5

Definition of happy triangle

On a unit circle, choose three uniformly random independent points $A,B,C$.
Inside the circle, choose a uniformly random independent point $D$.
Triangle $ABC$ is called "happy" if it contains $D$.

Conjecture

Is the following conjecture true: In a happy triangle, the expected product of the three side lengths is $\pi$.

I simulated $2\times10^7$ happy triangles, and the average product was $3.14184188\approx1.00007933\pi$.

My thoughts

Letting $X=\text{product of the three side lengths}$, I tried (but failed) to find the probability distribution function $P(X<x)$ so that I could find the probability density function and thus $E(X)$.

The expected product of the three side lengths is not the cube of the expected side length, because the three side lengths are not independent random variables. (A simulation suggests that the expected side length is approximately $1.509$.)

Other elegant properties?

A happy triangle seems to have many elegant properties.

Dan
  • 41,295
  • 1
    Why are the three side lengths independent in the "not necessarily happy" case? I would have thought if you knew two sides then the third is restricted to at most two possible values. – Henry Jul 26 '25 at 00:33
  • @Henry Good point, I have edited. – Dan Jul 26 '25 at 00:50
  • 1
    The product of the three side lengths is 4 times the area of the triangle (as the triangle is inscribed in a circle of radius 1). – Catalin Zara Jul 26 '25 at 01:26
  • 1
    I don’t know about happy. But the most confident triangles are those that have one angle equal to 90 degrees. That is because they know they are right. – Michael Jul 27 '25 at 20:32

2 Answers2

7

Given a random triangle $ABC$ with $A,B,C\in S^1$, the probability that $D$ lies inside $ABC$ is $\frac{[ABC]}{\pi}=\frac{abc}{4\pi}$, since $abc=4R\Delta$ by Euler's formula and $R=1$. It follows that the conditional expected value

$$\mathbb{E}[abc | D\in ABC] $$

just depends on the average area of the triangles we may build around $D$ (meaning: having $D$ as an interior point), with $D$ being uniformly distributed over the unit ball. Such area can be written as $2R^2\sin A\sin B\sin C$, which equals $2\sin A\sin B\sin C$.

As a continuation of Thomas Andrews' answer,

$$ 64\iint_{[0,\pi]^2}\sin^2(u-v)\sin^2(u)\sin^2(v)\,du\,dv = 6\pi^2 \tag{N}$$

while

$$ \int_{0}^{2\pi}\int_{0}^{2\pi}2\left|\sin\frac{x-y}{2}\sin\frac{x}{2}\sin\frac{y}{2}\right|\,dx\,dy = 8\iint_{[0,\pi]^2}\left|\sin(u-v)\sin(u)\sin(v)\right|\,du\,dv=6\pi\tag{D}$$

since $\sin(u)\sin(v)=\frac{1}{2}\left(\cos(u-v)-\cos(u+v)\right)$ and

$$ \left|\sin x\right| = \frac{2}{\pi}-\frac{4}{\pi}\sum_{n\geq 1}\frac{\cos(2nx)}{4n^2-1}.$$

This proves that the wanted expected value is indeed $\pi$.

Jack D'Aurizio
  • 372,937
  • 42
  • 421
  • 888
  • You wrote, "With $A$ given by the antipode of $A$", I think you mean "With $A'$ given by the antipode of $A$". About $\mathbb{E}[[ABC]]=\cdots=\mathbb{E}[[BAA']]$: The average area of a triangle inscribed in a unit circle is $\frac{3}{2\pi}$ (source), so $\mathbb{E}[[ABC]]=\frac{3}{2\pi}$, but $\mathbb{E}[[BAA']]$ should equal the average value of $\sin\theta$ for $0\le\theta<\pi$, which is $\frac{2}{\pi}$. What am I missing? – Dan Jul 26 '25 at 21:44
  • @Dan: $AA'$ fixed. The distance of $B$ from $AA'$ is not uniformly distributed, from which the discrepancy. – Jack D'Aurizio Jul 27 '25 at 00:55
  • I agree that the (shortest) distance from $B$ to $AA'$ is not uniformly distributed. It has the same distribution as $\sin\theta$ where $\theta$ is uniformly distributed in $[0,\pi)$. The area of triangle $BAA'$ equals the distance from $B$ to $AA'$, so the average area of triangle $BAA'$ should equal the average value of $\sin\theta$ where $\theta$ is uniformly distributed in $[0,\pi)$, which is $\frac{2}{\pi}$ (not $\frac{3}{2\pi}$). – Dan Jul 27 '25 at 01:41
  • @Dan: you're right. Then, what's wrong in $\mathbb{E}[[ABC]]=\frac{1}{2}\mathbb{E}[[ABC]+[A'BC]]=\frac{1}{2}\mathbb{E}[BA\cdot BA']=\mathbb{E}[[BAA']]$? – Jack D'Aurizio Jul 27 '25 at 15:15
  • The first and third equal signs are right, so the second equal sign must be wrong. (I don't know why you thought the second equal sign was right, so I can't pinpoint the mistake in reasoning.) – Dan Jul 27 '25 at 22:33
  • @Dan: the quadrilateral $ABA'C$ has two right angles at $B$ and $C$, so its area equals $\frac{1}{2}AB\cdot BA'+\frac{1}{2}AC\cdot CA'$, and I cannot believe that $\mathbb{E}[BA\cdot BA']$ and $\mathbb{E}[CA\cdot CA']$ are different things! – Jack D'Aurizio Jul 27 '25 at 23:04
  • I think you are assuming that $B$ and $C$ are on opposite sides of the diameter $\overline{AA'}$. – Dan Jul 27 '25 at 23:25
  • 1
    @Dan: correct. Thank you, that was driving me crazy ;P – Jack D'Aurizio Jul 27 '25 at 23:32
  • If $B$ and $C$ are on opposite sides of diameter $\overline{AA'}$, that doesn't necessarily mean that $\triangle ABC$ is acute. This shows that the average area of an acute triangle is $3$ times the average area of an obtuse triangle. – Dan Jul 28 '25 at 02:31
6

Not a complete answer, just a start.

Always choose one of the vectors as $1$ on the complex plane. You want to choose two more $w,z$ with $w=e^{ix},z=e^{iy}.$ Then the product of the edge lengths is:

$$p(x,y)=8|\sin((x-y)/2)\sin(x/2)\sin(y/2)|.$$

The probability of $D$ being inside the triangle is $\frac A\pi$ where $A=A(x,y)$ is the area of the triangle.

So you need a formula for $A(x,y)$ and then you want the integral:

$$\dfrac{\int_0^{2\pi}\int_0^{2\pi}A(x,y)p(x,y)\,dy\,dx}{\int_0^{2\pi}\int_0^{2\pi}{A(x,y)}\,dy\,dx}$$

You get: $$A(x,y)=\frac12|\sin((x-y)/2)(1-w)(1-z)|=2|\sin((x-y)/2)\sin(x/2)\sin(y/2)|.$$ This would indicate that if the area of a triangle inscribed in a unit circle is $\frac1{4}$ times the product of the lengths of the sides.

$$16\int_0^{2\pi}\int_0^{2\pi}\sin^2((x-y)/2)\sin^2(x/2)\sin^2(y/2)\,dy\,dx$$

Letting $x=2u,y=2v,$ the numerator becomes:

$$64\int_0^\pi\int_0^\pi\sin^2(u-v)\sin^2(u)\sin^2(v)\,dv\,du$$

That shouldn't be hard to integrate.

The denominator is harder, because of the absolute values.

Thomas Andrews
  • 188,543
  • Where you wrote $\dfrac{A(x,y)}{\pi}$, I think it should be $\dfrac{A(x,y)}{\frac{3}{2\pi}}$, because $\frac{3}{2\pi}$ is the average area of a triangle inscribed in the unit circle. (You're using probability but you should use probability density.) Then your final integral will equal $\pi$, which confirms my conjecture. – Dan Jul 26 '25 at 02:19
  • No, if you pick a uniform point inside the circle, the total are of the circle is $\pi,$ and the probability it is in a subset of area $A$ is $A/\pi.$ The average size of the triangle doesn't matter. @dan – Thomas Andrews Jul 26 '25 at 03:32
  • Oh, I see the reference to $1/16.$ Fixed, thanks @dan. – Thomas Andrews Jul 26 '25 at 03:38
  • And in my previous comment, it should be "total area." – Thomas Andrews Jul 26 '25 at 03:39
  • Your last double integral looks like it equals $\frac{3}{2\pi}$, but shouldn't it be $\pi$? – Dan Jul 26 '25 at 03:41
  • Oh, I forgot to do a division, I think. I need to divide by: $$\frac1{4\pi^2}\int\int \frac{A(x,y)}\pi,dy,dx$$ Let me think about it. – Thomas Andrews Jul 26 '25 at 03:49
  • I think I've updated it properly. The denominator is harder. – Thomas Andrews Jul 26 '25 at 03:56
  • 1
    Yes, in your latest edit you are essentially using $\dfrac{A(x,y)}{\frac{3}{2\pi}}$ instead of $\dfrac{A(x,y)}{\pi}$, as I mentioned in my first comment. You wrote "the denominator is harder..." but it is just the average area of a triangle inscribed in a unit circle, muliplied by $4\pi^2$, i.e. $6\pi$. Then the whole thing equals $\pi$, as conjectured. – Dan Jul 26 '25 at 04:45