined above coincide.<br />

Example 5.30. When X = N and Y = R with the euclidean metric, we clearly<br />

have Y X = C[X, Y ] = R ω . Since X is discrete, the uniform topology is finer than<br />

the other three.<br />


CHAPTER 6<br />

Complete Metric Spaces<br />

Definition 6.1. A sequence (x n ) in a metric space (X, d) is called a Cauchy<br />

sequence if for any ϵ > 0, there exists N ∈ N such that d(x m , x n ) < ϵ for all<br />

m, n ≥ N.<br />

Remark 6.2. A convergent sequence in a metric space X is always Cauchy.<br />

Definition 6.3. A metric space X is said to be complete if every Cauchy<br />

sequence in X converges.<br />

Theorem 6.4. A metric space (X, d) is complete if every Cauchy sequence in<br />

X has a convergent subsequence.<br />

Pro<strong>of</strong>. Let (x n ) be a Cauchy sequence in (X, d) and (x nk ) its subsequence<br />

converging to x. Note that n k ≥ k for all k ∈ N. Since (x nk ) converges to x<br />

and (x n ) is Cauchy, then for ϵ > 0, there exists a big enough K ∈ N such that<br />

d(x, x nk ) < ϵ/2 and d(x nk , x k ) < ϵ/2 for all k ≥ K. By the triangle inequality, we<br />

have d(x, x k ) ≤ d(x, x nk ) + d(x nk , x k ) < ϵ for all k ≥ K. Therefore, (x n ) converges<br />

to x as desired.<br />

□<br />

Corollary 6.5. For any k ∈ N, R k is complete (with respect to the euclidean<br />

metric).<br />

Pro<strong>of</strong>. Let d denote the euclidean metric on R k , and let (x n ) be a Cauchy<br />

sequence in (R k , d). Then there exists N ∈ N such that d(x m , x n ) ≤ 1 for all<br />

m, n ≥ N and hence, we have<br />

d(x n , 0) ≤ M = max{d(x 1 , 0), . . . , d(x N−1 , 0), d(x N , 0) + 1}<br />

for all n ∈ N. It follows that (x n ) is a sequence in the (sequentially) compact metric<br />

space B(0; M). Therefore, (x n ) has a convergent subsequence, and by the previous<br />

theorem, (R k , d) is complete.<br />

□<br />

Corollary 6.6. A compact metric space is complete.<br />

Pro<strong>of</strong>. Since compactness is equivalent to sequential compactness for metric<br />

spaces, the result then follows directly from the previous theorem.<br />

□<br />

Example 6.7. Q and R − Q are not complete with respect to the euclidean<br />

metric on R.<br />

Exercise 6.8. Prove that a subspace A <strong>of</strong> a complete metric space (X, d) is<br />

complete iff A is closed in X.<br />

Remark 6.9. It should be pointed out here that completeness is not topologically<br />

invariant. For example, the interval (0, 1) is not complete (by considering the<br />

Cauchy sequence ( 1 n<br />

)), but it is homeomorphic to R. The following example shows<br />

that the completeness <strong>of</strong> a metric space (X, d) really depends on the metric d.<br />

65


66 <strong>Topology</strong> (<strong>2301631</strong>)<br />

x<br />

Example 6.10. Let d(x, y) = ∣<br />

1+|x| − y<br />

1+|y|<br />

∣ for x, y ∈ R. It is straightforward<br />

to verify that d is a metric on R and it is equivalent to the euclidean metric on<br />

R. Since the sequence (n) is Cauchy with respect to d but it does not converge, it<br />

follows that (R, d) is not complete.<br />

Exercise 6.11. Let d and d ′ be two metrics on X. Suppose further that there<br />

are positive real numbers A, B and C such that<br />

Ad(x, y) ≤ Bd ′ (x, y) ≤ Cd(x, y)<br />

for any x, y ∈ X. Prove that a sequence (x n ) in X is Cauchy with respect to d iff<br />

it is Cauchy with respect to d ′ . Hence, (X, d) is complete iff (X, d ′ ) is complete.<br />

Notice also that d and d ′ are equivalent.<br />

The previous example immediately implies that R k is complete with respect to<br />

the square metric and the taxicab metric.<br />

Theorem 6.12. R ω is complete with respect to the metric<br />

D(x, y) = sup{ d(π i(x), π i (y))<br />

| i ∈ N}<br />

i<br />

as defined in Theorem 1.151.<br />

Pro<strong>of</strong>. Let (x n ) be a Cauchy sequence in (R ω , D). Notice that, for each i ∈ N<br />

and ϵ > 0, there exists N i ∈ N such that D(x m , x n ) < min{ϵ,1}<br />

i<br />

for all m, n ≥ N i .<br />

So we have d(π i (x m ), π i (x n )) = d(π i (x m ), π i (x n )) ≤ iD(x m , x n ) < min{ϵ, 1} ≤ ϵ<br />

for all m, n ≥ N i . It follows that (π i (x n )) is also a Cauchy sequence in (R, d) and<br />

hence converges to some a i ∈ R. To see that (x n ) converges to a = (a 1 , a 2 , . . . ), let<br />

B be a basis neighborhood <strong>of</strong> a. WLOG, we may write B = ∏ i∈N U i where U i is a<br />

neighborhood <strong>of</strong> a i for all i ∈ N, and U i = R for all i > m for some m ∈ N. Then<br />

for each i = 1, 2, . . . , m, there exists N i ∈ N such that π i (x n ) ∈ U i . It follows that<br />

x n ∈ B for all n ≥ max{N 1 , N 2 , . . . , N m }. Therefore, (x n ) → a and hence (R ω , D)<br />

is complete as desired.<br />

□<br />

Theorem 6.13. For a space X and a complete metric space (Y, d), the space<br />

Y X is complete with respect to the uniform metric ρ corresponding to d.<br />

Pro<strong>of</strong>. Let (f n ) be a Cauchy sequence in (Y X , ρ). For each x ∈ X, we can<br />

show that (f n (x)) is Cauchy in (Y, d) and hence converges to some y x ∈ Y by<br />

completeness. Now, define f ∈ Y X by f(x) = y x for each x ∈ X. To show that<br />

(f n ) → f, we let ϵ > 0. Since (f n ) is Cauchy in (Y X , ρ), there exists N ∈ N such<br />

that for m, n ≥ N and x ∈ X,<br />

d(f m (x), f n (x)) ≤ ρ(f m , f n ) < ϵ 2 .<br />

By letting m → ∞, we have d(f(x), f n (x)) ≤ ϵ 2<br />

for any n ≥ N and x ∈ X. It<br />

follows that ρ(f, f n ) ≤ ϵ 2 < ϵ for any n ≥ N. Therefore, (f n) → f and hence<br />

(Y X , ρ) is complete. □<br />

Corollary 6.14. Let X be a space and (Y, d) a complete metric space. Then,<br />

the subspaces B[X, Y ] and C[X, Y ] <strong>of</strong> (Y X , ρ) are complete.<br />

Pro<strong>of</strong>. According to the previous chapter, both B[X, Y ] and C[X, Y ] are<br />

closed in (Y X , ρ). It follows from the previous theorem that both subspaces are<br />

complete.<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 67<br />

Example 6.15. R ω , B[X, R n ] and C[X, R n ] are complete with respect to the<br />

uniform metric ρ. Moreover, it is not difficult to show that both B[X, R n ] and<br />

C[X, R n ] (when X is compact) are complete with respect to ρ.<br />

Definition 6.16. Let (X, d X ) and (Y, d Y ) be metric spaces. A map f : X → Y<br />

is called<br />

• an isometry if d Y (f(x), f(y)) = d X (x, y) for each x, y ∈ X.<br />

• a contraction if there exists α ∈ [0, 1) such that<br />

d Y (f(x), f(y)) ≤ αd X (x, y)<br />

for each x, y ∈ X.<br />

• contractive if d Y (f(x), f(y)) < d X (x, y) for each x, y ∈ X and x ≠ y.<br />

• nonexpansive if d Y (f(x), f(y)) ≤ d X (x, y) for each x, y ∈ X.<br />

Clearly, we have the following implications :<br />

isometry ⇒ nonexpansive ⇐ contractive ⇐ contraction.<br />

Example 6.17. Translations, rotations and reflections <strong>of</strong> R n are isometries.<br />

Exercise 6.18. Prove that a nonexpansive map is always uniformly continuous.<br />

Exercise 6.19. Prove that an isometry is always an imbedding.<br />

Exercise 6.20. Prove that f(x) = sin(x) on [0, 1] is contractive but not a<br />

contraction.<br />

Theorem 6.21. Any metric space can be isometrically imbedded into a complete<br />

metric space.<br />

Pro<strong>of</strong>. Let (X, d) be a metric space and x 0 ∈ X. We will show that X can<br />

be imbedded into the metric space (B[X, R], ρ) which is complete by the above<br />

example.<br />

For each a ∈ X, we define ϕ a : X → R by<br />

ϕ a (x) = d(x, a) − d(x, x 0 )<br />

for each x ∈ X. By the triangle inequality, we clearly have<br />

|ϕ a (x)| = |d(x, a) − d(x, x 0 )| ≤ d(a, x 0 )<br />

for each x ∈ X; i.e., ϕ a ∈ B[X, R].<br />

Since for each a, b ∈ X, we clearly have the inequality<br />

d(a, b) ≤ ρ(ϕ a , ϕ b ) = sup |ϕ a (x) − ϕ b (x)| = sup |d(x, a) − d(x, b)| ≤ d(a, b)<br />

x∈X<br />

x∈X<br />

which imples ρ(ϕ a , ϕ b ) = d(a, b). Therefore, the map a ↦→ ϕ a is the desired isometry.<br />

□<br />

Definition 6.22. A completion <strong>of</strong> a metric space X is a pair ( ˜X, ι) <strong>of</strong> a complete<br />

metric space ˜X and an isometric imbedding ι : X → ˜X such that ι(X) = ˜X.<br />

Notice that, for a given isometric imbedding h : X → Y <strong>of</strong> a metric space X<br />

into a complete metric space Y , we immediately obtain a completion <strong>of</strong> X simply by<br />

letting ˜X = h(X) and ι = h. We will call this completion (h(X), h) the completion<br />

induced by h.<br />

Theorem 6.23. A completion <strong>of</strong> a metric space is unique upto isometric homeomorphisms.


68 <strong>Topology</strong> (<strong>2301631</strong>)<br />

Pro<strong>of</strong>. Let ( ˜X, ι) and ( ˜X ′ , ι ′ ) be completions <strong>of</strong> a metric space X. Then,<br />

we clearly have an isometric homeomorphism ι ′ ◦ ι −1 : ι(X) → ι ′ (X), and after<br />

composing with the inclusion <strong>of</strong> ι ′ (X) into ι ′ (X) = ˜X ′ , we obtain an isometric<br />

imbedding f : ι(X) → ˜X ′ . Notice that for a Cauchy sequence (x n ) in ι(X), the<br />

sequence (f(x n )) is also Cauchy in ˜X ′ since f is an isometry. This allows us to<br />

define the desired isometric homeomorphism [exercise below] f : ˜X → ˜X ′ by<br />

f(x) = lim<br />

n→∞ f(x n)<br />

for each x ∈ ˜X and any sequence (x n ) converging to x.<br />

Exercise 6.24. Verify that the map f in the theorem above is a well-defined<br />

(independent from the sequence (x n )) isometric homeomorphism such that the<br />

following diagram commutes :<br />

˜X<br />

❅■ ι ❅<br />

f<br />

∼ =<br />

X.<br />

✲<br />

<br />

✒<br />

ι ′<br />

Exercise 6.25. Let (X, d) be a metric space and CS(X) the set <strong>of</strong> all Cauchy<br />

sequences in (X, d). Define a relation ∼ on CS(X) as follows :<br />

˜X′<br />

(x n ) ∼ (y n ) ⇔ lim<br />

n→∞ d(x n, y n ) = 0.<br />

(1) Prove that ∼ is an equivalence relation.<br />

(2) Prove that for any K ∈ N, we always have (x n ) ∼ (x K+n ).<br />

Now, let ˜X = CS(X)/ ∼ and for any [(x n )], [(y n )] ∈ ˜X,<br />

D([(x n )], [(y n )]) = lim<br />

n→∞ d(x n, y n ).<br />

(3)Prove that D is well-defined and it is a metric on ˜X.<br />

Define ι : (X, d) → ( ˜X, D) by ι(x) = [(x)].<br />

(4) Prove that ι is an isometric imbedding.<br />

(5) Prove that ι(X) is dense in ˜X.<br />

(6) Prove that that ( ˜X, D) is complete.<br />

Theorem 6.26 (Banach’s Fixed Point Theorem). Let (X, d) be a complete<br />

metric space and f : X → X a contraction. Then f has a unique fixed point, says<br />

x 0 , and the sequence (f n (x)) converges to x 0 for any x ∈ X.<br />

Pro<strong>of</strong>. Since f : X → X is a contraction, there exists α ∈ [0, 1) such that<br />

d(f(x), f(y)) ≤ αd(x, y) for any x, y ∈ X. Let x ∈ X. If f(x) = x, then x 0 = x.<br />

Suppose f(x) ≠ x. To obtain x 0 , we first show that the sequence (f n (x)) is Cauchy.<br />

Observe that<br />

• For m, n ∈ N with m > n, we have<br />

d(f m (x), f n (x)) ≤ αd(f m−1 (x), f n−1 (x)) ≤ · · · ≤ α n d(f m−n (x), x).<br />

• For k ∈ N, we have<br />

d(f k (x), x) ≤ Σ k i=1d(f i (x), f i−1 (x)) ≤ Σ k i=1α i−1 d(f(x), x)<br />

= 1−αk<br />

1<br />

1−α<br />

d(f(x), x) ≤<br />

1−αd(f(x), x).<br />


Let ϵ > 0 and N ∈ N be such that α N <<br />

assume that m > n and we have<br />

( α<br />

d(f m (x), f n (x)) ≤ α n d(f m−n n<br />

(x), x) ≤<br />

1 − α<br />

<strong>Phichet</strong> <strong>Chaoha</strong> 69<br />

ϵ(1−α)<br />

d(f(x),x)<br />

. Then for m, n ≥ N, we may<br />

)<br />

d(f(x), x) < ϵ.<br />

Hence, (f n (x)) is Cauchy and converges by the completeness <strong>of</strong> (X, d). Let<br />

x 0 = lim n→∞ f n (x). Since f is continuous, x 0 is a fixed point <strong>of</strong> f by<br />

f(x 0 ) = f( lim<br />

n→∞ f n (x)) = lim<br />

n→∞ f(f n (x)) = lim<br />

n→∞ f n+1 (x) = x 0 .<br />

Finally, suppose x ′ 0 is another fixed point <strong>of</strong> f. Since f is a contraction with<br />

α < 1, we have d(x 0 , x ′ 0) = d(f(x 0 ), f(x ′ 0)) ≤ αd(x 0 , x ′ 0) which is impossible unless<br />

d(x 0 , x ′ 0) = 0. Therefore, we must have x 0 = x ′ 0.<br />

□<br />

Definition 6.27. A metric space (X, d) is said to be totally bounded if for each<br />

ϵ > 0, X can be covered by finitely many ϵ-balls; i.e., there exists x 1 , x 2 , . . . , x n ∈ X,<br />

such that X = ∪ n<br />

i=1 B d(x i , ϵ).<br />

Theorem 6.28. A totally bounded metric space is always bounded.<br />

Pro<strong>of</strong>. Let (X, d) be a totally bounded metric space. Then, we can cover X by<br />

finitely many unit balls, says X = ∪ n<br />

i=1 B d(x i , 1). Let M = max{d(x i , x j ) | i, j =<br />

1, 2, . . . , n}. For any x, y ∈ X, we clearly have x ∈ B d (x i , 1) and y ∈ B d (x j , 1) for<br />

some i, j = 1, 2, . . . , n. Hence, d(x, y) ≤ d(x, x i ) + d(x i , x j ) + d(x j , y) ≤ M + 2; i.e.,<br />

(X, d) is bounded.<br />

□<br />

Example 6.29. The converse <strong>of</strong> the previous theorem is not true. Consider an<br />

infinite set X together with the discrete metric d. Clearly, (X, d) is bounded since<br />

d(x, y) ≤ 1 for all x, y ∈ X. However, it is impossible to cover X by finitely many<br />

open balls <strong>of</strong> radius 1 2 .<br />

Example 6.30. A compact metric space is always totally bounded.<br />

Theorem 6.31. A metric space is compact iff it is complete and totally bounded.<br />

Pro<strong>of</strong>. (⇒) Follows directly from Corollary 6.6 and the previous example.<br />

(⇐) Let (X, d) be a metric space that is complete and totally bounded. Then,<br />

it suffices to show that X is sequentially compact. Let (x n ) be a sequence in X.<br />

WLOG, we may assume that the set A 0 = {x n | n ∈ N} is infinite. Since X is<br />

totally bounded, we can cover X by finitely many open balls <strong>of</strong> radius 1, and hence,<br />

there exists a ball B 1 <strong>of</strong> radius 1 such that A 1 = B 1 ∩ A 0 is infinite. Now, assume<br />

that an infinite set A k and an open ball B k <strong>of</strong> radius 1 k<br />

are defined for some k ≥ 1,<br />

1<br />

we will construct an infinite set A k+1 and a ball B k+1 <strong>of</strong> radius<br />

k+1<br />

as follows<br />

1<br />

: cover X by finitely many open balls <strong>of</strong> radius<br />

k+1 and use the fact that A k is<br />

1<br />

infinite to obtain an open ball B k+1 <strong>of</strong> radius<br />

k+1 such that A k+1 = B k+1 ∩ A k is<br />

infinite. Now, for each k ∈ N, let J k = {n | x n ∈ A k }. Clearly, each J k is an infinite<br />

subset <strong>of</strong> N and J 1 ⊇ J 2 ⊇ J 3 ⊇ . . . Choose n 1 ∈ J 1 and for k > 1, we can always<br />

choose n k ∈ J k such that n k > n k−1 because J k is infinite. By this construction, it<br />

is easy to see that, for all i, j ≥ k, we have d(x ni , x nj ) < 2 k because x n i<br />

, x nj ∈ B k .<br />

That is (x nk ) forms a Cauchy subsequence <strong>of</strong> (x n ). Therefore, by completeness,<br />

(x nk ) is a convergent subsequence <strong>of</strong> (x n ). □


70 <strong>Topology</strong> (<strong>2301631</strong>)<br />

We will end this chapter by considering the well-known Ascoli’s theorem which<br />

characterizes compact subsets <strong>of</strong> the uniform space (C[X, R n ], ρ) when X is compact.<br />

Since the sup metric ρ also induces the uniform topology in this case, we will<br />

always consider C[X, R n ] in the sup metric throughout.<br />

Definition 6.32. Let X be a space, (Y, d) a metric space and F ⊆ C[X, Y ].<br />

For x 0 ∈ X, the set F is equicontinuous at x 0 if for each ϵ > 0, there exists a<br />

neighborhood U <strong>of</strong> x 0 such that<br />

for all x ∈ U and all f ∈ F.<br />

equicontinuous at each x 0 ∈ X.<br />

d(f(x), f(x 0 )) < ϵ<br />

We simply say that F is equicontinuous if it is<br />

Exercise 6.33. Show that F = {f n : I → I | f n (x) = x n , n ∈ N} is not<br />

equicontinuous at 1.<br />

Lemma 6.34. Let X be a compact space, (Y, d) a compact metric space and<br />

F ⊆ (C[X, Y ], ρ). Then F is equicontinuous iff F is totally bounded.<br />

Pro<strong>of</strong>. (⇒) Suppose F is equicontinuous and let ϵ > 0. Then for each x ∈ X,<br />

there is a neighborhood U x <strong>of</strong> x such that d(f(x), f(y)) < ϵ 3 for any y ∈ U x and<br />

f ∈ F. By the compactness <strong>of</strong> X, we then can cover X by U x1 , U x2 , . . . U xn for<br />

some x 1 , x 2 , . . . , x n ∈ X. Also, by the compactness <strong>of</strong> Y , we can cover Y by finitely<br />

many open sets V 1 , V 2 , . . . , V m with diam(V i ) < ϵ 3<br />

. Consider the set<br />

I = {α : {1, . . . , n} → {1, . . . , m} | ∃f ∈ F∀i ∈ {1, . . . , n}, f(x i ) ∈ V α(i) }<br />

which is clearly finite. For each α ∈ I, we fix a map f α ∈ F such that f α (x i ) ∈ V α(i)<br />

for all i ∈ {1, 2, . . . , n}.<br />

Now, we claim that F ⊆ ∪ α∈I B ρ(f α , ϵ). Let f ∈ F. Then we obtain a function<br />

α : {1, . . . , n} → {1, . . . , m} such that f(x i ) ∈ V α(i) for all i ∈ {1, 2, . . . , n}. That is<br />

α ∈ I (notice that f and f α may not be the same map). Moreover, for each x ∈ X,<br />

we have x ∈ U xi for some i ∈ {1, 2, . . . , n} and hence<br />

d(f(x), f α (x)) ≤ d(f(x), f(x i )) + d(f(x i ), f α (x i )) + d(f α (x i ), f α (x)) < ϵ.<br />

It follows that ρ(f, f α ) = max{d(f(x), f α (x)) | x ∈ X} < ϵ ; i.e., f ∈ B ρ (f α , ϵ).<br />

(⇐) Suppose F is totally bounded. Let ϵ > 0 and x ∈ X. Then there exist<br />

f 1 , f 2 , . . . , f n ∈ F such that F ⊆ ∪ n<br />

i=1 B ρ(f i , ϵ 3<br />

). For each i = 1, 2, . . . , n, by<br />

the continuity <strong>of</strong> each f i at x, we can find a neighborhood U i <strong>of</strong> x such that<br />

d(f i (x), f i (y)) < ϵ 3 for any y ∈ U i. So U = ∩ n<br />

i=1 U i is a neighborhood <strong>of</strong> x. For any<br />

y ∈ U and f ∈ F, we must have f ∈ B ρ (f i , ϵ 3<br />

) for some i = 1, 2, . . . , n and hence<br />

d(f(x), f(y)) ≤ d(f(x), f i (x)) + d(f i (x), f i (y)) + d(f i (y), f(y)) < ϵ.<br />

Therefore, F is equicontinuous at x.<br />

Exercise 6.35. Let X be a compact space and F ⊆ (C[X, R n ], ρ). Prove that<br />

if F is bounded, then F ⊆ C[X, Y ] for some compact subset Y <strong>of</strong> R n .<br />

Theorem 6.36 (Classical Ascoli’s Theorem). Let X be a compact space and<br />

F ⊆ (C[X, R n ], ρ). Then F is compact iff it is closed, bounded and equicontinuous.<br />

Pro<strong>of</strong>. (⇒) Since F is compact, it is closed and totally bounded by the previous<br />

theorem. Hence, F is also bounded. Since F ⊆ C[X, Y ] for some compact<br />


<strong>Phichet</strong> <strong>Chaoha</strong> 71<br />

subset Y <strong>of</strong> R n by the previous exercise, F must be equicontinuous by the previous<br />

lemma.<br />

(⇐) Since F is closed and (C[X, R n ], ρ) is complete, F is complete with respect<br />

to ρ. Since F is bounded with respect to ρ, the previous exercise implies that<br />

F ⊆ C[X, Y ] for some compact subset Y <strong>of</strong> R n . Finally, since F is equicontinuous,<br />

it must be totally bounded by the previous lemma.<br />

□<br />

Definition 6.37. Let X be a space and (Y, d) a metric space. A subset F <strong>of</strong><br />

C[X, Y ] is said to be pointwise bounded if for each x ∈ X, the subset<br />

is bounded.<br />

F x = {f(x) | f ∈ F}<br />

Exercise 6.38. Let X be a compact space and F ⊆ (C[X, R n ], ρ). Prove :<br />

(1) If F is equicontinuous and pointwise bounded, then F is bounded.<br />

(2) If F is equicontinuous, then so is F.<br />

(3) F has compact closure iff it is pointwise bounded and equicontinuous.<br />

Theorem 6.39 (Arzela’s Theorem). Let X be compact and (f n ) a sequence<br />

in (C[X, R n ], ρ). If the collection F = {f n | n ∈ N} is pointwise bounded and<br />

equicontinuous, then (f n ) has a uniformly convergent subsequence.<br />

Pro<strong>of</strong>. Since F is pointwise bounded and equicontinuous, the previous exercise<br />

implies that F is compact in (C[X, R n ], ρ) and hence sequentially compact.<br />

Since (f n ) a sequence in F and ρ induces the uniform topology, (f n ) has a uniform<br />

convergent subsequence.<br />

□<br />

Example 6.40. The family F = {f n : I → I | f n (x) = x n , n ∈ N} ⊆ C[I, R]<br />

is not equicontinuous by Arzela’s theorem since it is pointwise bounded (can you<br />

see why?) and the sequence (f n ) converges to a function f that is not continuous<br />

(hence it cannot have a uniform convergent subsequence).


Bibliography<br />

[1] S. W. Davis, <strong>Topology</strong>, McGraw-Hill, 2005.<br />

[2] M. A. Khamsi and W. A. Kirk, An introduction to metric spaces and fixed point theory, John<br />

Wiley & Sons, Inc., 2001.<br />

[3] J. G. Hocking and G. S. Young, <strong>Topology</strong>, Dover, 1998.<br />

[4] J. R. Munkres, <strong>Topology</strong> (second edition), Prentice Hall, Upper Saddle River, NJ, 2000.<br />

73

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!