forked from SLICOT/SLICOT-Reference
-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathAB08MD.f
More file actions
281 lines (281 loc) · 9.21 KB
/
Copy pathAB08MD.f
File metadata and controls
281 lines (281 loc) · 9.21 KB
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
SUBROUTINE AB08MD( EQUIL, N, M, P, A, LDA, B, LDB, C, LDC, D, LDD,
$ RANK, TOL, IWORK, DWORK, LDWORK, INFO )
C
C PURPOSE
C
C To compute the normal rank of the transfer-function matrix of a
C state-space model (A,B,C,D).
C
C ARGUMENTS
C
C Mode Parameters
C
C EQUIL CHARACTER*1
C Specifies whether the user wishes to balance the compound
C matrix (see METHOD) as follows:
C = 'S': Perform balancing (scaling);
C = 'N': Do not perform balancing.
C
C Input/Output Parameters
C
C N (input) INTEGER
C The number of state variables, i.e., the order of the
C matrix A. N >= 0.
C
C M (input) INTEGER
C The number of system inputs. M >= 0.
C
C P (input) INTEGER
C The number of system outputs. P >= 0.
C
C A (input) DOUBLE PRECISION array, dimension (LDA,N)
C The leading N-by-N part of this array must contain the
C state dynamics matrix A of the system.
C
C LDA INTEGER
C The leading dimension of array A. LDA >= MAX(1,N).
C
C B (input) DOUBLE PRECISION array, dimension (LDB,M)
C The leading N-by-M part of this array must contain the
C input/state matrix B of the system.
C
C LDB INTEGER
C The leading dimension of array B. LDB >= MAX(1,N).
C
C C (input) DOUBLE PRECISION array, dimension (LDC,N)
C The leading P-by-N part of this array must contain the
C state/output matrix C of the system.
C
C LDC INTEGER
C The leading dimension of array C. LDC >= MAX(1,P).
C
C D (input) DOUBLE PRECISION array, dimension (LDD,M)
C The leading P-by-M part of this array must contain the
C direct transmission matrix D of the system.
C
C LDD INTEGER
C The leading dimension of array D. LDD >= MAX(1,P).
C
C RANK (output) INTEGER
C The normal rank of the transfer-function matrix.
C
C Tolerances
C
C TOL DOUBLE PRECISION
C A tolerance used in rank decisions to determine the
C effective rank, which is defined as the order of the
C largest leading (or trailing) triangular submatrix in the
C QR (or RQ) factorization with column (or row) pivoting
C whose estimated condition number is less than 1/TOL.
C If the user sets TOL to be less than SQRT((N+P)*(N+M))*EPS
C then the tolerance is taken as SQRT((N+P)*(N+M))*EPS,
C where EPS is the machine precision (see LAPACK Library
C Routine DLAMCH).
C
C Workspace
C
C IWORK INTEGER array, dimension (2*N+MAX(M,P)+1)
C
C DWORK DOUBLE PRECISION array, dimension (LDWORK)
C On exit, if INFO = 0, DWORK(1) returns the optimal value
C of LDWORK.
C
C LDWORK INTEGER
C The length of the array DWORK.
C LDWORK >= (N+P)*(N+M) +
C MAX( MIN(P,M) + MAX(3*M-1,N), 1,
C MIN(P,N) + MAX(3*P-1,N+P,N+M) )
C For optimum performance LDWORK should be larger.
C
C If LDWORK = -1, then a workspace query is assumed;
C the routine only calculates the optimal size of the
C DWORK array, returns this value as the first entry of
C the DWORK array, and no error message related to LDWORK
C is issued by XERBLA.
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
C METHOD
C
C The routine reduces the (N+P)-by-(M+N) compound matrix (B A)
C (D C)
C
C to one with the same invariant zeros and with D of full row rank.
C The normal rank of the transfer-function matrix is the rank of D.
C
C REFERENCES
C
C [1] Svaricek, F.
C Computation of the Structural Invariants of Linear
C Multivariable Systems with an Extended Version of
C the Program ZEROS.
C System & Control Letters, 6, pp. 261-266, 1985.
C
C [2] Emami-Naeini, A. and Van Dooren, P.
C Computation of Zeros of Linear Multivariable Systems.
C Automatica, 18, pp. 415-430, 1982.
C
C NUMERICAL ASPECTS
C
C The algorithm is backward stable (see [2] and [1]).
C
C CONTRIBUTOR
C
C A. Varga, German Aerospace Center, Oberpfaffenhofen, May 2001.
C
C REVISIONS
C
C V. Sima, Research Institute for Informatics, Bucharest, June 2001,
C Dec. 2003, Jan. 2009, Mar. 2009, Apr. 2009.
C
C KEYWORDS
C
C Multivariable system, orthogonal transformation,
C structural invariant.
C
C ******************************************************************
C
C .. Parameters ..
DOUBLE PRECISION ZERO, ONE
PARAMETER ( ZERO = 0.0D0, ONE = 1.0D0 )
C .. Scalar Arguments ..
CHARACTER EQUIL
INTEGER INFO, LDA, LDB, LDC, LDD, LDWORK, M, N, P, RANK
DOUBLE PRECISION TOL
C .. Array Arguments ..
INTEGER IWORK(*)
DOUBLE PRECISION A(LDA,*), B(LDB,*), C(LDC,*), D(LDD,*), DWORK(*)
C .. Local Scalars ..
LOGICAL LEQUIL, LQUERY
INTEGER I, KW, MU, NINFZ, NKROL, NM, NP, NU, RO,
$ SIGMA, WRKOPT
DOUBLE PRECISION MAXRED, SVLMAX, THRESH, TOLER
C .. External Functions ..
LOGICAL LSAME
DOUBLE PRECISION DLAMCH, DLANGE
EXTERNAL DLAMCH, DLANGE, LSAME
C .. External Subroutines ..
EXTERNAL AB08NX, DLACPY, TB01ID, XERBLA
C .. Intrinsic Functions ..
INTRINSIC DBLE, INT, MAX, MIN, SQRT
C .. Executable Statements ..
C
NP = N + P
NM = N + M
INFO = 0
LEQUIL = LSAME( EQUIL, 'S' )
LQUERY = ( LDWORK.EQ.-1 )
WRKOPT = NP*NM
C
C Test the input scalar arguments.
C
IF( .NOT.LEQUIL .AND. .NOT.LSAME( EQUIL, 'N' ) ) THEN
INFO = -1
ELSE IF( N.LT.0 ) THEN
INFO = -2
ELSE IF( M.LT.0 ) THEN
INFO = -3
ELSE IF( P.LT.0 ) THEN
INFO = -4
ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
INFO = -6
ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
INFO = -8
ELSE IF( LDC.LT.MAX( 1, P ) ) THEN
INFO = -10
ELSE IF( LDD.LT.MAX( 1, P ) ) THEN
INFO = -12
ELSE
KW = WRKOPT + MAX( MIN( P, M ) + MAX( 3*M-1, N ), 1,
$ MIN( P, N ) + MAX( 3*P-1, NP, NM ) )
IF( LQUERY ) THEN
SVLMAX = ZERO
NINFZ = 0
CALL AB08NX( N, M, P, P, 0, SVLMAX, DWORK, MAX( 1, NP ),
$ NINFZ, IWORK, IWORK, MU, NU, NKROL, TOL, IWORK,
$ DWORK, -1, INFO )
WRKOPT = MAX( KW, WRKOPT + INT( DWORK(1) ) )
ELSE IF( LDWORK.LT.KW ) THEN
INFO = -17
END IF
END IF
C
IF ( INFO.NE.0 ) THEN
C
C Error return.
C
CALL XERBLA( 'AB08MD', -INFO )
RETURN
ELSE IF( LQUERY ) THEN
DWORK(1) = WRKOPT
RETURN
END IF
C
C Quick return if possible.
C
IF ( MIN( M, P ).EQ.0 ) THEN
RANK = 0
DWORK(1) = ONE
RETURN
END IF
C
DO 10 I = 1, 2*N+1
IWORK(I) = 0
10 CONTINUE
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
C Construct the compound matrix ( B A ), dimension (N+P)-by-(M+N).
C ( D C )
C Workspace: need (N+P)*(N+M).
C
CALL DLACPY( 'Full', N, M, B, LDB, DWORK, NP )
CALL DLACPY( 'Full', P, M, D, LDD, DWORK(N+1), NP )
CALL DLACPY( 'Full', N, N, A, LDA, DWORK(NP*M+1), NP )
CALL DLACPY( 'Full', P, N, C, LDC, DWORK(NP*M+N+1), NP )
C
C If required, balance the compound matrix (default MAXRED).
C Workspace: need N.
C
KW = WRKOPT + 1
IF ( LEQUIL ) THEN
MAXRED = ZERO
CALL TB01ID( 'A', N, M, P, MAXRED, DWORK(NP*M+1), NP, DWORK,
$ NP, DWORK(NP*M+N+1), NP, DWORK(KW), INFO )
WRKOPT = WRKOPT + N
END IF
C
C If required, set tolerance.
C
THRESH = SQRT( DBLE( NP*NM ) )*DLAMCH( 'Precision' )
TOLER = TOL
IF ( TOLER.LT.THRESH ) TOLER = THRESH
SVLMAX = DLANGE( 'Frobenius', NP, NM, DWORK, NP, DWORK(KW) )
C
C Reduce this system to one with the same invariant zeros and with
C D full row rank MU (the normal rank of the original system).
C Real workspace: need (N+P)*(N+M) +
C MAX( 1, MIN(P,M) + MAX(3*M-1,N),
C MIN(P,N) + MAX(3*P-1,N+P,N+M) );
C prefer larger.
C Integer workspace: 2*N+MAX(M,P)+1.
C
RO = P
SIGMA = 0
NINFZ = 0
CALL AB08NX( N, M, P, RO, SIGMA, SVLMAX, DWORK, NP, NINFZ, IWORK,
$ IWORK(N+1), MU, NU, NKROL, TOLER, IWORK(2*N+2),
$ DWORK(KW), LDWORK-KW+1, INFO )
RANK = MU
C
DWORK(1) = MAX( WRKOPT, INT( DWORK(KW) ) + KW - 1 )
RETURN
C *** Last line of AB08MD ***
END