[Trilinos-Users] Composing preconditioner?

[Federico Marini]

 Hi Trilinos Users,

I am an Italian PhD student working in Austin, Texas. I need to implement a Two-level Additive Schwarz Preconditioner:

P = R0 * Ac^-1 * R0^t + \sum R_i * A_i^-1 * R_i^t.

where Ri restricts to a subdomain and R0 to a coarse space. A_i is the subdomain matrix, Ac is the coarse one.

I use AztecOO and its Domain Decomposition Preconditioner for the first level (\sum R_i * A_i^-1 * R_i^t.), and it works fine. I would like to use an Epetra_Operator for the second level (R0 * Ac^-1 * R0^t). I explicitly build the R0 operator and Ac.

Is there a way to combine (in an additive way) the AztecOO 1-level Domain Decomposition preconditioner with a user-defined Epetra_Operator, so that in every iteration of CG or GMRES, the preconditioner is given by P as above?

Otherwise, can I do that using the ML package, providing my own Projection/Restriction R0/R0^t operators for the second level? I checked the ml_2level_DD.cpp example and ML documentation, but I can't figure out how to implement some geometric multigrid techniques (and not algebraic).

Thanks to anyone who wants to reply!
Federico

P.S. It's the first time I use a mail list, therefore don't hesitate to redirect me to a more appropriate mail address.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20140304/aa1795e3/attachment.html