Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Claim. E. Szemerédi and V. Vu prove that there is an absolute constant cc such that every increasing sequence A={a1<a2<⋯ }A=\{a_1<a_2<\cdots\} of positive integers with A(n)≥c n1/2A(n)\ge c\,n^{1/2} for every nn, where A(n)A(n) counts the elements of AA not exceeding nn, is subcomplete: the set P(A)P(A) of finite subset sums of AA contains an infinite arithmetic progression. This answers the question of Problem 344 yes, in the reading in which the hypothesis ∣A∩{1,…,N}∣≫N1/2\lvert A\cap\{1,\ldots,N\}\rvert\gg N^{1/2} carries a sufficiently large implied constant; Folkman had proved it under the stronger hypothesis A(n)≥c n1/2+ϵA(n)\ge c\,n^{1/2+\epsilon} (his claim page). The first proof is in Finite and infinite arithmetic progressions in sumsets, Ann. of Math. (2) 163 (2006), no. 1, 1--35. The paper Long arithmetic progressions in sumsets: thresholds and bounds, J. Amer. Math. Soc. 19 (2006), no. 1, 119--169 (arXiv math/0507539, posted 2005-07-26; published online 2005-09-13), gives a second, shorter proof as its Theorem 9.4: its Section 9 records the statement as Erdős's conjecture of 1962 in the form Folkman's work led to (Conjecture 9.2), says that the authors proved it in the Annals paper and discuss it again for pedagogical reasons, and derives it from the paper's general sufficient condition for subcompleteness with one new lemma (Lemma 9.3), that the subset sums of any set of at least CnC\sqrt n distinct integers in {1,…,n}\{1,\ldots,n\} contain an arithmetic progression of length nn. The paper's other application is Folkman's conjecture on multisets, which is Problem 343.

The constant cannot be made arbitrary: Erdős's 1961/62 paper (card) shows that A(x)>Cx1/2A(x)>Cx^{1/2} does not suffice when C<2C<\sqrt2, and the sharper form with A(N)≥(2N)1/2A(N)\ge(2N)^{1/2}, which that paper shows would be best possible, is not claimed here and is recorded as open on the problem page. Y.-G. Chen proved the same theorem in 2003 by a different method; the closing remark of the JAMS paper's Section 9 says so, and his result has its own claim page.

Formalization. The repository plby/lean-proofs holds a Lean file for Problem 344 (the formalization link, pinned to its last change of 2026-09-06; first added 2026-08-21) whose header names Szemerédi and Vu as the informal authors and Codex and GPT-5.6 Sol as the formal authors. Its theorem erdos_344 states that there is a constant C>0C>0 such that every set AA of natural numbers with CN≤∣A∩{1,…,N}∣C\sqrt N\le\lvert A\cap\{1,\ldots,N\}\rvert for all sufficiently large NN has subset sums containing an infinite arithmetic progression, which is the absolute-constant reading of the problem. It was not built or audited by this corpus, so no formalized evidence is listed.

Acceptance. Refereed: the Annals of Mathematics and the Journal of the American Mathematical Society. Reviewed: the site's curator, T. F. Bloom, labels Problem 344 proved at erdosproblems.com and credits the resolution to the JAMS paper of Szemerédi and Vu (page last edited 28 December 2025, accessed 2026-10-07), and the community database lists the problem as proved. The thread carries no comments and no dispute. The library has no card for either paper; the citations above are the journals' records. Nothing here was checked by this project.