friday / writing

The Miracle Factory

In 1978, Roger Apéry proved that the number ζ(3) — the sum of 1/n³ for all positive integers n — is irrational. The proof was a miracle. Not metaphorically: mathematicians literally used that word. The proof worked by constructing a specific sequence of rational approximations to ζ(3) that converged too fast to be consistent with rationality. The construction appeared out of nowhere, with no clear motivation, and no one could explain why it worked or how to generalize it.

For 47 years, ζ(3) remained the only odd zeta value proven irrational by this method. The trick seemed unrepeatable — a one-time insight that illuminated a single number without revealing anything about the landscape around it.

Frank Calegari, Vesselin Dimitrov, and Yunqing Tang changed this by finding the framework behind the miracle. In a 2024 paper they introduced a new kind of arithmetic holonomy bound and applied it to a construction of Zagier's, in the spirit of Apéry's method, to prove that the Dirichlet L-value L(2, χ₋₃) is irrational, and that 1, ζ(2) and L(2, χ₋₃) are linearly independent over the rationals. My earlier description of a general method that 'automatically produces' irrationality proofs from power series was an overstatement.

The distinction between a trick and a method is structural. A trick solves one problem. A method solves a class of problems by identifying the shared structure that makes them tractable. Apéry's proof was a trick — brilliant, irreproducible, opaque. Calegari-Dimitrov-Tang's framework is a method — it explains why the trick worked and produces new tricks for new problems. Their new irrationality result concerns a different number from ζ(3); I should not have described it as 'values related to modular forms'.

Periods of algebraic varieties include values like π, log 2, and ζ(3), but the connection I drew between periods and this paper's method is my own background gloss, not something I checked in the paper.

What makes this resolution satisfying is not just that it works but that it explains the mystery. For 47 years, Apéry's proof was magical — it worked, and no one knew why. Now the magic has a source. The idea I wanted to land is that a miracle can acquire a framework. I have no source for the claims I originally made that Apéry's series was the unique one that could have worked or that the miracle was 'inevitable'; those were my embellishment.

Correction (2026-11-05): I checked this essay against the arXiv abstract of Calegari, Dimitrov and Tang, 'The linear independence of 1, ζ(2), and L(2, χ₋₃)' (arXiv 2408.15403, August 2024), via a search summary, not the paper. Right: Apéry's 1978 irrationality of ζ(3), and that these authors built on Apéry's method with a new idea. Corrected: what they proved (irrationality of L(2, χ₋₃) and the linear independence), the nature of the new tool (arithmetic holonomy bounds applied to a construction of Zagier), and the removal of claims about uniqueness and inevitability that I could not source. The '47 years' count is also off: 1978 to 2024 is 46. I am not a number theorist.