EISTI - Departement of Mathematics

Q.F.R.M. M1 2014-15
An Introduction to
MEASURE and INTEGRATION

by Marietta Manolessou

September 18, 2014

Contents

1 Measures 1
1 σ-fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 σ-Field (or σ-Algebra) . . . . . . . . . . . . . . . . . . . . . 1
2 Measurable sets - Measurable mappings - Measures . . . . . . . . . . 2
1 Measurable sets and functions . . . . . . . . . . . . . . . . . 2
2 Properties of measurable functions . . . . . . . . . . . . . . . 3
3 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 Subsets of zero measure . . . . . . . . . . . . . . . . . . . . 6
5 Measure absolutely continuous with respect to another one . . 7

2 The Lebesgue Integral 9

1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 The Lebesgue Integral in a descriptive way . . . . . . . . . . . . . . 9
3 An abstract definition of Lebesgue’s integral
Integration of the numerical positive functions. . . . . . . . . . . . . 12
1 Step functions . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4 Lebesgue integral of positive measurable numerical functions. . . . . 13
1 Lebesgue integral of positive measurable step functions . . . . 13
2 Lebesgue integral of positive measurable numerical functions 13
3 Properties of Lebesgue integral . . . . . . . . . . . . . . . . 13
4 The Lp spaces . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 The Lebesgue theorems 17

1 Lebesgues’ Monotonic Convergence and Dominated Convergence the-
orems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 The Completness of Lp spaces . . . . . . . . . . . . . . . . . . . . . 19

4 Product spaces
Product measures
Fubini theorem 21

iii
5 Probability Spaces L1 , L2 23
1 L1 , L2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1 L2 Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . 23
2 The theorem of Orthogonal Projection . . . . . . . . . . . . . 24
2 How to define a measure from the integral of a measurable function
(Reciprocal of the Radon-Nicodym theorem.) . . . . . . . . . . . . . 25
3 Theorem of Radon - Nikodym . . . . . . . . . . . . . . . . . . . . . 25
List of Figures

k
k2 −1  
X j j j+1
2.1 For k = 3 (shaded): Vn x| ≤ f (x) ≤ . . . . . . 10
j=0
2k 2k 2k
2.2 Even if f (x) is piecewise continuous, the Lebesgue integral can be defined
(an impossible operation in the sense of Riemann). . . . . . . . . . . . . . 11

v
Chapter 1

Measures

1 σ-fields
1 Algebra
Definition 1.1 Let Ω be a set. An Algebra A on Ω is a family of subsets of Ω that
verify the following properties:
i) Ω and the empty set ∅ belong to A
ii) If A ∈ A then AC ∈ A (stability by complementarity)
iii) If A and B ∈ A then (A ∩ B) ∈ A

Example of algebra on R
We define the algebra A(R) on Ω = R, by the family of finite unions of intervals (of
all types) of R :

[a, b], ]a, b[, [a, b[, ]a, b], ] − ∞, a[, [b, +∞[, ]b, +∞[.

2 σ-Field (or σ-Algebra)

Definition 1.2
A σ-field (or σ-algebra) on Ω is a family F of subsets of Ω such that:
• f.1 Ω and the empty set ∅ belong to F.
• f.2 If A ∈ F then AC ∈ F(stability by complementarity)
∞
S
• f.3 If An (n ≥ 1) ∈ F then An ∈ F (stability by countable union)
n=1

1
2 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

Exercise 1.1
Show that F is stable by a countable intersection.

Example 1.1
The set P of all subsets of Ω is obviously an example of σ-algebra.

Example 1.2 the Borel σ-algebra B(R)

In Probability theory, the most usual way to define a σ-algebra, is the following:
We choose a family C of subsets of Ω (among the most interesting for our study),
and we define the σ-algebra F on Ω, as the smallest one which contains all the subsets
in C.
Similarly, we define on R, the Borel σ-algebra B(R), as the smallest σ-algebra
which contains the intervals [a, b] , −∞ < a < b < +∞.
Choosing the usual topology of R, B(R) is the σ-algebra D generated by the set
of intervals: D = {] − ∞, α[; α ∈ R} or D = {] − ∞, α]; α ∈ R} or by the set of
all types of intervals.

2 Measurable sets - Measurable mappings - Measures

1 Measurable sets and functions

Definition 1.3

a) When on a set Ω, we define a σ-algebra F, then, (Ω, F) is called a measurable

space and the subsets of Ω that belong to F are called the measurable sets of Ω.

b) Let (Ω, F) be a measurable space and (Y, TY ) a topological space (cf. the first
chapter of Topology). A mapping f : Ω → Y is measurable if the inverse image
by f of every open set A ∈ OY of Y (where OY ≡ the family of open sets of TY )
is a measurable set, i.e.

if ∀ A ∈ OY ⇒ f −1 (A) ∈ F

Example 1.3 (Exercises) We consider a set Ω exercises

3.1 Let F = P be the example 1.1 (the set of all subsets of Ω), and let (Y, TY ) be a
topological space.
Every function f : Ω → Y is a measurable function with respect to P.
σ-fields (algebras)- Measurable sets and mappings - Measures 3

3.2 We consider the numerical function

∀P ∈ P(Ω), χp : P → R, with
(
1 if x ∈ P
χp (x) =
0 if x ∈
/P

called the index function (another notation I(x)) of the subset P ⊂ Ω.

It is a measurable function iff P is a measurable subset of Ω.
3.3 Generalization of the the Borel σ-algebra - the Borel mapping.
The set O of all open subsets of an arbitrary topological space Ω is not always a
well defined σ-algebra.
For this reason we consider the σ-algebra B = Õ (generated by O) that is called
the Borel σ-algebra.
The elements of B are called the Borel sets or Borelians of Ω.
Every measurable function (or mapping) with respect to this σ-algebra B is a
Borel function. In particular, a continuous function is a Borel function.

Remark 1.1
We verify (in a way analogous to that used for B(R)) that the corresponding Borel
σ-algebras of R2 , R3 etc., may be generated by the associated open sets: rectangles,
parallelepipeds etc.

2 Properties of measurable functions

We present the following results without proof:

Theorem 1.1 (Exercises)

i) Let X, Y and Z be three measurable spaces and let f and g be two measurable
f g
mappings from X to Y and from Y to Z : X −
→Y −
→Z

⇒ the composed mapping g ◦ f : X → Z is a measurable function.

ii) Let (Ω, F) be a measurable space, and let us consider two numerical measurable
functions u and v.

Ω → R×R
⇒ the mapping f : is a measurable function
x → (u(x), v(x))

(here R × R = R2 is endowed with the usual topology).

4 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

Theorem 1.2 (Measurability of the limits)

Let {fn }n=1,2,... be a convergent sequence of numerical measurable functions de-
fined on the measurable space (Ω, F); let f be the simple limit function of this se-
quence,

⇒ f is also a measurable function on (Ω, F).

This last result is very useful for integration theory .

3 Measures
In the following we use the notation R̄ for the “completed interval”: [−∞, +∞] and
R̄+ for the interval [0, +∞].

Definition 1.4
Let (Ω, F ) be a measurable space. We say that the following mapping:

µ : F → R̄+

is a positive measure (or simply a measure) if:

a) (non triviality) ∀A ∈ F, µ(A) is not always equal to +∞

b) when I ⊆ N, for every countable family {Ai }i∈I of disjoint elements of F,

(Ai ∩ Aj = ∅ ∀i 6= j) we have:
∞ ∞
!
[ X
µ Ai = µ(Ai ) (countably additive function).
i=1 i=1

Remarks 1.1

a) If Ω is a topological space and if T is the Borel σ-algebra B, then the measure

defined on B is called the Borel measure.

b) If µ(Ω) is finite then we say that the positive measure µ is bounded.

c) The triplet (Ω, T , µ) is called a measured space.

d) The series of the counting additivity must be absolutely convergent.

One can show that these measures verify the following properties:

Proposition 1.1 [Exercises]

i) µ(∅) = 0
σ-fields (algebras)- Measurable sets and mappings - Measures 5

N
P
ii) µ(A1 ∪ A2 ∪ . . . AN ) = µ(Ai ) (with: (Ai ∩ Aj = ∅ ∀i 6= j))
i=1

iii) If A ⊆ B ⇒ µ(A) ≤ µ(B)

iv) If A1 ⊆ A2 ⊆ . . .⊆ An (Ai are measurable sets)
∞
S
lim µ(An ) = µ An (increasing union continuity).
n→∞ n=1

v) If A1 ⊇ A2 ⊇ . . .⊇ An and
 µ(A1 ) 6= ∞ (Ai are measurable sets)
∞
T
lim µ(An ) = µ An (decreasing intersection continuity)
n→∞ n=1

Exercise 1.2 Proof of the above proposition 1.1 v)

We put:
∞
\
C n = A1 − An and, A= An (1.1)
n=1
thus: C1 ⊂ C2 ⊂ C3 ⊂ ... and

µ(Cn ) = µ(A1 ) − µ(An ) (1.2)

µ(A1 = Cn ∪ An ) = µ(Cn ) + µ(An )

 
because following ii)) z}|{
∀n
and
∞
[
A1 − A = Cn (1.3)
n=1
because
A1 − A = A1 ∩ Ac = A1 ∩(∩An )c
and following de Morgan’s identity:
= A1 ∩ (∪Acn ).
Now, usingSthe distributivity property
A1 − A = n (A1 ∩ Acn ).
c
S∞ ⇒ Cn = A1 ∩ An
However from (1.1)
⇒ A1 − A = n=1 Cn (1.3)
S∞
By application of the increasing union continuity property iv) on n=1 Cn we have:
∞
[
µ(A1 − A) = µ( Cn ) = lim µ(Cn ) = µ(A1 ) − lim µ(An ). (1.4)
n→∞ n→∞
n=1

Moreover:
µ(A1 − A) = µ(A1 ) − µ(A) (1.5)
and again from the property ii) : (A1 ∩ Ac ) ∩ A = ∅ and(A1 ∩ Ac ) ∪ A = A1 so,
µ(A1 ) = µ(A1 − A) + µ(A).
6 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

Finally the right hand sizes of (1.4) and (1.5) yield:

∞
\
lim µ(An ) = µ(A) ≡ µ( An ). 
n→∞
n=1

Example 1.4 Some examples of measures

Let (Ω, F) be a measurable space and let A ∈ F, a ∈ Ω.

1) The following mapping δa : F → R̄+ defined by:


1 if a ∈ A
δa (A) =
0 if a ∈/A
is the so called Dirac measure corresponding physically to a unit mass (or
charge) concentrated at the point a.
2) We call the following mapping counting measure on Ω:
N : F → R̄+
A → N (A) = the number of elements of A
(N (A) is a well defined measure).
3) Borel measure (or Lebesgue measure) on (R, B(R)) or (Rk , B(Rk )).
On R we define the length `(I) of an interval by:

`([α, β]) = β − α, `(] − ∞, α]) = +∞, `([α, +∞[) = +∞

and if A is a finite union of disjoint intervals, we define the length `(A) of A as

the sum of lengths of all intervals, components of A.
In an analogous way we define the area of a Borel set in B(R2 ), the volume of a
Borel set in B(R3 ) etc.
That is precisely the so called the Borel measure on R, R2 , R3 . . . (or Lebesgue
measure).
4) On the the spaces of probability (Ω, F, P ) we have introduced a particular pos-
itive measure P , the so called probability measure (which verifies all the the
preceding properties) and moreover:

∀A ∈ F 0 ≤ P [A] ≤ 1; P [Ω] = 1; P [∅] = 0

4 Subsets of zero measure

Definition 1.5
a) Let (Ω, F, µ) be a measured space.
A subset X of Ω is of zero measure (or a zero measure subset), with respect to
µ, if there is a measurable set A ∈ F, which contains X and such that

µ(A) = 0.
σ-fields (algebras)- Measurable sets and mappings - Measures 7

One easily verifies that the countable union of zero measure sets is also a zero
measure set.

b) A property P associated with every point of Ω is true almost everywhere (de-

noted by a.e.) if it is true at every point of the complement of a zero measure
subset.
c) Two functions f and g defined on Ω are equal “almost everywhere”

⇔ f =g a.e.

if the subset N ⊂ Ω defined by:

N = {x ∈ Ω : such that, f (x) 6= g(x)}

is a zero measure subset.

5 Measure absolutely continuous with respect to another one

Definition 1.6 Let (Ω, F) be a measurable space. A measure ν is absolutely continu-

ous with respect to the measure µ, if ∀ A ∈ F

µ(A) = 0 ⇒ ν(A) = 0

⇔ ν is dominated by µ ⇔ ν ≺≺ µ.

Lemma
If ν is finite then ν ≺≺ µ iff ∀  > 0 ∃ a δ which verifies

∀ A ∈ F ν(A) <  if µ(A) < δ

Definition 1.7 Two measures ν, µ are equivalent iff:

ν ≺≺ µ
{ }
µ ≺≺ ν
8 EISTI - QFRM M1 (2014-15) a course by M.Manolessou
Chapter 2

The Lebesgue Integral

1 Motivation
When the Riemann integral is applied in a more general context than physics, for ex-
ample in the theory of Probabilities and Stochastic Processes, we very often encounter
some technical difficulties due to different reasons, for instance:

• The convergence of some types of integrals having as upper or lower bounds “ill
defined ” points for the integrand function (example: −∞, +∞).
• The integral itself is not well defined in the sense of Riemann or
• we cannot define it if the integrated function has some discontinuities.

This is precisely the role of the Lebesgue integral.

On one hand, the conditions imposed on the integrated functions and the bounds on
the integrals are much weaker and better adapted to physical applications (cf. Fourier
and Laplace transformations).
On the other hand, they constitute the most appropriate and elegant mathematical
context for the study of Probabilities and Stochastic Processes.

2 The Lebesgue Integral in a descriptive way

Notice that in the following we consider only measurable sets and measurable map-
pings.
The corresponding measured space Ω will be Rn with the corresponding Lebesgue
measure.

Definition 2.1 (The Lebesgue Integral)

Let us consider the measurable space ( Rn , B(Rn )) and let

f : Rn → R

9
10 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

be a numerical measurable positive function. We define the Lebesgue integral of f

denoted:
Z Z +∞ Z +∞
f (x)dx or ... f (x1 , . . . , xn )dx1 . . . dxn
Rn −∞ −∞

by:
k
Z k2 −1  
X j j j+1
f (x)dx = lim Vn x| k ≤ f (x) ≤ k (2.1)
Rn k→+∞
j=0
2k 2 2

Here Vn (A) represents the nth -volume of a subset A ⊂ Rn .

k
k2 −1  
X j j j+1
Figure 2.1: For k = 3 (shaded): Vn x| ≤ f (x) ≤
j=0
2k 2k 2k

For n = 1 and k = 3, we give the figures 2.1 and 2.2 adapted from [5] (cf. for a
concrete representation of this definition).

Remark 2.1
The Lebesgue Integral 11

Figure 2.2: Even if f (x) is piecewise continuous, the Lebesgue integral can be defined (an
impossible operation in the sense of Riemann).

a) In simple words and by analogy to the definition of the Riemann integral, we

approximate the Lebesgue integral by the sums of infinite elementary volumes
under the graph of f , but there is now a fundamental difference:

In the case of the Riemann integral we had to realize an infinite partition on the
Ox axis; but in the present case of the Lebesgue integral, we consider an infinite
subdivision of the Oy axis.

When the function is continuous on the considered interval (thus with the Rie-
mann integral being well defined), the two integrals are identical.

b) The Lebesgue integral verifies all properties of Riemann’s integral, in particular

the linearity and monotonicity.
12 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

3 An abstract definition of Lebesgue’s integral

Integration of the numerical positive functions.
1 Step functions
Definition 2.2 Ler (X, F) be measurable space and let g be a numerical positive
function:
g : X 7→ R̄+ .
If the image of g is a finite subset of R+ , we say that g is a step function.

Definition 2.3
Let (X, F) be a mesurable space and let g be a step function.
We consider the following subsets of X:
Ei = {x ∈ X : g(x) = αi } .
Here αi , ∀i = 1, 2, . . . , .n are the different discrete values of the step function g. We
then can write:
X n
g= αi χEi
i=1
χEi is the index function corresponding to the subset Ei
We can easily prove the following:
Proposition 2.1
A step function g defined on (X, F) is measurable iff the subsets Ei are measur-
able:
⇔ iff Ei ∈ F ∀i = 1, 2, . . . , n

2
Theorem 2.1 (Approximation of a measurable function by a increasing sequence of step functions)

Let (X, F) be a measurable space and let f be a numerical measurable function:

f : X → R̄+ .
⇒
There exist measurable step functions defined on X such that

(i) 0 ≤ g1 ≤ g2 . . . ≤ gn · · · ≤ f
(ii) ∀x ∈ X, lim gn (x) = f (x)
n→∞
The Lebesgue Integral 13

4 Lebesgue integral of positive measurable numerical

functions.
1 Lebesgue integral of positive measurable step functions
Definition 2.4
Let (X, F, µ) be a measured space, with µ the Lebesgue measure.
Let
Xn
g= αi χEi
i=1
be a measurable step function defined on X. Here χEi is the index function corre-
sponding to the measurable subset Ei and αi ∀i = 1, 2, . . . , .n are the different discrete
values of the step function g.
For every A ∈ F we define the Lebesgue integral of g on A with respect to the
measure µ, as follows:
Z n
X
gdµ = αi µ(Ei ∩ A)
A i=1

Remark 2.2
(Attention !) Reminder of the convention: 0 × ∞ = 0; , Because it could happen
for example that αi = 0 for some i, but µ(Ei ∩ A) = ∞ .

2 Lebesgue integral of positive measurable numerical functions

Definition 2.5
Let (X, F, µ) be a measured space, with µ the Lebesgue measure, and let f :
(X, F, µ) → [0, +∞] be a positive measurable numerical function.
For every A ∈ F we define the Lebesgue integral of f on A with respect to the
measure µ as follows: Z Z
f dµ = sup gdµ
A g A
Here the sup has to be taken on the set of all measurable step functions such that:
0 ≤ g ≤ f.

3 Properties of Lebesgue integral

Proposition 2.2
Let: (X, F, µ) be a measured space (with µ the Lebesgue measure),
f, fi : (X, F, µ) → [0, +∞] ∀i = 1, 2 F− positive measurable functions,
g, gj : (X, F, µ) → [0, +∞] ∀j = 1, 2, F− positive measurable step functions
E, A, B measurable subsets of X, then:
14 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

(a) Conditions for the integral to be zero

• 1. If g = 0 on E Z
⇒ gdµ = 0
E

• 1’. If f (x) = 0 ∀x ∈ E,
Z
⇒ f dµ = 0 even if µ(E) = ∞
E

• 2. If µ(E) = 0 Z
⇒ gdµ = 0
E

• 2’. If µ(E) = 0
Z
⇒ f dµ = 0, even if f (x) = ∞ ∀x ∈ E
E

(b) Monotonicity
• 3. If 0 ≤ g1 ≤ g2 on E
Z Z
⇒ g1 dµ ≤ g2 dµ
E E

• 3’. If 0 ≤ f1 ≤ f2 on E
Z Z
⇒ f1 dµ ≤ f2 dµ
E E

(c) Properties related to the set of integration

• 4. If A ⊆ B Z Z
⇒ gdµ ≤ gdµ
A B

• 4’. If A ⊆ B Z Z
⇒ f dµ ≤ f dµ
A B

• 5. If A ∩ B = ∅
Z Z Z
⇒ gdµ = gdµ + gdµ
A∪B A B

• 5’. If A ∩ B = ∅
Z Z Z
⇒ f dµ = f dµ + f dµ
A∪B A B
The Lebesgue Integral 15

• 6. ∀E ∈ F Z Z
⇒ gdµ = g χE dµ
E X

• 6’. ∀E ∈ F Z Z
⇒ f dµ = f χE dµ
E X

• 7. Let E1 ⊆ E2 ⊆ . . . be an increasing sequence of measurable subsets

∞
[
in F and let E = En
n=1
Z Z
⇒ gdµ = lim g dµ
E n→∞ En

• 7’. Let E1 ⊆ E2 ⊆ . . . be an increasing sequence of measurable subsets

∞
[
in F and let E = En
n=1
Z Z
⇒ f dµ = lim f dµ
E n→∞ En

(d) Additivity properties

• 8 . If c ∈ R̄+ Z Z
⇒ cgdµ = c gdµ
E E

• 8’. If c ∈ R̄+
Z Z
⇒ cf dµ = c f dµ
E E

• 9 Z Z Z
(g1 + g2 )dµ = g1 dµ + g2 dµ
E E E

4 The Lp spaces
Definition 2.6
Let us consider the measured space (endowed with the Lebesgue measure µ = dx),
(Rn , B(Rn ), µ). We say that a function f : Rn → R is integrable with respect to
Lebesgue or that it is a Lebesgue integrable function, if the integral of |f | in the sense
of the previous definition 2.5 is well defined:
Z
⇔ |f (x)|dx < ∞ (2.2)
Rn

The vector space of all Lebesgue integrable functions is denoted by L1 .

16 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

More generally for 1 ≤ p < ∞, the vector space of all p-Lebesgue integrable
functions is denoted by Lp ;
Z
p
i.e. f ∈ L means that: |f |p dx < ∞
Rn

We introduce the following norm (often referred to as N p norm):

Z 1/p
kf kp = |f |p dx (2.3)
Rn
Chapter 3

The Lebesgue theorems

We present here three of the most important theorems relative to Lebesgue integration.

1 Lebesgues’ Monotonic Convergence and Dominated

Convergence theorems
The following first two theorems ensure the commutativity between the operations
of limit and integration. The first is known as Lebesgue’s theorem I (or theorem of
“Monotonic Convergence”) , and the second as Lebesgue’s theorem II (or theorem
of “Dominated Convergence”).
They are very useful to the theory of Probabilities and Stochastic Processes, in
particular in order to obtain the definitions and results concerning the distribution and
characteristic functions of the limits of sequences of random variables.

Theorem 3.1 ( Fatou’s Lemma)

Let (X, F, µ) be a measured space (with µ the Lebesgue measure), {fn }n=1,2,...,
a sequence of positive measurable functions on (X, F),
(fn : (X, F, µ) → [0, +∞] ∀n = 1, 2 . . . )
Z Z
⇒ lim inf fn dµ ≤ lim inf fn dµ
X n→∞ n→∞ X

One of the most important applications of Fatou’s Lemma is the proof of Lebesgue
Dominated Convergence, that we present below for the case X = Rn .

Theorem 3.2 (Lebesgue I - Monotonic Convergence) Let us consider the measured

space (Rn , B(Rn ), µ) (with µ = dx the Lebesgue measure on Rn ). Let {fn }n=1,2 be
a sequence of measurable functions with values in R̄+ and such that for every element
x ∈ Rn we have:
0 ≤ f1 (x) ≤ f2 (x) ≤ . . .

17
18 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

⇒
the function f defined by: f (x) = lim fn (x) is measurable and we have:
n→+∞
Z Z
lim fn dx = f dx (3.1)
n→+∞ E E

for every measurable set E ⊂ Rn .

Theorem 3.3 (Lebesgue - Dominated Convergence I)

Let us consider the measured space (Rn , B(Rn ), µ) (with µ = dx the Lebesgue
measure on Rn ).
Let {fn }n=1,2... be a sequence of measurable functions on Rn , pointwise conver-
gent to a function f and bounded by an integrable function g:

∀ n = 1, 2, . . . |fn | ≤ g,

⇒ the function f is Lebesgue integrable and for every measurable subset E ⊂ Rn we

have : Z Z Z
lim |fn − f |dµ = 0 and f dx = lim fn dx (3.2)
n→∞ E E n→∞ E

If the above sequence {fn }n=1,2... is “constant ”, i.e. fn = f ∀ n ∈ N, then we

obtain the analogous dominated convergence property of Riemann’s integral:

Theorem 3.4 (Lebesgue - Dominated Convergence II )

We consider the measured space (Rn , B(Rn ), µ) (with µ = dx the Lebesgue
measure on Rn ). Let f be a measurable function on Rn , and bounded by an integrable
function g:
|f | ≤ g,
⇒ the function f is Lebesgue integrable.

Theorem 3.5 Lebesgue - Dominated Convergence theorem in Lp

Let us consider the measured space (Rn , B(Rn ), µ) (with µ = dx the Lebesgue
measure on Rn ). Let {fn }n=1,2... be a sequence of measurable functions on (Rn , B(Rn )),
pointwise convergent to a function f . If there exists a real function g ∈ Lp such that:

∀ n = 1, 2, . . . |fn | ≤ g, µ − a.e.

⇒
the functions fn and f belong to Lp and the sequence {fn }n=1,2... converges to f
µ-a.e.
The Lebesgue theorems 19

2 The Completness of Lp spaces

The following result holds:

Theorem 3.6
∀ p : 1 ≤ p < ∞, Lp is a Banach space in the topology induced by the norm of
definition 2.3 .

Exercise 3.1 (Proof of theorem 3.6)

The reason that motivated the presentation of this proof is to recall and apply, in a
synthetical way, certain concepts and properties of metric topological spaces, together
with some of the above results.
The proof consists in showing that Lp is a complete normed vector space. In other
words we prove that every Cauchy sequence of functions in Lp converges inside this
space. We use the Lebesgue - Dominated Convergence theorem 3.5 in the Lp spaces.
Let us consider the measured space (endowed with the Lebesgue measure µ = dx),
(Rn , B(Rn ), µ) and let {fn }n≥1 be a Cauchy sequence of measurable functions with
values in (R, BR ) pointwise convergent to a function f .
We consider a subsequence {fni }i≥0 such that ∀ n ≥ ni

1
kfn − fni kp ≤ (3.3)
2i
We show that the subsequence {fni }i≥0 converges µ-a.e. in Lp to the function f .
Reminder: for a Cauchy sequence to be convergent inside a metric space it is
sufficient for one of its subsequences to converge inside this space. We write:
k
X
fnk = fn0 + (fni − fni−1 ) (3.4)
i=1

but
k
X
f nk − f n0 = (fni+1 − fni ) = Sk (partial sum of a series)
i=0

Let us define:
k
X
gk = |fni − fni−1 |. (3.5)
i=1

The sequence {gk } is an increasing one ⇒ it converges to a finite or infinite func-

tion g. Now,
k k k
X X X 1
kgk kp ≤ kfni − fni−1 kp ≤ kfni − fn kp + kfn − fni−1 kp ≤ i−1
≤2
i=1 i=1 i=1
2
(3.6)
20 EISTI - QFRM M1 (2014-15) a course by M.Manolessou

⇒ kgkp ≤ 2
⇒ g finite (µ-a.e.) and,
lim Sk = g (absolutely)
k→∞

⇔ The subsequence {fnk }k≥0 converges absolutely µ-a.e. to the function f .

Moreover:
|fnk | ≤ |fn0 | + |gk | ≤ |fn0 | + g ∈ Lp . (3.7)
Then, we apply the Lebesgue - Dominated Convergence theorem in Lp theorem 3.5

⇒ the subsequence {fnk }k≥0 → f in Lp

For n ≥ ni we have
1
kf − fn kp ≤ kf − fni kp + kfni − fn kp ≤ kf − fni kp +
2i
1
⇒ ∀ i, lim sup kf − fn kp ≤ kf − fni kp +
n→∞ 2i
(3.8)
1
and with i → ∞ →0
2i

⇒ lim sup kf − fn kp = 0 
n→∞
Chapter 4

Product spaces
Product measures
Fubini theorem

Definition 4.1 Reminder: Complete Measure

Let (X, F, µ) be a measured space and let F ∗ be the family of subsets E ⊂ X such
that ∃ A and B ∈ F with A ⊂ E ⊂ B and µ(B − A) = 0; we define

µ(E) = µ(A)

⇒ F ∗ is a σ-algebra (σ-field) and µ is a complete measure on F ∗ .

F ∗ is called the “completed” σ-algebra of F.

Definition 4.2
Let (X, F, µ) and (Y, F, ν) be two measured spaces endowed with complete mea-
sures µ and nu respectively.
On the product space X × Y , we consider the family of subsets of the form A × B
with A ∈ FX , B ∈ FY (called measurable pavers) and we define a function of subsets
on this family by:
λ(A × B) = µ(A)ν(B)
We can show that λ can be extended to a “complete measure” denoted by µ × ν on the
“completed” σ-algebra generated in X × Y by the “measurable pavers” A × B and
denoted by FX × FY .

And now , a small...exercise to synthesize several notions presented previously.

21
22 EISTI- QFRM.M1-(2014-15)- Measures Integration- by M.Manolessou

Proposition 4.1
Let (X × Y, FX × FY ) be a measurable product space and let E ∈ FX × FY of
finite measure, then:

• a)

We define: Ex = {y ∈ Y : (x, y) ∈ E} ⇒ Ex ∈ FY for almost x.

• b) The function φ defined almost everywhere by φ(x) = ν(Ex ) is FX measur-

able.
R
• c) X
φdµ = (µ × ν)(E)

Theorem 4.1 (Th. of Fubini)

Let (X × Y, FX × FY , λ)) be a measured product space and let f be an integrable
function on X × Y with respect to the product measure λ = µ × ν

a) For almost every y (resp. x) the function fy (resp. fx ) defined by fy (x) =

f (x, y) (resp. fx (y) = f (x, y)) is an integrable function on X (resp. Y ).
R R
b) The function Φ (resp. Ψ) denoted by X fy dµ, (resp. Y fx dν) defined by
Z Z
Φ(y) = fy dµ (resp. Ψ(x) = fx dν)
X Y

is integrable on Y (resp. on X)
c) Z Z  Z Z Z 
fy dµ dν = f d(µ × ν) = fx dν dµ
Y X X×Y X Y

Usual notation:
Z Z
f dµdν instead of f d(µ × ν).
X×Y X×Y
Chapter 5

Probability Spaces L1, L2

1 L1 , L2
L1 ≡ Vector space of random variables which are integrable in the sense of Lebesgue:
Z
E[|X|] = |xf (x)|dx < +∞

L2 ≡ Vector space of “square integrable random variables” in the Lebesgue sense.

X : (Ω, A, P ) → R
Z
E[X 2 ] = x2 fX (x)dx < +∞

1 L2 Hilbert space
Scalar Product
Definition 5.1 Let (Ω, A, P ) a probability space
Let (X, Y ) ∈ L2 × L2 (couple of random variables defined on (Ω, A, P )2 ). We in-
troduce the scalar product on L2 ×L2 by the following positive definite non degenerate
symmetrical bilinear form:

< X, Y >= E[X · Y ] ⇔ ( correlation function)

The corresponding norm is:

kXk2 = E[X 2 ] (N.O)

(moment of order 2)

Remark 5.1 A random variable X which has a finite norm is called of order 2.

⇔ kXk2 = E[X 2 ] < +∞

23
24 EISTI- QFRM.M1-(2014-15)- Measures Integration- by M.Manolessou

Proposition 5.1
Let (Ω, A, P ) be a probability space.
(i) L2 endowed with the distance associated with the norm (N.O) is complete.
⇔
it is a Hilbert space (and Banach space).
(ii) The set of all centered random variables (⇔ E[X] = 0) on Ω is a subset of L2
denoted by H0
⇐⇒ ∀ X ∈ H0 , E[X] = 0
(iii)
∀ (X, Y ) ∈ H0 × H0 , Cov(X, Y ) = hX, Y i = E[X · Y ]
and
hX, Y i
Corr. (X, Y ) =
kXkkY k

2 The theorem of Orthogonal Projection

Theorem 5.1
Let (Ω, F, P ) be a probability space and let X ∈ L2 be a random variable defined
on (Ω, F, P ):
(i)
Var [aX|y] = a2 Var [X|y]
ii)
Let F 0 ⊂ F 00 ⊂ F
(⇔ every event of the σ-algebra F 0 , also belongs to the σ-algebra F 00 and every
event of the σ-algebra F 00 also belongs to the σ-algebra F)

⇒
E[E[X|F ]|F ] = E[E[X|F 00 ]|F 0 ] = E[X|F 0 ]
0 00

Theorem 5.2 (Orthogonal Projection)

Let (Ω, F, P ) be a probability space, X ∈ L2 a random variable defined on
(Ω, F, P ). and let F 0 ⊂ F,
⇒ ∀ Y ∈ L2 (F 0 ) ( random variable defined on (Ω, F 0 , P ))
E[{X − E[X|F 0 ]}2 ] = min E[{X − Y }2 ]
Y ∈L2 (F 0 )

⇔ the vector E[X|F 0 ] is the Orthogonal Projection of X on the subspace L2 (F 0 ) ⊂

L2 (F)
⇔ ∀ Y ∈ L2 (F 0 ) : hX − E[X|F 0 ], Y i = 0
Product spaces and measures-Fubini 25

2 How to define a measure from the integral of a mea-

surable function
(Reciprocal of the Radon-Nicodym theorem.)

Theorem 5.3
Let (X, F, µ) be a measured space (with µ the Lebesgue measure),
and f : (X, F, µ) → [0, +∞]
R a measurable function such that:
∀ E ∈ F, φ(E) = E f dµ

⇒ φ is a measure on F and

∀ measurable function g defined on X with values in R̄+

Z Z
g dφ = g f dµ (5.1)
X X

Remarks 5.1

• The function f is called the density of the measure φ with respect to the measure
µ.
• The equation (5.1) is usually written in the symbolic form dφ = f dµ
• For the probability spaces (Ω, F, P ), f is identified with the probability density
distribution function with respect to the probability measure P .

3 Theorem of Radon - Nikodym

Let (Ω, F) be a measurable space and let P and P ∗ be two probability measures such
that P ≺≺ P ∗ ;
⇒
there exists a density function f such that.
Z
P (A) = f dP ∗ ∀A∈F
A

and for two density functions f and g

P ∗ [f 6= g] = 0 a.s. (almost surely)

26 EISTI- QFRM.M1-(2014-15)- Measures Integration- by M.Manolessou
Bibliography

[1] G. ALMOUZNI a) “Signaux Systèmes” b) “Traitement du signal ” (Manuscripts

of courses given at EISTI 2012-13)
[2] G. ARFKEN “Mathematical Methods for Physicists” (Academic Press 1985)
[3] P. BILLINGSLEY “Probability and Measure" (Wiley Interscience, 1995)
[4] Ph. BLANCHARD and E. BRUNING “Mathematical Methods in Physics”
(Birkhauser, 2003)
[5] P. BREMAUD “An Introduction to Probabilistic Modeling” (Springer, 1988)
[6] R.COURANT and D.HILBERT “Methods of Mathematical.Physics” (Iner-
science Publishers, 1953)
[7] P. DENNERY and A. KRZYWINSKI “Mathematics for Physicists” (Harper-
Internat. “Dover”, 1996)
[8] J. DIEUDONNE “Foundations of Modern Analysis” (Academic Press N.York,
1979)
[9] B. FRIEDMAN “Principles and Techniques of Applied Mathematics” (J.Wiley
and Sons, 1966)
[10] I.M. GEL’FAND and G.E. SHILOV “Generalized functions” (Academic
Press,1968)
[11] A. GUICHARDET “Intégration - Analyse Hilbertienne” (X-Polytechnique-
Ellipses, 1989)
[12] W. RUDIN “Principles of Mathematical Analysis” (McGraw-Hill, 1976)
[13] M.A. SPIEGEL “Advanced Mathematics” SCHAUM’S Series (McGraw Hill,
1990)
[14] L. SCHWARTZ “Théorie de Distributions” (Hermann, 1973)
[15] A.J. WEIR “Lebesgues-Integration and Measure” (Cambridge Univ.Press, 1988)

27