Erdős Problem #267
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: irrationality.
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.
This is documentary exposition only. The protected V2 ledger determines manuscript acceptance; this text does not certify a theorem, a formalization, or an external proof artifact.
The question in plain language
Let \(F_1=F_2=1\) and \(F_{n+1}=F_n+F_{n-1}\), so the positive-index Fibonacci numbers begin \(1,1,2,3,5,8,\ldots\). Erdős and Graham ask about an infinite increasing sequence of positive integers
whose indices are separated at a fixed multiplicative rate: there is a real number \(c>1\) such that
for every \(k\). Must the reciprocal Fibonacci sum
be irrational? The original wording appears on p. 65 of Erdős and Graham’s 1980 scan, and the maintained Erdős Problem 267 entry reproduces it.
The quantifiers matter. The question concerns every admissible pair consisting of a sequence and a fixed rate \(c>1\); it is not asking whether one can choose a convenient rate after seeing the sequence. Equivalently, a disproof would need one \(c>1\) and one sequence satisfying the displayed inequality whose sum is rational. The known boundary is \(c=2\): the case \(c\ge 2\) is proved, while the interval \(1<c<2\) is the unresolved part. The sequence is positive-indexed in the original statement, so the quotients are ordinary real quotients of positive integers.
A small example
Take the especially sparse indices \(n_k=2^k\), starting at \(k=0\). The relevant Fibonacci denominators are
Using exact integer arithmetic, the first eight terms have sum
This is independently reproducible without floating-point arithmetic:
from fractions import Fraction
F = [0, 1]
for _ in range(2, 129):
F.append(F[-1] + F[-2])
S = sum((Fraction(1, F[2**k]) for k in range(8)), Fraction())
print(S)
The exact output is also recorded in the authenticated RESULT/1 source audit; the code above recomputes it independently.
The computation checks a finite rational identity, not the value or irrationality of the infinite sum. Good’s 1974 paper and Hoggatt–Bicknell’s 1976 paper establish the stronger special identity
so this particular sum is irrational. It is a useful calibration example: the ratio is exactly \(2\), and therefore it lies in the proved boundary case rather than in the difficult interval.
What is known
The first relevant special case is the power-of-two series just displayed. Good gives the identity on p. 346; Hoggatt–Bicknell derive a wider family on pp. 453–455 and recover the same value when their parameter is \(1\). Badea’s 1987 paper proves another special case, namely irrationality of \(\sum_{k\ge 0}1/F_{2^k+1}\) in Corollary 4 on p. 226. These results settle particular index patterns, not the full quantifier over all rapidly separated sequences.
The decisive published interface for the general question is C. Badea, “A theorem on irrationality of infinite series and applications,” Acta Arithmetica 63 (1993), 313–323, DOI 10.4064/aa-63-4-313-323. Badea considers the recurrence
with positive integers \(a\) and \(b\). Lemma 3.1(i), p. 317, gives
Corollary 3.2, p. 318, says that if \(n(k+1)\ge 2n(k)-1\) for all sufficiently large \(k\), then \(\sum_k1/x_{n(k)}\) is irrational. Setting \(a=b=1\) identifies the positive terms of this recurrence with the Fibonacci numbers. If the Erdős condition holds with \(c\ge2\), then \(n_{k+1}\ge2n_k\), hence Badea’s weaker eventual inequality applies. This proves the whole \(c\ge2\) subcase.
No inspected primary source resolves \(1<c<2\). The maintained entry therefore reports the problem as open. Its attribution of irrationality for \(\sum_{n\ge1}1/F_n\) to André-Jeannin is retained as a historical lead, but that cited source was not accessible in the source audit and is not promoted here as independently checked evidence.
GCL's current state
The protected catalogue registry records Problem 267 as open, with triage class PROTECTED_READY_BACKLOG and protected_fc_present: true; its exact registry issue-body hash is ffc93166f6cdfd0ecebef2574f8277965736839ec92ec31f34cbd696e8f86cc9. The pinned catalogue packet is the number: '267' entry in teorth/erdosproblems at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, Git blob 4e60263a80af6207aca494d2e1388e44fa372889; it records informal status open, formal status unformalized, and statement-formalized metadata yes.
The authenticated GitHub identity used for this readback was fyremael. The authenticated source audit on MATH-PROGRAMME issue #1537, exact RESULT/1 comment, returned DISPOSITION: PARTIAL. That audit confirms the live page still says “open,” records four discussion comments and one separate proof claim, and notes documentary drift in the page’s comment display. A public proof claim is evidence awaiting review, not a status change.
The pinned formal source is FormalConjectures/ErdosProblems/267.lean at commit 85f863718beeec7b58a3a1926ee92e3472bc2020, file blob aeb736e6df2f9972cf47226c3d6cb1b65873a771. Its main theorem has answer(sorry) and every displayed theorem body is sorry; it is a statement-bearing source, not a proof. It also uses zero-based n : ℕ → ℕ and Nat.fib, while the original uses positive indices and \(F_1=F_2=1\). Its ∀ᵉ notation is an explicit binder notation here, not an eventual-filter quantifier. The division and coercions are left to Lean elaboration, so semantic equivalence has not been certified.
What remains
The mathematical residue is precise: determine whether every sequence with a fixed ratio in \(1<c<2\) still gives an irrational sum, or produce one rational counterexample. A bounded next theorem would formalize only Badea’s \(c\ge2\) interface with explicit real reciprocals, positive indices, convergence, and the index shift between \(F_n\) and Nat.fib. Its acceptance test should include the recurrence bridge, the eventual inequality, and a kernel-checked proof with no sorry.
That task would not settle the original open interval. In particular, a proof for \(n_k=2^k\), for \(n_k=2^k+1\), or for the unreviewed external full-range claim would remain a special case or an unverified claim unless its exact semantics and proof were independently replayed. The separately cited external file at mo271/formal-conjectures commit 2663234a28260853790aa5752d8d4550ff0ab1ca, blob 9114bda467d209369b5eb07167f2273afec295f9, has not been kernel-replayed or checked for semantic equivalence. Self-reporting CI, a non-sorry declaration, or an external repository cannot by itself certify the mathematics.
Sources and claim boundaries
- Original statement: Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 65, original scan.
- Special power-of-two identity: I. J. Good, “A Reciprocal Series of Fibonacci Numbers,” The Fibonacci Quarterly 12 (1974), p. 346, scanned paper; V. E. Hoggatt Jr. and M. Bicknell, “A Reciprocal Series of Fibonacci Numbers with Subscripts \(2^n k\),” The Fibonacci Quarterly 14 (1976), pp. 453–455, scanned paper.
- Special shifted case: C. Badea, “The irrationality of certain infinite series,” Glasgow Mathematical Journal 29 (1987), pp. 221–228, especially Corollary 4, p. 226, DOI 10.1017/S0017089500006868 and full scan.
- General sparse-sequence result: Badea (1993), Lemma 3.1 and Corollary 3.2, pp. 317–318, DOI 10.4064/aa-63-4-313-323 and inspected full text.
- Current and pinned programme evidence: maintained Problem 267 entry, pinned catalogue record, authenticated issue #1537, and exact source-audit RESULT/1.
The status supported here is open, not proved or disproved. The \(c\ge2\) subcase is supported by Badea’s inspected theorem; the interval \(1<c<2\) remains unresolved in the evidence reviewed. The catalogue and formal-source hashes identify fixed evidence, but neither the registry, the formal statement, the live page, nor this manuscript supplies mathematical certification. The protected V2 ledger determines acceptance.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1537 — 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.