The Multidirectional Mean Value Theorem in Banach Spaces

From MaRDI portal





In 1994, \textit{F. H. Clarke} and \textit{Yu. S. Ledyaev} [Trans. Am. Math. Soc. 344, No. 1, 307-324 (1994; Zbl 0803.49018)] presented a so-called multidirectional mean value theorem in which an estimation for the difference \(f(y)- f(x)\) is given where \(y\) is no longer the end of an interval but ranges over a (compact) set \(Y\). The result was formulated by means of the notion of proximal subgradients. In case of a smooth function the assertion can be expressed shortly by the existence of a point \(z\in[x,Y]:= \text{conv}(\{x\}\cup Y)\) such that \[ \min_{y\in Y} f(y)- f(x)\leq \langle f'(z),y- x\rangle\quad\forall y\in Y. \] In the present paper, similar results are given for locally Lipschitz functions on a Banach space \(X\) which admits a smooth bump function (i.e., a smooth function with bounded nonempty support). In the main theorem it is shown that for any point \(x\in X\), for any closed, bounded, convex set \(Y\subseteq X\) and for any number \(\delta>0\) there exist a vector \(z\) in a \(\delta\)-neighborhood of \([x,Y]\) and a Fréchet subgradient \(\xi\in\partial_F f(z)\) such that \[ \inf_{y\in Y} f(y)- f(x)\leq \langle\xi, y-x\rangle+\delta\quad\forall y\in Y. \] The proof is based on a special version of the Borwein-Preiss smooth variational principle and on some Fréchet subdifferential sum and chain rules.











This page was built for publication: The Multidirectional Mean Value Theorem in Banach Spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4342785)