Skip to content

Erdős Problem #364

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: verifiable (not automatically the live result).
Subject areas: number theory, powerful.

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

A positive integer is called powerful when every prime that divides it occurs at least twice in its prime factorisation. Thus \(72=2^3\cdot3^2\) is powerful, while \(10=2\cdot5\) is not. Erdős Problem 364 asks whether there is a positive integer \(n\) for which all three numbers \(n\), \(n+1\), and \(n+2\) are powerful. The expected answer is no. This is the Erdős–Mollin–Walsh consecutive-triple question.

The question is easy to state but should not be confused with the nearby question about pairs. Consecutive powerful pairs do exist. The smallest is \(8,9\): the first number is \(2^3\) and the second is \(3^2\). The existence of infinitely many pairs follows from Pell-type equations, as Erdős records. The extra requirement that a third neighbour also be powerful is the difficult boundary.

The problem is about existence, not about finding a typical gap or proving that powerful numbers are rare. A proposed proof must rule out every positive integer, not merely all integers below a large computational cutoff.

A small example

Here is a reproducible finite check. For each integer from \(1\) through \(1000\), factor by trial division and declare the integer powerful exactly when every recorded exponent is at least two. The exact output was: 54 powerful numbers; consecutive pair starts \(8,288,675\); no triple starts; and no quadruple starts. The factorizations used as spot checks were \(8=2^3\), \(9=3^2\), \(10=2\cdot5\), \(48=2^4\cdot3\), \(49=7^2\), and \(50=2\cdot5^2\).

def powerful(n):
    if n == 1: return True
    for p, e in prime_factorisation(n):
        if e < 2: return False
    return True

P = [n for n in range(1, 1001) if powerful(n)]
T = [n for n in range(1, 999)
     if all(powerful(n+i) for i in range(3))]
print(len(P), [n for n in range(1, 1000)
               if powerful(n) and powerful(n+1)][:3], T)

With the stated trial-division factoriser, this prints 54 [8, 288, 675] []. This is an exact finite calculation, not a probabilistic search, and it is useful for catching definition or indexing mistakes. It does not establish the conjecture: a counterexample, if one exists, could be larger than 1000. It also does not verify the catalogue's much larger computational bound.

A simple universal observation does settle a neighbouring statement. Among four consecutive integers, one is \(2\) modulo \(4\). Such an integer has exactly one factor of \(2\) and therefore is not powerful. Hence four consecutive powerful integers are impossible. That modular obstruction leaves the three-term case open.

What is known

The primary source read for this audit is P. Erdős, “Problems and results on number theoretic properties of consecutive integers and related questions,” Proceedings of the Fifth Manitoba Conference on Numerical Mathematics (1976), pp. 25–44. On printed p. 31, Erdős writes the sequence of r-powerful numbers as \(u_k^{(r)}\), notes that \(x^2-8y^2=1\) supplies infinitely many consecutive powerful pairs, and says it is very probable that the distance between the third-powerful neighbours has no solutions. For the 2-powerful sequence he expects a positive-power gap of the form \(u_{k+2}^{(2)}-u_k^{(2)}>(u_k^{(2)})^c\) for some \(c>0\), but explicitly says that he cannot even prove that distance \(3\) occurs only finitely often. This is primary evidence for the historical conjectural boundary, not a proof of the no-triple assertion.

The maintained entry currently states the direct question and labels it “verifiable” and open. It reports that the abc conjecture implies only finitely many such triples, not that there are none. It also records two restricted nonexistence results for triples centred at a cube, one with neighbouring shapes \(p^3x^2\) and one with shapes \(p^2x^3\), and reports an OEIS-based search bound below approximately \(7.38\times10^{28}\). Those are narrower results or finite evidence; none is the universal theorem.

The live discussion contains six comments and one proof claim. The proof claim says that two mixed cube-centred shapes are excluded, while expressly acknowledging that other shapes remain and that Problem 364 is not settled. It is therefore an unverified external report of a partial result, not an accepted proof. The live page was last edited 13 April 2026, so its state is newer than the frozen catalogue packet.

GCL's current state

The frozen catalogue lock is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, with data/problems.yaml blob 4e60263a80af6207aca494d2e1388e44fa372889. Its row for problem 364 says informal status verifiable, last updated 2025-08-31; formal status unformalized; and formalized metadata yes. The GCL issue classifies the item as OUTSIDE_STRICT_OPEN_TRIAGE. The V2 ledger alone decides accepted coverage; this article is explanatory evidence, not a coverage decision.

The pinned formal source is google-deepmind/formal-conjectures commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/364.lean, blob 996aca9cff6d9a2c148912ed418868d6193cc606. Its main Lean theorem formalizes the no-triple proposition over Nat and ends by sorry. The stronger gap variant also ends by sorry. A separate weak variant proves the no-quadruple fact. The file's existence is statement formalization, not a proof, replay, semantic-equivalence certification, or GCL mathematical certification. In particular, the main formal statement does not syntactically include the positive hypothesis \(n>0\), and the stronger variant quantifies over all natural indices; those interfaces require specialist semantic review before being treated as exact translations of the prose.

What remains

The clean next step is a source-and-interface package, not an attempt to settle the conjecture. First, capture the positive-integer statement and the definition of powerfulness in a notation table. Second, compare each quantifier in the pinned Lean theorem with the primary and maintained statements, including the treatment of zero, indexing of Nat.nth, and whether the gap assertion is intended for all sufficiently large indices or all indices. Third, attach a bounded exact-arithmetic certificate for a declared bound, with factorisation witnesses for every rejected candidate and a reproducible generator-completeness argument.

The result should separate four layers: published mathematics, unverified literature or discussion reports, finite computation, and the formal theorem source with sorry. A successful package may show that the formal statement is a faithful interface and that a finite range is clear. It must not turn either fact into a universal proof. If the mixed cube-centred claims are reused, their exact hypotheses and source versions must be recorded rather than generalised to arbitrary powerful neighbours.

Sources and claim boundaries

Primary source: Erdős 1976, scanned proceedings paper, especially printed p. 31. Its downloaded PDF had SHA-256 2c098f2c3e5079ef09a44f693c7db43a29428bb1c8bf7df7ab57b020306a9ad0. The maintained entry is Erdős Problems #364, accessed 2026-10-13; the related primary bibliographic item is Mollin and Walsh, On powerful numbers (1986), pp. 801–806. The original assignment is MATH-PROGRAMME issue #1634. SOURCE RESULT/1 citation: authenticated original RESULT/1.

Frozen catalogue evidence is the commit and blob named above. Formal evidence is the exact commit and blob named above. The live page's status and proof claim are current external reports, not GCL acceptance. No claim here asserts that Problem 364 is solved, that the computational bound is independently certified, or that the Lean file proves its theorem. Any later editorial admission must preserve these distinctions and retain unresolved source ambiguity rather than silently upgrading a finite check or a sorry theorem.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1634 — 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.

Back to catalogue · Compendium introduction