Erdős Problem #382
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
Take a consecutive block of positive integers from \(u\) through \(v\), and multiply the block. Let \(p\) be the largest prime that divides this product. The block is “bad” when \(p\) occurs at least twice in its prime factorization, equivalently when \(p^2\) divides the product. Erdős Problem 382 asks two questions about the length \(v-u\) of such a block:
Here \(v^{o(1)}\) means that, for every fixed \(\varepsilon>0\), all sufficiently large relevant endpoints satisfy \(v-u\leq v^\varepsilon\). The natural normalization used by the current formal statement is \(1\leq u\leq v\) and \(v\geq2\). Writing \(P(u,v)=\prod_{m=u}^{v}m\), the hypothesis is that the largest prime divisor \(P^+(P(u,v))\) satisfies
The first question is an asymptotic upper bound on every bad interval. The second is an existence statement: for every \(k\), is there a bad interval with \(v-u\geq k\)? These are different quantifier patterns; evidence for one does not settle the other.
A small example
The smallest useful worked witness in the formal source is \(u=48\), \(v=50\). Exact integer arithmetic gives
Thus the largest prime divisor is \(7\), its exponent is \(2\), and \(7^2\) divides the product. The interval has width \(v-u=2\). This calculation was checked directly by factoring the integer \(117600\) and testing the remainder modulo \(49\); the remainder is \(0\).
This is a finite certificate that the hypothesis is nonempty, not evidence for either asymptotic answer. A finite search can find witnesses or disprove a proposed small bound, but cannot prove a statement quantified over all sufficiently large \(v\) or over every \(k\). The maintained discussion also records larger numerical examples supplied by Stijn Cambie, including a width of \(13\); those reports are useful leads, but the present article treats the exact \((48,50)\) calculation as the verified finite example.
What is known
The accessible primary source is P. Erdős and R. L. Graham, Old and new problems and results in combinatorial number theory (1980), whose scan is linked below. On printed p. 73 they define an integer as “bad” when it belongs to an interval \(u\leq n\leq v\) for which the largest prime factor of \(\prod_{i=0}^{v-u}(u+i)\) occurs with exponent greater than \(1\). They then say that results of K. Ramachandra imply that the interval length is at most \(v^{1/2-\varepsilon}\) for some positive constant \(\varepsilon\). The same page says it is expected to be bounded by \(v^\varepsilon\) for every \(\varepsilon>0\), and that it can be arbitrarily large, but states that the latter assertion had not been proved.
This primary page is stronger than the maintained page’s current summary, which reports the weaker-looking bound \(v-u\leq v^{1/2+o(1)}\). The discrepancy is preserved as a source wording issue: it does not change the fact that both the \(v^{o(1)}\) assertion and the arbitrary-width assertion are presented as open. Ramachandra’s cited paper is K. Ramachandra, “A Note on Numbers with a Large Prime Factor,” Journal of the London Mathematical Society (2) 1 (1969), 303–306, DOI 10.1112/jlms/s2-1.1.303. Its bibliographic record was verified, but the publisher PDF returned HTTP 403 in this audit, so Ramachandra’s proof and exact constants were not independently read.
The maintained page also explains a conditional route through prime gaps: a sufficiently strong short-interval prime statement would force a bad interval to be short. That is a conditional reduction or heuristic, not an unconditional solution. Likewise, the page’s positive-probability discussion near squares of primes is a heuristic for large widths, not a proof of the second question.
GCL's current state
The original issue is MATH-PROGRAMME #1652, assignment ERDOS-CATALOGUE-0382-S01. Its frozen catalogue packet identifies teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, blob 4e60263a80af6207aca494d2e1388e44fa372889, with source SHA-256 be65b0db56fe13becec3178f2937685ee7e6a2f5df9325712474722c279c4182. At that revision the entry says informal status open, last update 2025-08-31, formal status unformalized, and statement metadata formalized: yes.
The issue also pins the admitted Formal Conjectures provider FC-GDM-85F86371 at upstream commit 85f863718beeec7b58a3a1926ee92e3472bc2020. A lookup of FormalConjectures/ErdosProblems/382.lean at that exact commit returns no file, consistent with the issue’s explicit statement that the admitted snapshot contains no #382 statement. The later upstream file is documentary drift, not a replacement source lock. At current upstream commit e3e22360e1b917b0bc97cfcec70bfb0fe525b5d6, blob a9f327dd4aa4ffe700ddfb96526f707643f091f8, the file defines the natural-number predicate with explicit guards, records the \((48,50)\) test, and encodes both open parts plus the Ramachandra variant. The research declarations and the textbook inequality still end in by sorry; the completed finite test is not GCL mathematical certification.
What remains
A bounded successor task is to reconcile the exact source statement with the formal interface. Its input should be the frozen catalogue entry, Erdős–Graham p. 73, and the later upstream #382 file as a clearly marked non-admitted comparison. The target should be a source-faithful definition of \(P(u,v)\), largest prime divisor, and the multiplicity-two predicate, followed by a proof of the finite witness \((48,50)\) and a written comparison of the two asymptotic quantifier patterns.
Acceptance should require: exact agreement on endpoint conventions; a reproducible factorization check for \((48,50)\); explicit separation of the unconditional Ramachandra-reported bound, the two open assertions, and the Cramér-based conditional reasoning; and a record of the formal file’s sorry boundaries. Falsification is triggered by any endpoint convention that admits the degenerate products at \(0\) or \(1\), by confusing the largest factor with an arbitrary repeated factor, or by promoting the later formal file or a finite search to a proof. Stop when the semantic comparison and finite certificate are complete; do not launch a solver campaign or a Lean build merely to refresh stale provenance.
Sources and claim boundaries
- Erdős Problem 382 — maintained statement, dated open presentation, heuristics, and current formalization link.
- Erdős and Graham, 1980 scan, printed p. 73 — primary historical wording and the Ramachandra-reported upper bound.
- Ramachandra, 1969 — primary bibliographic record; proof text unavailable to this audit because the PDF request returned 403.
- Frozen catalogue commit and blob — provenance and status metadata, not proof authority.
- Later Formal Conjectures #382 file — formal statement and placeholders, not GCL certification.
- Original SOURCE RESULT/1 citation: authenticated original RESULT/1.
The article records evidence, finite verification, and limits of interpretation. V2 ledger state alone decides accepted coverage; neither the catalogue, the maintained page, the formal file, nor this article supplies GCL mathematical certification.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1652 — 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.