1*> \brief \b ZLARRV computes the eigenvectors of the tridiagonal matrix T = L D LT given L, D and the eigenvalues of L D LT.
2*
3*  =========== DOCUMENTATION ===========
4*
5* Online html documentation available at
6*            http://www.netlib.org/lapack/explore-html/
7*
8*> \htmlonly
9*> Download ZLARRV + dependencies
10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zlarrv.f">
11*> [TGZ]</a>
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zlarrv.f">
13*> [ZIP]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zlarrv.f">
15*> [TXT]</a>
16*> \endhtmlonly
17*
18*  Definition:
19*  ===========
20*
21*       SUBROUTINE ZLARRV( N, VL, VU, D, L, PIVMIN,
22*                          ISPLIT, M, DOL, DOU, MINRGP,
23*                          RTOL1, RTOL2, W, WERR, WGAP,
24*                          IBLOCK, INDEXW, GERS, Z, LDZ, ISUPPZ,
25*                          WORK, IWORK, INFO )
26*
27*       .. Scalar Arguments ..
28*       INTEGER            DOL, DOU, INFO, LDZ, M, N
29*       DOUBLE PRECISION   MINRGP, PIVMIN, RTOL1, RTOL2, VL, VU
30*       ..
31*       .. Array Arguments ..
32*       INTEGER            IBLOCK( * ), INDEXW( * ), ISPLIT( * ),
33*      $                   ISUPPZ( * ), IWORK( * )
34*       DOUBLE PRECISION   D( * ), GERS( * ), L( * ), W( * ), WERR( * ),
35*      $                   WGAP( * ), WORK( * )
36*       COMPLEX*16        Z( LDZ, * )
37*       ..
38*
39*
40*> \par Purpose:
41*  =============
42*>
43*> \verbatim
44*>
45*> ZLARRV computes the eigenvectors of the tridiagonal matrix
46*> T = L D L**T given L, D and APPROXIMATIONS to the eigenvalues of L D L**T.
47*> The input eigenvalues should have been computed by DLARRE.
48*> \endverbatim
49*
50*  Arguments:
51*  ==========
52*
53*> \param[in] N
54*> \verbatim
55*>          N is INTEGER
56*>          The order of the matrix.  N >= 0.
57*> \endverbatim
58*>
59*> \param[in] VL
60*> \verbatim
61*>          VL is DOUBLE PRECISION
62*>          Lower bound of the interval that contains the desired
63*>          eigenvalues. VL < VU. Needed to compute gaps on the left or right
64*>          end of the extremal eigenvalues in the desired RANGE.
65*> \endverbatim
66*>
67*> \param[in] VU
68*> \verbatim
69*>          VU is DOUBLE PRECISION
70*>          Upper bound of the interval that contains the desired
71*>          eigenvalues. VL < VU. Needed to compute gaps on the left or right
72*>          end of the extremal eigenvalues in the desired RANGE.
73*> \endverbatim
74*>
75*> \param[in,out] D
76*> \verbatim
77*>          D is DOUBLE PRECISION array, dimension (N)
78*>          On entry, the N diagonal elements of the diagonal matrix D.
79*>          On exit, D may be overwritten.
80*> \endverbatim
81*>
82*> \param[in,out] L
83*> \verbatim
84*>          L is DOUBLE PRECISION array, dimension (N)
85*>          On entry, the (N-1) subdiagonal elements of the unit
86*>          bidiagonal matrix L are in elements 1 to N-1 of L
87*>          (if the matrix is not split.) At the end of each block
88*>          is stored the corresponding shift as given by DLARRE.
89*>          On exit, L is overwritten.
90*> \endverbatim
91*>
92*> \param[in] PIVMIN
93*> \verbatim
94*>          PIVMIN is DOUBLE PRECISION
95*>          The minimum pivot allowed in the Sturm sequence.
96*> \endverbatim
97*>
98*> \param[in] ISPLIT
99*> \verbatim
100*>          ISPLIT is INTEGER array, dimension (N)
101*>          The splitting points, at which T breaks up into blocks.
102*>          The first block consists of rows/columns 1 to
103*>          ISPLIT( 1 ), the second of rows/columns ISPLIT( 1 )+1
104*>          through ISPLIT( 2 ), etc.
105*> \endverbatim
106*>
107*> \param[in] M
108*> \verbatim
109*>          M is INTEGER
110*>          The total number of input eigenvalues.  0 <= M <= N.
111*> \endverbatim
112*>
113*> \param[in] DOL
114*> \verbatim
115*>          DOL is INTEGER
116*> \endverbatim
117*>
118*> \param[in] DOU
119*> \verbatim
120*>          DOU is INTEGER
121*>          If the user wants to compute only selected eigenvectors from all
122*>          the eigenvalues supplied, he can specify an index range DOL:DOU.
123*>          Or else the setting DOL=1, DOU=M should be applied.
124*>          Note that DOL and DOU refer to the order in which the eigenvalues
125*>          are stored in W.
126*>          If the user wants to compute only selected eigenpairs, then
127*>          the columns DOL-1 to DOU+1 of the eigenvector space Z contain the
128*>          computed eigenvectors. All other columns of Z are set to zero.
129*> \endverbatim
130*>
131*> \param[in] MINRGP
132*> \verbatim
133*>          MINRGP is DOUBLE PRECISION
134*> \endverbatim
135*>
136*> \param[in] RTOL1
137*> \verbatim
138*>          RTOL1 is DOUBLE PRECISION
139*> \endverbatim
140*>
141*> \param[in] RTOL2
142*> \verbatim
143*>          RTOL2 is DOUBLE PRECISION
144*>           Parameters for bisection.
145*>           An interval [LEFT,RIGHT] has converged if
146*>           RIGHT-LEFT < MAX( RTOL1*GAP, RTOL2*MAX(|LEFT|,|RIGHT|) )
147*> \endverbatim
148*>
149*> \param[in,out] W
150*> \verbatim
151*>          W is DOUBLE PRECISION array, dimension (N)
152*>          The first M elements of W contain the APPROXIMATE eigenvalues for
153*>          which eigenvectors are to be computed.  The eigenvalues
154*>          should be grouped by split-off block and ordered from
155*>          smallest to largest within the block ( The output array
156*>          W from DLARRE is expected here ). Furthermore, they are with
157*>          respect to the shift of the corresponding root representation
158*>          for their block. On exit, W holds the eigenvalues of the
159*>          UNshifted matrix.
160*> \endverbatim
161*>
162*> \param[in,out] WERR
163*> \verbatim
164*>          WERR is DOUBLE PRECISION array, dimension (N)
165*>          The first M elements contain the semiwidth of the uncertainty
166*>          interval of the corresponding eigenvalue in W
167*> \endverbatim
168*>
169*> \param[in,out] WGAP
170*> \verbatim
171*>          WGAP is DOUBLE PRECISION array, dimension (N)
172*>          The separation from the right neighbor eigenvalue in W.
173*> \endverbatim
174*>
175*> \param[in] IBLOCK
176*> \verbatim
177*>          IBLOCK is INTEGER array, dimension (N)
178*>          The indices of the blocks (submatrices) associated with the
179*>          corresponding eigenvalues in W; IBLOCK(i)=1 if eigenvalue
180*>          W(i) belongs to the first block from the top, =2 if W(i)
181*>          belongs to the second block, etc.
182*> \endverbatim
183*>
184*> \param[in] INDEXW
185*> \verbatim
186*>          INDEXW is INTEGER array, dimension (N)
187*>          The indices of the eigenvalues within each block (submatrix);
188*>          for example, INDEXW(i)= 10 and IBLOCK(i)=2 imply that the
189*>          i-th eigenvalue W(i) is the 10-th eigenvalue in the second block.
190*> \endverbatim
191*>
192*> \param[in] GERS
193*> \verbatim
194*>          GERS is DOUBLE PRECISION array, dimension (2*N)
195*>          The N Gerschgorin intervals (the i-th Gerschgorin interval
196*>          is (GERS(2*i-1), GERS(2*i)). The Gerschgorin intervals should
197*>          be computed from the original UNshifted matrix.
198*> \endverbatim
199*>
200*> \param[out] Z
201*> \verbatim
202*>          Z is COMPLEX*16 array, dimension (LDZ, max(1,M) )
203*>          If INFO = 0, the first M columns of Z contain the
204*>          orthonormal eigenvectors of the matrix T
205*>          corresponding to the input eigenvalues, with the i-th
206*>          column of Z holding the eigenvector associated with W(i).
207*>          Note: the user must ensure that at least max(1,M) columns are
208*>          supplied in the array Z.
209*> \endverbatim
210*>
211*> \param[in] LDZ
212*> \verbatim
213*>          LDZ is INTEGER
214*>          The leading dimension of the array Z.  LDZ >= 1, and if
215*>          JOBZ = 'V', LDZ >= max(1,N).
216*> \endverbatim
217*>
218*> \param[out] ISUPPZ
219*> \verbatim
220*>          ISUPPZ is INTEGER array, dimension ( 2*max(1,M) )
221*>          The support of the eigenvectors in Z, i.e., the indices
222*>          indicating the nonzero elements in Z. The I-th eigenvector
223*>          is nonzero only in elements ISUPPZ( 2*I-1 ) through
224*>          ISUPPZ( 2*I ).
225*> \endverbatim
226*>
227*> \param[out] WORK
228*> \verbatim
229*>          WORK is DOUBLE PRECISION array, dimension (12*N)
230*> \endverbatim
231*>
232*> \param[out] IWORK
233*> \verbatim
234*>          IWORK is INTEGER array, dimension (7*N)
235*> \endverbatim
236*>
237*> \param[out] INFO
238*> \verbatim
239*>          INFO is INTEGER
240*>          = 0:  successful exit
241*>
242*>          > 0:  A problem occurred in ZLARRV.
243*>          < 0:  One of the called subroutines signaled an internal problem.
244*>                Needs inspection of the corresponding parameter IINFO
245*>                for further information.
246*>
247*>          =-1:  Problem in DLARRB when refining a child's eigenvalues.
248*>          =-2:  Problem in DLARRF when computing the RRR of a child.
249*>                When a child is inside a tight cluster, it can be difficult
250*>                to find an RRR. A partial remedy from the user's point of
251*>                view is to make the parameter MINRGP smaller and recompile.
252*>                However, as the orthogonality of the computed vectors is
253*>                proportional to 1/MINRGP, the user should be aware that
254*>                he might be trading in precision when he decreases MINRGP.
255*>          =-3:  Problem in DLARRB when refining a single eigenvalue
256*>                after the Rayleigh correction was rejected.
257*>          = 5:  The Rayleigh Quotient Iteration failed to converge to
258*>                full accuracy in MAXITR steps.
259*> \endverbatim
260*
261*  Authors:
262*  ========
263*
264*> \author Univ. of Tennessee
265*> \author Univ. of California Berkeley
266*> \author Univ. of Colorado Denver
267*> \author NAG Ltd.
268*
269*> \ingroup complex16OTHERauxiliary
270*
271*> \par Contributors:
272*  ==================
273*>
274*> Beresford Parlett, University of California, Berkeley, USA \n
275*> Jim Demmel, University of California, Berkeley, USA \n
276*> Inderjit Dhillon, University of Texas, Austin, USA \n
277*> Osni Marques, LBNL/NERSC, USA \n
278*> Christof Voemel, University of California, Berkeley, USA
279*
280*  =====================================================================
281      SUBROUTINE ZLARRV( N, VL, VU, D, L, PIVMIN,
282     $                   ISPLIT, M, DOL, DOU, MINRGP,
283     $                   RTOL1, RTOL2, W, WERR, WGAP,
284     $                   IBLOCK, INDEXW, GERS, Z, LDZ, ISUPPZ,
285     $                   WORK, IWORK, INFO )
286*
287*  -- LAPACK auxiliary routine --
288*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
289*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
290*
291*     .. Scalar Arguments ..
292      INTEGER            DOL, DOU, INFO, LDZ, M, N
293      DOUBLE PRECISION   MINRGP, PIVMIN, RTOL1, RTOL2, VL, VU
294*     ..
295*     .. Array Arguments ..
296      INTEGER            IBLOCK( * ), INDEXW( * ), ISPLIT( * ),
297     $                   ISUPPZ( * ), IWORK( * )
298      DOUBLE PRECISION   D( * ), GERS( * ), L( * ), W( * ), WERR( * ),
299     $                   WGAP( * ), WORK( * )
300      COMPLEX*16        Z( LDZ, * )
301*     ..
302*
303*  =====================================================================
304*
305*     .. Parameters ..
306      INTEGER            MAXITR
307      PARAMETER          ( MAXITR = 10 )
308      COMPLEX*16         CZERO
309      PARAMETER          ( CZERO = ( 0.0D0, 0.0D0 ) )
310      DOUBLE PRECISION   ZERO, ONE, TWO, THREE, FOUR, HALF
311      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0,
312     $                     TWO = 2.0D0, THREE = 3.0D0,
313     $                     FOUR = 4.0D0, HALF = 0.5D0)
314*     ..
315*     .. Local Scalars ..
316      LOGICAL            ESKIP, NEEDBS, STP2II, TRYRQC, USEDBS, USEDRQ
317      INTEGER            DONE, I, IBEGIN, IDONE, IEND, II, IINDC1,
318     $                   IINDC2, IINDR, IINDWK, IINFO, IM, IN, INDEIG,
319     $                   INDLD, INDLLD, INDWRK, ISUPMN, ISUPMX, ITER,
320     $                   ITMP1, J, JBLK, K, MINIWSIZE, MINWSIZE, NCLUS,
321     $                   NDEPTH, NEGCNT, NEWCLS, NEWFST, NEWFTT, NEWLST,
322     $                   NEWSIZ, OFFSET, OLDCLS, OLDFST, OLDIEN, OLDLST,
323     $                   OLDNCL, P, PARITY, Q, WBEGIN, WEND, WINDEX,
324     $                   WINDMN, WINDPL, ZFROM, ZTO, ZUSEDL, ZUSEDU,
325     $                   ZUSEDW
326      INTEGER            INDIN1, INDIN2
327      DOUBLE PRECISION   BSTRES, BSTW, EPS, FUDGE, GAP, GAPTOL, GL, GU,
328     $                   LAMBDA, LEFT, LGAP, MINGMA, NRMINV, RESID,
329     $                   RGAP, RIGHT, RQCORR, RQTOL, SAVGAP, SGNDEF,
330     $                   SIGMA, SPDIAM, SSIGMA, TAU, TMP, TOL, ZTZ
331*     ..
332*     .. External Functions ..
333      DOUBLE PRECISION   DLAMCH
334      EXTERNAL           DLAMCH
335*     ..
336*     .. External Subroutines ..
337      EXTERNAL           DCOPY, DLARRB, DLARRF, ZDSCAL, ZLAR1V,
338     $                   ZLASET
339*     ..
340*     .. Intrinsic Functions ..
341      INTRINSIC ABS, DBLE, MAX, MIN
342      INTRINSIC DCMPLX
343*     ..
344*     .. Executable Statements ..
345*     ..
346
347      INFO = 0
348*
349*     Quick return if possible
350*
351      IF( N.LE.0 ) THEN
352         RETURN
353      END IF
354*
355*     The first N entries of WORK are reserved for the eigenvalues
356      INDLD = N+1
357      INDLLD= 2*N+1
358      INDIN1 = 3*N + 1
359      INDIN2 = 4*N + 1
360      INDWRK = 5*N + 1
361      MINWSIZE = 12 * N
362
363      DO 5 I= 1,MINWSIZE
364         WORK( I ) = ZERO
365 5    CONTINUE
366
367*     IWORK(IINDR+1:IINDR+N) hold the twist indices R for the
368*     factorization used to compute the FP vector
369      IINDR = 0
370*     IWORK(IINDC1+1:IINC2+N) are used to store the clusters of the current
371*     layer and the one above.
372      IINDC1 = N
373      IINDC2 = 2*N
374      IINDWK = 3*N + 1
375
376      MINIWSIZE = 7 * N
377      DO 10 I= 1,MINIWSIZE
378         IWORK( I ) = 0
379 10   CONTINUE
380
381      ZUSEDL = 1
382      IF(DOL.GT.1) THEN
383*        Set lower bound for use of Z
384         ZUSEDL = DOL-1
385      ENDIF
386      ZUSEDU = M
387      IF(DOU.LT.M) THEN
388*        Set lower bound for use of Z
389         ZUSEDU = DOU+1
390      ENDIF
391*     The width of the part of Z that is used
392      ZUSEDW = ZUSEDU - ZUSEDL + 1
393
394
395      CALL ZLASET( 'Full', N, ZUSEDW, CZERO, CZERO,
396     $                    Z(1,ZUSEDL), LDZ )
397
398      EPS = DLAMCH( 'Precision' )
399      RQTOL = TWO * EPS
400*
401*     Set expert flags for standard code.
402      TRYRQC = .TRUE.
403
404      IF((DOL.EQ.1).AND.(DOU.EQ.M)) THEN
405      ELSE
406*        Only selected eigenpairs are computed. Since the other evalues
407*        are not refined by RQ iteration, bisection has to compute to full
408*        accuracy.
409         RTOL1 = FOUR * EPS
410         RTOL2 = FOUR * EPS
411      ENDIF
412
413*     The entries WBEGIN:WEND in W, WERR, WGAP correspond to the
414*     desired eigenvalues. The support of the nonzero eigenvector
415*     entries is contained in the interval IBEGIN:IEND.
416*     Remark that if k eigenpairs are desired, then the eigenvectors
417*     are stored in k contiguous columns of Z.
418
419*     DONE is the number of eigenvectors already computed
420      DONE = 0
421      IBEGIN = 1
422      WBEGIN = 1
423      DO 170 JBLK = 1, IBLOCK( M )
424         IEND = ISPLIT( JBLK )
425         SIGMA = L( IEND )
426*        Find the eigenvectors of the submatrix indexed IBEGIN
427*        through IEND.
428         WEND = WBEGIN - 1
429 15      CONTINUE
430         IF( WEND.LT.M ) THEN
431            IF( IBLOCK( WEND+1 ).EQ.JBLK ) THEN
432               WEND = WEND + 1
433               GO TO 15
434            END IF
435         END IF
436         IF( WEND.LT.WBEGIN ) THEN
437            IBEGIN = IEND + 1
438            GO TO 170
439         ELSEIF( (WEND.LT.DOL).OR.(WBEGIN.GT.DOU) ) THEN
440            IBEGIN = IEND + 1
441            WBEGIN = WEND + 1
442            GO TO 170
443         END IF
444
445*        Find local spectral diameter of the block
446         GL = GERS( 2*IBEGIN-1 )
447         GU = GERS( 2*IBEGIN )
448         DO 20 I = IBEGIN+1 , IEND
449            GL = MIN( GERS( 2*I-1 ), GL )
450            GU = MAX( GERS( 2*I ), GU )
451 20      CONTINUE
452         SPDIAM = GU - GL
453
454*        OLDIEN is the last index of the previous block
455         OLDIEN = IBEGIN - 1
456*        Calculate the size of the current block
457         IN = IEND - IBEGIN + 1
458*        The number of eigenvalues in the current block
459         IM = WEND - WBEGIN + 1
460
461*        This is for a 1x1 block
462         IF( IBEGIN.EQ.IEND ) THEN
463            DONE = DONE+1
464            Z( IBEGIN, WBEGIN ) = DCMPLX( ONE, ZERO )
465            ISUPPZ( 2*WBEGIN-1 ) = IBEGIN
466            ISUPPZ( 2*WBEGIN ) = IBEGIN
467            W( WBEGIN ) = W( WBEGIN ) + SIGMA
468            WORK( WBEGIN ) = W( WBEGIN )
469            IBEGIN = IEND + 1
470            WBEGIN = WBEGIN + 1
471            GO TO 170
472         END IF
473
474*        The desired (shifted) eigenvalues are stored in W(WBEGIN:WEND)
475*        Note that these can be approximations, in this case, the corresp.
476*        entries of WERR give the size of the uncertainty interval.
477*        The eigenvalue approximations will be refined when necessary as
478*        high relative accuracy is required for the computation of the
479*        corresponding eigenvectors.
480         CALL DCOPY( IM, W( WBEGIN ), 1,
481     $                   WORK( WBEGIN ), 1 )
482
483*        We store in W the eigenvalue approximations w.r.t. the original
484*        matrix T.
485         DO 30 I=1,IM
486            W(WBEGIN+I-1) = W(WBEGIN+I-1)+SIGMA
487 30      CONTINUE
488
489
490*        NDEPTH is the current depth of the representation tree
491         NDEPTH = 0
492*        PARITY is either 1 or 0
493         PARITY = 1
494*        NCLUS is the number of clusters for the next level of the
495*        representation tree, we start with NCLUS = 1 for the root
496         NCLUS = 1
497         IWORK( IINDC1+1 ) = 1
498         IWORK( IINDC1+2 ) = IM
499
500*        IDONE is the number of eigenvectors already computed in the current
501*        block
502         IDONE = 0
503*        loop while( IDONE.LT.IM )
504*        generate the representation tree for the current block and
505*        compute the eigenvectors
506   40    CONTINUE
507         IF( IDONE.LT.IM ) THEN
508*           This is a crude protection against infinitely deep trees
509            IF( NDEPTH.GT.M ) THEN
510               INFO = -2
511               RETURN
512            ENDIF
513*           breadth first processing of the current level of the representation
514*           tree: OLDNCL = number of clusters on current level
515            OLDNCL = NCLUS
516*           reset NCLUS to count the number of child clusters
517            NCLUS = 0
518*
519            PARITY = 1 - PARITY
520            IF( PARITY.EQ.0 ) THEN
521               OLDCLS = IINDC1
522               NEWCLS = IINDC2
523            ELSE
524               OLDCLS = IINDC2
525               NEWCLS = IINDC1
526            END IF
527*           Process the clusters on the current level
528            DO 150 I = 1, OLDNCL
529               J = OLDCLS + 2*I
530*              OLDFST, OLDLST = first, last index of current cluster.
531*                               cluster indices start with 1 and are relative
532*                               to WBEGIN when accessing W, WGAP, WERR, Z
533               OLDFST = IWORK( J-1 )
534               OLDLST = IWORK( J )
535               IF( NDEPTH.GT.0 ) THEN
536*                 Retrieve relatively robust representation (RRR) of cluster
537*                 that has been computed at the previous level
538*                 The RRR is stored in Z and overwritten once the eigenvectors
539*                 have been computed or when the cluster is refined
540
541                  IF((DOL.EQ.1).AND.(DOU.EQ.M)) THEN
542*                    Get representation from location of the leftmost evalue
543*                    of the cluster
544                     J = WBEGIN + OLDFST - 1
545                  ELSE
546                     IF(WBEGIN+OLDFST-1.LT.DOL) THEN
547*                       Get representation from the left end of Z array
548                        J = DOL - 1
549                     ELSEIF(WBEGIN+OLDFST-1.GT.DOU) THEN
550*                       Get representation from the right end of Z array
551                        J = DOU
552                     ELSE
553                        J = WBEGIN + OLDFST - 1
554                     ENDIF
555                  ENDIF
556                  DO 45 K = 1, IN - 1
557                     D( IBEGIN+K-1 ) = DBLE( Z( IBEGIN+K-1,
558     $                                 J ) )
559                     L( IBEGIN+K-1 ) = DBLE( Z( IBEGIN+K-1,
560     $                                 J+1 ) )
561   45             CONTINUE
562                  D( IEND ) = DBLE( Z( IEND, J ) )
563                  SIGMA = DBLE( Z( IEND, J+1 ) )
564
565*                 Set the corresponding entries in Z to zero
566                  CALL ZLASET( 'Full', IN, 2, CZERO, CZERO,
567     $                         Z( IBEGIN, J), LDZ )
568               END IF
569
570*              Compute DL and DLL of current RRR
571               DO 50 J = IBEGIN, IEND-1
572                  TMP = D( J )*L( J )
573                  WORK( INDLD-1+J ) = TMP
574                  WORK( INDLLD-1+J ) = TMP*L( J )
575   50          CONTINUE
576
577               IF( NDEPTH.GT.0 ) THEN
578*                 P and Q are index of the first and last eigenvalue to compute
579*                 within the current block
580                  P = INDEXW( WBEGIN-1+OLDFST )
581                  Q = INDEXW( WBEGIN-1+OLDLST )
582*                 Offset for the arrays WORK, WGAP and WERR, i.e., the P-OFFSET
583*                 through the Q-OFFSET elements of these arrays are to be used.
584*                  OFFSET = P-OLDFST
585                  OFFSET = INDEXW( WBEGIN ) - 1
586*                 perform limited bisection (if necessary) to get approximate
587*                 eigenvalues to the precision needed.
588                  CALL DLARRB( IN, D( IBEGIN ),
589     $                         WORK(INDLLD+IBEGIN-1),
590     $                         P, Q, RTOL1, RTOL2, OFFSET,
591     $                         WORK(WBEGIN),WGAP(WBEGIN),WERR(WBEGIN),
592     $                         WORK( INDWRK ), IWORK( IINDWK ),
593     $                         PIVMIN, SPDIAM, IN, IINFO )
594                  IF( IINFO.NE.0 ) THEN
595                     INFO = -1
596                     RETURN
597                  ENDIF
598*                 We also recompute the extremal gaps. W holds all eigenvalues
599*                 of the unshifted matrix and must be used for computation
600*                 of WGAP, the entries of WORK might stem from RRRs with
601*                 different shifts. The gaps from WBEGIN-1+OLDFST to
602*                 WBEGIN-1+OLDLST are correctly computed in DLARRB.
603*                 However, we only allow the gaps to become greater since
604*                 this is what should happen when we decrease WERR
605                  IF( OLDFST.GT.1) THEN
606                     WGAP( WBEGIN+OLDFST-2 ) =
607     $             MAX(WGAP(WBEGIN+OLDFST-2),
608     $                 W(WBEGIN+OLDFST-1)-WERR(WBEGIN+OLDFST-1)
609     $                 - W(WBEGIN+OLDFST-2)-WERR(WBEGIN+OLDFST-2) )
610                  ENDIF
611                  IF( WBEGIN + OLDLST -1 .LT. WEND ) THEN
612                     WGAP( WBEGIN+OLDLST-1 ) =
613     $               MAX(WGAP(WBEGIN+OLDLST-1),
614     $                   W(WBEGIN+OLDLST)-WERR(WBEGIN+OLDLST)
615     $                   - W(WBEGIN+OLDLST-1)-WERR(WBEGIN+OLDLST-1) )
616                  ENDIF
617*                 Each time the eigenvalues in WORK get refined, we store
618*                 the newly found approximation with all shifts applied in W
619                  DO 53 J=OLDFST,OLDLST
620                     W(WBEGIN+J-1) = WORK(WBEGIN+J-1)+SIGMA
621 53               CONTINUE
622               END IF
623
624*              Process the current node.
625               NEWFST = OLDFST
626               DO 140 J = OLDFST, OLDLST
627                  IF( J.EQ.OLDLST ) THEN
628*                    we are at the right end of the cluster, this is also the
629*                    boundary of the child cluster
630                     NEWLST = J
631                  ELSE IF ( WGAP( WBEGIN + J -1).GE.
632     $                    MINRGP* ABS( WORK(WBEGIN + J -1) ) ) THEN
633*                    the right relative gap is big enough, the child cluster
634*                    (NEWFST,..,NEWLST) is well separated from the following
635                     NEWLST = J
636                   ELSE
637*                    inside a child cluster, the relative gap is not
638*                    big enough.
639                     GOTO 140
640                  END IF
641
642*                 Compute size of child cluster found
643                  NEWSIZ = NEWLST - NEWFST + 1
644
645*                 NEWFTT is the place in Z where the new RRR or the computed
646*                 eigenvector is to be stored
647                  IF((DOL.EQ.1).AND.(DOU.EQ.M)) THEN
648*                    Store representation at location of the leftmost evalue
649*                    of the cluster
650                     NEWFTT = WBEGIN + NEWFST - 1
651                  ELSE
652                     IF(WBEGIN+NEWFST-1.LT.DOL) THEN
653*                       Store representation at the left end of Z array
654                        NEWFTT = DOL - 1
655                     ELSEIF(WBEGIN+NEWFST-1.GT.DOU) THEN
656*                       Store representation at the right end of Z array
657                        NEWFTT = DOU
658                     ELSE
659                        NEWFTT = WBEGIN + NEWFST - 1
660                     ENDIF
661                  ENDIF
662
663                  IF( NEWSIZ.GT.1) THEN
664*
665*                    Current child is not a singleton but a cluster.
666*                    Compute and store new representation of child.
667*
668*
669*                    Compute left and right cluster gap.
670*
671*                    LGAP and RGAP are not computed from WORK because
672*                    the eigenvalue approximations may stem from RRRs
673*                    different shifts. However, W hold all eigenvalues
674*                    of the unshifted matrix. Still, the entries in WGAP
675*                    have to be computed from WORK since the entries
676*                    in W might be of the same order so that gaps are not
677*                    exhibited correctly for very close eigenvalues.
678                     IF( NEWFST.EQ.1 ) THEN
679                        LGAP = MAX( ZERO,
680     $                       W(WBEGIN)-WERR(WBEGIN) - VL )
681                    ELSE
682                        LGAP = WGAP( WBEGIN+NEWFST-2 )
683                     ENDIF
684                     RGAP = WGAP( WBEGIN+NEWLST-1 )
685*
686*                    Compute left- and rightmost eigenvalue of child
687*                    to high precision in order to shift as close
688*                    as possible and obtain as large relative gaps
689*                    as possible
690*
691                     DO 55 K =1,2
692                        IF(K.EQ.1) THEN
693                           P = INDEXW( WBEGIN-1+NEWFST )
694                        ELSE
695                           P = INDEXW( WBEGIN-1+NEWLST )
696                        ENDIF
697                        OFFSET = INDEXW( WBEGIN ) - 1
698                        CALL DLARRB( IN, D(IBEGIN),
699     $                       WORK( INDLLD+IBEGIN-1 ),P,P,
700     $                       RQTOL, RQTOL, OFFSET,
701     $                       WORK(WBEGIN),WGAP(WBEGIN),
702     $                       WERR(WBEGIN),WORK( INDWRK ),
703     $                       IWORK( IINDWK ), PIVMIN, SPDIAM,
704     $                       IN, IINFO )
705 55                  CONTINUE
706*
707                     IF((WBEGIN+NEWLST-1.LT.DOL).OR.
708     $                  (WBEGIN+NEWFST-1.GT.DOU)) THEN
709*                       if the cluster contains no desired eigenvalues
710*                       skip the computation of that branch of the rep. tree
711*
712*                       We could skip before the refinement of the extremal
713*                       eigenvalues of the child, but then the representation
714*                       tree could be different from the one when nothing is
715*                       skipped. For this reason we skip at this place.
716                        IDONE = IDONE + NEWLST - NEWFST + 1
717                        GOTO 139
718                     ENDIF
719*
720*                    Compute RRR of child cluster.
721*                    Note that the new RRR is stored in Z
722*
723*                    DLARRF needs LWORK = 2*N
724                     CALL DLARRF( IN, D( IBEGIN ), L( IBEGIN ),
725     $                         WORK(INDLD+IBEGIN-1),
726     $                         NEWFST, NEWLST, WORK(WBEGIN),
727     $                         WGAP(WBEGIN), WERR(WBEGIN),
728     $                         SPDIAM, LGAP, RGAP, PIVMIN, TAU,
729     $                         WORK( INDIN1 ), WORK( INDIN2 ),
730     $                         WORK( INDWRK ), IINFO )
731*                    In the complex case, DLARRF cannot write
732*                    the new RRR directly into Z and needs an intermediate
733*                    workspace
734                     DO 56 K = 1, IN-1
735                        Z( IBEGIN+K-1, NEWFTT ) =
736     $                     DCMPLX( WORK( INDIN1+K-1 ), ZERO )
737                        Z( IBEGIN+K-1, NEWFTT+1 ) =
738     $                     DCMPLX( WORK( INDIN2+K-1 ), ZERO )
739   56                CONTINUE
740                     Z( IEND, NEWFTT ) =
741     $                  DCMPLX( WORK( INDIN1+IN-1 ), ZERO )
742                     IF( IINFO.EQ.0 ) THEN
743*                       a new RRR for the cluster was found by DLARRF
744*                       update shift and store it
745                        SSIGMA = SIGMA + TAU
746                        Z( IEND, NEWFTT+1 ) = DCMPLX( SSIGMA, ZERO )
747*                       WORK() are the midpoints and WERR() the semi-width
748*                       Note that the entries in W are unchanged.
749                        DO 116 K = NEWFST, NEWLST
750                           FUDGE =
751     $                          THREE*EPS*ABS(WORK(WBEGIN+K-1))
752                           WORK( WBEGIN + K - 1 ) =
753     $                          WORK( WBEGIN + K - 1) - TAU
754                           FUDGE = FUDGE +
755     $                          FOUR*EPS*ABS(WORK(WBEGIN+K-1))
756*                          Fudge errors
757                           WERR( WBEGIN + K - 1 ) =
758     $                          WERR( WBEGIN + K - 1 ) + FUDGE
759*                          Gaps are not fudged. Provided that WERR is small
760*                          when eigenvalues are close, a zero gap indicates
761*                          that a new representation is needed for resolving
762*                          the cluster. A fudge could lead to a wrong decision
763*                          of judging eigenvalues 'separated' which in
764*                          reality are not. This could have a negative impact
765*                          on the orthogonality of the computed eigenvectors.
766 116                    CONTINUE
767
768                        NCLUS = NCLUS + 1
769                        K = NEWCLS + 2*NCLUS
770                        IWORK( K-1 ) = NEWFST
771                        IWORK( K ) = NEWLST
772                     ELSE
773                        INFO = -2
774                        RETURN
775                     ENDIF
776                  ELSE
777*
778*                    Compute eigenvector of singleton
779*
780                     ITER = 0
781*
782                     TOL = FOUR * LOG(DBLE(IN)) * EPS
783*
784                     K = NEWFST
785                     WINDEX = WBEGIN + K - 1
786                     WINDMN = MAX(WINDEX - 1,1)
787                     WINDPL = MIN(WINDEX + 1,M)
788                     LAMBDA = WORK( WINDEX )
789                     DONE = DONE + 1
790*                    Check if eigenvector computation is to be skipped
791                     IF((WINDEX.LT.DOL).OR.
792     $                  (WINDEX.GT.DOU)) THEN
793                        ESKIP = .TRUE.
794                        GOTO 125
795                     ELSE
796                        ESKIP = .FALSE.
797                     ENDIF
798                     LEFT = WORK( WINDEX ) - WERR( WINDEX )
799                     RIGHT = WORK( WINDEX ) + WERR( WINDEX )
800                     INDEIG = INDEXW( WINDEX )
801*                    Note that since we compute the eigenpairs for a child,
802*                    all eigenvalue approximations are w.r.t the same shift.
803*                    In this case, the entries in WORK should be used for
804*                    computing the gaps since they exhibit even very small
805*                    differences in the eigenvalues, as opposed to the
806*                    entries in W which might "look" the same.
807
808                     IF( K .EQ. 1) THEN
809*                       In the case RANGE='I' and with not much initial
810*                       accuracy in LAMBDA and VL, the formula
811*                       LGAP = MAX( ZERO, (SIGMA - VL) + LAMBDA )
812*                       can lead to an overestimation of the left gap and
813*                       thus to inadequately early RQI 'convergence'.
814*                       Prevent this by forcing a small left gap.
815                        LGAP = EPS*MAX(ABS(LEFT),ABS(RIGHT))
816                     ELSE
817                        LGAP = WGAP(WINDMN)
818                     ENDIF
819                     IF( K .EQ. IM) THEN
820*                       In the case RANGE='I' and with not much initial
821*                       accuracy in LAMBDA and VU, the formula
822*                       can lead to an overestimation of the right gap and
823*                       thus to inadequately early RQI 'convergence'.
824*                       Prevent this by forcing a small right gap.
825                        RGAP = EPS*MAX(ABS(LEFT),ABS(RIGHT))
826                     ELSE
827                        RGAP = WGAP(WINDEX)
828                     ENDIF
829                     GAP = MIN( LGAP, RGAP )
830                     IF(( K .EQ. 1).OR.(K .EQ. IM)) THEN
831*                       The eigenvector support can become wrong
832*                       because significant entries could be cut off due to a
833*                       large GAPTOL parameter in LAR1V. Prevent this.
834                        GAPTOL = ZERO
835                     ELSE
836                        GAPTOL = GAP * EPS
837                     ENDIF
838                     ISUPMN = IN
839                     ISUPMX = 1
840*                    Update WGAP so that it holds the minimum gap
841*                    to the left or the right. This is crucial in the
842*                    case where bisection is used to ensure that the
843*                    eigenvalue is refined up to the required precision.
844*                    The correct value is restored afterwards.
845                     SAVGAP = WGAP(WINDEX)
846                     WGAP(WINDEX) = GAP
847*                    We want to use the Rayleigh Quotient Correction
848*                    as often as possible since it converges quadratically
849*                    when we are close enough to the desired eigenvalue.
850*                    However, the Rayleigh Quotient can have the wrong sign
851*                    and lead us away from the desired eigenvalue. In this
852*                    case, the best we can do is to use bisection.
853                     USEDBS = .FALSE.
854                     USEDRQ = .FALSE.
855*                    Bisection is initially turned off unless it is forced
856                     NEEDBS =  .NOT.TRYRQC
857 120                 CONTINUE
858*                    Check if bisection should be used to refine eigenvalue
859                     IF(NEEDBS) THEN
860*                       Take the bisection as new iterate
861                        USEDBS = .TRUE.
862                        ITMP1 = IWORK( IINDR+WINDEX )
863                        OFFSET = INDEXW( WBEGIN ) - 1
864                        CALL DLARRB( IN, D(IBEGIN),
865     $                       WORK(INDLLD+IBEGIN-1),INDEIG,INDEIG,
866     $                       ZERO, TWO*EPS, OFFSET,
867     $                       WORK(WBEGIN),WGAP(WBEGIN),
868     $                       WERR(WBEGIN),WORK( INDWRK ),
869     $                       IWORK( IINDWK ), PIVMIN, SPDIAM,
870     $                       ITMP1, IINFO )
871                        IF( IINFO.NE.0 ) THEN
872                           INFO = -3
873                           RETURN
874                        ENDIF
875                        LAMBDA = WORK( WINDEX )
876*                       Reset twist index from inaccurate LAMBDA to
877*                       force computation of true MINGMA
878                        IWORK( IINDR+WINDEX ) = 0
879                     ENDIF
880*                    Given LAMBDA, compute the eigenvector.
881                     CALL ZLAR1V( IN, 1, IN, LAMBDA, D( IBEGIN ),
882     $                    L( IBEGIN ), WORK(INDLD+IBEGIN-1),
883     $                    WORK(INDLLD+IBEGIN-1),
884     $                    PIVMIN, GAPTOL, Z( IBEGIN, WINDEX ),
885     $                    .NOT.USEDBS, NEGCNT, ZTZ, MINGMA,
886     $                    IWORK( IINDR+WINDEX ), ISUPPZ( 2*WINDEX-1 ),
887     $                    NRMINV, RESID, RQCORR, WORK( INDWRK ) )
888                     IF(ITER .EQ. 0) THEN
889                        BSTRES = RESID
890                        BSTW = LAMBDA
891                     ELSEIF(RESID.LT.BSTRES) THEN
892                        BSTRES = RESID
893                        BSTW = LAMBDA
894                     ENDIF
895                     ISUPMN = MIN(ISUPMN,ISUPPZ( 2*WINDEX-1 ))
896                     ISUPMX = MAX(ISUPMX,ISUPPZ( 2*WINDEX ))
897                     ITER = ITER + 1
898
899*                    sin alpha <= |resid|/gap
900*                    Note that both the residual and the gap are
901*                    proportional to the matrix, so ||T|| doesn't play
902*                    a role in the quotient
903
904*
905*                    Convergence test for Rayleigh-Quotient iteration
906*                    (omitted when Bisection has been used)
907*
908                     IF( RESID.GT.TOL*GAP .AND. ABS( RQCORR ).GT.
909     $                    RQTOL*ABS( LAMBDA ) .AND. .NOT. USEDBS)
910     $                    THEN
911*                       We need to check that the RQCORR update doesn't
912*                       move the eigenvalue away from the desired one and
913*                       towards a neighbor. -> protection with bisection
914                        IF(INDEIG.LE.NEGCNT) THEN
915*                          The wanted eigenvalue lies to the left
916                           SGNDEF = -ONE
917                        ELSE
918*                          The wanted eigenvalue lies to the right
919                           SGNDEF = ONE
920                        ENDIF
921*                       We only use the RQCORR if it improves the
922*                       the iterate reasonably.
923                        IF( ( RQCORR*SGNDEF.GE.ZERO )
924     $                       .AND.( LAMBDA + RQCORR.LE. RIGHT)
925     $                       .AND.( LAMBDA + RQCORR.GE. LEFT)
926     $                       ) THEN
927                           USEDRQ = .TRUE.
928*                          Store new midpoint of bisection interval in WORK
929                           IF(SGNDEF.EQ.ONE) THEN
930*                             The current LAMBDA is on the left of the true
931*                             eigenvalue
932                              LEFT = LAMBDA
933*                             We prefer to assume that the error estimate
934*                             is correct. We could make the interval not
935*                             as a bracket but to be modified if the RQCORR
936*                             chooses to. In this case, the RIGHT side should
937*                             be modified as follows:
938*                              RIGHT = MAX(RIGHT, LAMBDA + RQCORR)
939                           ELSE
940*                             The current LAMBDA is on the right of the true
941*                             eigenvalue
942                              RIGHT = LAMBDA
943*                             See comment about assuming the error estimate is
944*                             correct above.
945*                              LEFT = MIN(LEFT, LAMBDA + RQCORR)
946                           ENDIF
947                           WORK( WINDEX ) =
948     $                       HALF * (RIGHT + LEFT)
949*                          Take RQCORR since it has the correct sign and
950*                          improves the iterate reasonably
951                           LAMBDA = LAMBDA + RQCORR
952*                          Update width of error interval
953                           WERR( WINDEX ) =
954     $                             HALF * (RIGHT-LEFT)
955                        ELSE
956                           NEEDBS = .TRUE.
957                        ENDIF
958                        IF(RIGHT-LEFT.LT.RQTOL*ABS(LAMBDA)) THEN
959*                             The eigenvalue is computed to bisection accuracy
960*                             compute eigenvector and stop
961                           USEDBS = .TRUE.
962                           GOTO 120
963                        ELSEIF( ITER.LT.MAXITR ) THEN
964                           GOTO 120
965                        ELSEIF( ITER.EQ.MAXITR ) THEN
966                           NEEDBS = .TRUE.
967                           GOTO 120
968                        ELSE
969                           INFO = 5
970                           RETURN
971                        END IF
972                     ELSE
973                        STP2II = .FALSE.
974        IF(USEDRQ .AND. USEDBS .AND.
975     $                     BSTRES.LE.RESID) THEN
976                           LAMBDA = BSTW
977                           STP2II = .TRUE.
978                        ENDIF
979                        IF (STP2II) THEN
980*                          improve error angle by second step
981                           CALL ZLAR1V( IN, 1, IN, LAMBDA,
982     $                          D( IBEGIN ), L( IBEGIN ),
983     $                          WORK(INDLD+IBEGIN-1),
984     $                          WORK(INDLLD+IBEGIN-1),
985     $                          PIVMIN, GAPTOL, Z( IBEGIN, WINDEX ),
986     $                          .NOT.USEDBS, NEGCNT, ZTZ, MINGMA,
987     $                          IWORK( IINDR+WINDEX ),
988     $                          ISUPPZ( 2*WINDEX-1 ),
989     $                          NRMINV, RESID, RQCORR, WORK( INDWRK ) )
990                        ENDIF
991                        WORK( WINDEX ) = LAMBDA
992                     END IF
993*
994*                    Compute FP-vector support w.r.t. whole matrix
995*
996                     ISUPPZ( 2*WINDEX-1 ) = ISUPPZ( 2*WINDEX-1 )+OLDIEN
997                     ISUPPZ( 2*WINDEX ) = ISUPPZ( 2*WINDEX )+OLDIEN
998                     ZFROM = ISUPPZ( 2*WINDEX-1 )
999                     ZTO = ISUPPZ( 2*WINDEX )
1000                     ISUPMN = ISUPMN + OLDIEN
1001                     ISUPMX = ISUPMX + OLDIEN
1002*                    Ensure vector is ok if support in the RQI has changed
1003                     IF(ISUPMN.LT.ZFROM) THEN
1004                        DO 122 II = ISUPMN,ZFROM-1
1005                           Z( II, WINDEX ) = ZERO
1006 122                    CONTINUE
1007                     ENDIF
1008                     IF(ISUPMX.GT.ZTO) THEN
1009                        DO 123 II = ZTO+1,ISUPMX
1010                           Z( II, WINDEX ) = ZERO
1011 123                    CONTINUE
1012                     ENDIF
1013                     CALL ZDSCAL( ZTO-ZFROM+1, NRMINV,
1014     $                       Z( ZFROM, WINDEX ), 1 )
1015 125                 CONTINUE
1016*                    Update W
1017                     W( WINDEX ) = LAMBDA+SIGMA
1018*                    Recompute the gaps on the left and right
1019*                    But only allow them to become larger and not
1020*                    smaller (which can only happen through "bad"
1021*                    cancellation and doesn't reflect the theory
1022*                    where the initial gaps are underestimated due
1023*                    to WERR being too crude.)
1024                     IF(.NOT.ESKIP) THEN
1025                        IF( K.GT.1) THEN
1026                           WGAP( WINDMN ) = MAX( WGAP(WINDMN),
1027     $                          W(WINDEX)-WERR(WINDEX)
1028     $                          - W(WINDMN)-WERR(WINDMN) )
1029                        ENDIF
1030                        IF( WINDEX.LT.WEND ) THEN
1031                           WGAP( WINDEX ) = MAX( SAVGAP,
1032     $                          W( WINDPL )-WERR( WINDPL )
1033     $                          - W( WINDEX )-WERR( WINDEX) )
1034                        ENDIF
1035                     ENDIF
1036                     IDONE = IDONE + 1
1037                  ENDIF
1038*                 here ends the code for the current child
1039*
1040 139              CONTINUE
1041*                 Proceed to any remaining child nodes
1042                  NEWFST = J + 1
1043 140           CONTINUE
1044 150        CONTINUE
1045            NDEPTH = NDEPTH + 1
1046            GO TO 40
1047         END IF
1048         IBEGIN = IEND + 1
1049         WBEGIN = WEND + 1
1050 170  CONTINUE
1051*
1052
1053      RETURN
1054*
1055*     End of ZLARRV
1056*
1057      END
1058