Wiki
Wiki

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

Updated


Source. Theorem 7, p. 1175, and the Corollary, p. 1176, of P. Erdős, "Some remarks on polynomials," Bull. Amer. Math. Soc. 53 (1947), 1169-1176. Pages are the journal's own, as on the source card.

Statement

Theorem 7 (p. 1175). Let fn(z)f_n(z) be a polynomial of degree nn with real coefficients and ∣fn(z)∣<1\lvert f_n(z)\rvert<1 for −1≤z≤1-1\le z\le1. Then for ∣z0∣≥1\lvert z_0\rvert\ge1,

∣fn(z0)∣≤∣Tn(z0)∣,\lvert f_n(z_0)\rvert\le\lvert T_n(z_0)\rvert ,

where TnT_n is the Chebyshev polynomial of degree nn. The print adds that equality holds only for fn(z)=±T(z)f_n(z)=\pm T(z), although TnT_n itself does not satisfy the strict hypothesis. It notes that for real z0z_0 the result is well known.

The paper proves a more general form, (12) (p. 1175): the same inequality holds for ∣z0∣≥1\lvert z_0\rvert\ge1 when ∣fn∣≤1\lvert f_n\rvert\le1 is assumed only at the n+1n+1 points −1-1, 11 and the roots of Tn′T_n'.

Corollary (p. 1176). If ∣fn(z)∣≤1\lvert f_n(z)\rvert\le1 for −1≤z≤1-1\le z\le1 and fnf_n has real coefficients, then ∣fn(z)∣<∣Tn(i)∣\lvert f_n(z)\rvert<\lvert T_n(i)\rvert for ∣z∣≤1\lvert z\rvert\le1.

The paper remarks (p. 1176) that without real coefficients the corollary can fail, that it cannot determine max⁡∣fn(z)∣\max\lvert f_n(z)\rvert for ∣z∣≤1\lvert z\rvert\le1 in that case, and that the same method shows that ∑k∣ak∣\sum_k\lvert a_k\rvert is maximal for f=±Tnf=\pm T_n among real polynomials bounded by 11 on [−1,1][-1,1]; it quotes Szegő's stronger statement that ∣a2k∣+∣a2k+1∣\lvert a_{2k}\rvert+\lvert a_{2k+1}\rvert is maximal for f=±Tnf=\pm T_n.

Read depth. Claims checked: the statements were read on the print. The proof (pp. 1175-1176) was read but not checked step by step.

Proof pointer

Pages 1175-1176. Write fn(z0)=∑iyili(z0)f_n(z_0)=\sum_i y_il_i(z_0) by Lagrange interpolation at the extremal points of TnT_n, with real yiy_i of absolute value at most 11. A geometric argument shows that the vectors (−1)ili(z0)(-1)^il_i(z_0) pairwise make angles less than π/2\pi/2, because [−1,1][-1,1] subtends an angle at most π/2\pi/2 from z0z_0; so ∣fn(z0)∣\lvert f_n(z_0)\rvert is largest when yi=±(−1)iy_i=\pm(-1)^i, which gives ±Tn(z0)\pm T_n(z_0).