Optimality conditions for nonconvex nonsmooth optimization via global derivatives (Q1985279)

The lower and upper global directional derivatives of a proper function \(h:\mathbb{R}^{n}\rightarrow \overline{\mathbb{R}}\) at \(\overline{x}\in\mathrm{dom}\, h\) in the direction \(u\in \mathbb{R}^{n}\) are defined by \(h_{\epsilon }(\overline{x};u):=\inf_{t\in ]0,\epsilon ]}\frac{h(\overline{x}+tu)-h(\overline{x})}{t}\) and \(h^{\epsilon }(\overline{x};u):=\sup_{t\in ]0,\epsilon ]}\frac{h(\overline{x}+tu)-h(\overline{x})}{t}\), respectively. Among other properties of these notions, the author proves the following mean value results: if \(x,y\) are two different points in \(\mathbb{R}^{n}\) such that \([x,y]\subseteq\mathrm{dom}\, h\) and the restriction of \(h\) to \([x,y]\) is upper (lower) semicontinuous, then there exist \(z\in \lbrack x,y[\) (resp., \(w\in ]x,y]\)) and \(\epsilon >0\) such that \(h^{\epsilon }(z;y-x)\leq h(y)-h(x)\) (resp., \(h(y)-h(x)\leq h_{\epsilon }(w;y-x)\)). These results generalize the well known mean value theorem due to \textit{W. E. Diewert} [in: Generalized concavity in optimization and economics, Proc. NATO Adv. Study Inst., Vancouver/Can. 1980, 51--93 (1981; Zbl 0539.90088)]. The paper also presents local and global optimality conditions for both unconstrained and constrained nonconvex minimization problems in terms of global directional derivatives.