Erdős Problem #1196 — Primitive Sets Above a Threshold

At least 58 years old · documented by

Is it true that, for every xx, if A⊆[x,∞)A\subseteq [x,\infty) is a primitive set of integers—meaning that whenever a,b∈Aa,b\in A and a∣ba\mid b, the integers aa and bb are associated, hence equal—then

∑a∈A1alog⁡a<1+o(1),\sum_{a\in A}\frac{1}{a\log a}<1+o(1),

where the o(1)o(1) term tends to 00 as x→∞x\to\infty? Formally, does there exist a function o:N→Ro:\mathbb N\to\mathbb R with o=oat top(1)o=o_{\mathrm{at\ top}}(1) such that, for every natural number x>0x>0 and every primitive set A⊆{n∈N:x≤n}A\subseteq\{n\in\mathbb N:x\le n\},

∑a∈A1(alog⁡a)<1+o(x)?\sum_{a\in A}\frac{1}{(a\log a)}<1+o(x)?
References

Progress summary

Refreshed
Claimed solved

A 2026 paper proves the conjecture, with a stronger error bound, and a Lean formalization is also reported.

The Erdős–Sárközy–Szemerédi conjecture, recorded from 1966, asks whether every primitive set above xx has weighted reciprocal sum at most 1+o(1)1+o(1). The question is Erdős Problem #1196.

Known results

  • Lichtman (2023): upper bound eγπ/4+o(1)≈1.399+o(1)e^\gamma\pi/4+o(1)\approx1.399+o(1).
  • Lichtman (2020): for integers with exactly kk prime factors, a lower bound with error O(k−1/2+o(1))O(k^{-1/2+o(1)}).
  • Gorodetsky, Lichtman, and Wong (2024): asymptotic deficit (c+o(1))k22−k(c+o(1))k^22^{-k}, with c≈0.0656c\approx0.0656.

2026 quantitative proof

Alexeev, Barreto, Li, Lichtman, Price, Shah, Tang, and Tao state Theorem 1.1: for every primitive A⊆[x,∞)A\subseteq[x,\infty), ∑a∈A1/(alog⁡a)≤1+O(1/log⁡x)\sum_{a\in A}1/(a\log a)\leq1+O(1/\log x). Their von Mangoldt-chain method proves the conjecture; a Lean formalization is reported separately. The method was suggested by GPT-5.4 Pro, while the paper presents the proof as the authors’ work.

Current status (as of May 2026): The conjecture is settled by the quantitative theorem 1+O(1/log⁡x)1+O(1/\log x), with a reported Lean formalization.

Sources

Solutions 0

No solutions have been posted yet.