4
$\begingroup$

From Gawarecki and Mandrekar, Stochastic Differential Equations in Infinite Dimensions:

We call a family $\{ W_t \}_{t\geq 0}$ defined on a filtered probability space $(\Omega, \mathcal{F}, \{\mathcal{F}_t\}_{t \geq 0}, P)$ a cylindrical Wiener process in a Hilbert space $K$ if:

  1. For an arbitrary $t\geq0$, the mapping $W_t:K \rightarrow L^2(\Omega, \mathcal{F}, P)$ is linear.
  2. For an arbitrary $k \in K$, $W_t(k)$ is an $\mathcal{F}_t$-Brownian motion.
  3. For arbitrary $k, k' \in K$ and $t \geq 0$, $E(W_t(k)W_t(k')) = t \langle k, k'\rangle_K$.

Exercise 2.2 Show that $E(W_t(k)W_s(k')) = (t \wedge s) \langle k, k' \rangle_K$ and conclude that $W_t(f_j), j=1, 2, ...$, where $\{f_j\}_{j=1}^\infty$ is an orthonormal basis in $K$, are independent Brownian motions.


I am new to functional analysis and don't see how to start this exercise. I can see that the third point is useful, but how does one begin? My thought was the following (assuming $t \leq s$:

$$ \begin{align*} E(W_t(k)W_s(k')) &= E(E(W_t(k)W_s(k')|\mathcal{F}_t))\\ &= E(W_t(k)E(W_s(k')|\mathcal{F}))\\ &= E(W_t(k)W_t(k')\\ &= t\langle k, k' \rangle_K \end{align*} $$ by LIE, point two, and point three, respectively. The opposite holds for $s \leq t$. Is this the right approach? Then the independence follows from orthogonality.

$\endgroup$
  • $\begingroup$ Everything seems correct. Naturally, to deduce the independence from orthogonality, one has to mention that the joint distribution of $W(k)$ and $W(k')$ is Gaussian. This is easy to see from 1 and 2. $\endgroup$ – zhoraster May 10 '16 at 10:45
1
$\begingroup$

For posterity's sake, here's my answer (mainly copied from above and using the comment from @zhoraster):

When $t\leq s$ $$ \begin{align*} E(W_t(k)W_s(k')) &= E(E(W_t(k)W_s(k')|\mathcal{F}_t))\\ &= E(W_t(k)E(W_s(k')|\mathcal{F}))\\ &= E(W_t(k)W_t(k')\\ &= t\langle k, k' \rangle_K \end{align*} $$ by LIE, point two, and point three, respectively. The opposite holds for $s \leq t$. Thus $$ \begin{align*} E(W_t(k)W_s(k')) &= (t\wedge s)\langle k, k' \rangle_K \end{align*} $$ where $\wedge$ is the $\min$ operator. Since $\{f_j\}_{j=1}^\infty$ is an orthonormal basis, $W(f_j)$ is an $\mathcal{F}_t$-Brownian motion for all $j$ by 2. Then, using the above result, $E(W_t(f_j)W_s(f_i)) = (t\wedge s)\langle f_j, f_i \rangle_K = 0$ for all $i \neq j$ by orthoganality. Thus, the Brownian motions are independent.

$\endgroup$
  • $\begingroup$ Should the first set of equations read \begin{align*} E(W_t(k)W_s(k')) &= E\left(E(W_t(k)W_s(k')|\mathcal{F}_t)\right)\\ &= E\left(W_t(k)E(W_s(k')|\mathcal{F}_t)\right)\\ &= E\left(W_t(k)W_t(k')\right)\\ &= t\langle k, k' \rangle_K \end{align*} ? $\endgroup$ – man_in_green_shirt Oct 6 '17 at 10:06

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.