Existence of a sequence of deterministic measurable kernels (Skorohod Integral and Chaos Expansion)

60 Views Asked by At

I'm working on Di Nunno, Øksendal, and Proske's book on Malliavin Calculus and stuck in the problem below. I wrote a possible solution here and discussed it afterward.


Problem. Let $u(t)$, $0 \le t \le T$, be a measurable stochastic process such that $$ \mathbb{E} \left[ \int_0^T u^2(t) ~\mathrm{d}t \right] < \infty $$

Show that there exists a sequence of deterministic measurable kernels $f_n(t_1, \ldots, t_n, t)$ on $[0,T]^{n+1}$ $(n \ge 0 )$, with $$ \int_{[0,T]^{n+1}} f_n^2 (t_1, \ldots, t_n, t) ~\mathrm{d}t_1 \cdots \mathrm{d}t_n \mathrm{d}t < \infty $$ such that all $f_n$ are symmetric with respect to the variables $t_1, \ldots, t_n$ and such that $$ u(t) = u(\omega, t) = \sum_{n=0}^\infty I_n(f_n(\cdot, t))(\omega), \quad \omega \in \Omega, ~t \in [0,T] $$ with convergence in $L^2(\mathbb{P} \times \lambda)$.

Hint: Consider approximations of $u(t)$, $t \in [0,T]$, in $L^2(\mathbb{P} \times \lambda)$ of the form $\sum_{i=1}^m a_i(\omega) b_i(t)$, $m = 1, 2, \ldots$, where $a_i \in L^2(\mathbb{P})$ and $b_i \in L^2([0,T])$.


Using the hint, consider the approximations of $u(t)$ $$ \varphi_m(t) = \sum_{i=1}^m a_i(\omega) b_i(t) $$ with $a_i \in L^2(\mathbb{P})$ and $b_i \in L^2([0,T])$.

By the Wiener-Itô Chaos Expansion, for each $a_i$ there exists a unique sequence of functions $g_n^{(i)} \in \tilde{L}^2 ([0,T]^n)$ such that $$ a_i = \sum_{n=0}^\infty I_n(g_n^{(i)}) $$

Then, $$ \varphi_m(t) = \sum_{i=1}^m \sum_{n=0}^\infty b_i(t) I_n(g_n^{(i)}) $$

Just opening the iterated integral gives $$ \varphi_m(t) = \sum_{i=1}^m \sum_{n=0}^\infty b_i(t) \int_{[0,T]^n} g_n^{(i)} (t_1, \ldots, t_n) ~\mathrm{d}W(t_1) \cdots \mathrm{d}W(t_n) $$

Interchanging the series and the integral, $$ \varphi_m(t) = \sum_{n=0}^\infty \int_{[0,T]^n} \sum_{i=1}^m b_i(t) g_n^{(i)} (t_1, \ldots, t_n) ~\mathrm{d}W(t_1) \cdots \mathrm{d}W(t_n) $$

Taking $m \to \infty$, $\varphi_m \to u$. We define $$ f_n(t_1, \ldots, t_n, t) = \sum_{n=0}^\infty b_i(t) g_n^{(i)} (t_1, \ldots, t_n) $$ and obtain $$ u(t) = \sum_{n=0}^\infty b_i(t) \int_{[0,T]^n} f_n (t_1, \ldots, t_n, t) ~\mathrm{d}W(t_1) \cdots \mathrm{d}W(t_n) = \sum_{n=0}^\infty I_n(f_n(\cdot, t)) $$


There are two immediate problems here:

  1. Can I exchange the series and the integral?

  2. What guarantees the existence of $f_n$, i.e. the existence of that limit?

Are there any other problems? Is it possible to fix them?

I also tried to emulate the proof for the Wiener-Itô Chaos Expansion. Should I do it on the product $a_i b_i$? Is this on $L^2(\mathbb{P})$ to apply Itô Representation Theorem?