Open the Exascale Tool Chest for Predictive Plasma Modelling with PIConGPU

PIConGPU is reportedly the world's fastest, electro-magnetic 3D3V particle-in-cell code. With sustained multi-PFlop/s performance and demonstrated PByte-scale I/O performance 2 this open-source PIC code is able to fully exploit the computational power provided by leadership-scale manycore HPC systems. At the same time, modern single-source C++ meta-programming enables performance-portability without the need of rewriting or maintaining code for various programming models.

Originally developed for the domain of laser-plasma acceleration, the PIConGPU project addresses today's needs for reproducible, repeatable, large-scale parameter studies and fundamentally reshapes simulation workflows towards Exascale computing. Deploying in situ analysis for observables, we bridge the ever-growing gap between computational power and post-processing (IO) capabilities. For the first time, the dramatic increase in computational power allows self-consistent coupling of 3D PIC kinetics with non-LTE plasma physics for collisional-radiative modelling towards pump-probe experiments at the European XFEL.

We present the architecture, open environment, open standards, community and physical models of our open-source software PIConGPU 4 and its applications in laser-plasma physics, XFEL-matter interaction and computational astrophysics.