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" c == routine arguments == c == local variables == integer nn c == end of interface == c-- Headline. print* print*,' ==================================================' print*,' Large Scale Optimization with off-line capability.' print*,' ==================================================' print* print*,' Version 2.1.0' print* c-- Get the number of control variables. call OPTIM_NUMBMOD( nn ) cph( print *, 'pathei: vor optim_sub' cph) c-- Call the subroutine. call OPTIM_SUB( nn ) c-- Succesful termination. print* print*,' ======================================' print*,' Large Scale Optimization run finished.' print*,' ======================================' print* 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