Exact numerical studies of Hamiltonian maps: Iterating without roundoff error (Q1197499)

The study of iterations of maps in general and of Hamiltonian maps in particular, is often a matter of interpreting numerous numerical experiments. If these are done with floating-point arithmetic, as is usually the case, roundoff errors may well lead to an inaccurate picture, particularly with respect to long-terms behaviour. The authors present a well-argued case in favour of approaching the study of Hamiltonian maps via lattice maps which are themselves Hamiltonian and tend towards the given map when the lattice spacing is decreased. Using the standard map as an example, they extensively illustrate how a number of unnatural features are avoided with lattice maps. For example, chaotic trajectories will not artificially be trapped on a short regular orbit, or will not drift across invariant curves (in two dimensions). The lattice maps also exhibit a transition to global stochasticity in such a way that the critical value of the stochasticity parameter approaches the known critical value for the standard map. Another fourdimensional example is considered to illustrate how lattice maps can be used to study Arnold diffusion in a Hamiltonian system with three degrees of freedom. Finally, more general \(2N\)-dimensional lattice maps are discussed in the context of obtaining Hamiltonian numerical approximations to Hamiltonian flows. A peculiar map, which needs further study, is the one represented by the section numbers.