1
$\begingroup$

Consider three independent Poisson processes $N_1(t), N_2(t),N_3(t)$ with rates $\lambda_1,\lambda_2,\lambda_3$, respectively. What is the probability of reaching $(i,j,k)$?

I know that for the two-dimensional process, we have the probability of reaching $(i,j)$ as $$\frac{(i+j)!\lambda_1^i\lambda_2^j}{i!j!(\lambda_1+\lambda_2)^{i+j}}$$ Could the problem be easily extended to the three-dimensional Poisson process as $$\frac{(i+j+k)!\lambda_1^i\lambda_2^j\lambda_3^k}{i!j!k!(\lambda_1+\lambda_2+\lambda_3)^{i+j+k}}$$

I have reason to believe so since by considering $\lambda_1=\lambda_2=\lambda_3=1$; $i=j=k=n$, we have the probability of reaching $(n,n,n)$ as $$\frac{3n!\left(\frac{1}{3}\right)^{3n}}{n!n!n!}$$

$\endgroup$
  • $\begingroup$ Yes, this is correct. $\endgroup$ – Math1000 Oct 4 '16 at 1:50
0
$\begingroup$

$$\int_0^t P(N_1(t)=i,N_2(t)=j,N_3(t)=k) dt$$ is the average amount of time the process spends in $(i,j,k)$. This is equal to:

$$(*)\quad P(\mbox{ever reaching }(i,j,k)) \times \frac{1}{\lambda_1+\lambda_2+\lambda_3},$$

because time process spends at any state it arrives to is $\mbox{Exp}(\lambda_1+\lambda_2+\lambda_3)$.

This is also equal to

\begin{align*} &\int_0^\infty P(N_1(t) =i) P(N_2(t)=j) P(N_3(t)=k) dt\\ & =\int_0^\infty e^{-(\lambda_1+\lambda_2+ \lambda_3)t} t^{i+j+k} \frac{ \lambda_1^i\lambda_2^j \lambda_3^k}{i! j! k!}d t\\ &= \frac{ \lambda_1^i\lambda_2^j \lambda_3^k}{i! j! k!}(-1)^{i+j+k} \frac{\partial ^{i+j+k}}{\partial \lambda^{i+j+k}}\int_0^\infty e^{-(\lambda_1+\lambda_2+\lambda_3)t}dt \\ (**) \quad& = \frac{ \lambda_1^i\lambda_2^j \lambda_3^k}{i! j! k!}(i+j+k)!(\lambda_1+\lambda_2+\lambda_3)^{-(i+j+k+1)}. \end{align*}

Equating $(*)$ to $(**)$ we obtain

$$P(\mbox{ever reaching }(i,j,k))= \frac{(i+j+k)!}{i! j! k!}\frac{\lambda_1^i\lambda_2^j\lambda_3^k}{(\lambda_1+\lambda_2+\lambda_3)^{i+j+k}}.$$

$\endgroup$

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.