The Raptor code for radiation hydrodynamics is based on the BoxLib framework for structured adaptive mesh refinement (AMR). Each refinement level of a mesh is decomposed into logically rectangular grids, with one or more of these grids assigned to each active processor. In time-dependent calculations coarse levels are advanced at larger timesteps than fine levels.
Flux-limited diffusion and discrete ordinates transport models for radiation in-
volve different types of equations, but they face a number of the same issues when implemented in an AMR framework. The most notable of these are implicit coupling to the fluid energy, AMR timestepping, and energy-conserving relationships between refinement levels. Other issues are unique to one model or the other: diffusion involves
a parabolic equation - for which we use the hypre multigrid package - while the hyper-bolic discrete ordinate equations introduce the problems of efficient parallel sweeping and convergence acceleration. Among these topics I will give particular attention to effiicent parallel implementation of discrete ordinates, and will present parallel scaling results for both diffusion and transport on 2D and 3D adaptive meshes.
Copyright. All Rights Reserved.