3 * =========== DOCUMENTATION ===========
5 * Online html documentation available at
6 * http://www.netlib.org/lapack/explore-html/
9 *> Download DSTEDC + dependencies
10 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dstedc.f">
12 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dstedc.f">
14 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dstedc.f">
21 * SUBROUTINE DSTEDC( COMPZ, N, D, E, Z, LDZ, WORK, LWORK, IWORK,
24 * .. Scalar Arguments ..
26 * INTEGER INFO, LDZ, LIWORK, LWORK, N
28 * .. Array Arguments ..
30 * DOUBLE PRECISION D( * ), E( * ), WORK( * ), Z( LDZ, * )
39 *> DSTEDC computes all eigenvalues and, optionally, eigenvectors of a
40 *> symmetric tridiagonal matrix using the divide and conquer method.
41 *> The eigenvectors of a full or band real symmetric matrix can also be
42 *> found if DSYTRD or DSPTRD or DSBTRD has been used to reduce this
43 *> matrix to tridiagonal form.
45 *> This code makes very mild assumptions about floating point
46 *> arithmetic. It will work on machines with a guard digit in
47 *> add/subtract, or on those binary machines without guard digits
48 *> which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2.
49 *> It could conceivably fail on hexadecimal or decimal machines
50 *> without guard digits, but we know of none. See DLAED3 for details.
58 *> COMPZ is CHARACTER*1
59 *> = 'N': Compute eigenvalues only.
60 *> = 'I': Compute eigenvectors of tridiagonal matrix also.
61 *> = 'V': Compute eigenvectors of original dense symmetric
62 *> matrix also. On entry, Z contains the orthogonal
63 *> matrix used to reduce the original matrix to
70 *> The dimension of the symmetric tridiagonal matrix. N >= 0.
75 *> D is DOUBLE PRECISION array, dimension (N)
76 *> On entry, the diagonal elements of the tridiagonal matrix.
77 *> On exit, if INFO = 0, the eigenvalues in ascending order.
82 *> E is DOUBLE PRECISION array, dimension (N-1)
83 *> On entry, the subdiagonal elements of the tridiagonal matrix.
84 *> On exit, E has been destroyed.
89 *> Z is DOUBLE PRECISION array, dimension (LDZ,N)
90 *> On entry, if COMPZ = 'V', then Z contains the orthogonal
91 *> matrix used in the reduction to tridiagonal form.
92 *> On exit, if INFO = 0, then if COMPZ = 'V', Z contains the
93 *> orthonormal eigenvectors of the original symmetric matrix,
94 *> and if COMPZ = 'I', Z contains the orthonormal eigenvectors
95 *> of the symmetric tridiagonal matrix.
96 *> If COMPZ = 'N', then Z is not referenced.
102 *> The leading dimension of the array Z. LDZ >= 1.
103 *> If eigenvectors are desired, then LDZ >= max(1,N).
108 *> WORK is DOUBLE PRECISION array,
110 *> On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
116 *> The dimension of the array WORK.
117 *> If COMPZ = 'N' or N <= 1 then LWORK must be at least 1.
118 *> If COMPZ = 'V' and N > 1 then LWORK must be at least
119 *> ( 1 + 3*N + 2*N*lg N + 4*N**2 ),
120 *> where lg( N ) = smallest integer k such
122 *> If COMPZ = 'I' and N > 1 then LWORK must be at least
123 *> ( 1 + 4*N + N**2 ).
124 *> Note that for COMPZ = 'I' or 'V', then if N is less than or
125 *> equal to the minimum divide size, usually 25, then LWORK need
126 *> only be max(1,2*(N-1)).
128 *> If LWORK = -1, then a workspace query is assumed; the routine
129 *> only calculates the optimal size of the WORK array, returns
130 *> this value as the first entry of the WORK array, and no error
131 *> message related to LWORK is issued by XERBLA.
136 *> IWORK is INTEGER array, dimension (MAX(1,LIWORK))
137 *> On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK.
143 *> The dimension of the array IWORK.
144 *> If COMPZ = 'N' or N <= 1 then LIWORK must be at least 1.
145 *> If COMPZ = 'V' and N > 1 then LIWORK must be at least
146 *> ( 6 + 6*N + 5*N*lg N ).
147 *> If COMPZ = 'I' and N > 1 then LIWORK must be at least
149 *> Note that for COMPZ = 'I' or 'V', then if N is less than or
150 *> equal to the minimum divide size, usually 25, then LIWORK
153 *> If LIWORK = -1, then a workspace query is assumed; the
154 *> routine only calculates the optimal size of the IWORK array,
155 *> returns this value as the first entry of the IWORK array, and
156 *> no error message related to LIWORK is issued by XERBLA.
162 *> = 0: successful exit.
163 *> < 0: if INFO = -i, the i-th argument had an illegal value.
164 *> > 0: The algorithm failed to compute an eigenvalue while
165 *> working on the submatrix lying in rows and columns
166 *> INFO/(N+1) through mod(INFO,N+1).
172 *> \author Univ. of Tennessee
173 *> \author Univ. of California Berkeley
174 *> \author Univ. of Colorado Denver
177 *> \date November 2015
179 *> \ingroup auxOTHERcomputational
181 *> \par Contributors:
184 *> Jeff Rutter, Computer Science Division, University of California
185 *> at Berkeley, USA \n
186 *> Modified by Francoise Tisseur, University of Tennessee
188 * =====================================================================
189 SUBROUTINE DSTEDC( COMPZ, N, D, E, Z, LDZ, WORK, LWORK, IWORK,
192 * -- LAPACK computational routine (version 3.6.0) --
193 * -- LAPACK is a software package provided by Univ. of Tennessee, --
194 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
197 * .. Scalar Arguments ..
199 INTEGER INFO, LDZ, LIWORK, LWORK, N
201 * .. Array Arguments ..
203 DOUBLE PRECISION D( * ), E( * ), WORK( * ), Z( LDZ, * )
206 * =====================================================================
209 DOUBLE PRECISION ZERO, ONE, TWO
210 PARAMETER ( ZERO = 0.0D0, ONE = 1.0D0, TWO = 2.0D0 )
212 * .. Local Scalars ..
214 INTEGER FINISH, I, ICOMPZ, II, J, K, LGN, LIWMIN,
215 $ LWMIN, M, SMLSIZ, START, STOREZ, STRTRW
216 DOUBLE PRECISION EPS, ORGNRM, P, TINY
218 * .. External Functions ..
221 DOUBLE PRECISION DLAMCH, DLANST
222 EXTERNAL LSAME, ILAENV, DLAMCH, DLANST
224 * .. External Subroutines ..
225 EXTERNAL DGEMM, DLACPY, DLAED0, DLASCL, DLASET, DLASRT,
226 $ DSTEQR, DSTERF, DSWAP, XERBLA
228 * .. Intrinsic Functions ..
229 INTRINSIC ABS, DBLE, INT, LOG, MAX, MOD, SQRT
231 * .. Executable Statements ..
233 * Test the input parameters.
236 LQUERY = ( LWORK.EQ.-1 .OR. LIWORK.EQ.-1 )
238 IF( LSAME( COMPZ, 'N' ) ) THEN
240 ELSE IF( LSAME( COMPZ, 'V' ) ) THEN
242 ELSE IF( LSAME( COMPZ, 'I' ) ) THEN
247 IF( ICOMPZ.LT.0 ) THEN
249 ELSE IF( N.LT.0 ) THEN
251 ELSE IF( ( LDZ.LT.1 ) .OR.
252 $ ( ICOMPZ.GT.0 .AND. LDZ.LT.MAX( 1, N ) ) ) THEN
258 * Compute the workspace requirements
260 SMLSIZ = ILAENV( 9, 'DSTEDC', ' ', 0, 0, 0, 0 )
261 IF( N.LE.1 .OR. ICOMPZ.EQ.0 ) THEN
264 ELSE IF( N.LE.SMLSIZ ) THEN
268 LGN = INT( LOG( DBLE( N ) )/LOG( TWO ) )
273 IF( ICOMPZ.EQ.1 ) THEN
274 LWMIN = 1 + 3*N + 2*N*LGN + 4*N**2
275 LIWMIN = 6 + 6*N + 5*N*LGN
276 ELSE IF( ICOMPZ.EQ.2 ) THEN
277 LWMIN = 1 + 4*N + N**2
284 IF( LWORK.LT.LWMIN .AND. .NOT. LQUERY ) THEN
286 ELSE IF( LIWORK.LT.LIWMIN .AND. .NOT. LQUERY ) THEN
292 CALL XERBLA( 'DSTEDC', -INFO )
294 ELSE IF (LQUERY) THEN
298 * Quick return if possible
308 * If the following conditional clause is removed, then the routine
309 * will use the Divide and Conquer routine to compute only the
310 * eigenvalues, which requires (3N + 3N**2) real workspace and
311 * (2 + 5N + 2N lg(N)) integer workspace.
312 * Since on many architectures DSTERF is much faster than any other
313 * algorithm for finding eigenvalues only, it is used here
314 * as the default. If the conditional clause is removed, then
315 * information on the size of workspace needs to be changed.
317 * If COMPZ = 'N', use DSTERF to compute the eigenvalues.
319 IF( ICOMPZ.EQ.0 ) THEN
320 CALL DSTERF( N, D, E, INFO )
324 * If N is smaller than the minimum divide size (SMLSIZ+1), then
325 * solve the problem with another solver.
327 IF( N.LE.SMLSIZ ) THEN
329 CALL DSTEQR( COMPZ, N, D, E, Z, LDZ, WORK, INFO )
333 * If COMPZ = 'V', the Z matrix must be stored elsewhere for later
336 IF( ICOMPZ.EQ.1 ) THEN
342 IF( ICOMPZ.EQ.2 ) THEN
343 CALL DLASET( 'Full', N, N, ZERO, ONE, Z, LDZ )
348 ORGNRM = DLANST( 'M', N, D, E )
352 EPS = DLAMCH( 'Epsilon' )
356 * while ( START <= N )
359 IF( START.LE.N ) THEN
361 * Let FINISH be the position of the next subdiagonal entry
362 * such that E( FINISH ) <= TINY or FINISH = N if no such
363 * subdiagonal exists. The matrix identified by the elements
364 * between START and FINISH constitutes an independent
369 IF( FINISH.LT.N ) THEN
370 TINY = EPS*SQRT( ABS( D( FINISH ) ) )*
371 $ SQRT( ABS( D( FINISH+1 ) ) )
372 IF( ABS( E( FINISH ) ).GT.TINY ) THEN
378 * (Sub) Problem determined. Compute its size and solve it.
380 M = FINISH - START + 1
385 IF( M.GT.SMLSIZ ) THEN
389 ORGNRM = DLANST( 'M', M, D( START ), E( START ) )
390 CALL DLASCL( 'G', 0, 0, ORGNRM, ONE, M, 1, D( START ), M,
392 CALL DLASCL( 'G', 0, 0, ORGNRM, ONE, M-1, 1, E( START ),
395 IF( ICOMPZ.EQ.1 ) THEN
400 CALL DLAED0( ICOMPZ, N, M, D( START ), E( START ),
401 $ Z( STRTRW, START ), LDZ, WORK( 1 ), N,
402 $ WORK( STOREZ ), IWORK, INFO )
404 INFO = ( INFO / ( M+1 )+START-1 )*( N+1 ) +
405 $ MOD( INFO, ( M+1 ) ) + START - 1
411 CALL DLASCL( 'G', 0, 0, ONE, ORGNRM, M, 1, D( START ), M,
415 IF( ICOMPZ.EQ.1 ) THEN
417 * Since QR won't update a Z matrix which is larger than
418 * the length of D, we must solve the sub-problem in a
419 * workspace and then multiply back into Z.
421 CALL DSTEQR( 'I', M, D( START ), E( START ), WORK, M,
422 $ WORK( M*M+1 ), INFO )
423 CALL DLACPY( 'A', N, M, Z( 1, START ), LDZ,
424 $ WORK( STOREZ ), N )
425 CALL DGEMM( 'N', 'N', N, M, M, ONE,
426 $ WORK( STOREZ ), N, WORK, M, ZERO,
427 $ Z( 1, START ), LDZ )
428 ELSE IF( ICOMPZ.EQ.2 ) THEN
429 CALL DSTEQR( 'I', M, D( START ), E( START ),
430 $ Z( START, START ), LDZ, WORK, INFO )
432 CALL DSTERF( M, D( START ), E( START ), INFO )
435 INFO = START*( N+1 ) + FINISH
446 IF( ICOMPZ.EQ.0 ) THEN
450 CALL DLASRT( 'I', N, D, INFO )
454 * Use Selection Sort to minimize swaps of eigenvectors
461 IF( D( J ).LT.P ) THEN
469 CALL DSWAP( N, Z( 1, I ), 1, Z( 1, K ), 1 )