Newton-Goldstein convergence rates for convex constrained minimization problems with singular solutions (Q2266355)

The famous Newton iteration principle, which has been applied in a large number of equations, is a way to solve some constrained minimization problems in a Banach space, too. Following the general recursive quadratic minimization scheme: \(x_{n+1}=x_ n+\omega_ n(\hat x_ n- x_ n),\) \(\omega_ n\in [0,1],\) \(\hat x_ n\in \min_{y\in \Omega} Q(M_ n,x_ n,y)\) \(Q(M,x,y)=<F'(x),y-x>+1/2<M(y-x),y-x>,\) where M is a nonnegative bounded linear mapping from X to \(X^*\) (the dual of X), F and \(\Omega\) are the objective function and the feasible set of the minimization problem and \(<u,v>\) is the action of the linear functional u in \(X^*\) on a vector v in X, one can solve the problem of minimization of F in the feasible set \(\Omega\). In this very interesting work the authors test the convergence rates of the Newton-Goldstein sequence for the problem of minimization with singular solutions \(\xi\), that is to say the solutions at which the local quadratic approximation Q(\(\xi\),x) to the objective function F grows more slowly than \(\| x-\xi \|^ 2\), for admissible vectors x near \(\xi\). Generally for iterative minimization methods with quadratic subproblems the first author [Convergence rate analysis for iterative minimization schemes with quadratic subproblems, Ph. D. Thesis, Math. Dept. North Carolina State University (1981)] has shown that the values \(r_ n=F(x_ n)-\inf_{\Omega} F\) are at least of order \(O(n^{-1/3})\). In this work the authors estimate \(r_ n=O(n^{-1/2})\) when F'', the second Fréchet differential, is Lipschitz continuous and the admissible set is bounded. They prove that the Newton-Goldstein sequence can converge superlinearly to \(\xi\), if \(<F'(\xi),x-\xi >\geq A\| x-\xi \|^{\nu},\) for some \(A>0\) and \(\nu\in (2,2.5)\) and they prove that this condition on F'(\(\xi)\) is a natural one for a nontrivial class of constrained minimization problems. Full of nontrivial results, based on a large number of references, this work stimulates the reader.