Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $B_t,t\geq 0$ a brownian motion and $u(t,x)$ a function satisfying the following PDE $$\frac{\partial u}{\partial t}+\frac{1}{2}\frac{\partial^2 u}{\partial x^2}=0.$$ Then we prove that $\frac{\partial }{\partial t}\Bbb{E}(u(t,B_t))=0$, but I want to prove that $u(t,B_t)$ is a martingale (without using Itô formula). I think it suffices to show that $u(t,B_t)$ is a submartingale or supermartingale. Thanks.

share|improve this question
1  
Actually, you would have to use at some point that $u$ solves the PDE. With applying Ito formula, the only way you can use it, as I see, is to write the explicit expression of $u$ via the fundamental solution (=Gaussian density) and deal with integrals w.r.t. conditional distributions of Brownian motion. Perhaps, doable - but Ito's formula seems to be much easier. –  Ilya Jun 5 '13 at 15:22
add comment

1 Answer

You can apply the following theorem to prove $u(t,B_t)$ a martingale. The main idea is -as Ilya wrote- to use the fact that the density $p(t,x)$ of a Gaussian distribution with mean 0, variance $t$ solves $$\frac{\partial}{\partial t} p(t,x) = \frac{1}{2} \frac{\partial^2}{\partial x^2} p(t,x) \qquad (x \in \mathbb{R},t>0) \tag{1}$$.

Theorem Let $(B_t)_{t \geq 0}$ a Brownian motion and $u \in C^{1,2} \cap C([0,\infty) \times \mathbb{R})$ such that $u$ and its partial derivatives are exponentially bounded. Then $$M_t^u := u(t,B_t) - u(0,B_0) - \int_0^t \underbrace{\left( \frac{\partial }{\partial t} + \frac{1}{2} \frac{\partial^2}{\partial x^2} \right) u(r,B_r)}_{=:Lu(r,B_r)} \, dr$$ is a martingale.

Sketch of the proof: By the Markov property of the Brownian motion, one can show that

$$\begin{align*} \mathbb{E}(M_t^u-M_s^u \mid \mathcal{F}_s) &= \mathbb{E} \left( u(t,B_t)-u(s,B_s) - \int_s^t Lu(r,B_r) \, dr \mid \mathcal{F}_s \right) \\ &= \mathbb{E} \left( u(t,B_{t-s}+z)-u(s,B_0+z) - \int_0^{t-s} Lu(r+s,B_r+z) \, dr \right) \bigg|_{z=B_s} \\ &= \mathbb{E} \bigg( \underbrace{\varphi(t-s,B_{t-s}) - \varphi(0,B_s) - \int_0^{t-s} L\varphi(r,B_{r}) \, dr}_{M_{t-s}^{\varphi}-M_0^{\varphi}} \bigg) \bigg|_{z=B_s} \end{align*}$$

where $\varphi(t,x) := u(t+s,x+z)$. To prove that this expression equals $0$, use Fubini's Theorem, partial integration and $(1)$ to show that

$$\mathbb{E}(M_t^{\varphi}-M_{\varepsilon}^{\varphi}) = \int (p(t,x) \cdot \varphi(t,x)-p(\varepsilon,x) \cdot \varphi(\varepsilon,x)) \, dx \\ - \int_{\varepsilon}^t p(r,x) \cdot \left( \frac{\partial}{\partial t} + \frac{1}{2} \frac{\partial^2}{\partial x^2} \right) \varphi(r,x) \, dx \, dr = 0$$

for $t > \varepsilon>u$. Finally, by Doob's maximal inequality and the exponential boundedness, one can apply dominated converence to conclude that $$\mathbb{E}(M_t^{\varphi}-M_0^{\varphi}) = \lim_{\varepsilon \to 0} \mathbb{E}(M_t^{\varphi}-M_{\varepsilon}^{\varphi}) = 0$$

This finishs the proof.

(For more details, see e.g. René L. Schilling/Lothar Partzsch: Brownian Motion - An Introduction to Stochastic Processes, Theorem 5.6.)


Actually, the mentioned theorem is a special case of the following theorem:

Theorem Let $(X_t)_{t \geq 0}$ a Feller process with generator $A$. Then $$f(X_t)-f(X_0) - \int_0^t Af(X_r) \, dr$$ is a martingale for any $f$ in the domain of $A$.

So, in your case the process is given by $X_t := (t,B_t)$. With some knowledge about generators, it's not that difficult to show that the generator $A$ is given by $$A = \frac{\partial}{\partial t} + \frac{1}{2} \frac{\partial^2}{\partial x^2}.$$ (This leads to the topic of transition semigroup and its generator. It's also contained in the book, chapter 7.)

share|improve this answer
    
Thank you for your answer. I have a question maybe stupid why we don't compute directly $\Bbb{E}(u(t,B_t)-u(s,B_s)|\mathcal{F_s})$ and use the same arguments used in your proof. My objectif is to prove that $u(t,B_t)$ is a martingale not $M_t^u$. –  bmo Jun 6 '13 at 8:32
    
@bmo Yes, of course. In your example, we have $Lu = 0$, i.e. $M_t^u = u(t,B_t)-u(0,B_0)$. So you can either apply the theorem or adapt the proof. –  saz Jun 6 '13 at 10:42
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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