Number of exponentially distributed RVs is Poisson distributed (induction)

55 Views Asked by At

The problem is:

Let $\tau_1,\tau_2,\dots$ be independent exponentially distributed with parameter $\lambda >0$. Define $N_t:=|${$ k \geq 1 | \tau_1+\dots + \tau_k\leq t$}$ |$.

Show that $N_t$ is Poisson distributed with parameter $\lambda t$.

This is shown by induction:

Firstly $\mathbb{P}(N_t=0)=e^{-\lambda t}$. Now suppose that $\mathbb{P}(N_t=k)=e^{-\lambda t}\frac{(\lambda t)^k}{k!}$ for all $t>0$. Then by conditioning on $\tau_1$ we find

$$ \mathbb{P}(N_t=k+1)=\int_0^t\lambda e^{-\lambda x}\mathbb{P}(N_{t-x}=k)\text{d}x=\dots $$

which then yields the result easily, but I completely don't understand that step.

I tried starting with something like this:

$$ \mathbb{P}(N_t=k+1)=\int_0^{\infty}\lambda e^{-\lambda x}\mathbb{P}(N_t=k+1 | \tau_1=x)\text{d}x $$

but I couldn't figure out how to obtain the result from the solution. Can you please help me with that?

Thanks!

1

There are 1 best solutions below

2
On BEST ANSWER

Let $S_{n}:=\sum_{i=1}^{n}\tau_{i}$ and $R_{n}=\sum_{i=2}^{n+1}\tau_{i}$ and observe that $S_{n}$ and $R_{n}$ have the same distribution.

We have: $$N_{t}=\max\left\{ n\mid S_{n}\leq t\right\} =\max\left\{ n\mid\tau_{1}+R_{n-1}\leq t\right\} $$

Then under condition $\tau_{1}=x<t$ we have: $$N_{t}=\max\left\{ n\mid x+R_{n-1}\leq t\right\} =\max\left\{ n\mid R_{n-1}\leq t-x\right\} $$ where: $$\max\left\{ n\mid R_{n-1}\leq t-x\right\} \sim\max\left\{ n\mid S_{n-1}\leq t-x\right\} =$$$$1+\max\left\{ n-1\mid S_{n-1}\leq t-x\right\} =1+N_{t-x}$$(here $LHS\sim RHS$ means that both sides have the same distribution)

For $x<t$ this leads to: $$P\left(N_{t}=k+1\mid\tau_{1}=x\right)=P\left(1+N_{t-x}=k+1\right)=P\left(N_{t-x}=k\right)$$