Some numbers refuse to be fractions. One refuses harder than all the rest — provably — and you can watch a sunflower obey it.
Decimals lie about numbers — they round, they pretend to stop. A continued fraction doesn't. Any number can be written as a whole part plus one-over (a whole part plus one-over (a whole part plus …)), and that sequence of whole parts is unique: a number's true name. Rationals are the ones whose names end. Irrationals run forever. Pick one and watch its name unfold — and watch the fractions you get by stopping early, which are the best approximations of their size that exist.
| stop at | best fraction | value | error |
|---|
Look what happens to π. After 355/113 its true name has a huge entry — 292 — and a huge entry means the approximation just before it was already astonishingly good: 355/113 pins π to six decimal places — seven significant figures — with a three-digit denominator. A big number in the name is a place where a fraction nearly catches the number. So the numbers that are hardest to catch are the ones whose names hold no big numbers at all — nothing but 1s.
There is exactly one number whose true name is all ones: [1; 1, 1, 1, …]. Work out what that nested fraction must equal and it solves x = 1 + 1/x — the golden ratio, φ = (1+√5)/2 ≈ 1.6180339887. With no large entries ever, its fractions close in as slowly as fractions possibly can. Its convergents are the ratios of consecutive Fibonacci numbers — 1/1, 2/1, 3/2, 5/3, 8/5, 13/8, … — each the best rational approximation for its size, and each barely better than the last.
This is not loose talk. In 1891 Adolf Hurwitz proved that for every irrational number you can find infinitely many fractions p/q within 1/(√5·q²) of it — and that the constant √5 is the best one that works for all of them. Push for anything tighter and the claim breaks, and it breaks first and worst for the golden ratio and the numbers that share its all-ones tail. In the precise sense of how well it can be approximated by rationals, φ is the most irrational number there is — it, and its tail-sharing kin, mark the worst case of the whole real line, the point past which that √5 cannot be pushed. Below, every number's accuracy is plotted against how many steps you give it. Watch φ climb slowest.
“Most irrational” sounds like a figure of speech. It isn't. Give a number its own Lagrange number: the tightest constant L such that infinitely many fractions p/q still land within 1/(L·q²) of it. A larger L means fractions keep catching it closely; a smaller L means it holds them off. Hurwitz's √5 is exactly the smallest Lagrange number there is, and φ is the number that hits that floor. So irrationality has a bottom, and φ is standing on it.
Now the surprise. Ask for the second-hardest number and you don't get φ's near-neighbour. You get a clean jump to √8, the number whose true name is all 2s (that's 1+√2, and everything sharing its tail, the family of √2). Nothing at all lives between √5 and √8. Keep going and the hard numbers keep coming in a discrete ladder (√5, √8, √221/5, √1517/13, …) that climbs, rung by rung, toward 3 and never reaches it. Andrey Markov proved in 1879 that this ladder is indexed by something with no obvious right to be here: the whole-number solutions of x² + y² + z² = 3xyz. Each Markov solution m = 1, 2, 5, 13, 29, 34, … gives one rung, L = √(9m² − 4)/m. Grow the tree of solutions and watch the ladder fill in:
| Markov m | from the triple | rung L = √(9m²−4)/m |
|---|
Now the part that shouldn't be connected to any of the above, and is. A growing plant lays down each new seed, leaf, or floret at a fixed angle, turned from the last. Ask which angles keep new growth from ever stacking directly behind old — no radial pile-ups — and the answer turns on irrationality.
If the angle is a rational fraction of a full turn, say 3/8, then every eighth seed lands in the same direction and the whole head collapses into eight radial spokes. To avoid any such alignment for as long as possible, you want the angle whose fraction-of-a-turn is caught by simple fractions as badly as possible: the most irrational fraction of a circle. That is 137.50776°, the golden angle, 360° × (2 − φ) = 360° ÷ φ², the divergence angle seen most often in real plants. Dial it in, and let the flower count its own spirals for you:
This is the third layer of this archive to circle the same drain, and now the circle closes. Incommensurable showed that √2 can be no fraction: odd cannot equal even. The Comma showed that no stack of perfect fifths ever lands on a stack of octaves, the same odd-against-even refusal, this time as a sourness you can hear. And here is the same impossibility taken to its limit: of all the numbers that refuse to be fractions, φ refuses hardest, standing on the bottom rung of the whole ladder of refusal, and a sunflower is what that maximal refusal looks like when it is trying to live, turning each new seed by an angle that resists ever repeating, so the head fills evenly instead of stacking into spokes.
The strange part is the direction of it. We did not put it there; the pattern falls out of the way a growing tip lays down each new primordium — by what mechanism exactly is still argued — long before anyone wrote down √5. The most abstract fact in this whole place, a theorem about how badly fractions fail, turns out to be among the most physical, latent in every seed head you have ever idly looked at. The number that hides from arithmetic is the one a flower reaches for when it needs to leave nothing in line.