0
$\begingroup$

Let $L_j$ denote the lines of a planar Poisson line process. By Example 20 of

http://staffhome.ecm.uwa.edu.au/~00025879/reprints/chapter1-abstract.html http://staffhome.ecm.uwa.edu.au/~00025879/reprints/chapter1.ps

inclination angles $\omega_{j}$ of the lines $L_j$ relative to the $x$-axis are independent and identically distributed with density $\sin(\omega)/2$ on $[0,\pi]$. In particular, the vertical lines are weighted more than horizontal lines.

The R package spatstat has planar random process simulation capabilities. I can generate Poisson lines in a window via rpoisline and determine their inclination angles $\omega_{j}$ via angles.psp (with option directed=FALSE). To avoid edge effects, I restrict attention to only those lines fairly close to the window center. Surprisingly, the angles seem to follow not the sine density, but instead the uniform density on $[0,\pi]$.

What am I missing please? I would be grateful for help, e.g., R, Matlab or Mathematica code giving experimental results that match theoretical prediction.

Thank you very much.

$\endgroup$
0
$\begingroup$


The following code shows the paradox mentioned by @adrian-baddeley experimentally. In a circular sampling window we see a nearly uniform distribution of angles:

library(spatstat)
X  <- rpoisline(400, win = disc())
aa <- angles.psp(X)
plot(density(aa))

In a long thin sampling window we see a nearly sinusodial distribution of angles:

X  <- rpoisline(1000, win = owin(c(0,10), c(0,.01)))
aa <- angles.psp(X)
plot(density(aa))
plot(function(x){0.5*sin(x)},from=0,to=pi,add=TRUE,col="red")

$\endgroup$
  • $\begingroup$ Thank you for this elegant solution. Your technique of making the window "line-like" is certainly effective & helpful. $\endgroup$ – S. Finch Jan 22 '18 at 13:52
0
$\begingroup$

This is a classic paradox.

If you consider the random lines which intersect a given, fixed line, then these random lines have angles which are non-uniformly distributed with probability density proportional to the sine of the incidence angle. If you consider the random lines which intersect a given circle then these random lines have uniformly-distributed orientation angles. In each case the bold text describes a selection or sampling operation, and sampling operations introduce bias.

$\endgroup$
  • $\begingroup$ Thank you for writing! The theorem is indeed paradoxical. I am trying unsuccessfully to confirm this experimentally using "rpoisline" & "angles.psp". Are you able to successfully do this? I appreciate deeply the chance to correspond with you. $\endgroup$ – S. Finch Jan 19 '18 at 15:08

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.