3 * =========== DOCUMENTATION ===========
5 * Online html documentation available at
6 * http://www.netlib.org/lapack/explore-html/
9 *> Download DTRSEN + dependencies
10 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dtrsen.f">
12 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dtrsen.f">
14 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dtrsen.f">
21 * SUBROUTINE DTRSEN( JOB, COMPQ, SELECT, N, T, LDT, Q, LDQ, WR, WI,
22 * M, S, SEP, WORK, LWORK, IWORK, LIWORK, INFO )
24 * .. Scalar Arguments ..
25 * CHARACTER COMPQ, JOB
26 * INTEGER INFO, LDQ, LDT, LIWORK, LWORK, M, N
27 * DOUBLE PRECISION S, SEP
29 * .. Array Arguments ..
32 * DOUBLE PRECISION Q( LDQ, * ), T( LDT, * ), WI( * ), WORK( * ),
42 *> DTRSEN reorders the real Schur factorization of a real matrix
43 *> A = Q*T*Q**T, so that a selected cluster of eigenvalues appears in
44 *> the leading diagonal blocks of the upper quasi-triangular matrix T,
45 *> and the leading columns of Q form an orthonormal basis of the
46 *> corresponding right invariant subspace.
48 *> Optionally the routine computes the reciprocal condition numbers of
49 *> the cluster of eigenvalues and/or the invariant subspace.
51 *> T must be in Schur canonical form (as returned by DHSEQR), that is,
52 *> block upper triangular with 1-by-1 and 2-by-2 diagonal blocks; each
53 *> 2-by-2 diagonal block has its diagonal elements equal and its
54 *> off-diagonal elements of opposite sign.
63 *> Specifies whether condition numbers are required for the
64 *> cluster of eigenvalues (S) or the invariant subspace (SEP):
66 *> = 'E': for eigenvalues only (S);
67 *> = 'V': for invariant subspace only (SEP);
68 *> = 'B': for both eigenvalues and invariant subspace (S and
74 *> COMPQ is CHARACTER*1
75 *> = 'V': update the matrix Q of Schur vectors;
76 *> = 'N': do not update Q.
81 *> SELECT is LOGICAL array, dimension (N)
82 *> SELECT specifies the eigenvalues in the selected cluster. To
83 *> select a real eigenvalue w(j), SELECT(j) must be set to
84 *> .TRUE.. To select a complex conjugate pair of eigenvalues
85 *> w(j) and w(j+1), corresponding to a 2-by-2 diagonal block,
86 *> either SELECT(j) or SELECT(j+1) or both must be set to
87 *> .TRUE.; a complex conjugate pair of eigenvalues must be
88 *> either both included in the cluster or both excluded.
94 *> The order of the matrix T. N >= 0.
99 *> T is DOUBLE PRECISION array, dimension (LDT,N)
100 *> On entry, the upper quasi-triangular matrix T, in Schur
102 *> On exit, T is overwritten by the reordered matrix T, again in
103 *> Schur canonical form, with the selected eigenvalues in the
104 *> leading diagonal blocks.
110 *> The leading dimension of the array T. LDT >= max(1,N).
115 *> Q is DOUBLE PRECISION array, dimension (LDQ,N)
116 *> On entry, if COMPQ = 'V', the matrix Q of Schur vectors.
117 *> On exit, if COMPQ = 'V', Q has been postmultiplied by the
118 *> orthogonal transformation matrix which reorders T; the
119 *> leading M columns of Q form an orthonormal basis for the
120 *> specified invariant subspace.
121 *> If COMPQ = 'N', Q is not referenced.
127 *> The leading dimension of the array Q.
128 *> LDQ >= 1; and if COMPQ = 'V', LDQ >= N.
133 *> WR is DOUBLE PRECISION array, dimension (N)
137 *> WI is DOUBLE PRECISION array, dimension (N)
139 *> The real and imaginary parts, respectively, of the reordered
140 *> eigenvalues of T. The eigenvalues are stored in the same
141 *> order as on the diagonal of T, with WR(i) = T(i,i) and, if
142 *> T(i:i+1,i:i+1) is a 2-by-2 diagonal block, WI(i) > 0 and
143 *> WI(i+1) = -WI(i). Note that if a complex eigenvalue is
144 *> sufficiently ill-conditioned, then its value may differ
145 *> significantly from its value before reordering.
151 *> The dimension of the specified invariant subspace.
157 *> S is DOUBLE PRECISION
158 *> If JOB = 'E' or 'B', S is a lower bound on the reciprocal
159 *> condition number for the selected cluster of eigenvalues.
160 *> S cannot underestimate the true reciprocal condition number
161 *> by more than a factor of sqrt(N). If M = 0 or N, S = 1.
162 *> If JOB = 'N' or 'V', S is not referenced.
167 *> SEP is DOUBLE PRECISION
168 *> If JOB = 'V' or 'B', SEP is the estimated reciprocal
169 *> condition number of the specified invariant subspace. If
170 *> M = 0 or N, SEP = norm(T).
171 *> If JOB = 'N' or 'E', SEP is not referenced.
176 *> WORK is DOUBLE PRECISION array, dimension (MAX(1,LWORK))
177 *> On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
183 *> The dimension of the array WORK.
184 *> If JOB = 'N', LWORK >= max(1,N);
185 *> if JOB = 'E', LWORK >= max(1,M*(N-M));
186 *> if JOB = 'V' or 'B', LWORK >= max(1,2*M*(N-M)).
188 *> If LWORK = -1, then a workspace query is assumed; the routine
189 *> only calculates the optimal size of the WORK array, returns
190 *> this value as the first entry of the WORK array, and no error
191 *> message related to LWORK is issued by XERBLA.
196 *> IWORK is INTEGER array, dimension (MAX(1,LIWORK))
197 *> On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK.
203 *> The dimension of the array IWORK.
204 *> If JOB = 'N' or 'E', LIWORK >= 1;
205 *> if JOB = 'V' or 'B', LIWORK >= max(1,M*(N-M)).
207 *> If LIWORK = -1, then a workspace query is assumed; the
208 *> routine only calculates the optimal size of the IWORK array,
209 *> returns this value as the first entry of the IWORK array, and
210 *> no error message related to LIWORK is issued by XERBLA.
216 *> = 0: successful exit
217 *> < 0: if INFO = -i, the i-th argument had an illegal value
218 *> = 1: reordering of T failed because some eigenvalues are too
219 *> close to separate (the problem is very ill-conditioned);
220 *> T may have been partially reordered, and WR and WI
221 *> contain the eigenvalues in the same order as in T; S and
222 *> SEP (if requested) are set to zero.
228 *> \author Univ. of Tennessee
229 *> \author Univ. of California Berkeley
230 *> \author Univ. of Colorado Denver
235 *> \ingroup doubleOTHERcomputational
237 *> \par Further Details:
238 * =====================
242 *> DTRSEN first collects the selected eigenvalues by computing an
243 *> orthogonal transformation Z to move them to the top left corner of T.
244 *> In other words, the selected eigenvalues are the eigenvalues of T11
247 *> Z**T * T * Z = ( T11 T12 ) n1
251 *> where N = n1+n2 and Z**T means the transpose of Z. The first n1 columns
252 *> of Z span the specified invariant subspace of T.
254 *> If T has been obtained from the real Schur factorization of a matrix
255 *> A = Q*T*Q**T, then the reordered real Schur factorization of A is given
256 *> by A = (Q*Z)*(Z**T*T*Z)*(Q*Z)**T, and the first n1 columns of Q*Z span
257 *> the corresponding invariant subspace of A.
259 *> The reciprocal condition number of the average of the eigenvalues of
260 *> T11 may be returned in S. S lies between 0 (very badly conditioned)
261 *> and 1 (very well conditioned). It is computed as follows. First we
268 *> is the projector on the invariant subspace associated with T11.
269 *> R is the solution of the Sylvester equation:
271 *> T11*R - R*T22 = T12.
273 *> Let F-norm(M) denote the Frobenius-norm of M and 2-norm(M) denote
274 *> the two-norm of M. Then S is computed as the lower bound
276 *> (1 + F-norm(R)**2)**(-1/2)
278 *> on the reciprocal of 2-norm(P), the true reciprocal condition number.
279 *> S cannot underestimate 1 / 2-norm(P) by more than a factor of
282 *> An approximate error bound for the computed average of the
283 *> eigenvalues of T11 is
287 *> where EPS is the machine precision.
289 *> The reciprocal condition number of the right invariant subspace
290 *> spanned by the first n1 columns of Z (or of Q*Z) is returned in SEP.
291 *> SEP is defined as the separation of T11 and T22:
293 *> sep( T11, T22 ) = sigma-min( C )
295 *> where sigma-min(C) is the smallest singular value of the
296 *> n1*n2-by-n1*n2 matrix
298 *> C = kprod( I(n2), T11 ) - kprod( transpose(T22), I(n1) )
300 *> I(m) is an m by m identity matrix, and kprod denotes the Kronecker
301 *> product. We estimate sigma-min(C) by the reciprocal of an estimate of
302 *> the 1-norm of inverse(C). The true reciprocal 1-norm of inverse(C)
303 *> cannot differ from sigma-min(C) by more than a factor of sqrt(n1*n2).
305 *> When SEP is small, small changes in T can cause large changes in
306 *> the invariant subspace. An approximate bound on the maximum angular
307 *> error in the computed right invariant subspace is
309 *> EPS * norm(T) / SEP
312 * =====================================================================
313 SUBROUTINE DTRSEN( JOB, COMPQ, SELECT, N, T, LDT, Q, LDQ, WR, WI,
314 $ M, S, SEP, WORK, LWORK, IWORK, LIWORK, INFO )
316 * -- LAPACK computational routine (version 3.4.1) --
317 * -- LAPACK is a software package provided by Univ. of Tennessee, --
318 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
321 * .. Scalar Arguments ..
323 INTEGER INFO, LDQ, LDT, LIWORK, LWORK, M, N
324 DOUBLE PRECISION S, SEP
326 * .. Array Arguments ..
329 DOUBLE PRECISION Q( LDQ, * ), T( LDT, * ), WI( * ), WORK( * ),
333 * =====================================================================
336 DOUBLE PRECISION ZERO, ONE
337 PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 )
339 * .. Local Scalars ..
340 LOGICAL LQUERY, PAIR, SWAP, WANTBH, WANTQ, WANTS,
342 INTEGER IERR, K, KASE, KK, KS, LIWMIN, LWMIN, N1, N2,
344 DOUBLE PRECISION EST, RNORM, SCALE
349 * .. External Functions ..
351 DOUBLE PRECISION DLANGE
352 EXTERNAL LSAME, DLANGE
354 * .. External Subroutines ..
355 EXTERNAL DLACN2, DLACPY, DTREXC, DTRSYL, XERBLA
357 * .. Intrinsic Functions ..
358 INTRINSIC ABS, MAX, SQRT
360 * .. Executable Statements ..
362 * Decode and test the input parameters
364 WANTBH = LSAME( JOB, 'B' )
365 WANTS = LSAME( JOB, 'E' ) .OR. WANTBH
366 WANTSP = LSAME( JOB, 'V' ) .OR. WANTBH
367 WANTQ = LSAME( COMPQ, 'V' )
370 LQUERY = ( LWORK.EQ.-1 )
371 IF( .NOT.LSAME( JOB, 'N' ) .AND. .NOT.WANTS .AND. .NOT.WANTSP )
374 ELSE IF( .NOT.LSAME( COMPQ, 'N' ) .AND. .NOT.WANTQ ) THEN
376 ELSE IF( N.LT.0 ) THEN
378 ELSE IF( LDT.LT.MAX( 1, N ) ) THEN
380 ELSE IF( LDQ.LT.1 .OR. ( WANTQ .AND. LDQ.LT.N ) ) THEN
384 * Set M to the dimension of the specified invariant subspace,
385 * and test LWORK and LIWORK.
394 IF( T( K+1, K ).EQ.ZERO ) THEN
399 IF( SELECT( K ) .OR. SELECT( K+1 ) )
414 LWMIN = MAX( 1, 2*NN )
415 LIWMIN = MAX( 1, NN )
416 ELSE IF( LSAME( JOB, 'N' ) ) THEN
419 ELSE IF( LSAME( JOB, 'E' ) ) THEN
424 IF( LWORK.LT.LWMIN .AND. .NOT.LQUERY ) THEN
426 ELSE IF( LIWORK.LT.LIWMIN .AND. .NOT.LQUERY ) THEN
437 CALL XERBLA( 'DTRSEN', -INFO )
439 ELSE IF( LQUERY ) THEN
443 * Quick return if possible.
445 IF( M.EQ.N .OR. M.EQ.0 ) THEN
449 $ SEP = DLANGE( '1', N, N, T, LDT, WORK )
453 * Collect the selected blocks at the top-left corner of T.
463 IF( T( K+1, K ).NE.ZERO ) THEN
465 SWAP = SWAP .OR. SELECT( K+1 )
471 * Swap the K-th block to position KS.
476 $ CALL DTREXC( COMPQ, N, T, LDT, Q, LDQ, KK, KS, WORK,
478 IF( IERR.EQ.1 .OR. IERR.EQ.2 ) THEN
480 * Blocks too close to swap: exit.
497 * Solve Sylvester equation for R:
499 * T11*R - R*T22 = scale*T12
501 CALL DLACPY( 'F', N1, N2, T( 1, N1+1 ), LDT, WORK, N1 )
502 CALL DTRSYL( 'N', 'N', -1, N1, N2, T, LDT, T( N1+1, N1+1 ),
503 $ LDT, WORK, N1, SCALE, IERR )
505 * Estimate the reciprocal of the condition number of the cluster
508 RNORM = DLANGE( 'F', N1, N2, WORK, N1, WORK )
509 IF( RNORM.EQ.ZERO ) THEN
512 S = SCALE / ( SQRT( SCALE*SCALE / RNORM+RNORM )*
519 * Estimate sep(T11,T22).
524 CALL DLACN2( NN, WORK( NN+1 ), WORK, IWORK, EST, KASE, ISAVE )
528 * Solve T11*R - R*T22 = scale*X.
530 CALL DTRSYL( 'N', 'N', -1, N1, N2, T, LDT,
531 $ T( N1+1, N1+1 ), LDT, WORK, N1, SCALE,
535 * Solve T11**T*R - R*T22**T = scale*X.
537 CALL DTRSYL( 'T', 'T', -1, N1, N2, T, LDT,
538 $ T( N1+1, N1+1 ), LDT, WORK, N1, SCALE,
549 * Store the output eigenvalues in WR and WI.
556 IF( T( K+1, K ).NE.ZERO ) THEN
557 WI( K ) = SQRT( ABS( T( K, K+1 ) ) )*
558 $ SQRT( ABS( T( K+1, K ) ) )