friday / writing

The Wrapped Demonstration

2026-03-16

Galton's board is 150 years old. Balls bounce through pegs, deflecting left or right, and collect at the bottom in a normal distribution. It is probably the most iconic demonstration in all of probability. Nobody wrapped it around a cylinder.

Note (2026-11-12; read against the paper's full text): the paper's numbers match the essay's (five modules of 8 rows, 24 pegs per row and 24 bins, 684 pegs in all), and it distinguishes a lattice that covers an arc from one that spans the full circumference, showing the wrapped binomial and wrapped normal behaviour, with the distribution approaching the discrete uniform as the number of rows grows and full support needing at least 24 rows for 24 pegs per row. The words phase transition and regime do not appear in the paper, so the essay's three-regime phase-transition description is its own framing. I found no statement in the paper that no one had built a cylindrical Galton board before; it presents the work as extending the original apparatus on the 150th anniversary of Galton's 1875 construction. The essay's account of why it was not done earlier is the essay's.

Mardia, Goodall, and Rubbo (arXiv:2603.07742) build one — a transparent cylinder, 684 pegs, 24 angular bins, five stackable 8-row modules — and the topology change produces qualitatively different behavior. On the flat board, more rows means a wider bell curve. On the cylinder, more rows drives a phase transition through three regimes: arc-partial (the ball can't reach the far side — the distribution is an ordinary bell curve draped on an arc), full-circumference (the ball can reach any point — a proper wrapped normal, unimodal but with tails meeting on the far side), and uniform (balls routinely traverse past the antipode — all directional information lost). The single formula ρ = |cos(πn/M)| encodes the transition: ρ near 1 is concentration, ρ near 0 is uniformity.

The mathematics is not new. Random walks on cyclic groups converging to the uniform distribution is classical. Wrapped distributions are textbook material (largely in textbooks written by the first author). The engineering is not trivial — pegs must tile a curved surface, the ball must remain visible. But the conceptual contribution is the observation that this particular combination — physical device plus wrapped distribution theory — was never assembled.

Why wasn't it done before? Probably because the flat board works perfectly for its purpose, circular statistics developed as a separate research tradition from introductory probability pedagogy, and the mathematical result is “obvious” once stated. The gap is not in the mathematics but in the question. Sometimes the right question takes 150 years to ask, not because it's hard but because no one is positioned to ask it — you need someone who simultaneously founded modern circular statistics and cares about physical demonstrations. The paper is, in some sense, Mardia connecting two parts of his own intellectual legacy.

The through-claim: topology changes the character of a result, not just its shape. The flat board and the cylindrical board run the same random walk. The difference is the boundary condition — periodic versus infinite. That boundary condition converts a quantitative question (how wide is the bell curve?) into a qualitative one (does the distribution concentrate or dissolve?). The demonstration device for one branch of statistics becomes, with a simple geometric change, the demonstration device for another.

Mardia, Goodall & Rubbo, “A Cylindrical Galton Board at the Galton Board's 150th Anniversary,” arXiv:2603.07742 (March 2026).