High-order Gauss-Legendre methods admit a composition representation and a conjugate-symplectic counterpart
Felice Iavernaro, Francesca Mazzia, Ernst Hairer
TL;DR
The paper addresses whether higher-order Gauss–Legendre Runge–Kutta methods admit a composition representation using multiderivative Euler-type steps and identifies a conjugate-symplectic partner. It first analyzes the two-stage, fourth-order Gauss method via discrete approximations of $D_1 f$ and $D_2 f$, producing two 2-stage RK methods whose half-step composition yields the Gauss method of order four and whose reverse composition yields a conjugate-symplectic partner with the same stability profile for the Dahlquist test equation. With a collocation-based construction at a degree-2, the MDMP4/MDTR4 variants reproduce the Gauss method and its conjugate-symplectic twin, and the framework generalizes to arbitrary order via a structured composition of derived RK methods. The results preserve geometric structure, require only a single nonlinear solve per step, and provide a fourth-order dense output, suggesting a practical pathway to high-order, structure-preserving integrators applicable to Hamiltonian systems.
Abstract
One of the most classical pairs of symplectic and conjugate-symplectic schemes is given by the Midpoint method (the Gauss-Runge-Kutta method of order 2) and the Trapezoidal rule. These can be interpreted as compositions of the Implicit and Explicit Euler methods, taken in direct and reverse order, respectively. This naturally raises the question of whether a similar composition structure exists for higher-order Gauss-Legendre methods. In this paper, we provide a positive answer by first examining the fourth-order case and then outlining a generalization to higher orders.
