Work

Sharp Asymptotics for an Extremal Problem of Erdős on Lagrange Interpolation

Year: 2026
Type: preprint
Abstract: For nodes −1 ≤ x₁ < ⋯ < xₙ ≤ 1 let ℓ₁, …, ℓₙ be the fundamental polynomials of Lagrange interpolation and I = ∫ Σₖ ℓₖ² dx over [−1, 1]. Erdős, Szabados, Varma and Vértesi proved 2 − O((log n)²/n) ≤ min I ≤ 2 − 2/(2n − 1) and asked whether min I = 2 − (1 + o(1))/n (Erdős Problem #1131). We show that the answer is negative and determine the second term: min I = 2 − (c∗ + o(1))/n with c∗ = Γ(1/4)⁴/(16π²) = 1.0942…, the constant that Brutman and Toledano obtained by numerical extrapolation. Moreover min I > 2 − 5/n for every n ≥ 3. The upper bound comes from the Gauss–Lobatto nodes of the weight (1 − x²)^(1/4). More generally, the Gauss and Gauss–Lobatto nodes of every generalized Jacobi weight w with a C² factor satisfy I = 2 − Λ[w]/n + o(1/n) for an explicit functional Λ whose maximum c∗ is attained only at w = (1 − x²)^(1/4). For the lower bound we express the deficit of an arbitrary node system through its Prüfer phases and a comparison with Gauss–Jacobi nodes, and pass to the limit in exact identities for the Legendre coefficients of the polynomials whose squared norms make up the deficit. The reversed node polynomials converge to an outer function, which selects a lift of the limit phase, and an exact formula for the resulting quadratic functional gives lim sup n(2 − I) ≤ c∗ − D ≤ c∗, where D ≥ 0 is an explicit double integral that vanishes only for the extremal phase. Hence the phases of all asymptotic minimizers converge to those of the nodes of (1 − x²)^(1/4).
Language: English
Open Access status: green
OpenAlex ID: W7233794361
Cites:
Cited by:
Related to: 10