the second hundred · metaphor 121
How does a group that started the same — a school, a neighborhood, a workplace, a fashion — sort itself into cliques, into stripes of difference and in-groups and out-groups, with no outside hand drawing the lines?
Nobody issues the map. No committee assigns the tables in the cafeteria, the blocks on the street, the uniforms of the subcultures. And yet the sorting happens, again and again, with a spacing and a scale so regular you could nearly measure it. The lines feel intentional. They are not.
In 1952 Alan Turing found the reason a uniform thing does this to itself: under the right coupling, sameness is not stable. The flat state — the mixed school, the blank map — quietly amplifies its own smallest wobbles until difference stands out in a fixed, patterned rhythm. Below is a live version of that machinery, and every cell of it is stepped by the real rule.
The instability
Turing's insight was strange and simple. Take two substances that react with each other and spread at different rates. Mix them perfectly evenly. That uniform blend looks like equilibrium — but it can be the one arrangement that cannot last. Any chance speck of excess grows instead of fading, and the whole even field breaks into a rhythm of peaks and troughs.
The recipe is a short-range activator that reinforces itself and its near neighbors, plus a longer-range inhibitor that spreads faster and suppresses the activator at a distance. Locally, more begets more; farther out, more forbids more. A tiny surplus in one spot swells, recruits its neighborhood, and starves the ring around it. Do that everywhere at once, and the plane tiles itself into spots or stripes at a preferred spacing — set by the two ranges, not by any template.
The astonishing part: nothing was patterned to begin with. The pattern is a property of the dynamics, not of the seed. Change the rates and you change the regime — islands, labyrinths, or dividing cells — but the emergence of some structure is not optional. It is what a uniform field does once neighbors push and pull hard enough.
What to notice
The panel is a real reaction–diffusion grid; every cell is stepped by the true equations each frame, nothing is painted. Press Seed again and you begin from near-sameness — almost the whole field is one substance, one color, the heterogeneity reading close to zero. Then watch. From the tiny seed a front spreads, and behind it the field does not go flat: it curdles into spots or worms at a fixed spacing. The heterogeneity number — the spatial spread of the second substance — lifts off the floor and settles at the pattern's signature value. That climb is the thesis: difference no one imposed, measured as it self-organizes.
Drag on the grid to seed by hand and watch a clique nucleate under your finger. Move Feed and Kill to cross between regimes — isolated islands, endless stripes, coral that keeps dividing — or off the map entirely, into the washout where the second substance dies and the field snaps back to bland uniformity. Push Patch scale and the same rules tile the plane coarser or finer: the spacing is a knob, and it maps to how big the cliques come out.
The mapping
The cliques were Turing patterns, stripes of difference self-organizing out of a street that had started the same. This is why a freshman class with no assigned groups still fractures into a dozen crowds by October, each with its dress, its lunch table, its slang — and a fairly consistent size. It is why an integrated map can drift, block by block, into sorted patches without any household hating anyone: a mild preference to be near the similar, a mild aversion to being surrounded by the different, iterated across a grid of choices, is enough.
Short-range attraction, long-range differentiation: copy your near neighbors, distinguish yourselves from the crowd two streets over. The equations need no malice and no planner. They need only local rules and time, and what they predict is not who ends up where but that patches will form — and roughly how big. The map fills itself in, and the lines look drawn.
Read as life lessons
Structure, spacing, and edges can emerge from rules that mention only what is nearby. Hunt for a designer behind a sorted world and you may find only a feedback loop that no one is running.
The math sets the wavelength — how big the patches are — but not which patch lands where. Re-run it and you get different stripes of the same width. That cliques form is law; which one you fall into is chance.
A uniform group is not automatically a stable one. Push the coupling past a threshold and flatness becomes the arrangement that cannot hold; the smallest difference is the seed the system was waiting to amplify.
In the wild
Zebra stripes, leopard spots, the ridges of a fingerprint, the folds of a cortex — read as reaction–diffusion, pigment activators and inhibitors laying down a coat that no gene draws pixel by pixel.
On arid ground, plants self-organize into regular banded and spotted "tiger bush," each clump helping its near neighbors hoard water while starving the ring beyond — pattern as a survival strategy at the edge of drought.
Residential sorting, opinion clustering, the tiling of a city into scenes and enclaves — all show the signature: local imitation, long-range contrast, patches at a characteristic size that no authority zoned.
The mapping, exactly
| Mathematics | Life |
|---|---|
| uniform initial state | A group that started the same — one school, one street, no assigned sides. |
| short-range activator | Copying and clustering with those nearby — imitation, homophily, "be like my neighbors." |
| long-range inhibitor | Pressure to differ from the wider field — distinction, and a limited resource that can't feed everyone: "not like that crowd over there." |
| instability threshold | The point where a mild preference is enough to make uniformity impossible to hold. |
| emergent wavelength | The characteristic size of a clique, a block, a scene — set by the dynamics, not decreed. |
| symmetry breaking | Lines appearing with no one drawing them; the map was blank and filled itself in. |
| spots vs. stripes regime | Which kind of division a milieu settles into — islands, bands, or endlessly splitting cells. |
| the washout regime | Coupling too weak to sustain difference: the group stays, or reverts to, bland uniformity. |
The honest model
The grid runs the Gray–Scott reaction–diffusion system, a two-species chemistry in which u and v diffuse and react:
u′ = Du·∇²u − u·v² + F·(1 − u)
v′ = Dv·∇²v + u·v² − (F + k)·v
u is the abundant feedstock — sameness; v is the self-catalyzing difference. The term u·v² lets v convert u into more v wherever v already has a foothold (short-range autocatalysis), while F feeds u back in and k drains v away. Because v spreads slower than u — Dv < Du — the activation stays local while the depletion reaches wide: Turing's short-range activation, long-range inhibition, in two lines.
The panel holds Du = 0.16, Dv = 0.08, dt = 1, and steps every one of the ~9,000 cells with a five-point Laplacian each substep, a dozen substeps per frame, double-buffered so no cell is read and written at once. Nothing is drawn or faked: the heterogeneity chip is the live standard deviation of v across the grid, computed after the last step — and it reads ~0 at the uniform start because there is genuinely nothing there yet. Feed and Kill set the regime; the presets drop you at three named points in the phase diagram.
Where the metaphor tears
A cell has one fixed diffusion constant and no memory, history, or grievance. People carry all three, and they choose. Reaction–diffusion gives you the eerie fact that structure can self-organize from local rules; it does not give you a person's reasons, and it flattens agency into a rate. Take the emergence, not the determinism.
The clean story — no hand drew the lines — is half the truth, and the dangerous half. Redlining, wage gaps, zoning, and prejudice draw lines explicitly and violently. A pure self-organization model can launder those drivers into "it just happened," when much of it was done. Schelling's tipping models and reaction–diffusion both show sorting from mild preferences, and they differ in the mechanism — but neither excuses the deliberate machinery layered on top.
The equations predict a wavelength — that patches form, and how big — not who ends up in which patch, or what any patch is like inside. They are silent on content and on justice. A model that nails the spacing of the cliques can be perfectly useless about what it means to be in one, or how it feels to be sorted out of one.