2
$\begingroup$

I am trying to figure out how to solve problem:

Let $N(t)$ be a Poisson process of rate $λ$. Given that $N(t) = 3$, determine the conditional probability density functions of each of the three arrival times $S_1$, $S_2$, $S_3$.

I know that $S_1$ and $S_2$ are uniformly distributed at $[0,1]$ and $S_3$ seems to be independent of them, so I can find its pdf from joint density of $S_1$, $S_2$ and $S_3$ which is equal to $3!/t^3$. Then pdfs of $S_1$ and $S_2$ are $1/t$ and pdf of $S_3 = 6/t$. Or maybe they are all the same $(2/t)$?

$\endgroup$
3
$\begingroup$

$S_1$ and $S_2$ are not uniformly distributed, and just where you get $[0,1]$ you don't tell us. The support of the distribution of each of the three random variables $S_1$, $S_2$, $S_3$ is $[0,t]$, not $[0,1]$.

In fact $S_1,S_2,S_3$ have the same distribution as the order statistics from an i.i.d. sample of size $3$ from a uniform distribution. That implies that none of them is uniformly distributed.

Unconditionally, or marginally (if you prefer that word), we have $\Pr(S_1<s) = 1 - e^{-\lambda s}$ for $s>0.$ We want $\Pr(S_1<s\mid N(t)=3)$.

\begin{align} & \Pr(S_1<s \mid N(t) = 3) = \frac{\Pr(S_1<s\ \&\ N(t)=3)}{\Pr(N(t)=3)} \\[10pt] = {} & \Pr(N(s)\ge 1 \mid N(t)= 3) \\[10pt] = {} & \frac{\Pr\left( \begin{array}{r} \big(N(s)=1\ \&\ N(t)-N(s) = 2\big) \\ \text{ or } \big( N(s)=2\ \&\ N(t)-N(s)=1 \big) \\ \text{ or } \big( N(s)=3\ \&\ N(t)-N(s)=0 \big) \end{array} \right)}{\dfrac{\lambda^3 e^{-\lambda t}}{3!}} \\[10pt] = {} & \frac{ \left( \begin{array}{r} \dfrac{(\lambda s)^1 e^{-\lambda s}}{1!} \cdot\dfrac{(\lambda(t-s))^2 e^{-\lambda (t-s)}}{2!} \\[10pt] {} + \dfrac{(\lambda s)^2 e^{-\lambda s}}{2!} \cdot \dfrac{(\lambda(t-s))^1 e^{-\lambda (t-s)}}{1!} \\[10pt] {} + \dfrac{(\lambda s)^3 e^{-\lambda s}}{3!} \cdot\dfrac{(\lambda(t-s))^0 e^{-\lambda (t-s)}}{0!} \end{array} \right) }{\dfrac{\lambda^3 e^{-\lambda t}}{3!}} \\[10pt] = {} & \frac{\dfrac{s^1 (t-s)^2}{1!2!} + \dfrac{s^2(t-s)}{2!1!} + \dfrac{s^3}{3!}}{1/(3!)} \\[10pt] = {} & \binom 3 1 s^1 (t-s)^2 + \binom 3 2 s^2 (t-s)^1 + \binom 3 3 s^3. \end{align}

This is the c.d.f. on the interval $[0,t]$. Differentiating it with respect to $s$ gives you the p.d.f. on that interval. If you differentiate this with respect to $s$, using the product rule, you'll get five terms, and all but one of them will cancel.

$S_2$ and $S_3$ can be handled similarly.

$\endgroup$

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.