I am reading the book Continuous Martingales and Brownian Motion by Revuz and Yor and couldn't understand the "regular conditional distribution" in the following proposition. I will quote relevant definitions first.

Definition: The space $C(\mathbb{R}^+,\mathbb{R}^d)$ is denoted by $\mathbf{W}$, where $d \geq 1$. If $w(s), s \geq 0$ denote the coordinate mappings, we set $\mathscr{B}_t = \sigma(w(s),s \leq t)$.

Definition: Given two predictable functions $f$ and $g$ with values in $d \times r$ matrices and $d$-vectors, a solution of the stochastic differential equation $e(f,g)$ is a pair $(X,B)$ of adapted processes defined on a filtered probability space $(\Omega,\mathscr{F}_t,\mathbb{P})$ such that

  1. $B$ is a standard $\{\mathscr{F}_t\}$ Brownian motion in $\mathbb{R}^r$.
  2. $X_t = X_0 + \int_0^t f(t,X) d B_s + \int_0^t g(t,X) d s$ for any $t$.

Definition: There is uniqueness in law for $e(f,g)$ if whenever $(X,B)$ and $(X',B')$ are two solutions with possibly different Brownian mappings $B$ and $B'$ and $X_0 \stackrel{d}{=} X_0'$, then the laws of $X$ and $X'$ are equal.

Now we come to the proposition and its proof:

Proposition: There is uniqueness in law if, for every $x \in \mathbb{R}^d$, whenever $(X,B)$ and $(X',B')$ are two solutions such that $X_0 = x$ and $X_0' = x$ a.s., then the laws of $X$ and $X'$ are equal.

Proof: Let $P$ be the law of $(X,B)$ on the canonical space $C(\mathbb{R}^+,\mathbb{R}^{d+r})$. Since this is a Polish space, there is a regular conditional distribution $P(\omega,\cdot)$ for $P$ with respect to $\mathscr{B}_0$. For almost every $\omega$ the last $r$ coordinate mappings $\beta^i$ still form a $\text{BM}^r$ under $P(\omega,\cdot)$ and the integral $$ \int_0^t f(s,\xi) d \beta_s + \int_0^t g(s,\xi) d s $$ where $\xi$ stands for the vector of the first $d$ coordinate mappings, makes sense...

I stop quoting the thorough proof as my question has already come:

Question: My understanding is that the regular conditional distribution $P(\cdot, \cdot)$ satisfies that for any measurable $A \subset \mathbf{W}$, $$ P(\omega, A) = \mathbb{E}^P [ \mathbf{1}_A | \mathscr{B}_0] (\omega),\text{ for almost all $\omega$} $$ Then, how can we deduce that "For almost every $\omega$ the last $r$ coordinate mappings $\beta^i$ still form a $\text{BM}^r$ under $P(\omega,\cdot)"$? As far as I know, for any $t_1 < t_2 < \cdots < t_k$ and $\Gamma_j \in \mathscr{B}(\mathbb{R^r}), j = 1,\cdots,k$, $$ P(\omega,\bigcap\beta_{t_j} \in \Gamma_j) = \mathbb{E}^P [ \mathbf{1}_{(\bigcap\beta_{t_j} \in \Gamma_j)} | \mathscr{B}_0] (\omega)= P(\bigcap\beta_{t_j} \in \Gamma_j | \mathscr{B}_0) (\omega) $$ But I can't see if $$ P(\bigcap\beta_{t_j} \in \Gamma_j | \mathscr{B}_0) (\omega) = P(\bigcap\beta_{t_j} \in \Gamma_j) $$ holds so that we can conclude $\beta$ is a $\text{BM}^r$ under $P(\omega,\cdot)$. Maybe my understanding of the regular conditional distribution is completely wrong? Any hint will be greatly appreciated!


First note that $B$ is independent of $\mathcal F_0 = \mathcal B_0$, by the definition of Brownian Motion. This seems to be the main observation you are missing.

We know if $X$ is a random variable which is independent of a sigma algebra $\mathcal G$ then $E[X|\mathcal G] = E[X]$ almost surely.

So, let $\beta: C(\Bbb R_+, \Bbb R^{d+r}) \to C(\Bbb R_+, \Bbb R^{r})$ denote the last $r$ coordinate mappings and let $A$ be a Borel set in $C(\Bbb R_+, \Bbb R^{r})$. Then the preceding observations tell us that $$P(\omega, \beta^{-1}(A)) = E_P[1_{\beta^{-1}(A)} | \mathcal B_0](\omega) = P[B\in A| \mathcal B_0](\omega) = P(B\in A) \;\;\;\text{ a.s.}$$

  • $\begingroup$ Ah..! As the probability space in concern becomes the Wiener space, I was so careless as to overlook it.. Thanks very much! $\endgroup$ – Dormire Jan 31 '17 at 9:53

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.