Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I'm solving exercise 2.28 in Revuz/Yor. I was able to prove 1). Unfortunately at 2) I got stuck. I have to show:

Let $B$ be a d-dimensional Brownian motion and $A\in \mathcal{A}:=\cap_t \mathcal{A}_t$, where $\mathcal{A}_t:=\sigma(B_s;s\ge t)$. For any fixed $t$, there is an event $B\in \mathcal{A}$ such that $\mathbf1_{A}=\mathbf1_B\circ \theta_t$, ($\theta$ is the shift operator), then $$P_x[A]=\int P_y(B)P_t(x,dy)$$ and conclude that either $P_\cdot[A]=0$ or $P_\cdot[A]=1$.

I did the following:

By Proposition 1.7 (Markov Property) I get:

$$E_x[\mathbf1_B\circ \theta_t|\mathcal{F}_t^0] =E_{B_t}[\mathbf1_B]$$

Well integrating the LHS, we get:

$$E_x[E_x[\mathbf1_B\circ \theta_t|\mathcal{F}_t^0]]=E_x[\mathbf1_A]=P_x[A]$$

Therefore I'm done with the first part, if

$$E_x[E_{B_t}[\mathbf1_B]]=\int P_y(B)P_t(x,dy)$$

My problem is, I do not see why this is true. It would be very nice if someone could tell me, why this equation is true. I run always in trouble using regular conditional probability. Also some hints for the second statement would be appreciated. I'm very thankful for your help.



share|cite|improve this question
@ math : isn't this Chapman-Kolmogorov property of markovian kernel ? – TheBridge Mar 28 '12 at 18:56
@ TheBridge: You mean the equation $ E_x[E_{B_t}[\mathbf1_B]]=\int P_y(B)P_t(x,dy) $ ? Could you make this a little bit more precise in an answer, then I can accept it. If I read the article on Wikipedia to Chapman-Kolmogorov, I do not see why this should follow. – math Mar 28 '12 at 19:01

Define the function $\phi(y)=E_y[\mathbf 1_B]$, that is, $\phi(y)=P_y(B)$. Then $E_{B_t}[\mathbf 1_B]$ is the random variable $\phi(B_t)$. The expected value of this random variable can be expressed in terms of the distribution $\nu$ of $B_t$ via $$E(\phi(B_t))=\int \phi(y)\, \nu(dy).\tag1$$ Under $E_x$, the distribution of $B_t$ is $\nu(dy)=P_t(x,dy)$, so substituting back into (1) gives $$ E_x(E_{B_t}[\mathbf 1_B])=E(\phi(B_t))=\int P_y(B)\, P_t(x,dy). $$

share|cite|improve this answer

You are starting from $x$, not from $0$, so what you wrote about $E_x(\mathbf1(B)\circ \theta_t|\mathcal F^0_t)$ is a bit confusing.To clarify the expression, you'd better write it as $E(\mathbf1(B)\circ \theta_t|\mathcal F^x_t)$. And the equation can be established more clearly:Firstly,thanks to strong markov property, $$E(\mathbf1(B)\circ \theta_t|\mathcal F^x_t)=E(\mathbf1(B)\circ \theta_t| B^x_t) $$ which is the conditional probability of $B$ under the law of Brownian motion evaluated at time $t$ along with the starting point $x$.

And then you calculate the expectation of the above to achieve the probability, which is equal to

$$E(E(\mathbf1(B)\circ \theta_t| B^x_t))=\int P_y(B) P_t(x,dy)$$ The statement then holds. This question want you to confirm zero-one-law under any starting point of Brownian motion.

share|cite|improve this answer
The superscript zero in ${\cal F}_t^0$ has nothing to do with the starting point of the process. It simply denotes the "raw" $\sigma$-field ${\cal F}_t^0=\sigma(X_u,u\leq t)$. – Byron Schmuland Jul 25 '14 at 16:58
Yeah but I guess the superscript stands for the starting point if it exists, usually people don't put a superscript there. If the superscript doesn't count, then we can wrote the equation as $E_x(\mathbf 1(B)\circ \theta_t| \mathcal F_t)=E_x(\mathbf 1(B)\circ \theta_t| B_t) $ to express strong markov property in this case. – Cjay Jul 26 '14 at 2:24

Your Answer


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.