Measure theory

J. Palacios
IMCA-UNI, Peru.
[email protected]

September 3, 2020

Abstract
This is a fundamental course on measure theory.

Contents
1 Introduction 2
1.1 Boolean algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Semirings and semialgebras of subsets 4

3 Measurable spaces 4

4 Measurable mappings 5
4.1 Categorical point of view . . . . . . . . . . . . . . . . . . . . . . . . . 7

5 Measure space 7
5.1 Almost everywhere property . . . . . . . . . . . . . . . . . . . . . . . 9
5.2 Completeness and regularity . . . . . . . . . . . . . . . . . . . . . . . 10

6 Integration 12

7 Integrable functions 17

8 Lp -spaces 20
8.1 L ∞
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

9 Modes of convergence 26
p
9.1 Convergence in L . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
9.2 Convergence in measure . . . . . . . . . . . . . . . . . . . . . . . . . 27
9.3 Almost uniform convergence . . . . . . . . . . . . . . . . . . . . . . . 29

10 Exterior measure 32

11 Haar measure (existence) 34

1
 12 Decomposition of measures 37

13 Riesz representation for Lp 43

14 Riesz representation (continuous case) 48

15 Product measures 52

16 Change of variable in Rn 57

17 Properties of Haar measure 60

18 Sobolev spaces 62

19 Introduction to probability theory 64

1 Introduction
intro

1.1 Boolean algebras

Let R be a ring. We recall that an element a ∈ R is idempotent if a2 = a.
Definition 1. A Boolean ring (resp. algebra) is a ring (resp. unitary ring) such
that every element is idempotent.
Example 2 (Boolean algebra of subsets). Let X be a set. The set R = 2X is a
Boolean algebra with the operation of symmetric difference and intersection of sets
as the sum and multiplication respectively. Indeed, if A ∈ R then A2 = A ∩ A = A.
One easily show that (R, ∆, ∩) is a commutative unitary ring with 1R = X, as one
sees that X ∩ A = A for all A ∈ R.
Definition 3. A subring (resp. unitary subring) of the Boolean ring of subsets of
X (resp. aalgebra) is called ring (resp. algebra) of subsets of X
Proposition 4. Let R be a Boolean ring. Then
(a) R has characteristic 2.
(b) R is commutative.
Proof. (a). We have
2x = (2x)2
= 4x2
= 4x ,
from which we get 2x = 0. (b). We have
x + y = (x + y)2
= x2 + xy + yx + y 2
= x + xy + yx + y ,

2
 Examen de entrada:
1) Probar R=2^X es un algebra de Boole.
2) Probar
3) which we get xy + yx = 0. Hence yx = −xy = xy.
from
4) ........ Entregar en PDF
Proposition 5. Let A a nonempty subset of 2X . The following are equivalent:
(a) A is a ring of subsets of X.
(b) A is closed with respect to finite unions and set differences.
(c) A is closed with respect to finite unions and proper set differences.
(d) A is closed with respect to finite disjoint unions, (nonempty) finite intersec-
tions and proper set differences.
(e) A is closed with respect to finite unions, (nonempty) finite intersections and
set differences.
Proof. (a) ⇒ (b). If A is a Boolean subring of subsets of X, then A is closed with
respect to symmetric differences and intersections. For any pair of subsets A and B
of X, the identities:

2) A ∪ B = (A 4 B) 4 (A ∩ B)
3) A − B = A 4 (A ∩ B) ,

allow us to prove what we want.

(b) ⇒ (c). It is immediate.
(c) ⇒ (d). We apply the identity

4) A ∩ B = (A ∪ B) − (A 4 B) .

(d) ⇒ (e). Notice that empty set is an element of A . Notice that A−B = A−(A∩B)
and A ∩ B ⊂ A. Now, if A ∩ B = A then A − B = ∅ ∈ A ; if A ∩ B ( A then
A − (A ∩ B) is a proper set difference so it belongs to A . On the other hand, the
identity
A ∪ B = (A − B) + (A ∩ B) + (B − A)
shows that A is closed with respect to finite unions.
(e) ⇒ (a). It is clear from the definition that A is closed with respect to
symmetric differences.
boolsubalgx Corollary 6. Let A a nonempty subset of 2X . The following are equivalent:
(a) A is an algebra of subsets of X.
(b) A 3 X, A is closed with respect to finite unions and set differences.
(c) A 3 X, A is closed with respect to finite unions and proper set differences.
(d) A is closed with respect to finite disjoint unions, finite intersections and proper
set differences.
(e) A is closed with respect to finite unions, finite intersections and set differences.
boolsubalgxc Corollary 7. Let A a nonempty subset of 2X . The following are equivalent:

3
 (a) A is an algebra of subsets of X.

(b) A is closed with respect to finite unions and set complements.

(c) A 3 X and A is closed with respect to set differences.

Proof. (a) ⇒ (b). It is a particular case of Corollary 6 (b).

(b) ⇒ (c). It follows from the identities X = ∅c and A − B = (Ac ∪ B)c .
(c) ⇒ (a). In view of Corollary 6 (b), it follows from the identities

A ∩ B = A − (A − B) ,
A ∪ B = X − ((X − A) ∩ (X − B)) .

2 Semirings and semialgebras of subsets

Definition 8. A semiring of subsets of a set X is a subset S of 2X such that

(i) ∅ ∈ S .

(ii) A, B ∈ S implies A ∩ B ∈ S .

(iii) For any pair A, B ∈ SP

, there exist a finitely many disjoint subsets C1 , . . . Cn ∈
S such that A − B = ni=1 Ci .

If moreover X ∈ S , then S is called semialgebra.

Example 9. Any ring of subsets is a semiring of subsets. It is immediate as a ring

of subsets is closed with respect to set differences, see Corollary 6 (b).

Example 10. For X = Rn , the set S of cells j=1 haj , bj ], where −∞ ≤ aj ≤ bj ≤

Q
∞.

3 Measurable spaces
Definition 11. Boolean subring (resp. subalgebra) A of subsets of a set X is called
σ-ring (resp. σ-algebra) of subsets of X if A is closed Swith respect to countable
unions, i.e. if for any sequence of sets An in A one has ∞n=1 An ∈ A . An element
of a σ-algebra A is called A -measurable subset, or simply measurable subset of X

Corollary 12. Let A be a subset of 2X . The following are equivalent:

(a) A is a σ-algebra.

(b) A 3 X, A is closed with respect to complements and countable unions.

Proof. Use Corollary 7.

Lemma 13. Arbitrary intersections of σ-rings (resp. σ-algebras) is a σ-ring (resp.

σ-algebra).

4
 Definition 14. For a subset S ⊂ 2X , we denote by hSi the intersection of all σ-rings
(resp. σ-algebras) that contains S, and call it the σ-ring (resp. σ-algebra) generated
by S.
Definition 15. A measurable space is a pair (X, A ) formed by a set X and a
σ-algebra of subsets of X.
topological space (X,T )
Example 16. (measurable spaces)
1. For a set X, A = 2X is a σ-algebra and (X, A ) is a measurable space.

2. Let (R, τ ), where τ is the usual topology. Then B(R) =< τ > is called the σ-
alg of Borelian sets of R. Then (R, B(R)) is a meas. sp. It can be generalized,
taking (X, τ ) a top space and B(X) =< τ >. Then (X, B(X)) is a meas. sp.
B(X) sigma-algebra of Borelian sets of X
4 Measurable mappings
Definition 17. Let (X, A ) and (Y, B) be two measurable spaces. A mapping
f : X → Y is (A , B)-measurable (or simply measurable) if f −1 (B) ∈ A for all
B ∈ B.
Lemma 18. Let (X, A ) be a measurable space and let Y a topological space. A
mapping f : X → Y is (A , B(Y ))-measurable if and only if f −1 (V ) ∈ A for all V
open subset of Y .
Proof. The sufficiency is clear. For the necessity, consider the set S of elements
B ∈ B(Y ) such that f −1 (B) ∈ A . By hypothesis, S contain all open subsets of
Y . It is clear that S is a σ-algebra, so B(Y ) ⊆ S , i.e. S = B(Y ). Therefore, f
is (A , B(Y ))-measurable.

Para Lemma 19. Let (X, A ) be a measurable space and let Y a second countable topo-
logical space with sub-base V. A mapping f : X → Y is (A , B(Y ))-measurable if
despues and only if f −1 (V ) ∈ A for all V ∈ V.

Definition 20. The extended real number line is the set R̄ := R ∪ {−∞, +∞}.
Topology of R̄: base: <a,b>, a<b reals
Base= {< a, b >, [−∞, b >, < a, +∞]}
Definition 21. Let (X, A ) be a measurable space.
1. An (A , B(R))-measurable mapping f : X → R will be called A -measurable
function.

2. An (A , B(R̄))-measurable mapping f : X → R̄ will be called extended A -

measurable function.
Example 22. 1. Let X, Y be two topological spaces and let f : X → Y be a
continuous mapping. The f is (B(X), B(Y ))-measurable.
The following proposition shows that our definition of measurable functions co-
incide with the definition in several classic books.

5
Proposition 23. Let (X, A ) be a measurable space and f : X → R. The following
are equivalent:

(a) f is a measurable function.

(b) for all α ∈ R, f −1 (h−∞, αi) ∈ A . Bartle

... says f is (A , B(R))-meas iff f −1 (I) ∈ A for all open set
Proof. Indeed, Lemma 18
I of R. Base of the top of R : {< a, b >; a < b}. All open set of R is countable
union of intervals.

Question 24. Is it possible associate a meas space to a top space in a natural way?

Why not to take the generated topology?

Suppose that X is a set and S ⊂ 2X ⇒ σ-alg < S >. It is similar to topology
generated by S.

Question 25. It is possible to know explicitly the elements of < S >?

Given a top sp (X, τ ) ⇒ B(X) =< τ >. In algebra, it is possible

(linear combination of
Now, given a meas sp (X, A )
elements of S)
Question 26. What can you say of the top generated by A ?

M (X, A ) = {f : X → R; f is measurable }.
M̄ (X, A ) = {f : X → R̄; f is measurable }.

Proposition 27. M (X, A ) is an algebra over R.

Proof. f, g ∈ M (X, A ), (f + g)(x) = f (x) + g(x), x ∈ X,

(f.g)(x) = f (x).g(x), x ∈ X,
(α.f )(x) = α.f (x), x ∈ X, α real

Claim: f + g, f.g, α.f are measurable .

Let α real , (f + g)(x) > α ∈ f (x) > α − g(x). By Arq. property, there is a
rational r st
f (x) > r > α − g(x) iff f (x) > r and g(x) > α − r.
Then {x; (f + g)(x) > α} = ∪r rat {x, f (x) > r, g(x) > α − r} ∈ A , because f, g
are meas.
Claim: f.g is measurable. f+g is meas
2
Use that f is meas .
f.g = [(f + g)2 − (f − g)2 ]/4 is meas.
Let f : X → R meas. Define f + (x) := sup{f (x), 0}, f − (x) := sup{−f (x), 0}.
f = f + − f − , |f | = f + + f − .

Proposition 28. The functions f + , f − , |f | are measurable.

Ejerc

6
 4.1 Categorical point of view
Spaces Maps
Topology topological spaces continuous map
Measure theory measurable spaces measurable map

Category theory:

Definition 29. (informal) A category C is the following data:

1. objects

2. morphisms (arrows), they are assoc (with respect to the composition), there
are identity morph.

Example 30. 1. C = VectK (K is a field). Obj=K-vector sp. Mor=linear trans.

2. C = Top, Obj=top sp. Morf=continuos maps.

3. C = Meas, Obj=meas spaces. Morf=meas maps.

Example 31 (The category of measurable spaces Meas). Let X, Y, Z ∈ Meas, and

given two measurable maps f : X → Y , g : Y → Z. Then g ◦ f is a meas map.
Indeed, if C ∈ AZ , then (g ◦ f )−1 (C) = f −1 (g −1 (C)) ∈ AX . Observe that idX is
measurable.

Definition 32. (informal) Given two categories C and D. A functor from C to D

is a correspondence F : C → D, X 7→ F (X) (objects), f 7→ F (f )(morphisms)
such that
F (f ◦ g) = F (f ) ◦ F (g), F(id X)=idF (X) .

Example 33. C = VectK , F : C → C, V → F (V ) = V ?? , T → F (T ) = T ?? , where

T : V → W linear trans.

Example 34 (Functor from Top to Meas). X, Y top sp. f : X → Y cont. f −1 (V ) ∈

τX for all open subset V ∈ τY . Note that τX is contained in B(X). By lemma, f is
measurable. 18
In conclusion, cont ⇒ meas.
We have a functor F : Top → Meas, (X, τX ) 7→ (X, B(X)), f 7→ f .
Question: Is it possible construct a functor from Meas to Top in a natural way?
5 Measure space disjoint union
Definition 35. Let A be nonempty subset of 2X . APfunction µ : A → h−∞, Pn +∞] is
finitely additive, if A1 , . . . , An ∈ A are disjoint with i=1 Ai ∈ A then µ( i=1 Ai ) =
n
n
{Ai }∞
P
i=1 µ(A i ). We say that µ is σ-additive, ifP n+1 is a family of disjoint subsets in
∞ P∞ ∞
A with i=1 Ai ∈ A then µ( i=1 Ai ) = i=1 µ(Ai ).
P

Definition 36. Let S be a semiring of subsets. A function µ : S → h−∞, +∞]

is a finitely additive measure, if µ ≥ 0, µ(∅) = 0 and µ is finitely additive. If µ is
moreover σ-additive, then it is called a measure.

7
 The previous definition is also applied when S is a semialgebra, ring, algebra,
σ-ring, σ-algebra of subsets.

Definition 37. Let µ : A → [0, ∞] be a measure.

1. µ is called finite if it takes real values, i.e. µ(A ) ⊆ R.

2. µ is called σ-finite if there is a sequence {An } in A such that. µ(An ) < +∞

and X = ∪n An .

Example 38. Let (X, A ) be a measurable space. The function µ : A → [0, ∞]

defined by (
card(A), A is finite
A 7→ µ(A) =
+∞, A is infinite
is a measure called counting measure
P (X, A ). Let {An } be a sequence in A .
on P
We want to prove that µ ( n An ) = n µ(A Pn ). P
Suppose that X = N. We have always µ ( n An ) 6 n µ(An ).
Cases:
P
All An are finite . Then µ(An ) = card(An ) for all n. By set theory, card ( n An ) =
1. P
n card(An ).
P
a) n card(An ) < ∞.
P
b) n card(An ) = ∞.
P P P
2. Some An is infinite. ⇒ n An is infinite ⇒ µ ( n An ) = ∞ = n µ(An ).

Example 39. Let (X, A ) be a nonempty measurable space. Let x be an element

of X. The function δx : A → [0, ∞] defined by
(
1, x ∈ A
A 7→ δx (A) =
0, x 6∈ A
P
is
P a measure called Dirac measure at x. Indeed, we want to prove that δ x ( n An ) =
n xδ (A n ).
Cases: P
1. x ∈ n An ⇒ x ∈ An for some n. Then δx (An ) = 1, and is 0 for the others.
Hence P P
δx ( n P An ) = 1 = δx (An ) = n δx (An ). P P
2. x 6∈ n An ⇒ x 6∈ An for all n. Then δx ( n An ) = 0 = n δx (An ).

Example 40. 1. The Lebesgue measure on R is the unique measure λ : B(R) →

[0, ∞] such that λ((a, b)) = b − a for every open interval (a, b) of R.

2. Fix a real number γ > 0. The function λγ : B(R) → [0, ∞] defined by λγ = γ·λ
is a measure.

3. Let F : R → R be a nondecreasing function. There exists a unique measure

λF : B(R) → [0, ∞] such that λF ((a, b)) = F (b− ) − F (a+ ) for every open
interval (a, b) of R. This is called the Borel-Stieltjies measure generated by F .

8
Definition 41. A measure space is a triplet (X, A , µ) formed by measurable space
(X, A ) and a measure µ : A → [0, ∞].

Example 42. The measure space (R, L , λ), where L is the collection of measurable
sets of real numbers and λ is the Lebesgue measure.

Example 43. The measure space (R, B, λ), where B is the collection of Borel sets
of real numbers and λ is the Lebesgue measure.

Definition 44. A measure space (X, A , µ) such that µ(X) = 1 is called probability
space, and µ is called probability measure.

Example 45 (Probability spaces). 1. (X, A , δx ), where x ∈ X.

2. (X, A , µ), where X is finite and µ is the counting measure. Define P(A) =
µ(A)/µ(X). Then (X, A , P) is a probability space.
Rb R +∞ R +∞ √
3. (R, B(R), µ), where µ((a, b))= a e−x2 dx/ −∞ e−x2 dx. Here −∞ e−x2 dx = π.

4. (X, 2X , µ), where X = {c, s}, µ({c}) = 1/2, µ({s}) = 1/2, µ(X) = 1.

5. (X, 2X , µ), where X = {c, s}, µ({c}) = p, µ({s}) = q, µ(X) = 1. Here p, q >
0, p + q = 1.

Proposition 46. Let (X, A , µ) be a measure space.

(a) If A ⊂ B, then µ(A) 6 µ(B).

(b) µ(B − A) = µ(B) − µ(A), if A ⊂ B, µ(A) < ∞.

(c) Let {An } increasing seq . Then µ(∪n An ) = limn µ(An ).

(d) Let {An } decreasing seq . Then µ(∩n An ) = limn µ(An ).

Proof. (a). B = A q (B − A). Then µ(B) = µ(A) + µ(B − A) > µ(A), beacause
µ(B − A) > 0.
(b). Since µ(A) < ∞, from above, we have µ(B) − µ(A) = µ(B − A).
(c). If m 6 n, Am ⊂ An ⇒ µ(Am ) 6 µ(An ) 6 µ(∪n An ). P
Let B1 = A1 , BP n = An − A Pare disjoints and n Bn = ∪n An . Then
Pn−1 . The Bn ’s
µ(∪n An ) = µ ( n Bn ) = n µ(Bn ) = n (µ(An ) − µ(An−1 )) = limn µ(An ). The
last equality follows from the telescopic property.
(d). It is similar to the previous item.

5.1 Almost everywhere property

Definition 47. Let (X, A , µ) be a measure space. We say that a certain proposition
holds µ-almost everywhere if there is a subset N ∈ A with µ(N ) = 0 such that the
proposition holds on X − N .

Example 48. 1. Two functions f, g : X → R are equal µ-almost everywhere if

there is N ∈ A with µ(N ) = 0 such that f (x) = g(x) for all x ∈ X − N .

9
 2. f, g two functions, f 6 g µ-a.e. if ∃N with µ(N ) = 0 such that f (x) 6 g(x)
∀x ∈ X − N . Here the property is: x; f (x) 6 g(x)

3. Let f : Rn → R dif. A point a ∈ Rn is a critical point if f 0 (a) = 0. A point is

regular if it is not critical.
Then f is regular λ-a.e. (λ=Lebesgue) by Sard’s theorem: If N=crit points
and , then λ(N ) = 0.

4. (Counting measure) The property µ-a.e. means every point.

5. Let f : [a, b] → R bounded, N={p; f is not cont in p}. Then f is integrable iff
λ(N ) = 0.

6. Let fn : X → R seq. fn → f µ-a.e iff the set of points x ∈ X st fn (x) does

not converge to f (x) has measure zero .

Remark 49. (Borelian+Leb) density vs µ-a.e.

Consider X = [a, b], D = Q ∩ [a, b] dense in X.
Claim: µ(X − D) = µ(X) − µ(D) = b − a − 0 = b − a > 0.
Thus, density doesnt imply µ-a.e.

Example: K=Cantor (compact), µ(K) = 0. Note that R − K is open, in part,

R − K is not dense.
µ-a.e doesnt imply density

Example 50. Let f, g : R → R functions, let µ = ni=1 δxi . Prove that f = g

P
µ-a.e. iff f (xi ) = g(xi )∀i = 1, . . . , n.
Sol: Let N ⊂ R. µ(N ) = 0 xi 6∈ N for all i.
Suppose f = g µ-a.e. There is such N with µ(N ) = 0, st f = g in R − N . Note
xi ∈ R − N . In part, f (xi ) = g(xi )∀i = 1, . . . , n.
Recip. if f (xi ) = g(xi )∀i = 1, . . . , n. Take N = R − {x1 , . . . , xn }. We have
µ(N ) = 0. In part, f = g µ-a.e.

Example 51. a) Let f : R → R function st f (x) = 1 if x ∈ Q, f (x) = 0 if x 6∈ Q.

Prove that f is discontinouos a.e.
b) Let f : R → R function st f (x) = 1/q if x = p/q ∈ Q, f (x) = 0 if x 6∈ Q.
Prove that f is continouos a.e.

5.2 Completeness and regularity

Definition 52. Let (X, A , µ) be a measure space. A subset B ⊆ X is µ-null or
µ-negligible if there is A ∈ A such that µ(A) and B ⊂ A.
We say that µ is complete if for each A ∈ A such that µ(A) and each B ⊂ A,
one has B ∈ A , i.e. µ-null subset of X is A -measurable.

6 Integration
Definition 53. A real or complex function on a set X whose image consists of only
finitely many elements is called simple function.

10
 Let A ⊂ X, the characteristic function χA on A is defined as follows,
(
1, x ∈ A
x 7→ χA (x) =
0, x 6∈ A

Lemma 54. A function ϕ is simple if and only if it can be uniquely represented as

a linear combination of characteristic functions,
n
X
ϕ= ai χAi ,
i=1

where the Ai ’s form a partition of X and the ai ’s are distinct scalars. Suppose that
(X, A ) is a measurable space, then ϕ is measurable if and only if Ai ∈ A for all
i = 1, . . . , n.
Proof. Notice that any linear combination of characteristic functions is a simple
function. Reciprocally, if ϕ is a simple function with P
values a1 , . . . , an , then take
Ai = ϕ (ai ) for i = 1, . . . , n. Thus we obtain ϕ = ni=1 ai χAi . Moreover, ϕ is
−1

measurable if and only if Ai = ϕ−1 (ai ) ∈ A .

We denote by S the set of all simple real-valued A -measurable functions and
by S+ the set of nonnegative functions in S .
Lemma 55. S is a vector subspace of M (X, A ) and S+ is a convex cone of S .
ldensity Lemma 56. For an element f ∈ M̄ + (X, A ), there exists a sequence (ϕn ) in S+
that converges pointwise to f and 0 6 ϕn 6 ϕn+1 for all n > 1. In consequence, S+
is dense in M̄ + (X, A ).
Proof. Fix a natural number n > 1. For each k ∈ {0, . . . , n2n − 1}, we set

Ekn = {x ∈ X : k2−n 6 f (x) < (k + 1)2−n } ,

and for k = n2n we set Ekn = {x ∈ X : n 6 f (x)}. Note that {Ekn : k = 0, . . . , n2n }
is a collection of measurable sets and forms a partition of X. Hence, we set
n2n
X
−n
ϕn = 2 kχEkn
k=1

The sequence (ϕn ) is increasing and converges pointwise to f . Indeed, given a point
x ∈ X, say x ∈ Ekn for some k. Then ϕn (x) = k2−n . For ε > 0,we take a natural
number N such that 2−N < ε. for n > 0, we have

|ϕn (x) − f (x)| = |k2−n − f (x)| < 2−n 6 2−N < ε

Definition 57. Suppose that ϕ = ni=1 ai χAi in S+

P

Z X n
ϕdµ := ai µ(Ai )
X i=1

11
 R
R ∈ R̄ as µ may take +∞ as
Notice that X ϕdµ R its value.
We have a map X : S+ → R̄ given by ϕ 7→ X ϕdµ.
Proposition 58. The map X : S+ → R̄ is R-linear.
R

Goal: We want to extend the simple functions to integrable function.

For the good definition of the integral for simple functions, we need the following
arithmetic operations
a.(+∞) = +∞, a 6= 0.
0.(+∞) = 0
(+∞) + (+∞) = +∞
a + (+∞) = +∞,
P P R P P
Remark 59. If ϕ = i ai χAi = j bj χAj . Then X ϕdµ = i ai µ(Ai )= j bj µ(Bj ).
(Exerc)
Proposition 60. The map X : S+ → R̄+ is R+ -linear.
R
P R
RProof. 1) Product by scalar: ϕ = i ai χ A i α ∈ R+ . We want α X ϕdµ =
X
αϕdµ.
X
αϕ = αai χAi
i
R P R P P R
By def: X
ϕdµ = i ai µ(Ai ). Hence, α X
ϕdµ = α i ai µ(Ai ) = i αai µ(Ai )= αϕdµ.
P
2) Sum. φ = j bj χBj . Then
X
ϕ+φ= (ai + bj )χAi ∩Bj
i,j

Hence,
Z X X X X X X X
(ϕ+φ)dµ = (ai +bj )µ(Ai ∩Bj ) = ai µ(Ai ∩Bj )+ bj µ(Ai ∩Bj ) = ai µ(Ai ∩Bj )+ bj
X i,j i,j i,j i j j

X X Z Z
= ai µ(Ai ) + bj µ(Bj ) = ϕdµ + φdµ
i j X X

R
Lemma 61. Let ϕ be a simple function > 0. Define λ(A) = ϕdµ for all A. Then
λ is a measure.
Definition 62. Let f : X → R̄ nonnegative function. Define Φf as the set of simple
functions ϕ ∈ S+ st 0 6 ϕ 6 f .
The integral of f is: Z Z
f dµ = sup ϕdµ
X ϕ∈Φf X
R R
We write f dµ := X f χA dµ.
A
R
Lemma 63. (a) R X preserves
R order on functions i.e. If f 6 g functions from X
to R̄, then f dµ 6 gdµ.

12
 R R R
(a) X
preserves order on sets i.e. If A ⊂ B subsets of X, then A
f dµ 6 B
f dµ.
Proof. (a). We shall see Rthat Φf R⊂ Φg . Indeed, if ϕ ∈ Φf , then 0 6 ϕ 6 f 6 g.
Thus, ϕ ∈ Φg . Therefore f dµ 6 gdµ. R R
(b). Note that χA 6 χB , then f χA 6 f χB . By a), we have A f dµ 6 B f dµ.
Theorem 64 (Monotone Convergence). If (fn ) is a monotone increasing sequence
of measurable functions in M̄ + (X, A ) that converges pointwise to a function f ∈
M̄ + (X, A ). Then Z Z
f dµ = lim fn dµ

Proof. We have fn 6 fn+1 6 f for all n.

Claim:R f is measurable
R (exerc)
R R R
Then fn dµ 6 fn+1 dµ 6 f dµ , hence limn fn dµ 6 f dµ. Let ϕ ∈ Φf and
let 1 > α > 0 . Consider
An = {x ∈ X : fn (x) > αϕ(x)}
Since fn −αϕ is measurable, An is measurable, Moreover An ⊂ An+1 and X = ∪n An .
Hence Z Z Z Z
αϕdµ 6 fn dµ 6 f dµ 6 fn dµ
An An An
Z Z Z Z
α ϕdµ = α lim ϕdµ 6 lim fn dµ 6 lim fn dµ
n An n An n
R R R
When
R α → 1, we have ϕdµ 6 limn fn dµ. Applying sup, we have f dµ 6
limn fn dµ.
Remark 65. The thesis of theorem of the monotone convergence fails if we do not
assume that (fn ) is a monotone increasing sequence. For example, the sequence
(fn ), where fn = (1/n)χ[0,n] , converges uniformly to f = 0; however one has
Z Z
f dµ 6= lim fn dµ .

: M̄ + (X, A ) → R̄+ is R+ -linear.

R
Corollary 66. The map X

Proof. Let f, g ∈ M̄ + (X, A ). By Lemma 64, there are monotone increasing se-
quences (ϕn ), (ψn ) in S+ converging pointwise to f, g respectively. Then
Z Z
(f + g)dµ = lim (ϕn + ψn )dµ
n
Z Z
= lim ϕn dµ + lim ψn dµ
n n
Z Z
= f dµ + gdµ
R
Similarly one proves that X
preserves product by scalar in R+ .
Lemma 67 (Fatou). If fn : X → [0, ∞] is measurable for all integer n ≥ 1, then
Z   Z
lim inf fn dµ ≤ lim inf fn dµ .
X n→∞ n→∞ X

13
 Proof. Lemma(Fatou). Let (fn ) sequence > 0. Recall lim inf fn .

gm = inf{fm , fm+1 , . . .}

Note gm 6 gm+1 i.e. the seq (gm ) is increasing. We define

lim inf fn := sup gm = lim gm

m m

Observe that if m 6 n then gm 6 fn . Hence,

Z Z
gm dµ 6 fn dµ

Then Z Z
gm dµ 6 lim inf fn dµ
n

Since (gm ) is increasing, we use the MC theorem,

Z Z Z Z
lim inf fn dµ = lim gm dµ = lim gm dµ 6 lim inf fn dµ.
n m m n

Corollary 68. Let f a measurable function > 0. Define a function λ on the σ-alg
of X as Z
λ(A) = f dµ
A
Prove that λ is a measure.
R R
Proof. λ(∅) = 0. ∅ f dµ = f χ∅ dµ = 0, because χ∅ = 0. Suppose that (An ) is a seq
of disjoint measurable sets.
A = A1 ∪ A2 ∪ · · ·
Consider n
X
fn = f.χAn
k=1

This function is a truncation of f on A1 ∪ A2 ∪ · · · ∪ An . Hence,

Z n
Z X n Z
X n Z
X n
X
fn dµ = f.χAk dµ = f.χAk dµ = f dµ = λ(Ak )
k=1 k=1 k=1 Ak k=1

Note that the seq (fn ) is increasing and converges to f.χA . By MC th.
Z Z n
X ∞
X
λ(A) = f χA dµ = lim fn dµ = lim λ(Ak ) = λ(Ak )
n n
k=1 k=1

cfnull Corollary
R 69. Let f a measurable function > 0. Prove that f = 0 µ-a.e. iff
f dµ = 0.

14
Proof. ⇒ Let A = {x ∈ X : f (x) 6= 0}. By hypothesis µ(A) = 0.
RNote thatR f χAc = 0 R R R R
f dµ = f (χA + χAc )dµ = f χP A dµ + f χAc dµ = fP
χA dµ = supϕ ϕdµ
0 6 ϕ 6 f χA . Note that ϕ = k ak χAk , with A = k Ak . Since Ak ⊂ A,
µ(ARk ) = 0. Hence,
P R
φdµ = k ak µ(Ak ) = 0. Then f dµ = 0.

Recip . let An = {x ∈ X : f (x) R> 1/n}. Note that

R A = ∪n A1n .
1
f (x) > χAn .1/n . Hence 0 = f (x)dµ > n χAn dµ = n µ(An ) > 0. Then
µ(An ) = 0 for all n.
Finally, X
µ(A) 6 µ(An ) = 0.
n

Definition 70. Let (X, A ) be a measurable space and let λ, µ be two measures on
(X, A ). We say that λ is absolutely continuous with respect to µ if A ∈ A and
µ(A) = 0, then λ(A) = 0.

Corollary 71. Let (X, A , µ) be a measure space, f ∈ M̄ + (X, A ), and let λ the
measure on (X, A ) defined by Z
A 7→ f dµ .
A
Then λ is absolutely continuous with respect to µ

Proof. Suppose that µ(A) for some A ∈ A . Note that f χA vanishes µ-a.e. By
Corollary 77, we have
Z Z
λ(A) = f dµ = f χA dµ = 0 .
A

Corollary 72. If (fn ) is a monotone increasing sequence of measurable functions

in M̄ + (X, A ) that converges µ-a.e. to a function f ∈ M̄ + (X, A ). Then
Z Z
f dµ = lim fn dµ

Corollary 73 (Beppo Levi’s Theorem). If (fn ) is a sequence of functions in M̄ + (X, A ),

then
∞
Z X ! ∞ Z 
X
fn dµ = fn dµ
i=1 i=1
Pn
Proof. Apply the Monotone Convergence Theorem to the partial sums gn = k fk .

More properties of integrals

fn ↑ f µ-a.e.

15
 Let N measurable µ(N ) = 0 st fn ↑ f on X − N .
fn .χX−N ↑ f.χX−N . By MC Th,
Z Z
f.χX−N dµ = lim fn .χX−N dµ
n

Note f = f.χX−N + f.χN , fn = fn .χX−N + fn .χN . Hence,

Z Z Z
f dµ = f.χX−N dµ + f.χN dµ

Z Z Z
fn dµ = fn .χX−N dµ + fn .χN dµ

Moreover, f.χN = 0, fn .χN = 0 µ-a.e. because µ(N ) = 0. Therefore,

Z Z Z Z
f dµ = f.χX−N dµ = lim fn .χX−N dµ = fn dµ.
n

7 Integrable functions
Definition 74. Let (X, A , µ) be a measure space. A function f in M (X, A ) is
µ-integrable or simply integrable if both the positive and negative parts f + , f − , of f
have finite integrals with respect to µ. In this case, we define the integral of f with
respect to µ to be the number
Z Z Z
f dµ := f dµ − f − dµ
+

We denote by L (X, A , µ, R) the set of µ-integrable functions in M (X, A ).

lssumm Lemma 75. We have (f + g)+ 6 f + + g + and (f + g)− 6 f − + g − .

Proof. Indeed, (f + g)+ = max{f + g, 0}. Since f 6 f + , g 6 g + , we have f + g 6

f + + g + . Then max{f + g, 0} 6 f + + g +
Similarly, (f + g)− 6 f − + g − . Since −f 6 f − , −g 6 g − , we have −f − g 6
f + g − . Then (f + g)− = max{−f − g, 0} 6 f − + g − .
−

Example 76. Let f = idR , g = −f . Observe that f + g = 0, but

(f + g)+ = 0, f + = f.χR+ , g + = g.χR− . We have (f + + g + )(x) = |x| for all x.
Thus (f + g)+ 6= f + + g + .

Proposition 77. L is an R-vector space with the usual operations.

Proof. f, g ∈ L ⇒ fR + g ∈ L.R
Since f, g ∈ L, f + dµ, f −Rdµ, g + dµ, Rg − dµ are Rfinite. By Lemma
R R
R 83, we
have (f +g) 6 f +g . Hence, (f +g) dµ 6 f dµ+ g dµ. Thus (f +g)+ dµ
+ + + + + +

is finite.
− − −
(f + g)− dµ 6
R
R − Similarly,
R − by Lemma 83 we have (f + g) 6 f + g . Hence,
−
R
f dµ + g dµ. Thus (f + g) dµ is finite.

16
 Now, let λ ∈ R.
a)
R λ >+0 R R R
(λf ) dµ = λf + . Hence, (λf )+ dµ = λ f + dµ. Thus (λf )+ dµ is finite.
+ −
R λ <+0. (λf ) R = −−λf
b) R
(λf ) dµ = −λ f dµ. Thus (λf )+ dµ is finite.
Similarly for the negative part.
We have proved that L is a vector subspace of the function space F(X, R).

Theorem 78. f ∈ L iff |f | ∈ L. In consequence,

Z  Z
 
 f dµ 6 |f |dµ
 

Proof:
(⇒) |f | = f + + f − .
+ −
R |f+| = |fR|, |f+| = 0.R −
Note that
Hence, |f | dµ = f dµ + f dµ. It is finite. Thus |f | is integ.
(⇐) f + , f − 6 |f |. From this inequality, we deduce that f is integ.
Moreover,
Z  Z Z  Z  Z  Z Z Z
 f dµ =  f dµ − f dµ 6  f dµ+ f dµ = f + dµ+ f − dµ = |f |dµ
− −
   +
  +
  
       

Therefore, Z  Z
 
 f dµ 6 |f |dµ
 
Suppose that f measurable, g ∈ L with |f | 6 |g|. Prove that f ∈ L.
Sol: |f |+ = |f | 6 |g|. Thus |f | ∈ L. By the previous th. f ∈ L.

Theorem 79 (Lebesgue Dominated Convergence). Let (X, A , µ) be a measure

space. Let (fn ) be sequence in M (X, A ) that converges µ-a.e. to a function f
in M (X, A ). If there is a function g ∈ L (X, A , µ, R) such that |fn | 6 g µ-a.e. for
all n > 1, then f is integrable and
Z Z
f dµ = lim fn dµ .
n

Proof. Using the characteristic functions, we can suppose that fn → f and |fn | 6
g ∈ L. We have −g 6 fn 6 g. There are two cases:
i) −g 6 fn ⇒ g + fn > 0. By Fatou’s lemma,
Z Z Z Z Z
gdµ + f dµ = (g + f )dµ = lim inf (g + fn )dµ 6 lim inf (g + fn )dµ
n n
Z Z 
= lim inf gdµ + fn dµ
n
Z Z
= gdµ + lim inf fn dµ
n

17
 R
Since g is integ, gdµ is finite. From the equation above,
Z Z
f dµ 6 lim inf fn dµ (1) eqm1
n

ii) fn 6 g ⇒ g − fn > 0. Again by Fatou’s lemma,

Z Z Z Z Z
gdµ − f dµ = (g − f )dµ = lim inf (g − fn )dµ 6 lim inf (g − fn )dµ
n n
Z Z 
= lim inf gdµ − fn dµ
n
Z Z
= gdµ − lim sup fn dµ
n

It follows that
Z Z
lim sup fn dµ 6 f dµ (2) eqm2
n

. (2)
R R
From (1) and (2), we have limn fn dµ = f dµ.
Example 80. X = {x1 , . . . , xn } finite, µ counting meas. Let f : X → R function.
Is f measurable? Yes.

Suppose that f > 0. Is f integrable? Yes

Rf = f (xi )χ
P{xi }
f dµ = ni=1 f (xi )µ({xi }) = ni=1 f (xi )
P

In general, f = f + − f − .
Let P = {xi : f (xi ) > 0}, N = {xi : f (xi ) < 0}. X = P ∪ N
f + = f χP
−
Rf = −f RχN +
dµ = ni=1 f (xi )χ
R − P Pn
fPdµ = f dµ− fP P (xi )µ({xi })− i=1 −f (xi )χN (xi )µ({xi }) =
= ni=1,xi ∈P f (xi ) + ni=1,xi ∈N f (xi ) = ni=1 f (xi ).
P

R Pn
Thus f dµ = i=1 f (xi )

L = F(X, R) = Rn . In particular, dim L = n.

Example 81. X = {x1 , . . . , xn } finite, µ =counting meas.(1/n). In this case µ is a
probability measure.
Let f : X → R function. Is f measurable? Yes.

Rf = f (xi )χ
P{xi }
f dµ = ni=1 f (xi )µ({xi }) = 1
Pn
n i=1 f (xi ), f is integrable.
Theorem: continuous => integrable .
Suppose that X is a top space (Borel) R
We have a map C0 (X) → R given by f 7→ X f dµ. This map is a linear function.

18
8 Lp-spaces
Let (X, A , µ) be a measure space and given a real number 1 ≤ p ≤ +∞. We de-
note by L p (X, A , µ, R) (resp. L p (X, A , µ, C)) the set of A -measurable functions
f : X → R (resp. f : X → C) such that |f |p is integrable. In situations that are
valid for both real and complex cases, we will often use L p (X, A , µ) to represent
L p (X, A , µ, R) or L p (X, A , µ, C).
Lemma 82. With the usual operations, L p (X, A , µ, R) is a R-vector space and
L p (X, A , µ, C) is a C-vector space.
Proof. Let f, g ∈ Lp .
Note that |f + g| 6 2 max{|f |, |g|}. (use the tr. ineq)
|f + g|p 6 (2 max{|f |, |g|})p 6 2p (|f |p + |g|p )
Hence Z Z Z 
p p p p
|f + g| dµ 6 2 |f | dµ + |g| dµ

Thus, f + g ∈ Lp .
Let λ ∈ R
|λf |p = |λ|p |f |p . One sees that λf ∈ Lp . Therefore Lp is a R-vector space.
Definition 83 (Norm and seminorm). Let V vector space. Norm N : V → R if
satisfies
1) N > 0
2) N (v) = 0 iff v = 0
3) N (λv) = |λ|N (v)
4) N (u + v) 6 N (u) + N (v)

N is seminorm if it satisfies 1), 3) and 4) and 2’) N (v) = 0 if v = 0.

R
Example 84. Let V = L. Define N (f ) = |f |dµ.
N is a seminorm.
R
1) N (f ) = R|f |dµ > 0 R R
3) N (λf ) = |λfR |dµ = |λ||f
R |dµ = |λ| |f |dµ.
R R
4) N (f + g) = |f + g|dµ 6 |f | + |g|dµ = |f |dµ + |g|dµ.
For each 1 ≤ p < +∞, we define a function

k · kp : L p (X, A , µ) → R
R
given by f 7→ kf kp := ( |f |p dµ)1/p .
Two real number p, q > 1 are conjugate exponents if

1/p + 1/q = 1 .

We also say that 1 and +∞ are conjugate exponents.

Lemma 85. Let p, q > 1 be two conjugate exponents. Then for any pair of non
negative real numbers x and y, we have
xp y q
xy 6 + .
p q

19
 Proof. The inequality is clear when x = 0 or y = 0. We suppose that x and y are
p −q
both positive. Consider the function φ : h0, +∞i → R given by φ(t) = tp + t q .
Differentiating, we obtain
φ0 (t) = tp−1 − t−q−1 .
Notice that t = 1 is the only critical point of φ. Since

lim φ(t) = lim φ(t) = +∞ ,

t→0+ t→+∞

the value of φ at t = 1 is a minimum. Hence,

1 1 tp t−q
1= + = φ(1) 6 φ(t) = + .
p q p q

Now, taking t = x1/q /y 1/p , we have

xp/q y q/p xp−1 y q−1 xp y q

16 + = + ⇒ xy 6 + .
py qx py qx p q

pmeasac Proposition 86. Let (X, A µ) be a measure space and let f : X → [0, +∞] be an
A -measurable function. Then for each real number t > 0, we have
Z Z
1 1
µ ({x ∈ X : f (x) > t}) 6 f dµ 6 f dµ .
t {x∈X:f (x)>t} t

Proof. Let At = {x ∈ X : f (x) > t}. Then the inequalities 0 6 tχAt 6 f χAt 6 f
implies the inequalities
Z Z Z
0 6 tχAt dµ 6 f χAt dµ 6 f dµ .
R
Since tχAt dµ = tµ(At ), the required inequalities follows.

csffinite Corollary 87. Let (X, A µ) be a measure space and let f ∈ L (X, A , µ). Then the
measurable set {x ∈ X : f (x) 6= 0} is σ-finite.

Proof. For each natural number n > 1, let An = {x ∈ X : |f (x)| > 1/n}. By
Proposition 94, we have
Z Z
µ (An ) 6 n f dµ 6 n f dµ .
An
S
Thus each An is finite, and since {x ∈ X : f (x) 6= 0} = n An , the result follows.

8.1 L∞
Definition 88. Let (X, A , µ) be a measure space. A function f in M (X, A ) is
essentially bounded if f is bounded µ-a.e. We denote by L ∞ (X, A , µ) the set of all
essentially bounded A -measurable functions.

20
Lemma 89. With the usual operations, L ∞ (X, A , µ, R) is a R-vector space and
L ∞ (X, A , µ, C) is a C-vector space.

Proof. Let f, g ∈ L∞ . There are N1 , N2 meas. st µ(N1 ) = µ(N2 ) = 0 and constants

M1 , M2 > 0 satisfying f 6 M1 on X − N1 and g 6 M2 on X − N2 . Then f + g 6
max{M1 , M2 } on X − (N1 ∪ N2 ), but µ(N1 ∪ N2 ) 6 µ(N1 ) + µ(N2 ) = 0. Thus
f + g ∈ L∞ .
Product by scalar: (exerc)
Let N ∞
= {f ∈ L∞ : f = 0 µ − a.e.}

Lemma 90. N ∞
is a R-vector subspace of L∞

Proof. Let f, g ∈ N ∞ . There are N1 , N2 meas. st µ(N1 ) = µ(N2 ) = 0 satisfying

f = 0 on X − N1 and g = 0 on X − N2 .
f + g = 0 on X − (N1 ∪ N2 ). Thus f + g ∈ N ∞ .

Let λ ∈ R, f ∈N ∞
⇒ λf ∈ N ∞
.

Definition 91. L∞ = L∞ /N ∞
is a R-vector space.

Let f ∈ L∞ , N ∈ A with µ(N ) = 0. Define

S(N ) = sup{|f (x)| : x ∈ X − N }

kf k∞ = inf{S(N ) : N ∈ A , µ(N ) = 0}.

Take sequence {Nn } st S(Nn ) → kf k∞

Lemma 92. |f | 6 kf k∞ µ-a.e.

T f is bounded on X −
Proof. S Nn . Then |f | 6SS(Nn ) on P
X − Nn . Then |f | 6 kf k∞
on n (X − Nn ) = X − ( n Nn ). But µ ( n Nn ) 6 n µ(Nn ) = 0. Therefore
|f | 6 kf k∞ µ-a.e.

Lemma 93. k − k∞ is a norm in L∞ .

Proof. 1) kf k∞ > 0
2) kf k∞ = 0 iff f = 0 µ-a.e. Supp kf k∞ = 0. Since |f | 6 kf k∞ = 0 µ-a.e. Then
f = 0 µ-a.e.
3) kλf k∞ = |λ|kf k∞
kλf k∞ = inf N,µ(N )=0 supx∈X−N |λf (x)| = |λ| inf N,µ(N )=0 supx∈X−N |f (x)| = λkf k∞
4) kf + gk∞ 6 kf k∞ + kgk∞

kf + gk∞ = inf N,µ(N )=0 supx∈X−N |(f + g)(x)| 6 inf N,µ(N )=0 supx∈X−N (|f (x)| +
|g(x)|) =
= inf N,µ(N )=0 {supx∈X−N |f (x)| + supx∈X−N |g(x)|} = kf k∞ + kgk∞ .

Definition 94. A subset N of X is locally µ-null if for every A ∈ A with µ(A) <
+∞, the set A ∩ N is µ-null.

Lemma 95. One has the following:

21
 (a) Every µ-null subset of X is locally µ-null.
(b) If X is σ-finite, every locally µ-null subset of X is µ-null.
(c) The union of a sequence of locally µ-null subset of X is locally µ-null.
Proposition 96 (Holder’s inequality). Let (X, A µ) be a measure space and let 1 6
p, q 6 +∞ be two conjugate exponents. If f ∈ L p (X, A , µ) and g ∈ L q (X, A , µ),
then f g ∈ L 1 (X, A , µ) and satisfies
Z
|f g|dµ 6 kf kp kgkq .
p q
Proof. By Lemma xy 6 xp + yq ∀x, y > 0. Let f ∈ Lp , g ∈ Lq .
Take x = |f |/kf kp , y = |g|/kgkq .
|f |p |g|q
|f ||g|/kf kp ||g||q 6 +
pkf kpp qkgkqq
Z Z Z
1 1 p 1 1 1
|f g|dµ 6 p |f | dµ + q |g|q dµ = + = 1
kf kp ||g||q pkf kp qkgk q p q
Hence Z
|f g|dµ 6 kf kp ||g||q

In particular f g ∈ L1 .

Proposition 97 (Minkowski’s inequality). Let (X, A µ) be a measure space and let

1 6 p 6 +∞. If f, g ∈ L p (X, A , µ), then f + g ∈ L p (X, A , µ) and satisfies
kf + gkp 6 kf kp + kgkp .
Proof. f, g ∈ Lp ⇒ kf + gkp 6 kf kp + kgkp .
Proof: We have seen that f + g ∈ Lp .
Consider |f + g|p =|f + g||f + g|p−1 6 |f ||f + g|p−1 + |g||f + g|p−1 . Let q =
p/(p − 1) ⇒ p, q are conjugate exponents.
Note |f + g|p−1 ∈ Lq , because |f + g|(p−1)q = |f + g|p . By Holder’s ineq:

Z Z 1/q
p−1 p−1 (p−1)q
|f ||f + g| dµ 6 kf kp ||(f + g) ||q = kf kp |f + g|
Z 1/p.p/q
p
= kf kp |f + g|

= kf kp kf + gkp/q
p

Similarly,
Z Z 1/q
p−1 p−1 (p−1)q
|g||f + g| dµ 6 kf kp ||(f + g) ||q = kgkp |f + g|
Z 1/p.p/q
p
= kgkp |f + g|

= kgkp kf + gkp/q
p

22
Hence,
Z Z
kf + gkpp = p
|f + g| dµ 6 (|f | + |g|)|f + g|p−1 dµ 6 {kf kp + kg||p }kf + gkp/q
p

There two cases:

1) kf + gkp = 0
p−p/q
2) kf + gkp 6= 0. Dividing we have, kf kp = kf + gkp 6 kf kp + kg||p .

Corollary 98. k · kp is a seminorm on L p (X, A , µ).

Proof. V = Lp
R
1/p
Define N (f ) = |f |p dµ .
N is a seminorm.
R
1/p
1) N (f ) = |f |p dµ >0
R p

1/p R p p
1/p R
1/p
3) N (λf ) = |λf | dµ = |λ| |f | dµ = |λ| |f |p dµ = |λ|N (f ).
R
1/p
4) N (f + g) = |f + g|p dµ 6 . . .. This is Minkowski’s inequality.
Let N p (X, A , µ) the subset of f ∈ L p (X, A , µ) such that kf kp = 0.
Definition 99. We denote by Lp (X, A , µ) the quotient vector space L p (X, A , µ)/N p (X, A , µ)
An element of Lp (X, A , µ) is a class [f ] = f +N p (X, A , µ) with f ∈ L p (X, A , µ),
i.e.
The seminorm k · kp on L p (X, A , µ) induces a function on Lp (X, A , µ), denoted
abusively by k · kp .
Therefore, k − kp is a seminorm in Lp .
Lemma 100. k − kp is a norm in Lp .
Proof. k[f ]kp = kf kp well-defined, N (f ) = k[f ]kp
1) N > 0
2) N (f ) = 0 iff [f ] = 0.
Supp k[f ]kp = 0=¿kf kp = 0 ⇒ f ∈ N p ⇒ [f ] = 0.
3) N (λf ) = |λ|N (f )
4) N (f + g) 6 N (f ) + N (g)
Therefore (Lp , k − kp ) is a normed vector space.
Writing abusively, we write f instead of [f ] in Lp .
Theorem 101 (Completeness theorem). (Lp (X, A , µ), k · kp ) is a Banach space for
1 ≤ p ≤ +∞.
Proof. We shall prove that any Cauchy sequence is convergent. Let (fn ) be a Cauchy
seq. For any ε > 0, there is N = N (ε) > 0 such that if m, n > N , we have
kfm − fn kp < ε. Hence Z
|fm − fn |p dµ < εp

For ε = 2−k . There is a subsequence (gk ) of (fn ) st kgk+1 − gk k < 2−k .

Define g = |g1 | + |g2 − g1 | + |g3 − g2 | + · · · . Note that g > 0 is measurable,

23
 g = limn (|g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn |).
g p = limn (|g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn |)p .
By Fatou’s lemma,
Z Z
p
g dµ 6 lim inf (|g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn |)p dµ
n

Z 1/p Z 1/p
p p
g dµ 6 lim inf (|g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn |) dµ
n

By Minkowski,
Z 1/p
p
(|g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn |) dµ 6 kg1 kp +kg2 −g1 kp +· · ·+kgn+1 −gn kp < kg1

R p 1/p
Now, lim inf n kg1 kp + 2−1 + · · · + 2−n = kg1 kp + 1. Then g dµ 6 kg1 kp + 1.
Let A = {x ∈ X : g(x) < +∞}
gχA ∈ Lp . Note µ(X − A) = 0.
Let f = g1 + (g2 − g1 ) + · · · + (gn+1 − gn ) + · · · on A, and f = 0 on X − A.

Claim: f ∈ Lp .
|f | 6 |g1 | + |g2 − g1 | + |g3 − g2 | + · · · + |gn+1 − gn | + · · · + = g µ − a.e.=¿ |f |p 6 g p
µ-a.e. By Dominated C.T. f ∈ Lp .

Claim: (gk ) converges to f in Lp .

We have |f − gk |p 6 2p g p , then f − gk ∈ Lp . By Dominated C.T.
Z Z
0 = (f − f ) dµ = lim |f − gk |p dµ = kf − gk kp .
p
k

Claim: (fn ) converges to f in Lp .

For m > N , k large, we have
Z
|fm − gk |p dµ < εp

By Fatou, R R
kfm − f kp = |fm − f |p dµ 6 lim inf k |fm − gk |p dµ 6 εp .
this implies that (fn ) converges to f in Lp .
Now, let (fn ) be a Cauchy sequence in L∞ . Each fn is essentially bounded. We
can take N ∈ A with µ(N ) = 0 and |fn | 6 kfn k∞ on X − N for n > 1. Since
|fn − fm | 6 kfn − fm k∞ µ-a.e. we can suppose that

|fn (x) − fm (x)| 6 kfn − fm k∞

for all x ∈ X − N . Then (fn ) is uniformly convergent on X − N . Let f (x) =

limn fn (x) if x ∈ X − N , f (x) = 0 if x ∈ N .
Claim: kfn − f k∞ → 0 when n → ∞.
Note first that f is measurable. (exerc).

24
 Example 102. Consider X = N, A = 2N , µ=counting measure.
`p = Lp (N+ , 2N+ , µ)
1) 1 6 p < +∞
Any function f : N+ → R is measurable. f = (an ), where an = f (n).

Note f ∈ Lp iff |f |p dµ = ∞
R P p
n=1 |an | < +∞
In this case
k(an )k = kf kp = ( ∞ p 1/p
P
n=1 |an | )
Therefore ( ∞
)
X
`p = (an ) : |an |p < +∞
n=1

Moreover, (`p , k − kp ) is a Banach.

2) p = +∞.
Observer N st µ(N ) = 0 iff N = ∅.
Note f ∈ Lp iff f= (an ) bounded
In this case
k(an )k∞ = kf k∞ = supn |an |
Therefore
`∞ = {(an ) : (an ) bounded}
Moreover, (`p , k − k∞ ) is a Banach.

9 Modes of convergence
9.1 Convergence in Lp
Proposition 103. Suppose that µ(X) < +∞ and that (fn ) is a sequence in Lp
which converges uniformly to f . Then f belongs Lp and the sequence converges in
Lp to f .

Proof. Since (fn ) converges uniformly to f , for ε > 0, there exists a natural number
N (ε) > 1 such that |fn (x) − f (x)| < ε for all n > N (ε) and all x ∈ X. If n > N (ε),
then Z  1/p
Z  1/p
p p
kfn − f kp = |fn − f | 6 ε = εµ(X)1/p .

Thus (fn ) converges in Lp to f . In particular, (fn ) is a Cauchy sequence in Lp which

is a Banach space, therefore f ∈ Lp .

domconvLp Proposition 104 (Dominated convergence in Lp ). Let (fn ) be a sequence in Lp which

converges µ-a.e. to a real-valued A -measurable function f , let g ∈ Lp such that
|fn (x)| 6 g(x) for all x ∈ X and n > 1. Then f ∈ Lp and (fn ) converges in Lp to
f.

25
 9.2 Convergence in measure
Definition 105. Let (X, A , µ) be a measure space. Let (fn ) be a sequence of real-
valued A -measurable functions on X. The sequence (fn ) converges in measure to
f if
lim µ ({x ∈ X : |fn (x) − f (x)| > ε}) = 0
n→+∞

holds for all real ε > 0.

We say that (fn ) is Cauchy in measure if

lim µ ({x ∈ X : |fm (x) − fn (x)| > ε}) = 0

m,n→+∞

holds for all real ε > 0.

Lemma 106. Uniform convergence implies convergence in measure.

Proof. If (fn ) converges uniformly to f , then the set

{x ∈ X : |fn (x) − f (x)| > ε}

is empty for any sufficiently large n.

Lemma 107. Convergence in Lp implies convergence in measure.

Proof. Let An (ε) = {x ∈ X : |fn (x) − f (x)| > ε}. Then

Z Z
p p
kfn − f kp = |fn − f | dµ > |fn − f |p dµ > εp µ(An (ε)) .
An (ε)

Now, if kfn − f kp → 0 then µ(An (ε)) → 0 as n → +∞.

propcaucm Proposition 108. Let (fn ) be a sequence of real-valued A -measurable functions

which is Cauchy in measure. Then there is a subsequence which converges µ-a.e.
and in measure to a A -measurable function f .

Proof. By definition, we have

lim µ ({x ∈ X : |fm (x) − fn (x)| > ε}) = 0

m,n→+∞

for all real ε > 0. Then for each natural number k > 1, there exists a subsequence
) such that, if Ak = {x ∈ X : |gk+1 (x) − gk (x)| > 2−k } then µ(Ak ) < 2−k .
(gk ) of (fnS
Set Bk = ∞ j=k Aj . Notice that Bk ∈ A and

∞
X ∞
X
µ(Bk ) 6 µ(Aj ) 6 2−j = 2−(k−1) .
j=k j=k

Now, for i > j > k and x 6∈ Bk , one has

|gi (x) − gj (x)| 6 |gi (x) − gi−1 (x)| + · · · + |gj+1 (x) − gj (x)| < 2−(i−1) + · · · + 2−j < 2−(j−1) .
(3) measconvs

26
 T∞
Let B = k=1 Bk . Since {Bk } is a decreasing sequence, we have

µ(B) = lim µ(Bk ) 6 lim 2−k = 0 .

n→+∞ n→+∞

Notice that if x 6∈ B then (gk (x)) is a Cauchy sequence in R, so it is convergent, i.e.

(gk ) is pointwise convergent on X − B. Define
(
lim gk (k), x 6∈ B
f (x) =
0, x∈B

Then (gk ) converges µ-a.e. to f . Taking i → +∞ in (3), we obtain

|f (x) − gj (x)| 6 2−(j−1) 6 2−(k−1)

Thus (gk ) converges uniformly to f on each X − Bk .

Claim: (gk ) converges in measure to f .
Claim: gk converges in measure to f .
We know
µ(Bk ) < 2−(k−1) . Given ε0 > 0. There is k suff. large st 2−(k−1) < min{ε, ε0 }
If j > k, then |f (x) − gj (x)| > ε ⇒ |f (x) − gj (x)| > 2−(k−1)

{x ∈ X : |f (x) − gj (x)| > ε} ⊂ {x ∈ X : |f (x) − gj (x)| > 2−(k−1) } ⊂ Bk

Taking measure

µ({x ∈ X : |f (x) − gj (x)| > ε}) 6 µ(Bk ) < 2−(k−1) < ε0 ∀j > k

Therefore (gk ) converges in measure to f .

corcauchm Corollary 109. Suppose that (fn ) Cauchy in measure. Then (fn ) converges in
measure to a function f .
Proof: By the previous prop, there exists a subseq (gk ) st it converges in meas.
to a function f . Let fnk = gk .
Hence,
|f (x) − fn (x)| 6 |f (x) − fnk (x)| + |fnk (x) − fn (x)|
Given ε > 0, notice that

{x ∈ X : |f (x)−fn (x)| > ε} ⊂ {x ∈ X : |f (x)−fnk (x)| > ε/2}∪{x ∈ X : |fnk (x)−fn (x)| > ε/2}

Taking measure,

µ({x ∈ X : |f (x)−fn (x)| > ε}) 6 µ({x ∈ X : |f (x)−fnk (x)| > ε/2})+µ({x ∈ X : |fnk (x)−fn (x)| > ε/2

From here, we deduce (fn ) converges in measure to f .

Claim: f is µ-a.e. unique

Suppose that (fn ) converges in measure to g. We want that g = f µ-a.e.

27
 We have,
|f (x) − g(x)| 6 |f (x) − fn (x)| + |fn (x) − g(x)|
Hence, given ε > 0, notice that

{x ∈ X : |f (x)−g(x)| > ε} ⊂ {x ∈ X : |f (x)−fn (x)| > ε/2}∪{x ∈ X : |fn (x)−g(x)| > ε/2}

Taking measure,

µ({x ∈ X : |f (x)−g(x)| > ε}) 6 µ({x ∈ X : |f (x)−fn (x)| > ε/2})+µ({x ∈ X : |fn (x)−g(x)| > ε/2})

When n → ∞, one has µ({x ∈ X : |f (x) − g(x)| > ε}) = 0 for ε > 0. In
particular µ({x ∈ X : |f (x) −S g(x)| > 1/n}) = 0 for n > 0. Observe that
{x ∈ X : |f (x) − g(x)| =
6 0} = n {x ∈ X : |f (x) − g(x)| > 1/n}has zero measure.
Therefore f = g µ-a.e.

Proposition 110 (Dominated convergence in measure). Let (fn ) seq in Lp which

converges in measure to f , let g ∈ Lp st |fn (x)| 6 g(x) µ-a.e. Then f ∈ Lp and (fn )
converges in Lp to f .

Proof: By contradiction, suppose that (fn ) does not converge in Lp to f . Then

there exists a subsequence (gk ) of (fn ) and ε > 0 st kgk − f kp > ε for k. Since
(gk ) is a subsequence of (fn ), (gk ) converges in measure to f . Hence there exists
a subsequence (hl ) of (gk ) which converges µ-a.e. and in measure to a function h.
Since hl converges in measure to h and f , h = f µ-a.e. Now, since (hl ) converges
µ-a.e. to f and |hl (x)| 6 g(x) µ-a.e., we have khl − f kp → 0 when l → ∞ (see
Proposition 112)).

9.3 Almost uniform convergence

Definition 111. A sequence (fn ) of real-valued functions is called almost uniformly
convergent to a function f if for all real number ε > 0 there is a set Aε ∈ A with
µ(Aε ) < ε such that (fn ) converges uniformly to f on X − Aε .
We say that (fn ) is almost uniformly Cauchy sequence if for every real number
ε > 0 there is a set Aε ∈ A with µ(Aε ) < ε such that (fn ) is a uniformly Cauchy
sequence on X − Aε , which means that

lim sup |fm (x) − fn (x)| = 0 .

m,n x∈X−Aε

Proposition 112. A sequence (fn ) is almost uniformly convergent if and only if it

is an almost uniformly Cauchy sequence. In consequence, it converges µ-a.e. to the
same function.

Proof. Suppose that (fn ) is almost uniformly convergent to a function f . We have

sup |fm (x) − fn (x)| 6 sup |fm (x) − f (x)| + sup |f (x) − fn (x)|
x∈X−Aε x∈X−Aε x∈X−Aε

28
Proposition 113. If a sequence (fn ) converges almost uniformly to a function f ,
then it converges in measure to f . Reciprocally, if a sequence (fn ) converges in
measure to a function f , then there is a subsequence that converges almost uniformly
to f .
Proof. Suppose that (fn ) converges almost uniformly to f . By definition for a given
ε > 0, there exists Aε ∈ A with µ(Aε ) < ε such thta (fn ) converges uniformly to f
on X − Aε . Given a real number α > 0, for n sufficiently large, we have
{x ∈ X : |fn (x) − f (x)| > α} ⊂ Aε
which shows that (fn ) converges in measure to f .
Claim: (fn ) conv in measure to f then ∃ subseq that converges a.u. to f .
We have seen that ∃ subseq (gk ) of (fn ) which converges µ-a.e. and in measure
to g.
We have that g = f µ-a.e. by Corollary 117.
Claim: (gk ) converges a.u. to f .
∀ε > 0, ∃Aε with µ(Aε ) < ε st (gk ) converges unif to f on X − Aε .

We have µ(B) = limk µ(Bk ) decreasing. Given ε > 0, ∃k such that µ(Bk ) <
−(k−1)
2 < ε. In the proof of Proposition 116, we have
|f (x) − gj (x)| < 2−(k−1) < ε
for x ∈ X − Bk , j > k. Taking Aε = Bk , (gj ) converges unif to f on X − Aε .
i.e. (gk ) converges a.u. to f .
Proposition 114 (Egoroff’s theorem). Let (X, A , µ) be a measure space and let (fn )
be real-valued A -measurable functions on X. If µ is finite and if (fn ) converges µ-
a.e. to a real-valued A -measurable function f , then (fn ) converges almost uniformly
and in measure to f .
Proof. µ(X) < +∞.
Suppose that (fn ) converges µ-a.e. to f .
∃N with µ(N ) = 0 st (fn ) converges to f on X − N . For m, n > 0, let
∞
[
An (m) = {x ∈ X − N : |fk (x) − f (x)| > 1/m}
k=n

Note that An (m) is measurable, and An (m) is decreasing in n. Since fn (x) → f (x)
for all x ∈ X − N ,
∞
\
An (m) = ∅
n=1

Since µ(X) < +∞,

lim µ(An (m)) = 0
n→∞

Given ε > 0, there is km st µ(Akm (m)) < ε/2m . Let Aε = ∞

S
m=1 Akm (m). Note that
Aε is measurable
P∞ and P∞ m
µ(Aε ) 6 m=1 µ(Akm (m)) 6 m=1 ε/2 = ε. Hence, if x 6∈ Aε , then x 6∈
Akm (m),
|fk (x) − f (x)| < 1/m
for all k > km . Thus (fn ) converges a.u. to f .

29
Theorem 115 (Vitali convergence theorem). Let (fn ) seq in Lp , 1 6 p < ∞. Then
(fn ) converges in Lp iff:
(a) (fn ) converges in measure to f .
(b) For each ε > 0 there is set Aε ∈ A with µ(Aε ) < ∞ st if B ∈ A and B∩Aε = ∅,
then Z
|fn |p dµ < εp
B
for all n > 1.
(c) For each ε > 0 there is set Aε ∈ A with µ(Aε ) > 0 st if B ∈ A and µ(B) < Aε ,
then Z
|fn |p dµ < εp
B
for all n > 1.
Proof. ⇒. We know
R (a). kfpn − f kp → 0 when nR → ∞. p
kfn − f kp = |fn − f | dµ. Given ε > 0, |fn − f | dµ < εp for n suff large.
p

(exerc)

Recip. use (b) and apply Minkowski to kfn − fm k, hence use (c) and convergence
in measure (a).
.....
General case

a.e. ks
KS ck ;3 a.u.
KS (4)


Lp +3 µ

Finite measure space

+3
a.e.
KS ks 3; a.u.
KS (5)
ck

#+ 
Lp +3 µ

30
 Dominated converge
+3
a.e.
KS ks 3; a.u.
SK (6)
ck

 +3 #+ 
s{
Lp ks µ

10 Exterior measure
X set
Def. An outer measure on X is a function µ∗ : 2X → [0, +∞] such that
1) µ∗ (∅) = 0
2) (monotony) If A ⊆ B, then µ∗ (A) 6 µ∗ (B)
∗
S
P ∗3) (subadditivity) If {An } is a sequence of subsets of X, then µ ( n An ) 6
n µ (An )

Example 116. Lebesgue outer measure on R: λ∗ : 2R → [0, +∞]

A 7→ λ∗ (A) = inf { i (bi − ai ) : {(ai , bi )} ∈ CA }
P
where CA is the set of all covering {(ai , bi )} of A formed by bounded open inter-
vales.
Lebesgue outer measure on R is an outer measure.
n
Example 117. Lebesgue outer measure on Rn : λ∗ : 2R → [0, +∞]
A 7→ λ∗ (A) = inf { i vol(Ri ) : {Ri } ∈ CA }
P
where CA is the set of all covering {Ri } of A formed by bounded open n-dimensional
intervales or cubes.
Lebesgue outer measure on Rn is an outer measure.

Let µ∗ be an outer measure on a set X. A subset B of X is called µ∗ -measurable

if
µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c )
for every subset A of X.

A Lebesgue measurable subset of Rn is a λ∗ -measurable subset.

Let Mµ∗ the set of all of µ∗ -measurable subsets.

Question: Is Mµ∗ a σ-algebra?
Yes.

Proposition 118. (Caratheodory’s method) Mµ∗ is a σ-algebra, moreover the re-

striction µ∗ : Mµ∗ → [0, +∞] is a measure.

31
Proof. µ∗ (A) = µ∗ (A ∩ ∅) + µ∗ (A ∩ ∅c ), µ∗ (A) = µ∗ (A ∩ X) + µ∗ (A ∩ X c )
Let B ∈ Mµ∗ . µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c ). Then µ∗ (A) = µ∗ (A ∩ B c ) +
∗
µ (A ∩ B cc ). Thus B c ∈ Mµ∗ .
Let (Bn ) seq in Mµ∗ .

µ∗ (A) = µ∗ (A ∩ Bn ) + µ∗ (A ∩ Bnc )

for every subset A of X.

First let us prove that B1 ∪ B2 ∈ Mµ∗ .
Since B1 is µ∗ -measurable, we have

µ∗ (A∩(B1 ∪B2 )) = µ∗ (A∩(B1 ∪B2 )∩B1 )+µ∗ (A∩(B1 ∪B2 )∩B1c ) = µ∗ (A∩B1 )+µ∗ (A∩B1c ∩B2 )

On the other hand, B2 is µ∗ -measurable

µ∗ (A∩(B1 ∪B2 ))+µ∗ (A∩(B1 ∪B2 )c ) = µ∗ (A∩B1 )+µ∗ (A∩B1c ∩B2 )+µ∗ (A∩B1c ∩B2c ) =

= µ∗ (A ∩ B1 ) + µ∗ (A ∩ B1c ) = µ∗ (A)
Thus µ∗ (A) = µ∗ (A ∩ (B1 ∪ B2 )) + µ∗ (A ∩ (B1 ∪ B2 )c ) i.e. B1 ∪ B2 ∈ Mµ∗ .
Assume that (Bn ) is a disjoint
Pn seq.
µ (A) = i=1 µ (A ∩ BSi ) + µ∗ (A
∗ ∗
∩ ni=1 Bic ) for all n > 1.
T
By induction,
µ∗ (A) > ∞ A∩( ∞
P ∗ ∗ c

i=1 µ (A ∩ Bi ) + µ i=1 Bi ) .....(*)
(exerc) S
µ∗ (A ∩ ( ∞ ∗
S∞ P∞ ∗
i=1 Bi )) = µ ( i=1 A ∩ Bi ) 6 i=1 µ (A ∩ Bi )
Hence
∞
!! ∞
!c ! ∞ ∞
!c !
[ [ X [
∗ ∗ ∗
µ (A) 6 µ A ∩ Bi +µ A ∩ Bi 6 µ∗ (A∩Bi )+µ∗ A ∩ Bi 6 µ∗ (A)
i=1 i=1 i=1 i=1

Therefore µ∗ (A) = µ∗ (A ∩ ( ∞ A∩( ∞

S ∗
S c

i=1 Bi )) + µ i=1 Bi ) .

In the general case, we take the seq:

B1 , B1c ∩ B2 , . . . , B1c ∩ B2c ∩ · · · ∩ Bn−1

c
∩ Bnc , . . .

is a disjoint seq in Mµ∗ , and its union is ∞

S S∞
i=1 Bi . Thus i=1 Bi ∈ Mµ∗ . Therefore
Mµ∗ is a σ-algebra.
Finally,
S suppose P that (Bn ) is a disjoint seq. We have
∗ ∗
µ (Sn Bn ) 6 Pn µ (Bn ). S From (*), we have
∞ S∞ c
P∞ ∗
µ∗ ( n Bn ) > ∗ ∗
S
i=1 µ ( n Bn ∩ Bi ) + µ n Bn ∩ ( i=1 Bi ) = i=1 µ (Bi ).
Then " #
[ X
µ∗ Bn = µ∗ (Bn ).
n n
∗
Therefore µ : Mµ∗ → [0, +∞] is a measure.
Corollary 119. Mλ∗ (the set of Lebesgue measurable subsets) is a σ-algebra, more-
over the restriction λ∗ : Mλ∗ → [0, +∞] is a measure.
In these case, the restriction λ∗ on Mλ∗ will be denoted by λ and is called
Lebesgue measure.

32
Proposition 120. Every Borel subset of Rn is Lebesgue measurable.

Proof: It is enough to prove when n = 1. Let B be an interval of the form

(−∞, b]. We want
µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c )
for every subset A of R. One use the definition of infimum and covering. (exerc).

Conclusion:
We have
n
B(Rn ) ⊂ Mλ∗ ⊂ 2R
From these the Lebesgue measure λ can considered on B(Rn ).

Exercise 121. Prove that these inclusions are strict.

Exercise 122. Let A a Lebesgue meas. subset of Rn . Prove the following:

a) λ(A) = inf{λ(U ) : U open, A ⊂ U }
b) λ(A) = sup{λ(K) : K compact, K ⊂ A}

Exercise 123. The Lebesgue measure is invariant by translation i.e. λ(x + A) =

λ(A) for x ∈ Rn , A Borel subset.

Note that (Rn , +) is an abelian group.

As topological space, Rn is locally compact.

11 Haar measure (existence)

Definition 124. A topological group is a group G provided with a topology such that
the product
G × G → G (x, y) 7→ xy and the G → G x 7→ x−1 are continuous.

Example 125. (Rn , +) is a top. group.

Example 126. (GLn (R), ·) is a top group.

Exercise 127. Prove that (S 1 , ·) is a top group.

Exercise 128. Let G top group, let a ∈ G. Prove that the maps x 7→ ax and
x 7→ xa are homeomorphisms

Question: How to define a measure on a top group?

This question is difficult in general. We want to answer this question when this
top group is locally compact.

Let G a locally compact group. B(G) set of Borel subsets.

Definition 129. A (left) Haar measure on G is a measure µ : B(G) → [0, +∞] if

µ is invariant by left translation i.e. µ(xA) = µ(A) for all x ∈ G, A ∈ B(G).

33
Example 130. The Lebesgue measure on (Rn , +) is a Haar measure.

Let K be a compact subset of G, let V be a subset of G with interior V ◦ 6= ∅.

Note {xV ◦ }x∈G is a open covering of K. Then S
there is a finite sequence {xi }ni=1 such that K ⊂ ni=1 xi V .......(*)
Let
](K : V ) = min{n > 1 : satisfies(∗)}
By convention, ](K : V ) = 0 if K = ∅.

Let K0 be a compact subset of G with interior K ◦ 6= ∅.

Let C = {compact subsets of G}
U = {open neigborhoods of e}
For each U ∈ U, we define a function hU : C → R given by K 7→ ](K : U )/](K0 :
U)
We have the following properties:
(a) 0 6 hU (K) 6 ](K : K0 ) Sm
m n
Indeed, if {x i } i=1 and {y j } j=1 such that K ⊂ i=1 xi K0 S
and K0 ⊂ j=1 yj U , then by transitivity we have K ⊂ m
Sn Sn
i=1 j=1 xi yj U . Then

](K : U ) 6 ](K : K0 ) · ](K0 : U )

hence, hU (K) = ](K : U )/](K0 : U ) 6 ](K : K0 ).

Similarly, we have
(b) hU (K0 ) = 1
(c) hU (xK) = hU (K)
(d) If K1 ⊂ K2 , then hU (K1 ) 6 hU (K2 )
(e) hU (K1 ∪ K2 ) 6 hU (K1 ) + hU (K2 )
(f) (additive)
02/07
For each K ∈ C, let
IK = [0, ](K : K0 )]
This is an interval of R. Consider
Y
X= IK
K∈C

By the Tychonoff’s th, X is compact. By (a), hU (K) ∈ IK , i.e. hU ∈ X.

For each V ∈ U, let S(V
T ) the closure in X of {hU : U ∈ U, U ⊂ V }.T
Since X is compact, V ∈U S(V ) 6= ∅. We choose an element h• of V ∈U S(V ).
We have the following:
(a) 0 6 h• (K)
(b) h• (K0 ) = 1
(b’) h• (∅) = 0
(c) h• (xK) = h• (K)
(d) If K1 ⊂ K2 , then h• (K1 ) 6 h• (K2 )
(e) h• (K1 ∪ K2 ) 6 h• (K1 ) + h• (K2 )
(f) (additive) h• (K1 ∪ K2 ) = h• (K1 ) + h• (K2 ) if K1 ∩ K2 = ∅.

34
 0
Now consider τG the collection of open subsets of G (topology). Define µ : τG →
R̄ by
0
µ (U ) = sup{h• (K) : K ⊂ U, K ∈ C}
Hence define µ∗ : 2G → R̄ by
0
µ∗ (A) = inf{µ (U ) : A ⊂ U, U ∈ τG }
Claim: µ∗ is an outer measure

1) µ∗ (∅) = 0
2) (monotony) If A ⊆ B, then µ∗ (A) 6 µ∗ (B)
∗
S
P 3)∗ (subadditivity) If {An } is a sequence of subsets of X, then µ ( n An ) 6
n µ (An )

Finally, if denote µ the restriction of µ∗ to B(G), µ is a measure which is invariant

by left translation i.e. it is a Haar measure. Therefore we have the following:
Theorem 131. Let G be a locally compact group. Then there exists a Haar measure
on G.
Question: Is it unique?
No, but yes in the following sense.

Theorem 132. Let G be a locally compact group, and let µ, ν be two Haar measures
on G. Then there exists c > 0 such that ν = cµ.
Example 133. (measure on S 1 ) Consider ϕ : [0, 1i → S 1 by t 7→ e2πit
Let λ the Lebesgue measure on [0, 1i. We define a function µ on B(S 1 ) as
µ(A) = λ(ϕ−1 (A))
Claim: µ is a measure on S 1 .
Question: Is µ a Haar measure? Yes.
Given z ∈ S 1 , A ∈ B(S 1 ),
ϕ−1 (zA) = ϕ−1 (z) + ϕ−1 (A), because ϕ is the restriction of a group hom.
Hence µ(zA) = λ(ϕ−1 (zA)) = λ(ϕ−1 (z) + ϕ−1 (A)) = λ(ϕ−1 (A)) = µ(A).

An application to arithmetics (continuous fractions):

A continuous fraction has the form:

1
[3; 7, 16] = 3 + 1
7 + 16
For example, 355/113=[3; 7, 16]. The algorithm is as follows:
p = 355, q = 113. By the the alg of the division,
p = tq + r
(p, q) →t (q, r). We iterate this procedure until up to r = 0.
3 7 16
(355, 113) →
− (113, 16) →
− (16, 1) −
→ (1, 0)

35
 √
1+ 5 1
Example 134. 2
= [1; 1, 1, 1, ...] = 1 + 1+ 1
1+ 1
...
1
1
 1 X = [0, 1] − Q. Consider a function T : X → X such that T (x) =
Let x
:=
x
− x , where
1
x
is the entire part of x1 .
If x = [0; a1 , a2, . . .] = [a1 , a2 , . . .] = 0 + a1 + 1 1 , then
a2 + 1
..
.
1 1
= a1 +
x a2 + . 1
..

then
T (x) = [a2 , a3 , . . .] i.e. T acts as a shift to the left.

Is T measurable?
Yes,
Let µ a measure on X. When µ is T -invariant i.e. µ(T −1 (A)) = µ(A) for all A?

12 Decomposition of measures
Signed measure (charge)
Let (X, A) be a measurable space. A signed measure (or charge) is a function
µ : A → [−∞, +∞]
which satisfies the same axioms of measure.

Lemma 135. The sum and difference of charges is also a charge. The product of a
charge by a scalar is also a charge. Then the set of real-valued charges is a R-vector
space.

Proof: Let µ, λ charges

(µ + λ)(A) = µ(A) + λ(A)
(µ − λ)(A) = µ(A) − λ(A)
If A = ∅,Pthen (µ ± λ)(A)
P = 0. P P P P
(µ + λ) (
P i i A ) = µ ( i Ai ) + λ ( i Ai ) = i µ(Ai ) + i λ(Ai ) = i (µ(Ai ) +
λ(Ai )) = i (µ + λ)(Ai )
Similarly with the difference.

Example:
Let f ∈ L1 . For every A ∈ A, define
Z
λ(A) = f dµ,
A

then λ is a signed measure.

36
 Indeed, set λ+ (A) = A f + dµ, λ− (A) = A f − dµ. Now, λ+ , λ− are measures.
R R
Therefore
λ = λ+ − λ− is a signed measure.

Definition: λ signed measure on (X, A). A set P ∈ A is called positive if

λ(A ∩ P ) > 0 for all A ∈ A.
A set N ∈ A is called negative if λ(A ∩ N ) 6 0 for all A ∈ A.
A set M ∈ A is called null if λ(A ∩ M ) = 0 for all A ∈ A.

Theorem 136. (Hahn decomposition) λ signed measure on (X, A). Then there exist
P, N ∈ A which form a partition of X such that P is positive and N is negative.

Proof: Let P the set of positive sets. Note that ∅ ∈ P, then P 6= ∅. Let
α = sup{λ(P ) : P ∈ P}.
Consider a sequence (Pn ) in P such that limn λ(Pn ) = α. Take P = ∞
S
n=1 Pn .
We can assume that (Pn ) is an increasing seq.
Claim: P is positive
S∞ S∞
λ(A ∩ P ) = λ (A ∩ n=1 Pn ) = λ ( n=1 A ∩ Pn ) = limn λ(A ∩ Pn ) > 0.

Note that λ(P ) = α.

Claim: N = P c is negative
By contradiction, suppose that N is not negative. Then there is A ∈ A st
λ(A ∩ N ) > 0.
Set A0 = A ∩ N .
Question: Is A0 positive?
If A0 positive, then A0 ∪ P is positive. Hence λ(A0 ∪ P ) = λ(A0 ) + λ(P ) > α.
This is a contrad. Thus A0 is not positive. Then A0 contains a subset with negative
charge. i.e. there is a subset A1 ⊂ A0 st λ(A1 ) < 0. Let n1 be the smallest natural
number st there is a subset A1 ⊂ A0 st λ(A1 ) 6 −1/n1 .
Note
λ(A0 − A1 ) = λ(A0 ) − λ(A1 ) > λ(A0 ) > 0,
but A0 − A1 is not positive, if so P1 = P ∪ (A0 − A1 ) would be postive with
λ(P1 ) > α. Then A0 − A1 contains subsets with negative charge.
By induction, let n2 be the smallest natural number st there is a subset A2 ⊂
A0 − A1 such that λ(A1 ) 6 −1/n2S .
We get a seq (A Pk ). Take B = Pn An (disjoint).
Hence λ(B) = k λ(Ak ) 6 − k 1/nk 6 0
Then limk 1/nk = 0.

Claim: λ(A0 − B) > 0 ⇒ P ∪ A0 − B positive (contradiction with the def of α)

This concludes the proof.

Definition: The pair (P, N ) is called Hahn decomposition of X with respect to

λ.

Remark 137. The Hahn decomposition theorem does not give us the uniqueness of
such decomposition.

37
Lemma 138. Let (P1 , N1 ) and (P2 , N2 ) be two Hahn decomp for λ. Then for every
A ∈ A, we have
λ(A ∩ P1 ) = λ(A ∩ P2 ), λ(A ∩ N1 ) = λ(A ∩ N2 ).
Proof: Note that A ∩ (P1 − P2 ) ⊂ P1 ∩ N2 , because N2 = P2c . Then
λ(A ∩ (P1 − P2 )) = 0 (by defintion of positive and negative set).
Note that A ∩ (P1 − P2 ) = (A ∩ P1 ) − (A ∩ P1 ∩ P2 )
Hence,
λ(A ∩ P1 ) = λ(A ∩ P1 ∩ P2 )..... (1)

Similarly, A ∩ (P2 − P1 ) ⊂ P2 ∩ N1 , because N1 = P1c .

Then λ(A ∩ P2 ) = λ(A ∩ P1 ∩ P2 ).......(2)
From (1) and (2), λ(A ∩ P1 ) = λ(A ∩ P2 ).
Finally, A ∩ (N1 − N2 ) ⊂ N1 ∩ P2 , because P2 = N2c . Then λ(A ∩ N1 ) =
λ(A ∩ N1 ∩ N2 ). Similarly,
λ(A ∩ N2 ) = λ(A ∩ N1 ∩ N2 ). Thus λ(A ∩ N1 ) = λ(A ∩ N2 ).
The previous lemma gives us the following def:
Definition: λ signed measure on (X, A). Let (P, N ) Hahn decomp of λ. Define
λ+ (A) = λ(A ∩ P )
λ− (A) = −λ(A ∩ N )

Lemma: λ+ , λ− are measures on (X, A).

Proof: λ+ , λ− > 0, they satifies the axioms of measure.

We have λ(A) = λ((A ∩ P ) + (A ∩ N )) = λ(A ∩ P ) + λ(A ∩ N ) = λ+ (A) − λ− (A).

Thus λ = λ+ − λ− .

The total variation of λ is |λ| defined by

|λ|(A) = λ+ (A) + λ− (A)
for all A ∈ A.
Observe that |λ| is a measure, as it is sum of measures.
Example: Let f ∈ L1 . For every A ∈ A, define
Z
λ(A) = f dµ,
A

then λ is a signed measure. If λ (A) = A f + dµ, λ− (A) = A f − dµ, then λ+ , λ− are

+
R R
measures and
λ = λ+ − λ− . Moreover, R
|λ|(A) = λ+ (A) + λ− (A) = A f + dµ + A f − dµ = A |f |dµ
R R

16/07
Theorem 139. (Jordan decomposition) λ charge on (X, A). Then λ is a difference
of measures. Moreover, if λ = µ − ν, where µ, ν are measures on (X, A), then
µ(A) > λ+ (A), ν(A) > λ− (A)
for all A ∈ A.

38
Proof. We already know that λ = λ+ − λ− . Suppose that λ = µ − ν. Then
λ+ (A) = λ(A ∩ P ) = (µ − ν)(A ∩ P ) = µ(A ∩ P ) − ν(A ∩ P ) 6 µ(A ∩ P ) 6 µ(A).
Similarly,
λ− (A) = −λ(A ∩ N ) = −(µ − ν)(A ∩ P ) = −µ(A ∩ P ) + ν(A ∩ P ) 6 ν(A ∩ P ) 6
ν(A).

Theorem 140. (Radon-Nikodym) λ abs cont with respect to µ (σ-finite measures).

Then there exists a function f ∈ M + (X, A) such that
Z
λ(A) = f dµ,
A

for all A ∈ A. Moreover, f is unique µ-a.e.

Proof. First suppose that λ, µ are finite measures. Considere the charge λ − cµ,
where c > 0. By the Hahn decomp, there is a decomp (P (c), N (c)) of λ − cµ. Set
D1 = N (c), D2 = N (2c) − D1 . If k > 1, we set Dk+1 = N ((k + 1)c) − kj=1 Dj . Note
S
that (Dk ) is a disjoint seq, and
k
[ k
[
N (jc) = Dj
j=1 j=1

Let E = X − ∞
S
j=1 Dj .
1) Claim: µ(E) = 0
Sk−1 Tk−1
We have Dk = N (kc) − j=1 N (jc) = N (kc) ∩ j=1 P (jc). By the other hand,
S∞ T∞
E = X − j=1 Dj = j=1 P (jc). Then E ⊂ P (jc) for all j > 1. Hence,
(λ − jcµ)(E) > 0, then

0 6 jcµ(E) 6 λ(E) 6 λ(X) < +∞

for all j > 1. Thus µ(E) = 0.

Since λ  µ,Pλ(E) = 0.
Define fc = k (k − 1)cχDk . Note that each fc is measurable > 0.
If A ⊂ Dk , then A ⊂ N (kc) and A ⊂ P ((k − 1)c), hence

(λ − kcµ)(A) 6 0, (λ − (k − 1)cµ)(A) > 0

Thus
(k − 1)cµ(A) 6 λ(A) 6 kcµ(A).
Let A ∈ PA, we have
A= ∞ k=1 (A ∩ Dk ) + (A ∩ E).
Hence,
Z X∞ Z X∞ Z ∞
X ∞
X
fc dµ = fc dµ = (k−1)cχDk dµ = (k−1)cµ(A∩Dk ) 6 λ(A∩Dk ) = λ(A)
A k=1 A∩Dk k=1 A∩Dk k=1 k=1
R
Thus A
fc dµ 6 λ(A)

39
 Similarly (using the other ineq),
Z Z
λ(A) 6 (fc + c)dµ 6 fc dµ + cµ(X)
A A
−n
Reeplacing c = 2 , n > 1, we have, fn = fc , and
Z Z Z
−n
fn dµ 6 λ(A) 6 (fn + 2 )dµ 6 fn dµ + 2−n µ(X)
A A A

....(*)
Claim: (fn ) is Cauchy in L1 ⇒ (fn ) is convergent.
If m > n,
Z Z Z Z
−n −n
fn dµ 6 λ(A) 6 (fn + 2 )dµ 6 fn dµ + 2 µ(X) 6 fm dµ + 2−m µ(X)
A A A A

Hence
R 
 (fn − fm )dµ 6 2−n µ(X)
A

Hence, from (*) Z Z

λ(A) = lim fn dµ = f dµ
n A A
Claim: f is unique µ-a.e.
SupposeRthat R
λ(A) = A f dµ = A gdµ, where f, g > 0 measurable. Let A1 = {x ∈ A : f (x) >
R A2 = {x
g(x)}, R ∈ A : f (x)
R < g(x)}.R Hence
f dµ = f dµ + f dµ + A−(A1 ∪A2 ) f dµ
RA R A1 R A2 R
A
gdµ = A1
gdµ + A2R
gdµ + A−(A1 ∪A2 )
gdµ
R
Then A1 (f − g)dµ = A2 (g − f )dµ = 0, hence f = g on A1 ∪ A2 with µ(X −
(A1 ∪ A2 )) = 0.

Finally, we prove S
the general case (λ, µ σ-finite). There exists a increasing seq
(Xn ) such that X = n Xn and
λ(Xn ) < +∞, µ(Xn ) < +∞, for all n > 1.
By the previous case, for each n, there exists gn ∈ M + and such that
Z
λ(A) = gn dµ
A

If m > n, A ⊂ Xn ⊂ Xm , then
Z Z
gn dµ = λ(A) = gm dµ
A A

Then gn = gm µ-a.e. in Xn .
Define fn = max{g1 , . . . , gn } mearurable > 0. (fn ) is increasing. Let f = limn fn .
Then Z Z
λ(A ∩ Xn ) = gn dµ = fn dµ
A A

40
By monotone cnvergence,
Z Z
λ(A) = lim λ(A ∩ Xn ) = lim fn dµ = fdµ.
n n A A

The function f is called the Radon-Nikodym derivative de λ with respect to µ.

Notation: f = dλ/dµ.

21/07
Exercise:
1) (Chain rule) Let ν  λ  µ σ-finite measures. Prove that
dν dλ
dν/dµ = . µ − a.e.
dλ dµ
2) Let λ1 , λ2  µ σ-finite measures. Prove that
d dλ1 dλ2
(λ1 + λ2 ) = + µ − a.e.
dµ dµ dµ
3) Let λ  µ, µ  λ, σ-finite measures. Then
dλ 1
= µ − a.e.
dµ dµ/dλ
Definition: (X, A) meas space. Let λ, µ measures on X. We say that λ, µ
are mutually singular if there are disjoint subsets A, B of X st X = A ∪ B and
λ(A) = µ(B) = 0.
Notation: λ⊥µ
The relation ⊥ is symmetric

Theorem 141. (Lebesgue decomposition) Let λ, µ σ-finite measures on (X, A).

Then there exists a measure λ1 with λ1 ⊥µ and a measure λ2 with λ2  µ such
that λ = λ1 + λ2 . The measures λ1 abd λ2 are unique with that property.
Proof: Consider ν = λ + µ. Note that ν is σ-finite, and λ  ν, µ  ν. By R-N.
∃f, g ∈ M + (X, A) such that
Z Z
λ(A) = f dν, µ(A) = gdν,
A A

for all A. Let E = {x ∈ X : g(x) = 0}, F = {x ∈ X : g(x) > 0}. Note that
X = E + F . Define
λ1 (A) := λ(A ∩ E), λ2 (A) := λ(A ∩ F ).
Observe that λ1 ⊥µ, because µ(E) = 0 and λ1 (F ) = 0. Let us see that λ2  µ. If
µ(A) = 0, then R
0 = µ(A) = A gdν ⇒ g = 0 ν-a.e. Note that ν(A ∩ F ) = 0. Now, since λ  ν,
we have λ(A ∩ F ) = 0. Then
λ2 (A) = λ(A ∩ F ) = 0.
Moreover, (λ1 + λ2 )(A) = λ1 (A) + λ2 (A) = λ(A ∩ E) + λ(A ∩ F ) = λ(A). Thus
λ = λ1 + λ2 .
Uniqueness (exerc)

41
13 Riesz representation for Lp
Linear algebra

(V, h.i) vector space with inner product

Riesz representation: Given φ ∈ V ∗ . Then ∃!w ∈ V such that φ(v) = hv, wi.
Moreover, kφk = kwk.

It is true for example when V is finite dimensional.

We want to analyse when V = Lp .

Bounded linear operator:

Let V , W (normed) vector space.
T : V → W linear map.
The following are equivalent:
a) T is continuous
b) T is bounded i.e. ∃ M > 0 such that kT (v)kW 6 M kvkV for all v ∈ V .

In this, we can define kT k = sup{kT (v)kW : kvkV 6 1, v ∈ V }

Suppose that V + is a cone. We write that v > 0 iff v ∈ V + .
Definition: Let φ ∈ V ∗ . We say that φ is positive if φ(v) > 0 for all v ∈ V such
that v > 0.
Lemma 142. V = Lp . For every φ ∈ V ∗ bounded, there are two positive bounded
linear functional φ+ , φ− such that φ = φ+ − φ− .
Proof. Let f ∈ V with f > 0. Define

φ+ (f ) = sup{φ(g) : g ∈ V, 0 6 g 6 f }

Claim: φ+ is R+ -linear i.e. φ+ (cf ) = cφ+ (f ), φ+ (f1 + f2 ) = φ+ (f1 ) + φ+ (f2 ) for

f1 , f2 > 0, c > 0.
Indeed,
If c = 0, then φ+ (cf ) = cφ+ (f ). Suppose that c > 0.

φ+ (cf ) = sup{φ(g) : g ∈ V, 0 6 g 6 cf } = c. sup{φ(g/c) : g ∈ V, 0 6 g/c 6 f } = cφ+ (f ).

φ+ (f1 + f2 ) = sup{φ(g) : g ∈ V, 0 6 g 6 f1 + f2 }
φ+ (f1 ) = sup{φ(g) : g ∈ V, 0 6 g 6 f1 }
φ+ (f2 ) = sup{φ(g) : g ∈ V, 0 6 g 6 f2 }
g − f2 6 f1
Consider g1 = max(0, g − f2 ), g2 = min(g, f2 ). Then g1 + g2 = g. Hence,
0 6 g1 6 f1 , 0 6 g2 6 f2 , φ(g) = φ(g1 + g2 ) = φ(g1 ) + φ(g2 ) 6 φ+ (f1 ) + φ+ (f2 ).
Thus φ+ (f1 +f2 ) 6 φ+ (f1 )+φ+ (f2 ). On the other hand, let 0 6 g1 6 f1 , 0 6 g2 6 f2 .
Then 0 6 g1 + g2 6 f1 + f2 ,

φ(g1 ) + φ(g2 ) = φ(g1 + g2 ) 6 φ+ (f1 + f2 )

Taking sup, we have φ+ (g1 ) + φ+ (g2 ) 6 φ+ (f1 + f2 ).

42
 Let f ∈ V . Define
φ+ (f ) := φ+ (f + ) − φ+ (f − )
Note that φ+ is R-linear and bounded (exerc).
Define φ− (f ) := φ+ (f ) − φ(f ), note that φ− bounded linear functional. Thus
φ = φ+ − φ− .
23/07

Theorem 143. (Riesz representation) (X, A, µ) measure space, µ σ-finite. Let φ be

a bounded linear functional on Lp . Then there exists g ∈ Lq , p and q are conjugate
and such that Z
φ(f ) = f gdµ

for all f ∈ Lp . Moreover, kφk = kgkq .

Proof.
Case: p = 1, q = +∞.
Suppose that µ is a finite measure and φ is positive. Define λ : A → R by
λ(A) = φ(χA ).
Claim: λ is a measure and λ  µ.
Indeed, λ(∅) = 0. We have
∞
! ∞
! ∞ ∞
X
X X X
λ An = φ χP∞ n=1 An
= φ χ An = φ(χ An ) = λ(An )
n=1 n=1 n=1 n=1

λ(A) > 0 for all A. Thus λ is a measure. R

Suppose µ(A) = 0. Then χA = 0 µ-a.e.⇔ χA dµ = 0 i.e. χA = 0 in L1 . Then
λ(A) = φ(χA ) = 0.
Therefore λ  µ.
R
By R-N. ∃g ∈ M + st λ(A) = A
gdµ ......(*)
R 1
Claim: φ(f ) = f gdµ for all f ∈RL.
In particular, from (*) φ(χA ) = χA gdµ.
R
By the linearity of φ, we have φ(ϕ) = ϕgdµ for all measurable simple function
ϕ.
Let f ∈ L1 , ∃(ϕn ) seq of measurable simple functions st ϕn ↑ f . Since φ is
bounded, φ(f ) = limn φ(ϕn ).
By the monotone convergence th,
Z Z
φ(f ) = lim φ(ϕn ) = lim ϕn gdµ = f gdµ
n n

Case: µ is σ-finite. Let X = ∪n Bn st (Bn ) increasing and µ(Bn ) < +∞ for all n.
By the previous case, for each n, ∃gn ∈ L∞ st
Z
φ(f χBn ) = f χBn gn dµ

43
 1
R all f ∈ L . R Observe that if m 6 n then Bm ⊂ Bn . From the equation above,
for
χBm gm dµ = χBm gn dµ i.e. gm = gn on Bm µ-a.e. then there exists g ∈ L∞ st
g |Bn = gP
n µ-a.e. for all n. Then
g= ∞ n=1 χBn gn .
Hence
∞
! ∞ ∞ Z ∞
Z X Z
X X X
φ(f ) = φ f χBn = φ(f χBn ) = f χBn gn dµ = f χBn gn dµ = f gdµ.
n=1 n=1 n=1 n=1

Now, we suppose that φ is arbitrary. By the previous lemma, φ = φ+ − φ− . By

the previous case,
∃g + , g − ∈ L∞ st −
+
R + R − + −
R φ (f ) = f g dµ and φ (f ) = f g dµ. Taking g := g − g ,
we obtain φ(f ) = f gdµ.
R
Case: p, q > 1. Doing the same as before, by R-N. ∃g ∈ M + st λ(A) = A
gdµ.

Claim: g ∈ Lq .
Consider An = {x ∈ X : |g(x)| 6 n}, it is measurable, X = ∪n An increasing.
Then gχAn ∈ Lq .
Define φn : Lp → R by φn (f ) = φ(f χAn ). Note that φn is bounded linear
functional.
Assume R
φn (f ) = f gχAn dµ (use the previous idea)
Then kgχAn kq = kφn k 6 kφk. By the monotone conv th. kgkq 6 kφk. Then
g ∈ Lq .

R
Consider b : Lp × Lq → R given by b(f, g) = f gdµ.

Lemma 144. b is a bilinear form.

Proof. (exerc)
Question: Is b non-degenerate?
b : V × W → R is non-degenerated if
1) b(v, w) = 0 for all v, then w = 0
2) b(v, w) = 0 for all w, then v = 0.

Example: The inner product h, i is non-degenerate.

R
Supp
R b(f, g) = f gdµ=0 for all f ∈ Lp . Then in part,
χA gdµ = 0 for all A. Then g = 0 µ-a.e. i.e g = 0 in Lq .
Thus b is non-degenerate.

Define
ψ : V → W ∗ given by v 7→ b(v, −)
ψ 0 : W → V ∗ given by w 7→ b(−, v)
Claim: b non-degenerated => ψ, ψ 0 injective
ψ injective: Suppose ψ(v) = 0 then b(v, w) = 0 for w. This implies v = 0.

44
 Application: V = Lp , W = Lq
In case that p = q = 2, then V = W = L2 . Moreover b is a inner product. In
conclusion L2 is Hilbert space (Banach space+norm is defined from hi).

ψ : V → W ∗ induces ψ ∗ : W ∗∗ → V ∗ . If W is algebraic reflexive, ie W ∗∗ = W ,

then we have ψ ∗ : W → V ∗ .

V ∗ algebraic dual of V .
0
V = {ϕ ∈ V ∗ : ϕ bounded} topological dual of V . We have V 0 ⊂ V ∗ .
Remark: In dim < +∞, V 0 = V ∗ .

We say that V is topological reflexive if V 00 = V .

Question: Is Lp (algebraic) reflexive?
Question: Is Lp (toplogical) reflexive?
Question: Is Lp separable (i.e. it has a countable dense subset)?

30/07

Corollary 145. (X, A, µ) measure space, let p and q conjugate. Then the induced
map
Lp → (Lq )∗ is an isometry (injective but not necessarily surjective for p = 1).

Generation of measures:

A algebra of subsets of a set X, µ : A → [0, +∞] measure

Question:
Is there a σ-algebra A∗ containing A and a measure µ∗ : A∗ → [0, +∞] such
that
µ∗ |A = µ?
Yes. (Caratheodory extension)
Definition: A algebra of subsets of a set X, µ : A → [0, +∞] measure. For each
B ⊂ X. (∞ ∞
)
X [
µ∗ (B) = inf µ(Aj ) : (Aj ) seq in A st B ⊂ Aj
j=1 j=1

Lemma 146. µ∗ is an outer measure and µ∗ |A = µ.

Proof. 1) µ∗ (∅) = 0
2) (monotony) If BS∞⊆ C, then µ∗ (B) 6 µ∗ (C).
Indeed, B ⊂ C ⊂ j=1 Aj .
∗
S
P 3)∗ (subadditivity) If {Bn } is a sequence of subsets of X, then µ ( n Bn ) 6
n µ (Bn )
Given ε > 0. We use the definition of µ∗ (Bn ). For each n, ∃ (Ank ) in A st
∞ ∞
[ X ε
B⊂ Ank , µ(Ank ) 6 µ∗ (Bn ) +
k=1 k=1
2n

45
Hence, considering the seq (Ank )
! ∞ ∞  ∞
[ X X ε X ∗
µ∗ Bn 6 µ(Ank ) = µ∗ (Bn ) + n 6 µ (Bn ) + ε
n n,k=1 n=1
2 n=1

Since ε is arb, we get µ∗ ( n Bn ) 6 n µ∗ (Bn ).

S P

Definition 147. A subset B ⊂ X is µ∗ − measurable with respect to A if

µ∗ (A) = µ∗ (A ∩ B) + µ∗ (A ∩ B c )

for every subset A ⊂ X.

We denote by A∗ the set of µ∗ − measurable with respect to A.

Theorem 148. (Caratheodory extension) A∗ is a σ-algebra which contains A. More-

over the restriction µ∗ on A∗ is a measure.

Proof. It is a consequence from the Caratheodory method.

04/08

Theorem 149. (Hahn extension) Assume that µ is a σ-finite measure on an algebra

A. Then there is a unique extension of µ to a measure on A∗ .

Proof. The existence is given by the Caratheodory ext. Let us see the uniqueness.
Suppose that there is a measure µ0 on A∗ such that µ0 |A = µ. We want to see that
µ0 = µ∗ .
Suppose that µ is finite. We have µ∗ (X) = µ(X) = µ0 (X) is finite, thus µ∗ , µ0
are also finite.
Recall that nP o
∞ S∞
µ∗ (B) = inf j=1 µ(Aj ) : (A j ) seq in A st B ⊂ A
j=1 j
S∞
Let (Aj ) seq in A st B ⊂ j=1 Aj . Then

∞
! ∞ ∞
[ X X
0 0 0
µ (B) 6 µ Aj 6 µ (Aj ) = µ(Aj )
j=1 j=1 j=1

By the def of infimum, µ0 (B) 6 µ∗ (B) for all B ∈ A∗ . On the other hand, µ0 , µ∗
measures on A∗ , then for B ∈ A∗ , we have

µ∗ (B) + µ∗ (X − B) = µ(X) = µ0 (B) + µ0 (X − B) 6 µ0 (B) + µ∗ (X − B)

Then µ∗ (B) 6 µ0 (B). Therefore µ∗ (B) = µ0 (B) for all B ∈ A∗ .

General case (exerc)

46
14 Riesz representation (continuous case)
X topological space.
C 0 (X) = {f : X → R : f cont}
supp(f ) = {x ∈ X : f (x) 6= 0}
K(X) = {f ∈ C 0 (X) : f has compact support}
If X is compact, then K(X) = C 0 (X).

Lemma 150. K(X) is a vector subspace of C 0 (X).

Proof. Let f, g ∈ K(X). Then

supp(f + g) = {x ∈ X : (f + g)(x) 6= 0}
We have {x ∈ X : (f + g)(x) 6= 0} ⊂ {x ∈ X : f (x) 6= 0} ∪ {x ∈ X : g(x) 6= 0}.
The n
supp(f + g) ⊂ supp(f ) ∪ supp(g). Therefore f + g ∈ K(X).
If α ∈ R, then supp(αf ) = supp(f ) when α 6= 0.

Theorem 151. (Riesz representation) Let X be a locally compact Hausdorff space.

Let φ ∈ K(X)∗ positivo. Then there is a unique regular Borel measure µ on X such
that Z
φ(f ) = f dµ

for all f ∈ K(X).

First of all let us see some preliminaries.

Let X Hausdorff. Let A be a σ-algebra on X such that B(X) ⊂ A, µ : A →
[0, +∞] measure.

Definition 152. µ is regular if:

1) µ(K) < +∞ for all compact subset K ⊂ X
2) For each A ∈ A, we have

µ(A) = inf{µ(U ) : A ⊂ U, U open}

3) For each U open, we have

µ(U ) = sup{µ(K) : K ⊂ U, K compact}

A regular Borel measure on X is a regular measure µ : B(X) → [0, +∞].

Proposition 153. Let X be a locally compact Hausdorff space that has a countable
base. Let µ be a Borel measure on X that is finite on compact sets. Then µ is
regular.

Proposition 154. Let X be a locally compact Hausdorff space that has a countable
base. Then every regular measure on X is σ-finite.

47
 Strategy of the proof:
Let φ ∈ K(X)∗ positivo.
Let O(X) = {open subsets of X}
Define µ∗ : O(X) → [0, +∞] given by

µ∗ (U ) := sup{φ(f ) : f ∈ K(X), f ≺ U }

where f ≺ U means 0 6 f 6 χU and supp(f ) ⊂ U .

The function µ∗ extends to 2X , defining by for A ⊂ X as

µ∗ (A) := inf{µ∗ (U ) : U open, A ⊂ U }

It is proven that µ∗ is an outer measure, and B(X) ⊂ Mµ∗ .

By Caratheodory method, µ∗ is a measure on Mµ∗ . Take µ = µ∗ |B(X) . Moreover
Z
φ(f ) = f dµ

for all f ∈ K(X).

06/08

Definition: A topological space X is locally compact if each point x ∈ X has a

compact neighborhood of x. Equiv. if each point x ∈ X has an open neighborhood
of x whose closure is compact.
Example: Rn is locally compact.
Definition: A topological space X is normal if it is Hausdorff (T2 ) and satisfies the
axiom T4 : every two disjoint closed subsets of X have disjoint open neighborhood.
Example: Rn is normal.

Proposition 155. Let X Hausdorff space, and let K and L be disjoint compact
subsets of X. Then there are disjoint open subsets U and V such K ⊂ U and
L⊂V.

Proof. Assume K 6= ∅ and L 6= ∅. Suppose that K = {x}. Since X is Hausdorff,

for each y ∈SL, there are disjoint open subsets Uy , Vy such that x ∈ Uy and y ∈ Vy .
Since L ⊂ y∈L Vy and L is compact, there are finite points y1 , . . . , yn such that
L ⊂ Vy1 ∪ · · · ∪ Vyn . Put U = ni=1 Uyi and V = ni=1 Vyi . We have K ⊂ U and
T S
L ⊂ V , and U ∩ V = ∅.
General case: For each x ∈ K, S there are disjoint open subsets Ux and Vx such
K ⊂ Ux and L ⊂ Vx . Since K ⊂ x∈K Ux and K is compact, Sm there are finite
Tmpoints
x1 , . . . , xm such that K ⊂ Ux1 ∪ · · · ∪ Uxm . Put U = i=1 Uxi and V = i=1 Vxi .
We have K ⊂ U and L ⊂ V , and U ∩ V = ∅.
Definition: The subsets U and V in the previous prop. are said to separate the
sets K and L.

Proposition 156. Let X be a locally compact Hausdorff space. Let x ∈ X, and

let U open neighborhood of x. Then x has an open neighborhood whose closure is
compact and contained in U .

48
Proof. Since X is locally compact, there is an open neighborhood W of x such that
W̄ is compact. Consider W ∩ U . Then W ∩ U ⊂ W̄ , then W ∩ U is compact. Then
W ∩ U − (W ∩ U ) ⊂ W ∩ U ⊂ W̄ , thus W ∩ U − (W ∩ U ) is compact. Note that
x ∈ W ∩ U.
In proposition above, consider {x} and W ∩ U − (W ∩ U ). Then there are open
subsets V1 and V2 that separate te sets {x} and W ∩ U − (W ∩ U ). i.e. x ∈ V1 and
W ∩ U − (W ∩ U ) ⊂ V2 .
Consider V1 ∩ W ∩ U . Note that x ∈ V1 ∩ W ∩ U and V1 ∩ W ∩ U ⊂ W ∩ U ⊂ U
(exerc)
Proposition 157. Let X be a locally compact Hausdorff space. Let K ⊂ X compact,
and let U open subset with K ⊂ U . Then there is an open subset V of X that has
a compact closure and K ⊂ V ⊂ V̄ ⊂ U .
Proof. It is generalization of the previous prop.
Proposition 158. Every compact Hausdorff space is normal.
Definition: 1) A subset A of a topological space X is Gδ if it is countable
intersection of open subsets.
2) A subset A of a topological space X is Fσ if it is countable union of closed
subsets.
Theorem 159. (Urysohn’s lemma) Let X be a normal topological space. Let E
and F be two disjoint closed subsets of X. Then there exists a continuous function
f : X → [0, 1] such that 
0, x ∈ E;
f (x) =
1, x ∈ F.
Proposition 160. Let X be a locally compact Hausdorff space. Let K compact
subset of X, and let U open subset of X with K ⊂ U . Then there is a function
f ∈ K(X) such that χK 6 f 6 χU and supp(f ) ⊂ U .
Proof. By a previous prop. there is an open subset V of X that has a compact
closure and
K ⊂ V ⊂ V̄ ⊂ U.
Note that V̄ is a compact Hausdorff space, thus V̄ is normal. We apply the Urysohn’s
lemma to V̄ . Then there is a continuous function g : V → [0, 1] such that

0, x ∈ V̄ − V ;
g(x) =
1, x ∈ K.

g(x), x ∈ V̄ ;
Define f : X → [0, 1] such that f (x) =
0, x 6∈ V̄ .
f is continuos, moreover f ∈ K(X) such that χK 6 f 6 χU and supp(f ) ⊂
U.

Proof of the Riesz representation:

Assume that µ∗ is an outer measure, and B(X) ⊂ Mµ∗ . Take µ = µ∗ |B(X)
measure.

49
 For each compact K of X, there is a function f ∈ K(X) such that χK 6 f .
Claim: If χA 6 f then µ∗ (A) 6 φ(f ). If 0 6 f 6 χA and if A is compact then
φ(f ) 6 µ∗ (A).
Let A ⊂ X subset such that χA 6 f . Let ε st 0 < ε < 1, and define

Uε = {x ∈ X : f (x) > 1 − ε}
1
Note Uε is open. If g ∈ K(X) is such that g 6 χUε , then g 6 1−ε
f. Then φ(g) 6
1
1−ε
φ(f ). Recall that

µ∗ (U ) := sup{φ(f ) : f ∈ K(X), f ≺ U }

where f ≺ U means 0 6 f 6 χU and supp(f ) ⊂ U . Then µ∗ (Uε ) 6 1−ε1

φ(f ). Since
χA 6 f , A ⊂ Uε . Taking ε → 0, we get µ∗ (A) 6 φ(f ).
Reciprocally, let U open st K ⊂ U . By definition φ(f ) 6 µ∗ (U ). Since U is
arbitrary, φ(f ) 6 µ∗ (K).

Let µ1 := µ∗ |Mµ∗ . It is a measure.

Claim: Z Z
φ(f ) = f dµ = f dµ1

for all f ∈ K(X).

Let f ∈ K(X). Assume that f > 0. Let ε > 0. For each n > 1, define
fn : X → R,

 0, if f (x) 6 (n − 1)ε;
fn (x) = f (x) − (n − 1)ε, if(n − 1)ε < f (x) 6 nε;
ε, if nε < f (x).

P
Then fn ∈ K(X) and f = n fn . Since f is continuous with compact support, it
is bounded. Hence there is a positive integer N such that fn = 0 for all n > N .
Put K0 = supp(f ), and for each n > 0 let Kn = {x ∈ X : f (x) > nε}. We have
Kn ⊂ Kn−1 for n > 1. Notice that one has εχKn 6 fn 6 εχKn−1 for all n > 1.
Taking integral we have
Z
εµ(Kn ) 6 fn dµ 6 εµ(Kn−1 )

On the other hand, by the previous claim,

N
X N
X −1
ε µ(Kn ) 6 φ(f ) 6 ε µ(Kn )
n=1 n=0

PN
Now, we have f = n=1 fn . Then,
N
X Z N
X −1
ε µ(Kn ) 6 f dµ 6 ε µ(Kn )
n=1 n=0

50
 h P PN −1 i
R N
Thus, we observe that φ(f ) and f dµ both lie in the interval ε n=1 µ(Kn ), ε n=0 µ(Kn )
R
which has length ε(µ(K0 ) − µ(KN )). Therefore φ(f ) = f dµ.
Unicity (exerc)

11/08

Corollary 161. Let X be a locally compact Hausdorff space. Then there is a one-
to-one correspondence

{positive linear functional on K(X)} ∼

= {regular BorelmeasureonX}

15 Product measures
Product of measurable spaces
(Xα , Aα )α∈Λ family of measurable spaces.
Q
How to define a natural σ-algebra
Q on α Xα ?
Consider the projections prα : α Xα → Xα defined by (xα ) 7→ xα .

Let A ∈ Aα . Then pr−1

Q
α (A) ⊂ α Xα .

We can ask the following: Given a set Y and a family of maps fα : Y → Xα for
α∈Λ
How to define a natural σ-algebra on Y ?
Let A ∈ Aα . Then fα−1 (A) ⊂ Y .

Define AY = h{fα−1 (A) : α ∈ Λ, A ∈ Aα }i σ-algebra on Y .

We have a measurable space (Y, AY ).Q

In particular, we get a σ-algebra on Q α Xα .
Notation: AQα Xα will be denoted by α Aα .
Let Ai ∈ Aαi for i = 1, . . . , n.
fα−1
1
(A1 ) ∩ · · · ∩ fα−1
n
(An ) ∈ h{fα−1 (A) : α ∈ Λ, A ∈ Aα }i
Then
h{fα−11
(A1 ) ∩ · · · ∩ fα−1
n
(An ) : n > 1, Ai ∈ Aαi , αi ∈ A, i = 1, . . . , n}i ⊂ h{fα−1 (A) :
α ∈ Λ, A ∈ Aα }i
Do we have an equality?
Yes.

pr−1 −1
Xα0 , where
Q
α1 (A1 ) ∩ · · · ∩ prαn (An ) = {(xα ) : xα1 ∈ A1 , . . . , xαn ∈ An } = α
Xα = Aα if α ∈ {α1 , . . . , αn }, and Xα0 = Xα if α 6∈ {α1 , . . . , αn },
0

pr−1 −1
α1 (A1 ) ∩ · · · ∩ prαn (An ) can be seen as a cylinder.

Particular case: (X, A), (Y, B) measurable spaces. A × B if the σ-algebra on

X ×Y.
prX : X × Y → X, prY : X × Y → Y .
Let A ∈ A, B ∈ B. Then A × B = pr−1 −1
X (A) ∩ prY (B) this cylinder is called
rectangle.

51
 Let (X, A ) and (Y, B) be two measurable spaces. A subset of X × Y is called
rectangle with measurable sides if it has the form A×B for some A ∈ A and B ∈ B.
We denote by A × B the σ-algebra on X × Y generated by the collection of all
rectangles with measurable sides.
Consider the projections

X ×Y
prX prY

{ #
X Y
Let x ∈ X and y ∈ Y . For a subset E ⊆ X × Y , we write

Ex = prY (E ∩ pr−1
X (x)) = {y ∈ Y : (x, y) ∈ E}
E y = prX (E ∩ pr−1
Y (y)) = {x ∈ X : (x, y) ∈ E}

They are called the x-section and y-section of E, respectively.

If f : X × Y → R̄ is a function, then the x-section of f is the function fx : Y → R̄
defined by
y 7→ fx (y) = f (x, y) .
Similarly, the y-section of f is the function f y : X → R̄ defined by

x 7→ f y (x) = f (x, y) .

Let f : X × Y → R̄ function. Let x ∈ X, y ∈ Y . Define

fx : Y → R̄ given y 7→ f (x, y)
x-section of f
f y : X → R̄ given x 7→ f (x, y)
y-section of f .
Let E ⊂ X × Y , denote
Ex = {y ∈ Y : (x, y) ∈ E} x-section of E
E y = {x ∈ X : (x, y) ∈ E} y-section of E
Question: If f is measurable, is it true that fx , f y are measurable for all x ∈
X, y ∈ Y ?
Yes,
Let α ∈ R. Then

{y ∈ Y : fx (y) > α} = {y ∈ Y : f (x, y) > α} = {(x, y) ∈ X × Y : f (x, y) > α}x

Since f is measurable then {(x, y) ∈ X × Y : f (x, y) > α} is measurable.

It is enough to see that each section of a measurable set is measurable. Indeed,
let E measurable in X × Y . We want to see that Ex (similarly E y ) is measurable.
If E = A × B, where A ∈ A, B ∈ B. Then Ex = {y ∈ Y : (x, y) ∈ A × B} = B
(if x ∈ A) is measurable, and Ex = ∅ (if x 6∈ A) is measurable.
AX×Y is generated by rectangles. Consider E = {F ∈ AX×Y : Fx is meas i.e.Fx ∈
B}.
We have seen that {rectangles} ⊂ E. Then AX×Y ⊂ E. Thus E = AX×Y .
Lemma 162. Let (X, A ) and (Y, B) be two measurable spaces.

52
 (a) If E is an (A × B)-measurable set, then each x-section of E is B-measurable
and each y-section of E is A -measurable.

(b) If f is an extended (A ×B)-measurable function on X ×Y , then each x-section

of f is B-measurable and each y-section of f is A -measurable.

13/08

Proposition 163. Let (X, A, µ), (Y, B, ν) σ-finite measure spaces. If E ∈ A × B,

7 µ(E y ) are
then the functions X → R̄ given by x 7→ ν(Ex ) and Y → R̄ given by y →
measurable.

Proof. Suppose that ν is finite. Let

F = {E ∈ A × B : x 7→ ν(Ex ) is measurable}

Claim: F is a σ-algebra
(exerc)
Claim: F = A × B (this proves the first part of the prop.)
Indeed, we know that Ex is B-measurable. Consider rectangle E = A × B in
A × B. Note that
ν((A × B)x ) = ν(B)χA (x),
and the function x 7→ ν(B)χA (x) is measurable. Then E = A × B ∈ F.
Since F is a σ-algebra containing of rectangles in A × B, we conclude that
F = A × B.
Similarly, one defines

G = {E ∈ A × B : y 7→ µ(E y ) is measurable}

and hence one proves that G = A × B.

Theorem 164. Let (X, A, µ), (Y, B, ν) σ-finite measure spaces. Then there is a
unique measure µ × ν on A × B such that

(µ × ν)(A × B) = µ(A)ν(B)

for all A ∈ A and B ∈ B. Moreover, if E ∈ A × B, then

Z Z
(µ × ν)(E) = ν(Ex )µ(dx) = µ(E y )ν(dy)
X Y

Proof. By the previous prop., for each E ∈ A × B the functions x 7→ ν(Ex ) and
y 7→ µ(E y ) are measurable. We define two functions (µ × ν)1 , (µ × ν)2 : A × B → R̄
given by Z
(µ × ν)1 (E) = ν(Ex )µ(dx)
X
Z
(µ × ν)2 (E) = µ(E y )ν(dy)
Y

Claim: (µ × ν)1 , (µ × ν)2 are the same measure.

Let us prove that (µ × ν)1 is a measure.

53
 P
(µ × ν)1 (∅) = 0. Suppose that E = n En is a disjoint union in A × B. For
x ∈ X, wePhave
Ex = n (En )x .
Let y ∈ (Em )x ∩ (En )x . Since y ∈ (Em )x , (x, y) ∈ Em . Since y ∈ (En )x ,
(x, y) ∈ En . Then (x,
P y) ∈ Em ∩ En . This happens iff m = n.
Hence ν(Ex ) = n ν((En )x ). Then
Z Z X XZ X
(µ×ν)1 (E) = ν(Ex )µ(dx) = ν((En )x )µ(dx) = ν((En )x )µ(dx) = (µ×ν)1 (En )
X X n n X n

This proves that (µ × ν)1 is σ-additive.

Similarly (µ × ν)2 is a measure.
Now,
Z Z Z
(µ×ν)1 (A×B) = ν((A×B)x )µ(dx) = ν(B)χA (x)µ(dx) = ν(B) χA (x)µ(dx) = ν(B)µ(A)
X X X

Similarly,
Z Z Z
y
(µ×ν)2 (A×B) = µ((A×B) )ν(dy) = µ(A)χB (y)ν(dy) = µ(A) χB (y)ν(dy) = µ(A)ν(B)
Y Y Y

Thus (µ × ν)1 (A × B) = µ(A)ν(B) = (µ × ν)2 (A × B) i.e. (µ × ν)1 , (µ × ν)2 coincide

in rectangles. Therefore (µ × ν)1 = (µ × ν)2 .
We denote µ × ν := (µ × ν)1 = (µ × ν)2 .
Uniqueness (exerc)
Proposition 165. (Tonelli’s theorem) Let (X, A, µ), (Y, B, ν) σ-finite measure spaces.
Let f : X × Y → [0, +∞]
R A × B-measurable. Then R
a) The function x 7→ Y fx dν is A-measurable, and the function y 7→ X f y dµ is
B-measurable.
b) The function f satisfies
Z Z Z  Z Z 
y
f d(µ × ν) = fx dν µ(dx) = f dµ ν(dy)
X×Y X Y Y X

Proof. Let E ∈ A × B, and suppose that f = χE . Then

fx (y) = f (x, y) = χE (x, y) = χEx (y)

R R
Thus fx = χEx and similiarly f y = χE y . Hence, Y fx dν = ν(Ex ) and X f y dµ =
µ(E y ) for all x ∈ X, y ∈ Y . Then,
Z Z Z Z 
f d(µ × ν) = (µ × ν)(E) = ν(Ex )µ(dx) = fx dν µ(dx)
X×Y X X Y

Similarly
Z Z Z Z 
y y
f d(µ × ν) = (µ × ν)(E) = µ(E )ν(dy) = f dµ ν(dy)
X×Y Y Y X

Hence these equality hold for simple functions, and taking limits they hold for non-
negative A × B-measurable functions.

54
Proposition 166. (Fubini’s theorem) Let (X, A, µ), (Y, B, ν) σ-finite measure spaces.
Let f : X × Y → [−∞, +∞] A × B-measurable and µ × ν-integrable. Then
a) fx is ν-integrable µ-a.e. x ∈ X, and f y is µ-integrable ν-a.e. y ∈ Y .
b) The functions If and Jf defined by
 R
f dν, if fx is ν − integrable;
Y x
If (x) =
0, otherwise .

and  R
X
f y dµ, if f y is µ − integrable;
Jf (x) =
0, otherwise .
belong to L1 (X, A, µ) and L1 (Y, B, ν), respectively.
c) We have Z Z Z
f d(µ × ν) = If dµ = Jf dν.
X×Y X Y

Proof. Consider f + and f − the positive and negative of f . We know that the
sections
fx , (f +R)x , (f − )x are B-measurable. We also know that the functions x 7→ (f + )x dν
R
and x 7→ (f − )x dν are A-measurable.
Then by Tonelli
Z Z Z Z Z  Z Z 
+ − + −
f d(µ×ν) = f d(µ×ν)− f d(µ×ν) = fx dν µ(dx)− fx dν µ(dx)
X×Y X×Y X×Y X Y X Y
Z Z  Z Z  Z
= (fx+ − fx− )dν µ(dx) = fx dν µ(dx) = If dµ
X Y X Y X

The last equality follows because the set N = x : Y fx+ dν = +∞ or Y fx− dν = +∞

 R R
is meas. and If (x) = 0 if x ∈ N .
18/08
Applications:

1) Let (X, A, µ) σ-finite measure space, f : X → [0, +∞] A-measurable

Consider
E = {(x, y) ∈ X × R : 0 6 y 6 f (x)}
Note that E is the region under the graph of f .
(R, B(R), λ) λ Lebesgue measure.
We have a σ-algebra A × B(R) on X × R.
If x ∈ X, then Ex = {y ∈ R : 0 6 y 6 f (x)} = [0, f (x)]. Hence λ(Ex ) = f (x).
If y ∈ R, then E y = {x ∈ X : 0 6 y 6 f (x)} = f −1 ([y, +∞i).
Claim: E ∈ A × B(R)
F : X × R → [−∞, +∞], F (x, y) = y − f (x)
E = {(x, y) : F (x, y) 6 0, 0 6 y}
R R∞
Proposition 167. f dµ = 0 µ({x ∈ X : f (x) > y})dy.

55
Proof. Consider µ × λ on X × R. We have
Z Z Z
(µ × λ)(E) = λ(Ex )µ(dx) = f (x)µ(dx) = f dµ.
X X

On the other hand,

Z Z Z ∞
y
(µ×λ)(E) = µ(E )λ(dy) = µ({x ∈ X : f (x) > y, 0 6 y})dy = µ({x ∈ X : f (x) > y})dy.
R R 0

Therefore, the proposition is true.

2) (Discret case) X = N+ . Consider (X, A, µ), A = 2X , µ=counting measure.

Y = X in the Fubini theorem. Let f : N+ × N+ → [−∞, +∞] µ × µ-integrable.
P
Remark 168. f : N+ × N+ → [−∞, +∞] µ × µ-integrable iff m,n>1 f (m, n)
absolutely convergent.
Then
R P
X×X
f d(µ × µ) = m,n>1 f (m, n).
We have:
fm (n) = f (m, n)
f n (m) = f (m, n)
By Fubini:
Z Z Z  Z ! !
X X X
f d(µ×µ) = fm dµ µ(dm) = fm (n) µ(dm) = f(m, n)
X×X X X X n>1 m>1 n>1

Z Z Z  Z ! !
X X X
f d(µ×µ) = f n dµ µ(dn) = f n (m) µ(dn) = f(m, n)
X×X X X X m>1 n>1 m>1

We conclude that
! !
X X X X X
f (m, n) = f(m, n) = f(m, n)
m,n>1 m>1 n>1 n>1 m>1

16 Change of variable in Rn
Let U, V ⊂ Rn open subsets. Recall that T : U → V is a difeomorphism of class C 1
if T is bijective and
T, T −1 ∈ C 1 .
Example: T : Rn → Rn isomorphism of vector spaces. Then T is a difeom.
Moreover, the Jacobian matrix JT (x) is the associated matrix of T i.e. T 0 (x) = T
for all x ∈ Rn .

Lemma 169. T : Rn → Rn isomorphism of vector spaces. Then

Z
λ(T (B)) = | det(T )|λ(B) = | det(T )|λ(dx)
B
n
for all B Borel subset in R .

56
Proof. First of all, notice that T is a homeomorphism. In part, T, T −1 is Borel
measurable. If B is a Borel subset, then T (B) is a Borel subset.
Note that B(Rn ) = B(R) × · · · × B(R) (n-times).
Particular cases (n = 2):  
1 0
1) T (x, y) = (x, αy), α 6= 0. In this case [T ] = , det(T ) = α
  0 α
0 1
2) T (x, y) = (y, x). In this case [T ] = , det(T ) = −1
1 0 
1 0
3) T (x, y) = (x, x + y). In this case [T ] = , det(T ) = 1.
1 1
Suppose that B = [a, b] × [c, d]. Note that λ(B) = (b − a)(d − c).
1) α > 0.
T (B) = [a, b] × [αc, αd]. Hence, λ(T (B)) = (b − a)(αd − αc) = α(b − a)(d − c) =
αλ(B) = | det(T )|λ(B).
α < 0 we also have λ(T (B)) = | det(T )|λ(B)
2) T (B) = [c, d] × [a, b]. Hence, λ(T (B)) = (d − c)(b − a) = (b − a)(d − c) =
λ(B) = | det(T )|λ(B). S
3) T (B) = {(x, x + y) : a 6 x 6 b, c 6 y 6 d} = x∈[a,b] {x} × [x + c, x + d].
T (B)x = [x R+ c, x + d] R R
λ(T (B)) = [a,b] λ(T (B)x )dx = [a,b] λ([x + c, x + d])dx = [a,b] (d − c)dx = (d −
c)(b − a) = λ(B)=| det(T )|λ(B).

If T : R2 → R2 isomorphism, then T is a product T = T1 ◦ T2 ◦ · · · ◦ Tk , where

each Ti is of the form 1) or 2) or 3).
In this case, det(T ) = det(T1 ) · · · det(Tk ). Hence
λ(T (B)) = λ((T1 ◦ T2 ◦ · · · ◦ Tk )(B)) = | det(T1 )|λ((T2 ◦ · · · ◦ Tk )(B)) = · · · =
| det(T1 ) · · · det(Tk )|λ(B).
This is equal to | det(T )|λ(B).

20/08
Lemma 170. Let T : U → V be a difeomorphism of class C 1 between two open
subsets U, V ⊂ Rn . Let α > 0 be a real number and B a Borel subset of U . Then
a) If | det(JT (x))| 6 α for all x ∈ B, then λ(T (B)) 6 αλ(B).
b) If | det(JT (x))| > α for all x ∈ B, then λ(T (B)) > αλ(B).
Proposition 171. (Change of variables) Let T : U → V be a difeomorphism of
class C 1 between two open subsets U, V ⊂ Rn .
Then for every Borel subset B ⊂ U , one has
Z
λ(T (B)) = | det(JT (x))|λ(dx),
B

and for every Borel measurable function f : V → R, we have

Z Z
f dλ = f (T (x))| det(JT (x))|λ(dx)
V U

whenever these integrals exist.

57
Proof. Let B be a Borel subset subset of U such that µ(B) < +∞. For every
n, k > 1, define
 
k−1 k
Bn,k = x ∈ B : 6 | det(JT (x))| <
n n
k−1
Then n
· χBn,k 6 | det(JT (x))| · χBn,k < nk · χBn,k . Integrating, we have
k−1
Z
k
µ(Bn,k ) 6 | det(JT (x))|λ(dx) 6 µ(Bn,k )
n Bn,k n

On the other hand, by Lemma above we have

k−1 k
µ(Bn,k ) 6 λ(T (Bn,k )) 6 µ(Bn,k )
n n
From these two equalities, we get
 
k−1
Z  k
 1
λ(T (Bn,k )) − | det(JT (x))|λ(dx) 6 µ(Bn,k ) − µ(Bn,k ) = µ(Bn,k )
 
 Bn,k  n n n
P
Since B = k Bn,k , we obtain
 Z 
  1
λ(T (B)) − | det(J T (x))|λ(dx) 6 µ(B)
  n
B

Since µ(B) < +∞, taking limit n → +∞, we obtain,

Z
λ(T (B)) = | det(JT (x))|λ(dx).
B

In the general case, we can write B = ∪k Bk union of a increasing seq (Bk ) such that
µ(Bk ) < +∞. We have for each k,
Z
λ(T (Bk )) = | det(JT (x))|λ(dx).
Bk
R
Taking limit k → +∞, we get λ(T (B)) = B | det(JT (x))|λ(dx).
Let f : V → R Borel measurable function. If f = χT (B) , then

Z Z Z
f dλ = χT (B) dλ = λ(T (B)) = | det(JT (x))|λ(dx)
V V ZB
= χT (B) (T (x))| det(JT (x))|λ(dx)
ZU
= f (T (x))| det(JT (x))|λ(dx)
U

Therefore, Z Z
f dλ = f (T (x))| det(JT (x))|λ(dx)
V U
Hence, we we extend this formula when f is simple, then when f measurable non-
negative. Finally we consider f = f + − f − .

58
 Example: Let R > 0. Define U = {(r, θ) ∈ R2 : 0 < r < R, 0 < θ < 2π} =
h0, Ri × h0, 2πi. Consider T : U → R2 given by
(r, θ) 7→ (r cos(θ), r sin(θ)). Let V = T (U ). Observe V = B(0, R)−{nonnegative x−
axis}, and T : U → V is a difeomorphism of class C 1 . Let f : V → R Borel mea-
surable function. By Change of Variables,
Z Z
f dλ = f (T (x))| det(JT (x))|λ(dx)
V U

We have | det(JT (x))| = r. Hence,

Z Z Z 2π Z R
f dλ = f dλ = f (r cos(θ), r sin(θ))rdrdθ
B(0,R) V 0 0

We get the formula,

Z Z 2π Z R
f dλ = f (r cos(θ), r sin(θ))rdrdθ
B(0,R) 0 0

17 Properties of Haar measure

Let G locally compact group, µ regular Borel measure on G. Let A ⊂ G Borel
subset i.e. A ∈ B(G).
A−1 = {x−1 ∈ G : x ∈ A}.
Write µ̌(A) = µ(A−1 )
Exercise: µ̌ is a regular Borel measure on G.
Proposition 172. µ is a left (resp. right) Haar measure iff µ̌ is a right (resp. left)
Haar measure.
Proof. µ̌(Ax) = µ((Ax)−1 ) = µ(x−1 A−1 ) = µ(A−1 ) = µ̌(A).
Proposition 173. Let G locally compact group, µ is a left Haar measure. Then µ
is finite iff G is compact.
Remark: If G is a compact, then its Haar measure can be normalized i.e. we
obtain a probablity measure. i.e. µ(G) = 1.
Let G locally compact group, µ is a left Haar measure. For each x ∈ G, we
define
µx (A) = µ(Ax)
Exercise: µx is a regular measure on G (=¿µx is a left Haar measure on G)
(Sugg. The map G → G given by y 7→ yx is a homeomorphism. On the other
hand, µx (yA) = µ(yAx) = µ(Ax) = µx (A) ).
We have two Haar measures µ and µx on G (x fix). There is ∆(x) > 0 such that
µx = ∆(x)µ. We obtain a function ∆ : G → R given by x 7→ ∆(x).
∆ is called modular function of G.
  
a b
Example: G = : a > 0, b ∈ R (G, ·) is a loc compact group. We
0 1
have

59
 R 1
Define
 µ(A)  A a2 dadb for A ∈ B(G). µ is a left Haar measure. We have,
=
a b
∆ = 1/a.
0 1
Let ν another left Haar meas. on G. There is c > 0 such that ν = cµ. Then

νx = cµx = c∆(x)µ = ∆(x)ν

Proposition 174. ∆ does not depend on the left Haar measure.

Proposition 175. We have

a) ∆ is continuous
b) ∆(xy) = ∆(x)∆(y) for all x, y ∈ G.

Proof. b) ∆(xy)µ(A) = µ(Axy) = ∆(y)µ(Ax) = ∆(y)∆(x)µ(A).

Then ∆(xy) = ∆(x)∆(y).
a) exerc.
Def: A loaclly compact group G is unimodular if ∆(x) = 1 for all x ∈ G. iff
µx = µ iff µ(Ax) = µx (A) = µ(A) for x ∈ G iff every left Haar measure is a right
Haar measure.

Proposition 176. Every compact group is unimodular.

Proof. G compact group, ∆ modular function. We have

∆(xn ) = ∆(x)n for all x ∈ G, n > 1.
If ∆(x) > 1, then ∆(xn ) → +∞. In part, ∆(G) is unbounded.
If 0 < ∆(x) < 1, then ∆(x−1 ) = ∆(x)−1 > 1, so ∆(x−n ) → +∞. Thus ∆(G) is
unbounded.
Since ∆ is continuous and G is compact then ∆(G) is compact, in part ∆(G) is
bounded.
Therefore, ∆(x) = 1 for all x ∈ G.
Example: On (R) is a compact group, hence unimodular.
Example: SLn (R) is unimodular.

Structure of L1 (G)
Let G locally compact group, µ is a left Haar measure.
In this part, we want to show that L1 (G) is a Banach algebra.

Def: f, g ∈ L1 (G). We define the convolution product as

f (s)g(s−1 t)µ(ds), if s 7→ f (s)g(s−1 t) is integrable;

 R
(f ∗ g)(t) =
0, otherwise .

Proposition 177.
a) The function s 7→ f (s)g(s−1 t) is integrable µ-a.e.
b) f ∗ g is integrable and kf ∗ gk1 6 kf k1 kgk1 .

60
 This proposition implies that L1 (G) is a Banach algebra.

Cohn. Measure theory.

More generally, Rudin. Real and complex analysis.

18 Sobolev spaces
Functional spaces
We want to describe the so called Sobolev spaces.

In this section, we write Lp = Lp (Rn )

Let u ∈ Lp (Rn )
u is a class of a meas. function Rn → R.
How to define partial derivatives of u even when u (its representative) is not
differentiable?
For each j = 1, . . . , n, we define the weak derivatives of u as:
∂j u = fj which satisfies
Z Z
∂ϕ
− udx = ϕfj dx
∂xj

for all ϕ ∈ C0∞ = C0∞ (Rn ) = {C ∞ − functions Rn → R with compact supp}.

ϕ is called test function.

Lemma 178. Let u ∈ C01 . Then

Z
∂u
dx = 0
∂xj

Proof. It follows from the Calculus fund theorem.

Example: Suppose that u ∈ C0∞ . What is ∂j u?
For any ϕ ∈ C0∞ , we have
∂(ϕu) ∂ϕ ∂u
∂xj
= ∂x j
u + ϕ ∂xj
By lemma, Z Z Z
∂(ϕu) ∂ϕ ∂u
0= dx = udx + ϕ dx
∂xj ∂xj ∂xj
R ∂ϕ R ∂u ∂u
Then − ∂x j
udx = ϕ ∂xj dx. Thus ∂j u = ∂x j
is the usual partial derivative.
Therefore, the weak derivatives generalize the usual derivatives.

Def:(Sobolev spaces) We define

H 1,p = {u ∈ Lp : ∂j u ∈ Lp ∀j = 1, .., n}

The index 1 means that the derivatives are of first order.

Lemma 179. H 1,p is vector subspace of Lp .

61
Proof. u, v ∈ H 1,p . Then
Z Z Z Z Z Z
∂ϕ ∂ϕ ∂ϕ
− (u+v)dx = − udx− vdx = ϕ∂j udx+ ϕ∂j vdx = ϕ(∂j u+∂j v)dx
∂xj ∂xj ∂xj

for test funtion ϕ. Thus ∂j (u+v) = ∂j u+∂j v. Since ∂j u, ∂j v ∈ Lp . Then ∂j (u+v) =

∂j u + ∂j v ∈ Lp .
Similarly, if α ∈ R then ∂j (αu) = α∂j u ∈ Lp .

We set for u ∈ H 1,p ,

n
X
kukH 1,p = kukp + k∂j ukp
j=1

Lemma 180. k − kH 1,p is a norm in H 1,p .

Proof. 1) kukH 1,p = 0, then kukp + nj=1 k∂j ukp = 0. Then kukp = 0 =¿ u = 0 in
P

Lp . Therefore u = 0 in HP 1,p
.
2) kαukH 1,p = kαukp + nj=1 k∂j (αu)kp = |α|kukp +|α| nj=1 k∂j ukp = |α|kukH 1,p
P

3) ku + vkH 1,p = ku + vkp + j=1 k∂j (u + v)kp 6 kukp + kvkp + nj=1 (k∂j ukp +
Pn P
k∂j vkp ) = kukH 1,p + kvkH 1,p
Proposition 181. H 1,p is a Banach
Proof. By the previous lemmas H 1,p is a normed vector space.
Let (um ) a Cauchy sequence in H 1,p .
Since kum − ul kp 6 kum − ul kH 1,p , (um ) a Cauchy sequence in Lp . Then there is
u ∈ Lp such that um → u in Lp . Now, since um ∈ H 1,p , ∂j um = fj,m ∈ Lp .
Claim: For each j, (fj,m ) is Cauchy in Lp .
kfj,m − fj,l kp = k∂j um − ∂j ul kp 6 kum − ul kH 1,p .
Then there is fj ∈ Lp such that fj,m → fj . Hence,

Z Z Z
∂ϕ ∂ϕ ∂ϕ
− udx = − lim um dx = − lim um dx
∂xj ∂xj m m ∂xj
Z Z Z
= lim ϕfj,m dx = ϕ lim fj,m dx = ϕfj dx
m m

Thus ∂j u = fj ∈ Lp . Therefore u ∈ H 1,p . Finally

n
X n
X
kum − ukH 1,p = kum − ukp + k∂j um − ∂j ukp = kum − ukp + kfj,m − fj kp
j=1 j=1

Thus um → u in H 1,p .
Proposition 182. For each real number p > 1, C0∞ is dense in H 1,p .
Theorem 183. (Sobolev embedding) If p > n or n = p = 1, then

H 1,p (Rn ) ⊂ C(Rn ) ∩ L∞ (Rn )

62
 Next, we define the Sobolev spaces H k,p (Rn ), where k > 1.
By convention, we set H 0,p (Rn ) = Lp (Rn ).
Notation: α = (α1 , . . . , αn ) ∈ Nn , we denote |α| = α1 + · · · + αn .
∂ α = ∂ α1 ∂ α2 . . . ∂ αn .
Let u ∈ Lp . For each α ∈ Nn , we define the weak derivatives of u as:
∂ α u = fα which satisfies
Z Z
|α| α
(−1) ∂ ϕudx = ϕfα dx

for all ϕ ∈ C0∞ .

Def:(Generalized Sobolev spaces) We define

H k,p = {u ∈ Lp : ∂ α u ∈ Lp ∀|α| 6 k}

From the definition H 1,p coincides with the previous definition, because ∂j u =
∂ (0,...,1,...,0) u, where
(0, . . . , 1, . . . , 0) (1 lies is the position j).

Notation: ∂ (0,...,0,...,0) u = u
We can define u ∈ H k,p ,
X
kukH k,p = k∂ α ukp
|α|6k

Question: Is k − kH k,p a norm?

(exerc)
We have a generalization

Theorem 184. (Sobolev embedding) If kp > n, then

H k,p (Rn ) ⊂ C(Rn ) ∩ L∞ (Rn )

Application:
In PDE, Sobolev spaces are used to find weak solutions of partial diff equation.
The next idea is to regularize these weak solutions.

19 Introduction to probability theory

Various concepts of measure theory are changed in probability
(Ω, A, P ) probability space if it is measure space and P is a probability measure
i.e. P (Ω) = 1.
A measurable set i.e. an element of A is called event.
A measurable function X : Ω → R is called a (real-valued) random variable.
More generally, Let (S, B) be a measurable space. A measurable function X : Ω → S
is called a random variable.
Example:

63
 1) Ω = {0, 1}, A = 2Ω , P ({0}) = 1/2, P ({1}) = 1/2, P (Ω) = 1. This represents
when we toss a fair coin.
2) Ω = {0, 1}2 , A = 2Ω ,
P ({(0, 0)}) = 1/4, P ({(0, 1)}) = 1/4, P ({(1, 0)}) = 1/4, P ({(1, 1)}) = 1/4, P ({(0, 0), (0, 1)}) =
1/2,etc.
Consider a random variable X : Ω → {0, 1, 2} given by

 0, ω = (0, 0);
X(ω) = 1, ω = (1, 0) or ω = (0, 1);
2, ω = (1, 1)


Def: Let (S, B) be a measurable space. The distribution of a random variable

X : Ω → S is the measure PX = P X −1 : B → R given by A 7→ P (X −1 (A)).
Example: Consider the previous example. Then PX : 2{0,1,2} → R is given by
PX = 14 δ0 + 21 δ1 + 14 δ2 .
We have PX (A) = P (X −1 (A)).
If A = {0}, then PX ({0}) = P (X −1 (0)) = P ({0, 0}) = 1/4.
If A = {1}, then PX ({1}) = P (X −1 (1)) = P ({(1, 0), (0, 1)}) = 1/2.
If A = {2}, then PX ({2}) = P (X −1 (2)) = P ({1, 1}) = 1/4.

Notation: Let X : Ω → R random variable.

{X > 0} = {ω ∈ Ω : X(ω) > 0} = X −1 ([0, +∞i)
{X = limn Xn } = {ω ∈ Ω : X(ω) = limn Xn (ω)}, etc
P (X > 0) = P ({X > 0}).
If X : Ω → R is integrable, then its integral is llamado expectation
Z Z
E(X) = XdP = xPX (dx)
R

More generally, f : R → R Borel measurable. then

Z Z Z
E(f ◦ X) = f ◦ XdP = f (x)PX (dx) = f dPX
R R

Def: E(X 2 ) is called second moment of X.

var(X) = E((X − E(X))2 ) is called variance of X.

Lemma 185. If m 6 n, then E(X n ) < +∞ implies E(X n ) < +∞.

Proof. It is consequence of the Jensen inequality: if ϕ is a convex function, then

ϕ(E(X)) 6 E(ϕ(X)).

Lemma 186. Suppose moreover that E(X 2 ) < +∞.

a) var(X) = E(X 2 ) − (E(X))2
b) If a, b ∈ R, then var(aX + b) = a2 var(X)

Proof. a) E((X − E(X))2 ) = E(X 2 − 2XE(X) + E(X)2 ) = E(X 2 ) − 2E(X)E(X) +

E(X)2 = E(X 2 ) − E(X)2 .
b) var(aX + b) = E(((aX + b) − E(aX + b))2 ) = E((aX + b − aE(X) − b)2 ) =
E(a2 (X − E(X))2 ) = a2 var(X).

64
 Def: The distribution function of a r.v. X : Ω → R is a function FX : R → R
given by
FX (x) = P (X 6 x) = PX (h−∞, x])
FX is a nondecreasing right-continuous function.
R
Remark: If F > 0 is a nondecreasing
R Borel function such that F dλ. Then we
have probability P given by P (A) = A F dλ
2
For example, F = √12π R e−x /2 dx.
R

Def: Let (Ai )i∈I family of events in (Ω, A, P ). (Ai )i∈I is independent if
!
\ Y
P Ai = P (Ai )
i∈I0 i∈I0

for all finite subset I0 ⊂ I.

Def: Let (Xi )i∈I family of random variables (Ω, A, P ) → (S, B). (Xi )i∈I are
independent if (Xi−1 (Ai ))i∈I are independent for Ai ∈ B.
Def: Let (Ω, A, P ) prob space. A family of sub-σ-algebras (Ai )i∈I of A is
independent if for each Ai ∈ Ai , the family (Ai )i∈I is independent.

X r.v. σ(X) = {X −1 (A), A ∈ B} is a σ-algebra

Lemma 187. (Xi )i∈I is independent iff (σ(Xi ))i∈I is independent

Proof. Follows from the def.

Proposition 188. Let (Ω, A, P ) prob space, (Ai )i∈I family independent of sub-
σ-algebrasEof A. Let (Sj )j∈J be a partition of I, and for each j ∈ J, let Bj =
D S
i∈Sj Ai . Then (Bj )j∈J is independent.

Example: Let (Xi )i>1 family of independent random variables.

The random var X2i−1 + X2i are independent for i > 1. Consider the partition
{{1, 2}, {3, 4}, . . .} of {1, 2, 3, . . .} and apply the prop above.

References
B [1] Bartle, Elements of integration.

IMCA, lima-Peru
E-mail address: [email protected]

65