1: *> \brief <b> DGGSVD computes the singular value decomposition (SVD) for OTHER matrices</b>
2: *
3: * =========== DOCUMENTATION ===========
4: *
5: * Online html documentation available at
6: * http://www.netlib.org/lapack/explore-html/
7: *
8: *> \htmlonly
9: *> Download DGGSVD + dependencies
10: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dggsvd.f">
11: *> [TGZ]</a>
12: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dggsvd.f">
13: *> [ZIP]</a>
14: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dggsvd.f">
15: *> [TXT]</a>
16: *> \endhtmlonly
17: *
18: * Definition:
19: * ===========
20: *
21: * SUBROUTINE DGGSVD( JOBU, JOBV, JOBQ, M, N, P, K, L, A, LDA, B,
22: * LDB, ALPHA, BETA, U, LDU, V, LDV, Q, LDQ, WORK,
23: * IWORK, INFO )
24: *
25: * .. Scalar Arguments ..
26: * CHARACTER JOBQ, JOBU, JOBV
27: * INTEGER INFO, K, L, LDA, LDB, LDQ, LDU, LDV, M, N, P
28: * ..
29: * .. Array Arguments ..
30: * INTEGER IWORK( * )
31: * DOUBLE PRECISION A( LDA, * ), ALPHA( * ), B( LDB, * ),
32: * $ BETA( * ), Q( LDQ, * ), U( LDU, * ),
33: * $ V( LDV, * ), WORK( * )
34: * ..
35: *
36: *
37: *> \par Purpose:
38: * =============
39: *>
40: *> \verbatim
41: *>
42: *> DGGSVD computes the generalized singular value decomposition (GSVD)
43: *> of an M-by-N real matrix A and P-by-N real matrix B:
44: *>
45: *> U**T*A*Q = D1*( 0 R ), V**T*B*Q = D2*( 0 R )
46: *>
47: *> where U, V and Q are orthogonal matrices.
48: *> Let K+L = the effective numerical rank of the matrix (A**T,B**T)**T,
49: *> then R is a K+L-by-K+L nonsingular upper triangular matrix, D1 and
50: *> D2 are M-by-(K+L) and P-by-(K+L) "diagonal" matrices and of the
51: *> following structures, respectively:
52: *>
53: *> If M-K-L >= 0,
54: *>
55: *> K L
56: *> D1 = K ( I 0 )
57: *> L ( 0 C )
58: *> M-K-L ( 0 0 )
59: *>
60: *> K L
61: *> D2 = L ( 0 S )
62: *> P-L ( 0 0 )
63: *>
64: *> N-K-L K L
65: *> ( 0 R ) = K ( 0 R11 R12 )
66: *> L ( 0 0 R22 )
67: *>
68: *> where
69: *>
70: *> C = diag( ALPHA(K+1), ... , ALPHA(K+L) ),
71: *> S = diag( BETA(K+1), ... , BETA(K+L) ),
72: *> C**2 + S**2 = I.
73: *>
74: *> R is stored in A(1:K+L,N-K-L+1:N) on exit.
75: *>
76: *> If M-K-L < 0,
77: *>
78: *> K M-K K+L-M
79: *> D1 = K ( I 0 0 )
80: *> M-K ( 0 C 0 )
81: *>
82: *> K M-K K+L-M
83: *> D2 = M-K ( 0 S 0 )
84: *> K+L-M ( 0 0 I )
85: *> P-L ( 0 0 0 )
86: *>
87: *> N-K-L K M-K K+L-M
88: *> ( 0 R ) = K ( 0 R11 R12 R13 )
89: *> M-K ( 0 0 R22 R23 )
90: *> K+L-M ( 0 0 0 R33 )
91: *>
92: *> where
93: *>
94: *> C = diag( ALPHA(K+1), ... , ALPHA(M) ),
95: *> S = diag( BETA(K+1), ... , BETA(M) ),
96: *> C**2 + S**2 = I.
97: *>
98: *> (R11 R12 R13 ) is stored in A(1:M, N-K-L+1:N), and R33 is stored
99: *> ( 0 R22 R23 )
100: *> in B(M-K+1:L,N+M-K-L+1:N) on exit.
101: *>
102: *> The routine computes C, S, R, and optionally the orthogonal
103: *> transformation matrices U, V and Q.
104: *>
105: *> In particular, if B is an N-by-N nonsingular matrix, then the GSVD of
106: *> A and B implicitly gives the SVD of A*inv(B):
107: *> A*inv(B) = U*(D1*inv(D2))*V**T.
108: *> If ( A**T,B**T)**T has orthonormal columns, then the GSVD of A and B is
109: *> also equal to the CS decomposition of A and B. Furthermore, the GSVD
110: *> can be used to derive the solution of the eigenvalue problem:
111: *> A**T*A x = lambda* B**T*B x.
112: *> In some literature, the GSVD of A and B is presented in the form
113: *> U**T*A*X = ( 0 D1 ), V**T*B*X = ( 0 D2 )
114: *> where U and V are orthogonal and X is nonsingular, D1 and D2 are
115: *> ``diagonal''. The former GSVD form can be converted to the latter
116: *> form by taking the nonsingular matrix X as
117: *>
118: *> X = Q*( I 0 )
119: *> ( 0 inv(R) ).
120: *> \endverbatim
121: *
122: * Arguments:
123: * ==========
124: *
125: *> \param[in] JOBU
126: *> \verbatim
127: *> JOBU is CHARACTER*1
128: *> = 'U': Orthogonal matrix U is computed;
129: *> = 'N': U is not computed.
130: *> \endverbatim
131: *>
132: *> \param[in] JOBV
133: *> \verbatim
134: *> JOBV is CHARACTER*1
135: *> = 'V': Orthogonal matrix V is computed;
136: *> = 'N': V is not computed.
137: *> \endverbatim
138: *>
139: *> \param[in] JOBQ
140: *> \verbatim
141: *> JOBQ is CHARACTER*1
142: *> = 'Q': Orthogonal matrix Q is computed;
143: *> = 'N': Q is not computed.
144: *> \endverbatim
145: *>
146: *> \param[in] M
147: *> \verbatim
148: *> M is INTEGER
149: *> The number of rows of the matrix A. M >= 0.
150: *> \endverbatim
151: *>
152: *> \param[in] N
153: *> \verbatim
154: *> N is INTEGER
155: *> The number of columns of the matrices A and B. N >= 0.
156: *> \endverbatim
157: *>
158: *> \param[in] P
159: *> \verbatim
160: *> P is INTEGER
161: *> The number of rows of the matrix B. P >= 0.
162: *> \endverbatim
163: *>
164: *> \param[out] K
165: *> \verbatim
166: *> K is INTEGER
167: *> \endverbatim
168: *>
169: *> \param[out] L
170: *> \verbatim
171: *> L is INTEGER
172: *>
173: *> On exit, K and L specify the dimension of the subblocks
174: *> described in Purpose.
175: *> K + L = effective numerical rank of (A**T,B**T)**T.
176: *> \endverbatim
177: *>
178: *> \param[in,out] A
179: *> \verbatim
180: *> A is DOUBLE PRECISION array, dimension (LDA,N)
181: *> On entry, the M-by-N matrix A.
182: *> On exit, A contains the triangular matrix R, or part of R.
183: *> See Purpose for details.
184: *> \endverbatim
185: *>
186: *> \param[in] LDA
187: *> \verbatim
188: *> LDA is INTEGER
189: *> The leading dimension of the array A. LDA >= max(1,M).
190: *> \endverbatim
191: *>
192: *> \param[in,out] B
193: *> \verbatim
194: *> B is DOUBLE PRECISION array, dimension (LDB,N)
195: *> On entry, the P-by-N matrix B.
196: *> On exit, B contains the triangular matrix R if M-K-L < 0.
197: *> See Purpose for details.
198: *> \endverbatim
199: *>
200: *> \param[in] LDB
201: *> \verbatim
202: *> LDB is INTEGER
203: *> The leading dimension of the array B. LDB >= max(1,P).
204: *> \endverbatim
205: *>
206: *> \param[out] ALPHA
207: *> \verbatim
208: *> ALPHA is DOUBLE PRECISION array, dimension (N)
209: *> \endverbatim
210: *>
211: *> \param[out] BETA
212: *> \verbatim
213: *> BETA is DOUBLE PRECISION array, dimension (N)
214: *>
215: *> On exit, ALPHA and BETA contain the generalized singular
216: *> value pairs of A and B;
217: *> ALPHA(1:K) = 1,
218: *> BETA(1:K) = 0,
219: *> and if M-K-L >= 0,
220: *> ALPHA(K+1:K+L) = C,
221: *> BETA(K+1:K+L) = S,
222: *> or if M-K-L < 0,
223: *> ALPHA(K+1:M)=C, ALPHA(M+1:K+L)=0
224: *> BETA(K+1:M) =S, BETA(M+1:K+L) =1
225: *> and
226: *> ALPHA(K+L+1:N) = 0
227: *> BETA(K+L+1:N) = 0
228: *> \endverbatim
229: *>
230: *> \param[out] U
231: *> \verbatim
232: *> U is DOUBLE PRECISION array, dimension (LDU,M)
233: *> If JOBU = 'U', U contains the M-by-M orthogonal matrix U.
234: *> If JOBU = 'N', U is not referenced.
235: *> \endverbatim
236: *>
237: *> \param[in] LDU
238: *> \verbatim
239: *> LDU is INTEGER
240: *> The leading dimension of the array U. LDU >= max(1,M) if
241: *> JOBU = 'U'; LDU >= 1 otherwise.
242: *> \endverbatim
243: *>
244: *> \param[out] V
245: *> \verbatim
246: *> V is DOUBLE PRECISION array, dimension (LDV,P)
247: *> If JOBV = 'V', V contains the P-by-P orthogonal matrix V.
248: *> If JOBV = 'N', V is not referenced.
249: *> \endverbatim
250: *>
251: *> \param[in] LDV
252: *> \verbatim
253: *> LDV is INTEGER
254: *> The leading dimension of the array V. LDV >= max(1,P) if
255: *> JOBV = 'V'; LDV >= 1 otherwise.
256: *> \endverbatim
257: *>
258: *> \param[out] Q
259: *> \verbatim
260: *> Q is DOUBLE PRECISION array, dimension (LDQ,N)
261: *> If JOBQ = 'Q', Q contains the N-by-N orthogonal matrix Q.
262: *> If JOBQ = 'N', Q is not referenced.
263: *> \endverbatim
264: *>
265: *> \param[in] LDQ
266: *> \verbatim
267: *> LDQ is INTEGER
268: *> The leading dimension of the array Q. LDQ >= max(1,N) if
269: *> JOBQ = 'Q'; LDQ >= 1 otherwise.
270: *> \endverbatim
271: *>
272: *> \param[out] WORK
273: *> \verbatim
274: *> WORK is DOUBLE PRECISION array,
275: *> dimension (max(3*N,M,P)+N)
276: *> \endverbatim
277: *>
278: *> \param[out] IWORK
279: *> \verbatim
280: *> IWORK is INTEGER array, dimension (N)
281: *> On exit, IWORK stores the sorting information. More
282: *> precisely, the following loop will sort ALPHA
283: *> for I = K+1, min(M,K+L)
284: *> swap ALPHA(I) and ALPHA(IWORK(I))
285: *> endfor
286: *> such that ALPHA(1) >= ALPHA(2) >= ... >= ALPHA(N).
287: *> \endverbatim
288: *>
289: *> \param[out] INFO
290: *> \verbatim
291: *> INFO is INTEGER
292: *> = 0: successful exit
293: *> < 0: if INFO = -i, the i-th argument had an illegal value.
294: *> > 0: if INFO = 1, the Jacobi-type procedure failed to
295: *> converge. For further details, see subroutine DTGSJA.
296: *> \endverbatim
297: *
298: *> \par Internal Parameters:
299: * =========================
300: *>
301: *> \verbatim
302: *> TOLA DOUBLE PRECISION
303: *> TOLB DOUBLE PRECISION
304: *> TOLA and TOLB are the thresholds to determine the effective
305: *> rank of (A',B')**T. Generally, they are set to
306: *> TOLA = MAX(M,N)*norm(A)*MAZHEPS,
307: *> TOLB = MAX(P,N)*norm(B)*MAZHEPS.
308: *> The size of TOLA and TOLB may affect the size of backward
309: *> errors of the decomposition.
310: *> \endverbatim
311: *
312: * Authors:
313: * ========
314: *
315: *> \author Univ. of Tennessee
316: *> \author Univ. of California Berkeley
317: *> \author Univ. of Colorado Denver
318: *> \author NAG Ltd.
319: *
320: *> \date November 2011
321: *
322: *> \ingroup doubleOTHERsing
323: *
324: *> \par Contributors:
325: * ==================
326: *>
327: *> Ming Gu and Huan Ren, Computer Science Division, University of
328: *> California at Berkeley, USA
329: *>
330: * =====================================================================
331: SUBROUTINE DGGSVD( JOBU, JOBV, JOBQ, M, N, P, K, L, A, LDA, B,
332: $ LDB, ALPHA, BETA, U, LDU, V, LDV, Q, LDQ, WORK,
333: $ IWORK, INFO )
334: *
335: * -- LAPACK driver routine (version 3.4.0) --
336: * -- LAPACK is a software package provided by Univ. of Tennessee, --
337: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
338: * November 2011
339: *
340: * .. Scalar Arguments ..
341: CHARACTER JOBQ, JOBU, JOBV
342: INTEGER INFO, K, L, LDA, LDB, LDQ, LDU, LDV, M, N, P
343: * ..
344: * .. Array Arguments ..
345: INTEGER IWORK( * )
346: DOUBLE PRECISION A( LDA, * ), ALPHA( * ), B( LDB, * ),
347: $ BETA( * ), Q( LDQ, * ), U( LDU, * ),
348: $ V( LDV, * ), WORK( * )
349: * ..
350: *
351: * =====================================================================
352: *
353: * .. Local Scalars ..
354: LOGICAL WANTQ, WANTU, WANTV
355: INTEGER I, IBND, ISUB, J, NCYCLE
356: DOUBLE PRECISION ANORM, BNORM, SMAX, TEMP, TOLA, TOLB, ULP, UNFL
357: * ..
358: * .. External Functions ..
359: LOGICAL LSAME
360: DOUBLE PRECISION DLAMCH, DLANGE
361: EXTERNAL LSAME, DLAMCH, DLANGE
362: * ..
363: * .. External Subroutines ..
364: EXTERNAL DCOPY, DGGSVP, DTGSJA, XERBLA
365: * ..
366: * .. Intrinsic Functions ..
367: INTRINSIC MAX, MIN
368: * ..
369: * .. Executable Statements ..
370: *
371: * Test the input parameters
372: *
373: WANTU = LSAME( JOBU, 'U' )
374: WANTV = LSAME( JOBV, 'V' )
375: WANTQ = LSAME( JOBQ, 'Q' )
376: *
377: INFO = 0
378: IF( .NOT.( WANTU .OR. LSAME( JOBU, 'N' ) ) ) THEN
379: INFO = -1
380: ELSE IF( .NOT.( WANTV .OR. LSAME( JOBV, 'N' ) ) ) THEN
381: INFO = -2
382: ELSE IF( .NOT.( WANTQ .OR. LSAME( JOBQ, 'N' ) ) ) THEN
383: INFO = -3
384: ELSE IF( M.LT.0 ) THEN
385: INFO = -4
386: ELSE IF( N.LT.0 ) THEN
387: INFO = -5
388: ELSE IF( P.LT.0 ) THEN
389: INFO = -6
390: ELSE IF( LDA.LT.MAX( 1, M ) ) THEN
391: INFO = -10
392: ELSE IF( LDB.LT.MAX( 1, P ) ) THEN
393: INFO = -12
394: ELSE IF( LDU.LT.1 .OR. ( WANTU .AND. LDU.LT.M ) ) THEN
395: INFO = -16
396: ELSE IF( LDV.LT.1 .OR. ( WANTV .AND. LDV.LT.P ) ) THEN
397: INFO = -18
398: ELSE IF( LDQ.LT.1 .OR. ( WANTQ .AND. LDQ.LT.N ) ) THEN
399: INFO = -20
400: END IF
401: IF( INFO.NE.0 ) THEN
402: CALL XERBLA( 'DGGSVD', -INFO )
403: RETURN
404: END IF
405: *
406: * Compute the Frobenius norm of matrices A and B
407: *
408: ANORM = DLANGE( '1', M, N, A, LDA, WORK )
409: BNORM = DLANGE( '1', P, N, B, LDB, WORK )
410: *
411: * Get machine precision and set up threshold for determining
412: * the effective numerical rank of the matrices A and B.
413: *
414: ULP = DLAMCH( 'Precision' )
415: UNFL = DLAMCH( 'Safe Minimum' )
416: TOLA = MAX( M, N )*MAX( ANORM, UNFL )*ULP
417: TOLB = MAX( P, N )*MAX( BNORM, UNFL )*ULP
418: *
419: * Preprocessing
420: *
421: CALL DGGSVP( JOBU, JOBV, JOBQ, M, P, N, A, LDA, B, LDB, TOLA,
422: $ TOLB, K, L, U, LDU, V, LDV, Q, LDQ, IWORK, WORK,
423: $ WORK( N+1 ), INFO )
424: *
425: * Compute the GSVD of two upper "triangular" matrices
426: *
427: CALL DTGSJA( JOBU, JOBV, JOBQ, M, P, N, K, L, A, LDA, B, LDB,
428: $ TOLA, TOLB, ALPHA, BETA, U, LDU, V, LDV, Q, LDQ,
429: $ WORK, NCYCLE, INFO )
430: *
431: * Sort the singular values and store the pivot indices in IWORK
432: * Copy ALPHA to WORK, then sort ALPHA in WORK
433: *
434: CALL DCOPY( N, ALPHA, 1, WORK, 1 )
435: IBND = MIN( L, M-K )
436: DO 20 I = 1, IBND
437: *
438: * Scan for largest ALPHA(K+I)
439: *
440: ISUB = I
441: SMAX = WORK( K+I )
442: DO 10 J = I + 1, IBND
443: TEMP = WORK( K+J )
444: IF( TEMP.GT.SMAX ) THEN
445: ISUB = J
446: SMAX = TEMP
447: END IF
448: 10 CONTINUE
449: IF( ISUB.NE.I ) THEN
450: WORK( K+ISUB ) = WORK( K+I )
451: WORK( K+I ) = SMAX
452: IWORK( K+I ) = K + ISUB
453: ELSE
454: IWORK( K+I ) = K + I
455: END IF
456: 20 CONTINUE
457: *
458: RETURN
459: *
460: * End of DGGSVD
461: *
462: END
CVSweb interface <joel.bertrand@systella.fr>