*DECK MAIN C**** File MI00MAIN FORTRAN. C C MAIN FUNOBJ FUNCON MATMOD C C ***************************************************************** C * * C * M I N O S * C * * C * A Modular In-core Nonlinear Optimization System * C * * C * Version 5.1 June 11, 1987 * C * * C * * C * Bruce A. Murtagh, University of New South Wales * C * Michael A. Saunders, Stanford University * C * * C *---------------------------------------------------------------* C * * C * MINOS is distributed by OTL, Stanford University. Enquiries * C * should be directed to the following address: * C * * C * Office of Technology Licensing * C * Software Distribution Center * C * 350 Cambridge Avenue, Suite 250 * C * Palo Alto, California 94306, U.S.A. * C * * C * Telephone: (415) 723-0651 Telex: 348402 STANFRD STNU * C * * C *---------------------------------------------------------------* C * * C * Copyright 1987 Stanford University. * C * * C * This material may be reproduced by or for the U.S. Government * C * pursuant to the copyright license under DAR CLAUSE 7-104.9(A) * C * (1979 MAR). * C * * C * This material is based upon work partially supported by the * C * National Science Foundation under grants MCS-7926009 and * C * ECS-8012974; the Department of Energy contract AM03-76SF00326,* C * PA NO. DE-AT03-76ER72018; and the Army Research Office * C * contract DAAG29-81-K-0156. * C * * C ***************************************************************** C C This is the default main program for MINOS. C It provides all of the necessary workspace. C PARAMETER(NWCORE=30000) DOUBLE PRECISION Z(NWCORE) COMMON /ZCOMM / Z C DATA NWCORE/41000/ C CALL MINOS1( Z,NWCORE ) STOP C C End of MAIN END *DECK FUNOBJ SUBROUTINE FUNOBJ( MODE, N, X, F, G, NSTATE, NPROB, Z, NWCORE ) IMPLICIT REAL*8(A-H,O-Z) DOUBLE PRECISION X(N), G(N), Z(NWCORE) C C ------------------------------------------------------------------ C This is the default version of FUNOBJ for MINOS. C It belongs to the nonlinear test problem MANNE, C which will specify PROBLEM NUMBER 1111 C in order to identify itself. C C For test purposes, we look at DERIVATIVE LEVEL C and sometimes pretend that we dont know the first C three elements of the gradient. C ------------------------------------------------------------------ C COMMON /M1FILE/ IREAD,IPRINT,ISUMM COMMON /M8DIFF/ DIFINT(2),GDUMMY,LDERIV,LVLDIF,KNOWNG(2) COMMON /MANNE / B,AT(100),BT(100) LOGICAL GKNOWN DATA ZERO /0.0D+0/ C IF (NPROB .NE. 1111) GO TO 900 GKNOWN = LDERIV .EQ. 1 .OR. LDERIV .EQ. 3 NT = N/2 F = ZERO C DO 50 J = 1, NT XCON = X(NT+J) F = F + BT(J)*DLOG(XCON) G(J) = ZERO IF (GKNOWN .OR. J .GT. 3) G(NT+J) = BT(J)/XCON 50 CONTINUE RETURN C C C It looks like some other FUNOBJ is needed. C 900 WRITE(IPRINT, 9000) IF (ISUMM .GT. 0) WRITE(ISUMM, 9000) MODE = -1 RETURN C 9000 FORMAT(/ ' XXX Subroutine FUNOBJ has not been loaded.') C End of FUNOBJ for MANNE END *DECK FUNCON SUBROUTINE FUNCON( MODE, M, N, NJAC, X, F, G, * NSTATE, NPROB, Z, NWCORE ) IMPLICIT REAL*8(A-H,O-Z) DOUBLE PRECISION X(N), F(M), G(NJAC), Z(NWCORE) C C ------------------------------------------------------------------ C This is the default version of FUNCON for MINOS. C It belongs to the nonlinear test problem MANNE, C which will specify PROBLEM NUMBER 1111 C in order to identify itself. C C For test purposes, we look at DERIVATIVE LEVEL C and sometimes pretend that we dont know the first C three elements of the gradient. C ------------------------------------------------------------------ C COMMON /M1FILE/ IREAD,IPRINT,ISUMM COMMON /M8DIFF/ DIFINT(2),GDUMMY,LDERIV,LVLDIF,KNOWNG(2) COMMON /MANNE / B,AT(100),BT(100) LOGICAL GKNOWN C IF (NPROB .NE. 1111) GO TO 900 GKNOWN = LDERIV .GE. 2 NT = N IF (NSTATE .NE. 1) GO TO 100 C C ------------ C First entry. C ------------ ONE = 1.0 GROW = 0.03 BETA = 0.95 XK0 = 3.0 XC0 = 0.95 XI0 = 0.05 B = 0.25 WRITE(IPRINT, 1000) B C A = (XC0 + XI0) / XK0**B GFAC = (ONE + GROW)**(ONE - B) AT(1) = A*GFAC BT(1) = BETA DO 10 J = 2, NT AT(J) = AT(J-1)*GFAC BT(J) = BT(J-1)*BETA 10 CONTINUE BT(NT) = BT(NT)/(ONE - BETA) C C ------------- C Normal entry. C ------------- 100 DO 150 J = 1, NT XKAP = X(J) FJ = AT(J) * XKAP**B F(J) = FJ IF (GKNOWN .OR. J .GT. 3) G(J) = B*FJ / XKAP 150 CONTINUE IF (NSTATE .LT. 2) RETURN C C ------------ C Final entry. C ------------ WRITE(IPRINT, 2000) (F(J), J = 1, NT) RETURN C C C It looks like some other FUNCON is needed. C 900 WRITE(IPRINT, 9000) IF (ISUMM .GT. 0) WRITE(ISUMM, 9000) MODE = -1 RETURN C 1000 FORMAT(// ' This is problem MANNE. B =', F8.3) 2000 FORMAT(// ' Final nonlinear function values' / (5F12.5)) 9000 FORMAT( / ' XXX Subroutine FUNCON has not been loaded.') C End of FUNCON for MANNE END *DECK MATMOD SUBROUTINE MATMOD( NCYCLE, NPROB, FINISH, * M, N, NB, NE, NKA, NS, NSCL, * A, HA, KA, BL, BU, * ASCALE, HS, ID1, ID2, * X, PI, Z, NWCORE ) C IMPLICIT REAL*8(A-H,O-Z) INTEGER*2 HA(NE), HS(NB) INTEGER KA(NKA), ID1(NB), ID2(NB) DOUBLE PRECISION A(NE), ASCALE(NSCL), BL(NB), BU(NB) DOUBLE PRECISION X(NB), PI(M), Z(NWCORE) LOGICAL FINISH C C ------------------------------------------------------------------ C This is the default version of MATMOD for MINOS. C It belongs to the nonlinear test problem MANNE, C which will specify PROBLEM NUMBER 1111 C in order to identity itself. C C MATMOD is called at the end of each CYCLE except the last. C ------------------------------------------------------------------ C COMMON /M1FILE/ IREAD,IPRINT,ISUMM COMMON /M5LOBJ/ SINF,WTOBJ,MINIMZ,NINF,IOBJ,JOBJ,KOBJ COMMON /CYCLCM/ CNVTOL,JNEW,MATERR,MAXCY,NEPHNT,NPHANT,NPRINT C IF (NPROB .NE. 1111) GO TO 900 BU(N) = 0.1 WRITE(IPRINT, 1000) BU(N) IF (ISUMM .GT. 0) WRITE(ISUMM, 1000) BU(N) RETURN C C C It looks like some other MATMOD is needed. C 900 WRITE(IPRINT, 9000) IF (ISUMM .GT. 0) WRITE(ISUMM, 9000) FINISH = .TRUE. RETURN C 1000 FORMAT(/ ' MATMOD changes last upper bound to', F5.2) 9000 FORMAT(/ ' XXX Subroutine MATMOD has not been loaded.') C C End of MATMOD END C**** File MI10MACH FORTRAN. C C MIFILE MINOS1 MINOS2 MINOS3 C M1CHAR M1HASH M1INIT M1READ C C C Contents of this and other MINOS source files C ============================================= C C C 0. File MI00MAIN Main program and user subroutines: C ------------- C C MAIN FUNOBJ FUNCON MATMOD C C C 1. File MI10MACH Machine-dependent routines: C ------------- C C MIFILE MINOS1 MINOS2 MINOS3 C M1CHAR M1HASH M1INIT M1READ C C C 2. File MI15BLAS Basic Linear Algebra Subprograms (a subset): C ------------- C C DASUM DAXPY DCOPY DDOT DNRM2 DSCAL IDAMAX C C These correspond to members of the BLAS package (Basic Linear C Algebra Subprograms, Lawson et al. (1979), ACM TOMS 5, 3). C If possible, they should be replaced by authentic BLAS routines C tuned to the machine being used. C Beware that the versions here work correctly only when C INCX and INCY are 1 (which is the only way MINOS uses them). C C C The following utilities are also used: C C DDDIV DDSCL DNORM DZERO HCOPY ICOPY C C These could be tuned to the machine being used. C DZERO is used the most. C C C 3. File MI20AMAT Core allocation and manipulation of the C ------------- constraint matrix ( A I ): C C M2CORE M2AMAT M2APRD M2APR1 M2APR5 C M2CRSH M2SCAL M2SCLA M2UNPK M2UNP2 MATCOL C C C 4. File MI25BFAC Basis routines for handling the factorization C ------------- of the basis matrix B: C C M2BFAC M2BMAP M2BELM M2BSOL M2SING C LU1FAC LU1MAR LU1OR1 LU1OR2 LU1OR3 LU1OR4 C LU1PQ1 LU1PQ2 LU1REC C LU6CHK LU6SOL LU7ADD LU7ELM LU7FOR LU7ZAP LU8RPC C C C 5. File MI30SPEC SPECS file routines: C ------------- C C M3SPC0 M3SPC1 M3SPC2 C C C 6. FilE MI35INPT MPS file routines: C ------------- C C M3INPT M3MPSA M3MPSB M3MPSC M3READ C C C 7. File MI40BFIL Basis file and solution output routines: C ------------- C C M4GETB M4NAME M4INST M4LOAD M4OLDB M4CHEK C M4SAVB M4DUMP M4NEWB M4PNCH M4SOLN M4SOLP C M4STAT REPORT C C C 8. File MI50LP Routines for the primal simplex method: C ------------- C C M5BSX M5CHZR M5FRMC M5LOG M5LPIT M5PRIC C M5SETP M5SETX M5SOLV C C C 9. File MI60SRCH Routines for the linesearch and subproblem obj: C ------------- C C M6DMMY M6FOBJ M6FCON M6FUN M6FUN1 M6GRD M6GRD1 C M6DOBJ M6DCON M6SRCH GETPTC GETPTQ C C C 10. File MI65RMOD For maintaining R, the quasi-Newton C ------------- approximation to the reduced Hessian: C C M6BFGS M6BSWP M6RADD M6RCND M6RDEL C M6RMOD M6RSET M6RSOL M6SWAP C C C 11. File MI70NOBJ To handle a nonlinear objective function C ------------- via the reduced-gradient algorithm: C C M7BSG M7CHKD M7CHKG M7CHZQ C M7RG M7RGIT M7SDIR M7SSCV C C C 12. File MI80NCON to handle nonlinear constraints C ------------- via the projected augmented Lagrangian algorithm: C C M8AJAC M8AUGL M8AUG1 M8CHKJ M8PRTJ M8SCLJ C M8SETJ M8VIOL C C ****************************************************************** C