Hopf's ergodic theorem for nonlinear operators (Q916069): Difference between revisions

We show for any order preserving, integral preserving, positively homogeneous and \(L^{\infty}\) nonexpansive mapping on \(L^ 1(E,\mu)\) with \(\mu (E)<\infty\), that \((1/n+1)S_ nf\) is a.e. convergent, where \(S_ nf\) is inductively defined by \(S_ 0f=f\) and \(S_{n+1}f=f+T(S_ nf)\). Furthermore, if \(f\in L^ 2\), then the limit function is just the best approximation in \(L^ 2\) of f with respect to the convex cone of invariant functions. Weaker results are given for mappings, which are not positively homogeneous and only \(L^{\infty}\) norm decreasing.