Convergence, Integration and Probability

Exercises
John Cagnol, Lionel Gabet, Erick Herbin, Pauline Lafitte, Alexandre Richard

2023-2024
 Contents

I T OPOLOGY OF METRIC SPACES AND NORMED VECTOR SPACES . . . . . . . . . . . 2

II H ILBERT SPACES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
III M EASURABLE SPACES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
IV M EASURE SPACES , INTEGRALS AND L p SPACES . . . . . . . . . . . . . . . . . . 21
V T HE L EBESGUE MEASURE AND INTEGRAL . . . . . . . . . . . . . . . . . . . . 28
VI P ROBABILITY AND MEASURE . . . . . . . . . . . . . . . . . . . . . . . . . 35
VII P RODUCT MEASURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
VIII F OURIER TRANSFORM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
IX P ROBABILITY IN d AND INDEPENDENCE . . . . . . . . . . . . . . . . . . . . 53
X C HARACTERISTIC FUNCTIONS AND G AUSSIAN VECTORS . . . . . . . . . . . . . . 57
XI C ONVERGENCE OF RANDOM VARIABLES . . . . . . . . . . . . . . . . . . . . 62
XII C ONDITIONAL EXPECTATION AND INTRODUCTION TO STOCHASTIC PROCESSES . . . . 67

1
 Chapter I : Topology of metric spaces and normed vector spaces

NB1: At the beginning of each lab, Section A) contains the list of minimal knowledge and know-hows that
are required to pass the course.
NB2: Questions marked with a ⇤ or ⇤⇤ are optional. They may go beyond the scope of this course and are not
examinable. However, they may help you have a better understanding of the material. If you choose to skip these
questions (which you are allowed to do), you can use the result of this question to answer the next questions of
the exercise.

A) Aims of this class

After this class,

• in a normed vector space, or more generally in a metric space, I know the definitions of an open
set, a closed set, the closure of a set and of compact set;

• I understand the formal concept of convergence, I know how to study the convergence of a
sequence and the continuity of a function;

• I am able to find the limit superior and limit inferior of a sequence and of a function;

• I know what a Cauchy sequence and a complete normed vector space are, and more generally
what a complete metric space is;

• I am able to prove that a function is a distance on a set or a norm on a vector space;

• I am able to determine the balls for a given distance, understand the proofs using double inclu-
sion and the equivalence of norms.

2
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions I.1 and I.2 must be done before the examples class. The solutions for these exercises are
available online.

Topology is a branch of mathematics at the core of modern analysis. It is interesting to know

some of its basic principles even though it will not be evaluated in the CIP course. Besides, it will
be interesting to observe the similarities between the axioms of topology and the axioms of measure
theory.

Question I.1 (Some questions of topology in )

Q. I.1.1 Let X = and T be the usual topology on X. Which ones of these sets are open?

1. (0; 4)
2. ( •; 4]
3. ( •; 0) [ (2; 4)
4. {0}
Q. I.1.2 For x = ( x1 , x2 ) in 2 define

1. N1 ( x ) = | x1 |
2. N2 ( x ) = | x1 | + | x2 |
3. N3 ( x ) = | x1 | + | x2 |2
4. N4 ( x ) = | x1 | + | x2 | + 1
5. N5 ( x ) = x1 + x2
Which mappings are norms on ?

Question I.2 (Completeness in )

1
Q. I.2.1 Prove that the sequence (un )n2 ⇤ defined by un = n is a Cauchy sequence in .

CS, 1st year 3 2023-2024

CIP 2023-2024

C) Exercises

Exercise I.1 (Metrics and limits in n )

Questions are independent from one another.
E. I.1.1 Let d E denote the Euclidean distance of d . For any a 2 d , consider the mapping Na :
x 7! d E ( a, x ), defined for any x 2 d . For which value(s) of a is Na a norm on d ?
E. I.1.2 Consider the mapping d : ⇥ ! defined by d( x, y) = |e x e y| for all ( x, y) 2 2.

Prove that d is a distance on .

E. I.1.3 Let (un )n2 2 . Check the following properties:

(i ) lim sup un < l ) 9n 2 such that 8k n, uk < l

n!+•
(ii ) 9n 2 such that 8k n, uk < l ) lim sup un  l
n!+•
(iii ) lim sup un > l ) 8n 2 , 9k n such that uk > l
n!+•
(iv) 8n 2 , 9k n such that uk > l ) lim sup un l.
n!+•

Write the corresponding properties for the lim inf.

Exercise I.2 (Topology on n )

Questions are independent from one another.
E. I.2.1 For any x, y 2 2 , define
(
k x yk if x, y and 0 are aligned
d( x, y) =
k x k + kyk otherwise.

(a) Prove that d is a distance. SNCF is the French railway company. Look at the French rail route
map and guess why d is facetiously called the SNCF distance function.

(b) Describe geometrically the open balls.

(c) Prove that any open ball for the usual topology can be written as a union of open balls for the
topology induced by d. Is the converse statement true? We say the topology of d is finer than the
usual topology.

*E. I.2.2 Let B( x, r ) denote the Euclidean ball of n, centred in x with radius r > 0. Let us define
the following collection of sets:
n
O = {O ⇢ : 8 x 2 O, 9r > 0 such that B( x, r ) ⇢ O} .

Show that O is a topology on n (which is called the usual topology of n ).

Completeness is a very convenient property. Complete spaces have various useful properties: the
Cauchy criterion, normal convergence of series of functions, fixed point theorems, extension theorem,
projections, etc.

CS, 1st year 4 2023-2024

CIP 2023-2024

Exercise I.3 (Completeness and applications)

E. I.3.1 Prelude (example of a simple non-complete space): Consider the distance d on that was
defined in question E.I.1.2. Prove that the sequence of integers (i.e. un = n, 8n 2 ) is a Cauchy
sequence for this distance. Is ( , d) a complete metric space?
Let E be a complete normed vector space.
*E. I.3.2 Prove that for any series of E-valued functions, normal convergence implies uniform con-
vergence.
*E. I.3.3 Prove that the space of E-valued continuous functions defined on an interval [ a, b] is com-
plete for the metric induced by the supremum norm.
E. I.3.4 Deduce the continuity on of the mapping
+•
sin(nx )
x 7! Â n2
.
n =1

Exercise I.4 (The space of continuous functions)

Consider the space C of continuous functions from [0, 1] to , and the three following norms: for any
f 2 C,
Z 1 ✓Z 1
◆ 12
2
k f k• = sup | f ( x )|, k f k1 = | f ( x )|dx, k f k2 = | f ( x )| dx .
x 2[0,1] 0 0

E. I.4.1 Compare these three norms.

E. I.4.2 Let E0 = { f 2 C : f (0) = 0}. Find the closure of E0 for all the previous norms.
*E. I.4.3 Let E1 = { f 2 C : f is differentiable} and EP = { f 2 C : f is a polynomial on [0, 1]}.
Find the interior of E1 and EP for k · k• .

D) Going further

Exercise I.5 (Examples in topology)

*E. I.5.1 Let X = { a, b, c}. Among the following sets, which ones are topologies?

1. {∆, { a, b, c}}
2. {∆, { a}, {b}, {c}, { a, b}, { a, c}, {b, c}, { a, b, c}}
3. {∆, { a, b}, {b, c}, { a, b, c}}
4. {∆, { a}, {c}, { a, c}, { a, b, c}}
*E. I.5.2 Let X = {1, 2, 3, 4, 5} and T = {∆, {1, 2}, {5}, {1, 2, 5}, {1, 2, 3, 4}, X } a topology on X. Let
n
(un )n2 be defined by un = 3+(2 1) .
1. Does (un )n2 converge toward 1?

CS, 1st year 5 2023-2024

CIP 2023-2024

2. Does (un )n2 converge toward 2?

3. Does (un )n2 converge toward 3?
4. Does (un )n2 converge toward 4?
5. Does (un )n2 converge toward 5?

Exercise I.6 (Topology)

The two questions are independent.
*E. I.6.1 Let X and Y be two topological spaces and f be a mapping from X to Y.

(a) Prove that if f is continuous, then for any subset A ⇢ X, f ( A) ⇢ f ( A). You may want to check
that f is continuous if and only if for any closed subset F of Y, f 1 ( F ) is a closed subset of X.
Then, you may want to use this characterization of continuity by closed sets.
We wish to prove the reverse statement. Thus in the sequel, we always assume that for any A ⇢ X,
f ( A ) ⇢ f ( A ).
⇣ ⌘
(b) Assume f 1 f ( A) is not a closed subset. Show this assumption leads to a contradiction.

(c) Let B 1 ( B)
⇣ be a ⌘closed subset of Y. Prove that there exists A ⇢ X such that f ( A) ⇢ B and f =
f 1 f ( A) . Deduce that f is continuous.

E. I.6.2 Let ( X, d) be a metric space. Let F ⇢ X be a non-empty closed set. For any x 2 X, set
d( x, F ) = infy2 F d( x, y).
(a) We wish to study the continuity of x 7! d( x, F ). What are the topologies being used?

(b) Prove that x 7! d( x, F ) is a continuous mapping and that d( x, F ) = 0 , x 2 F.

Exercise I.7 (How useful lim⇣sup ⌘and lim inf can be)
n
Let a > 0 and set Sn = Ânk=1 a+k
n . We wish to compute limn!+• Sn .
e a +1
E. I.7.1 Note that for any n 1 and x > n, (1 + nx )n  e x . Prove that lim supn!+• Sn  e 1.
a j n
E. I.7.2 Now prove that for any fixed k 2 and any n > k, Sn Âkj=0 (1 + n ) . Deduce an
inequality on lim infn!+• Sn and conclude.

Exercise I.8 (Compactness)

The questions are independent.
*E. I.8.1 Let ( X, O) and ( X 0 , O 0 ) be two Hausdorff topological spaces, and let f : X ! X 0 be a
continuous mapping.
(a) Is the preimage of a compact set by f compact?

(b) Is the image of a compact set by f compact?

*E. I.8.2 Let ( X, O) be a Hausdorff topological space. Let K be a compact subset of X and x 2 X \ K.
Prove that there exist two open subsets U, U 0 ⇢ X such that K ⇢ U, { x } ⇢ U 0 and U \ U 0 = ∆. [Hint:
by the Hausdorff property, one may consider for each k 2 K some open sets Uk and Vk , such that k 2 Uk , x 2 Vk
and Uk \ Vk = ∆.]
Extend this property for K and K 0 which are two disjoint compact sets.

CS, 1st year 6 2023-2024

CIP 2023-2024

E. I.8.3 Let ( X, d) be a metric space. Prove Cantor’s intersection theorem: if (Kn )n2 is a non-
increasing sequence of nonempty compact sets of X, then \n2 Kn 6= ∆. Prove that any open set
containing \n2 Kn also contains all the Kn for n large enough.

Exercise I.9 (Completeness and applications (continued))

Let E be a Banach space.
**E. I.9.1 Let X be a metric space and X0 be any dense subset of X. Prove the following statement,
that is a generalization of Theorem I.4.7: any uniformly continuous mapping defined on X0 with
values in E admits a unique extension to X and this extension is also uniformly continuous.
*E. I.9.2 Let F be the set of all sequences with values in [0, 1].
(a) Prove that
+•
| xn yn |
d( x, y) = Â 2n
n =0

is a distance on F.

(b) Prove that ( F, d) is a complete metric space.

Exercise I.10 (Picard iterations)

Let d 2 ⇤ , T > 0 and f be an d -valued fonction defined on d such that f is globally Lipschitz
continuous with constant `, i.e.
d
8 x, y 2 , k f (x) f (y)k  `k x y k.

Let y0 2 d .
Let us define the mapping
⇣ ⌘ ⇣ ⌘
d d
F : C([0, T ], ), k · k• ! C([0, T ], ), k · k •
j 7! F ( j)

where
Z t
8t 2 [0, T ], F ( j)(t) = y0 + f ( j(s)) ds.
0

E. I.10.1 Assume in this question that T = 1 and ` < 1. Prove that F is a contraction and deduce
that the sequence of functions ( jk )k2 defined by
⇢ 0
y if k = 0
jk (t) = 0
Rt
y + 0 f ( jk 1 (s)) ds if k 1,

converges uniformly toward a function y and that

Z t
8t 2 [0, T ], y ( t ) = y0 + f (y(s))ds.
0

This is known as Picard iterations.

E. I.10.2 It is no longer assumed that T = 1 et ` < 1. Prove that there exists n 2 ⇤ such that
F n = F · · · F is a contraction.

CS, 1st year 7 2023-2024

CIP 2023-2024

E. I.10.3 Deduce that the sequence of functions (yk )k2 defined by

⇢ 0
y if k = 0
yk (t) =
F n (yk 1 )(t) if k 1,

converges uniformly toward a function y and that

Z t
8t 2 [0, T ], y ( t ) = y0 + f (y(s))ds.
0

*E. I.10.4 Now let d 2 ⇤, T > 0 and f be a d -valued fonction defined on [0, T ] ⇥ d such that

8 x 2 d , t 7! f (t, x ) is continuous in t
8t 2 [0, T ], x 7! f (t, x ) is globally Lipschitz continuous, meaning that:

d
8t 2 [0, T ], 9`(t) > 0 such that 8 x, y 2 , | f (t, x ) f (t, y)|  `(t) | x y |.

Let t0 2 [0, T ] and y0 2 d. Define the sequence of function ( jk )k2 by

(
y0 if k = 0
jk (t) = Rt
y0 + t0 f (s, jk 1 (s)) ds if k 1.

Prove ( jk )k2 converges uniformly toward a function y and that

Z t
8t 2 [0, T ], y ( t ) = y0 + f (s, y(s))ds.
t0

Exercise I.11 (Is there a fixed point?)

Let ( E, k · k) be a normed vector space, K a compact subset of E and f : K ! K a continuous mapping.
Assume that f satisfies for any x 6= y 2 K,

k f (x) f (y)k < k x y k.

*E. I.11.1 Prove that f admits a unique fixed point (you may use the mapping x 7! k x f ( x )k instead
of the iterative procedure used in class). What happens if K is only assumed to be closed?

Exercise I.12 (Closed graph theorem, compact version)

Let ( X, d) be a metric space and ( X 0 , d0 ) be a compact metric space. Note that X ⇥ X 0 is endowed
with the product topology, i.e. the coarsest topology that makes the projection mappings continuous
(in particular, a sequence ( xn , yn ) 2 X ⇥ X 0 converges if and only if ( xn ) and (yn ) converge) Let
f : X ! X 0 be a function and let G f = {( x, f ( x )), x 2 X } be its graph (a subset of X ⇥ X 0 ).
E. I.12.1 If f is continuous, prove that G f is closed in X ⇥ X 0 .
E. I.12.2 Conversely, assume that G f is closed in X ⇥ X 0 . Prove that f is continuous.

CS, 1st year 8 2023-2024

 Chapter II : Hilbert spaces

A) Aims of this class

After this class,

• I know the characterisation of the continuous linear mappings;

• I can recognise a Hilbert space and prove that a sequence converges (with Cauchy sequences);

• I can write a vector in a Hilbert basis;

• I can compute the orthogonal projection of a vector on a closed convex subset of a Hilbert space.

9
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Question II.1 must be done before the examples class. The solutions for these exercises are available
online.

Question II.1 (Various questions on Hilbert spaces)

Les questions sont indépendantes.
Q. II.1.1 Let H be a Hilbert space and F, G ⇢ H. Prove the following statements:

(a) F ⇢ G ) G ? ⇢ F ? ;

(b) F ? \ G ? ⇢ ( F + G )? ;

(c) F ⇢ F ?? ;

(d) F + G = H ) F ? \ G ? = {0};

Q. II.1.2 Let x, x 0 2 H and r, r 0 > 0 such that the closed balls B( x, r ) and B( x 0 , r 0 ) are equal. Prove
that x = x and r = r 0 . [Remark: observe that this property holds in general in normed vector spaces.]
0

Q. II.1.3 Let H be a Hilbert space and ( f n )n2 be a sequence of orthogonal vectors. Prove that
• f
n =0 n converges in H if and only if • 2
n=0 k f n k H converges in .

Q. II.1.4 Let H be a Hilbert space and B its unit closed ball. Check that the projection theorem can
be applied, then prove that the projection P from H to B satisfies P( x ) = k xxk for any x 2 H \ B and
P( x ) = x for x 2 B.

CS, 1st year 10 2023-2024

CIP 2023-2024

C) Exercises

Exercise II.1 (The `2 ( ) space)

Denote by `2 ( ) the space of square integrable sequences, i.e. `2 ( ) = {(un )n2 2 : Ân2 u2n <
• }.
Questions 1 to 4 are independent, but are all dealing with the `2 ( ) space.
E. II.1.1 Let us verify that `2 ( ) is a Hilbert space.

(a) Prove that `2 ( ) is a pre-Hilbert space for the inner product hu, vi = Ân2 un vn (you will first
check that this quantity is well defined for u, v 2 `2 ( )).
We shall now prove that `2 ( ) is complete. Let (u(k) )k2 be a Cauchy sequence in `2 ( ).

(b) Let e > 0. Since (u(k) )k2 is a Cauchy sequence, there exists K > 0 such that for any j, k K,
j k ( j) (k)
ku ( ) u k  e. Prove that for any n 2 , |un
( ) un |  e. Deduce that for any n 2 ,
(k) (•)
limk!• un exists. Denote by un this limit.
(K )
(c) Prove that there exists N 2 such that Ân N | u n |2  e2 .
(•) 1 (•)
(d) Deduce that for any M N, one has (Â N n M |un |2 ) 2  2e, where the terms un have been
defined in question (b).
(•)
(e) Prove that u(•) = (un )n2 is in `2 ( ), and deduce that (u(k) )k2 converges towards u(•) in
`2 ( ).
E. II.1.2 Prove that any separable Hilbert is isometrically isomorphic to `2 ( ).
j(n)
*E. II.1.3 Let j : ⇤ ! ⇤ be a one-to-one mapping. Prove that •
n =1 n2
= •. [Hint: you may
study the sign of Ânk=1 1
k Ânk=1 j(1k) , for n 2 ⇤ .]
*E. II.1.4 Set C = { x = ( xn )n2 2 `2 ( ) : 8n, xn 0}.

(a) Prove that C is a closed convex set.

(b) Determine the projection mapping on C. [Hint: First, you can try to guess the projection in dimension
2, then verify that your solution works also in higher dimensions.]

Exercise II.2 (Fréchet-von Neumann-Jordan Theorem)

Consider a real Banach space ( E, k · k) and assume that the norm fulfills the parallelogram identity:

8 x, y 2 E, k x + y k2 + k x y k2 = 2 k x k2 + k y k2 .

We now define
1
8 x, y 2 E, k x + y k2 k x k2 k y k2
h x, yi =
2
and we will prove that h·, ·i is an inner product.
E. II.2.1 (a) Let x, y, z 2 E. Prove that h x + y, zi + h x y, zi = 2h x, zi. Deduce that

8 x, y, z 2 E, h x + y, zi = h x, zi + hy, zi.

CS, 1st year 11 2023-2024

CIP 2023-2024

(b) Prove that h x, yi = h x, yi and h2x, yi = 2h x, yi.

(c) Prove that hlx, yi = lh x, yi, for any l 2 and any x, y 2 E. You will start proving the result
for l 2 , then for l 2 and finally l 2 .

(d) Conclude.

Exercise II.3 (Adjoint operator)

Let ( H, h·, ·i) be a real Hilbert space and let u : H ! H be a continuous linear mapping.
E. II.3.1 Let x 2 H. Prove that there exists a unique y 2 H such that for any z 2 H, hu(z), x i =
hz, yi. Denote by u⇤ the mapping which takes x and maps it to y. We call this mapping the adjoint of
u.
E. II.3.2 Prove that u⇤ is a continuous linear mapping and that ku⇤ k  kuk.
E. II.3.3 Prove that u⇤⇤ = u and deduce that ku⇤ k = kuk.
E. II.3.4 Prove that ku⇤ u k = k u k2 = k u u ⇤ k.
*E. II.3.5 Prove that Ker(u⇤ ) = Im(u)? and determine Ker(u⇤ )? (you can use that the orthogonal of a
sub-vector space of H is closed, see question E.II.4.3).
E. II.3.6 Consider now H = `2 ( ) endowed with the inner product studied in Exercise II.1.

(a) Let

S: H !H
( un )n2 7 ! (0, u0 , u1 , . . . ).

Prove that S is a continuous linear mapping and compute its adjoint.

(b) Let (an )n2 be a bounded sequence of real numbers, and let

T: H !H
( un )n2 7 ! ( an un )n2 .

Prove that T is a continuous linear mapping and compute its adjoint.

D) Going further

Exercise II.4 (Questions about the topology of Hilbert spaces)

Let H be a Hilbert space. The following questions are independent.
*E. II.4.1 Assume that H is separable, i.e. that there exists a countable dense subset of H. Prove that
the unit closed ball of H is not compact. [Hint: consider a Hilbert base and compute the distance between
two elements of this base.]
*E. II.4.2 Recall that any sub-vector space of a finite-dimensional vector space is closed (prove it
if you cannot remember why). Assuming that H is separable, consider a Hilbert base (en )n2 of H.

CS, 1st year 12 2023-2024

CIP 2023-2024

Recall that Vect{en , n 2 } = H. Prove that Vect{en , n 2 } is not closed. [Hint: consider u N =
ÂnN=1 n1 en .]
Deduce that Vect{en , n 2 } 6= H.
*E. II.4.3 Let M be a sub-vector space of H (not necessarily closed). Prove that M? is closed and
?
that M = M? .

Exercise II.5 (Féjer’s Theorem)

In this exercise, the following notations are used: for n 2 , en is the mapping defined by en ( x ) = einx ;
S N = Â|n| N en and VN = Vect{en : |n|  N }. Set also K N = N1+1 ÂnN=0 Sn for any N 2 , which is the
sequence of Féjer’s kernels.
Recall that for 2p-periodic, continuous functions defined on , the convolution product is given
by Z
1
( f ⇤ g)( x ) = f ( x y) g(y) dy .
2p [0,2p ]
E. II.5.1 We will prove Féjer’s theorem. Let f : ! be a 2p-periodic, continuous function.

(a) Prove that f ⇤ S N is a trigonometric polynomial (that is a linear combination of the en ), that we
will call Fourier series of f .

(b) If x 2 \ 2p , prove the following identity:

!2
1 sin( 12 ( N + 1) x )
KN (x) = .
N+1 sin( 2x )

1
R 2p
(c) Prove that K N 0 and that 2p 0
K N ( x ) dx = 1. Deduce from the previous questions that for
any t 2]0, p ], K N converges uniformly to 0 on ]t, 2p t[.

⇤(d) Deduce from the previous questions that f ⇤ K N converges uniformly to f on .

E. II.5.2 Application of the previous theorem: let f be 2p-periodic, continuous function. Denote by
S N ( f ) the partial sums of its Fourier series. Prove that if f satisfies kS N ( f )k•  1 for all N 2 , then
k f k•  1.

CS, 1st year 13 2023-2024

 Chapter III : Measurable spaces

A) Aims of this class

After this class,

• I can verify that a collection of sets is a s-algebra;

• I can determine the s-algebra generated by a collection of sets;

• I know the Borel s-algebra;

• I can prove that a function is measurable;

• I know the link between continuous and measurable functions;

• I can prove that a function is measurable by approximation with a sequence of measurable func-
tions.

• I can define and construct a probability measure on a discrete set.

14
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions III.1 and III.2 must be done before the examples class. The solutions for these exercises
are available online.

Question III.1

Q. III.1.1 Define W = { a, b, c, d}.

(a) What is the s-algebra generated by { a} ?

(b) What is the s-algebra generated by { a, b} ?

(c) What is the s-algebra generated by { a} and { a, b} ?

(d) Compare the s-algebras obtained in (a), (b) and (c) with respect to the inclusion property.

Q. III.1.2 What is the s-algebra of generated by 1 , i.e. the smallest s-algebra such that 1 is
measurable?
Q. III.1.3 Verify that any constant function is measurable.
Q. III.1.4 Is the set of all open sets of a s-algebra?

Question III.2

Q. III.2.1 In this question, we consider functions from ( , B( )) to ( , B( )).

(a) Prove that the indicator function of is measurable.

(b) Prove that the sine function is measurable.

(c) Set the function f such that f ( x ) = exp( x ) for all x > 0 and f ( x ) = 0 for all x  0. Prove that f
is measurable.

CS, 1st year 15 2023-2024

CIP 2023-2024

C) Exercises

In measure theory (thus analysis) and probability, the concept of s-algebra is quite important. In
the classes of Analysis and PDEs, for the majority of the time, we will see the discrete s-algebra of
and the Borel s-algebra of n .
In probability however, s-algebra represent the set of possible outcomes over a set W, and thus
there is no particular reason to choose or . First, the probability exercises will allow you to bridge
the gap with the discrete probability you have seen in your previous studies. We will then quickly
move on to the modern approach.

Exercise III.1

E. III.1.1 Two six-sided dice are rolled. Let’s consider the mapping j that takes the two numbers
written on the dices and returns their sum.

(a) What is the codomain of this mapping? Propose a s-algebra for this codomain.

(b) What is the domain of this mapping? Propose a s-algebra for this domain.

(c) Depending on your choice of s-algebras, is j a measurable mapping? (in probability theory, a
measurable mapping will be called a random variable.)

Exercise III.2 (Continuity and measurability)

E. III.2.1 Denote by T the usual topology of . Prove that if f : ( , T ) ! ( , T ) is continuous,

then f : ( , B( )) ! ( , B( )) is measurable. You can start by proving first the following facts: 1)
the pre-image by f of any open set is an open set; 2) F = { B 2 B( ) : f 1 ( B) 2 B( )} is a s-algebra.
Prove that the reciprocal statement is wrong by providing a counter-example.
E. III.2.2 Let f and g be functions from to . Prove that if f is continuous and g is (Borel-
)measurable, then f g is measurable. Is f g continuous?

Exercise III.3 (The Borel s-algebra)

E. III.3.1 ⇤(a) Prove that any open set of can be written as an at most countable union of open
intervals.

(b) Deduce that the Borel s-algebra of , denoted by B( ), is generated by the set of all intervals of
the form ( •, a), a 2 .
E. III.3.2 Prove that B( ) is generated by the set of all compacts subsets of (i.e. the closed and
bounded subsets of ).
E. III.3.3 (a) Prove that the following set

C = {A ⇢ : A or \ A is finite or countable}

is a s-algebra on . Recall that a countable union of countable sets is countable.

(b) Is this s-algebra equal to B( )?

CS, 1st year 16 2023-2024

CIP 2023-2024

(c) Is the s-algebra generated by the singletons equal to the Borel s-algebra B( )?
*E. III.3.1 Is the set of all finite or countable unions of intervals a s-algebra?

Measurable functions are the central objects of integration theory. In probability theory, they are
called random variables.
The three following exercises develop classical methods around measurable functions.

Exercise III.4 (Measurability)

E. III.4.1 In this question, we consider the functions from a measurable space (W, F ) to ( , B( )).

(a) Let ( f n )n2 be a sequence of measurable functions. Prove that for any t 2 , { x : lim supn!+• f n ( x ) > t}
is measurable. Same for { x : lim infn!+• f n ( x ) > t}.

(b) Deduce that { x : limn!+• f n ( x ) exists} is measurable and that if the function limn!+• f n exists,
then it is measurable.

(c) Assume now that (W, F ) = ( , B( )). Prove that if f is differentiable, then its derivative f 0 is
measurable.

E. III.4.2 In this question, we consider functions from ( , B( )) to ( , B( )).

(a) Define the function f such that f (0) = 0 and f ( x ) = sin(1/x2 ) if x 6= 0. Prove that f is measur-
able (you may use question 1)(b)).

(b) Prove that the function f ( x ) = x1{ x2 } is measurable but not piecewise continuous.

⇤(c) Prove that if f is monotonic, then it is measurable.

E. III.4.3 In this question, we consider functions from ( 2 , B( 2 )) to ( , B( )).

(a) Prove that the mapping ( x, y) 7! sin( xy) is measurable.

(b) Define f such that f (0, 0) = 0 and

x 2 y2
f ( x, y) = if ( x, y) 6= (0, 0).
x 2 + y2

Prove that f is measurable.

E. III.4.4 In this question, we consider the functions from a measurable space (W, F ) to ( , B( )).

(a) Determine the measurable functions when F = {∆, W}.

(b) Determine the measurable functions when F = P (W).

Exercise III.5 (Approximation of a measurable function)

Let f : E ! + [ {+•} be some mapping. For all n 1 and all i 2 {0, 1, . . . , n 2n 1}, set An = { x 2
E : f ( x ) n} and Bn,i = { x 2 E : i 2 n  f ( x ) < (i + 1) 2 n }.
n2n 1
For all n 2 , define the mapping f n = Â i2 n
1 Bn,i + n1 An .
i =0

CS, 1st year 17 2023-2024

CIP 2023-2024

E. III.5.1 Prove that ( f n ) is pointwise-convergent to f (i.e. for any x 2 E, f n ( x ) converges to f ( x )).

Can the convergence be uniform?
E. III.5.2 Deduce that any Borel function h : ! + can be approached by a sequence of simple
functions.
E. III.5.3 Can this result be extended to any Borel function h : ! , which is not necessarily
non-negative?

Until now, probability theory looks like a mere application of measure theory where measures
have mass 1 and may potentially be discrete (i.e they have support in and this is combinatorics).
New concepts will appear later to show that probability theory is much more than that (independence
is a first glimpse). In the meantime, here are some elementary probability exercises.

Exercise III.6 (Counting)

E. III.6.1 In a meeting with n people, for which value of n would you accept to bet that at least 2
people were born on the same day?

Knowing precisely the law of a random variable allows to solve some practical problems.

Exercise III.7 (Binomial law and insurance)

With the intent of changing its rates, an insurance company conducted a risk evaluation survey. It
found its policy-holders probability to get involved in a car accident depends on how long they have
had their driver’s license:

• 20% of the policy-holders have had their license for less than 5 years.

• The study reports that policy-holders that had their license for less than 5 years have a probabil-
ity 0.4 to get involved in an accident within a year.

• This probability becomes 0.125 for someone who has had his license for more than 5 years.

E. III.7.1 Picking randomly 10 people having their driver’s license for less than 5 years, what is the
probability that at least one of them gets involved in an accident within a year?
E. III.7.2 Same question with 10 people having their driver’s license for more than 5 years.
E. III.7.3 Now if 10 policy-holders are picked randomly among all policy-holders, what is the
probability that at least one of them gets involved in an accident within a year?

Exercise III.8 (Bernoulli law, binomial law)

Let ( Xn )n2 ⇤ be a sequence of independent random variables distributed according to the same Bernoulli
law of parameter p 2 (0, 1), i.e. ( Xn = 1) = p and ( Xn = 0) = 1 p for all n 1.
E. III.8.1 Prove that the random variable Sn = X1 + X2 + · · · + Xn follows a binomial law B(n, p),
defined by ✓ ◆
n
8k 0, ( Sn = k ) = p k (1 p ) n k .
k

CS, 1st year 18 2023-2024

CIP 2023-2024

E. III.8.2 Prove that if limn!• (n pn ) = l > 0, then the binomial law B(n, pn ) can be approximated
by the Poisson law of parameter l, i.e. that for any fixed k 0,

l lk
( Sn = k ) ! e as n ! •.
k!
Hint: You may write
✓ ◆
n nk n n 1 n k+1
= ... .
k k! n n n

Exercise III.9 (Beyond a reasonable doubt...)

In the United States, the “due process clauses” of the Fifth and Fourteenth Amendments protect the
defendant against conviction except “upon proof beyond a reasonable doubt”. The reasonable doubt
standard is a key to the criminal procedure that aims at reducing the risk of convictions resting on
factual error. While the reasonable doubt stardard is usually not (yet?) quantified and not (yet?)
expressed in terms of probability (and appelate courts have condemned such attempts whenever they
have encountered them), mathematical modeling and probability can still be used in court to convince
a jury. The Chicago metropolitan area (also known as Chicagoland) has a population of about ten
million people. Assume a suspect has unusual characteristics with a probability of 10 7 independently
of other individuals.
E. III.9.1 What is the average number of people in Chicagoland with these unusual characteris-
tics? Knowing that the police have found someone with these unusual characteristics, what is the
probability that there is someone else who was then in Chicagoland?
E. III.9.2 If the police find two such people, what is the probability that there is a third one in
Chicagoland?
E. III.9.3 How many people with these characteristics should the police find to be convinced that
they have found them all?
E. III.9.4 How unlikely should the characteristics be to convince you the defendant can (reason-
ably) be identified uniquely?

D) Going further

Exercise III.10 (Infinite s-algebras)

The goal of this exercise is to prove that there is no infinite s-algebra which is countable.
Let ( E, E ) be a measurable space such that E contains an inifinite number of elements. For any x 2 E,
define A x = \ A2E :x2 A A.
*E. III.10.1 Prove that if A x \ Ay 6= ∆, then A x = Ay .
*E. III.10.2 Assuming that E is at most countable, prove that:
(a) 8 x 2 E, A x 2 E ;

(b) any B 2 E satisfies B = [ x2 B A x ;

CS, 1st year 19 2023-2024

CIP 2023-2024

(c) the previous union is countable and the decomposition of B on the A x ’s is unique.

**E. III.10.3 Consider A = { A x } x2E . Prove that A contains an infinite number of elements (you
may assume the opposite and obtain a contradiction). Deduce that E is not countable.

Exercise III.11

E. III.11.1 Let f be a function defined on the measurable space (W, F ). Prove that if { x : f ( x ) r}
is measurable for all r 2 , then f is measurable from (W, F ) to ( , B( )).

Exercise III.12 (Bertrand’s paradox)

A chord of the unit circle is chosen at random. What is the probability that an equilateral triangle
having this chord as one of its sides is contained in the disk? In this exercise, we shall prove that
different constructions of the “random chord” lead to different solutions, hence the problem is ill-
posed.
Denote by U the unit circle centred in O, and let X be the length of the random chord [ A, B], where
A and B are on the circle U . Determine the probability that an equilateral triangle with side [ A, B] is
contained in the disk, in the following three situations:
E. III.12.1 A point P is picked uniformly at random inside the disk, and the chord is constructed
so that P is the middle of [ A, B].
E. III.12.2 Choose uniformly at random a radius of U , and choose uniformly at random a point P
of this radius. The chord is then constructed so that P is the middle of [ A, B].
E. III.12.3 The points A and B are chosen uniformly at random and independently on the circle U .
E. III.12.4 Comment on the difference between these results.

CS, 1st year 20 2023-2024

 Chapter IV : Measure spaces, integrals and L p spaces

A) Aims of this class

After this class,

• I can verify that a function defined on a collection of sets is a measure;

• I know how to construct the integral of a non-negative function with respect to a measure;

• I know how to interchange limits and integrals for a non-decreasing non-negative sequence of
functions;

• I can apply the framework of measure theory to probability;

• I know what a probability measure is and I can determine the distribution of a discrete random
variable;

• I am able to apply the monotone and dominated convergence theorems;

• I know the vector space L p .

21
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions IV.1 and IV.2 must be done before the examples class. The solutions for these exercises
are available online.

Question IV.1 (Various questions about measures)

Q. IV.1.1 Let E1 , E2 be two sets and G be a s-algebra over E2 . Let f be a mapping from E1 to E2 .
Prove that { f 1 ( B) : B 2 G} is a s-algebra over E1 (called the s-algebra generated by f , denoted by
s( f ). You may check that this is the smallest s-algebra that makes f measurable).
Q. IV.1.2 Let ( E, F ) be a measurable space. For x 2 E, define

dx :F ! +
(
=1 if x 2 A
A 7! dx ( A) =
=0 otherwise.

Prove that dx is a measure.

Q. IV.1.3 Is the sum of two measures a measure?

Question IV.2

Q. IV.2.1 Let (W, F , ) be a probability space, and let ( An )n2 be a sequence of events such that
S
( An ) = 0, 8n 2 . Prove that B = n2 An is an event having probability 0.
Q. IV.2.2 More generally, consider now ( E, F , µ) a measure space such that µ( E) < •. Let
( An )n2 2 F be such that for any n 2 , µ( An ) = µ( E). Prove that µ (\n2 An ) = µ( E).

CS, 1st year 22 2023-2024

CIP 2023-2024

C) Exercises

The integral of a function f : E ! with respect to some measure n is denoted indinstinctly by

Z Z
f ( x ) n(dx ) or f dn.
E E

The following exercises involve the integral that has been constructed during the lectures. In prob-
ability theory, the integral with respect to a probability measure is called expectation and is denoted by
[·].

First, let us prove that the integral is general enough to coincide with sums of sequences on an
appropriate measure space.

Exercise IV.1 (Measure on )

Let W = and F = P ( ).
E. IV.1.1 Verify that F is a s-algebra on W.
E. IV.1.2 Verify that n = Card is a measure on F (called the counting measure).
E. IV.1.3 Verify that any function f from ( , P ( )) to ( , B( )) is measurable.
E. IV.1.4 Let f be a non-negative mapping from ( , P ( )) to ( , B( )).
n
(a) Prove that  f ( k ) 1{k} " f .
k =0
Z
(b) Deduce the value of f dn.

E. IV.1.5 Determine the function defined from ( , P ( ), n) to ( , B( )) which are integrable.

E. IV.1.6 Prove that for any + -valued family ( xi,j )(i,j)2 ⇥ , the following equality holds in +:

  xi,j =   xi,j .
i2 j2 j2 i2

Exercise IV.2 (Markov’s inequality)

E. IV.2.1 Let ( E, F , µ) be a measure space and f be a measurable function with values in . Prove
the Markov inequality:
Z
1
8a > 0, µ ({ x 2 E : | f ( x )| > a})  | f | dµ .
a E

E. IV.2.2 Let f be a measurable function from E to . Prove that

µ ({ x 2 E : f ( x ) = +•})  lim µ ({ x 2 E : | f ( x )| n}) .

n!+•

Using the Markov inequality, deduce that if f 2 L1 ( E, F , µ), then µ ({ x 2 E : f ( x ) = +•}) = 0. Is

the reciprocal statement true?

CS, 1st year 23 2023-2024

CIP 2023-2024

Exercise IV.3 (Pushforward measure)

Let ( E, E ) and ( F, F ) be two measurable spaces. Let f be a measurable mapping defined from ( E, E )
to ( F, F ). Let n be a measure defined on the s-algebra E .
Set
8 B 2 F , n f ( B) = n f 1 ( B) .
Consider a measurable mapping j : F ! + .
E. IV.3.1 Verify that n f defines a measure on the s-algebra F .
Z
E. IV.3.2 Recall how the integral is constructed in the particular case of j( x ) n f (dx ).
F
E. IV.3.3 Deduce that
Z Z
j( x ) n f (dx ) = j( f ( x )) n(dx ).
F E

The measure n f is called the pushforward measure of n by f .

Exercise IV.4 (Construction of measures)

Let ( E, F , µ) be a measure space and f be a non-negative measurable function. For all A 2 F , define
Z
n( A) = f dµ .
A

E. IV.4.1 Justify rigorously that n is a measure.

We say that f is the density of n with respect to µ.

To start with probability exercises requiring measure theory.

Exercise IV.5
Let W = {w1 , w2 , w3 }.
E. IV.5.1 Is there a probability measure on W such that ({w1 }) = ({w2 }) = ({w3 })?
E. IV.5.2 Define the mappings X and Y from W to such that

X (w1 ) = 1, X (w2 ) = 2, X (w3 ) = 3,

Y (w1 ) = 2, Y (w2 ) = 3, Y (w3 ) = 1.

(a) Justify that X and Y are random variables.

(b) Prove that X and Y have the same law.

(c) Determine the law of X + Y and XY.

Exercise IV.6
Set X = {1, 2, 3, 4} and F = {{1, 2}, {2, 3}, X }.
E. IV.6.1 Prove that F generates P ( X ) (i.e. s(F ) = P ( X )). Is F a s-algebra?
E. IV.6.2 Denote by M1 ( X ) the set of all probability measures over ( X, P ( X )). Prove that there is a
bijective mapping between M1 ( X ) and { x = ( x1 , x2 , x3 , x4 ) 2 [0, 1]4 : Â4i=1 xi = 1}.

CS, 1st year 24 2023-2024

CIP 2023-2024

E. IV.6.3 Use the previous question to find two probability measures (i.e. such that µ1 ( X ) =
µ2 ( X ) = 1) wich are distinct on P ( X ) while equal on F .

Exercise IV.7 (Probability of limits of events)

Let ( An )n2 ⇤ be a sequence of events of the probability space (W, F , ). Define, for every n 1,
[ \
Bn = Am , Cn = Am .
m n m n

For any n 1, Cn ⇢ An ⇢ Bn and the sequences ( Bn )n2 ⇤ and (Cn )n2 ⇤ are respectively decreasing
and increasing. Then define
\ \ [
lim Bn = B = Bn = Am ,
n!•
n2 ⇤ n2 ⇤ m n
[ [ \
lim Cn = C = Cn = Am .
n!•
n2 ⇤ n2 ⇤ m n

E. IV.7.1 Justify that

(a) B = {w 2 W : w 2 An , for infinitely many n 2 ⇤ }.

(b) C = {w 2 W : w 2 An , for all n 2 ⇤ except for finitely many n}.

E. IV.7.2 The sequence ( An )n2 ⇤ is said to converge to A if B and C are equal to A. In this question,
assume that An converges to A, as n ! •.
(a) Prove that A is an event, meaning that A 2 F .
(b) Prove that ( An ) converges to ( A), as n ! •.
E. IV.7.3 Assume that for any n 2 ⇤, the events Bn and Cn are independent.
(a) Prove that B and C are independent.
(b) Deduce that when An converges to A, ( A) equals either 0 or 1.

D) Going further

Exercise IV.8 (Uniform continuity of the integral)

Let ( E, F , µ) be a measure space and let f 2 L1 (µ).
*E. IV.8.1 Prove that:
Z
8e > 0, 9d > 0 such that µ( A) < d ) | f | dµ < e .
A

Exercise IV.9 (Hölder’s inequality)

Let ( E, F , µ) be a measure space, and f , g be two real-valued measurable mappings. We will prove
Hölder’s inequality: for any p, q 2 [1, •] such that 1p + 1q = 1,
Z ✓Z ◆ 1p ✓Z ◆ 1q
p q
| f g| dµ  | f | dµ | g| dµ . (IV.1)
E E E

CS, 1st year 25 2023-2024

CIP 2023-2024

E. IV.9.1 Prove it in the case k f k p = 0 and k f k p = •.

Assume now that k f k p 2 (0, +•) and k gkq 2 (0, +•).
E. IV.9.2 Prove the inequality when p = 1.
Assume now that 1 < p < •.
E. IV.9.3 (a) Let a 2 (0, 1). Prove that 8 x 2 +, xa ax  1 a. Deduce that for any u, v 0,
ua v1 a  au + (1 a)v.
| f ( x )| p | g( x )|q
(b) Apply the previous inequality to a = 1p , u = k f kp
p and v = k gkq
q . Conclude.

*E. IV.9.4 When 1 < p < •, find a necessary and sufficient condition for the equality to hold in
(IV.1).

Exercise IV.10 (Measure or not? )

Let `• = u = (un )n2 2 : kuk• := supn2 |un | < • be the set of bounded real-valued se-
quences.
*E. IV.10.1 Prove that (`• , k · k• ) is a Banach space.

We admit the following property (which is a consequence of the Hahn-Banach theorem), which
states that there exists a continuous linear form F : `• ! such that: for all u 2 `• ,

• F (u)  kuk• ,

• if limn!• un = a, then F (u) = a.

E. IV.10.2 For any A ⇢ , recall that 1 A is the indicator function of A, defined by the following
relation: 1 A (n) = 1 if n 2 A, and 1 A (n) = 0 otherwise. Hence one can consider 1 A as an element of
`• .
Define µ : P ( ) ! by µ( A) = F (1 A ). Prove that

• µ(∆) = 1 µ( ) = 0,

• µ( Ac ) = 1 µ ( A ),

• µ( A [ B) = µ( A) + µ( B) if A \ B = ∆.

E. IV.10.3 Prove that µ is not a measure.

Exercise IV.11
Let X be an uncountable set and define

M = { E ⇢ X : E is countable or Ec is countable}.

E. IV.11.1 Prove that M is a s-algebra.

E. IV.11.2 Define µ : M ! {0, 1} as µ( E) = 0 if E is countable and µ( E) = 1 otherwise. Prove that
µ is a measure on ( X, M).
**E. IV.11.3 Describe the measurable functions from ( X, M) to ( , B( )) and their integral with
respect to µ.

CS, 1st year 26 2023-2024

CIP 2023-2024

Exercise IV.12 (Euler’s totient function)

Pick uniformly at random an integer in 1, 2, . . . , n. For any integer p such that 0 < p  n, consider the
event
A p = {the integer picked at the beginning can be divided by p} .
E. IV.12.1 Compute ( A p ) when p divides n.
E. IV.12.2 Prove that if p1 , p2 , . . . , pk are prime factors of n which are distinct, then the events
A p1 , A p2 , . . . , A pk are independent.
E. IV.12.3 Recall that Euler’s totient function f : ! is defined as: for all n 2 , f(n) counts the
number of integers in {1, . . . , n} that are relatively prime to n.
Prove that for any n 2 ,
f(n) = n ’ (1 1p ).
p2P
p|n

Exercise IV.13 (Conditional independence)

Two events E and F are conditionally independent with respect to the event C if

(E \ F | C) = ( E | C ) ( F | C ).

E. IV.13.1 Prove that E and F can be independent, but not conditionally independent with respect
to C.

CS, 1st year 27 2023-2024

 Chapter V : The Lebesgue measure and integral

A) Aims of this class

After this class,

• I know the characterisation of the Lebesgue measure on the intervals of ;

• I can explain the difference between the Lebesgue and the Riemann integrals;

• I can compute integrals of real-valued functions;

• I know that functions which are equal almost everywhere have the same integral;

• I know the relation between integration and differentiation;

• I can define and study the convergence of Fourier series in L2 .

28
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions V.1 and V.2 must be done before the examples class. The solutions for these exercises
are available online.

The most common measure on endowed with the Borel s-algebra is the Lebesgue measure,
denoted by l. It naturally generalizes the concept of length.

Question V.1 (Various questions)

Q. V.1.1 Denote by l the Lebesgue measure on .

(a) Justify that l({ x }) = 0 for any x 2 .

(b) Based on (a), one could make the following calculation:

!
[
l( ) = l {x} = Â l({ x }) = 0.
x2 x2

But this is in contradiction with the fact that l( ) = +•.

Where is the mistake?

Q. V.1.2 Let F and G be two s-algebras. Is F [ G a s-algebra?

Question V.2

Q. V.2.1 Let f be theR identity function of [0, 1]. Prove that f 2 L1 ([0, 1], l).
Q. V.2.2 Compute [0,1] f dl without using a Riemann integral. State the theorem you use and
check carefully its assumptions.

CS, 1st year 29 2023-2024

CIP 2023-2024

C) Exercises

We recall that the integral of a function f : E ! with respect to some measure n is denoted
indinstinctly by
Z Z
f ( x ) n(dx ) or f dn.
E E
One of the most common measures is of course the Lebesgue measure, which generalizes length and
volumes. When the term integral is used, by default it means the integral with respect to a measure.
Hence if the measure is the Lebesgue measure, then we may talk about the Lebesgue integral. While it
will be specified systematically that an integral is a Riemann integral when it appears. The following
five exercises focus on the Lebesgue measure.

Exercise V.1 (A warning concerning the interchange of limits and integrals: a few examples on
( , B( ), l).)

E. V.1.1 (a) For all n 2 ⇤ , set f n ( x ) = 1[0,n] ( x ). Prove that f n converges pointwise to some
R
function f that you will determine. Prove that f dl = +•. Does the monotone convergence
theorem apply here?
Z R
(b) For all n 2 ⇤ , set f n ( x ) = n1 1[n,•[ ( x ). Compute lim f n dl and lim f n dl. Explain why
n!+• n!+•
the monotone convergence theorem does not apply here?
(c) For all n 2 ⇤, set f n ( x ) = n1 1[0,n] ( x ) and f ( x ) = 0. Prove that f n converges uniformly to f , but
Z Z
that lim f n dl 6= lim f n dl. Explain why the monotone convergence theorem does not
n!+• n!+•
apply here.

Exercise V.2 (Measure defined by a density)

Let f : ! + a measurable mapping and µ a measure on ( , B(R )). R
E. V.2.1 Assume that for any measurable mapping j : ! + , j( x ) µ(dx ) = j( x ) f ( x ) l(dx ).
Compute µ( A), for any A 2 B( ).
R
RE. V.2.2 Assume now that for any bounded continuous function j : ! + , j( x ) µ(dx ) =
j( x ) f ( x ) l(dx ). Compute µ(( a, b)), for any a < b 2 .

Exercise V.3 (Fourier series)

1
We consider here the Hilbert space H = L2 [0, 2p ], B([0, 2p ]), 2p l . Let f 2 H. Recall that the
Fourier coefficients are denoted by
Z 2p
1 inx
cn ( f ) = f ( x )e l(dx ).
2p 0
E. V.3.1 Recall Parseval’s identity. Prove that cn ( f ) converges to 0 as |n| ! •.
E. V.3.2 Compute the Fourier series of x 7! cos(5x ). What can you say about the convergence of
this series?
E. V.3.3 Compute the Fourier series of the function defined on [0, 2p ] by f ( x ) = 1[0,p ) ( x )
1[p,2p ] ( x ). What can you say about the convergence of this series? Is there normal convergence of
the Fourier series towards f on [0, 2p ]?

CS, 1st year 30 2023-2024

CIP 2023-2024

Exercise V.4 (Interchanging limit and integral)

The questions below are independent but are all concerned with interchanging a limit and an integral.
E. V.4.1 In this question, we will prove the following assertion:

For a sequenceRof non-negative functions ( f n )n2 which converges l-a.e. Rto a function f ,
assuming that f n dl converges to a constant c 2 + does not imply that f dl = c.

Consider f n = 1[n,n+1] for all n 2 .

(a) Study the convergence of the sequence ( f n )n2 (pointwise convergence, l-a.e. convergence.).
Z
(b) Determine f n dl for all n 2 .

(c) Conclude.

E. V.4.2 Study the convergence of the sequence (un )n2 ⇤ defined by

Z
⇤ sin(tn )
8n 2 , un = l(dt).
]0,+•[ t n (1 + t )

E. V.4.3 Let ( E, F , µ) be a measure space and ( f n )n2 be a sequence of real-valued bounded mea-
surable functions defined on E. Assume that this sequence converges uniformly to f .

(a) Assume further that µ( E) < •. Prove that

Z Z
lim f n dµ = f dµ .
n!+• E E

⇤(b) What happens if we no longer assume that µ( E) < •?

In the following exercise, we prove that for any continuous function defined on a closed interval,
the Riemann and Lebesgue integrals coincide.

Exercise V.5 (Integral of a continuous function on a closed interval)

Let f be a continuous function on the interval [ a, b], a < b 2 . Define ak,n = a + k b n a and consider:
n
f n (x) = Â f ( ak,n )1(ak 1,n ,ak,n ]
(x)
k =1
R
E. V.5.1 (a) Compute [ a,b] f n dl.
R Rb
(b) Prove that [ a,b] f n dl converges to f ( x ) dx. a
R R
E. V.5.2 Prove that [ a,b] f n dl converges to [a,b] f dl.
E. V.5.3 Deduce that: Z b Z
f ( x ) dx = f dl.
a [ a,b]

CS, 1st year 31 2023-2024

CIP 2023-2024

In the following exercise, we compute the Gauss integral:

Z p
u2 /2
e l(du) = 2p

using the basic properties of the Lebesgue integral.

Exercise V.6 (Computing the Gauss integral)

Set ✓ ◆
Z 2 Z x 2 (1+ t2 )
t2 e
F(x) = e l(dt) and G(x) = l(dt).
[0,x ] [0,1] 1 + t2
E. V.6.1 Prove that F is differentiable on and compute F 0 (it can remain in integral form).
E. V.6.2 Prove that G is differentiable on and compute G 0 (it can remain in integral form).
E. V.6.3 Deduce F + G.
E. V.6.4 Compute lim G ( x ) = 0.
x !+•
Z +•
t2
E. V.6.5 Deduce the value of e dt.
0

There are many functions which are not Riemann-integrable, while they are Lebesgue-integrable.
The following exercise proposes to study the example of the indicator function of .

Exercise V.7 (The indicator function of )

E. V.7.1 (a) Prove that 1 is not piecewise continuous.

⇤(b) Prove that 1 is not a regulated function (i.e. the uniform limit of step functions).
⇤(c) Prove that 1 is not Riemann-integrable.
E. V.7.2 (a) Prove that is negligible set (i.e. l( ) = 0).
Z Z
(b) Prove that 1 dl = 1 ( x ) l(dx ) exists and give its value.

Exercise V.8 (Comparison between the Riemann and Lebesgue integrals)

Z
t
E. V.8.1 (a) Compute e l(dt).
+
Z p
(b) Compute limn!+• sinn (t) dt.
0
E. V.8.2 Let f be a continuous function on .
Z x
(a) Is the function x 7! f (t) dt differentiable on ?
0
Z
(b) Is the function x 7! f (t) l(dt) differentiable on ?
[0,x ]
Z
(c) Is the function x 7! f ( t )1 \ (t) l(dt) differentiable on ?
[0,x ]

CS, 1st year 32 2023-2024

CIP 2023-2024
sin(t)
E. V.8.3 (a) Is the function t 7! t Riemann-integrable on +? Lebesgue-integrable?
sin(t)
(b) Same questions with t 7! t 1 \ ( t ).

D) Going further

Exercise V.9 (Féjer’s theorem in L p )

This exercise follows Exercise II.5 and the notations are taken from there.
E. V.9.1 Let p 2 [1, •[. In this question, we extend Féjer’s theorem to
p
L ([ p, p ], B([ p, p ]), l), hence f is not assumed to be continuous anymore.

(a) Let r 2 [1, •]. Prove that for any g 2 Lr ( ) and any h 2 L1 ( ), k g ⇤ hkr  k gkr k hk1 . [Hint: first
apply Hölder’s inequality to | g| ⇤ | h|( x ).]

⇤⇤(b) Use Féjer’s theorem as stated in E.II.5.1 to prove that if fe is continuous on [ p, p ], then for any
e > 0, there exists N 2 such that for all n N, k fe Kn ⇤ fek p  e.

⇤(c) Let f 2 L p ([ p, p ], B([ p, p ]), l). Using the previous questions, prove that k f KN ⇤ f k p ! 0
as N ! •.

E. V.9.2 Application of the previous theorem: Let f 2 L1 ([ p, p ], B([ p, p ]), l). Prove the injec-
tivity of the Fourier coefficients (i.e. if 8n 2 , cn ( f ) = 0, then f = 0 a.e.).

The following exercise provides some counter-intuitive properties of the Lebesgue measure.

Exercise V.10 (Topological properties of the Lebesgue measure)

*E. V.10.1 Prove that for any e > 0, there exists a dense open subset Oe ⇢ such that l(Oe )  e.
E. V.10.2 Let A be an open subset of . Is there an equivalence between the assertions “A is
bounded” and “A has finite Lebesgue measure”?
E. V.10.3 Let A 2 B( ). Is there an equivalence between the assertions “l( A) > 0” and “A
contains a non-empty open set” ?

Exercise V.11 (Hardy’s inequality)

Let f 2 L2 ( + ; ). For any x > 0, set:
Z
1
F(x) = f (t) l(dt).
x [0,x ]

E. V.11.1 Prove that F is well-defined on ⇤.

+
E. V.11.2 Assume that f is continuous with compact support in ⇤.
+

(a) Prove that F is the solution to a simple linear differential equation.

CS, 1st year 33 2023-2024

CIP 2023-2024

(b) Deduce that: Z Z

F2 ( x ) l(dx ) = 2 F ( x ) f ( x ) l(dx ).
+ +

(c) Prove that:

k F k2  2k f k2 .

E. V.11.3 f is no longer assumed to be continuous with compact support. Let ( f n )n2 be a sequence
of continuous functions with compact support that converges to f in L2 .

(a) Justify that such a sequence ( f n ) exists.

Z
1
(b) Let Fn ( x ) = f n (t) l(dt). Prove that ( Fn )n2 converges in L2 .
x [0,x ]

(c) Let F̂ be the limit of the sequence ( Fn )n2 . Prove that:

|| F̂ ||2  2|| f ||2 .

(d) Prove that F̂ = F a.e.

(e) Deduce that || F ||2  2|| f ||2 and thus that the mapping f 7! F is continuous.

Remark: more generally, one can prove that if f 2 L p , then so is F and

p
|| F || p  || f || p .
p 1

CS, 1st year 34 2023-2024

 Chapter VI : Probability and measure

A) Aims of this class

After this class,

• I know and understand the concepts of probability space, random variable, distribution/law of
a random variable;

• I can compute the probability of an event when the probability measure is given;

• I know and understand the concepts of cumulative distribution function and of probability den-
sity function;

• I can determine the distribution of a random variable;

• I am able to check that a given random variable is measurable with respect to a sub-s-algebra;

• I can compute the expectation and variance of a random variable when they exist;

• I am able to apply the transfer theorem to compute laws of random variables.

35
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions VI.1 and VI.2 must be done before the examples class. The solutions for these exercises
are available online.

Question VI.1
Let (W, F , ) be a probability space and N be a Poisson random variable P (l) with parameter l > 0:

lk l
( N = k) = e .
k!
Q. VI.1.1 (a) Recall the definition of [ N ] (i.e. write it as an integral over W).
(b) Using the transfer theorem, write [ N ] as a sum.
(c) Compute the expectation of N.

Question VI.2
Let (W, F , ) be a probability space and X be a Gaussian random variable N (m, s2 ):
Z x ✓ ◆
1 1 ( t m )2
(X  x) = p exp dt.
2ps • 2 s2

Q. VI.2.1 (a) Recall the definition of [ X ] (i.e. write it as an integral over W).
(b) Using the transfer theorem, write [ X ] as an integral over .

(c) Compute the expectation of X.

CS, 1st year 36 2023-2024

CIP 2023-2024

C) Exercises

Here are a couple of exercices to practice with the definition of random variables, the transfer
theorem and change of variables.

Exercise VI.1 (Computing some expectations)

The two questions are independent.
E. VI.1.1 For any p 2 (0, 1), consider a probability measure µ on ( , B( )) which is proportional
• n
to Ân=0 p dn , where dn is the Dirac measure at n.

(a) Justify the existence of µ.

(b) Determine the expectation of a random variable with law µ.

E. VI.1.2 Set U = X s m , where X is a Gaussian random variable N (m, s). What is the law of U?
Determine the law of the random variable Y = U 2 . [Hint: you may consider [ h(Y )] for any bounded
measurable function h : ! .]

Exercise VI.2 (Probability density, log-normal law)

2
Let X be a real-valued random variable with a law proportional to e x /2 l(dx ), where l is the
Lebesgue measure on .
E. VI.2.1 Prove that Y = e X is a random variable which has a probability density.

The three following exercises focus on random variables. They allow you to grasp how important
measurability is and also make a link with almost sure equalities. The three results that are proven
below could almost be part of the lectures.

Exercise VI.3

E. VI.3.1 Prove that if X and Y are two random variables which are equal a.s., then they have the
same law. Prove that the reverse statement is wrong.

Exercise VI.4

E. VI.4.1 Let (W, F , ) be a probability space, and let G and H be two independent sub-s-algebras
of F . Two sub-s-algebras G and H are independent if for any A 2 G and any B 2 H,

( A \ B) = ( A) ( B ).

Prove that if X is a real-valued random variable which is both G -measurable and H-measurable, then
X is constant a.s., i.e. there exists c 2 such that ( X = c) = 1.

Exercise VI.5
Let (W, F , ) be a probability space, let X be an p -valued random variable and let Y be an q -valued
random variable.
The goal of this exercise is to prove that Y is X-measurable (i.e. Y is s( X )-measurable) if and only
if there exists a Borel mapping Y : p ! q such that Y = Y( X ).

CS, 1st year 37 2023-2024

CIP 2023-2024

E. VI.5.1 Assume first that Y is a simple function, i.e. Y = Âik=1 ai Ai , and is s( X )-measurable.
Without loss of generality, we assume that the Ai are pairwise disjoint.

(a) What consequence on Ai can you draw for the s( X )-measurability of Y?

(b) Find a simple function Y : p ! q such that Y = Y( X ).

E. VI.5.2 In the general case,

(a) justify that there exists a sequence of simple s( X )-measurable random variables (Yn )n2 which
converges to Y.

(b) Write Yn = Yn ( X ) where Yn is a Borel mapping, and consider the set C = { x 2 p : limn!• Yn ( x ) exists}.
Prove that C 2 B( p ).

(c) Observe that X (W) ⇢ C and find a Borel mapping Y : p ! q such that Y = Y( X ).

E. VI.5.3 Conversely, prove that if Y = Y( X ) with Y : p ! q a Borel mapping, then Y is

s( X )-measurable.

The exponential law is frequently used in the modelling of systems failures. The following exercise
shows its characteristics with respect to the memory phenomenon.

Exercise VI.6 (Exponential law)

A laptop manufacturer wants to check the warranty period associated to its hard drives to ensure it is
not too long with respect to the number of returns. Tests have shown that the lifespan X (in years) of
a hard drive is distributed according to an exponential law with mean 4.
E. VI.6.1 What is the probability that a hard drive works fine for at least 4 years?
E. VI.6.2 What is the probability that a hard drive works for six years, knowing that it has already
been working fine for five years?
E. VI.6.3 What is the probability that the lifespan is between [X] s( X ) and [ X ] + s ( X )?
E. VI.6.4 For how long can at least 50% of the drives work without failure?
E. VI.6.5 Provided the law of the country does not mandate a minimum warranty period, what is
the optimal warranty period to ensure that less than 15% of the hard drives have to be replaced during
their warranty period.

D) Going further

Exercise VI.7 (Exponential law)

Let T be a real-valued random variable such that

8s, t 0; ( T > t + s) = ( T > t) ( T > s ). (VI.1)

The goal of this exercise is to prove that either ( T > 0) = 0 or T follows an exponential law.

CS, 1st year 38 2023-2024

CIP 2023-2024

E. VI.7.1 Verify that if ( T > 0) = 0, then (VI.1) is satisfied.

E. VI.7.2 Hence assume that ( T > 0) > 0.
(a) Determine the limit of the sequence ( T > n1 ) n2
.

(b) Deduce that there exists e > 0 such that { T > e} > 0.
(c) Prove that for any t > 0, ( T > t) > 0. For any t > 0, you may consider n 2 ⇤ such that
t < ne.

E. VI.7.3 Define the function f : t 7! log ( T > t) on ⇤.

+

(a) Based on (VI.1), write the relation satisfied by f .

(b) Prove that for any n 2 , f ( n ) = n f (1).

(c) Deduce that for any x 2 +, f ( x ) = x f (1) and that this relation holds for any x 2 +.

E. VI.7.4 Determine limn!• f (n). Deduce that f (1) < 0 and conclude.

Exercise VI.8 (Gamma, Beta, and c2 laws)

For a > 0, recall that Z +•
x a 1
G( a) = e x dx.
0

The Gamma law of parameters a > 0 and l > 0, denoted by G ( a, l), is the probability measure on
whose density ga,l is given by

la lx a 1
x 7! ga,l ( x ) = e x ( x ).
G( a) +

E. VI.8.1 Let X be a r.v. with law G ( a, l). Compute [ X ] and Var( X ).

E. VI.8.2 Let X and Y be two independent r.v. with law G ( a, l) and G (b, l).
X
(a) Prove that X + Y and X +Y are independent and determine their law. Deduce that
Z 1
1 1 G( a)G(b)
B( a, b) = xa (1 x )b dx = .
0 G( a + b)

X
(b) Determine the law of Y.

(c) If X1 , . . . , Xn are i.i.d. random variables of exponential law E (l), find the law of Sn = X1 + · · · +
Xn .

E. VI.8.3 Let Y be a standard Gaussian r.v. Prove that Y 2 follows the law G (1/2, 1/2). Deduce the
value of G(1/2).
E. VI.8.4 If Y1 , . . . , Yn are i.i.d. random variables with law N (0, 1), find the law of Z = Y12 + · · · + Yn2
and compute [ Z ] and Var( Z ).

CS, 1st year 39 2023-2024

 Chapter VII : Product measures

A) Aims of this class

After this class,

• I know what a product s-algebra is;

• I know that a product measure is characterised by its values on rectangles (i.e. on the Cartesian
products of sets);

• I can verify that a function of several variables is measurable and integrable;

• I am able to apply the Fubini-Tonelli and Fubini-Lebesgue theorems correctly, in order to com-
pute integrals of functions of several variables;

• I am able to apply the integration with respect to a product measure to the case of random
variables;

• I am able to perform a change of variables for a multiple integral.

40
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions VII.1 and VII.2 must be done before the examples class. The solutions for these exercises
are available online.

Question VII.1

Q. VII.1.1 Find a function f 2 L1 ( , B( ), l) which is not in L p ( , B( ), l), 8 p > 1. Find a

function f 2 L p ( , B( ), l) for some p > 1 which is not in L1 ( , B( ), l).

Question VII.2
Let l(2) denote the Lebesgue measure in 2 and consider the domain D = [0, 1] ⇥ +.
Let f be the function defined by
y
f ( x, y) = sin(2px )e D ( x, y ).

Q. VII.2.1 (a) Explain carefully why f 2 L1 ( 2 , l (2) ) .

R
(b) Compute 2 f dl(2) .

CS, 1st year 41 2023-2024

CIP 2023-2024

C) Exercises

Measure theory provides a more natural and consistent framework than the Riemann integral for
the study of multiple integrals. We observe below, on some simple examples, that Tonelli and Fubini’s
theorems are efficient tools to study the integrability of multivariable functions.

Exercise VII.1 (Integrability)

1
E. VII.1.1 Is the function g defined by g( x, y, z) = 1 xyz integrable on (0, 1)3 ? [i.e. is there g (0,1)3 2
L1 ( 3 )?]
1
E. VII.1.2 Find the values of a such that the function h defined by h( x, y) = ( x 2 + y2 ) a
is integrable
on D = {( x, y) 2 (0, +•)2 : x2 + y2 < 1}.

Exercise VII.2 (s-algebra and measure of 3 )

Let f : 2 ! be a Borel function. Consider G = {( x, f ( x )); x 2 2} the graph of f .
E. VII.2.1 Prove that G is a Borel set of 3 .
E. VII.2.2 Prove that G is negligible in 3 for the Lebesgue measure (i.e. l(3) (G) = 0).

The goal of the following exercise is to extend the Riemann-Lebesgue lemma from Riemann-
integrable functions on a closed interval to Lebesgue-integrable functions on any interval. Besides
the result, you may remember the density-based method.

Exercise VII.3 (The Riemann-Lebesgue lemma)

E. VII.3.1 (a) Let f be a step function defined on [ a, b]. Prove that:

Z b Z b
lim f ( x ) cos(nx ) dx = 0 = lim f ( x ) sin(nx ) dx.
n!• a n!• a

(b) Extend the result to any regulated function (recall that regulated functions are the functions
which are uniform limit of step functions). [Regulated functions form a large class of Riemann-
integrable functions. You may consider proving the RL lemma for any Riemann-integable function.]
E. VII.3.2 Let I be an interval of .
(a) Let f be an infinitely differentiable function with compact support in I. Prove that:
Z Z
lim f ( x ) cos(nx ) l(dx ) = 0 = lim f ( x ) sin(nx ) l(dx ).
n!• I n!• I

(b) Extend the result to any Lebesgue-integrable function.

Exercise VII.4
x 2 y2
E. VII.4.1 Is the function f defined by f (0, 0) = 0 and f ( x, y) = ( x 2 + y2 )2
integrable on [0, 1]2 ?

Exercise VII.5
Let X be a non-negative random variable with probability density f .

CS, 1st year 42 2023-2024

CIP 2023-2024

E. VII.5.1 Prove that for any r > 0,

⇥ ⇤ Z 1
Xr = r xr ( X > x ) l(dx ).
+

The two following exercises show that the usual theorems for multiple integrals allow to obtain
easily some classical results.

Exercise VII.6 (Computing simple integrals)

Let 0 < a < b. Z
E. VII.6.1 Compute e xy l(dy). Deduce the value of
[ a,b]
Z ax bx
e e
I= l(dx ).
+ x
E. VII.6.2 Adapt the previous method to compute
Z +•
cos( ax ) cos(bx )
J= dx.
0 x

Exercise VII.7 (A classical integral)

Let a > 0 and Da = (0, a) ⇥ (0, +•).
sin(u)
E. VII.7.1 Let f be the function defined on 2 by f (u, v) = u e
v for any (u, v) 2 ⇤ ⇥ and
v
f (0, v) = e for v 2 . Prove that f is integrable on Da .
E. VII.7.2 Prove that the mapping T defined by T ( x, y) = ( x, xy) is a diffeomorphism of Da .
E. VII.7.3 Let g be the function defined on 2 by g( x, y) = e xy sin( x ). Prove that g is integrable
on Da . Compare its integral to the integral of f .
Z a Z +• Z +•
sin(u) e ay y e ay
E. VII.7.4 Express du in terms of dy and dy.
0 u 0 1 + y2 0 1 + y2
Z +•
sin(u)
E. VII.7.5 Deduce that du exists and find its value.
0 u

Exercise VII.8

E. VII.8.1 Let X and Y be two real-valued r.v. whose joint law has a density f : ⇥ ! +.
Determine the law of Y/X.

D) Going further

The integration by parts formula holds classically for C1 functions. We will extend this formula to
a larger class of functions.

Exercise VII.9 ( Generalisation of the integration by parts formula)

In this exercise, denote L1 = L1 ([0, 1]).

CS, 1st year 43 2023-2024

CIP 2023-2024

Let f 2 L1 . For any x 2 [0, 1], set:

Z
F(x) = f (t) l(dt).
[0,x ]

E. VII.9.1 Does the integration by parts formula hold for a.e.-differentiable (or even everywhere
differentiable) functions?
E. VII.9.2 (a) Is F a.e.-differentiable on (0, 1) ?

(b) Is F necessarily differentiable everywhere on (0, 1)?

E. VII.9.3 Let g 2 L1 . For any x 2 [0, 1], set
Z
G(x) = g(t) l(dt).
[0,x ]

(a) Prove that for any (u, t) 2 [0, 1]2 ,

⇣ ⌘
1[0,t] (u) 1[0,x] (t) = 1[0,x] (u) 1[0,x] (t) 1[0,u[ (t) .

(b) Prove that (t, u) 7! | f (t) g(u)| is integrable on [0, 1]2 .

(c) Deduce that for any x 2 [0, 1],

Z Z
G (t) f (t) l(dt) = G ( x ) F ( x ) g(t) F (t) l(dt).
[0,x ] [0,x ]

(d) Then prove that for any ( a, b) 2 (0, 1)2 ,

Z Z
G (t) f (t) l(dt) = [ GF ]ba g(t) F (t) l(dt).
[ a,b] [ a,b]

CS, 1st year 44 2023-2024

 Chapter VIII : Fourier transform

– Exercises to be studied in autonomy after Lecture VIII (24 October 2023) –

A) Aims of this class

After this class,

• I know the definition and the properties of the Fourier transform of an integrable function;

• I know the Fourier transform of a Gaussian function;

• I understand the construction of the Fourier transform in L2 and I know the inverse formula;

• I can state the isometry property (Parseval) of the Fourier transform in L2 ;

• I know the link between Fourier transform and differentiation, and between Fourier transform
and convolution;

• I know how to compute the characteristic function of a real-valued random variable.

45
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions VIII.1 and VIII.2 must be done before the examples class. The solutions for these exer-
cises are available online.

In class, we have seen that the Fourier transform is a linear mapping from L1 to C0 or from L2 to
L2 . We will check that the codomain does depend on the chosen domain.

Question VIII.1 (Fourier transform in L1 and L2 )

Q. VIII.1.1 (a) Compute F 1[ 1,1] .

(b) Deduce that F is not a mapping from L1 to L1 .

sin(t) sin(t)
Q. VIII.1.2 (a) Prove that F t and F̄ t exist.
sin(t) sin(t)
(b) Use 1)(a) to compute F̄ t , and deduce F t .

(c) Deduce that F is not a mapping from L2 to C0 .

Question VIII.2
Let X be a random variable following an exponential law of parameter l > 0 and let Y be a Poisson
r.v. with parameter µ > 0.
Q. VIII.2.1 Give the characteristic function of X.
Q. VIII.2.2 Give the characteristic function of Y.

CS, 1st year 46 2023-2024

CIP 2023-2024

C) Exercises

We now study the classical example of the Fourier transform of a Gaussian function.

Exercise VIII.1 (Fourier transform of a Gaussian function)

E. VIII.1.1 Let f ( x ) = exp( cx2 ), with c > 0. Prove that the Fourier transform of f exists.
E. VIII.1.2 (a) Find a simple ODE satisfied by f .

(b) Deduce a simple equation satisfied by F f .

(c) Determine F f .
E. VIII.1.3 Find an invariant for the Fourier transform F .

The Fourier transform of a square integrable function exists but the usual formula
Z
1 ixy
F f (y) = p f ( x )e l(dx )
2p
is no longer available. We shall establish appropriate formulas in that case.

Exercise VIII.2 (The Fourier-Plancherel transform)

Let f 2 L2 ( ). Denote f n = f 1[ n,n] .
E. VIII.2.1 Prove that the following limit holds in L2 :
⇢ Z
1
y 7! p f ( x )e ixy l(dx ) ! F f,
2p [ n,n] n!+•

or equivalently: lim kF f F f n k L2 ( ) = 0.
n!+•

E. VIII.2.2 Prove that for any g 2 L2 ( ),

Z Z
lim g F f n dl = g F f dl.
n!+•

E. VIII.2.3 Choosing g = 1[0,x] , prove that:

Z ixy
1 d 1 e
F f (y) = p f ( x ) l(dx ) a.e.
2p dy ix

Exercise VIII.3 (Non-surjectivity of the Fourier transform)

Consider the function q : ! defined by
Z +•
sin u
8x 2 , q (x) = du,
x u
where the integral is in the sense of Riemann.
E. VIII.3.1 Prove that q is well-defined, and that it is continuous and bounded on [0, +•).

CS, 1st year 47 2023-2024

CIP 2023-2024

E. VIII.3.2 Let f be an odd and integrable function on . Define, for all z 1,

Z z
F f (y)
f(z) = dy.
1 y

Prove that the limit of f(z) exists and is finite, as z ! +•.

E. VIII.3.3 Let g be the (odd) function defined by
arctan x
8x 2 , g( x ) = .
log(2 + x2 )

(a) Prove that g is a continuous function that converges to 0 in • and +•.

(b) Prove that if g is the Fourier transform of a function f , then f is necessarily odd. [Hint: you may
use the injectivity of the Fourier transform, proving it first.]

(c) Prove that g is not the Fourier transform of an integrable function.

Exercise VIII.4 (The Fourier transform in L1 )

Let L1 stand for L1 ( ), F denote the Fourier transform and ⇤ denote the convolution product.
E. VIII.4.1 Prove that there is no function f in L1 such that F f = 1.
E. VIII.4.2 Find the functions f 2 L1 such that f ⇤ f = f . [Hint: you may use the Fourier transform.]
E. VIII.4.3 Prove that there does not exist g 2 L1 such that, for any f 2 L1 , one has g ⇤ f = f . [Hint:
you may use the Fourier transform.]
E. VIII.4.4 Let f be a function of L1 such that the function y 7! y F f (y) is in L1 too.

(a) Prove that F f 2 L1 .

(b) Deduce that f has a continuously differentiable representative.

D) Going further

Exercise VIII.5 (A basis of eigenvectors of F )

Denote by F the Fourier transform in L2 ( ) and yn the nth Hermite function. Recall that for x 2 ,

( 1) n x2 d
n
x2
yn ( x ) = p p e2 e .
2n n! p dx n

E. VIII.5.1 (a) Compute F 4 .

(b) Deduce the possible eigenvalues of F .

Z Z
ixy y2 dn y2 /2+ixy
*E. VIII.5.2 (a) Compute e yn (y) l(dy) in terms of e e l(dy).
dyn

CS, 1st year 48 2023-2024

CIP 2023-2024
Z
y2 dn (ix+y)2 /2
(b) Deduce F̄ yn in terms of e e l(dy).
dx n
(c) Prove that F̄ yn = in yn .
E. VIII.5.3 (a) Deduce a Hilbert basis of L2 made of eigenvectors of F .

(b) Is this also an algebraic basis?

E. VIII.5.4 Let g 2 L2 ( ). Compute g and F g in terms of the yn .

The characteristic function is a very common tool in probability. Note the obvious link with the
Fourier transform.

Exercise VIII.6 (The Cauchy law)

A r.v. X follows a Cauchy law of parameter c > 0 if it has the following density

1 c
f : x 7! .
p x 2 + c2
E. VIII.6.1 Determine the characteristic function of the symmetric exponential law of parameter
l > 0, as defined by its density
l
g : x 7 ! e l| x | .
2
E. VIII.6.2 Deduce the characteristic function of X.
E. VIII.6.3 Prove that if X and Y are independent r.v. following Cauchy laws of parameters c and
c , then X + Y follows a Cauchy law of parameter c + c0 .
0

E. VIII.6.4 Prove that if X follows a Cauchy law of parameter c > 0 and if a > 0, then aX follows
a Cauchy law of parameter ac. In particular, prove that 2X has the same law as the sum of two
independent Cauchy r.v. with the same parameter.
X +Y
E. VIII.6.5 Prove that if X and Y are i.i.d. r.v. following a Cauchy distribution, then 2 has the
same law as X.
E. VIII.6.6 Prove that if X and Y are two i.i.d. r.v. which are not constant a.s., whose law is
symmetric and such that
8a, a0 > 0; aX + a0 Y ⇠ (a + a0 ) X,
then they follow a Cauchy law.

CS, 1st year 49 2023-2024

CIP 2023-2024

E) Three problems with links to other fields

Exercise VIII.7 (The Heisenberg inequality)

Let f be a differentiable mapping defined on such that f , f 0 , t 7! t f (t) and t 7! t f 0 (t) are square
integrable on .
*E. VIII.7.1 Let D( ) denote the space of infinitely differentiable functions on with compact
support. This space and the following result will be presented in greater details in the PDE lectures.
Prove the injectivity of the linear mapping
R T· : L1loc ( ) ! D 0 ( ) which maps f 2 L1loc ( ) to the
regular distribution T f : f 2 D( ) 7! f f. [Indication: you may use Question E.IX.6.2].
E. VIII.7.2 ⇤⇤(a) Compute F ( f 0 ) in terms of F ( f ).
(b) Prove that:
Z 2
t f (t) f 0 (t) l(dt)  ht f (t), t f (t)i L2 hyF f (y), yF f (y)i L2 ,

where in the previous inner product, “t f (t)00 denotes abusively the function t 7! t f (t).
E. VIII.7.3 (a) Define g(t) = t f (t) f¯(t). Z
Prove that the limits of g in ±• exist and that g0 (t) l(dt) = 0.

(b) Using g0 , prove that

Z Z Z
f (t) f (t) l(dt) = t f (t) f 0 (t) l(dt) t f 0 (t) f (t) l(dt).

E. VIII.7.4 (a) Deduce that for E = k f k2L2 = kF f k2L2 , then

E2
kt f (t)k2L2 kyF f (y)k2L2 .
4

⇤(b) By considering the function t 7! g(t) = exp( t2 /2), prove that 1/4 is optimal.
E. VIII.7.5 (a) Define g(t) = f (t + t0 ) exp( ity0 ) for some fixed real numbers t0 and y0 . Com-
pute F g.
(b) Deduce that
Z
! Z
!
| f (t)| 2 |F f (y)| 2 1
(t t0 )2 l(dt) (y y0 )2 l(dy) .
E E 4

(c) Find the values of t0 and y0 that minimise the integrals in the left hand side of the previous
inequality.
Interpret these values, say tm and ym , as mean values.
Deduce that the integrals represent variances of t and y.
We can deduce the Heisenberg inequality : there exists a constant C such that
Variance(time).Variance(energy) C
or in other words, there exists a constant K such that
Variance(time).Variance(frequency) K.

CS, 1st year 50 2023-2024

CIP 2023-2024

Interpretation : R R
The functions a and b (see the correction) can be interpreted as densities ( 0, L1 , a dl = b dl = 1).
Hence tm is the mean value of t and ym is the mean value of y and the inequality 5b) displays some
variances : Var(t) and Var(y).
Since y can be interpreted as a pulsation w, then by multiplying by h̄2 and noticing that E = h̄w is
an energy, the inequality 5b) reads:

h̄2
Var(time) Var(energy) C= .
4
Similarly, observing that the frequence is given by f = w/2p, one has:

1
Var(time) Var(frequency) K= .
16p 2
In signal processing, this can interpreted as follows:
A signal which is localised in time cannot be localised in frequency (or in energy) and con-
versely.

Now we shall establish a fundamental result in signal processing:

Any signal which is bounded in frequency can be reconstructed from a discrete sample
with adapted bandwidth.

Exercise VIII.8 (Shannon sampling)

Let f 2 L2 ( ) and assume that the Fourier transform of f is compactly supported. Let a > 0 be such
that [ a, a] contains the support of F f .
E. VIII.8.1 (a) Prove that f has a continuous representative.

⇤(b) Prove that f has an infinitely differentiable representative.

E. VIII.8.2 Let b > a and let g be the 2b-periodic function which equals F f on [ a, a] and 0 on
[ b, a) [ ( a, b].
(a) Compute the Fourier series of g.

(b) Does the Fourier series of g converge to g in L2 ([ b, b]) ?

E. VIII.8.3 Deduce that: Z +•

p np
| f (t)|2 l(dt) = Â | f ( )|2 .
• b b
E. VIII.8.4 Denote sinc (u) = sin(u)/u, and prove that the series
np
 f(
b
) sinc (tb np )
n2

converges to f in L2 ( ).
**E. VIII.8.5 Prove that the convergence is uniform.
E. VIII.8.6 State precisely the Shannon theorem that we have just proven.

CS, 1st year 51 2023-2024

CIP 2023-2024

Exercise VIII.9 (Poisson summation formula)

Let f be an integrable function on . We shall use temporarily the following alternative form of the
Fourier transform: Z
fˆ( x ) = e 2ipxt f (t) l(dt).

Assume moreover that:

• f is continuous in .
M
• 9 M > 0, 9a > 1 : 8 x 2 , | f ( x )|  (1+| x |)a
.

• Â+• ˆ
• | f ( n )| < •.

Define F ( x ) = Â+•• f ( x + n ), for all x 2 .

E. VIII.9.1 ⇤(a) Prove that F is continuous.
(b) Prove that F has period 1.
E. VIII.9.2 Compute the Fourier series of F.
*E. VIII.9.2 Prove that the Fourier series of F converges pointwise to F. [Hint: you may use Féjer’s
theorem.]
E. VIII.9.4 Deduce the Poisson summation formula:
+• +•
 f (n) =  fˆ(n).
• •

E. VIII.9.5 Deduce the formula for the usual definition of the Fourier transform:
p +• +•
2p  f (2pn) =  F f ( n ).
• •

E. VIII.9.6 Let a > 0. Compute the Fourier transform of exp( 2pa|t|). Deduce the value of
Â+• 1
• n2 + a2 .

CS, 1st year 52 2023-2024

 Chapter IX : Probability in d and independence

A) Aims of this class

After this class,

• I can characterise the independence of random variables in terms of product measure;

• I can verify that two random variables are independent;

• I am able to compute the law of a random variable which is defined as a function of two inde-
pendent random variables;

• I know the difference between independence and uncorrelatedness;

• I am able to study real-valued random vectors whose distribution is given (law of each compo-
nent, independence);

• I am able to study the convolution of two functions (integrability, convergence).

53
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions IX.1 and IX.2 must be done before the examples class. The solutions for these exercises
are available online.

Question IX.1
Let m1 , m2 2 and s1 , s2 > 0. Let X and Y be two independent random variables such that X ⇠
N (m1 , s12 ) and Y ⇠ N (m2 , s22 ), i.e. X (resp. Y) is distributed as a Gaussian with mean m1 (resp. m2 )
and standard deviation s1 (resp. s2 ).
Q. IX.1.1 Give the law of X + Y.

The following exercise shows that two uncorrelated random variables need not be independent.

Question IX.2
Let X and Y be two independent r.v. following a Bernoulli law of parameter 12 .
Q. IX.2.1 Give the distribution of X + Y, | X Y | and ( X + Y )| X Y |.
Q. IX.2.2 Prove that X + Y and | X Y | are uncorrelated.
Q. IX.2.3 Prove that X + Y and | X Y | are not independent. [Hint: you may compute (X + Y =
0, | X Y | = 0)].

CS, 1st year 54 2023-2024

CIP 2023-2024

C) Exercises

Exercise IX.1
Let X and Y be two independent r.v. following the same exponential distribution of parameter 1.
X
E. IX.1.1 Determine the joint law of U = X + Y and V = .
X+Y
E. IX.1.2 Are the random variables U and V independent?
E. IX.1.3 Deduce that V follows the uniform law on [0, 1].

Exercise IX.2
Let (W, F , ) be a probability space. Let N be an ⇤ -valued r.v. and ( Xn )n2 ⇤ be a sequence of r.v.
Define X N by
8w 2 W, X N (w ) = X N (w ) (w ).
E. IX.2.1 Prove that X N is a random variable.

Exercise IX.3
Assume that the intensities of earthquakes are given by independent random variables X1 , X2 , . . .
which follow an exponential law of parameter a and that the number of earthquakes per year is given
by a r.v. N which follows a Poisson law of parameter l that is independent of the intensities of the
earthquakes.
E. IX.3.1 Consider the quantity Y which represents the maximal yearly intensity of the earthquakes.
What is the law of Y?

Now we study the links between independence, marginal distributions and conditional distribu-
tions.

Exercise IX.4
Let (W, F , ) be a probability space and consider a r.v. ( X, Y ) whose law is absolutely continuous (wrt
the Lebesgue measure), and defined on the domain
2
D = ( x, y) 2 ; 0 < y  1 and 0  x  y ,

by the probability density function f : 2 ! +,

⇢
1/y if ( x, y) 2 D
f ( x, y) =
0 otherwise.

E. IX.4.1 Prove that f does define a probability density function.

E. IX.4.2 Compute the cumulative distribution function of ( X, Y ).
E. IX.4.3 Compute the marginal density functions of ( X, Y ).
E. IX.4.4 Compute the probability density functions of the r.v. Y | X = x and X | Y = y.
E. IX.4.5 Are the variables X and Y independent?

CS, 1st year 55 2023-2024

CIP 2023-2024

D) Going further

Additional exercises on the convolution product.

Exercise IX.5 (Convolution)

Let f 2 L1 ( ) and g, h 2 L2 ( ). Define fˇ by fˇ( x ) = f ( x ).
E. IX.5.1 Prove that | f | ⇤ | g| 2 L2 ( ).
E. IX.5.2 Prove that the function ( x, y) 7! f ( x y) g(y)h( x ) is integrable on 2.

E. IX.5.3 Deduce that: Z Z

( f ⇤ g)h dl = g( fˇ ⇤ h) dl.

*E. IX.5.4 Assume in addition that f and g are compactly supported with supports denoted by A
and B respectively. Then prove that f ⇤ g is compactly supported with support included in A + B.

Exercise IX.6 (Mollifiers (ctd) - Urysohn’s lemma and consequences)

*E. IX.6.1 We shall prove Urysohn’s lemma: Let W be an open set of d and K ⇢ W a compact set,
then there exists y 2 Cc• (W) such that y( x ) = 1 for any x 2 K.

(a) Let d > 0 be such that the set K2d = { x : | x y|  2d for some y 2 K } is included in W. Prove
that K2d is compact.

(b) Let j be the mollifier defined in the Exercise of Homework #3 and { je }e>0 the corresponding
approximation of identity. Prove that for e = d, the function yK = je ⇤ 1Kd provides a solution to
Urysohn’s problem.

*E. IX.6.2 You may admit that C • ( d ) is dense in L p ( d ). Using the previous question, prove that
•
Cc ( ) is dense in L p ( d ). [Hint: consider a non-decreasing sequence of compact sets {Kn }n2 such that
d

[Kn = and the sequence of functions {yKn }n2 .]

CS, 1st year 56 2023-2024

 Chapter X : Characteristic functions and Gaussian vectors

A) Aims of this class

After this class,

• I am able to give a necessary and sufficient condition for the independence of random variables
in terms of characteristic functions;

• I understand the definition of Gaussian vector and I do not mistake it for a vector with real-
valued Gaussian entries;

• I can recognise a Gaussian vector, based on its characteristic function or its density;

• I can give the necessary and sufficient condition for independence of the components of a Gaus-
sian vector in terms of its covariance matrix;

• I am able to perform changes of variables for Gaussian vectors.

57
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions X.1 and X.2 must be done before the examples class. The solutions for these exercises
are available online.

Question X.1

Q. X.1.1 Let n 2 ⇤ . Let X1 , X2 , . . . , Xn be independent random variables with X j ⇠ P ( j) (Poisson

law with parameter j) for all j = 1, . . . , n.
Using the characteristic function, give the distribution of Ânj=1 X j .
Q. X.1.2 Let Y ⇠ P (1) and Z ⇠ P (2) be Poisson random variables. Assume that for any (t1 , t2 ) 2
2,

n o
j(Y,Z) (t1 , t2 ) = exp eit1 + 2eit2 3 .

What can you say about Y and Z?

Question X.2

Q. X.2.1 For each matrix S below, argue whether there exists a Gaussian vector with mean zero
and covariance matrix S:
0 1 0 1
✓ ◆ 2 0 0 ✓ ◆ 3 1 0
1 0 2 1 2
S= ; S = @0 1 0A ; S = ; S = @1 3 0A .
2 0 1 2 1
0 0 3 0 0 6

CS, 1st year 58 2023-2024

CIP 2023-2024

C) Exercises

Exercise X.1
Let r 2 ( 1, 1) and ( X, Y ) be real-valued random variables, whose law has the density f : 2 ! +
defined by
✓ ◆
1 1 2 2
8 x, y 2 , f ( x, y) = p exp (x 2rxy + y ) .
2p 1 r2 2(1 r2 )

Y rX
E. X.1.1 Prove that the r.v. X and Z = p are independent and have the same distribution
1 r2
N (0, 1). Deduce the value of ( X > 0, Y > 0).

The following exercise describes a classical method to simulate Gaussian random variables starting
from uniform samples (which is roughly what a computer can do). This provides an alternative to the
method based on cumulative distribution functions that we have seen in class. Unlike the latter, the
Box-Muller method allows to simulate Gaussian vectors in N .

Exercise X.2 (The Box-Muller method)

Let (U, V ) be two independent r.v., each following the uniform law on [0, 1].
E. X.2.1 Prove that the following quantities

X = ( 2 ln U )1/2 cos(2pV )
Y = ( 2 ln U )1/2 sin(2pV )

are real-valued random variables.

E. X.2.2 Let h : 2 ! be a measurable and bounded mapping.

(a) Give [h( X, Y )] as an integral wrt the Lebesgue measure of 2.

(b) Using a change of variables, deduce that the law of ( X, Y ) is absolutely continuous wrt the
Lebesgue measure. Determine its density.

(c) Deduce that X and Y are independent and have same law N (0, 1).

We have already encountered random variables which are uncorrelated but not independent. Here
is another example.

Exercise X.3
Let X be a real-valued r.v. with law N (0, 1). For any a > 0, consider

Ya = X {| X |< a} X {| X | a} .

E. X.3.1 Compute [h(Y a )] for any bounded measurable function h. Deduce that the r.v. Y a is
Gaussian.
E. X.3.2 Is the 2-dimensional vector ( X, Y a ) Gaussian? (It may help to study the r.v. X + Y a .)

CS, 1st year 59 2023-2024

CIP 2023-2024

E. X.3.3 Prove that there exists b > 0 such that

Z b
1 t2 1
p t2 e 2 dt = .
2p 0 4

Compute Cov( X, Y b ).
Are the r.v. X and Y b independent?

In the following exercise, we will study the independence between linear combinations of a Gaus-
sian vector.

Exercise X.4
Let X be an n -valued Gaussian vector, with covariance matrix K.
E. X.4.1 Let M1 and M2 be matrices with dimension n1 ⇥ n and n2 ⇥ n respectively. Consider the
random vectors Y1 = M1 X and Y2 = M2 X.
✓ ◆
Y1
(a) Prove that is a Gaussian vector.
Y2

(b) Let SY1 Y2 = [Cov((Y1 )i , (Y2 ) j )]i,j . Find a condition on SY1 Y2 so that the random vectors Y1 and
Y2 are independent.

(c) Prove that Y1 and Y2 are independent if and only if M1 KM2T = 0. Generalise this to k vectors
Yi = Mi X (1  i  k).

E. X.4.2 Assume now that X has law N (0, In ). Let E be a sub-linear space of n and denote by E?
its orthogonal complementary subspace. Denote by A and B the matrices representing the orthogonal
projections p E and p E? on E and E? respectively.
Prove that the random vectors AX and BX are independent.
E. X.4.3 Deduce from the previous question that the following random variables are independent:

1 n
n iÂ
X= Xi and ( X1 X, . . . , Xn X ),
=1

where X = ( X1 , . . . , Xn ).

D) Going further

Exercise X.5
Let (W, F , ) be a probability space, let X1 , . . . , Xn be a family of independent real-valued standard
Gaussian r.v. N (0, 1). Let a1 , . . . , an and b1 , . . . , bn be real numbers.
The goal of this exercise is to find a necessary and sufficient condition for the independence of the
random variables Y = Âin=1 ai Xi and Z = Âin=1 bi Xi .
E. X.5.1 Is ( X1 , . . . , Xn ) a Gaussian vector?

CS, 1st year 60 2023-2024

CIP 2023-2024

E. X.5.2 Is (Y, Z ) a Gaussian vector?

E. X.5.3 Deduce a necessary and sufficient condition on a1 , . . . , an , b1 , . . . , bn to ensure independence
of Y and Z.

Exercise X.6 (Canonical form of a Gaussian vector)

E. X.6.1 For any n 2 ⇤, consider the family of n ⇥ n matrices ( Jk,n )1kn defined by Jn,n = In , and
for 1  k  n 1, ✓ ◆
Ik 0
Jk,n =
0 0
Prove that if K is a symmetric positive matrix of rank k, then there exists an invertible matrix A such
that K = AJk,n A T .
E. X.6.2 Let Y be an n -valued random vector, µ 2 n and K an n ⇥ n symmetric positive matrix.
Prove that Y follows the Gaussian distribution N (µ, K ) if and only if Y reads

Y = AJk,n X + µ

where X is an n -valued Gaussian vector N (0, In ), and A is an invertible matrix satisfying K =

AJk,n A T .

CS, 1st year 61 2023-2024

 Chapter XI : Convergence of random variables

A) Aims of this class

After this class,

• I know the following definitions of convergence for a sequence of random variables: conver-
gence in probability, almost sure convergence and convergence in L p ;

• I can prove that a sequence of r.v. converges;

• I am able to apply the Law of Large Numbers, after verifying carefully the assumptions of the
theorem;

• I know what convergence in law means;

• I can apply the Central Limit Theorem, after verifying carefully the assumptions of the theorem;

• I know clearly the difference between the conclusions of the LLN and the CLT.

62
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions XI.1 and XI.2 must be done before the examples class. The solutions for these exercises
are available online.

Question XI.1
Let ( Xn )n2 ⇤ be a sequence of square-integrable random variables (i.e. 8n 2 ⇤, X
n 2 L2 (W, F , )),
which converges in L2 to a r.v. X 2 L2 (W, F , ).
Q. XI.1.1 Prove that Var( Xn ) converges to Var( X ) as n ! •.

Question XI.2
Let ( Xn )n2 be a sequence of real-valued Gaussian random variables such that Xn ⇠ N (µn , sn2 ), n 2
. Assume that limn µn = µ and limn sn = s.
Q. XI.2.1 Prove that the sequence ( Xn )n2 converges in law to a Gaussian random variable X ⇠
N (µ, s2 ).

CS, 1st year 63 2023-2024

CIP 2023-2024

C) Exercises

Exercise XI.1
Let ( Xn )n2 ⇤ be a sequence of r.v. with probability densities ( f n )n2 ⇤ defined by
n
8x 2 , f n (x) = ,
p (1 + n2 x 2 )

for all n 1.
E. XI.1.1 What types of convergence hold for Xn , as n ! +•?

Exercise XI.2 (The Weierstrass theorem)

Let f : [0, 1] ! be a continuous mapping and let x 2 [0, 1]. For any n 2 ⇤ , consider the real-valued
r.v. Sn with binomial law B(n, x ).
In this exercise, we use the convergence of Sn /n to prove the Weierstrass theorem.
E. XI.2.1 Prove that pn : x 7! [ f ( Snn )] is a polynomial in x (called Bernstein’s polynomial of f ).
E. XI.2.2 Using the uniform continuity of f , prove that for any e > 0, there exists d > 0 such that
for any n 2 ⇤ and any x 2 [0, 1],

Sn
| pn ( x ) f ( x )|  f ( ) f (x)
n
✓ ◆ ✓ ◆
Sn Sn
e x < d +2 x d sup | f ( x )].
n n x 2[0,1]

Deduce that for any e > 0, there exists d > 0 such that for any n 2 ⇤ and any x 2 [0, 1],

x (1 x )
| pn ( x ) f ( x )|  e + 2 sup | f ( x )].
nd2 x2[0,1]

E. XI.2.3 Prove the Weierstrass theorem: Any continuous mapping from [0, 1] to is the uniform
limit of a sequence of polynomials.

Exercise XI.3 (Stirling’s formula)

Let ( Xn )n2 ⇤ be a sequence of i.i.d. random variables distributed according to the Poisson law P (1).
Set
n
Sn n
Sn = Â Xk and Tn = p .
k =1 n
Let a > 0 and define f a : x 7! inf(| x |, a).
E. XI.3.1 Prove that for X a real-valued r.v. in L2 (W, F , ),
1/2
[| X inf( X, a)|]  [X { X a} ]  [X2 ] (X a) .

E. XI.3.2 For all n 1, find the law of Sn and compute [ Tn2 ].

E. XI.3.3 Prove that ( Tn )n2 ⇤ converges in law to some r.v. Y and give its law. Deduce that [ f a ( Tn )]
admits a limit as n ! •.
Using Question 1., prove that [| Tn |] converges to [|Y |].

CS, 1st year 64 2023-2024

CIP 2023-2024

E. XI.3.4 Prove that [ Tn+ ] converges to [Y + ], where x + = max( x, 0).

E. XI.3.5 Compute [Y + ] and [ Tn+ ] for all n 2 ⇤.
Deduce Stirling’s formula
p ⇣ n ⌘n
n! ⇠ 2pn .
e

D) Going further

Exercise XI.4 (Convergence in law and density)

Let ( Xn )n2 and X be N -valued r.v. whose probability densities are repsectively denoted by ( f n )n2
and f . Assume that the sequence ( f n )n2 converges pointwise to f .
E. XI.4.1 Given any bounded measurable function h : N ! , consider the functions h1 and h2
defined by
d
8x 2 ; h1 ( x ) = h ( x ) + Mh ; h2 ( x ) = Mh h( x )

where Mh = supx | h( x )|.

Prove that for i = 1, 2

[hi ( X )]  lim inf [hi ( Xn )].

n!•

E. XI.4.2 Write [ h( X )] and [ h( Xn )] in terms of [h1 ( X )] and [h1 ( Xn )] and then in terms of
[h2 ( X )] and [h2 ( Xn )]. Then deduce that

[h( X )]  lim inf [h( Xn )]

n!•

and

lim sup [ h( Xn )]  [h( X )].

n!•

L
E. XI.4.3 Deduce that Xn ! X.
E. XI.4.4 What happens if ( f n )n2 does not converge almost everywhere?

Exercise XI.5
Let X be an ( E, E )-valued random variable. We want to estimate the probability ( X 2 A) for A 2 E .
Consider a sequence of i.i.d. random variables ( Xn )n2 ⇤ with the same law as X. For all n 1, we
study
Rn = Card {1  i  n : Xi 2 A} .
E. XI.5.1 Prove that for all n 1, Rn is a real-valued random variable.
E. XI.5.2 Compute [ Rn ] and Var( Rn ).
E. XI.5.3 Study the convergence in law of ( Rn )n2 ⇤ .

CS, 1st year 65 2023-2024

CIP 2023-2024

E. XI.5.4 Application : A preliminary study has proven that in a mass production, 3% of the pieces
manufactured by a particular machine were defective. A customer receives 500 pieces coming from
that machine.

(a) What is the probability that less than 1% of the pieces received by the customer are defective?

(b) The contract signed with the manufacturer allows the client to send back the 500 pieces if more
than 5% of the pieces are defective. What is the probability that the client sends back the 500
pieces?

CS, 1st year 66 2023-2024

 Chapter XII : Conditional expectation and introduction to stochastic processes

– Exercises to be studied in autonomy after Lecture 11 (12 December 2023) –

A) Aims of this class

After this class,

• I know the characterisation of the conditional expectation of an integrable r.v. given a sub-
s-algebra;

• I understand the conditional expectation in L2 as an orthogonal projection;

• I can find the conditional expectation of a random variable;

• I know the properties of the conditional expectation and I am able to work with this object.

67
CIP 2023-2024

B) To become familiar with this class’ concepts (to prepare before the examples class)

Questions XII.1 and XII.2 must be done before the examples class. The solutions for these exercises
are available online.

Question XII.1
Let
⇥ X be a real-valued ⇤ r.v. in L2 (W, F , ) and let G be a sub-s-algebra of F . Define Var( X | G) =
(X [ X | G])2 | G .
Q. XII.1.1 Prove that
⇥ ⇤
Var( X ) = Var( X | G) + Var [ X | G] .

Question XII.2 (Conditionning with respect to a discrete random variable)

Let (W, F , ) be a probability space, let X 2 L1 (W, F , ) and let Y be a real-valued r.v. such that
(Y 2 {yn ; n 2 }) = 1 and (Y = yn ) > 0 for all n.
Q. XII.2.1 Determine [ X | Y ].

CS, 1st year 68 2023-2024

CIP 2023-2024

C) Exercises

In the case of Gaussian random variables, conditional expectations can be written using covariance
matrices. This observation is important in statistics when dealing with linear regressions.

Exercise XII.1 (Gaussian expectations)

Let X and Y be respectively n and p -valued random variables, such that ( X, Y ) is a Gaussian vector.
Assume that X is a non-degenerate Gaussian vector.
E. XII.1.1 In this question, assume that [ X ] = [Y ] = 0. Set

U=Y SYX KX 1 X,

where KX = [Cov( Xi , X j )]i,j and SYX = [Cov(Yi , X j )]i,j .

(a) Prove that (U, X ) is Gaussian.

(b) Prove that X and U are independent.

(c) Deduce that

[Y | X ] = SYX KX 1 X.

E. XII.1.2 Prove that [Y | X ] can be written as follows

[Y | X ] = [Y ] + SYX KX 1 ( X [ X ]).

E. XII.1.3 Prove that Y [Y | X ] and X are independent.

The conditional expectation allows to characterise the independence of random variables.

Exercise XII.2

E. XII.2.1 Let X and Y be real-valued random variables. Prove that X and Y are independent if and
only if for any bounded and Borel-measurable mapping g : ! , one has

[ g (Y ) | X ] = [ g(Y )] a.s.

E. XII.2.2 Application : Let ( X, Y ) be a r.v. with probability density function p : ( x, y) 7! e y {0< x<y} ( x, y).
Compute the conditional distribution of Y given that X = x. Deduce that X and Y X are indepen-
dent.

Exercise XII.3 (Sum of a random number of random variables)

Let ( Xn )n2 ⇤ be a sequence of i.i.d. real-valued random variables and let N be an -valued r.v. which
is independent of ( Xn )n2 ⇤ . Assume that the random variables N and Xn , n 2 ⇤have a finite first
moment.
Define
N (w )
Y (w ) = Â Xi ( w ) .
i =1

E. XII.3.1 Prove that Y is a random variable. Compute [Y ] .

CS, 1st year 69 2023-2024

CIP 2023-2024

D) Going further

Exercise XII.4 (The Radon-Nikodym derivative)

Let (W, F , ) be a probability space and let X be a real-valued r.v. with a continuous probability
density f . Let A 2 F .
Assume that there exists a continuous function g such that for any x 2 ,

( A \ { X 2 [ x, x + h)})
! g ( x ).
( X 2 [ x, x + h)) h !0

E. XII.4.1 Prove that ( A | X ) = g( X ) -a.s.

Exercise XII.5
Let ( Xn )n2 be a sequence of r.v. and (Fn )n2 be a filtration. Assume that for any n, Xn is Fn -
measurable and [ Xn+1 | Fn ] = 0. Set Sn = X0 + · · · + Xn for all n 2 .
E. XII.5.1 Prove that {Sn ; n 2 } is an Fn -martingale.

Exercise XII.6 (Doob’s decomposition)

Let { Xn ; n 2 } ba a square-integrable martingale for the filtration (Fn )n2 .
E. XII.6.1 Prove that there exists a non-decreasing predictable process { An ; n 2 } and an Fn -
martingale {Yn ; n 2 } such that ( X02 ) = A0 and

8n 2 ; Xn2 = An + Yn .

E. XII.6.2 Application. Let ( Tn )n2 ⇤ be a sequence of real-valued i.i.d square-integrable random

variables such that [ Tn ] = 0 and [ Tn2 ] = s2 . Set S0 = 0 and Sn = T1 + · · · + Tn for all n 1.
Prove that {Sn2 ns2 ; n 2 } is a martingale.

Exercise XII.7
Let ( Xn )n2 be a sequence of real-valued i.i.d. random variables defined on the probability space
(W, F , ), and with values in the measurable space ( E, E ). Denote by µ the law of Xn .
Let N1 and N2 be ⇤ -valued random variables such that 1  N1 < N2 . Assume that for any
m 2 ⇤ , the event { Ni = m} depends only on X0 , X1 , . . . , Xm 1 .
E. XII.7.1 Prove that the random variables X N1 and X N2 are i.i.d.

CS, 1st year 70 2023-2024