Continuous subdifferential approximation and its construction (Q1267796)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Continuous subdifferential approximation and its construction
scientific article

    Statements

    Continuous subdifferential approximation and its construction (English)
    0 references
    0 references
    4 May 1999
    0 references
    It is known that the lack of continuity of the Clarke subdifferential mapping of Lipschitz functions create difficulties in the study of optimization methods. In this manner, the author presents a new approach for the construction of continuous approximations of this mapping. Let \(f: \mathbb{R}^n\to \mathbb{R}\) be a locally Lipschitz function, \(\partial f\) be the Clarke subdifferential mapping. Then the convex-compact-valued mapping \(V(u,\varepsilon)\) (\(u\in\mathbb{R}^n\), \(\varepsilon>0\)) is called a continuous approximation to \(\partial f(u)\) iff \hskip 17mm (1) \(V(u,\varepsilon)\) is Hausdorff continuous with respect to \(u\), \hskip 17mm (2) \(\partial f(u)=\text{conv }V_L(u)\), where \(V_L(u)\) is generated by the upper Kuratowski limit of \(V(u,\varepsilon)\), i.e., \[ V_L(u)= \{v\mid \exists u_k\to u,\;\varepsilon_k\to+0,\;v_k\to v: v_k\in V(u_,\varepsilon_k)\}. \] Some properties of this notion are pointed out and it is shown that any uniform continuous approximation to \(\partial f\) introduced 1996 by Xu, Rubinov and Glover is a continuous approximation according to the above definition. Conversely, if \(V(u,\varepsilon)\) is a continuous approximation to \(\partial f\), then by additional assumptions, the mapping \[ Q(u,\varepsilon)= \text{conv }\bigcup\{V(u,t)\mid 0\leq t\leq\varepsilon\} \] is a uniform continuous approximation to \(\partial f\). In the second part of the paper, the author presents a method to construct a special continuous approximation to the subdifferential of a semismooth function using the notion of discrete gradients.
    0 references
    0 references
    Clarke subdifferentials
    0 references
    continuous approximations
    0 references
    locally Lipschitz function
    0 references
    discrete gradients
    0 references