Generalized derivatives of distance functions and the existence of nearest points (Q1006692): Difference between revisions

The authors study the relevance of the generalized derivatives of Clarke, Michel--Penot and Dini of the distance function, for the existence and uniqueness of nearest points, a direction of research initiated by \textit{S.\,Fitzpatrick} [Bull.\ Aust.\ Math.\ Soc.\ 22, 291--312 (1980; Zbl 0437.46012); ibid.\ 39, No.\,2, 233--238 (1989; Zbl 0674.46011)]. For a nonempty closed subset \(G\) of a real Banach space \(X\), put \(d_G(x)=\inf\{\|g-x\|: g\in G\}\) and \(P_G(x)=\{g\in G: \|g-x\|=d_G(x)\}\). For \(v\in S_X\) (the unit sphere of \(X\)), one considers the Clarke derivatives \( d^\circ_G(x;v)\) and \( (d_G)_\circ(x;v)\), the Dini derivatives \( d^+_G(x;v)\) and \( (d_G)_+(x;v)\), and the Michel--Penot derivatives \(d^\diamond_G(x;v)\) and \( (d_G)_\diamond(x;v)\). The authors consider also the modified Dini derivatives given by \( (d_G)^D(x;v)=\sup_{u\in X}\{ (d_G)^+(x;v+u)- (d_G)_+(x;v+u)\), and \( (d_G)_D(x;v)\) defined similarly, but with the infimum instead of the supremum. For convenience, put \(\mathcal R=\{\circ,+,\diamond,D\}\) and denote by \(\mathcal C(X)\) the family of all nonempty closed subsets of \(X\). As a sample of the results proved in the paper, we present Theorem 3.1, asserting that for \(v\in S_X\) and \(\beta \in \mathcal R\setminus \{+\}\), the following are equivalent: {\parindent=8mm \begin{itemize}\item[(i)] \(\forall G\in \mathcal C(X)\), \(\forall x\in X\setminus G\), if there exists \(\alpha \in \mathcal R\) such that \(d_G^\alpha(x;v)=1\), then \(P_G(x)\neq \emptyset\); \item[(ii)] \(\forall G\in \mathcal (X)\), \(\forall x\in X\setminus G\), if \(d_G^\beta(x;v)=1\), then \(P_G(x)\neq \emptyset\); \item[(iii)] \(\forall G\in \mathcal (X)\), \(\forall x\in X\setminus G\), if \(d_G^\alpha(x;v)=1\) for all \(\alpha \in \mathcal R\), then \(P_G(x)\neq \emptyset\); \item[(iv)] \(\forall G\in \mathcal (X)\), \(\forall x\in X\setminus G\), if \(d_G^\alpha(x;v)=1\) for all \(\alpha \in \mathcal R\), then \(G\) is approximatively compact for \(x\); \item[(v)] \(X\) is compactly uniformly convex at \(v.\) \end{itemize}} The set \(G\) is called approximatively compact for \(x\) if every minimizing sequence for \(d_G(x)\) contains a convergent subsequence. If this happens for every \(x\in X\), then \(G\) is called approximatively compact. The space \(X\) is called compactly locally uniformly convex for \(v\in S_X\) if any sequence \((x_n)\) in \(S_X\) such that \(\lim_n\|x_n+v\|=2\) has a convergent subsequence.