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
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
|
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
<meta http-equiv="X-UA-Compatible" content="IE=9"/>
<title>FreeMat: LEFTDIVIDE Matrix Equation Solver/Divide Operator</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
$(document).ready(initResizable);
</script>
<link href="doxygen.css" rel="stylesheet" type="text/css" />
</head>
<body>
<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
<div id="titlearea">
<table cellspacing="0" cellpadding="0">
<tbody>
<tr style="height: 56px;">
<td style="padding-left: 0.5em;">
<div id="projectname">FreeMat
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.1.1 -->
<div id="navrow1" class="tabs">
<ul class="tablist">
<li><a href="index.html"><span>Main Page</span></a></li>
<li class="current"><a href="pages.html"><span>Related Pages</span></a></li>
</ul>
</div>
</div><!-- top -->
<div id="side-nav" class="ui-resizable side-nav-resizable">
<div id="nav-tree">
<div id="nav-tree-contents">
</div>
</div>
<div id="splitbar" style="-moz-user-select:none;"
class="ui-resizable-handle">
</div>
</div>
<script type="text/javascript">
$(document).ready(function(){initNavTree('operators_leftdivide.html','');});
</script>
<div id="doc-content">
<div class="header">
<div class="headertitle">
<div class="title">LEFTDIVIDE Matrix Equation Solver/Divide Operator </div> </div>
</div><!--header-->
<div class="contents">
<div class="textblock"><p>Section: <a class="el" href="sec_operators.html">Mathematical Operators</a> </p>
<h1><a class="anchor" id="Usage"></a>
Usage</h1>
<p>The divide operator <code></tt> is really a combination of three operators, all of which have the same general syntax: </p>
<pre class="fragment"> Y = A \ B
</pre><p> where <code>A</code> and <code>B</code> are arrays of numerical type. The result <code>Y</code> depends on which of the following three situations applies to the arguments <code>A</code> and <code>B</code>: </p>
<ol>
<li>
<code>A</code> is a scalar, <code>B</code> is an arbitrary <code>n</code>-dimensional numerical array, in which case the output is each element of <code>B</code> divided by the scalar <code>A</code>. </li>
<li>
<code>A,B</code> are matrices with the same number of rows, i.e., <code>A</code> is of size <code>M x K</code>, and <code>B</code> is of size <code>M x L</code>, in which case the output is of size <code>K x L</code>. </li>
</ol>
<p>The output follows the standard type promotion rules, although in the first two cases, if <code>A</code> and <code>B</code> are integers, the output is an integer also, while in the third case if <code>A</code> and <code>B</code> are integers, the output is of type <code>double</code>.</code></p>
<p><code>A few additional words about the third version, in which <code>A</code> and <code>B</code> are matrices. Very loosely speaking, <code>Y</code> is the matrix that satisfies <img class="formulaInl" alt="$A * Y = B$" src="form_109.png"/>. In cases where such a matrix exists. If such a matrix does not exist, then a matrix <code>Y</code> is returned that approximates <img class="formulaInl" alt="$A * Y \approx B$" src="form_110.png"/>. </code></p>
<h1><a class="anchor" id="Function"></a>
Internals</h1>
<p><code>There are three formulae for the times operator. For the first form </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ Y(m_1,\ldots,m_d) = \frac{B(m_1,\ldots,m_d)}{A}. \]" src="form_111.png"/>
</p>
<p> In the second form, the calculation of the output depends on the size of <code>A</code>. Because each column of <code>B</code> is treated independantly, we can rewrite the equation <img class="formulaInl" alt="$A Y = B$" src="form_112.png"/> as </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ A [y_1, y_2,\ldots, y_l] = [b_1, b_2, \ldots, b_l] \]" src="form_113.png"/>
</p>
<p> where <code>y_i</code> are the columns of <code>Y</code>, and <code>b_i</code> are the columns of the matrix <code>B</code>. If <code>A</code> is a square matrix, then the LAPACK routine <code>*gesvx</code> (where the <code>*</code> is replaced with <code>sdcz</code> depending on the type of the arguments) is used, which uses an LU decomposition of <code>A</code> to solve the sequence of equations sequentially. If <code>A</code> is singular, then a warning is emitted.</code></p>
<p><code>On the other hand, if <code>A</code> is rectangular, then the LAPACK routine <code>*gelsy</code> is used. Note that these routines are designed to work with matrices <code>A</code> that are full rank - either full column rank or full row rank. If <code>A</code> fails to satisfy this assumption, a warning is emitted. If <code>A</code> has full column rank (and thus necessarily has more rows than columns), then theoretically, this operator finds the columns <code>y_i</code> that satisfy: </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ y_i = \arg \min_y \| A y - b_i \|_2 \]" src="form_114.png"/>
</p>
<p> and each column is thus the Least Squares solution of <code>A y = b_i</code>. On the other hand, if <code>A</code> has full row rank (and thus necessarily has more columns than rows), then theoretically, this operator finds the columns <code>y_i</code> that satisfy </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ y_i = \arg \min_{A y = b_i} \| y \|_2 \]" src="form_115.png"/>
</p>
<p> and each column is thus the Minimum Norm vector <code>y_i</code> that satisfies <img class="formulaInl" alt="$A y_i = b_i$" src="form_116.png"/>. In the event that the matrix <code>A</code> is neither full row rank nor full column rank, a solution is returned, that is the minimum norm least squares solution. The solution is computed using an orthogonal factorization technique that is documented in the LAPACK User's Guide (see the References section for details). </code></p>
<h1><a class="anchor" id="Examples"></a>
Examples</h1>
<p><code>Here are some simple examples of the divide operator. We start with a simple example of a full rank, square matrix:</code></p>
<p><code></p>
<pre class="fragment">--> A = [1,1;0,1]
A =
1 1
0 1
</pre><p></code></p>
<p><code>Suppose we wish to solve </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ \begin{bmatrix} 1 & 1 \\ 0 & 1 \end{bmatrix} \begin{bmatrix} y_1 \\ y_2 \end{bmatrix} = \begin{bmatrix} 3 \\ 2 \end{bmatrix} \]" src="form_117.png"/>
</p>
<p> (which by inspection has the solution <code>y_1 = 1</code>, <code>y_2 = 2</code>). Thus we compute:</code></p>
<p><code></p>
<pre class="fragment">--> B = [3;2]
B =
3
2
--> Y = A\B
Y =
1
2
</pre><p></code></p>
<p><code>Suppose we wish to solve a trivial Least Squares (LS) problem. We want to find a simple scaling of the vector <code>[1;1]</code> that is closest to the point <code>[2,1]</code>. This is equivalent to solving </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ \begin{bmatrix} 1 \\ 1 \end{bmatrix} y = \begin{bmatrix} 2 \\ 1 \end{bmatrix} \]" src="form_118.png"/>
</p>
<p> in a least squares sense. For fun, we can calculate the solution using calculus by hand. The error we wish to minimize is </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ \varepsilon(y) = (y - 2)^2 + (y-1)^2. \]" src="form_119.png"/>
</p>
<p> Taking a derivative with respect to <code>y</code>, and setting to zero (which we must have for an extrema when <code>y</code> is unconstrained) </p>
<p class="formulaDsp">
<img class="formulaDsp" alt="\[ 2 (y-2) + 2 (y-1) = 0 \]" src="form_120.png"/>
</p>
<p> which we can simplify to <code>4y = 6</code> or <code>y = 3/2</code> (we must, technically, check to make sure this is a minimum, and not a maximum or an inflection point). Here is the same calculation performed using FreeMat:</code></p>
<p><code></p>
<pre class="fragment">--> A = [1;1]
A =
1
1
--> B = [2;1]
B =
2
1
--> A\B
ans =
1.5000
</pre><p></code></p>
<p><code>which is the same solution. </code></p>
</div></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="navelem"><a class="el" href="index.html">FreeMat Documentation</a></li><li class="navelem"><a class="el" href="sec_operators.html">Mathematical Operators</a></li>
<li class="footer">Generated on Thu Jul 25 2013 17:17:45 for FreeMat by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.1.1 </li>
</ul>
</div>
</body>
</html>
|
