Annotation of rpl/lapack/lapack/ztgsja.f, revision 1.1.1.1
1.1 bertrand 1: SUBROUTINE ZTGSJA( JOBU, JOBV, JOBQ, M, P, N, K, L, A, LDA, B,
2: $ LDB, TOLA, TOLB, ALPHA, BETA, U, LDU, V, LDV,
3: $ Q, LDQ, WORK, NCYCLE, INFO )
4: *
5: * -- LAPACK routine (version 3.2.1) --
6: * -- LAPACK is a software package provided by Univ. of Tennessee, --
7: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
8: * -- April 2009 --
9: *
10: * .. Scalar Arguments ..
11: CHARACTER JOBQ, JOBU, JOBV
12: INTEGER INFO, K, L, LDA, LDB, LDQ, LDU, LDV, M, N,
13: $ NCYCLE, P
14: DOUBLE PRECISION TOLA, TOLB
15: * ..
16: * .. Array Arguments ..
17: DOUBLE PRECISION ALPHA( * ), BETA( * )
18: COMPLEX*16 A( LDA, * ), B( LDB, * ), Q( LDQ, * ),
19: $ U( LDU, * ), V( LDV, * ), WORK( * )
20: * ..
21: *
22: * Purpose
23: * =======
24: *
25: * ZTGSJA computes the generalized singular value decomposition (GSVD)
26: * of two complex upper triangular (or trapezoidal) matrices A and B.
27: *
28: * On entry, it is assumed that matrices A and B have the following
29: * forms, which may be obtained by the preprocessing subroutine ZGGSVP
30: * from a general M-by-N matrix A and P-by-N matrix B:
31: *
32: * N-K-L K L
33: * A = K ( 0 A12 A13 ) if M-K-L >= 0;
34: * L ( 0 0 A23 )
35: * M-K-L ( 0 0 0 )
36: *
37: * N-K-L K L
38: * A = K ( 0 A12 A13 ) if M-K-L < 0;
39: * M-K ( 0 0 A23 )
40: *
41: * N-K-L K L
42: * B = L ( 0 0 B13 )
43: * P-L ( 0 0 0 )
44: *
45: * where the K-by-K matrix A12 and L-by-L matrix B13 are nonsingular
46: * upper triangular; A23 is L-by-L upper triangular if M-K-L >= 0,
47: * otherwise A23 is (M-K)-by-L upper trapezoidal.
48: *
49: * On exit,
50: *
51: * U'*A*Q = D1*( 0 R ), V'*B*Q = D2*( 0 R ),
52: *
53: * where U, V and Q are unitary matrices, Z' denotes the conjugate
54: * transpose of Z, R is a nonsingular upper triangular matrix, and D1
55: * and D2 are ``diagonal'' matrices, which are of the following
56: * structures:
57: *
58: * If M-K-L >= 0,
59: *
60: * K L
61: * D1 = K ( I 0 )
62: * L ( 0 C )
63: * M-K-L ( 0 0 )
64: *
65: * K L
66: * D2 = L ( 0 S )
67: * P-L ( 0 0 )
68: *
69: * N-K-L K L
70: * ( 0 R ) = K ( 0 R11 R12 ) K
71: * L ( 0 0 R22 ) L
72: *
73: * where
74: *
75: * C = diag( ALPHA(K+1), ... , ALPHA(K+L) ),
76: * S = diag( BETA(K+1), ... , BETA(K+L) ),
77: * C**2 + S**2 = I.
78: *
79: * R is stored in A(1:K+L,N-K-L+1:N) on exit.
80: *
81: * If M-K-L < 0,
82: *
83: * K M-K K+L-M
84: * D1 = K ( I 0 0 )
85: * M-K ( 0 C 0 )
86: *
87: * K M-K K+L-M
88: * D2 = M-K ( 0 S 0 )
89: * K+L-M ( 0 0 I )
90: * P-L ( 0 0 0 )
91: *
92: * N-K-L K M-K K+L-M
93: * ( 0 R ) = K ( 0 R11 R12 R13 )
94: * M-K ( 0 0 R22 R23 )
95: * K+L-M ( 0 0 0 R33 )
96: *
97: * where
98: * C = diag( ALPHA(K+1), ... , ALPHA(M) ),
99: * S = diag( BETA(K+1), ... , BETA(M) ),
100: * C**2 + S**2 = I.
101: *
102: * R = ( R11 R12 R13 ) is stored in A(1:M, N-K-L+1:N) and R33 is stored
103: * ( 0 R22 R23 )
104: * in B(M-K+1:L,N+M-K-L+1:N) on exit.
105: *
106: * The computation of the unitary transformation matrices U, V or Q
107: * is optional. These matrices may either be formed explicitly, or they
108: * may be postmultiplied into input matrices U1, V1, or Q1.
109: *
110: * Arguments
111: * =========
112: *
113: * JOBU (input) CHARACTER*1
114: * = 'U': U must contain a unitary matrix U1 on entry, and
115: * the product U1*U is returned;
116: * = 'I': U is initialized to the unit matrix, and the
117: * unitary matrix U is returned;
118: * = 'N': U is not computed.
119: *
120: * JOBV (input) CHARACTER*1
121: * = 'V': V must contain a unitary matrix V1 on entry, and
122: * the product V1*V is returned;
123: * = 'I': V is initialized to the unit matrix, and the
124: * unitary matrix V is returned;
125: * = 'N': V is not computed.
126: *
127: * JOBQ (input) CHARACTER*1
128: * = 'Q': Q must contain a unitary matrix Q1 on entry, and
129: * the product Q1*Q is returned;
130: * = 'I': Q is initialized to the unit matrix, and the
131: * unitary matrix Q is returned;
132: * = 'N': Q is not computed.
133: *
134: * M (input) INTEGER
135: * The number of rows of the matrix A. M >= 0.
136: *
137: * P (input) INTEGER
138: * The number of rows of the matrix B. P >= 0.
139: *
140: * N (input) INTEGER
141: * The number of columns of the matrices A and B. N >= 0.
142: *
143: * K (input) INTEGER
144: * L (input) INTEGER
145: * K and L specify the subblocks in the input matrices A and B:
146: * A23 = A(K+1:MIN(K+L,M),N-L+1:N) and B13 = B(1:L,,N-L+1:N)
147: * of A and B, whose GSVD is going to be computed by ZTGSJA.
148: * See Further Details.
149: *
150: * A (input/output) COMPLEX*16 array, dimension (LDA,N)
151: * On entry, the M-by-N matrix A.
152: * On exit, A(N-K+1:N,1:MIN(K+L,M) ) contains the triangular
153: * matrix R or part of R. See Purpose for details.
154: *
155: * LDA (input) INTEGER
156: * The leading dimension of the array A. LDA >= max(1,M).
157: *
158: * B (input/output) COMPLEX*16 array, dimension (LDB,N)
159: * On entry, the P-by-N matrix B.
160: * On exit, if necessary, B(M-K+1:L,N+M-K-L+1:N) contains
161: * a part of R. See Purpose for details.
162: *
163: * LDB (input) INTEGER
164: * The leading dimension of the array B. LDB >= max(1,P).
165: *
166: * TOLA (input) DOUBLE PRECISION
167: * TOLB (input) DOUBLE PRECISION
168: * TOLA and TOLB are the convergence criteria for the Jacobi-
169: * Kogbetliantz iteration procedure. Generally, they are the
170: * same as used in the preprocessing step, say
171: * TOLA = MAX(M,N)*norm(A)*MAZHEPS,
172: * TOLB = MAX(P,N)*norm(B)*MAZHEPS.
173: *
174: * ALPHA (output) DOUBLE PRECISION array, dimension (N)
175: * BETA (output) DOUBLE PRECISION array, dimension (N)
176: * On exit, ALPHA and BETA contain the generalized singular
177: * value pairs of A and B;
178: * ALPHA(1:K) = 1,
179: * BETA(1:K) = 0,
180: * and if M-K-L >= 0,
181: * ALPHA(K+1:K+L) = diag(C),
182: * BETA(K+1:K+L) = diag(S),
183: * or if M-K-L < 0,
184: * ALPHA(K+1:M)= C, ALPHA(M+1:K+L)= 0
185: * BETA(K+1:M) = S, BETA(M+1:K+L) = 1.
186: * Furthermore, if K+L < N,
187: * ALPHA(K+L+1:N) = 0
188: * BETA(K+L+1:N) = 0.
189: *
190: * U (input/output) COMPLEX*16 array, dimension (LDU,M)
191: * On entry, if JOBU = 'U', U must contain a matrix U1 (usually
192: * the unitary matrix returned by ZGGSVP).
193: * On exit,
194: * if JOBU = 'I', U contains the unitary matrix U;
195: * if JOBU = 'U', U contains the product U1*U.
196: * If JOBU = 'N', U is not referenced.
197: *
198: * LDU (input) INTEGER
199: * The leading dimension of the array U. LDU >= max(1,M) if
200: * JOBU = 'U'; LDU >= 1 otherwise.
201: *
202: * V (input/output) COMPLEX*16 array, dimension (LDV,P)
203: * On entry, if JOBV = 'V', V must contain a matrix V1 (usually
204: * the unitary matrix returned by ZGGSVP).
205: * On exit,
206: * if JOBV = 'I', V contains the unitary matrix V;
207: * if JOBV = 'V', V contains the product V1*V.
208: * If JOBV = 'N', V is not referenced.
209: *
210: * LDV (input) INTEGER
211: * The leading dimension of the array V. LDV >= max(1,P) if
212: * JOBV = 'V'; LDV >= 1 otherwise.
213: *
214: * Q (input/output) COMPLEX*16 array, dimension (LDQ,N)
215: * On entry, if JOBQ = 'Q', Q must contain a matrix Q1 (usually
216: * the unitary matrix returned by ZGGSVP).
217: * On exit,
218: * if JOBQ = 'I', Q contains the unitary matrix Q;
219: * if JOBQ = 'Q', Q contains the product Q1*Q.
220: * If JOBQ = 'N', Q is not referenced.
221: *
222: * LDQ (input) INTEGER
223: * The leading dimension of the array Q. LDQ >= max(1,N) if
224: * JOBQ = 'Q'; LDQ >= 1 otherwise.
225: *
226: * WORK (workspace) COMPLEX*16 array, dimension (2*N)
227: *
228: * NCYCLE (output) INTEGER
229: * The number of cycles required for convergence.
230: *
231: * INFO (output) INTEGER
232: * = 0: successful exit
233: * < 0: if INFO = -i, the i-th argument had an illegal value.
234: * = 1: the procedure does not converge after MAXIT cycles.
235: *
236: * Internal Parameters
237: * ===================
238: *
239: * MAXIT INTEGER
240: * MAXIT specifies the total loops that the iterative procedure
241: * may take. If after MAXIT cycles, the routine fails to
242: * converge, we return INFO = 1.
243: *
244: * Further Details
245: * ===============
246: *
247: * ZTGSJA essentially uses a variant of Kogbetliantz algorithm to reduce
248: * min(L,M-K)-by-L triangular (or trapezoidal) matrix A23 and L-by-L
249: * matrix B13 to the form:
250: *
251: * U1'*A13*Q1 = C1*R1; V1'*B13*Q1 = S1*R1,
252: *
253: * where U1, V1 and Q1 are unitary matrix, and Z' is the conjugate
254: * transpose of Z. C1 and S1 are diagonal matrices satisfying
255: *
256: * C1**2 + S1**2 = I,
257: *
258: * and R1 is an L-by-L nonsingular upper triangular matrix.
259: *
260: * =====================================================================
261: *
262: * .. Parameters ..
263: INTEGER MAXIT
264: PARAMETER ( MAXIT = 40 )
265: DOUBLE PRECISION ZERO, ONE
266: PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 )
267: COMPLEX*16 CZERO, CONE
268: PARAMETER ( CZERO = ( 0.0D+0, 0.0D+0 ),
269: $ CONE = ( 1.0D+0, 0.0D+0 ) )
270: * ..
271: * .. Local Scalars ..
272: *
273: LOGICAL INITQ, INITU, INITV, UPPER, WANTQ, WANTU, WANTV
274: INTEGER I, J, KCYCLE
275: DOUBLE PRECISION A1, A3, B1, B3, CSQ, CSU, CSV, ERROR, GAMMA,
276: $ RWK, SSMIN
277: COMPLEX*16 A2, B2, SNQ, SNU, SNV
278: * ..
279: * .. External Functions ..
280: LOGICAL LSAME
281: EXTERNAL LSAME
282: * ..
283: * .. External Subroutines ..
284: EXTERNAL DLARTG, XERBLA, ZCOPY, ZDSCAL, ZLAGS2, ZLAPLL,
285: $ ZLASET, ZROT
286: * ..
287: * .. Intrinsic Functions ..
288: INTRINSIC ABS, DBLE, DCONJG, MAX, MIN
289: * ..
290: * .. Executable Statements ..
291: *
292: * Decode and test the input parameters
293: *
294: INITU = LSAME( JOBU, 'I' )
295: WANTU = INITU .OR. LSAME( JOBU, 'U' )
296: *
297: INITV = LSAME( JOBV, 'I' )
298: WANTV = INITV .OR. LSAME( JOBV, 'V' )
299: *
300: INITQ = LSAME( JOBQ, 'I' )
301: WANTQ = INITQ .OR. LSAME( JOBQ, 'Q' )
302: *
303: INFO = 0
304: IF( .NOT.( INITU .OR. WANTU .OR. LSAME( JOBU, 'N' ) ) ) THEN
305: INFO = -1
306: ELSE IF( .NOT.( INITV .OR. WANTV .OR. LSAME( JOBV, 'N' ) ) ) THEN
307: INFO = -2
308: ELSE IF( .NOT.( INITQ .OR. WANTQ .OR. LSAME( JOBQ, 'N' ) ) ) THEN
309: INFO = -3
310: ELSE IF( M.LT.0 ) THEN
311: INFO = -4
312: ELSE IF( P.LT.0 ) THEN
313: INFO = -5
314: ELSE IF( N.LT.0 ) THEN
315: INFO = -6
316: ELSE IF( LDA.LT.MAX( 1, M ) ) THEN
317: INFO = -10
318: ELSE IF( LDB.LT.MAX( 1, P ) ) THEN
319: INFO = -12
320: ELSE IF( LDU.LT.1 .OR. ( WANTU .AND. LDU.LT.M ) ) THEN
321: INFO = -18
322: ELSE IF( LDV.LT.1 .OR. ( WANTV .AND. LDV.LT.P ) ) THEN
323: INFO = -20
324: ELSE IF( LDQ.LT.1 .OR. ( WANTQ .AND. LDQ.LT.N ) ) THEN
325: INFO = -22
326: END IF
327: IF( INFO.NE.0 ) THEN
328: CALL XERBLA( 'ZTGSJA', -INFO )
329: RETURN
330: END IF
331: *
332: * Initialize U, V and Q, if necessary
333: *
334: IF( INITU )
335: $ CALL ZLASET( 'Full', M, M, CZERO, CONE, U, LDU )
336: IF( INITV )
337: $ CALL ZLASET( 'Full', P, P, CZERO, CONE, V, LDV )
338: IF( INITQ )
339: $ CALL ZLASET( 'Full', N, N, CZERO, CONE, Q, LDQ )
340: *
341: * Loop until convergence
342: *
343: UPPER = .FALSE.
344: DO 40 KCYCLE = 1, MAXIT
345: *
346: UPPER = .NOT.UPPER
347: *
348: DO 20 I = 1, L - 1
349: DO 10 J = I + 1, L
350: *
351: A1 = ZERO
352: A2 = CZERO
353: A3 = ZERO
354: IF( K+I.LE.M )
355: $ A1 = DBLE( A( K+I, N-L+I ) )
356: IF( K+J.LE.M )
357: $ A3 = DBLE( A( K+J, N-L+J ) )
358: *
359: B1 = DBLE( B( I, N-L+I ) )
360: B3 = DBLE( B( J, N-L+J ) )
361: *
362: IF( UPPER ) THEN
363: IF( K+I.LE.M )
364: $ A2 = A( K+I, N-L+J )
365: B2 = B( I, N-L+J )
366: ELSE
367: IF( K+J.LE.M )
368: $ A2 = A( K+J, N-L+I )
369: B2 = B( J, N-L+I )
370: END IF
371: *
372: CALL ZLAGS2( UPPER, A1, A2, A3, B1, B2, B3, CSU, SNU,
373: $ CSV, SNV, CSQ, SNQ )
374: *
375: * Update (K+I)-th and (K+J)-th rows of matrix A: U'*A
376: *
377: IF( K+J.LE.M )
378: $ CALL ZROT( L, A( K+J, N-L+1 ), LDA, A( K+I, N-L+1 ),
379: $ LDA, CSU, DCONJG( SNU ) )
380: *
381: * Update I-th and J-th rows of matrix B: V'*B
382: *
383: CALL ZROT( L, B( J, N-L+1 ), LDB, B( I, N-L+1 ), LDB,
384: $ CSV, DCONJG( SNV ) )
385: *
386: * Update (N-L+I)-th and (N-L+J)-th columns of matrices
387: * A and B: A*Q and B*Q
388: *
389: CALL ZROT( MIN( K+L, M ), A( 1, N-L+J ), 1,
390: $ A( 1, N-L+I ), 1, CSQ, SNQ )
391: *
392: CALL ZROT( L, B( 1, N-L+J ), 1, B( 1, N-L+I ), 1, CSQ,
393: $ SNQ )
394: *
395: IF( UPPER ) THEN
396: IF( K+I.LE.M )
397: $ A( K+I, N-L+J ) = CZERO
398: B( I, N-L+J ) = CZERO
399: ELSE
400: IF( K+J.LE.M )
401: $ A( K+J, N-L+I ) = CZERO
402: B( J, N-L+I ) = CZERO
403: END IF
404: *
405: * Ensure that the diagonal elements of A and B are real.
406: *
407: IF( K+I.LE.M )
408: $ A( K+I, N-L+I ) = DBLE( A( K+I, N-L+I ) )
409: IF( K+J.LE.M )
410: $ A( K+J, N-L+J ) = DBLE( A( K+J, N-L+J ) )
411: B( I, N-L+I ) = DBLE( B( I, N-L+I ) )
412: B( J, N-L+J ) = DBLE( B( J, N-L+J ) )
413: *
414: * Update unitary matrices U, V, Q, if desired.
415: *
416: IF( WANTU .AND. K+J.LE.M )
417: $ CALL ZROT( M, U( 1, K+J ), 1, U( 1, K+I ), 1, CSU,
418: $ SNU )
419: *
420: IF( WANTV )
421: $ CALL ZROT( P, V( 1, J ), 1, V( 1, I ), 1, CSV, SNV )
422: *
423: IF( WANTQ )
424: $ CALL ZROT( N, Q( 1, N-L+J ), 1, Q( 1, N-L+I ), 1, CSQ,
425: $ SNQ )
426: *
427: 10 CONTINUE
428: 20 CONTINUE
429: *
430: IF( .NOT.UPPER ) THEN
431: *
432: * The matrices A13 and B13 were lower triangular at the start
433: * of the cycle, and are now upper triangular.
434: *
435: * Convergence test: test the parallelism of the corresponding
436: * rows of A and B.
437: *
438: ERROR = ZERO
439: DO 30 I = 1, MIN( L, M-K )
440: CALL ZCOPY( L-I+1, A( K+I, N-L+I ), LDA, WORK, 1 )
441: CALL ZCOPY( L-I+1, B( I, N-L+I ), LDB, WORK( L+1 ), 1 )
442: CALL ZLAPLL( L-I+1, WORK, 1, WORK( L+1 ), 1, SSMIN )
443: ERROR = MAX( ERROR, SSMIN )
444: 30 CONTINUE
445: *
446: IF( ABS( ERROR ).LE.MIN( TOLA, TOLB ) )
447: $ GO TO 50
448: END IF
449: *
450: * End of cycle loop
451: *
452: 40 CONTINUE
453: *
454: * The algorithm has not converged after MAXIT cycles.
455: *
456: INFO = 1
457: GO TO 100
458: *
459: 50 CONTINUE
460: *
461: * If ERROR <= MIN(TOLA,TOLB), then the algorithm has converged.
462: * Compute the generalized singular value pairs (ALPHA, BETA), and
463: * set the triangular matrix R to array A.
464: *
465: DO 60 I = 1, K
466: ALPHA( I ) = ONE
467: BETA( I ) = ZERO
468: 60 CONTINUE
469: *
470: DO 70 I = 1, MIN( L, M-K )
471: *
472: A1 = DBLE( A( K+I, N-L+I ) )
473: B1 = DBLE( B( I, N-L+I ) )
474: *
475: IF( A1.NE.ZERO ) THEN
476: GAMMA = B1 / A1
477: *
478: IF( GAMMA.LT.ZERO ) THEN
479: CALL ZDSCAL( L-I+1, -ONE, B( I, N-L+I ), LDB )
480: IF( WANTV )
481: $ CALL ZDSCAL( P, -ONE, V( 1, I ), 1 )
482: END IF
483: *
484: CALL DLARTG( ABS( GAMMA ), ONE, BETA( K+I ), ALPHA( K+I ),
485: $ RWK )
486: *
487: IF( ALPHA( K+I ).GE.BETA( K+I ) ) THEN
488: CALL ZDSCAL( L-I+1, ONE / ALPHA( K+I ), A( K+I, N-L+I ),
489: $ LDA )
490: ELSE
491: CALL ZDSCAL( L-I+1, ONE / BETA( K+I ), B( I, N-L+I ),
492: $ LDB )
493: CALL ZCOPY( L-I+1, B( I, N-L+I ), LDB, A( K+I, N-L+I ),
494: $ LDA )
495: END IF
496: *
497: ELSE
498: ALPHA( K+I ) = ZERO
499: BETA( K+I ) = ONE
500: CALL ZCOPY( L-I+1, B( I, N-L+I ), LDB, A( K+I, N-L+I ),
501: $ LDA )
502: END IF
503: 70 CONTINUE
504: *
505: * Post-assignment
506: *
507: DO 80 I = M + 1, K + L
508: ALPHA( I ) = ZERO
509: BETA( I ) = ONE
510: 80 CONTINUE
511: *
512: IF( K+L.LT.N ) THEN
513: DO 90 I = K + L + 1, N
514: ALPHA( I ) = ZERO
515: BETA( I ) = ZERO
516: 90 CONTINUE
517: END IF
518: *
519: 100 CONTINUE
520: NCYCLE = KCYCLE
521: *
522: RETURN
523: *
524: * End of ZTGSJA
525: *
526: END
CVSweb interface <joel.bertrand@systella.fr>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to ztgsja.f CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [local] / rpl / lapack / lapack