Purpose
To compute the eigenvalues of a real N-by-N skew-Hamiltonian/
skew-Hamiltonian pencil aS - bT with
( B F ) ( 0 I )
S = J Z' J' Z and T = ( ), where J = ( ). (1)
( G B' ) ( -I 0 )
Optionally, if JOB = 'T', the pencil aS - bT will be transformed
to the structured Schur form: an orthogonal transformation matrix
Q and an orthogonal symplectic transformation matrix U are
computed, such that
( Z11 Z12 )
U' Z Q = ( ) = Zout, and
( 0 Z22 )
(2)
( Bout Fout )
J Q' J' T Q = ( ),
( 0 Bout' )
where Z11 and Z22' are upper triangular and Bout is upper quasi-
triangular. The notation M' denotes the transpose of the matrix M.
Optionally, if COMPQ = 'I', the orthogonal transformation matrix Q
will be computed.
Optionally, if COMPU = 'I' or COMPU = 'U', the orthogonal
symplectic transformation matrix
( U1 U2 )
U = ( )
( -U2 U1 )
will be computed.
Specification
SUBROUTINE MB04ED( JOB, COMPQ, COMPU, N, Z, LDZ, B, LDB, FG, LDFG,
$ Q, LDQ, U1, LDU1, U2, LDU2, ALPHAR, ALPHAI,
$ BETA, IWORK, LIWORK, DWORK, LDWORK, INFO )
C .. Scalar Arguments ..
CHARACTER COMPQ, COMPU, JOB
INTEGER INFO, LDB, LDFG, LDQ, LDU1, LDU2, LDWORK, LDZ,
$ LIWORK, N
C .. Array Arguments ..
INTEGER IWORK( * )
DOUBLE PRECISION ALPHAI( * ), ALPHAR( * ), B( LDB, * ),
$ BETA( * ), DWORK( * ), FG( LDFG, * ),
$ Q( LDQ, * ), U1( LDU1, * ), U2( LDU2, * ),
$ Z( LDZ, * )
Arguments
Mode Parameters
JOB CHARACTER*1
Specifies the computation to be performed, as follows:
= 'E': compute the eigenvalues only; Z and T will not
necessarily be put into the forms in (2);
= 'T': put Z and T into the forms in (2), and return the
eigenvalues in ALPHAR, ALPHAI and BETA.
COMPQ CHARACTER*1
Specifies whether to compute the orthogonal transformation
matrix Q as follows:
= 'N': Q is not computed;
= 'I': the array Q is initialized internally to the unit
matrix, and the orthogonal matrix Q is returned.
COMPU CHARACTER*1
Specifies whether to compute the orthogonal symplectic
transformation matrix U as follows:
= 'N': U is not computed;
= 'I': the array U is initialized internally to the unit
matrix, and the orthogonal matrix U is returned;
= 'U': the arrays U1 and U2 contain the corresponding
submatrices of an orthogonal symplectic matrix U0
on entry, and the updated submatrices U1 and U2
of the matrix product U0*U are returned, where U
is the product of the orthogonal symplectic
transformations that are applied to the pencil
aS - bT to reduce Z and T to the forms in (2), for
COMPU = 'I'.
Input/Output Parameters
N (input) INTEGER
The order of the pencil aS - bT. N >= 0, even.
Z (input/output) DOUBLE PRECISION array, dimension (LDZ, N)
On entry, the leading N-by-N part of this array must
contain the matrix Z.
On exit, if JOB = 'T', the leading N-by-N part of this
array contains the matrix Zout; otherwise, it contains the
matrix Z just before the application of the periodic QZ
algorithm. The entries in the rows N/2+1 to N and the
first N/2 columns are unchanged.
LDZ INTEGER
The leading dimension of the array Z. LDZ >= MAX(1, N).
B (input/output) DOUBLE PRECISION array, dimension
(LDB, N/2)
On entry, the leading N/2-by-N/2 part of this array must
contain the matrix B.
On exit, if JOB = 'T', the leading N/2-by-N/2 part of this
array contains the matrix Bout; otherwise, it contains the
matrix B just before the application of the periodic QZ
algorithm.
LDB INTEGER
The leading dimension of the array B. LDB >= MAX(1, N/2).
FG (input/output) DOUBLE PRECISION array, dimension
(LDFG, N/2+1)
On entry, the leading N/2-by-N/2 strictly lower triangular
part of this array must contain the strictly lower
triangular part of the skew-symmetric matrix G, and the
N/2-by-N/2 strictly upper triangular part of the submatrix
in the columns 2 to N/2+1 of this array must contain the
strictly upper triangular part of the skew-symmetric
matrix F.
On exit, if JOB = 'T', the leading N/2-by-N/2 strictly
upper triangular part of the submatrix in the columns 2 to
N/2+1 of this array contains the strictly upper triangular
part of the skew-symmetric matrix Fout.
If JOB = 'E', the leading N/2-by-N/2 strictly upper
triangular part of the submatrix in the columns 2 to N/2+1
of this array contains the strictly upper triangular part
of the skew-symmetric matrix F just before the application
of the QZ algorithm.
The entries on the diagonal and the first superdiagonal of
this array are not referenced, but are assumed to be zero.
Moreover, the diagonal and the first subdiagonal of this
array on exit coincide to the corresponding diagonals of
this array on entry.
LDFG INTEGER
The leading dimension of the array FG.
LDFG >= MAX(1, N/2).
Q (output) DOUBLE PRECISION array, dimension (LDQ, N)
On exit, if COMPQ = 'I', the leading N-by-N part of this
array contains the orthogonal transformation matrix Q.
On exit, if COMPQ = 'N', the leading N-by-N part of this
array contains the orthogonal matrix Q1, such that
( Z11 Z12 )
Z*Q1 = ( ),
( 0 Z22 )
where Z11 and Z22' are upper triangular (the first step
of the algorithm).
LDQ INTEGER
The leading dimension of the array Q. LDQ >= MAX(1, N).
U1 (input/output) DOUBLE PRECISION array, dimension
(LDU1, N/2)
On entry, if COMPU = 'U', then the leading N/2-by-N/2 part
of this array must contain the upper left block of a
given matrix U0, and on exit, the leading N/2-by-N/2 part
of this array contains the updated upper left block U1 of
the product of the input matrix U0 and the transformation
matrix U used to transform the matrices Z and T.
On exit, if COMPU = 'I', then the leading N/2-by-N/2 part
of this array contains the upper left block U1 of the
orthogonal symplectic transformation matrix U.
If COMPU = 'N' this array is not referenced.
LDU1 INTEGER
The leading dimension of the array U1.
LDU1 >= 1, if COMPU = 'N';
LDU1 >= MAX(1, N/2), if COMPU = 'I' or COMPU = 'U'.
U2 (input/output) DOUBLE PRECISION array, dimension
(LDU2, N/2)
On entry, if COMPU = 'U', then the leading N/2-by-N/2 part
of this array must contain the upper right block of a
given matrix U0, and on exit, the leading N/2-by-N/2 part
of this array contains the updated upper right block U2 of
the product of the input matrix U0 and the transformation
matrix U used to transform the matrices Z and T.
On exit, if COMPU = 'I', then the leading N/2-by-N/2 part
of this array contains the upper right block U2 of the
orthogonal symplectic transformation matrix U.
If COMPU = 'N' this array is not referenced.
LDU2 INTEGER
The leading dimension of the array U2.
LDU2 >= 1, if COMPU = 'N';
LDU2 >= MAX(1, N/2), if COMPU = 'I' or COMPU = 'U'.
ALPHAR (output) DOUBLE PRECISION array, dimension (N/2)
The real parts of each scalar alpha defining an eigenvalue
of the pencil aS - bT.
ALPHAI (output) DOUBLE PRECISION array, dimension (N/2)
The imaginary parts of each scalar alpha defining an
eigenvalue of the pencil aS - bT.
If ALPHAI(j) is zero, then the j-th eigenvalue is real; if
positive, then the j-th and (j+1)-st eigenvalues are a
complex conjugate pair, with ALPHAI(j+1) = -ALPHAI(j).
BETA (output) DOUBLE PRECISION array, dimension (N/2)
The scalars beta that define the eigenvalues of the pencil
aS - bT.
Together, the quantities alpha = (ALPHAR(j),ALPHAI(j)) and
beta = BETA(j) represent the j-th eigenvalue of the pencil
aS - bT, in the form lambda = alpha/beta. Since lambda may
overflow, the ratios should not, in general, be computed.
Due to the skew-Hamiltonian/skew-Hamiltonian structure of
the pencil, every eigenvalue occurs twice and thus it has
only to be saved once in ALPHAR, ALPHAI and BETA.
Workspace
IWORK INTEGER array, dimension (LIWORK)
On exit, if INFO = 3, IWORK(1) contains the number of
(pairs of) possibly inaccurate eigenvalues, q <= N/2, and
IWORK(2), ..., IWORK(q+1) indicate their indices.
Specifically, a positive value is an index of a real or
purely imaginary eigenvalue, corresponding to a 1-by-1
block, while the absolute value of a negative entry in
IWORK is an index to the first eigenvalue in a pair of
consecutively stored eigenvalues, corresponding to a
2-by-2 block. A 2-by-2 block may have two complex, two
real, two purely imaginary, or one real and one purely
imaginary eigenvalue.
For i = q+2, ..., 2*q+1, IWORK(i) contains a pointer to
the starting location in DWORK of the i-th triplet of
1-by-1 blocks, if IWORK(i-q) > 0, or 2-by-2 blocks,
if IWORK(i-q) < 0, defining unreliable eigenvalues.
IWORK(2*q+2) contains the number of the 1-by-1 blocks, and
IWORK(2*q+3) contains the number of the 2-by-2 blocks,
corresponding to unreliable eigenvalues. IWORK(2*q+4)
contains the total number t of the 2-by-2 blocks.
If INFO = 0, then q = 0, therefore IWORK(1) = 0.
LIWORK INTEGER
The dimension of the array IWORK. LIWORK >= N+9.
DWORK DOUBLE PRECISION array, dimension (LDWORK)
On exit, if INFO = 0 or INFO = 3, DWORK(1) returns the
optimal LDWORK, and DWORK(2), ..., DWORK(4) contain the
Frobenius norms of the factors of the formal matrix
product used by the algorithm. In addition, DWORK(5), ...,
DWORK(4+3*s) contain the s triplet values corresponding
to the 1-by-1 blocks. Their eigenvalues are real or purely
imaginary. Such an eigenvalue is obtained as a1/a2/a3,
where a1, ..., a3 are the corresponding triplet values.
Moreover, DWORK(5+3*s), ..., DWORK(4+3*s+12*t) contain the
t groups of triplet 2-by-2 matrices corresponding to the
2-by-2 blocks. Their eigenvalue pairs are either complex,
or placed on the real and imaginary axes. Such an
eigenvalue pair is the spectrum of the matrix product
A1*inv(A2)*inv(A3), where A1, ..., A3 define the
corresponding 2-by-2 matrix triplet.
On exit, if INFO = -23, DWORK(1) returns the minimum value
of LDWORK.
LDWORK INTEGER
The dimension of the array DWORK.
If JOB = 'E' and COMPQ = 'N' and COMPU = 'N',
LDWORK >= 3/4*N**2+MAX(3*N, 27);
else, LDWORK >= 3/2*N**2+MAX(3*N, 27).
For good performance LDWORK should generally be larger.
If LDWORK = -1, then 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 related to LDWORK is issued by
XERBLA.
Error Indicator
INFO INTEGER
= 0: succesful exit;
< 0: if INFO = -i, the i-th argument had an illegal value;
= 1: problem during computation of the eigenvalues;
= 2: periodic QZ algorithm did not converge in the SLICOT
Library subroutine MB03BD;
= 3: some eigenvalues might be inaccurate, and details can
be found in IWORK and DWORK. This is a warning.
Method
The algorithm uses Givens rotations and Householder reflections to
annihilate elements in Z and T such that Z is in a special block
triangular form and T is in skew-Hamiltonian Hessenberg form:
( Z11 Z12 ) ( B1 F1 )
Z = ( ), T = ( ),
( 0 Z22 ) ( 0 B1' )
with Z11 and Z22' upper triangular and B1 upper Hessenberg.
Subsequently, the periodic QZ algorithm is applied to the pencil
aZ22' Z11 - bB1 to determine orthogonal matrices Q1, Q2 and U such
that U' Z11 Q1, Q2' Z22' U are upper triangular and Q2' B1 Q1 is
upper quasi-triangular. See also page 35 in [1] for more details.
References
[1] Benner, P., Byers, R., Mehrmann, V. and Xu, H.
Numerical Computation of Deflating Subspaces of Embedded
Hamiltonian Pencils.
Tech. Rep. SFB393/99-15, Technical University Chemnitz,
Germany, June 1999.
Numerical Aspects
3 The algorithm is numerically backward stable and needs O(N ) real floating point operations.Further Comments
NoneExample
Program Text
* MB04ED EXAMPLE PROGRAM TEXT
*
* .. Parameters ..
INTEGER NIN, NOUT
PARAMETER ( NIN = 5, NOUT = 6 )
INTEGER NMAX
PARAMETER ( NMAX = 60 )
INTEGER LDB, LDFG, LDQ, LDU1, LDU2, LDWORK, LDZ,
$ LIWORK
PARAMETER ( LDB = NMAX/2, LDFG = NMAX/2,
$ LDQ = NMAX, LDU1 = NMAX/2, LDU2 = NMAX/2,
$ LDWORK = 3*NMAX**2/2 + MAX( NMAX, 24 ) + 3,
$ LDZ = NMAX, LIWORK = NMAX + 9 )
*
* .. Local Scalars ..
CHARACTER COMPQ, COMPU, JOB
INTEGER I, INFO, J, N
*
* .. Local Arrays ..
INTEGER IWORK( LIWORK )
DOUBLE PRECISION ALPHAI( NMAX/2 ), ALPHAR( NMAX/2 ),
$ B( LDB, NMAX/2 ), BETA( NMAX/2 ),
$ DWORK( LDWORK ), FG( LDFG, NMAX/2+1 ),
$ Q( LDQ, NMAX ), U1( LDU1, NMAX/2 ),
$ U2( LDU2, NMAX/2 ), Z( LDZ, NMAX )
*
* .. External Functions ..
LOGICAL LSAME
EXTERNAL LSAME
*
* .. External Subroutines ..
EXTERNAL MB04ED
*
* .. Intrinsic Functions ..
INTRINSIC MAX, MOD
*
* .. Executable statements ..
*
WRITE( NOUT, FMT = 99999 )
*
* Skip first line in data file.
*
READ( NIN, FMT = * )
READ( NIN, FMT = * ) JOB, COMPQ, COMPU, N
READ( NIN, FMT = * ) ( ( Z( I, J ), J = 1, N ), I = 1, N )
READ( NIN, FMT = * ) ( ( B( I, J ), J = 1, N/2 ), I = 1, N/2 )
READ( NIN, FMT = * ) ( ( FG( I, J ), J = 1, N/2+1 ), I = 1, N/2 )
IF( LSAME( COMPU, 'U' ) ) THEN
READ( NIN, FMT = * ) ( ( U1( I, J ), J = 1, N/2 ), I = 1, N/2 )
READ( NIN, FMT = * ) ( ( U2( I, J ), J = 1, N/2 ), I = 1, N/2 )
END IF
IF( N.LT.0 .OR. N.GT.NMAX .OR. MOD( N, 2 ).NE.0 ) THEN
WRITE( NOUT, FMT = 99998 ) N
ELSE
*
* Test of MB04ED.
*
CALL MB04ED( JOB, COMPQ, COMPU, N, Z, LDZ, B, LDB, FG, LDFG, Q,
$ LDQ, U1, LDU1, U2, LDU2, ALPHAR, ALPHAI, BETA,
$ IWORK, LIWORK, DWORK, LDWORK, INFO )
IF( INFO.NE.0 ) THEN
WRITE( NOUT, FMT = 99997 ) INFO
ELSE
WRITE( NOUT, FMT = 99996 )
DO 10 I = 1, N
WRITE( NOUT, FMT = 99995 ) ( Z(I,J), J = 1, N )
10 CONTINUE
WRITE( NOUT, FMT = 99994 )
DO 20 I = 1, N/2
WRITE( NOUT, FMT = 99995 ) ( B(I,J), J = 1, N/2 )
20 CONTINUE
WRITE( NOUT, FMT = 99993 )
DO 30 I = 1, N/2
WRITE( NOUT, FMT = 99995 ) ( FG(I,J), J = 1, N/2+1 )
30 CONTINUE
WRITE( NOUT, FMT = 99992 )
WRITE( NOUT, FMT = 99995 ) ( ALPHAR(I), I = 1, N/2 )
WRITE( NOUT, FMT = 99995 ) ( ALPHAI(I), I = 1, N/2 )
WRITE( NOUT, FMT = 99995 ) ( BETA(I), I = 1, N/2 )
WRITE( NOUT, FMT = 99991 )
DO 40 I = 1, N
WRITE( NOUT, FMT = 99995 ) ( Q( I, J ), J = 1, N )
40 CONTINUE
IF ( .NOT.LSAME( COMPU, 'N' ) ) THEN
WRITE( NOUT, FMT = 99990 )
DO 50 I = 1, N/2
WRITE( NOUT, FMT = 99995 ) ( U1( I, J ), J = 1, N/2 )
50 CONTINUE
WRITE( NOUT, FMT = 99989 )
DO 60 I = 1, N/2
WRITE( NOUT, FMT = 99995 ) ( U2( I, J ), J = 1, N/2 )
60 CONTINUE
END IF
END IF
END IF
STOP
99999 FORMAT ( 'MB04ED EXAMPLE PROGRAM RESULTS', 1X )
99998 FORMAT ( 'N is out of range.', /, 'N = ', I5 )
99997 FORMAT ( 'INFO on exit from MB04ED = ', I2 )
99996 FORMAT (/' The transformed matrix Z is' )
99995 FORMAT ( 60( 1X, F8.4 ) )
99994 FORMAT (/' The transformed matrix B is' )
99993 FORMAT (/' The transformed matrix FG is' )
99992 FORMAT (/' The real, imaginary, and beta parts of eigenvalues are'
$ )
99991 FORMAT (/' The matrix Q is ' )
99990 FORMAT (/' The upper left block of the matrix U is ' )
99989 FORMAT (/' The upper right block of the matrix U is ' )
END
Program Data
MB04ED EXAMPLE PROGRAM DATA T I I 8 0.0949 3.3613 -4.7663 -0.5534 0.6408 -3.2793 3.4253 2.9654 0.1138 -1.5903 2.1837 -4.1648 -4.3775 -1.7454 0.1744 2.3262 2.7505 4.4048 4.4183 3.0478 2.7728 2.3048 -0.6451 -1.2045 3.6091 -4.1716 3.4461 3.6880 -0.0985 3.8458 0.2528 -1.3859 0.4352 -3.2829 3.7246 0.4794 -0.3690 -1.5562 -3.4817 -2.2902 1.3080 -3.9881 -3.5497 3.5020 2.2582 4.4764 -4.4080 -1.6818 1.1308 -1.5087 2.4730 2.1553 -1.7129 -4.8669 -2.4102 4.2274 4.7933 -4.3671 -0.0473 -2.0092 1.2439 -4.7385 3.4242 -0.2764 2.0936 1.5510 4.5974 2.5127 2.5469 -3.3739 -1.5961 -2.4490 -2.2397 -3.8100 0.8527 0.0596 1.7970 -0.0164 -2.7619 1.9908 1.0000 2.0000 -4.0500 1.3353 0.2899 -0.4318 2.0000 2.0000 -2.9860 -0.0160 1.0241 0.9469 2.0000 2.0000 1.3303 0.0946 -0.1272 -4.4003 2.0000 2.0000Program Results
MB04ED EXAMPLE PROGRAM RESULTS The transformed matrix Z is -2.5678 -2.9888 0.4304 -2.8719 2.7331 1.3072 1.7565 2.8246 0.0000 -3.8520 -6.0992 6.2935 -3.0386 -5.5317 -1.2189 3.9973 0.0000 0.0000 4.4560 4.4602 0.6080 -4.4326 3.7959 -0.6297 0.0000 0.0000 0.0000 7.0155 1.5557 2.1441 3.6649 -2.3864 0.4352 -3.2829 3.7246 0.4794 -5.3205 0.0000 0.0000 0.0000 1.3080 -3.9881 -3.5497 3.5020 2.2466 6.9633 0.0000 0.0000 1.1308 -1.5087 2.4730 2.1553 -1.7204 -0.8164 8.1468 0.0000 4.7933 -4.3671 -0.0473 -2.0092 -3.9547 0.2664 1.0382 5.5977 The transformed matrix B is 3.8629 -1.3266 0.1253 2.1882 0.0000 3.7258 -3.5913 -2.4583 0.0000 -3.6551 -2.5063 -0.8378 0.0000 0.0000 0.0000 -6.7384 The transformed matrix FG is 1.0000 2.0000 -0.7448 -1.2359 -1.3653 0.0158 2.0000 2.0000 3.4030 3.2344 -1.1665 2.5791 2.0000 2.0000 -0.4096 3.3823 -1.2344 3.9016 2.0000 2.0000 The real, imaginary, and beta parts of eigenvalues are 1.1310 -0.0697 -0.0697 -0.6864 0.0000 0.6035 -0.6035 0.0000 4.0000 4.0000 4.0000 4.0000 The matrix Q is -0.6042 -0.4139 -0.4742 0.1400 -0.2947 0.3462 -0.0980 0.0534 -0.3706 0.1367 0.4442 -0.1381 -0.1210 0.2913 0.7248 -0.0524 0.1325 -0.2735 -0.0515 -0.5084 -0.3163 -0.2855 0.1638 0.6619 0.2373 0.5514 -0.4988 0.3373 -0.3852 -0.0007 0.3329 0.1339 0.4777 -0.4517 0.2739 0.5172 -0.0775 0.3874 0.1088 0.2395 -0.0116 -0.4372 -0.1843 0.2474 0.1236 -0.6052 0.4772 -0.3228 0.1237 -0.0310 -0.4300 -0.2090 0.7209 0.3408 0.2898 0.1883 0.4245 -0.1871 -0.1803 -0.4655 -0.3304 0.2849 0.0623 -0.5843 The upper left block of the matrix U is 0.0154 -0.5058 -0.5272 0.6826 0.4829 -0.1519 -0.2921 -0.3491 0.4981 0.1532 0.1019 0.1810 -0.0188 0.6270 -0.5260 0.0587 The upper right block of the matrix U is 0.0000 0.0000 0.0000 0.0000 0.3179 -0.4312 -0.1802 -0.4659 0.5644 0.0873 0.4480 0.3979 -0.3137 -0.3330 0.3413 0.0239