Software design of SHARP

SHARP (Simulation-based High-efficiency Advanced Reactor Prototyping) is a modern suite of codes to simulate the key components of a fast reactor core. The SHARP toolkit is organized as a collection of modules, each representing the key components of the physics to be modeled - neutron transport, thermal hydraulics, fuel/structure behavior - together with pre and post-processing for geometry definition, mesh generation, visualization, user interface, etc. The physics models are designed to make minimal possible use of lumped parameter models, homogenization, and empirical correlations in favor of more direct solution of the fundamental governing equations, when sufficient computing resources are available. Thus, one of the key design goals is to effectively leverage leadership class computing resources - viz. BG/P and Cray Supercomputers that are on the current trajectory to delivering sustained petaflops performance. Further, the nature of the physical problem to be investigated will require either strong or weak coupling between some or all of the existing modules (e.g. operator split vs. fully coupled), while multiple implementations of each physics module, representing different algorithms, will also be required (e.g. deterministic versus Monte Carlo) for verification and to explore different physical regimes. Accomplishing these goals in the context of ultra-scalable architectures and multidisciplinary and possibly distributed development teams is a daunting task. In this paper we explain our inital lighweight and loosely coupled framework, its initial design, and a number of current open research questions in this area.