Investigation of an Algebraic Multigrid Method with IMEX

Geoffrey L. Brown
Supervisors: Dingjun Chen (CMG), Dave Collins(CMG)


Algebraic Multi-grid (AMG) as an optional pre-conditioner of iterative solution speedup of linear systems has some advantages  for elliptic and parabolic problems. It has a faster reduction of low frequency errors and has no requirement of knowledge of the model geometry. Here, the Fraunhofer’s SAMG software package coupled to IMEX. The relevant technical issues include using FGMRES rather than GMRES acceleration and AMG sensitive to the decoupling technique are presented.

 Finally, the initial SAMG and parallel SAMG tests are performed using recent version of IMEX with SAMG. The results show that speedup of several data sets can be achieved with AMG; AMG has difficulties with some data sets such as less reduction in outer iterations and then leading to many AMG cycles, which possibly can be attributed to the non-parabolic nature of PDE’s.