On the existence of saddle points for nonlinear second-order cone programming problems (Q2351524): Difference between revisions

The following non-linear second-order cone programming problem is considered: Minimize \(f(x)\) subject to \[ g_j(x) ~ \geq_{ K_{m_j}+1} ~0,~ 1 \leq j \leq J,~ h(x) = 0, \] where \( f: R^n \longrightarrow R, h: R^n \longrightarrow R^l,~ g_j: R^n \longrightarrow R^{m_j+1}\) are twice continuously differentiable functions, and \(K_{m_j + 1} \) is the second order cone in \(R^{m_j + 1}\) for all \(j\), which is defined as follows: \[ K_{m_j + 1} = \{ (x_1, x_2) \in R \times R^{m_j} ~|~ \left\|x_2 \right\| \leq x_1 \} . \] The existence of local and global saddle points for this problem is proved using the second-order sufficient conditions with an augmented Lagrangian. By a perturbation technique, the existence of global saddle points is established, which can be applied for multiple optimal solutions. The relationship between global saddle points and exact penalty representations is discussed in the concluding part of the paper.