Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Claim. For each nn let Xn⊂[−1,1]X_n\subset[-1,1] have nn distinct points, and let rn≥0r_n\ge0 be integers with rn/n→0r_n/n\to0. Theorem 1 of the write-up Almost everywhere divergence of polynomial interpolation with sublinear excess degree (Zenodo record 10.5281/zenodo.22549777, version 7 of 2026-09-06 under the concept record 10.5281/zenodo.22312935, whose record lists its creator as anonymous) states that there is a continuous f:[−1,1]→Rf:[-1,1]\to\mathbb R such that every sequence of real polynomials pnp_n of degree at most n+rnn+r_n with pn=fp_n=f on XnX_n satisfies

lim sup⁡n→∞∣pn(x)∣=∞for almost every x∈[−1,1],\limsup_{n\to\infty}\lvert p_n(x)\rvert=\infty \quad\text{for almost every }x\in[-1,1],

the exceptional null set depending on the sequence. This answers Problem 1152 yes, in a stronger form: given ϵ(n)→0\epsilon(n)\to0, put rn=max⁡{0,⌈ϵ(n)n⌉−1}r_n=\max\{0,\lceil\epsilon(n)n\rceil-1\}, so that deg⁡pn<(1+ϵ(n))n\deg p_n<(1+\epsilon(n))n implies deg⁡pn≤n+rn\deg p_n\le n+r_n, and unboundedness at xx excludes pn(x)→f(x)p_n(x)\to f(x). The same ff serves every admissible sequence, as the question requires, and the quantifier order matches: the null set may depend on (pn)(p_n). The write-up presents the theorem as the extension of the Erdős–Vértesi almost-everywhere divergence of Lagrange interpolation, for arbitrary arrays, to non-unique interpolation with o(n)o(n) excess degree, in contrast with the theorem of Erdős, Kroó and Szabados [EKS89] in the site's commentary, that for a fixed ϵ>0\epsilon>0 suitable arrays give every continuous ff uniformly convergent interpolants of degree below (1+ϵ)n(1+\epsilon)n. Its inputs are a construction of Olevskii and Ulanovskii in Bernstein spaces and the uniform Christoffel–Darboux kernel asymptotics of Kriecherbauer, Schubert, Schüler and Venker; a finite interpolation construction forces large values on a fixed fraction of a local interval uniformly over the admissible corrections, finitely many such data sets rule out common sets of bounded values, and the Baire category theorem gives one ff.

Submission note. Posted to erdosproblems.com as a proof claim by Qiyuan Gu (account fireflysentinel) on 4 September 2026, giving "GPT 6 Astra, GPT 5.6 Sol and Claude Fable 5.1" as the AI used:

Let Xₙ be any set of n distinct points in [−1, 1], and let rₙ = o(n). We prove that there exists f ∈ C([−1, 1]; ℝ) such that every sequence of polynomials pₙ of degree at most n + rₙ interpolating f on Xₙ satisfies lim supₙ→∞ |pₙ(x)| = ∞ almost everywhere. This strengthens the divergence assertion in Erdős Problem #1152 and extends the Erdős–Vértesi phenomenon to o(n) excess degree. The proof combines logarithmic potential theory, localized Bernstein-space functions, weighted Christoffel-Darboux kernel asymptotics, a sign-change argument, and the Baire category theorem. Notes: The proof is drafted using GPT 6 Astra, running on top of GPT 5.6 Sol and Claude Fable 5.1. The author is solely responsible for all mathematical content.

Depends on. No page of this wiki: the inputs are cited from the literature.

Standing. A manuscript claim, claimed. The claimant is Qiyuan Gu, named on the site's proof-claims tab, where the claim was submitted on 2026-09-04 by another account on Gu's behalf, with the disclosure that the proof was drafted with GPT 6 Astra on top of GPT 5.6 Sol and Claude Fable 5.1 and that the author is solely responsible for the mathematics; the Zenodo record lists its creator as anonymous. The tab links the record's first version (22312936), which Zenodo reported as gone (HTTP 410) on 2026-10-07; the concept DOI resolves to version 7, linked above, which carries the paper and a Lean 4 archive that the paper calls a partial formalization. The archive covers the sign-change bound of Lemma 3, the scalar calculations of Lemma 4, the deduction from (6.3) through Remez's inequality, and the finite construction and Baire argument of Sections 7-8. It takes Remez's inequality and the remaining analytic estimates as hypotheses. No build or audit of the archive is recorded, so it is recorded as a link and contributes no formalized evidence. Not reviewed: the site's label is OPEN (page last edited 23 January 2026), the tab carries no comments as of 2026-10-07, its commentary does not mention the claim, and no outside review is known. Not refereed: no journal or arXiv version is known.