1
       2
       3
       4
       5
       6
       7
       8
       9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      23
      24
      25
      26
      27
      28
      29
      30
      31
      32
      33
      34
      35
      36
      37
      38
      39
      40
      41
      42
      43
      44
      45
      46
      47
      48
      49
      50
      51
      52
      53
      54
      55
      56
      57
      58
      59
      60
      61
      62
      63
      64
      65
      66
      67
      68
      69
      70
      71
      72
      73
      74
      75
      76
      77
      78
      79
      80
      81
      82
      83
      84
      85
      86
      87
      88
      89
      90
      91
      92
      93
      94
      95
      96
      97
      98
      99
     100
     101
     102
     103
     104
     105
     106
     107
     108
     109
     110
     111
     112
     113
     114
     115
     116
     117
     118
     119
     120
     121
     122
     123
     124
     125
     126
     127
     128
     129
     130
     131
     132
     133
     134
     135
     136
     137
     138
     139
     140
     141
     142
     143
     144
     145
     146
     147
     148
     149
     150
     151
     152
     153
     154
     155
     156
     157
     158
     159
     160
     161
     162
     163
     164
     165
     166
     167
     168
     169
     170
     171
     172
     173
     174
     175
     176
     177
     178
     179
     180
     181
     182
     183
     184
     185
     186
     187
     188
     189
     190
     191
     192
     193
     194
     195
     196
     197
     198
     199
     200
     201
     202
     203
     204
     205
     206
     207
     208
     209
     210
     211
     212
     213
     214
     215
     216
     217
     218
     219
     220
     221
     222
     223
     224
     225
     226
     227
     228
     229
     230
     231
     232
     233
     234
     235
     236
     237
     238
     239
     240
     241
     242
     243
     244
     245
     246
     247
     248
     249
     250
     251
     252
     253
     254
     255
     256
     257
     258
     259
     260
     261
     262
     263
     264
     265
     266
     267
     268
     269
     270
     271
     272
     273
     274
     275
     276
     277
     278
     279
     280
     281
     282
     283
     284
     285
     286
     287
     288
     289
     290
     291
     292
     293
     294
     295
     296
     297
     298
     299
     300
     301
     302
     303
     304
     305
     306
     307
     308
     309
     310
     311
     312
     313
     314
     315
     316
     317
     318
     319
     320
     321
     322
     323
     324
     325
     326
     327
     328
     329
     330
     331
     332
     333
     334
     335
     336
     337
     338
     339
     340
     341
     342
     343
     344
     345
     346
     347
     348
     349
     350
     351
     352
     353
     354
     355
     356
     357
     358
     359
     360
     361
     362
     363
     364
     365
     366
     367
     368
     369
     370
     371
     372
     373
     374
     375
     376
     377
     378
     379
     380
     381
     382
     383
     384
     385
     386
     387
     388
     389
     390
     391
     392
     393
     394
     395
     396
     397
     398
     399
     400
     401
     402
     403
     404
     405
     406
     407
     408
     409
     410
     411
     412
     413
     414
     415
     416
     417
     418
     419
     420
     421
     422
     423
     424
     425
     426
     427
     428
     429
     430
     431
     432
     433
     434
     435
     436
     437
     438
     439
     440
     441
     442
     443
     444
     445
     446
     447
     448
     449
     450
     451
     452
     453
     454
     455
     456
     457
     458
     459
     460
     461
     462
     463
     464
     465
     466
     467
     468
     469
     470
     471
     472
     473
     474
     475
     476
     477
     478
     479
     480
     481
     482
     483
     484
     485
     486
     487
     488
     489
     490
     491
     492
     493
     494
     495
     496
      SUBROUTINE DTRSNA( JOB, HOWMNY, SELECT, N, T, LDT, VL, LDVL, VR,
     $                   LDVR, S, SEP, MM, M, WORK, LDWORK, IWORK,
     $                   INFO )
*
*  -- LAPACK routine (version 3.3.1) --
*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
*  -- April 2011                                                      --
*
*     Modified to call DLACN2 in place of DLACON, 5 Feb 03, SJH.
*
*     .. Scalar Arguments ..
      CHARACTER          HOWMNY, JOB
      INTEGER            INFO, LDT, LDVL, LDVR, LDWORK, M, MM, N
*     ..
*     .. Array Arguments ..
      LOGICAL            SELECT* )
      INTEGER            IWORK( * )
      DOUBLE PRECISION   S( * ), SEP( * ), T( LDT, * ), VL( LDVL, * ),
     $                   VR( LDVR, * ), WORK( LDWORK, * )
*     ..
*
*  Purpose
*  =======
*
*  DTRSNA estimates reciprocal condition numbers for specified
*  eigenvalues and/or right eigenvectors of a real upper
*  quasi-triangular matrix T (or of any matrix Q*T*Q**T with Q
*  orthogonal).
*
*  T must be in Schur canonical form (as returned by DHSEQR), that is,
*  block upper triangular with 1-by-1 and 2-by-2 diagonal blocks; each
*  2-by-2 diagonal block has its diagonal elements equal and its
*  off-diagonal elements of opposite sign.
*
*  Arguments
*  =========
*
*  JOB     (input) CHARACTER*1
*          Specifies whether condition numbers are required for
*          eigenvalues (S) or eigenvectors (SEP):
*          = 'E': for eigenvalues only (S);
*          = 'V': for eigenvectors only (SEP);
*          = 'B': for both eigenvalues and eigenvectors (S and SEP).
*
*  HOWMNY  (input) CHARACTER*1
*          = 'A': compute condition numbers for all eigenpairs;
*          = 'S': compute condition numbers for selected eigenpairs
*                 specified by the array SELECT.
*
*  SELECT  (input) LOGICAL array, dimension (N)
*          If HOWMNY = 'S', SELECT specifies the eigenpairs for which
*          condition numbers are required. To select condition numbers
*          for the eigenpair corresponding to a real eigenvalue w(j),
*          SELECT(j) must be set to .TRUE.. To select condition numbers
*          corresponding to a complex conjugate pair of eigenvalues w(j)
*          and w(j+1), either SELECT(j) or SELECT(j+1) or both, must be
*          set to .TRUE..
*          If HOWMNY = 'A', SELECT is not referenced.
*
*  N       (input) INTEGER
*          The order of the matrix T. N >= 0.
*
*  T       (input) DOUBLE PRECISION array, dimension (LDT,N)
*          The upper quasi-triangular matrix T, in Schur canonical form.
*
*  LDT     (input) INTEGER
*          The leading dimension of the array T. LDT >= max(1,N).
*
*  VL      (input) DOUBLE PRECISION array, dimension (LDVL,M)
*          If JOB = 'E' or 'B', VL must contain left eigenvectors of T
*          (or of any Q*T*Q**T with Q orthogonal), corresponding to the
*          eigenpairs specified by HOWMNY and SELECT. The eigenvectors
*          must be stored in consecutive columns of VL, as returned by
*          DHSEIN or DTREVC.
*          If JOB = 'V', VL is not referenced.
*
*  LDVL    (input) INTEGER
*          The leading dimension of the array VL.
*          LDVL >= 1; and if JOB = 'E' or 'B', LDVL >= N.
*
*  VR      (input) DOUBLE PRECISION array, dimension (LDVR,M)
*          If JOB = 'E' or 'B', VR must contain right eigenvectors of T
*          (or of any Q*T*Q**T with Q orthogonal), corresponding to the
*          eigenpairs specified by HOWMNY and SELECT. The eigenvectors
*          must be stored in consecutive columns of VR, as returned by
*          DHSEIN or DTREVC.
*          If JOB = 'V', VR is not referenced.
*
*  LDVR    (input) INTEGER
*          The leading dimension of the array VR.
*          LDVR >= 1; and if JOB = 'E' or 'B', LDVR >= N.
*
*  S       (output) DOUBLE PRECISION array, dimension (MM)
*          If JOB = 'E' or 'B', the reciprocal condition numbers of the
*          selected eigenvalues, stored in consecutive elements of the
*          array. For a complex conjugate pair of eigenvalues two
*          consecutive elements of S are set to the same value. Thus
*          S(j), SEP(j), and the j-th columns of VL and VR all
*          correspond to the same eigenpair (but not in general the
*          j-th eigenpair, unless all eigenpairs are selected).
*          If JOB = 'V', S is not referenced.
*
*  SEP     (output) DOUBLE PRECISION array, dimension (MM)
*          If JOB = 'V' or 'B', the estimated reciprocal condition
*          numbers of the selected eigenvectors, stored in consecutive
*          elements of the array. For a complex eigenvector two
*          consecutive elements of SEP are set to the same value. If
*          the eigenvalues cannot be reordered to compute SEP(j), SEP(j)
*          is set to 0; this can only occur when the true value would be
*          very small anyway.
*          If JOB = 'E', SEP is not referenced.
*
*  MM      (input) INTEGER
*          The number of elements in the arrays S (if JOB = 'E' or 'B')
*           and/or SEP (if JOB = 'V' or 'B'). MM >= M.
*
*  M       (output) INTEGER
*          The number of elements of the arrays S and/or SEP actually
*          used to store the estimated condition numbers.
*          If HOWMNY = 'A', M is set to N.
*
*  WORK    (workspace) DOUBLE PRECISION array, dimension (LDWORK,N+6)
*          If JOB = 'E', WORK is not referenced.
*
*  LDWORK  (input) INTEGER
*          The leading dimension of the array WORK.
*          LDWORK >= 1; and if JOB = 'V' or 'B', LDWORK >= N.
*
*  IWORK   (workspace) INTEGER array, dimension (2*(N-1))
*          If JOB = 'E', IWORK is not referenced.
*
*  INFO    (output) INTEGER
*          = 0: successful exit
*          < 0: if INFO = -i, the i-th argument had an illegal value
*
*  Further Details
*  ===============
*
*  The reciprocal of the condition number of an eigenvalue lambda is
*  defined as
*
*          S(lambda) = |v**T*u| / (norm(u)*norm(v))
*
*  where u and v are the right and left eigenvectors of T corresponding
*  to lambda; v**T denotes the transpose of v, and norm(u)
*  denotes the Euclidean norm. These reciprocal condition numbers always
*  lie between zero (very badly conditioned) and one (very well
*  conditioned). If n = 1, S(lambda) is defined to be 1.
*
*  An approximate error bound for a computed eigenvalue W(i) is given by
*
*                      EPS * norm(T) / S(i)
*
*  where EPS is the machine precision.
*
*  The reciprocal of the condition number of the right eigenvector u
*  corresponding to lambda is defined as follows. Suppose
*
*              T = ( lambda  c  )
*                  (   0    T22 )
*
*  Then the reciprocal condition number is
*
*          SEP( lambda, T22 ) = sigma-min( T22 - lambda*I )
*
*  where sigma-min denotes the smallest singular value. We approximate
*  the smallest singular value by the reciprocal of an estimate of the
*  one-norm of the inverse of T22 - lambda*I. If n = 1, SEP(1) is
*  defined to be abs(T(1,1)).
*
*  An approximate error bound for a computed right eigenvector VR(i)
*  is given by
*
*                      EPS * norm(T) / SEP(i)
*
*  =====================================================================
*
*     .. Parameters ..
      DOUBLE PRECISION   ZERO, ONE, TWO
      PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0, TWO = 2.0D+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            PAIR, SOMCON, WANTBH, WANTS, WANTSP
      INTEGER            I, IERR, IFST, ILST, J, K, KASE, KS, N2, NN
      DOUBLE PRECISION   BIGNUM, COND, CS, DELTA, DUMM, EPS, EST, LNRM,
     $                   MU, PROD, PROD1, PROD2, RNRM, SCALE, SMLNUM, SN
*     ..
*     .. Local Arrays ..
      INTEGER            ISAVE( 3 )
      DOUBLE PRECISION   DUMMY( 1 )
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      DOUBLE PRECISION   DDOT, DLAMCH, DLAPY2, DNRM2
      EXTERNAL           LSAME, DDOT, DLAMCH, DLAPY2, DNRM2
*     ..
*     .. External Subroutines ..
      EXTERNAL           DLACN2, DLACPY, DLAQTR, DTREXC, XERBLA
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          ABSMAXSQRT
*     ..
*     .. Executable Statements ..
*
*     Decode and test the input parameters
*
      WANTBH = LSAME( JOB, 'B' )
      WANTS = LSAME( JOB, 'E' ) .OR. WANTBH
      WANTSP = LSAME( JOB, 'V' ) .OR. WANTBH
*
      SOMCON = LSAME( HOWMNY, 'S' )
*
      INFO = 0
      IF.NOT.WANTS .AND. .NOT.WANTSP ) THEN
         INFO = -1
      ELSE IF.NOT.LSAME( HOWMNY, 'A' ) .AND. .NOT.SOMCON ) THEN
         INFO = -2
      ELSE IF( N.LT.0 ) THEN
         INFO = -4
      ELSE IF( LDT.LT.MAX1, N ) ) THEN
         INFO = -6
      ELSE IF( LDVL.LT.1 .OR. ( WANTS .AND. LDVL.LT.N ) ) THEN
         INFO = -8
      ELSE IF( LDVR.LT.1 .OR. ( WANTS .AND. LDVR.LT.N ) ) THEN
         INFO = -10
      ELSE
*
*        Set M to the number of eigenpairs for which condition numbers
*        are required, and test MM.
*
         IF( SOMCON ) THEN
            M = 0
            PAIR = .FALSE.
            DO 10 K = 1, N
               IF( PAIR ) THEN
                  PAIR = .FALSE.
               ELSE
                  IF( K.LT.N ) THEN
                     IF( T( K+1, K ).EQ.ZERO ) THEN
                        IFSELECT( K ) )
     $                     M = M + 1
                     ELSE
                        PAIR = .TRUE.
                        IFSELECT( K ) .OR. SELECT( K+1 ) )
     $                     M = M + 2
                     END IF
                  ELSE
                     IFSELECT( N ) )
     $                  M = M + 1
                  END IF
               END IF
   10       CONTINUE
         ELSE
            M = N
         END IF
*
         IF( MM.LT.M ) THEN
            INFO = -13
         ELSE IF( LDWORK.LT.1 .OR. ( WANTSP .AND. LDWORK.LT.N ) ) THEN
            INFO = -16
         END IF
      END IF
      IF( INFO.NE.0 ) THEN
         CALL XERBLA( 'DTRSNA'-INFO )
         RETURN
      END IF
*
*     Quick return if possible
*
      IF( N.EQ.0 )
     $   RETURN
*
      IF( N.EQ.1 ) THEN
         IF( SOMCON ) THEN
            IF.NOT.SELECT1 ) )
     $         RETURN
         END IF
         IF( WANTS )
     $      S( 1 ) = ONE
         IF( WANTSP )
     $      SEP( 1 ) = ABS( T( 11 ) )
         RETURN
      END IF
*
*     Get machine constants
*
      EPS = DLAMCH( 'P' )
      SMLNUM = DLAMCH( 'S' ) / EPS
      BIGNUM = ONE / SMLNUM
      CALL DLABAD( SMLNUM, BIGNUM )
*
      KS = 0
      PAIR = .FALSE.
      DO 60 K = 1, N
*
*        Determine whether T(k,k) begins a 1-by-1 or 2-by-2 block.
*
         IF( PAIR ) THEN
            PAIR = .FALSE.
            GO TO 60
         ELSE
            IF( K.LT.N )
     $         PAIR = T( K+1, K ).NE.ZERO
         END IF
*
*        Determine whether condition numbers are required for the k-th
*        eigenpair.
*
         IF( SOMCON ) THEN
            IF( PAIR ) THEN
               IF.NOT.SELECT( K ) .AND. .NOT.SELECT( K+1 ) )
     $            GO TO 60
            ELSE
               IF.NOT.SELECT( K ) )
     $            GO TO 60
            END IF
         END IF
*
         KS = KS + 1
*
         IF( WANTS ) THEN
*
*           Compute the reciprocal condition number of the k-th
*           eigenvalue.
*
            IF.NOT.PAIR ) THEN
*
*              Real eigenvalue.
*
               PROD = DDOT( N, VR( 1, KS ), 1, VL( 1, KS ), 1 )
               RNRM = DNRM2( N, VR( 1, KS ), 1 )
               LNRM = DNRM2( N, VL( 1, KS ), 1 )
               S( KS ) = ABS( PROD ) / ( RNRM*LNRM )
            ELSE
*
*              Complex eigenvalue.
*
               PROD1 = DDOT( N, VR( 1, KS ), 1, VL( 1, KS ), 1 )
               PROD1 = PROD1 + DDOT( N, VR( 1, KS+1 ), 1, VL( 1, KS+1 ),
     $                 1 )
               PROD2 = DDOT( N, VL( 1, KS ), 1, VR( 1, KS+1 ), 1 )
               PROD2 = PROD2 - DDOT( N, VL( 1, KS+1 ), 1, VR( 1, KS ),
     $                 1 )
               RNRM = DLAPY2( DNRM2( N, VR( 1, KS ), 1 ),
     $                DNRM2( N, VR( 1, KS+1 ), 1 ) )
               LNRM = DLAPY2( DNRM2( N, VL( 1, KS ), 1 ),
     $                DNRM2( N, VL( 1, KS+1 ), 1 ) )
               COND = DLAPY2( PROD1, PROD2 ) / ( RNRM*LNRM )
               S( KS ) = COND
               S( KS+1 ) = COND
            END IF
         END IF
*
         IF( WANTSP ) THEN
*
*           Estimate the reciprocal condition number of the k-th
*           eigenvector.
*
*           Copy the matrix T to the array WORK and swap the diagonal
*           block beginning at T(k,k) to the (1,1) position.
*
            CALL DLACPY( 'Full', N, N, T, LDT, WORK, LDWORK )
            IFST = K
            ILST = 1
            CALL DTREXC( 'No Q', N, WORK, LDWORK, DUMMY, 1, IFST, ILST,
     $                   WORK( 1, N+1 ), IERR )
*
            IF( IERR.EQ.1 .OR. IERR.EQ.2 ) THEN
*
*              Could not swap because blocks not well separated
*
               SCALE = ONE
               EST = BIGNUM
            ELSE
*
*              Reordering successful
*
               IF( WORK( 21 ).EQ.ZERO ) THEN
*
*                 Form C = T22 - lambda*I in WORK(2:N,2:N).
*
                  DO 20 I = 2, N
                     WORK( I, I ) = WORK( I, I ) - WORK( 11 )
   20             CONTINUE
                  N2 = 1
                  NN = N - 1
               ELSE
*
*                 Triangularize the 2 by 2 block by unitary
*                 transformation U = [  cs   i*ss ]
*                                    [ i*ss   cs  ].
*                 such that the (1,1) position of WORK is complex
*                 eigenvalue lambda with positive imaginary part. (2,2)
*                 position of WORK is the complex eigenvalue lambda
*                 with negative imaginary  part.
*
                  MU = SQRTABS( WORK( 12 ) ) )*
     $                 SQRTABS( WORK( 21 ) ) )
                  DELTA = DLAPY2( MU, WORK( 21 ) )
                  CS = MU / DELTA
                  SN = -WORK( 21 ) / DELTA
*
*                 Form
*
*                 C**T = WORK(2:N,2:N) + i*[rwork(1) ..... rwork(n-1) ]
*                                          [   mu                     ]
*                                          [         ..               ]
*                                          [             ..           ]
*                                          [                  mu      ]
*                 where C**T is transpose of matrix C,
*                 and RWORK is stored starting in the N+1-st column of
*                 WORK.
*
                  DO 30 J = 3, N
                     WORK( 2, J ) = CS*WORK( 2, J )
                     WORK( J, J ) = WORK( J, J ) - WORK( 11 )
   30             CONTINUE
                  WORK( 22 ) = ZERO
*
                  WORK( 1, N+1 ) = TWO*MU
                  DO 40 I = 2, N - 1
                     WORK( I, N+1 ) = SN*WORK( 1, I+1 )
   40             CONTINUE
                  N2 = 2
                  NN = 2*( N-1 )
               END IF
*
*              Estimate norm(inv(C**T))
*
               EST = ZERO
               KASE = 0
   50          CONTINUE
               CALL DLACN2( NN, WORK( 1, N+2 ), WORK( 1, N+4 ), IWORK,
     $                      EST, KASE, ISAVE )
               IF( KASE.NE.0 ) THEN
                  IF( KASE.EQ.1 ) THEN
                     IF( N2.EQ.1 ) THEN
*
*                       Real eigenvalue: solve C**T*x = scale*c.
*
                        CALL DLAQTR( .TRUE..TRUE., N-1, WORK( 22 ),
     $                               LDWORK, DUMMY, DUMM, SCALE,
     $                               WORK( 1, N+4 ), WORK( 1, N+6 ),
     $                               IERR )
                     ELSE
*
*                       Complex eigenvalue: solve
*                       C**T*(p+iq) = scale*(c+id) in real arithmetic.
*
                        CALL DLAQTR( .TRUE..FALSE., N-1, WORK( 22 ),
     $                               LDWORK, WORK( 1, N+1 ), MU, SCALE,
     $                               WORK( 1, N+4 ), WORK( 1, N+6 ),
     $                               IERR )
                     END IF
                  ELSE
                     IF( N2.EQ.1 ) THEN
*
*                       Real eigenvalue: solve C*x = scale*c.
*
                        CALL DLAQTR( .FALSE..TRUE., N-1, WORK( 22 ),
     $                               LDWORK, DUMMY, DUMM, SCALE,
     $                               WORK( 1, N+4 ), WORK( 1, N+6 ),
     $                               IERR )
                     ELSE
*
*                       Complex eigenvalue: solve
*                       C*(p+iq) = scale*(c+id) in real arithmetic.
*
                        CALL DLAQTR( .FALSE..FALSE., N-1,
     $                               WORK( 22 ), LDWORK,
     $                               WORK( 1, N+1 ), MU, SCALE,
     $                               WORK( 1, N+4 ), WORK( 1, N+6 ),
     $                               IERR )
*
                     END IF
                  END IF
*
                  GO TO 50
               END IF
            END IF
*
            SEP( KS ) = SCALE / MAX( EST, SMLNUM )
            IF( PAIR )
     $         SEP( KS+1 ) = SEP( KS )
         END IF
*
         IF( PAIR )
     $      KS = KS + 1
*
   60 CONTINUE
      RETURN
*
*     End of DTRSNA
*
      END