This library contains a cross-platform Eigen wrapper for many different external linear solvers including (but not limited to):
- CHOLMOD
- Hypre
- AMGCL
- Pardiso
const std::string solver_name = "Hypre"
auto solver = Solver::create(solver_name, "");
// Configuration parameters like iteration or accuracy for iterative solvers
// solver->set_parameters(params);
// System sparse matrix
Eigen::SparseMatrix<double> A;
// Right-hand side
Eigen::VectorXd b;
// Solution
Eigen::VectorXd x(b.size());
solver->analyze_pattern(A, A.rows());
solver->factorize(A);
solver->solve(b, x);You can use Solver::available_solvers() to obtain the list of available solvers.
Polysolve uses a json file to provide parameters to the individual solvers. The following template can be used as a starting points, and a more detailed explanation of the parameters is below.
{
"Eigen::LeastSquaresConjugateGradient": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Eigen::DGMRES": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Eigen::ConjugateGradient": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Eigen::BiCGSTAB": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Eigen::GMRES": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Eigen::MINRES": {
"max_iter": 1000,
"tolerance": 1e-6
},
"Pardiso": {
"mtype": -1
},
"Hypre": {
"max_iter": 1000,
"pre_max_iter": 1000,
"tolerance": 1e-6
},
"AMGCL": {
"precond": {
"relax": {
"degree": 16,
"type": "chebyshev",
"power_iters": 100,
"higher": 2,
"lower": 0.008333333333,
"scale": true
},
"class": "amg",
"max_levels": 6,
"direct_coarse": false,
"ncycle": 2,
"coarsening": {
"type": "smoothed_aggregation",
"estimate_spectral_radius": true,
"relax": 1,
"aggr": {
"eps_strong": 0
}
}
},
"solver": {
"tol": 1e-10,
"maxiter": 1000,
"type": "cg"
}
}
}max_itercontrols the solver's iterations, default1000conv_tol,tolerancecontrols the convergence tolerance, default1e-10
pre_max_iter, number of pre iterations, default1
The default parameters of the AMGCL solver are:
{
"precond": {
"relax": {
"degree": 16,
"type": "chebyshev",
"power_iters": 100,
"higher": 2,
"lower": 0.008333333333,
"scale": true
},
"class": "amg",
"max_levels": 6,
"direct_coarse": false,
"ncycle": 2,
"coarsening": {
"type": "smoothed_aggregation",
"estimate_spectral_radius": true,
"relax": 1,
"aggr": {
"eps_strong": 0
}
}
},
"solver": {
"tol": 1e-10,
"maxiter": 1000,
"type": "cg"
}
}For a more details and options refer to the AMGCL documentation.
mtype, sets the matrix type, default 11
| mtype | Description |
|---|---|
| 1 | real and structurally symmetric |
| 2 | real and symmetric positive definite |
| -2 | real and symmetric indefinite |
| 3 | complex and structurally symmetric |
| 4 | complex and Hermitian positive definite |
| -4 | complex and Hermitian indefinite |
| 6 | complex and symmetric |
| 11 | real and nonsymmetric |
| 13 | complex and nonsymmetric |
This error is cause by having a more recent version of SuiteSparse (≥ v7.0.0) installed on your system than the version we download and build. We use @sergiud's fork of SuiteSparse which includes CMake support. However, the fork is not up to date with the latest version of SuiteSparse (currently v5.12.0 while the official release is at version v7.0.1). Version v7.0.0 changed the SuiteSparse_config.h header and no longer includes the necessary struct definitions.
For now, if you can, please downgrade (< v7.0.0) or uninstall your system version of SuiteSparse. In the meantime, we will work with the SuiteSparse developers to resolve this issue.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent