9.2.3 FGMRES
This iterative linear solver, from the MKL library, implements the GMRES algorithm, which is an efficient Krylov-based iterative solution strategy. It requires several configuration parameters, listed below, and only works with unsymmetric matrices. In the table below, N refers to the number of equations.
parameter | description | default |
max_iter | The maximum number of iterations | min(N,150) |
tol | relative residual tolerance | 1e-6 |
abs_tol | absolute residual tolerance | 1e-12 |
print_level | The amount of information printed (0,1, or 2) | 0 |
fail_max_iters | Indicates whether reaching max iters is a failure or not (1) | 1 (true) |
pc_left | The left preconditioner | (none) |
pc_right | The right preconditioner | (none) |
Comments:
- When using FGMRES as a preconditioner it is recommended to set the fail_max_iters flag to zero.