The Chemical Basis of Morphogenesis Alan Turing · reaction–diffusion · the mathematics of pattern formation

Alan Turing, The Chemical Basis of Morphogenesis

Alan Mathison Turing, 1912–1954 · morphogenesis paper published 1952

In the last major paper he published, Alan Turing asked how a uniform, featureless mass of cells could spontaneously develop structure, spots, stripes, whorls, the arrangement of an embryo's parts. He called the origin of biological form morphogenesis, and he gave it a mathematical account so unexpected it took decades to be confirmed in the laboratory.

His answer was diffusion-driven instability. Two chemicals, Turing called them morphogens, react with each other while spreading through tissue. Intuition says diffusion should smooth differences away. Turing proved the opposite can happen: if one morphogen diffuses faster than the other, diffusion can destabilise the uniform state and drive it into a stable spatial pattern. That counterintuitive result is the heart of what the world now calls a Turing pattern.

The equations

One Turing-type activator–inhibitor realisation suitable for numerical simulation, a discrete, grid-indexed form in the spirit of Turing's reaction–diffusion framework, the shape you would implement to run it on a lattice and watch pattern emerge. (Turing's 1952 paper introduced the general reaction–diffusion mechanism and worked specific kinetic examples; this compact pair is a representative implementation, not a verbatim transcription of his.) Two coupled fields a and b over grid cells (i, j):

∂a/∂t = s (16 − a_i,j b_i,j) + D_a ∇²a ∂b/∂t = s (a_i,j b_i,j − b_i,j − β_i,j) + D_b ∇²b

Each equation has the same two-part shape:

A local reaction term, the s(…) part, where the product a·b is the nonlinear coupling between the two morphogens. The constant 16 is a feed term, β a decay/removal parameter, s a reaction-rate scaling.
A diffusion term, the D ∇² part, the spatial spreading. The two coefficients D_a and D_b being different is a characteristic requirement of the classical two-species Turing mechanism, but not, on its own, sufficient: pattern formation also depends on the reaction kinetics and parameter regime. Diffusion must destabilise an equilibrium that is stable without spatial variation (trace J < 0, det J > 0 at the reaction-only fixed point), and then render some nonzero spatial mode unstable.

From a nearly-uniform start, systems of this form spontaneously produce spots, stripes, labyrinths, and spirals, motifs resembling those seen on animal coats, seashells, and in developing tissue (a visual resemblance; establishing that a given biological pattern is actually produced by a Turing mechanism requires molecular evidence). Turing, who was among the first to think about simulating such systems on a computer, opened an entire field of mathematical biology with them.

Watch it happen

Turing's claim is easiest to believe once you have seen it. Below is a live reaction–diffusion field, a representative Turing-type (Gray–Scott) system running in your browser. It starts from a near-uniform state seeded with a little noise; from that, diffusion-driven instability alone organises it into spots, stripes, labyrinths, and worms. Nothing is drawn by hand: every feature you see is the two morphogens reacting and spreading. Drag the sliders and watch the morphology change, especially the diffusion ratio, which is the unequal-spreading condition at the heart of the mechanism.

preset:

What you are watching. Two chemicals reacting and diffusing on a wrap-around grid, integrating Turing's equations directly. This is a demonstration of Turing's mechanism on its own terms, a tribute widget, not a RAPT claim. It computes no kernel quantity and asserts nothing about the framework; the RAPT reading is the separate, clearly-marked section below.

Runs entirely in your browser; nothing is sent anywhere. If the field goes flat, press Reset, some slider combinations relax back to uniformity, which is itself Turing's point: not every parameter regime forms a pattern.

Reading Turing through the RAPT kernel

Downstream section · not part of Turing's work.

Everything above this line describes Turing's mechanism and a representative Turing-type implementation, on its own terms. Everything below is a RAPT redescription of it. The kernel is downstream of nothing: Turing's system does not support or validate the kernel, and the redescription below earns orientation, not analytic purchase, see the two notes.

Because a reaction–diffusion system is already a field theory, its parts can be redescribed in the kernel's vocabulary. What follows is exactly that, a translation table, offered for orientation, not an analysis that earns anything Turing's own account lacks. Read it as a glossary between two languages describing the same system, and hold it to the honest standard stated just below it.

Turing's system	Kernel reading	Primitive / construct
The two morphogen fields `a`, `b`	The RAPT domain fields	`RAPT-RFT` § 16
Nonlinear coupling `a·b` (each field conditioning the other)	Candidate correspondence to state-conditioned feedback (state dependence, not, on its own, self-reference in the kernel's sense)	`T0-M` Self-Reference (candidate)
Differential diffusion `D_a ≠ D_b` destabilising uniformity	Candidate correspondence: a directional symmetry-breaking condition that can be redescribed through asymmetric admissibility (a physical transport asymmetry, not identical to the kernel's structural constraint on allowed transitions)	`T0-P` Asymmetric Admissibility (candidate)
Local neighbour coupling of the `∇²` lattice	Influence restricted to bounded neighbour relations	`T0-σ` Adjacency
A pattern, once formed, persisting	A coordinated basin exhibiting structural + temporal stability	`ATX-COORD` / `RST-COMP`

What this table does not do. Each row is a redescription, not a discrimination. Every nonlinear coupled field system has an a·b-style feedback term; naming it T0-M adds a label, not a test. Turing already told us the unequal diffusion rates are what break uniformity; calling that break T0-P restates his insight in new words, it does not predict anything he did not or forbid anything he would allow. The honest tell is falsifiability: there is no reaction–diffusion system that would fail to map onto these primitives. A mapping that cannot fail is not saying anything about the specific system, it confirms only that the primitives are broad enough to absorb it. So this table earns orientation, not analytic purchase, and it should not be read as the kernel “explaining” Turing.

On the attractor / attractlet distinction, and its lineage. One could ask whether a given Turing pattern is a self-maintaining structure or one sustained only by continuous external feed. This is adjacent to, but not identical with, the open/dissipative-vs-closed distinction of non-equilibrium thermodynamics (Prigogine's dissipative structures, order sustained by throughflow that relaxes when the flow stops). The adjacency is real. But the equivalence does not hold, and the kernel forbids it: a sovereign attractor still requires external throughput, so sovereignty does not mean thermodynamic closure, a sovereign is itself physically open. The kernel's proposed cut is therefore a different one than open-vs-closed: not whether throughput is supplied from outside (it always is), but what is supplied, bare throughput versus the identity-maintaining organisation required for continuation. Who supplies the throughput is not the same question as who supplies the organisation; it is that second question the kernel tries to isolate (and it is what makes the mitochondrion case interesting). The kernel's SOV / attractlet vocabulary is adjacent to the dissipative-structures distinction, not a rediscovery of it and not identical to it; whether its throughput-vs-organisation cut ultimately earns its keep against that literature is exactly the sort of thing this page declines to assert.

Boundary Notice

The tribute (top) presents Alan Turing's 1952 morphogenesis mathematics on its own terms, correctly attributed to him. The lower section is a downstream RAPT-RFT reading, candidate structural correspondence (V1–V2), not empirical alignment, introducing no primitives and asserting no domain ontology. It redescribes Turing's system in canonical vocabulary; it does not analyse it, and it does not yet earn a discrimination Turing's own account lacks. It cannot be cited upward: no external system validates or supports the kernel. Where any summary here differs from the canonical text, the kernel governs.

Reference. A. M. Turing, “The Chemical Basis of Morphogenesis,” Philosophical Transactions of the Royal Society of London, Series B, vol. 237, no. 641, pp. 37–72, 1952. DOI 10.1098/rstb.1952.0012 (volume, issue, pages and year stated with confidence; DOI digits believed correct but not independently re-verified here).