2
$\begingroup$

My book outlines a procedure but a preliminary part of it is unclear to me.

Let X be the number of occurences of an event over a unit of time and assume that it has a Poisson distribution with mean $m=\lambda $. Let $T_1, T_2 , T_3, \ldots $ be the interarrival times of the occurences and they are iid with an exponential $\lambda $ distribution. Note that $ X=k $ iff $$ \sum_{j=1}^k T_j \leq 1 \quad \text{and} \quad \sum_{j=1}^{k+1} T_j >1. $$

This is precisely what I do not understand. Why does the total waiting tme until $k$ occurences have to be less than or equal to 1? Any help is greatly appreciated. Thank you.

|cite|improve this question
$\endgroup$
  • 2
    $\begingroup$ "Let $X$ be the number of occurrences of an event over a unit of time..." $\endgroup$ – cardinal Nov 24 '13 at 22:25
3
$\begingroup$

So if $T_1, T_2, \dots$ are the interval times of the occurrences, the first event occurs at $t = T_1$, the second at $t = T_1 + T_2$, etc.

Then $X = {}$ the number of occurrences during the first unit of elapsed time is $$ X = \max_k T_1 + \cdots + T_k \le 1.$$

Hence if $X = k$, $T_1 + \cdots + T_k \le 1$ and $T_1 + \cdots + T_{k+1} > 1$.

PS. I think the simplest/fastest way to simulate a Poisson process with constant rate $\lambda$ on the interval $(0, t)$ is

  1. draw $X \sim \mathcal P (\lambda t)$ the number of events

  2. draw $u_1, \dots, u_X$ uniformly in $(0,t)$ and order them: these are the times of occurrences of the $X$ events.

|cite|improve this answer
$\endgroup$
  • $\begingroup$ Thank you but my book suggests another algorithm, which I have not figured out yet. 1. Set X=0 and T=0 2. Generate U uniform (0,1) and let $Y=-(1/ \lambda) log(1-U)$, 3. SET T=T+Y 4. If T>1, output X, else set X=X+1 and go to step 2. I have not been able to understand it yet. $\endgroup$ – JohnK Nov 25 '13 at 14:17
  • $\begingroup$ I was assuming that uou can use a function for generation of random Poisson deviates, such as rpois in R. Your book has a more elementary approach. If you let $F(x) = 1 - e^{-\lambda x}$ be the cdf of the exponential distribution $\mathcal E(\lambda)$, then you have $Y = F^{-1}(U)$, from which you should be able to prove that, $U$ being uniform, $Y \sim \mathcal E(\lambda)$. Then the successive values of $T$ along the iterations of steps 2 and 3 are $T_1$, $T_1+ T_2$, etc. You should be able to conclude. $\endgroup$ – Elvis Nov 25 '13 at 15:00
  • $\begingroup$ Oh that's right. Thanks again man. My book, is an intermediate one in Mathematical Statistics, but I am a novice studying on my own. $\endgroup$ – JohnK Nov 25 '13 at 15:04

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.