Purpose
To compute beta(A), the 2-norm distance from a real matrix A to the nearest complex matrix with an eigenvalue on the imaginary axis. If A is stable in the sense that all eigenvalues of A lie in the open left half complex plane, then beta(A) is the complex stability radius, i.e., the distance to the nearest unstable complex matrix. The value of beta(A) is the minimum of the smallest singular value of (A - jwI), taken over all real w. The value of w corresponding to the minimum is also computed.Specification
SUBROUTINE AB13FD( N, A, LDA, BETA, OMEGA, TOL, DWORK, LDWORK,
$ CWORK, LCWORK, INFO )
C .. Scalar Arguments ..
INTEGER INFO, LCWORK, LDA, LDWORK, N
DOUBLE PRECISION BETA, OMEGA, TOL
C .. Array Arguments ..
DOUBLE PRECISION A(LDA,*), DWORK(*)
COMPLEX*16 CWORK(*)
Arguments
Input/Output Parameters
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input) DOUBLE PRECISION array, dimension (LDA,N)
The leading N-by-N part of this array must contain the
matrix A.
LDA INTEGER
The leading dimension of array A. LDA >= MAX(1,N).
BETA (output) DOUBLE PRECISION
The computed value of beta(A), which actually is an upper
bound.
OMEGA (output) DOUBLE PRECISION
The value of w such that the smallest singular value of
(A - jwI) equals beta(A).
Tolerances
TOL DOUBLE PRECISION
Specifies the accuracy with which beta(A) is to be
calculated. (See the Numerical Aspects section below.)
If the user sets TOL to be less than EPS, where EPS is the
machine precision (see LAPACK Library Routine DLAMCH),
then the tolerance is taken to be EPS.
Workspace
DWORK DOUBLE PRECISION array, dimension (LDWORK)
On exit, if INFO = 0, DWORK(1) returns the optimal value
of LDWORK.
If DWORK(1) is not needed, the first 2*N*N entries of
DWORK may overlay CWORK.
LDWORK INTEGER
The length of the array DWORK.
LDWORK >= MAX( 1, 3*N*(N+2) ).
For optimum performance LDWORK should be larger.
CWORK COMPLEX*16 array, dimension (LCWORK)
On exit, if INFO = 0, CWORK(1) returns the optimal value
of LCWORK.
If CWORK(1) is not needed, the first N*N entries of
CWORK may overlay DWORK.
LCWORK INTEGER
The length of the array CWORK.
LCWORK >= MAX( 1, N*(N+3) ).
For optimum performance LCWORK should be larger.
Error Indicator
INFO INTEGER
= 0: successful exit;
< 0: if INFO = -i, the i-th argument had an illegal
value;
= 1: the routine fails to compute beta(A) within the
specified tolerance. Nevertheless, the returned
value is an upper bound on beta(A);
= 2: either the QR or SVD algorithm (LAPACK Library
routines DHSEQR, DGESVD or ZGESVD) fails to
converge; this error is very rare.
Method
AB13FD combines the methods of [1] and [2] into a provably reliable, quadratically convergent algorithm. It uses the simple bisection strategy of [1] to find an interval which contains beta(A), and then switches to the modified bisection strategy of [2] which converges quadratically to a minimizer. Note that the efficiency of the strategy degrades if there are several local minima that are near or equal the global minimum.References
[1] Byers, R.
A bisection method for measuring the distance of a stable
matrix to the unstable matrices.
SIAM J. Sci. Stat. Comput., Vol. 9, No. 5, pp. 875-880, 1988.
[2] Boyd, S. and Balakrishnan, K.
A regularity result for the singular values of a transfer
matrix and a quadratically convergent algorithm for computing
its L-infinity norm.
Systems and Control Letters, Vol. 15, pp. 1-7, 1990.
Numerical Aspects
In the presence of rounding errors, the computed function value
BETA satisfies
beta(A) <= BETA + epsilon,
BETA/(1+TOL) - delta <= MAX(beta(A), SQRT(2*N*EPS)*norm(A)),
where norm(A) is the Frobenius norm of A,
epsilon = p(N) * EPS * norm(A),
and
delta = p(N) * SQRT(EPS) * norm(A),
and p(N) is a low degree polynomial. It is recommended to choose
TOL greater than SQRT(EPS). Although rounding errors can cause
AB13FD to fail for smaller values of TOL, nevertheless, it usually
succeeds. Regardless of success or failure, the first inequality
holds.
Further Comments
NoneExample
Program Text
* AB13FD EXAMPLE PROGRAM TEXT
*
* .. Parameters ..
INTEGER NIN, NOUT
PARAMETER ( NIN = 5, NOUT = 6 )
INTEGER NMAX
PARAMETER ( NMAX = 20 )
INTEGER LDA
PARAMETER ( LDA = NMAX )
INTEGER LDWORK
PARAMETER ( LDWORK = 3*NMAX*( NMAX + 2 ) )
INTEGER LCWORK
PARAMETER ( LCWORK = NMAX*( NMAX + 3 ) )
* .. Local Scalars ..
INTEGER I, INFO, J, N
DOUBLE PRECISION BETA, OMEGA, TOL
* .. Local Arrays ..
DOUBLE PRECISION A(LDA,NMAX), DWORK(LDWORK)
COMPLEX*16 CWORK(LCWORK)
* .. External Subroutines ..
EXTERNAL AB13FD, UD01MD
* ..
* .. Executable Statements ..
*
WRITE ( NOUT, FMT = 99999 )
* Skip the heading in the data file and read the data.
READ ( NIN, FMT = '()' )
* Read N, TOL and next A (row wise).
READ ( NIN, FMT = * ) N, TOL
IF ( N.LT.0 .OR. N.GT.NMAX ) THEN
WRITE ( NOUT, FMT = 99995 ) N
ELSE
DO 10 I = 1, N
READ ( NIN, FMT = * ) ( A(I,J), J = 1, N )
10 CONTINUE
*
WRITE ( NOUT, FMT = 99998 ) N, TOL
CALL UD01MD( N, N, 5, NOUT, A, LDA, 'A', INFO )
*
CALL AB13FD( N, A, LDA, BETA, OMEGA, TOL, DWORK, LDWORK, CWORK,
$ LCWORK, INFO )
*
IF ( INFO.NE.0 )
$ WRITE ( NOUT, FMT = 99996 ) INFO
WRITE ( NOUT, FMT = 99997 ) BETA, OMEGA
END IF
*
99999 FORMAT (' AB13FD EXAMPLE PROGRAM RESULTS', /1X)
99998 FORMAT (' N =', I2, 3X, 'TOL =', D10.3)
99997 FORMAT (' Stability radius :', D18.11, /
* ' Minimizing omega :', D18.11)
99996 FORMAT (' INFO on exit from AB13FD = ', I2)
99995 FORMAT (/' N is out of range.',/' N = ',I5)
END
Program Data
AB13FD EXAMPLE PROGRAM DATA
4 0.0D-00 0.0D-00
246.500 242.500 202.500 -197.500
-252.500 -248.500 -207.500 202.500
-302.500 -297.500 -248.500 242.500
-307.500 -302.500 -252.500 246.500
Program Results
AB13FD EXAMPLE PROGRAM RESULTS
N = 4 TOL = 0.000D+00
A ( 4X 4)
1 2 3 4
1 0.2465000D+03 0.2425000D+03 0.2025000D+03 -0.1975000D+03
2 -0.2525000D+03 -0.2485000D+03 -0.2075000D+03 0.2025000D+03
3 -0.3025000D+03 -0.2975000D+03 -0.2485000D+03 0.2425000D+03
4 -0.3075000D+03 -0.3025000D+03 -0.2525000D+03 0.2465000D+03
Stability radius : 0.39196472317D-02
Minimizing omega : 0.98966520430D+00