dspigm.f

Go to the documentation of this file.
00001 C Work performed under the auspices of the U.S. Department of Energy
00002 C by Lawrence Livermore National Laboratory under contract number 
00003 C W-7405-Eng-48.
00004 C
00005       SUBROUTINE DSPIGM (NEQ, TN, Y, YPRIME, SAVR, R, WGHT, MAXL,
00006      *   MAXLP1, KMP, EPLIN, CJ, RES, IRES, NRE, PSOL, NPSL, Z, V,
00007      *   HES, Q, LGMR, WP, IWP, WK, DL, RHOK, IFLAG, IRST, NRSTS,
00008      *   RPAR, IPAR)
00009 C
00010 C***BEGIN PROLOGUE  DSPIGM
00011 C***DATE WRITTEN   890101   (YYMMDD)
00012 C***REVISION DATE  900926   (YYMMDD)
00013 C***REVISION DATE  940927   Removed MNEWT and added RHOK in call list.
00014 C
00015 C
00016 C-----------------------------------------------------------------------
00017 C***DESCRIPTION
00018 C
00019 C This routine solves the linear system A * Z = R using a scaled
00020 C preconditioned version of the generalized minimum residual method.
00021 C An initial guess of Z = 0 is assumed.
00022 C
00023 C      On entry
00024 C
00025 C          NEQ = Problem size, passed to PSOL.
00026 C
00027 C           TN = Current Value of T.
00028 C
00029 C            Y = Array Containing current dependent variable vector.
00030 C
00031 C       YPRIME = Array Containing current first derivative of Y.
00032 C
00033 C         SAVR = Array containing current value of G(T,Y,YPRIME).
00034 C
00035 C            R = The right hand side of the system A*Z = R.
00036 C                R is also used as work space when computing
00037 C                the final approximation and will therefore be
00038 C                destroyed.
00039 C                (R is the same as V(*,MAXL+1) in the call to DSPIGM.)
00040 C
00041 C         WGHT = The vector of length NEQ containing the nonzero
00042 C                elements of the diagonal scaling matrix.
00043 C
00044 C         MAXL = The maximum allowable order of the matrix H.
00045 C
00046 C       MAXLP1 = MAXL + 1, used for dynamic dimensioning of HES.
00047 C
00048 C          KMP = The number of previous vectors the new vector, VNEW,
00049 C                must be made orthogonal to.  (KMP .LE. MAXL.)
00050 C
00051 C        EPLIN = Tolerance on residuals R-A*Z in weighted rms norm.
00052 C
00053 C           CJ = Scalar proportional to current value of 
00054 C                1/(step size H).
00055 C
00056 C           WK = Real work array used by routine DATV and PSOL.
00057 C
00058 C           DL = Real work array used for calculation of the residual
00059 C                norm RHO when the method is incomplete (KMP.LT.MAXL)
00060 C                and/or when using restarting.
00061 C
00062 C           WP = Real work array used by preconditioner PSOL.
00063 C
00064 C          IWP = Integer work array used by preconditioner PSOL.
00065 C
00066 C         IRST = Method flag indicating if restarting is being
00067 C                performed.  IRST .GT. 0 means restarting is active,
00068 C                while IRST = 0 means restarting is not being used.
00069 C
00070 C        NRSTS = Counter for the number of restarts on the current
00071 C                call to DSPIGM.  If NRSTS .GT. 0, then the residual
00072 C                R is already scaled, and so scaling of R is not
00073 C                necessary.
00074 C
00075 C
00076 C      On Return
00077 C
00078 C         Z    = The final computed approximation to the solution
00079 C                of the system A*Z = R.
00080 C
00081 C         LGMR = The number of iterations performed and
00082 C                the current order of the upper Hessenberg
00083 C                matrix HES.
00084 C
00085 C         NRE  = The number of calls to RES (i.e. DATV)
00086 C
00087 C         NPSL = The number of calls to PSOL.
00088 C
00089 C         V    = The neq by (LGMR+1) array containing the LGMR
00090 C                orthogonal vectors V(*,1) to V(*,LGMR).
00091 C
00092 C         HES  = The upper triangular factor of the QR decomposition
00093 C                of the (LGMR+1) by LGMR upper Hessenberg matrix whose
00094 C                entries are the scaled inner-products of A*V(*,I)
00095 C                and V(*,K).
00096 C
00097 C         Q    = Real array of length 2*MAXL containing the components
00098 C                of the givens rotations used in the QR decomposition
00099 C                of HES.  It is loaded in DHEQR and used in DHELS.
00100 C
00101 C         IRES = Error flag from RES.
00102 C
00103 C           DL = Scaled preconditioned residual, 
00104 C                (D-inverse)*(P-inverse)*(R-A*Z). Only loaded when
00105 C                performing restarts of the Krylov iteration.
00106 C
00107 C         RHOK = Weighted norm of final preconditioned residual.
00108 C
00109 C        IFLAG = Integer error flag..
00110 C                0 Means convergence in LGMR iterations, LGMR.LE.MAXL.
00111 C                1 Means the convergence test did not pass in MAXL
00112 C                  iterations, but the new residual norm (RHO) is
00113 C                  .LT. the old residual norm (RNRM), and so Z is
00114 C                  computed.
00115 C                2 Means the convergence test did not pass in MAXL
00116 C                  iterations, new residual norm (RHO) .GE. old residual
00117 C                  norm (RNRM), and the initial guess, Z = 0, is
00118 C                  returned.
00119 C                3 Means there was a recoverable error in PSOL
00120 C                  caused by the preconditioner being out of date.
00121 C               -1 Means there was an unrecoverable error in PSOL.
00122 C
00123 C-----------------------------------------------------------------------
00124 C***ROUTINES CALLED
00125 C   PSOL, DNRM2, DSCAL, DATV, DORTH, DHEQR, DCOPY, DHELS, DAXPY
00126 C
00127 C***END PROLOGUE  DSPIGM
00128 C
00129       INTEGER NEQ,MAXL,MAXLP1,KMP,IRES,NRE,NPSL,LGMR,IWP,
00130      1   IFLAG,IRST,NRSTS,IPAR
00131       DOUBLE PRECISION TN,Y,YPRIME,SAVR,R,WGHT,EPLIN,CJ,Z,V,HES,Q,WP,WK,
00132      1   DL,RHOK,RPAR
00133       DIMENSION Y(*), YPRIME(*), SAVR(*), R(*), WGHT(*), Z(*),
00134      1   V(NEQ,*), HES(MAXLP1,*), Q(*), WP(*), IWP(*), WK(*), DL(*),
00135      2   RPAR(*), IPAR(*)
00136       INTEGER I, IER, INFO, IP1, I2, J, K, LL, LLP1
00137       DOUBLE PRECISION RNRM,C,DLNRM,PROD,RHO,S,SNORMW,DNRM2,TEM
00138       EXTERNAL  RES, PSOL
00139 C
00140       IER = 0
00141       IFLAG = 0
00142       LGMR = 0
00143       NPSL = 0
00144       NRE = 0
00145 C-----------------------------------------------------------------------
00146 C The initial guess for Z is 0.  The initial residual is therefore
00147 C the vector R.  Initialize Z to 0.
00148 C-----------------------------------------------------------------------
00149       DO 10 I = 1,NEQ
00150  10     Z(I) = 0.0D0
00151 C-----------------------------------------------------------------------
00152 C Apply inverse of left preconditioner to vector R if NRSTS .EQ. 0.
00153 C Form V(*,1), the scaled preconditioned right hand side.
00154 C-----------------------------------------------------------------------
00155       IF (NRSTS .EQ. 0) THEN
00156          CALL PSOL (NEQ, TN, Y, YPRIME, SAVR, WK, CJ, WGHT, WP, IWP,
00157      1      R, EPLIN, IER, RPAR, IPAR)
00158          NPSL = 1
00159          IF (IER .NE. 0) GO TO 300
00160          DO 30 I = 1,NEQ
00161  30         V(I,1) = R(I)*WGHT(I)
00162       ELSE
00163          DO 35 I = 1,NEQ
00164  35         V(I,1) = R(I)
00165       ENDIF
00166 C-----------------------------------------------------------------------
00167 C Calculate norm of scaled vector V(*,1) and normalize it
00168 C If, however, the norm of V(*,1) (i.e. the norm of the preconditioned
00169 C residual) is .le. EPLIN, then return with Z=0.
00170 C-----------------------------------------------------------------------
00171       RNRM = DNRM2 (NEQ, V, 1)
00172       IF (RNRM .LE. EPLIN) THEN
00173         RHOK = RNRM
00174         RETURN
00175         ENDIF
00176       TEM = 1.0D0/RNRM
00177       CALL DSCAL (NEQ, TEM, V(1,1), 1)
00178 C-----------------------------------------------------------------------
00179 C Zero out the HES array.
00180 C-----------------------------------------------------------------------
00181       DO 65 J = 1,MAXL
00182         DO 60 I = 1,MAXLP1
00183  60       HES(I,J) = 0.0D0
00184  65     CONTINUE
00185 C-----------------------------------------------------------------------
00186 C Main loop to compute the vectors V(*,2) to V(*,MAXL).
00187 C The running product PROD is needed for the convergence test.
00188 C-----------------------------------------------------------------------
00189       PROD = 1.0D0
00190       DO 90 LL = 1,MAXL
00191         LGMR = LL
00192 C-----------------------------------------------------------------------
00193 C Call routine DATV to compute VNEW = ABAR*V(LL), where ABAR is
00194 C the matrix A with scaling and inverse preconditioner factors applied.
00195 C Call routine DORTH to orthogonalize the new vector VNEW = V(*,LL+1).
00196 C call routine DHEQR to update the factors of HES.
00197 C-----------------------------------------------------------------------
00198         CALL DATV (NEQ, Y, TN, YPRIME, SAVR, V(1,LL), WGHT, Z,
00199      1     RES, IRES, PSOL, V(1,LL+1), WK, WP, IWP, CJ, EPLIN,
00200      1     IER, NRE, NPSL, RPAR, IPAR)
00201         IF (IRES .LT. 0) RETURN
00202         IF (IER .NE. 0) GO TO 300
00203         CALL DORTH (V(1,LL+1), V, HES, NEQ, LL, MAXLP1, KMP, SNORMW)
00204         HES(LL+1,LL) = SNORMW
00205         CALL DHEQR (HES, MAXLP1, LL, Q, INFO, LL)
00206         IF (INFO .EQ. LL) GO TO 120
00207 C-----------------------------------------------------------------------
00208 C Update RHO, the estimate of the norm of the residual R - A*ZL.
00209 C If KMP .LT. MAXL, then the vectors V(*,1),...,V(*,LL+1) are not
00210 C necessarily orthogonal for LL .GT. KMP.  The vector DL must then
00211 C be computed, and its norm used in the calculation of RHO.
00212 C-----------------------------------------------------------------------
00213         PROD = PROD*Q(2*LL)
00214         RHO = ABS(PROD*RNRM)
00215         IF ((LL.GT.KMP) .AND. (KMP.LT.MAXL)) THEN
00216           IF (LL .EQ. KMP+1) THEN
00217             CALL DCOPY (NEQ, V(1,1), 1, DL, 1)
00218             DO 75 I = 1,KMP
00219               IP1 = I + 1
00220               I2 = I*2
00221               S = Q(I2)
00222               C = Q(I2-1)
00223               DO 70 K = 1,NEQ
00224  70             DL(K) = S*DL(K) + C*V(K,IP1)
00225  75           CONTINUE
00226             ENDIF
00227           S = Q(2*LL)
00228           C = Q(2*LL-1)/SNORMW
00229           LLP1 = LL + 1
00230           DO 80 K = 1,NEQ
00231  80         DL(K) = S*DL(K) + C*V(K,LLP1)
00232           DLNRM = DNRM2 (NEQ, DL, 1)
00233           RHO = RHO*DLNRM
00234           ENDIF
00235 C-----------------------------------------------------------------------
00236 C Test for convergence.  If passed, compute approximation ZL.
00237 C If failed and LL .LT. MAXL, then continue iterating.
00238 C-----------------------------------------------------------------------
00239         IF (RHO .LE. EPLIN) GO TO 200
00240         IF (LL .EQ. MAXL) GO TO 100
00241 C-----------------------------------------------------------------------
00242 C Rescale so that the norm of V(1,LL+1) is one.
00243 C-----------------------------------------------------------------------
00244         TEM = 1.0D0/SNORMW
00245         CALL DSCAL (NEQ, TEM, V(1,LL+1), 1)
00246  90     CONTINUE
00247  100  CONTINUE
00248       IF (RHO .LT. RNRM) GO TO 150
00249  120  CONTINUE
00250       IFLAG = 2
00251       DO 130 I = 1,NEQ
00252  130     Z(I) = 0.D0
00253       RETURN
00254  150  IFLAG = 1
00255 C-----------------------------------------------------------------------
00256 C The tolerance was not met, but the residual norm was reduced.
00257 C If performing restarting (IRST .gt. 0) calculate the residual vector
00258 C RL and store it in the DL array.  If the incomplete version is 
00259 C being used (KMP .lt. MAXL) then DL has already been calculated.
00260 C-----------------------------------------------------------------------
00261       IF (IRST .GT. 0) THEN
00262          IF (KMP .EQ. MAXL) THEN
00263 C
00264 C           Calculate DL from the V(I)'s.
00265 C
00266             CALL DCOPY (NEQ, V(1,1), 1, DL, 1)
00267             MAXLM1 = MAXL - 1
00268             DO 175 I = 1,MAXLM1
00269                IP1 = I + 1
00270                I2 = I*2
00271                S = Q(I2)
00272                C = Q(I2-1)
00273                DO 170 K = 1,NEQ
00274  170              DL(K) = S*DL(K) + C*V(K,IP1)
00275  175        CONTINUE
00276             S = Q(2*MAXL)
00277             C = Q(2*MAXL-1)/SNORMW
00278             DO 180 K = 1,NEQ
00279  180           DL(K) = S*DL(K) + C*V(K,MAXLP1)
00280          ENDIF
00281 C
00282 C        Scale DL by RNRM*PROD to obtain the residual RL.
00283 C
00284          TEM = RNRM*PROD
00285          CALL DSCAL(NEQ, TEM, DL, 1)
00286       ENDIF
00287 C-----------------------------------------------------------------------
00288 C Compute the approximation ZL to the solution.
00289 C Since the vector Z was used as work space, and the initial guess
00290 C of the Newton correction is zero, Z must be reset to zero.
00291 C-----------------------------------------------------------------------
00292  200  CONTINUE
00293       LL = LGMR
00294       LLP1 = LL + 1
00295       DO 210 K = 1,LLP1
00296  210    R(K) = 0.0D0
00297       R(1) = RNRM
00298       CALL DHELS (HES, MAXLP1, LL, Q, R)
00299       DO 220 K = 1,NEQ
00300  220    Z(K) = 0.0D0
00301       DO 230 I = 1,LL
00302         CALL DAXPY (NEQ, R(I), V(1,I), 1, Z, 1)
00303  230    CONTINUE
00304       DO 240 I = 1,NEQ
00305  240    Z(I) = Z(I)/WGHT(I)
00306 C Load RHO into RHOK.
00307       RHOK = RHO
00308       RETURN
00309 C-----------------------------------------------------------------------
00310 C This block handles error returns forced by routine PSOL.
00311 C-----------------------------------------------------------------------
00312  300  CONTINUE
00313       IF (IER .LT. 0) IFLAG = -1
00314       IF (IER .GT. 0) IFLAG = 3
00315 C
00316       RETURN
00317 C
00318 C------END OF SUBROUTINE DSPIGM-----------------------------------------
00319       END
 All Classes Files Functions Variables Typedefs Enumerations Enumerator Friends Defines