Discretization of dynamical systems with first integrals (Q379514)

It is very well known that numerical methods preserve hyperbolic invariant subsets of ordinary differential equations but can fail in approaching the right solutions when non-hyperbolic dynamics is present. The simplest and classical example is the use of the Euler method to simulate the dynamics of a linear center: discretization wrongly shows an expanding behaviour. The main goal of this paper is to show that there is a simple principle behind such facts. The authors prove that, under appropriate assumptions, a first integral of an ordinary differential equation becomes a discrete Lyapunov function for its discretization. They also prove that if a periodic orbit of a planar conservative system is convex, hence the dynamics of the Euler discretization is monotone near that orbit. Higher-dimensional cases are also treated. Moreover, they show that increasing the order of a numerical Runge-Kutta method does not lead in general to improve its efficiency for numerical simulations. To show this, the authors study the discretization of linear equations having a quadratic first integral. The results are also applied to mechanical systems with central forces and to symmetric mechanical systems.