1
$\begingroup$

On page 217 - 218 of the pdf of this book, the author derives the Poisson Distribution using gamma and exponential densities. The author defines $S_n$ to be the sum of a sequence of independent exponentially distributed random variables ($X_1 + X_2 .... + X_n$) with parameter $\lambda$ and $Y$ to count the number of emissions in a time interval $t$, and then states the following:

$$P(Y = n) = P(S_n < t \space and \space S_{n+1} > t) = P(S_n < t) - P(S_{n+1} <t)$$

The reasoning being that $S_{n+1} < t$ is a subset of the event $S_{n} < t$

  1. I don't understand why $P(Y = n)$ is equal to $P(S_{n} < t < S_{n+1})$, why the event $S_{n+1} < t$ is a subset of $S_{n} < t$ (and not the other way around?), and why we take the difference of the two?

  2. Is the $\lambda$ defined for the sequence of exponential densities, in the gamma density (used to model $S_n$), and the final Poisson distribution, all the same?

  3. To simulate a Poisson random variable ($W$) with parameter $\lambda$ using exponential densities, the author proposes that $-\frac{1}{\lambda}log(rnd)$ (which simulates an exponential density with parameter $\lambda$) be run sequentially until $S_{n} < 1 < S_{n+1}$ and the $n$ returned is the simulated value for $W$. Would I replace the $1$ with $t$ if I want to simulate a Poisson random variable with parameter $\lambda t$?

$\endgroup$
1
$\begingroup$

If $Y(t) = \sum_{n=1}^\infty \mathsf 1_{(0,t]}(S_n)$ is the counting process associated with the renewal times $\{S_n\}$, then by definition $$\{Y(t) = n\} = \{S_n\leqslant Y(t) < S_{n+1}\}. $$ This is simply because $n$ renewals have occured by time $t$, and the $(n+1)^{\mathrm{th}}$ renewal has yet to occur. As for $\{S_{n+1}<t\}\subset\{S_n<t\}$, we have $$ \{S_{n+1}<t\} = \left\{\sum_{i=1}^{n+1}X_i<t \right\} = \left\{X_{n+1}+\sum_{i=1}^n X_i<t\right\}\subset\left\{\sum_{i=1}^n X_i<t\right\} = \{S_n<t\} $$ since $\mathbb P(X_{n+1}\geqslant 0)=1$.

$\lambda$ is a constant, and is the same everywhere.

To simulate a Poisson random variable with rate $\lambda t$, you would use $-\frac1{\lambda t}\log U$ where $U$ is uniformly distributed over $(0,1)$.

$\endgroup$
  • $\begingroup$ 1. Is it fair to argue that if $S_{n+1} \leq t$ means $S_n \leq t$ but the reverse is not always true, hence the event $S_n \leq t$ includes the event where $S_{n+1} \leq t$ and thus the latter is a subset of the former? 2. To generate a Poisson r.v. $W$, what if I tweak the procedure in my book: generate a sequence of exponential r.v. each with $-\frac{1}{\lambda}log(rnd)$, keep track of subtotals $S_k$ (sum of the exponential r.v.), and if $W$ has parameter $\lambda t$ instead of $\lambda$, I take $n$ such that $S_n < t < S_{n+1}$ instead of $S_n < 1 < S_{n+1}$ per my book. Is this valid? $\endgroup$ – Yandle Jan 21 at 18:50

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.