2

Given $n \ge 3$ points with real number coordinates, distributed randomly in a unit square, what is the probability $P$ that 3 points exist which are collinear?

My reading of this answer to a similar question indicates to me that $P=0$ for any finite $n$. Is this correct?

What happens when $n \to \infty$? Does $\lim \limits_{n \to \infty} P = 1$?

Jens
  • 5,786
  • 1
    If $P=0$ for any finite $n$, then the limit as $n\to\infty$ of $P$ is certainly zero as well. This is equivalent to picking a countably infinite number of points in the unit square, and wondering if any three are collinear. However, a more interesting question to ask would be... what if we pick an uncountably infinite random subset with area $A\lt 1$? Then what is the probability of three collinear points? – Franklin Pezzuti Dyer Oct 05 '18 at 21:01
  • @Frpzzd So $P=0$ for any $n$? Regarding your interesting question, is the answer known? – Jens Oct 05 '18 at 21:06
  • Oh, I don't know if $P=0$ for any $n$, I just trusted what you said in the question (however, that would certainly be my guess). I have no clue about the uncountable version of the problem, though... I can't even grasp the concept of choosing an "uncountable sample," lest it be something well-behaved like a random interval or disk with random radius and center. – Franklin Pezzuti Dyer Oct 05 '18 at 21:10
  • 1
    There is no such thing as a "random subset" without specifying a particular distribution. – Robert Israel Oct 05 '18 at 21:13
  • @Robert Israel: Assume a uniform random distribution. – Jens Oct 05 '18 at 21:15
  • @Laz: Hmm... I don't know. What can I say then? – Jens Oct 05 '18 at 21:17
  • @Laz: No, sorry, smooth manifolds and Grassmanians are terms I have to google. – Jens Oct 05 '18 at 22:31
  • 1
    @Laz: Not sure how you elevated the question to more than $2$ dimensions. But if your comment is an answer, please post it. I probably won't understand it, but if others do and upvote, I'll feel confident the answer is correct. Which is what I'm looking for; a correct answer. :) – Jens Oct 05 '18 at 22:38
  • There is a uniform distribution for points in a unit square, but there is no "uniform random distribution" for sets of points. – Robert Israel Oct 06 '18 at 00:47

1 Answers1

4

Let $A$ be the event "$P_1,P_2,P_3$ are collinear". Then $A \iff 0, P_2-P_1, P_3-P_1$ are collinear (I just performed a translation). Then if we could prove that for $v,w$ varying uniformly in the plane, $P(B)=0$ for $B$ the event "$0,v,w$ are collinear", we would have $p(A)=0$. Now observe $B\iff det(v,w)=0$ (I first suppose we are talking about points in the plane, and look at $v,w$ as vectors or as points). But since $det(v,w)$ can take any value in $\mathbb{R}$ with the same probability $0$ as $v,w$ vary uniformly in the plane, this probability is just $\frac{\mu({0})}{\mu({\mathbb{R}})}=0$, where $\mu$ denotes the Lebesgue measure in $\mathbb{R}$, wich measures the "lenght" of subsets.
Now, for the general case, if you have $k$ points ($k\geq 3$) varying uniformly in Euclidean space $\mathbb{R}^n$ ($n\geq 2$), they being collinear implies their projections onto $\mathbb{R}^2\times 0_{\mathbb{R}^{n-2}}$ (or another suitable $2$-plane) are collinear in this plane, and we just proved that three of them being collinear has probability $0$, let alone more than three.
Did this help you?
$\textbf{Added}:$ With the reasoning above, you can treat a similar problem: compute the probability that $n+1$ points in $\mathbb{R}^n$ lie in a hyperplane. First solve the problem for $n=3$:
$P_1, P_2, P_3, P_4$ lie in a plane $\iff 0, P_2-P_1, P_3-P_1, P_4-P_1$ lie in a plane. Now construct the map $det(v_1, v_2, v_3)$ and treat this as before, for $v_1, v_2, v_3$ varying uniformly in $\mathbb{R}^3$. Then you can extend it to the general case again using projections onto $\mathbb{R}^3\times 0_{\mathbb{R}^{n-3}}$ (or another suitable $3$-plane).

Laz
  • 1,796