On sufficiency and duality theorems for nonsmooth semi-infinite mathematical programming problem with equilibrium constraints (Q2089243)

The semi-infinite MPEC \(f(x)\rightarrow \min_{x\in C}\) subject to the semi-infinite constraint \(g(x,t)\le 0 \;\forall t\in \Delta\), the equality constraint \(h(x)=0\) and the MPEC constraint \(G(x)\ge 0\), \(H(x)\ge 0\), \(G(x)^TH(x)=0\) is considered. Using the \(\Psi\)-invexity of \(C\) and the \(\Psi\)-preinvexity of the problem functions \(f,g(\cdot,t), \pm h_j, -G_i, -H_i\) sufficient conditions for global optimality are derived. The main tool is the contingent epi-derivative \(D_\uparrow f(\bar x)v\) in direction \(v\) of the contingend/Bouligand cone \(T(C,\bar x)\) [\textit{J.-P. Aubin} and \textit{H. Frankowska}, Set-valued analysis. Boston etc.: Birkhäuser (1990; Zbl 0713.49021)]. (GA) stationary points together with above \(\Psi\) pre-invexity (Theorem 1) or \(\Psi\)-quasi-convexity (Theorem 2) are global optimal whenever the last sum in the (GA) definition (see below) vanishes (\(\Leftarrow L_u=\emptyset\)). Since the inequality of differentials in Definition 2 ((GA)-stationary vector) of the paper is incomplete, the paper is badly to understand. The inequality for (GA)-stationary \(\bar x\) (generalized alternatively stationary point) should be written (see, e.g., [\textit{Y. Pandey} and \textit{S. K. Mishra}, J. Optim. Theory Appl. 171, No. 2, 694--707 (2016; Zbl 1349.90837), Definition 2.8]) \begin{multline*} D_{\uparrow} f(\bar x)v+\sum_{i\in I_k}\lambda^g_iD_{\uparrow}g(\bar x, t_i)v+\sum_{j\in I_n}\left[\lambda^h_jD_{\uparrow}h_j(\bar x)v+\mu^h_jD_{\uparrow}(-h_j)(\bar x)v\right] \\ +\sum_{i\in I_p}\left[\lambda^G_iD_{\uparrow}(-G_i)(\bar x)v+\lambda^H_iD_{\uparrow}(-H_i)(\bar x)v\right]+\sum_{i\in I_p}\left[\mu^G_iD_{\uparrow}(G_i)(\bar x)v+\mu^H_iD_{\uparrow}(H_i)(\bar x)v\right]\ge 0 \end{multline*} for all \(v\in T(C,\bar x)\). In Section 4, weak (Theorem 3, Theorem 5) and strong Mond-Weir/Wolfe duality (for local solutions in Theorem 4, Theorem 6) results are shown under the above invexity/quasi-convexity conditions and \(L_\mu=\emptyset\). For strong duality results, some generalized Slater condition is additionally necessary. Also, in Definition 4 of the Mond-Weir/Wolfe duality, the above last sum with the multipliers \(\mu_i^G,\mu_i^H\) must be added for understanding the subsequent formulations. However, there is no discussion whenever for a given problem the index set \(L_\mu\) is empty being a crucial condition for the validity of Theorem 1--Theorem 6.