Purpose
To solve for X = op(U)'*op(U) either the stable non-negative definite continuous-time Lyapunov equation 2 op(A)'*X + X*op(A) = -scale *op(B)'*op(B) (1) or the convergent non-negative definite discrete-time Lyapunov equation 2 op(A)'*X*op(A) - X = -scale *op(B)'*op(B) (2) where op(K) = K or K' (i.e., the transpose of the matrix K), A is an N-by-N matrix, op(B) is an M-by-N matrix, U is an upper triangular matrix containing the Cholesky factor of the solution matrix X, X = op(U)'*op(U), and scale is an output scale factor, set less than or equal to 1 to avoid overflow in X. If matrix B has full rank then the solution matrix X will be positive-definite and hence the Cholesky factor U will be nonsingular, but if B is rank deficient then X may be only positive semi-definite and U will be singular. In the case of equation (1) the matrix A must be stable (that is, all the eigenvalues of A must have negative real parts), and for equation (2) the matrix A must be convergent (that is, all the eigenvalues of A must lie inside the unit circle).Specification
SUBROUTINE SB03OD( DICO, FACT, TRANS, N, M, A, LDA, Q, LDQ, B, $ LDB, SCALE, WR, WI, DWORK, LDWORK, INFO ) C .. Scalar Arguments .. CHARACTER DICO, FACT, TRANS INTEGER INFO, LDA, LDB, LDQ, LDWORK, M, N DOUBLE PRECISION SCALE C .. Array Arguments .. DOUBLE PRECISION A(LDA,*), B(LDB,*), DWORK(*), Q(LDQ,*), WI(*), $ WR(*)Arguments
Mode Parameters
DICO CHARACTER*1 Specifies the type of Lyapunov equation to be solved as follows: = 'C': Equation (1), continuous-time case; = 'D': Equation (2), discrete-time case. FACT CHARACTER*1 Specifies whether or not the real Schur factorization of the matrix A is supplied on entry, as follows: = 'F': On entry, A and Q contain the factors from the real Schur factorization of the matrix A; = 'N': The Schur factorization of A will be computed and the factors will be stored in A and Q. TRANS CHARACTER*1 Specifies the form of op(K) to be used, as follows: = 'N': op(K) = K (No transpose); = 'T': op(K) = K**T (Transpose).Input/Output Parameters
N (input) INTEGER The order of the matrix A and the number of columns in matrix op(B). N >= 0. M (input) INTEGER The number of rows in matrix op(B). M >= 0. A (input/output) DOUBLE PRECISION array, dimension (LDA,N) On entry, the leading N-by-N part of this array must contain the matrix A. If FACT = 'F', then A contains an upper quasi-triangular matrix S in Schur canonical form; the elements below the upper Hessenberg part of the array A are not referenced. On exit, the leading N-by-N upper Hessenberg part of this array contains the upper quasi-triangular matrix S in Schur canonical form from the Shur factorization of A. The contents of array A is not modified if FACT = 'F'. LDA INTEGER The leading dimension of array A. LDA >= MAX(1,N). Q (input or output) DOUBLE PRECISION array, dimension (LDQ,N) On entry, if FACT = 'F', then the leading N-by-N part of this array must contain the orthogonal matrix Q of the Schur factorization of A. Otherwise, Q need not be set on entry. On exit, the leading N-by-N part of this array contains the orthogonal matrix Q of the Schur factorization of A. The contents of array Q is not modified if FACT = 'F'. LDQ INTEGER The leading dimension of array Q. LDQ >= MAX(1,N). B (input/output) DOUBLE PRECISION array, dimension (LDB,N) if TRANS = 'N', and dimension (LDB,max(M,N)), if TRANS = 'T'. On entry, if TRANS = 'N', the leading M-by-N part of this array must contain the coefficient matrix B of the equation. On entry, if TRANS = 'T', the leading N-by-M part of this array must contain the coefficient matrix B of the equation. On exit, the leading N-by-N part of this array contains the upper triangular Cholesky factor U of the solution matrix X of the problem, X = op(U)'*op(U). If M = 0 and N > 0, then U is set to zero. LDB INTEGER The leading dimension of array B. LDB >= MAX(1,N,M), if TRANS = 'N'; LDB >= MAX(1,N), if TRANS = 'T'. SCALE (output) DOUBLE PRECISION The scale factor, scale, set less than or equal to 1 to prevent the solution overflowing. WR (output) DOUBLE PRECISION array, dimension (N) WI (output) DOUBLE PRECISION array, dimension (N) If FACT = 'N', and INFO >= 0 and INFO <= 2, WR and WI contain the real and imaginary parts, respectively, of the eigenvalues of A. If FACT = 'F', WR and WI are not referenced.Workspace
DWORK DOUBLE PRECISION array, dimension (LDWORK) On exit, if INFO = 0, or INFO = 1, DWORK(1) returns the optimal value of LDWORK. LDWORK INTEGER The length of the array DWORK. If M > 0, LDWORK >= MAX(1,4*N + MIN(M,N)); If M = 0, LDWORK >= 1. For optimum performance LDWORK should sometimes 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: successful exit; < 0: if INFO = -i, the i-th argument had an illegal value; = 1: if the Lyapunov equation is (nearly) singular (warning indicator); if DICO = 'C' this means that while the matrix A (or the factor S) has computed eigenvalues with negative real parts, it is only just stable in the sense that small perturbations in A can make one or more of the eigenvalues have a non-negative real part; if DICO = 'D' this means that while the matrix A (or the factor S) has computed eigenvalues inside the unit circle, it is nevertheless only just convergent, in the sense that small perturbations in A can make one or more of the eigenvalues lie outside the unit circle; perturbed values were used to solve the equation; = 2: if FACT = 'N' and DICO = 'C', but the matrix A is not stable (that is, one or more of the eigenvalues of A has a non-negative real part), or DICO = 'D', but the matrix A is not convergent (that is, one or more of the eigenvalues of A lies outside the unit circle); however, A will still have been factored and the eigenvalues of A returned in WR and WI. = 3: if FACT = 'F' and DICO = 'C', but the Schur factor S supplied in the array A is not stable (that is, one or more of the eigenvalues of S has a non-negative real part), or DICO = 'D', but the Schur factor S supplied in the array A is not convergent (that is, one or more of the eigenvalues of S lies outside the unit circle); = 4: if FACT = 'F' and the Schur factor S supplied in the array A has two or more consecutive non-zero elements on the first sub-diagonal, so that there is a block larger than 2-by-2 on the diagonal; = 5: if FACT = 'F' and the Schur factor S supplied in the array A has a 2-by-2 diagonal block with real eigenvalues instead of a complex conjugate pair; = 6: if FACT = 'N' and the LAPACK Library routine DGEES has failed to converge. This failure is not likely to occur. The matrix B will be unaltered but A will be destroyed.Method
The method used by the routine is based on the Bartels and Stewart method [1], except that it finds the upper triangular matrix U directly without first finding X and without the need to form the normal matrix op(B)'*op(B). The Schur factorization of a square matrix A is given by A = QSQ', where Q is orthogonal and S is an N-by-N block upper triangular matrix with 1-by-1 and 2-by-2 blocks on its diagonal (which correspond to the eigenvalues of A). If A has already been factored prior to calling the routine however, then the factors Q and S may be supplied and the initial factorization omitted. If TRANS = 'N', the matrix B is factored as (QR factorization) _ _ _ _ _ B = P ( R ), M >= N, B = P ( R Z ), M < N, ( 0 ) _ _ where P is an M-by-M orthogonal matrix and R is a square upper _ _ _ _ _ triangular matrix. Then, the matrix B = RQ, or B = ( R Z )Q (if M < N) is factored as _ _ B = P ( R ), M >= N, B = P ( R Z ), M < N. If TRANS = 'T', the matrix B is factored as (RQ factorization) _ _ _ ( Z ) _ B = ( 0 R ) P, M >= N, B = ( _ ) P, M < N, ( R ) _ _ where P is an M-by-M orthogonal matrix and R is a square upper _ _ _ _ _ triangular matrix. Then, the matrix B = Q'R, or B = Q'( Z' R' )' (if M < N) is factored as _ _ B = ( R ) P, M >= N, B = ( Z ) P, M < N. ( R ) These factorizations are utilised to either transform the continuous-time Lyapunov equation to the canonical form 2 op(S)'*op(V)'*op(V) + op(V)'*op(V)*op(S) = -scale *op(F)'*op(F), or the discrete-time Lyapunov equation to the canonical form 2 op(S)'*op(V)'*op(V)*op(S) - op(V)'*op(V) = -scale *op(F)'*op(F), where V and F are upper triangular, and F = R, M >= N, F = ( R Z ), M < N, if TRANS = 'N'; ( 0 0 ) F = R, M >= N, F = ( 0 Z ), M < N, if TRANS = 'T'. ( 0 R ) The transformed equation is then solved for V, from which U is obtained via the QR factorization of V*Q', if TRANS = 'N', or via the RQ factorization of Q*V, if TRANS = 'T'.References
[1] Bartels, R.H. and Stewart, G.W. Solution of the matrix equation A'X + XB = C. Comm. A.C.M., 15, pp. 820-826, 1972. [2] Hammarling, S.J. Numerical solution of the stable, non-negative definite Lyapunov equation. IMA J. Num. Anal., 2, pp. 303-325, 1982.Numerical Aspects
3 The algorithm requires 0(N ) operations and is backward stable.Further Comments
The Lyapunov equation may be very ill-conditioned. In particular, if A is only just stable (or convergent) then the Lyapunov equation will be ill-conditioned. A symptom of ill-conditioning is "large" elements in U relative to those of A and B, or a "small" value for scale. A condition estimate can be computed using SLICOT Library routine SB03MD. SB03OD routine can be also used for solving "unstable" Lyapunov equations, i.e., when matrix A has all eigenvalues with positive real parts, if DICO = 'C', or with moduli greater than one, if DICO = 'D'. Specifically, one may solve for X = op(U)'*op(U) either the continuous-time Lyapunov equation 2 op(A)'*X + X*op(A) = scale *op(B)'*op(B), (3) or the discrete-time Lyapunov equation 2 op(A)'*X*op(A) - X = scale *op(B)'*op(B), (4) provided, for equation (3), the given matrix A is replaced by -A, or, for equation (4), the given matrices A and B are replaced by inv(A) and B*inv(A), if TRANS = 'N' (or inv(A)*B, if TRANS = 'T'), respectively. Although the inversion generally can rise numerical problems, in case of equation (4) it is expected that the matrix A is enough well-conditioned, having only eigenvalues with moduli greater than 1. However, if A is ill-conditioned, it could be preferable to use the more general SLICOT Lyapunov solver SB03MD.Example
Program Text
* SB03OD EXAMPLE PROGRAM TEXT * Copyright (c) 2002-2017 NICONET e.V. * * .. Parameters .. DOUBLE PRECISION ZERO PARAMETER ( ZERO = 0.0D0 ) INTEGER NIN, NOUT PARAMETER ( NIN = 5, NOUT = 6 ) INTEGER NMAX, MMAX PARAMETER ( NMAX = 20, MMAX = 20 ) INTEGER LDA, LDB, LDQ, LDX, LDWORK PARAMETER ( LDA = NMAX, LDB = MAX( MMAX,NMAX ), $ LDQ = NMAX, LDX = NMAX ) PARAMETER ( LDWORK = 4*NMAX+MIN(MMAX,NMAX) ) * .. Local Scalars .. DOUBLE PRECISION SCALE, TEMP INTEGER I, INFO, J, K, M, N CHARACTER*1 DICO, FACT, TRANS * .. Local Arrays .. DOUBLE PRECISION A(LDA,NMAX), B(LDB,LDB), DWORK(LDWORK), $ Q(LDQ,NMAX), WR(NMAX), WI(NMAX), X(LDX,NMAX) * .. External Functions .. LOGICAL LSAME EXTERNAL LSAME * .. External Subroutines .. EXTERNAL SB03OD * .. Intrinsic Functions .. INTRINSIC MAX, MIN * .. Executable Statements .. * WRITE ( NOUT, FMT = 99999 ) * Skip the heading in the data file and read the data. READ ( NIN, FMT = '()' ) READ ( NIN, FMT = * ) N, M, DICO, FACT, TRANS IF ( N.LT.0 .OR. N.GT.NMAX ) THEN WRITE ( NOUT, FMT = 99994 ) N ELSE READ ( NIN, FMT = * ) ( ( A(I,J), J = 1,N ), I = 1,N ) IF ( LSAME( FACT, 'F' ) ) READ ( NIN, FMT = * ) $ ( ( Q(I,J), J = 1,N ), I = 1,N ) IF ( M.LT.0 .OR. M.GT.MMAX ) THEN WRITE ( NOUT, FMT = 99993 ) M ELSE IF ( LSAME( TRANS, 'N' ) ) THEN READ ( NIN, FMT = * ) ( ( B(I,J), J = 1,N ), I = 1,M ) ELSE READ ( NIN, FMT = * ) ( ( B(I,J), J = 1,M ), I = 1,N ) END IF * Find the Cholesky factor U. CALL SB03OD( DICO, FACT, TRANS, N, M, A, LDA, Q, LDQ, B, $ LDB, SCALE, WR, WI, DWORK, LDWORK, INFO ) * IF ( INFO.NE.0 ) THEN WRITE ( NOUT, FMT = 99998 ) INFO ELSE WRITE ( NOUT, FMT = 99997 ) DO 20 J = 1, N WRITE ( NOUT, FMT = 99996 ) ( B(I,J), I = 1,J ) 20 CONTINUE * Form the solution matrix X = op(U)'*op(U). IF ( LSAME( TRANS, 'N' ) ) THEN DO 80 I = 1, N DO 60 J = I, N TEMP = ZERO DO 40 K = 1, I TEMP = TEMP + B(K,I)*B(K,J) 40 CONTINUE X(I,J) = TEMP X(J,I) = TEMP 60 CONTINUE 80 CONTINUE ELSE DO 140 I = 1, N DO 120 J = I, N TEMP = ZERO DO 100 K = J, N TEMP = TEMP + B(I,K)*B(J,K) 100 CONTINUE X(I,J) = TEMP X(J,I) = TEMP 120 CONTINUE 140 CONTINUE END IF WRITE ( NOUT, FMT = 99995 ) DO 160 J = 1, N WRITE ( NOUT, FMT = 99996 ) ( X(I,J), I = 1,N ) 160 CONTINUE WRITE ( NOUT, FMT = 99992 ) SCALE END IF END IF END IF STOP * 99999 FORMAT (' SB03OD EXAMPLE PROGRAM RESULTS',/1X) 99998 FORMAT (' INFO on exit from SB03OD = ',I2) 99997 FORMAT (' The transpose of the Cholesky factor U is ') 99996 FORMAT (20(1X,F8.4)) 99995 FORMAT (/' The solution matrix X = op(U)''*op(U) is ') 99994 FORMAT (/' N is out of range.',/' N = ',I5) 99993 FORMAT (/' M is out of range.',/' M = ',I5) 99992 FORMAT (/' Scaling factor = ',F8.4) ENDProgram Data
SB03OD EXAMPLE PROGRAM DATA 4 5 C N N -1.0 37.0 -12.0 -12.0 -1.0 -10.0 0.0 4.0 2.0 -4.0 7.0 -6.0 2.0 2.0 7.0 -9.0 1.0 2.5 1.0 3.5 0.0 1.0 0.0 1.0 -1.0 -2.5 -1.0 -1.5 1.0 2.5 4.0 -5.5 -1.0 -2.5 -4.0 3.5Program Results
SB03OD EXAMPLE PROGRAM RESULTS The transpose of the Cholesky factor U is 1.0000 3.0000 1.0000 2.0000 -1.0000 1.0000 -1.0000 1.0000 -2.0000 1.0000 The solution matrix X = op(U)'*op(U) is 1.0000 3.0000 2.0000 -1.0000 3.0000 10.0000 5.0000 -2.0000 2.0000 5.0000 6.0000 -5.0000 -1.0000 -2.0000 -5.0000 7.0000 Scaling factor = 1.0000