Here are two heaps of stones. Take as many as you like from one heap, or the same number from both. Whoever clears the table wins. It is a children's game. Its secret is the most irrational number there is.
In 1907 the Dutch mathematician Willem Wythoff wrote down a tiny variation on the ancient game of Nim. Two piles of tokens sit between two players. On your turn you do exactly one of three things: take any number of tokens from the first pile, take any number from the second, or take the same number from both at once. The player who removes the last token wins.
That third move — the one that empties from both piles together — is the whole story. Without it the game is trivial. With it, the game has a perfect strategy hiding inside it, and the shape of that strategy is two straight lines whose slope is φ = 1.618…, the golden ratio. Not approximately. Exactly. The losing positions are the integer points (⌊nφ⌋, ⌊nφ²⌋), and below the golden ratio falls out of a stone-counting game the way a struck bell finds its pitch.
Play first. Then we'll take the machine apart.
Picture the two pile sizes as a single chess queen on a quarter-infinite board, at the square (a, b). The three moves are exactly the queen's moves toward the corner: slide left (take from pile a), slide down (take from pile b), or slide diagonally down-left (take the same from both). Whoever lands the queen on the corner (0, 0) wins. You move; then a machine that has solved the whole board moves back. It will not lose. Find out why.
If you played a few rounds you found the rule the machine obeys: from most squares there is a move that drops the queen onto a gold square — and once it sits there, every escape only lands on a non-gold square, where your opponent can put it back. The gold squares are the cold positions: lose them and, against perfect play, you have already lost. Everything else is hot — a square from which the winning move exists. The entire theory of the game is the list of cold squares. Reveal them, and they line up.
Plot every cold position. They are not a cloud — they are two rays leaving the corner, mirror images across the diagonal, and their slopes are φ and 1/φ. Each cold position is (⌊nφ⌋, ⌊nφ²⌋) for n = 0, 1, 2, …: the n-th multiple of the golden ratio, floored, paired with the n-th multiple of its square. Below the dots, the two coordinate-sequences are unrolled along the number line — and something quietly miraculous happens.
Run your eye along the number line. Every positive integer is exactly one colour. No integer is skipped; no integer is claimed twice. The lower sequence 1, 3, 4, 6, 8, 9, 11, … and the upper sequence 2, 5, 7, 10, 13, 15, 18, … interlock perfectly to tile all of ℕ. This is a theorem older than the game: Rayleigh's theorem (Lord Rayleigh, 1894), better known as Beatty's theorem after the problem Samuel Beatty set in 1926. For any two irrational numbers r and s with 1/r + 1/s = 1, the sequences ⌊nr⌋ and ⌊ns⌋ together hit every positive integer once and only once.
And 1/φ + 1/φ² = 1 — exactly, because φ² = φ + 1, so (φ+1)/φ² = φ²/φ² = 1. The golden ratio is a Beatty pair with its own square. That's rare. Which raises the real question: of all the irrational pairs that tile the integers, why does this game land on φ specifically?
The answer is the diagonal move. Look back at the cold positions: the gap between the two coordinates of the n-th one is always exactly n. ⌊nφ²⌋ − ⌊nφ⌋ = n, because φ² = φ + 1. That is forced by the rule "take the same number from both piles": the move along the diagonal can never change the difference between the piles, so the n-th cold position must be the one whose piles differ by n, and there can be only one per difference. So the upper sequence is pinned to be the lower sequence plus n — meaning s = r + 1. Now demand that the two sequences still tile the integers, 1/r + 1/s = 1:
1/r + 1/(r+1) = 1 ⟹ r² = r + 1 ⟹ r = φ
One equation, one positive solution. The golden ratio is not chosen, not approximated, not aesthetic — it is the only number for which "take the same from both piles" and "leave no integer uncovered" can both hold. Turn the dial below and watch every other value fail: leave gaps in the number line, or double-cover it. Only at φ does the coverage become seamless.
Recomputed live, and checked offline first. The machine in Instrument I does not consult a formula to play — it solves the board the hard way, by retrograde analysis: starting from the corner and working outward, every square is labelled cold or hot purely by the game's own rules (a square is cold exactly when no move escapes to another cold square). Only afterwards do we overlay the golden-ratio formula and find the two agree on every square. The Beatty partition and the forced identity r²=r+1 are likewise rechecked above as you watch. All of it is independently proven in /research/wythoffs-game/verify.mjs (12/12): the brute-force cold positions equal (⌊nφ⌋,⌊nφ²⌋) and equal the classical mex construction; the two sequences partition 1…100,000 with zero gaps and zero collisions; the integer floors are proven exact against square-root bounds (no floating-point slack); the first terms match OEIS A000201 and A001950 fetched from the catalogue; and the Sprague–Grundy value is zero on exactly the cold positions.
What is and isn't claimed. "The golden ratio wins" is shorthand: φ describes the losing positions, and the winning strategy is to always hand your opponent one of them. The strategy is perfect under normal play (last to move wins) for this specific move-set — change the rules (misère play, a third pile, a bound on how much you may take) and the clean golden structure can break. The dial in Instrument III shows the tiling condition geometrically; it is a faithful illustration of the algebra 1/r+1/(r+1)=1 ⟹ r=φ, not an independent proof of Beatty's theorem (which is cited). And φ here is the same φ that is the "most irrational" number — the one hardest to approximate by fractions — which is why its multiples spread most evenly along the line and leave no integer behind. That is no coincidence; it is the deep reason the game is so clean.
Sources. W. A. Wythoff, “A modification of the game of Nim,” Nieuw Archief voor Wiskunde (2) 7:199–202, 1907 (the game and its golden-ratio solution). Lord Rayleigh (J. W. Strutt), The Theory of Sound, vol. 1, 2nd ed., Macmillan, 1894, p. 123 (the complementary-sequence statement). S. Beatty, problem 3173, American Mathematical Monthly 33:159, 1926, with solutions ibid. 34:159–160, 1927 (whence “Beatty sequences”). H. S. M. Coxeter, “The golden section, phyllotaxis, and Wythoff's game,” Scripta Mathematica 19:135–143, 1953. A. S. Fraenkel, “How to Beat Your Wythoff Games' Opponent on Three Fronts,” Amer. Math. Monthly 89:353–361, 1982. OEIS sequences A000201 (lower Wythoff) and A001950 (upper Wythoff), values verified against their b-files. The floor identity ⌊nφ²⌋ = ⌊nφ⌋ + n and the partition follow from φ²=φ+1 and Beatty's theorem.