Erdős Problem #334
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (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
For an integer \(n\), call a positive integer smooth at level \(y\) if every prime dividing it is at most \(y\). Erdős Problem 334 asks how small one can make a common smoothness bound while still representing every sufficiently large \(n\) as
with both \(a\) and \(b\) having no prime divisor exceeding \(f(n)\). Equivalently, if \(P^+(m)\) denotes the largest prime factor of \(m\), with \(P^+(1)=1\), the target is to understand the least scale of \(f(n)\) for which
can always be arranged. The quantifier “every sufficiently large” matters: checking many values of \(n\) cannot by itself establish the asymptotic assertion.
The original source phrases the challenge through the least integer not representable with prime factors below a given bound. In the source’s notation, Erdős conjectured that this least exception eventually exceeds every fixed power of the bound. The modern problem page presents the equivalent smooth-summand formulation and records the historical \(n^{1/3}\) question.
A small example
Here is an exact finite check, included to make the definitions concrete. Direct trial division of every summand pair \(a,n-a\) for \(2\leq n\leq20\) gives the least common smoothness bound
in order for \(n=2,3,\ldots,20\). For example, the entries requiring \(3\) can be witnessed by
The factors are explicit: \(9=3^2\), \(12=2^2\cdot3\), and \(18=2\cdot3^2\), while \(1\), \(2\), and \(3\) have no prime factor above \(3\). The same exact enumeration supplies the other entries; for instance \(20=4+16\) has smoothness bound \(2\). This verifies a finite lemma, namely that every integer from \(2\) through \(20\) has a representation with bound at most \(3\). It does not establish a bound for all larger integers, identify the optimal function, or reproduce Balog’s asymptotic argument.
What is known
The primary historical source is Erdős, “Some new problems and results in number theory,” Number Theory (Mysore, 1981), Lecture Notes in Mathematics 938 (1982), p. 55. Its problem 4 asks about the least exception for binary sums and conjectures growth beyond every fixed power. The retrieved PDF has SHA-256 e3fb7fca8b27e2fb8ca63f4063a61a1603ab368c41e420310ce2159842930622.
The principal verified theorem is A. Balog, “On additive representation of integers,” Acta Mathematica Hungarica 54 (1989), 297–301, especially the theorem on p. 297. For each fixed \(\varepsilon>0\), Balog proves that every sufficiently large \(N\) can be written as \(N=a+b\) with
The exponent is approximately \(0.2695691821\), so this is stronger than the originally mentioned \(N^{1/3}\) scale. The proof uses a deep distribution result of Fouvry for sequences in residue classes with large moduli; Balog explicitly says that the method does not reach even an \(N^\varepsilon\) bound in the conjectural subpolynomial direction. The complete volume scan used here is available from REAL-J, with SHA-256 12db2f333c5cd10c4f50dd152fed0a13a9d0d1af42c2da82fff83c0f7c935095.
Thus the external record supports a proved polynomial upper bound and an open improvement problem. The live page also says that \(f(n)\leq n^{o(1)}\) is considered likely, but that sentence is a conjectural outlook, not a theorem. Its two forum comments are user reports and explicitly not verified proof expositions; they do not change the dated status.
GCL's current state
Issue #1604 pins the catalogue commit 536408716724a2402bdfa7c1d764bf83d64a8c89 and the problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889. At that frozen revision, problem 334 is marked open, last updated 2025-08-31, with formal status unformalized and statement-formalized metadata no. The maintained live entry agrees on open status and was last edited 3 April 2026; the retrieved HTML has SHA-256 a4608c0a2c007c94e0ca68b9ed518a0ba99e447950908b5aef1b784f8f8f6a32.
The live entry cites Erdős’s 1976, 1980, and 1982 sources and Balog 1989, and reports no proof exposition and no proof claim. Its displayed current statement is the binary smooth-summand question above. No accepted coverage is inferred from these pages: V2 ledger state alone decides accepted coverage. The original issue had zero GitHub comments when audited, so the original SOURCE RESULT/1 citation is https://github.com/grandchallenge/MATH-PROGRAMME/issues/1604#issuecomment-6138460562.
No pinned FormalConjectures statement file was available in the issue’s named FC-GDM-85F86371 provider snapshot. That identifier is not a Git commit accepted by the upstream repository, and FormalConjectures/ErdosProblems/334.lean is absent at the checked upstream head. Consequently there is no Lean theorem, by sorry source, or formal semantic comparison to report.
What remains
The smallest useful successor is a source-locked normalization and bounded formalization-preparation task. It should first fix whether \(a,b\) range over positive naturals, define \(P^+(1)\), and choose one exact asymptotic statement: either Balog’s theorem with its \(\varepsilon\) and sufficiently-large threshold, or Erdős’s least-exception conjecture. It should then translate only that statement into a theorem interface and test the translation against the wording and page locations in the two primary sources.
Acceptance should require a line-by-line quantifier record, a source quotation or transcription checked against the scans, and an executable finite checker for a declared range. Falsification is any mismatch in positivity, the product condition, the order of quantifiers, or the exponent. The task should stop when the interface and finite checker agree with the sources, or immediately when the primary statement cannot be read unambiguously. A formal theorem proved with sorry would be a source artifact, not a proof of the mathematics.
Sources and claim boundaries
The source lock for this article is the original issue, the frozen catalogue commit/blob above, the live Erdős Problems entry, Erdős’s 1982 source, and Balog’s 1989 scan. The issue URL is the audit container, not mathematical certification. Balog’s theorem is reported as an external published result verified against the scan; no independent reconstruction or Lean replay is claimed. The finite calculation is an exact local check only. Live-page commentary, catalogue metadata, and absent formal files establish status and boundaries, not new mathematics.
Original SOURCE RESULT/1 citation: https://github.com/grandchallenge/MATH-PROGRAMME/issues/1604#issuecomment-6138460562.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1604 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: STATEMENT_CAPTURE_REQUIRED.
- 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.