Attainment and (sub)differentiability of the supremal convolution of a function and square of the norm (Q432421): Difference between revisions
From MaRDI portal
Latest revision as of 10:03, 5 July 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Attainment and (sub)differentiability of the supremal convolution of a function and square of the norm |
scientific article |
Statements
Attainment and (sub)differentiability of the supremal convolution of a function and square of the norm (English)
0 references
4 July 2012
0 references
Let \((X,\|\cdot\|)\) be a real Banach space. For a function \(f:X\to \mathbb R\cup\{\infty\}\) one defines the infimal (supremal) convolution of \(f\) with the \(p\)'th power of the norm at a point \(x\in X\) by \((f\square \|\cdot\|^p)(x)=\inf\{f(z)+\|x-z\|^p : z\in X\}\) (resp. \((f\Delta \|\cdot\|^p)(x)=\sup\{-f(z)+\|x-z\|^p : z\in X\}\)). It is obvious that \(-(f\Delta \|\cdot\|^p)=f\square(-\|\cdot\|^p).\) Also, taking \(f\) to be the indicator function of a nonempty subset \(K\) of \(X\), \(f\square \|\cdot\|^p\) is the \(p\)'th power of the distance function to \(K\), while \(f\Delta \|\cdot\|^p \) is the \(p\)'th power of the farthest distance function to \(K\). The differentiability properties of the infimal convolution were studied by the authors in a previous paper [J. Math. Anal. Appl. 368, No. 2, 538--550 (2010; Zbl 1195.46043)]. The present paper is concerned with the differentiability properties of the supremal convolution of a function \(f\) with the square of the norm. Throughout the paper it is assumed that \(f\) has a certain property, called property (Q), meaning that \(f\) is proper, lsc, bounded from below and \(\liminf_{\|z\|\to \infty}f(z)/(\|z\|^2)>1.\) Also, one says that \((f\Delta \|\cdot\|^p)(x)\) is attained if there exists \(z\in X\) such that \((f\Delta \|\cdot\|^p)(x)=-f(z)+\|x-z\|^2,\) and that it is attained strongly if, further, \(\lim\|x-z_n\|=0\) for every maximizing sequence \((z_n)\) in \(X.\) Under these hypotheses, \(\partial(f\Delta \|\cdot\|^2)(x)=\partial(\|\cdot\|^2)(x-z).\) Moreover, \(\|\cdot\|^2\) is Gâteaux (Fréchet) differentiable at the point \(x-z\) iff \(f\Delta \|\cdot\|^2 \) is Gâteaux (Fréchet) differentiable at the point \(x,\) and, in this case, \((f\Delta \|\cdot\|^2 )'(x)=2\|x-z\|\, \|\cdot\|'(x-z)\) (Proposition 1). In Theorem 3 one proves that if \(\|\cdot\|\) is Fréchet differentiable and LUR, then \((f\Delta \|\cdot\|^2)(x) \) is attained strongly at \(z\) iff \(f\Delta \|\cdot\|^2\) is Fréchet differentiable at \(x-z.\) Some generic differentiability results, containing as particular cases generic differentiability results of the farthest point mapping, are also proved. The obtained results extend some results obtained for the farthest point mapping by \textit{S. Dutta} and \textit{P. Bandyopadhyay} [Bull. Aust. Math. Soc. 71, No. 3, 425--433 (2005; Zbl 1084.46010)] and \textit{S. Dutta} [Proc. Am. Math. Soc. 133, No. 10, 2949--2955 (2005; Zbl 1088.49012)].
0 references
supremal convolution
0 references
strong attainment
0 references
dense \(G_\delta\) set
0 references
subdifferential
0 references
farthest distance function
0 references
densely remotal set
0 references
Fréchet and Gâteaux differentiability
0 references
locally uniformly rotund norm
0 references
0 references