Work in progress. The interactive demo runs; the writeup below sketches the core ideas. A full treatment lives in the companion PDF.


The Problem of Noise

Every physical system that drifts, diffuses, or fluctuates is governed — at the density level — by a Fokker-Planck equation. The standard story writes particle dynamics as

dx(t) = -∇V(x) dt + σ dW(t)

where W is Brownian motion: structureless, memoryless, Gaussian. The density p(t,x) satisfies a PDE whose stochastic character enters solely through the driving signal.

If the randomness enters only through a signal, then the signal need not be random at all. It need only look random in the right statistical sense.


The Central Proposal

Replace the Brownian realisation with a signal generated by iterating f_c(z) = z^d + c on its Julia set. Three pillars support this:

Pillar 1 — Almost-Sure Invariance Principle. Birkhoff sums of Hölder observables along hyperbolic rational maps approximate Brownian motion with error O(n^(1/2 - ε)) (Dupont 2010). The orbit is deterministic; the statistics are Brownian.

Pillar 2 — Rough Path Homogenization. Chaotic fast-slow systems converge to SDEs in rough path topology, with a computable Lévy area correction (Kelly-Melbourne 2016-17). The rescaled process converges to an Itô diffusion with effective diffusion coefficient given by the Green-Kubo formula.

Pillar 3 — Orthogonal Polynomials on Julia Sets. The Bessis-Geronimo-Moussa construction provides a canonical orthogonal polynomial family on the equilibrium measure — a natural chaos expansion basis intrinsic to the deterministic system.


Live Demo

Three Julia sets provide six independent noise channels (Re/Im each), driving a three-body gravitational system. Switch tabs to see the different layers of the framework.

DETERMINISTIC NOISE EXPLORER n = 3 product construction
Progressive Nebulabrot (3 depths: 60/250/1000). Three n=3 parameters pulsing. The Buddhabrot = what deterministic dynamics look like after you forget which Julia set.
Information Hierarchy
Moduli point c
Complete: J(fc), μc, (an,bn), σ², λ, all statistics
Spectral fingerprint (BGM)
Jacobi sequence (an,bn) → all moments of μc, fractal geometry
Green–Kubo + autocorrelation
σ²(c) and decay rate λ(c) — two structural parameters
Fokker–Planck (σ²-matched)
Leading-order density evolution, correct to O(ε)
Brownian motion
σ² only — no geometry, no memory, no spectral structure

The Buddhabrot as Projection of Moduli

Each Julia set J(f_c) is a point in the moduli space of noise — the family parametrising deterministic noise models whose diffusion coefficients vary real-analytically within hyperbolic components.

The Buddhabrot arises as the fibre integral B(z) = ∫ O_c(z) dλ(c) — the pushforward of the orbit flow across moduli. Aggregating over c erases the moduli information.

The Buddhabrot shows what deterministic dynamics look like after moduli information is erased. Julia sets show that the erased information was geometrically real.


Information Hierarchy

  • Brownian motion — σ² only.
  • Fokker-Planck (σ²-matched) — Leading-order density evolution.
  • Green-Kubo + autocorrelation — σ²(c) and decay rate λ(c).
  • Spectral fingerprint (BGM) — Full Jacobi sequence encoding all moments of μ_c and fractal geometry.
  • Moduli point c — Complete specification.

Known Obstacles

  1. The Lévy area is generically nonzero — the limiting SDE is Marcus-type for vector-valued observables.
  2. CLT failure at the Mandelbrot boundary — saving grace: Collet-Eckmann parameters carry full harmonic measure (Smirnov 2000).
  3. Singular measures kill exponential polynomial chaos convergence — capped at algebraic rates.
  4. Dimensional ceiling — one Julia set gives at most 2 real signals; need ⌈d/2⌉ sets.
  5. Computational overhead — potentially ~600x more efficient than 1000-trajectory Monte Carlo, but rough path scheme adds complexity.

Obstacles 1, 4, and 5 are essentially resolved for hyperbolic parameters. Obstacles 2 and 3 remain harder.


The orbit is deterministic. The statistics are Brownian. The structure is fractal. Somewhere in the liminal space between chaos and randomness, these limits meet.