Reaction-Diffusion Relief (3D)

The same Gray-Scott field as a raised relief in a Petri dish, or wrapped on a torus.

Switch to 2D
The science

What is modelled

The Gray-Scott model: two chemicals U and V on a square grid with periodic edges. U is fed in at rate F; V is produced by the autocatalytic step U + 2V → 3V and removed at rate F+k. U and V diffuse at different speeds (Du > Dv), and that difference is what lets stripes, spots and mazes appear from a nearly uniform start (Turing, 1952).

Equations

  • ∂u/∂t = Du∇²u - u v² + F(1 - u)
  • ∂v/∂t = Dv∇²v + u v² - (F + k) v
  • Numerics: 128×128 grid, spacing 1, explicit Euler with dt = 1, 5-point Laplacian (stable while 4 Du < 1).
  • Homogeneous states: u v = F+k and v = [1 ± √(1 - 4(F+k)²/F)] F / (2(F+k)), with u = 1 - (F+k) v / F. They exist only when 4(F+k)² ≤ F (white line on the map).
  • Linear stability: with Jacobian J = [[-v²-F, -2(F+k)], [v², F+k]], a Fourier mode eiqx grows at σ(q) = max Re eig(J - q² diag(Du, Dv)). A Turing instability needs σ(0) < 0 (stable to uniform change) but σ(q) > 0 for some q; the fastest mode has wavelength λ = 2π/qmax.

Checks run in this page

  • Turing wavelength: start at the homogeneous state plus 1e-3 noise, run until the V amplitude is 0.015 (still linear), take the power-weighted centroid of the 2D spectrum (own FFT) and compare with qmax from the Jacobian above. At the default F=0.055, k=0.062 theory gives q = 0.529 (wavelength 11.9 cells) and the grid gives about 0.515, within 3 %.
  • Integrator: a diffusion-only sine wave must decay as (1 - D(2 - 2cos q))T (exact for explicit Euler) and, to lattice accuracy, as e-D q² T.
  • The map and the two graphs use the analytic formulas only; they never read the grid.

Honest simplifications

  • Most famous Gray-Scott patterns (spots that divide, mazes, worms) arise from a finite-amplitude instability of the trivial state u=1, v=0 and are not linear Turing patterns; linear theory only describes the shaded yellow region of the map. Outside it the wavelength readout is measured, not predicted.
  • Diffusivities and rates are dimensionless lattice units, not measured chemistry. Real Turing patterns in tissue or chemical gels need much more detail.
  • Explicit Euler at dt = 1 is accurate enough for pattern shape but not for quantitative transient timing; the integrator check shows its diffusion error.
  • The 3D view maps the same flat periodic field as a colour/height texture. On the torus the pattern is stretched on the inner ring by about 2.5 times because the true curved-surface Laplacian is not solved.
  • Pseudo-random seeds make every preset repeatable.

The dish, plate, torus and pedestal are Blender-built meshes; the chemistry is solved live in your browser and mapped onto them as vertex colour and height.

Loading model...
Loading Blender model...
Measured spectrum of V vs linear theory
Dispersion relation sigma(q) from the Jacobian
Parameter map: yellow = Turing-unstable (click to set F, k)
Made with 🟠 Blender · Three.js