Lecture notes on Topology

Huỳnh Quang Vũ

Version of February 10, 2019

2
 i

This is a set of lecture notes for a series of introductory courses in topology for under-
graduate students at the University of Science, Ho Chi Minh City. It is written to be deliv-
ered by myself, tailored to my students. I did not write it with other lecturers or self-study
readers in mind.
In writing these notes I intend that more explanations and discussions will be carried
out in class. I hope by presenting only the essentials these notes will be more suitable for
use in classroom. Some details are left for students to fill in or to be discussed in class.
A sign X in front of a problem notifies the reader that this is a typical problem or an im-
portant one (the result is used later). A sign * indicates a relatively more difficult problem.
The latest version of this set of notes is available at
http://www.math.hcmus.edu.vn/∼hqvu/n.pdf, and the source file is at
http://www.math.hcmus.edu.vn/∼hqvu/n.zip. This work is released to Public Domain
(CC0) wherever applicable, see
http://creativecommons.org/publicdomain/zero/1.0/, otherwise it is licensed under the
Creative Commons Attribution 4.0 International License, see
http://creativecommons.org/licenses/by/4.0/.
Address: Faculty of Mathematics and Computer Science, University of Science, Vietnam
National University - Ho Chi Minh City. Email: hqvu@hcmus.edu.vn
ii
Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

I General Topology 3
1 Infinite sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Topological space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6 Compact space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7 Product of spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
8 Real functions and Spaces of functions . . . . . . . . . . . . . . . . . . . . . . 54
9 Quotient space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Other topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

II Algebraic Topology 77
10 Structures on topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
11 Classification of compact surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 86
12 Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
13 The fundamental group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
14 The fundamental group of the circle . . . . . . . . . . . . . . . . . . . . . . . . 104
15 Van Kampen theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
16 Simplicial homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
17 Singular homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
18 Homology of cell complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Other topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

III Differential Topology 129

19 Smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
20 Tangent spaces and derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
21 Regular values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
22 Critical points and the Morse lemma . . . . . . . . . . . . . . . . . . . . . . . . 147
23 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
24 Manifold with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
25 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
26 Topological degrees of maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
27 Integration of real functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

iii
iv CONTENTS

Guide for further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

Suggestions for some problems 183

Bibliography 188
CONTENTS 1

Introduction
Topology is a mathematical subject that studies shapes. The term comes from the Greek
words “topos” (place) and “ology” (study). A set becomes a topological space when each
element of the set is given a collection of neighborhoods. Operations on topological spaces
must be continuous, bringing certain neighborhoods into neighborhoods. There is no no-
tion of distance. Topology is a part of geometry that does not concern distance.

Figure 0.1: How to make a closed trip such that every bridge is crossed exactly once? This is
the problem “Seven Bridges of Konigsberg”, studied by Leonard Euler in the 18th century.
It does not depend on the sizes of the bridges.

Characteristics of topology
Operations on topological objects are more relaxed than in geometry: beside moving around
(allowed in geometry), stretching or bending are allowed in topology (not allowed in ge-
ometry). For example, in topology circles - big or small, anywhere - are same. Ellipses and
circles are same. On the other hand in topology tearing or breaking are not allowed: circles
are still different from lines. While topological operations are more flexible they still retain
some essential properties of spaces.

Contributions of topology
Topology provides basic notions to areas of mathematics where there is a need for a notion
of continuity. Topology focuses on some essential properties of spaces. It can be used
in qualitative study. It can be useful where metrics or coordinates are not available, not
natural, or not necessary.
Initially developed in the late nineteenth century and early twentieth century to provide
basis for abstract mathematical analysis, topology gradually became an influential subject,
reaching many achievements in the mid and late twentieth century.
Topology often does not stand alone: nowadays there are fields such as algebraic topol-
ogy, differential topology, geometric topology, combinatorial topology, quantum topol-
ogy, . . . Topology often does not solve a problem by itself, but contributes important un-
derstanding, settings, and tools. Topology features prominently in differential geometry,
global analysis, algebraic geometry, theoretical physics . . .
2 CONTENTS
Part I General Topology

3
1. INFINITE SETS 5

1 Infinite sets
General Topology, also called Point-set Topology, studies fundamentals of neighborhoods,
limits, and continuity – basic notions used throughout mathematics. In General Topology
we often work in very general settings, in particular we often deal with infinite sets.
We will not define what a set is. In other words, we will work on the level of “naive set
theory”, pioneered by Georg Cantor in the late 19th century. We will use familiar notions
such as unions, intersections, maps, Cartesian product of two sets, . . . without recalling
definitions. We will not go back to definitions of the natural numbers and of the real num-
bers.
Still, we should be aware of certain problems in naive set theory.

Example (Russell’s paradox). Consider the set S = { x | x ∈

/ x } (the set of all sets which are
not members of themselves). Then whether S ∈ S or not is undecidable, since answering
either yes or no to this question leads to contradiction.1
Axiomatic systems for theory of sets have been developed since then. In the Von Neu-
mann–Bernays–Godel system a more general notion than set, called class (lớp), is used. In
this course, we do not distinguish set, class, or collection (họ), but in occasions where we
deal with “set of sets” we often prefer the term collection. For more one can read [End77,
p. 6], [Dug66, p. 32].

Indexed collection
A map f : I → A where I is a set and A is a collection is called an indexed collection, or
indexed family (họ được đánh chỉ số). We often write f i = f (i ), and denote the indexed
collection f by ( f i )i∈ I or { f i }i∈ I . Notice that it can happen that f i = f j for some i 6= j.

Example. A sequence of elements in a set A is a collection of elements of A indexed by the

set Z+ of positive integer numbers, written as ( an )n∈Z+ .

Relation
A relation (quan hệ) R on a set S is a non-empty subset of the set S × S.
When ( a, b) ∈ R we often say that a is related to b and often write a ∼ R b.
A relation said to be:

(a) reflexive (phản xạ) if ∀ a ∈ S, ( a, a) ∈ R.

(b) symmetric (đối xứng) if ∀ a, b ∈ S, ( a, b) ∈ R ⇒ (b, a) ∈ R.

(c) anti-symmetric (phản đối xứng) if ∀ a, b ∈ S, (( a, b) ∈ R ∧ (b, a) ∈ R) ⇒ a = b.

(d) transitive (bắc cầu) if ∀ a, b, c ∈ S, (( a, b) ∈ R ∧ (b, c) ∈ R) ⇒ ( a, c) ∈ R.

An equivalence relation on S is a relation that is reflexive, symmetric and transitive.

If R is an equivalence relation on S then an equivalence class (lớp tương đương) repre-
sented by a ∈ S is the subset [ a] = {b ∈ S | ( a, b) ∈ R}. Two equivalence classes are either
coincident or disjoint. The set S is partitioned (phân hoạch) into the disjoint union of its
equivalence classes.
1 Discovered in 1901 by Bertrand Russell. A famous version of this paradox is the barber paradox: In a village

there is a barber; his job is to do hair cut for a villager if and only if the villager does not cut his hair himself.
Consider the set of all villagers who had their hairs cut by the barber. Is the barber himself a member of that set?
6

Order
An order (thứ tự) on a set S is a relation R on S that is reflexive, anti-symmetric and transi-
tive.
Note that two arbitrary elements a and b do not need to be comparable; that is, the pair
( a, b) may not belong to R. For this reason an order is often called a partial order.
When ( a, b) ∈ R we often write a ≤ b. When a ≤ b and a 6= b we write a < b.
If any two elements of S are related then the order is called a total order (thứ tự toàn
phần) and (S, ≤) is called a totally ordered set.

Example. The set R of all real numbers with the usual order ≤ is totally ordered.

Example. Given a set S the collection of all subsets of S is denoted by P (S) or 2S . With the
inclusion relation, (2S , ⊆) is a partially ordered set, but is not a totally ordered set if S has
more than one element.

Example. Let (S1 , ≤1 ) and (S2 , ≤2 ) be two ordered sets. The following is an order on
S1 × S2 : ( a1 , b1 ) ≤ ( a2 , b2 ) if ( a1 < a2 ) or (( a1 = a2 ) ∧ (b1 ≤ b2 )). This is called the
dictionary order (thứ tự từ điển).

In an ordered set, the smallest element (phần tử nhỏ nhất) is the element that is smaller
than all other elements. More concisely, if S is an ordered set, the smallest element of S is
an element a ∈ S such that ∀b ∈ S, a ≤ b. The smallest element, if exists, is unique.
A minimal element (phần tử cực tiểu) is an element which no element is smaller than.
More concisely, a minimal element of S is an element a ∈ S such that ∀b ∈ S, b ≤ a ⇒ b = a.
There can be more than one minimal element.
A lower bound (chặn dưới) of a subset of an ordered set is an element of the set that is
smaller than or equal to any element of the subset. More concisely, if A ⊂ S then a lower
bound of A in S is an element a ∈ S such that ∀b ∈ A, a ≤ b.
The definitions of largest element, maximal element, and upper bound are similar.

Set equivalence
Two sets are said to be set-equivalent if there is a bijection from one set to the other set.
A set is said to be finite if it is either empty or is set-equivalent to a subset {1, 2, 3, . . . , n}
of the set all positive integers Z+ for some n ∈ Z+ . More precisely, a set S is finite if S = ∅
or there exists N ∈ Z+ such that S is set-equivalent to the set {n ∈ Z+ | n ≤ N }.
If a set is not finite we say that it is infinite.
A set is called countably infinite (vô hạn đếm được) if it is equivalent to Z+ . A set is
called countable if it is either finite or countably infinite.
Intuitively, a countably infinite set can be “counted” by the positive integers. The ele-
ments of such a set can be indexed by the set of all positive integers as a sequence a1 , a2 , a3 , . . . .

Example. The set Z of all integer numbers is countable. We can count alternatively the
positive and the negative integers, by

Z+ → Z

n, n is even,
2
n 7→ m =
 1− n , n is odd.
2
1. INFINITE SETS 7

The inverse map brings positive integers to even positive integers, and negative integers to
odd positive integers:

Z → Z+

2m, m > 0,
m 7→ n =
−2m + 1, m ≤ 0.

Proposition. A subset of a countable set is countable.

Proof. The statement is equivalent to the statement that a subset of Z+ is countable. Sup-
pose that A is an infinite subset of Z+ . Let a1 be the smallest number in A. Let an be the
smallest number in A \ { a1 , a2 , . . . , an−1 }. Then an−1 < an and the set B = { an | n ∈ Z+ } is
a countably infinite subset of A.
We show that any element m of A is an an for some n, and therefore B = A.
Let C = { an | an ≥ m}. Then C 6= ∅ since B is infinite. Let min C = an0 . We are using
the well-ordering property of the set of all positive integers: every non-empty subset has
a smallest element. Then an0 ≥ m. Further, since an0 −1 < an0 we have an0 −1 < m. This


implies m ∈ A \ { a1 , a2 , . . . , an0 −1 }. Since an0 = min A \ { a1 , a2 , . . . , an0 −1 } we must have
an0 ≤ m. Thus an0 = m.

Corollary. If there is an injective map from a set S to Z+ then S is countable.

1.1 Proposition. If there is a surjective map from Z+ to a set S then S is countable.

Proof. Suppose that there is a surjective map φ : Z+ → S. For each s ∈ S the set φ−1 (s) is
non-empty. Let ns = min φ−1 (s). The map s 7→ ns is an injective map from S to a subset of
Z+ , therefore S is countable.

Proposition. Z+ × Z+ is countable.

Proof. We can enumerate Z+ × Z+ by counting along diagonals, such as by the method

shown in the following diagram:

(1, 1) / (1, 2)
5 (1, 3) 7 (1, 4)

{ { {
(2, 1) (2, 2) (2, 3)

{ {
(3, 1) (3, 2) ...

{
(4, 1)

For detail we can derive the explicit formula for the counting:

Z+ × Z+ → Z+
(m + n − 2)(m + n − 1)
(m, n) 7→ (1 + 2 + · · · + ((m + n − 1) − 1)) + m = + m.
2
8

(k−2)(k−1)
We will check that this map is injective. Let k = m + n. Suppose that 2 +m =
(k0 −2)(k0 −1)
2 + m0 . If k = k0 then the equation certainly leads to m = m and n = n . If k < k0
0 0

then

(k − 2)(k − 1) (k − 2)(k − 1) ( k − 1) k
+m ≤ + ( k − 1) = <
2 2 2
( k − 1) k (k0 − 2)(k0 − 1)
< +1 ≤ + m0 ,
2 2

a contradiction.

1.2 Theorem. The union of a countable collection of countable sets is a countable set.

Proof. The collection can be indexed as A1 , A2 , . . . , Ai , . . . (if the collection is finite we can
let Ai be the same set for all i starting from a certain index). The elements of each set Ai
can be indexed as ai,1 , ai,2 , . . . , ai,j , . . . (if Ai is finite we can let ai,j be the same element for
all j starting from a certain index). This means there is a surjective map from the index set
Z+ × Z+ to the union Ai by (i, j) 7→ ai,j .
S
i∈ I

Theorem. The set Q of all rational numbers is countable.

p
Proof. One way to prove this result is by writing Q = | p ∈ Z , then use 1.2.
S
q ∈Z+ q
p
Another way is by observing that if we write each rational number in the form q with
p
q > 0 and gcd( p, q) = 1 then the map q 7→ ( p, q) from Q to Z × Z is injective.

1.3 Theorem. The set R of all real numbers is uncountable.

Proof. The proof uses the Cantor diagonal argument. We use the property that every real
number can be expressed in decimal forms. Notice that there are real numbers whose
1
decimal presentations are not unique, such as 2 = 0.5000 . . . = 0.4999 . . . . (See 1.17 for
more details on this topic.)
Suppose that set of all real numbers in the interval [0, 1] is countable, and is enumerated
as a sequence { ai | i ∈ Z+ }. Let us write

a1 = 0.a1,1 a1,2 a1,3 . . .

a2 = 0.a2,1 a2,2 a2,3 . . .
a3 = 0.a3,1 a3,2 a3,3 . . .
..
.

Choose a number b = 0.b1 b2 b3 . . . such that bn 6= 0, bn 6= 9, and bn 6= an,n . Then b 6= an for

all n. Thus the number b is not in the above table, a contradiction.

Example. Two intervals [ a, b] and [c, d] on the real number line are equivalent. The bijection
d−c
can be given by a linear map x 7→ b− a ( x − a) + c. Similarly, any two intervals ( a, b) and
(c, d) are equivalent.

1.4 Example. The interval (−1, 1) is equivalent to R via a map related to the tan function:

x
x 7→ √ .
1 − x2
1. INFINITE SETS 9

1
x
tan α = √
1− x 2

x α

0 1
√
1− x2

−1

1.5 Theorem. A non-empty set is not equivalent to the set of all of its subsets. More specifically,
given a set S 6= ∅ there is no surjective map from S to 2S .

Note that there is an injective map from S to 2S , namely a 7→ { a}.

Proof. Let φ be any map from S to 2S . Let X = { a ∈ S | a ∈

/ φ( a)}. Suppose that there is
x ∈ S such that φ( x ) = X. Then the truth of the statement x ∈ X (whether it is true or false)
is undecidable, a contradiction. Therefore there is no x ∈ S such that φ( x ) = X, so φ is not
surjective.

This result implies that any set is “strictly smaller” than the set of all of its subsets. So
there can not be a set that is “larger” than any other set. There is no “universal set”, “set
which contains everything”, or “set of all sets”.
There is a notion of sizes of sets, called cardinality (lực lượng), however we do not
discuss it here.

The Axiom of choice

Proposition. The following statements are equivalent:

(a) Axiom of choice: Given a collection of non-empty sets, there is a function defined on this
collection, called a choice function, associating each set in the collection with an element of
that set.

(b) Zorn lemma: If any totally ordered subset of an ordered set X has an upper bound then X has
a maximal element.

Zorn lemma is often a convenient form of the Axiom of choice.

Intuitively, a choice function “chooses” an element from each set in a given collection of
non-empty sets. The Axiom of choice allows us to make infinitely many arbitrary choices.2
This is also often used in constructions of functions, sequences, or nets, see one example
at 5.5. One common application is the use of the product of an infinite family of sets – the
Cartesian product, discussed below.
2 Bertrand
Russell said that choosing one shoe from each pair of shoes from an infinite collection of pairs of
shoes does not need the Axiom of choice (because in a pair of shoes the left shoe is different from the right one
so we can define our choice), but usually in a pair of socks the two socks are identical, so choosing one sock from
each pair of socks from an infinite collection of pairs of socks needs the Axiom of choice.
10

The Axiom of choice is needed for many important results in mathematics, such as the
Tikhonov theorem in Topology, the Hahn-Banach theorem and Banach-Alaoglu theorem in
Functional analysis, the existence of a Lebesgue unmeasurable set in Real analysis, . . . .
There are cases where this axiom could be avoided. For example in the proof of 1.1 we
used the well-ordered property of Z+ instead. See for instance [End77, p. 151] for further
material on this subject.

Cartesian product
Let ( Ai )i∈ I be a collection of sets indexed by a set I. The Cartesian product (tích Decartes)
∏i∈ I Ai of this indexed collection is defined to be the collection of all maps a : I → i∈ I Ai
S

such that a(i ) ∈ Ai for every i ∈ I. The existence of such a map is a consequence of the
Axiom of choice. An element a of ∏i∈ I Ai is often denoted by ( ai )i∈ I , with ai = a(i ) ∈ Ai
being the coordinate of index i, in analog to the finite product case.

Problems
1.6. Let f be a function. Show that:

(a) f ( f ( A i ).
S S
i Ai ) = i

(b) f ( Ai ) ⊂ f ( Ai ). If f is injective (one-one) then equality happens.

T T
i i

(c) f −1 ( f −1 ( A i ).
S S
i Ai ) = i

(d) f −1 ( f −1 ( A i ).
T T
i Ai ) = i

1.7. Let f be a function. Show that:

(a) f ( f −1 ( A)) ⊂ A. If f is surjective (onto) then equality happens.

(b) f −1 ( f ( A)) ⊃ A. If f is injective then equality happens.

1.8. Show that if A is infinite and B is countable then A ∪ B is equivalent to A.

1.9. Give another proof of 1.2 by checking that the map Z+ × Z+ → Z+ , (m, n) 7→ 2m 3n is injective.

1.10. X Show that the set of points in Rn with rational coordinates is countable.

1.11. Show that if A has n elements then |2 A | = 2n .

1.12. Show that the set of all functions f : A → {0, 1} is equivalent to 2 A .

1.13. A real number α is called an algebraic number if it is a root of a polynomial with integer coeffi-
cients. Show that the set of all algebraic numbers is countable.
A real number which is not algebraic is called transcendental. For example it is known that π and
e are transcendental. Show that the set of all transcendental numbers is uncountable.

1.14. A continuum set is a set which is set-equivalent to R. Show that a countable union of continuum
sets is a continuum set.

1.15 (Cantor set). Deleting the open interval ( 13 , 23 ) from the interval of real numbers [0, 1], one gets
a space consisting of two intervals [0, 13 ] ∪ [ 32 , 1]. Continuing, delete the intervals ( 19 , 29 ) and ( 79 , 89 ).
1
In general on each of the remaining intervals, delete the middle open interval of 3 the length of that
interval. The Cantor set is the set of remaining points. It can be described as the set of real numbers
∑∞
an
n=1 3n , an = 0, 2. In other words, it is the set of real numbers in [0, 1] which in base 3 could be
written without the digit 1.
Show that the total length of the deleted intervals is 1. Is the Cantor set countable?
1. INFINITE SETS 11

1.16 (Cantor–Schoeder–Bernstein theorem). We will prove: If A is set-equivalent to a subset of B

and B is set-equivalent to a subset of A then A and B are set-equivalent ([KF75, p. 17], [End77, p.
148]).
Suppose that f : A 7→ B and g : B 7→ A are injective maps. Let A0 = A and B0 = B. For n ∈ Z,
n ≥ 0, let Bn+1 = f ( An ) and An+1 = g( Bn ).
(a) Show that An+1 ⊂ An and Bn+1 ⊂ Bn .
(b) Show that An+2 is set-equivalent to An , and An \ An+1 is set-equivalent to An+2 \ An+3 .
(c) By writing

∞
\
A = [( A0 \ A1 ) ∪ ( A1 \ A2 ) ∪ ( A2 \ A3 ) ∪ ( A3 \ A4 ) ∪ · · · ] ∪ ( An )
n =1
∞
\
A1 = [( A1 \ A2 ) ∪ ( A2 \ A3 ) ∪ ( A3 \ A4 ) ∪ ( A4 \ A5 ) ∪ · · · ] ∪ ( An )
n =1

deduce that A is set-equivalent to A1 .

1.17. Show that any real number could be written in base d with any d ∈ Z, d ≥ 2. More specifically,
any positive real number x can be written as

a1 a a
x = a0 .a1 a2 a3 · · · = a0 + + 22 + 33 + · · ·
d d d
where ai ∈ Z, 0 ≤ a0 , if i ≥ 1 then 0 ≤ ai ≤ d − 1. However two forms in base d can represent the
same real number, as seen in 1.3. This happens only if starting from certain digits, all digits of one
form are 0 and all digits of the other form are d − 1. (This result is used in 1.3.)

1.18 (Rn is equivalent to R). Here we prove that R2 is equivalent to R, in other words, a plane is
set-equivalent to a line, or the set C of complex numbers is equivalent to the set R of real numbers.
As a corollary, Rn is equivalent to R.
Let us construct a map from [0, 1) × [0, 1) to [0, 1) as follows: the pair of two real numbers in
decimal forms 0.a1 a2 . . . and 0.b1 b2 . . . corresponds to the real number 0.a1 b1 a2 b2 . . .. In view of 1.17,
we only allow decimal presentations in which not all digits are 9 starting from a certain digit. Check
that this map is injective.

1.19. We prove that 2Z is set-equivalent to R.

+

(a) Show that 2Z is set-equivalent to the set of all sequences of binary digits. Deduce that there is
+

an injective map from [0, 1] to 2N .

(b) Consider a map f : 2Z → [0, 2], for each binary sequence a = a1 a2 a3 · · · , if starting from a
+

certain digit all digits are 1 then let f ( a) = 1.a1 a2 a3 · · · in binary form, otherwise let f ( a) =
0.a1 a2 a3 · · · . Show that f is injective.

The statement that there does not exist a set that is strictly more than Z+ but is strictly less than R
(meaning a set S for which there is an injective map from Z+ to S but there is not a bijective map,
and there is an injective map from S to R but there is not a bijective map) is called the Continuum
hypothesis.

1.20 (transfinite induction principle). An ordered set S is well-ordered (được sắp tốt) if every non-
empty subset A of S has a smallest element, i.e. ∃ a ∈ A, ∀b ∈ A, a ≤ b. For example with the usual
order, N is well-ordered while R is not. Notice that a well-ordered set must be totally ordered. Ernst
Zermelo proved in 1904, based on the Axiom of choice, that any set can be well-ordered.
Prove the following generalization of the principle of induction. Let A be a well-ordered set. Let
P( a) be a statement whose truth depends on a ∈ A. Suppose that if P( a) is true for all a < b then
P(b) is true. Then P( a) is true for all a ∈ A.
12

2 Topological space
When we discuss dependence of an object on another object we are often interested in
continuity of the dependence. This notion is often understood as that the dependent object
is confined in a given set if the independent object is confined in a certain corresponding
set. The sets that appear in this discussion are called open sets. Briefly, a topology is a
system of open sets. Topology is a general setting for discussion on continuity. This is a
generalization of the notion of continuity in metric spaces. Let us recall the notion of open
sets in metric spaces.

Metric space
Recall that, briefly, a metric space is a set equipped with a distance between every two
points. Namely, a metric space is a set X with a map d : X × X 7→ R such that for all
x, y, z ∈ X:

(a) d( x, y) ≥ 0 (distance is non-negative),

(b) d( x, y) = 0 ⇐⇒ x = y (distance is zero if and only if the two points coincide),

(c) d( x, y) = d(y, x ) (distance is symmetric),

(d) d( x, y) + d(y, z) ≥ d( x, z) (triangular inequality).

A ball is a set of the form B( x, r ) = {y ∈ X | d(y, x ) < r } where r ∈ R, r > 0.

In the theory of metric spaces, a subset U of X is said to be open if for all x in U there
is e > 0 such that B( x, e) is contained in U. This is equivalent to saying that a non-empty
open set is a union of balls.
We see at once that a union of open sets is an open set.
We can also see that the intersection of two balls is a union of balls. Indeed, let z ∈
B( x, r x ) ∩ B(y, ry ), let rz = min{r x − d(z, x ), ry − d(z, y)}, then the ball B(z, rz ) will be
inside both B( x, r x ) and B(y, ry ). Then we can deduce by induction that a finite intersection
of balls is a union of balls.
T∞
However, an infinite intersection of balls, such as n =1 B( x, n1 ) = { x }, may not be an
open set.
The notion of open sets and continuity in metric spaces have been very successful. We
shall deduce what we consider essential properties of these notions in order to generalize
them.

Topology
Continuity can be expressed through a notion of neighborhoods of points. In metric space,
the notion of neighborhood is expressed using distance. To generalize without the present
of distance, we directly specify which sets are the neighborhoods. Taking properties of
open sets in metric spaces, we propose:

Definition. A topology on a set X is a collection τ of subsets of X satisfying

(a) the sets ∅ and X are elements of τ,

(b) a union of elements of τ is an element of τ,

2. TOPOLOGICAL SPACE 13

(c) a finite intersection of elements of τ is an element of τ.

The elements of τ are called the open sets of X in the topology τ.

In short, a topology on a set X is a collection of subsets of X which includes ∅ and X

and is “closed” under unions and finite intersections.
A set X together with a topology τ is called a topological space, denoted by( X, τ ) or X
alone if we do not need to specify the topology. An element of X is often called a point.
A neighborhood (lân cận) of a point x ∈ X is a subset of X which contains an open set
containing x. Note that a neighborhood does not need to be open 3 .

Example. On any set X there is the trivial topology (tôpô hiển nhiên) {∅, X }. There is also
the discrete topology (tôpô rời rạc) whereas any subset of X is open. Thus on a set there
can be many topologies.

Notice that the statement “intersection of any finitely many open sets is open” is equiv-
alent to the statement “intersection of any two open sets is open”.
This generalization of metric space has shown to be general enough yet simple enough
for many purposes.

Example. A metric space is canonically a topological space with the topology generated by
the metric, i.e. a non-empty open sets are unions of balls in this metric. When we speak
about topology on a metric space we mean this topology.

Example (normed space). Recall that a normed space (không gian định chuẩn) is briefly
a vector space equipped with lengths of vectors. Namely, a normed space over the field of
real numbers is a set X with a structure of a vector space over the real numbers and a real
function X → R, x 7→ k x k, called a norm (chuẩn), satisfying:

(a) k x k ≥ 0 and k x k = 0 ⇐⇒ x = 0 (length is non-negative),

(b) kcx k = |c| k x k for c ∈ R (length is proportionate to vector),

(c) k x + yk ≤ k x k + kyk (triangle inequality).

A normed space is canonically a metric space with metric d( x, y) = k x − yk. Therefore a

normed space is canonically a topological space with the topology generated by the norm.

Example (Euclidean topology). In Rn = {( x1 , x2 , . . . , xn ) | xi ∈ R}, the Euclidean norm of

a point x = ( x1 , x2 , . . . , xn ) is k x k = ∑in=1 xi2 1/2 . The topology generated by this norm is
 

called the Euclidean topology (tôpô Euclid) of Rn .

A complement of an open set is called a closed set. We can check immediately:

Proposition (dual description of topology). In a topological space X:

(a) ∅ and X are closed,

(b) A finite union of closed sets is closed,

(c) An intersection of closed sets is closed.

3 Be careful that not everyone uses this convention. For instance Kelley [Kel55] uses this convention but

Munkres [Mun00] requires a neighborhood to be open.

14

Interior – Closure – Boundary

Let X be a topological space and let A be a subset of X. A point x in X is said to be:

(a) an interior point (điểm trong) of A in X if there is an open set of X containing x that
is contained in A,

(b) a contact point (điểm dính) (or point of closure) of A in X if any open set of X con-
taining x contains a point of A,

(c) a limit point (điểm tụ) (or cluster point, or accumulation point) of A in X if any open
set of X containing x contains a point of A other than x,

(d) a boundary point (điểm biên) of A in X if every open set of X containing x contains a
point of A and a point of the complement of A. In other words, a boundary point of
A is a contact point of both A and the complement of A.

With these notions we define:

(a) The set of all interior points of A is called the interior (phần trong) of A in X, denoted
by Å or int( A),

(b) The set of all contact points of A in X is called the closure (bao đóng) of A in X,
denoted by A or cl( A),

(c) The set of all boundary points of A in X is called the boundary (biên) of A in X,
denoted by ∂A.

Example. On the Euclidean line R, consider the subset A = [0, 1) ∪ {2}. Its interior is
intA = (0, 1), the closure is clA = [0, 1] ∪ {2}, the boundary is ∂A = {0, 1, 2}, the set of all
limit points is [0, 1].

Bases of a topology
Definition. Given a topology, a collection of open sets is a basis (cơ sở) for that topology if
every non-empty open set is a union of members of that collection.
More concisely, let τ be a topology of X, then a collection B ⊂ τ is called a basis for τ if
for any ∅ 6= V ∈ τ there is C ⊂ B such that V =
S
O∈C O.
So a basis of a topology is a subset of the topology that generates the entire topology via
unions. Specifying a basis is a more “efficient” way to give a topology.

Example. In a metric space the collection of all balls is a basis for the topology generated
by the metric.

Example. The Euclidean plane has a basis consisting of all open disks. It also has a basis
consisting of all open rectangles.

Definition. A collection S ⊂ τ is called a subbasis (tiền cơ sở) for the topology τ if the
collection of all finite intersections of members of S is a basis for τ.

Clearly a basis for a topology is also a subbasis for that topology. Briefly, given a topol-
ogy, a subbasis is a subset of the topology that can generate the entire topology by unions
and finite intersections.
2. TOPOLOGICAL SPACE 15

Example. Let X = {1, 2, 3}. The topology τ = {∅, {1, 2}, {2, 3}, {2}, {1, 2, 3}} has a basis
{{1, 2}, {2, 3}, {2}} and a subbasis {{1, 2}, {2, 3}}.

2.1 Example. The collection of all open rays, that are, sets of the forms ( a, ∞) and (−∞, a),
is a subbasis for the Euclidean topology of R.

Comparing topologies
On the set of all topologies on a set X we use the order relation by set inclusion. Namely,
let τ1 and τ2 be two topologies on X, if τ1 ⊂ τ2 we say that τ2 is finer (mịn hơn) (or stronger,
bigger) than τ1 and τ1 is coarser (thô hơn) (or weaker, smaller) than τ2 .

Example. On any set the trivial topology is the coarsest topology and the discrete topology
is the finest one.

Generating topologies
Suppose that we have a set and we want a topology such that certain subsets of that set are
open sets, how do find a topology for that purpose?

Theorem. Let S be a collection of subsets of X. The collection τ consisting of ∅, X, and all unions of
finite intersections of members of S is the smallest topology on X containing S, called the topology
generated by S. Concisely, letting B = { O | I ⊂ S, | I | < ∞} then the topology generated
T
O∈ I
by S is τ = { U | F ⊂ B}. The collection S ∪ { X } is a subbasis for this topology.
S
U∈F

Remark. In several textbooks to avoid adding the element X to S it is required that the
union of all members of S is X.

Proof. It is clear that any topology containing S must contain τ. We only need to check that
τ is a topology. First we check that τ is closed under unions. Let σ ⊂ τ, consider
S
A∈σ A.


We write A∈σ A = A∈σ U ∈ FA U , where FA ⊂ B. Since
S S S

 
[ [ [
 U = U,
A∈σ U ∈ FA U ∈(
S
A∈σ FA )

and since FA ⊂ B, we conclude that A ∈ τ.

S S
A∈σ A∈σ
We check that τ is closed under intersections of two elements. Let U and
S S
U∈F V ∈G V
be two elements of τ, where F, G ⊂ B. We can write
[ [ [
( U) ∩ ( V) = (U ∩ V ) .
U∈F V ∈G U ∈ F,V ∈ G

Let J = {U ∩ V | U ∈ F, V ∈ G }. Then J ⊂ B, and we can write

[ [ [
( U) ∩ ( V) = W,
U∈F V ∈G W∈J

showing that ( U) ∩ ( V ) ∈ τ.
S S
U∈F V ∈G

By this theorem, given a set, any collection of subsets generates a topology. This is a
remarkable property: if for a reason we want certain sets to be open, we can always build a
topology for that purpose.
16

Example. Let X = {1, 2, 3, 4}. The set {{1}, {2, 3}, {3, 4}} generates the topology

{∅, {1}, {3}, {1, 3}, {2, 3}, {3, 4}, {1, 2, 3}, {1, 3, 4}, {2, 3, 4}, {1, 2, 3, 4}}.

A basis for this topology is {{1}, {3}, {2, 3}, {3, 4}}.

Example (ordering topology). Let ( X, ≤) be a totally ordered set. The collection of subsets
of the forms { β ∈ X | β < α} and { β ∈ X | β > α} generates a topology on X, called the
ordering topology.

Example. The Euclidean topology on R is the ordering topology with respect to the usual
order of real numbers. (This is just a different way to state 2.1.)

Problems
2.2 (finite complement topology). X The finite complement topology on X consists of the empty set
and all subsets of X whose complements are finite. Check that this is indeed a topology.

2.3 (countable complement topology). X The countable complement topology on X consists of the
empty set and all subsets of X whose complements are countable. Check that this is indeed a topol-
ogy. Compare the countable complement topology to the finite complement topology.

2.4. Let X be a set and p ∈ X. Show that the collection consisting of ∅ and all subsets of X containing
p is a topology on X. This topology is called the Particular Point Topology on X, denoted by PPX p .
Describe the closed sets in this space.

2.5. Show that

(a) The interior of A in X is the largest open subset of X that is contained in A. A subset is open if
and only if all of its points are interior points.
(b) The closure of A in X is the smallest closed subset of X containing A. A subset is closed if and
only if it contains all of its contact points.

2.6. Let X be a topological space and A ⊂ X.

(a) Show that A is the disjoint union of Å and ∂A.
(b) Show that X is the disjoint union of Å, ∂A, and X \ A.

2.7. In a metric space X, a point x ∈ X is a limit point of the subset A of X if and only if there is a
sequence in A \ { x } converging to x. (This is not true in general topological spaces, see 5.5. )

2.8. Find the closures, interiors and the boundaries of the interval [0, 1) under the Euclidean, discrete
and trivial topologies of R.

2.9. Show that

(a) In a normed space, the boundary of the ball B( x, r ) is the sphere {y | k x − yk = r }, and so the
ball B0 ( x, r ) = {y | k x − yk ≤ r } is the closure of B( x, r ).
(b) In a metric space, the boundary of the ball B( x, r ) is a subset of the sphere {y | d( x, y) = r }. Is
the ball B0 ( x, r ) = {y | d( x, y) ≤ r } the closure of B( x, r )?

2.10. Let On = {k ∈ Z+ | k ≥ n}. Check that {∅} ∪ {On | n ∈ Z+ } is a topology on Z+ . Find the
closure of the set {5}. Find the closure of the set of all even positive integers.

2.11. Show that any open set in the Euclidean line is a countable union of open intervals.

2.12. In the real number line with the Euclidean topology, is the Cantor set (see 1.15) closed or open,
or neither? Find the boundary and the interior of the Cantor set (see 2.12).
2. TOPOLOGICAL SPACE 17

2.13. X Show that the intersection of a collection of topologies on a set X is a topology on X. If S is

a subset of X, then the intersection of all topologies of X containing S is the smallest topology that
contains S. Show that this is exactly the topology generated by S.

2.14. X A collection B of open sets is a basis if for each point x and each open set O containing x there
is a U in B such that U contains x and U is contained in O.

2.15. X Show that two bases generate the same topology if and only if each member of one basis is a
union of members of the other basis.

2.16. Let B be a collection of subsets of X. Then B ∪ { X } is a basis for a topology on X if and only if
the intersection of two members of B is either empty or is a union of some members of B. (In several
textbooks to avoid adding the element X to B it is required that the union of all members of B is X.)

2.17. In a metric space the set of all balls with rational radii is a basis for the topology. The set of all
1
balls with radii 2m , m ≥ 1 is another basis.

2.18 (Rn has a countable basis). X The set of all balls each with rational radius whose center has
rational coordinates forms a basis for the Euclidean topology of Rn .

2.19. Let d1 and d2 be two metrics on X. If there are α, β > 0 such that for all x, y ∈ X, αd1 ( x, y) ≤
d2 ( x, y) ≤ βd1 ( x, y) then the two metrics are said to be equivalent. Show that two equivalent metrics
generate same topologies.

2.20. Let ( X, d) be a metric space.

(a) Let d1 ( x, y) = min{d( x, y), 1}. Show that d1 is a metric on X generating the same topology as
that generated by d. Is d1 equivalent to d?
d( x,y)
(b) Let d2 ( x, y) = 1+d( x,y)
. Show that d2 is a metric on X generating the same topology as the
topology generated by d. Is d2 equivalent to d?

2.21 (all norms in Rn generate the Euclidean topology). In Rn denote by k·k2 the Euclidean norm,
and let k·k be any norm.

(a) Check that the map x 7→ k x k from (Rn , k·k2 ) to (R, k·k2 ) is continuous.
(b) Let Sn be the unit sphere under the Euclidean norm. Show that the restriction
of the map above
to Sn has a maximum value β and a minimum value α. Hence α ≤ k xxk ≤ β for all x 6= 0.

2

(c) Deduce that any two norms in Rn generate equivalent metrics, hence all norms in Rn generate
the Euclidean topology.

2.22. Is the Euclidean topology on R2 the same as the ordering topology on R2 with respect to the
dictionary order? If it is not the same, can the two be compared?

2.23 (the Sorgenfrey’s line). The collection of all intervals of the form [ a, b) generates a topology on
R. Is it the Euclidean topology?

2.24. * On the set of all integer numbers Z, consider arithmetic progressions

Sa,b = a + bZ,

where a ∈ Z and b ∈ Z+ .

(a) Show that these sets form a basis for a topology on Z.

(b) Show that with this topology each set Sa,b is closed.
(c) Show that if there are only finitely many prime numbers then the set {±1} is open.
(d) Conclude that there are infinitely many prime numbers. (This argument was given by Hillel
Furstenberg in 1955.)
18

3 Continuity

Continuous maps
Example (metric space). Let ( X, d X ) and (Y, dY ) be metric spaces. Recall that in the theory
of metric spaces, a map f : ( X, d X ) → (Y, dY ) is continuous at x ∈ X if and only if

∀e > 0, ∃δ > 0, ∀y ∈ X, d X (y, x ) < δ =⇒ dY ( f (y), f ( x )) < e,

equivalently,
∀e > 0, ∃δ > 0, ∀y ∈ BX ( x, δ), f (y) ∈ BY ( f ( x ), e),

or even shorter,
∀e > 0, ∃δ > 0, f ( BX ( x, δ)) ⊂ BY ( f ( x ), e),

in words, given any ball B( f ( x ), e) centered at f ( x ), there is a ball B( x, δ) centered at x

such that f brings B( x, δ) into B( f ( x ), e). So in metric spaces a function f is considered
continuous at x if f (y) can be arbitrarily close to f ( x ) provided that y is sufficiently close
to x. Thus continuity only needs a notion of neighborhood.
With a topology, we have specified a collection of sets to be the collection of neighbor-
hoods, instead of neighborhoods given via distance. Thus we can propose a generalization:

Definition. Let X and Y be topological spaces. We say a map f : X → Y is continuous

at a point x in X if for any open set U of Y containing f ( x ) there is an open set V of X
containing x such that f (V ) is contained in U. Writing Ux to indicate that x ∈ U, we can
write the definition in symbolic form as:

∀U f (x) ∈ τY , ∃Vx ∈ τX , f (V ) ⊂ U.

Clearly, in the definition we can replace open sets by neighborhoods.

We say that a map is continuous on the space if it is continuous at every point in the
space.

Example. It is apparent that this definition is equivalent to the definition of continuity in

metric spaces where the topologies are generated by the metric. In other words, if we look
at a metric space as a topological space with the topology generated by the metric then
continuity in the metric space is the same as continuity in the topological space. Therefore
we inherit all results concerning continuity in metric spaces.

Example. Let X and Y be topological spaces.

(a) The identity function, idX : X → X, x 7→ x, is continuous.

(b) The constant function, with a given a ∈ Y, x 7→ a, is continuous.

(c) If Y has the trivial topology then any map f : X → Y is continuous.

(d) If X has the discrete topology then any map f : X → Y is continuous.

Theorem. A map is continuous if and only if the inverse image of an open set is an open set.

Proof. (⇒) Suppose that f : X → Y is continuous. Let U be an open set in Y. Let x ∈

f −1 (U ). Since f is continuous at x and U is an open neighborhood of f ( x ), there is an open
3. CONTINUITY 19

set Vx containing x such that Vx is contained in f −1 (U ). Therefore f −1 (U ) =

S
x ∈ f −1 (U ) Vx
is open.
(⇐) Suppose that the inverse image of any open set is an open set. Let x ∈ X. Let U be
an open neighborhood of f ( x ). Then V = f −1 (U ) is an open set containing x, and f (V ) is
contained in U. Therefore f is continuous at x.

Topology generated by maps

In many circumstances we have situations where we have maps first and we want to build
structures so that these maps become continuous.
In one situation, let ( X, τX ) be a topological space, Y be a set, and f : X → Y be a map,
we want to find a topology on Y such that f is continuous. The requirement for such a
topology τY is that if U ∈ τY then f −1 (U ) ∈ τX . The trivial topology on Y satisfies that
requirement. It is the coarsest topology satisfying that requirement. On the other hand the
collection {U ⊂ Y | f −1 (U ) ∈ τX } is actually a topology on Y. This is the finest topology
satisfying that requirement.
In another situation, let X be a set, (Y, τY ) be a topological space, and f : X → Y be
a map, we want to find a topology on X such that f is continuous. The requirement for
such a topology τX is that if U ∈ τY then f −1 (U ) ∈ τX . The discrete topology on X is the
finest topology satisfying that requirement. The collection τX = { f −1 (U ) | U ∈ τY } is the
coarsest topology satisfying that requirement. We can observe further that if the collection
SY generates τY then τX is generated by the collection { f −1 (U ) | U ∈ SY }.

Example. On the set R with the finite complement topology, any polynomial is continuous,
but a trigonometric function such as the sin function is not continuous.

Subspace
Let ( X, τ ) be a topological space and let Y be a subset of X. We want to define a topology
on Y that can be naturally considered as being “inherited” from X. Thus any open set of X
that is contained in Y should be considered open in Y. If an open set of X is not contained
in Y then its restriction to Y should be considered open in Y. We can easily check that the
collection of restrictions of the open sets in X to Y is a topology on Y.

Definition. Let Y be a subset of the topological space X. The subspace topology on Y,

also called the relative topology (tôpô tương đối) with respect to X, is defined to be the
collection of restrictions of the open sets of X to Y, that is, the collection {O ∩ Y | O ∈ τ }.
With this topology we say that Y is a subspace (không gian con) of X.

In brief, a subset of a subspace Y of X is open in Y if and only if it is a restriction of an

open set of X to Y.

Remark. An open or a closed subset of a subspace Y of a space X is not necessarily open

or closed in X. For example, under the Euclidean topology of R, the set [0, 21 ) is open in the
subspace [0, 1], but is not open in R. When we say that a set is open, we must know which
topology we are using.

Proposition. Let X be a topological space and let Y ⊂ X. The subspace topology on Y is the
coarsest topology on Y such that the inclusion map i : Y ,→ X, x 7→ x is continuous. In other
words, the subspace topology on Y is the topology generated by the inclusion map from Y to X.
20

Proof. If O is a subset of X then i−1 (O) = O ∩ Y. Thus the topology generated by i is

{O ∩ Y | O ∈ τX }, exactly the subspace topology of Y.

Example (subspaces of a metric space). The notion of topological subspaces is compatible

with the earlier notion of metric subspaces. Let ( X, d) be a metric space and let Y ⊂ X.
Then, as we know, Y is a metric space with the metric inherited from X, namely dY =
d X |Y ×Y . A ball in Y is a set of the form, for a ∈ Y, r > 0:

BY ( a, r ) = {y ∈ Y | d(y, a) < r } = { x ∈ X | d( x, a) < r } ∩ Y = BX ( a, r ) ∩ Y.

Any open set A in Y is the union of a collection of balls in Y, i.e.

!
[ [ [
A= BY ( ai , ri ) = ( BX ( a i , r i ) ∩ Y ) = BX ( a i , r i ) ∩ Y,
i ∈ I, ai ∈Y i∈ I i∈ I

thus A is the intersection of an open set of X with Y. Conversely, if B is open in X, then for
each x ∈ B there is r x > 0 such that BX ( x, r x ) ⊂ B, so
! !
[ [
B∩Y = BX ( x, r x ) ∩Y = BX ( x, r x ) ∩Y
x∈B x ∈ B ∩Y
[ [
= ( BX ( x, r x ) ∩ Y ) = BY ( x, r x ).
x ∈ B ∩Y x ∈ B ∩Y

This implies that B ∩ Y is an open set in Y.

Example. For n ∈ Z+ define the sphere Sn to be the subspace of the Euclidean space Rn+1
consisting of all points of distance 1 to the origin:

Sn = {( x1 , x2 , . . . , xn+1 ) ∈ Rn+1 | x12 + x22 + · · · + xn2 +1 = 1}.

Homeomorphism
A map from one topological space to another is said to be a homeomorphism (phép đồng
phôi) if it is a bijection, is continuous and its inverse map is also continuous. Two spaces are
said to be homeomorphic (đồng phôi) if there is a homeomorphism from one to the other.
This is a basic relation among topological spaces.
It is easy to see that a homeomorphism brings an open set onto an open set, a closed set
onto a closed set, see 3.6.

Proposition. A homeomorphism between two spaces induces a bijection between the two topologies.

Proof. A homeomorphism f : ( X, τX ) → (Y, τY ) induces a map

f˜ : τX → τY
O 7→ f (O),

which is a bijection.

Roughly speaking, in the field of Topology, when two spaces are homeomorphic they
are considered the same. For example a “topological sphere” means a topological space
which is homeomorphic to a sphere.
3. CONTINUITY 21

3.1 Example. Any two balls in a normed space are homeomorphic via translations (tịnh
tiến) and dilations (scaling) (co dãn, vị tự).
Any ball in a normed space is homeomorphic to the whole space. We can consider
a map from the unit ball B(0, 1) to the whole space in the form x 7→ ϕ(k x k) k xxk , where
ϕ is a bijection from (0, 1) onto (0, ∞). Such a map ϕ has appeared in 1.4: we can take
√ t , or ϕ(t) = t
ϕ(t) =
1− t2 1− t .

An embedding (or imbedding) (phép nhúng) from a topological space X to a topological

space Y is a homeomorphism from X to a subspace of Y, i.e. it is a map f : X → Y such
that the restriction f˜ : X → f ( X ) is a homeomorphism. If there is an imbedding from X to
Y then we say that X can be embedded in Y.

Example. With the subspace topology the inclusion map is an embedding.

Example. The Euclidean line R can be embedded in the Euclidean plane R2 as a line in the
plane.

Example. Suppose that f : R → R is continuous under the Euclidean topology. Then R

can be embedded into the plane as the graph of f .

Example. The sphere Sn minus one point is homeomorphic to the Euclidean space Rn via
the stereographic projection (phép chiếu nổi), see fig 3.2.

x n +1
N

Figure 3.2: The stereographic projection.

Specifically, on the sphere Sn minus the North Pole N = (0, 0, . . . , 0, 1), each point x =
( x1 , x2 , . . . , xn+1 ) corresponds to the intersection y = (y1 , y2 , . . . , yn , 0) between the straight
line from N to x with the hyperplane xn+1 = 0. An intersection equation between the line
and the hyperplane is:
(1 − t) N + tx = y,
1
giving t = 1 − x n +1 . We get a formula for the stereographic projection:

Sn \ {(0, 0, . . . , 0, 1)} → Rn × { 0}
1
( x 1 , x 2 , . . . , x n +1 ) 7 → ( y 1 , y 2 , . . . , y n , 0 ), y i = x.
1 − x n +1 i
22

We can deduce the inverse map:

2yi −1 + ∑in=1 y2i

xi = , 1 ≤ i ≤ n, xn+1 = .
1 + ∑in=1 yi 2 1 + ∑in=1 y2i

Both maps are continuous thus the the stereographic projection is a homeomorphism.
We can also say that the n-dimensional Euclidean space can be embedded onto the n-
sphere minus one point.

Problems
3.3. X If f : X → Y and g : Y → Z are continuous then g ◦ f is continuous.

3.4. A map is continuous if and only if the inverse image of a closed set is a closed set.

3.5. X Suppose that f : X → Y and S is a subbasis for the topology of Y. Show that f is continuous
if and only if the inverse image of any element of S is an open set in X.

3.6. Define an open map to be a map such that the image of an open set is an open set. A closed map
is a map such that the image of a closed set is a closed set. Show that a homeomorphism is an open
map and is also a closed map.

3.7. Show that a continuous bijection is a homeomorphism if and only if it is an open map.

3.8. Show that ( X, PPX p ) and ( X, PPXq ) (see 2.4) are homeomorphic.

3.9.  Let X be a set and (Y, τ ) be a topological space. Let f i : X → Y, i ∈ I be a collection of maps.
Find the coarsest topology on X such that all maps f i , i ∈ I are continuous.
In Functional Analysis this construction is used to construct the weak topology on a normed
space. It is the coarsest topology such that all linear functionals which are continuous under the
norm are still continuous under the topology. See for instance [Con90].

3.10. Suppose that X is a normed space. Prove that the topology generated by the norm is exactly
the coarsest topology on X such that the norm and the translations (maps of the form x 7→ x + a) are
continuous.

3.11. X Show that a subset of a subspace Y of X is closed in Y if and only if it is a restriction of a

closed set in X to Y.
√ √
3.12. The set { x ∈ Q | − 2<x< 2} is both closed and open in Q under the Euclidean topology
of R.

3.13.  Suppose that X is a topological space and Z ⊂ Y ⊂ X. Then the relative topology of Z with
respect to Y is the same as the relative topology of Z with respect to X.

3.14. X Let X and Y be topological spaces and let f : X → Y.

(a) If Z is a subspace of X, denote by f | Z the restriction of f to Z. Show that if f is continuous then

f | Z is continuous.
(b) Let Z be a space containing Y as a subspace. Consider f as a function from X to Z, that is, let
f˜ : X → Z, f˜( x ) = f ( x ). Show that f is continuous if and only if f˜ is continuous.

3.15 (gluing continuous function). X Let X = A ∪ B where A and B are both open or are both closed
in X. Suppose f : X → Y, and f | A and f | B are both continuous. Then f is continuous.
Another way to phrase this is the following. Let g : A → Y and h : B → Y be continuous and
g( x ) = h( x ) on A ∩ B. Define 
 g ( x ), x∈A
f (x) =
 h ( x ), x ∈ B.
3. CONTINUITY 23

Then f is continuous.
Is it still true if the restriction that A and B are both open or are both closed in X is removed?

3.16. On the Euclidean plane R2 , show that the subspace ({0} × [0, 1]) ∪ ([0, 1] × {0}) is homeomor-
phic to the subspace {0} × [0, 1].

x2 y2
3.17. In the Euclidean plane an ellipse a2
+ b2
= 1 is homeomorphic to a circle.

3.18. In the Euclidean plane the upper half-plane {( x, y) ∈ R2 | y > 0} is homeomorphic to the
plane.

3.19. Is it true that any two balls in a metric space are homeomorphic?

3.20. Show that any two finite-dimensional normed spaces of same dimensions over the same fields
are homeomorphic.

3.21. Let kk1 and kk2 be two norms on Rn . Consider the map x 7→ k x k1 k xxk for x 6= 0 and sending
2
0 to 0.
(a) Prove that this map is a homeomorphism from (Rn , kk1 ) to (Rn , kk2 ).
(b) Deduce that balls with respect to different norms on Rn are homeomorphic.
(c) In particular deduce that in Rn any Euclidean ball is homeomorphic to an n-dimensional rect-
angle.

3.22. X If f : X → Y is a homeomorphism and Z ⊂ X then X \ Z and Y \ f ( Z ) are homeomorphic.

3.23. On the Euclidean plane R2 , show that:

(a) R2 \ {(0, 0)} and R2 \ {(1, 1)} are homeomorphic.

(b) R2 \ {(0, 0), (1, 1)} and R2 \ {(1, 0), (0, 1)} are homeomorphic.

Can you generalize these results?

3.24. Show that N and Z are homeomorphic under the Euclidean topology. More generally, prove
that any two set-equivalent discrete spaces are homeomorphic.

3.25. Among the spaces Z, Q, R, each with the Euclidean topology, which one is homeomorphic to
another?

3.26. Show that R with the finite complement topology and R2 with the finite complement topology
are homeomorphic.

3.27. Show that any homeomorphism from Sn−1 onto Sn−1 can be extended to a homeomorphism
from the unit disk D n = B0 (0, 1) onto D n .

3.28. Under the Euclidean topology the map ϕ : [0, 2π ) → S1 given by t 7→ (cos t, sin t) is a bijection
but is not a homeomorphism.

3.29. A space is homogeneous (đồng nhất) if given two points there exists a homeomorphism from
the space to itself bringing one point to the other point.

(a) Show that the sphere S2 is homogeneous.

(b) Show that being homogeneous is a topological property, meaning that if two spaces are home-
omorphic and one space is homogeneous then the other space is also homogeneous.

3.30. * An isometry (phép đẳng cấu metric, phép đẳng cấu hình học, hay phép đẳng cự) from a
metric space X to a metric space Y is a surjective map f : X → Y that preserves distance, that is
d( f ( x ), f (y)) = d( x, y) for all x, y ∈ X. If there exists such an isometry then X is said to be isometric
to Y.
24

(a) Show that an isometry is a homeomorphism.

(b) Show that being isometric is an equivalence relation among metric spaces.
(c) Show that (R2 , k · k∞ ) and (R2 , k · k1 ) are isometric, but they are not isometric to (R2 , k · k2 ),
although the three spaces are homeomorphic (see 2.21). (For generalization to higher dimen-
sions one may use the Mazur-Ulam theorem, see [P. Lax, Functional Analysis, John Wiley and
Sons, 2002, p. 49].)
4. CONNECTEDNESS 25

4 Connectedness
Intuitively when we say a space is connected to indicate that it has “one piece”, whereas a
space is disconnected when it has several disjoint, “independent pieces”. In this section we
will develop this notion precisely.
A topological space is said to be connected (liên thông) if it is not a union of two non-
empty disjoint open subsets.
Equivalently, a topological space is connected if and only if its only subsets which are
both closed and open are the empty set and the space itself.

Example. A space containing only one point is connected.

When we say that a subset of a topological space is connected we mean that the subset
under the subspace topology is a connected space.

Example. The Euclidean real number line minus a point is not connected.

4.1 Proposition. If a collection of connected subspaces of a space has non-empty intersection then
its union is connected.

Proof. Consider a topological space and let F be a collection of connected subspaces whose
intersection is non-empty. Let A be the union of the collection, A = D. Suppose that
S
D∈ F
C is a subset of A that is both open and closed in A. If C 6= ∅ then there is D ∈ F such that
C ∩ D 6= ∅. Then C ∩ D is a subset of D and is both open and closed in D (we are using
3.13 here). Since D is connected and C ∩ D 6= ∅, we must have C ∩ D = D. This implies
C contains the intersection of F. Therefore C ∩ D 6= ∅ for all D ∈ F. The argument above
shows that C contains all D in F, that is, C = A. We conclude that A is connected.

Connected component
Let us define a relation on a topological space X whereas two points are related if both
belong to a connected subspace of X (we then say that the two points are connected). Then
this relation is an equivalence relation, by 4.1.

Proposition. Under the above equivalence relation the equivalence class containing a point x is
equal to the union of all connected subspaces containing x, thus it is the largest connected subspace
containing x.

Proof. Consider the equivalence class [ x ] represented by a point x. By definition, y ∈ [ x ] if

and only if there is a connected set Oy containing both x and y. Since Oy ⊂ [ x ], we have
y∈[ x ] Oy . By 4.1, [ a] is connected.
S
[x] =

Under the above equivalence relation, the equivalence classes are called the connected
components of the space. Thus any space is a disjoint union of its connected components.

Proposition (continuous image of connected space is connected). If f : X → Y is continu-

ous and X is connected then f ( X ) is connected.

Proof. Suppose that U and V are non-empty disjoint open subset of f ( X ). Since f : X →
f ( X ) is continuous (see 3.14), f −1 (U ) and f −1 (V ) are open in X, and are non-empty and
disjoint. This contradicts the connectedness of X.
26

Theorem. If two spaces are homeomorphic then there is a bijection between the collections of con-
nected components of the two spaces.

Proof. Let f : X → Y be a homeomorphism. Set

f˜ : {[ x ] | x ∈ X } → {[y] | y ∈ Y }
[ x ] 7→ [ f ( x )].

We check that f˜ is a well-defined map. Suppose that y ∈ [ x ]. Since f is continuous, the

image f ([ x ]) is connected. Thus f (y) and f ( x ) belong to a connected subspace f ([ x ]), so
[ f (y)] = [ f ( x )].
The map f˜ has an inverse map

f −1 : {[y] | y ∈ Y } → {[ x ] | x ∈ X }
g

[y] 7→ [ f −1 (y)].

Therefore we have a bijection between the two sets.

From this result we say that connectedness is a topological property. We also say that
the number of connected components is a topological invariant. If two spaces have differ-
ent numbers of connected components then they must be different (that is, not homeomor-
phic).

Example (line not homeomorphic to plane). Suppose that R and R2 under the Euclidean
topologies are homeomorphic via a homeomorphism f . Delete any point x from R. By
3.22 the subspaces R \ { x } and R2 \ { f ( x )} are homeomorphic. But R \ { x } is not con-
nected while R2 \ { f ( x )} is connected (see 4.24), a contradiction. Thus R and R2 are not
homeomorphic.

Example (circle not homeomorphic to line). Suppose that there is a homemomorphism f

from the Euclidean line R to the circle S1 . Let x ∈ R. Then R \ { x } is homeomorphic to
S1 \ { f ( x )}. But they have different number of connected components, a contradiction.

4.2 Proposition. A connected subspace with a limit point added is still connected. Consequently
the closure of a connected subspace is connected, and any connected component is closed.

Proof. Let A be a connected subspace of a space X and let a ∈

/ A be a limit point of A,
we show that A ∪ { a} is connected. Suppose that A ∪ { a} = U ∪ V where U and V are
non-empty disjoint open subsets of A ∪ { a}. Suppose that a ∈ U. Then a ∈
/ V, so V ⊂ A.
Since a is a limit point of A, U ∩ A is non-empty. Then U ∩ A and V are open subsets of
A, by 3.13, which are non-empty and disjoint, while A = (U ∩ A) ∪ V. This contradicts the
assumption that A is connected.

Connected sets in the Euclidean real number line

Theorem. A subspace of the Euclidean real number line is connected if and only if it is an interval.

Proof. Suppose that a subset A of R is connected. Suppose that x, y ∈ A and x < y. If

x < z < y we must have z ∈ A, otherwise the set { a ∈ A | a < z} = { a ∈ A | a ≤ z} will be
both closed and open in A. Thus A contains the whole interval [ x, y].
Let a = inf A if A is bounded from below and a = −∞ otherwise. Similarly let b =
sup A if A is bounded from above and b = ∞ otherwise. Suppose that A contains more
4. CONNECTEDNESS 27

than one element, so that a < b. There are sequences ( an )n∈Z+ and (bn )n∈Z+ of elements
in A such that a < an < bn < b, and an → a while bn → b. By the above argument,
[ an , bn ] ⊂ A for all n. So ( a, b) ⊂ ∞ n=1 [ an , bn ] ⊂ A ⊂ [ a, b ]. It follows that A is either ( a, b )
S

or [ a, b) or ( a, b] or [ a, b].
We prove that any interval is connected. By homeomorphisms we just need to consider
the intervals (0, 1), (0, 1], and [0, 1]. Since [0, 1] is the closure of (0, 1), and (0, 1] = (0, 43 ) ∪
[ 12 , 1], it is sufficient to prove that (0, 1) is connected. Instead we prove that R is connected.
Suppose that R contains a non-empty, proper, open and closed subset C. Let x ∈ / C and
let D = C ∩ (−∞, x ) = C ∩ (−∞, x ]. Then D is both open and closed in R, and is bounded
from above.
If D 6= ∅, consider s = sup D. Since D is closed and s is a contact point of D, s ∈ D.
Since D is open s must belong to an open interval contained in D. But then there are points
in D which are bigger than s, a contradiction.
If D = ∅ we let E = C ∩ ( x, ∞), consider t = inf E and proceed similarly.

Example. Since the Euclidean Rn is the union of all lines passing through the origin, it is
connected.

Below is a simple application, a form of Intermediate value theorem (4.8):

Theorem (Borsuk–Ulam theorem). For any continuous real function on the sphere Sn there
must be antipodal (i.e. opposite) points where the values of the function are same.4

Proof. Let f : Sn → R be continuous. Let g( x ) = f ( x ) − f (− x ). Then g is continuous and

g(− x ) = − g( x ). Since Sn is connected (see 4.7), the range g(Sn ) is a connected subset of
the Euclidean R, and so it is an interval, containing the interval between g( x ) and g(− x ) =
− g( x ). Therefore 0 is in the range of g, so there is x0 ∈ Sn such that f ( x0 ) = f (− x0 ).

Path-connectedness
Path-connectedness is a more intuitive notion than connectedness. Shortly, a space is path-
connected if for any two points there is a path connecting them.

Definition. A path (đường đi) in a topological space X from a point x to a point y is a

continuous map α : [ a, b] → X such that α( a) = x and α(b) = y, where the interval of real
numbers [ a, b] has the Euclidean topology. The space X is said to be path-connected (liên
thông đường) if for any two points x and y in X there is a path in X from x to y.

Example. Any normed space is path-connected, and so is any convex subspace of that
space: any two points x and y are connected by a straight line segment x + t(y − x ), t ∈
[0, 1].

Example. In a normed space, the sphere S = { x | k x k = 1} is path-connected. One way

to show this is as follow. If two points x and y are not opposite then they can be connected
x +t(y− x )
by the arc k x +t(y− x )k
, t ∈ [0, 1]. If x and y are opposite, we can take a third point z, then
compose a path from x to z with a path from z to y.

Every point is path-connected to itself via a constant path.

If α is a path defined on [ a, b] then there is a path β defined on [0, 1] with the same
images (also called the traces of the paths): we can just use the linear homeomorphism
4 On the surface of the Earth at any moment there two opposite places where temperatures are same!
28

(1 − t) a + tb from [0, 1] to [ a, b] and let β(t) = α((1 − t) a + tb). For convenience we can
assume that the domains of paths is the interval [0, 1].
If there is a path α : [0, 1] → X from x to y then there is a path from y to x, for example
β : [0, 1] → X, β(t) = α(1 − t).
If α : [0, 1] → X is a path from x to y and β : [0, 1] → X is a path from y to z then there
is a path from x to z, for example

α(2t), 0 ≤ t ≤ 21 ,
γ(t) =
 β(2t − 1), 1
2 ≤ t ≤ 1.

This path follows α at twice the speed, then follow β at twice the speed and at half of a
unit time later. It is continuous by 3.15. There are other ways. For example we may take
β 1 : [1, 2] → X, β 1 (t) = β(t − 1), then take γ1 : [0, 2] → X,

 α ( t ), 0 ≤ t ≤ 1,
γ1 (t) =
 β ( t ),
1 1 ≤ t ≤ 2,

to go from x to y then to z.
We immediately obtain:

Proposition. The relation on a topological space X whereas a point x is related to a point y if there
is a path in X from x to y is an equivalence relation.

An equivalence class under the above equivalence relation is called a path-connected

component.

Theorem (path-connected =⇒ connected). Any path-connected space is connected.

Proof. This is a consequence of the fact that an interval on the Euclidean real number line is
connected. Let X be path-connected. Let x, y ∈ X. There is a path from x to y. The image of
this path is a connected subspace of X. That means every point y belongs to the connected
component containing x. Therefore X has only one connected component.

On the reverse direction we have:

4.3 Proposition. If a space is connected and every point has a path-connected neighborhood then it
is path-connected.

Proof. Suppose that X is connected and every point of X has a path-connected neighbor-
hood. Let C be a path-connected component of X. If x ∈ X is a contact point of C then
there is a path-connected neighborhood U in X of x such that U ∩ C 6= ∅. By 4.20, U ∪ C is
path-connected, so U ∪ C ⊂ C, thus U ⊂ C. This implies that every contact point of C is an
interior point of C, thus C is both open and closed in X. Since X is connected, C = X.

A topological space is said to be locally path-connected if every neighborhood of a point

contains an open path-connected neighborhood of that point.

Example. Open sets in a normed space are locally path-connected.

Corollary. A connected, locally path-connected space is path-connected.

4. CONNECTEDNESS 29

Topologist’s sine curve

The closure in the Euclidean plane of the graph of the function y = sin 1x , x > 0 is often
called the Topologist’s sine curve. This is a classic example of a space which is connected
but is not path-connected.

1
sin(1/x)

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.4: Topologist’s sine curve.

Denote A = {( x, sin 1x ) | x > 0} and B = {0} × [−1, 1]. Then the Topologist’s sine curve
is X = A ∪ B.

Proposition (connected 6= path-connected). The Topologist’s sine curve is connected but is not
path-connected.

Proof. By 4.9 the set A is connected. Each point of B is a limit point of A, so by 4.2 X is
connected.
Suppose that there is a path γ(t) = ( x (t), y(t)), t ∈ [0, 1] from the origin (0, 0) on B to a
point on A, we show that there is a contradiction.
Let t0 = sup{t ∈ [0, 1] | x (t) = 0}. Then x (t0 ) = 0, t0 < 1, and x (t) > 0 for all t > t0 .
Thus t0 is the moment when the path γ departs from B. We can see that the path jumps
immediately when it departs from B. Thus we will show that y(t) cannot be continuous
at t0 by showing that for any δ > 0 there are t1 , t2 ∈ (t0 , t0 + δ) such that y(t1 ) = 1 and
y(t2 ) = −1.
To find t1 , note that the set x ([t0 , t0 + 2δ ]) is an interval [0, x0 ] where x0 > 0. There exists
1 1
an x1 ∈ (0, x0 ) such that sin x1 = 1: we just need to take x1 = π + k2π with sufficiently large
2
1
k. There is t1 ∈ (t0 , t0 + δ
2] such that x (t1 ) = x1 . Then y(t1 ) = sin x ( t1 )
= 1. We can find t2
similarly.

Problems
4.5. A space is connected if whenever it is a union of two non-empty disjoint subsets, then at least
one set must contain a contact point of the other set.

4.6. Here is a different proof that any interval of real numbers is connected. Suppose that A and B
are non-empty, disjoint subsets of (0, 1) whose union is (0, 1). Let a ∈ A and b ∈ B. Let a0 = a,
a n + bn
b0 = b, and for each n ≥ 1 consider the middle point of the segment from an to bn . If 2 ∈ A then
a n + bn a n + bn
let an+1 = 2 and bn+1 = bn ; otherwise let an+1 = an and bn+1 = 2 . Then:

(a) The sequence ( an )n≥1 is a Cauchy sequence, hence is convergent to a number c.

30

(b) The sequence (bn )n≥1 is also convergent to c. This implies that (0, 1) is connected.

4.7. Find several ways to show that the sphere Sn is connected and is path-connected.

4.8 (intermediate value theorem). If X is a connected space and f : X → R is continuous, where R

has the Euclidean topology, then the image f ( X ) is an interval.
A consequence is the following familiar theorem in Calculus: Let f : [ a, b] → R be continuous
under the Euclidean topology. If f ( a) and f (b) have opposite signs then the equation f ( x ) = 0 has a
solution.

4.9. X If f : R → R is continuous under the Euclidean topology then its graph is connected in the
Euclidean plane. Moreover the graph is homeomorphic to R.

4.10. Let X be a topological space and let Ai , i ∈ I be connected subspaces. If Ai ∩ A j 6= ∅ for all
i, j ∈ I then Ai is connected.
S
i∈ I

4.11. Let X be a topological space and let Ai , i ∈ Z+ be connected subsets. If Ai ∩ Ai+1 6= ∅ for all
S∞
i ≥ 1 then i =1 Ai is connected.

4.12. Let A be a subspace of X with the particular point topology ( X, PPX p ) (see 2.4). Find the
connected components of A.

4.13. Let X be connected and let f : X → Y be continuous. If f is locally constant on X (meaning that
every point has a neighborhood on which f is a constant map) then f is constant on X.

4.14. Let X be a topological space. A map f : X → Y is called a discrete map if Y has the discrete
topology and f is continuous. Show that X is connected if and only if all discrete maps on X are
constant.

4.15. What are the connected components of N and Q under the Euclidean topology?

4.16. What are the connected components of Q2 as a subspace of the Euclidean plane?

4.17. Find the connected components of the Cantor set (see 2.12).

4.18. Show that if a space has finitely many components then each component is both open and
closed. Is it still true if there are infinitely many components?

4.19. X Suppose that a space X has finitely many connected components. Show that a map defined
on X is continuous if and only if it is continuous on each components. Is it still true if X has infinitely
many components?

4.20. If a collection of path-connected subspaces of a space has non-empty intersection then its union
is path-connected.

4.21. If f : X → Y is continuous and X is path-connected then f ( X ) is path-connected.

4.22. The path-connected component containing a point x is the union of all path-connected sub-
spaces containing x, thus it is the largest path-connected subspace containing x.

4.23. If two space are homeomorphic then there is a bijection between the collections of path-connected
components of the two spaces. In particular, if one space is path-connected then the other space is
also path-connected.

4.24. The plane with countably many points removed is path-connected under the Euclidean topol-
ogy.

4.25. Is the Euclidean R3 minus countably many straight lines connected?

4.26. A topological space is locally path-connected if and only if the collection of all open path-
connected subsets is a basis for the topology.
4. CONNECTEDNESS 31

4.27. Let X = {( x, x sin 1x ) | x > 0} ∪ {(0, 0)}, that is, the graph of the function x sin 1x , x > 0 with
the origin added. Under the Euclidean topology of the plane, is the space X connected or path-
connected?
1

0.8

0.6

0.4
sin(1/x)*x

0.2

-0.2

-0.4
0 0.2 0.4 0.6 0.8 1
x

4.28. The Topologist’s sine curve is not locally path-connected.

4.29. The set of n × n-matrix with real coefficients, denoted by M (n, R), could be naturally considered
2
as a subset of the Euclidean space Rn by considering entries of a matrix as coordinates, via the map

( ai,j ) 7−→ ( a1,1 , a2,1 , . . . , an,1 , a1,2 , a2,2 , . . . , an,2 , a1,3 , . . . , an−1,n , an,n ).

The Orthogonal Group O(n) is defined to be the group of matrices representing orthogonal linear
maps of Rn , that is, linear maps that preserve inner product. Thus

O(n) = { A ∈ M (n, R)| A · A T = In }.

The Special Orthogonal Group SO(n) is the subgroup of O(n) consisting of all orthogonal matri-
ces with determinant 1.

(a) Show that any element of SO(2) is of the form

!
cos ϕ − sin ϕ
R( ϕ) = .
sin ϕ cos ϕ

This is a rotation in the plane around the origin with an angle ϕ. Thus SO(2) is the group of
rotations on the plane around the origin.

(b) Show that SO(2) is path-connected.

(c) How many connected components does O(2) have?

(d) * It is known [F. Gantmacher, Theory of matrices, vol. 1, Chelsea, 1959, p. 285], that for any
matrix A ∈ O(n) there is a matrix P ∈ O(n) such that A = PBP−1 where B is a matrix of the
form  
R ( ϕ1 )
..
 
.
 
 
 
 R( ϕk ) 
.
 


 1 

 .. 

 . 

1
Using this fact, prove that SO(n) is path-connected.
32

4.30. Classify the alphabetical characters up to homeomorphisms, that is, which of the following
characters are homeomorphic to each other as subspaces of the Euclidean plane? Try to provide
rigorous arguments.

ABCDEFGHIJKLMNOPQRSTUVWXYZ
Note that the result depends on the font you use!
Do the same for the Vietnamese alphabetical characters:

AĂÂBCDĐEÊGHIKLMNOÔƠPQRSTUƯVXY
5. CONVERGENCE 33

5 Convergence

Separation
In this section we begin to put restrictions on topologies in terms of separation properties.
For example, we know that any metric on a set induces a topology on that set. If a
topology can be induced from a metric, we say that the topological space is metrizable.

Example. On any set X, the discrete topology is generated by the metric


1, x 6= y
d( x, y) =
0, x = y.

Indeed for any x ∈ X, the set containing one point { x } = B( x, 1) is open, therefore any
subset of X is open.
On the other hand, no metric can generate the trivial topology on X if X has more than
d( x,y)
one element. Indeed, if X has two different elements x and y then the ball B( x, 2 ) is a
non-empty open proper subset of X. Thus the trivial topology is not metrizable.

Definition (separation axioms). We define the following types of topological spaces:

T1 A topological space is called a T1 -space if for any two points x 6= y there is an open set
containing x but not y and an open set containing y but not x.

T2 A topological space is called a T2 -space or Hausdorff if for any two points x 6= y there
are disjoint open sets U and V such that x ∈ U and y ∈ V.

T3 A T1 -space is called a T3 -space or regular (chính tắc) if for any point x and a closed set
F not containing x there are disjoint open sets U and V such that x ∈ U and F ⊂ V. 5

T4 A T1 -space is called a T4 -space or normal (chuẩn tắc) if for any two disjoint closed sets
F and G there are disjoint open sets U and V such that F ⊂ U and G ⊂ V.

Proposition. A space is a T1 space if and only if any subset containing exactly one point is a closed
set.

Proof. If a space X is T1 , given x ∈ X, for any y 6= x there is an open set Uy that does not
contain x. Then X \ { x } = Uy . So X \ { x } is open. If any point constitutes a closed
S
y6= x
set, then two points y 6= x can be separated by the open sets X \ { x } and X \ {y}.

Proposition (T4 ⇒ T3 ⇒ T2 ⇒ T1 ). If a space is Ti then it is Ti−1 , for 2 ≤ i ≤ 4.

Proof. The inclusion of a T1 requirement to T3 and T4 ensures that a set containing exactly
one point is a special case of a closed set.

Example. Any space with the discrete topology is normal.

d( x,y)
Any metric space is a Hausdorff space, since if x 6= y then the balls B( x, 2 ) and
d( x,y)
B(y, 2 ) separate x and y. Thus if a space is not a Hausdorff space then it cannot be
metrizable.
We have a stronger result:
5 We include T requirement for regular and normal spaces, as in Munkres [Mun00]. Some authors such as
1
Kelley [Kel55] do not include the T1 requirement.
34

Proposition. Any metric space is normal.

Proof. Let ( X, d) be a metric space. For x ∈ X and A ⊂ X we define the distance from x to
A to be d( x, A) = inf{d( x, y) | y ∈ A}. This is a continuous function with respect to x, see
5.14.
Suppose that A and B are disjoint closed subsets of X. Let U = { x | d( x, A) < d( x, B)}
and V = { x | d( x, A) > d( x, B)}. Then A ⊂ U, B ⊂ V (using the fact that A and B are
closed, see 5.14), U ∩ V = ∅, and both U and V are open.

5.1 Example. The set of all real numbers under the finite complement topology is T1 but is
not T2 .

There are examples of a T2 -space which is not T3 , and a T3 -space which is not T4 , but
they are rather difficult, see 5.13, 5.24, [Mun00, p. 197] and [SJ70].

5.2 Proposition. A T1 -space is normal if and only if given a closed set C and an open set U
containing C there is an open set V such that C ⊂ V ⊂ V ⊂ U.

Roughly, this means that in a normal space given a closed set inside an open set we
always can enlarge the closed set.

Proof. Suppose that X is normal. Since X \ U is closed and disjoint from C there is an open
set V containing C and an open set W containing X \ U such that V and W are disjoint.
Then V ⊂ ( X \ W ), so V ⊂ ( X \ W ) ⊂ U.
In the reverse direction, given a closed set F and a closed set G disjoint from F, let
U = X \ G. There is an open set V containing F such that V ⊂ V ⊂ U. Then V and X \ V
separate F and G.

Sequence
Recall that a sequence in a set X is a map Z+ → X, a countably indexed family of elements
of X. Given a sequence x : Z+ → X the element x (n) is often denoted as xn , and sequence
is often denoted by ( xn )n∈Z+ or { xn }n∈Z+ .
When X is a metric space, the sequence is said to be convergent to x if xn can be as
closed to x as we want provided n is sufficiently large, i.e.

∀e > 0, ∃ N ∈ Z+ , n ≥ N =⇒ d( xn , x ) < e.

It is easy to see that this statement is equivalent to:

∀U open 3 x, ∃ N ∈ Z+ , n ≥ N =⇒ xn ∈ U.

This definition can be used in topological spaces.

However, in general topological spaces we need to use a notion more general than se-
quence. Roughly speaking, sequences (having countable indexes) might not be sufficient
for describing the neighborhood systems at a point. Arbitrary index is needed.

5.3 Example (sequence is not adequate for convergence). Consider R with the countable
complement topology. The point 0 is a contact point of the set (−∞, 0) since every open set
containing 0 has countable complement therefore cannot be disjoint from (−∞, 0) which is
uncountable. Let ( xn )n∈Z+ be any sequence in (−∞, 0). The set R \ { xn | n ∈ Z+ } is an
5. CONVERGENCE 35

open set containing 0 containing no element of the sequence ( xn )n∈Z+ . Thus the sequence
( xn )n∈Z+ cannot converge to 0. We conclude that in this space the notion of contact point
cannot be expressed by the notion of sequence.

Net
A directed set (tập được định hướng) is a (partially) ordered set such that for any two
indices there is an index greater than or equal to both. In symbols: ∀i, j ∈ I, ∃k ∈ I, k ≥
i ∧ k ≥ j.
A net (lưới), also called a generalized sequence (dãy suy rộng), in a space is a map from
a directed set to that space. In symbols, a net on a space X with index set a directed set I is
a map x : I → X. It is an element of the Cartesian product set ∏i∈ I X. Writing xi = x (i ) we
often denote the net as ( xi )i∈ I or { xi }i∈ I .

Example. Nets with index set I = Z+ with the usual order are exactly sequences.

Example. Let X be a topological space and x ∈ X. Let I be the collection of all open
neighborhoods of x. Define an order on I by U ≤ V ⇐⇒ U ⊃ V. Then I becomes a
directed set.

Definition. A net ( xi )i∈ I is said to be convergent (hội tụ) to x ∈ X if for each neighborhood
U of x there is an index i ∈ I such that if j ≥ i then x j belongs to U. The point x is called a
limit of the net ( xi )i∈ I and we often write xi → x.

Example. Convergence of nets with index set I = Z+ with the usual order is exactly con-
vergence of sequences.

Example. Let X = { x1 , x2 , x3 } with topology {∅, X, { x1 , x3 }, { x2 , x3 }, { x3 }}. The net ( x3 )

converges to x1 , x2 , x3 . The net ( x1 , x2 ) converges to x2 .

Example. If X has the trivial topology then any net in X is convergent to any point in X.

5.4 Proposition. A space is a Hausdorff space if and only if any net has at most one limit.

Proof. Suppose the space is a Hausdorff space. Suppose that a net ( xi ) is convergent to two
different limits x and y. Since the space is Hausdorff, there are disjoint open neighborhoods
U and V of x and y. There is i ∈ I such that for γ ≥ i we have xγ ∈ U, and there is j ∈ I
such that for γ ≥ j we have xγ ∈ U. Since there is a γ ∈ I such that γ ≥ i and γ ≥ j, the
point xγ will be in U ∩ V, a contradiction.
Suppose that the space is not a Hausdorff space, then there are two points x and y
that could not be separated by open sets. Consider the index set I whose elements are pairs
(U, V ) of open neighborhoods of x and y, with the order (U1 , V1 ) ≤ (U2 , V2 ) if U1 ⊃ U2 and
V1 ⊃ V2 . With this order the index set is directed. Since U ∩ V 6= ∅, take z(U,V ) ∈ U ∩ V.
 
Then the net z(U,V ) is convergent to both x and y, contradicting the uniqueness of
(U,V )∈ I
limit.

5.5 Proposition. A point x ∈ X is a contact point of a subset A ⊂ X if and only if there is a net
in A convergent to x. Consequently a subset is closed if and only if any limit of any net in that set
belongs to that set. A subset is open if and only if no limit of a net outside of that set belong to that
set.
36

This proposition allows us to describe topologies in terms of convergences. With it

many statements about convergence in metric spaces could be carried to topological spaces
by simply replacing sequences by nets.

Proof. (⇐=) Suppose that there is a net ( xi )i∈ I in A convergent to x. Let U be an open
neighborhood of x. There is an i ∈ I such that for j ≥ i we have x j ∈ U, in particular
xi ∈ U ∩ A. Thus x is a contact point of A.
( =⇒ ) Suppose that x is a contact point of A. Consider the directed set I consisting
of all the open neighborhoods of x with the partial order U ≤ V if U ⊃ V. For any open
neighborhood U of x there is an element xU ∈ U ∩ A. For any V ≥ U, xV ∈ V ⊂ U. Thus
( xU )U ∈ I is a net in A convergent to x. (This construction of the net ( xU )U ∈ I involves the
Axiom of choice.)

Remark. When can nets be replaced by sequences? By examining the above proof we can
see that the term net can be replaced by the term sequence if there is a countable collection F
of neighborhoods of x such that any neighborhood of x contains a member of F. In this case
the point x is said to have a countable neighborhood basis. A space having this property at
every point is said to be a first countable space. A metric space is such a space, where for
example each point has a neighborhood basis consisting of balls of rational radii. See also
5.3.

5.6 Proposition. Let τ1 and τ2 be two topologies on X. If convergence in τ1 implies convergence in

τ1 τ 2
τ2 then τ1 is finer than τ2 . In symbols: if for all nets xi and all points x, xi → x =⇒ xi → x, then
τ2 ⊂ τ1 . As a consequence, if convergences are same then topologies are same.

Proof. If convergence in τ1 implies convergence in τ2 then contact points in τ1 are contact

points in τ2 . Therefore closed sets in τ2 are closed sets in τ1 , and so are open sets.

Similarly to the case of metric spaces, we have:

Theorem. A function f is continuous at x if and only if for all nets ( xi ), xi → x =⇒ f ( xi ) →

f ( x ).

Proof. The proof is simply a repeat of the proof for the case of metric spaces.
( =⇒ ) Suppose that f is continuous at x. Let U is a neighborhood of f ( x ). Then f −1 (U )
is a neighborhood of x in X. Since ( xi ) is convergent to x, there is an i ∈ I such that for all
j ≥ i we have x j ∈ f −1 (U ), which implies f ( x j ) ∈ U.
(⇐=) We will show that if U is an open neighborhood in Y of f ( x ) then f −1 (U ) is a
neighborhood in X of x. Suppose the contrary, then x is not an interior point of f −1 (U ),
so it is a limit point of X \ f −1 (U ). By 5.5 there is a net ( xi ) in X \ f −1 (U ) convergent to
x. Since f is continuous, f ( xi ) ∈ Y \ U is convergent to f ( x ) ∈ U. This contradicts the
assumption that U is open.

Problems
5.7. If a finite set is a T1 -space then the topology is the discrete topology.

5.8. Is the space ( X, PPX p ) (see 2.4) a Hausdorff space?

5.9. Show that in a Hausdorff space three different points can be separated by open sets. Can this
statement be generalized?
5. CONVERGENCE 37

5.10. A T1 -space X is regular if and only if given a point x and an open set U containing x there is an
open set V such that x ∈ V ⊂ V ⊂ U.

5.11. Show that a subspace of a Hausdorff space is a Hausdorff space.

5.12. Show that a closed subspace of a normal space is normal.

5.13 (T2 but not T3 ). Show that the set R with the topology generated by all the subsets of the form
( a, b) and ( a, b) ∩ Q is a Hausdorff space but is not a regular space.

5.14 (distance function). Let ( X, d) be a metric space. For x ∈ X and A ⊂ X we define the distance
from x to A to be d( x, A) = inf{d( x, y) | y ∈ A}. More generally, for two subsets A and B of X, define
d( A, B) = inf {d( a, b) | a ∈ A, b ∈ B}.

(a) Check that d( x, A) is a continuous function with respect to x.

(b) Check that d( x, A) = 0 ⇐⇒ x ∈ A.
(c) Check that d( A, B) = inf{d( a, B) | a ∈ A} = inf{d( A, b) | b ∈ B}.
(d) When is d( A, B) = 0? Is d a metric?

5.15. Show that the separation properties Ti , 1 ≤ i ≤ 4, are topological properties, that means, if a
space X is Ti and is homeomorphic to a space Y, then Y is also Ti .

5.16. Let X be a normal space. Suppose that U1 and U2 are open sets in X satisfying U1 ∪ U2 = X.
Show that there is an open sets V1 such that V1 ⊂ U1 and V1 ∪ U2 = X. (This means in a normal
space given any open cover we can get a smaller cover.)

5.17. * Let X be normal, let f : X → Y be a surjective, continuous, and closed map. Prove that Y is a
normal space.

5.18. Let I = (0, ∞) ⊂ R. For i, j ∈ I, define i ≤ I j if i ≥R j (i is less than or is equal to j as indexes if

i is greater than or is equal to j as real numbers). On R with the Euclidean topology, consider the net
( xi = i )i∈ I . Is this net convergent?

5.19. On R with the finite complement topology and the usual order of real numbers, consider the
net ( xi = i )i∈R . Where does this net converge to?

5.20. Reconsider problems 2.19, 2.20 using 5.6.

5.21. Let Y be a T1 -space, and let f : X → Y be continuous. Suppose that A ⊂ X and f ( x ) = c on A,

where c is a constant. Show that f ( x ) = c on A, by:

(a) using nets.

(b) not using nets.

5.22. X Let Y be a Hausdorff space and let f , g : X → Y be continuous. Show that the set { x ∈
X | f ( x ) = g( x )} is closed in X, by:

(a) using nets.

(b) not using nets.

Show that, as a consequence, if f and g agree on a dense (dày đặc, trù mật) subspace of X (meaning
the closure of that subspace is X) then they agree on X.

5.23. * Let ( A, ≤) be a well-ordered uncountable set (see 1.20). The smallest element of A is be
denoted by 0. If A does not have a biggest element then add an element to A and define that element
to be the biggest one, denoted by ∞. For a, b ∈ A denote [ a, b] = { x ∈ A | a ≤ x ≤ b} and
[ a, b) = { x ∈ A | a ≤ x < b}. Thus we can write A = [0, ∞].
Let Ω be the smallest element of the set { a ∈ A | [0, a] is uncountable} (this set is non-empty since
it contains ∞).
38

(a) Show that [0, Ω) is uncountable, and for all a ∈ A, a < Ω the set [0, a] is countable.
(b) Consider [0, Ω] with the order topology. Show that Ω is a limit point of [0, Ω].
(c) Show that every countable subset of [0, Ω) is bounded in [0, Ω), therefore a sequence in [0, Ω)
cannot converge to Ω.

5.24 (Niemytzki space). * Let H = {( x, y) ∈ R2 | y ≥ 0} be the upper half-plane. Equip H with

the topology generated by the Euclidean open disks (i.e. open balls) in K = {( x, y) ∈ R2 | y > 0},
together with sets of the form { p} ∪ D where p is a point on the line L = {( x, y) ∈ R2 | y = 0} and D
is an open disk in K tangent to L at p. This is called the Niemytzki space.

(a) Check that this is a topological space.

(b) What is the subspace topology on L?
(c) What are the closed sets in H?
(d) Show that H is Hausdorff.
(e) Show that H is regular.
(f) Show that H is not normal.

5.25 (filter). * A filter (lọc) on a set X is a collection F of non-empty subsets of X such that:

(a) if A, B ∈ F then A ∩ B ∈ F,
(b) if A ⊂ B and A ∈ F then B ∈ F.

For example, given a point, the collection of all neighborhoods of that point is a filter.
A filter is said to be convergent to a point if every neighborhood of that point is an element of the
filter.
A filter-base (cơ sở lọc) is a collection G of non-empty subsets of X such that if A, B ∈ G then
there is C ∈ G such that C ⊂ ( A ∩ B).
If G is a filter-base in X then the filter generated by G is defined to be the collection of all subsets
of X each containing an element of G: { A ⊂ X | ∃ B ∈ G, B ⊂ A}.
For example, in a metric space, the collection of all open balls centered at a point is the filter-base
for the filter consisting of all neighborhoods of that point.
A filter-base is said to be convergent to a point if the filter generated by it converges to that point.

(a) Show that a filter-base is convergent to x if and only if every neighborhood of x contains an
element of the filter-base.
(b) Show that a point x ∈ X is a limit point of a subset A of X if and only if there is a filter-base of
X whose elements are subsets of A \ { x } convergent to x.
(c) Show that a map f : X → Y is continuous at x if and only if for any filter-base F which is
convergent to x, the filter-base f ( F ) is convergent to f ( x ).

Filter gives an alternative way to net for describing convergence. For more see [Dug66, p. 209], [Eng89,
p. 49], [Kel55, p. 83].

5.26 (Hausdorff distance). The Hausdorff distance between two bounded subsets A and B of a metric
space X is defined to be d H ( A, B) = max{supx∈ A d( x, B), supy∈ B d(y, A)}. Show that d H is a metric
on the set of all closed bounded subsets of X. (For more, see [BBI01, ch. 7].)

5.27 (Riemann integral by nets). * Let us define an n-dimensional rectangle to be a subset of Eu-
clidean space Rn of the form I = [ a1 , b1 ] × [ a2 , b2 ] × · · · × [ an , bn ] with ai < bi for all 1 ≤ i ≤ n. The
volume of the rectangle is defined to be the number | I | = (b1 − a1 )(b2 − a2 ) · · · (bn − an ). A partition
of the interval [ a, b] is a finite subset of [ a, b] which contains both a and b. We can label the elements
of the partition as x0 , x1 , . . . , xm with a = x0 < x1 < x2 < · · · < xm = b. Each interval [ xi−1 , xi ] is a
sub-interval of [ a, b] with respect to the partition. A partition of the the rectangle I = ∏in=1 [ ai , bi ] is a
5. CONVERGENCE 39

Cartesian product of partitions of [ ai , bi ], namely if Pi is a partition of [ ai , bi ] then P = ∏in=1 Pi is a par-

tition of I. A sub-rectangle with respect to the partition P of a rectangle I is a product ∏in=1 Ti where
Ti is a sub-interval of [ ai , bi ] with respect to the partition Pi . Let Rec( P) be the set of all sub-rectangles
with respect to the partition P.
(a) A choice of sample points in each sub-rectangle corresponding to P is a map

x ∗ : Rec( P) → Rn
R 7→ x ∗ ( R) ∈ R.

Let Par( I ) be the set of all pairs ( P, x ∗ ) consisting of a partition P a choice of sample points x ∗
corresponding to P. On Par( I ) we define the following relation:

( P, x ∗ ) ≤ ( P0 , x 0∗ ) ⇐⇒ P0 % P ∨ ( P, x ∗ ) = ( P0 , x 0∗ ) .

It represents a coarser and finer relation among partitions. Show that this is an order relation
on Par( I ) .
(b) If P = ∏in=1 Pi and P0 = ∏in=1 Pi0 , let P00 = ∏in=1 Pi00 with Pi00 = Pi ∪ Pi0 . Show that ( P, x ∗ ) ≤
( P00 , x 00∗ ) and ( P0 , x 0∗ ) ≤ ( P00 , x 00∗ ). Deduce that (Par( I ), ≤) is a directed set.
(c) With ( P, x ∗ ) ∈ Par( I ), let

Rsum( f , ( P, x ∗ )) = ∑ ( f ( x ∗ ( R))| R|,

R∈Rec( P)

be a Riemann sum corresponding to ( P, x ∗ ). Show that (Rsum( f , ( P, x ∗ )))( P,x∗ )∈Par( I ) is a net
in R.
(d) We say that f : I → R is Riemann integrable if lim Rsum( f , ( P, x ∗ )) exists, and in that case
Riemann integral of f is defined to be that real number. Show that the Riemann integral, if
exists, is unique.
(e) With ( P, x ∗ ) ∈ Par( I ), if f is bounded, we set

L( f , ( P, x ∗ )) = ∑ (inf f )| R|,
R
R∈Rec( P)

as a lower sum and

U ( f , ( P, x ∗ )) = ∑ (sup f )| R|
R∈Rec( P) R

as an upper sum. The function f : I → R is callled Darboux integrable if it is bounded and

lim L( f , ( P, x ∗ )) = lim U ( f , ( P, x ∗ )). In that case the Darboux integral of f is defined to be that
limit. Show that a function is Darboux integrable if and only if it is Riemann integrable, and
the two integrals are equal.
40

6 Compact space
A cover (phủ) of a set X is a collection of subsets of X whose union is X. A subset of a cover
which is itself a cover is called a subcover (phủ con). A cover is said to be an open cover if
each member of the cover is an open subset of X.

Definition. A space is compact if every open cover has a finite subcover. In symbols: a
space ( X, τ ) is compact if

O = X =⇒ (∃ J ⊂ I, | J | < ∞,
[ [
∀ I ⊂ τ, O = X ).
O∈ I O∈ J

The notion of compactness is important and common in mathematics. It often allows

for finite constructions.

Example. Any finite space is compact. Any space whose topology is finite (that is, the
space has finitely many open sets) is compact.

Example. On the Euclidean line R the collection {(−n, n) | n ∈ Z+ } is an open cover

without a finite subcover. Therefore the Euclidean line R is not compact.

Remark. Let A be a subspace of a topological space X. Let I be an open cover of A. Each

O ∈ I is an open set of A, so it is the restriction of an open set UO of X. Thus we have
a collection {UO | O ∈ I } of open sets of X whose union contains A. On the other hand
if we have a collection I of open sets of X whose union contains A then the collection
{U ∩ A | U ∈ I } is an open cover of A. For this reason, to simplify notation, we often use
the term open cover of a subspace A of X in both senses: either as an open cover of A or as
a collection of open subsets of the space X whose union contains A.

Theorem (continuous image of compact space is compact). If X is compact and f : X → Y

is continuous then f ( X ) is compact.

Proof. Let I be a cover of f ( X ) by open sets of Y (see the above remark). Then { f −1 (O) | O ∈
I } is an open cover of X. Since X is compact there is a finite subcover, so there is a finite set
J ⊂ I such that { f −1 (O) | O ∈ J } covers X. This implies f −1 ( O) = X, so applying f
S
O∈ J
to both sides we get O ⊃ f ( X ), hence J is a subcover of I.
S
O∈ J

In particular, compactness is preserved under homeomorphism. We say that compact-

ness is a topological property.

Proposition. Any closed subspace of a compact space is compact.

Proof. Suppose that X is compact and A ⊂ X is closed. Let I be an open cover of A by open
set of X. By adding the open set X \ A to I we get an open cover of X. This open cover has
a finite subcover. This subcover of X must contain X \ A, thus omitting this set we get a
finite subcover of A from I.

6.1 Proposition. Any compact subspace of a Hausdorff space is closed.

Proof. Let A be a compact set in a Hausdorff space X. We show that X \ A is open.

Let x ∈ X \ A. For each a ∈ A there are disjoint open sets Ua containing x and Va
containing a. The collection {Va | a ∈ A} covers A, so there is a finite subcover {Vai | 1 ≤
Tn Sn
i ≤ n}. Let U = i =1 Ua i and V = i =1 Vai . Then U is an open neighborhood of x disjoint
from V, a neighborhood of A.
6. COMPACT SPACE 41

Example. Notice this result is not true in general topological space. Any subspace of R
with the finite complement topology is compact. (Note that this space is not Hausdorff, see
5.1.)

Characterization of compact spaces in terms of closed subsets

In the definition of compact spaces by writing open sets as complements of closed sets, we
get a dual statement: A space is compact if for every collection of closed subsets whose in-
tersection is empty there is a a finite subcollection whose intersection is empty. A collection
of subsets of a set has the finite intersection property (tính giao hữu hạn) if the intersection
of every finite subcollection is non-empty. We get:

6.2 Proposition. A space is compact if and only if every collection of closed subsets with the finite
intersection property has non-empty intersection.

Proof. Let F be a collection of subsets of X, then C = ∅ if and only if C∈ F ( X \ C )

T S
C∈ F = X.
The collection F does not have the finite intersection property if and only if there is a finite
set I ⊂ F such that C∈ I C = ∅, which is equivalent to that C∈ I ( X \ C ) = X. Hence the
T S

existence of a collection of closed subsets with finite intersection property and with empty
intersection is equivalent to the existence of a collection of open subsets which covers X
with no finite subcover.

Compact metric spaces

A space is called sequentially compact (compắc dãy) if every sequence has a convergent
subsequence.

6.3 Lemma (Lebesgue’s number). In a sequentially compact metric space, for any open cover
there exists a number e > 0 such that any ball of radius e is contained in an element of the cover.

Proof. Let O be a cover of a sequentially compact metric space X. Suppose the opposite of
the conclusion, that is for any number e > 0 there is a ball B( x, e) not contained in any of
the element of O. Take a sequence of such balls B( xn , 1/n). The sequence ( xn )n∈Z+ has a
subsequence ( xnk )k∈Z+ converging to x. There is e > 0 such that B( x, 2e) is contained in
an element U of O. Take k sufficiently large such that nk > 1/e and xnk is in B( x, e). Then
B( xnk , 1/nk ) ⊂ B( xnk , e) ⊂ B( x, 2e) ⊂ U, a contradiction.

In general topological spaces compactness and sequentially compactness are different

notions, none implies the other. However for metric spaces the two notions coincide.

Theorem. A metric space is compact if and only if it is sequentially compact.

Proof. (⇒) Suppose that X is compact and let ( xn )n be a sequence in X. Suppose that
there is a point x ∈ X such that for every neighborhood Ux of x and every integer N
there is n > N such that xn in Ux . Let xn1 ∈ B( x, 1), and for each k ≥ 2 let nk be such
that nk > nk−1 and xnk ∈ B( x, 1k ). Then ( xnk )k is a subsequence of ( xn )n convergent to x.
Thus if the sequence ( xn )n∈Z+ has no convergent subsequence then for any point x ∈ X
there is an open neighborhood Ux of x and there is Nx ∈ Z+ such that if n ≥ Nx then
xn ∈
/ Ux . Since the collection {Ux | x ∈ X } covers the compact space X it has a finite
subcover {Uxk | 1 ≤ k ≤ m}. Let N = max{ Nxk | 1 ≤ k ≤ m}. If n ≥ N then xn ∈
/ Uxk for
all k, so xn ∈
/ X, a contradiction.
42

(⇐) First we show that for any e > 0 the space X can be covered by finitely many balls
of radii e (a property called total boundedness or pre-compact (tiền compắc)). Suppose the
contrary. Take x1 ∈ X, and inductively take xn+1 ∈ B( xi , e). Since d( xm , xn ) ≥ e if
S
/ 1≤ i ≤ n
m 6= n, the sequence ( xn )n≥1 cannot have any convergent subsequence, a contradiction.
Now let O be any open cover of X. By 6.3 there is a corresponding Lebesgue’s number
e such that a ball of radius e is contained in an element of O. The space X is covered by
finitely many balls of radii e. The collection of finitely many corresponding elements of O
covers X. Thus O has a finite subcover.

The above theorem means that we inherit all results obtained previously on compact-
ness in metric spaces. For examples, if a subspace of a metric space is compact then it is
closed and bounded, and a subspace of the Euclidean space Rn is compact if and only if it
is closed and bounded.

Compactification
A compactification (compắc hóa) of a space X is a compact space Y such that X is homeo-
morphic to a dense subspace of Y.

Example. The Euclidean interval [0, 1] is a compactification of the Euclidean interval [0, 1).

Example. A compactification of the Euclidean interval (0, 1) is the Euclidean interval [0, 1].

Example. A compactification of the Euclidean interval (0, 1) is the circle S1 , by adding just
one point, see figure 6.4.

We see that in some cases it is possible to compactify a space by adding just one point,
obtaining a one-point compactification. We now investigate whether it is possible to do
this in general. Figure 6.4 is a template for this construction.
Let X be a non-empty space. Since the set P ( X ) of all subsets of X cannot be contained
in X (see 1.5) there is an element of P ( X ) which is not in X. Let us denote that element by
∞, and let X ∞ = X ∪ {∞}. Let us see what a topology on X ∞ should be in order for X ∞
to contain X as a subspace and to be compact. If such a topology exists, then if an open
subset U of X ∞ does not contain ∞ then U is contained in X, therefore U is an open subset
of X in the subspace topology of X, which is the same as the original topology of X. If U
contains ∞ then its complement X ∞ \ U must be a closed subset of X ∞ , hence is compact,
furthermore X ∞ \ U is contained in X and is therefore a closed subset of X.

Theorem. The collection consisting of all open subsets of X and all complements in X ∞ of closed
compact subsets of X is the finest topology on X ∞ such that X ∞ is compact and contains X as a
subspace. If X is not compact then X is dense in X ∞ , and X ∞ is called the Alexandroff compact-
ification of X.6

Proof. We go through several steps.

(a) We check that we really have a topology.

Let I be a collection of closed compact sets in X. Then
S ∞ \ C) = X∞ \
T
C∈ I ( X C∈ I C,
where C ∈ I C is closed compact.
T

If O is open in X and C is closed compact in X then O ∪ ( X ∞ \ C ) = X ∞ \ (C \ O),

where C \ O is a closed and compact subset of X.
6 Proved in the early 1920s by Pavel Sergeyevich Alexandrov. Alexandroff is another way to spell his name.
6. COMPACT SPACE 43

Figure 6.4: A one-point compactification of the Euclidean interval (0, 1).

Also O ∩ ( X ∞ \ C ) = O ∩ ( X \ C ) is open in X.
If C1 and C2 are closed compact in X then ( X ∞ \ C1 ) ∩ ( X ∞ \ C2 ) = X ∞ \ (C1 ∪ C2 ),
where C1 ∪ C2 is closed compact.
So we do have a topology. With this topology X is a subspace of X ∞ .

(b) We show that X ∞ is compact. Let F be an open cover of X ∞ . Then an element O ∈ F

will cover ∞. The complement of O in X ∞ is a closed compact set C in X. Then F \ {O}
is an open cover of C. From this cover there is a finite cover. This finite cover together
with O is a finite cover of X ∞ .

(c) Since X is not compact and X ∞ is compact, X cannot be closed in X ∞ , therefore the
closure of X in X ∞ is X ∞ .

A space X is called locally compact if every point has a compact neighborhood. For
Hausdorff spaces, local compactness means every neighborhood of any point contains a
compact neighborhood of that point, see 6.29.

Example. The Euclidean space Rn is locally compact.

44

Proposition. The Alexandroff compactification of a locally compact Hausdorff space is Hausdorff.

Proof. Suppose that X is locally compact and is Hausdorff. We check that ∞ and x ∈ X can
be separated by open sets. Since X is locally compact there is a compact set C containing an
open neighborhood O of x. Since X is Hausdorff, C is closed in X. Then X ∞ \ C is open in
the Alexandroff compactification X ∞ . So O and X ∞ \ C separate x and ∞.

The need for the locally compact assumption is discussed in 6.28.

Proposition. If X is homeomorphic to Y then a Hausdorff one-point compactification of X is home-

omorphic to a Hausdorff one-point compactification of Y.

In particular, Hausdorff one-point compactification is unique up to homeomorphisms.

For this reason we can talk about the one-point compactification of a locally compact Haus-
dorff space.

Proof. Suppose that h : X → Y is a homeomorphism. Let X ∪ { a} and Y ∪ {b} be Hausdorff

one-point compactifications of X and Y. Let h̃ : X ∪ { a} → Y ∪ {b} be defined by h̃( x ) =
h( x ) if x 6= a and h̃( a) = b. We show that h̃ is a homeomorphism. We will prove that h̃ is
continuous. That the inverse map is continuous is similar, or we can use 6.11 instead.
Let U be an open subset of Y ∪ {b}. If U does not contain b then U is open in Y, so
h −1 (U ) is open in X, and so is open in X ∪ { a}. If U contains b then (Y ∪ {b}) \ U is closed
in Y ∪ {b}, which is compact, so (Y ∪ {b}) \ U = Y \ U is compact. Then h̃−1 ((Y ∪ {b}) \
U ) = h−1 (Y \ U ) is a compact subspace of X and therefore of X ∪ { a}. Since X ∪ { a} is a
Hausdorff space, h̃−1 ((Y ∪ {b}) \ U ) is closed in X ∪ { a}. Thus h̃−1 (U ) must be open in
X ∪ { a }.

Example. The Euclidean line R is homeomorphic to the circle S1 minus a point. The circle
is of course a Hausdorff one-point compactification of the circle minus a point. Thus a
Hausdorff one-point compactification (in particular, the Alexandroff compactification) of
the Euclidean line is homeomorphic to the circle.

Example. The one-point compactification of the Euclidean plane R2 is the sphere S2 . When
R2 is identified with the complex plane C its compactification S2 is often called the Rie-
mann sphere.

Problems
6.5. Any discrete compact topological space is finite.

6.6. In a topological space any finite unions of compact subspaces is compact.

6.7. In a Hausdorff space any intersection of compact subspaces is compact.

6.8 (extension of Cantor lemma in Calculus). Show that if X is compact and X ⊃ A1 ⊃ A2 ⊃ · · · ⊃

T∞
An ⊃ . . . is a descending sequence of closed, non-empty sets, then n =1 An 6= ∅.

6.9 (extreme value theorem). If X is a compact space and f : X → (R, Euclidean) is continuous then
f has a maximum value and a minimum value.

6.10 (uniformly continuous). A function f from a metric space to a metric space is uniformly con-
tinuous if for any e > 0, there is δ > 0 such that if d( x, y) < δ then d( f ( x ), f (y)) < e. Show that a
continuous function from a compact metric space to a metric space is uniformly continuous, by using
open covering.
6. COMPACT SPACE 45

6.11. X If X is compact, Y is Hausdorff, f : X → Y is bijective and continuous, then f is a homeo-

morphism.

6.12. In a compact space any infinite set has a limit point.

6.13. In a Hausdorff space a point and a disjoint compact set can be separated by open sets.

6.14. In a regular space a closed set and a disjoint compact set can be separated by open sets.

6.15. X In a Hausdorff space two disjoint compact sets can be separated by open sets.

6.16. X Any compact Hausdorff space is normal.

6.17. Prove 6.2.

6.18. Prove the existence of Lebesgue’s number 6.3 by using open coverings.

6.19. Find the one-point compactification of (0, 1) ∪ (2, 3) with the Euclidean topology, that is, de-
scribe this space more concretely.

6.20. A compactification of the Euclidean interval (0, 1) is the Topologist’s sine curve {( x, sin 1x ) | 0 <
x ≤ 1} ∪ {(0, y) | − 1 ≤ y ≤ 1} (see 4.4).

6.21. Find the one-point compactification of { n1 | n ∈ Z+ } under the Euclidean topology?

6.22. Find the one-point compactification of Z+ under the Euclidean topology? How about Z?

6.23. Show that Q is not locally compact (under the Euclidean topology of R). Is its Alexandroff
compactification Hausdorff?

6.24. What is the one-point compactification of the Euclidean open ball B(0, 1)? Find the one-point
compactification of the Euclidean space Rn .

6.25. What is the one-point compactification of the Euclidean annulus {( x, y) ∈ R2 | 1 < x2 + y2 <
2}?

6.26. Define a topology on R ∪ {±∞} such that it is a compactification of the Euclidean R.

6.27. Consider R with the Euclidean topology. Find a necessary and sufficient condition for a con-
tinuous function from R to R to have an extension to a continuous function from the one-point com-
pactification R ∪ {∞} to R.

6.28. If there is a topology on the set X ∞ = X ∪ {∞} such that it is compact, Hausdorff, and contain-
ing X as a subspace, then X must be Hausdorff, locally compact, and there is only one such topology
– the topology of the Alexandroff compactification.

6.29. We could have noticed that the notion of local compactness as we have defined is not appar-
ently a local property. For a property to be local, every neighborhood of any point must contain a
neighborhood of that point with the given property (as in the cases of local connectedness and local
path-connectedness). Show that for Hausdorff spaces local compactness is indeed a local property,
i.e., every neighborhood of any point contains a compact neighborhood of that point.

6.30. Any locally compact Hausdorff space is a regular space.

6.31. In a locally compact Hausdorff space, if K is compact, U is open, and K ⊂ U, then there is an
open set V such that V is compact and K ⊂ V ⊂ V ⊂ U. (Compare with 5.2.)

6.32. * A space is locally compact Hausdorff if and only if it is homeomorphic to an open subspace
of a compact Hausdorff space.

6.33. * Let X be a compact Hausdorff space. Let X1 ⊃ X2 ⊃ · · · be a nested sequence of closed

T∞
connected subsets of X. Show that i =1 Xi is connected.
46

6.34. A set with both a group structure and a topology such that the group operations are continuous
is called a topological group.
(a) Show that the circle S1 identified with the set of unit complex numbers under complex number
multiplication and the Euclidean topology is a topological group. Which is the identity element
in this group?
(b) Let G be a topological group. Let A and B be two subset of G. Let AB = { ab | a ∈ A, b ∈ B}.
Show that if A is open then AB is open. Show that if A and B are compact then AB is compact.
(c) Let e be the identity element of G. Show that if {e} is a closed set in G then G is a T1 -space.
(d) Let U be an open set containing e. Show that there is an open set V such that e ∈ V ⊂ U and
V · V ⊂ U. Let W = V ∩ V −1 . Suppose x ∈
/ U, show that xW ∩ W = ∅. Deduce that a T1
topological group is T2 .
(e) Show that if H is a subgroup of G then H is also a subgroup of G.
For more on topological group, see e.g.S. Morris, Topology without tears, Appendix 5].
7. PRODUCT OF SPACES 47

7 Product of spaces

Finite products of spaces

Let X and Y be two topological spaces, and consider the Cartesian product X × Y. The
product topology on X × Y is the topology generated by the collection F of sets of the form
U × V where U is an open set of X and V is an open set of Y. Since the intersection of two
members of F is also a member of F, the collection F is a basis for the product topology.
Thus every open set in the product topology is a union of products of open sets of X with
open sets of Y.

Remark. Notice a common error: to assume that an aribitrary open set in the product
topology is a product U × V.

The product topology on ∏in=1 ( Xi , τi ) is defined similarly to be the topology generated

by the collection {∏in=1 Ui | Ui ∈ τi }.

Proposition. If each bi is a basis for Xi then {∏in=1 Ui | Ui ∈ bi } is a basis for the product topology
on ∏in=1 Xi .

Proof. Consider an element in the above basis of the product topology, which is of the form
∏in=1 Vi where Vi ∈ τi . Each Vi can be written Vi = Ui j , where Ui j ∈ bi . Then
S
i j ∈ Ii

n n
∏ Vi = ∏ Uij .
[

i =1 i j ∈ Ii ,1≤i ≤n i =1

This proves our assertion.

Example (Euclidean topology). Recall that Rn = R × R × · · · × R. Let R have Euclidean

| {z }
n copies of R
topology, generated by open intervals. An open set in the product topology of Rn is a union
of products of open intervals. Since a product of open intervals is an open rectangle, and
an open rectangle is a union of open balls and vice versa, the product topology on Rn is
exactly the Euclidean topology.

Arbitrary products of spaces

Definition. Let (( Xi , τi )) i∈ I be an indexed family of topological spaces. The product topol-
ogy on the set ∏i∈ I Xi is the topology generated by the collection F consisting of all sets of
the form ∏i∈ I Ui , where Ui ∈ τi and Ui = Xi for all except finitely many i ∈ I. In symbols:

F = {∏ Ui | Ui ∈ τi , ∃ J ⊂ I, | J | < ∞, ∀i ∈ I \ J, Ui = Xi }.
i∈ I

Notice that the collection F above is a basis of the product topology, since it is closed
under finite intersections. The subcollection

{∏ Ui | Ui ∈ τi , ∃ j ∈ I, ∀i ∈ I \ { j}, Ui = Xi },
i∈ I

that is collection of all sets of the form ∏i∈ I Ui , where Ui ∈ τi and Ui = Xi for all except
one i ∈ I, is a subbasis for the product topology.
48

Recall that an element of the set ∏i∈ I Xi is written ( xi )i∈ I (see section 1). For j ∈ I the
projection to the j-coordinate is defined by p j : ∏i∈ I Xi → X j , p j (( xi )i∈ I ) = x j .
The definition of the product topology is explained in the following:

7.1 Theorem (product topology is the topology such that projections are continuous).
The product topology is the coarsest topology on ∏i∈ I Xi such that all the projection maps pi are
continuous. In other words, the product topology is the topology generated by the projection maps.

This result can be taken as a definition for the product topology.

Proof. Notice that if O j ∈ X j then p− 1

j (O j ) = ∏i ∈ I Ui with Ui = Xi for all i except j, and
Uj = O j . The topology generated by all the maps pi is the topology generated by all sets of
the form pi−1 (Oi ) with Oi ∈ τi , see 3.9.

7.2 Theorem (map to a product space is continuous if and only if each component map is
continuous). A map f : Y → ∏i∈ I Xi is continuous if and only if each component f i = pi ◦ f is
continuous.

Proof. If f : Y → ∏i∈ I Xi is continuous then pi ◦ f is continuous because pi is continuous.

On the other hand, let us assume that every f i is continuous. Let U = ∏i∈ I Ui where
Ui ∈ τi and ∃ j ∈ I such that Ui = Xi , ∀i 6= j be an element of the subbasis of ∏i∈ I Xi . Then

f −1 (U ) = f j−1 (Uj )

is an open set since f j is continuous. So f is continuous.

7.3 Theorem (convergence in product topology is coordinate-wise convergence). A net

n : J → ∏i∈ I Xi is convergent if and only if all of its projections pi ◦ n are convergent.

Proof. (⇐) Suppose that each pi ◦ n is convergent to ai , we show that n is convergent to

a = ( ai )i ∈ I .
A neighborhood of a contains an open set of the form U = ∏i∈ I Oi with Oi are open
sets of Xi and Oi = Xi except for i ∈ K, where K is a finite subset of I.
For each i ∈ K, pi ◦ n is convergent to ai , therefore there exists an index ji ∈ J such that
for j ≥ ji we have pi (n( j)) ∈ Oi . Take an index j0 such that j0 ≥ ji for all i ∈ K. Then for
j ≥ j0 we have n( j) ∈ U.

Tikhonov theorem
Theorem (Tikhonov theorem). The product of any family of compact spaces is compact. More
concisely, if Xi is compact for all i ∈ I then ∏i∈ I Xi is compact.7

Example. Let [0, 1] have the Euclidean topology. The space ∏i∈Z+ [0, 1] is called the Hilbert
cube. By Tikhonov theorem the Hilbert cube is compact.

Applications of Tikhonov theorem include the Banach-Alaoglu theorem in Functional

Analysis and the Stone-Cech compactification.
The proof we present below is rather difficult, however in the case of finite product it
can be proved more easily (7.19) and different techniques can be used in special cases of
this theorem (7.24 and 7.8).
7 Proved by Andrei Nicolaievich Tikhonov around 1926. The product topology was defined by him. His name

is also spelled as Tychonoff.

7. PRODUCT OF SPACES 49

Proof of Tikhonov theorem. Let Xi be compact for all i ∈ I. We will show that X = ∏i∈ I Xi
is compact by showing that if a collection of closed subsets of X has the finite intersection
property then it has non-empty intersection (see 6.2).8
Let F be a collection of closed subsets of X that has the finite intersection property. We
will show that A 6= ∅.
T
A∈ F
Have a look at the following argument, which suggests that proving the Tikhonov the-
orem might not be easy. If we take the closures of the projections of the collection F to the
i-coordinate then we get a collection { pi ( A), A ∈ F } of closed subsets of Xi having the
finite intersection property. Since Xi is compact, this collection has non-empty intersection.
From this it is tempting to conclude that F must have non-empty intersection itself. But that
is not true, see the figure.

In what follows we will overcome this difficulty by first enlarging the collection F.

(a) We show that there is a maximal collection F̃ of subsets of X such that F̃ contains F
and still has the finite intersection property. We will use Zorn lemma for this pur-
pose.9
Let K be the collection of collections G of subsets of X such that G contains F and has
the finite intersection property. On K we define an order by the usual set inclusion.
Now suppose that L is a totally ordered subcollection of K. Let H = G. We will
S
G∈ L
show that H ∈ K, therefore H is an upper bound of L.
First H contains F. We need to show that H has the finite intersection property. Sup-
pose that Hi ∈ H, 1 ≤ i ≤ n. Then Hi ∈ Gi for some Gi ∈ L. Since L is totally ordered,
there is an i0 , 1 ≤ i0 ≤ n such that Gi0 contains all Gi , 1 ≤ i ≤ n. Then Hi ∈ Gi0 for all
Tn
1 ≤ i ≤ n, and since Gi0 has the finite intersection property, we have i =1 Hi 6= ∅.

(b) Since F̃ is maximal, it is closed under finite intersection. Moreover if a subset of X has
non-empty intersection with every element of F̃ then it belongs to F̃.

(c) Since F̃ has the finite intersection property, for each i ∈ I the collection { pi ( A) | A ∈
F̃ } also has the finite intersection property, and so does the collection { pi ( A) | A ∈ F̃ }.
Since Xi is compact, pi ( A) is non-empty.
T
A∈ F̃

(d) Let xi ∈ pi ( A) and let x = ( xi )i∈ I ∈ ∏i∈ I [ A∈ F̃ pi ( A)]. We will show that
T T
A∈ F̃
x ∈ A for all A ∈ F̃, in particular x ∈ A for all A ∈ F.
We need to show that any neighborhood of x has non-empty intersection with every
A ∈ F̃. It is sufficient to prove this for neighborhoods of x belonging to the basis
of X, namely finite intersections of sets of the form pi−1 (Oi ) where Oi is an open
8A proof based on open covers is also possible, see [Kel55, p. 143].
9 This is a routine step; it might be easier for the reader to carry it out instead of reading.
50

neighborhood of xi = pi ( x ). For any A ∈ F̃, since xi ∈ pi ( A) we have Oi ∩ pi ( A) 6= ∅.

Therefore pi−1 (Oi ) ∩ A 6= ∅. By the maximality of F̃ we have pi−1 (Oi ) ∈ F̃, and the
desired result follows.

Stone-Cech compactification
Let X be a topological space. Denote by C ( X ) the set of all bounded continuous func-
tions from X to R where R has the Euclidean topology. By Tikhonov theorem the space
∏ f ∈C(X ) [inf f , sup f ] is compact. Define

Φ:X → ∏ [inf f , sup f ]

f ∈C ( X )
x 7→ ( f ( x )) f ∈C(X ) .

Thus for each x ∈ X and each f ∈ C ( X ), the f -coordinate of the point Φ( x ) is Φ( x ) f = f ( x ).

This means the f -component of Φ is f , i.e. p f ◦ Φ = f , where p f is the projection to the
f -coordinate.
Notice that the closure Φ( X ) is compact. Under additional assumptions we can check
that Φ is an embedding.
A space is said to be completely regular (also called a T3 1 -space) if it is a T1 -space and
2
for each point x and each closed set A with x ∈
/ A there is a map f ∈ C ( X ) such that
f ( x ) = a and f ( A) = {b} where a 6= b. Thus in a completely regular space a point and a
closed set disjoint from it can be separated by a continuous real function.

7.4 Theorem. If X is completely regular then Φ : X → Φ( X ) is a homeomorphism, i.e. Φ is an

embedding. In this case Φ( X ) is called the Stone-Cech compactification of X. It is a Hausdorff
space.

Proof. We go through several steps.

(a) Φ is injective: If x 6= y then since X is completely regular there is f ∈ C ( X ) such that

f ( x ) 6= f (y), therefore Φ( x ) 6= Φ(y).

(b) Φ is continuous: Since the f -component of Φ is f , which is continuous, the result

follows from 7.2.

(c) Φ−1 is continuous: We prove that Φ brings an open set onto an open set. Let U be an
open subset of X and let x ∈ U. There is a function f ∈ C ( X ) that separates x and
X \ U. In particular there is an interval ( a, b) containing f ( x ) such that f −1 (( a, b)) ∩
( X \ U ) = ∅. We have f −1 (( a, b)) = ( p f ◦ Φ)−1 (( a, b)) = Φ−1 ( p−f 1 (( a, b))) ⊂ U.
Apply Φ to both sides, we get p− 1 −1
f (( a, b )) ∩ Φ ( X ) ⊂ Φ (U ). Since p f (( a, b )) ∩ Φ ( X )
is an open set in Φ( X ) containing Φ( x ), we see that Φ( x ) is an interior point of Φ(U ).
We conclude that Φ(U ) is open.

That Φ( X ) is a Hausdorff space follows from that Y is a Hausdorff space, by 7.16, and
5.11.

Theorem. A bounded continuous real function on a completely regular space has a unique exten-
sion to the Stone-Cech compactification of the space.
7. PRODUCT OF SPACES 51

More concisely, if X is a completely regular space and f ∈ C ( X ) then there is a unique function
f˜ ∈ C (Φ( X )) such that f = f˜ ◦ Φ.

Φ / Φ( X )
X
f
 } f˜
R

Proof. A continuous extension of f , if exists, is unique, by 5.22.

Since p f ◦ Φ = f the obvious choice for f˜ is the projection p f .

Problems
7.5. Note that, as sets:

(a) ( A × B) ∩ (C × D ) = ( A ∩ C ) × ( B ∩ D ).

(b) ( A × B) ∪ (C × D ) ( A ∪ C ) × ( B ∪ D ) = ( A × B ) ∪ ( A × D ) ∪ ( C × B ) ∪ ( C × D ).

7.6. Check that in topological sense (i.e. up to homeomorphisms):

(a) R3 = R2 × R = R × R × R.

(b) More generally, is the product associative? Namely, is ( X × Y ) × Z = X × (Y × Z )? Is ( X ×

Y ) × Z = X × Y × Z?

7.7. Show that the sphere S2 with the North Pole and the South Pole removed is homeomorphic to
the infinite cylinder S1 × R.

7.8. Let ( Xi , di ), 1 ≤ i ≤ n be metric spaces. Let X = ∏in=1 Xi . For x = ( x1 , x2 , . . . , xn ) ∈ X and

y = (y1 , y2 , . . . , yn ) ∈ X, define

δ1 ( x, y) = max{di ( xi , yi ) | 1 ≤ i ≤ n},

!1/2
n
δ2 ( x, y) = ∑ d i ( x i , y i )2 .
i =1
Show that δ1 and δ2 are metrics on X generating the product topology.

7.9. If for each i ∈ I the space Xi is homeomorphic to the space Yi then ∏i∈ I Xi is homeomorphic to
∏i∈ I Yi .

7.10. X Show that a space X is Hausdorff if and only if the diagonal ∆ = {( x, x ) ∈ X × X } is closed
in X × X, by:

(a) using nets,

(b) not using nets.

7.11. Show that if Y is Hausdorff and f : X → Y is continuous then the graph of f , that is, the set
{( x, f ( x )) | x ∈ X }, is closed in X × Y, by:

(a) using nets,

(b) not using nets.

7.12. X Show that the graph of a continuous function is homeomorphic to its domain. Namely, let
f : X → Y be continuous, and let graph( f ) = {( x, f ( x )) | x ∈ X } ⊂ X × Y, then graph( f ) is
homemomorphic to X.
52

7.13. Show that each projection map pi is a an open map, mapping an open set onto an open set. Is it
a closed map?

7.14. X Let A1 and A2 be topological spaces. On the set ( A1 × {1}) ∪ ( A2 × {2}) consider the topol-
ogy generated by subsets of the form U × {1} and V × {2} where U is open in A1 and V is open in
A2 . Show that A1 × {1} is homeomorphic to A1 , while A2 × {2} is homeomorphic to A2 . Notice that
( A1 × {1}) ∩ ( A2 × {2}) = ∅. The space ( A1 × {1}) ∪ ( A2 × {2}) is called the disjoint union (hội
rời) of A1 and A2 , denoted by A1 t A2 . We can use this construction when for example we want to
consider a space consisting of two disjoint circles.
More generally, given a collection of spaces (( Xi , τi ))i∈ I the disjoint union of this collection is the
space × {i }) with the topology × {i }. Check that this is indeed a topology.
S S
i ∈ I ( Xi i ∈ I τi

7.15. X Fix a point O = (Oi ) ∈ ∏i∈ I Xi . Define the inclusion map f : Xi → ∏i∈ I Xi by

O
j if j 6= i
x 7→ f ( x ) with f ( x ) j = .
x if j = i

Show that f is a homeomorphism onto its image X̃i (an embedding of Xi ). Thus X̃i is a copy of Xi in
∏i∈ I Xi . The spaces X̃i have O as the common point. This is analogous to the coordinate system Oxy
on R2 .

7.16. X Show that

(a) If each Xi , i ∈ I is a Hausdorff space then ∏i∈ I Xi is a Hausdorff space.

(b) If ∏i∈ I Xi is a Hausdorff space then each Xi is a Hausdorff space.

7.17. Show that

(a) If ∏i∈ I Xi is path-connected then each Xi is path-connected.

(b) If each Xi , i ∈ I is path-connected then ∏i∈ I Xi is path-connected.

7.18. X Show that

(a) If ∏i∈ I Xi is connected then each Xi is connected.

(b) If X and Y are connected then X × Y is connected.

(c) * If each Xi , i ∈ I is connected then ∏i∈ I Xi is connected.

7.19. Show that

(a) If ∏i∈ I Xi is compact then each Xi is compact.

(b) * If X and Y are compact then X × Y is compact (of course without using the Tikhonov theo-
rem).

7.20. Let X be a normed space over a field F which is R or C. Check that the addition ( x, y) 7→ x + y
is a continuous map from the product space X × X to X, while the scalar multiplication (c, x ) 7→ c · x
is a continuous map from F × X to X. This is an example of a topological vector space, and is a
special case of topological groups (see 7.22).

7.21. Let X = [0, 1]N = ∏n∈N [0, 1] where [0, 1] ⊂ R is equipped with the Euclidean topology (the
Hilbert cube).

(a) Show that X is separable, meaning that it contains a countable dense subset.

(b) Show that X contains a countable basis.

(c) Show that X is normal. (Thus it is metrizable, by Urysohn metrizability theorem 9.49.)
7. PRODUCT OF SPACES 53

(d) With x = ( xn )n∈N and y = (yn )n∈N let

∞
1
d( x, y) = ∑ 2 n
d ( x n , y n ).
n =0

Show that this is a metric generating the topology of X.

7.22. The set of n × n-matrix with real coefficients, denoted by M(n, R), could be naturally considered
2
as a subset of the Euclidean space Rn by considering entries of a matrix as coordinates, via the map

( ai,j ) 7−→ ( a1,1 , a2,1 , . . . , an,1 , a1,2 , a2,2 , . . . , an,2 , a1,3 , . . . , an−1,n , an,n ).

Let GL(n, R) be the set of all invertible n × n-matrices with real coefficients.

(a) Show that taking product of two matrices is a continuous map on GL(n, R).
(b) Show that taking inverse of a matrix is a continuous map on GL(n, R).
(c) Show that GL(n, R) is a topological group (see 6.34). It is called the General Linear Group.

7.23 (Zariski topology). * Let F = R or F = C.

A polynomial in n variables on F is a function from Fn to F that is a finite sum of terms of the
form ax1m1 x2m2 · · · xnmn , where a, xi ∈ F and mi ∈ N. Let P be the set of all polynomials in n variables
on F.
If S ⊂ P then define Z (S) to be the set of all common zeros of all polynomials in S, thus Z (S) =
{ x ∈ Fn | ∀ p ∈ S, p( x ) = 0}. Such a set is called an algebraic set.

(a) Show that if we define that a subset of Fn is closed if it is algebraic, then this gives a topology
on Fn , called the Zariski topology.
(b) Show that the Zariski topology on F is exactly the finite complement topology.
(c) Show that if both F and Fn have the Zariski topology then all polynomials on Fn are continu-
ous.
(d) Is the Zariski topology on Fn the product topology?

The Zariski topology is used in Algebraic Geometry.

7.24. Using the characterization of compact subsets of Euclidean spaces, prove the Tikhonov theorem
for finite products of compact subsets of Euclidean spaces.

7.25. Using the characterization of compact metric spaces in terms of sequences, prove the Tikhonov
theorem for finite products of compact metric spaces.

7.26. Any completely regular space is a regular space.

7.27. Prove 7.4 using nets.

7.28. A space is completely regular if and only if it is homeomorphic to a subspace of a compact

Hausdorff space. As a corollary, a locally compact Hausdorff space is completely regular.
By 7.28 if a space has a Hausdorff Alexandroff compactification then it also has a Hausdorff Stone-
Cech compactification. In a certain sense, for a noncompact space the Alexandroff compactification
is the “smallest” Hausdorff compactification of the space and the Stone-Cech compactification is the
“largest” one. For more discussions on this topic see for instance [Mun00, p. 237].
54

8 Real functions and Spaces of functions

In this section the set R is assumed to have the Euclidean topology.

Urysohn lemma
8.1 Theorem (Urysohn lemma). If X is normal, F is closed, U is open, and F ⊂ U ⊂ X, then
there exists a continuous map f : X → [0, 1] such that f ( x ) = 0 on F and f ( x ) = 1 on X \ U.

Equivalently, if X is normal, A and B are two disjoint closed subsets of X, then there is a
continuous function f from X to [0, 1] such that f ( x ) = 0 on A and f ( x ) = 1 on B. Thus in
a normal space two disjoint closed subsets can be separated by a continuous real function.

Example. For metric spaces we can take

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

If we take g = 1 − f we get a continuous function g from X to [0, 1] such that g| F = 1

and g| X \U = 0. The graph of g can be visualized as a continuous connection between a
plateau F at height 1 with the lowland outside U at height 0.

Proof. The proof goes through the following steps:

(a) Construct a family of open sets in the following manner (recalling 5.2):
Let U1 = U.

n = 0 F ⊂ U0 ⊂ U0 ⊂ U1 .
n = 1 U0 ⊂ U 1 ⊂ U 1 ⊂ U1 .
2 2

n = 2 U0 ⊂ U 1 ⊂ U 1 ⊂ U 2 = U 1 ⊂ U 2 ⊂ U 3 ⊂ U 3 ⊂ U 4 = U1 .
4 4 4 2 4 4 4 4

Inductively,

F ⊂ U0 ⊂ U0 ⊂ U 1 ⊂ U 1n ⊂ U 2n ⊂ U 2n ⊂ U 3n ⊂ U 3n ⊂ · · · ⊂
2n 2 2 2 2 2

⊂ U 2n −1 ⊂ U 2n −1 ⊂ U 2n = U1 .
2n 2n 2n

(b) Let I = { 2mn | m, n ∈ N; 0 ≤ m ≤ 2n }. We have a family of open sets {Ur | r ∈ I }

having the property r < s ⇒ Ur ⊂ Us . We can check that I is dense in [0, 1] (this is
really the same thing as that any real number in [0, 1] can be written in binary form,
compare 1.17).
8. REAL FUNCTIONS AND SPACES OF FUNCTIONS 55

(c) Define f : X → [0, 1],


inf{r ∈ I | x ∈ U } if x ∈ U,
r
f (x) =
1 if x ∈
/ U.

In this way if x ∈ Ur then f ( x ) ≤ r, while if x ∈

/ Ur then f ( x ) ≥ r. So f ( x ) gives the
“level” of x on the scale from 0 to 1, while Ur is like a sublevel set of f .

x f

Us U1 = U
F U0 Ur 0 r f (x) s 1

We prove that f is continuous, so f is the function we are looking for. It is enough to

prove that sets of the form { x | f ( x ) < a} and { x | f ( x ) > a} are open.

(d) If a ≤ 1 then f ( x ) < a if and only if there is r ∈ I such that r < a and x ∈ Ur . Thus
{ x | f ( x ) < a} = { x ∈ Ur | r < a} = is open.
S
r < a Ur

(e) If a < 1 then f ( x ) > a if and only if there is r ∈ I such that r > a and x ∈
/ Ur . Indeed,
if f ( x ) > a then there is s > a such that f ( x ) ≥ s, which in turn implies that if x ∈ Ur
then r ≥ s, so for a < r < s we have x ∈
/ Ur . On the other direction, if r > a and
x∈
/ Ur then x ∈ Us implies s ≥ r, so f ( x ) ≥ r > a.
Thus { x | f ( x ) > a} = { x ∈ X \ Ur | r > a} = X \ Ur . Now we show that
S
r>a
r > a X \ Ur = r > a X \ Ur , which implies that r > a X \ Ur is open. Indeed, if r ∈ I
S S S

and r > a then there is s ∈ I such that r > s > a. Then Us ⊂ Ur , therefore X \ Ur ⊂
X \ Us . This implies r> a X \ Ur ⊂ r> a X \ Ur . The opposite inclusion is true since
S S

X \ Ur ⊃ X \ Ur .

Partition of unity
An important application of the Urysohn lemma is the existence of partition of unity (phân
hoạch đơn vị). A partition of unity is a collection of real functions which are non-negative,
each function is zero outside a given open set, and the sum of these functions is 1 every-
where. Roughly, a partition of unity is a system of relative weights of each open set at each
point (so that total weight at each point is 100%, that is, 1). Figure 8.2 illustrate a simple
case.
56

0 1 2 3 4 5 6

Figure 8.2: A partition of unity for the interval [0, 6] corresponding to the two open sets
[0, 5) and (1, 6].

For f : X → R, the support (giá) of f , denoted by supp( f ), is defined to be the closure

of the subset { x ∈ X | f ( x ) 6= 0}.

8.3 Theorem (partition of unity). Let X be a normal space. Suppose that X has a finite open cover
O. Then there is a collection of continuous maps ( f U : X → [0, 1])U ∈O such that supp( f U ) ⊂ U
and for every x ∈ X we have ∑U ∈O f U ( x ) = 1.

00 )
Proof. From problem 5.16, there is an open cover (UU U ∈O of X such that for each U ∈ O
0 satisfying U 00 ⊂ U 00 ⊂ U 0 ⊂ U 0 ⊂ U. By Urysohn lemma there is
there is an open UU U U U u
a continuous map ϕU : X → [0, 1] such that ϕU |U 00 = 1 and ϕU | X \U 0 = 0. This implies
U U
00 contains x, therefore
supp( ϕU ) ⊂ Uu0 ⊂ U. For each x ∈ X there is U ∈ O such that UU
ϕU ( x ) = 1. Let
ϕU
fU = .
∑U ∈O ϕU

Partition of unity allows us to extend some local properties to global ones, by “patching”
neighborhoods.

The compact-open topology

Let X and Y be two topological spaces. We say that a net ( f i )i∈ I of functions from X to Y
converges point-wise to a function f : X → Y if for each x ∈ X the net ( f i ( x ))i∈ I converges
to f ( x ).
Now let Y be a metric space. Recall that a function f : X → Y is said to be bounded if
the set of values f ( X ) is a bounded subset of Y. We consider the set B( X, Y ) of all bounded
functions from X to Y. If f , g ∈ B( X, Y ) then we define a metric on B( X, Y ) by d( f , g) =
sup{d( f ( x ), g( x )) | x ∈ X }. The topology generated by this metric is called the topology of
uniform convergence, often studied in Functional Analysis. If a net ( f i )i∈ I converges to f in
the metric space B( X, Y ) then we say that ( f i )i∈ I converges to f uniformly.

Proposition. Suppose that ( f i )i∈ I converges to f uniformly. Then:

(a) ( f i )i∈ I converges to f point-wise.

(b) If each f i is continuous then f is continuous.

Proof. The proof is the same as the proof for metric spaces. We write down a proof for the
second part. Suppose that each f i is continuous. Let x ∈ X, we prove that f is continuous
at x. The key step is the following inequality:

d( f ( x ), f (y)) ≤ d( f ( x ), f i ( x )) + d( f i ( x ), f i (y)) + d( f i (y), f (y)).

8. REAL FUNCTIONS AND SPACES OF FUNCTIONS 57

Given e > 0, fix an i ∈ I such that d( f i , f ) < e. For this i, there is a neighborhood U of x
such that if y ∈ U then d( f i ( x ), f i (y)) < e. The above inequality implies that for y ∈ U we
have d( f ( x ), f (y)) < 3e.

Definition. Let X and Y be two topological spaces. Let C ( X, Y ) be the set of all continuous
functions from X to Y. The topology generated by all sets of the form

S( A, U ) = { f ∈ C ( X, Y ) | f ( A) ⊂ U }

where A ⊂ X is compact and U ⊂ Y is open is called the compact-open topology on

C ( X, Y ).

Proposition. If X is compact and Y is a metric space then on C ( X, Y ) the compact-open topology

is the same as the uniform convergence topology.

This says that the compact-open topology is a generalization of the uniform convergence
topology to topological spaces.

Proof. Given f ∈ C ( X, Y ) and e > 0, we show that the ball B( f , e) ⊂ C ( X, Y ) in the uni-
form metric contains an open neighborhood of f in the compact-open topology. For each
x ∈ X there is an open set Vx containing x such that f (Vx ) ⊂ B( f ( x ), e/3). Since X is
Sn
compact, there are finitely many xi , 1 ≤ i ≤ n, such that i =1 Vxi ⊃ X and f (Vxi ) ⊂
B( f ( xi ), e/3). Noticing that f (V xi ) ⊂ f (Vxi ) ⊂ B( f ( xi ), e/3) ⊂ B( f ( xi ), e/2), we deduce
that f ∈ in=1 S(V xi , B( f ( xi ), e/2)). Now if g ∈ S(V xi , B( f ( xi ), e/2)) then for x ∈ Vxi we
T

have d( g( x ), f ( x )) ≤ d( g( x ), f ( xi )) + d( f ( xi ), f ( x )) < e. Thus in=1 S(V xi , B( f ( xi ), e/2)) ⊂

T

B( f , e). See a demonstration of this idea in figure 8.4.

B( f , e)

f (x)

Vx
x X

Figure 8.4:

In the opposite direction, we need to show that every open neighborhood of f in the
compact-open topology contains a ball B( f , e) in the uniform metric. It is sufficient to
show that for open neighborhoods of f of the form S( A, U ). For each x ∈ A there is a ball
B( f ( x ), ex ) ⊂ U. Since f ( A) is compact, there are finitely many xi ∈ A and ei > 0, 1 ≤ i ≤
Sn
n, such that B( f ( xi ), ei ) ⊂ U and i =1 B( f ( xi ), ei /2) ⊃ f ( A). Let e = min{ei /2 | 1 ≤ i ≤
58

n}. Suppose that g ∈ B( f , e). For each x ∈ A, there is an i such that f ( x ) ∈ B( f ( xi ), ei /2).
Then
ei
d( g( x ), f ( xi )) ≤ d( g( x ), f ( xi )) + d( f ( xi ), f ( x )) < e + ≤ ei ,
2
so g( x ) ∈ U. Thus g ∈ S( A, U ), and so B( f , e) ⊂ S( A, U ). See a demonstration of this idea
in figure 8.5.

U B( f , e)

f ( A)

A X

Figure 8.5:

Tiestze extension theorem

Below is another application of Urysohn lemma.

Theorem (Tiestze extension theorem). Let X be a normal space. Let F be closed in X. Let
f : F → R be continuous. Then there is a continuous map g : X → R such that g| F = f .

Thus in a normal space a continuous real function on a closed subspace can be extended
continuously to the whole space.

Proof. First consider the case where f is bounded.

(a) The general case can be reduced to the case when infF f = 0 and supF f = 1. We will
restrict our attention to this case.

(b) By Urysohn lemma, there is a continuous function g1 : X → [0, 13 ] such that


0 if x ∈ f −1 ([0, 13 ])
g1 ( x ) =
1
3 if x ∈ f −1 ([ 32 , 1]).

Let f 1 = f − g1 . Then supX g1 = 13 , supF f 1 = 23 , and infF f 1 = 0.

8. REAL FUNCTIONS AND SPACES OF FUNCTIONS 59

(c) Inductively, once we have a function f n : F → R, for a certain n ≥ 1 we will obtain a

n
function gn+1 : X → [0, 31 23 ] such that

n
0 if x ∈ f n−1 ([0, 31 23 ])
g n +1 ( x ) =
2 n

n +1 2
n
1 if x ∈ f n−1 ([ 23


3 3 , 3 ]).

1 2 n 2 n +1



Let f n+1 = f n − gn+1 . Then supX gn+1 = 3 3 , sup F f n+1 = 3 , and infF f n+1 =
0.

(d) The series ∑∞

n=1 gn converges uniformly to a continuous function g.

(e) Since f n = f − ∑in=1 gi , the series ∑nn=1 gn | F converges uniformly to f . Therefore

g| F = f .

(f) Note that with this construction infX g = 0 and supX g = 1.

Now consider the case when f is not bounded.

(a) Suppose that f is neither bounded from below nor bounded from above. Let h be a
homeomorphism from (−∞, ∞) to (0, 1). Then the range of f 1 = h ◦ f is a subset of
(0, 1), therefore it can be extended as in the previous case to a continuous function g1
such that infx∈X g1 ( x ) = infx∈ F f 1 ( x ) = 0 and supx∈X g1 ( x ) = supx∈ F f 1 ( x ) = 1.
If the range of g1 includes neither 0 nor 1 then g = h−1 ◦ g1 will be the desired func-
tion.
It may happens that the range of g1 includes either 0 or 1. In this case let C =
g1−1 ({0, 1}). Note that C ∩ F = ∅. By Urysohn lemma, there is a continuous func-
tion k : X → [0, 1] such that k|C = 0 and k| F = 1. Let g2 = kg1 + (1 − k) 21 . Then
g2 | F = g1 | F and the range of g2 is a subset of (0, 1) (g2 ( x ) is a certain convex combi-
nation of g1 ( x ) and 12 ). Then g = h−1 ◦ g2 will be the desired function.

(b) If f is bounded from below then similarly to the previous case we can use a homeo-
morphism h : [ a, ∞) → [0, 1), and we let C = g1−1 ({1}).
The case when f is bounded from above is similar.

Problems
8.6. Show that a normal space is completely regular. So: normal ⇒ completely regular ⇒ regular. In
other words: T4 ⇒ T3 1 ⇒ T3 .
2

8.7. Prove the following version of Urysohn lemma, as stated in [Rud86]. Suppose that X is a locally
compact Hausdorff space, V is open in X, K ⊂ V, and K is compact. Then there is a continuous
function f : X → [0, 1] such that f ( x ) = 1 for x ∈ K and supp( f ) ⊂ V.

8.8 (point-wise convergence topology). Now we view a function from X to Y as an element of the
set Y X = ∏ x∈ X Y. In this view a function f : X → Y is an element f ∈ Y X , and for each x ∈ X the
value f ( x ) is the x-coordinate of the element f .

(a) Let ( f i )i∈ I be a net of functions from X to Y, i.e. a net of points in Y X . Show that ( f i )i∈ I
converges to a function f : X → Y point-wise if and only if the net of points ( f i )i∈ I converges
to the point f in the product topology of Y X .
60

(b) Define the point-wise convergence topology on the set Y X of functions from X to Y as the
topology generated by sets of the form

S( x, U ) = { f ∈ Y X | f ( x ) ∈ U }

with x ∈ X and U ⊂ Y is open. Show that the point-wise convergence topology is exactly the
product topology on Y X .

8.9. Let X and Y be two topological spaces. Let C ( X, Y ) be the set of all continuous functions from
X to Y. Show that if a net ( f i )i∈ I converges to f in the compact-open topology of C ( X, Y ) then it
converges to f point-wise.

8.10 (continuity of functions of two variables). * Let Y, Z be topological spaces and let X be a locally
compact Hausdorff space. Recall that it is not true that a map on a product space is continuous if it is
continuous on each variable. Prove that a map

f : X×Z → Y
( x, t) 7→ f ( x, t)

is continuous if and only if the map

ft : X → Y
x 7→ f t ( x ) = f ( x, t)

is continuous for each t ∈ Z and the map

Z → C ( X, Y )
t 7→ ft

is continuous with the compact-open topology on C ( X, Y ).

8.11. Show that the Tiestze extension theorem implies the Urysohn lemma.

8.12. The Tiestze extension theorem is not true without the condition that the set F is closed.

8.13. Show that the Tiestze extension theorem can be extended to maps to the space ∏i∈ I R where R
has the Euclidean topology.

8.14. Let X be a normal space and F be a closed subset of X. Then any continuous map f : F → Sn
can be extended to an open set containing F.
9. QUOTIENT SPACE 61

9 Quotient space

We study gluing spaces to form new spaces.

Example. Gluing the two endpoints of a line segment together we get a circle, see figure
9.1.

Figure 9.1: Gluing the two end-points of a line segment gives a circle.

Mathematically speaking, gluing elements mean to identify them as one. If two ele-
ments are identified as one, they have a relation which should be reflective, symmetric,
transitive, that is, an equivalence relation. For a set X and an equivalence relation ∼ on X,
taking the quotient set X/ ∼ is exactly what we mean by gluing. The gluing map is exactly
the quotient map bringing each element to its equivalence class, p : X → X/ ∼, x 7→ [ x ].
We want gluing to be a continuous operation. That means when X is a topological
space we want to equip the quotient set X/ ∼ with a topology such that the gluing map is
continuous. We take the finest such topology, namely, a subset U of X/ ∼ is open if and
only if the preimage p−1 (U ) is open in X (see 3.9). The set X/ ∼ with this topology is called
the quotient space of X by this equivalence relation.

Example. In a special case, if A is a subspace of X then there is this equivalence relation on

X: x ∼ x if x ∈ X, and x ∼ y if x, y ∈ A. The quotient space X/ ∼ is often written as X/A,
and we can think of it as being obtained from X by collapsing the whole subset A to one
point.

9.2 Proposition. Let Y be a topological space. A map f : X/ ∼→ Y is continuous if and only if

62

f ◦ p is continuous.
p
X / X/ ∼

f
" 
f ◦p
Y

Proof. The map f ◦ p is continuous if and only if for each open subset U of Y, the set ( f ◦
p)−1 (U ) = p−1 ( f −1 (U )) is open in X. The latter statement is equivalent to that f −1 (U ) is
open for every U, that is, f is continuous.

The following result will provide us a tool for identifying quotient spaces:

Theorem. Suppose that X is compact and ∼ is an equivalence relation on X. Suppose that Y is

Hausdorff, and f : X → Y is continuous and onto. Suppose that f ( x1 ) = f ( x2 ) if and only if
x1 ∼ x2 . Then f induces a homeomorphism from X/ ∼ onto Y.

Proof. Define h : X/ ∼→ Y by h([ x ]) = f ( x ). Then h is onto and is injective, thus it is a

bijection.

p
X / X/ ∼

h
" 
f
Y
Notice that f = h ◦ p (in such a case people often say that the above diagram is com-
mutative, and that the map f can be factored). By 9.2 h is continuous. By 6.11, h is a
homeomorphism.

Example. Gluing the two end-points of a line segment gives a circle, see figure 9.1. More
precisely, consider the interval [0, 1] on the Euclidean real number line. Take the minimal
equivalence relation ∼ for which 0 ∼ 1, namely we only need to add ∀ x ∈ (0, 1), x ∼ x
(see 9.46 for more on minimal equivalence relation). We often denote the quotient space by
this relation by [0, 1]/0 ∼ 1. Then [0, 1]/0 ∼ 1 is homeomorphic to S1 via the map h in the
following diagram:
p
[0, 1] / [0, 1]/0 ∼ 1

h
% 1
f
S
Here f is the map t 7→ (cos(2πt), sin(2πt)). The map f is continuous, onto, and it fails to
be injective only at t = 0 and t = 1. Since in the quotient set 0 and 1 are identified, the
induced map h on the quotient set becomes a bijection. The above theorem allows us to
check that h is a homeomorphism.

Example. Gluing a pair of opposite edges of a square gives a cylinder. Let X = [0, 1] ×
[0, 1]/ ∼ where (0, t) ∼ (1, t) for all 0 ≤ t ≤ 1, then X is homeomorphic to the cylinder
[0, 1] × S1 . The homeomorphism is induced by the map (s, t) 7→ (s, cos(2πt), sin(2πt)).
9. QUOTIENT SPACE 63

Figure 9.3: This space is homemomorphic to the cylinder.

Example. Gluing opposite edges of a square gives a torus. The torus10 (mặt xuyến) is
[0, 1] × [0, 1]/ ∼ where (s, 0) ∼ (s, 1) and (0, t) ∼ (1, t) for all 0 ≤ s, t ≤ 1.

Figure 9.4: The torus.

A of subspace R3 homeomorphic to T 2 can be obtained as the surface of revolution

obtained by revolving a circle around a line not intersecting it. Suppose that the circle is on

Figure 9.5: The torus embedded in R3 .

the Oyz-plane, the center is on the y-axis and the axis for the rotation is the z-axis. Let a be
the radius of the circle, b be the distance from the center of the circle to O, (a < b). Let S be
the surface of revolution, then the embedding can be given by

[0, 2π ] × [0, 2π ] / T2

h
& 
f
S

where f (φ, θ ) = ((b + a cos θ ) cos φ, (b + a cos θ ) sin φ, a sin θ ). Thus the torus can be em-
bedded in R3 .
10 The plural form of the word torus is tori.
64

a
O b y
φ

Figure 9.6: Equation for this embedded torus: ( x, y, z) = ((b + a cos θ ) cos φ, (b +
p 2
a cos θ ) sin φ, a sin θ ), or x 2 + y2 − b + z2 = a2 .

We can also describe the torus as the quotient of the Euclidean plane by the relation
∀m, n ∈ Z, ( x, y) ∼ ( x + m, y + n), see problem 9.44 and figure 9.7.

Figure 9.7: The torus as a quotient of the plane.

9.8 Example. Gluing a pair of opposite edges of a square in opposite directions gives the
Mobius band (dải, lá, mặt Mobius 11 ). More precisely the Mobius band is X = [0, 1] ×
[0, 1]/ ∼ where (0, t) ∼ (1, 1 − t) for all 0 ≤ t ≤ 1.

11 Möbius or Moebius are other spellings for this name.

9. QUOTIENT SPACE 65

Figure 9.9: The Mobius band.

Figure 9.10: The Mobius band embedded in R3 .

The Mobius band could be embedded in R3 . It is homeomorphic to a subspace of R3

obtained by rotating a straight segment around the z-axis while also turning that segment
“up side down”. The embedding can be induced by the map (see Figure 9.11)

(s, t) 7→ (( a + t cos(s/2)) cos s, ( a + t cos(s/2)) sin s, t sin(s/2)),

with 0 ≤ s ≤ 2π and −1 ≤ t ≤ 1.

t
a

s/2

Figure 9.11: An embedding of the Mobius band in R3 .

66

The Mobius band is famous as an example of un-orientable surfaces (mặt không định
hướng được). It is also one-sided (một phía). A proof is at 25.4.

Example. Identifying one pair of opposite edges of a square and the other pair in opposite
directions gives a topological space called the Klein bottle. More precisely it is [0, 1] ×
[0, 1]/ ∼ with (0, t) ∼ (1, t) and (s, 0) ∼ (1 − s, 1).

Figure 9.12: The Klein bottle.

It is known that the Klein bottle cannot be embedded in R3 . The figure 9 does not
present an embedded Klein bottle in R3 , instead only an immersed Klein bottle in R3 . An
immersion (phép nhúng chìm) is a local embedding. More concisely, f : X → Y is an
immersion if each point in X has a neighborhood U such that f |U : U → f (U ) is a homeo-
morphism. Intuitively, an immersion allows self-intersection (tự cắt).

Figure 9.13: The Klein bottle immersed in R3 .

Example. Identifying opposite points on the boundary of a disk (they are called antipodal
points) we get a topological space called the projective plane (mặt phẳng xạ ảnh) RP2 . The
real projective plane cannot be embedded in R3 . It can be embedded in R4 . More generally,
identifying antipodal boundary points of D n gives us the projective space (không gian xạ
ảnh) RPn .

Example. Identifying opposite faces of the cube [0, 1]3 by ( x, y, 0) ∼ ( x, y, 1), ( x, 0, z) ∼

( x, 1, z), (0, y, z) ∼ (1, y, z) we get a space called the three-dimensional torus. It can also
be described, similarly to the two-dimensional torus, as the quotient space of R3 by the
relation ∀m, n, p ∈ Z, ( x, y, z) ∼ ( x + m, y + n, z + p), see figure 9.14.

9.15 Example. Let ( Xi )i∈ I be a collection of spaces together with a collection of points xi ∈
Xi . The disjoint union of this collection of spaces (see 7.14) quotiented by the relation such
that all the points xi are identified, that is, Xi /( xi )i∈ I , is denoted by ∨i∈ I ( Xi , xi ), and
F
i∈ I
is called the wedge sum (tổng nêm, chèn) of ( Xi )i∈ I with respect to the points ( xi )i∈ I .
9. QUOTIENT SPACE 67

Figure 9.14: The three-dimensional torus. A person living in this space may be able to see
infinitely many copies of her/himself.

Figure 9.16: S1 ∨ S1 ∨ S1 ∨ S1 .

For example a wedge sum of circles S1 ∨ S1 ∨ · · · ∨ S1 (which does not depend on how
the identified points are chosen) is called a bouquet of circles (chùm, đóa đường tròn). See
9.25.

Cut-and-paste Topology

Example. We see that the boundary of a Mobius band is a circle. Gluing two Mobius
bands along their boundaries gives the Klein bottle 12 . See figure 9.17. The process in this
example, where parts of spaces are cut and glued, is appropriately called “cut and paste”.

12 There is a humorous poem (limerick):

A mathematician named Klein

Thought the Mobius band was divine
Said he, “If you glue
The edges of two,
You’ll get a weird bottle like mine.”
68

d d

a a a a
c

b c
c b
b b
d
d
d
a e
a a
e

b b b
d b

d
a
e

f f

Figure 9.17: Gluing two Mobius bands along their boundaries gives the Klein bottle.

Example. Gluing a disk to the Mobius band gives the projective plane. In other words,
deleting a disk from the projective plane gives the Mobius band. See figure 9.18
9. QUOTIENT SPACE 69

a c

a a
c b
a

a c
b
c a a c
b
c a

b b

c c
c a c
a
b
b b b

b b

Figure 9.18: Gluing a disk to the Mobius band gives the projective plane.

Problems
1
9.19. Describe the space [0, 1]/0 ∼ 2 ∼ 1.

9.20. Describe the space that is the quotient of the sphere S2 by its equator S1 .

9.21. X Show that the torus T 2 is homeomorphic to S1 × S1 .

9.22. Show that the torus is homogeneous.

9.23. Show that gluing the boundary circle of a disk together gives a sphere, that is D2 /∂D2 is home-
omorphic to S2 .

Figure 9.24: A map from D2 onto S2 which becomes injective on D2 /∂D2 .

70

9.25. Show that S1 ∨ S1 does not depend on how the identified points are chosen, and is homeo-
morphic to the figure-8, namely, this subspace of the Euclidean plane {( x, y) ∈ R2 | ( x − 1)2 + y2 =
1} ∪ {( x, y) ∈ R2 | ( x + 1)2 + y2 = 1}.

Figure 9.26: S1 ∨ S1 is homeomorphic to the figure-8.

9.27. X Show that the following space is homeomorphic to the Mobius band.

9.28. X Show that the following two spaces are homeomorphic (one of the two is the Klein bottle).

b a

a a a b

b b

9.29. X What do we obtain after we cut a Mobius band along its middle circle? Try it with an
experiment. Mathematically, to cut a subset S from a space X means to delete S from X, the resulting
space is the subspace X \ S. In figure 9.29 the curve CC 0 is deleted.
9. QUOTIENT SPACE 71

B A0

C C0

A B0

C
A

Figure 9.30: Cutting a Mobius band along the middle circle.

9.31. To obtain a physical Mobius band we can take a long rectangular piece of paper, twist one side
once (an angle of 180 degree), then glue to the opposite side. What happen if we twist twice? What
happen if we twist many times? Do a physical experiment and a computer experiment. See figure
9.32.

Figure 9.32: A rectangle twisted twice then glued.

9.33. X Following the description of the torus a the quotient space of a rectangle, a line with slope 2/3
in the rectangle after the quotient will be a closed curve on the torus that goes around the torus 2 times
in one direction and 3 times in another direction, see figure 9.34. From this it is not difficult to obtain
a parametrization of an embedding of that space in R3 as for example ((2 + cos(t/2)) cos(t/3), (2 +
cos(t/2)) sin(t/3), sin(t/2)), 0 ≤ t ≤ 12π, see figure 9.33, and compare with figure 9.6. From these
pictures we can see why this space is often called the trefoil knot.
72

Figure 9.34: The trefoil knot on the torus.

Figure 9.35: The trefoil knot in R3 .

Show that the trefoil knot is homeomorphic to the circle S1 .

9.36. In general, the image of a simple closed curve in a Hausdorff topological space X is called a
knot (nút) in X. Thus a knot is the subspace γ([0, 1]) where γ : [0, 1] → X is a continuous map such
that γ(0) = γ(1) and γ|[0,1) is injective. Show that any knot is homeomorphic to the circle.

Figure 9.37: A knot is the image of a simple closed curve.

9.38. Show that the projective space RP1 is homeomorphic to S1 .

9.39. The one-point compactification of the open Mobius band (the Mobius band without the bound-
ary circle) is the projective space RP2 .

9.40. * Show that identifying antipodal boundary points of D n is equivalent to identifying antipodal
points of Sn . In other words, the projective space RPn is homeomorphic to Sn /x ∼ − x.

9.41. Show that if X has one of the properties connected, path-connected, compact, then so is the
quotient space X/ ∼.
9. QUOTIENT SPACE 73

9.42. Show that in order for the quotient space X/ ∼ to be a Hausdorff space, a necessary condition
is that each equivalence class [ x ] must be a closed subset of X. Is this condition sufficient?

9.43. On the Euclidean R define x ∼ y if x − y ∈ Z. Show that R/ ∼ is homeomorphic to S1 . The

space R/ ∼ is also described as “R quotiented by the action of the group Z”.

9.44. On the Euclidean R2 , define ( x1 , y1 ) ∼ ( x2 , y2 ) if ( x1 − x2 , y1 − y2 ) ∈ Z × Z. Show that R2 / ∼

is homeomorphic to the torus.

9.45. Let ∼ be an equivalence relation on X. Suppose a map g : X → Y is continuous and if x1 ∼ x2

then g( x1 ) = g( x2 ). Then there exists a unique continuous map f : X/ ∼→ Y such that g = f ◦ p.

p
X / X/ ∼

f
" 
g= f ◦ p
Y

9.46. Given a set X and a set Y ⊂ X × X, show that there is an equivalence relation on X that contains
Y and is contained in every equivalence relation that contains Y, called the minimal equivalence
relation containing Y.
For example, when we write [0, 1]/0 ∼ 1 we mean the quotient of the set [0, 1] by the minimal
equivalence relation on [0, 1] such that 0 ∼ 1. In this case that minimal equivalence relation is clearly
{(0, 1), (1, 0), ( x, x ) | x ∈ [0, 1]}.

9.47. * A question can be raised: In quotient spaces, if identifications are carried out in steps rather
than simultaneously, will the results be different? More precisely, let R1 and R2 be two equivalence
relations on a space X and let R be the minimal equivalence relation containing R1 ∪ R2 is also an
equivalence relation on space X. On the space X/R1 we define an equivalence relation R̃2 induced
from R2 by: [ x ] R1 ∼ R̃2 [y] R1 if ( x ∼ R1 ∪ R2 y). Prove that the map

X/( R1 ∪ R2 ) → ( X/R1 )/ R̃2

[ x ] R1 ∪ R2 7→ [[ x ] R1 ] R̃2

is a homeomorphism. Thus in this sense the results are same.

9.48. Show that if H is a normal subgroup of a topological group G (see 6.34) then the quotient group
G/H under the quotient topology is a topological group.
74

Other topics
Below there are several topics for further study and a guide for further reading.

Strategy for a proof of Tikhonov theorem based on net

The proof that we will outline here is based on further developments of the theory of nets
and a characterization of compactness in terms of nets.

Definition (subnet). Let I and I 0 be directed sets, and let h : I 0 → I be a map such that

∀k ∈ I, ∃k0 ∈ I 0 , (i0 ≥ k0 ⇒ h(i0 ) ≥ k).

If n : I → X is a net then n ◦ h is called a subnet of n.

The notion of subnet is an extension of the notion of subsequence. If we take ni ∈ Z+

such that ni < ni+1 then ( xni ) is a subsequence of ( xn ). In this case the map h : Z+ → Z+
given by h(i ) = ni is a strictly increasing function. Thus a subsequence of a sequence is a
subnet of that sequence. On the other hand a subnet of a sequence does not need to be a
subsequence, since for a subnet the map h is only required to satisfy limi→∞ h(i ) = ∞.
A net ( xi )i∈ I is called eventually in A ⊂ X if there is j ∈ I such that i ≥ j ⇒ xi ∈ A.
Universal net A net n in X is universal if for any subset A of X either n is eventually in
A or n is eventually in X \ A.

Proposition. If f : X → Y is continuous and n is a universal net in X then f (n) is a universal

net.

Proposition. The following statements are equivalent:

(a) X is compact.

(b) Every universal net in X is convergent.

(c) Every net in X has a convergent subnet.

The proof of the two propositions above could be found in [Bre93].

Then we finish the proof of Tikhonov theorem as follows.

Proof of Tikhonov theorem. Let X = ∏i∈ I Xi where each Xi is compact. Suppose that ( x j ) j∈ J
is a universal net in X. By 7.3 the net ( x j ) is convergent if and only if the projection ( pi ( x j ))
is convergent for all i. But that is true since ( pi ( x j )) is a universal net in the compact set
Xi .

Metrizability
The following is a typical result:

9.49 Theorem (Urysohn metrizability theorem). A regular space with a countable basis is
metrizable.

The proof uses the Urysohn lemma. For more see [Mun00, p. 243].
9. QUOTIENT SPACE 75

The weak topology

On a topological vector space the weak topology is the topology generated by continuous
linear functionals. This topology has an important role in Functional Analysis and its ap-
plications. See [H. Brezis, Functional analysis, Sobolev spaces and partial differential equations,
Springer, 2011, chapter 3], and [Con90, ch. 5].

Space filling curves

A rather curious and surprising result is:

Theorem. There is a continuous curve filling a rectangle on the plane. More concisely, there is a
continuous map from the interval [0, 1] onto the square [0, 1]2 under the Euclidean topology.

Note that this map cannot be injective, in other words the curve cannot be simple.
Such a curve is called a Peano curve. It could be constructed as a limit of an iteration of
piecewise linear curves [Mun00, tr. 271].

Guide for further reading

The book by Kelley [Kel55] has been a classics and a standard reference although it was
published in 1955. Its presentation is rather abstract: the book contains no figure!
Munkres’ book [Mun00] is presently a standard textbook. The treatment there is some-
what more modern than that in Kelley’s book, with many examples, figures and exercises.
Hocking and Young’s book [HY61] contains many deep and difficult results. This book
together with Kelley’s and Munkres’ books contain many topics not discussed in our lec-
tures.
A more recent textbook [Ros99] has more concrete presentations. The textbook [AF08]
contains many interesting recent applications of topology.
76
Part II Algebraic Topology

77
10. STRUCTURES ON TOPOLOGICAL SPACES 79

Briefly, Algebraic Topology associates algebraic objects, such as numbers, groups, vector
spaces, modules, . . . to topological objects, then studies these algebraic objects in order to
understand more about the topological objects.

10 Structures on topological spaces

Topological manifold
If we only stay around our small familiar neighborhood then we might not be able to rec-
ognize that the surface of the Earth is curved, and to us it is indistinguishable from a plane.
When we begin to travel farther and higher, we can realize that the surface of the Earth is
a sphere, not a plane. In mathematical language, a sphere and a plane are locally same but
globally different.
Briefly, a manifold is a space that is locally Euclidean. In a manifold each element can
be described by a list of independent parameters. The description can vary from one part
to another part of the manifold. 13

Definition. A topological manifold (đa tạp tôpô) of dimension n is a topological space

each point of which has a neighborhood homeomorphic to the Euclidean space Rn .

We can think of a manifold as a space that could be covered by a collection of open

subsets each of which homeomorphic to Rn .

Remark. In this chapter we assume Rn has the Euclidean topology unless we mention
otherwise.

The statement below is often convenient in practice:

Proposition. A manifold of dimension n is a space such that each point has an open neighborhood
homeomorphic to an open subset of Rn .

Proof. Let M be a topological space, x ∈ M, and suppose that x has an open neighborhood
U in M with a homeomorphism h : U → V where V is an open subset of Rn . We will
show that x has a neighborhood homeomorphic to Rn . There is a ball B( f ( x ), e) ⊂ V.
Since h is a homeomorphism, h−1 ( B( f ( x ), e)) is open in U, so is open in M. So x has
an open neighborhood h−1 ( B( f ( x ), e)) homeomorphic to a ball B( f ( x ), e), which in turns
homeomorphic to Rn (see 3.1).
13 Bernhard Riemann proposed the idea of manifold in 1854.
80

Remark. By Invariance of dimension (18.3), Rn and Rm are not homeomorphic unless m =

n, therefore a manifold has a unique dimension.

Example. Any open subspace of Rn is a manifold of dimension n.

Example. Let f : D → R be a continuous function where D ⊂ Rn is an open set, then the

graph of f , the set {( x, f ( x )) | x ∈ D }, as a subspace of Rn+1 , is homeomorphic to D, see
7.12, therefore is an n-dimensional manifold.
Thus manifolds generalizes curves and surfaces.

Example. The sphere Sn is an n-dimensional manifold. One way to show this is by cover-
ing Sn with two neighborhoods Sn \ {(0, . . . , 0, 1)} and Sn \ {(0, . . . , 0, −1)}. Each of these
neighborhoods is homeomorphic to Rn via stereographic projections. Another way is by
covering Sn by hemispheres {( x1 , x2 , . . . , xn+1 ) ∈ Sn | xi > 0} and {( x1 , x2 , . . . , xn+1 ) ∈
Sn | xi < 0}, 1 ≤ i ≤ n + 1. Each of these hemispheres is a graph, homeomorphic to an
open n-dimensional unit ball.

Example. The torus is a two-dimensional manifold. Let us consider the torus as the quo-
tient space of the square [0, 1]2 by identifying opposite edges. Each point has a neighbor-
hood homeomorphic to an open disk, as can be seen easily in the following figure, though
explicit description would be time consuming. We can also view the torus as a surface

Figure 10.1: The sets with same colors are glued to form a neighborhood of a point on the
torus. Each such neighborhood is homeomorphic to an open ball.

in R3 , given by the equation (

p
x2 + y2 − a)2 + z2 = b2 , see figure 9.6. As such it can be
covered by the open subsets of R3 corresponding to z > 0, z < 0, x2 + y2 < a2 , x2 + y2 > a2 .

Remark. The interval [0, 1] is not a manifold, it is a “manifold with boundary”, see section
24.

Simplicial complex
For an integer n ≥ 0, an n-dimensional simplex (đơn hình) is a subspace of a Euclidean
space Rm , m ≥ n which is the set of all convex linear combinations of (n + 1) points in
Rm which do not belong to any n-dimensional hyperplane. As a set it is given by {t0 v0 +
t1 v1 + · · · + tn vn | t0 , t1 , . . . , tn ∈ [0, 1], t0 + t1 + · · · + tn = 1} where v0 , v1 , . . . , vn ∈ Rm
and v1 − v0 , v2 − v0 , . . . , vn − v0 are n linearly independent vectors (it can be checked in
problem 10.15 that this condition does not depend on the order of the points). The points
v0 , v1 , . . . , vn are called the vertices of the simplex.
10. STRUCTURES ON TOPOLOGICAL SPACES 81

Example. A 0-dimensional simplex is just a point. A 1-dimensional simplex is a straight

segment in Rm , m ≥ 1. A 2-dimensional simplex is a triangle in Rm , m ≥ 2. A 3-
dimensional simplex is a tetrahedron in Rm , m ≥ 3.

In particular, the standard n-dimensional simplex (đơn hình chuẩn) ∆n is the convex
linear combination of the (n + 1) vectors (1, 0, 0, . . . ), (0, 1, 0, 0, . . . ), . . . , (0, 0, . . . , 0, 1) in
Rn+1 . Thus

∆n = {(t0 , t1 , . . . , tn ) | t0 , t1 , . . . , tn ∈ [0, 1], t0 + t1 + · · · + tn = 1}.

The convex linear combination of any subset of the set of vertices of a simplex is called
a face of the simplex.

Example. For a 2-dimensional simplex (a triangle) its faces are the vertices, the edges, and
the triangle itself.

Definition. An n-dimensional simplicial complex (phức đơn hình) in Rm is a finite collec-

tion S of simplexes in Rm of dimensions at most n and at least one simplex is of dimension
n and such that:

(a) any face of an element of S is an element of S,

(b) the intersection of any two elements of S is a common face.

The union of all elements of S is called its underlying space, denoted by |S|, a subspace of
Rm . A space which is the underlying space of a simplicial complex is also called a polyhe-
dron (đa diện).

Example. A 1-dimensional simplicial complex is often called a (combinatorial) graph in

graph theory.

Triangulation
A triangulation (phép phân chia tam giác) of a topological space X is a homeomorphism
from the underlying space of a simplicial complex to X, the space X is then said to be
triangulated.
For example, a triangulation of a surface is an expression of the surface as the union of
finitely many triangles, with a requirement that two triangles are either disjoint, or have
one common edge, or have one common vertex.

Figure 10.2: This is an octahedron, giving a triangulation of the 2-dimensional sphere.

82

1
0.8
1 0.6
0.4
0.2
0.5 0
-0.2
0 -0.4
-0.6
-0.5 -0.8
-1
-1
3
2
1
-3 0
-2
-1 -1
0
1 -2
2
3 -3

Figure 10.3: A triangulation of the torus.

Figure 10.4: Description of a triangulation of the torus.

A simplicial complex is specified by a finite set of points, if a space can be triangulated

then we can study that space combinatorially, using constructions and computations in
finitely many steps.

Cell complex
For n ≥ 1 a cell (ô) is an open ball in the Euclidean space Rn . A 0-dimensional cell is a
point (this is consistent with the general case with the convention that R0 = {0}).
Recall a familiar term that an n-dimensional disk is a closed ball in Rn , in particular
when n = 0 it is a point. The unit disk centered at the origin B0 (0, 1) is denoted by D n .
Thus for n ≥ 1 the boundary ∂D n is the sphere Sn−1 and the interior int( D n ) is an n-cell.
By attaching a cell to a topological space X we mean taking a continuous function
f : ∂D n → X then forming the quotient space ( X t D n )/( x ∼ f ( x ), x ∈ ∂D n ) (for dis-
joint union see 7.14). Intuitively, we attach a cell to the space by gluing each point on the
boundary of the disk to a point on the space in a certain way.
We can attach finitely many cells toX in 
the same 
manner. Precisely, attaching k n-cells
Fk
to X means taking the quotient space X t i =1 Din /( x ∼ f i ( x ), x ∈ ∂Din , 1 ≤ i ≤ k )
where f i : ∂Din → X is continuous, 1 ≤ i ≤ k.

Definition. A (finite) n-dimensional cell complex (phức ô) or CW-complex X is a topolog-

ical space with a structure as follows:
10. STRUCTURES ON TOPOLOGICAL SPACES 83

(a) X 0 is a finite collection of 0-cells, with the discrete topology,

(b) for 1 ≤ i ≤ n ∈ Z+ , X i is obtained by attaching finitely many i-cells to X i−1 , and

X n = X.

Briefly, a cell complex is a topological space with an instruction for building it by attaching
cells. The subspaces X i are called the i-dimensional skeleton (khung) of X.14

Example. A topological circle has a cell complex structure as a triangle with three 0-cells
and three 1-cells. There is another cell complex structure with only one 0-cell and one 1-cell.

Example. The 2-dimensional sphere has a cell complex structure as a tetrahedron with four
0-cells, six 1-cells, and four 2-cells. There is another cell complex structure with only one
0-cell and one 2-cell, obtained by gluing the boundary of a 2-disk to a point.

Example. The torus, as we can see directly from its definition (figure 9.4), has a cell complex
structure with one 0-cells, two 1-cells, and one 2-cells.

A simplicial complex gives rise to a cell complex:

Proposition. Any triangulated space is a cell complex.

Proof. Let X be a simplicial complex. Let X i be the union of all simplexes of X of dimensions
at most i. Then X i+1 is the union of X i with finitely many (i + 1)-dimensional simplexes.
Let ∆i+1 be such an (i + 1)-dimensional simplex. The i-dimensional faces of ∆i+1 are sim-
plexes of X, so the union of those faces, which is the boundary of ∆i+1 , belongs to X i . There
is a homeomorphism from an (i + 1)-disk to ∆i+1 (see 10.16), bringing the boundary of the
disk to the boundary of ∆i+1 . Thus including ∆i+1 in X implies attaching an (i + 1)-cell to
Xi .

The example of the torus indicates that cell complexes may require less cells than sim-
plicial complexes. On the other hand we loose the combinatorial setting, because we need
to specify the attaching maps.
It is known that any compact manifold of dimension different from 4 has a cell complex
structure. Whether that is true or not in dimension 4 is not known yet [Hat01, p. 529].

Euler characteristic
Definition. The Euler characteristic (đặc trưng Euler) of a cell complex is defined to be the
alternating sum of the number of cells in each dimension of that cell complex. Namely, let
X be an n-dimensional cell complex and let ci , 0 ≤ i ≤ n, be the number of i-dimensional
cells of X, then the Euler characteristic of X is
n
χ( X ) = ∑ (−1)i ci .
i =0

Example. For a triangulated surface, its Euler characteristic is the number v of vertices
minus the number e of edges plus the number f of triangles (faces):

χ(S) = v − e + f .
14 The term CW-complex is more general than the term cell-complex, can be used when there are infinitely many

cells.
84

10.5 Theorem. The Euler characteristic of homeomorphic cell complexes are equal.

In particular two cell complex structures on a topological space have same Euler char-
acteristic. The Euler characteristic is a topological invariant (bất biến tôpô).
We have χ(S2 ) = 2. A consequence is the famous formula of Leonhard Euler:

Theorem (Euler’s formula). For any polyhedron homeomorphic to a 2-dimensional sphere, v −

e + f = 2.

Figure 10.6: This is the dodecahedron. Its Euler characteristic is 2.

Example. The Euler characteristic of a surface is defined and does not depend on the choice
of triangulation. If two surfaces have different Euler characteristic they are not homeomor-
phic.
From any triangulation of the torus, we get χ( T 2 ) = 0. For the projective plane, χ(RP2 ) =
1. As a consequence, the sphere, the torus, and the projective plane are not homeomorphic
to each other: they are different surfaces.

Problems
10.7. Show that if two spaces are homeomorphic and one space is an n-dimensional manifold then
the other is also an n-dimensional manifold.

10.8. Show that an open subspace of a manifold is a manifold.

10.9. Show that if X and Y are manifolds then X × Y is also a manifold.

10.10. Show that a connected manifold must be path-connected. Thus for manifold connectedness
and path-connectedness are same.

10.11. Show that RPn is an n-dimensional topological manifold.

10.12. Show that a manifold is a locally compact space.

10.13. Show that the Mobius band, without its boundary circle, is a manifold.

10.14. Give examples of topological spaces which are not manifolds. Discuss necessary conditions
for a topological space to be a manifold.

10.15. Given a set of n points in Rn , list the points as v0 , v1 , . . . , vn . Check that the condition that
v1 − v0 , v2 − v0 , . . . , vn − v0 are linearly independent vectors does not depend on the choice of orders
for the points.
10. STRUCTURES ON TOPOLOGICAL SPACES 85

10.16. X Show that any n-dimensional simplex is homeomorphic to an n-dimensional disk.

10.17. Give a triangulation for the cylinder, find a simpler cell complex structure for it, and compute
its Euler characteristic.

10.18. Give a triangulation for the Mobius band, find a simpler cell complex structure for it, and
compute its Euler characteristic.

10.19. Give a triangulation for the Klein bottle, find a simpler cell complex structure for it, and com-
pute its Euler characteristic.

10.20. Give examples of spaces which are not homeomorphic but have same Euler characteristics.

10.21. Draw a cell complex structure on the torus with two holes.

10.22. Find a cell complex structure on RPn .

10.23 (Platonic solids). In this problem, using a term in classical geometry, by a convex polyhedron
(đa diện lồi) in R3 we mean the union of a finite number of polygons (đa giác) in R3 such that any
two such polygons intersect in either an empty set, or a common vertex, or a common edge, that each
edge belong to exactly two polygons, and that it is the boundary of a convex compact subset of R3
with non-empty interior, called the corresponding solid. Notice that this is like the underlying space
of a 2-dimensional simplicial complex except that the 2-dimensional faces are allowed to be polygons
instead of only triangles. A regular convex polyhedron (đa diện lồi đều) is a convex polyhedron
whose faces are the same regular polygons (đa giác đều) and each vertex belongs to the same number
of faces.
Consider a regular convex polyhedron. Let p be the number of regular polygons at each vertex,
and let q be the number of vertices of the regular polygon.
(a) Counting the number of vertices v from the number of faces f , show that pv = q f .
(b) Counting the number of edges e from the number of faces, show that 2e = q f .
(c) It is known that a convex polyhedron is homeomorphic to a sphere (Problem 17.16). Using
4p
Euler characteristic, deduce that f = 2( p+q)− pq , and hence 2( p + q) − pq > 0.
2q
(d) Deduce that 3 ≤ p < q−2 .
(e) Deduce that there are only 5 possibilies for ( p, q): (3, 3), (4, 3), (5, 3), (3, 4), (3, 5).
It was known since ancient time that there exists those regular convex polyhedron: (3, 3) gives the
regular tetrahedron, (4, 3) gives the regular octahedron (Figure 10.2), (5, 3) gives the regular icosa-
hedron, (3, 4) gives the cube, (3, 5) gives the regular dodecahedron (Figure 10.6). The corresponding
solids are called the Platonic solids.
86

11 Classification of compact surfaces

In this section by a surface (mặt) we mean a two-dimensional topological manifold (without
boundary).

Connected sum
Let S and T be two surfaces. From each surface deletes an open disk, then glue the two
boundary circles. The resulting surface is called the connected sum (tổng liên thông) of the
two surfaces, denoted by S#T.

◦ ◦
S \ D2 T \ D2 S #T

It is known that the connected sum does not depend on the choices of the disks.

Example. If S is any surface then S#S2 = S.

Classification
Theorem (classification of compact surfaces). A connected compact surface is homeomorphic
to either the sphere, or a connected sum of tori, or a connected sum of projective planes.

We denote by Tg the connected sum of g tori, and by Mg the connected sum of g projec-
tive planes. The number g is called the genus (giống) of the surface.
The sphere and the surfaces Tg are orientable (định hướng được) surface , while the
surfaces Mg are non-orientable (không định hướng được) surfaces. We will not give a
precise definition of orientability here (compare 25).

Figure 11.1: Orientable surfaces: S2 , T1 , T2 , . . .

11. CLASSIFICATION OF COMPACT SURFACES 87

Recall that a torus T1 is the quotient space of a square by the identification on the bound-
ary described by the word aba−1 b−1 . In general, given a polygon on the plane (the under-
lying space that is homeomorphic to a disk of a simplicial complex), and suppose that the
edges of the polygon are labeled and oriented. Choose one edge as the initial one then fol-
low the edge of the polygon in a predetermined direction (the boundary of the polygon is
homeomorphic to a circle). If an edge a is met in the opposite direction then write it down
as a−1 . In this way we associate each polygon with a word. We also write aa as a2 .
In the reverse direction, a word gives a polygon with labeled and oriented edges. If a
label appears more than once, then the edges with this label can be identified in the ori-
entations assigned to the edges. Let us consider two words equivalent if they give rise to
homeomorphic spaces. For examples, changing labels and cyclic permutations are equiva-
lence operations on words.
The classification theorem is a direct consequence of the following result:

11.2 Theorem. A connected compact surface is homeomorphic to the space obtained by identifying
the edges of a polygon in one of the following ways:

(a) aa−1 ,

(b) a1 b1 a1−1 b1−1 a2 b2 a2−1 b2−1 · · · a g bg a− 1 −1

g bg ,

(c) c21 c22 · · · c2g .

Proof of 11.2
We will use the following fact: any compact surface can be triangulated (a proof is available
in [Moi77]).
Take a triangulation X of the compact connected surface S. Let T be the set of triangles
in X. Label the edges of the triangles and mark an orientation for each edge. Each edge will
appear twice, on two different triangles.
We now build a new space P out of these triangles. Let P1 be any element of T, which is
homeomorphic to a disk. Inductively suppose that we have built a space Pi from a subset Ti
of T, where Pi is homeomorphic to a disk and its boundary consists of labeled and directed
edges. If Ti 6= T then there exists an element ∆ ∈ T \ Ti such that ∆ has at least one
edge with the same label α as an edge on the boundary of Pi , otherwise | X | could not be
connected. Glue ∆ to Pi via this edge to get Pi+1 , in other words Pi+1 = ( Pi t ∆) /α ∼
α. Since two disks glued along a common arc on the boundaries is still a disk, Pi+1 is
homeomorphic to a disk. Also, let Ti+1 = Ti ∪ {∆}. When this process stops we get a
space P homeomorphic to a disk, whose boundary consists of labeled, directed edges. If
these edges are identified following the instruction by the labels and the directions, we get
a space homeomorphic to the original surface S. The polygon P is called a fundamental
polygon of the surface S. See an example in figure 11.3.
Theorem 11.2 is a direct consequence of the following:

11.4 Proposition. An associated word to a connected compact surface is equivalent to a word of the
forms:

(a) aa−1 ,

(b) a1 b1 a1−1 b1−1 a2 b2 a2−1 b2−1 · · · a g bg a− 1 −1

g bg ,
88

c
c
a c

b a f
b e
f

f
d
d e d

Figure 11.3: A triangulation of the sphere and an associated fundamental polygon.

(c) c21 c22 · · · c2g .

We will prove 11.4 through a sequence of lemmas. The technique is cut-and-paste topol-
ogy in section 9.

Let w a the word associated to a fundamental polygon of a compact connected surface.

If w1 and w2 are two equivalent words then we write w1 ∼ w2 .

11.5 Lemma. aa−1 α ∼ α, if α 6= ∅.

Proof. See figure 11.6.

a a
a

Figure 11.6: Lemma 11.5.

11.7 Lemma. The word w is equivalent to a word whose all of the vertices of the associated polygon
is identified to a single point on the associated surface (w is said to be “reduced”).

Proof. When we do the operation in figure 11.8, the number of P vertices is decreased.
When there is only one P vertex left, we arrive at the situation in lemma 11.5.
11. CLASSIFICATION OF COMPACT SURFACES 89

P P
b a a
b
c Q c
Q
Q
c

b P P
b

c Q

b P
a
c Q

Figure 11.8: Lemma 11.7.

11.9 Lemma. aαaβ ∼ aaαβ−1 .

Proof. See figure 11.10.

β β

c a
a c
a c a

α α

c c
a

β α

Figure 11.10: Lemma 11.9.

11.11 Lemma. Suppose that w is reduced. If w = aαa−1 β where α 6= ∅ then ∃b ∈ α such that
b ∈ β or b−1 ∈ β.

Proof. If all labels in α appear in pairs then the vertices in the part of the polygon associated
to α are identified only with themselves, and are not identified with a vertex outside of that
part. This contradicts the assumption that w is reduced.
90

11.12 Lemma. aαbβa−1 γb−1 δ ∼ aba−1 b−1 αδβγ.

Proof. See figure 11.13.

a a

b b
c b c c b

a a
a
a

c c
b c
d
c
a
a

a
d

c c
d a

c c
d
a
d

Figure 11.13: Lemma 11.12.

11.14 Lemma. aba−1 b−1 αc2 β ∼ a2 b2 αc2 β.

Proof. Do the operation in figure 11.15, after that we are in a situation where we can apply
lemma 11.9 three times.
11. CLASSIFICATION OF COMPACT SURFACES 91

b b
a a
a

b
d
d

c c

a b

Figure 11.15: Lemma 11.14.

Proof of 11.4. The proof follows the following steps.

1. Bring w to reduced form by using 11.7 finitely many times.

2. If w has the form aa−1 − then go to 2.1, if not go to 3.

2.1. If w has the form aa−1 then stop, if not go to 2.2.

2.2. w has the form aa−1 α where α 6= ∅. Repeatedly apply 11.5 finitely many times,
deleting pairs of the form aa−1 in w until no such pair is left or w has the form
aa−1 . If no such pair is left go to 3.

3. w does not have the form − aa−1 −. If we apply 11.9 then a pair of of the form
− aαa− with α 6= ∅ could become a pair of the form − a − a−1 −, but a pair of
the form − aa− will not be changed. Therefore 11.9 could be used finitely many
times until there is no pair − aαa− with α = ∅ left. Notice from the proof of
11.9 that this step will not undo the steps before it.

4. If there is no pair of the form − aαa−1 where α 6= ∅, then stop: w has the form
a21 a22 · · · a2g .

5. w has the form − aαa−1 where α 6= ∅. By 11.11 w must has the form − a − b −
a−1 − b−1 −, since after Step 3 there could be no −b − a−1 − b−.

6. Apply 11.12 finitely many times until w no longer has the form − aαbβa−1 γb−1 −
where at least one of α, β, or γ is non-empty. If w is not of the form − aa− then
stop: w has the form a1 b1 a1−1 b1−1 · · · a g bg a− 1 −1
g bg .

7. w has the form − aba−1 b−1 − cc−. Use 11.14 finitely many times to transform w
to the form a21 a22 · · · a2g .
92

Example. What is the topological space in figure 11.16?

a c

d d

c a

Figure 11.16:

Applying lemma 11.12 we get abcda−1 b−1 c−1 d−1 ∼ aba−1 b−1 cdc−1 d−1 . Thus this space
is homeomorphic to the genus 2 torus.
For more on this topic one can read [Mas91, ch. 1].

Problems
11.17. Show that RP2 #RP2 = K, in other words, gluing two Mobius bands along their boundaries
gives the Klein bottle. See figure 9.18.

11.18. Show that T 2 #RP2 = K#RP2 , where K is the Klein bottle.

11.19. What is the topological space in figure 11.20?

b
a c

d d

a b

Figure 11.20:

11.21. (a) Show that Tg #Th = Tg+h .

(b) Show that Mg #Mh = Mg+h .

(c) Mg #Th =?

11.22. Show that for any two compact surfaces S1 and S2 we have χ(S1 #S2 ) = χ(S1 ) + χ(S2 ) − 2.

11.23. Compute the Euler characteristics of all connected compact surfaces. Deduce that the ori-
entable surfaces S2 and Tg , for different g, are distinct, meaning not homeomorphic to each other.
Similarly the non-orientable surfaces Mg are all distinct.

11.24. From 11.2, describe a cell complex structure on any compact surface. From that compute the
Euler characteristics.
11. CLASSIFICATION OF COMPACT SURFACES 93

11.25. Let S be a triangulated compact connected surface with v vertices, e edges, and f triangles.
Prove that:

(a) 2e = 3 f . The number of triangles is even.

(b) v ≤ f .
(c) e ≤ 12 v(v − 1).
j  k
(d) v ≥ 21 7 + 49 − 24χ(S)
p
= H (S). (The integer H (S), called the Heawood number, gives
the minimal number of colors needed to color a map on the surface S, except in the case of the
Klein bottle. When S is the sphere this is The four colors problem [MT01, p. 230].)
(e) Show that a triangulation of the torus needs at least 14 triangles. Indeed 14 is the minimal
number: there are triangulations of the torus with exactly 14 triangles, see e.g. [MT01, p. 142].
94

12 Homotopy

Homotopy of maps
Definition. Let X and Y be topological spaces and f , g : X → Y be continuous. We say
that f and g are homotopic (đồng luân) if there is a continuous map

F : X × [0, 1] → Y
( x, t) 7→ F ( x, t)

such that F ( x, 0) = f ( x ) and F ( x, 1) = g( x ) for all x ∈ X. The map F is called a homotopy

(phép đồng luân) from f to g.

We can think of t as a time parameter and F as a continuous process in time that starts
with f and ends with g. To suggest this view F ( x, t) is often written as Ft ( x ). So F0 = f and
F1 = g.
Here we are looking at [0, 1] ⊂ R with the Euclidean topology and X × [0, 1] with the
product topology.

12.1 Proposition. Homotopic relation on the set of continuous maps between two given topological
spaces is an equivalence relation.

Proof. If f : X → Y is continuous then f is homotopic to itself via the map Ft = f , ∀t ∈ [0, 1].
If f is homotopic to g via F then g is homotopic to f via G ( x, t) = F ( x, 1 − t). We can think
of G is an inverse process to H. If f is homotopic to g via F and g is homotopic to h via G
then f is homotopic to h via

1
F ,
2t 0≤t≤ 2
Ht =
1
G
2t−1 , 2 ≤ t ≤ 1.

Thus we obtain H by following F at twice the speed, then continuing with a copy of G also
at twice the speed. To check the continuity of H it is better to write it as

 F ( x, 2t), ( x, t) ∈ X × [0, 21 ],
H ( x, t) =
 G ( x, 2t − 1), ( x, t) ∈ X × [ 12 , 1],

then use 3.15. Another, more advanced way to check continuity of H is by using the prop-
erty 8.10 about the compact-open topology.

In this section, and whenever it is clear from the context, we indicate this relation by the
usual notation for equivalence relation ∼.

12.2 Lemma. If f , g : X → Y and f ∼ g then f ◦ h ∼ g ◦ h for any h : X 0 → X, and k ◦ f ∼ k ◦ g

for any k : Y → Y 0 .

Proof. Let F be a homotopy from f to g then Gt = Ft ◦ h is a homotopy from f ◦ h to g ◦ h.

Rewriting G ( x 0 , t) = F (h( x 0 ), t) we see that G is continuous.

Homotopic spaces
Definition. Two topological spaces X and Y are homotopic if there are continuous maps
f : X → Y and g : Y → X such that g ◦ f is homotopic to IdX and f ◦ g is homotopic to IdY .
12. HOMOTOPY 95

Each of the maps f and g is called a homotopy equivalence.

Immediately we have:

Proposition (homeomorphic ⇒ homotopic). Homeomorphic spaces are homotopic.

Proof. If f is a homeomorphism then we can take g = f −1 .

Proposition. Homotopic relation on the set of all topological spaces is an equivalence relation.

Proof. Any space is homotopic to itself. From the definition it is clear that this relation
is symmetric. Now we check that it is transitive. Suppose that X is homotopic to Y via
f : X → Y, g : Y → X and suppose that Y is homotopic to Z via h : Y → Z, k : Z → Y. By
12.2:
( g ◦ k) ◦ (h ◦ f ) = ( g ◦ (k ◦ h)) ◦ f ∼ ( g ◦ IdY ) ◦ f = g ◦ f ∼ IdX .

Similarly
(h ◦ f ) ◦ ( g ◦ k) = (h ◦ ( f ◦ g)) ◦ k ∼ (h ◦ IdY ) ◦ k = h ◦ k ∼ IdZ .

Thus X is homotopic to Z.

Deformation retraction
Let X be a space, and let A be a subspace of X. We say that A is a retract (rút) of X if there
is a continuous map r : X → A such that r | A = id A , called a retraction (phép rút) from X
to A. In other words A is a retract of X if the identity map id A can be extended to X.
A deformation retraction (phép co rút) from X to A is a homotopy F that starts with
idX , ends with a retraction from X to A, and fixes A throughout, i.e., F0 = idX , F1 ( X ) = A,
and Ft | A = id A , ∀t ∈ [0, 1]. If there is such a deformation retraction we say that A is a
deformation retract (co rút) of X.
In such a deformation retraction each point x ∈ X \ A “moves” along the path Ft ( x ) to
a point in A, while every point of A is fixed.

Example. A subset A of a normed space is called a star-shaped region if there is a point

x0 ∈ A such that for any x ∈ A the straight segment from x to x0 is contained in A.
For example, any convex subset of the normed space is a star-shaped region. The map
Ft = (1 − t) x + tx0 is a deformation retraction from A to x0 , so a star-shaped region has a
deformation retraction to a point.

Example. A normed space minus a point has a deformation retraction to a sphere. For
example, a normed space minus the origin has a deformation retraction Ft ( x ) = (1 − t) x +
t || xx|| to the unit sphere at the origin.

Example. An annulus S1 × [0, 1] has a deformation retraction to one of its circle boundary
S1 × {0}.

Proposition. If a space X has a deformation retraction to a subspace A then X is homotopic to A.

Proof. Suppose that Ft is a deformation retraction from X to A. Consider F1 : X → A and

the inclusion map g : A → X, g( x ) = x. Then idX is homotopic to g ◦ F1 via Ft , while
F1 ◦ g = id A .

Example. The circle, the annulus, and the Mobius band are homotopic to each other, al-
though they are not homeomorphic to each other.
96

A space which is homotopic to a space containing only one point is said to be con-
tractible (thắt được). It can be check immediately that this means there is a homotopy from
the identity map to a constant map. Thus a space X is contractible if and only if there exists
x0 ∈ X and a continuous map F : X × [0, 1] → X such that F0 = idX and F1 = x0 .

Corollary. A space which has a deformation retraction to a point is contractible.

The converse is not true, since the point x0 is not required to be fixed during the homo-
topy of a contractible space, see [Hat01, p. 18].

Example. Any star-shaped region is contractible.

Problems
12.3. Show in detail that the Mobius band has a deformation retraction to a circle.

12.4. Show that contractible spaces are path-connected.

12.5. Let X be a topological space and let Y be a retract of X. Show that any continuous map from Y
to a topological space Z can be extended to X.

12.6. Let X be a topological space and let Y be a retract of X. Show that if every continuous map from
X to itself has a fixed point then every continuous map from Y to itself has a fixed point.

12.7. Show that a retract of a Hausdorff space must be a closed subspace. Deduce that in a normed
space, there exists a retraction from the space the the closed unit ball, but there is no retraction from
the space to the open unit ball.

12.8. Show that if B is contractible then A × B is homotopic to A.

12.9. Show that if a space X is contractible then all continuous maps from a space Y to X are homo-
topic.

12.10. Let X be a topological space and let f : X → Sn , n ≥ 1, be a continuous map which is not
surjective. Show that f is homotopic to a constant map.

12.11. Show that if f : Sn → Sn is not homotopic to the identity map then there is x ∈ Sn such that
f ( x ) = − x.

12.12 (isotopy). An isotopy is a homotopy by homeomorphisms. Namely two homeomorphisms f

and g from X to Y are isotopic if there is a continuous map

F : X × [0, 1] → Y
( x, t) 7→ F ( x, t)

such that F ( x, 0) = f ( x ) and F ( x, 1) = g( x ) for all x ∈ X, and for each t ∈ [0, 1] the map F (·, t) is a
homeomorphism. Check that being isotopic is an equivalence relation.

12.13 (Alexander trick). * Any homeomorphism of the disk fixing the boundary is isotopic to the
identity.
This result can be proved in the following way. Suppose that f is homeomorphism from D n onto
D n which is the identity on ∂D n . Let F : D n × [0, 1] → D n ,

 x, k x k ≥ t,
F ( x, t) =
x


t f
t , k x k < t.

At each t > 0 the map F (·, t) replicates f on a smaller ball B0 (0, t), so as t → 0 gradually F (·, t)
becomes the identity map. Check that F is indeed an isotopy from f to the identity.
12. HOMOTOPY 97

12.14. Classify the alphabetical characters according to homotopy types, that is, which of the charac-
ters are homotopic to each other as subspaces of the Euclidean plane? Do the same for the Vietnamese
alphabetical characters. Note that the result depends on the font you use.
98

13 The fundamental group

A path (đường đi) in a space X is a continuous map α from the Euclidean interval [0, 1] to
X. The point α(0) is called the initial end point, and α(1) is called the final end point. In
this section for simplicity of presentation we assume the domain of a path is the Euclidean
interval [0, 1] instead of any Euclidean closed interval as before.
A loop (vòng) or a closed path (đường đi kín) based at a point a ∈ X is a path whose
initial point and end point are a. In other words it is a continuous map α : [0, 1] → X such
that α(0) = α(1) = a. The constant loop at a is the loop α(t) = a for all t ∈ [0, 1].
If α(t), 0 ≤ t ≤ 1 is a path from a to b then we write as α−1 the path given by α−1 (t) =
α(1 − t), called the inverse path of α, going from b to a.
If α is a path from a to b, and β is a path from b to c, then we define the composition
(hợp) of α with β, written α · β, to be the path

1
α(2t), 0≤t≤ 2
α · β(t) =
 β(2t − 1), 1
2 ≤ t ≤ 1.

Notice that α · β is continuous by 3.15.

Homotopy of paths
Definition. Let α and β be two paths from a to b in X. A path-homotopy (phép đồng luân
đường) from α to β is a continuous map F : [0, 1] × [0, 1] → X, F (s, t) = Ft (s), such that
F0 = α, F1 = β, and for each t the path Ft goes from a to b, i.e. Ft (0) = a, Ft (1) = b.

If there is a path-homotopy from α to β we say that α is path-homotopic (đồng luân

đường) to β.

Remark. A homotopy of path is a homotopy of maps defined on [0, 1], with the further re-
quirement that the homotopy fixes the initial point and the terminal point. To emphasize
this we have used the word path-homotopy, but some sources (e.g. [Hat01, p. 25]) simply
use the term homotopy, taking this further requirement implicitly.

F0 = α

Ft

a b

F1 = β

Figure 13.1: We can think of a path-homotopy from α to β as a way to continuously brings

α to β, keeping the endpoints fixed, similar to a motion picture.

13.2 Example. In a convex subset of a normed space any two paths α and β with the same
initial points and end points are homotopic, via the homotopy (1 − t)α + tβ.

Proposition. Path-homotopic relation on the set of all paths from a to b is an equivalence relation.

The proof is the same as the proof of 12.1.

13. THE FUNDAMENTAL GROUP 99

13.3 Lemma. If α is path-homotopic to α1 and β is path-homotopic to β 1 then α · β is path-homotopic

to α1 · β 1 .

Proof. Let F be a path-homotopy from αto β and G be a homotopy from β to γ. Let Ht =

Ft · Gt , that is: 
 F (2s, t), 0 ≤ s ≤ 12 , 0 ≤ t ≤ 1
H (s, t) =
 G (2s − 1, t), 1
2 ≤ s ≤ 1, 0 ≤ t ≤ 1.

Then H is continuous by 3.15 and is a homotopy from α · β to α1 · β 1 .

13.4 Lemma. If α is a path from a to b then α · α−1 is path-homotopic to the constant loop at a.

Proof. A homotopy F from α · α−1 to the constant loop at a can be described as follows. At
a fixed t, the loop Ft starts at time 0 at a, goes along α but at twice the speed of α, until time
1
2 − 2t , stays there until time 12 + 2t , then catches the inverse path α−1 at twice its speed to
come back to a. More precisely,

1 t


 α(2s), 0≤s≤ 2 − 2

F (s, t) = α(1 − t), 1 t 1 t
2 − 2 ≤s≤ 2 + 2


α−1 (2s − 1) = α(2 − 2s),
 1
+ t
≤ s ≤ 1.
2 2

Again by 3.15, F is continuous.

13.5 Lemma (reparametrization does not change homotopy class). With a continuous map
ϕ : [0, 1] → [0, 1], ϕ(0) = 0, ϕ(1) = 1, for any path α the path α ◦ ϕ is path-homotopic to α.

Proof. Let Ft = (1 − t) ϕ + tId[0,1] , a homotopy from ϕ to Id[0,1] on [0, 1]. We can check that
Gt = α ◦ Ft , i.e., G = α ◦ F is a path-homotopy from α ◦ ϕ to α.

13.6 Corollary. (α · β) · γ is path-homotopic to α · ( β · γ).

Proof. We can check directly that (α · β) · γ is a reparametrization of α · ( β · γ). Indeed,


α · β(2t), 1
0≤t≤ 2
(α · β) · γ(t) =
γ(2t − 1), 1
2 ≤t≤1

1


 α(4t), 0≤t≤ 4

= β(4t − 1), 1 1
4 ≤t≤ 2


γ(2t − 1), 1
≤ t ≤ 1,

2

and

1
α(2t), 0≤t≤ 2
α · ( β · γ)(t) =
 β · γ(2t − 1) 1 ≤ t ≤ 1
2



 α(2t), 0 ≤ t ≤ 12

= β(4t − 2), 12 ≤ t ≤ 34


γ(4t − 3), 3 ≤ t ≤ 1.

4
100

We can see immediately that if (α · β) · γ = α · ( β · γ) ◦ ϕ then ϕ has to be


1


 2t, 0≤t≤ 4

ϕ(t) = t + 41 , 1
4 ≤t≤ 1
2


1t + 1,
 1
≤ t ≤ 1.
2 2 2

The fundamental group

Recall that a group is a set G with a map

G×G → G
( a, b) 7→ a · b,

called a multiplication, such that:

(a) multiplication is associative: ( ab)c = a(bc), ∀ a, b, c ∈ G,

(b) there exists a unique element of G called the identity element esatisfying ae = ea = a,
∀ a ∈ G,

(c) for each element a of G there exists an element of G called the inverse element of a,
denoted by a−1 , satisfying a · a−1 = a−1 · a = e.

See e.g. [Gal10] for more on groups.

Consider the set of loops of X based at a point x0 under the path-homotopy relation.
On this set we define a multiplication operation [α] · [ β] = [α · β]. By 13.3 this operation is
well-defined.

13.7 Theorem. The set of all path-homotopy classes of loops of X based at a point x0 is a group
under the above operation.

This group is called the fundamental group (nhóm cơ bản) of X at x0 , denoted by

π1 ( X, x0 ). The point x0 is called the base point.

Proof. Denote by 1 the constant loop at x0 . Since 1 · α is a reparametrization of α via the

map 
1
0, 0≤t≤ 2
ϕ(t) =
2t − 1, 1
2 ≤t≤1

by lemma 13.5 1 · α is path-homotopic to α, thus [1] · [α] = [α]. So [1] is the identity in
π1 ( X, x0 ).
Define [α]−1 = [α−1 ]. It is easy to check that [α]−1 is well-defined, and is the inverse
element of [α], by using 13.4.
By 13.6 we have associativity:

([α] · [ β]) · [γ] = [α · β] · [γ] = [(α · β) · γ] = [α · ( β · γ)] = [α] · ([ β] · [γ]).

Example. The fundamental group of a space containing only one point is trivial. Namely,
if X = { x0 } then π1 ( X, x0 ) = 1, the group with only one element.
13. THE FUNDAMENTAL GROUP 101

Recall that a map h : G1 → G2 between two groups is called a group homomorphism if it

preserves group operations, i.e. ∀ a, b ∈ G1 , h( a · b) = h( a) · h(b). It is a group isomorphism
if is a bijective homomorphism (in this case the inverse map is also a homomorphism).

13.8 Proposition (dependence on base point). If there is a path from x0 to x1 then π1 ( X, x0 )

is isomorphic to π1 ( X, x1 ).

Proof. Let α be a path from x0 to x1 . Consider the map

hα : π1 ( X, x1 ) → π1 ( X, x0 )
[ γ ] 7 → [ α · γ · α −1 ] .

Using 13.3 and 13.6 we can check that this is a well-defined map, and is a group homomor-
phism with an inverse homomorphism:

h− 1
α : π1 ( X, x1 ) → π1 ( X, x0 )
[ γ ] 7 → [ α −1 · γ · α ] .

Thus for a path-connected space the fundamental group is the same up to group iso-
morphisms for any choice of base point. Therefore if X is a path-connected space we often
drop the base point in the notation and just write π1 ( X ).

Induced homomorphism
Let X and Y be topological spaces, and f : X → Y. Then f induces the following map

f ∗ : π1 ( X, x0 ) → π1 (Y, f ( x0 ))
[ γ ] 7 → [ f ◦ γ ].

This is a well-defined map (problem 13.11). Notice that f ∗ depends on the base point x0 .
Furthermore f ∗ is a group homomorphism, indeed:

f ∗ ([γ1 ] · [γ2 ]) = f ∗ ([γ1 · γ2 ]) = [ f ◦ (γ1 · γ2 )] = ( f ◦ γ1 ) · ( f ◦ γ2 ) = f ∗ ([γ1 ]) · f ∗ ([γ2 ]).

Example. For id : X → X, the induced map is id∗ = id. For a constant map f : X → { x0 },
the induced map is f ∗ = 1.

Proposition. If f : X → Y and g : Y → Z then ( g ◦ f )∗ = g∗ ◦ f ∗ .

Proof.

( g ◦ f )∗ ([γ]) = [( g ◦ f ) ◦ γ] = [ g ◦ ( f ◦ γ)] = g∗ ([ f ◦ γ)]) = g∗ ( f ∗ ([γ])).

We deduce immediately that if f : X → Y is a homeomorphism then f ∗ : π1 ( X, x0 ) →

π1 (Y, f ( x0 )) is a an isomorphism with inverse ( f ∗ )−1 = ( f −1 )∗ . This implies that the
fundamental group is a topological invariant of path-connected spaces.
102

Homotopy invariance
Lemma. If f : X → X is homotopic to the identity map then f ∗ : π1 ( X, x0 ) → π1 ( X, f ( x0 )) is
an isomorphism.

Proof. From the assumption there is a homotopy F from f to idX . Then α(t) = Ft ( x0 ),
0 ≤ t ≤ 1, is a continuous path from f ( x0 ) to x0 . We will show that f ∗ = hα where hα
is the map used in the proof of 13.8, which was shown there to be an isomorphism. Since
f ∗ ([γ]) = [ f ◦ γ] while hα (γ) = [α · γ · α−1 ], this is reduced to showing that f ◦ γ is path-
homotopic to α · γ · α−1 .

γ
x0

Ft ◦ γ
α(t) = Ft ( x0 )

βt

f ◦γ
f ( x0 )

For each fixed 0 ≤ t ≤ 1, let β t be a path which goes along α from α(0) = f ( x0 )
to α(t), for example given by β t (s) = α(st), 0 ≤ s ≤ 1. For any loop γ at x0 , let Gt =
β t · ( Ft ◦ γ) · β− 1
t . That G is continuous can be checked by writing down the formula for G
explicitly. Then G gives a path-homotopy from f ◦ γ to α · γ · α−1 .

Theorem. If f : X → Y is a homotopy equivalence then f ∗ : π1 ( X, x0 ) → π1 (Y, f ( x0 )) is an

isomorphism.

Proof. Since f is a homotopy equivalence there is g : Y → X such that g ◦ f is homotopic to

idX and f ◦ g is homotopic to idY . By the above lemma, the composition

f∗ g∗
π1 ( X, x0 ) → π1 (Y, f ( x0 )) → π1 ( X, g( f ( x0 )))

is an isomorphism, which implies that g∗ is surjective. Similarly the composition

0
g∗ f∗
π1 (Y, f ( x0 )) → π1 ( X, g( f ( x0 ))) → π1 (Y, f ( g( f ( x0 ))))

is an isomorphism, where f ∗0 denotes the induced homomorphism by f with base point

g( f ( x0 )). This implies that g∗ is injective. Since g∗ is bijective, from the first composition
we see that f ∗ is bijective.

Corollary (homotopy invariance). If two path-connected spaces are homotopic then their funda-
mental groups are isomorphic.

We say that for path-connected spaces, the fundamental group is a homotopy invariant.

Example. The fundamental group of a contractible space is trivial.

13. THE FUNDAMENTAL GROUP 103

Problems
13.9. Suppose that f : X → Y is continuous. Let γ1 and γ2 be two paths in X which are path-
homotopic. Show that f ◦ γ1 and f ◦ γ2 are path-homotopic.

13.10. Show that if X0 is a path-connected component of X and x0 ∈ X0 then π1 ( X, x0 ) is isomorphic

to π1 ( X0 , x0 ).

13.11. Let X and Y be topological spaces, f : X → Y, f ( x0 ) = y0 . Show that the induced map

f ∗ : π1 ( X, x0 ) → π1 (Y, y0 )
[γ] 7→ [ f ◦ γ]

is a well-defined.

13.12. Suppose that Y is a retract of X via a retraction r : X → Y. Let i : Y ,→ X be the inclusion

map. With y0 ∈ Y, show that r∗ : π1 ( X, y0 ) → π1 (Y, y0 ) is surjective and i∗ : π1 (Y, y0 ) → π1 ( X, y0 )
is injective.

13.13. Show that

π1 ( X × Y, ( x0 , y0 )) ∼
= π1 ( X, x0 ) × π1 (Y, y0 ).

13.14. A space is said to be simply connected (đơn liên) if it is path-connected and any loop is path-
homotopic to a constant loop. Show that a space is simply-connected if and only if it is path-connected
and its fundamental group is trivial.

13.15. Show that any contractible space is simply connected.

13.16. Show that a space is simply connected if and only if all paths with same initial points and same
terminal points are path-homotopic, in other words, there is exactly one path-homotopy class from
one point to another point.

13.17. * Show that in a space X the following statements are equivalent:

(a) π1 ( X, x0 ) = 1, ∀ x0 ∈ X.
(b) Every continuous map S1 → X is homotopic to a constant map.
(c) Every continuous map S1 → X has a continuous extension to a map D2 → X.
104

14 The fundamental group of the circle

Theorem (π1 (S1 ) ∼

= Z). The fundamental group of the circle is infinite cyclic.

Let γn be the loop (cos(n2πt), sin(n2πt)), 0 ≤ t ≤ 1, the loop on the circle S1 based
at the point (1, 0) that goes n times around the circle at uniform speed in the counter-
clockwise direction if n > 0 and in the clockwise direction if n < 0. Consider the map

Φ:Z → π1 (S1 , (1, 0))

n 7 → [ γn ] .

This map associates each integer n with the path-homotopy class of γn . We will show that
Φ is a group isomorphism, where Z has the usual additive structure. This implies that the
fundamental group of the circle is generated by a loop that goes once around the circle in
the counter-clockwise direction, and the homotopy class of a loop in the circle corresponds
to an integer representing the “number of times” that the loop goes around the circle, with
the counter-clockwise direction being the positive direction.

Φ is a group homomorphism

This means γm+n is path-homotopic to γm · γn . If m = −n then we can verify from the

−1 ( t ). We have checked earlier 13.4 that
formulas that γn (t) = γ−m (t) = γm (1 − t) = γm
−1 is path-homotopic to the constant loop γ .
γm · γm 0

If m 6= −n then γm · γn is a reparametrization of γm+n so they are path-homotopic by

13.5. Indeed, with

γm+n (t) = (cos((m + n)2πt), sin((m + n)2πt)), 0 ≤ t ≤ 1,

and

(cos(m2π2t), sin(m2π2t)), 0 ≤ t ≤ 21 ,
γm · γn ( t ) =
(cos(n2π (2t − 1)), sin(n2π (2t − 1))), 1
2 ≤ t ≤ 1,

we can find ϕ such that γm+n ◦ ϕ = γm · γn , namely,


m

m+n 2t, 0 ≤ t ≤ 12 ,
ϕ(t) =
 n(2t−1)+m , 1
≤ t ≤ 1.
m+n 2

Covering spaces

Let p : R → S1 , p(t) = (cos(2πt), sin(2πt)), a map that wraps the line around the circle
countably infinitely many times in the counter-clockwise direction, called the projection
map. This is related to the usual parametrization of the circle by angle. The map p is called
the covering map associated with of the covering space (không gian phủ) R of S1 . For a
path γ : [0, 1] → S1 , a path γ̃ : [0, 1] → R such that p ◦ γ̃ = γ is called a lift of γ. For
example γn (t) = p(nt), 0 ≤ t ≤ 1, so γn has a lift γ
fn with γ
fn (t) = nt.
14. THE FUNDAMENTAL GROUP OF THE CIRCLE 105

=R
γ̃
p

[0, 1] / S1
γ

A reason for considering covering spaces is that paths in S1 are projections of paths in
R, while the homotopy of paths in R can be simpler than that in S1 .
As a demonstration, we use this idea of covering space to show again that γm+n is path-
homotopic to γm · γn . Consider the case m + n > 0. Let

m+n : [0, 1]
γ] → [0, m + n]
t 7→ (m + n)t,

then this is a lift of γm+n . Let

m · γn : [0, 1]
γ^ → [0, m + n]

m2t, 0 ≤ t ≤ 21 ,
t 7→
n(2t − 1) + m, 1 ≤ t ≤ 1,
2

then this is a lift of γm · γn . The two paths γ] m · γn are paths in R with same
m+n and γ^
endpoints, so they are path-homotopic, via a path-homotopy such as Fs = (1 − s)γ] m+n +
m · γn , 0 ≤ s ≤ 1 (see 13.2). Then γm+n is path-homotopic to γm · γn via the path-
sγ^
homotopy p ◦ F.

Φ is surjective
This means every loop γ on the circle based at (1, 0) is path-homotopic to a loop γn . Our
argument is based on the fact, proved below at 14.1, that there is a path γ̃ on R starting at
0 which is a lift of γ. Then γ̃(1) is an integer n. On R the path γ̃ is path-homotopic to the
fn , via a path-homotopy F, so γ is path-homotopic to γn via the path-homotopy p ◦ F.
path γ

14.1 Lemma (existence of lift). Every path in S1 has a lift to R. Furthermore if the initial point
of the lift is specified then the lift is unique.

Proof. Let us write S1 = U ∪ V with U = S1 \ {(0, −1)} and V = S1 \ {(0, 1)}. Then
p −1 (U ) = n∈Z ( n − 4 , n + 4 ), consisting of infinitely many disjoint open subsets of R, each
1 3
S

of which is homeomorphic to U via p, i.e. p : (n − 41 , n + 43 ) → U is a homeomorphism, in

particular the inverse map exists and is continuous. The same happens with respect to V.
Let γ : [0, 1] → S1 , γ(0) = (1, 0). We can divide [0, 1] into sub-intervals with endpoints
0 = t0 < t1 < · · · < tn = 1 such that on each sub-interval [ti−1 , ti ], 1 ≤ i ≤ n, the path γ
is either contained in U or in V. This is guaranteed by the existence of a Lebesgue number
(see 6.3) with respect to the open cover γ−1 (U ) ∪ γ−1 (V ) of [0, 1].
Suppose an initial point γ̃(0) is chosen, an integer. Suppose that γ̃ has been uniquely
extended to [0, ti−1 ] for a certain 1 ≤ i ≤ n. If γ([ti−1 , ti ]) ⊂ U then there is a unique ni ∈ Z
such that γ̃(ti−1 ) ∈ (ni − 14 , ni + 34 ). There is only one way to continuously extend γ̃ to
[ti−1 , ti ], that is by defining γ̃ = p|−1 ◦ γ. See figure 14.2. In this way inductively γ̃
(ni − 14 ,ni + 34 )
is extended continuously to [0, 1], uniquely.
106

ni n i +1 p −1 (U )

R
p

γ̃

γ
S1
0 1
(1, 0)
t i −1 ti

Figure 14.2: Construction of a lift of a path.

Examining the proof above we can see that the key property of the covering space p :
R → S1 is the following: each point on the circle has an open neighborhood U such that the
preimage p−1 (U ) is the disjoint union of open subsets of R, each of which is homeomorphic
to U via p. This is the defining property of general covering spaces.

Φ is injective
This is reduced to showing that if γm is path-homotopic to γn then m = n. Our proof is
based on another important result below, 14.3, which says that if γm is path-homotopic to
γn then γf
m is path-homotopic to γ
fn . This implies the terminal point m of γfm must be the
same as the terminal point n of γ
fn .

14.3 Lemma (homotopy of lifts). Lifts of path-homotopic paths with same initial points are path-
homotopic.

Proof. The proof is similar to the above proof of 14.1. Let F : [0, 1] × [0, 1] → S1 be a path-
homotopy from the path F0 to the path F1 . If the two lifts Fe0 and Fe1 have same initial points
then that initial point is the lift of the point F ((0, 0)).
As we noted earlier, the circle has an open cover O such that each U ∈ O we have p−1 (U )
is the disjoint union of open subsets of R, each of which is homeomorphic to U via p. The
collection { F −1 (U ) | U ∈ O} is an open cover of the square [0, 1] × [0, 1]. By the existence
of Lebesgue’s number, there is a partition of [0, 1] × [0, 1] into sub-rectangles such that each
sub-rectangle is contained in an element of F −1 (O). More concisely, we can divide [0, 1]
into sub-intervals with endpoints 0 = t0 < t1 < · · · < tn = 1 such that for each 1 ≤ i, j ≤ n
there is U ∈ O such that F ([ti−1 , ti ] × [t j−1 , t j ]) ⊂ U.
We will build up F̃ sub-rectangle by sub-rectangle, going on columns from left to right
and on each column going from bottom up. We already have F̃ ((0, 0)). Suppose that now
it is the turn of the sub-rectangle [ti−1 , ti ] × [t j−1 , t j ], for some 1 ≤ i, j ≤ n, on which F̃ is
to be extended to. We have F ([ti−1 , ti ] × [t j−1 , t j ]) ⊂ U for a certain U ∈ O. Let Ũ be the
unique open subset of R such that Ũ contains the point F̃ ((ti−1 , t j−1 )) and p|Ũ : Ũ → U
14. THE FUNDAMENTAL GROUP OF THE CIRCLE 107

Ũ p −1 (U )

R
p

F̃

F
S1

1 (1, 0)

t j −1

0 t i −1 1

Figure 14.4: Construction of a lift of a path-homotopy.

is a homeomorphism. By the way we proceed one sub-rectangle at a time, the intersection

A of the previous domain of F̃ and the sub-rectangle [ti−1 , ti ] × [t j−1 , t j ] is either one point,
one edge, or the union of two edges with a common vertex, in all cases A is connected
containing the vertex (ti−1 , t j−1 ), therefore F̃ ( A) is connected. Since F̃ ( A) ⊂ p−1 (U ), a
disjoint union of open sets, and F̃ ( A) ∩ Ũ 6= ∅, we must have F̃ ( A) ⊂ Ũ. We extend F̃
−1
by defining F̃ on the sub-rectangle [ti−1 , ti ] × [t j−1 , t j ] to be p|Ũ ◦ F. This extension agrees
with the previous function on the common domain A. See figure 14.4. The continuity of F̃
follows from gluing of continuous functions, see 3.15.
Thus we obtained a continuous lift F̃ of F. Since the initial point is given, by uniqueness
of lifts of paths in 14.1, the restriction of F̃ to [0, 1] × {0} is Fe0 while the restriction of F̃ to
[0, 1] × {1} is Fe1 . Also by uniqueness of lifts, the lifts of F on {0} × [0, 1] and {1} × [0, 1] are
constants. Thus F̃ is a path-homotopy from Fe0 to Fe1 .

That the fundamental group of the circle is non-trivial gives us:

Corollary. The circle is not contractible.

14.5 Corollary. There cannot be any retraction from the disk D2 to its boundary S1 .

Proof. Suppose there is a retraction r : D2 → S1 . Let i : S1 ,→ D2 be the inclusion map.

i r
From the diagram S1 → D2 → S1 we have r ◦ i = idS1 , therefore on the fundamental
groups (r ◦ i )∗ = r∗ ◦ i∗ = idπ1 (S1 ) . A consequence is that r∗ is onto, but this is not possible
since π1 ( D2 ) is trivial while π1 (S1 ) is non-trivial.

A proof of this result for higher dimensions is presented in 17.2.

The important Brouwer fixed point theorem follows from that simple result:

14.6 Theorem (Brouwer fixed point theorem for dimension two). Any continuous map from
the disk D2 to itself has a fixed point.

Proof. Suppose that f : D2 → D2 does not have a fixed point, i.e. f ( x ) 6= x for all x ∈ D2 .
The straight line from f ( x ) to x will intersect the boundary ∂D2 at a point g( x ).
We can check that g is continuous (see 14.10). Then g : D2 → ∂D2 is a retraction,
contradicting 14.5.
108

x g( x )

f (x)

Problems
14.7. Find the fundamental groups of:

(a) the Mobius band,

(b) the cylinder.

14.8. Show that the plane minus a point is not simply connected.

14.9. Compute the fundamental group of the torus.

14.10. Check that the map g in the proof of the Brouwer fixed point theorem is indeed continuous.
15. VAN KAMPEN THEOREM 109

15 Van Kampen theorem

Van Kampen theorem is about giving the fundamental group of a union of subspaces from
the fundamental groups of the subspaces.

Example (S1 ∨ S1 ). Two circles with one common point (the figure 8) is called a wedge
product S1 ∨ S1 , see 9.15. In problem 15.4 we see that this is homotopic to the plane minus
two points. Let x0 be the common point, let a be a loop starting at x0 going once around the
first circle and let b the a loop starting at x0 going once around the second circle. Then a and
b generate the fundamental groups of the two circles with based points at x0 . Intuitively
we can see that π1 (S1 ∨ S1 , x0 ) consists of path-homotopy classes of loops like a, ab, bba,
aabab−1 a−1 a−1 , . . . This is a group called the free group generated by a and b, denoted by
h a, bi.
a b

x0

Free group
Let S be a set. Let S−1 be a disjoint set having a bijection with S. Corresponding to each
element x ∈ S is an element in S−1 denoted by x −1 . A word with letters in S is a finite
sequence of elements in S ∪ S−1 . The sequence with no element is called the empty word.
Given two words we form a new word by juxtaposition (đặt kề): (s1 s2 · · · sn ) · (s10 s20 · · · s0m ) =
s1 s2 · · · sn s10 s20 · · · s0m . With this operation the set of all words with letters in S becomes a
group. The identity element 1 is the empty word. The inverse element of a word s1 s2 · · · sn
is the word s− 1 −1 −1
n sn−1 · · · s1 . This group is called the free group generated by the set S,
denoted by hSi.

Example. The free group h{ a}i generated by the set { a} is often written as h ai. As a set h ai
is often written as { an | n ∈ Z}. The product is given by am · an = am+n . The identity is a0 .
Thus as a group h ai is an infinite cyclic group, isomorphic to (Z, +).

Let G be a set and let R be a set of words with letters in G, i.e. R is a finite subset of the
free group h G i. Let N be the smallest normal subgroup of h G i containing R. The quotient
group h G i /N is written h G | Ri. Elements of G are called generators of this group and
elements of R are called relators of this group. We can think of h G | Ri as consisting of
words in G subjected to the relations r = 1 for all r ∈ R.

a | a2 = { a0 , a1 } ∼
= Z2 .


Example.

Free product of groups

Let G and H be groups. Form the set of all words with letters in G or H. In such a word,
two consecutive elements from the same group can be reduced by the group operation. For
example ba2 ab3 b−5 a4 = ba3 b−2 a4 . In particular if x and x −1 are next to each other then they
will be canceled. The identities of G and H are also reduced. For example abb−1 c = a1c =
ac.
110

As with free groups, given two words we form a new word by juxtaposition. For ex-
ample ( a2 b3 a−1 ) · ( a3 ba) = a2 b3 a−1 a3 ba = a2 b3 a2 ba. This is a group operation, with the
identity element 1 being the empty word, the inverse of a word s1 s2 · · · sn is the word
s− 1 −1 −1
n sn−1 · · · s1 . This group is called the free product of G with H.

Example (G ∗ H 6= G × H). We have

h gi ∗ hhi = h g, hi = { gm1 hn1 gm2 hn2 · · · gmk hnk | m1 , n1 , . . . , mk , nk ∈ Z, k ∈ Z+ }.

Compare that to h gi × hhi = {( gm , hn ) | m, n ∈ Z} with component-wise multiplication.

This group can be identified with h g, h | gh = hgi = { gm hn | m, n ∈ Z}. Thus Z ∗ Z 6=
Z × Z.

For more details on free group and free product, see textbooks on Algebra such as
[Gal10] or [Hun74].

Van Kampen theorem

Theorem (Van Kampen theorem). Suppose that U, V ⊂ X are open, path-connected, U ∩ V is
path-connected, and x0 ∈ U ∩ V. Let iU : U ∩ V ,→ U and iV : U ∩ V ,→ V be inclusion maps.
Then
π1 (U, x0 ) ∗ π1 (V, x0 )
π1 (U ∪ V, x0 ) ∼
= .
h{(iU )∗ (α)(iV )∗ (α)−1 | α ∈ π1 (U ∩ V, x0 )}i

Proof. Let γ be any loop in X at x0 . By the existence of a Lebesgue number in association

with the open cover {γ−1 (U ), γ−1 (V )} of [0, 1], there is a partition of [0, 1] by 0 = t0 <
t1 < · · · < tn = 1 such that γ([ti−1 , ti ]) ⊂ U or γ([ti−1 , ti ]) ⊂ V for every 1 ≤ i ≤ n,
and furthermore
 we can arrange so that γ(ti ) ∈ U ∩ V for every 1 ≤ i ≤ n. Let γi be
the path γ [ti−1 ,ti ] reparametrized to the domain [0, 1]. Then γ has a reparametrization as


γ1 · γ2 · · · γn . Let β i be a path in U ∩ V from γ(ti ) to x0 , 1 ≤ i ≤ n − 1. In terms of

path-homotopy in U ∪ V we have:

γ ∼ γ1 · γ2 · · · γn
∼ (γ1 · β 1 ) · ( β− 1 −1 −1
1 · γ2 · β 2 ) · · · ( β n−2 · γn−1 · β n−1 )( β n−1 · γn ).

Thus every loop at x0 in U ∪ V is path-homotopic to a product of loops at x0 each of which

is contained entirely either in U or in V.

U V

γ1 γ2
γ ( t1 )
γ ( t2 )

γ3 β1
β2
γ ( t3 )

γ
x0 γ4
β3
15. VAN KAMPEN THEOREM 111

Let jU : U ,→ U ∪ V and jV : V ,→ U ∪ V be inclusion maps. Let

Φ : π1 (U, x0 ) ∗ π1 (V, x0 ) → π1 (U ∪ V, x0 )
a1 b1 · · · an bn 7→ ( jU )∗ ( a1 ) ( jV )∗ (b1 ) · · · ( jU )∗ ( an ) ( jV )∗ (bn ).

It is immediate that Φ is a homomorphism. The above discussion implies that Φ is surjec-

tive.
For the kernel of Φ, let α ∈ π1 (U ∩ V, x0 ), then

Φ((iU )∗ (α)(iV )∗ (α)−1 ) = ( jU )∗ ((iU )∗ (α)) ( jV )∗ ((iV )∗ (α)−1 ))

= ( jU ◦ iU )∗ (α) ( jV ◦ iV )∗ (α)−1 = 1,

by noticing that jU ◦ iU = jV ◦ iV . Thus ker Φ contains the normal subgroup generated by

all elements (iU )∗ (α)(iV )∗ (α)−1 , α ∈ π1 (U ∩ V, x0 ).
That ker Φ is equal to that group is more difficult to prove, we will not present a proof.
The reader can read for instance in [Vic94].

Corollary. The spheres of dimensions greater than one are simply connected:

Z, n=1
π1 ( S n ) ∼
=
1, n > 1.

Proof. Let A = Sn \ {(0, 0, . . . , 0, 1)} and B = Sn \ {(0, 0, . . . , 0, −1)}. Then A and B are
contractible. If n ≥ 2 then A ∩ B is path-connected. By Van Kampen theorem, π1 (S2 ) = ∼
π1 ( A) ∗ π1 ( B) = 1.

Corollary. If X = U ∪ V with U, V open, path-connected, U ∩ V is simply connected, and

x0 ∈ U ∩ V, then π1 ( X, x0 ) ∼
= π1 (U, x0 ) ∗ π1 (V, x0 ).

Example (S1 ∨ S1 ). Let U be the union of the first circle with an open arc on the second
circle containing the common point. Similarly let V be the union of the second circle with an
open arc on the first circle containing the common point. Clearly U and V have deformation
retractions to the first and the second circles respectively, while U ∩ V is simply connected
(has a deformation retraction to the common point). Applying the Van Kampen theorem,
or the corollary above, we get

π1 ( S1 ∨ S1 ) ∼
= π1 ( S1 ) ∗ π1 ( S1 ) ∼
= Z ∗ Z.

The fundamental group of a cell complex

15.1 Theorem. Let X be a path-connected topological space and consider the space X t f D n ob-
tained by attaching an n-dimensional cell to X via the map f : ∂D n = Sn−1 → X.

(a) If n > 2 then π1 ( X t f D n ) ∼

= π1 ( X ).

(b) If n = 2: Let γ1 be the loop (cos 2πt, sin 2πt) on the circle ∂D2 . Let x0 = f (1, 0). Then
π1 ( X t f D n , x0 ) ∼
= π1 ( X, x0 )/ h[ f ◦ γ1 ]i.

Intuitively, gluing a 2-disk destroys the boundary circle of the disk homotopically,
while gluing disks of dimensions greater than 2 does not affect the fundamental group.
112

Proof. Let Y = X t f D n . Let U = X t f { x ∈ D n | k x k > 12 } ⊂ Y. There is a deformation

retraction from U to X. Let V be the image of the imbedding of the interior of D n in Y. Then
1
V is contractible. Also, U ∩ V is homeomorphic to { x ∈ D n |
< k x k < 1}, which has a 2
deformation retraction to S n −1 .
We now apply Van Kampen theorem to the pair (U, V ).
When n > 2 the fundamental group of U ∩ V is trivial, therefore π1 (Y ) ∼
= π 1 (U ) ∼
=
π1 ( X ).
Consider the case n = 2. Let y0 ∈ U ∩ V be the image of the point ( 32 , 0) ∈ D2 , let γ
be imbedding of the loop 32 (cos 2πt, sin 2πt) starting at y0 going once around the annulus
U ∩ V. Then [γ] is a generator of π1 (U ∩ V, y0 ). In V the loop γ is homotopically trivial,
therefore π1 (Y, y0 ) ∼
= π1 (U, y0 )/ h[γ]i. Under the deformation retraction from U to X, [γ]
becomes [ f ◦ γ1 ]. Therefore π1 (Y, x0 ) ∼
= π1 ( X, x0 )/ h[ f ◦ γ1 ]i.

This result shows that the fundamental group only gives information about the two-
dimensional skeleton of a cell complex, it does not give information on cells of dimensions
greater than 2.

15.2 Example. Consider the space below. As a cell-complex it consists of one 0-cell, one

1-cell (represented by a), forming the 1-dimensional skeleton (a circle, also denoted by a),
and one 2-cell attached to the 1-dimensional skeleton by wrapping the boundary of the
disk around a three times. Thus the fundamental group of the space is isomorphic to
a | a3 = 1 ∼= Z3 . In practice, we get the fundamental group immediately by seeing that

the boundary of the disk is a3 and gluing this disk homotopically destroys a3 .

The fundamental groups of surfaces

By the classification theorem, any compact without boundary surface is obtained by iden-
tifying the edges of a polygon following a word as in 11.2. As such it has a cell complex
structure with a two-dimensional disk glued to the boundary of the polygon under the
equivalence relation, which is a wedge of circles. An application of 15.1 gives us:

Theorem. The fundamental group of a connected compact surface S is isomorphic to one of the
following groups:

(a) trivial group, if S = S2 ,

D E
(b) a1 , b1 a2 , b2 , . . . , a g , bg | a1 b1 a1−1 b1−1 a2 b2 a2−1 b2−1 · · · a g bg a−
g
1 b−1 , if S is the orientable sur-
g
face of genus g,
D E
(c) c1 , c2 , . . . , c g | c21 c22 · · · c2g , if S is the unorientable surface of genus g.

Example. For the torus: π1 ( T1 ) ∼

= h a, b | ab = bai ∼
= Z × Z.
For the projective plane: π1 (RP ) =
2 ∼ c|c =1 ∼
2
= Z2 .


15. VAN KAMPEN THEOREM 113

Problems
15.3. Use the Van Kampen theorem to find the fundamental groups of the following spaces:

(a) A wedge of finitely many circles.

(b) S1 ∨ S2 .
(c) S2 ∨ S3 .

15.4. Show that the plane minus finitely many points has a deformation retraction to a bouquet of
circles, then find its fundamental group.

15.5. Find the fundamental group of the Euclidean space R3 minus finitely many points.

15.6. By problem 9.28 we have two different presentations for the fundamental group of the Klein
bottle: π1 (K ) ∼= a, b | aba−1 b and π1 (K ) ∼
= a, b | a2 b2 . Show directly that a, b | aba−1 b ∼




=
a, b | a2 b2 .

15.7. Is the fundamental group of the Klein bottle abelian?

15.8. Show that the fundamental groups of the one-hole torus and the two-holes torus are not iso-
morphic. Therefore the two surfaces are different.

15.9. Find a space whose fundamental group is isomorphic to Z ∗ Z2 ∗ Z3 .

15.10. Show that if X = A ∪ B, where A and B are open, simply connected, and A ∩ B is path-
connected, then X is simply connected.

15.11. Is R3 \ {(0, 0, 0)} simply connected? How about R3 \ {(0, 0, 0), (1, 0, 0)}?

15.12. If a compact surface is simply connected then it is homeomorphic to the two-dimensional

sphere. This is a motivational analogy for the three-dimensional Poincaré conjecture.

15.13. Consider the three-dimensional torus T 3 , obtained from the cube [0, 1]3 by identifying opposite
faces by projection maps, that is ( x, y, 0) ∼ ( x, y, 1), (0, y, z) ∼ (1, y, z), ( x, 0, z) ∼ ( x, 1, z), ∀( x, y, z) ∈
[0, 1]3 .

(a) Show that T 3 is homeomorphic to S1 × S1 × S1 .

(b) Show that T 3 is a 3-dimensional manifold.
(c) Construct a cellular structure on T 3 .
(d) Compute the fundamental group of T 3 .

15.14. * Consider the space X obtained from the cube [0, 1]3 by rotating each lower face (i.e. faces on
the planes xOy, yOz, zOx) an angle of π/2 about the normal line that goes through the center of the
face, then gluing this face to the opposite face. For example the point (1,1,0) will be glued to the point
(0, 1, 1).

(a) Show that X has a cellular structure consisting of 2 0-cells, 4 1-cells, 3 2-cells, 1 3-cell.
(b) Compute the fundamental group of X.
(c) Check that the fundamental group of X is isomorphic to the quaternion group.

15.15. If
G = h g1 , g2 , . . . , g m 1 | r 1 , r 2 , . . . , r n 1 i

and
H = h h 1 , h 2 , . . . , h m2 | s1 , s2 , . . . , s n2 i

then
G ∗ H = h g1 , g2 , . . . , g m 1 , h 1 , h 2 , . . . , h m 2 | r 1 , r 2 , . . . , r n 1 , s 1 , s 2 , . . . , s n 2 i .
114

16 Simplicial homology
In homology theory we associate each space (with certain additional structures) with a
sequence of groups.

Oriented simplex
On a simplex of dimension greater than 0, consider the relation on the collection of ordered
sets of vertices of this simplex whereas two order sets of vertices are related if they differ by
an even permutation. This is an equivalence relation. Each of the two equivalence classes
is called an orientation of the simplex. If we choose an orientation, then the simplex is said
to be oriented.
In more details, an orientation for a set A consisting of n elements is represented by
bijective map from the set of integers I = {1, 2, . . . , n} to A. If o and o 0 are two such maps,
then o = o 0 ◦ σ where σ = o 0−1 ◦ o : I → I is bijective, a permutation. Any permutation falls
into one of two classes: even or odd permutations. We say that o and o 0 represent the same
orientations if σ is an even permutation.
For each simplex of dimension greater than 0 there are two oriented simplexes. For
convenience we say that for a 0-dimensional simplex (a vertex) there is only one orientation.

Remark. In this section, to be clearer we reserve the notation v0 v1 · · · vn for an un-oriented

simplex with the set of vertices {v0 , v1 , . . . , vn }, and the notation [v0 , v1 , . . . , vn ] for an ori-
ented simplex with the same set of vertices, i.e. the simplex v0 v1 · · · vn with this particular
order of vertices. (Later, and in other sources, this convention can be dropped.)

Example. A 1-dimensional simplex in Rn is a straight segment connecting two points.

Choosing one point as the first point and the other point as the second gives an orienta-
tion to this simplex. Intuitively, this means to give a direction to the straight segment. If
the two vertices are labeled v0 and v1 , then an ordered pair [v0 , v1 ] gives an orientation for
the simplex v0 v1 , while an ordered pair [v1 , v0 ] gives an oriented simplex with the opposite
orientation.
Consider a 2-dimensional simplex. Let v0 , v1 , v2 be the vertices. The oriented simplexes
[v0 , v1 , v2 ], [v1 , v2 , v0 ], [v2 , v0 , v1 ] have same orientations, opposite to the orientations of the
oriented simplexes [v1 , v0 , v2 ], [v2 , v1 , v0 ], [v0 , v2 , v1 ].

v2

v0 v1

Figure 16.1:

Chain
Let X be a simplicial complex in a Euclidean space. For each integer n let Sn ( X ) be the
free abelian group generated by all n-dimensional oriented simplexes in X modulo the
16. SIMPLICIAL HOMOLOGY 115

relation that if σ and σ0 are oriented simplexes with opposite orientations then σ = −σ0 .
Each element of Sn ( X ) is an equivalence class, called an n-dimensional chain (xích), is
represented by a finite sum of integer multiples of n-dimensional oriented simplexes, i.e.
of the form ∑im=1 ni σi where σi is an n-dimensional oriented simplex of X and ni ∈ Z.
We often drop the notation for equivalence class and use a representative to indicate its
equivalence class.
The group Sn ( X ) is isomorphic to the free abelian group generated by the set of all
n-simplexes of X.
If n < 0 or n > dim X then Sn ( X ) is assigned to be the trivial group 0.

Example. Consider the simplicial complex whose underlying space is a line segment. So
X = { AB, A, B}. The following is a 0-chain: 2A − 3B. The group S0 ( X ) is generated by A
and B. An example of a 1-chain is 2[ A, B] + 5[ B, A] = 2[ A, B] − 5[ A, B] = −3[ A, B]. The
group S1 ( X ) is generated by [ A, B] and [ B, A] together with the relation [ B, A] = −[ A, B].
We can identify S1 ( X ) with the group generated by [ A, B] and write an element of S1 ( X )
as n[ A, B] with n ∈ Z.

16.2 Example. Consider the simplicial complex whose underlying space is the triangle in
figure 16.1:
X = { v0 v1 v2 , v0 v1 , v1 v2 , v2 v0 , v0 , v1 , v2 }.

For n = 0:
S0 ( X ) = { n 0 v 0 + n 1 v 1 + n 2 v 2 | n 0 , n 1 , n 2 ∈ Z } .

For n = 1: In the group S1 ( X ) we have the relations [v1 , v0 ] = −[v0 , v1 ], [v2 , v1 ] =

−[v1 , v2 ], [v0 , v2 ] = −[v2 , v0 ]. We write

S1 ( X ) = { n 0 [ v 0 , v 1 ] + n 1 [ v 1 , v 2 ] + n 2 [ v 2 , v 0 ] | n 0 , n 1 , n 2 ∈ Z } .

For n = 2: Here for example [v0 , v1 , v2 ] = −[v2 , v1 , v0 ] = [v1 , v2 , v0 ]. We can write

S2 ( X ) = { n 0 [ v 0 , v 1 , v 2 ] | n 0 ∈ Z } .

Boundary
Let σ = [v0 , v1 , . . . , vn ] be an n-dimensional oriented simplex. Define the boundary of σ to
be the following (n − 1)-dimensional chain, the alternating sum of the (n − 1)-dimensional
faces of σ:

n
∂n σ = ∑ (−1)i [v0 , v1 , . . . , vi−1 , vbi , vi+1 , . . . , vn ],
i =0

where the notation vbi is traditionally used to indicate that this element is dropped. This
map is extended linearly to become a map from Sn ( X ) to Sn−1 ( X ), namely
!
m m
∂n ∑ ni σi = ∑ ni ∂n (σi ) .
i =1 i =1

We can check easily that ∂n (−σ) = −∂n (σ ), thus the map ∂n , which is defined on ori-
ented simplexes, induces a well-defined map on Sn ( X ), which is a group homomorphism
from Sn ( X ) to Sn−1 ( X ).
116

Remark. If n < 0 or n > dim X then ∂n = 0 since Sn ( X ) = 0.

Example. Continuing example 16.2:

∂1 ([v0 , v1 ]) = v1 − v0 ,

∂1 ([v1 , v2 ]) = v2 − v1 ,

∂1 ([v2 , v0 ]) = v0 − v2 ,

∂2 ([v0 , v1 , v2 ]) = [v1 , v2 ] − [v0 , v2 ] + [v0 , v1 ] = [v1 , v2 ] + [v2 , v0 ] + [v0 , v1 ].

Notice that:

∂1 (∂2 ([v0 , v1 , v2 ])) = ∂1 ([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ])

= (v1 − v0 ) + (v2 − v1 ) + (v0 − v2 ) = 0.

This illustrates that intuitively “a boundary has empty boundary”:

16.3 Proposition (boundary of boundary is zero). ∂n−1 ◦ ∂n = 0.

Proof. Let σ = [v0 , v1 , . . . , vn ], an oriented n-simplex. By definition,

n
∂n (σ ) = ∑ (−1)i [v0 , v1 , . . . , vi−1 , vbi , vi+1 , . . . , vn ].
i =0

Then
n
∂ n −1 ∂ n ( σ ) = ∑ (−1)i ∂n−1 ([v0 , v1 , . . . , vi−1 , vbi , vi+1 , . . . , vn ])
i =0
n i −1
= ∑ (−1)i ( ∑ (−1) j [v0 , . . . , vbj , . . . , vbi , . . . , vn ] +
i =0 j =0
n −1
+ ∑ (−1) j [v0 , . . . , vbi , . . . , vd
j+1 , . . . , vn ])
j =i

= ∑ (−1)i+ j [v0 , . . . , vbj , . . . , vbi , . . . , vn ] +

0≤ j < i ≤ n

+ ∑ (−1)i+ j [v0 , . . . , vbi , . . . , vd

j +1 , . . . , v n ]
0≤ i ≤ j ≤ n −1

= ∑ (−1)i+ j [v0 , . . . , vbj , . . . , vbi , . . . , vn ] +

0≤ j < i ≤ n

+ ∑ (−1)i+k−1 [v0 , . . . , vbi , . . . , vbk , . . . , vn ]

0≤i <k ≤n, (k = j+1)

= ∑ (−1)i+ j [v0 , . . . , vbj , . . . , vbi , . . . , vn ] +

0≤ j < i ≤ n

+ ∑ −(−1)i+ j [v0 , . . . , vbj , . . . , vbi , . . . , vn ]

0≤ j < i ≤ n
= 0.

That ∂n−1 ◦ ∂n = 0 can be interpreted as

Im(∂n+1 ) ⊂ ker(∂n ).
16. SIMPLICIAL HOMOLOGY 117

Elements of ker(∂n ) are often called closed chains or cycles, while elements of Im(∂n+1 ) are
called exact chains or boundaries. We have just observed that

exact ⇒ closed

boundaries are cycles

Homology
In general, a sequence of groups and homomorphisms

∂ n +2 ∂ n +1 ∂ ∂ n −1 ∂
· · · → Sn+1 → Sn →n Sn−1 → · · · →1 S0

satisfying Im(∂n+1 ) ⊂ ker(∂n ), i.e. ∂n ◦ ∂n+1 = 0 for all n ≥ 0 is called a chain complex
(phức xích). If furthermore Im(∂n+1 ) = ker(∂n ), ∀n ≥ 0 then the chain complex is called
exact (khớp).
Notice that if the groups Sn are abelian then Im(∂n+1 ) is a normal subgroup of ker(∂n ).
The n-dimensional simplicial homology group (nhóm đồng điều) of an oriented simpli-
cial complex X is defined to be the quotient group

ker(∂n )
Hn ( X ) = .
Im(∂n+1 )

Briefly:
Hn = closed n-chains/exact n-chains = n-cycles/n-boundaries.

Thus homology measures the set of cycles which are not boundaries. Even more intuitively,
a cycle which is not a boundary is a “hole” in space, so homology counts higher dimensional
holes.

Remark. If n < 0 or n > dim X then Hn ( X ) = 0 since Sn ( X ) = 0. If n = 0 then H0 ( X ) =

S0 ( X )/ Im ∂1 . If n = dim X then Hn ( X ) = ker ∂n .

Example. Continuing example 16.2, we compute the homology groups of X.

From

∂1 (n0 [v0 , v1 ] + n1 [v1 , v2 ] + n2 [v2 , v0 ]) = n0 ( v1 − v0 ) + n1 ( v2 − v1 ) + n2 ( v0 − v2 ),

we see that an element of Im ∂1 has the form

n0 ( v1 − v0 ) + n1 ( v2 − v1 ) + n2 ( v0 − v2 ) = n0 ( v1 − v0 ) + n1 ( v2 − v0 + v0 − v1 ) + n2 ( v0 − v2 )
= (n0 − n1 )(v1 − v0 ) + (n1 − n2 )(v2 − v0 ).

Thus Im ∂1 = h{v1 − v0 , v2 − v0 }i. So

H0 ( X ) = S0 ( X )/ Im ∂1 ∼= h v0 , v1 , v2 | v1 = v0 , v2 = v0 i
∼
= h v0 i ∼
= h v1 i ∼
= h v2 i ∼
= Z.

Since

n0 ( v1 − v0 ) + n1 ( v2 − v1 ) + n2 ( v0 − v2 ) = ( n2 − n0 ) v0 + ( n0 − n1 ) v1 + ( n1 − n2 ) v2
118

we see that ∂1 (n0 [v0 , v1 ] + n1 [v1 , v2 ] + n2 [v2 , v0 ]) = 0 if and only if n0 = n1 = n2 . So

ker ∂1 = {n([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ]) | n ∈ Z} = h([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ])i .

Since
∂2 (n[v0 , v1 , v2 ]) = n([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ])

we see that

Im ∂2 = {n([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ]) | n ∈ Z} = h([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ])i .

Thus Im ∂2 = ker ∂1 , so
H1 ( X ) = 0.

We also see immediately that ∂2 (n[v0 , v1 , v2 ]) = 0 if and only if n = 0. Thus ker ∂2 = 0 and

H2 ( X ) = ker ∂2 = 0.

Example. Instead of a 2-dimensional triangle in example 16.2, let us consider a 1-dimensional

triangle, namely
X = { v0 v1 , v1 v2 , v2 v0 , v0 , v1 , v2 }.

v2

v0 v1

In this case H1 ( X ) = ker ∂1 = h([v0 , v1 ] + [v1 , v2 ] + [v2 , v0 ])i ∼

= Z, H0 ( X ) = hv0 i ∼
= Z,
and Hn ( X ) = 0 for n 6= 0, 1.
For more on simplicial homology one can read [Mun84], [Cro78], or [Ams83]. In [Hat01]
Hatcher used a modified notion called ∆-complex, different from simplicial complex. For
algorithms for computation of simplicial homology, see [KMM04, chapter 3].

Problems
16.4. Compute the homology of the simplicial complex described in the picture:

(a)

(b)

(c)

(d)

(e)
16. SIMPLICIAL HOMOLOGY 119

16.5. Compute the simplicial homology group H2 ( X ) where X is the simplicial complex representing
a tetrahedron in R3 , namely

X = { v0 , v1 , v2 , v3 , v0 v1 , v1 v2 , v2 v0 , v3 v0 , v3 v1 , v3 v2 , v0 v1 v2 , v1 v2 v3 , v2 v3 v0 , v0 v1 v3 }.

16.6. Show that if the underlying space of a simplical complex is connected then for any two vertices
there is a continuous path on the edges connecting the two vertices.

16.7. Given a simplicial complex X.

(a) Show that if the underlying space | X | is connected then H0 ( X ) ∼

= hvi ∼
= Z where v is any
vertex of X.
(b) Show that if | X | has k connected components then H0 ( X ) ∼
= ik=1 hvi i ∼
= Zk , where vi , 1 ≤ i ≤
L

k are arbitrary vertices in different components of | X |.

16.8. Show that if the underlying space of a simplicial complex X has k connected components Xi ,
∼ Lk Hn ( Xi ).
1 ≤ i ≤ k, then for any n ∈ Z we have Hn ( X ) = i =1
120

17 Singular homology

A singular simplex (đơn hình suy biến, kì dị) is a continuous map from a standard simplex
to a topological space. More precisely, an n-dimensional singular simplex in a topological
space X is a continuous map σ : ∆n → X, where ∆n is the standard n-simplex.
Let Sn ( X ) be the free abelian group generated by all n-dimensional singular simplexes
in X. As a set ( )
m
Sn ( X ) = ∑ ni σi | σi : ∆n → X, ni ∈ Z, m ∈ Z+ .
i =1

Each element of Sn ( X ) is a finite sum of integer multiples of n-dimensional singular sim-

plexes, called a singular n-chain.
Let σ be an n-dimensional singular simplex in X, i.e., a map

σ : ∆n → X
( t0 , t1 , . . . , t n ) 7 → σ ( t0 , t1 , . . . , t n ).

For 0 ≤ i ≤ n define the ith face of σ to be the (n − 1)-singular simplex

∂in (σ ) : ∆n−1 → X
(t0 , t1 , . . . , tn−1 ) 7→ σ(t0 , t1 , . . . , ti−1 , 0, ti , . . . , tn−1 ).

Define the boundary of σ to be the singular (n − 1)-chain

n
∂n (σ ) = ∑ (−1)i ∂in (σ).
i =0

Intuitively one can think of ∂n (σ ) as σ |∂∆n . The map ∂n is extended linearly to become a
group homomorphism from Sn ( X ) to Sn−1 ( X ).

Example. A 0-dimensional singular simplex in X is a point in X.

A 1-dimensional singular simplex is a continuous map σ(t0 , t1 ) with t0 , t1 ∈ [0, 1] and
t0 + t1 = 1. Its image is a curve between the points A = σ (1, 0) and B = σ(0, 1). Its
boundary is − A + B.
A 2-dimensional singular simplex is a continuous map σ(t0 , t1 , t2 ) with t0 , t1 , t2 ∈ [0, 1]
and t0 + t1 + t2 = 1. Its image is a “curved triangle” between the points A = σ (1, 0, 0),
B = σ (0, 1, 0), and C = (0, 0, 1). Intuitively, the image of the face ∂0 is the “curved edge” BC,
the image of ∂1 is AC, and the image of ∂2 is AB. The boundary is ∂0 − ∂1 + ∂2 . Intuitively,
it is BC − AC + AB.

Similar to the case of simplicial complex 16.3, we have:

17.1 Proposition. ∂n−1 ◦ ∂n = 0, ∀n ≥ 2.

Proof. Let σ be a singular n-simplex. From definition:

17. SINGULAR HOMOLOGY 121

!
n
( ∂ n −1 ∂ n ) ( σ ) = ∂ n −1 ∑ (−1)i ∂in (σ)
i =0
!
n −1 n
∑ ∑
j
= (−1) j ∂n−1 (−1)i ∂in (σ)
j =0 i =0

∑
j
= (−1)i+ j ∂n−1 ∂in (σ)
0≤i ≤n,0≤ j≤n−1

∑
j
= (−1)i+ j ∂n−1 ∂in (σ) +
0≤ j < i ≤ n

∑
j
+ (−1)i+ j ∂n−1 ∂in (σ)
0≤ i ≤ j ≤ n −1

∑
j
= (−1)i+ j ∂n−1 ∂in (σ) +
0≤ j < i ≤ n

+ ∑ (−1)i+k−1 ∂kn− 1 i
−1 ∂ n ( σ )
0≤i <k ≤n, (k = j+1)

∑
j
= (−1)i+ j ∂n−1 ∂in (σ) +
0≤ j < i ≤ n

∑
j
+ −(−1)i+ j ∂in−−11 ∂n (σ).
0≤ j < i ≤ n

For 0 ≤ j < i ≤ n:

j
∂n−1 ∂in (σ )(t0 , . . . , tn−2 ) = ∂in σ(t0 , . . . , t j−1 , 0, t j , . . . , tn−2 )
= σ(t0 , . . . , t j−1 , 0, t j , . . . , ti−2 , 0, ti−1 , . . . , tn−2 ),

while

j j
∂in−−11 ∂n (σ )(t0 , . . . , tn−2 ) = ∂n σ(t0 , . . . , ti−2 , 0, ti−1 , . . . , tn−2 )
= σ(t0 , . . . , t j−1 , 0, t j , . . . , ti−2 , 0, ti−1 , . . . , tn−2 ),

j j
thus ∂n−1 ∂in = ∂in−−11 ∂n . So ∂n−1 ◦ ∂n = 0.

Definition. The n-dimensional singular homology group of a topological space X is de-

fined to be the quotient group
ker(∂n )
Hn ( X ) = .
Im(∂n+1 )

Example. Denoting by {pt} a space containing only one point, then Hn ({pt}) = 0, n ≥ 1
∼ Z.
and H0 ({pt}) = hpti =

Proposition. If X is path-connected then H0 ( X ) ' Z, generated by any point of X. In general

H0 ( X ) is generated by one point in each path-connected components of X.

Proof. Let x0 and x1 be two points in X. A continuous path from x0 to x1 gives rise to a
singular 1-simplex σ : ∆1 → X such that σ(0, 1) = x0 and σ (1, 0) = x1 . The boundary of
this singular simplex is ∂σ = x1 − x0 ∈ Im ∂1 . Thus [ x0 ] = [ x1 ] ∈ H0 ( X ) = S0 ( X )/ Im ∂1 .
122

Induced homomorphism
Let X and Y be topological spaces and let f : X → Y be continuous. For any n-singular
simplex σ let f # (σ ) = f ◦ σ, then extend f # linearly, we get a group homomorphism f # :
S n ( X ) → S n (Y ) .

Example. Id∗ = Id.

Lemma. ∂ ◦ f # = f # ◦ ∂. As a consequence f # brings cycles to cycles, boundaries to boundaries.

Proof. Because both f # and ∂ are linear we only need to prove ∂( f # (σ )) = ( f # (∂(σ )) for any
n-singular simplex σ : ∆n → X. We have
!
n n
f # (∂(σ )) = f # ∑ (−1) ∂ (σ) i i
= ∑ (−1)i f ◦ ∂i (σ).
i =0 i =0

On the other side: ∂( f # (σ )) = ∑in=0 (−1)i ∂i ( f ◦ σ ). Notice that

∂i ( f ◦ σ )(t0 , . . . , tn−1 ) = ( f ◦ σ)(t0 , . . . , ti−1 , 0, ti , . . . , tn−1 )

= f (σ (t0 , . . . , ti−1 , 0, ti , . . . , tn−1 ))
= f (∂i σ (t0 , . . . , tn−1 )).

Thus ∂i ( f ◦ σ) = f ◦ ∂i (σ ). From this the result follows.

As a consequence f # induces a group homomorphism:

f ∗ : Hn ( X ) → Hn (Y ).
[c] 7→ [ f # (c)].

Lemma. ( g ◦ f )∗ = g∗ ◦ f ∗ .

Proof. Since the maps involved are linear it is sufficient to check this property on each n-
singular simplex σ:

( g ◦ f )∗ ([σ]) = [( g ◦ f )# (σ)] = [( g ◦ f ) ◦ σ] = [ g ◦ ( f ◦ (σ))] = g∗ ([ f ◦ σ]) =

= g∗ ( f ∗ ([σ])).

A simple application of the above lemma gives us an important result:

Theorem (topological invariance of homology). If f : X → Y is a homeomorphism then

f ∗ : Hn ( X ) → Hn (Y ) is an isomorphism.

Proof. Apply the above lemma to the pair f and f −1 .

We will use, but will not prove, the following stronger result :

Theorem (homotopy invariance of homology). If f : X → Y is a homotopy equivalence then

f ∗ : Hn ( X ) → Hn (Y ) is an isomorphism.
17. SINGULAR HOMOLOGY 123

Mayer-Vietoris sequence
Theorem. Let X be a topological space. Suppose U, V ⊂ X and int(U ) ∪ int(V ) = X. Then there
is an exact chain complex, called the Mayer-Vietoris sequence:

(i∗ ,j∗ ) ψ∗ ∆
··· → Hn (U ∩ V ) → Hn (U ) ⊕ Hn (V ) → Hn (U ∪ V ) → Hn−1 (U ∩ V ) → · · ·
· · · → H0 (U ∪ V ) → 0.

Here i and j are the inclusion maps from U ∩ V to U and V respectively.

The Mayer-Vietoris sequence allows us to study the homology of a space from homolo-
gies of subspaces, in a similar manner to the Van Kampen theorem.
We will use the Mayer-Vietoris sequence to compute the homology of the sphere:

Theorem. For m ≥ 1, 
Z, if n = 0, m
Hn (Sm ) ∼
=
0, otherwise.

Proof. Let U and V be the upper hemisphere and the lower hemisphere slightly enlarged,
for example, U = Sm \ {(0, . . . , 0, −1)} and V = Sm \ {(0, . . . , 0, 1)}. The Mayer-Vietoris
sequence for this pair gives an exact short sequence:

Hn (U ) ⊕ Hn (V ) → Hn (Sm ) → Hn−1 (U ∩ V ) → Hn−1 (U ) ⊕ Hn (V ).

Notice that for m ≥ 1, U and V are contractible, while U ∩ V is homotopic to Sm−1 . We get
for n ≥ 2 a short exact sequence

0 → Hn (Sm ) → Hn−1 (Sm−1 ) → 0.

This implies Hn (Sm ) ∼

= Hn−1 (Sm−1 ) for n ≥ 2. The problem now reduces to computation
of H1 (Sm ).
Consider the exact sequence:

∆ (i∗ ,j∗ )
H1 (U ) ⊕ H1 (V ) → H1 (Sm ) → H0 (U ∩ V ) → H0 (U ) ⊕ H0 (V ).

Since H1 (U ) = H1 (V ) = 0, it follows that ∆ is injective.

For m ≥ 2 a point x ∈ U ∩ V generates H0 (U ∩ V ) as well as H0 (U ) and H0 (V ). There-
fore the maps i∗ and j∗ are injective. This implies Im(∆) = 0. This can happen only when
H1 (Sm ) = 0.
For m = 1 the intersection U ∩ V has two path-connected components. Let x and y
be points in each connected component. If mx + ny ∈ H0 (U ∩ V ) then i∗ (mx + ny) =
j∗ (mx + ny) = mx + nx = (m + n) x. Thus ker(i∗ , j∗ ) = {mx − my = m( x − y) | m ∈ Z}.
This implies H1 (S1 ) ∼
= Im(∆) = ker(i∗ , j∗ ) ∼
= Z.

17.2 Corollary. For n ≥ 2 there cannot be any retraction from the disk D n to its boundary Sn−1 .

The proof is similar to the 2-dimensional case (14.5).

Proof. Suppose there is a retraction r : D n → Sn−1 . Let i : Sn−1 ,→ D n be the inclusion

i r
map. From the diagram Sn−1 → D n → Sn−1 we have r ◦ i = idSn−1 , therefore on the
(n − 1)-dimensional homology groups (r ◦ i )∗ = r∗ ◦ i∗ = id Hn−1 (Sn−1 ) , implying that
124

r∗ : Hn−1 ( D n ) → Hn−1 (Sn−1 ) is onto. But this is not possible for n ≥ 2, since Hn−1 ( D n ) is
trivial while Hn−1 (Sn−1 ) is not.

A proof of this result in differentiable setting is presented in 24.3.

Just as in the case of dimension two (14.6), the Brouwer fixed point theorem follows,
with the same proof:

17.3 Theorem (Brouwer fixed point theorem). Any continuous map from the disk D n to itself
has a fixed point.

By 17.16 and 17.15 we get a more general version:

Theorem (generalized Brouwer fixed point theorem). Any continuous map from a compact
convex subset of a Euclidean space to itself has a fixed point.

For more on singular homology, one can read [Vic94] and [Hat01].

Problems
17.4. Prove 17.1.

17.5. Compute the homology groups of S2 × [0, 1].

17.6. Compute the fundamental group and the homology groups of the Euclidean space R3 minus a
straight line.

17.7. Compute the fundamental group and the homology groups of the Euclidean space R3 minus
two intersecting straight lines.

17.8. Compute the fundamental group and the homology groups of R3 \ S1 .

17.9. Show that if X has k connected components Xi , 1 ≤ i ≤ k, then Sn ( X ) ∼

= ik=1 Sn ( Xi ) and
L

Hn ( X ) ∼
= k Hn ( Xi ).
L
i =1

17.10. Show that if A and B are open, A ∩ B is contractible then Hi ( A ∪ B) ∼

= Hi ( A) ⊕ Hi ( B) for i ≥ 2.
Is this true if i = 0, 1?

17.11. Using problem 17.10, compute the homology groups of S2 ∨ S4 .

17.12. Compute the fundamental group and the homology groups of S2 ∪ {(0, 0, z) | − 1 ≤ z ≤ 1}.

17.13. Let U be an open subset of R2 and a ∈ U. Show that

H1 (U \ { a}) ∼
= H1 (U ) ⊕ H1 (R2 \ { a}).

17.14. Let A be a retract of X via a retraction r : X → A. Let i : A ,→ X be the inclusion map. Show
that the induced homomorphism i∗ : Hn ( A) → Hn ( X ) is injective while r∗ : Hn ( X ) → Hn ( A) is
surjective.

17.15. Show that if every continuous map from X to itself has a fixed point and Y is homeomorphic
to X then every continuous map from Y to itself has a fixed point.

17.16. * Show that:

(a) Any convex compact subset of Rn with empty interior is homeomorphic to the closed interval
D1 .
(b) Any convex compact subset of Rn with non-empty interior is homeomorphic to the disk D n .
(This is a generalization of 10.16.)
(c) The boundary of any convex compact subset of Rn with non-empty interior is homeomorphic
to the sphere Sn−1 .

17.17. Is the Brouwer fixed point theorem correct for open balls? for spheres? for tori?
18. HOMOLOGY OF CELL COMPLEXES 125

18 Homology of cell complexes

In this section we consider homology for a special class of topological spaces: cell com-
plexes, including simplicial complexes.

Degrees of maps on spheres

A continuous map f : Sn → Sn induces a homomorphism f ∗ : Hn (Sn ) → Hn (Sn ). We
know Hn (Sn ) ∼= Z, so there is a generator a such that Hn (Sn ) = h ai. Then f ∗ ( a) = ma for a
certain integer m, called the topological degree of f , denoted by deg f .

Example. If f is the identity map then deg f = 1. If f is the constant map then deg f = 0.

Relative homology groups

Let A be a subspace of X. Viewing each singular simplex in A as a singular simplex in X, we
have a natural inclusion Sn ( A) ,→ Sn ( X ). In this way Sn ( A) is a normal subgroup of Sn ( X ).
The boundary map ∂n induces a homomorphism ∂n : Sn ( X )/Sn ( A) → Sn−1 ( X )/Sn−1 ( A),
giving a chain complex

∂ ∂ n −1
· · · → Sn ( X )/Sn ( A) →n Sn−1 ( X )/Sn−1 ( A) → Sn−2 ( X )/Sn−2 ( A) → · · ·

The homology groups of this chain complex is called the relative homology groups of the
pair ( X, A), denoted by Hn ( X, A).
If f : X → Y is continuous and f ( A) ⊂ B then as before it induces a homomorphism
f ∗ : Hn ( X, A) → Hn (Y, B).

Homology of cell complexes

Let X be a cellular complex. Recall that X n denote the n-dimensional skeleton of X. Sup-
pose that X n is obtained from X n−1 by attaching the n-dimensional disks D1n , D2n , . . . , Dcnn .
Let e1n , e2n , . . . , ecnn be the corresponding cells. Then

cn
= e1n , e2n , . . . , ecnn = { ∑ mi ein | mi ∈ Z}.
Hn ( X n , X n−1 ) ∼


i =1

If cn = 0 then let the group be 0.

Consider the sequence

d n +1 d d n −1
C ( X ) = · · · → Hn ( X n , X n−1 ) →n
Hn−1 ( X n−1 , X n−2 ) → · · ·
d d
· · · →2 H1 ( X 1 , X 0 ) →1 H0 ( X ).

Here the map dn is given by

c n −1
dn (ein ) = ∑ di,j enj −1 ,
j =1

where the integer number di,j is the degree of the following map on spheres:

Sin−1 = ∂Din → X n−1 → X n−1 /X n−2 = S1n−1 ∨ S2n−1 ∨ · · · ∨ Scnn−−11 → Snj −1 .

126

Of course if cn−1 = 0 then dn = 0.

Below is the main tool for computing homology of cellular complexes:

Theorem. The sequence C ( X ) is a chain complex and its homology coincides with the homology of
X.

As an application we get:

Theorem (homology groups of surfaces). The homology groups of a connected compact ori-
entable surface S of genus g ≥ 0 is

Z, if n = 0, 2
Hn (S) ∼
=
Z g if n = 1.

For more on cellular homology one can read [Hat01, p. 137].

Problems
18.1. Using cellular homology compute the homology groups of the following spaces:

(a) The Klein bottle.

(b) S1 ∨ S1 .
(c) S1 ∨ S2 .
(d) S2 ∨ S3 .

18.2. Show that two compact connected surfaces are homeomorphic if and only if their homologies
are isomorphic.
18. HOMOLOGY OF CELL COMPLEXES 127

Other topics

Invariance of dimension
That the Euclidean spaces R2 and R3 are not homeomorphic is not easy. It is a consequence
of the following difficult theorem of L. Brouwer in 1912:

18.3 Theorem (Invariance of domain). If two subsets of the Euclidean Rn are homeomorphic
and one set is open then the other is also open.

This theorem is often proved using Algebraic Topology, see for instance [Mun00, p. 381],
[Vic94, p. 34], [Hat01, p. 126].

Corollary. The Euclidean spaces Rm and Rn are not homeomorphic if m 6= n.

Proof. Suppose that m < n. We can check that the inclusion map Rm → Rn given by
( x1 , x2 , . . . , xm ) 7→ ( x1 , x2 , . . . , xm , 0, . . . , 0) is a homeomorphism onto its image A ⊂ Rn .
Since Rn is open in Rn , if A is homeomorphic to Rn then by Invariance of dimension A is
open in Rn . But A is clearly not open in Rn .

Jordan curve theorem

The following is an important and deep result of plane topology.

Theorem (Jordan curve theorem). A simple, continuous, closed curve separates the plane into
two disconnected regions. More concisely, if C is a subset of the Euclidean plane homeomorphic to
the circle then R2 \ C has two connected components.

Nowadays this theorem is usually proved in a course in Algebraic Topology.

The Poincaré conjecture

One of the most celebrated achievements in topology is the resolution of the Poincaré con-
jecture:

Theorem (Poincaré conjecture). A compact manifold which is homotopic to the sphere is homeo-
morphic to the sphere.

The proof of this statement is the result of a cumulative effort of many mathematicians,
including Stephen Smale (for dimension ≥ 5, early 1960s), Michael Freedman (for dimen-
sion 4, early 1980s), and Grigory Perelman (for dimension 3, early 2000s). For dimension 2
it is elementary, see problem 15.12.

Guide for further study

There are several books presenting aspects of algebraic topology to undergraduate stu-
dents, such as [Cro78], [Ams83], [J84].
The book of Munkres [Mun00], also aims at undergraduate students, has a part on
algebraic topology, but stops before homology.
The book [Vas01] gives a modern overview of many aspects of both algebraic and dif-
ferential topology, aiming at undergraduate students. Although it often provides only
sketches of proofs, it introduces general ideas very well.
128

For cellular homology the book of Hatcher [Hat01] is popular, however it aims at grad-
uate students.
It turns out that our choice of topics is similar to that in the book [Lee11], which aims at
graduate students. The reader can find more details and more advanced treatments there.
Recently algebraic topology has begun to be applied to science and engineering. One
can read about a new field called “Computational Topology” in [EH10].
Part III Differential Topology

129
19. SMOOTH MANIFOLDS 131

Differential Topology studies topology in the smooth category, meaning that all maps
considered are smooth. The main objects are smooth manifolds. For simplicity we only
consider manifolds in Euclidean spaces, following [Mil97].
Some proofs in this part are more technically difficult, already treated in more advanced
sources. The readers may choose to focus on attempting to understand the main ideas of
the arguments.

19 Smooth manifolds
In this chapter we always assume that Rn has the Euclidean topology.
Roughly, a smooth manifold is a space that is locally diffeomorphic to Rn . This allows us
to bring the differential and integral calculus from Rn to manifolds. Smooth manifolds pro-
vide a setting for generalizations of vector calculus of curves and surfaces (see e.g. [Vugt3])
to higher dimensions. Smooth manifolds constitute a subset of the set of topological mani-
folds, introduced in section 10.
We will discuss smooth manifolds embedded in Euclidean spaces. There is a notion of
abstract smooth manifold, but we do not discuss it.

Smooth maps on open subsets of Rn

We summarize here several results about derivatives of functions defined on open sets in
Rn , see for instance [Spi65] or [Lan97] for more details.
Let D ⊂ Rk and let x be an interior point of D. Recall that e1 = (1, 0, 0, . . . , 0),
e2 = (0, 1, 0, . . . , 0), . . . , en = (0, . . . , 0, 1) form the canonical vector basis for Rn . The partial
derivative of the function f : D → R with respect to the ith variable at x is defined to be
the real number
∂f f ( x + hei ) − f ( x )
( x ) = lim .
∂xi h →0 h
This is the rate of change of the value of f compared to the value of its ith variable.
∂2 f
We can take higher order partial derivatives. For example, ∂xi ∂x j is a partial derivative
of order 2. If f is continuous and all partial derivatives of all orders of f exist and are
continuous at x then we say that f is smooth at x. (In common notation in analysis, f is
T∞
smooth means f ∈ C ∞ = k =0 C
k .)

The matrix of all first order partial derivativesof f iscalled the Jacobian matrix of f ,
∂ fi
denoted by J f ( x ) or J ( f )( x ). Precisely J f ( x ) =
∂x ( x ) . When m = 1 the
j 1≤i ≤m, 1≤ j≤n
Jacobian matrix J f ( x ) is the gradient vector
 
∂f ∂f
grad f ( x ) = ∇ f ( x ) = ( x ), . . . , (x) .
∂x1 ∂xn

Let D be an open subset of Rk . Consider a smooth map f : D → Rl . If there is a linear

map f 0 ( x ) : Rk → Rl such that there is a ball B( x, e) ⊂ D and a function r : B( x, e) → Rl
satisfying:
f ( x + h) = f ( x ) + f 0 ( x )(h) + r (h), ∀h ∈ B( x, e)
r (h)
and limh→0 = 0, then f 0 ( x ) is called the derivative of f at x. Thus the derivative is a
|h|
linear approximation of the function:

f ( x + h) ≈ f ( x ) + f 0 ( x )(h).
132

If f is smooth at x then f has derivative at x, and the linear map f 0 ( x ) can be represented
in the canonical linear bases by the Jacobian matrix J f ( x ), i.e. f 0 ( x )(v) = J f ( x ) · v.
We will also denoted f 0 ( x ) by d f ( x ) or d f x . It should be emphasized the derivative is a
linear map, not a number or a matrix.
If v is a unit vector then we can deduce immediately that

f ( x + tv) − f ( x )
d f x (v) = lim .
t →0 t

Thus d f x (v) is the directional derivative of f at x in the direction of v, measuring rate of

change of f in the direction of v at x.
Let U, V, W be open subsets of Rk , Rl , R p respectively, let f : U → V and g : V → W
be smooth maps, and let y = f ( x ). Then we have the chain rule

d( g ◦ f ) x = dgy ◦ d f x .

In other words, the following commutative diagram

?V
f g

U /W
g◦ f

induces the following commutative diagram

>R
l

d fx dgy

Rk / Rp
d( g◦ f ) x

Lemma. Let U and V be open subsets of Rk and Rl respectively. If f : U → V is a diffeomorphism

then the derivative d f x is a linear isomorphism, and k = l. As a corollary, Rk and Rl are not
diffeomorphic if k 6= l.

Smooth manifolds
We extend the notion of smooth function:

Definition. If x is a boundary point of D ⊂ Rk , then f : D → Rl is said to be smooth

at x if f can be extended to a function which is smooth in an open neighborhood in Rk
of x. Precisely, f is smooth at x if there is an open set U ⊂ Rk containing x and function
F : U → Rl such that F is smooth at every point of U and F |U ∩ D = f .

If f is smooth at every point of D then we say that f is smooth on D, or f ∈ C ∞ ( D ).

It follows immediately that compositions of smooth functions are smooth. Notice that
we do not define derivatives for functions which are smooth in this extended sense, since
there is no obvious way to do it yet.
Let X ⊂ Rk and Y ⊂ Rl . Then f : X → Y is a diffeomorphism (vi đồng phôi, hay vi
phôi) if it is bijective and both f and f −1 are smooth. If there is a diffeomorphism from X
to Y then we say that X and Y are diffeomorphic.
19. SMOOTH MANIFOLDS 133

If a function is smooth at a point, it must be continuous at that point, therefore a diffeo-

morphism is a homeomorphism.

Example. Any open ball B( x, r ) in Rn is diffeomorphic to Rn (compare 3.1).

Definition. A subspace M ⊂ Rk is a smooth manifold of dimension m ∈ Z+ if every point

in M has a neighborhood in M which is diffeomorphic to Rm .

Since Rm cannot be diffeomorphic to Rn if m 6= n, a non-empty smooth manifold has a

unique dimension.
A diffeomorphism is a homeomorphism, therefore a smooth manifold is a topological
manifold. In differential topology unless stated otherwise manifolds mean smooth man-
ifolds. By convention, a smooth manifold of dimension 0 is a discrete subspace of a Eu-
clidean space.
The following is a simple but convenient observation, although it seems to be less intu-
itive than our original definition, it is technically more convenient to use:

Proposition. A subspace M ⊂ Rk is a smooth manifold of dimension m if every point in M has an

open neighborhood in M which is diffeomorphic to an open subset of Rm .

Proof. Suppose that (U, φ) is a local coordinate on M where U is a neighborhood of x in

M and φ : U → Rm is a diffeomorphism. There is an open subset U 0 of M such that
x ∈ U 0 ⊂ U. Since φ is a homeomorphism, φ(U 0 ) is an open neighborhood of φ( x ). There
is a ball B(φ( x ), r ) ⊂ φ(U 0 ). Let U 00 = φ−1 ( B(φ( x ), r )). Then U 00 is open in U 0 , so is open
in M. Furthermore φ|U 00 : U 00 → B(φ( x ), r ) is a diffeomorphism.
We have just shown that any point in the smooth manifold has an open neighborhood
diffeomorphic to an open ball in Rm . For the reverse direction, we recall that any open ball
in Rm is diffeomorphic to Rm .

By this result, each point x in a smooth manifold has an open neighborhood U in M and
a diffeomorphism ϕ : U → V where V is an open subset of Rm . The pair (V, ϕ−1 ) is called
a local parametrization at x. The pair (U, ϕ) is called a local coordinate or a chart (tờ bản
đồ) at x. The set of pairs (U, ϕ) at all points is called an atlas (tập bản đồ) for M. We can
think of a manifold as a space which has an atlas.
As a consequence of the definition of manifold, a function f defined on a manifold is
smooth if and only if f ◦ ϕ is smooth for any local parametrization of the manifold.
If M and N are two smooth manifolds in Rk and M ⊂ N then we say that M is a
submanifold of N.

Example. Any open subset of Rm is a smooth manifold of dimension m.

19.1 Proposition. The graph of a smooth function f : D → Rl , where D ⊂ Rk is an open set, is a

smooth manifold of dimension k.

Proof. Let G = {( x, f ( x )) | x ∈ D } ⊂ Rk+l be the graph of f . The map x 7→ ( x, f ( x ))

from D to G is smooth. Its inverse is the projection ( x, y) 7→ x. This projection is the
restriction of the projection given by the same formula from Rk+l to Rk , which is a smooth
map. Therefore D is diffeomorphic to G.

Example (curves and surfaces). The graph of a smooth function y = f ( x ) for x ∈ ( a, b) is

a 1-dimensional smooth manifold, often called a smooth curve.
The graph of a smooth function z = f ( x, y) for ( x, y) ∈ D, where D is an open set in R2
is a 2-dimensional smooth manifold, often called a smooth surface.
134

Example (circle). Let S1 = {( x, y) ∈ R2 | x2 + y2 = 1}. It is covered by four neighborhoods

which are half circles, corresponding to points ( x, y) ∈ S1 such that x > 0, x < 0, y > 0 and
y < 0. Each of these neighborhoods is diffeomorphic to (−1, 1). For example consider the
projection U = {( x, y) ∈ S1 | x > 0} → (−1, 1) given by ( x, y) 7→ y. The map ( x, y) 7→ y
is smooth on R2 , so it is smooth on U. The inverse map y 7→ ( 1 − y2 , y) is smooth on
p

(−1, 1). Therefore the projection is a diffeomorphism.

The Inverse function theorem

Recall the following important result in Analysis (see for instance [Spi65] ):

Theorem (Inverse function theorem). Let f : Rk → Rk be smooth. If det( J f x ) 6= 0 then there

is an open neighborhood U of x and an open neighborhood V of f ( x ) such that f |U : U → V is a
diffeomorphism.
Briefly, if d f x is an isomorphism then f is a diffeomorphism locally at x.

Usually the result is stated for continuously differentiable function (i.e. C1 ), but the
result for smooth functions follows, since the Jacobian matrix of the inverse map is the
inverse matrix of the Jacobian of the original map, and the entries of an inverse matrix can
be obtained from the entries of the original matrix via smooth operations, namely A−1 =
1 ∗ ∗ = (−1)i + j det( A j,i ), and A j,i is obtained from A by omitting the ith row
where Ai,j
det A A ,
and jth column.
Smooth invariance of domain is a simple corollary. (Compare the continuous case 18.3.)

Corollary (smooth invariance of domain). If two subsets of the Euclidean Rk are diffeomorphic
and one set is open then the other set is also open. .

Proof. Let U, V ⊂ Rk and let U be open. Let f : U → V be a diffeomorphism. Let y ∈ V.

There is x ∈ U such that f ( x ) = y. Since f −1 is smooth at y, there is an open subset
W of Rk containing y and a smooth function g : W → Rk such that g|W ∩V = f −1 . Let
U 0 = f −1 (W ∩ V ), then U 0 is open in U, hence in Rk . We have g ◦ ( f |U 0 ) = idU 0 . Taking
derivative at x we get dgy ◦ d f x = idRk . This implies d f x is an isomorphism. By the Inverse
function theorem, there is an open set V 0 of Rk containing y on which f −1 is defined. So
V 0 ⊂ V. Thus V is open.

Proposition. A subspace M of Rk is an m-dimensional smooth manifold if and only if for each

x ∈ M there is an open neighborhood V of x in M, an open set U in Rm , and a bijective map
ϕ : U → V, such that

(a) ϕ is a homeomorphism,

(b) ϕ is smooth,

(c) dϕ(u) has rank m for each u ∈ U.

This is the definition of smooth manifold used in [Mun91, p. 196]. When m = 2 and
k = 3 this is the definition of regular surfaces in [dC76, p. 52].

Proof. The necessary direction: Since ϕ−1 ◦ ϕ = idRm , taking derivative, we get dϕ− 1
ϕ(u)
◦
dϕ(u) = idRm , thus rank dϕ(u) = m.
19. SMOOTH MANIFOLDS 135

The sufficient direction: We need to prove that ϕ−1 is smooth. Let ϕ(u) = x. Since
the Jacobian matrix J ( ϕ)(u) is of rank m, it has m independent rows, corresponding to m
components xi1 , xi2 , . . . , xim . Thus the m × m-matrix
 
∂xi1 ∂xi2 ∂xim
∂u (u) ∂u (u) · · · ∂u (u)


is non-singular. Denote the correspondence u 7→ xi1 (u), xi2 (u), . . . , xim (u) by ψ. Apply
the Inverse function theorem to ψ, there is an open neighborhood U of u in Rm and an open
neighborhood V of xi1 (u), xi2 (u), . . . , xim (u) in Rm such that ψ is a diffeomorphism from

U onto V. We can take U inside the domain of ϕ.

Rk

ϕ (U ) M
x

pr

u
ψ −1 U

V Rm
( x i1 , x i2 , . . . , x i m )

Rm

Let pr be the projection ( x1 , x2 , . . . , xk ) 7→ xi1 , xi2 , . . . , xim . We have that x ∈ ϕ(U ) if
and only if pr ( x ) ∈ V. In other words the following composition

pr ψ −1
ϕ (U ) → V → U


( x1 , x2 , . . . , x k ) 7 → x i1 , x i2 , . . . , x i m 7→ u

gives ϕ−1 = ψ−1 ◦ pr on ϕ(U ).

At this point it is tempting to conclude that ϕ−1 is smooth since ψ−1 ◦ pr is smooth on
pr −1 (V ). However we do need to use the hypothesis that ϕ is a homeomorphism, so that
ϕ(U ) is an open neighborhood of x. Otherwise the argument does not work, as can be seen
from problem 19.9 (the figure 8), looking at the point (0, 0). Now there is an open set W of
Rk such that W ∩ M = ϕ(U ). The set pr −1 (V ) ∩ W is open in Rk and pr −1 (V ) ∩ W ∩ M =

ϕ(U ). On pr −1 (V ) ∩ W the map ψ−1 ◦ pr is defined and is smooth, therefore the restriction
of the function to ϕ(U ), which is ϕ−1 , is smooth.

Example. The graph of the function y = | x | is not a smooth 1-dimensional manifold in R2 .

Indeed, if the graph, denoted by G, is a smooth 1-manifold then there is a parametrization
ϕ(t) = ( x (t), y(t)) of a neighborhood of ϕ(0) = (0, 0) on G. Since ϕ0 (0) = ( x 0 (0), y0 (0)) 6=
0, either x 0 (0) 6= 0 or y0 (0) 6= 0. Repeating the above argument, we find that either there
is a neighborhood of (0, 0) in G which is the graph of a smooth function on x, or there is a
neighborhood of (0, 0) in G which is the graph of a smooth function on y. Obviously the
second possibility cannot happen. For the first possibility, the smooth function must be
136

y = | x |, but this function is not smooth.

Problems
19.2. From our definition, a smooth function f defined on D ⊂ Rk does not necessarily have partial
derivatives defined at boundary points of D. However, show that if D is the closure of an open
subspace of Rk then the partial derivatives of f are defined and are continuous on D. For example, f :
[ a, b] → R is smooth if and only if f has right-derivative at a and left-derivative at b, or equivalently,
f is smooth on an open interval (c, d) containing [ a, b].

19.3. If X and Y are diffeomorphic and X is an m-dimensional manifold then so is Y.

19.4. Show that any open subset of a smooth manifold is a smooth manifold.

19.5. Show that the sphere Sn = {( x1 , x2 , . . . , xn+1 ) ∈ Rn+1 | x12 + x22 + · · · + xn2 +1 = 1} is a smooth
manifold of dimension n, covered by the hemispheres.
Prove that in another way, by using two stereographic projections, one from the North Pole and
one from the South Pole.

19.6. Show that the hyperboloid x2 + y2 − z2 = 1 is a manifold.

19.7. The torus can be obtained by rotating around the z axis a circle on the xOz plane not intersecting
the z axis (see 9.6). Show that the torus is a smooth manifold.

19.8. Consider the union of the curve y = x3 sin 1x , x 6= 0 and the point (0, 0). Is it a smooth manifold?

19.9. Is the trace of the path γ(t) = (sin(2t), sin(t)), t ∈ (0, 2π ) (the figure 8) a smooth manifold?

19.10. Consider the union of the curve y = sin 1x , x > 0 and the segment {(0, y)| − 1 ≤ y ≤ 1} (the
Topologist’s sine curve, see section 4.4). Is it a smooth manifold?

19.11. Is the surface x2 + y2 − z2 = 0 a manifold?

19.12. A simple closed regular path is a map γ : [ a, b] → Rm such that γ is injective on [ a, b), γ is
smooth, γ(k) ( a) = γ(k) (b) for all integer k ≥ 0, and γ0 (t) 6= 0 for all t ∈ [ a, b]. Show that the trace of a
simple closed regular path is a smooth 1-dimensional manifold.

19.13. Show that the trefoil knot (see 9.33) is a smooth 1-dimensional manifold (in fact it is diffeomor-
phic to the circle S1 , but this is more difficult).

19.14. Show that a connected smooth manifold is also path-connected.

19.15. Show that on a path-connected smooth manifold any two points can be connected by a piece-
wise smooth path.

19.16. Let M be a smooth manifold and let ϕ be a local parametrization of a neighborhood of x ∈ M.

Show that a map f : M → Rl is smooth at x if and only if f ◦ ϕ is smooth at ϕ−1 (u).

19.17. * Show that our definition of smooth manifold coincides with the definitions in [Spi65, p. 109]
.
20. TANGENT SPACES AND DERIVATIVES 137

20 Tangent spaces and derivatives

Tangent spaces of manifolds

To motivate the definition of tangent spaces of manifolds we recall the notion of tangent
spaces of surfaces. Consider a parametrized surface in R3 given by ϕ(u, v) = ( x (u, v), y(u, v), z(u, v)).
Consider a point ϕ(u0 , v0 ) on the surface. Near to (u0 , v0 ) if we fix v = v0 and only allow u
to change then we get a parametrized path ϕ(u, v0 ) passing through ϕ(u0 , v0 ). The veloc-
ity vector of the path ϕ(u, v0 ) is a “tangent vector” to the path at the point ϕ(u0 , v0 ), and
∂ϕ
is given by the partial derivative with respect to u, that is, ∂u ( u0 , v0 ). Similarly we have
∂ϕ
another “tangent vector” ∂v ( u0 , v0 ). Then the “tangent space” of the surface at ϕ(u0 , v0 ) is
the plane spanned by the above two tangent vectors.

ϕ ( u0 , v0 )
ϕ

y
( u0 , v0 )

u
x

We can think of a manifold as a high dimensional surface. Therefore our definition of

tangent space of manifold is a natural generalization.

Definition. Let M be an m-dimensional manifold in Rk . Let x ∈ M and let ϕ be a parametriza-

tion of a neighborhood of x, and x = ϕ(u). We define the tangent space of M at x, denoted
∂ϕ
by TMx (or Tx M, or T ( M )( x )), to be the vector space spanned by ∂ui ( u ), 1 ≤ i ≤ m. Each
element of the tangent space is called a tangent vector.

Since
∂ϕ
= dϕ(u)(ei ), we get TMx = dϕu (Rm ). It means the derivative of the local
∂ui ( u )
parametrization brings the Euclidean space onto the tangent space.

dϕu (γ0 (t))

Rm ϕ◦γ
x
γ0 (t) ϕ

t R
138

Here is another interpretation of tangent vector. A smooth path in Rm is a smooth map

γ from an open interval of R to Rm . The vector γ0 (t) ∈ Rk is called the velocity vector of the
path at t. Each vector in Rm is the velocity vector of a path in Rm . The local parametrization
ϕ then brings this path to ϕ ◦ γ, a smooth path on M. Since ◦ γ)(t) = dϕ(u)(γ0 (t)),
d
dt ( ϕ
the velocity vector of ϕ ◦ γ is a tangent vector of M. Conversely, every tangent vector of M
at x is the velocity vector of a smooth path in M at x, see problem 20.5.
Example. Consider a surface z = f ( x, y) in R3 . Take the parametrization ϕ( x, y) = ( x, y, f ( x, y)).
The tangent plane at ( x, y, f ( x, y)) consists of the linear combinations of the vectors ϕ x ( x, y) =
(1, 0, f x ( x, y)) and ϕy ( x, y) = (0, 1, f y ( x, y)).
Example. Consider the circle S1 . Let ( x (t), y(t)) be any path on S1 . The tangent space
of S1 at ( x, y) is spanned by the velocity vector ( x 0 (t), y0 (t)) if this vector is not 0. Since
x (t)2 + y(t)2 = 1, differentiating both sides with respect to t we get x (t) x 0 (t) + y(t)y0 (t) =
0, or in other words ( x 0 (t), y0 (t)) is perpendicular to ( x (t), y(t)). Thus the tangent space is
the line perpendicular to the radius.
Lemma. The tangent space does not depend on the choice of parametrization.
Proof. Consider the following diagram, where U, U 0 are open, ϕ and ϕ0 are parametriza-
tions of open neighborhood of x ∈ M.

>M`
ϕ ϕ0

U / U0
ϕ0−1 ◦ ϕ

Notice that the map ϕ0−1 ◦ ϕ is to be understood as follows. We have that ϕ(U ) ∩ ϕ0 (U 0 )
is a neighborhood of x ∈ M. Restricting to ϕ−1 ( ϕ(U ) ∩ ϕ0 (U 0 )), the map ϕ0−1 ◦ ϕ is well-
defined, and is a diffeomorphism.
The above diagram gives us, with any v ∈ Rm :

dϕu (v) = dϕ0ϕ0−1 ◦ ϕ(u) d( ϕ0−1 ◦ ϕ)u (v) .

Thus any tangent vector with respect to the parametrization ϕ is also a tangent vector with
respect to the parametrization ϕ0 . We conclude that the tangent space does not depend on
the choice of parametrization.

The above statement can also be seen from the interpretation of tangent vectors as ve-
locity vectors of smooth paths.
Proposition. If M is an m-dimensional manifold then the tangent space TMx is an m-dimensional
linear space.
Proof. Since a parametrization ϕ is a diffeomorphism, there is a smooth map F from an
open set in Rk to Rm such that F ◦ ϕ = Id. So dFϕ(u) ◦ dϕu = IdRm . This implies that dϕu is
an injective linear map, thus the image of dϕu is an m-dimensional vector space.

Derivatives of maps on manifolds

Let M ⊂ Rk and N ⊂ Rl be manifolds of dimensions m and n respectively. Let f : M → N
be smooth. Let x ∈ M. There is a neighborhood W of x in Rk and a smooth extension F of
f to W. The derivative of f is defined to be the restriction of the derivative of F. Precisely:
20. TANGENT SPACES AND DERIVATIVES 139

Definition. The derivative of f at x is defined to be d f x = dFx | TMx , i.e.

d f x : TMx → TN f (x)
h 7→ d f x (h) = dFx (h).

We need to show that the derivative is well-defined:

Lemma. d f x (h) ∈ TN f ( x) and does not depend on the choice of F.

Thus, although as noted in the previous section a smooth map defined on a general
subset of Rk may not have derivative, on a manifold the derivative can be defined.

Proof. Consider the following commutative diagram

W ∩O M
F /
; NO
f ◦ϕ
ϕ ψ

U /V
ψ −1 ◦ f ◦ ϕ

Let us explain this diagram. Assume that ϕ(u) = x, ψ(v) = f ( x ), h = dϕu (w). Take a
parametrization ψ of a neighborhood of f ( x ). Then f −1 (ψ(V )) is an open neighborhood of
x in M. We can find an open set W in Rk such that (W ∩ M ) ⊂ f −1 (ψ(V )) and W ∩ M is
an open neighborhood of x in M parametrized by ϕ, and f has an extension to a smooth
function F on W.
The diagram induces two ways to write d f x . For any w ∈ Rm :


d f x (dϕu (w)) = dFx dϕu (w) = d( F ◦ ϕ)u (w) = d( f ◦ ϕ)u (w),

which shows that d f x does not depend on the choice of F, and furthermore

d f x (dϕu (w)) = d( f ◦ ϕ)u (w) = d(ψ ◦ (ψ−1 ◦ f ◦ ϕ))u (w) = dψv d(ψ−1 ◦ f ◦ ϕ)u (w) ,

which shows that d f x has value in TN f ( x) .

The above statement can also be seen in the following way. A tangent vector at x ∈ M
is the velocity γ0 (t) of a path γ with γ(t) = x. The map f brings this path to N, giving the
path f ◦ γ. The velocity vector of this new path at f ( x ) is

( f ◦ γ)0 (t) = ( F ◦ γ)0 (t) = dFx (γ0 (t)) = d f x (γ0 (t)).

Thus the derivative brings the velocity vector of a path in to the velocity of the corre-
sponding path. This interpretation also gives another proof that d f x is well-defined.

Example. • If M is a smooth manifold and Id : M → M is the identity map then

dIdx = IdTMx for any x ∈ M.

• If N is another smooth manifold and f : M → N is a constant map then d f x = 0.

Proposition (chain rule). If f : M → N and g : N → P are smooth functions between manifolds,

then
d( g ◦ f ) x = dg f ( x) ◦ d f x .
140

M N
γ0 (t) γ f
d f x (γ0 (t))
x
f ◦γ
f (x)

Proof. There is an open neighborhood V of y in Rl and a smooth extension G of g to V.

There is an open neighborhood U of x in Rk such that U ⊂ F −1 (V ) and there is a smooth
extension F of f U. Then d( g ◦ f ) x = d( G ◦ F ) x | TMx = (dGy ◦ dFx )| TMx = dGy | TNy ◦
dFx | TMx = dgy ◦ d f x .

Proposition. If f : M → N is a diffeomorphism then d f x : TMx → TN f ( x) is a linear isomor-

phism. In particular the dimensions of the two manifolds are same.

Proof. Let m = dim M and n = dim N. Since d f x ◦ d f f−(1x) = IdTN f (x) and d f f−(1x) ◦ d f x =
IdTMx we deduce, via the rank of d f x that m ≥ n. Doing the same with d f f−(1x) we get m ≤ n,
hence m = n. From that d f x must be a linear isomorphism.

Problems
20.1. Calculate the tangent spaces of Sn .

20.2. Calculate the tangent spaces of the hyperboloid x2 + y2 − z2 = 1.

20.3. Let V be an m-dimensional vector subspace of Rk , let p ∈ Rk , and let M = p + V, called an

affine subspace of Rk .

(a) Show that M is an m-dimensional smooth manifold.

(b) Find the tangent spaces of M.

(c) Similarly let W be an n-dimensional vector subspace of Rl , let q ∈ Rk , and let N = q + W. Let
T : V → W be a linear map. Define T̃ : M → N by T̃ ( p + v) = q + T (v). Find the derivatives
of T̃.

20.4. Show that if M is a submanifold of N then TMx is a linear subspace of TNx .

20.5. Show that the velocity vector at x of a smooth path in M going through x is a tangent vector of
M at x.

20.6 (Cartesian products of manifolds). If X ⊂ Rk and Y ⊂ Rl are manifolds then X × Y ⊂ Rk+l is

also a manifold. Furthermore T ( X × Y )( x,y) = TXx × TYy . As an application, find the tangent space
of the cylinder S1 × R.

20.7. Calculate the derivative of the maps:

(a) f : (0, 2π ) → S1 , f (t) = (cos t, sin t).

(b) f : S1 → R, f ( x, y) = y.
20. TANGENT SPACES AND DERIVATIVES 141

20.8. Calculate the derivative of the map f : S1 → S1 , f (z) = zn , where we are writing a point
z = ( x, y) ∈ R2 as a complex number z = x + iy ∈ C, and n ∈ Z. Show that the derivative d f z brings
any tangent vector of S1 to a tangent vector of S1 of n-times the length.

20.9. Let M ⊂ Rk be a closed subset and a smooth manifold and let p ∈ Rk .

(a) Show that there exists q ∈ M such that d( p, q) = d( p, M) = inf{d( p, x ) | x ∈ M} (see 5.14).
(b) Show that ( p − q) ⊥ TMq (in the Euclidean inner product of Rk ).
142

21 Regular values
Let f : M → N be smooth. A point in M is called a critical point (điểm dừng, điểm tới
hạn) of f if the derivative of f at that point is not surjective. Otherwise the point is called a
regular point (điểm thường, điểm chính qui) of f .

Example. Let U be an open set in Rn and let f : U → R be smooth. Then x ∈ U is a critical

point of f if and only if ∇ f ( x ) = 0.

A point in N is called a critical value of f if it is the value of f at a critical point.

Otherwise the point is called a regular value of f .
Thus y is a critical value of f if and only if f −1 (y) contains a critical point. In particular,
if f −1 ( y ) = ∅ then y is considered a regular value in this convention.

Example. If f : M → N where dim( M ) < dim( N ) then every x ∈ M is a critical point and
every y ∈ f ( M ) is a critical value of f .

The Implicit function theorem

Theorem (Implicit function theorem). Suppose that

f : Rm × Rn → Rn
( x, y) 7→ f ( x, y)

∂f
is smooth and the n × n-matrix ∂y ( x0 , y0 ) is non-singular then there is a neighborhood U × V of
( x0 , y0 ) such that for each x ∈ U there is a unique g( x ) ∈ V satisfying f ( x, g( x )) = 0, and the
function g is smooth.

Proof. Set F ( x, y) = ( x, f ( x, y)) then apply the Inverse function theorem to F.

The conclusion of the theorem implies that on U × V the implicit equation f ( x, y) = 0

has a unique solution y = g( x ). It also implies that {( x, y) ∈ U × V | f ( x, y) = 0} =
{( x, g( x )) | x ∈ U }. In other words, on the open neighborhood U × V of ( x0 , y0 ) the level
set f −1 (0) is the graph of a smooth function.
 
∂f ∂f
R ∂x , ∂y ( x0 , y0 )

V
y = g( x )
y0

f ( x, y) = 0

x0 R
U
21. REGULAR VALUES 143

21.1 Corollary. Let f : Rm+n → Rn be smooth and f ( x ) = y. If d f x is surjective then there is a

an open neighborhood of x in the level set f −1 (y) which is diffeomorphic to an open subset of Rm .

Proof. All we need is a permutation of variables to bring the function to the form used by
the Implicit function theorem. Since d f x is onto, the Jacobian matrix J ( f )( x ) is of rank
n, therefore it has n independent columns, corresponding to n variables xi1 , xi2 , . . . , xin .
Let σ be any permutation of {1, 2, . . . , (m + n)} such that σ(m + 1) = i1 , σ(m + 2) =
i2 , . . . , σ (m + n) = in . Let ϕ : Rm+n → Rm+n be the map that permutes the variables
as ( x1 , x2 , . . . , xm+n ) 7→ ( xσ(1) , xσ(2) , . . . , xσ(m+n) ). Consider the function f ◦ ϕ. By the chain
rule J ( f ◦ ϕ)( ϕ−1 ( x )) = J f ( x ) · J ϕ( ϕ−1 ( x )). Computing J ( ϕ)( ϕ−1 ( x )) it can be verified
that the last n columns of the matrix J ( f ◦ ϕ)( ϕ−1 ( x )) are the previous n independent
columns of J ( f )( x ).
Now we can apply the Implicit function theorem to f ◦ ϕ. Write (u0 , v0 ) = ϕ−1 ( x ) ∈
f ◦ ϕ)
Rm × Rn . Then ∂( ∂v (u0 , v0 ) is non-singular. There is an open neighborhood U of u0 in
R and an open neighborhood V of v0 in Rn such that
m

(U × V ) ∩ ( f ◦ ϕ)−1 (y) = {(u, h(u)) | u ∈ U }.

The latter set is the graph of the smooth function h on U, therefore it is diffeomorphic to U
(see 19.1). On the other hand the former set is (U × V ) ∩ ϕ−1 ( f −1 (y)) = ϕ−1 ϕ(U × V ) ∩ f −1 (y) .

Since ϕ is a diffeomorphism, we can conclude that ϕ(U × V ) ∩ f −1 (y) is diffeomorphic to

U.

Preimage of a regular value

21.2 Proposition. If dim( M) = dim( N ) and y is a regular value of f then f −1 (y) is a discrete
set. In other words, f −1 (y) is a zero dimensional manifold. Furthermore if M is compact then
f −1 (y) is a finite set.

Proof. If x ∈ f −1 (y) then there is a neighborhood of x on which f is a bijection (a conse-

quence of the Inverse function theorem, 21.16). That neighborhood contains no other point
in f −1 (y). Thus f −1 (y) is a discrete set, i.e. each point has a neighborhood containing no
other points.
If M is compact then the set f −1 (y) is compact. A discrete compact space must be finite
(6.5).

The following theorem is the Implicit function theorem for manifolds.

21.3 Theorem (level set at a regular value is a manifold). If y is a regular value of f : M → N

then f −1 (y) is a manifold of dimension dim( M ) − dim( N ).

Proof. Let m = dim( M) and n = dim( N ). The case m = n is already considered in 21.2.
Now we assume m > n. Let x0 ∈ f −1 (y0 ). Consider the diagram

f
MO /N
O
ϕ ψ

O /W
g

where g = ψ−1 ◦ f ◦ ϕ and ψ(w0 ) = y0 . Since d f x0 is onto, dg ϕ−1 ( x0 ) is also onto. Denote
ϕ−1 ( x0 ) = (u0 , v0 ) ∈ Rm−n × Rn . Applying 21.1 to g, there is an open neighborhood U of
144

u0 in Rm−n and an open neighborhood V of v0 in Rn such that U × V is contained in O and

(U × V ) ∩ g−1 (w0 ) is diffeomorphic to U. Now
   
ϕ (U × V ) ∩ g−1 (w0 ) = ϕ(U × V ) ∩ ϕ( g−1 (w0 ))
 
= ϕ(U × V ) ∩ ϕ( ϕ−1 ( f −1 (ψ(w0 ))))

= ϕ (U × V ) ∩ f −1 ( y ) .

Thus ϕ(U × V ) ∩ f −1 (y), which is an open neighborhood of x0 in f −1 (y), is diffeomorphic

to U. This ends the proof.

Example. To be able to follow the proof more easily the reader can try to work it out for an
example, such as the case where M is the graph of the function z = x2 + y2 , and f is the
height function f (( x, y, z)) = z defined on M.

Example. The n-sphere Sn is a subset of Rn+1 determined by the implicit equation ∑in=+11 xi2 =
1. Since 1 is a regular value of the function f ( x1 , x2 , . . . , xn+1 ) = ∑in=+11 xi2 we conclude that
Sn is a manifold of dimension n.

Lie groups
2
The set Mn (R) of n × n matrices over R can be identified with the Euclidean manifold Rn .
Consider the map det : Mn (R) → R. Let A = [ ai,j ] ∈ Mn (R). Since

det( A) = ∑ (−1)σ a1,σ(1) a2,σ(2) · · · an,σ(n) = ∑(−1)i+ j ai,j det( Ai,j )

σ ∈ Sn j

we can see that det is a smooth function.

Let us find the critical points of det. A critical point is a matrix A = [ ai,j ] at which
∂ det
= (−1)i+ j det( Ai,j ) = 0 for all i, j. In particular, det( A) = 0. So 0 is the only critical
∂ai,j ( A )
value of det.
Therefore SLn (R) = det−1 (1) is a manifold of dimension n2 − 1.
Furthermore we note that the group multiplication in SLn (R) is a smooth map from
SLn (R) × SLn (R) to SLn (R). The inverse operation is a smooth map from SLn (R) to itself.
We then say that SLn (R) is a Lie group.

Definition. A Lie group is a smooth manifold which is also a group, for which the group
operations are compatible with the smooth structure, namely the group multiplication and
inversion are smooth.

Let O(n) be the group of orthogonal n × n matrices, the group of linear transformation
of Rn that preserves distances.

Proposition. The orthogonal group O(n) is a Lie group.

Proof. Let S(n) be the set of symmetric n × n matrices. This is clearly a manifold of dimen-
n2 + n
sion 2 .
Consider the smooth map f : M (n) → S(n), f ( A) = AAt . We have O(n) = f −1 ( I ). We
will show that I is a regular value of f .
We compute the derivative of f at A ∈ f −1 ( I ):

f ( A + tB) − f ( A)
d f A ( B) = lim = BAt + ABt .
t →0 t
21. REGULAR VALUES 145

We note that the tangents spaces of M(n) and S(n) are themselves. To check whether
d f A is onto for A ∈ O(n), we need to check that given C ∈ S(n) there is a B ∈ M (n) such
that C = BAt + ABt . We can write C = 21 C + 12 C, and the equation 12 C = BAt will give a
solution B = 12 CA, which is indeed a solution to the original equation.

Problems
21.4. Let f : R2 → R, f ( x, y) = x2 − y2 . Show that if a 6= 0 then f −1 ( a) is a 1-dimensional manifold,
but f −1 (0) is not. Show that if a and b are both positive or both negative then f −1 ( a) and f −1 (b) are
diffeomorphic.

21.5. Let f : R3 → R, f ( x, y) = x2 + y2 − z2 . Show that if a 6= 0 then f −1 ( a) is a 2-dimensional

manifold, but f −1 (0) is not. Show that if a and b are both positive or both negative then f −1 ( a) and
f −1 (b) are diffeomorphic.

21.6. Show that the equation x6 + y4 + z2 = 1 determine a manifold in R3 .

21.7. Is the intersection of the two surfaces z = x2 + y2 and z = 1 − x2 − y a manifold?

21.8. Show that the height function ( x, y, z) 7→ z on the sphere S2 has exactly two critical points.

21.9. Show that if f achieves local extremum at x then x is a critical point of f .

21.10. Show that a smooth function on a compact manifold must have at least two critical points.

21.11. Let dim( M) = dim( N ), M be compact and S be the set of all regular values of f : M → N. For
y ∈ S, let | f −1 (y)| be the number of elements of f −1 (y). Show that the map

S → N
y 7→ | f −1 (y)|.

is locally constant. In other words, each regular value has a neighborhood where the number of
preimages of regular values is constant.

21.12. Let M be a compact manifold and let f : M → R be smooth. Show that the set of regular
values of f is open.

1
21.13. Find the regular values of the function f ( x, y, z) = [4x2 (1 − x2 ) − y2 ]2 + z2 − 4 and draw a
corresponding level set.

21.14. Find a counter-example to show that 21.2 is not correct if regular value is replaced by critical
value.

21.15. X If f : M → N is smooth, y is a regular of f , and x ∈ f −1 (y), then ker d f x = Tx f −1 (y).

21.16 (inverse function theorem for manifolds). Let M and N be two manifolds of the same dimen-
sions, and let f : M → N be smooth. Show that if x is a regular point of f then there is a neighborhood
in M of x on which f is a diffeomorphism onto its image.

21.17. Use regular value to show that the torus T 2 is a smooth manifold.

21.18. X Let M be a smooth m-dimensional manifold and suppose that ϕ : Rm → M is injective,

smooth, and dϕ(u) has rank m for each u ∈ Rm . Show that ϕ is a local parametrization of M.
For example, once we know that the torus is a smooth manifold, we can deduce that map ψ(φ, θ ) =
((b + a cos θ ) cos φ, (b + a cos θ ) sin φ, a sin θ ) with 0 < θ < 2π and 0 < φ < 2π (see 9.6) is indeed a
local parametrization of the torus, just by directly checking that this map is injective and its Jacobian
matrix is non-singular. There is no need to write down the inverse map of ψ then check that it is
smooth.
146

21.19. Show that S1 is a Lie group.

21.20. Show that the set of all invertible n × n-matrices GL(n; R) is a Lie group and find its dimension.

21.21. In this problem we find the tangent spaces of SLn (R).

(a) Check that the derivative of the determinant map det : Mn (R) → R is represented by a gradi-
ent vector whose (i, j)-entry is (−1)i+ j det( Ai,j ).
(b) Determine the tangent space of SLn (R) at A ∈ SLn (R).
(c) Show that the tangent space of SLn (R) at the identity matrix is the set of all n × n matrices with
zero traces.
22. CRITICAL POINTS AND THE MORSE LEMMA 147

22 Critical points and the Morse lemma

Partial derivatives

Let M ⊂ Rk be an m-dimensional smooth manifold and let f : M → R be smooth. Let U

be an open neighborhood in M parametrized by ϕ. For each x = ϕ(u) ∈ U we define the
first partial derivative:
 
∂
f :U→R
∂xi
 
∂ ∂
x = ϕ(u) 7→ f (x) = ( f ◦ ϕ)(u).
∂xi ∂ui
 
This definition depends on local coordinates. By this definition, ∂
∂xi f ◦ ϕ is smooth,
 
therefore ∂x∂ f is smooth (see 19.16).
i

Example. If M = Rm and ϕ = id this is the usual partial derivative.

 
∂
To understand f ( x ) better, we can think that the parametrization ϕ brings the
∂xi
 
coordinate system of Rm to the neighborhood U, then ∂x∂ f ( x ) is the rate of change of f
i
when the variable x changes along the path in U which is the composition of the path tei
along the ith axis of Rm with ϕ.
We can write
 
∂ ∂
f (x) = ( f ◦ ϕ)(u) = d( f ◦ ϕ)(u)(ei )
∂xi ∂ui
= (d f ( x ) ◦ dϕ(u)) (ei ) = d f ( x )(dϕ(u)(ei )).
 
∂
Thus ∂xi f ( x ) is the value of the derivative map d f ( x ) at the image of the unit vector ei
of Rm .
∂ϕ
In the local parametrization by ϕ the vectors dϕ(u)(ei ) = ∂ui ( u ) , 1 ≤ i ≤ m constitutes
a vector basis for TMx . In this basis the linear map d f ( x ) is represented by a 1 × m-matrix
whose coordinates are
∂f
d f x (dϕ(u)(ei )) = ( x ).
∂xi
 
∂f ∂f ∂f
Thus in that basis we have the familiar formula [d f ( x )] = ∂x , ∂x , . . . , ∂xm . This formula
1 1
depends on local coordinates.

Gradient vector

The tangent space TMx is a linear subspace of the ambient space Rk , therefore it inherits
the Euclidean inner product from Rk . In this inner product space the linear map d f x :
TMx → R is represented by an element in TMx which we called the gradient vector ∇ f ( x )
or grad f ( x ). This vector is determined by the property h∇ f ( x ), vi = d f x (v) for all v ∈
TMx . Notice that the gradient vector field ∇ f is defined on the whole manifold, and does
not depending on local coordinates.
 
∂ϕ ∂ϕ
Let us see how ∇ f is written in local coordinates. Take the linear basis ∂u1 ( u ), . . . , ∂um ( u )
148
D E
. Letting ∇ f ( ϕ(u)) = ∑im=1 ai (u) ∂u (u), we have
∂ϕ ∂ϕ ∂ϕ
of TMx . Let gij = ∂ui , ∂ui i

* +
  m m
∂ϕ ∂ϕ ∂ϕ
∇ f ( ϕ(u)), (u) = ∑ a j (u) , (u) = ∑ a j (u) gji (u)
∂ui j =1
∂u j ∂ui j =1
 
∂ϕ ∂f
= d f (x) (u) = ( x ).
∂ui ∂xi

So we get a matrix equation

 t
∂f
gij i,j · ( ai )it


= ,
∂xi i

thus

 t

−1 ∂f
( ai )t i = gij i,j
· .
∂xi i

In other words in this linear basis we have

 t
t

−1 ∂f ∂f ∂f
[∇ f ] = gij i,j
· , ,..., .
∂x1 ∂x1 ∂xm

These formulas show that the gradient is a smooth function on the manifold.
We have a simple observation:

Proposition. Let f : M → R be smooth and let x ∈ M.

(a) The point x is a critical point of f if and only if all partial derivatives are zero at x, that is
∂f
∂xi ( x ) = 0, ∀i.

(b) The point x is a critical point of f if and only if the gradient vector is zero at x, that is
∇ f ( x ) = 0.

Proposition. At a local extremum point the gradient vector must be zero.

Proof. Suppose that f has a local maximum at x. There is an open neighborhood U of x

in M parametrized by ϕ, ϕ(u) = x, and f ( x ) ≥ f (y), ∀y ∈ U. For each 1 ≤ i ≤ m, the
dg
function g(t) = f ( ϕ(u + tei )) has a local maximum at t = 0. Therefore dt (0) = 0. From the
chain rule
dg ∂f
(0) = d f ( ϕ(u))(dϕ(u)(ei )) = d f ( x )(dϕ(u)(ei )) = ( x ).
dt ∂xi
The result follows.

Second derivatives
∂
Since ∂xi f is a smooth function on U, we can take its partial derivatives. Thus we define the
second partial derivatives:
!
∂2 f ∂ ∂
(x) = f ( x ).
∂xi ∂x j ∂xi ∂x j

In other words,
22. CRITICAL POINTS AND THE MORSE LEMMA 149

! !
∂2 f ∂ ∂ ∂ ∂f
( ϕ(u)) = f ( ϕ(u)) = ◦ϕ (u)
∂xi ∂x j ∂xi ∂x j ∂ui ∂x j
!
∂ ∂ ∂2
= ( f ◦ ϕ) (u) = ( f ◦ ϕ)(u).
∂ui ∂u j ∂ui ∂u j

Non-degenerate critical points

Consider the Hessian matrix of second partial derivatives:
!
∂2 f
H f (x) = (x) .
∂xi ∂x j
1≤i,j≤m

If this matrix is non-singular then we say that x is a non-degenerate critical point of f .

22.1 Lemma. The non-degeneracy of a critical point does not depend on choices of local coordinates.

Proof. We can see that the problem is reduced to the case of functions on Rm . If f : Rm → R
and ϕ is a change of variables (i.e. a diffeomorphism) of Rm then we have

∂ ∂f ∂ϕk
∂ui
( f ◦ ϕ)(u) = ∑ ∂xk (x) · ∂ui
( u ).
k

Then
" ! #
∂2 ∂2 f ∂ϕ ∂ϕ ∂f ∂2 ϕ
∂u j ∂ui
( f ◦ ϕ)(u) = ∑ ∑ ∂xl ∂xk
( x ) · l (u)
∂u j
· k (u) +
∂ui ∂xk
(x) ·
∂u j ∂ui
(u)
k l
2
∂ f ∂ϕ ∂ϕk
= ∑ ∂xl ∂xk (x) · ∂u jl (u) · ∂ui
( u ).
k,l

In other words
H ( f ◦ ϕ)(u) = J ϕ(u)t · H f ( ϕ(u)) · J ϕ(u). (22.2)

This formula immediately gives us the conclusion.

Morse lemma
Theorem (Morse’s lemma). Suppose that f : M → R is smooth and p is a non-degenerate
critical point of f . There is a local coordinate ϕ in a neighborhood of p such that ϕ( p) = 0 and in
that neighborhood

f ( x ) = f ( p) − ϕ( x )21 − ϕ( x )22 − · · · − ϕ( x )2k + ϕ( x )2k+1 + ϕ( x )2k+2 + · · · + ϕ( x )2m .

In other words, in a neighborhood of 0,

( f ◦ ϕ−1 )(u) = ( f ◦ ϕ−1 )(0) − u21 − u22 − · · · − u2k + u2k+1 + u2k+2 + · · · + u2m .

It is customary to abuse notations by determining a point x on the manifold by its local coordinates
ϕ( x )i , so that we can write
150

f ( x ) = f ( p) − x12 − x22 − · · · − xk2 + xk2+1 + xk2+2 + · · · + xm

2
.

The number k does not depend on the choice of such local coordinates and is called the index of the
non-degenerate critical point p.

Example. Non-degenerate critical points of indexes 0 are local minima, ones of indexes m
are local maxima, and ones of indexes larger than 0 and less than m are neither minima nor
maxima.

Proof. Since we only need to prove the formula for f ◦ ϕ−1 , we only need to work in Rm .
First we show that we can write
m
f ( x ) = f (0) + ∑ xi x j hi,j ( x )
i,j=1

where hi,j = h j,i . We have

Z 1
d
f ( x ) = f (0) + f (tx ) dt
0 dt
m Z 1
 
∂f
= f (0) + ∑ (tx ) xi dt
i =1 0
∂xi
m Z 1 
∂f
= f (0) + ∑ x i (tx ) dt
i =1 0 ∂xi
m
= f ( 0 ) + ∑ x i gi ( x ) ,
i =1

R1 ∂

where the functions gi ( x ) = 0 ∂xi f (tx ) dt are smooth (see for example [Lan97, p. 276]).
∂f
Notice that gi (0) = = 0. Apply the above construction once again to gi we obtain
∂xi (0)
smooth functions gi,j such that

m
f ( x ) = f (0) + ∑ xi x j gi,j ( x ).
i,j=1

m
Set hi,j = ( gi,j + g j,i )/2 then hi,j = h j,i and f ( x ) = f (0) + ∑i,j =1 xi x j hi,j ( x ).
We can obtain the same expression if we use Taylor expansion, which might be already
familiar to the reader, so we will give a presentation. Let g(t) = f (tx ), t ∈ R. Applying the
Taylor expansion to g we get

1 0 1 1
g (1) = g (0) + g (0) + g00 (0) + g000 (c),
1! 2! 3!

where c ∈ (0, 1). Notice that

m
∂f
g0 (t) = ∇ f (tx ) · x = ∑ ∂xi (tx)xi ,
i =1
!
m m m m
∂2 f ∂2 f
 
d ∂f
g (t) = ∑
00
(tx ) xi = ∑ ∑ ∂x j ∂xi (tx)x j xi = ∑ (tx ) xi x j ,
i =1
dt ∂xi i =1 j =1 i,j=1
∂xi ∂x j

m
∂3 f
g000 (t) = ∑ ∂xi ∂x j ∂xk (tx)xi x j xk .
i,j,k =1
22. CRITICAL POINTS AND THE MORSE LEMMA 151

Plugin we get
!
m
1 ∂2 f 1 m ∂3 f
f ( x ) = f (0) + ∑ · (0) 1 + ∑ (c) xk xi x j .
i,j=1
2 ∂xi ∂x j 3 k=1 ∂xi ∂x j ∂xk

Let hi,j be the coefficient of xi x j .

2
1 ∂ f
The rest of the proof is completing the square. Notice that hi,j (0) = 2 ∂xi ∂x j (0). First
we show that we can assume that h1,1 (0) 6= 0. Indeed, suppose that h1,1 (0) = 0. Since the


matrix hi,j (0) is non-singular, there is an index i0 6= 1 such that h1,i0 (0) 6= 0. Make the
change of variable xi0 = xi00 + x1 , then in this new variable h1,1
0 (0) = 2h
1,i0 (0) 6 = 0.
When h1,1 (0) 6= 0 there is a neighborhood of 0 such that h1,1 ( x ) does not change its sign.
In that neighborhood, if h1,1 (0) > 0 then

f (x) = f (0)+ h1,1 ( x ) x12 + ∑ (h1,j ( x ) + h j,1 ( x )) x1 x j + ∑ hi,j ( x ) xi x j

1< j 1<i,j

= f (0)+ h1,1 ( x ) x12 + 2 ∑ h1,j ( x ) x1 x j + ∑ hi,j ( x ) xi x j

1< j 1<i,j

∑1< j h1,j ( x ) x j
q 2 q
= f (0)+ h1,1 ( x ) x1 +2 h1,1 ( x ) x1 p + ∑ hi,j ( x ) xi x j
h1,1 ( x ) 1<i,j
" #2
q h1,j ( x )
= f (0)+ h1,1 ( x ) x1 + ∑ ph xj
1< j 1,1 ( x )
!2
h1,j ( x )
− ∑ p
h1,1 ( x )
xj + ∑ hi,j ( x ) xi x j .
1< j 1<i,j

Similarly, if h1,1 (0) < 0 then

" #2
q h1,j ( x )
f (x) = f (0)− −h1,1 ( x ) x1 − ∑ p xj
1< j −h1,1 ( x )
!2
h1,j ( x )
+ ∑p xj + ∑ hi,j ( x ) xi x j .
1< j −h1,1 ( x ) 1<i,j

Combining both cases, we define the new variables:

q h1,j ( x )
v1 = |h1,1 ( x )| x1 + sign(h1,1 (0)) ∑ p xj,
1< j |h1,1 ( x )|
vi = xi , i > 1.

Since q
∂v1
(0) = |h1,1 (0)| 6= 0
∂x1
 
∂vi
the Jacobian matrix ∂x j (0) is non-singular. By the Inverse function theorem, there is a
neighborhood of 0 where the correspondence x 7→ v is a diffeomorphism, that is, a change
of variables. With the new variables we have

f (v) = f (0) + sign(h1,1 (0))v21 + ∑ 0

hi,j ( v ) vi v j .
1<i,j
152

By a direct calculation, we can check that in these variables

!
sign(h1,1 (0)) 0
H f (0) = 0 (0)) .
0 (2hi,j 1<i,j≤m

0 (0))
Using (22.2) we conclude that the matrix (hi,j 1<i,j≤m must be non-singular. Thus the
induction process can be carried out. Finally we can permute the variables such that in the
final form of f the negative signs are in front.

We now show that the number k of negative signs depend only on the function f and
does not depend on the choice of change of variable. Suppose that ϕ is a change of variables
with ϕ(0) = 0. Recall (22.2), we have

H ( f ◦ ϕ)(0) = J ϕ(0)t · H f ( ϕ(0)) · J ϕ(0) = J ϕ(0)t · H f (0) · J ϕ(0).

The number k is precisely the number of negative eigenvalues of H ( f ◦ ϕ)(0). The Sylvester
law on quadratic forms in Algebra (see e.g. [Lan97, p. 577]) says that the numbers of
negative eigenvalues of H ( f ◦ ϕ)(0) and H f (0) are equal, as the corresponding quadratic
forms are related by a change of variable.

For convenience here we provide a sketch of a proof for the Sylvester law. Let V be a
subspace of Rm of the maximum dimension on which the bilinear form represented by the
matrix H ( f ◦ ϕ)(0) is negatively definite, i.e. vt · H ( f ◦ ϕ)(0) · v < 0 for all 0 6= v ∈ V. Since

vt · H ( f ◦ ϕ)(0) · v = vt · J ϕ(0)t · H f (0) · J ϕ(0) · v = ( J ϕ(0) · v)t · H f (0) · ( J ϕ(0) · v),

we deduce that H f (0) is negatively definite on dϕ(0)(V ). Thus the maximum dimension
of a negatively definite space for H f (0) is bigger than or equal to the maximum dimension
of a negatively definite space for H ( f ◦ ϕ)(0). By symmetry (namely, by composing with
ϕ−1 and using the same argument) the two maximum dimensions must be same, equal to
a number l. Since H ( f ◦ ϕ)(0) is a diagonal matrix with k negative items it is easy to see
that l ≥ k. A similar argument gives that the maximum dimension of a positively definite
space for H ( f ◦ ϕ)(0) is larger than or equal to (m − k). This implies that l = k.

Example. On the the sphere x2 + y2 + z2 = 1 consider the height function f ( x, y, z) = z.

p
The upper-half of the sphere is the graph of the function z = 1 − x2 − y2 . With this
∂f
local parametrization we get ∂
∂x ( x, y, z ) = ∂x (1 − x2 − y2 )1/2 = − x (1 − x2 − y2 )−1/2 and
∂f
∂y ( x, y, z ) = ∂
∂y (1 − x2 − y2 )1/2 = −y(1 − x2 − y2 )−1/2 . The only critical point of f is
∂2 f
(0, 0, 1). With this local parametrization we get for example ∂x2
( x, y, z) = ∂
∂x2
(1 − x2 −
y2 )1/2 , and we find !
−1 0
H f (0, 0, 1) = .
0 −1

Thus (0, 0, 1) is a non-degenerate critical point. The index of this critical point is the number
of negative eigenvalues of the Hessian matrix, which is 2.

Similarly we find that f has another non-degenerate critical point on the lower-half of
the sphere at (0, 0, −1), of index 0.
22. CRITICAL POINTS AND THE MORSE LEMMA 153

Problems
22.3. Let f : M → R be smooth where M is a compact smooth manifold of dimension m ≥ 1. Show
that f has at least two critical points.

22.4. Let f : M → R be smooth. Show that ∇ f = 0 if and only if f is constant on each connected
component of M.

22.5. Show that the direction of the gradient vector is the direction in which the function increases
the greatest.

22.6 (gradient vector is normal to level set). X Suppose that M ⊂ Rk is a smooth manifold, f : M →
R is smooth, c is a regular value of f , N = f −1 (c), and x ∈ N. Show that ∇ f ( x ) ⊥ TNx (in the
Euclidean inner product of Rk ).

22.7. Show that the following equation determines a surface (i.e. a two dimensional smooth mani-
fold) in R3 :
x2 + y2 + z2 + x4 y4 + x4 z4 + y4 z4 − 9z = 21

Give an equation for the tangent plane of this surface at the point (1, 1, 2).

22.8 (method of Lagrange multipliers). Suppose that f : M → R is smooth. Suppose that g : M → R

is smooth, c a regular value of g, and N = g−1 (c). Suppose that f | N has a local extremum at x ∈ N.
(a) Show that ∇ f ( x ) ⊥ TNx and ∇ g( x ) ⊥ TNx
(b) Deduce that there is λ ∈ R such that ∇ f ( x ) = λ∇ g( x ).
In Calculus, the above statement can be phrased as the method of Lagrange multipliers: To find the
maximum and minimum values of f subject to the constraint g( x ) = c (assuming that these extreme
values exist and ∇ g( x ) 6= 0 on the set g( x ) = c) we need to solve the system ∇ f ( x ) = λ∇ g( x ) and
g( x ) = c for x and λ.

22.9. * Generalize the method of Lagrange multipliers to two constraints.

22.10. Prove the Morse lemma for the special case of a quadratic function f ( x ) = ∑1≤i,j≤m ai,j xi x j ,
where ai,j are real numbers. In this case we can use a diagonalization of a symmetric matrix or a
quadratic form considered in Linear Algebra. The change of variable corresponds to using a new
vector basis consisting of eigenvectors of the matrix.

22.11. Show that the set of non-degenerate critical points of a smooth function is discrete, i.e. each
element of the set has a neighborhood in the manifold containing no other element of the set (ex-
plaining the name, the subspace topology which this set inherits from the manifold coincide with the
discrete topology).

22.12 (the classification of critical points in Calculus). Recover the following sufficient condition
for extrema in Calculus from the Morse lemma. Suppose that f : U → R is smooth where U is an
open set in R2 and ( a, b) is a critical point of f . Compute

2
det H f ( a, b) = f xx ( a, b) f yy ( a, b) − f xy ( a, b) .

(a) If det H f ( a, b) > 0 and f xx ( a, b) > 0, then f ( a, b) is a local minimum.

(b) If det H f ( a, b) > 0 and f xx ( a, b) < 0, then f ( a, b) is a local maximum.
(c) If det H f ( a, b) = 0, then f ( a, b) is not a local maximum or minimum.

22.13. Deduce the following sufficient condition for extrema from Calculus from the Morse lemma.
Suppose that f : U → R is smooth where U is an open subset in Rm and x is a critical point of f . Let
B(u, v) = ut · H f ( x ) · v for u, v ∈ Rm . Then B is a quadratic form on Rm .
(a) If B is a positive definite, i.e. B(u, u) > 0 for all u 6= 0 then f has a local minimum at x.
154

(b) If B is a negative definite, i.e. B(u, u) < 0 for all u 6= 0 then f has a local maximum at x.
(c) If there are u 6= 0 such that B(u, u) > 0 and v 6= 0 such that B(v, v) < 0 then f does not have a
local extremum at x.
p 2
22.14. Consider the torus in R3 given by the equation x 2 + y2 − b + z2 = a2 , 0 < a < b. Let
f ( x, y, z) = x (height function in the direction of the x-axis). Find the non-degenerate critical points
of f and their indexes.

22.15 (height function). Let M ⊂ Rk be a smooth manifold. Let v ∈ Rk with unit length. Consider
the function f on M given by f ( x ) = h x, vi (the projection to v) where the inner product is that of
Rk . This is a height function in the direction of v. Show that x is a critical point of f if and only if
v ⊥ TMx .
23. FLOWS 155

23 Flows

Ordinary differential equations

Let V be a smooth vector field defined on an open subset U ⊂ Rm , i.e. a smooth map
V : U → Rm . We define an integral curve of V at a point x0 ∈ U to be a smooth path
γ : ( a, b) → U such that 0 ∈ ( a, b), γ(0) = x0 , and γ0 (t) = V (γ(t)) for all t ∈ ( a, b). It is a
path going through x and at every moment takes the vector of the given vector field as its
velocity vector. An integral curve of a vector field is tangent to the vector field.
Thus, by using local coordinate, we can consider a local integral curve as a solution to
the differential equation γ0 (t) = V (γ(t)) in Rm subjected to the initial condition γ(0) = x0 .
In more common notations in differential equations, a local integral curve is a local solution
to the following initial value problem:

 dx = V ( x ),
dt
(23.1)
 x (0) = x .
0

Followings are fundamental theorems in the subject of Ordinary Differential Equations

(ODE) on the existence, uniqueness, and dependence on initial conditions of solutions to
differential equations, see for example [HS74], [Lan97].

Theorem (existence). If V is smooth in a neighborhood of x0 then the equation (23.1) has a local
solution, that is, there is a smooth function x defined on an interval Ix that satisfies (23.1).

Theorem (uniqueness). If V is smooth in a neighborhood of x0 then the equation (23.1) has a

unique local solution, meaning that if x is a local solution defined on an interval Ix and y is a local
solution defined on an interval Iy , then x = y on Ix ∩ Iy .

Let I = where the union is over the set S of all solutions to (23.1). Let t ∈ I, then
S
x ∈ S Ix
there is x ∈ S such that t ∈ Ix . Define x̃ : I → U by x̃ (t) = x (t). Then x̃ (t) is defined not
depending on the choice of xby uniqueness of solution. Furthermore x̃ is also a solution
itself, called the maximal solution of (23.1), as it has the maximal domain.
Given a a smooth vector field V, for each x ∈ U, let φ(t, x ) or φt ( x ) be the maximal
integral curve of V at x, with t belonging to an interval Ix . We have a map

φ : D = {(t, x ) | x ∈ U, t ∈ Ix } ⊂ R × U → U
(t, x ) 7→ φt ( x ),

d
with the properties φ0 ( x ) = x, and dt φt ( x ) = V (φ(t, x )). This map φ is called the flow
(dòng) generated by the vector field V. We can think of φt as moving every point along the
integral curve for an amount of time t.

Theorem (dependence on initial condition). The flow of a smooth vector field has an open
domain and is smooth.

Flows on manifolds
Definition. A smooth tangent vector field on a manifold M ⊂ Rk is a smooth map V :
M → Rk such that V ( x ) ∈ TMx for each x ∈ M.

Example. If f : M → R is smooth then the gradient ∇ f is a smooth vector field on M.

156

In a local parametrized neighborhood around a point on the manifold, a smooth vec-

tor field on that neighborhood corresponds to a smooth vector field on an open subset of
Rm , and an integral curve in that neighborhood corresponds to an integral curve on Rm .
Namely, let ϕ : U ⊂ Rm → M be a local parametrization at x0 = ϕ(u0 ) then the corre-
sponding ODE in Rm is: 
 du = (dϕ )−1 (V ( ϕ(u))) ,
dt u
(23.2)
 u (0) = u .
0

This system has a unique maximal solution u defined on an interval Iu . We check that ϕ ◦ u
is a solution to 23.1. First,

ϕ ◦ u(0) = ϕ(u(0)) = ϕ(u0 ) = x0 .

Next,

d   −1
( ϕ ◦ u)(t) = dϕu(t) (u0 (t)) = dϕu(t) dϕu(t) (V ( ϕ(u(t))))
dt
= V ( ϕ(u(t))) = V ( ϕ ◦ u(t)),

as expected. We have just shown that 23.1 has a local solution.

Next we show that this solution does not depend on the choice of local parametrization.
Suppose that ψ is another local parametrization and with it a solution ψ ◦ v is found. We
show that ( ϕ−1 ◦ ψ) ◦ v is a solution to 23.2. First, ( ϕ−1 ◦ ψ) ◦ v(0) = ( ϕ−1 ◦ ψ)(v0 ) =
ϕ−1 (ψ(v0 )) = ϕ−1 ( x0 ) = u0 . Next
h i0 h i
( ϕ−1 ◦ ψ) ◦ v (t) = d( ϕ−1 ◦ ψ)v(t) (v0 (t)) = d( ϕ−1 ◦ ψ)v(t) (dψv )−1 (V (ψ(v(t)))
  −1
= dϕ− 1
ψ(v(t))
( V ( ψ ( v ( t ))) = dϕ − 1
( ϕ ◦ψ)◦v(t) (V (ψ(v(t)))
  −1  
= dϕ( ϕ−1 ◦ψ)◦v(t) V ( ϕ(( ϕ−1 ◦ ψ) ◦ v(t)) ,

as expected. As u is already the maximal solution, the solution ( ϕ−1 ◦ ψ) ◦ v must agree
with u on its domain. Therefore ψ ◦ v must agree with ϕ ◦ u on its domain. We deduce that
ϕ ◦ u is the maximal solution for 23.1.
Now for each x ∈ M let φ(t, x ) or φt ( x ) be the maximal integral curve of V at x, with t
belonging to an interval Ix . We have a map

φ : D = {(t, x ) | x ∈ M, t ∈ Ix } ⊂ R × M → M
(t, x ) 7→ φt ( x ),

d
with the properties φ0 ( x ) = x, and dt φ ( t, x ) = V (φ(t, x )). This map φ is called the flow
(dòng) generated by the vector field V. We have:

Theorem. The flow of a smooth vector field on a smooth manifold has an open domain and is
smooth.

Let ψ : U → M be a local parametrization of an open neighborhood of x = ψ(u).

Then (id, ψ) is a local parametrization of a neighborhood
 of (t, x ). Then ψ−1 ◦ φ satisfies
( ψ −1 ◦ φ)(0, x ) = u and d
dt ( ψ
−1 ◦ φ)(t, x ) = dψφ−(1t,x) d
dt φ ( t, x )
By the result on dependence on initial condition in Rm ,
23. FLOWS 157

Theorem (Group law). Given x ∈ M, if φt ( x ) is defined on (−e, e) then for s ∈ (−e, e) such
that s + t ∈ (−e, e) we have
φt+s ( x ) = φt (φs ( x )).

Proof. Define γ(t) = φt+s ( x ). Then γ(0) = φs ( x ), and γ0 (t) = d

dt φ ( t
+ s, x ) = V (φ(t +
s, x )) = V (γ(t)). Thus γ is an integral curve at φs ( x ). But if β(t) = φt (φs ( x )) then β is
another integral curve at φs ( x ). By uniqueness of integral curves, γ must agree with β on
their common domains.

Theorem. If φt : M → M is defined for t ∈ (−e, e) then it is a diffeomorphism.

Proof. Since the flow φ is smooth the map φt is smooth. Its inverse map φ−t is also smooth.

When every integral curve can be extended without bound in both directions, that is,
for every x ∈ M the point φt ( x ) is defined for every t ∈ R, we say that the flow is complete.

Theorem. On a compact manifold any flow is complete.

Proof. Although a priori each integral curve has its own domain, we will show that for a
compact manifold all integral curves can have same domains. Since the domain D of the
flow can be taken to be an open subset of R × M, each x ∈ M has an open neighborhood
Ux and a corresponding interval (−ex , ex ) such that (−ex , ex ) × Ux is contained in D. The
collection {Ux | x ∈ M } is an open cover of M, so there is a finite subcover corresponding
to xi , 1 ≤ i ≤ n. Take e = min {ei | 1 ≤ i ≤ n}, then for every x ∈ M the integral curve
φt ( x ) is defined on (−e, e).
Now φt ( x ) can be extended inductively by intervals of length e/2 to be defined on R.
For example, if t > 0 then there is n ∈ N such that n 2e ≤ t < (n + 1) 2e , and if n ≥ 2 we
define  
φt ( x ) = φt−n e φn e ( x ) ,
2 2
 
where inductively φn e ( x ) = φ e φ(n−1) e ( x ) .
2 2 2

23.3 Theorem. Let M be a compact smooth manifold and let f : M → R be smooth. If the interval
[ a, b] only contains regular values of f then the level sets f −1 ( a) and f −1 (b) are diffeomorphic.

Briefly, if between the two levels there is no critical point then the two levels are diffeo-
morphic.

Proof. The idea of the proof is to construct a diffeomorphism from f −1 ( a) to f −1 (b) by

pushing points from one level to the other level along the flow lines of the gradient vector
field. To do this we want the gradient vector at each point to be non-zero. Also we need
a complete flow in order to follow the flow lines as long as necessary. We want all the
flow lines from one level to reach the other level at the same time, so we want the speed of
change of the level f (φt ( x )) with respect to time to be constant for all x.
By 21.12 there are intervals [ a, b] ⊂ (c, d) ⊂ [c, d] ⊂ (h, k) such that (h, k ) contains only
regular values of f . On the submanifold N = f −1 ((h, k )) the gradient vector ∇ f never
vanish. By 23.8 there is a smooth function ψ that is 1 on f −1 ([c, d]) and is 0 outside of N.
∇f
Let F = ψ ||∇ f ||2 on N and F = 0 outside of N, then F is a well-defined smooth vector field
on M. Notice that F is basically a re-scale of ∇ f .
158

Let φ be the flow generated by F. Fix x ∈ f −1 ( a), on level a. We have

 
d d
f (φt ( x )) = d f φt ( x) = h∇ f (φt ( x )), F (φt ( x ))i
φt ( x )
dt dt
* +
∇ f (φt ( x ))
= ∇ f (φt ( x )), ψ(φt ( x )) = ψ(φt ( x )),
k∇ f (φt ( x ))k2

which is equal to 1 if the point φt ( x ) remains in f −1 ([c, d]). This means the level of the flow
line from x is going up at constant speed 1. So if we can travel along this flow line for an
interval of time of length (b − a) then we will reach level b. We now show that this is indeed
the case.
Since φt ( x ) is continuous with respect to t and φ0 ( x ) = x, there is an e > 0 such that
φt ( x ) ∈ f −1 ((c, d)) for t ∈ [0, e). Let e0 be the supremum (or ∞) of the set of such e. For
t ∈ [0, e0 ) we get f (φt ( x )) = t + a. If e0 ≤ b − a then by continuity f (φe0 ( x )) = e0 + a ≤
b < d. This implies there is e0 > e0 such that f (φt ( x )) < d for t ∈ [e0 , e0 ), a contradiction.
Thus e0 > b − a. Observe that f (φb−a ( x )) = b.
Thus φb−a maps f −1 ( a) to f −1 (b), so it is the desired diffeomorphism.

23.4 Theorem. On a connected manifold there is a self diffeomorphism that brings any given point
to any given point.

We say that any connected manifold is homogeneous (đồng nhất).

Proof. First we show that we can locally bring any point to any given point without outside
disturbance. That translates to a problem on Rn : we will show that for any c ∈ B(0, 1) there
is a diffeomorphism h : Rn → Rn such that h|Rn \ B(0,1) = id and h(0) = c.
It is possible to explicitly construct a diffeomorphism from B(0, 1) to itself bringing 0 to
c, but to have one that fixes the complement we use an existence proof, following [Mil97].
Take kck < δ < 1. By 23.6 there is a smooth function ϕ : Rn → [0, 1] such that ϕ| B0 (0,δ) = 1
and ϕ|Rn \ B(0,1) = 0. Consider the vector field V : Rn → Rn , V ( x ) = ϕ( x )c. This is a smooth
vector field with compact support. The flow generated by this vector field is complete, is
the unique smooth map F : R × Rn → Rn such that

F0 ( x ) = x,
d
Ft ( x ) = V ( Ft ( x )).
dt

We know that for each t the map Ft : Rn → Rn is a diffeomorphism. Notice that Ft (0) = ct
is a solution for − kδck < t < δ
kck
. Since F1 (0) = c, we see that F1 is a diffeomorphism of
Rn bringing 0 to c. For each x ∈ Rn \ B(0, 1), Ft ( x ) = x is a solution, so Ft is the identity
outside B(0, 1). Thus F1 is isotopic to F0 via the isotopy Ft fixing the complement of B(0, 1).
Now we prove the result globally using connnectedness. Let M be a connected mani-
fold. Let x ∈ M and let S be the set of points of M such that there exists a diffeomorphism
of M bringing x to y. The first part of this proof shows that S is an open set. If z ∈ M is a
limit point of S, take a sufficiently small neighborhood of z such that there is a diffeomor-
phism of M bringing any point in that neighborhood to z. That neighborhood contains a
point y in S. There is diffeomorphism of M bringing x to y and a diffeomorphism bringing
y to z, the composition brings xto z, thus z ∈ S. So S is both open and closed in M, implying
S = M. Alternatively we can check that the relation that there is a diffeomorphism of M
23. FLOWS 159

B(0, 1)

B(0, δ)

0 c

Figure 23.5: A flow pushing 0 to c, fixing the complement of the ball B(0, 1).

bringing x to y is an equivalence relation, and each equivalence class is open, so there is

only one class.

Problems
23.6 (test function, mollifier). X Consider the following special function:

e−1/x , if x > 0
f1 (x) =
0, if x ≤ 0.

(a) Show that f (n) ( x ) = e−1/x Pn (1/x ) for x > 0, where Pn is a polynomial. Show that f is smooth.

f1

(b) Let
f 1 ( x − a)
f2 (x) =
f 1 ( x − a) + f 1 (b − x )
 
then f 2 is smooth, 0 ≤ f 2 ≤ 1, f 2 (b,∞) ≡ 1 , f 2 (−∞,a) ≡ 0 .
 
160

1 f2

a b

(c) With 0 < a < b, the function f 3 ( x ) = f 2 ( x2 ) is smooth, 0 ≤ f 3 ≤ 1, f 3 |(−√ a,√ a) = 0,

f 3 |R\(−√b,√b) = 1.

1 f3

√ √ √ √
− b− a a b

(d) In Rn , with 0 < a < b, let f 4 ( x ) = 1 − f 3 (k x k2 ). Show that f 4 is smooth, 0 ≤ f 4 ≤ 1,

f 4  B(0, √ √
a) = 1 , f 4 Rn \ B(0, 4 b) = 0 This is a smooth function whose values are between 0 and 1,
 
4

equal to 1 inside a smaller ball, equal to 0 outside of a bigger ball. This function can be used to
connect different level sets smoothly.

23.7 (smooth Urysohn lemma). Let A ⊂ U ⊂ Rn where A is compact and U is open. We will show
that there exists a smooth function ϕ : Rn → R such that 0 ≤ ϕ ≤ 1, ϕ | A = 1 , ϕ Rn \U = 0 .


Sk Sk
(a) Show that there exists a family of balls ( B( xi , ei ))1≤i≤k such that i =1 B( xi , ei ) ⊃ A and i =1 B( xi , 2ei ) ⊂
U.
23. FLOWS 161
 
(b) Let ϕi : Rn → R be smooth, 0 ≤ ϕi ≤ 1, ϕi  B( xi ,ei ) ≡ 1 , ϕi Rn \ B( xi ,2ei ) ≡ 0 . Let ψ = ∑ik=1 ϕi .
 

Show that ψ | A > 0 , ψ Rn \U ≡ 0 .

 
(c) Let c = min A ψ. Let h be a smooth function such that 0 ≤ h ≤ 1, h (−∞,0] ≡ 0 , h [c,∞) ≡ 1 . Let
 

ϕ = h ◦ ψ. Show that this is a function that we are looking for.

23.8 (smooth Urysohn lemma for manifolds). Let M be a smooth m-dimensional manifold, A ⊂ U ⊂
M where A is compact and U is open  in M. We will show that there is a smooth function ϕ : M → R
such that 0 ≤ ϕ ≤ 1, ϕ | A ≡ 1 , ϕ  M\U ≡ 0 .


(a) Show that there exists a family of local parametrization (ψi )1≤i≤k and a family of balls ( B( xi , ei ))1≤i≤k
Sk
in Rm such that ψi ( B( xi , ei )) ⊃ A and ψi ( B( xi , 2ei )) ⊂ U.
i =1
 
(b) Let ϕi : Rn → R be smooth, 0 ≤ ϕi ≤ 1, ϕi  B( xi ,ei ) ≡ 1 , ϕi Rn \ B( xi ,2ei ) ≡ 0 . Let ϕi0 = ϕi ◦ ψi−1
 
 
on ψi (Rm ) and ϕi0  M\U ≡ 0 . Let ψ = ∑ik=1 ϕi0 . Show that ψ | A > 0 , ψ  M\U ≡ 0 .
 
 
(c) Let c = min A ψ. Let h be a smooth function such that 0 ≤ h ≤ 1, h (−∞,0] ≡ 0 , h [c,∞) ≡ 1 . Let
 

ϕ = h ◦ ψ. Show that this is a function we are looking for.

162

24 Manifold with boundary

The closed half-space Hm = {( x1 , x2 , . . . , xm ) ∈ Rm | xm ≥ 0} ⊂ Rm whose topological
boundary is ∂Hm = {( x1 , x2 , . . . , xm ) ∈ Rm | xm = 0} is our model for a manifold with
boundary.

Definition. A subspace M of Rk is called a manifold with boundary of dimension m if

each point in M has a neighborhood diffeomorphic to either Rm or Hm .
The set of all points having neighborhoods diffeomorphic to Rm is called the interior of
M.
If a point has a neighborhood diffeomorphic to Hm via a diffeomorphism that send the
point to ∂Hm then it is called a boundary point of M. The set of all boundary points of M
is called the boundary of M, denoted by ∂M. Thus x ∈ ∂M if and only if there is an open
set U in M, x ∈ U, and there is ϕ : U → Hm a diffeomorphism, such that ϕ( x ) ∈ ∂Hm .

A point belongs to either the interior or the boundary, not both, since Hm is not diffeo-
morphic to Rm , by smooth invariance of domain 19.

Remark. We will follow a common convention that the term “manifold” is reserved for
manifold without boundary. A manifold with boundary can have empty boundary, in
which case it is a manifold. When there is a need to avoid confusion, we will say clearly
as “let M be a manifold without boundary”. We can also say “let M be a manifold with or
without boundary”.
The boundary of a manifold is generally not the same as its topological boundary.

Proposition. The interior of an m-manifold with boundary is an m-manifold without boundary.

The boundary of an m-manifold with boundary is an (m − 1)-manifold without boundary.

Proof. The part about the interior is clear. Let us consider the part about the boundary.
Let M be an m-manifold and let x ∈ ∂M. Let ϕ be a diffeomorphism from a neighbor-
hood U of x in M to Hm . If y ∈ U then ϕ(y) ∈ ∂Hm if and only if y ∈ ∂M. Thus the
restriction ϕ|U ∩∂M is a diffeomorphism from a neighborhood of x in ∂M to ∂Hm , which is
diffeomorphic to Rm−1 .

Thetangent space of a manifold with boundary M is defined as follows. It x is an in-

terior point of M then TMx is defined as before. If x is a boundary point then there is a
parametrization ϕ : Hm → M, where ϕ(0) = x. Notice that by continuity ϕ has well-
defined partial derivatives at 0 (see 19.2). This implies that the derivative dϕ0 : Rm → Rk is
well-defined. Then TMx is still defined as dϕ0 (Rm ). The chain rule for functions still holds.

Regular values for manifolds with boundaries

The notion of critical point is defined exactly as for manifolds without boundaries.

24.1 Theorem. Let M be an m-dimensional manifold without boundary. Let f : M → R be

smooth and let y be a regular value of f . Then the set f −1 ([y, ∞)) is an m-dimensional manifold
with boundary f −1 (y).

Proof. Let N = f −1 ([y, ∞)). Since f −1 ((y, ∞)) is an open subspace of M, it is an m-manifold
without boundary.
The crucial case is when x ∈ f −1 (y). Let ϕ : Rm → M be a parametrization of a neigh-
borhood of x in M, with ϕ(0) = x. Let g = f ◦ ϕ. As in the proof of 21.3, by the Implicit
24. MANIFOLD WITH BOUNDARY 163

function theorem, there is an open ball U in Rm−1 containing 0 and an open interval V in
R containing 0 such that in U × V the set g−1 (y) is a graph {(u, h(u)) | u ∈ U } where h is
smooth.

f
y
x

ϕ g

Since (U × V ) \ g−1 (y) consists of two connected components, exactly one of the two is
mapped via g to (y, ∞), otherwise x will be a local extremum point of f , and so d f x = 0, vi-
olating the assumption. In order to be definitive, let us assume that W = {(u, v) | v ≥ h(u)}
is mapped by g to [y, ∞). Then ϕ(W ) = ϕ(U × V ) ∩ f −1 ([y, ∞)) is a neighborhood of x in
N parametrized by ϕ|W . On the other hand W is diffeomorphic to an open neighborhood
of 0 in Hm . To show this, consider the map ψ(u, v) = (u, v − h(u)) on U × V. Then ψ is a
smooth bijection on open subspaces of Rm , whose Jacobian is non-singular, therefore is a
diffeomorphism. The restriction ψ|W is a diffeomorphism to ψ(U × V ) ∩ Hm . Thus x is a
boundary point of N.

Example. The closed disk D n is an n-manifold with boundary Sn−1 . Indeed, let f : Rn →
R, f ( x ) = k x k2 . Since 1 is a regular value of f , the sub-level set D n = f −1 ((−∞, 1]) is an
n-manifold with boundary f −1 (1) = Sn−1 .

Example. Let f be the height function on S2 , f ( x, y, z) = z. Since 0 is a regular value of

f the set f −1 ([0, ∞)), which is the upper hemisphere, is a 2-dimensional manifold with
boundary f −1 (0), which is diffeomorphic to S1 .

24.2 Theorem. Let M be an m-dimensional manifold with boundary, let N be an n-manifold with
or without boundary. Let f : M → N be smooth. Suppose that y ∈ N is a regular value of both f
and f |∂M . Then f −1 (y) is an (m − n)-manifold with boundary ∂M ∩ f −1 (y).

Proof. That f −1 (y) \ ∂M is an (m − n)-manifold without boundary is already proved in

21.3.
We consider the crucial case of x ∈ ∂M ∩ f −1 (y).
Let ϕ : Hm → M be a parametrization of a neighborhood of x in M, with ϕ(0) = x. Let
g = f ◦ ϕ. The map g can be extended to g̃ defined on an open neighborhood Ũ of 0 in Rm .
Notice that d g̃0 = dg0 , we have 0 is a regular point of g̃. By the Implicit function theorem,
if we take a small enough neighborhood then g̃−1 (y) is a graph of a function of (m − n)
variables so it is an (m − n)-manifold without boundary containing on regular points of g̃.
Let p : g̃−1 (y) → R be the projection to the last coordinate (the height function), namely
( x1 , x2 , . . . , xm ) 7→ xm . We have g−1 (y) = p−1 ([0, ∞)) therefore if we can show that 0 is a
regular value of p then the desired result follows from 24.1 applied to g̃−1 (y) and p. For
164

x f
y

ϕ g̃

∂Hm
Ũ
T g̃−1 (y)u

x f
y

U
ϕ g̃

∂Hm
T g̃−1 (y)u
Ũ

each u ∈ p−1 (0) ⊂ ∂Hm the derivative dpu is given by the vector em , so dpu is onto if and
only if the tangent space T g̃−1 (y)u is not contained in ∂Hm .
Since g̃ is regular at u, the null space of d g̃u on T Ũu = Rm is exactly T g̃−1 (y)u , of
dimension m − n. On the other hand, g̃|∂Hm is regular at u, which implies that the null
space of d g̃u restricted to T (∂Hm )u = ∂Hm has dimension (m − 1) − n. Thus T g̃−1 (y)u is
not contained in ∂Hm .

Example. Let f : D2 → R, f ( x, y) = y. Then 0 is a regular value of f and f −1 (0) is a

1-dimensional manifold with boundary on ∂D2 .

Sard theorem
We use the following result from Analysis:

Theorem (Sard theorem). The set of critical values of a smooth map from Rm to Rn is of Lebesgue
measure zero.

For a proof see for instance [Mil97]. Sard theorem also holds for smooth functions from
Hm to Rn , see 24.10.
Since a set of Lebesgue measure zero must have empty interior, we have:

Corollary. The set of regular values of a smooth map from Rm to Rn is dense in Rn .

An application of Sard theorem for manifolds is the following:

24. MANIFOLD WITH BOUNDARY 165

Theorem. If M and N are manifolds with boundaries and f : M → N is smooth then the set of all
regular values of f is dense in N. Thus, any smooth function has a regular value.

Proof. Consider any open subset V of N parametrized by ψ : V 0 → V. Then f −1 (V ) is an

open submanifold of M. We only need to prove that f | f −1 (V ) has a regular value in V. Let
C be the set of all critical points of f | f −1 (V ) .
We can cover f −1 (V ) (or any manifold) by a countable collection I of parametrized open
neighborhoods. This is possible because a Euclidean space has a countable topological basis
(see 2.18).
For each U ∈ I we have a commutative diagram:

f
UO /V
O
ϕU ψ
gU
U0 / V0

where U 0 is an open subset of Hm and V 0 is an open subset of Hn . From this diagram, x is

−1
a critical point of f in U if and only if ϕU ( x ) is a critical point of gU . Thus the set of critical
−1
points of gU is ϕU ( C ∩ U ).
Now we write

!
−1 −1
[ [ [
f (C ) = f (C ∩ U ) = ψ( gU ( ϕ (C ∩ U ))) = ψ gU ( ϕ (C ∩ U )) .
U∈ I U∈ I U∈ I

−1
By Sard Theorem the set gU ( ϕU (C ∩ U )) is of measure zero. This implies that the set
−1
D = U ∈ I gU ( ϕU (C ∩ U )) is of measure zero, since a countable union of sets of measure
S

zero is a set of measure zero. As a consequence D must have empty topological interior.
Since ψ is a homeomorphism, ψ( D ) = f (C ) must also have empty topological interior.
Thus f (C ) V, so there must be a regular value of f in V.

Brouwer fixed point theorem

If N ⊂ M and f : M → N such that f | N = id N then f is called a retraction from M to N
and N is a retract of M.

24.3 Lemma. Let M be a compact manifold with boundary. There is no smooth map f : M → ∂M
such that f |∂M = id∂M . In other words there is no smooth retraction from M to its boundary.

Proof. Suppose that there is such a map f . Let y be a regular value of f . Since f |∂M is the
identity map, y is also a regular value of f |∂M . By theorem 24.2 the inverse image f −1 (y) is
a 1-manifold with boundary f −1 (y) ∩ ∂M = {y}. But a 1-manifold cannot have boundary
consisting of exactly one point, see 24.4.

24.4 Theorem (classification of compact one-dimensional manifolds). A smooth compact

connected one-dimensional manifold is diffeomorphic to either a circle, in which case it has no bound-
ary, or an arc, in which case its boundary consists of two points.

For a proof, see e.g. [Mil97] .

Repeating of the proof for the continuous Brouwer fixed point theorem as in 17.3, we
get:
166

24.5 Corollary (smooth Brouwer fixed point theorem). A smooth map from the disk D n to
itself has a fixed point.

Proof. Suppose that f does not have a fixed point, i.e. f ( x ) 6= x for all x ∈ D n . The straight
line from f ( x ) to x will intersect the boundary ∂D n at a point g( x ). Then g : D n → ∂D n is
a smooth function which is the identity on ∂D n . That is impossible, by 24.3.

A proof for the continuous version of the theorem using the smooth version can be
found in [Mil97].

Problems
24.6. Show that the subspace Rm
+ = {( x1 , x2 , . . . , xm ) ∈ R | xm > 0} is diffeomorphic to R .
m m

24.7. A simple regular path is a map γ : [ a, b] → Rm such that γ is injective, smooth, and γ0 (t) 6= 0 for
all t ∈ [ a, b]. Show that the trace of a simple closed regular path is a smooth 1-dimensional manifold
with boundary.

24.8. Suppose that M is an n-manifold without boundary. Show that M × [0, 1] is an (n + 1)-manifold
with boundary. Show that the boundary of M × [0, 1] consists of two connected components, each of
which is diffeomorphic to M.

24.9. Let M be a compact smooth manifold and f : M → R be smooth. Show that if the interval
[ a, b] only contains regular values of f then the sub-level sets f −1 ((−∞, a]) and f −1 ((−∞, b]) are
diffeomorphic.

24.10. Show that Sard theorem also holds for smooth functions from Hm to Rn .

24.11. Show that a smooth loop on S2 (i.e. a smooth map from S1 to S2 ) cannot cover S2 . Similarly,
there is no smooth surjective maps from R to Rn with n > 1. In other words, there is no smooth
space filling curves, in contrast to the continuous case (compare 9).

24.12. Check that the function g in the proof of 24.5 is smooth.

24.13. Let A be an n × n matrix whose entries are all non-negative real numbers. We will derive the
Frobenius theorem which says that A must have a real non-negative eigenvalue.
Av
(a) Suppose that A is not singular. Check that the map v 7→ k Av k
brings Q = {( x1 , x2 , . . . , xn ) ∈
S n − 1 | xi ≥ 0, 1 ≤ i ≤ n} to itself.
(b) Prove that Q is homeomorphic to the closed ball D n−1 .
(c) Use the continuous Brouwer fixed point theorem to prove that A has a real non-negative eigen-
value.
25. ORIENTATION 167

25 Orientation

Orientations on vector spaces

On a finite dimensional real vector space, two vector bases are said to determine the same
orientation of the space if the change of bases matrix has positive determinant. Being of
the same orientation is an equivalence relation on the set of all bases. With this equivalence
relation the set of all bases is divided into two equivalence classes. If we choose one of the
two classes as the preferred one, then we say the vector space is oriented and the chosen
equivalence class is called the orientation (or the positive orientation).
Thus any finite dimensional real vector space is orientable (i.e. can be oriented) with
two possible orientations.

Example. The standard, canonical positive orientation of Rm is represented by the basis

(e1 = (1, 0, . . . , 0), e2 = (0, 1, 0, . . . , 0), . . . , em = (0, . . . , 0, 1)).

Unless stated otherwise, Rm is oriented this way.

Let T be an isomorphism from an oriented finite dimensional real vector space V to an

oriented finite dimensional real vector space W. Then T brings a basis of V to a basis of W.
There are only two possibilities:

Lemma. Either T brings any positive basis of V to a positive basis of W, or T brings any positive
basis of V to a negative basis of W.

Proof. Let b = (b1 , . . . , bn ) and b0 = (b10 , . . . , bn0 ) be two bases of V of same orientations.
That means the change of bases matrix A = ( ai,j )1≤i≤n,1≤ j≤n has positive determinant.
Since T (bi0 ) = T (∑nj=1 a j,i b j ) = ∑nj=1 a j,i T (b j ), we see that the change of bases matrix from
( T (b1 ), . . . , T (bn )) to ( T (b10 ), . . . , T (bn0 )) is also A, so the two bases are of same orientations
in W.

In the first case we say that T is orientation-preserving, and in the second case we say
that T is orientation-reversing.

Orientations on manifolds
Roughly, a manifold is oriented if at each point an orientation for the tangent space is cho-
sen and this orientation should be smoothly depended on the point.

Definition. A smooth manifold M of dimension m, with or without boundary is said to be

oriented if at each point x an orientation for the tangent space TMx is chosen and at each
point there exists a local parametrization ϕ such that the derivative dϕu : Rm → TM ϕ(u) is
orientation-preserving for all u.
 
∂ϕ ∂ϕ ∂ϕ
Thus the orientation of TM ϕ(u) is given by the basis ∂u1 ( u ), ∂u2 ( u ), . . . , ∂um ( u ) where
the orientation of Rm to the mani-
∂ϕ
∂ui ( u ) = dϕu (ei ). Roughly, local parametrizations bring
fold.
If a manifold is oriented then the set of orientations of its tangent spaces is called an
orientation of the manifold and the the manifold is said to be orientable.
168

Another approach to orientation of manifold is to orient each parametrized neighbor-

hood first then require that the orientations on overlapping neighborhoods agree. Con-
cisely, suppose that ϕ : Rm → M is a parametrization of a neighborhood of M and
suppose that ψ : Rm → M parametrizes an overlapping neighborhood. Since dψv =
d(ψ ◦ ϕ−1 )v ◦ dϕu , the requirement for consistency of orientation is that the map d(ψ ◦ ϕ−1 )v
must be orientation preserving on Rm . Thus:

Proposition. A manifold is orientable if and only if it has an atlas whose all change of coordinate
functions are orientation-preserving.

Example. If a manifold is parametrized by one parametrization, that is, it is covered by

one local coordinate, then it is orientable, since we can take the unique parametrization to
bring an orientation of Rm to the entire manifold. In particular, any open subset of Rk is an
orientable manifold.
Let M and N be oriented manifolds. A map f : M → N is said to be orientation-
preserving if at each x ∈ M the derivative d f x is orientation-preserving.

Example. The graph of a smooth function f : D → Rl , where D ⊂ Rk is an open set, is

an orientable manifold, since this graph can be parametrized by a single parametrization,
namely x 7→ ( x, f ( x )).

25.1 Proposition. A (k − 1)-dimensional manifold in Rk is orientable if there is a smooth normal

unit vector field on the manifold.

A submanifold of one dimension less is often called a hypersurface (siêu phẳng). For
normal unit vector here we are using the Euclidean inner product of Rk .

Proof. Suppose M is a (k − 1)-dimensional manifold in Rk having a smooth normal vector

field n. At each point x ∈ M choose a linear basis b( x ) = (b1 ( x ), . . . , bk−1 ( x )) for Tx M
such that b0 ( x ) = (b1 ( x ), . . . , bk−1 ( x ), n( x )) is a linear basis in the positive orientation of
Rk , that is det b0 ( x ) > 0. We check that this property depends only on the orientation
class of b( x ). Indeed, suppose c( x ) = (c1 ( x ), . . . , ck−1 ( x )) is another basis of Tx M having
the same orientation with b( x ), meaning the change of bases matrix A( x ) from b( x ) to
c( x ) has det A( x ) > 0. Since n( x ) ⊥ T x M the vector n( x ) must be linearly independent
from vectors in Tx M, therefore the change of bases matrix of Rk from (b1 , . . . , bk−1 , n( x )) to
c0 ( x ) = (c1 ( x ), . . . , ck−1 ( x ), n( x )) is exactly
!
A( x ) 0
0 1

having determinant equal to det A( x ) > 0. This implies det c0 ( x ) = det A( x ) det b0 ( x ) > 0.
Thus this rule using the unit normal determine an orientation for Tx M.
Let ϕ : Rk−1 → M be a local parametrization of a neighborhood of ϕ(u0 ) = x0 . At
∂ϕ ∂ϕ
x = ϕ(u) we have a basis c( x ) = ∂u (u), . . . , ∂u (u) for Tx M. If c( x0 ) is not in the same
1 k −1
orientation as b( x0 ) we can interchange two variables of ϕ, that is to replace ϕ by ψ with
ψ(u1 , u2 , . . . , uk−1 ) = ϕ(u2 , u1 , . . . , uk−1 ) then we get a basis in the same orientation with
b( x0 ). Thus we can assume c( x0 ) has the same orientation with b( x0 ), that is det A( x0 ) > 0.
Above we got det c0 ( x ) = det A( x ) det b0 ( x ). Since the function u 7→ det (c0 ( ϕ(u)) is
smooth on Rk−1 and is positive at u0 it must be always positive, and since det (b0 (( ϕ(u)))
is always positive, we deduce that det A(( ϕ(u)) must always be positive. So c(( ϕ(u)) is
25. ORIENTATION 169

always in the same orientation with b(( ϕ(u)). Thus ϕ is an orientation-preserving local
parametrization and b gives an orientation for the manifold. We conclude that M is ori-
entable.

25.2 Proposition. If f : Rk → R is smooth and a is a regular value of f then f −1 ( a) is an

orientable manifold.

Proof. Let M = f −1 ( a). If x ∈ M then ker d f x = TMx , so the gradient vector ∇ f ( x ) non-
zero and is perpendicular to TMx (the gradient vector is always perpendicular to the level
set, see 22.6).
∇f
Thus ∇ f is a no-where zero smooth normal vector field and k∇ f k
is a unit smooth
normal vector field on f −1 ( a). We can now use 25.1.

Example. The sphere and the torus are orientable.

Proposition. A connected orientable manifold with or without boundary has exactly two orienta-
tions.

Proof. Suppose the manifold M is orientable. There is an orientation o on M. Then −o is

a different orientation on M. Suppose that o1 is an orientation on M, we show that o1 is
either o or −o.
If two orientations agrees at a point they must agree locally around that point. Indeed,
from the definition there is a neighborhood V of x and a local coordinates ϕ : V → Rm
that brings the orientation o1 to the standard orientation of Rm , and a local coordinates
ψ : V → Rk that brings the orientation o to the standard orientation of Rm . Assuming
ϕ( x ) = ψ( x ) = 0, then det J (ψ−1 ◦ ϕ) is smooth on Rm and is positive at 0, therefore it is
always positive. That implies o1 and o agree on V.
Let U be the set of all points x in M such that the orientation of TMx with respect to o1
is the same with the orientation of TMx with respect to o. Then U is open in M. Similarly
the complement M \ U is also open. Since M is connected, either U = M or U = ∅.

Let f : M → N be a diffeomorphism between oriented manifolds. If d f x is orientation-

preserving for all x ∈ M then we say that f is an orientation-preserving diffeomorphism.
If d f x is orientation-reversing for all x ∈ M then we say that f is an orientation-reversing
diffeomorphism.

Example. In Rk any translation is orientation-preserving, while any reflection over a hy-

perplane is orientation-reversing.

Orientable surfaces
A two dimensional smooth manifold in R3 is called a (smooth) surface. A surface is two-
sided if there is a smooth way to choose a unit normal vector N ( p) at each point p ∈ S. That
is, there is a smooth map N : S → R3 such that at each p ∈ S the vector N ( p) has length 1
and is perpendicular to TS p .

Proposition. A surface is orientable if and only if it is two-sided.

Proof. If the surface S is orientable then its tangent spaces could be oriented smoothly. That
means at each point p ∈ S there is a local parametrization r (u, v) such that {ru (u, v), rv (u, v)}
ru (u, v) × rv (u, v)
gives the orientation of TS p . Then the unit normal vector is defined
||ru (u, v) × rv (u, v)||
smoothly on the surface.
170

The converse is given by 25.1 above.

Now we are able to prove a famous fact, that the Mobius surface (see 9.8) is not ori-
entable.

Figure 25.3: The Mobius surface is not orientable and is not two-sided.

Visually, if we pick a normal vector to the surface at a point in the center of the Mobius
band, then move that normal vector smoothly along the center circle of the band. When we
come back at the initial point after one loop, we realize that the normal vector is now in the
opposite direction. That demonstrate that the Mobius surface is not two-sided. Similarly if
we choose an orientation at a point then move that orientation continuously along the band
then when we comeback the orientation has been switched.
We can write this argument rigorously now. We define the Mobius surface (without
boundary) as the image of the map (recall figure 9.11):

[0, 2π ] × (−1, 1) → R3
 s s s
(s, t) 7→ (2 + t cos ) cos s, (2 + t cos ) sin s, t sin .
2 2 2

25.4 Theorem. The Mobius surface is not orientable.

Proof. Let M be the Mobius surface and let

ϕ1 : (0, 2π ) × (−1, 1) → M
 s s s
(s, t) 7→ (2 + t cos ) cos s, (2 + t cos ) sin s, t sin .
2 2 2

We can check (though it is a bit tedious, as for polar coordinates) that this map is a local
parametrization of M. This parametrization misses a subset of M, namely the interval [1, 3]
on the x-axis. So we need one more parametrization to cover this part. We can take

ϕ2 : (−π, π ) × (−1, 1) → M
 s s s
(s, t) 7→ (2 + t cos ) cos s, (2 + t cos ) sin s, t sin .
2 2 2

This ϕ2 is given by the same formula as ϕ1 , but on a different domain. This parametrization
misses the subset {−2} × {0} × [−1, 1] of M.
Suppose that M is orientable. Take an orientation for M. Then either ϕ1 agrees with this
orientation or disagrees with this orientation over the entire connected domain of ϕ1 . The
same is true for ϕ2 . That implies that ϕ1 and ϕ2 either induce the same orientations over
their entire domains, or they induces the opposite orientations over their domains.
25. ORIENTATION 171

Calculating directly, we get the normal vector given by ϕ1 at the point ϕ1 (s, 0) on the
center circle is:
 s s s
( ϕ1 )s × ( ϕ1 )t (s, 0) = 2 cos s sin , 2 sin s sin , −2 cos .
2 2 2

The normal vector given by ϕ2 at the point ϕ2 (s, 0) on the center circle is by the same
formula:  s s s
( ϕ2 )s × ( ϕ2 )t (s, 0) = 2 cos s sin , 2 sin s sin , −2 cos .
2 2 2
At the point (0, 2, 0) = ϕ1 ( π2 , 0) = ϕ2 ( π2 , 0) the two normal vectors agree, but at (0, −2, 0) =
ϕ1 ( 3π π
2 , 0) = ϕ2 (− 2 , 0) they are opposite. Thus ϕ1 and ϕ2 do not give the same orientation,
a contradiction.

Orientation on the boundary of an oriented manifold

The canonical orientation at every point of Hm is the canonical positive orientation of Rm .
The boundary ∂Hm is oriented as follows. At each point x ∈ ∂Hm the tangent space
Tx (∂Hm ) = Rm−1 × {0} is oriented by a basis b = (b1 , b2 , . . . , bm−1 ) such that the ordered
set
(−em , b1 , b2 , . . . , bm−1 )

is a positive basis of Rm . This is called the outer normal first orientation of the boundary.
R

Hm

∂Hm
− em

Lemma. This construction does not depend on the choice of representative basis b.

Proof. Suppose c = (c1 , . . . , cm−1 ) is another basis of Tx (∂Hm ) such that the change of bases
matrix A from basis b to basis c has positive determinant. Then the change of bases matrix
from basis (−em , b1 , . . . , bm−1 ) to basis (−em , c1 , . . . , cm−1 ) is
!
1 0
,
0 A

whose determinant is equal to det A > 0. Thus c induced the same orientation on Tx (∂Hm )
as b.

25.5 Example (∂Hm = (−1)m Rm−1 ). Take the canonical basis (e1 , . . . , em−1 ) of Rm−1 , and
consider the basis (i (e1 ), . . . , i (em−1 )) of Rm−1 × {0} where i is the inclusion map Rm−1 ,→
Rm−1 × {0}, x 7→ ( x, 0). We have

det(−em , i (e1 ), . . . , i (em−1 )) = (−1)m .

From this people often say that the boundary orientation of ∂Hm coincides with the canon-
ical orientation of Rm−1 if m is even and is opposite if m is odd, although this is an abuse
of language.
172

Suppose that M is an oriented manifold with boundary. The boundary of M receives

a special induced boundary orientation as follows. Suppose that ϕ : Hm → M is an
orientation-preserving local parametrization of M at the point ϕ(u) = x ∈ ∂M. Then
ϕ|∂Hm : ∂Hm → ∂M is a local parametrization of ∂M at ϕ(u). The derivative dϕ(u) : Rm →
Tx M has a restriction dϕ(u) : Tx ∂Hm → Tx ∂M. Let b be a basis of Tx ∂Hm giving the bound-
ary orientation of ∂Hm . The basis dϕ(u)(b) gives an orientation of Tx ∂M. This is called the
boundary orientation of ∂M associated to the orientation of M.
Tx ∂M

n( x )
x
dϕ(u)(−em )
M

Hm ϕ
∂M
b
∂Hm
− em

Lemma. This orientation does not depend on the choice of local parametrization.

Proof. Suppose that ψ : Hm → M is another orientation-preserving local parametriza-

tion of M at the point ψ(0) = x0 ∈ ∂M. Writing ψ = ϕ ◦ ( ϕ−1 ◦ ψ) we get dψ(v)(b) =
dϕ(u)(d( ϕ−1 ◦ ψ)(v)(b)). We need to check that the vector d( ϕ−1 ◦ ψ)(v)(b) is in the same
orientation as b, i.e. ϕ−1 ◦ ψ is orientation-preserving on ∂Hm . So this question is reduced
to: Given an orientation-preserving ϕ : Hm → Hm , show that ϕ̃ = ϕ |∂Hm : ∂Hm → ∂Hm
is orientation-preserving. Let u ∈ ∂Hm . The change of bases matrix from (e1 , . . . , em ) to
(dϕ(u)(e1 ), . . . , dϕ(u)(em )) is J ϕ (u), having positive determinant. Noting that dϕ(u)(ei ) =
d ϕ̃(u)(ei ) for 1 ≤ i ≤ m − 1, we have
!
A
J ϕ (u) = ∂ϕm ,
0 ∂um ( u )

where A is the change of bases matrix from (e1 , . . . , em−1 ) to (d ϕ̃(u)(e1 ), . . . , d ϕ̃(u)(em−1 )).
∂ϕm
Now we have det J ϕ (u) = ∂um ( u ) det A > 0. We write

∂ϕm ϕm (u + tem ) − ϕm (u) ϕm (u + tem )

(u) = lim = lim ,
∂um t →0+ t t →0+ t

noticing that since ϕ brings u to a point in ∂Hm we must have ϕm (u) = 0. Since u + tem ∈
int(Hm ) when t > 0 we have ϕ(u + tem ) ∈ int(Hm ), so ϕm (u + tem ) > 0. Thus
∂ϕm
∂um ( u ) ≥ 0.
∂ϕm
This implies actually that ∂um ( u ) > 0, and that det A > 0. Thus d ϕ̃(u) is orientation-
preserving.
∂ϕm
Notice that, since = dϕ(u)(em ) · em , the problem has been essentially reduced to
∂um ( u )
checking that the derivative dϕ(u) brings em to an inward pointing vector, i.e. dϕ(u)(em ) ·
em > 0.

We say that a vector v ∈ Rm is called inward pointing relative to Hm if v · em > 0,

25. ORIENTATION 173

and is outward pointing relative to Hm if v · em < 0. At x ∈ ∂M, a vector w ∈ Tx M is

called outward pointing if there is an orientation-preserving local parametrization ϕ such
that x = ϕ(u) and w = dϕ(u)(v) where v ∈ Rm is outward pointing relative to Hm . As
observed in the last paragraph of the above proof, this notion does not depend on the choice
of parametrization.
In the Euclidean inner product of Rk , there is a unique unit vector n( x ) ∈ Tx M such
that n( x ) ⊥ Tx ∂M and n( x ) is outward pointing. This vector n( x ) is called the outer unit
normal vector of the boundary.
Now we can describe the boundary orientation of ∂M as the one represented by a basis
c( x ) of Tx ∂M such that (n( x ), c( x )) represents the orientation of Tx M, in similarity to the
boundary orientation of ∂Hm . Therefore this orientation of the boundary is called the outer
normal first orientation of the boundary.

Example. The closed disk D2 has the orientation of R2 . Its boundary S1 receives the in-
duced outer normal first orientation, which is counter-clockwise.
The sphere S2 receives the boundary orientation from the ball D3 . As such the sphere
is oriented counter-clockwise viewed from outside the ball.

Problems
25.6. Show that two diffeomorphic manifolds are are either both orientable or both unorientable. For
example, this implies that the Mobius band and the cylinder are not diffeomorphic.

25.7. Suppose that f : M → N is a diffeomorphism of connected oriented manifolds with bound-

aries. Show that if there is an x such that d f x : TMx → TN f ( x) is orientation-preserving then f is
orientation-preserving.

25.8. Let f : Rk → Rl be smooth and let a be a regular value of f . Show that f −1 ( a) is an orientable
manifold.

25.9. Consider the map −id : Sn → Sn with x 7→ − x. Show that −id is orientation-preserving if and
only if n is odd.

25.10. Discuss the boundary orientation for a 1-dimensional manifold.

25.11. Discuss the boundary orientation for the cylinder S1 × [0, 1].

25.12. A smooth isotopy is an isotopy (see 12.12) by diffeomorphisms. Namely two diffeomorphisms
f and g from smooth manifold M to smooth manifold N are smoothly isotopic if there is a smooth
map

F : M × [0, 1] → N
( x, t) 7→ F ( x, t)

such that F ( x, 0) = f ( x ) and F ( x, 1) = g( x ) for all x ∈ M, and for each t ∈ [0, 1] the map F (·, t) is
a diffeomorphism. Show that two isotopic diffeomorphisms must be both orientation-preserving or
both orientation-reversing.
174

26 Topological degrees of maps

Let M and N be boundaryless, oriented manifolds of the same dimensions m. Further
suppose that M is compact.
Let f : M → N be smooth. Suppose that x is a regular point of f . Then d f x is an
isomorphism from TMx to TN f ( x) . Let sign(d f x ) = 1 if d f x preserves orientations, and
sign(d f x ) = −1 otherwise.
For any regular value y of f , let

deg( f , y) = ∑ sign(d f x ).
x ∈ f −1 ( y )

Notice that the set f −1 (y) is finite because M is compact (see 21.2).
This number deg( f , y) is called the Brouwer degree 15 or topological degree of the map
f with respect to the regular value y.
From the Inverse Function Theorem 21.16, each regular value y has a neighborhood V
and each preimage x of y has a neighborhood Ux on which f is a diffeomorphism onto V,
either preserving or reversing orientation. Therefore we can interpret that deg( f , y) counts
the algebraic number of times the function f covers the value y.

Example. Consider f : R → R, f ( x ) = x2 . Then deg( f , 1) = 0. This could be explained

geometrically from the graph of f , as f covers the value 1 twice in opposite directions at
x = −1 and x = 1.

Example. Consider f ( x ) = x3 − x with the regular value 0. From the graph of f we see
that f covers the value 0 three times in positive direction at x = −1 and x = 1 and negative
direction at x = 0, therefore we see right away that deg( f , 0) = 1.
On the other hand, if we consider the regular value −1 then f covers this value only
once in positive direction, thus deg( f , 1) = 1.

Homotopy invariance
In this section we will show that the Brouwer degree does not depend on the choice of
regular values and is invariant under smooth homotopy.

Lemma. Let M be the boundary of a compact oriented manifold X, oriented as the boundary of X.
If f : M → N extends to a smooth map F : X → N then deg( f , y) = 0 for every regular value y
of f .

Proof. (a) Assume that y is a regular value of F. Then F −1 (y) is a 1-dimensional manifold
of dimension 1 whose boundary is F −1 (y) ∩ M = f −1 (y), by Theorem 24.1.
By the Classification of one-dimensional manifolds 24.4, F −1 (y) is the disjoint union of
arcs and circles. Let A be a component that intersects M. Then A is an arc with boundary
{ a, b} ⊂ M.
We will show that sign(det(d f a )) = − sign(det(d f b )). Taking sum over all arc compo-
nents of F −1 (y) would give us deg( f , y) = 0.
Let x ∈ A. Recall that TA x is the kernel of dFx : TXx → TNy . Choose an orientation
for A as follows. Let (v1 , v2 , v3 , . . . , vn+1 ) be a positive oriented basis for TXx such that
15 L. E. J. Brouwer (1881–1966) is a Dutch mathematician. He had many important contributions in the early

development of topology, and founded Intuitionism.

26. TOPOLOGICAL DEGREES OF MAPS 175

v1 ∈ TA x . At x = a or x = b the vectors v2 , v3 , . . . , vn+1 belong to TMx . The linear map

dFx is bijective on the vector subspace generated v2 , v3 , . . . , vn+1 . We define v1 to give the
positive direction of TA x if (dFx (v2 ), . . . , dFx (vn+1 )) gives the positive orientation of TNy .
It is possible to check that this gives a smooth orientation for A, using the definition of
orientation by local coordinates.

TMb
N

TMa

( v 2 , . . . , v n +1 )
f
y
a
b
v1 A

M X

At x = a or at x = b we have d f x = dFx | TMx . Therefore d f x is orientation-preserving on

TMx oriented by the basis (v2 , v3 , . . . , vn+1 ). The key observation is that at exactly one of
the two points x = a or x = b this orientation of TMx is opposite to the orientation of TMx
as the boundary of X. This implies that sign(det(d f a )) = − sign(det(d f b )). The reason is
a simple visual observation: if at a the orientation of TA a is pointing outward with respect
to X then at b the orientation of TAb is pointing inward, and vice versa. Indeed, since A
is a smooth arc it can be parametrized by a smooth path γ. Assume that γ(0) = a and
γ(1) = b. In local coordinates ϕ : Hn+1 → X, at a = ϕ(0) a corresponding path α in Hn+1
has α(0) = 0, and αn+1 (t) > 0 for t > 0, therefore α0n+1 (0) ≥ 0, and since α0 (0) ∈
/ ∂Hn+1
we have α0n+1 (0) > 0, thus α0 (0) points inward. Similarly α0 (1) points outward.
(b) Suppose now that y is not a regular value of F. From 21.11 there is a neighbor-
hood of y in the set of all regular values of f such that deg( f , z) does not change in this
neighborhood. By Sard theorem, let z be a regular value of F in this neighborhood, then
deg( f , z) = deg( F, z) = 0 by (a), and deg( f , z) = deg( f , y). Thus deg( f , y) = 0.

Lemma. If f is smoothly homotopic to g then deg( f , y) = deg( g, y) for any common regular value
y.

Proof. Let I = [0, 1] and X = M × I. Since f be homotopic to g there is a smooth map

F : X → N such that F ( x, 0) = f ( x ) and F ( x, 1) = g( x ).
The boundary of X is ( M × {0}) t ( M × {1}). Then F is an extension of the pair f , g
from ∂X to X, thus deg( F |∂X , y) = 0 by the above lemma.
Note that one of the two orientations of M × {0} or M × {1} as the boundary of X is
opposite to the orientation of M (this is essentially for the same reason as in the proof of
the above lemma). Therefore deg( F |∂X , y) = ±(deg( f , y) − deg( g, y)) = 0, so deg( f , y) =
deg( g, y).
176

26.1 Theorem. Let M and N be boundaryless, oriented manifolds of the same dimensions. Further
suppose that M is compact and N is connected. The Brouwer degree of a map from M to N does not
depend on the choice of regular values and is invariant under smooth homotopy.

Therefore from now on we will write deg( f ) instead of deg( f , y).

Proof. We have already shown that degree is invariant under homotopy.

Let y and z be two regular values for f : M → N. Since N is homogeneous, see 23.4,
there is a diffeomorphism h from N to N that carries y to z. Furthermore, examining the
proof of that result, we can see that h constructed there is isotopic to the identity.
Note that h preserves orientation. Indeed, there is a smooth isotopy F : N × [0, 1] → N
such that F0 = h and F1 = id. Let x ∈ N, and let ϕ : Rm → N be an orientation-preserving
parametrization of a neighborhood of x with ϕ(0) = x. Since dFt ( x ) ◦ dϕ0 : Rm × R is
smooth with respect to t, the sign of dFt ( x ) does not change with t.
As a consequence, deg( f , y) = deg(h ◦ f , h(y)).
Finally since h ◦ f is homotopic to id ◦ f , we have deg(h ◦ f , h(y)) = deg(id ◦ f , h(y)) =
deg( f , h(y)) = deg( f , z).

The above proof follows the exposition in [Mil97]. There are more advanced treatments,
in terms of transversal theory and intersection number, such as in [GP74, p. 100, 107].

Applications
Example. Let M be a compact, oriented and boundaryless manifold. Then the degree of
the identity map on M is 1. On the other hand the degree of a constant map on M is 0.
Therefore the identity map is not homotopic to a constant map.

26.2 Example (Proof of the Brouwer fixed point theorem via the Brouwer degree). We can
prove that D n+1 cannot retract to its boundary (this is 24.3 for the case of D n+1 ) as follows.
Suppose that there is such a retraction, a smooth map f : D n+1 → Sn that is the identity on
Sn . Define F : [0, 1] × Sn by F (t, x ) = f (tx ). Then F is a smooth homotopy from a constant
map to the identity map on the sphere. But these two maps have different degrees.

Theorem (The fundamental theorem of Algebra). Any non-constant polynomial with real
coefficients has at least one complex root.

Proof. Let p(z) = zn + a1 zn−1 + a2 zn−2 + · · · + an−1 z + an , with ai ∈ R, 1 ≤ i ≤ n. Suppose

that p has no root, that is, p(z) 6= 0 for all z ∈ C. As a consequence, an 6= 0.
For t ∈ [0, 1], let

qt (z) = (1 − t)n zn + a1 (1 − t)n−1 tzn−1 + · · · + an−1 (1 − t)tn−1 z + an tn .

Then qt (z) is continuous with respect to the pair (t, z). Notice that if t 6= 0 then qt (z) =
t p((1 − t)t−1 z), and q0 (z) = zn while q1 (z) = an .
n
q (z)
If we restrict z to the set {z ∈ C | |z| = 1} = S1 then qt (z) has no roots, so |qt (z)| is a
t
continuous homotopy of maps from S1 to itself, starting with the polynomial zn and ending
an
with the constant polynomial | an |
. But these two polynomials have different degrees, a
contradiction.

Example. Let v : S1 → R2 , v(( x, y)) = (−y, x ), then it is a nonzero (not zero anywhere)
tangent vector field on S1 .
Similarly we can find a nonzero tangent vector field on Sn with odd n.
26. TOPOLOGICAL DEGREES OF MAPS 177

26.3 Theorem (The Hairy Ball Theorem). If n is even then every smooth tangent vector field on
Sn has a zero.
v( x )
Proof. Suppose that v is a nonzero tangent smooth vector field on Sn . Let w( x ) = kv( x )k
,
then w is a unit smooth tangent vector field on Sn .
Notice that w( x ) is perpendicular to x. On the plane spanned by x and w( x ) we can
easily rotate vector x to vector − x. Precisely, let Ft ( x ) = cos(t) · x + sin(t) · w( x ) with
0 ≤ t ≤ π, then F is a homotopy on Sn from x to − x. But the degrees of these two maps are
different, see 26.13.

Problems
26.4. Find the topological degree of a polynomial on R. Notice that although the domain R is not
compact, the topological degree is well-defined for polynomial.

26.5. Let f : S1 → S1 , f (z) = zn , where n ∈ Z. We can also consider f as a vector-valued function

f : R2 → R2 , f ( x, y) = ( f 1 ( x, y), f 2 ( x, y)). Then f = f 1 + i f 2 .

(a) Recalling the notion of complex derivative and the Cauchy-Riemann condition, check that
det( J f z ) = | f 0 (z)|2 .

(b) Check that all values of f are regular.

(c) Check that deg( f , y) = n for all y ∈ S1 .

26.6. Show that deg( f , y) is locally constant on the subspace of all regular values of f .

26.7. What happens if we drop the condition that N is connected in Theorem 26.1? Where do we use
this condition?

26.8. Let M and N be oriented boundaryless manifolds, M is compact and N is connected. Let
f : M → N. Show that if deg( f ) 6= 0 then f is onto, i.e. the equation f ( x ) = y always has a solution.

26.9. Let ri : Sn → Sn be the reflection map

ri (( x1 , x2 , . . . , xi , . . . , xn+1 )) = ( x1 , x2 , . . . , − xi , . . . , xn+1 ).

Compute deg(ri ).

26.10. Let f : Sn → Sn be the map that interchanges two coordinates:

f (( x1 , x2 , . . . , xi , . . . , x j , . . . , xn+1 )) = ( x1 , x2 , . . . , x j , . . . , xi , . . . , xn+1 ).

Compute deg( f ).

f
26.11. Suppose that M, N, P are compact, oriented, connected, boundaryless m-manifolds. Let M →
g
N → P. Then deg( g ◦ f ) = deg( f ) deg( g).

26.12. Let M be a compact connected smooth manifold. Let f : M → M be smooth.

(a) Show that if f is bijective then deg f = ±1.

(b) Let f 2 = f ◦ f . Show that deg( f 2 ) ≥ 0.

26.13. Let r : Sn → Sn be the antipodal map

r (( x1 , x2 , . . . , xn+1 )) = (− x1 , − x2 , . . . , − xn+1 ).

Compute deg(r ).
178

26.14. Let f : S4 → S4 , f (( x1 , x2 , x3 , x4 , x5 )) = ( x2 , x4 , − x1 , x5 , − x3 ). Find deg( f ).

26.15. Find a map from S2 to itself of any given degree.

26.16. If f , g : Sn → Sn be smooth such that f ( x ) 6= − g( x ) for all x ∈ Sn then f is smoothly homotopic

to g.

26.17. Let f : M → Sn be smooth. Show that if dim( M) < n then f is homotopic to a constant map.

26.18 (Brouwer fixed point theorem for the sphere). Let f : Sn → Sn be smooth. If deg( f ) 6=
(−1)n+1 then f has a fixed point.

26.19. Show that any map of from Sn to Sn of odd degree carries a certain pair of antipodal points to
a pair of antipodal point.
27. INTEGRATION OF REAL FUNCTIONS 179

27 Integration of real functions

Let M be a smooth m-manifold in Rk and let f : M → R be smooth. The purpose of this

R
section is to construct the integral M f .
In the special case when M is an open subset of Rm it is clear that M f should be the
R

usual Lebesgue integral of f . For the integral be a real number, not ∞, we may impose that
either f has compact support or M is compact.
Now suppose that M is parametrized by a single parametrization ϕ : Rm → M. By the
same arguments in Calculus of several variables when we construct integrals over curves
and surfaces, what we need is the
 notion
 of the m-dimensional volume for parallelepiped
spanned by the m vectors in Rk ,
∂ϕ
∂ui .
1≤ i ≤ m

Volume of parallelepiped

We take this as the principle for the definition of m-dimensional measure in Rk : If P is an m-

dimensional vector subspace of Rk and F = ( f 1 , f 2 , . . . , f m ) is an orthonormal vector basis
for P under the Euclidean inner product, then the m-dimensional measure in P written the
basis F should be same as the m-dimensional measure in Rm written in the canonical basis
E = (e1 , e2 , . . . , em ). More precisely, the m-dimensional measure of a set D = { a1 f 1 + a2 f 2 +
· · · + am f m | ( a1 , a2 , . . . , an ) ∈ D 0 ⊂ Rm } should be equal to the m-dimensional Lebesgue
measure of D 0 . In other words, if T is the linear isometry from P to Rm bringing F to E,
then T preserves measure.
This principle implies, and is consistent with, the property that the m-dimensional
Lebesgue measure is invariant under orthonormal transformations of Rm .
Now we concentrate on the case of parallelepiped. A parallelepiped spanned by m
vectors (v1 , v2 , . . . , vm ) in P is the set [v1 , v2 , · · · , vm ] = {∑1≤i≤m ai vi | ai ∈ [0, 1]}. The linear
transformation T will bring this set to the subset of Rm given by {∑1≤i≤m ai T (vi ) | ai ∈ R}.
Thus now we should find the volume in the case of Rm .
In Rm , the change of variable formula for Lebesgue integration contains the important
property that if vi = ∑m
j=1 v ji e j then vol[ v1 , v2 , · · · , vm ] = | det( vij )1≤i ≤m,1≤ j≤m |. We have
the following expression:
  1/2
| det(vij )| = det (vij )T · (vij )
!!1/2
m
= det ∑ vki vkj
k =1

1/2
= det vi · v j ij
.

Example. When m = 2 we have had this formula derived by an elementary method:

q q
a, b) = | a|2 |b|2 (1 − cos2 ([
| a × b| = | a||b| sin ([ a, b)) = | a|2 |b|2 − h a, bi2
" !#1/2
a·a a·b
= det .
b·a b·b


1/2
Returning to the case of m vectors in Rk , now we have vol[v1 , v2 , · · · , vm ] = det T (vi ) · T (v j ) ij .
180

Notice that T preserves inner product, finally we obtain an answer:


1/2
vol[v1 , v2 , · · · , vm ] = det vi · v j ij .

Integration of real functions for single parametrizations

Now it is clear what we should do:

Definition. Suppose that M ⊂ Rk is parametrized by a single parametrization ϕ : Rm →

M. Let f : M → R. Let gij =
∂ϕ ∂ϕ
∂ui · ∂u j . We define

Z Z q
f = f ◦ϕ det( gij )ij .
M Rm

To guarantee that this integral exists as a real number we may require f to have compact
support.

Example. When m = 1 this is path integral, and when m = 2 this is surface integral, as
studied in Calculus.

Lemma. This definition is independence from the choice of parametrization.

Proof. Let ψ : Rm → M be another parametrization. Applying the change of variable

ϕ−1 ◦ ψ and using the change of variable formula in Lebesgue integration:
Z q Z  r 
f ◦ϕ det( gij ) = f ◦ ϕ det J ϕT J ϕ
Rm Rm
Z  q 
◦ ( ϕ−1 ◦ ψ)| det J ϕ−1 ◦ψ |


= f ◦ ϕ det J ϕ T J ϕ
Rm
Z r  
−1
( f ◦ ϕ) ◦ ( ϕ ◦ ψ) det J ϕ−1 ◦ψ T J ϕ T J ϕ ◦ ( ϕ−1 ◦ ψ) J ϕ−1 ◦ψ



=
Rm
s 
Z T  
= ( f ◦ ψ) det − 1
J ϕ ◦ ( ϕ ◦ ψ ) J ϕ −1 ◦ ψ − 1
J ϕ ◦ ( ϕ ◦ ψ ) J ϕ −1 ◦ ψ
Rm
Z q

= ( f ◦ ψ) det Jψ T Jψ .
Rm

Remark. In the language of measure theory, we may say that the measurable sets on M
where V is Lebesgue measurable on Rm , and the measure of such
consists of the sets ϕ(V )q
a set is vol( ϕ(V )) = V det( gij )dµ where µ is the Lebesgue measure of Rm .
R

Partition of Unity
When it is not possible to cover the manifold by a single parametrization, we will use par-
titions of unity.

27.1 Theorem (smooth partition of unity). Let M be a compact smooth manifold. Let O be any
open cover of M. There exists a family of functions ( f i )1≤i≤k such that f i : M → R is smooth,
0 ≤ f i ≤ 1, supp( f i ) = { x ∈ M | f i ( x ) 6= 0} ⊂ U for some U ∈ O, and ∑1≤i≤k f i = 1.

Proof. The proof follows the steps below:

27. INTEGRATION OF REAL FUNCTIONS 181

(a) For each x ∈ M, show that there are open sets Vx and Wx in M such that x ∈ Vx ⊂
V x ⊂ Wx ⊂ W x ⊂ U for some U ∈ O.

(b) The collection (Vx ) x∈ M is an open cover of M with a finite subcover (Vi )1≤i≤k . There is
a smooth function ϕi : M → R such that 0 ≤ ϕi ≤ 1, ϕi V ≡ 1 , supp( ϕi ) ⊂ W i ⊂ U

i
for some U ∈ O.
ϕi
(c) Let f i = . Then ( f i )1≤i≤k is the family of functions we are looking for.
∑ik=1 ϕi

Integration of real functions: general cases

Let ( pi )1≤i≤k be any finite partition of unity associated
 to any
 cover of M by coordinate
k k
neighborhoods. Since ∑i=1 pi = 1, we can write f = ∑i=1 pi f = ∑ik=1 pi f . Since integral
should be linear, we can propose:

Definition. Let M be a compact manifold. Let ( pi )1≤i≤n be any finite partition of unity
associated to any cover of M by coordinate neighborhoods. We define
Z n Z

M
f = ∑ pi f .
i =1 M

Lemma. The definition of integral is independence from the choice of partition of unity.

Proof. Let (q j )1≤ j≤l be any partition of unity associated to any cover of M by coordinate
neighborhoods. We have
Z Z Z Z
∑ M
pi f = ∑ M
(∑ q j )( pi f ) = ∑
M
∑ pi q j f =∑
M
pi q j f .
i i j i j ij

Similarly Z Z
∑ M
pj f = ∑ M
pi q j f .
j ij

Thus ∑i pi f = ∑ j
R R
M M pj f .

As a special case, we can define the volume of a compact manifold M ⊂ Rk to be

R
M 1.
Integration of real functions is a special case of a far-reaching theory of integration of
differential forms on manifolds. In this theory important results of differential and inte-
gral calculus such as Newton-Leibniz formula, Stokes formula are generalized to higher
dimensions. The reader can study this theory from many sources, such as [Spi65], [GP74],
[VuSt].
182

Guide for further reading

There are not many textbooks presenting differential topology to undergraduate students.
We have closely followed John Milnor’s lectures in [Mil97]. Other good textbooks include
[GP74], [Sas11], [Tu13].
For more advanced treatments, the book [DFN85] is a masterful presentation of modern
topology and geometry, with some enlightening explanations, but it sometimes requires
comprehensive knowledge. The book [Hir76] can serve as a technical reference for some
advanced topics. A more recent textbook [Lee13] supplies lot of detailed proofs.
Suggestions for some problems

1.8 There exists an infinite countable subset of A.

S∞
1.14 n=1 [ n, n + 1] = [1, ∞).

1.15 Use the idea of the Cantor diagonal argument in the proof of 1.3. In this case the issue of
different presentations of same real numbers does not appear.

1.16 Use the injective map g ◦ f .

1.19 For A ⊂ Z+ , if n ∈ A let an = 1, otherwise let an = 0. Consider the map A 7→ a = a1 a2 · · · an . . . .

Use 1.17 and 1.16.

1.20 Proof by contradiction.

2.19 Show that each ball in one metric contains a ball in the other metric with the same center.

3.16 Use 3.15 to construct a homeomorphism bringing {0} × [0, 1] to {0} × [ 21 , 1], and [0, 1] × {0} to
{0} × [0, 21 ].

3.21 See 2.21 and 3.1.

3.27 Let f : ∂D n → ∂D n be a homeomorphism. Consider F : D n → D n , F ( x ) = k x k f ( k xxk ) (radial

extension).

3.28 Compare the sub-interval [1, 2π ) and its image via ϕ.

4.17 Use the characterization of connected subspaces of the Euclidean line.

4.24 Let A be countable and x ∈ R2 \ A. There is a line passing through x that does not intersect A
(by an argument involving countability of sets).

4.30 Use 3.15 to modify each letter part by part. See 3.16. Use connectedness to distinguish spaces.

5.3 Let C be a countable subset of [0, Ω). The set is countable while the set [0, Ω) is un-
S
c∈C [0, c )
countable. This implies C is bounded from above.

5.13 Consider the set of all irrational numbers.

6.8 This is a special case of 6.2.

6.10 Use Lebesgue’s number.

6.13 See the proof of 6.1.

6.15 Use 6.13.

6.16 Use 6.15.

183
184

6.18 Let X be a compact metric space, and let I be an open cover of X. For each x ∈ X there is an open
set Ux ∈ I containing x. There is a number ex > 0 such that the ball B( x, 2ex ) is contained in
Ux . The collection { B( x, ex ) | x ∈ X } is an open cover of X, therefore there is a finite subcover
{ B( xi , ei ) | 1 ≤ i ≤ n}. Let e = min{ei | 1 ≤ i ≤ n}. Let x ∈ X and consider B( x, e). There is
an i0 , 1 ≤ i0 ≤ n, such that x ∈ B( xi0 , ei0 ). Then B( x, e) ⊂ B( xi0 , 2ei0 ) ⊂ Ui0 .

6.19 A bouquet of circles.

6.22 Use 6.21.

6.29 Use 6.13

6.30 Use 6.29.

6.32 (⇐) Use 6.16 and 5.10.

T∞
6.33 Use 6.16 to show that if Y = i =1 Xi is not connected then there are two disjoint sets U and V
which are open in X such that Y ⊂ U ∪ V, U ∩ Y 6= ∅, and V ∩ Y 6= ∅. Show that if O is an
open set of X containing Y then O contains Xn for some n, by using the nested sequence of
closed sets ( Xn \ U )n and 6.8. In the the case of path-connectedness, look at the Topologist’s
sine curve.

7.6 Look at their bases.

7.13 Only need to show that the projection of an element of the basis is open.

7.15 Use 7.2 to prove that the inclusion map is continuous.

7.16 Use 7.15.

7.17 Let ( xi ) and (yi ) be in ∏i∈ I Xi . Let γi be a continuous path from xi to yi . Let γ = (γi ).

7.18 (b) Use 7.15. (c) Fix a point x ∈ ∏i∈ I Xi . Use (b) to show that the set A x of points that differs
from x at at most finitely many coordinates is connected. Furthermore A x is dense in ∏i∈ I Xi .

7.19 Use 7.15. It is enough to prove for the case an open cover of X × Y by open sets of the form
a product of an open set in X with an open set in Y. For each “slice” { x } × Y there is finite
Tnx
subcover {Ux,i × Vx,i | 1 ≤ i ≤ n x }. Take Ux = i =1 Ux,i . The collection {Ux | x ∈ X } covers X
so there is a subcover {Ux j | 1 ≤ j ≤ n}. The collection {Ux j ,i × Vx j ,i | 1 ≤ i ≤ n x j , 1 ≤ j ≤ n}
is a finite subcover of X × Y.

7.21 Use 6.16.

7.28 (⇐) Use 6.16 and the Urysohn lemma 8.1.

8.7 Use 6.30 and 6.14 and the proof of Urysohn lemma.

8.14 Use 8.13.

8.8 See 7.1 and 7.3.

8.10 Use 6.29. Use a technique similar to the one in 7.19.

9.27 Cut the square by a suitable diagonal, then glue back the resulting two triangles at a different
pair of edges.

9.28 Cut one of the two squares by a suitable diagonal, then glue a different pair of edges of the
resulting two triangles.
27. INTEGRATION OF REAL FUNCTIONS 185

9.29 To give a rigorous argument we can simply describe the figure below.

B A0

C C0

A B0

C C0 C C0

A B0 B A0

 
The map from X = ([0, 1] × [0, 1]) \ [0, 1] × { 12 } to Y = [0, 2] × [0, 21 ) given by

( x, y), y < 12 ,
( x, y) 7→
( x + 1, 1 − y), y > 12 ,

is bijective and is continuous. The induced map to Y/(0, y) ∼ (2, y) is surjective and is contin-
uous. Then its induced map on X/(0, y) ∼ (1, 1 − y) is bijective and is continuous, hence is a
homeomorphism between X/(0, y) ∼ (1, 1 − y) and Y/(0, y) ∼ (2, y).

9.40 The idea is easy to be visualized in the cases n = 1 and n = 2. Let S+ = { x = ( x1 , x2 , . . . , xn+1 ) ∈
Sn | x1 ≥ 0}, the upper hemisphere. Let S0 = { x = ( x1 , x2 , . . . , xn+1 ) ∈ Sn | x1 = 0}, the
equator. Let f : Sn → S+ be given by f ( x ) = x if x ∈ S+ and f ( x ) = − x otherwise. Then the
following diagram is commutative:

f
Sn / S+
p◦ f
p
 f˜
( 
Sn /x ∼ − x / S+ /x ∼ − x, x ∈ S0

Then it is not difficult to show that S+ /x ∼ − x, x ∈ S0 is homeomorphic to RPn = D n /x ∼

− x, x ∈ ∂D n .

9.43 Use this diagram and 9.45:

[0, 1] /R

 ' 
[0, 1]/0∼1 / R/ x∼x+n, n∈Z

9.47 Examine the following diagram, and use 9.2 to check that the maps are continuous.

X / X/R1 / ( X/R1 )/ R̃2

4

 x
X/( R1 ∪ R2 ) q

10.16 One approach is by using ideas in 3.21. A generalization is in 17.16.

186

11.22 Deleting an open disk is the same as deleting the interior of a triangle.

11.25 Count edges from of the set of triangles. Count vertices from the set of triangles. Count edges
from the set of vertices, notice that each vertex belongs to at most (v − 1) edges.

14.9 Use the fact that the torus T 2 is homeomorphic to S1 × S1 (9.21).

16.6 Let X be the simplicial complex, A be the set of vertices path-connected to the vertex v0 by
edges, U be the union of simplexes containing vertices in A. Let B be the set of vertices not
path-connected to the vertex v0 by edges, V be the union of simplexes containing vertices in B.
Then U and V are closed and disjoint.

15.4 Take a rectangle containing the set of points to be deleted. The plane has a deformation re-
traction to that rectangle. Divide the rectangle into finitely many subrectangles such that each
subrectangle contains at most one point to be deleted, which is in the interior. A rectangle with
an interior point deleted has a deformation retraction to its boundary. The new space has a de-
formation retraction to a bouquet of circles, for example by first taking a homotopy collapsing
the all the vertical edges of the subrectangles.

15.7 Consider the dihedral group D3 . In particular

√ !
consider the subgroup of the group of invertible
!
− 21 − 23 2π −1 0
2 × 2 matrices generated by √ (a rotation of an angle 3 ) and (a
− 23 − 12 0 1
reflection).

16.7 Use problem 16.6.

17.7 First take a deformation retraction to a sphere.

17.8 Show that R3 \ S1 is homotopic to Y which is a closed ball minus a circle inside. Show that
Y = S1 ∨ S2 , see [Hat01, p. 46]. Or write Y as a union of two halves, each of which is a closed
ball minus a straight line, and use the Van Kampen theorem.

17.10 Use Mayer-Vietoris sequence.

17.16 Let S be a convex compact subset of Rn . Suppose that p0 ∈ IntS. For x ∈ Sn−1 , let f ( x ) be the
intersection of the ray from p0 in the direction of x with the boundary of S, namely let t x =
max {t ∈ R+ | p0 + tx ∈ S} and let f ( x ) = p0 + t x x. Notice that the straight segment [ p0 , f ( x )]
has only f ( x ) as a boundary point of S, since if we take B( p0 , e) ⊂ S then the convex hull
of the set { f ( x )} ∪ B( p0 , e) contains [ p0 , f ( x )) in the interior. Check that the map g : D n → S
sending the straight segment [0, x ] linearly to the straight segment [ p0 , f ( x )] is well-defined and
is bijective. To check that g is a homeomorphism it might be easier to consider the continuity
of the inverse map. Also f : Sn−1 → ∂S is a homeomorphism.

19.9 Consider a neighborhood of (0, 0). Any continuous path from a point ( x1 , y1 ) to a point ( x2 , y2 ),
where x1 < 0, x2 > 0 must pass through (0, 0).

19.14 See 4.3.

19.17 Use the Implicit function theorem.

20.9 A closed bounded subset of Rn is compact. Consider any smooth path passing through q. Dif-
ferentiate the square of the distance function.

21.16 Consider
f
MO /N
O
ϕ ψ
ψ −1 ◦ f ◦ ϕ
U /V
27. INTEGRATION OF REAL FUNCTIONS 187

Since d f x is surjective, it is bijective. Then d(ψ−1 ◦ f ◦ ϕ)u = dψ− 1

f (x)
◦ d f x ◦ dϕu is an isomor-
phism. The Inverse function theorem can be applied to ψ−1 ◦ f ◦ ϕ.

21.11 Each x ∈ f −1 (y) has a neighborhood Ux on which f is a diffeomorphism. Let V = [ x∈ f −1 (y) f (Ux )] \
T

f ( M \ x∈ f −1 (y) Ux ). Consider V ∩ S.
S

21.21 Use problem 21.15.

21.18 We only need to prove that ϕ−1 is continuous. Let ψ be a local parametrization of a neigh-
borhood of x = ϕ(u) ∈ M. Since ϕ is continuous the set U = ϕ−1 ( ϕ(Rm ) ∩ ψ(Rm )) is open,
non-empty in Rm . The function ψ−1 ◦ ϕ is defined on U. Apply the Inverse function theorem

−1
to ψ−1 ◦ ϕ there are open sets V and W of Rm such that the inverse map ψ−1 ◦ ϕ = ϕ −1 ◦ ψ
is continuous from W onto V. This implies ϕ−1 = ( ϕ−1 ◦ ψ) ◦ ψ−1 is continuous on ψ(W ), an
open neighborhood of x in M.

22.6 Use 21.15, or the following direct argument. Let α be a smooth path in f −1 (c) through x = α(0)
then for all t we have f (α(t)) = a, implying ∇ f (α(t)) · α0 (t) = 0, thus ∇ f (α(t)) ⊥ α0 (t),
in particular ∇ f ( x ) ⊥ α0 (0). Since α0 (0) is an arbitrary tangent vector of f −1 (c) we have
∇ f ( x ) ⊥ Tx f −1 (c).

23.6 Show that f (n) ( x ) = e−1/x Pn (1/x ) for x > 0, where Pn is a polynomial.

23.7 Cover A by finitely many balls Bi ⊂ U. For each i there is a smooth function ϕi which is positive
in Bi and is zero outside of Bi .

∂F
25.12 det ∂x ( x, t ) is continuous with respect to t.

26.16 Note that f ( x ) and g( x ) will not be antipodal points. Use the homotopy

(1 − t) f ( x ) + tg( x )
Ft ( x ) = .
k(1 − t) f ( x ) + tg( x )k

26.18 If f does not have a fixed point then f will be homotopic to the reflection map.

26.17 Using Sard Theorem show that f cannot be onto.

188
Bibliography

[AF08] Colin Adams and David Fransoza, Introduction to topology: Pure and applied, Pren-
tice Hall, 2008.

[Ams83] M. A. Amstrong, Basic topology, Springer, 1983.

[BBI01] Dmitri Burago, Yuri Burago, and Sergei Ivanov, A course in metric geometry,
American Mathematical Society, 2001.

[Bre93] Glen E. Bredon, Topology and Geometry, Graduate Texts in Mathematics, vol. 139,
Springer, 1993.

[Cai94] George L. Cain, Introduction to general topology, Addison-Wesley, 1994.

[Con90] John B. Conway, A course in functional analysis, Springer-Verlag, 1990.

[Cro78] Fred H. Croom, Basic concepts of algebraic topology, Springer-Verlag, 1978.

[dC76] Manfredo do Carmo, Differential geometry of curves and surfaces, Prentice-Hall,

1976.

[DFN85] Boris A. Dubrovin, Anatoly T. Fomenko, and Sergey P. Novikov, Modern Geome-
try–Methods and Applications, Part II, Springer-Verlag, 1985.

[DFN90] Boris A. Dubrovin, Anatoly T. Fomenko, and Sergey P. Novikov, Modern Geome-
try–Methods and Applications, Part III, Springer-Verlag, 1990.

[Duc05] Dương Minh Đức, Giải tích hàm, Nhà xuất bản Đại học Quốc gia Thành phố Hồ
Chí Minh, 2005.

[Dug66] James Dugundji, Topology, Allyn and Bacon, 1966.

[EH10] Herbert Edelsbrunner and John L Harer, Computational topology: an introduction,

AMS, 2010.

[End77] Herbert B. Enderton, Elements of set theory, Academic Press, 1977.

[Eng89] Ryszard Engelking, General topology, Heldermann Verlag, 1989.

[Gal10] Joseph A. Gallian, Contemporary abstract algebra, Brooks/Cole, 2010.

[GP74] Victor Guillemin and Alan Pollack, Differential topology, Prentice-Hall, 1974.

[Hat01] Allen Hatcher, Algebraic Topology, Cambridge University Press, 2001, available at
http://www.math.cornell.edu/~hatcher.

[Hir76] Morris W. Hirsch, Differential Topology, Springer, 1976.

189
190 BIBLIOGRAPHY

[HS74] Morris W. Hirsch and Stephen Smale, Differential equations, dynamical systems,
and linear algebra, Academic Press, 1974.

[Hun74] Thomas W. Hungerford, Algebra, Holt, Rinehart and Winston, Inc., 1974.

[HY61] John G. Hocking and Gail S. Young, Topology, Dover, 1961.

[J84] Klaus Jänich, Topology, Undergraduate Texts in Mathematics, Springer-Verlag,

1984.

[Kel55] John L. Kelley, General topology, Van Nostrand, 1955.

[KF75] A. N. Kolmogorov and S. V. Fomin, Introductory real analysis, Dover Publications

Inc., 1975.

[KMM04] Tomasz Kaczynski, Konstantin Mischaikow, and Marian Mrozek, Computational

homology, Springer, 2004.

[Lan97] Serge Lang, Undergraduate analysis, 2nd ed., Springer, 1997.

[Lee11] John M. Lee, Introduction to topological manifolds, 2nd ed., Springer, 2011.

[Lee13] John M. Lee, Introduction to smooth manifolds, 2nd ed., Springer, 2013.

[Mas91] William S. Massey, A basic course in algebraic topology, Springer, 1991.

[Mil97] John Milnor, Topology from the differentiable viewpoint, Princeton University Press,
1997.

[Moi77] Edwin E. Moise, Geometric topology in dimensions 2 and 3, Springer-Verlag, 1977.

[MT01] Bojan Mohar and Carsten Thomassen, Graphs on surfaces, JHU Press, 2001.

[Mun84] James Munkres, Elements of algebraic topology, Addison-Wesley, 1984.

[Mun91] James Munkres, Analysis on manifolds, Addison-Wesley, 1991.

[Mun00] James Munkres, Topology a first course, 2nd ed., Prentice-Hall, 2000.

[Ros99] Dennis Roseman, Elementary topology, Prentice-Hall, 1999.

[Rud86] Walter Rudin, Real and complex analysis, 3rd ed., McGraw Hill, 1986.

[Sas11] Anant R. Sastri, Elements of Differential Topology, CRC Press, 2011.

[SJ70] Lynn A. Steen and J. Arthur Seebach Jr., Counterexamples in topology, Holt, Rine-
hart and Winston, 1970.

[Spi65] Michael Spivak, Calculus on manifolds, Addison-Wesley, 1965.

[Spi99] Michael Spivak, A comprehensive introduction to differential geometry, vol. 2, Pub-

lish or Perish, 1999.

[Tu13] Loring W. Tu, Introduction to manifolds, Springer, 2nd ed., 2013.

[Vas01] V. A. Vassiliev, Introduction to Topology, American Mathematical Society, 2001.

[Vic94] James W. Vick, Homology theory: an introduction to algebraic topology, 2nd ed.,
Springer-Verlag, 1994.
BIBLIOGRAPHY 191

[VINK08] O. Ya. Viro, O. A. Ivanov, N. Yu. Netsvetaev, and V. M. Kharlamov, Elementary

topology problem textbook, preprint, 2008.

[Vugt3] Huỳnh Quang Vũ, Bài giảng Giải tích 3, Đại học Khoa học Tự nhiên Thành phố
Hồ Chí Minh, http://www.math.hcmus.edu.vn/~hqvu/gt3.pdf.

[VuSt] Huỳnh Quang Vũ, Công thức Stokes, Đại học Khoa học Tự nhiên Thành phố Hồ
Chí Minh, http://www.math.hcmus.edu.vn/~hqvu/Stokes.pdf.
Index

2S , 6 continuous map, 18
contractible space, 96
Alexander trick, 96 convergent, 35
atlas, 133 covering map, 104
attaching cells, 82 covering space, 104
Axiom of choice, 9 critical point, 142
non-degenerate, 149
basis, 14
critical value, 142
Borsuk–Ulam theorem, 27
CW-complex, 82
boundary, 14
boundary orientation, 172 deformation retract, 95
bouquet of circles, 67 deformation retraction, 95
Brouwer degree, 174 dense
Brouwer fixed point theorem, 107, 124, 165 subspace, 37
derivative, 139
Cantor diagonal argument, 8
diagram
Cantor set, 10
commutative, 62
Cartesian product, 10
diffeomorphic, 132
cell, 82
diffeomorphism, 132
cell complex, 82
disjoint union, 52
skeleton, 83
disk, 82
chain, 115
boundary, 117 embedding, 21
closed, 117 equivalence
cycle, 117 homotopy, 95
exact, 117 Euler characteristic, 83
chain complex, 117
exact, 117 filter, 38
chart, 133 filter-base, 38
finite intersection property, 41
Classification of compact one-dimensional man-
ifolds, 165 first partial derivative, 147
closure, 14 flow, 155, 156
collection of all subsets, 6 complete, 157
compact free product, 110
locally, 43, 45 fundamental group, 100
compactification, 42
gradient vector, 131
Alexandroff, 42
graph, 51
one-point, 42
connected Hairy Ball Theorem, 177
simply, 103 Hausdorff distance, 38

192
INDEX 193

Hilbert cube, 48 closed, 22

homeomorphism, 20 derivative, 131
homogeneous, 23 discrete, 30
homology homotopic, 94
simplicial, 117 open, 22
singular, 121 smooth, 131, 132
homology group P ( S ), 6
relative, 125 Mayer-Vietoris sequence, 123
homotopy, 94 metric
hypersurface, 168 equivalent, 17
metrizable, 33
imbedding, 21
Mobius band, 65, 170
immersion, 66
interior, 14 neighborhood, 13
Invariance of dimension, 80 net, 35
invariance of domain, 127 universal, 74
smooth, 134 norm, 13
invariant
homotopy, 102 order
topological, 26, 84 dictionary, 6
inward pointing, 172 lower bound, 6
isometry, 23 minimal, 6
isotopy, 96 smallest, 6
smooth, 173 total, 6
orientation
Jacobian matrix, 131 boundary, 172
Jordan curve theorem, 127 outer normal first orientation of the bound-
ary, 171
Klein bottle, 66
outer unit normal vector of the boundary,
knot, 72
173
figure-8, 72
outward pointing, 173
trefoil, 72
partion of unity, 56
Lebesgue’s number, 41
partition of unity, 55
Lie group, 144
smooth, 180
lift of path, 104
path, 27, 98
local coordinate, 133
composition, 98
local parametrization, 133
homotopy, 98
loop, 98
inverse, 98
manifold Peano curve, 75
homogeneous, 158 Poincaré conjecture, 127
orientable, 167 point
orientation, 167 contact, 14
smooth, 133 interior, 14
submanifold, 133 limit, 14
topological, 79 point-wise convergence topology, 56, 59
with boundary, 162 polyhedron, 81
map pre-compact, 42
194 INDEX

projective plane, 66 support, 55

projective space, 66 surface, 86
connected sum, 86
regular point, 142
fundamental polygon, 87
regular value, 142
genus, 86
relation, 5
non-orientable, 86
equivalence, 5
orientable, 86
minimal equivalence, 73
two-sided, 169
retract, 95, 165
retraction, 95, 165 tangent vector, 137
test function, 160
Sard Theorem, 164
The fundamental theorem of Algebra, 176
second partial derivatives, 148
Tiestze extension theorem, 58
separation axioms, 33
topological degree, 125, 174
set
topological group, 46
closed, 13
topological invariant, 84
countable, 6
topological space
directed, 35
connected, 25
equivalence, 6
connected component, 25
indexed collection, 5
topological vector space, 52
indexed family, 5
Topologist’s sine curve, 29
open, 13
topology
ordered, 6
coarser, 15
well-ordered, 11
compact-open, 57
simplex, 80
countable complement, 16
standard, 81
discrete, 13
simplicial complex, 81
Euclidean, 13
singular chain, 120
finer, 15
singular simplex, 120
finite complement, 16
Sorgenfrey’s line, 17
generated by subsets, 15
space
ordering, 16
completely regular, 50
product, 47
first countable, 36
quotient, 61
Hausdorff, 33
relative, 19
homeomorphic, 20
subspace, 19
homotopic, 94
trivial, 13
Niemytzki, 38
Zariski, 53
normal, 33
torus, 63
quotient, 61
total boundedness, 42
regular, 33
triangulation, 81
subspace, 19
tangent, 137 Urysohn lemma, 54
Space filling curve, 75 manifold, 161
sphere, 20 smooth, 160
star-shaped, 95 Urysohn Metrizability Theorem, 74
stereographic projection, 21
Stone-Cech compactification, 50 vector field, 155
subbasis, 14 volume, 181
INDEX 195

wedge sum, 66
well-ordering property, 7

Zorn lemma, 9