Measure Theory
V. Liskevich
1998
1 Introduction
We always denote by X our universe, i.e. all the sets we shall consider are subsets of X.
Recall some standard notation. 2
X
everywhere denotes the set of all subsets of a given
set X. If A ∩ B = ∅ then we often write A . B rather than A ∪ B, to underline the
disjointness. The complement (in X) of a set A is denoted by A
c
. By A´B the symmetric
diﬀerence of A and B is denoted, i.e. A´B = (A¸ B) ∪ (B ¸ A). Letters i, j, k always
denote positive integers. The sign is used for restriction of a function (operator etc.) to
a subset (subspace).
1.1 The Riemann integral
Recall how to construct the Riemannian integral. Let f : [a, b] →R. Consider a partition
π of [a, b]:
a = x
0
< x
1
< x
2
< . . . < x
n−1
< x
n
= b
and set ∆x
k
= x
k+1
− x
k
, [π[ = max¦∆x
k
: k = 0, 1, . . . , n − 1¦, m
k
= inf¦f(x) : x ∈
[x
k
, x
k+1
]¦, M
k
= sup¦f(x) : x ∈ [x
k
, x
k+1
]¦. Deﬁne the upper and lower Riemann—
Darboux sums
s(f, π) =
n−1
k=0
m
k
∆x
k
, ¯ s(f, π) =
n−1
k=0
M
k
∆x
k
.
One can show (the Darboux theorem) that the following limits exist
lim
π→0
s(f, π) = sup
π
s(f, π) =
_
b
a
fdx
lim
π→0
¯ s(f, π) = inf
π
¯ s(f, π) =
_
b
a
fdx.
1
Clearly,
s(f, π) ≤
_
b
a
fdx ≤
_
b
a
fdx ≤ ¯ s(f, π)
for any partition π.
The function f is said to be Riemann integrable on [a, b] if the upper and lower integrals
are equal. The common value is called Riemann integral of f on [a, b].
The functions cannot have a large set of points of discontinuity. More presicely this
will be stated further.
1.2 The Lebesgue integral
It allows to integrate functions from a much more general class. First, consider a very
useful example. For f, g ∈ C[a, b], two continuous functions on the segment [a, b] = ¦x ∈
R : a x b¦ put
ρ
1
(f, g) = max
axb
[f(x) −g(x)[,
ρ
2
(f, g) =
_
b
a
[f(x) −g(x)[dx.
Then (C[a, b], ρ
1
) is a complete metric space, when (C[a, b], ρ
2
) is not. To prove the latter
statement, consider a family of functions ¦ϕ
n
¦
∞
n=1
as drawn on Fig.1. This is a Cauchy
sequence with respect to ρ
2
. However, the limit does not belong to C[a, b].
2
E
T
L
L
L
L
L
L
L
L
LL
−
1
2
+
1
n
1
2
−
1
n
−
1
2
1
2
Figure 1: The function ϕ
n
.
2 Systems of Sets
Deﬁnition 2.1 A ring of sets is a nonempty subset in 2
X
which is closed with respect
to the operations ∪ and ¸.
Proposition. Let K be a ring of sets. Then ∅ ∈ K.
Proof. Since K ,= ∅, there exists A ∈ K. Since K contains the diﬀerence of every two
its elements, one has A ¸ A = ∅ ∈ K.
Examples.
1. The two extreme cases are K = ¦∅¦ and K = 2
X
.
2. Let X = R and denote by K all ﬁnite unions of semisegments [a, b).
Deﬁnition 2.2 A semiring is a collection of sets P ⊂ 2
X
with the following properties:
1. If A, B ∈ P then A ∩ B ∈ P;
3
2. For every A, B ∈ P there exists a ﬁnite disjoint collection (C
j
) j = 1, 2, . . . , n of
sets (i.e. C
i
∩ C
j
= ∅ if i ,= j) such that
A ¸ B =
n
_
j=1
C
j
.
Example. Let X = R, then the set of all semisegments, [a, b), forms a semiring.
Deﬁnition 2.3 An algebra (of sets) is a ring of sets containing X ∈ 2
X
.
Examples.
1. ¦∅, X¦ and 2
X
are the two extreme cases (note that they are diﬀerent from the
corresponding cases for rings of sets).
2. Let X = [a, b) be a ﬁxed interval on R. Then the system of ﬁnite unions of subin
tervals [α, β) ⊂ [a, b) forms an algebra.
3. The system of all bounded subsets of the real axis is a ring (not an algebra).
Remark. A is algebra if (i) A, B ∈ A =⇒ A ∪ B ∈ A, (ii) A ∈ A =⇒ A
c
∈ A.
Indeed, 1) A ∩ B = (A
c
∪ B
c
)
c
; 2) A ¸ B = A ∩ B
c
.
Deﬁnition 2.4 A σring (a σalgebra) is a ring (an algebra) of sets which is closed with
respect to all countable unions.
Deﬁnition 2.5 A ring (an algebra, a σalgebra) of sets, K(U) generated by a collection
of sets U ⊂ 2
X
is the minimal ring (algebra, σalgebra) of sets containing U.
In other words, it is the intersection of all rings (algebras, σalgebras) of sets containing
U.
4
3 Measures
Let X be a set, A an algebra on X.
Deﬁnition 3.1 A function µ: A −→R
+
∪ ¦∞¦ is called a measure if
1. µ(A) 0 for any A ∈ A and µ(∅) = 0;
2. if (A
i
)
i1
is a disjoint family of sets in A ( A
i
∩ A
j
= ∅ for any i ,= j) such that
∞
i=1
A
i
∈ A, then
µ(
∞
_
i=1
A
i
) =
∞
i=1
µ(A
i
).
The latter important property, is called countable additivity or σadditivity of the measure
µ.
Let us state now some elementary properties of a measure. Below till the end of this
section A is an algebra of sets and µ is a measure on it.
1. (Monotonicity of µ) If A, B ∈ A and B ⊂ A then µ(B) µ(A).
Proof. A = (A ¸ B) . B implies that
µ(A) = µ(A ¸ B) + µ(B).
Since µ(A ¸ B) ≥ 0 it follows that µ(A) ≥ µ(B).
2. (Subtractivity of µ). If A, B ∈ A and B ⊂ A and µ(B) < ∞ then µ(A ¸ B) =
µ(A) −µ(B).
Proof. In 1) we proved that
µ(A) = µ(A ¸ B) + µ(B).
If µ(B) < ∞ then
µ(A) −µ(B) = µ(A ¸ B).
3. If A, B ∈ A and µ(A ∩ B) < ∞ then µ(A ∪ B) = µ(A) + µ(B) −µ(A ∩ B).
Proof. A ∩ B ⊂ A, A ∩ B ⊂ B, therefore
A ∪ B = (A ¸ (A ∩ B)) . B.
Since µ(A ∩ B) < ∞, one has
µ(A ∪ B) = (µ(A) −µ(A ∩ B)) + µ(B).
5
4. (Semiadditivity of µ). If (A
i
)
i≥1
⊂ A such that
∞
i=1
A
i
∈ A then
µ(
∞
_
i=1
A
i
)
∞
i=1
µ(A
i
).
Proof. First let us proove that
µ(
n
_
i=1
A
i
)
n
i=1
µ(A
i
).
Note that the family of sets
B
1
= A
1
B
2
= A
2
¸ A
1
B
3
= A
3
¸ (A
1
∪ A
2
)
. . .
B
n
= A
n
¸
n−1
_
i=1
A
i
is disjoint and
n
i=1
B
i
=
n
i=1
A
i
. Moreover, since B
i
⊂ A
i
, we see that µ(B
i
) ≤
µ(A
i
). Then
µ(
n
_
i=1
A
i
) = µ(
n
_
i=1
B
i
) =
n
i=1
µ(B
i
) ≤
n
i=1
µ(A
i
).
Now we can repeat the argument for the inﬁnite family using σadditivity of the
measure.
3.1 Continuity of a measure
Theorem 3.1 Let A be an algebra, (A
i
)
i≥1
⊂ A a monotonically increasing sequence of
sets (A
i
⊂ A
i+1
) such that
i≥1
∈ A. Then
µ(
∞
_
i=1
A
i
) = lim
n→∞
µ(A
n
).
Proof. 1). If for some n
0
µ(A
n
0
) = +∞then µ(A
n
) = +∞∀n ≥ n
0
and µ(
∞
i=1
A
i
) = +∞.
2). Let now µ(A
i
) < ∞ ∀i ≥ 1.
6
Then
µ(
∞
_
i=1
A
i
) = µ(A
1
. (A
2
¸ A
1
) . . . . . (A
n
¸ A
n−1
) . . . .)
= µ(A
1
) +
∞
k=2
µ(A
k
¸ A
k−1
)
= µ(A
1
) + lim
n→∞
n
k=2
(µ(A
k
) −µ(A
k−1
)) = lim
n→∞
µ(A
n
).
3.2 Outer measure
Let a be an algebra of subsets of X and µ a measure on it. Our purpose now is to extend
µ to as many elements of 2
X
as possible.
An arbitrary set A ⊂ X can be always covered by sets from A, i.e. one can always ﬁnd
E
1
, E
2
, . . . ∈ A such that
∞
i=1
E
i
⊃ A. For instance, E
1
= X, E
2
= E
3
= . . . = ∅.
Deﬁnition 3.2 For A ⊂ X its outer measure is deﬁned by
µ
∗
(A) = inf
∞
i=1
µ(E
i
)
where the inﬁmum is taken over all Acoverings of the set A, i.e. all collections (E
i
), E
i
∈
A with
i
E
i
⊃ A.
Remark. The outer measure always exists since µ(A) 0 for every A ∈ A.
Example. Let X = R
2
, A = K(P), σalgebra generated by P, P = ¦[a, b) R
1
¦.
Thus A consists of countable unions of strips like one drawn on the picture. Put µ([a, b)
R
1
) = b − a. Then, clearly, the outer measure of the unit disc x
2
+ y
2
1 is equal to 2.
The same value is for the square [x[ 1, [y[ 1.
Theorem 3.2 For A ∈ A one has µ
∗
(A) = µ(A).
In other words, µ
∗
is an extension of µ.
Proof. 1. A is its own covering. This implies µ
∗
(A) µ(A).
2. By deﬁnition of inﬁmum, for any ε > 0 there exists a Acovering (E
i
) of A such that
i
µ(E
i
) < µ
∗
(A) + ε. Note that
A = A ∩ (
_
i
E
i
) =
_
i
(A ∩ E
i
).
7
E
T
a b
Using consequently σsemiadditivity and monotonicity of µ, one obtains:
µ(A)
i
µ(A ∩ E
i
)
i
µ(E
i
) < µ
∗
(A) + ε.
Since ε is arbitrary, we conclude that µ(A) µ
∗
(A).
It is evident that µ
∗
(A) 0, µ
∗
(∅) = 0 (Check !).
Lemma. Let A be an algebra of sets (not necessary σalgebra), µ a measure on A. If
there exists a set A ∈ A such that µ(A) < ∞, then µ(∅) = 0.
Proof. µ(A ¸ A) = µ(A) −µ(A) = 0.
Therefore the property µ(∅) = 0 can be substituted with the existence in A of a set
with a ﬁnite measure.
Theorem 3.3 (Monotonicity of outer measure). If A ⊂ B then µ
∗
(A) µ
∗
(B).
Proof. Any covering of B is a covering of A.
Theorem 3.4 (σsemiadditivity of µ
∗
). µ
∗
(
∞
j=1
A
j
)
∞
j=1
µ
∗
(A
j
).
8
Proof. If the series in the righthand side diverges, there is nothing to prove. So assume
that it is convergent.
By the deﬁnition of outer measur for any ε > 0 and for any j there exists an Acovering
k
E
kj
⊃ A
j
such that
∞
k=1
µ(E
kj
) < µ
∗
(A
j
) +
ε
2
j
.
Since
∞
_
j,k=1
E
kj
⊃
∞
_
j=1
A
j
,
the deﬁnition of µ
∗
implies
µ
∗
(
∞
_
j=1
A
j
)
∞
j,k=1
µ(E
kj
)
and therefore
µ
∗
(
∞
_
j=1
A
j
) <
∞
j=1
µ
∗
(A
j
) + ε.
3.3 Measurable Sets
Let A be an algebra of subsets of X, µ a measure on it, µ
∗
the outer measure deﬁned in
the previous section.
Deﬁnition 3.3 A ⊂ X is called a measurable set (by Carath`eodory) if for any E ⊂ X
the following relation holds:
µ
∗
(E) = µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
).
Denote by
˜
A the collection of all set which are measurable by Carath`eodory and set
˜ µ = µ
∗
˜
A.
Remark Since E = (E ∩ A) ∪ (E ∩ A
c
), due to semiadditivity of the outer measure
µ
∗
(E) ≤ µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
).
Theorem 3.5
˜
A is a σalgebra containing A, and ˜ µ is a measure on
˜
A.
9
Proof. We devide the proof into several steps.
1. If A, B ∈
˜
A then A ∪ B ∈
˜
A.
By the deﬁnition one has
µ
∗
(E) = µ
∗
(E ∩ B) + µ
∗
(E ∩ B
c
). (1)
Take E ∩ A instead of E:
µ
∗
(E ∩ A) = µ
∗
(E ∩ A ∩ B) + µ
∗
(E ∩ A ∩ B
c
). (2)
Then put E ∩ A
c
in (1) instead of E
µ
∗
(E ∩ A
c
) = µ
∗
(E ∩ A
c
∩ B) + µ
∗
(E ∩ A
c
∩ B
c
). (3)
Add (2) and (3):
µ
∗
(E) = µ
∗
(E ∩ A ∩ B) + µ
∗
(E ∩ A ∩ B
c
) + µ
∗
(E ∩ A
c
∩ B) + µ
∗
(E ∩ A
c
∩ B
c
). (4)
Substitute E ∩ (A ∪ B) in (4) instead of E. Note that
1) E ∩ (A ∪ B) ∩ A ∩ B = E ∩ A ∩ B
2) E ∩ (A ∪ B) ∩ A
c
∩ B = E ∩ A
c
∩ B
3) E ∩ (A ∪ B) ∩ A ∩ B
c
= E ∩ A ∩ B
c
4) E ∩ (A ∪ B) ∩ A
c
∩ B
c
= ∅.
One has
µ
∗
(E ∩ (A ∪ B)) = µ
∗
(E ∩ A ∩ B) + µ
∗
(E ∩ A
c
∩ B) + µ
∗
(E ∩ A ∩ B
c
). (5)
From (4) and (5) we have
µ
∗
(E) = µ
∗
(E ∩ (A ∪ B)) + µ
∗
(E ∩ (A ∪ B)
c
).
2. If A ∈
˜
A then A
c
∈
˜
A.
The deﬁnition of measurable set is symmetric with respect to A and A
c
.
Therefore
˜
A is an algebra of sets.
3.
Let A, B ∈ A, A ∩ B = ∅. From (5)
µ
∗
(E ∩ (A . B)) = µ
∗
(E ∩ A
c
∩ B) + µ
∗
(E ∩ A ∩ B
c
) = µ
∗
(E ∩ B) + µ
∗
(E ∩ A).
10
4.
˜
A is a σalgebra.
From the previous step, by induction, for any ﬁnite disjoint collection (B
j
) of sets:
µ
∗
(E ∩ (
n
_
j=1
B
j
)) =
n
j=1
µ
∗
(E ∩ B
j
). (6)
Let A =
∞
j=1
A
j
, A
j
∈ A. Then A =
∞
j=1
B
j
, B
j
= A
j
¸
j−1
k=1
A
k
and
B
i
∩ B
j
= ∅ (i ,= j). It suﬃces to prove that
µ
∗
(E) µ
∗
(E ∩ (
∞
_
j=1
B
j
)) + µ
∗
(E ∩ (
∞
_
j=1
B
j
)
c
). (7)
Indeed, we have already proved that µ
∗
is σsemiadditive.
Since
˜
A is an algebra, it follows that
n
j=1
B
j
∈
˜
A(∀n ∈ N) and the following inequality
holds for every n:
µ
∗
(E) µ
∗
(E ∩ (
n
_
j=1
B
j
)) + µ
∗
(E ∩ (
n
_
j=1
B
j
)
c
). (8)
Since E ∩ (
∞
j=1
B
j
)
c
⊂ E ∩ (
n
j=1
B
j
)
c
, by monotonicity of the mesasure and (8)
µ
∗
(E) ≥
n
j=1
µ
∗
(E ∩ B
j
) + µ
∗
(E ∩ A
c
). (9)
Passing to the limit we get
µ
∗
(E) ≥
∞
j=1
µ
∗
(E ∩ B
j
) + µ
∗
(E ∩ A
c
). (10)
Due to semiadditivity
µ
∗
(E ∩ A) = µ
∗
(E ∩ (
∞
_
j=1
B
j
)) = µ
∗
(
∞
_
j=1
(E ∩ B
j
)) ≤
∞
j=1
µ
∗
(E ∩ B
j
).
Compare this with (10):
µ
∗
(E) ≥ µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
).
Thus, A ∈
˜
A, which means that
˜
A is a σalgebra.
5. ˜ µ = µ
∗
˜
A is a measure.
11
We need to prove only σadditivity. Let E =
∞
j=1
A
j
. From(10) we get
µ
∗
(
∞
_
j=1
A
j
)
∞
j=1
µ
∗
(A
j
).
The oposite inequality follows from σsemiadditivity of µ
∗
.
6.
˜
A ⊃ A.
Let A ∈ A, E ⊂ X. We need to prove:
µ
∗
(E) µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
). (11)
If E ∈ A then (11) is clear since E ∩ A and E ∩ A
c
are disjoint and both belong to A
where µ
∗
= µ and so is additive.
For E ⊂ X for ∀ε > 0 there exists a Acovering (E
j
) of E such that
µ
∗
(E) + ε >
∞
j=1
µ(E
j
). (12)
Now, since E
j
= (E
j
∩ A) ∪ (E
j
∩ A
c
), one has
µ(E
j
) = µ(E
j
∩ A) + µ(E
j
∩ A)
and also
E ∩ A ⊂
∞
_
j=1
(E
j
∩ A)
E ∩ A
c
⊂
∞
_
j=1
(E
j
∩ A
c
)
By monotonicity and σsemiadditivity
µ
∗
(E ∩ A)
∞
j=1
µ(E
j
∩ A),
µ
∗
(E ∩ A
c
)
∞
j=1
µ(E
j
∩ A
c
).
Adding the last two inequalities we obtain
µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
) ≤
∞
j=1
µ
∗
(E
j
) < µ
∗
(E) + ε.
Since ε > 0 is arbitrary, (11) is proved.
The following theorem is a direct consequence of the previous one.
12
Theorem 3.6 Let A be an algebra of subsets of X and µ be a measure on it. Then there
exists a σalgebra A
1
⊃ A and a measure µ
1
on A
1
such that µ
1
A = µ.
Remark. Consider again an algebra A of subsets of X. Denot by A
σ
the generated
σalgebra and construct the extension µ
σ
of µ on A
σ
. This extension is called minimal
extension of measure.
Since
˜
A ⊃ A therefore A
σ
⊂
˜
A. Hence one can set µ
σ
= ˜ µ A
σ
. Obviously µ
σ
is a
minimal extension of µ. It always exists. On can also show (see below) that this extension
is unique.
Theorem 3.7 Let µ be a measure on an algebra A of subsets of X, µ
∗
the corresponding
outer measure. If µ
∗
(A) = 0 for a set A ⊂ X then A ∈
˜
A and ˜ µ(A) = 0.
Proof. Clearly, it suﬃces to prove that A ∈
˜
A. Further, it suﬃces to prove that µ
∗
(E)
µ
∗
(E ∩ A) + µ
∗
(E ∩ A
c
). The latter statement follows from monotonicity of µ
∗
. Indeed,
one has µ
∗
(E ∩ A) µ
∗
(A) = 0 and µ
∗
(E ∩ A
c
) µ
∗
(E).
Deﬁnition 3.4 A measure µ on an algebra of sets A is called complete if conditions
B ⊂ A, A ∈ A, µ(A) = 0 imply B ∈ A and µ(B) = 0.
Corollary. ˜ µ is a complete measure.
Deﬁnition 3.5 A measure µ on an algebra A is called ﬁnite if µ(X) < ∞. It is called
σﬁnite if the is an increasing sequence (F
j
)
j≥1
⊂ A such that X =
j
F
j
and µ(F
j
) < ∞
∀j.
Theorem 3.8 Let µ be a σﬁnite measure on an algebra A. Then there exist a unique
extension of µ to a measure on
˜
A.
Proof. It suﬃces to sjow uniqueness. Let ν be another extension of µ (ν A = µ A).
First, let µ (and therefore ν, µ
∗
) be ﬁnite. Let A ∈
˜
A. Let (E
j
) ⊂ A such that
A ⊂
j
E
j
. We have
ν(A) ≤ ν(
∞
_
j=1
E
j
) ≤
∞
j=1
ν(E
j
) =
∞
j=1
µ(E
j
).
Therefore
ν(A) ≤ µ
∗
(A) ∀A ∈
˜
A.
13
Since µ
∗
and ν are additive (on
˜
A) it follows that
µ
∗
(A) + µ
∗
(A
c
) = ν(A) + ν(A
c
).
The terms in the RHS are ﬁnite and ν(A) ≤ µ
∗
(A), ν(A
c
) ≤ µ
∗
(A
c
). From this we infer
that
ν(A) = µ
∗
(A) ∀A ∈
˜
A.
Now let µ be σﬁnite, (F
j
) be an increasing sequence of sets from A such that µ(F
j
) <
∞ ∀j and X =
∞
j=1
F
j
. From what we have already proved it follows that
µ
∗
(A ∩ F
j
) = ν(A ∩ F
j
) ∀A ∈
˜
A.
Therefore
µ
∗
(A) = lim
j
µ
∗
(A ∩ F
j
) = lim
j
ν(A ∩ F
j
) = ν(A).
Theorem 3.9 (Continuity of measure). Let A be a σalgebra with a measure µ, ¦A
j
¦ ⊂
A a monotonically increasing sequence of sets. Then
µ(
∞
_
j=1
A
j
) = lim
j→∞
µ(A
j
).
Proof. One has:
A =
∞
_
j=1
A
j
=
∞
_
j=2
(A
j+1
¸ A
j
) . A
1
.
Using σadditivity and subtractivity of µ,
µ(A) =
∞
j=1
(µ(A
j+1
) −µ(A
j
)) + µ(A
1
) = lim
j→∞
µ(A
j
).
Similar assertions for a decreasing sequence of sets in A can be proved using de Morgan
formulas.
Theorem 3.10 Let A ∈
˜
A. Then for any ε > 0 there exists A
ε
∈ A such that µ
∗
(A ´
A
ε
) < ε.
Proof. 1. For any ε > 0 there exists an A cover
E
j
⊃ A such that
j
µ(E
j
) < µ
∗
(A) +
ε
2
= ˜ µ(A) +
ε
2
.
14
On the other hand,
j
µ(E
j
) ˜ µ(
_
j
E
J
).
The monotonicity of ˜ µ implies
˜ µ(
∞
_
j=1
E
J
) = lim
n→∞
˜ µ(
n
_
j=1
E
j
),
hence there exists a positive integer N such that
˜ µ(
∞
_
j=1
E
j
) − ˜ µ(
N
_
j=1
E
j
) <
ε
2
. (13)
2. Now, put
A
ε
=
N
_
j=1
E
j
and prove that µ
∗
(A ´A
ε
) < ε.
2a. Since
A ⊂
∞
_
j=1
E
j
,
one has
A ¸ A
ε
⊂
∞
_
j=1
E
j
¸ A
ε
.
Since
A
ε
⊂
∞
_
j=1
E
j
,
one can use the monotonicity and subtractivity of ˜ µ. Together with estimate (13), this
gives
˜ µ(A ¸ A
ε
) ≤ ˜ µ(
∞
_
j=1
E
j
¸ A
ε
) <
ε
2
.
2b. The inclusion
A
ε
¸ A ⊂
∞
_
j=1
E
j
¸ A
implies
˜ µ(A
ε
¸ A) ˜ µ(
∞
_
j=1
E
j
¸ A) = ˜ µ(
∞
_
j=1
E
j
) − ˜ µ(A) <
ε
2
.
15
Here we used the same properties of ˜ µ as above and the choice of the cover (E
j
).
3. Finally,
˜ µ(A ´A
ε
) ˜ µ(A ¸ A
ε
) + ˜ µ(A
ε
¸ A).
16
4 Monotone Classes
and Uniqueness of Extension of Measure
Deﬁnition 4.1 A collection of sets, M is called a monotone class if together with any
monotone sequence of sets M contains the limit of this sequence.
Example. Any σring. (This follows from the Exercise 1. below).
Exercises.
1. Prove that any σring is a monotone class.
2. If a ring is a monotone class, then it is a σring.
We shall denote by M(K) the minimal monotone class containing K.
Theorem 4.1 Let K be a ring of sets, K
σ
the σring generated by K. Then M(K) = K
σ
.
Proof. 1. Clearly, M(K) ⊂ K
σ
. Now, it suﬃces to prove that M(K) is a ring. This follows
from the Exercise (2) above and from the minimality of K
σ
.
2. M(K) is a ring.
2a. For B ⊂ X, set
K
B
= ¦A ⊂ X : A ∪ B, A ∩ B, A ¸ B, B ¸ A ∈ M(K)¦.
This deﬁnition is symmetric with respect to A and B, therefore A ∈ K
B
implies B ∈ K
A
.
2b. K
B
is a monotone class.
Let (A
j
) ⊂ K
B
be a monotonically increasing sequence. Prove that the union, A =
A
j
belongs to K
B
.
Since A
j
∈ K
B
, one has A
j
∪ B ∈ K
B
, and so
A ∪ B =
∞
_
j=1
(A
j
∪ B) ∈ M(K).
In the same way,
A ¸ B = (
∞
_
j=1
A
j
) ¸ B =
∞
_
j=1
(A
j
¸ B) ∈ M(K);
17
B ¸ A = B ¸ (
∞
_
j=1
A
j
) =
∞
j=1
(B ¸ A
j
) ∈ M(K).
Similar proof is for the case of decreasing sequence (A
j
).
2c. If B ∈ K then M(K) ⊂ K
B
.
Obviously, K ⊂ K
B
. Together with minimality of M(K), this implies M(K) ⊂ K
B
.
2d. If B ∈ M(K) then M(K) ⊂ K
B
.
Let A ∈ K. Then M(K) ⊂ K
A
. Thus if B ∈ M(K), one has B ∈ K
A
, so A ∈ K
B
.
Hence what we have proved is K ⊂ K
B
. This implies M(K) ⊂ K
B
.
2e. It follows from 2a. — 2d. that if A, B ∈ M(K) then A ∈ K
B
and so A ∪ B, A ∩ B,
A ¸ B and B ¸ A all belong to M(K).
Theorem 4.2 Let A be an algebra of sets, µ and ν two measures deﬁned on the σ
algebra A
σ
generated by A. Then µ A = ν A implies µ = ν.
Proof. Choose A ∈ A
σ
, then A = lim
n→∞
A
n
, A
n
∈ A, for A
σ
= M(A). Using continuity
of measure, one has
µ(A) = lim
n→∞
µ(A
n
) = lim
n→∞
ν(A
n
) = ν(A).
Theorem 4.3 Let A be an algebra of sets, B ⊂ X such that for any ε > 0 there exists
A
ε
∈ A with µ
∗
(B ´A
ε
) < ε. Then B ∈
˜
A.
Proof. 1. Since any outer measure is semiadditive, it suﬃces to prove that for any E ⊂ X
one has
µ
∗
(E) µ
∗
(E ∩ B) + µ
∗
(E ∩ B
c
).
2a. Since A ⊂
˜
A, one has
µ
∗
(E ∩ A
ε
) + µ
∗
(E ∩ A
c
ε
) µ
∗
(E). (14)
2b. Since A ⊂ B ∪ (A ´ B) and since the outer measure µ
∗
is monotone and semi
additive, there is an estimate [µ
∗
(A) −µ
∗
(B)[ µ
∗
(A ´B) for any A, B ⊂ X. (C.f. the
proof of similar fact for measures above).
2c. It follows from the monotonicity of µ
∗
that
[µ
∗
(E ∩ A
ε
) −µ
∗
(E ∩ B)[ µ
∗
((E ∩ A
ε
) ´(E ∩ B)) µ(A
ε
∩ B) < ε.
18
Therefore, µ
∗
(E ∩ A
ε
) > µ
∗
(E ∩ B) −ε.
In the same manner, µ
∗
(E ∩ A
c
ε
) > µ
∗
(E ∩ B
c
) −ε.
2d. Using (14), one obtains
µ
∗
(E) > µ
∗
(E ∩ B) + µ
∗
(E ∩ B
c
) −2ε.
19
5 The Lebesgue Measure on the real line R
1
5.1 The Lebesgue Measure of Bounded Sets of R
1
Put A for the algebra of all ﬁnite unions of semisegments (semiintervals) on R
1
, i.e. all
sets of the form
A =
k
_
j=1
[a
j
, b
j
).
Deﬁne a mapping µ : A −→R by:
µ(A) =
k
j=1
(b
j
−a
j
).
Theorem 5.1 µ is a measure.
Proof. 1. All properties including the (ﬁnite) additivity are obvious. The only thing to
be proved is the σadditivity.
Let (A
j
) ⊂ A be such a countable disjoint family that
A =
∞
_
j=1
A
j
∈ A.
The condition A ∈ A means that
A
j
is a ﬁnite union of intervals.
2. For any positive integer n,
n
_
j=1
A
j
⊂ A,
hence
n
j=1
µ(A
j
) µ(A),
and
∞
j=1
µ(A
j
) = lim
n→∞
n
j=1
µ(A
j
) µ(A).
3. Now, let A
ε
a set obtained from A by the following construction. Take a connected
component of A. It is a semisegment of the form [s, t). Shift slightly on the left its
righthand end, to obtain a (closed) segment. Do it with all components of A, in such a
way that
µ(A) < µ(A
ε
) + ε. (15)
20
Apply a similar procedure to each semisegment shifting their left end point to the left
A
j
= [a
j
, b
j
), and obtain (open) intervals, A
ε
j
with
µ(A
ε
j
) < µ(A
j
) +
ε
2
j
. (16)
4. By the construction, A
ε
is a compact set and (A
ε
j
) its open cover. Hence, there exists
a positive integer n such that
n
_
j=1
A
ε
j
⊃ A
ε
.
Thus
µ(A
ε
)
n
j=1
µ(A
ε
j
).
The formulas (15) and (16) imply
µ(A) <
n
j=1
µ(A
ε
j
) + ε
n
j=1
µ(A
j
) +
n
j=1
ε
2
j
+ ε,
thus
µ(A) <
∞
j=1
µ(A
j
) + 2ε.
Now, one can apply the Carath`eodory’s scheme developed above, and obtain the mea
sure space (
˜
A, ˜ µ). The result of this extension is called the Lebesgue measure. We shall
denote the Lebesgue measure on R
1
by m.
Exercises.
1. A one point set is measurable, and its Lebesgue measure is equal to 0.
2. The same for a countable subset in R
1
. In particular, m(Q∩ [0, 1]) = 0.
3. Any open or closed set in R
1
is Lebesgue measurable.
Deﬁnition 5.1 Borel algebra of sets, B on the real line R
1
is a σalgebra generated by
all open sets on R
1
. Any element of B is called a Borel set.
Exercise. Any Borel set is Lebesgue measurable.
Theorem 5.2 Let E ⊂ R
1
be a Lebesgue measurable set. Then for any ε > 0 there
exists an open set G ⊃ E such that m(G¸ E) < ε.
21
Proof. Since E is measurable, m
∗
(E) = m(E). According the deﬁnition of an outer
measure, for any ε > 0 there exists a cover A =
[a
k
, b
k
) ⊃ E such that
m(A) < m(E) +
ε
2
.
Now, put
G =
_
(a
k
−
ε
2
k+1
, b
k
).
Problem. Let E ⊂ R
1
be a bounded Lebesgue measurable set. Then for any ε > 0
there exists a compact set F ⊂ E such that m(E ¸ F) < ε. (Hint: Cover E with a
semisegment and apply the above theorem to the σalgebra of measurable subsets in this
semisegment).
Corollary. For any ε > 0 there exist an open set G and a compact set F such that
G ⊃ E ⊃ F and m(G¸ F) < ε.
Such measures are called regular.
5.2 The Lebesgue Measure on the Real Line R
1
We now abolish the condition of boundness.
Deﬁnition 5.2 A set A on the real numbers line R
1
is Lebesgue measurable if for any
positive integer n the bounded set A ∩ [−n, n) is a Lebesgue measurable set.
Deﬁnition 5.3 The Lebesgue measure on R
1
is
m(A) = lim
n→∞
m(A ∩ [−n, n)).
Deﬁnition 5.4 A measure is called σﬁnite if any measurable set can be represented as
a countable union of subsets each has a ﬁnite measure.
Thus the Lebesgue measure m is σﬁnite.
Problem. The Lebesgue measure on R
1
is regular.
5.3 The Lebesgue Measure in R
d
Deﬁnition 5.5 We call a ddimensional rectangle in R
d
any set of the form
¦x : x ∈ R
d
: a
i
x
i
< b
i
¦.
22
Using rectangles, one can construct the Lebesque measure in R
d
in the same fashion as
we deed for the R
1
case.
23
6 Measurable functions
Let X be a set, A a σalgebra on X.
Deﬁnition 6.1 A pair (X, A) is called a measurable space.
Deﬁnition 6.2 Let f be a function deﬁned on a measurable space (X, A), with values in
the extended real number system. The function f is called measurable if the set
¦x : f(x) > a¦
is measurable for every real a.
Example.
Theorem 6.1 The following conditions are equivalent
¦x : f(x) > a¦ is measurable for every real a. (17)
¦x : f(x) ≥ a¦ is measurable for every real a. (18)
¦x : f(x) < a¦ is measurable for every real a. (19)
¦x : f(x) ≤ a¦ is measurable for every real a. (20)
Proof. The statement follows from the equalities
¦x : f(x) ≥ a¦ =
∞
n=1
¦x : f(x) > a −
1
n
¦, (21)
¦x : f(x) < a¦ = X ¸ ¦x : f(x) ≥ a¦, (22)
¦x : f(x) ≤ a¦ =
∞
n=1
¦x : f(x) < a +
1
n
¦, (23)
¦x : f(x) > a¦ = X ¸ ¦x : f(x) ≤ a¦ (24)
Theorem 6.2 Let (f
n
) be a sequence of measurable functions. For x ∈ X set
g(x) = sup
n
f
n
(x)(n ∈ N)
h(x) = limsup
n→∞
f
n
(x).
Then g and h are measurable.
24
Proof.
¦x : g(x) ≤ a¦ =
∞
n=1
¦x : f
n
(x) ≤ a¦.
Since the LHS is measurable it follows that the RHS is measurable too. The same proof
works for inf.
Now
h(x) = inf g
m
(x),
where
g
m
(x) = sup
n≥m
f
n
(x).
Theorem 6.3 Let f and g be measurable realvalued functions deﬁned on X. Let F be
real and continuous function on R
2
. Put
h(x) = F(f(x), g(x)) (x ∈ X).
Then h is measurable.
Proof. Let G
a
= ¦(u, v) : F(u, v) > a¦. Then G
a
is an open subset of R
2
, and thus
G
a
=
∞
_
n=1
I
n
where (I
n
) is a sequence of open intervals
I
n
= ¦(u, v) : a
n
< u < b
n
, c
n
< v < d
n
¦.
The set ¦x : a
n
< f(x) < b
n
¦ is measurable and so is the set
¦x : (f(x), g(x)) ∈ I
n
¦ = ¦x : a
n
< f(x) < b
n
¦ ∩ ¦x : c
n
< g(x) < d
n
¦.
Hence the same is true for
¦x : h(x) > a¦ = ¦x : (f(x), g(x)) ∈ G
a
¦ =
∞
_
n=1
¦x : (f(x), g(x)) ∈ I
n
¦.
Corollories. Let f and g be measurable. Then the following functions are measurable
(i)f + g (25)
(ii)f g (26)
(iii)[f[ (27)
(iv)
f
g
(ifg ,= 0) (28)
(v) max¦f, g¦, min¦f, g¦ (29)
(30)
since max¦f, g¦ = 1/2(f + g +[f −g[), min¦f, g¦ = 1/2(f + g −[f −g[).
25
6.1 Step functions (simple functions)
Deﬁnition 6.3 A real valued function f : X → R is called simple function if it takes
only a ﬁnite number of distinct values.
We will use below the following notation
χ
E
(x) =
_
1 if x ∈ E
0 otherwise
Theorem 6.4 A simple function f =
n
j=1
c
j
χ
E
j
is measurable if and only if all the
sets E
j
are measurable.
Exercise. Prove the theorem.
Theorem 6.5 Let f be real valued. There exists a sequence (f
n
) of simple functions such
that f
n
(x) −→ f(x) as n → ∞, for every x ∈ X. If f is measurable, (f
n
) may be chosen
to be a sequence of measurable functions. If f ≥ 0, (f
n
) may be chosen monotonically
increasing.
Proof. If f ≥ 0 set
f
n
(x) =
n·2
n
i=1
i−1
2
n
χ
E
n
i
+ nχ
F
n
where
E
n
i
= ¦x :
i−1
2
n
≤ f(x) <
i
2
n
¦, F
n
= ¦x : f(x) ≥ n¦.
The sequence (f
n
) is monotonically increasing, f
n
is a simple function. If f(x) < ∞then
f(x) < n for a suﬃciently large n and [f
n
(x) − f(x)[ < 1/2
n
. Therefore f
n
(x) −→ f(x).
If f(x) = +∞ then f
n
(x) = n and again f
n
(x) −→ f(x).
In the general case f = f
+
−f
−
, where
f
+
(x) := max¦f(x), 0¦, f
−
(x) := −min¦f(x), 0¦.
Note that if f is bounded then f
n
−→ f uniformly.
26
7 Integration
Deﬁnition 7.1 A triple (X, A, µ), where A is a σalgebra of subsets of X and µ is a
measure on it, is called a measure space.
Let (X, A, µ) be a measure space. Let f : X →R be a simple measurable function.
f(x) =
n
i=1
c
i
χ
E
i
(x) (31)
and
n
_
i=1
E
i
= X, E
i
∩ E
j
= ∅ (i ,= j).
There are diﬀerent representations of f by means of (31). Let us choose the represen
tation such that all c
i
are distinct.
Deﬁnition 7.2 Deﬁne the quantity
I(f) =
n
i=1
c
i
µ(E
i
).
First, we derive some properties of I(f).
Theorem 7.1 Let f be a simple measurable function. If X =
k
j=1
F
j
and f takes the
constant value b
j
on F
j
then
I(f) =
k
j=1
b
j
µ(F
j
).
Proof. Clearly, E
i
=
j: b
j
=c
i
F
j
.
i
c
i
µ(E
i
) =
n
i=1
c
i
µ(
_
j: b
j
=c
i
F
j
) =
n
i=1
c
i
j: b
j
=c
i
µ(F
j
) =
k
j=1
b
j
µ(F
j
).
This show that the quantity I(f) is well deﬁned.
27
Theorem 7.2 If f and g are measurable simple functions then
I(αf + βg) = αI(f) + βI(g).
Proof. Let f(x) =
n
j=1
b
j
χ
F
j
(x), X =
n
j=1
F
j
, g(x) =
m
k=1
c
k
χ
G
k
(x), X =
n
k=1
G
k
.
Then
αf + βg =
n
j=1
m
k=1
(αb
j
+ βc
k
)χ
E
jk
(x)
where E
jk
= F
j
∩ G
k
.
Exercise. Complete the proof.
Theorem 7.3 Let f and g be simple measurable functions. Suppose that f ≤ g every
where except for a set of measure zero. Then
I(f) ≤ I(g).
Proof. If f ≤ g everywhere then in the notation of the previous proof b
j
≤ c
k
on E
jk
and
I(f) ≤ I(g) follows.
Otherwise we can assume that f ≤ g + φ where φ is nonnegative measurable simple
function which is zero every exept for a set N of measure zero. Then I(φ) = 0 and
I(f) ≤ I(g + φ) = I(f) + I(φ) = I(g).
Deﬁnition 7.3 If f : X → R
1
is a nonnegative measurable function, we deﬁne the
Lebesgue integral of f by
_
fdµ := sup I(φ)
where sup is taken over the set of all simple functions φ such that φ ≤ f.
Theorem 7.4 If f is a simple measurable function then
_
fdµ = I(f).
Proof. Since f ≤ f it follows that
_
fdµ ≥ I(f).
On the other hand, if φ ≤ f then I(φ) ≤ I(f) and also
sup
φ≤f
I(φ) ≤ I(f)
which leads to the inequality
_
fdµ ≤ I(f).
28
Deﬁnition 7.4 1. If A is a measurable subset of X (A ∈ A)and f is a nonnegative
measurable function then we deﬁne
_
A
fdµ =
_
fχ
A
dµ.
2.
_
fdµ =
_
f
+
dµ −
_
f
−
dµ
if at least one of the terms in RHS is ﬁnite. If both are ﬁnite we call f integrable.
Remark. Finiteness of the integrals
_
f
+
dµ and
_
f
−
dµ is equivalent to the ﬁnitenes of
the integral
_
[f[dµ.
If it is the case we write f ∈ L
1
(X, µ) or simply f ∈ L
1
if there is no ambiguity.
The following properties of the Lebesgue integral are simple consequences of the deﬁ
nition. The proofs are left to the reader.
• If f is measurable and bounded on A and µ(A) < ∞ then f is integrable on A.
• If a ≤ f(x) ≤ b (x ∈ A)), µ(A) < ∞ then
aµ(A) ≤
_
A
fdµ ≤ bµ(a).
• If f(x) ≤ g(x) for all x ∈ A then
_
A
fdµ ≤
_
A
gdµ.
• Prove that if µ(A) = 0 and f is measurable then
_
A
fdµ = 0.
The next theorem expresses an important property of the Lebesgue integral. As a con
sequence we obtain convergence theorems which give the main advantage of the Lebesgue
approach to integration in comparison with Riemann integration.
29
Theorem 7.5 Let f be measurable on X. For A ∈ A deﬁne
φ(A) =
_
A
fdµ.
Then φ is countably additive on A.
Proof. It is enough to consider the case f ≥ 0. The general case follows from the
decomposition f = f
+
−f
−
.
If f = χ
E
for some E ∈ A then
µ(A ∩ E) =
_
A
χ
E
dµ
and σadditivity of φ is the same as this property of µ.
Let f(x) =
n
k=1
c
k
χ
E
k
(x),
n
k=1
E
k
= X. Then for A =
∞
i=1
A
i
, A
i
∈ A we have
φ(A) =
_
A
fdµ =
_
fχ
A
dµ =
n
k=1
c
k
µ(E
k
∩ A)
=
n
k=1
c
k
µ(E
k
∩ (
∞
_
i=1
A
i
)) =
n
k=1
c
k
µ(
∞
_
i=1
(E
k
∩ A
i
))
=
n
k=1
c
k
∞
i=1
µ(E
k
∩ A
i
) =
∞
i=1
n
k=1
c
k
µ(E
k
∩ A
i
)
(the series of positive numbers)
=
∞
i=1
_
A
i
fdµ =
∞
i=1
φ(A
i
).
Now consider general positive f’s. Let ϕ be a simple measurable function and ϕ ≤ f.
Then
_
A
ϕdµ =
∞
i=1
_
A
i
ϕdµ ≤
∞
i=1
φ(A
i
).
Therefore the same inequality holds for sup, hence
φ(A) ≤
∞
i=1
φ(A
i
).
Now if for some i φ(A
i
) = +∞ then φ(A) = +∞ since φ(A) ≥ φ(A
n
). So assume that
φ(A
i
) < ∞∀i. Given ε > 0 choose a measurable simple function ϕ such that ϕ ≤ f and
_
A
1
ϕdµ ≥
_
A
1
fdµ −ε,
_
A
2
ϕdµ ≥
_
A
2
f −ε.
30
Hence
φ(A
1
∪ A
2
) ≥
_
A
1
∪A
2
ϕdµ =
_
A
1
+
_
A
2
ϕdµ ≥ φ(A
1
) + φ(A
2
) −2ε,
so that φ(A
1
∪ A
2
) ≥ φ(A
1
) + φ(A
2
).
By induction
φ(
n
_
i=1
A
i
) ≥
n
i=1
φ(A
i
).
Since A ⊃
n
i=1
A
i
we have that
φ(A) ≥
n
i=1
φ(A
i
).
Passing to the limit n → ∞ in the RHS we obtain
φ(A) ≥
∞
i=1
φ(A
i
).
This completes the proof.
Corollary. If A ∈ A, B ⊂ A and µ(A ¸ B) = 0 then
_
A
fdµ =
_
B
fdµ.
Proof.
_
A
fdµ =
_
B
fdµ +
_
A\B
fdµ =
_
B
fdµ + 0.
Deﬁnition 7.5 f and g are called equivalent (f ∼ g in writing) if µ(¦x : f(x) ,=
g(x)¦) = 0.
It is not hard to see that f ∼ g is relation of equivalence.
(i) f ∼ f , (ii) f ∼ g, g ∼ h ⇒ f ∼ h, (iii) f ∼ g ⇔ g ∼ f.
Theorem 7.6 If f ∈ L
1
then [f[ ∈ L
1
and
¸
¸
¸
¸
_
A
fdµ
¸
¸
¸
¸
≤
_
A
[f[dµ
31
Proof.
−[f[ ≤ f ≤ [f[
Theorem 7.7 (Monotone Convergence Theorem)
Let (f
n
) be nondecreasing sequence of nonnegative measurable functions with limit f. Then
_
A
fdµ = lim
n→∞
_
A
f
n
dµ, A ∈ A
Proof. First, note that f
n
(x) ≤ f(x) so that
lim
n
_
A
f
n
dµ ≤
_
fdµ
It is remained to prove the opposite inequality.
For this it is enough to show that for any simple ϕ such that 0 ≤ ϕ ≤ f the following
inequality holds
_
A
ϕdµ ≤ lim
n
_
A
f
n
dµ
Take 0 < c < 1. Deﬁne
A
n
= ¦x ∈ A : f
n
(x) ≥ cϕ(x)¦
then A
n
⊂ A
n+1
and A =
∞
n=1
A
n
.
Now observe
c
_
A
ϕdµ =
_
A
cϕdµ = lim
n→∞
_
A
n
cϕdµ ≤
(this is a consequence of σadditivity of φ proved above)
≤ lim
n→∞
_
A
n
f
n
dµ ≤ lim
n→∞
_
A
f
n
dµ
Pass to the limit c → 1.
Theorem 7.8 Let f = f
1
+ f
2
, f
1
, f
2
∈ L
1
(µ). Then f ∈ L
1
(µ) and
_
fdµ =
_
f
1
dµ +
_
f
2
dµ
32
Proof. First, let f
1
, f
2
≥ 0. If they are simple then the result is trivial. Otherwise, choose
monotonically increasing sequences (ϕ
n,1
), (ϕ
n,2
) such that ϕ
n,1
→ f
1
and ϕ
n,2
→ f
2
.
Then for ϕ
n
= ϕ
n,1
+ ϕ
n,2
_
ϕ
n
dµ =
_
ϕ
n,1
dµ +
_
ϕ
n,2
dµ
and the result follows from the previous theorem.
If f
1
≥ 0 and f
2
≤ 0 put
A = ¦x : f(x) ≥ 0¦, B = ¦x : f(x) < 0¦
Then f, f
1
and −f
2
are nonnegative on A.
Hence
_
A
f
1
=
_
A
fdµ +
_
A
(−f
2
)dµ
Similarly
_
B
(−f
2
)dµ =
_
B
f
1
dµ +
_
B
(−f)dµ
The result follows from the additivity of integral.
Theorem 7.9 Let A ∈ A, (f
n
) be a sequence of nonnegative measurable functions and
f(x) =
∞
n=1
f
n
(x), x ∈ A
Then
_
A
fdµ =
∞
n=1
_
A
f
n
dµ
Exercise. Prove the theorem.
Theorem 7.10 (Fatou’s lemma)
If (f
n
) is a sequence of nonnegative measurable functions deﬁned a.e. and
f(x) = lim
n→∞
f
n
(x)
then
_
A
fdµ ≤ lim
n→∞
_
A
f
n
dµ
A ∈ A
33
Proof. Put g
n
(x) = inf
i≥n
f
i
(x)
Then by deﬁnition of the lower limit lim
n→∞
g
n
(x) = f(x).
Moreover, g
n
≤ g
n+1
, g
n
≤ f
n
. By the monotone convergence theorem
_
A
fdµ = lim
n
_
A
g
n
dµ = lim
n
_
A
g
n
dµ ≤ lim
n
_
A
f
n
dµ.
Theorem 7.11 (Lebesgue’s dominated convergence theorem)
Let A ∈ A, (f
n
) be a sequence of measurable functions such that f
n
(x) → f(x) (x ∈ A.)
Suppose there exists a function g ∈ L
1
(µ) on A such that
[f
n
(x)[ ≤ g(x)
Then
lim
n
_
A
f
n
dµ =
_
A
fdµ
Proof. From [f
n
(x)[ ≤ g(x) it follows that f
n
∈ L
1
(µ). Sinnce f
n
+ g ≥ 0 and f + g ≥ 0,
by Fatou’s lemma it follows
_
A
(f + g)dµ ≤ lim
n
_
A
(f
n
+ g)
or
_
A
fdµ ≤ lim
n
_
A
f
n
dµ.
Since g −f
n
≥ 0 we have similarly
_
A
(g −f)dµ ≤ lim
n
_
A
(g −f
n
)dµ
so that
−
_
A
fdµ ≤ −lim
n
_
A
f
n
dµ
which is the same as
_
A
fdµ ≥ lim
n
_
A
f
n
dµ
This proves that
lim
n
_
A
f
n
dµ = lim
n
_
A
f
n
dµ =
_
A
fdµ.
34
8 Comparison of the Riemann
and the Lebesgue integral
To distinguish we denote the Riemann integral by (R)
_
b
a
f(x)dx and the Lebesgue integral
by (L)
_
b
a
f(x)dx.
Theorem 8.1 If a ﬁnction f is Riemann integrable on [a, b] then it is also Lebesgue
integrable on [a, b] and
(L)
_
b
a
f(x)dx = (R)
_
b
a
f(x)dx.
Proof. Boundedness of a function is a necessary condition of being Riemann integrable.
On the other hand, every bounded measurable function is Lebesgue integarble. So it is
enough to prove that if a function f is Riemann integrable then it is measurable.
Consider a partition π
m
of [a, b] on n = 2
m
equal parts by points a = x
0
< x
1
< . . . <
x
n−1
< x
n
= b and set
f
m
(x) =
2
m
−1
k=0
m
k
χ
k
(x), f
m
(x) =
2
m
−1
k=0
M
k
χ
k
(x),
where χ
k
is a charactersitic function of [x
k
, x
k+1
) clearly,
f
1
(x) ≤ f
2
(x) ≤ . . . ≤ f(x),
f
1
(x) ≥ f
2
(x) ≥ . . . ≥ f(x).
Therefore the limits
f(x) = lim
m→∞
f
m
(x), f(x) = lim
m→∞
f
m
(x)
exist and are measurable. Note that f(x) ≤ f(x) ≤ f(x). Since f
m
and f
m
are simple
measurable functions, we have
(L)
_
b
a
f
m
(x)dx ≤ (L)
_
b
a
f(x)dx ≤ (L)
_
b
a
f(x)dx ≤ (L)
_
b
a
f
m
(x)dx.
Moreover,
(L)
_
b
a
f
m
(x)dx =
2
m
−1
k=0
m
k
∆x
k
= s(f, π
m
)
35
and similarly
(L)
_
b
a
f
m
(x) = s(f, π
m
).
So
s(f, π
m
) ≤ (L)
_
b
a
f(x)dx ≤ (L)
_
b
a
f(x)dx ≤ s(f, π
m
).
Since f is Riemann integrable,
lim
m→∞
s(f, π
m
) = lim
m→∞
s(f, π
m
) = (R)
_
b
a
f(x)dx.
Therefore
(L)
_
b
a
(f(x) −f(x))dx = 0
and since f ≥ f we conclude that
f = f = f almost everywhere.
From this measurability of f follows.
36
9 L
p
spaces
Let (X, A, µ) be a measure space. In this section we study L
p
(X, A, µ)spaces which occur
frequently in analysis.
9.1 Auxiliary facts
Lemma 9.1 Let p and q be real numbers such that p > 1,
1
p
+
1
q
= 1 (this numbers are
called conjugate). Then for any a > 0, b > 0 the inequality
ab ≤
a
p
p
+
b
q
q
.
holds.
Proof. Note that ϕ(t) :=
t
p
p
+
1
q
−t with t ≥ 0 has the only minimum at t = 1. It follows
that
t ≤
t
p
p
+
1
q
.
Then letting t = ab
−
1
p−1
we obtain
a
p
b
−q
p
+
1
q
≥ ab
−
1
p−1
,
and the result follows.
Lemma 9.2 Let p ≥ 1, a, b ∈ R. Then the inequality
[a + b[
p
≤ 2
p−1
([a[
p
+[b[
p
).
holds.
Proof. For p = 1 the statement is obvious. For p > 1 the function y = x
p
, x ≥ 0 is convex
since y
≥ 0. Therefore
_
[a[ +[b[
2
_
p
≤
[a[
p
+[b[
p
2
.
37
9.2 The spaces L
p
, 1 ≤ p < ∞. Deﬁnition
Recall that two measurable functions are said to be equaivalent (with respect to the
measure µ) if they are equal µa;most everywhere.
The space L
p
= L
p
(X, A, µ) consists of all µequaivalence classes of Ameasurable
functions f such that [f[
p
has ﬁnite integral over X with respect to µ.
We set
f
p
:=
__
X
[f[
p
dµ
_
1/p
.
9.3 H¨older’s inequality
Theorem 9.3 Let p > 1,
1
p
+
1
q
= 1. Let f and g be measurable functions, [f[
p
and [g[
q
be integrable. Then fg is integrable andthe inequality
_
X
[fg[dµ ≤
__
X
[f[
p
dµ
_
1/p
__
X
[g[
q
dµ
_
1/q
.
Proof. It suﬃces to consider the case
f
p
> 0, g
q
> 0.
Let
a = [f(x)[f
−1
p
, b = [g(x)[g
−1
q
.
By Lemma 1
[f(x)g(x)[
f
p
g
q
≤
[f(x)[
p
pf
p
p
+
[g(x)[
q
qg
q
q
.
After integration we obtain
f
−1
p
g
−1
q
_
X
[fg[dµ ≤
1
p
+
1
q
= 1.
9.4 Minkowski’s inequality
Theorem 9.4 If f, g ∈ L
p
, p ≥ 1, then f + g ∈ L
p
and
f + g
p
≤ f
p
+g
p
.
38
Proof. If f
p
and g
p
are ﬁnite then by Lemma 2 [f +g[
p
is integrable and f +g
p
is welldeﬁned.
[f(x)+g(x)[
p
= [f(x)+g(x)[[f(x)+g(x)[
p−1
≤ [f(x)[[f(x)+g(x)[
p−1
+[g(x)[[f(x)+g(x)[
p−1
.
Integratin the last inequality and using H¨older’s inequality we obtain
_
X
[f+g[
p
dµ ≤
__
X
[f[
p
dµ
_
1/p
__
X
[f + g[
(p−1)q
dµ
_
1/q
+
__
X
[g[
p
dµ
_
1/p
__
X
[f + g[
(p−1)q
dµ
_
1/q
.
The result follows.
9.5 L
p
, 1 ≤ p < ∞, is a Banach space
It is readily seen from the properties of an integral and Theorem 9.3 that L
p
, 1 ≤ p < ∞,
is a vector space. We introduced the quantity f
p
. Let us show that it deﬁnes a norm
on L
p
, 1 ≤ p < ∞,. Indeed,
1. By the deﬁnition f
p
≥ 0.
2. f
p
= 0 =⇒ f(x) = 0 for µalmost all x ∈ X. Since L
p
consists of µeqivalence
classes, it follows that f ∼ 0.
3. Obviously, αf
p
= [α[f
p
.
4. From Minkowski’s inequality it follows that f + g
p
≤ f
p
+g
p
.
So L
p
, 1 ≤ p < ∞, is a normed space.
Theorem 9.5 L
p
, 1 ≤ p < ∞, is a Banach space.
Proof. It remains to prove the completeness.
Let (f
n
) be a Cauchy sequence in L
p
. Then there exists a subsequence (f
n
k
)(k ∈ N)
with n
k
increasing such that
f
m
−f
n
k

p
<
1
2
k
∀m ≥ n
k
.
Then
k
i=1
f
n
i+1
−f
n
i

p
< 1.
39
Let
g
k
:= [f
n
1
[ +[f
n
2
−f
n
1
[ + . . . +[f
n
k+1
−f
n
k
[.
Then g
k
is monotonocally increasing. Using Minkowski’s inequality we have
g
p
k

1
= g
k

p
p
≤
_
f
n
1

p
+
k
i=1
f
n
i+1
−f
n
i

p
_
p
< (f
n
1

p
+ 1)
p
.
Put
g(x) := lim
k
g
k
(x).
By the monotone convergence theorem
lim
k
_
X
g
p
k
dµ =
_
A
g
p
dµ.
Moreover, the limit is ﬁnite since g
p
k

1
≤ C = (f
n
1

p
+ 1)
p
.
Therefore
[f
n
1
[ +
∞
j=1
[f
n
j+1
−f
n
j
[ converges almost everywhere
and so does
f
n
1
+
∞
j=1
(f
n
j+1
−f
n
j
),
which means that
f
n
1
+
N
j=1
(f
n
j+1
−f
n
j
) = f
n
N+1
converges almost everywhere as N → ∞.
Deﬁne
f(x) := lim
k→∞
f
n
k
(x)
where the limit exists and zero on the complement. So f is measurable.
Let > 0 be such that for n, m > N
f
n
−f
m

p
p
=
_
X
[f
n
−f
m
[
p
dµ < /2.
Then by Fatou’s lemma
_
X
[f −f
m
[
p
dµ =
_
X
lim
k
[f
n
k
−f
m
[
p
dµ ≤ lim
k
_
X
[f
n
k
−f
m
[
p
dµ
which is less than for m > N. This proves that
f −f
m

p
→ 0 as m → ∞.
40
Clearly,
b b
s(f, π) ≤
a
f dx ≤
a
f dx ≤ s(f, π) ¯
for any partition π. The function f is said to be Riemann integrable on [a, b] if the upper and lower integrals are equal. The common value is called Riemann integral of f on [a, b]. The functions cannot have a large set of points of discontinuity. More presicely this will be stated further.
1.2
The Lebesgue integral
It allows to integrate functions from a much more general class. First, consider a very useful example. For f, g ∈ C[a, b], two continuous functions on the segment [a, b] = {x ∈ R : a x b} put ρ1 (f, g) = max f (x) − g(x),
a x b b
ρ2 (f, g) =
a
f (x) − g(x)dx.
Then (C[a, b], ρ1 ) is a complete metric space, when (C[a, b], ρ2 ) is not. To prove the latter statement, consider a family of functions {ϕn }∞ as drawn on Fig.1. This is a Cauchy n=1 sequence with respect to ρ2 . However, the limit does not belong to C[a, b].
2
T
.
.
.
.
.
.
.
b). there exists A ∈ K. Let X = R and denote by K all ﬁnite unions of semisegments [a. Let K be a ring of sets. Since K contains the diﬀerence of every two its elements.
3
. 1. 2.1 A ring of sets is a nonempty subset in 2X which is closed with respect to the operations ∪ and \. Deﬁnition 2.
1 −2
L L L L L L L L L
1 n 1 2
E
−1 + 2
−
1 n
1 2
Figure 1: The function ϕn . one has A \ A = ∅ ∈ K. Since K = ∅. The two extreme cases are K = {∅} and K = 2X . Examples. Proof.
2
Systems of Sets
Deﬁnition 2. Proposition. If A. Then ∅ ∈ K. B ∈ P then A ∩ B ∈ P.2 A semiring is a collection of sets P ⊂ 2X with the following properties: 1.
σalgebras) of sets containing U. (ii) A ∈ A =⇒ Ac ∈ A. 2. then the set of all semisegments.e.5 A ring (an algebra. B ∈ A =⇒ A ∪ B ∈ A. 1) A ∩ B = (Ac ∪ B c )c . forms a semiring. 1. In other words. Deﬁnition 2. For every A. b) be a ﬁxed interval on R. n of sets (i. Examples. Deﬁnition 2. X} and 2X are the two extreme cases (note that they are diﬀerent from the corresponding cases for rings of sets).
4
. . Deﬁnition 2. a σalgebra) of sets. Then the system of ﬁnite unions of subintervals [α.
Example. Ci ∩ Cj = ∅ if i = j) such that
n
A\B =
j=1
Cj . it is the intersection of all rings (algebras.2. b) forms an algebra. Remark. 3. . [a. 2) A \ B = A ∩ B c . A is algebra if (i) A. 2. . β) ⊂ [a. K(U) generated by a collection of sets U ⊂ 2X is the minimal ring (algebra.3 An algebra (of sets) is a ring of sets containing X ∈ 2X . σalgebra) of sets containing U. b). Let X = R.4 A σring (a σalgebra) is a ring (an algebra) of sets which is closed with respect to all countable unions. Let X = [a. Indeed. The system of all bounded subsets of the real axis is a ring (not an algebra). {∅. B ∈ P there exists a ﬁnite disjoint collection (Cj ) j = 1. .
The latter important property. 1. 2. µ(A). (Subtractivity of µ). 3. A ∩ B ⊂ B. therefore A ∪ B = (A \ (A ∩ B)) Since µ(A ∩ B) < ∞. If A.
. A = (A \ B) B implies that µ(A) = µ(A \ B) + µ(B). is called countable additivity or σadditivity of the measure µ. B ∈ A and B ⊂ A and µ(B) < ∞ then µ(A \ B) = µ(A) − µ(B). If µ(B) < ∞ then µ(A) − µ(B) = µ(A \ B).1 A function µ: A −→ R+ ∪ {∞} is called a measure if 1. Since µ(A \ B) ≥ 0 it follows that µ(A) ≥ µ(B). A ∩ B ⊂ A. 5 B. Let us state now some elementary properties of a measure. (Monotonicity of µ) If A. A an algebra on X.
2. Deﬁnition 3.3
Measures
Let X be a set. Below till the end of this section A is an algebra of sets and µ is a measure on it. If A. then
∞ ∞
µ(
i=1
Ai ) =
i=1
µ(Ai ). B ∈ A and B ⊂ A then µ(B) Proof. Proof. In 1) we proved that µ(A) = µ(A \ B) + µ(B). B ∈ A and µ(A ∩ B) < ∞ then µ(A ∪ B) = µ(A) + µ(B) − µ(A ∩ B). one has µ(A ∪ B) = (µ(A) − µ(A ∩ B)) + µ(B). Proof. µ(A) 0 for any A ∈ A and µ(∅) = 0. if (Ai )i 1 is a disjoint family of sets in A ( Ai ∩ Aj = ∅ for any i = j) such that ∞ i=1 Ai ∈ A.
If for some n0 µ(An0 ) = +∞ then µ(An ) = +∞ ∀n ≥ n0 and µ( 2). (Ai )i≥1 ⊂ A a monotonically increasing sequence of sets (Ai ⊂ Ai+1 ) such that i≥1 ∈ A.. we see that µ(Bi ) ≤
n n n
µ(
i=1
Ai ) = µ(
i=1
Bi ) =
i=1
µ(Bi ) ≤
i=1
µ(Ai ).
Proof.
n→∞ ∞ i=1
Proof. First let us proove that
n n
µ(
i=1
Ai )
i=1
µ(Ai ).
Ai ) = +∞.4. Let now µ(Ai ) < ∞ ∀i ≥ 1. 1).
Now we can repeat the argument for the inﬁnite family using σadditivity of the measure. Then
n i=1
Bi =
n
n i=1
Ai . since Bi ⊂ Ai . (Semiadditivity of µ).
n−1
Bn = An \
i=1
Ai
is disjoint and µ(Ai ).
Note that the family of sets B1 = A1 B2 = A2 \ A1 B3 = A3 \ (A1 ∪ A2 ) . Then
∞
µ(
i=1
Ai ) = lim µ(An ).1
Continuity of a measure
Theorem 3.
6
.1 Let A be an algebra.
3.. If (Ai )i≥1 ⊂ A such that
∞ ∞
∞ i=1
Ai ∈ A then
µ(
i=1
Ai )
i=1
µ(Ai ). Moreover.
b) × R1 ) = b − a. Put µ([a. one can always ﬁnd E1 . The outer measure always exists since µ(A) 0 for every A ∈ A.. b) × R1 }. Thus A consists of countable unions of strips like one drawn on the picture. Our purpose now is to extend µ to as many elements of 2X as possible.. For instance. An arbitrary set A ⊂ X can be always covered by sets from A. Let X = R2 .Then
∞
µ(
i=1
Ai ) = µ(A1
(A2 \ A1 )
.
Example. In other words.e. y 1. . clearly. P = {[a. The same value is for the square x 1. Remark.e.
k=2 n→∞
3. . This implies µ∗ (A) µ(A). E2 .2
Outer measure
Let a be an algebra of subsets of X and µ a measure on it. E2 = E3 = .2 For A ∈ A one has µ∗ (A) = µ(A). the outer measure of the unit disc x2 + y 2 1 is equal to 2. Proof. σalgebra generated by P. E1 = X.)
= µ(A1 ) +
k=2 n
µ(Ak \ Ak−1 )
= µ(A1 ) + lim
n→∞
(µ(Ak ) − µ(Ak−1 )) = lim µ(An ). . 1. Ei ∈ A with i Ei ⊃ A. A is its own covering. . all collections (Ei ). i=1 Deﬁnition 3.
7
. By deﬁnition of inﬁmum.2 For A ⊂ X its outer measure is deﬁned by
∞
µ∗ (A) = inf
i=1
µ(Ei )
where the inﬁmum is taken over all Acoverings of the set A. for any ε > 0 there exists a Acovering (Ei ) of A such that ∗ i µ(Ei ) < µ (A) + ε. A = K(P). ∈ A such that ∞ Ei ⊃ A. . i. Then. i. = ∅. Note that A=A∩(
i
Ei ) =
i
(A ∩ Ei ). Theorem 3. µ∗ is an extension of µ.
2.
(An \ An−1 )
∞
. . .
Aj )
∞ j=1
µ∗ (Aj ). then µ(∅) = 0.
µ∗ (∅) = 0 (Check !).T
a b
E
Using consequently σsemiadditivity and monotonicity of µ.
8
.3 (Monotonicity of outer measure). µ∗ (A). we conclude that µ(A) It is evident that µ∗ (A) 0. Let A be an algebra of sets (not necessary σalgebra). µ(A \ A) = µ(A) − µ(A) = 0.4 (σsemiadditivity of µ∗ ). µ a measure on A. µ∗ (
∞ j=1
µ∗ (B). If A ⊂ B then µ∗ (A) Proof. Proof.
Lemma. Any covering of B is a covering of A. Theorem 3. one obtains: µ(A)
i
µ(A ∩ Ei )
i
µ(Ei ) < µ∗ (A) + ε. Therefore the property µ(∅) = 0 can be substituted with the existence in A of a set with a ﬁnite measure. Theorem 3. If there exists a set A ∈ A such that µ(A) < ∞.
Since ε is arbitrary.
˜ Remark Since E = (E ∩ A) ∪ (E ∩ Ac ).k=1 ∞
µ(Ekj )
j=1
and therefore µ∗ (
∞
Aj ) <
j=1
µ∗ (Aj ) + ε. 2 k=1 Since
∞ ∞
Ekj ⊃
j.3 A ⊂ X is called a measurable set (by Carath`odory) if for any E ⊂ X e the following relation holds: µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). Deﬁnition 3. So assume that it is convergent.3
Measurable Sets
Let A be an algebra of subsets of X. ˜ Denote by A the collection of all set which are measurable by Carath`odory and set e ∗ ˜ µ = µ A. due to semiadditivity of the outer measure µ∗ (E) ≤ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). there is nothing to prove. and µ is a measure on A.
j=1
3.
the deﬁnition of µ implies µ(
∗
∗
∞
Aj )
j. ˜
9
. If the series in the righthand side diverges. µ∗ the outer measure deﬁned in the previous section. By the deﬁnition of outer measur for any ε > 0 and for any j there exists an Acovering k Ekj ⊃ Aj such that ∞ ε µ(Ekj ) < µ∗ (Aj ) + j . ˜ ˜ Theorem 3. µ a measure on it.k=1 j=1 ∞
Aj .5 A is a σalgebra containing A.Proof.
Let A. ˜ Therefore A is an algebra of sets. B ∈ A then A ∪ B ∈ A. By the deﬁnition one has µ∗ (E) = µ∗ (E ∩ B) + µ∗ (E ∩ B c ). A ∩ B = ∅. If A ∈ A then Ac ∈ A. Take E ∩ A instead of E: µ∗ (E ∩ A) = µ∗ (E ∩ A ∩ B) + µ∗ (E ∩ A ∩ B c ). Note that 1) 2) 3) 4) One has µ∗ (E ∩ (A ∪ B)) = µ∗ (E ∩ A ∩ B) + µ∗ (E ∩ Ac ∩ B) + µ∗ (E ∩ A ∩ B c ). Add (2) and (3): µ∗ (E) = µ∗ (E ∩ A ∩ B) + µ∗ (E ∩ A ∩ B c ) + µ∗ (E ∩ Ac ∩ B) + µ∗ (E ∩ Ac ∩ B c ). If A. ˜ ˜ 1. Then put E ∩ Ac in (1) instead of E µ∗ (E ∩ Ac ) = µ∗ (E ∩ Ac ∩ B) + µ∗ (E ∩ Ac ∩ B c ).Proof. 3. From (5) µ∗ (E ∩ (A B)) = µ∗ (E ∩ Ac ∩ B) + µ∗ (E ∩ A ∩ B c ) = µ∗ (E ∩ B) + µ∗ (E ∩ A). B ∈ A. The deﬁnition of measurable set is symmetric with respect to A and Ac . Substitute E ∩ (A ∪ B) in (4) instead of E. (4) (3) (2) (1)
. 10 (5) E ∩ (A ∪ B) ∩ A ∩ B = E ∩ A ∩ B E ∩ (A ∪ B) ∩ Ac ∩ B = E ∩ Ac ∩ B E ∩ (A ∪ B) ∩ A ∩ B c = E ∩ A ∩ B c E ∩ (A ∪ B) ∩ Ac ∩ B c = ∅. From (4) and (5) we have µ∗ (E) = µ∗ (E ∩ (A ∪ B)) + µ∗ (E ∩ (A ∪ B)c ). ˜ ˜ 2. We devide the proof into several steps.
Then A = ∞ Bj . µ = µ∗ A is a measure. ˜ ˜ Since A is an algebra.˜ 4.
(8)
Bj )c ⊂ E ∩ (
∗
Bj )c . which means that A is a σalgebra. Aj ∈ A. A is a σalgebra. ˜ 11
.
(6)
Let A = ∞ Aj . ˜ ˜ Thus. by monotonicity of the mesasure and (8)
n
µ (E) ≥
j=1
µ∗ (E ∩ Bj ) + µ∗ (E ∩ Ac ). Bj = Aj \ j=1 j=1 Bi ∩ Bj = ∅ (i = j). we have already proved that µ∗ is σsemiadditive.
Compare this with (10): µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). It suﬃces to prove that
∞ ∞
j−1 k=1
Ak and
µ (E)
∗
µ (E ∩ (
j=1
∗
Bj )) + µ (E ∩ (
j=1
∗
Bj )c ).
(10)
Due to semiadditivity
∞ ∞ ∞
µ∗ (E ∩ A) = µ∗ (E ∩ (
j=1
Bj )) = µ∗ (
j=1
(E ∩ Bj )) ≤
j=1
µ∗ (E ∩ Bj ). From the previous step. it follows that n Bj ∈ A(∀n ∈ N) and the following inequality j=1 holds for every n:
n n
µ (E) Since E ∩ (
∞ j=1
∗
µ (E ∩ (
j=1 n j=1
∗
Bj )) + µ (E ∩ (
j=1
∗
Bj )c ).
(9)
Passing to the limit we get
∞
µ (E) ≥
j=1
∗
µ∗ (E ∩ Bj ) + µ∗ (E ∩ Ac ). ˜ 5. for any ﬁnite disjoint collection (Bj ) of sets:
n n
µ (E ∩ (
j=1
∗
Bj )) =
j=1
µ∗ (E ∩ Bj ). A ∈ A.
(7)
Indeed. by induction.
˜ 6.
The oposite inequality follows from σsemiadditivity of µ∗ . Let E =
∞ ∞
∞ j=1
Aj .We need to prove only σadditivity. 12
. We need to prove: µ∗ (E) µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). The following theorem is a direct consequence of the previous one. From(10) we get
µ(
j=1
∗
Aj )
j=1
µ∗ (Aj ).
Adding the last two inequalities we obtain
∞
µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ≤
j=1
µ∗ (Ej ) < µ∗ (E) + ε. one has µ(Ej ) = µ(Ej ∩ A) + µ(Ej ∩ A) and also
∞
E∩A⊂
j=1 ∞
(Ej ∩ A) (Ej ∩ Ac )
E ∩ Ac ⊂
j=1
By monotonicity and σsemiadditivity
∞
µ∗ (E ∩ A)
j=1 ∞
µ(Ej ∩ A).
(12)
Now.
µ∗ (E ∩ Ac )
j=1
µ(Ej ∩ Ac ). For E ⊂ X for ∀ε > 0 there exists a Acovering (Ej ) of E such that
∞
µ∗ (E) + ε >
j=1
µ(Ej ). A ⊃ A. since Ej = (Ej ∩ A) ∪ (Ej ∩ Ac ). (11)
If E ∈ A then (11) is clear since E ∩ A and E ∩ Ac are disjoint and both belong to A where µ∗ = µ and so is additive.
Since ε > 0 is arbitrary. (11) is proved. E ⊂ X. Let A ∈ A.
Theorem 3. Denot by Aσ the generated σalgebra and construct the extension µσ of µ on Aσ . It suﬃces to sjow uniqueness.Theorem 3. µ∗ ) be ﬁnite.8 Let µ be a σﬁnite measure on an algebra A. Proof. Consider again an algebra A of subsets of X. Deﬁnition 3. µ(A) = 0 imply B ∈ A and µ(B) = 0.7 Let µ be a measure on an algebra A of subsets of X. ˜ ˜ Since A ⊃ A therefore Aσ ⊂ A. Obviously µσ is a ˜ minimal extension of µ. Let (Ej ) ⊂ A such that A ⊂ j Ej . Let A ∈ A. The latter statement follows from monotonicity of µ∗ . let µ (and therefore ν. Clearly. Hence one can set µσ = µ Aσ . 13
.
Therefore ˜ ν(A) ≤ µ∗ (A) ∀A ∈ A. Corollary. ˜ Deﬁnition 3. Indeed. it suﬃces to prove that A ∈ A. On can also show (see below) that this extension is unique. µ∗ the corresponding ˜ outer measure. This extension is called minimal extension of measure. We have
∞ ∞ ∞
ν(A) ≤ ν(
j=1
Ej ) ≤
j=1
ν(Ej ) =
j=1
µ(Ej ). Remark.6 Let A be an algebra of subsets of X and µ be a measure on it.5 A measure µ on an algebra A is called ﬁnite if µ(X) < ∞. Further. one has µ∗ (E ∩ A) µ∗ (A) = 0 and µ∗ (E ∩ Ac ) µ∗ (E). A ∈ A. Theorem 3. If µ∗ (A) = 0 for a set A ⊂ X then A ∈ A and µ(A) = 0. It is called σﬁnite if the is an increasing sequence (Fj )j≥1 ⊂ A such that X = j Fj and µ(Fj ) < ∞ ∀j. ˜ ˜ Proof.4 A measure µ on an algebra of sets A is called complete if conditions B ⊂ A. it suﬃces to prove that µ∗ (E) ∗ ∗ c µ (E ∩ A) + µ (E ∩ A ). µ is a complete measure. Then there exists a σalgebra A1 ⊃ A and a measure µ1 on A1 such that µ1 A = µ. Then there exist a unique ˜ extension of µ to a measure on A. It always exists. Let ν be another extension of µ (ν A = µ A). ˜ First.
{Aj } ⊂ A a monotonically increasing sequence of sets. ˜ 2 2
14
. From this we infer that ˜ ν(A) = µ∗ (A) ∀A ∈ A.˜ Since µ∗ and ν are additive (on A) it follows that µ∗ (A) + µ∗ (Ac ) = ν(A) + ν(Ac ). Therefore µ∗ (A) = lim µ∗ (A ∩ Fj ) = lim ν(A ∩ Fj ) = ν(A). Then for any ε > 0 there exists Aε ∈ A such that µ∗ (A Aε ) < ε. From what we have already proved it follows that j=1 ˜ µ∗ (A ∩ Fj ) = ν(A ∩ Fj ) ∀A ∈ A. Proof.
∞
µ(A) =
j=1
(µ(Aj+1 ) − µ(Aj )) + µ(A1 ) = lim µ(Aj ). The terms in the RHS are ﬁnite and ν(A) ≤ µ∗ (A). ν(Ac ) ≤ µ∗ (Ac ).
j→∞
Proof. Now let µ be σﬁnite.
j j
Theorem 3.10 Let A ∈ A. (Fj ) be an increasing sequence of sets from A such that µ(Fj ) < ∞ ∀j and X = ∞ Fj .9 (Continuity of measure). For any ε > 0 there exists an A cover µ(Ej ) < µ∗ (A) +
j
Ej ⊃ A such that ε ε = µ(A) + .
j→∞
Similar assertions for a decreasing sequence of sets in A can be proved using de Morgan formulas. Let A be a σalgebra with a measure µ. Then
∞
µ(
j=1
Aj ) = lim µ(Aj ). 1. ˜ Theorem 3.
Using σadditivity and subtractivity of µ. One has: A=
∞
∞
Aj =
j=1 j=2
(Aj+1 \ Aj )
A1 .
The monotonicity of µ implies ˜
∞ n
µ( ˜
j=1
EJ ) = lim µ( ˜
n→∞ j=1
Ej ). ˜ 2 j=1
∞
.
j=1 ∞
Since Aε ⊂
Ej . put
N
Aε =
j=1
Ej
and prove that µ∗ (A 2a. 2 j=1
N
(13)
2. Together with estimate (13).
∞
one has A \ Aε ⊂
Ej \ Aε .On the other hand.
hence there exists a positive integer N such that
∞
µ( ˜
j=1
Ej ) − µ( ˜
ε Ej ) < . Now. this ˜ gives ∞ ε µ(A \ Aε ) ≤ µ( Ej \ Aε ) < . Since
Aε ) < ε.
j=1
one can use the monotonicity and subtractivity of µ.
∞
A⊂
j=1
Ej . The inclusion Aε \ A ⊂
j=1 ∞
Ej \ A
implies µ(Aε \ A) ˜ µ( ˜
∞
Ej \ A) = µ( ˜ 15
j=1
ε Ej ) − µ(A) < . µ(Ej )
j
µ( ˜
j
EJ ). ˜ ˜ 2 j=1 2b.
˜ ˜
16
.Here we used the same properties of µ as above and the choice of the cover (Ej ). ˜ 3. Finally. µ(A ˜ Aε ) µ(A \ Aε ) + µ(Aε \ A).
2. This follows from the Exercise (2) above and from the minimality of Kσ .
In the same way. below). Prove that the union. Prove that any σring is a monotone class. B \ A ∈ M(K)}. Kσ the σring generated by K. Let (Aj ) ⊂ KB be a monotonically increasing sequence. (This follows from the Exercise 1. one has Aj ∪ B ∈ KB .
∞ ∞
A\B =(
j=1
Aj ) \ B =
j=1
(Aj \ B) ∈ M(K). If a ring is a monotone class. Any σring. Exercises. 1. A ∩ B.
17
. 2a. Then M(K) = Kσ .4
Monotone Classes and Uniqueness of Extension of Measure
Deﬁnition 4. This deﬁnition is symmetric with respect to A and B. 2. M is called a monotone class if together with any monotone sequence of sets M contains the limit of this sequence. set KB = {A ⊂ X : A ∪ B. 1. Example.1 Let K be a ring of sets. Clearly. Now. Proof. Theorem 4. Since Aj ∈ KB . therefore A ∈ KB implies B ∈ KA . A = belongs to KB . KB is a monotone class. 2b. it suﬃces to prove that M(K) is a ring.1 A collection of sets. For B ⊂ X. and so
∞
Aj
A∪B =
j=1
(Aj ∪ B) ∈ M(K). A \ B. We shall denote by M(K) the minimal monotone class containing K. M(K) ⊂ Kσ . M(K) is a ring. then it is a σring.
This implies M(K) ⊂ KB . 1. Then M(K) ⊂ KA . then A = limn→∞ An . that if A.∞
∞
B\A=B\(
j=1
Aj ) =
j=1
(B \ Aj ) ∈ M(K). Since A ⊂ A. — 2d. for Aσ = M(A). so A ∈ KB .3 Let A be an algebra of sets.
n→∞ n→∞
Theorem 4. 2d. Since any outer measure is semiadditive. Together with minimality of M(K). Let A ∈ K.2 Let A be an algebra of sets. 2c. one has B ∈ KA . ˜ 2a. Since A ⊂ B ∪ (A B) and since the outer measure µ∗ is monotone and semiadditive. Thus if B ∈ M(K). A \ B and B \ A all belong to M(K). A ∩ B. (14)
2b. B ⊂ X. Choose A ∈ Aσ .
Similar proof is for the case of decreasing sequence (Aj ). An ∈ A. B ⊂ X such that for any ε > 0 there exists ˜ Aε ∈ A with µ∗ (B Aε ) < ε. one has µ(A) = lim µ(An ) = lim ν(An ) = ν(A). If B ∈ K then M(K) ⊂ KB . it suﬃces to prove that for any E ⊂ X one has µ∗ (E) µ∗ (E ∩ B) + µ∗ (E ∩ B c ). there is an estimate µ∗ (A) − µ∗ (B) µ∗ (A B) for any A. Proof. Then B ∈ A.
. 2c. Obviously. one has µ∗ (E ∩ Aε ) + µ∗ (E ∩ Ac ) ε µ∗ (E). If B ∈ M(K) then M(K) ⊂ KB .f. B ∈ M(K) then A ∈ KB and so A ∪ B. Then µ A = ν A implies µ = ν. Using continuity of measure. the proof of similar fact for measures above). Hence what we have proved is K ⊂ KB . Theorem 4. It follows from 2a. this implies M(K) ⊂ KB . µ and ν two measures deﬁned on the σalgebra Aσ generated by A. K ⊂ KB . It follows from the monotonicity of µ∗ that µ∗ (E ∩ Aε ) − µ∗ (E ∩ B) µ∗ ((E ∩ Aε ) 18 (E ∩ B)) µ(Aε ∩ B) < ε. (C. 2e. Proof.
Using (14). ε 2d. µ∗ (E ∩ Aε ) > µ∗ (E ∩ B) − ε.
19
. one obtains µ∗ (E) > µ∗ (E ∩ B) + µ∗ (E ∩ B c ) − 2ε. µ∗ (E ∩ Ac ) > µ∗ (E ∩ B c ) − ε. In the same manner.Therefore.
Let (Aj ) ⊂ A be such a countable disjoint family that
∞
A=
j=1
Aj ∈ A. let Aε a set obtained from A by the following construction. Now. to obtain a (closed) segment.5
5. t).
j=1
hence
n
µ(Aj )
j=1
µ(A).e. Proof. bj ). (15) 20
. Do it with all components of A. in such a way that µ(A) < µ(Aε ) + ε. All properties including the (ﬁnite) additivity are obvious. i.1
The Lebesgue Measure on the real line R1
The Lebesgue Measure of Bounded Sets of R1
Put A for the algebra of all ﬁnite unions of semisegments (semiintervals) on R1 . Take a connected component of A.
n
Aj ⊂ A.1 µ is a measure.
Theorem 5. The only thing to be proved is the σadditivity. all sets of the form
k
A=
j=1
[aj .
n
and
∞
µ(Aj ) = lim
j=1
n→∞
µ(Aj )
j=1
µ(A).
Deﬁne a mapping µ : A −→ R by:
k
µ(A) =
j=1
(bj − aj ).
3.
The condition A ∈ A means that 2. Shift slightly on the left its righthand end. For any positive integer n.
Aj is a ﬁnite union of intervals. 1. It is a semisegment of the form [s.
Exercises. Aε with j µ(Aε ) < µ(Aj ) + j ε . Any open or closed set in R1 is Lebesgue measurable.
j=1
Now. 2j
thus µ(A) <
µ(Aj ) + 2ε. and its Lebesgue measure is equal to 0. 1]) = 0. 21
. 3. and obtain the meae ˜ µ). Then for any ε > 0 there exists an open set G ⊃ E such that m(G \ E) < ε. Aε is a compact set and (Aε ) its open cover. Deﬁnition 5.1 Borel algebra of sets. j
j=1
Thus µ(Aε )
n
µ(Aε ). 2j (16)
4. Any element of B is called a Borel set. Hence. A one point set is measurable. one can apply the Carath`odory’s scheme developed above. ˜ denote the Lebesgue measure on R1 by m. The result of this extension is called the Lebesgue measure. We shall sure space (A. Theorem 5. there exists j a positive integer n such that
n
Aε ⊃ Aε .2 Let E ⊂ R1 be a Lebesgue measurable set. j
j=1
The formulas (15) and (16) imply
n n n
µ(A) <
j=1
µ(Aε ) j
+ε
j=1 ∞
µ(Aj ) +
j=1
ε + ε. Exercise. 2. 1. m(Q ∩ [0. Any Borel set is Lebesgue measurable. bj ). B on the real line R1 is a σalgebra generated by all open sets on R1 . By the construction. and obtain (open) intervals. The same for a countable subset in R1 . In particular.Apply a similar procedure to each semisegment shifting their left end point to the left Aj = [aj .
put G= (ak − ε 2k+1 . Problem.2
The Lebesgue Measure on the Real Line R1
We now abolish the condition of boundness. Corollary.4 A measure is called σﬁnite if any measurable set can be represented as a countable union of subsets each has a ﬁnite measure. Deﬁnition 5.
Problem. 2 Now. Thus the Lebesgue measure m is σﬁnite. According the deﬁnition of an outer measure. For any ε > 0 there exist an open set G and a compact set F such that G ⊃ E ⊃ F and m(G \ F ) < ε. Since E is measurable.2 A set A on the real numbers line R1 is Lebesgue measurable if for any positive integer n the bounded set A ∩ [−n. Deﬁnition 5. n)). Such measures are called regular.Proof.
5.5 We call a ddimensional rectangle in Rd any set of the form {x : x ∈ Rd : ai 22 xi < bi }. n) is a Lebesgue measurable set. Let E ⊂ R1 be a bounded Lebesgue measurable set. for any ε > 0 there exists a cover A = [ak . The Lebesgue measure on R1 is regular. bk ).
n→∞
Deﬁnition 5.
5.
.3 The Lebesgue measure on R1 is m(A) = lim m(A ∩ [−n. (Hint: Cover E with a semisegment and apply the above theorem to the σalgebra of measurable subsets in this semisegment). bk ) ⊃ E such that ε m(A) < m(E) + . Then for any ε > 0 there exists a compact set F ⊂ E such that m(E \ F ) < ε.3
The Lebesgue Measure in Rd
Deﬁnition 5. m∗ (E) = m(E).
23
. one can construct the Lebesque measure in Rd in the same fashion as we deed for the R1 case.Using rectangles.
A) is called a measurable space. n
(21) (22) (23) (24)
{x : f (x) < a} = X \ {x : f (x) ≥ a}. A).2 Let (fn ) be a sequence of measurable functions.6
Measurable functions
Let X be a set. with values in the extended real number system. The statement follows from the equalities
∞
{x : f (x) ≥ a} =
n=1 ∞
{x : f (x) > a −
1 }. a. Deﬁnition 6. The function f is called measurable if the set {x : f (x) > a} is measurable for every real a. 24
.
n→∞
Then g and h are measurable. For x ∈ X set g(x) = sup fn (x)(n ∈ N)
n
h(x) = lim sup fn (x). (17) (18) (19) (20)
Proof.2 Let f be a function deﬁned on a measurable space (X. n 1 }. Theorem 6. Example. a.1 The following conditions are equivalent {x : f (x) > a} {x : f (x) ≥ a} {x : f (x) < a} {x : f (x) ≤ a} is is is is measurable measurable measurable measurable for for for for every every every every real real real real a. A a σalgebra on X. {x : f (x) ≤ a} =
n=1
{x : f (x) < a +
{x : f (x) > a} = X \ {x : f (x) ≤ a} Theorem 6. a. Deﬁnition 6.1 A pair (X.
The same proof works for inf. g} since max{f. {x : g(x) ≤ a} =
∞
{x : fn (x) ≤ a}.3 Let f and g be measurable realvalued functions deﬁned on X. Let Ga = {(u. g} = 1/2(f + g − f − g). cn < v < dn }. g} = 1/2(f + g + f − g).Proof. where gm (x) = sup fn (x). g(x)) ∈ Ga } =
n=1
{x : (f (x). g}.
Corollories. v) : an < u < bn . Then the following functions are measurable (i)f + g (ii)f · g (iii)f  f (iv) (ifg = 0) g (v) max{f. Proof. The set {x : an < f (x) < bn } is measurable and so is the set {x : (f (x). 25 (25) (26) (27) (28) (29) (30)
. Now h(x) = inf gm (x). Put h(x) = F (f (x). Let F be real and continuous function on R2 . g(x)) ∈ In }. Hence the same is true for
∞
{x : h(x) > a} = {x : (f (x). and thus
∞
Ga =
n=1
In
where (In ) is a sequence of open intervals In = {(u. v) : F (u. min{f. Then Ga is an open subset of R2 . min{f.
n≥m
Theorem 6. g(x)) (x ∈ X). v) > a}. Let f and g be measurable.
n=1
Since the LHS is measurable it follows that the RHS is measurable too. Then h is measurable. g(x)) ∈ In } = {x : an < f (x) < bn } ∩ {x : cn < g(x) < dn }.
If f is measurable.
The sequence (fn ) is monotonically increasing. If f (x) < ∞ then f (x) < n for a suﬃciently large n and fn (x) − f (x) < 1/2n . f − (x) := − min{f (x).5 Let f be real valued. We will use below the following notation χE (x) = Theorem 6. If f ≥ 0. where f + (x) := max{f (x). If f ≥ 0 set fn (x) = where Eni = {x :
i−1 2n n·2n i−1 i=1 2n χEni
1 if x ∈ E 0 otherwise
n j=1 cj χEj
is measurable if and only if all the
+ nχFn
i }. Proof. (fn ) may be chosen to be a sequence of measurable functions.
26
.4 A simple function f = sets Ej are measurable. There exists a sequence (fn ) of simple functions such that fn (x) −→ f (x) as n → ∞. 0}. In the general case f = f + − f − . Note that if f is bounded then fn −→ f uniformly.1
Step functions (simple functions)
Deﬁnition 6. Exercise. for every x ∈ X. Therefore fn (x) −→ f (x). Theorem 6.6.3 A real valued function f : X → R is called simple function if it takes only a ﬁnite number of distinct values. 0}. (fn ) may be chosen monotonically increasing. Prove the theorem. fn is a simple function. 2n
≤ f (x) <
Fn = {x : f (x) ≥ n}. If f (x) = +∞ then fn (x) = n and again fn (x) −→ f (x).
µ). µ) be a measure space. If X = constant value bj on Fj then
k k j=1
Fj and f takes the
I(f ) =
j=1
bj µ(Fj ). A. Let us choose the representation such that all ci are distinct. Deﬁnition 7.7
Integration
Deﬁnition 7.
i=1
There are diﬀerent representations of f by means of (31).1 Let f be a simple measurable function. Clearly.2 Deﬁne the quantity
n
I(f ) =
i=1
ci µ(Ei ).1 A triple (X. Theorem 7. we derive some properties of I(f ). is called a measure space. Let (X.
Proof. where A is a σalgebra of subsets of X and µ is a measure on it. Ei =
j: bj =ci n
Fj .
27
.
First. A.
n
f (x) =
i=1
ci χEi (x)
(31)
and
n
Ei = X.
This show that the quantity I(f ) is well deﬁned. Ei ∩ Ej = ∅ (i = j).
n k
ci µ(Ei ) =
i i=1
ci µ(
j: bj =ci
Fj ) =
i=1
ci
j: bj =ci
µ(Fj ) =
j=1
bj µ(Fj ). Let f : X → R be a simple measurable function.
Since f ≤ f it follows that f dµ ≥ I(f ). Suppose that f ≤ g everywhere except for a set of measure zero. Exercise.
28
. Proof.
X=
n k=1
Gk .
X=
n
n j=1 m
Fj .2 If f and g are measurable simple functions then I(αf + βg) = αI(f ) + βI(g).4 If f is a simple measurable function then Proof. Proof. if φ ≤ f then I(φ) ≤ I(f ) and also sup I(φ) ≤ I(f )
φ≤f
which leads to the inequality f dµ ≤ I(f ).
On the other hand.3 Let f and g be simple measurable functions. f dµ = I(f ). Then I(φ) = 0 and I(f ) ≤ I(g + φ) = I(f ) + I(φ) = I(g).3 If f : X → R1 is a nonnegative measurable function.Theorem 7. Theorem 7. we deﬁne the Lebesgue integral of f by f dµ := sup I(φ) where sup is taken over the set of all simple functions φ such that φ ≤ f . Complete the proof. If f ≤ g everywhere then in the notation of the previous proof bj ≤ ck on Ejk and I(f ) ≤ I(g) follows. Otherwise we can assume that f ≤ g + φ where φ is nonnegative measurable simple function which is zero every exept for a set N of measure zero. Theorem 7.
(αbj + βck )χEjk (x)
where Ejk = Fj ∩ Gk . Deﬁnition 7. Then I(f ) ≤ I(g). Let f (x) = Then αf + βg =
j=1 k=1 n j=1 bj χFj (x). g(x) =
m k=1 ck χGk (x).
If it is the case we write f ∈ L1 (X. If A is a measurable subset of X (A ∈ A)and f is a nonnegative measurable function then we deﬁne f dµ =
A
f χA dµ. f dµ = f + dµ − f − dµ
if at least one of the terms in RHS is ﬁnite.
• If f (x) ≤ g(x) for all x ∈ A then f dµ ≤
A A
gdµ.
29
.Deﬁnition 7. As a consequence we obtain convergence theorems which give the main advantage of the Lebesgue approach to integration in comparison with Riemann integration. • If a ≤ f (x) ≤ b (x ∈ A)).
• Prove that if µ(A) = 0 and f is measurable then f dµ = 0. Finiteness of the integrals the integral f + dµ and f dµ.
A
The next theorem expresses an important property of the Lebesgue integral. µ(A) < ∞ then aµ(A) ≤
A
f − dµ is equivalent to the ﬁnitenes of
f dµ ≤ bµ(a). The following properties of the Lebesgue integral are simple consequences of the deﬁnition. • If f is measurable and bounded on A and µ(A) < ∞ then f is integrable on A. Remark. µ) or simply f ∈ L1 if there is no ambiguity. If both are ﬁnite we call f integrable.4 1.
2. The proofs are left to the reader.
Let ϕ be a simple measurable function and ϕ ≤ f .
Now if for some i φ(Ai ) = +∞ then φ(A) = +∞ since φ(A) ≥ φ(An ).5 Let f be measurable on X.
Then φ is countably additive on A. For A ∈ A deﬁne φ(A) =
A
f dµ. It is enough to consider the case f ≥ 0. Proof. hence
∞
φ(A) ≤
i=1
φ(Ai ). n k=1
Ek = X. So assume that φ(Ai ) < ∞∀i.
30
. Given ε > 0 choose a measurable simple function ϕ such that ϕ ≤ f and ϕdµ ≥
A1 A1
f dµ − ε.Theorem 7. Ai ∈ A we have
φ(A) =
A n
f dµ =
∞
f χA dµ =
k=1 n
ck µ(Ek ∩ A)
∞
=
k=1 n
ck µ(Ek ∩ (
i=1 ∞
Ai )) =
k=1 ∞
ck µ(
i=1 n
(Ek ∩ Ai )) ck µ(Ek ∩ Ai )
=
k=1
ck
i=1
µ(Ek ∩ Ai ) =
i=1 k=1
(the series of positive numbers)
∞ ∞
=
i=1 Ai
f dµ =
i=1
φ(Ai ). Then for A =
n
∞ i=1
Ai . The general case follows from the decomposition f = f + − f − .
Therefore the same inequality holds for sup. Let f (x) =
n k=1 ck χEk (x).
Now consider general positive f ’s. Then ∞ ∞ ϕdµ =
A i=1 Ai
ϕdµ ≤
i=1
φ(Ai ).
A2
ϕdµ ≥
A2
f − ε. If f = χE for some E ∈ A then µ(A ∩ E) =
A
χE dµ
and σadditivity of φ is the same as this property of µ.
By induction
n n
φ(
i=1
Ai ) ≥
i=1
φ(Ai ). (iii) f ∼ g ⇔ g ∼ f. Theorem 7. g ∼ h ⇒ f ∼ h. (ii) f ∼ g.
This completes the proof. Corollary. f dµ =
A B
f dµ +
A\B
f dµ =
B
f dµ + 0.
Deﬁnition 7.6 If f ∈ L1 then f  ∈ L1 and f dµ ≤
A A
f (x) =
f dµ
31
.5 f and g are called equivalent (f ∼ g in writing) if µ({x : g(x)}) = 0.
Passing to the limit n → ∞ in the RHS we obtain
∞
φ(A) ≥
i=1
φ(Ai ). (i) f ∼ f .
Proof.Hence φ(A1 ∪ A2 ) ≥
A1 ∪A2
ϕdµ =
A1
+
A2
ϕdµ ≥ φ(A1 ) + φ(A2 ) − 2ε.
so that φ(A1 ∪ A2 ) ≥ φ(A1 ) + φ(A2 ). It is not hard to see that f ∼ g is relation of equivalence. B ⊂ A and µ(A \ B) = 0 then f dµ =
A B
f dµ. If A ∈ A.
Since A ⊃
n i=1
Ai we have that
n
φ(A) ≥
i=1
φ(Ai ).
8 Let f = f1 + f2 .
ϕdµ =
A
cϕdµ = lim
n→∞
cϕdµ ≤
An
(this is a consequence of σadditivity of φ proved above)
≤ lim Pass to the limit c → 1.
n→∞
fn dµ ≤ lim
An
n→∞
fn dµ
A
Theorem 7. note that fn (x) ≤ f (x) so that lim
n A
fn dµ ≤
f dµ
It is remained to prove the opposite inequality.Proof. First. Then f dµ = lim
A
n→∞
fn dµ. f2 ∈ L1 (µ). f1 . A ∈ A
A
Proof. −f  ≤ f ≤ f  Theorem 7. Then f ∈ L1 (µ) and f dµ = f1 dµ + f2 dµ
32
.7 (Monotone Convergence Theorem) Let (fn ) be nondecreasing sequence of nonnegative measurable functions with limit f. Deﬁne An = {x ∈ A : fn (x) ≥ cϕ(x)} then An ⊂ An+1 and A = Now observe c
A ∞ n=1
An . For this it is enough to show that for any simple ϕ such that 0 ≤ ϕ ≤ f the following inequality holds ϕdµ ≤ lim
A n A
fn dµ
Take 0 < c < 1.
let f1 . Then for ϕn = ϕn. (fn ) be a sequence of nonnegative measurable functions and
∞
f (x) =
n=1
fn (x). (ϕn. If they are simple then the result is trivial. and f (x) = limn→∞ fn (x) then f dµ ≤ limn→∞
A A
fn dµ
A∈A
33
. Hence Similarly (−f2 )dµ =
B B A
f1 =
A
f dµ +
A
(−f2 )dµ f1 dµ +
B
(−f )dµ
The result follows from the additivity of integral. f2 ≥ 0. x ∈ A
Then f dµ =
A
∞
fn dµ
n=1 A
Exercise. choose monotonically increasing sequences (ϕn. B = {x : f (x) < 0} Then f.2 ) such that ϕn.10 (Fatou’s lemma) If (fn ) is a sequence of nonnegative measurable functions deﬁned a.Proof.2 → f2 . f1 and −f2 are nonnegative on A.1 → f1 and ϕn. Theorem 7. If f1 ≥ 0 and f2 ≤ 0 put A = {x : f (x) ≥ 0}.e. Otherwise.9 Let A ∈ A.1 + ϕn.1 ). First. Theorem 7.2 dµ
and the result follows from the previous theorem. Prove the theorem.1 dµ + ϕn.2 ϕn dµ = ϕn.
gn ≤ fn . by Fatou’s lemma it follows (f + g)dµ ≤ limn
A A
(fn + g)
or f dµ ≤ limn
A A
fn dµ. From fn (x) ≤ g(x) it follows that fn ∈ L1 (µ).Proof. By the monotone convergence theorem f dµ = lim
A n A
gn dµ = limn
A
gn dµ ≤ limn
A
fn dµ.
34
. (fn ) be a sequence of measurable functions such that fn (x) → f (x) (x ∈ A.) Suppose there exists a function g ∈ L1 (µ) on A such that fn (x) ≤ g(x) Then lim
n A
fn dµ =
A
f dµ
Proof.11 (Lebesgue’s dominated convergence theorem) Let A ∈ A. gn ≤ gn+1 . Put gn (x) = inf i≥n fi (x) Then by deﬁnition of the lower limit limn→∞ gn (x) = f (x). Sinnce fn + g ≥ 0 and f + g ≥ 0.
Theorem 7. Moreover.
Since g − fn ≥ 0 we have similarly (g − f )dµ ≤ limn
A A
(g − fn )dµ
so that −
A
f dµ ≤ −limn
A
fn dµ
which is the same as f dµ ≥ limn
A A
fn dµ
This proves that limn
A
fn dµ = limn
A
fn dµ =
A
f dµ.
f (x)dx and the Lebesgue integral
Theorem 8.
b 2m −1
(L)
a
f m (x)dx =
k=0
mk ∆xk = s(f. every bounded measurable function is Lebesgue integarble. . f 1 (x) ≤ f 2 (x) ≤ . b] on n = 2m equal parts by points a = x0 < x1 < . Consider a partition πm of [a. we have
b b b b
(L)
a
f m (x)dx ≤ (L)
f (x)dx ≤ (L)
a a
f (x)dx ≤ (L)
a
f m (x)dx.8
Comparison of the Riemann and the Lebesgue integral
b a
To distinguish we denote the Riemann integral by (R) b by (L) a f (x)dx. b] then it is also Lebesgue integrable on [a.1 If a ﬁnction f is Riemann integrable on [a. πm )
35
. ≤ f (x). f (x) = lim f m (x)
m→∞ m→∞
exist and are measurable.
where χk is a charactersitic function of [xk . Therefore the limits f (x) = lim f m (x). . . So it is enough to prove that if a function f is Riemann integrable then it is measurable.
Moreover. ≥ f (x). f m (x) =
k=0
Mk χk (x). Boundedness of a function is a necessary condition of being Riemann integrable. Note that f (x) ≤ f (x) ≤ f (x). Since f m and f m are simple measurable functions. . b] and
b b
(L)
a
f (x)dx = (R)
a
f (x)dx. < xn−1 < xn = b and set
2m −1 2m −1
f m (x) =
k=0
mk χk (x). . On the other hand. xk+1 ) clearly. .
Proof. f 1 (x) ≥ f 2 (x) ≥ .
b
So s(f. πm ) = (R)
m→∞ a b
f (x)dx.
Since f is Riemann integrable. πm ) ≤ (L)
a
b
f (x)dx ≤ (L)
a
f (x)dx ≤ s(f. πm ).
36
.
Therefore (L)
a
(f (x) − f (x))dx = 0
and since f ≥ f we conclude that f =f =f From this measurability of f follows. πm ). πm ) = lim s(f.and similarly
b
(L)
a
f m (x) = s(f. almost everywhere.
b m→∞
lim s(f.
p q
Then letting t = ab− p−1 we obtain
1 ap b−q 1 + ≥ ab− p−1 . p q
+ 1 − t with t ≥ 0 has the only minimum at t = 1. For p = 1 the statement is obvious. Proof. Then the inequality a + bp ≤ 2p−1 (ap + bp ). called conjugate).1 Let p and q be real numbers such that p > 1. For p > 1 the function y = xp .2 Let p ≥ 1.9
Lpspaces
Let (X. Therefore p a + b ap + bp ≤ . 2 2
37
. b ∈ R. Note that ϕ(t) := that
tp p
+
1 q
= 1 (this numbers are
ap b q + . µ) be a measure space.1
Auxiliary facts
1 p
Lemma 9. holds.
9. a. µ)spaces which occur frequently in analysis. x ≥ 0 is convex since y ≥ 0. In this section we study Lp (X. It follows q t≤ tp 1 + . Then for any a > 0. A. Lemma 9. A. b > 0 the inequality ab ≤ holds. Proof. p q
1
and the result follows.
g ∈ Lp . p
> 0.3
H¨lder’s inequality o
1 Theorem 9. then f + g ∈ Lp and f +g
p
≤ f
p
+ g p.
38
. Let f and g be measurable functions.4 If f. 1 ≤ p < ∞.2
The spaces Lp .
9. We set f
p 1/p p
:=
X
f  dµ
. A. The space Lp = Lp (X.
Proof.3 Let p > 1. p q
9. f p g q p f p q g q p q After integration we obtain f
−1 p −1 p . Deﬁnition
Recall that two measurable functions are said to be equaivalent (with respect to the measure µ) if they are equal µa.
−1 q . It suﬃces to consider the case f Let a = f (x) f By Lemma 1 f (x)g(x) f (x)p g(x)q ≤ + . Then f g is integrable andthe inequality 1/p 1/q
f gdµ ≤
X X
f  dµ
X
p
g dµ
q
.most everywhere. f p and gq q be integrable.9. p ≥ 1. µ) consists of all µequaivalence classes of Ameasurable functions f such that f p has ﬁnite integral over X with respect to µ. p + 1 = 1.4
Minkowski’s inequality
Theorem 9. g
q
> 0.
b = g(x) g
g
−1 q X
f gdµ ≤
1 1 + = 1.
We introduced the quantity f p . Integratin the last inequality and using H¨lder’s inequality we obtain o
1/p 1/q 1/p 1/q
f +g dµ ≤
X X
p
f  dµ
X
p
f + g
(p−1)q
dµ
+
X
g dµ
X
p
f + g
(p−1)q
dµ
. it follows that f ∼ 0. 1 ≤ p < ∞.5
Lp . Theorem 9. 1 ≤ p < ∞. Let us show that it deﬁnes a norm on Lp .3 that Lp . 1 ≤ p < ∞. Indeed.
fni+1 − fni
i=1
p
< 1.
9. is a Banach space
It is readily seen from the properties of an integral and Theorem 9. By the deﬁnition f
p
≥ 0. From Minkowski’s inequality it follows that f + g So Lp .5 Lp .. 1 ≤ p < ∞. f p = 0 =⇒ f (x) = 0 for µalmost all x ∈ X. p
4. is a Banach space. Since Lp consists of µeqivalence classes. Proof. Then there exists a subsequence (fnk )(k ∈ N) with nk increasing such that fm − fnk Then
k p
<
1 2k
∀m ≥ nk .Proof. If f is welldeﬁned. is a normed space. 3. αf
p
= α f
p.
2. 1. Obviously. 1 ≤ p < ∞.
The result follows. is a vector space.
Let (fn ) be a Cauchy sequence in Lp .
39
.
p
and g
p
are ﬁnite then by Lemma 2 f + gp is integrable and f + g
p
f (x)+g(x)p = f (x)+g(x)f (x)+g(x)p−1 ≤ f (x)f (x)+g(x)p−1 +g(x)f (x)+g(x)p−1 .
≤ f
p
+ g p. It remains to prove the completeness.
the limit is ﬁnite since Therefore fn1  +
j=1 ∞
X p gk 1
≤ C = ( fn1
+ 1)p . + fnk+1 − fnk . .
Deﬁne f (x) := lim fnk (x)
k→∞
where the limit exists and zero on the complement.
Put g(x) := lim gk (x). 40
.
p
Moreover. This proves that f − fm
p
→ 0 as m → ∞. Using Minkowski’s inequality we have
k p gk 1 p
=
gk p p
≤
fn1
p
+
i=1
fni+1 − fni
p
< ( fn1
p
+ 1)p . m > N fn − fm Then by Fatou’s lemma f − fm p dµ =
X X p p
=
X
fn − fm p dµ < /2. Let > 0 be such that for n.
k
By the monotone convergence theorem lim
k p gk dµ = A
g p dµ. .Let gk := fn1  + fn2 − fn1  + . Then gk is monotonocally increasing.
j=1
which means that
N
fn1 +
j=1
(fnj+1 − fnj ) = fnN +1 converges almost everywhere as N → ∞. So f is measurable.
fnj+1 − fnj  converges almost everywhere
∞
and so does fn1 +
(fnj+1 − fnj ).
lim fnk − fm p dµ ≤ limk
k X
fnk − fm p dµ
which is less than
for m > N .