On the convergence of the penalty function method with respect to a functional (Q1281223)

A set of mathematical programming problems depending on a vector parameter \(\lambda=(\lambda_1,\lambda_2,\ldots,\lambda_m)\) \[ \begin{aligned} R(\lambda)&:\sup\{ f(x)\mid x\in A(\lambda)\} \;(=\varphi (\lambda))\\ A(\lambda)&=\{ x\in E_{n}\mid f_{i} (x)\leq\lambda_{i},\;i\in I=\{ 1, 2, \ldots , m \} \}, \end{aligned}\tag{1} \] where \( f, f_{i} \) are functions defined and finite in each point of \(E_n\) , is considered. The problem of obtaining \(R(0)\) is approximated by a set of unconditional extreme problems \[ \sup\{ F(x, \alpha)\mid x\in E_{n} \} (=\widehat{O}(\alpha)), \quad F(x, \alpha)=f(x)- \alpha \cdot v(x), \tag{2} \] where \(v(x)=\sum_{i \in I} [\max \{ f_{i} (x), 0 \} ]^{p} , p>0 \) is the penalty function of a set of \(A'(0)\) from (1). It is proved that the problem (2) converges to the problem (1) at \(\alpha \to \infty \). In this paper also two theorems of convergence of the penalty functions method in a functional are proved and the theorem of convergence of the method in relation to an argument is formulated.