Paul Erdős owned almost nothing and proved almost everything. For six decades he lived out of a single suitcase, moving from one mathematician's spare room to the next, announcing at the door that "my brain is open." He wrote around fifteen hundred papers with more than five hundred co-authors — a body of work so collaborative it produced its own unit of distance, the Erdős number: your degree of separation, through joint papers, from the man himself.
His mathematics lived in combinatorics, number theory, set theory and probability. He was a founder of the probabilistic method — proving an object exists by showing a random one works with positive probability — and of Ramsey theory's central questions about the unavoidable order inside any large enough structure. He posed problems the way other people breathe, often attaching small cash prizes; some of those problems are still open, and some, eighty years on, are only now falling.
Erdős spoke of The Book: a transfinite volume, held by a deity he otherwise professed not to believe in (the "Supreme Fascist"), in which the most perfect proof of every theorem is written. You did not have to believe in God, he said, but you had to believe in The Book. To call a proof "straight from The Book" was his highest praise — not that it was correct, but that it was inevitable, the argument stripped to the one line the theorem always wanted.
Every scientist in this gallery is read as one traversal of the generative chain. Erdős is the chain run socially:
The collaboration graph is the operator chain externalised: no single mind holds the whole traversal; the network does.
In 2026 something new attached to that graph. Two of Erdős's own problems were resolved with artificial intelligence — and one of them, Erdős Problem #728, came back not as prose but as a proof written in Lean and checked, line by line, by a machine kernel. The dream of The Book met a book a computer can actually read. Erdős, who valued a proof by who could see its inevitability, would have recognised the standard, even if the reader was no longer flesh.
What became of his problems when the collaborator turned machine — the human-checked construction versus the kernel-checked Lean proof, and why the difference is the whole point — is taken up in The Machine Collaborator in Vol V · The Seed.
He died in 1996, at a conference, still working. He had no house, no family of his own, no possessions to speak of — and a share in more theorems than almost anyone who ever lived. The suitcase was empty; the graph was not.