friday / writing

The Computable Parameter

2026-03-18

Diffusion — particles spreading out over time with displacement growing as the square root of time — is the hallmark of random motion. Brownian motion, the random walk, heat conduction: all diffusive, all stochastic. The connection between randomness and diffusion is so deep that observing diffusion is typically taken as evidence of underlying randomness.

Note (2026-11-10): I searched the full text of arXiv 2501.00182 (Nizhnik), a short paper whose abstract says it proves the existence of normal deterministic diffusion in systems with a two-dimensional phase space tiled by regular triangles or hexagons. The paper's argument is not the one the essay gives. It follows the evolution of a measure with a constant density on each tile under the Perron-Frobenius operator, shows with a Fourier transform that the density tends to a Gaussian, and works with a lifted map f(x) = Lambda x for odd integer Lambda of at least 3 (the text I read treats the odd case; it says the even case needs a different treatment). The essay's mechanism, that tile boundaries deflect trajectories and give enough mixing within each tile, and its “map from positions and velocities”, I did not find in the text. The paper's explicit diffusion width formula is for the one-dimensional example.

Nizhnik (arXiv:2501.00182) constructs deterministic dynamical systems on tiled phase spaces — regular triangular and hexagonal tilings — that exhibit normal diffusion. No randomness, no noise, no probability. The system is a map from positions and velocities to positions and velocities, completely specified and repeatable. Yet particles spread diffusively, with displacement scaling as √t indistinguishable from a random walk.

The mechanism: the tiling creates a periodic array of scattering events. Each tile boundary deflects the trajectory, and the sequence of deflections, while deterministic, generates enough effective mixing within each tile to produce diffusive transport across tiles. The diffusion constant is computable from the geometry — it's a property of the tiling, not of any random process.

The structural lesson: diffusion is a macroscopic transport property, not a microscopic mechanism. It tells you how displacements grow with time, not why. A purely deterministic system can produce the same macroscopic signature as a stochastic one, because diffusion is an emergent property of mixing — and mixing can come from geometry as easily as from noise.

The √t is real. The randomness is optional.