1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
|
<HTML>
<HEAD><TITLE>SB02PD - SLICOT Library Routine Documentation</TITLE>
</HEAD>
<BODY>
<H2><A Name="SB02PD">SB02PD</A></H2>
<H3>
Solution of continuous-time algebraic Riccati equations (matrix sign function method) with error bounds and condition estimates
</H3>
<A HREF ="#Specification"><B>[Specification]</B></A>
<A HREF ="#Arguments"><B>[Arguments]</B></A>
<A HREF ="#Method"><B>[Method]</B></A>
<A HREF ="#References"><B>[References]</B></A>
<A HREF ="#Comments"><B>[Comments]</B></A>
<A HREF ="#Example"><B>[Example]</B></A>
<P>
<B><FONT SIZE="+1">Purpose</FONT></B>
<PRE>
To solve the real continuous-time matrix algebraic Riccati
equation
op(A)'*X + X*op(A) + Q - X*G*X = 0,
where op(A) = A or A' = A**T and G, Q are symmetric (G = G**T,
Q = Q**T). The matrices A, G and Q are N-by-N and the solution X
is an N-by-N symmetric matrix.
An error bound on the solution and a condition estimate are also
optionally provided.
It is assumed that the matrices A, G and Q are such that the
corresponding Hamiltonian matrix has N eigenvalues with negative
real parts.
</PRE>
<A name="Specification"><B><FONT SIZE="+1">Specification</FONT></B></A>
<PRE>
SUBROUTINE SB02PD( JOB, TRANA, UPLO, N, A, LDA, G, LDG, Q, LDQ, X,
$ LDX, RCOND, FERR, WR, WI, IWORK, DWORK, LDWORK,
$ INFO )
C .. Scalar Arguments ..
CHARACTER JOB, TRANA, UPLO
INTEGER INFO, LDA, LDG, LDQ, LDWORK, LDX, N
DOUBLE PRECISION FERR, RCOND
C .. Array Arguments ..
INTEGER IWORK( * )
DOUBLE PRECISION A( LDA, * ), DWORK( * ), G( LDG, * ),
$ Q( LDQ, * ), WI( * ), WR( * ), X( LDX, * )
</PRE>
<A name="Arguments"><B><FONT SIZE="+1">Arguments</FONT></B></A>
<P>
<B>Mode Parameters</B>
<PRE>
JOB CHARACTER*1
Specifies the computation to be performed, as follows:
= 'X': Compute the solution only;
= 'A': Compute all: the solution, reciprocal condition
number, and the error bound.
TRANA CHARACTER*1
Specifies the option op(A):
= 'N': op(A) = A (No transpose);
= 'T': op(A) = A**T (Transpose);
= 'C': op(A) = A**T (Conjugate transpose = Transpose).
UPLO CHARACTER*1
Specifies which triangle of the matrices G and Q is
stored, as follows:
= 'U': Upper triangles of G and Q are stored;
= 'L': Lower triangles of G and Q are stored.
</PRE>
<B>Input/Output Parameters</B>
<PRE>
N (input) INTEGER
The order of the matrices A, G, Q, and X. N >= 0.
A (input) DOUBLE PRECISION array, dimension (LDA,N)
The leading N-by-N part of this array must contain the
coefficient matrix A of the equation.
LDA INTEGER
The leading dimension of the array A. LDA >= max(1,N).
G (input) DOUBLE PRECISION array, dimension (LDG,N)
If UPLO = 'U', the leading N-by-N upper triangular part of
this array must contain the upper triangular part of the
matrix G.
If UPLO = 'L', the leading N-by-N lower triangular part of
this array must contain the lower triangular part of the
matrix G.
LDG INTEGER
The leading dimension of the array G. LDG >= max(1,N).
Q (input) DOUBLE PRECISION array, dimension (LDQ,N)
If UPLO = 'U', the leading N-by-N upper triangular part of
this array must contain the upper triangular part of the
matrix Q.
If UPLO = 'L', the leading N-by-N lower triangular part of
this array must contain the lower triangular part of the
matrix Q.
LDQ INTEGER
The leading dimension of the array Q. LDQ >= max(1,N).
X (output) DOUBLE PRECISION array, dimension (LDX,N)
If INFO = 0, INFO = 2, or INFO = 4, the leading N-by-N
part of this array contains the symmetric solution matrix
X of the algebraic Riccati equation.
LDX INTEGER
The leading dimension of the array X. LDX >= max(1,N).
RCOND (output) DOUBLE PRECISION
If JOB = 'A', the estimate of the reciprocal condition
number of the Riccati equation.
FERR (output) DOUBLE PRECISION
If JOB = 'A', the estimated forward error bound for the
solution X. If XTRUE is the true solution, FERR bounds the
magnitude of the largest entry in (X - XTRUE) divided by
the magnitude of the largest entry in X.
WR (output) DOUBLE PRECISION array, dimension (N)
WI (output) DOUBLE PRECISION array, dimension (N)
If JOB = 'A' and TRANA = 'N', WR and WI contain the real
and imaginary parts, respectively, of the eigenvalues of
the matrix A - G*X, i.e., the closed-loop system poles.
If JOB = 'A' and TRANA = 'T' or 'C', WR and WI contain the
real and imaginary parts, respectively, of the eigenvalues
of the matrix A - X*G, i.e., the closed-loop system poles.
If JOB = 'X', these arrays are not referenced.
</PRE>
<B>Workspace</B>
<PRE>
IWORK INTEGER array, dimension (LIWORK), where
LIWORK >= 2*N, if JOB = 'X';
LIWORK >= max(2*N,N*N), if JOB = 'A'.
DWORK DOUBLE PRECISION array, dimension (LDWORK)
On exit, if INFO = 0 or INFO = 2, DWORK(1) contains the
optimal value of LDWORK. If JOB = 'A', then DWORK(2:N*N+1)
and DWORK(N*N+2:2*N*N+1) contain a real Schur form of the
closed-loop system matrix, Ac = A - G*X (if TRANA = 'N')
or Ac = A - X*G (if TRANA = 'T' or 'C'), and the
orthogonal matrix which reduced Ac to real Schur form,
respectively.
LDWORK INTEGER
The dimension of the array DWORK.
LDWORK >= 4*N*N + 8*N + 1, if JOB = 'X';
LDWORK >= max( 4*N*N + 8*N + 1, 6*N*N ), if JOB = 'A'.
For good performance, LDWORK should be larger, e.g.,
LDWORK >= 4*N*N + 6*N +( 2*N+1 )*NB, if JOB = 'X',
where NB is the optimal blocksize.
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.
</PRE>
<B>Error Indicator</B>
<PRE>
INFO INTEGER
= 0: successful exit;
< 0: if INFO = -i, the i-th argument had an illegal
value;
= 1: the Hamiltonian matrix has eigenvalues on the
imaginary axis, so the solution and error bounds
could not be computed;
= 2: the iteration for the matrix sign function failed to
converge after 50 iterations, but an approximate
solution and error bounds (if JOB = 'A') have been
computed;
= 3: the system of linear equations for the solution is
singular to working precision, so the solution and
error bounds could not be computed;
= 4: the matrix A-G*X (or A-X*G) cannot be reduced to
Schur canonical form and condition number estimate
and forward error estimate have not been computed.
</PRE>
<A name="Method"><B><FONT SIZE="+1">Method</FONT></B></A>
<PRE>
The Riccati equation is solved by the matrix sign function
approach [1], [2], implementing a scaling which enhances the
numerical stability [4].
</PRE>
<A name="References"><B><FONT SIZE="+1">References</FONT></B></A>
<PRE>
[1] Bai, Z., Demmel, J., Dongarra, J., Petitet, A., Robinson, H.,
and Stanley, K.
The spectral decomposition of nonsymmetric matrices on
distributed memory parallel computers.
SIAM J. Sci. Comput., vol. 18, pp. 1446-1461, 1997.
[2] Byers, R., He, C., and Mehrmann, V.
The matrix sign function method and the computation of
invariant subspaces.
SIAM J. Matrix Anal. Appl., vol. 18, pp. 615-632, 1997.
[3] Higham, N.J.
Perturbation theory and backward error for AX-XB=C.
BIT, vol. 33, pp. 124-136, 1993.
[4] Petkov, P.Hr., Konstantinov, M.M., and Mehrmann, V.,
DGRSVX and DMSRIC: Fortran 77 subroutines for solving
continuous-time matrix algebraic Riccati equations with
condition and accuracy estimates.
Preprint SFB393/98-16, Fak. f. Mathematik, Technical
University Chemnitz, May 1998.
</PRE>
<A name="Numerical Aspects"><B><FONT SIZE="+1">Numerical Aspects</FONT></B></A>
<PRE>
The solution accuracy can be controlled by the output parameter
FERR.
</PRE>
<A name="Comments"><B><FONT SIZE="+1">Further Comments</FONT></B></A>
<PRE>
The condition number of the Riccati equation is estimated as
cond = ( norm(Theta)*norm(A) + norm(inv(Omega))*norm(Q) +
norm(Pi)*norm(G) ) / norm(X),
where Omega, Theta and Pi are linear operators defined by
Omega(W) = op(Ac)'*W + W*op(Ac),
Theta(W) = inv(Omega(op(W)'*X + X*op(W))),
Pi(W) = inv(Omega(X*W*X)),
and the matrix Ac (the closed-loop system matrix) is given by
Ac = A - G*X, if TRANA = 'N', or
Ac = A - X*G, if TRANA = 'T' or 'C'.
The program estimates the quantities
sep(op(Ac),-op(Ac)') = 1 / norm(inv(Omega)),
norm(Theta) and norm(Pi) using 1-norm condition estimator.
The forward error bound is estimated using a practical error bound
similar to the one proposed in [3].
</PRE>
<A name="Example"><B><FONT SIZE="+1">Example</FONT></B></A>
<P>
<B>Program Text</B>
<PRE>
* SB02PD EXAMPLE PROGRAM TEXT.
*
* .. Parameters ..
INTEGER NIN, NOUT
PARAMETER ( NIN = 5, NOUT = 6 )
INTEGER NMAX
PARAMETER ( NMAX = 20 )
INTEGER LDA, LDG, LDQ, LDX
PARAMETER ( LDA = NMAX, LDG = NMAX, LDQ = NMAX,
$ LDX = NMAX )
INTEGER LIWORK
PARAMETER ( LIWORK = MAX( 2*NMAX, NMAX*NMAX ) )
INTEGER LDWORK
PARAMETER ( LDWORK = MAX( 4*NMAX*NMAX + 8*NMAX,
$ 6*NMAX*NMAX ) + 1 )
* .. Local Scalars ..
DOUBLE PRECISION FERR, RCOND
INTEGER I, INFO, J, N
CHARACTER JOB, TRANA, UPLO
* .. Local Arrays ..
DOUBLE PRECISION A(LDA,NMAX), DWORK(LDWORK), G(LDG,NMAX),
$ Q(LDQ,NMAX), WI(NMAX), WR(NMAX),
$ X(LDX,NMAX)
INTEGER IWORK(LIWORK)
* .. External Functions ..
LOGICAL LSAME
EXTERNAL LSAME
* .. External Subroutines ..
EXTERNAL SB02PD
* .. Intrinsic Functions ..
INTRINSIC MAX
* .. Executable Statements ..
*
WRITE ( NOUT, FMT = 99999 )
* Skip the heading in the data file and read the data.
READ ( NIN, FMT = '()' )
READ ( NIN, FMT = * ) N, JOB, TRANA, UPLO
IF ( N.LT.0 .OR. N.GT.NMAX ) THEN
WRITE ( NOUT, FMT = 99995 ) N
ELSE
READ ( NIN, FMT = * ) ( ( A(I,J), J = 1,N ), I = 1,N )
READ ( NIN, FMT = * ) ( ( Q(I,J), J = 1,N ), I = 1,N )
READ ( NIN, FMT = * ) ( ( G(I,J), J = 1,N ), I = 1,N )
* Find the solution matrix X.
CALL SB02PD( JOB, TRANA, UPLO, N, A, LDA, G, LDG, Q, LDQ, X,
$ LDX, RCOND, FERR, WR, WI, IWORK, DWORK, LDWORK,
$ INFO )
*
IF ( INFO.NE.0 ) THEN
WRITE ( NOUT, FMT = 99998 ) INFO
END IF
IF ( INFO.EQ.0 .OR. INFO.EQ.2 .OR. INFO.EQ.4 ) THEN
WRITE ( NOUT, FMT = 99997 )
DO 20 I = 1, N
WRITE ( NOUT, FMT = 99996 ) ( X(I,J), J = 1,N )
20 CONTINUE
IF ( LSAME( JOB, 'A' ) .AND. INFO.NE.4 ) THEN
WRITE ( NOUT, FMT = 99994 ) RCOND
WRITE ( NOUT, FMT = 99993 ) FERR
END IF
END IF
END IF
STOP
*
99999 FORMAT (' SB02PD EXAMPLE PROGRAM RESULTS',/1X)
99998 FORMAT (' INFO on exit from SB02PD = ',I2)
99997 FORMAT (' The solution matrix X is ')
99996 FORMAT (20(1X,F8.4))
99995 FORMAT (/' N is out of range.',/' N = ',I5)
99994 FORMAT (/' Estimated reciprocal condition number = ',F8.4)
99993 FORMAT (/' Estimated error bound = ',F20.16)
END
</PRE>
<B>Program Data</B>
<PRE>
SB02PD EXAMPLE PROGRAM DATA
2 A N U
0.0 1.0
0.0 0.0
1.0 0.0
0.0 2.0
0.0 0.0
0.0 1.0
</PRE>
<B>Program Results</B>
<PRE>
SB02PD EXAMPLE PROGRAM RESULTS
The solution matrix X is
2.0000 1.0000
1.0000 2.0000
Estimated reciprocal condition number = 0.1333
Estimated error bound = 0.0000000000000063
</PRE>
<HR>
<p>
<A HREF=..\libindex.html><B>Return to index</B></A></BODY>
</HTML>
|
