When I want to solve a large-scale sparse problem with the multigrid Solver, how do I set up its components and parameters? The current setup needs to be iterated too many times

[cuihaoran21]

```

using ir = gko::solver::Ir;
using bj = gko::preconditioner::Jacobi<double, int>;
using pgm = gko::multigrid::Pgm<double, int>;
using mg = gko::solver::Multigrid;

/* Create smoother factory*/
smoother_gen = gko::share(
ir::build()
.with_solver(bj::build().with_max_block_size(4u))
.with_relaxation_factor(static_cast(0.8))
.with_criteria(gko::stop::Iteration::build().with_max_iters(10u))
.on(exec));

/* Create RestrictProlong factory*/
mg_level_gen = gko::share(pgm::build().with_deterministic(true).on(exec));

/* Create CoarsestSolver factory */
coarsest_solver_gen = gko::share(
ir::build()
    .with_solver(bj::build().with_max_block_size(4u))
    .with_relaxation_factor(static_cast<double>(0.8))
    .with_criteria(gko::stop::Iteration::build().with_max_iters(1u))
    .on(exec));

/* Create MultigridSolver factory*/
multigrid_gen = mg::build()
                    .with_max_levels(10u) //Limit the maximum number of layers for a multigrid
                    .with_min_coarse_rows(4u) //Specifies the minimum number of rows for the coarsest layer
                    .with_pre_smoother(smoother_gen)//Set the pre smoother
                    .with_post_uses_pre(true)//Specify that the post smoother is the same as the pre smoother
                    .with_mg_level(mg_level_gen)//Specifies RestrictProlong
                    .with_coarsest_solver(coarsest_solver_gen)//Sets the coarsest layer solver
                    .with_criteria(iter_stop, tol_stop)//Set stopping criteria including the number of iterations and tolerance
                    .on(exec);

[yhmtsai]

Hi @cuihaoran21 , good to hear you would like to try multigrid solver.
It might be worth setting the multigrid is a preconditioner of CG solver if the system is symmetric.

[rbourgeois33]

Hi,

If you want to use the multigrid solver as a stand-alone method:
https://github.com/ginkgo-project/ginkgo/blob/develop/examples/mixed-multigrid-solver/mixed-multigrid-solver.cpp
I have not tried it myself, I wonder how much the mixed precision helps.

If you want to use it as a preconditionner to the GC method (this is a much more common practice):
https://github.com/ginkgo-project/ginkgo/tree/develop/examples/multigrid-preconditioned-solver-customized

[MarcelKoch]

Hi @cuihaoran21, I've converted your original issue into a discussion. Please feel free to update this if you have any further comments.