Erdős Problem #326
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: number theory, additive basis.
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 326 asks for a very economical additive building set. Write its elements in increasing order as \(A=\{a_1<a_2<\cdots\}\). It should be a minimal basis of order \(2\): every sufficiently large natural number must be expressible as the sum of two elements of \(A\), while deleting any element must destroy that eventual covering property. The question is whether such an \(A\) can have a stable quadratic scale,
for a nonzero constant \(c\). The word “minimal” is doing substantial work. Without it, the problem is about constructing a thin basis with regular growth. With it, the same set must both cover all sufficiently large integers and have every element remain essential.
The live problem page gives this minimal-basis formulation and says that Erdős and Graham conjectured a negative answer. It also records that Erdős originally asked the broader question for an arbitrary basis, not necessarily a minimal one.
A small example
Here is an exact finite calculation illustrating the covering mechanism, not a solution of the infinite problem. In the Fibonacci instance of Cassels’s construction take \(q_1=q_2=1\), \(q_3=2\), and \(q_4=3\). The first two block lengths are \(m_1=q_2+q_3=3\) and \(m_2=q_3+q_4=5\), and the second block begins at \(Q_2=m_1q_1=3\). Thus
All computations are in the integers. Since each block is an interval,
Their union is exactly \(\{0,1,\ldots,16\}\). This is reproducible by listing the endpoint sums and observing that sums of two integer intervals are integer intervals. It demonstrates finite coverage and the useful overlap between adjacent blocks. It does not establish coverage beyond 16, minimality, or convergence of an infinite counting sequence. It also uses nonnegative integers, as does the standard construction; changing the convention for whether zero belongs to the natural numbers is a harmless issue for this finite illustration, but must be fixed in a formal statement.
What is known
The primary reference identified by the maintained page is J. W. S. Cassels, “Über Basen der natürlichen Zahlenreihe,” Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg 21 (1957), 247–257, DOI 10.1007/BF02941936. The University of Hamburg archive confirms the author, title, volume, year, and page range. The primary article’s landing page was reachable, but its full text was not readable in this audit because the publisher redirected the document to an access-controlled Springer identity flow.
An accessible secondary source is Melvyn B. Nathanson, “Cassels bases,” arXiv:0905.3144, especially Section 6. It gives the hypotheses for Cassels’s arithmetic-progression blocks, proves that the resulting infinite set is a basis of order \(2\), and states the consequence that there is an order-2 basis \(C=\{c_n\}\) with \(c_n=\lambda n^2+O(n)\) for some positive \(\lambda\). This verifies the broad, non-minimal construction through a readable exposition. It does not prove the minimal-basis question in the live entry, and no claim here treats the secondary exposition as a substitute for reading Cassels’s original proof.
GCL's current state
The protected source packet pins the catalogue at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, with catalogue blob 4e60263a80af6207aca494d2e1388e44fa372889. Its entry number: "326" says status: open, last update 2025-08-31, formal_status: unformalized, and formalized: yes as of 2025-09-01. The maintained page was last edited 17 April 2026 and now displays the minimal formulation, the Cassels historical note, and “Formalised statement? Yes”. That is live-source drift, not evidence that the mathematical problem has been solved.
The pinned formal provider is google-deepmind/formal-conjectures at commit 85f863718beeec7b58a3a1926ee92e3472bc2020; the file FormalConjectures/ErdosProblems/326.lean has blob 1b3712bcf1ea6d6f059281a257cc7428a42dc57a. Its open theorem quantifies over every exact order-2 additive basis and asks for a strictly increasing subsequence whose range is an additive basis of some order and whose quadratic ratio has no real limit. It has no minimality predicate and no explicit nonzero constant. Both that theorem and the documented positive-limit variant are proved by sorry. The formal file therefore records an intended interface, not a proof or semantic equivalence with the live question.
What remains
The smallest unresolved interface is a source-normalization comparison: determine whether the minimal-basis question in the live page is intended to be formalized, or whether the current Lean theorem is a different strengthening about basis-preserving subsequences. A bounded successor can generate the first eight Fibonacci Cassels blocks exactly, verify each finite two-sum interval by integer arithmetic, and test element-deletion essentiality against the covered finite tail. A failure is a concrete counterexample to the finite certificate; success is only a checked finite prefix. It cannot decide infinite minimality or the limit question.
The primary Cassels scan remains necessary before attributing a theorem, proof step, or page number directly to Cassels. Until that scan is available, the responsible status is open with partial source coverage. V2 ledger disposition, rather than this article, decides accepted coverage.
Sources and claim boundaries
The assignment record is MATH-PROGRAMME issue #1596; it had zero comments at audit time, so the original SOURCE RESULT/1 citation is authenticated original RESULT/1. The live entry is erdosproblems.com/326, and its displayed references are Ca57 and ErGr80. The frozen catalogue and formal blobs, and their hashes, are stated above so that later readers can distinguish pinned evidence from live drift.
No Lean replay, mathematical certification, publication decision, priority claim, or acceptance claim is supplied here. The finite interval calculation is independently reproducible arithmetic; the Cassels conclusion is an externally reported result verified through Nathanson’s accessible exposition; the minimal-basis problem remains an open claim in the maintained source; and the exact primary-paper proof remains unread in this bounded audit.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1596 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: PROTECTED_READY_BACKLOG.
- Formal-statement coverage: A provider statement was included in the pinned snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.