General Theory of Stochastic Process
General Theory of Stochastic Process
General Theory of Stochastic Process
1. I.i.d. Noise. Let {Xn : n Z} be independent and identically distributed (i.i.d.) random
variables. This stochastic process is sometimes called the i.i.d. noise. A realization of this
process is shown in Figure 1, left.
3
10
5
1
0
20 40 60 80 100 20 40 60 80 100
-1
-5
-2
-10
-3
Figure 1. Left: A sample path of the i.i.d. noise. Right: A sample path of
the random walk. In both cases, the variables Xn are standard normal
We will call F the cylinder -algebra. Equivalently, one could define F as the smallest
-algebra on which makes the coordinate mappings Xt : R measurable.
Sometimes cylinder sets are defined as sets of the form
AB
t1 ,...,tn := {f : (f (t1 ), . . . , f (tn )) B},
where t1 , . . . , tn T and B B(Rn ). One can show that the -algebra generated by these
sets coincides with F.
Step 4. We define a probability measure P on (, F). We want the distribution of the
random vector (Xt1 , . . . , Xtn ) to coincide with the given probability measure Pt1 ,...,tn , for all
t1 , . . . , tn T . Equivalently, we want the probability of the event {Xt1 B1 , . . . , Xtn Bn }
to be equal to Pt1 ,...,tn (B1 . . . Bn ), for every t1 , . . . , tn T and B1 , . . . , Bn B(R).
However, with our definition of Xt as coordinate mappings, we have
{Xt1 B1 , . . . , Xtn Bn } = {f : Xt1 (f ) B1 , . . . , Xtn (f ) Bn }
= {f : f (t1 ) B1 , . . . , f (tn ) Bn }
= AB 1 ,...,Bn
t1 ,...,tn .
5
Hence, we must define the probability of a cylinder set AB 1 ,...,Bn
t1 ,...,tn as follows:
P[AtB11,...,t
,...,Bn
n
] = Pt1 ,...,tn (B1 . . . Bn ).
It can be shown that P can be extended to a well-defined probability measure on (, F).
This part of the proof is non-trivial but similar to the extension of the Lebesgue measure
from the semi-ring of all rectangles to the Borel -algebra. We will omit this argument here.
The properties of permutation invariance and projection invariance are used to show that P
is well-defined.
Example 1.3.1 (Independent random variables). Let T be an index set. For all t T let a
probability measure Pt on R be given. Can we construct a probability space (, F, P) and
a collection of independent random variables {Xt : t T } on this probability space such
that Xt has distribution Pt for all t T ? We will show that the answer is yes. Consider the
family of probability distributions P = {Pt1 ,...,tn : n N, t1 , . . . , tn T } defined by
(1.3.1) Pt1 ,...,tn (B1 . . . Bn ) = Pt1 (B1 ) . . . Ptn (Bn ),
where B1 , . . . , Bn B(R). It is an exercise to check that permutation invariance and projec-
tion invariance hold for this family. By Kolmogorovs theorem, there is a probability space
(, F, P) and a collection of random variables {Xt : t T } on this probability space such
that the distribution of (Xt1 , . . . , Xtn ) is Pt1 ,...,tn . In particular, the one-dimensional distri-
bution of Xt is Pt . Also, it follows from (1.3.1) that the random variables Xt1 , . . . , Xtn are
independent. Hence, the random variables {Xt : t T } are independent.
is given by
1 (AB 1 ,...,Bn 1 1
t1 ,...,tn ) = { : Xt1 () B1 , . . . , Xtn () Bn } = Xt1 (B1 ) . . . Xtn (Bn ).
Definition 1.5.2. Let {Xt : t T } and {Yt : t T } be two stochastic processes defined on
the same probability space (, F, P) and having the same index set T . We say that X is a
modification of Y if
t T : P[Xt = Yt ] = 1.
With other words: For the random events At = { : Xt () = Yt ()} it holds that
t T : P[At ] = 1.
Note that in this definition the random event At may depend on t.
The next definition looks very similar to Definition 1.5.2. First we formulate a preliminary
version of the definition and will argue later why this preliminary version has to be modified.
Definition 1.5.3. Let {Xt : t T } and {Yt : t T } be two stochastic processes defined
on the same probability space (, F, P) and having the same index set T . We say that the
processes X and Y are indistinguishable if
P[t T : Xt = Yt ] = 1.
(b) X and Y are not indistinguishable because for every the sample paths t 7 Xt ()
and t 7 Yt () are not equal as functions on T . Namely, YU () () = 1 while XU () () = 0.
Proposition 1.5.6. Let {Xt : t T } and {Yt : t T } be two stochastic processes defined on
the same probability space (, F, P) and having the same index set T . Consider the following
statements:
1. X and Y are indistinguishable.
2. X and Y are modifications of each other.
3. X and Y have the same finite-dimensional distributions.
Then, 1 2 3 and none of the implications can be inverted, in general.
Proof. Exercise.
Exercise 1.5.7. Let {Xt : t T } and {Yt : t T } be two stochastic processes defined on
the same probability space (, F, P) and having the same countable index set T . Show that
X and Y are indistinguishable if and only if they are modifications of each other.
8
1.6. Measurability of subsets of RT
Let {Xt : t T } be a stochastic process defined on a probability space (, F, P). To every
outcome we can associate a trajectory of the process which is the function t 7 Xt ().
Suppose we would like to compute the probability that the trajectory is everywhere equal
to zero. That is, we would like to determine the probability of the set
Z := { : Xt () = 0 for all t T } = tT { : Xt () = 0} = tT Xt1 (0).
But first we need to figure out whether Z is a measurable set, that is whether Z F. If
T is countable, then Z is measurable since any of the sets Xt1 (0) is measurable (because
Xt is a measurable function) and a countable intersection of measurable sets is measurable.
However, if the index set T is not countable (for example T = R), then the set Z may be
non-measurable, as the next example shows.
Example 1.6.1. We will construct a stochastic process {Xt : t R} for which the set Z
is not measurable. As in the proof of Kolmogorovs theorem, our stochastic process will be
defined on the canonical probability space = RR = {f : R R}, with F = cyl being
the cylinder -algebra. Let Xt : RR R be defined as the canonical coordinate mappings:
Xt (f ) = f (t), f RR . Then, the set Z consists of just one element, the function which is
identically 0.
We show that Z does not belong to the cylinder -algebra. Let us call a set A RR
countably generated if one can find t1 , t2 , . . . R and a set B RN such that
(1.6.1) f A {i 7 f (ti )} RN .
With other words, a set A is countably generated if we can determine whether a given
function f : R R belongs to this set just by looking at the values of f at a countable
number of points t1 , t2 , . . . and checking whether these values have some property represented
by the set B.
One can easily check that the countably generated sets form a -algebra (called cg ) and
that the cylinder sets belong to this -algebra. Since the cylinder -algebra is the minimal
-algebra containing all cylinder sets, we have cyl cg .
Let us now take some (nonempty) set A cyl . Then, A cg . Let us show that A is
infinite. Indeed, since A is non-empty, it contains at least one element f A. We will show
that it is possible to construct infinitely many modifications of f (called fa , a R) which
are still contained in A. Since A is countably generated we can find t1 , t2 , . . . R and a set
B RN such that (1.6.1) holds. Since the sequence t1 , t2 , . . . is countable while R is not, we
can find t0 R such that t0 is not a member of the sequence t1 , t2 , . . .. For every a R let
fa : R R be the function given by
(
a, if t = t0 ,
fa (t) =
f (t), if t 6= t0 .
The function fa belongs to A because f belongs to A and the functions i 7 f (ti ), i N,
and i 7 fa (ti ), i N, coincide; see (1.6.1). Hence, the set A contains infinitely many
elements, namely fa , a R. In particular, the set A cannot contain exactly one element. It
follows that the set Z (which contains exactly one element) does not belong to the cylinder
-algebra.
9
Exercise 1.6.2. Show that the following subsets of RR do not belong to the cylinder -
algebra:
(1) C = {f : R R : f is continuous}.
(2) B = {f : R R : f is bounded}.
(3) M = {f : R R : f is monotone increasing}.
Definition 1.7.4. We say that the process X is continuous in Lp , where p 1, if for all
t R it holds that
Lp
Xs Xt as s t.
That is,
t R : lim E|Xt Xs |p = 0.
st
Example 1.7.5. Let U be a random variable which has continuous distribution function
F . For concreteness, one can take the uniform distribution on [0, 1]. Let (, F, P) be the
probability space on which U is defined. Consider a stochastic process {Xt : t R} defined
as follows: For all t R and let
(
1, if t > U (),
Xt () =
0, if t U ().
10
1. For every outcome the trajectory t 7 Xt () is discontinuous because it has a
jump at t = U (). Thus, the process X does not have continuous sample paths.
2. However, we will show that the process X is continuous in probability. Take some
(0, 1). Then, for any t, s [0, 1],
P[|Xt Xs | > ] = P[|Xt Xs | = 1] = P[U is between t and s] = |F (t) F (s)|,
which converges to 0 as s t because the distribution function F was supposed to be
continuous. Hence, the process X is continuous in probability.
3. We show that X is continuous in Lp , for every p 1. Since the random variable |Xt Xs |
takes only values 0 and 1 and since the probability of the value 1 is |F (t) F (s)|, we have
E|Xt Xs |p = |F (t) F (s)|,
which goes to 0 as s t.
Exercise 1.7.6. Show that if a process {X(t) : t R} has continuous sample paths, the it
is stochastically continuous. (The converse is not true by Example 1.7.5).
We have seen in Section 1.6 that for general stochastic processes some very natural events
(for example, the event that the trajectory is everywhere equal to 0) may be non-measurable.
This nasty problem disappears if we are dealing with processes having continuous sample
paths.
Example 1.7.7. Let {Xt , t R} be a process with continuous sample paths. We show that
the set
A := { : Xt () = 0 for all t R}
is measurable. A continuous function is equal to 0 for all t R if and only if it is equal to 0
for all t Q. Hence, we can write
A = { : Xt () = 0 for all t Q} = tQ { : Xt () = 0} = tQ Xt1 (0)
which is a measurable set because Xt1 (0) F for every t (since Xt : R is a measurable
function) and because the intersection over t Q is countable.
Exercise 1.7.8. Let {X : t R} be a stochastic process with continuous sample paths. The
probability space on which X is defined is denoted by (, F, P). Show that the following
subsets of belong to the -algebra F:
(1) B = { : the function t 7 Xt () is bounded}.
(2) M = { : the function t 7 Xt () is monotone increasing}
(3) I = { : limt+ Xt () = +}.
11
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: