Computational
Morphogenesis

A high-resolution atlas of pattern formation in Turing's reaction-diffusion space

2,304 simulations 48×48 parameter grid 101.8 min compute 9 pattern classes
Simulations Run
2,304
Pattern Regions
508
Theory vs. Reality (r)
−0.244
Pixel Wavelength
~7
// The Premise

In 1952, Alan Turing published his final paper — The Chemical Basis of Morphogenesis — proposing that the patterns on animal coats, seashells, and fish scales emerge from the interplay of two chemicals diffusing at different rates. He called this reaction-diffusion, and he worked out the mathematics on paper, years before computers could simulate it.

In 1993, John Pearson ran the Gray-Scott equations on a 1990s supercomputer and discovered a zoo of patterns no one had predicted: self-replicating spots, growing labyrinths, coral-like chaos, and oscillating worms. His Science paper revealed a system where simple rules produce stunning complexity.

Two chemicals. Two parameters. Infinite beauty.

This project re-creates and extends that exploration using modern hardware: a systematic sweep of 2,304 parameter combinations, each simulated from first principles in NumPy, then analyzed with linear stability theory to test what theory predicts and what actually happens.

⚡
Key Finding: Theory fails. Nature doesn't care. The Turing linear stability criterion predicts pattern wavelengths that show zero correlation with the actual measured wavelengths (r = −0.244). Gray-Scott patterns are not classical Turing patterns — they are nonlinear structures that emerge through mechanisms linear theory cannot capture.
// Featured Patterns
Self-Replicating Spots
Self-Replicating Spots
F=0.0367, k=0.0649
Maze / Labyrinth
Maze / Labyrinth
F=0.0290, k=0.0570
Coral / Chaos
Coral / Chaos
F=0.0200, k=0.0540

View Full Gallery →

// Key Findings
  • 01
    Gray-Scott is NOT a classical Turing system The trivial steady state (1, 0) is linearly stable everywhere — its Jacobian is diagonal. Patterns cannot arise from linear instability of the uniform state. They are finite-amplitude, nonlinear phenomena.
  • 02
    Linear theory fails to predict wavelengths Where nontrivial steady states exist and are Turing-unstable, the predicted critical wavelength λ_c shows essentially no correlation with measured wavelengths (Pearson r = −0.244, n = 180).
  • 03
    The saddle-node boundary organizes everything The analytic bifurcation curve k_c = √F/4 − F precisely bounds the pattern-forming region. Inside the curve: complex patterns. Outside: death or saturation. The transition is pixel-sharp.
  • 04
    Complexity peaks far from the boundary Shannon entropy of the pattern field increases with distance from the saddle-node bifurcation, contradicting the "critical slowing down" expectation. The richest patterns live deep in the pattern region, not at its edge.
// The Phase Atlas

Below: Shannon entropy of the concentration field across the full (F, k) parameter space. Bright regions are maximally complex patterns. Dark regions are uniform states. The white dashed curve is the analytically-derived saddle-node bifurcation — note how it precisely traces the phase boundary.

Entropy heatmap with theoretical bifurcation overlays
Figure: Shannon entropy heatmap with saddle-node bifurcation (white line) and Turing instability region (cyan dots) overlaid.
// Explore