计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2014年
13期
234-238
,共5页
吴建平%赵军%宋君强%张卫民%马怀发
吳建平%趙軍%宋君彊%張衛民%馬懷髮
오건평%조군%송군강%장위민%마부발
混凝土细观数值模拟%稀疏线性方程组%并行计算%区域分解%预条件
混凝土細觀數值模擬%稀疏線性方程組%併行計算%區域分解%預條件
혼응토세관수치모의%희소선성방정조%병행계산%구역분해%예조건
meso-scale simulation of concrete%sparse linear system%parallel computing%domain decomposition%preconditioner
细观数值模拟是混凝土性能研究的一种重要手段,但稀疏线性方程组求解在总体模拟时间中所占比重很大。由于属于三维问题,且规模很大,所以采用预条件Krylov子空间迭代是必由之路。Aztec是国际上专门设计用于求解稀疏线性方程组的软件包之一,由于目前混凝土细观数值模拟中的稀疏线性方程组对称正定,所以利用Aztec中提供的CG迭代法进行求解,并对多种能保持对称性的预条件选项进行了实验比较。结果表明,在基于区域分解的并行不完全Cholesky分解、无重叠对称化GS迭代、最小二乘等预条件技术中,第一种的效率最高,且在重叠度为0,填充层次为0时,效果最好;实验结果还表明,在本应用问题中,用RCM排序一般导致求解时间更长,从而没有必要采用。
細觀數值模擬是混凝土性能研究的一種重要手段,但稀疏線性方程組求解在總體模擬時間中所佔比重很大。由于屬于三維問題,且規模很大,所以採用預條件Krylov子空間迭代是必由之路。Aztec是國際上專門設計用于求解稀疏線性方程組的軟件包之一,由于目前混凝土細觀數值模擬中的稀疏線性方程組對稱正定,所以利用Aztec中提供的CG迭代法進行求解,併對多種能保持對稱性的預條件選項進行瞭實驗比較。結果錶明,在基于區域分解的併行不完全Cholesky分解、無重疊對稱化GS迭代、最小二乘等預條件技術中,第一種的效率最高,且在重疊度為0,填充層次為0時,效果最好;實驗結果還錶明,在本應用問題中,用RCM排序一般導緻求解時間更長,從而沒有必要採用。
세관수치모의시혼응토성능연구적일충중요수단,단희소선성방정조구해재총체모의시간중소점비중흔대。유우속우삼유문제,차규모흔대,소이채용예조건Krylov자공간질대시필유지로。Aztec시국제상전문설계용우구해희소선성방정조적연건포지일,유우목전혼응토세관수치모의중적희소선성방정조대칭정정,소이이용Aztec중제공적CG질대법진행구해,병대다충능보지대칭성적예조건선항진행료실험비교。결과표명,재기우구역분해적병행불완전Cholesky분해、무중첩대칭화GS질대、최소이승등예조건기술중,제일충적효솔최고,차재중첩도위0,전충층차위0시,효과최호;실험결과환표명,재본응용문제중,용RCM배서일반도치구해시간경장,종이몰유필요채용。
Meso-scale simulation is an important way for performance study of concrete. But the solution of sparse linear systems occupies most of the total simulation time. Due to its three-dimensional origination and large scale, preconditioned Krylov subspace iterations are the best choices. Aztec is a software package developed in the community and designed specially to solve sparse linear systems. For the sparse linear systems in the meso-simulation of concrete are symmetric positive definite currently, the CG iteration provided in Aztec, is selected to solve the linear systems. Several preconditioning options which can preserve the symmetric characteristic are tested with experiments. The results show that among the par-allel incomplete Cholesky factorization based on domain decomposition, symmetrical Gauss-seidel iteration without over-lapping, least square preconditioners, the first is the most efficient and when the degree of overlapping and the level of fill are both selected as 0, the preconditioner is the best. At the same time, the results show that the time elapsed with RCM reordering is lager in general. Therefore, it should not be exploited in the simulation.