\(\varepsilon\)-optimality criteria for convex programming problems via exact penalty functions (Q1199753)

The main result states that, for the convex programming problem (P): minimize \(f(x)\), subject to \(g_ i(x)\leq 0\) \((1\leq i\leq m)\), where \(f,g_ i: R^ n\to R\) are convex functions, if the feasible set is nonempty and \(\bar x\in R^ n\) is an \(\varepsilon\)-minimal point of the exact penalty function \(\theta(x,\rho)=f(x)+\rho\sum^ m_{i=0}\max(0,g_ i(x))\) for any \(\rho\geq \rho_ 0\) then it is also an \(\varepsilon\)-optimal solution for (P) and there exists a nonnegative vector \(\lambda=(\lambda_ 1,\dots,\lambda_ m)\in R^ m\) such that \(0\in\partial_{\varepsilon_ 0}f(\bar x)+\sum^ m_{i=1}\partial_{\varepsilon_ i}(\lambda_ ig_ i(\bar x)\) for some \(\varepsilon_ i\geq 0\) \((i=0,\dots,m)\) with \(\sum^ m_{i=0}\varepsilon_ i-\varepsilon\leq\sum^ m_{i=1}\lambda_ i g_ i(\bar x)\leq 0\) \((\partial_{\varepsilon_ i}\) stands for \(\varepsilon_ i\)-subdifferential in the sense of convex analysis).