petsc-3.11.4 2019-09-28
Report Typos and Errors


Sets the objective and (gradient'*stepdirection) evaluation routine for the line search. Sometimes it is more efficient to compute the inner product of the gradient and the step direction than it is to compute the gradient, and this is all the line search typically needs of the gradient.


#include "petsctaolinesearch.h" 
PetscErrorCode TaoLineSearchSetObjectiveAndGTSRoutine(TaoLineSearch ls, PetscErrorCode(*func)(TaoLineSearch ls, Vec x, Vec s, PetscReal *, PetscReal *, void*), void *ctx)
Logically Collective on TaoLineSearch

Input Parameter

ls - the TaoLineSearch context
func - the objective and gradient evaluation routine
ctx - the (optional) user-defined context for private data

Calling sequence of func

     func (TaoLinesearch ls, Vec x, PetscReal *f, PetscReal *gts, void *ctx);

x - input vector
s - step direction
f - function value
gts - inner product of gradient and step direction vectors
ctx (optional) user- defined context

Note: The gradient will still need to be computed at the end of the line search, so you will still need to set a line search gradient evaluation routine

Note: Bounded line searches (those used in bounded optimization algorithms) don't use g's directly, but rather (g'x - g'x0)/steplength. You can get the x0 and steplength with TaoLineSearchGetStartingVector() and TaoLineSearchGetStepLength()


Some algorithms (lcl, gpcg) set their own objective routine for the line search, application programmers should be wary of overriding the default objective routine.

See Also

TaoLineSearchCreate(), TaoLineSearchSetObjective(), TaoLineSearchSetGradient(), TaoLineSearchUseTaoRoutines()




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