This file is indexed.

/usr/share/doc/pyxplot/html/sect0243.html is in pyxplot-doc 0.8.4-3.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below.

  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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
<!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" xml:lang="en" lang="en">
<head>
<meta name="generator" content="plasTeX" />
<meta content="text/html; charset=utf-8" http-equiv="content-type" />
<title>PyXPlot Users' Guide: Estimating the Error in </title>

<link href="sect0244.html" title="The Covariance Matrix" rel="next" />
<link href="sec-bayes_pdf.html" title="The Probability Density Function" rel="prev" />
<link href="ch-fit_maths.html" title="The fit Command: Mathematical Details" rel="up" />
<link rel="stylesheet" href="styles/styles.css" />
</head>
<body>

<div class="navigation">
<table cellspacing="2" cellpadding="0" width="100%">
<tr>
<td><a href="sec-bayes_pdf.html" title="The Probability Density Function"><img alt="Previous: The Probability Density Function" border="0" src="icons/previous.gif" width="32" height="32" /></a></td>

<td><a href="ch-fit_maths.html" title="The fit Command: Mathematical Details"><img alt="Up: The fit Command: Mathematical Details" border="0" src="icons/up.gif" width="32" height="32" /></a></td>

<td><a href="sect0244.html" title="The Covariance Matrix"><img alt="Next: The Covariance Matrix" border="0" src="icons/next.gif" width="32" height="32" /></a></td>

<td class="navtitle" align="center">PyXPlot Users' Guide</td>
<td><a href="index.html" title="Table of Contents"><img border="0" alt="" src="icons/contents.gif" width="32" height="32" /></a></td>

<td><a href="sect0255.html" title="Index"><img border="0" alt="" src="icons/index.gif" width="32" height="32" /></a></td>

<td><img border="0" alt="" src="icons/blank.gif" width="32" height="32" /></td>
</tr>
</table>
</div>

<div class="breadcrumbs">
<span>
<span>
<a href="index.html">PyXPlot Users' Guide</a> <b>:</b>
</span>

</span><span>
<span>
<a href="sect0230.html">Appendices</a> <b>:</b>
</span>

</span><span>
<span>
<a href="ch-fit_maths.html">The <tt class="tt">fit</tt> Command: Mathematical Details</a> <b>:</b>
</span>

</span><span>

<span>
<b class="current">Estimating the Error in <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" /></b>
</span>
</span>
<hr />
</div>

<div><h1 id="a0000000244">C.3 Estimating the Error in <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" /></h1>
<p>To estimate the error in the best-fitting parameter values that we find, we assume <img src="images/img-0798.png" alt="$\mathrm{P}\left( \mathbf{u} | \left\{  \mathbf{x}_ i, f_ i, \sigma _ i \right\}  \right)$" style="vertical-align:-5px; 
                                     width:129px; 
                                     height:19px" class="math gen" /> to be approximated by an <img src="images/img-0794.png" alt="$n_\mathrm {u}$" style="vertical-align:-2px; 
                                     width:19px; 
                                     height:10px" class="math gen" />-dimensional Gaussian distribution around <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" />. Taking a Taylor expansion of <img src="images/img-0805.png" alt="$L(\mathbf{u})$" style="vertical-align:-4px; 
                                     width:37px; 
                                     height:18px" class="math gen" /> about <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" />, we can write: </p><table id="&lt;plasTeX.TeXFragment object at 0xb216cec&gt;" cellpadding="7" width="100%" cellspacing="0" class="eqnarray">
<tr id="a0000001670">
    
    <td style="width:40%">&nbsp;</td>
    
    
        <td style="vertical-align:middle;                                    text-align:right"><img src="images/img-0806.png" alt="$\displaystyle  L(\mathbf{u})  $" style="vertical-align:-4px; width:37px;                     height:18px" class="math gen" /></td>
    
    
    
        <td style="vertical-align:middle;                                    text-align:center"><img src="images/img-0058.png" alt="$\displaystyle  =  $" style="vertical-align:2px; width:12px;                     height:4px" class="math gen" /></td>
    
    
    
        <td style="vertical-align:middle;                                    text-align:left"><img src="images/img-0807.png" alt="$\displaystyle  L(\mathbf{u}^0) + \underbrace{ \sum _{i=0}^{n_\mathrm {u}-1} \left( u_ i - u^0_ i \right) \left.\frac{\partial L}{\partial u_ i}\right|_{\mathbf{u}^0} }_{\textrm{Zero at $\mathbf{u}^0$ by definition}} + \label{eqa:L_ taylor_ expand} $" style="vertical-align:-53px; width:253px;                     height:85px" class="math gen" /></td>
    
    
    
    <td style="width:40%">&nbsp;</td>
    <td style="width:20%" class="eqnnum"><span>(<span>C.4</span>)</span></td>
</tr><tr id="a0000001671">
    
    <td style="width:40%">&nbsp;</td>
    
    
        <td style="vertical-align:middle;                                    text-align:right"><img src="images/img-0175.png" alt="$\displaystyle  $" style="vertical-align:0px; width:1px;                     height:1px" class="math gen" /></td>
    
    
    
        <td style="vertical-align:middle;                                    text-align:center"><img src="images/img-0175.png" alt="$\displaystyle  $" style="vertical-align:0px; width:1px;                     height:1px" class="math gen" /></td>
    
    
    
        <td style="vertical-align:middle;                                    text-align:left"><img src="images/img-0808.png" alt="$\displaystyle  \sum _{i=0}^{n_\mathrm {u}-1} \sum _{j=0}^{n_\mathrm {u}-1} \frac{\left( u_ i - u^0_ i \right) \left( u_ j - u^0_ j \right)}{2} \left.\frac{\partial ^2 L}{\partial u_ i \partial u_ j}\right|_{\mathbf{u}^0} + \mathcal{O}\left( \mathbf{u} - \mathbf{u}^0\right)^3 \nonumber  $" style="vertical-align:-25px; width:438px;                     height:65px" class="math gen" /></td>
    
    
    
    <td style="width:40%">&nbsp;</td>
    <td style="width:20%" class="eqnnum">&nbsp;</td>
</tr>
</table><p>Since the logarithm of a Gaussian distribution is a parabola, the quadratic terms in the above expansion encode the Gaussian component of the probability distribution <img src="images/img-0798.png" alt="$\mathrm{P}\left( \mathbf{u} | \left\{  \mathbf{x}_ i, f_ i, \sigma _ i \right\}  \right)$" style="vertical-align:-5px; 
                                     width:129px; 
                                     height:19px" class="math gen" /> about <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" />.<a href="#a0000001672" class="footnote"><sup class="footnotemark">1</sup></a> We may write the sum of these terms, which we denote <img src="images/img-0809.png" alt="$Q$" style="vertical-align:-4px; 
                                     width:14px; 
                                     height:16px" class="math gen" />, in matrix form: </p><table id="&lt;plasTeX.TeXFragment object at 0xb22d9bc&gt;" class="equation" width="100%" cellspacing="0" cellpadding="7">
<tr>
    
    <td style="width:40%">&nbsp;</td>
    <td><img src="images/img-0810.png" alt="\begin{equation}  Q = \frac{1}{2} \left(\mathbf{u} - \mathbf{u}^0\right)^\mathbf {T} \mathbf{A} \left(\mathbf{u} - \mathbf{u}^0\right) \label{eqn:Q_ vector} \end{equation}" style="width:412px; 
                            height:37px" class="math gen" /></td>
    
    <td style="width:40%">&nbsp;</td>
    <td class="eqnnum" style="width:20%"><span>(<span>C.5</span>)</span></td>
</tr>
</table><p>where the superscript <img src="images/img-0811.png" alt="$^\mathbf {T}$" style="vertical-align:7px; 
                                     width:11px; 
                                     height:9px" class="math gen" /> represents the transpose of the vector displacement from <img src="images/img-0804.png" alt="$\mathbf{u}^0$" style="vertical-align:0px; 
                                     width:19px; 
                                     height:16px" class="math gen" />, and <img src="images/img-0812.png" alt="$\mathbf{A}$" style="vertical-align:0px; 
                                     width:15px; 
                                     height:12px" class="math gen" /> is the Hessian matrix of <img src="images/img-0378.png" alt="$L$" style="vertical-align:0px; 
                                     width:12px; 
                                     height:12px" class="math gen" />, given by: </p><table id="a0000001673" class="equation" width="100%" cellspacing="0" cellpadding="7">
<tr>
    
    <td style="width:40%">&nbsp;</td>
    <td><img src="images/img-0813.png" alt="\begin{equation}  A_{ij} = \nabla \nabla L = \left.\frac{\partial ^2 L}{\partial u_ i \partial u_ j}\right|_{\mathbf{u}^0} \end{equation}" style="width:399px; 
                            height:46px" class="math gen" /></td>
    
    <td style="width:40%">&nbsp;</td>
    <td class="eqnnum" style="width:20%"><span>(<span>C.6</span>)</span></td>
</tr>
</table><p> <a name="a0000001674" id="a0000001674"></a> </p><p>This is the Hessian matrix which is output by the <tt class="tt">fit</tt> command. In general, an <img src="images/img-0794.png" alt="$n_\mathrm {u}$" style="vertical-align:-2px; 
                                     width:19px; 
                                     height:10px" class="math gen" />-dimensional Gaussian distribution such as that given by Equation (<a></a>) yields elliptical contours of equi-probability in parameter space, whose principal axes need not be aligned with our chosen coordinate axes – the variables <img src="images/img-0814.png" alt="$u_0 ... u_{n_ u-1}$" style="vertical-align:-4px; 
                                     width:77px; 
                                     height:12px" class="math gen" />. The eigenvectors <img src="images/img-0815.png" alt="$\mathbf{e}_ i$" style="vertical-align:-2px; 
                                     width:14px; 
                                     height:10px" class="math gen" /> of <img src="images/img-0812.png" alt="$\mathbf{A}$" style="vertical-align:0px; 
                                     width:15px; 
                                     height:12px" class="math gen" /> are the principal axes of these ellipses, and the corresponding eigenvalues <img src="images/img-0816.png" alt="$\lambda _ i$" style="vertical-align:-2px; 
                                     width:14px; 
                                     height:15px" class="math gen" /> equal <img src="images/img-0817.png" alt="$1/\sigma _ i^2$" style="vertical-align:-5px; 
                                     width:35px; 
                                     height:21px" class="math gen" />, where <img src="images/img-0792.png" alt="$\sigma _ i$" style="vertical-align:-2px; 
                                     width:14px; 
                                     height:10px" class="math gen" /> is the standard deviation of the probability density function along the direction of these axes. </p><p>This can be visualised by imagining that we diagonalise <img src="images/img-0812.png" alt="$\mathbf{A}$" style="vertical-align:0px; 
                                     width:15px; 
                                     height:12px" class="math gen" />, and expand Equation (<a></a>) in our diagonal basis. The resulting expression for <img src="images/img-0378.png" alt="$L$" style="vertical-align:0px; 
                                     width:12px; 
                                     height:12px" class="math gen" /> is a sum of square terms; the cross terms vanish in this basis by definition. The equations of the equi-probability contours become the equations of ellipses: </p><table id="a0000001675" class="equation" width="100%" cellspacing="0" cellpadding="7">
<tr>
    
    <td style="width:40%">&nbsp;</td>
    <td><img src="images/img-0818.png" alt="\begin{equation}  Q = \frac{1}{2} \sum _{i=0}^{n_\mathrm {u}-1} A_{ii} \left(u_ i - u^0_ i\right)^2 = k \end{equation}" style="width:414px; 
                            height:53px" class="math gen" /></td>
    
    <td style="width:40%">&nbsp;</td>
    <td class="eqnnum" style="width:20%"><span>(<span>C.7</span>)</span></td>
</tr>
</table><p>where <img src="images/img-0651.png" alt="$k$" style="vertical-align:0px; 
                                     width:8px; 
                                     height:13px" class="math gen" /> is some constant. By comparison with the equation for the logarithm of a Gaussian distribution, we can associate <img src="images/img-0819.png" alt="$A_{ii}$" style="vertical-align:-2px; 
                                     width:23px; 
                                     height:14px" class="math gen" /> with <img src="images/img-0820.png" alt="$-1/\sigma _ i^2$" style="vertical-align:-5px; 
                                     width:49px; 
                                     height:21px" class="math gen" /> in our eigenvector basis. </p><p>The problem of evaluating the standard deviations of our variables <img src="images/img-0821.png" alt="$u_ i$" style="vertical-align:-2px; 
                                     width:14px; 
                                     height:10px" class="math gen" /> is more complicated, however, as we are attempting to evaluate the width of these elliptical equi-probability contours in directions which are, in general, not aligned with their principal axes. To achieve this, we first convert our Hessian matrix into a covariance matrix. </p></div>



<div id="footnotes">
<p><b>Footnotes</b></p>
<ol>
<li id="a0000001672">The use of this is called <i class="itshape">Gauss’ Method</i>. Higher order terms in the expansion represent any non-Gaussianity in the probability distribution, which we neglect. See MacKay, D.J.C., <i class="itshape">Information Theory, Inference and Learning Algorithms</i>, CUP (2003).</li>
</ol>
</div>

<div class="navigation">
<table cellspacing="2" cellpadding="0" width="100%">
<tr>
<td><a href="sec-bayes_pdf.html" title="The Probability Density Function"><img alt="Previous: The Probability Density Function" border="0" src="icons/previous.gif" width="32" height="32" /></a></td>

<td><a href="ch-fit_maths.html" title="The fit Command: Mathematical Details"><img alt="Up: The fit Command: Mathematical Details" border="0" src="icons/up.gif" width="32" height="32" /></a></td>

<td><a href="sect0244.html" title="The Covariance Matrix"><img alt="Next: The Covariance Matrix" border="0" src="icons/next.gif" width="32" height="32" /></a></td>

<td class="navtitle" align="center">PyXPlot Users' Guide</td>
<td><a href="index.html" title="Table of Contents"><img border="0" alt="" src="icons/contents.gif" width="32" height="32" /></a></td>

<td><a href="sect0255.html" title="Index"><img border="0" alt="" src="icons/index.gif" width="32" height="32" /></a></td>

<td><img border="0" alt="" src="icons/blank.gif" width="32" height="32" /></td>
</tr>
</table>
</div>

<script language="javascript" src="icons/imgadjust.js" type="text/javascript"></script>

</body>
</html>