Consider $B=(B_t)_{t\geq 0}$ $\mathcal F_t$ - brownian motion in $\mathbb R ^n, \ (n\geq 2)$ starting at zero, in a probability space $(\Omega, \mathcal F, (\mathcal F_t)_{t\geq 0}, \mathbb P)$. Then, consider $X_t ^x = x + B_t $ and $S_t = \left|X^x_t\right|$. I was trying to show that $(T_t)_{t\geq 0}$ defined by $$T_t := S_t -\left| x \right| -\frac {n-1}{2} \int _0 ^t \frac {1}{S_u}~du$$

is a real $\mathcal F_t$ - brownian motion.

It's not difficult to see that $W_t = S_t -\left| x \right|$ is a real $\mathcal F_t$ - brownian motion.

Inded, Ito's lemma implies that

$$ W_t = \int _0 ^t \frac{X^x_t}{\left| X^x_t \right| }~ dB_s $$ wich is well defined since $\mathbb P (\exists t>0 : S_t =\left| X^x_t \right| =0) =0$. Then by Pauls-Lévy Theorem we have that $(W_t)$ is a real $\mathcal F_t$ - brownian motion.

So, $$T_t = W_t -\frac {n-1}{2} \int _0 ^t \frac {1}{S_u}~du.$$

Have anybody a smarter idea that compute its mean and covariance in order to apply again Paul-Lévy theorem ? If not, any smart advices to simplify the calculation?

Thank's in advance.

Edit: Maybe this other result could helpful to crack this proof.

I could show that $\mathbb E \left\{ \int _0 ^t \frac {1}{S_u}~du\right\} =0$ since $M_t := \int _0 ^t \frac {1}{S_u}~du$ is a bounded local martingale. Then for the covariation we have $$ \mathbb E \left \{T_s T_t \right\}=t \wedge s -c (\mathbb E \left \{W_s M_t \right\} +\mathbb E \left \{M_s W_t \right\}) + \mathbb E \left \{M_s M_t \right\}$$

New Edit: Could someone please check the question at the end of my solution try at the answer ?


I'd like to show that $M^\lambda = (M_t^\lambda)_{t\geq0}$ $$ M_t^\lambda := \exp\left(i\lambda T_t-\frac{\lambda^2}{2}t \right) $$ is a complex martingale so $T$ is a brownian motion.

Indeed, by Ito's lemma we have that

$$dM_t ^\lambda = \frac{\lambda^2}{2}M_t ^\lambda dt + i\lambda M_t ^\lambda dT_t - \frac{\lambda^2}{2} d\langle T \rangle_t$$

but also we have that

$$ \langle T \rangle_t = \langle W \rangle_t + \langle \int _0 ^ . \frac{1}{S_u}~du \rangle_t = t$$

so $$dM_t ^\lambda = i\lambda M_t ^\lambda dT_t$$ and $$\mathbb E \left\{ \int _0 ^t \lambda^2 (M_t^ \lambda)^2 d\langle T \rangle_t\right\}\leq \exp(\lambda^2 t)< +\infty$$ However, we must remember $T$ is a local martingale. Then even if $\phi_t := i\lambda M_t ^\lambda \in \mathbb H ^2(T)$ can we conclude that $M^\lambda$ is a martingale ?


First, your sentence "$W_t=S_t-x$ is a Brownian motion" is wrong, or Brownian motion is bounded below by |x| almost surely. The aim of the integral is precisely to compensate for the absolute value making this not to be a Brownian motion.

Have you checked out Revuz-Yor or Chaumont-Yor on this ? (I think your exercise in the latter book)


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.