Two disjoint unit discs $D_1$ and $D_2$. Inside them there are random poisson points with intensity $\lambda$. For a given real $r>0$, what is the probability that there exist a poisson point $x_1\in D_1$ and a poisson point $x_2\in D_2$ such that $\|x_1-x_2\|_2 \le r$? Couldn't find any existing results on this.

  • $\begingroup$ How about 1D case, i.e., $D_1$ and $D_2$ are line segments? $\endgroup$
    – user32828
    Apr 19 '14 at 0:51

If $M$ is the $4$-dimensional Lebesgue measure of $\{(x,y) \in D_1 \times D_2: \|x - y\|_2 \le r\}$, then the number of such pairs of points is a Poisson random variable with parameter $M \lambda^2$, so the probability that there is at least one is $1 - \exp(-M \lambda^2)$.

EDIT: Oops, this is wrong. It is not Poisson.

EDIT: Consider the case where $D_1$ and $D_2$ are line segments, say $[0,1]$ and $[a,a+1]$ where $a > 1$. Of course we need $a-1 < r < a+1$ to make the question nontrivial. Let $X$ be the maximum of the Poisson points in $D_1$ ( $-\infty$ if there are none) and $Y$ the minimum of the Poisson points in $D_2$ ($+\infty$ if there are none). $X$ and $Y$ are independent, with densities $f_X(x) = \lambda e^{-\lambda(1-x)}$ and $f_Y(y) = \lambda e^{-\lambda(y-a)})$ on $[0,1]$ and $[a,a+1]$ respectively. Thus if $a-1 < r \le a$ $$P(Y - X \le r) = \lambda^2 \int_{a-r}^1 dx \int_a^{x+r} dy\; e^{-\lambda(1-x)} e^{-\lambda(y-a)} = {1 + \left( \lambda\,(a- r-1)-1 \right)\; {{ e}^{\lambda\, \left( a-r-1 \right) }}} $$ while if $a \le r < a + 1$ $$P(Y - X \le r) = (1-e^{-\lambda})^2 - \lambda^2 \int_0^{a+1-r} dx \int_{x+r}^{a+1} dy\; e^{-\lambda(1-x)} e^{-\lambda(y-a)} = 1 - 2 e^{-\lambda} + (1 - \lambda(a-r+1)) e^{\lambda(a-r+1)}$$

  • $\begingroup$ But how to compute $M$? $\endgroup$
    – user32828
    Apr 7 '13 at 11:58
  • $\begingroup$ Integrate over $x \in D_1$ the area of the intersection of $D_2$ with the disk of radius $r$ about $x$. I doubt that you'll end up with a closed-form expression. $\endgroup$ Apr 8 '13 at 18:41
  • $\begingroup$ Some anon user wished to comment via an edit and an answer that "The number of pairs (x,y) is not a Poisson random variable." and that this creates problems of the answer. I leave this purely as a service as the technicalities of the site prevent to record it otherwise. $\endgroup$
    – user9072
    Apr 17 '14 at 15:54
  • $\begingroup$ Oops, it looks like the anon user is correct. The product of two independent Poisson processes is not a Poisson process. $\endgroup$ Apr 18 '14 at 3:46
  • $\begingroup$ However the idea here could be used to find the expected number of pairs of points within a distance r. With that, the Aldous Poisson Clumping Heuristic could be used to get a decent approximation to the probability. I doubt that its been done for this particular problem though. $\endgroup$
    – guest
    Apr 20 '14 at 11:21

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.