SLATEC Routines --- LSSUDS ---


*DECK LSSUDS
      SUBROUTINE LSSUDS (A, X, B, N, M, NRDA, U, NRDU, IFLAG, MLSO,
     +   IRANK, ISCALE, Q, DIAG, KPIVOT, S, DIV, TD, ISFLG, SCALES)
C***BEGIN PROLOGUE  LSSUDS
C***SUBSIDIARY
C***PURPOSE  Subsidiary to BVSUP
C***LIBRARY   SLATEC
C***TYPE      SINGLE PRECISION (LSSUDS-S, DLSSUD-D)
C***AUTHOR  Watts, H. A., (SNLA)
C***DESCRIPTION
C
C    LSSUDS solves the underdetermined system of equations  A Z = B,
C    where A is N by M and N .LE. M.  In particular, if rank A equals
C    IRA, a vector X and a matrix U are determined such that X is the
C    UNIQUE solution of smallest length, satisfying A X = B, and the
C    columns of U form an orthonormal basis for the null space of A,
C    satisfying A U = 0 .  Then all solutions Z are given by
C              Z = X + C(1)*U(1) + ..... + C(M-IRA)*U(M-IRA)
C    where U(J) represents the J-th column of U and the C(J) are
C    arbitrary constants.
C    If the system of equations are not compatible, only the least
C    squares solution of minimal length is computed.
C
C *********************************************************************
C   INPUT
C *********************************************************************
C
C     A -- Contains the matrix of N equations in M unknowns, A remains
C          unchanged, must be dimensioned NRDA by M.
C     X -- Solution array of length at least M.
C     B -- Given constant vector of length N, B remains unchanged.
C     N -- Number of equations, N greater or equal to 1.
C     M -- Number of unknowns, M greater or equal to N.
C     NRDA -- Row dimension of A, NRDA greater or equal to N.
C     U -- Matrix used for solution, must be dimensioned NRDU by
C          (M - rank of A).
C          (storage for U may be ignored when only the minimal length
C           solution X is desired)
C     NRDU -- Row dimension of U, NRDU greater or equal to M.
C             (if only the minimal length solution is wanted,
C              NRDU=0 is acceptable)
C     IFLAG -- Status indicator
C           =0  for the first call (and for each new problem defined by
C               a new matrix A) when the matrix data is treated as exact
C           =-K for the first call (and for each new problem defined by
C               a new matrix A) when the matrix data is assumed to be
C               accurate to about K digits.
C           =1  for subsequent calls whenever the matrix A has already
C               been decomposed (problems with new vectors B but
C               same matrix A can be handled efficiently).
C     MLSO -- =0 if only the minimal length solution is wanted.
C             =1 if the complete solution is wanted, includes the
C                linear space defined by the matrix U.
C     IRANK -- Variable used for the rank of A, set by the code.
C     ISCALE -- Scaling indicator
C               =-1 if the matrix A is to be pre-scaled by
C               columns when appropriate.
C               If the scaling indicator is not equal to -1
C               no scaling will be attempted.
C            For most problems scaling will probably not be necessary.
C     Q -- Matrix used for the transformation, must be dimensioned
C            NRDA by M.
C     DIAG,KPIVOT,S, -- Arrays of length at least N used for internal
C      DIV,TD,SCALES    storage (except for SCALES which is M).
C     ISFLG -- Storage for an internal variable.
C
C *********************************************************************
C   OUTPUT
C *********************************************************************
C
C     IFLAG -- Status indicator
C            =1 if solution was obtained.
C            =2 if improper input is detected.
C            =3 if rank of matrix is less than N.
C               To continue, simply reset IFLAG=1 and call LSSUDS again.
C            =4 if the system of equations appears to be inconsistent.
C               However, the least squares solution of minimal length
C               was obtained.
C     X -- Minimal length least squares solution of A Z = B
C     IRANK -- Numerically determined rank of A, must not be altered
C              on succeeding calls with input values of IFLAG=1.
C     U -- Matrix whose M-IRANK columns are mutually orthogonal unit
C          vectors which span the null space of A. This is to be ignored
C          when MLSO was set to zero or IFLAG=4 on output.
C     Q -- Contains the strictly upper triangular part of the reduced
C           matrix and transformation information.
C     DIAG -- Contains the diagonal elements of the triangular reduced
C             matrix.
C     KPIVOT -- Contains the pivotal information.  The row interchanges
C               performed on the original matrix are recorded here.
C     S -- Contains the solution of the lower triangular system.
C     DIV,TD -- Contains transformation information for rank
C               deficient problems.
C     SCALES -- Contains the column scaling parameters.
C
C *********************************************************************
C
C***SEE ALSO  BVSUP
C***REFERENCES  H. A. Watts, Solving linear least squares problems
C                 using SODS/SUDS/CODS, Sandia Report SAND77-0683,
C                 Sandia Laboratories, 1977.
C***ROUTINES CALLED  J4SAVE, OHTROL, ORTHOR, R1MACH, SDOT, XERMAX,
C                    XERMSG, XGETF, XSETF
C***REVISION HISTORY  (YYMMDD)
C   750601  DATE WRITTEN
C   890531  Changed all specific intrinsics to generic.  (WRB)
C   890831  Modified array declarations.  (WRB)
C   891214  Prologue converted to Version 4.0 format.  (BAB)
C   900315  CALLs to XERROR changed to CALLs to XERMSG.  (THJ)
C   900328  Added TYPE section.  (WRB)
C   900510  Fixed an error message.  (RWC)
C   910408  Updated the AUTHOR and REFERENCES sections.  (WRB)
C   920501  Reformatted the REFERENCES section.  (WRB)
C***END PROLOGUE  LSSUDS