Math 626 1 23 -

01 -
13

Measure
1.

theory
↓ Measure
spaces
and
probability spaces.
def. Let 20. A collection of subsets of his said to be
a
w-algebra on to
if
(If FC, then F'EF, where Fc=MIE
(2) If FntF, h21, then
NEn-F.

mark: ·
A Nalgebra is closed under the
complement, countable union
and countable intersection
&
EFn =) nEFnY) "I
·
Let
4B:biet be a
family of N-algebras on 1.
Then
also salgebra
Nor is a on
a

Def. Let to be a collection

of subsets of wh
Let
5(*) denote the smallest W-algebra on 1 that contains N.
we call 5)) the W-algebra generated by 1.

ample.
# Let X be a topological space. Let B(X) denote the
5-algebra generated by the collection of open sets in X.
we call ((X) the Borel Malgebra.
 Let B(IR) denote the Borel Falgebra on IR. Each element in BCR) is

called a Borelset in M.
mem

I able (v.) called measurable

Def. space) space if
is a
a m

&** and B is a 5-algebra on 1.

Def. ) measure A
function 4: F-50,01 is called a

measure on
(R, 5) if

(i)((x) 0.
=

(ii) M) An) =
M)An) if
An, n21, are
disjoint elements in

Prop 1.1. Let

me be a measure on (, 5). Then

(i) MSA) M(B) if Al B smonotonicity)

(ii) MSEAn) <tMJAn) for Ant F. (sub-additivity

viiIfANNA,thenfilAult MCAT acontinuity from below as

loM(An) =
M(A). (continuity from above).
Def.· J2b, F, M) called
space if
is
A
triple a measure
i is
a measure on (2, F).
If M(2) =1, call i prob. measure. correspondingly,
· we a

&, F, M) called
is a
prob space.
mu

Usually a
prob. measure is denoted as
I

ample. (discrete prob. space

#

Let t be a countable set. Let

# 2e=

(A:A5r).
=

Then Sr, F) is measurable

a
space.
Let [P(msSwer be a
prob.vector, i.e. PCW) 20 and
EP(W) =1
Define
(A)
P) =

EqP(w) for all Ath

then J, oF,P) is a
(discrete) prob. space.

Example (Bovel measure on IR) A measure

(on JIR, B(R1) is

called a Borel measure on H.

 Let be Borel R. Set
Prop 1.2.
I prob. measure
a on

F(x) =

4)(-0,x1) for set.

Then
(1) F is
non-decreasing, i.e. FCXLEF(2) If x<Y.
2)
F is right-continuous, i.e.

E(y) (
lim
=
F

(3) lim
F(x) 1, limo F(x) 0.
=

*
- +r

Pf. (1) is trivial. (2)&(3) follow from the

continuity property
of a
prob. measure.

1. 2
-
Random variables and their distributions.

Let Jv3, F,P) be a

probability space.
ef.
& A
function X: --> A is said to be F-measurable
if
*JASEF for every Borel set ACI.

If so, we call X a random variable (v.U.).

C2, F,P) be discrete

Example: · Let a
prob. space.
Then
any function
X: -> 1 is a r.U.
 ·
Let Ji, B,P) be a
general prob. space and let At F.
efine
& A: 2 -> R
by
XEA,
S
H if
A(x)
# =

0 otherwise.

then A is a rr. which is called the indicator function

of A.

(check: 1,913 =
A.1,'90] =
A"

Defo Let X be a rc. on (1, F,P). Then X induces a

prob
1
measure
I on
by
↑(A) =
((X + A): P(X (As),
=

AE B(R).
We call
in
the
famibution of X.

Moreover, set F(x) =

PSX<x} =
M(70, x2) for xER;
we call it the
defamation
function of X.

If Fox has the form

F(x) =
Sfloldy,
then we
say X
has the
density function fo
Example: 0
(uniform distribution on
(0, 1).
·

f(x) =

9% for n*5OT
pro,

if
4
x = (
f(x)
·
=

if 0 < X ->
1
0
if Xs0.

&Sexponential distribution with

parameter x
(

f(x) xex
s if
=

X30

O otherwise

j-xx
S if
1-
f(x) =
x>0

O
if X40

(standard
& worm distribution (

f(x) =
e, NEIR.
 ↓3. Random elements and random vectors.

Now we
generalize the
concept of r.U.

Def. A
map X: 2-T from JM, F.P) to a measurable
space
said measurable
IT,J) is to be
if
*"JACE F for every AtT.

random element (T, 5).

In this case, we call X a
of
If(T, Y) =
(RRP, BURds), then we call X a random vector.

Def. Let X: A+T be a random element. Set

5(X) =
9X SA): AtY].
we call it the main
generated by X.

Below we
give a
useful result to check the measurability of X: 1-T.

#1.3. Let X: -->T be a

map. Suppose A is a collection

of subsets of such that 5(N) =

J. Then

*
is F-measurable () X(A)EF for all AEA.

of. EU:= PACT: X"SALE 5). Then D is a

5-algebra.
and contains A. Hene 5c 5(N) J.M. =
Prop 14. If X: (1,4) -> (5, 5)
and f: (T, 5) ->
U, 81) are
measurable,
then is
so
f(x). (n,F) ->
(V,f).

Pf. Let At81. Then f"CALES. Thus

x"(f"(Al) - F.

Hence
(f(X))(A) =
X" (f"(A))-F. M

·
Extended real line (R* F-r0,01.
=

ndow
=> IR* with the
topology generated by
t0,a),(a,b),jb, + 0]

Let B(IR*) denote the Borel 5-algebra on

M*.

Ameasurable X: (2, F) -> (1R*, B(R*)) is also

called a
map
random variable.

r.r's. Then
Prop. 1.5. Let X., Xn, ..., be

infXn, supXn, HimXn, Tim An

are all v.v.'s.

1.4
Integration
Let J, F, M) be a measure
space
and
fil+** be

measurable.

Then we can
define
& fdp =
(ftdm SFdm -

if one
of Iftdl, Sfidy is
finite.

We call
of integrable if (Ifldhesco, and write

=(*(m, 5,M) or ((*).

Moreover are write for(M if Sif1"dress
llflp: ((1f1dn)" spnorm
and >

of f
Basic
inequalities:
rinequality:
# Let
p, C be such that
T+ =1. Then

& Ife1dm<)SIf(°dn) *)(191dn) **

ski
inequality:
1) f+
allp =11fllp +1181lp for all pal.
 Let to be a
prob. measure.

inequality ag:IR
->IR be convex, i.e.
#

pg(x) + (1 -

p)g(z) 9(Px
=
+ (1 -

p(y)
for all
expel and x.3t1R.
Suppose I and
91f) are

integrable. Then

3) SfdN) < (oof dri

Pf. Write CFSfdM.
Since
I is convex, there exists a function
f(x) = ax +
b

such that l() =

P() and 9(x) = ((X) for all otR.

see the following picture.

Xsa
*
ex
Hence

3f(x) l(f(x) afxx

=
=

+ b

Taking integration gives

<pofdm = ((a f(x) + b)dm aSfdn

=

+b

=
l(x) g(). =
Next we recall some
convergence results.

· (Monotone convergence thm) Let r, n21, be non-negative functions

such that In face. Then

SEndn ->Sfdm as n+ n

Faton's Lemma:Let r, n21, be

non-negative measurable functions.
then
tmoStudia (tm fudi.

ted
convergence
thm:

frtface Ifrl>8 for all

Suppose and n

and lady so. Then

Studie ->Sfdm.