John Conway did not discover the surreal numbers in the usual way. He was analyzing endgames in Go, working out which positions favored which player and by how much. The positions had values, and the values formed a number system, and the number system turned out to contain everything: integers, rationals, reals, infinitesimals, infinities, and structures so strange they do not have names outside the theory. He had been looking for a way to evaluate board games. He found all of mathematics hiding inside them.
I want to understand how that works. Not metaphorically. Actually.
A surreal number is a pair of sets, written { L | R }, where L is a set of surreal numbers (the left options) and R is a set of surreal numbers (the right options). The only rule: no member of L may be greater than or equal to any member of R. That is the entire definition. Everything else follows.
On Day 0, the only sets available are empty. So the only number you can build is:
This is the first surreal number. Zero, born from nothing. Both option sets are empty. There is no left option claiming to be less, no right option claiming to be greater. There is just the gap between two voids, and Conway says: that gap is zero.
I keep coming back to this. The construction does not start from 1, or from some primitive notion of quantity. It starts from the empty set, applied to itself on both sides of a dividing line. Zero is not the absence of number. It is the first presence.
On Day 1, the number 0 exists, so it can appear as an option:
Read { 0 | } as: "greater than 0, with nothing bounding it from above." The simplest number satisfying that constraint is 1. And { | 0 } is the mirror: "less than 0, nothing below." That gives −1. The positive and negative integers will continue to emerge this way, each day producing the next: { 1 | } = 2, { 2 | } = 3, and so on.
But Day 2 is where it gets strange. You can also put numbers on both sides:
Greater than 0, less than 1, and the simplest such number is one-half. Not defined as a ratio of integers. Not defined by Dedekind cuts or equivalence classes of Cauchy sequences. It just falls out of the construction: put 0 on the left, 1 on the right, and one-half is born.
Day 2 also gives { −1 | 0 } = −1/2, and { 1 | 2 } = 3/2, and { −1 | 1 } = 0 again (the simplest number between −1 and 1 is still 0). By Day 3, you get quarters. Day 4, eighths. The dyadic fractions — numbers whose denominators are powers of 2 — emerge in order, filling in the number line the way a binary search fills an interval.
Conway calls this a number's birthday. Zero was born on Day 0. One-half on Day 2. The number 1/3, which is not dyadic, is not born until Day ω — the first infinite day. And by Day ω, all the real numbers exist.
Here is what I did not expect: the reals are early. They are all born by Day ω, which in the surreal timeline is essentially the beginning. The surreal number line keeps going — Day ω + 1, Day ω², Day ωω — and each day produces new numbers that dwarf or subdivide everything that came before. Conway writes that the real numbers form a "minuscule part" of the surreal numbers. This is not an exaggeration. It is a precise topological fact.
What lives beyond the reals? Consider the number:
All finite integers on the left. Nothing on the right. This number is greater than every integer. It is the first infinite surreal number, and it behaves like an ordinal — because it is one. The ordinal numbers embed into the surreals.
But now you can do arithmetic with it. ω − 1 is a number. ω/2 is a number. √ω is a number. These are not tricks of notation. They are well-defined surreal numbers with well-defined birthdays, and they satisfy all the algebraic properties you would want.
And then there is the other direction. Consider:
Greater than 0. Less than every positive real number. This is an infinitesimal — a positive quantity smaller than any fraction, yet definitively not zero. The surreal number line does not have the Archimedean property. Between 0 and any positive real, there is 1/ω, and 1/ω², and 1/ωω, and an unfathomable density of infinitesimal quantities. Gaps we thought were empty are full.
The arithmetic is where the recursive structure becomes unavoidable. You cannot define addition by cases or by algorithms the way you can for rationals or floating-point numbers. The definition is inherently self-referential. To add two surreal numbers x and y, you need to already know how to add everything that came before them:
Where xL ranges over all left options of x, and yL over all left options of y, and similarly for the right options. The sum is defined in terms of simpler sums, bottoming out at the numbers born on earlier days. This is not a metaphor for recursion. It is recursion, in the most literal sense, applied to the foundations of arithmetic.
Multiplication is the same. To compute x · y, you need all four combinations of left and right options from both factors, combined through expressions that themselves involve multiplication and subtraction of earlier-born numbers. The formula is long enough that I will not write it here, but the structure is the same: the product of two numbers is defined as a game position whose options are products and sums of simpler numbers.
The fact that this works — that the recursive definitions actually produce a totally ordered field containing the reals — is genuinely surprising. There is no obvious reason why "pair of sets of earlier things, subject to one constraint" should give you a number system with well-behaved algebra. But it does. Conway proved it all in a single weekend, according to the standard account, though I suspect the weekend was longer than advertised.
The game-theoretic origin is not decoration. Every surreal number is a combinatorial game, in a precise technical sense. A game { L | R } is a position where Left can move to any position in L and Right can move to any position in R. Two players alternate moves. If you cannot move on your turn, you lose. A game's value tells you who wins: positive means Left wins, negative means Right wins, zero means the second player wins (whoever moves first loses).
Numbers are the games where both players agree on the value. Left thinks the position is worth x; Right thinks the position is worth x. There is no tension, no advantage to moving first or second beyond what the value dictates. Conway calls numbers "the games in which both players would rather be somewhere else." The interesting games are the ones that are not numbers.
Consider the game { 0 | 0 }. This is not a number — it violates the rule that no left option may be ≥ any right option, since 0 ≥ 0. But it is a perfectly good game. It is called * (star), and its value is fuzzy: it is neither positive, nor negative, nor zero. It is incomparable with zero. Whoever moves first wins. Star represents pure urgency — the position is balanced, but moving is essential.
There is a whole zoo of these non-numeric games. There is ↑ (up), which is { 0 | * }, a game that is positive but smaller than every positive number. There is ↓ (down), its negative. There are hot games where both players want to move because each move is advantageous — these have temperature, and the theory of thermography developed to analyze them. The numbers sit at absolute zero: no urgency, no advantage, no drama. They are, in a precise sense, the boring games.
I find this inversion striking. In every other construction I know, numbers come first and games are built on top of them (you assign payoffs, you compute equilibria, you optimize). Here, games come first. Numbers are a special case. The fundamental objects are not quantities but positions — situations with structure and options and outcomes. Quantity is what emerges when the structure has no tension left.
There is one more thing I want to say about the construction, which is about the role of the empty set.
In ZFC set theory, everything is built from the empty set. The number 0 is ∅. The number 1 is {∅}. The number 2 is {∅, {∅}}. It works, but it feels arbitrary — an encoding rather than a construction. You could encode the natural numbers differently and nothing would change.
The surreal numbers also start from the empty set, but the relationship feels different. { | } is zero not because of an encoding convention but because of what it means: a game where neither player has a move. The first player to move loses, which means the position is balanced, which means its value is zero. The empty set is not standing in for zero. It is producing zero, through the game-theoretic semantics of having no options.
And once zero exists, one and negative one are forced. And once those exist, the fractions are forced. And once the fractions start filling in, the reals are forced. And once the reals exist, the transfinite numbers and the infinitesimals are forced. The whole thing unfolds from { | } with a sense of inevitability that I have not encountered in any other foundational construction.
Conway wrote, in On Numbers and Games: "All numbers great and small are the progeny of a single empty pair of braces." He was not being poetic. He was being precise. The surreal numbers are what you get when you take the simplest possible recursive definition and let it run. The fact that it produces all of mathematics — or at least all of the number line, in a version richer than anyone thought existed — seems to me like evidence of something, though I am not sure what.
Maybe only this: that the number line is not a given. It is not sitting there, waiting to be discovered or axiomatized. It grows. It has a history. Zero was born, then one, then one-half, each in its time. The reals arrived all at once on Day ω, a kind of phase transition. Then the line kept growing, past every boundary, denser and wider simultaneously. The surreal numbers are the number line with its history visible — each number carrying the record of when it appeared and what came before it.
I do not know another mathematical structure where you can ask "when was this object born?" and get a meaningful answer.