On the positivity of Poisson integrators for the Lotka-Volterra equations (Q2350717): Difference between revisions

The authors study numerical methods for solving the Lotka-Volterra system. This system is not itself Hamiltonian but it has a structure similar to a Hamiltonian system. Due to this fact the equations can be rewritten as a non-canonical Hamiltonian system or, more generally, as a Poisson system. The numerical simulations show the same qualitative behavior as the exact solution, namely, the approximate solutions are cyclic and positive. Let us note that Hamiltonian systems are symplectic, which in the two-dimensional case means area-preserving, and that suitable methods to get good qualitative behavior are the so-called symplectic methods. The simplest symplectic method is the symplectic Euler method, and an explicit variant of it. The authors define a Poisson integrator and prove that both the symplectic Euler method and its explicit variant are Poisson integrators for the Lotka-Volterra system. Moreover, these methods have an interesting property that for suitable values of the step-size the numerical result stays in the first quadrant. For the symplectic Euler method the authors prove a very simple condition that yields the desired result. A bound of the local error of the numerical result as well as a bound of the Hamiltonian error are given. As a consequence the authors state the main result of the paper. They say that for given initial conditions, they can compute for which step-size values the numerical solution given by the explicit variant of the symplectic Euler method is ensured to remain positive over exponentially long time intervals.