2

I was given this question as a task by a teacher at a Maths Circle:

If three points be taken at random in a given plane, prove that the probability of their being the vertices of an acute triangle is ${4 \over \pi^2} - {1 \over 8}.$

Original Question by W. Woolhouse as far as I can find.

Needless to say, me being clueless resulted in a few searches where I arrived at this MSE question-answer where a user rightly mentions that the question is incomplete without knowing the distribution.

So, is that answer also applicable for my question? That is, do I conclude the same and can be assured that the given proposition cannot be shown without the information regarding the distribution.

Thank you.

Edit: I must mention, this is a challenge "takeaway" and the teacher has explicitly mentioned that the question is above the class level so it is not expected that we solve it. For the "takeaway" challenges, we are free to use web/books/discuss.

  • 1
    Yeah, the MSE question should be right. You can't define a uniform probability distribution, which is the default assumption when nothing is specified, over the entire plane (given the integral of the PDF over the whole plane would have to be $1$, which is impossible for a constant function). I guess I'll detail this more in an answer. – Debalanced Jun 27 '25 at 03:38
  • 2
    It's not even obvious what "natural" distribution the writers had in mind. My guess, but it is just a guess, is that they meant to bound the possible selections. Thus, take the uniform distribution on a disk of radius $r$ for some $r$. Since rescaling a triangle preserves acuteness (or non-acuteness), you can just take $r=1$. But, again, the writers should have specified whatever they intended. – lulu Jun 27 '25 at 04:11

1 Answers1

1

Here's the issue: when we speak of a "random number from $0$ to $1$," for instance, we are usually speaking of a uniform probability distribution, which means that "every number has an equal probability of getting picked." That's actually a very weird thing to say, because any particular number has a $0$ probability of being picked, so we have to go back and explore what it means to have a random number from $0$ to $1$.

If I have a die, it has six sides, each with $\frac{1}{6}$ probability of getting picked. A probability density function $p(x)$ tells us that the probability of the result of the die roll being $n$ is $p(n)$. Obviously the die has to give some result with a $100\%$ ($p=1$) probability. So, $$\sum_{i=1}^{6}p(x_i)=1$$ if there are $6$ possible results and $x_1=1$, $x_2=2$, $x_3=3$ etc. If $n$ is not a possible result, $p(n)= 0$. The natural generalization of this is to continuous random variables, and you probably know that the generalization of discrete Riemann sums is integrals, so we are forced to take $$\int_{-\infty}^{\infty}p(x)\ dx = 1$$ A uniform probability distribution is one where the probability density function $p(x)$ is just a constant value (the same way the probability density function for a die is always $\frac{1}{6}$ for any number from $1$ through $6$). We can always find a uniform probability distribution over finite intervals. For instance, for $[1,3]$, we must have $p(x)=C$ such that $$\int_1^3 C\ dx = 1 \Rightarrow 3C-C=1 \Rightarrow C=\frac{1}{2}$$ But, for infinite intervals, this will give us an indeterminate value. If we want a uniform probability distribution over the reals, we need $$\int_{-\infty}^{\infty}C \ dx = 1$$ for some real $C$. But, you see, $$\int_{-\infty}^{\infty}C \ dx = \lim_{b \to \infty}\int_{-b}^{b}C\ dx = \lim_{b\to \infty} 2Cb$$ which is an indeterminate limit! There's no way $2Cb=1$ for any $C$! The same problem presents if we attempt to find a uniform probability distribution over $\mathbb{R}^2$, which is what's being assumed in this problem. We need $$\int_{-\infty}^{\infty}\int_{-\infty}^{\infty}C\ dx=1$$ but writing out the limits gives us $$\lim_{b\to \infty}\int_{-b}^{b}\int_{-b}^{b}C\ dx = \lim_{b\to \infty} 4Cb^2$$ which is also indeterminate, so it can never be equal to $1$ for any real $C$!! That's why this problem is malformed, as the distribution you would default to doesn't exist, and can't be answered unless you ask for the real probability distribution over the plane.

Debalanced
  • 1,304
  • 2
  • 15