3
$\begingroup$

The Brownian motion has the following (Levy-Ciesielski?) construction via Hilbert space isomorphisms:

  1. Let $\{ Z_i \}_{i \in \mathbb{Z}}$ be i.i.d. $N(0,1)$ random variables defined on $(\Omega, \mathcal{F}, P)$.

  2. Let $\{ \phi_i \}_{i \in \mathbb{Z}}$ be an orthonormal basis for $L^2[0, 1]$.

  3. The map $\phi_i \mapsto Z_i $ extends to an isometry $B$ from $L^2[0, 1]$ to the subspace of $L^2(\Omega)$ generated by $\{ Z_i \}_{i \in \mathbb{Z}}$.

  4. The process defined by $B_t(\omega) = B(1_{[0,t]})(\omega)$ is a version of Brownian motion, and $\int_0 ^1 \phi_i dB_t = Z_i$ in the Ito sense.

Does this extend in some way to the fractional Brownian motion?

$\endgroup$
1
  • 1
    $\begingroup$ Initially asked on SE. $\endgroup$
    – Michael
    Aug 18, 2014 at 18:58

1 Answer 1

3
$\begingroup$

Yes, similar constructions essentially work for the fractional Brownian motion. The space $L^2[0,1]$ needs to be replaced by the space $\mathcal{H}$ which is the closure of the linear span of space of indicators $1_{[s,t]}$ with respect to the inner product

$ \langle 1_{[0,s]}, 1_{[0,t]} \rangle_\mathcal{H}=R(s,t)$

where $R(s,t)$ is the covariance function of the fractional Brownian motion. Details may be found in

Stochastic calculus for the fractional Brownian motion

which also contains relevant references.

$\endgroup$
3
  • $\begingroup$ Thank you. What about the choice of basis? Can they be chosen in such a way that their integrals against fBm are, say, approximately uncorrelated? $\endgroup$
    – Michael
    Aug 19, 2014 at 3:47
  • $\begingroup$ Yes, if $\phi_i$ is an orthonormal basis of $\mathcal{H}$, then $\mathbb{E}(B(\phi_i)B(\phi_j))=\delta_{ij}$. So the $B(\phi_i)$'s are uncorrelated. $\endgroup$ Aug 19, 2014 at 14:30
  • $\begingroup$ I seem to be missing something. So the construction maps $1_{[s,t]}$ to $B^H_t - B^H_s$. Extending to the closure, for $f \in \mathcal{H}$, $\int f dB^H_t$ is defined to be the image of $f$ under this map. If we have an ONB $\{ \phi_i\}$ of $L^2[0,1]$ that happen to also be in $\mathcal{H}$, how do we know $\{ \int \phi_i dB^H_t \}$ are uncorrelated? Thanks for your patience. $\endgroup$
    – Michael
    Aug 19, 2014 at 20:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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