LAPACK: SRC/dlasd5.f Source File

Go to the documentation of this file.
00001       SUBROUTINE DLASD5( I, D, Z, DELTA, RHO, DSIGMA, WORK )
00002 *
00003 *  -- LAPACK auxiliary routine (version 3.2) --
00004 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
00005 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
00006 *     November 2006
00007 *
00008 *     .. Scalar Arguments ..
00009       INTEGER            I
00010       DOUBLE PRECISION   DSIGMA, RHO
00011 *     ..
00012 *     .. Array Arguments ..
00013       DOUBLE PRECISION   D( 2 ), DELTA( 2 ), WORK( 2 ), Z( 2 )
00014 *     ..
00015 *
00016 *  Purpose
00017 *  =======
00018 *
00019 *  This subroutine computes the square root of the I-th eigenvalue
00020 *  of a positive symmetric rank-one modification of a 2-by-2 diagonal
00021 *  matrix
00022 *
00023 *             diag( D ) * diag( D ) +  RHO *  Z * transpose(Z) .
00024 *
00025 *  The diagonal entries in the array D are assumed to satisfy
00026 *
00027 *             0 <= D(i) < D(j)  for  i < j .
00028 *
00029 *  We also assume RHO > 0 and that the Euclidean norm of the vector
00030 *  Z is one.
00031 *
00032 *  Arguments
00033 *  =========
00034 *
00035 *  I      (input) INTEGER
00036 *         The index of the eigenvalue to be computed.  I = 1 or I = 2.
00037 *
00038 *  D      (input) DOUBLE PRECISION array, dimension ( 2 )
00039 *         The original eigenvalues.  We assume 0 <= D(1) < D(2).
00040 *
00041 *  Z      (input) DOUBLE PRECISION array, dimension ( 2 )
00042 *         The components of the updating vector.
00043 *
00044 *  DELTA  (output) DOUBLE PRECISION array, dimension ( 2 )
00045 *         Contains (D(j) - sigma_I) in its  j-th component.
00046 *         The vector DELTA contains the information necessary
00047 *         to construct the eigenvectors.
00048 *
00049 *  RHO    (input) DOUBLE PRECISION
00050 *         The scalar in the symmetric updating formula.
00051 *
00052 *  DSIGMA (output) DOUBLE PRECISION
00053 *         The computed sigma_I, the I-th updated eigenvalue.
00054 *
00055 *  WORK   (workspace) DOUBLE PRECISION array, dimension ( 2 )
00056 *         WORK contains (D(j) + sigma_I) in its  j-th component.
00057 *
00058 *  Further Details
00059 *  ===============
00060 *
00061 *  Based on contributions by
00062 *     Ren-Cang Li, Computer Science Division, University of California
00063 *     at Berkeley, USA
00064 *
00065 *  =====================================================================
00066 *
00067 *     .. Parameters ..
00068       DOUBLE PRECISION   ZERO, ONE, TWO, THREE, FOUR
00069       PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0, TWO = 2.0D+0,
00070      $                   THREE = 3.0D+0, FOUR = 4.0D+0 )
00071 *     ..
00072 *     .. Local Scalars ..
00073       DOUBLE PRECISION   B, C, DEL, DELSQ, TAU, W
00074 *     ..
00075 *     .. Intrinsic Functions ..
00076       INTRINSIC          ABS, SQRT
00077 *     ..
00078 *     .. Executable Statements ..
00079 *
00080       DEL = D( 2 ) - D( 1 )
00081       DELSQ = DEL*( D( 2 )+D( 1 ) )
00082       IF( I.EQ.1 ) THEN
00083          W = ONE + FOUR*RHO*( Z( 2 )*Z( 2 ) / ( D( 1 )+THREE*D( 2 ) )-
00084      $       Z( 1 )*Z( 1 ) / ( THREE*D( 1 )+D( 2 ) ) ) / DEL
00085          IF( W.GT.ZERO ) THEN
00086             B = DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
00087             C = RHO*Z( 1 )*Z( 1 )*DELSQ
00088 *
00089 *           B > ZERO, always
00090 *
00091 *           The following TAU is DSIGMA * DSIGMA - D( 1 ) * D( 1 )
00092 *
00093             TAU = TWO*C / ( B+SQRT( ABS( B*B-FOUR*C ) ) )
00094 *
00095 *           The following TAU is DSIGMA - D( 1 )
00096 *
00097             TAU = TAU / ( D( 1 )+SQRT( D( 1 )*D( 1 )+TAU ) )
00098             DSIGMA = D( 1 ) + TAU
00099             DELTA( 1 ) = -TAU
00100             DELTA( 2 ) = DEL - TAU
00101             WORK( 1 ) = TWO*D( 1 ) + TAU
00102             WORK( 2 ) = ( D( 1 )+TAU ) + D( 2 )
00103 *           DELTA( 1 ) = -Z( 1 ) / TAU
00104 *           DELTA( 2 ) = Z( 2 ) / ( DEL-TAU )
00105          ELSE
00106             B = -DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
00107             C = RHO*Z( 2 )*Z( 2 )*DELSQ
00108 *
00109 *           The following TAU is DSIGMA * DSIGMA - D( 2 ) * D( 2 )
00110 *
00111             IF( B.GT.ZERO ) THEN
00112                TAU = -TWO*C / ( B+SQRT( B*B+FOUR*C ) )
00113             ELSE
00114                TAU = ( B-SQRT( B*B+FOUR*C ) ) / TWO
00115             END IF
00116 *
00117 *           The following TAU is DSIGMA - D( 2 )
00118 *
00119             TAU = TAU / ( D( 2 )+SQRT( ABS( D( 2 )*D( 2 )+TAU ) ) )
00120             DSIGMA = D( 2 ) + TAU
00121             DELTA( 1 ) = -( DEL+TAU )
00122             DELTA( 2 ) = -TAU
00123             WORK( 1 ) = D( 1 ) + TAU + D( 2 )
00124             WORK( 2 ) = TWO*D( 2 ) + TAU
00125 *           DELTA( 1 ) = -Z( 1 ) / ( DEL+TAU )
00126 *           DELTA( 2 ) = -Z( 2 ) / TAU
00127          END IF
00128 *        TEMP = SQRT( DELTA( 1 )*DELTA( 1 )+DELTA( 2 )*DELTA( 2 ) )
00129 *        DELTA( 1 ) = DELTA( 1 ) / TEMP
00130 *        DELTA( 2 ) = DELTA( 2 ) / TEMP
00131       ELSE
00132 *
00133 *        Now I=2
00134 *
00135          B = -DELSQ + RHO*( Z( 1 )*Z( 1 )+Z( 2 )*Z( 2 ) )
00136          C = RHO*Z( 2 )*Z( 2 )*DELSQ
00137 *
00138 *        The following TAU is DSIGMA * DSIGMA - D( 2 ) * D( 2 )
00139 *
00140          IF( B.GT.ZERO ) THEN
00141             TAU = ( B+SQRT( B*B+FOUR*C ) ) / TWO
00142          ELSE
00143             TAU = TWO*C / ( -B+SQRT( B*B+FOUR*C ) )
00144          END IF
00145 *
00146 *        The following TAU is DSIGMA - D( 2 )
00147 *
00148          TAU = TAU / ( D( 2 )+SQRT( D( 2 )*D( 2 )+TAU ) )
00149          DSIGMA = D( 2 ) + TAU
00150          DELTA( 1 ) = -( DEL+TAU )
00151          DELTA( 2 ) = -TAU
00152          WORK( 1 ) = D( 1 ) + TAU + D( 2 )
00153          WORK( 2 ) = TWO*D( 2 ) + TAU
00154 *        DELTA( 1 ) = -Z( 1 ) / ( DEL+TAU )
00155 *        DELTA( 2 ) = -Z( 2 ) / TAU
00156 *        TEMP = SQRT( DELTA( 1 )*DELTA( 1 )+DELTA( 2 )*DELTA( 2 ) )
00157 *        DELTA( 1 ) = DELTA( 1 ) / TEMP
00158 *        DELTA( 2 ) = DELTA( 2 ) / TEMP
00159       END IF
00160       RETURN
00161 *
00162 *     End of DLASD5
00163 *
00164       END