Erdős Problem #260
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
Choose a strictly increasing sequence of positive integers
that becomes sparse in the average sense
The question is whether the binary-looking weighted sum
must be irrational. The quantifiers matter: the assertion is about every such sequence, not about one convenient choice. The same integer \(a_n\) appears both as the coefficient and as the exponent. The growth condition says that the \(n\) th selected exponent eventually lies farther than any fixed multiple of \(n\); it does not require the individual gaps \(a_{n+1}-a_n\) to grow. Long clusters of small gaps can be separated by occasional very large gaps.
The sum converges because the exponential denominator dominates the linear numerator, and strict increase makes the exponents distinct. A finite change in the first few terms changes the sum by a rational number, so the irrationality question is fundamentally about the tail. The problem is therefore a question about how sparsity interacts with the base-two expansion, not merely about estimating a numerical series.
A small example
The sequence \(a_n=2n\) is strictly increasing, but it does not satisfy the problem's hypothesis: \(a_n/n=2\). It is useful as a warning that a sparse-looking series can still be rational. Directly,
The finite calculation is also exact. For every \(m\geq1\),
For example, an exact rational-arithmetic check gives the partial sum at \(m=20\) as \(7635497415/8589934592\), with tail \(1/77309411328\).
from fractions import Fraction
m = 20
partial = sum((Fraction(2*n, 4**n) for n in range(1, m + 1)), Fraction(0))
closed = Fraction(8, 9) * (1 - Fraction(m + 1, 4**m) + Fraction(m, 4**(m + 1)))
assert partial == closed
assert Fraction(8, 9) - partial == Fraction(1, 77309411328)
This is not a counterexample to Problem 260, because its average density is positive. It isolates the real boundary: rationality is easy for a regular, non-sparse support, while the problem asks what happens when the selected exponents have zero average density.
What is known
Erdős's primary paper, “Sur l'irrationalité d'une certaine série,” C. R. Acad. Sci. Paris 292 (1981), 765–768, proves irrationality under stronger hypotheses. On p. 765 he proves the result when
and also records a proof under the growth condition
The paper says that the desired condition \(a_n/n\to\infty\) very likely suffices, but does not supply that proof or a rational example with unbounded limsup gaps. Thus the classical source is a proof interface for two stronger regimes, not a proof of the exact problem. The same paper notes that the base \(2\) can be replaced by an integer base \(t>1\) in its stronger results.
Borwein and Loring's paper, “Some Questions of Erdős and Graham on Numbers of the Form \(\sum g_n/2^{g_n}\),” Mathematics of Computation 54 (1990), 377–394, p. 378, records a related irrationality condition involving
That condition is also not implied by \(a_n/n\to\infty\): average sparsity allows local clusters of short gaps. These results explain why the original formulation remained nontrivial.
The current external claim is Han Wang's arXiv preprint, version 4, “Sparse polynomial-weighted expansions.” Its Theorem 1.1 states that for an integer \(b\geq2\), a nonzero polynomial \(p\in\mathbb{Q}[x]\), and an infinite set \(S\subseteq\mathbb{N}\), rationality of
forces, for every fixed \(0<\theta\leq1\), constants \(c>0\) and \(N_0\) such that
whenever \(N\geq N_0\) and \(N^\theta\leq W\leq N\). Corollary 1.2 applies this to \(b=2\), \(p(n)=n\), and \(S=\{a_1,a_2,\ldots\}\), yielding positive lower density and hence contradicting \(a_n/n\to\infty\). The preprint describes a carry recurrence and polynomial-height argument; its Lemma 2.1 is the local gap interface used by the corollary. This is a claimed external proof, not an accepted GCL proof.
GCL's current state
The authenticated latest RESULT/1 source audit on original issue #1530 has disposition STATUS_CONFLICT. It read the original packet, the 1981 paper, the related 1990 result, the live page, arXiv version 4, and the external Lean repository. It records that the live Erdős Problems #260 page, last edited 1 February 2026, still says open, offers no proof exposition, and shows one unreviewed proof claim. The audit also records the pinned catalogue as open and formally unformalized.
The pinned catalogue entry is the 260 record at commit 536408716724a2402bdfa7c1d764bf83d64a8c89, blob 4e60263a80af6207aca494d2e1388e44fa372889; it says open, with last update 2025-08-31, while marking the statement as formalized. The pinned Formal Conjectures declaration, commit 85f863718beeec7b58a3a1926ee92e3472bc2020, blob 3530bdd3ec1113071a2c73f48ff68944f1c69555, has the matching theorem statement but uses answer(sorry) and sorry; its TODOs explicitly mention the stronger Erdős assumptions. It is statement evidence only.
The external Lean repository claims a no-sorry formal endpoint at commit e708b584c9a3b54857f050d6b7efbecb8a5ea27a. That self-reported claim and the preprint's formalization are outside protected GCL mathematical certification. No GCL theorem, MathCert certificate, or independent replay is asserted here.
What remains
The central unresolved question is status adjudication, not statement access. A specialist review must check the full Theorem 1.1 argument and its specialization, including denominator normalization, the carry orbit, the polynomial-window density conclusion, and the passage from positive lower density to a contradiction. A bounded successor could first audit the binary linear case of Lemma 2.1 and reproduce its exact recurrence before considering the longer theorem. Until that work is completed, “the preprint claims a proof” is the strongest safe wording.
One elementary consequence makes the proposed contradiction transparent. Let \(A(x)\) count the selected exponents at most \(x\). If \(a_n/n\to\infty\), then for every \(M>0\), eventually \(a_n\geq Mn\). Consequently, for large \(x\),
so \(A(x)/x\to0\) after letting \(M\) grow. Wang's claimed Corollary 1.2 would instead give \(A(x)\geq c'x\) for all sufficiently large \(x\). The contradiction is rigorous conditional on that external theorem, but it does not validate the theorem's proof. It also does not turn the sorry-bearing pinned declaration or an unreviewed repository into certification.
Sources and claim boundaries
- Erdős, “Sur l'irrationalité d'une certaine série,” 1981, pp. 765–768, especially p. 765: primary proof for the two stronger assumptions and its stated limitation.
- Borwein–Loring, “Some Questions of Erdős and Graham on Numbers of the Form \(\sum g_n/2^{g_n}\),” Math. Comp. 54 (1990), 377–394, especially p. 378; JSTOR record: related gap condition and base-change context.
- Wang, “Sparse polynomial-weighted expansions,” arXiv:2606.24972v4, Theorem 1.1, Corollary 1.2, and Lemma 2.1: the 2026 external claimed route; not GCL-certified.
- Erdős–Graham, Old and new problems and results in combinatorial number theory (1980) and Erdős, “On the irrationality of certain series: problems and results” (1988), pp. 102–109: bibliography attached to the original problem. The authenticated audit reports that the 1988 scan was not directly retrievable in its environment, so it is not used here as sole support for a proof claim.
- Pinned catalogue source and exact problem record and pinned formal declaration: external statement and formal-interface evidence, not proof authority.
- Live Problem 260 page, external formalization claim, and authenticated GCL RESULT/1: current status surfaces and documentary evidence. They do not replace mathematical review.
The article preserves the unresolved conflict conspicuously: the 1981 classical results are source-reported theorems with stated stronger hypotheses; the 2026 resolution is an external claim requiring mathematical review; and GCL currently certifies neither that claim nor the target problem as solved.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1530 — 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.