Erdős Problem #368
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 a positive integer \(n\), let \(F(n)\) be the largest prime divisor of the consecutive-integer product \(n(n+1)\). Erdős Problem 368 asks how small this largest prime can be as \(n\) grows. The motivating tension is simple: \(n\) and \(n+1\) cannot share a prime factor, but each might separately be built from relatively small primes. The question is therefore about the extreme smoothness of two consecutive integers at once.
The original formulation in Erdős's 1976 paper records two conjectural thresholds. For every positive \(epsilon\), infinitely many \(n\) should satisfy \(F(n)<(\log n)^{2+\epsilon}\), while eventually every \(n\) should satisfy \(F(n)>(\log n)^{2-\epsilon}\). The maintained problem page presents the same question as open. These are asymptotic statements: a handful of small values cannot settle either one.
A small example
Here is an exact check that can be reproduced using only integer arithmetic. Take \(n=4374\). Direct factorization gives
so
The largest prime factor is therefore \(F(4374)=7\). The same calculation gives \(F(8)=3\) from \(8\cdot9=2^3\cdot3^2\), and \(F(80)=5\) from \(80\cdot81=2^4\cdot3^4\cdot5\). These three identities were also retained as kernel-checked examples in the later Formal Conjectures file. They verify the definition and provide regression tests for a factorization program; they do not establish a bound for all sufficiently large \(n\), or the existence of infinitely many \(n\) in any asymptotic range.
What is known
Erdős's primary source, Problems and results on number theoretic properties of consecutive integers and related questions, states on printed page 27 that a classical result gives \(P(n(n+1))>c\log\log n\), and records the stronger \((\log n)^{2-\epsilon}\) lower-bound expectation and the infinitely-often \((\log n)^{2+\epsilon}\) upper-bound expectation. The paper is careful that the conjecture rests on probabilistic intuition, not a proof.
The strongest directly relevant primary result located in this audit is Hector Pasten's paper The largest prime factor of \(n^2+1\) and improvements on subexponential \(ABC\). Its Corollary 1.5 states that for coprime positive integers \(x<y\), the largest prime factor of \(xy(x+y)\) is at least a constant multiple of \((\log_2 y)^2/\log_3 y\) for sufficiently large variables. Setting \(x=1\) and \(y=n\) satisfies the hypotheses for \(n>1\) and gives the unconditional lower bound
This is a genuine external written proof, not a GCL result. It remains far below Erdős's conjectured order \((\log n)^2\). The live-page text also attributes earlier divergence, logarithmic-logarithmic bounds, and a subpower upper result to Pólya, Mahler, and Schinzel. Those attributions are recorded as source context here; this audit did not independently inspect all three older primary papers.
There is also a normalization issue worth keeping visible. The original paper expresses the conjecture with an arbitrary \(epsilon\) and an eventual quantifier, whereas the later Lean declaration uses one constant \(c\) and quantifies over every natural number with \(1\leq n\). The declaration is therefore useful as an interface for formal work, but it is not evidence that the intended asymptotic theorem has been proved, and the two formulations should not be called identical without an explicit equivalence argument.
GCL's current state
The pinned catalogue blob is 4e60263a80af6207aca494d2e1388e44fa372889 in teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89. Its entry for 368 says informal_status: open, formal_status: unformalized, and formalized: yes with metadata dated 2026-10-08. The protected GCL registry binds the problem to issue #1638, records catalogue status open, triage class STATEMENT_CAPTURE_REQUIRED, and protected_fc_present: false; the issue-body SHA-256 is 76e250cb2a4a67d857cd49a35945555dcafb2fb614b861162bd8de30ceecf5c9.
The later upstream Formal Conjectures commit b1f53fee831eff6853e3d82fc58b28a4f393e0bc adds FormalConjectures/ErdosProblems/368.lean, blob 5ae4205544786446db0875632ece319c28d55ebd. It defines \(F\) with Nat.maxPrimeFac, formalizes a universal constant lower bound and the infinitely-often upper bound, and includes the finite examples above. Both main theorem declarations and the Pólya/Schinzel variants end with sorry; the sample equalities use decide +kernel. Thus this is a formal statement source, not a checked proof. The issue's original protected provider snapshot explicitly said no such file was present, so the newer file is reported as drift, not silently treated as protected GCL evidence.
What remains
The smallest useful bounded successor is a source-locked semantic bridge: verify line by line that Pasten's Corollary 1.5 permits \(x=1\), \(y=n\), then record the resulting statement with its exact asymptotic threshold and no stronger quantifiers. As a finite companion, an exact-arithmetic checker could factor every \(n(n+1)\) for \(2\leq n\leq10000\), emit a factorization identity and primality certificate for every factor, and reproduce the three examples above. Acceptance would require all identities to multiply back and the source-hypothesis bridge to be correct; any mismatch or hypothesis failure would falsify the task. The run should stop at \(10000\), at the first failed identity, or when the source text needed for a claimed stronger result is unavailable. No finite run can decide the infinitely-often clause or the eventual lower bound.
Sources and claim boundaries
The main sources are the original issue #1638, the pinned catalogue entry, Erdős's 1976 paper, Pasten's arXiv paper and its published DOI, OEIS A074399, and the Formal Conjectures file at the exact later commit. Retrieved PDF SHA-256 values were 2c098f2c3e5079ef09a44f693c7db43a29428bb1c8bf7df7ab57b020306a9ad0 for Erdős and 7060cd7468344cc2dfdcf496b1f00d08b8a77c9e87dbf92f7f2135c466eab770 for Pasten.
Direct access to the maintained erdosproblems.com/368 page and its forum thread returned HTTP 403; indexed text reports the page as open and says no claimed solution is in the comments, but the exact live revision and all comment metadata could not be read back. No comment exists on the GCL issue at this audit. Original SOURCE RESULT/1 citation: authenticated original RESULT/1. This article reports external proofs, source metadata, finite arithmetic, and formal statements with sorry separately. It supplies no GCL mathematical certification; V2 ledger evidence alone decides accepted coverage.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1638 — 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.