Authors: Jean-Luc Vay (Lawrence Berkeley National Laboratory), Leroy Anthony Drummond (Lawrence Berkeley National Laboratory), Alice Koniges (Lawrence Berkeley National Laboratory), Brendan Godfrey (University of Maryland and Lawrence Berkeley National Laboratory), Irving Haber (University of Maryland)
Abstract: Numerical simulations have been critical in the recent rapid developments of advanced accelerator concepts. Among the various available numerical techniques, the Particle-In-Cell (PIC) approach is the method of choice for self-consistent simulations from first principles. While spectral methods have been popular in the early PIC codes, the finite-difference time-domain method has become dominant. Recently, a novel parallelization strategy was proposed that takes advantage of the local nature of Maxwell equations that has the potential to combine spectral accuracy with finite-difference favorable parallel scaling. Due to its compute-intensive nature combined with adjustable accuracy and locality, the new solver promises to be especially well suited for emerging exascale systems. The new solver was recently extended to enable user-programmability of the spatial and temporal order of accuracy at runtime, enabling a level of scalability and flexibility that is unprecedented for such codes.