Newton's methods from a geometric point of view (Q2732314)

Newton's method of computation of the zeros of a mapping \(f\in C^k(\mathbb{R}^n,\mathbb{R}^n)\) is generalized to the manifold setting. Many results obtained in \(\mathbb{R}^n\) are translated to a manifold \(M\) equipped with a vector bundle \(E\to M\) and a linear connection on \(E\). Newton's method is defined as a transformation associated to Newton's vector field \(N(s)\) on \(E\) with a section \(s\) of \(E\) such that \(N(s)\) keeps the equilibrium points of \(s\). Properties of Newton's vector field are analyzed and results about the convergence (and probabilistic convergence) of trajectories of \(N(s)\) are given. The behaviour of the singular set of \(s\) is also characterized via Newton's vector field.