A kid counts the cars passing a certain street according to a Poisson process of rate λ. When a given α time units passes without any car arriving, he gets bored and leaves.

(a) How much time does he spend there on average? Hint: Condition on the time of the first car.

(b) How many cars did he count on average?

I'm not sure how to approach question a), so far i have something like this: $s_i$ is the time it takes for the $i$-th event to happen and has a memoryless property. $\alpha P(s_1>\alpha) + E[(s_1+\alpha \mid s_1<\alpha)]+E[(s_2+\alpha \mid s_2<\alpha)]+\cdots$ and i think this will converge to some finite value, however I'm not sure if this approach is the right way of doing it.

for part b) I did: $\sum^\infty_{k=0} k P(T_{k+1}>\alpha)P(T_k<\alpha)\cdots P(T_1<\alpha)$, which will be $k$ times a geometric series times $P(T_{k+1}>\alpha)$ but the problem I encountered is that after writing the geometric series and did the multiplication my answer to the question is $k.$ I did this: $P(T_{k+1}>\alpha)\frac{k}{e^{-\lambda \alpha}} = e^{-\lambda \alpha} \frac{k}{e^{-\lambda \alpha}}$, $e^{-\lambda \alpha}$ cancels out and leaves me $k$, which I know it is definitely wrong but I'm not sure how I would approach this question differently.

  • $\begingroup$ "$s_i$ is the time it takes for the $i$-th event to happen and has a memoryless property." That is unclearlly written. The time from the $(i-1)\text{th}$ event until the $i\text{th}$ event is what has a memoryless distribution. The time from the beginning until the $i\text{th}$ event does not. $\endgroup$ Mar 14 '18 at 1:56
  • $\begingroup$ You never defined $T_i.$ Is it the same thing as $s_i$? $\endgroup$ Mar 14 '18 at 1:57
  • $\begingroup$ @MichaelHardy yes it is the time of one arrival. Do you think the approach above is correct? $\endgroup$
    – PiCubed
    Mar 14 '18 at 2:44
  • 1
    $\begingroup$ I've deleted my answer; maybe tomorrow I'll edit it further and then undelete it. $\endgroup$ Mar 14 '18 at 3:13
  • $\begingroup$ @MichaelHardy I found out what I did wrong for part b), but still need a little help with part a), thank you! $\endgroup$
    – PiCubed
    Mar 14 '18 at 4:32

We will need to know the conditional expected time of the first car's arrival given that at least one car arrives before time $\alpha.$

Let $S$ be the time of the first car's arrival, so $\operatorname E S = 1/\lambda.$

We have $\Pr(S>t) = e^{-\lambda t}$ (for $t\ge0$).

Therefore $\Pr(S>t\mid S<\alpha) = \dfrac{\Pr(t<S<\alpha)}{\Pr(S<\alpha)} = \dfrac{e^{-\lambda t} - e^{-\lambda\alpha}}{1-e^{-\lambda\alpha}}.$

From this we get the conditional distribution $$ f_{S\,\mid\,S\,<\,\alpha} (t) \, dt = \left(-\frac d {dt}\, \frac{e^{-\lambda t} - e^{-\lambda\alpha}}{1-e^{-\lambda\alpha}} \right) dt = \frac{e^{-\lambda t}}{1-e^{-\lambda\alpha}} (\lambda\,dt) \quad \text{for } 0<t<\alpha, $$ and from this we get the conditional expected value: $$ \operatorname E(S\mid S<\alpha) = \int_0^\alpha t \, \frac{e^{-\lambda t}}{1-e^{-\lambda\alpha}} (\lambda \, dt) = \frac 1 \lambda \cdot \frac{1-e^{-\lambda\alpha}- \lambda\alpha e^{-\lambda\alpha}}{1-e^{-\lambda\alpha}}. $$

The boy counts cars until an interval of length $\alpha$ appears with no cars.

Let $\beta$ be the average time the boy is there. Let $T$ be the total time the boy is there.

Then \begin{align} \beta & = \operatorname ET = \operatorname E(T\mid S>\alpha) \Pr(S>\alpha) + \operatorname E(T\mid S<\alpha) \Pr(S<\alpha) \\[10pt] & = \alpha \Pr(S>\alpha) + \operatorname E(T + S\mid S<\alpha) \Pr(S<\alpha) \\[10pt] & = \alpha\Pr(S>\alpha) + \Big( \operatorname E(T) + \operatorname E(S\mid S<\alpha) \Big) \Pr(S<\alpha) \\[10pt] & = \alpha\Pr(S>\alpha) + \Big( \beta + \operatorname E(S\mid S<\alpha) \Big) \Pr(S<\alpha). \end{align}

Since we know how to express $\Pr(S>\alpha),$ $\Pr(S>\alpha),$ and $\operatorname E(S\mid S<\alpha)$ as functions of $\alpha$ and $\lambda,$ we can substitute those functions of $\alpha$ and $\lambda$ into that last line, and then we have an algebraic equation that can be solved for $\beta$ as a function of $\alpha$ and $\lambda.$

I got $\displaystyle \beta = \alpha + \frac 1 \lambda\left( 1 - e^{-\lambda\alpha} - \lambda\alpha e^{-\lambda\alpha} \right).$


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.