Skip to content

Erdős Problem #306

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

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 #306 asks whether every positive rational number can be written as a finite sum of distinct unit fractions whose denominators are semiprimes. In the normalized formulation, write \(q=a/b\) in lowest terms, require \(b\) to be squarefree, and ask whether there are integers

\[ 1<n_1<\cdots<n_k \]

such that each \(n_i\) is the product of two distinct primes and

\[ q=\frac{1}{n_1}+\cdots+\frac{1}{n_k}. \]

“Distinct primes” excludes squares such as \(9\); the allowed denominators include \(6=2\cdot3\), \(10=2\cdot5\), and \(15=3\cdot5\). The squarefree condition belongs to the reduced denominator of \(q\), not to an arbitrary unreduced way of writing the same rational. The question is therefore an existence statement with several important quantifiers: for every eligible \(q\), there must exist some finite length \(k\) and some increasing list of eligible denominators.

The restriction is on the denominators, not on the number of prime factors appearing across the whole list: different terms may reuse a prime, as the example below does. Also, the problem asks for a finite sum, so an argument based only on an infinite convergent series would not be enough. These distinctions explain why a verified example for \(q=1\) is informative without being a solution of the universal question. They also give a practical checklist for reading a claimed proof: identify the exact domain of \(q\), check whether the denominator is canonical and squarefree, and check that every selected denominator is a product of two different primes.

A small example

The primary source Watanabe, arXiv:2009.03275v2 gives the following 47-term example for \(q=1\):

\[ \begin{aligned} 1={}&\frac1{6}+\frac1{10}+\frac1{14}+\frac1{15}+\frac1{21}+\frac1{22}+\frac1{26}+\frac1{33}+\frac1{34}+\frac1{35}+\frac1{38}+\frac1{39}\\ &+\frac1{46}+\frac1{51}+\frac1{55}+\frac1{57}+\frac1{58}+\frac1{62}+\frac1{65}+\frac1{69}+\frac1{74}+\frac1{82}+\frac1{85}+\frac1{86}\\ &+\frac1{87}+\frac1{91}+\frac1{93}+\frac1{95}+\frac1{106}+\frac1{111}+\frac1{123}+\frac1{133}+\frac1{145}+\frac1{155}+\frac1{159}\\ &+\frac1{185}+\frac1{203}+\frac1{215}+\frac1{253}+\frac1{265}+\frac1{287}+\frac1{319}+\frac1{493}+\frac1{583}+\frac1{731}+\frac1{851}+\frac1{1073}. \end{aligned} \]

An exact-arithmetic check using rational fractions returned count = 47, strictly_increasing = True, all_distinct_prime_products = True, and sum = 1; it also found no denominator failing the two-distinct-prime test. For example, \(159=3\cdot53\) and \(1073=29\cdot37\). This verifies one finite representation and makes no claim about arbitrary \(a/b\), about a minimum number of terms, or about the truth of the general problem. Watanabe reports 17 such 47-term examples and says that no shorter example had been found; that is a computational report, not a proof that 47 is minimal.

The calculation is reproducible without floating-point rounding: construct an exact rational 1/n for each listed integer, add the 47 fractions, sort the denominators, remove duplicates, and factor each integer. The resulting equality is exact, rather than a decimal approximation. That distinction matters for Egyptian fractions, where a tiny numerical residual could conceal an incorrect identity. It still checks only the certificate printed in the paper; it does not validate the paper’s search procedure or prove that no other, shorter certificate exists.

What is known

The source page Erdős Problems #306 gives the same general question. It records that when denominators are products of three distinct primes, the integer case \(b=1\) was proved by Butler, Erdős, and Graham in “Egyptian fractions with each denominator having three distinct prime divisors,” Integers 15 (2015), A51; the authors’ primary paper states that every natural number has such a representation. That theorem changes “two” to “three” and does not settle #306.

For the semiprime case, Watanabe’s 2020 paper is primary evidence for explicit decompositions of \(1\), not a proof for every squarefree-denominator rational. The maintained page was last edited on 21 June 2026 and now displays one proof claim linking to 2026 arXiv work. The page itself warns that a proof claim is not a guarantee of correctness. The claim and the associated preprints are therefore recorded here as unverified external reports, not as accepted coverage.

GCL's current state

The frozen catalogue packet is teorth/erdosproblems at commit 536408716724a2402bdfa7c1d764bf83d64a8c89. Its data/problems.yaml entry for number: "306" has Git blob 4e60263a80af6207aca494d2e1388e44fa372889, informal and overall status open last updated 2025-08-31, formal status unformalized, and statement-formalized metadata yes. The entry notes that the AI-assisted formalization reference concerns Watanabe’s 47-term example, not the general rational question.

The protected FormalConjectures file at commit 85f863718beeec7b58a3a1926ee92e3472bc2020 has blob b7ffb5f739a36391874fe86a92006aeef4ffbc57. Its main theorem quantifies over positive rationals with squarefree canonical denominators and encodes an increasing array beginning with 1; the later entries have exactly two distinct prime factors, and their reciprocals sum to the rational. That is a useful formal interface, but the theorem is answer(sorry) ↔ ... and ends with by sorry. Its separate three-prime integer variant also ends with by sorry. Neither file is a proof or a GCL certification.

What remains

The immediate missing work is a source-semantic review of the recent proof claims: compare their exact hypotheses, quantifiers, denominator distinctness conditions, and dependency assumptions with the normalized statement above. A report that proves all integers, or rationals above a threshold, is materially weaker than the registered all-positive-rationals statement. A proof of the three-prime analogue is a different theorem even when its method is relevant.

A bounded successor task can independently reproduce Watanabe’s displayed list with exact rational arithmetic, factor every denominator, and compare the resulting finite certificate with the formal file’s indexing convention. Acceptance would require exact sum \(1\), 47 distinct increasing denominators, and two different prime factors for every denominator. Falsification is immediate at the first bad factor, duplicate, ordering failure, or nonzero residual. This finite check is useful evidence and a regression test, but it cannot establish the infinite quantified statement.

Sources and claim boundaries

The original assignment is MATH-PROGRAMME issue #1576. The original SOURCE RESULT/1 citation is authenticated original RESULT/1; no future comment or link is invented here. The article relies on the frozen catalogue commit and blob above, the pinned FormalConjectures commit and blob above, the maintained live problem entry, Watanabe’s 2020 primary source, and Butler–Erdős–Graham’s 2015 primary paper.

The displayed finite identity is established by exact arithmetic; the general problem remains an open catalogue claim in the frozen packet. Recent proof claims are external and unverified in this audit. A Lean statement containing sorry is reported as a formal statement with an admitted proof placeholder, not as a replayed theorem. V2 ledger state alone decides accepted coverage. This source audit supplies no mathematical certification, publication decision, priority claim, or award claim.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1576 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: PROTECTED_READY_BACKLOG.
  • 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