Discontinous Galerkin methods for Vlasov models of plasma

Abstract

The Vlasov-Poisson equations describe the evolution of a collisionless plasma, represented through a probability density function (PDF) that self-interacts via an electrostatic force. One of the main difficulties in numerically solving this system is the severe time-step restriction that arises from parts of the PDF associated with moderate-to-large velocities. The dominant approach in the plasma physics community for removing these time-step restrictions is the so-called particle-in-cell (PIC) method, which discretizes the distribution function into a set of macro-particles, while the electric field is represented on a mesh. Several alternatives to this approach exist, including fully Lagrangian, fully Eulerian, and so-called semi-Lagrangian methods. The focus of this work is the semi-Lagrangian approach, which begins with a grid-based Eulerian representation of both the PDF and the electric field, then evolves the PDF via Lagrangian dynamics, and finally projects this evolved field back onto the original Eulerian mesh.

We present a semi-Lagrangian and a hybrid semi-Lagrangian method for solving the Vlasov Poisson equations, based on high-order discontinuous Galerkin (DG) spatial representations of the solution. The Poisson equation is solved via a high-order local discontinuous Galerkin (LDG) scheme. The resulting methods are high-order accurate, which is demonstrably important for this problem in order to retain the rich phase-space structure of the solution; mass conservative; and provably positivity-preserving. We argue that our approach is a promising method that can produce very accurate results at relatively low computational expense. We demonstrate this through several examples for the (1+1)D case, using both the hybrid as well as the full semi-Lagrangian method. In particular, the methods are validated on several numerical test cases, including the two-stream instability problem, Landau damping, and the formation of a plasma sheath. In addition, we propose a (2+2)D method that promises to be a productive avenue of future research. The (2+2)D method incorporates local time-stepping methods on unstructured grids in physical space and semi-Lagrangian time stepping on Cartesian grids in velocity space. This method is again high-order, mass conservative, and provably positivity-preserving.