| 1 |
subroutine hqr(nm,n,low,igh,h,wr,wi,ierr) |
| 2 |
C RESTORED CORRECT INDICES OF LOOPS (200,210,230,240). (9/29/89 BSG) |
| 3 |
c |
| 4 |
integer i,j,k,l,m,n,en,ll,mm,na,nm,igh,itn,its,low,mp2,enm2,ierr |
| 5 |
double precision h(nm,n),wr(n),wi(n) |
| 6 |
double precision p,q,r,s,t,w,x,y,zz,norm,tst1,tst2 |
| 7 |
logical notlas |
| 8 |
c |
| 9 |
c this subroutine is a translation of the algol procedure hqr, |
| 10 |
c num. math. 14, 219-231(1970) by martin, peters, and wilkinson. |
| 11 |
c handbook for auto. comp., vol.ii-linear algebra, 359-371(1971). |
| 12 |
c |
| 13 |
c this subroutine finds the eigenvalues of a real |
| 14 |
c upper hessenberg matrix by the qr method. |
| 15 |
c |
| 16 |
c on input |
| 17 |
c |
| 18 |
c nm must be set to the row dimension of two-dimensional |
| 19 |
c array parameters as declared in the calling program |
| 20 |
c dimension statement. |
| 21 |
c |
| 22 |
c n is the order of the matrix. |
| 23 |
c |
| 24 |
c low and igh are integers determined by the balancing |
| 25 |
c subroutine balanc. if balanc has not been used, |
| 26 |
c set low=1, igh=n. |
| 27 |
c |
| 28 |
c h contains the upper hessenberg matrix. information about |
| 29 |
c the transformations used in the reduction to hessenberg |
| 30 |
c form by elmhes or orthes, if performed, is stored |
| 31 |
c in the remaining triangle under the hessenberg matrix. |
| 32 |
c |
| 33 |
c on output |
| 34 |
c |
| 35 |
c h has been destroyed. therefore, it must be saved |
| 36 |
c before calling hqr if subsequent calculation and |
| 37 |
c back transformation of eigenvectors is to be performed. |
| 38 |
c |
| 39 |
c wr and wi contain the real and imaginary parts, |
| 40 |
c respectively, of the eigenvalues. the eigenvalues |
| 41 |
c are unordered except that complex conjugate pairs |
| 42 |
c of values appear consecutively with the eigenvalue |
| 43 |
c having the positive imaginary part first. if an |
| 44 |
c error exit is made, the eigenvalues should be correct |
| 45 |
c for indices ierr+1,...,n. |
| 46 |
c |
| 47 |
c ierr is set to |
| 48 |
c zero for normal return, |
| 49 |
c j if the limit of 30*n iterations is exhausted |
| 50 |
c while the j-th eigenvalue is being sought. |
| 51 |
c |
| 52 |
c questions and comments should be directed to burton s. garbow, |
| 53 |
c mathematics and computer science div, argonne national laboratory |
| 54 |
c |
| 55 |
c this version dated september 1989. |
| 56 |
c |
| 57 |
c ------------------------------------------------------------------ |
| 58 |
c |
| 59 |
ierr = 0 |
| 60 |
norm = 0.0d0 |
| 61 |
k = 1 |
| 62 |
c .......... store roots isolated by balanc |
| 63 |
c and compute matrix norm .......... |
| 64 |
do 50 i = 1, n |
| 65 |
c |
| 66 |
do 40 j = k, n |
| 67 |
40 norm = norm + dabs(h(i,j)) |
| 68 |
c |
| 69 |
k = i |
| 70 |
if (i .ge. low .and. i .le. igh) go to 50 |
| 71 |
wr(i) = h(i,i) |
| 72 |
wi(i) = 0.0d0 |
| 73 |
50 continue |
| 74 |
c |
| 75 |
en = igh |
| 76 |
t = 0.0d0 |
| 77 |
itn = 30*n |
| 78 |
c .......... search for next eigenvalues .......... |
| 79 |
60 if (en .lt. low) go to 1001 |
| 80 |
its = 0 |
| 81 |
na = en - 1 |
| 82 |
enm2 = na - 1 |
| 83 |
c .......... look for single small sub-diagonal element |
| 84 |
c for l=en step -1 until low do -- .......... |
| 85 |
70 do 80 ll = low, en |
| 86 |
l = en + low - ll |
| 87 |
if (l .eq. low) go to 100 |
| 88 |
s = dabs(h(l-1,l-1)) + dabs(h(l,l)) |
| 89 |
if (s .eq. 0.0d0) s = norm |
| 90 |
tst1 = s |
| 91 |
tst2 = tst1 + dabs(h(l,l-1)) |
| 92 |
if (tst2 .eq. tst1) go to 100 |
| 93 |
80 continue |
| 94 |
c .......... form shift .......... |
| 95 |
100 x = h(en,en) |
| 96 |
if (l .eq. en) go to 270 |
| 97 |
y = h(na,na) |
| 98 |
w = h(en,na) * h(na,en) |
| 99 |
if (l .eq. na) go to 280 |
| 100 |
if (itn .eq. 0) go to 1000 |
| 101 |
if (its .ne. 10 .and. its .ne. 20) go to 130 |
| 102 |
c .......... form exceptional shift .......... |
| 103 |
t = t + x |
| 104 |
c |
| 105 |
do 120 i = low, en |
| 106 |
120 h(i,i) = h(i,i) - x |
| 107 |
c |
| 108 |
s = dabs(h(en,na)) + dabs(h(na,enm2)) |
| 109 |
x = 0.75d0 * s |
| 110 |
y = x |
| 111 |
w = -0.4375d0 * s * s |
| 112 |
130 its = its + 1 |
| 113 |
itn = itn - 1 |
| 114 |
c .......... look for two consecutive small |
| 115 |
c sub-diagonal elements. |
| 116 |
c for m=en-2 step -1 until l do -- .......... |
| 117 |
do 140 mm = l, enm2 |
| 118 |
m = enm2 + l - mm |
| 119 |
zz = h(m,m) |
| 120 |
r = x - zz |
| 121 |
s = y - zz |
| 122 |
p = (r * s - w) / h(m+1,m) + h(m,m+1) |
| 123 |
q = h(m+1,m+1) - zz - r - s |
| 124 |
r = h(m+2,m+1) |
| 125 |
s = dabs(p) + dabs(q) + dabs(r) |
| 126 |
p = p / s |
| 127 |
q = q / s |
| 128 |
r = r / s |
| 129 |
if (m .eq. l) go to 150 |
| 130 |
tst1 = dabs(p)*(dabs(h(m-1,m-1)) + dabs(zz) + dabs(h(m+1,m+1))) |
| 131 |
tst2 = tst1 + dabs(h(m,m-1))*(dabs(q) + dabs(r)) |
| 132 |
if (tst2 .eq. tst1) go to 150 |
| 133 |
140 continue |
| 134 |
c |
| 135 |
150 mp2 = m + 2 |
| 136 |
c |
| 137 |
do 160 i = mp2, en |
| 138 |
h(i,i-2) = 0.0d0 |
| 139 |
if (i .eq. mp2) go to 160 |
| 140 |
h(i,i-3) = 0.0d0 |
| 141 |
160 continue |
| 142 |
c .......... double qr step involving rows l to en and |
| 143 |
c columns m to en .......... |
| 144 |
do 260 k = m, na |
| 145 |
notlas = k .ne. na |
| 146 |
if (k .eq. m) go to 170 |
| 147 |
p = h(k,k-1) |
| 148 |
q = h(k+1,k-1) |
| 149 |
r = 0.0d0 |
| 150 |
if (notlas) r = h(k+2,k-1) |
| 151 |
x = dabs(p) + dabs(q) + dabs(r) |
| 152 |
if (x .eq. 0.0d0) go to 260 |
| 153 |
p = p / x |
| 154 |
q = q / x |
| 155 |
r = r / x |
| 156 |
170 s = dsign(dsqrt(p*p+q*q+r*r),p) |
| 157 |
if (k .eq. m) go to 180 |
| 158 |
h(k,k-1) = -s * x |
| 159 |
go to 190 |
| 160 |
180 if (l .ne. m) h(k,k-1) = -h(k,k-1) |
| 161 |
190 p = p + s |
| 162 |
x = p / s |
| 163 |
y = q / s |
| 164 |
zz = r / s |
| 165 |
q = q / p |
| 166 |
r = r / p |
| 167 |
if (notlas) go to 225 |
| 168 |
c .......... row modification .......... |
| 169 |
do 200 j = k, EN |
| 170 |
p = h(k,j) + q * h(k+1,j) |
| 171 |
h(k,j) = h(k,j) - p * x |
| 172 |
h(k+1,j) = h(k+1,j) - p * y |
| 173 |
200 continue |
| 174 |
c |
| 175 |
j = min0(en,k+3) |
| 176 |
c .......... column modification .......... |
| 177 |
do 210 i = L, j |
| 178 |
p = x * h(i,k) + y * h(i,k+1) |
| 179 |
h(i,k) = h(i,k) - p |
| 180 |
h(i,k+1) = h(i,k+1) - p * q |
| 181 |
210 continue |
| 182 |
go to 255 |
| 183 |
225 continue |
| 184 |
c .......... row modification .......... |
| 185 |
do 230 j = k, EN |
| 186 |
p = h(k,j) + q * h(k+1,j) + r * h(k+2,j) |
| 187 |
h(k,j) = h(k,j) - p * x |
| 188 |
h(k+1,j) = h(k+1,j) - p * y |
| 189 |
h(k+2,j) = h(k+2,j) - p * zz |
| 190 |
230 continue |
| 191 |
c |
| 192 |
j = min0(en,k+3) |
| 193 |
c .......... column modification .......... |
| 194 |
do 240 i = L, j |
| 195 |
p = x * h(i,k) + y * h(i,k+1) + zz * h(i,k+2) |
| 196 |
h(i,k) = h(i,k) - p |
| 197 |
h(i,k+1) = h(i,k+1) - p * q |
| 198 |
h(i,k+2) = h(i,k+2) - p * r |
| 199 |
240 continue |
| 200 |
255 continue |
| 201 |
c |
| 202 |
260 continue |
| 203 |
c |
| 204 |
go to 70 |
| 205 |
c .......... one root found .......... |
| 206 |
270 wr(en) = x + t |
| 207 |
wi(en) = 0.0d0 |
| 208 |
en = na |
| 209 |
go to 60 |
| 210 |
c .......... two roots found .......... |
| 211 |
280 p = (y - x) / 2.0d0 |
| 212 |
q = p * p + w |
| 213 |
zz = dsqrt(dabs(q)) |
| 214 |
x = x + t |
| 215 |
if (q .lt. 0.0d0) go to 320 |
| 216 |
c .......... real pair .......... |
| 217 |
zz = p + dsign(zz,p) |
| 218 |
wr(na) = x + zz |
| 219 |
wr(en) = wr(na) |
| 220 |
if (zz .ne. 0.0d0) wr(en) = x - w / zz |
| 221 |
wi(na) = 0.0d0 |
| 222 |
wi(en) = 0.0d0 |
| 223 |
go to 330 |
| 224 |
c .......... complex pair .......... |
| 225 |
320 wr(na) = x + p |
| 226 |
wr(en) = x + p |
| 227 |
wi(na) = zz |
| 228 |
wi(en) = -zz |
| 229 |
330 en = enm2 |
| 230 |
go to 60 |
| 231 |
c .......... set error -- all eigenvalues have not |
| 232 |
c converged after 30*n iterations .......... |
| 233 |
1000 ierr = en |
| 234 |
1001 return |
| 235 |
end |