source: trunk/MagicSoft/slalib/svdsol.c

Last change on this file was 732, checked in by tbretz, 24 years ago
*** empty log message ***
  • Property svn:executable set to *
File size: 3.9 KB
Line 
1#include "slalib.h"
2#include "slamac.h"
3void slaSvdsol ( int m, int n, int mp, int np, double *b, double *u,
4 double *w, double *v, double *work, double *x )
5/*
6** - - - - - - - - - -
7** s l a S v d s o l
8** - - - - - - - - - -
9**
10** From a given vector and the SVD of a matrix (as obtained from
11** the slaSvd routine), obtain the solution vector.
12**
13** (double precision)
14**
15** This routine solves the equation:
16**
17** a . x = b
18**
19** where:
20**
21** a is a given m (rows) x n (columns) matrix, where m.ge.n
22** x is the n-vector we wish to find
23** b is a given m-vector
24**
25** By means of the singular value decomposition method (SVD). In
26** this method, the matrix a is first factorized (for example by
27** the routine slaSvd) into the following components:
28**
29** a = u x w x vt
30**
31** where:
32**
33** a is the m (rows) x n (columns) matrix
34** u is an m x n column-orthogonal matrix
35** w is an n x n diagonal matrix with w(i,i).ge.0
36** vt is the transpose of an nxn orthogonal matrix
37**
38** Note that m and n, above, are the logical dimensions of the
39** matrices and vectors concerned, which can be located in
40** arrays of larger physical dimensions mp and np.
41**
42** The solution is found from the expression:
43**
44** x = v . [diag(1/wj)] . ( transpose(u) . b )
45**
46** Notes:
47**
48** 1) If matrix a is square, and if the diagonal matrix w is not
49** adjusted, the method is equivalent to conventional solution
50** of simultaneous equations.
51**
52** 2) If m>n, the result is a least-squares fit.
53**
54** 3) If the solution is poorly determined, this shows up in the
55** SVD factorization as very small or zero wj values. Where
56** a wj value is small but non-zero it can be set to zero to
57** avoid ill effects. The present routine detects such zero
58** wj values and produces a sensible solution, with highly
59** correlated terms kept under control rather than being allowed
60** to elope to infinity, and with meaningful values for the
61** other terms.
62**
63** Given:
64** m,n int numbers of rows and columns in matrix a
65** mp,np int physical dimensions of array containing matrix a
66** *b double[m] known vector b
67** *u double[mp][np] array containing mxn matrix u
68** *w double[n] nxn diagonal matrix w (diagonal elements only)
69** *v double[np][np] array containing nxn orthogonal matrix v
70**
71** Returned:
72** *work double[n] workspace
73** *x double[n] unknown vector x
74**
75** Note: If the relative sizes of m, n, mp and np are inconsistent,
76** the vector x is returned unaltered. This condition should
77** have been detected when the SVD was performed using slaSvd.
78**
79** Reference:
80** Numerical Recipes, Section 2.9.
81**
82** Example call (note handling of "adjustable dimension" 2D arrays):
83**
84** double a[MP][NP], w[NP], v[NP][NP], work[NP], b[MP], x[NP];
85** int m, n;
86** :
87** slaSvdsol ( m, n, MP, NP, b, (double *) a, w, (double *) v, work, x );
88**
89** Last revision: 20 February 1995
90**
91** Copyright P.T.Wallace. All rights reserved.
92*/
93{
94 int j, i, jj;
95 double s;
96 double *ui;
97 double *vj;
98
99/* Check that the matrix is the right size and shape */
100 if ( m >= n && m <= mp && n <= np ) {
101
102 /* Calculate [diag(1/wj)] . transpose(u) . b (or zero for zero wj) */
103 for ( j = 0; j < n; j++ ) {
104 s = 0.0;
105 if ( w[j] != 0.0 ) {
106 for ( i = 0, ui = u;
107 i < m;
108 i++, ui += np ) {
109 s += ui[j] * b[i];
110 }
111 s /= w[j];
112 }
113 work[j] = s;
114 }
115
116 /* Multiply by matrix v to get result */
117 for ( j = 0, vj = v;
118 j < n;
119 j++, vj += np ) {
120 s = 0.0;
121 for ( jj = 0; jj < n; jj++ ) {
122 s += vj[jj] * work[jj];
123 }
124 x[j] = s;
125 }
126 }
127}
Note: See TracBrowser for help on using the repository browser.