skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Fast algorithms for transport models. Final report

Technical Report ·
DOI:https://doi.org/10.2172/10190744· OSTI ID:10190744

This project has developed a multigrid in space algorithm for the solution of the S{sub N} equations with isotropic scattering in slab geometry. The algorithm was developed for the Modified Linear Discontinuous (MLD) discretization in space which is accurate in the thick diffusion limit. It uses a red/black two-cell {mu}-line relaxation. This relaxation solves for all angles on two adjacent spatial cells simultaneously. It takes advantage of the rank-one property of the coupling between angles and can perform this inversion in O(N) operations. A version of the multigrid in space algorithm was programmed on the Thinking Machines Inc. CM-200 located at LANL. It was discovered that on the CM-200 a block Jacobi type iteration was more efficient than the block red/black iteration. Given sufficient processors all two-cell block inversions can be carried out simultaneously with a small number of parallel steps. The bottleneck is the need for sums of N values, where N is the number of discrete angles, each from a different processor. These are carried out by machine intrinsic functions and are well optimized. The overall algorithm has computational complexity O(log(M)), where M is the number of spatial cells. The algorithm is very efficient and represents the state-of-the-art for isotropic problems in slab geometry. For anisotropic scattering in slab geometry, a multilevel in angle algorithm was developed. A parallel version of the multilevel in angle algorithm has also been developed. Upon first glance, the shifted transport sweep has limited parallelism. Once the right-hand-side has been computed, the sweep is completely parallel in angle, becoming N uncoupled initial value ODE`s. The author has developed a cyclic reduction algorithm that renders it parallel with complexity O(log(M)). The multilevel in angle algorithm visits log(N) levels, where shifted transport sweeps are performed. The overall complexity is O(log(N)log(M)).

Research Organization:
Colorado Univ., Boulder, CO (United States). Applied Mathematics Program
Sponsoring Organization:
USDOE, Washington, DC (United States)
DOE Contract Number:
FG02-90ER25086
OSTI ID:
10190744
Report Number(s):
DOE/ER/25086-T2; ON: DE95001702; BR: KC0701010/KC0701030; TRN: 94:022675
Resource Relation:
Other Information: PBD: [1994]
Country of Publication:
United States
Language:
English