Wiki
Wiki

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

Updated


Setting

Notation (1), p. 51. For a natural number nn, complex numbers c0,…,cn−1c_0,\ldots,c_{n-1} and r>0r>0,

Pn(z)=zn+cn−1zn−1+⋯+c1z+c0,L(Pn,r)={z:∣Pn(z)∣=rn}.P_n(z)=z^n+c_{n-1}z^{n-1}+\cdots+c_1z+c_0,\qquad L(P_n,r)=\{z:|P_n(z)|=r^n\}.

The paper notes (p. 51) that L(Pn,r)L(P_n,r) consists of m≤nm\le n closed Jordan curves σ1,…,σm\sigma_1,\ldots,\sigma_m with pairwise disjoint interiors, which may meet only at zeros of Pn′P_n', and its length is ∣L(Pn,r)∣=∣σ1∣+⋯+∣σm∣|L(P_n,r)|=|\sigma_1|+\cdots+|\sigma_m|.

Statement

Theorem 1 (p. 52). For every such PnP_n and r>0r>0,

∣L(Pn,r)∣≤2πnr.(3)|L(P_n,r)|\le 2\pi nr. \tag{3}

The paper places this against earlier upper bounds (p. 51): Dolzhenko's ∣L(Pn,r)∣≤4πnr|L(P_n,r)|\le 4\pi nr, Pommerenke's ∣L(Pn,1)∣≤74n2|L(P_n,1)|\le 74n^2, Borwein's ∣L(Pn,1)∣≤8πen|L(P_n,1)|\le 8\pi en and Eremenko and Hayman's ∣L(Pn,1)∣≤9.173n|L(P_n,1)|\le 9.173n.

The conjectured extremal case (p. 52). The paper records as a natural conjecture, crediting it to its references [2]--[4] and [6], that among all L(Pn,1)L(P_n,1) the Bernoulli-type lemniscate L(Bn,1)L(B_n,1), Bn(z)=zn−1B_n(z)=z^n-1, is longest, and computes in polar coordinates

∣L(Bn,1)∣=21+1/n∫0π/2cos⁡1/n−1t dt=21/nB(12,12n)=21/nπ Γ(12n)Γ(12+12n)=2n+4ln⁡2+O(1n).|L(B_n,1)|=2^{1+1/n}\int_0^{\pi/2}\cos^{1/n-1}t\,dt =2^{1/n}B\Bigl(\frac12,\frac1{2n}\Bigr) =2^{1/n}\sqrt\pi\,\frac{\Gamma(\frac1{2n})}{\Gamma(\frac12+\frac1{2n})} =2n+4\ln2+O\Bigl(\frac1n\Bigr).

Proof pointer

P. 52: the inequality follows at once from Lemmas 1--3. By Lemma 1, ∣L∣≤Ψ(L)γ(L)|L|\le\Psi(L)\gamma(L), the secant variation times the analytic capacity. Lemma 2 (p. 55) gives γ(L(Pn,r))≤r\gamma(L(P_n,r))\le r: with ff the Ahlfors function of the unbounded complementary component, fnPnf^nP_n is bounded by rnr^n there and equals γn\gamma^n at infinity. Lemma 3 (p. 55) gives Ψ(L(Pn,r))≤2πn\Psi(L(P_n,r))\le 2\pi n for every r>0r>0: after a Möbius change of variable the image of the lemniscate is a level line of a rational function of degree at most nn, which meets each line through the origin in at most 2n2n points. Multiplying the two bounds gives (3).

Dependencies

Lemma 1 (p. 53); Lemma 2 and Lemma 3 (p. 55); the definitions of Ψ\Psi and γ\gamma (pp. 52--53).

Source. V. I. Danchenko, The lengths of lemniscates. Variations of rational functions, Mat. Sb. 198 (2007), no. 8, 51--58 (in Russian); pages are the journal's, as on the source card.

Read depth. Claims checked: the statement, the setting (1) and the Bernoulli computation read on the print; the proofs of Lemmas 1--3 (pp. 53--55) read but not checked step by step. Nothing here is independently reviewed.

Bears on

  • #114: at r=1r=1 the theorem bounds the length of {z:∣p(z)∣=1}\{z:|p(z)|=1\} by 2πn2\pi n for every monic pp of degree nn. The conjectured maximiser zn−1z^n-1 has length 2n+4ln⁡2+O(1/n)2n+4\ln2+O(1/n), so the bound exceeds it by a factor tending to π\pi as nn grows; the theorem does not decide whether zn−1z^n-1 maximises the length.