Math 202A: Topology and Analysis I Lecture Notes

Joseph Pagadora
Instructor: Marc Rieffel

Fall 2018
2
Contents

1 Metric Spaces 5
1.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Completion of a Metric Space . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Openness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Introduction to Topology 13
2.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Quotient and Product Topologies . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Special Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Compactness 21
3.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Tychonoff’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Compact and Hausdorff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Compactness for Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Locally Compact Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4 Measure Theory 31
4.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Borel Measures on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Outer Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 From Outer Measure to Measure . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5 Non-Measurable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5 Integration 41
5.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Towards L1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 The Space L1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3
4 CONTENTS
Chapter 1

Metric Spaces

1.1 Fundamentals

Def: Let X be a set. A metric on X is a function d : X × X → [0, ∞) that satisfies:

(a) d(x, y) = d(y, x) for any x, y ∈ X

(b) d(x, y) ≤ d(x, z) + d(z, y) for any x, y, z ∈ X

(c) d(x, y) = 0 iff x = y

If a function d satisfies (a), (b) above, and d(x, x) = 0 for all x ∈ X, then d is a semi-metric.

Ex: On Cn , the following are common metrics:

n
P 1/p
• dp (v, w) = |vj − wj |p for p ≥ 1
j=1

• d∞ (v, w) = sup{|vj − wj | : 1 ≤ j ≤ n}

(Verify that these are metrics.)

Fact: If S ⊆ X, and d is a metric on X, then d is a metric on S.

Def: Let V be a vector space over R or C. A norm on V is a function k·k : V → [0, ∞)

such that:

(a) kcvk = |c| · kvk for c ∈ R or C and v ∈ V

(b) kv + wk ≤ kvk + kwk for v, w ∈ V

(c) kvk = 0 implies v = 0

A function that satisfies only (a) and (b) above is called a seminorm.

5
6 CHAPTER 1. METRIC SPACES

Remark: Any norm k·k on X induces the metric d(x, y) := kx − yk .

Ex: Let V be the space of continuous functions on [0, 1]. Then

kf k∞ = sup{|f (x)| : x ∈ [0, 1]} is a norm on V.
R 1/p
1 p
It can also be shown that kf kp := 0 |f (x)| dx is a norm on V.

Def: Let (X, dx ) and (Y, dy ) be metric spaces. A function f : X → Y is isometric if

dy (f (v), f (w)) = dx (v, w) for all v, w ∈ X.

• Note that all isometries are injective.

Ex: If S ⊆ X, and f : S → X is defined by f (x) = x (inclusion), then f is an isometry.

If f is onto, then f is viewed as an isomorphism between (X, dx ) and (Y, dy ). f −1 is also

an isomorphism.

Def: A function f : X → Y is Lipschitz if there is a constant k ≥ 0 such that

dy (f (x1 ), f (x2 )) ≤ k · dx (x1 , x2 ). The smallest such constant is the Lipschitz constant
for f.

Def: f : X → Y is uniformly continuous if for any  > 0, there exists δ > 0 such that
dy (f (x1 ), f (x2 )) <  whenever dx (x1 , x2 ) < δ.

• It is easy to see that if f is Lipschitz, then it is uniformly continuous.

Def: f : X → Y is continuous at x0 if for any  > 0, there exists δ > 0 such that
dy (f (x), f (x0 )) <  whenever dx (x, x0 ) < δ. We say f is continuous if it is continuous at
every x ∈ X.

Def: A sequence {xn } in X converges to x∗ ∈ X if for any  > 0, there exists N ∈ N such
that for all n ≥ N, we have d(xn , x∗ ) < .

Proposition: A function f : X → Y is continuous iff xn → x implies f (xn ) → f (x).

Proof: Exercise. 
1.2. COMPLETION OF A METRIC SPACE 7

Def: S ⊆ X is dense in X if for any x ∈ X and  > 0, there exists s ∈ S such that
d(x, s) < .

Proposition: Let S be dense in X, and let f : X → Y and g : X → Y be continuous

functions such that f (s) = g(s) for all s ∈ S. Then f = g on X.

Proof: Let x ∈ X \ S, and let  > 0. Then there exists δ > 0 and s ∈ S such
that d(f (x), f (s)) < /2, and d(g(x), g(s)) < /2 for d(x, s) < δ, by continuity and density.
Then
d(f (x), g(x)) ≤ d(f (x), f (s)) + d(g(s), g(x)) < /2 + /2 = ,
since f (s) = g(s). Thus d(f (x), g(x)) = 0, so f (x) = g(x). 

Def: A sequence {xn } is Cauchy if for any  > 0, there exists N ∈ N such that n, m ≥ N
implies d(xn , xm ) < .
A metric space in which every Cauchy sequence converges is complete.

√
Ex: Consider (Q, | · |).√We know there exists a Cauchy sequence converging to 2 ∈ R, but
in this metric space, 2 is not an element, so this sequence does not converge, hence this
metric space is not complete.

1.2 Completion of a Metric Space

Proposition: If f : X → Y is uniformly continuous, and {xn } is Cauchy in X, then

{f (xn )} is Cauchy in Y.

Proof: Exercise. 

Def: Let (X, d) be a metric space. A complete metric space (X, d), together with an
isometric function f : X → X with dense range is a completion of (X, d).

• Remark: Completions are unique up to isomorphism.

Proposition: If ((Y1 , d1 ), f1 ) and ((Y2 , d2 ), f2 ) are completions of (X, d), then there exists
an onto isometry (metric space isomorphism) g : Y1 → Y2 with f2 = g ◦ f1 . This can be
visualized by the following commutative diagram:
8 CHAPTER 1. METRIC SPACES

f1 Y1

X g

f2
Y2

Every metric space has a completion, and the proof will be constructive. The completion
will be defined using equivalence classes of Cauchy sequences. We will need the following
lemmas to support the construction.

Lemma 1: If {sn } and {tn } are Cauchy sequences in X, then the sequence {d(sn , tn )} in
R converges.
Proof: Exercise. Hint: {d(sn , tn )} is a Cauchy sequence in a complete metric space. 

Lemma 2: Let Cau(X) denote the set of all Cauchy sequences in X. Then the relation
{sn } ∼ {tn } iff d(sn , tn ) → 0 is an equivalence relation.

Proof: Reflexivity and symmetry are trivial. Suppose d(sn , rn ) → 0 and d(rn , tn ) → 0.
Then d(sn , tn ) ≤ d(sn , rn ) + d(rn , tn ) for all n ∈ N. The result follows immediately. 

Lemma 3: Let X be the set of all equivalence classes of Cau(X) under the equivalence rela-
tion above. Then d : X → [0, ∞) defined by d({sn }, {tn }) := lim d(sn , tn ) is a metric on X.
n→∞

Proof: First, note that by Lemma 1, d is always defined. Since we are dealing
with equivalence classes, we must show that d is also well-defined. Let ξ, η ∈ X, and let
{xn }, {sn } ∈ ξ, and {yn }, {tn } ∈ η. We have lim d(xn , sn ) = lim d(yn , tn ) = 0. Thus,
d(sn , tn ) ≤ d(sn , xn ) + d(xn , yn ) + d(yn , tn ). For any  > 0, we can find N ∈ N such that
both d(sn , xn ) < /2 and d(yn , tn ) < /2 for n ≥ N. Then |d(sn , tn ) − d(xn , yn )| < . It
follows that d(ξ, η) = lim d(xn , yn ) = lim d(sn , tn ), so that d is indeed well-defined.

Symmetry is trivial. The triangle inequality follows from the proof to Lemma 2. If
d(ξ, η) = 0, then for any {xn } ∈ ξ, {yn } ∈ η, we have lim d(xn , yn ) = 0, so in particular,
{yn } ∈ ξ, hence ξ = η. 

Theorem: Let (X, dx ) and (Y, dy ) be metric spaces with Y complete. If S ⊆ X is dense,
and f : S → Y is uniformly continuous, then there exists a unique continuous extension
f : X → Y of f. In fact, f is uniformly continuous.
1.2. COMPLETION OF A METRIC SPACE 9

Proof: (Existence only) For x ∈ X, choose a Cauchy sequence {sn } in S converg-

ing to x. Then {f (sn )} is Cauchy in Y, so it converges to a point p ∈ Y. Set f (x) := p. We
show that f is well-defined. Indeed, if {tn } ∈ Cau(S) and converges to x, then we have
lim dx (sn , tn ) = 0, implying that lim dy (f (sn ), f (tn )) = 0. Therefore lim dy (f (tn ), p) = 0, so
{f (tn )} converges to p also. It remains to show continuity, which is left as an exercise. 

Theorem: Every metric space (X, d) has a completion.

Proof: As in Lemma 3, (X, d) is a completion of (X, d). We embed X in X by the

isometry ι : X → X defined by ι(x) := [{x, x, x, ...}], where [·] denotes the corresponding
equivalence class. Note that d = d, i.e., d(ι(x), ι(y)) = d(x, y).
X
It remains to show that d has dense range, and that (X, d) is complete.

• Let ξ ∈ X,  > 0, {xn } ∈ ξ. There exists N ∈ N such that n, m ≥ N implies

d(xn , xm ) < . Then d(ι(xN ), ξ) = lim d(xN , xn ) < . Therefore d has dense range
n→∞
by considering ι(xN ).

• Let {ξn } be a Cauchy sequence in X. For each m ∈ N, pick xm ∈ X such that

d(ι(xm ), ξm ) < 1/m. Then {xm } is a Cauchy sequence, and it follows that {ξm }
converges to the equivalence class of {xm }.

Remark on functions:
Denote C([0, 1]) the space of continuous functions on [0, 1]. Consider the metric space
C([0, 1]) induced by the norms k·k∞ or k·kp . This space is not complete. It is easy to come
up with a sequence of continuous functions converging under these norms to a function that
is not continuous.

Vector Space Remark:

Let V be a vector space with norm k·k . Consider V ∞ , the space of all sequences of elements
in V. This is also a vector space. It can be shown that Cau(V ) is a sebspace of V ∞ .
Now let N (V ) denote the set of all Cauchy sequences in V converging to 0. Then N (V ) is
a subspace of Cau(V ). If {vn } and {wn } are equivalent Cauchy sequences, then
kvn − wm k → 0, so {vn − wn } ∈ N (V ). Thus V is in fact the quotient space Cau(V )/N (V ).

Fact: Any two norms k·k1 , k·k2 on a finite dimensional vector space are equivalent, mean-
ing that there are constants c, C > 0 such that c kxk1 ≤ kxk2 ≤ C kxk1 for all x. If a
function is continuous with respect to a particular norm, then it is easily seen that it is
continuous with respect to any equivalent norm.
10 CHAPTER 1. METRIC SPACES

1.3 Openness

Def: If (X, d) is a metric space, the open ball of radius r around x is

Br (x) := {y ∈ X : d(x, y) < r}.

Def: We can rephrase continuity

  as f : X → Y is continuous at x0 if for any  > 0, there
exists δ > 0 such that f Bδ (x0 ) ⊆ B (f (x0 )).

If y ∈ B (f (x0 )) and y = f (x) for some x ∈ X, let 0 =  − d(y, f (x0 )) > 0. Then
B0 (y) ⊆ B (f (x0 )), so there exists δ 0 > 0 such that f Bδ0 (x) ⊆ B (f (x0 )).
   
If x1 ∈ f −1 B (f (x)) , there is an open ball Bδ0 (x) such that Bδ0 (x1 ) ⊆ f −1 B (f (x)) .
 
Thus f −1 B (f (x)) is a union of open balls in X. Similarly, f −1 (B (y)) is a union of open
balls in X.

Def: A subset O ⊆ X (where X is a metric space) is open if it is a union of open balls in

X.

By the arguments above, it can be shown that f : X → Y is continuous iff for any open
ball B (y) ⊆ Y, we have that f −1 (B (y)) is open in X.

Set theoretic facts:

Let f : X → Y be a function, and let {Aα } be a family of subsets of X. Then:
S 
• f −1 −1 (A
S
α Aα = αf α)

T 
• f −1 −1 (A
T
α Aα = αf α)


• f −1 A1 \ A2 ) = f −1 (A1 ) \ f −1 (A2 )
S 
• f
S
α Aα = α f (Aα )
T 
• f α Aα ⊆
T
α f (Aα )

• f (A1 \ A2 ) ⊇ f (A1 ) \ f (A2 )

 
Let O ⊆ Y be open. Then f −1 (O) = f −1 f −1 (U ).
S S
U =
U ⊆O,U open ball U ⊆O,U open ball
Thus, if f : X → Y is continuous, and O is open in Y, then f −1 (O) is open in X.

Theorem: If (X, d) is a metric space, and τd is the collection of all open sets, then:
1.3. OPENNESS 11

(1) If {Oα } is an arbitrary collection of subsets in τd , then α Oα

S
is open.
n
(2) If O1 , . . . , On is a finite collection of subsets in τd , then Oj is open.
T
j=1

(3) X ∈ τd (X is open)

Proof:

(1) Trivial
n
(2) Let x ∈ Oj . For each j, there exists δj > 0 such that Bδj (x) ⊆ Oj . Let
T
j=1
δ = min{δj : 1 ≤ j ≤ n}. Then Bδ (x) ⊆ Oj for each j. This yields the desired result.
S
(3) Let  > 0. Then X = B (x). The result follows from (1).
x∈X

By convention, ∅ ∈ τd (∅ is open).
12 CHAPTER 1. METRIC SPACES
Chapter 2

Introduction to Topology

2.1 Fundamentals

Def: Let X be a set. A topology on X is a collection T ⊆ P(X) that satisfies:

(1) For any arbitrary family {Oα } ⊆ T , we have Oα ∈ T .

S
α
n
(2) For O1 , . . . , On ∈ T , we have Oj ∈ T .
T
j=1

(3) X ∈ T , ∅ ∈ T .

Def:

• If T is a topology on X, then (X, T ) is a topological space. The sets in T are called

open, and the complements of the sets in T are closed.

• For any A ⊆ X, there is a smallest closed set containing A, namely, the intersection of
all closed sets containing A (by DeMorgan’s Laws, the arbitrary intersection of closed
sets is closed). We denote such a set by the closure of A, denoted A.

• If T1 and T2 are topologies on X, and T1 ⊆ T2 , we say that T1 is weaker/coarser

than T2 . T2 is stronger/finer than T1 .

Proposition: For any sets A, B in a topological space, A ∪ B = A ∪ B.

Proof: Exercise. 

Let C be a collection of topologies on X. It can be shown that T is a topology on X.

T
T ∈C
From this fact, it follows that for any non-empty collection S of subsets of X, there is a
smallest topology on X that contains S.

13
14 CHAPTER 2. INTRODUCTION TO TOPOLOGY

Def: Let B ⊆ T . We say that B is a base for T if every element of T is a union of elements
in B.

Proposition: Let X be a set, and let B be a collection of subsets of X. If B has the property
that for any U, V ∈ B, U ∩ V is a union of elements of B, then the collection of unions of ele-
ments of B, then the collection of unions of elements in B is a topology for which B is a base.

Proof: Exercise. 

Def: Let S be any collection of subsets of X such that

S
V = X, and the set of finite
V ∈S
intersections of elements of S is a base for a topology, then S is a sub-base for that
topology.

Ex: The standard metric topology on Rn has the base {Br (x) : r > 0, x ∈ Rn }.
A sub-base for R with the metric topology is the collection of open rays: (−∞, a) and
(b, ∞).
Ex: Given a metric space (X, d), we will always assume that it has a standard topology
whose base consists of the open balls: {Br (x) : r > 0, x ∈ X}.

2.2 Continuity

Def: Let (X, Tx ) and (Y, Ty ) be topological spaces. A function f : X → Y is continuous

for Tx and Ty if for any U ∈ Ty , we have f −1 (U ) ∈ Tx .

• It is clear that any composition of continuous functions on topological spaces is con-

tinuous.

• For any topological space (X, T ), the identity map ι : X → X is continuous.

• The collection of topological spaces with continuous functions between them is a

category.

Def: f : X → Y is homeomorphic if f is continuous, bijective, and has a continuous

inverse.

Proposition: f : (X, Tx ) → (Y, Ty ) is continuous iff for a base or sub-base C for Ty , we

have f −1 (U ) ∈ Tx for U ∈ C.
2.3. QUOTIENT AND PRODUCT TOPOLOGIES 15

Proof: The forward direction is obvious. For the converse, we prove the case for C
a base. The case for C a sub-base follows similarly (add on finite intersections).
Suppose f −1 (U ) ∈ Tx for any U ∈ C. Let V ∈ Ty . Since C is a base, there

is a collection of
open sets B ⊆ C such that V = A. Then f −1 (V ) = f −1 f −1 (A) ∈ Tx ,
S S S
A =
A∈B A∈B A∈B
since each f −1 (A) is open. 

• Let X be a set and let (Yα , Tα ) be a collection of topological spaces, and for each α,
let fα : X → Yα be a function. Then there is a smallest topology on X for which each
fα is continuous, namely, the smallest topology having as sub-base all sets fα−1 (U ),
where U ∈ Tα for all α.

• Let (X, Tx ) be a topological space, and let Y be a set with f : X → Y a function.

What is the strongest topology on Y making f continuous?

– We need that if A ⊆ Y is open, then f −1 (A) ∈ Tx .

Let T := {A ⊆ Y : f −1 (A) ∈ Tx }. Then T is easily seen to be a topology on Y.
This is called the quotient topology on Y for f.

• Note that if y 6∈ f (X), then f −1 ({y}) = ∅, so {y} is open. Also, f −1 ({y}c ) = X, so

{y} is also closed. Therefore, on f (X)c , the quotient topology is discrete. Thus, we
usually require f : X → Y to be onto.

2.3 Quotient and Product Topologies

Let f : X → Y be onto, and define the equivalence relation on X by x1 ∼ x2 iff
f (x1 ) = f (x2 ). Conversely, given any set X, and an equivalence relation on X, let Y be the
set of equivalence classes for ∼ . We often denote Y by X/ ∼ . Thus, given a topology on
X, we get the quotient topology on X/ ∼ .

Given a collection {(Xα , Tα )} of topological spaces and a set Y, and for each α a func-
tion fα : Xα → Y, the strongest topology on Y making all fα continuous is the intersection
of all quotient topologies for each fα . This is called the final topology.

Ex: Let X = [0, 1]. Define the equivalence relation s ∼ t iff s = t, and have 0 ∼ 1.
That is, {0, 1} is an equivalence class.

X/ ∼

X
0 1 {0, 1}
16 CHAPTER 2. INTRODUCTION TO TOPOLOGY

Define f : X → {z ∈ C : |z| = 1} by f (t) = e2πit , for t ∈ [0, 1].

Note that f is continuous but f −1 is not: there is a discontinuity at 1 ∈ C. However, the
corresponding function f : X/ ∼→ {z ∈ C : |z| = 1} is a homeomorphism with the usual
topology from C.

Ex: Let X1 = {x ∈ R2 : kxk2 ≤ 1}, the closed unit disk in R2 . Let X2 be a copy of
X1 . Consider the disjoint union X1 ∪ X2 , and consider the following equivalence relation.
Let x, y be interior points of the same disk. Then x ∼ y iff x = y. If x is a boundary point
of X1 , and y is a boundary point of X2 , then x ∼ y iff they correspond to the same point
in the plane. Similarly as in the previous example, one can visualize that X1 ∪ X2 / ∼ is
homeomorphic to the 2-sphere.

Def: Let X be a set, and let {Yα } be a family of topological spaces. For each α, let
fα : X → Yα be a function. The corresponding weak topology on X is the smallest
topology making each fα continuous.
Fact: The weak topology has as sub-base all sets of the form fα−1 (U ), where U is open in
Yα .

Def: Given (X, T ), and a subset Y ⊆ X, the topology on Y induced by T is the

relative topology, which has {Y ∩ O : O ∈ T } as open sets.

Proposition: If A ⊆ X is closed, and C ⊆ A is closed in the relative topology of A, then

C is closed in X.

Proof: A \ C is open in A. There is an open set O ∈ T such that A \ C = A ∩ O ∈ T , so

C = A ∩ Oc which is closed in X. 

Def: Let {(Xα , Tα } be a collection of topological spaces indexed by A. The product space
Xα := {h : A → Xα | h(α) ∈ Xα , ∀α}.
Q S
is defined by
α∈A α
Xβ → Xα , defined by πα (h) = h(α).
Q
The α-th projection map is πα :
β∈A

Q
Def: The product topology on Xα is the weakest topology making all projections
α∈A
continuous. That is, it is the weak topology with respect to all the projection maps.

Q
Fact: In general, the product topology will have as a base all sets of the form α∈A Uα ,
where Uα ∈ Tα , and also Uα = Xα for all but finitely-many α.
2.4. SPECIAL TOPOLOGICAL SPACES 17

Proposition: Consider fα : X → Yα for α ∈ A. Let Tx be the corresponding weak

topology on X. Let (Z, Tz ) be a topological space, and let g : Z → X. Then g is continuous
iff fα ◦ g is continuous for all α.

Proof: Suppose fα ◦ g is continuous for all α. It suffices to check on the sub-base.

Let O ∈ Tα . Then g −1 (fα−1 (O)) = (fα ◦ g)−1 (O) is open, hence g is continuous.
Conversely, if g is continuous, then (fα ◦ g)−1 (O) = g −1 (fα−1 (O)) is open since fα−1 (O) is
open in Tx , thus (fα ◦ g) is continuous. 

2.4 Special Topological Spaces

Def:

• A topological space X is Hausdorff if for any two distinct points x1 , x2 ∈ X, there

are disjoint open sets O1 and O2 such that x1 ∈ O1 , and x2 ∈ O2 .

• X is normal if for any two disjoint closed sets C1 , C2 , there are disjoint open sets
O1 , O2 such that C1 ⊆ O1 , and C2 ⊆ O2 .

• A topological space is metrizable if its topology comes from a metric, i.e., its base
consists of open balls from some metric.

Clearly, every metrizable space with more than one element is Hausdorff. Suppose the
topology is induced by a metric d, and take two distinct points x, y. Let r = d(x, y). Then
the open balls Br/3 (x) and Br/3 (y) are disjoint. More is true:

Proposition: Every metrizable topological space is normal.

Proof: It suffices to consider a metric space (X, d). Let C1 , C2 be disjoint closed
subsets of X. For each x ∈ C1 choose x > 0 such that Bx (x) ⊆ C2c , and for each y ∈ C2 ,
choose y > 0 such that By (y) ⊆ C1c .
Let O1 = Bx /3 (x), and O2 = By /3 (y). Clearly, O1 and O2 are open.
S S
x∈C1 y∈C2
Since C1 ⊆ C2c , and C2 ⊆ C1c , we have C1 ⊆ O1 and C2 ⊆ O2 . Toward contradiction,
suppose z ∈ O1 ∩ O2 . Then there are x0 ∈ C1 and y 0 ∈ C2 such that z ∈ Bx0 /3 (x0 ) and
z ∈ By0 /3 (y 0 ). But then

x0 y0 2
d(x0 , y 0 ) ≤ d(x0 , z) + d(z, y 0 ) < + ≤ max(x0 , y0 ).
3 3 3
This implies that C1 ∩ C2 6= ∅, a contradiction, hence O1 ∩ O2 = ∅, as desired. 

We now develop two important results: Urysohn’s Lemma and Tietze’s Theorem. Given
topological spaces, there may not be many continuous functions between them, but in the
18 CHAPTER 2. INTRODUCTION TO TOPOLOGY

case of normal spaces, these results demonstrate their abundance.

Urysohn’s Lemma: Let (X, T ) be a normal topological space. Then for any two disjoint
closed sets C0 , C1 ⊆ X, there exists a continuous function f : X → R such that f (x) = 0
for x ∈ C0 , and f (x) = 1 for x ∈ C1 .

Urysohn’s Lemma is easy for metric spaces. Let d denote the metric, and let A, B be disjoint
closed subsets. For any non-empty subset E, we can define ρE (x) = inf{d(x, y) : y ∈ E},
which can be shown to be continuous. Furthermore, ρE (x) = 0 iff x ∈ E.
ρA (x)
Define f (x) = . One can easily check that this function yields the desired
ρA (x) + ρB (x)
result of Urysohn’s Lemma for metric spaces.

Lemma: Let (X, T ) be a normal space, and let C be a closed subset. Let O be an open
subset such that C ⊆ O. Then there exists an open set U such that C ⊆ U ⊆ U ⊆ O.

Proof: C and Oc are disjoint closed sets, so there are disjoint open sets U, V such
that C ⊆ U and Oc ⊆ V. Then C ⊆ U ⊆ V c ⊆ O. V c is a closed set containing U ; it
therefore contains the closure U , so that C ⊆ U ⊆ U ⊆ O. 

Proof of Urysohn’s Lemma: By the lemma, there is an open set O1/2 such that
C0 ⊆ O1/2 ⊆ O1/2 ⊆ C1c . Applying the lemma again, there are open sets O1/4 and O3/4
such that C0 ⊆ O1/4 ⊆ O1/4 ⊆ O1/2 ⊆ O1/2 ⊆ O3/4 ⊆ O3/4 ⊆ C1c . Then there are open
sets O1/8 , O3/8 , O5/8 , O7/8 such that C0 ⊆ O1/8 ⊆ O1/8 ⊆ O1/4 ⊆ · · · ⊆ O7/8 ⊆ C1c .
Continuing the pattern, for each dyadic rational number (numbers of the form k2−n , for
n ∈ N, 0 < k ≤ 2n − 1), we get an open set Ok2−n , and if s, t are dyadic rationals in (0, 1)
such that s < t, then Os ⊆ Ot .
Define f : X → [0, 1] by f (x) = inf{r : r is a dyadic rational, x ∈ Or }.
Clearly, if x ∈ C0 , then x ∈ O2−n for any n ∈ N, so it follows that f (x) = 0. On the other
hand, if x ∈ C1 , then x 6∈ Ok2−n for any n, k, hence f (x) = 1 on C1 . Thus, it remains to
show that f is continuous. Recall that it suffices to consider the sub-base of open rays.

For a ≤ 0 and b ≥ 1, we get f −1 ((−∞, a)) = f −1 ((b, ∞)) = ∅. For a > 1 and
b < 0, f −1 ((−∞, a)) = f −1 ((b, ∞)) = X.
Suppose 0 ≤ a < 1. If x ∈ X and f (x) < a, then there is a dyadic rational r such that
f (x) < r < a, so x ∈ Or , so f −1 ((−∞, a)) = r<a Or , which is open.
S

Similarly, suppose 0 ≤ b < 1. If f (x) > b, then there is a dyadic rational r such that
b < r < f (x), so x 6∈ Or , so there is a dyadic rational s such that b < s < r, so Os ⊆ Or ,
c c
so x 6∈ Os , so x ∈ Os , which is open. Then f −1 ((b, ∞)) = s<b Os , which is open. 
S

Before we discuss Tietze’s Theorem, we digress to talk about Banach spaces.

2.4. SPECIAL TOPOLOGICAL SPACES 19

Def: A Banach space is a complete, normed vector space.

Let X be a set, and let V be a normed vector space. Let B(X, V ) denote the set of all
bounded functions from X to V, that is, functions whose range is contained in an open ball.
Then it can easily be checked that B(X, V ) is a vector space for pointwise operations, and
that kf k∞ := sup{kf (x)kV : x ∈ X} is a norm on B(X, V ).

Proposition: If V is a Banach space, then B(X, V ) with k·k∞ is a Banach space.

Proof: Let {fn } be a Cauchy sequence in B(X, V ). For each x ∈ X, the sequence
{fn (x)} is Cauchy in V, so by the completeness of V, call the limit f (x) = lim fn (x).
It is easy to see that since all the fn ’s are bounded, the limit f is bounded as well.
We need to show that fn → f in norm. Let  > 0. There exists N1 ∈ N such that
for n, m ≥ N1 , we have kfn − fm k∞ < /2. For fixed x ∈ X, there exists N2 ∈ N such
that for n ≥ N2 , we have kfn (x) − f (x)k < /2. Then for n ≥ max(N1 , N2 ), we have
kfn (x) − f (x)k∞ ≤ kfn − fn+1 k∞ + kfn+1 (x) − f (x)k < . Therefore kfn − f k < . 

Proposition: Let (X, T ) be a topological space, and let Cb (X, V ) be the set of bounded
continuous functions from X to V. Then Cb (X, V ) is a closed subspace.

Proof: Exercise. 

Tietze Extension Theorem: Let (X, T ) be a normal topological space, and let A be
a closed subset of X. Let f : A → R be continuous. Then f has a continuous extension
f˜ : X → R, i.e., f˜|A = f. If f : A → [a, b], then we can arrange the extension f˜ : X → [a, b].

Proof: First, we will prove the case f : A → [0, 1]. For E0 , F0 disjoint closed sets
in X, by Urysohn’s Lemma, let hE0 ,F0 : X → [0, 1] be a continuous function such that
hE0 ,F0 |E0 = 0 and hE0 ,F0 |F0 = 1.
Let f0 = f, and let A0 = {x ∈ A : f0 (x) ≤ 1/3}, B0 = {x ∈ A : f0 (x) ≥ 2/3}.
Clearly A0 and B0 are disjoint. Let g1 = 31 hA0 ,B0 .
Now let f1 = f0 − g1 |A . That is, f1 : A → [0, 2/3], and g1 : X → [0, 1/3].
Inductively, let fn : A → [0, (2/3)n ]. Let An = {x ∈ A : f (x) ≤ 13 (2/3)n },
 n
Bn = {x ∈ A : f (x) ≥ 32 (2/3)n }, with gn+1 = 1
3
2
3 hAn ,Bn , so
h  n i h  n+1 i
gn+1 : X → 0, 13 2
3 . Let fn+1 = fn − gn+1 |A , so fn+1 : A → 0, 13 2
3 .
 n−1 ∞
1 2
Note that kgn k∞ =
P
3 3 . Let g = gn . We will show that the sequence of partial
n=1
∞
P
sums is Cauchy in Cb (X, R), thus gn converges.
n=1
n n
gj . For m < n, consider kn − km =
P P
Let kn = gj .
j=1 j=m+1
20 CHAPTER 2. INTRODUCTION TO TOPOLOGY

n Pn  j−1
1 2
Then kkn − km k∞ ≤ kgj k∞ =
P
j=m+1 3 3 .
j=m+1
Clearly, for large enough n, m, we can get this arbitrarily small. Therefore g is well-defined
and continuous, by the previous proposition. Then
n
X
fn = fn−1 − gn = fn−2 − gn−1 − gn = · · · = f0 − gj ,
j=1

 n
so kfn k∞ = 32 , so kfn k∞ → 0, thus f − g|A = 0, i.e., g|A = f.
Finally, we want to check that the range of g is contained in [0, 1]. Note that
∞ ∞  n−1 ∞  n
X 1X 2 1X 2 1 1
g(x) = gn (x) ≤ = = · = 1.
n=1
3 n=1 3 3 n=0 3 3 1 − 2/3

Therefore 0 ≤ g(x) ≤ 1 for all x ∈ X.

Now suppose that f : A → R is unbounded. Let h be a homeomorphism of R

with (0, 1). Let g = h ◦ f, so g : A → (0, 1) ⊂ [0, 1]. By the arguments above, we can find an
extension g̃ : X → [0, 1]. Let D = g̃ −1 ({0, 1}). Since g̃ is continuous, D is closed in X and
is disjoint from A. By Urysohn’s Lemma, there exists a continuous function k : X → [0, 1]
such that k|D = 0 and k|A = 1. Define f˜ = g̃k (pointwise product). Then the function
h−1 ◦ f˜ is a continuous extension of f to X. 
Chapter 3

Compactness

3.1 Fundamentals

Def:

• Let X be a set, and let C be a collection of subsets of X. We say C covers X if

S
A = X.
A∈C

• If C is a cover of X, and D ⊆ C is also a cover of X, then D is a subcover of X.

• For a topological space (X, T ), an open cover is a cover of X that is contained in

T.

Def: (X, T ) is compact if every open cover of X has a finite subcover.

Def: Let F be a collection of subsets of X. Then F has the finite intersection property
if the intersection of any finite collection of sets in F is nonempty.

Proposition: (X, T ) is compact iff it has the property that whenever F is a collection of
A 6= ∅.
T
closed subsets of X with the finite intersection property, then
A∈F

Proof: Exercise. 

Proposition: Let (X, T ) be a topological space. Then A ⊆ X is compact for the relative
topology iff for any open cover of A, there is a finite subcover of A.

Proof: The open sets in the relative topology are exactly the sets of the form
A ∩ O, where O ∈ T . 

21
22 CHAPTER 3. COMPACTNESS

Proposition: If (X, Tx ) is compact, and (Y, Ty ) is a topological space, and f : X → Y is

continuous, then f (X) is compact in Y.

Proof: Let C ⊆ Ty be an open cover of f (X). Then {f −1 (O)}O∈C is an open

cover of X, so there is a finite subcover {f −1 (O1 ), ..., f −1 (On )} of X. It follows that
f −1 (O1 ) ∪ · · · ∪ f −1 (On ) = f −1 (O1 ∪ · · · ∪ On ) = X, so {O1 , . . . , On } is an open cover of
f (X). 

Proposition: If (X, T ) is compact, and A ⊆ X is closed, then A is compact.

Proof: Let C be an open cover of A. Since Ac is open, C ∪ {Ac } covers X, and

since (X, T ) is compact, there is a finite subcover of X, so clearly there is a finite subcover
for A. 

Proposition: If (X, T ) is Hausdorff, then any compact subset is closed.

Proof: Let A ⊆ X be compact, and let x∗ 6∈ A. For any y ∈ A, there are disjoint
open sets Oy , Uy such that y ∈ Oy and x∗ ∈ Uy . Then {Oy : y ∈ A} is an open cover of A,
n
so by compactness there is a finite subcover {Oy1 , . . . , Oyn }. Let U =
T
Uyj . Then U is
j=1
open and covers Ac and is also disjoint from A. It follows that U = Ac , and so A is closed.


Recall from standard real analysis the fact that in Rn , a subset is compact iff it is closed
and bounded (Heine-Borel).

Def: (X, T ) is regular if for any closed set A ⊆ X and any x 6∈ A, there are disjoint open
sets O, U such that A ⊆ O and x ∈ U.

3.2 Tychonoff’s Theorem

We now develop an important compactness theorem of Tychonoff, which says that a product
of compact spaces is compact in the product topology. In order to start, we need some set
theory.

Axiom of Choice: Given any family of non-empty sets, there is a set containing an ele-
ment from each of these sets.

We will see later that the axiom of choice is in fact equivalent to Tychonoff’s Theorem.
3.2. TYCHONOFF’S THEOREM 23

Def: A partially-ordered set P is a set with a partial order ≤, which is a relation that
satisfies

(i) x ≤ x for all x ∈ P

(ii) x ≤ y and y ≤ z implies x ≤ z

(iii) If x ≤ y, and y ≤ x, then x = y.

A totally/linearly-ordered set satisfies the extra condition

(iv) For any x, y ∈ P, either x ≤ y or y ≤ x.

Ex: If X is a set, consider its power set P(X). Then the relation A ≤ B iff A ⊆ B is a
partial order on P(X) but not a total order.
Ex: In the plane R2 , the relation x ≤ y iff kxk2 ≤ kyk2 is not a partial order, since two
points could have the same norm, but be unequal.
Ex: The usual relation ≤ on R is a total order.

Def:

• A chain in P is a totally-ordered subset of P.

• A maximal element in P is an element x ∈ P such that if y ≥ x, then y = x.

• An upper bound for a subset A ⊆ P is an element x ∈ P such that y ≤ x for all

y ∈ A.

• P is inductively ordered if every chain in P has an upper bound.

Zorn’s Lemma: If P is inductively ordered, then every chain C has a maximal element b
for C with a ≤ b for all a ∈ C.

Zorn’s Lemma seems quite obscure, but it is incredibly practical in many important results
in mathematics and particularly in analysis.

Tychonoff’s Theorem: Let {(Xα , Tα )} be a family of compact spaces indexed by A.

Q
Then X = α Xα with the product topology is compact.

Proof: We will show compactness of X through the finite intersection property.

Let C be a collection of closed subsets of X with the finite intersection property. We wish
to show that C∈C C 6= ∅.
T
24 CHAPTER 3. COMPACTNESS

Let Θ = {D ⊆ P(X) : C ⊆ D, D has finite intersection property}.

Then Θ is partially ordered by set inclusion. To show that Θ is inductively ordered, let
Φ ⊆ Θ be a chain, and let E = D∈Φ D. We show that E has the finite intersection property
S

so that E ∈ Θ and is thus an upper bound for Φ.

Let Z1 , . . . , Zn ∈ E. Then there are D1 , . . . , Dn ∈ Φ with Zj ∈ Dj . Since Φ is totally ordered,
one of the Dj ’s will be the largest, say, Dn , and each Dj ⊆ Dn . Then Z1 , . . . , Zn ∈ Dn . But
n
Dn has the finite intersection property, so Zj 6= ∅. Clearly C ⊆ E, so E ∈ Θ, and E is
T
j=1
an upper bound for Φ. Thus Θ is inductively ordered, as desired.

By Zorn’s Lemma, Θ contains a maximal element, say, D∗ , which will have the
following properties:

• If Z1 , Z2 ∈ D∗ , then Z1 ∩ Z2 ∈ D∗ .
Indeed, for Y1 , . . . , Yn ∈ D∗ , we have (Z1 ∩ Z2 ) ∩ (Y1 ∩ · · · ∩ Yn ) 6= ∅. Therefore
D∗ ∪ {Z1 ∩ Z2 } has the finite intersection property. But by maximality of D∗ , in fact
D∗ = D∗ ∪ {Z1 ∩ Z2 }, therefore Z1 ∩ Z2 ∈ D∗ .

• For Y ⊆ X, if Y ∩ Z 6= ∅ for all Z ∈ D∗ , then Y ∈ D∗ .

n
Indeed, for Z1 , . . . , Zn ∈ Z, we have Y ∩ (Z1 ∩ · · · ∩ Zn ) = (Y ∩ Zj ) 6= ∅, and by
T
j=1
maximality, Y ∈ D∗ .

For any D ∈ Θ and for any α ∈ A, we claim that Fα := {πα (Y ) : Y ∈ D} has the finite
intersection property. By the finite intersection property of D, for any Z1 , . . . , Zn ∈ D,
there exists x ∩ Z1 , . . . ∩ Zn . Then πn
α (x) ∈ πα (Z1 ) ∩o· · · ∩ πα (Zn ), thus Fα has the finite
intersection property. It follows that πα (Z) : Z ∈ D has the finite intersection property.
This is a collection of closed subsets with the finite intersection property in Xα , which is
πα (Z) 6= ∅.
T
compact, so
Z∈D
Apply this to each D ⊆ D∗ . For each α ∈ A, by the axiom of choice, pick xα ∈ Xα .
Let x = (xα ) ∈ α∈A Xα . We claim that x ∈
Q T
C.
C∈C
It suffices to show that if O is an open set containing x, then O ∩ C 6= ∅ for all C ∈ C.
It further suffices to have O be a basis element for the product topology. That is, suppose
x ∈ O = Uα1 × · · · × Uαn ×
Q
Xα .
α6=αi ∀i
Note that xαj ∈ Uαj for j = 1, . . . , n. Then xαj ∈ παj (Z) for all Z ∈ D∗ , so Uαj ∩πα (Z) 6= ∅
for all Z ∈ D∗ and for all 1 ≤ j ≤ n. So πα−1 j
(Uαj ) ∩ Z 6= ∅ for all Z ∈ D∗ , so by maximality,
n
πα−1 (Uαj ) ∈ D∗ , thus πα−1 (Uαj ) ∈ D∗ , so O ∈ D∗ .
T
j j
j=1
C 6= ∅, hence X is compact.
T
This proves that C∈C 

Ex: Here is an interesting application of Tychonoff’s Theorem.

Let H be a Hilbert space (complete inner product space) with its norm induced by its
inner product. Given η ∈ H, define φη : H → R by φη (ξ) = hη, ξi, and thus we can
put on H the weakest topology making all {φη }η∈H continuous. The closed unit ball
B := {ξ ∈ H : kξk ≤ 1} is compact with the relative weak topology. The proof of this uses
3.3. COMPACT AND HAUSDORFF 25

Tychonoff’s Theorem. To start, for each η ∈ H, let Dη = {r ∈ R : |r| ≤ kξk}, with the
Q
usual topology. Then form η∈H Dη , and one can show that this product is equal to B.
Clearly each Dη is compact, so B is thus compact.

Tychonoff’s Theorem uses the axiom choice for its proof, but in fact, it is equivalent to
the axiom of choice as well! Note that the axiom of choice essentially says that the product
of non-empty sets is non-empty.

Theorem: Let {Xα }α∈A be any collection of non-empty sets. Then without the axiom of
choice, and assuming Tychonoff’s Theorem, α∈A Xα 6= ∅.
Q

S S
Proof: Let X = α∈A Xα . Now let ω be some set not in α∈A Xα . For each α, let
Yα = Xα ∪ {ω}, and define the topology Tα for Yα by Tα = {Xα , {ω}, Yα , ∅}. Clearly,
(Yα , Tα ) is compact. Then Y := α∈A Yα with the product topology is compact. For each
Q

α, let Cα = πα−1 (Xα ), where πα : Y → Yα is the standard projection map. Note that Cα is
closed.
We will show that {Cα } has the finite intersection property. Given Cα1 , . . . , Cαn , where
xαj ∈ Xαj , define y ∈ Y by
(
xαj , if α = αj ,
yα =
ω, if α 6= αj ∀j
Then y ∈ nj=1 Cαj , so {Cα } has the finite intersection property, as desired. By compact-
T

ness, α∈A Cα 6= ∅. Let z ∈ α∈A Cα . Then z ∈ Xα for all α, so z ∈ α∈A Xα . Thus we

T T Q

can deduce the axiom of choice from Tychonoff’s Theorem. 

3.3 Compact and Hausdorff

Proposition: If (X, T ) is compact and Hausdorff, then it is normal.

Proof: Let C, D be disjoint closed sets of X. Since X is compact, C and D are

also compact, and since X is Hausdorff, it follows that it is regular. Therefore, for any
x ∈ D, there are disjoint open sets Ox , Ux with x ∈ Ux and C ⊆ Ox . Then {Ux }x∈D is an
open cover for D, so there is a finite subcover. That is, there are points x1 , . . . , xn ∈ D
such that {Ux1 , . . . , Uxn } covers D. Then U := nj=1 Uxj is an open set containing D, and
S

O := nj=1 Oxj is an open set containing C, and U ∩ C = ∅. 

T

Proposition: Let X be a set, and let T1 , T2 be topologies on X with T1 ⊇ T2 . Then

(i) If (X, T1 ) is compact, so is (X, T2 ).

(ii) If (X, T2 ) is Hausdorff, so is (X, T1 ).

Proof: Exercise. 
26 CHAPTER 3. COMPACTNESS

Corollary: If (X, T1 ) and (X, T2 ) are both compact and Hausdorff, then T1 = T = 2.

Proof: If C is closed for T, then it is compact for T1 , so is compact for T2 , so is

closed for T2 . 

Proposition: Let (X, Tx ) and (Y, Ty ) be compact Hausdorff spaces. If f : X → Y is

continuous and bijective, then it is a homeomorphism.

Proof: It suffices to show that f (C) is closed for any closed set C ⊆ X. Since X
is compact, C is compact, so f (C) is compact by continuity of f. Now since Y is Hausdorff,
it follows that f (C) is closed. 

3.4 Compactness for Metric Spaces

We now study specifically the importance of compactness for metric spaces, starting with
a simple necessary and sufficient condition for a metric space to be compact, and leading
to the Arzela-Ascoli theorem.

First, observe that if (X, d) is a compact metric space, and if A ⊆ X is dense in X,

then the balls B (y) for  > 0 and y ∈ A form an open cover for X, so there is a finite
subcover.

Def: A metric space (X, d) is totally bounded if for any  > 0, there is a finite collection
of open balls of radius  that covers X.

Proposition: Let (X, d) be a metric space, and let A ⊆ X. If A is totally bounded, then
so is A.

Proof: Let  > 0. Then there are points y1 , . . . , yn ∈ A such that {B/2 (yj )}nj=1
covers A. For each z ∈ A, there exists y ∈ A such that z ∈ B/2 (y), and there is some j
such that y ∈ B/2 (yj ). Therefore z ∈ B (yj ), so that {B (yj )}nj=1 covers A. 

Proposition: Let (X, d) be a metric space. If X is compact, then it is complete.

Proof: We prove the contrapositive. Suppose X is not complete. Let {xn } be a

Cauchy sequence that does not converge. For any x ∈ X, there exists x > 0 such that for
any N ∈ N, there exists n ≥ N such that d(xn , x) ≥ x . But since {xn } is Cauchy, there
is M ∈ N such that for any n, m ≥ M, we have d(xn , xm ) < x . Pick Mx > M such that
there is nx ≥ Mx with d(x, xnx ) ≥ x . So for n > Mx , we have d(x, xn ) ≥ /2. Therefore
3.4. COMPACTNESS FOR METRIC SPACES 27

for each x ∈ X, Bx (x) contains at most a finite number of elements in the sequence {xn }.
Clearly, the balls {Bx (x)} cover X, and no finite subcollection can cover X. 

Compactness clearly implies totally bounded, but the converse is not true. The following
theorem says that a metric space must be both totally bounded and complete. For example,
the non-compact space (0, 1) is totally bounded, but not complete. Furthermore, R is
complete, but not totally bounded.

Theorem: Let (X, d) be a complete metric space. If X is totally bounded, then it is

compact.

Proof: Let C be an open cover of X, and let B11 , . . . , Bn1 be a finite cover of X by
closed balls of radius 1. Toward contradiction, suppose X is not compact, so at least one of
these closed balls, denote it by A1 , has no finite subcover. Let B12 , . . . , Bn2 2 be closed balls
of radius 1/2 that cover A1 . One of these, say, B∗2 has no finite subcover. Let A2 = A1 ∩ B∗2 .
Let B13 , . . . , Bn3 3 be balls of radius 1/4 that cover A2 . One of these has no finite subcover,
etc. By continuing this process, we get a sequence {An } of non-empty closed sets with
An+1 ⊆ An for all n. Furthermore, by construction, diam(An ) → 0.
For each n, choose xn ∈ An . Then {xn } is a Cauchy sequence, and by completeness, there
is x ∈ X such that xn → x. Since C is a cover, there is O ∈ C with x ∈ O, and then there
is  > 0 such that B (x) ⊆ O. There is also N ∈ N such that for n ≥ N, xn ∈ B/2 (x).
For large n, we can get diam(An ) < /2, An ⊆ B (xn ) ⊆ O, so An is covered by C, a
contradiction. 

Recall that for a set X and normed vector space V, B(X, V ) denotes the set of bounded
functions from X to V.

Def: Let X be a set, and let (Y, d) be a metric space. A function f : X → Y is bounded
if its range is bounded in Y.

Let B ∗ (X, Y ) denote the set of bounded functions from X to Y (Y is not necessarily a
normed vector space). One can verify that d∞ (f, g) = sup {d(f (x), g(x))} is a metric on
x∈X
B ∗ (X, Y ). Furthermore, one can check that the bounded continuous functions, denoted by
Cb (X, Y ), is a closed subspace of B ∗ (X, Y ). Given a collection F ⊆ Cb (X, Y ), when is F
totally bounded?
• Suppose it is. Then for  > 0, there are g1 , . . . , gn ∈ Cb (X, Y ) such that the balls
B (gj ) cover F. Let x ∈ X. Then for each j, there is Oj ⊆ Tx such that x ∈ Oj and
y ∈ Oj imply d(gj (x), gj (y)) < .
n
Let Ox = Oj ∈ Tx , where clearly x ∈ Ox , and if y ∈ Ox , then d(gj (x), gj (y)) < 
T
j=1
for all j = 1, . . . , n. For any f ∈ F there is j such that d∞ (f, g) < . Then for any
y ∈ Ox , we have
d(f (x), f (y)) ≤ d(f (x), g(x)) + d(g(x), g(y)) + d(g(y), f (y)) < 3.
28 CHAPTER 3. COMPACTNESS

Then for each x and each 0 > 0, there is an open set Ox such that if y ∈ Ox , then
d(f (x), f (y)) < 0 for all f ∈ F.

Def: Let (X, T ) be a topological space, and let (Y, d) be a metric space. Let F ⊆ C(X, Y )
(a collection of continuous functions from X to Y ). F is equicontinuous at x if for any
 > 0, there exists an open set Ox in X such that for all f ∈ F and any x0 ∈ Ox , we have
d(f (x), f (x0 )) < .
F is equicontinuous if it is equicontinuous at every x ∈ X.

Continuing the previous argument, note that d(f (x), gj (x)) < , i.e., the open balls B (gj (x))
cover {f (x) : f ∈ F}, so F is “pointwise totally bounded.”

Arzela-Ascoli Theorem: If (X, T ) is compact, (M, d) is a metric space, and

F ⊆ Cb (X, Y ), and if F is equicontinuous and pointwise totally bounded, then F is
totally bounded for d∞ .

Proof: Let  > 0. Since F is equicontinuous, for each x ∈ X there is an open set
Ox with x ∈ Ox such that for y ∈ Ox , we have d(f (x), f (y)) <  for all f ∈ F. Since X is
compact, there are points x1 , . . . , xn such that X ⊆ nj=1 Oxj . For each j, {f (x) : f ∈ F}
S

is totally bounded since F is pointwise totally bounded. Let Sj ⊆ {f (xj ) : f ∈ F} ⊆ M

be a finite subset of M such that the balls of radius  about the points of Sj cover
{f (xj ) : f ∈ F}. Let S = nj=1 Sj , and let Ψ = {ψ : {1, . . . , n} → S}, a finite set. Let
S

Bψ = {f ∈ F : d(f (xj ), ψ(j)) < , ∀j}. Then F = ψ∈Ψ Bψ .

S

Let ψ be given. Let f, g ∈ Bψ , and let x ∈ X, so that x ∈ Oxj for some j. Then

d(f (x), g(x)) ≤ d(f (x), f (xj )) + d(f (xj ), g(xj )) + d(g(xj ), g(x)) ≤

≤ d(f (x), f (xj )) + d(f (xj ), ψ(j)) + d(ψ(j), g(xj )) + d(g(xj ), g(x)) < 4.
Therefore Bψ is contained in the ball of radius 4 about any of its points. Since Ψ is finite,
it follows at once that F is totally bounded for d∞ . 

Corollary: Let M be a complete metric space, so that C(X, M ) is complete. Then

F ⊆ C(X, M ) is compact iff F is equicontinuous, pointwise totally bounded, and closed in
C(X, M ).

3.5 Locally Compact Spaces

Although compact spaces form a nice class of topological spaces with many properties, there
are many important spaces that share a similar structure: these are the locally compact
spaces. Even more important are the locally compact Hausdorff spaces, which we discuss.
3.5. LOCALLY COMPACT SPACES 29

Def: A topological space (X, T ) is locally compact if for each x ∈ X, there is an open
set O containing x such that O is compact.

Proposition: Let (X, T ) be locally compact, and let C be a compact subset of X. Then
there exists an open set O such that C ⊆ O, and O is compact.

Proof: For each x ∈ C, there is Ox ∈ T such that x ∈ Ox and Ox is compact. {Ox }x∈X cov-
ers C, so there is a finite subcover, say, O1 , . . . , On . Then C ⊆ O1 ∪· · ·∪On = O1 ∪ · · · ∪ On ,
which is compact. 

From here, let LCH mean locally compact Hausdorff.

Proposition: Let (X, T ) be a LCH space, let C ⊆ X be compact, and let O ∈ T , C ⊆ O.

Then there is an open set U such that C ⊆ U ⊆ U ⊆ O and U is compact.

Proof: We know that we can find an open set V such that C ⊆ V and V is com-
pact. Let W = V ∩ O. Then C ⊆ W, and W is open. Furthermore, W is compact,
and so the relative topology of W makes W compact and Hausdorff, hence normal. Let
B = W \ W, so B is closed and disjoint from C, so by normality, there are disjoint
open sets U, Z such that C ⊆ U and B ⊆ Z. Then U ⊆ Z c ∩ W , so U ⊆ Z c ∩ W , so
U ⊆ B c = (W \ W )c ∩ W = W. Thus U ⊆ W ⊆ O and U is compact. 

Def: For a continuous function f on X to a normed vector space, its support is the set
supp(f ) := {x : f (x) 6= 0}.
f has compact support if its support is compact.

Let V be a normed vector space. We will let Cc (X, V ) denote the set of all continuous
functions from X to V with compact support.

Proposition: Let (X, T ) be a LCH space. Let C ⊆ X be compact, and let O be open
with C ⊆ O. Then there is a continuous function f : X → [0, 1] such that f = 1 on C, and
f = 0 outside O, and f has compact support.

Proof: There exists U ∈ T such that C ⊆ U and U is compact, and U ⊆ O.

There also exists V ∈ T with C ⊆ V ⊆ V ⊆ U. Let B = U \ V, so that B is closed and
disjoint from C. Since U is compact and Hausdorff, we apply Urysohn’s Lemma to obtain
a continuous function f : U → [0, 1] such that f = 1 on C, and f = 0 on B. For x 6∈ U, set
f (x) = 0. Since U ⊆ O, it follows that f = 0 outside O, as desired. It remains to show
that f : X → [0, 1] is continuous, but this is left as an exercise. 
30 CHAPTER 3. COMPACTNESS

Remark: Cc (X) ⊆ Cb (X), which denotes the set of bounded continuous functions. Equip
Cb (X) with the norm k·k∞ , so that kf gk∞ ≤ kf k∞ kgk∞ , making Cb (X) a Banach algebra.
One can verify that the closure of Cc (X) in Cb (X) is the space of all continuous functions
that vanish at ∞, denote C∞ (X).
f vanishes at ∞ if for any  > 0 there exists a compact K such that kf (x)k <  for
x 6∈ K.
Chapter 4

Measure Theory

4.1 Fundamentals

Def: Let X be a set, and let M be a collection of subsets of X. We say M is a ring if

(1) E ∪ F ∈ M for any E, F ∈ M.

(2) E \ F ∈ M for any E, F ∈ M.

M is an algebra/field if M is a ring, and X ∈ M.

Note that (1) and (2) imply that if E, F ∈ M, then E ∩ F ∈ M.

Ex: Clearly, P(X) is a ring and an algebra on X.

Ex: The set M of all finite subsets of X is a ring, but it is not an algebra if X is infinite.
Ex: If X = R, then M := {[a, b) : a < b} is a ring.

Def: F is a σ-ring on X if

(1) F is a ring
∞
(2) If {En }∞
n=1 is a countable collection of elements in F, then En ∈ F.
S
n=1

F is a σ-algebra/field if it is a σ-ring, and X ∈ F.

Note that σ-rings are also closed under countable intersections.

Proposition: Let X be a set, and let {Fα } be a collection of rings/algebras/σ-rings/σ-

algebras on X. Then α Fα is also a ring/algebra/σ-ring/σ-algebra, respectively.
T

Proof: Exercise. 

31
32 CHAPTER 4. MEASURE THEORY

Proposition: For any family of sets A ⊆ P(X), there is a smallest ring/algebra/σ-ring/σ-

algebra that contains A.

Proof: The smallest such ring/algebra/σ-ring/σ-algebra is in fact the intersection

of all rings/algebras/σ-rings/σ-algebras that contain A. 

The smallest such ring/algebra/σ-ring/σ-algebra containing A is the ring/algebra/σ-ring/σ-

algebra generated by A.
The smallest σ-ring generated by A will be denoted by S(A).

Def: If (X, T ) is a topological space, then the σ-algebra generated by T is called the
Borel σ-algebra.
If (X, T ) is a locally compact space, let C denote the collection of all compact sets. The
σ-ring generated by C is called the Borel σ-ring.

Ex: Let X be an uncountable set with the discrete topology. Then compact sets are finite
sets, so the Borel σ-ring consists of all countable subsets of X.

Def: Let A be a collection of subsets of a set X. A function µ : A → [0, ∞] is additive if

for all disjoint E, F ∈ A such that E ∪ F ∈ A, we have µ(E ∪ F ) = µ(E) + µ(F ).
µ is countable additive if whenever {En }∞ n=1 is a countable collection of mutually disjoint
∞ S  ∞
∞
En ∈ A, we have µ
S P
sets such that n=1 En = µ(En ).
n=1 n=1

Def: A measure on X consists of a σ-ring S on X, together with a function µ : S → [0, ∞]

that is countably additive.

Ex: Let X be a set with the σ-field P(X). Then µ(E) = |E|, where |E| = ∞ if E is infinite,
is a measure (we could also have the σ-field be all countable subsets). This measure is called
counting measure.

4.2 Borel Measures on R

We will develop Borel measures in more generality. That is, let α : R → R be any function
that is non-decreasing and left continuous, i.e., for any t, we have lim α(t − ) = α(t). If
↓0
we weren’t general, we would have used α(t) = t. These will lead to the “usual” measure
on R.

Let P = {[a, b) : a < b ∈ R}, and define µ on P by µ([a, b)) = α(b) − α(a). We will
show that µ is countably additive.
4.2. BOREL MEASURES ON R 33

Def: Let X be a set, P a collection of subsets. Then P is a pre-ring/semi-ring if

(1) E, F ∈ P implies E ∩ F ∈ P.

(2) E, F ∈ P implies E \ F is a finite disjoint union of elements in P.

Then it is easy to see that P = {[a, b) : a < b} is a pre-ring on R.

Def: For a pre-ring P, a function µ : P → [0, ∞] is a premeasure on P if it is countably

additive.

Theorem: The function µ on P defined above is countably additive. That is, µ is a

pre-measure.

Proof: Let {[aj , bj )}∞ disjoint, and let [a, b) = ∞

S
j=1 beP j=1 [aj , bj ) be their union.
∞
We will first show that µ([a ,
j j b )) ≤ µ([a, b)). To show this, let n ∈ N.
Pn Pj=1
n
Then j=1 µ([aj , bj )) = j=1 [α(bj ) − α(aj )]. Without loss of generality, suppose
a1 < a2 < · · · < an . Since the intervals are disjoint, bj ≤ aj+1 for all j, so α(bj ) ≤ α(aj+1 ).
Then
n
X
µ([aj , bj )) ≤ [α(bn ) − α(an )] + [α(an ) − α(an−1 )] + · · · + [α(a2 ) − α(a1 )] ≤ α(b) − α(a).
j=1

Since this holds for any n, it holds for the limit.

∞ 0
Conversely, let  > 0. Let {j } be a sequence such that
P
j=1 j < /2, and let b
0
be a number such that α(b ) + /2 ≥ α(b). For each j, choose aj < aj such that 0

α(a0j ) + j ≥ α(aj ).
Then [a, b0 ] ⊆ [a, b) = ∞
S∞ 0 0 S∞ 0
j=1 [aj , bj ) ⊆ j=1 (aj , bj ). Since [a, b ] is compact, and j=1 (aj , bj ) is
S

an open cover, there must be a finite subcover. Relabel these intervals: (a01 , b1 ), . . . , (a0n , bn ),
such that a0 ∈ (a01 , b1 ), and bj ∈ (a0j+1 , bj+1 ), so a0j+1 < bj , and also b0 ≤ bn . Then
∞
[α(bj ) − α(a0j )] = α(bn ) + α(bn−1 ) − α(a0n ),
X

j=1

where α(bn−1 ) − α(a0n ) ≥ 0. Then

  
α(b) − α(a) ≤ α(b0 ) + − α(a) ≤ α(bn ) + − α(a) ≤ α(bn ) − α(a1 ) + ≤
2 2 2

≤ α(bn ) + [α(bn−1 ) − α(a0n )] + [α(bn−2 ) − α(a0n−1 )] + · · · + α(a01 ) + =
2
n n ∞

[α(bj ) − α(a0j )] ≤
X X X
= [α(bj ) − α(aj ) + j ] + ≤ µ([aj , bj )) + .
j=1 j=1
2 j=1
∞
Therefore, µ([a, b)) ≤
P
µ([aj , bj )). 
j=1
34 CHAPTER 4. MEASURE THEORY

Note that if P is a pre-ring, then for E, F1 , F2 ∈ P, there exist {Gjk }nj,k=1

1 ,n2
⊆ P, all disjoint,
such that n n n
[1  [1 [2
(E \ F1 ) \ F2 = Gj \ F 2 = Gjk .
j=1 j=1 k=1
S` Sp
Then for F1 , . . . , F` ∈ P, we have E \ i=1 Fi = j=1 Gj , where Gj ∈ P.

Sn
Proposition: If µ : P → [0, ∞], and µ is finitely additive, and if E ⊆ j=1 Fj , where the
Fj ’s are disjoint, and E, Fj ∈ P, then µ(E) ≤ nj=1 µ(Fj ).
P

Proof:
n
[ n
[  n1 [
[ n2 
Fj = E ∪ Fj \ E = E ∪ Gjk ,
j=1 j=1 j=1 k=1

for some {Gjk } ⊆ P disjoint. Furthermore, E and nj=1

S 1 Sn2
k=1 Gjk are disjoint, so

n
X X n2
n1 X
µ(Fj ) = µ(E) + µ(Gjk ) ≥ µ(E).
j=1 j=1 k=1

Corollary: If E ⊆ F, then µ(E) ≤ µ(F ).

Def: A function µ satisfying µ(E) ≤ µ(F ) whenever E ⊆ F is called monotone.

For a family of sets F, we say µ : F → R is countably subadditive if whenever

E⊆ ∞
P∞
j=1 Fj , where E, Fj ∈ F, then µ(E) ≤
S
j=1 µ(Fj ).

Proposition: If (P, µ) is a pre-measure, then µ is countably subadditive.

S 
Proof: Suppose E ⊆ ∞ j=1 Fj . Then E = E ∩
∞ ∞
j=1 Fj = j=1 (E ∩ Fj ).
S S

Note that µ(E ∩ Fj ) ≤ µ(Fj ), so it suffices to show the result for E = ∞

S
j=1 Fj .
Define H1 = F1 , and for n > 1, define Hn = Fn \ n−1
S
F
j=1 j . Then clearly H1 , H2 , . . . are all
S∞ S∞ nj
disjoint, and clearly j=1 Hj = j=1 Fj . Now for each j, there are sets {Gjk }k=1 such that
Snj
Hj = k=1 Gjk , so
∞
[  ∞
X ∞ 
X j−1
[ 
µ(E) = µ Hj = µ(Hj ) = µ Fj \ Fk =
j=1 j=1 j=1 k=1
nj
∞ X  nj
∞ X ∞
X X X
= µ Gjk = µ(Gjk ) ≤ µ(Fj ).
j=1 k=1 j=1 k=1 j=1

4.3. OUTER MEASURES 35

4.3 Outer Measures

Def: Let F be a family of subsets of X. Let A ⊆ X. Then A is countably covered by F

if there is a sequence {Fj } of subsets of F with A ⊆ ∞
S
j=1 Fj .

Let H(F) denote the collection of all subsets of X that are countably covered by F.
Properties of H(F):
(i) H(F) is a σ-ring

(ii) H(F) is hereditary, i.e., if A ∈ H(F), and B ⊆ A, then B ∈ H(F).

Given H(F), let µ : F → [0, ∞] be any function. Define µ∗ : H(F) → [0, ∞] by
∞
nX ∞ o
µ∗ (A) = inf
[
µ(Fj ) : A ⊆ Fj , F j ∈ F .
j=1 j=1

Def: Let H be a hereditary σ-ring of subsets on X. A function ν : H → [0, ∞] is an

outer measure if

(i) ν(∅) = 0

(ii) ν is monotone

(iii) ν is countably subadditive

Proposition: µ∗ : H(F) → [0, ∞] defined as above is an outer measure.

Proof: Clearly, µ∗ (∅) = 0, and monotonicity is trivial.

Suppose A ⊆ ∞ ∞ P∞
j=1 Bj . Let  > 0, and suppose {j }j=1 satisfies
S
j=1 j < . S
∞
For each j, there is a collection {Bjk }∞ k=1 ⊆ F such that Bj ⊆ k=1 Bjk and
∞
µ∗ (Bj ) + j >
P
µ(Bjk ). Then
k=1

∞
[  ∞ X
∞ ∞   ∞
µ∗ (A) ≤ µ∗ µ∗ (Bj ) + j < µ∗ (Bj ) + .
X X X
Bj ≤ µ(Bjk ) <
j=1 j=1 k=1 j=1 j=1
P∞
It follows that µ∗ (A) ≤ j=1 µ
∗ (B ),
j so µ∗ is countably subadditive. 

Proposition: Let (P, µ) be a pre-measure. Then µ∗ agrees with µ on P.

S∞
Proof: Let E ∈ P. Clearly, µ∗ (E) ≤ µ(E). However, if E ⊆ j=1 Ej , then
µ(E) ≤ ∞ ∗ ∗
j=1 µ(Ej ), so µ (E) ≥ µ(E), hence µ (E) = µ(E). 
P
36 CHAPTER 4. MEASURE THEORY

4.4 From Outer Measure to Measure

Def: Let ν be an outer measure on a hereditary σ-ring H. Then E ∈ H is ν-measurable

if for any A ∈ H,
ν(A) = ν(A ∩ E) + ν(A ∩ E c ).

Note that by subadditivity, we always have ν(A) ≤ ν(A ∩ E) + ν(A ∩ E c ).

From now on, let ⊕ denote a disjoint union.

Let M(ν) denote the set of all ν-measurable sets in H.

Caratheodory’s Theorem M(ν) is a σ-ring, and ν restricted to M(ν) is a measure.

Proof: First, we show that M(ν) is a ring.

Let E, F ∈ M(ν). Let A ∈ H. Then

ν(A ∩ (E ∪ F )) + ν(A ∩ (E ∪ F )c ) =
   
= ν (A ∩ E) ⊕ (A ∩ E c ∩ F ) + ν (A ∩ E c ) ∩ F c ≤
≤ ν(A ∩ E) + ν((A ∩ E c ) ∩ F ) + ν((A ∩ E c ) ∩ F c ) = ν(A ∩ E) + ν(A ∩ E c ) = ν(A),
so E ∪ F ∈ M(ν). On the other hand,
   
ν A ∩ (E ∩ F c ) + ν A ∩ (E c ∪ F ) =
   
= ν (A ∩ E) ∩ F c + ν (A ∩ E c ) ⊕ (A ∩ E ∩ F ) ≤
≤ ν(A ∩ E ∩ F c ) + ν(A ∩ E c ) + ν(A ∩ E ∩ F ) = ν(A ∩ E) + ν(A ∩ E c ) = ν(A),
so E \ F ∈ M(ν). Therefore M(ν) is a ring.

We now show that ν is finitely additive on M(ν).

Let E, F ∈ M(ν) be disjoint, and let A = E ∪ F. Then E “splits” A:

ν(A) = ν(A ∩ E) + ν(A ∩ E c ) = ν(E) + ν(F ).

Furthermore, for E, F disjoint and for any A ∈ H,

 
ν (A ∩ E) ∪ (A ∩ F ) = ν(A ∩ E) + ν(A ∩ F ).

Finally, we show that M(ν) is a σ-ring.

Let {Ej }∞j=1 ⊆ M(ν). Since M(ν) is a ring, we can construct, as done previously, disjoint
sets {Fj }j=1 such that j=1 Ej = ∞
∞ S∞
j=1 Fj , where Fj ∈ M(ν). Let A ∈ H. For any m ≥ 1,
S

we know that
 m
[   m
[ c  m
X  m
[ c 
ν(A) = ν A ∩ Fj +ν A∩ Fj ≥ ν(A ∩ Fj ) + ν A ∩ Fj ≥
j=1 j=1 j=1 j=1
4.4. FROM OUTER MEASURE TO MEASURE 37

m
[   m
[ c   m
[   m
[ c 
≥ν (A ∩ Fj ) + ν A ∩ Fj =ν A∩ Fj +ν A∩ Fj .
j=1 j=1 j=1 j=1
S∞
In particular, this holds for A = j=1 Fj , so
∞
X ∞
X
ν(A) = ν(A ∩ Fj ) = ν(Fj ),
j=1 j=1

so ν restricted to its ν-measurable sets M(ν) is a measure. 

Proposition: Let (P, µ) be a pre-measure. From H(P ) define the outer measure µ∗ as
done previously. Let M(µ∗ ) denote the σ-ring of µ∗ -measurable sets contained in H(P ).
then P ⊆ M(µ∗ ).

Proof: Let E, F ∈ P, so that there are disjoint sets G1 , . . . , Gn ∈ P such that

E \ F = j=1 Gj . Then µ∗ (E \ F ) ≤ nj=1 µ(Gj ). By measurability,
Sn P

n
µ∗ (E) = µ∗ (E ∩ F ) + µ∗ (Gj ) ≥ µ∗ (E ∩ F ) + µ∗ (E \ F ).
X

j=1

Therefore µ(E) ≥ µ(E ∩ F ) + µ∗ (E \ F ). Now we must show that, given A ∈ H(P ) and
E ∈ P, then µ∗ (A) ≥ µ∗ (A ∩ E) + µ∗ (A \ E).
⊆ ∞
P∞
Let  > 0. There exists {Fj }∞
j=1 ⊆ P such that A S
∗
S
j=1 Fj , and µ (A) +  > j=1 µ(Fj ).
Notice that A ∩ E ⊆ j=1 (Fj ∩ E), and A \ E ⊆ ∞
S∞
j=1 (Fj \ E). Therefore
∞ ∞  
µ∗ (A) +  ≥ µ∗ (Fj ∩ E) + µ∗ (Fj \ E) ≥ µ∗ (A ∩ E) + µ∗ (A \ E).
X X
µ(Fj ) =
j=1 j=1

It follows that µ∗ (A) ≥ µ∗ (A ∩ E) + µ∗ (A \ E), so E is µ∗ -measurable. 

Proposition: Let (H, ν) be any outer measure. If A ∈ H, and ν(A) = 0, then A ∈ M(ν).

Proof: Let B ∈ H. Then ν(B ∩ A) = 0, since B ∩ A ⊆ A, and also ν(B ∩ Ac ) ≤ ν(B),

since B ∩ Ac ⊆ B. Measurability of A follows immediately. 

In other words, sets of zero outer measure are measurable.

Def: A measure (S, µ) is complete if whenever E ⊆ S, and µ(E) = 0, then for all A ⊆ E,
we have A ∈ S and µ(A) = 0.
Given a complete measure, sets whose measure is zero are called null sets.

• For a complete measure, the null sets form a hereditary σ-ring.

38 CHAPTER 4. MEASURE THEORY

• For any outer measure (H, ν) the corresponding measure on M(ν) is complete (from
the last proposition).
• For a pre-measure (P, µ), µ∗ on S(P ) may not be complete. However, S(P ) ⊆ M(µ∗ ),
where M(µ∗ ) is complete.
• It’s easy to see that for P = {[a, b) : a < b}, that S(P ) is exactly the Borel σ-algebra
on R (countable operations on open intervals can yield these half-open intervals).
• Extensions of measures are in general not unique.

Def: Let F be a family of subsets of X, and let ν : F → [0, ∞]. Then A ∈ F is σ-finite if
S∞
there is {Bj }∞
j=1 ⊆ F such that A ⊆ j=1 Bj , and ν(Bj ) < ∞ for all j.

We say that (F, ν) is σ-finite if every A ∈ F is σ-finite. If X ∈ F and X is σ-finite, then

we say that (F, ν) is totally σ-finite.

Lemma: Let (P, µ) be a pre-measure. Let S be any σ-ring with P ⊆ S ⊆ M(µ∗ ).

Let ν : S → [0, ∞] and let ν be countably subadditive, monotone, and ν|P = µ. Then
ν(A) ≤ µ∗ (A) for all A ∈ S such that A ∈ H(P ).
S∞
Proof: For any {Ej }∞
j=1 ⊆ P such that A ⊆ j=1 Ej , we have
∞
X ∞
X
ν(A) ≤ ν(Ej ) = µ(Ej ).
j=1 j=1
nP o
∞ S∞
Since µ∗ (A) = inf j=1 µ(Fj ) : A⊆ j=1 Fj , it follows that ν(A) ≤ µ∗ (A). 

Theorem: Let (P, µ) be a σ-finite pre-measure. Then for any σ-ring S such that
P ⊆ S ⊆ M(µ∗ ), and any measure ν on S such that ν|P = µ, we have ν = µ∗ |S .

Proof: Let A ∈ S.

• Step 1: Suppose A ⊆ E, E ∈ P, µ(E) < ∞. Then E = A⊕(E \A), so by measurability

of A, we have ν(E) = ν(A) + ν(E \ A) ≤ µ∗ (A) + µ∗ (E \ A) = µ∗ (E) = µ(E) = ν(E).
Thus ν(A)+ν(E\A) = µ∗ (A)+µ∗ (E\A). But ν(A) ≤ µ∗ (A), and ν(E\A) ≤ µ∗ (E\A),
and since µ∗ (A) ≤ µ∗ (E) < ∞, and µ∗ (E \ A) ≤ µ∗ (E) < ∞, it follows that in fact,
ν(A) = µ∗ (A), and ν(E \ A) = µ∗ (E \ A).

• Step 2: For general A ∈ S, there is {Ej } ⊆ P, A ⊆ ∞ j ) < ∞ for all

S
j=1 Ej , and µ(ES
j. Without loss of generality, we can let {Ej } be disjoint. Then A = ∞ j=1 (A ∩ Ej ),
so ν(A) = ∞ ∩ ∩ ⊆ ∩ µ∗ (A ∩ Ej ),
P
j=1 ν(A Ej ), but A E j Ej , so by step 1, ν(A Ej ) =
implying that µ∗ (A) = ∞ ∗
j=1 µ (A ∩ Ej ).
P


4.5. NON-MEASURABLE SETS 39

Def: On R, define α(x) = x, and consider the pre-measure µα defined by µα ([a, b)) = b − a,
which can be extended to M(µ∗α ). Then µα restricted to M(µ∗α ) is Lebesgue measure on
R.

• Lebesgue measure is translation invariant, i.e., if A ∈ M(µ∗α ), and r0 ∈ R, then if

r0 + A = {r0 + a : a ∈ A}, we have µα (r0 + A) = µα (A).

• Above, we can let α be any non-decreasing, left-continuous function. Then the re-
sulting measure is the general Lebesgue-Stieltjes measure.

Proposition: (Continuity from below): Let (S, µ) be a measure. If {Ej } ⊆ S, and

Ej ⊆ Ej+1 for all j, and if E = ∞j=1 Ej , then µ(Ej ) → µ(E).
S

L 
∞
Proof: Observe that E = E1 ⊕ j=1 Ej+1 \ Ej , so
P∞
µ(E) = µ(E1 ) + j=1 µ(Ej+1 \ Ej ) = lim µ(Em+1 ). 
m→∞

Observe that if Ej is a decreasing sequence of subsets, i.e., Ej+1 ⊆ Ej for all j, and if
E= ∞
T
j=1 Ej , then the analogue of the above proposition for intersections is not necessarily
true. Consider Ej = [j, ∞). However, we can say the following:

Proposition: (Continuity from above): If {Ej } ⊆ S, and Ej+1 ⊆ Ej for all j, with at
least one µ(Ej ) < ∞, and if E = ∞j=1 Ej , then µ(Ej ) → µ(E).
T

Proof: Note that µ(E) < ∞, and furthermore, suppose µ(EN ) < ∞. Then for all
n ≥ N, µ(En ) < ∞. Then the proof follows in a similar manner as above. 

4.5 Non-Measurable Sets

The construction of non-measurable sets often makes use of the axiom of choice. The fol-
lowing is a well-known example.
Ex: Let T = {z ∈ C : |z| = 1}. Let H = {z ∈ T : ∃n ∈ Z such that z n = 1}. Then T is a
compact group under multiplication, and H is a subgroup, so we can form the cosets of H.
Let A be a set that contains exactly one element from each coset of H. By rotation in-
variance, µ(hA) = µ(A) for any h ∈ H. Also, since H is countable, T = h∈H hA, so
L
P P
µ(T ) = h∈H µ(hA) = h∈H µ(A).
Clearly, 0 < µ(T ) < ∞. If µ(A) = 0, then this would yield µ(T ) = 0, a contradiction. But
if µ(A) > 0, then µ(T ) = ∞, a contradiction. Therefore A is not measurable.
40 CHAPTER 4. MEASURE THEORY
Chapter 5

Integration

5.1 Fundamentals

Let (X, S, µ) be a measure space, and let B be a Banach space.

To start simply, suppose S is a ring, and µ is finitely-additive.
(
b, if x ∈ E
Let f = bχE , for b ∈ R, and E ∈ S. That is, f (x) = .
0, if x 6∈ E

Def: Given (X, S), a function f : X → B is a simple S-measurable function if its range
is finite, and for each b ∈ Im(f ), we have {x : f (x) = b} ∈ S. Then f can be written as
f = nj=1 bj χEj , where the bj ’s are distinct, and the Ej ’s are disjoint.
P

It is clear that if f = ∞
P
j=1 bj χEj , where the Ej ’s are disjoint, but the bj ’s are not all
distinct, then f is simple S-measurable.

Proposition: If f, g are simple S-measurable, then so is f + g.

P ∞
Proof: Suppose f = j=1 bj χEj , and for simplicity, g = cχF . The general case fol-
lows easily.
Let En+1 = F \ nj=1 Ej , and let bn+1 = 0, so f = n+1
Ln+1
j=1 bj χEj with F ⊆
L P
j=1 Ej .
Ln+1
Therefore F = j=1 (F ∩ Ej ).
Then f = n+1
P
j=1 bj (χF ∩Ej + χEj \F ), so that

n+1
X n
X
f +g = (bj + c)χEj ∩F + bj χEj \F .
j=1 j=1

Then f + g is simple S-measurable. 

41
42 CHAPTER 5. INTEGRATION

Def: The integral (with respect to the measure µ) of a simple S-measurable function
f = nj=1 bj χEj is
P
Z n
X
f dµ = bj µ(Ej ).
j=1

Simple measurable function will be abbreviated as SMF.

Def: If a simple measurable function has a finite integral with respect to the measure µ, it
is called a simple µ-integrable function, abbreviated as SIF.

The following facts have easy proofs.

Proposition:
Z Z Z
(1) If f, g are SIFs and c ∈ R, then (cf + g) dµ = c f dµ + g dµ.

(2) If f is a SMF, then the map x 7→ kf (x)kB is a SMF.

Z
(3) If f is a real-valued SIF, and if f ≥ 0, then f dµ ≥ 0.
Z Z
(4) If f, g are real-valued SIFs with f ≥ g, then f dµ ≥ g dµ.

Proof: Exercise. 

Observe that the simple measurable functions

Z form a vector space (closed under addition
and scalar multiplication). Also, f 7→ f dµ is a linear functional.

Z
Def: The L1 norm of a measurable function is kf k1 := kf (x)kB dµ(x).

Proposition: If f, g are B-valued SIFs, then kf + gk1 ≤ kf k1 + kgk1 .

Proof:
Z Z Z
kf (x) + g(x)kB dµ(x) ≤ kf (x)kB dµ(x) + kg(x)kB dµ(x) = kf k1 + kgk1 .


5.2. TOWARDS L1 43

5.2 Towards L1
Let SIF(S, µ) denote the vector space of simple µ-integrable functions.

• If f is a SIF, and r ∈ R, then krf k1 = |r| · kf k1 . Therefore k·k1 is a seminorm on

SIF(S, µ).

• For any f : X → B, let C(f ) = {x : f (x) 6= 0}. If f is a SIF, then C(f ) ∈ S.

• Let N = {f ∈ SIF(S, µ) : µ(C(f )) = 0}. Then N is a vector subspace of SIF(S, µ).

Z
• If f ∈ N , then clearly f dµ = 0. Conversely, if f ∈ SIF(S, µ), and kf k1 = 0, then
kbj k χEj = 0 iff µ(Ej ) = 0 whenever bj 6= 0, implying that f ∈ N .
P

Z
• Thus, on SIF(S, µ)/N , k·k1 becomes a norm, and f 7→ f dµ is well-defined. There-
fore, SIF(S, µ)/N is a normed vector space, so we can then consider its abstract
completion (recall this is done by equivalence classes of Cauchy sequences).

• We will try to find a useful “concrete realization” of the elements of the completion,
which we will see is the space L1 (X, S, µ).

• Let {bn } be a Cauchy sequence in B, and let E ∈ S with 0 < µ(E) < ∞. Then
{bn χE } is a Cauchy sequence in SIF(S, µ)/N .
L∞
• Suppose {Ej } is a sequence of disjoint sets with 0 < µ(Ej ) < 2−j . Let E = j=1 Ej
and let Fn = nj=1 Ej .
L
Z
Let b ∈ B, b 6= 0. Suppose {bχFn } is a Cauchy sequence. Then bχFn dµ = b µ(Fn ),
Z Z
so that bχFn dµ → bχE dµ = b µ(E).

Def: Given a measurable space (X, S), let f : X → B. We say that f is S-measurable if
there is a sequence of SMFs {fn } converging pointwise to f.

Def: Let (X, S, µ) be a measure space. A ⊆ X is a null set if there is a set E ∈ S such
that A ⊆ E and µ(E) = 0. The µ-null sets form a σ-ring.

We say fn → f µ-almost everywhere (a.e.) if fn → f pointwise, except on a null

set.

Def: Let B be a Banach space. Let f : X → B (or at least f is defined on X \ A, for a

null set A). We say that f is µ-measurable if there is a sequence {fn } of B-valued SMFs
such that fn → f µ-almost everywhere.

• The S-measurable or µ-measurable functions form a vector space.

44 CHAPTER 5. INTEGRATION

• The function x 7→ kf (x)k is S-or-µ-measurable.

• If f, g are measurable real-valued functions, then max(f, g) and min(f, g) are mea-
surable.

• If f, g are measurable, f is real-valued and g is B-valued, then f g is measurable.

Let M(X, S, B) denote the vector space of S-measurable functions from X to B.

Let M(X, S, µ, B) denote the vector space of µ-measurable functions from X to B.

Note: If {fn } is a sequence of SMFs, and if fn → f pointwise, then the closure of the
range of f contains a countably-dense subset.

Def: Let (X, T ) be a topological space. A subset A ⊆ X is separable if it contains a

countably-dense subset, or its closure contains a countably-dense subset.

Then if f is measurable, range(f ) is separable.

Proposition: If {fn } is a sequence of functions such that range(fn ) is separable for all n,
and if fn → f pointwise, then range(f ) is separable.
S 
∞
Proof: Let E = n=1 range(fn ) . Then clearly E is separable, and furthermore
range(f ) ⊆ E, so range(f ) is separable. 

Observe that if f is a B-valued SMF, and if O ⊆ B is open with 0 6∈ O, then f −1 (O) ∈ S.

Lemma: Let {fn } be a sequence of B-valued functions, and assume that each fn has the
property that fn−1 (O) ∈ S for every open O ⊆ B, 0 6∈ O. If fn → f pointwise, then f has
this same property.

Proof: Let x ∈ X. Then x ∈ f −1 (O) iff f (x) ∈ O.

For each n, let On = {b ∈ O : distance(b, Oc ) > 1/n}.
Observe that On ⊆ On+1 . Then f (x) ∈ O iff there exists n such that f (x) ∈ On , and there
T −1
is K > 0 such that for all k ≥ K, fk (x) ∈ On iff there exist n, K such that x ∈ fk (On )
k≥K
fk (On ) ∈ S. This implies that x ∈ f −1 (O) iff x ∈ fk−1 (O), thus
S S T S S T
iff
n∈N K∈N k≥K n∈N K∈N k≥K
f −1 (O) ∈ S. 

Corollary: If {fn } is a sequence of measurable functions and fn → f pointwise, then for

any open O ∈ B with 0 6∈ O, we have f −1 (O) ∈ S. That is, if f is S-measurable, then f
satisfies the above property.
5.2. TOWARDS L1 45

Theorem: Let f : X → B satisfy (i) for any open O ∈ B, 0 6∈ O, we have f −1 (O) ∈ S,

(ii) range(f ) is separable. Then f is S-measurable.

Proof: Let {bn }∞ n=1 be a sequence of points in 

B that is dense in the range of f.
For i, j, let Cj,i = {x : kf (x) − bj k < 1/j} = f −1 B1/j (bi ) \ {0}, where f (x) =
6 0. Order
the pairs (i, j) lexicographically, i.e.,
(
j < k, or
(j, i) ≤ (k, n) if .
j = k and i < n

This is a total order. Given n, consider the pairs (i, j) ≤ (n, n), and disjointize the Cji ’s.
Set Eji = Cji \ {Ck` : (j, i) < (k, `) ≤ (n, n)} ⊆ S.
S
Pn
Let fn = j,i=1 bi χEji be a SMF. We claim that fn → f pointwise. Suppose x satisfies
f (x) 6= 0. Let  > 0. Choose j0 such that j0−1 < .
Then choose i0 such that kf (x) − bi0 k < j0−1 > Let N = max(i0 , j0 ), and let n ≥ N. To
show kfn (x) − f (x)k < , given n ≥ N we claim that x ∈ Cj0 ,i0 .
Let (`, k) = max{(j, i) : x ∈ Cji , j ≤ n, i ≤ n}.
Then x ∈ E`k , so kf (x) − bk k < `−1 ≤ j0−1 < .
Therefore fn (x) = bk , so kf (x) − fn (x)k < . 

Before continuing, it is important to introduce and discuss the different modes of conver-
gence of functions.

Egoroff’s Theorem: Let (X, S, µ) be a measure space. Let {fn } be a sequence of

µ-measurable functions. Let E ∈ S, µ(E) < ∞. Assume that fn → f pointwise on E.
Then for any  > 0, there is a subset F ⊆ E, with F ∈ S, such that µ(E \ F ) <  and
fn → f uniformly on F. That is, given δ > 0, there is N such that for n > N, we have
kfn (x) − f (x)k < δ for all x ∈ F.

Proof: For m, k, let Gkm = {x ∈ E : kf (x) − fk (x)k > m−1 } ∈ S.

n = S Gk ∈ Sm.
Let Fm m
k≥n

Fix m. As n → ∞, Fm n ↓ ∅. Since F n ⊆ E, µ(E) < ∞, we have µ(F n ) → 0.

m m
nm ) < 2−m .
Let  > 0. For each m, choose nm such that µ(Fm
∞
. Then µ(E) < , and let F = E \ H, so that µ(E \ F ) < . We claim that
S
Let H =
m=1
{fn } converges uniformly to f on F.
nm0
Let δ > 0. Choose m0 such that m−1 0 < δ. Then for x ∈ F, x 6∈ H, so x 6∈ Fm0 , so for all
k ≥ nm0 , we have kf (x) − fk (x)k ≤ m−1
0 < δ. 

Note: The result does not hold for µ(E) = ∞. Let X = R with Lebesgue measure. Then
fn = χ[n,n+1] converges pointwise to f = 0, but not uniformly.
46 CHAPTER 5. INTEGRATION

Def: Given (X, S, µ) and measurable function {fn }, f, and E ∈ S, we say that {fn }
converges to f almost uniformly on E if for any  > 0, there is F ∈ S, F ⊆ E, such that
µ(E \ F ) < , and fn → f uniformly on F.

Note: Almost uniform convergence implies converges a.e.

Proposition: If {fn } converges to f uniformly on E, then fn → f pointwise except

possibly on a null set.

Proof: For each n, let Fn ⊆ E be such that µ(E \ Fn ) < 1/n, and fm → f uni-
formly on Fn . Let G = ∞ n=1 Fn , so E \ G ⊆ E \ Fn for each n, µ(E \ G) = 0. But fm → f
S

uniformly, hence pointwise, on each Fn so fm → f pointwise on G. 

Def: Let (X, S, µ) be a measure space, and let B be a Banach space. Let E ∈ S.
Let {fn : X → B} be a sequence of µ-measurable functions. We say that {fn } is
almost uniformly Cauchy on E if for any  > 0, there is F ⊆ E such that µ(E \ F ) < ,
and {fn } is Cauchy on F, i.e., for any δ > 0, there is N such that for m, n ≥ N, we have
kfn (x) − fm (x)k < δ for all x ∈ F.

Proposition: If {fn } is almost uniformly Cauchy on E, then there is f on E such that

fn → f almost uniformly.

Proof: Given  > 0, find F ⊆ E with µ(E \ F ) < , and {fn } is uniformly Cauchy on F.
By completeness, there is a limist f, but this depends on f. Take Fn , µ(E \ Fn ) < 1/n, so
we get f defined on all of G := ∞ n=1 Fn . Therefore µ(E \ G) = 0. 
S

Def: Let (X, S, µ) be a measure space, and let {fn } be a sequence of measurable functions,
and let f be a measurable function. We say that {fn } converges in measure to f if for
any  > 0,  
µ {x : kf (x) − fn (x)k ≥ } → 0 as n → ∞.

Proposition: If {fn } converges to f almost uniformly on E, then {fn } converges to f in

measure.

Proof:
 Let , δ > 0.  Choose N ∈ N such that for n ≥ N, we have
µ {x : kf (x) − fn (x)k ≥ } < δ. Let F ⊆ E satisfy

(i) µ(E \ F ) < δ

5.2. TOWARDS L1 47

(ii) fn → f uniformly on F.

Now “update” N so that for n ≥ N, we also kfn (x) − f (x)k <  for all x ∈ F. Then for
n ≥ N, we have
 
{x ∈ E : kf (x) − fn (x)k ≥ } ⊆ E \ F =⇒ µ {x ∈ E : kf (x) − fn (x)k ≥ } .

Proposition: Suppose there are two measurable functions f, g such that fn → f and
fn → g in measure on E. Then f = g a.e.

Proof: Let  > 0. Then for n ∈ N, the set {x ∈ E : kf (x) − g(x)k ≥ } is con-
tained in the set {x ∈ E : kfn (x) − f (x)k ≥ 2 } ∪ {x ∈ E : kfn (x) − g(x)k ≥ 2 }. By
monotonicity
 and subadditivityof measure, it follows that
µ {x ∈ E : kf (x) − g(x)k > 0} = 0, by letting  → 0, since fn → f and fn → g converge
in measure. 

Def: {fn } is Cauchy in measure on E if for any  > 0,

 
µ {x : kfn (x) − fm (x)k ≥ } → 0 as n, m → ∞.

Proposition: Let {fn } be a sequence of SIFs that is Cauchy for k·k1 . Then {fn } is
Cauchy in measure.

Proof: Given  > 0, let A = {x ∈ E : kfn (x) − fm (x)k ≥ }. Then

1
Z Z
µ(A ) = χA (x) dµ(x) ≤ kfn (x) − fm (x)k dµ(x) → 0

as n, m → ∞. 

Riesz-Weyl Theorem: If {fn } is a sequence Cauchy in measure on E, then there is a

subsequence that is almost uniformly Cauchy.

Proof: Define {nk } as follows:

Let n1 = 1. Given nk−1 , consider Em,` = {x ∈ E : kfm (x) − f` (x)k ≥ 2−k }.
Then choose nk such that for m, ` > nk , we have µ(Em,` ) < 2−k and nk > nk−1 . We now
show that {fnk } is uniformly Cauchy.
Let  > 0. Choose K such that ∞ −k < .
P
k=K 2
48 CHAPTER 5. INTEGRATION

S∞
Let F = E \ k=K {x : fnk (x) − fnk+1 (x) ≥ 2−k }.
 
For k ≥ K, we have µ {x : fnk (x) − fnk+1 (x) ≥ 2−k } < 2−k , so
P∞
µ(E \ F ) < k=K 2−k < .
It remains to show that {fnk } is Cauchy on F.
Let δ > 0. Choose N > K such that ∞ −n < δ.
P
n=N 2
Then for j > ` > N, and for x ∈ F,
j−1 j
2−i < δ.
X X
fnj (x) − fn` (x) ≤ fni+1 (x) − fni (x) ≤
i=` i=`+1

Proposition: Let {fn } be Cauchy in measure, and suppose it has a subsequence {fnk }
that converges almost uniformly to some f. Then {fn } converges to f in measure.

Proof: Given  > 0, consider {x ∈ E : kf (x) − fn (x)k > }. Observe that this set
is contained in
{x ∈ E : kf (x) − fnk (x)k > /2} ∪ {x ∈ E : kfnk (x) − fn (x)k > /2}. Since fnk → f almost
uniformly,
 we can choose N such that for k ≥ N, and for δ > 0, we have
µ {x ∈ E : kf (x) − fnk (x)k > /2} < δ/2, and then we can choose N2 ≥ N such that
 
µ {x ∈ E : kfnk (x) − fn (x)k > /2} < δ/2. Thus for n > N2 , we have
 
µ {x ∈ E : kf (x) − fn (x)k > } < δ, 

Let {fn } be a sequence of SIFs, Cauchy for k·k1 . Then {fn } is Cauchy in measure, and it
has a subsequence almost uniformly Cauchy, hence converging to some f almost uniformly.
Finally, fn → f in measure, where f is unique a.e.

Proposition: If {fn } and {gn } are Cauchy for k·k1 , and are “equivalent” in the sense
that kfn − gn k1 → 0, and if fn → f in measure, then so does {gn }.

Proof: Let {hn } be the sequence {f1 , g1 , f2 , g2 , . . .}. Then clearly hn is Cauchy for
k·k1 and thus has a subsequence, namely, {fn }, converging to f in measure, and so hn → f
in measure. It follows that gn → f in measure. 

On the space M(X, S, µ, B) of µ-measurable, B-valued functions, define f ∼ g iff f = g

a.e.

Def: {fn } is mean Cauchy if it is Cauchy for k·k1 .

5.2. TOWARDS L1 49

Lemma: If {hn } is a mean Cauchy sequence converging to 0 almost uniformly, then

khn k1 → 0.

Proof: Let  > 0. Choose N such that for m, n ≥ N, khn − hm k < .

Let E = {x : hN (x) 6= 0}, so µ(E) < ∞. Now find F ⊆ E with µ(F ) 6= 0 so that
µ(E \ F ) < 1+khN k , and {hn } converges uniformly to 0 on F. For given n ≥ N,
∞
Z Z
khn (x)k dµ(x) = khn (x) − hN (x)k dµ(x) ≤ khn − hN k1 < .
Ec Ec

Choose N1 > N so that for n > N1 and x ∈ F, by uniform convergence, khn (x)k < µ(F ) ,

Z Z
so for n > N, khn (x)k dµ(x) ≤ dµ(x) = . Therefore,
F F µ(F )
Z Z Z
khn (x)k dµ(x) ≤ khn (x) − hN (x)k dµ(x) + khN (x)k dµ(x) ≤
E\F E\F E\F


≤ khn − hN k1 + µ(E \ F ) khN k∞ < 2 + · khN k∞ < 4.
1 + khN k∞


Proposition: If {fn } and {gn } are mean Cauchy sequences, and if they both converge to
the a.e.-same function in measure, then {fn } and {gn } are equivalent Cauchy sequences.

Proof: {fn } and {gn } have subsequences that converge almost uniformly, and the
limit is unique a.e. It suffices to show that the subsequences are equivalent Cauchy
sequences. Let hk = fnk − gnk . Then {hk } is a mean Cauchy sequence that converges
uniformly to 0. But by the lemma, it follows that {fnk } and {gnk } are equivalent Cauchy
sequences. 

Theorem: Let f ∈ M(X, S, µ, B). The following are equivalent. There is a mean Cauchy
sequence {fn } of SIFs that converges to f

(i) in measure

(ii) almost uniformly

(iii) almost everywhere

Proof: (i) =⇒ (ii) by Riesz-Weyl. The other two implications are exercises. 
50 CHAPTER 5. INTEGRATION

5.3 The Space L1

Def: Let L1 (X, S, µ, B) be the subset of M(X, S, µ, B) of µ-integrable functions, where

a function is µ-integrable if it satisfies any of properties (i), (ii), (iii) above.

That is, for f ∈ L1 (X, S, µ), there is a mean Cauchy sequence of SIFs that converges to f
in measure, almost uniformly, and almost everywhere.

Def: Let {fn } be a mean Cauchy sequence converging to f. Then the integral of f is
Z Z
f dµ := lim fn dµ.
n→∞

The following propositions have easy proofs.

Proposition: If f, g ∈ L1 (X, S, µ, B), and a ∈ R, then:

Z Z Z
(i) f + g dµ = f dµ + g dµ.
Z Z
(ii) af dµ = a f dµ.

(iii) If f ∈ L1 , then the map x 7→ kf (x)k is in L1 (X, S, µ, R).

Z
(iv) If f ∈ L1 (X, S, µ, R), and if f ≥ 0 a.e., then f dµ ≥ 0.
Z Z
(v) If f, g ∈ L1 (X, S, µ, R), and if f ≥ g a.e., then f dµ ≥ g dµ.

(vi) If f, g ∈ L1 (X, S, µ, R), then kf + gk1 ≤ kf k1 + kgk1 .

(vii) If f ∈ L1 , and {fn } is a mean Cauchy sequence of SIFs converging to f, then kfn k1 →
kf k1 .
Thus, L1 (X, S, µ, B) is complete for k·k1 .
Z Z Z
(viii) If E, F ∈ S and E ∩ F = ∅, then f dµ = f dµ + f dµ.
E∩F E F
Z Z
(ix) If f ∈ L1 (X, S, µ, B), and f ≥ 0, and if E, F ∈ S and F ⊆ E, then f dµ ≤ f dµ.
F E

Proof: Exercise. 
5.3. THE SPACE L1 51

Def: The indefinite integral of f, denoted µf is the function on S defined by

Z Z
µf (E) = f dµ = f χE dµ.
E

In fact, µf is a B-valued measure, i.e., countably additive (proven later).

Proposition: If f ∈ L1 , then C(f ) := {x ∈ X : f (x) 6= 0} is σ-finite.

Proof: Let {fn } be a mean Cauchy sequence of SIFs converging to f a.e. Then
µ({x ∈ X : fn (x) 6= 0}) < ∞ for all n. Let D = ∞ n=1 {x ∈ X : fn (x) 6= 0}, so D is
S

σ-finite. Clearly, C(f ) ⊆ D, except possibly on a null set. 

Thus, we can see that for f ∈ L1 (X, S, µ, B), the indefinite integral µf is indeed a B-valued
measure.

Proposition: If f ∈ L1 (X, S, µ, B), then for any  > 0, there exists E ∈ S such that
Z
µ(E) < ∞ and f dµ < .
X\E
Z
Proof: Find a SIF g such that kf − gk1 = kf (x) − g(x)k dµ(x) < , where
X\E
E = {x : g(x) 6= 0}. 

Proposition: (Absolute continuity) If f ∈ L1 (X, S, µ, B), then for any  > 0, there exists
δ > 0 such that if µ(E) < δ, then kµf (E)k < .

Proof: Choose a SIF g such that kf − gk1 < /2.


Then kf k1 ≤ kf − gk1 + kgk1 . Therefore if µ(E) < , then
2 kgk∞

 
Z Z
kµf (E)k = f dµ ≤ + g dµ < · µ(E) · kgk∞ < .
E 2 E 2


Lebesgue-Dominated Convergence Theorem: Let fn ∈ L1 (X, S, µ, B) for each n,

and suppose fn → f a.e. If there is a R-valued g ∈ L1 such that kfn (x)kB ≤ g(x) a.e. for
all n, then {fn } is a mean-Cauchy sequence.
52 CHAPTER 5. INTEGRATION

Z
Proof: Let  > 0. Find E such that µ(E) < ∞, and g dµ < /6. Then
X\E


Z Z Z
kfn (x) − fm (x)kB dµ(x) ≤ kfn (x)kB +kfm (x)kB dµ(x) ≤ 2g(x) dµ(x) < .
X\E X\E X\E 3

By absolute continuity of µg , choose δ > 0 such that if µ(F ) < δ, then µg (F ) < /6.
By Egoroff’s Theorem, there exists G ⊆ E such that µ(E \ G) < δ, and {fn } converges
uniformly to f on G. Then

Z Z
kfn (x) − fm (x)k dµ(x) ≤ 2 g(x) dµ(x) = 2µg (E \ G) < .
E\G E\G 3

Then choose N such that for n, m ≥ N, we have kfn (x) − fm (x)k ≤ , where x ∈ G.
Z 3µ(G)
Thus, kfn − fm k dµ < /3.
G
Therefore if m, n ≥ N, then kfn − fm k1 < . 

Proposition: Let f be a B-valued µ-measurable function. If there is g ∈ L1 that is

R-valued, with kf (x)k ≤ g(x) a.e., then f ∈ L1 .

Proof: Since f is measurable, there is a sequence {fn } of SMFs converging to f

a.e. Choose increasing En ’s with µ(En ) < ∞, and ∞ n=1 En = {x : g(x) 6= 0}.
S
(
fn (x), if kfn (x)k < 2g(x), x ∈ En
Set hn (x) =
0, o.w.
Since {x : g(x) 6= 0} is σ-finite, we see that the support of hn is measurable, thus hn is a
SIF. Clearly hn → f a.e., and by definition, khn (x)k < 2g(x). By dominated convergence,
f is mean-Cauchy, hence f ∈ L1 . 

Monotone Convergence Theorem: Let {fn } be a sequence of real-valued

Z
integrable
functions on E with fn ≥ 0. If fn+1 (x) ≥ fn (x) a.e. for all n, and supn fn dµ < ∞, then
fn ∈ L1 , and {fn } is a mean-Cauchy sequence that converges to a finite f ∈ L1 a.e.
Z Z nZ o∞
Proof: fn+1 dµ ≥ fn dµ, so the sequence fn dµ is monotonically in-
Z n=1
creasing, but since supn fn dµ < ∞, it follows that it must converge. For a.e. x,
we
Z have that
Z {fn (x)} is monotonically increasing, so let f (x) = lim fn (x). Clearly,
fn dµ ≤ f dµ for all n.
For the reverse inequality, let φ be any SIF such that φ ≤ f. Let 0 < α < 1, and define
En = {x : fn (x) ≥ αφ(x)}. Then
Z Z Z Z
fn dµ ≥ fn dµ ≥ αφ dµ = α φ dµ.
En En En
5.3. THE SPACE L1 53

Z Z
Note that since fn ↑ f, it follows that also En ↑ E a.e. Thus lim fn dµ ≥ α φ dµ.
Z Z
Now since α ∈ (0, 1) was arbitrary, we have lim fn dµ ≥ φ dµ. Finally, since φ was an
Z Z
arbitrary SIF with φ ≤ f, it is easily seen that lim fn dµ ≥ f dµ. 

Finally, we will show that the indefinite integral of an integrable function is a measure by
establishing countable additivity.

Proposition: Let f ∈ L1 (X, S, µ, B). Then µf is a B-valued measure on X.

Proof: We will show countable additivity.

Let E = ∞ n=1 . Let  > 0. There exists a sequence of SIFs {fn } converging in mean to f.
L

Then choose N such that for n ≥ N, we have kfn − f k1 < .

Note that for SIFs fn , µfn is clearly a measure:
∞
X m
X
µfn (E) = µfn (Ej ) = lim µfn (Ej ).
m→∞
j=1 j=1

Then for m > N,

∞
X ∞
X
µf (E) − µf (Ej ) ≤ kµf (E) − µfm (E)k + µfm (E) − µf (Ej ) .
j=1 j=1

We have the bounds

Z
• kµf (E) − µfm (E)k = f − fm dµ ≤ kf − fm k1 < ,
E

∞
X ∞
X ∞ Z
X
• µfm (E) − µf (Ej ) = [µfm (Ej ) − µf (Ej )] = lim fm − f dµ =
m→∞
j=1 j=1 j=1 Ej
Z
= lim Lm fm − f dµ ≤ kfn − f k1 < .
m→∞ Ej
j=1
P∞
It follows that µf (E) = j=1 µf (Ej ), as desired.