Assume there is post office that is run by two clerks $1$ and $2$. Customers arrive at times that follow exponential distribution with rate $λ_a$. The amount of time that the clerks 1 and 2 serve are exponentially distributed with mean $1/λ_1$ and $1/λ_2$. The first customer that arrives in the post office is served by clerk $1$ and the second by clerk $2$. The three exponential random variables are independent. Two customers A and B have decided to go to the post office. What is the probability that A arrives before and departs after B?

So, $T_1$ = time taken by clerk $1$ and $T_2$ = time taken by clerk 2. To solve for the probability, if A is the first one there, she goes to clerk $1$ and clerk $1$ has to be faster than clerk $2$ ($T_1 < T_2$). Now what?

  • $\begingroup$ How do you know that A is the first one to arrive? Are you just giving the name A to the first one to arrive? If so, your question would be: What is the probability that the first one to arrive is also the first one to depart? Construed that way, it would be a well-formed math problem, and it would explain why the number of exponential random variables is three, but it's not completely clear from your phrasing that that is what you meant. $\endgroup$ – Michael Hardy Nov 29 '17 at 17:24
  • $\begingroup$ @MichaelHardy I fixed it, does it make more sense now? I still want the probability that A arrives before and departs after B, I just don't know how to go about that. $\endgroup$ – AmaC Nov 30 '17 at 17:02
  • $\begingroup$ You refer to three exponential random variables. One of them is the time taken by clerk $1.$ Another is the time taken by clerk $2.$ What is the third? I am guessing that its the time between the arrival of A and the arrival of B, but so far that's only a guess. $\endgroup$ – Michael Hardy Nov 30 '17 at 17:12
  • $\begingroup$ Yes, that makes sense, $λ_a$. $\endgroup$ – AmaC Nov 30 '17 at 17:24
  • $\begingroup$ So apparently the question is to be contrued as follows: Let $T_1$ and $T_2$ be the service times of the first and second customers respectively. Let $T_0$ be the time from the arrival of the first customer until the arrival of the second. Then the time of departure of the first customer is $T_1$ and the time of departure of the second customer is $T_0+T_2.$ The question then is: What is $\Pr( T_1<T_0+T_2 )\text{ ?} \qquad$ $\endgroup$ – Michael Hardy Nov 30 '17 at 18:46

The question of the probability density of $T_0+T_2$ arises: \begin{align} f_{T_0+T_2}(t) & = \int_0^t f_{T_0}(s) f_{T_1} (t-s) \, ds = \int_0^t (e^{-\lambda_a s} \cdot\lambda_a)\, e^{-\lambda_2(t-s)} (\lambda_2\,ds) \\[10pt] & = \lambda_a\lambda_2 e^{-\lambda_2 t} \int_0^t e^{(\lambda_2-\lambda_a)s} \,ds =\lambda_a \lambda_2 e^{-\lambda_2 t} \left[ \frac{e^{(\lambda_2-\lambda_a)s}}{\lambda_2 -\lambda_a} \right]_{s\,:=\,0}^{s\,:=\,t} \\[10pt] & = \lambda_a\lambda_2 e^{-\lambda_2 t} \cdot \frac{e^{(\lambda_2 - \lambda_a) t} - 1}{\lambda_2 - \lambda_a} = \frac{\lambda_a\lambda_2}{\lambda_2 - \lambda_a} \left( e^{-\lambda_a t} - e^{-\lambda_2 t} \right) \text{ for } t \ge 0. \end{align} In case $\lambda_a=\lambda_2,$ one might think that the limit of that last expression as $\lambda_a\to\lambda_2$ would be the thing to find. Maybe what is a bit simpler is that the integral on the second line is easy to evaluate in that case.

Then we have \begin{align} \Pr(T_1 < T_0+T_2) & = \operatorname E(\Pr(T_1 < T_0 + T_2) \mid T_0+T_2) = \operatorname E\left( 1 - e^{-\lambda_1(T_0+T_2)}\right) \\[10pt] & = 1 - \int_0^\infty t f_{T_0+T_2} (t) \, dt \end{align} and so on.

To find this last integral, one should recall that

$$ \int_0^\infty t e^{-ct} \, dt = \frac 1 {c^2} \int_0^\infty (ct) e^{-ct} (c\, dt) = \frac 1 {c^2} \int_0^\infty ue^{-u} \, du = \frac 1 {c^2}. $$ (One can derive this by integrating by parts $\displaystyle \int u \Big(e^{-u}\,du\Big) = \int u\,dv = \cdots$


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.