On properties of different notions of centers for convex cones (Q977073)

The paper is concerned with the extension of the notion of axis of a revolution cone to arbitrary cones. The authors consider four notions of centers, each of them having its advantages and inconveniences. The motivation for this study comes from numerical linear algebra, the study of nonsmooth convex bodies, among others. The framework is that of a real reflexive Banach space \(X\) with dual \(X^*\) and duality pairing \(\langle f,x\rangle =f(x)\), \(x\in X\), \(f\in X^*\). By a cone one understands a nontrivial closed convex cone \(K\) in \(X\). An incenter of a cone \(K\) is the center of a certain largest ball inscribed in \(K\). Formally, put \(\rho(K)=\sup_{x\in K\cap S_X} \operatorname{dist}(x,\partial k)\) and let \(\Pi_{\text{inc}}(K) =\{x\in K\cap S_X : \rho(K)= \operatorname{dist}(x,\partial k)\}\) be the set of incenters of \(K\). It turns out that the space \(X\) is rotund iff each solid cone has a unique incenter. In this case, one denotes by \(\pi_{\text{inc}}(K)\) the unique incenter of \(K\) and one shows that the mapping \(\pi_{\text{inc}}\) from the hyperspace of solid cones equipped with the Pompeiu-Hausdorff metric to \((X,\|\cdot\|)\) is continuous, provided that the space \(X\) is uniformly rotund. For \(w\in S_X\) and \(s\in[0,1]\), let \(M(w,s)= \text{cl}[\mathbb R_+(w+sB_X)]\) be the cone generated by the ball \(w+sB_X\). For \(w\in S_X\), put \(g_K(w)=\inf\{s: s\in[0,1]\), \(K\subset M(w,s)\}\) and call a minimizer \(w_0\) of \(g_K\) a circumcenter of \(K\), i.e., a circumcenter of the cone \(K\) is the center of some smallest ball generating \(K\). One shows that the space \(X\) is smooth iff any cone \(K\) is contained in some ball-generated cone. Finally, an inner center of a cone \(K\) is the center of some largest revolution cone contained in \(K\), while an outer center is the center of some smallest revolution cone containing \(K\). A revolution cone is defined by analogy with a revolution cone in \(\mathbb R^n\) by the formula \(\Gamma(y,\theta)=\{x\in X: \langle y,x\rangle\geq \| x\| \cos \theta\}\), for some \(y\in X^*\), \(y\neq 0\), and \(\theta \in[0,\pi/2]\). As the authors mention, the special case of the Euclidean space \(X=\mathbb R^n\) is treated in another paper (an as yet unpublished manuscript), containing also the computations of incenters and circumcenters of several cones arising in concrete applications.