| 1 |
C dgbfa.f |
| 2 |
C is freely available from netlib. It is not subject to any GNU License |
| 3 |
C set by the authors of the ASCEND math programming system. |
| 4 |
C $Date: 1996/04/30 18:17:11 $ $Revision: 1.1.1.1 $ |
| 5 |
C |
| 6 |
subroutine dgbfa(abd,lda,n,ml,mu,ipvt,info) |
| 7 |
integer lda,n,ml,mu,ipvt(1),info |
| 8 |
double precision abd(lda,1) |
| 9 |
c |
| 10 |
c dgbfa factors a double precision band matrix by elimination. |
| 11 |
c |
| 12 |
c dgbfa is usually called by dgbco, but it can be called |
| 13 |
c directly with a saving in time if rcond is not needed. |
| 14 |
c |
| 15 |
c on entry |
| 16 |
c |
| 17 |
c abd double precision(lda, n) |
| 18 |
c contains the matrix in band storage. the columns |
| 19 |
c of the matrix are stored in the columns of abd and |
| 20 |
c the diagonals of the matrix are stored in rows |
| 21 |
c ml+1 through 2*ml+mu+1 of abd . |
| 22 |
c see the comments below for details. |
| 23 |
c |
| 24 |
c lda integer |
| 25 |
c the leading dimension of the array abd . |
| 26 |
c lda must be .ge. 2*ml + mu + 1 . |
| 27 |
c |
| 28 |
c n integer |
| 29 |
c the order of the original matrix. |
| 30 |
c |
| 31 |
c ml integer |
| 32 |
c number of diagonals below the main diagonal. |
| 33 |
c 0 .le. ml .lt. n . |
| 34 |
c |
| 35 |
c mu integer |
| 36 |
c number of diagonals above the main diagonal. |
| 37 |
c 0 .le. mu .lt. n . |
| 38 |
c more efficient if ml .le. mu . |
| 39 |
c on return |
| 40 |
c |
| 41 |
c abd an upper triangular matrix in band storage and |
| 42 |
c the multipliers which were used to obtain it. |
| 43 |
c the factorization can be written a = l*u where |
| 44 |
c l is a product of permutation and unit lower |
| 45 |
c triangular matrices and u is upper triangular. |
| 46 |
c |
| 47 |
c ipvt integer(n) |
| 48 |
c an integer vector of pivot indices. |
| 49 |
c |
| 50 |
c info integer |
| 51 |
c = 0 normal value. |
| 52 |
c = k if u(k,k) .eq. 0.0 . this is not an error |
| 53 |
c condition for this subroutine, but it does |
| 54 |
c indicate that dgbsl will divide by zero if |
| 55 |
c called. use rcond in dgbco for a reliable |
| 56 |
c indication of singularity. |
| 57 |
c |
| 58 |
c band storage |
| 59 |
c |
| 60 |
c if a is a band matrix, the following program segment |
| 61 |
c will set up the input. |
| 62 |
c |
| 63 |
c ml = (band width below the diagonal) |
| 64 |
c mu = (band width above the diagonal) |
| 65 |
c m = ml + mu + 1 |
| 66 |
c do 20 j = 1, n |
| 67 |
c i1 = max0(1, j-mu) |
| 68 |
c i2 = min0(n, j+ml) |
| 69 |
c do 10 i = i1, i2 |
| 70 |
c k = i - j + m |
| 71 |
c abd(k,j) = a(i,j) |
| 72 |
c 10 continue |
| 73 |
c 20 continue |
| 74 |
c |
| 75 |
c this uses rows ml+1 through 2*ml+mu+1 of abd . |
| 76 |
c in addition, the first ml rows in abd are used for |
| 77 |
c elements generated during the triangularization. |
| 78 |
c the total number of rows needed in abd is 2*ml+mu+1 . |
| 79 |
c the ml+mu by ml+mu upper left triangle and the |
| 80 |
c ml by ml lower right triangle are not referenced. |
| 81 |
c |
| 82 |
c linpack. this version dated 08/14/78 . |
| 83 |
c cleve moler, university of new mexico, argonne national lab. |
| 84 |
c |
| 85 |
c subroutines and functions |
| 86 |
c |
| 87 |
c blas daxpy,dscal,idamax |
| 88 |
c fortran max0,min0 |
| 89 |
c |
| 90 |
c internal variables |
| 91 |
c |
| 92 |
double precision t |
| 93 |
integer i,idamax,i0,j,ju,jz,j0,j1,k,kp1,l,lm,m,mm,nm1 |
| 94 |
c |
| 95 |
c |
| 96 |
m = ml + mu + 1 |
| 97 |
info = 0 |
| 98 |
c |
| 99 |
c zero initial fill-in columns |
| 100 |
c |
| 101 |
j0 = mu + 2 |
| 102 |
j1 = min0(n,m) - 1 |
| 103 |
if (j1 .lt. j0) go to 30 |
| 104 |
do 20 jz = j0, j1 |
| 105 |
i0 = m + 1 - jz |
| 106 |
do 10 i = i0, ml |
| 107 |
abd(i,jz) = 0.0d0 |
| 108 |
10 continue |
| 109 |
20 continue |
| 110 |
30 continue |
| 111 |
jz = j1 |
| 112 |
ju = 0 |
| 113 |
c |
| 114 |
c gaussian elimination with partial pivoting |
| 115 |
c |
| 116 |
nm1 = n - 1 |
| 117 |
if (nm1 .lt. 1) go to 130 |
| 118 |
do 120 k = 1, nm1 |
| 119 |
kp1 = k + 1 |
| 120 |
c |
| 121 |
c zero next fill-in column |
| 122 |
c |
| 123 |
jz = jz + 1 |
| 124 |
if (jz .gt. n) go to 50 |
| 125 |
if (ml .lt. 1) go to 50 |
| 126 |
do 40 i = 1, ml |
| 127 |
abd(i,jz) = 0.0d0 |
| 128 |
40 continue |
| 129 |
50 continue |
| 130 |
c |
| 131 |
c find l = pivot index |
| 132 |
c |
| 133 |
lm = min0(ml,n-k) |
| 134 |
l = idamax(lm+1,abd(m,k),1) + m - 1 |
| 135 |
ipvt(k) = l + k - m |
| 136 |
c |
| 137 |
c zero pivot implies this column already triangularized |
| 138 |
c |
| 139 |
if (abd(l,k) .eq. 0.0d0) go to 100 |
| 140 |
c |
| 141 |
c interchange if necessary |
| 142 |
c |
| 143 |
if (l .eq. m) go to 60 |
| 144 |
t = abd(l,k) |
| 145 |
abd(l,k) = abd(m,k) |
| 146 |
abd(m,k) = t |
| 147 |
60 continue |
| 148 |
c |
| 149 |
c compute multipliers |
| 150 |
c |
| 151 |
t = -1.0d0/abd(m,k) |
| 152 |
call dscal(lm,t,abd(m+1,k),1) |
| 153 |
c |
| 154 |
c row elimination with column indexing |
| 155 |
c |
| 156 |
ju = min0(max0(ju,mu+ipvt(k)),n) |
| 157 |
mm = m |
| 158 |
if (ju .lt. kp1) go to 90 |
| 159 |
do 80 j = kp1, ju |
| 160 |
l = l - 1 |
| 161 |
mm = mm - 1 |
| 162 |
t = abd(l,j) |
| 163 |
if (l .eq. mm) go to 70 |
| 164 |
abd(l,j) = abd(mm,j) |
| 165 |
abd(mm,j) = t |
| 166 |
70 continue |
| 167 |
call daxpy(lm,t,abd(m+1,k),1,abd(mm+1,j),1) |
| 168 |
80 continue |
| 169 |
90 continue |
| 170 |
go to 110 |
| 171 |
100 continue |
| 172 |
info = k |
| 173 |
110 continue |
| 174 |
120 continue |
| 175 |
130 continue |
| 176 |
ipvt(n) = n |
| 177 |
if (abd(m,n) .eq. 0.0d0) info = n |
| 178 |
return |
| 179 |
end |