\documentclass{article}
\title{Connected Sets}
\author{Matt McCutchen}
\date{April 10, 2005}

\usepackage{amssymb}

% Page layout parameters
% Don't ask me why these settings work, but apparently they do.
% Horizontal...
\paperwidth=8.5in
\textwidth=7.5in
\oddsidemargin=-.5in
\evensidemargin=-.5in
% Vertical...
\paperheight=11in
\textheight=9.5in
\topmargin=-.5in
\headheight=0in
\headsep=0in

\def\term#1{\textbf{#1}}

\begin{document}

\maketitle

\tableofcontents

\bigskip
Note: In this document, a juxtaposition of two points denotes the distance between them.

\section{Kinds of connectedness}

\textbf{Unsplittable.}  Two point sets $T_1$ and $T_2$ \term{split} a point set $S$ if the following conditions all hold:
\begin{itemize}
\item $T_1$ and $T_2$ are disjoint open sets.
\item $T_1$ and $T_2$ both intersect $S$ ($T_1 \cap S \not= \emptyset$ and $T_2 \cap S \not= \emptyset$).
\item $T_1$ and $T_2$ together cover $S$ ($S \subseteq T_1 \cup T_2$).
\end{itemize}
A point set $S$ is \term{unsplittable} if there do not exist sets $T_1$ and $T_2$ that split it.  According to several sources, when mathematicians say ``connected'', they usually mean this definition.

\bigskip\noindent
\textbf{Arcwise connected.}  A point set $S$ is \term{arcwise connected} if, for all $a, b \in S$, there exists a continuous curve lying entirely in $S$ that connects $a$ and $b$.  That is, there is a continuous function $f : [0,1] \to S$ with $f(0) = a$ and $f(1) = b$.

\bigskip\noindent
\textbf{Capacitor-connected.}  A point set $S$ is \term{capacitor-connected} if, for all $a, b \in S$ and all positive real numbers $\delta$, there exists a sequence of points $a_0, \ldots, a_n \in S$ with $a_0 = a$, $a_n = b$, and $a_i a_{i+1} < \delta$ for $i = 0, \ldots, n-1$.

\section{All unsplittable sets are capacitor-connected}

Let $S$ be an unsplittable point set, let $a \in S$, and let $\delta$ be a positive real number.  Generate a sequence of point sets $A_0, A_1, \ldots$ as follows: $A_0$ = $\{a\}$, and for all $i \ge 0$,
\[A_{i+1} = S \cap \bigcup_{x \in A_i} Ball_\delta(x).\]
Let $A^* = \bigcup_{i \ge 0} A_i$.

Now, let
\[U = \bigcup_{x \in A^*} Ball_{\delta/2}(x)\quad\textrm{and}\quad V = \bigcup_{x \in S \setminus A^*} Ball_{\delta/2}(x).\]

Some observations about the sets $U$ and $V$:
\begin{itemize}
\item $U$ and $V$ are both open because they are both unions of open balls.
\item Together they cover $S$ because $A^* \subseteq U$ and $S \setminus A^* \subseteq V$.
\item They are disjoint.  To prove this, suppose there were a point $x$ in both sets.  By the definition of $U$, there must be a point $x_U \in A^*$ with $xx_U < \delta/2$, and since $x_U \in A^*$, $x_U \in A_n$ for some $n$.  By the definition of $V$, there must also be a point $x_V \in S \setminus A^*$ with $xx_V < \delta/2$.  But
\[x_Ux_V \le xx_U + xx_V < \delta/2 + \delta/2 = \delta,\]
so $x_V \in Ball_\delta(x_U) \subseteq A_{n+1}$ by the definition of the $A_i$.  But then $x_V \in A^*$, contradicting the fact that $x_V \in S \setminus A^*$.
\item $U \cap S$ is nonempty (it contains $a$).
\end{itemize}

If $V \cap S$ were also nonempty, $U$ and $V$ would split $S$, but we know $S$ is unsplittable, so $V$ doesn't intersect $S$.  This means $S \setminus A^* = \emptyset$, because if there existed $x \in S \setminus A^*$, we'd have $x \in V \cap S$.  Thus $A^* = S$.

By the definition of $A^*$, any point $b \in S$ must be in $A_n$ for some $n$.  Generate a sequence of points $b_n, \ldots, b_0$ as follows: $b_n = b$, and $b_{i-1} \in A_{i-1}$ with $b_i b_{i-1} < \delta$ for $i = n, \ldots, 1$.  A $b_i$ must exist at every step because of the way the $A_i$ were constructed, and $b_0 = a$ because $a$ is the only point in $A_0$.

The points $a\!=\!b_0, \ b_1, \ldots, \ b_n\!=\!b$ form a sequence of hops in $S$ from $a$ to $b$ of lengths less than $\delta$.  The foregoing argument can be used to create such a sequence for all $a, b \in S$ and all positive real numbers $\delta$, so $S$ is capacitor-connected.  $\blacksquare$

\section{All compact, capacitor-connected sets are unsplittable}

Let $S$ be a compact, capacitor-connected set split by sets $T_1$ and $T_2$; we'll prove the section title by reaching a contradiction.

\bigskip
\textbf{Lemma.} \textsl{Let $O$ and $C$ be open and compact point sets, respectively.  If $O \supseteq C$, there exists a positive real number $\delta$ such that for all points $x \in C$ and $y \notin O$, $xy \ge \delta$.}

\medskip
\textsl{Proof:} For $i = 0, 1, \ldots$, let $O_i$ be the set of all points $x$ such that $Ball_{2^{-i}}(x) \subseteq O$.  Clearly $O_i \subseteq O_j$ when $i \le j$.  If $x$ is any point in $O$, since $O$ is open, there exists a positive real number $r$ such that $Ball_r(x) \subseteq O$, so $x \in O_i$ where $i = max(0, \lceil -log_2 r \rceil)$; thus $\bigcup_i O_i = O$.

The $O_i$ form an infinite set of open sets that together cover $O$ and therefore $C$, so by the Heine-Borel Theorem, there also exists a finite set $U$ of $O_i$ whose members together cover $C$.  Let $n$ be the highest index of any of the sets $O_i$ in $U$.  Since $O_n \supseteq O_i$ for all $i \le n$, $O_n = \bigcup U$ and therefore $O_n \supseteq C$.

Let $\delta = 2^{-n}$.  Let $x$ be any point in $C$.  Since $C \subseteq O_n$, the definition of $O_n$ tells us that $Ball_{2^{-n}}(x) \subseteq O$.  Let $y$ be any point not in $O$.  Since $Ball_{2^{-n}}(x) \subseteq O$, $y \notin Ball_{2^{-n}}(x)$, so $xy \ge 2^{-n} = \delta$.  $\Box$

\bigskip
Applying Lemma 1 with $O \mapsto T_1 \cup T_2$ and $C \mapsto S$, we find that any point $x$ in $S$ is always at least a fixed distance $\delta$ from any point $y$ not in either $T_i$.  I claim now that any point $x$ in $S \cap T_1$ is further than $\delta$ from any point $y \notin T_1$; it is enough to prove this for $y \in T_2$.

Let $f : xy \to \mathbb{R}^n$ parametrize the line segment connecting $x$ and $y$ in the obvious fashion with $f(0) = x$ and $f(xy) = y$.  Clearly $f(t_1)f(t_2) = |t_1 - t_2|$ for all $t_1, t_2 \in [0,xy]$.  Let $P$ be the set of all $t \in [0,xy]$ for which $f(t) \in T_1$.

$P$ is nonempty (it contains $0$) and has an upper bound ($xy$), so by the completeness axiom, it has a supremum $s$.  That is, $t \notin P$ for all $t > s$, and there exists no $s' < s$ for which $t \notin P$ for all $t > s'$.  The second part implies that, for all $s' < s$, there exists $t \in (s',s]$ with $t \in P$.  I now claim that $f(s)$ is a point in neither $T_1$ nor $T_2$.

If $f(s)$ were in $T_1$, then, since $T_1$ is open, there would exist a positive real number $\epsilon$ such that $Ball_\epsilon(f(s)) \subseteq T_1$.  In particular, $f(s+\epsilon/2)$ would be in $T_1$, so $s + \epsilon/2 \in P$, contradicting the last paragraph.  If $f(s)$ were in $T_2$, there would again exist a positive real number $\epsilon$ such that $Ball_\epsilon(f(s)) \subseteq T_2$.  Setting $s' = s - \epsilon$, we find that there is a number $t \in (s',s]$ with $t \in P$, which means $f(t) \in T_1$.  But $f(s)f(t) < \epsilon$, so $f(t) \in T_2$, contradicting the fact that $T_1$ and $T_2$ are disjoint.

We know that $x \in S$ and $f(s) \in \mathbb{R}^n \setminus (T_1 \cup T_2)$, so Lemma 1 tells us that $xf(s) \ge \delta$.  $f(s)$ is strictly between $x$ and $y$ on a line segment, so $xy > xf(s) \ge \delta$, proving the claim that $xy \ge \delta$ for all $x \in S \cap T_1$ and $y \notin T_1$.

Since $T_1$ and $T_2$ split $S$, there exist points $a \in S \cap T_1$ and $b \in S \cap T_2$.  Since $S$ is capacitor-connected, there must exist points $a_0, \ldots, a_k \in S$ with $a_0 = a$, $a_k = b$, and $a_i a_{i+1} < \delta$ for $i = 0, \ldots, n-1$.  Let $j$ be the smallest index for which $a_j \in T_1$ but $a_{j+1} \notin T_1$; some such $j$ must exist because $a_0 \in T_1$ and $a_k \notin T_1$.  Applying the claim, we find that $a_j a_{j+1} > \delta$, contradicting the validity of the sequence of hops $a_i$.  $\blacksquare$

\end{document}