I would like to compute $\mathbb P\left(\sup_{t\in [a,b]}B_t>x\right)$ where $(B_t)$ is a Brownian motion and $0<a<b$. What I would say using Markov property is $$\mathbb P\left(\sup_{t\in [a,b]}B_t>x \right)=\mathbb P\left(\sup_{t\in [0,b-a]}B_t>x\mid B_0=B_a\right),$$ but something looks strange since the LHS is a number whereas the RHS is a random variable. Can someone tel me how to manage ?
Asked
Active
Viewed 59 times
0
-
They're both numbers... – Ian May 22 '21 at 15:44
-
@Ian: Are you sure ? For me if $f(u)=\mathbb P(Y\in A\mid X=u)$, then if $Z$ is a random variable, then so is $f(Z)$. So, why do you think that the RHS is not a random variable but really a number ? – Surb May 22 '21 at 16:46
-
@Surb Maybe if $B_0$ is also random then that makes sense, but that would be somewhat unusual for Brownian motion. – Ian May 22 '21 at 16:48
-
@Ian: How do you interpret $\mathbb P\left{\sup_{t\in [0,b-a]}B_t>x\mid B_0=B_a\right}$ ? I do it as : there is $\Omega '$ s.t. $\mathbb P(\Omega ')=1$ and for all $\omega '\in \Omega '$, $$\mathbb P\left{\sup_{t\in [0,b-a]}B_t>x\mid B_0=B_a\right}(\omega ')=\mathbb P\left{\sup_{t\in [0,b-a]}B_t>x\mid B_0=B_a(\omega ')\right}.$$ In other words, it's a "brownian motion" that start at $B_a$ a.s. The OP is right at this point, the RHS of his last equality is a random variable, whereas the LHS is a number... – Surb May 22 '21 at 19:14
-
@joshua : your equality is indeed not correct (for the reason you mentionned). Maybe something as $$\mathbb P\left{\sup_{t\in [a,b]}B_t>x\right}=\int_{\mathbb R}\mathbb P\left{\sup_{t\in [0,b-a]}(y+B_t)>x\mid B_0=y\right}\mathbb P\left{B_a\in ,\mathrm d y\right},$$ should be better... – Surb May 22 '21 at 19:22
-
It's not a random variable, it's just conditioned on returning to where it started at time $a$. This is an event, so it's something you can condition on. This is easier to see when $B_0$ is just a number (which it usually is), in which case this is just the usual Markov process restarting at time $a$. – Ian May 22 '21 at 19:45
-
That said, it doesn't achieve what is desired as you say. What is desired is to split the process between time $[0,a]$ and time $[a,b]$. This requires conditioning on going to a specific value at time $a$ and integrating over the appropriate distribution of those values. – Ian May 22 '21 at 19:52