Sie sind auf Seite 1von 2

Section 3

Kolmogorovs Theorem about


consistent distributions.

Given a probability space (, A, P), one can dene for a r.v. X : R a sample space (R, B, P X ) and
for a random vector X : Rk we can dene a sample space (Rk , B, PX ), where B are Borel sets in Rk .
Suppose we have an innite sequence or r.v.s {Xi }i1 such that the distribution of any nite subset of them
is well dened. We would like to dene events that are expressed in terms of the entire sequence, for example,
{supi1 Xi 1}.
More generally, let us consider a family of r.v.

{Xt }tT RT = Rt = {f : T R}.
tT

For any nite subset N T let PN be the law of {Xt }tN on the Borel -algebra BN on RN . Suppose that
these laws satisfy a natural consistency assumption: for any nite subsets N M and any Borel set B B N ,
PN (B) = PM (B RM N ).
We would like to dene a sample space for the entire family {Xt }tT . First, let us consider the following
algebra of sets on RT :
A = {B RT N : B BN }.
It is easy to check that A is an algebra. Let us dene a function P on A by
P(B RT N ) = PN (B).
By consistency, P is well dened. Given two nite subsets N1 , N2 T and B BN1 , the same set can be
respresented as
B1 RT N1 = B1 R(N1 N2 )\N1 RT \(N1 N2 ) .
However, by consistency, P will not depend on the representation.
Let A = (A) be a -algebra generated by algebra A. A is called a cylindrical algebra and A is a
cylindrical -algebra. A set B RT N is called a cylinder and B is a base of the cylinder.
Can we extend P from A to A = (A)? To use Caratheodory extension theorem, we need to show that
P is -additive on A or, equivalently, continuity of measure property: given a sequence B n A :

Bn Bn+1 , Bi = = P(Bn ) 0.
i1

Proof. We will prove that if there exists > 0 such that P(Bn ) > for all n then n1 . We
Bn =
have
Bn = Cn RT Nn , Nn - nite subset of T and Cn BNn .

9
Since Bn Bn+1 , we can assume that Nn Nn+1 . First of all, by regularity of measure PNn there exists a
compact set Kn Cn such that

PNn (Cn \ Kn ) n+1 .
2
We have,
Ci RT Ni \ Ki RT Ni (Ci \ Ki ) RT Ni
in in in

and, therefore,

P Ci RT Ni \ Ki RT Ni P (Ci \ Ki ) RT Ni

in in in


P (Ci \ Ki ) RT Ni i+1
.
2 2
in in

Since
P(Bn ) = P Ci RT Ni >
in

this implies that


P Ki RT Ni > 0.
2
in

We can write
Ki RT Ni = (Ki RNn Ni ) RT Nn = K n RT Nn
in in

n
 Nn Ni
where K = in (Ki R ) is a compact in RNn , since Kn is a compact in RNn . We proved that

PNn (K n ) = P(K n RT Nn ) = P Ki RT Ni > 0
in

and, therefore, there exists a point

xn = (x1n , . . . , xnNn , . . .) K n RT Nn .

We also have the following inclusion property. For m > n,

xm K m RT Nm K n RT Nn

and, therefore, (xm m n


1 , ..., xNn ) K .
We proceed using triangulation method. Any sequence on a compact has a converging subsequence.
n1 k
Let {n1k }k1 be such that x1 k x1 K 1 . Then we can take a subsequence {n2k }k1 {n1k }k1 such that
n1 n2 m1
(x1 k , x2 k ) (x1 , x2 ) K 2 By iteration, we can nd a subsequence {nm
k }k1 {nk }k1 , such that
nm nm
(x1 k , ..., xNkm ) (x1 , ..., xNm ) K m .

Therefore, a point
(x1 , x2 , . . .) K m RT Nm ,
m1

so this last set is not empty.

10

Das könnte Ihnen auch gefallen