For a Brownian motion $(B_t)_{t \geq 0}$ consider $$ X_t=\int_0^t\sin\left(\frac{1}{B_s}\right)\;ds+B_t. $$ I want to show that $(X_t)_t$ is a Brownian motion.

Clearly, one can check the definition of BM and invoke the fact that $\int_{t_i}^{t_{i+1}}\sin(B_s^{-1})\;ds$ is measurable wrt. $\sigma(\{B_s,t_i\leq s\leq t_{i+1}\})$ together with the Markov property. However, this might be a bit tedious.

I was wondering whether I can show that $(X_t)_t$ is a martingale and therefore conclude with Levy's characterization.

  • $\begingroup$ You want to show that $(X_t)_t$ is a BM with respect to some probability measure or with respect to the same probability measure as for your original BM? $\endgroup$ – saz Feb 13 '18 at 14:53
  • $\begingroup$ W.r.t. some probability measure would be simply Girsanov's theorem. $\endgroup$ – julian Feb 13 '18 at 15:05

Let $(B_t)_{t \geq 0}$ be a Brownian motion on a probabiliy space $(\Omega,\mathcal{A},\mathbb{P})$. The process

$$X_t := \int_0^t \sin \left( \frac{1}{B_s} \right) \,ds + B_t$$

is a Brownian motion with respect to some probability measure $\mathbb{Q}$ on $(\Omega,\mathcal{A})$ (this is a direct consequence of Girsaonov's theorem); as far as I can see, it is, however, not a Brownian motion on the original probability space.

To show that $(X_t)_{t \geq 0}$ is not a Brownian motion on $(\Omega,\mathcal{A},\mathbb{P})$ it clearly suffices to show that $(X_t)_{t \geq 0}$ is not a $\mathbb{P}$-martingale. Since $(B_t)_{t \geq 0}$ is a martingale, we know that $(X_t)_{t \geq 0}$ is a martingale iff

$$M_t := \int_0^t \sin \left( \frac{1}{B_s} \right) \, ds$$

is a martingale. Denote by $(\mathcal{F}_t)_{t \geq 0}$ the canonical filtration of $(B_t)_{t \geq 0}$. If we fix $s \leq t$, then it follows from the independence and stationarity of the increments of $(B_t)_{t \geq 0}$ that

$$\begin{align*} \mathbb{E}(M_t \mid \mathcal{F}_s) &= \int_0^s \sin(B_r^{-1}) \, dr + \int_s^t \mathbb{E}(\sin(B_r^{-1}) \mid \mathcal{F}_s) \, dr \\ &= M_s+ f(t-s,B_s) \end{align*}$$


$$f(u,x) := \int_0^{u} \mathbb{E} \left( \sin \left[ \frac{1}{x+ B_r} \right] \right) \, dr.$$

Thus, $(M_t)_{t \geq 0}$ is a martingale iff

$$f(t-s,B_s) = 0 \quad \text{$\mathbb{P}$-almost surely for all $s \leq t$}. \tag{1}$$

Clearly, we can choose $\delta_1,\delta_2>0$ such that

$$ \forall \left| x- \frac{\pi}{2} \right| \leq \delta_1, |y| \leq \delta_2: \quad \frac{1}{x+y} \in \left[ \frac{\pi}{4}, \frac{3\pi}{4} \right].$$

Moreover, the continuity of the sample paths of Brownian motion implies that there exists $u>0$ such that

$$\mathbb{P}(|B_r| \leq \delta_2) \geq \frac{3}{4} \quad \text{for all $r \in [0,u]$.}$$

This implies

$$\mathbb{E}\left( 1_{|B_r| \leq \delta_2} \sin \left[ \frac{1}{x+B_r} \right] \right) \, dr \geq \sin \left( \frac{\pi}{4} \right) \frac{3}{4}$$

for all $|x-\pi/2|\leq \delta_1$ and $r \in [0,u]$. On the other hand,

$$\left| \mathbb{E}\left( 1_{|B_r| > \delta_2} \sin \left[ \frac{1}{x+B_r} \right] \right) \right| \leq \mathbb{P}(|B_r|>\delta_2) = 1- \mathbb{P}(|B_r| \leq \delta_2)\leq \frac{1}{4}$$

for all $r \in [0,u]$. Hence,

$$f(u,x) \geq \int_0^u \left[ \sin \left( \frac{\pi}{4} \right) \frac{3}{4} - \frac{1}{4} \right] \, dr > \frac{u}{4} >0$$

for any $|x-\pi/2| \leq \delta_1$. Since $\mathbb{P}(|B_s-\pi/2| \leq \delta_1)>0$ for any $s>0$, this implies that $(1)$ does not hold true for $t :=s+u$.

  • $\begingroup$ Why is $\int_s^t E[\sin(B_r^{-1})|\mathcal{F}_s]\;dr=\int_0^{t-s}E[\sin((B_s+B_r)^{-1})]\;dr$? $\endgroup$ – julian Feb 13 '18 at 16:37
  • 1
    $\begingroup$ @julian I'm not claiming that this holds, please check carefully what I wrote. The idea is to write $$\frac{1}{B_r} = \frac{1}{(B_r-B_s)+B_s}$$ and use that $(B_r-B_s)_{r \geq s}$ is independent from $\mathcal{F}_s$. $\endgroup$ – saz Feb 13 '18 at 16:40
  • $\begingroup$ @saz: I'm not sure I understand why this argument is necessary. Since $|\sin(1/x)| \leq 1$ it is clear that $|M_t-M_s| \leq \int_s^t|\sin(1/B_u)|du \leq |t-s|$ for all $s,t$ (and thus $M$ has Lipchitz paths almost surely). Therefore $M$ is a continuous process whose quadratic variation is zero (because clearly any Lipchitz function has zero quadratic variation). Hence $M$ cannot be a martingale. What is wrong with this argument? $\endgroup$ – Shalop Feb 19 '18 at 6:53
  • $\begingroup$ @Shalop Nothing wrong about it, as far as I can see. $\endgroup$ – saz Feb 19 '18 at 8:13

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.