Efficient algorithms for discrete-ordinates transport iterations in massively parallel computers /
We have demonstrated the effectiveness of several transport synthetic acceleration (TSA) methods: β-TSA, stretched TSA, and stretched and filtered (SF) TSA. For SF TSA, we have shown that the filtering parameter affects the performance and the optimum value is dependent on the scattering ratio. Ou...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2004.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=790270111&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | We have demonstrated the effectiveness of several transport synthetic acceleration (TSA) methods: β-TSA, stretched TSA, and stretched and filtered (SF) TSA. For SF TSA, we have shown that the filtering parameter affects the performance and the optimum value is dependent on the scattering ratio. Our analyses reveal that simple S[N] low-order acceleration of S[N] high-order transport problems (β-TSA with β = 0 or stretched TSA with [epsilon] = 1) are divergent for problems with modest or strong heterogeneities. Although SF TSA with the correct filtering parameter converges for problems with heterogeneities, the performance of SF TSA degrades for large heterogeneities. We have demonstrated how to "wrap" Generalized Minimal Residual (GMRES) iterative method around transport solution methods, such as source iteration and various TSA methods. Our analyses reveal that SF TSA with the proper filtering parameter and GMRES is an effective iterative method to solve difficult transport problems. We have demonstrated how easy it is to incorporate GMRES and TSA-type preconditioners into a sweep-based parallel transport code by implementing GMRES, β-TSA, stretched TSA, and SF TSA into the Texas A&M University parallel transport code. We present numerical results of simple test problems on the Texas A&M University parallel transport code. We also show the performance of several different spatial domain partitioning strategies on parallel efficiencies and how parallel speedup is affected by our iterative methods. |
|---|---|
| Item Description: | Vita. "Major Subject: Nuclear Engineering". |
| Physical Description: | ix, 145 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 139-144). |