Second-order directional derivatives for nonsmooth functions (Q1119898)

This is the first in a series of four related papers. The other three have already appeared, and are listed below; for maximum benefit, all four should be studied together. The author defines new upper and lower second-order directional derivatives for a locally Lipschitzian function f: \({\mathbb{R}}^ n\to R\). These are real numbers which depend not only on the function f, the base point \(x^*\), and the direction u, but also on a vector v in Clarke's generalized gradient set \(\partial f(x^*)\). The paper includes sections dealing with convex functions and necessary conditions in unconstrained minimization problems, and develops a formula for the new second-order derivatives of a class of functions introduced by \textit{A. Ben-Tal} and \textit{J. Zowe} [J. Optimization Theory Appl. 47, 483-490 (1985; Zbl 0556.90074)]. (These are the finite sums of functions like \(f(x)=h_ 0(\max \{h_ 1(x),...,h_ m(x)\})\), where each \(h_ j\) is of class \(C^ 2.)\) The other papers of interest are all by the same author and appeared in Math. Program., Ser. A 40, No.1, 95-109 (1988), SIAM J. Control. Optimization 25, 1072-1081 (1987; Zbl 0635.49013), and Math. Oper. Res. 13, 660-673 (1988; Zbl 0671.49013).