Necessary and sufficient KKT optimality conditions in non-convex optimization (Q2361126): Difference between revisions

The author provides necessary and sufficient Karush-Kuhn-Tucker (KKT) optimality conditions for the non-convex optimization problem: \[ \min_x\{f(x): g_j(x)\leq 0, \,j=1,2,\dots, m\},\quad x\in \mathbb{R}^n, \] where \(f,g_j:\mathbb{R}^n\to \mathbb{R}\), \(j=1,2,\dots, m\). The author proves that if Slater's condition and a non-degeneracy assumption are satisfied together with a suitable condition of the feasible set, then, without any convexity of the functions involved as well of the feasible set, the KKT optimality conditions are necessary. Furthermore, the non-trivial KKT optimality conditions are globally sufficient for the point \(x\) to be a global minimizer, provided that the strict level set \(L^<_f(x)\) of the objective function is convex. In particular, if \(L_f(x)\) is strictly convex at \(x\), then \(x\) is a strict global minimizer for the optimization problem.