1       SUBROUTINE DSTEVX( JOBZ, RANGE, N, D, E, VL, VU, IL, IU, ABSTOL,
  2      $                   M, W, Z, LDZ, WORK, IWORK, IFAIL, INFO )
  3 *
  4 *  -- LAPACK driver routine (version 3.2) --
  5 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  6 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  7 *     November 2006
  8 *
  9 *     .. Scalar Arguments ..
 10       CHARACTER          JOBZ, RANGE
 11       INTEGER            IL, INFO, IU, LDZ, M, N
 12       DOUBLE PRECISION   ABSTOL, VL, VU
 13 *     ..
 14 *     .. Array Arguments ..
 15       INTEGER            IFAIL( * ), IWORK( * )
 16       DOUBLE PRECISION   D( * ), E( * ), W( * ), WORK( * ), Z( LDZ, * )
 17 *     ..
 18 *
 19 *  Purpose
 20 *  =======
 21 *
 22 *  DSTEVX computes selected eigenvalues and, optionally, eigenvectors
 23 *  of a real symmetric tridiagonal matrix A.  Eigenvalues and
 24 *  eigenvectors can be selected by specifying either a range of values
 25 *  or a range of indices for the desired eigenvalues.
 26 *
 27 *  Arguments
 28 *  =========
 29 *
 30 *  JOBZ    (input) CHARACTER*1
 31 *          = 'N':  Compute eigenvalues only;
 32 *          = 'V':  Compute eigenvalues and eigenvectors.
 33 *
 34 *  RANGE   (input) CHARACTER*1
 35 *          = 'A': all eigenvalues will be found.
 36 *          = 'V': all eigenvalues in the half-open interval (VL,VU]
 37 *                 will be found.
 38 *          = 'I': the IL-th through IU-th eigenvalues will be found.
 39 *
 40 *  N       (input) INTEGER
 41 *          The order of the matrix.  N >= 0.
 42 *
 43 *  D       (input/output) DOUBLE PRECISION array, dimension (N)
 44 *          On entry, the n diagonal elements of the tridiagonal matrix
 45 *          A.
 46 *          On exit, D may be multiplied by a constant factor chosen
 47 *          to avoid over/underflow in computing the eigenvalues.
 48 *
 49 *  E       (input/output) DOUBLE PRECISION array, dimension (max(1,N-1))
 50 *          On entry, the (n-1) subdiagonal elements of the tridiagonal
 51 *          matrix A in elements 1 to N-1 of E.
 52 *          On exit, E may be multiplied by a constant factor chosen
 53 *          to avoid over/underflow in computing the eigenvalues.
 54 *
 55 *  VL      (input) DOUBLE PRECISION
 56 *  VU      (input) DOUBLE PRECISION
 57 *          If RANGE='V', the lower and upper bounds of the interval to
 58 *          be searched for eigenvalues. VL < VU.
 59 *          Not referenced if RANGE = 'A' or 'I'.
 60 *
 61 *  IL      (input) INTEGER
 62 *  IU      (input) INTEGER
 63 *          If RANGE='I', the indices (in ascending order) of the
 64 *          smallest and largest eigenvalues to be returned.
 65 *          1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0.
 66 *          Not referenced if RANGE = 'A' or 'V'.
 67 *
 68 *  ABSTOL  (input) DOUBLE PRECISION
 69 *          The absolute error tolerance for the eigenvalues.
 70 *          An approximate eigenvalue is accepted as converged
 71 *          when it is determined to lie in an interval [a,b]
 72 *          of width less than or equal to
 73 *
 74 *                  ABSTOL + EPS *   max( |a|,|b| ) ,
 75 *
 76 *          where EPS is the machine precision.  If ABSTOL is less
 77 *          than or equal to zero, then  EPS*|T|  will be used in
 78 *          its place, where |T| is the 1-norm of the tridiagonal
 79 *          matrix.
 80 *
 81 *          Eigenvalues will be computed most accurately when ABSTOL is
 82 *          set to twice the underflow threshold 2*DLAMCH('S'), not zero.
 83 *          If this routine returns with INFO>0, indicating that some
 84 *          eigenvectors did not converge, try setting ABSTOL to
 85 *          2*DLAMCH('S').
 86 *
 87 *          See "Computing Small Singular Values of Bidiagonal Matrices
 88 *          with Guaranteed High Relative Accuracy," by Demmel and
 89 *          Kahan, LAPACK Working Note #3.
 90 *
 91 *  M       (output) INTEGER
 92 *          The total number of eigenvalues found.  0 <= M <= N.
 93 *          If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1.
 94 *
 95 *  W       (output) DOUBLE PRECISION array, dimension (N)
 96 *          The first M elements contain the selected eigenvalues in
 97 *          ascending order.
 98 *
 99 *  Z       (output) DOUBLE PRECISION array, dimension (LDZ, max(1,M) )
100 *          If JOBZ = 'V', then if INFO = 0, the first M columns of Z
101 *          contain the orthonormal eigenvectors of the matrix A
102 *          corresponding to the selected eigenvalues, with the i-th
103 *          column of Z holding the eigenvector associated with W(i).
104 *          If an eigenvector fails to converge (INFO > 0), then that
105 *          column of Z contains the latest approximation to the
106 *          eigenvector, and the index of the eigenvector is returned
107 *          in IFAIL.  If JOBZ = 'N', then Z is not referenced.
108 *          Note: the user must ensure that at least max(1,M) columns are
109 *          supplied in the array Z; if RANGE = 'V', the exact value of M
110 *          is not known in advance and an upper bound must be used.
111 *
112 *  LDZ     (input) INTEGER
113 *          The leading dimension of the array Z.  LDZ >= 1, and if
114 *          JOBZ = 'V', LDZ >= max(1,N).
115 *
116 *  WORK    (workspace) DOUBLE PRECISION array, dimension (5*N)
117 *
118 *  IWORK   (workspace) INTEGER array, dimension (5*N)
119 *
120 *  IFAIL   (output) INTEGER array, dimension (N)
121 *          If JOBZ = 'V', then if INFO = 0, the first M elements of
122 *          IFAIL are zero.  If INFO > 0, then IFAIL contains the
123 *          indices of the eigenvectors that failed to converge.
124 *          If JOBZ = 'N', then IFAIL is not referenced.
125 *
126 *  INFO    (output) INTEGER
127 *          = 0:  successful exit
128 *          < 0:  if INFO = -i, the i-th argument had an illegal value
129 *          > 0:  if INFO = i, then i eigenvectors failed to converge.
130 *                Their indices are stored in array IFAIL.
131 *
132 *  =====================================================================
133 *
134 *     .. Parameters ..
135       DOUBLE PRECISION   ZERO, ONE
136       PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0 )
137 *     ..
138 *     .. Local Scalars ..
139       LOGICAL            ALLEIG, INDEIG, TEST, VALEIG, WANTZ
140       CHARACTER          ORDER
141       INTEGER            I, IMAX, INDIBL, INDISP, INDIWO, INDWRK,
142      $                   ISCALE, ITMP1, J, JJ, NSPLIT
143       DOUBLE PRECISION   BIGNUM, EPS, RMAX, RMIN, SAFMIN, SIGMA, SMLNUM,
144      $                   TMP1, TNRM, VLL, VUU
145 *     ..
146 *     .. External Functions ..
147       LOGICAL            LSAME
148       DOUBLE PRECISION   DLAMCH, DLANST
149       EXTERNAL           LSAME, DLAMCH, DLANST
150 *     ..
151 *     .. External Subroutines ..
152       EXTERNAL           DCOPY, DSCAL, DSTEBZ, DSTEIN, DSTEQR, DSTERF,
153      $                   DSWAP, XERBLA
154 *     ..
155 *     .. Intrinsic Functions ..
156       INTRINSIC          MAXMINSQRT
157 *     ..
158 *     .. Executable Statements ..
159 *
160 *     Test the input parameters.
161 *
162       WANTZ = LSAME( JOBZ, 'V' )
163       ALLEIG = LSAME( RANGE'A' )
164       VALEIG = LSAME( RANGE'V' )
165       INDEIG = LSAME( RANGE'I' )
166 *
167       INFO = 0
168       IF.NOT.( WANTZ .OR. LSAME( JOBZ, 'N' ) ) ) THEN
169          INFO = -1
170       ELSE IF.NOT.( ALLEIG .OR. VALEIG .OR. INDEIG ) ) THEN
171          INFO = -2
172       ELSE IF( N.LT.0 ) THEN
173          INFO = -3
174       ELSE
175          IF( VALEIG ) THEN
176             IF( N.GT.0 .AND. VU.LE.VL )
177      $         INFO = -7
178          ELSE IF( INDEIG ) THEN
179             IF( IL.LT.1 .OR. IL.GT.MAX1, N ) ) THEN
180                INFO = -8
181             ELSE IF( IU.LT.MIN( N, IL ) .OR. IU.GT.N ) THEN
182                INFO = -9
183             END IF
184          END IF
185       END IF
186       IF( INFO.EQ.0 ) THEN
187          IF( LDZ.LT.1 .OR. ( WANTZ .AND. LDZ.LT.N ) )
188      $      INFO = -14
189       END IF
190 *
191       IF( INFO.NE.0 ) THEN
192          CALL XERBLA( 'DSTEVX'-INFO )
193          RETURN
194       END IF
195 *
196 *     Quick return if possible
197 *
198       M = 0
199       IF( N.EQ.0 )
200      $   RETURN
201 *
202       IF( N.EQ.1 ) THEN
203          IF( ALLEIG .OR. INDEIG ) THEN
204             M = 1
205             W( 1 ) = D( 1 )
206          ELSE
207             IF( VL.LT.D( 1 ) .AND. VU.GE.D( 1 ) ) THEN
208                M = 1
209                W( 1 ) = D( 1 )
210             END IF
211          END IF
212          IF( WANTZ )
213      $      Z( 11 ) = ONE
214          RETURN
215       END IF
216 *
217 *     Get machine constants.
218 *
219       SAFMIN = DLAMCH( 'Safe minimum' )
220       EPS = DLAMCH( 'Precision' )
221       SMLNUM = SAFMIN / EPS
222       BIGNUM = ONE / SMLNUM
223       RMIN = SQRT( SMLNUM )
224       RMAX = MINSQRT( BIGNUM ), ONE / SQRTSQRT( SAFMIN ) ) )
225 *
226 *     Scale matrix to allowable range, if necessary.
227 *
228       ISCALE = 0
229       IF( VALEIG ) THEN
230          VLL = VL
231          VUU = VU
232       ELSE
233          VLL = ZERO
234          VUU = ZERO
235       END IF
236       TNRM = DLANST( 'M', N, D, E )
237       IF( TNRM.GT.ZERO .AND. TNRM.LT.RMIN ) THEN
238          ISCALE = 1
239          SIGMA = RMIN / TNRM
240       ELSE IF( TNRM.GT.RMAX ) THEN
241          ISCALE = 1
242          SIGMA = RMAX / TNRM
243       END IF
244       IF( ISCALE.EQ.1 ) THEN
245          CALL DSCAL( N, SIGMA, D, 1 )
246          CALL DSCAL( N-1, SIGMA, E( 1 ), 1 )
247          IF( VALEIG ) THEN
248             VLL = VL*SIGMA
249             VUU = VU*SIGMA
250          END IF
251       END IF
252 *
253 *     If all eigenvalues are desired and ABSTOL is less than zero, then
254 *     call DSTERF or SSTEQR.  If this fails for some eigenvalue, then
255 *     try DSTEBZ.
256 *
257       TEST = .FALSE.
258       IF( INDEIG ) THEN
259          IF( IL.EQ.1 .AND. IU.EQ.N ) THEN
260             TEST = .TRUE.
261          END IF
262       END IF
263       IF( ( ALLEIG .OR. TEST ) .AND. ( ABSTOL.LE.ZERO ) ) THEN
264          CALL DCOPY( N, D, 1, W, 1 )
265          CALL DCOPY( N-1, E( 1 ), 1, WORK( 1 ), 1 )
266          INDWRK = N + 1
267          IF.NOT.WANTZ ) THEN
268             CALL DSTERF( N, W, WORK, INFO )
269          ELSE
270             CALL DSTEQR( 'I', N, W, WORK, Z, LDZ, WORK( INDWRK ), INFO )
271             IF( INFO.EQ.0 ) THEN
272                DO 10 I = 1, N
273                   IFAIL( I ) = 0
274    10          CONTINUE
275             END IF
276          END IF
277          IF( INFO.EQ.0 ) THEN
278             M = N
279             GO TO 20
280          END IF
281          INFO = 0
282       END IF
283 *
284 *     Otherwise, call DSTEBZ and, if eigenvectors are desired, SSTEIN.
285 *
286       IF( WANTZ ) THEN
287          ORDER = 'B'
288       ELSE
289          ORDER = 'E'
290       END IF
291       INDWRK = 1
292       INDIBL = 1
293       INDISP = INDIBL + N
294       INDIWO = INDISP + N
295       CALL DSTEBZ( RANGE, ORDER, N, VLL, VUU, IL, IU, ABSTOL, D, E, M,
296      $             NSPLIT, W, IWORK( INDIBL ), IWORK( INDISP ),
297      $             WORK( INDWRK ), IWORK( INDIWO ), INFO )
298 *
299       IF( WANTZ ) THEN
300          CALL DSTEIN( N, D, E, M, W, IWORK( INDIBL ), IWORK( INDISP ),
301      $                Z, LDZ, WORK( INDWRK ), IWORK( INDIWO ), IFAIL,
302      $                INFO )
303       END IF
304 *
305 *     If matrix was scaled, then rescale eigenvalues appropriately.
306 *
307    20 CONTINUE
308       IF( ISCALE.EQ.1 ) THEN
309          IF( INFO.EQ.0 ) THEN
310             IMAX = M
311          ELSE
312             IMAX = INFO - 1
313          END IF
314          CALL DSCAL( IMAX, ONE / SIGMA, W, 1 )
315       END IF
316 *
317 *     If eigenvalues are not in order, then sort them, along with
318 *     eigenvectors.
319 *
320       IF( WANTZ ) THEN
321          DO 40 J = 1, M - 1
322             I = 0
323             TMP1 = W( J )
324             DO 30 JJ = J + 1, M
325                IF( W( JJ ).LT.TMP1 ) THEN
326                   I = JJ
327                   TMP1 = W( JJ )
328                END IF
329    30       CONTINUE
330 *
331             IF( I.NE.0 ) THEN
332                ITMP1 = IWORK( INDIBL+I-1 )
333                W( I ) = W( J )
334                IWORK( INDIBL+I-1 ) = IWORK( INDIBL+J-1 )
335                W( J ) = TMP1
336                IWORK( INDIBL+J-1 ) = ITMP1
337                CALL DSWAP( N, Z( 1, I ), 1, Z( 1, J ), 1 )
338                IF( INFO.NE.0 ) THEN
339                   ITMP1 = IFAIL( I )
340                   IFAIL( I ) = IFAIL( J )
341                   IFAIL( J ) = ITMP1
342                END IF
343             END IF
344    40    CONTINUE
345       END IF
346 *
347       RETURN
348 *
349 *     End of DSTEVX
350 *
351       END