Given $\lbrace Y_i\rbrace$ a non-homogenous Poisson process with mean density $\theta y^{-1}e^{-y}$ where $y>0$ $(\theta>0)$. I.e., the number of points of $\lbrace Y_i\rbrace$ in $(a,b)$ with $0<a<b\le \infty$ has a Poisson distribution with mean $\int_a ^b \theta y^{-1}e^{-y}dy$.

How can the density of $Y_j$ be expressed as $$\theta y^{-1}\frac{(\theta E_1(y))^{j-1}}{(j-1)!}e^{-\theta E_1(y)}, y>0 \tag{$*$}$$

where $E_1(y)$ is the exponential integral $$E_1(y)=\int_y ^\infty \frac{e^{-w}}{w}dw\,\,?$$ I found this in page 3 of http://www.stats.ox.ac.uk/~griff/pd.pdf

I know how to find the number of $Y_j$ in a given interval: $$\mathbb{P}(N(b)-N(a)=n)=\frac{(\int_a ^b \theta y^{-1}e^{-y}dy)^n \exp(-\int_a ^b\theta y^{-1}e^{-y}dy )}{n!}$$

Although this formula has some similarities to the desired formula, I am not sure how to derive $(*)$.


the desired density $\rho_j(y)d\epsilon$ is the probability that there is an event in the interval $(y,y+d\epsilon)$, with $d\epsilon$ infinitesimal, multiplied by the independent probability that there are $j-1$ events in the interval $(y,\infty)$:

$$\rho_j(y)=\left(\lim_{\epsilon\rightarrow 0}\frac{1}{\epsilon}\mathbb{P}(N(y+\epsilon)-N(y)=1)\right)\left(\lim_{b\rightarrow\infty}\mathbb{P}(N(b)-N(y)=j-1)\right)$$ $$=\theta y^{-1}e^{-y}\frac{1}{(j-1)!}\bigl(\theta E_1(y)\bigr)^{j-1} \exp\bigl(-\theta E_1(y) \bigr)$$

the formula $(*)$ in the OP is missing a factor $e^{-y}$.


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.