Mat M
Mat M
Mat M
Jan Dereziński
Department of Mathematical Methods in Physics
Warsaw University
Hoża 74, 00-682, Warszawa, Poland
Contents
1 Measurability 3
1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Rings and fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Ordered spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Elementary functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 σ-rings and σ-fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.6 Transport of subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Transport of σ-rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.8 Measurable transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.9 Measurable real functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.10 Spaces L∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1
3 Extension of a measure 20
3.1 Hereditary families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Extension of a measure by null sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Complete measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 External measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.5 External measure generated by a measure . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 Extension of a measure to localizable sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.7 Sum-finite measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.8 Boolean rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.9 Measures on Boolean rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6 Measures in Rn 35
6.1 Regular contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2 Borel sets in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3 Borel premeasures on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.4 Borel measures on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.5 The Cantor set and devil’s staircase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.6 Transport of the Lebesgue measure in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.7 The Lebesgue measure in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.8 Transport of the Lebesgue measure in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2
8 Measures on topological spaces 49
8.1 δ-open and σ-closed sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
8.2 Baire and Borel sets of 1st kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
8.3 Baire and Borel sets of 2nd kind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
8.4 Baire measures on compact spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
8.5 Borel measures on compact spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
8.6 Baire measures on locally compact spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
8.7 Borel measures on locally compact spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1 Measurability
1.1 Notation
X
2 denotes the family of subsets of the set X. The symmetric difference is defined as
Let A1 , A2 , · · · ∈ X.
We write An % A, if An ⊂ An+1 , n ∈ N and ∪∞
n=1 An = A.
We write An & A, if An ⊃ An+1 , n ∈ N and ∩∞
n=1 An = A.
Proof. A ∩ B = A\(A\B). 2
If (Ri )i∈I is a family of rings in X, then so is ∩i∈I Ri . Hence for any T ⊂ 2X there exists the smallest
ring containing T . We denote it by Ring(T ).
Equivalently, a field is a ring containing X. (Field is a ring, because A\B = X\((X\A) ∪ B)).
For T ⊂ 2X , Field(T ) denotes the smallest field of sets containing T .
3
1.3 Ordered spaces
Suppose that (X, ≤) is an ordered set. Let U be a nonempty subset of X.
We say that u0 is a largest minorant of U if
(1) u ∈ U implies u0 ≤ u
(2) u1 ≤ u for all u ∈ U implies u1 ≤ u
If U possesses a largest minorant, then it is uniquely defined. The largest minorant of a set {x1 , x2 }
is often denoted x1 ∧ x2 and of a set U is denoted ∧ x.
x∈U
Analogously we define the smallest majorant of U . The smallest majorant of a set {x1 , x2 } is often
denoted x1 ∨ x2 and of a set U is denoted ∨ x.
x∈U
We say that (X, ≤) is a lattice if every two-element (hence every finite) set of elements of X possess
the smallest majorant and the largest minorant. It is a countably complete lattice if every countable
subset that has a majorant and a minorant has the smallest majorant and the largest minorant. It is a
complete lattice if every countable subset that has a majorant and a minorant has the smallest majorant
and the largest minorant.
Let X be a vector space. (X , ≤) is an ordered vector space iff
(1) x, y, z ∈ X , x ≤ y ⇒ x + z ≤ y + z;
(2) x ∈ X , x ≥ 0, λ ∈ R, λ ≥ 0 ⇒ λx ≥ 0.
X+ := {x ∈ X : x ≥ 0} is a cone called the positive cone.
We say that an ordered vector space (X , ≤) is a Riesz space if it is a lattice. It is enough to check
that it has ∨ of two elements, since
x ∧ y := −(−x) ∨ (−y).
4
(3) A1 , A2 , · · · ∈ F, An % A ⇒ A ∈ F,
∞
Proof. Let us prove (1). Clearly, A := ∪ Ai ∈ F. Now by the de Morgan’s law
i=1
∞ ∞
∩ Ai = A\ ∪ (A\Ai ) ∈ F.
i=1 i=1
2
For T ⊂ 2X , σ−Ring(T ) denotes the smallest σ-ring of sets containing T .
Theorem 1.8 Let T ⊂ 2X and A ∈ σ − Ring(T ). Then there exists a countable T0 ⊂ T such that
A ∈ σ−Ring(T0 ).
Proof. Let F be the family of A ⊂ X such that there exists a countable T0 ⊂ T with A ∈ σ−Ring(T0 ).
Then T ⊂ F and F is a σ-ring. Hence σ−Ring(T ) ⊂ F. 2
Theorem 1.10 (1) For A ⊂ X, F −1 F (A) ⊃ A and we have the equality for all A iff F is injective.
(2) For A0 ∈ X 0 , F F −1 (A0 ) ⊂ A0 and we have the equality for all A0 iff F is surjective.
(3) F −1 (∅) = ∅, F −1 (X 0 ) = X, F −1 (A0 ∪ B 0 ) = F −1 (A0 ) ∪ F −1 (B 0 ), F −1 (A0 \B 0 ) = F −1 (A0 )\F −1 (B 0 ),
0
Let F ∗ : 2X → 2X be the map given by (1.1). (We prefer not to denote it by F −1 to avoid ambiguous
notation).
0
For C 0 ⊂ 2X , we can write
F ∗ (C 0 ) = {F −1 (A0 ) : A0 ∈ C 0 }.
Let C ⊂ 2X . We will write
0
F∗ (C) := (F ∗ )−1 (C) = {A0 ∈ 2X : F −1 (A0 ) ∈ C}.
The following facts follow from Theorem 1.10 (1), (2) applied to F ∗ :
Theorem 1.11 (1) F∗ F ∗ (C 0 ) ⊃ C 0 ;
(2) F ∗ F∗ (C) ⊂ C.
5
1.7 Transport of σ-rings
Theorem 1.12 (1) If F 0 is a σ-ring over X 0 , then F ∗ (F 0 ) is a σ-ring over X.
(2) If F is a σ-ring over X, then F∗ (F) is a σ-ring over X 0 .
0
(3) If C 0 ⊂ 2X , then
F ∗ (σ−Ring(C 0 )) = σ−Ring(F ∗ (C 0 )).
Proof. To see (1) and (2), we use Theorem 1.10 (3), which says that F ∗ is a homomorphism for
set-theoretical operations.
Let us prove (3). By (1), F ∗ (σ−Ring(C 0 )) is a σ-ring. It contains F ∗ (C 0 ). Hence
F∗ (σ−Ring(F ∗ (C 0 ))) ⊃ F∗ (F ∗ (C 0 )) ⊃ C 0 .
2
For A ∈ 2X and C ⊂ 2X , we set
C := {A ∩ C : C ∈ C}.
A
Proof.
F ∗ (F 0 ) = F ∗ (σ−Ring(C 0 )) = σ−Ring(F ∗ (C 0 )) ⊂ σ−Ring(F) = F,
where we used Theorem 1.12 (3) in the second equality. 2
6
1.9 Measurable real functions
R ∪ {−∞, ∞} =: [−∞, ∞] is a topological space in the obvious way. We can extend the addition to
[−∞, ∞] except that ∞ − ∞ is undefined. We extend the multiplication to [−∞, ∞], adopting the
convention 0(±∞) = 0. Let Borel([−∞, ∞]) denote the σ-field of Borel subsets of [−∞, ∞], that is the
σ-field generated by open subsets of [−∞, ∞]. If Y ⊂ R, then Borel(Y ) will denote the σ-field in X
generated by open subsets in X. Note in particular that Borel([−∞, 0[∪]0, ∞]) is generated by the sets
[−∞, −α[ and ]α, ∞[ for 0 ≤ α.
Let (X, F) be a space with a σ-ring. We say that
f : X → [−∞, ∞]
is a F-measurable function iff for any A ∈ Borel([−∞, 0[∪]0, ∞]), f −1 (A) ∈ F. The set of such functions
will be denoted M(X, F), or for shortness, M(X). The set of measurable functions with values in [0, ∞]
will be called M+ (X).
Let A ⊂ X. Its characteristic function is denoted by
1, x ∈ A
1A (x) =
0, x ∈ X\A,
1A is F-measurable iff A ∈ F.
If f, g are real functions on X, we will write
{f ≥ g} := {x ∈ X : f (x) ≥ g(x)}.
Proof. (2) For simplicity, we assume in addition that F is a σ-field. Using the countability of Q we see
that [
{f + g > α} = {f > α + β} ∩ {g > −β}.
β∈Q
7
Proposition 1.19 Let f1 , f2 , · · · ∈ M(X). Then
are measurable. If there exists the pointwise limit of fn , then also limn→∞ fn is measurable.
Finally,
lim fn = lim sup fn = lim inf fn .
n→∞ n→∞ n→∞
Theorem 1.20 Let f : X → R. Then f ∈ M+ (X) iff there exists an increasing sequence un ∈ E+ (X)
such that
f = sup un
n∈N
1.10 Spaces L∞
Assume that I ⊂ F ⊂ 2X are rings. We say that say that I is an ideal in F if
A ∈ I, B ∈ F ⇒ A ∩ B ∈ I.
8
Theorem 1.22 (1) Given f ∈ M(X, F), we can always find N ∈ I such that sup |f | = kf k∞ ;
X\N
(2) kf + gk∞ ≤ kf k∞ + kgk∞ ;
(3) kαf k = |α|kf k∞ .
sup |f |X\N = kf k∞ ,
≤ sup(|f | + |g|)
X\(N ∪M )
Let
L∞ (X, F, I) := {f ∈ M(X, F) : kf k∞ < ∞},
Theorem 1.23 (Riesz-Fischer) Let (fn )n∈N be a sequence in L∞ (µ) satisfying the Cauchy condition,
that is for any > 0 there exists N such that for n, m ≥ N
kfn − fm k∞ ≤ .
Proof. There exists a subsequence (fnk )k∈N such that kfnk+1 − fnk k∞ ≤ 2−k , for any k. We set
∞
X
gk := fnk+1 − fnk , g := |gk |.
k=1
Then
∞
X ∞
X
kgk∞ ≤ kgk k∞ ≤ 2−k = 1.
k=1 k=1
∞
P∞
Hence g ∈ L and therefore g is finite outside of a set N in I. Hence the series k=1 gk is convergent
outside of N . This means that the sequence (fnk )k∈N is convergent to a function f outside of N . Inside
N we set f := 0. We check that f ∈ L∞ and kf − fn k∞ → 0. 2
9
2 Measure and integral
2.1 Contents
Let R be a ring. ν : R → [0, ∞] is a content if
(1) ν(∅) = 0;
(2) A1 , A2 ∈ R, A1 ∩ A2 = ∅, ⇒ ν(A1 ∪ A2 ) = ν(A1 ) + ν(A2 ).
Theorem 2.1 Let (X, R, ν) be a content on a ring. Then if A1 , A2 , · · · ∈ R are disjoint and A =
∞
∪ Ai ∈ R, then
i=1
∞
X
ν(A) ≥ ν(Ai ).
i=1
2.2 Measures
Let (X, F) be a space with a σ-ring. A function µ : F → [0, ∞] is called a measure if
(1) µ(∅) = 0,
P∞
(2) A1 , A2 · · · ∈ F, Ai ∩ Aj = ∅ for i 6= j ⇒ µ (∪∞
n=1 An ) = n=1 µ(An ).
The triple (X, F, µ) is called a space with a measure.
Proposition 2.2 (1) If A ⊂ B, then µ(A) ≤ µ(B).
P∞
(2) A1 , A2 , · · · ∈ F ⇒ µ((∪∞
n=1 An ) ≤ n=1 µ(An ).
(3) A1 , A2 , · · · ∈ F, An % A ⇒ µ(An ) % µ(A).
(4) If A1 , A2 , · · · ∈ F, An & A and for some n, µ(An ) < ∞, then µ(An ) & µ(A).
Definition 2.3 Let (X, F, µ) be a space with a measure. and P (x) be a property defined on X. We say
that P (x) is true µ-almost everywhere (µ-a.e.) if
µ ({x ∈ X : P (x) is not true }) = 0.
Fµf := {A ∈ F : A is µ-finite}.
Fµσf := {A ∈ F : A is µ − σ-finite}.
We say that µ is σ-finite iff F = Fµσf and µ is finite iff F = Fµf . We say that µ is probabilistic iff
µ(X) = 1.
10
Theorem 2.4 (1) Fµ0 is a σ-ring and an ideal in Fµf , Fµσf , F.
(2) Fµσf is a σ-ring and an ideal in F
(3) Fµf is a ring and an ideal in Fµσf , F. We have Fµσf = σ−Ring(Fµf ).
(4) If F is a σ-field, then µ is σ-finite iff X is µ-σ-finite; µ is finite if X is µ-finite.
Note that if (X, F, µ) is any measure, then X, F µf , µ µf ) is a σ-finite measure. We will call say that
F
the latter measure has been obtained from the former by restricting to µ − σ-finite sets.
satisfies
R
(1) 1A dµ = µ(A)
R R
(2) α ≥ 0 implies (αu)dµ = α udµ;
R R R
(3) (u + v)dµ = udµ + vdµ.
R R
(4) u ≤ v implies udµ ≤ udµ.
Lemma 2.6 Let (un )n∈N be an increasing sequence in E+ (X) and v ∈ E+ (X). Then
Z Z
v ≤ sup un ⇒ vdµ ≤ sup un dµ.
n∈N n∈N
An := {un ≥ v − } ∩ A.
11
But the lhs tends to (α − )∞ = ∞. Therefore,
Z
lim un dµ = ∞.
n→∞
2) µ(A) < ∞. Set Bn := A\An . Then Bn ∈ F, µ(Bn ) < ∞ and Bn & ∅. Thus µ(Bn ) & 0.
Adding v1An ≤ un + 1An and 1Bn v ≤ β1Bn we get
v ≤ 1An + β1Bn + un
Hence Z Z
vdµ ≤ µ(An ) + βµ(Bn ) + un dµ.
Lemma 2.7 Let (un )n∈N and (vn )n∈N be increasing sequences from E+ (X). Then
Z Z
sup un = sup vn ⇒ sup un dµ = sup vn dµ.
n∈N n∈N n∈N n∈N
Thus Z Z
sup vm dµ ≤ sup un dµ.
m∈N n∈N
2
satisfies
12
R
R un ∈ E+ (X) is an increasing sequence such that f = sup un (which always exists), then
(1) If un dµ →
f dµ.
R
(2) 1A dµ = µ(A);
R R
(3) λf dµ = λ f dµ, f ∈ M+ (X), λ ∈ [0, ∞[;
R R R
(4) (f + g)dµ = f dµ + gdµ;
(5) on E+ (X) it coincides with the previously defined integral.
R R
(6) if f, g ∈ M+ (X), f ≤ g, then f dµ ≤ gdµ.
Theorem 2.9 (Beppo Levi) Let (fn )n∈N be an increasing sequence from M+ (X). Then supn∈N fn ∈
M+ (X) and Z Z
sup fn dµ = sup fn dµ.
n∈N n∈N
Hence Z Z
sup fn dµ ≤ f dµ.
Using vn ≤ fn , we obtain Z Z
vn dµ ≤ fn dµ.
Hence Z Z
f dµ ≤ sup fn dµ.
n
2
Proof. Set f := lim inf n→∞ fn , gn := inf m≥n fm . We have f, gn ∈ M+ (X) and gn % f . Hence
Z Z
f dµ = sup gn dµ.
n∈N
13
gn ≤ fm for m ≥ n, therefore Z Z
gn dµ ≤ inf fm dµ.
m≥n
2
Proof. Set
M := {f 6= 0}.
⇒ Let
Mn := {f ≥ n−1 }.
Then Mn ∈ F and Z
n−1 µ(Mn ) ≤ f dµ = 0.
f dµ = 0. 2
R
Hence
14
2.9 Transport of a measure—the change of variables in an integral
If (X, F), (X 0 , F 0 ) are spaces with σ-rings, µ is a measure on (X, F) and T : X → X 0 is a measurable
transformation, then T∗ µ is the measure on (X 0 , F 0 ) defined as
T ∗ µ0 (A) := µ0 (T (A)), A ∈ F.
2.10 Integrability
Definition 2.15 Let f ∈ M(X). If
Z Z
f+ dµ < ∞, f− dµ < ∞,
|f |dµ < ∞, then we say that f ∈ L1 , or integrable (in the sense of L1 ) and we write
R
or equvalently, if
Z Z Z
1
f ∈ L (µ), f dµ := f+ dµ − f− dµ.
f = u − v,
then f ∈ L1 (µ).
Proof. We have
f ≤ u ≤ u + v, −f ≤ v ≤ u + v.
Therefore,
|f | ≤ u + v ∈ L1 (µ).
2
15
(3) sup(f, g), inf(f, g) ∈ L1 (µ);
R R
(4) f ≤ g ⇒ f dµ ≤ gdµ
R R
(5) | f dµ| ≤ |f |dµ.
16
We substitute
|f |p (x) |g|q (x)
a= , b = .
kf kpp kf kqq
We get
1 |f |p (x) 1 |g|q (x) |f |(x)|g|(x)
p + q ≥ .
p kf kp q kf kq kf kp kgkq
We integrate R
|f ||g|dµ
1≥ .
kf kp kgkq
The case p = 1, q = ∞ is straightforward. 2
α 1−α
Theorem 2.21 Let 1 ≤, r ≤ ∞, 0 ≤ α ≤ 1, q + r = p1 . Then
kf kp ≤ kf kα 1−α
q kkf kr ≤ αkf kq + (1 − α)kf kr .
Proof. Z Z
|f |p = |f |pα |f |p(1−α) ≤ k|f |pα k αp
q k|f |
p(1−α)
k (1−α)p
r .
Theorem 2.22 (The generalized Minkowski inequality) Let X, Y be spaces with measures µ and
ν, 1 ≤ p < ∞.
1 p1
Z Z p p Z Z
dν(y) f (x, y)dµ(x) ≤ dµ(x) |f |p (x, y)dν(y)
1 1
Proof. Let p + q = 1. It suffices to assume that f ≥ 0. We will restrict ourselves to the case p > 1.
R R p
dy f (x, y)dx
R R p−1 R
= dy
f (x1 , y)dx1 f (x2 , y)dx2
R R R p−1
= dx2 dy f (x1 , y)dx1 f (x2 , y)
R R R q(p−1) q1 R p 1
≤ dx2 dy1 f (x1 , y1 )dx1 f (x2 , y2 )dy2 p (the Hölder inequality)
R R p 1− p1 R R p 1
= dy1 f (x1 , y1 )dx1 dx2 f (x2 , y2 )dy2 p .
kf − fn kp → 0.
17
Proof. We define
limn→∞ fn (x) if limn→∞ fn (x) exists
f (x) :=
0 if limn→∞ fn (x) does not exists
h = lim (h − hn )
n→∞
Thus Z
lim sup hn dµ ≤ 0
n→∞
Using hn ≥ 0 we get Z
lim hn dµ = 0.
n→∞
2.13 Lp spaces
Theorem 2.26 (Riesz-Fischer) Let 1 ≤ p ≤ ∞. Let (fn )n∈N be a sequence in Lp (µ) satisfying the
Cauchy condition, that is for any > 0 there exists N such that for n, m ≥ N
kfn − fm kp ≤ .
18
Proof. There exists a subsequence (fnk )k∈N such that kfnk+1 − fnk kp ≤ 2−k , for any k. We set
∞
X
gk := fnk+1 − fnk , g := |gk |.
k=1
Then
∞
X ∞
X
kgkp ≤ kgk kp ≤ 2−k = 1.
k=1 k=1
p
P∞
Hence g ∈ L (µ) and therefore g is finite µ-a.e. Hence the series k=1 gk is µ-a.e. convergent. This
means that the sequence (fnk )k∈N is µ-a.e. convergent.
In the case p = ∞ it is sufficient to take the limit and to check that it is the limit in the L∞ sense.
In the case 1 ≤ p < ∞, we need to apply the Lebesgue theorem. We first check hat
limk→∞ kfnk − f kp = 0,
Proof. (4) Let f ∈ Lp (µ). Then f+ , f− ∈ Lp (µ). We know that there exist sequences u±,n ∈ E+ (µ)
with u±,n % f± . By the Lebesgue theorem, kf± − u±,n kp → 0. 2
19
2.14 Egorov theorem
Theorem 2.28 Let f, f1 , f2 , · · · ∈ M(X). Consider the following statements:
(1) fn (x) → f (x) for a.a. x ∈ X.
(2) For all > 0
∞ [
\ ∞
µ |fj − f | ≥ = 0.
n=1 j=n
(4) For every δ > 0, there exists A ∈ F with µ(A) < δ and
lim sup |f − fn | = 0.
n→∞ x∈X\A
Then (1)⇔(2)⇐(3)⇐(4). If µ(X) < ∞, then (1)⇔(2)⇔(3)⇔(4). (The implication (1)⇒(4) is called
the Egorov theorem).
Proof. Only (3)⇒(4) is not immediate, and we are going to prove this implication. Let δ > 0 and k ∈ N.
Then by (3) there exists nk such that for
∞
[
Bk := {|f − fj | > 1/k},
j=nk
∞
we have µ(Bk ) < δ2−k . Set A := ∪ Bk . We have µ(A) < δ and on X\A, |f (x) − fj (x)| ≤ 1/k for
k=1
j ≥ nk . Hence on X\A, fn converges uniformly to f . 2
3 Extension of a measure
3.1 Hereditary families
T ⊂ 2X . We say that T is hereditary if A ⊂ B ∈ T implies A ∈ T . For any T we denote by Her(T ) the
smallest hereditary family containing T .
= {A\N : A ∈ F, N ∈ Her(I)}.
20
3.2 Extension of a measure by null sets
Let X be a set and (F2 , µ2 ), (F1 , µ1 ) be measures on X. We say that (F2 , µ2 ) extends (F1 , µ1 ) by null
sets iff
(1) σ−Ring(F1 ∪ (F2 )0µ2 ) = F2 ;
(2) µ2 = µ1 .
F1
Theorem 3.4 Suppose that (F2 , µ2 ) extends (F1 , µ1 ) by null sets. Then
(1) F2 = {A ∪ N : A ∈ F1 , N ∈ (F2 )0µ2 )}.
(2) N (µ2 ) ∩ M(F1 ) = N (µ2 ). Hence, we can identify M(X, F1 )/N (µ1 ) with M(X, F2 )/N (µ2 ) in the
obvious way.
(3) We can identify Lp (µ1 ) with Lp (µ2 ).
21
The family of sets measurable wrt µ∗ will be denoted F ms . µ∗ restricted to F ms will be denoted µms .
Theorem 3.8 Let µ∗ be an external measure on X. Let F ms and µms be defined as above. Then
(1) F ms is a σ-field
(2) (X, F ms , µms ) is a complete measure;
(3)
A ∈ 2X , µ∗ (A) = 0 ⇔ A ∈ F ms , µms (A) = 0.
µ∗ (Q ∩ A) = µ∗ (Q ∩ A ∩ B) + µ∗ (Q ∩ A\B). (3.6)
Thus by (3.5)
Thus, by (3.8),
µ∗ (Q) = µ∗ (Q\(A\B)) + µ∗ (Q ∩ (A\B)). (3.12)
Hence A\B ∈ F ms .
Step 2. A, B ∈ F ms ⇒ B ∪ A ∈ F ms .
We have, applying the measurability condition to Q ∩ (A ∪ B) and A.
Hence
µ∗ (Q) = µ∗ (Q\(A ∪ B)) + µ∗ (Q ∩ (A ∪ B)). (3.15)
Therefore A ∪ B ∈ F ms . Thus we proved that F ms is a field.
22
Step 3.
A1 , A2 , · · · ∈ F ms ⇒ ∪∞
j=1 Aj =: A ∈ F
ms
. (3.16)
It suffices to assume that Aj are disjoint. For any n we have
Pn (3.17)
= j=1 µ∗ (Q ∩ Aj ) + µ∗ (Q\A)).
µ∗ (Q) = µ∗ (Q ∩ A) + µ∗ (Q\A)),
which shows A ∈ F ms .
Step 4. As a by-product we get
∞
X
µ∗ (Q ∩ A) = µ∗ (Q ∩ Aj ).
j=1
µ∗ (Q) ≥ µ∗ (Q ∩ A) + µ∗ (Q\A).
Hence A ∈ F ms and µms (A) = 0. This proves (3), which implies the completeness of the measure µms . 2
Then
(1) µ∗ is an external measure;
(2) µ∗ = µ on F.
(3) Let Fµms , µms be defined from µ∗ as in the previous subsection. Then Fµms is a σ-field containing F.
(4) In the definition of µ∗ we can replace F with Fµf , Fµσf , Fµcp , Fµms , etc., obtaining the same µ∗ .
23
Proof. (1) The properties (1) and (3) of the definition of an external measure are obvious. Let us show
the property (2).
Let Q1 , Q2 , · · · ∈ 2X . For any > 0 we will find a sequence A1 , A2 , · · · ∈ F such that
Qn ⊂ An ,
µ(An ) ≤ µ∗ (Qn ) + 2−n .
Then
∞
X ∞
X
µ∗ (∪∞ ∞
n=1 Qn ) ≤ µ(∪n=1 An ) ≤ µ(An ) ≤ µ∗ (Qn ) + .
n=1 n=1
Hence
∞
X
µ∗ (∪∞
n=1 Qn ) ≤ µ∗ (Qn ).
n=1
(2) is obvious.
Let us prove (3). Let B ∈ F, Q ∈ 2X . For any > 0 and suitable A ∈ F such that Q ⊂ A we have
µ∗ (Q) ≥ µ(A) −
µS (A) := µ∗ (A).
Theorem 3.11 Suppose (X, F, µ) and (X, Fµms , µms ) are as in Theorem 3.9. Then (Fµms , µms ) can be
obtained by applying consecutively to (F, µ) the following constructions:
1. restricting to σ-finite subsets,
2. completion
3. extending to localizable sets.
24
3.7 Sum-finite measures
Let (X, F, µ) be a measure. Define the set of locally µ-measurable sets by
Fµloc := {A ∈ 2X : A ∩ B ∈ F, B ∈ Fµσf }
= {A ∈ 2X : A ∩ B ∈ F, B ∈ Fµf }.
In other words, Fµloc is the family of all sets localizable in Fµσf (or in Fµf ). Clearly, Fµloc is a σ-field
containing F as an ideal.
A family {Xi : i ∈ I} of disjoint elements of Fµf such that ∪ Xi = X and
i∈I
X
µ(A) = µ(A ∩ Xi ), A ∈ F,
i∈I
Theorem 3.12 If {Xi : i ∈ I} and {Yj : j ∈ J} are localizing families for µ, then so is {Xi ∩ Yj :
(i, j) ∈ I × J}
Theorem 3.13 If (X, F, µ) is a measure possessing a localizing family {Xi : i ∈ I}, then if for
A ∈ Fµloc we set
X
µloc (A) = µ(A ∩ Xi ),
i∈I
A ⊂ B ⇔ B ⊃ A ⇔ A = A ∩ B.
If there exists an identity element for ∩, called X, then (R, ∆, ∅, ∩, X) is called a Boolean field.
Clearly, every ring/field in 2X is a Boolean ring/field.
In the obvious way we introduce the notion of Boolean σ-rings, Boolean σ-fields, etc. In what follows
we concentrate on σ-rings/fields.
If I ⊂ F are σ-rings and I is an ideal in F, then F/I is a Boolean σ-ring.
25
If F is a Boolean σ-ring, we can define the space M(F) as the set of all function
such that for α 6= β, F (α) ∩ F (β) = ∅. If F ⊂ 2X , then we identify f ∈ M(X, F) with F ∈ M(F) where
F (α) = f −1 ({α}).
If F is a Boolean σ-ring and F ∈ M(F), we set kF k∞ := sup{|α| : F (α) 6= ∅} and define the
space L∞ (F). If I ⊂ F ⊂ 2X are σ-rings and I is an ideal in F, then we can identify L∞ (X, F, I) with
L∞ (F/I).
Theorem 3.15 Let X be a set and F ⊂ F1 are σ-rings over X. Let (F, µ) and (F1 , µ1 ) be measures
such that
(1) µ = µ1 on F;
(2) σ−Ring F ∪ (F1 )0µ1 = F1 .
If F̃ and F̃1 are faithful measures defined as above, then they are isomorphic.
Dyn(C) ⊂ Ring(C).
K(B) := {A ∈ 2X : A ∩ B ∈ Dyn(C)}.
26
Note that
A ∈ K(B) ⇔ B ∈ K(A). (4.19)
Using the fact that Dyn(C) is a Dynkin class we check that K(B) is a Dynkin class.
Using the fact that C is ∩-stable we see that
B ∈ C ⇒ C ⊂ K(B) (4.20)
Hence,
B ∈ C ⇒ Dyn(C) ⊂ K(B) (4.21)
From (4.19) and (4.21) we get
A ∈ Dyn(C) ⇒ C ⊂ K(A) (4.22)
Hence
A ∈ Dyn(C) ⇒ Dyn(C) ⊂ K(A) (4.23)
Therefore, Dyn(C) is ∩-stable. Hence, by Theorem 4.1 it is a ring. 2
4.2 Semirings
Definition 4.3 T ⊂ 2X is called a semiring if
(1) A, B ∈ T ⇒ A ∩ B ∈ T ;
n
(2) A, B ∈ T ⇒ A\B = ∪ Ci , where Ci are disjoint elements of T
i=1
Theorem 4.4 Let T be a semiring. Then A ∈ Ring(T ) iff A is a disjoint union of elements of T .
Proof. Let R be the family of finite unions of disjoint elements of T . It is obvious that R ⊂ Ring(T ).
Let us prove the converse inclusion. To this end it is enough to prove that R is a ring.
n
Step 1. Let A ∈ R, B ∈ T . Then A = ∪ Ai with disjoint Ai ∈ T . Now
i=1
n
A\B = ∪ (Ai \B),
i=1
where Ai \B ∈ T are disjoint and each Ai \B is a finite union of disjoint elements of T . Hence A\B ∈ R.
n
Step 2. Let A ∈ R, B ∈ R. Then B = ∪ Bi with disjoint Bi ∈ T . Now
i=1
A ∪ B = (A\B) ∪ (A ∩ B) ∪ (B\A).
thus by Steps 2 and 3, the left hand side is a union of three disjoint elements of R. Therefore, it is a
union of a finite family of elements of T . Hence, A ∪ B ∈ R.
Thus we proved that R is a ring. 2
27
4.3 σ-Dynkin classes
We say that D is a σ-Dynkin class if
(1) A, B ∈ D, A ⊂ B ⇒ B\A ∈ D;
∞
(2) A1 , A2 , · · · ∈ D, Ai ∩ Aj = ∅, i 6= j, ⇒ ∪ Ai ∈ D
i=1
Mon(R) = σ−Ring(R).
Proof. Since a σ-ring is a monotone class and since R ⊂ σ − Ring(R), it follows that
Mon(R) ⊂ σ − Ring(R).
Note that
A ∈ K(B) ⇔ B ∈ K(A). (4.24)
We easily check that for every A ∈ 2X , K(A) is a monotone class. Clearly,
A ∈ R ⇒ R ⊂ K(A). (4.25)
Hence
A ∈ R ⇒ Mon(R) ⊂ K(A). (4.26)
From (4.24) and (4.26), we get
A ∈ Mon(R) ⇒ R ⊂ K(A).
28
Hence,
A ∈ Mon(R) ⇒ Mon(R) ⊂ K(A).
Therefore, Mon(R) is a ring. By Proposition 4.7 (2) it is a σ-ring. Hence,
Mon(R) ⊃ σ−Ring(R).
Theorem 4.9 Let T be a ∩-stable family containing ∅ and R = Ring(T ). Let ν1 , ν2 be finite contents
on R coinciding on T . Then ν1 = ν2 .
Theorem 4.10 Suppose that T is a semiring and ν is a content on T . Then there exists a unique
content on Ring(T ) extending ν.
n
Proof. Every A ∈ Ring(T ) can be written as A = ∪ Bi for some disjoint Bi ∈ T . Then we set
i=1
n
X
ν(A) := ν(Bi ).
i=1
n m
Suppose now that A = ∪ Bi = ∪ Cj are decompositions of the above type. Then
i=1 j=1
n m
A = ∪ ∪ Bi ∩ Cj
i=1 j=1
Therefore,
n
X n X
X m m
X
ν(Bi ) = ν(Bi ∩ Cj ) = ν(Cj ).
i=1 i=1 j=1 j=1
29
4.6 Uniqueness of a measure
Theorem 4.11 Let (X, F) be a set with a σ-ring. Let µ1 and µ2 be two measures defined on (X, F)
Suppose that T ⊂ F is a ∩-stable family such that µ1 = µ2 on T and is finite on T . Then µ1 = µ2 on
σ−Ring(T ).
Proof. Let W be the family of sets whose characteristic functions can be approximated in Lp (µ) by
linear combinations of characteristic functions of elements in T .
Step 1. Assume first that µ is finite. Clearly, W is then a σ-Dynkin class. Hence F ⊂ W.
Step 2. Let µ be arbitrary. Let A ∈ Fµf . Let {Xi : i ∈ I} be Pa∞localizing family for µ contained in
T . Then there exists a sequence i1 , i2 , · · · ∈ I such that µ(A) = j=1 µ(A ∩ Xij ). We apply Step 1. to
µ restricted to T . We conclude that A ∈ W. Hence Fµf ⊂ W Consequently, linear combinations of
Xij
characteristic functions of elements in T are dense in E ∩ Lp (µ).
Step 3. Let f ∈ Lp (µ). There exist sequences u± ± ± p
n ∈ E+ such that un % f± . Clearly, un ∈ L (µ),
+ − p + −
un −un is dominated by |f | ∈ L (µ), hence by the Lebesgue dominated convergence theorem un −un → f
in the Lp (µ) sense. 2
4.8 Premeasures
Definition 4.13 Let (X, R) be a set with a ring. A function ν : R → [0, ∞] is called a premeasure if
(1) ν(∅) = 0,
P∞
(2) A1 , A2 · · · ∈ R, ∪∞ ∞
n=1 An ∈ R, Ai ∩ Aj = ∅ for i 6= j ⇒ ν (∪n=1 An ) = n=1 ν(An ).
30
∞
Theorem 4.14 Let (X, R, ν) be a premeasure, A1 , A2 , · · · ∈ R , A ∈ R and A ⊂ ∪ Ai . Then
i=1
∞
X
ν(A) ≤ ν(Ai ).
i=1
n−1 ∞
Proof. Bn := (An \ ∪ Ai ) ∩ A are disjoint elements of R, Bi ⊂ Ai and A = ∪ Bi . Hence
i=1 i=1
∞
X ∞
X
ν(A) = ν(Bi ) ≤ ν(Ai ).
i=1 i=1
Then
(1) µ∗ is an external measure;
(2) ν = µ∗ on R;
(3) Let F ms be the σ-field of µ∗ -measurable sets and µms the corresponding measure. Then (X, F ms , µms )
is a complete measure extending the premeasure (X, R, ν).
(4) Let F := σ−Ring(R). Let the restriction of µ∗ to F be denoted µ. Then for Q ∈ 2X
Proof. (1) The properties (1) and (3) of the definition of an external measure are obvious. Let us show
the property (2). Let Q1 , Q2 , · · · ∈ 2X . For any > 0 we will find a double sequence (Anm )m∈N such that
Qn ⊂ ∪∞
m=1 Anm ,
∞
X
µ(Anm ) ≤ µ∗ (Qn ) + 2−n .
m=1
Then
∞
X ∞
X
µ∗ (∪∞
n=1 Qn ) ≤ µ(Anm ) ≤ µ∗ (Qn ) + .
n,m=1 n=1
Hence
∞
X
µ∗ (∪∞
n=1 Qn ) ≤ µ∗ (Qn ).
n=1
∗
(2) It is obvious that µ (A) ≤ µ(A). The converse inequality follows by Theorem 4.14
31
∞
(3) Let A ∈ R, Q ∈ 2X . For any > 0 and suitable A1 , A2 , · · · ∈ R such that Q ⊂ ∪ Ai , we have
i=1
P∞
µ∗ (Q) ≥ j=1 µ(Aj ) −
P∞ P∞
= j=1 µ(Aj ∩ A) + j=1 µ(Aj \A) −
≥ µ∗ (Q ∩ A) + µ∗ (Q\A) − .
Therefore,
µ∗ (Q) ≥ µ∗ (Q ∩ A) + µ∗ (Q\A).
Thus, A ∈ F ms . 2
T1 ∗ T2 := {A1 × A2 : Ai ∈ Ti , i = 1, 2}.
Then T1 ∗ T2 is a semiring.
Hence
W := {B ⊂ X1 × X2 : π1x2 (B) ∈ F1 }
is a σ-ring. Clearly W contains F1 ∗ F2 . Hence F1 ⊗ F2 ⊂ W. 2
32
Proof. Set
sB (x1 ) := µ2 (π2x1 (B)).
Set
W := {B ⊂ X1 × X2 : sB is measurable }.
Step 1. Assume that µ(X2 ) < ∞. Clearly, F1 ∗ Fµσf2 ⊂ W. If A, B ∈ W with A ⊂ B, then sB\A =
P∞
sB − sA , Hence B\A ∈ W. Let B1 , B2 , · · · ∈ W be disjoint and B = ∪∞ B
j=1 j . Then s B = j=1 sBj .
∞
Hence ∪ Bj ∈ W. Therefore, W is a σ-Dynkin class. Hence it contains σ−Ring(F1 ∗ Fµσf2 ).
j=1
Alternative versionPof Step 1. Assume that µ(X2 ) < ∞. If disjoint B1 , B2 , · · · ⊂ W and B =
∞
∪∞
j=1 Bj , then s ∞
∪ B
= σf
j=1 sBj . We know that Ring(F1 ∗ Fµ2 ) are disjoint unions of elements in
j
j=1
F1 ∗ Fµσf2
⊂ W. Hence Ring(F1 ∗ Fµσf2 ) ⊂ W.
Clearly, if A1 , A2 , · · · ∈ W and An % A, then sAn % sA . Hence, A ∈ W.
Likewise, if A1 , A2 , · · · ∈ W and An & A, then sAn & sA . Hence, using the finiteness of X, A ∈ W.
Therefore, W is a monotone class. Hence it contains σ−Ring(F1 ∗ Fµσf2 ) = F1 ⊗ Fµσf2 ).
Step 2. Now drop the assumption µ2 (X2 ) < ∞. Let B ∈ F1 ⊗ Fµσf2 . Then there exists a disjoint
∞
family A1 , A2 , · · · ∈ F2 such that µ2 (Ai ) < ∞ and B ⊂ X1 × ∪ Aj . Set Bj := B ∩ X1 × Aj . Then
j=1
∞
X
sB = sBj ,
j=1
Proof. The formula (5.27) is well defined by Proposition 5.4. Then we check that it is a measure.
The uniqueness follows by Theorem 4.11, because both measures coincide on the semiring Fµ1 ∗ Fµ2 .
2
33
5.3 Multiple integrals
For any x2 ∈ X2 , the function
X1 3 x1 7→ (x1 , x2 ) ∈ X1 × X2
is measurable. Hence if f ∈ M(X1 × X2 ), then
X1 3 x1 7→ f (x1 , x2 ) ∈ [−∞, ∞]
belongs to M(X1 ).
Theorem 5.6 Let (Xi , Fi , µi ) be spaces with measures. Let f ∈ M+ (X1 × X2 ).
(1) R
x1 7→ f (x1 , x2 )dµ2 (x2 )
belongs to M+ (X1 ),
(2) R R R R R
f d(µ1 ⊗ µ2 ) = ( f (x1 , x2 )dµ2 (x2 ))dµ1 (x1 ) = ( f (x1 , x2 )dµ1 (x1 ))dµ2 (x2 ). (5.29)
Proof. For elementary functions the theorem is obvious. For an arbitrary function from M+ (X1 × X2 )
we use the monotone convergence. 2
f+ (x1 , x2 )dµ2 (x2 ) belongs to L1 (µ1 ). Hence, by Theorem 2.12, for µ1 -a.a. x1 ,
R
Thus x1 7→
Z
f+ (x1 , x2 )dµ2 (x2 ) < ∞.
Theorem 5.8 If ⊗ denotes the tensor product in the sense of Hilbert spaces, then L2 (µ1 ) ⊗ L2 (µ2 ) =
L2 (µ1 ⊗ µ2 ).
34
5.4 Layer-cake representation
Let (X, F, µ) be a measure. Let ν be a Borel measure on [0, ∞[ and f ∈ M+ (X). For a, t ≥ 0, set
Z t
φ(t) := dν(s), u(a) := µ{f > a}.
0
Proposition 5.9 Z Z ∞
φ(f (x))dµ(x) = u(a)dν(a).
0
Proof.
R∞ R R
0
u(a)dν(a) = 1{f >a} (x)dµ(x) dν(a)
R R R R f (x) R
= 1{f >a} (x)dν(a) dµ(x) = 0
dν(a) dµ(x) = φ(f (x))dµ(x).
2
R∞
Corollary 5.10 (1) |f (x)|p dµ(x) = p 0 u(a)ap−1 da.
R
R
(2) f (x) = 1{f >a} (x)da.
6 Measures in Rn
6.1 Regular contents
Suppose that X is a topological space. Let R be a ring over X and ν a content on R. We say that ν is
regular iff for F ∈ R the following two conditions hold:
= inf{ν(H) : F ⊂ H o }.
∞
X
ν(F ) ≥ ν(Fj ).
j=1
Let us show the converse inequality. Let > 0. For any j = 1, . . . , n we can find Hj ∈ R such that
Fj ⊂ Hjo and ν(Hj \Fj ) < 2−j−1 . Likewise, we can find G ∈ R such that Gcl ⊂ F , Gcl is compact and
ν(F \G) < /2. Thus {Hjo : j = 1, 2, . . . } is an open cover of the compact set Gcl . We can choose a
finite subcover {Hjok : k = 1, . . . , m}, so that
m
Gcl ⊂ ∪ Hjok .
k=1
35
Consequently,
m
G ⊂ ∪ Hjk .
k=1
Thus
m
X ∞
X
ν(F ) ≤ ν(G) + /2 ≤ ν(Hjk ) + /2 ≤ ν(Fj ) − .
k=1 j=1
2
Assume now in addition that f : R → R is continuous from the right, that means
Lemma 6.4 Let F ∈ Ring(T ) and > 0. Then there exist H, G ∈ Ring(T ) such that
2
By Theorem 6.1, we get:
36
6.4 Borel measures on R
Theorem 6.6 (1) Let f : R → R be an increasing function continuous from the right. Then there
exists a unique measure µf on (R, Borel(R)) such that
Proof. The premeasure νf can be extended by the Caratheodory construction to a σ-field containing
Borel(R). 2
Definition 6.7 The measure on Borel(R) with the distribution function f (x) = x is called the Borel-
Lebesgue measure, and denoted λ. Its complete extension is called the Lebesgue measure. In integrals, if
the generic variable in R will be denoted by x, then instead of dλ(x) we will usually write dx.
Theorem 6.8 The Borel-Lebesgue measure is the only measure on Borel(R) invariant wrt translations
such that µ(]0, 1]) = 1.
Proof. Let (R, Borel(R), µ) be translation invariant. This means µ(]a, b]) = µ(]a + x, b + x]). Using this
and µ(]0, 1]) we get µ(]k/n, (k + 1)/n]) = 1/n. This easily implies µ(]a, b]) = b − a for any a ≤ b. 2
Theorem 6.9 (1) Let µ be a measure on 2R invariant wrt translations (that means if τt (x) = x − t,
then τt∗ µ = µ). Suppose that µ is finite on compact sets. Then µ = 0.
(2) Let T := R/Z denote the 1-dimensional torus. Let µ be a finite measure on 2T invariant wrt
translations. Then µ = 0.
x ∼ y ⇔ x − y ∈ Q.
Let I be the set of equivalence classes. Choose from every class a representant xi . We have
[
[0, 1[= Ay
y∈Q
for
Ay := {xi + y : i ∈ I}.
But τy2 −y1 Ay2 = Ay1 . Hence
X X
µ(T) = µ(Ay ) = µ(A0 ) = ∞µ(A0 ).
y∈Q y∈Q
37
6.5 The Cantor set and devil’s staircase
Definition 6.10 Let q ∈ N. To every sequence of numbers (pj )j∈N with values in {0, 1, . . . , q − 1} we
assign a number from the interval [0, 1]:
∞
X pj
0.p1 p2 · · · := = x.
j=1
qj
Note that every x ∈ [0, 1] has such a representation. It is ambiguous only if for some n we have
q n x ∈ N. Then
0.p1 . . . pn−1 pn 00 · · · = 0.p1 p2 . . . pn−1 (pn − 1)(q − 1)(q − 1) . . . .
Definition 6.11 The Cantor set C, is the subset of [0, 1] consisting of the numbers that in the tri-
1 2
nary system have only 0 and 2. It can
1 2 7 8
T∞be defined also as follows: C0 = [0, 1], C1 = C0 \] 3 , 3 [,
C2 = C1 \] 9 , 9 [∪] 9 , 9 [, etc. We set C = n=1 Cn .
It is a closed set with an empty interior, uncountable and has zero Lebesgue measure (because λ(Cn ) =
2n
3n ).
Definition 6.12 Define the transformation, called devil’s staircase, F : [0, 1] → [0, 1] as follows. If
x = 0.p1 p2 · · · ∈ C in the trinary system, where pi ∈ {0, 2}, then F (x) = 0. p21 p22 . . . in the binary system.
If x ∈ [0, 1]\C, then x ∈]x− , x+ [ where x− = 0.p1 . . . pn 022 . . . and x+ = 0.p1 . . . pn 200 . . . . We see that
F (x− ) = F (x+ ) and we set F (x) = F (x− ) = F (x+ )
The function F is increasing, continuous, locally constant beyond C, and F (1) − F (0) = 1. It defines
a Borel measure µ, which is continuous and singular wrt the Lebesgue measure, since µ([0, 1]\C) = 0.
g ∗ λ(]α, β]) = λ(g −1 (]α, β]) = λ(]f (α), f (β)]) = f (β) − f (α).
Theorem 6.13 (1) The n-dimensional Borel-Lebesgue measure λn is the unique measure on Borel(Rn )
such that n n
λn × ]ai , bi ] = Π |bi − ai |.
i=1 i=1
(2) It is also the unique translation invariant measure on Borel(Rn ) such that
n
λn × ]0, 1] = 1.
i=1
38
We can also consider its completion, called the n-dimensional Lebesgue measure. There are several
equivalent ways to construct the Lebesgue measure, described in the following theorem.
n
Then ν is a premeasure. We consider the measure obtained by the Caratheodory construction.
(y 1 , . . . , y n ) 7→ (x1 , . . . , xn ) = (f (y 1 , . . . , y n ), y 2 , . . . , y n ),
∂f (y 1 ,x2 ,...,xn )
dxn · · · dx2 dy 1 F (f (y 1 , x2 , . . . , xn ), x2 , . . . , xn )
R R R
= ∂y 1
∂f (y 1 ,y 2 ,...,y n )
dy n · · · dy 2 dy 1 F (f (y 1 , y 2 , . . . , xn ), y 2 , . . . , y n )
R R R
= ∂y 1
39
Step 4. We proceed by induction wrt n. We assume that the theorem is true for n − 1.
If F ∈ M+ (U ), then we can find a sequence of functions Fn ∈ M+ (U ) of compact support with
Fn % F .
Therefore, it is sufficient to assume that the support of F is compact.
∂xi
Let a = (a1 , . . . , an ) ∈ suppF . There exist i, j such that ∂y j (a) 6= 0. We can find δ > 0 such that for
i 1 n i 1 n
|y i − ai | < δ, i = 1, . . . , n, ∂x (y∂y,...,y
j
)
> 0, or ∂x (y∂y,...,y
j
)
< 0. Set Wa := [a1 − δ, a1 + δ] × · · · × [an −
n n
δ, a + δ]. Clearly, we can Pnfind a finite family of a1 , . . . , an ∈ R such that Wa1 , . . . , Wan covers suppF .
Then we can write F = i=1 Fi with suppFi ⊂ Wi .
i 1 n
In what follows we assume that on the support of F , ∂x (y∂y,...,y j
)
> 0. By Step 2, we can assume that
i = j = 1. Define
ψ
(y 1 , . . . , y n ) 7→(z 1 , . . . , z n ),
where
z 1 (y 1 , . . . , y n ) = x1 (y 1 , . . . , y n ), z 2 = y 2 , . . . , z n = y n .
The map ψ is injective. Define ρ := φψ −1 , that is
ρ
(z 1 , . . . , z n ) 7→(x1 , . . . , xn ).
Note that x1 = z 1 .
The map ψ is of the type considered in Step 3. Hence it satisfies the change of variables formula. We
have φ = ρψ. By Step 1, it is thus sufficient to prove that ρ satisfies the change of variables formula.
We have
1 0 ... 0
∂x12 ∂x22 . . . ∂xn2
∂x
ρ0 = = ∂z ∂z ∂z ,
∂z ...
∂xn ∂xn ∂xn
∂z 1 ∂z 2 . . . ∂zn
and hence
∂x2 ∂x2
∂x ∂z 2 ... ∂z n
det = det ...
∂z ∂xn ∂xn
∂z 2 ... ∂z n
Thus
R 1
dx · · · dxn F (x1 , . . . , xn )
R
∂x2 ∂x2
∂z 2 ... ∂z n
1
dz 2 · · · dz n F (x1 , x2 (x1 , z 2 , . . . , z n ), . . . , xn (x1 , z 2 , . . . , z n )) det
R R R
= dx ...
∂xn ∂xn
∂z 2 2 ... ∂z n
∂x2
∂x
∂z 2 ... ∂z n
1 2 n 1 1 n n 1 n
R R R
= dz dz · · · dz F (x (z , . . . , z ), . . . , x (z , . . . , z )) det ...
∂xn ∂xn
∂z 2 ... ∂z n
F (x(z)) det ∂x(z)
R
= ∂z dz.
40
7 Charges and the Radon-Nikodym theorem
7.1 Extension of a measure from a σ-ring
Let F, I be σ-rings over X and let I be an ideal in F. Let (X, I, µ) be a space with a measure. We can
then extend the measure µ from I to F. We can do this in many ways.
Then µmax is a measure. We have µmax (A) = µ(A), A ∈ I, and µmax (A) = ∞, A ∈ F\I. µmax is
the largest measure on F extending µ onto F. σ-finite and null sets coincide for µ and µmax .
(2) Define µmin : F → [0, ∞] by
Then µmin is a measure. We have µmin (A) = µ(A), A ∈ I. µmin is the smallest measure extending
µ onto F.
Hence
∞
X
min
µ (A) ≤ µmin (Aj ).
j=1
∞
X ∞
µ(Bj ) = µ( ∪ Bj ) ≤ µmin (A).
j=1
j=1
Thus
∞
X
µmin (Aj ) ≤ µmin (A).
j=1
2
41
(More generally, if ν is a charge, we say it is I-continuous if (7.32) is true).
In particular, if (X, F, µ) is also a space with a measure, then
Fµ0 := {A ∈ F : µ(A) = 0}
(2) If ν is σ-finite, then the decomposition of ν into a I-singular and a I-continuous measure is unique.
(3) If X is ν − σ-finite, then there exists a set N ∈ I such that
Proof. The fact that νIs is a measure follows from Theorem 7.1 applied to ν .
I
∞
We need to show that νIc is a measure. Let A1 , A2 , · · · ∈ F be disjoint and A = ∪ Aj . Let us prove
j=1
that
∞
X
νIc (A) ≤ νIc (Aj ). (7.34)
j=1
It is sufficient to assume that νIc (Aj ) < ∞, j = 1, 2 . . . . Let > 0. We will find Bj ∈ I, Bj ⊂ Aj with
νIc (Aj ) > ν(Aj \Bj ) − 2−j . Then
∞
X ∞
X
νIc (A) ≤ ν(A\ ∪∞
j=1 Bj ) = ν(Aj \Bj ) ≤ νIc (Aj ) + .
j=1 j=1
It is sufficient to assume that νIc (A) < ∞. Let > 0. We will find B ∈ I, B ⊂ A with νIc (A) >
ν(A\B) − . Then
X∞ ∞
X
νIc (Aj ) ≤ ν(Aj \B) = ν(A\B) < νIc (A) + .
j=1 j=1
α := sup {ν(A) : A ∈ I} .
42
Then α < ∞. We can find a sequence (Nj )j∈N in I such that limj→∞ ν(Nj ) = α. We can assume that
the sequence Nj % N . Then N ∈ I and ν(N ) = α.
It is obvious that ν(A ∩ N ) ≤ νIs (A). Suppose that for some A ∈ F,
Then B ∪ N ∈ I and
which is a contradiction
If X is ν − σ-finite, then we can find a sequence Xn % X such that ν(Xn ) < ∞. We will also find
sets Nn ⊂ Xn constructed as above. We easily check that ν(A ∩ N ) = νIs (A).
The decomposition of ν is uniquely determined on σ-finite sets. Hence it is unique. 2
{{x} : x ∈ X} ⊂ F. (7.36)
ν is continuous (diffuse) if
ν({x}) = 0, x ∈ X.
The I-continuous part does not have to be unique, but there is a canonical choice given by
(2) If ν is σ-finite, then the decomposition of ν into a point and a continuous measure is unique.
(3) If X is ν − σ-finite, there exists a countable set N ∈ F such that
43
Corollary 7.5 Let (X, F) be a set with a σ-field Let
{{x} : x ∈ X} ⊂ F
Let ν, µ be measures on (X, F) and let µ be continuous. Then there exists a decomposition
ν = νp + νsc + νac
such that
νp is pure point ,
νac is µ-continuous.
If ν is σ-finite, then the decomposition is unique.
Proof. (1) µ(B) = µ(A) + µ(B\A) and µ(B\A) > −∞. Hence µ(A) = µ(B) − µ(B\A) with both
summands less than ∞.
(2) We group the sets Ai into two subfamilies: those with a positive charge and a nonpositive charge.
After renumbering we can call the former B1 , B2 , . . . and the latter C1 , C2 , . . . . We have
∞
X
− µ(Bn ) = −µ (∪∞
n=1 Bn ) < ∞
n=1
and
∞
X
µ(Cn ) = µ (∪∞ ∞ ∞
n=1 Cn ) = µ (∪n=1 An ) − µ (∪n=1 Bn ) < ∞.
n=1
2
Theorem 7.7 Fµ± are σ-rings and ideals of F. ±µ restricted to Fµ± are measures.
44
For A ∈ F, we set
µ+ (A) := sup{µ(B) : B ⊂ A, B ∈ Fµ+ },
(9)
µ+ (A) := sup{µ(B) : B ⊂ A, B ∈ F},
Proof. (1) and (2) follow immediately from Theorem 7.1 (2).
Let β := µ− (X). Then there exist negative E1 , E2 , . . . such that µ− (En ) → β. Since negative sets
n
form a σ-ring, E := ∪ Ej ∈ Fµ− . Clearly, µ− (E) ≤ µ− (En ). Hence β = µ− (E). This implies (4) and
j=1
(3). A similar argument yields (5).
We interrupt the proof. 2
Lemma 7.9 Suppose that µ(X) < µ+ (X). Then there exists B ∈ Fµ− with µ(B) < 0.
Proof. If µ+ (X) = 0, then X ∈ Fµ− and µ(X) < 0. We can thus set B := X
The condition µ(X) < ∞ implies that µ+ (X) < ∞.
Let µ+ (X) > 0. We can find q < 1 such that
We can find E ∈ Fµ+ such that µ(E) ≥ qµ+ (X). Set X1 := X\E. Then
45
Hence,
µ+ (Xn ) ≤ (1 − q)n µ+ (X0 ).
Moreover, Pn
µ(Xn+1 ) = µ(X1 )\ j=1 µ(Ej ) ≤ µ(X1 ) < 0.
Set B := ∩ Xj . Then
j
µ+ (B) = lim µ+ (Xj ) = 0,
j→∞
is a measure. If f ∈ M(X) and f− ∈ L1 (µ), then (7.37) is a bounded from below charge.
Definition 7.12 The measure ν is called the measure with the density f wrt the measure µ and is denoted
dν
ν = f µ. We will also write f := dµ .
46
Theorem 7.13 (1) For f, g ∈ M+ (X) we have
f = g µ-a.e. ⇒ f µ = gµ.
h := g1N − f 1N .
Clearly,
R f 1N ≤ f and g1N ≤ g. Hence, f 1N ∈ L1 (µ), g1N ∈ L1 (µ). Therefore, h ∈ L1 (µ). Besides,
hdµ = 0 and h ≥ 0. Thus h = 0 µ-a.e. But N = {h > 0}. Hence µ(N ) = 0.
Assume now that µ is sum-finite. Let Xi be a localizing family. Then f µ = gµ restricted to Xi . Hence
fi = gi on Xi almost everywhere wrt the measure µ restricted to Xi . This implies that f = g µ-a.e. 2
Recall that the charge ν is called continuous wrt µ (or µ-continuous), if
µ(N ) = 0 ⇒ ν(N ) = 0, N ∈ F.
Theorem 7.14 (Radon-Nikodym) Let µ be a sum-finite measure on (X, F) and let ν be a charge.
Then the following conditions are equivalent:
(1) there exists f ∈ M(X) such that ν = f µ and f− ∈ L1 (µ);
(2) ν is µ-continuous.
G := {g ∈ M+ (F) : gµ ≤ ν}.
Let R
γ := sup{ gdµ : g ∈ G}. (7.38)
Then γ ≤ ν(X) < ∞. We can find gn0 ∈ G such that
limn→∞ gn0 dµ = γ.
R
Let
f := sup(gn0 )n∈N .
We claim that Z
f ∈ G and gdµ = γ. (7.39)
47
In fact, we have
gn := sup(g10 , . . . , gn0 ) ∈ G, gn % f,
which immediately implies (7.39).
Suppose now that
(f µ)(X) < ν(X). (7.40)
Using µ(X) < ∞, we can find β > 0 such that
Set
ρ(A) := ν(A) − (f µ)(A) − βµ(A), A ∈ F.
ρ is a bounded µ-continuous charge satisfying ρ(X) > 0. By Lemma 7.9, we can find a ρ-positive set
E ∈ F such that ρ(E) > 0. Recall that ρ-positivity of E means that
A ∈ 2E ∩ F ⇒ ρ(A) ≥ 0.
Hence f0 = f + β1E ∈ G.
Note that the µ-continuity of the charge ρ and ρ(E) > 0 implies µ(E) > 0. Hence
Z Z
f0 dµ = f dµ + βµ(E) = γ + βµ(E) > γ,
or µ(A) = ν(A) = 0,
In fact, if A ⊂ X1 , µ(A) > 0 and ν(A) < ∞, then An ∪ A ∈ Fνf and µ(An ∪ A) % α + µ(A), which means
µ(A) = 0. We apply Step 2. to X0 and on X1 we put ν = ∞µ.
Step 4. ν is arbitrary and µ sum-finite. We decompose X into a union of disjoint sets with a finite
measure µ and use Step 3. 2
Then
(1) vg ∈ (Lp (µ))# and kvg k = kgkq . Thus
is an isometry.
48
(2) Z
f˜ := f |f |p−2 /( |f |p )1/q (7.43)
Proof. (1) Using the Hölder inequality we check that if g ∈ Lq (µ), then vg ∈ Lp (µ)# and kvg k ≤ kgkq .
Setting Z
g̃ := g|g| /( |g|q )1/p ,
q−2
(7.44)
F 3 A 7→ hv|1A i
is a µ-continuous finite charge. By the Radon-Nikodym Theorem, there exists g ∈ M(X) such that
Z
hv|1A i = g1A dµ,
Assume that g 6∈ Lq (µ) Clearly, g+ 6∈ Lq (µ) or g− 6∈ Lq (µ). Hence it is sufficient toRassume that g ≥ 0.
We can find gn ∈ E+ (X) such that gn % g. Clearly, kgn kq → ∞. Set g̃n := gnq−1 /( |gn |q )1/p . Clearly,
g̃n ∈ Lp (µ), kg̃n k = 1 and
Z Z
hv|g̃n i = ggn dµ/( |gn |q )1/p ≥ kgn kq → ∞.
q−1
Theorem 8.2 Let X be a metrizable space. Then every closed set is δ-open.
49
Proof. Let C be closed. Define
Proof. We have
∞
[
−1
f (]a, ∞[) = f −1 ([a + 1/n, ∞[) (8.45)
n=1
be open σ-closed and An let be closed. We can then find fn ∈ C(X) such that 0 ≤ f ≤ 1, fn = 1 on An
and fn = 0 outside A. We define
X∞
f := 2−n fn .
n=1
Then
A = f −1 (]0, ∞[).
2
Definition 8.4 Let X be a topological space. Then the σ−field of Baire sets of 1st kind, denoted
Baire1 (X), is the smallest σ−field such that all elements of C(X, R) are measurable.
Theorem 8.5 Let X be normal, A ⊂ B ⊂ X, A be closed and B open. Then there exist A0 ∈
Closed ∩ Baire1 (X) and B0 ∈ Open ∩ Baire1 (X) such that A ⊂ B0 ⊂ A0 ⊂ B.
Theorem 8.6 (1) In any topological space, Baire1 (X) is generated e.g. by
50
(2) If X is normal, then Baire1 (X) is generated by
Open ∩ σ−Closed(X),
Closed ∩ δ−Open(X).
is a semimetric on X.
Let (X̃, d)˜ be the reduced metric space and T : X → X̃ the corresponding reduction. Clearly, x ∈ Cn ,
d(x, y) = 0 imply y ∈ Cn . Therefore, T ∗ 2X̃ contains C1 , C2 , . . . .
T ∗ 2X̃ is a σ-ring. Hence
T ∗ 2X̃ ⊃ σ−Ring(C1 , C2 , . . . ).
Thus there exists C̃ ∈ 2X̃ with C = T −1 C̃. But C is compact and T continuous. Therefore, C̃ is compact
as well. Thus C̃ is a closed subset of a metric space, and hence it is σ-open. Hence there exist open
Ũ1 , Ũ2 , . . . in X̃ with Ũn & C̃. Now T −1 Ũn are open in X and T −1 Ũn & C. 2
Definition 8.8 σ-field of Borel sets of 1st kind, denoted Borel1 (X), is the σ−field generated by open
sets.
Q
Example 8.10 Let I be uncountable and Xi be sets of at least two elements. Let X = Xi . Then
i∈I
one-element sets in X are closed (hence Borel) but not δ-open
Q (hence not Baire). In fact, let x ∈ Y ⊂ X
and Y be δ-open. Then Y contains a subset of the form Yi with Yi = Xi for all but a countable number
i∈I
of i ∈ I.
Definition 8.12 Let X be a topological space. Then the σ-ring of Baire sets of 2nd kind, denoted
Baire2 (X), is the smallest σ−ring such that all elements of Cc (X) are measurable.
51
Lemma 8.13 Let X be Tikhonov, A ⊂ B ⊂ X, A be compact and B open. Then there exist A0 ∈
Compact ∩ Baire2 (X) and B0 ∈ Open ∩ Baire2 (X) such that A ⊂ B0 ⊂ A0 ⊂ B.
{f −1 (U ) : U ∈ Open(R\{0}), f ∈ Cc (X)};
Open ∩ σ−Compact(X),
Compact ∩ δ−Open(X).
Definition 8.16 σ-ring of Borel sets of 2nd kind, denoted Borel2 (X), is the σ−ring generated by compact
sets.
Theorem 8.18 For σ-compact spaces Baire1 (X) = Baire2 (X) and Borel1 (X) = Borel2 (X).
In what follows, we will consider σ-rings of sets only on locally compact Hausdorff spaces. We will
use the σ-rings Baire2 (X) and Borel2 (X). We will call them simply Baire and Borel σ-rings and denote
by Baire(X) and Borel(X) respectively. For σ-compact spaces they are in fact σ-fields and coincide with
Baire1 (X) and Borel1 (X) respectively.
f ∈ C(X), f ≥ 0 ⇒ λ(f ) ≥ 0.
52
(2) If U ∈ Open ∩ Baire(X), then
R
ν(U ) = sup{ f dν : f ∈ C(X), suppf ⊂ U, 0 ≤ f ≤ 1}.
Theorem 8.20 (Riesz-Markov) Let λ be a positive linear functional on C(X). Then there exists a
unique Baire measure ν satisfying
Z
λ(f ) = f dν, f ∈ C(X). (8.47)
The proof of Theorem 8.20 will be split into a number of steps. Let us assume that we are given a
positive functional λ.
Proof. By Theorem 8.19 (1), ν is uniquely determined by λ on Closed ∩ Baire(X). This is a ∩-stable
family that generates Baire(X). Hence ν is uniquely determined. 2
For U ∈ Open ∩ Baire(X) we set
If the sum on the right is infinite, the inequality is obvious. Assume it is finite. We will find Uj ∈
Open ∩ Baire(X) with Aj ⊂ Uj and
53
By the compactness of suppf , for some n,
n
[
suppf ⊂ Uj .
j=1
Pn
We will find h1 , . . . , hn ∈ C(X) such that hj ≥ 0, supphj ⊂ Uj and j=1 hj = 1 on suppf . Hence,
S∞ S∞
ν ∗ ( n=1 An ) ≤ ν ∗ ( n=1 Un )
Pn
≤ λ(f ) + = j=1 λ(f hj ) +
P∞
≤ n=1 ν ∗ (Un ) +
P∞ P∞ P∞
≤ n=1 ν ∗ (An ) + n=1 2−n + = n=1 ν ∗ (An ) + 2.
2
Proof. It suffices to show that open Baire sets are ν ∗ -measurable. Let U be open and Q ∈ 2X .
Let A ∈ Open ∩ Baire(X) such that Q ⊂ A and ν ∗ (A) ≤ ν ∗ (Q) + . Consider the set A1 := A ∩ U ∈
Open ∩ Baire(X). Let f1 ∈ C(A) such that 0 ≤ f1 ≤ 1, suppf1 ⊂ A1 and
ν ∗ (A1 ) ≤ λ(f1 ) + .
Next consider set A2 := A\suppf1 ∈ Open. We will find f2 ∈ C(X) such that 0 ≤ f2 ≤ 1, suppf2 ⊂ A2
and
ν(A2 ) ≤ λ(f2 ) + .
We have 0 ≤ f1 + f2 ≤ 1 and supp(f1 + f2 ) ⊂ A. Besides, Q ∩ U ⊂ A1 and Q\U ⊂ A2 . Hence
ν ∗ (Q ∩ U ) + ν ∗ (Q\U )
≤ ν ∗ (A1 ) + ν ∗ (A2 )
Hence,
ν ∗ (Q ∩ U ) + ν ∗ (Q\U ) ≤ ν ∗ (A).
Hence U is ν ∗ -measurable. 2
Set ν to be equal to ν ∗ restricted to Baire(X). By Lemmas 8.22 and 8.23 it is a Baire measure.
54
Let gn,j ∈ C(X) with suppgn,j ⊂ Un,j and 0 ≤ gn,j ≤ 1. Set
n−1
1X
gn := gn,j .
n j=1
Then gn ≤ f . Hence
n−1
1X
λ(f ) ≥ λ(gn ) = λ(gn,j ).
n j=1
Thus
n−1 Z
1X
λ(f ) ≥ ν(Un,j ) = fn dν,
n j=1
1
Pn−1 R R
where fn := n j=1 1Un,j . But fn → f and 0 ≤ fn ≤ 1. Hence fn dν → f dν. Thus
Z
λ(f ) ≥ f dν.
We will find a sequence of open sets Wn such that Wn & suppf . Choose functions g̃n,j such that
Pn−1
suppg̃n,0 ⊂ Wn , g̃n,0 = 1 on suppf ; suppg̃n,j ⊂ Un,j , g̃n,j = 1 on Cn,j+1 . Set g̃n := n1 j=0 g̃n,j . Then
f ≤ g̃n . Thus Pn
λ(f ) ≤ λ(g̃n ) = n1 j=0 λ(g̃n,j )
Pn−1
≤ n1 ν(Wn ) + n1 j=0 ν(Un,j ) = n1 ν(Wn ) + f˜n dν,
R
Pn−1
where f˜n := n1 j=1 1Un,j . We have f˜n → f and 0 ≤ f˜n ≤ 1. Hence, by the Lebesgue theorem,
f˜n dν → f dν. Hence
R R
Z
λ(f ) ≤ f dν.
2
Proof of Theorem 8.20. We define the Baire measure ν as described before Lemma 8.24. By Lemma
8.24, λ(f ) = f dν. By Lemma 8.21, ν is uniquely defined. 2
R
Theorem 8.25 Let ν be a Baire measure on X. Then it satisfies the following regularity properties:
(1) ν(A) = inf{ν(U ) : A ⊂ U, U ∈ Open ∩ Baire(X)}, A ∈ Baire(X);
(2) ν(A) = sup{ν(C) : C ⊂ A, C ∈ Closed ∩ Baire(X)} A ∈ Baire(X).
55
8.5 Borel measures on compact spaces
Let X be a compact Hausdorff space. A finite measure on Borel(X) is called a Borel measure on X.
Theorem 8.26 Let µ be a Borel measure on X. The following condtions are equivalent:
(1) µ(A) = inf{µ(U ) : A ⊂ U, U ∈ Open(X)}, A ∈ Borel(X).
(2) µ(A) = sup{µ(C) : C ⊂ A, C ∈ Closed(X)}, A ∈ Borel(X)..
If the above conditions are satisfied then µ is called a regular Borel (or Radon) measure on X.
Proof of Theorem 8.26 Using µ(X) < ∞, we get
Theorem 8.27 Let ν be a Baire measure on X. Then there exists a unique regular Borel measure µ
extending ν. It has the following properties:
(1) If U ∈ Open(X), then
Theorems 8.19 and8.27 imply the following version of the Riesz-Markov theorem:
Theorem 8.28 Let λ be a positive linear functional on C(X). Then there exists a unique regular Borel
measure µ satisfying Z
λ(f ) = f dµ, f ∈ C(X). (8.50)
56
(For U ∈ Open(X), (8.51) agrees with (8.52)).
Exactly as in the previous subsection, we show that µ∗ is an external measure and Borel sets are
µ -measurable. We define µ to be the restriction of µ∗ to Borel(X). By (8.52), it is a regular Borel
∗
measure.
Clearly, if U ∈ Open ∩ Baire(X), then
ν(U ) = ν ∗ (U ) = µ∗ (U ) = µ(U ).
Thus ν coincides with µ on Open ∩ Baire(X). But this is a ∩-closed family generating Baire(X). Hence
ν coincides with µ on Baire(X). 2
f ∈ Cc (X), f ≥ 0 ⇒ λ(f ) ≥ 0.
Theorem 8.31 Let ν be a Baire measure on X. Then it satisfies the following regularity properties:
(1) ν(A) = inf{ν(U ) : A ⊂ U, U ∈ Open0 ∩ Baire(X)}, A ∈ Baire(X);
(2) ν(A) = sup{ν(C) : C ⊂ A, C ∈ Compact ∩ Baire(X)} A ∈ Baire(X).
57
8.7 Borel measures on locally compact spaces
A measure on Borel(X) finite on compact sets is called a Borel measure on X.
Theorem 8.32 Let µ be a Borel measure on X. The following condtions are equivalent:
(1) µ(A) = inf{µ(U ) : A ⊂ U, U ∈ Open0 (X)}, A ∈ Borel(X).
(2) µ(A) = sup{µ(C) : C ⊂ A, C ∈ Compact(X)}, A ∈ Borel(X).
If the above conditions are satisfied then µ is called a regular Borel (or Radon) measure on X.
Theorem 8.33 Let ν be a Baire measure on X. Then there exists a unique regular Borel measure µ
extending ν. It has the following properties:
(1) If U ∈ Open0 (X), then
Theorems 8.29 and8.33 imply the following version of the Riesz-Markov theorem:
Theorem 8.34 Let λ be a positive linear functional on Cc (X). Then there exists a unique regular Borel
measure µ satisfying Z
λ(f ) = f dµ, f ∈ Cc (X). (8.55)
π KJ : × Xj → × Xk ,
j∈J k∈K
π M K π KJ = π M J .
If (Xi , Fi ), i ∈ I is a family of sets with σ-fields, then for J ⊂ I we set ∗ Fj to be the family of
j∈J
subsets of × Xj of the form × Aj with Aj ∈ Fj and Aj = Xj for all but a finite number of j ∈ J. We
j∈J j∈J
set ⊗ Fj := σ−Field( ∗ Fj ).
j∈J j∈J
Clearly, the maps π KJ for K ⊂ J ⊂ I are measurable.
58
9.2 Compatible measures
Let (Xi , Fi ), i ∈ I is a family of sets with σ-fields. Let K ⊂ J ⊂ I and µJ , µK are probabilistic measures
on ( × Xj , ⊗ Fj ) and ( × Xk , ⊗ Fk ) respectively. We say that they are compatible iff πKJ ∗ µJ = µK ,
j∈J j∈J k∈K k∈K
that means Z Z
f (xK )dµK (xK ) = f (xK )dµJ (xK , xJ\K ).
Theorem 9.3 Suppose that for any finite set J ∈ 2I we are given a Baire measure νJ on × Xj .
j∈J
Assume that for any finite K, J ∈ 2I with K ⊂ J, νK is compatible with νJ . Then there exists a unique
Baire measure νI on × Xi compatible with all νJ for finite J.
i∈I
Proof. It is easy to see that the family of measures µJ defines a regular content on Ring ∗ Baire(Xi ) .
i∈I
By Theorem 6.1, it is a premeasure. Hence it admits a unique extension to
σ−Ring Ring ∗ Baire(Xi ) = Baire( × Xi ).
i∈I i∈I
2
Q
Theorem 9.4 Let Xi , i ∈ I be a family of topological spaces. Then Baire1 ( Xi ) = ⊗ Baire1 (Xi ).
i∈I i∈I
References
[Bau] Bauer, H.: Wahrscheinlichkeitstheorie und Grundzüge der Masstheorie, Walter de Gruyter & Co,
Berlin 1968
[Bour] Bourbaki, N.: Elements de mathematique. Les structures fondamentales de l’analyse. Livre VI.
Integration
[Hal] Halmos, P.: Measure theory, Van Nostrand 1950
[LL] Lieb, E., Loss, M.: Analysis, AMS 1997
59