KSPRichardsonSetSelfScale#

Sets Richardson to automatically determine optimal scaling at each iteration to minimize the 2-norm of the preconditioned residual

Synopsis#

#include "petscksp.h" 
PetscErrorCode KSPRichardsonSetSelfScale(KSP ksp, PetscBool scale)

Logically Collective

Input Parameters#

ksp - the iterative context
scale - PETSC_TRUE or the default of PETSC_FALSE

Options Database Key#

-ksp_richardson_self_scale - Use self-scaling

Notes#

Requires two extra work vectors. Uses an extra VecAXPY() and VecDotNorm2() per iteration.

Changing this flag runs KSPSetUp() again, since the number of work vectors depends on it. The self-scaled variant has no block analog, so turning it on makes KSPMatSolve() solve the right-hand sides one at a time instead of iterating on the whole block at once, and releases the work space that block iteration cached

Developer Note#

Could also minimize the 2-norm of the true residual with one less work vector

Level#

intermediate

Location#

src/ksp/ksp/impls/rich/richscale.c

Implementations#

KSPRichardsonSetSelfScale_Richardson() in src/ksp/ksp/impls/rich/rich.c

Index of all KSP routines
Table of Contents for all manual pages
Index of all manual pages