Aleph: Highly Scalable Unstructured PIC-DSMC Low Temperature Plasma Code

Aleph is an unstructured 3D Particle-In-Cell (PIC) low temperature kinetic plasma code developed to model plasma discharges in low pressure systems. Self-consistent plasma evolution within electrostatic fields is treated using the Galerkin Finite-Element-Method (FEM) with particle collisions treated using the Direct Simulation Monte Carlo (DSMC) method. Modeling the extreme range of length and time scales is made possible by unstructured meshes together with a two-decomposition treatment of the mesh. A static uniform decomposition of the mesh is used for the electrostatic field solves, and a dynamically evolving decomposition of the same mesh is employed for evolving the particles. Rebalancing the particle mesh allows problems involving hundreds of millions of elements and billions of particles to scale well up to thousands of processors on high-performance computers (HPCs). The data structures to enable unstructured PIC are built on the Trilinos solver package which in turn provides direct access to several solver/preconditioner algorithms, robust load balancing and input-output services supporting simulation restarts. Aleph's object-oriented C++ design leverages these data structures via Evaluations to provide extensible functionality supporting various approaches to unstructured PIC and a variety of post-processing and multi-physics extensions to the baseline PIC-DSMC capability. We present design details of the Aleph code together with representative results emphasizing algorithm correctness, performance, and scalability of various 3D unstructured PIC problems on a range of HPCs available at Sandia National Laboratories.

Aleph is an unstructured 3D Particle-In-Cell (PIC) low temperature kinetic plasma code developed to model plasma discharges in low pressure systems. Self-consistent plasma evolution within electrostatic fields is treated using the Galerkin Finite-Element-Method (FEM) with particle collisions treated using the Direct Simulation Monte Carlo (DSMC) method. Modeling the extreme range of length and time scales is made possible by unstructured meshes together with a two-decomposition treatment of the mesh. A static uniform decomposition of the mesh is used for the electrostatic field solves, and a dynamically evolving decomposition of the same mesh is employed for evolving the particles. Rebalancing the particle mesh allows problems involving hundreds of millions of elements and billions of particles to scale well up to thousands of processors on high-performance computers (HPCs). The data structures to enable unstructured PIC are built on the Trilinos solver package which in turn provides direct access to several solver/preconditioner algorithms, robust load balancing and input-output services supporting simulation restarts. Aleph's object-oriented C++ design leverages these data structures via Evaluations to provide extensible functionality supporting various approaches to unstructured PIC and a variety of post-processing and multi-physics extensions to the baseline PIC-DSMC capability. We present design details of the Aleph code together with representative results emphasizing algorithm correctness, performance, and scalability of various 3D unstructured PIC problems on a range of HPCs available at Sandia National Laboratories.