Lecture 1

Preliminaries

We present the basic notions of measure theory, with the aim of fixing the notation and
making the exposition self-contained. We deal only with finite measures, even though
of course positive infinite measures are the first natural examples. But, assuming that
the basic notions relative to positive measures are known, we go straight to finite real
measures because all measures we are going to discuss are so. In the first section we
present real measures and the related notions of Lp space, absolute continuous and singular
measure, the Radon-Nikodym theorem, weak convergence and product measures. In the
case of topological spaces we introduce Borel and Radon measures. Next, we introduce
characteristic functions (or Fourier transforms) of measures and Gaussian measures in Rd .

1.1 Abstract Measure Theory

We start by introducing measurable spaces, i.e., sets equipped with a -algebra.

Definition 1.1.1 ( -algebras and measurable spaces). Let X be a nonempty set and let
F be a collection of subsets of X.

(a) We say that F is an algebra if ; 2 F , E1 [ E2 2 F and X \ E1 2 F whenever

E1 , E 2 2 F .

(b) S
We say that an algebra F is a -algebra if for any sequence (Eh ) ⇢ F its union
h Eh belongs to F .

(c) For any collection G of subsets of X, the -algebra generated by G is the smallest
-algebra containing G . If (X, ⌧ ) is a topological space, we denote by B(X) the
-algebra of Borel subsets of X, i.e., the -algebra generated by the open subsets of
X.

(d) If F is a -algebra in X, we call the pair (X, F ) a measurable space.

1
2

It is obvious by the De Morgan laws that algebras are closed under finite intersections,
and -algebras under countable intersections. Moreover, since the intersection of any
family of -algebras is a -algebra and the set of all subsets of X is a -algebra, the
definition of generated -algebra is well posed. Once a -algebra has been fixed, it is
possible to introduce positive measures.
Definition 1.1.2 (Finite measures). Let (X, F ) be a measurable space and µ : F !
[0, +1). We say that µ is a positive finite measure if µ(;) = 0 and µ is -additive on F ,
i.e., for any sequence (Eh ) of pairwise disjoint elements of F the equality
1
! 1
[ X
µ Eh = µ(Eh ) (1.1.1)
h=0 h=0

holds. We say that µ is a probability measure if µ(X) = 1.

We say that µ : F ! R is a (finite) real measure if µ = µ1 µ2 , where µ1 and µ2 are
positive finite measures.
If µ is a real measure, we define its total variation |µ| for every E 2 F as follows:
(1 1
)
X [
|µ|(E) := sup |µ(Eh )| : Eh 2 F pairwise disjoint, E = Eh (1.1.2)
h=0 h=0

and it turns out to be a positive measure, see Exercise 1.2. If µ is real, then (1.1.1) still
holds, and the series converges absolutely, as the union is independent of the order. Notice
also that the following equality holds:
nZ o
|µ|(Rd ) = sup f dµ : f 2 Cb (Rd ), kf k1  1 , (1.1.3)
Rd

see Exercise 1.3.

Remark 1.1.3. (Monotonicity) Any positive finite measure µ is monotone with respect
to set inclusion and continuous along monotone sequences, i.e., if (Eh ) is an increasing
sequence of sets (resp. a decreasing sequence of sets), then
1
! 1
!
[ \
µ Eh = lim µ(Eh ), resp. µ Eh = lim µ(Eh ),
h!1 h!1
h=0 h=0

see Exercise 1.1.

Definition 1.1.4 (Radon measures). A real measure µ on the Borel sets of a topological
space X is called a real Radon measure if for every B 2 B(X) and " > 0 there is a
compact set K ⇢ B such that |µ|(B \ K) < ".
A measure is tight if the same property holds with B = X.
Proposition 1.1.5. If (X, d) is a separable complete metric space then every real measure
on (X, B(X)) is Radon.
 3

Proof. Observe that it is enough to prove the result for finite positive measures. The
general case follows splitting the given real measure into its positive and negative parts.
Let then µ be a positive finite measure on (X, B(X)). Let us first show that it is a regular
measure, i.e., for any B 2 B(X) and for any " > 0 there are an open set G B and a
closed set F ⇢ B such that µ(G \ F ) < ". Indeed, for a given " > 0, if B = F is closed
T to consider open sets G = {x 2 X : d(x, F ) = inf y2F d(x, y) < }, getting
it suffices
F = >0 G . As µ(G ) ! µ(F ) as ! 0, fixed " > 0, for small enough by Remark
1.1.3 we have µ(G \ F ) < ". Next, we show that the family G containing ; and all sets
B such that for any " > 0 there are an open set G B and a closed set F ⇢ B such that
µ(G\F ) < " is a -algebra. To this aim, given a sequence (Bn ) ⇢ G , consider open sets Gn
and closed sets Fn such that Fn ⇢ Bn ⇢ Gn and µ(Gn \ Fn ) < "/2n+1 . For G = [1 n=1 Gn
and F = [N F , with N 2 N such that µ([ 1 F \ F ) < "/2, we have F ⇢ [ B ⇢ G
n=1 n n=1 n n n
and µ(G \ F ) < ". Therefore, G is closed under countable unions, and, since it is closed
under complementation as well, it is a -algebra.
Since we have proved that all closed sets belong to G , then it coincides with B(X).
As a consequence, we prove that any positive finite measure on (X, B(X)) is Radon
i↵ it is tight. If µ is Radon then it is tight by definition. Conversely, assuming that µ is
tight, for every " > 0 and every Borel set B ⇢ X, we may take a compact set K1 such
that µ(X \ K1 ) < " and a closed set F ⇢ B such that µ(B \ F ) < ". Then, defining the
compact set K := K1 \ F we have µ(B \ K) < 2".
Therefore, to prove our statement it suffices to show that every Borel measure on X is
tight. Let (xn ) be a dense sequence and notice that X ⇢ [1 n=1 B(xn , 1/k) for every k 2 N.
Then, given " > 0, for every k 2 N there is Nk 2 N such that
⇣[
Nk ⌘
µ B(xn , 1/k) > µ(X) "/2k .
n=1
Then, the compact set
Nk
1 [
\
K := B(xn , 1/k)
k=1 n=1
verifies µ(K) > µ(X) ".
Notice that property (1.1.3) holds in any metric space, but we shall use it only in Rd .
Let us come to measurable functions.
Definition 1.1.6 (Measurable functions). Let (X, F , µ) be a measure space and let Y be
a topological space. A function f : X ! Y is said to be F -measurable (or µ-measurable)
if f 1 (A) 2 F for every open set A ⇢ Y .
If (Y, G ) is a measurable space, a function f : X ! Y is said to be F -measurable (or
µ-measurable) if f 1 (A) 2 F for every A 2 G .
In particular, if f is F -measurable then f 1 (B) 2 F for every B 2 B(Y ). For E ⇢ X
we define the indicator (or characteristic) function of E, denoted by 1lE , by
(
1 if x 2 E
1lE (x) :=
0 if x 62 E.
4

and we say that f : X ! R is a simple function if the image of f is finite, i.e., if f belongs
to the vector space generated by the indicator functions. We assume that the readers are
familiar with the usual notion of integral of a measurable function. We now define the Lp
(semi)-norms and spaces as follows,
✓Z ◆1/p
p
kukLp (X,µ) := |u| dµ
X

if 1  p < 1, and

kukL1 (X,µ) := inf {C 2 [0, +1] : |u(x)|  C for µ-a.e. x 2 X} .

We define the space Lp (X, µ) as the space of equivalence classes of functions agreeing
µ-a.e. such that kukLp (X,µ) < 1. In this way, k · kLp (X,µ) is a norm and Lp (X, µ) is a
Banach space, see e.g. [D, Theorem 5.2.1]. When there is no risk of confusion, we use the
shorter notation k · kp .
We assume that the reader is familiar also with the properties of integrals, measur-
able functions and Lp spaces as well as the main convergence theorems of Levi, Fatou,
Lebesgue, see e.g. [D, Section 4.3]. We just recall the Lebesgue-Vitali theorem on uni-
formly integrable sequences, see Exercise 1.4.
Theorem 1.1.7 (Lebesgue-Vitali Convergence Theorem). Let (X, F ) be a measurable
space, let µ be a positive finite measure on it and let (fk ) be a sequence of measurable
functions such that Z
lim sup |fk | dµ = 0.
M !1 k2N {|fk |>M }

If fk ! f in measure, i.e.,

lim µ({x 2 X : |fk (x) f (x)| > "}) = 0 for every " > 0 (1.1.4)
k!1
R
and |f (x)| < 1 µ-a.e., then f 2 L1 (X, µ) and limk!1 X |f fk |dµ = 0.
Given a -algebra, we have defined the class of measurable functions. Conversely, given
a family of functions, it is possible to define a suitable -algebra.
Definition 1.1.8. Given a family F of functions f : X ! R, let us define the -algebra
E (X, F ) generated by F on X as the smallest -algebra such that all the functions f 2 F
are measurable, i.e., the -algebra generated by the sets {f < t}, with f 2 F and t 2 R.
Given a metric space, the set of real Borel measures µ is a vector space in an obvi-
uos way. All continuous and bounded functions are in L1 (X, µ) and we define the weak
convergence of measures by
Z Z
µj ! µ () f dµj ! f dµ 8f 2 Cb (X). (1.1.5)
X X

Let us now introduce the notions of absolute continuity and singularity of measures. Let
µ be a positive finite measure and ⌫ a real measure on the measurable space (X, F ). We
 5
this paragrapgh is very important
say that ⌫ is absolutely continuous with respect to µ, and write ⌫ ⌧ µ, if µ(B) = 0 =)
|⌫|(B) = 0 for every B 2 F . If µ, ⌫ are real measures, we say that they are mutually
singular, and write ⌫ ? µ, if there exists E 2 F such that |µ|(E) = 0 and |⌫|(X \ E) = 0.
Notice that for mutually singular measures µ, ⌫ the equality |µ + ⌫| = |µ| + |⌫| holds.
If µ ⌧ ⌫ and ⌫ ⌧ µ we say that µ and ⌫ are equivalent and write µ ⇡ ⌫. If µ is a
positive measure and f 2 L1 (X, µ), then the measure ⌫ := f µ defined below is absolutely
continuous with respect to µ and the following integral representations hold, see Exercise
1.2: Z Z
⌫(B) = f dµ, |⌫|(B) = |f | dµ 8B 2 F . (1.1.6)
B B
In the following classical result we see that if a real measure ⌫ is absolutely continuous
with respect to µ, then the above integral representation holds, with a suitable f .
Theorem 1.1.9 (Radon-Nikodym). Let µ be a positive finite measure and let ⌫ be a real
measure. Then there is a unique pair of real measures ⌫ a , ⌫ s such that ⌫ a ⌧ µ, ⌫ s ? µ
and ⌫ = ⌫ a + ⌫ s . Moreover, there is a unique function f 2 L1 (X, µ) such that ⌫ a = f µ.
The function f is called the density (or Radon-Nikodym derivative) of ⌫ with respect to µ
and is denoted by d⌫/dµ.
Since trivially each real measure µ is absolutely continuous with respect to |µ|, from
the Radon-Nikodym theorem the polar decomposition of µ follows: there exists a unique
real valued function f 2 L1 (X, |µ|) such that µ = f |µ| and |f | = 1 |µ|-a.e.
The following result is a useful criterion of mutual singularity.
Theorem 1.1.10 (Hellinger). Let µ, ⌫ be two probability measures on a measurable space
(X, F ), and let be a positive measure such that µ ⌧ , ⌫ ⌧ . Then the integral
Z r
dµ d⌫
H(µ, ⌫) := d
X d d
is independent of and
p
2(1 H(µ, ⌫))  |µ ⌫|(X)  2 1 H(µ, ⌫)2 . (1.1.7)
Proof. Let us first take = µ + ⌫ and notice that µ, ⌫ ⌧ . Then, setting f := dµ/d and
g := d⌫/d , i.e., µ = f and ⌫ = g , we have |µ ⌫|(X) = kf gkL1 (X, ) and integrating
the inequalities p p
p 2 p p p
( f g)  |f g| = | f g| | f + g|
we get
Z p Z
p 2
( f g) d = 2(1 H(µ, ⌫))  |f g| d = |µ ⌫|(X)
X Z p X
p p p
= | f g| | f + g| d
X
⇣Z p p 2 ⌘1/2 ⇣
Z p
p ⌘1/2
 | f g| d | f + g|2 d
X X
p
= (2 2H(µ, ⌫))1/2 (2 + 2H(µ, ⌫))1/2 = 2 1 H(µ, ⌫)2 ,
6

where we have used the Cauchy-Schwarz inequality. If 0 is another measure such that
µ = f 0 0 ⌧ 0 and ⌫ = g 0 0 ⌧ 0 , then ⌧ 0 : setting := dd 0 , we have f 0 = f, g 0 = g
and then
Z r Z p Z p Z p Z r
dµ d⌫ dµ d⌫
d = fg d = fg d 0 = f 0g0 d 0 = d 0.
X d d X X X X d 0d 0

Corollary 1.1.11. If µ and ⌫ are probability measures, then µ ? ⌫ i↵ H(µ, ⌫) = 0.

Proof. It is obvious from Hellinger’s theorem that |µ ⌫|(X) = 2 if and only if H(µ, ⌫) = 0.
Let us show that this is equivalent to µ ? ⌫. Using the notation in the proof of Theorem
1.1.10, notice that H(µ, ⌫) = 0 if and only if the set F defined by F := {f g 6= 0}
verifies (F ) = 0 (hence also µ(F ) = ⌫(F ) = 0). Therefore, for the measurable set
E = {f = 0, g > 0} we have µ(E) = ⌫(X \ E) = 0 and the thesis follows.

We recall the notions of push-forward of a measure (or image measure) and the con-
structions and main properties of product measure. The push-forward of a measure gen-
eralises the classical change of variable formula.
Definition 1.1.12 (Push-forward). Let (X, F ) and (Y, G ) be measurable spaces, and let
f : X ! Y be such that f 1 (F ) 2 F whenever F 2 G . For any positive or real measure µ
on (X, F ) we define the push-forward measure or the law of µ under f µ f 1 , sometimes
denoted by f# µ, in (Y, G ) by
1 1
µ f (F ) := µ f (F ) 8F 2 G .
The change of variables formula immediately follows from the previous definition. If
u 2 L1 (Y, µ f 1 ), then u f 2 L1 (X, µ) and we have the equality
Z Z
1
u d(µ f ) = (u f ) dµ. (1.1.8)
Y X

The above relation is nothing but the definition for simple functions, and is immediately
extended to the whole of L1 by density.
We consider now two measure spaces and describe the natural resulting structure on
their cartesian product.
Definition 1.1.13 (Product -algebra). Let (X1 , F1 ) and (X2 , F2 ) be measure spaces.
The product -algebra of F1 and F2 , denoted by F1 ⇥ F2 , is the -algebra generated in
X1 ⇥ X2 by
G = {E1 ⇥ E2 : E1 2 F1 , E2 2 F2 } .
Remark 1.1.14. Let E 2 F1 ⇥ F2 ; then for every x 2 X1 the section Ex := {y 2
X2 : (x, y) 2 E} belongs to F2 , and for every y 2 X2 the section E y := {x 2 X1 : (x, y) 2
E} belongs to F1 . In fact, it is easily checked that the families
Gx := {F 2 F1 ⇥ F2 : Fx 2 F2 } , G y := {F 2 F1 ⇥ F2 : F y 2 F1 }
are -algebras in X1 ⇥ X2 and contain G see Exercise 1.5.
 7

Theorem 1.1.15 (Fubini). Let (X1 , F1 , µ1 ), (X2 , F2 , µ2 ) be measure spaces with µ1 , µ2

positive and finite. Then, there is a unique positive finite measure µ on (X1 ⇥X2 , F1 ⇥F2 ),
denoted also by µ1 ⌦ µ2 , such that
µ(E1 ⇥ E2 ) = µ1 (E1 ) · µ2 (E2 ) 8E1 2 F1 , 8E2 2 F2 .
Furthermore, for any µ-measurable function u : X1 ⇥ X2 ! [0, 1] the functions
Z Z
x 7! u(x, y) µ2 (dy) and y 7! u(x, y) µ1 (dx)
X2 X1

are respectively µ1 -measurable and µ2 -measurable and

Z Z ✓Z ◆
u dµ = u(x, y) µ2 (dy) µ1 (dx)
X1 ⇥X2 X1 X2
Z ✓Z ◆
= u(x, y) µ1 (dx) µ2 (dy).
X2 X1

Remark 1.1.16. More generally, it is possible to construct a product measure on infinite

cartesian products. If I is a set of indices, typically I = [0, 1] or I = N, and (Xt , Ft , µt ),
t 2 I, is a family of measure spaces, the product -algebra is that generated by the family
of sets of the form
°
B = B1 ⇥ · · · ⇥ Bn ⇥ X t , Bk 2 F t k ,
t2I\{t1 ,...,tk }

whose measure is µ(B) = µt1 (B1 ) · · · µtn (Bn ).

In the sequel we shall sometimes encounter some ideas coming from probability theory
and stochastic analysis. In order to simplify several computations concerning probability
measures on Rd , it is often useful to use characteristic functions of measures. This is the
probabilistic counterpart of Fourier transform. Indeed, given a probability measure µ on
(Rd , B(Rd )), we define its characteristic function by setting
Z
µ̂(⇠) := eix·⇠ µ(dx), ⇠ 2 Rd . (1.1.9)
Rd

We list the main elementary properties of characteristic functions, whose proofs are in
Exercise 1.6.
1. µ̂ is uniformly continuous on Rd ;
2. µ̂(0) = 1;
3. if µ̂1 = µ̂2 then µ1 = µ2 ;
4. if µj ! µ in the sense of (1.1.5), then µ̂j ! µ̂ uniformly on compacts;
5. if (µj ) is a sequence of probability measures and there is : Rd ! C continuous in
⇠ = 0 such that µ̂j ! pointwise, then there is a probability measure µ such that
µ̂ = .
8

1.2 Gaussian measures

Gaussian (probability) measures are the main reference measures we shall encounter in the
Lectures. Let us start from the finite dimensional case. We recall the following elementary
equality
Z n (x a)2 o
1
p exp dx = 1 (1.2.1)
2⇡ R 2 2
that holds for all a 2 R and > 0. An easy way to prove (1.2.1) is to compute the double
integral Z
exp{ (x2 + y 2 )} dxdy
R2

in polar coordinates and apply Fubini Theorem 1.1.15 and the change of variables formula.

Definition 1.2.1 (Gaussian measures on R). A probability measure on (R, B(R)) is

called Gaussian if it is either a Dirac measure a at a point a (in this case, we put = 0),
or a measure absolutely continuous with respect to the Lebesgue measure 1 with density

1 n (x a)2 o
p exp .
2⇡ 2 2

In this case we call a the mean, the mean-square deviation and 2 the variance of and
we say that is centred or symmetric if a = 0 and standard if = 1.

By elementary computations we get

Z Z
2
a= x (dx), = (x a)2 (dx).
R R

1 2 2
Remark 1.2.2. For every a, 2 R we have ˆ (⇠) = eia⇠ 2 ⇠ , see Exercise 1.7. Con-
versely, by property 3 of characteristic functions, a probability measure on R is Gaussian
i↵ its characteristic function has this form. Therefore, it is easy to recognise a Gaussian
measure from its characteristic function. This is true in Rd , as we are going to see in
Proposition 1.2.4, and also in infinite dimensions, as we shall see in the next Lecture.

Let us come to Gaussian measures in Rd .

Definition 1.2.3 (Gaussian measures on Rd ). A probability measure on Rd is said to

be Gaussian if for every linear functional ` on Rd the measure ` 1 is Gaussian on R.
2
The first example of Gaussian measure in Rd is d := (2⇡) d/2 e |x| d , that is called
standard Gaussian measure. We denote by Gd the standard Gaussian density in Rd , i.e.,
the density of d with respect to d . Notice also that if d = h + k then d = h ⌦ k .
The following result gives an useful characterisation of a Gaussian measure through
its characteristic function.
 9

Proposition 1.2.4. A measure on Rd is Gaussian if and only if its characteristic

function is
n 1 o
ˆ (⇠) = exp ia · ⇠ Q⇠ · ⇠ (1.2.2)
2
for some a 2 Rd and Q nonnegative d ⇥ d symmetric matrix. Moreover, is absolutely
continuous with respect to the Lebesgue measure d if and only if Q is nondegenerate. In
this case, the density of is

1 n 1 o
1
p exp Q (x a) · (x a) . (1.2.3)
(2⇡)d detQ 2

Proof. Let be a measure such that (1.2.2) holds. Then, for every linear funtional ` :
Rd ! R (here we identify ` with the vector in Rd such that `(x) = ` · x) we may compute
the characteristic function of the measure µ` := ` 1 on R:
Z Z n o
i⌧ t ⌧2
µb` (⌧ ) = e µ` (dt) = ei⌧ `(x) (dx) = ˆ (⌧ `) = exp i⌧ a · ` Q` · `
R Rd 2

by (1.2.2). Therefore, by Remark 1.2.2 µ` is a Gaussian measure with mean a` = a · ` and

variance `2 = Q` · `, and also is a Gaussian measure by the arbitrariness of `.
Conversely, assume that µ` is Gaussian for every ` as above. Its mean a` and its
variance `2 are given by
Z Z ⇣Z ⌘
a` := t µ` (dt) = `(x) (dx) = ` · x (dx) (1.2.4)
ZR Rd Z Rd
2
` := (t a` )2 µ` (dt) = (`(x) a` )2 (dx). (1.2.5)
Rd Rd

These formulas show that the map ` 7! a` is linear and the map ` 7! `2 is a nonnegative
quadratic form. Therefore, there are a vector a 2 Rd and a nonnegative definite symmetric
matrix Q = (Qij ) such that a` = a · ` and `2 = Q` · `, whence (1.2.2) follows. Notice that
Z Z
a= x (dx), Qij = (xi ai )(xj aj ) (dx).
Rd Rd

To prove the last part of the statement, let us assume that ⌧ d , i.e. =f d. We
want to show that Q` · ` = 0 i↵ ` = 0. From (1.2.4), (1.2.5) we have
Z
that Q is nondegenrate Q` · ` = (` · (x a))2 f (x)dx,
Rd

then Q` · ` = 0 i↵ (` · (x a))2 = 0 for a.e. x 2 Rd , i.e. i↵ ` = 0, as f 6⌘ 0. Hence Q is

nondegenerate.
Viceversa, if Q is nondegenerate, we consider the measure ⌫ = f d with f given by (1.2.3)
and we compute its characteristic function. Using the change of variable z = Q 1/2 (x a),
10

since d = ⌦dj=1 1 , we have:

Z Z n o
1 1
⌫ˆ(⇠) = exp{i⇠ · x}f (x)dx = p exp i⇠ · x Q 1 (x a) · (x a) dx
Rd (2⇡)d detQ Rd 2
Z Yd Z
= exp{i⇠ · a} exp{iQ1/2 ⇠ · z} d (dz) = exp{i⇠ · a} exp{i(Q1/2 ⇠)j t} 1 (dt)
Rd j=1 R
d
Y d
Y ⇢
1
= exp{i⇠ · a} ˆ1 ((Q1/2 ⇠)j ) = exp(i⇠ · a) exp ((Q1/2 ⇠)j )2
2
j=1 j=1
⇢
1
= exp i⇠ · a Q⇠ · ⇠ = ˆ (⇠).
2

Hence, by property 3 of the characteristic function, = ⌫.

Remark 1.2.5. If is a Gaussian measure and (1.2.2) holds, we call a the mean and Q
the covariance of , and we write = N (a, Q) when it is useful to emphasise the relevant
parameters. If the matrix Q is invertible then the Gaussian measure = N (a, Q) is said
to be nondegenerate. Its density, given by (1.2.3), is denoted Ga,Q . The nondegeneracy is
equivalent to the fact that ` ⌧ 1 for every ` 2 Rd .

Proposition 1.2.6. Every centred Gaussian measure on Rd is invariant under the

rotation map defined, for every ✓ 2 R, by : Rd ⇥ Rd ! Rd by (x, y) := sin ✓x + cos ✓y;
then, the image measure ( ⌦ ) 1 in Rd is .

Proof. We use characteristic functions and Proposition 1.2.4. Indeed, the characteristic
function of is exp{ 12 Q⇠ ·⇠} for some nonnegative d⇥d matrix Q. Then we may compute
the characteristic function of µ := 1 as follows:

Z
µ̂(⇠) = eiz⇠ µ(dz)
ZR
d

= ei(sin ✓x+cos ✓y)⇠ ⌦ (d(x, y))

ZR ⇥R
d d
Z
i(sin ✓x)⇠
= e (dx) ei(cos ✓y)⇠ (dy)
Rd Rd
1 1
= exp{ sin2 ✓Q⇠ · ⇠} exp{ cos2 ✓Q⇠ · ⇠}
2 2
1
= exp{ Q⇠ · ⇠},
2
and the thesis follows from property 3 of characteristic functions.

Remark 1.2.7. We point out that the property stated in Proposition 1.2.6 is not the
invariance of under rotations in Rd . Indeed, rotation invariance holds i↵ the covariance
of is a positive multiple of an orthogonal matrix.
 11

1.3 Exercises
Exercise 1.1. Let µ be a positive finite measure on (X, F ). Prove the monotonicity
properties stated in Remark 1.1.3.

Exercise 1.2. Prove that the set function |µ| defined in (1.1.2) is a positive finite measure
and that the integral representation for |⌫| = |f µ| in (1.1.6) holds. Prove also that if µ ? ⌫
then |µ + ⌫| = |µ| + |⌫|.

Exercise 1.3. Prove the equality (1.1.3).

Exercise 1.4. Prove the Vitali–Lebesgue Theorem 1.1.7.

Exercise 1.5. Prove that the families Gx and G y defined in Remark 1.1.14 are -algebras.

Exercise 1.6. Prove the properties of characteristic functions listed in Section 1.1.
1 2 ⇠2
Exercise 1.7. Prove the equality ˆ (⇠) = eia⇠ 2 stated in Remark 1.2.2.

Exercise 1.8. (Layer cake formula) Prove that if µ is a positive finite measure on (X, F )
and 0  f 2 L1 (X, µ) then
Z Z 1
f dµ = µ {x 2 X : f (x) > t} dt.
X 0

Bibliography
[B] V. I. Bogachev: Gaussian Measures. American Mathematical Society, 1998.

[D] R. M. Dudley: Real Analysis and Probabillity, Cambridge University Press, 2004.