14

The vertices of a triangle are three unifomly random points on a unit circle. The side lengths are, in random order, $a,b,c$.

Show that $P(ab>c)=\frac12$.

The result is strongly suggested by simulations, and by my attempt shown below.

The simplicity of the result suggests that there may be an intuitive explanation. I am hoping for an intuitive explanation, but if that's not possible then any answer is welcome. (Examples of intuitive explanations are here and here.)

My attempt

Assume that the circle is centred at the origin, and the vertices of the triangle are:
$A(\cos(-2Y),\sin(-2Y))$ where $0\le Y\le\pi$
$B(\cos(2X),\sin(2X))$ where $0\le X\le\pi$
$C(1,0)$

Let:
$a=BC=2\sin X$
$b=AC=2\sin Y$
$c=AB=\left|2\sin\left(\frac{2\pi-2X-2Y}{2}\right)\right|=|2\sin(X+Y)|$

$P\left[ab>c\right]=P\left[2(\sin X)(\sin Y)>|\sin(X+Y)|\right]$

This probability is the ratio of the area of the shaded region to the area of the square in the graph below.

enter image description here

Rotate these regions $45^\circ$ clockwise about the origin, then shrink by factor $\frac{1}{\sqrt2}$, by letting $X=x-y$ and $Y=x+y$.

enter image description here

Using symmetry, we only need to consider the left half of the blue "diamond". Note that in the left half, $0<x<\pi/2$, so $|\sin(2x)|=\sin(2x)$.

$P\left[2(\sin X)(\sin Y)>|\sin(X+Y)|\right]$

$=P\left[2(\sin (x-y))(\sin (x+y))>\sin(2x)\right]$

$=P\left[(\sin^2x)(\cos^2y)-(\cos^2 x)(\sin^2 y)>(\cos x)(\sin x)\right]$

$=P\left[(\sin^2x)(\cos^2y)-(\cos^2 x)(1-\cos^2 y)>(\cos x)(\sin x)\right]$

$=P\left[\cos^2 y>\cos^2 x+(\cos x)(\sin x)\right]$

$=P\left[ -f(x)<y<f(x)\right]$

where $\color{red}{f(x)=\arccos\left((\cos x)\sqrt{1+\tan x}\right)}$.

Noting that $f\left(\frac{\pi}{4}\right)=0$, the probability is

$$\dfrac{\int_{\pi/4}^{\pi/2}f(x)\mathrm dx}{\frac12\left(\frac{\pi}{2}\right)^2}$$

Numerical evidence suggests that $\int_{\pi/4}^{\pi/2}f(x)\mathrm dx=\frac{\pi^2}{16}$, but I don't know how to prove this.

If that's true, then the probability is indeed $1/2$.

Context

This question was inspired by a question, "If $(a,b,c)$ are the sides of a triangle, what is the probability that $ac>b^2$?"

Edit

I found a generalization, which may or may not help in finding an intuitive explanation.

For $k\in\mathbb{R^+}$, we have

$$P(ab<kc)=\frac{2}{\pi}\arctan k$$

or equivalently,

$$P\left(\frac{ab}{c}<k\right)=\frac{\arctan k}{\frac{\pi}{2}}$$

This can be proved using the method in "My attempt" above, which gives

$$P(ab>kc)=\dfrac{\int_{\arctan k}^{\pi/2}\arccos\left((\cos x)\sqrt{1+k\tan x}\right)\mathrm dx}{\frac12\left(\frac{\pi}{2}\right)^2}$$

The integral in the numerator can be evaluated, using the method in @Zacky's answer, as $\frac{\pi}{4}\arctan\frac1k$. So $P(ab<kc)=1-P(ab>kc)=1-\frac{2}{\pi}\arctan\frac1k=\frac{2}{\pi}\arctan k$.

Dan
  • 22,158
  • 6
    The radius of your circle should be part of the question - the result for the unit circle will not be the same as for circles of other radius, because $ab>c$ is not dimensionally homogenous. For example, if the circle has radius 100, $ab$ will almost certainly be greater than $c$. – mcd Mar 10 '24 at 14:44
  • @mcd Good point. I have edited. – Dan Mar 10 '24 at 14:50
  • 2
    A different form of $2\sin X\sin Y>|\sin(X+Y)|$ is $2>|\tan Y+1/\tan X|$. Don't know if it would help. I got it by expanding the RHS and dividing by the LHS. – Al.G. Mar 10 '24 at 15:39
  • 2
    It looks like if you fix $A=(1,0)$ and $p\in[0,2\pi)$, and then randomly choose $B$ to be one of either $(\cos(p),\sin(p))$ or $(\cos(\pi-p),\sin(\pi-p))$, and then set $c=AB$ and then randomly choose the last point, the probability is still $1/2$. This is stronger than the given claim (i.e. if you could show this, then you get an answer to the question). – Varun Vejalla Mar 11 '24 at 01:51
  • It is clear enough that @mcd's comment should lead to $P(ab>cR)$ where $R$ is the radius of the circle we start from, and the circumradius of the triangle. Now we are comparing quantities of the same "physical dimension". But can it lead to an intuitive geometric interpretation? – Jeppe Stig Nielsen Mar 17 '24 at 20:31

1 Answers1

11

Here's a way to finish the last integral making use of its symmetry.

$$I=\int_\frac{\pi}{2}^\frac{\pi}{4} \arccos\left(\cos x \sqrt{1+\tan x}\right)dx\overset{\cot x\to x}=\int_0^1 \frac{\arccos \sqrt{\frac{x(1+x)}{1+x^2}}}{1+x^2}dx$$

$$=\int_0^1 \frac{\arctan \sqrt{\frac{1}{x}\frac{1-x}{1+x}}}{1+x^2} dx\overset{\large \frac{1-x}{1+x}\to x}=\int_0^1 \frac{\operatorname{arccot} \sqrt{\frac{1}{x}\frac{1-x}{1+x}}}{1+x^2}dx$$

$$\Rightarrow 2I=\frac{\pi}{2}\int_0^1 \frac{1}{1+x^2}dx\Rightarrow \boxed{I=\frac{\pi^2}{16}}$$

Above it was utilized that $\, \arccos x =\arctan \left(\frac{\sqrt{1-x^2}}{x}\right)$ and $\arctan x+\operatorname{arccot} x=\frac{\pi}{2}$.

Zacky
  • 27,674