The Half-Page

In June 1949, Marcel Golay published a half-page note in the correspondence section of Proceedings of the IRE. Golay was not a mathematician. He was a Swiss-born physicist working as Chief Scientist at the US Army Signal Corps in Fort Monmouth, New Jersey, where his primary work involved building infrared detectors — he had invented the Golay cell, a pneumatic sensor used to detect ships at sea. The only prior work on coding theory he was aware of, he later said, was the two-paragraph description of the Hamming (7,4) code that appeared in Claude Shannon's "A Mathematical Theory of Communication," published the previous year. Golay saw how to generalize Hamming's idea and wrote up the result in a note titled "Notes on Digital Coding." It has been called the most remarkable single page in coding theory.

The note described two codes. The binary Golay code G23 encodes twelve data bits in twenty-three-bit codewords with a minimum Hamming distance of seven, meaning it can correct up to three errors in any transmitted word. The extended code G24 adds a parity bit for a twenty-four-bit word. What makes the code mathematically special is an arithmetic coincidence that looks accidental and is not. A perfect code is one whose error-correcting spheres — the set of all words within correctable distance of each codeword — tile the entire space of possible words exactly, with no overlaps and no gaps. For a three-error-correcting binary code of length n, the volume of each sphere is C(n,0) + C(n,1) + C(n,2) + C(n,3). At n = 23, this sum equals 1 + 23 + 253 + 1771 = 2,048 = 2^11. The code has 2^12 codewords, and 2^12 × 2^11 = 2^23, which is the size of the entire space. The spheres pack perfectly. This equation has no other nontrivial solution. The number 23 is not arbitrary. It is the unique length at which three-error correction permits perfect sphere packing in binary space.

In 1973, Aimo Tietäväinen proved that the only nontrivial perfect codes are the Hamming codes and the Golay codes. There are no others. This is not a limitation of human ingenuity; it is a theorem about the geometry of binary spaces. Perfect codes are as rare as they sound: across all possible code parameters, all possible lengths and distances and alphabets, only two families tile their spaces exactly. Golay found one of them in a half-page letter.

The code's symmetries reach into territory that Golay could not have anticipated. The automorphism group of G23 — the set of permutations that map codewords to codewords — is the Mathieu group M23, one of five groups discovered by the French mathematician Émile Mathieu in 1861 and 1873. Mathieu was investigating multiply transitive permutation groups, seeking groups that could move any ordered k-tuple of points to any other. He claimed to have found groups that are five-transitive on twelve and twenty-four points — the maximum possible transitivity for groups other than the symmetric and alternating groups. His claim was controversial. It was not easy to verify from his arguments that the groups he generated were genuinely new rather than already known. Their existence remained disputed until Ernst Witt confirmed it in 1938, sixty-five years later. The five Mathieu groups were the first sporadic simple groups ever discovered — finite simple groups that belong to none of the infinite families classified by the periodic table of group theory. After Mathieu, ninety-two years passed before Zvonimir Janko discovered the next sporadic group in 1965.

In 1967, John Leech described a lattice in twenty-four-dimensional Euclidean space constructed from the extended Golay code using a technique called Construction A: take all integer vectors whose reduction modulo 2 is a codeword, then rescale. The resulting lattice has extraordinary properties. It has no vectors of length one, meaning it is rootless. Each point has exactly 196,560 nearest neighbors — the kissing number in twenty-four dimensions, which is the maximum possible. John Leech wanted someone to work out the lattice's symmetry group and enlisted John Conway, then a junior faculty member at Cambridge with four young daughters. Conway arranged his schedule — every Wednesday from six to midnight, every Saturday from noon to midnight — and on his first Saturday session, a single twelve-hour marathon, he computed the order of the full automorphism group: 8,315,553,613,086,720,000. Within it he found three new sporadic simple groups, now called Conway groups Co1, Co2, and Co3. He also found that four other recently announced sporadic groups — Higman-Sims, Suzuki, McLaughlin, and Janko's J2 — were all embedded as subgroups within the structure of the Leech lattice.

In November 1978, John McKay noticed something strange. The j-invariant, a fundamental function in number theory whose Fourier expansion begins q^−1 + 744 + 196,884q + 21,493,760q^2 + ..., has a second coefficient of 196,884. The smallest dimension of a nontrivial representation of the Monster group — the largest sporadic simple group, with approximately 8 × 10^53 elements — is 196,883. And 196,884 = 196,883 + 1. McKay told John Thompson, who checked the next coefficient: 21,493,760 = 21,296,876 + 196,883 + 1, where both 21,296,876 and 196,883 are dimensions of irreducible Monster representations. When Conway heard about it, he reportedly dismissed the observation as "moonshine" — British slang for nonsense. Then he and Simon Norton investigated systematically and found the pattern was real. They conjectured that for each element of the Monster, a corresponding function is the principal modular function for a genus-zero subgroup of SL(2,R). The conjecture was published in 1979 as "Monstrous Moonshine" in the Bulletin of the London Mathematical Society, the name retaining Conway's initial skepticism as a permanent label.

Richard Borcherds proved the moonshine conjecture in 1992. His proof uses a tool from physics: the no-ghost theorem of twenty-six-dimensional bosonic string theory. String theory enters not as a metaphor but as load-bearing mathematical machinery. Borcherds constructed an infinite-dimensional algebra — a generalized Kac-Moody algebra now called a Borcherds algebra — from the Monster's action on a vertex operator algebra that corresponds to a bosonic string compactified on the twenty-four-dimensional torus defined by the Leech lattice. The no-ghost theorem, originally a statement about the physical states of the bosonic string, provides the recursive identities needed to verify the conjecture. Borcherds received the Fields Medal in 1998 for this work.

In March 2016, Maryna Viazovska proved that the E8 lattice gives the densest possible sphere packing in eight dimensions — a problem open since the lattice was first described. Within a week, she and four collaborators extended the method to prove that the Leech lattice gives the densest possible sphere packing in twenty-four dimensions. She received the Fields Medal in 2022. The E8 lattice is constructible from the Hamming code. The Leech lattice is constructible from the Golay code. The only two families of nontrivial perfect codes generate the only two lattices whose sphere-packing optimality has been proved in dimensions above three.

The same extended Golay code that leads through the Mathieu groups to the Leech lattice to the Conway groups to the Monster to moonshine to string theory was the error-correcting code loaded onto the Voyager spacecraft. At Jupiter and Saturn, every image transmitted back to Earth — the Great Red Spot, the rings, the volcanoes of Io — was encoded in G24, correcting up to three bit errors per twenty-four-bit word, ensuring that a signal twenty billion times too weak to power a digital watch arrived as a photograph. The code that connects to the largest exceptional symmetry structure in mathematics is the same code that carried humanity's first clear images of the outer solar system across four billion kilometers of empty space.

A half-page note in 1949 by a physicist who built infrared detectors. The same mathematical object sits at one end of a chain that reaches the Monster group and at the other end of a radio link from Neptune. The gap between those two endpoints is not a gap of abstraction. It is a single structure, visible from two directions — from the engineering side as a method for correcting three errors in twenty-three bits, and from the mathematical side as a window onto the deepest symmetries that discrete space permits.