KSPSetNormType#

Sets the norm that is used for convergence testing.

Synopsis#

#include "petscksp.h"
PetscErrorCode KSPSetNormType(KSP ksp, KSPNormType normtype)


Logically Collective

Input Parameters#

• ksp - Krylov solver context

• normtype - one of

   KSP_NORM_NONE - skips computing the norm, this should generally only be used if you are using
the Krylov method as a smoother with a fixed small number of iterations.
Implicitly sets KSPConvergedSkip() as KSP convergence test.
Note that certain algorithms such as KSPGMRES ALWAYS require the norm calculation,
for these methods the norms are still computed, they are just not used in
the convergence test.
KSP_NORM_PRECONDITIONED - the default for left preconditioned solves, uses the l2 norm
of the preconditioned residual P^{-1}(b - A x)
KSP_NORM_UNPRECONDITIONED - uses the l2 norm of the true b - Ax residual.
KSP_NORM_NATURAL - supported  by KSPCG, KSPCR, KSPCGNE, KSPCGS


Options Database Key#

• -ksp_norm_type <none,preconditioned,unpreconditioned,natural> - set KSP norm type

Note#

Not all combinations of preconditioner side (see KSPSetPCSide()) and norm type are supported by all Krylov methods. If only one is set, PETSc tries to automatically change the other to find a compatible pair. If no such combination is supported, PETSc will generate an error.

Developer Note#

Supported combinations of norm and preconditioner side are set using KSPSetSupportedNorm().

KSPSetUp(), KSPSolve(), KSPDestroy(), KSPConvergedSkip(), KSPSetCheckNormIteration(), KSPSetPCSide(), KSPGetPCSide(), KSPNormType