Professional Documents
Culture Documents
Fabio Maccheroni1
Department of Decision Sciences - U. Bocconi - Milano
1I thank Erio Castagnoli who read these notes over and over again and added punctuation;
I also thank Laura Maspero who helped me to write them and took notes during the Advanced
Analysis course held by Luigi Paganoni at "Università degli Studi di Milano" during the academic
year 1991-1992, notes and course that in‡uenced these notes very much. Finally I thank Giulia
Brancaccio who drafted this …rst English version of them in 2010. I take responsibility of all the
mistakes.
Contents
1 PREREQUISITES 2
3 SET FUNCTIONS 9
4 EXTENSION OF A MEASURE 12
5 COMPLETIONS 18
6 LEBESGUE-STIELTJES MEASURES ON R 20
7 MEASURABLE FUNCTIONS 23
11 SUMMABLE FUNCTIONS 40
16 RADON-NIKODYM THEOREM 62
17 ESSENTIAL BIBLIOGRAPHY 66
1
1 PREREQUISITES
Set operations, topological and order properties of R, ("; ) arguments, properties of series
with nonnegative terms.
(a.i) X 2 A,
(a.ii) A 2 A ) Ac 2 A,
(a.iii) A; B 2 A ) A [ B 2 A.
1. ; 2 A:
[In fact, ; = X c and X 2 A.]
SN
2. A1 ; A2 ; :::; AN 2 A ) j=1 Aj 2 A:
[By induction.]
3. A; B 2 A ) A \ B 2 A.
[In fact, A \ B = (Ac [ B c )c .]
TN
4. A1 ; A2 ; :::; AN 2 A ) j=1 Aj 2 A:
[By induction.]
5. A; B 2 A ) A B 2 A.
[Exercise.]
6. A; B 2 A ) A M B 2 A.
[Exercise.]
2
Proposition. A P (X) is an algebra if and only if (a.i) and (a.ii) hold, and A; B 2
A ) A \ B 2 A.
Proof. It is enough to note that A [ B = (Ac \ B c )c .
is an algebra of subsets of R.
Proof. By assumption, R 2 E.
If I 2 I, then I c is a …nite disjoint union of elements of I. In fact, if I = ;, then I c = R; if
I = R, then I c = ;; if I = (a; b] for a 2 R, then I c = ( 1; a] [ (b; 1); if I = ( 1; b], then
I c = (b; 1); …nally if I = (b; 1), then I c = ( 1; b].
I; J 2 I ) I \ J 2 I. If I = ; or J = ;, and If I = R or J = R, the statement is straight-
forward;
If I = (a; b] and J = (c; d], x 2 I \ J i¤ x > a; x b; x > c; x d i.e. x > max fa; cg and
x min fb; dg i.e. x 2 (max fa; cg ; min fb; dg], that is I \ J = (max fa; cg ; min fb; dg] 2 I.1
If I = (a; b] and J = (c; 1), x 2 I \ J i¤ x > a; x b; x > c i.e. x > max fa; cg and x b
i¤ x 2 (max fa; cg ; b], that is I \ J = (max fa; cg ; b] 2 I.
If I = (a; 1) and J = (c; 1), then I \ J = I or J 2 I.
S SM
E; F 2 E ) E \ F 2 E . In fact, E = N j=1 Ij and F = i with Ij ; Ji 2 I for each j
i=1 Ji
SN SM SN h SM S SM
and i, E \ F = j=1 Ij \ i=1 Ji = j=1 Ij \ i=1 Ji = N j=1 i=1 Ij \ Ji =
SN;M
j=1;i=1 Ij \ Ji 2 E since Ij \ Ji 2 I for each i; j.
T
By induction, E1 ; E2 ; :::; EN 2 E ) N Ej 2 E.
SN j=1 T
E 2 E ) E 2 E . In fact, E = j=1 Ij with Ij 2 I for each j and E c = N
c c c
j=1 Ij , but Ij 2 E
for each j, therefore E c 2 E:
(s.i) ;; X 2 S,
(s.iii) A; B 2 S ) A \ B 2 S.
1
With the convention ( ; ] = ; if .
3
With little abuse, we call the class I de…ned in the previous Proposition: class of real
semi-intervals. I is a semialgebra (why?).
is an algebra of subsets of X.
Proof. Exercise. Hint: look at the …nal steps of the previous proof.
( .i) X 2 F,
( .ii) A 2 F ) Ac 2 F,
S1
( .iii) fAn gn2N F) n=1 An 2 F.
Some trivial -algebras are: P (X), f;; Xg, f;; A; Ac ; Xg for ; A X. Each algebra
of subsets of a …nite set is a -algebra. E is not a -algebra (why?).
1. F is an algebra.
[In fact, A; B 2 F ) A [ B = A [ B [ B [ B:::: 2 F.]
T
2. fAn gn2N F ) 1 n=1 An 2 F.
T1 S c c
[In fact, n=1 An = ( 1n=1 An ) .]
T1
Proposition. F P (X) is a -algebra i¤ ( .i), ( .ii) and fAn gn2N F) n=1 An 2 F
hold.
Proof. Exercise.
Let fAn gn2N P (X); if An An+1 for all n 2 N, the sequence fAn gn2N is increasing,
if An An+1 for all n 2 N, the sequence fAn gn2N is decreasing. If fAn g is increasing or
S
decreasing, it is said to be monotone. If fAn gn2N is increasing and A = 1n=1 An , we write
T1
An " A; analogously, if fAn gn2N is decreasing and A = n=1 An , we write An # A.
4
(i) fAn gn2N M and An " A ) A 2 M,
Clearly a -algebra is a monotone class. The reverse is not true (why?), but:
Definition. Let C be a non empty class of subsets of X. We call -algebra (resp. algebra,
resp. monotone class) generated by C the smallest -algebra (resp. algebra, resp.
monotone class) T containing C. Moreover, C is said to be a system of generators of T .
In other words, the -algebra (resp. algebra, resp. monotone class) generated by C is a
-algebra (resp. algebra, resp. monotone class) T s.t.
1. T C,
Notice that, if T exists, then it is unique: If it existed T 0 6= T with the same properties, since
T is a -algebra containing C, then T T 0 ; vice-versa, since T 0 is a -algebra containing
C, then T 0 T . Therefore T 0 = T ! (= paradox). Such a -algebra is denoted by
(C) (resp. A (C), resp. M (C)). The next proposition shows that (C) (resp. A (C), resp.
M (C)) always exists.
Proposition. Let C be a non empty class of subsets of X, the -algebra (resp. algebra,
resp. monotone class) generated by C is the intersection of all -algebras (resp. algebras,
resp. monotone classes) containing C.
Proof. We prove the “ -algebra”case: the “algebra”and “monotone class”cases are similar
(verify it). Let fF g 2 be the set of all -algebras F such that F C. P (X) C and
T
it is a -algebra, therefore fF g 2 6= ;. Let T = 2 F . Since, for every , F is a
-algebra, then X 2 F for every and X 2 T . If A 2 T , then A 2 F for every , and
therefore Ac 2 F for every (F is a -algebra), then Ac 2 T . If An 2 T for every n 2 N,
S
then An 2 F for every n 2 N and for every , therefore 1 n=1 An 2 F for every (F is
S1
a -algebra), …nally n=1 An 2 T . Therefore T is a -algebra. If C 2 C, then C 2 F for
every , and C 2 T . Thus, T is a -algebra and contains C and, by de…nition, if F is a
-algebra s.t. F C, then F T.
Facts. Let ; =
6 C K P (X).
5
1. A (C) (C).
[Exercise.]
2. M (C) (C).
[Exercise.]
nS o
N
Proof. Let A = j=1 Aj : N 2 N and Aj 2 S for j = 1; 2; :::; N and A0 = f…nite disjoint
unions of elements of Sg. We have already shown that A is an algebra and since A S,
then A A (S). Moreover, being A (S) an algebra that contains S, it also contains all
the …nite unions of elements of S, i.e. A (S) A. Clearly A (S) A0 . It remains to
S
be shown that A0 is an algebra. Of course X 2 A0 . If E; F 2 A0 , E = N j=1 Aj and
SM
F = i=1 Bi with A1 ; A2 ; :::; AN 2 S pairwise disjoint sets and B1 ; B2 ; :::; BM 2 S pairwise
SN SM S
disjoint sets, then E \ F = j=1 Aj \ i=1 Bi = N;M
j=1;i=1 Aj \ Bi . Let (j; i) ; (k; h) 2
f1; 2; :::; N g f1; 2; :::; M g and (j; i) 6= (k; h), then either j 6= k or i 6= h, in the …rst case
Aj \ Ak = ;, in the second Bi \ Bh = ;, in both cases (Aj \ Bi ) \ (Ak \ Bh ) = ;. Thus
S TN
E \ F = N;M 0
j=1;i=1 Aj \ Bi 2 A . By induction, E1 ; E2 ; :::; EN 2 A )
0 0
j=1 Ej 2 A . Let
S TN
E 2 A0 , then E = N j=1 Aj with A1 ; A2 ; :::; AN 2 S pairwise disjoint sets and E =
c c
j=1 Aj ,
but, being S a semialgebra, for each j the Acj is a disjoint union of elements of S, i.e. Acj 2 A0 ,
T
and N c
j=1 Aj 2 A .
0
M (A) = (A) :
Proof. Clearly A M (A) (A). It remains to show that M (A) is an algebra, in that
case, since it is also a monotone class, M (A) is a -algebra, and, containing A, it also
contains (A).
For each B X, set DB = fA 2 P (X) : A \ B c ; B \ Ac ; A [ B 2 M (A)g. If fAn gn2N DB
6
and An # A, then An \B c # A\B c , B \Acn " B \Ac , and An [B # A[B (verify it); moreover
fAn \ B c g; fB \ Acn g; fAn [ Bg M (A). It follows that A \ B c ; B \ Ac ; A [ B 2 M (A),
and hence A 2 DB . Analogously, if fAn gn2N DB and An " A, then A 2 DB (verify it).
Thus DB is a monotone class for each B X.
If B 2 A, A DB (in fact, for each A 2 A, A \ B c ; B \ Ac ; A [ B 2 A M (A)), but DB is
a monotone class and therefore M (A) DB . This implies that, if B 2 A and C 2 M (A),
it is true that C \ B c ; B \ C c ; C [ B 2 M (A); that is, if B 2 A and C 2 M (A), then
B \ C c ; C \ B c ; B [ C 2 M (A); that is, if B 2 A and C 2 M (A), then B 2 DC . Thus, if
C 2 M (A), A DC , but DC is a monotone class and M (A) DC .
Finally, if C; D 2 M (A), then D 2 DC , and D \ C c ; C \ Dc ; D [ C 2 M (A); moreover
X 2 A M (A) and if C 2 M (A), then C c = X \ C c 2 M (A).
( .i) X 2 H,
( .ii) A; B 2 H, A B)B A 2 H,
( .i) A; B 2 D ) A \ B 2 D.
1. If A 2 H, then Ac 2 H.
[Exercise.]
2. If A; B 2 H and A \ B = ;, then A [ B 2 H.
[In fact, A B c , and (A [ B)c = B c \ Ac = B c A.]
3. H is a monotone class.
[Exercise. Hint: if An 2 H and An # A, then Acn 2 H and Acn " Ac .]
7
4. If H is a -class, then H is a -algebra.
[Exercise. Hint: (A [ B)c = Ac \ B c .]
G1 is a -class (why?) and contains D, then G1 (D). It follows that for each A 2 (D)
and each D 2 D, A \ D 2 (D). Let
Facts. Let ; =
6 Y X and ; =
6 C P (X).
2. (C) \ Y = (C \ Y ) ; A (C) \ Y = A (C \ Y ).
[Exercise.]
Let I be the semialgebra of semi-intervals of R. The -algebra (I) is called the Borel
-algebra of R and it is denoted by B or B (R). The following classes generate B (prove it):
E = A (I).
8
Closed sets in R.
Open sets in R.
Open intervals in R.
Closed intervals in R.
f( 1; a] : a 2 Rg.
f( 1; a) : a 2 Rg.
f(b; 1) : b 2 Rg.
f[b; 1) : b 2 Rg.
Let ; =
6 I R; the -algebra (I) \ I = (I \ I) is said Borel -algebra of I and it
is denoted by B (I). B (I) is the -algebra generated by open sets in I, or by closed sets in
I,...(prove it).
3 SET FUNCTIONS
If A; B X and A\B = ;, we write AtB instead of A[B, analogously if fAn gn2N P (X)
F S1
and Ai \ Aj = ; for i 6= j, we write 1 n=1 An instead of n=1 An .
Let C P (X) and A X. A …nite partition (resp. countable) of A in C is a set
S
fA1 ; A2 ; ::; AN g (resp. fAn gn2N ) of pairwise disjoint elements of C s.t. A = Nj=1 Aj (resp.
S1
A = n=1 Aj ):
additive, if
X
N
(A) = (Aj ) .
j=1
9
-additive, if X
(A) = (Aj )
j2J
(c.i) (;) = 0,
(c.ii) A; B 2 C and A \ B = ; ) (A [ B) = (A) + (B).
[Exercise.]
(m.i) (;) = 0,
S1 S P1
(m.ii) fAn gn2N C, n=1 An 2 C, Ai \ Aj = ; if i 6= j ) ( 1n=1 An ) = n=1 (An ).
[Exercise.]
1. is monotone.
10
SN PN
3. j=1 Aj j=1 (Aj ) for each A1 ; A2 ; :::; AN 2 A.
for each n 2 N, Bn 2 A,
Bi \ Bj = ; if i 6= j,
F1 S1
n=1 Bn = n=1 An = A,
Fn
for each n 2 N, An = j=1 Bj .
2.) If An # A, then A1 An " A1 A (verify it), then limn (A1 ) (An ) = (A1 ) (A),
or limn (An ) = (A).
Sn 1
3.) Let B1 = A1 and Bn = An j=1 Aj for each n 2. Notice that,
Bi \ Bj = ; if i 6= j,
11
F1 S1
n=1 Bn = n=1 An ,
P1 P1
(Verify these statements.) Therefore (A) = n=1 (Bn ) n=1 (An ) .
(c.iii’) is -subadditive.
Proof. If is a measure requirements (c.iii) and (c.iii’) are satis…ed. Let fAn gn2N A and
F FN
A= 1 n=1 A n 2 A; for each N 2 N set B N = n=1 An 2 A. Suppose (c.iii) holds, since
Bn " A, it follows that
X
N X
1
(A) = lim (BN ) = lim (An ) = (An ) :
N N
n=1 n=1
FN FN
Suppose (c.iii’) holds, monotonicity implies, as n=1 An A, that n=1 An (A) for
each N 2 N, then
!
X
1 X
N G
N
(An ) = lim (An ) = lim An (A) :
N N
n=1 n=1 n=1
F P1
Finally -subadditivity implies that (A) = ( 1n=1 An ) n=1 (An ).
4 EXTENSION OF A MEASURE
Let C D be two non empty classes of subsets of X, and consider the functions : C ! [0; 1]
and : D ! [0; 1]; is an extension of if jC = .
12
FN FM
Let A = j=1 Aj with Aj 2 S and A = k=1 Bk with Bk 2 S, then
! !
X
N X
N G
M X
N G
M
(Aj ) = Aj \ Bk = (Aj \ Bk ) =
j=1 j=1 k=1 j=1 k=1
X
N X
M X
M X
N
= (Aj \ Bk ) = (Aj \ Bk ) =
j=1 k=1 k=1 j=1
! !
X
M G
N X
M G
N
= (Aj \ Bk ) = Bk \ Aj =
k=1 j=1 k=1 j=1
X
M
= (Bk ) .
k=1
X
N X
M
~ (A [ B) = (Aj ) + (Bk ) = ~ (A) + ~ (B) .
j=1 k=1
X
N X
N
(A) = (Aj ) = (Aj ) = ~ (A) .
j=1 j=1
13
P1
to the limit, ~ (A) = 1. When instead n=1 ~ (An ) converges:
X
1 1 jn+1
X X1 X
1
~ (An ) = (Bj ) = (Bj ) =
n=1 n=1 j=jn j=1
X
1 X
1 X
L
= (A \ Bj ) = (Ck \ Bj ) =
j=1 j=1 k=1
X
L X
1 X
L
= (Ck \ Bj ) = (Ck ) = ~ (A) ,
k=1 j=1 k=1
that is ~ is -additive.
is called the outer measure induced by (beware! It might fail to be a measure on P (X)).
Clearly, : P (X) ! [0; 1] is a set function (verify it). Notice that, if (A) < 1, then for
S1 P1
each " > 0 there exists fBn g A s.t. A n=1 Bn and n=1 (Bn ) < (A) + " (why?).
Proposition.
1. is monotone.
S P1
2. fAn gn2N P (X) ) ( 1n=1 An ) n=1 (An ).
3. jA = .
S1 S1
Proof. 1.) Let A B X, for each fBn g A s.t. B n=1 Bn , then A n=1 Bn , and
P1
hence (A) n=1 (Bn ), then
(1 )
X [
1
(A) inf (Bn ) : fBn g A and B Bn = (B) :
n=1 n=1
S1 P1
2.) Set A = n=1 An . If n=1 (An ) = 1, the result is trivial. Then suppose (An )
P1
j=1 (Aj ) < 1 for each n 2 N. Let " > 0. For each n 2 N there exists fBn;m g A s.t.
S1 P1 S S S1
An m=1 Bn;m and m=1 (Bn;m ) < (An ) + 2"n . n;m Bn;m = 1 n=1 ( m=1 Bn;m ) A
P P1 P1 P1
(why?). It follows that (A) n;m (Bn;m ) = (Bn;m ) (An ) + 2"n =
P1 P1n=1 m=1 n=1
14
3.) For each A 2 A, (A) (A) (consider fBn g = fA; ;; ;; ;; ;; ::::g). If fBn g A
S1 S1 P1
and A n=1 Bn , then A = n=1 (Bn \ A) for Bn \A 2 A, then (A) n=1 (Bn \ A)
P1
n=1 (Bn ) and (A) (A).
(E) = (E \ A) + (E \ Ac )
for each E X.
(E) = (E \ B) + (E \ B c ) (1)
Substituting E with E \ A
(E \ A) = (E \ A \ B) + (E \ A \ B c ) (2)
(E \ Ac ) = (E \ Ac \ B) + (E \ Ac \ B c ) (3)
Adding (2) and (3):
(E) = (E \ A \ B) + (E \ A \ B c ) + (E \ Ac \ B) + (E \ Ac \ B c ) (4)
(E \ (A [ B)) = (E \ (A [ B) \ A \ B) + (E \ (A [ B) \ A \ B c )
+ (E \ (A [ B) \ Ac \ B) + (E \ (A [ B) \ Ac \ B c )
= (E \ A \ B) + (E \ A \ B c ) + (E \ Ac \ B) (5):
15
Substituting (4) in (5),
Step 2. If A 2 C, then Ac 2 C.
Let E X, by de…nition
(E) = (E \ A) + (E \ Ac ) = (E \ Ac ) + (E \ (Ac )c ):
Thus C is an algebra .
Step 3. C is a -algebra.
Let E X, and consider A; B 2 C s.t. A \ B = ;. It follows from (5) that
(E \ (A [ B)) = (E \ A) + (E \ B);
SN c S1 c
Since E \ j=1 Bj E\ j=1 Bj = E \ Ac , monotonicity implies:
!!
[
N
(E) E\ Bj + (E \ Ac ) =
j=1
X
N
= (E \ Bj ) + (E \ Ac ) (9):
j=1
16
S1 S1 P1
But (E \ A) = E\ j=1 Bj = j=1 (E \ Bj ) j=1 (E \ Bj ), this,
together with (10), implies:
(E) (E \ A) + (E \ Ac ) (10):
Thus A 2 C.
Step 4. jC is a measure. In fact it is -subadditive.
Step 5. C A.
Let A 2 A and E X: It remains to show that
(E) (E \ A) + (E \ Ac ) (11):
moreover
[
1 [
1
E\A (Ej \ A) and E \ Ac (Ej \ Ac ) .
j=1 j=1
Thus
X
1
c
(E \ A) + (E \ A ) ( (Ej \ A) + (Ej \ Ac )) =
j=1
X
1
= (Ej ) < (E) + ":
j=1
17
In particular, a measure on a semialgebra S can be extended to a measure on (S). Next
theorems concern the uniqueness of this extension.
5 COMPLETIONS
Definition. A measure on a -algebra F is complete if A 2 F, (A) = 0 and N A
imply N 2 F (and (N ) = 0).
18
Proof. Let A 2 C (A; ), (A) = 0 and N A; then, by monotonicity, for each E X:
0 (E \ N ) (E \ A) (A) = 0:
Again by monotonicity:
(E) 0+ (E \ N c ) = (E \ N ) + (E \ N c ) .
Thus N 2 C (A; ).
Then
F 0 = fA 4 N : A 2 F; N 2 N g
is a -algebra containing F and the set function
0
(A 4 N ) = (A)
Theorem. Let
Then
(A)0 = C (A; ) and 0
= ~.
19
6 LEBESGUE-STIELTJES MEASURES ON R
Consider the semialgebra
F ( 1) = limt! 1 F (t),
Lemma. Let F : R ! R be an increasing and right continuous function; consider (a; b] and
S1
(an ; bn ] 2 I s.t. (a; b] n=1 (an ; bn ], then
X
1
F (b) F (a) (F (bn ) F (an )) :
n=1
20
a = a1 < b1 a2 < b2 :::::bN 1 aN < bN = b, but if bj < aj+1 for some j N 1, (a; b]
wouldn’t be connected, which is absurd. It follows that bj = aj+1 for every j N 1; set
aN +1 = bN obtaining:
X
N X
N X
N
X
N +1 X
N
= F (aj ) F (aj ) = F (aN +1 ) F (a1 ) = F (b) F (a) =
j=2 j=1
= F ((a; b]) .
and as a consequence
X
F ((a; b]) = F ((aj ; bj ]) =
j:(aj ;bj ]6=;
X
N
= F ((aj ; bj ]) .
j=1
X
N
The reverse inequality is granted by the previous Lemma. Thus F is a measure on the
S
semialgebra I and, as R = 1 n=1 ( n; n], F is -…nite. As a consequence there exists an
unique measure F that extends F to B. Clearly, if B 2 B and B is bounded, then there
exist a; b 2 R s.t. B (a; b] and F (B) F ((a; b]) = F (b) F (a) < 1.
Notice that, if t 0, then:
F (t) = F ((0; t]) + F (0) :
If t < 0, then:
F (t) = F (0) F ((t; 0]) .
21
Let F = G. If t 0, then:
If t < 0, then:
Then F (t) = G (t) + F (0) G (0) for each t 2 R. Vice-versa, if F (t) = G (t) + c for each
t 2 R, then F ( 1) = G ( 1) + c, and
for each 1 a b 1.
Let F : R ! R be an increasing and right continuous fnction and let F be the extension
of F to B. We use F instead of F , and we call it Lebesgue-Stieltjes measure induced by
F . The Lebesgue-Stieltjes measure induced by id is called Borel measure, its completion
Lebesgue measure and the -algebra of 0id -measurable sets is called Lebesgue -algebra.
Attention: not every subset of R belong to the Lebesgue -algebra.
Facts. Let F : R ! R be an increasing and right continuous function and F be the induced
Lebesgue-Stieltjes measure. We write F (b ) = limt!b F (t) for each b 2 R, then
1. (fbg) = F (b) F (b ) :
F
[Exercise; in this regards, why fbg 2 B?]
2. F (fbg) = 0 i¤ F is continuous in b:
4. ((a; b)) = F (b )
F F (a) :
[Exercise.]
22
Proposition. Let be a Lebesgue-Stieltjes measure on B, then there exists a function
F : R ! R increasing and right continuous, s.t. = (F ) .
Proof. If t 0, set
F (t) = ((0; t]) ;
if t < 0,
F (t) = ((t; 0]) .
Prove (as an exercise) that F is increasing and right continuous.
Instead of considering the whole real line R we can consider an interval I, and the -
algebra B (I), the the Lebesgue-Stieltjes measure induced by F on I, is precisely ( F )B(I) .
7 MEASURABLE FUNCTIONS
Let X be a set and F a -algebra of subsets of X. The pair (X; F) is called measurable
space.
1
Notice that fx 2 X : f (x) > ag = f ((a; 1)): it can be simply denoted by ff > ag.
Proposition. Let (X; F) be a measurable space, and consider the function f : X ! R (or
[ 1; 1]). The following are equivalent:
1. f is measurable.
It follows that
23
Proof. Exercise. Use (after verifying) the following identities
\
1
1
fx 2 X : f (x) ag = x 2 X : f (x) > a ;
n=1
n
fx 2 X : f (x) < ag = X fx 2 X : f (x) ag ;
\
1
1
fx 2 X : f (x) ag = x 2 X : f (x) < a + ;
n=1
n
fx 2 X : f (x) > ag = X fx 2 X : f (x) ag ;
fx 2 X : f (x) = ag = fx 2 X : f (x) ag fx 2 X : f (x) > ag if a 2 R,
\
1
fx 2 X : f (x) = 1g = fx 2 X : f (x) > ng ;
n=1
\
1
fx 2 X : f (x) = 1g = fx 2 X : f (x) < ng :
n=1
1
1. The class T = fA R t.c. f (A) 2 Fg is a -algebra.
[Exercise.]
1
2. If H is a -algebra, f (H) is -algebra.
[Exercise.]
3. (f 1 (H)) = f 1 ( (H)).
[In fact, according to 2., f 1 ( (H)) is a -algebra, moreover if H 2 H, H 2 (H),
and f 1 (H) 2 f 1 ( (H)); thus f 1 (H) f 1 ( (H)) and therefore (f 1 (H))
f 1 ( (H)). Vice-versa, we want to show that each -algebra C containing f 1 (H) also
contains f 1 ( (H)). Set G = fA R : f 1 (A) 2 Cg, according to1. G is a -algebra,
since f 1 (H) C, H G, and (H) G, it follows that f 1 ( (H)) C.]
1. f is measurable.
1
2. f (B) F.
24
1
3. f (G) F for each G P (R) s.t. B = (G).
1
4. f (G) F for each G P (R) s.t. B (G).
[
1
fx 2 X : g(x) > ag = fx 2 X : fn (x) > ag
n=1
Remember that f + = max ff; 0g, f = min ff; 0g, and the following equalities hold:
jf j = f + + f , f = f + f .
25
Proposition. Let (X; F) be a measurable space; consider f; g : X ! R (or [ 1; 1]
setting 0 = 1 1 = 1 + 1) measurable functions and c 2 R. Then the functions:
h (x) = c for each x 2 X, f + c, cf , f + g, f g, max ff; gg, min ff; gg, f + , f , and jf j, are
measurable.
Proof. For each a 2 R,
n ao
fx 2 X : cf (x) > ag = x 2 X : f (x) > if c > 0,
c
= fx 2 X : 0 > ag if c = 0,
n ao
= x 2 X : f (x) < if c > 0,
c
and therefore cf is measurable.
Let a 2 R, if f (x) + g (x) < a, then f (x) < a g (x), and there exists q 2 Q s.t. f (x) <
q < a g (x), then f (x) < q and g (x) < a q, and hence x 2 fx 2 X : f (x) < qg \
fx 2 X : g (x) < a qg. Thus
[
fx 2 X : f (x) + g (x) < ag = fx 2 X : f (x) < qg \ fx 2 X : g (x) < a qg 2 F
q2Q
and f + g is measurable.
For each a 2 R,
26
Let A X, denote by A the characteristic function of A de…ned as
(
1 if x 2 A
A (x) = .
0 if x 2 Ac
X
N
'= i Ai
j=1
gn is monotone increasing (gn ") if gn (x) gn+1 (x) for each x 2 X and each n 2 N,
if also gn ! g, we write gn " g.
unif:
gn converges uniformly to g (gn ! g) if supx2X jg (x) gn (x)j ! 0 (this de…nition
applies only for real valued functions).
27
Proof. Let f 0. For each n 2 N and x 2 X, set
8
>
> n if x 2 f 1 ([n; 1]) i.e. f (x) n
>
<
'n (x) = j :
>
>
> 2n
if x 2 f 1 2jn ; j+1
2n
i.e. 2jn f (x) < j+1
2n
:
for j = 0; 2; :::; n2n 1
If m f (x) < n, then 'm (x) = m and 'n (x) = 2jn . Suppose it were 'm (x) > 'n (x),
that is m > 2jn ; then we would have 2jn < m f (x) < j+1
2n
, and therefore 2jn < m < j+1
2n
,
n
and j < m2 < j + 1 ! .
If 0 f (x) < m, then 'm (x) = 2im and 'n (x) = 2jn , in this case 2im f (x) < 2im + 21m
and 2jn f (x) < 2jn + 21n . Then 2n m i 2n f (x) < 2n m (i + 1) and j 2n f (x) <
j + 1. Suppose it were 'm (x) > 'n (x), i.e. 2im > 2jn , then 2n m i > j j + 1, it would
imply that 2n f (x) 2n m i j + 1 > 2n f (x) ! .
Concluding 'm (x) 'n (x) for each x 2 X and 'n " f .
If f has a lower bound (not necessarily 0) there exists a > 0 s.t. f + a is greater than zero,
therefore there exists a sequence f'n g of simple and measurable functions s.t. 'n " f + a,
then 'n a " f (why?).
Let f 0 and bounded, then there exists n 2 N s.t. f (x) < n for each x 2 X. For
each k n and each x 2 X, f (x) < k, therefore there exists i 2 0; 2; :::; k 2k 1
s.t. 2ik f (x) < i+1
2k
and 'k (x) = 2ik . Thus, for each k n and each x 2 X it is
1
jf (x) 'k (x)j < 2k , then for each k n supx2X jf (x) 'k (x)j < 21k and 'n converges
pointwise to f .
If f is bounded (not necessarily 0), then there exists a > 0 s.t. f + a is greater than zero.
unif:
Then there exists a sequence f'n g of simple and measurable functions s.t. 'n ! f + a,
unif:
and therefore 'n a ! f (why?).
Finally for e generic measurable function f , f = f + f , and there exist two sequences
f'n g,f n g of simple and measurable functions s.t. 'n ! f + and n ! f , then 'n n !
f + f = f (why?). We have already proved that the limit of measurable functions is itself
measurable.
28
8 INTEGRATION OF SIMPLE AND MEASURABLE
FUNCTIONS WITH RESPECT TO A FINITE CHARGE
Let (X; F) be a measurable space and a …nite charge (s.t. (X) < 1) on F, then let
B0 (X; F) denote the set of all simple and measurable functions X.
P
N
Definition. Let ' 2 B0 (X; F), and ' = i Ai be its standard representation. the real
i=1
number
Z X
N
'd = i (Ai )
X i=1
Facts.
[Obvious.]
[Exercise.]
Lemma. If B1 ; B2 ; :::; BM 2 F, there exist C1 ; C2 ; :::; CK 2 F pairwise disjoint s.t. for each
F
l = 1; 2; :::; K, there exists j s.t. Cl Bj and Bj = Cl for each j = 1; 2; :::; M .
l:Cl Bj
Proof. For M = 1 it is obvious. Assume the equality holds for M elements of F. Consider
B1 ; B2 ; :::; BM ; BM +1 2 F, by induction, there exist C1 ; C2 ; :::; CK 2 F pairwise disjoint s.t.
S
for each l = 1; 2; :::; K there exist j s.t. Cl Bj and Bj = Cl for each j = 1; 2; :::; M .
l:Cl Bj
C C
Consider the 2K +1 sets C1 \BM +1 ; C1 \BM C2 \BM +1 ; C2 \BM
+1 ; +1 ; :::; CK \BM +1 ; CK \
C
S
M
BM +1 ; BM +1 n Bj 2 F, and rename them D1 ; D2 ; :::; D2K+1 : these are pairwise disjoint and
j=1
29
S
each of them is contained in Bj for some j. If j < M +1 each Bj = Dh and, moreover:
h:Dh Bj
!
[
M
BM +1 = D2K+1 [ BM +1 \ Bj =
j=1
0 0 11
[
M [
= D2K+1 [ @BM +1 \ @ Dh AA =
j=1 h:Dh Bj
0 0 11
[
M [
= D2K+1 [ @ @ BM +1 \ Dh AA =
j=1 h:Dh Bj
0 0 11
[
M [
= D2K+1 [ @ @ BM +1 \ Dh AA :
j=1 h odd and Dh Bj
S
Then BM +1 = Dh .
h:Dh BM +1
P
M
Proposition. Let ' 2 B0 (X; F), and let ' = j Bj , with Bj 2 F for each j =
j=1
1; 2; :::; M , be a generic representation of '. Then
Z X
M
'd = j (Bj ) .
X j=1
P
N
Proof. Let ' = i Ai be the standard representation of '.
i=1
Let B1 ; B2 ; :::; BM be pairwise disjoint sets. Assume, without loss of generality, that j 6= 0
S
M S
for each j = 1; 2; :::; M . Then Bj = Ai and i (Ai \ Bj ) = j (Ai \ Bj ), for each
j=1 i: i 6=0
j = 1; 2; :::; M and each i s.t. i 6= 0. Then
Z !
X
N X [
M
'd = i (Ai ) = i Ai \ Bj =
X i=1 i: i 6=0 j=1
X X
M X
M X
= i (Ai \ Bj ) = i (Ai \ Bj ) =
i: i 6=0 j=1 j=1 i: i 6=0
X
M X X
M X
= j (Ai \ Bj ) = j (Ai \ Bj ) =
j=1 i: i 6=0 j=1 i: i 6=0
X
M
= j (Bj ) .
j=1
30
If B1 ; B2 ; :::; BM are not pairwise disjoint, then there exist C1 ; C2 ; :::; CK 2 F pairwise disjoint
F
s.t. for each l = 1; 2; :::; K, there exists j s.t. Cl Bj and Bj = Cl for each j =
l:Cl Bj
1; 2; :::; M (it follows from the lemma we have just shown).
Let (
j 1 Cl Bj
l = :
0 else
P
K
j P
K
j
Bj = l Cl and (Bj ) = l (Cl ).
l=1 l=1
Consider the following representation of '
X
M X
M X
K
j
X
M X
K
j
'= j Bj = j l Cl = j l Cl =
j=1 j=1 l=1 j=1 l=1
!
X
K X
M
j
X
K X
M
j
= j l Cl = j l Cl ;
l=1 j=1 l=1 j=1
P
N P
M
Proof. Let ' = i Ai , = j Bj :
i=1 j=1
Z Z !
X
N X
M
( '+ )d = i Ai + j Bj : d =
X X i=1 j=1
X
N X
M
= i (Ai ) + j (Bj ) =
i=1 j=1
Z Z
= 'd + d :
X X
31
Proposition. Let ' and 2 B0 (X; F), if ' , then
Z Z
'd d :
X X
P
N
Proof. Clearly ' 0, let i Ai be its standard representation, then i 0 for each
i=1
i = 1; 2; :::; N and
Z Z Z X
N
'd d = (' )d = i (Ai ) 0:
X X X i=1
dn jf (x) fn (x)j f (x) fn (x) for each x 2 X, and hence f (x) fn (x) + dn ,
dn jf (x) fn (x)j fn (x) f (x) for each x 2 X, and hence fn (x) dn f (x).
gn d0n = fn 3dn fn dn
f fn + dn = gn + d0n ;
unif: unif:
therefore fn dn ! f , similarly, fn dn ! f .
32
Proposition (E). Let f be a measurable and bounded function, then:
8 9 8 9
<Z = <Z =
sup 'd : ' 2 B0 (X; F) ; ' f = inf 'd : ' 2 B0 (X; F) ; ' f .
: ; : ;
X X
unif:
If f'n g is a sequence of simple and measurable functions s.t. 'n ! f , then the common
R
value for the two terms is limn 'n d .
X
Proof. Set: 8 9
<Z =
I (f ) = sup 'd : ' 2 B0 (X; F) ; ' f
: ;
X
and 8 9
<Z =
I (f ) = inf 'd : ' 2 B0 (X; F) ; ' f :
: ;
X
Obviously 1 < I (f ) I (f ) < 1. Let f'n g be a sequence of simple and measurable
unif:
functions s.t. 'n ! f , then there exists fdn g R+ s.t. dn ! 0 and 'n dn f 'n + dn ,
then for each n 2 N;
Z Z
('n dn ) d I (f ) I (f ) ('n + dn ) d
X X
and therefore:
R
dn (X) I (f ) 'n d dn (X) and
X
R
dn (X) I (f ) 'n d dn (X),
X
R
thus I (f ) = limn 'n d = I (f ) :
X
If f is simple considering the sequence 'n = f for each n it’s easy to obtain
Z Z Z
S f d = lim 'n d = f d .
n
X X X
R
R
Consistently with that, for f 2 B (X; F) we use f d instead of S f d .
X X
33
R R R
1. ( f + g) d = fd + gd :
X X X
R R
2. f g ) fd gd :
X X
unif: R unif: R
3. fn ! f ) fn d ! fd .
X X
Proof. Exercise. Hints: 1.) Notice that if f'n g ; f n g are sequences of simple and measurable
unif: unif: unif:
functions s.t. 'n ! f and n ! g, then 'n ! f for each 2 R (verify it), and
unif:
n + n ! f + g (verify it). 2.) Notice that ' 2 B0 (X; F) and ' f implies ' g. 3.)
Use the Lemma we have shown at the beginning of the section.
1. I ( f + g) = I (f ) + I (g) :
2. f g ) I (f ) I (g) :
Then there exists one and only one …nite charge on F s.t.
Z
I (f ) = f d
X
34
is called integral of ' with respect to .
Facts.
[Obvious.]
[Exercise.]
P
M
Proposition. Let ' 2 B0+ (X; F), and let ' = j Bj , with Bj 2 F for each j =
j=1
1; 2; :::; M be a generic representation for '. Then
Z X
M
'd = j (Bj ) .
X j=1
Proof. Repeat exactly that designed for the case of …nite charge.
Proof. Exercise. Hints: 1.) Repeat exactly that designed for the case with …nite charge.
PN P
M
2.) Let ' = i Ai , = j Bj , and let C1 ; C2 ; :::; CK be a partition for fAi \ Bj g; it
i=1 j=1
P
K P
K
can be written ' = k Ck , = k Ck , moreover ' implies k k for each
k=1 k=1
k = 1; 2; :::; K, then....
35
Definition. If ' 2 B0+ (X; F) and E 2 F, set
Z Z
'd = ' Ed :
E X
This extended real number is called Lebesgue integral of ' with respect to over E.
P
N R P
N
1. If ' = i Ai , then 'd = i (Ai \ E) for each E 2 F:
i=1 E i=1
R R R
2. ( '+ )d = 'd + d for each E 2 F:
E E E
R R
3. ' ) 'd d for each E 2 F:
E E
R
4. The function de…ned as ' (E) = 'd for each E 2 F is a measure.
E
R
Proof. Steps 1.,2.,3. are an exercise. Obviously ' (E) = 'd 2 [0; 1] for each E 2 F and
R R R E
' (;) = 'd = ' ; d = 0d = 0: If A; B 2 F and A \ B = ;, then
; X X
Z Z
' (A [ B) = ' AtB d = '( A + B) d =
ZX Z X
P
N
Let ' = i Ai and fBn gn2N F, Bn " B. Notice that, for each i = 1; 2; :::; N ,
i=1
fAi \ Bn gn2N F, Ai \ Bn " Ai \ B, and limn (Ai \ Bn ) = (Ai \ B). Then:
Z !
XN
lim ' (Bn ) = lim i Ai Bn d =
n n X i=1
Z X
N Z X
N
= lim i Ai Bn d = lim i Ai \Bn d =
n X i=1 n X i=1
X
N X
N
= lim i (Ai \ Bn ) = i lim (Ai \ Bn ) =
n n
i=1 i=1
X
N
= i (Ai \ B) = ' (B) :
i=1
36
Definition. Consider a measurable function f : X ! [0; 1]. The extended real number
8 9
Z <Z =
+
L f d = sup 'd : ' 2 B0 (X; F) ; ' f
: ;
X X
R R
is called the Lebesgue integral of f with respect to . If E 2 F, we set f d = f Ed :
E X
R R
If f is simple, clearly f d L f d . For the reverse inequality consider that for each
X R X R R R
' 2 B0+ (X; F), ' f implies 'd f d , and therefore L f d f d . If f is bounded
X X X X
R R R
and …nite, by de…nition S f d = sup 'd : ' 2 B0 (X; F) ; ' f L f d ; on
X X X
the other hand, for each ' 2 B0 (X; F), ' f , '+ 2 B0+ (X; F), ' '+ f , then
R R + R R R R
'd ' d L f d and S f d = sup 'd : ' 2 B0 (X; F) ; ' f L fd .
X X X X X R X
Consistently with this, for f 0 measurable function it is possible to write f d instead of
R X
L fd .
X
R R
Proposition. Let f; g : X ! [0; 1] be measurable functions. If f g, then f d gd :
X X
Proof. Notice that ' 2 B0+ (X; F) and ' f implies ' g.
37
The inequality holds for each 0 < c < 1 and ' 2 B0+ (X; F) s.t. ' f . Then if you choose
c = 1 n1 and pass again to the limit,
Z
'd :
X
R
for each ' 2 B0+ (X; F) s.t. ' f . Finally fd :
X
Notice that, if fn # f , the Theorem doesn’t hold (Consider as an example R with the
Lebesgue measure and the sequences fn (x) = n1 or fn (x) = [n;1) ).
R P
1 1 R
P
2. un d = un d .
X n=1 n=1X
R
3. The function de…ned as f (E) = f d for each E 2 F is a measure.
E
R
4. If (E) = 0, then f (E) = f d = 0.
E
R
5. If f d = 0, then fx 2 X : f (x) 6= 0g = 0.
X
Proof. 1.) Let f'n g ; f n g be sequences of simple measurable and nonnegative functions s.t.
'n " f , n " g, then 'n + n " f + g (why?), and
Z Z
(f + g) d = lim ('n + n ) d =
n
X X
0 1
Z Z
= lim @ 'n d + nd
A=
n
X X
Z Z
= lim 'n d + lim nd =
n n
X X
Z Z
= fd + gd .
X X
R R
Similarly it is possible to prove that: fd = fd .
X X
P
1 P
N
2.) We can write u = un i¤ un " u, and therefore u : X ! [0; 1] is measurable (why?)
n=1 n=1
38
and
Z Z !
X
N
ud = lim un d =
n
n=1
X X
N Z
X
= lim un d =
n
n=1
X
1 Z
X
= un d .
n=1 X
Proof. First of all observe that f = limn fn : X ! [0; 1] is measurable (why?). For each
x 2 X, limn fn (x) = supk2N (inf n k fn (x)), then for each k 2 N set gk (x) = inf n k fn (x).
gk " f and fk gk for each k 2 N. By the Monotone Convergence Theorem
Z Z
limn fn d = lim gn d
n
X X
Z Z
= limn gn d limn fn d .
X X
39
R R
It is limn fn d = 0 and limn fn d
= min ( (A) ; (B)) (why?). The same doesn’t
X X R
hold for limn ( as an example, with fn as above it is limn fn d = (A) + (B) >
R X R
max ( (A) ; (B)) = limn fn d ; on the other hand, for fn (x) = n1 , limn fn d = 0 <
R X X
1 = limn fn d ).
X
11 SUMMABLE FUNCTIONS
Let (X; F) be a measurable space and be a measure on F. If we consider a generic
measurable function f : X ! R (or [ 1; 1]), it is f = f + f where f + ; f : X ! [0; 1]
are measurable functions, smart as a fox...
= f +d f d = L fd .
X X X
R R
Consistently with this, for f measurable we can write (if it exists) f d instead of L f d .
X X
40
Proof. As jf j = f + + f then
Z Z Z
+
jf j d < 1 , f d + f d < 1
X
ZX X
Z
+
, f d and f d <1
X X
Facts.
1. If f; g 2 L and ; 2 R, then f + g 2 L.
[Exercise. Hint: j f + gj j j jf j + j j jgj.]
R R R
2. If f; g 2 L, then ( f + g) d = f d + gd .
X X X
[Exercise. Hint: let h = f + g, h = h+ h , then h+ h = f + f + g + g i.e.
h+ + f + g = f + + g + + h , thus
Z Z
+
h +f +g d = f + + g+ + h d ,
X X
and therefore
Z Z Z Z Z Z
+ + +
h d + f d + g d = f d + g d + h d .
X X X X X X
R R R
Reordering we can conclude (f + g) d = fd + gd (verify that each step
X X X
is legitimate thanks to properties of integrals for nonnegative functions). To show
R R
( f) d = f d , verify it for > 0,and for = 0, = 1, then put them
X X
together.]
R R
3. If f; g 2 L and f g, then f d gd .
X X
[Exercise. Hint: f g i¤ f g 0.]
R R
4. If f 2 L, then fd jf j d .
X X
[Exercise. Hint: f; f jf j.]
41
6. If f is summable on A; B 2 F with A \ B = ;, then f is summable on A t B and
Z Z Z
fd = fd + fd :
AtB A B
[Exercise.]
R
7. If f 2 L and f d = 0 for each A 2 F, then (fx 2 X : f (x) 6= 0g) = 0.
A S
[Exercise. Hint: ff > 0g = n f > n1 .]
R R R R R
In particular, fd fn d jf fn j d ! 0, that is fn d ! f d .
X X X X X
42
12 PROPERTIES HOLDING ALMOST EVERYWHERE
Let (X; F) be a measurable space and be a complete measure on F.
Since is complete it is enough to show that there exists N 2 F s.t. fx 2 X : on x P doesn’t holdg
N and (N ) = 0.
43
R R R R R R R R
2.) fd = fd + fd = f Ncd and gd = gd + gd = g Ncd , but
X X N N X X X N N X
f Nc = g Nc.
R
3.) Set M = fx 2 X : jf (x)j =6 jg (x)jg N , then jf j = jgj a.e., if f 2 L, then jgj d =
R X
jf j d < 1 and g 2 L. Then setting M + = fx 2 X : f + (x) 6= g + (x)g and M =
X R R
fx 2 X : f (x) 6= g (x)g, it is M + ; M N (why?), and therefore f d = gd (why?).
X X
Facts. Let f; fn ; g; f 0 ; fn0 ; g 0 : X ! R (or [ 1; 1]) s.t. f = f 0 a.e., fn = fn0 and g = g 0 a.e.
R
[Set A = fx : f (x) = 1g, f n A for each n 2 N, if (A) > 0, then fd n (A)
R X
for each n 2 N implies X f d = 1 ! .]
Proof. Let:
44
N be the set on which g is not de…ned,
(A) = 0. Let g; f ; fn be the functions that coincide with g; f; fn on Ac and are zero on A.
g; f ; fn satisfy the hypothesis of the Dominated Convergence Theorem, therefore
Z
f fn d ! 0;
X
R R
but f fn d = jf fn j d for each n 2 N (why?).
X X
Consider …nite measure and bounded functions, then the following "counterpart" of Pro-
position (E) holds
Then f is measurable.
Proof. There exists a sequence f' bn g B0 ( ; F) s.t. ' bn f for each n 2 N, and
R
lim ' bn d = I (f ) (why?). Let 'n = max f' b1 ; '
b2 ; :::; '
bn g for each n 2 N, then f'n g
n!1
R R R
B0 ( ; F), and ' bn 'n 'n+1 f for each n 2 N. Thus ' bn d 'n d 'n+1 d
R X X X
I (f ) for each n 2 N, and hence 'n d " I (f ). Similarly, it is possible to build a sequence
R
f n g B0 ( ; F) s.t. f n+1 n and n d # I (f ). Set f = lim 'n and f = lim n ,
X
then
'n f f f n
for each n 2 N, and f f = 0 a.e. (why?), then f = f a.e.. Notice that x 2 X : f (x) 6= f (x)
x 2 X : f (x) 6= f (x) , and therefore f = f a.e.
45
13 RIEMANN (STIELTJES) Vs LEBESGUE (STIELTJES)
Let X = [a; b]. (E) delivers the following:
Corollary. Let be the Lebesgue measure on [a; b], and F be the Lebesgue -algebra. If
the function f : [a; b] ! R is Riemann integrable, then f is summable and, if we denote by
R
R f (x) dx the Riemann integral of f , it is
[a;b]
Z Z
fd = R f (x) dx.
[a;b] [a;b]
R R
Proof. Let R f (x) dx and R f (x) dx be the upper and lower Riemann integral of f . Notice
[a;b] [a;b]
that
8 9
Z <Z =
R f (x) dx sup 'd : ' 2 B0 ([a; b] ; F) ; ' f
: ;
[a;b] X
8 9
<Z =
inf 'd : ' 2 B0 ([a; b] ; F) ; ' f
: ;
X
Z
R f (x) dx (why?),
[a;b]
R R
but R f (x) dx = R f (x) dx.
[a;b] [a;b]
Proposition. Let be the Lebesgue measure on [a; b], F be the Lebesgue -algebra;
consider a bounded function f : [a; b] ! R. Then, f : [a; b] ! R is Riemann integrable i¤ it
is continuous a.e..
Proof. See, as an example, Royden p.85.
46
integral of f , it is
Z Z
f d = f (a) F (a) F a +R f (x) dF (x) .
[a;b] [a;b]
R R
Proof. Let R f (x) dF (x) and R f (x) dF (x) the upper and lower Riemann-Stieltjes integ-
[a;b] [a;b]
ral of f . Notice that
8 9
Z <Z =
f (a) F (a) F a +R f (x) dF (x) sup 'd : ' 2 B0 ([a; b] ; F) ; ' f
: ;
[a;b] X
8 9
<Z =
inf 'd : ' 2 B0 ([a; b] ; F) ; ' f
: ;
X
Z
f (a) F (a) F a +R f (x) dF (x) (why?),
[a;b]
R R
but R f (x) dF (x) = R f (x) dF (x).
[a;b] [a;b]
F G = fA B : A 2 F; B 2 Gg
: F G ! [0; 1]
A B 7! (A) (B)
is a measure on F G.
Proof. Since X 2 F and Y 2 G, X Y 2 F G; similarly ; = ; Y 2 F G. If
A B; C D 2 F G, then (A B)\(C D) = (A \ C) (B \ D) 2 F G. Finally, if A
B 2 F G, then (A B)c = (A B c )t(Ac B)t(Ac B c ). Thus F G is a semialgebra.
It can be easily veri…ed that ( ) (;) = 0. Consider the sequence fAn Bn gn2N F G
47
F
and set A B = 1 n=1 An Bn 2 F G. Then let fAn Bn g = 6 ; for each n 2 N.
Fix x 2 A. Set Nx = fn 2 N : x 2 An g = fn 2 N : An (x) = 1g. For each y 2 B, (x; y) 2
A B, and there exists a unique n s.t. (x; y) 2 An Bn . This implies that for each y 2 B
there exists an unique n s.t. x 2 An and y 2 Bn . It follows that for each y 2 B there exists
S
an unique n 2 Nx s.t. y 2 Bn . Therefore B Bn , vice-versa, if there exists n 2 Nx s.t.
n2Nx S
y 2 Bn , then x 2 An and y 2 Bn , i.e. (x; y) 2 An Bn thus y 2 B and Bn B. If there
n2Nx
existed n; m 2 Nx s.t. y 2 Bn \ Bm , then x 2 An and y 2 Bn , so (x; y) 2 An Bn , but also
F
x 2 Am and y 2 Bm , so (x; y) 2 Am Bm , which is absurd if n 6= m. Thus, B = Bn and
n2Nx
X X
(B) A (x) = (B) = (Bn ) = (Bn ) An (x) .
n2Nx n2N
If and are -…nite, then is -…nite (why?) and it is the unique extension of
to F G.
From now on we will consider -…nite and .
Qx = fy 2 Y : (x; y) 2 Qg and
Qy = fx 2 X : (x; y) 2 Qg .
48
It is straightforward to see that maps Q 7! Qx and Q 7! Qy preserve all the usual operations
between sets.
T = fQ 2 F G : Qx 2 G and Qy 2 F 8 (x; y) 2 X Yg
if x 2 A, Qx = fy 2 Y : (x; y) 2 A Bg = B 2 G,
if x 2
= A, Qx = fy 2 Y : (x; y) 2 A Bg = ; 2 G,
thus Qx 2 G for each x 2 X, a similar argument can be used for Qy for y 2 Y , then
F G T.
It remain to show that T is a -algebra.
X Y 2F G T.
S
Let fQn g T and Q = 1 n=1 Qn . Then Q 2 F G. Moreover: for each x 2 X,
Qx = fy 2 Y : (x; y) 2 Qg =
= fy 2 Y : 9n s.t. (x; y) 2 Qn g =
= fy 2 Y : 9n s.t. y 2 (Qn )x g =
[1
= (Qn )x 2 G;
n=1
S1
similarly, for each y 2 Y , Q = n=1 (Qn )y 2 F. Then, Q 2 T .
y
(Qc )x = fy 2 Y : (x; y) 2 Qc g =
= fy 2 Y : (x; y) 2
= Qg =
= (Qx )c 2 G;
49
fx is called x-section of f , analogously f y is called y-section of f . It is easy to see that maps
f 7! fx and f 7! f y preserve all the usual operations.
fx 1 (A) = fy 2 Y : fx (y) 2 Ag =
= fy 2 Y : f (x; y) 2 Ag =
1
= y 2 Y : (x; y) 2 f (A) =
1
= f (A) x
:
R R y
Notice that p (x) = Y
( Q )x d and q (y) = X
( Q) d .
Proof. It is enough to show that the class T of subsets of Q 2 F G s.t. p and q are
measurable and s.t. it is Z Z
pd = ( ) (Q) = qd ,
X Y
coincides with F G. The proof is long: we develop it through several steps.
Step 0. F G T .
If Q = A B 2 F G, then A 2 F and B 2 G, and Q 2 F G. Moreover,
(
(B) if x 2 A
p (x) = (Qx ) = = (B) A , and
0 if x 2
=A
(
(A) if y 2 B
q (y) = (Qy ) = = (A) B .
0 if y 2
=B
50
Thus, p and q are measurable and
Z Z
pd = (A) (B) = qd ,
X Y
therefore Q 2 T .
Step 1. If fQn g T , and Qn " Q, then Q 2 T .
(Obviously Q 2 F G.) Let pn (x) = ((Qn )x ) and qn (y) = ((Qn )y ). For each n 2 N, pn
and qn are measurable and
Z Z
pn d = ( ) (Qn ) = qn d .
X Y
(Q R)x = fy 2 Y : (x; y) 2 Q Rg =
= fy 2 Y : (x; y) 2 Q and (x; y) 2 Rc g =
= fy 2 Y : (x; y) 2 Qg \ fy 2 Y : (x; y) 2 Rc g =
= Qx \ (Rc )x = Qx \ (Rx )c = Qx Rx
51
qQ R (x) = ((Q R)y ) 8y 2 Y .
Notice that
pQ R = pQ pR and qQ R = qQ qR ;
but pQ , pR , qQ and qR are summable, thus pQ pR and qQ R = qQ qR are summable, and
it is
Z Z
pQ d = ( ) (Q) = qQ d and
X Y
Z Z
pR d = ( ) (R) = qR d .
X Y
If Q 2 F G, then Q = A B, and
Q \ (An Bn ) = (A \ An ) (B \ Bn ) 2 T
thus F G T 0 .
If fQm g T 0 , and Qm " Q, then Q 2 T 0 . Fixing n 2 N, for each m 2 N Qm \(An Bn ) 2 T ,
but (for m ! 1) Qm \ (An Bn ) " Q \ (An Bn ) and by Step 1, Q \ (An Bn ) 2 T .
Since n 2 N is arbitrary, Q 2 T 0 .
If Q; R 2 T 0 , Q R, then Q R 2 T 0 . Fixing n 2 N, Q\(An Bn ) 2 T , R\(An Bn ) 2 T ,
R \ (An Bn ) Q \ (An Bn ) An Bn , by Step 2, Q \ (An Bn ) R \ (An Bn ) 2 T ,
then (Q R) \ (An Bn ) = Q \ (An Bn ) R \ (An Bn ) 2 T . Since n 2 N is arbitrary,
Q R 2 T 0.
Thus T 0 is a -class containing F G and, by Dynkin’s Lemma, T 0 (F G) = F G,
0
the reverse inclusion is trivial, and T = F G. It follows that, for each Q 2 F G,
Q \ (An Bn ) 2 T for each n 2 N, but Q \ (An Bn ) " Q and therefore, by Step 1, Q 2 T .
Then we have F G T ; The reverse inclusion is trivial.
52
R R
If h : X ! R (or [ 1; 1]) we sometimes write X
h (x) d (x) instead of X
hd .
53
Then, p and q are measurable (with respect to F and G) and
Z Z
pd = ( ) (Q) = qd .
X Y
But
Z
p (x) = (Qx ) = Qx (y) d (y) for each x 2 X, and
ZY
q (y) = (Qy ) = Qy (x) d (x) for each y 2 Y ,
X
then
Z Z Z
Qx (y) d (y) d (x) = Q (x; y) d ( ) (x; y)
X Y X Y
Z Z
= Qy (x) d (x) d (y) .
Y X
thus ( Q )x = Qx , analogously Qy = ( Q )y .
Therefore, for each Q 2 F G, the functions de…ned as
Z
p (x) = ( Q )x (y) d (y) for each x 2 X, and
ZY
are measurable (with respect to F and G), thus c) and d) holds and the functions are
nonnegative, moreover
Z Z Z
( Q )x (y) d (y) d (x) = Q (x; y) d ( ) (x; y)
X Y X Y
Z Z
= ( Q )y d (x) d (y) ,
Y X
54
b) Analogously the function ( f + g)y = f y + g y is measurable with respect to F for
each y 2 Y .
R R
c) pf (x) = Y fx d is measurable with respect to F and pg (x) = Y gx d is measurable
with respect to F, then is measurable with respect to F the function pf + pg de…ned
as
Z Z
( pf + pg ) (x) = fx d + gx d =
Z Y Y
= ( fx + gx ) d =
Y
Z
= ( f + g)x d
Y
R
or as x 7! Y
( f + g)x d .
R
d) Analogously the function y 7! X
( f + g)y d is measurable with respect to G.
gx d d (x) = gd ( )= gy d d (y)
X Y X Y Y X
it follows
Z Z Z
( f + g)x d d (x) = ( pf + pg ) (x) d (x) =
X Y X
Z Z
= pf (x) d (x) + pg (x) d (x) =
X X
Z Z Z Z
= fx d d (x) + gx d d (x) =
X Y X Y
Z Z
= fd ( )+ gd ( )=
X Y X Y
Z
= ( f + g) d ( )=
X Y
Z Z
= ( f + g)y d d (y) .
Y X
55
b) Analogously the function f y is measurable with respect to F for each y 2 Y .
R
c) pfn (x) = Y (fn )x d is measurable with respect to F for each n 2 N. For each x 2 X,
R
(fn )x " fx ; then, by the Monotone Convergence Theorem, pfn (x) = Y (fn )x d "
R R
f d = pf (x), that is pfn " pf , and pf (x) = Y fx d is measurable with respect to
Y x
F.
R
d) Analogously the function y 7! X f y d is measurable with respect to G.
f is summable,
R R
X Y
jf (x; y)j d (y) d (x) < 1;
R R
X Y
jf (x; y)j d (y) d (x) < 1:
Proof. Exercise.
56
b) the function x 7! f (x; y) is summable with respect to for a.e. every y 2 Y ,
R
c) the function x 7! Y f (x; y) d (y) is de…ned a.e. and is summable with respect to ,
R
d) the function y 7! X f (x; y) d (x) is de…ned a.e. and is summable with respect to ,
Proof. We use the notation already introduced for the Proof of Tonelli’s theorem. We only
prove claims a), c), and the …rst line of the equality e) (the others can be obtained in a
similar fashion, using Y instead of X).
Notice that jf jx = jfx j, (f + )x = (fx )+ and (f )x = (fx ) for each x 2 X. f + and f are
summable and nonnegative since 0 f + ; f jf j. The functions de…ned for each x 2 X as
R + R
pf + (x) = Y fx d and pf (x) = Y fx d are F measurable and nonnegative,moreover it is
Z Z Z Z
+
pf + d = fx d d (x) = f +d ( ) < 1 and
X X Y X Y
Z Z
pf d = f d( ) < 1.
X X Y
Thus pf + and pf are summable and …nite a.e.. For each x s.t. pf + (x) ; pf (x) < 1, it
R R R
is Y fx+ d ; Y fx d < 1 (in particular fx is summable for a.e. x 2 X), then Y fx d =
R + R R
f
Y x
d f d
Y x
, is well-de…ned and pf : x 7! Y f (x; y) d (y) is de…ned a.e.. Moreover
it is
Z Z Z Z Z Z
jpf j d = fx d d (x) jfx j d d (x) = jf j d ( ) < 1.
X X Y X Y X Y
R
And x 7! Y f (x; y) d (y) is summable.
R R R
Finally, note that Y fx d = Y fx+ d f d is the same as pf = pf +
Y x
pf , and
Z Z Z
f (x; y) d (y) d (x) = pf d =
X Y X
Z Z
= pf + d pf + d =
X Y
Z Z
+
= f d( ) f d( )=
X Y X Y
Z
= fd ( ):
X Y
57
If and are not complete, it is always possible to complete them. If and are
complete, in general is not complete but is always possible to complete it. Denoted by
0
( ) the completion of , consider functions f that are measurable with respect to
0
the -algebra (F G) of subsets ( )0 -measurable, then Tonelli ad Fubini theorems still
hold with the following
N
When instead of 2, we are facing N measurable spaces f(Xi ; Fi ; i )gi=1 it is straightfor-
ward to extend de…nitions and theorems. (Useful exercise.)
8E 2 F, then:
1. f (;) = 0.
58
S P1
2. fAn gn2N F and Ai \ Aj = ; for each i 6= j implies f ( 1n=1 A n ) = n=1 f (An ).
(i) (;) = 0,
S P1
(ii) fAn gn2N F and Ai \ Aj = ; for each i 6= j implies ( 1n=1 An ) = n=1 (An ).
Facts.
3. If is a signed measure, fAn gn2N F and An " A (or An # A), then (A) =
limn (An ).
F1
[Exercise. Hint: set A0 = ; and Bn = An An 1 , A = n=1 Bn .]
Definition. A 2 F is:
59
Lemma. If A is positive, then each E 2 F s.t. E A is positive. If An is positive 8n 2 N,
S
then n An is positive.
Sn 1
Proof. The …rst part is obvious. Let B1 = A1 and Bn = An j=1 Aj for each n 2.
F1 S1 S1
For each n 2 N, Bn is positive and n=1 Bn = n=1 An (why?). 8E 2 F s.t. E n=1 An ,
P1
(E) = n=1 (E \ Bn ) 0 .
1
(B)
n1 1
(why?).
Suppose that for …xed k there existed E1 ; E2 ; :::; Ek 2 F e n1 ; :::; nk 2 N s.t.:
Sj 1
Ej E i=1 Ei for each j = 1; :::; k,
1
(Ej ) nj
for each j = 1; :::; k,
Sj
for each B E i=1 Ei ,
1
(B)
nj 1
for each j = 1; :::; k.
Fk Pk Fk
Then, E i=1 Ei = (E) i=1 (Ei ) > (E), and E i=1 Ei is not positive;
Fk
let nk+1 the smallest integer s.t. it exists F 3 Ek+1 E i=1 Ei with
1
(Ek+1 ) :
nk+1
1 Fk Fk+1
With the convention = 1; for each B E i=1 Ei Ek+1 = E i=1 Ei
0
1
(B) :
nk+1 1
60
F1 P1
Let A = E k=1 E k , then (A) = (E) k=1 (Ek ) (E) and A is not positive.
P1 1
Since k=1 (Ek ) converges, (Ek ) ! 0 as k ! 1. Thus (Ek ) nk
0 implies n1k ! 0
Sk
for k ! 1, nk ! 1 for k ! 1, and nk1 1 ! 0. If B A, then B E j=1 Ek for
each k 2 N, therefore
1
(B)
nk 1
and, passing to the limit for k ! 1, (B) 0. It follows that A is positive ! .
(Exercise.)
Hahn’s Decomposition theorem. Let (X; F) be a measurable space and a signed
measure on F, then there exist A; B 2 F s.t. A is positive, B is negative, A \ B = ;,
A[B =X
Proof. Let = supA2F (A). If = 0, X is negative and we are done. Otherwise, consider
a sequence fAn g 2 F s.t. (An ) ! . For n big enough (An ) > 0 and, choosing Bn 2 F
S
with Bn An , Bn positive, (Bn ) (An ); it follows that (Bn ) ! . 1k=1 Bk is positive
S1 S1
and (Bn ) ( k=1 Bk ) for each n 2 N, passing to the limit ( k=1 Bk ) = . Set
S1
A = k=1 Bk . If C A and (C) > 0, then (A t C) > ! . Set B = Ac .
c
Facts.
1. If fA; Bg and fA0 ; B 0 g are two Hahn decomposition of , then A \ B 0 and B \ A0 are
null sets. Thus
A = A\A0 = A0 and B = B\B 0 = B 0 .
3. Let and be two …nite measures and fA; Bg be a partition of X s.t. (B) = (A) =
0 (we say and are mutually singular), then fA; Bg is an Hahn decomposition of
= . Moreover = A and = B.
61
[Exercise.]
+ +
= A; = B; j j= +
16 RADON-NIKODYM THEOREM
Let (X; F) be a measurable space.
Facts.
62
2. If is a signed measure and , then + ; ; j j .
[Exercise. Hint: let fA; Bg be an Hahn decomposition of , if E 2 F and (E) = 0,
then + (E) = (E \ A), but (E) = 0 ) (E \ A) = 0.]
Lemma. If and are …nite (positive) measures on F, 6= 0 and , then there exist
1
n 2 N and E 2 F s.t. (E) > 0 and E is positive with respect to n
.
1
Proof. Let X = An t Bn be an Hahn decomposition of n
for each n 2 N (with An
S1 T
positive and Bn negative with respect to 1
n
). Set A = n=1 An and B = 1 n=1 Bn
T1 c c
(= n=1 An = A ). For each n 2 N,
1
(Bn ) (Bn ) ;
n
1 1
then (B) (Bn ) n
(Bn ) n
(X), therefore (B) = 0. Then (A) > 0 ( (A) +
P 1
(B) = (X) > 0), but (A) n=1 (An ), and there exists n s.t. (An ) > 0. An is
1
positive with respect to n
and (An ) > 0 implies (An ) > 0. Set E = An .
d
Such f is sometimes denoted by and is called the Radon-Nikodym derivative or
d
density of with respect to .
Proof. It is enough to prove the theorem for …nite (positive) measure and 6= 0 (why?).
Set 8 9
< Z =
K = f : X ! [0; 1] measurable : fd (A) 8A 2 F .
: ;
A
R R
K 0 and 8f 2 K, f d (X) < 1 i.e. f 2 L (X; F; ). Let M = supf 2K f d , there
R X X
exists fn 2 K s.t. fn d ! M for n ! 1. For each n 2 N and each x 2 X set
X
63
and denote by N the set f1; 2; :::; ng. For each E 2 F set
\
E1 = fx 2 E : f1 (x) fj (x) 8j 2 N g = E \ ff1 fj g
j2N
\
E2 = (E E1 ) \ ff2 fj g
j2N
:::
n[1
!
\
En = E E1 \ ffn fj g
j=1 j2N
and gn 2 K for each n 2 N. Moreover gn (x) " sup fn (x) for each x 2 X, set f = sup fn .
Applying the Monotone Convergence Theorem, for each E 2 F,
Z Z
f d = lim gn d (E)
n
E E
and f 2 K, moreover Z Z
fn d gn d M
X X
R
implies f d = M .
X R
The set function de…ned as (E) = (E) f d , for each E 2 F, is a …nite (positive)
E
measure and it is absolutely continuous with respect to . If 6= 0, then there exist " > 0
and A 2 F such that (A) > 0 and A is positive with respect to " , that is
(E \ A) " (E \ A)
64
and h 2 K, but Z Z
hd = f d + " (A) > M
X X
Corollary. Let (X; F) be a measurable space and let and be -…nite measures on
F with absolutely continuous with respect to . Then there exists a measurable function
f : X ! [0; 1) such that Z
(E) = fd
E
if g 2 X ! [0; 1] is measurable,
Z Z
gd = (g f ) d ;
X X
Proof. It is an useful exercise. Hint: consider the countable partition fAn g of X in F s.t.
d An
(An ) + (An ) < 1 and set f (x) = (x) if x 2 An .
d An
65
17 ESSENTIAL BIBLIOGRAPHY
Berberian, S.K. (1965), Measure and Integration, The Macmillan Company, New York.
Kolmogorov, A.N., e S.V. Fomin (1975), Introductory Real Analysis, Dover, New York.
(Translation of Elementy Teorii Funktsij i Funktsional’nogo Analiza, Nauka, Mosca,
1968.)
66