Clark Barwick

LECTURE NOTES FOR TOPOLOGY I
Clark Barwick
Department of Mathematics, Harvard University, One Oxford Street, Cambridge, MA 02138, USA.
E-mail : clarkbar@gmail.com
LECTURE NOTES FOR TOPOLOGY I
Clark Barwick
CONTENTS
Act I. General topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1. The category of topological spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1. Topological spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2. First examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3. Continuous maps and homeomorphisms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2. Constructing topologies and continuous maps. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1. Quotients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2. Coverings and locality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3. Compactness I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1. Compact spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2. The Tychonoff Product Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3. The compact-open topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4. Compactness II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1. Forms of compactness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2. Locally compact Hausdorff spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5. Countability and separation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.1. Taxonomy of separation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2. Normality, Urysohn’s lemma, and the Tietze extension theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6. Metrization theorems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.1. First and second countability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2. Urysohn’s metrization theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3. The Bing–Nagata–Smirnov metrization theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Entr’Acte. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7. Sheaf theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.1. Presheaves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2. Stalks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.3. Sheaves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.4. The espace étalé and sheafification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.5. Direct and inverse images of sheaves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Act II. An introduction to homotopy theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
8. Covering space theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
8.1. Connectedness and π
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6 CONTENTS
8.2. Covering spaces and locally constant sheaves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
8.3. The étale fundamental group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
8.4. Homotopy classes of loops and lifts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
8.5. Universal covering spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
9. Computing the étale fundamental group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
9.1. The étale fundamental group of S
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
9.2. Seifert-van Kampen Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
9.3. Simple connectedness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10. Homotopy theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.1. Paths and π
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.2. Homotopies of maps. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.3. The homotopy-theoretic fundamental group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.4. Homotopy invariance of the étale fundamental group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
11. The comparison theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
11.1. The two fundamental groups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
ACT I
GENERAL TOPOLOGY
CHAPTER 1
THE CATEGORY OF TOPOLOGICAL SPACES
1.1. Topological spaces
Notation 1.1.1. — Given a set S, we denote by ´(S) the power set of S, i.e., the set of all subsets of S. Any map
f : T

S induces a map
f
−1
: ´(S)

´(T) ,
where for any subset A⊂ S,
f
−1
A:=¦t ∈ T | f (t ) ∈ A] ⊂T.
One may verify that f
−1
respects arbitrary intersections and arbitrary unions: for any subset · ⊂´(S),
f
−1

¸
¸
¸
A∈·
A

=
¸
A∈·
f
−1
A and f
−1

¸
¸
¸
A∈·
A

=
¸
A∈·
f
−1
A;
moreover, f
−1
respects complements: for any set A∈ ´(S), one may verify that
f
−1
(
S
A) =
T
( f
−1
A).
Definition 1.1.2. — A topological space (or more simply a space) is a pair (X, Op(X)) comprised of a set X and a
subset Op(X) ⊂´(X) satisfying the following properties.
(1.1.2.1) For any finite subset ¹ ⊂Op(X), the intersection
¸
U∈¹
U ∈ Op(X).
(1.1.2.2) For any subset ¹
/
⊂Op(X), the union
¸
U
/
∈¹
/
U
/
∈ Op(X).
In this case, the set Op(X) is said to be a topology on X. The elements U ∈ Op(X) will be called open in X. A subset
V ∈ ´(X) will be called closed if its complement V :=X −V is an open set.
1.1.3. — In the definition above, note that when ¹ = ∅, condition (1.1.2.1) translates to the condition that X ∈
Op(X). Similarly, when ¹
/
=∅, condition (1.1.2.2) translates to the condition that ∅∈ Op(X).
1.1.4. — We have described a topology on a set X by prescribing the set Op(X) of open sets; we may also describe
a topology on a set X by prescribing the set (l(X) of closed sets.
A topology on a set X is, equivalently, a subset (l(X) ⊂´(X) satisfying the following axioms.
(1.1.4.1) For any subset · ⊂(l(X), the intersection
¸
V∈·
V ∈ (l(X).
10 CHAPTER 1. THE CATEGORY OF TOPOLOGICAL SPACES
(1.1.4.2) For any finite subset ·
/
⊂(l(X), the union
¸
V
/
∈·
/
V
/
∈ (l(X).
As a result, we sometimes refer to the pair (X, (l(X)) as a topological space. In practice, it will always be clear
whether we are describing the set of open subsets of X or the set of closed subsets of X.
Definition 1.1.5. — Suppose X a set, and suppose Op(X) and Op
/
(X) two topologies on X. If Op
/
(X) ⊂ Op(X),
then we say that Op(X) is finer than Op
/
(X) and that Op
/
(X) is coarser than Op(X).
Definition 1.1.6. — Suppose (X, Op(X)) a topological space. Then there are many different kinds of subsets of X.
(1.1.6.1) For any point x ∈ X, an open neighborhood of x is an open set U ∈ Op(X) containing x. A neighborhood
of x is a subset V ⊂X containing an open neighborhood of x.
(1.1.6.2) For any subset A⊂X, the closure of Ais the set
A:=¦x ∈ X | for any neighborhood V of x, V ∩A(=∅].
In other words, the points of A are those points x ∈ X with the property that every neighborhood of x
meets A.
(1.1.6.3) Consider two subsets A⊂B ⊂X. We say that Ais dense in B if A∩B =B.
(1.1.6.4) Dual to the closure is the interior. For any subset A⊂X, the interior of Ais the set
A

:=¦x ∈ A| there exists a neighborhood V of x such that V ⊂A].
In other words, the points of A

are those points x ∈ A with the property that there is a neighborhood of
x contained in A.
Exercise 1. — Suppose (X, Op(X)) a topological space. Suppose that A⊂X is a subset. Prove that the closure A⊂X
can be characterized in the following two ways.
(1.1) The closure Ais the set
A:=¦x ∈ X | for any open neighborhood V of x, V ∩A(=∅].
(1.2) The closure Ais the smallest closed subset containing A.
(1.3) The closure Ais the intersection of all the closed subsets containing A:
A=
¸
V∈(l(X),A⊂V
V.
Dually, prove that the interior A

⊂X can be characterized in the following two ways.
(1.4) The interior A

is the set
A

:=¦x ∈ A| there exists an open neighborhood V of x such that V ⊂A].
(1.5) The interior A

is the largest open subset contained in A.
(1.6) The interior A

is the union of all the open subsets contained in A:
A

=
¸
U∈Op(X),U⊂A
U.
Show that the closure and the interior operations are dual operations in the sense that, for any subset A⊂X,
((A)

) =A and (A) =A

.
Exercise

2. — Suppose X a set. A Kuratowski closure operator on X is a map
κ : ´(X)

´(X)
satisfying the following properties.
(2.1) For any subset U ⊂X, one has U ⊂κ(U).
(2.2) For any subset U ⊂X, one has κ(κ(U)) =κ(U).
1.2. FIRST EXAMPLES 11
(2.3) For any finite subset ¹ ⊂´(X), one has
κ

¸
¸
¸
U∈¹
U

=
¸
U∈¹
κ(U).
Given a Kuratowski closure operator κ, define subsets
(l(X
κ
) :=¦V ∈ ´(X) | κ(V) =V] ⊂´(X) and thus Op(X
κ
) :=¦U ∈ ´(X) | U ∈ (l(X
κ
)] ⊂´(X)
Show that the pair X
κ
:=(X, Op(X
κ
)) is a topological space.
Moreover, show that the map
¦Kuratowski closure operators on X]

¦topologies on X]
κ

Op(X
κ
)
is a bijection, where the inverse is the operation A

A
described in 1.1.6. That is, a topology on a set X can be
specified simply by describing a Kuratowski closure operator.
1.2. First examples
Example 1.2.1. — For any set S, there are two topologies that exist automatically on S.
(1.2.1.1) The indiscrete topology S
ι
is the pair (S, ¦∅, S]).
(1.2.1.2) The discrete topology S
δ
is the pair (S, ´(S)).
Example 1.2.2. — The Sierpi´ nski space Σ
0
is the set ¦0, 1] equipped with the subset:
Op(Σ
0
) :=¦∅, ¦1], ¦0, 1]] ⊂´ (¦0, 1]) .
One may verify directly that the Sierpi´ nski space is in fact a topological space. The set ¦0, 1] can also be given a
topology given by the subset
Op(Σ
1
) :=¦∅, ¦0], ¦0, 1]] ⊂´ (¦0, 1]) .
This defines a space Σ
1
, which is — as we will explain — homeomorphic to Σ
0
. The topologies
¦0, 1]
ι
, Σ
0
, Σ
1
, and ¦0, 1]
δ
specify all the topologies on the set ¦0, 1].
Exercise 3. — Describe explicitly all the topologies on the set ¦0, 1, 2].
Example 1.2.3. — Any set S can be given the cofinite topology, in which a subset U ⊂ S is declared to be open if
and only if either S =∅, or else the complement
S
U is finite. In this case, the closed subsets of S are the finite sets
and S itself.
Similarly, any set S can be given the cocountable topology, in which a subset U ⊂ S is declared to be open if and
only if either S =∅, or else the complement
S
U is countable. In this case, the closed subsets of S are the countable
sets and S itself.
Example 1.2.4. — Suppose (X, Op(X)) a topological space, and suppose A⊂X a subset of X. Then Amay be given
the subspace topology, defined by
Op(A) :=¦U ∈ ´(U) | there exists U
/
∈ Op(X) such that U = U
/
∩A].
A subspace of X is thus a subset A⊂X, equipped with the subspace topology.
Example 1.2.5. — Let us consider the vector space R
n
. For any point x ∈ R
n
and any real number >0, the (open)
ball of radius r centered at x is the subset
B(x, ) :=¦y ∈ R
n
| ||y −x|| <].
Now the set R
n
can be given a topology in the following manner: we will say that a subset U ⊂R
n
is open if for any
point x ∈ U, there is a real number >0 such that B(x, ) ⊂ U. This is known as the standard topology on R
n
.
12 CHAPTER 1. THE CATEGORY OF TOPOLOGICAL SPACES
Example 1.2.6. — We can now combine the previous two examples to get a whole world of interesting topological
spaces.
(1.2.6.1) The open ball of radius 1 at 0 is sometimes called the n-cell.
(1.2.6.2) The closed n-disk is the set
D
n
:=¦x ∈ R
n
| ||x|| ≤1].
It is equal to the closure B(0, 1) of the ball of radius 1 at 0.
(1.2.6.3) The (n −1)-sphere is the set
S
n−1
:=¦x ∈ R
n
| ||x|| ≤1].
It is equal to the complement of the n-cell in this n-disk.
(1.2.6.4) The rational numbers Q⊂R can be given the subspace topology.
(1.2.6.5) The irrational numbers R−Q⊂R can also be given the subspace topology.
Exercise 4. — Show that the subspace topology on Q is not the discrete topology on Q. Show also that Q is dense
in R.
1.2.7. — One of the most difficult aspects of topology is that, despite the fact that the subset Op(X) ⊂ ´(X) is
the information that is needed in order to specify a topology on a set X, it is frequently difficult to give an explicit
description of the elements of Op(X).
This leads us to the notion of a subbase for a topology.
Proposition 1.2.8. — Suppose X a set, and suppose · ⊂ ´(X) a collection of subsets of X. Then there exists a unique
coarsest topology Op
·
(X) containing · .
Proof. — Define a topology in the following manner: let 0 be the set of all finite intersections of elements of ·
(including the empty intersection, which is all of X). Then let Op
·
(X) be the set of all unions of elements of 0.
This is indeed a topology. Unions of unions of elements of 0 are unions of elements of 0. Moreover, for any
finite subset ¹ ⊂Op
·
(X), one may write any element U ∈ ¹ as a union of elements of 0:
U =
¸
i ∈I
U
B
i
,
where B
i
∈ 0. Now
¸
U∈¹
U =
¸
U∈¹
¸
i ∈I
U
B
i
=
¸
U∈¹, i
U
∈I
U
¸
U∈¹
B
i
U
,
and since each set
¸
U∈¹
B
i
U
is an element of 0 by construction, this intersection is a union of elements of 0.
(1)
Any topology containing · must contain all unions of finite intersections of elements of · , so any such topol-
ogy must be finer than Op
·
(X). Thus Op
·
(X) is the unique coarsest topology containing · .
Definition 1.2.9. — In the situation of the previous proposition, the topology Op
·
(X) is said to be generated by · ,
and the set · is said to be a subbase for the topological space (X, Op
·
(X)).
Exercise 5. — Show that the standard topology on R [1.2.5] is generated by the set
¦(a, ∞) | a ∈ R] ∪¦(−∞, b) | b ∈ R] ⊂´(R).
Example 1.2.10. — More generally, on any totally ordered set X, one may define the order topology, which is gen-
erated by the set
¦(a, ∞) | a ∈ X] ∪¦(−∞, b) | b ∈ X] ⊂´(X),
where (a, ∞) and (−∞, b) are shorthand:
(a, ∞) :=¦x ∈ X | x >a] and (−∞, b) :=¦x ∈ X | x < b].
(1)
If this formula seems confusing, contemplate the case when ¹ has cardinality 2.
1.3. CONTINUOUS MAPS AND HOMEOMORPHISMS 13
Example 1.2.11. — Suppose X and Y two spaces. Then their product
X ×Y :=¦(x, y) | x ∈ X, y ∈ Y]
has a topology, called the product topology generated by the set
¦U ×Y | U ∈ Op(X)] ∪¦X ×V | V ∈ Op(Y)].
Exercise 6. — Show that the standard topology on R
n
can be described inductively in the following way.
(6.1) If n =0, then R
0
=, and the standard topology is just the discrete topology.
(6.2) If n =1, then R
1
=R, and the standard topology is just the order topology, as you showed above.
(6.3) If n > 1, then R
n
= R
n−1
×R, and the standard topology on R
n
is just the product topology of the standard
topology on R
n−1
with the standard topology on R.
1.3. Continuous maps and homeomorphisms
Definition 1.3.1. — Suppose X and Y topological spaces. Then a map f : X

Y is continuous if the inverse
image of any open set is open, that is, if for any U ∈ Op(Y), the set f
−1
U ∈ Op(X).
Example 1.3.2. — Of course the identity map of any space is a homeomorphism. More generally, suppose X a
space, and suppose A ⊂ X is a subset, equipped with the subspace topology. Then the inclusion map i : A

X
(given by the rule i (a) =a) is continuous.
Exercise 7. — In fact, the subspace topology is rigged precisely to make this happen. Show that if X is a space,
and A⊂ X is a subset, then the subspace topology on A is the coarsest topology on A such that the inclusion map
i : A

X is continuous.
Example 1.3.3. — The composition of two continuous functions is continuous.
Example 1.3.4. — Suppose X is a set with two topologies Op(X) and Op
/
(X). Then Op(X) is finer than Op
/
(X) if
and only if the identity map on X is a continuous map
(X, Op(X))

(X, Op
/
(X)) .
Example 1.3.5. — Suppose X and Y two spaces, and consider the product X ×Y with the product topology. Then
we have two projection maps
pr
1
: X ×Y

X and pr
2
: X ×Y

Y ,
given by the rules pr
1
(x, y) = x and pr
2
(x, y) = y. Both of these maps are continuous.
Exercise 8. — Once again, the way we defined the product topology guaranteed that this property was as sharp as
possible. If X and Y are spaces, then the product topology is the coarsest topology on X ×Y such that the two
projection maps
pr
1
: X ×Y

X and pr
2
: X ×Y

Y ,
are continuous.
Definition 1.3.6. — A map f : X

Y is open if the image of any open set is open, that is, if for any U ∈ Op(X),
f U ∈ Op(Y).
Example 1.3.7. — Open maps need not be continuous. Suppose X a topological space, and suppose S any set. Let
us endow S with the discrete topology S
δ
. Then one sees that any map p : X

S
δ
is open; however, in order for
p to be continuous, for any s ∈ S the fiber over s — i.e., the inverse image f
−1
¦s ] — would have to be an open set.
Hence the floor function |·| : R

Z (which just rounds any real number down to the nearest integer) is open, but
not continuous.
Example 1.3.8. — If X and Y are topological spaces, then the projection maps pr
1
and pr
2
are both continuous (as
we have seen) and open.
Example 1.3.9. — Consider the function f : R

R given by the rule f (x) = x
2
. This is continuous, but not
open.
14 CHAPTER 1. THE CATEGORY OF TOPOLOGICAL SPACES
Exercise 9. — Show that if X is a space, and A⊂X is a subset with the subspace topology, then the inclusion map
i : A

X is open if and only if Ais an open subset of X.
Definition 1.3.10. — A continuous map f : X

Y is a homeomorphism if it is both a bijection and an open map.
Two spaces X and Y are said to be homeomorphic if there is a homeomorphism X

Y .
1.3.11. — A homeomorphism between two topological spaces is a guarantee that those two spaces are indistin-
guishable from the point of view of topology. In group theory or set theory, an isomorphism or bijection between two
objects (groups in the first case, sets in the second) guaranteed that the objects involved were the same, up to some
“relabeling,” so anything you knew about the first group or set was reflected in the second group of set.
As we will see, the same phenomenon happens here. In the case of homeomorphisms, not only do you have the
same points up to some relabeling, but you have the same open sets up to that relabeling!
1.3.12. — Here’s another way to think of homeomorphism. If f : X

Y is a continuous map, then the inverse
image can be viewed as a map
f
−1
: Op(Y)

Op(X) .
The statement that f is a homeomorphism is the statement that both f and f
−1
are bijections.
Alternately, if f : X

Y is a bijection, then it has an inverse f
−1
: Y

X . The statement that f is a
homeomorphism is the statement that both f and f
−1
are continuous.
(We’re being totally sloppy about notation here, but only because everyone else is too, and you might as well
get used to it now. Mathematicians use the symbol f
−1
in two different ways, often in close proximity to one
another. (1) In the first description, we were assuming that f is continuous, and we were talking about the function
Op(Y)

Op(X) that takes an open set U to its inverse image f
−1
U, which is again an open set. (2) In the second
description, we were assuming that f is a bijection, and we were talking about the function Y

X which is the
inverse to f .)
Example 1.3.13. — Of course the identity map is a homeomorphism.
Example 1.3.14. — The two Sierpi´ nski spaces Σ
0
and Σ
1
are homeomorphic.
Example 1.3.15. — Note that, in order for f : X

Y to be a homeomorphism, we do not only require that f
be a bijective continuous map. That is not enough: let us think of the set
U(1) :=¦z ∈ C | |z| =1]
with the subspace topology from C

=R
2
. Now contemplate the map
ρ : [0, 2π)

U(1)
given by the rule ρ(x) = e
i x
. This wraps the half-open interval around the circle. It is a bijective continuous map,
but it had better not be a homeomorphism!
Example 1.3.16. — Here’s a very important example to contemplate. Consider the tangent function as a map
tan :


π
2
,
π
2

R.
It’s a homeomorphism! This suggests something that may at first seem surprising: topology doesn’t “see” the differ-
ence between an infinitely long line and an open interval of finite length.
This might seem a little odd, since we’ve been thinking of topological spaces as sets with a concept of “nearness”
of points. But this leads to an important idea about what topology actually is: one might say that topology is what
is left over from geometry when you throw out the notion of distance.
Exercise 10. — Continuing with that idea, find a homeomorphism between R and the ray (0, ∞).
Exercise 11. — Suppose n is a nonnegative integer. Show that any two n-dimensional balls B
n
(x, δ) and B
n
(y, ) in
R
n
are homeomorphic, and so are their closures.
Exercise

12. — Suppose n is a nonnegative integer. Show that the following three spaces are all homeomorphic:
(12.1) the n-dimensional ball B
n
(x, ) for any x ∈ R
n
and any >0,
(12.2) the euclidean space R
n
, and
1.3. CONTINUOUS MAPS AND HOMEOMORPHISMS 15
(12.3) the product space
¸
m
i =1
B
k
i
(x
i
,
i
) for any decomposition n =
¸
m
i =1
k
i
, and points x
i
∈ R
k
i
, and any real
numbers
i
>0.
Example 1.3.17. — Here’s a challenging example. Consider the spaces R
m
and R
n
with the standard topology. It is
a fact that they are homeomorphic if and only if m = n. Certainly one direction here is obvious! But how on earth
could you show that there is no homeomorphism between two spaces? What kind of contradiction could you get
from the existence of a homeomorphism? These kinds of questions will lead us naturally to
Example 1.3.18. — Here’s another example that merits some contemplation. Consider the function f (x) = 1/x.
This can regarded as a map
f : R−¦0]

R−¦0].
Since f is its own inverse, f here is a homeomorphism. (In general, a homeomorphism from a space to itself is
called an automorphism.) Of course we have removed the point 0 ∈ R, because in elementary school we were told
that 1/0 is “undefined.” But let’s try to define it anyhow.
We note that, as x approaches 0 from the right, 1/x increases without bound; as x approaches 0 from the left,
1/x decreases without bound. If we wanted to add a point that would play the role of 1/0, then this leads us to the
following idea: consider the set R|¦∞], where ∞here is just a formal symbol that we’ve tacked on. Let’s topologize
this set in the following way: we will contemplate the set
¦(a, b) | a, b ∈ R] ∪¦(−∞, c) ∪¦∞] ∪(d, ∞) | c, d ∈ R] ⊂´(R∪¦∞]),
and we will contemplate the topology generated by this set. That is, we will look at all unions of all finite intersec-
tions of open intervals in R along with sets
(−∞, c) ∪¦∞] ∪(d, ∞)
that “wrap around” our added point ∞.
Now we think of our map f : R−¦0]

R−¦0], and we note that we can extend it to a map F :
R∪¦∞]

R∪¦∞], where we set F (x) := 1/x for any x ∈ R−¦0], we set F (0) := ∞, and we set F (∞) := 0.
With the topology we’ve given R∪¦∞], this is continuous!
The space R∪¦∞] we’ve constructed here is called a compactification of R. (We’ll explain more about that later.)
The subspace R⊂R∪¦∞] is open, and ∞is contained in the closure of any ray (−∞, c) or (d, ∞).
Exercise 13. — A little bit of thought should convince you that the space R∪¦∞] we’ve described in the previous
example is in fact a circle. Let’s find a homeomorphism that makes this explicit. Translate our circle S
1
so that we
are looking at the unit circle around the point (0, 1) in R
2
:
S
1
(1) :=¦θ ∈ R
2
| ||θ −(0, 1)|| =1].
Now let us define a map g : S
1
(1) −¦(0, 2)]

R in the following way. For any point θ ∈ S
1
(1) −¦(0, 2)], let

θ
be the unique line passing through (0, 2) and θ in R
2
. Now define g(θ) to be the x-coordinate of the point of
intersection between
θ
and the x-axis ¦(x, y) ∈ R
2
| y =0].
(13.1) Write an explicit formula for g.
(13.2) Show that g is a homeomorphism.
(13.3) Show that g can be extended to a homeomorphism
G : S
1
(1)

R∪¦∞]
by setting G(θ) := g(θ) for any θ ∈ S
1
(1) −¦(0, 2)] and G((0, 2)) :=∞.
(13.4) In the previous example, we constructed a homeomorphism F : R∪¦∞]

R∪¦∞] (which is not the
identity). Using the homeomorphism G, we can rewrite this as a homeomorphism
G
−1
◦ F ◦ G : S
1
(1)

S
1
(1) .
What homeomorphism is this? How might you draw its graph?
16 CHAPTER 1. THE CATEGORY OF TOPOLOGICAL SPACES
Exercise 14. — The space R∪¦∞] is not the only compactification of R; we were led to consider it by our example.
If we thought of another example, we might be led to a different compactification. Think, for example, of the
function h(x) = x
3
, which we regard as a map
h : R

R.
(14.1) Show that h is a homeomorphism. (Be careful if you decide to study h
−1
!)
(14.2) Show that h(x) increases without bound as x does, and likewise h(x) decreases without bound as x does.
(14.3) Describe a space R∪¦−∞, ∞] (where −∞and ∞are each just formal symbols) with the following properties.
(1) R is an open subspace of R∪¦−∞, ∞]. (2) The closure of a ray (−∞, a) ⊂R in R∪¦−∞, ∞] is the set
[−∞, a] :=¦−∞] ∪(−∞, a],
and the closure of a ray (b, ∞) ⊂R in R∪¦−∞, ∞] is the set
[b, ∞] :=[b, ∞) ∪¦∞].
(14.4) Describe an extension of h to a homeomorphism
H : R∪¦−∞, ∞]

R∪¦−∞, ∞].
(14.5) Show that the space R∪¦−∞, ∞] is homeomorphic to the interval [−1, 1].
(14.6) Show that the map
p : R∪¦−∞, ∞]

R∪¦∞]
(with the topologies defined here and above) is continuous. Show, moreover, that for any topological space
X, and any continuous map f : R∪¦−∞, ∞]

X such that f (−∞) = f (∞), there exists a unique map
f
/
: R∪¦∞]

X such that f = f
/
◦ p.
CHAPTER 2
CONSTRUCTING TOPOLOGIES AND CONTINUOUS MAPS
2.1. Quotients
Definition 2.1.1. — Suppose X a space, and suppose A⊂X a subspace. Then a quotient of X by Ais a pointed space
(X/A, α) and a continuous map q : X

X/A such that q(A) =¦α], and for any space Y and any continuous map
f : X

Y such that f (A) = ¦a] for some element a ∈ Y, there exists a unique continuous map f
/
: X/A

Y
such that f
/
(α) =a and f = f
/
◦ q.
2.1.2. — Suppose X a space, and suppose A⊂ X a subspace. The definition above is nonconstructive, in the sense
that it does not make it apparent that such a space X/A exists. In fact, it always does. Here’s one way to construct
it.
Let X/A=(X −A) |¦α]; consider the map q : X

X/A, given by the formula
q(x) :=

x if x ∈ X −A;
α otherwise.
We topologize X/Ain the following manner. We say that a subset U ⊂X/Ais open if and only if the inverse image
q
−1
U is open in X. This forces q to be continuous. (In fact, this is the finest possible topology on X/Asuch that q
is continuous.)
I claim that the space X/A and the continuous map q : X

X/A satisfy the conditions of the definition
of a quotient. To see this, suppose Y a topological space, and suppose f : X

Y a continuous map such that
f (A) =¦a] for some element a ∈ Y. Let us define f
/
: X/A

Y in the following way
f
/
(x) :=

x if x ∈ X −A;
a otherwise.
One sees that this is the only possible map such that f = f
/
◦q. Moreover, f
/
is continuous: for any open set V ⊂Y,
f
/−1
V is open in X/Aif and only if f
−1
V = q
−1
f
/−1
V is open in X, but this is automatic since f is continuous.
Exercise 15. — Show that the above description of X/A is unique up to unique homeomorphism. That is, show
that if X is a space, A⊂ X is a subspace, and both q : X

W and q
/
: X

W
/
are each quotients of X by A,
then there is a unique homeomorphism h : W

W
/
such that h ◦ q = q
/
.
Exercise

16. — For any integer n ≥0, show that the quotient D
n
/S
n−1
is homeomorphic to S
n
.
2.1.3. — Contemplate the quotient X/∅. The answer may surprise you!
2.2. Coverings and locality
Exercise 17. — Subbases for topologies permit the easy identification of continuous maps. Suppose X and Y topo-
logical spaces, and suppose · a subbase for the topology on Y. Then show that a map f : X

Y is continuous
if and only if the inverse image of any element U ∈ · under f is open in X.
18 CHAPTER 2. CONSTRUCTING TOPOLOGIES AND CONTINUOUS MAPS
2.2.1. — The previous result is helpful because it permits one to check only a small number of conditions on subsets
of Y in order to guarantee the continuity of a function.
Definition 2.2.2. — Suppose X and Y two topological spaces, suppose f : X

Y a map, and suppose x ∈ X.
Then one says that f is continuous at x if for any neighborhood V of f (x) ∈ Y, there is a neighborhood U of x ∈ X
such that f (U) ⊂V.
2.2.3. — Note that I have written “neighborhood” in the above definition. I could easily as well have written “open
neighborhood”: f is continuous at x if and only if for any open neighborhood V of f (x) ∈ Y, there is an open
neighborhood U of x ∈ X such that f (U) ⊂V.
Lemma 2.2.4. — Suppose X a topological space, and suppose U ⊂X a subset thereof. Then U is open if and only if, for
any point x ∈ U, there exists an open neighborhood W of x with W ⊂ U.
Proof. — If U is open, then of course U itself is an open neighborhood of any point x ∈ U.
Conversely, one may for the interior U

of U, which is the union of all the open sets W ⊂ U. One sees that
U

= U: indeed, U

⊂ U, and for any point x ∈ U, there exists an open neighborhood W of x with W ⊂ U,
whence x ∈ U

.
Proposition 2.2.5. — Suppose X and Y two topological spaces, and suppose f : X

Y a map. Then f is continuous
if and only if it is continuous for every point x ∈ X.
Proof. — Suppose f continuous, and suppose x ∈ X a point. Then for any neighborhood V of f (x) ∈ Y, one may
contemplate the interior V

, which is open, and its inverse image f
−1
(V

), which is open as well. Now the image
f ( f
−1
(V

)) ⊂V

⊂V. Hence f is continuous at every point x ∈ X.
Suppose, conversely, that f is continuous at every point x ∈ X. Suppose V ⊂ Y an open set. The claim is that
f
−1
(V) is open. For any point x ∈ f
−1
(V), there is an open neighborhood W of x ∈ f
−1
(V) with f (W) ⊂V, or
equivalently, with W ⊂ f
−1
(V). Now we apply the lemma above to see that f
−1
(V) is open.
Definition 2.2.6. — Suppose X a space. A collection · ⊂ ´(X) of subspaces of X is said to cover X — or to be a
covering of X — if X =
¸
V∈·
V. One says that · is an open covering of X if every V ∈ · is an open set; likewise,
one says that · is an closed covering of X if every V ∈ · is an closed set.
For a subspace A ⊂ X, we say that a collection · ⊂ ´(X) of subspaces of X is said to cover A — or to be a
covering of A— if A⊂
¸
V∈·
V, or, equivalently, of A=
¸
V∈·
A∩V.
Exercise 18. — Suppose X a space, and suppose · an open covering of X. Now suppose, for any V ∈ · , that
f
V
: V

Y is a continuous function, and suppose that, for any elements V, W ∈ · , one has
f
V
|
V∩W
= f
W
|
V∩W
: V ∩W

Y .
Then there exists a unique continuous map f : X

Y such that for any V ∈ · ,
f |
V
= f
V
: V

Y .
Definition 2.2.7. — Suppose X a topological space, and suppose x ∈ X. Then a fundamental system of neighborhoods
of x ∈ X is a collection ·
x
of neighborhoods of x with the following property. For any neighborhood W of x, there
is a neighborhood V ∈ ·
x
with V ⊂W.
Definition 2.2.8. — Suppose X a topological space. Then a base for X is a collection 0 ⊂Op(X) such that for any
point x ∈ X, the set
0
x
:=¦U ∈ 0 | x ∈ U]
is a fundamental system of neighborhoods of x ∈ X.
2.2.9. — A base for a topological space is also a subbase for the topology Op(X).
Proposition 2.2.10. — Suppose X is a topological space, and suppose 0 a base for the topology. The the following are
equivalent for a subset U ⊂X.
(2.2.10.1) The set U is open in X.
(2.2.10.2) For any point x ∈ U, there exists an element V ∈ 0
x
such that V ⊂ U.
2.2. COVERINGS AND LOCALITY 19
(2.2.10.3) The set U can be written as the union of elements of 0.
Proof. — Suppose U is open in X; then for any point x ∈ U, U is an open neighborhood of x. Since 0
x
is an
open system of neighborhoods of x ∈ X, there is an element V ∈ 0
x
such that V ⊂ U. This shows that the first
condition implies the second.
Suppose that, for any point x ∈ U, there exists an element V ∈ 0
x
such that V ⊂ U. Then (using the axiom of
choice) one may select, for every point x ∈ U, one element V
x
∈ 0
x
such that V
x
⊂ U. The union U
/
:=
¸
x∈U
V
x
contains every point of U by definition, so that U
/
⊃ U, and since V
x
⊂ U for every x ∈ U, it follows that U
/
⊂ U.
Thus U is a union of elements of 0. This shows that the second condition implies the third.
Finally, suppose that U can be written as the union of elements of 0. By definition, every element of 0 is open,
so by the axioms for a topological space, U is open as well. This completes the proof.
Proposition 2.2.11. — Suppose X a topological space. Then a subset 0 ⊂Op(X) is a base for the topology if and only if
the following conditions are satisfied.
(2.2.11.1) The elements of 0 cover X.
(2.2.11.2) For any U, V ∈ 0 and any element x ∈ U ∩V, there is an element W ∈ 0
x
such that W ⊂ U ∩V.
Example 2.2.12. — For any totally ordered set X, the set
¦(a, b) | a, b ∈ X]
is a base for the order topology on X. (Here, one writes
(a, b) :=¦x ∈ X | a < x < b].)
Example 2.2.13. — Suppose n ≥0 an integer. Then the set
¦B
n
(x, ) | x ∈ Q
n
, ε ∈ Q]
is a base for the standard topology on euclidean space R
n
.
Exercise 19. — Suppose X a set with two topologies Op(X) and Op
/
(X); suppose 0 a base for (X, Op(X)) and 0
/
a base for (X, Op
/
(X)). Show that Op(X) is finer than Op
/
(X) if and only if, for any x ∈ X and any element U ∈ 0
/
x
,
there exists an element V ∈ 0
x
such that V ⊂ U.
Example 2.2.14. — Here is a standard counterexample in topology; everyone must see it once in their lives. The
Sorgenfrey line R

is the set Rwith the lower limit topology. This is the topology generated by the half-open intervals
[a, b) for a, b ∈ R. The set
0

:=¦[a, b) | a, b, ∈ R]
is then a basis for R

. (One may also define the upper limit topology R
u
, which is homeomorphic to R

via the map
x


−x .) One can check fairly easily that these basic open sets [a, b) are both open and closed.
The previous exercise can be used to show that the Sorgenfrey line is finer than the standard (= order) topology
on R. Indeed, for any open interval (a, b), one may choose any point a
/
∈ (a, b), and then [a
/
, b) ⊂(a, b).
A map f : R

R is continuous if and only if, for any x ∈ R,
f (x) = lim
→0
+
f (x +).
In particular, the floor function |·| : R

R is continuous.
CHAPTER 3
COMPACTNESS I
3.1. Compact spaces
Definition 3.1.1. — A topological space X is compact if any open covering has a finite subcovering. A compact
topological space is sometimes called a compactum.
For any space X, write (p(X) for the set of all compact subsets of X.
Example 3.1.2. — Any finite space is compact, and the indiscrete topology on any set is compact.
Example 3.1.3. — The euclidean spaces R
n
are not compact; indeed, the open balls B
n
(x, ) form an open covering,
but there is no finite subcovering. To see this, consider any finite covering of R
n
by balls B
n
(x
i
,
i
) for i =1, 2, . . . , m,
and suppose x a point of one of the balls. Then one may choose a real number r so that B
n
(x, r ) ⊃
¸
m
i =1
B
n
(x
i
,
i
).
Lemma 3.1.4. — Suppose X and Y spaces, and suppose f : X

Y a continuous function. Then the image f (A) of
any compact subspace A⊂X is compact.
Proof. — Suppose ¹ an open covering of f (A); then the set
¦ f
−1
U | U ∈ ¹]
is an open covering of A, which thus has a finite subcovering · . Now the claim is that the subset
¦U ∈ ¹ | f
−1
U ∈ · ],
which is finite, is a covering of f (A). Indeed, suppose y = f (a) ∈ f (A) a point with a ∈ A. Then there exists some
U ∈ ¹ such that f
−1
U ∈ · and a ∈ f
−1
U, whence f (a) ∈ U.
Lemma 3.1.5. — A closed subspace of a compact space is compact.
Proof. — Suppose X a compactum, and suppose A ⊂ X closed. Suppose ¹ an open covering of A. Since each
element U ∈ ¹ is open in A, one may choose a collection ¹
/
⊂Op(X) such that
¹ =¦U ∈ Op(A) | U = U
/
∩Afor someU
/
∈ ¹
/
].
Now the set ¹
/
∪¦X −A] is an open covering of A; hence there is a finite subcovering ·
/
⊂¹
/
. Now
· =¦U ∈ Op(A) | U = U
/
∩Afor someU
/
∈ ·
/
]
is a finite subcovering of ¹.
Exercise 20. — Suppose we are given a finite collection ¦X
i
] of compact spaces, i = 1, 2, . . . , n. Show that the
product space
¸
n
i =1
X
i
is compact. (This is the easy case of Tychonoff’s Theorem, to which we shall return.)
Theorem 3.1.6. — The following are equivalent for a subspace A⊂R
n
.
(3.1.6.1) Ais compact.
(3.1.6.2) [Bolzano–Weierstraß] Every sequence of points in Ahas a convergent subsequence.
(3.1.6.3) [Heine–Borel] Ais closed and bounded.
22 CHAPTER 3. COMPACTNESS I
Proof. — Let us show that the first of these conditions implies the second. Suppose Acompact, and suppose (x
i
)
i ≥0
a sequence of points in A. If there were no convergent subsequence of (x
i
)
i ≥0
, then the set ¦x
i
]
i ≥0
would be a closed
subset of A, hence compact, and there would be a sequence
i
of positive real numbers such that the balls B
n
(x
i
,
i
)
would be disjoint. But then ¦B
n
(x
i
,
i
)]
i ≥0
would be an open cover of ¦x
i
]
i ≥0
with no finite subcover.
Let us show that the second property implies the third. Suppose A has the property that every sequence has a
convergent subsequence. Then A must be bounded, since otherwise there exists a sequence (x
i
)
i ≥0
of points in A
such that ||x
i
|| → ∞. It must also be closed, since if x ∈ A−A is a point “at the boundary,” one can construct a
sequence of points of Aconverging to x.
Finally, let us show that the third condition implies the first. Suppose A is closed and bounded. Since A is
bounded, it is contained in a box [−a, a]
n
for some a > 0. Since every closed subspace of a compact space is
compact, it is enough to show that [−a, a]
n
is compact. By the previous exercise, it is enough to show that [−a, a]
is compact. Suppose ¹ an open covering of [−a, a]; then let
c =sup¦x ∈ R | [−a, x] can be covered by finitely many elements of ¹].
Suppose, to generate a contradiction, that c < a; then let U ∈ ¹ be an element of the open covering containing
c. Then for some > 0, one has B(c, ) ⊂ U. By the definition of c, one may cover [−a, a − c/2] with finitely
many elements of ¹; adding in U to that set yields a finite open covering of [−a, a + c/2], yielding the desired
contradiction.
Exercise 21. — Suppose X a compact space, and suppose f : X

R a continuous function. Then f attains both
a maximum and a minimum value.
3.2. The Tychonoff Product Theorem
Definition 3.2.1. — A collection ¨ of sets is said to be of finite character if the following condition holds: a set Ais
an element of ¨ if and only if every finite subset of Ais an element of ¨.
Lemma 3.2.2 (Tukey). — Suppose ¨ a nonempty collection of sets. Then if ¨ is of finite character, then every element
A∈ ¨ is contained in a maximal set M ∈ ¨, i.e., a set M such that no element of ¨ properly contains M.
Proof. — Give ¨ a partial ordering by inclusion. The union of every chain of elements of ¨ must also belong to
¨, so by Zorn’s Lemma, every element A∈ ¨ is contained in a maximal element.
Theorem 3.2.3 (Alexander Subbase). — Suppose X a topological space, and suppose · a subbase for Op(X) that is
also an open cover of Op(X). If every subcover of · has a finite subcover, then X is compact.
Proof. — The language here can get a tad confusing, so let’s introduce some new terminology to make things a little
simpler. Let’s call a family ¹ ⊂´(X) exiguous if it does not cover X, and let’s call ¹ bourgeois if any finite subset
of ¹ is exiguous.
Using our new terminology, we see that X is compact if and only if any bourgeois family of open sets is exiguous.
We can also rewrite our assumption on the subbase · : we are assuming that any bourgeois subfamily of · is
exiguous.
Our aim is to show that any bourgeois family · of open sets is exiguous. Write
¨ :=¦¹ ∈ ´(Op(X)) | ¹ is bourgeois].
This family is obviously of finite character. So suppose · a bourgeois family of open sets, and let ¨(·) be the
collection of all bourgeois families of open sets of X that contain ·:
¨(·) :=¦¹ ∈ ¨ | · ⊂¨].
Tukey’s Lemma says that it contains a maximal element . ∈ ¨(·). It is enough to show that . is exiguous, and
this is now our goal.
Exercise 22. — Prove the following Key Fact about our maximal element .:
If ¦U
i
]
n
i =1
is a finite collection of open sets of X such that the intersection
¸
n
i =1
U
i
is contained in
an element of ., then one of the U
i
is itself an element of ..
3.3. THE COMPACT-OPEN TOPOLOGY 23
(Hint: if U is an open set of X that is not contained in ., show that there is a finite collection ¦V
j
]
m
j =1
of elements
of . such that ¦U] ∪¦V
j
]
m
j =1
is a covering of X.)
All right, now it’s time to consider our subbase · ; again, we are assuming that any bourgeois subcollection of
· is exiguous. Let’s contemplate the intersection . ∩ · ; it is bourgeois, and thus (by our assumption on · )
exiguous.
Now I claim that
¸
U∈.
U ⊂
¸
U∈.∩·
U.
Suppose x ∈ U ∈ .; then since · is a subbase, there is a finite intersection W :=
¸
n
i =1
U
i
of elements U
i
∈ ·
such that x ∈ W and W ⊂ U. By the Key Fact above, one of the U
i
is itself and element of .. This means that
x ∈ U
i
∈ . ∩· , just as we had hoped.
But remember the intersection . ∩· is exiguous, so we conclude that . is also!
Exercise

23. — Suppose X a totally ordered set. Let us say that X is order-complete if every subset of X has a
supremum in X. Show that X is compact (with the order topology) if and only if it is order-complete.
Theorem 3.2.4 (Tychonoff Product). — The product of compact spaces is compact.
Proof. — Suppose ¦X
α
| α ∈ A] a collection of compact spaces, and give
¸
α∈A
X
α
the product topology. This is the
topology generated by the subbase
· :=¦pr
−1
α
(U) | α ∈ A, U ∈ Op(X
α
)].
Let’s use the Alexander Subbase Theorem. We need to show that any bourgeois subcollection ¹ of · is exigu-
ous. Suppose ¹ a bourgeois subcollection of · . For every α ∈ A, set
¹
α
:=¦U ∈ Op(X
α
) | pr
−1
α
(U) ∈ ¹].
It is obvious that ¹
α
is bourgeois and hence exiguous (since X
α
is compact). Hence for each α ∈ A, we can choose a
point x
α
∈ X
α
such that x
α
/ ∈
¸
U∈¹
α
U. But now the point
x :=(x
α
)
α∈A
is not a member of any element of ¹. So ¹ is exiguous.
3.3. The compact-open topology
Definition 3.3.1. — Suppose X and Y two topological spaces. Denote by ((X, Y) the set of all continuous maps
X

Y . For any subsets A⊂X and B ⊂Y, write
· (A, B) :=¦ f ∈ ((X, Y) | f (A) ⊂B].
Now we topologize the set ((X, Y) in the following manner. Take the topology generated by the subbase
¦· (K, U) | K ∈ (p(X), and U ∈ Op(Y)].
This is called the compact-open topology.
Exercise 24. — Suppose X any compact space, and suppose (Y, d) a metric space. A sequence of continuous func-
tions f
n
∈ ((X, Y) is said to converge uniformly to a continuous function f ∈ ((X, Y) if
lim
n→∞
sup
x∈X
d( f
n
(x), f (x)) =0;
in other words, for every >0, there exists a natural number N such that for any x ∈ X and any n ≥N,
d( f
n
(x), f (x)) <.
Prove that a sequence f
n
of continuous functions X

Y converges uniformly if and only if it converges in
((X, Y) with the compact-open topology. Is the same true if X is not compact?
Exercise 25. — Suppose X = S
δ
a discrete space. Showthat the compact-open topology on ((X, Y) coincides with
the product topology on
¸
s ∈S
Y.
CHAPTER 4
COMPACTNESS II
4.1. Forms of compactness
We have already seen that for subspaces of a euclidean space, compactness is equivalent to the Heine–Borel prop-
erty (closed and bounded) of the Bolzano–Weierstraß property (the property every sequence has a convergent subse-
quence). This doesn’t quite cut it for more general spaces. In this section, we’ll discuss analogues and generalizations
of these notions, and we’ll also talk about the notion of a net or Moore–Smith sequence.
Definition 4.1.1. — Suppose X a topological space.
(4.1.1.1) One says that X is limit point compact if and only if every infinite subset A of X has a limit point (i.e., a
point x ∈ X that is contained in A−¦x]).
(4.1.1.2) One says that X is countably compact if and only if every countable open cover has a finite subcover.
(4.1.1.3) One says that X is sequentially compact if and only if every sequence x : N

X has a convergent subse-
quence.
4.1.2. — These three notions seemremarkably similar, and in R
n
(and more generally, in any metrizable space) they
are indistinct and, moreover, equivalent to compactness. However, they are different conditions in general, and they
are very different from compactness. To see this, will contemplate some examples, including a key counterexample.
First, let’s see what implications we do have.
Exercise 26. — Show that a space X is countably compact if and only if every nested sequence X ⊃ V
1
⊃ V
2

V
3
⊃· · · of closed, nonempty subspaces of X has a nonempty intersection.
Proposition 4.1.3. — Consider the following properties for a topological space X.
(4.1.3.1) X is compact.
(4.1.3.2) X is sequentially compact.
(4.1.3.3) X is countably compact.
(4.1.3.4) X is limit point compact.
Then we have a diagram of implications
(1)

I
I
I
I
(2)
·u
u
u
u
(3)

(4)
Proof. — It is obvious that a compact spaces is countably compact.
Let us now see that a sequentially compact space is countably compact. Suppose ¹ =¦U
i
]
i ≥0
a countable open
cover of X. Proceeding inductively, write V
0
:= U
0
, and let V
p
be the first of the sequence of U
i
’s not covered by
the V
0
∪V
1
∪· · · ∪V
p−1
. If this choice is impossible at any (finite) stage, then we have our required finite subcover;
if this choice is always possible, then for every n ∈ N we may select an element x
n
∈ V
n
such that for any i < n,
26 CHAPTER 4. COMPACTNESS II
x
n
/ ∈ V
i
. Now sequential compactness guarantees that the sequence (x
n
)
n≥0
has a convergent subsequence (x
n
i
)
i ≥0
;
let y be the limit of this subsequence. Then y ∈V
n
for some integer n, so for every i >0 (in particular for i so that
n
i
> n), we have x
n
i
∈V
m
, a contradiction.
Let us now see that a countably compact space is limit point compact. Suppose X a countably compact space,
and suppose B ⊂X a subset with no limit point. Let us show that every countable subset A⊂B is finite, whence B
is itself finite. Of course any countable subset A⊂B has no limit point; so Ais closed, and for every point a ∈ A, we
can choose an open neighborhood U
a
such that U
a
∩A=¦a]. Now write
X =(X −A) ∪
¸
a∈A
U
a
;
this is an open cover of X, which must admit a finite subcover. But it admits no proper subcover at all, so A must
be finite.
Example 4.1.4. — To construct this example, we have to appeal to some basic facts about ordinals.
First, let’s recall that a strict well-ordering on a set A is a well-ordering < that is irreflexive, in the sense that for
any element a ∈ A, it is not the case that a <a. Now we come to von Neumann’s definition of ordinal: a set Ais an
ordinal if
(4.1.4.1) every element of Ais also a subset of A, and
(4.1.4.2) Ais strictly well-ordered with respect to set membership.
Put differently, an ordinal is a set Asuch that the following conditions hold.
(4.1.4.1’) If β∈ A, then βA.
(4.1.4.2’) If β, γ ∈ A, then one of the following is true: (1) β=γ, (2) β∈ γ, or (3) γ ∈ β.
(4.1.4.3’) If ∅(=βA, then there exists an element β∈ Asuch that γ ∩β=∅.
With this definition, we define 0 := ∅, the minimal ordinal. Now recursively, we define, for every integer n > 0
the ordinal
n :=¦0, 1, . . . , n−1].
These are the finite ordinals.
Every ordinal α has a successor, namely
α +1 :=α ∪¦α].
Some ordinals are not successors; these are called limit ordinals. The minimal ordinal 0 is a limit ordinal.
The first infinite ordinal is also the first nonzero limit ordinal; it is
ω :=¦0, 1, . . . ].
It has a successor ω+1, and its successor has a successor ω+2, etc.
If α and β are ordinals, then we say that α <β if α ∈ β. Suppose Aany set of ordinals; then Ais well-ordered by
this relation. Consequently, a set Aof ordinals is itself an ordinal if and only if, for every element α ∈ A, any ordinal
β<α is also an element of A. Using this, there is a supremum of any set A of ordinals: it is simply the union of the
elements of A.
We won’t prove it here, but in fact every well-ordered set is order-isomorphic to a unique ordinal in a unique fash-
ion. As a result, if we build a well-ordered set A, there is an ordinal ord(A) and an order-isomorphism A

ord(A) .
In particular, every ordinal α is equal to the set [0, α) of all ordinals less than α.
We can define operations of ordinal arithmetic in the following manner. Given two ordinals α and β, we can
concatenate them by well ordering the disjoint union α|βso that the order within α and βis preserved, and every
element of α is strictly less than every element of β. Now set
α +β:=ord(α |β).
Similarly, one can give the product α ×β the lexicographic ordering, so that (a, b) < (a
/
, b
/
) if either a < a
/
or else
a =a
/
and b < b
/
. This is well-ordered, and so one may set
α · β:=ord(α ×β).
4.2. LOCALLY COMPACT HAUSDORFF SPACES 27
We can generalize the lexicographic order to a well ordering on the set Map(α, β) of all maps f : α

β such that
only finitely many elements of α are mapped to an element larger than 0 ∈ β; one sets
β
α
:=ord(Map(α, β)).
Note that none of these operations is commutative, though both addition and multiplication are commutative.
If we apply these operations to the finite ordinals and to ω, we get a sequence of ordinals:
0, 1, 2, . . . , ω, ω+1, ω+2, . . . , ω· 2, ω· 2 +1, ω· 2 +2, . . . , ω
2
, . . . , ω
3
, . . . , ω
ω
, . . . , ω
ω
ω
, . . . , ω
ω
ω
ω
, . . . ,
0
, . . . .
The last one listed there was studied by Cantor: it is the smallest ordinal satisfying the equation

0

0
;
it can be expressed as the supremum of the sequence

0
=sup¦ω, ω
ω
, ω
ω
ω
, ω
ω
ω
ω
, . . . ].
The surprising fact is that these ordinals are all countable, because every supremum of a countable set of countable
ordinals is countable. So the infinite ordinals all have the same cardinality, but they are non-isomorphic as well-
ordered sets.
We can give any of these sets the order topology, but it might not be an interesting enterprise:
Exercise 27. — Show that for any ordinal α, the order topology and the discrete topology coincide if and only if α
contains no limit ordinal apart from 0.
Now the supremum of all countable cardinals is an uncountable cardinal ω
1
. This is the first uncountable cardi-
nal, and it is going to give us a useful counterexample now and in the future.
Every increasing sequence in ω
1
converges to an element of ω
1
. Hence the space ω
1
(when equipped with the
order topology) is sequentially compact, but it cannot be compact, since it contains no maximal element. The set
ω
1
+1 is, however, compact.
Exercise 28. — Consider the discrete space ¦0, 1]
δ
. Show that the product
X :=
¸
x∈[0,1]
¦0, 1]
δ
is compact, but not sequentially compact.
Exercise

29. — Find a space that is limit point compact, but not countably compact. (Hint: such a space must have
the property that it contains a finite subspace that is not closed.)
4.2. Locally compact Hausdorff spaces
The next variant of compactness we want to consider, local compactness, is most interesting when coupled with a
particular separation axiom, called Hausdorffness. We will discuss a variety of separation axioms later, but for now,
let us concentrate on this one, since it appears so often.
Definition 4.2.1. — A space X is said to be Hausdorff if for any two distinct points x, y ∈ X, there are neighbor-
hoods U of x and V of y such that U ∩V =∅.
Proposition 4.2.2. — A space X is Hausdorff if and only if, for any point x ∈ X, the intersection I
x
of all closed
neighborhoods of x is the singleton ¦x].
Proof. — Suppose X Hausdorff, and suppose x ∈ X. Surely x ∈ I
x
, and I claim that any point y ∈ X −¦x] is not
in I . Indeed, for any such point, one may find disjoint open neighborhoods U of x and V of y, whence X −V is a
closed neighborhood of x not containing y. Thus I =¦x].
Conversely, suppose that, for any point x ∈ X, one has I
x
= ¦x]. Suppose x and y two distinct points of X.
Then since I
x
=¦x], there exists a closed neighborhood W of x not containing y, and now the interior W

and the
complement X −W are disjoint open neighborhoods of x and y, respectively.
28 CHAPTER 4. COMPACTNESS II
Exercise 30. — Show that a space X is Hausdorff if and only if the diagonal

X
:=¦(x, x) ∈ X ×X]
is a closed subspace of X ×X.
Example 4.2.3. — Any metric space (in particular any euclidean space R
n
) is Hausdorff. The cofinite topology on
any infinite set is not.
Proposition 4.2.4. — Any subspace of a Hausdorff space is Hausdorff, and any product of any collection of Hausdorff
spaces is Hausdorff.
Proposition 4.2.5. — If X and Y are spaces, and Y is Hausdorff, then for any continuous maps f , g : X

Y , the
equalizer
E
( f , g)
:=¦(x, x
/
) ∈ X ×X | f (x) = g(x
/
)]
is closed in X ×X, and
Proof. — The set E
( f , g)
is the inverse image of ∆
Y
under the map ( f , g) : X ×X

Y ×Y .
Corollary 4.2.6. — If S and T are spaces, and T is Hausdorff, then for any continuous map h : S

T , the graph
Γ
h
:=¦(s , t ) ∈ S ×T | h(s ) = t ]
is closed in S ×T.
Proof. — Apply the proposition to the case when X = S ×T, Y = T, f (s , t ) = t , and g(s , t ) = h(s ) to obtain a
closed subspace of S ×T ×S ×T, and intersect this with the diagonal ∆
S×T
.
Proposition 4.2.7. — If X is a Hausdorff space, then any compact subspace K ⊂X is closed.
Proof. — Suppose K ⊂ X a compact subspace. Write U = X −K. Suppose x ∈ U. For any element y ∈ K, choose
disjoint open neighborhoods V
y
of x and W
y
of y. The collection ¦W
y
| y ∈ K] is an open cover of K, so there
is a finite collection of points y
1
, y
2
, . . . , y
n
such that the subcollection ¦W
y
i
| i ∈ ¦1, 2, . . . , n]] covers n. Now the
intersection
¸
n
i =1
V
y
i
is an open neighborhood of x that is disjoint from K, i.e., completely contained in U.
Exercise 31. — Suppose X a Hausdorff space, and suppose K, L ⊂ X two disjoint compact subspaces. Show that
there exist open sets U, V ⊂X such that K ⊂ U, L ⊂V, and U ∩V =∅.
Definition 4.2.8. — A space is said to be locally compact if and only if any point has a compact neighborhood.
Example 4.2.9. — Any compact space is of course locally compact. The euclidean spaces R
n
are locally compact,
and any discrete space is locally compact.
More exotically, suppose X any set, and suppose x ∈ X. Then the particular point topology for (X, x) is the
topology in which
Op(X) :=¦U ∈ ´(X) | x ∈ U] ∪¦∅].
The particular point topology is always locally compact.
Example 4.2.10. — The space Q with the subspace topology is not locally compact, since its compact subspaces
have empty interior.
Exercise 32. — Suppose X a locally compact Hausdorff space X, and suppose K ⊂ X a compact subspace. Show
that for any point x ∈ X − K, there exists a neighborhood U of x and an open set V containing K such that
U ∩V =∅, and the closure V is compact.
Lemma 4.2.11. — Suppose X a locally compact Hausdorff space X, and suppose U an open subset that contains a
compact subspace K. Then there exists an open set W whose closure is compact such that K ⊂W ⊂W ⊂ U.
4.2. LOCALLY COMPACT HAUSDORFF SPACES 29
Proof. — If U = X, then exercise 32 applies directly. If not, then for every point x ∈ X −U, the exercise 32 gives
an open neighborhood V
x
of x and an open set W
x
whose closure is compact such that
K ⊂V
x
and V
x
∩W
x
=∅.
Now the family ¦(X −U) ∩W
x
| x ∈ X −U] is a family of compact subspaces of X whose intersection is empty.
Consequently there are a finite number of elements x
1
, x
2
, . . . , x
n
∈ X − U such that the intersection (X − U) ∩
¸
n
i =1
W
x
is empty. Now W :=
¸
n
i =1
W
x
does the job.
Proposition 4.2.12. — Suppose X a Hausdorff space. Then the following are equivalent.
(4.2.12.1) Every point x ∈ X has a fundamental system of compact neighborhoods.
(4.2.12.2) Every point x ∈ X has a closed compact neighborhood.
(4.2.12.3) X is locally compact.
Proof. — The first two conditions imply the third by definition. The third also implies the second, since compact
subspaces of a Hausdorff space are closed. Thus it remains only to show that the third condition implies the first.
So suppose X is locally compact, and suppose V an open neighborhood of x; I can apply the previous lemma to get
an open neighborhood W of x such that W is compact and contained in V.
4.2.13. — We often think of locally compact and Hausdorff as a single condition, which we abbreviate as LCH.
Lemma 4.2.14. — A closed subspace of an LCH space is itself LCH.
Proof. — The intersection of a closed set and a compact set is compact. (Why does our result follow from this?)
Exercise 33. — More generally, a subspace of an LCH space X is itself LCH if and only if it can be written as the
complement of a closed subspace inside another closed subspace.
4.2.15. — We now turn to the Alexandroff or one-point compactification of a space X.
Definition 4.2.16. — Suppose X a space. Then define X

to be the set X |¦∞], with the following topology: a set
U ⊂X

will be said to be open if and only if either U is an open subset of X, or else U contains ∞, and X −U is
closed and compact.
Exercise 34. — Show that this indeed defines a topology on X

, and the resulting space is compact.
Theorem 4.2.17. — Suppose X a space. We have the following facts about the Alexandroff compactification.
(4.2.17.1) The inclusion map i : X

X

is continuous and open.
(4.2.17.2) The image of i is a dense open set in X unless X is already compact.
(4.2.17.3) The Alexandroff compactification X

is Hausdorff if and only if X is LCH.
Exercise 35. — Prove the previous theorem.
Exercise 36. — Suppose Y an LCH space, and suppose Z ⊂ Y a closed subspace. Compare the Alexandroff com-
pactification (Y −Z)

and the quotient space Y/Z. Are they homeomorphic? Prove it or give a counterexample. If
it’s true, is the LCH condition necessary? If it’s false, can you add conditions on Z and Y so that it will be true?
Exercise 37. — Describe the Alexandroff compactifications of the following spaces in terms of more familiar spaces.
(37.1) R
n
.
(37.2) A discrete space.
(37.3) The subspace ¦(n, x) | n ∈ Zx ∈ R] ⊂R
2
.
(37.4) Q with the subspace topology.
CHAPTER 5
COUNTABILITY AND SEPARATION
5.1. Taxonomy of separation
Definition 5.1.1. — Suppose X a space. Then two points x, y ∈ X are topologically distinguishable if they do not
have the same collection of open neighborhoods. Otherwise, the points are said to be topologically indistinguishable.
Exercise 38. — Observe that for any space X, if x, y ∈ X topologically distinguishable points, then x (= y. Give an
example to show that the converse need not hold.
Definition 5.1.2. — Suppose X a space. Then a neighborhood of a subset A⊂ X is a set containing whose interior
contains A.
Definition 5.1.3. — Suppose X a space, and suppose A, B ⊂X two subspaces of X.
(5.1.3.1) The sets Aand B are said to be separated if both A∩B and A∩B are empty.
(5.1.3.2) The sets A and B are said to be separated by neighborhoods if there exists disjoint neighborhoods U and V
and Aand B, respectively.
(5.1.3.3) The sets Aand B are said to be separated by closed neighborhoods there exists disjoint closed neighborhoods
U and V and Aand B, respectively.
(5.1.3.4) The sets A and B are said to be separated by a function if there exists a continuous map f : X

[0, 1]
such that f |
A
=0 and f |
B
=1.
(5.1.3.5) The sets A and B are said to be precisely separated by a function if there exists a continuous map f :
X

[0, 1] such that A= f
−1
¦0] and B = f
−1
¦1].
5.1.4. — Suppose X a space, and suppose A, B ⊂ X. It is easy to see that each of these conditions follows from the
next.
(5.1.4.1) Aand B are separated.
(5.1.4.2) Aand B are separated by neighborhoods.
(5.1.4.3) Aand B are separated by closed neighborhoods.
(5.1.4.4) Aand B are separated by a function.
(5.1.4.5) Aand B are precisely separated by a function.
Definition 5.1.5. — Suppose X a space. Then we introduce the Trennungsaxiome.
T
0
One says that X is Kolmogoroff or T
0
if any two distinct points are topologically distinguishable.
T
1
One says that X is T
1
if any two distinct points are separated.
T
2
One says that X is Hausdorff or T
2
if any two distinct points are separated by neighborhoods.
T
2
1
2
One says that X is Urysohn or T
2
1
2
if any two points are separated by closed neighborhoods.
T
3
One says that X is regular Hausdorff or T
3
if it is T
0
and if any closed subset Z ⊂ X and any point x / ∈ Z are
separated by neighborhoods.
T
3
1
2
One says that X is Tychonoff or T
3
1
2
if it is T
0
and if any closed subset Z ⊂X and any point x / ∈ Z are separated
by a function.
32 CHAPTER 5. COUNTABILITY AND SEPARATION
T
4
One says that X is normal Hausdorff or T
4
if it is T
1
and if any two closed subsets can be separated by
neighborhoods.
T
5
One says that X is completely normal Hausdorff or T
5
if it is T
1
and if any two separated sets are separated by
neighborhoods.
T
6
One says that X is perfectly normal Hausdorff or T
6
if it is T
1
and if any two closed sets are precisely separated
by a function.
5.1.6. — For any space X, topological indistinguishability is an equivalence relation on the points of X. The quo-
tient space under this equivalence relation is the Kolmogoroff quotient KX of X. This space is automatically Kol-
mogoroff. This is frequently done in analysis.
Lemma 5.1.7. — The following are equivalent for a space X.
(5.1.7.1) X is T
1
.
(5.1.7.2) X is T
0
, and any two topologially distinguishable points are separated.
(5.1.7.3) Points of X are closed.
(5.1.7.4) Every finite set in X is closed.
Theorem 5.1.8. — For any i , j ∈ ¦0, 1, 2, 2
1
2
, 3, 3
1
2
, 4, 5, 6], if i ≤ j , then any T
j
space is also T
i
.
Example 5.1.9. — Let us convince ourselves that these conditions are distinct.
(5.1.9.1) You have already given an example of a space that is not T
0
.
(5.1.9.2) The Sierpi´ nski space Σ
0
is T
0
but not T
1
.
(5.1.9.3) The cofinite topology on an infinite set is T
1
but not T
2
.
(5.1.9.4) Let X =¦(x, y) ∈ Q
2
| y ≥0], and fix an irrational number θ. Topologize this set in the following manner:
for any point (x, y) ∈ X, and any >0, let
N

(x, y) =¦(x, y)] ∪
¸
x +
y
θ
−, x +
y
θ
+

x −
y
θ
−, x −
y
θ
+

∩Q
.
,
and contemplate the topology generated by these sets; the sets N

(x, y) form a base for this topology. The
closure of each basic neighborhood N

(x, y) contains the union of four strips of slope ±θ extending from
the intervals (x +
y
θ
−, x +
y
θ
+) and (x −
y
θ
−, x −
y
θ
+). Hence the closures of every two open sets
intersect. This is a T
2
space that is not T
2
1
2
.
(5.1.9.5) Let H be the upper half-plane ¦(x, y) ∈ R
2
| y ≥ 0]. Topologize this set in the following manner. Begin
with all the open sets in the set P =¦(x, y) ∈ R
2
| y >0], and add in sets of the form¦ p] ∪(U ∩P), where
p =(x, 0), and U is an open neighborhood of p in the topology on R
2
. This is a topology on H; and it is
not difficult to see that it is T
2
1
2
. However, the complement of a neighborhood of p = (0, 0) intersects the
closure of every neighborhood of p. So H is not T
3
.
(5.1.9.6) The Tychonoff corkscrew is the standard example of a space that is T
3
but not T
3
1
2
, but it’s an absurd space
that just makes you sorry you ever tried to contemplate it. Here’s one that’s slightly simpler (though
still pretty absurd), due to A. B. Raha. Let’s call it the Raha space. For every even integer n, set T
n
:=
¦n] ×(−1, 1), and let X
1
=
¸
n even
T
n
. Now let (t
k
)
k≥1
be an increasing sequence of positive real numbers
converging to 1. For every odd integer n, set T
n
:=
¸
k≥1
¦(x, y) ∈ R
2
| (x − n)
2
+ y
2
= t
2
k
], and let
X
2
=
¸
n odd
T
n
. Now let X = ¦a, b] ∪
¸
n∈Z
T
n
. Topologize X so that: (1) every point of X
2
except the
points (n, t
k
) are isolated; (2) a neighborhood of (n, t
k
) consists of all but finitely many elements of ¦(x, y) ∈
R
2
| (x −n)
2
+y
2
= t
2
k
]; (3) a neighborhood of a point (n, y) ∈ X
1
consists of all but a finite number of
points of ¦(z, y) | n −1 < z < n +1] ∩(T
n−1
∪T
n
); (4) a neighborhood of a is a set U
c
containing a and all
points of X
1
∪X
2
with x-coordinate greater than a number c; (5) a neighborhood of b is a set V
d
containing
b and all points of X
1
∪X
2
with x-coordinate less than a number d. This is a space that is T
3
, but every
continuous map f : X

R has the property that f (a) = f (b), so it is not T
3
1
2
.
(5.1.9.7) Consider the space Map(R, R) of all maps R

R with the topology of pointwise convergence. This
space is T
3
1
2
but not T
4
.
5.2. NORMALITY, URYSOHN’S LEMMA, AND THE TIETZE EXTENSION THEOREM 33
(5.1.9.8) The Tychonoff plank is the space P :=(ω+1) ×(ω
1
+1). It is compact and Hausdorff, and hence it is T
4
.
It is not, however T
5
, because the space P −¦(ω, ω
1
)] is not normal. (So you walk the plank . . . Get it? It’s
clever.)
(5.1.9.9) Suppose X an uncountable set, and suppose x ∈ X. Then we have the variant on the particular point
topology, which is called the Fort space topology. The closed sets are declared to be those sets V such that
either V is finite or x ∈ V. Then it is simple to verify that X is T
5
, and a short argument shows it cannot
be T
6
.
The moral of the story is not, of course, to memorize exotic counterexamples. The point is to realize that there
are examples that prevent the reversal of any of the implications T
j
⇒T
i
for i ≤ j . Also, in each case the example
listed is the simplest one I could find. This suggests something interesting: the differences between T
2
1
2
and T
2
and
T
3
is pretty small, as is the difference between T
3
1
2
and T
3
. However, the difference between T
4
and T
3
1
2
is pretty
significant, contrary to what might have been your expectation. This shows us two relatively bright lines separating
the “easy” separation axioms T
0
−T
1
fromthe “midrange” separation axioms T
2
−T
3
1
2
, and separating the “midrange”
axioms from the “hard” separation axioms T
4
−T
6
.
5.1.10. — There is another condition, called sobriety, that is better suited to applications in algebraic geometry.
Definition 5.1.11. — A space X is said to be irreducible (or hyperconnected) if no two nonempty open subsets of X
are disjoint.
Exercise 39. — Show that the following are equivalent for a space X.
(39.1) X is irreducible.
(39.2) Every nonempty open set of X is dense.
(39.3) X cannot be written as the union of two proper nonempty closed subsets.
(39.4) The interior of every proper closed subset is empty.
Definition 5.1.12. — A point of a space X is said to be generic if it is dense in X.
Definition 5.1.13. — A space X is said to be sober if every irreducible closed subspace Z ⊂ X has a unique generic
point.
5.1.14. — Every Hausdorff space is sober, and every sober space is Kolmogoroff, but there is no comparison be-
tween sobriety and T
1
.
Exercise

40. — Suppose X and Y sober spaces, and consider the posets (Op(X), ⊂) and (Op(Y), ⊂). Show that if
(Op(X), ⊂) and (Op(Y), ⊂) are isomorphic as posets, then X and Y are homeomorphic.
5.2. Normality, Urysohn’s lemma, and the Tietze extension theorem
Definition 5.2.1. — A space X is said to be normal if disjoint closed sets have disjoint neighborhoods.
5.2.2. — A normal T
1
space is normal Hausdorff, or T
4
. The results of this section do not require the T
1
property,
so we do not include it in our definition.
Lemma 5.2.3. — Suppose D ⊂ R
>0
a dense subset, and suppose that for each element t ∈ D, there is a subset F
t
of a set
X such that
(5.2.3.1) if t < s , then F
t
⊂ F
s
, and
(5.2.3.2) the sets F
t
cover X.
Then for any x ∈ X, let f (x) =inf¦t | x ∈ F
t
]. Then for any element s ∈ R, one has
¦x ∈ X | f (x) < s ] =
¸
¦F
t
| t ∈ D and t < s ]
and
¦x ∈ X | f (x) ≤ s ] =
¸
¦F
t
| t ∈ D and t > s ].
Proof. — An exercise in set theory (not to be handed in).
34 CHAPTER 5. COUNTABILITY AND SEPARATION
Lemma 5.2.4. — Suppose now X a space, and suppose D ⊂ R
>0
a dense subset. If for each element t ∈ D, there is an
open subset F
t
⊂X such that
(5.2.3.1) if t < s , then F
t
⊂ F
s
, and
(5.2.3.2) the sets F
t
cover X,
then the function f defined by f (x) =inf¦t | x ∈ F
t
] is continuous.
Proof. — It’s enough to check that the inverse image of the sets (−∞, s ) and (−∞, s ] are open and closed (respec-
tively) for every s ∈ R.
The previous lemma shows that the inverse images of sets of the first kind, (−∞, s ), are the union of open sets,
hence open.
It now follows from the previous lemma that the inverse images of sets of the second kind, (−∞, s ], are closed
(and hence we are done) if the following equality obtains:
¸
¦F
t
| t ∈ D and t > s ] =
¸
¦F
t
| t ∈ D and t > s ].
It is clear that the first set is contained in the second. On the other hand, for any t ∈ D with t > s , there is an
element r ∈ D with s < r < t , whence F
r
⊂ F
t
. This completes the proof.
Lemma 5.2.5 (Urysohn). — Suppose X a normal space, and suppose A and B two disjoint closed subsets of X. Then A
and B are separated by functions, i.e., there is a continuous map f : X

[0, 1] such that f |
A
=0 and f |
B
=1.
Proof. — We wish to apply the previous lemma with the set
D =¦x ∈ Q| x = p2
−q
, where p, q ∈ N].
We define our sets F
t
for t ∈ D thus.
(5.2.5.1) If t >1, then let F
t
=X.
(5.2.5.2) Let F
1
=X −B.
(5.2.5.3) Let F
0
be an open set containing Asuch that F
0
is disjoint from B.
(5.2.5.4) For 0 < t < 1, we proceed by recursion. Write t = (2m +1)2
−n
and let F
t
be an open set containing
F
(2m)2
−n whose closure F
t
⊂ F
(2m+2)2
−n (by normality).
It is plain to see that the conditions of the previous lemma apply, and it constructs our continuous function f .
By construction, f |
A
=0 and f |
B
=1.
Theorem 5.2.6 (Tietze extension). — Suppose X is a normal space, and suppose A a closed subspace of X. Then any
continuous map A

[−1, 1] can be extended to a continuous map X

[−1, 1].
CHAPTER 6
METRIZATION THEOREMS
6.1. First and second countability
Definition 6.1.1. — Suppose X a space. Then X is said to satisfy the first axiom of countability or to be first-countable
if for every point x ∈ X there is a countable fundamental system of neighborhoods of x.
The space X is said to satisfy the second axiom of countability or to be second-countable if the topology on X has
a countable base.
Lemma 6.1.2. — The second axiom of countability implies the first.
Example 6.1.3. — The ordinal ω
1
(with the order topology) satisfies the first but not the second axiom of count-
ability, and the ordinal ω
1
+1 satisfies neither axiom of countability.
Lemma 6.1.4. — Any subspace of a first-countable (respectively, second-countable) space is first-countable (resp., second-
countable), as is any countable product of first-countable (resp., second-countable) spaces. Additionally, any continuous
open image of a second-countable space is second-countable.
Exercise 41. — Give an example demonstrating that an uncountable product of second-countable spaces need not
be first-countable.
Exercise 42. — Suppose X a first-countable space. Show that for any subset A⊂ X, a point x ∈ X is contained in
the closure Aif and only if there is a sequence of points of Aconverging to x. Show also that a map f : X

Y to a
topological space Y is continuous if and only if, for any convergent sequence x
n
→x, the sequence f (x
n
) converges
to f (x). Finally, show that X is sequentially compact if and only if it is countably compact.
Example 6.1.5. — The real line R is second countable, as is any countable product of R with itself.
Exercise 43. — Suppose X a second-countable space; show that the set Op(X) has cardinality not greater than

1
=#R (i.e., the cardinality of the continuum).
Definition 6.1.6. — A space X is said to be separable if it has a countable dense subset. A space X is said to be
Lindelöf if every open cover has a countable subcover.
Theorem 6.1.7. — Second-countable spaces are both separable and Lindelöf.
Proof. — Suppose 0 =¦B
1
, B
2
, . . . ] a countable base for a space X.
We show first that X is Lindelöf. Suppose ¹ an open covering of X. For every natural number j , select, if
possible, an element U
j
∈ ¹ containing B
j
. This defines a countable subset ¹
/
:=¦U
j
1
, U
j
2
, . . . ] ⊂¹ indexed by a
subset J ⊂N. I claim that this is a subcover. Indeed, for any point x ∈ X, an element U ∈ ¹ contains x, and so for
some j , we have x ∈ B
j
⊂ U. Hence j ∈ J , whence some U
j
∈ ¹
/
contains x.
Showing that X is separable is even easier. Simply select a point x
j
∈ B
j
for every j ∈ N. This set is dense.
(Why?)
36 CHAPTER 6. METRIZATION THEOREMS
Exercise 44. — Suppose X a metrizable space. Show that the following conditions are equivalent on X: (1) X
satisfies the second axiom of countability; (2) X is separable; and (3) X is Lindelöf. Check that the Sorgenfrey line
is first-countable, separable, and Lindelöf, but not second-countable.
Lemma 6.1.8 (Tychonoff). — Every T
3
Lindelöf space is T
4
.
Proof. — Suppose A and B disjoint closed subsets of X. Since X is regular, for any point a ∈ A there is a neighbor-
hood of a whose closure fails to intersect B, and thus the family ¹ of open sets whose closures do not intersect B
cover A. Similarly, the family · of open sets whose closures do not intersect Acover B. Now we form a cover
¹ ∪· ∪¦X −(A∪B)]
of X.
It now follows from the Lindelöf property that there are countable subcovers ¦U
1
, U
2
, . . . ] ⊂ ¹ of A and
¦V
1
, V
2
, . . . ] ⊂· of B. Now set
U
/
r
:= U
r

¸
m≤r
V
m
and V
/
s
:=V
s

¸
m≤n
U
m
for every natural number r and s . One verifies immediately that U
/
r
∩V
/
s
=∅for every pair (r, s ) of natural numbers,
and so
U =
¸
r ≥1
U
/
r
and V =
¸
s ≥1
V
/
s
are disjoint open neighborhoods of Aand B, respectively.
Exercise 45. — Show that any locally compact Hausdorff space is Tychonoff.
Example 6.1.9. — The product of two Lindelöf spaces need not be Lindelöf; indeed, the product of the Sorgenfrey
line with itself is the Sorgenfrey plane, which is not Lindelöf. Similarly, a subspace of a Lindelöf space need not be
Lindelöf.
6.2. Urysohn’s metrization theorem
Exercise

46. — Prove the following
Theorem 6.2.1 (Embedding). — A space is a Tychonoff space if and only if it is homeomorphic to a subspace of a cube
[0, 1]
A
for some set A.
Begin by proving the following
Lemma 6.2.2 (Embedding). — Suppose X a space, and suppose ¨ a family of continuous maps f : X

Y
f
(valued
in possibly many different spaces).
(6.2.2.1) The evaluation map is a continuous map E : X

¸
f ∈¨
Y
f
.
(6.2.2.2) The evaluation map E is open if for any closed subset A⊂X and any point x ∈ X −A, there exists an element
f ∈ ¨ such that f (x) / ∈ f (A).
(6.2.2.3) The evaluation map E is injective if and only if for any two distinct point x, y ∈ X, there exists an element
f ∈ ¨ such that f (x) (= f (y).
Theorem 6.2.3 (Urysohn metrization). — The following are equivalent for a space X.
(6.2.3.1) X is second countable and T
3
.
(6.2.3.2) X is homeomorphic to a subspace of the Hilbert cube [0, 1]
N
.
(6.2.3.3) X is separable and metrizable.
6.3. THE BING–NAGATA–SMIRNOV METRIZATION THEOREM 37
Proof. — It is not hard to see that the Hilbert cube is separable and metrizable; hence so is any subspace. (Why?)
Any metrizable space is T
3
, and any separable metrizable space is second countable. (Check this!)
Hence it remains to show that a T
3
space that satisfies the second axiom of countability is homeomorphic to a
subspace of the Hilbert cube. Using the embedding lemma above, it is enough to construct a countable family ¨ of
continuous functions X

[0, 1] such that for any closed subset A⊂ X and any point x ∈ X −A, there exists an
element f ∈ ¨ such that f (x) / ∈ f (A). Suppose 0 a countable base for the topology of X, and set
· :=¦(U, V) ∈ 0×0 | U ⊂V].
One sees that · is a countable set. Since X is T
3
and Lindelöf, it is also T
4
by Tychonoff’s lemma. So, using
Urysohn’s lemma, choose, for any (U, V) ∈ ·, a Urysohn function f
(U,V)
: X

[0, 1] such that f |
U
= 0 and
f |
X−V
=1; set
¨ :=¦ f
(U,V)
| (U, V) ∈ ·].
This is a countable set of functions, and it remains to show that it has the desired properties. Suppose A⊂X a closed
subset, and suppose x ∈ X −A. We may now choose a basic open V ∈ 0 such that x ∈ U ⊂X −Aand a basic open
U ∈ 0 such that x ∈ U ⊂V. (Why?) Since (U, V) ∈ ·, the corresponding function f
(U,V)
does the job.
Exercise 47. — Fill in the details I skipped in the proof of the Urysohn metrization theorem, and write up a com-
plete proof.
6.3. The Bing–Nagata–Smirnov metrization theorem
Definition 6.3.1. — A family ¹ of subspaces of a topological space X is said to be locally finite if every point is
contained in a neighborhood that intersects only finitely many elements of ¹. One says that ¹ is discrete if every
point is contained in a neighborhood that intersects only one element of ¹.
Afamily · of subspaces of a space X is said to be σ-locally finite (respectively, σ-locally discrete) if it can be written
as the union of a countable collection of locally finite (resp., locally discrete) families of subspaces of X.
Theorem 6.3.2 (Bing–Nagata–Smirnov metrization). — The following are equivalent for a space X.
(6.3.2.1) X is metrizable.
(6.3.2.2) X is T
3
, and there is a σ-locally finite base for X.
(6.3.2.3) X is T
3
, and there is a σ-locally discrete base for X.
Outline of proof. — First, observe that the final condition implies the penultimate condition. The first two lemmas
here will show that the second condition implies the first.
Lemma 6.3.3. — A T
3
space whose topology has a σ-locally finite base is T
4
.
Lemma 6.3.4. — A T
3
space whose topology has a σ-locally finite base is metrizable.
Lemma 6.3.5. — Any open cover of a metrizable space has a σ-discrete open refinement.
It is now an immediate consequence that a metrizable space has a σ-discrete base.
6.3.6. — One question that might naturally arise here is: so? I can show that if a space satisfies some abstract
conditions, then it can be given a metric that gives rise to its topology. Let’s list some of the purely topological
corollaries of the metrizability of a space X.
(6.3.6.1) A subset U ⊂X is open if and only if no sequence of X −U converges to a point of U.
(6.3.6.2) Dually, for any subspace A ⊂ X, a point x ∈ X is a member of the closure A if and only if there is a
sequence of points in Athat converge to A.
(6.3.6.3) For any space Y, function f : X

Y is continuous if and only if, for any convergent sequence x
n
→x
in X, the sequence f (x
n
) → f (x).
(6.3.6.4) Asubspace A⊂X is compact if and only if it is countably compact, if and only if it is sequentially compact.
(6.3.6.5) X is T
6
.
(6.3.6.6) X is paracompact, i.e., every open cover of X has a locally finite open refinement.
ENTR’ACTE
CHAPTER 7
SHEAF THEORY
7.1. Presheaves
Example 7.1.1. — Let us begin right away with a motivational example. Suppose X is a topological space, and let’s
contemplate, for any open set U ∈ Op(X), the set
O
cts
X
(U) :=¦ f : U

R | f is continuous].
It is not difficult to see that if I have a continuous function f on U, then I may restrict it to a continuous function
f |
V
on any open subset V ⊂ U. This defines a restriction map
ρ
V⊂U
: O
cts
X
(U)

O
cts
X
(V) .
I can, of course, be rather silly in how I restrict: I might have taken V = U; then my restriction map is simply the
identity.
On the other hand, what if I consider three open sets W ⊂ V ⊂ U in X? If I restrict a function f on U to a
function on V, and thence to a function on W, I see that I get the same answer as if I had restricted f all the way
from U to W. That is, we get an equality of maps
ρ
W⊂V
◦ ρ
V⊂U

W⊂U
.
We can now attempt to study, not the individual sets O
cts
X
(U) or the restriction maps ρ
V⊂U
, but the entire system
made up of these:
O
cts
X
:=

(O
cts
X
(U))
U∈Op(X)
, (ρ
V⊂U
)
V⊂U∈Op(X)

.
Example 7.1.2. — There was nothing particularly special about the topological space R in the above example. I
could try the same trick with any “target” topological space Y: for any open set U ∈ Op(X), write
O
Y
X
(U) :=¦ f : U

Y | f is continuous].
Again, it’s pretty clear that we can restrict continuous maps, and so we have a similar structure
O
Y
X
:=

(O
Y
X
(U))
U∈Op(X)
, (ρ
V⊂U
)
V⊂U∈Op(X)

.
Example 7.1.3. — More interestingly, suppose p : Y

X a continuous map. Then we can make a similar defini-
tion: for any open set U ∈ Op(X), set
Γ( p)(U) =Γ(X/Y)(U) :=¦s : U

Y | s is continuous and p ◦ s =id
U
].
We say that Γ(X/Y)(U) is the set of sections of p over U.
Definition 7.1.4. — Suppose X a topological space. Then a presheaf ¨ on X is the following data:
(7.1.4.A) for any open set U ∈ Op(X), a set ¨(U), and
(7.1.4.B) for any open sets U, V ∈ Op(X) with V ⊂ U, a restriction map ρ
V⊂U
: ¨(V)

¨(U) .
These data are subject to the following conditions.
(7.1.4.1) For any open set U ∈ Op(X), the map ρ
U⊂U
is the identity map on ¨(U).
42 CHAPTER 7. SHEAF THEORY
(7.1.4.2) For any three open sets U, V, W ∈ Op(X) with W ⊂V ⊂ U, we have
ρ
W⊂V
◦ ρ
V⊂U

W⊂U
: ¨(U)

¨(W) .
In other words, the data of a presheaf is a pair
¨ :=

(¨(U))
U∈Op(X)
, (ρ
V⊂U
)
V⊂U∈Op(X)

,
and these data are subject to the axioms ρ
U⊂U
=id, and ρ
W⊂V
◦ ρ
V⊂U

W⊂U
.
For an open set U ∈ Op(X), an element s ∈ ¨(U) is sometimes called a section of ¨ over U.
7.1.5. — This sort of pattern appears all the time: to any “object” of some collection (in this case Op(X)), you
assign a set, and to any “relation” between two objects (in this case set inclusion), you assign a map between the
corresponding sets. Sometimes that map goes in the same direction as your relation; sometimes (as here), the map
goes in the opposite direction.
When you have these data, then you want to impose some conditions. These conditions always amount to the
same thing: when the relation is trivial (U ⊂ U, for example), the corresponding map has to be the identity, and
composing relations (W ⊂V ⊂ U) gives rise to compositions of maps.
These data, with these conditions, are collectively called a functor in general. When the maps go in the same
direction as the relation, then the functor is said to be covariant; when the maps go in the opposite direction, the
functor is said to be contravariant. Thus presheaves are particular kinds of functors. We won’t say more about this
here, but you can expect functors to appear again in your mathematics education!
7.1.6. — One of the most challenging aspects of modern mathematics for beginners is to maintain a clear idea
which of a list of bullet points are data, and which are conditions. In the case of a presheaf ¨ on a space X, it’s a
little more subtle than anything we’ve seen so far: the data are the choice of all the sets ¨(U) and the choice of all
the restriction maps ρ
V⊂U
: ¨(V)

¨(U) . It isn’t enough to say that there are suitable maps out there; one has
to specify them as part of the information of a presheaf.
Example 7.1.7. — Our definition is general enough to allow for some “degenerate” examples of presheaves. For
any set S, one may form the constant presheaf ´
S
at S, which assigns to any open set U the set S, and to any open
sets U, V ∈ Op(X) with V ⊂ U the identity map on S.
Example 7.1.8. — It’s not difficult to see that for any topological space X, the objects O
Y
X
(and hence the object
O
cts
X
) are presheaves on X.
Example 7.1.9. — Likewise, for any continuous map p : Y

X , the object Γ(Y/X) is a presheaf on X. Let’s
contemplate this example in a few particular cases.
(7.1.9.1) Suppose p is a homeomorphism. Then Γ(Y/X) is not so very interesting. For any open set U ∈ Op(X),
there is exactly one section of p over U. It’s not hard to see that, in this case, Γ(Y/X) is the constant
presheaf on one point.
(7.1.9.2) Now if p is not a homeomorphism, but a local homeomorphism (so that every point of Y in contained
in a neighborhood V such that the restriction of p to V is open and injective), then the story becomes
a lot more interesting. First, consider the folding map ∇ : S
1
|S
1
S
1
; in this case Γ(∇) is somewhat
more interesting: to any interval (a, b) it assigns a two-point set. But if the open set is more complicated,
the value of Γ(∇) is more complicated as well. If, for example, U is the disjoint union of two intervals,
Γ(∇)(U) is a four point set. And in general, Γ(∇)(V) has cardinal 2
n
, where n is the number of connected
components of V. Note, in particular, that Γ(∇)(S
1
) is a two-point set also.
(7.1.9.3) But now consider the map p : S
1
S
1
given by ξ


ξ
2
(with respect to complex multiplication,
thinking of S
1
⊂C). This too is a local homeomorphism, but observe that the presheaf Γ( p) has no global
sections, i.e., Γ( p)(S
1
) = ∅. One way to think of this is to think that there is no single, consistent way to
extract square roots of complex numbers. This is quite striking: the fiber over each point of p is two points,
just as in the previous example. But the presheaf Γ( p) has shown us that there is something nontrivial about
our map p.
7.1. PRESHEAVES 43
(7.1.9.4) Here’s yet another example in the same vein: consider the exponential map exp : C

C
×
:=C−¦0].
We can contemplate the presheaf Γ(exp), and we see immediately that over a sufficiently small open neigh-
borhood U of any point in C
×
, there are countably infinitely many elements of Γ(exp)(U); these corre-
spond to the branches of the logarithm: if we write a complex number z = r exp(i θ), then we might write
log(z) = log|r | +i θ; however, this is ambiguous, I can add 2kπ to θ. Once again, however, one sees that
one cannot make a consistent choice of a branch of the logarithm: there are no global sections of Γ(exp)!
Example 7.1.10. — Suppose X a space, and suppose A∈ Op(X) a particular fixed open set. We have an associated
presheaf h
A
defined by the rule
h
A
(U) :=

if U ⊂A
∅ else.
In this case, the restriction maps are forced on us: for any open sets V, U ∈ Op(X) with V ⊂ U, there is necessarily
only one map h
A
(U)

h
A
(V) . Observe that we cannot expect this in general: ordinarily, we’ll have to put some
effort into describing the restriction maps!
This gives us a way (in fact, a very powerful way) of “representing” open sets A ∈ Op(X) as presheaves; conse-
quently, the presheaves h
A
are known as representable presheaves, or presheaves represented by A∈ Op(X).
Exercise 48. — Suppose X a space, A ∈ Op(X) an open subset. Suppose I attempted to define a presheaf h
A
the
other way around from our representable friend:
h
A
(U) :=

∅ if U ⊂A
else.
What goes wrong here? Why does this not define a presheaf?
Example 7.1.11. — Suppose S is a set, and suppose X a space with a distinguished point x ∈ X. Then the skyscraper
presheaf at x with value S is defined by the rule
S
x
(U) :=

S if x ∈ U
else.
7.1.12. — We can also ask for additional structure on a presheaf ¨. If all the sets ¨(U) are equipped with a group
structure, and all the restriction maps ρ
V⊂U
are group homomorphisms, then the presheaf ¨ is said to be a presheaf
in groups. Similarly, if all the sets ¨(U) are equipped with an abelian group (respectively, ring, vector space, algebra,
...) structure, and all the restriction maps ρ
V⊂U
are group (resp., ring, vector space, algebra, ...) homomorphisms,
then the presheaf ¨ is said to be a presheaf in abelian groups (resp., rings, vector spaces, algebras, ...). The key point
here is that both the sets ¨(U) and the maps ρ
V⊂U
carry this added structure.
Example 7.1.13. — Suppose X =R
n
with the euclidean topology. Then for any open set U ⊂R
n
, we can contem-
plate the set
O
an
X
(U) =(
ω
(U) :=¦ f : U

R | f is analytic].
(Recall that a function f on an open set U is analytic if it is infinitely differentiable and, for any point x ∈ U, there
is a neighborhood of x in U such that the Taylor series of f at x converges to f . This is surely a presheaf as well;
moreover, since the sum, difference, and product of any two analytic functions is analytic, we can see that O
an
X
is a
sheaf of rings (even of R-algebras).
Definition 7.1.14. — A morphism or simply map of presheaves φ: ¨

¼ is the following data: for any open set
U ∈ Op(X), a map (called the component of φ)
φ
U
: ¨(U)

¼(U) ,
44 CHAPTER 7. SHEAF THEORY
subject to the following condition: for any open sets U, V ∈ Op(X) with V ⊂ U, the following diagram commutes:
¨(U)
ρ
V⊂U

φ
U

¼(U)
ρ
V⊂U

¨(V)
φ
V

¼(V),
that is, for any section s ∈ ¨(U), one has
ρ
V⊂U

U
(s )) =φ
V

V⊂U
(s )).
Given morphisms of presheaves φ: ¨

¼ and ψ : ¼

· , we can form the composite ψ◦ φ: ¨

·
in the following manner: for any open set U ∈ Op(X), set
(ψ◦ φ)
U

U
◦ φ
U
.
This defines a morphism of presheaves ¨

· as desired.
A morphism of presheaves φ : ¨

¼ is said to be an isomorphism if there exists a morphism of sheaves
ψ: ¼

¨ such that both ψ◦ φ=id and φ◦ φ=id.
Exercise 49. — (49.1) Check that our description of the composite of two morphisms of presheaves is indeed again
a morphism of presheaves.
(49.2) Verify that if we have three morphisms of presheaves φ: ¨

¼ , ψ: ¼

· , and χ : ·

· , then
one has
(χ ◦ ψ) ◦ φ=χ ◦ (ψ◦ φ).
(49.3) Finally, show that a morphism of presheaves φ: ¨

¼ is an isomorphism if and only if, for any open set
U ∈ Op(X), the components φ
U
are all bijections.
Definition 7.1.15. — We say that a morphism ¨

¼ of presheaves is injective if for every open set U ∈ Op(X),
the corresponding map ¨(U)

¼(U) is injective. (We’ll need to be a little more cautious when we define sur-
jectivity.)
Notation 7.1.16. — For any space X, and any presheaves ¨ and ¼, write Mor
X
(¨, ¼) for the set of all morphisms
of presheaves ¨

¼ .
Exercise 50. — Show that for any space X, and any pair of open sets U, V ∈ Op(X) with V ⊂ U, there is a unique
morphism of presheaves h
V

h
U
.
Exercise

51. — Show that for any space X, any open set U ∈ Op(X), and any presheaf ¨ on X, there is a natural
bijection
Φ
U
: ¨(U)

=Mor
X
(h
U
, ¨).
This word natural gets thrown around a lot. In this case, we can say exactly what we mean by it: for any pair of
open sets U, V ∈ Op(X) with V ⊂ U, we have our unique morphism of presheaves j : h
V

h
U
. Composition
with j induces a map −◦ j : Mor
X
(h
U
, ¨)

Mor
X
(h
V
, ¨) . (How?) For an extra challenge, show that in this
case, the following diagram of sets commutes:
¨(U)
ρ
U⊂V

Φ
U

Mor
X
(h
U
, ¨)
−◦j

¨(V)
Φ
V

Mor
X
(h
V
, ¨).
Exercise 52. — Show that for any two spaces X and Y, the presheaf O
Y
X
of local Y-valued functions is isomorphic
to the presheaf Γ(Y ×X/X) of local sections of the projection map Y ×X

X .
7.3. SHEAVES 45
7.2. Stalks
Definition 7.2.1. — Suppose X a space, and suppose ¨ a presheaf on X. Then for any point x ∈ X, consider the
set
¸
x∈U∈Op(X)
¨(U) =¦(U, s ) | x ∈ U ∈ Op(X), s ∈ ¨(U)].
On this set we may impose an equivalence relation ∼ in the following manner. For any two elements (U, s ) and
(V, t ), we say that (U, s ) ∼ (V, t ) if and only if there exists an open neighborhood W ⊂ U ∩V of x such that
ρ
W⊂U
(s ) =ρ
W⊂V
(t ). Now define the stalk of ¨ at x to be the set
¨
x
:=

¸
¸
¸
x∈U∈Op(X)
¨(U)

∼.
The equivalence class of a section s under this equivalence relation is called the germ of s , and is dented s
x
.
Example 7.2.2. — Suppose p : Y

X a covering space, i.e., a surjective continuous map with the property that
every point x ∈ X is contained in an open neighborhood U such that p
−1
U is a disjoint union
¸
α∈Λ
V
α
of open
subsets V
α
⊂ Y such that p|
V
α
: V
α

U is a homeomorphism. It is easy to see that such a map must be a local
homeomorphism, but here we have required much more: we have required that we can choose the neighborhoods
of various points in a fiber in a consistent manner.
For any point x ∈ X, we can compute the stalk Γ(Y/X)
x
easily: there exists an open neighborhood U of x whose
inverse image is homeomorphic (via p) to a disjoint union of copies of U. But now it’s easy to see that a section of
Γ(Y/X) over this open set U is completely determined by where it sends out point x. For every sub-neighborhood
V ⊂ U, we’ll be in the same situation, so the equivalence relation doesn’t change anything from this point on as the
neighborhoods get smaller. Hence the stalk Γ(Y/X)
x
is simply the fiber p
−1
(x) again.
We will undertake a more systematic study of coverings spaces very soon.
Exercise 53. — Suppose p : Y

X any continuous map. Is it true that for any point x ∈ X, the stalk Γ(Y/X)
x
is the fiber p
−1
(x)? Prove or give a counterexample.
Example 7.2.3. — Let us consider the case of two topological spaces X and Y, and let us consider the presheaf O
Y
X
of local continuous functions on X valued in Y. What is the stalk of this presheaf at a point x ∈ X? It’s difficult to say
in familiar language. The elements of O
Y
X,x
are equivalence classes of continuous functions defined in a neighborhood
of x, where we declare two such functions to be equivalent when there is a neighborhood of x on which they agree.
These equivalence classes are known as germs of continuous functions.
Example 7.2.4. — Suppose X =R
n
, and consider the presheaf O
an
X
of local analytic functions on X. For any point
x ∈ X, the stalk O
an
X,x
is the set of all power series at x with a positive radius of convergence. This is a ring, sometimes
denoted R¦x
1
, x
2
, . . . , x
n
].
7.3. Sheaves
Example 7.3.1. — To motivate the sheaf axiom, let’s briefly recall a little lemma we proved in Exercise 18 some
time back. If X is a space with an open covering · , and if for any V ∈ · we are given a continuous function
f
V
: V

Y with the property that for any elements V, W ∈ · , one has
f
V
|
V∩W
= f
W
|
V∩W
: V ∩W

Y ,
then there exists a unique continuous map f : X

Y such that for any V ∈ · ,
f |
V
= f
V
: V

Y .
7.3.2. — Suppose ¨ a presheaf on a space X. For any open set U, and any cover ¦U
α
]
α∈Λ
of U, we can contemplate
the following diagram:
(7.3.2.1)
¨(U)

¸
α∈Λ
¨(U
α
)

¸
β,γ∈Λ
¨(U
β
∩U
γ
).
46 CHAPTER 7. SHEAF THEORY
Let us describe these maps in more detail. The first map is the assignment
¨(U)

¸
α∈Λ
¨(U
α
)
s

ρ
U
α
⊂U
(s )

α∈Λ
.
The other two maps are the assignments
¸
α∈Λ
¨(U
α
)

¸
β,γ∈Λ
¨(U
β
∩U
γ
)
(s
α
)
α∈Λ

ρ
U
β
∩U
γ
⊂U
β
(s
β
)

β,γ∈Λ
and
¸
α∈Λ
¨(U
α
)

¸
β,γ∈Λ
¨(U
β
∩U
γ
)
(s
α
)
α∈Λ

ρ
U
β
∩U
γ
⊂U
γ
(s
γ
)

β,γ∈Λ
We can ask whether the diagram (7.3.2.1) is an equalizer, i.e., whether the map ¨(U)

¸
¨(U
α
) is
an injective map whose image consists of exactly those elements that have the same values under the maps
¸
¨(U
α
)

¸
¨(U
β
∩U
γ
)
. In other words, (7.3.2.1) is an equalizer if and only if the following condition is
satisfied: for every α ∈ Λ there is a section s
α
∈ ¨(U
α
) such that for any pair β, γ ∈ Λ, the restrictions
ρ
U
β
∩U
γ
⊂U
β
(s
β
) =ρ
U
β
∩U
γ
⊂U
γ
(s
γ
)
coincide, then there exists a unique section s ∈ ¨(U) such that ρ
U
α
⊂U
(s ) = s
α
.
Definition 7.3.3. — A presheaf ¨ on a space X is said to be a sheaf if, for any open set U ∈ Op(X) and any open
cover ¦U
α
]
α∈Λ
of U, the diagram
¨(U)

¸
α∈Λ
¨(U
α
)

¸
β,γ∈Λ
¨(U
β
∩U
γ
).
is an equalizer in the sense above.
7.3.4. — This is a little tricky, so let’s discuss this. A sheaf ¨ on a space X is a presheaf that has a powerful local-to-
global property.
To illustrate this local-to-global property, consider the case of an open cover with just two open sets. Suppose
that U ⊂X is an open set, which we have written as a union U =V ∪W for some open sets V, W ⊂X. Then the
sheaf condition for ¨ says that if we are given sections s
V
∈ ¨(V) and s
W
∈ ¨(W) that match up on V ∩W (so
that ρ
V∩W⊂V
(s
V
) = ρ
V∩W⊂W
(s
W
)), then there is one and only one set s ∈ ¨(U) who restrictions to V and to W
are s
V
and s
W
, respectively.
What does this say? It says that sections glue uniquely. In other words, specifying a section of ¨ over an open set
U ⊂X is the same as specifying a compatible family of sections of ¨ over each element of an open cover of ¨.
Example 7.3.5. — Our first example of this section shows that, indeed, for any two topological spaces X and Y,
the presheaf O
Y
X
is a sheaf, called the sheaf of local continuous functions on X with values in Y.
Example 7.3.6. — More generally, one can see that for any continuous map p : Y

X , the associated presheaf
Γ(Y/X) is indeed a sheaf, called the sheaf of local sections of p.
Example 7.3.7. — For an space X, any point x ∈ X, and any set S, the skyscraper presheaf S
x
is a sheaf as well.
Exercise 54. — Show that for any sheaf ¨, the map
¨(U)

¸
x∈U
¨
x
is injective. Give a counterexample to show that this is not true of presheaves.
Lemma 7.3.8. — Suppose ¨ a sheaf on a space X. Then ¨(∅) =.
Lemma 7.3.9. — Suppose ¨ and ¼ sheaves on a space X. Then if f , g : ¨

¼ are two morphisms such that for
every point x ∈ X, the induced maps f
x
, g
x
: ¨
x

¼
x
on stalks coincide (so that f
x
= g
x
), then f = g.
Proof. — This follows immediately from the previous exercise.
7.4. THE ESPACE ÉTALÉ AND SHEAFIFICATION 47
Proposition 7.3.10. — Suppose ¨ and ¼ sheaves on a space X. Then a morphism φ: ¨

¼ is an isomorphism, or
an injection, if and only if, for every point x ∈ X, the map on stalks ¨
x

¼
x
is so.
Proof. — The injectivity statement follows from the fact that if the composite of two maps is injective, then the
first of the two must also be injective.
Now suppose U ∈ Op(X) any open set, and consider the component φ
U
: ¨(U)

¼(U) . The claim is that
it is surjective. To demonstrate this, suppose s ∈ ¼(U) a section. For every point x ∈ U, the element (U, s ) ∈ ¼
x
has an inverse image in ¨
x
, i.e., for each point x ∈ U there is a pair (V
x
, t
x
), consisting of an open neighborhood
V
x
⊂ U of x along with a section t
x
∈ ¨(V
x
) such that φ
V
x
(t
x
) =ρ
V
x
⊂U
(s ). Since φ is injective, for any two points
x, y ∈ U, we have
ρ
V
x
∩V
y
⊂V
x
(t
x
) =ρ
V
x
∩V
y
⊂V
y
(t
y
).
The sheaf axiom applied to the open cover ¦V
x
| x ∈ U] of U now implies that there is a global section t ∈ ¨(U)
such that t
x
= ρ
V
x
⊂U
(t ). And now φ(t ) ∈ ¼(U) is the unique section whose restriction to any V
x
is φ
V
x
(t
x
) =
ρ
V
x
⊂U
(s ), namely, s itself!
Warning 7.3.11. — Note that what we have not said is that two sheaves ¨ and ¼, all of whose stalks are iso-
morphic, are themselves isomorphic. This is false. We must have a morphism ¨

¼ inducing the stalkwise
isomorphism in order to make this conclusion. This sounds like a minor point, but it is extremely important.
Indeed, recall our two examples of double covers of the circle: we were given two local homeomorphisms ∇ :
S
1
|S
1
S
1
and p : S
1
S
1
. The former map is simply the folding map, and the second is z


z
2
. We saw
that, while Γ(∇) was simply the constant sheaf on a two point set, Γ( p) was a more subtle object, with no global
sections. In this case, the stalk over any point x ∈ S
1
is always the same: simply two points. But the two sheaves
differ dramatically! The point is that there is no morphism of sheaves between Γ(∇) and Γ( p) inducing bijections
on the stalks. This situation is quite typical, and it’s one that sheaves are particularly well-adapted to.
Exercise 55. — Suppose n a positive integer. Consider the continuous map p
n
: S
1
S
1
given by z


z
n
(where again we think of S
1
as embedded in C, where we may use complex multiplication). Consider also the
n-fold folding map ∇
n
: S
1
|S
1
|· · · |S
1
S
1
. Prove that Γ( p
n
) and Γ(∇
n
) are both sheaves, that their stalks at
any point are isomorphic, but the sheaves themselves are not isomorphic.
7.4. The espace étalé and sheafification
Definition 7.4.1. — A continuous map p : Y

X is said to be a local homeomorphism if every point y ∈ Y is
contained in a neighborhood V such that p is open and injective.
7.4.2. — Suppose X a space and ¨ a presheaf on X. Consider the set
´
Et(¨) :=
¸
x∈X
¨
x
; there is an obvious map
p
¨
:
´
Et(¨)

X whose fibers are precisely the stalks of ¨. For any open set U and any section s ∈ ¨(U), there
is a corresponding map σ
s
: U

´
Et(¨)
, x


s
x
, such that p ◦ s =id. (This is the origin of the term “section.”)
Definition 7.4.3. — Suppose X a space and ¨ a presheaf on X. The espace étalé of ¨ is the set
´
Et(¨) equipped with
the finest topology such that for any section s ∈ ¨(U), the corresponding map σ
s
: U

´
Et(¨)
is continuous.
That is, we declare a subset V ⊂
´
Et(¨) to be open if and only if, for any open set U ∈ Op(X) and any section
s ∈ ¨(U), the inverse image σ
−1
s
(V) is open in U.
Exercise 56. — Show that for any space X and any presheaf ¨ on X, the natural morphism p
¨
:
´
Et(¨)

X is
a local homeomorphism.
Exercise 57. — Suppose S a set, and suppose ¨
S
is the constant presheaf at S on a space X. Then show that the
éspace étalé of ¨
S
is the fold map ∇
(S)
: X
|(S)
X , where X
|(S)
is the S-fold disjoint union of X with itself:
X
|(S)
:=
¸
s ∈S
X =X ×S
δ
.
48 CHAPTER 7. SHEAF THEORY
Lemma 7.4.4. — Suppose p : Y

X a local homeomorphism. Then the éspace étalé of the sheaf Γ(Y/X) is canoni-
cally homeomorphic over X to Y. That is, there is a unique homeomorphism Y

´
Et(Γ(Y/X))
such that the diagram
´
Et(Γ(Y/X))
p
Γ(Y/X)

?
?
?
?
?
?
?
?
w

Y
p
·








X
commutes.
Proof. — Suppose s ∈
´
Et(Γ(Y/X)) a germ of a section near x ∈ X. Then its value w(s ) = s (x) at x is well-defined.
This defines a map w :
´
Et(Γ(Y/X))

Y such that p ◦ w = p
Γ(Y/X)
.
In the other direction, suppose y ∈ Y, and suppose x = p(y). Then there exist open neighborhoods U of x
and V of y such that p restricts to a homeomorphism V

U ; denote by s : U

V its inverse, and define
v(y) ∈ Γ(Y/X)
x
as the germ of the section s . The class v(y) does not depend upon the choice of U or V. This
defines an inverse v : Y

´
Et(Γ(Y/X))
to the map w.
It remains to show that w is continuous and open. We leave this to the reader.
Definition 7.4.5. — Suppose X a space and ¨ a presheaf on X. The sheafification of ¨ is the sheaf a¨ :=
Γ(
´
Et(¨)/X) of sections of the projection map p :
´
Et(¨)

X . The canonical morphismof presheaves ¨


that assigns to any section s ∈ ¨(U) the local section x


s
x
is called the unit morphism.
Proposition 7.4.6. — For any presheaf ¨ on a space X, the natural morphism ¨

a¨ induces an isomorphism on
all stalks.
Proof. — By 7.4.4, the stalks of a¨ correspond under the unit morphism to the fibers of
´
Et(¨)

X , which in
turn are the stalks of ¨.
Corollary 7.4.7. — For any sheaf ¨ on a space X, the unit morphism ¨

a¨ is an isomorphism.
Corollary 7.4.8. — There is a bijective correspondence between isomorphism classes of sheaves on X and homeomor-
phism classes of local homeomorphisms Y

X .
Exercise 58. — Recall that on a space X, the presheaf O
bdd
X
of local bounded continuous functions (valued in R)
may not be a sheaf. Describe its sheafification.
Example 7.4.9. — The constant sheaf ¨
S
at a set S on a space X is the sheafification of the constant presheaf ´
S
at
S. Observe that the constant sheaf is not really constant: it can take many different values on an open set U ⊂X.
Theorem 7.4.10. — Suppose X a space, suppose ¨ a presheaf on X, and suppose ¼ a sheaf on X. Then any morphism
¨

¼ factors uniquely through the unit. In other words, for any morphism ¨

¼ , there exists a unique morphism


¼ such that the diagram
¨

?
?
?
?
?
?
?
?
?

¼










commutes.
Proof. — Let us show that unicity is guaranteed once the existence of the morphism a¨

¼ satisfying the
diagram above is confirmed. Indeed, any two morphisms a¨

¼ satisfying the diagram above must agree an all
stalks, since the unit morphism ¨

a¨ is a bijection on all stalks. Now we may employ 7.3.9 to see that the
morphisms a¨

¼ coincide.
To verify existence, observe that we can simply apply a to the morphism ¨

¼ to get a morphism of sheaves


a¼ ; since the unit morphism ¼

a¼ is an isomorphism, we can compose a¨

a¼ with the inverse


¼ to get the desired morphism a¨

¼ .
7.5. DIRECT AND INVERSE IMAGES OF SHEAVES 49
7.4.11. — There is a more sophisticated way of talking about all this. The construction of the espace étalé is a functor
´
Et from the category ´re·h(X) of presheaves on a space X to the category ('·/X) of local homeomorphisms
to X. Now ('·/X) is actually equivalent (via the sections sheaf functor Γ) to the category ·h(X) of sheaves
on X. Now the composite a = Γ ◦
´
Et is the sheafification functor, which is left adjoint to the inclusion ·h(X) ⊂
´re·h(X).
7.5. Direct and inverse images of sheaves
Definition 7.5.1. — For any two continuous maps f : X

Y and h : Z

Y , the fiber product X ×
Y
Z is the
set
X ×
Y
Z :=¦(x, z) ∈ X ×Z | f (x) = h(z)] ⊂X ×Z,
equipped with the subspace topology. (Note that the notation here is ambiguous, since it makes no explicit mention
of the maps f and h.) The (continuous) projection X ×
Y
Z

X is called the pullback of the map h.
Exercise 59. — For any map f : X

Y , contemplate the continuous map ∆
f
: X

X ×
Y
X given by the
assignment x


(x, x) . Show that f is a local homeomorphism if and only if both f and ∆
f
are open maps.
Exercise 60. — Show that the pullback of a local homeomorphism is a local homeomorphism.
Definition 7.5.2. — Suppose f : X

Y a continuous map.
(7.5.2.1) For any sheaf ¨ on X, define the direct image f

¨ of ¨ as the sheaf that assigns to any open set V ∈ Op(Y)
the set
f

¨(V) :=¨( f
−1
V).
(7.5.2.2) For any sheaf ¼ on Y, we define the inverse image f

¼ as the sheaf of sections of the pullback
X ×
Y
´
Et(¼)

X of the map p
¼
:
´
Et(¼)

Y
Example 7.5.3. — Suppose A⊂X a subspace of a topological space X. Then for any sheaf ¨ on X, if i denotes the
inclusion map, the sheaf i

¨ on A is denoted ¨|
A
and is called the restriction of ¨ to A. If, in particular, A is an
open set, then the restriction ¨|
A
assigns to any open set U ⊂Athe set ¨(U). On the other hand, if A=¦x], then
¨|
¦x]
is simply the stalk of ¨ at x, regarded as a sheaf on ¦x].
Exercise 61. — Any sheaf on the one-point space is uniquely determined by a set S. Show that the direct image of
such a sheaf along the inclusion

X of a point x of a space X is the skyscraper sheaf S
x
. Show that the inverse
image of such a sheaf under the unique map X

is the constant sheaf a¨
S
.
ACT II
AN INTRODUCTION TO HOMOTOPY THEORY
CHAPTER 8
COVERING SPACE THEORY
8.1. Connectedness and π
0
Warning 8.1.1. — The definitions we give here do not coincide with definitions in many other texts. The defini-
tions agree in “good cases,” i.e., in all the cases with which we will have to deal in the sequel; however, there are
important ways in which our definitions differ from those found elsewhere in the literature.
8.1.2. — For any set S, consider the constant presheaf ´
S
on a space X; its associated sheaf is the constant sheaf ¨
S
.
The unit morphism ´
S

¨
S
induces a canonical map S

¨
S
(X) on global sections. If S is the one-point set,
then this map is an isomorphism. If S is empty, then this map is a bijection unless X is empty.
An alternate description of the sheaf a¨
S
is the following. It is the sheaf O
S
δ
X
of local continuous functions valued
in the discrete space S
δ
.
Definition 8.1.3. — A space X is said to be connected if for any set S, the obvious map S

¨
S
(X) (which assigns
to any element s ∈ S the global section x


(x, s ) ) is a bijection.
8.1.4. — Thus a space X is connected if and only if it is nonempty and every global section of the constant sheaf at
S is the inclusion X צs ]
δ
⊂X ×S
δ
for some s ∈ S. Equivalently, X is connected if and only if the only continuous
maps X

S
δ
are constant.
Definition 8.1.5. — Suppose X a space. Then π
´et
0
X is a set along with a global section u ∈ ¨
π
´et
0
X
(X) such that for
any set S and any global section σ ∈ ¨
S
(X), there exists a unique morphism of sheaves ¯ σ :

π
´et
0
X


S
such
that ¯ σ(u) =σ.
8.1.6. — Note that π
´et
0
X is a set with a continuous map X


´et
0
X)
δ
such that for any set S and any continuous
map X

S
δ
, there exists a unique map π
´et
0
X

S such that the diagram
X

?
?
?
?
?
?
?
?

S
δ
.

´et
0
X)
δ








8.1.7. — Note that a space X is connected if and only if π
´et
0
X =.
Warning 8.1.8. — Note that this definition of π
0
is a little strange. In particular, the definition is nonconstructive
in the sense that it is not clear from the definition alone that π
´et
0
X exists for a given space X. In fact, it may very
well not exist: in the case of Q with the subspace topology, for instance, there is no π
´et
0
.
However, it’s much easier to compute this version of π
0
than it is to compute other variants, at least once you get
used to checking universal properties. This is primarily because it involves finding maps out of your space.
54 CHAPTER 8. COVERING SPACE THEORY
Exercise 62. — Recall that for any collection of spaces ¦V
α
]
α∈Λ
, the topological disjoint union
¸
α∈Λ
V
α
is the set-
theoretic disjoint union
¸
α∈Λ
V
α
equipped with the following topology: an open set U ⊂
¸
α∈Λ
V
α
is open if and
only if each intersection U ∩V
α
is open in V
α
.
Now for any space V, a decomposition of V is a collection of subspaces V
α
⊂ V — called summands — such
that V can be written as
¸
α∈Λ
V
α
. Show that there is a bijective correspondence between decompositions V =
¸
α∈Λ
V
α
and continuous maps V

Λ
δ
. Let us call a decomposition V =
¸
α∈Λ
V
α
nondegenerate if and only if
the corresponding map V

Λ
δ
is surjective, or, equivalently, if no summand V
α
is empty.
Show that a space X is connected if and only if it is nonempty and admits no nondegenerate decomposition.
Show that if X is a space such that π
´et
0
X exists, then the continuous map X

π
´et
0
X is surjective. Conclude that
π
´et
0
X exists if and only if there exists a maximal nondegenerate decomposition of X, i.e., a decomposition of X such
that each summand is connected.
8.1.9. — Let us consider Q with the subspace topology. The only connected subspaces of Q are points. (Proof: if
A⊂Q is a connected subspace containing two distinct points a and b, choose an irrational number a < c < b. Now
consider the map g : A

¦0, 1]
δ
given by the rule
g(x) :=

0 if x < c;
1 else.
This is continuous, since g
−1
¦0] =(−∞, c) ∩Aand g
−1
¦1] =(c, ∞) ∩A.) But it is not the case that Q is the disjoint
union of its points, for then points would be open in Q. Hence there is no maximal nondegenerate decomposition
of Q, so π
´et
0
Q does not exist.
Exercise 63. — Compute π
´et
0
of the following spaces. (In all these cases, π
´et
0
exists, so there are no tricks here.)
(63.1) a set S with the discrete topology,
(63.2) the sphere S
n
for any n >0,
(63.3) the set GL
n
R of n ×n invertible matrices with real entries, with the subspace topology from the inclusion
GL
n
R⊂R
n
2
, for n >0,
(63.4) the set SL
n
R of n × n matrices with real entries of determinant 1, with the subspace topology from the
inclusion SL
n
R⊂R
n
2
, for n >0, and
(63.5) the set GL
n
C of n ×n invertible matrices with complex entries, with the subspace topology from the inclu-
sion GL
n
C⊂C
n
2
, for n >0.
Exercise 64. — Consider the letters of the (English) alphabet A–Z as subspaces of the plane R
2
. (Consider them in
their simplest possible form, without any thickness of any lines, and without any serifs!) Can you classify all of them
up to homeomorphism? (Hint: if two spaces X and Y are homeomorphic, then their π
´et
0
agree, and if f : X

Y
is a homeomorphism, then so is f : X −¦x]

Y −¦ f (x)] for any x ∈ X.)
8.1.10. — The construction X


π
´et
0
X is functorial in the sense that for any continuous map f : X

Y , there
is a corresponding map π
´et
0
f : π
´et
0
X

π
´et
0
Y if both π
´et
0
X and π
´et
0
Y exist. Indeed, the definition implies that there
exists a unique map π
´et
0
f : π
´et
0
X

π
´et
0
Y such that the diagram
X

f

Y


´et
0
X)
δ
π
´et
0
f


´et
0
Y)
δ
commutes.
8.2. COVERING SPACES AND LOCALLY CONSTANT SHEAVES 55
8.2. Covering spaces and locally constant sheaves
Definition 8.2.1. — A covering space over a topological space B (called the base) is a continuous map p : E

B
such that every point b ∈ B is contained in an open neighborhood U for which there is a homeomorphism
γ
U
: p
−1
U

U ×( p
−1
b)
δ
such that the diagam
p
−1
U
p|
p
−1
U

?
?
?
?
?
?
?
?
γ
U

U ×( p
−1
b)
δ
pr
1
·







U
commutes.
Exercise 65. — Show that a covering space is a local homeomorphism.
Exercise 66. — Check that the pullback E ×
B
B
/

B
/
of a covering space E

B along any continuous map
B
/

B is a covering space.
Definition 8.2.2. — A morphism of covering spaces from p
/
: E
/

B to p : E

B (covering spaces over the same
base) is simply a continuous map f : E
/

E such that the diagram
E
/
p
/

?
?
?
?
?
?
?
?
?
f

E
p
·








B
commutes.
Example 8.2.3. — (8.2.3.1) For any space X and any set S, the map ∇
(S)
: X
|(S) ∼
=X ×S
δ
X is a covering
space. These covering spaces are known as the trivial covering spaces. The corresponding locally constant
sheaves are constant.
(8.2.3.2) Consider again the continuous map p
n
: S
1
S
1
given by z


z
n
. This is a covering space too. The
corresponding sheaf assigns to any open set U ⊂ S
1
the set of n-th root functions defined on U.
(8.2.3.3) The exponential map exp : C

C
×
:=C−¦0] is a covering map. One can pull this example back to
S
1
⊂ C
×
to get the covering space R

S
1
given by θ


e
2πθ

−1
. The corresponding sheaf assigns to
any open set U a logarithm defined on U.
Exercise 67. — Consider the subspace (C−¦0])
×n
⊂ C
n
. Show that for any n-tuple (k
1
, . . . , k
n
) of nonnegative
integers, the map
p
(k
1
,...,k
n
)
: (C−¦0])
×n

(C−¦0])
×n
(z
1
, . . . , z
n
)


(z
k
1
1
, . . . , z
k
n
n
)
is a covering space. Pull this back to get a covering space on the n-dimensional torus T
n
. Now contemplate the fiber
over (1, . . . , 1). Can you draw a picture of this when n =2?
Definition 8.2.4. — A sheaf ¨ on a space X is locally constant if every point of X is contained in an open neigh-
borhood U such that the restriction ¨|
U
is a constant sheaf.
Theorem 8.2.5. — A local homeomorphism p : E

B is a covering map if and only if the sheaf Γ(E/B) is locally
constant; correspondingly, a sheaf ¨ is locally constant if and only if the espace étalé
´
Et(¨) is a covering space.
56 CHAPTER 8. COVERING SPACE THEORY
8.2.6. — Under our bijective correspondence between homeomorphism classes of local homeomorphisms and iso-
morphism classes of sheaves, we see that locally constant sheaves correspond (bijectively) to covering spaces. As we
have seen, constant sheaves correspond to trivial covering spaces.
To put this more formally, for any space B, let ('(/B) be the category of locally constant sheaves on B, and let
(ov(B) denote the category of covering spaces with base B. Then our equivalence of categories ('·/B) ~·h(B)
restricts to an equivalence of categories ('(/B) ~(ov(B).
Exercise

68. — Show that the only possible locally constant sheaves on the interval [0, 1] are constant. (Hint:
Suppose ¨ a locally constant sheaf on [0, 1]. Show that there is a finite open cover of [0, 1] comprised of intervals
U on which ¨|
U
. Now induct on the size of ¹ in order to show that ¨ must be constant. Why doesn’t this
argument work for S
1
?)
Using this, suppose X any space, and suppose γ a path in X — i.e., a continuous map γ : [0, 1]

X . For any
locally constant sheaf ¼ on X, the pullback γ

¼ is a constant sheaf. More precisely, for each t ∈ [0, 1], show that
there is a unique isomorphism between γ

¼ and the constant sheaf ¨
S(t )
at the stalk S(t ) := ¼
γ(t )
. Now if γ is a
loop in X — i.e., a path such that γ(0) =γ(1) = x ∈ X —, we have specified two isomorphisms:
¨
x

S(0)

γ

¼ ¨
S(1)

x

,
so we have an automorphism of the stalk ¨
x
. Give an example to show that this autmorphism need not be the
identity.
Definition 8.2.7. — We say that a covering space p : E

X is finite if the fibers of p are finite. If π
´et
0
X exists,
then we say that it is of finite type if π
´et
0
E exists, and the fibers of π
´et
0
p : π
´et
0
E

π
´et
0
X are finite. Let us denote by
((ov
ft
/X) the set of all isomorphism classes of covering spaces of finite type over X.
8.3. The étale fundamental group
8.3.1. — We are now interested in classifying all locally constant sheaves/covering spaces over a connected space
X. In order to do this, we will begin by constructing, for any point x ∈ X, the étale fundamental group π
´et
1
(X, x)
out of the collection of all covering spaces of finite type over X.
Definition 8.3.2. — Suppose X a space, and suppose x ∈ X. For any covering space E

X , a monodromy element
for E at x is an automorphism of the fiber E
x
.
A uniform monodromy element at x is an element
γ =(γ
E
) ∈
¸
E∈((ov
ft
/X)
Aut(E
x
)
such that for any morphism φ: E
/

E of covering spaces of finite type,
E
/
x
γ
E
/

φ
x

E
x
γ
E

E
/
x
φ
x

E
x
.
For any two uniform monodromy elements γ and γ
/
, one may compose them to obtain
γ
/
◦ γ :=(γ
/
E
◦ γ
E
) ∈
¸
E∈((ov
ft
/X)
Aut(¨
x
)
The étale fundamental group π
´et
1
(X, x) is the set of uniform monodromy elements.
Lemma 8.3.3. — The set π
´et
1
(X, x) is a group under composition.
8.5. UNIVERSAL COVERING SPACES 57
Proof. — Since π
´et
1
(X, x) is given as a subset of a group, we only have to check that π
´et
1
(X, x) contains the identity
(obvious) and that it is closed under composition. This latter point follows from inspection of the diagram
E
/
x
γ
E
/

φ
x

E
x
γ
E

E
/
x
γ
/
E
/

φ
x

E
x
γ
/
E

E
/
x
φ
x

E
x
.
8.4. Homotopy classes of loops and lifts
Lemma 8.4.1. — Consider the following commutative diagram of spaces

E
p

Y
f

B
in which p : E

B is a covering space and Y is connected space. Then a lift of f , if it exists, is unique.
Proof. — Suppose g and h two lifts. Define a map r : Y

¦0, 1]
δ
by
r (y) =

1 if g(y) = h(y)
0 if not.
I claim r is continuous. Indeed, suppose y ∈ Y; let U be an open neighborhood of f (y) such that p
−1
U

= U×E
f (y)
.
Let e
g
, e
h
∈ E
f (y)
be the projections of the element g(y), h(y) to E
f (y)
; then the set
V := g
−1
(U צe
g
]) ∩ h
−1
(U צe
h
])
is open in Y, and one verifies that V ⊂ r
−1
( f (y)). This shows that r is continuous, and the result follows from the
connectedness of Y.
Lemma 8.4.2. — Unique path lifting ...
Lemma 8.4.3. — In the situation above, assume that Y is locally path connected. Then a lift exists if and only if the
image of π
1
(Y) in π
1
(X) is contained in the image of π
1
(E).
Proof. — Necessity is obvious. For sufficiency, define a lift φ : Y

E in the following manner. For any point
y ∈ Y, choose a path γ from the basepoint to y. Then there is a unique lift γ
/
of this path to E; define φ(y) =γ
/
(1).
The assumption guarantees that this is independent of the choice of path.
8.5. Universal covering spaces
Definition 8.5.1. — Suppose X a space, and suppose x ∈ X. Then a universal locally constant sheaf on (X, x) is a
locally constant sheaf ¿ on X along with a germ of a section υ ∈ ¨
x
such that for any locally constant sheaf ¨ and
any germ σ ∈ ¨
x
, there is a unique morphism φ: ¿

¨ of sheaves such that φ
x
(υ) =σ.
Exercise 69. — Suppose X a space, and suppose x ∈ X. Show that the universal locally constant sheaf (¿, υ) on
(X, x), if it exists, has the following property: for any locally constant sheaf ¨ on X, the map Mor
X
(¿, ¨)

¨
x
,
φ

φ
x
(υ) , is a bijection. Deduce that if a universal locally constant sheaf on (X, x) exists, then it is unique up to
a unique isomorphism.
58 CHAPTER 8. COVERING SPACE THEORY
8.5.2. — There is a natural candidate for the universal locally constant sheaf on a pointed space (X, x).
Definition 8.5.3. — Suppose X a space, and suppose x ∈ X. Then a universal covering space over (X, x) is a covering
space u :
¯
X

X along with a point ¯x ∈ u
−1
(x) such that for any covering space p : E

X and any point
e ∈ p
−1
(x), there exists a unique covering space map f :
¯
X

E with the property that f (¯x) = e.
Definition 8.5.4. — A space X is said to be well connected if it is path connected, it admits a basis comprised of
path-connected neighborhoods, and any point x is contained in a neighborhood U such that the homomorphism
π
1
(U, x)

π
1
(X, x) is trivial.
Theorem 8.5.5. — Suppose X a well connected space. Then X admits a universal covering space. Moreover, it is the
unique connected pointed covering space (
¯
X, ¯x) of X such that π
1
(
¯
X, ¯x) is trivial.
Proof. — Let us construct the corresponding locally constant sheaf. Begin by contemplating the based path space
Π
x
(X) ⊂ ((I , X), whose points are those paths γ such that γ(0) = x, equipped with the subspace topology. Eval-
uation at 1 gives a continuous map e : Π
x
(X)

X Consider now the sheaf of sections Γ(e); for any open set U
of X, define ¼(U) as the quotient Γ(e)(U)/ ∼, where we say that two sections s , t ∈ Γ(e)(U) are equivalent if they
are homotopic as maps U ×I

X rel U צ0, 1]. The result is a presheaf ¼; the claim is that its sheafification is
locally constant. To this end, observe that since X is well connected, it admits a base consisting of path connected
open sets U such that the induced homomorphism π
1
(U)

π
1
(X) is trivial. Over these open sets, the espace
étalé of ¼ is a trivial covering space.
So now we have the corresponding covering space
¯
X. It is the space of homotopy (rel ¦0, 1]) classes of paths
starting at x; evaluation at 1 gives the structure map
¯
X

X . Write ¯x for the constant map at x. If now we can
check that
¯
X is path connected and that π
1
(
¯
X, ¯x) is trivial, then we are done by 8.4.3.
To show that
¯
X is path connected, consider an equivalence class α of paths. Let f be a representative thereof.
Now for any real number t , let f
t
(s ) = f (s t ). This gives a path from ¯x to α.
To show that π
1
(
¯
X, ¯x) is trivial, ...
Theorem 8.5.6. — Suppose X admits a universal covering space. Then there is a bijective correspondence between
((ov
ft
/X) and the set of isomorphism classes of π
´et
1
(X, x)-sets with finitely many orbits, and under this correspondence,
the connected covering spaces correspond to the transitive π
´et
1
(X, x)-sets.
CHAPTER 9
COMPUTING THE ÉTALE FUNDAMENTAL GROUP
9.1. The étale fundamental group of S
1
9.2. Seifert-van Kampen Theorem
9.3. Simple connectedness
CHAPTER 10
HOMOTOPY THEORY
10.1. Paths and π
0
10.2. Homotopies of maps
10.3. The homotopy-theoretic fundamental group
10.4. Homotopy invariance of the étale fundamental group
CHAPTER 11
THE COMPARISON THEOREM
11.1. The two fundamental groups

Clark Barwick Department of Mathematics, Harvard University, One Oxford Street, Cambridge, MA 02138, USA. E-mail : clarkbar@gmail.com

LECTURE NOTES FOR TOPOLOGY I Clark Barwick

.

. . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sheaves. .CONTENTS Act I. . . . . . . . . . . . 11 1. . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Compactness II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . 35 35 36 37 Entr’Acte. . . . . . . . . . . . . . . . . . . . 17 2. . . . . 9 1. . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Connectedness and π0 . . . . . . . . . . . . . . . .1. . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 4. . . . . . Normality. 53 . . . . . . . . . . 31 5. . . . . . . . . . . . . . . . . . . 7 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Continuous maps and homeomorphisms. . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . Stalks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Forms of compactness. . . . . . . . . . . . . . . First examples. . . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sheaf theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 41 45 45 47 49 Act II. . . . 31 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. . . . . The category of topological spaces. . . . . . . . . . . . . . . . . . . . . . . . . . Taxonomy of separation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Topological spaces. . . . . . . . . . . . . . . . . . . . . . . 17 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . An introduction to homotopy theory. . . . . . . 6. . . . . . 13 2. . . . . . . . . . Countability and separation. . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . The espace étalé and sheafification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Presheaves. . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Urysohn’s metrization theorem. .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . Locally compact Hausdorff spaces. . . . . . . . . . . . . . . . . . . Constructing topologies and continuous maps. . . . . . . . . . . . 21 21 22 23 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 4. . . . . . . . . . . . . . . . . . . . . . . .1. . . .1. . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quotients. . . . . . . . . . . . Urysohn’s lemma. . . . . . 53 8. . . . . 27 5. . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Compact spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . and the Tietze extension theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The compact-open topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . The Tychonoff Product Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coverings and locality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 8. . . . . . . . . . . . . . . . . 33 6. . . . 3. . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . Compactness I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 7. . . . . . . . . . . . . . . . . . . . . 7. . . . . The Bing–Nagata–Smirnov metrization theorem. . Metrization theorems. . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . Direct and inverse images of sheaves. . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Covering space theory. . . . . . . . . . . . . . . First and second countability. . . . . . . . . . . . . . . . 9 1. . . . . . . . . . . . . . . . . . . .

. . . . . . Homotopy invariance of the étale fundamental group. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . .6 CONTENTS 8. . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homotopies of maps. . . 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . Seifert-van Kampen Theorem. . . . . . . . . . . . . 63 11. . . . . . . . . . . . Homotopy theory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Covering spaces and locally constant sheaves. . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The étale fundamental group. . . . . . . . The étale fundamental group of S 1 . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . The two fundamental groups. . . . . . . . . . . .2. . . . . 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The comparison theorem. . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Computing the étale fundamental group. . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homotopy classes of loops and lifts. . 8. . Universal covering spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simple connectedness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Paths and π0 . . . . 10. . The homotopy-theoretic fundamental group. . . . . . . . . . . . . . 9. . . . . . . . . . .3. 55 56 57 57 59 59 59 59 61 61 61 61 61 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . . . . . . .

ACT I GENERAL TOPOLOGY .

.

— We have described a topology on a set X by prescribing the set p(X ) of open sets.4..1. f −1 respects complements: for any set A ∈ f −1 (S). A topology on a set X is.2.1.e. the union U ∈ p(X ). Any map / S induces a map f :T / (T ). f −1 : (S) where for any subset A ⊂ S.1. 1.1.1) For any subset ⊂ l(X ). equivalently.1. A = f −1 A and f −1  A = f −1  A∈ A∈ A∈ A∈ ⊂ (S). — A topological space (or more simply a space) is a pair (X . i. (1.1. Topological spaces Notation 1. f −1 A := {t ∈ T | f (t ) ∈ A} ⊂ T . U ∈ In this case.2) translates to the condition that ∅ ∈ p(X ).1. A subset V ∈ (X ) will be called closed if its complement V := X − V is an open set.4.1. a subset l(X ) ⊂ (X ) satisfying the following axioms. . we denote by (S) the power set of S.3.2. (1.1. the set of all subsets of S. p(X )) comprised of a set X and a subset p(X ) ⊂ (X ) satisfying the following properties.2.2.2) For any subset ⊂ p(X ). when = ∅.1) For any finite subset ⊂ p(X ). we may also describe a topology on a set X by prescribing the set l(X ) of closed sets. The elements U ∈ p(X ) will be called open in X .1. condition (1.1. One may verify that f −1 respects arbitrary intersections and arbitrary unions: for any subset         f −1 A. 1. the intersection V∈ V∈ l(X ). condition (1. the set p(X ) is said to be a topology on X . — Given a set S. Definition 1. Similarly. — In the definition above.2.CHAPTER 1 THE CATEGORY OF TOPOLOGICAL SPACES 1. U∈ (1. one may verify that T (f −1 ( S A) = A).1) translates to the condition that X ∈ p(X ). note that when = ∅. moreover. the intersection U ∈ p(X ).

.1) The closure A is the set A := {x ∈ X | for any open neighborhood V of x. A neighborhood of x is a subset V ⊂ X containing an open neighborhood of x.6) The interior A◦ is the union of all the open subsets contained in A: A◦ = U ∈ p(X ).2) The closure A is the smallest closed subset containing A. (1. Prove that the closure A ⊂ X can be characterized in the following two ways. (1. prove that the interior A ⊂ X can be characterized in the following two ways.6. In other words. V ∩ A = ∅}.3) The closure A is the intersection of all the closed subsets containing A: A= V ∈ l(X ).5. — Suppose (X . (1.5) The interior A◦ is the largest open subset contained in A.1.6.1.A⊂V ◦ V. Show that the closure and the interior operations are dual operations in the sense that. We say that A is dense in B if A ∩ B = B. In other words.1. the union V ∈ V ∈ l(X ). an open neighborhood of x is an open set U ∈ p(X ) containing x. (2.3) Consider two subsets A ⊂ B ⊂ X . we sometimes refer to the pair (X . it will always be clear whether we are describing the set of open subsets of X or the set of closed subsets of X .4.1.4) Dual to the closure is the interior. l(X )) as a topological space.2) For any subset A ⊂ X . the points of A are those points x ∈ X with the property that every neighborhood of x meets A. As a result. (1. one has U ⊂ κ(U ). Definition 1. (1. — Suppose (X .1) For any subset U ⊂ X .1. p (X ) ⊂ p(X ). (1. (1. one has κ(κ(U )) = κ(U ). Suppose that A ⊂ X is a subset.10 CHAPTER 1. Exercise 2.1. (1.6. (( A)◦ ) = A and ( A) = A◦ . — Suppose X a set. (2.2) For any finite subset ⊂ l(X ). (1.6. Then there are many different kinds of subsets of X . For any subset A ⊂ X .2) For any subset U ⊂ X . A Kuratowski closure operator on X is a map / (X ) κ : (X ) satisfying the following properties. In practice. — Suppose X a set. Definition 1. V ∩ A = ∅}. If then we say that p(X ) is finer than p (X ) and that p (X ) is coarser than p(X ). Dually. p(X )) a topological space.1) For any point x ∈ X .6.U ⊂A U. the points of A◦ are those points x ∈ A with the property that there is a neighborhood of x contained in A. the interior of A is the set A◦ := {x ∈ A | there exists a neighborhood V of x such that V ⊂ A}. the closure of A is the set A := {x ∈ X | for any neighborhood V of x. p(X )) a topological space. THE CATEGORY OF TOPOLOGICAL SPACES (1. for any subset A ⊂ X . and suppose p(X ) and p (X ) two topologies on X . Exercise 1. (1.4) The interior A◦ is the set A◦ := {x ∈ A | there exists an open neighborhood V of x such that V ⊂ A}.1.

1. ) := {y ∈ Rn | ||y − x|| < }. Example 1.2. in which a subset U ⊂ S is declared to be open if and only if either S = ∅. which is — as we will explain — homeomorphic to Σ0 . The topologies {0. show that the map / {topologies on X } {Kuratowski closure operators on X } / p(X κ ) κ / A described in 1. in which a subset U ⊂ S is declared to be open if and only if either S = ∅.1. {0. 1}δ specify all the topologies on the set {0. For any point x ∈ Rn and any real number > 0. FIRST EXAMPLES 11 (2. the (open) ball of radius r centered at x is the subset B(x. 1. S}).2.4.2. One may verify directly that the Sierpi´ ski space is in fact a topological space. 1}} ⊂ ({0. Σ0 . a topology on a set X can be is a bijection. 1} can also be given a n topology given by the subset p(Σ1 ) := {∅. (S)). the closed subsets of S are the finite sets and S itself. In this case. In this case. Similarly.2. one has   κ U∈   U = U∈ κ(U ).1) The indiscrete topology S ι is the pair (S. 2}. 1}} ⊂ ({0. 1}) . there is a real number > 0 such that B(x.2.5. Example 1. The set {0. or else the complement S U is finite. — Describe explicitly all the topologies on the set {0. Example 1.2. p(X )) a topological space. 1}ι .2. ) ⊂ U . the closed subsets of S are the countable sets and S itself. First examples Example 1. p(X κ )) is a topological space. and suppose A ⊂ X a subset of X . equipped with the subspace topology.1. define subsets l(X κ ) := {V ∈ (X ) | κ(V ) = V } ⊂ (X ) and thus p(X κ ) := {U ∈ (X ) | U ∈ l(X κ )} ⊂ (X ) Show that the pair X κ := (X . {1}. defined by p(A) := {U ∈ (U ) | there exists U ∈ p(X ) such that U = U ∩ A}. there are two topologies that exist automatically on S. {0}. — Suppose (X . 1}.1. 1.2. or else the complement S U is countable. where the inverse is the operation A  specified simply by describing a Kuratowski closure operator. 1}) . This defines a space Σ1 . 1} equipped with the subset: n p(Σ0 ) := {∅.2) The discrete topology S δ is the pair (S. (1. Now the set Rn can be given a topology in the following manner: we will say that a subset U ⊂ Rn is open if for any point x ∈ U . and {0. — Let us consider the vector space Rn . Example 1.3) For any finite subset ⊂ (X ). (1. — The Sierpi´ ski space Σ0 is the set {0. . — For any set S.3. {∅.1. A subspace of X is thus a subset A ⊂ X . any set S can be given the cocountable topology. Moreover.2. Exercise 3. {0. — Any set S can be given the cofinite topology. This is known as the standard topology on Rn . Σ1 . That is. Then A may be given the subspace topology.2. Given a Kuratowski closure operator κ.6.

1. which is generated by the set {(a. p (X )).7. which is all of X ). Then let p (X ) be the set of all unions of elements of . Now U= U∈ U∈ i ∈IU Bi = U∈ .10. Show also that Q is dense in R. — Define a topology in the following manner: let be the set of all finite intersections of elements of (including the empty intersection. This leads us to the notion of a subbase for a topology. one may write any element U ∈ as a union of elements of : U= i∈IU Bi . ∞) and (−∞. b ) are shorthand: (a. one may define the order topology. it is frequently difficult to give an explicit description of the elements of p(X ). — We can now combine the previous two examples to get a whole world of interesting topological spaces. — Suppose X a set. ∞) | a ∈ R} ∪ {(−∞. b ) | b ∈ X } ⊂ (X ). where Bi ∈ . this intersection is a union of elements of . Definition 1. Thus p (X ) is the unique coarsest topology containing . Exercise 5. It is equal to the closure B(0.6.5) The irrational numbers R − Q ⊂ R can also be given the subspace topology.9.2.2. (1.2.6. — More generally.6. iU ∈IU U ∈ B iU . 1) of the ball of radius 1 at 0.5] is generated by the set {(a. Example 1. — One of the most difficult aspects of topology is that. so any such topology must be finer than p (X ). (1.6. Moreover. . — Show that the standard topology on R [1. b ) | b ∈ R} ⊂ (R). where (a.2.2) The closed n-disk is the set D n := {x ∈ Rn | ||x|| ≤ 1}.8. b ) := {x ∈ X | x < b }.2. and suppose coarsest topology p (X ) containing . — In the situation of the previous proposition. Unions of unions of elements of are unions of elements of .2. THE CATEGORY OF TOPOLOGICAL SPACES Example 1.6. despite the fact that the subset p(X ) ⊂ (X ) is the information that is needed in order to specify a topology on a set X .(1) Any topology containing must contain all unions of finite intersections of elements of .2. (1. ∞) | a ∈ X } ∪ {(−∞.6. Then there exists a unique Proof. and since each set U ∈ BiU is an element of by construction. contemplate the case when has cardinality 2. (1. — Show that the subspace topology on Q is not the discrete topology on Q.3) The (n − 1)-sphere is the set S n−1 := {x ∈ Rn | ||x|| ≤ 1}.4) The rational numbers Q ⊂ R can be given the subspace topology. . ⊂ (X ) a collection of subsets of X . (1. Exercise 4. the topology p (X ) is said to be generated by and the set is said to be a subbase for the topological space (X .2.1) The open ball of radius 1 at 0 is sometimes called the n-cell.2. ∞) := {x ∈ X | x > a} (1) If and (−∞.2. Proposition 1. this formula seems confusing. This is indeed a topology. It is equal to the complement of the n-cell in this n-disk.2. on any totally ordered set X . for any finite subset ⊂ p (X ).12 CHAPTER 1.

Then the inclusion map i : A (given by the rule i(a) = a) is continuous.3.3. — Suppose X and Y two spaces.1) If n = 0.3.3.2. and the standard topology is just the order topology. Then we have two projection maps / X and pr : X × Y /Y .3. (6.3. If X and Y are spaces.3. then R0 = . that is.3. the way we defined the product topology guaranteed that this property was as sharp as possible. Exercise 6. Let / S δ is open. Example 1.3.11. Definition 1.3. Exercise 8. (6. then the product topology is the coarsest topology on X × Y such that the two projection maps /Y .3. Example 1. — Of course the identity map of any space is a homeomorphism. — A map f : X f U ∈ p(Y ). — Suppose X and Y two spaces. p (X )). and the standard topology is just the discrete topology. then Rn = Rn−1 × R. More generally.1. then R1 = R. Both of these maps are continuous. however.9. Then p(X ) is finer than p (X ) if and only if the identity map on X is a continuous map / (X . — Open maps need not be continuous. / R given by the rule f (x) = x 2 . and suppose A ⊂ X is a subset. — Once again. i:A Example 1. y) | x ∈ X . Continuous maps and homeomorphisms / Y is continuous if the inverse Definition 1. Example 1. the set f U ∈ p(X ).3. the subspace topology is rigged precisely to make this happen. and consider the product X × Y with the product topology.5. 1. — Consider the function f : R open. that is. — In fact. Example 1. but Hence the floor function · : R not continuous.3) If n > 1. y) = x and pr2 (x. in order for us endow S with the discrete topology S δ .1.2) If n = 1. — Suppose X is a set with two topologies p(X ) and p (X ). then the projection maps pr1 and pr2 are both continuous (as we have seen) and open. — If X and Y are topological spaces. equipped with the subspace topology. — Show that the standard topology on Rn can be described inductively in the following way. if for any U ∈ p(Y ). and the standard topology on Rn is just the product topology of the standard topology on Rn−1 with the standard topology on R. — Suppose X and Y topological spaces. / Z (which just rounds any real number down to the nearest integer) is open. This is continuous. if for any U ∈ p(X ). then the subspace topology on A is the coarsest topology on A such that the inclusion map / X is continuous. Show that if X is a space. — The composition of two continuous functions is continuous. Suppose X a topological space. pr1 : X × Y 2 given by the rules pr1 (x.2. Then their product X × Y := {(x. y ∈ Y } has a topology..7. called the product topology generated by the set {U × Y | U ∈ p(X )} ∪ {X × V | V ∈ p(Y )}. but not . p(X )) Example 1. suppose X a /X space. Then one sees that any map p : X p to be continuous. for any s ∈ S the fiber over s — i. Then a map f : X −1 image of any open set is open. and suppose S any set. the inverse image f −1 {s} — would have to be an open set. CONTINUOUS MAPS AND HOMEOMORPHISMS 13 Example 1. and A ⊂ X is a subset.e. (6. (X .6. Example 1. / X and pr : X × Y pr1 : X × Y 2 are continuous.4. / Y is open if the image of any open set is open.8. y) = y. as you showed above. Exercise 7.

3.14 CHAPTER 1. the same phenomenon happens here. which is again an open set.12. / Y is a bijection.14. As we will see. f −1 : p(Y ) / Y is a continuous map. sets in the second) guaranteed that the objects involved were the same. ) for any x ∈ Rn and any > 0. Show that any two n-dimensional balls B n (x. Now contemplate the map with the subspace topology from C = / U (1) ρ : [0. In group theory or set theory. find a homeomorphism between R and the ray (0.3. — Of course the identity map is a homeomorphism. Example 1. an isomorphism or bijection between two objects (groups in the first case. — Suppose n is a nonnegative integer. (1) In the first description.3.16. Two spaces X and Y are said to be homeomorphic if there is a homeomorphism X 1.10. — Continuing with that idea. if f : X −1 homeomorphism is the statement that both f and f are continuous. — The two Sierpi´ ski spaces Σ0 and Σ1 are homeomorphic. i:A / Y is a homeomorphism if it is both a bijection and an open map.3. — Here’s another way to think of homeomorphism. we were assuming that f is continuous. 2π) given by the rule ρ(x) = e i x . — A continuous map f : X /Y .1) the n-dimensional ball B n (x. then the inverse The statement that f is a homeomorphism is the statement that both f and f −1 are bijections. Exercise 11. but you have the same open sets up to that relabeling! 1. then the inclusion map / X is open if and only if A is an open subset of X . and A ⊂ X is a subset with the subspace topology. — Here’s a very important example to contemplate. in order for f : X be a bijective continuous map. — Show that if X is a space. up to some “relabeling. THE CATEGORY OF TOPOLOGICAL SPACES Exercise 9. ) in Rn are homeomorphic.3. since we’ve been thinking of topological spaces as sets with a concept of “nearness” of points. δ) and B n (y. It is a bijective continuous map. Consider the tangent function as a map tan : − π . but it had better not be a homeomorphism! Example 1. we do not only require that f Example 1.11.) Example 1. — A homeomorphism between two topological spaces is a guarantee that those two spaces are indistinguishable from the point of view of topology. — Note that.13. — Suppose n is a nonnegative integer. This might seem a little odd. and you might as well get used to it now. In the case of homeomorphisms. (12. and we were talking about the function Y inverse to f .3. ∞). Definition 1. It’s a homeomorphism! This suggests something that may at first seem surprising: topology doesn’t “see” the difference between an infinitely long line and an open interval of finite length. (We’re being totally sloppy about notation here. and so are their closures. often in close proximity to one another. n / Y to be a homeomorphism. Mathematicians use the symbol f −1 in two different ways. But this leads to an important idea about what topology actually is: one might say that topology is what is left over from geometry when you throw out the notion of distance. but only because everyone else is too.2) the euclidean space Rn . we were assuming that f is a bijection. and . not only do you have the same points up to some relabeling. The statement that f is a Alternately. Exercise 10. This wraps the half-open interval around the circle.” so anything you knew about the first group or set was reflected in the second group of set. then it has an inverse f −1 : Y / X . If f : X image can be viewed as a map / p(X ). (2) In the second p(Y ) / X which is the description. Show that the following three spaces are all homeomorphic: (12.3. π 2 2 / R. and we were talking about the function / p(X ) that takes an open set U to its inverse image f −1 U . Exercise 12.15. That is not enough: let us think of the set U (1) := {z ∈ C | |z| = 1} ∼ R2 .

1) in R2 : S 1 (1) := {θ ∈ R2 | ||θ − (0. ∞) | c. and we set F (∞) := 0. — A little bit of thought should convince you that the space R ∪ {∞} we’ve described in the previous example is in fact a circle. i ) i=1 for any decomposition n = m k.3. For any point θ ∈ S 1 (1) − {(0. where we set F (x) := 1/x for any x ∈ R − {0}.3) Show that g can be extended to a homeomorphism G : S 1 (1) / R ∪ {∞} by setting G(θ) := g (θ) for any θ ∈ S 1 (1) − {(0.4) In the previous example.1) Write an explicit formula for g .) The subspace R ⊂ R ∪ {∞} is open.17. this is continuous! The space R ∪ {∞} we’ve constructed here is called a compactification of R.1. because in elementary school we were told that 1/0 is “undefined. 1)|| = 1}. and we note that we can extend it to a map F : Now we think of our map f : R − {0} / R ∪ {∞} . b ) | a.2) Show that g is a homeomorphism. Let’s topologize this set in the following way: we will contemplate the set {(a.18. let Now let us define a map g : S 1 (1) − {(0. If we wanted to add a point that would play the role of 1/0. (13. — Here’s a challenging example. 2)} and G((0. 2) and θ in R . c) ∪ {∞} ∪ (d . f : R − {0} Since f is its own inverse. Translate our circle S 1 so that we are looking at the unit circle around the point (0. we will look at all unions of all finite intersections of open intervals in R along with sets (−∞. — Here’s another example that merits some contemplation. a homeomorphism from a space to itself is called an automorphism. we constructed a homeomorphism F : R ∪ {∞} identity).) Of course we have removed the point 0 ∈ R. (13. (We’ll explain more about that later. / R in the following way. Consider the function f (x) = 1/x. / R ∪ {∞} (which is not the (13. (13. d ∈ R} ⊂ (R ∪ {∞}). as x approaches 0 from the left.3. b ∈ R} ∪ {(−∞. That is. i =1 i and points xi ∈ Rki . c) ∪ {∞} ∪ (d . ∞). CONTINUOUS MAPS AND HOMEOMORPHISMS 15 (12. m B ki (xi . Let’s find a homeomorphism that makes this explicit. 1/x increases without bound.3. 2)}. we can rewrite this as a homeomorphism G −1 ◦ F ◦ G : S 1 (1) / S 1 (1). and any real Example 1. Now define g (θ) to be the x-coordinate of the point of 2 intersection between θ and the x-axis {(x. Consider the spaces R m and Rn with the standard topology. c) or (d . where ∞ here is just a formal symbol that we’ve tacked on. Using the homeomorphism G. What homeomorphism is this? How might you draw its graph? . 2)} 2 θ be the unique line passing through (0. y) ∈ R | y = 0}. then this leads us to the following idea: consider the set R {∞}. 2)) := ∞. we set F (0) := ∞. / R − {0} . and we will contemplate the topology generated by this set.” But let’s try to define it anyhow. It is a fact that they are homeomorphic if and only if m = n. 1/x decreases without bound. Exercise 13. Certainly one direction here is obvious! But how on earth could you show that there is no homeomorphism between two spaces? What kind of contradiction could you get from the existence of a homeomorphism? These kinds of questions will lead us naturally to Example 1. ∞) that “wrap around” our added point ∞. We note that. and ∞ is contained in the closure of any ray (−∞. as x approaches 0 from the right. f here is a homeomorphism.3) the product space numbers i > 0. R ∪ {∞} With the topology we’ve given R ∪ {∞}. This can regarded as a map / R − {0} . (In general.

1) Show that h is a homeomorphism. moreover. 1].16 CHAPTER 1. which we regard as a map /R. Think.6) Show that the map / R ∪ {∞} p : R ∪ {−∞. that for any topological space / X such that f (−∞) = f (∞). (14. ∞) ⊂ R in R ∪ {−∞. there exists a unique map X .4) Describe an extension of h to a homeomorphism H : R ∪ {−∞. ∞} (where −∞ and ∞ are each just formal symbols) with the following properties. Show. THE CATEGORY OF TOPOLOGICAL SPACES Exercise 14. (14. f : R ∪ {∞} . h:R (14.3) Describe a space R∪{−∞. a] := {−∞} ∪ (−∞.2) Show that h(x) increases without bound as x does. ∞} (with the topologies defined here and above) is continuous. ∞} is the set [−∞. of the function h(x) = x 3 . (2) The closure of a ray (−∞. (14.5) Show that the space R ∪ {−∞. ∞] := [b . If we thought of another example. ∞} . (1) R is an open subspace of R ∪ {−∞. (Be careful if you decide to study h −1 !) (14. ∞} / X such that f = f ◦ p. (14. for example. we were led to consider it by our example. ∞} is the set [b . ∞} / R ∪ {−∞. and any continuous map f : R ∪ {−∞. ∞} is homeomorphic to the interval [−1. ∞) ∪ {∞}. — The space R∪{∞} is not the only compactification of R. ∞}. a]. and likewise h(x) decreases without bound as x does. and the closure of a ray (b . we might be led to a different compactification. a) ⊂ R in R ∪ {−∞.

. but this is automatic since f is continuous. f −1V is open in X /A if and only if f −1V = q −1 f −1V is open in X . suppose Y a topological space. — Subbases for topologies permit the easy identification of continuous maps. and suppose A ⊂ X a subspace. it always does.1. We say that a subset U ⊂ X /A is open if and only if the inverse image q −1 U is open in X . this is the finest possible topology on X /A such that q is continuous. otherwise. This forces q to be continuous. — Show that the above description of X /A is unique up to unique homeomorphism. and suppose A ⊂ X a subspace. and suppose f : X / Y in the following way f (A) = {a} for some element a ∈ Y . 2. Moreover. The definition above is nonconstructive. Here’s one way to construct it. show that the quotient D n /S n−1 is homeomorphic to S n .1. — Contemplate the quotient X /∅. and suppose a subbase for the topology on Y . there exists a unique continuous map f : X /A /Y f :X such that f (α) = a and f = f ◦ q. α) and a continuous map q : X / Y such that f (A) = {a} for some element a ∈ Y .2. — Suppose X a space. and both q : X / W such that h ◦ q = q . — For any integer n ≥ 0. Let us define f : X /A f (x) := x a if x ∈ X − A. consider the map q : X q(x) := x α if x ∈ X − A. To see this.2. otherwise. That is. 2. then there is a unique homeomorphism h : W Exercise 16. Then show that a map f : X if and only if the inverse image of any element U ∈ under f is open in X . — Suppose X a space.) / X /A satisfy the conditions of the definition I claim that the space X /A and the continuous map q : X / Y a continuous map such that of a quotient. In fact. The answer may surprise you! 2.1. Quotients Definition 2.1. / X /A.CHAPTER 2 CONSTRUCTING TOPOLOGIES AND CONTINUOUS MAPS 2. Suppose X and Y topo/ Y is continuous logical spaces. Coverings and locality Exercise 17. given by the formula Let X /A = (X − A) {α}. (In fact. One sees that this is the only possible map such that f = f ◦ q. Then a quotient of X by A is a pointed space / X /A such that q(A) = {α}. A ⊂ X is a subspace. / W and q : X that if X is a space. show / W are each quotients of X by A. Exercise 15. in the sense that it does not make it apparent that such a space X /A exists. f is continuous: for any open set V ⊂ Y . and for any space Y and any continuous map (X /A. We topologize X /A in the following manner.3.1.

W ∈ . there is a neighborhood V ∈ x with V ⊂ W .4. one has fV : V fV |V ∩W = fW |V ∩W : V ∩ W Then there exists a unique continuous map f : X f |V = fV : V /Y .7. Proposition 2. — Suppose X and Y two topological spaces. — Suppose X a space. Then U is open if and only if.9. which is open. Exercise 18. — The previous result is helpful because it permits one to check only a small number of conditions on subsets of Y in order to guarantee the continuity of a function. for any elements V . which is the union of all the open sets W ⊂ U . and suppose x ∈ X . Hence f is continuous at every point x ∈ X . Then a base for X is a collection point x ∈ X . Proposition 2. The claim is that f −1 (V ) is open. one may for the interior U ◦ of U . One says that is an open covering of X if every V ∈ is an open set. and suppose that. whence x ∈ U ◦ . One sees that ◦ U = U : indeed. / Y a map. The the following are equivalent for a subset U ⊂ X . that f is continuous at every point x ∈ X . of A = V ∈ A ∩ V .2.1) The set U is open in X . Then f is continuous Proof. and suppose U ⊂ X a subset thereof. Lemma 2. Conversely. U ◦ ⊂ U . Definition 2. / Y such that for any V ∈ . 2.2. suppose f : X Then one says that f is continuous at x if for any neighborhood V of f (x) ∈ Y . For any neighborhood W of x. For a subspace A ⊂ X . / Y a map. /Y . for any point x ∈ U . or equivalently.3. Suppose. CONSTRUCTING TOPOLOGIES AND CONTINUOUS MAPS 2. — Suppose f continuous. Proof. 2. — A base for a topological space is also a subbase for the topology p(X ). there exists an element V ∈ x such that V ⊂ U . and suppose x ∈ X a point.2. . and for any point x ∈ U . — Suppose X a topological space. and suppose f : X if and only if it is continuous for every point x ∈ X . likewise.2. with W ⊂ f −1 (V ). — Suppose X a space.18 CHAPTER 2.10. which is open as well.2. the set | x ∈ U} x := {U ∈ is a fundamental system of neighborhoods of x ∈ X . that Definition 2. Definition 2. there exists an open neighborhood W of x with W ⊂ U . and its inverse image f −1 (V ◦ ). — Suppose X a topological space. Suppose V ⊂ Y an open set. Now the image f ( f −1 (V ◦ )) ⊂ V ◦ ⊂ V . (2. — Suppose X is a topological space. — If U is open. — Suppose X and Y two topological spaces.2. ⊂ p(X ) such that for any . — Suppose X a topological space. For any point x ∈ f −1 (V ). Then a fundamental system of neighborhoods of x ∈ X is a collection x of neighborhoods of x with the following property.1. there is a neighborhood U of x ∈ X such that f (U ) ⊂ V . and suppose x ∈ X . Then for any neighborhood V of f (x) ∈ Y . I could easily as well have written “open neighborhood”: f is continuous at x if and only if for any open neighborhood V of f (x) ∈ Y .2.10.2. there exists an open neighborhood W of x with W ⊂ U . one may contemplate the interior V ◦ . there is an open neighborhood W of x ∈ f −1 (V ) with f (W ) ⊂ V . one says that is an closed covering of X if every V ∈ is an closed set. for any V ∈ / Y is a continuous function.2.2) For any point x ∈ U .6. and suppose a base for the topology.10. Now suppose.2. equivalently. then of course U itself is an open neighborhood of any point x ∈ U . — Note that I have written “neighborhood” in the above definition. A collection ⊂ (X ) of subspaces of X is said to cover X — or to be a covering of X — if X = V ∈ V . conversely.8. and suppose an open covering of X . Now we apply the lemma above to see that f −1 (V ) is open. we say that a collection ⊂ (X ) of subspaces of X is said to cover A — or to be a covering of A — if A ⊂ V ∈ V . Definition 2. or.2.2. there is an open neighborhood U of x ∈ X such that f (U ) ⊂ V . (2.2.5.

so that U ⊃ U .1) The elements of cover X . there is an element W ∈ x such that W ⊂ U ∩ V . there exists an element V ∈ x such that V ⊂ U . — Suppose U is open in X . the floor function · : R / R is continuous. — For any totally ordered set X .2.11.2. one may choose any point a ∈ (a. Since x is an open system of neighborhoods of x ∈ X .14. — Suppose X a set with two topologies p(X ) and p (X ). Then the set {B n (x. / R is continuous if and only if. for any open interval (a. Exercise 19. for any x ∈ X and any element U ∈ there exists an element V ∈ x such that V ⊂ U .2) For any U . U is an open neighborhood of x.11.V ∈ and any element x ∈ U ∩ V . so by the axioms for a topological space. p(X )) and a base for (X . This shows that the second condition implies the third. Proposition 2. By definition.2. — Here is a standard counterexample in topology. (One may also define the upper limit topology R u . (Here. everyone must see it once in their lives. b ). This completes the proof. — Suppose n ≥ 0 an integer. Indeed. + →0 In particular. there is an element V ∈ x such that V ⊂ U . suppose a base for (X .12.) Example 2.2. COVERINGS AND LOCALITY 19 (2. (2.11. b ) for a. b ∈ X } is a base for the order topology on X . Suppose that. then for any point x ∈ U . suppose that U can be written as the union of elements of . . Then a subset ⊂ p(X ) is a base for the topology if and only if the following conditions are satisfied. Finally. for every point x ∈ U .2. b ) | a. one writes (a. ∈ R} is then a basis for R . the set {(a. b ). b ).10. A map f : R f (x) = lim f (x + ).2. ) | x ∈ Qn . The Sorgenfrey line R is the set R with the lower limit topology. ε ∈ Q} is a base for the standard topology on euclidean space Rn .2. b . This shows that the first condition implies the second. Then (using the axiom of choice) one may select. x The previous exercise can be used to show that the Sorgenfrey line is finer than the standard (= order) topology on R. b ) | a. The union U := x∈U V x contains every point of U by definition. b ) ⊂ (a. Proof. for any x ∈ R.13. Thus U is a union of elements of . it follows that U ⊂ U . b ∈ R. p (X )). which is homeomorphic to R via the map / − x . x Example 2. This is the topology generated by the half-open intervals [a. and then [a . Show that p(X ) is finer than p (X ) if and only if. and since V x ⊂ U for every x ∈ U .) One can check fairly easily that these basic open sets [a.2. one element V x ∈ x such that V x ⊂ U . U is open as well. — Suppose X a topological space. Example 2. b ) are both open and closed. . (2. b ) := {x ∈ X | a < x < b }. The set := {[a. for any point x ∈ U .2.3) The set U can be written as the union of elements of . every element of is open.

.

Then there exists some U∈ such that f −1 U ∈ and a ∈ f −1 U . whence f (a) ∈ U . i ) for i = 1. one may choose a collection ⊂ p(X ) such that = {U ∈ p(A) | U = U ∩ A for someU ∈ Now the set = {U ∈ p(A) | U = U ∩ A for someU ∈ is a finite subcovering of . m and suppose x a point of one of the balls. — A closed subspace of a compact space is compact. suppose y = f (a) ∈ f (A) a point with a ∈ A.1. — A topological space X is compact if any open covering has a finite subcovering.1. .3. Now } ∪ {X − A} is an open covering of A. A compact topological space is sometimes called a compactum. but there is no finite subcovering. Indeed. Example 3.6. Proof.1.1. n. Since each element U ∈ is open in A.1. i = 1. and suppose A ⊂ X closed.1. Now the claim is that the subset }. ) form an open covering. and the indiscrete topology on any set is compact. / Y a continuous function.4. (3. Lemma 3. Lemma 3. indeed.2.2) [Bolzano–Weierstraß] Every sequence of points in A has a convergent subsequence.1) A is compact. 2.1.6. Show that the product space n=1 Xi is compact. To see this. r ) ⊃ i=1 B n (xi . write p(X ) for the set of all compact subsets of X . and suppose f : X any compact subspace A ⊂ X is compact.1.5. . Then one may choose a real number r so that B n (x. hence there is a finite subcovering Exercise 20.CHAPTER 3 COMPACTNESS I 3.1. consider any finite covering of Rn by balls B n (xi .) i Theorem 3. (This is the easy case of Tychonoff’s Theorem. . m. . Compact spaces Definition 3. then the set { f −1 U | U ∈ is an open covering of A.1.6. ⊂ . . .3) [Heine–Borel] A is closed and bounded. to which we shall return. Proof. (3. . For any space X . which thus has a finite subcovering {U ∈ | f −1 U ∈ } . — The following are equivalent for a subspace A ⊂ Rn . 2. i ). . is a covering of f (A). the open balls B n (x. Example 3. — Suppose X a compactum. Suppose an open covering of A. — Suppose X and Y spaces. — Any finite space is compact. . — The euclidean spaces Rn are not compact. Then the image f (A) of which is finite.6.1. }. — Suppose an open covering of f (A). (3. — Suppose we are given a finite collection {Xi } of compact spaces.

If every subcover of has a finite subcover. it is contained in a box [−a. and let collection of all bourgeois families of open sets of X that contain : ( ) := { ∈ ∈ | ( ⊂ }. and suppose a subbase for also an open cover of p(X ). yielding the desired contradiction. ( ) be the This family is obviously of finite character. then every element A ∈ is contained in a maximal set M ∈ . since if x ∈ A − A is a point “at the boundary. we see that X is compact if and only if any bourgeois family of open sets is exiguous. We can also rewrite our assumption on the subbase : we are assuming that any bourgeois subfamily of is exiguous. Lemma 3. a − c/2] with finitely many elements of . a] is compact. Suppose an open covering of [−a. Finally. since otherwise there exists a sequence (xi )i≥0 of points in A such that ||xi || → ∞. i. Using our new terminology. and let’s call bourgeois if any finite subset of is exiguous.3 (Alexander Subbase). then one of the Ui is itself an element of . Let’s call a family ⊂ (X ) exiguous if it does not cover X . So suppose a bourgeois family of open sets. one has B(c. Proof.2. and there would be a sequence i of positive real numbers such that the balls B n (xi .2.2 (Tukey). Then f attains both Theorem 3. Suppose. ). a set M such that no element of properly contains M . and suppose f : X a maximum and a minimum value. The union of every chain of elements of .2. — Prove the following Key Fact about our maximal element : If {Ui }n=1 is a finite collection of open sets of X such that the intersection i an element of . it is enough to show that [−a. COMPACTNESS I Proof. that c < a. then let c = sup{x ∈ R | [−a.” one can construct a sequence of points of A converging to x. Since A is bounded. so by Zorn’s Lemma. — Suppose X a compact space. 3.1. But then {B n (xi . It is enough to show that Exercise 22. Write := { ∈ ( p(X )) | is bourgeois}. — A collection of sets is said to be of finite character if the following condition holds: a set A is an element of if and only if every finite subset of A is an element of . The Tychonoff Product Theorem Definition 3. Since every closed subspace of a compact space is compact. it is enough to show that [−a. x] can be covered by finitely many elements of }. Then if is of finite character.e. Suppose A is closed and bounded. to generate a contradiction. Then for some > 0. Our aim is to show that any bourgeois family of open sets is exiguous. Then A must be bounded. one may cover [−a. let us show that the third condition implies the first. a]n for some a > 0. — Let us show that the first of these conditions implies the second. — Give a partial ordering by inclusion. then X is compact. a]. must also belong to p(X ) that is / R a continuous function. Proof. is exiguous. so let’s introduce some new terminology to make things a little simpler. i )}i ≥0 would be an open cover of {xi }i ≥0 with no finite subcover. By the definition of c. ) ⊂ U . a + c/2]. then let U ∈ be an element of the open covering containing c. — The language here can get a tad confusing. — Suppose X a topological space. Suppose A compact. Exercise 21. n U i =1 i is contained in . then the set {xi }i ≥0 would be a closed subset of A. i ) would be disjoint. a]n is compact.22 CHAPTER 3. Let us show that the second property implies the third. and suppose (xi )i≥0 a sequence of points in A. It must also be closed.. and Tukey’s Lemma says that it contains a maximal element this is now our goal. If there were no convergent subsequence of (xi )i ≥0 .2. By the previous exercise. every element A ∈ is contained in a maximal element. — Suppose a nonempty collection of sets. adding in U to that set yields a finite open covering of [−a. Suppose A has the property that every sequence has a convergent subsequence. hence compact.

U ) | K ∈ This is called the compact-open topology. then since is a subbase. there is a finite intersection W := i=1 Ui of elements Ui ∈ such that x ∈ W and W ⊂ U . It is obvious that α is bourgeois and hence exiguous (since Xα is compact). d ) a metric space. and suppose (Y. The compact-open topology Definition 3. there exists a natural number N such that for any x ∈ X and any n ≥ N . — Suppose X a totally ordered set.3. But remember the intersection ∩ is exiguous. in other words. and thus (by our assumption on ) U⊂ U∈ ∩ U. — Suppose X = S δ a discrete space.4 (Tychonoff Product). Show that the compact-open topology on the product topology on s ∈S Y . A sequence of continuous functions fn ∈ (X . again. Y ) coincides with . p(X ). just as we had hoped. now it’s time to consider our subbase is exiguous. we are assuming that any bourgeois subcollection of ∩ . B) := { f ∈ Now we topologize the set { (K. Suppose a bourgeois subcollection of . Proof. n Suppose x ∈ U ∈ . Y ) the set of all continuous maps (X .3. Let’s contemplate the intersection exiguous. — Suppose X and Y two topological spaces. Y ) if n→∞ x∈X . (X . But now the point / α x := (xα )α∈A is not a member of any element of 3. Show that X is compact (with the order topology) if and only if it is order-complete. f (x)) < . Let us say that X is order-complete if every subset of X has a supremum in X .3. and give topology generated by the subbase α∈A Xα the product topology. Y ) with the compact-open topology.1. Take the topology generated by the subbase lim sup d ( fn (x). α Let’s use the Alexander Subbase Theorem. This means that x ∈ Ui ∈ ∩ . f (x)) = 0. / Y converges uniformly if and only if it converges in Prove that a sequence fn of continuous functions X (X . Y ) is said to converge uniformly to a continuous function f ∈ (X . For every α ∈ A. write X (A. — Suppose {Xα | α ∈ A} a collection of compact spaces. we can choose a point xα ∈ Xα such that xα ∈ U ∈ U . it is bourgeois. This is the := {pr−1 (U ) | α ∈ A. Is the same true if X is not compact? Exercise 25. so we conclude that is also! Exercise 23. d ( fn (x). and U ∈ p(Y )}. Theorem 3. U ∈ p(Xα )}.) j =1 All right. Exercise 24. one of the Ui is itself and element of . For any subsets A ⊂ X and B ⊂ Y . — The product of compact spaces is compact. (X . Y ) in the following manner. Hence for each α ∈ A. (X . By the Key Fact above. So is exiguous. — Suppose X any compact space. for every > 0. set α of is exigu- := {U ∈ p(Xα ) | pr−1 (U ) ∈ α }. Y ) | f (A) ⊂ B}. THE COMPACT-OPEN TOPOLOGY 23 (Hint: if U is an open set of X that is not contained in of such that {U } ∪ {V j } m is a covering of X . Denote by / Y . We need to show that any bourgeois subcollection ous. show that there is a finite collection {V j } m of elements j =1 .2. Now I claim that U∈ .3.

.

1) X is compact.1. In this section.. — Consider the following properties for a topological space X . (4.1.1. Then we have a diagram of implications (1) I II I( (3)  (4) Proof. in any metrizable space) they are indistinct and. — These three notions seem remarkably similar.1. will contemplate some examples. First. — Show that a space X is countably compact if and only if every nested sequence X ⊃ V1 ⊃ V2 ⊃ V3 ⊃ · · · of closed. compactness is equivalent to the Heine–Borel property (closed and bounded) of the Bolzano–Weierstraß property (the property every sequence has a convergent subsequence).1. Let us now see that a sequentially compact space is countably compact. if this choice is always possible. and let V p be the first of the sequence of Ui ’s not covered by the V0 ∪ V1 ∪ · · · ∪ V p−1 .2) X is sequentially compact. This doesn’t quite cut it for more general spaces.1.1) One says that X is limit point compact if and only if every infinite subset A of X has a limit point (i. Proceeding inductively. Definition 4. (4. nonempty subspaces of X has a nonempty intersection.4) X is limit point compact.e. (4.3. moreover. If this choice is impossible at any (finite) stage.3) X is countably compact.CHAPTER 4 COMPACTNESS II 4. then we have our required finite subcover. a point x ∈ X that is contained in A − {x}). (4. and they are very different from compactness. and in Rn (and more generally. — It is obvious that a compact spaces is countably compact. (4.1. To see this. let’s see what implications we do have. they are different conditions in general.1. — Suppose X a topological space. uu v~ uu (2) .3. Exercise 26. However.3) One says that X is sequentially compact if and only if every sequence x : N quence.1. (4.2) One says that X is countably compact if and only if every countable open cover has a finite subcover. Proposition 4.3. / X has a convergent subse(4.1. then for every n ∈ N we may select an element xn ∈ Vn such that for any i < n.1. and we’ll also talk about the notion of a net or Moore–Smith sequence.1. we’ll discuss analogues and generalizations of these notions.3. 4.1. write V0 := U0 .1.2.1.3. including a key counterexample. Forms of compactness We have already seen that for subspaces of a euclidean space. Suppose = {Ui }i ≥0 a countable open cover of X . equivalent to compactness.

4. }. With this definition. b ) if either a < a or else a = a and b < b .1. It has a successor ω + 1. Let us now see that a countably compact space is limit point compact. every ordinal α is equal to the set [0.4. (4. then β A. this is an open cover of X . (2) β ∈ γ . we can choose an open neighborhood Ua such that Ua ∩ A = {a}. Now write X = (X − A) ∪ a∈A Ua . which must admit a finite subcover.4. an ordinal is a set A such that the following conditions hold. But it admits no proper subcover at all. we can concatenate them by well ordering the disjoint union α β so that the order within α and β is preserved. the minimal ordinal. COMPACTNESS II xn ∈ Vi . Suppose A any set of ordinals. ion. a contradiction. Then y ∈ Vn for some integer n. Now sequential compactness guarantees that the sequence (xn )n≥0 has a convergent subsequence (xni )i ≥0 . any ordinal β < α is also an element of A.1. then A is well-ordered by this relation.1. then one of the following is true: (1) β = γ . so that (a. and every element of α is strictly less than every element of β. and its successor has a successor ω + 2. and suppose B ⊂ X a subset with no limit point. . for every integer n > 0 the ordinal n := {0. As a result. and so one may set α · β := ord(α × β). This is well-ordered. it is not the case that a < a. / let y be the limit of this subsequence. whence B is itself finite. let’s recall that a strict well-ordering on a set A is a well-ordering < that is irreflexive.1.1. there is a supremum of any set A of ordinals: it is simply the union of the elements of A. Given two ordinals α and β. . . we define 0 := ∅. (4. We won’t prove it here. Of course any countable subset A ⊂ B has no limit point. and for every point a ∈ A. The minimal ordinal 0 is a limit ordinal. so A must be finite. . First. (4. so for every i 0 (in particular for i so that ni > n). then there exists an element β ∈ A such that γ ∩ β = ∅. in the sense that for any element a ∈ A. Some ordinals are not successors. 1.1’) If β ∈ A.4. Every ordinal α has a successor. so A is closed. for every element α ∈ A. we have to appeal to some basic facts about ordinals.26 CHAPTER 4. α) of all ordinals less than α. one can give the product α × β the lexicographic ordering. if we build a well-ordered set A.1. γ ∈ A. but in fact every well-ordered set is order-isomorphic to a unique ordinal in a unique fash/ ord(A). Suppose X a countably compact space.2) A is strictly well-ordered with respect to set membership. Consequently.3’) If ∅ = β A. The first infinite ordinal is also the first nonzero limit ordinal. n − 1}. we define. Example 4. . These are the finite ordinals. b ) < (a . If α and β are ordinals.2’) If β. We can define operations of ordinal arithmetic in the following manner. a set A of ordinals is itself an ordinal if and only if. namely α + 1 := α ∪ {α}. there is an ordinal ord(A) and an order-isomorphism A In particular. Using this. Now set α + β := ord(α β). we have xni ∈ V m . . and (4. . Put differently. . Now we come to von Neumann’s definition of ordinal: a set A is an ordinal if (4. Now recursively.1) every element of A is also a subset of A. 1. Let us show that every countable subset A ⊂ B is finite. Similarly. these are called limit ordinals. it is ω := {0.4. — To construct this example. etc.4. then we say that α < β if α ∈ β. or (3) γ ∈ β.

— Consider the discrete space {0. Surely x ∈ I x . let us concentrate on this one. and suppose x ∈ X . . . but not countably compact. ω + 1. . however. . .4. ω ωω it can be expressed as the supremum of the sequence 0 = sup{ω. 1}δ .1] {0. one has I x = {x}. ω ω . Thus I = {x}.2. . . but they are non-isomorphic as wellordered sets. . . . ω 3 . the intersection I x of all closed neighborhoods of x is the singleton {x}. / β such that Note that none of these operations is commutative. respectively. = ω 0. ω 2 . Now the supremum of all countable cardinals is an uncountable cardinal ω1 .. y ∈ X . 1. Proof. but it might not be an interesting enterprise: Exercise 27. Proposition 4. called Hausdorffness. . since it contains no maximal element. and it is going to give us a useful counterexample now and in the future. This is the first uncountable cardinal. . . . . — A space X is said to be Hausdorff if for any two distinct points x. β)). ω · 2. whence X − V is a closed neighborhood of x not containing y. . . . .2. Exercise 28. ω + 2. We can give any of these sets the order topology. ω ω . for any point x ∈ X . Suppose x and y two distinct points of X . for any point x ∈ X . . . . ω ω The last one listed there was studied by Cantor: it is the smallest ordinal satisfying the equation 0 ω ωω . 0. and I claim that any point y ∈ X − {x} is not in I . Locally compact Hausdorff spaces The next variant of compactness we want to consider. So the infinite ordinals all have the same cardinality. — Show that for any ordinal α. suppose that. the order topology and the discrete topology coincide if and only if α contains no limit ordinal apart from 0. for any such point. 2. Show that the product X := x∈[0. since it appears so often. . . .. . .2. — Suppose X Hausdorff. ω ω .) 4.2. ω · 2 + 1. though both addition and multiplication are commutative.. Then since I x = {x}. there are neighborhoods U of x and V of y such that U ∩ V = ∅. ω ω . If we apply these operations to the finite ordinals and to ω. . local compactness. The surprising fact is that these ordinals are all countable. but it cannot be compact. one may find disjoint open neighborhoods U of x and V of y.. Indeed. one sets βα := ord(Map(α. Exercise 29. Definition 4. — A space X is Hausdorff if and only if. ω · 2 + 2. . and now the interior W ◦ and the complement X − W are disjoint open neighborhoods of x and y. we get a sequence of ordinals: 0. ω. . there exists a closed neighborhood W of x not containing y. compact. (Hint: such a space must have the property that it contains a finite subspace that is not closed. . . ω ω . . but not sequentially compact. is most interesting when coupled with a particular separation axiom.1. but for now. The set ω1 + 1 is. .2. LOCALLY COMPACT HAUSDORFF SPACES 27 We can generalize the lexicographic order to a well ordering on the set Map(α. }. . Every increasing sequence in ω1 converges to an element of ω1 . because every supremum of a countable set of countable ordinals is countable. . β) of all maps f : α only finitely many elements of α are mapped to an element larger than 0 ∈ β. We will discuss a variety of separation axioms later. Conversely. — Find a space that is limit point compact. 1}δ is compact. Hence the space ω1 (when equipped with the order topology) is sequentially compact.

so there is a finite collection of points y1 . .2. Exercise 32. g : X equalizer E( f . — The set E( f . — If S and T are spaces. — If X and Y are spaces. Example 4. — Any metric space (in particular any euclidean space Rn ) is Hausdorff. Exercise 31. COMPACTNESS II Exercise 30. . x) is the topology in which p(X ) := {U ∈ (X ) | x ∈ U } ∪ {∅}. f (s .2. .2. and suppose K ⊂ X a compact subspace.8. g ) := {(x. x) ∈ X × X } is a closed subspace of X × X . completely contained in U .e. t ) ∈ S × T | h(s) = t } is closed in S × T . y2 . — Show that a space X is Hausdorff if and only if the diagonal ∆X := {(x. More exotically.28 CHAPTER 4.9. and any product of any collection of Hausdorff spaces is Hausdorff. then for any continuous map h : S Γ h := {(s. the Corollary 4. x ) ∈ X × X | f (x) = g (x )} is closed in X × X . and T is Hausdorff. Show that there exist open sets U . and suppose K. Suppose x ∈ U .2. . Write U = X − K. The collection {Wy | y ∈ K} is an open cover of K. Proof. — If X is a Hausdorff space.2.6. and any discrete space is locally compact. — Apply the proposition to the case when X = S × T . . Show that for any point x ∈ X − K.V ⊂ X such that K ⊂ U . . and suppose U an open subset that contains a compact subspace K. Proposition 4. — Any subspace of a Hausdorff space is Hausdorff. yn such that the subcollection {Wyi | i ∈ {1. Then there exists an open set W whose closure is compact such that K ⊂ W ⊂ W ⊂ U . — Suppose K ⊂ X a compact subspace. L ⊂ V . L ⊂ X two disjoint compact subspaces.2. i.5. and Y is Hausdorff.3.2. . The cofinite topology on any infinite set is not. t ) = h(s) to obtain a closed subspace of S × T × S × T . — Suppose X a Hausdorff space. then any compact subspace K ⊂ X is closed. the graph / Y . Proposition 4. . then for any continuous maps f . For any element y ∈ K. and suppose x ∈ X . Proof. since its compact subspaces have empty interior. and U ∩ V = ∅. and intersect this with the diagonal ∆S×T . Y = T . Example 4. and Proof.g ) is the inverse image of ∆Y under the map ( f ..2. t ) = t . — Any compact space is of course locally compact. Definition 4. Now the n intersection i =1 Vyi is an open neighborhood of x that is disjoint from K. g ) : X × X /Y × Y . 2. and the closure V is compact. — A space is said to be locally compact if and only if any point has a compact neighborhood. Lemma 4. Proposition 4. — The space Q with the subspace topology is not locally compact. n}} covers n. Example 4.10. and g (s. / T .11. The particular point topology is always locally compact. — Suppose X a locally compact Hausdorff space X . suppose X any set. — Suppose X a locally compact Hausdorff space X . Then the particular point topology for (X .2.4. choose disjoint open neighborhoods Vy of x and Wy of y. . there exists a neighborhood U of x and an open set V containing K such that U ∩ V = ∅. The euclidean spaces Rn are locally compact.7.

— A closed subspace of an LCH space is itself LCH.13. can you add conditions on Z and Y so that it will be true? Exercise 37. / X is continuous and open.12. (37. — Suppose X a Hausdorff space.2. (4.16. x2 . (Why does our result follow from this?) Exercise 33. . If it’s true. a subspace of an LCH space X is itself LCH if and only if it can be written as the complement of a closed subspace inside another closed subspace. or else U contains ∞. — We now turn to the Alexandroff or one-point compactification of a space X .2. Proof.3) The subspace {(n.2) A discrete space.1) Every point x ∈ X has a fundamental system of compact neighborhoods. — Suppose X a space. the exercise 32 gives an open neighborhood V x of x and an open set W x whose closure is compact such that K ⊂ Vx and V x ∩ W x = ∅. Exercise 34. 4.12. 4.2. . and the resulting space is compact.12. (4. then for every point x ∈ X − U . — Describe the Alexandroff compactifications of the following spaces in terms of more familiar spaces. Consequently there are a finite number of elements x1 . — We often think of locally compact and Hausdorff as a single condition. — Suppose X a space. Then the following are equivalent.2. x) | n ∈ Zx ∈ R} ⊂ R2 . . (4. then exercise 32 applies directly. Are they homeomorphic? Prove it or give a counterexample. Compare the Alexandroff compactification (Y − Z) and the quotient space Y /Z.2. Thus it remains only to show that the third condition implies the first. .2. xn ∈ X − U such that the intersection (X − U ) ∩ n n W x is empty. (37.2.14.2) Every point x ∈ X has a closed compact neighborhood. — Prove the previous theorem.1) The inclusion map i : X (4.4. Now W := i =1 W x does the job.17. and X − U is closed and compact.4) Q with the subspace topology. (4. — More generally.15. and suppose Z ⊂ Y a closed subspace. (37. — Suppose Y an LCH space. is the LCH condition necessary? If it’s false. Exercise 36. — The first two conditions imply the third by definition. Then define X to be the set X {∞}. We have the following facts about the Alexandroff compactification.2. — The intersection of a closed set and a compact set is compact. Theorem 4.17. — Show that this indeed defines a topology on X .2.3) The Alexandroff compactification X is Hausdorff if and only if X is LCH. Proof. . which we abbreviate as LCH.2.2. with the following topology: a set U ⊂ X will be said to be open if and only if either U is an open subset of X . Exercise 35.17. Lemma 4.3) X is locally compact. If not.1) Rn . LOCALLY COMPACT HAUSDORFF SPACES 29 Proof. i=1 Proposition 4. since compact subspaces of a Hausdorff space are closed. — If U = X . and suppose V an open neighborhood of x. (4. Now the family {(X − U ) ∩ W x | x ∈ X − U } is a family of compact subspaces of X whose intersection is empty.2.2. Definition 4. So suppose X is locally compact. (37.17. The third also implies the second. I can apply the previous lemma to get an open neighborhood W of x such that W is compact and contained in V .12.2) The image of i is a dense open set in X unless X is already compact.

.

(5.1. Taxonomy of separation Definition 5. (5.3. T3 1 One says that X is Tychonoff or T3 1 if it is T0 and if any closed subset Z ⊂ X and any point x ∈ Z are separated / 2 2 by a function. and suppose A. B ⊂ X two subspaces of X . respectively.3. then x = y. (5. Exercise 38.1) A and B are separated.3. — Observe that for any space X .5) The sets A and B are said to be precisely separated by a function if there exists a continuous map f : / [0. T2 1 One says that X is Urysohn or T2 1 if any two points are separated by closed neighborhoods.3. Otherwise. — Suppose X a space. 2 2 T3 One says that X is regular Hausdorff or T3 if it is T0 and if any closed subset Z ⊂ X and any point x ∈ Z are / separated by neighborhoods. if x.1.4) The sets A and B are said to be separated by a function if there exists a continuous map f : X such that f |A = 0 and f |B = 1.4. Then a neighborhood of a subset A ⊂ X is a set containing whose interior contains A. Then two points x. respectively. Give an example to show that the converse need not hold.1.5) A and B are precisely separated by a function. Definition 5. .1) The sets A and B are said to be separated if both A ∩ B and A ∩ B are empty. X 5.4.3.4. T2 One says that X is Hausdorff or T2 if any two distinct points are separated by neighborhoods.CHAPTER 5 COUNTABILITY AND SEPARATION 5.1.4. — Suppose X a space.3) A and B are separated by closed neighborhoods. (5. Then we introduce the Trennungsaxiome. (5. y ∈ X are topologically distinguishable if they do not have the same collection of open neighborhoods.1.1. the points are said to be topologically indistinguishable.3) The sets A and B are said to be separated by closed neighborhoods there exists disjoint closed neighborhoods U and V and A and B. and suppose A.1. y ∈ X topologically distinguishable points. / [0. — Suppose X a space. T1 One says that X is T1 if any two distinct points are separated. Definition 5. (5.1.1. It is easy to see that each of these conditions follows from the next. T0 One says that X is Kolmogoroff or T0 if any two distinct points are topologically distinguishable.1.1. Definition 5.1.2) The sets A and B are said to be separated by neighborhoods if there exists disjoint neighborhoods U and V and A and B. 1] such that A = f −1 {0} and B = f −1 {1}. 1] (5. — Suppose X a space.4. B ⊂ X .5.3.1. — Suppose X a space.1.1.4.2) A and B are separated by neighborhoods. (5. (5. (5.1.4) A and B are separated by a function.2.1.

but every / R has the property that f (a) = f (b ). due to A. and any > 0. Hence the closures of every two open sets intersect. T6 One says that X is perfectly normal Hausdorff or T6 if it is T1 and if any two closed sets are precisely separated by a function. and let X2 = n odd Tn . topological indistinguishability is an equivalence relation on the points of X . This is frequently done in analysis. 3 1 .1. COUNTABILITY AND SEPARATION T4 One says that X is normal Hausdorff or T4 if it is T1 and if any two closed subsets can be separated by neighborhoods. y) ∈ R2 | (x − n)2 + y 2 = tk }. 2 2 . 2 2 Example 5. y) ∈ X1 consists of all but a finite number of points of {(z. (2) a neighborhood of (n.1) (5. 1). For every even integer n.4) X is T1 . Raha. Lemma 5.3) (5.8.7.9.1.1.9. and add in sets of the form { p} ∪ (U ∩ P ).1. tk ) consists of all but finitely many elements of {(x. 4. 6}.9. (3) a neighborhood of a point (n. n The cofinite topology on an infinite set is T1 but not T2 . 3. R) of all maps R space is T3 1 but not T4 . 1.2) (5.5) Let H be the upper half-plane {(x. — The following are equivalent for a space X .1.4) You have already given an example of a space that is not T0 . This (5. but it’s an absurd space 2 that just makes you sorry you ever tried to contemplate it.9. y) ∈ Q2 | y ≥ 0}. This is a T2 space that is not T2 1 . x − θ + ). and any two topologially distinguishable points are separated. This is a topology on H . y) ∈ X . the complement of a neighborhood of p = (0.2) (5. set Tn := k≥1 {(x. then any T j space is also Ti . if i ≤ j . y) | n − 1 < z < n + 1} ∩ (Tn−1 ∪ Tn ).1) (5.3) (5.1. y) form a base for this topology. Topologize this set in the following manner.6.9.7. Let’s call it the Raha space. (5.32 CHAPTER 5. the sets N (x. (5.x − y θ + ∩Q . The Sierpi´ ski space Σ0 is T0 but not T1 . and contemplate the topology generated by these sets.1. Let X = {(x. 0). Topologize this set in the following manner: for any point (x.7. (5. set Tn := {n} × (−1.6) The Tychonoff corkscrew is the standard example of a space that is T3 but not T3 1 . and let X1 = n even Tn . This is a space that is T3 . — For any i. y) = {(x. continuous map f : X 32 / R with the topology of pointwise convergence. The quotient space under this equivalence relation is the Kolmogoroff quotient KX of X . This space is automatically Kolmogoroff. y) contains the union of four strips of slope ±θ extending from y y y y the intervals (x + θ − . and fix an irrational number θ. tk ) are isolated. (4) a neighborhood of a is a set Uc containing a and all points of X1 ∪X2 with x-coordinate greater than a number c. Points of X are closed.1.1. 2.7. T5 One says that X is completely normal Hausdorff or T5 if it is T1 and if any two separated sets are separated by neighborhoods. and it is not difficult to see that it is T2 1 .1. let N (x.9. y) ∈ R2 | y > 0}. where p = (x. However.x + y θ + ∪ x− y θ − . Every finite set in X is closed. — Let us convince ourselves that these conditions are distinct. j ∈ {0. y) ∈ 2 R2 | (x − n)2 + y 2 = tk }. y) ∈ R2 | y ≥ 0}.9. So H is not T3 . so it is not T 1 . 2 1 . Theorem 5. x + θ + ) and (x − θ − . and U is an open neighborhood of p in the topology on R2 . Begin with all the open sets in the set P = {(x. X is T0 . y)} ∪ x+ y θ − . 0) intersects the 2 closure of every neighborhood of p. For every odd integer n. Topologize X so that: (1) every point of X2 except the points (n. The closure of each basic neighborhood N (x. Here’s one that’s slightly simpler (though still pretty absurd). Now let (tk )k≥1 be an increasing sequence of positive real numbers 2 converging to 1. (5. — For any space X .1. 5. B. (5) a neighborhood of b is a set Vd containing b and all points of X1 ∪ X2 with x-coordinate less than a number d . 5.1. b } ∪ n∈Z Tn .1.7) Consider the space Map(R.7.1.1.9. Now let X = {a.

— There is another condition. (So you walk the plank . (39. Definition 5. AND THE TIETZE EXTENSION THEOREM 33 (5. let f (x) = inf{t | x ∈ F t }. Then we have the variant on the particular point topology. because the space P − {(ω.2) the sets F t cover X . 5.1.2. Exercise 39. called sobriety. to memorize exotic counterexamples.2) Every nonempty open set of X is dense.8) The Tychonoff plank is the space P := (ω + 1) × (ω1 + 1).2. ω1 )} is not normal. and every sober space is Kolmogoroff. that is better suited to applications in algebraic geometry.5. The point is to realize that there are examples that prevent the reversal of any of the implications T j ⇒ Ti for i ≤ j .2. which is called the Fort space topology. {F t | t ∈ D and t > s }. However. . — A normal T1 space is normal Hausdorff.10. Then it is simple to verify that X is T5 .1) X is irreducible.1. — A space X is said to be normal if disjoint closed sets have disjoint neighborhoods. Get it? It’s clever. — Suppose X and Y sober spaces. or T4 . ⊂) are isomorphic as posets. Show that if ( p(X ).2. It is not.4) The interior of every proper closed subset is empty.9.1. and separating the “midrange” 2 axioms from the “hard” separation axioms T4 − T6 . and suppose that for each element t ∈ D. .3.2.) (5.12.11. Lemma 5. — A space X is said to be sober if every irreducible closed subspace Z ⊂ X has a unique generic point. NORMALITY. there is a subset F t of a set X such that (5. and hence it is T4 . (39.3. then X and Y are homeomorphic. and consider the posets ( p(X ). contrary to what might have been your expectation.2. — A point of a space X is said to be generic if it is dense in X .1. 5.2. in each case the example listed is the simplest one I could find. Urysohn’s lemma. so we do not include it in our definition. as is the difference between T3 1 and T3 . Normality. — An exercise in set theory (not to be handed in).3) X cannot be written as the union of two proper nonempty closed subsets. This suggests something interesting: the differences between T2 1 and T2 and 2 T3 is pretty small. Definition 5.1. {F t | t ∈ D and t < s} . Also. and suppose x ∈ X . 5.1. of course. however T5 . and the Tietze extension theorem Definition 5. — Every Hausdorff space is sober. The results of this section do not require the T1 property. — A space X is said to be irreducible (or hyperconnected) if no two nonempty open subsets of X are disjoint. — Suppose D ⊂ R>0 a dense subset. Then for any element s ∈ R.14.1. the difference between T4 and T3 1 is pretty 2 2 significant. ⊂) and ( p(Y ). but there is no comparison between sobriety and T1 . It is compact and Hausdorff.13. ⊂).3. Exercise 40. Definition 5.1. (39.2. and a short argument shows it cannot be T6 . URYSOHN’S LEMMA. Then for any x ∈ X . The closed sets are declared to be those sets V such that either V is finite or x ∈ V . and (5.9. one has {x ∈ X | f (x) < s} = and {x ∈ X | f (x) ≤ s} = Proof. then F t ⊂ F s .9) Suppose X an uncountable set.1) if t < s . — Show that the following are equivalent for a space X . 5. ⊂) and ( p(Y ). (39. This shows us two relatively bright lines separating the “easy” separation axioms T0 −T1 from the “midrange” separation axioms T2 −T3 1 . The moral of the story is not.

f |A = 0 and f |B = 1.2. s ] are open and closed (respectively) for every s ∈ R. where p. Theorem 5. This completes the proof. (−∞. (−∞. s ). (5. there is an element r ∈ D with s < r < t . (5. Lemma 5.2.4) For 0 < t < 1. — It’s enough to check that the inverse image of the sets (−∞.2.2. — We wish to apply the previous lemma with the set D = {x ∈ Q | x = p2−q . It is clear that the first set is contained in the second. — Suppose now X a space.2. The previous lemma shows that the inverse images of sets of the first kind. then F t ⊂ F s .1) if t < s . continuous map A .2) the sets F t cover X . and suppose D ⊂ R>0 a dense subset. there is an open subset F t ⊂ X such that (5.6 (Tietze extension). Proof.34 CHAPTER 5. By construction.5. Then A / [0. 1] can be extended to a continuous map X / [−1. 1] such that f | = 0 and f | = 1.1) If t > 1. there is a continuous map f : X A B Proof.2. and B are separated by functions. (5.2. and suppose A a closed subspace of X .5. If for each element t ∈ D. — Suppose X is a normal space. COUNTABILITY AND SEPARATION Lemma 5. then the function f defined by f (x) = inf{t | x ∈ F t } is continuous. q ∈ N}. We define our sets F t for t ∈ D thus. It is plain to see that the conditions of the previous lemma apply. then let F t = X . Then any / [−1.e. s) and (−∞. It now follows from the previous lemma that the inverse images of sets of the second kind. (5. 1] . and suppose A and B two disjoint closed subsets of X . and (5. hence open. On the other hand. for any t ∈ D with t > s.5. s]. — Suppose X a normal space. Write t = (2m + 1)2−n and let F t be an open set containing F (2m)2−n whose closure F t ⊂ F(2m+2)2−n (by normality). and it constructs our continuous function f .5 (Urysohn).2. we proceed by recursion. i. are closed (and hence we are done) if the following equality obtains: {F t | t ∈ D and t > s} = {F t | t ∈ D and t > s}.2) Let F1 = X − B.2.3) Let F0 be an open set containing A such that F 0 is disjoint from B. whence F r ⊂ F t .4. are the union of open sets.5.3.3..

5. Theorem 6. Example 6. Additionally. We show first that X is Lindelöf. . This set is dense. — Give an example demonstrating that an uncountable product of second-countable spaces need not be first-countable. First and second countability Definition 6. Exercise 43. Then X is said to satisfy the first axiom of countability or to be first-countable if for every point x ∈ X there is a countable fundamental system of neighborhoods of x. and so for some j . if possible. This defines a countable subset := {Uj1 .7. any continuous open image of a second-countable space is second-countable. Example 6. whence some Uj ∈ contains x. Simply select a point x j ∈ B j for every j ∈ N... . Show that for any subset A ⊂ X . Exercise 42. for any convergent sequence xn → x. for any point x ∈ X .4. the cardinality of the continuum). Showing that X is separable is even easier. as is any countable product of R with itself. and the ordinal ω1 + 1 satisfies neither axiom of countability. .1. — The real line R is second countable. Lemma 6.. Proof. we have x ∈ B j ⊂ U . For every natural number j . Hence j ∈ J .e. Exercise 41. Indeed. Lemma 6. (Why?) . . — Suppose X a space.3. the sequence f (xn ) converges to f (x).2. — Suppose X a first-countable space. The space X is said to satisfy the second axiom of countability or to be second-countable if the topology on X has a countable base. — Second-countable spaces are both separable and Lindelöf. — Suppose = {B1 .1. as is any countable product of first-countable (resp. secondcountable). — The second axiom of countability implies the first.1. show that the set 1 = #R (i. . A space X is said to be Lindelöf if every open cover has a countable subcover.CHAPTER 6 METRIZATION THEOREMS 6. — Suppose X a second-countable space. an element U ∈ contains x. p(X ) has cardinality not greater than Definition 6. } ⊂ indexed by a subset J ⊂ N. Finally. show that X is sequentially compact if and only if it is countably compact. } a countable base for a space X .1. — The ordinal ω1 (with the order topology) satisfies the first but not the second axiom of countability. B2 . an element Uj ∈ containing B j . . Show also that a map f : X topological space Y is continuous if and only if. second-countable) spaces. — A space X is said to be separable if it has a countable dense subset. second-countable) space is first-countable (resp.1. Uj2 . I claim that this is a subcover. select.6.1.1.1. — Any subspace of a first-countable (respectively. a point x ∈ X is contained in / Y to a the closure A if and only if there is a sequence of points of A converging to x. Suppose an open covering of X .1.

— Show that any locally compact Hausdorff space is Tychonoff.3) The evaluation map E is injective if and only if for any two distinct point x. .2) The evaluation map E is open if for any closed subset A ⊂ X and any point x ∈ X − A. (6. .2. — The product of two Lindelöf spaces need not be Lindelöf.2.2. Similarly. which is not Lindelöf. there exists an element f ∈ such that f (x) = f (y). U2 . there exists an element f ∈ such that f (x) ∈ f (A). Lemma 6. Proof. . / f ∈ Yf . Now we form a cover ∪ ∪ {X − (A ∪ B)} of A and of X . y ∈ X .2. (6. — The following are equivalent for a space X . 1]N . s ) of natural numbers. Since X is regular.1. and thus the family of open sets whose closures do not intersect B cover A.2.3 (Urysohn metrization).V2 . the family of open sets whose closures do not intersect A cover B.1) The evaluation map is a continuous map E : X (6. Theorem 6.8 (Tychonoff). It now follows from the Lindelöf property that there are countable subcovers {U1 . and (3) X is Lindelöf. .2. — A space is a Tychonoff space if and only if it is homeomorphic to a subspace of a cube [0. (6.2. .2. One verifies immediately that Ur ∩V s = ∅ for every pair (r. .3. — Suppose X a space.3. (2) X is separable. indeed. — Suppose X a metrizable space.9. } ⊂ of B. for any point a ∈ A there is a neighborhood of a whose closure fails to intersect B. — Prove the following Theorem 6. separable. METRIZATION THEOREMS Exercise 44.2.1.2. Urysohn’s metrization theorem Exercise 46. — Every T3 Lindelöf space is T4 .3) X is separable and metrizable.36 CHAPTER 6. Exercise 45. 1]A for some set A. } ⊂ {V1 . Now set Ur := Ur − m≤r Vm and V s := V s − m≤n Um for every natural number r and s. respectively.2. and suppose a family of continuous maps f : X in possibly many different spaces). and so U= r ≥1 Ur and V= s ≥1 Vs are disjoint open neighborhoods of A and B.2.1 (Embedding). Check that the Sorgenfrey line is first-countable. Similarly. Show that the following conditions are equivalent on X : (1) X satisfies the second axiom of countability. the product of the Sorgenfrey line with itself is the Sorgenfrey plane.1) X is second countable and T3 . but not second-countable. — Suppose A and B disjoint closed subsets of X .3. and Lindelöf. Example 6.2 (Embedding). 6. Begin by proving the following / Y f (valued Lemma 6. a subspace of a Lindelöf space need not be Lindelöf. .2) X is homeomorphic to a subspace of the Hilbert cube [0. (6. / (6.2.

(6. The Bing–Nagata–Smirnov metrization theorem Definition 6. choose.1. if and only if it is sequentially compact. and it remains to show that it has the desired properties. Since X is T3 and Lindelöf.3. (6. (Why?) Since (U ..3. It is now an immediate consequence that a metrizable space has a σ-discrete base.V ) ∈ . it is enough to construct a countable family of / [0.V ) does the job.6.2. 1] such that for any closed subset A ⊂ X and any point x ∈ X − A. — First. a point x ∈ X is a member of the closure A if and only if there is a sequence of points in A that converge to A.2. := {(U . observe that the final condition implies the penultimate condition. Lemma 6.6) X is paracompact.V ) ∈ }.4) A subspace A ⊂ X is compact if and only if it is countably compact.6.6. Theorem 6. (Check this!) Hence it remains to show that a T3 space that satisfies the second axiom of countability is homeomorphic to a subspace of the Hilbert cube. the sequence f (xn ) → f (x).3. every open cover of X has a locally finite open refinement. — A T3 space whose topology has a σ-locally finite base is metrizable.3. This is a countable set of functions. and set × | U ⊂ V }.2) X is T3 .V ) ∈ . hence so is any subspace. function f : X n in X .2 (Bing–Nagata–Smirnov metrization). (6. / Y is continuous if and only if. there exists an continuous functions X element f ∈ such that f (x) ∈ f (A). (6. and write up a complete proof. Lemma 6. Suppose A ⊂ X a closed subset.3) X is T3 . Exercise 47. a Urysohn function f(U .5) X is T6 . Outline of proof. σ-locally discrete) if it can be written as the union of a countable collection of locally finite (resp. (6.3.V ) | (U .1) X is metrizable. 6.6.3. set := { f(U ..3.2. A family of subspaces of a space X is said to be σ-locally finite (respectively. One says that is discrete if every point is contained in a neighborhood that intersects only one element of . for any convergent sequence x → x (6. and there is a σ-locally finite base for X . for any subspace A ⊂ X .3.6. (6. — The following are equivalent for a space X .6. — One question that might naturally arise here is: so? I can show that if a space satisfies some abstract conditions. for any (U . — A family of subspaces of a topological space X is said to be locally finite if every point is contained in a neighborhood that intersects only finitely many elements of . it is also T4 by Tychonoff’s lemma. The first two lemmas here will show that the second condition implies the first. and there is a σ-locally discrete base for X .3.3) For any space Y . and suppose x ∈ X − A. i. We may now choose a basic open V ∈ such that x ∈ U ⊂ X − A and a basic open U∈ such that x ∈ U ⊂ V . locally discrete) families of subspaces of X .3.3. . THE BING–NAGATA–SMIRNOV METRIZATION THEOREM 37 Proof.V ) : X U f |X −V = 1. 6.2) Dually.4. 1] such that f | = 0 and Urysohn’s lemma. (6. the corresponding function f(U .3. — Any open cover of a metrizable space has a σ-discrete open refinement. Lemma 6.e.V ) ∈ One sees that is a countable set. (Why?) Any metrizable space is T3 .3. — A T3 space whose topology has a σ-locally finite base is T4 . — Fill in the details I skipped in the proof of the Urysohn metrization theorem. So.1) A subset U ⊂ X is open if and only if no sequence of X − U converges to a point of U . then it can be given a metric that gives rise to its topology.6. Let’s list some of the purely topological corollaries of the metrizability of a space X . Suppose / a countable base for the topology of X .3.6.3.3.3. and any separable metrizable space is second countable. — It is not hard to see that the Hilbert cube is separable and metrizable.5.3. using / [0. (6. Using the embedding lemma above.

.

ENTR’ACTE .

.

a restriction map ρV ⊂U : (V ) These data are subject to the following conditions.1.4.3. On the other hand. and / (U ). write Y / Y | f is continuous}. Example 7. — Let us begin right away with a motivational example.1. . I could try the same trick with any “target” topological space Y : for any open set U ∈ p(X ).4. — Suppose X a topological space. (7.1. then I may restrict it to a continuous function f |V on any open subset V ⊂ U . the set cts / R | f is continuous}. Then a presheaf on X is the following data: (7. a set (U ). Then we can make a similar defini/ Y | s is continuous and p ◦ s = id }. (U ) := { f : U X Again. (7. set Γ( p)(U ) = Γ(X /Y )(U ) := {s : U We say that Γ(X /Y )(U ) is the set of sections of p over U . Presheaves Example 7.1. and let’s contemplate.1.4. it’s pretty clear that we can restrict continuous maps. / X a continuous map. suppose p : Y tion: for any open set U ∈ p(X ). what if I consider three open sets W ⊂ V ⊂ U in X ? If I restrict a function f on U to a function on V . This defines a restriction map ρV ⊂U : cts (U ) X / cts (V ) . Suppose X is a topological space. X I can. Example 7.A) for any open set U ∈ p(X ). and so we have a similar structure Y X := ( Y (U ))U ∈ p(X ) .4.V ∈ p(X ) with V ⊂ U .2. we get an equality of maps ρW ⊂V ◦ ρV ⊂U = ρW ⊂U . — There was nothing particularly special about the topological space R in the above example.1. and thence to a function on W . X (ρV ⊂U )V ⊂U ∈ p(X ) . I see that I get the same answer as if I had restricted f all the way from U to W .1. of course. the map ρU ⊂U is the identity map on (U ).1.1) For any open set U ∈ p(X ).1.B) for any open sets U .CHAPTER 7 SHEAF THEORY 7. U Definition 7. (U ) := { f : U X It is not difficult to see that if I have a continuous function f on U . We can now attempt to study. then my restriction map is simply the identity. or the restriction maps ρV ⊂U . not the individual sets made up of these: cts cts := ( X (U ))U ∈ X cts (U ) X p(X ) . — More interestingly. for any open set U ∈ p(X ). That is. but the entire system p(X ) (ρV ⊂U )V ⊂U ∈ . be rather silly in how I restrict: I might have taken V = U .

8. and to any “relation” between two objects (in this case set inclusion).4.42 CHAPTER 7.1. that Γ(∇)(S 1 ) is a two-point set also. but you can expect functors to appear again in your mathematics education! 7. / S 1 given by ξ  / ξ 2 (with respect to complex multiplication. These data. and to any open sets U . the objects cts ) are presheaves on X . If.1. — One of the most challenging aspects of modern mathematics for beginners is to maintain a clear idea which of a list of bullet points are data.9. the value of Γ(∇) is more complicated as well. This too is a local homeomorphism. one has the restriction maps ρV ⊂U : (V ) to specify them as part of the information of a presheaf. For an open set U ∈ p(X ). consistent way to extract square roots of complex numbers. SHEAF THEORY (7. consider the folding map ∇ : S 1 S 1 more interesting: to any interval (a. U is the disjoint union of two intervals. just as in the previous example. When you have these data. and these data are subject to the axioms ρU ⊂U = id. with these conditions. but observe that the presheaf Γ( p) has no global sections. when the maps go in the opposite direction. an element s ∈ (U ) is sometimes called a section of 7. . for example. Example 7. For any set S.9. It isn’t enough to say that there are suitable maps out there. b ) it assigns a two-point set.7. First.1.3) But now consider the map p : S 1 thinking of S 1 ⊂ C). — Our definition is general enough to allow for some “degenerate” examples of presheaves. the data of a presheaf is a pair := ( (U ))U ∈ p(X ) . We won’t say more about this here. the corresponding map has to be the identity. one may form the constant presheaf S at S.2) Now if p is not a homeomorphism. the object Γ(Y /X ) is a presheaf on X . for any continuous map p : Y contemplate this example in a few particular cases.V ∈ p(X ) with V ⊂ U the identity map on S. for example). then the story becomes / S 1 . sometimes (as here). Example 7. Γ( p)(S 1 ) = ∅. which assigns to any open set U the set S. are collectively called a functor in general. These conditions always amount to the same thing: when the relation is trivial (U ⊂ U .1) Suppose p is a homeomorphism.6. — Likewise. Γ(∇)(V ) has cardinal 2n . and composing relations (W ⊂ V ⊂ U ) gives rise to compositions of maps. (U ) / (W ). (7. the map goes in the opposite direction. there is exactly one section of p over U . One way to think of this is to think that there is no single. we have ρW ⊂V ◦ ρV ⊂U = ρW ⊂U : In other words.V .1.W ∈ p(X ) with W ⊂ V ⊂ U . Γ(Y /X ) is the constant presheaf on one point. In the case of a presheaf on a space X .. When the maps go in the same direction as the relation. Note. — This sort of pattern appears all the time: to any “object” of some collection (in this case p(X )). Γ(∇)(U ) is a four point set. and ρW ⊂V ◦ ρV ⊂U = ρW ⊂U .1. For any open set U ∈ p(X ). but a local homeomorphism (so that every point of Y in contained in a neighborhood V such that the restriction of p to V is open and injective). over U . And in general. Sometimes that map goes in the same direction as your relation. Then Γ(Y /X ) is not so very interesting. it’s a little more subtle than anything we’ve seen so far: the data are the choice of all the sets (U ) and the choice of all / (U ). in this case Γ(∇) is somewhat a lot more interesting. But the presheaf Γ( p) has shown us that there is something nontrivial about our map p. you assign a set.1. Y X (and hence the object / X . i.1. It’s not hard to see that.5. the functor is said to be contravariant. and which are conditions.9. Thus presheaves are particular kinds of functors.1. This is quite striking: the fiber over each point of p is two points. But if the open set is more complicated.1. then the functor is said to be covariant. where n is the number of connected components of V . in particular. Let’s (7.e.2) For any three open sets U . in this case. then you want to impose some conditions. (ρV ⊂U )V ⊂U ∈ p(X ) .9. — It’s not difficult to see that for any topological space X . you assign a map between the corresponding sets. X Example 7. (7.

Suppose I attempted to define a presheaf h A the other way around from our representable friend: h A(U ) := ∅ if U ⊂ A else. a map (called the component of φ) φU : (U ) / (U ). ring.7. Exercise 48. one sees that one cannot make a consistent choice of a branch of the logarithm: there are no global sections of Γ(exp)! Example 7. we can contemplate the set an / R | f is analytic}.1. Then for any open set U ⊂ Rn . — Suppose S is a set. — Suppose X a space. vector space. Then the skyscraper presheaf at x with value S is defined by the rule S x (U ) := S if x ∈ U else.. however. If all the sets (U ) are equipped with a group structure. consequently. there are countably infinitely many elements of Γ(exp)(U ). a very powerful way) of “representing” open sets A ∈ p(X ) as presheaves.). . there is a neighborhood of x in U such that the Taylor series of f at x converges to f . and all the restriction maps ρV ⊂U are group homomorphisms. — We can also ask for additional structure on a presheaf . 7. What goes wrong here? Why does this not define a presheaf? Example 7.12.. algebra. if all the sets (U ) are equipped with an abelian group (respectively.1..1. / is the following data: for any open set .1.) structure. then we might write log(z) = log |r | + i θ. and suppose A ∈ p(X ) a particular fixed open set. A ∈ p(X ) an open subset. Similarly.1. . however. This is surely a presheaf as well. — A morphism or simply map of presheaves φ : U ∈ p(X ). In this case.) homomorphisms. Definition 7.13. algebra. and we see immediately that over a sufficiently small open neighborhood U of any point in C× . then the presheaf is said to be a presheaf in abelian groups (resp. algebras. there is necessarily / h (V ).1. an moreover. vector space.11.9.4) Here’s yet another example in the same vein: consider the exponential map exp : C We can contemplate the presheaf Γ(exp).. this is ambiguous. we can see that X is a sheaf of rings (even of R-algebras). the presheaves hA are known as representable presheaves.. and all the restriction maps ρV ⊂U are group (resp. Example 7. and product of any two analytic functions is analytic. the restriction maps are forced on us: for any open sets V . (U ) = ω (U ) := { f : U X (Recall that a function f on an open set U is analytic if it is infinitely differentiable and. I can add 2kπ to θ. U ∈ p(X ) with V ⊂ U . The key point here is that both the sets (U ) and the maps ρV ⊂U carry this added structure. since the sum. for any point x ∈ U . We have an associated presheaf hA defined by the rule hA(U ) := ∅ if U ⊂ A else. vector spaces. PRESHEAVES 43 / C× := C − {0} ..10. rings. we’ll have to put some only one map hA(U ) A effort into describing the restriction maps! This gives us a way (in fact.14. these correspond to the branches of the logarithm: if we write a complex number z = r exp(iθ). Observe that we cannot expect this in general: ordinarily. then the presheaf is said to be a presheaf in groups. and suppose X a space with a distinguished point x ∈ X .. — Suppose X a space. ring. — Suppose X = Rn with the euclidean topology.. . Once again. difference.1. (7. or presheaves represented by A ∈ p(X ).

write MorX ( . and any presheaves / . there is a natural This word natural gets thrown around a lot. of presheaves and . — Show that for any space X .1. X U  / Mor (h . show that in this with j induces a map − ◦ j : MorX (hU .V ∈ p(X ) with V ⊂ U . (How?) For an extra challenge. Definition 7. the following diagram of sets commutes: (U ) ρU ⊂V ΦU / Mor (h . there is a unique /h . for any section s ∈ (U ). to the presheaf Γ(Y × X /X ) of local sections of the projection map Y × X . ρV ⊂U  (V ) φV that is. we have our unique morphism of presheaves j : hV U / Mor (h .ψ: / . set (ψ ◦ φ)U = ψU ◦ φU . show that a morphism of presheaves φ : U ∈ p(X ).15. SHEAF THEORY subject to the following condition: for any open sets U . / . In this case. — For any space X . ). we can form the composite ψ ◦ φ : / / and ψ : Given morphisms of presheaves φ : in the following manner: for any open set U ∈ p(X ). / is said to be an isomorphism if there exists a morphism of sheaves A morphism of presheaves φ : / such that both ψ ◦ φ = id and φ ◦ φ = id. the presheaf X of local Y -valued functions is isomorphic /X . / .) Notation 7. and χ : / . ) for the set of all morphisms Exercise 50. ψ: Exercise 49. — We say that a morphism / (U ) is injective. (We’ll need to be a little more cautious when we define surthe corresponding map (U ) jectivity. / of presheaves is injective if for every open set U ∈ p(X ). Composition open sets U . ). X V −◦ j ) ). morphism of presheaves hV U Exercise 51.1. we can say exactly what we mean by it: for any pair of / h . = on X . — (49. any open set U ∈ p(X ). and any presheaf bijection ΦU : (U ) ∼ MorX (hU . / is an isomorphism if and only if. ) X V case. and any pair of open sets U .1) Check that our description of the composite of two morphisms of presheaves is indeed again a morphism of presheaves. the components φU are all bijections. / This defines a morphism of presheaves as desired. for any open set (49. then (49.V ∈ p(X ) with V ⊂ U . — Show that for any two spaces X and Y .V ∈ p(X ) with V ⊂ U . — Show that for any space X . the following diagram commutes: (U ) ρV ⊂U φU / / (U )  (V ).3) Finally.2) Verify that if we have three morphisms of presheaves φ : one has (χ ◦ ψ) ◦ φ = χ ◦ (ψ ◦ φ).44 CHAPTER 7.16.  (V ) ΦV Y Exercise 52. one has ρV ⊂U (φU (s)) = φV (ρV ⊂U (s )).

. s ) and (V . For any point an x ∈ X . Hence the stalk Γ(Y /X ) x is simply the fiber p −1 (x) again.1) β. For any open set U .3. — Suppose p : Y x −1 is the fiber p (x)? Prove or give a counterexample. It is easy to see that such a map must be a local subsets Vα ⊂ Y such that p|Vα : Vα homeomorphism. we can compute the stalk Γ(Y /X ) x easily: there exists an open neighborhood U of x whose inverse image is homeomorphic (via p) to a disjoint union of copies of U . For any two elements (U . x2 . But now it’s easy to see that a section of Γ(Y /X ) over this open set U is completely determined by where it sends out point x. and any cover {Uα }α∈Λ of U .2. . — Suppose X = Rn .2. s ∈ (U )}. we say that (U . — Suppose a presheaf on a space X .7. so the equivalence relation doesn’t change anything from this point on as the neighborhoods get smaller. For any point x ∈ X .1. xn }. — Suppose X a space. consider the set (U ) = {(U .e.2. we’ll be in the same situation. and let us consider the presheaf X of local continuous functions on X valued in Y . and suppose a presheaf on X .4. sometimes denoted R{x1 .γ ∈Λ α∈Λ .3. and is dented s x . . . f | =f | : V ∩W V V ∩W W V ∩W then there exists a unique continuous map f : X / Y such that for any V ∈ /Y . x∈U ∈ p(X ) On this set we may impose an equivalence relation ∼ in the following manner.3. What is the stalk of this presheaf at a point x ∈ X ? It’s difficult to say Y in familiar language. Then for any point x ∈ X . f |V = fV : V .3.3. we can contemplate the following diagram: / / (Uβ ∩ Uγ ).2. x :=  x∈U ∈ p(X ) The equivalence class of a section s under this equivalence relation is called the germ of s . i. 7. a surjective continuous map with the property that Example 7. s) | x ∈ U ∈ p(X ). an Example 7.x are equivalence classes of continuous functions defined in a neighborhood of x. where we declare two such functions to be equivalent when there is a neighborhood of x on which they agree. — To motivate the sheaf axiom. SHEAVES 45 7. s) ∼ (V .x is the set of all power series at x with a positive radius of convergence.2.2. t ) if and only if there exists an open neighborhood W ⊂ U ∩ V of x such that ρW ⊂U (s) = ρW ⊂V (t ). Now define the stalk of at x to be the set     (U ) ∼ . Stalks Definition 7. let’s briefly recall a little lemma we proved in Exercise 18 some time back. but here we have required much more: we have required that we can choose the neighborhoods of various points in a fiber in a consistent manner. the stalk Γ(Y /X ) Exercise 53. For every sub-neighborhood V ⊂ U . — Suppose p : Y every point x ∈ X is contained in an open neighborhood U such that p −1 U is a disjoint union α∈Λ Vα of open / U is a homeomorphism.2. . 7. one has fV : V /Y .2. / X any continuous map. (Uα ) (U ) / (7. and consider the presheaf X of local analytic functions on X .1. The elements of X . Sheaves Example 7. / X a covering space. — Let us consider the case of two topological spaces X and Y .3. These equivalence classes are known as germs of continuous functions.W ∈ . This is a ring. We will undertake a more systematic study of coverings spaces very soon. t ). the stalk X . If X is a space with an open covering . Y Example 7. Is it true that for any point x ∈ X . and if for any V ∈ we are given a continuous function / Y with the property that for any elements V .

— For an space X . called the sheaf of local sections of p. then there is one and only one set s ∈ (U ) who restrictions to V and to W are sV and sW .1) is an equalizer.3.3.2. which we have written as a union U = V ∪ W for some open sets V . (7. A sheaf on a space X is a presheaf that has a powerful local-toglobal property. What does this say? It says that sections glue uniquely. Lemma 7. To illustrate this local-to-global property. ρUβ ∩Uγ ⊂Uβ (sβ ) = ρUβ ∩Uγ ⊂Uγ (sγ ) coincide. — A presheaf cover {Uα }α∈Λ of U .e. the diagram on a space X is said to be a sheaf if. for any open set U ∈ p(X ) and any open (U ) is an equalizer in the sense above. Example 7. any point x ∈ X . — This is a little tricky.3. and any set S. the skyscraper presheaf S x is a sheaf as well. g : / every point x ∈ X . . — Suppose and sheaves on a space X .. g x : x x on stalks coincide (so that f x = g x ). whether the map an injective map whose image consists of exactly those elements that have the same values under the maps / (Uβ ∩ Uγ ) .γ ∈Λ (Uβ ∩ Uγ ) and α∈Λ (Uα ) / β. the associated presheaf / α∈Λ (Uα ) / / β.6. i. Proof.3. called the sheaf of local continuous functions on X with values in Y . In other words. one can see that for any continuous map p : Y Γ(Y /X ) is indeed a sheaf.3. — More generally.1) is an equalizer if and only if the following condition is (U ) satisfied: for every α ∈ Λ there is a section sα ∈ α (Uα ) such that for any pair β. so let’s discuss this.W ⊂ X . — Our first example of this section shows that. (Uα ) / β.3. Then if f .3. Example 7. — Show that for any sheaf .γ ∈Λ (Uβ ∩ Uγ ) (sα )α∈Λ  / ρU ∩U ⊂U (sβ ) β γ β β. respectively. / are two morphisms such that for Lemma 7.9. The first map is the assignment (U ) s  The other two maps are the assignments α∈Λ / α∈Λ (Uα ) α∈Λ / ρU ⊂U (s ) α . γ ∈ Λ.2.3.3. Y the presheaf X is a sheaf. Example 7.8.7. the induced maps f x . Exercise 54. — Suppose a sheaf on a space X .γ ∈Λ / (Uα ) is (U ) We can ask whether the diagram (7.46 CHAPTER 7. indeed. Suppose that U ⊂ X is an open set.γ ∈Λ (sα )α∈Λ  / ρU ∩U ⊂U (sγ ) β γ γ β. then there exists a unique section s ∈ Definition 7. specifying a section of over an open set U ⊂ X is the same as specifying a compatible family of sections of over each element of an open cover of .γ ∈Λ (Uβ ∩ Uγ ).5. consider the case of an open cover with just two open sets. for any two topological spaces X and Y . Give a counterexample to show that this is not true of presheaves. then f = g .4. In other words. the map (U ) / x∈U x is injective. 7. Then the sheaf condition for says that if we are given sections sV ∈ (V ) and sW ∈ (W ) that match up on V ∩ W (so that ρV ∩W ⊂V (sV ) = ρV ∩W ⊂W (sW )). SHEAF THEORY Let us describe these maps in more detail.3. Then (∅) = . / X . the restrictions (U ) such that ρUα ⊂U (s) = sα . — This follows immediately from the previous exercise.

The former map is simply the folding map. are themselves isomorphic.7. Consider also the / S 1 . This situation is quite typical.2. This is false. where X (S) is the S-fold disjoint union of X with itself: := s∈S (S) X = X × Sδ . — Suppose n a positive integer. for any open set U ∈ p(X ) and any section s ∈ (U ).4. there is an obvious map / X whose fibers are precisely the stalks of . In this case. — Show that for any space X and any presheaf a local homeomorphism.e. For any open set U and any section s ∈ (U ). we have ρVx ∩Vy ⊂Vx (t x ) = ρVx ∩Vy ⊂Vy (ty ).. — Suppose X a space and a presheaf on X . 7. recall our two examples of double covers of the circle: we were given two local homeomorphisms ∇ : / S 1 and p : S 1 / S 1 . if and only if. Consider the continuous map pn : S 1 1 (where again we think of S as embedded in C. THE ESPACE ÉTALÉ AND SHEAFIFICATION 47 Proposition 7. for each point x ∈ U there is a pair (V x . The sheaf axiom applied to the open cover {V x | x ∈ U } of U now implies that there is a global section t ∈ (U ) such that t x = ρVx ⊂U (t ). The claim is that Now suppose U ∈ p(X ) any open set. We must have a morphism isomorphism in order to make this conclusion. — Suppose S a set. ´ the finest topology such that for any section s ∈ (U ). — A continuous map p : Y contained in a neighborhood V such that p is open and injective. i. Prove that Γ( p ) and Γ(∇n ) are both sheaves. but the sheaves themselves are not isomorphic. all of whose stalks are iso/ inducing the stalkwise morphic. while Γ(∇) was simply the constant sheaf on a two point set. We saw S1 S1 that. and the second is z  / z 2 . the inverse image σ s−1 (V ) is open in U . but it is extremely important. Exercise 56. then the first of the two must also be injective.1. and consider the component φU : (U ) it is surjective. For every point x ∈ U .3. consisting of an open neighborhood V x ⊂ U of x along with a section t x ∈ (V x ) such that φVx (t x ) = ρVx ⊂U (s). the map on stalks x x is so. Then a morphism φ : / an injection. and it’s one that sheaves are particularly well-adapted to. The espace étalé and sheafification / X is said to be a local homeomorphism if every point y ∈ Y is Definition 7. Consider the set Et( ) := x∈X x .4. s) ∈ x has an inverse image in x . and suppose éspace étalé of S is the fold map ∇(S) : X (S) X S ´ on X . or Proof. suppose s ∈ (U ) a section. such that p ◦ s = id. the corresponding map σ s : U ´ That is. for any two points x. with no global sections. Since φ is injective. the element (U . that their stalks at n-fold folding map ∇n : S 1 S 1 · · · S 1 n any point are isomorphic. — Suppose and sheaves on a space X .3. for every point x ∈ X . And now φ(t ) ∈ (U ) is the unique section whose restriction to any V x is φVx (t x ) = ρVx ⊂U (s ). y ∈ U . But the two sheaves differ dramatically! The point is that there is no morphism of sheaves between Γ(∇) and Γ( p) inducing bijections on the stalks. where we may use complex multiplication). . — Suppose X a space and a presheaf on X . the natural morphism p : Et( ) / X is is the constant presheaf at S on a space X . Indeed. s itself! Warning 7. / (U ).11. This sounds like a minor point. (This is the origin of the term “section. ´ 7.10. the stalk over any point x ∈ S 1 is always the same: simply two points. / S 1 given by z  / zn Exercise 55. Γ( p) was a more subtle object. Then show that the / X .”) ´ is a corresponding map σ s : U x ´ Definition 7. there ´ p : Et( ) / Et( ) .4. we declare a subset V ⊂ Et( ) to be open if and only if. To demonstrate this.4.3. / is an isomorphism. Exercise 57. — Note that what we have not said is that two sheaves and . namely. t x ). — The injectivity statement follows from the fact that if the composite of two maps is injective. x  / s .4. The espace étalé of is the set Et( ) equipped with / Et( ) is continuous.

48

CHAPTER 7. SHEAF THEORY

/ X a local homeomorphism. Then the éspace étalé of the sheaf Γ(Y /X ) is canoniLemma 7.4.4. — Suppose p : Y / Et(Γ(Y /X )) such that the diagram ´ cally homeomorphic over X to Y . That is, there is a unique homeomorphism Y ´ Et(Γ(Y /X )) ?? ?? ? pΓ(Y /X ) ? ??  commutes. ´ Proof. — Suppose s ∈ Et(Γ(Y /X )) a germ of a section near x ∈ X . Then its value w(s) = s(x) at x is well-defined. / Y such that p ◦ w = p ´ This defines a map w : Et(Γ(Y /X )) Γ(Y /X ) . In the other direction, suppose y ∈ Y , and suppose x = p(y). Then there exist open neighborhoods U of x / U ; denote by s : U / V its inverse, and define and V of y such that p restricts to a homeomorphism V v(y) ∈ Γ(Y /X ) x as the germ of the section s. The class v(y) does not depend upon the choice of U or V . This / Et(Γ(Y /X )) to the map w. ´ defines an inverse v : Y It remains to show that w is continuous and open. We leave this to the reader. Definition 7.4.5. — Suppose X a space and a presheaf on X . The sheafification of is the sheaf a := ´ / X . The canonical morphism of presheaves /a ´ Γ(Et( )/X ) of sections of the projection map p : Et( ) / s is called the unit morphism. that assigns to any section s ∈ (U ) the local section x  x Proposition 7.4.6. — For any presheaf all stalks. Proof. — By 7.4.4, the stalks of a turn are the stalks of . Corollary 7.4.7. — For any sheaf on a space X , the natural morphism /a induces an isomorphism on ) / X , which in /Y     p 
w

X

´ correspond under the unit morphism to the fibers of Et( on a space X , the unit morphism /a

is an isomorphism.

Corollary 7.4.8. — There is a bijective correspondence between isomorphism classes of sheaves on X and homeomor/X . phism classes of local homeomorphisms Y Exercise 58. — Recall that on a space X , the presheaf may not be a sheaf. Describe its sheafification.
bdd X

of local bounded continuous functions (valued in R)

Example 7.4.9. — The constant sheaf S at a set S on a space X is the sheafification of the constant presheaf S at S. Observe that the constant sheaf is not really constant: it can take many different values on an open set U ⊂ X . Theorem 7.4.10. — Suppose X a space, suppose a presheaf on X , and suppose / factors uniquely through the unit. In other words, for any morphism / such that the diagram a / ?? ? ??  ??  ??  ?  a commutes. a sheaf on X . Then any morphism / , there exists a unique morphism

/ satisfying the Proof. — Let us show that unicity is guaranteed once the existence of the morphism a / satisfying the diagram above must agree an all diagram above is confirmed. Indeed, any two morphisms a / a is a bijection on all stalks. Now we may employ 7.3.9 to see that the stalks, since the unit morphism / coincide. morphisms a / to get a morphism of sheaves To verify existence, observe that we can simply apply a to the morphism / a ; since the unit morphism / a is an isomorphism, we can compose a / a with the inverse a / to get the desired morphism a / . a

7.5. DIRECT AND INVERSE IMAGES OF SHEAVES

49

7.4.11. — There is a more sophisticated way of talking about all this. The construction of the espace étalé is a functor ´ Et from the category re h(X ) of presheaves on a space X to the category ( /X ) of local homeomorphisms to X . Now ( /X ) is actually equivalent (via the sections sheaf functor Γ) to the category h(X ) of sheaves ´ on X . Now the composite a = Γ ◦ Et is the sheafification functor, which is left adjoint to the inclusion h(X ) ⊂ re h(X ). 7.5. Direct and inverse images of sheaves / Y and h : Z / Y , the fiber product X × Z is the Definition 7.5.1. — For any two continuous maps f : X Y set X ×Y Z := {(x, z) ∈ X × Z | f (x) = h(z)} ⊂ X × Z, equipped with the subspace topology. (Note that the notation here is ambiguous, since it makes no explicit mention / X is called the pullback of the map h. of the maps f and h.) The (continuous) projection X ×Y Z / X × X given by the / Y , contemplate the continuous map ∆ : X Exercise 59. — For any map f : X Y f  / (x, x). Show that f is a local homeomorphism if and only if both f and ∆ are open maps. assignment x f Exercise 60. — Show that the pullback of a local homeomorphism is a local homeomorphism. / Y a continuous map. Definition 7.5.2. — Suppose f : X (7.5.2.1) For any sheaf on X , define the direct image f of as the sheaf that assigns to any open set V ∈ p(Y ) the set f (V ) := ( f −1V ). (7.5.2.2) For any sheaf on Y , we define the inverse image f as the sheaf of sections of the pullback / X of the map p : Et( ) /Y ´ ´ X ×Y Et( ) Example 7.5.3. — Suppose A ⊂ X a subspace of a topological space X . Then for any sheaf on X , if i denotes the inclusion map, the sheaf i on A is denoted |A and is called the restriction of to A. If, in particular, A is an open set, then the restriction |A assigns to any open set U ⊂ A the set (U ). On the other hand, if A = {x}, then |{x} is simply the stalk of at x, regarded as a sheaf on {x}. Exercise 61. — Any sheaf on the one-point space is uniquely determined by a set S. Show that the direct image of / X of a point x of a space X is the skyscraper sheaf S . Show that the inverse such a sheaf along the inclusion x / is the constant sheaf a . image of such a sheaf under the unique map X S

ACT II AN INTRODUCTION TO HOMOTOPY THEORY .

.

If S is empty. — Thus a space X is connected if and only if it is nonempty and every global section of the constant sheaf at S is the inclusion X ×{s }δ ⊂ X × S δ for some s ∈ S. The unit morphism S S induces a canonical map S S then this map is an isomorphism. for instance. If S is the one-point set.4. however. It is the sheaf X of local continuous functions valued in the discrete space S δ . — A space X is said to be connected if for any set S.7.1. This is primarily because it involves finding maps out of your space. The definitions agree in “good cases. Equivalently. Definition 8. — Suppose X a space.1. In particular.5. X is connected if and only if the only continuous / S δ are constant. — The definitions we give here do not coincide with definitions in many other texts.8. Sδ An alternate description of the sheaf a S is the following. it’s much easier to compute this version of π0 than it is to compute other variants.2. the definition is nonconstructive e in the sense that it is not clear from the definition alone that π´t X exists for a given space X . then this map is a bijection unless X is empty. in all the cases with which we will have to deal in the sequel. Connectedness and π0 Warning 8. — Note that this definition of π0 is a little strange. Then π´t X is a set along with a global section u ∈ 0 e π´t X (X ) 0 e π´t X 0 such that for /a S any set S and any global section σ ∈ that σ(u) = σ. . there is no π´t . the obvious map S / (x.1. / / (X ) on global sections. — Note that a space X is connected if and only if π´t X = .1.e.1. s)) is a bijection.1. S (X ). at least once you get used to checking universal properties.CHAPTER 8 COVERING SPACE THEORY 8.. — Note that π´t X is a set with a continuous map X 0 0 e / S δ . 0 However.3.1. its associated sheaf is the constant sheaf S . there exists a unique morphism of sheaves σ : a such e e / (π´t X )δ such that for any set S and any continuous 8. X? ?? ? ??  ??  ?  e (π´t X )δ 0 e 8. 8.1.” i. it may very 0 e well not exist: in the case of Q with the subspace topology.6.1. — For any set S. there are important ways in which our definitions differ from those found elsewhere in the literature. there exists a unique map π´t X / S such that the diagram map X 0 / Sδ .1. In fact. maps X e Definition 8. consider the constant presheaf S on a space X . to any element s ∈ S the global section x  / S (X ) (which assigns 8. 0 Warning 8.

with the subspace topology from the 2 inclusion SLn R ⊂ Rn . c) ∩ A and g −1 {1} = (c. Exercise 64. for then points would be open in Q. so π´t Q does not exist.) 0 0 (63. 1}δ given by the rule consider the map g : A g (x) := 0 1 if x < c. with the subspace topology from the inclu2 sion GLn C ⊂ Cn . Show that there is a bijective correspondence between decompositions V = / Λδ . (Proof: if A ⊂ Q is a connected subspace containing two distinct points a and b . — The construction X  e / π´t X is functorial in the sense that for any continuous map f : X / Y . Now for any space V . the corresponding map V α Show that a space X is connected if and only if it is nonempty and admits no nondegenerate decomposition. and without any serifs!) Can you classify all of them e /Y up to homeomorphism? (Hint: if two spaces X and Y are homeomorphic. Let us call a decomposition V = α∈Λ Vα and continuous maps V α∈Λ Vα nondegenerate if and only if / Λδ is surjective. — Consider the letters of the (English) alphabet A–Z as subspaces of the plane R2 . then the continuous map X 0 0 e π´t X exists if and only if there exists a maximal nondegenerate decomposition of X . Hence there is no maximal nondegenerate decomposition e of Q. if no summand V is empty.2) the sphere S n for any n > 0.1.1) a set S with the discrete topology. (63..10. with the subspace topology from the inclusion 2 GLn R ⊂ Rn . for n > 0. choose an irrational number a < c < b . Now / {0. and (63. — Let us consider Q with the subspace topology. (63. since g −1 {0} = (−∞. then their π´t agree. π´t exists.4) the set SLn R of n × n matrices with real entries of determinant 1.) is a homeomorphism. (In all these cases. ∞) ∩ A.5) the set GLn C of n × n invertible matrices with complex entries.1. The only connected subspaces of Q are points. then so is f : X − {x} 8. i. (63. so there are no tricks here. — Compute π´t of the following spaces. 8.54 CHAPTER 8. the topological disjoint union α∈Λ Vα is the settheoretic disjoint union α∈Λ Vα equipped with the following topology: an open set U ⊂ α∈Λ Vα is open if and only if each intersection U ∩ Vα is open in Vα . the definition implies that there is a corresponding map π´t f : π´t X 0 0 0 0 0 e e e / π´t Y such that the diagram exists a unique map π´t f : π´t X 0 0 0 f X  e (π´t X )δ 0 commutes. equivalently. — Recall that for any collection of spaces {Vα }α∈Λ . /Y  e / (π´t Y )δ 0 e π´t f 0 .3) the set GLn R of n × n invertible matrices with real entries. Indeed. there 0 e e e e e / π´t Y if both π´t X and π´t Y exist. (Consider them in their simplest possible form. e e / π´t X is surjective. and if f : X 0 / Y − { f (x)} for any x ∈ X . or. a decomposition of V is a collection of subspaces Vα ⊂ V — called summands — such that V can be written as α∈Λ Vα .) But it is not the case that Q is the disjoint union of its points. for n > 0. This is continuous. a decomposition of X such 0 that each summand is connected. 0 e e Exercise 63. Conclude that Show that if X is a space such that π´t X exists. else.e. without any thickness of any lines.9. COVERING SPACE THEORY Exercise 62. for n > 0.

The (8. Exercise 67.2. Can you draw a picture of this when n = 2? Definition 8. . z kn ) 1 n is a covering space.3. . — A sheaf on a space X is locally constant if every point of X is contained in an open neighborhood U such that the restriction |U is a constant sheaf. . kn ) of nonnegative integers. The corresponding sheaf assigns to S 1 ⊂ C× to get the covering space R any open set U a logarithm defined on U .2. . The corresponding locally constant sheaves are constant. Exercise 66.5. — A covering space over a topological space B (called the base) is a continuous map p : E such that every point b ∈ B is contained in an open neighborhood U for which there is a homeomorphism γU : p −1 U such that the diagam p −1 U ?? ?? ?? ?? p| p −1 U  commutes. f B /E     p   / X is a covering Example 8. Pull this back to get a covering space on the n-dimensional torus Tn .1) For any space X and any set S. 1).3.2.2. .. . . . . — Check that the pullback E ×B B / B is a covering space. Show that for any n-tuple (k1 . / S 1 given by z  / z n .8. .3.. — Show that a covering space is a local homeomorphism. correspondingly. Exercise 65. — (8.2.3. — A morphism of covering spaces from p : E / E such that the diagram base) is simply a continuous map f : E E ? ?? ?? ?? ?? p  commutes. a sheaf / B is a covering map if and only if the sheaf Γ(E/B) is locally ´ is locally constant if and only if the espace étalé Et( ) is a covering space.2) Consider again the continuous map pn : S 1 1 corresponding sheaf assigns to any open set U ⊂ S the set of n-th root functions defined on U .3) The exponential map exp : C / S 1 given by θ  / e 2πθ −1 . — Consider the subspace (C − {0})×n ⊂ Cn . . . the map / (C − {0})×n p : (C − {0})×n (k1 .2. This is a covering space too.1. — A local homeomorphism p : E constant. / C× := C − {0} is a covering map. . Theorem 8. Now contemplate the fiber over (1. These covering spaces are known as the trivial covering spaces.. zn )  / (z k1 .2. COVERING SPACES AND LOCALLY CONSTANT SHEAVES 55 8.2. the map ∇(S) : X (S) ∼ X × S δ = space.2. Covering spaces and locally constant sheaves Definition 8.kn ) (z1 .2. . B / B of a covering space E / B along any continuous map / B (covering spaces over the same γU /B / U × ( p −1 b )δ / U × ( p −1 b )δ   pr1   U / B to p : E Definition 8. . .4.2. One can pull this example back to (8. ..

COVERING SPACE THEORY 8. for any space B.2. More precisely. 1].3. 1 e Lemma 8..56 CHAPTER 8. 8. To put this more formally. x) is the set of uniform monodromy elements. — Suppose X a space. For any covering space E for E at x is an automorphism of the fiber E x . 1] locally constant sheaf on X . 1] comprised of intervals U on which |U . — We say that a covering space p : E ( ovft /X ) the set of all isomorphism classes of covering spaces of finite type over X . Ex γE φx / Ex  γE  Ex φx / Ex . so we have an automorphism of the stalk identity. — We are now interested in classifying all locally constant sheaves/covering spaces over a connected space e X . for any point x ∈ X .6. Then our equivalence of categories ( /B) h(B) restricts to an equivalence of categories ( /B) ov(B). — Under our bijective correspondence between homeomorphism classes of local homeomorphisms and isomorphism classes of sheaves. For any two uniform monodromy elements γ and γ . — The set π´t (X . and let ov(B) denote the category of covering spaces with base B. Now induct on the size of in order to show that must be constant. we see that locally constant sheaves correspond (bijectively) to covering spaces. Give an example to show that this autmorphism need not be the e / X is finite if the fibers of p are finite. In order to do this. Definition 8. /γ o S(1) = x. x) 1 out of the collection of all covering spaces of finite type over X . and suppose x ∈ X . the étale fundamental group π´t (X . Exercise 68. If π´t X exists.7. 1 . Now if γ is a loop in X — i. 1].2. one may compose them to obtain γ ◦ γ := (γE ◦ γE ) ∈ E∈( ov /X ) ft Aut( x) e The étale fundamental group π´t (X . 1] are constant.e..3. let ( /B) be the category of locally constant sheaves on B. for each t ∈ [0. show that there is a unique isomorphism between γ and the constant sheaf S(t ) at the stalk S(t ) := γ (t ) . A uniform monodromy element at x is an element γ = (γE ) ∈ E∈( ovft /X ) / X . the pullback γ is a constant sheaf.2.3. a continuous map γ : [0. For any Using this.3. The étale fundamental group 8. As we have seen.e. Let us denote by then we say that it is of finite type if π´t E exists. constant sheaves correspond to trivial covering spaces. x) is a group under composition. and the fibers of π´t p : π´t E 0 0 0 0 Definition 8. — Show that the only possible locally constant sheaves on the interval [0. Why doesn’t this argument work for S 1 ?) / X .3. we will begin by constructing. a path such that γ (0) = γ (1) = x ∈ X —. we have specified two isomorphisms: x = S(0) x. suppose X any space. and suppose γ a path in X — i. a monodromy element Aut(E x ) such that for any morphism φ : E / E of covering spaces of finite type.1. (Hint: Suppose a locally constant sheaf on [0. Show that there is a finite open cover of [0. 0 e e e e / π´t X are finite.

= Let e g . the map MorX ( . Deduce that if a universal locally constant sheaf on (X . Then a lift of f .4. if it exists. has the following property: for any locally constant sheaf on X .8.4. — Suppose X a space. suppose y ∈ Y . x). — Suppose g and h two lifts. x) exists. — In the situation above.2. 8. . Lemma 8. Then there is a unique lift γ of this path to E. Then a universal locally constant sheaf on (X . Lemma 8.1. h(y) to E f (y) . Proof. then the set V := g −1 (U × {e g }) ∩ h −1 (U × {e h }) is open in Y .3. — Consider the following commutative diagram of spaces /E  Y in which p : E  /B p f / B is a covering space and Y is connected space.. — Necessity is obvious. This shows that r is continuous. / {0. The assumption guarantees that this is independent of the choice of path. — Suppose X a space. This latter point follows from inspection of the diagram Ex γE φx / Ex  / Ex  γE  Ex  Ex φx γ E γE φx / Ex . / E in the following manner. any germ σ ∈ x . ) x. if it exists. we only have to check that π´t (X . and suppose x ∈ X . then it is unique up to φ x a unique isomorphism. let U be an open neighborhood of f (y) such that p −1 U ∼ U ×E f (y) . x) contains the identity 1 1 (obvious) and that it is closed under composition.4. x) is given as a subset of a group. there is a unique morphism φ : x Exercise 69. 8. define φ(y) = γ (1).4. UNIVERSAL COVERING SPACES 57 e e Proof. and the result follows from the connectedness of Y . define a lift φ : Y y ∈ Y .5. assume that Y is locally path connected.. Homotopy classes of loops and lifts Lemma 8. Universal covering spaces Definition 8. Show that the universal locally constant sheaf ( . For sufficiency. For any point Proof.5. choose a path γ from the basepoint to y.1. — Since π´t (X .5. is a bijection. and one verifies that V ⊂ r −1 ( f (y)). x) is a locally constant sheaf on X along with a germ of a section υ ∈ x such that for any locally constant sheaf and / of sheaves such that φ (υ) = σ.  / φ (υ). and suppose x ∈ X . — Unique path lifting . 1}δ by if g (y) = h(y) if not. Indeed. Then a lift exists if and only if the image of π1 (Y ) in π1 (X ) is contained in the image of π1 (E). e h ∈ E f (y) be the projections of the element g (y). Define a map r : Y r (y) = 1 0 I claim r is continuous. υ) on / (X . is unique.

5. Then X admits a universal covering space. x)-sets. 1}) classes of paths / X .2. x)-sets with finitely many orbits. Write x for the constant map at x. Begin by contemplating the based path space Π x (X ) ⊂ (I . — Suppose X a well connected space.. define (U ) as the quotient Γ(e)(U )/ ∼.5. x) is trivial. there exists a unique covering space map f : X Definition 8. — Suppose X a space. So now we have the corresponding covering space X . x). . — There is a natural candidate for the universal locally constant sheaf on a pointed space (X . e ∈ p (x). Definition 8. — Let us construct the corresponding locally constant sheaf. whose points are those paths γ such that γ (0) = x. x) is trivial. X ). It is the space of homotopy (rel {0.58 CHAPTER 8.. the claim is that its sheafification is are homotopic as maps U × I locally constant. Let f be a representative thereof. To show that X is path connected. 1 . then we are done by 8. 1}.3. let f t (s) = f (s t ). the espace open sets U such that the induced homomorphism π1 (U ) 1 étalé of is a trivial covering space. where we say that two sections s. — Suppose X admits a universal covering space. Theorem 8. it admits a basis comprised of path-connected neighborhoods.5.5.6. — A space X is said to be well connected if it is path connected. To show that π1 (X . evaluation at 1 gives the structure map X check that X is path connected and that π1 (X . and suppose x ∈ X . observe that since X is well connected. consider an equivalence class α of paths.4. 1 e the connected covering spaces correspond to the transitive π´t (X . Eval/ X Consider now the sheaf of sections Γ(e). it is the unique connected pointed covering space (X . Proof. Then a universal covering space over (X .4. COVERING SPACE THEORY 8. it admits a base consisting of path connected / π (X ) is trivial. The result is a presheaf . t ∈ Γ(e)(U ) are equivalent if they / X rel U × {0. To this end. x) is trivial. x) 1 Theorem 8. x) is trivial. Moreover. Now for any real number t . and any point x is contained in a neighborhood U such that the homomorphism / π (X . x) of X such that π1 (X . equipped with the subspace topology. If now we can starting at x. x) is a covering / X and any point / X along with a point x ∈ u −1 (x) such that for any covering space p : E space u : X −1 / E with the property that f (x) = e. for any open set U uation at 1 gives a continuous map e : Π x (X ) of X .5. π1 (U . This gives a path from x to α. Then there is a bijective correspondence between e ( ovft /X ) and the set of isomorphism classes of π´t (X .5.3. Over these open sets. and under this correspondence.

1.CHAPTER 9 COMPUTING THE ÉTALE FUNDAMENTAL GROUP 9. The étale fundamental group of S 1 9. Simple connectedness .3. Seifert-van Kampen Theorem 9.2.

.

Homotopies of maps 10. Paths and π0 10.3.CHAPTER 10 HOMOTOPY THEORY 10. The homotopy-theoretic fundamental group 10. Homotopy invariance of the étale fundamental group .2.4.1.

.

CHAPTER 11 THE COMPARISON THEOREM 11.1. The two fundamental groups .

Sign up to vote on this title
UsefulNot useful