10
$\begingroup$

How to generate numerically a set of random points $(x_1,y_1), (x_2,y_2),\cdots, (x_N,y_N)$ such that the pair-wise distances

$d = \sqrt { (x_i-x_j)^2 + (y_i-y_j)^2}$, for all $ 0<i\le N, 0<j\le N $

satisfy some given distribution $P(d)$ (e.g., Gaussian, exponential, etc.).

$\endgroup$
  • $\begingroup$ Interesting question - where is it from? $\endgroup$ – nbubis Dec 17 '13 at 6:58
  • 3
    $\begingroup$ Also note that the distribution of distances can never be Gaussian, since $d>0$ :) $\endgroup$ – nbubis Dec 17 '13 at 6:59
  • $\begingroup$ If you want to exclude $i=j$ then it might be neater to write $1 \le i \lt j \le N$ $\endgroup$ – Henry Dec 17 '13 at 8:04
  • 1
    $\begingroup$ The question is from mixing two components, e.g., fine particles in the air; stars distribution in the universe, etc. I distilled it into this general form. The question may not be well posed but to me it made some sense. $\endgroup$ – user43286 Dec 17 '13 at 17:53
  • 1
    $\begingroup$ Yes, the number of samples is usually large. $\endgroup$ – user43286 Dec 17 '13 at 21:50
1
$\begingroup$

If you sample all Points independently from a fixed spatial Distribution with density $g:\mathbb R^2 \to [0,\infty)$ then every two distinct Points have the same distance Distribution. Now if $F$ is the cumulative density function of the Distribution $P(d)$, it will have to solve the following equation $$ F(r) = \mathbb P(\|x_1-x_2\|_2 \leq r) = \int_{\mathbb R^2} \int_{B_r(x_1)} g(x_1) g(x_2) dx_2 dx_1 = \int_{\mathbb R^2} g(x) (g \ast \mathbf 1_{B_r})(x) dx.$$ Here $B_r(x)$ is the Ball around $x$ with radius $r$. Now this is some Kind of integral equation, which to solve numerically is another question in the numerics section.

$\endgroup$

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.