Sie sind auf Seite 1von 40

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

i, j, k always

denote positive integers. The sign is used for restriction of a function (operator etc.) to

a subset (subspace).

diﬀerence of A and B is denoted, i.e. A B = ( A \ B ) ( B \ A). Letters

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 ,

[x k , x k +1 ]}, M k = sup {f ( x ) : x [x k , x k +1 ]} . Deﬁne the upper and lower Riemann— Darboux sums

, n 1 }, m k = inf { f ( x ) : x

s ( f, π ) =

n

1

k

=0

m k x k , s¯( f, π ) =

n1

k

=0

M

k x k .

One can show (the Darboux theorem) that the following limits exist

lim 0 s ( f, π ) =

|π |→

sup s ( f, π ) = b f dx

π

a

lim 0 s¯( f, π ) = inf s¯( f, π ) = b f dx.

| π |→

π

a

1

Clearly,

s ( f, π ) b f dx b f dx s¯( f, π )

a

a

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 b | f ( x )

a

x

g ( x ) | ,

ρ 2 ( f, g ) = b | f ( x ) g ( x ) | dx.

a

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 } =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 ].

n

2

 ✻ ☞ ☞ ☞ ☞ ☞ ☞ ☞ ☞☞ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲ ☞ ✲ − 1 1 1 1 − 2 2 + n 1 2 − 1 n 2

Figure 1: The function ϕ n .

2 Systems of Sets

Deﬁnition 2.1 A ring of sets is a non-empty 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 semi-segments [ a, b ).

Deﬁnition 2.2 A semi-ring 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 , sets (i.e. C i C j = if i = j ) such that

A \ B =

n

j =1

C j .

, n of

Example. Let X = R , then the set of all semi-segments, [ a, b ), forms a semi-ring.

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 σ -ring (a σ -algebra) is a ring (an algebra) of sets which is closed with respect to all countable unions.

Deﬁnition 2.5 A ring (an algebra, a σ -algebra) of sets , K ( U ) generated by a collection of sets U 2 X is the minimal ring (algebra, σ -algebra) of sets containing U .

In other words, it is the intersection of all rings (algebras, σ -algebras) 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 ) i 1 is a disjoint family of sets in A ( A i A j = for any i

= j ) such that

=1 A i A , then

i

∞ ∞

µ

(

i

=1

A i ) =

i

=1

µ

( A i ) .

The latter important property, is called countable additivity or σ -additivity 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

If µ ( B ) < then

µ ( A) = µ ( A \ B ) + µ ( B ) .

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. (Semi-additivity of µ ). If ( A i ) i 1 A such that =1 A i A then

i

∞ ∞

µ

(

i

=1

A i )

i

=1

µ

(

A i ) .

Proof. First let us proove that

Note that the family of sets

µ

(

n

i

=1

A i )

n

i

=1

µ

(

A i ) .

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 µ ( A i ). Then

i =1 B i = i n

=1 A i . Moreover, since B i A i , we see that µ ( B i )

µ

(

n

A i ) = µ (

n

B i ) =

n

µ ( B i )

n

µ

( A i ) .

i

=1

i

=1

i

=1

i

=1

Now we can repeat the argument for the inﬁnite family using σ -additivity 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 ( =1 A i ) = + .

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 ) + lim

n→∞

n

k =2

= µ ( A 1 ) +

k =2

µ ( A k \ A k 1 )

( 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

An arbitrary set A X can be always covered by sets from A , i.e. one can always ﬁnd

to as many elements of 2 X as possible.

E 1 , E 2 ,

A such that =1 E i A . For instance, E 1 = X, E 2 = E 3 =

i

= .

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 A -coverings 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 ) , - σ -algebra 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 A -covering ( E i ) of A such that i µ ( E i ) < µ ( A) + ε. Note that

A = A ( E i ) = ( A E i ) .

i

7

i

 ✻ ✲ a b

Using consequently σ -semiadditivity and monotonicity of µ , one obtains:

µ ( A) µ ( A E i ) µ ( E i ) < µ ( A) + ε.

i

i

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 σ -algebra), 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 (σ -semiadditivity of µ ). µ ( =1 A j ) =1 µ ( A j ) .

j

j

8

Proof. If the series in the right-hand 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 A -covering k E kj A j such that

µ ( E kj ) < µ ( A j ) +

ε

2 j .

k =1

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 σ -algebra 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

10

A) .

˜

4 . A is a σ -algebra.

From the previous step, by induction, for any ﬁnite disjoint collection ( B j ) of sets:

µ ( E (

n n

j

=1

B j )) =

j

=1

µ ( E B j ) .

Let A = =1 A j , A j A . Then A =

j

=1 B j , B j = A j \ j 1

j

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 ) .

(6)

(7)

˜

Since A is an algebra, it follows that

holds for every n :

n j =1 B j

˜

A ( n N ) and the following inequality

µ ( E ) µ ( E (

n

j

=1

B j )) + µ ( E (

n

j

=1

B j ) c ) .

(8)

Since E ( =1 B j ) c E ( =1 B j ) c , by monotonicity of the mesasure and (8)

j

j

n

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)

µ ( E A ) = µ ( E (

j

=1

Compare this with (10):

B j )) = µ (

j

=1

( E B j ))

j

=1

µ ( E B j ) .

µ ( E ) µ ( E A) + µ ( E A c ) .

˜

˜

Thus, A A , which means that A is a σ -algebra.

˜

5 . µ˜ = µ A is a measure.

11

We need to prove only σ -additivity. Let E = =1 A j . From(10) we get

j

∞ ∞

µ

(

j

=1

A j )

j

=1

µ ( A j ) .

The oposite inequality follows from σ -semiadditivity of µ .

˜

6 . A A .

Let A A , E X .

We need to prove:

µ ( E ) µ ( E A) + µ ( E A c ) .

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.

(11)

For E X for ε > 0 there exists a A -covering ( E j ) of E such that

µ ( E ) + ε >

j =1

µ ( E j ) .

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

(12)

 E ∩ A ⊂ ( E j ∩ A ) j =1 ∞ E ∩ A c ⊂ ( E j ∩ A c ) j =1 By monotonicity and σ -semiadditivity ∞ µ ∗ ( E ∩ A) µ ( E j ∩ A) , j =1 ∞ µ ∗ ( E ∩ A c ) µ ( E j ∩ A c ) .

j

=1

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 σ -algebra 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 σ -algebra and construct the extension µ σ of µ on A σ . This extension is called minimal extension of measure .

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.

˜

˜

Since A A therefore A σ A .

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 σ -ﬁnite 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 σ -ﬁnite 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

Therefore

ν ( A) ν (

j

=1

E j )

j

=1

ν ( E j ) =

˜

j

=1

µ ( E j ) .

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 σ -ﬁnite, ( F j ) be an increasing sequence of sets from A such that µ ( F j ) < ∞ ∀ j and X = =1 F j . From what we have already proved it follows that

j

˜

µ ( A F j ) = ν ( A F j ) A A .

Therefore

µ ( A ) = lim µ ( A F j ) = lim ν ( A F j ) = ν ( A) .

j

j

Theorem 3.9 (Continuity of measure). Let A be a σ -algebra with a measure µ , { A j } ⊂ A a monotonically increasing sequence of sets. Then

Proof. One has:

A =

µ

(

j =1

A j ) = lim

j

µ ( A j ) .

∞ ∞

j

=1

A j =

j

=2

( A j +1 \ A j ) A 1 .

Using σ -additivity 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) + 2

ε = µ˜ ( A ) + 2 ε .

14

On the other hand,

The monotonicity of µ˜ implies

µ˜ (

j

j

=1

µ ( E j ) µ˜ ( E J ) .

j

E J ) = n→∞ lim µ˜ (

n

j

=1

E j ) ,