On quadratic invariants and symplectic structure (Q1338532)

It is known [see e.g. \textit{J. M. Sanz-Serna}, BIT 28, No. 4, 877-883 (1988; Zbl 0655.70013)] that symplectic Runge-Kutta (RK) methods, i.e. RK methods which applied to Hamiltonian systems provide a discrete canonical flow, have the property that they preserve all quadratic invariants of the original Hamiltonian system. In this paper the authors analyze more deeply the relation between the preservation of quadratic invariants by a numerical method and its symplecticity. More specifically they prove that a numerical method is symplectic if it satisfies the following assumptions: 1) It preserves the quadratic invariants generated by quadratic invariants of Hamiltonian systems. 2) It is closed under restriction to closed subsystems, i.e. the result of the method when the method is applied to the full system and then evaluated on the variables of the subsystem is the same as the result of the method on the closed subsystem. 3) The method is closed under differentiation, which means that if \(y_ 1 = F(y_ 0)\) is the numerical flow of a Hamiltonian equation, \(z_ 1 = dF(y_ 0) z_ 0\) is the flow associated to its variational equation. Finally, it is shown that assumptions 2) and 3) are always satisfied by all RK, linear multistep and even more general linear methods, therefore assumption 1) characterizes symplecticity for these types of methods.