MB04HD

Reducing a special real block (anti-)diagonal skew-Hamiltonian/Hamiltonian pencil to generalized Schur form

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

Purpose

  To compute the transformed matrices A and B, using orthogonal
  matrices Q1 and Q2 for a real N-by-N regular pencil

                ( A11   0  )     (  0   B12 )
    aA - bB = a (          ) - b (          ),                   (1)
                (  0   A22 )     ( B21   0  )

  where A11, A22 and B12 are upper triangular, B21 is upper
  quasi-triangular and the generalized matrix product
     -1        -1
  A11   B12 A22   B21 is in periodic Schur form, such that the
  matrix Q2' A Q1 is upper triangular, Q2' B Q1 is upper
  quasi-triangular and the transformed pencil
  a(Q2' A Q1) - b(Q2' B Q1) is in generalized Schur form. The
  notation M' denotes the transpose of the matrix M.

Specification
      SUBROUTINE MB04HD( COMPQ1, COMPQ2, N, A, LDA, B, LDB, Q1, LDQ1,
     $                   Q2, LDQ2, IWORK, LIWORK, DWORK, LDWORK, BWORK,
     $                   INFO )
C     .. Scalar Arguments ..
      CHARACTER          COMPQ1, COMPQ2
      INTEGER            INFO, LDA, LDB, LDQ1, LDQ2, LDWORK, LIWORK, N
C     .. Array Arguments ..
      LOGICAL            BWORK( * )
      INTEGER            IWORK( * )
      DOUBLE PRECISION   A( LDA, * ), B( LDB, * ), DWORK( * ),
     $                   Q1( LDQ1, * ), Q2( LDQ2, * )

Arguments

Mode Parameters

  COMPQ1  CHARACTER*1
          Specifies whether to compute the orthogonal transformation
          matrix Q1, as follows:
          = 'N':  Q1 is not computed;
          = 'I':  the array Q1 is initialized internally to the unit
                  matrix, and the orthogonal matrix Q1 is returned;
          = 'U':  the array Q1 contains an orthogonal matrix Q01 on
                  entry, and the matrix Q01*Q1 is returned, where Q1
                  is the product of the orthogonal transformations
                  that are applied on the right to the pencil
                  aA - bB in (1).

  COMPQ2  CHARACTER*1
          Specifies whether to compute the orthogonal transformation
          matrix Q2, as follows:
          = 'N':  Q2 is not computed;
          = 'I':  the array Q2 is initialized internally to the unit
                  matrix, and the orthogonal matrix Q2 is returned;
          = 'U':  the array Q2 contains an orthogonal matrix Q02 on
                  entry, and the matrix Q02*Q2 is returned, where Q2
                  is the product of the orthogonal transformations
                  that are applied on the left to the pencil aA - bB
                  in (1).

Input/Output Parameters
  N       (input) INTEGER
          Order of the pencil aA - bB, N >= 0, even.

  A       (input/output) DOUBLE PRECISION array, dimension (LDA, N)
          On entry, the leading N-by-N block diagonal part of this
          array must contain the matrix A in (1). The off-diagonal
          blocks need not be set to zero.
          On exit, the leading N-by-N part of this array contains
          the transformed upper triangular matrix.

  LDA     INTEGER
          The leading dimension of the array A.  LDA >= MAX(1, N).

  B       (input/output) DOUBLE PRECISION array, dimension (LDB, N)
          On entry, the leading N-by-N block anti-diagonal part of
          this array must contain the matrix B in (1). The diagonal
          blocks need not be set to zero.
          On exit, the leading N-by-N part of this array contains
          the transformed upper quasi-triangular matrix.

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

  Q1      (input/output) DOUBLE PRECISION array, dimension (LDQ1, N)
          On entry, if COMPQ1 = 'U', then the leading N-by-N part of
          this array must contain a given matrix Q01, and on exit,
          the leading N-by-N part of this array contains the product
          of the input matrix Q01 and the transformation matrix Q1
          used to transform the matrices A and B.
          On exit, if COMPQ1 = 'I', then the leading N-by-N part of
          this array contains the orthogonal transformation matrix
          Q1.
          If COMPQ1 = 'N' this array is not referenced.

  LDQ1    INTEGER
          The leading dimension of the array Q1.
          LDQ1 >= 1,         if COMPQ1 = 'N';
          LDQ1 >= MAX(1, N), if COMPQ1 = 'I' or COMPQ1 = 'U'.

  Q2      (input/output) DOUBLE PRECISION array, dimension (LDQ2, N)
          On entry, if COMPQ2 = 'U', then the leading N-by-N part of
          this array must contain a given matrix Q02, and on exit,
          the leading N-by-N part of this array contains the product
          of the input matrix Q02 and the transformation matrix Q2
          used to transform the matrices A and B.
          On exit, if COMPQ2 = 'I', then the leading N-by-N part of
          this array contains the orthogonal transformation matrix
          Q2.
          If COMPQ2 = 'N' this array is not referenced.

  LDQ2    INTEGER
          The leading dimension of the array Q2.
          LDQ2 >= 1,         if COMPQ2 = 'N';
          LDQ2 >= MAX(1, N), if COMPQ2 = 'I' or COMPQ2 = 'U'.

Workspace
  IWORK   INTEGER array, dimension (LIWORK)

  LIWORK  INTEGER
          The dimension of the array IWORK.
          LIWORK >= MAX( N/2+1, 32 ).

  DWORK   DOUBLE PRECISION array, dimension (LDWORK)
          On exit, if INFO = 0, DWORK(1) returns the optimal LDWORK.
          On exit, if INFO = -16, DWORK(1) returns the minimum value
          of LDWORK.

  LDWORK  INTEGER
          The dimension of the array DWORK.
          LDWORK >= 2*N*N + MAX( N/2 + 168, 272 ).
          For good performance LDWORK should be generally larger.

          If LDWORK = -1  a 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. 

  BWORK   LOGICAL array, dimension (N/2)

Error Indicator
  INFO    INTEGER
          = 0: succesful exit;
          < 0: if INFO = -i, the i-th argument had an illegal value;
          = 1: the periodic QZ algorithm failed to reorder the
               eigenvalues (the problem is very ill-conditioned) in
               the SLICOT Library routine MB03KD;
          = 2: the standard QZ algorithm failed in the LAPACK
               routines DGGES or DHGEQZ, called by the SLICOT
               routines MB03DD or MB03FD;
          = 3: the eigenvalue reordering failed in the LAPACK
               routine DTGEX2, called by the SLICOT routine MB03FD;
          = 4: the standard QZ algorithm failed to reorder the
               eigenvalues in the LAPACK routine DTGSEN, called by
               the SLICOT routine MB03DD.

Method
  First, the periodic QZ algorithm (see also [2] and [3]) is applied
                                  -1        -1
  to the formal matrix product A11   B12 A22   B21 to reorder the
  eigenvalues, i.e., orthogonal matrices V1, V2, V3 and V4 are
  computed such that V2' A11 V1, V2' B12 V3, V4' A22 V3 and
  V4' B21 V1 keep the triangular form, but they can be partitioned
  into 2-by-2 block forms and the last diagonal blocks correspond to
  all nonpositive real eigenvalues of the formal product, and the
  first diagonal blocks correspond to the remaining eigenvalues.

  Second, Q1 = diag(V1, V3), Q2 = diag(V2, V4) and

                   ( AA11 AA12   0    0  )
                   (                     )
                   (   0  AA22   0    0  )
  A := Q2' A Q1 =: (                     ),
                   (   0    0  AA33 AA34 )
                   (                     )
                   (   0    0    0  AA44 )

                   (   0    0  BB13 BB14 )
                   (                     )
                   (   0    0    0  BB24 )
  B := Q2' B Q1 =: (                     ),
                   ( BB31 BB32   0    0  )
                   (                     )
                   (   0  BB42   0    0  )

                         -1          -1
  are set, such that AA22   BB24 AA44   BB42 has only nonpositive
  real eigenvalues.

  Third, the permutation matrix

      (  I  0  0  0  )
      (              )
      (  0  0  I  0  )
  P = (              ),
      (  0  I  0  0  )
      (              )
      (  0  0  0  I  )

  where I denotes the identity matrix of appropriate size, is used
  to transform aA - bB to block upper triangular form

                ( AA11   0  | AA12   0  )
                (           |           )
                (   0  AA33 |   0  AA34 )   ( AA1  *  )
  A := P' A P = (-----------+-----------) = (         ),
                (   0    0  | AA22   0  )   (  0  AA2 )
                (           |           )
                (   0    0  |   0  AA44 )

                (   0  BB13 |   0  BB14 )
                (           |           )
                ( BB31   0  | BB32   0  )   ( BB1  *  )
  B := P' B P = (-----------+-----------) = (         ).
                (   0    0  |   0  BB24 )   (  0  BB2 )
                (           |           )
                (   0    0  | BB42   0  )

  Then, further orthogonal transformations that are provided by
  MB03FD and MB03DD are used to triangularize the subpencil
  aAA1 - bBB1.

  Finally, the subpencil aAA2 - bBB2 is triangularized by applying a
  special permutation matrix.

  See also page 31 in [1] for more details.

References
  [1] Benner, P., Byers, R., Losse, P., Mehrmann, V. and Xu, H.
      Numerical Solution of Real Skew-Hamiltonian/Hamiltonian
      Eigenproblems.
      Tech. Rep., Technical University Chemnitz, Germany,
      Nov. 2007.

  [2] Bojanczyk, A., Golub, G. H. and Van Dooren, P.
      The periodic Schur decomposition: algorithms and applications.
      In F.T. Luk (editor), Advanced Signal Processing Algorithms,
      Architectures, and Implementations III, Proc. SPIE Conference,
      vol. 1770, pp. 31-42, 1992.

  [3] Hench, J. J. and Laub, A. J.
      Numerical Solution of the discrete-time periodic Riccati
      equation. IEEE Trans. Automat. Control, 39, 1197-1210, 1994.

Numerical Aspects
                                                            3
  The algorithm is numerically backward stable and needs O(N ) real
  floating point operations.

Further Comments
  None
Example

Program Text

  None
Program Data
  None
Program Results
  None

Return to Supporting Routines index