I have a quesition about an isonormal Gaussian process and measurability.

Let $\mathcal{H}$ be a real separable Hilbert space with inner product $\langle \cdot,\cdot \rangle$ and norm $\|\cdot\|=\langle\cdot,\cdot\rangle^{1/2}$. Let $X=\{X_{h}:h \in \mathcal{H}\}$ be an isonormal Gaussian process over $\mathcal{H}$ i.e. $X$ is a collection of jointly Gaussian random variables defined on a probability space $(\Omega,\mathcal{F},P)$ and such that $E \left[X_{h}X_{g} \right]=\langle h,g\rangle$ for all $h,g \in \mathcal{H}$, $E\left[X_{h} \right]=0$ for all $h \in \mathcal{H}$.

We shall assume $\mathcal{F}=\sigma[X]$.

My question

Let $(h_{j})_{j =1}^{\infty}$ be an orthonormal basis of $\mathcal{H}$ and $\mathcal{F}_{m}\,(m\in\mathbb{N})$ be the $\sigma$-algebra generated by $X_{h_{1}},\ldots,X_{h_{m}}$. In this case can we deduce $\mathcal{F}=\sigma \left[ \bigcup_{m=1}^{\infty} \mathcal{F}_{m} \right]$?

My attempt:

I think it is enough to show that $\forall h \in \mathcal{H}$, $X_{h}$ is $\sigma \left[ \bigcup_{m=1}^{\infty} \mathcal{F}_{m} \right]$-measurable.

Since $h=\lim_{n \to \infty} \hat{h}_{n}$ in $\mathcal{H}\quad(\hat{h}_{n}=\sum_{j=1}^{n}\langle h,h_{j} \rangle h_{j})$,

\begin{align*} E \left[\left|X_{h}-X_{\hat{h}_{n}} \right|^{2} \right]=\|h-\hat{h}_{n}\|^{2}\to0\quad(n \to \infty) \end{align*}

Therefore $X_{h}=\lim_{k \to \infty}X_{\hat{h}_{n_{k}}}$ $P$-a.s. but not in pointwise. If we could prove $X_{h}(\omega)=\lim_{k \to \infty}X_{\hat{h}_{n_{k}}}(\omega)$ for all $\omega \in \Omega$, $X_{h}$ is $\sigma \left[ \bigcup_{m=1}^{\infty} \mathcal{F}_{m} \right]$-measurable.

(since $X_{\hat{h}_{n}}$ is $\sigma \left[ \bigcup_{m=1}^{\infty} \mathcal{F}_{m} \right]$-measurable.)

Thank you in advance.



Take a non-zero $g\in\mathcal{H}$ such that $g$ is not an element of the orthonormal basis $(h_j)_{j=1}^\infty$. Let $B$ be a Bernoulli random variable with parameter $1/2$, independent of $X$. Define, for $h\in\mathcal{H}$, $$ \tilde{X}_h = \begin{cases}X_g\mathbf{1}(B = 0)\qquad &\text{if $g=h$}\\ X_h \qquad &\text{otherwise} \end{cases} $$

$\tilde{X}_g$ is not measurable with respect to $\sigma(\tilde{X}_{h_1},\tilde{X}_{h_2},\cdots)=\sigma(X_{h_1},X_{h_2},\cdots)$, since it depends on the outcome of $B$! (Prove this if it's not obvious.)

Now instead choose the parameter of $B$ to zero (then $B=0$, a.s.). This doesn't change the measurability of $\tilde{X}_g$. Now $\tilde{X}$ is an isonormal process, but $\tilde{X}_g$ still isn't measurable with respect to $\sigma(\tilde{X}_{h_1},\tilde{X}_{h_2},\cdots)$.

This is why it is typical to work with a complete $\sigma$-algebra.

  • $\begingroup$ Thanks for your reply. I hardly understand. I try to understand your answer. $\endgroup$ – sharpe Mar 23 '15 at 2:58
  • $\begingroup$ @sharpe You're welcome. If you don't understand anything, let me know and I'll add detail to the answer. $\endgroup$ – Ben Derrett Mar 23 '15 at 9:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.