Performance of two methods for solving separable Hamiltonian systems (Q1841946)

The authors discuss two different approaches to numerically solving the equations of motion of a separable Hamiltonian. The first approach consists of using a fourth-order symplectic partitioned Runge-Kutta method with fixed step size; in the second approach an explicit non-symplectic Runge-Kutta-Nyström method of sixth-order with a fourth-order embedded error estimate is used with variable step size. As test problems two different sets of initial conditions for the Hénon-Heiles system are chosen: one set leading to a chaotic and the other one to a quasi-periodic trajectory. No direct comparison of the two approaches is made; instead the sensitivity with respect to the precision of the computer arithmetic and in the second case with respect to the chosen tolerance is studied separately for each approach. It turns out that for certain choices of the step size and the precision, the symplectic method yields a wrong picture of the long-term solution behaviour.