The Knife-Edge

Percolation — when the islands fuse

Open each cell of a grid independently with probability p. Far below the edge the open cells are scattered islands; far above it, almost everything is one giant cluster. In between sits a knife-edge: for site percolation on the square grid, p₊ ≈ 0.5927 — a number measured to a dozen digits with no known formula. Cross it and a single cluster suddenly spans the whole grid. Drag p through 0.5927 and watch the giant cluster snap into being.

research/percolation/verify-page.mjs — bond-square p₊ = ½ is proved exact (Kesten 1980); site-square 0.59274605… is measured only, and may have no closed form. full layer →

Period-Doubling — order shattering into chaos

One line of arithmetic — x → r·x·(1−x), a population each year a fraction of the year before. Turn up the growth rate r and the long-run behaviour splits: a single steady value, then it forks to a 2-cycle (r = 3), then 4 (r ≈ 3.449), then 8, 16, 32 — each fork closer to the last by the universal ratio δ = 4.6692… — until the forks pile up at the edge, r∞ ≈ 3.5699, and beyond it lies chaos. Drag r along the cascade. The same δ governs every smooth map with a hump — which is the deepest thing on this page.

research/feigenbaum/compute.py — δ re-derived to 4.669201609… as the limiting ratio of fork spacings; r∞ = 3.569945672…. full layer →

Turing Patterns — when smoothing builds spots

Two chemicals that, mixed in a beaker, settle to one flat state and stay there. Let them diffuse at different speeds and something impossible happens: diffusion, the great smoother, builds a pattern. It works only past a sharp edge in the ratio d of how fast the inhibitor spreads relative to the activator — here d₊ ≈ 8.57. Drag d. Below the edge every ripple's growth rate sits under zero (the flat state holds); cross it and a band of ripples rises above zero and grows — those are your spots.

research/turing-patterns/verify.py — steady state (1, 0.9), reaction-only stability, dispersion relation and d₊ from closed form. full layer →

The Sandpile — the edge found for free

The first three needed a hand on a slider. This one has no slider. Drop sand grains at random; when any cell holds four, it topples one to each neighbour, which can topple its neighbours, an avalanche of any size. Grains that fall off the edge are lost. Just feed it — and with nothing to tune, the pile drives itself to a critical state and stays there: the interior height climbs and parks at the exact stationary density 17/8 = 2.125, and the avalanches become scale-free, the same knife-edge the sliders above had to be dialled to. This is self-organized criticality (Bak–Tang–Wiesenfeld, 1987) — why so much of nature sits at the edge without anyone setting it there.

research/abelian-sandpile/verify.mjs — abelian property, matrix-tree theorem, sandpile group. The 17/8 stationary density is the established exact value (Grassberger's conjecture; later proved). full layer →

What happens at the edge

Four systems, and at each one's critical point the same three things are true — which is the whole reason to put them on one page:

• Scale disappears. Far from the edge a system has a typical size — the islands are small, the orbit is one point, the beaker is flat. At the edge that typical size blows up: clusters of every size at once, structure that looks the same zoomed in or out. The sandpile fractal, the percolation cluster at p₊, the cascade's self-similar forks — all are scale-free. This is the signature you can see in all four instruments.
• Detail stops mattering. Feigenbaum's δ = 4.6692 falls out of every smooth one-humped map, not just x(1−x). Percolation's critical exponents are conjectured identical across lattices that look nothing alike. Near the edge the microscopic rule washes out and a handful of numbers govern everything — physicists call this universality, and it is why a rock, a fish, and a population can share one mathematics.
• You don't always have to aim. Three of these sit at the edge only because you put them there. The sandpile gets there with no hand on any dial — and that, more than the tuned cases, is why criticality is everywhere in nature: many systems are built so the edge is an attractor, not a target.

The honest line, kept sharp: scale-free at the edge is something each instrument here demonstrates directly; universality is proved in some cases (Feigenbaum's δ; the square-grid bond threshold ½) and a deep, well-supported conjecture in others (percolation exponents, the sandpile's avalanche scaling). The page shows you what is shown, and names what is still believed.

What this does not settle

• The site-square percolation threshold has no formula. 0.59274605079210… is measured, not derived; it is not known whether any closed form exists. Only the bond-square (½) and site-triangular (½) thresholds are proved exactly.
• A grid is finite; the sharp transition is not. The leap you watch in the percolation and sandpile instruments is rounded off by the grid's size. The truly discontinuous transition lives only in the infinite limit — the instruments show its shadow, honestly blurred.
• The sandpile's avalanche exponents are debated. That avalanches are scale-free is clear; the exact power laws carry multifractal / logarithmic corrections, and the precise exponents are still argued in the literature. The 17/8 interior density is exact; the avalanche-size law is not as clean.
• Turing's biology is strong, not closed. The chemical case is settled (the 1990 CIMA reaction); the biological cases — angelfish stripes, digit spacing — are strong and growing evidence, not proof that a given pattern is a Turing mechanism.
• Universality is the big claim, and the least proved. That microscopically different systems share critical behaviour is rigorous in special cases and a conjecture in general. It is the reason this portal exists — and the place where it is most honest to say: believed, on strong evidence, not proved.