Skip to content

Erdős Problem #380

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

Erdős Problem #380 asks how often a product of consecutive integers has an unusually repeated largest prime factor. For positive integers \(u\le v\), write

\[ Q(u,v)=\prod_{u\le m\le v}m. \]

Call \([u,v]\) bad when the largest prime factor of \(Q(u,v)\) occurs with exponent at least two. Equivalently, if \(L(u,v)\) is that largest prime factor, then \(L(u,v)^2\mid Q(u,v)\). Let \(B_{\le x}\) be the set of integers \(n\le x\) that lie in at least one bad interval. For \(n\ge2\), let \(P(n)\) be the largest prime factor of \(n\); the comparison set is

\[ S_{\le x}=\{n\le x:P(n)^2\mid n\}. \]

The problem asks whether, as \(x\) tends to infinity, the two counts are asymptotic:

\[ \#B_{\le x}\sim\#S_{\le x}. \]

The interval is inclusive. The original wording is naturally normalized by putting \(N=u-1\) and \(H=v-u+1\), so Tao's notation \(\{N+1,\ldots,N+H\}\) describes exactly the same objects. The singleton interval is allowed; the product \(1\) is not treated as bad because it has no largest prime factor.

A small example

Take the interval \([24,25]\). Exact multiplication and factorization give

\[ 24\cdot25=600=2^3\cdot3\cdot5^2. \]

The largest prime factor is \(5\), and it occurs twice, so this interval is bad. Consequently \(24\) and \(25\) belong to \(B_{\le x}\) for every \(x\ge25\). The interval \([8,9]\) is even stronger:

\[ 8\cdot9=72=2^3\cdot3^2, \]

so every prime occurring in the product occurs at least twice; this is called very bad in the later literature.

The following finite check was run with exact integer arithmetic: factor every product \(Q(u,v)\) for \(1\le u\le v\le25\), discard the empty-factorization case \(Q(1,1)=1\), and union the elements of the bad intervals. It returns

\[ \{4,8,9,16,18,24,25\}. \]

This is a check on intervals wholly contained in \([1,25]\), not the complete value of \(B_{\le25}\), because an interval containing a small integer could extend beyond \(25\). It illustrates the distinction between a finite witness and an asymptotic assertion: finding bad intervals at small scales neither proves the limit nor checks Tao's analytic estimates.

What is known

The maintained Erdős Problems page records that Erdős and Graham posed the question in Old and new problems and results in combinatorial number theory (1980), page 73. It reports that Terence Tao resolved it in Products of consecutive integers with unusual anatomy, arXiv:2603.27990v3 (2026).

Tao uses \(\mathcal B\) for the set of integers lying in at least one bad interval and \(\mathcal B^1\) for the simpler set of integers divisible by the square of their own largest prime factor. Theorem 1.7, on page 6 of the version dated 26 September 2026, proves that the contribution from \(\mathcal B\setminus\mathcal B^1\) is smaller than \(\mathcal B^1\) by a factor \((\log x)^{1-o(1)}\). In particular,

\[ \#(\mathcal B\cap[1,x])\sim\#(\mathcal B^1\cap[1,x]). \]

That is the registered question after the notation is aligned. The result is an external published-preprint claim and source report; this article does not independently reproduce its 42-page proof. Tao's paper also treats the related very-bad intervals, but those are additional statements rather than a change to Problem #380.

GCL's current state

The original GCL issue #1650, assignment ERDOS-CATALOGUE-0380-S01, is the source-audit record for this article. Its frozen packet points to teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, entry number: "380", Git blob 4e60263a80af6207aca494d2e1388e44fa372889. That snapshot says informal_status: proved, last updated 2026-03-31, while formal_status is unformalized and formalized is no.

The live page was edited 10 April 2026 and now includes Tao's theorem and citation. Thus the snapshot status and the live explanatory record are consistent on the mathematical outcome but differ in detail and date. The exact pinned FormalConjectures provider snapshot FC-GDM-85F86371, upstream revision 85f863718beeec7b58a3a1926ee92e3472bc2020, was searched in both admitted catalogue shards; no ErdosProblems/380 statement file occurs. There is consequently no pinned Lean theorem, with or without sorry, for GCL to report here.

The catalogue and live page are evidence, not GCL mathematical certification. V2 ledger state alone decides accepted coverage. The original SOURCE RESULT/1 citation is authenticated original RESULT/1.

What remains

For the mathematics, the registered yes-or-no question has an affirmative answer in Tao's source. The bounded remaining work is documentary and formal-interface work: preserve the inclusive interval convention, the largest-prime-factor condition, the \(n\ge2\) convention on the comparison set, and the limit quantifier when translating the result. A future formalization would also need definitions for finite products, prime-factor exponents, and asymptotic counting functions before it could express Theorem 1.7.

A useful small successor is to formalize or independently verify the elementary family \([p^2-1,p^2]\) for odd primes \(p\), then record exactly which part is a finite lemma and which part still depends on Tao's analytic proof. That task can improve the interface without pretending to replay the asymptotic theorem.

Sources and claim boundaries

The article distinguishes Tao's externally reported proof, the maintained catalogue's status label, the finite arithmetic check above, and GCL's separate certification process. None of the last three is a substitute for reading or certifying the primary proof.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1650 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: OUTSIDE_STRICT_OPEN_TRIAGE.
  • 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.

Back to catalogue · Compendium introduction