linesrch.f90 File Reference

Line search. More...

Go to the source code of this file.

Modules
module	linesrch
	Line search data.

Functions/Subroutines
subroutine	ptline (n, x, f, g, s, step, info)
	Perform linesearch. More...
subroutine	ptldef (gtole, stmax, minfe, maxfe)
	Initialize line search. More...
subroutine	ptlopt (nf, m, slopes, steps)
	Get details. More...
subroutine	ptlprt (lunp)
	Print line search data. More...

Variables
integer(mpi), parameter	linesrch::msfd =20
integer(mpi)	linesrch::nsfd
	number of function calls More...
integer(mpi)	linesrch::idgl
	index of smallest negative slope More...
integer(mpi)	linesrch::idgr
	index of smallest positive slope More...
integer(mpi)	linesrch::idgm
	index of minimal slope More...
integer(mpi)	linesrch::minf =1
	min. More...
integer(mpi)	linesrch::maxf =5
	max. More...
integer(mpi)	linesrch::lsinfo
	(status) information More...
real(mpd), dimension(4, msfd)	linesrch::sfd
	abscissa; function value; slope; predicted zero More...
real(mpd)	linesrch::stmx =0.9
	maximum slope ratio More...
real(mpd)	linesrch::gtol
	slope ratio More...

Detailed Description

Line search.

Author

Volker Blobel, University Hamburg, 2005-2009 (initial Fortran77 version)

Claus Kleinwort, DESY (maintenance and developement)

Copyright

Copyright (c) 2009 - 2015 Deutsches Elektronen-Synchroton, Member of the Helmholtz Association, (DESY), HAMBURG, GERMANY

This library is free software; you can redistribute it and/or modify it under the terms of the GNU Library General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public License for more details.

You should have received a copy of the GNU Library General Public License along with this program (see the file COPYING.LIB for more details); if not, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

Line search routine with sufficient decrease of slope.

In many minimization problems the objective function is close to quadratic, except far from the solution. Close to the minimum the behaviour may be almost quadratic or, due to round-off errors, it may have a non-smooth behaviour, which often complicates any further progress and the recognition of convergence. Round-off errors affect the function value, which may be large and small parameter changes result in small relative changes of the function value. Close to the minimum the gradient becomes small and the behaviour is not so much affected by Round-off errors.

     CALL PTLDEF(0.0,0.0, 0,0) ! init line search
     N=...
     X(.)=...
     D(.)=...
     ALPHA=1.0D0
10   F(X)=...
     G(X)=...
     IF(.) S(X)=..
     CALL PTLINE(N,X,F,G,D,ALPHA,INFO)
     IF(INFO.LT.0) GOTO 10

Definition in file linesrch.f90.

Function/Subroutine Documentation

ptldef()

subroutine ptldef	(	real(mps), intent(in)	gtole,
		real(mps), intent(in)	stmax,
		integer(mpi), intent(in)	minfe,
		integer(mpi), intent(in)	maxfe
	)

Initialize line search.

Parameters

[in]	gtole	slope ratio
[in]	stmax	total step limit
[in]	minfe	minimum number of evaluations
[in]	maxfe	maximum number of evaluations --- range ---- default slope ratio 1.0E-4 ... 0.9 0.9 min. F-calls 1 ... 2 1 max. F-calls 2 ... 10 5

Definition at line 232 of file linesrch.f90.

References linesrch::gtol, linesrch::maxf, linesrch::minf, linesrch::msfd, linesrch::nsfd, and linesrch::stmx.

Referenced by xloopn().

ptline()

subroutine ptline	(	integer(mpi), intent(in)	n,
		real(mpd), dimension(n), intent(inout)	x,
		real(mpd), intent(inout)	f,
		real(mpd), dimension(n), intent(inout)	g,
		real(mpd), dimension(n), intent(inout)	s,
		real(mpd), intent(out)	step,
		integer(mpi), intent(out)	info
	)

Perform linesearch.

Parameters

[in]	N	dimension of problem
[in,out]	X	current iterate
[in,out]	F	associated function value
[in,out]	G	associated gradient
[in,out]	S	search vector
[out]	STEP	step factor (initially = 1.0)
[out]	INFO	information = -1 repeat function evaluation = 0 input error (e.g. gradient not negative) = 1 convergence reached = 2 convergence assumed, but round-off errors = 3 too many function calls = 4 step factor ALPHA to small (ALPHA <= TOL)

Definition at line 89 of file linesrch.f90.

Referenced by xloopn().

ptlopt()

subroutine ptlopt	(	integer(mpi), intent(out)	nf,
		integer(mpi), intent(out)	m,
		real(mps), dimension(3), intent(out)	slopes,
		real(mps), dimension(3), intent(out)	steps
	)

Get details.

Parameters

[out]	NF	number of function values
[out]	M	index of function value with smallest slope
[out]	SLOPES	initial, current, smallest slope
[out]	STEPS	initial position, current, smallest step

Definition at line 258 of file linesrch.f90.

References linesrch::nsfd, and linesrch::sfd.

Referenced by ploopb(), and ploopc().

ptlprt()

subroutine ptlprt

(

integer(mpi), intent(in)

lunp

)

Print line search data.

Parameters

[in]

lunp

unit number

Definition at line 294 of file linesrch.f90.

References linesrch::gtol, linesrch::idgl, linesrch::idgr, linesrch::lsinfo, linesrch::maxf, linesrch::minf, linesrch::nsfd, linesrch::sfd, and linesrch::stmx.

Referenced by xloopn().