The "north pole problem" and random orthogonal matrices

TL;DR

This paper investigates how applying random orthogonal matrices to a point on the sphere affects its distribution, revealing that certain transformed points tend to stay near the poles more than expected, with exact distribution formulas derived.

Contribution

It provides the first exact distributional results for the components of points obtained by powers of Haar-distributed orthogonal matrices acting on the sphere.

Findings

Points after multiple rotations are more likely to be near the poles.

Derived explicit distributions for the components of transformed points.

Confirmed simulation observations with rigorous proofs.

Abstract

This paper is motivated by the following observation. Take a 3 x 3 random (Haar distributed) orthogonal matrix $\Gamma$, and use it to "rotate" the north pole, $x_0$ say, on the unit sphere in $R^3$. This then gives a point $u=\Gamma x_0$ that is uniformly distributed on the unit sphere. Now use the same orthogonal matrix to transform u, giving $v=\Gamma u=\Gamma^2 x_0$. Simulations reported in Marzetta et al (2002) suggest that v is more likely to be in the northern hemisphere than in the southern hemisphere, and, morever, that $w=\Gamma^3 x_0$ has higher probability of being closer to the poles $\pm x_0$ than the uniformly distributed point u. In this paper we prove these results, in the general setting of dimension $p\ge 3$, by deriving the exact distributions of the relevant components of u and v. The essential questions answered are the following. Let x be any fixed point on the unit sphere in $R^p$, where $p\ge 3$. What are the distributions of $U_2=x'\Gamma^2 x$ and $U_3=x'\Gamma^3 x$? It is clear by orthogonal invariance that these distribution do not depend on x, so that we can, without loss of generality, take x to be $x_0=(1,0,...,0)'\in R^p$. Call this the "north pole". Then $x_0'\Gamma^ k x_0$ is the first component of the vector $\Gamma^k x_0$. We derive stochastic representations for the exact distributions of $U_2$ and $U_3$ in terms of random variables with known distributions.