Proto-derivative formulas for basic subgradient mappings in mathematical programming (Q1332550)

The notion of proto-differentiability for point-to-set mappings [\textit{R. T. Rockafellar}, Ann. Inst. Henri Poincaré, Anal. Non Linéaire 6, No. Suppl., 449-482 (1989; Zbl 0674.90082)] is applied to the subgradient mapping of so-called amenable functions. \(f: \mathbb{R}^ n\to\overline{\mathbb{R}}\) is fully amenable at \(\bar x\) with finite \(f(\bar x)\) iff on some open neighbourhood \(V\) of \(\bar x\) there is a \({\mathbb{C}}^ 2\) mapping \(F: V\to \mathbb{R}^ m\) and a convex, l.s.c. function \(g: \mathbb{R}^ m\to \mathbb{R}\) such that the following abstract constraint qualification is satisfied at \(\bar x\) \[ \text{there is no vector} y\neq 0\text{ in } N_{\text{dom},g} (F(\bar x))\text{ with }\nabla F(\bar x)^* y= 0, \] where \(N_ M (q)\) denotes the normal cone of the convex set \(M\) at the point \(q\in M\) (and additionally \(g\) is piecewise linear- quadratic). The pointwise maximum of a collection of finitely many \({\mathbb{C}}^ 2\) functions, the indicator of a set defined by finitely may \(\mathbb{C}^ 2\) constraints under a constraint qualification, and the sum of a \(\mathbb{C}^ 2\) function with the last mentioned indicator function are fully amenable functions. For these examples specialized formulas for the first and second order epi-derivatives as well as the proto-derivatives of \(\partial f\) are derived. In section 3 comparisons are done with papers of Auslender/Cominetti (1991), Penot (1992) also dealing with derivatives of \(\partial f\).