SUBROUTINE SB03OU( DISCR, LTRANS, N, M, A, LDA, B, LDB, TAU, U,
$ LDU, SCALE, DWORK, LDWORK, INFO )
C
C SLICOT RELEASE 5.0.
C
C Copyright (c) 2002-2009 NICONET e.V.
C
C This program is free software: you can redistribute it and/or
C modify it under the terms of the GNU General Public License as
C published by the Free Software Foundation, either version 2 of
C the License, or (at your option) any later version.
C
C This program is distributed in the hope that it will be useful,
C but WITHOUT ANY WARRANTY; without even the implied warranty of
C MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
C GNU General Public License for more details.
C
C You should have received a copy of the GNU General Public License
C along with this program. If not, see
C .
C
C PURPOSE
C
C To solve for X = op(U)'*op(U) either the stable non-negative
C definite continuous-time Lyapunov equation
C 2
C op(A)'*X + X*op(A) = -scale *op(B)'*op(B) (1)
C
C or the convergent non-negative definite discrete-time Lyapunov
C equation
C 2
C op(A)'*X*op(A) - X = -scale *op(B)'*op(B) (2)
C
C where op(K) = K or K' (i.e., the transpose of the matrix K), A is
C an N-by-N matrix in real Schur form, op(B) is an M-by-N matrix,
C U is an upper triangular matrix containing the Cholesky factor of
C the solution matrix X, X = op(U)'*op(U), and scale is an output
C scale factor, set less than or equal to 1 to avoid overflow in X.
C If matrix B has full rank then the solution matrix X will be
C positive-definite and hence the Cholesky factor U will be
C nonsingular, but if B is rank deficient then X may only be
C positive semi-definite and U will be singular.
C
C In the case of equation (1) the matrix A must be stable (that
C is, all the eigenvalues of A must have negative real parts),
C and for equation (2) the matrix A must be convergent (that is,
C all the eigenvalues of A must lie inside the unit circle).
C
C ARGUMENTS
C
C Mode Parameters
C
C DISCR LOGICAL
C Specifies the type of Lyapunov equation to be solved as
C follows:
C = .TRUE. : Equation (2), discrete-time case;
C = .FALSE.: Equation (1), continuous-time case.
C
C LTRANS LOGICAL
C Specifies the form of op(K) to be used, as follows:
C = .FALSE.: op(K) = K (No transpose);
C = .TRUE. : op(K) = K**T (Transpose).
C
C Input/Output Parameters
C
C N (input) INTEGER
C The order of the matrix A and the number of columns in
C matrix op(B). N >= 0.
C
C M (input) INTEGER
C The number of rows in matrix op(B). M >= 0.
C
C A (input) DOUBLE PRECISION array, dimension (LDA,N)
C The leading N-by-N upper Hessenberg part of this array
C must contain a real Schur form matrix S. The elements
C below the upper Hessenberg part of the array A are not
C referenced. The 2-by-2 blocks must only correspond to
C complex conjugate pairs of eigenvalues (not to real
C eigenvalues).
C
C LDA INTEGER
C The leading dimension of array A. LDA >= MAX(1,N).
C
C B (input/output) DOUBLE PRECISION array, dimension (LDB,N)
C if LTRANS = .FALSE., and dimension (LDB,M), if
C LTRANS = .TRUE..
C On entry, if LTRANS = .FALSE., the leading M-by-N part of
C this array must contain the coefficient matrix B of the
C equation.
C On entry, if LTRANS = .TRUE., the leading N-by-M part of
C this array must contain the coefficient matrix B of the
C equation.
C On exit, if LTRANS = .FALSE., the leading
C MIN(M,N)-by-MIN(M,N) upper triangular part of this array
C contains the upper triangular matrix R (as defined in
C METHOD), and the M-by-MIN(M,N) strictly lower triangular
C part together with the elements of the array TAU are
C overwritten by details of the matrix P (also defined in
C METHOD). When M < N, columns (M+1),...,N of the array B
C are overwritten by the matrix Z (see METHOD).
C On exit, if LTRANS = .TRUE., the leading
C MIN(M,N)-by-MIN(M,N) upper triangular part of
C B(1:N,M-N+1), if M >= N, or of B(N-M+1:N,1:M), if M < N,
C contains the upper triangular matrix R (as defined in
C METHOD), and the remaining elements (below the diagonal
C of R) together with the elements of the array TAU are
C overwritten by details of the matrix P (also defined in
C METHOD). When M < N, rows 1,...,(N-M) of the array B
C are overwritten by the matrix Z (see METHOD).
C
C LDB INTEGER
C The leading dimension of array B.
C LDB >= MAX(1,M), if LTRANS = .FALSE.,
C LDB >= MAX(1,N), if LTRANS = .TRUE..
C
C TAU (output) DOUBLE PRECISION array of dimension (MIN(N,M))
C This array contains the scalar factors of the elementary
C reflectors defining the matrix P.
C
C U (output) DOUBLE PRECISION array of dimension (LDU,N)
C The leading N-by-N upper triangular part of this array
C contains the Cholesky factor of the solution matrix X of
C the problem, X = op(U)'*op(U).
C The array U may be identified with B in the calling
C statement, if B is properly dimensioned, and the
C intermediate results returned in B are not needed.
C
C LDU INTEGER
C The leading dimension of array U. LDU >= MAX(1,N).
C
C SCALE (output) DOUBLE PRECISION
C The scale factor, scale, set less than or equal to 1 to
C prevent the solution overflowing.
C
C Workspace
C
C DWORK DOUBLE PRECISION array, dimension (LDWORK)
C On exit, if INFO = 0, or INFO = 1, DWORK(1) returns the
C optimal value of LDWORK.
C
C LDWORK INTEGER
C The length of the array DWORK. LDWORK >= MAX(1,4*N).
C For optimum performance LDWORK should sometimes be larger.
C
C Error Indicator
C
C INFO INTEGER
C = 0: successful exit;
C < 0: if INFO = -i, the i-th argument had an illegal
C value;
C = 1: if the Lyapunov equation is (nearly) singular
C (warning indicator);
C if DISCR = .FALSE., this means that while the matrix
C A has computed eigenvalues with negative real parts,
C it is only just stable in the sense that small
C perturbations in A can make one or more of the
C eigenvalues have a non-negative real part;
C if DISCR = .TRUE., this means that while the matrix
C A has computed eigenvalues inside the unit circle,
C it is nevertheless only just convergent, in the
C sense that small perturbations in A can make one or
C more of the eigenvalues lie outside the unit circle;
C perturbed values were used to solve the equation
C (but the matrix A is unchanged);
C = 2: if matrix A is not stable (that is, one or more of
C the eigenvalues of A has a non-negative real part),
C if DISCR = .FALSE., or not convergent (that is, one
C or more of the eigenvalues of A lies outside the
C unit circle), if DISCR = .TRUE.;
C = 3: if matrix A has two or more consecutive non-zero
C elements on the first sub-diagonal, so that there is
C a block larger than 2-by-2 on the diagonal;
C = 4: if matrix A has a 2-by-2 diagonal block with real
C eigenvalues instead of a complex conjugate pair.
C
C METHOD
C
C The method used by the routine is based on the Bartels and
C Stewart method [1], except that it finds the upper triangular
C matrix U directly without first finding X and without the need
C to form the normal matrix op(B)'*op(B) [2].
C
C If LTRANS = .FALSE., the matrix B is factored as
C
C B = P ( R ), M >= N, B = P ( R Z ), M < N,
C ( 0 )
C
C (QR factorization), where P is an M-by-M orthogonal matrix and
C R is a square upper triangular matrix.
C
C If LTRANS = .TRUE., the matrix B is factored as
C
C B = ( 0 R ) P, M >= N, B = ( Z ) P, M < N,
C ( R )
C
C (RQ factorization), where P is an M-by-M orthogonal matrix and
C R is a square upper triangular matrix.
C
C These factorizations are used to solve the continuous-time
C Lyapunov equation in the canonical form
C 2
C op(A)'*op(U)'*op(U) + op(U)'*op(U)*op(A) = -scale *op(F)'*op(F),
C
C or the discrete-time Lyapunov equation in the canonical form
C 2
C op(A)'*op(U)'*op(U)*op(A) - op(U)'*op(U) = -scale *op(F)'*op(F),
C
C where U and F are N-by-N upper triangular matrices, and
C
C F = R, if M >= N, or
C
C F = ( R ), if LTRANS = .FALSE., or
C ( 0 )
C
C F = ( 0 Z ), if LTRANS = .TRUE., if M < N.
C ( 0 R )
C
C The canonical equation is solved for U.
C
C REFERENCES
C
C [1] Bartels, R.H. and Stewart, G.W.
C Solution of the matrix equation A'X + XB = C.
C Comm. A.C.M., 15, pp. 820-826, 1972.
C
C [2] Hammarling, S.J.
C Numerical solution of the stable, non-negative definite
C Lyapunov equation.
C IMA J. Num. Anal., 2, pp. 303-325, 1982.
C
C NUMERICAL ASPECTS
C 3
C The algorithm requires 0(N ) operations and is backward stable.
C
C FURTHER COMMENTS
C
C The Lyapunov equation may be very ill-conditioned. In particular,
C if A is only just stable (or convergent) then the Lyapunov
C equation will be ill-conditioned. "Large" elements in U relative
C to those of A and B, or a "small" value for scale, are symptoms
C of ill-conditioning. A condition estimate can be computed using
C SLICOT Library routine SB03MD.
C
C CONTRIBUTOR
C
C Release 3.0: V. Sima, Katholieke Univ. Leuven, Belgium, May 1997.
C Supersedes Release 2.0 routine SB03CZ by Sven Hammarling,
C NAG Ltd, United Kingdom.
C Partly based on routine PLYAPS by A. Varga, University of Bochum,
C May 1992.
C
C REVISIONS
C
C Dec. 1997, April 1998, May 1999.
C
C KEYWORDS
C
C Lyapunov equation, orthogonal transformation, real Schur form.
C
C ******************************************************************
C
C .. Parameters ..
DOUBLE PRECISION ZERO, ONE
PARAMETER ( ZERO = 0.0D0, ONE = 1.0D0 )
C .. Scalar Arguments ..
LOGICAL DISCR, LTRANS
INTEGER INFO, LDA, LDB, LDU, LDWORK, M, N
DOUBLE PRECISION SCALE
C .. Array Arguments ..
DOUBLE PRECISION A(LDA,*), B(LDB,*), DWORK(*), TAU(*), U(LDU,*)
C .. Local Scalars ..
INTEGER I, J, K, L, MN, WRKOPT
C .. External Subroutines ..
EXTERNAL DCOPY, DGEQRF, DGERQF, DLACPY, DLASET, SB03OT,
$ XERBLA
C .. Intrinsic Functions ..
INTRINSIC MAX, MIN
C .. Executable Statements ..
C
INFO = 0
C
C Test the input scalar arguments.
C
IF( N.LT.0 ) THEN
INFO = -3
ELSE IF( M.LT.0 ) THEN
INFO = -4
ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
INFO = -6
ELSE IF( ( LDB.LT.MAX( 1, M ) .AND. .NOT.LTRANS ) .OR.
$ ( LDB.LT.MAX( 1, N ) .AND. LTRANS ) ) THEN
INFO = -8
ELSE IF( LDU.LT.MAX( 1, N ) ) THEN
INFO = -11
ELSE IF( LDWORK.LT.MAX( 1, 4*N ) ) THEN
INFO = -14
END IF
C
IF ( INFO.NE.0 ) THEN
C
C Error return.
C
CALL XERBLA( 'SB03OU', -INFO )
RETURN
END IF
C
C Quick return if possible.
C
MN = MIN( N, M )
IF ( MN.EQ.0 ) THEN
SCALE = ONE
DWORK(1) = ONE
RETURN
END IF
C
C (Note: Comments in the code beginning "Workspace:" describe the
C minimal amount of real workspace needed at that point in the
C code, as well as the preferred amount for good performance.
C NB refers to the optimal block size for the immediately
C following subroutine, as returned by ILAENV.)
C
IF ( LTRANS ) THEN
C
C Case op(K) = K'.
C
C Perform the RQ factorization of B.
C Workspace: need N;
C prefer N*NB.
C
CALL DGERQF( N, M, B, LDB, TAU, DWORK, LDWORK, INFO )
C
C The triangular matrix F is constructed in the array U so that
C U can share the same memory as B.
C
IF ( M.GE.N ) THEN
CALL DLACPY( 'Upper', MN, N, B(1,M-N+1), LDB, U, LDU )
ELSE
C
DO 10 I = M, 1, -1
CALL DCOPY( N-M+I, B(1,I), 1, U(1,N-M+I), 1 )
10 CONTINUE
C
CALL DLASET( 'Full', N, N-M, ZERO, ZERO, U, LDU )
END IF
ELSE
C
C Case op(K) = K.
C
C Perform the QR factorization of B.
C Workspace: need N;
C prefer N*NB.
C
CALL DGEQRF( M, N, B, LDB, TAU, DWORK, LDWORK, INFO )
CALL DLACPY( 'Upper', MN, N, B, LDB, U, LDU )
IF ( M.LT.N )
$ CALL DLASET( 'Upper', N-M, N-M, ZERO, ZERO, U(M+1,M+1),
$ LDU )
END IF
WRKOPT = DWORK(1)
C
C Solve the canonical Lyapunov equation
C 2
C op(A)'*op(U)'*op(U) + op(U)'*op(U)*op(A) = -scale *op(F)'*op(F),
C
C or
C 2
C op(A)'*op(U)'*op(U)*op(A) - op(U)'*op(U) = -scale *op(F)'*op(F)
C
C for U.
C
CALL SB03OT( DISCR, LTRANS, N, A, LDA, U, LDU, SCALE, DWORK,
$ INFO )
IF ( INFO.NE.0 .AND. INFO.NE.1 )
$ RETURN
C
C Make the diagonal elements of U non-negative.
C
IF ( LTRANS ) THEN
C
DO 30 J = 1, N
IF ( U(J,J).LT.ZERO ) THEN
C
DO 20 I = 1, J
U(I,J) = -U(I,J)
20 CONTINUE
C
END IF
30 CONTINUE
C
ELSE
K = 1
C
DO 50 J = 1, N
DWORK(K) = U(J,J)
L = 1
C
DO 40 I = 1, J
IF ( DWORK(L).LT.ZERO ) U(I,J) = -U(I,J)
L = L + 1
40 CONTINUE
C
K = K + 1
50 CONTINUE
C
END IF
C
DWORK(1) = MAX( WRKOPT, 4*N )
RETURN
C *** Last line of SB03OU ***
END