#include "slepcpep.h" PetscErrorCode PEPSetScale(PEP pep,PEPScale scale,PetscReal alpha,Vec Dl,Vec Dr,PetscInt its,PetscReal lambda)Collective
pep | - the eigensolver context | |
scale | - scaling strategy | |
alpha | - the scaling factor used in the scalar strategy | |
Dl | - the left diagonal matrix of the diagonal scaling algorithm | |
Dr | - the right diagonal matrix of the diagonal scaling algorithm | |
its | - number of iterations of the diagonal scaling algorithm | |
lambda | - approximation to wanted eigenvalues (modulus) |
-pep_scale <type> | - scaling type, one of <none,scalar,diagonal,both> | |
-pep_scale_factor <alpha> | - the scaling factor | |
-pep_scale_its <its> | - number of iterations | |
-pep_scale_lambda <lambda> | - approximation to eigenvalues |
In the scalar strategy, scaling is applied to the eigenvalue, that is, mu = lambda/alpha is the new eigenvalue and all matrices are scaled accordingly. After solving the scaled problem, the original lambda is recovered. Parameter 'alpha' must be positive. Use PETSC_DETERMINE to let the solver compute a reasonable scaling factor, and PETSC_CURRENT to retain a previously set value.
In the diagonal strategy, the solver works implicitly with matrix Dl*A*Dr, where Dl and Dr are appropriate diagonal matrices. This improves the accuracy of the computed results in some cases. The user may provide the Dr and Dl matrices represented as Vec objects storing diagonal elements. If not provided, these matrices are computed internally. This option requires that the polynomial coefficient matrices are of MATAIJ type. The parameter 'its' is the number of iterations performed by the method. Parameter 'lambda' must be positive. Use PETSC_DETERMINE or set lambda = 1.0 if no information about eigenvalues is available. PETSC_CURRENT can also be used to leave its and lambda unchanged.