Let $N_1(t),~N_2(t),~N_3(t)$ be independent Poisson processes of rates $\lambda_1,~\lambda_2,~\lambda_3>0$, respectively. Evaluate the probability between two successive arrivals of $N_1$ we get at least one arrival of $N_2$ and at least one arrival of $N_3$.

Attempt. The probability of getting an arrival of $N_i$ (and not of $N_j,~j\neq i$) is $\frac{\lambda_i}{\lambda_1+\lambda_2+\lambda_3}.$ If we already have an arrival of $N_1$, then for

$E_i$= the event we get at least one arrival of $N_i$,

we seek $P(E_2\cap E_3)=P(E_2)P(E_3).$ So:


and similar we work for $E_3.$ Am I on the right path?

Thanks in advance.

  • $\begingroup$ I think that the problem can be rephrased as: find $P(X_1<X_3\text{ and }X_2<X_3)$ where $X_i$ has exponential distribution with parameter $\lambda_i$. Here $X_i$ corresponds with the first arrival time connected with process $N_i(t)$. $\endgroup$ – drhab Sep 5 '18 at 9:38
  • $\begingroup$ @drhab well, wouldn't it be $\mathbb{P}(X_2 < X_1 \text{ and } X_3 < X_1)$ ? $\endgroup$ – P. Quinton Sep 5 '18 at 10:21
  • 2
    $\begingroup$ I don't think $E_2$ and $E_3$ are independent. They both make early arrivals of $N_1$ less likely, so they should correlate positively? $\endgroup$ – joriki Sep 5 '18 at 10:26
  • $\begingroup$ @P.Quinton Yes, that is what I meant to write of course. $\endgroup$ – drhab Sep 5 '18 at 11:33

After the initial $N_1$ you need to either first get $N_2$ and then $N_3$ before $N_1$, or first $N_3$ and then $N_2$ before $N_1$. Thus the probability is

$$ \frac{\lambda_2}{\lambda_1+\lambda_2+\lambda_3}\frac{\lambda_3}{\lambda_1+\lambda_3}+\frac{\lambda_3}{\lambda_1+\lambda_2+\lambda_3}\frac{\lambda_2}{\lambda_1+\lambda_2}=\frac{\lambda_2\lambda_3}{\lambda_1+\lambda_2+\lambda_3}\left(\frac1{\lambda_1+\lambda_3}+\frac1{\lambda_1+\lambda_2}\right)\;. $$

  • $\begingroup$ If you are using the memoryless property for this, perhaps you might say so $\endgroup$ – Henry Sep 5 '18 at 11:02
  • 1
    $\begingroup$ @Henry: I'm not sure whether you'd count this as "using the memoryless property". I'm using the fact that Poisson processes with rates $\lambda_i$ can be combined into a single Poisson process (with rate $\sum_k\lambda_k$) in which each arrival is independently of type $i$ with probability $\lambda_i/\sum_k\lambda_k$. This is stated in the question, so I thought it would be OK to use it. $\endgroup$ – joriki Sep 5 '18 at 12:23
  • $\begingroup$ I would say that your $\frac{\lambda_3}{\lambda_1+\lambda_3}$ term uses the memoryless property, since you are saying once $N_2$ has happened first, the probabilities in the subsequent race between $N_3$ and $N_1$ are the same as they would have been at the start $\endgroup$ – Henry Sep 5 '18 at 12:56
  • $\begingroup$ @Henry: I see. In that case, most of what I know about Poisson processes uses the memoryless property. $\endgroup$ – joriki Sep 5 '18 at 13:01

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.