| 1 |
subroutine balanc(nm,n,a,low,igh,scale) |
| 2 |
c |
| 3 |
integer i,j,k,l,m,n,jj,nm,igh,low,iexc |
| 4 |
double precision a(nm,n),scale(n) |
| 5 |
double precision c,f,g,r,s,b2,radix |
| 6 |
logical noconv |
| 7 |
c |
| 8 |
c this subroutine is a translation of the algol procedure balance, |
| 9 |
c num. math. 13, 293-304(1969) by parlett and reinsch. |
| 10 |
c handbook for auto. comp., vol.ii-linear algebra, 315-326(1971). |
| 11 |
c |
| 12 |
c this subroutine balances a real matrix and isolates |
| 13 |
c eigenvalues whenever possible. |
| 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 matrix. |
| 22 |
c |
| 23 |
c a contains the input matrix to be balanced. |
| 24 |
c |
| 25 |
c on output |
| 26 |
c |
| 27 |
c a contains the balanced matrix. |
| 28 |
c |
| 29 |
c low and igh are two integers such that a(i,j) |
| 30 |
c is equal to zero if |
| 31 |
c (1) i is greater than j and |
| 32 |
c (2) j=1,...,low-1 or i=igh+1,...,n. |
| 33 |
c |
| 34 |
c scale contains information determining the |
| 35 |
c permutations and scaling factors used. |
| 36 |
c |
| 37 |
c suppose that the principal submatrix in rows low through igh |
| 38 |
c has been balanced, that p(j) denotes the index interchanged |
| 39 |
c with j during the permutation step, and that the elements |
| 40 |
c of the diagonal matrix used are denoted by d(i,j). then |
| 41 |
c scale(j) = p(j), for j = 1,...,low-1 |
| 42 |
c = d(j,j), j = low,...,igh |
| 43 |
c = p(j) j = igh+1,...,n. |
| 44 |
c the order in which the interchanges are made is n to igh+1, |
| 45 |
c then 1 to low-1. |
| 46 |
c |
| 47 |
c note that 1 is returned for igh if igh is zero formally. |
| 48 |
c |
| 49 |
c the algol procedure exc contained in balance appears in |
| 50 |
c balanc in line. (note that the algol roles of identifiers |
| 51 |
c k,l have been reversed.) |
| 52 |
c |
| 53 |
c questions and comments should be directed to burton s. garbow, |
| 54 |
c mathematics and computer science div, argonne national laboratory |
| 55 |
c |
| 56 |
c this version dated august 1983. |
| 57 |
c |
| 58 |
c ------------------------------------------------------------------ |
| 59 |
c |
| 60 |
radix = 16.0d0 |
| 61 |
c |
| 62 |
b2 = radix * radix |
| 63 |
k = 1 |
| 64 |
l = n |
| 65 |
go to 100 |
| 66 |
c .......... in-line procedure for row and |
| 67 |
c column exchange .......... |
| 68 |
20 scale(m) = j |
| 69 |
if (j .eq. m) go to 50 |
| 70 |
c |
| 71 |
do 30 i = 1, l |
| 72 |
f = a(i,j) |
| 73 |
a(i,j) = a(i,m) |
| 74 |
a(i,m) = f |
| 75 |
30 continue |
| 76 |
c |
| 77 |
do 40 i = k, n |
| 78 |
f = a(j,i) |
| 79 |
a(j,i) = a(m,i) |
| 80 |
a(m,i) = f |
| 81 |
40 continue |
| 82 |
c |
| 83 |
50 go to (80,130), iexc |
| 84 |
c .......... search for rows isolating an eigenvalue |
| 85 |
c and push them down .......... |
| 86 |
80 if (l .eq. 1) go to 280 |
| 87 |
l = l - 1 |
| 88 |
c .......... for j=l step -1 until 1 do -- .......... |
| 89 |
100 do 120 jj = 1, l |
| 90 |
j = l + 1 - jj |
| 91 |
c |
| 92 |
do 110 i = 1, l |
| 93 |
if (i .eq. j) go to 110 |
| 94 |
if (a(j,i) .ne. 0.0d0) go to 120 |
| 95 |
110 continue |
| 96 |
c |
| 97 |
m = l |
| 98 |
iexc = 1 |
| 99 |
go to 20 |
| 100 |
120 continue |
| 101 |
c |
| 102 |
go to 140 |
| 103 |
c .......... search for columns isolating an eigenvalue |
| 104 |
c and push them left .......... |
| 105 |
130 k = k + 1 |
| 106 |
c |
| 107 |
140 do 170 j = k, l |
| 108 |
c |
| 109 |
do 150 i = k, l |
| 110 |
if (i .eq. j) go to 150 |
| 111 |
if (a(i,j) .ne. 0.0d0) go to 170 |
| 112 |
150 continue |
| 113 |
c |
| 114 |
m = k |
| 115 |
iexc = 2 |
| 116 |
go to 20 |
| 117 |
170 continue |
| 118 |
c .......... now balance the submatrix in rows k to l .......... |
| 119 |
do 180 i = k, l |
| 120 |
180 scale(i) = 1.0d0 |
| 121 |
c .......... iterative loop for norm reduction .......... |
| 122 |
190 noconv = .false. |
| 123 |
c |
| 124 |
do 270 i = k, l |
| 125 |
c = 0.0d0 |
| 126 |
r = 0.0d0 |
| 127 |
c |
| 128 |
do 200 j = k, l |
| 129 |
if (j .eq. i) go to 200 |
| 130 |
c = c + dabs(a(j,i)) |
| 131 |
r = r + dabs(a(i,j)) |
| 132 |
200 continue |
| 133 |
c .......... guard against zero c or r due to underflow .......... |
| 134 |
if (c .eq. 0.0d0 .or. r .eq. 0.0d0) go to 270 |
| 135 |
g = r / radix |
| 136 |
f = 1.0d0 |
| 137 |
s = c + r |
| 138 |
210 if (c .ge. g) go to 220 |
| 139 |
f = f * radix |
| 140 |
c = c * b2 |
| 141 |
go to 210 |
| 142 |
220 g = r * radix |
| 143 |
230 if (c .lt. g) go to 240 |
| 144 |
f = f / radix |
| 145 |
c = c / b2 |
| 146 |
go to 230 |
| 147 |
c .......... now balance .......... |
| 148 |
240 if ((c + r) / f .ge. 0.95d0 * s) go to 270 |
| 149 |
g = 1.0d0 / f |
| 150 |
scale(i) = scale(i) * f |
| 151 |
noconv = .true. |
| 152 |
c |
| 153 |
do 250 j = k, n |
| 154 |
250 a(i,j) = a(i,j) * g |
| 155 |
c |
| 156 |
do 260 j = 1, l |
| 157 |
260 a(j,i) = a(j,i) * f |
| 158 |
c |
| 159 |
270 continue |
| 160 |
c |
| 161 |
if (noconv) go to 190 |
| 162 |
c |
| 163 |
280 low = k |
| 164 |
igh = l |
| 165 |
return |
| 166 |
end |