scientific article; zbMATH DE number 3908176

From MaRDI portal
Publication:3686508





For a system of convex inequalities \(\phi_ i(x)\leq 0\) \((i=1,...,m)\), which satisfies Slater's condition (\(\exists z: \phi_ i(z)<0\), \(\forall i)\), the method is proposed: \[ x^{(k+1)}=x^{(k)}- 2(\sum^{m}_{i=1}\nu_{ik}\phi_ i(x^{(k)})/\| \sum^{m}_{i=1}\nu_{ik}g_ i(x^{(k)})\|^ 2)\sum^{m}_{i=1}\nu_{ik}g_ i(x^{(k)}), \] where \(\{\nu_{ik}\}^ m_{i=1}\) must satisfy two requirements for every k: (1) if \(\phi_ i(x^{(k)})>0\), then \(\nu_{ik}\geq 0\), else \(\nu_{ik}=0\); (2) \(\sum^{m}_{i=1}\nu_{ik}>0.\) Theorem 1. If \(\{\phi_ i\}^ m_{i=1}\) are defined on the real Hilbert space H, \(g_ i(x)\) is the continued Frechet's derivative of \(\phi_ i\) in x, and \(x^{(0)}\) is some point from H, then the method gives some solution of the system in a finite number of steps. Theorem 2. If \(\{\phi_ i\}^ m_{i=1}\) are defined on the Euclidean real space \(R^ n\), \(g_ i(x)\) is some subgradient of \(\phi_ i\) in x and \(x^{(0)}\) is some point from \(R^ n\), then the method gives some solution of the system in a finite number of steps. Remark. Later the authors found out that the requirement on \(\{\nu_{ik}\}^ m_{i=1}\) may be replaced by: (1) if \(\phi_ i(x^{(k)})\geq 0\), then \(\nu_{ik}\geq 0\), else \(\nu_{ik}=0\); (2) \(\sum^{m}_{i=1}\nu_{ik}>0\).











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3686508)