Skip to content

Erdős Problem #263

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.

Documentary explanation. Editorial acceptance is determined by the protected V2 admissions ledger and the exact manuscript source hash; this text does not confer mathematical certification.

The question in plain language

Take a strictly increasing sequence of positive integers

\[ a_1<a_2<a_3<\cdots. \]

Call it a Type-2 irrationality sequence when every positive-integer sequence \((b_n)\) satisfying \(b_n/a_n\to1\) has an irrational reciprocal sum. In other words, the requirement is universal: every multiplicative perturbation that becomes asymptotically negligible must still leave

\[ \sum_{n=1}^{\infty}\frac{1}{b_n} \]

irrational. The problem asks two questions. Is \(a_n=2^{2^n}\) such a sequence? Must every Type-2 sequence satisfy \(a_n^{1/n}\to\infty\)?

The increasing hypothesis is essential. The live problem page corrected an earlier version on 2 April 2026 after a counterexample was reported for the non-monotone formulation. Thus the corrected question is not interchangeable with the older formal statement or with a counterexample to that older statement. The maintained catalogue still labels the problem open.

A small example

For a reproducible finite check, use zero-based indexing for the same double-exponential formula. The first five terms are \(2,4,16,256,65536\). They satisfy the exact recurrence \(a_{n+1}=a_n^2\), so the boundary ratio is \(a_{n+1}/a_n^2=1\). The five-term reciprocal sum is

\[ \frac12+\frac14+\frac1{16}+\frac1{256}+\frac1{65536}=\frac{53505}{65536}. \]

The following Python calculation checks both facts with exact integers and rational arithmetic:

from fractions import Fraction

a = [2 ** (2 ** n) for n in range(5)]
partial = sum((Fraction(1, x) for x in a), Fraction())
ratios = [Fraction(a[n + 1], a[n] ** 2) for n in range(4)]
assert a == [2, 4, 16, 256, 65536]
assert partial == Fraction(53505, 65536)
assert ratios == [Fraction(1, 1)] * 4
print(a, partial, ratios)

This verifies only a finite partial sum and the recurrence. A finite rational partial sum says nothing about whether the infinite sum is rational, and the recurrence says nothing by itself about all possible perturbations \((b_n)\).

What is known

The classical source is Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 63. Erdős restated the issue in “On the irrationality of certain series: problems and results,” New advances in transcendence theory (1988), p. 105. Those sources give the Type-2 definition and present the double-exponential sequence as an unresolved, nontrivial case. Erdős’s 1975 paper, “Some problems and results on the irrationality of the sum of infinite series,” Journal of Mathematical Sciences 10 (1975), 1–7, supplies related classical context, but is not a proof of this specific Type-2 question.

The sharpest directly relevant proved interface in the audited literature is Kovač and Tao, On several irrationality problems for Ahmes series, arXiv:2406.17593v4, Theorem 2.4, p. 6. If \((a_n)\) is strictly increasing, \(sum 1/a_n\) converges, and

\[ \frac{a_{n+1}}{a_n^2}\longrightarrow0, \]

then \((a_n)\) is not Type-2. Its conclusion is a genuine obstruction: under those hypotheses there is a positive-integer perturbation with ratio tending to \(1\) whose reciprocal sum is rational. It is not a classification theorem, and it does not cover the boundary case where the displayed ratio tends to \(1\).

The same paper records the positive folklore criterion that sufficiently rapid root growth, such as \(a_n^{1/2^n}\to\infty\), forces the relevant reciprocal sums to be irrational; stronger growth assumptions, for example \(\liminf a_{n+1}/a_n^{2+\varepsilon}>0\) for some \(\varepsilon>0\), imply that criterion. For \(a_n=2^{2^n}\), however, \(a_n^{1/2^n}=2\) (with either consistent index convention), not infinity. Both broad interfaces therefore stop exactly where Problem #263 is interesting.

Koizumi’s Irrationality of the reciprocal sum of doubly exponential sequences, arXiv:2504.05933v1, Theorem 4, p. 3, proves that \(\lfloor\alpha^{2^n}\rfloor\) is Type-2 for all but countably many \(\alpha>1\), using a nondecreasing convention. This is a powerful almost-everywhere result, but it does not decide the exceptional-looking specific value \(\alpha=2\). It is an external mathematical result, not GCL certification.

GCL's current state

CONSPICUOUS EXCEPTION — STATUS_CONFLICT. The authenticated latest source audit is RESULT/1 on original issue #1533. It returns DISPOSITION: STATUS_CONFLICT, not SOURCE_LOCK_READY and not a proof. The conflict is documentary and semantic: the pinned Formal Conjectures declaration encoded the pre-correction, non-monotone question and marked the second part answer(False), while the corrected source problem requires strict increase and leaves that second part open. A declaration containing sorry is not a proof, and the old counterexample cannot settle the corrected question.

The pinned catalogue record is teorth/erdosproblems at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, entry number: '263', Git blob 4e60263a80af6207aca494d2e1388e44fa372889. It records open, with formal-solution status unformalized and statement-formalized metadata yes. The pinned Formal Conjectures file uses \(a_n/b_n\to1\) rather than the source’s \(b_n/a_n\to1\), omits monotonicity, starts indexing at zero, and uses unrestricted Lean tsum. The current provider’s revised file adds the monotonicity correction and explicitly keeps the corrected second question open, but its theorem bodies remain sorry.

The exact source-audit proposal is narrower than solving #263: reconcile the index shift, the two ratio orientations under positivity, and the meaning of tsum when convergence is not explicit. That interface task may produce a compiling semantic lemma or a concrete countermodel; it cannot be promoted to a solution of either question.

What remains

Question 1, whether \(2^{2^n}\) is Type-2, remains open in the checked primary evidence. Kovač–Tao do not apply because the ratio is exactly \(1\), and the folklore sufficient condition does not apply because the relevant root is constantly \(2\). Koizumi’s all-but-countably-many statement also leaves the particular base \(2\) undecided. Question 2 remains open after the monotonicity correction; the pre-correction negative result is not evidence against it.

The smallest responsible next step is a source/formalization reconciliation, not a claim of completion. It should use positive-natural indexing, prove the reciprocal-ratio equivalence rather than assume it, and state explicitly whether the reciprocal series is required to converge. Any later mathematical work would need a new theorem or a counterexample satisfying strict increase and the corrected quantifiers. Finite computation cannot settle either infinite assertion.

The live page also reports user-submitted proof claims and discussion proposals, including third-party Lean developments. They are unreviewed external claims: they may be useful leads, but none is treated here as an accepted proof, a GCL result, or a change in status. In particular, a formal declaration, a compilable fragment, or a proof claim with hidden axioms does not supply the missing mathematical certification.

Sources and claim boundaries

  • Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), p. 63: exact scan.
  • Erdős, “On the irrationality of certain series: problems and results,” New advances in transcendence theory (1988), pp. 102–109, especially p. 105: exact PDF.
  • Erdős, “Some problems and results on the irrationality of the sum of infinite series,” Journal of Mathematical Sciences 10 (1975), pp. 1–7: exact PDF.
  • Kovač and Tao, On several irrationality problems for Ahmes series, arXiv:2406.17593v4, Theorem 2.4 p. 6: arXiv record and exact PDF.
  • Koizumi, Irrationality of the reciprocal sum of doubly exponential sequences, arXiv:2504.05933v1, Theorem 4 p. 3: exact PDF.
  • Current catalogue presentation: official Erdős Problem #263 page, which records the corrected increasing formulation and open status.
  • Maintained source catalogue: pinned data/problems.yaml entry; catalogue entry blob 4e60263a80af6207aca494d2e1388e44fa372889.
  • Formal declarations: pinned 263.lean, blob e45d17f365d79d054885ce9eee82cc86272a46f3; current main declaration, whose current blob was identified by the audit as 6235aac34661f9f327d47ecefb9af1cf017f7e01.
  • GCL evidence: original issue #1533, exact issue-body SHA-256 cc677a9179fa1e0aa354783ec3a62795b82c25010aaa9146d86db773e2b47000, and the authenticated RESULT/1 source audit.

These references preserve the source and evidence boundaries. The papers provide the stated theorem interfaces; the catalogue reports a maintained status; the formal files expose statements and unresolved proof obligations; and the authenticated issue comment records the GCL source audit. None, alone or in combination, is a GCL mathematical certification of Problem #263. Its admission state is determined separately by the protected V2 admissions ledger.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1533 — 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