Conditioning of an event related to Brownian Motion

101 Views Asked by At

I'm currently reading the book Brownian Motion by Mörters and Peres and on p. 39 the following situation comes up:

We have a Brownian Motion $B(t)$ and two intervals $[a_1,b_1]$ and $[a_2,b_2]$ where $b_1<a_2$. We define $m_1 = \max_{a_1\leq t \leq b_1}B(t)$ and $m_2 = \max_{a_2\leq t \leq b_2}B(t)$. It then follows that $B(a_2)-B(b_1)$, $m_1-B(b_1)$ and $m_2-B(a_2)$ are independent.

We wan't to show that $\mathbb{P}(m_1 = m_2) = 0$ and therefore they consider the event $$\{\omega:B(a_2)-B(b_1)= m_1-B(b_1)-(m_2-B(a_2))\}$$

They say that they condition on the values of the random variables $m_1-B(b_1)$ and $m_2-B(a_2)$ and that the output from the LHS is a continuous random variable while the RHS outputs a constant.

I don't understand what they mean by conditioning in this case. What I've seen before is conditional expectation and I believe that $$\mathbb{E}[B(a_2)-B(b_1)\vert m_1-B(b_1)] = \mathbb{E}[B(a_2)-B(b_1)]$$ $$\mathbb{E}[m_1-B(b_1)-(m_2-B(a_2))\vert m_1-B(b_1)] =m_1-B(b_1)$$ holds by the independence however I don't see how this implies the above where they are dealing with a specific event.

1

There are 1 best solutions below

2
On BEST ANSWER

Define $X=B(a_2)-B(b_1)$, $Y=m_1-B(b_1)$, $Z=m_2-B(a_2)$, and $m=m_2-m_1=X-Y+Z$. As noted, $X,Y,Z$ are mutually independent. Let $\mu$ be the distribution of $Y$ and $\nu$ the distribution of $Z$. Then, using the independence, because $m=X-Y+Z$, $$ P[m=0]=\int\int P[X-y+z=0]\mu(dy)\nu(dz).\qquad\qquad (1) $$ But the random variable $X$ has a continuous distribution, so $$ P[X=y-z]=0, $$ for each pair $(y,z)$. It follows that the integral on the right in (1) vanishes, hence so does the left; that is, $P[m=0]=0$.