Centerlines of regions in the sphere (Q1030180)

Let \(\mathbb S\) denote a Euclidean 2-sphere. For two disjoint closed sets \(X,Y \subseteq \mathbb S\), it is shown that the equidistant set \(E(X,Y)\) is a 1-manifold if and only if \(X\) and \(Y\) are non-interlaced, which means that the \(X\)-neighbours and the \(Y\)-neighbours of any equidistant point always belong to two disjoint intervals in a circle. This implies a theorem of \textit{E. T. Bell} [Rev.~Columb. Mat.~9, 125--135 (1975; Zbl 0331.54023) and Rev.~Columb. Mat.~10, 93 (1976)] which asserts that the equidistant set of two disjoint continua in \(\mathbb S\) is a connected 1-manifold. The result is applied to the study of open sets \(U\subseteq \mathbb S\) such that every maximal disk in \(U\) hits the boundary in at most two points and \({\mathbb S}\setminus U\) has at least two points. The collection of these sets is denoted by \({\mathcal U}_2\). For example, a circular ring and an ellipse that is not a circle belong to \({\mathcal U}_2\). It is shown that the boundary of \(U\in {\mathcal U}_2\) has at most two components. If the number of components is exactly two, then \(\partial U\) is locally connected, and its components have cut points of order two only. For any \(U \in {\mathcal U}_2\), the set of points in \(U\) having exactly two neighbours in the complement is a connected smooth non-empty 1-manifold.