Study on parallel processing of fast multipole boundary element method
| Project/Area Number |
11650481
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
構造工学・地震工学
|
|---|
| Research Institution | Fukui University |
|---|
Principal Investigator |
FUKUI Takuo Fukui University, Department of Archtecture and Civil Engineering, Professor, 工学部, 教授 (30026299)
|
|---|
| Project Period (FY) |
1999 – 2001
|
| Project Status |
Completed (Fiscal Year 2001)
|
| Budget Amount *help |
¥3,200,000 (Direct Cost: ¥3,200,000)
Fiscal Year 2001: ¥500,000 (Direct Cost: ¥500,000)
Fiscal Year 2000: ¥1,400,000 (Direct Cost: ¥1,400,000)
Fiscal Year 1999: ¥1,300,000 (Direct Cost: ¥1,300,000)
|
|---|
| Keywords | Computational mechanics / Boundary element method / Fast multipole method / Huge size simulation / Preconditioning / Parallel processing / Data sampling |
|---|
| Research Abstract |
The purpose of this project is to extend the appicability of the boundary element method (BEM) to analize huge size mechanical simulation problems. The fast multipole method (FMM) is used to accelerate the boundary element analysis and to reduce the required memory (fast multipole boundary element method : FMBEM).
The turget problems concerned here are the simulation of construction of under ground cavern (3D elastostatic problem), and the simulation of the earthquake motion in the plane of Fukui at Fukui earthquake (3D elastodynamic problem). The following studies were performed in the project :
(a) Improvement of FMBEM algorithm : FMM needs a long multipole expansion in the wave problem with a large wavenumber, thus the translation computation needs much time. To avoid this difficulty the diagonal form is introcuded in FMBEM algorithm. This is effective to analyze a large region compared with the wave length.
(b) Preprocessing method in FMBEM : An applicable preprocessing method by Haar wavelet transfor mation is proposed, and a FMM like algorithm is constructed to apply this to FMBEM. This method is effective both in Dirichlet and Neumann boundary conditions even if BEM is based on Burton ＆ Miller integral equation.
c Parallel processing of FMBEM : A Beowulf by PC cluster is composed. FMBEM code in C is paral lelized on the system by MPI. The solvable problem size is propotional to the total memory size of PC cluster.
(d) Problems of half-space and inhomogeneous region, and aiftomatic boundary data generation : Some topics in practical applications are studied. FMM for a half-space and an inhomogeneous region is proposed. A boundary data generation method by the stochastic sampling method is studied.
|
Report
(4 results)
Research Products
(35 results)
