Thue-Morse sequence and groups of intermediate growth
Extracted source text · arXiv 2605.30605
\documentclass[12pt]{article}
%
\usepackage{amsmath,amssymb,amsthm}
\usepackage{graphicx}
\usepackage{hyperref}
%
\textwidth=6.5in
\oddsidemargin=0.2in
\textheight=8.8in
\topmargin=-0.6in
%
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{problem}[theorem]{Problem}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
%
\def\bC{\mathbb{C}}
\def\bN{\mathbb{N}}
\def\bQ{\mathbb{Q}}
\def\bR{\mathbb{R}}
\def\bZ{\mathbb{Z}}
\def\cA{\mathcal{A}}
\def\cB{\mathcal{B}}
\def\cD{\mathcal{D}}
\def\cG{\mathcal{G}}
\def\cH{\mathcal{H}}
\def\cI{\mathcal{I}}
\def\cK{\mathcal{K}}
\def\cL{\mathcal{L}}
\def\cT{\mathcal{T}}
\def\al{\alpha}
\def\be{\beta}
\def\ga{\gamma}
\def\de{\delta}
\def\eps{\varepsilon}
\def\la{\lambda}
\def\om{\omega}
\def\si{\sigma}
\def\La{\Lambda}
\def\Om{\Omega}
\def\dT{\partial\mathcal{T}}
\def\St{\mathrm{St}}
%
\def\subst{\mathsf{s}}
\def\TM{\subst_{\mathrm{TM}}}
\def\pd{\subst_{\mathrm{pd}}}
\def\sub{\subst_{\mathrm{Mu}}}
\def\xiTM{\xi_{\mathrm{TM}}}
\def\xipd{\xi_{\mathrm{pd}}}
\def\xiGE{\xi_{\mathrm{Mu}}}
\def\GE{\mathrm{Mu}}
\def\Homeo{\mathop{\mathrm{Homeo}}}
\def\Sch{\mathop{\mathrm{Sch}}}
\def\id{\mathrm{id}}
\def\0{\mathbf{0}}
%
\title{Thue-Morse sequence and groups of intermediate growth}
\author{Rostislav Grigorchuk \and Yaroslav Vorobets}
\date{}
%
\begin{document}
\maketitle
\begin{abstract}
We consider the substitution subshift generated by the Thue-Morse
substitution $0\to01$, $1\to10$. We prove that the topological full group
of the subshift contains a subgroup of intermediate growth. Namely, one
group from the family known as the Grigorchuk groups embeds into this group.
To obtain our main result, we prove an embedding theorem for topological
full groups, and also develop a technique to prove isomorphism of groups
using the Schreier graphs.
\end{abstract}
%
\section{Introduction}\label{sect-intro}
There are many topics where the interests of theory of dynamical systems
and algebra converge. The interconnection between these two mathematical
disciplines goes both ways. Theory of dynamical systems is looking for
algebraic invariants to use in various classification problems. Besides, a
lot of dynamical systems are of algebraic origin (see \cite{Schmidt,
LindSchmidt}). On the other hand, algebra (first of all, group theory)
uses ideas and methods of theory of dynamical systems for new constructions
and for solving old problems as demonstrated, for instance, in the recent
book by V. Nekrashevych \cite{Nek2022}.
This paper is centered around one specific topic on the intersection of
theory of dynamical systems with group theory, namely, the topological full
groups associated with Cantor minimal systems.
The \emph{Cantor minimal systems} are classical objects of study in
topological dynamics. Such a system $(X,T)$ consists of a Cantor set $X$
and a homeomorphism $T:X\to X$, which defines a continuous action of the
infinite cyclic group $\bZ$ on $X$. The action is to be minimal in the
sense that it allows no closed invariant set other than the empty set and
$X$. We are interested in systems coming from symbolic dynamics. Recall
that the two-sided \emph{shift} over a finite alphabet $\cA$ (of at least
two symbols) is the transformation $\si$ of the space $\cA^{\bZ}$ of
bi-infinite sequences of symbols from $\cA$ that acts by the rule
$\si(\ldots x_{-2}x_{-1}x_0.x_1x_2\ldots)=\ldots
y_{-2}y_{-1}y_0.y_1y_2\ldots$, where $y_n=x_{n+1}$ for all $n\in\bZ$.
A \emph{subshift} is the restriction of $\si$ to a closed shift-invariant
set $\Om\subset\cA^{\bZ}$. Any subshift is uniquely determined by the set
of all forbidden words (i.e., finite strings of symbols that must not occur
in elements of $\Om$) or, equivalently, by the set of all admissible
words. Two major classes of subshifts are subshifts of finite type and
minimal subshifts. They represent two different kinds of dynamic behavior
(chaotic vs.\@ regular). A subshift of finite type is determined by
specifying a finite number of forbidden words. The minimal subshifts are
not so easy to construct. The main source of examples are the substitution
subshifts generated by primitive substitutions. A \emph{substitution} over
an alphabet $\cA$ is a procedure that replaces every symbol from the
alphabet with a certain word over $\cA$. Suppose that the substitution
preserves an infinite sequence $\xi=x_1x_2x_3\ldots$ of symbols from $\cA$,
and let $\bigl(\Om(\xi),\si\bigr)$ be the subshift for which the admissible
words are words that occur infinitely often in $\xi$. Then it is called a
\emph{substitution subshift}. The substitution is \emph{primitive} if,
starting with any symbol and applying the substitution several times, we
eventually end up with a word that includes all symbols in the alphabet.
If this is the case then the subshift $\bigl(\Om(\xi),\si\bigr)$ is
necessarily a Cantor minimal system (see, e.g., the books
\cite{AlloucheShallit:automatic, Fogg:substitutions, LindMarcus}).
The \emph{Thue-Morse} substitution $\TM$ over the binary alphabet
$\{0,1\}$, given by $0\to01$, $1\to10$, is among the simplest and most
famous substitutions that appear in many areas of mathematics. Starting
with $0$ and repeatedly applying the substitution $\TM$, we obtain a
sequence of finite binary strings $0,01,0110,01101001,\dots$ where each
string extends the previous one. This results in an infinite binary string
$\xiTM=01101001\dots$ invariant under the substitution rule. The string
$\xiTM$ is known as the \emph{Thue-Morse sequence}. Since the substitution
$\TM$ is primitive, the sequence $\xiTM$ generates a minimal subshift over
the binary alphabet $\{0,1\}$, called the \emph{Morse subshift}. One
instance where the sequence $\xiTM$ appears in algebra is a criterion of
nilpotency of a group in terms of semigroup identities, which uses the idea
of iteration in the spirit of the Thue-Morse substitution as shown
implicitly by A. Maltsev \cite{Maltsev} and repeated in an explicit form in
\cite{BoffaPoint}. This is also related to the so-called Zimin words (see,
e.g., the book \cite{Sapir}), which are used in theory of rings. L. Rowen
\cite{Rowen:Koethe} suggested that $\xiTM$ may be the key to solving one of
the most famous problems in theory of rings, the Koethe conjecture. For
many more applications of the Thue-Morse sequence in many different
contexts, see the survey \cite{AlloucheShallit:ubiquitous}.
\begin{figure}[t]
\centerline{\includegraphics[scale=0.8]{fig-automata.eps}}
\caption{The automaton computing $\xiTM$ and the automaton generating the
lamplighter group.}
\label{fig-automata}
\end{figure}
Here is another curious connection of the Thue-Morse sequence to group
theory. The sequence $\xiTM$ is $2$-automatic, which means there is an
automaton that, given a number $n$ in its binary form as input, computes
the $n$-th symbol of $\xiTM$. If the symbols are counted starting from
$0$, the task is done by a $2$-state automaton shown in Figure
\ref{fig-automata} on the left (see \cite{AlloucheShallit:automatic}).
It is an acceptor-like automaton in which the result is determined by the
terminal state. Now let us consider a transducer automaton with the same
transition machine shown in Figure \ref{fig-automata} on the right. This
automaton outputs one symbol per each symbol of input. Choosing one of the
states as initial, the automaton generates a transformation on the set of
binary strings. The transformation happens to be related to a Gray code.
Recall that a Gray code is an ordering of numbers $0,1,\dots,2^k-1$ such
that the binary representations of any two neighbors differ in a single
digit. Given a number $n$ as a $k$-digit binary input, the automaton
outputs the position of $n$ in the Gray code. The two states of the
automaton give rise to two transformations, which in turn generate a
transformation group. The group turns out to be the lamplighter group,
i.e., the wreath product of an infinite cyclic group with a group of order
$2$ (see Theorem 7 in \cite{Classification}).
The notion of a \emph{topological full group} (TFG) is relatively recent.
Such a group can be associated to any homeomorphism $T:X\to X$, where $X$
is a Cantor set. Informally, the TFG of $T$, denoted $[[T]]$, consists of
all homeomorphisms of $X$ that locally coincide with elements of the cyclic
group generated by $T$ (see Definition \ref{def-TFG}). After the paper
\cite{GPS} by T. Giordano, I. Putnam and C. Skau, the TFG emerged as an
algebraic invariant of topological dynamics of the system $(X,T)$ provided
it is a Cantor minimal system (see Theorem \ref{flip-conjugacy}).
The TFG's of Cantor minimal systems are groups with unique properties.
Already in \cite{GPS} it was shown that they are indicable (admit a
homomorphism onto $\bZ$). The systematic study of algebraic properties of
groups $[[T]]$ was spearheaded by H. Matui \cite{Matui} who showed that the
commutator subgroup $[[T]]'$ is simple and, in the case of a minimal
subshift, finitely generated. The studies were continued by Matui
\cite{Matui2}, K. Medynets and the first author \cite{GrigMed,GrigMed2}, K.
Juschenko and N. Monod \cite{JuschMonod}, and others. Any finite group and
the free abelian group of infinite rank embed into $[[T]]$. If the system
$(X,T)$ is not topologically conjugate to a subshift, then the group
$[[T]]$ is locally virtually abelian (i.e., every finitely generated
subgroup $H\subset [[T]]$ contains an abelian subgroup of finite index in
$H$). Otherwise the lamplighter group embeds into $[[T]]'$ (see
\cite{Matui2}). The latter implies exponential growth of the group
$[[T]]'$ as the lamplighter group contains a free semigroup on two
generators. Further remarkable properties of TFG's include non-elementary
amenability \cite{JuschMonod} and factorization of $[[T]]'$ into a product
of two locally finite subgroups \cite{Matui,GrigMed}. Neither $[[T]]$ nor
$[[T]]'$ can be finitely presented but an effective description of the
system $(X,T)$ may help to obtain an infinite presentation. For instance,
if $(X,T)$ is a subshift such that all admissible words form a recursive
language, then the commutator subgroup $[[T]]'$ allows a recursive
presentation and, as a consequence, has a decidable word problem (see
\cite{GrigMed2}). The maximal subgroups of TFG's were studied by the
authors in \cite{GrigVor24}.
N. Matte Bon \cite{MatteBon} observed that the TFG of a Cantor minimal
system may contain a subgroup of intermediate (faster than polynomial but
slower than exponential) growth. Namely, each group from a family
$\cG_\om$, $\om\in\{0,1,2\}^{\bN}$ constructed in \cite{Grig84} embeds into
the TFG of a minimal subshift. The growth of $\cG_\om$ is intermediate if
the sequence $\om$ is not eventually constant. The groups $\cG_\om$ have
many other unusual properties. They are branch, just-infinite groups, most
of them are torsion groups (that is, every element has finite order). Each
group acts transitively on levels of the binary rooted tree, and the
Schreier graphs of those actions have line-like shapes. This defines a
linear order on each level that gives rise to a Gray code (see
\cite{MatteBon}). The code is the same for all groups $\cG_\om$ and,
somehow, it is related to the Gray code computed by the automaton in Figure
\ref{fig-automata}. Each group $\cG_\om$ also acts naturally on the
boundary of the binary rooted tree, and the Schreier graphs associated with
the orbits of that action also have line-like shapes. The latter fact was
used in numerous articles for various purposes, in particular, to initiate
the study of a joint spectrum of self-similar groups (see \cite{BarGri00})
or to construct first examples of simple groups of intermediate growth
\cite{Nek16}. Linear structure of the Schreier graphs will be important in
this paper as well.
Among groups $\cG_\om$, a special role belongs to the groups
$\cG_{(012)^\infty}$ and $\cG_{(01)^\infty}$ associated to periodic
sequences $012012\dots$ and $010101\dots$. Unlike generic groups in the
family, these two are self-similar groups, namely, each of them naturally
embeds as a subgroup of finite index into the wreath product of itself with
the group of order $2$. The two groups share many properties but there are
also essential differences. For example, the first group is torsion while
the second one has elements of infinite order. Each maximal subgroup in
$\cG_{(012)^\infty}$ has finite index while $\cG_{(01)^\infty}$ admits
maximal subgroups of infinite index.
The Schreier graphs of the group $\cG_{(012)^\infty}$ were studied by the
second author in \cite{Vor12}. D. Lenz, T. Nagnibeda and the first author
observed in \cite{GLN17} that $\cG_{(012)^\infty}$ can be embedded into the
TFG of the substitution subshift generated by a substitution
$\subst_{\mathrm{Ly}}$ given by $a\to aca$, $b\to d$, $c\to b$, $d\to c$.
They used this fact to link the spectral problem for Schreier graphs of
$\cG_{(012)^\infty}$ to the spectral theory of random Schroedinger
operators. The substitution $\subst_{\mathrm{Ly}}$ is not primitive but,
in fact, the associated subshift coincides with the subshift generated by a
primitive substitution $a\to ac$, $b\to ad$, $c\to ab$, $d\to ac$ (see
\cite{AlloucheDQ:hidden}). Note that the TFG of this subshift is finitely
generated. An explicit generating set was found in \cite{Vor20}.
The substitution $\subst_{\mathrm{Ly}}$ occurs naturally in the study of
the group $\cG_{(012)^\infty}$. Indeed, the group is generated by four
elements denoted $a,b,c,d$ and the substitution $\subst_{\mathrm{Ly}}$
induces an injective endomorphism of $\cG_{(012)^\infty}$. This was first
observed by I. Lysenok \cite{Lys85} who obtained the following
presentation of the group $\cG_{(012)^\infty}$ by generators and relators:
\[
\bigl\langle a,b,c,d \mid 1=a^2=b^2=c^2=d^2=bcd
=\subst_{\mathrm{Ly}}^k\bigl((ad)^4\bigr)
=\subst_{\mathrm{Ly}}^k\bigl((adacac)^4\bigr), \ k=1,2,\dots \bigr\rangle.
\]
Recursive group presentations of this kind are called
\emph{$L$-presentations} by analogy with the $L$-systems that are used for
the study of formal languages and in some other areas of science. Note
that the above presentation is optimal in the sense that no relator is
redundant (as shown in \cite{Grig99}). Interestingly, this optimality is
preserved after the profinite completion of the group $\cG_{(012)^\infty}$
(as proved by M. G. Benli \cite{Benli}).
Remarkably, the group $\cG_{(01)^\infty}$ also admits an $L$-presentation
\[
\bigl\langle a,c,d \mid \sub^k(r)=1, \ r\in R, \ k\ge0 \bigr\rangle,
\]
where the set of basic relators $R$ is
\[
\{a^2,(cd)^2,(ad)^4,[d,ac]^2,[d,cac]^2,[d,acaca]^2,
[d,acacac]^2,[d,cacacac]^2\}
\]
and $\sub$ is the substitution given by $a\to aca$, $c\to d$, $d\to c$.
It was obtained by Y. Muntyan \cite{Muntyan} in his dissertation. It is
not known yet whether this presentation is optimal. We do not need the
presentation in this paper, but the substitution $\sub$ will be very
important in what follows. Just like $\subst_{\mathrm{Ly}}$, the
substitution $\sub$ is not primitive. However, as we will show in this
paper, the associated substitution subshift is topologically conjugate to
the subshift generated by the \emph{period doubling} substitution $\pd$:
$x\to xy$, $y\to xx$, which is primitive. Note that the substitution
subshifts generated by the substitutions $\subst_{\mathrm{Ly}}$ and $\sub$
belong to a family of Toeplitz subshifts studied in \cite{Vor10}.
The main result of this paper adds another item to the list of properties
of the group $\cG_{(01)^\infty}$ showing its connection to the classical
Thue-Morse sequence.
\begin{theorem}\label{main}
The group of intermediate growth $\cG_{(01)^\infty}$ embeds into the
topological full group of the substitution subshift generated by the
Thue-Morse substitution $\TM$ as well as into the TFG of the substitution
subshift generated by the period doubling substitution $\pd$.
\end{theorem}
The embeddings in Theorem \ref{main} can be described explicitly (see
Proposition \ref{two-more-inj} at the end of the paper).
The proof of Theorem \ref{main} relies on the following general result.
\begin{theorem}\label{main0}
Let $(X_1,T_1)$ and $(X_2,T_2)$ be aperiodic Cantor systems. If
$(X_2,T_2)$ is a continuous factor of the system $(X_1,T_1)$ then the TFG
$[[T_2]]$ embeds into $[[T_1]]$.
\end{theorem}
Given a factor map $f:X_1\to X_2$, the induced embedding
$\mathcal{E}_f:[[T_2]]\to[[T_1]]$ in Theorem \ref{main0} can
be described explicitly. Namely, every homeomorphism $S_2\in[[T_2]]$ is
given by $S_2(y)=T_2^{\nu(y)}(y)$, $y\in X_2$ for some continuous function
$\nu:X_2\to\bZ$. Then the homeomorphism $S_1=\mathcal{E}_f(S_2)$ is given
by $S_1(x)=T_1^{\nu(f(x))}(x)$, $x\in X_1$.
Theorem \ref{main} inspires a number of questions and topics to explore.
We would like to formulate some of them.
\begin{problem}
Suppose $T_{\subst}$ is a substitution subshift generated by a primitive
substitution $\subst$.
\begin{itemize}
\item[(i)]
Under what condition on the substitution $\subst$ the topological full
group $[[T_{\subst}]]$ of the subshift $T_{\subst}$ contains a group of
intermediate growth?
\item[(ii)]
Under what conditions on $\subst$ the group $[[T_{\subst}]]$ contains a
subgroup of Burnside type (i.e., finitely generated infinite torsion group)?
\end{itemize}
In particular, what can be said about the TFG of the subshift generated by
the Fibonacci substitution $0\to01$, $1\to0$? Or the Rudin-Shapiro
substitution $a\to ab$, $b\to ac$, $c\to db$, $d\to dc$? Or the Chacon
substitution $0\to 0012$, $1\to 12$, $2\to 012$?
\end{problem}
There are other remarkable groups that have finite $L$-presentations. One
example is the iterated monodromy group (IMG) of the polynomial $z^2+i$.
This group first appeared in \cite{BarGrigNek} and then in many other
articles including \cite{GrigSavSun} where a finite $L$-presentation for
$\mathrm{IMG}(z^2+i)$ was found. It is quite similar to the above
presentation of $\cG_{(01)^\infty}$ but uses a substitution $a\to b$, $b\to
c$, $c\to aba$. This substitution preserves no infinite sequence.
Instead, there are three sequences cyclically permuted by the
substitution. Each of the three sequences generates the same minimal
subshift. Note that the group $\mathrm{IMG}(z^2+i)$ acts naturally on the
boundary of a rooted binary tree but, unlike the groups $\cG_\om$, the
Schreier graphs of orbits are not of linear shape.
\begin{problem}\label{problem-IMG}
Does the group $\mathrm{IMG}(z^2+i)$ embed into the TFG of a minimal Cantor
system? Does it embed into the TFG of a substitution subshift generated by
a primitive substitution?
\end{problem}
We should also mention that connections between the Thue-Morse substitution
and self-similar groups were previously explored in the paper
\cite{Barth:Thue-Morse} where the substitution $\TM$ was used to construct
a certain self-similar group that turned out to be the IMG of the rational
map $f(z)=\bigl(z-\frac12 z^2\bigr)^{-1}$. Note that this group can be
generated by a $3$-state automaton and, in fact, it is featured in the
classification \cite{Classification} of such groups (see Automaton no.\@
2853 in \cite{Classification}).
The paper is organized as follows. In Section \ref{sect-sub} we recall
basic facts about substitution subshifts, discuss a few specific examples
(including the subshift generated by the Thue-Morse substitution) and
establish relations between them (Proposition \ref{3-subshifts-factors}).
For the reader's convenience, this section is self-contained though some
material in it is known and can be found in textbooks on symbolic dynamics.
In Section \ref{sect-TFG} we consider topological full groups associated
with aperiodic homeomorphisms of a Cantor set and prove the embedding
Theorem \ref{main0} (Theorem \ref{TFG-embeds}). This allows to determine
at the end of the section a certain group generated by three involutive
homeomorphisms that has intermediate growth and embeds into the TFG of the
Thue-Morse subshift (Theorem \ref{sub-main}). In Section \ref{sect-growth}
we describe the family $\cG_\om$, $\om\in\{0,1,2\}^{\bN}$ of groups of
intermediate growth constructed in \cite{Grig84} and single out the group
$\cG_{(01)^\infty}$ as the one that is isomorphic to the group described in
Section \ref{sect-TFG} (Proposition \ref{inj-homomorph}). In Section
\ref{sect-Schr} we consider the Schreier graphs of groups actions. We
study the Schreier graphs with linear structure and use them to prove
Proposition \ref{inj-homomorph}. Then we prove Theorem \ref{sub-main} and
the main Theorem \ref{main}. The section is concluded with the proof of
Proposition \ref{two-more-inj}.
\medskip
\emph{Acknowledgements}.
The first author is supported by the Travel Support for Mathematicians
grant MP-TSM-00002045 from the Simons Foundation.
%
\section{Substitution subshifts}\label{sect-sub}
An \emph{alphabet} $\cA$ is a finite set consisting of more than one
element. Elements of $\cA$ are referred to as \emph{letters} or
\emph{symbols}. Let $\cA^*$ denote the set of all finite strings
$x_1x_2\ldots x_n$ of letters from $\cA$ (including the empty one
$\varnothing$). We refer to them as \emph{words} over $\cA$ and write
without any delimiters. The number of letters in a word is called its
\emph{length}. Elements of $\cA$ are identified with words of length $1$.
For any words $u,w\in\cA^*$ let $uw$ denote their \emph{concatenation}: if
$u=x_1x_2\ldots x_m$ and $w=y_1y_2\ldots y_n$, where each $x_i$ and $y_j$
is a letter, then $uw=x_1x_2\ldots x_my_1y_2\ldots y_n$. This operation is
associative and turns $\cA^*$ into a monoid. Likewise, for any finite list
of words $w_1,w_2,\dots,w_k$ their concatenation is denoted $w_1w_2\ldots
w_k$. The concatenation of $k$ copies of the same word $w$ is denoted
$w^k$. We say that a word $u\in\cA^*$ is a \emph{prefix} of a word
$w\in\cA^*$ if $w=uv$ for some $v\in\cA^*$. We say that $u$ is a
\emph{subword} of $w$ if $w=vuv'$ for some $v,v'\in\cA^*$.
For any nonempty set $S$ let $\cA^S$ denote the set of all functions
$\xi:S\to\cA$. In the cases $S=\bN$ (positive integers), $S=-\bN$
(negative integers), and $S=\bZ$ (all integers), we represent an element
$\xi\in\cA^S$ as respectively an infinite string $x_1x_2x_3\ldots$, a
left-infinite string $\ldots x_{-3}x_{-2}x_{-1}$, and a bi-infinite string
$\ldots x_{-2}x_{-1}x_0.x_1x_2x_3\ldots$, where $x_i=\xi(i)$ for each $i\in
S$. Notation for bi-infinite strings includes a dot that serves as a
reference point. We say that a nonempty word $w\in\cA^*$ is a
\emph{subword} of the string $\xi$ if $w=x_nx_{n+1}\ldots x_m$ for some
integers $n$ and $m$, $n\le m$ such that $\{n,n+1,\dots,m\}\subset S$.
Given a word $w=x_1x_2\ldots x_n\in\cA^*$, an infinite string
$\eta=y_1y_2\ldots\in\cA^{\bN}$, and a left-infinite string $\zeta=\ldots
z_2z_1\in\cA^{-\bN}$, we define \emph{concatenations} $w\eta$, $\zeta w$,
and $\zeta.\eta$ as respectively the infinite string $x_1x_2\ldots x_n
y_1y_2\ldots$, the left-infinite string $\ldots z_2z_1x_1x_2\ldots x_n$,
and the bi-infinite string $\ldots z_2z_1.y_1y_2\ldots$. We say that a
word $w\in\cA^*$ is a \emph{prefix} of an infinite string $\xi\in\cA^{\bN}$
if $\xi=w\eta$ for some $\eta\in\cA^{\bN}$. Note that for any words
$u,w\in\cA^*$ and strings $\eta\in\cA^{\bN}$ and $\zeta\in\cA^{-\bN}$, we
have $(uw)\eta=u(w\eta)$ and $\zeta(uw)=(\zeta u)w$ so that expressions
$uw\eta$ and $\zeta uw$ make sense. On the other hand, usually
$\zeta.(u\eta)\ne(\zeta u).\eta$.
Let $w_1,w_2,w_3,\dots$ be an infinite list of words in $\cA^*$ in which at
most finitely many words are empty. Then there exists a unique string
$\eta\in\cA^{\bN}$ such that each finite concatenation $w_1w_2\ldots w_k$,
$k\ge1$ is a prefix of $\eta$. We denote $\eta$ as an \emph{infinite
concatenation} $w_1w_2w_3\ldots$. If all $w_i$ are copies of the same word
$w$, we also use notation $w^\infty$. Likewise, the \emph{left-infinite
concatenation} $\ldots w_3w_2w_1$ is defined as the only left-infinite
string $\zeta\in\cA^{-\bN}$ such that for any $k\ge1$ we have
$\zeta=\zeta'(w_k\ldots w_2w_1)$ for some $\zeta'\in\cA^{-\bN}$.
Next we are going to introduce the \emph{shift} and \emph{subshifts} over
an alphabet $\cA$, which are topological dynamical systems on the set
$\cA^{\bZ}$ and its subsets. First we need a topology on $\cA^{\bZ}$. We
regard $\cA$ as a discrete topological space, then $\cA^{\bZ}$ is endowed
with the \emph{product topology}. Given any nonempty finite set
$S_0\subset\bZ$ and any function $\om_0:S_0\to\cA$, let
$\Sigma_{\cA}(S_0,\om_0)$ denote the set of all functions $\om:\bZ\to\cA$
such that the restriction of $\om$ to $S_0$ coincides with $\om_0$. Sets
of the form $\Sigma_{\cA}(S_0,\om_0)$, referred to as \emph{cylinders}, are
open subsets of $\cA^{\bZ}$. Moreover, they form a base of the topology on
$\cA^{\bZ}$. The topological space $\cA^{\bZ}$ is compact and metrizable.
Convergence in $\cA^{\bZ}$ is the pointwise convergence of functions. In
terms of bi-infinite strings, a sequence of strings $\om^{(k)}=\ldots
x^{(k)}_{-2}x^{(k)}_{-1}x^{(k)}_0.x^{(k)}_1x^{(k)}_2x^{(k)}_3\ldots$,
$k=1,2,\dots$ converges to a string $\eta=\ldots
y_{-2}y_{-1}y_0.y_1y_2y_3\ldots$ in $\cA^{\bZ}$ if for every $n\in\bZ$ the
sequence of letters $x^{(1)}_n,x^{(2)}_n,x^{(3)}_n,\dots$ eventually
coincides with the constant sequence $y_n,y_n,y_n,\dots$.
\begin{definition}\label{def-shift}
The (two-sided) \textbf{shift} over an alphabet $\cA$ is a transformation
of $\cA^{\bZ}$, denoted $\si_{\cA}$ or simply $\si$, that sends any
function $\om:\bZ\to\cA$ to a function $\eta:\bZ\to\cA$ given by
$\eta(n)=\om(n+1)$ for all $n\in\bZ$. In terms of bi-infinite strings,
$\si(\ldots x_{-2}x_{-1}x_0.x_1x_2x_3\ldots)=\ldots x_{-1}x_0x_1.x_2x_3x_4
\ldots$. A \textbf{subshift} over $\cA$ is the restriction $\si_{\cA}|Y$
of the shift $\si_{\cA}$ to a nonempty closed set $Y\subset\cA^{\bZ}$ such
that $\si_{\cA}(Y)=Y$.
\end{definition}
The shift $\si_{\cA}$ is invertible and maps cylinders onto cylinders,
hence it is a homeomorphism of the topological space $\cA^{\bZ}$ onto
itself. Any subshift $\si_{\cA}|Y$ is a homeomorphism of the compact
topological space $Y$ onto itself.
Given an infinite string $\xi\in\cA^{\bN}$, let $\Omega(\xi)$ denote the
set of all bi-infinite strings $\om\in\cA^{\bZ}$ such that every subword of
$\om$ is also a subword of $\xi$.
\begin{lemma}\label{omega-xi}
For any $\xi\in\cA^{\bN}$, the set $\Omega(\xi)$ is nonempty and closed,
and $\si_{\cA}(\Omega(\xi))=\Omega(\xi)$. Every subword of any string in
$\Omega(\xi)$ occurs infinitely many times as a subword of $\xi$.
\end{lemma}
\begin{proof}
For any bi-infinite string $\om\in\cA^{\bZ}$, the shifted string
$\si_{\cA}(\om)$ has exactly the same subwords as $\om$. It follows that
$\si_{\cA}(\Omega(\xi))=\Omega(\xi)$. Next we prove that $\Omega(\xi)$ is
a closed set. Given a nonempty word $w\in\cA^*$ of length $k\ge1$ and any
$n\in\bZ$, let $\Sigma_{w,n}$ be the set of all bi-infinite strings $\ldots
y_{-1}y_0.y_1y_2\ldots$ in $\cA^{\bZ}$ such that $y_ny_{n+1}\ldots
y_{n+k-1}=w$. The set $\Sigma_{w,n}$ is a cylinder. The complement of
$\Omega(\xi)$ is the union of cylinders $\Sigma_{w,n}$ over all words
$w\in\cA^*$ that are not subwords of $\xi$ and over all $n\in\bZ$. Hence
the complement is an open set. Then the set $\Omega(\xi)$ is closed.
Next we prove that every subword $u$ of any string $\om\in\Om(\xi)$ occurs
infinitely often as a subword of $\xi$. Let $\xi=x_1x_2x_3\ldots$ and
$\om=\ldots y_{-1}y_0.y_1y_2\ldots$. If $u$ is a subword of length $s\ge1$
of $\om$, we have $y_my_{m+1}\ldots y_{m+s-1}=u$ for some $m\in\bZ$. Then
for any $k\ge1$ the word $u_k=y_{m-k}y_{m-k+1}\ldots
y_{m-1}y_my_{m+1}\ldots y_{m+s-1}$ is a subword of length $s+k$ of $\om$.
Since $\om\in\Om(\xi)$, the same word $u_k$ is also a subword of the string
$\xi$. That is, $x_{n_k}x_{n_k+1}\ldots x_{n_k+k+s-1}=u_k$ for some
$n_k\ge1$. Then $x_{n_k+k}x_{n_k+k+1}\ldots x_{n_k+k+s-1}$ is another
occurrence of $u$ as a subword of $\xi$. Since $n_k+k\ge k+1$, there are
infinitely many numbers of the form $n_k+k$, $k\ge1$. Therefore $u$ occurs
infinitely often as a subword of the string $\xi$.
It remains to prove that the set $\Om(\xi)$ is not empty. Let $W_\xi$ be
the set of all nonempty words that occur infinitely many times as subwords
of $\xi$. Since the alphabet $\cA$ is finite while the string $\xi$ is
infinite, $W_\xi$ contains at least one $1$-letter word. Let us show that
for any word $w\in W_\xi$ there exist letters $l,r\in\cA$ such that the
word $lwr$ belongs to $W_\xi$ as well. Let $s$ be the length of $w$.
There is a sequence $\{n_k\}$, $2\le n_1<n_2<n_3<\ldots$, such that
$x_{n_k}x_{n_k+1}\ldots x_{n_k+s-1}=w$ for all $k\ge1$. Since the
alphabet $\cA$ is finite, we can find letters $l,r\in\cA$ such that
$(x_{n_k-1},x_{n_k+s})=(l,r)$ for infinitely many values of $k$. Then
$lwr\in W_\xi$.
Now it follows by induction from the above that there exist two sequences
of letters $r_0,r_1,r_2,\dots$ and $l_1,l_2,l_3,\dots$ such that for any
$k\ge1$ the word $w_k=l_kl_{k-1}\ldots l_2l_1r_0r_1r_2\ldots r_k$ belongs
to $W_\xi$. Then the bi-infinite string $\om=\ldots l_2l_1r_0.r_1r_2
\ldots$ belongs to the set $\Om(\xi)$ so that this set is not empty.
Indeed, any subword $u$ of $\om$ is also a subword of $w_k$ for all
sufficiently large $k$. Since each $w_k$ is a subword of $\xi$, so is $u$.
\end{proof}
In view of Lemma \ref{omega-xi}, the restriction of the shift $\si_{\cA}$
to the set $\Om(\xi)$, where $\xi\in\cA^{\bN}$, is a subshift. We refer to
it as the subshift generated by the string $\xi$. Even though $\xi$ does
not belong to $\cA^{\bZ}$, the shift directly relates it to the set
$\Om(\xi)$ via the following construction.
\begin{lemma}\label{limit-points}
For any infinite string $\xi\in\cA^{\bN}$ and left-infinite string
$\zeta\in\cA^{-\bN}$, the set of all limit points of the sequence
$\zeta.\xi,\si_{\cA}(\zeta.\xi),\si_{\cA}^2(\zeta.\xi),\dots$ in
$\cA^{\bZ}$ coincides with $\Omega(\xi)$.
\end{lemma}
\begin{proof}
Let $\ldots x_{-1}x_0.x_1x_2\ldots$ be the bi-infinite string $\zeta.\xi$.
Then $\xi=x_1x_2\ldots$. For any $k\ge0$ we have $\si_{\cA}^k(\zeta.\xi)=
\ldots x^{(k)}_{-1}x^{(k)}_0.x^{(k)}_1x^{(k)}_2\ldots$, where
$x^{(k)}_n=x_{n+k}$ for all $n\in\bZ$. Consider an arbitrary bi-infinite
string $\om=\ldots y_{-1}y_0.y_1y_2\ldots\in\cA^{\bZ}$. First assume that
$\om\in\Om(\xi)$. For any $k\ge1$ the word $w_k=y_{-k}\ldots
y_{-1}y_0.y_1\ldots y_k$ is a subword of $\om$ of length $2k+1$. Hence it
is also a subword of $\xi$. That is, $x_{n_k}x_{n_k+1}\ldots
x_{n_k+2k}=w_k$ for some $n_k\ge1$. Moreover, $n_k$ can be chosen
arbitrarily large since $w_k$ occurs infinitely many times as a subword of
$\xi$ (due to Lemma \ref{omega-xi}). Therefore we can build the sequence
$\{n_k\}_{k\ge1}$ inductively in such a way that $1<n_1+1<n_2+2<n_3+3<
\ldots$. Then $\{\si_{\cA}^{n_k+k}(\zeta.\xi)\}_{k\ge1}$ is a subsequence
of the sequence $\zeta.\xi,\si_{\cA}(\zeta.\xi),\si_{\cA}^2(\zeta.\xi),
\dots$. This subsequence converges to $\om$ since
$x^{(n_k+k)}_m=x_{m+n_k+k}=y_m$ whenever $k\ge|m|$.
Conversely, assume that $\om$ is a limit point of the sequence
$\zeta.\xi,\si_{\cA}(\zeta.\xi),\si_{\cA}^2(\zeta.\xi),\dots$. That is,
$\si_{\cA}^{k_i}(\zeta.\xi)\to\om$ as $i\to\infty$, where $0\le
k_1<k_2<k_3<\ldots$. Take any nonempty subword $u$ of the string $\om$.
We have $y_{n_1}y_{n_1+1}\ldots y_{n_2}=u$ for some integers $n_1$ and
$n_2$, $n_1\le n_2$. The convergence implies that the word
$x^{(k_i)}_{n_1}x^{(k_i)}_{n_1+1}\ldots x^{(k_i)}_{n_2}=
x_{n_1+k_i}x_{n_1+1+k_i}\ldots x_{n_2+k_i}$ coincides with $u$ for all
sufficiently large $i$. If $i$ is large enough then, additionally,
$n_1+k_i\ge1$ so that $u$ occurs as a subword of the string $\xi$. Thus
every subword of $\om$ is also a subword of $\xi$, which means that
$\om\in\Om(\xi)$.
\end{proof}
\begin{definition}\label{def-sub}
A \textbf{substitution} over an alphabet $\cA$ is a map
$\mathsf{s}:\cA\to\cA^*\setminus\{\varnothing\}$. The substitution can be
applied term by term to any finite or infinite string $x_1x_2x_3\ldots$ of
letters from $\cA$, which yields a finite or infinite concatenation
$\mathsf{s}(x_1)\mathsf{s}(x_2)\mathsf{s}(x_3)\ldots$. Hence the map
$\mathsf{s}$ induces a transformation of $\cA^*$ and a transformation of
$\cA^{\bN}$. We use the same notation for the substitution and both
induced transformations.
\end{definition}
Let $\mathsf{s}$ be a substitution over an alphabet $\cA$. Suppose $z$ is
a letter in $\cA$ such that the word $\mathsf{s}(z)$ begins with $z$ and
has length greater than $1$. That is, $\mathsf{s}(z)=zw$ for some nonempty
word $w\in\cA^*$. Consider a sequence of words
$z,\mathsf{s}(z),\mathsf{s}^2(z),\dots$ obtained by repeatedly applying the
substitution to the letter $z$. For any $k\ge1$ we have $\mathsf{s}^k(z)=
\mathsf{s}^{k-1}(zw)=\mathsf{s}^{k-1}(z)\mathsf{s}^{k-1}(w)$. Note that
$\mathsf{s}^{k-1}(w)\ne\varnothing$ since $w\ne\varnothing$. Hence every
word in the sequence is a prefix of the next word and of smaller length
than the next word. It follows that, as $k\to\infty$, words
$\mathsf{s}^k(z)$ in a sense converge to an infinite string. To be
precise, there exists a unique string $\xi\in\cA^{\bN}$ such that each
$\mathsf{s}^k(z)$ is a prefix of $\xi$. Clearly, $\mathsf{s}(\xi)=\xi$.
Moreover, $\xi$ is the only infinite string that begins with the letter $z$
and is preserved by the substitution $\mathsf{s}$.
\begin{definition}\label{def-subshift}
The \textbf{substitution subshift} generated by a substitution $\mathsf{s}$
over an alphabet $\cA$ is a subshift $\si_{\cA}|\Omega(\xi)$, where $\xi$
is an infinite string preserved by $\mathsf{s}$.
\end{definition}
In this paper we are interested in substitution subshifts generated by
three substitutions. The first one is the \emph{Thue-Morse} substitution
$\TM$ over the alphabet $\{0,1\}$ given by $\TM(0)=01$, $\TM(1)=10$. The
second one is the so-called \emph{period doubling} substitution $\pd$ over
the alphabet $\{x,y\}$ given by $\pd(x)=xy$, $\pd(y)=xx$. The third one is
a substitution $\sub$ over the alphabet $\{a,c,d\}$ given by $\sub(a)=aca$,
$\sub(c)=d$, $\sub(d)=c$.
Since the word $\TM(0)=01$ begins with $0$ and the word $\TM(1)=10$ begins
with $1$, the substitution $\TM$ preserves a unique infinite string
$\xiTM\in\{0,1\}^{\bN}$ that begins with $0$ and another infinite string
$\xiTM'$ that begins with $1$.
\begin{lemma}\label{TM-unique}
$\xiTM$ and $\xiTM'$ are obtained from each other by changing all 0s to 1s
and all 1s to 0s. Both strings generate the same substitution subshift.
\end{lemma}
\begin{proof}
Consider a substitution $\mathsf{s}_{(01)}$ over the alphabet $\{0,1\}$
given by $\mathsf{s}_{(01)}(0)=1$, $\mathsf{s}_{(01)}(1)=0$. We obtain
that $\mathsf{s}_{(01)}\TM(0)=\TM\mathsf{s}_{(01)}(0)=10$ and
$\mathsf{s}_{(01)}\TM(1)=\TM\mathsf{s}_{(01)}(1)=01$. It follows that
$\mathsf{s}_{(01)}\TM(\xi)=\TM\mathsf{s}_{(01)}(\xi)$ for any word or
infinite string $\xi$ over the alphabet $\{0,1\}$. In particular,
$\TM(\mathsf{s}_{(01)}(\xiTM))=\mathsf{s}_{(01)}(\TM(\xiTM))
=\mathsf{s}_{(01)}(\xiTM)$. Since the string $\mathsf{s}_{(01)}(\xiTM)$
begins with $1$, we have $\mathsf{s}_{(01)}(\xiTM)=\xiTM'$ due to
uniqueness of $\xiTM'$. In other words, $\xiTM'$ is obtained from $\xiTM$
by changing all 0s to 1s and all 1s to 0s. Then also
$\mathsf{s}_{(01)}(\xiTM')=\xiTM$.
For any $k\ge1$ we obtain $\TM^{k+1}(0)=\TM^k(01)=\TM^k(0)\TM^k(1)$ and
$\TM^{k+1}(1)=\TM^k(10)=\TM^k(1)\TM^k(0)$. As a consequence, the word
$\TM^k(1)$ is a subword of $\TM^{k+1}(0)$ while the word $\TM^k(0)$ is a
subword of $\TM^{k+1}(1)$. Any subword $w'$ of the string $\xiTM'$ is also
a subword of its prefix $\TM^k(1)$ for all sufficiently large $k$. Since
$\TM^k(1)$ is a subword of $\TM^{k+1}(0)$, which is a prefix of the string
$\xiTM$, it follows that $w'$ is a subword of $\xiTM$ as well. Conversely,
if $w$ is a subword of the string $\xiTM$, then for some $k\ge1$ it is a
subword of the prefix $\TM^k(0)$, which is a subword of $\TM^{k+1}(1)$,
which is a prefix of $\xiTM'$. Hence $w$ also a subword of $\xiTM'$. Thus
the strings $\xiTM$ and $\xiTM'$ have the same subwords. Therefore
$\Om(\xiTM)=\Om(\xiTM')$ so that both strings generate the same subshift.
\end{proof}
Since the word $\pd(x)=xy$ begins with $x$, the substitution $\pd$
preserves a unique infinite string $\xipd\in\{x,y\}^{\bN}$ that begins with
$x$. Since the word $\pd(y)=xx$ does not begin with $y$, the substitution
$\pd$ preserves no infinite string that begins with $y$. Likewise, the
substitution $\sub$ preserves a unique infinite string
$\xiGE\in\{a,c,d\}^{\bN}$ that begins with $a$ and no infinite string that
begins with $c$ or $d$.
Let us explore the dynamics of the substitution subshifts
$\si_{\{0,1\}}|\Om(\xiTM)$, $\si_{\{x,y\}}|\Om(\xipd)$ and
$\si_{\{a,c,d\}}|\Om(\xiGE)$. First we need detailed descriptions of the
strings $\xiTM$, $\xipd$ and $\xiGE$.
\begin{lemma}\label{xi-TM}
The $n$-th symbol of the string $\xiTM$ is $0$ if the number of 1s in the
binary expansion of the number $n-1$ is even, and $1$ otherwise.
\end{lemma}
\begin{proof}
For any $n\in\bN$ let $x_n$ denote the $n$-th symbol of $\xiTM$. Then
$\xiTM=x_1x_2x_3\ldots$. As $\TM(\xiTM)=\xiTM$, we also have
$\xiTM=\TM(x_1)\TM(x_2)\TM(x_3)\ldots$. Since $\TM(0)$ and $\TM(1)$ are
both words of length $2$, we obtain that $\TM(x_i)=x_{2i-1}x_{2i}$ for all
$i\ge1$. Since $\TM(0)=01$ and $\TM(1)=10$, we further obtain that
$x_{2i-1}=x_i$ while $x_{2i}\ne x_i$.
For any $i\ge2$ the binary expansion of the number $(2i-1)-1=2(i-1)$ is
obtained from the binary expansion of $i-1$ by appending $0$ to it while
the binary expansion of $2i-1$ is obtained by appending $1$. Since
$x_{2i-1}=x_i$ and $x_{2i}\ne x_i$, it follows that the lemma holds for
$n=2i-1$ and $n=2i$ whenever it holds for $n=i$. Besides, the lemma holds
for $n=1$ and $n=2$ as $x_1=0$ and $x_2=1$. Since any $n\ge3$ can be
represented as $2i-1$ or $2i$ for some integer $i$, $2\le i<n$, the lemma
follows by strong induction on $n$.
\end{proof}
\begin{lemma}\label{xi-pd}
Let $n=2^km$, where $k$ is a nonnegative integer and $m$ is a positive odd
integer. Then the $n$-th letter of the string $\xipd$ is $x$ if $k$ is
even, and $y$ if $k$ is odd.
\end{lemma}
\begin{proof}
For any $n\in\bN$ let $x_n$ denote the $n$-th letter of $\xipd$. Then
$\xipd=x_1x_2x_3\ldots$. As $\pd(\xipd)=\xipd$, we also have
$\xipd=\pd(x_1)\pd(x_2)\pd(x_3)\ldots$. Since $\pd(x)$ and $\pd(y)$ are
both words of length $2$, we obtain that $\pd(x_i)=x_{2i-1}x_{2i}$ for all
$i\ge1$. Since $\pd(x)=xy$ and $\pd(y)=xx$, we further obtain that
$x_{2i}=y$ if $x_i=x$, and $x_{2i}=x$ if $x_i=y$. Besides, $x_n=x$ for any
odd $n$. Now the lemma easily follows by induction on $k$.
\end{proof}
\begin{lemma}\label{xi-GE}
Let $n=2^km$, where $k$ is a nonnegative integer and $m$ is a positive odd
integer. Then the $n$-th letter of the string $\xiGE$ is $a$ if $k=0$, $c$
if $k$ is odd, and $d$ if $k$ is even and positive.
\end{lemma}
\begin{proof}
Let $W$ be the set of all words of the form $ay_1ay_2\ldots ay_s$, where
each $y_i$ is $c$ or $d$. We have $\sub(ac)=acad$ and $\sub(ad)=acac$. It
follows that the set $W$ is invariant under the substitution $\sub$.
Besides, $\sub$ doubles the length of every word in $W$. Since $ac$ is a
prefix of the infinite string $\xiGE$ and $\sub(\xiGE)=\xiGE$, each of the
words $\sub(ac),\sub^2(ac),\sub^3(ac),\dots$ is also a prefix of $\xiGE$.
All these words belong to $W$ and their length gets arbitrarily large. We
conclude that $\xiGE=x_1x_2x_3\ldots$, where $x_n=a$ if $n$ is odd, and
$x_n=c$ or $d$ if $n$ even.
We have $\xiGE=\sub(\xiGE)=\sub(x_1x_2)\sub(x_3x_4)\sub(x_5x_6)\ldots$.
Since $\sub(ac)$ and $\sub(ad)$ are both words of length $4$, it follows
that $\sub(x_{2i-1}x_{2i})=x_{4i-3}x_{4i-2}x_{4i-1}x_{4i}$ for all
$i\ge1$. Since $\sub(ac)=acad$ and $\sub(ad)=acac$, it further follows
that $x_{4i}=d$ if $x_{2i}=c$, and $x_{4i}=c$ if $x_{2i}=d$. Besides,
$x_n=c$ for any even $n$ not divisible by $4$. Now the lemma easily
follows by induction on $k$.
\end{proof}
The description of the string $\xiGE$ obtained in Lemma \ref{xi-GE} allows
to produce explicit examples of bi-infinite strings in the set $\Om(\xiGE)$.
\begin{lemma}\label{xiGE-examples}
Let $\overleftarrow{\xiGE}$ denote the left-infinite string obtained by
writing $\xiGE$ backwards. Then the bi-infinite strings
$\overleftarrow{\xiGE}c.\xiGE$ and $\overleftarrow{\xiGE}d.\xiGE$ belong to
$\Om(\xiGE)$.
\end{lemma}
\begin{proof}
Let $\om=\ldots x_{-1}x_0.x_1x_2\ldots$ be one of the strings
$\overleftarrow{\xiGE}c.\xiGE$ and $\overleftarrow{\xiGE}d.\xiGE$. That
is, $\xiGE=x_1x_2x_3\ldots$, $x_{-n}=x_n$ for all $n\ge1$, and $x_0=c$ or
$d$. We need to show that every subword of $\om$ is also a subword of
$\xiGE$. Lemma \ref{xi-GE} implies that for any $n\ge1$ the letter
$x_n$ depends only on the largest power of $2$ that divides $n$. As a
consequence, $x_n=x_{2^k-n}=x_{2^k+n}$ for any integers $n,k\ge1$ such that
$n<2^k$. It follows that for any $k\ge1$ the words $w_k=x_1x_2\ldots
x_{2^k-1}$, $w^-_k=x_{1-2^k}\ldots x_{-2}x_{-1}$ and
$w^+_k=x_{2^k+1}x_{2^k+2}\ldots x_{2^{k+1}-1}$ are the same.
Any subword $u$ of the string $\om$ is also a subword of $w^-_kx_0w_k$ for
all sufficiently large $k$. By the above, $w^-_kx_0w_k=w_kx_0w^+_k$. By
Lemma \ref{xi-GE}, $x_{2^k}=c$ if $k$ is odd, and $x_{2^k}=d$ if $k$ is
even and positive. It follows that $u$ is a subword of $w_kx_{2^k}w^+_k$
for some $k$. Since $w_kx_{2^k}w^+_k$ is a prefix of the string $\xiGE$,
the word $u$ is also a subword of $\xiGE$.
\end{proof}
\begin{definition}\label{def-minimal}
A homeomorphism $T:X\to X$ of a compact topological space $X$ onto itself
is called \textbf{minimal} if there is no closed set $Y\subset X$ different
from the empty set and $X$ such that $T(Y)=Y$. An equivalent condition is
that for any point $x\in X$ the \emph{two-sided orbit} $\{T^n(x)\mid
n\in\bZ\}$ is dense in $X$.
\end{definition}
Minimality of a homeomorphism $T$ is a sign of very regular recurrence in
the dynamical system defined by $T$.
\begin{proposition}\label{3-subshifts-minimal}
The subshifts $\si_{\{0,1\}}|\Om(\xiTM)$, $\si_{\{x,y\}}|\Om(\xipd)$ and
$\si_{\{a,c,d\}}|\Om(\xiGE)$ are minimal.
\end{proposition}
Note that minimality of the first two subshifts in the proposition is well
known. We are going to present a unified proof for all three subshifts.
First let us observe that minimality of a subshift can be described in
purely combinatorial terms.
\begin{lemma}\label{same-subwords}
A subshift $\si_{\cA}|Y$ is minimal if and only if any two bi-infinite
strings in $Y$ have the same subwords.
\end{lemma}
\begin{proof}
For any nonempty word $w\in\cA^*$ of length $k\ge1$ and any
$n\in\bZ$, let $\Sigma_{w,n}$ be the set of all bi-infinite strings $\ldots
y_{-1}y_0.y_1y_2\ldots$ in $\cA^{\bZ}$ such that $y_ny_{n+1}\ldots
y_{n+k-1}=w$. Each of the sets $\Sigma_{w,n}$ is a cylinder. It is easy
to observe that these cylinders form a base of the topology on
$\cA^{\bZ}$. Further, for any $w\in\cA^*\setminus\{\varnothing\}$ let
$U_w$ be the set of all strings in $\cA^{\bZ}$ that admit $w$ as a
subword. Obviously, $U_w$ is the union of the cylinders $\Sigma_{w,n}$
over all $n\in\bZ$. Hence $U_w$ is an open set. Besides,
$\si_{\cA}(\Sigma_{w,n})=\Sigma_{w,n-1}$ for all $n\in\bZ$, which implies
that $\si_{\cA}(U_w)=U_w$.
Consider an arbitrary subshift $\si_{\cA}|Y$. It follows from the above
that for any nonempty word $w\in\cA^*$, the set difference $Y\setminus U_w$
is a closed set and $\si_{\cA}(Y\setminus U_w)=Y\setminus U_w$. Assuming
the subshift $\si_{\cA}|Y$ is minimal, we obtain that $Y\setminus U_w$ is
either the empty set or $Y$. Hence $w$ is either a subword of every string
in $Y$ or a subword of no string in $Y$. Since this holds for any nonempty
word $w$, we conclude that any two strings in $Y$ have the same subwords.
Now assume the subshift $\si_{\cA}|Y$ is not minimal, that is, there is a
nonempty closed set $Y_0\subset Y$ different from $Y$ such that
$\si_{\cA}(Y_0)=Y_0$. Take any point $\om\in Y\setminus Y_0$. Since the
cylinders of the form $\Sigma_{w,n}$ constitute a base of the topology, we
can find one of them that contains $\om$ and is disjoint from the closed
set $Y_0$. Since $\si_{\cA}(Y_0)=Y_0$, it follows that the set $U_w$ is
also disjoint from $Y_0$. We obtain that $w$ is a subword of the string
$\om$ but not a subword of any string in $Y_0$. Thus two strings in $Y$ do
not always have the same subwords.
\end{proof}
For subshifts of the form $\si_{\cA}|\Om(\xi)$, where $\xi\in\cA^{\bZ}$,
minimality is equivalent to a very restrictive combinatorial property of
the string $\xi$ (usually called \emph{uniform recurrence}).
\begin{lemma}\label{subwords-in-subwords}
The subshift $\si_{\cA}|\Om(\xi)$, where $\xi\in\cA^{\bZ}$, is minimal if
and only if for any nonempty word $w\in\cA^*$ that occurs infinitely many
times as a subword of $\xi$, there exists $N\ge1$ such that $w$ is a
subword of every word of length at least $N$ that is a subword of the
string $\xi$.
\end{lemma}
\begin{proof}
Take any left-infinite string $\zeta\in\cA^{-\bN}$. Let $\ldots
x_{-1}x_0.x_1x_2\ldots$ be the bi-infinite string $\zeta.\xi$. First we
show that any nonempty word $w$ that occurs infinitely often as a subword
of $\xi$ is also a subword of some string $\om\in\Om(\xi)$. Let $s$ be the
length of $w$. There is a sequence of integers $\{n_k\}$, $0\le
n_1<n_2<n_3<\ldots$, such that $x_{n_k+1}x_{n_k+2}\ldots x_{n_k+s}=w$ for
all $k\ge1$. Since $\cA^{\bZ}$ is compact, the sequence
$\{\si^{n_k}(\zeta.\xi)\}_{k\ge1}$ has a limit point $\om$. By
construction, $\om=\zeta_1.w\xi_1$ for some $\xi_1\in\cA^{\bN}$ and
$\zeta_1\in\cA^{-\bN}$. In particular, $w$ is a subword of the string
$\om$. Clearly, $\om$ is also a limit point of the sequence
$\zeta.\xi,\si_{\cA}(\zeta.\xi),\si_{\cA}^2(\zeta.\xi),\dots$. Hence
$\om\in\Om(\xi)$ due to Lemma \ref{limit-points}.
Next we show that for any word $w$ that is not a subword of all
sufficiently long subwords of $\xi$, there is a string $\om'\in\Om(\xi)$
such that $w$ is not a subword of $\om'$. For any $N\ge1$ let $w_N$ be a
subword of $\xi$ of length at least $N$ that does not admit $w$ as a
subword. We have $w_N=x_{n_N}x_{n_N+1}\ldots x_{m_N}$ for some $n_N$ and
$m_N$, where $n_N\ge1$ and $m_N-n_N\ge N-1$. The sequence
$\{\si^{n_{2k}+k}(\zeta.\xi)\}_{k\ge1}$ has a limit point $\om'=\ldots
y_{-1}y_0.y_1y_2\ldots$. Since $n_{2k}+k\to\infty$ as $k\to\infty$, it
follows that the string $\om'$ is also a limit point of the sequence
$\zeta.\xi,\si_{\cA}(\zeta.\xi),\si_{\cA}^2(\zeta.\xi),\dots$. Hence
$\om'\in\Om(\xi)$ due to Lemma \ref{limit-points}. Any subword $u$ of
$\om'$ is also a subword of the word $u_s=y_{-s}y_{-s+1}\ldots y_{s-1}y_s$
for some $s\ge1$. By construction, $x_{n_{2k}+k-s}\,x_{n_{2k}+k-s+1}\ldots
x_{n_{2k}+k+s}=u_s$ for infinitely many values of $k$. If $k\ge s+1$ then
$n_{2k}+k-s>n_{2k}$ and $n_{2k}+k+s\le n_{2k}+2k-1\le m_{2k}$. It follows
that $u_s$ is a subword of the word $w_{2k}$ for some $k$. Then $u$ is
also a subword of $w_{2k}$ so that $u\ne w$.
Now we can prove the lemma. Assume that the string $\xi$ does not have the
combinatorial property formulated in the lemma. That is, some nonempty
word $w$ occurs infinitely many times as a subword of $\xi$, but $w$ is not
a subword of all sufficiently long subwords of $\xi$. By the above there
exist strings $\om,\om'\in\Om(\xi)$ such that $w$ is a subword of $\om$ but
not a subword of $\om'$. Then it follows from Lemma \ref{same-subwords}
that the subshift $\si_{\cA}|\Om(\xi)$ is not minimal. Conversely, assume
that the string $\xi$ has the combinatorial property formulated in the
lemma. Take an arbitrary $\om\in\Om(\xi)$ and let $w$ be a subword of
$\om$. By Lemma \ref{omega-xi}, the word $w$ occurs infinitely often as a
subword of $\xi$. Hence for some $N\ge1$ every subword of $\xi$ of length
at least $N$ admits $w$ as a subword. Clearly, any bi-infinite string
$\om'\in\Om(\xi)$ has a subword $w'$ of length $N$. Since $w'$ is also a
subword of $\xi$, it admits $w$ as a subword. Then $w$ is a subword of
$\om'$ as well. We obtain that every subword of $\om$ is, in fact, a
subword of all strings in $\Om(\xi)$. As the choice of $\om$ was
arbitrary, we conclude that any two strings in $\Om(\xi)$ have the same
subwords. By Lemma \ref{same-subwords}, the subshift $\si_{\cA}|\Om(\xi)$
is minimal.
\end{proof}
\begin{proof}[Proof of Proposition \ref{3-subshifts-minimal}]
Let $\xi=x_1x_2x_3\ldots$, where $\xi=\xiTM$, $\xipd$ or $\xiGE$. In the
cases $\xi=\xipd$ and $\xi=\xiGE$, it follows from respectively Lemmas
\ref{xi-pd} and \ref{xi-GE} that $x_{n_1}=x_{n_2}$ whenever $1\le n_1<2^k$
and $n_2-n_1=2^km$ for some integers $k,m\ge1$. As a consequence, the word
$x_1x_2\ldots x_{2^k-1}$ coincides with $x_{2^km+1}x_{2^km+2}\ldots
x_{2^km+2^k-1}$ for any $k,m\ge1$.
In the case $\xi=\xiTM$, it follows from Lemma \ref{xi-TM} that
$x_{n_1}=x_{n_2}$ whenever $1\le n_1\le2^k$ and $n_2-n_1=2^km$ for some
integers $k,m\ge1$ such that the binary expansion of the number $m$ has an
even number of 1s. Note that for any $s\ge1$ exactly one of the numbers
$2s$ and $2s+1$ has an even number of 1s in its binary expansion. As a
consequence, the word $x_1x_2\ldots x_{2^k}$ coincides with the word
$x_{M+1}x_{M+2}\ldots x_{M+2^k}$ for $M=2^k(2s)$ or $M=2^k(2s+1)$.
For any $k\ge1$ let $Z_k$ be the set of all integers $m\ge1$ such that the
prefix of length $2^k-1$ of the string $\xi$ coincides with the subword
$x_{2^km+1}x_{2^km+2}\ldots x_{2^km+2^k-1}$. By the above, $Z_k=\bN$ if
$\xi=\xipd$ or $\xiGE$. If $\xi=\xiTM$ then for any $s\ge1$ the set $Z_k$
contains $2s$ or $2s+1$. It follows that $Z_k$ contains at least one of
any three consecutive positive integers.
Now we are going to show that the string $\xi$ has the combinatorial
property formulated in Lemma \ref{subwords-in-subwords}, which will
complete the proof of the proposition. Any subword $w$ of $\xi$ is also a
subword of the prefix $x_1x_2\ldots x_{2^k-1}$ for some $k$. Let
$N=2^{k+2}$ and consider an arbitrary subword $w'$ of $\xi$ of length at
least $N$. We have $w'=x_{n_1}x_{n_1+1}\ldots x_{n_2}$ for some $n_1$ and
$n_2$, where $n_1\ge1$ and $n_2-n_1\ge N-1$. Let $2^km$ be the largest
multiple of $2^k$ not exceeding $n_2$. Then $n_2\le 2^k(m+1)-1$, which
implies that $n_1\le n_2-N+1\le2^k(m-3)$. In particular, $m-3\ge1$. Hence
at least one of the integers $m-3$, $m-2$ and $m-1$ belongs to the set
$Z_k$. Let $m_0$ be such an integer. Then $x_{2^km_0+1}x_{2^km_0+2}\ldots
x_{2^km_0+2^k-1}=x_1x_2\ldots x_{2^k-1}$. Since $2^km_0+1>2^k(m-3)\ge n_1$
and $2^km_0+2^k-1\le 2^km-1<n_2$, it follows that $x_1x_2\ldots x_{2^k-1}$
is a subword of $w'$. Then $w$ is a subword of $w'$ as well.
\end{proof}
In the remainder of this section we are going to establish relations
between the subshifts $\si_{\{0,1\}}|\Om(\xiTM)$,
$\si_{\{x,y\}}|\Om(\xipd)$ and $\si_{\{a,c,d\}}|\Om(\xiGE)$.
\begin{definition}\label{def-factor}
Let $T_1:X_1\to X_1$ and $T_2:X_2\to X_2$ be homeomorphisms of topological
spaces. Suppose there exists a continuous map $f:X_1\to X_2$ such that $f$
is onto and $fT_1=T_2f$ so that the following diagram is commutative:
\[\begin{array}{ccc}
X_1 & \stackrel{T_1}{\longrightarrow} & X_1\\[0.2em]
f\Big\downarrow\phantom{f} && \phantom{f}\Big\downarrow f\\[0.25em]
X_2 & \stackrel{T_2}{\longrightarrow} & X_2
\end{array}
\]
Then we say that $T_2$ is a \textbf{continuous factor} of $T_1$ while $T_1$
is a \textbf{continuous extension} of $T_2$. In the case the map $f$ is
$k$-to-one for some $k\ge1$, we say that $T_2$ is a $k$-to-one continuous
factor of $T_1$. In the case $f$ is a homeomorphism, we say that the map
$T_2$ is \textbf{topologically conjugate} to $T_1$.
\end{definition}
\begin{proposition}\label{3-subshifts-factors}
The subshift $\si_{\{x,y\}}|\Om(\xipd)$ is a two-to-one continuous factor
of the subshift $\si_{\{0,1\}}|\Om(\xiTM)$. The subshift
$\si_{\{a,c,d\}}|\Om(\xiGE)$ is topologically conjugate to
$\si_{\{x,y\}}|\Om(\xipd)$.
\end{proposition}
To prove Proposition \ref{3-subshifts-factors}, we are going to use the
so-called \emph{block factor maps}, which are helpful for establishing
relations between subshifts. Let $\cA_1$ and $\cA_2$ be two alphabets.
Suppose that for some $k\ge1$ we have a function $\phi:\cA_1^k\to\cA_2$.
This function induces a map $f_\phi:\cA_1^{\bZ}\to\cA_2^{\bZ}$ that sends
any bi-infinite string $\ldots x_{-1}x_0.x_1x_2\ldots$ over the alphabet
$\cA_1$ to a bi-infinite string $\ldots y_{-1}y_0.y_1y_2\ldots$ over
$\cA_2$ given by $y_n=\phi(x_n,x_{n+1},\dots,x_{n+k-1})$ for all
$n\in\bZ$. It follows from the construction that for any cylinder
$\Sigma\subset\cA_2^{\bZ}$ the pre-image $f_\phi^{-1}(\Sigma)$ is a finite
union of cylinders in $\cA_1^{\bZ}$. Hence the map $f_\phi$ is
continuous. Also by construction, $f_\phi$ intertwines the shifts
$\si_{\cA_1}$ and $\si_{\cA_2}$, that is,
$f_\phi\si_{\cA_1}=\si_{\cA_2}f_\phi$. The map $f_\phi$ is called a
$k$-block factor map. Note that each shift is itself a $2$-block factor
map.
\begin{lemma}\label{block-factor-0}
Given an infinite string $\xi_1=x_1x_2x_3\ldots\in\cA_1^{\bN}$, let
$\xi_2=y_1y_2y_3\ldots$ be an infinite string in $\cA_2^{\bN}$ such that
$y_n=\phi(x_n,x_{n+1},\dots,x_{n+k-1})$ for all $n\ge1$. Then
$f_\phi(\Om(\xi_1))=\Om(\xi_2)$.
\end{lemma}
\begin{proof}
Take any left-infinite string $\zeta_1\in\cA_1^{-\bN}$. Then
$f_\phi(\zeta_1.\xi_1)=\zeta_2.\xi_2$ for some left-infinite string
$\zeta_2\in\cA_2^{-\bN}$. By Lemma \ref{limit-points}, $\Om(\xi_1)$ is the
set of all limit points of the sequence $\zeta_1.\xi_1,
\si_{\cA_1}(\zeta_1.\xi_1),\si_{\cA_1}^2(\zeta_1.\xi_1),\dots$ while
$\Om(\xi_2)$ is the set of all limit points of the sequence $\zeta_2.\xi_2,
\si_{\cA_2}(\zeta_2.\xi_2),\si_{\cA_2}^2(\zeta_2.\xi_2),\dots$. Since
$f_\phi\si_{\cA_1}=\si_{\cA_2}f_\phi$, it follows by induction on $m$ that
$\si_{\cA_2}^m(\zeta_2.\xi_2)=
f_\phi\bigl(\si_{\cA_1}^m(\zeta_1.\xi_1)\bigr)$ for all $m\ge0$. Hence the
first sequence is mapped by $f_\phi$ onto the second sequence. Assume
$\om\in\Om(\xi_1)$. Then $\om$ is the limit of some subsequence
$\{\si_{\cA_1}^{m_i}(\zeta_1.\xi_1)\}_{i\ge1}$ of the first sequence.
Continuity of the map $f_\phi$ implies that
$\si_{\cA_2}^{m_i}(\zeta_2.\xi_2)\to f_\phi(\om)$ as $i\to\infty$ so that
$f_\phi(\om)\in\Om(\xi_2)$. Therefore
$f_\phi(\Om(\xi_1))\subset\Om(\xi_2)$.
Conversely, assume $\om'\in\Om(\xi_2)$. Let
$\{\si_{\cA_2}^{m_j}(\zeta_2.\xi_2)\}_{j\ge1}$ be a subsequence of the
second sequence converging to $\om'$. The respective subsequence
$\{\si_{\cA_1}^{m_j}(\zeta_1.\xi_1)\}_{j\ge1}$ of the first sequence need
not converge, but it has a limit point due to compactness of
$\cA_1^{\bZ}$. Any limit point $\om$ is also a limit point of the
first sequence so that $\om\in\Om(\xi_1)$. Besides, $f_\phi(\om)=\om'$ due
to continuity of the map $f_\phi$. We conclude that $\Om(\xi_2)\subset
f_\phi(\Om(\xi_1))$. Thus $f_\phi(\Om(\xi_1))=\Om(\xi_2)$.
\end{proof}
First consider a function $\phi_1:\{0,1\}^2\to\{x,y\}$ given by
$\phi_1(0,1)=\phi_1(1,0)=x$ and $\phi_1(0,0)=\phi_1(1,1)=y$.
\begin{lemma}\label{block-factor-1}
The $2$-block factor map $f_{\phi_1}$ is two-to-one when restricted to the
set $\Om(\xiTM)$, and $f_{\phi_1}(\Om(\xiTM))=\Om(\xipd)$.
\end{lemma}
\begin{proof}
For any $n\ge1$ let $x_n$ denote the $n$-th symbol of $\xiTM$ and $y_n$
denote the $n$-th symbol of $\xipd$. First we show that
$y_n=\phi_1(x_n,x_{n+1})$ for all $n\ge1$. If $n$ is odd, the last digit
in the binary expansion of $n$ is $1$. Changing that digit to $0$, we
obtain the binary expansion of $n-1$. Hence the numbers of 1s in those
expansions differ by $1$. It follows from Lemma \ref{xi-TM} that
$x_{n+1}\ne x_n$. Then $\phi_1(x_n,x_{n+1})=x$. Lemma \ref{xi-pd} implies
that $y_n=x$ as well.
If $n$ is even, it is of the form $2^km$, where $k,m\ge1$ and $m$ is odd.
Then the last $k+1$ digits in the binary expansion of the number $n$ are 1
followed by $k$ 0s. The last $k+1$ digits in the binary expansion of $n-1$
are 0 followed by $k$ 1s. The other digits are the same in both
expansions. Hence the numbers of 1s in those expansions differ by $k-1$.
Lemma \ref{xi-TM} implies that $x_{n+1}\ne x_n$ if $k$ is even, and
$x_{n+1}=x_n$ if $k$ is odd. By Lemma \ref{xi-pd}, $y_n=x$ if $k$ is even,
and $y_n=y$ if $k$ is odd. Either way, $\phi_1(x_n,x_{n+1})=y_n$.
Since $y_n=\phi_1(x_n,x_{n+1})$ for all $n\ge1$, Lemma \ref{block-factor-0}
implies that $f_{\phi_1}(\Om(\xiTM))=\Om(\xipd)$. It remains to show that
the map $f_{\phi_1}$ is two-to-one when restricted to the set
$\Om(\xiTM)$. Consider the substitution $\mathsf{s}_{(01)}$ over the
alphabet $\{0,1\}$ given by $\mathsf{s}_{(01)}(0)=1$,
$\mathsf{s}_{(01)}(1)=0$. Let $f_{(01)}$ denote the induced $1$-block
factor map. Lemma \ref{block-factor-0} implies that
$f_{(01)}(\Om(\xiTM))=\Om\bigl(\mathsf{s}_{(01)}(\xiTM)\bigr)$. It follows
from Lemma \ref{TM-unique} that $\mathsf{s}_{(01)}(\xiTM)=\xiTM'$ and
$\Om(\xiTM')=\Om(\xiTM)$. Hence $f_{(01)}(\Om(\xiTM))=\Om(\xiTM)$. Note
that $\phi_1\bigl(\mathsf{s}_{(01)}(z_1),\mathsf{s}_{(01)}(z_2)\bigr)=
\phi_1(z_1,z_2)$ for all $z_1,z_2\in\{0,1\}$. It follows that
$f_{\phi_1}\bigl(f_{(01)}(\om)\bigr)=f_{\phi_1}(\om)$ for any string
$\om=\ldots z_{-1}z_0.z_1z_2\ldots$ in $\{0,1\}^{\bZ}$. Clearly,
$f_{(01)}(\om)\ne\om$. Now assume that $f_{\phi_1}(\om')=f_{\phi_1}(\om)$
for some $\om'=\ldots z'_{-1}z'_0.z'_1z'_2\ldots$. Then
$\phi_1(z_n,z_{n+1})=\phi_1(z'_n,z'_{n+1})$ for all $n\in\bZ$. This
implies that $z'_n=z_n$ if and only if $z'_{n+1}=z_{n+1}$. It follows by
induction that either $z'_n=z_n$ for all $n$ or $z'_n\ne z_n$ for all $n$.
Hence $\om'=\om$ or $\om'=f_{(01)}(\om)$. Since
$f_{(01)}(\Om(\xiTM))=\Om(\xiTM)$, we conclude that the map $f_{\phi_1}$ is
two-to-one when restricted to $\Om(\xiTM)$.
\end{proof}
Next consider a function $\phi_2:\{a,c,d\}\to\{x,y\}$ given by
$\phi_2(a)=\phi_2(d)=x$ and $\phi_2(c)=y$.
\begin{lemma}\label{block-factor-2}
The $1$-block factor map $f_{\phi_2}$ is one-to-one when restricted to the
set $\Om(\xiGE)$, and $f_{\phi_2}(\Om(\xiGE))=\Om(\xipd)$.
\end{lemma}
\begin{proof}
For any $n\ge1$ let $x_n$ denote the $n$-th letter of $\xiGE$ and $y_n$
denote the $n$-th letter of $\xipd$. It follows from the descriptions of
the strings $\xipd$ and $\xiGE$ (Lemmas \ref{xi-pd} and \ref{xi-GE}) that
$\phi_2(x_n)=y_n$ for all $n\ge1$. Therefore
$f_{\phi_2}(\Om(\xiGE))=\Om(\xipd)$ due to Lemma \ref{block-factor-0}.
It remains to prove that the map $f_{\phi_2}$ is one-to-one when restricted
to $\Om(\xiGE)$. Consider any strings $\om=\ldots z_{-1}z_0.z_1z_2\ldots
\in\Om(\xipd)$ and $\om'=\ldots z'_{-1}z'_0.z'_1z'_2\ldots \in\Om(\xiGE)$
such that $f_{\phi_2}(\om')=\om$, that is, $\phi_2(z'_n)=z_n$ for all
$n\in\bZ$. We need to show that the string $\om'$ is uniquely determined
by $\om$. By Lemma \ref{xi-pd}, $y_n=y$ if $n$ is even but not divisible
by $4$. Since any four consecutive integers include such a number, it
follows that $xxxx$ is not a subword of $\xipd$. Then $xxxx$ is also not a
subword of $\om$. Hence the string $\om$ contains the letter $y$. Choose
any $k\in\bZ$ such that $z_k=y$. Further, Lemma \ref{xi-GE} implies that
$x_n=a$ if and only if $n$ is odd. Therefore the words $aa$, $cc$, $cd$,
$dc$ and $dd$ are not subwords of $\xiGE$. Then they are also not subwords
of $\om'$. As a consequence, either $z'_n=a$ if $n$ is odd and $z'_n\ne a$
if $n$ is even, or else $z'_n=a$ if $n$ is even and $z'_n\ne a$ if $n$ is
odd. Since $\phi_2(z'_k)=z_k=y$, we obtain that $z'_k=c$. It follows that
$z'_n=a$ if and only if $n-k$ is odd. In the case $n-k$ is even, we have
$z'_n=c$ if $z_n=y$ and $z'_n=d$ if $z_n=x$. Thus the string $\om'$ is
uniquely determined by the string $\om$.
\end{proof}
Now we are ready to complete the proof of Proposition
\ref{3-subshifts-factors}.
\begin{proof}[Proof of Proposition \ref{3-subshifts-factors}]
Since the map $f_{\phi_1}$ defined above is a $2$-block factor map of
$\{0,1\}^{\bZ}$ to $\{x,y\}^{\bZ}$, it is continuous and satisfies
$f_{\phi_1}\si_{\{0,1\}}=\si_{\{x,y\}}f_{\phi_1}$. By Lemma
\ref{block-factor-1}, $f_{\phi_1}(\Om(\xiTM))=\Om(\xipd)$. Hence the
subshift $\si_{\{x,y\}}|\Om(\xipd)$ is a continuous factor of the subshift
$\si_{\{0,1\}}|\Om(\xiTM)$. Besides, $f_{\phi_1}$ is two-to-one when
restricted to $\Om(\xiTM)$ so that $\si_{\{x,y\}}|\Om(\xipd)$ is a
two-to-one continuous factor of $\si_{\{0,1\}}|\Om(\xiTM)$.
The map $f_{\phi_2}$ defined above is a $1$-block factor map of
$\{a,c,d\}^{\bZ}$ to $\{x,y\}^{\bZ}$. Hence it is continuous and satisfies
$f_{\phi_2}\si_{\{a,c,d\}}=\si_{\{x,y\}}f_{\phi_1}$. By Lemma
\ref{block-factor-2}, $f_{\phi_2}$ is one-to-one when restricted to the
set $\Om(\xiGE)$, and $f_{\phi_2}(\Om(\xiGE))=\Om(\xipd)$. Hence the
restriction of $f_{\phi_2}$ to $\Om(\xiGE)$ is a continuous invertible map
onto $\Om(\xipd)$. Since $\Om(\xiGE)$ is a closed subset of the compact
topological space $\{a,c,d\}^{\bZ}$, it is compact as well. It follows
that the restriction of $f_{\phi_2}$ to $\Om(\xiGE)$ is a homeomorphism
onto $\Om(\xipd)$. Therefore the subshift $\si_{\{x,y\}}|\Om(\xipd)$ is
topologically conjugate to the subshift $\si_{\{a,c,d\}}|\Om(\xiGE)$. Then
$\si_{\{a,c,d\}}|\Om(\xiGE)$ is topologically conjugate to
$\si_{\{x,y\}}|\Om(\xipd)$ as well.
\end{proof}
%
\section{Topological full groups}\label{sect-TFG}
Let $X$ be a \emph{Cantor set}. This means that $X$ is a compact
metrizable topological space that is totally disconnected and has no
isolated points. The topology on the Cantor set is generated by
\emph{clopen} (i.e., both closed and open) sets. For instance, the space
$\cA^{\bZ}$ of bi-infinite strings over an alphabet $\cA$ is a Cantor set.
Each cylinder is a clopen subset of $\cA^{\bZ}$.
\begin{definition}\label{def-TFG}
Let $T:X\to X$ be a homeomorphism of the Cantor set $X$ onto itself. The
\textbf{topological full group} of $T$, denoted $[[T]]$, is a transformation
group consisting of all homeomorphisms $F:X\to X$ that can be given by
$F(x)=T^{\nu(x)}(x)$, $x\in X$ for some continuous function $\nu:X\to\bZ$.
\end{definition}
If a function $\nu:X\to\bZ$ is continuous then it is locally constant and
takes only finitely many values. Then nonempty level sets of $\nu$ form a
finite partition of the Cantor set $X$ into clopen sets. Thus every
element of the topological full group $[[T]]$ is ``piecewise'' a power of
$T$. The group $[[T]]$ is countable (as there are only countably many
clopen subsets of $X$).
The topological full group $[[T]]$ is \emph{ample}, which means that any
homeomorphism of the Cantor set $X$ that locally coincides with elements of
the group is itself in $[[T]]$. Moreover, it is the smallest ample
collection of homeomorphisms that contains the cyclic group generated by
$T$. For more details and proofs of these facts, see Section 2 of the
paper \cite{GrigVor24}.
In the case $T$ is a minimal homeomorphism of a Cantor set (such a
transformation is called a \emph{Cantor minimal system}), the (isomorphism
class of the) topological full group $[[T]]$ is an almost complete
invariant of the dynamics of $T$ as shown by Giordano, Putnam and Skau
\cite{GPS}.
\begin{theorem}[\cite{GPS}]\label{flip-conjugacy}
Suppose $T_1:X_1\to X_1$ and $T_2:X_2\to X_2$ are minimal homeomorphisms of
Cantor sets. Then the topological full groups $[[T_1]]$ and $[[T_2]]$ are
isomorphic if and only if $T_1$ is topologically conjugate to $T_2$ or
$T_2^{-1}$.
\end{theorem}
Next we show that Theorem \ref{flip-conjugacy} applies to the substitution
subshifts $\si_{\{0,1\}}|\Om(\xiTM)$, $\si_{\{x,y\}}|\Om(\xipd)$ and
$\si_{\{a,c,d\}}|\Om(\xiGE)$, and, moreover, the topological full groups of
these subshifts are complete invariants of their dynamics.
\begin{proposition}\label{3-subshifts-Cantor}
The subshifts $\si_{\{0,1\}}|\Om(\xiTM)$, $\si_{\{x,y\}}|\Om(\xipd)$ and
$\si_{\{a,c,d\}}|\Om(\xiGE)$ are Cantor minimal systems. Each of these
subshifts is topologically conjugate to its inverse.
\end{proposition}
Minimality of all three subshifts is already established by Proposition
\ref{3-subshifts-minimal}, but we need to show that $\Om(\xiTM)$,
$\Om(\xipd)$ and $\Om(\xiGE)$ are Cantor sets.
\begin{lemma}\label{finite-or-Cantor}
Let $T$ be a homeomorphism of a Cantor set $X$ and let $Y\subset X$ be a
nonempty closed set such that $T(Y)=Y$. Suppose that the restriction $T|Y$
is a minimal homeomorphism of $Y$. Then the subset $Y$ is either finite or
a Cantor set.
\end{lemma}
\begin{proof}
Since $Y$ is a closed subset of the Cantor set $X$, it is compact,
metrizable and totally disconnected. Therefore $Y$ is itself a Cantor set
if and only if it has no isolated points. Since $T$ is a homeomorphism of
$X$ and $T(Y)=Y$, isolated points of $Y$ are mapped to isolated points of
$Y$ while non-isolated points are mapped to non-isolated points. It
follows that $T(Y_0)=Y_0$, where $Y_0$ is the set of all non-isolated
points of $Y$. As $Y_0$ is a closed subset of $Y$, minimality of the
restriction $T|Y$ implies that $Y_0$ is the empty set or $Y$. In the case
$Y_0$ is empty, every point of $Y$ is isolated. Then compactness of $Y$
implies that this set is finite. In the case $Y_0=Y$, there are no
isolated points in $Y$. Then $Y$ is a Cantor set.
\end{proof}
\begin{lemma}\label{when-finite}
The set $\Om(\xi)$ is finite if and only if the infinite string $\xi$ is
eventually periodic.
\end{lemma}
\begin{proof}
Let $\xi=x_1x_2x_3\ldots$ be an infinite string over an alphabet $\cA$.
First assume that $\xi$ is eventually periodic. That is, $\xi=wuuu\ldots$
for some words $w,u\in\cA^*$, where $u\ne\varnothing$. Let $n_0$ be the
length of $w$ and $p$ be the length of $u$. Then for any left-infinite
string $\zeta\in\cA^{-\bN}$ the sequence
$\si^{n_0}(\zeta.\xi),\si^{n_0+p}(\zeta.\xi),\si^{n_0+2p}(\zeta.\xi),\dots$
converges in $\cA^{\bZ}$ to the bi-infinite string $\om=\ldots
uu.uu\ldots$. Consequently, for every $i$, $0\le i\le p-1$, we have
$\si^{n_0+i+kp}(\zeta.\xi)\to\si^i(\om)$ as $k\to\infty$. Any subsequence
of the sequence $\zeta.\xi,\si(\zeta.\xi),\si^2(\zeta.\xi),\dots$ shares
infinitely many terms with at least one of the subsequences
$\{\si^{n_0+i+kp}(\zeta.\xi)\}_{k\ge1}$, $0\le i\le p-1$. It follows that
$\om,\si(\om),\dots,\si^{p-1}(\om)$ are the only limit points of that
sequence. By Lemma \ref{limit-points}, these are also the only points of
the set $\Om(\xi)$ so that $\Om(\xi)$ is finite.
Conversely, assume that the set $\Om(\xi)$ is finite. Take any
$\om\in\Om(\xi)$. Since $\si(\Om(\xi))=\Om(\xi)$, it follows that
$\si^p(\om)=\om$ for some $p\ge1$. Then $\om=\ldots uu.uu\ldots$, where
$u$ is a word of length $p$. We claim that for some $k\ge1$ the string
$\xi$ has no subwords of the form $u^kv=uu\ldots uv$, where $v$ is a word
of length $p$ different from $u$. Assume the contrary. Then there exists
a sequence of words $v_1,v_2,v_3,\dots$ of length $p$ different from $u$
such that for any $k\ge1$ the word $u^kv_k$ is a subword of $\xi$. This
means existence of nonnegative integers $n_1,n_2,n_3,\dots$ such that
$x_{n_k+1}x_{n_k+2}\ldots x_{n_k+(k+1)p}=u^kv_k$ for all $k\ge1$. Take
any left-infinite string $\zeta\in\cA^{-\bN}$. Since $\cA^{\bZ}$ is
compact, the sequence $\{\si^{n_k+kp}(\zeta.\xi)\}_{k\ge1}$ has a limit
point $\om'$. By construction, $\om'=\ldots uuu.v\eta$ for some word $v$
of length $p$ different from $u$ and some $\eta\in\cA^{\bN}$. Since
$n_k+kp\to\infty$ as $k\to\infty$, the string $\om'$ is also a limit point
of the sequence $\zeta.\xi,\si(\zeta.\xi),\si^2(\zeta.\xi),\dots$. Hence
$\om'\in\Om(\xi)$ due to Lemma \ref{limit-points}. Then each of the
strings $\si^p(\om'),\si^{2p}(\om'),\si^{3p}(\om'),\dots$ belongs to
$\Om(\xi)$ as well. It is easy to observe that all these strings are
distinct, which contradicts the fact that $\Om(\xi)$ is a finite set.
We have established that for some $k\ge1$ words of the form $u^kv$, where
$v$ is a word of length $p$ different from $u$, do not occur as subwords of
the string $\xi$. The word $u^k$ does occur as a subword of $\xi$ since it
is a subword of $\om$. Hence $x_{n+1}x_{n+2}\ldots x_{n+kp}=u^k$ for some
$n\ge0$. Note that whenever $x_{n+ip+1}x_{n+ip+2}\ldots x_{n+ip+p}=u$ for
$k$ consecutive values of $i$, the same should hold for the next value of
$i$. Now it follows by induction that $x_{n+ip+1}x_{n+ip+2}\ldots
x_{n+ip+p}=u$ for all $i\ge0$. Thus the string $\xi$ is eventually
periodic.
\end{proof}
For any word $w$, let $\overleftarrow{w}$ denote the same word written
backwards. The word $w$ is called a \emph{palindrome} if
$\overleftarrow{w}=w$.
\begin{lemma}\label{palindrome}
Suppose that any subword of a string $\xi\in\cA^{\bN}$ can be extended to a
palindrome subword of $\xi$. Then the subshift $\si_{\cA}|\Om(\xi)$ is
topologically conjugate to its own inverse.
\end{lemma}
\begin{proof}
Consider a map $R:\cA^{\bZ}\to\cA^{\bZ}$ that flips any bi-infinite string
$\ldots y_{-2}y_{-1}y_0.y_1y_2\ldots$ in $\cA^{\bZ}$ around turning it into
the string $\ldots y_2y_1y_0.y_{-1}y_{-2}\ldots$. The map $R$ is an
involution that maps cylinders onto cylinders. This implies that $R$ is a
homeomorphism. It is easy to see that $R\si_{\cA}=\si_{\cA}^{-1}R$. It
follows that for any subshift $\si_{\cA}|Y$, the restriction
$\si_{\cA}|R(Y)$ is also a subshift. Moreover, $\si_{\cA}|R(Y)$ is
topologically conjugate to the inverse of $\si_{\cA}|Y$. Hence to prove
the lemma, it is enough to show that $R(\Om(\xi))=\Om(\xi)$. A string
$\om\in\cA^{\bZ}$ belongs to $\Om(\xi)$ if and only if every subword of
$\om$ is also a subword of $\xi$. By construction, a word $u$ is a subword
of $R(\om)$ if and only if $\overleftarrow{u}$ is a subword of $\om$. It
follows that a string $\eta\in\cA^{\bZ}$ belongs to $R(\Om(\xi))$ if and
only if for every subword $u$ of $\eta$, the word $\overleftarrow{u}$ is a
subword of $\xi$.
If a word $u$ is a subword of a word $w$, then the word $\overleftarrow{u}$
is a subword of $\overleftarrow{w}$. In the case $w$ is a palindrome, $u$
is a subword of $w$ if and only if $\overleftarrow{u}$ is a subword of
$w$. Since any subword of the string $\xi$ can be extended to a palindrome
subword of $\xi$, any word $u\in\cA^*$ is a subword of $\xi$ if and only if
$\overleftarrow{u}$ is a subword of $\xi$. Now it follows from the above
that $R(\Om(\xi))=\Om(\xi)$.
\end{proof}
\begin{proof}[Proof of Proposition \ref{3-subshifts-Cantor}]
By Proposition \ref{3-subshifts-minimal}, the subshifts
$\si_{\{0,1\}}|\Om(\xiTM)$, $\si_{\{x,y\}}|\Om(\xipd)$ and
$\si_{\{a,c,d\}}|\Om(\xiGE)$ are minimal. Let us prove that $\Om(\xiTM)$,
$\Om(\xipd)$ and $\Om(\xiGE)$ are Cantor sets. In view of Lemma
\ref{finite-or-Cantor}, it is enough to show that all three sets are
infinite. By Proposition \ref{3-subshifts-factors}, the subshift
$\si_{\{x,y\}}|\Om(\xipd)$ is a continuous factor of
$\si_{\{0,1\}}|\Om(\xiTM)$ while the subshift $\si_{\{a,c,d\}}|\Om(\xiGE)$
is topologically conjugate to $\si_{\{x,y\}}|\Om(\xipd)$. This means, in
particular, that the set $\Om(\xiTM)$ can be mapped onto $\Om(\xipd)$ while
$\Om(\xipd)$ can be mapped onto $\Om(\xiGE)$. It follows that the sets
$\Om(\xiTM)$ and $\Om(\xipd)$ are infinite if $\Om(\xiGE)$ is infinite.
It remains to prove that the set $\Om(\xiGE)$ is infinite. In view of
Lemma \ref{when-finite}, it is enough to show that the string $\xiGE$ is
not eventually periodic. If an infinite string $\xi=x_1x_2x_3\ldots$ is
eventually periodic with period $p$, then for any $n\ge1$ the sequence of
letters $x_n,x_{n+p},x_{n+2p},\dots$ is eventually constant. We are going
to show that the latter is not the case when $\xi=\xiGE$. Any integer
$p\ge1$ can be represented as $2^km$, where $k\ge0$, $m\ge1$ and $m$ is
odd. Let $n_p=2^k$. Then for any $s\ge1$ the number
$n_p+2sp=2^k+2^{k+1}sm$ is divisible by $2^k$ but not divisible by
$2^{k+1}$. Since $m$ is odd, the number $m^2$ leaves remainder $1$ after
division by $4$, and so does the number $(4s+1)m^2$ for any $s\ge1$.
Therefore the number $n_p+(4s+1)mp=2^k+2^k(4s+1)m^2$ is divisible by
$2^{k+1}$ but not divisible by $2^{k+2}$. Now it follows from Lemma
\ref{xi-GE} that $x_{n_p+2sp}=x_{2^k}$ for all $s\ge1$,
$x_{n_p+(4s+1)mp}=x_{2^{k+1}}$ for all $s\ge1$, and $x_{2^{k+1}}\ne
x_{2^k}$. As a consequence, the sequence
$x_{n_p},x_{n_p+p},x_{n_p+2p},\dots$ is not eventually constant.
We proceed to the second statement of the proposition. Suppose
$\mathsf{s}$ is a substitution over an alphabet $\cA$ that substitutes
every symbol of $\cA$ with a palindrome. Then for any word $w=x_1x_2\ldots
x_n\in\cA^*$,
\[
\overleftarrow{\mathsf{s}(w)}=
\overleftarrow{\mathsf{s}(x_1)\mathsf{s}(x_2)\ldots\mathsf{s}(x_n)}
=\overleftarrow{\mathsf{s}(x_n)}\ldots\overleftarrow{\mathsf{s}(x_2)}
\,\overleftarrow{\mathsf{s}(x_1)}= \mathsf{s}(x_n)\ldots\mathsf{s}(x_2)
\mathsf{s}(x_1)=\mathsf{s}\bigl(\overleftarrow{w}\bigr).
\]
As a consequence, the substitution $\mathsf{s}$ transforms palindromes into
palindromes. Note that the words $\sub(a)=aca$, $\sub(c)=d$ and
$\sub(d)=c$ are palindromes. The words $\TM^2(0)=0110$ and $\TM^2(1)=1001$
are also palindromes. By the above the substitutions $\sub$ and $\TM^2$
transform palindromes into palindromes. It follows by induction that for
any $k\ge1$ the words $\sub^k(a)$ and $\TM^{2k}(0)$ are palindromes.
Recall that each $\sub^k(a)$ is a subword of the string $\xiGE$. Moreover,
any subword of $\xiGE$ is also a subword of $\sub^k(a)$ for all
sufficiently large $k$. Likewise, each $\TM^{2k}(0)$ is a subword of the
string $\xiTM$ and any subword of $\xiTM$ is also a subword of
$\TM^{2k}(0)$ for all sufficiently large $k$. Now it follows from Lemma
\ref{palindrome} that each of the subshifts $\si_{\{a,c,d\}}|\Om(\xiGE)$
and $\si_{\{0,1\}}|\Om(\xiTM)$ is topologically conjugate to its inverse.
By Proposition \ref{3-subshifts-factors}, the subshift
$\si_{\{a,c,d\}}|\Om(\xiGE)$ is topologically conjugate to the subshift
$\si_{\{x,y\}}|\Om(\xipd)$. By the above $\si_{\{a,c,d\}}|\Om(\xiGE)$ is
topologically conjugate to its inverse. Hence there exist homeomorphisms
$f_1:\Om(\xipd)\to\Om(\xiGE)$ and $f_2:\Om(\xiGE)\to\Om(\xiGE)$ such that
$f_1\si_{\{x,y\}}=\si_{\{a,c,d\}}f_1$ on $\Om(\xipd)$ and
$f_2\si_{\{a,c,d\}}=\si_{\{a,c,d\}}^{-1}f_2$ on $\Om(\xiGE)$. Then the map
$f=f_1^{-1}f_2f_1$ is a homeomorphism of $\Om(\xipd)$ onto itself. It is
easy to check that $f\si_{\{x,y\}}=\si_{\{x,y\}}^{-1}f$ on $\Om(\xipd)$.
Thus the subshift $\si_{\{x,y\}}|\Om(\xipd)$ is topologically conjugate to
its inverse.
\end{proof}
\begin{definition}\label{def-aperiodic}
A transformation $T:X\to X$ of a set $X$ is called \textbf{aperiodic} if it
has no periodic points, that is, $T^n(x)\ne x$ for all points $x\in X$ and
integers $n\ge1$.
\end{definition}
The main result of this section is the following embedding theorem.
\begin{theorem}\label{TFG-embeds}
Suppose that $T_1$ and $T_2$ are aperiodic homeomorphisms of Cantor sets
and $T_2$ is a continuous factor of $T_1$. Then the topological full group
$[[T_2]]$ embeds into $[[T_1]]$.
\end{theorem}
To prove Theorem \ref{TFG-embeds}, we are going to use an isomorphic model
of the group $[[T]]$, where $T$ is an aperiodic homeomorphism of a Cantor
set (onto itself). The model is due to Putnam \cite{Putnam89} (see the
construction of a group $G$ right before Theorem 5.2 in \cite{Putnam89})
and it actually predates the group $[[T]]$.
\begin{lemma}\label{variable-power}
Suppose $T$ is a homeomorphism of a Cantor set $X$ and $\ell:X\to\bZ$ is a
continuous function. Then the map $h_{T,\ell}:X\to X$ defined by
$h_{T,\ell}(x)=T^{\ell(x)}(x)$, $x\in X$, is continuous.
\end{lemma}
\begin{proof}
For any $n\in\bZ$ the map $h_{T,\ell}$ coincides with a continuous map
$T^n$ on the set $\ell^{-1}(n)$. Since the function $\ell$ is continuous,
the set $\ell^{-1}(n)$ is clopen, which implies that $h_{T,\ell}$ is
continuous on that set as well. As $X$ is the disjoint union of the sets
$\ell^{-1}(n)$, $n\in\bZ$, the lemma follows.
\end{proof}
Let $C(X,\bZ)$ denote the set of all continuous functions $\ell:X\to\bZ$.
Given a homeomorphism $T$ of a Cantor set $X$, for any functions
$\ell_1,\ell_2\in C(X,\bZ)$ we define a function
$\ell_1\oplus_T\ell_2:X\to\bZ$ by $(\ell_1\oplus_T\ell_2)(x)=
\ell_1\bigl(T^{\ell_2(x)}(x)\bigr)+\ell_2(x)$ for all $x\in X$. Note that
the formula involves addition and composition of functions. We never use
multiplication of functions in what follows.
\begin{lemma}
$\oplus_T$ is a well defined operation on $C(X,\bZ)$.
\end{lemma}
\begin{proof}
We need to show that for any continuous functions $\ell_1,\ell_2:X\to\bZ$,
the function $\ell_1\oplus_T\ell_2$ is continuous as well. Note that
$\ell_1\oplus_T\ell_2=\ell_1h_{T,\ell_2}+\ell_2$. The map $h_{T,\ell_2}$
is continuous due to Lemma \ref{variable-power}. Then $\ell_1h_{T,\ell_2}$
is continuous as a composition of two continuous functions. Finally,
$\ell_1\oplus_T\ell_2$ is continuous as the sum of two continuous functions
$\ell_1h_{T,\ell_2}$ and $\ell_2$.
\end{proof}
\begin{lemma}\label{h-composed}
$h_{T,\ell_1\oplus_T\ell_2}=h_{T,\ell_1}h_{T,\ell_2}$ for all
$\ell_1,\ell_2\in C(X,\bZ)$.
\end{lemma}
\begin{proof}
For any $x\in X$ we have $(\ell_1\oplus_T\ell_2)(x)=
\ell_1(h_{T,\ell_2}(x))+\ell_2(x)$. Then
\[
h_{T,\ell_1\oplus_T\ell_2}(x)=T^{\ell_1(h_{T,\ell_2}(x))+\ell_2(x)}(x)
=T^{\ell_1(h_{T,\ell_2}(x))}T^{\ell_2(x)}(x).
\]
On the other hand,
\[
h_{T,\ell_1}(h_{T,\ell_2}(x))=T^{\ell_1(h_{T,\ell_2}(x))}h_{T,\ell_2}(x)
=T^{\ell_1(h_{T,\ell_2}(x))}T^{\ell_2(x)}(x).
\]
Hence $h_{T,\ell_1\oplus_T\ell_2}=h_{T,\ell_1}h_{T,\ell_2}$ everywhere on
$X$.
\end{proof}
\begin{lemma}
$\oplus_T$ is an associative operation on $C(X,\bZ)$.
\end{lemma}
\begin{proof}
For any functions $\ell,\ell'\in C(X,\bZ)$, we have $\ell\oplus_T\ell'=\ell
h_{T,\ell'}+\ell'$. Besides, $h_{T,\ell\oplus_T\ell'}=
h_{T,\ell}h_{T,\ell'}$ due to Lemma \ref{h-composed}. Given arbitrary
functions $\ell_1,\ell_2,\ell_3\in C(X,\bZ)$, we obtain
\begin{gather*}
(\ell_1\oplus_T\ell_2)\oplus_T\ell_3=
(\ell_1h_{T,\ell_2}+\ell_2)\oplus_T\ell_3=
(\ell_1h_{T,\ell_2}+\ell_2)h_{T,\ell_3}+\ell_3=
\ell_1h_{T,\ell_2}h_{T,\ell_3}+\ell_2h_{T,\ell_3}+\ell_3, \\
\ell_1\oplus_T(\ell_2\oplus_T\ell_3)=
\ell_1h_{T,\ell_2\oplus_T\ell_3}+(\ell_2\oplus_T\ell_3)=
\ell_1h_{T,\ell_2}h_{T,\ell_3}+\ell_2h_{T,\ell_3}+\ell_3.
\end{gather*}
Thus $(\ell_1\oplus_T\ell_2)\oplus_T\ell_3=
\ell_1\oplus_T(\ell_2\oplus_T\ell_3)$.
\end{proof}
Let $\0_X$ denote the zero function on $X$. It is easy to see that $\0_X$
is the identity element of the operation $\oplus_T$. Hence $C(X,\bZ)$ is a
monoid with respect to $\oplus_T$. We denote by $C^\times_T(X,\bZ)$ the
set of all invertible elements of this monoid. Then $C^\times_T(X,\bZ)$ is
a group with respect to the operation $\oplus_T$.
Next we define a map $H_T:C(X,\bZ)\to C(X,X)$ by $H_T(\ell)=h_{T,\ell}$ for
all $\ell\in C(X,\bZ)$. The map $H_T$ is well defined due to Lemma
\ref{variable-power}. In view of Lemma \ref{h-composed}, $H_T$ is a
homomorphism of the monoid $C(X,\bZ)$ with the operation $\oplus_T$ to the
monoid $C(X,X)$ with the operation of composition.
\begin{lemma}\label{image-of-group}
$H_T\bigl(C^\times_T(X,\bZ)\bigr)=[[T]]$.
\end{lemma}
\begin{proof}
It follows from Definition \ref{def-TFG} that
$[[T]]=H_T(C(X,\bZ))\cap\Homeo(X)$, where $\Homeo(X)$ is the group of all
homeomorphisms of $X$. Hence it is enough to show that for any function
$\ell\in C(X,\bZ)$, the map $H_T(\ell)=h_{T,\ell}$ is a homeomorphism if
and only if $\ell$ is invertible with respect to the operation $\oplus_T$.
First assume that $\ell$ is invertible, and let $\ell'$ be its inverse. We
have $\ell\oplus_T\ell'=\ell'\oplus_T\ell=\0_X$. Then it follows from
Lemma \ref{h-composed} that $h_{T,\ell}h_{T,\ell'}=h_{T,\ell'}h_{T,\ell}
=h_{T,\0_X}=\id_X$. Hence the map $h_{T,\ell}$ is invertible and its
inverse $h_{T,\ell'}$ is continuous (due to Lemma \ref{variable-power}).
Therefore $h_{T,\ell}$ is a homeomorphism.
Conversely, assume that $h_{T,\ell}$ is a homeomorphism of $X$. Define a
function $\ell':X\to\bZ$ by $\ell'(x)=-\ell\bigl(h_{T,\ell}^{-1}(x)\bigr)$
for all $x\in X$. Since $h_{T,\ell}$ is a homeomorphism, the function
$\ell'$ is well defined and continuous. Hence $\ell'\in C(X,\bZ)$. By
construction,
\[
\ell'\oplus_T\ell=\ell'h_{T,\ell}+\ell=
(-\ell)h_{T,\ell}^{-1}h_{T,\ell}+\ell= -\ell+\ell=\0_X.
\]
Then $h_{T,\ell'}h_{T,\ell}=h_{T,\0_X}=\id_X$ due to Lemma
\ref{h-composed}. Since the map $h_{T,\ell}$ is invertible, it follows
that $h_{T,\ell'}=h_{T,\ell}^{-1}$. Consequently, $\ell\oplus_T\ell'=\ell
h_{T,\ell'}+\ell'=\ell h_{T,\ell}^{-1}+\ell'=-\ell'+\ell'=\0_X$. We
conclude that the function $\ell$ is invertible with respect to the
operation $\oplus_T$ and $\ell'$ is its inverse.
\end{proof}
\begin{lemma}\label{model-isomorphism}
Suppose that the homeomorphism $T$ is aperiodic. Then the homomorphism
$H_T$ is one-to-one. Moreover, the restriction of $H_T$ to the group
$C^\times_T(X,\bZ)$ is an isomorphism onto the group $[[T]]$.
\end{lemma}
\begin{proof}
If $T$ is aperiodic, then $T^n(x)\ne x$ for all points $x\in X$ and
positive integers $n$. As $T$ is invertible, it follows that for any $x\in
X$ and $n_1,n_2\in\bZ$ we have $T^{n_1}(x)\ne T^{n_2}(x)$ whenever $n_1\ne
n_2$. In particular, $h_{T,\ell_1}(x)=T^{\ell_1(x)}(x)\ne
T^{\ell_2(x)}(x)=h_{T,\ell_2}(x)$ whenever $\ell_1(x)\ne\ell_2(x)$. As a
consequence, the homomorphism $H_T$ is one-to-one.
By Lemma \ref{image-of-group}, $H_T\bigl(C^\times_T(X,\bZ)\bigr)=[[T]]$.
Hence the restriction of $H_T$ to the group $C^\times_T(X,\bZ)$ is a
homomorphism onto the group $[[T]]$. In the case when the homomorphism
$H_T$ is one-to-one, the restriction is an isomorphism onto $[[T]]$.
\end{proof}
Any continuous function $f:X_1\to X_2$ induces a map $\Phi_f:C(X_2,\bZ)\to
C(X_1,\bZ)$ given by $\Phi_f(\ell)=\ell f$ for all $\ell\in C(X_2,\bZ)$.
\begin{lemma}\label{monoid-embed}
Suppose $T_1:X_1\to X_1$ and $T_2:X_2\to X_2$ are homeomorphisms of Cantor
sets and $f:X_1\to X_2$ is a continuous function such that $fT_1=T_2f$.
Then the map $\Phi_f$ is a homomorphism of the monoid $C(X_2,\bZ)$ with
operation $\oplus_{T_2}$ to the monoid $C(X_1,\bZ)$ with operation
$\oplus_{T_1}$.
\end{lemma}
\begin{proof}
First we prove that $fT_1^n=T_2^nf$ for all $n\in\bZ$. For $n\ge0$, the
proof is by induction on $n$. Indeed, the case $n=0$ is trivial, and if
$fT_1^k=T_2^kf$ for some $k\ge0$ then
\[
fT_1^{k+1}=(fT_1^k)T_1=(T_2^kf)T_1=T_2^k(fT_1)=T_2^k(T_2f)=T_2^{k+1}f.
\]
In the case $n<0$, we have $-n>0$ so that $fT_1^{-n}=T_2^{-n}f$ by the
above. Then
\[
fT_1^n=(T_2^nT_2^{-n})fT_1^n=T_2^n(T_2^{-n}f)T_1^n=T_2^n(fT_1^{-n})T_1^n
=T_2^nf(T_1^{-n}T_1^n)=T_2^nf.
\]
Given functions $\ell_1,\ell_2\in C(X_2,\bZ)$, we need to show that
$\Phi_f(\ell_1\oplus_{T_2}\ell_2)=\Phi_f(\ell_1)\oplus_{T_1}
\Phi_f(\ell_2)$. For any $x\in X_1$ we obtain
\begin{gather*}
\bigl(\Phi_f(\ell_1\oplus_{T_2}\ell_2)\bigr)(x)=
(\ell_1\oplus_{T_2}\ell_2)(f(x))=
\ell_1\bigl(T_2^{\ell_2(f(x))}(f(x))\bigr)+\ell_2(f(x)), \\
\bigl(\Phi_f(\ell_1)\oplus_{T_1}\Phi_f(\ell_2)\bigr)(x)=
(\ell_1f\oplus_{T_1}\ell_2f)(x)=
\ell_1f\bigl(T_1^{\ell_2f(x)}(x)\bigr)+\ell_2f(x).
\end{gather*}
Since $\ell_2(f(x))=\ell_2f(x)$ is an integer, the map $fT_1^{\ell_2f(x)}$
coincides with the map $T_2^{\ell_2f(x)}f$. It follows that
$\bigl(\Phi_f(\ell_1\oplus_{T_2}\ell_2)\bigr)(x)=
\bigl(\Phi_f(\ell_1)\oplus_{T_1}\Phi_f(\ell_2)\bigr)(x)$.
\end{proof}
\begin{lemma}\label{group-embed}
Under the assumptions of Lemma \ref{monoid-embed}, the restriction of
$\Phi_f$ to the group $C^\times_{T_2}(X_2,\bZ)$ is a homomorphism to the
group $C^\times_{T_1}(X_1,\bZ)$. This group homomorphism is one-to-one
provided that the function $f$ is onto.
\end{lemma}
\begin{proof}
By Lemma \ref{monoid-embed}, the map $\Phi_f$ is a homomorphism of the
monoid $C(X_2,\bZ)$ with operation $\oplus_{T_2}$ to the monoid
$C(X_1,\bZ)$ with operation $\oplus_{T_1}$. Take any function $\ell\in
C(X_2,\bZ)$ invertible with respect to $\oplus_{T_2}$ and let $\ell'$ be
its inverse. Then $\ell\oplus_{T_2}\ell'=\ell'\oplus_{T_2}\ell
=\0_{X_2}$, which implies that $\Phi_f(\ell)\oplus_{T_1}\Phi_f(\ell')=
\Phi_f(\ell')\oplus_{T_1}\Phi_f(\ell)=\Phi_f(\0_{X_2})=\0_{X_1}$. Hence
the function $\Phi_f(\ell)$ is invertible with respect to $\oplus_{T_1}$
and $\Phi_f(\ell')$ is its inverse. We obtain that $\Phi_f$ maps
invertible elements of the monoid $C(X_2,\bZ)$ to invertible elements of
the monoid $C(X_1,\bZ)$. Consequently, the restriction of $\Phi_f$ to the
group $C^\times_{T_2}(X_2,\bZ)$ is a homomorphism to the group
$C^\times_{T_1}(X_1,\bZ)$.
If the map $f$ is onto then for any functions $\ell_1$ and $\ell_2$ on
$X_2$ we have $\ell_1f=\ell_2f$ if and only if $\ell_1=\ell_2$. It follows
that the map $\Phi_f$ is one-to-one, and so is its restriction to
$C^\times_{T_2}(X_2,\bZ)$.
\end{proof}
Now we are ready to prove Theorem \ref{TFG-embeds}.
\begin{proof}[Proof of Theorem \ref{TFG-embeds}]
Suppose $T_1:X_1\to X_1$ and $T_2:X_2\to X_2$ are aperiodic homeomorphisms
of Cantor sets. If $T_2$ is a continuous factor of $T_1$, there exists a
continuous map $f:X_1\to X_2$ such that $f$ is onto and $fT_1=T_2f$. Lemma
\ref{group-embed} implies that the group $C^\times_{T_2}(X_2,\bZ)$ with
operation $\oplus_{T_2}$ embeds into the group $C^\times_{T_1}(X_1,\bZ)$
with operation $\oplus_{T_1}$. Since the homeomorphisms $T_1$ and $T_2$
are aperiodic, it follows from Lemma \ref{model-isomorphism} that the group
$C^\times_{T_1}(X_1,\bZ)$ is isomorphic to the topological full group
$[[T_1]]$ while the group $C^\times_{T_2}(X_2,\bZ)$ is isomorphic to
$[[T_2]]$. We conclude that the group $[[T_2]]$ embeds into $[[T_1]]$.
\end{proof}
Cantor minimal systems are obviously aperiodic. In view of Proposition
\ref{3-subshifts-factors}, Theorem \ref{TFG-embeds} implies that the
topological full groups of the subshifts $\si|\Om(\xipd)$ and
$\si|\Om(\xiGE)$ embed into the topological full group of the subshift
$\si|\Om(\xiTM)$. In particular, every subgroup of $[[\si|\Omega(\xipd)]]$
or $[[\si|\Omega(\xiGE)]]$ embeds into $[[\si|\Omega(\xiTM)]]$. In the
remainder of this section we are going to describe a subgroup of the group
$[[\si|\Omega(\xiGE)]]$ that will be later proved to have intermediate
growth.
We begin with a general construction of involutions in a topological full
group.
\begin{definition}\label{def-2-cycle}
Let $U$ be a clopen subset of a Cantor set $X$. For any homeomorphism
$T:X\to X$ such that the image $T(U)$ is disjoint from $U$, we define a
\textbf{generalized 2-cycle} $\de_{U;T}:X\to X$ by
\[
\de_{U;T}(x)=
\begin{cases}
T(x) & \text{if $x\in U$,}\\
T^{-1}(x) & \text{if $x\in T(U)$,}\\
x & \text{otherwise.}
\end{cases}
\]
\end{definition}
The generalized $2$-cycle $\de_{U;T}$ maps disjoint clopen sets $U$ and
$T(U)$ onto each other while fixing the other points of $X$. By
construction, $\de_{U;T}$ is continuous and $\de_{U;T}^{-1}=\de_{U;T}$.
Therefore, whenever $\de_{U;T}$ is defined, it belongs to the topological
full group $[[T]]$.
Given an alphabet $\cA$, for any letter $z\in\cA$ let $[.z]_{\cA}$ denote
the set of all bi-infinite strings $\ldots x_{-1}x_0.x_1x_2\ldots$ in
$\cA^{\bZ}$ such that $x_1=z$, and let $[z.]_{\cA}$ denote the set of all
strings such that $x_0=z$. The sets $[.z]_{\cA}$ and $[z.]_{\cA}$ are
cylinders in $\cA^{\bZ}$. Clearly, the shift $\si_{\cA}$ maps $[.z]_{\cA}$
onto $[z.]_{\cA}$. Let $\si_{\cA}|Y$ be a subshift such that $Y$ is a
Cantor set. Suppose that the letter $z$ is never doubled in elements of
$Y$, that is, the word $zz$ is a subword of no string in $Y$. Then the
intersection $[.z]_{\cA}\cap [z.]_{\cA}\cap Y$ is empty. Since
$\si_{\cA}(Y)=Y$, it follows that the set $[.z]_{\cA}\cap Y$, which is a
clopen subset of the Cantor set $Y$, is disjoint from
$\si_{\cA}([.z]_{\cA}\cap Y)=[z.]_{\cA}\cap Y$. Hence the generalized
$2$-cycle $\de_{U;T}$, where $U=[.z]_{\cA}\cap Y$ and $T=\si_{\cA}|Y$, is
defined. We will use the short notation $\de_z$ for this involution.
The generalized $2$-cycle $\de_z$ is uniquely determined when its domain
$Y$ is specified. It belongs to the topological full group of the subshift
$\si|Y$. The action of $\de_z$ on a bi-infinite string $\ldots
x_{-1}x_0.x_1x_2\ldots$ can be described as follows. If one of two letters
$x_0$ and $x_1$ adjacent to the dot is $z$, then the dot is carried across
that letter. Otherwise the string is not changed.
It follows from Lemma \ref{xi-GE} that for any two consecutive letters in
the infinite string $\xiGE$, one is $a$ while the other one is $c$ or $d$.
Hence the string $\xiGE$ contains no double letters, that is, the words
$aa$, $cc$ and $dd$ are not subwords of $\xiGE$. Consequently, the same is
true for any bi-infinite string in $\Om(\xiGE)$. Therefore each of the
generalized $2$-cycles $\de_a$, $\de_c$ and $\de_d$ is defined on
$\Om(\xiGE)$.
\begin{theorem}\label{sub-main}
The subgroup of the topological full group $[[\si|\Omega(\xiGE)]]$
generated by involutions $\de_a$, $\de_c$ and $\de_d$ has intermediate
growth.
\end{theorem}
To prove Theorem \ref{sub-main}, we are going to identify a known group of
intermediate growth that is isomorphic to the subgroup in the theorem.
This will be done in the next section.
%
\section{Groups of intermediate growth}\label{sect-growth}
Let $G$ be an infinite, finitely generated group. Suppose $A$ is a finite
generating set for $G$. Let $A_\pm$ denote the subset of $G$ consisting of
all generators in $A$ and their inverses. Any word over the alphabet
$A_\pm$ can be interpreted as a product in the group $G$. This gives rise
to a map $h_{G,A}:A_\pm^*\to G$, which is a homomorphism of the monoid
$A_\pm^*$ onto $G$. For any element $g\in G$, the length of the shortest
word in the pre-image $h_{G,A}^{-1}(g)$ is called the \emph{length} of $g$
(relative to the generating set $A$). For any integer $n\ge1$ let
$\ga_{G,A}(n)$ be the number of all elements of the group $G$ of length at
most $n$. The function $\ga_{G,A}$ is called the \emph{growth function} of
$G$. The asymptotic behavior of $\ga_{G,A}(n)$ as $n\to\infty$ is referred
to as the \emph{growth} of the group $G$. Clearly, $\ga_{G,A}(n)$ does not
exceed the number of words of length at most $n$ in $A_\pm^*$, which is
$1+m+m^2+\dots+m^n=(m^{n+1}-1)/(m-1)$, where $m$ is the number of elements
in $A_\pm$. Hence the growth of $G$ is at most exponential.
\begin{definition}\label{def-growth}
Let $G$ be an infinite, finitely generated group with a finite generating
set $A$. We say that the group $G$ has \textbf{exponential growth} if
$\ga_{G,A}(n)\ge c^n$ for some $c>0$ and all sufficiently large $n$. We
say that the group $G$ has \textbf{polynomial growth} if $\ga_{G,A}(n)\le
n^\alpha$ for some $\alpha>0$ and all sufficiently large $n$. We say that
the group $G$ has \textbf{intermediate growth} if the growth of $G$ is
slower than exponential but faster than polynomial, namely, for any
$c,\alpha>0$ we have $n^\alpha<\ga_{G,A}(n)<c^n$ for all sufficiently large
$n$.
\end{definition}
Definition \ref{def-growth} is not sensitive to the choice of a generating
set for the group $G$. Suppose $B$ is a different finite generating set
for $G$. Let $k_1$ be the maximal length of elements of $B$ relative to
the generating set $A$. Then any product of $n$ elements of $B$ or their
inverses can be written as a product of at most $k_1n$ elements of $A$ or
their inverses. Therefore $\ga_{G,B}(n)\le\ga_{G,A}(k_1n)$ for all
$n\ge1$. Likewise, $\ga_{G,A}(n)\le\ga_{G,B}(k_2n)$ for all $n\ge1$, where
$k_2$ is the maximal length of elements of $A$ relative to the generating
set $B$. It follows that whenever the growth function $\ga_{G,A}$ is of
exponential, polynomial or intermediate growth, the growth function
$\ga_{G,B}$ exhibits the same kind of growth.
Note that if an infinite, finitely generated group is not of exponential or
polynomial growth, it need not have intermediate growth. Indeed, there are
groups with growth oscillating between exponential and sub-exponential.
Now we are going to describe a family of groups of intermediate growth
constructed by the first author in \cite{Grig84}. Each group $\cG_\om$ of
the family is labeled by an infinite ternary string
$\om\in\{0,1,2\}^{\bN}$. The group $\cG_\om$ is generated by a set $A_\om$
consisting of four transformations $a$, $b_\om$, $c_\om$ and $d_\om$. In
the original construction in \cite{Grig84}, the generators were acting on
the interval $[0,1]$. We consider a modification in which the generators
act on the set $\{0,1\}^{\bN}$ of infinite binary strings. This
modification agrees with the original construction when we identify any
binary string $x_1x_2x_3\ldots$ with the number
$2^{-1}x_1+2^{-2}x_2+2^{-3}x_3+\dots$.
The generator $a$ is common for all groups in the family. It is defined by
$a(0\xi)=1\xi$ and $a(1\xi)=0\xi$ for all $\xi\in\{0,1\}^{\bN}$. That is,
$a$ changes the first symbol in every binary string. The generators
$b_\om$, $c_\om$ and $d_\om$ are defined using the following recursive
rules: for any $\xi\in\{0,1\}^{\bN}$, $l\in\{0,1,2\}$ and
$\om\in\{0,1,2\}^{\bN}$ we have
\begin{eqnarray*}
& b_{2\om}(0\xi)=c_{1\om}(0\xi)=d_{0\om}(0\xi)=0\xi,\\
& b_{0\om}(0\xi)=b_{1\om}(0\xi)=c_{0\om}(0\xi)=c_{2\om}(0\xi)
=d_{1\om}(0\xi)=d_{2\om}(0\xi)=0a(\xi),\\
& b_{l\om}(1\xi)=1b_\om(\xi), \qquad c_{l\om}(1\xi)=1c_\om(\xi), \qquad
d_{l\om}(1\xi)=1d_\om(\xi).
\end{eqnarray*}
The string $1^\infty=111\ldots$ is the only element of $\{0,1\}^{\bN}$
fixed by $b_\om$, $c_\om$ and $d_\om$ for all $\om\in\{0,1,2\}^{\bN}$. Any
other string $\xi\in\{0,1\}^{\bN}$ is uniquely represented as
$\xi=1^k0x\eta$, where $x\in\{0,1\}$, $\eta\in\{0,1\}^{\bN}$, $k\ge0$, and
$1^k$ denotes the symbol $1$ repeated $k$ times ($1^0=\varnothing$). Then
two of the transformations $b_\om$, $c_\om$ and $d_\om$ change only the
symbol $x$ in the string $\xi$ while the third one does not change $\xi$ at
all. Which one of the three fixes $\xi$ is determined by the $(k+1)$-th
symbol of the ternary string $\om$. Namely, $\xi$ is fixed by $b_\om$,
$c_\om$ or $d_\om$ if the $(k+1)$-th symbol of $\om$ is respectively $2$,
$1$ or $0$.
It follows from the construction that $a$, $b_\om$, $c_\om$ and $d_\om$ are
involutions. Besides, $b_\om$, $c_\om$ and $d_\om$ commute with one
another, and $b_\om c_\om=d_\om$, $b_\om d_\om=c_\om$ and $c_\om
d_\om=b_\om$. Therefore one of the transformations $b_\om$, $c_\om$ and
$d_\om$ can be dropped from the generating set $A_\om$ as redundant.
It is easy to see that for any $\xi\in\{0,1\}^{\bN}$ and any $n\ge1$ the
first $n$ symbols of strings $a(\xi$), $b_\om(\xi)$, $c_\om(\xi)$ and
$d_\om(\xi)$ are uniquely determined by the first $n$ symbols of the string
$\xi$ and the first $n-1$ symbols of the string $\om$. As a consequence,
for any $n\ge1$ the natural action of the group $\cG_\om$ on
$\{0,1\}^{\bN}$ induces an action on binary words of length $n$ regarded as
prefixes of infinite binary strings. Thus we obtain a length-preserving
action of $\cG_\om$ on the set $\{0,1\}^*$ of all binary words. Note that
the action of the generators $a$, $b_\om$, $c_\om$ and $d_\om$ can be given
by the same recursive formulas as their action on $\{0,1\}^{\bN}$, only
this time $\xi$ is an arbitrary word in $\{0,1\}^*$ and we also need to let
$a(\varnothing)=b_\om(\varnothing)= c_\om(\varnothing)=d_\om(\varnothing)
=\varnothing$ for all $\om\in\{0,1,2\}^{\bN}$.
\begin{lemma}\label{transitive-on-levels}
The group $\cG_\om$ acts transitively on binary words of any given length.
\end{lemma}
\begin{proof}
The proof is by induction on the length $k$ of the binary words,
simultaneously for all $\om\in\{0,1,2\}$. The case $k=0$ is trivial as
$\varnothing$ is the only word of length $0$. Now assume that the lemma
holds for words of some length $k\ge0$ and every $\om\in\{0,1,2\}$. Any
binary word of length $k+1$ can be written as $0w$ or $1w$, where $w$ is a
word of length $k$. Any infinite ternary string $\om$ can be written as
$l\om'$, where $l\in\{0,1,2\}$ and $\om'\in\{0,1,2\}^{\bN}$. We obtain
that $b_\om(1w)=1b_{\om'}(w)$, $c_\om(1w)=1c_{\om'}(w)$ and
$d_\om(1w)=1d_{\om'}(w)$. Further, out of three words $b_\om(0w)$,
$c_\om(0w)$ and $d_\om(0w)$, two coincide with $0a(w)$ while the third one
is $0w$. Consequently, two of three words $ab_\om a(1w)$, $ac_\om a(1w)$
and $ad_\om a(1w)$ coincide with $1a(w)$. Let $S'$ be the set of all
elements $g'\in\cG_{\om'}$ such that $g(1w)=1g'(w)$ for some $g\in\cG_\om$
and all words $w$ of length $k$. It is easy to see that $S'$ is a subgroup
of $\cG_{\om'}$. By the above, $a,b_{\om'},c_{\om'},d_{\om'}\in S'$.
Therefore $S'=\cG_{\om'}$. By the inductive assumption, the group
$\cG_{\om'}$ acts transitively on words of length $k$. It follows that all
words of length $k+1$ that begin with $1$ are in the same orbit of the
action of $\cG_\om$. Since the generator $a$ changes only the first symbol
in each word of length $k+1$, we conclude that the group $\cG_\om$ acts
transitively on binary words of length $k+1$.
\end{proof}
The set $\{0,1\}^{\bN}$ of infinite binary strings is endowed with the
product topology. The topology is generated by the cylinder sets of the
form $w\{0,1\}^{\bN}=\bigl\{w\eta\mid\eta\in\{0,1\}^{\bN}\bigr\}$, where
$w\in\{0,1\}^*$. The fact that the action of the group $\cG_\om$ on
$\{0,1\}^{\bN}$ induces a length-preserving action on finite binary strings
implies that $\cG_\om$ acts on $\{0,1\}^{\bN}$ by homeomorphisms.
\begin{lemma}\label{density-of-orbits}
Every orbit of the group $\cG_\om$ is dense in $\{0,1\}^{\bN}$.
\end{lemma}
\begin{proof}
Let $O$ be any orbit of $\cG_\om$. By Lemma \ref{transitive-on-levels},
the group $\cG_\om$ acts transitively on binary words of any given length.
It follows that any binary word $w$ occurs as a prefix of some string in
$O$. In other words, the orbit $O$ has a point in the cylinder
$w\{0,1\}^{\bN}$. Thus $O$ is dense in $\{0,1\}^{\bN}$.
\end{proof}
In the case a ternary string $\om\in\{0,1,2\}^{\bN}$ is eventually
constant, it is not hard to show that the group $\cG_\om$ has a cyclic
subgroup of finite index. Hence the growth of $\cG_\om$ is polynomial in
this case. In all other cases, the growth turns out to be intermediate.
\begin{theorem}[\cite{Grig84}]\label{Gri-growth}
If a string $\om\in\{0,1,2\}^{\bN}$ is not eventually constant, then the
group $\cG_\om$ has intermediate growth. Moreover,
$\gamma_{\cG_\om,A_\om}(n)\ge\exp(\alpha\sqrt{n})$ for some $\alpha>0$ and
all sufficiently large $n$. If, additionally, for some $m\ge1$ any $m$
consecutive symbols of $\om$ include $0$, $1$ and $2$ then
$\gamma_{\cG_\om,A_\om}(n)\le\exp(n^\beta)$ for some $\beta<1$ and all
sufficiently large $n$.
\end{theorem}
In this paper we are interested in the group $\cG_{(01)^\infty}$ labeled by
the periodic string $(01)^\infty=010101\ldots$. The growth of this group
is much faster than the upper bound in Theorem \ref{Gri-growth}. The
precise rate of growth was established by Erschler \cite{Erschler04} as a
particular case of the following general result.
\begin{theorem}[\cite{Erschler04}]\label{GE-growth}
Suppose that a string $\om\in\{0,1,2\}^{\bN}$ contains no $2$ and for
some $m\ge1$ any $m$ consecutive symbols of $\om$ include both $0$ and
$1$. Then
\[
\exp\left(\frac{n}{\log^{2+\eps}(n)}\right)\le \gamma_{\cG_\om,A_\om}(n)\le
\exp\left(\frac{n}{\log^{1-\eps}(n)}\right)
\]
for any $\eps>0$ and all sufficiently (depending on $\eps$) large $n$.
\end{theorem}
Note that our notation for the groups $\cG_\om$ and their generators is
different from Erschler's notation (cf.\@ Remark 1 in \cite{Erschler04}).
We reformulated Theorem \ref{GE-growth} (Corollary 2$'$ in
\cite{Erschler04}) accordingly. The group $\cG_{(01)^\infty}$ was the
primary example for the theorem. It turns out that $\cG_{(01)^\infty}$ is
isomorphic to the subgroup of the topological full group
$[[\si|\Omega(\xiGE)]]$ constructed at the end of Section \ref{sect-TFG}.
\begin{proposition}\label{inj-homomorph}
There exists a one-to-one group homomorphism
$h:\cG_{(01)^\infty}\to[[\si|\Omega(\xiGE)]]$ such that $h(a)=\de_a$,
$h(c)=\de_c$ and $h(d)=\de_d$, where $c=c_{(01)^\infty}$,
$d=d_{(01)^\infty}$.
\end{proposition}
%
\section{Schreier graphs}\label{sect-Schr}
In this section we prove Proposition \ref{inj-homomorph}, which will allow
us to prove Theorem \ref{sub-main} as well as Theorem \ref{main}. The
proof of Proposition \ref{inj-homomorph} is going to rely on the use of the
Schreier graphs.
A \emph{graph} is a combinatorial object that consists of \emph{vertices}
and \emph{edges} related so that every edge joins two vertices or a vertex
to itself (in the latter case the edge is called a \emph{loop} at that
vertex). Vertices joined by an edge are called its \emph{endpoints}. We
allow multiple edges joining the same pair of vertices as well as multiple
loops at the same vertex. We consider graphs with \emph{undirected} edges,
which means that the two endpoints of a non-loop edge are not ordered.
Also, our graphs have labeled edges, that is, every edge is assigned a
\emph{label}.
The vertices of a graph are pictured as dots or small circles. An edge is
pictured as an arc joining its endpoints. The label of an edge is written
next to the edge.
\begin{definition}\label{def-covering}
Let $\Gamma_1$ be a graph with vertex set $V_1$ and edge set $E_1$. Let
$\Gamma_2$ be a graph with vertex set $V_2$ and edge set $E_2$. An onto
map $f:V_1\to V_2$ is called a \textbf{covering} of the graph $\Gamma_2$ by
the graph $\Gamma_1$ (and then $\Gamma_1$ is said to \textbf{cover}
$\Gamma_2$) if there exists an onto map $h:E_1\to E_2$ such that for any
edge $e\in E_1$ the endpoints of $h(e)$ are images of the endpoints of $e$
under $f$ and the label of $h(e)$ matches the label of $e$. The covering
$f:V_1\to V_2$ is called an \textbf{isomorphism} of the graph $\Gamma_1$
onto the graph $\Gamma_2$ (and then $\Gamma_1$ is said to be
\textbf{isomorphic} to $\Gamma_2$) if $f$ is bijective and the associated
map $h:E_1\to E_2$ can be chosen to be bijective.
\end{definition}
A \emph{path} of length $n\ge0$ in a graph $\Gamma$ is a sequence of its
vertices $v_0,v_1,\dots,v_n$ along with a sequence of edges $e_1,\dots,e_n$
such that for any $i$, $1\le i\le n$ the edge $e_i$ joins the vertex
$v_{i-1}$ to $v_i$. We say that the path connects the vertex $v_0$ to
$v_n$. The graph $\Gamma$ is called \emph{connected} if any two vertices
of $\Gamma$ are connected by a path. The length of the shortest path
connecting a vertex $u$ to a vertex $v$ is called the \emph{distance} from
$u$ to $v$ in the graph. This distance function turns the vertex set of a
connected graph into a metric space.
\begin{definition}\label{def-Schr}
Let $G$ be a group with a finite generating set $A$. Let
$\alpha:G\curvearrowright X$ be an action of $G$ on a set $X$ and $O$ be an
orbit of that action. The \textbf{Schreier graph}
$\Gamma_{\Sch}(G,A;\al,O)$ of the orbit $O$ of the action $\alpha$ of the
group $G$ relative to the generating set $A$ is a graph with vertex set $O$
and edges labeled by elements of $A$. For any $u,v\in O$ and $l\in A$ the
graph $\Gamma_{\Sch}(G,A;\al,O)$ has a (unique) edge with label $l$
connecting the vertex $u$ to $v$ if and only if $lu=v$ or $lv=u$ within the
action $\al$. In the case the generators in $A$ are indexed by letters of
an alphabet $\cA$, we might regard edge labels in
$\Gamma_{\Sch}(G,A;\al,O)$ as elements of $\cA$.
\end{definition}
If $l\in A$ is a generator of the group $G$ then any vertex $v\in O$ of the
Schreier graph $\Gamma_{\Sch}(G,A;\al,O)$ is joined to $lv$ and $l^{-1}v$
by edges with label $l$. It follows that the graph
$\Gamma_{\Sch}(G,A;\al,O)$ is connected. Moreover, for any $v\in O$ and
$g\in G$ the distance between vertices $v$ and $gv$ in this graph does not
exceed the length of the element $g$ relative to the generating set $A$.
We denote by $\Sch(G,A)$ the set of all Schreier graphs of the form
$\Gamma_{\Sch}(G,A;\al,O)$. If a generator $l\in A$ is an involution
($l^{-1}=l$), then the conditions $lu=v$ and $lv=u$ are equivalent for all
$u,v\in O$. Hence the action of $l$ on the orbit $O$ is uniquely recovered
from the graph $\Gamma_{\Sch}(G,A;\al,O)$. In the case when all generators
in $A$ are involutions, the graph $\Gamma_{\Sch}(G,A;\al,O)$ allows to
reconstruct the action of the entire group $G$ on $O$. Therefore in that
case, elements of the set $\Sch(G,A)$ encode all transitive actions of the
group $G$.
\begin{definition}\label{def-I_A}
Given an alphabet $\cA$, we define a group $\cI_{\cA}$ by
$\cI_{\cA}=\langle\cA\mid z^2=1,\ z\in\cA\rangle$. That is, elements of
$\cA$ are considered nontrivial involutions, and the group $\cI_{\cA}$ is
freely generated by them.
\end{definition}
\begin{lemma}\label{Sch-I_A}
A graph $\Gamma$ with labeled edges belongs to $\Sch(\cI_{\cA},\cA)$ if and
only if the following conditions hold:
\begin{itemize}
\item
the graph $\Gamma$ is connected;
\item
all edge labels are in $\cA$;
\item
for any vertex $v$ of $\Gamma$ and any $z\in\cA$ there is a unique edge
with $v$ as an endpoint and labeled by $z$.
\end{itemize}
\end{lemma}
\begin{proof}
Any Schreier graph is connected. For a graph of the form
$\Gamma_{\Sch}(\cI_{\cA},\cA;\al,O)$, all edge labels are in $\cA$.
Furthermore, for any $u,v\in O$ and $z\in\cA$, the relations $zu=v$ and
$zv=u$ are equivalent since $z$ is an involution. Hence for any $v\in O$
and $z\in\cA$, the vertex $v$ is an endpoint of a unique edge with label
$z$, which joins $v$ to $zv$.
Conversely, assume that a graph $\Gamma$ satisfies all three conditions of
the lemma. Let $V$ be the vertex set of $\Gamma$. For any $z\in\cA$ and
$v\in V$, the vertex $v$ is an endpoint of a unique edge with label $z$.
Let $T_z(v)$ denote the vertex of $\Gamma$ joined to $v$ by that edge.
This defines a family of transformations $T_z$, $z\in\cA$ of the set $V$.
Since edges of the graph $\Gamma$ are undirected, it follows that each
$T_z$ is an involution. Then there exists a homomorphism
$\phi:\cI_{\cA}\to\mathrm{Sym}(V)$ of the group $\cI_{\cA}$ to the group
$\mathrm{Sym}(V)$ of all invertible transformations of $V$ such that
$\phi(z)=T_z$ for all $z\in\cA$. The homomorphism $\phi$ induces an action
$\al:\cI_{\cA}\curvearrowright V$ given by $gv=\bigl(\phi(g)\bigr)(v)$ for
all $g\in\cI_{\cA}$ and $v\in V$. Note that any two vertices of $\Gamma$
joined by an edge are in the same orbit of the action $\al$. Since the
graph $\Gamma$ is connected, it follows that the action is transitive. By
construction, $\Gamma=\Gamma_{\Sch}(\cI_{\cA},\cA;\al,V)$.
\end{proof}
Let $\xi$ be a word $x_1x_2\ldots x_n\in\cA^*$, an infinite string
$x_1x_2x_3\ldots\in\cA^{\bN}$ or a bi-infinite string $\ldots
x_{-1}x_0.x_1x_2\ldots\in\cA^{\bZ}$. Suppose that there are no double
letters in $\xi$. Then we define a graph $\Gamma_{\cA}(\xi)$ with labeled
edges as follows. The vertex set $V$ of the graph is $\{0,1,\dots,n\}$ if
$\xi$ is a word of length $n\ge0$, $\{0,1,2,\dots\}$ if $\xi$ is an
infinite string, and $\bZ$ if $\xi$ is a bi-infinite string. For any
$i\in\bZ$ such that $\{i-1,i\}\subset V$, we connect the vertex $i$ to
$i-1$ by a unique edge with label $x_i$. These are the only non-loop edges
of $\Gamma_{\cA}(\xi)$. Further, for any $i\in V$ and $x\in\cA$ such that
the vertex $i$ is not connected to $i-1$ or $i+1$ by an edge with label
$x$, we add a unique loop at $i$ with label $x$. These are the only loops
of $\Gamma_{\cA}(\xi)$.
By construction, $\Gamma_{\cA}(\xi)$ is a connected graph with a linear
structure. All edge labels are letters in $\cA$. Since $\xi$ contains no
double letters, for any $i\in V$ and $x\in\cA$ the vertex $i$ is an
endpoint of a unique edge with label $x$. Hence the graph
$\Gamma_{\cA}(\xi)$ belongs to $\Sch(\cI_{\cA},\cA)$ due to Lemma
\ref{Sch-I_A}. For examples of such graphs, see Figures \ref{fig-lvl123},
\ref{fig-halfline} and \ref{fig-line}.
Suppose $\si_{\cA}|Y$ is a subshift over an alphabet $\cA$ such that for
any letter $z\in\cA$ the generalized $2$-cycle $\de_z$ is defined on $Y$.
Let $G_Y$ be the subgroup of the topological full group $[[\si_{\cA}|Y]]$
generated by involutions $\de_z$, $z\in\cA$. The group $G_Y$ is uniquely
determined by $\cA$ and $Y$.
\begin{lemma}\label{Schr-subshift}
Under the above assumptions, for any bi-infinite string $\xi\in Y$ the
orbit of $\xi$ under the action of the group $G_Y$ is
$\{\si_{\cA}^i(\xi)\mid i\in\bZ\}$. If the string $\xi$ is not periodic
then the Schreier graph of its orbit relative to the generating set
$\{\de_z\mid z\in\cA\}$ is isomorphic to $\Gamma_{\cA}(\xi)$.
\end{lemma}
\begin{proof}
Since $\xi\in Y$ and each of the generalized $2$-cycles $\de_z$, $z\in\cA$
is defined on $Y$, the string $\xi$ has no double letters. Let $\xi=\ldots
x_{-1}x_0.x_1x_2\ldots$. Then $\si_{\cA}^i(\xi)=\ldots
x_{i-1}x_i.x_{i+1}x_{i+2}\ldots$ for any $i\in\bZ$. In the Schreier graph
of the orbit of $\si_{\cA}^i(\xi)$ under the action of the group $G_Y$, for
any $z\in\cA$ the vertex $\si_{\cA}^i(\xi)$ is an endpoint of a unique edge
with label $z$. This edge joins $\si_{\cA}^i(\xi)$ to
$\si_{\cA}^{i-1}(\xi)$ if $z=x_i$, to $\si_{\cA}^{i+1}(\xi)$ if
$z=x_{i+1}$, and to itself otherwise.
Let $O$ be the orbit of the string $\xi$ under the action of the group
$G_Y$ and $\Gamma$ be the Schreier graph of that orbit relative to the
generating set $\{\de_z\mid z\in\cA\}$ (with edge labels in $\cA$). Since
each generator is piecewise a power of the subshift $\si_{\cA}|Y$, the set
$\{\si_{\cA}^i(\xi)\mid i\in\bZ\}$ is invariant under the action of $G_Y$.
Therefore the orbit $O$ is contained in this set. On the other hand, it
follows from the above that for any $i\in\bZ$ the strings
$\si_{\cA}^i(\xi)$ and $\si_{\cA}^{i-1}(\xi)$ are in the same orbit of the
group $G_Y$. We conclude that $O=\{\si_{\cA}^i(\xi)\mid i\in\bZ\}$.
Consider a map $f:\bZ\to O$ given by $f(i)=\si_{\cA}^i(\xi)$ for all
$i\in\bZ$. The map $f$ is onto. It follows from the definition of the
graph $\Gamma_{\cA}(\xi)$ that $f$ is a covering of the Schreier graph
$\Gamma$ by $\Gamma_{\cA}(\xi)$. In the case when the string $\xi$ is not
periodic, the map $f$ is bijective. Then $f$ is an isomorphism of the
graph $\Gamma_{\cA}(\xi)$ onto $\Gamma$.
\end{proof}
Our next goal is to describe some Schreier graphs of the group
$\cG_{(01)^\infty}$ relative to the generating set $\{a,c,d\}$ (see Lemmas
\ref{Schr-level} and \ref{Schr-half-line} below).
\begin{figure}[t]
\centerline{\includegraphics[scale=1.25]{fig-edgesub.eps}}
\caption{Edge substitution acting on a non-loop edge with label $a$.}
\label{fig-edgesub}
\end{figure}
\begin{lemma}\label{edge-sub}
Suppose $w$ is a nonempty word without double letters over the alphabet
$\{a,c,d\}$. Then the graph $\Gamma_{\{a,c,d\}}(w)$ can be transformed
into a graph isomorphic to $\Gamma_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ via the
following edge substitution procedure:
\begin{itemize}
\item
every old edge with label $a$ joining two different vertices $u$ and $v$ is
removed; instead we insert two new vertices $u'$ and $v'$, and add five new
edges: an edge with label $a$ joining $u$ to $u'$, an edge with label $c$
joining $u'$ to $v'$, an edge with label $a$ joining $v'$ to $v$, and two
loops with label $d$ at the vertices $u'$ and $v'$ (see Figure
\ref{fig-edgesub});
\item
every old loop with label $a$ is preserved;
\item
for every old edge with label $c$, the label is changed to $d$;
\item
for every old edge with label $d$, the label is changed to $c$.
\end{itemize}
\end{lemma}
\begin{proof}
The procedure transforms the graph $\Gamma_{\{a,c,d\}}(w)$ into a graph
$\Gamma$. First we show that the graph $\Gamma$ satisfies all three
conditions in Lemma \ref{Sch-I_A} so that
$\Gamma\in\Sch(\cI_{\{a,c,d\}},\{a,c,d\})$. Any two vertices of the graph
$\Gamma_{\{a,c,d\}}(w)$ joined by an edge are also connected in $\Gamma$
(by an edge or by a path of length $3$). Any newly added vertex in
$\Gamma$ is connected to an old vertex by an edge. Since
$\Gamma_{\{a,c,d\}}(w)$ is a connected graph, it follows that the graph
$\Gamma$ is connected as well. Any vertex $v$ of $\Gamma_{\{a,c,d\}}(w)$
is an endpoint of three edges, which are labeled $a$, $c$ and $d$. In the
graph $\Gamma$, the edges with labels $c$ and $d$ are retained while their
labels are exchanged. The edge with label $a$ is retained if it is a
loop. Otherwise it is removed but another edge with an endpoint $v$ is
added, which has label $a$. Any newly added vertex of $\Gamma$ is an
endpoint of three edges, labeled $a$, $c$ and $d$. Thus the three
conditions in Lemma \ref{Sch-I_A} are satisfied.
Let $\Gamma'$ and $\Gamma'_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ be graphs
obtained from respectively $\Gamma$ and
$\Gamma_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ by removing all loops. The action
of the edge substitution on non-loop edges mimics the action of
the substitution $\sub$ on labels of those edges. It follows that the
graphs $\Gamma'$ and $\Gamma'_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ are
isomorphic. Since the graphs $\Gamma$ and
$\Gamma_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ are both in
$\Sch(\cI_{\{a,c,d\}},\{a,c,d\})$, for any vertex $v$ of either of the
graphs and any letter $z\in\{a,c,d\}$ the loop with label $z$ at $v$ exists
if and only if there is no edge with label $z$ joining $v$ to another
vertex. As a consequence, any isomorphism of the graph $\Gamma'$ onto
$\Gamma'_{\{a,c,d\}}\bigl(\sub(w)\bigr)$ is also an isomorphism of $\Gamma$
onto $\Gamma_{\{a,c,d\}}\bigl(\sub(w)\bigr)$.
\end{proof}
\begin{figure}[t]
\centerline{\includegraphics{fig-lvl123.eps}}
\caption{Graphs $\Gamma_{\{a,c,d\}}(a)$,
$\Gamma_{\{a,c,d\}}\bigl(\sub(a)\bigr)$ and
$\Gamma_{\{a,c,d\}}\bigl(\sub^2(a)\bigr)$.}
\label{fig-lvl123}
\end{figure}
\begin{lemma}\label{Schr-level}
The Schreier graph of the action of the group $\cG_{(01)^\infty}$ on binary
words of length $k\ge1$ is isomorphic to the graph
$\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$.
\end{lemma}
\begin{proof}
For any $k\ge1$ let $V_k$ denote the set of all binary words of length
$k$. By Lemma \ref{transitive-on-levels}, the group $\cG_{(01)^\infty}$
acts
transitively on $V_k$. Let $\Gamma_k$ be the Schreier graph of the only
orbit of that action. We prove that $\Gamma_k$ is isomorphic to the graph
$\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$ by induction on $k$. The
case $k=1$ is easy. Indeed, both $\Gamma_1$ and $\Gamma_{\{a,c,d\}}(a)$
have two vertices, $0$ and $1$. Since $a(0)=1$, $a(1)=0$, $c(0)=d(0)=0$
and $c(1)=d(1)=1$, it follows that the identity map on the set $\{0,1\}$ is
an isomorphism of the graph $\Gamma_1$ onto $\Gamma_{\{a,c,d\}}(a)$.
Now assume that the graph $\Gamma_k$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$ for some $k\ge1$. We need to
derive from this that the graph $\Gamma_{k+1}$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\sub^k(a)\bigr)$. Let $\Gamma$ be the graph
obtained from $\Gamma_k$ by applying the edge substitution procedure
described in Lemma \ref{edge-sub}. Since $\Gamma_k$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$, it follows from Lemma
\ref{edge-sub} that the graph $\Gamma$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\sub^k(a)\bigr)$. It remains to prove that
$\Gamma$ is also isomorphic to $\Gamma_{k+1}$. Let $V$ be the vertex set
of the graph $\Gamma$. Then $V_k\subset V$ and any vertex of $\Gamma$ not
in $V_k$ is joined to a vertex in $V_k$ by a unique edge, which has label
$a$. Since the generator $a$ fixes no nonempty word, the graph $\Gamma_k$
has no loops with label $a$. It follows that the graph $\Gamma$ has twice
as many vertices as $\Gamma_k$, and any vertex $v\in V_k$ is joined in
$\Gamma$ by an edge to a unique vertex $v'\in V\setminus V_k$. We let
$f(v)=1v$ and $f(v')=0v$. This defines a bijective map $f:V\to V_{k+1}$.
We are going to show that $f$ is an isomorphism of the graph $\Gamma$ onto
$\Gamma_{k+1}$. After analyzing the structure of $\Gamma$, this task can
be reformulated as follows. For any $u,v\in V_k$ we have to show that
$a(v)=u$ implies $c(0v)=0u$, $c(v)=u$ implies $d(1v)=1u$, and $d(v)=u$
implies $c(1v)=1u$. Equivalently, $c(0v)=0a(v)$, $c(1v)=1d(v)$ and
$d(1v)=1c(v)$ for all $v\in V_k$. Also, we have to show that $a(1v)=0v$
and $d(0v)=0v$ for all $v\in V_k$.
Recall that $c=c_{(01)^\infty}$ and $d=d_{(01)^\infty}$. It follows from
the definition of the groups $\cG_\om$ in Section \ref{sect-growth} that
$c_{(10)^\infty}=d_{(01)^\infty}=d$ and $d_{(10)^\infty}=c_{(01)^\infty}
=c$. Now for any $v\in V_k$ we obtain
\begin{eqnarray*}
& c(1v)=c_{(01)^\infty}(1v)=c_{0(10)^\infty}(1v)
=1c_{(10)^\infty}(v)=1d(v),\\
& d(1v)=d_{(01)^\infty}(1v)=d_{0(10)^\infty}(1v)
=1d_{(10)^\infty}(v)=1c(v),\\
& c(0v)=c_{(01)^\infty}(0v)=0a(v),\qquad
d(0v)=d_{(01)^\infty}(0v)=0v
\end{eqnarray*}
and, obviously, $a(1v)=0v$, just as required.
\end{proof}
\begin{lemma}\label{half-line-covers}
Suppose that for some infinite string $\xi$ and nonempty word $w$ without
double letters over an alphabet $\cA$, the graph $\Gamma_{\cA}(\xi)$ covers
the graph $\Gamma_{\cA}(w)$. Then the string $\xi$ is of the form
$wa_1\overleftarrow{w}a_2wa_3\overleftarrow{w}a_4\ldots$ or
$\overleftarrow{w}a_1wa_2\overleftarrow{w}a_3wa_4\ldots$, where
$\overleftarrow{w}$ is the word $w$ written backwards and each $a_i$ is a
letter in $\cA$.
\end{lemma}
\begin{proof}
Let $\xi=x_1x_2x_3\ldots$ and $w=y_1y_2\ldots y_n$, where each $x_i$ and
$y_j$ is in $\cA$. The vertex set $V$ of the graph $\Gamma_{\cA}(\xi)$ is
$\{0,1,2,\dots\}$. The vertex set $V_0$ of the graph $\Gamma_{\cA}(w)$ is
$\{0,1,\dots,n\}$. For any vertex $v$ of either of the graphs and any
$z\in\cA$, the vertex $v$ is an endpoint of a unique edge with label $z$ in
this graph. We denote that edge by $e_{v,z}$ for the graph
$\Gamma_{\cA}(\xi)$ and by $\eps_{v,z}$ for the graph $\Gamma_{\cA}(w)$.
Further, for any integer $i\ge1$ let $\tilde e_i$ denote the only edge of
$\Gamma_{\cA}(\xi)$ joining vertices $i$ and $i-1$. In the case $i\le n$,
let $\tilde\eps_i$ denote the only edge joining the same vertices in the
graph $\Gamma_{\cA}(w)$. Note that the edge $\tilde e_i$ has label $x_i$.
Hence $\tilde e_i=e_{i,x_i}=e_{i-1,x_i}$. Likewise, the edge
$\tilde\eps_i$ has label $y_i$ so that
$\tilde\eps_i=\eps_{i,y_i}=\eps_{i-1,y_i}$.
Suppose $f:V\to V_0$ is a covering of the graph $\Gamma_{\cA}(w)$ by the
graph $\Gamma_{\cA}(\xi)$ and $h:E\to E_0$ is the associated map of the
edge set $E$ of $\Gamma_{\cA}(\xi)$ onto the edge set $E_0$ of
$\Gamma_{\cA}(w)$. Then $h(e_{v,z})=\eps_{f(v),z}$ for all $v\in V$ and
$z\in\cA$. Hence for any $v\in V$ the map $h$ establishes a one-to-one
correspondence between edges of $\Gamma_{\cA}(\xi)$ that have $v$ as an
endpoint and edges of $\Gamma_{\cA}(w)$ that have $f(v)$ as an endpoint.
For any integer $i\ge1$, the edge $\tilde e_i$ joins vertices $i$ and $i-1$
in the graph $\Gamma_{\cA}(\xi)$. Therefore the edge $h(\tilde e_i)$ joins
vertices $f(i)$ and $f(i-1)$ in the graph $\Gamma_{\cA}(w)$. It follows
that $f(i-1)$ is either $f(i)-1$ or $f(i)$ or $f(i)+1$. If $f(i-1)=f(i)-1$
then $h(\tilde e_i)=\tilde\eps_{f(i)}$. Hence $x_i=y_{f(i)}$ since the
edge $h(\tilde e_i)$ has the same label as $\tilde e_i$. If
$f(i-1)=f(i)+1$ then $h(\tilde e_i)=\tilde\eps_{f(i)+1}$ so that
$x_i=y_{f(i)+1}$.
Any vertex $v\in V$ is an endpoint of one or two non-loop edges in $E$.
Likewise, $f(v)$ is an endpoint of one or two non-loop edges in $E_0$. We
are going to explore all possibilities here. First assume that
$0<f(v)<n$. Then $f(v)$ is an endpoint of two non-loop edges
$\tilde\eps_{f(v)}$ and $\tilde\eps_{f(v)+1}$. We have
$h(e)=\tilde\eps_{f(v)}$ and $h(e')=\tilde\eps_{f(v)+1}$ for some edges
$e,e'\in E$ that have $v$ as an endpoint. Since $h$ clearly maps loops to
loops, neither $e$ nor $e'$ is a loop. It follows that $v>0$, one of the
edges $e$ and $e'$ is $\tilde e_v$, and the other is $\tilde e_{v+1}$. If
$h(\tilde e_v)=\tilde\eps_{f(v)}$ and $h(\tilde
e_{v+1})=\tilde\eps_{f(v)+1}$, then $f(v-1)=f(v)-1$ and $f(v+1)=f(v)+1$.
If $h(\tilde e_v)=\tilde\eps_{f(v)+1}$ and $h(\tilde
e_{v+1})=\tilde\eps_{f(v)}$, then $f(v-1)=f(v)+1$ and $f(v+1)=f(v)-1$.
Next consider the case $v=0$. It follows from the above that $f(0)=0$ or
$n$ so that $f(0)$ is an endpoint of a single non-loop edge $\eps\in E_0$.
We have $h(e)=\eps$ for some edge $e\in E$ that has $0$ as an endpoint.
The edge $e$ is not a loop. Therefore $e=\tilde e_1$. Then $f(1)$ is the
other endpoint of the edge $\eps$. That is, $f(1)=1$ if $f(0)=0$, and
$f(1)=n-1$ if $f(0)=n$.
Next assume that $v>0$ and $f(v)=0$. The edge $\tilde\eps_1$ is the only
non-loop edge in $E_0$ with an endpoint $0$. We have $h(e)=\tilde\eps_1$
for some edge $e\in E$ with an endpoint $v$. The edge $e$ is not a loop.
Therefore $e=\tilde e_v$ or $\tilde e_{v+1}$. If $f(\tilde
e_v)=\tilde\eps_1$ then $f(\tilde e_{v+1})\ne\tilde\eps_1$, which implies
that $f(\tilde e_{v+1})$ is a loop at $0$. Hence $f(v-1)=1$ and
$f(v+1)=0$. If $f(\tilde e_{v+1})=\tilde\eps_1$ then $f(\tilde e_v)$ is a
loop so that $f(v-1)=0$ and $f(v+1)=1$.
The only remaining possibility is $v>0$ and $f(v)=n$. Similarly to the
previous case, here we obtain that one of the vertices $f(v-1)$ and
$f(v+1)$ is $n-1$ while the other is $n$.
By the above any triple of consecutive elements of the sequence
$f(0),f(1),f(2),\dots$ is of the form $i-1,i,i+1$ or $i+1,i,i-1$ or $0,0,1$
or $1,0,0$ or $n-1,n,n$ or $n,n,n-1$. Besides, the pair $f(0),f(1)$ is
either $0,1$ or $n,n-1$. As a consequence, the sequence naturally splits
into blocks of $n+1$ consecutive elements where each block is either
$0,1,2,\dots,n-1,n$ or $n,n-1,\dots,2,1,0$ and adjacent blocks are never
the same. Any of these blocks is of the form
$f\bigl(m(n+1)\bigr),f\bigl(m(n+1)+1\bigr),\dots,f\bigl(m(n+1)+n\bigr)$ for
some integer $m\ge0$. If the block coincides with $0,1,2,\dots,n-1,n$ then
it follows from the above that the subword $x_{m(n+1)}x_{m(n+1)+1}\ldots
x_{m(n+1)+n-1}$ of the string $\xi$ coincides with the word $w$. If the
block coincides with $n,n-1,\dots,2,1,0$, it follows that the same subword
coincides with the word $\overleftarrow{w}$. We conclude that
$\xi=wa_1\overleftarrow{w}a_2wa_3\overleftarrow{w}a_4\ldots$ or
$\xi=\overleftarrow{w}a_1wa_2\overleftarrow{w}a_3wa_4\ldots$, where
$a_i=x_{(n+1)i-1}$ for all $i\ge1$.
\end{proof}
\begin{figure}[t]
\centerline{\includegraphics[scale=0.95]{fig-halfline.eps}}
\caption{Graph $\Gamma_{\{a,c,d\}}(\xiGE)$.}
\label{fig-halfline}
\end{figure}
\begin{lemma}\label{Schr-half-line}
The Schreier graph of the action of the group $\cG_{(01)^\infty}$ on the
orbit of the string $1^\infty$ is isomorphic to the graph
$\Gamma_{\{a,c,d\}}(\xiGE)$.
\end{lemma}
\begin{proof}
Let $O$ be the orbit of $1^\infty$ under the action of the group
$\cG_{(01)^\infty}$ and $\Gamma$ be the Schreier graph of that orbit
relative to the generating set $\{a,c,d\}$. We have
$a(1^\infty)=01^\infty$ and $c(1^\infty)=d(1^\infty)=1^\infty$. Hence
$1^\infty$ is joined to $01^\infty$ in the graph $\Gamma$ by an edge with
label $a$, and there are two loops at $1^\infty$ with labels $c$ and $d$.
Any string $\zeta\in O$ different from $1^\infty$ can be written as
$1^k0\eta$ for some $k\ge0$ and $\eta\in\{0,1\}^{\bN}$. Recall that
$c=c_{(01)^\infty}$ and $d=d_{(01)^\infty}$. If $k$ even then
$c(\zeta)=1^k0a(\eta)$ and $d(\zeta)=\zeta$. If $k$ odd then
$c(\zeta)=\zeta$ and $d(\zeta)=1^k0a(\eta)$. Since $cd=dc$, in either case
we have $cd(\zeta)=1^k0a(\eta)$. Note that $\zeta$, $a(\zeta)$ and
$cd(\zeta)$ are three different strings. Hence the vertex $\zeta$ of the
graph $\Gamma$ is joined to $a(\zeta)$ by an edge with label $a$, to
$cd(\zeta)$ by an edge with label $c$ or $d$, and to itself by one loop.
We obtained that every vertex $\zeta\ne1^\infty$ of the Schreier graph
$\Gamma$ is joined by edges to two other vertices while $1^\infty$ is
joined by an edge to only one other vertex. Besides, $\Gamma$ has no
multiple non-loop edges. Since the graph $\Gamma$ is connected, it follows
that it is isomorphic to the graph $\Gamma_{\{a,c,d\}}(\xi)$ for some
infinite string $\xi\in\{a,c,d\}^{\bN}$. Moreover, the isomorphism
$f:\{0,1,2,\dots\}\to O$ of $\Gamma_{\{a,c,d\}}(\xi)$ onto $\Gamma$ is
unique and satisfies $f(0)=1^\infty$. To be precise, it follows from the
above that $f(2m)=(cda)^m(1^\infty)$ and $f(2m+1)=a(cda)^m(1^\infty)$ for
all integers $m\ge0$. Now it remains to show that $\xi=\xiGE$.
For any $k\ge1$ let $V_k$ denote the set of all binary words of length
$k$. The action of the group $\cG_{(01)^\infty}$ on $V_k$ is transitive
due to Lemma \ref{transitive-on-levels}. Let $\Gamma_k$ be the Schreier
graph of the only orbit of that action. Consider a map $f_k:O\to V_k$ that
assigns to any infinite string in $O$ its prefix of length $k$. As the
orbit $O$ is dense in $\{0,1\}^{\bN}$ (due to Lemma
\ref{density-of-orbits}), the map $f_k$ is onto. Since the action of the
group $\cG_{(01)^\infty}$ on $V_k$ is induced by its action on
$\{0,1\}^{\bN}$, it follows that $f_k$ is a covering of the graph
$\Gamma_k$ by the Schreier graph $\Gamma$ of the orbit $O$. By Lemma
\ref{Schr-level}, the graph $\Gamma_k$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$. Since $\Gamma$ is
isomorphic to the graph $\Gamma_{\{a,c,d\}}(\xi)$, we obtain that the
latter covers the graph $\Gamma_{\{a,c,d\}}\bigl(\sub^{k-1}(a)\bigr)$.
Then Lemma \ref{half-line-covers} implies that the string $\xi$ is of the
form $wa_1\overleftarrow{w}a_2wa_3\overleftarrow{w}a_4\ldots$ or
$\overleftarrow{w}a_1wa_2\overleftarrow{w}a_3wa_4\ldots$, where
$w=\sub^{k-1}(a)$, $\overleftarrow{w}$ is $w$ written backwards, and each
$a_i$ is a letter in $\{a,c,d\}$. Note that $\overleftarrow{w}=w$ (as
shown in the proof of Proposition \ref{3-subshifts-Cantor}). We conclude
that $w=\sub^{k-1}(a)$ is a prefix of the string $\xi$. Thus $\xi=\xiGE$
since this is the only string that has each of the words $\sub^{k-1}(a)$,
$k\ge1$ as a prefix.
\end{proof}
\begin{definition}\label{def-kernel}
Let $\Gamma\in\Sch(\cI_{\cA},\cA)$ and $V$ be the vertex set of the graph
$\Gamma$. Then $\Gamma=\Gamma_{\Sch}(\cI_{\cA},\cA;\al_\Gamma,V)$ for a
unique transitive action $\al_\Gamma:\cI_{\cA}\curvearrowright V$. The
action $\al_\Gamma$ gives rise to a group homomorphism
$\psi_\Gamma:\cI_{\cA}\to\mathrm{Sym}(V)$, where $\mathrm{Sym}(V)$ is the
group of all invertible transformations of the set $V$. Namely,
$\psi_\Gamma(g)$ is how the element $g\in\cI_{\cA}$ acts on $V$. We denote
by $\cK(\Gamma)$ the \emph{kernel} of $\psi_\Gamma$. The kernel consists
of all elements of $\cI_{\cA}$ that act trivially on $V$ within the action
$\al_\Gamma$.
\end{definition}
Let $G$ be a group generated by a finite set $A$ of involutions. Suppose
that the generators of $G$ are labeled by elements of an alphabet $\cA$:
$A=\{g_z\mid z\in\cA\}$. Then for any Schreier graph in $\Sch(G,A)$ we
assume that edge labels are in $\cA$.
\begin{lemma}\label{factor-by-kernel}
Under the above assumptions, let $\Gamma=\Gamma_{\Sch}(G,A;\al,O)$ for some
orbit $O$ of an action $\al$ of the group $G$. Then
$\Gamma\in\Sch(\cI_{\cA},\cA)$ and there exists a group homomorphism
$\phi:G\to\cI_{\cA}/\cK(\Gamma)$ such that $\phi(g_z)=z\cK(\Gamma)$ for all
$z\in\cA$. In the case when the restriction of the action $\al$ to the
orbit $O$ is faithful, $\phi$ is an isomorphism.
\end{lemma}
\begin{proof}
Since the group $G$ is generated by involutions $g_z$, $z\in\cA$, there
exists a group homomorphism $\pi:\cI_{\cA}\to G$ such that $\pi(z)=g_z$ for
all $z\in\cA$. This homomorphism is onto. The restriction of the action
$\al$ to the orbit $O$ gives rise to a group homomorphism
$\psi_O:G\to\mathrm{Sym}(O)$. Then the composition $\psi=\psi_O\pi$ is a
homomorphism of the group $\cI_{\cA}$ to $\mathrm{Sym}(O)$. The
homomorphism $\psi$ induces an action $\beta:\cI_{\cA}\curvearrowright O$
given by $gv=\bigl(\psi(g)\bigr)(v)$ for all $g\in\cI_{\cA}$ and $v\in O$.
Since $O$ is an orbit of the action $\al$, the natural action of the group
$\psi_O(G)$ on $O$ is transitive. As the map $\pi$ is onto, we have
$\psi(\cI_{\cA})=\psi_O(G)$. Hence the action $\beta$ is transitive as
well.
Since $\psi(z)=\psi_O(g_z)$ for all $z\in\cA$, it follows that the graph
$\Gamma=\Gamma_{\Sch}(G,A;\al,O)$ is also the Schreier graph
$\Gamma_{\Sch}(\cI_{\cA},\cA;\beta,O)$. In particular,
$\Gamma\in\Sch(\cI_{\cA},\cA)$. Besides, $\cK(\Gamma)$ is the kernel of
the homomorphism $\psi$. Therefore $\psi$ factors to a group isomorphism
$\rho:\cI_{\cA}/\cK(\Gamma)\to\psi(\cI_{\cA})$ such that
$\rho\bigl(h\cK(\Gamma)\bigr)=\psi(h)$ for all $h\in\cI_{\cA}$. As
$\psi(\cI_{\cA})=\psi_O(G)$, the composition $\phi=\rho^{-1}\psi_O$ is a
well-defined map of $G$ to $\cI_{\cA}/\cK(\Gamma)$. Since both $\psi_O$
and $\rho$ are group homomorphisms, so is $\phi$. By construction,
$\phi(g_z)=\rho^{-1}(\psi(z))=z\cK(\Gamma)$ for all $z\in\cA$.
Note that the homomorphism $\phi$ is always onto. In the case when the
restriction of the action $\al$ to the orbit $O$ is faithful, the
homomorphism $\psi_O$ is one-to-one. Then $\phi$ is also one-to-one, which
implies that it is an isomorphism.
\end{proof}
\begin{lemma}\label{kernel-cover}
Suppose $\Gamma_1,\Gamma_2\in\Sch(\cI_{\cA},\cA)$ are graphs such that
$\Gamma_1$ covers $\Gamma_2$. Then $\cK(\Gamma_1)\subset\cK(\Gamma_2)$.
\end{lemma}
\begin{proof}
Let $V_1$ be the vertex set of $\Gamma_1$ and $V_2$ be the vertex set of
$\Gamma_2$. There exist transitive actions
$\al_1:\cI_{\cA}\curvearrowright V_1$ and $\al_2:\cI_{\cA}\curvearrowright
V_2$ such that $\Gamma_1=\Gamma_{\Sch}(\cI_{\cA},\cA;\al_1,V_1)$ and
$\Gamma_2=\Gamma_{\Sch}(\cI_{\cA},\cA;\al_2,V_2)$. The actions give rise
to group homomorphisms $\psi_1:\cI_{\cA}\to\mathrm{Sym}(V_1)$ and
$\psi_2:\cI_{\cA}\to\mathrm{Sym}(V_2)$. Then $\cK(\Gamma_1)$ is the kernel
of $\psi_1$ and $\cK(\Gamma_2)$ is the kernel of $\psi_2$. Suppose
$f:V_1\to V_2$ is a covering of the graph $\Gamma_2$ by the graph
$\Gamma_1$. Then the equality $\bigl(\psi_1(z)\bigr)(v)=u$ implies
$\bigl(\psi_2(z)\bigr)(f(v))=f(u)$ for all $u,v\in V_1$ and $z\in\cA$.
Equivalently, $\psi_2(z)f=f\psi_1(z)$ for all $z\in\cA$. It is easy to
verify that the set of all elements $g\in\cI_{\cA}$ satisfying the relation
$\psi_2(g)f=f\psi_1(g)$ is a subgroup. Since every generator $z\in\cA$ of
the group $\cI_{\cA}$ belongs to this set, we conclude that
$\psi_2(g)f=f\psi_1(g)$ for all $g\in\cI_{\cA}$.
If $g\in\cK(\Gamma_1)$ then $\psi_1(g)$ is the identity map on $V_1$.
Hence $\psi_2(g)f=f\psi_1(g)=f$. Since the map $f$ is onto, it follows
that $\psi_2(g)$ is the identity map on $V_2$. Equivalently,
$g\in\cK(\Gamma_2)$. Thus $\cK(\Gamma_1)\subset\cK(\Gamma_2)$.
\end{proof}
\begin{figure}[t]
\centerline{\includegraphics[scale=1]{fig-line.eps}}
\caption{Graph $\Gamma_{\{a,c,d\}}
\bigl(\protect\overleftarrow{\xiGE}d.\xiGE\bigr)$.}
\label{fig-line}
\end{figure}
\begin{lemma}\label{Sch-line-halfline}
Let $\xi$ be an infinite string without double letters over an alphabet
$\cA$, $x$ be a letter in $\cA$ different from the first letter of $\xi$,
and $\overleftarrow{\xi}$ be the left-infinite string obtained by writing
$\xi$ backwards. Then
$\cK\bigl(\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)\bigr)
\subset\cK\bigl(\Gamma_{\cA}(\xi)\bigr)$. Moreover, if
$\al:\cI_{\cA}\curvearrowright\bZ$ is the transitive action such that
$\Gamma_{\Sch}(\cI_{\cA},\cA;\al,\bZ)=
\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)$, then every element of
$\cK\bigl(\Gamma_{\cA}(\xi)\bigr)$ moves at most finitely many points
within that action.
\end{lemma}
\begin{proof}
Since the string $\xi$ has no double letters and $x$ is different from the
first letter of $\xi$, the bi-infinite string $\overleftarrow{\xi}\!x.\xi$
also has no double letters. Hence the graph
$\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)$ is defined. The
graphs $\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)$ and
$\Gamma_{\cA}(\xi)$ both belong to $\Sch(\cI_{\cA},\cA)$. Therefore there
exist unique transitive actions $\al:\cI_{\cA}\curvearrowright\bZ$ and
$\beta:\cI_{\cA}\curvearrowright V$, where $V=\{0,1,2\dots\}$, such that
$\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)=
\Gamma_{\Sch}(\cI_{\cA},\cA;\al,\bZ)$ and
$\Gamma_{\cA}(\xi)=\Gamma_{\Sch}(\cI_{\cA},\cA;\beta,V)$. The actions give
rise to group homomorphisms $\psi_\al:\cI_{\cA}\to\mathrm{Sym}(\bZ)$ and
$\psi_\beta:\cI_{\cA}\to\mathrm{Sym}(V)$. Then
$\cK\bigl(\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)\bigr)$ is the
kernel of $\psi_\al$ and $\cK\bigl(\Gamma_{\cA}(\xi)\bigr)$ is the kernel
of $\psi_\beta$.
Consider a map $f:\bZ\to V$ given by $f(i)=i$ if $i\ge0$ and $f(i)=-1-i$ if
$i<0$. It is easy to observe that $f$ is a covering of the graph
$\Gamma_{\cA}(\xi)$ by the graph
$\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)$. Then Lemma
\ref{kernel-cover} implies that
$\cK\bigl(\Gamma_{\cA}\bigl(\overleftarrow{\xi}\!x.\xi\bigr)\bigr)
\subset\cK\bigl(\Gamma_{\cA}(\xi)\bigr)$. Moreover, just as in the proof
of Lemma \ref{kernel-cover}, it follows that $\psi_\beta(g)f=f\psi_\al(g)$
for all $g\in\cI_{\cA}$.
Suppose $g\in\cK\bigl(\Gamma_{\cA}(\xi)\bigr)$. Then $f(gu)=f(u)$ for all
$u\in\bZ$, where $gu=\bigl(\psi_\al(g)\bigr)(u)$. Any $v\in V$ has exactly
two pre-images under the map $f$: $v$ and $-1-v$. Therefore within the
action $\al$, the element $g$ either exchanges $v$ and $-1-v$, or else
fixes them both. Recall that the distance between vertices $v$ and $gv$ in
the Schreier graph $\Gamma_{\Sch}(\cI_{\cA},\cA;\al,\bZ)$ cannot exceed the
length $|g|$ of $g$ relative to the generating set $\cA$. The distance
from $v$ to $-1-v$ is clearly $2v+1$. Hence $g$ fixes $v$ and $-1-v$
whenever $2v+1>|g|$. This leaves only finitely many vertices in
$\Gamma_{\Sch}(\cI_{\cA},\cA;\al,\bZ)$ that the element $g$ can possibly
move within the action $\al$.
\end{proof}
\begin{lemma}\label{GE-line-halfline}
$\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}c.\xiGE\bigr)\bigr)
=\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}d.\xiGE\bigr)\bigr)
=\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)$.
\end{lemma}
\begin{proof}
Let $x$ be either of the letters $c$ and $d$. The graph
$\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)$ is also the
Schreier graph $\Gamma_{\Sch}(\cI_{\{a,c,d\}},\{a,c,d\};\al,\bZ)$ for some
transitive action $\al:\cI_{\{a,c,d\}}\curvearrowright\bZ$. Then
$\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)\bigr)$
is the kernel of the group homomorphism $\psi_\al:\cI_{\{a,c,d\}}\to
\mathrm{Sym}(\bZ)$ associated to the action $\al$. By Lemma
\ref{Sch-line-halfline},
$\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)\bigr)
\subset\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)$ and, moreover,
every element of $\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)$ moves at most
finitely many points within the action $\al$. To prove that
$\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)\bigr)=
\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)$, it is enough to show that every
element of the group $\cI_{\{a,c,d\}}$ that moves at most finitely many
points within the action $\al$ is, in fact, acting trivially.
By Lemma \ref{xiGE-examples}, the bi-infinite string
$\overleftarrow{\xiGE}x.\xiGE$ belongs to the set $\Om(\xiGE)$. Let $G$ be
the subgroup of the topological full group of the subshift
$\si|\Omega(\xiGE)$ generated by the generalized $2$-cycles $\de_a$,
$\de_c$ and $\de_d$. Let $O$ be the orbit of
$\overleftarrow{\xiGE}x.\xiGE$ under the action of $G$ and $\Gamma$ be the
Schreier graph of that orbit relative to the generating set
$\{\de_a,\de_c,\de_d\}$ (with edge labels in $\{a,c,d\}$). Lemma
\ref{Schr-subshift} implies that $O$ coincides with the orbit of
$\overleftarrow{\xiGE}x.\xiGE$ under the action of the cyclic group
generated by the subshift $\si|\Omega(\xiGE)$. Since the subshift is a
Cantor minimal system (due to Proposition \ref{3-subshifts-Cantor}), it
follows that the orbit $O$ is dense in $\Om(\xiGE)$. As $\Om(\xiGE)$ is a
Cantor set, the orbit $O$ is infinite, which implies that the string
$\overleftarrow{\xiGE}x.\xiGE$ is not periodic. Then it follows from Lemma
\ref{Schr-subshift} that the Schreier graph $\Gamma$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)$. Let
$f:O\to\bZ$ be an isomorphism of $\Gamma$ onto
$\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}x.\xiGE\bigr)$. Then the
equality $\de_z(v)=u$ is equivalent to $\bigl(\psi_\al(z)\bigr)(f(v))=f(u)$
for all $u,v\in O$ and $z\in\{a,c,d\}$. Consequently, everywhere on $O$ we
have $\psi_\al(z)f=f\de_z$ for all $z\in\{a,c,d\}$. Since the generators
of the group $G$ are involutions, there exists a group homomorphism
$\pi:\cI_{\{a,c,d\}}\to G$ such that $\pi(z)=\de_z$ for all
$z\in\{a,c,d\}$. It is easy to verify that the set of all elements
$g\in\cI_{\cA}$ satisfying the relation $\psi_\al(g)f=f\pi(g)$ everywhere
on $O$ is a subgroup. Since every generator $z\in\cA$ of the group
$\cI_{\cA}$ belongs to this set, we conclude that $\psi_\al(g)f=f\pi(g)$ on
$O$ for all $g\in\cI_{\cA}$.
Suppose $g\in\cI_{\{a,c,d\}}$ is an element that moves at most finitely
many points within the action $\al$. Since $\psi_\al(g)f=f\pi(g)$ on $O$
and the map $f$ is bijective, it follows that $\pi(g)$ moves a point $v\in
O$ if and only if $g$ moves $f(v)$ within the action $\al$. Hence $\pi(g)$
moves at most finitely many points of the orbit $O$. As $\Om(\xiGE)$ is a
Cantor set, any dense subset of $\Om(\xiGE)$ remains dense after removing
finitely many points from it. Therefore the set of all points of the orbit
$O$ fixed by $\pi(g)$ is dense in $\Om(\xiGE)$. Since $\pi(g)$ is a
homeomorphism of $\Om(\xiGE)$, it has to coincide with the identity map.
Hence $g$ acts trivially on $\bZ$ within the action $\al$.
\end{proof}
Now we are ready to prove Proposition \ref{inj-homomorph}.
\begin{proof}[Proof of Proposition \ref{inj-homomorph}]
To distinguish the generators of the group $\cG_{(01)^\infty}$ from the
generators of the group $\cI_{\{a,c,d\}}$, in this proof we use notation
$\tilde a,\tilde c,\tilde d$ for the former ones.
Let $O_1$ denote the orbit of the infinite string $1^\infty$ under the
natural action of the group $\cG_{(01)^\infty}$ on $\{0,1\}^{\bN}$. Let
$\Gamma_1$ be the Schreier graph of that orbit. By Lemma
\ref{factor-by-kernel}, there exists a group homomorphism
$\phi_1:\cG_{(01)^\infty}\to\cI_{\{a,c,d\}}/\cK(\Gamma_1)$ such that
$\phi_1(\tilde a)=a\cK(\Gamma_1)$, $\phi_1(\tilde c)=c\cK(\Gamma_1)$ and
$\phi_1(\tilde d)=d\cK(\Gamma_1)$. The orbit $O_1$ is dense in
$\{0,1\}^{\bN}$ due to Lemma \ref{density-of-orbits}. Since the group
$\cG_{(01)^\infty}$ acts on $\{0,1\}^{\bN}$ by homeomorphisms, it follows
that each element of $\cG_{(01)^\infty}$ is uniquely determined by its
restriction to $O_1$. In other words, the restriction of the natural
action of $\cG_{(01)^\infty}$ to the orbit $O_1$ is faithful. Then Lemma
\ref{factor-by-kernel} implies that $\phi_1$ is an isomorphism.
Let $G$ be the subgroup of the topological full group
$[[\si|\Omega(\xiGE)]]$ generated by the generalized $2$-cycles $\de_a$,
$\de_c$ and $\de_d$. By Lemma \ref{xiGE-examples}, the bi-infinite string
$\overleftarrow{\xiGE}d.\xiGE$ belongs to $\Om(\xiGE)$. Let $O_2$ denote
the orbit of this string under the natural action of $G$ on $\Om(\xiGE)$.
Let $\Gamma_2$ be the Schreier graph of the orbit $O_2$. By Lemma
\ref{factor-by-kernel}, there exists a group homomorphism
$\phi_2:G\to\cI_{\{a,c,d\}}/\cK(\Gamma_2)$ such that
$\phi_2(\de_a)=a\cK(\Gamma_2)$, $\phi_2(\de_c)=c\cK(\Gamma_2)$ and
$\phi_2(\de_d)=d\cK(\Gamma_2)$. It was shown in the proof of Lemma
\ref{GE-line-halfline} that the orbit $O_2$ is dense in $\Om(\xiGE)$.
Therefore each element of the group $G$ is uniquely determined by its
restriction to $O_2$ so that the restriction of the natural action of $G$
to the orbit $O_2$ is faithful. Then Lemma \ref{factor-by-kernel} implies
that $\phi_2$ is an isomorphism.
By Lemma \ref{Schr-half-line}, the Schreier graph $\Gamma_1$ is isomorphic
to $\Gamma_{\{a,c,d\}}(\xiGE)$. It was shown in the proof of Lemma
\ref{GE-line-halfline} that the Schreier graph $\Gamma_2$ is isomorphic to
$\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}d.\xiGE\bigr)$. Since
isomorphic graphs cover each other, it follows from Lemma
\ref{kernel-cover} that $\cK(\Gamma_1)=
\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)$ and $\cK(\Gamma_2)=
\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}d.\xiGE\bigr)\bigr)$.
Besides, $\cK\bigl(\Gamma_{\{a,c,d\}}(\xiGE)\bigr)=
\cK\bigl(\Gamma_{\{a,c,d\}}\bigl(\overleftarrow{\xiGE}d.\xiGE\bigr)\bigr)$
due to Lemma \ref{GE-line-halfline}. Therefore
$\cK(\Gamma_1)=\cK(\Gamma_2)$ so that the factor groups
$\cI_{\{a,c,d\}}/\cK(\Gamma_1)$ and $\cI_{\{a,c,d\}}/\cK(\Gamma_2)$ are the
same. Then the composition $h=\phi_2^{-1}\phi_1$ is a well-defined map of
the group $\cG_{(01)^\infty}$ to $G$. Since both $\phi_1$ and $\phi_2$ are
group isomorphisms, so is $h$. As $G$ is a subgroup of
$[[\si|\Omega(\xiGE)]]$, we may regard $h$ as a one-to-one homomorphism of
$\cG_{(01)^\infty}$ to $[[\si|\Omega(\xiGE)]]$. By construction, $h(\tilde
a)=\de_a$, $h(\tilde c)=\de_c$ and $h(\tilde d)=\de_d$.
\end{proof}
Now we are going to derive Theorems \ref{sub-main} and \ref{main} from
Proposition \ref{inj-homomorph}.
\begin{proof}[Proof of Theorem \ref{sub-main}]
It follows from Theorem \ref{GE-growth} that the group $\cG_{(01)^\infty}$
has intermediate growth. The group $\cG_{(01)^\infty}$ is generated by
three involutions $a$, $c$ and $d$. Proposition \ref{inj-homomorph}
implies that $\cG_{(01)^\infty}$ is isomorphic to the subgroup of the
topological full group $[[\si|\Omega(\xiGE)]]$ generated by involutions
$\de_a$, $\de_c$ and $\de_d$. Since isomorphic groups have the same
growth, the theorem follows.
\end{proof}
\begin{proof}[Proof of Theorem \ref{main}]
By Proposition \ref{3-subshifts-factors}, the subshift
$\si_{\{x,y\}}|\Om(\xipd)$ is a continuous factor of the subshift
$\si_{\{0,1\}}|\Om(\xiTM)$ while the subshift $\si_{\{a,c,d\}}|\Om(\xiGE)$
is topologically conjugate to $\si_{\{x,y\}}|\Om(\xipd)$. By Proposition
\ref{3-subshifts-Cantor}, all three subshifts are Cantor minimal systems.
In particular, all three are aperiodic. Then it follows from Theorem
\ref{TFG-embeds} that the topological full group
$[[\si_{\{x,y\}}|\Om(\xipd)]]$ embeds into the topological full group
$[[\si_{\{0,1\}}|\Om(\xiTM)]]$ while the
topological full group $[[\si_{\{a,c,d\}}|\Om(\xiGE)]]$ embeds into
$[[\si_{\{x,y\}}|\Om(\xipd)]]$. As a consequence, any subgroup of
$[[\si_{\{a,c,d\}}|\Om(\xiGE)]]$ is isomorphic to a subgroup of
$[[\si_{\{0,1\}}|\Om(\xiTM)]]$ and a subgroup of
$[[\si_{\{x,y\}}|\Om(\xipd)]]$. Proposition \ref{inj-homomorph} implies
that the group $\cG_{(01)^\infty}$ embeds into the topological full group
$[[\si|\Omega(\xiGE)]]$. It follows that $\cG_{(01)^\infty}$ also embeds
into $[[\si_{\{0,1\}}|\Om(\xiTM)]]$ and $[[\si_{\{x,y\}}|\Om(\xipd)]]$, the
topological full groups of the substitution subshifts generated by the
Thue-Morse substitution and the period doubling substitution.
\end{proof}
The group $\cG_{(01)^\infty}$ can be embedded into the topological full
group $[[\si_{\{0,1\}}|\Om(\xiTM)]]$ of the Thue-Morse subshift in three
steps. First $\cG_{(01)^\infty}$ is embedded into the group
$[[\si_{\{a,c,d\}}|\Om(\xiGE)]]$, then $[[\si_{\{a,c,d\}}|\Om(\xiGE)]]$ is
embedded into (in fact, mapped isomorphically onto) the group
$[[\si_{\{x,y\}}|\Om(\xipd)]]$, and finally $[[\si_{\{x,y\}}|\Om(\xipd)]]$
is embedded into $[[\si_{\{0,1\}}|\Om(\xiTM)]]$. The first step is
described explicitly by Proposition \ref{inj-homomorph}. The other two
steps can be described explicitly as well. Each time the generators $a$,
$c$ and $d$ of the group $\cG_{(01)^\infty}$ are mapped to generalized
$2$-cycles of the form $\de_{U;S}$, where $S$ is an appropriate subshift.
To describe the clopen sets $U$, we need another notation. Given a
nonempty word $w$ over an alphabet $\cA$, let $[.w]_{\cA}$ denote the set
of all bi-infinite strings $\ldots x_{-1}x_0.x_1x_2\ldots$ in $\cA^{\bZ}$
such that $x_1x_2\ldots x_n=w$, where $n$ is the length of $w$.
\begin{proposition}\label{two-more-inj}
(i) There exists a one-to-one group homomorphism $h_1:\cG_{(01)^\infty}\to
[[\si_{\{x,y\}}|\Om(\xipd)]]$ such that
$h_1(z)=\de_{U_z\cap\Om(\xipd);S_1}$ for all $z\in\{a,c,d\}$, where
$S_1=\si_{\{x,y\}}|\Om(\xipd)$, $U_a=[.xy]_{\{x,y\}}\cup [.xxxy]_{\{x,y\}}$,
$U_c=[.y]_{\{x,y\}}$ and $U_d=[.xxy]_{\{x,y\}}$.
(ii) There exists a one-to-one group homomorphism $h_2:\cG_{(01)^\infty}\to
[[\si_{\{0,1\}}|\Om(\xiTM)]]$ such that
$h_2(z)=\de_{W_z\cap\Om(\xiTM);S_2}$ for all $z\in\{a,c,d\}$, where
$S_2=\si_{\{0,1\}}|\Om(\xiTM)$,
\begin{eqnarray*}
W_a &=& [.011]_{\{0,1\}}\cup [.100]_{\{0,1\}}\cup [.01011]_{\{0,1\}}\cup
[.10100]_{\{0,1\}},\\
W_c &=& [.00]_{\{0,1\}}\cup [.11]_{\{0,1\}},\\
W_d &=& [.0100]_{\{0,1\}}\cup [.1011]_{\{0,1\}}.
\end{eqnarray*}
\end{proposition}
\begin{proof}
Suppose $T_1:X_1\to X_1$ and $T_2:X_2\to X_2$ are aperiodic homeomorphisms
of Cantor sets, and $f:X_1\to X_2$ is a continuous map that is onto and
satisfies $fT_1=T_2f$. By Theorem \ref{TFG-embeds}, there exists a
one-to-one homomorphism $\Psi:[[T_2]]\to[[T_1]]$. An explicit description
of $\Psi$ can be extracted from Lemmas \ref{model-isomorphism} and
\ref{group-embed}. Namely, if $F\in[[T_2]]$ is a homeomorphism given by
$F(x)=T_2^{\nu(x)}(x)$, $x\in X_2$, where $\nu:X_2\to\bZ$ is a continuous
function, then its image $\Psi(F)$ is given by
$\bigl(\Psi(F)\bigr)(x)=T_1^{\nu(f(x))}(x)$ for all $x\in X_1$. In the
case when $F=\de_{U;T_2}$ for some clopen set $U\subset X_2$, we obtain
that $\Psi(F)=\de_{f^{-1}(U);T_1}$.
By Proposition \ref{inj-homomorph}, there is a one-to-one homomorphism
$h:\cG_{(01)^\infty}\to[[\si_{\{a,c,d\}}|\Om(\xiGE)]]$ such that
$h(z)=\de_z$ for all $z\in\{a,c,d\}$. Recall that $\de_z$ is the short
notation for the generalized $2$-cycle
$\de_{[.z]_{\{a,c,d\}}\cap\Om(\xiGE);S}$, where
$S=\si_{\{a,c,d\}}|\Om(\xiGE)$. Let
$f_{\phi_2}:\{a,c,d\}^{\bZ}\to\{x,y\}^{\bZ}$ be the block factor map
considered in Lemma \ref{block-factor-2}. In view of the lemma, the
restriction of $f_{\phi_2}$ to the set $\Om(\xiGE)$ can be regarded a
homeomorphism onto $\Om(\xipd)$. Let $g_1$ be the inverse of that
homeomorphism. Then $g_1S_1=Sg_1$, where $S_1=\si_{\{x,y\}}|\Om(\xipd)$.
Hence $g_1$ induces a one-to-one homomorphism $\Psi_1:[[S]]\to[[S_1]]$ as
described above. Then $h_1=\Psi_1h$ is a one-to-one homomorphism of the
group $\cG_{(01)^\infty}$ to $[[S_1]]$. For any $z\in\{a,c,d\}$ we obtain
that $h_1(z)=\Psi_1(\de_z)=\de_{U'_z;S_1}$, where $U'_z=
g_1^{-1}\bigl([.z]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)
=f_{\phi_2}\bigl([.z]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)$.
It follows from the description of the string $\xiGE$ (Lemma \ref{xi-GE})
that any occurrence of the letter $d$ in $\xiGE$ is followed by $ac$.
Hence $[.d]_{\{a,c,d\}}\cap\Om(\xiGE)=[.dac]_{\{a,c,d\}}\cap\Om(\xiGE)$.
Any occurrence of the letter $a$ is followed by either $c$ or $dac$. Hence
the set $[.a]_{\{a,c,d\}}\cap\Om(\xiGE)$ is the union of
$[.ac]_{\{a,c,d\}}\cap\Om(\xiGE)$ and $[.adac]_{\{a,c,d\}}\cap\Om(\xiGE)$.
Recall that the $1$-block factor map $f_{\phi_2}$ is induced by the
function $\phi_2:\{a,c,d\}\to\{x,y\}$ given by $\phi_2(a)=\phi_2(d)=x$ and
$\phi_2(c)=y$. Therefore $f_{\phi_2}\bigl([.c]_{\{a,c,d\}}\bigr)\subset
[.y]_{\{x,y\}}$, $f_{\phi_2}\bigl([.ac]_{\{a,c,d\}}\bigr)\subset
[.xy]_{\{x,y\}}$, $f_{\phi_2}\bigl([.dac]_{\{a,c,d\}}\bigr)\subset
[.xxy]_{\{x,y\}}$ and $f_{\phi_2}\bigl([.adac]_{\{a,c,d\}}\bigr)\subset
[.xxxy]_{\{x,y\}}$. Note that $[.c]_{\{a,c,d\}}$, $[.ac]_{\{a,c,d\}}$,
$[.dac]_{\{a,c,d\}}$ and $[.adac]_{\{a,c,d\}}$ are disjoint sets that cover
$\Om(\xiGE)$. As observed in the proof of Lemma \ref{block-factor-2}, the
word $xxxx$ is not a subword of the string $\xipd$. Hence the set
$[.xxxx]_{\{x,y\}}$ is disjoint from $\Om(\xipd)$. Consequently,
$\Om(\xipd)$ is covered by disjoint sets $[.y]_{\{x,y\}}$,
$[.xy]_{\{x,y\}}$, $[.xxy]_{\{x,y\}}$ and $[.xxxy]_{\{x,y\}}$. Since the
restriction of $f_{\phi_2}$ to $\Om(\xiGE)$ is a one-to-one map onto
$\Om(\xipd)$, it follows that
$f_{\phi_2}\bigl([.c]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)=
[.y]_{\{x,y\}}\cap\Om(\xipd)$,
$f_{\phi_2}\bigl([.ac]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)=
[.xy]_{\{x,y\}}\cap\Om(\xipd)$,
$f_{\phi_2}\bigl([.dac]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)=
[.xxy]_{\{x,y\}}\cap\Om(\xipd)$ and
$f_{\phi_2}\bigl([.adac]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)=
[.xxxy]_{\{x,y\}}\cap\Om(\xipd)$. As a consequence, for any
$z\in\{a,c,d\}$ we have
$f_{\phi_2}\bigl([.z]_{\{a,c,d\}}\cap\Om(\xiGE)\bigr)=U_z\cap\Om(\xipd)$,
where $U_z$ is as defined in the proposition. This completes the proof of
the statement (i).
We proceed to the statement (ii). Let
$f_{\phi_1}:\{0,1\}^{\bZ}\to\{x,y\}^{\bZ}$ be the block factor map
considered in Lemma \ref{block-factor-1}. The map $f_{\phi_1}$ is
two-to-one when restricted to $\Om(\xiTM)$ and $f_{\phi_1}(\Om(\xiTM))=
\Om(\xipd)$. It follows from the proof of Lemma \ref{block-factor-1} that
$f_{\phi_1}$ is two-to-one on $\{0,1\}^{\bZ}$. As a consequence,
$f_{\phi_1}^{-1}(\Om(\xipd))=\Om(\xiTM)$. Let $g_2$ be the restriction of
$f_{\phi_1}$ to $\Om(\xiTM)$. Then $g_2S_2=S_1g_2$, where
$S_2=\si_{\{0,1\}}|\Om(\xiTM)$. Hence $g_2$ induces a one-to-one
homomorphism $\Psi_2:[[S_1]]\to[[S_2]]$ as described above. Then
$h_2=\Psi_2h_1$ is a one-to-one homomorphism of the group
$\cG_{(01)^\infty}$ to $[[S_2]]$. For any $z\in\{a,c,d\}$ we have
$h_2(z)=\Psi_2\bigl(\de_{U_z\cap\Om(\xipd);S_1}\bigr)
=\de_{g_2^{-1}(U_z\cap\Om(\xipd));S_2}$. Note that
$g_2^{-1}(U_z\cap\Om(\xipd))=f_{\phi_1}^{-1}(U_z)\cap\Om(\xiTM)$ since
$f_{\phi_1}^{-1}(\Om(\xipd))=\Om(\xiTM)$. Recall that the $2$-block factor
map $f_{\phi_1}$ is induced by the function $\phi_1:\{0,1\}^2\to\{x,y\}$
given by $\phi_1(0,1)=\phi_1(1,0)=x$ and
$\phi_1(0,0)=\phi_1(1,1)=y$. We obtain that
\begin{align*}
&f_{\phi_1}^{-1}([.y]_{\{x,y\}}) = [.00]_{\{0,1\}}\cup [.11]_{\{0,1\}},\\
&f_{\phi_1}^{-1}([.xy]_{\{x,y\}}) = [.011]_{\{0,1\}}\cup
[.100]_{\{0,1\}},\\
&f_{\phi_1}^{-1}([.xxy]_{\{x,y\}}) = [.0100]_{\{0,1\}}\cup
[.1011]_{\{0,1\}},\\
&f_{\phi_1}^{-1}([.xxxy]_{\{x,y\}}) = [.01011]_{\{0,1\}}\cup
[.10100]_{\{0,1\}}.
\end{align*}
It follows that for any $z\in\{a,c,d\}$ we have $f_{\phi_1}^{-1}(U_z)=W_z$,
where $W_z$ is as defined in the proposition. This completes the proof of
the statement (ii).
\end{proof}
%
\begin{thebibliography}{BGN}
\bibitem[ADQ]{AlloucheDQ:hidden}
J.-P. Allouche, M. Dekking and M. Queff{\'e}lec, Hidden automatic
sequences. \emph{Comb. Theory} \textbf{1} (2021), Article ID 20, 15 pp.
\bibitem[AS1]{AlloucheShallit:ubiquitous}
J.-P. Allouche and J. Shallit, The ubiquitous Prouhet-Thue-Morse sequence.
In \emph{Sequences and their applications (Singapore 1998)}, 1--16.
Springer Ser. Discrete Math. Theor. Comput. Sci., Springer, London, 1999.
\bibitem[AS2]{AlloucheShallit:automatic}
J.-P. Allouche and J. Shallit, \emph{Automatic sequences. Theory,
applications, generalizations}. Cambridge Univ. Press, Cambridge, 2003.
\bibitem[BCT]{Barth:Thue-Morse}
L. Bartholdi, J. M. R. Caballero and A. Tanbir, The Thue-Morse
substitutions and self-similar groups and algebras. \emph{Algebra Discrete
Math.} \textbf{34} (2022), no. 1, 1--14.
\bibitem[BG]{BarGri00}
L. Bartholdi and R. I. Grigorchuk, On the spectrum of Hecke type operators
related to some fractal groups. \emph{Proc. Steklov Inst. Math.}
\textbf{231} (2000), 1--41 [translated from \emph{Tr. Mat. Inst. Steklova}
\textbf{231} (2000), 5--45].
\bibitem[BGN]{BarGrigNek}
L. Bartholdi, R. Grigorchuk and V. Nekrashevych, From fractal groups to
fractal sets. In \emph{Fractals in Graz 2001}, 25--118. Trends Math.
Birkh\"{a}user, Basel, 2003.
\bibitem[Be]{Benli}
M. G. Benli, Profinite completion of Grigorchuk's group is not finitely
presented. \emph{Int. J. Algebra Comput.} \textbf{22} (2012), no. 5,
Article ID 1250045, 26 pp.
\bibitem[BP]{BoffaPoint}
M. Boffa and F. Point, Identit\'{e}s de Thue-Morse dans les groupes.
\emph{C. R. Acad. Sci. Paris S\'{e}r. I Math.} \textbf{312} (1991), no. 9,
667--670.
\bibitem[BG+]{Classification}
I. Bondarenko, R. Grigorchuk, R. Kravchenko, Y. Muntyan, V. Nekrashevych,
D. Savchuk and Z. \v{S}uni{\'c}, On classification of groups generated by
3-state automata over a 2-letter alphabet. \emph{Algebra Discrete Math.}
(2008), no. 1, 1--163.
\bibitem[Er]{Erschler04}
A. Erschler, Boundary behavior for groups of subexponential growth.
\emph{Ann. Math. (2)} \textbf{160} (2004), no. 3, 1183--1210.
\bibitem[Fo]{Fogg:substitutions}
N. P. Fogg, \emph{Substitutions in dynamics, arithmetics and
combinatorics}. Lect. Notes in Math. \textbf{1794}. Springer-Verlag,
Berlin, 2002.
\bibitem[GPS]{GPS}
T. Giordano, I. F. Putnam and C. F. Skau, Full groups of Cantor minimal
systems. \emph{Israel J. Math.} \textbf{111} (1999), no. 1, 285--320.
\bibitem[Gr1]{Grig84}
R. I. Grigorchuk, Degrees of growth of finitely generated groups, and the
theory of invariant means. \emph{Math. USSR, Izv.} \textbf{25} (1985),
259--300 [translated from \emph{Izv. Akad. Nauk SSSR, Ser. Mat.}
\textbf{48} (1984), no. 5, 939--985].
\bibitem[Gr2]{Grig99}
R. I. Grigorchuk, On the system of defining relations and the Schur
multiplier of periodic groups generated by finite automata. In \emph{Groups
St. Andrews 1997 in Bath, Vol. 1}, 290--317. Lond. Math. Soc. Lect. Note
Ser. \textbf{260}. Cambridge Univ. Press, Cambridge, 1999.
\bibitem[GLN]{GLN17}
R. Grigorchuk, D. Lenz and T. Nagnibeda, Spectra of Schreier graphs of
Grigorchuk’s group and Schroedinger operators with aperiodic order.
\emph{Math. Ann.} \textbf{370} (2018), no. 3--4, 1607--1637.
\bibitem[GM1]{GrigMed}
R. I. Grigorchuk and K. S. Medynets, On algebraic properties of topological
full groups. \emph{Sb. Math.} \textbf{205} (2014), no. 6, 843--861
[translated from \emph{Mat. Sb.} \textbf{205} (2014), no. 6, 87--108].
\bibitem[GM2]{GrigMed2}
R. Grigorchuk and K. Medynets, Presentations of topological full groups by
generators and relations. \emph{J. of Algebra} \textbf{500} (2018), 46--68.
\bibitem[GSS]{GrigSavSun}
R. Grigorchuk, D. Savchuk and Z. \v{S}uni\'{c}, The spectral problem,
substitutions and iterated monodromy. In \emph{Probability and mathematical
physics}, 225--248. CRM Proc. Lecture Notes \textbf{42}. Amer. Math. Soc.,
Providence, RI, 2007.
\bibitem[GV]{GrigVor24}
R. Grigorchuk and Y. Vorobets, On maximal subgroups of ample groups.
Preprint, 2024 (available at \url{https://arxiv.org/abs/2403.16364}).
\bibitem[JM]{JuschMonod}
K. Juschenko and N. Monod, Cantor systems, piecewise translations and
simple amenable groups. \emph{Ann. of Math. (2)} \textbf{178} (2013), no.
2, 775--787.
\bibitem[LM]{LindMarcus}
D. Lind and B. Marcus, \emph{An introduction to symbolic dynamics and
coding}, 2nd updated ed. Cambridge Math. Library. Cambridge University
Press, Cambridge, 2021.
\bibitem[LS]{LindSchmidt}
D. Lind and K. Schmidt, Symbolic and algebraic dynamical systems. In
\emph{Handbook of dynamical systems, Vol. 1A}, 765--812. North-Holland,
Amsterdam, 2002.
\bibitem[Ly]{Lys85}
I. G. Lysenok, A system of defining relations for a Grigorchuk group.
\emph{Math. Notes} \textbf{38} (1985), 784--792.
\bibitem[M-B]{MatteBon}
N. Matte Bon, Topological full groups of minimal subshifts with subgroups
of intermediate growth. \emph{J. Modern Dynamics} \textbf{9} (2015), 67--80.
\bibitem[Mal]{Maltsev}
A. I. Mal'tsev, Nilpotent semigroups. \emph{Ivanov. Gos. Ped. Inst. U\v{c}.
Zap. Fiz.-Mat. Nauki} \textbf{4} (1953), 107--111.
\bibitem[Ma1]{Matui}
H. Matui, Some remarks on topological full groups of Cantor minimal
systems. \emph{Internat. J. Math.} \textbf{17} (2006), no. 2, 231--251.
\bibitem[Ma2]{Matui2}
H. Matui, Some remarks on topological full groups of Cantor minimal
systems II. \emph{Ergodic Theory Dyn. Syst.} \textbf{33} (2013), no. 5,
1542--1549.
\bibitem[Mu]{Muntyan}
Y. Muntyan, Automata groups. Ph.D. Dissertation, Texas A\&M University, 2009.
\bibitem[Ne1]{Nek16}
V. Nekrashevych, Palindromic subshifts and simple periodic groups of
intermediate growth. \emph{Ann. of Math. (2)} \textbf{187} (2018), no. 3,
667--719.
\bibitem[Ne2]{Nek2022}
V. Nekrashevych, \emph{Groups and topological dynamics}. Grad. Stud. Math.
\textbf{223}. Amer. Math. Soc., Providence, RI, 2022.
\bibitem[Pu]{Putnam89}
I. F. Putnam, The $C^*$-algebras associated with minimal homeomorphisms of
the Cantor set. \emph{Pacific J. Math.} \textbf{136} (1989), no. 2,
329--353.
\bibitem[Ro]{Rowen:Koethe}
L. H. Rowen, Koethe's conjecture. In \emph{Ring theory 1989 (Ramat Gan and
Jerusalem, 1988/1989)}, 193--202. Israel Math. Conf. Proc. \textbf{1}.
Weizmann, Jerusalem, 1989.
\bibitem[Sa]{Sapir}
M. V. Sapir, \emph{Combinatorial algebra: syntax and semantics}. Springer
Monographs in Mathematics. Springer, Cham, 2014.
\bibitem[Sch]{Schmidt}
K. Schmidt, \emph{Algebraic ideas in ergodic theory}. Reg. Conf. Ser.
Math. \textbf{76}. Amer. Math. Soc., Providence, RI, 1990.
\bibitem[Vo1]{Vor10}
Y. Vorobets, On a substitution subshift related to the Grigorchuk group.
\emph{Proc. Steklov Inst. Math.} {\bf 271} (2010), 306--321.
\bibitem[Vo2]{Vor12}
Y. Vorobets, Notes on the Schreier graphs of the Grigorchuk group. In
\emph{Dynamical systems and group actions}, 221--248. Contemp. Math.
\textbf{567}. Amer. Math. Soc., Providence, RI, 2012.
\bibitem[Vo3]{Vor20}
Y. Vorobets, On the topological full group containing the Grigorchuk group.
Preprint, 2020 (available at \url{https://arxiv.org/abs/2007.06789}).
\end{thebibliography}
\bigskip
{\sc
\begin{raggedright}
Department of Mathematics\\
Texas A\&M University\\
College Station, TX 77843--3368
\end{raggedright}
}
\end{document}