An iterative semi-implicit scheme with robust damping

TL;DR

This paper introduces a second-order accurate semi-implicit iterative scheme for stiff wave systems that offers robust damping, error control, and significant computational speed-ups over explicit methods, applicable to complex plasma simulations.

Contribution

The paper develops a convergent iterative semi-implicit scheme with robust damping capabilities, enabling efficient and accurate simulation of stiff wave systems with error monitoring.

Findings

Achieves CPU speed-up factors of 20 to several hundreds over explicit methods.

Maintains second-order accuracy even with non-commuting operators.

Demonstrates effectiveness on kinetic Alfvén wave magnetic reconnection simulations.

Abstract

An efficient, iterative semi-implicit (SI) numerical method for the time integration of stiff wave systems is presented. Physics-based assumptions are used to derive a convergent iterative formulation of the SI scheme which enables the monitoring and control of the error introduced by the SI operator. This iteration essentially turns a semi-implicit method into a fully implicit method. Accuracy, rather than stability, determines the timestep. The scheme is second-order accurate and shown to be equivalent to a simple preconditioning method. We show how the diffusion operators can be handled so as to yield the property of robust damping, i.e., dissipating the solution at all values of the parameter $\mathcal D\dt$, where $\mathcal D$ is a diffusion operator and $\dt$ the timestep. The overall scheme remains second-order accurate even if the advection and diffusion operators do not commute. In the limit of no physical dissipation, and for a linear test wave problem, the method is shown to be symplectic. The method is tested on the problem of Kinetic Alfv\'en wave mediated magnetic reconnection. A Fourier (pseudo-spectral) representation is used. A 2-field gyrofluid model is used and an efficacious k-space SI operator for this problem is demonstrated. CPU speed-up factors over a CFL-limited explicit algorithm ranging from $\sim20$ to several hundreds are obtained, while accurately capturing the results of an explicit integration. Possible extension of these results to a real-space (grid) discretization is discussed.