Nonlinear duality and best approximations in metric linear spaces (Q580612): Difference between revisions

The object of this paper is the construction of what the author calls ``a nonlinear dual space'' which can be used in the characterization of best approximations from linear subspaces of a real metric linear space as well as for the computation, in special cases, of best approximation with a certain type of algorithm. Let X be a real linear metric space with a translation invariant metric d and let \(q(x):=d(x,0)\) for every \(x\in X\). The mapping \(q: X\to R_+\) is the canonical quasinorm on X and has all the properties of the norm but positive homogeneity. For a real valued function \(\ell\) defined on X, one defines the norm of \(\ell\) by \(\| \ell \|:=\sup \{| \ell (x)| /q(x);\quad x\neq 0\}\) whenever this supremum is finite. Let \(Lip_ k(X)\) denote the set of all Lipschitz continuous odd functions with Lipschitz constant \(k>0\), i.e. the functions \(\ell\) for which \(| \ell (x)-\ell (y)| \leq k\cdot q(x-y),\ell (-x)=-\ell (x)\), for all x,y\(\in X\). The sets \(Lip_ k(X)\) are not closed under the linear operations and therefore are not linear spaces. But if one takes \(Lip(X):=\cup_{k>0}Lip_ k(X),\) the pair (Lip(X), \(\| \cdot \|)\) is a normed linear space although not a Banach space in general. For a fixed \(k>0\) the set \(Lip_ k(X)\) is called ``the nonlinear dual space with index k''. This set is always complete in the topology induced by \(\| \cdot \|\) and is closely connected with the dual spaces of normed linear spaces. As one of the main properties of the nonlinear dual space \(Lip_ k(X)\) the author proves the following Krein-Milman-type theorem: If \(k>0\) is fixed and \(K\subset Lip_ k(X)\) is convex and compact relative to the topology \(T_ k\), then (a) ext(K)\(\neq \emptyset\) and (b) \(K=\overline{co}(ext(K))\), where, as usual, ext(A) denotes the set of all extreme points of a set A and co(B) the convex hull of a set B. Another result is a version for the nonlinear dual spaces of the well- known Hahn-Banach extension theorem. As a consequence of this result it is proved that every functional \(\ell \in Lip_ k(V)\), where V is a linear subspace of X, has a norm-preserving extension to all of X. After investigating various consequences of the Hahn-Banach theorem in the present context related to the problem of characterizing best approximations in metric linear spaces, the author gives characterizations of best approximations by elements of the nonlinear dual space \(Lip_ k(X)\). He remarks that the characterizing properties and their proofs are very similar to the ones known from the theory of normed linear spaces and that the nonlinear dual space behaves, in many ways, like the linear dual space of a normed linear space. However, as the author observes, the question of what the functionals and in particular the extreme functionals look like when one considers special metric linear spaces is open.