DISCLAIMER

This note is based on Malliavin Calculus and Related Topics by David Nualart and is intended solely for personal academic study and should not be regarded as a substitute for the original text.

1 Wiener Chaos Decomposition

We seek a generalized form of Gaussian process, say changing the set of time indices T to a "more abstract" space H . Suppose H is a real separable Hilbert space with innrt product ⟨⋅,⋅⟩H and norm ‖h‖H for h∈H .

Wiener chaos expansion

Theorem 1.1.6. (Wiener chaos expansion) The space L2(Ω,G,P) can be decomposed to infinite orthogonal sums of Hn :

L2(Ω,G,P)=⨁n=0∞Hn

Proof. Consider a similar approach in lemma 1.1.5. Pick X∈L2(Ω,F,P) such that X is orthogonal to all Hn , i.e. E[XHn(W(h))]=0 for all n≥0 and h∈H,‖h‖H=1 . As xn can be written by a linear combination of Hm in which m≤n , thus we have E[X(W(h))n]=0 for all n and h , and therefore E[Xexp⁡[tW(h)]]=0 for all t∈R . Acording to lemma 1.1.5, we deduce X=0 . ◻

Next we consider constructing a orthogonal system of spaces Hn and admit the case when H is infinite dimensional. We want to utilize the orthogonal properties of Hermite polynomials, such that every F∈Hn could be written in a linear combination (or series) of something related to Hermite polynomial.

We start with polynomials of degrees up to n . Let Pn0 be the space spanned by random variables p(W(h1),…,W(hk)) , where k⩾1 , h1,…,hk∈H and p is a real polynomials in k variables such that deg(p)⩽n . Note that Pn0 is not a closed set in L2(Ω,F,P) , take Sn=∑i=1n2−iW2(ei) , in which ei are orthogonal base vectors in H . This means W(ei) s are independent standard Gaussian random variables, as the definition of isonormal Gaussian process puts: E[W(ei)W(ej)]=⟨ei,ej⟩H=δij . Sn has a limit in L2 , take m<n , we have

E[(Sn−Sm)2]=E[(∑i=mn2−iW2(ei))2]=∑i=mn2−2iE[W4(ei)]+∑j<k21−j−kE[W2(ej)W2(ek)]⩽2−2(m−1)E[W4(ei)]+2−2mE[W2(ej)W2(ek)]

we conclude that Sn is a Cauchy sequence in L2 and therefore has a limit S . We then prove S is not in Pn0 . Assume S=p(W(h1),…,W(hk)) for some k and h1,…,hk∈H , then one could always find a sufficiently large j such that ej⊥{h1,…,hk} , which leads to contradiction.

We use the term "orthogonal system" instead of "basis" sometimes. The latter one says each h∈H can be written by linear combinations of finite base vectors.

In summary, taking a closure of Pn0 is crutial, we denote the closure Pn0― of Pn0 in L2 be Pn , and we claim that it is the orthogonal sum of H0 through Hn .

Lemma 1.1.7. The closure of Pn0 in L2 can be decomposed to orthogonal sum of Wiener chaos subspaces H0 through Hn :

Pn=⨁i=0nHi

Proof. On the one hand, we begin with Hi(W(h)) , which is apprently a polynomial and belongs to Pn0 . Furthermore, we have the finite linear combination of Hi(W(hi)) is still in Pn0 , i.e. Hi0=span{Hi(W(h)):h∈H}⊂Pn0 . When taking the closure of both Hi0 and Pn0 , one can see that all clusterpoints in Hi are also cluster points in Pn , thus Hi⊂Pn . Finally, since i is arbitrary, Hi,i=0,…,n are subsets of Pn , which gives their finite orthogonal sum is also contained by Pn . On the other hand, To show Pn⊂⨁i=0nHi , it is enough to show that Pn is orthogonal to all Hm such that m>n , in other words, we need to show E[p(W(h1),…,W(hk))Hm(W(h0))]=0 for all h0,…,hk,‖h‖=1 , m>n and polynomial p with degree less than or equals n . By the linearity of W(⋅) we can always rewrite p(W(h1),…,W(hk)) using q(W(e1),…,W(ej),W(h)) , where {e1,…,ej,h} is a orthognormal family for some polynomial q whose degree is also less than or equals n . Expanding Ep(W(e1),…,W(ej),W(h))Hm(W(h)) would yield a linear combinations of mononomials like Wα1(e1)⋯Wαj(ej)Wαh(h)Hm(W(h)) , which could be decomposed since W(h) and other W(ei) s are independent:

E[Wα1(e1)⋯Wαj(ej)Wr(h)Hm(W(h))]=E[∏i=1jWαi(ei)]E[Wr(h)Hm(W(h))],

in which α1+⋯+αj+r⩽n . To show E[p(W(h1),…,W(hk))Hm(W(h0))]=0 , it is enough to show for all r⩽n<m , we have E[Wr(h)Hm(h)]=0 , which is immediate since Wr(h) could be expressed as linear combinations of Hermite polynomials with degrees less than r . ◻

Taking closure of Pn0 and Hi0 s let Pn and Hi contains series made by similar mononomials, however to ensure we find orthonormal system in Hn we seek to link this series of mononomials with Hermite polynomials. With this idea in mind, we need to construct random variables like Ha1(W(e1))Ha2(W(e2))⋯Han(W(en)) to construct aforementioned polynomials (as p(W(h1),…,W(hk)) could always be expressed as p(∑i1=1∞a1iW(ei),…,∑ik=1∞akiW(ei)) in which ∑i1=1∞ani<∞,n=1,…,k ).