Self-adaptive Event-Driven Simulations of Multi-Scale Plasma Systems

Abstract

Multi-scale systems pose a formidable computational challenge. Explicit time stepping suffers from the global CFL restriction. Efficient application of adaptive mesh refinement (AMR) to systems with irregular dynamics (e.g. turbulence, reactive systems, particle acceleration etc.) may be problematic. To address these issues, we developed an alternative approach (Karimabadi et al. 2005; Omelchenko & Karimabadi 2006a,b) to time-stepped integration of physics-based systems: Discrete-Event Simulation (DES). We combine finite difference and particle-in-cell techniques with this new methodology by assuming two caveats: (1) a local time increment, ∆f for a discrete quantity f can be expressed in terms of a physically meaningful increment, ∆f; (2) f is considered to be modified only when its change exceeds ∆f. Event-driven asynchronous time advance makes use of local causality rules. This technique enables flux conserving integration of the solution, removes the curse of global CFL condition, and eliminates unnecessary computation in inactive regions. It can be naturally combined with various mesh refinement techniques. DES results in robust and fast simulation codes, which can be efficiently parallelized when implemented via a Preemptive Event Processing (PEP) technique (Omelchenko & Karimabadi 2006c). We discuss this novel technology in the context of diffusion reaction and computational fluid dynamics (CFD) applications, as well as general model-model (flux) coupling.