SB02MX

Conversion of optimal problems with coupling weighting terms to standard problems

[Specification] [Arguments] [Method] [References] [Comments] [Example]

Purpose

  To compute the following matrices

             -1
      G = B*R  *B',

      -               -1
      A = A +/- op(B*R  *L'),

      -            -1
      Q = Q +/- L*R  *L',

  where A, B, Q, R, L, and G are N-by-N, N-by-M, N-by-N, M-by-M,
  N-by-M, and N-by-N matrices, respectively, with Q, R and G
  symmetric matrices, and op(W) is one of

      op(W) = W  or  op(W) = W'.

  When R is well-conditioned with respect to inversion, standard
  algorithms for solving linear-quadratic optimization problems will
  then also solve optimization problems with coupling weighting
  matrix L. Moreover, a gain in efficiency is possible using matrix
  G in the deflating subspace algorithms (see SLICOT Library routine
  SB02OD) or in the Newton's algorithms (see SLICOT Library routine
  SG02CD).

Specification
      SUBROUTINE SB02MX( JOBG, JOBL, FACT, UPLO, TRANS, FLAG, DEF, N, M,
     $                   A, LDA, B, LDB, Q, LDQ, R, LDR, L, LDL, IPIV,
     $                   OUFACT, G, LDG, IWORK, DWORK, LDWORK, INFO )
C     .. Scalar Arguments ..
      CHARACTER         DEF, FACT, FLAG, JOBG, JOBL, TRANS, UPLO
      INTEGER           INFO, LDA, LDB, LDG, LDL, LDQ, LDR, LDWORK, M,
     $                  N, OUFACT
C     .. Array Arguments ..
      INTEGER           IPIV(*), IWORK(*)
      DOUBLE PRECISION  A(LDA,*), B(LDB,*), DWORK(*), G(LDG,*),
     $                  L(LDL,*), Q(LDQ,*), R(LDR,*)

Arguments

Mode Parameters

  JOBG    CHARACTER*1
          Specifies whether or not the matrix G is to be computed,
          as follows:
          = 'G':  Compute G;
          = 'N':  Do not compute G.

  JOBL    CHARACTER*1
          Specifies whether or not the matrix L is zero, as follows:
          = 'Z':  L is zero;
          = 'N':  L is nonzero.

  FACT    CHARACTER*1
          Specifies how the matrix R is given (factored or not), as
          follows:
          = 'N':  Array R contains the matrix R;
          = 'C':  Array R contains the Cholesky factor of R;
          = 'U':  Array R contains the factors of the symmetric
                  indefinite UdU' or LdL' factorization of R.

  UPLO    CHARACTER*1
          Specifies which triangle of the matrices R, Q (if
          JOBL = 'N'), and G (if JOBG = 'G') is stored, as follows:
          = 'U':  Upper triangle is stored;
          = 'L':  Lower triangle is stored.

  TRANS   CHARACTER*1
          Specifies the form of op(W) to be used in the matrix
          multiplication, as follows:
          = 'N':  op(W) = W;
          = 'T':  op(W) = W';
          = 'C':  op(W) = W'.

  FLAG    CHARACTER*1
          Specifies which sign is used, as follows:
          = 'P':  The plus  sign is used;
          = 'M':  The minus sign is used.

  DEF     CHARACTER*1
          If FACT = 'N', specifies whether or not it is assumed that
          matrix R is positive definite, as follows:
          = 'D':  Matrix R is assumed positive definite;
          = 'I':  Matrix R is assumed indefinite.
          Both values can be used to perform the computations,
          irrespective to the R definiteness, but using the adequate
          value will save some computational effort (see FURTHER
          COMMENTS).

Input/Output Parameters
  N       (input) INTEGER
          The order of the matrices A, Q, and G, and the number of
          rows of the matrices B and L.  N >= 0.

  M       (input) INTEGER
          The order of the matrix R, and the number of columns of
          the matrices B and L.  M >= 0.

  A       (input/output) DOUBLE PRECISION array, dimension (LDA,N)
          On entry, if JOBL = 'N', the leading N-by-N part of this
          array must contain the matrix A.
          On exit, if JOBL = 'N', and INFO = 0, the leading N-by-N
                                                 -
          part of this array contains the matrix A.
          If JOBL = 'Z', this array is not referenced.

  LDA     INTEGER
          The leading dimension of array A.
          LDA >= MAX(1,N) if JOBL = 'N';
          LDA >= 1        if JOBL = 'Z'.

  B       (input/output) DOUBLE PRECISION array, dimension (LDB,M)
          On entry, the leading N-by-M part of this array must
          contain the matrix B.
          On exit, if OUFACT = 1, and INFO = 0, the leading N-by-M
                                                          -1
          part of this array contains the matrix B*chol(R)  .
          On exit, B is unchanged if OUFACT <> 1 (hence also when
          FACT = 'U').

  LDB     INTEGER
          The leading dimension of array B.  LDB >= MAX(1,N).

  Q       (input/output) DOUBLE PRECISION array, dimension (LDQ,N)
          On entry, if JOBL = 'N', the leading N-by-N upper
          triangular part (if UPLO = 'U') or lower triangular part
          (if UPLO = 'L') of this array must contain the upper
          triangular part or lower triangular part, respectively, of
          the symmetric matrix Q. The stricly lower triangular part
          (if UPLO = 'U') or stricly upper triangular part (if
          UPLO = 'L') is not referenced.
          On exit, if JOBL = 'N' and INFO = 0, the leading N-by-N
          upper triangular part (if UPLO = 'U') or lower triangular
          part (if UPLO = 'L') of this array contains the upper
          triangular part or lower triangular part, respectively, of
                               -            -1
          the symmetric matrix Q = Q +/- L*R  *L'.
          If JOBL = 'Z', this array is not referenced.

  LDQ     INTEGER
          The leading dimension of array Q.
          LDQ >= MAX(1,N) if JOBL = 'N';
          LDQ >= 1        if JOBL = 'Z'.

  R       (input/output) DOUBLE PRECISION array, dimension (LDR,M)
          On entry, if FACT = 'N', the leading M-by-M upper
          triangular part (if UPLO = 'U') or lower triangular part
          (if UPLO = 'L') of this array must contain the upper
          triangular part or lower triangular part, respectively,
          of the symmetric input weighting matrix R.
          On entry, if FACT = 'C', the leading M-by-M upper
          triangular part (if UPLO = 'U') or lower triangular part
          (if UPLO = 'L') of this array must contain the Cholesky
          factor of the positive definite input weighting matrix R
          (as produced by LAPACK routine DPOTRF).
          On entry, if FACT = 'U', the leading M-by-M upper
          triangular part (if UPLO = 'U') or lower triangular part
          (if UPLO = 'L') of this array must contain the factors of
          the UdU' or LdL' factorization, respectively, of the
          symmetric indefinite input weighting matrix R (as produced
          by LAPACK routine DSYTRF).
          If FACT = 'N' and DEF = 'D', the stricly lower triangular
          part (if UPLO = 'U') or stricly upper triangular part
          (if UPLO = 'L') of this array is used as workspace (filled
          in by symmetry). If FACT = 'N' and DEF = 'I', the stricly
          lower triangular part (if UPLO = 'U') or stricly upper
          triangular part (if UPLO = 'L') is unchanged.
          On exit, if OUFACT = 1, and INFO = 0 (or INFO = M+1),
          the leading M-by-M upper triangular part (if UPLO = 'U')
          or lower triangular part (if UPLO = 'L') of this array
          contains the Cholesky factor of the given input weighting
          matrix.
          On exit, if OUFACT = 2, and INFO = 0 (or INFO = M+1),
          the leading M-by-M upper triangular part (if UPLO = 'U')
          or lower triangular part (if UPLO = 'L') of this array
          contains the factors of the UdU' or LdL' factorization,
          respectively, of the given input weighting matrix.
          On exit R is unchanged if FACT = 'C' or 'U'.

  LDR     INTEGER
          The leading dimension of array R.  LDR >= MAX(1,M).

  L       (input/output) DOUBLE PRECISION array, dimension (LDL,M)
          On entry, if JOBL = 'N', the leading N-by-M part of this
          array must contain the matrix L.
          On exit, if JOBL = 'N', OUFACT = 1, and INFO = 0, the
          leading N-by-M part of this array contains the matrix
                   -1
          L*chol(R)  .
          On exit, L is unchanged if OUFACT <> 1 (hence also when
          FACT = 'U').
          L is not referenced if JOBL = 'Z'.

  LDL     INTEGER
          The leading dimension of array L.
          LDL >= MAX(1,N) if JOBL = 'N';
          LDL >= 1        if JOBL = 'Z'.

  IPIV    (input/output) INTEGER array, dimension (M)
          On entry, if FACT = 'U', this array must contain details
          of the interchanges performed and the block structure of
          the d factor in the UdU' or LdL' factorization of matrix R
          (as produced by LAPACK routine DSYTRF).
          On exit, if OUFACT = 2, this array contains details of
          the interchanges performed and the block structure of the
          d factor in the UdU' or LdL' factorization of matrix R,
          as produced by LAPACK routine DSYTRF.
          This array is not referenced if FACT = 'C'.

  OUFACT  (output) INTEGER
          Information about the factorization finally used.
          OUFACT = 0:  no factorization of R has been used (M = 0);
          OUFACT = 1:  Cholesky factorization of R has been used;
          OUFACT = 2:  UdU' (if UPLO = 'U') or LdL' (if UPLO = 'L')
                       factorization of R has been used.

  G       (output) DOUBLE PRECISION array, dimension (LDG,N)
          If JOBG = 'G', and INFO = 0, the leading N-by-N upper
          triangular part (if UPLO = 'U') or lower triangular part
          (if UPLO = 'L') of this array contains the upper
          triangular part (if UPLO = 'U') or lower triangular part
                                                              -1
          (if UPLO = 'L'), respectively, of the matrix G = B*R  B'.
          If JOBG = 'N', this array is not referenced.

  LDG     INTEGER
          The leading dimension of array G.
          LDG >= MAX(1,N) if JOBG = 'G';
          LDG >= 1        if JOBG = 'N'.

Workspace
  IWORK   INTEGER array, dimension (M)
          If FACT = 'C' or FACT = 'U', this array is not referenced.

  DWORK   DOUBLE PRECISION array, dimension (LDWORK)
          On exit, if INFO = 0 or LDWORK = -1, DWORK(1) returns the
          optimal value of LDWORK; if FACT = 'N' and LDWORK is set
          as specified below, DWORK(2) contains the reciprocal
          condition number of the given matrix R. DWORK(2) is set to
          zero if M = 0.
          On exit, if LDWORK = -2 on input or INFO = -26, then
          DWORK(1) returns the minimal value of LDWORK.

  LDWORK  INTEGER
          The length of the array DWORK.
          LDWORK >= 1              if FACT = 'C' or  (FACT = 'U' and
                                      JOBG = 'N' and  JOBL = 'Z');
          LDWORK >= MAX(2,3*M)     if FACT = 'N' and  JOBG = 'N' and
                                                      JOBL = 'Z';
          LDWORK >= MAX(2,3*M,N*M) if FACT = 'N' and (JOBG = 'G' or
                                                      JOBL = 'N');
          LDWORK >= MAX(1,N*M)     if FACT = 'U' and (JOBG = 'G' or
                                                      JOBL = 'N').
          For optimum performance LDWORK should be larger than 3*M,
          if FACT = 'N'.

          If LDWORK = -1, an optimal workspace query is assumed; the
          routine only calculates the optimal size of the DWORK
          array, returns this value as the first entry of the DWORK
          array, and no error message is issued by XERBLA.

          If LDWORK = -2, a minimal workspace query is assumed; the
          routine only calculates the minimal size of the DWORK
          array, returns this value as the first entry of the DWORK
          array, and no error message is issued by XERBLA.

Error Indicator
  INFO    INTEGER
          = 0:  successful exit;
          < 0:  if INFO = -i, the i-th argument had an illegal
                value;
          = i:  if the i-th element (1 <= i <= M) of the d factor is
                exactly zero; the UdU' (or LdL') factorization has
                been completed, but the block diagonal matrix d is
                exactly singular;
          = M+1:  if the matrix R is numerically singular.

Method
                         -     -
  The matrices G, and/or A and Q are evaluated using the given or
  computed symmetric factorization of R.

Numerical Aspects
  The routine should not be used when R is ill-conditioned.

Further Comments
  Using argument TRANS allows to avoid the transposition of matrix A
  needed to solve optimal filtering/estimation problems by the same
  routines solving optimal control problems.
  If DEF is set to 'D', but R is indefinite, the computational
  effort for factorization will be approximately double, since
  Cholesky factorization, tried first, will fail, and symmetric
  indefinite factorization will then be used.
  If DEF is set to 'I', but R is positive definite, the
  computational effort will be slightly higher than that when using
  Cholesky factorization. It is recommended to use DEF = 'D' also if
  the definiteness is not known, but M is (much) smaller than N.

Example

Program Text

  None
Program Data
  None
Program Results
  None

Return to index