RBG-Maxwell: A GPU-accelerated software package for simulation of collisional plasmas with open boundaries

We present RBG-Maxwell ( R elativistic B oltzmann equation on G PUs coupled with Maxwell’s equations), an open-source GPU-accelerated package for simulating relativistic collisional plasmas with self-consistent electromagnetic fields. The software integrates a grid-based relativistic Boltzmann solver with a Jefimenko-equation electromagnetic solver, providing an integral-field formulation for open electromagnetic-field treatment without explicitly engineered absorbing layers. Implemented in Python with CuPy, Numba, and Ray, RBG-Maxwell supports domain-decomposed multi-GPU execution, configurable collision processes, prescribed external forces, and standard plasma diagnostics. We describe the software architecture, numerical workflow, and representative examples including particle diffusion with self-consistent fields and finite-time collision-induced relaxation. Strong- and weak-scaling tests are also presented to quantify the performance of the GPU implementation. RBG-Maxwell is intended for kinetic plasma problems where smooth distribution-function evolution, self-consistent fields, and open-boundary electromagnetic treatment are important.