lu_normest.m
3.51 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
function rho = lu_normest (A, L, U)
%LU_NORMEST estimates norm (L*U-A, 1) without forming L*U-A
%
% Example:
%
% rho = lu_normest (A, L, U)
%
% which estimates the computation of the 1-norm:
%
% rho = norm (A-L*U, 1)
%
% Authors: William W. Hager, Math Dept., Univ. of Florida
% Timothy A. Davis, CISE Dept., Univ. of Florida
% Gainesville, FL, 32611, USA.
% based on normest1, contributed on November, 1997
%
% This code can be quite easily adapted to estimate the 1-norm of any
% matrix E, where E itself is dense or not explicitly represented, but the
% computation of E (and E') times a vector is easy. In this case, our matrix
% of interest is:
%
% E = A-L*U
%
% That is, L*U is the LU factorization of A, where A, L and U
% are sparse. This code works for dense matrices A and L too,
% but it would not be needed in that case, since E is easy to compute
% explicitly. For sparse A, L, and U, computing E explicitly would be quite
% expensive, and thus normest (A-L*U) would be prohibitive.
%
% For a detailed description, see Davis, T. A. and Hager, W. W.,
% Modifying a sparse Cholesky factorization, SIAM J. Matrix Analysis and
% Applications, 1999, vol. 20, no. 3, 606-627.
%
% See also normest
% The three places that the matrix-vector multiply E*x is used are highlighted.
% Note that E is never formed explicity.
% Copyright 1995-2007 by William W. Hager and Timothy A. Davis
[m n] = size (A) ;
if (m ~= n)
% pad A, L, and U with zeros so that they are all square
if (m < n)
U = [ U ; (sparse (n-m,n)) ] ;
L = [ L , (sparse (m,n-m)) ; (sparse (n-m,n)) ] ;
A = [ A ; (sparse (n-m,n)) ] ;
else
U = [ U , (sparse (n,m-n)) ; (sparse (m-n,m)) ] ;
L = [ L , (sparse (m,m-n)) ] ;
A = [ A , (sparse (m,m-n)) ] ;
end
end
[m n] = size (A) ; %#ok
notvisited = ones (m, 1) ; % nonvisited(j) is zero if j is visited, 1 otherwise
rho = 0 ; % the global rho
for trial = 1:3 % {
x = notvisited ./ sum (notvisited) ;
rho1 = 0 ; % the current rho for this trial
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% COMPUTE Ex1 = E*x EFFICIENTLY: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Ex1 = (A*x) - L*(U*x) ;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
rho2 = norm (Ex1, 1) ;
while rho2 > rho1 % {
rho1 = rho2 ;
y = 2*(Ex1 >= 0) - 1 ;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% COMPUTE z = E'*y EFFICIENTLY: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
z = (A'*y) - U'*(L'*y) ;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[zj, j] = max (abs (z .* notvisited)) ;
j = j (1) ;
if (abs (z (j)) > z'*x) % {
x = zeros (m, 1) ;
x (j) = 1 ;
notvisited (j) = 0 ;
else % } {
break ;
end % }
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% COMPUTE Ex1 = E*x EFFICIENTLY: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Ex1 = (A*x) - L*(U*x) ;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
rho2 = norm (Ex1, 1) ;
end % }
rho = max (rho, rho1) ;
end % }