1       SUBROUTINE DLASD5( I, D, Z, DELTA, RHO, DSIGMA, WORK )
  2 *
  3 *  -- LAPACK auxiliary routine (version 3.2) --
  4 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  5 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  6 *     November 2006
  7 *
  8 *     .. Scalar Arguments ..
  9       INTEGER            I
 10       DOUBLE PRECISION   DSIGMA, RHO
 11 *     ..
 12 *     .. Array Arguments ..
 13       DOUBLE PRECISION   D( 2 ), DELTA( 2 ), WORK( 2 ), Z( 2 )
 14 *     ..
 15 *
 16 *  Purpose
 17 *  =======
 18 *
 19 *  This subroutine computes the square root of the I-th eigenvalue
 20 *  of a positive symmetric rank-one modification of a 2-by-2 diagonal
 21 *  matrix
 22 *
 23 *             diag( D ) * diag( D ) +  RHO *  Z * transpose(Z) .
 24 *
 25 *  The diagonal entries in the array D are assumed to satisfy
 26 *
 27 *             0 <= D(i) < D(j)  for  i < j .
 28 *
 29 *  We also assume RHO > 0 and that the Euclidean norm of the vector
 30 *  Z is one.
 31 *
 32 *  Arguments
 33 *  =========
 34 *
 35 *  I      (input) INTEGER
 36 *         The index of the eigenvalue to be computed.  I = 1 or I = 2.
 37 *
 38 *  D      (input) DOUBLE PRECISION array, dimension ( 2 )
 39 *         The original eigenvalues.  We assume 0 <= D(1) < D(2).
 40 *
 41 *  Z      (input) DOUBLE PRECISION array, dimension ( 2 )
 42 *         The components of the updating vector.
 43 *
 44 *  DELTA  (output) DOUBLE PRECISION array, dimension ( 2 )
 45 *         Contains (D(j) - sigma_I) in its  j-th component.
 46 *         The vector DELTA contains the information necessary
 47 *         to construct the eigenvectors.
 48 *
 49 *  RHO    (input) DOUBLE PRECISION
 50 *         The scalar in the symmetric updating formula.
 51 *
 52 *  DSIGMA (output) DOUBLE PRECISION
 53 *         The computed sigma_I, the I-th updated eigenvalue.
 54 *
 55 *  WORK   (workspace) DOUBLE PRECISION array, dimension ( 2 )
 56 *         WORK contains (D(j) + sigma_I) in its  j-th component.
 57 *
 58 *  Further Details
 59 *  ===============
 60 *
 61 *  Based on contributions by
 62 *     Ren-Cang Li, Computer Science Division, University of California
 63 *     at Berkeley, USA
 64 *
 65 *  =====================================================================
 66 *
 67 *     .. Parameters ..
 68       DOUBLE PRECISION   ZERO, ONE, TWO, THREE, FOUR
 69       PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0, TWO = 2.0D+0,
 70      $                   THREE = 3.0D+0, FOUR = 4.0D+0 )
 71 *     ..
 72 *     .. Local Scalars ..
 73       DOUBLE PRECISION   B, C, DEL, DELSQ, TAU, W
 74 *     ..
 75 *     .. Intrinsic Functions ..
 76       INTRINSIC          ABSSQRT
 77 *     ..
 78 *     .. Executable Statements ..
 79 *
 80       DEL = D( 2 ) - D( 1 )
 81       DELSQ = DEL*( D( 2 )+D( 1 ) )
 82       IF( I.EQ.1 ) THEN
 83          W = ONE + FOUR*RHO*( Z( 2 )*Z( 2 ) / ( D( 1 )+THREE*D( 2 ) )-
 84      $       Z( 1 )*Z( 1 ) / ( THREE*D( 1 )+D( 2 ) ) ) / DEL
 85          IF( W.GT.ZERO ) THEN
 86             B = DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
 87             C = RHO*Z( 1 )*Z( 1 )*DELSQ
 88 *
 89 *           B > ZERO, always
 90 *
 91 *           The following TAU is DSIGMA * DSIGMA - D( 1 ) * D( 1 )
 92 *
 93             TAU = TWO*/ ( B+SQRTABS( B*B-FOUR*C ) ) )
 94 *
 95 *           The following TAU is DSIGMA - D( 1 )
 96 *
 97             TAU = TAU / ( D( 1 )+SQRT( D( 1 )*D( 1 )+TAU ) )
 98             DSIGMA = D( 1 ) + TAU
 99             DELTA( 1 ) = -TAU
100             DELTA( 2 ) = DEL - TAU
101             WORK( 1 ) = TWO*D( 1 ) + TAU
102             WORK( 2 ) = ( D( 1 )+TAU ) + D( 2 )
103 *           DELTA( 1 ) = -Z( 1 ) / TAU
104 *           DELTA( 2 ) = Z( 2 ) / ( DEL-TAU )
105          ELSE
106             B = -DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
107             C = RHO*Z( 2 )*Z( 2 )*DELSQ
108 *
109 *           The following TAU is DSIGMA * DSIGMA - D( 2 ) * D( 2 )
110 *
111             IF( B.GT.ZERO ) THEN
112                TAU = -TWO*/ ( B+SQRT( B*B+FOUR*C ) )
113             ELSE
114                TAU = ( B-SQRT( B*B+FOUR*C ) ) / TWO
115             END IF
116 *
117 *           The following TAU is DSIGMA - D( 2 )
118 *
119             TAU = TAU / ( D( 2 )+SQRTABS( D( 2 )*D( 2 )+TAU ) ) )
120             DSIGMA = D( 2 ) + TAU
121             DELTA( 1 ) = -( DEL+TAU )
122             DELTA( 2 ) = -TAU
123             WORK( 1 ) = D( 1 ) + TAU + D( 2 )
124             WORK( 2 ) = TWO*D( 2 ) + TAU
125 *           DELTA( 1 ) = -Z( 1 ) / ( DEL+TAU )
126 *           DELTA( 2 ) = -Z( 2 ) / TAU
127          END IF
128 *        TEMP = SQRT( DELTA( 1 )*DELTA( 1 )+DELTA( 2 )*DELTA( 2 ) )
129 *        DELTA( 1 ) = DELTA( 1 ) / TEMP
130 *        DELTA( 2 ) = DELTA( 2 ) / TEMP
131       ELSE
132 *
133 *        Now I=2
134 *
135          B = -DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
136          C = RHO*Z( 2 )*Z( 2 )*DELSQ
137 *
138 *        The following TAU is DSIGMA * DSIGMA - D( 2 ) * D( 2 )
139 *
140          IF( B.GT.ZERO ) THEN
141             TAU = ( B+SQRT( B*B+FOUR*C ) ) / TWO
142          ELSE
143             TAU = TWO*/ ( -B+SQRT( B*B+FOUR*C ) )
144          END IF
145 *
146 *        The following TAU is DSIGMA - D( 2 )
147 *
148          TAU = TAU / ( D( 2 )+SQRT( D( 2 )*D( 2 )+TAU ) )
149          DSIGMA = D( 2 ) + TAU
150          DELTA( 1 ) = -( DEL+TAU )
151          DELTA( 2 ) = -TAU
152          WORK( 1 ) = D( 1 ) + TAU + D( 2 )
153          WORK( 2 ) = TWO*D( 2 ) + TAU
154 *        DELTA( 1 ) = -Z( 1 ) / ( DEL+TAU )
155 *        DELTA( 2 ) = -Z( 2 ) / TAU
156 *        TEMP = SQRT( DELTA( 1 )*DELTA( 1 )+DELTA( 2 )*DELTA( 2 ) )
157 *        DELTA( 1 ) = DELTA( 1 ) / TEMP
158 *        DELTA( 2 ) = DELTA( 2 ) / TEMP
159       END IF
160       RETURN
161 *
162 *     End of DLASD5
163 *
164       END