line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
#include |
2
|
|
|
|
|
|
|
/* |
3
|
|
|
|
|
|
|
Copyright © 2002, University of Tennessee Research Foundation. |
4
|
|
|
|
|
|
|
All rights reserved. |
5
|
|
|
|
|
|
|
|
6
|
|
|
|
|
|
|
Redistribution and use in source and binary forms, with or without |
7
|
|
|
|
|
|
|
modification, are permitted provided that the following conditions are met: |
8
|
|
|
|
|
|
|
|
9
|
|
|
|
|
|
|
* Redistributions of source code must retain the above copyright notice, this |
10
|
|
|
|
|
|
|
list of conditions and the following disclaimer. |
11
|
|
|
|
|
|
|
|
12
|
|
|
|
|
|
|
Redistributions in binary form must reproduce the above copyright notice, |
13
|
|
|
|
|
|
|
this list of conditions and the following disclaimer in the documentation |
14
|
|
|
|
|
|
|
and/or other materials provided with the distribution. |
15
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
* Neither the name of the University of Tennessee nor the names of its |
17
|
|
|
|
|
|
|
contributors may be used to endorse or promote products derived from this |
18
|
|
|
|
|
|
|
software without specific prior written permission. |
19
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
21
|
|
|
|
|
|
|
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
22
|
|
|
|
|
|
|
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
23
|
|
|
|
|
|
|
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
24
|
|
|
|
|
|
|
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
25
|
|
|
|
|
|
|
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
26
|
|
|
|
|
|
|
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
27
|
|
|
|
|
|
|
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
28
|
|
|
|
|
|
|
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
29
|
|
|
|
|
|
|
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
30
|
|
|
|
|
|
|
POSSIBILITY OF SUCH DAMAGE. |
31
|
|
|
|
|
|
|
*/ |
32
|
|
|
|
|
|
|
/* |
33
|
|
|
|
|
|
|
* Modified by moocow for PDL::SVDLIBC distribution |
34
|
|
|
|
|
|
|
*/ |
35
|
|
|
|
|
|
|
|
36
|
|
|
|
|
|
|
#include |
37
|
|
|
|
|
|
|
#include |
38
|
|
|
|
|
|
|
#include |
39
|
|
|
|
|
|
|
#include |
40
|
|
|
|
|
|
|
#include |
41
|
|
|
|
|
|
|
#include |
42
|
|
|
|
|
|
|
#include "svdlib.h" |
43
|
|
|
|
|
|
|
#include "svdutil.h" |
44
|
|
|
|
|
|
|
|
45
|
|
|
|
|
|
|
#define MAXLL 2 |
46
|
|
|
|
|
|
|
|
47
|
|
|
|
|
|
|
#define LMTNW 100000000 /* max. size of working area allowed */ |
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
enum storeVals {STORQ = 1, RETRQ, STORP, RETRP}; |
50
|
|
|
|
|
|
|
|
51
|
|
|
|
|
|
|
static char *error_msg[] = { /* error messages used by function * |
52
|
|
|
|
|
|
|
* check_parameters */ |
53
|
|
|
|
|
|
|
NULL, |
54
|
|
|
|
|
|
|
"", |
55
|
|
|
|
|
|
|
"ENDL MUST BE LESS THAN ENDR", |
56
|
|
|
|
|
|
|
"REQUESTED DIMENSIONS CANNOT EXCEED NUM ITERATIONS", |
57
|
|
|
|
|
|
|
"ONE OF YOUR DIMENSIONS IS LESS THAN OR EQUAL TO ZERO", |
58
|
|
|
|
|
|
|
"NUM ITERATIONS (NUMBER OF LANCZOS STEPS) IS INVALID", |
59
|
|
|
|
|
|
|
"REQUESTED DIMENSIONS (NUMBER OF EIGENPAIRS DESIRED) IS INVALID", |
60
|
|
|
|
|
|
|
"6*N+4*ITERATIONS+1 + ITERATIONS*ITERATIONS CANNOT EXCEED NW", |
61
|
|
|
|
|
|
|
"6*N+4*ITERATIONS+1 CANNOT EXCEED NW", NULL}; |
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
double **LanStore, *OPBTemp; |
64
|
|
|
|
|
|
|
double eps, eps1, reps, eps34; |
65
|
|
|
|
|
|
|
__SVDLIBC_LONG ierr; |
66
|
|
|
|
|
|
|
/* |
67
|
|
|
|
|
|
|
double rnm, anorm, tol; |
68
|
|
|
|
|
|
|
FILE *fp_out1, *fp_out2; |
69
|
|
|
|
|
|
|
*/ |
70
|
|
|
|
|
|
|
|
71
|
|
|
|
|
|
|
void purge(__SVDLIBC_LONG n, __SVDLIBC_LONG ll, double *r, double *q, double *ra, |
72
|
|
|
|
|
|
|
double *qa, double *wrk, double *eta, double *oldeta, __SVDLIBC_LONG step, |
73
|
|
|
|
|
|
|
double *rnmp, double tol); |
74
|
|
|
|
|
|
|
void ortbnd(double *alf, double *eta, double *oldeta, double *bet, __SVDLIBC_LONG step, |
75
|
|
|
|
|
|
|
double rnm); |
76
|
|
|
|
|
|
|
double startv(SMat A, double *wptr[], __SVDLIBC_LONG step, __SVDLIBC_LONG n); |
77
|
|
|
|
|
|
|
void store(__SVDLIBC_LONG, __SVDLIBC_LONG, __SVDLIBC_LONG, double *); |
78
|
|
|
|
|
|
|
void imtql2(__SVDLIBC_LONG, __SVDLIBC_LONG, double *, double *, double *); |
79
|
|
|
|
|
|
|
void imtqlb(__SVDLIBC_LONG n, double d[], double e[], double bnd[]); |
80
|
|
|
|
|
|
|
void write_header(__SVDLIBC_LONG, __SVDLIBC_LONG, double, double, __SVDLIBC_LONG, double, __SVDLIBC_LONG, __SVDLIBC_LONG, |
81
|
|
|
|
|
|
|
__SVDLIBC_LONG); |
82
|
|
|
|
|
|
|
__SVDLIBC_LONG check_parameters(SMat A, __SVDLIBC_LONG dimensions, __SVDLIBC_LONG iterations, |
83
|
|
|
|
|
|
|
double endl, double endr, __SVDLIBC_LONG vectors); |
84
|
|
|
|
|
|
|
int lanso(SMat A, __SVDLIBC_LONG iterations, __SVDLIBC_LONG dimensions, double endl, |
85
|
|
|
|
|
|
|
double endr, double *ritz, double *bnd, double *wptr[], |
86
|
|
|
|
|
|
|
__SVDLIBC_LONG *neigp, __SVDLIBC_LONG n); |
87
|
|
|
|
|
|
|
__SVDLIBC_LONG ritvec(__SVDLIBC_LONG n, SMat A, SVDRec R, double kappa, double *ritz, |
88
|
|
|
|
|
|
|
double *bnd, double *alf, double *bet, double *w2, |
89
|
|
|
|
|
|
|
__SVDLIBC_LONG steps, __SVDLIBC_LONG neig); |
90
|
|
|
|
|
|
|
__SVDLIBC_LONG lanczos_step(SMat A, __SVDLIBC_LONG first, __SVDLIBC_LONG last, double *wptr[], |
91
|
|
|
|
|
|
|
double *alf, double *eta, double *oldeta, |
92
|
|
|
|
|
|
|
double *bet, __SVDLIBC_LONG *ll, __SVDLIBC_LONG *enough, double *rnmp, |
93
|
|
|
|
|
|
|
double *tolp, __SVDLIBC_LONG n); |
94
|
|
|
|
|
|
|
void stpone(SMat A, double *wrkptr[], double *rnmp, double *tolp, __SVDLIBC_LONG n); |
95
|
|
|
|
|
|
|
__SVDLIBC_LONG error_bound(__SVDLIBC_LONG *, double, double, double *, double *, __SVDLIBC_LONG step, |
96
|
|
|
|
|
|
|
double tol); |
97
|
|
|
|
|
|
|
void machar(__SVDLIBC_LONG *ibeta, __SVDLIBC_LONG *it, __SVDLIBC_LONG *irnd, __SVDLIBC_LONG *machep, __SVDLIBC_LONG *negep); |
98
|
|
|
|
|
|
|
|
99
|
|
|
|
|
|
|
/*********************************************************************** |
100
|
|
|
|
|
|
|
* * |
101
|
|
|
|
|
|
|
* main() * |
102
|
|
|
|
|
|
|
* Sparse SVD(A) via Eigensystem of A'A symmetric Matrix * |
103
|
|
|
|
|
|
|
* (double precision) * |
104
|
|
|
|
|
|
|
* * |
105
|
|
|
|
|
|
|
***********************************************************************/ |
106
|
|
|
|
|
|
|
/*********************************************************************** |
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
Description |
109
|
|
|
|
|
|
|
----------- |
110
|
|
|
|
|
|
|
|
111
|
|
|
|
|
|
|
This sample program uses landr to compute singular triplets of A via |
112
|
|
|
|
|
|
|
the equivalent symmetric eigenvalue problem |
113
|
|
|
|
|
|
|
|
114
|
|
|
|
|
|
|
B x = lambda x, where x' = (u',v'), lambda = sigma**2, |
115
|
|
|
|
|
|
|
where sigma is a singular value of A, |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
B = A'A , and A is m (nrow) by n (ncol) (nrow >> ncol), |
118
|
|
|
|
|
|
|
|
119
|
|
|
|
|
|
|
so that {u,sqrt(lambda),v} is a singular triplet of A. |
120
|
|
|
|
|
|
|
(A' = transpose of A) |
121
|
|
|
|
|
|
|
|
122
|
|
|
|
|
|
|
User supplied routines: svd_opa, opb, store, timer |
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
svd_opa( x,y) takes an n-vector x and returns A*x in y. |
125
|
|
|
|
|
|
|
svd_opb(ncol,x,y) takes an n-vector x and returns B*x in y. |
126
|
|
|
|
|
|
|
|
127
|
|
|
|
|
|
|
Based on operation flag isw, store(n,isw,j,s) stores/retrieves |
128
|
|
|
|
|
|
|
to/from storage a vector of length n in s. |
129
|
|
|
|
|
|
|
|
130
|
|
|
|
|
|
|
User should edit timer() with an appropriate call to an intrinsic |
131
|
|
|
|
|
|
|
timing routine that returns elapsed user time. |
132
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
External parameters |
135
|
|
|
|
|
|
|
------------------- |
136
|
|
|
|
|
|
|
|
137
|
|
|
|
|
|
|
Defined and documented in las2.h |
138
|
|
|
|
|
|
|
|
139
|
|
|
|
|
|
|
|
140
|
|
|
|
|
|
|
Local parameters |
141
|
|
|
|
|
|
|
---------------- |
142
|
|
|
|
|
|
|
|
143
|
|
|
|
|
|
|
(input) |
144
|
|
|
|
|
|
|
endl left end of interval containing unwanted eigenvalues of B |
145
|
|
|
|
|
|
|
endr right end of interval containing unwanted eigenvalues of B |
146
|
|
|
|
|
|
|
kappa relative accuracy of ritz values acceptable as eigenvalues |
147
|
|
|
|
|
|
|
of B |
148
|
|
|
|
|
|
|
vectors is not equal to 1 |
149
|
|
|
|
|
|
|
r work array |
150
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B (ncol) |
151
|
|
|
|
|
|
|
dimensions upper limit of desired number of singular triplets of A |
152
|
|
|
|
|
|
|
iterations upper limit of desired number of Lanczos steps |
153
|
|
|
|
|
|
|
nnzero number of nonzeros in A |
154
|
|
|
|
|
|
|
vectors 1 indicates both singular values and singular vectors are |
155
|
|
|
|
|
|
|
wanted and they can be found in output file lav2; |
156
|
|
|
|
|
|
|
0 indicates only singular values are wanted |
157
|
|
|
|
|
|
|
|
158
|
|
|
|
|
|
|
(output) |
159
|
|
|
|
|
|
|
ritz array of ritz values |
160
|
|
|
|
|
|
|
bnd array of error bounds |
161
|
|
|
|
|
|
|
d array of singular values |
162
|
|
|
|
|
|
|
memory total memory allocated in bytes to solve the B-eigenproblem |
163
|
|
|
|
|
|
|
|
164
|
|
|
|
|
|
|
|
165
|
|
|
|
|
|
|
Functions used |
166
|
|
|
|
|
|
|
-------------- |
167
|
|
|
|
|
|
|
|
168
|
|
|
|
|
|
|
BLAS svd_daxpy, svd_dscal, svd_ddot |
169
|
|
|
|
|
|
|
USER svd_opa, svd_opb, timer |
170
|
|
|
|
|
|
|
MISC write_header, check_parameters |
171
|
|
|
|
|
|
|
LAS2 landr |
172
|
|
|
|
|
|
|
|
173
|
|
|
|
|
|
|
|
174
|
|
|
|
|
|
|
Precision |
175
|
|
|
|
|
|
|
--------- |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
All floating-point calculations are done in double precision; |
178
|
|
|
|
|
|
|
variables are declared as __SVDLIBC_LONG and double. |
179
|
|
|
|
|
|
|
|
180
|
|
|
|
|
|
|
|
181
|
|
|
|
|
|
|
LAS2 development |
182
|
|
|
|
|
|
|
---------------- |
183
|
|
|
|
|
|
|
|
184
|
|
|
|
|
|
|
LAS2 is a C translation of the Fortran-77 LAS2 from the SVDPACK |
185
|
|
|
|
|
|
|
library written by Michael W. Berry, University of Tennessee, |
186
|
|
|
|
|
|
|
Dept. of Computer Science, 107 Ayres Hall, Knoxville, TN, 37996-1301 |
187
|
|
|
|
|
|
|
|
188
|
|
|
|
|
|
|
31 Jan 1992: Date written |
189
|
|
|
|
|
|
|
|
190
|
|
|
|
|
|
|
Theresa H. Do |
191
|
|
|
|
|
|
|
University of Tennessee |
192
|
|
|
|
|
|
|
Dept. of Computer Science |
193
|
|
|
|
|
|
|
107 Ayres Hall |
194
|
|
|
|
|
|
|
Knoxville, TN, 37996-1301 |
195
|
|
|
|
|
|
|
internet: tdo@cs.utk.edu |
196
|
|
|
|
|
|
|
|
197
|
|
|
|
|
|
|
***********************************************************************/ |
198
|
|
|
|
|
|
|
|
199
|
|
|
|
|
|
|
/*********************************************************************** |
200
|
|
|
|
|
|
|
* * |
201
|
|
|
|
|
|
|
* check_parameters() * |
202
|
|
|
|
|
|
|
* * |
203
|
|
|
|
|
|
|
***********************************************************************/ |
204
|
|
|
|
|
|
|
/*********************************************************************** |
205
|
|
|
|
|
|
|
|
206
|
|
|
|
|
|
|
Description |
207
|
|
|
|
|
|
|
----------- |
208
|
|
|
|
|
|
|
Function validates input parameters and returns error code (__SVDLIBC_LONG) |
209
|
|
|
|
|
|
|
|
210
|
|
|
|
|
|
|
Parameters |
211
|
|
|
|
|
|
|
---------- |
212
|
|
|
|
|
|
|
(input) |
213
|
|
|
|
|
|
|
dimensions upper limit of desired number of eigenpairs of B |
214
|
|
|
|
|
|
|
iterations upper limit of desired number of lanczos steps |
215
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
216
|
|
|
|
|
|
|
endl left end of interval containing unwanted eigenvalues of B |
217
|
|
|
|
|
|
|
endr right end of interval containing unwanted eigenvalues of B |
218
|
|
|
|
|
|
|
vectors 1 indicates both eigenvalues and eigenvectors are wanted |
219
|
|
|
|
|
|
|
and they can be found in lav2; 0 indicates eigenvalues only |
220
|
|
|
|
|
|
|
nnzero number of nonzero elements in input matrix (matrix A) |
221
|
|
|
|
|
|
|
|
222
|
|
|
|
|
|
|
***********************************************************************/ |
223
|
|
|
|
|
|
|
|
224
|
10
|
|
|
|
|
|
__SVDLIBC_LONG check_parameters(SMat A, __SVDLIBC_LONG dimensions, __SVDLIBC_LONG iterations, |
225
|
|
|
|
|
|
|
double endl, double endr, __SVDLIBC_LONG vectors) { |
226
|
|
|
|
|
|
|
__SVDLIBC_LONG error_index; |
227
|
10
|
|
|
|
|
|
error_index = 0; |
228
|
|
|
|
|
|
|
|
229
|
10
|
50
|
|
|
|
|
if (endl >/*=*/ endr) error_index = 2; |
230
|
10
|
50
|
|
|
|
|
else if (dimensions > iterations) error_index = 3; |
231
|
10
|
50
|
|
|
|
|
else if (A->cols <= 0 || A->rows <= 0) error_index = 4; |
|
|
50
|
|
|
|
|
|
232
|
|
|
|
|
|
|
/*else if (n > A->cols || n > A->rows) error_index = 1;*/ |
233
|
10
|
50
|
|
|
|
|
else if (iterations <= 0 || iterations > A->cols || iterations > A->rows) |
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
234
|
0
|
|
|
|
|
|
error_index = 5; |
235
|
10
|
50
|
|
|
|
|
else if (dimensions <= 0 || dimensions > iterations) error_index = 6; |
|
|
50
|
|
|
|
|
|
236
|
10
|
50
|
|
|
|
|
if (error_index) |
237
|
0
|
|
|
|
|
|
svd_error("svdLAS2 parameter error: %s\n", error_msg[error_index]); |
238
|
10
|
|
|
|
|
|
return(error_index); |
239
|
|
|
|
|
|
|
} |
240
|
|
|
|
|
|
|
|
241
|
|
|
|
|
|
|
/*********************************************************************** |
242
|
|
|
|
|
|
|
* * |
243
|
|
|
|
|
|
|
* write_header() * |
244
|
|
|
|
|
|
|
* Function writes out header of output file containing ritz values * |
245
|
|
|
|
|
|
|
* * |
246
|
|
|
|
|
|
|
***********************************************************************/ |
247
|
|
|
|
|
|
|
|
248
|
0
|
|
|
|
|
|
void write_header(__SVDLIBC_LONG iterations, __SVDLIBC_LONG dimensions, double endl, double endr, |
249
|
|
|
|
|
|
|
__SVDLIBC_LONG vectors, double kappa, __SVDLIBC_LONG nrow, __SVDLIBC_LONG ncol, |
250
|
|
|
|
|
|
|
__SVDLIBC_LONG vals) { |
251
|
0
|
|
|
|
|
|
printf("SOLVING THE [A^TA] EIGENPROBLEM\n"); |
252
|
0
|
|
|
|
|
|
printf("NO. OF ROWS = %6ld\n", nrow); |
253
|
0
|
|
|
|
|
|
printf("NO. OF COLUMNS = %6ld\n", ncol); |
254
|
0
|
|
|
|
|
|
printf("NO. OF NON-ZERO VALUES = %6ld\n", vals); |
255
|
0
|
|
|
|
|
|
printf("MATRIX DENSITY = %6.2f%%\n", |
256
|
0
|
|
|
|
|
|
((float) vals / nrow) * 100 / ncol); |
257
|
|
|
|
|
|
|
/* printf("ORDER OF MATRIX A = %5ld\n", n); */ |
258
|
0
|
|
|
|
|
|
printf("MAX. NO. OF LANCZOS STEPS = %6ld\n", iterations); |
259
|
0
|
|
|
|
|
|
printf("MAX. NO. OF EIGENPAIRS = %6ld\n", dimensions); |
260
|
0
|
|
|
|
|
|
printf("LEFT END OF THE INTERVAL = %9.2E\n", endl); |
261
|
0
|
|
|
|
|
|
printf("RIGHT END OF THE INTERVAL = %9.2E\n", endr); |
262
|
0
|
|
|
|
|
|
printf("KAPPA = %9.2E\n", kappa); |
263
|
|
|
|
|
|
|
/* printf("WANT S-VECTORS? [T/F] = %c\n", (vectors) ? 'T' : 'F'); */ |
264
|
0
|
|
|
|
|
|
printf("\n"); |
265
|
0
|
|
|
|
|
|
return; |
266
|
|
|
|
|
|
|
} |
267
|
|
|
|
|
|
|
|
268
|
|
|
|
|
|
|
|
269
|
|
|
|
|
|
|
/*********************************************************************** |
270
|
|
|
|
|
|
|
* * |
271
|
|
|
|
|
|
|
* landr() * |
272
|
|
|
|
|
|
|
* Lanczos algorithm with selective orthogonalization * |
273
|
|
|
|
|
|
|
* Using Simon's Recurrence * |
274
|
|
|
|
|
|
|
* (double precision) * |
275
|
|
|
|
|
|
|
* * |
276
|
|
|
|
|
|
|
***********************************************************************/ |
277
|
|
|
|
|
|
|
/*********************************************************************** |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
Description |
280
|
|
|
|
|
|
|
----------- |
281
|
|
|
|
|
|
|
|
282
|
|
|
|
|
|
|
landr() is the LAS2 driver routine that, upon entry, |
283
|
|
|
|
|
|
|
(1) checks for the validity of input parameters of the |
284
|
|
|
|
|
|
|
B-eigenproblem |
285
|
|
|
|
|
|
|
(2) determines several machine constants |
286
|
|
|
|
|
|
|
(3) makes a Lanczos run |
287
|
|
|
|
|
|
|
(4) calculates B-eigenvectors (singular vectors of A) if requested |
288
|
|
|
|
|
|
|
by user |
289
|
|
|
|
|
|
|
|
290
|
|
|
|
|
|
|
|
291
|
|
|
|
|
|
|
arguments |
292
|
|
|
|
|
|
|
--------- |
293
|
|
|
|
|
|
|
|
294
|
|
|
|
|
|
|
(input) |
295
|
|
|
|
|
|
|
n dimension of the eigenproblem for A'A |
296
|
|
|
|
|
|
|
iterations upper limit of desired number of Lanczos steps |
297
|
|
|
|
|
|
|
dimensions upper limit of desired number of eigenpairs |
298
|
|
|
|
|
|
|
nnzero number of nonzeros in matrix A |
299
|
|
|
|
|
|
|
endl left end of interval containing unwanted eigenvalues of B |
300
|
|
|
|
|
|
|
endr right end of interval containing unwanted eigenvalues of B |
301
|
|
|
|
|
|
|
vectors 1 indicates both eigenvalues and eigenvectors are wanted |
302
|
|
|
|
|
|
|
and they can be found in output file lav2; |
303
|
|
|
|
|
|
|
0 indicates only eigenvalues are wanted |
304
|
|
|
|
|
|
|
kappa relative accuracy of ritz values acceptable as eigenvalues |
305
|
|
|
|
|
|
|
of B (singular values of A) |
306
|
|
|
|
|
|
|
r work array |
307
|
|
|
|
|
|
|
|
308
|
|
|
|
|
|
|
(output) |
309
|
|
|
|
|
|
|
j number of Lanczos steps actually taken |
310
|
|
|
|
|
|
|
neig number of ritz values stabilized |
311
|
|
|
|
|
|
|
ritz array to hold the ritz values |
312
|
|
|
|
|
|
|
bnd array to hold the error bounds |
313
|
|
|
|
|
|
|
|
314
|
|
|
|
|
|
|
|
315
|
|
|
|
|
|
|
External parameters |
316
|
|
|
|
|
|
|
------------------- |
317
|
|
|
|
|
|
|
|
318
|
|
|
|
|
|
|
Defined and documented in las2.h |
319
|
|
|
|
|
|
|
|
320
|
|
|
|
|
|
|
|
321
|
|
|
|
|
|
|
local parameters |
322
|
|
|
|
|
|
|
------------------- |
323
|
|
|
|
|
|
|
|
324
|
|
|
|
|
|
|
ibeta radix for the floating-point representation |
325
|
|
|
|
|
|
|
it number of base ibeta digits in the floating-point significand |
326
|
|
|
|
|
|
|
irnd floating-point addition rounded or chopped |
327
|
|
|
|
|
|
|
machep machine relative precision or round-off error |
328
|
|
|
|
|
|
|
negeps largest negative integer |
329
|
|
|
|
|
|
|
wptr array of pointers each pointing to a work space |
330
|
|
|
|
|
|
|
|
331
|
|
|
|
|
|
|
|
332
|
|
|
|
|
|
|
Functions used |
333
|
|
|
|
|
|
|
-------------- |
334
|
|
|
|
|
|
|
|
335
|
|
|
|
|
|
|
MISC svd_dmax, machar, check_parameters |
336
|
|
|
|
|
|
|
LAS2 ritvec, lanso |
337
|
|
|
|
|
|
|
|
338
|
|
|
|
|
|
|
***********************************************************************/ |
339
|
|
|
|
|
|
|
|
340
|
0
|
|
|
|
|
|
SVDRec svdLAS2A(SMat A, __SVDLIBC_LONG dimensions) { |
341
|
0
|
|
|
|
|
|
double end[2] = {-1.0e-30, 1.0e-30}; |
342
|
0
|
|
|
|
|
|
double kappa = 1e-6; |
343
|
0
|
0
|
|
|
|
|
if (!A) { |
344
|
0
|
|
|
|
|
|
svd_error("svdLAS2A called with NULL array\n"); |
345
|
0
|
|
|
|
|
|
return NULL; |
346
|
|
|
|
|
|
|
} |
347
|
0
|
|
|
|
|
|
return svdLAS2(A, dimensions, 0, end, kappa); |
348
|
|
|
|
|
|
|
} |
349
|
|
|
|
|
|
|
|
350
|
|
|
|
|
|
|
|
351
|
10
|
|
|
|
|
|
SVDRec svdLAS2(SMat A, __SVDLIBC_LONG dimensions, __SVDLIBC_LONG iterations, double end[2], |
352
|
|
|
|
|
|
|
double kappa) { |
353
|
10
|
|
|
|
|
|
char transpose = FALSE; |
354
|
|
|
|
|
|
|
__SVDLIBC_LONG ibeta, it, irnd, machep, negep, n, i, steps, nsig, neig, m; |
355
|
|
|
|
|
|
|
double *wptr[10], *ritz, *bnd; |
356
|
10
|
|
|
|
|
|
SVDRec R = NULL; |
357
|
10
|
|
|
|
|
|
ierr = 0; // reset the global error flag |
358
|
|
|
|
|
|
|
|
359
|
10
|
|
|
|
|
|
svdResetCounters(); |
360
|
|
|
|
|
|
|
|
361
|
10
|
|
|
|
|
|
m = svd_imin(A->rows, A->cols); |
362
|
10
|
50
|
|
|
|
|
if (dimensions <= 0 || dimensions > m) |
|
|
100
|
|
|
|
|
|
363
|
7
|
|
|
|
|
|
dimensions = m; |
364
|
10
|
50
|
|
|
|
|
if (iterations <= 0 || iterations > m) |
|
|
50
|
|
|
|
|
|
365
|
10
|
|
|
|
|
|
iterations = m; |
366
|
10
|
50
|
|
|
|
|
if (iterations < dimensions) iterations = dimensions; |
367
|
|
|
|
|
|
|
|
368
|
|
|
|
|
|
|
/* Write output header */ |
369
|
10
|
50
|
|
|
|
|
if (SVDVerbosity > 0) |
370
|
0
|
|
|
|
|
|
write_header(iterations, dimensions, end[0], end[1], TRUE, kappa, A->rows, |
371
|
|
|
|
|
|
|
A->cols, A->vals); |
372
|
|
|
|
|
|
|
|
373
|
|
|
|
|
|
|
/* Check parameters */ |
374
|
10
|
50
|
|
|
|
|
if (check_parameters(A, dimensions, iterations, end[0], end[1], TRUE)) |
375
|
0
|
|
|
|
|
|
return NULL; |
376
|
|
|
|
|
|
|
|
377
|
|
|
|
|
|
|
#if 0 |
378
|
|
|
|
|
|
|
/* Fri, 24 Jul 2015 11:31:27 +0200 moocow |
379
|
|
|
|
|
|
|
* + disable transposition for PDL::SVDLIBC, since we're getting segfaults whenever this code runs |
380
|
|
|
|
|
|
|
*/ |
381
|
|
|
|
|
|
|
/* If A is wide, the SVD is computed on its transpose for speed. */ |
382
|
|
|
|
|
|
|
if (A->cols >= A->rows * 1.2) { |
383
|
|
|
|
|
|
|
if (SVDVerbosity > 0) printf("TRANSPOSING THE MATRIX FOR SPEED\n"); |
384
|
|
|
|
|
|
|
transpose = TRUE; |
385
|
|
|
|
|
|
|
A = svdTransposeS(A); |
386
|
|
|
|
|
|
|
} |
387
|
|
|
|
|
|
|
#endif |
388
|
|
|
|
|
|
|
|
389
|
10
|
|
|
|
|
|
n = A->cols; |
390
|
|
|
|
|
|
|
/* Compute machine precision */ |
391
|
10
|
|
|
|
|
|
machar(&ibeta, &it, &irnd, &machep, &negep); |
392
|
10
|
|
|
|
|
|
eps1 = eps * sqrt((double) n); |
393
|
10
|
|
|
|
|
|
reps = sqrt(eps); |
394
|
10
|
|
|
|
|
|
eps34 = reps * sqrt(reps); |
395
|
|
|
|
|
|
|
|
396
|
|
|
|
|
|
|
/* Allocate temporary space. */ |
397
|
10
|
50
|
|
|
|
|
if (!(wptr[0] = svd_doubleArray(n, TRUE, "las2: wptr[0]"))) goto abort; |
398
|
10
|
50
|
|
|
|
|
if (!(wptr[1] = svd_doubleArray(n, FALSE, "las2: wptr[1]"))) goto abort; |
399
|
10
|
50
|
|
|
|
|
if (!(wptr[2] = svd_doubleArray(n, FALSE, "las2: wptr[2]"))) goto abort; |
400
|
10
|
50
|
|
|
|
|
if (!(wptr[3] = svd_doubleArray(n, FALSE, "las2: wptr[3]"))) goto abort; |
401
|
10
|
50
|
|
|
|
|
if (!(wptr[4] = svd_doubleArray(n, FALSE, "las2: wptr[4]"))) goto abort; |
402
|
10
|
50
|
|
|
|
|
if (!(wptr[5] = svd_doubleArray(n, FALSE, "las2: wptr[5]"))) goto abort; |
403
|
10
|
50
|
|
|
|
|
if (!(wptr[6] = svd_doubleArray(iterations, FALSE, "las2: wptr[6]"))) |
404
|
0
|
|
|
|
|
|
goto abort; |
405
|
10
|
50
|
|
|
|
|
if (!(wptr[7] = svd_doubleArray(iterations, FALSE, "las2: wptr[7]"))) |
406
|
0
|
|
|
|
|
|
goto abort; |
407
|
10
|
50
|
|
|
|
|
if (!(wptr[8] = svd_doubleArray(iterations, FALSE, "las2: wptr[8]"))) |
408
|
0
|
|
|
|
|
|
goto abort; |
409
|
10
|
50
|
|
|
|
|
if (!(wptr[9] = svd_doubleArray(iterations + 1, FALSE, "las2: wptr[9]"))) |
410
|
0
|
|
|
|
|
|
goto abort; |
411
|
|
|
|
|
|
|
/* Calloc may be unnecessary: */ |
412
|
10
|
50
|
|
|
|
|
if (!(ritz = svd_doubleArray(iterations + 1, TRUE, "las2: ritz"))) |
413
|
0
|
|
|
|
|
|
goto abort; |
414
|
|
|
|
|
|
|
/* Calloc may be unnecessary: */ |
415
|
10
|
50
|
|
|
|
|
if (!(bnd = svd_doubleArray(iterations + 1, TRUE, "las2: bnd"))) |
416
|
0
|
|
|
|
|
|
goto abort; |
417
|
10
|
|
|
|
|
|
memset(bnd, 127, (iterations + 1) * sizeof(double)); |
418
|
|
|
|
|
|
|
|
419
|
10
|
50
|
|
|
|
|
if (!(LanStore = (double **) calloc(iterations + MAXLL, sizeof(double *)))) |
420
|
0
|
|
|
|
|
|
goto abort; |
421
|
10
|
50
|
|
|
|
|
if (!(OPBTemp = svd_doubleArray(A->rows, FALSE, "las2: OPBTemp"))) |
422
|
0
|
|
|
|
|
|
goto abort; |
423
|
|
|
|
|
|
|
|
424
|
|
|
|
|
|
|
/* Actually run the lanczos thing: */ |
425
|
10
|
|
|
|
|
|
steps = lanso(A, iterations, dimensions, end[0], end[1], ritz, bnd, wptr, |
426
|
|
|
|
|
|
|
&neig, n); |
427
|
|
|
|
|
|
|
|
428
|
|
|
|
|
|
|
/* Print some stuff. */ |
429
|
10
|
50
|
|
|
|
|
if (SVDVerbosity > 0) { |
430
|
0
|
|
|
|
|
|
printf("NUMBER OF LANCZOS STEPS = %6ld\n" |
431
|
|
|
|
|
|
|
"RITZ VALUES STABILIZED = %6ld\n", steps + 1, neig); |
432
|
|
|
|
|
|
|
} |
433
|
10
|
50
|
|
|
|
|
if (SVDVerbosity > 2) { |
434
|
0
|
|
|
|
|
|
printf("\nCOMPUTED RITZ VALUES (ERROR BNDS)\n"); |
435
|
0
|
0
|
|
|
|
|
for (i = 0; i <= steps; i++) |
436
|
0
|
|
|
|
|
|
printf("%3ld %22.14E (%11.2E)\n", i + 1, ritz[i], bnd[i]); |
437
|
|
|
|
|
|
|
} |
438
|
|
|
|
|
|
|
|
439
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[0]); |
440
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[1]); |
441
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[2]); |
442
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[3]); |
443
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[4]); |
444
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[7]); |
445
|
10
|
50
|
|
|
|
|
SAFE_FREE(wptr[8]); |
446
|
|
|
|
|
|
|
|
447
|
|
|
|
|
|
|
/* Compute eigenvectors */ |
448
|
10
|
|
|
|
|
|
kappa = svd_dmax(fabs(kappa), eps34); |
449
|
|
|
|
|
|
|
|
450
|
10
|
|
|
|
|
|
R = svdNewSVDRec(); |
451
|
10
|
50
|
|
|
|
|
if (!R) { |
452
|
0
|
|
|
|
|
|
svd_error("svdLAS2: allocation of R failed"); |
453
|
0
|
|
|
|
|
|
goto cleanup; |
454
|
|
|
|
|
|
|
} |
455
|
10
|
|
|
|
|
|
R->d = /*svd_imin(nsig, dimensions)*/dimensions; |
456
|
10
|
|
|
|
|
|
R->Ut = svdNewDMat(R->d, A->rows); |
457
|
10
|
|
|
|
|
|
R->S = svd_doubleArray(R->d, TRUE, "las2: R->s"); |
458
|
10
|
|
|
|
|
|
R->Vt = svdNewDMat(R->d, A->cols); |
459
|
10
|
50
|
|
|
|
|
if (!R->Ut || !R->S || !R->Vt) { |
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
460
|
0
|
|
|
|
|
|
svd_error("svdLAS2: allocation of R failed"); |
461
|
0
|
|
|
|
|
|
goto cleanup; |
462
|
|
|
|
|
|
|
} |
463
|
|
|
|
|
|
|
|
464
|
10
|
|
|
|
|
|
nsig = ritvec(n, A, R, kappa, ritz, bnd, wptr[6], wptr[9], wptr[5], steps, |
465
|
|
|
|
|
|
|
neig); |
466
|
|
|
|
|
|
|
|
467
|
10
|
50
|
|
|
|
|
if (SVDVerbosity > 1) { |
468
|
0
|
|
|
|
|
|
printf("\nSINGULAR VALUES: "); |
469
|
0
|
|
|
|
|
|
svdWriteDenseArray(R->S, R->d, "-", FALSE); |
470
|
|
|
|
|
|
|
|
471
|
0
|
0
|
|
|
|
|
if (SVDVerbosity > 2) { |
472
|
0
|
|
|
|
|
|
printf("\nLEFT SINGULAR VECTORS (transpose of U): "); |
473
|
0
|
|
|
|
|
|
svdWriteDenseMatrix(R->Ut, "-", SVD_F_DT); |
474
|
|
|
|
|
|
|
|
475
|
0
|
|
|
|
|
|
printf("\nRIGHT SINGULAR VECTORS (transpose of V): "); |
476
|
0
|
|
|
|
|
|
svdWriteDenseMatrix(R->Vt, "-", SVD_F_DT); |
477
|
|
|
|
|
|
|
} |
478
|
|
|
|
|
|
|
} |
479
|
10
|
50
|
|
|
|
|
if (SVDVerbosity > 0) { |
480
|
0
|
|
|
|
|
|
printf("SINGULAR VALUES FOUND = %6d\n" |
481
|
|
|
|
|
|
|
"SIGNIFICANT VALUES = %6ld\n", R->d, nsig); |
482
|
|
|
|
|
|
|
} |
483
|
|
|
|
|
|
|
|
484
|
|
|
|
|
|
|
cleanup: |
485
|
110
|
100
|
|
|
|
|
for (i = 0; i <= 9; i++) |
486
|
100
|
100
|
|
|
|
|
SAFE_FREE(wptr[i]); |
487
|
10
|
50
|
|
|
|
|
SAFE_FREE(ritz); |
488
|
10
|
50
|
|
|
|
|
SAFE_FREE(bnd); |
489
|
10
|
50
|
|
|
|
|
if (LanStore) { |
490
|
90
|
100
|
|
|
|
|
for (i = 0; i < iterations + MAXLL; i++) |
491
|
80
|
50
|
|
|
|
|
SAFE_FREE(LanStore[i]); |
492
|
10
|
50
|
|
|
|
|
SAFE_FREE(LanStore); |
493
|
|
|
|
|
|
|
} |
494
|
10
|
50
|
|
|
|
|
SAFE_FREE(OPBTemp); |
495
|
|
|
|
|
|
|
|
496
|
|
|
|
|
|
|
/* This swaps and transposes the singular matrices if A was transposed. */ |
497
|
10
|
50
|
|
|
|
|
if (R && transpose) { |
|
|
50
|
|
|
|
|
|
498
|
|
|
|
|
|
|
DMat T; |
499
|
0
|
|
|
|
|
|
svdFreeSMat(A); |
500
|
0
|
|
|
|
|
|
T = R->Ut; |
501
|
0
|
|
|
|
|
|
R->Ut = R->Vt; |
502
|
0
|
|
|
|
|
|
R->Vt = T; |
503
|
|
|
|
|
|
|
} |
504
|
|
|
|
|
|
|
|
505
|
10
|
|
|
|
|
|
return R; |
506
|
|
|
|
|
|
|
abort: |
507
|
0
|
|
|
|
|
|
svd_error("svdLAS2: fatal error, aborting"); |
508
|
10
|
|
|
|
|
|
return NULL; |
509
|
|
|
|
|
|
|
} |
510
|
|
|
|
|
|
|
|
511
|
|
|
|
|
|
|
|
512
|
|
|
|
|
|
|
/*********************************************************************** |
513
|
|
|
|
|
|
|
* * |
514
|
|
|
|
|
|
|
* ritvec() * |
515
|
|
|
|
|
|
|
* Function computes the singular vectors of matrix A * |
516
|
|
|
|
|
|
|
* * |
517
|
|
|
|
|
|
|
***********************************************************************/ |
518
|
|
|
|
|
|
|
/*********************************************************************** |
519
|
|
|
|
|
|
|
|
520
|
|
|
|
|
|
|
Description |
521
|
|
|
|
|
|
|
----------- |
522
|
|
|
|
|
|
|
|
523
|
|
|
|
|
|
|
This function is invoked by landr() only if eigenvectors of the A'A |
524
|
|
|
|
|
|
|
eigenproblem are desired. When called, ritvec() computes the |
525
|
|
|
|
|
|
|
singular vectors of A and writes the result to an unformatted file. |
526
|
|
|
|
|
|
|
|
527
|
|
|
|
|
|
|
|
528
|
|
|
|
|
|
|
Parameters |
529
|
|
|
|
|
|
|
---------- |
530
|
|
|
|
|
|
|
|
531
|
|
|
|
|
|
|
(input) |
532
|
|
|
|
|
|
|
nrow number of rows of A |
533
|
|
|
|
|
|
|
steps number of Lanczos iterations performed |
534
|
|
|
|
|
|
|
fp_out2 pointer to unformatted output file |
535
|
|
|
|
|
|
|
n dimension of matrix A |
536
|
|
|
|
|
|
|
kappa relative accuracy of ritz values acceptable as |
537
|
|
|
|
|
|
|
eigenvalues of A'A |
538
|
|
|
|
|
|
|
ritz array of ritz values |
539
|
|
|
|
|
|
|
bnd array of error bounds |
540
|
|
|
|
|
|
|
alf array of diagonal elements of the tridiagonal matrix T |
541
|
|
|
|
|
|
|
bet array of off-diagonal elements of T |
542
|
|
|
|
|
|
|
w1, w2 work space |
543
|
|
|
|
|
|
|
|
544
|
|
|
|
|
|
|
(output) |
545
|
|
|
|
|
|
|
xv1 array of eigenvectors of A'A (right singular vectors of A) |
546
|
|
|
|
|
|
|
ierr error code |
547
|
|
|
|
|
|
|
0 for normal return from imtql2() |
548
|
|
|
|
|
|
|
k if convergence did not occur for k-th eigenvalue in |
549
|
|
|
|
|
|
|
imtql2() |
550
|
|
|
|
|
|
|
nsig number of accepted ritz values based on kappa |
551
|
|
|
|
|
|
|
|
552
|
|
|
|
|
|
|
(local) |
553
|
|
|
|
|
|
|
s work array which is initialized to the identity matrix |
554
|
|
|
|
|
|
|
of order (j + 1) upon calling imtql2(). After the call, |
555
|
|
|
|
|
|
|
s contains the orthonormal eigenvectors of the symmetric |
556
|
|
|
|
|
|
|
tridiagonal matrix T |
557
|
|
|
|
|
|
|
|
558
|
|
|
|
|
|
|
Functions used |
559
|
|
|
|
|
|
|
-------------- |
560
|
|
|
|
|
|
|
|
561
|
|
|
|
|
|
|
BLAS svd_dscal, svd_dcopy, svd_daxpy |
562
|
|
|
|
|
|
|
USER store |
563
|
|
|
|
|
|
|
imtql2 |
564
|
|
|
|
|
|
|
|
565
|
|
|
|
|
|
|
***********************************************************************/ |
566
|
|
|
|
|
|
|
|
567
|
10
|
|
|
|
|
|
void rotateArray(double *a, int size, int x) { |
568
|
|
|
|
|
|
|
int i, j, n, start; |
569
|
|
|
|
|
|
|
double t1, t2; |
570
|
10
|
100
|
|
|
|
|
if (x == 0) return; |
571
|
3
|
|
|
|
|
|
j = start = 0; |
572
|
3
|
|
|
|
|
|
t1 = a[0]; |
573
|
108
|
100
|
|
|
|
|
for (i = 0; i < size; i++) { |
574
|
105
|
100
|
|
|
|
|
n = (j >= x) ? j - x : j + size - x; |
575
|
105
|
|
|
|
|
|
t2 = a[n]; |
576
|
105
|
|
|
|
|
|
a[n] = t1; |
577
|
105
|
|
|
|
|
|
t1 = t2; |
578
|
105
|
|
|
|
|
|
j = n; |
579
|
105
|
100
|
|
|
|
|
if (j == start) { |
580
|
21
|
|
|
|
|
|
start = ++j; |
581
|
21
|
|
|
|
|
|
t1 = a[j]; |
582
|
|
|
|
|
|
|
} |
583
|
|
|
|
|
|
|
} |
584
|
|
|
|
|
|
|
} |
585
|
|
|
|
|
|
|
|
586
|
10
|
|
|
|
|
|
__SVDLIBC_LONG ritvec(__SVDLIBC_LONG n, SMat A, SVDRec R, double kappa, double *ritz, double *bnd, |
587
|
|
|
|
|
|
|
double *alf, double *bet, double *w2, __SVDLIBC_LONG steps, __SVDLIBC_LONG neig) { |
588
|
|
|
|
|
|
|
__SVDLIBC_LONG js, jsq, i, k, /*size,*/ id2, tmp, nsig, x; |
589
|
10
|
|
|
|
|
|
double *s, *xv2, tmp0, tmp1, xnorm, *w1 = R->Vt->value[0]; |
590
|
|
|
|
|
|
|
|
591
|
10
|
|
|
|
|
|
js = steps + 1; |
592
|
10
|
|
|
|
|
|
jsq = js * js; |
593
|
|
|
|
|
|
|
/*size = sizeof(double) * n;*/ |
594
|
|
|
|
|
|
|
|
595
|
10
|
|
|
|
|
|
s = svd_doubleArray(jsq, TRUE, "ritvec: s"); |
596
|
10
|
|
|
|
|
|
xv2 = svd_doubleArray(n, FALSE, "ritvec: xv2"); |
597
|
|
|
|
|
|
|
|
598
|
|
|
|
|
|
|
/* initialize s to an identity matrix */ |
599
|
70
|
100
|
|
|
|
|
for (i = 0; i < jsq; i+= (js+1)) s[i] = 1.0; |
600
|
10
|
|
|
|
|
|
svd_dcopy(js, alf, 1, w1, -1); |
601
|
10
|
|
|
|
|
|
svd_dcopy(steps, &bet[1], 1, &w2[1], -1); |
602
|
|
|
|
|
|
|
|
603
|
|
|
|
|
|
|
/* on return from imtql2(), w1 contains eigenvalues in ascending |
604
|
|
|
|
|
|
|
* order and s contains the corresponding eigenvectors */ |
605
|
10
|
|
|
|
|
|
imtql2(js, js, w1, w2, s); |
606
|
|
|
|
|
|
|
|
607
|
|
|
|
|
|
|
/*fwrite((char *)&n, sizeof(n), 1, fp_out2); |
608
|
|
|
|
|
|
|
fwrite((char *)&js, sizeof(js), 1, fp_out2); |
609
|
|
|
|
|
|
|
fwrite((char *)&kappa, sizeof(kappa), 1, fp_out2);*/ |
610
|
|
|
|
|
|
|
/*id = 0;*/ |
611
|
10
|
|
|
|
|
|
nsig = 0; |
612
|
|
|
|
|
|
|
|
613
|
10
|
50
|
|
|
|
|
if (ierr) { |
614
|
0
|
|
|
|
|
|
R->d = 0; |
615
|
|
|
|
|
|
|
} else { |
616
|
10
|
|
|
|
|
|
x = 0; |
617
|
10
|
|
|
|
|
|
id2 = jsq - js; |
618
|
70
|
100
|
|
|
|
|
for (k = 0; k < js; k++) { |
619
|
60
|
|
|
|
|
|
tmp = id2; |
620
|
60
|
50
|
|
|
|
|
if (bnd[k] <= kappa * fabs(ritz[k]) && k > js-neig-1) { |
|
|
50
|
|
|
|
|
|
621
|
60
|
100
|
|
|
|
|
if (--x < 0) x = R->d - 1; |
622
|
60
|
|
|
|
|
|
w1 = R->Vt->value[x]; |
623
|
480
|
100
|
|
|
|
|
for (i = 0; i < n; i++) w1[i] = 0.0; |
624
|
420
|
100
|
|
|
|
|
for (i = 0; i < js; i++) { |
625
|
360
|
|
|
|
|
|
store(n, RETRQ, i, w2); |
626
|
360
|
|
|
|
|
|
svd_daxpy(n, s[tmp], w2, 1, w1, 1); |
627
|
360
|
|
|
|
|
|
tmp -= js; |
628
|
|
|
|
|
|
|
} |
629
|
|
|
|
|
|
|
/*fwrite((char *)w1, size, 1, fp_out2);*/ |
630
|
|
|
|
|
|
|
|
631
|
|
|
|
|
|
|
/* store the w1 vector row-wise in array xv1; |
632
|
|
|
|
|
|
|
* size of xv1 is (steps+1) * (nrow+ncol) elements |
633
|
|
|
|
|
|
|
* and each vector, even though only ncol __SVDLIBC_LONG, |
634
|
|
|
|
|
|
|
* will have (nrow+ncol) elements in xv1. |
635
|
|
|
|
|
|
|
* It is as if xv1 is a 2-d array (steps+1) by |
636
|
|
|
|
|
|
|
* (nrow+ncol) and each vector occupies a row */ |
637
|
|
|
|
|
|
|
|
638
|
|
|
|
|
|
|
/* j is the index in the R arrays, which are sorted by high to low |
639
|
|
|
|
|
|
|
singular values. */ |
640
|
|
|
|
|
|
|
|
641
|
|
|
|
|
|
|
/*for (i = 0; i < n; i++) R->Vt->value[x]xv1[id++] = w1[i];*/ |
642
|
|
|
|
|
|
|
/*id += nrow;*/ |
643
|
60
|
|
|
|
|
|
nsig++; |
644
|
|
|
|
|
|
|
} |
645
|
60
|
|
|
|
|
|
id2++; |
646
|
|
|
|
|
|
|
} |
647
|
10
|
50
|
|
|
|
|
SAFE_FREE(s); |
648
|
|
|
|
|
|
|
|
649
|
|
|
|
|
|
|
/* Rotate the singular vectors and values. */ |
650
|
|
|
|
|
|
|
/* x is now the location of the highest singular value. */ |
651
|
10
|
|
|
|
|
|
rotateArray(R->Vt->value[0], R->Vt->rows * R->Vt->cols, |
652
|
10
|
|
|
|
|
|
x * R->Vt->cols); |
653
|
10
|
|
|
|
|
|
R->d = svd_imin(R->d, nsig); |
654
|
67
|
100
|
|
|
|
|
for (x = 0; x < R->d; x++) { |
655
|
|
|
|
|
|
|
/* multiply by matrix B first */ |
656
|
57
|
|
|
|
|
|
svd_opb(A, R->Vt->value[x], xv2, OPBTemp); |
657
|
57
|
|
|
|
|
|
tmp0 = svd_ddot(n, R->Vt->value[x], 1, xv2, 1); |
658
|
57
|
|
|
|
|
|
svd_daxpy(n, -tmp0, R->Vt->value[x], 1, xv2, 1); |
659
|
57
|
|
|
|
|
|
tmp0 = sqrt(tmp0); |
660
|
57
|
|
|
|
|
|
xnorm = sqrt(svd_ddot(n, xv2, 1, xv2, 1)); |
661
|
|
|
|
|
|
|
|
662
|
|
|
|
|
|
|
/* multiply by matrix A to get (scaled) left s-vector */ |
663
|
57
|
|
|
|
|
|
svd_opa(A, R->Vt->value[x], R->Ut->value[x]); |
664
|
57
|
|
|
|
|
|
tmp1 = 1.0 / tmp0; |
665
|
57
|
|
|
|
|
|
svd_dscal(A->rows, tmp1, R->Ut->value[x], 1); |
666
|
57
|
|
|
|
|
|
xnorm *= tmp1; |
667
|
57
|
|
|
|
|
|
bnd[i] = xnorm; |
668
|
57
|
|
|
|
|
|
R->S[x] = tmp0; |
669
|
|
|
|
|
|
|
} |
670
|
|
|
|
|
|
|
} |
671
|
|
|
|
|
|
|
|
672
|
10
|
50
|
|
|
|
|
SAFE_FREE(s); |
673
|
10
|
50
|
|
|
|
|
SAFE_FREE(xv2); |
674
|
10
|
|
|
|
|
|
return nsig; |
675
|
|
|
|
|
|
|
} |
676
|
|
|
|
|
|
|
|
677
|
|
|
|
|
|
|
/*********************************************************************** |
678
|
|
|
|
|
|
|
* * |
679
|
|
|
|
|
|
|
* lanso() * |
680
|
|
|
|
|
|
|
* * |
681
|
|
|
|
|
|
|
***********************************************************************/ |
682
|
|
|
|
|
|
|
/*********************************************************************** |
683
|
|
|
|
|
|
|
|
684
|
|
|
|
|
|
|
Description |
685
|
|
|
|
|
|
|
----------- |
686
|
|
|
|
|
|
|
|
687
|
|
|
|
|
|
|
Function determines when the restart of the Lanczos algorithm should |
688
|
|
|
|
|
|
|
occur and when it should terminate. |
689
|
|
|
|
|
|
|
|
690
|
|
|
|
|
|
|
Arguments |
691
|
|
|
|
|
|
|
--------- |
692
|
|
|
|
|
|
|
|
693
|
|
|
|
|
|
|
(input) |
694
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
695
|
|
|
|
|
|
|
iterations upper limit of desired number of lanczos steps |
696
|
|
|
|
|
|
|
dimensions upper limit of desired number of eigenpairs |
697
|
|
|
|
|
|
|
endl left end of interval containing unwanted eigenvalues |
698
|
|
|
|
|
|
|
endr right end of interval containing unwanted eigenvalues |
699
|
|
|
|
|
|
|
ritz array to hold the ritz values |
700
|
|
|
|
|
|
|
bnd array to hold the error bounds |
701
|
|
|
|
|
|
|
wptr array of pointers that point to work space: |
702
|
|
|
|
|
|
|
wptr[0]-wptr[5] six vectors of length n |
703
|
|
|
|
|
|
|
wptr[6] array to hold diagonal of the tridiagonal matrix T |
704
|
|
|
|
|
|
|
wptr[9] array to hold off-diagonal of T |
705
|
|
|
|
|
|
|
wptr[7] orthogonality estimate of Lanczos vectors at |
706
|
|
|
|
|
|
|
step j |
707
|
|
|
|
|
|
|
wptr[8] orthogonality estimate of Lanczos vectors at |
708
|
|
|
|
|
|
|
step j-1 |
709
|
|
|
|
|
|
|
|
710
|
|
|
|
|
|
|
(output) |
711
|
|
|
|
|
|
|
j number of Lanczos steps actually taken |
712
|
|
|
|
|
|
|
neig number of ritz values stabilized |
713
|
|
|
|
|
|
|
ritz array to hold the ritz values |
714
|
|
|
|
|
|
|
bnd array to hold the error bounds |
715
|
|
|
|
|
|
|
ierr (globally declared) error flag |
716
|
|
|
|
|
|
|
ierr = 8192 if stpone() fails to find a starting vector |
717
|
|
|
|
|
|
|
ierr = k if convergence did not occur for k-th eigenvalue |
718
|
|
|
|
|
|
|
in imtqlb() |
719
|
|
|
|
|
|
|
ierr = 0 otherwise |
720
|
|
|
|
|
|
|
|
721
|
|
|
|
|
|
|
|
722
|
|
|
|
|
|
|
Functions used |
723
|
|
|
|
|
|
|
-------------- |
724
|
|
|
|
|
|
|
|
725
|
|
|
|
|
|
|
LAS stpone, error_bound, lanczos_step |
726
|
|
|
|
|
|
|
MISC svd_dsort2 |
727
|
|
|
|
|
|
|
UTILITY svd_imin, svd_imax |
728
|
|
|
|
|
|
|
|
729
|
|
|
|
|
|
|
***********************************************************************/ |
730
|
|
|
|
|
|
|
|
731
|
10
|
|
|
|
|
|
int lanso(SMat A, __SVDLIBC_LONG iterations, __SVDLIBC_LONG dimensions, double endl, |
732
|
|
|
|
|
|
|
double endr, double *ritz, double *bnd, double *wptr[], |
733
|
|
|
|
|
|
|
__SVDLIBC_LONG *neigp, __SVDLIBC_LONG n) { |
734
|
|
|
|
|
|
|
double *alf, *eta, *oldeta, *bet, *wrk, rnm, tol; |
735
|
10
|
|
|
|
|
|
__SVDLIBC_LONG ll, first, last, ENOUGH, id2, id3, i, l, neig, j = 0, intro = 0; |
736
|
|
|
|
|
|
|
|
737
|
10
|
|
|
|
|
|
alf = wptr[6]; |
738
|
10
|
|
|
|
|
|
eta = wptr[7]; |
739
|
10
|
|
|
|
|
|
oldeta = wptr[8]; |
740
|
10
|
|
|
|
|
|
bet = wptr[9]; |
741
|
10
|
|
|
|
|
|
wrk = wptr[5]; |
742
|
|
|
|
|
|
|
|
743
|
|
|
|
|
|
|
/* take the first step */ |
744
|
10
|
|
|
|
|
|
stpone(A, wptr, &rnm, &tol, n); |
745
|
10
|
50
|
|
|
|
|
if (!rnm || ierr) return 0; |
|
|
50
|
|
|
|
|
|
746
|
10
|
|
|
|
|
|
eta[0] = eps1; |
747
|
10
|
|
|
|
|
|
oldeta[0] = eps1; |
748
|
10
|
|
|
|
|
|
ll = 0; |
749
|
10
|
|
|
|
|
|
first = 1; |
750
|
10
|
|
|
|
|
|
last = svd_imin(dimensions + svd_imax(8, dimensions), iterations); |
751
|
10
|
|
|
|
|
|
ENOUGH = FALSE; |
752
|
|
|
|
|
|
|
/*id1 = 0;*/ |
753
|
20
|
100
|
|
|
|
|
while (/*id1 < dimensions && */!ENOUGH) { |
754
|
10
|
50
|
|
|
|
|
if (rnm <= tol) rnm = 0.0; |
755
|
|
|
|
|
|
|
|
756
|
|
|
|
|
|
|
/* the actual lanczos loop */ |
757
|
10
|
|
|
|
|
|
j = lanczos_step(A, first, last, wptr, alf, eta, oldeta, bet, &ll, |
758
|
|
|
|
|
|
|
&ENOUGH, &rnm, &tol, n); |
759
|
10
|
50
|
|
|
|
|
if (ENOUGH) j = j - 1; |
760
|
10
|
|
|
|
|
|
else j = last - 1; |
761
|
10
|
|
|
|
|
|
first = j + 1; |
762
|
10
|
|
|
|
|
|
bet[j+1] = rnm; |
763
|
|
|
|
|
|
|
|
764
|
|
|
|
|
|
|
/* analyze T */ |
765
|
10
|
|
|
|
|
|
l = 0; |
766
|
20
|
50
|
|
|
|
|
for (id2 = 0; id2 < j; id2++) { |
767
|
20
|
100
|
|
|
|
|
if (l > j) break; |
768
|
60
|
100
|
|
|
|
|
for (i = l; i <= j; i++) if (!bet[i+1]) break; |
|
|
50
|
|
|
|
|
|
769
|
10
|
50
|
|
|
|
|
if (i > j) i = j; |
770
|
|
|
|
|
|
|
|
771
|
|
|
|
|
|
|
/* now i is at the end of an unreduced submatrix */ |
772
|
10
|
|
|
|
|
|
svd_dcopy(i-l+1, &alf[l], 1, &ritz[l], -1); |
773
|
10
|
|
|
|
|
|
svd_dcopy(i-l, &bet[l+1], 1, &wrk[l+1], -1); |
774
|
|
|
|
|
|
|
|
775
|
10
|
|
|
|
|
|
imtqlb(i-l+1, &ritz[l], &wrk[l], &bnd[l]); |
776
|
|
|
|
|
|
|
|
777
|
10
|
50
|
|
|
|
|
if (ierr) { |
778
|
0
|
|
|
|
|
|
svd_error("svdLAS2: imtqlb failed to converge (ierr = %ld)\n", ierr); |
779
|
0
|
|
|
|
|
|
svd_error(" l = %ld i = %ld\n", l, i); |
780
|
0
|
0
|
|
|
|
|
for (id3 = l; id3 <= i; id3++) |
781
|
0
|
|
|
|
|
|
svd_error(" %ld %lg %lg %lg\n", |
782
|
0
|
|
|
|
|
|
id3, ritz[id3], wrk[id3], bnd[id3]); |
783
|
|
|
|
|
|
|
} |
784
|
70
|
100
|
|
|
|
|
for (id3 = l; id3 <= i; id3++) |
785
|
60
|
|
|
|
|
|
bnd[id3] = rnm * fabs(bnd[id3]); |
786
|
10
|
|
|
|
|
|
l = i + 1; |
787
|
|
|
|
|
|
|
} |
788
|
|
|
|
|
|
|
|
789
|
|
|
|
|
|
|
/* sort eigenvalues into increasing order */ |
790
|
10
|
|
|
|
|
|
svd_dsort2((j+1) / 2, j + 1, ritz, bnd); |
791
|
|
|
|
|
|
|
|
792
|
|
|
|
|
|
|
/* for (i = 0; i < iterations; i++) |
793
|
|
|
|
|
|
|
printf("%f ", ritz[i]); |
794
|
|
|
|
|
|
|
printf("\n"); */ |
795
|
|
|
|
|
|
|
|
796
|
|
|
|
|
|
|
/* massage error bounds for very close ritz values */ |
797
|
10
|
|
|
|
|
|
neig = error_bound(&ENOUGH, endl, endr, ritz, bnd, j, tol); |
798
|
10
|
|
|
|
|
|
*neigp = neig; |
799
|
|
|
|
|
|
|
|
800
|
|
|
|
|
|
|
/* should we stop? */ |
801
|
10
|
50
|
|
|
|
|
if (neig < dimensions) { |
802
|
0
|
0
|
|
|
|
|
if (!neig) { |
803
|
0
|
|
|
|
|
|
last = first + 9; |
804
|
0
|
|
|
|
|
|
intro = first; |
805
|
0
|
|
|
|
|
|
} else last = first + svd_imax(3, 1 + ((j - intro) * (dimensions-neig)) / |
806
|
|
|
|
|
|
|
neig); |
807
|
0
|
|
|
|
|
|
last = svd_imin(last, iterations); |
808
|
10
|
|
|
|
|
|
} else ENOUGH = TRUE; |
809
|
10
|
50
|
|
|
|
|
ENOUGH = ENOUGH || first >= iterations; |
|
|
0
|
|
|
|
|
|
810
|
|
|
|
|
|
|
/* id1++; */ |
811
|
|
|
|
|
|
|
/* printf("id1=%d dimen=%d first=%d\n", id1, dimensions, first); */ |
812
|
|
|
|
|
|
|
} |
813
|
10
|
|
|
|
|
|
store(n, STORQ, j, wptr[1]); |
814
|
10
|
|
|
|
|
|
return j; |
815
|
|
|
|
|
|
|
} |
816
|
|
|
|
|
|
|
|
817
|
|
|
|
|
|
|
|
818
|
|
|
|
|
|
|
/*********************************************************************** |
819
|
|
|
|
|
|
|
* * |
820
|
|
|
|
|
|
|
* lanczos_step() * |
821
|
|
|
|
|
|
|
* * |
822
|
|
|
|
|
|
|
***********************************************************************/ |
823
|
|
|
|
|
|
|
/*********************************************************************** |
824
|
|
|
|
|
|
|
|
825
|
|
|
|
|
|
|
Description |
826
|
|
|
|
|
|
|
----------- |
827
|
|
|
|
|
|
|
|
828
|
|
|
|
|
|
|
Function embodies a single Lanczos step |
829
|
|
|
|
|
|
|
|
830
|
|
|
|
|
|
|
Arguments |
831
|
|
|
|
|
|
|
--------- |
832
|
|
|
|
|
|
|
|
833
|
|
|
|
|
|
|
(input) |
834
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
835
|
|
|
|
|
|
|
first start of index through loop |
836
|
|
|
|
|
|
|
last end of index through loop |
837
|
|
|
|
|
|
|
wptr array of pointers pointing to work space |
838
|
|
|
|
|
|
|
alf array to hold diagonal of the tridiagonal matrix T |
839
|
|
|
|
|
|
|
eta orthogonality estimate of Lanczos vectors at step j |
840
|
|
|
|
|
|
|
oldeta orthogonality estimate of Lanczos vectors at step j-1 |
841
|
|
|
|
|
|
|
bet array to hold off-diagonal of T |
842
|
|
|
|
|
|
|
ll number of intitial Lanczos vectors in local orthog. |
843
|
|
|
|
|
|
|
(has value of 0, 1 or 2) |
844
|
|
|
|
|
|
|
enough stop flag |
845
|
|
|
|
|
|
|
|
846
|
|
|
|
|
|
|
Functions used |
847
|
|
|
|
|
|
|
-------------- |
848
|
|
|
|
|
|
|
|
849
|
|
|
|
|
|
|
BLAS svd_ddot, svd_dscal, svd_daxpy, svd_datx, svd_dcopy |
850
|
|
|
|
|
|
|
USER store |
851
|
|
|
|
|
|
|
LAS purge, ortbnd, startv |
852
|
|
|
|
|
|
|
UTILITY svd_imin, svd_imax |
853
|
|
|
|
|
|
|
|
854
|
|
|
|
|
|
|
***********************************************************************/ |
855
|
|
|
|
|
|
|
|
856
|
10
|
|
|
|
|
|
__SVDLIBC_LONG lanczos_step(SMat A, __SVDLIBC_LONG first, __SVDLIBC_LONG last, double *wptr[], |
857
|
|
|
|
|
|
|
double *alf, double *eta, double *oldeta, |
858
|
|
|
|
|
|
|
double *bet, __SVDLIBC_LONG *ll, __SVDLIBC_LONG *enough, double *rnmp, |
859
|
|
|
|
|
|
|
double *tolp, __SVDLIBC_LONG n) { |
860
|
10
|
|
|
|
|
|
double t, *mid, rnm = *rnmp, tol = *tolp, anorm; |
861
|
|
|
|
|
|
|
__SVDLIBC_LONG i, j; |
862
|
|
|
|
|
|
|
|
863
|
60
|
100
|
|
|
|
|
for (j=first; j
|
864
|
50
|
|
|
|
|
|
mid = wptr[2]; |
865
|
50
|
|
|
|
|
|
wptr[2] = wptr[1]; |
866
|
50
|
|
|
|
|
|
wptr[1] = mid; |
867
|
50
|
|
|
|
|
|
mid = wptr[3]; |
868
|
50
|
|
|
|
|
|
wptr[3] = wptr[4]; |
869
|
50
|
|
|
|
|
|
wptr[4] = mid; |
870
|
|
|
|
|
|
|
|
871
|
50
|
|
|
|
|
|
store(n, STORQ, j-1, wptr[2]); |
872
|
50
|
100
|
|
|
|
|
if (j-1 < MAXLL) store(n, STORP, j-1, wptr[4]); |
873
|
50
|
|
|
|
|
|
bet[j] = rnm; |
874
|
|
|
|
|
|
|
|
875
|
|
|
|
|
|
|
/* restart if invariant subspace is found */ |
876
|
50
|
50
|
|
|
|
|
if (!bet[j]) { |
877
|
0
|
|
|
|
|
|
rnm = startv(A, wptr, j, n); |
878
|
0
|
0
|
|
|
|
|
if (ierr) return j; |
879
|
0
|
0
|
|
|
|
|
if (!rnm) *enough = TRUE; |
880
|
|
|
|
|
|
|
} |
881
|
50
|
50
|
|
|
|
|
if (*enough) { |
882
|
|
|
|
|
|
|
/* added by Doug... */ |
883
|
|
|
|
|
|
|
/* These lines fix a bug that occurs with low-rank matrices */ |
884
|
0
|
|
|
|
|
|
mid = wptr[2]; |
885
|
0
|
|
|
|
|
|
wptr[2] = wptr[1]; |
886
|
0
|
|
|
|
|
|
wptr[1] = mid; |
887
|
|
|
|
|
|
|
/* ...added by Doug */ |
888
|
0
|
|
|
|
|
|
break; |
889
|
|
|
|
|
|
|
} |
890
|
|
|
|
|
|
|
|
891
|
|
|
|
|
|
|
/* take a lanczos step */ |
892
|
50
|
|
|
|
|
|
t = 1.0 / rnm; |
893
|
50
|
|
|
|
|
|
svd_datx(n, t, wptr[0], 1, wptr[1], 1); |
894
|
50
|
|
|
|
|
|
svd_dscal(n, t, wptr[3], 1); |
895
|
50
|
|
|
|
|
|
svd_opb(A, wptr[3], wptr[0], OPBTemp); |
896
|
50
|
|
|
|
|
|
svd_daxpy(n, -rnm, wptr[2], 1, wptr[0], 1); |
897
|
50
|
|
|
|
|
|
alf[j] = svd_ddot(n, wptr[0], 1, wptr[3], 1); |
898
|
50
|
|
|
|
|
|
svd_daxpy(n, -alf[j], wptr[1], 1, wptr[0], 1); |
899
|
|
|
|
|
|
|
|
900
|
|
|
|
|
|
|
/* orthogonalize against initial lanczos vectors */ |
901
|
50
|
100
|
|
|
|
|
if (j <= MAXLL && (fabs(alf[j-1]) > 4.0 * fabs(alf[j]))) |
|
|
50
|
|
|
|
|
|
902
|
0
|
|
|
|
|
|
*ll = j; |
903
|
50
|
50
|
|
|
|
|
for (i=0; i < svd_imin(*ll, j-1); i++) { |
904
|
0
|
|
|
|
|
|
store(n, RETRP, i, wptr[5]); |
905
|
0
|
|
|
|
|
|
t = svd_ddot(n, wptr[5], 1, wptr[0], 1); |
906
|
0
|
|
|
|
|
|
store(n, RETRQ, i, wptr[5]); |
907
|
0
|
|
|
|
|
|
svd_daxpy(n, -t, wptr[5], 1, wptr[0], 1); |
908
|
0
|
|
|
|
|
|
eta[i] = eps1; |
909
|
0
|
|
|
|
|
|
oldeta[i] = eps1; |
910
|
|
|
|
|
|
|
} |
911
|
|
|
|
|
|
|
|
912
|
|
|
|
|
|
|
/* extended local reorthogonalization */ |
913
|
50
|
|
|
|
|
|
t = svd_ddot(n, wptr[0], 1, wptr[4], 1); |
914
|
50
|
|
|
|
|
|
svd_daxpy(n, -t, wptr[2], 1, wptr[0], 1); |
915
|
50
|
50
|
|
|
|
|
if (bet[j] > 0.0) bet[j] = bet[j] + t; |
916
|
50
|
|
|
|
|
|
t = svd_ddot(n, wptr[0], 1, wptr[3], 1); |
917
|
50
|
|
|
|
|
|
svd_daxpy(n, -t, wptr[1], 1, wptr[0], 1); |
918
|
50
|
|
|
|
|
|
alf[j] = alf[j] + t; |
919
|
50
|
|
|
|
|
|
svd_dcopy(n, wptr[0], 1, wptr[4], 1); |
920
|
50
|
|
|
|
|
|
rnm = sqrt(svd_ddot(n, wptr[0], 1, wptr[4], 1)); |
921
|
50
|
|
|
|
|
|
anorm = bet[j] + fabs(alf[j]) + rnm; |
922
|
50
|
|
|
|
|
|
tol = reps * anorm; |
923
|
|
|
|
|
|
|
|
924
|
|
|
|
|
|
|
/* update the orthogonality bounds */ |
925
|
50
|
|
|
|
|
|
ortbnd(alf, eta, oldeta, bet, j, rnm); |
926
|
|
|
|
|
|
|
|
927
|
|
|
|
|
|
|
/* restore the orthogonality state when needed */ |
928
|
50
|
|
|
|
|
|
purge(n, *ll, wptr[0], wptr[1], wptr[4], wptr[3], wptr[5], eta, oldeta, |
929
|
|
|
|
|
|
|
j, &rnm, tol); |
930
|
50
|
100
|
|
|
|
|
if (rnm <= tol) rnm = 0.0; |
931
|
|
|
|
|
|
|
} |
932
|
10
|
|
|
|
|
|
*rnmp = rnm; |
933
|
10
|
|
|
|
|
|
*tolp = tol; |
934
|
10
|
|
|
|
|
|
return j; |
935
|
|
|
|
|
|
|
} |
936
|
|
|
|
|
|
|
|
937
|
|
|
|
|
|
|
/*********************************************************************** |
938
|
|
|
|
|
|
|
* * |
939
|
|
|
|
|
|
|
* ortbnd() * |
940
|
|
|
|
|
|
|
* * |
941
|
|
|
|
|
|
|
***********************************************************************/ |
942
|
|
|
|
|
|
|
/*********************************************************************** |
943
|
|
|
|
|
|
|
|
944
|
|
|
|
|
|
|
Description |
945
|
|
|
|
|
|
|
----------- |
946
|
|
|
|
|
|
|
|
947
|
|
|
|
|
|
|
Funtion updates the eta recurrence |
948
|
|
|
|
|
|
|
|
949
|
|
|
|
|
|
|
Arguments |
950
|
|
|
|
|
|
|
--------- |
951
|
|
|
|
|
|
|
|
952
|
|
|
|
|
|
|
(input) |
953
|
|
|
|
|
|
|
alf array to hold diagonal of the tridiagonal matrix T |
954
|
|
|
|
|
|
|
eta orthogonality estimate of Lanczos vectors at step j |
955
|
|
|
|
|
|
|
oldeta orthogonality estimate of Lanczos vectors at step j-1 |
956
|
|
|
|
|
|
|
bet array to hold off-diagonal of T |
957
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
958
|
|
|
|
|
|
|
j dimension of T |
959
|
|
|
|
|
|
|
rnm norm of the next residual vector |
960
|
|
|
|
|
|
|
eps1 roundoff estimate for dot product of two unit vectors |
961
|
|
|
|
|
|
|
|
962
|
|
|
|
|
|
|
(output) |
963
|
|
|
|
|
|
|
eta orthogonality estimate of Lanczos vectors at step j+1 |
964
|
|
|
|
|
|
|
oldeta orthogonality estimate of Lanczos vectors at step j |
965
|
|
|
|
|
|
|
|
966
|
|
|
|
|
|
|
|
967
|
|
|
|
|
|
|
Functions used |
968
|
|
|
|
|
|
|
-------------- |
969
|
|
|
|
|
|
|
|
970
|
|
|
|
|
|
|
BLAS svd_dswap |
971
|
|
|
|
|
|
|
|
972
|
|
|
|
|
|
|
***********************************************************************/ |
973
|
|
|
|
|
|
|
|
974
|
50
|
|
|
|
|
|
void ortbnd(double *alf, double *eta, double *oldeta, double *bet, __SVDLIBC_LONG step, |
975
|
|
|
|
|
|
|
double rnm) { |
976
|
|
|
|
|
|
|
__SVDLIBC_LONG i; |
977
|
50
|
50
|
|
|
|
|
if (step < 1) return; |
978
|
50
|
50
|
|
|
|
|
if (rnm) { |
979
|
50
|
100
|
|
|
|
|
if (step > 1) { |
980
|
40
|
|
|
|
|
|
oldeta[0] = (bet[1] * eta[1] + (alf[0]-alf[step]) * eta[0] - |
981
|
80
|
|
|
|
|
|
bet[step] * oldeta[0]) / rnm + eps1; |
982
|
|
|
|
|
|
|
} |
983
|
110
|
100
|
|
|
|
|
for (i=1; i<=step-2; i++) |
984
|
180
|
|
|
|
|
|
oldeta[i] = (bet[i+1] * eta[i+1] + (alf[i]-alf[step]) * eta[i] + |
985
|
120
|
|
|
|
|
|
bet[i] * eta[i-1] - bet[step] * oldeta[i])/rnm + eps1; |
986
|
|
|
|
|
|
|
} |
987
|
50
|
|
|
|
|
|
oldeta[step-1] = eps1; |
988
|
50
|
|
|
|
|
|
svd_dswap(step, oldeta, 1, eta, 1); |
989
|
50
|
|
|
|
|
|
eta[step] = eps1; |
990
|
50
|
|
|
|
|
|
return; |
991
|
|
|
|
|
|
|
} |
992
|
|
|
|
|
|
|
|
993
|
|
|
|
|
|
|
/*********************************************************************** |
994
|
|
|
|
|
|
|
* * |
995
|
|
|
|
|
|
|
* purge() * |
996
|
|
|
|
|
|
|
* * |
997
|
|
|
|
|
|
|
***********************************************************************/ |
998
|
|
|
|
|
|
|
/*********************************************************************** |
999
|
|
|
|
|
|
|
|
1000
|
|
|
|
|
|
|
Description |
1001
|
|
|
|
|
|
|
----------- |
1002
|
|
|
|
|
|
|
|
1003
|
|
|
|
|
|
|
Function examines the state of orthogonality between the new Lanczos |
1004
|
|
|
|
|
|
|
vector and the previous ones to decide whether re-orthogonalization |
1005
|
|
|
|
|
|
|
should be performed |
1006
|
|
|
|
|
|
|
|
1007
|
|
|
|
|
|
|
|
1008
|
|
|
|
|
|
|
Arguments |
1009
|
|
|
|
|
|
|
--------- |
1010
|
|
|
|
|
|
|
|
1011
|
|
|
|
|
|
|
(input) |
1012
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
1013
|
|
|
|
|
|
|
ll number of intitial Lanczos vectors in local orthog. |
1014
|
|
|
|
|
|
|
r residual vector to become next Lanczos vector |
1015
|
|
|
|
|
|
|
q current Lanczos vector |
1016
|
|
|
|
|
|
|
ra previous Lanczos vector |
1017
|
|
|
|
|
|
|
qa previous Lanczos vector |
1018
|
|
|
|
|
|
|
wrk temporary vector to hold the previous Lanczos vector |
1019
|
|
|
|
|
|
|
eta state of orthogonality between r and prev. Lanczos vectors |
1020
|
|
|
|
|
|
|
oldeta state of orthogonality between q and prev. Lanczos vectors |
1021
|
|
|
|
|
|
|
j current Lanczos step |
1022
|
|
|
|
|
|
|
|
1023
|
|
|
|
|
|
|
(output) |
1024
|
|
|
|
|
|
|
r residual vector orthogonalized against previous Lanczos |
1025
|
|
|
|
|
|
|
vectors |
1026
|
|
|
|
|
|
|
q current Lanczos vector orthogonalized against previous ones |
1027
|
|
|
|
|
|
|
|
1028
|
|
|
|
|
|
|
|
1029
|
|
|
|
|
|
|
Functions used |
1030
|
|
|
|
|
|
|
-------------- |
1031
|
|
|
|
|
|
|
|
1032
|
|
|
|
|
|
|
BLAS svd_daxpy, svd_dcopy, svd_idamax, svd_ddot |
1033
|
|
|
|
|
|
|
USER store |
1034
|
|
|
|
|
|
|
|
1035
|
|
|
|
|
|
|
***********************************************************************/ |
1036
|
|
|
|
|
|
|
|
1037
|
50
|
|
|
|
|
|
void purge(__SVDLIBC_LONG n, __SVDLIBC_LONG ll, double *r, double *q, double *ra, |
1038
|
|
|
|
|
|
|
double *qa, double *wrk, double *eta, double *oldeta, __SVDLIBC_LONG step, |
1039
|
|
|
|
|
|
|
double *rnmp, double tol) { |
1040
|
50
|
|
|
|
|
|
double t, tq, tr, reps1, rnm = *rnmp; |
1041
|
|
|
|
|
|
|
__SVDLIBC_LONG k, iteration, flag, i; |
1042
|
|
|
|
|
|
|
|
1043
|
50
|
100
|
|
|
|
|
if (step < ll+2) return; |
1044
|
|
|
|
|
|
|
|
1045
|
40
|
|
|
|
|
|
k = svd_idamax(step - (ll+1), &eta[ll], 1) + ll; |
1046
|
40
|
100
|
|
|
|
|
if (fabs(eta[k]) > reps) { |
1047
|
10
|
|
|
|
|
|
reps1 = eps1 / reps; |
1048
|
10
|
|
|
|
|
|
iteration = 0; |
1049
|
10
|
|
|
|
|
|
flag = TRUE; |
1050
|
30
|
100
|
|
|
|
|
while (iteration < 2 && flag) { |
|
|
50
|
|
|
|
|
|
1051
|
20
|
50
|
|
|
|
|
if (rnm > tol) { |
1052
|
|
|
|
|
|
|
|
1053
|
|
|
|
|
|
|
/* bring in a lanczos vector t and orthogonalize both |
1054
|
|
|
|
|
|
|
* r and q against it */ |
1055
|
0
|
|
|
|
|
|
tq = 0.0; |
1056
|
0
|
|
|
|
|
|
tr = 0.0; |
1057
|
0
|
0
|
|
|
|
|
for (i = ll; i < step; i++) { |
1058
|
0
|
|
|
|
|
|
store(n, RETRQ, i, wrk); |
1059
|
0
|
|
|
|
|
|
t = -svd_ddot(n, qa, 1, wrk, 1); |
1060
|
0
|
|
|
|
|
|
tq += fabs(t); |
1061
|
0
|
|
|
|
|
|
svd_daxpy(n, t, wrk, 1, q, 1); |
1062
|
0
|
|
|
|
|
|
t = -svd_ddot(n, ra, 1, wrk, 1); |
1063
|
0
|
|
|
|
|
|
tr += fabs(t); |
1064
|
0
|
|
|
|
|
|
svd_daxpy(n, t, wrk, 1, r, 1); |
1065
|
|
|
|
|
|
|
} |
1066
|
0
|
|
|
|
|
|
svd_dcopy(n, q, 1, qa, 1); |
1067
|
0
|
|
|
|
|
|
t = -svd_ddot(n, r, 1, qa, 1); |
1068
|
0
|
|
|
|
|
|
tr += fabs(t); |
1069
|
0
|
|
|
|
|
|
svd_daxpy(n, t, q, 1, r, 1); |
1070
|
0
|
|
|
|
|
|
svd_dcopy(n, r, 1, ra, 1); |
1071
|
0
|
|
|
|
|
|
rnm = sqrt(svd_ddot(n, ra, 1, r, 1)); |
1072
|
0
|
0
|
|
|
|
|
if (tq <= reps1 && tr <= reps1 * rnm) flag = FALSE; |
|
|
0
|
|
|
|
|
|
1073
|
|
|
|
|
|
|
} |
1074
|
20
|
|
|
|
|
|
iteration++; |
1075
|
|
|
|
|
|
|
} |
1076
|
70
|
100
|
|
|
|
|
for (i = ll; i <= step; i++) { |
1077
|
60
|
|
|
|
|
|
eta[i] = eps1; |
1078
|
60
|
|
|
|
|
|
oldeta[i] = eps1; |
1079
|
|
|
|
|
|
|
} |
1080
|
|
|
|
|
|
|
} |
1081
|
40
|
|
|
|
|
|
*rnmp = rnm; |
1082
|
40
|
|
|
|
|
|
return; |
1083
|
|
|
|
|
|
|
} |
1084
|
|
|
|
|
|
|
|
1085
|
|
|
|
|
|
|
|
1086
|
|
|
|
|
|
|
/*********************************************************************** |
1087
|
|
|
|
|
|
|
* * |
1088
|
|
|
|
|
|
|
* stpone() * |
1089
|
|
|
|
|
|
|
* * |
1090
|
|
|
|
|
|
|
***********************************************************************/ |
1091
|
|
|
|
|
|
|
/*********************************************************************** |
1092
|
|
|
|
|
|
|
|
1093
|
|
|
|
|
|
|
Description |
1094
|
|
|
|
|
|
|
----------- |
1095
|
|
|
|
|
|
|
|
1096
|
|
|
|
|
|
|
Function performs the first step of the Lanczos algorithm. It also |
1097
|
|
|
|
|
|
|
does a step of extended local re-orthogonalization. |
1098
|
|
|
|
|
|
|
|
1099
|
|
|
|
|
|
|
Arguments |
1100
|
|
|
|
|
|
|
--------- |
1101
|
|
|
|
|
|
|
|
1102
|
|
|
|
|
|
|
(input) |
1103
|
|
|
|
|
|
|
n dimension of the eigenproblem for matrix B |
1104
|
|
|
|
|
|
|
|
1105
|
|
|
|
|
|
|
(output) |
1106
|
|
|
|
|
|
|
ierr error flag |
1107
|
|
|
|
|
|
|
wptr array of pointers that point to work space that contains |
1108
|
|
|
|
|
|
|
wptr[0] r[j] |
1109
|
|
|
|
|
|
|
wptr[1] q[j] |
1110
|
|
|
|
|
|
|
wptr[2] q[j-1] |
1111
|
|
|
|
|
|
|
wptr[3] p |
1112
|
|
|
|
|
|
|
wptr[4] p[j-1] |
1113
|
|
|
|
|
|
|
wptr[6] diagonal elements of matrix T |
1114
|
|
|
|
|
|
|
|
1115
|
|
|
|
|
|
|
|
1116
|
|
|
|
|
|
|
Functions used |
1117
|
|
|
|
|
|
|
-------------- |
1118
|
|
|
|
|
|
|
|
1119
|
|
|
|
|
|
|
BLAS svd_daxpy, svd_datx, svd_dcopy, svd_ddot, svd_dscal |
1120
|
|
|
|
|
|
|
USER store, opb |
1121
|
|
|
|
|
|
|
LAS startv |
1122
|
|
|
|
|
|
|
|
1123
|
|
|
|
|
|
|
***********************************************************************/ |
1124
|
|
|
|
|
|
|
|
1125
|
10
|
|
|
|
|
|
void stpone(SMat A, double *wrkptr[], double *rnmp, double *tolp, __SVDLIBC_LONG n) { |
1126
|
|
|
|
|
|
|
double t, *alf, rnm, anorm; |
1127
|
10
|
|
|
|
|
|
alf = wrkptr[6]; |
1128
|
|
|
|
|
|
|
|
1129
|
|
|
|
|
|
|
/* get initial vector; default is random */ |
1130
|
10
|
|
|
|
|
|
rnm = startv(A, wrkptr, 0, n); |
1131
|
10
|
50
|
|
|
|
|
if (rnm == 0.0 || ierr != 0) return; |
|
|
50
|
|
|
|
|
|
1132
|
|
|
|
|
|
|
|
1133
|
|
|
|
|
|
|
/* normalize starting vector */ |
1134
|
10
|
|
|
|
|
|
t = 1.0 / rnm; |
1135
|
10
|
|
|
|
|
|
svd_datx(n, t, wrkptr[0], 1, wrkptr[1], 1); |
1136
|
10
|
|
|
|
|
|
svd_dscal(n, t, wrkptr[3], 1); |
1137
|
|
|
|
|
|
|
|
1138
|
|
|
|
|
|
|
/* take the first step */ |
1139
|
10
|
|
|
|
|
|
svd_opb(A, wrkptr[3], wrkptr[0], OPBTemp); |
1140
|
10
|
|
|
|
|
|
alf[0] = svd_ddot(n, wrkptr[0], 1, wrkptr[3], 1); |
1141
|
10
|
|
|
|
|
|
svd_daxpy(n, -alf[0], wrkptr[1], 1, wrkptr[0], 1); |
1142
|
10
|
|
|
|
|
|
t = svd_ddot(n, wrkptr[0], 1, wrkptr[3], 1); |
1143
|
10
|
|
|
|
|
|
svd_daxpy(n, -t, wrkptr[1], 1, wrkptr[0], 1); |
1144
|
10
|
|
|
|
|
|
alf[0] += t; |
1145
|
10
|
|
|
|
|
|
svd_dcopy(n, wrkptr[0], 1, wrkptr[4], 1); |
1146
|
10
|
|
|
|
|
|
rnm = sqrt(svd_ddot(n, wrkptr[0], 1, wrkptr[4], 1)); |
1147
|
10
|
|
|
|
|
|
anorm = rnm + fabs(alf[0]); |
1148
|
10
|
|
|
|
|
|
*rnmp = rnm; |
1149
|
10
|
|
|
|
|
|
*tolp = reps * anorm; |
1150
|
|
|
|
|
|
|
|
1151
|
10
|
|
|
|
|
|
return; |
1152
|
|
|
|
|
|
|
} |
1153
|
|
|
|
|
|
|
|
1154
|
|
|
|
|
|
|
/*********************************************************************** |
1155
|
|
|
|
|
|
|
* * |
1156
|
|
|
|
|
|
|
* startv() * |
1157
|
|
|
|
|
|
|
* * |
1158
|
|
|
|
|
|
|
***********************************************************************/ |
1159
|
|
|
|
|
|
|
/*********************************************************************** |
1160
|
|
|
|
|
|
|
|
1161
|
|
|
|
|
|
|
Description |
1162
|
|
|
|
|
|
|
----------- |
1163
|
|
|
|
|
|
|
|
1164
|
|
|
|
|
|
|
Function delivers a starting vector in r and returns |r|; it returns |
1165
|
|
|
|
|
|
|
zero if the range is spanned, and ierr is non-zero if no starting |
1166
|
|
|
|
|
|
|
vector within range of operator can be found. |
1167
|
|
|
|
|
|
|
|
1168
|
|
|
|
|
|
|
Parameters |
1169
|
|
|
|
|
|
|
--------- |
1170
|
|
|
|
|
|
|
|
1171
|
|
|
|
|
|
|
(input) |
1172
|
|
|
|
|
|
|
n dimension of the eigenproblem matrix B |
1173
|
|
|
|
|
|
|
wptr array of pointers that point to work space |
1174
|
|
|
|
|
|
|
j starting index for a Lanczos run |
1175
|
|
|
|
|
|
|
eps machine epsilon (relative precision) |
1176
|
|
|
|
|
|
|
|
1177
|
|
|
|
|
|
|
(output) |
1178
|
|
|
|
|
|
|
wptr array of pointers that point to work space that contains |
1179
|
|
|
|
|
|
|
r[j], q[j], q[j-1], p[j], p[j-1] |
1180
|
|
|
|
|
|
|
ierr error flag (nonzero if no starting vector can be found) |
1181
|
|
|
|
|
|
|
|
1182
|
|
|
|
|
|
|
Functions used |
1183
|
|
|
|
|
|
|
-------------- |
1184
|
|
|
|
|
|
|
|
1185
|
|
|
|
|
|
|
BLAS svd_ddot, svd_dcopy, svd_daxpy |
1186
|
|
|
|
|
|
|
USER svd_opb, store |
1187
|
|
|
|
|
|
|
MISC random |
1188
|
|
|
|
|
|
|
|
1189
|
|
|
|
|
|
|
***********************************************************************/ |
1190
|
|
|
|
|
|
|
|
1191
|
10
|
|
|
|
|
|
double startv(SMat A, double *wptr[], __SVDLIBC_LONG step, __SVDLIBC_LONG n) { |
1192
|
|
|
|
|
|
|
double rnm2, *r, t; |
1193
|
|
|
|
|
|
|
__SVDLIBC_LONG irand; |
1194
|
|
|
|
|
|
|
__SVDLIBC_LONG id, i; |
1195
|
|
|
|
|
|
|
|
1196
|
|
|
|
|
|
|
/* get initial vector; default is random */ |
1197
|
10
|
|
|
|
|
|
rnm2 = svd_ddot(n, wptr[0], 1, wptr[0], 1); |
1198
|
10
|
|
|
|
|
|
irand = 918273 + step; |
1199
|
10
|
|
|
|
|
|
r = wptr[0]; |
1200
|
10
|
50
|
|
|
|
|
for (id = 0; id < 3; id++) { |
1201
|
10
|
50
|
|
|
|
|
if (id > 0 || step > 0 || rnm2 == 0) |
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
1202
|
80
|
100
|
|
|
|
|
for (i = 0; i < n; i++) r[i] = svd_random2(&irand); |
1203
|
10
|
|
|
|
|
|
svd_dcopy(n, wptr[0], 1, wptr[3], 1); |
1204
|
|
|
|
|
|
|
|
1205
|
|
|
|
|
|
|
/* apply operator to put r in range (essential if m singular) */ |
1206
|
10
|
|
|
|
|
|
svd_opb(A, wptr[3], wptr[0], OPBTemp); |
1207
|
10
|
|
|
|
|
|
svd_dcopy(n, wptr[0], 1, wptr[3], 1); |
1208
|
10
|
|
|
|
|
|
rnm2 = svd_ddot(n, wptr[0], 1, wptr[3], 1); |
1209
|
10
|
50
|
|
|
|
|
if (rnm2 > 0.0) break; |
1210
|
|
|
|
|
|
|
} |
1211
|
|
|
|
|
|
|
|
1212
|
|
|
|
|
|
|
/* fatal error */ |
1213
|
10
|
50
|
|
|
|
|
if (rnm2 <= 0.0) { |
1214
|
0
|
|
|
|
|
|
ierr = 8192; |
1215
|
0
|
|
|
|
|
|
return(-1); |
1216
|
|
|
|
|
|
|
} |
1217
|
10
|
50
|
|
|
|
|
if (step > 0) { |
1218
|
0
|
0
|
|
|
|
|
for (i = 0; i < step; i++) { |
1219
|
0
|
|
|
|
|
|
store(n, RETRQ, i, wptr[5]); |
1220
|
0
|
|
|
|
|
|
t = -svd_ddot(n, wptr[3], 1, wptr[5], 1); |
1221
|
0
|
|
|
|
|
|
svd_daxpy(n, t, wptr[5], 1, wptr[0], 1); |
1222
|
|
|
|
|
|
|
} |
1223
|
|
|
|
|
|
|
|
1224
|
|
|
|
|
|
|
/* make sure q[step] is orthogonal to q[step-1] */ |
1225
|
0
|
|
|
|
|
|
t = svd_ddot(n, wptr[4], 1, wptr[0], 1); |
1226
|
0
|
|
|
|
|
|
svd_daxpy(n, -t, wptr[2], 1, wptr[0], 1); |
1227
|
0
|
|
|
|
|
|
svd_dcopy(n, wptr[0], 1, wptr[3], 1); |
1228
|
0
|
|
|
|
|
|
t = svd_ddot(n, wptr[3], 1, wptr[0], 1); |
1229
|
0
|
0
|
|
|
|
|
if (t <= eps * rnm2) t = 0.0; |
1230
|
0
|
|
|
|
|
|
rnm2 = t; |
1231
|
|
|
|
|
|
|
} |
1232
|
10
|
|
|
|
|
|
return(sqrt(rnm2)); |
1233
|
|
|
|
|
|
|
} |
1234
|
|
|
|
|
|
|
|
1235
|
|
|
|
|
|
|
/*********************************************************************** |
1236
|
|
|
|
|
|
|
* * |
1237
|
|
|
|
|
|
|
* error_bound() * |
1238
|
|
|
|
|
|
|
* * |
1239
|
|
|
|
|
|
|
***********************************************************************/ |
1240
|
|
|
|
|
|
|
/*********************************************************************** |
1241
|
|
|
|
|
|
|
|
1242
|
|
|
|
|
|
|
Description |
1243
|
|
|
|
|
|
|
----------- |
1244
|
|
|
|
|
|
|
|
1245
|
|
|
|
|
|
|
Function massages error bounds for very close ritz values by placing |
1246
|
|
|
|
|
|
|
a gap between them. The error bounds are then refined to reflect |
1247
|
|
|
|
|
|
|
this. |
1248
|
|
|
|
|
|
|
|
1249
|
|
|
|
|
|
|
|
1250
|
|
|
|
|
|
|
Arguments |
1251
|
|
|
|
|
|
|
--------- |
1252
|
|
|
|
|
|
|
|
1253
|
|
|
|
|
|
|
(input) |
1254
|
|
|
|
|
|
|
endl left end of interval containing unwanted eigenvalues |
1255
|
|
|
|
|
|
|
endr right end of interval containing unwanted eigenvalues |
1256
|
|
|
|
|
|
|
ritz array to store the ritz values |
1257
|
|
|
|
|
|
|
bnd array to store the error bounds |
1258
|
|
|
|
|
|
|
enough stop flag |
1259
|
|
|
|
|
|
|
|
1260
|
|
|
|
|
|
|
|
1261
|
|
|
|
|
|
|
Functions used |
1262
|
|
|
|
|
|
|
-------------- |
1263
|
|
|
|
|
|
|
|
1264
|
|
|
|
|
|
|
BLAS svd_idamax |
1265
|
|
|
|
|
|
|
UTILITY svd_dmin |
1266
|
|
|
|
|
|
|
|
1267
|
|
|
|
|
|
|
***********************************************************************/ |
1268
|
|
|
|
|
|
|
|
1269
|
10
|
|
|
|
|
|
__SVDLIBC_LONG error_bound(__SVDLIBC_LONG *enough, double endl, double endr, |
1270
|
|
|
|
|
|
|
double *ritz, double *bnd, __SVDLIBC_LONG step, double tol) { |
1271
|
|
|
|
|
|
|
__SVDLIBC_LONG mid, i, neig; |
1272
|
|
|
|
|
|
|
double gapl, gap; |
1273
|
|
|
|
|
|
|
|
1274
|
|
|
|
|
|
|
/* massage error bounds for very close ritz values */ |
1275
|
10
|
|
|
|
|
|
mid = svd_idamax(step + 1, bnd, 1); |
1276
|
|
|
|
|
|
|
|
1277
|
60
|
100
|
|
|
|
|
for (i=((step+1) + (step-1)) / 2; i >= mid + 1; i -= 1) |
1278
|
50
|
50
|
|
|
|
|
if (fabs(ritz[i-1] - ritz[i]) < eps34 * fabs(ritz[i])) |
1279
|
0
|
0
|
|
|
|
|
if (bnd[i] > tol && bnd[i-1] > tol) { |
|
|
0
|
|
|
|
|
|
1280
|
0
|
|
|
|
|
|
bnd[i-1] = sqrt(bnd[i] * bnd[i] + bnd[i-1] * bnd[i-1]); |
1281
|
0
|
|
|
|
|
|
bnd[i] = 0.0; |
1282
|
|
|
|
|
|
|
} |
1283
|
|
|
|
|
|
|
|
1284
|
|
|
|
|
|
|
|
1285
|
10
|
50
|
|
|
|
|
for (i=((step+1) - (step-1)) / 2; i <= mid - 1; i +=1 ) |
1286
|
0
|
0
|
|
|
|
|
if (fabs(ritz[i+1] - ritz[i]) < eps34 * fabs(ritz[i])) |
1287
|
0
|
0
|
|
|
|
|
if (bnd[i] > tol && bnd[i+1] > tol) { |
|
|
0
|
|
|
|
|
|
1288
|
0
|
|
|
|
|
|
bnd[i+1] = sqrt(bnd[i] * bnd[i] + bnd[i+1] * bnd[i+1]); |
1289
|
0
|
|
|
|
|
|
bnd[i] = 0.0; |
1290
|
|
|
|
|
|
|
} |
1291
|
|
|
|
|
|
|
|
1292
|
|
|
|
|
|
|
/* refine the error bounds */ |
1293
|
10
|
|
|
|
|
|
neig = 0; |
1294
|
10
|
|
|
|
|
|
gapl = ritz[step] - ritz[0]; |
1295
|
70
|
100
|
|
|
|
|
for (i = 0; i <= step; i++) { |
1296
|
60
|
|
|
|
|
|
gap = gapl; |
1297
|
60
|
100
|
|
|
|
|
if (i < step) gapl = ritz[i+1] - ritz[i]; |
1298
|
60
|
|
|
|
|
|
gap = svd_dmin(gap, gapl); |
1299
|
60
|
50
|
|
|
|
|
if (gap > bnd[i]) bnd[i] = bnd[i] * (bnd[i] / gap); |
1300
|
60
|
50
|
|
|
|
|
if (bnd[i] <= 16.0 * eps * fabs(ritz[i])) { |
1301
|
60
|
|
|
|
|
|
neig++; |
1302
|
60
|
50
|
|
|
|
|
if (!*enough) *enough = endl < ritz[i] && ritz[i] < endr; |
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
1303
|
|
|
|
|
|
|
} |
1304
|
|
|
|
|
|
|
} |
1305
|
10
|
|
|
|
|
|
return neig; |
1306
|
|
|
|
|
|
|
} |
1307
|
|
|
|
|
|
|
|
1308
|
|
|
|
|
|
|
/*********************************************************************** |
1309
|
|
|
|
|
|
|
* * |
1310
|
|
|
|
|
|
|
* imtqlb() * |
1311
|
|
|
|
|
|
|
* * |
1312
|
|
|
|
|
|
|
***********************************************************************/ |
1313
|
|
|
|
|
|
|
/*********************************************************************** |
1314
|
|
|
|
|
|
|
|
1315
|
|
|
|
|
|
|
Description |
1316
|
|
|
|
|
|
|
----------- |
1317
|
|
|
|
|
|
|
|
1318
|
|
|
|
|
|
|
imtqlb() is a translation of a Fortran version of the Algol |
1319
|
|
|
|
|
|
|
procedure IMTQL1, Num. Math. 12, 377-383(1968) by Martin and |
1320
|
|
|
|
|
|
|
Wilkinson, as modified in Num. Math. 15, 450(1970) by Dubrulle. |
1321
|
|
|
|
|
|
|
Handbook for Auto. Comp., vol.II-Linear Algebra, 241-248(1971). |
1322
|
|
|
|
|
|
|
See also B. T. Smith et al, Eispack Guide, Lecture Notes in |
1323
|
|
|
|
|
|
|
Computer Science, Springer-Verlag, (1976). |
1324
|
|
|
|
|
|
|
|
1325
|
|
|
|
|
|
|
The function finds the eigenvalues of a symmetric tridiagonal |
1326
|
|
|
|
|
|
|
matrix by the implicit QL method. |
1327
|
|
|
|
|
|
|
|
1328
|
|
|
|
|
|
|
|
1329
|
|
|
|
|
|
|
Arguments |
1330
|
|
|
|
|
|
|
--------- |
1331
|
|
|
|
|
|
|
|
1332
|
|
|
|
|
|
|
(input) |
1333
|
|
|
|
|
|
|
n order of the symmetric tridiagonal matrix |
1334
|
|
|
|
|
|
|
d contains the diagonal elements of the input matrix |
1335
|
|
|
|
|
|
|
e contains the subdiagonal elements of the input matrix in its |
1336
|
|
|
|
|
|
|
last n-1 positions. e[0] is arbitrary |
1337
|
|
|
|
|
|
|
|
1338
|
|
|
|
|
|
|
(output) |
1339
|
|
|
|
|
|
|
d contains the eigenvalues in ascending order. if an error |
1340
|
|
|
|
|
|
|
exit is made, the eigenvalues are correct and ordered for |
1341
|
|
|
|
|
|
|
indices 0,1,...ierr, but may not be the smallest eigenvalues. |
1342
|
|
|
|
|
|
|
e has been destroyed. |
1343
|
|
|
|
|
|
|
ierr set to zero for normal return, j if the j-th eigenvalue has |
1344
|
|
|
|
|
|
|
not been determined after 30 iterations. |
1345
|
|
|
|
|
|
|
|
1346
|
|
|
|
|
|
|
Functions used |
1347
|
|
|
|
|
|
|
-------------- |
1348
|
|
|
|
|
|
|
|
1349
|
|
|
|
|
|
|
UTILITY svd_fsign |
1350
|
|
|
|
|
|
|
MISC svd_pythag |
1351
|
|
|
|
|
|
|
|
1352
|
|
|
|
|
|
|
***********************************************************************/ |
1353
|
|
|
|
|
|
|
|
1354
|
10
|
|
|
|
|
|
void imtqlb(__SVDLIBC_LONG n, double d[], double e[], double bnd[]) |
1355
|
|
|
|
|
|
|
|
1356
|
|
|
|
|
|
|
{ |
1357
|
|
|
|
|
|
|
__SVDLIBC_LONG last, l, m, i, iteration; |
1358
|
|
|
|
|
|
|
|
1359
|
|
|
|
|
|
|
/* various flags */ |
1360
|
|
|
|
|
|
|
__SVDLIBC_LONG exchange, convergence, underflow; |
1361
|
|
|
|
|
|
|
|
1362
|
|
|
|
|
|
|
double b, test, g, r, s, c, p, f; |
1363
|
|
|
|
|
|
|
|
1364
|
10
|
50
|
|
|
|
|
if (n == 1) return; |
1365
|
10
|
|
|
|
|
|
ierr = 0; |
1366
|
10
|
|
|
|
|
|
bnd[0] = 1.0; |
1367
|
10
|
|
|
|
|
|
last = n - 1; |
1368
|
60
|
100
|
|
|
|
|
for (i = 1; i < n; i++) { |
1369
|
50
|
|
|
|
|
|
bnd[i] = 0.0; |
1370
|
50
|
|
|
|
|
|
e[i-1] = e[i]; |
1371
|
|
|
|
|
|
|
} |
1372
|
10
|
|
|
|
|
|
e[last] = 0.0; |
1373
|
70
|
100
|
|
|
|
|
for (l = 0; l < n; l++) { |
1374
|
60
|
|
|
|
|
|
iteration = 0; |
1375
|
210
|
100
|
|
|
|
|
while (iteration <= 30) { |
1376
|
460
|
50
|
|
|
|
|
for (m = l; m < n; m++) { |
1377
|
460
|
|
|
|
|
|
convergence = FALSE; |
1378
|
460
|
100
|
|
|
|
|
if (m == last) break; |
1379
|
|
|
|
|
|
|
else { |
1380
|
360
|
|
|
|
|
|
test = fabs(d[m]) + fabs(d[m+1]); |
1381
|
360
|
100
|
|
|
|
|
if (test + fabs(e[m]) == test) convergence = TRUE; |
1382
|
|
|
|
|
|
|
} |
1383
|
360
|
100
|
|
|
|
|
if (convergence) break; |
1384
|
|
|
|
|
|
|
} |
1385
|
150
|
|
|
|
|
|
p = d[l]; |
1386
|
150
|
|
|
|
|
|
f = bnd[l]; |
1387
|
150
|
100
|
|
|
|
|
if (m != l) { |
1388
|
90
|
50
|
|
|
|
|
if (iteration == 30) { |
1389
|
0
|
|
|
|
|
|
ierr = l; |
1390
|
0
|
|
|
|
|
|
return; |
1391
|
|
|
|
|
|
|
} |
1392
|
90
|
|
|
|
|
|
iteration += 1; |
1393
|
|
|
|
|
|
|
/*........ form shift ........*/ |
1394
|
90
|
|
|
|
|
|
g = (d[l+1] - p) / (2.0 * e[l]); |
1395
|
90
|
|
|
|
|
|
r = svd_pythag(g, 1.0); |
1396
|
90
|
|
|
|
|
|
g = d[m] - p + e[l] / (g + svd_fsign(r, g)); |
1397
|
90
|
|
|
|
|
|
s = 1.0; |
1398
|
90
|
|
|
|
|
|
c = 1.0; |
1399
|
90
|
|
|
|
|
|
p = 0.0; |
1400
|
90
|
|
|
|
|
|
underflow = FALSE; |
1401
|
90
|
|
|
|
|
|
i = m - 1; |
1402
|
400
|
50
|
|
|
|
|
while (underflow == FALSE && i >= l) { |
|
|
100
|
|
|
|
|
|
1403
|
310
|
|
|
|
|
|
f = s * e[i]; |
1404
|
310
|
|
|
|
|
|
b = c * e[i]; |
1405
|
310
|
|
|
|
|
|
r = svd_pythag(f, g); |
1406
|
310
|
|
|
|
|
|
e[i+1] = r; |
1407
|
310
|
50
|
|
|
|
|
if (r == 0.0) underflow = TRUE; |
1408
|
|
|
|
|
|
|
else { |
1409
|
310
|
|
|
|
|
|
s = f / r; |
1410
|
310
|
|
|
|
|
|
c = g / r; |
1411
|
310
|
|
|
|
|
|
g = d[i+1] - p; |
1412
|
310
|
|
|
|
|
|
r = (d[i] - g) * s + 2.0 * c * b; |
1413
|
310
|
|
|
|
|
|
p = s * r; |
1414
|
310
|
|
|
|
|
|
d[i+1] = g + p; |
1415
|
310
|
|
|
|
|
|
g = c * r - b; |
1416
|
310
|
|
|
|
|
|
f = bnd[i+1]; |
1417
|
310
|
|
|
|
|
|
bnd[i+1] = s * bnd[i] + c * f; |
1418
|
310
|
|
|
|
|
|
bnd[i] = c * bnd[i] - s * f; |
1419
|
310
|
|
|
|
|
|
i--; |
1420
|
|
|
|
|
|
|
} |
1421
|
|
|
|
|
|
|
} /* end while (underflow != FALSE && i >= l) */ |
1422
|
|
|
|
|
|
|
/*........ recover from underflow .........*/ |
1423
|
90
|
50
|
|
|
|
|
if (underflow) { |
1424
|
0
|
|
|
|
|
|
d[i+1] -= p; |
1425
|
0
|
|
|
|
|
|
e[m] = 0.0; |
1426
|
|
|
|
|
|
|
} |
1427
|
|
|
|
|
|
|
else { |
1428
|
90
|
|
|
|
|
|
d[l] -= p; |
1429
|
90
|
|
|
|
|
|
e[l] = g; |
1430
|
90
|
|
|
|
|
|
e[m] = 0.0; |
1431
|
|
|
|
|
|
|
} |
1432
|
|
|
|
|
|
|
} /* end if (m != l) */ |
1433
|
|
|
|
|
|
|
else { |
1434
|
|
|
|
|
|
|
|
1435
|
|
|
|
|
|
|
/* order the eigenvalues */ |
1436
|
60
|
|
|
|
|
|
exchange = TRUE; |
1437
|
60
|
100
|
|
|
|
|
if (l != 0) { |
1438
|
50
|
|
|
|
|
|
i = l; |
1439
|
110
|
50
|
|
|
|
|
while (i >= 1 && exchange == TRUE) { |
|
|
100
|
|
|
|
|
|
1440
|
60
|
100
|
|
|
|
|
if (p < d[i-1]) { |
1441
|
10
|
|
|
|
|
|
d[i] = d[i-1]; |
1442
|
10
|
|
|
|
|
|
bnd[i] = bnd[i-1]; |
1443
|
10
|
|
|
|
|
|
i--; |
1444
|
|
|
|
|
|
|
} |
1445
|
50
|
|
|
|
|
|
else exchange = FALSE; |
1446
|
|
|
|
|
|
|
} |
1447
|
|
|
|
|
|
|
} |
1448
|
60
|
100
|
|
|
|
|
if (exchange) i = 0; |
1449
|
60
|
|
|
|
|
|
d[i] = p; |
1450
|
60
|
|
|
|
|
|
bnd[i] = f; |
1451
|
60
|
|
|
|
|
|
iteration = 31; |
1452
|
|
|
|
|
|
|
} |
1453
|
|
|
|
|
|
|
} /* end while (iteration <= 30) */ |
1454
|
|
|
|
|
|
|
} /* end for (l=0; l
|
1455
|
10
|
|
|
|
|
|
return; |
1456
|
|
|
|
|
|
|
} /* end main */ |
1457
|
|
|
|
|
|
|
|
1458
|
|
|
|
|
|
|
/*********************************************************************** |
1459
|
|
|
|
|
|
|
* * |
1460
|
|
|
|
|
|
|
* imtql2() * |
1461
|
|
|
|
|
|
|
* * |
1462
|
|
|
|
|
|
|
***********************************************************************/ |
1463
|
|
|
|
|
|
|
/*********************************************************************** |
1464
|
|
|
|
|
|
|
|
1465
|
|
|
|
|
|
|
Description |
1466
|
|
|
|
|
|
|
----------- |
1467
|
|
|
|
|
|
|
|
1468
|
|
|
|
|
|
|
imtql2() is a translation of a Fortran version of the Algol |
1469
|
|
|
|
|
|
|
procedure IMTQL2, Num. Math. 12, 377-383(1968) by Martin and |
1470
|
|
|
|
|
|
|
Wilkinson, as modified in Num. Math. 15, 450(1970) by Dubrulle. |
1471
|
|
|
|
|
|
|
Handbook for Auto. Comp., vol.II-Linear Algebra, 241-248(1971). |
1472
|
|
|
|
|
|
|
See also B. T. Smith et al, Eispack Guide, Lecture Notes in |
1473
|
|
|
|
|
|
|
Computer Science, Springer-Verlag, (1976). |
1474
|
|
|
|
|
|
|
|
1475
|
|
|
|
|
|
|
This function finds the eigenvalues and eigenvectors of a symmetric |
1476
|
|
|
|
|
|
|
tridiagonal matrix by the implicit QL method. |
1477
|
|
|
|
|
|
|
|
1478
|
|
|
|
|
|
|
|
1479
|
|
|
|
|
|
|
Arguments |
1480
|
|
|
|
|
|
|
--------- |
1481
|
|
|
|
|
|
|
|
1482
|
|
|
|
|
|
|
(input) |
1483
|
|
|
|
|
|
|
nm row dimension of the symmetric tridiagonal matrix |
1484
|
|
|
|
|
|
|
n order of the matrix |
1485
|
|
|
|
|
|
|
d contains the diagonal elements of the input matrix |
1486
|
|
|
|
|
|
|
e contains the subdiagonal elements of the input matrix in its |
1487
|
|
|
|
|
|
|
last n-1 positions. e[0] is arbitrary |
1488
|
|
|
|
|
|
|
z contains the identity matrix |
1489
|
|
|
|
|
|
|
|
1490
|
|
|
|
|
|
|
(output) |
1491
|
|
|
|
|
|
|
d contains the eigenvalues in ascending order. if an error |
1492
|
|
|
|
|
|
|
exit is made, the eigenvalues are correct but unordered for |
1493
|
|
|
|
|
|
|
for indices 0,1,...,ierr. |
1494
|
|
|
|
|
|
|
e has been destroyed. |
1495
|
|
|
|
|
|
|
z contains orthonormal eigenvectors of the symmetric |
1496
|
|
|
|
|
|
|
tridiagonal (or full) matrix. if an error exit is made, |
1497
|
|
|
|
|
|
|
z contains the eigenvectors associated with the stored |
1498
|
|
|
|
|
|
|
eigenvalues. |
1499
|
|
|
|
|
|
|
ierr set to zero for normal return, j if the j-th eigenvalue has |
1500
|
|
|
|
|
|
|
not been determined after 30 iterations. |
1501
|
|
|
|
|
|
|
|
1502
|
|
|
|
|
|
|
|
1503
|
|
|
|
|
|
|
Functions used |
1504
|
|
|
|
|
|
|
-------------- |
1505
|
|
|
|
|
|
|
UTILITY svd_fsign |
1506
|
|
|
|
|
|
|
MISC svd_pythag |
1507
|
|
|
|
|
|
|
|
1508
|
|
|
|
|
|
|
***********************************************************************/ |
1509
|
|
|
|
|
|
|
|
1510
|
10
|
|
|
|
|
|
void imtql2(__SVDLIBC_LONG nm, __SVDLIBC_LONG n, double d[], double e[], double z[]) |
1511
|
|
|
|
|
|
|
|
1512
|
|
|
|
|
|
|
{ |
1513
|
|
|
|
|
|
|
__SVDLIBC_LONG index, nnm, j, last, l, m, i, k, iteration, convergence, underflow; |
1514
|
|
|
|
|
|
|
double b, test, g, r, s, c, p, f; |
1515
|
10
|
50
|
|
|
|
|
if (n == 1) return; |
1516
|
10
|
|
|
|
|
|
ierr = 0; |
1517
|
10
|
|
|
|
|
|
last = n - 1; |
1518
|
60
|
100
|
|
|
|
|
for (i = 1; i < n; i++) e[i-1] = e[i]; |
1519
|
10
|
|
|
|
|
|
e[last] = 0.0; |
1520
|
10
|
|
|
|
|
|
nnm = n * nm; |
1521
|
70
|
100
|
|
|
|
|
for (l = 0; l < n; l++) { |
1522
|
60
|
|
|
|
|
|
iteration = 0; |
1523
|
|
|
|
|
|
|
|
1524
|
|
|
|
|
|
|
/* look for small sub-diagonal element */ |
1525
|
150
|
50
|
|
|
|
|
while (iteration <= 30) { |
1526
|
460
|
50
|
|
|
|
|
for (m = l; m < n; m++) { |
1527
|
460
|
|
|
|
|
|
convergence = FALSE; |
1528
|
460
|
100
|
|
|
|
|
if (m == last) break; |
1529
|
|
|
|
|
|
|
else { |
1530
|
360
|
|
|
|
|
|
test = fabs(d[m]) + fabs(d[m+1]); |
1531
|
360
|
100
|
|
|
|
|
if (test + fabs(e[m]) == test) convergence = TRUE; |
1532
|
|
|
|
|
|
|
} |
1533
|
360
|
100
|
|
|
|
|
if (convergence) break; |
1534
|
|
|
|
|
|
|
} |
1535
|
150
|
100
|
|
|
|
|
if (m != l) { |
1536
|
|
|
|
|
|
|
|
1537
|
|
|
|
|
|
|
/* set error -- no convergence to an eigenvalue after |
1538
|
|
|
|
|
|
|
* 30 iterations. */ |
1539
|
90
|
50
|
|
|
|
|
if (iteration == 30) { |
1540
|
0
|
|
|
|
|
|
ierr = l; |
1541
|
0
|
|
|
|
|
|
return; |
1542
|
|
|
|
|
|
|
} |
1543
|
90
|
|
|
|
|
|
p = d[l]; |
1544
|
90
|
|
|
|
|
|
iteration += 1; |
1545
|
|
|
|
|
|
|
|
1546
|
|
|
|
|
|
|
/* form shift */ |
1547
|
90
|
|
|
|
|
|
g = (d[l+1] - p) / (2.0 * e[l]); |
1548
|
90
|
|
|
|
|
|
r = svd_pythag(g, 1.0); |
1549
|
90
|
|
|
|
|
|
g = d[m] - p + e[l] / (g + svd_fsign(r, g)); |
1550
|
90
|
|
|
|
|
|
s = 1.0; |
1551
|
90
|
|
|
|
|
|
c = 1.0; |
1552
|
90
|
|
|
|
|
|
p = 0.0; |
1553
|
90
|
|
|
|
|
|
underflow = FALSE; |
1554
|
90
|
|
|
|
|
|
i = m - 1; |
1555
|
400
|
50
|
|
|
|
|
while (underflow == FALSE && i >= l) { |
|
|
100
|
|
|
|
|
|
1556
|
310
|
|
|
|
|
|
f = s * e[i]; |
1557
|
310
|
|
|
|
|
|
b = c * e[i]; |
1558
|
310
|
|
|
|
|
|
r = svd_pythag(f, g); |
1559
|
310
|
|
|
|
|
|
e[i+1] = r; |
1560
|
310
|
50
|
|
|
|
|
if (r == 0.0) underflow = TRUE; |
1561
|
|
|
|
|
|
|
else { |
1562
|
310
|
|
|
|
|
|
s = f / r; |
1563
|
310
|
|
|
|
|
|
c = g / r; |
1564
|
310
|
|
|
|
|
|
g = d[i+1] - p; |
1565
|
310
|
|
|
|
|
|
r = (d[i] - g) * s + 2.0 * c * b; |
1566
|
310
|
|
|
|
|
|
p = s * r; |
1567
|
310
|
|
|
|
|
|
d[i+1] = g + p; |
1568
|
310
|
|
|
|
|
|
g = c * r - b; |
1569
|
|
|
|
|
|
|
|
1570
|
|
|
|
|
|
|
/* form vector */ |
1571
|
2170
|
100
|
|
|
|
|
for (k = 0; k < nnm; k += n) { |
1572
|
1860
|
|
|
|
|
|
index = k + i; |
1573
|
1860
|
|
|
|
|
|
f = z[index+1]; |
1574
|
1860
|
|
|
|
|
|
z[index+1] = s * z[index] + c * f; |
1575
|
1860
|
|
|
|
|
|
z[index] = c * z[index] - s * f; |
1576
|
|
|
|
|
|
|
} |
1577
|
310
|
|
|
|
|
|
i--; |
1578
|
|
|
|
|
|
|
} |
1579
|
|
|
|
|
|
|
} /* end while (underflow != FALSE && i >= l) */ |
1580
|
|
|
|
|
|
|
/*........ recover from underflow .........*/ |
1581
|
90
|
50
|
|
|
|
|
if (underflow) { |
1582
|
0
|
|
|
|
|
|
d[i+1] -= p; |
1583
|
0
|
|
|
|
|
|
e[m] = 0.0; |
1584
|
|
|
|
|
|
|
} |
1585
|
|
|
|
|
|
|
else { |
1586
|
90
|
|
|
|
|
|
d[l] -= p; |
1587
|
90
|
|
|
|
|
|
e[l] = g; |
1588
|
90
|
|
|
|
|
|
e[m] = 0.0; |
1589
|
|
|
|
|
|
|
} |
1590
|
|
|
|
|
|
|
} |
1591
|
60
|
|
|
|
|
|
else break; |
1592
|
|
|
|
|
|
|
} /*...... end while (iteration <= 30) .........*/ |
1593
|
|
|
|
|
|
|
} /*...... end for (l=0; l
|
1594
|
|
|
|
|
|
|
|
1595
|
|
|
|
|
|
|
/* order the eigenvalues */ |
1596
|
60
|
100
|
|
|
|
|
for (l = 1; l < n; l++) { |
1597
|
50
|
|
|
|
|
|
i = l - 1; |
1598
|
50
|
|
|
|
|
|
k = i; |
1599
|
50
|
|
|
|
|
|
p = d[i]; |
1600
|
200
|
100
|
|
|
|
|
for (j = l; j < n; j++) { |
1601
|
150
|
100
|
|
|
|
|
if (d[j] < p) { |
1602
|
10
|
|
|
|
|
|
k = j; |
1603
|
10
|
|
|
|
|
|
p = d[j]; |
1604
|
|
|
|
|
|
|
} |
1605
|
|
|
|
|
|
|
} |
1606
|
|
|
|
|
|
|
/* ...and corresponding eigenvectors */ |
1607
|
50
|
100
|
|
|
|
|
if (k != i) { |
1608
|
10
|
|
|
|
|
|
d[k] = d[i]; |
1609
|
10
|
|
|
|
|
|
d[i] = p; |
1610
|
70
|
100
|
|
|
|
|
for (j = 0; j < nnm; j += n) { |
1611
|
60
|
|
|
|
|
|
p = z[j+i]; |
1612
|
60
|
|
|
|
|
|
z[j+i] = z[j+k]; |
1613
|
60
|
|
|
|
|
|
z[j+k] = p; |
1614
|
|
|
|
|
|
|
} |
1615
|
|
|
|
|
|
|
} |
1616
|
|
|
|
|
|
|
} |
1617
|
10
|
|
|
|
|
|
return; |
1618
|
|
|
|
|
|
|
} /*...... end main ............................*/ |
1619
|
|
|
|
|
|
|
|
1620
|
|
|
|
|
|
|
/*********************************************************************** |
1621
|
|
|
|
|
|
|
* * |
1622
|
|
|
|
|
|
|
* machar() * |
1623
|
|
|
|
|
|
|
* * |
1624
|
|
|
|
|
|
|
***********************************************************************/ |
1625
|
|
|
|
|
|
|
/*********************************************************************** |
1626
|
|
|
|
|
|
|
|
1627
|
|
|
|
|
|
|
Description |
1628
|
|
|
|
|
|
|
----------- |
1629
|
|
|
|
|
|
|
|
1630
|
|
|
|
|
|
|
This function is a partial translation of a Fortran-77 subroutine |
1631
|
|
|
|
|
|
|
written by W. J. Cody of Argonne National Laboratory. |
1632
|
|
|
|
|
|
|
It dynamically determines the listed machine parameters of the |
1633
|
|
|
|
|
|
|
floating-point arithmetic. According to the documentation of |
1634
|
|
|
|
|
|
|
the Fortran code, "the determination of the first three uses an |
1635
|
|
|
|
|
|
|
extension of an algorithm due to M. Malcolm, ACM 15 (1972), |
1636
|
|
|
|
|
|
|
pp. 949-951, incorporating some, but not all, of the improvements |
1637
|
|
|
|
|
|
|
suggested by M. Gentleman and S. Marovich, CACM 17 (1974), |
1638
|
|
|
|
|
|
|
pp. 276-277." The complete Fortran version of this translation is |
1639
|
|
|
|
|
|
|
documented in W. J. Cody, "Machar: a Subroutine to Dynamically |
1640
|
|
|
|
|
|
|
Determine Determine Machine Parameters," TOMS 14, December, 1988. |
1641
|
|
|
|
|
|
|
|
1642
|
|
|
|
|
|
|
|
1643
|
|
|
|
|
|
|
Parameters reported |
1644
|
|
|
|
|
|
|
------------------- |
1645
|
|
|
|
|
|
|
|
1646
|
|
|
|
|
|
|
ibeta the radix for the floating-point representation |
1647
|
|
|
|
|
|
|
it the number of base ibeta digits in the floating-point |
1648
|
|
|
|
|
|
|
significand |
1649
|
|
|
|
|
|
|
irnd 0 if floating-point addition chops |
1650
|
|
|
|
|
|
|
1 if floating-point addition rounds, but not in the |
1651
|
|
|
|
|
|
|
ieee style |
1652
|
|
|
|
|
|
|
2 if floating-point addition rounds in the ieee style |
1653
|
|
|
|
|
|
|
3 if floating-point addition chops, and there is |
1654
|
|
|
|
|
|
|
partial underflow |
1655
|
|
|
|
|
|
|
4 if floating-point addition rounds, but not in the |
1656
|
|
|
|
|
|
|
ieee style, and there is partial underflow |
1657
|
|
|
|
|
|
|
5 if floating-point addition rounds in the ieee style, |
1658
|
|
|
|
|
|
|
and there is partial underflow |
1659
|
|
|
|
|
|
|
machep the largest negative integer such that |
1660
|
|
|
|
|
|
|
1.0+float(ibeta)**machep .ne. 1.0, except that |
1661
|
|
|
|
|
|
|
machep is bounded below by -(it+3) |
1662
|
|
|
|
|
|
|
negeps the largest negative integer such that |
1663
|
|
|
|
|
|
|
1.0-float(ibeta)**negeps .ne. 1.0, except that |
1664
|
|
|
|
|
|
|
negeps is bounded below by -(it+3) |
1665
|
|
|
|
|
|
|
|
1666
|
|
|
|
|
|
|
***********************************************************************/ |
1667
|
|
|
|
|
|
|
|
1668
|
10
|
|
|
|
|
|
void machar(__SVDLIBC_LONG *ibeta, __SVDLIBC_LONG *it, __SVDLIBC_LONG *irnd, __SVDLIBC_LONG *machep, __SVDLIBC_LONG *negep) { |
1669
|
|
|
|
|
|
|
|
1670
|
|
|
|
|
|
|
volatile double beta, betain, betah, a, b, ZERO, ONE, TWO, temp, tempa, |
1671
|
|
|
|
|
|
|
temp1; |
1672
|
|
|
|
|
|
|
__SVDLIBC_LONG i, itemp; |
1673
|
|
|
|
|
|
|
|
1674
|
10
|
|
|
|
|
|
ONE = (double) 1; |
1675
|
10
|
|
|
|
|
|
TWO = ONE + ONE; |
1676
|
10
|
|
|
|
|
|
ZERO = ONE - ONE; |
1677
|
|
|
|
|
|
|
|
1678
|
10
|
|
|
|
|
|
a = ONE; |
1679
|
10
|
|
|
|
|
|
temp1 = ONE; |
1680
|
540
|
100
|
|
|
|
|
while (temp1 - ONE == ZERO) { |
1681
|
530
|
|
|
|
|
|
a = a + a; |
1682
|
530
|
|
|
|
|
|
temp = a + ONE; |
1683
|
530
|
|
|
|
|
|
temp1 = temp - a; |
1684
|
530
|
|
|
|
|
|
b += a; /* to prevent icc compiler error */ |
1685
|
|
|
|
|
|
|
} |
1686
|
10
|
|
|
|
|
|
b = ONE; |
1687
|
10
|
|
|
|
|
|
itemp = 0; |
1688
|
20
|
100
|
|
|
|
|
while (itemp == 0) { |
1689
|
10
|
|
|
|
|
|
b = b + b; |
1690
|
10
|
|
|
|
|
|
temp = a + b; |
1691
|
10
|
|
|
|
|
|
itemp = (__SVDLIBC_LONG)(temp - a); |
1692
|
|
|
|
|
|
|
} |
1693
|
10
|
|
|
|
|
|
*ibeta = itemp; |
1694
|
10
|
|
|
|
|
|
beta = (double) *ibeta; |
1695
|
|
|
|
|
|
|
|
1696
|
10
|
|
|
|
|
|
*it = 0; |
1697
|
10
|
|
|
|
|
|
b = ONE; |
1698
|
10
|
|
|
|
|
|
temp1 = ONE; |
1699
|
540
|
100
|
|
|
|
|
while (temp1 - ONE == ZERO) { |
1700
|
530
|
|
|
|
|
|
*it = *it + 1; |
1701
|
530
|
|
|
|
|
|
b = b * beta; |
1702
|
530
|
|
|
|
|
|
temp = b + ONE; |
1703
|
530
|
|
|
|
|
|
temp1 = temp - b; |
1704
|
|
|
|
|
|
|
} |
1705
|
10
|
|
|
|
|
|
*irnd = 0; |
1706
|
10
|
|
|
|
|
|
betah = beta / TWO; |
1707
|
10
|
|
|
|
|
|
temp = a + betah; |
1708
|
10
|
50
|
|
|
|
|
if (temp - a != ZERO) *irnd = 1; |
1709
|
10
|
|
|
|
|
|
tempa = a + beta; |
1710
|
10
|
|
|
|
|
|
temp = tempa + betah; |
1711
|
10
|
50
|
|
|
|
|
if ((*irnd == 0) && (temp - tempa != ZERO)) *irnd = 2; |
|
|
50
|
|
|
|
|
|
1712
|
|
|
|
|
|
|
|
1713
|
10
|
|
|
|
|
|
*negep = *it + 3; |
1714
|
10
|
|
|
|
|
|
betain = ONE / beta; |
1715
|
10
|
|
|
|
|
|
a = ONE; |
1716
|
570
|
100
|
|
|
|
|
for (i = 0; i < *negep; i++) a = a * betain; |
1717
|
10
|
|
|
|
|
|
b = a; |
1718
|
10
|
|
|
|
|
|
temp = ONE - a; |
1719
|
40
|
100
|
|
|
|
|
while (temp-ONE == ZERO) { |
1720
|
30
|
|
|
|
|
|
a = a * beta; |
1721
|
30
|
|
|
|
|
|
*negep = *negep - 1; |
1722
|
30
|
|
|
|
|
|
temp = ONE - a; |
1723
|
|
|
|
|
|
|
} |
1724
|
10
|
|
|
|
|
|
*negep = -(*negep); |
1725
|
|
|
|
|
|
|
|
1726
|
10
|
|
|
|
|
|
*machep = -(*it) - 3; |
1727
|
10
|
|
|
|
|
|
a = b; |
1728
|
10
|
|
|
|
|
|
temp = ONE + a; |
1729
|
50
|
100
|
|
|
|
|
while (temp - ONE == ZERO) { |
1730
|
40
|
|
|
|
|
|
a = a * beta; |
1731
|
40
|
|
|
|
|
|
*machep = *machep + 1; |
1732
|
40
|
|
|
|
|
|
temp = ONE + a; |
1733
|
|
|
|
|
|
|
} |
1734
|
10
|
|
|
|
|
|
eps = a; |
1735
|
10
|
|
|
|
|
|
return; |
1736
|
|
|
|
|
|
|
} |
1737
|
|
|
|
|
|
|
|
1738
|
|
|
|
|
|
|
/*********************************************************************** |
1739
|
|
|
|
|
|
|
* * |
1740
|
|
|
|
|
|
|
* store() * |
1741
|
|
|
|
|
|
|
* * |
1742
|
|
|
|
|
|
|
***********************************************************************/ |
1743
|
|
|
|
|
|
|
/*********************************************************************** |
1744
|
|
|
|
|
|
|
|
1745
|
|
|
|
|
|
|
Description |
1746
|
|
|
|
|
|
|
----------- |
1747
|
|
|
|
|
|
|
|
1748
|
|
|
|
|
|
|
store() is a user-supplied function which, based on the input |
1749
|
|
|
|
|
|
|
operation flag, stores to or retrieves from memory a vector. |
1750
|
|
|
|
|
|
|
|
1751
|
|
|
|
|
|
|
|
1752
|
|
|
|
|
|
|
Arguments |
1753
|
|
|
|
|
|
|
--------- |
1754
|
|
|
|
|
|
|
|
1755
|
|
|
|
|
|
|
(input) |
1756
|
|
|
|
|
|
|
n length of vector to be stored or retrieved |
1757
|
|
|
|
|
|
|
isw operation flag: |
1758
|
|
|
|
|
|
|
isw = 1 request to store j-th Lanczos vector q(j) |
1759
|
|
|
|
|
|
|
isw = 2 request to retrieve j-th Lanczos vector q(j) |
1760
|
|
|
|
|
|
|
isw = 3 request to store q(j) for j = 0 or 1 |
1761
|
|
|
|
|
|
|
isw = 4 request to retrieve q(j) for j = 0 or 1 |
1762
|
|
|
|
|
|
|
s contains the vector to be stored for a "store" request |
1763
|
|
|
|
|
|
|
|
1764
|
|
|
|
|
|
|
(output) |
1765
|
|
|
|
|
|
|
s contains the vector retrieved for a "retrieve" request |
1766
|
|
|
|
|
|
|
|
1767
|
|
|
|
|
|
|
Functions used |
1768
|
|
|
|
|
|
|
-------------- |
1769
|
|
|
|
|
|
|
|
1770
|
|
|
|
|
|
|
BLAS svd_dcopy |
1771
|
|
|
|
|
|
|
|
1772
|
|
|
|
|
|
|
***********************************************************************/ |
1773
|
|
|
|
|
|
|
|
1774
|
440
|
|
|
|
|
|
void store(__SVDLIBC_LONG n, __SVDLIBC_LONG isw, __SVDLIBC_LONG j, double *s) { |
1775
|
|
|
|
|
|
|
/* printf("called store %ld %ld\n", isw, j); */ |
1776
|
440
|
|
|
|
|
|
switch(isw) { |
1777
|
|
|
|
|
|
|
case STORQ: |
1778
|
60
|
50
|
|
|
|
|
if (!LanStore[j + MAXLL]) { |
1779
|
60
|
50
|
|
|
|
|
if (!(LanStore[j + MAXLL] = svd_doubleArray(n, FALSE, "LanStore[j]"))) |
1780
|
0
|
|
|
|
|
|
svd_fatalError("svdLAS2: failed to allocate LanStore[%d]", j + MAXLL); |
1781
|
|
|
|
|
|
|
} |
1782
|
60
|
|
|
|
|
|
svd_dcopy(n, s, 1, LanStore[j + MAXLL], 1); |
1783
|
60
|
|
|
|
|
|
break; |
1784
|
|
|
|
|
|
|
case RETRQ: |
1785
|
360
|
50
|
|
|
|
|
if (!LanStore[j + MAXLL]) |
1786
|
0
|
|
|
|
|
|
svd_fatalError("svdLAS2: store (RETRQ) called on index %d (not allocated)", |
1787
|
|
|
|
|
|
|
j + MAXLL); |
1788
|
360
|
|
|
|
|
|
svd_dcopy(n, LanStore[j + MAXLL], 1, s, 1); |
1789
|
360
|
|
|
|
|
|
break; |
1790
|
|
|
|
|
|
|
case STORP: |
1791
|
20
|
50
|
|
|
|
|
if (j >= MAXLL) { |
1792
|
0
|
|
|
|
|
|
svd_error("svdLAS2: store (STORP) called with j >= MAXLL"); |
1793
|
0
|
|
|
|
|
|
break; |
1794
|
|
|
|
|
|
|
} |
1795
|
20
|
50
|
|
|
|
|
if (!LanStore[j]) { |
1796
|
20
|
50
|
|
|
|
|
if (!(LanStore[j] = svd_doubleArray(n, FALSE, "LanStore[j]"))) |
1797
|
0
|
|
|
|
|
|
svd_fatalError("svdLAS2: failed to allocate LanStore[%d]", j); |
1798
|
|
|
|
|
|
|
} |
1799
|
20
|
|
|
|
|
|
svd_dcopy(n, s, 1, LanStore[j], 1); |
1800
|
20
|
|
|
|
|
|
break; |
1801
|
|
|
|
|
|
|
case RETRP: |
1802
|
0
|
0
|
|
|
|
|
if (j >= MAXLL) { |
1803
|
0
|
|
|
|
|
|
svd_error("svdLAS2: store (RETRP) called with j >= MAXLL"); |
1804
|
0
|
|
|
|
|
|
break; |
1805
|
|
|
|
|
|
|
} |
1806
|
0
|
0
|
|
|
|
|
if (!LanStore[j]) |
1807
|
0
|
|
|
|
|
|
svd_fatalError("svdLAS2: store (RETRP) called on index %d (not allocated)", |
1808
|
|
|
|
|
|
|
j); |
1809
|
0
|
|
|
|
|
|
svd_dcopy(n, LanStore[j], 1, s, 1); |
1810
|
0
|
|
|
|
|
|
break; |
1811
|
|
|
|
|
|
|
} |
1812
|
440
|
|
|
|
|
|
return; |
1813
|
|
|
|
|
|
|
} |