← All problems

Erdős Problem #3 — Arithmetic progressions in sets with divergent reciprocal sum

52 years old · traced to

I conjectured long ago that if

∑i1/ai=∞(6)\sum_i 1/a_i = \infty \tag{6}

then the aia_i's contain arbitrarily long arithmetic progressions. If true this of course implies that there are arbitrarily long arithmetic progressions all whose terms are primes. I offer 3000 dollars for a proof or disproof of (6).

Equivalent formulations
  1. Let A⊂Z+A\subset\mathbb{Z^{+}} satisfy

    ∑n∈An−1=+∞.\sum_{n\in A}n^{-1}=+\infty.

    Erdős's conjecture. Then AA contains arbitrarily long arithmetic progressions. This generalizes the corresponding conjecture for the primes; the source does not state a resolution, so the problem remains open.

    source: Weiwen Zhang, “Roth-type theorems in additive combinatroics”, arXiv:2512.08455 (2025).

References

P. Erdős, A survey of problems in combinatorial number theory, Ann. Discrete Math. 6 (1980), 89-115.

Source: P. Erdős, A survey of problems in combinatorial number theory, Ann. Discrete Math. 6 (1980), 89-115.

Progress summary

Refreshed
Open

The conjecture remains open: a 2020 breakthrough established only three-term progressions, not progressions of every length.

Erdős conjectured that every set of positive integers with divergent reciprocal sum contains arithmetic progressions of arbitrarily large length, offering a $3000 prize. The conjecture is recorded in his 1980 survey.

Known results

  • Roth (1952): every set of positive density contains infinitely many three-term arithmetic progressions.
  • Szemerédi (1975): every set of positive density contains arbitrarily long arithmetic progressions.
  • Green and Tao (2004): the primes contain arbitrarily long arithmetic progressions, a special case of the conjecture’s consequence.

July 7, 2020 three-term result

Thomas Bloom and Olof Sisask reported a proof that divergent reciprocal sum implies infinitely many three-term arithmetic progressions. The full arbitrarily-long progression statement remains unresolved.

Community submission (unverified), October 7, 2026

A submitted claim says an OpenAI preprint, dated September 23, 2026, resolves the conjecture; no independent mathematical corroboration was found.

Current status (as of October 2026): The three-term case is reported proved, while the assertion of arbitrarily long progressions remains open; the submitted OpenAI claim is unverified.

Sources

Solutions 1

Claimed by OpenAI: Erdős’s reciprocal-sum conjecture on arithmetic progressions.See full solutionHide full solution

Claimed by OpenAI. Attached is OpenAI’s preprint, Quasipolynomial Bounds for Arithmetic Progressions (September 23, 2026).

Source: OpenAI math GitHub repository.

Original PDF: paper.pdf.

  • 159.pdf1,510,724 bytesOpen