lbfgs_jpl_version/optim.2/optim_main.F


c     ==================================================================
c
c     prgopti.F: Routines for doing an off-line optimization after the
c                ECCO forward and adjoint model have been run.
c
c     main     - Driver routine.
c     opti     - Mid-level routine to do the spin up and spin down.
c     optimum  - Routine that calls the minimization.
c
c     Documentation:
c
c     The collection of these routines originated mainly from Ralf
c     Giering. Patrick Heimbach improved and corrected considerable
c     parts of the original code. Christian Eckert contributed the
c     interface to the ECCO release of the MITgcmUV in order to get
c     the offline version going.
c
c     How to use the off-line optimization.
c
c       Doing an off-line optimization means that one alternately
c       calls the adjoint model and the optimization routines.
c
c       The adjoint model yields at iteration i the cost function
c       value and the gradient of the cost function with respect to
c       the control variables. The optimization routines then use
c       this information to reduce the cost function and give a
c       new estimate of the control variables which can then be used
c       in the next cycle to yield a new cost function and the
c       corresponding gradient.
c
c     started:  Ralf Giering      (lsoptv1)
c
c               Patrick Heimbach  heimbach@mit.edu  28-Feb-2000
c
c               - Corrected and restructured the original lsoptv1
c                 code.
c
c               Christian Eckert  eckert@mit.edu    15-Feb-2000
c
c               - Off-line capability and some cosmetic changes
c                 of the optimization wrapper.
c
c     changed:
c
c     ==================================================================


      program optim_main

c     ==================================================================
c     PROGRAM optim_main
c     ==================================================================
c
c     o Driver routine for the ECCO optimization package.
c
c     started: Christian Eckert eckert@mit.edu 15-Feb-2000
c
c     changed: Christian Eckert eckert@mit.edu 10-Mar-2000
c
c              - Added ECCO layout.
c
c     ==================================================================
c     SUBROUTINE
c     ==================================================================

      implicit none

c     == global variables ==

#include "blas1.h"
#include "mpif.h"
      integer nx
      integer ierr

c     == routine arguments ==

c     == local variables ==

      integer nn

      integer myid,nprocs,mystart,myend
      common /mpi_parm/ myid,nprocs,mystart,myend
c     == end of interface ==

      call MPI_INIT(ierr)
      call MPI_COMM_RANK(MPI_COMM_WORLD,myid,ierr)
      call MPI_COMM_SIZE(MPI_COMM_WORLD,nprocs,ierr)

      if (myid .eq. 0) then
c--   Headline.
      print*
      print*,' =================================================='
      print*,' Large Scale Optimization with off-line capability.'
      print*,' =================================================='
      print*
      print*,'                Version 2.1.0'
      print*
      endif

c--   Get the number of control variables.
      call optim_numbmod( nn )

cph(
      if (myid .eq. 0)
     & print *, 'pathei: vor optim_sub'
cph)
c--   Call the subroutine.
      call optim_sub( nn )

      if (myid .eq. 0) then 
c--   Succesful termination.
      print*
      print*,' ======================================'
      print*,' Large Scale Optimization run finished.'
      print*,' ======================================'
      print*
      endif

      call MPI_FINALIZE(ierr)

      end

CStartOfInterface
      INTEGER FUNCTION IFNBLNK( string )
C     /==========================================================\
C     | FUNCTION IFNBLNK                                         |
C     | o Find first non-blank in character string.              |
C     \==========================================================/
      IMPLICIT NONE
C
      CHARACTER*(*) string
CEndOfInterface
C
      INTEGER L, LS
C
      LS     = LEN(string)
      IFNBLNK = 0
      DO 10 L = 1, LS
       IF ( string(L:L) .EQ. ' ' ) GOTO 10
        IFNBLNK = L
        GOTO 11
   10 CONTINUE
   11 CONTINUE
C
      RETURN
      END

CStartOfInterface
      INTEGER FUNCTION ILNBLNK( string )
C     /==========================================================\
C     | FUNCTION ILNBLNK                                         |
C     | o Find last non-blank in character string.               |
C     \==========================================================/
      IMPLICIT NONE
      CHARACTER*(*) string
CEndOfInterface
      INTEGER L, LS
C
      LS      = LEN(string)
      ILNBLNK = LS
      DO 10 L = LS, 1, -1
        IF ( string(L:L) .EQ. ' ' ) GOTO 10
         ILNBLNK = L
         GOTO 11
   10 CONTINUE
   11 CONTINUE
C
      RETURN
      END

1	heimbach	1.1
2			c ==================================================================
3			c
4			c prgopti.F: Routines for doing an off-line optimization after the
5			c ECCO forward and adjoint model have been run.
6			c
7			c main - Driver routine.
8			c opti - Mid-level routine to do the spin up and spin down.
9			c optimum - Routine that calls the minimization.
10			c
11			c Documentation:
12			c
13			c The collection of these routines originated mainly from Ralf
14			c Giering. Patrick Heimbach improved and corrected considerable
15			c parts of the original code. Christian Eckert contributed the
16			c interface to the ECCO release of the MITgcmUV in order to get
17			c the offline version going.
18			c
19			c How to use the off-line optimization.
20			c
21			c Doing an off-line optimization means that one alternately
22			c calls the adjoint model and the optimization routines.
23			c
24			c The adjoint model yields at iteration i the cost function
25			c value and the gradient of the cost function with respect to
26			c the control variables. The optimization routines then use
27			c this information to reduce the cost function and give a
28			c new estimate of the control variables which can then be used
29			c in the next cycle to yield a new cost function and the
30			c corresponding gradient.
31			c
32			c started: Ralf Giering (lsoptv1)
33			c
34			c Patrick Heimbach heimbach@mit.edu 28-Feb-2000
35			c
36			c - Corrected and restructured the original lsoptv1
37			c code.
38			c
39			c Christian Eckert eckert@mit.edu 15-Feb-2000
40			c
41			c - Off-line capability and some cosmetic changes
42			c of the optimization wrapper.
43			c
44			c changed:
45			c
46			c ==================================================================
47
48
49			program optim_main
50
51			c ==================================================================
52			c PROGRAM optim_main
53			c ==================================================================
54			c
55			c o Driver routine for the ECCO optimization package.
56			c
57			c started: Christian Eckert eckert@mit.edu 15-Feb-2000
58			c
59			c changed: Christian Eckert eckert@mit.edu 10-Mar-2000
60			c
61			c - Added ECCO layout.
62			c
63			c ==================================================================
64			c SUBROUTINE
65			c ==================================================================
66
67			implicit none
68
69			c == global variables ==
70
71			#include "blas1.h"
72			#include "mpif.h"
73			integer nx
74			integer ierr
75
76			c == routine arguments ==
77
78			c == local variables ==
79
80			integer nn
81
82			integer myid,nprocs,mystart,myend
83			common /mpi_parm/ myid,nprocs,mystart,myend
84			c == end of interface ==
85
86			call MPI_INIT(ierr)
87			call MPI_COMM_RANK(MPI_COMM_WORLD,myid,ierr)
88			call MPI_COMM_SIZE(MPI_COMM_WORLD,nprocs,ierr)
89
90			if (myid .eq. 0) then
91			c-- Headline.
92			print*
93			print*,' =================================================='
94			print*,' Large Scale Optimization with off-line capability.'
95			print*,' =================================================='
96			print*
97			print*,' Version 2.1.0'
98			print*
99			endif
100
101			c-- Get the number of control variables.
102			call optim_numbmod( nn )
103
104			cph(
105			if (myid .eq. 0)
106			& print *, 'pathei: vor optim_sub'
107			cph)
108			c-- Call the subroutine.
109			call optim_sub( nn )
110
111			if (myid .eq. 0) then
112			c-- Succesful termination.
113			print*
114			print*,' ======================================'
115			print*,' Large Scale Optimization run finished.'
116			print*,' ======================================'
117			print*
118			endif
119
120			call MPI_FINALIZE(ierr)
121
122			end
123
124			CStartOfInterface
125			INTEGER FUNCTION IFNBLNK( string )
126			C /==========================================================\
127			C \| FUNCTION IFNBLNK \|
128			C \| o Find first non-blank in character string. \|
129			C \==========================================================/
130			IMPLICIT NONE
131			C
132			CHARACTER() string
133			CEndOfInterface
134			C
135			INTEGER L, LS
136			C
137			LS = LEN(string)
138			IFNBLNK = 0
139			DO 10 L = 1, LS
140			IF ( string(L:L) .EQ. ' ' ) GOTO 10
141			IFNBLNK = L
142			GOTO 11
143			10 CONTINUE
144			11 CONTINUE
145			C
146			RETURN
147			END
148
149			CStartOfInterface
150			INTEGER FUNCTION ILNBLNK( string )
151			C /==========================================================\
152			C \| FUNCTION ILNBLNK \|
153			C \| o Find last non-blank in character string. \|
154			C \==========================================================/
155			IMPLICIT NONE
156			CHARACTER() string
157			CEndOfInterface
158			INTEGER L, LS
159			C
160			LS = LEN(string)
161			ILNBLNK = LS
162			DO 10 L = LS, 1, -1
163			IF ( string(L:L) .EQ. ' ' ) GOTO 10
164			ILNBLNK = L
165			GOTO 11
166			10 CONTINUE
167			11 CONTINUE
168			C
169			RETURN
170			END
171