| 1 |
SUBROUTINE DLAEV2( A, B, C, RT1, RT2, CS1, SN1 ) |
| 2 |
* |
| 3 |
* -- LAPACK auxiliary routine (version 2.0) -- |
| 4 |
* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., |
| 5 |
* Courant Institute, Argonne National Lab, and Rice University |
| 6 |
* October 31, 1992 |
| 7 |
* |
| 8 |
* .. Scalar Arguments .. |
| 9 |
DOUBLE PRECISION A, B, C, CS1, RT1, RT2, SN1 |
| 10 |
* .. |
| 11 |
* |
| 12 |
* Purpose |
| 13 |
* ======= |
| 14 |
* |
| 15 |
* DLAEV2 computes the eigendecomposition of a 2-by-2 symmetric matrix |
| 16 |
* [ A B ] |
| 17 |
* [ B C ]. |
| 18 |
* On return, RT1 is the eigenvalue of larger absolute value, RT2 is the |
| 19 |
* eigenvalue of smaller absolute value, and (CS1,SN1) is the unit right |
| 20 |
* eigenvector for RT1, giving the decomposition |
| 21 |
* |
| 22 |
* [ CS1 SN1 ] [ A B ] [ CS1 -SN1 ] = [ RT1 0 ] |
| 23 |
* [-SN1 CS1 ] [ B C ] [ SN1 CS1 ] [ 0 RT2 ]. |
| 24 |
* |
| 25 |
* Arguments |
| 26 |
* ========= |
| 27 |
* |
| 28 |
* A (input) DOUBLE PRECISION |
| 29 |
* The (1,1) element of the 2-by-2 matrix. |
| 30 |
* |
| 31 |
* B (input) DOUBLE PRECISION |
| 32 |
* The (1,2) element and the conjugate of the (2,1) element of |
| 33 |
* the 2-by-2 matrix. |
| 34 |
* |
| 35 |
* C (input) DOUBLE PRECISION |
| 36 |
* The (2,2) element of the 2-by-2 matrix. |
| 37 |
* |
| 38 |
* RT1 (output) DOUBLE PRECISION |
| 39 |
* The eigenvalue of larger absolute value. |
| 40 |
* |
| 41 |
* RT2 (output) DOUBLE PRECISION |
| 42 |
* The eigenvalue of smaller absolute value. |
| 43 |
* |
| 44 |
* CS1 (output) DOUBLE PRECISION |
| 45 |
* SN1 (output) DOUBLE PRECISION |
| 46 |
* The vector (CS1, SN1) is a unit right eigenvector for RT1. |
| 47 |
* |
| 48 |
* Further Details |
| 49 |
* =============== |
| 50 |
* |
| 51 |
* RT1 is accurate to a few ulps barring over/underflow. |
| 52 |
* |
| 53 |
* RT2 may be inaccurate if there is massive cancellation in the |
| 54 |
* determinant A*C-B*B; higher precision or correctly rounded or |
| 55 |
* correctly truncated arithmetic would be needed to compute RT2 |
| 56 |
* accurately in all cases. |
| 57 |
* |
| 58 |
* CS1 and SN1 are accurate to a few ulps barring over/underflow. |
| 59 |
* |
| 60 |
* Overflow is possible only if RT1 is within a factor of 5 of overflow. |
| 61 |
* Underflow is harmless if the input data is 0 or exceeds |
| 62 |
* underflow_threshold / macheps. |
| 63 |
* |
| 64 |
* ===================================================================== |
| 65 |
* |
| 66 |
* .. Parameters .. |
| 67 |
DOUBLE PRECISION ONE |
| 68 |
PARAMETER ( ONE = 1.0D0 ) |
| 69 |
DOUBLE PRECISION TWO |
| 70 |
PARAMETER ( TWO = 2.0D0 ) |
| 71 |
DOUBLE PRECISION ZERO |
| 72 |
PARAMETER ( ZERO = 0.0D0 ) |
| 73 |
DOUBLE PRECISION HALF |
| 74 |
PARAMETER ( HALF = 0.5D0 ) |
| 75 |
* .. |
| 76 |
* .. Local Scalars .. |
| 77 |
INTEGER SGN1, SGN2 |
| 78 |
DOUBLE PRECISION AB, ACMN, ACMX, ACS, ADF, CS, CT, DF, RT, SM, |
| 79 |
$ TB, TN |
| 80 |
* .. |
| 81 |
* .. Intrinsic Functions .. |
| 82 |
INTRINSIC ABS, SQRT |
| 83 |
* .. |
| 84 |
* .. Executable Statements .. |
| 85 |
* |
| 86 |
* Compute the eigenvalues |
| 87 |
* |
| 88 |
SM = A + C |
| 89 |
DF = A - C |
| 90 |
ADF = ABS( DF ) |
| 91 |
TB = B + B |
| 92 |
AB = ABS( TB ) |
| 93 |
IF( ABS( A ).GT.ABS( C ) ) THEN |
| 94 |
ACMX = A |
| 95 |
ACMN = C |
| 96 |
ELSE |
| 97 |
ACMX = C |
| 98 |
ACMN = A |
| 99 |
END IF |
| 100 |
IF( ADF.GT.AB ) THEN |
| 101 |
RT = ADF*SQRT( ONE+( AB / ADF )**2 ) |
| 102 |
ELSE IF( ADF.LT.AB ) THEN |
| 103 |
RT = AB*SQRT( ONE+( ADF / AB )**2 ) |
| 104 |
ELSE |
| 105 |
* |
| 106 |
* Includes case AB=ADF=0 |
| 107 |
* |
| 108 |
RT = AB*SQRT( TWO ) |
| 109 |
END IF |
| 110 |
IF( SM.LT.ZERO ) THEN |
| 111 |
RT1 = HALF*( SM-RT ) |
| 112 |
SGN1 = -1 |
| 113 |
* |
| 114 |
* Order of execution important. |
| 115 |
* To get fully accurate smaller eigenvalue, |
| 116 |
* next line needs to be executed in higher precision. |
| 117 |
* |
| 118 |
RT2 = ( ACMX / RT1 )*ACMN - ( B / RT1 )*B |
| 119 |
ELSE IF( SM.GT.ZERO ) THEN |
| 120 |
RT1 = HALF*( SM+RT ) |
| 121 |
SGN1 = 1 |
| 122 |
* |
| 123 |
* Order of execution important. |
| 124 |
* To get fully accurate smaller eigenvalue, |
| 125 |
* next line needs to be executed in higher precision. |
| 126 |
* |
| 127 |
RT2 = ( ACMX / RT1 )*ACMN - ( B / RT1 )*B |
| 128 |
ELSE |
| 129 |
* |
| 130 |
* Includes case RT1 = RT2 = 0 |
| 131 |
* |
| 132 |
RT1 = HALF*RT |
| 133 |
RT2 = -HALF*RT |
| 134 |
SGN1 = 1 |
| 135 |
END IF |
| 136 |
* |
| 137 |
* Compute the eigenvector |
| 138 |
* |
| 139 |
IF( DF.GE.ZERO ) THEN |
| 140 |
CS = DF + RT |
| 141 |
SGN2 = 1 |
| 142 |
ELSE |
| 143 |
CS = DF - RT |
| 144 |
SGN2 = -1 |
| 145 |
END IF |
| 146 |
ACS = ABS( CS ) |
| 147 |
IF( ACS.GT.AB ) THEN |
| 148 |
CT = -TB / CS |
| 149 |
SN1 = ONE / SQRT( ONE+CT*CT ) |
| 150 |
CS1 = CT*SN1 |
| 151 |
ELSE |
| 152 |
IF( AB.EQ.ZERO ) THEN |
| 153 |
CS1 = ONE |
| 154 |
SN1 = ZERO |
| 155 |
ELSE |
| 156 |
TN = -CS / TB |
| 157 |
CS1 = ONE / SQRT( ONE+TN*TN ) |
| 158 |
SN1 = TN*CS1 |
| 159 |
END IF |
| 160 |
END IF |
| 161 |
IF( SGN1.EQ.SGN2 ) THEN |
| 162 |
TN = CS1 |
| 163 |
CS1 = -SN1 |
| 164 |
SN1 = TN |
| 165 |
END IF |
| 166 |
RETURN |
| 167 |
* |
| 168 |
* End of DLAEV2 |
| 169 |
* |
| 170 |
END |