Porting the PUMImbbl library to GPUs and integration in the hPIC2 Particle-in-Cell code*

GPU-capable PUMImbbl library, which provides multi-block boundary layer (MBBL) mesh capability under the Parallel Unstructured Mesh Infrastructure (PUMI), is integrated with the hPIC2 code. PUMImbbl library employs an implicit representation of the multi-block mesh and exploits tensor product mesh structure in multiple dimensions. In addition, PUMImbbl library supports complex geometry such as tile gaps by making certain blocks inactive in the mesh as desired by the application. The implicit representation of such a mesh is based on parameters that describe every block in each direction (including blocks with geometrically graded/boundary layer elements). The PUMImbbl allows us to incorporate a GPU-capable non-uniform mesh into the hPIC2 Particle-in-Cell code. hPIC2 is a full-orbit, hybrid Particle-in-Cell (PIC) code targeting Plasma-Surface Interaction problems. hPIC2 is developed using the C++ library of  Kokkos performance portability framework for hybrid CPU/GPU parallelism. Thanks to the new mesh, hPIC2 can simulate 2D plasma problems on non-uniform meshes using GPUs, and solve problems on large plasma domains including the steep gradients normally encountered in plasma sheaths, with a relatively small number of cells, effectively reducing the total number of particles needed for the simulation and the overall computational cost.