The Flint Hills series is the series
(Pickover 2002, p. 59). Convergence has not been established, since can have sporadic large values. The plots above show
its behavior up to
.
The positive integer values of
giving incrementally largest values of
are given by 1, 3, 22, 333, 355, 103993, ... (OEIS A046947), which are precisely the numerators
of the convergents of
, corresponding to the values 1.1884, 7.08617, 112.978, 113.364,
33173.7, ....
Alekseyev (2011) has shown that the question of the convergence of the Flint Hills series is related to the irrationality
measure of ,
and in particular, convergence would imply
, which is much stronger than the best established
upper bound.
Meiburg (2022) proved the near-converse that implies convergence.
The boundary case
is not settled by these implications.
A purported proof of convergence was given by Mantzakouras and López Zapata (2025). López Zapata (2026) withdrew the manuscript on Sep. 14, 2026 after identifying serious errors in the argument.
OpenAI (2026) announced and, as a consequence, convergence
of the Flint Hills series. The release attributes the proof
to an unreleased internal OpenAI model and supplies a Lean formalization of the claimed
irrationality measure. As of Oct. 7,
2026, the formal statement had been read by VibeMathed (2026), but an independent
rebuild and specialist verification of the complete proof
had not been reported.