Turing's other great idea. Two chemicals that sit perfectly still when stirred together break into stripes and spots the moment they are allowed to diffuse — the maths of leopard coats and zebra flanks. Diffusion, which everywhere else smooths, here creates structure. Down the center the state flows: the well-mixed system goes in, the dispersion engine tests every spatial wavelength, the pattern verdict comes out. Blue builds it; red breaks it.
source A. M. Turing, The Chemical Basis of Morphogenesis (1952), Phil. Trans. R. Soc. B 237(641), 37–72 — doi.org/10.1098/rstb.1952.0012. Rendered, not quoted.
Two morphogens on a line: activator u (makes more of itself and its inhibitor), inhibitor v (suppresses the activator). Near the uniform steady state the reaction is the Jacobian J:
| ∂/∂u | ∂/∂v | |
|---|---|---|
| ∂u/∂t | +1.0 | -1.0 |
| ∂v/∂t | +2.0 | -1.5 |
Add diffusion: ∂u/∂t = f + Du∇²u, ∂v/∂t = g + Dv∇²v. A mode of wavenumber k grows at rate λ(k), the top eigenvalue of J−k²diag(Du,Dv). AMBER the J entries are illustrative kinetics, chosen stable-when-mixed; the flip is exact.
Well-mixed, the state is a settled premise: perturb it and it returns to uniform. Let diffusion act and that same premise forces a new conclusion — a wavelength. Each sphere is the next one's premise.
The move Turing needed is long-range inhibition, short-range activation: the inhibitor must out-diffuse the activator. The same author who made reasoning computable in the frame of the syllogism made form computable here — logic of the mind, then logic of the body.
The blue team's live check: re-derive the dispersion relation on the canonical system and confirm the four facts — stable mixed, unstable with diffusion, dead when the ratio reverses, dead when equal. If red tampers, this badge is where it shows.
Feed the engine the homogeneous steady state where f = g = 0 — the tissue is a uniform bath, no spots anywhere — together with the reaction Jacobian J and two diffusion constants. The well-mixed verdict must be STABLE: trace(J) < 0 and det(J) > 0, so both reaction eigenvalues have negative real part and any stir relaxes back to flat.
Nothing here is a pattern yet. That is the whole surprise below — the pattern is not fed in; diffusion manufactures it from a state that was provably calm.
Turing threshold for this J: Dvfu + Dugv > 0 — the inhibitor must out-diffuse the activator.
Curve: growth rate λ(k) over spatial wavenumber k — computed live, never looked up. It sits below zero at k=0 (well-mixed stays calm) and lifts into a positive band at nonzero k when the inhibitor out-diffuses the activator. The peak is the fastest-growing mode; its k sets the stripe spacing shown beneath.
What the machine proves: the same state that is STABLE when mixed becomes UNSTABLE once it diffuses — but only for a finite band of wavenumbers, and only when Dinhibitor > Dactivator. The dispersion relation peaks at a nonzero k*, and that peak fixes a real pattern wavelength = 2π/k*. Diffusion, the great smoother, is the thing that breaks the symmetry.
The blue witness (left) confirms the flip live; the red team (right) tries to keep the "pattern" claim while quietly killing the instability.
And the biology fights back: a robust Turing mechanism needs Dv/Du large, yet most real morphogen pairs diffuse at similar rates — few clean animal examples are confirmed (fish stripes, digit spacing are the strong cases). Rival mechanisms — positional-information "French flag", mechanochemical folding — make many of the same patterns without any diffusion-driven instability. Turing is a route to form, not the route.
"Diffusion always smooths things out." Cut. True for one species; couple two at different rates and diffusion drives the instability. That paradox is the whole paper.
"Any two reacting, diffusing chemicals can stripe." Cut. Only if the inhibitor out-diffuses the activator (and the kinetics are right). Equal D — no pattern, proven in window 6.
"Turing patterns explain animal coats, full stop." Kept, corrected. A compelling, partly-confirmed model — strong for fish and digits, contested elsewhere; not a closed case.
The red move: set the two diffusion constants equal (kill the ratio) while still flashing "pattern forming". With Dv = Du the band collapses below zero — no mode grows — yet the claim stands. The blue witness (window 7) recomputes the dispersion relation and catches it.
Equalise the diffusion and the diffusion-driven instability vanishes — the state stays uniform. The witness disagrees with the known flip and turns red. Nothing is faked; the attack is real and it is caught.