New constructions for local approximation of Lipschitz functions. II (Q875256)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | New constructions for local approximation of Lipschitz functions. II |
scientific article |
Statements
New constructions for local approximation of Lipschitz functions. II (English)
0 references
13 April 2007
0 references
In a former paper, the author presented a new generalized differentiability notion for Lipschitz functions \(f: \mathbb{R}^n\to\mathbb{R}\) according to \[ \begin{multlined} Df(x_0)= \text{conv}\Biggl\{v\in\mathbb{R}^n\mid\exists g\in \mathbb{R}^n,\| g\|= 1,\exists r(x_0,\cdot,g)\in \eta(x_0),\;\exists\alpha_k\to +0:\\ v= \lim_{k\to\infty}\,{1\over \alpha_k} \int^{\alpha_k}_0\nabla f(r(x_0,\tau, g))\,d\tau\Biggr\},\end{multlined} \] where \(\eta(x_0)\) is the set of all smooth curves \(r(x_0,\tau,g)\) which converges (for \(\tau\to+ 0\)) against \(x_0\) from the direction \(g\). It was shown that \(Df(x_0)\) is a closed bounded set which is contained in the Clarke subdifferential \(\partial_{\text{Cl}}f(x_0)\). Moreover, if \(f\) is a d.c.-function (i.e. if \(f\) is represented as the difference of two convex functions \(f_1\) and \(f_2\)) then \[ \partial f_1(x_0)-\partial f_2(x_0)\subset Df(x_0)= \partial_{Cl}f(x_0). \] In the present paper, two further approximations are introduced according to \[ \begin{multlined} D_\alpha f(x_0)= \overline{\text{conv}}\Biggl\{v\in\mathbb{R}^n\mid \exists g\in\mathbb{R}^n,\| f\|= 1,\exists r(x_0,\cdot g)\in \eta(x_0):\\ v={1\over \alpha} \int^\alpha_0\nabla f(r(x_0, \tau,g))\,d\tau\Biggr\}\end{multlined} \] and \(\widetilde D_\delta f(x_0)= \overline{\text{conv}}\bigcup_{\alpha\in[0, \delta]} D_\alpha f(x_0)\). Obviously, for \(\alpha= 0\) it is \(D_0 f(x_0)= \widetilde D_0 f(x_0)= Df(x_0)\). For convex functions, the set \(\widetilde D_\delta f(x_0)\) recalls the \(\varepsilon\)-subdifferential of \(f\). It is shown that \(D_\alpha f(\cdot)\) is a Lipschitz set-valued mapping. Moreover, the set-valued mapping \(\widetilde D_\delta f(\cdot)\) is a uniform continuous extension of the Clarke subdifferential mapping \(\partial_{\text{Cl}}f(\cdot)\) if \(f\) is a d.c.-function. At the end of the paper, the author presents a simple optimization algorithm where each accumulation point of the generated sequence is a stationary point with respect of the Clarke subdifferential.
0 references
Lipschitz functions
0 references
Lipschitz set-valued mappings
0 references
directional derivatives
0 references
\(\varepsilon\) -subdifferential for convex functions
0 references
continuous extension of Clarke subdifferential
0 references
optimization process
0 references