# Stochastic calculus and its Stochastic calculus and its applications Stochastic calculus and its...

date post

12-Jul-2020Category

## Documents

view

7download

1

Embed Size (px)

### Transcript of Stochastic calculus and its Stochastic calculus and its applications Stochastic calculus and its...

Stochastic calculus and its applications

Stochastic calculus and its applications Andrey A. Dorogovtsev Department of Random Processes Institute of Mathematics NAS Ukraine andrey.dorogovtsev@gmail.com Sarajevo, 2019

Lecture 1. Ito-Wiener expansion

De�nition

A random variable ξ is Gaussian or normally distributed with parameters a and σ2if its density has a form

p(u) = 1√ 2πσ

exp−1 2

(u−a)2.

The case a = 1, σ = 1 is called standard.

Fact

Eξ = a, V ξ = σ2. Characteristic function E exp itξ = exp ita− σ2t2 2

.

Theorem

Let ξ and η be independent Gaussian random variabless with parameters aξ , σξ , aη , ση correspondingly. Then for arbitrary α and β random variable αξ + βη is Gaussian with parameters αaξ + βaη , α2σ2ξ + β

2σ2η . Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Corollary

Let ξ1, ...,ξn be independent standard Gaussian variables. Then for arbitrary α1, ...,αn the sum

n

∑ k=1

αkξk

has a Gaussian distribution with the parameters 0 and ∑nk=1α2k .

This gives us possibility to de�ne a linear map from Rn to space of

random variables

Rn 3 α 7→ Uα = n

∑ k=1

αkξk

with the properties

1) for every α ∈ Rn, Uα is a Gaussian random variable, 2) EUα = 0, VUα = ||α||2, where ||α||, is a Euqlidean norm of α

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Fact

U saves the norm.

Corollary

For orthogal α,β ∈ Rn the random variables Uα and Uβ are independent.

Proof.

E exp i(λ1Uα + λ2Uβ ) =

= E exp

( i

n

∑ k=1

(λ1αk + λ2βk)ξk

) =

= exp−1 2

n

∑ k=1

(λ1αk + λ2βk)2 =

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Proof.

exp

{ −λ

2

1

2

n

∑ k=1

α2k − λ 2 2

2

n

∑ k=1

β 2k −λ1λ2 n

∑ k=1

αkβk

} =

= E exp iλ1UαE exp iλ2Uβ

due to the equality

(α,β ) = n

∑ k=1

αkβk = 0.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

De�nition

White noise in the Hilbert space H is a linear map from H to space

of random variables

H 3 α 7→ (α,ξ )

1) for every α ∈ H, (α,ξ ) is a Gaussian random variable, 2) E (α,ξ ) = 0, V (α,ξ ) = ||α||2, where ||α||, is a norm of α .

Example

H = l2, {ξn;n ≥ 1} be the sequence of the independent standard Gaussian random variables. De�ne

(h,ξ ) := ∞

∑ n=1

hnξn

Now ξ is not a random element in H.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

How about H = L2([0; 1])? Let ξ be a white noise in L2([0; 1]). Recall that for f ,g ∈ L2([0; 1]) their product is de�ned as

(f , g) = ∫

1

0

f (t)g(t)dt.

Consequently, one can denote

(f , ξ ) = ∫

1

0

f (t)ξ (t)dt.

But since dimL2([0; 1]) = ∞ , then ξ is not a random function (some people say that ξ is a generalized function)!

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

The integral from ξ has a right meaning.

De�nition

w(t) = ∫ t 0

ξ (s)ds = (1[0; t], ξ ), t ∈ [0; 1]. w is called by the Wiener process or the process of Brownian

motion.

Fact

Properties of the Wiener process

1. w is a Gaussian process, which means that for arbitrary

t1, ..., tn ∈ [0; 1], α1, ...,αn ∈ R the sum

n

∑ k=1

αkw(tk)

is a Gaussian random variable.

2. w has independent and stationary increments.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Note that one can write

(f , ξ ) = ∫

1

0

f (t)ξ (t)dt = ∫

1

0

f (t)dw(t).

It follows from the de�nition of white noise, that

E

∫ 1

0

f (t)dw(t) = 0, V ∫

1

0

f (t)dw(t) = ∫

1

0

f (t)2dt.

In particular,

Ew(t) = 0, Vw(t) = t.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

De�nition

n-th degree Hermite polynomial is

Hn(x) = (−1)ne x2

2

( d

dx

)n e−

x2

2 .

The �rst Hermite polynomials are

H0(x) = 1, H1(x) = x , H2(x) = x 2−1.

Fact

For arbitrary n ≥ 0 the following statement holds. 1) Hn is odd if n is odd and even if n is even,

2) Hn+2(x) = xHn+1(x)− (n+1)Hn(x), 3) H ′n+1(x) = (n+1)Hn(x).

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Fact

For every n,m ≥ 0

1√ 2π

∫ R Hn(x)Hm(x)e

− x22 dx = δmnn!

The sequence {

1√ n! Hn;n ≥ 1

} is an orthonormal basis in

L2

( R, 1√

2π e − x22 dx

) .

Translation of this fact on the �probability language� will be useful.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Theorem

Every square-integrable random variable α which is measurable with respect to the standard Gaussian variable ξ can be uniquely represented by the series

α = ∞

∑ n=0

anHn(ξ ).

The series converges in the square mean and

Eα = a0, Eα2 = ∞

∑ n=0

n!a2n.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

The properties of Hermite polynomials can be derived from their

generating function. Let us note that for arbitrary x ,y ∈ R

exy− y2

2 = ∞

∑ n=0

Hn(x)y n

n! .

From this formula one can get

Hn(ξ ) = ∂ n

∂yn eξy−

y2

2 |y=0.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Let ξ be white noise in H.

De�nition

Stochastic exponent for h ∈ H is de�ned as follows

E (h) = exp((h,ξ )− 1 2 ||h||2)

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Multi-dimensional Hermite polinomials from white noise ξ . Let An be a symmetric n-multiple linear form on H.

De�nition

An is a Hilbert-Shmidt form if for arbitrary orthonormal basis

{en; n ≥ 1} the following sum

||An||2n = ∞

∑ k1,...,kn=1

An(ek1 , ...,ekn) 2

is �nite. The value of the sum does not depend on choice of basis.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

De�nition

The value of the form An on the white noise ξ (the in�nite-dimensional Hermite polinomial corresponding to An) is the

following product

An(ξ , ...,ξ ) = (An, ∇nhE )|h=0.

Fact

Properties of An(ξ , ...,ξ ) 1.

EAn(ξ , . . . ,ξ ) = 0, EAn(ξ , . . . ,ξ )Bn(ξ , . . . ,ξ ) = n!(An,Bn)n.

2. Let m 6= n. Then

EAn(ξ , . . . ,ξ )Bm(ξ , . . . ,ξ ) = 0.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Fact

3. Ito-Wiener expansion. Every square integrable random variable

α which is measurable with respect to ξ has a unique orthogonal expansion

α = ∞

∑ n=0

An(ξ , . . . ,ξ ),

Eα = A0, Eα2 = ∞

∑ n=0

n!‖An‖2n.

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Examples

1. Let A = e⊗n, where e ∈ H and ‖e‖= 1. Then

e⊗n(ξ , . . . ,ξ ) = Hn((e,ξ )).

If ‖e‖ 6= 1 then

e⊗n(ξ , . . . ,ξ ) = ‖e‖nHn ((

e

‖e‖ ,ξ ))

.

2. Let {e1, . . . ,en} be an orthonormal system. Then

e1⊗ . . .⊗ en(ξ , . . . ,ξ ) = n

∏ k=1

(ek ,ξ ).

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

Examples

3.

e⊗r1 1 ⊗ . . .e⊗rnn (ξ , . . . ,ξ ) =

n

∏ k=1

Hrk ((ek ,ξ )).

4. For arbitrary f1, f2 ∈ H

f1⊗ f2(ξ ,ξ ) = (f1,ξ )(f2,ξ )− (f1, f2).

Andrey Dorogovtsev White noise analysis and coalescing stochastic �ows

Lecture 1. Ito-Wiener expansion

The c

Recommended

*View more*