| This thesis is based on the development and testing of the PENTRAN (Parallel Environment Neutral particle TRANsport) discrete ordinates code, designed for distributed memory and distributed computing parallel environments. The code can iteratively solve complex problems in nuclear design with complete, automatic decomposition of the angular, energy, and spatial variables. Written in ANSI FORTRAN-77 using the new standard Message Passing Interface library, PENTRAN is 28,500 lines and solves multigroup, anisotropic, three-dimensional discrete ordinates neutron and photon transport problems. Using a Block-Jacobi or Red-Black algorithm with automatic decomposition onto a distributed memory array of n processors, the code also has options to automatically load balance subdomain assignments. Memory in PENTRAN is partitioned so that each processor is only required to store the subdomain needed for local computation of the angular, energy, and spatial discretization, making the code scalable in memory by using more processors with increased decomposition.;Further, as a direct result of this research, new numerical approaches that enhance the flexibility and accuracy of 3-D transport calculations have been developed and incorporated into PENTRAN. These include an adaptive differencing strategy that contains a new hybrid discrete ordinates differencing scheme, Exponential-Directional Weighted (EDW) differencing, a new discontinuous grid treatment with Taylor Projected Mesh Coupling (TPMC), and a simplified TPMC-multigrid acceleration. Tests show that PENTRAN has a 97% parallel code fraction, is scalable, and has been successfully benchmarked in parallel against conventional, single CPU production codes with appreciable speedups. With the completion of parallel code development, testing, and benchmarking, PENTRAN is now ready to be used to solve actual 3-D problems that are not practical (or tractable) to solve on single CPU computers. |
