Multi-revolution composition methods for highly oscillatory differential equations (Q740813)

The paper deals with the approximation of the \(N\)th iterates of a near-identity smooth map by composition methods. The considered smooth map has the form \(\varphi _{\varepsilon }(y)=y+\varepsilon \Theta _{e}(y),\) where the iteration \(\varphi _{\varepsilon }^{N}\) is seen as one step with step-size \(\varepsilon \) of the first-order integrator of the ordinary differential equation \[ dz(t)/dt=\Theta _{0}(z(t)),\quad \Theta _{0}(t)=d\varphi _{\varepsilon }(z)/d\varepsilon |_{\text{ }y=0}. \] \noindent Thus \(\varphi _{\varepsilon }^{N}(y)\) may be interpreted as an approximation at \(t=N\varepsilon \) of the solution \(z(t)\) with the initial condition \(z(0)=y\). \ Within this context, the authors introduce a new class of multi-revolution composition methods (MRCM) which are applied to time integration of highly oscillating problems with a single harmonic frequency \( \omega =2\pi /\varepsilon\). In particular, MRCM share the properties of standard composition methods, since, for instance, they are intrinsically geometric and well suited for Hamiltonian or divergence-free equations. Remarkably, error estimates are proved with error constants that are independent of the oscillatory frequency. Then, the paper reports numerical experiments on a Fermi-Ulam-Pasta-like problem and on a cubic nonlinear Schrödinger equation. These numerical results show the effectiveness and robustness of the proposed methods.