Erdős Problem #438
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: solved (not automatically the live result).
Subject areas: number theory.
Read this status correctly
The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.
The question in plain language
Erdős Problem 438 asks how large a set \(A\subseteq\{1,\ldots,N\}\) can be if none of the sums \(a+b\) with \(a,b\in A\) is a perfect square. The definition of \(A+A\) includes sums formed using the same element twice; ruling out only sums of distinct elements would be a different extremal problem. The relevant objective is to maximize \(|A|\), not to count the number of non-square sums. The sharp limiting upper density is \(11/32\), and the difficult part is showing that a modular construction cannot be improved asymptotically by a nonperiodic set.
This question is a particularly transparent example of how a local modular obstruction can govern a global extremal density. Square integers occupy only a restricted set of residue classes modulo any fixed modulus. If every sum of two admissible residue classes avoids those square residues, selecting integers from those classes produces a genuinely square-sum-free set of positive integers. But proving that no other construction has a larger leading density calls for additional mathematical arguments beyond checking finitely many residues.
A small example
The simplest example takes all positive integers congruent to \(1\) modulo \(3\). Every pair sum is congruent to \(2\) modulo \(3\), while every square is congruent to \(0\) or \(1\). This creates square-free sums with limiting density \(1/3\). For a better example, consider the eleven residue classes \(R=\{1,5,9,13,14,17,21,25,26,29,30\}\) modulo \(32\), as observed by Massias.
The square residues modulo \(32\) are exactly \(\{0,1,4,9,16,17,25\}\). I independently checked all \(121\) ordered pairs \((a,b)\in R^2\); none of the residues \((a+b)\bmod32\) belongs to the square list. Hence \(A=\{m\geq1:m\bmod32\in R\}\) has no square in \(A+A\), and \(|A\cap[1,N]|=(11/32)N+O(1)\). This is a complete exact construction. The \(O(1)\) accounts for the unfinished last block of \(32\) integers, and does not obscure the leading constant.
What is known
The problem appears in several writings of Erdős and in Erdős–Graham's Old and New Problems and Results in Combinatorial Number Theory (1980), printed page 87. The maintained Erdős Problems #438 page describes the modular construction and credits Massias with the eleven classes. Lagarias, Odlyzko and Shearer (1983) showed that this proportion is optimal for the relevant modular problem. They also established a general upper bound of approximately \(0.475N\) for the integer problem.
The global theorem came from Khalfalah, Lodha and Szemerédi (2002), who established that the largest square-sum-free subset of \([1,N]\) has size \(\left(11/32+o(1)\right)N\). The matching upper bound shows that no infinite sequence of finite sets can maintain a greater limiting proportion than the modular construction. This does not claim an exact formula such as \(\lfloor11N/32\rfloor\) for every individual \(N\).
There is a relationship with Erdős #439, a chromatic problem asking whether every finite coloring contains a monochromatic square-sum pair. One extremal bound on a single large square-sum-free set is not by itself an immediate proof of the all-colorings statement; the stronger Ramsey conclusion requires separate reasoning.
GCL's current state
The original GCL source issue #1708 maps to the fixed teorth/erdosproblems revision 536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889. Its source packet classifies the problem as solved and records no #438 statement file in the admitted FormalConjectures provider revision 85f863718beeec7b58a3a1926ee92e3472bc2020. The absence of that pinned file is an evidence limitation, not a reason to disregard the established published extremal theorem.
The authenticated issue-specific source return is https://github.com/grandchallenge/MATH-PROGRAMME/issues/1708#issuecomment-6158423003. That return reports the mathematical bounds and exact finite residue audit but does not confer GCL formal certification. Accepted Compendium coverage remains controlled by the later protected V2 ledger, not by the presence of a staging manuscript or an original issue comment.
What remains
A bounded independent regression should enumerate the square classes modulo \(32\) and verify every sum from the eleven allowed residue classes avoids them. A second program should search for maximum square-sum-free subsets of \([1,N]\) for \(N\leq20\) by backtracking, testing diagonal sums as well as sums of distinct values. Both computations use exact integer square roots and must state their complete finite search space. This establishes a trustworthy computational interface; it cannot prove the asymptotic bound.
The mathematical source review should inspect the Lagarias–Odlyzko–Shearer modular theorem and the Khalfalah–Lodha–Szemerédi global upper theorem. Record the exact quantifiers on \(N\), the error term, and whether diagonal sums are excluded in either source. Acceptance requires a valid modular certificate, two matching finite search methods, and a citation-grounded upper bound. A residue collision, a hidden assumption about diagonal pairs, or conflation of finite exact maxima with leading asymptotic density is a falsification. Stop at the bounded tests and documentary comparison.
Sources and claim boundaries
Historical question: Erdős–Graham (1980), printed p.87. Primary resolving line: Lagarias–Odlyzko–Shearer (1983) and Khalfalah–Lodha–Szemerédi (2002), catalogued at Erdős Problem #438. Related #439. Frozen GCL source issue #1708, teorth catalogue, and authenticated audit.
The residue calculation is independently exact and the asymptotic theorem is credited to its published authors. The article does not claim mathematical novelty, a GCL-replayed formal proof, or protected editorial admission.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1708 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.