Continuous approximations to generalized jacobians (Q4944357)

The well-known properties of \(\varepsilon -\)subdifferentials as some kind of numerically stable, and upper semi-continuous approximation of subdifferentials are used to define three Hausdorff upper semicontinuous approximations (not necessarily convex valued) of the Jacobian of a non-smooth Lipschitz map between finite dimensional spaces defined on a compact set (sect. 3 \textit{uniform}, sect. 4 \textit{strong}, sect. 5 \textit{weak} \textit{approximation}, weaker in this order). If such an approximation is known for the nonsmooth part of a composition the approximation of the composition is given similar to that one of the usual subgradient (chain rule). If the considered subdifferential is a Cartesian product then its above approximations have this property, too. Above that a \textit{pertinent weak} approximation of a composition of two Lipschitz maps is given by the product of the pertinent weak approximation of each map (chain rule). They can be used to construct point-based set-valued approximations. The strong approximation is also uniform for Lipschitz maps. In this case the family of \(\varepsilon -\)generalized Jacobians gives a strong (and uniform) approximation of Clarke's generalized Jacobian. The paper generalizes considerations in [\textit{A. M. Rubinov}, Lect. Notes Econ. Math. Syst. 255, 74-86 (1985; Zbl 0601.49007)], [\textit{V. F. Demyanov} and \textit{A. M. Rubinov}, ``Constructive nonsmooth analysis'' (1995; Zbl 0887.49014)], [\textit{A. D. Ioffe}, Trans. Am. Math. Soc. 266, 1-56 (1981; Zbl 0651.58007)] and [\textit{H. Xu} and \textit{X. W. Chang}, J. Optimization Theory Appl. 93, No. 2, 373-394 (1997; Zbl 0899.90150)].