-
Notifications
You must be signed in to change notification settings - Fork 0
/
sChebyshev.tex
306 lines (273 loc) · 9.93 KB
/
sChebyshev.tex
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
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
% Options for packages loaded elsewhere
\PassOptionsToPackage{unicode}{hyperref}
\PassOptionsToPackage{hyphens}{url}
%
\documentclass[
12pt,
]{article}
\usepackage{amsmath,amssymb}
\usepackage{iftex}
\ifPDFTeX
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{textcomp} % provide euro and other symbols
\else % if luatex or xetex
\usepackage{unicode-math} % this also loads fontspec
\defaultfontfeatures{Scale=MatchLowercase}
\defaultfontfeatures[\rmfamily]{Ligatures=TeX,Scale=1}
\fi
\usepackage{lmodern}
\ifPDFTeX\else
% xetex/luatex font selection
\setmainfont[]{Times New Roman}
\fi
% Use upquote if available, for straight quotes in verbatim environments
\IfFileExists{upquote.sty}{\usepackage{upquote}}{}
\IfFileExists{microtype.sty}{% use microtype if available
\usepackage[]{microtype}
\UseMicrotypeSet[protrusion]{basicmath} % disable protrusion for tt fonts
}{}
\makeatletter
\@ifundefined{KOMAClassName}{% if non-KOMA class
\IfFileExists{parskip.sty}{%
\usepackage{parskip}
}{% else
\setlength{\parindent}{0pt}
\setlength{\parskip}{6pt plus 2pt minus 1pt}}
}{% if KOMA class
\KOMAoptions{parskip=half}}
\makeatother
\usepackage{xcolor}
\usepackage[margin=1in]{geometry}
\usepackage{longtable,booktabs,array}
\usepackage{calc} % for calculating minipage widths
% Correct order of tables after \paragraph or \subparagraph
\usepackage{etoolbox}
\makeatletter
\patchcmd\longtable{\par}{\if@noskipsec\mbox{}\fi\par}{}{}
\makeatother
% Allow footnotes in longtable head/foot
\IfFileExists{footnotehyper.sty}{\usepackage{footnotehyper}}{\usepackage{footnote}}
\makesavenoteenv{longtable}
\usepackage{graphicx}
\makeatletter
\def\maxwidth{\ifdim\Gin@nat@width>\linewidth\linewidth\else\Gin@nat@width\fi}
\def\maxheight{\ifdim\Gin@nat@height>\textheight\textheight\else\Gin@nat@height\fi}
\makeatother
% Scale images if necessary, so that they will not overflow the page
% margins by default, and it is still possible to overwrite the defaults
% using explicit options in \includegraphics[width, height, ...]{}
\setkeys{Gin}{width=\maxwidth,height=\maxheight,keepaspectratio}
% Set default figure placement to htbp
\makeatletter
\def\fps@figure{htbp}
\makeatother
\setlength{\emergencystretch}{3em} % prevent overfull lines
\providecommand{\tightlist}{%
\setlength{\itemsep}{0pt}\setlength{\parskip}{0pt}}
\setcounter{secnumdepth}{5}
\usepackage{tcolorbox}
\usepackage{amssymb}
\usepackage{yfonts}
\usepackage{bm}
\newtcolorbox{greybox}{
colback=white,
colframe=blue,
coltext=black,
boxsep=5pt,
arc=4pt}
\newcommand{\ds}[4]{\sum_{{#1}=1}^{#3}\sum_{{#2}=1}^{#4}}
\newcommand{\us}[3]{\mathop{\sum\sum}_{1\leq{#2}<{#1}\leq{#3}}}
\newcommand{\ol}[1]{\overline{#1}}
\newcommand{\ul}[1]{\underline{#1}}
\newcommand{\amin}[1]{\mathop{\text{argmin}}_{#1}}
\newcommand{\amax}[1]{\mathop{\text{argmax}}_{#1}}
\newcommand{\ci}{\perp\!\!\!\perp}
\newcommand{\mc}[1]{\mathcal{#1}}
\newcommand{\mb}[1]{\mathbb{#1}}
\newcommand{\mf}[1]{\mathfrak{#1}}
\newcommand{\eps}{\epsilon}
\newcommand{\lbd}{\lambda}
\newcommand{\alp}{\alpha}
\newcommand{\df}{=:}
\newcommand{\am}[1]{\mathop{\text{argmin}}_{#1}}
\newcommand{\ls}[2]{\mathop{\sum\sum}_{#1}^{#2}}
\newcommand{\ijs}{\mathop{\sum\sum}_{1\leq i<j\leq n}}
\newcommand{\jis}{\mathop{\sum\sum}_{1\leq j<i\leq n}}
\newcommand{\sij}{\sum_{i=1}^n\sum_{j=1}^n}
\ifLuaTeX
\usepackage{selnolig} % disable illegal ligatures
\fi
\IfFileExists{bookmark.sty}{\usepackage{bookmark}}{\usepackage{hyperref}}
\IfFileExists{xurl.sty}{\usepackage{xurl}}{} % add URL line breaks if available
\urlstyle{same}
\hypersetup{
pdftitle={Smacof Meets Chebyshev},
pdfauthor={Jan de Leeuw - University of California Los Angeles},
hidelinks,
pdfcreator={LaTeX via pandoc}}
\title{Smacof Meets Chebyshev}
\author{Jan de Leeuw - University of California Los Angeles}
\date{Started October 08 2023, Version of October 09, 2023}
\begin{document}
\maketitle
{
\setcounter{tocdepth}{4}
\tableofcontents
}
\textbf{Note:} This is a working paper which will be expanded/updated frequently. All suggestions for improvement are welcome. The Rmd file, the pdf, all R files, a LaTeX version and so on are available at \url{https://github.com/deleeuw/sChebyshev}.
\section*{Notation}\label{notation}
\addcontentsline{toc}{section}{Notation}
\subsection*{Conventions}\label{conventions}
\addcontentsline{toc}{subsection}{Conventions}
Since we only work in finite dimensional vector spaces, and since our emphasis is on computation, we adopt the following conventions.
\begin{itemize}
\tightlist
\item
A vector \emph{is} a matrix with one column.
\item
A row-vector \emph{is} a matrix with one row.
\item
Derivatives \emph{are} matrices.
\end{itemize}
\subsection*{Notations}\label{notations}
\addcontentsline{toc}{subsection}{Notations}
The length of vectors and the dimension of matrices will generally
be clear from the context.
\begin{itemize}
\tightlist
\item
\(e_i\quad\) unit vector (element \(i\) is one, other elements zero).
\item
\(e\quad\) vector with all elements one.
\item
\(E\quad\) matrix with all elements one.
\item
\(0\quad\) real number zero, also vector or matrix with all elements \(0\).
\item
\(I\quad\) identity matrix.
\item
\(J=I-\frac{ee'}{e'e}\quad\) centering matrix.
\item
\(A\otimes B\quad\) Kronecker product of matrices \(A\) and \(B\).
\item
\(X\oplus Y\quad\) direct sum of matrices \(X\) and \(Y\).
\item
\(X\times Y\quad\) elementwise (Hadamard) product of matrices \(X\) and \(Y\).
\item
\(\text{vec}(X)\quad\) matrix \(X\) to vector (columns on top of each other).
\item
\(\text{vecr}(X)\quad\) elements below diagonal of matrix \(X\) to vector (columns on top of each other).
\item
\(X'\quad\) transpose of matrix \(X\).
\item
\(X^+\quad\) Moore-Penrose inverse of matrix \(X\).
\item
\(X^{-T}\quad\) inverse of the transpose \(X'\) (and transpose of the inverse).
\item
\(X\gtrsim Y\quad\) Loewner order of symmetric matrices (\(X-Y\) is positive semi-definite).
\item
\(X\lesssim Y\quad\) Loewner order of symmetric matrices (\(Y-X\) is positive semi-definite).
\item
\(:=\quad\) definition.
\item
\(X\times Y\quad\) Cartesian product of sets \(X\) and \(Y\).
\item
\((x,y)\quad\) is an ordered pair, i.e.~an element of \(X\times Y\).
\item
\(x_{is}\) or \(\{X\}_{is}\quad\) element \((i,s)\) of matrix \(X\).
\item
\(a_{\bullet s}\quad\) column \(s\) of matrix \(A\).
\item
\(a_{i\bullet\quad}\) row \(i\) of matrix \(A\).
\item
\([A]_{is}\quad\) submatrix \((i,s)\) of block-matrix \(A\).
\item
\(A^{[p]}\quad\) direct sum of \(p\) copies of matrix \(A\).
\item
\(a^{(k)}\quad\) the \(k^{th}\) element of the sequence \(\{a\}=a^{(1)},\cdots,a^{(k)},\cdots\).
\item
\(\mathbb{R}^n\quad\) space of all vectors of length \(n\).
\item
\(\overline{\mathbb{R}}^n\quad\) space of all centered vectors of length \(n\) (i.e.~\(x'e=0\)).
\item
\(\mathbb{R}^{n\times p}\quad\) space of all \(n\times p\) matrices.
\item
\(\overline{\mathbb{R}}^{n\times p}\quad\) space of all column-centered \(n\times p\) matrices (i.e.~with \(X'e=0\)).
\item
\(f:X\Rightarrow Y\quad\) function with arguments in \(X\) and values in \(Y\).
\item
If \(f:X\times Y\Rightarrow Z\) then \(f(\bullet,y):X\Rightarrow Z\)
and \(f(x,\bullet):Y\Rightarrow Z\).
\item
\(x'y\quad\) inner product in \(\mathbb{R}^n\).
\item
\(\text{tr}\ X'Y\quad\) inner product in \(\mathbb{R}^{n\times p}\).
\item
\(\|x\|=\sqrt{x'x}\quad\) Euclidean norm of \(x\in\mathbb{R}^n\).
\item
\(\|X\|=\sqrt{\text{tr}\ X'X}\quad\) Euclidean norm of \(X\in\mathbb{R}^{n\times p}\).
\item
\(\mathcal{D}f(x)\quad\) derivative of \(f\) at \(x\).
\item
\(\mathcal{D}^2f(x)\quad\) second derivative of \(f\) at \(x\).
\item
\(\mathcal{D}_sf(x)=\{\mathcal{D}f(x)\}_s\quad\) partial derivative with respect to \(x_s\) at \(x\).
\item
\(\mathcal{D}_{st}f(x)=\{\mathcal{D}^2f(x)\}_{st}\quad\) second partial with respect to \(x_s\) and \(x_t\) at \(x\).
\end{itemize}
\section{Introduction}\label{introduction}
In this paper we study techniques to speed up the basic smacof iteration
\[
X^{(k+1)}=\Gamma(X^{(k)})
\]
with \(\Gamma\) the Guttman transform, by using updates of the form
\[
X^{(k+1)}=\sum_{r=1}^s\alpha_r\Gamma^r(X^{(k)})
\]
with \(\Gamma^0(X)=X\) and \(\Gamma^r(X)=\Gamma(\Gamma^{r-1}(X))\)
\subsection{Matrix Basis}\label{matrix-basis}
An important special case of DCDD imposes the constraint
\begin{equation}
X=\sum_{v=1}^r\theta_vG_v,
\label{eq:matbasis}
\end{equation}
where the \(G_s\) are \(n\times p\) matrices. To see that this is indeed a special case
of DCDD define \(Y_s\) as the matrix collecting the \(s^{th}\) columns of all \(G_v\). Thus
there are \(r\) of these \(n\times r\) matrices \(Y_s\). Now \(\vec(X)=Y\theta\), with \(Y\)
the direct sum of the \(Y_s\), as usual, and
\begin{equation}
\theta=\left.\begin{bmatrix}\theta\\\vdots\\\theta\end{bmatrix}\right\}r\ \text{times}.
\label{eq:thetamat}
\end{equation}
The distinguishing DCDD characteristic in using the \emph{matrix basis} \eqref{eq:matbasis} is that all \(r\) pieces of \(\theta\) in \eqref{eq:thetamat} must be equal.
Better in configuration space
No \(V=I\)
\begin{verbatim}
One important application of the matrix basis is finding the optimal step size in
\end{verbatim}
iterative procedures, or, more generally, finding optimal weights in multistep
procedures. For the steepest descent method, for example, we choose \(G_1=X\)
and \(G_2=\nabla\sigma(X)\).
\[
\{C_{ij}\}_{vw}=\text{tr}\ G_v'A_{ij}G_w
\]
Thus
\[
\{V_s\}_{vw}=\text{tr}\ G_v'V_0G_w
\]
\[
\{B_s(\theta)\}_{vw}=\text{tr}\ G_v'B_0(\theta)G_w
\]
\[\sigma(\theta):=\frac12\{1-2\theta'B(\theta)\theta+\theta'V\theta\}\]
\subsection{Matrix Based}\label{matrix-based}
\[
X^{(k+1)}=\theta_1 X^{(k)}+\theta_2\Gamma(X^{(k)})+\theta_3\Gamma^2(X^{(k)})+\cdots+\theta_r\Gamma^{r-1}(X^{(k)})
\]
Problem: near a fixed point \(B\) and \(V\) become almost singular (of rank one)
Chebyshev: \[\min_\theta\max_s|\theta_1+\theta_2\lambda_s+\cdots+\theta_r\lambda_s^{r-1}|\]
\begin{verbatim}
Sidi
\end{verbatim}
Generalizes relax.
\end{document}