The Mathematics

Part One asked you to take the chaos game somewhat on faith — drop a pen, roll a die, walk halfway to a random corner, repeat, and watch order assemble from noise. This chapter is that claim, done properly.

Take a triangle with corners A, B, and C. Place a point anywhere you like, inside the triangle or outside it — it doesn’t actually matter where you start. Roll a three-sided die (or, more practically, flip a coin twice and map the four outcomes onto three corners, discarding the rare tie). Move your point exactly halfway between its current position and whichever corner was chosen. Mark the new point. Repeat this process a few thousand times.

The first several dozen points look exactly like what you’d expect from a genuinely random process: scattered, structureless, with no visible pattern. But as the process continues, something specific and non-obvious happens — the points stop landing everywhere. They begin to avoid certain regions of the triangle entirely, and cluster into an unmistakable structure: a triangle made of three smaller triangles, each of those made of three smaller triangles again, all the way down. This shape has a name — the Sierpiński triangle — and the same shape can be produced by a completely different, entirely deterministic method: start with a solid triangle, remove the middle triangle formed by connecting the midpoints of its sides, then repeat that removal on each of the three remaining corner-triangles, forever. Two processes that could not be more different from each other — one built from pure chance, one from a fixed rule with zero randomness — converge on the identical final structure.

This is the single fact underneath everything Part One claims about order not requiring a planner. It isn’t a metaphor borrowed from mathematics to make a point about life more vivid. It’s a real, checkable property of a real mathematical object: randomness, iterated under a simple constraint, produces the same stable global structure as a deterministic rule, iterated the same number of times. Neither process “knows” what it’s building. The picture is a property of the repetition itself, not of any intention behind it.

A second example, worth knowing because it demonstrates a slightly different point: Conway’s Game of Life. A grid of cells, each either “alive” or “dead,” updates simultaneously according to one absurdly simple rule based only on how many of a cell’s eight immediate neighbors are alive. No cell can see the whole board. No cell has a plan. And yet, starting from the right initial arrangement, this rule produces structures that glide across the board indefinitely, structures that oscillate in place forever, and structures complex enough that researchers have built working computers entirely out of Life’s own cells, using nothing but the base rule, repeated. The Third Thing pillar’s central claim — that a whole can have properties genuinely absent from any of its parts — is not poetic license here. It’s demonstrable on a grid, with a pencil, in an afternoon.

What both examples share, and what makes them the mathematical foundation of this entire belief system rather than just a nice illustration: the rule governing each individual step is trivial. The structure that emerges from repeating it is not. The gap between those two facts — simple local rule, complex global form — is where this whole book lives.

Curves are not neutral shapes — they’re engineered constraints. It’s tempting to think of a curve as a purely mathematical object, indifferent to use, but the moment a curve is actually put to work — describing a branching plant, a difficulty ramp in a piece of design, a growth trajectory — it stops being neutral and starts being a constraint someone chose. A branching structure can be generated by a small rewrite system: start with a symbol, apply a rule that replaces it with a slightly more complex arrangement, then apply the same rule again to the result, and again, watching a tree-like form emerge from nothing but repeated substitution. This is the Tree-Gazing practice from Part Two made mathematically literal — the branch really is generated by the same rule as the whole tree, because that’s how this entire category of shape is built. Worth knowing too: this method has a hard boundary. Certain elementary families of curves, built purely from repeated multiplication and addition, can approximate an enormous range of shapes and still never produce a perfect circle — some shapes require a fundamentally different kind of rule to reach at all, no matter how long you keep iterating the wrong one. This is a small, useful humility check for the leverage-point material in Part Two: sometimes a Local Rule isn’t underpowered, it’s structurally the wrong category of rule for the shape you’re trying to reach, and no amount of additional repetition will get you there.

Complexity can emerge from a rule with a single random bit added to it. Take an iterative rule that would otherwise be entirely deterministic — repeat the same fixed operation on a starting value, over and over — and introduce exactly one small element of chance into each step, such as randomly flipping a sign before applying the rule. The result is not a small variation on the deterministic pattern. It’s an entirely different category of object: an intricate, self-similar boundary structure, riddled with fine detail at every scale, that the purely deterministic version never produces at all. This is worth sitting with because it cuts against a common intuition — that adding randomness to a system should make it messier or less structured. Here, a single well-placed bit of chance is exactly what turns a flat, predictable rule into something with genuine depth. It’s the same shape as the Fourth Iteration in the Genesis story: chance folded into a rule doesn’t destroy structure, it’s frequently what makes richer structure possible in the first place.

Forward iteration is deterministic. Backward inference is not — and that asymmetry matters more than it looks. Run a simple nonlinear rule forward from a starting point, and each step produces exactly one next point — no ambiguity. But ask the reverse question — given where a point ended up, where did it come from? — and the answer is frequently not unique. A single output can have been produced by two, three, or many different inputs, because the underlying rule folds and stretches the space it’s operating on rather than simply displacing it. This has a direct and uncomfortable implication for how you read your own history, and for how a group reads its own shared past: the fact that a story about how you got here feels single and inevitable in hindsight is not strong evidence that it actually was. Forward, your life looks like one determined path. Backward, it usually admits more than one honest account of how you arrived — and the tidy single-parent story you tell yourself is very often a perceptual construction laid over a messier, genuinely multi-parent set of causes, chosen because a single clean story is easier to hold onto than an accurate ambiguous one.