Preservation of phase space structure in symplectic integration : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Mathematics at Massey University, Palmerston North, New Zealand
This thesis concerns the study of geometric numerical integrators and how they preserve
phase space structures of Hamiltonian ordinary differential equations.
We examine the invariant sets of differential equations and investigate which numerical
integrators preserve these sets, and under what conditions. We prove that when periodic
orbits of Hamiltonian differential equations are discretized by a symplectic integrator they
are preserved in the numerical solution when the integrator step size is not resonant with
the frequency of the periodic orbit.
The preservation of periodic orbits is the result of a more general theorem which proves
preservation of lower dimensional invariant tori from dimension zero (fixed points) up to
full dimension (the same as the number of degrees of freedom for the differential equation).
The proof involves first embedding the numerical trajectory in a non-autonomous
then applying a KAM type theorem for
flows to achieve the result. This avoids having to
prove a KAM type theorem directly for the symplectic map which is generally difficult to do.
We also numerically investigate the break up of periodic orbits when the integrator's
step size is resonant with the frequency of the orbit.
We study the performance of trigonometric integrators applied to highly oscillatory
Hamiltonian differential equations with constant frequency. We show that such integrators
may not be as practical as was first thought since they suffer from higher order resonances
and can perform poorly at preserving various properties of the di fferential equation. We
show that, despite not being intended for such systems, the midpoint rule performs no
worse than many of the trigonometric integrators, and indeed, better than some.
Lastly, we present a numerical study of a Hamiltonian system consisting of two magnetic
moments in an applied magnetic field. We investigate the effect of both the choice of
integrator and the choice of coordinate system on the numerical solutions of the system.
We show that by a good choice of integrator (in this case the generalised leapfrog method)
one can preserve phase space structures of the system without having to resort to a change
of coordinates that introduce a coordinate singularity.