In the proof of the reflection principle in Durrett's textbook (Probability: Theory and Examples (4e), Theorem 8.4.1, page 317), there's a step which I'm a little shaky on. Basically, this proof invokes the strong Markov property so to set this up in Durrett's notation, let $B_s$ denote Brownian motion and define (I think for fixed $t$ here, but correct me if I'm wrong): $$ Y_s(\omega) = \begin{cases} 1, & \textrm{if } s<t,\, \omega(t-s)>a\\ 0, & \textrm{otherwise} \end{cases} $$ Then defining the stopping time $S = \inf\{s<t: B_s=a\}$ (and $\inf \emptyset = \infty$), we get: $$ Y_S(\theta_S \omega) = \begin{cases} 1, & \textrm{if } S<t,\, B_t>a\\ 0, & \textrm{otherwise} \end{cases} $$ where $\theta_S$ is the usual random shift operator for elements of $\mathcal{C}[0,\infty)$. So far, so good. This took a while to wrap my head around but it makes sense. Applying the strong Markov property we get $$ E_0(Y_S \circ \theta_S | \mathcal{F}_S) = E_{B_S} Y_S \textrm{ on } \{S < \infty\}$$ Now taking expectations, Durrett gets: $$ P_0(T_a < t, B_t \geq a) = E_0(Y_S \circ \theta_S; S < \infty) $$ I'm not seeing the right hand side. I'm trying to see how $$ E_0 (E_{B_S} Y_S; S < \infty) = E_0(Y_S \circ \theta_S; S<\infty) $$ but I'm getting confused. Does this somehow involve translation invarance of $B_s$ or is it straight from the definition of $Y_S$ and I'm just missing something obvious?

EDIT: Just putting @saz's solution here to clear up exactly where the strong Markov property is used. Essentially integrating $Y_S(\theta_S \omega)$ on $\{S < \infty\}$ and writing it as: $$E_0(Y_S \circ \theta_S; S < \infty) = E_0(E_0(Y_S \circ \theta_S | \mathcal{F}_S); S < \infty)$$ It becomes obvious where to apply the Markov property.

  • $\begingroup$ How is $T_a$ defined? $\endgroup$
    – saz
    Jul 30, 2015 at 14:23
  • $\begingroup$ Sorry, $T_a = \inf\{t: B_t = a\}$, so essentially the same as $S$. $\endgroup$
    – gogurt
    Jul 30, 2015 at 16:25
  • 1
    $\begingroup$ $ E_0 (E_{B_S} Y_S; S < \infty) = E_0(Y_S \circ \theta_S; S<\infty) $ follows directly by taking expectations of the equation (two lines above) that defines the strong Markov property. $\endgroup$
    – user940
    Jul 30, 2015 at 17:34
  • $\begingroup$ @ByronSchmuland Ouch. That's obvious. Thanks for straightening that out. $\endgroup$
    – gogurt
    Jul 30, 2015 at 17:51
  • $\begingroup$ @ByronSchmuland ... but that's exactly what I wrote in the first line of my answer; I simply added some more details. $\endgroup$
    – saz
    Jul 30, 2015 at 18:34

1 Answer 1


Actually, Durrett takes the expectation in the previous identity:

$$Y_S(\theta_S \omega) = \begin{cases} 1, & \text{if} \, S(\omega)<t, B_t(\omega)>a \\ 0, & \text{otherwise} \end{cases}. \tag{1}$$

Indeed: By $(1)$, we have $$Y_S(\theta_S \omega)= 1_{\{S<t,B_t>a\}}(\omega). $$

Since $T_a(\omega) = S(\omega)$ for all $\omega \in \{S<t\}$, this shows

$$Y_S(\theta_S \omega) = 1_{\{T_a<t, B_t>a\}}(\omega). \tag{2}$$

As $\{T_a < t\} = \{S<\infty\}$, this implies

$$Y_S(\theta_S \omega) 1_{\{S<\infty\}} \stackrel{(2)}{=} 1_{\{T_a<t, B_t>a\}}(\omega) 1_{\{S<\infty\}}(\omega) = 1_{\{T_a<t, B_t>a\}}(\omega). \tag{3}$$

Integrating both sides yields

$$\begin{align*} \mathbb{P}_0(T_a < t, B_t>a) &\stackrel{(3)}{=} \mathbb{E}_0(Y_S \circ \theta_S 1_{\{S<\infty\}}) \\ &= \mathbb{E}_0(Y_S \circ \theta_S; S< \infty). \end{align*}$$

Finally, note that $\mathbb{P}_0(B_t=a)=0$ and therefore the claim follows.

  • $\begingroup$ ARGH. I see. I just got confused by the way it's worded in the book. Thanks. $\endgroup$
    – gogurt
    Jul 30, 2015 at 17:48

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.