How Hard is it to be a Star? Convex Geometry and the Real Hierarchy
Marcus Schaefer, Daniel Štefankovič
TL;DR
This work proves that testing star-shapedness for non-empty compact smooth semialgebraic regions is $\forall \mathbb{R}$-complete, by leveraging Krasnosel'skiĭ's duality between star-shapedness and common visibility of points. It presents a $orall \mathbb{R}$-membership framework for smooth regions based on tangent-half-space representations and gradient conditions, and develops a robust $\forall \mathbb{R}$-hardness reduction using effective Sard-type perturbations to encode polynomial-zero existence within a unit ball. The results illustrate how convex-geometry dualities can yield unexpected complexity reductions in the real hierarchy, complementing classical Carathéodory/Steinitz/Kirchberger theorems, and open multiple avenues for extending the approach to broader semialgebraic classes and related geometric properties.
Abstract
A set is star-shaped if there is a point in the set that can see every other point in the set in the sense that the line-segment connecting the points lies within the set. We show that testing whether a non-empty compact smooth region is star-shaped is $\forall\mathbb{R}$-complete. Since the obvious definition of star-shapedness has logical form $\exists\forall$, this is a somewhat surprising result, based on Krasnosel'skiĭ's theorem from convex geometry; we study several related complexity classifications in the real hierarchy based on other results from convex geometry.
