On the precise form of the inverse Markov factor for convex sets

Abstract

Let $K\subset \mathbb{C}$ be a convex compact set, and let $\Pi_n(K)$ be the class of polynomials of exact degree $n$, all of whose zeros lie in $K$. The Tur\'an type inverse Markov factor is defined by $M_n(K)=\inf_{P\in \Pi_n(K)} \left(\|P'\|_{C(K)}/\|P\|_{C(K)}\right)$. A combination of two well-known results due to Levenberg and Poletsky (2002) and R\'ev\'esz (2006) provides the lower bound $M_n(K)\ge c\left(wn/d^2+\sqrt{n}/d\right)$, $c:=0.00015$, where $d>0$ is the diameter of $K$ and $w\ge 0$ is the minimal width (the smallest distance between two parallel lines between which $K$ lies). We prove that this bound is essentially sharp, namely, $M_n(K)\le 28\left(wn/d^2+\sqrt{n}/d\right)$ for all $n,w,d$.