A Kantorovich-type convergence analysis for the Gauss-Newton-method (Q1079923): Difference between revisions

The author proves a Newton-Kantorovich type theorem for the Gauss-Newton method (GNM): \(x_{k+1}=x_ k-J^+(x_ k)F(x_ k)\) where \(F: D\subset {\mathbb{R}}^ n\to {\mathbb{R}}^ m\) (n\(\leq m)\) is a smooth map, \(J(x)=F'(x)\) and \(J^+\) is the Moore-Penrose inverse. The theorem in question is analogous to that in the book of \textit{J. Ortega} and \textit{W. C. Rheinboldt} [Iterative solution of nonlinear equations in several variables (1970; Zbl 0241.65046)]. The conclusion is that for an adequate starting point \(x_ 0\) the GNM converges to a zero \(\hat x\) of \(J^+(x)F(x)\). In general \(\hat x\) is not locally unique, but if in addition rank \(J(x_ 0)=n\), then \(\hat x\) is also locally unique.