Back to blog
js puzzleprobabilitygeometry

Planetary Parade

·5 min read

Pyrknot Planet Visibility Problem

An alien on planet Pyrknot (a perfect sphere of radius RR) observes six planets that appear nightly in uniformly random locations across the sky. The alien says:

"If at a given moment there exists somewhere on Pyrknot that all six planets are visible, then from my friend's position on the surface they have an α\alpha probability of also seeing all of the planets."

A tower is then considered. It aggregates visibility: the tower can see every planet that is visible from at least one surface point within distance rr of the tower's base.

Task: In the limit where rRr \ll R, the new observation probability is α+β(r/R)\alpha + \beta \cdot (r/R). Find exact values of α\alpha and β\beta.

Setup and Key Definitions#

Planet Visibility#

Pyrknot is a sphere of radius RR. A planet at direction dS2\mathbf{d} \in S^2 (unit vector from the center) is visible from surface point pS2\mathbf{p} \in S^2 if it is above the local horizon, i.e.:

pd>0\mathbf{p} \cdot \mathbf{d} > 0

The set of surface points from which planet ii is visible is the open hemisphere:

H(di)={pS2:pdi>0}H(\mathbf{d}_i) = \{ \mathbf{p} \in S^2 : \mathbf{p} \cdot \mathbf{d}_i > 0 \}

The All-Visible Region#

The region where all 6 planets are simultaneously visible is:

V=i=16H(di)V = \bigcap_{i=1}^{6} H(\mathbf{d}_i)

The condition "there exists somewhere on Pyrknot that all six planets are visible" is exactly VV \neq \emptyset.

Geometrically, VV \neq \emptyset iff the six direction vectors d1,,d6\mathbf{d}_1, \ldots, \mathbf{d}_6 all lie in some common open hemisphere — since a point p\mathbf{p} sees all 6 iff p\mathbf{p} is in the hemisphere "opposite" to all di\mathbf{d}_i.

Part 1: Computing α\alpha (No Tower)

The problem asks for a conditional probability:

α=P(friend sees all 6pS2:all 6 visible from p)\alpha = P(\text{friend sees all 6} \mid \exists\, \mathbf{p} \in S^2 : \text{all 6 visible from } \mathbf{p})

By Bayes' theorem (since "friend sees all 6" \Rightarrow "someone sees all 6"):

α=P(friend sees all 6)P(V)\alpha = \frac{P(\text{friend sees all 6})}{P(V \neq \emptyset)}

Numerator: P(friend sees all 6)P(\text{friend sees all 6})

The six planets are placed independently and uniformly on S2S^2. For a fixed friend position q\mathbf{q}, each planet independently satisfies:

P(qdi>0)=12P(\mathbf{q} \cdot \mathbf{d}_i > 0) = \frac{1}{2}

by symmetry (half the sphere is above any horizon). By independence:

P(friend sees all 6)=(12)6=164P(\text{friend sees all 6}) = \left(\frac{1}{2}\right)^6 = \frac{1}{64}

Denominator: P(V)P(V \neq \emptyset)

We need the probability that 6 uniform random points on S2S^2 all lie in some common open hemisphere. This is a classical result.

Wendel / Cover–Efron Formula: For nn i.i.d. uniform points on Sd1S^{d-1} (a sphere in Rd\mathbb{R}^d), the probability they all lie in some open hemisphere is:

p(n,d)=12n1k=0d1(n1k)p(n, d) = \frac{1}{2^{n-1}} \sum_{k=0}^{d-1} \binom{n-1}{k}

For our sphere S2S^2 (so d=3d = 3) with n=6n = 6 planets:

p(6,3)=125[(50)+(51)+(52)]=132(1+5+10)=1632=12p(6, 3) = \frac{1}{2^5} \left[ \binom{5}{0} + \binom{5}{1} + \binom{5}{2} \right] = \frac{1}{32}(1 + 5 + 10) = \frac{16}{32} = \frac{1}{2}

So P(V)=12P(V \neq \emptyset) = \dfrac{1}{2}.

Result#

α=1/641/2=132\boxed{\alpha = \frac{1/64}{1/2} = \frac{1}{32}}

Part 2: Computing β\beta (With Tower)

Tower Visibility Condition#

The tower at base position q\mathbf{q} aggregates visibility from all surface points within distance rr of the base. Letting ε=r/R\varepsilon = r/R (the angular radius of the disk), the tower sees planet ii iff:

pD(q,ε):pdi>0\exists\, \mathbf{p} \in D(\mathbf{q}, \varepsilon) : \mathbf{p} \cdot \mathbf{d}_i > 0

where D(q,ε)={pS2:(q,p)<ε}D(\mathbf{q}, \varepsilon) = \{ \mathbf{p} \in S^2 : \angle(\mathbf{q}, \mathbf{p}) < \varepsilon \} is the spherical cap of angular radius ε\varepsilon centered at q\mathbf{q}.

The maximum of pdi\mathbf{p} \cdot \mathbf{d}_i over pD(q,ε)\mathbf{p} \in D(\mathbf{q}, \varepsilon) is achieved at the point in the cap closest to di\mathbf{d}_i. If θ=(q,di)\theta = \angle(\mathbf{q}, \mathbf{d}_i):

maxpD(q,ε)pdi=cos ⁣(max(0,θε))\max_{\mathbf{p} \in D(\mathbf{q}, \varepsilon)} \mathbf{p} \cdot \mathbf{d}_i = \cos\!\bigl(\max(0, \theta - \varepsilon)\bigr)

This is positive iff max(0,θε)<π/2\max(0, \theta - \varepsilon) < \pi/2, i.e., θ<π/2+ε\theta < \pi/2 + \varepsilon, i.e.:

qdi>sin(ε)\mathbf{q} \cdot \mathbf{d}_i > -\sin(\varepsilon)

So: planet ii is visible from the tower iff (q,di)<π/2+ε\angle(\mathbf{q}, \mathbf{d}_i) < \pi/2 + \varepsilon.

Per-Planet Tower Probability (3D Geometry)#

Since di\mathbf{d}_i is uniform on S2S^2, the probability that a uniform point lies within angle π/2+ε\pi/2 + \varepsilon of the fixed point q\mathbf{q} is:

P ⁣((q,d)<π2+ε)=1cos(π/2+ε)2=1+sin(ε)212+ε2P\!\left(\angle(\mathbf{q}, \mathbf{d}) < \frac{\pi}{2} + \varepsilon\right) = \frac{1 - \cos(\pi/2 + \varepsilon)}{2} = \frac{1 + \sin(\varepsilon)}{2} \approx \frac{1}{2} + \frac{\varepsilon}{2}

Note on 3D vs 2D: In a 2D cross-sectional model (circular sky), the analogous formula gives 12+επ\frac{1}{2} + \frac{\varepsilon}{\pi}. This is incorrect for our 3D sphere. The correct coefficient is 12\frac{1}{2}, not 1π\frac{1}{\pi}, because we integrate over the spherical measure sinθdθ\sin\theta\, d\theta, not the circular measure dθd\theta.

All-6 Tower Probability#

Since the 6 planets are independent:

P(tower sees all 6)=i=16P ⁣(qdi>sinε)=(1+sinε2)6P(\text{tower sees all 6}) = \prod_{i=1}^{6} P\!\left(\mathbf{q} \cdot \mathbf{d}_i > -\sin\varepsilon\right) = \left(\frac{1 + \sin\varepsilon}{2}\right)^6

Expanding to first order in ε=r/R\varepsilon = r/R:

(12+ε2)6=164(1+ε)6164(1+6ε)=164+6ε64=164+332rR\left(\frac{1}{2} + \frac{\varepsilon}{2}\right)^6 = \frac{1}{64}(1 + \varepsilon)^6 \approx \frac{1}{64}\left(1 + 6\varepsilon\right) = \frac{1}{64} + \frac{6\varepsilon}{64} = \frac{1}{64} + \frac{3}{32}\cdot\frac{r}{R}

Validity of the First-Order Approximation#

One must verify that the event ¬A\lnot A (no common viewpoint exists) does not contribute at first order. When one planet is in the thin band qd1(ε,0)\mathbf{q} \cdot \mathbf{d}_1 \in (-\varepsilon, 0) and the remaining five are strictly above the equator of q\mathbf{q}, a small tilt of the hemisphere toward d1\mathbf{d}_1 generically includes all six. The probability of the tilt failing (some dj\mathbf{d}_j dropping below the horizon) requires two planets near the equator simultaneously, contributing O(ε2)O(\varepsilon^2). Therefore, the correction from ¬A\lnot A does not affect the first-order term.

Conditional Probability with Tower#

P(tower sees all 6V)164+332rR12=132+316rRP(\text{tower sees all 6} \mid V \neq \emptyset) \approx \frac{\dfrac{1}{64} + \dfrac{3}{32}\cdot\dfrac{r}{R}}{\dfrac{1}{2}} = \frac{1}{32} + \frac{3}{16}\cdot\frac{r}{R}

β=316\boxed{\beta = \frac{3}{16}}

Final Answers#

α=132,β=316\alpha = \frac{1}{32}, \qquad \beta = \frac{3}{16}

Summary of Errors in the 2D Approach#

Issue2D ApproachCorrect 3D Approach
Missing conditionalRead off single-planet fraction as α\alphaDivide by P(V)=1/2P(V \neq \emptyset) = 1/2
Sky geometryCircular sky → coefficient 1/π1/\piSpherical sky → coefficient 1/21/2
Number of planetsComputed for 1 planetRaise to the 6th power
α\alpha result1/21/2 (wrong)1/321/32
β\beta result1/π1/\pi (wrong)3/163/16

The denominator P(V)=1/2P(V \neq \emptyset) = 1/2 comes from the Wendel formula, a non-trivial combinatorial geometry result. Without it, the conditional probability structure of the problem is missed entirely.