Skip to content

Erdős Problem #285

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: proved (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.

Editorial authority: The protected GCL V2 admissions ledger alone determines publication status. This source-bounded explanation is not mathematical certification.

The question in plain language

Write \(k\) distinct positive integers in increasing order, \(n_1<\cdots<n_k\), whose reciprocal sum is \(1\). Define \(f(k)\) to be the smallest possible largest denominator \(n_k\):

\[ f(k)=\min\left\{n_k:1\leq n_1<\cdots<n_k,\ \sum_{i=1}^k\frac{1}{n_i}=1\right\}. \]

Erdős Problem 285 asks whether

\[ f(k)=(1+o(1))\frac{e}{e-1}k \]

as \(k\) tends to infinity. The quantifier is asymptotic: it does not say that every finite value of \(f(k)\) equals the displayed main term, but that the ratio tends to \(1\) along the values of \(k\) for which the representation exists. There is no two-term representation of \(1\) by distinct positive unit fractions, while the splitting identity

\[ \frac{1}{n}=\frac{1}{n+1}+\frac{1}{n(n+1)} \]

produces longer representations once a suitable one exists. Thus the natural eventual domain starts at \(k=3\).

The live problem page attributes the resolution to Martin. The exact source packet and authenticated audit are issue #1555 and its RESULT/1 comment.

A small example

Exact search gives

\[ f(3)=6\quad\text{from }\quad 1=\frac12+\frac13+\frac16, \]

and

\[ f(4)=12\quad\text{from }\quad 1=\frac12+\frac14+\frac16+\frac1{12}. \]

The following calculation checks these values by enumerating every increasing denominator tuple whose largest entry is at most the search bound. Fraction makes every sum exact; no floating-point approximation is involved.

from fractions import Fraction
from itertools import combinations

def first_by_max(k, bound):
    for largest in range(k, bound + 1):
        for earlier in combinations(range(1, largest), k - 1):
            ns = earlier + (largest,)
            if sum((Fraction(1, n) for n in ns), Fraction()) == 1:
                return ns
    return None

assert first_by_max(3, 20) == (2, 3, 6)
assert first_by_max(4, 100) == (2, 4, 6, 12)
print(first_by_max(3, 20), first_by_max(4, 100))

The output is (2, 3, 6) (2, 4, 6, 12). Since the routine tests smaller possible maxima first, these are finite certificates for the two stated values. They illustrate the definition only; no finite search can establish an asymptotic theorem.

What is known

Greg Martin’s Denser Egyptian fractions, Acta Arithmetica 95 (2000), 231–260, proves a stronger statement. In the paper’s Theorem 2, equation (5), on published p. 232, Martin defines \(M_t(r)\) as the least largest denominator in a representation with \(t\) terms for a positive rational \(r\). He proves that, for every positive rational \(r\) and every integer \(t\geq t_0(r)\),

\[ M_t(r)=\frac{t}{1-e^{-r}}+O_r\left(\frac{t\log\log(3t)}{\log(3t)}\right), \]

and states that this is best possible. He explicitly takes \(t_0(1)=3\). Therefore \(f(k)=M_k(1)\) for \(k\geq3\), and Martin’s theorem implies

\[ f(k)=\frac{e}{e-1}k+O\left(\frac{k\log\log(3k)}{\log(3k)}\right), \]

which yields the problem’s \((1+o(1))\) formulation. The proof is a published external mathematical result. This audit checked the theorem statement and definitions in the accessible source; it did not independently reproduce Martin’s analytic proof.

The original Erdős–Graham source is cited by the maintained page as [ErGr80, p. 33], and the linked scan was reachable. It is the historical problem source, not a substitute for Martin’s resolving theorem. Croot’s related interval result is discussed in Martin’s introduction, but related consequences must not be promoted to the exact theorem for all values of \(t\) without checking their quantifiers.

GCL's current state

The exact original issue body pins the catalogue snapshot teorth/erdosproblems@536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, blob 4e60263a80af6207aca494d2e1388e44fa372889, and the FormalConjectures snapshot google-deepmind/formal-conjectures@85f863718beeec7b58a3a1926ee92e3472bc2020, file FormalConjectures/ErdosProblems/285.lean, blob 1d601222a7b5097ebe420507ba9044d27f66b5a0. The catalogue reports informal status proved, formal status Lean, and overall status proved (Lean); these are catalogue reports, not GCL certification.

The authenticated return is explicitly DISPOSITION: STATUS_CONFLICT, so it is not a SOURCE_LOCK_READY result. Its evidence is necessarily partial at the semantic boundary. The return also describes the issue as having zero comments and no earlier RESULT/1; that was a pre-posting observation and is not the current live state, because the same return is now the issue’s one visible comment. This fidelity discrepancy is preserved rather than silently corrected.

The pinned Lean file is not index-faithful to the original definition. It uses n : Fin k.succ → ℕ, so there are \(k+1\) denominators, and minimizes n (Fin.last k). Its f k therefore corresponds to the original \(f(k+1)\), or \(M_{k+1}(1)\), not \(f(k)\). Its admissible set also restricts the assertion to indices having a representation. In the visible theorem body, the project-specific ∀ᵉ binder is retained; the body quantifies over a function f, a set S, an equality identifying S with the representable indices, and a leastness hypothesis h, then asserts that there exists a little-o error function and that the asymptotic equality holds for every k ∈ S. Both the main theorem and its lower-bound variant end in by sorry. The answer(True) wrapper and category research solved annotation are metadata, not a kernel-checked proof. Any external Lean proof claim remains unsupported by this audit; none was verified, and GCL has issued no internal mathematical certification.

What remains

The literature resolves the mathematics, but a source-faithful formalization still needs an explicit index repair. A bounded successor task is to define the least-largest-denominator quantity with Fin t, prove the correspondence between the existing Fin k.succ formulation and \(M_{k+1}(1)\), and check the finite values \(f(3)=6\) and \(f(4)=12\) without sorry or newly introduced axioms. That task should retain Martin’s analytic theorem as an external dependency rather than implying that the Lean file formalizes its proof.

Two cross-problem warnings are mandatory scope boundaries. For #284, a Croot-related claim that a bound holds for infinitely many \(k\) is not automatically a claim for all sufficiently large \(k\); the bridge between those quantifiers needs scrutiny and must not be imported into #285. For #276, the weak gcd-seed condition is not automatically equivalent to the stronger pinned Lean condition

\[ \forall d>1\ \exists k\ \gcd(d,a_k)=1. \]

That equivalence remains unresolved here. Neither warning changes the published status of #285, and neither is evidence for a #285 formal proof.

Sources and claim boundaries

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1555 — 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