Exercise 1.3.4 in Nualart's book

62 Views Asked by At

Let $W$ be an isonormal process over a real separable Hilbert space. Let $D, \delta$ be the associated Malliavin derivative, divergence operators.

Exercise 1.3.4.

Let $F$ be in $\mathbb{D}^{1,2}$ (meaning $F\in L^2(\Omega)$ and $\mathbb{E}\big[ || DF||_H \big] < \infty$)

suppose $\mathbb{E}[ |F|^{-2} ]$ is finite.

Show $\mathbb{P}(F>0)\in\{0,1\}$.

It is clear that $\mathbb{P}(F =0 ) = 0$. And the hint suggests

(i) we take $\psi_\epsilon$ an approximation of the sign function $$ {\rm sign}(x) = 1_{\{x>0 \}} - 1_{\{x < 0 \}}. $$ and

(ii) take $u$ (arbitrary) in $\text{domain}(\delta)$ with $\| u \|_H$ uniformly bounded

and

(iii) use duality relation to compute

$$ \mathbb{E}\big[ \psi_\epsilon(F) \delta(u) \big] = \mathbb{E}\big[ \langle u, DF \rangle_H \psi_\epsilon'(F) \big.] $$ Left side converges to $\mathbb{E}[ {\rm sign}(F) \delta(u) ]$.

(A) Right side seems to converge to zero

so that we would have $$ \mathbb{E}[ {\rm sign}(F) \delta(u) ] =0. \quad(\#) $$ The above equality holds for arbitrary $u$ with a.s. bounded $\| u\|_H$.

(B) This (#) seems to imply ${\rm sign}(F)$ is constant, which implies $\mathbb{P}(F>0)\in\{0,1\}$

For (A): $\psi_\epsilon'(F)\to 0$ almost surely since $|F|>0$ a.s. But I do not know how to justify the convergence of the expectation, ($\psi_\epsilon'(F)$ is not uniformly bounded as $\epsilon\downarrow 0$.) I do not know how to use the assumption $\mathbb{E} |F|^{-2}<\infty$.

For (B): So the set of random variables of the form $\delta(u)$, with $\| u\|_H$ a.s. bounded, is $L^2$-orthogonal complement of the space of constant random variables?

Any comment or clarification is appreciated. Thanks.