Let $f: \mathbb{R}^d \longrightarrow \mathbb{R}^d$ be twice continuously differentiable and $\{B(t): t \ge 0 \}$ be a d-dimensional Brownian motion. Further suppose that, for all $t>0$ and $x \in \mathbb{R}^d$ we have $$\mathbb{E}_x \int_{0}^{t}e^{-\lambda s}|f(B(s))| \, ds< \infty $$ and $$\mathbb{E}_x \int_{0}^{t}e^{-\lambda s}|\Delta f(B(s))| \, ds < \infty.$$ Prove that $X(t)=e^{-\lambda t}f(B(t))- \int_{0}^{t} e^{-\lambda s} (\dfrac{1}{2}\Delta f(B(s))- \lambda f(B(s))) \, ds$ is a martingale.

I solved the easier case $\lambda=0$ in the following way but I got stuck to generalize it: For any $0 \le s \le t$,

$$\mathbb{E}[X(t)|\mathcal{F}(s)]=\mathbb{E}_{B(s)}[f(B(t-s))]-\dfrac{1}{2} \int_{0}^{s} \Delta f(B(u)) du - \int_{0}^{t-s} \mathbb{E}_{B(s)}[\dfrac{1}{2} \Delta f(B(u)) du.$$

Using integration by parts and the fact that the probability density of $B$ satisfies the heat equation , i.e. $\dfrac{1}{2} \Delta \mathfrak{p}(t,x,y)=\dfrac{\partial}{\partial t}p(t,x,y)$, we get

\begin{align*} \mathbb{E}_{B(s)}[\dfrac{1}{2} \Delta f(B(u))] &=\dfrac{1}{2} \int \mathfrak{p}(u, B(s), x) \Delta f(x)\, dx \\ &=\dfrac{1}{2} \int \Delta \mathfrak{p}(u, B(s), x) f(x) \, dx \\ &=\int \dfrac{\partial}{\partial u} \mathfrak{p}(u,B(s),x) f(x) \, dx \end{align*}

and then

\begin{align*} \int_{0}^{t-s} \mathbb{E}_{B(s)}[\dfrac{1}{2} \Delta f(B(u)) \, du &= \lim_{\epsilon \downarrow 0} \int \left[\int_{\epsilon}^{t-s} \dfrac{\partial}{\partial u} \mathfrak{p}(u, B(s), x) \, du \right] f(x) \, dx \\ &= \int \mathfrak{p}(t-s, B(s), x) f(x) dx - \lim_{\epsilon \downarrow 0} \int \mathfrak{p}(\epsilon, B(s), x) f(x) dx \\ &= \mathbb{E}_{B(s)} [f(B(t-s))]-f(B(s)) \end{align*}

and hence we get $$\mathbb{E}[X(t) \mid \mathcal{F}(s)]=f(B(s))-\dfrac{1}{2}\int_{0}^{s} \Delta f(B(u)) \, du$$ which confirms the martingale property.

  • $\begingroup$ Yes, I fixed the typo, sorry about this! $\endgroup$ – Romeo123 Feb 24 at 14:05

You can use the fact that

$$M_t := f(B_t)- \frac{1}{2} \int_0^t \Delta f(B_s) \, ds$$ is a martingale to prove the martingale property for $\lambda>0$. The proof goes as follows:

Lemma 1: For any $\alpha \in C_b^1$ the process $$N_t := M_t \alpha_t - \int_0^t M_r \alpha'(r) \, dr$$ is a martingale.

Proof: Fix $0 \leq s \leq t$. Since $$M_t \alpha_t = (M_t-M_s) \alpha_t + M_s \alpha_t$$ we have $$\mathbb{E}(M_t \alpha_t \mid \mathcal{F}_s) = M_s \alpha_t.\tag{1}$$ On the other hand, it follows from the (conditional) Fubini theorem that \begin{align*} \mathbb{E} \left( \int_0^t M_r \alpha'(r) \, dr \mid \mathcal{F}_s \right) &= \int_0^t M_r \alpha'(r) \, dr + \int_s^t \mathbb{E}(M_r \mid \mathcal{F}_s) \alpha'(r) \, dr \\ &= \int_0^s M_r \alpha'(r) \, dr+ M_s (\alpha(t)-\alpha(s)). \tag{2} \end{align*} Subtracting $(2)$ from $(1)$ proves the assertion.

Lemma 2: Let $\alpha \in C_b^1$. Then $$X_t := \alpha_t f(B_t) - \int_0^t (\alpha'(s) f(B_s) + \alpha(s) \Delta f(B_s)) \, ds$$ is a martingale.

Proof: The integration by parts formula shows that

$$\int_0^t \alpha(s) \Delta f(B_s) \, ds = \alpha(t) \int_0^t \Delta f(B_s) \, ds - \int_0^t \alpha'(s) \left(\int_0^s \Delta f(B_r) \, dr \right) \, ds,$$ and this implies $$X_t = M_t \alpha(t) - \int_0^t \alpha'(s) M_s \, ds.$$ Now the assertion follows from Lemma 1.


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.