Duality between Fréchet differentiability and strong convexity (Q655980): Difference between revisions

Let \(f:X\to \bar{\mathbb R}\) be a proper convex lower semicontinuous (lsc) function defined on a Banach space \(X\). The function \(f\) is called: (i)\; essentially strictly convex if it is strictly convex on every convex subset of dom\(\,\partial f\) and \(\,(\partial f)^{-1}\) is locally bounded on its domain, and (ii)\; essentially Gâteaux differentiable if int dom\(\,f\neq \emptyset,\; f\) is Gâteaux differentiable on int dom\(\,f\), and \(\|\nabla f(x_j)\|\to\infty\) for every sequence \((x_j)\) in int dom\(\,f\) converging to a boundary point of dom\(\,f\). Rockafellar proved that in the case \(X=\mathbb R^n\) a function \(f\) is essentially strictly convex iff its conjugate \(f^*\) is essentially Gâteaux differentiable. This result was extended to reflexive Banach spaces \(X\) by \textit{H. H. Bauschke, J. M. Borwein} and \textit{P. Combettes} [Commun. Contemp. Math. 3, No. 4, 615--647 (2001; Zbl 1032.49025)]. The author proves in this paper a similar result for Fréchet differentiability. The function \(f\) is called essentially Fréchet differentiable if the conditions from (ii) above hold with Fréchet instead of Gâteaux. The function \(f\) is called essentially strongly convex if it is essentially strictly convex and for every \(x_0\in\) dom\(\,\partial f\) there exist \(\xi_0\in X^*\) and a convex lsc function \(\gamma:[0,\infty)\to[0,\infty],\) with \(\gamma(0) =0\) and \(\gamma(r)>0\) for \(r>0,\) such that \(f(x)\geq f(x_0)+\langle x-x_0,\xi_o\rangle + \gamma(\|x-x_0\|)\) for all \(x\in X.\) If \(f\) is a lsc function defined on an arbitrary Banach space \(X\) and \(f^*\) is proper and essentially Fréchet differentiable, then \(f\) is essentially strongly convex (Theorem 3). If \(X\) is reflexive, then the converse is also true (Theorem 4), that is, \(f\) is essentially strongly convex iff \(f^*\) is proper and essentially Fréchet differentiable. If \(X\) is a locally uniformly convex nonreflexive Banach space, then the function \(f=2^{-1}\|\cdot\|^2\) is essentially strongly convex, but \(f^*=2^{-1}\|\cdot\|_*^2\) is not Fréchet differentiable on \(X^*\) because the Fréchet differentiability of \(f^*\) on \(X^*\setminus\{0\}\) would imply the reflexivity of \(X\). This shows that the reflexivity condition is essential for the validity of Theorem 4.