Massively parallel transport sweeps on meshes with cyclic dependencies

Abstract When solving the first-order form of linear Boltzmann equation, a common misconception is that matrix-free computational method “sweeping mesh”, used in conjunction with Discrete Ordinates method, too complex or does not scale well enough to be implemented modern high performance computing codes. This has led considerable efforts development matrix-based methods are computationally expensive and partly driven by requirements placed on spatial discretizations. In particular, transport codes required support higher order elements, concept invariably adds lot complexity sweeps because introduction cyclic dependencies curved mesh cells. this article we will present comprehensive implementation sweeping, piecewise-linear DFEM discretization particular focus handling possible extensions These new C++ simulation framework called Chi-Tech ( ? ? T e c h ). We some typical results aspects one can expect during real world simulations, also scaling study >100k processes where maintains greater than 80% efficiency total 87.7 trillion angular flux unknowns for 116 group simulation.