大型稀疏矩阵计算的现代方法介绍

--  

    前面有2篇博文：矩阵迭代， 稀疏矩阵相关。初学的认识。最近又来咀嚼回头草了，仍然参考两本书：

    Iterative Methods for linear sparse systems 

    Lectcutres on computational numerical analysis of PDE. J.M.McDonough, Kentucky.

    第一本是hpc课，老师主推荐；第二本的作者，写了几个lectures都非常好，e.g. <cfd>, <turbulenence>

    

－  经典迭代方法

*    Jacobi, Gauss-Seidel, SOR serials :

     - to solve Au=b,  decompose A = D - L - U

D: dialogue part of A, L : lower part of A; U : upper part of A

        u^(n+1) = G u^n + k

     - for convergence test, contrast fixed-point iteration form, and through contraction mapping principle( 压缩映射定理). requiring spectral of system less than 1 .

        namely  rho(G) < 1     

  Jacobi  G_ja =  D;  

  G-S      G_gs =  D - L;

  SOR     G_sor =  (D- wL)/w,  w is relaxation paramter

  SSOR  G_ssor = (D - wL)D^(-1)(D - wF)/ (w(2-w))  

   

 *    ADI

      decompose A = H + V + S

H: x-direction term; V: y/z-direction term; S: 0-order term