1
$\begingroup$

How can I find the expectation of arrival times on an interval $[0,t]$ of a non-homogeneous poisson process.

For example if customers arrive to a shop following a non-homogeneous poisson process with a rate function $\lambda(t)$, how can I find the expected arrival time of customers in the time interval $[0,t]$.

That is, given that some customers have arrived at the shop between time $0$ and time $t$, what is the average time of their arrivals?

I know that with homogeneous poisson processes, the arrival times on an interval are uniformly distributed due to Campbell's theorem. However, I can't find anything similar for non-homogeneous poisson processes.

$\endgroup$

1 Answer 1

1
$\begingroup$

Let $\mathcal{P}_n=\Big\{[t_{k-1},t_k),t_k^*\Big\}_{k=1}^n$ be a sequence of tagged partitions of $[0,t)$ such that $\max_{1\leq k \leq n}\Delta t_k\rightarrow 0$ as $n \rightarrow +\infty$. Take $X_k$ as the number of arrivals on the time interval $[t_{k-1},t_k)$. When $n$ is large we have $$X_k\approx \text{Poisson}\big(\lambda(t_k^*)\Delta t_k\big)$$

Then $S_k=X_1 + \dots + X_k$ counts the number of arrivals on $[0,t)$ and is approximately $\text{Poisson}\Big(\sum_{k=1}^n\lambda(t_k^*)\Delta t_k\Big)$ which clearly tends to $\text{Poisson}\Big(\int_0^t\lambda(x)\mathrm{d}x\Big)$ as $n\rightarrow \infty$.

Take $T_k$ as the arrival time of the $k^{\text{th}}$ arrival. Then for $t>0$ fixed, $$\{T_k<t\}=\bigcup_{j=k}^{\infty} \{N(t)=j\}$$ where $N(t)$ is the number of arrivals on $[0,t)$. This union is disjoint, so that $$P(T_k<t)=\sum_{j=k}^{\infty}P(N(t)=j)=1-e^{-\int_0^t\lambda(x)\mathrm{d}x}\sum_{j=0}^{k-1}\frac{\Big(\int_0^t\lambda(x)\mathrm{d}x\Big)^j}{j!}$$ Taking a derivative and simplifying yields the pdf for the $k^{\text{th}}$ arrival time, namely $ f_{T_k}$: $$f_{T_k}(t)=\lambda(t)e^{-\int_0^t\lambda(x)\mathrm{d}x}\frac{\Big( \int_0^t\lambda(x)\mathrm{d}x \Big)^{k-1}}{(k-1)!}$$

$\endgroup$
5
  • $\begingroup$ Thanks for the answer but I'm not sure I understand. Does this not just tell you that the number of arrivals in [0,t] is poisson distributed with its parameter being the integral of the rate function between 0 and t. How do I use this to know the distribution of the times of these arrivals as opposed to the distribution of the number of arrivals? $\endgroup$
    – Jamal
    Commented Nov 10, 2021 at 20:15
  • $\begingroup$ My bad. I read your question too quickly. I only illustrated the distribution for the number of arrivals on $[0,t)$ Let me get back to your on the distribution of the arrival times. $\endgroup$
    – user801306
    Commented Nov 10, 2021 at 20:23
  • $\begingroup$ no worries, thank you very much $\endgroup$
    – Jamal
    Commented Nov 10, 2021 at 20:29
  • $\begingroup$ I updated my answer to illustrate the pdf of the $k^{\text{th}}$ arrival. Note taking $\lambda(t)$ to be constant yields the appropriate Erlang distribution. $\endgroup$
    – user801306
    Commented Nov 10, 2021 at 21:29
  • $\begingroup$ Thanks, so to find the expectation, I would take the integral of this pdf multiplied by t and then after that plug in my value of t? $\endgroup$
    – Jamal
    Commented Nov 11, 2021 at 8:44

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .