Beyond Electrodynamics with PIConGPU: Performance Portable, Open Multi-Physics HPC Simulations for Laser-Plasma Experiments at the European XFEL

PIConGPU is reportedly the fastest electro-magnetic particle-in-cell code in the world in terms of sustained Flop/s. Its computational power does not only enable 3D3V simulations with unprecedented detail and fast time-to-solution but also allows improving predictive capabilities of simulations by estimating stochastic and systematic errors via repeated simulations. Synthetic in-situ diagnostics drive the exploration of high-detail simulations whose signatures, e.g. emitted radiation spectra, would be inaccessible in post-processing due to sheer data size. Upcoming experiments at the European XFEL require us to take PIConGPU even one step further: modeling XFEL-matter interaction on top of laser-driven particle acceleration processes requires the introduction of non-trivial X-ray photon scattering, photon generation and advanced non-LTE atomic models. Coupled with an open-science centered strategy, from open performance portable source code over open standardized data formats to documented workflows for PByte scale simulation we strive towards a new quality of predictive, reproducible simulations within our community.