Calculation of fixed points of extremal mappings by gradient-type methods (Q1974721): Difference between revisions

A class of skew-symmetric functions similar to the class of convex functions in nonlinear programming is introduced for an extremal problem in the form \[ \nu^*\in\text{argmin}\bigl \{\Phi(\nu^*,w)\mid w\in\Omega \bigr\} \] where \(\nu,w\in\mathbb{R}^n\), \(\Omega\subset\mathbb{R}^n\) is a convex closed set. It is shown that a condition holds in this class that ensures the stability of a fixed point. The gradient method for finding fixed point \(\nu^*\) of problem is proposed in the form \[ \nu^{k+1} =\pi_\Omega (\nu^k)-\alpha \nabla_w \Phi(\nu^k, \nu^k) \] where \(\pi_\Omega (\cdot)\) is the operator of projection onto \(\Omega\). \(\alpha>0\) is a parameter and \(\nabla_w \Phi(\nu,w)\) is the partial gradient of \(\Phi(\nu,w)\) with respect to \(w\). Convergence of the gradient methods to sharp, quadratic and degenerate equilibrium points of extremal mapping is proved.