Erdős Problem #370
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (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
Let \(P(r)\) mean the largest prime factor of an integer \(r\). Erdős Problem 370 asks whether there are infinitely many positive integers \(n\) for which both consecutive numbers are unusually smooth:
Here “smooth” means that no prime factor is too large. The strict inequality matters. A number such as \(12=2^2\cdot3\) has \(P(12)=3<\sqrt{12}\), whereas a prime number \(q\) has \(P(q)=q\), so it cannot pass the test. The two conditions must hold for the same \(n\), not merely for infinitely many unrelated integers.
The original source is P. Erdős and R. L. Graham, Old and New Problems and Results in Combinatorial Number Theory (L’Enseignement Mathématique, Monographie 28, 1980), printed page 69. Their wording is the question above. The source discusses a density observation of Pomerance and says that very little is known. The maintained problem page later records an elementary construction attributed to Stefan Steinerberger, which changes how the literal wording should be read: as written, the problem has a sparse infinite family.
A small example
Take \(m=9\) and \(n=m^2-1=80\). Direct factorization gives
Thus \(P(80)=5\) and \(P(81)=3\). Exact integer arithmetic verifies
Since both sides are nonnegative, these squared inequalities are exactly the desired comparisons with the square roots. So \(n=80\) is a genuine finite witness. It does not, by itself, establish infinitude, and a finite search cannot decide a statement quantified over all integers.
The example also shows the mechanism. The identities \(80=(9-1)(9+1)\) and \(81=9^2\) place the prime factors of \(n\) and \(n+1\) below the relevant square-root scales. The choice \(m=9\) is especially convenient because \(8\), \(9\), and \(10\) are all composite.
What is known
The maintained entry at the frozen catalogue commit 536408716724a2402bdfa7c1d764bf83d64a8c89 marks Problem 370 “proved” and reports a Lean formalization. Its exact YAML entry is Git blob 4e60263a80af6207aca494d2e1388e44fa372889, last updated there on 2025-11-24. The live catalogue revision used in this audit is commit 503d0a58ff3b6a12d7ff1665aa37fb953416d85f, whose entry blob is 97ae243a7fc056bb9c6413ce665b575eadd27ff3. The status agrees with the frozen packet, but catalogue status is evidence about the record, not GCL mathematical certification.
The live problem page gives the construction \(n=m^2-1\). If \(m\) and \(m+1\) are composite, then the largest prime factor of \(m^2\) is strictly less than \(m=\sqrt{m^2}\), while every prime factor of \((m-1)(m+1)\) is at most \(m+1\) and, with the composite choice, below \(\sqrt{m^2-1}\) after the elementary factor bounds are applied. A uniform choice is \(m=k!+3\) for every \(k\ge3\): \(m-1\) and \(m+1\) are even and composite, and \(m\) is divisible by \(3\) and larger than \(3\). This supplies distinct values \(n=m^2-1\) without relying on a search.
The original Erdős–Graham document remains important because it fixes the historical statement and its intended context. The later explanation may indicate that the printed problem was mistyped or that a stronger density version was intended, but that speculation does not alter the registered literal question.
GCL's current state
The source packet’s protected FormalConjectures reference is google-deepmind/formal-conjectures@85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/370.lean, Git blob 13bf5540ba08c5e6734a0b2f86f0d9f32395a15d. Its theorem has the shape
The declaration ends in by sorry; its annotation points to an external proof source. The helper definition gives the junk value \(0\) for the largest prime factor of \(0\) and \(1\), while the historical question concerns positive integers where the notion is meaningful. This finite-domain convention does not obviously affect an infinitude statement, but it is a semantic detail that should be retained in any formal comparison.
The current external Lean source cited by the page contains a longer construction proof and has source blob 661ac044731ecb32dcb8cb405ee0b2508cc7f813. I did not run Lean or treat that source as a GCL certificate. The live page has three discussion comments, no proof exposition, and zero submitted proof claims. Therefore the editorial state is a documented source resolution with an explicit verification boundary, not an independent mathematical adjudication.
What remains
For the literal statement, the useful remaining work is provenance and semantic maintenance rather than an open search for examples. A bounded successor could compare the elementary construction line by line with the formal theorem, prove the composite-number lemmas in a pinned Lean environment, and record whether the imported Nat.maxPrimeFac convention is harmless. It should also check whether the intended historical problem was a density or longer-block variant, since the original page links Problems 369 and 928 as related material.
No such variant should be silently substituted for Problem 370. In particular, the original source’s density remark and the live page’s warning about a possible typographical error are context, not a replacement theorem. The V2 ledger alone decides accepted coverage; this article does not decide that coverage.
Sources and claim boundaries
- Original GCL issue #1640, assignment
ERDOS-CATALOGUE-0370-S01; its original SOURCE RESULT/1 citation is authenticated original RESULT/1. - Erdős–Graham, 1980 source PDF, printed p. 69; retrieved PDF SHA-256
0cbf0c32f0ab1e1c71db5121a88bac905bf976c4a6ab6bb6d7d9cf9ddd184ed3. - Maintained Problem 370 page and its discussion, accessed 2026-10-13; comments are reports and suggestions, not verified proof expositions.
- Frozen catalogue entry and pinned formal declaration.
The exact witness \(80\) was checked by integer factorization above. The construction, catalogue labels, discussion comments, and Lean declaration support a status report and a reusable research interface. They do not by themselves provide GCL certification, publication authority, or an accepted-coverage decision.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1640 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
- 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.