High Performance Linear Solver Library Using Information Obtained in Previous Solution Steps
Project/Area Number |
26540052
|
Research Category |
Grant-in-Aid for Challenging Exploratory Research
|
Allocation Type | Multi-year Fund |
Research Field |
High performance computing
|
Research Institution | Hokkaido University |
Principal Investigator |
|
Co-Investigator(Renkei-kenkyūsha) |
MIFUNE TAKESHI 京都大学, 大学院工学研究科, 助教 (20362460)
|
Project Period (FY) |
2014-04-01 – 2016-03-31
|
Project Status |
Completed (Fiscal Year 2015)
|
Budget Amount *help |
¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
Fiscal Year 2015: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2014: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
|
Keywords | 高性能計算 / 線形反復法 / 誤差修正法 / 自動生成 / ハイパフォーマンスコンピューティング / 数値解析 / 電磁場解析 / 有限要素解析 / ハイパフォーマンス・コンピューティング / 情報システム / 機械学習 |
Outline of Final Research Achievements |
In this research, we focused on a series of linear systems having an identical or similar coefficient matrix. We examined the improvement of convergence of the linear solver by means of error correction methods. In the research, we proposed an automatic construction method for a mapping operator for the error correction method based on the information obtained in the previous solution steps. Because the proposed method constructs an appropriate mapping operator with minimum memory requirement, it is easily applied to practical simulations. The proposed method was examined in the context of time-dependent electromagnetic field problems. Numerical tests of 2-d and 3-d eddy-current finite element analyses with T-Omega formulation confirmed the effectiveness of the method.
|
Report
(3 results)
Research Products
(6 results)