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
      SUBROUTINE SPBSVX( FACT, UPLO, N, KD, NRHS, AB, LDAB, AFB, LDAFB,
     $                   EQUED, S, B, LDB, X, LDX, RCOND, FERR, BERR,
     $                   WORK, IWORK, INFO )
*
*  -- LAPACK driver 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                                                      --
*
*     .. Scalar Arguments ..
      CHARACTER          EQUED, FACT, UPLO
      INTEGER            INFO, KD, LDAB, LDAFB, LDB, LDX, N, NRHS
      REAL               RCOND
*     ..
*     .. Array Arguments ..
      INTEGER            IWORK( * )
      REAL               AB( LDAB, * ), AFB( LDAFB, * ), B( LDB, * ),
     $                   BERR( * ), FERR( * ), S( * ), WORK( * ),
     $                   X( LDX, * )
*     ..
*
*  Purpose
*  =======
*
*  SPBSVX uses the Cholesky factorization A = U**T*U or A = L*L**T to
*  compute the solution to a real system of linear equations
*     A * X = B,
*  where A is an N-by-N symmetric positive definite band matrix and X
*  and B are N-by-NRHS matrices.
*
*  Error bounds on the solution and a condition estimate are also
*  provided.
*
*  Description
*  ===========
*
*  The following steps are performed:
*
*  1. If FACT = 'E', real scaling factors are computed to equilibrate
*     the system:
*        diag(S) * A * diag(S) * inv(diag(S)) * X = diag(S) * B
*     Whether or not the system will be equilibrated depends on the
*     scaling of the matrix A, but if equilibration is used, A is
*     overwritten by diag(S)*A*diag(S) and B by diag(S)*B.
*
*  2. If FACT = 'N' or 'E', the Cholesky decomposition is used to
*     factor the matrix A (after equilibration if FACT = 'E') as
*        A = U**T * U,  if UPLO = 'U', or
*        A = L * L**T,  if UPLO = 'L',
*     where U is an upper triangular band matrix, and L is a lower
*     triangular band matrix.
*
*  3. If the leading i-by-i principal minor is not positive definite,
*     then the routine returns with INFO = i. Otherwise, the factored
*     form of A is used to estimate the condition number of the matrix
*     A.  If the reciprocal of the condition number is less than machine
*     precision, INFO = N+1 is returned as a warning, but the routine
*     still goes on to solve for X and compute error bounds as
*     described below.
*
*  4. The system of equations is solved for X using the factored form
*     of A.
*
*  5. Iterative refinement is applied to improve the computed solution
*     matrix and calculate error bounds and backward error estimates
*     for it.
*
*  6. If equilibration was used, the matrix X is premultiplied by
*     diag(S) so that it solves the original system before
*     equilibration.
*
*  Arguments
*  =========
*
*  FACT    (input) CHARACTER*1
*          Specifies whether or not the factored form of the matrix A is
*          supplied on entry, and if not, whether the matrix A should be
*          equilibrated before it is factored.
*          = 'F':  On entry, AFB contains the factored form of A.
*                  If EQUED = 'Y', the matrix A has been equilibrated
*                  with scaling factors given by S.  AB and AFB will not
*                  be modified.
*          = 'N':  The matrix A will be copied to AFB and factored.
*          = 'E':  The matrix A will be equilibrated if necessary, then
*                  copied to AFB and factored.
*
*  UPLO    (input) CHARACTER*1
*          = 'U':  Upper triangle of A is stored;
*          = 'L':  Lower triangle of A is stored.
*
*  N       (input) INTEGER
*          The number of linear equations, i.e., the order of the
*          matrix A.  N >= 0.
*
*  KD      (input) INTEGER
*          The number of superdiagonals of the matrix A if UPLO = 'U',
*          or the number of subdiagonals if UPLO = 'L'.  KD >= 0.
*
*  NRHS    (input) INTEGER
*          The number of right-hand sides, i.e., the number of columns
*          of the matrices B and X.  NRHS >= 0.
*
*  AB      (input/output) REAL array, dimension (LDAB,N)
*          On entry, the upper or lower triangle of the symmetric band
*          matrix A, stored in the first KD+1 rows of the array, except
*          if FACT = 'F' and EQUED = 'Y', then A must contain the
*          equilibrated matrix diag(S)*A*diag(S).  The j-th column of A
*          is stored in the j-th column of the array AB as follows:
*          if UPLO = 'U', AB(KD+1+i-j,j) = A(i,j) for max(1,j-KD)<=i<=j;
*          if UPLO = 'L', AB(1+i-j,j)    = A(i,j) for j<=i<=min(N,j+KD).
*          See below for further details.
*
*          On exit, if FACT = 'E' and EQUED = 'Y', A is overwritten by
*          diag(S)*A*diag(S).
*
*  LDAB    (input) INTEGER
*          The leading dimension of the array A.  LDAB >= KD+1.
*
*  AFB     (input or output) REAL array, dimension (LDAFB,N)
*          If FACT = 'F', then AFB is an input argument and on entry
*          contains the triangular factor U or L from the Cholesky
*          factorization A = U**T*U or A = L*L**T of the band matrix
*          A, in the same storage format as A (see AB).  If EQUED = 'Y',
*          then AFB is the factored form of the equilibrated matrix A.
*
*          If FACT = 'N', then AFB is an output argument and on exit
*          returns the triangular factor U or L from the Cholesky
*          factorization A = U**T*U or A = L*L**T.
*
*          If FACT = 'E', then AFB is an output argument and on exit
*          returns the triangular factor U or L from the Cholesky
*          factorization A = U**T*U or A = L*L**T of the equilibrated
*          matrix A (see the description of A for the form of the
*          equilibrated matrix).
*
*  LDAFB   (input) INTEGER
*          The leading dimension of the array AFB.  LDAFB >= KD+1.
*
*  EQUED   (input or output) CHARACTER*1
*          Specifies the form of equilibration that was done.
*          = 'N':  No equilibration (always true if FACT = 'N').
*          = 'Y':  Equilibration was done, i.e., A has been replaced by
*                  diag(S) * A * diag(S).
*          EQUED is an input argument if FACT = 'F'; otherwise, it is an
*          output argument.
*
*  S       (input or output) REAL array, dimension (N)
*          The scale factors for A; not accessed if EQUED = 'N'.  S is
*          an input argument if FACT = 'F'; otherwise, S is an output
*          argument.  If FACT = 'F' and EQUED = 'Y', each element of S
*          must be positive.
*
*  B       (input/output) REAL array, dimension (LDB,NRHS)
*          On entry, the N-by-NRHS right hand side matrix B.
*          On exit, if EQUED = 'N', B is not modified; if EQUED = 'Y',
*          B is overwritten by diag(S) * B.
*
*  LDB     (input) INTEGER
*          The leading dimension of the array B.  LDB >= max(1,N).
*
*  X       (output) REAL array, dimension (LDX,NRHS)
*          If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X to
*          the original system of equations.  Note that if EQUED = 'Y',
*          A and B are modified on exit, and the solution to the
*          equilibrated system is inv(diag(S))*X.
*
*  LDX     (input) INTEGER
*          The leading dimension of the array X.  LDX >= max(1,N).
*
*  RCOND   (output) REAL
*          The estimate of the reciprocal condition number of the matrix
*          A after equilibration (if done).  If RCOND is less than the
*          machine precision (in particular, if RCOND = 0), the matrix
*          is singular to working precision.  This condition is
*          indicated by a return code of INFO > 0.
*
*  FERR    (output) REAL array, dimension (NRHS)
*          The estimated forward error bound for each solution vector
*          X(j) (the j-th column of the solution matrix X).
*          If XTRUE is the true solution corresponding to X(j), FERR(j)
*          is an estimated upper bound for the magnitude of the largest
*          element in (X(j) - XTRUE) divided by the magnitude of the
*          largest element in X(j).  The estimate is as reliable as
*          the estimate for RCOND, and is almost always a slight
*          overestimate of the true error.
*
*  BERR    (output) REAL array, dimension (NRHS)
*          The componentwise relative backward error of each solution
*          vector X(j) (i.e., the smallest relative change in
*          any element of A or B that makes X(j) an exact solution).
*
*  WORK    (workspace) REAL array, dimension (3*N)
*
*  IWORK   (workspace) INTEGER array, dimension (N)
*
*  INFO    (output) INTEGER
*          = 0:  successful exit
*          < 0:  if INFO = -i, the i-th argument had an illegal value
*          > 0:  if INFO = i, and i is
*                <= N:  the leading minor of order i of A is
*                       not positive definite, so the factorization
*                       could not be completed, and the solution has not
*                       been computed. RCOND = 0 is returned.
*                = N+1: U is nonsingular, but RCOND is less than machine
*                       precision, meaning that the matrix is singular
*                       to working precision.  Nevertheless, the
*                       solution and error bounds are computed because
*                       there are a number of situations where the
*                       computed solution can be more accurate than the
*                       value of RCOND would suggest.
*
*  Further Details
*  ===============
*
*  The band storage scheme is illustrated by the following example, when
*  N = 6, KD = 2, and UPLO = 'U':
*
*  Two-dimensional storage of the symmetric matrix A:
*
*     a11  a12  a13
*          a22  a23  a24
*               a33  a34  a35
*                    a44  a45  a46
*                         a55  a56
*     (aij=conjg(aji))         a66
*
*  Band storage of the upper triangle of A:
*
*      *    *   a13  a24  a35  a46
*      *   a12  a23  a34  a45  a56
*     a11  a22  a33  a44  a55  a66
*
*  Similarly, if UPLO = 'L' the format of A is as follows:
*
*     a11  a22  a33  a44  a55  a66
*     a21  a32  a43  a54  a65   *
*     a31  a42  a53  a64   *    *
*
*  Array elements marked * are not used by the routine.
*
*  =====================================================================
*
*     .. Parameters ..
      REAL               ZERO, ONE
      PARAMETER          ( ZERO = 0.0E+0, ONE = 1.0E+0 )
*     ..
*     .. Local Scalars ..
      LOGICAL            EQUIL, NOFACT, RCEQU, UPPER
      INTEGER            I, INFEQU, J, J1, J2
      REAL               AMAX, ANORM, BIGNUM, SCOND, SMAX, SMIN, SMLNUM
*     ..
*     .. External Functions ..
      LOGICAL            LSAME
      REAL               SLAMCH, SLANSB
      EXTERNAL           LSAME, SLAMCH, SLANSB
*     ..
*     .. External Subroutines ..
      EXTERNAL           SCOPY, SLACPY, SLAQSB, SPBCON, SPBEQU, SPBRFS,
     $                   SPBTRF, SPBTRS, XERBLA
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          MAXMIN
*     ..
*     .. Executable Statements ..
*
      INFO = 0
      NOFACT = LSAME( FACT, 'N' )
      EQUIL = LSAME( FACT, 'E' )
      UPPER = LSAME( UPLO, 'U' )
      IF( NOFACT .OR. EQUIL ) THEN
         EQUED = 'N'
         RCEQU = .FALSE.
      ELSE
         RCEQU = LSAME( EQUED, 'Y' )
         SMLNUM = SLAMCH( 'Safe minimum' )
         BIGNUM = ONE / SMLNUM
      END IF
*
*     Test the input parameters.
*
      IF.NOT.NOFACT .AND. .NOT.EQUIL .AND. .NOT.LSAME( FACT, 'F' ) )
     $     THEN
         INFO = -1
      ELSE IF.NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
         INFO = -2
      ELSE IF( N.LT.0 ) THEN
         INFO = -3
      ELSE IF( KD.LT.0 ) THEN
         INFO = -4
      ELSE IF( NRHS.LT.0 ) THEN
         INFO = -5
      ELSE IF( LDAB.LT.KD+1 ) THEN
         INFO = -7
      ELSE IF( LDAFB.LT.KD+1 ) THEN
         INFO = -9
      ELSE IF( LSAME( FACT, 'F' ) .AND. .NOT.
     $         ( RCEQU .OR. LSAME( EQUED, 'N' ) ) ) THEN
         INFO = -10
      ELSE
         IF( RCEQU ) THEN
            SMIN = BIGNUM
            SMAX = ZERO
            DO 10 J = 1, N
               SMIN = MIN( SMIN, S( J ) )
               SMAX = MAX( SMAX, S( J ) )
   10       CONTINUE
            IF( SMIN.LE.ZERO ) THEN
               INFO = -11
            ELSE IF( N.GT.0 ) THEN
               SCOND = MAX( SMIN, SMLNUM ) / MIN( SMAX, BIGNUM )
            ELSE
               SCOND = ONE
            END IF
         END IF
         IF( INFO.EQ.0 ) THEN
            IF( LDB.LT.MAX1, N ) ) THEN
               INFO = -13
            ELSE IF( LDX.LT.MAX1, N ) ) THEN
               INFO = -15
            END IF
         END IF
      END IF
*
      IF( INFO.NE.0 ) THEN
         CALL XERBLA( 'SPBSVX'-INFO )
         RETURN
      END IF
*
      IF( EQUIL ) THEN
*
*        Compute row and column scalings to equilibrate the matrix A.
*
         CALL SPBEQU( UPLO, N, KD, AB, LDAB, S, SCOND, AMAX, INFEQU )
         IF( INFEQU.EQ.0 ) THEN
*
*           Equilibrate the matrix.
*
            CALL SLAQSB( UPLO, N, KD, AB, LDAB, S, SCOND, AMAX, EQUED )
            RCEQU = LSAME( EQUED, 'Y' )
         END IF
      END IF
*
*     Scale the right-hand side.
*
      IF( RCEQU ) THEN
         DO 30 J = 1, NRHS
            DO 20 I = 1, N
               B( I, J ) = S( I )*B( I, J )
   20       CONTINUE
   30    CONTINUE
      END IF
*
      IF( NOFACT .OR. EQUIL ) THEN
*
*        Compute the Cholesky factorization A = U**T *U or A = L*L**T.
*
         IF( UPPER ) THEN
            DO 40 J = 1, N
               J1 = MAX( J-KD, 1 )
               CALL SCOPY( J-J1+1, AB( KD+1-J+J1, J ), 1,
     $                     AFB( KD+1-J+J1, J ), 1 )
   40       CONTINUE
         ELSE
            DO 50 J = 1, N
               J2 = MIN( J+KD, N )
               CALL SCOPY( J2-J+1, AB( 1, J ), 1, AFB( 1, J ), 1 )
   50       CONTINUE
         END IF
*
         CALL SPBTRF( UPLO, N, KD, AFB, LDAFB, INFO )
*
*        Return if INFO is non-zero.
*
         IF( INFO.GT.0 )THEN
            RCOND = ZERO
            RETURN
         END IF
      END IF
*
*     Compute the norm of the matrix A.
*
      ANORM = SLANSB( '1', UPLO, N, KD, AB, LDAB, WORK )
*
*     Compute the reciprocal of the condition number of A.
*
      CALL SPBCON( UPLO, N, KD, AFB, LDAFB, ANORM, RCOND, WORK, IWORK,
     $             INFO )
*
*     Compute the solution matrix X.
*
      CALL SLACPY( 'Full', N, NRHS, B, LDB, X, LDX )
      CALL SPBTRS( UPLO, N, KD, NRHS, AFB, LDAFB, X, LDX, INFO )
*
*     Use iterative refinement to improve the computed solution and
*     compute error bounds and backward error estimates for it.
*
      CALL SPBRFS( UPLO, N, KD, NRHS, AB, LDAB, AFB, LDAFB, B, LDB, X,
     $             LDX, FERR, BERR, WORK, IWORK, INFO )
*
*     Transform the solution matrix X to a solution of the original
*     system.
*
      IF( RCEQU ) THEN
         DO 70 J = 1, NRHS
            DO 60 I = 1, N
               X( I, J ) = S( I )*X( I, J )
   60       CONTINUE
   70    CONTINUE
         DO 80 J = 1, NRHS
            FERR( J ) = FERR( J ) / SCOND
   80    CONTINUE
      END IF
*
*     Set INFO = N+1 if the matrix is singular to working precision.
*
      IF( RCOND.LT.SLAMCH( 'Epsilon' ) )
     $   INFO = N + 1
*
      RETURN
*
*     End of SPBSVX
*
      END