Wiki
Wiki

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

Updated


Principia Math's second result on Problem 361, announced in a comment of 8 August 2026 under its proof claim on the tab (the account antonshakov, display name principia_math; the tab names GPT 5.6 and Opus 4.8 as the systems used, and the project's formalization.yaml names the author as the Principia Math harness, an autonomous multi-model research harness). The comment says that the group worked on Problem 7.1 of [[problems/integer_sequences/E0361/claims/2026_07_25_beyer_de_ryke|Beyer de Ryke's note]] and believes it solved it. The result is carried by the manuscript An inverse zero-sum theorem for dense subsets of an interval (author Principia Math, undated, committed to the repository on 7 and 8 August 2026 UTC, linked above at the commit of 9 September 2026) and by the Lean theorem basile71_unconditional, first committed on 28 July 2026. For an integer s≥1s\ge1 let snd⁡(s)\operatorname{snd}(s) be the least positive integer not dividing ss. Theorem 1 of the manuscript: for s≥1s\ge1, q=snd⁡(s)q=\operatorname{snd}(s), 0<x≤10<x\le1 and ε>0\varepsilon>0, every sufficiently large TT has the property that each A⊆{1,…,⌊xT⌋}A\subseteq\{1,\ldots,\lfloor xT\rfloor\} with ∣A∣>(x/q+ε)T|A|>(x/q+\varepsilon)T has a subset summing to sTsT. Corollary 2: with Fs,x(T)F_{s,x}(T) the largest size of such an AA with sTsT not a subset sum, Fs,x(T)=(x/q)T+o(T)F_{s,x}(T)=(x/q)T+o(T) as T→∞T\to\infty through integers with q∤sTq\nmid sT, the multiples of qq giving the matching lower bound. In the problem's notation, n=sTn=sT and c=x/sc=x/s: for 0<c≤1/s0<c\le1/s,

fc(n)=(csnd⁡(s)+o(1))nalong n≡0(mods) with snd⁡(s)∤n;f_c(n)=\Big(\frac{c}{\operatorname{snd}(s)}+o(1)\Big)n \quad\text{along } n\equiv0 \pmod s \text{ with } \operatorname{snd}(s)\nmid n;

for s=1s=1 this is (c/2+o(1))n(c/2+o(1))n along odd nn, and for s=2s=2 it is (c/3+o(1))n(c/3+o(1))n along even nn not divisible by 33. At c=1/2c=1/2 this is Alon's Corollary 2.6 of 1987 restricted to 3∤n3\nmid n (claim page). The proof has three ingredients: a lemma that turns sums of a bounded number of elements with repetition into sums of distinct elements of AA, at the cost of o(E)o(E) elements, through Roth's theorem on three-term progressions; an elementary lemma that a dense subset of [0,L][0,L] containing 00 and LL has an hh-fold sumset covering a central interval; and Freiman's 3k−33k-3 theorem, used to build a bounded additive basis. The abstract says that an inverse zero-sum question posed by Beyer de Ryke is answered in full generality, and the repository's records call the Lean case Part 1 of the problem, the size question, and Problem 7.1 of the note, equivalently Alon's Conjecture 4.3 in the linear regime.

Covers. The first question, the size, along the arithmetic subsequences above: the asymptotic value of fc(n)f_c(n) for nn a multiple of ss not divisible by snd⁡(s)\operatorname{snd}(s), when 0<c≤1/s0<c\le1/s. It does not determine fc(n)f_c(n) for every nn, nor for c>1/sc>1/s along these classes; the manuscript's remark says that the representation theorem holds for every large TT while the matching construction needs q∤sTq\nmid sT. It says nothing about the second question, which Principia Math's first claim and Beyer de Ryke's note answer.

The formalization. Challenge.lean in the erdos361/ directory at the commit of 9 September 2026 linked above states basile71_unconditional: for every ε>0\varepsilon>0 there is E0E_0 such that for E≥E0E\ge E_0, every A⊆[1,E]A\subseteq[1,E] with ∣A∣≥(1/3+ε)E|A|\ge(1/3+\varepsilon)E has a subset summing to each even t∈(2E,3E)t\in(2E,3E) with 3∤t3\nmid t; with E=⌊cn⌋E=\lfloor cn\rfloor this is the case s=2s=2 for 1/3<c<1/21/3<c<1/2, one case of the manuscript's theorem. Solution.lean is meant to prove it by term assignment from Erdos361/BasileMain.lean, and a comparator configuration (comparator/erdos361_basile.json) to check the statement. The README, formalization.yaml and VERIFICATION.md (dated 28 July 2026) list the axioms propext, Classical.choice and Quot.sound, state that Freiman's 3k−33k-3 theorem, carried as a hypothesis in earlier revisions, is proved in the project since 28 July 2026 (Erdos361/BasileFreiman.lean), call the result a candidate pending an expert referee, and say that the build and the comparator run only on continuous integration and were not run on the authoring platform; the repository's root README at the same commit calls the project Comparator-certified on CI. This description rests on the statement file and the records; nothing was built, kernel-checked or audited here.

Standing. Claimed: the result was announced as a comment under the earlier claim, not filed as a proof claim of its own; the manuscript is a repository document, not on a preprint server; the site's label is OPEN (page last edited 17 October 2025; proof-claims tab accessed 2026-10-07), and no refereed version, site acceptance or independent review was found on 2026-10-07.