Lipschitz condition in minimum norm problems on bounded functions (Q1067133): Difference between revisions

This paper considers the problem of finding optimal Lipschitzian selections in minimum norm problems on bounded functions; it is conceptually similar to the well-known problem of finding continuous selections. Consider the Banach space B of bounded functions f on a set S with uniform norm \(\| \cdot \|\). For each f in B, let \(K_ f\), possibly dependent on f, be a closed convex subset of B. The minimum norm problem is to find an element in \(K_ f\) nearest to f. Let \(G_ f\) be the closed convex set of all elements in \(K_ f\) which are nearest to f. In general, \(G_ f\) is not singleton. The problem is to find an optimal Lipschitzian selection operator \(T: B\to B\) defined by the following three conditions. (i) For each f in B, T(f) is in \(G_ f\). (ii) T is Lipschitzian, i.e., \(\| T(f)-T(h)\| \leq C\| f-h\|\) for all f,h in B for some least number C. (iii) C is smallest among all Lipschitzian operators mapping f to an element in \(G_ f.\) Another problem considered is to find a smallest number D satisfying \(\| \Delta (f)-\Delta (h)\| \leq D\| f-h\|\) for all f,h in B, where \(\Delta (f)=\| f-T(f)\|\) is the shortest distance from f to \(K_ f\). Three cases are considered. Case 1: The greatest convex minorant (gcm). \(S=I=[a,b]\), a real interval. \(K_ f=\{k\) in B: k convex, \(k\leq f\}\). \(\bar f(\)s)\(=\sup \{k(s):\) k in \(K_ f\}\), s in S, is the gcm of f. Then \(T(f)=\bar f\), \(C=1\), \(D=2\). Case 2: Approximation by convex functions. \(S=I\) as above, \(K_ f=K=\{k\) in B: k convex on \(I\}\). Then \(\Delta (f)=(1/2)\| f-\bar f\|\). If \(f'=\bar f+\Delta (f)\) then f' is in \(G_ f\). Then \(T(f)=f'\), \(C=2\), \(D=1\). Case 3: Generalized isotone optimization. S is a partially ordered set and \(K_ f=K\) is the set of all functions in B which satisfy certain isotonicity and nonnegativity conditions on subsets of S. In this case there exist two nearest functions \(f_*\) and \(f^*\), respectively, the minimal and the maximal with \(f_*\leq f^*\). Then \(T(f)=(1/2)(f_*+f^*),\) \(C=1\).