﻿

# ─ 2.2. Solver parameters

The recommended values for the solver configuration are explained in this section.

This example may be solved with the default parameters. The most important parameters within the Solver → Parameters panel are listed below:

• Method: Subdomains is the default solver for periodical structures. Green’s Function is another experimental approach to solve the problem calculating the integral of these functions.
• Architecture strategy: both MPI and OpenMP provide similar results whenever the machine supports them, as OpenMP is not compatible with distributed memory systems. However OpenMP is more efficient in the cases when it can be applied.
• Solver: Indicates the approach used to solve the linear system of equations. BICGSTAB (BiConjugate Gradient STAbilized method) and GMRES (Generalized Minimal Residual method) are available. If no convergence is achieved by using any of these methods try to use the other one. The Max. number of unknowns for the Direct Solver option is the threshold of the maximum allowed unknowns to compute the currents of the Method of Moments by using a direct solution method, instead of using the iterative process. In order to use always the iterative solver, this parameter can be set equal to zero. Note that the direct solution method may require huge memory and time resources when a large number of unknowns is considered.
• Preconditioner: The user can enable the preconditioner to speed up the resolution of the problem with the Enable Precondicioner option. The user can choose between two different preconditioners:
• Diagonal preconditioner: The diagonal preconditioner is fast to compute and requires a reduced amount of memory, although the improvement in the convergence rate it produces is normally moderate. This preconditioner it is only recommended when more than 8 divisions per wavelenth is set in the meshing process, as a lower number of divisions can reduce the convergence rate compared to not using any preconditioner.
• Sparse Approximate Inverse (SAI): This preconditioner will generally the results faster than the diagonal preconditioner. The Sparsity Distance, expressed in wavelengths (0.5 as the default value) indicates how accurately this preconditioner will resemble the inverse of the rigorous MoM matrix. Higher values will normally involve a faster convergence, but the memory required to store the preconditioner data will grow fast, non-linearly. We advice to keep the default value or increase it slightly in case of specially ill-conditioned systems.

Due to its numerical nature, the SAI preconditioner is better suited for the case of shared memory parallelization (OpenMP), while the conventional diagonal preconditioner can be used either for OpenMP or for the MPI paradigm.

• Other parameters: The Relative error may be reduced to ensure a higher accuracy in the resolution of the problem. At most an error of 0.01 or even smaller is specially recommended when the SAI preconditioner is selected.

Solver parameters

We use cookies on this website to improve your navigation experience on this site. By using this site, you agree to our cookie policy.

I agree