| 1 |
subroutine reduc(nm,n,a,b,dl,ierr) |
| 2 |
c |
| 3 |
integer i,j,k,n,i1,j1,nm,nn,ierr |
| 4 |
double precision a(nm,n),b(nm,n),dl(n) |
| 5 |
double precision x,y |
| 6 |
c |
| 7 |
c this subroutine is a translation of the algol procedure reduc1, |
| 8 |
c num. math. 11, 99-110(1968) by martin and wilkinson. |
| 9 |
c handbook for auto. comp., vol.ii-linear algebra, 303-314(1971). |
| 10 |
c |
| 11 |
c this subroutine reduces the generalized symmetric eigenproblem |
| 12 |
c ax=(lambda)bx, where b is positive definite, to the standard |
| 13 |
c symmetric eigenproblem using the cholesky factorization of b. |
| 14 |
c |
| 15 |
c on input |
| 16 |
c |
| 17 |
c nm must be set to the row dimension of two-dimensional |
| 18 |
c array parameters as declared in the calling program |
| 19 |
c dimension statement. |
| 20 |
c |
| 21 |
c n is the order of the matrices a and b. if the cholesky |
| 22 |
c factor l of b is already available, n should be prefixed |
| 23 |
c with a minus sign. |
| 24 |
c |
| 25 |
c a and b contain the real symmetric input matrices. only the |
| 26 |
c full upper triangles of the matrices need be supplied. if |
| 27 |
c n is negative, the strict lower triangle of b contains, |
| 28 |
c instead, the strict lower triangle of its cholesky factor l. |
| 29 |
c |
| 30 |
c dl contains, if n is negative, the diagonal elements of l. |
| 31 |
c |
| 32 |
c on output |
| 33 |
c |
| 34 |
c a contains in its full lower triangle the full lower triangle |
| 35 |
c of the symmetric matrix derived from the reduction to the |
| 36 |
c standard form. the strict upper triangle of a is unaltered. |
| 37 |
c |
| 38 |
c b contains in its strict lower triangle the strict lower |
| 39 |
c triangle of its cholesky factor l. the full upper |
| 40 |
c triangle of b is unaltered. |
| 41 |
c |
| 42 |
c dl contains the diagonal elements of l. |
| 43 |
c |
| 44 |
c ierr is set to |
| 45 |
c zero for normal return, |
| 46 |
c 7*n+1 if b is not positive definite. |
| 47 |
c |
| 48 |
c questions and comments should be directed to burton s. garbow, |
| 49 |
c mathematics and computer science div, argonne national laboratory |
| 50 |
c |
| 51 |
c this version dated august 1983. |
| 52 |
c |
| 53 |
c ------------------------------------------------------------------ |
| 54 |
c |
| 55 |
ierr = 0 |
| 56 |
nn = iabs(n) |
| 57 |
if (n .lt. 0) go to 100 |
| 58 |
c .......... form l in the arrays b and dl .......... |
| 59 |
do 80 i = 1, n |
| 60 |
i1 = i - 1 |
| 61 |
c |
| 62 |
do 80 j = i, n |
| 63 |
x = b(i,j) |
| 64 |
if (i .eq. 1) go to 40 |
| 65 |
c |
| 66 |
do 20 k = 1, i1 |
| 67 |
20 x = x - b(i,k) * b(j,k) |
| 68 |
c |
| 69 |
40 if (j .ne. i) go to 60 |
| 70 |
if (x .le. 0.0d0) go to 1000 |
| 71 |
y = dsqrt(x) |
| 72 |
dl(i) = y |
| 73 |
go to 80 |
| 74 |
60 b(j,i) = x / y |
| 75 |
80 continue |
| 76 |
c .......... form the transpose of the upper triangle of inv(l)*a |
| 77 |
c in the lower triangle of the array a .......... |
| 78 |
100 do 200 i = 1, nn |
| 79 |
i1 = i - 1 |
| 80 |
y = dl(i) |
| 81 |
c |
| 82 |
do 200 j = i, nn |
| 83 |
x = a(i,j) |
| 84 |
if (i .eq. 1) go to 180 |
| 85 |
c |
| 86 |
do 160 k = 1, i1 |
| 87 |
160 x = x - b(i,k) * a(j,k) |
| 88 |
c |
| 89 |
180 a(j,i) = x / y |
| 90 |
200 continue |
| 91 |
c .......... pre-multiply by inv(l) and overwrite .......... |
| 92 |
do 300 j = 1, nn |
| 93 |
j1 = j - 1 |
| 94 |
c |
| 95 |
do 300 i = j, nn |
| 96 |
x = a(i,j) |
| 97 |
if (i .eq. j) go to 240 |
| 98 |
i1 = i - 1 |
| 99 |
c |
| 100 |
do 220 k = j, i1 |
| 101 |
220 x = x - a(k,j) * b(i,k) |
| 102 |
c |
| 103 |
240 if (j .eq. 1) go to 280 |
| 104 |
c |
| 105 |
do 260 k = 1, j1 |
| 106 |
260 x = x - a(j,k) * b(i,k) |
| 107 |
c |
| 108 |
280 a(i,j) = x / dl(i) |
| 109 |
300 continue |
| 110 |
c |
| 111 |
go to 1001 |
| 112 |
c .......... set error -- b is not positive definite .......... |
| 113 |
1000 ierr = 7 * n + 1 |
| 114 |
1001 return |
| 115 |
end |