19. mai 2015 kl. 22:09 skrev Tobias Ritschel <[email protected]>:
> Hi again > > I want to solve the fully implicit residual equations using Newtons method > with an iterative method for solving the linear system. I am using the > AutoDiffBlock objects and thus the matrix and right-hand-side is in the V and > M formats, or Array and SparseMatrix in Eigen, respectively. > > As far as I can see, the only iterative solver for non-symmetric matrices in > Eigen is BiCGSTAB and it is a bit hard to work with. Besides, I would like > some freedom of choice with regard to algorithm. > > I can see that dune-istl has several solvers and eWoms as well (perhaps the > same?). Can you give some recommendation or advice as to what functions to > use? In addition, it would be preferable if the solver can run in parallel. Hi Tobias, The solvers of dune-istl are able to run in parallel, so I would recommend that you try those. Which solver and preconditioner to choose would depend on the problem, but we do have an implementation using a CPR-type two-stage preconditioning for fully implicit systems in the classes NewtonIterationBlackoilCPR (overall solution) and CPRPreconditioner (preconditioner part) that are used by the 'flow' fully implicit black-oil simulator. Those are implemented with dune-istl, and have some choice of options/solvers, so you may find them useful to look at. There is a class (DuneMatrix) that allows you to construct a matrix that can be used with the Dune solvers from an Eigen matrix, it could also be useful to you. Good luck with your project! Atgeirr > > Kind regards > Tobias Ritschel > _______________________________________________ > Opm mailing list > [email protected] > http://www.opm-project.org/mailman/listinfo/opm _______________________________________________ Opm mailing list [email protected] http://www.opm-project.org/mailman/listinfo/opm
