1 | /*************************************************************************
|
---|
2 | Copyright (c) 1992-2007 The University of Tennessee. All rights reserved.
|
---|
3 |
|
---|
4 | Contributors:
|
---|
5 | * Sergey Bochkanov (ALGLIB project). Translation from FORTRAN to
|
---|
6 | pseudocode.
|
---|
7 |
|
---|
8 | See subroutines comments for additional copyrights.
|
---|
9 |
|
---|
10 | >>> SOURCE LICENSE >>>
|
---|
11 | This program is free software; you can redistribute it and/or modify
|
---|
12 | it under the terms of the GNU General Public License as published by
|
---|
13 | the Free Software Foundation (www.fsf.org); either version 2 of the
|
---|
14 | License, or (at your option) any later version.
|
---|
15 |
|
---|
16 | This program is distributed in the hope that it will be useful,
|
---|
17 | but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
18 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
19 | GNU General Public License for more details.
|
---|
20 |
|
---|
21 | A copy of the GNU General Public License is available at
|
---|
22 | http://www.fsf.org/licensing/licenses
|
---|
23 |
|
---|
24 | >>> END OF LICENSE >>>
|
---|
25 | *************************************************************************/
|
---|
26 |
|
---|
27 | using System;
|
---|
28 |
|
---|
29 | namespace alglib
|
---|
30 | {
|
---|
31 | public class srcond
|
---|
32 | {
|
---|
33 | /*************************************************************************
|
---|
34 | Condition number estimate of a symmetric matrix
|
---|
35 |
|
---|
36 | The algorithm calculates a lower bound of the condition number. In this
|
---|
37 | case, the algorithm does not return a lower bound of the condition number,
|
---|
38 | but an inverse number (to avoid an overflow in case of a singular matrix).
|
---|
39 |
|
---|
40 | It should be noted that 1-norm and inf-norm condition numbers of symmetric
|
---|
41 | matrices are equal, so the algorithm doesn't take into account the
|
---|
42 | differences between these types of norms.
|
---|
43 |
|
---|
44 | Input parameters:
|
---|
45 | A - symmetric definite matrix which is given by its upper or
|
---|
46 | lower triangle depending on IsUpper.
|
---|
47 | Array with elements [0..N-1, 0..N-1].
|
---|
48 | N - size of matrix A.
|
---|
49 | IsUpper - storage format.
|
---|
50 |
|
---|
51 | Result:
|
---|
52 | 1/LowerBound(cond(A))
|
---|
53 | *************************************************************************/
|
---|
54 | public static double smatrixrcond(ref double[,] a,
|
---|
55 | int n,
|
---|
56 | bool isupper)
|
---|
57 | {
|
---|
58 | double result = 0;
|
---|
59 | int i = 0;
|
---|
60 | int j = 0;
|
---|
61 | double[,] a1 = new double[0,0];
|
---|
62 |
|
---|
63 | a1 = new double[n+1, n+1];
|
---|
64 | for(i=1; i<=n; i++)
|
---|
65 | {
|
---|
66 | if( isupper )
|
---|
67 | {
|
---|
68 | for(j=i; j<=n; j++)
|
---|
69 | {
|
---|
70 | a1[i,j] = a[i-1,j-1];
|
---|
71 | }
|
---|
72 | }
|
---|
73 | else
|
---|
74 | {
|
---|
75 | for(j=1; j<=i; j++)
|
---|
76 | {
|
---|
77 | a1[i,j] = a[i-1,j-1];
|
---|
78 | }
|
---|
79 | }
|
---|
80 | }
|
---|
81 | result = rcondsymmetric(a1, n, isupper);
|
---|
82 | return result;
|
---|
83 | }
|
---|
84 |
|
---|
85 |
|
---|
86 | /*************************************************************************
|
---|
87 | Condition number estimate of a matrix given by LDLT-decomposition
|
---|
88 |
|
---|
89 | The algorithm calculates a lower bound of the condition number. In this
|
---|
90 | case, the algorithm does not return a lower bound of the condition number,
|
---|
91 | but an inverse number (to avoid an overflow in case of a singular matrix).
|
---|
92 |
|
---|
93 | It should be noted that 1-norm and inf-norm condition numbers of symmetric
|
---|
94 | matrices are equal, so the algorithm doesn't take into account the
|
---|
95 | differences between these types of norms.
|
---|
96 |
|
---|
97 | Input parameters:
|
---|
98 | L - LDLT-decomposition of matrix A given by the upper or lower
|
---|
99 | triangle depending on IsUpper.
|
---|
100 | Output of SMatrixLDLT subroutine.
|
---|
101 | Pivots - table of permutations which were made during LDLT-decomposition,
|
---|
102 | Output of SMatrixLDLT subroutine.
|
---|
103 | N - size of matrix A.
|
---|
104 | IsUpper - storage format.
|
---|
105 |
|
---|
106 | Result:
|
---|
107 | 1/LowerBound(cond(A))
|
---|
108 | *************************************************************************/
|
---|
109 | public static double smatrixldltrcond(ref double[,] l,
|
---|
110 | ref int[] pivots,
|
---|
111 | int n,
|
---|
112 | bool isupper)
|
---|
113 | {
|
---|
114 | double result = 0;
|
---|
115 | int i = 0;
|
---|
116 | int j = 0;
|
---|
117 | double[,] l1 = new double[0,0];
|
---|
118 | int[] p1 = new int[0];
|
---|
119 |
|
---|
120 | l1 = new double[n+1, n+1];
|
---|
121 | for(i=1; i<=n; i++)
|
---|
122 | {
|
---|
123 | if( isupper )
|
---|
124 | {
|
---|
125 | for(j=i; j<=n; j++)
|
---|
126 | {
|
---|
127 | l1[i,j] = l[i-1,j-1];
|
---|
128 | }
|
---|
129 | }
|
---|
130 | else
|
---|
131 | {
|
---|
132 | for(j=1; j<=i; j++)
|
---|
133 | {
|
---|
134 | l1[i,j] = l[i-1,j-1];
|
---|
135 | }
|
---|
136 | }
|
---|
137 | }
|
---|
138 | p1 = new int[n+1];
|
---|
139 | for(i=1; i<=n; i++)
|
---|
140 | {
|
---|
141 | if( pivots[i-1]>=0 )
|
---|
142 | {
|
---|
143 | p1[i] = pivots[i-1]+1;
|
---|
144 | }
|
---|
145 | else
|
---|
146 | {
|
---|
147 | p1[i] = -(pivots[i-1]+n+1);
|
---|
148 | }
|
---|
149 | }
|
---|
150 | result = rcondldlt(ref l1, ref p1, n, isupper);
|
---|
151 | return result;
|
---|
152 | }
|
---|
153 |
|
---|
154 |
|
---|
155 | public static double rcondsymmetric(double[,] a,
|
---|
156 | int n,
|
---|
157 | bool isupper)
|
---|
158 | {
|
---|
159 | double result = 0;
|
---|
160 | int i = 0;
|
---|
161 | int j = 0;
|
---|
162 | int im = 0;
|
---|
163 | int jm = 0;
|
---|
164 | double v = 0;
|
---|
165 | double nrm = 0;
|
---|
166 | int[] pivots = new int[0];
|
---|
167 |
|
---|
168 | a = (double[,])a.Clone();
|
---|
169 |
|
---|
170 | nrm = 0;
|
---|
171 | for(j=1; j<=n; j++)
|
---|
172 | {
|
---|
173 | v = 0;
|
---|
174 | for(i=1; i<=n; i++)
|
---|
175 | {
|
---|
176 | im = i;
|
---|
177 | jm = j;
|
---|
178 | if( isupper & j<i )
|
---|
179 | {
|
---|
180 | im = j;
|
---|
181 | jm = i;
|
---|
182 | }
|
---|
183 | if( !isupper & j>i )
|
---|
184 | {
|
---|
185 | im = j;
|
---|
186 | jm = i;
|
---|
187 | }
|
---|
188 | v = v+Math.Abs(a[im,jm]);
|
---|
189 | }
|
---|
190 | nrm = Math.Max(nrm, v);
|
---|
191 | }
|
---|
192 | ldlt.ldltdecomposition(ref a, n, isupper, ref pivots);
|
---|
193 | internalldltrcond(ref a, ref pivots, n, isupper, true, nrm, ref v);
|
---|
194 | result = v;
|
---|
195 | return result;
|
---|
196 | }
|
---|
197 |
|
---|
198 |
|
---|
199 | public static double rcondldlt(ref double[,] l,
|
---|
200 | ref int[] pivots,
|
---|
201 | int n,
|
---|
202 | bool isupper)
|
---|
203 | {
|
---|
204 | double result = 0;
|
---|
205 | double v = 0;
|
---|
206 |
|
---|
207 | internalldltrcond(ref l, ref pivots, n, isupper, false, 0, ref v);
|
---|
208 | result = v;
|
---|
209 | return result;
|
---|
210 | }
|
---|
211 |
|
---|
212 |
|
---|
213 | public static void internalldltrcond(ref double[,] l,
|
---|
214 | ref int[] pivots,
|
---|
215 | int n,
|
---|
216 | bool isupper,
|
---|
217 | bool isnormprovided,
|
---|
218 | double anorm,
|
---|
219 | ref double rcond)
|
---|
220 | {
|
---|
221 | int i = 0;
|
---|
222 | int kase = 0;
|
---|
223 | int k = 0;
|
---|
224 | int km1 = 0;
|
---|
225 | int km2 = 0;
|
---|
226 | int kp1 = 0;
|
---|
227 | int kp2 = 0;
|
---|
228 | double ainvnm = 0;
|
---|
229 | double[] work0 = new double[0];
|
---|
230 | double[] work1 = new double[0];
|
---|
231 | double[] work2 = new double[0];
|
---|
232 | int[] iwork = new int[0];
|
---|
233 | double v = 0;
|
---|
234 | int i_ = 0;
|
---|
235 |
|
---|
236 | System.Diagnostics.Debug.Assert(n>=0);
|
---|
237 |
|
---|
238 | //
|
---|
239 | // Check that the diagonal matrix D is nonsingular.
|
---|
240 | //
|
---|
241 | rcond = 0;
|
---|
242 | if( isupper )
|
---|
243 | {
|
---|
244 | for(i=n; i>=1; i--)
|
---|
245 | {
|
---|
246 | if( pivots[i]>0 & (double)(l[i,i])==(double)(0) )
|
---|
247 | {
|
---|
248 | return;
|
---|
249 | }
|
---|
250 | }
|
---|
251 | }
|
---|
252 | else
|
---|
253 | {
|
---|
254 | for(i=1; i<=n; i++)
|
---|
255 | {
|
---|
256 | if( pivots[i]>0 & (double)(l[i,i])==(double)(0) )
|
---|
257 | {
|
---|
258 | return;
|
---|
259 | }
|
---|
260 | }
|
---|
261 | }
|
---|
262 |
|
---|
263 | //
|
---|
264 | // Estimate the norm of A.
|
---|
265 | //
|
---|
266 | if( !isnormprovided )
|
---|
267 | {
|
---|
268 | kase = 0;
|
---|
269 | anorm = 0;
|
---|
270 | while( true )
|
---|
271 | {
|
---|
272 | estnorm.iterativeestimate1norm(n, ref work1, ref work0, ref iwork, ref anorm, ref kase);
|
---|
273 | if( kase==0 )
|
---|
274 | {
|
---|
275 | break;
|
---|
276 | }
|
---|
277 | if( isupper )
|
---|
278 | {
|
---|
279 |
|
---|
280 | //
|
---|
281 | // Multiply by U'
|
---|
282 | //
|
---|
283 | k = n;
|
---|
284 | while( k>=1 )
|
---|
285 | {
|
---|
286 | if( pivots[k]>0 )
|
---|
287 | {
|
---|
288 |
|
---|
289 | //
|
---|
290 | // P(k)
|
---|
291 | //
|
---|
292 | v = work0[k];
|
---|
293 | work0[k] = work0[pivots[k]];
|
---|
294 | work0[pivots[k]] = v;
|
---|
295 |
|
---|
296 | //
|
---|
297 | // U(k)
|
---|
298 | //
|
---|
299 | km1 = k-1;
|
---|
300 | v = 0.0;
|
---|
301 | for(i_=1; i_<=km1;i_++)
|
---|
302 | {
|
---|
303 | v += work0[i_]*l[i_,k];
|
---|
304 | }
|
---|
305 | work0[k] = work0[k]+v;
|
---|
306 |
|
---|
307 | //
|
---|
308 | // Next k
|
---|
309 | //
|
---|
310 | k = k-1;
|
---|
311 | }
|
---|
312 | else
|
---|
313 | {
|
---|
314 |
|
---|
315 | //
|
---|
316 | // P(k)
|
---|
317 | //
|
---|
318 | v = work0[k-1];
|
---|
319 | work0[k-1] = work0[-pivots[k-1]];
|
---|
320 | work0[-pivots[k-1]] = v;
|
---|
321 |
|
---|
322 | //
|
---|
323 | // U(k)
|
---|
324 | //
|
---|
325 | km1 = k-1;
|
---|
326 | km2 = k-2;
|
---|
327 | v = 0.0;
|
---|
328 | for(i_=1; i_<=km2;i_++)
|
---|
329 | {
|
---|
330 | v += work0[i_]*l[i_,km1];
|
---|
331 | }
|
---|
332 | work0[km1] = work0[km1]+v;
|
---|
333 | v = 0.0;
|
---|
334 | for(i_=1; i_<=km2;i_++)
|
---|
335 | {
|
---|
336 | v += work0[i_]*l[i_,k];
|
---|
337 | }
|
---|
338 | work0[k] = work0[k]+v;
|
---|
339 |
|
---|
340 | //
|
---|
341 | // Next k
|
---|
342 | //
|
---|
343 | k = k-2;
|
---|
344 | }
|
---|
345 | }
|
---|
346 |
|
---|
347 | //
|
---|
348 | // Multiply by D
|
---|
349 | //
|
---|
350 | k = n;
|
---|
351 | while( k>=1 )
|
---|
352 | {
|
---|
353 | if( pivots[k]>0 )
|
---|
354 | {
|
---|
355 | work0[k] = work0[k]*l[k,k];
|
---|
356 | k = k-1;
|
---|
357 | }
|
---|
358 | else
|
---|
359 | {
|
---|
360 | v = work0[k-1];
|
---|
361 | work0[k-1] = l[k-1,k-1]*work0[k-1]+l[k-1,k]*work0[k];
|
---|
362 | work0[k] = l[k-1,k]*v+l[k,k]*work0[k];
|
---|
363 | k = k-2;
|
---|
364 | }
|
---|
365 | }
|
---|
366 |
|
---|
367 | //
|
---|
368 | // Multiply by U
|
---|
369 | //
|
---|
370 | k = 1;
|
---|
371 | while( k<=n )
|
---|
372 | {
|
---|
373 | if( pivots[k]>0 )
|
---|
374 | {
|
---|
375 |
|
---|
376 | //
|
---|
377 | // U(k)
|
---|
378 | //
|
---|
379 | km1 = k-1;
|
---|
380 | v = work0[k];
|
---|
381 | for(i_=1; i_<=km1;i_++)
|
---|
382 | {
|
---|
383 | work0[i_] = work0[i_] + v*l[i_,k];
|
---|
384 | }
|
---|
385 |
|
---|
386 | //
|
---|
387 | // P(k)
|
---|
388 | //
|
---|
389 | v = work0[k];
|
---|
390 | work0[k] = work0[pivots[k]];
|
---|
391 | work0[pivots[k]] = v;
|
---|
392 |
|
---|
393 | //
|
---|
394 | // Next k
|
---|
395 | //
|
---|
396 | k = k+1;
|
---|
397 | }
|
---|
398 | else
|
---|
399 | {
|
---|
400 |
|
---|
401 | //
|
---|
402 | // U(k)
|
---|
403 | //
|
---|
404 | km1 = k-1;
|
---|
405 | kp1 = k+1;
|
---|
406 | v = work0[k];
|
---|
407 | for(i_=1; i_<=km1;i_++)
|
---|
408 | {
|
---|
409 | work0[i_] = work0[i_] + v*l[i_,k];
|
---|
410 | }
|
---|
411 | v = work0[kp1];
|
---|
412 | for(i_=1; i_<=km1;i_++)
|
---|
413 | {
|
---|
414 | work0[i_] = work0[i_] + v*l[i_,kp1];
|
---|
415 | }
|
---|
416 |
|
---|
417 | //
|
---|
418 | // P(k)
|
---|
419 | //
|
---|
420 | v = work0[k];
|
---|
421 | work0[k] = work0[-pivots[k]];
|
---|
422 | work0[-pivots[k]] = v;
|
---|
423 |
|
---|
424 | //
|
---|
425 | // Next k
|
---|
426 | //
|
---|
427 | k = k+2;
|
---|
428 | }
|
---|
429 | }
|
---|
430 | }
|
---|
431 | else
|
---|
432 | {
|
---|
433 |
|
---|
434 | //
|
---|
435 | // Multiply by L'
|
---|
436 | //
|
---|
437 | k = 1;
|
---|
438 | while( k<=n )
|
---|
439 | {
|
---|
440 | if( pivots[k]>0 )
|
---|
441 | {
|
---|
442 |
|
---|
443 | //
|
---|
444 | // P(k)
|
---|
445 | //
|
---|
446 | v = work0[k];
|
---|
447 | work0[k] = work0[pivots[k]];
|
---|
448 | work0[pivots[k]] = v;
|
---|
449 |
|
---|
450 | //
|
---|
451 | // L(k)
|
---|
452 | //
|
---|
453 | kp1 = k+1;
|
---|
454 | v = 0.0;
|
---|
455 | for(i_=kp1; i_<=n;i_++)
|
---|
456 | {
|
---|
457 | v += work0[i_]*l[i_,k];
|
---|
458 | }
|
---|
459 | work0[k] = work0[k]+v;
|
---|
460 |
|
---|
461 | //
|
---|
462 | // Next k
|
---|
463 | //
|
---|
464 | k = k+1;
|
---|
465 | }
|
---|
466 | else
|
---|
467 | {
|
---|
468 |
|
---|
469 | //
|
---|
470 | // P(k)
|
---|
471 | //
|
---|
472 | v = work0[k+1];
|
---|
473 | work0[k+1] = work0[-pivots[k+1]];
|
---|
474 | work0[-pivots[k+1]] = v;
|
---|
475 |
|
---|
476 | //
|
---|
477 | // L(k)
|
---|
478 | //
|
---|
479 | kp1 = k+1;
|
---|
480 | kp2 = k+2;
|
---|
481 | v = 0.0;
|
---|
482 | for(i_=kp2; i_<=n;i_++)
|
---|
483 | {
|
---|
484 | v += work0[i_]*l[i_,k];
|
---|
485 | }
|
---|
486 | work0[k] = work0[k]+v;
|
---|
487 | v = 0.0;
|
---|
488 | for(i_=kp2; i_<=n;i_++)
|
---|
489 | {
|
---|
490 | v += work0[i_]*l[i_,kp1];
|
---|
491 | }
|
---|
492 | work0[kp1] = work0[kp1]+v;
|
---|
493 |
|
---|
494 | //
|
---|
495 | // Next k
|
---|
496 | //
|
---|
497 | k = k+2;
|
---|
498 | }
|
---|
499 | }
|
---|
500 |
|
---|
501 | //
|
---|
502 | // Multiply by D
|
---|
503 | //
|
---|
504 | k = n;
|
---|
505 | while( k>=1 )
|
---|
506 | {
|
---|
507 | if( pivots[k]>0 )
|
---|
508 | {
|
---|
509 | work0[k] = work0[k]*l[k,k];
|
---|
510 | k = k-1;
|
---|
511 | }
|
---|
512 | else
|
---|
513 | {
|
---|
514 | v = work0[k-1];
|
---|
515 | work0[k-1] = l[k-1,k-1]*work0[k-1]+l[k,k-1]*work0[k];
|
---|
516 | work0[k] = l[k,k-1]*v+l[k,k]*work0[k];
|
---|
517 | k = k-2;
|
---|
518 | }
|
---|
519 | }
|
---|
520 |
|
---|
521 | //
|
---|
522 | // Multiply by L
|
---|
523 | //
|
---|
524 | k = n;
|
---|
525 | while( k>=1 )
|
---|
526 | {
|
---|
527 | if( pivots[k]>0 )
|
---|
528 | {
|
---|
529 |
|
---|
530 | //
|
---|
531 | // L(k)
|
---|
532 | //
|
---|
533 | kp1 = k+1;
|
---|
534 | v = work0[k];
|
---|
535 | for(i_=kp1; i_<=n;i_++)
|
---|
536 | {
|
---|
537 | work0[i_] = work0[i_] + v*l[i_,k];
|
---|
538 | }
|
---|
539 |
|
---|
540 | //
|
---|
541 | // P(k)
|
---|
542 | //
|
---|
543 | v = work0[k];
|
---|
544 | work0[k] = work0[pivots[k]];
|
---|
545 | work0[pivots[k]] = v;
|
---|
546 |
|
---|
547 | //
|
---|
548 | // Next k
|
---|
549 | //
|
---|
550 | k = k-1;
|
---|
551 | }
|
---|
552 | else
|
---|
553 | {
|
---|
554 |
|
---|
555 | //
|
---|
556 | // L(k)
|
---|
557 | //
|
---|
558 | kp1 = k+1;
|
---|
559 | km1 = k-1;
|
---|
560 | v = work0[k];
|
---|
561 | for(i_=kp1; i_<=n;i_++)
|
---|
562 | {
|
---|
563 | work0[i_] = work0[i_] + v*l[i_,k];
|
---|
564 | }
|
---|
565 | v = work0[km1];
|
---|
566 | for(i_=kp1; i_<=n;i_++)
|
---|
567 | {
|
---|
568 | work0[i_] = work0[i_] + v*l[i_,km1];
|
---|
569 | }
|
---|
570 |
|
---|
571 | //
|
---|
572 | // P(k)
|
---|
573 | //
|
---|
574 | v = work0[k];
|
---|
575 | work0[k] = work0[-pivots[k]];
|
---|
576 | work0[-pivots[k]] = v;
|
---|
577 |
|
---|
578 | //
|
---|
579 | // Next k
|
---|
580 | //
|
---|
581 | k = k-2;
|
---|
582 | }
|
---|
583 | }
|
---|
584 | }
|
---|
585 | }
|
---|
586 | }
|
---|
587 |
|
---|
588 | //
|
---|
589 | // Quick return if possible
|
---|
590 | //
|
---|
591 | rcond = 0;
|
---|
592 | if( n==0 )
|
---|
593 | {
|
---|
594 | rcond = 1;
|
---|
595 | return;
|
---|
596 | }
|
---|
597 | if( (double)(anorm)==(double)(0) )
|
---|
598 | {
|
---|
599 | return;
|
---|
600 | }
|
---|
601 |
|
---|
602 | //
|
---|
603 | // Estimate the 1-norm of inv(A).
|
---|
604 | //
|
---|
605 | kase = 0;
|
---|
606 | while( true )
|
---|
607 | {
|
---|
608 | estnorm.iterativeestimate1norm(n, ref work1, ref work0, ref iwork, ref ainvnm, ref kase);
|
---|
609 | if( kase==0 )
|
---|
610 | {
|
---|
611 | break;
|
---|
612 | }
|
---|
613 | ssolve.solvesystemldlt(ref l, ref pivots, work0, n, isupper, ref work2);
|
---|
614 | for(i_=1; i_<=n;i_++)
|
---|
615 | {
|
---|
616 | work0[i_] = work2[i_];
|
---|
617 | }
|
---|
618 | }
|
---|
619 |
|
---|
620 | //
|
---|
621 | // Compute the estimate of the reciprocal condition number.
|
---|
622 | //
|
---|
623 | if( (double)(ainvnm)!=(double)(0) )
|
---|
624 | {
|
---|
625 | v = 1/ainvnm;
|
---|
626 | rcond = v/anorm;
|
---|
627 | }
|
---|
628 | }
|
---|
629 | }
|
---|
630 | }
|
---|