Complexity among the finitely generated subgroups of Thompson's group
Extracted source text · arXiv 1711.10998
\documentclass[12pt]{amsart}
\usepackage{amssymb}
\usepackage{amsfonts}
\usepackage{latexsym}
\usepackage{wasysym}
\usepackage{mathrsfs}
\usepackage{accents}
\usepackage{xcolor}
\usepackage{fancyvrb}
\usepackage{relsize}
\usepackage{ifthen}
\usepackage{verbatim}
\usepackage{epsfig}
\usepackage[all]{xy}
\newcommand{\SST}{\scriptstyle}
\numberwithin{equation}{section}
\newtheorem{thm}{Theorem}
\newtheorem{conj}{Conjecture}
\newtheorem{lem}{Lemma}[section]
\newtheorem{prop}[lem]{Proposition}
\newtheorem{claim}[lem]{Claim}
\newtheorem{cor}[lem]{Corollary}
\theoremstyle{remark}
\newtheorem{remark}[lem]{Remark}
\newtheorem*{rem}{Remark}
\newcommand{\Marking}{marking}
\newcommand{\Markings}{markings}
\newcommand{\Marked}{marked}
\newcommand{\Marker}{marker}
\newcommand{\Markers}{markers}
\newcommand{\Mark}{mark}
\newcommand{\gen}[1]{\langle #1 \rangle}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\R}{\mathbb{R}}
\newcommand\supt{\operatorname{supt}}
\renewcommand\exp{\operatorname{exp}}
\newcommand{\infl}[2]{#1^{\underaccent{\,\rightharpoondown}{#2}}}
\newcommand\Homeo{\operatorname{Homeo}}
\newcommand{\HomeoI}{\Homeo_+(I)}
\newcommand\Ascr{\mathscr{A}}
\newcommand\Bscr{\mathscr{B}}
\newcommand\Cscr{\mathscr{C}}
\newcommand\Dscr{\mathscr{D}}
\newcommand\Escr{\mathscr{E}}
\newcommand\Fscr{\mathscr{F}}
\newcommand\Gscr{\mathscr{G}}
\newcommand\Hscr{\mathscr{H}}
\newcommand\Iscr{\mathscr{I}}
\newcommand\Jscr{\mathscr{J}}
\newcommand\Kscr{\mathscr{K}}
\newcommand\Lscr{\mathscr{L}}
\newcommand\Mscr{\mathscr{M}}
\newcommand\Nscr{\mathscr{N}}
\newcommand\Pscr{\mathscr{P}}
\newcommand\Qscr{\mathscr{Q}}
\newcommand\Rscr{\mathscr{R}}
\newcommand\Sscr{\mathscr{S}}
\newcommand\Tscr{\mathscr{T}}
\newcommand\Uscr{\mathscr{U}}
\newcommand\Vscr{\mathscr{V}}
\newcommand\Wscr{\mathscr{W}}
\newcommand\Xscr{\mathscr{X}}
\newcommand\Yscr{\mathscr{Y}}
\newcommand\Zscr{\mathscr{Z}}
\newcommand\Acal{\mathcal{A}}
\newcommand\Bcal{\mathcal{B}}
\newcommand\Ccal{\mathcal{C}}
\newcommand\Dcal{\mathcal{D}}
\newcommand\Ecal{\mathcal{E}}
\newcommand\Fcal{\mathcal{F}}
\newcommand\Gcal{\mathcal{G}}
\newcommand\Hcal{\mathcal{H}}
\newcommand\Ical{\mathcal{I}}
\newcommand\Jcal{\mathcal{J}}
\newcommand\Kcal{\mathcal{K}}
\newcommand\Lcal{\mathcal{L}}
\newcommand\Mcal{\mathcal{M}}
\newcommand\Ncal{\mathcal{N}}
\newcommand\Pcal{\mathcal{P}}
\newcommand\Qcal{\mathcal{Q}}
\newcommand\Rcal{\mathcal{R}}
\newcommand\Scal{\mathcal{S}}
\newcommand\Tcal{\mathcal{T}}
\newcommand\Ucal{\mathcal{U}}
\newcommand\Vcal{\mathcal{V}}
\newcommand\Wcal{\mathcal{W}}
\newcommand\Xcal{\mathcal{X}}
\newcommand\Ycal{\mathcal{Y}}
\newcommand\Zcal{\mathcal{Z}}
\newcommand{\wta}{\widetilde{\alpha}}
\newcommand{\wtb}{\widetilde{\beta}}
\renewcommand\Im{\operatorname{Im}}
\newcommand{\cut}{{}^\circ}
\newcommand{\Ffrak}{\mathfrak{F}}
\newcommand{\Sfrak}{\mathfrak{S}}
\newcommand\emb{\hookrightarrow}
\newcommand\PLoI{\mathrm{PL}_+(I)}
\renewcommand\Z{\mathbf{Z}}
\renewcommand\R{\mathbf{R}}
\newcommand\formula[1]{{}^\ulcorner #1 {}^\urcorner}
\newcommand\EA{\operatorname{EA}}
\newcommand\As{\mathsf{A}}
\newcommand\Bs{\mathsf{B}}
\newcommand\Cs{\mathsf{C}}
\newcommand\Ds{\mathsf{D}}
\newcommand\Es{\mathsf{E}}
\newcommand\Fs{\mathsf{F}}
\newcommand\Rs{\mathsf{R}}
\newcommand\Xs{\mathsf{X}}
\newcommand\Ys{\mathsf{Y}}
\newcommand\Zs{\mathsf{Z}}
\newcommand\zero{\mathbf{0}}
\newcommand\one{\mathsf{Z}}
\newcommand{\underdots}[1]{\lowerddot{#1}}
\newcommand\Bsig{\Bscr}
\newcommand\Rsig{\Rscr}
\newcommand\Ssig{\Sscr}
\newcommand\Sgen{\Scal}
\renewcommand\epsilon{\varepsilon}
\newcommand{\varbump}[3]
{
\xy
(0,0); (#3,0)**\crv{(#1,#1)&(#2,#1)};
\endxy
}
\newcommand{\varnbump}[3]
{
\xy
(0,0); (#3,0)**\crv{(#1,-#1)&(#2,-#1)};
\endxy
}
\newcommand{\vardotbump}[3]
{
\xy
(0,0); (#3,0)**\crv{~*=<3pt>{.}(#1,#1)&(#2,#1)};
\endxy
}
\newcommand{\varndotbump}[3]
{
\xy
(0,0); (#3,0)**\crv{~*=<3pt>{.}(#1,-#1)&(#2,-#1)};
\endxy
}
\newcommand{\Th}{{}^{\textrm{th}}}
\newcommand\mand{\textrm{ and }}
\newcommand\mor{\textrm{ or }}
\title[Subgroups of Thompson's group]
{Complexity among the finitely generated subgroups of Thompson's group}
\keywords{elementary amenable,
elementary group,
geometrically fast,
homeomorphism group,
ordinal,
Peano Arithmetic,
piecewise linear,
Thompson's group,
transition chain}
\subjclass[2010]{
20E22, % Extensions, wreath products, and other compositions
20B07, % General theory of infinite groups
20B10, % Characterization theorems
20E07}
\thanks{
This paper is in part a product of a visit of the first and third author to
the \emph{Mathematisches Forschungsinstitut Oberwolfach}, Germany in December 2016
as part of their \emph{Research In Pairs} program.
The third author was partially supported by
NSF grant DMS--1600635.
}
\author[Bleak]{Collin~Bleak}
\author[Brin]{Matthew~G.~Brin}
\author[Moore]{Justin~Tatch~Moore}
\address{
Collin~Bleak \\
School of Mathematics and Statistics \\
University of St. Andrews \\
St. Andrews, Fife KY16 9SS \\
}
\email{{\tt cb211@st-andrews.ac.uk}}
\address{
Matthew~G.~Brin \\
Department of Mathematical Sciences \\
Binghamton University \\
Binghamton, NY 13902-6000 \\
}
\email{{\tt matt@math.binghamton.edu}}
\address{
Justin~Tatch~Moore \\
Department of Mathematics \\
Cornell University \\
Ithaca, NY 14853-4201 \\
}
\email{{\tt justin@math.cornell.edu}}
\begin{document}
\begin{abstract}
We demonstrate the existence of a family of finitely generated
subgroups of Richard Thompson's group \(F\) which is strictly well-ordered
by the embeddability relation in type \(\epsilon_0 +1\).
All except the maximum element of this family (which is \(F\)
itself) are elementary amenable groups.
In fact we also obtain, for each \(\alpha < \epsilon_0\), a finitely generated
elementary amenable subgroup of \(F\) whose EA-class is \(\alpha + 2\).
These groups all have simple, explicit descriptions and can be
viewed as a natural continuation of the progression which starts
with \(\Z + \Z\), \(\Z \wr \Z\), and the Brin-Navas group \(B\).
We also give an example of a pair of finitely generated elementary
amenable subgroups of \(F\) with the property that neither is
embedable into the other.
\end{abstract}
\maketitle
\section{Introduction}
\label{intro:sec}
The aim of this article is to initiate a program to classify the
finitely generated subgroups of Richard Thompson's group \(F\).
While we are far from completing this program, we isolate a
class \(\Sfrak\) of finitely generated subgroups of \(F\) which
exhibits a high degree of complexity, but which admits
a complete structural analysis
and seems likely to play a central role in the classification of all
finitely generated subgroups of \(F\).
The groups in \(\Sfrak\) all
have generating sets with simple descriptions in the language of
\cite{fast_gen}. Using \cite{fast_gen} one can specify certain
subgroups of \(\HomeoI\), the group of orientation preserving self
homeomorphisms of the unit interval \(I\), by the qualitative
dynamics of their generating sets. All groups given in this way embed
homomorphically into \(F\). The groups in \(\Sfrak\) provide simple
and natural examples of elementary amenable groups of high EA-class
that we feel are of interest independent of being subgroups of
\(F\).
There are two main features of this paper. The first is the shift of
attention away from the usual ``isomorphism type and containment
relation'' (the Hasse diagram) of subgroups, and toward
the coarser ``biembeddability class and embeddability relation'' where
two groups are biembeddable if each embeds in the other.
A finer analysis of the
isomorphism types of subgroups of \(F\) does not seem feasible at
this time.
The second feature is the discovery of a rich arithmetic that lives
on \(\Sfrak\) that greatly facilitates transfinite induction and
recursion. The usual ingredients of transfinite recursion are base,
successor, and limit stage: a base object \(A_0\) must be built, an
object \(A_{\alpha+1}\) must be built from the object \(A_\alpha\),
and for a limit \(\alpha\), an object \(A_\alpha\) must be built
from the objects \(A_\beta\) with \(\beta<\alpha\).
We show that \(\Sfrak\) can be equipped with
arithmetic operations that allow us to
easily build from \(B_\alpha\in \Sfrak\) not only \(B_{\alpha+1}\),
but also
\(B_{\alpha\cdot\omega}\) and even
\(B_{\omega^\alpha}\) with equal ease. This has two consequences.
First, our groups are remarkably easy to ``write down.'' Second,
the bulk of the work in the paper is shifted from construction to
analysis. In fact, it is still a wonder to the authors that these
groups can be analyzed at all.
\subsection{The results}
We now state and discuss our results in somewhat more detail. We
give indication of the meaning of terminology in what follows;
full definitions are given in Sections \ref{objects:sec} and
\ref{prelim:sec} and as noted.
Elementary amenable groups form a class \(EG\) and are those groups
that can be built from finite and abelian groups by a (possibly
transfinite) process using extension and directed union. The
EA-class of a group \(G\) in \(EG\) is a measure of the complexity of the
construction process for \(G\).
Thompson's group \(F\) is not elementary amenable;
it is finitely generated and every nontrivial normal
subgroup of \(F\) contains isomorphic copies of \(F\) (see \cite{CFP}).
The second author and Mark Sapir have made the
following conjecture which was the original motivation for the
present article.
\begin{conj} \label{BrinSapir} \cite{BrinEG} \cite{SapirProblems} If
\(G\) is a subgroup of \(F\), then either \(G\) is elementary
amenable or else \(G\) contains a copy of \(F\).
\end{conj}
\noindent
Our basic thesis is that this conjecture will eventually be a
corollary of a more complete understanding of the partial order
\((\Ffrak, \emb)\) where \(\Ffrak\) is the set of biembeddability classes of
finitely generated subgroups of \(F\) and \(A \emb B\) asserts that members of the class
\(A\) embed into members of the class \(B\). (We will sometimes abuse language and notation and confuse
equivalence classes with their representatives ---e.g. write \(A \emb B\) when \(A\) and \(B\) are groups.)
The complex nature of \((\Ffrak,\emb)\) is demonstrated by our main result:
%
\begin{figure}
\[
\xy
(45,5); (5,5)**@{-}; (5,45)**@{-};
(5,5); (45,45)**@{.}; %(40,8)*{(G_3)};
(5,5); (15,15)**\crv{(11,6)&(14,9)};
(15,15); (25,25)**\crv{(21,16)&(24,19)};
(25,25); (35,35)**\crv{(26,31)&(29,34)};
(35,35); (45,45)**\crv{(36,41)&(39,44)};
(25,19)*{g_4};
(10,10); (20,20)**\crv{~*=<3pt>{.}(16,11)&(19,14)};
(20,20); (30,30)**\crv{~*=<3pt>{.}(21,26)&(24,29)};
(30,30); (40,40)**\crv{~*=<3pt>{.}(31,36)&(34,39)};
(21,28)*{f_4};
(45,5); (45,3)**@{-};
(45,1)*{\scriptstyle 1};
(5,45); (3,45)**@{-};
(1,45)*{\scriptstyle 1};
(5,5); (5,3)**@{-};
(5,1)*{\scriptstyle 0};
(5,5); (3,5)**@{-};
(1,5)*{\scriptstyle 0};
\endxy
\qquad\qquad
\xy
(50,0); (0,0)**@{-}; (0,50)**@{-};
(0,0); (50,50)**@{.}; %(45,3)*{(G_4)};
(0,0); (10,10)**\crv{(6,1)&(9,4)};
(10,10); (20,20)**\crv{(16,11)&(19,14)};
(20,20); (30,30)**\crv{(21,26)&(24,29)};
(30,30); (40,40)**\crv{(31,36)&(34,39)};
(40,40); (50,50)**\crv{(41,46)&(44,49)};
(21,28)*{g_5};
(5,5); (15,15)**\crv{~*=<3pt>{.}(11,6)&(14,9)};
(15,15); (25,25)**\crv{~*=<3pt>{.}(21,16)&(24,19)};
(25,25); (35,35)**\crv{~*=<3pt>{.}(26,31)&(29,34)};
(35,35); (45,45)**\crv{~*=<3pt>{.}(36,41)&(39,44)};
(25,19)*{f_5};
(50,0); (50,-2)**@{-};
(50,-4)*{\scriptstyle 1};
(0,50); (-2,50)**@{-};
(-4,50)*{\scriptstyle 1};
(0,0); (0,-2)**@{-};
(0,-4)*{\scriptstyle 0};
(0,0); (-2,0)**@{-};
(-4,0)*{\scriptstyle 0};
\endxy
\]
\caption{\(G_{\tau_4} := \gen{f_4,g_4}\) and \(G_{\tau_5} :=
\gen{f_5,g_5}\). The EA-classes of these groups are \(\omega^\omega
+ 2\) and \(\omega^{\omega^\omega}+2\), respectively.}
\label{Gt4Gt5_fig} \end{figure}
\begin{thm} \label{main_thm}
There is a transfinite sequence \((G_\xi \mid \xi < \epsilon_0)\) of finitely
generated elementary amenable subgroups of \(F\) such that:
\begin{itemize}
\item \(G_0\) is the trivial group and \(G_{\xi+1} \cong G_\xi + \Z\);
\item \(G_\xi\) embeds into \(G_\eta\) if and only if \(\xi \le \eta\);
\item Given \(0 \le \alpha < \epsilon_0\) and \(n<\omega\), let
\(\xi = \omega^{(\omega^\alpha) \cdot (2^n)}\). If \(\alpha>0\), then
the EA-class of \(G_{\xi}\) is \(\omega \cdot \alpha + n + 2\).
If \(\alpha=0\), then the EA-class of \(G_\xi\) is \(n+1\).
\end{itemize}
\end{thm}
\noindent
In particular, for each \(\alpha < \epsilon_0\), there is a \(\xi\) such that
the EA-class of \(G_\xi\) is \(\alpha + 2\).
(If the EA-class of a finitely generated group is infinite, it is always of the form \(\alpha+2\).)
Thus Theorem \ref{main_thm} improves
previous work of the second author \cite{BrinEG},
who demonstrated that there are finitely generated subgroups of
\(F\) in \(EG\) of class \(\xi + 2\) for each \(\xi < \omega^2\).
With \(\omega\) the smallest infinite ordinal, the ordinal
\(\epsilon_0\) is the smallest ordinal solution to the equation
\(\omega^{x}= x\).
If we define a sequence \((\tau_k)_{k\in \omega}\) of
ordinals recursively by \(\tau_0:=2\), \(\tau_1:=\omega\) and
\(\tau_{k+1}:=\omega^{\tau_k}\) for \(k>1\), then \(\epsilon_0\) can
be described as
\[
\epsilon_0 = \sup\{\tau_k\mid k\in\omega\} =
\omega^{\omega^{\omega^{\omega^{\cdot^{\cdot^\cdot}}}}}.
\]
This countable ordinal is well known to play a central role in proof theory
and in particular in understanding the limitations and consistency
of Peano Arithmetic (see e.g. \cite[\#\! 4]{Gentzen_collected} \cite{MR0491157} \cite{MR663480} \cite[\S D8]{MR0491063}).
The groups in \(\Sfrak:=\{G_\xi \mid \xi < \epsilon_0\}\) are built
from \(\Z\) using certain familiar group-theoretic operations ---
direct sums and wreath products --- as well as a new operation which
is analogous to ordinal exponentiation base \(\omega\).
Whether this new operation
is meaningful in a broader setting
is unclear but even in our rather restrictive setting,
it already yields a wealth of examples.
The operations also make the construction of the groups in \(\Sfrak\)
straightforward and highly analogous to the construction of ordinals below
\(\epsilon_0\) from \(0\) using exponentiation base \(\omega\) and addition.
Specifically, given the Cantor normal form for an ordinal \(\xi<\epsilon_0\)
there is an efficient algorithm that lets one write down a finite number of generators
(explicitly as words in the generators of \(F\) if desired)
for a group with EA-class \(\omega \cdot \xi +2\).
While the results of this paper concern groups, the focus of the
analysis is on generating sets.
The groups in \(\Sfrak\) are specified by a family of generating
sets \(\Sgen\).
This collection has the property that \(A\) is in \(\Sgen\) if and only if
each of its two element subsets is in \(\Sgen\).
The 2-element sets in \(\Sgen\) generate precisely the groups
\(G_{\tau_k}\) in the family \(\Sfrak = \{G_\xi \mid \xi <
\epsilon_0\}\);
this is the reason for setting \(\tau_0 := 2\).
Theorem \ref{main_thm} implies, in particular, that the
\(G_{\tau_k}\) are an infinite
family of elementary amenable \(2\)-generated subgroups of \(F\)
which are not pairwise biembeddable.
Two of these generating pairs
are illustrated in Figure \ref{Gt4Gt5_fig}.
The isomorphism types of the \(G_{\tau_k}\) are
parametrized by the nonnegative integer \(k\) which we refer to as
the \emph{oscillation} of the generating pair from \(\Sgen\).
Figure \ref{Gt4Gt5_fig} illustrates pairs with
oscillation 4 and 5.
The function giving the oscillations of the pairs from an
\(A\in\Sgen\) is the \emph{signature} of \(A\).
Each generating set in \(\Sgen\) is equipped with a total order, and
the signature serves as a complete invariant for all of
\(\Sgen\).
\begin{thm} \label{sig_thm}
If \(A,B \in \Sgen\) have the same
signature, then the order preserving bijection from \(A\) and \(B\)
extends to an isomorphism from \(\gen{A}\) to \(\gen{B}\).
\end{thm}
\noindent
Thus one may analyze \(\Sfrak\) by analyzing the set \(\Ssig\) of all signatures of \(\Sgen\).
We also algebraically characterize the relation \(\leq\) on \(\Ssig\) which comes from the embeddability relation
on \(\Sfrak\).
The family \(\Sgen\) is %also
robust at a group-theoretic level:
if \(A \in \Sgen\), then \(\gen{A}\) is an HNN extension of a group which is itself an increasing union
subgroups of the form \(\gen{B}\) for \(B \in \Sgen\).
On the other hand, while the closure properties of \(\Sgen\) --- and thus of \(\Ssig\) ---
are important in the group-theoretic analysis of \(\Sfrak\), they introduce redundancies which
obscure the structure of the order on these classes.
This is resolved by introducing algebraic operations \(+\), \(*\), and \(\exp\) on \(\Ssig\) and using them to define
a subclass \(\Rsig\) of \(\Ssig\).
The next theorem is at the core of the proof of Theorem \ref{main_thm}.
It shows that \(\Rsig\) provides a
notion of ``normal form'' for \(\Ssig\) and consequently for \(\Sgen\)
(here \(\As \equiv \Bs\) denotes \(\As \leq \Bs \leq \As\)).
\begin{thm} \label{R_thm}
For each \(\As\) is in \(\Ssig\) there is a unique \(\Bs\) in \(\Rsig\) such that \(\As \equiv \Bs\).
Moreover there is a natural isomorphism
\[
(\Rsig,<,+,\exp) \cong (\epsilon_0,\in,+,\zeta \mapsto \omega^{-1 + \zeta})
\]
provided \(+\) is restricted to those pairs in \(\Rsig\) for which
the sum remains in \(\Rsig\).
\end{thm}
\noindent
Thus each biembeddability class in \(\Sfrak\) has a distinguished representative
--- unique up to marked isomorphism --- identified by the form of its signature.
Moreover, this representative can be viewed as being built up from \(\Z\) using simple
arithmetic operations which are analogs of the fundamental operations of ordinal arithmetic.
The well-foundedness and linearity of \((\Ssig,<)\) are subtle matters and likely to be of
independent interest.
In fact while it can be phrased in the language of arithmetic, the well-foundedness of \((\Ssig,<)\) is not
provable in Peano Arithmetic.
This is a consequence of Gentzen's analysis of the consistency of Peano Arithmetic \cite[\#\! 4]{Gentzen_collected} and
G\"odel's second incompleteness theorem \cite{incompleteness} \cite{Godel_collected1}.
At a more pragmatic level, future methods of proof
may lend themselves more naturally to induction on
\(\Ssig\) than to induction on \(\epsilon_0\).
Extending the chain \(\Sfrak\) by setting \(G_{\epsilon_0} = F\), we
make the following conjectures. Recall that \(\PLoI\) is the group
of all piecewise linear elements of \(\HomeoI\).
\begin{conj} \label{max_chain_conj}
\(\{G_\xi \mid \xi \leq \epsilon_0\}\) is a maximal chain in \(\Ffrak\).
\end{conj}
\begin{conj} \label{univ_conj}
If \(H\) is a finitely generated subgroup of \(\PLoI\), then either
\(F\) embeds into \(H\) or else there is an \(\eta < \epsilon_0\)
such that \(H\) embeds into \(G_\eta\).
\end{conj}
\begin{conj} \label{WQO} The partial order \((\Ffrak,\emb)\) is a
\emph{well-quasi-order} --- it contains no infinite decreasing
sequences and no infinite antichains.
\end{conj}
Observe that Conjecture \ref{univ_conj} immediately implies
Conjecture \ref{BrinSapir} since each \(G_\eta\) for \(\eta <
\epsilon_0\) is elementary amenable. It also implies another
conjecture of the second author which complements Conjecture
\ref{BrinSapir}: every elementary amenable subgroup of \(\PLoI\)
embeds into \(F\). Moreover, this would imply that \(\epsilon_0\)
is a strict upper bound for the EA-class of every finitely generated
elementary amenable subgroup of \(F\).
Conjecture \ref{WQO} is really about understanding those finitely
generated subgroups of \(F\) which do not contain \(F\). The second
author has shown that not containing an isomorphic copy of \(F\) is
a strong restriction on subgroups of \(F\) (and more generally
subgroups of \(\PLoI\)) \cite{BrinU}. Motivation for Conjecture
\ref{WQO} stems in part from the heuristic that forbidding a
ubiquitous substructure often portends a well developed structure
theory (see, for instance \cite{Fraisse_conj}).
There are limitations, however, as to what one can expect in the direction
of these conjectures.
We show that there is a finitely generated subgroup of \(F\)
which is not biembeddable with any \(G_\xi\) for \(\xi \leq \epsilon_0\).
Moreover, we show that \((\Ffrak,\emb)\) is not a linear ordering.
\begin{thm}\label{AntiChain}
There are finitely generated subgroups \(H_0\) and \(H_1\) of \(F\) of EA-class
\(\omega+2\) such that \(H_0\) does not embed into \(H_1\) and vice versa.
\end{thm}
Lastly we remark, without proof, that the sequence of groups
\(G_{\tau_k}\) with the generating pairs \((f_k, g_k)\) converges in
the space of 2-marked groups to the free group on 2 generators. We
refer the reader to \cite{MR2653970} for definitions and relevant
arguments.
\subsection{Related results}
All solvable groups are necessarily elementary amenable. The
solvable subgroups of \(F\) have been thoroughly analyzed by the
first author in \cite{Bpasc}. In particular, he proves that
Conjecture \ref{univ_conj} holds for the finitely generated solvable
subgroups \(H\) of \(\PLoI\) \cite{Bpasc}: every finitely generated
solvable subgroup of \(\PLoI\) embeds into \(G_{\omega^\omega}\),
which is the Brin-Navas group \(B\) (\cite{BrinEG} Fig. 5, \cite{MR2135961}
Example 6.3).
In \cite{ATaylor}, A.~Taylor finds uncountably many pairwise
nonisomorphic elementary amenable subgroups of \(F\). These
examples are not finitely generated, are not solvable but are
locally solvable, and all have EA-class \(\omega+1\).
In \cite{MR3079257}, Ol'shanskii and Osin show that for every countable ordinal \(\alpha\),
there is a finitely generated group \(G\in EG\) with \(\EA(G)=\alpha+2\).
The construction in \cite{MR3079257} is recursive and based on HNN extensions.
The operations on \(\Sfrak\) in the current paper accelerate the recursive process and lead to
easier descriptions of the elements of \(\Sfrak\).
\subsection{Organization}
Section \ref{objects:sec} gives definitions and examples sufficient
to introduce the reader to the groups that we build and how we build them.
It does not give any hints as to their analysis.
Section \ref{prelim:sec} fixes more notation and terminology which
is used in the paper. It also contains a review of a number of
prerequisites for the paper:
details from \cite{fast_gen};
ordinals and
their arithmetic; elementary amenable groups and EA-class;
wreath products of permutation groups. The reader may wish to skim or skip
Section \ref{prelim:sec} and then refer back to the various subsections as needed.
In Section \ref{osc:sec}, the oscillation function is developed.
This function is further developed in the context of standard generating sets
in Section \ref{signature:sec}, where we study the signature of an element of \(\Sgen\)
and prove Theorem \ref{sig_thm}.
In Section \ref{inflation:sec} we introduce the notion of an \emph{inflation} of a standard generating
set by one of its elements and show that the result is again a standard generating set.
The interaction of the family \(\Sgen\) with wreath products is detailed
in Section \ref{wreath:sec}.
Section \ref{arithmetic:sec} develops analogs of the operations of ordinal arithmetic
for signatures of elements of \(\Sgen\).
This analysis is then used to show that
\((\Sfrak,\emb)\) is a well-order with ordertype \(\epsilon_0\) in
Section \ref{main:sec}, where the proofs of Theorems \ref{main_thm}
and \ref{R_thm} are completed.
Finally, Section \ref{nonlinear:sec}
contains a proof of Theorem \ref{AntiChain},
establishing that \((\Ffrak,\emb)\) is not a linear order.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The objects of study}
\label{objects:sec}
In this section we describe the generating sets in \(\Sgen\)
and the functions that are elements of such sets. We also
describe the signature associated to a generating set. The class of
signatures \(\Rsig\) of Theorem \ref{R_thm} is described as
well as the isomorphism from \(\Rsig\) to \(\epsilon_0\).
An aim of this section is to indicate how, given a suitable EA class
\(\alpha<\epsilon_0\), one can write down a set of generators of a
group with EA class \(\alpha\).
\subsection{The anatomy of a homeomorphism}\label{anat:sec}
To describe the elements of \(\HomeoI\) that we work with, we set
some terminology. We write homeomorphisms to the right of their
arguments and compose from left-to-right. The {\itshape support}
\(\supt(f)\) of \(f\in \HomeoI\) is the set \(\{t\in [0,1]\mid tf\ne
t\}\), and the {\itshape extended support} of \(f\) is the interior
of the closure of \(\supt(f)\).
For \(f\in \HomeoI\), a component \(J\) of \(\supt(f)\) is an
{\itshape orbital} of \(f\), and a function with exactly one orbital
is be called a {\itshape bump}. The \emph{transition points} of
\(f\) are the endpoints of the orbitals of \(f\). A bump \(f\) with
orbital \(J\) is {\itshape positive} if \(tf>t\) for one
(equivalently all) \(t\in J\), and negative otherwise. If \(g\in
\HomeoI\) has multiple orbitals one of which is \(J\), then the bump
\(f\in \HomeoI\) with \(f|_J=g|_J\) is called a {\itshape bump
of} \(g\).
%
If \(f \in \HomeoI\) and \(X \subseteq I\) is a union of orbitals
and fixed points of \(f\), then we write \(f |_X\) to denote
the homeomorphism which coincides with \(f\) on \(X\) and which is
the identity outside of \(X\).
To conserve space in drawing functions, we do not use horizontal and
vertical \(x\) and \(y\) axes as in Figure \ref{Gt4Gt5_fig}, but
draw as if the \(x\) and \(y\) axes are at 45 degrees and the part
of the line \(y=x\) in the first quadrant stretches horizontally to
the right from the origin. The axes themselves and the line \(y=x\)
are suppressed, as are intervals of fixed points. As in the following,
%
\begin{equation}\label{BumpsPic}
\xy
(48,0); (64,0)**\crv{(52,4)&(60,4)}; (46,0.5)*{f};
\endxy
\qquad
\qquad
\xy
(8,0); (24,0)*\crv{(12,-4)&(20,-4)};
(40,0)**\crv{(28,4)&(36,4)};
(56,0)**\crv{(44,4)&(52,4)};
(6,0.5)*{g};
\endxy
\end{equation}
%
positive bumps become arcs above the horizontal, and negative bumps
are arcs below the horizontal. The function \(f\) is a positive
bump, and the function \(g\) has one negative bump and two positive
bumps.
\subsection{The attribute fast}\label{FastAttrSec}
To control the isomorphism type of the
groups that we generate, we need some mild controls on the
dynamics of our generating sets. We use some concepts from
\cite{fast_gen}.
Suppose that \(A \subset \HomeoI\) is a set of bumps. A
\emph{\Marking{}} of \(A\) is an assignment \(a \mapsto s_a\) of an
element of the support of \(a\) to each \(a \in A\). The
\emph{feet} of \(a\) with respect to this \Marking{} are the
intervals \((u,s_a)\) and \([t_a,v)\) where \((u,v)\) is the support
of \(a\) and \(t_a:= s_a a\) if \(a\) is positive and \(t_a:= s_a
a^{-1}\) if \(a\) is negative. This is illustrated below for
positive \(a\).
%
\[
\xy
(0,0); (32,0)**\crv{(10,10)&(22,10)}; (4,5.5)*{a}; %(16,7.3)*{\cdot};
(8.7, 0); (16,7.3)**@{.}; (23.3,0)**@{.};
(0,-2)*{u}; (8.7, -2.5)*{s_a}; (23.3, -2.5)*{t_a}; (32,-2)*{v};
\endxy
\]
An \(f\in \HomeoI\) equipped with a \Marking{} of its bumps is a
\emph{\Marked{} function}. The expression ``a \Marked{} function
\(f \in \HomeoI\)'' always means that \(f\) comes with a fixed
\Marking{} even if the \Marking{} is not specified.
The finiteness assumptions in the next definitions are more
restrictive than in \cite{fast_gen}, but are sufficient for us and
add some conveniences. A collection \(B\) of \Marked{} bumps is
\emph{geometrically fast} (or just \emph{fast}) if for every \(f\ne
g\) in \(B\) the feet of \(f\) are disjoint from the feet of \(g\).
If \(S \subset \HomeoI\) is a finite set of \Marked{} functions and
each element of \(S\) has only finitely many bumps, then we define
\(S\) to be \emph{fast} if the set of bumps of \(S\) is fast and no
bump occurs in more than one element of \(S\). Given a fast \(S\),
we always view \(S\) as being ordered according to the order on
the maximum transition points of its elements. Note that no two
different elements of such an \(S\) are equal under this order.
The point of these concepts is that the isomorphism type of a
subgroup of \(\HomeoI\) generated by a fast
set \(S\) of functions is determined by a very small amount of
information from \(S\). Specifically, if \(S'\) is another
fast set and \(h:S\rightarrow S'\) is a bijection that
``preserves enough of the combinatorics'' (including the order given
above, the order of the feet, and the signs of the bumps), then \(h\)
extends to an isomorphism from \(\gen S\) to \(\gen{S'}\).
A more detailed statement is given in Section \ref{FastGenSec}.
We exploit this in two ways. We can specify groups up to
isomorphism by giving somewhat sloppy descriptions of the generating
sets. Further, a result of \cite{fast_gen} says that if \(S\) is
fast, then \(\gen S\) embeds in Thompson's group \(F\). Thus under
the assumption that the pairs \(\{f_4, g_4\}\) and \(\{f_5, g_5\}\)
in Figure \ref{Gt4Gt5_fig} are fast, we can regard the groups
\(G_{\tau_4}\) and \(G_{\tau_5}\) as subgroups of \(F\) with very
specific isomorphism types.
\subsection{Standard functions and standard pairs}
Thompson's group \(F\) is not elementary amenable, and to build a
subgroup of \(F\) that is elementary amenable one must avoid
including an isomorphic copy of \(F\) in the subgroup. By the
Ubiquity Theorem \cite{BrinU}, this places severe restrictions on
the generating sets and the functions that can be used in the
generating sets. The restrictions in the next definition
reflect this.
%
A {\itshape standard function} is a \Marked{} function \(f\) in
\(\HomeoI\) with finitely many bumps satisfying all following
properties.
%
\begin{enumerate}
\item The extended support of \(f\) is an interval;
\item every positive bump of \(f\) is to the right of every negative bump;
\item the number of positive and negative bumps of \(f\) differ by
at most one and there are at least as many positive bumps as
negative bumps.
\end{enumerate}
%
The functions \(f\) and \(g\) in (\ref{BumpsPic}) are both standard.
A set \(A\) of standard functions forms an element of \(\Sgen\)
if each pair of elements of \(A\) forms a \emph{standard pair}. We need
some notation to define standard pairs. If \(f, g\in \HomeoI\)
are standard, then we write \(f\ll g\) if their extended supports
are disjoint and the extended support of \(f\) is to the left in
\([0,1]\) of the extended support of \(g\). We write \(f\sqsubset
g\) if the closure of the extended support of \(f\) is contained in
the extended support of \(g\). We write \(f<g\) if \(f\ll g\) or
\(f\sqsubset g\).
If \(S\in \HomeoI\) is fast and is totally
ordered by \(<\) as just defined, then \(<\) is identical to the
order on \(S\) from Section \ref{FastAttrSec} determined by the
order of the rightmost transition points.
The diagram below illustrates two pairs \((f,g)\) of standard
functions, where the left pair satisfies
\(f \ll g\) and the right pair satisfies \(f \sqsubset g\).
\begin{equation}\label{PairsInOrder}
\xy
(-14,4.0)*{f};
(-6,1.0)*{\vardotbump{4}{12}{16}};
(14, 1.0)*{\varbump{4}{12}{16}};
(23.5,3)*{g};
\endxy
\qquad
\xy
(0.0,8.0); (0.0,-8.0)**@{.};
\endxy
\qquad
\xy
(-8,3.5)*{g};
(0,-.85)*{\varnbump{4}{12}{16}};
(16,2.6)*{\varbump{4}{12}{16}};
(32,2.6)*{\varbump{4}{12}{16}};
(9,-0.8)*{\varndotbump{4}{12}{16}};
(25,2.6)*{\vardotbump{4}{12}{16}};
(0.0,4.0)*{f};
\endxy
\end{equation}
Standard pairs of functions are defined recursively using a
reduction operation that will be heavily used during our analysis in
later sections. Suppose that \(f\) is a standard function. If
\(f\) has more than two orbitals, then let \(X\) be the union of the
orbitals of \(f\) except the maximum (rightmost) and minimum
(leftmost) orbitals, and
define \(f \cut\) to be \(f|_X\).
Observe that \(f
\cut\) is again a standard function --- we \Mark{} \(f \cut\) with
the \Markers{} of \(f\) on the orbitals which remain in \(f \cut\).
If \(f\) has one or two orbitals and the left foot of the positive orbital is
\((r,s)\), define \(f\cut\) to be any positive bump with support
\((r,s)\) and an arbitrary \Marker{}. It will not be necessary
to be more specific.
It will be important later that in all cases each foot of \(f \cut\) is
a subset of a foot of \(f\).
A pair \((f,g)\) of standard functions is a \emph{standard pair}
if it satisfies the following recursive definition:
\begin{enumerate}
\item The set \(\{f,g\}\) is fast, and
\item either \(f \ll g\), or else \(f\sqsubset g\) and \((g\cut,f)\)
a standard pair.
\end{enumerate}
The following diagram illustrates the recursive clause:
\[
\xy
(0,3)*{g};
(0,0); (8,0)**\crv{(2,-3)&(6,-3)};
(16,0)**\crv{(10,3)&(14,3)}; (24,0)**\crv{(18,3)&(22,3)};
(4,0); (12,0)**\crv{~*=<3pt>{.}(6,-3)&(10,-3)};
(20,0)**\crv{~*=<3pt>{.}(14,3)&(18,3)};
(4,3.5)*{f};
\endxy
\quad
\leadsto
\quad
\xy
(9,4)*{g\cut};
(8,0); (16,0)**\crv{(10,3)&(14,3)};
(4,0); (12,0)**\crv{~*=<3pt>{.}(6,-3)&(10,-3)};
(20,0)**\crv{~*=<3pt>{.}(14,3)&(18,3)};
(4,3.5)*{f};
\endxy
\quad
\leadsto
\quad
\xy
(13,4)*{g\cut};
(12,0); (24,0)**\crv{(15,3)&(21,3)};
(15.5,0); (20.5,0)**\crv{~*=<3pt>{.}(17.5,1.5)&(18.5,1.5)};
(20.5,-3.5)*{f\cut};
\endxy
\quad
\leadsto
\quad
\xy
(10,3.5)*{{g\cut}\cut};
(7.5,0); (12.5,0)**\crv{(9.5,1.5)&(10.5,1.5)};
(15.5,0); (20.5,0)**\crv{~*=<3pt>{.}(17.5,1.5)&(18.5,1.5)};
(18,4)*{f\cut};
\endxy
\]
%
It is easily checked that both pairs \((f,g)\) in
(\ref{PairsInOrder}), and the pairs \((f_4,g_4)\) and \((f_5,g_5)\)
in Figure \ref{Gt4Gt5_fig} are standard and that, up to topological
conjugacy, \((g_5 \cut,f_5)\) coincides with \((f_4,g_4)\).
A \emph{standard generating set} is now defined mean a finite set of marked
functions in \(\HomeoI\) which is pairwise standard; the collection of all
standard generating sets is denoted \(\Sgen\).
\subsection{Oscillation and signature}\label{OscSig:sec}
In our analysis, we reduce the information in a standard generating
set to a finite matrix of integers. If \(f < g\) is a fast pair of
\Marked{} functions, then we define their \emph{oscillation}
\(o(f,g)\) to be the number of orbitals of \(g\) that contain at
least one transition point of \(f\).
If \(A\) is in \(\Sgen\), then the {\itshape signature} of \(A\) is
the function \(\As\) defined by
\(\As(f,g) = o(f,g)\) whenever \(f < g\) are in \(A\).
We refer to \(A\) as the \emph{base} of the signature \(\As\).
Figure \ref{osc_fig} shows the signature of a
standard generating set with four elements.
Since elements of \(\Sgen\) may have either one or even no elements,
we include the base as part of the data of a signature.
If \(\As\) and \(\Bs\) are signatures with bases \(A\) and \(B\), then we say that \(\As\) and \(\Bs\) are
\emph{equivalent} if \(|A| = |B|\) and the order
preserving bijection \(\theta : A \to B\)
satisfies \(\As(f,g) = \Bs(\theta(f), \theta(g))\) whenever \(f < g\) are in \(A\).
We also extend this notion of equivalence to when \(\As\) and \(\Bs\) are
just integer functions defined on pairs from ordered sets \(A\) and \(B\);
we use \emph{signature} to refer to a function which is equivalent in this way to a signature
of an element of \(\Sgen\).
In particular, every signature is uniquely equivalent to a signature with base
\(\{0,\ldots,n-1\}\) for some \(n\).
These canonical representatives allow us say without guilt
that a signature \(\As\) {\emph is} the signature of some \(S\in \Sgen\) when in reality \(\As\)
is only equivalent to the signature of \(S\).
Theorem \ref{sig_thm} (proven in Section \ref{signature:sec}) says that if \(A,B \in \Sgen\)
have the same signature, then the order preserving bijection between \(A\) and \(B\)
extends to an isomorphism between \(\gen A\) and \(\gen B\).
If \(\As\) is a signature, we define \(\gen{ \As } := \gen{A}\) where \(A \in \Sgen\) has
signature (equivalent to) \(\As\).
\begin{figure}
\[
\xy
%
(0,0); (40,0)**\crv{(15,-15)&(25,-15)};
(40,0); (60,0)**\crv{(47,7)&(53,7)};
%
(5,0); (35,0)**\crv{~*=<3pt>{.}(15,-10)&(25,-10)};
(35,0); (55,0)**\crv{~*=<3pt>{.}(42,7)&(48,7)};
%
(10,0); (20,0)**\crv{(13,-3)&(17,-3)};
(20,0); (30,0)**\crv{(23,3)&(27,3)};
(30,0); (50,0)**\crv{(37,7)&(43,7)};
%
(15,0); (25,0)**\crv{~*=<3pt>{.}(18,-3)&(22,-3)};
(25,0); (45,0)**\crv{~*=<3pt>{.}(32,7)&(38,7)};
%
(0,2)*{d};
(55,-2)*{c};
(10,2)*{b};
(45,-2)*{a};
%
%
(70,18); (99, 18)**@{-}; (99,-11)**@{-}; (70,18)**@{-};
%
(79,21)*{c}; (87,21)*{b}; (95,21)*{a};
(69,14)*{d}; (79,14)*{2}; (87,14)*{2}; (95,14)*{2};
(69,6)*{c}; (87,6)*{2}; (95,6)*{2};
(69,-2)*{b}; (95,-2)*{3};
\endxy
\]
\caption{Four functions that are pairwise standard with various
oscillation numbers.
Their signature is displayed on the right.
They generate a group of
EA-class \(\omega^2+\omega\cdot 2+2\).}
\label{osc_fig} \end{figure}
\subsection{Arithmetic on signatures and normal forms}\label{BasicOp:sec}
We now introduce arithmetic operations on the set of
signatures which allow us to readily specify
complex standard generating sets.
The following theorem of Cantor is important motivation.
(Ordinal arithmetic is reviewed in Section \ref{ordinal:sec}.)
\begin{thm}\label{CNF} \cite[\S 19 Theorem B]{MR0045635}
If \(\alpha\) is a nonzero ordinal, there is a unique sequence
\(\beta_0\ge \beta_1 \ge \cdots \ge \beta_n\) such that
\[
\alpha = \sum_{i \leq n} \omega^{\beta_i} =
\omega^{\beta_0} + \omega^{\beta_1} + \ldots + \omega^{\beta_n}.
\]
\end{thm}
If \(0 < \alpha<\epsilon_0\), each positive \(\beta_i\) is less than \(\alpha\)
and may be further expanded in the form
of Theorem \ref{CNF}.
Iterating the process, one obtains an expression of
\(\alpha\) in terms of \(0\), \(+\) and exponentiation base
\(\omega\) which is known as the \emph{Cantor normal form} of \(\alpha\).
We now define analogs of these arithmetic operations on the set of signatures.
We use \(\zero\) to denote the signature with empty base
and \(\one\) to denote the signature with singleton base;
notice that if \(|Z| = 1\), then \(\gen{Z} \cong \Z\).
If \(A,B,C\) is in \(\Sgen\) then we define \(A = B + C\) to mean:
\begin{itemize}
\item \(A = B \cup C\) and \(b < c\) for all \(b \in B\) and \(c \in C\), and
\item \(o(b,c) = 0\) for all \(b \in B\) and \(c \in C\).
\end{itemize}
Notice that \(b \ll c\) is equivalent to \(o(b,c) = 0\) and \(b < c\).
Thus if \(A = B + C\), then \(\gen{A} = \gen{B} + \gen{C}\).
Observe that while \(A=B+C\) expresses a ternary relation on \(\Sgen\),
it induces a well defined binary operation \(+\) on the set \(\Ssig\) of signatures.
Additionally, we will show in Section \ref{signature:sec}
that if \(\As \in \Ssig\), then the function \(\exp(\As)(a,b) = \As(a,b) + 1\) defined for \(a < b\) are in \(A\)
is equivalent to signature of some standard generating set.
Observe that the ordinals below \(\epsilon_0\) can all be generated
from \(\{0,1\}\) via the operations \(+\) and \(\zeta \mapsto \omega^{-1 + \zeta}\).
Moreover a slight modification of the Cantor normal form specifies, for each \(\xi < \epsilon_0\),
a unique term in these operations which generates \(\xi\).
Let \(\Rs_\xi \in \Ssig\) denote the result of evaluating the same term
but in the structure \((\Ssig,\zero,\one,+,\exp)\) in place of \((\epsilon_0,0,1,+,\zeta \mapsto \omega^{-1 + \zeta})\).
The family \(\Rsig\) (of \emph{reduced signatures}) mentioned in
the introduction is defined to be the collection \(\{\Rs_\xi \mid \xi < \epsilon_0\}\);
the groups \(G_\xi\) from Theorem \ref{main_thm} are given by \(G_\xi := \gen{\Rs_\xi}\).
To summarize, we will eventually prove Theorems \ref{main_thm} and \ref{R_thm}, showing that:
\begin{itemize}
\item every \(\Sgen\)-generated group is biembeddable with some \(G_\xi\);
\item \(G_\xi\) embeds into \(G_\eta\) if and only if \(\xi < \eta < \epsilon_0\);
\item if \(\xi = \omega^{\omega^\alpha \cdot 2^n}\) for \(0 < \alpha < \epsilon_0\) and \(0 < n < \omega\),
then the EA-class of \(G_\xi\) is
\(\omega \cdot \alpha + n + 2\).
\end{itemize}
In fact we can take the analogy with ordinal arithmetic further.
If \(A,B,C \in \Sgen\) we define \(A = B * C\) to mean:
\begin{itemize}
\item \(C = A \cup B\) and \(a < b\) for all \(a \in A\) and \(b \in B\), and
\item \(o(a,b) = 1\) for all \(a \in A\) and \(b \in B\).
\end{itemize}
The ternary relation \(A = B*C\) on \(\Sgen\) allows us to define an operation \(*\)
at the level of signatures as in the case of \(+\).
We show in Section \ref{wreath:sec} that if \(A = B* C\) for nonempty \(B,C\)
and \(o(f,g) > 0\) for all \(f < g\) in \(A\), then \(\gen{A} \cong \gen{B} \wr \gen{C}\).
We caution that \(\wr\) is a \emph{permutation wreath product} and typically not the \emph{standard wreath product}
--- see Section \ref{wreath:subsec}.
Also, it can be shown that if \(1 < \beta \leq \alpha < \epsilon_0\) are indecomposable ordinals,
then \(\Rs_{\alpha \cdot \beta} \equiv \Rs_\alpha * \Rs_\beta\) with equality holding if \(\Rs_\alpha * \Rs_\beta \in \Rsig\).
Finally, if \(\As \in \Ssig\), define \(E(\As) = \exp(\exp((\As))\).
Unlike \(\exp\), \(E\) also comes from a natural binary relation on standard generating sets:
if \(A,B \in \Sgen\) then \(A = E(B)\)
asserts that \(o(f,g) \geq 2\) for all \(f < g \in A\) and
\(B = \{f \cut \mid f \in A\}\).
The operations \(+\), \(*\), and \(E\) generate \(\Rsig\) in the following strong sense.
\begin{prop}
if \(\As \in \Rsig\), then exactly one of the following hold:
\begin{itemize}
\item \(\As = E(\Bs)\) for some \(\Bs \in \Rsig\);
\item there are \(\Bs,\Cs \ne \zero\) in \(\Rsig\) such that
\(\As = \Bs + \Cs\), in which case \(\gen{\As} = \gen{\Bs} + \gen{\Cs}\);
\item there are \(\Bs,\Cs \ne \zero\) in \(\Rsig\) such that
\(\As = \Bs * \Cs\), in which case \(\gen{\As}\) is the permutation wreath product
\(\gen{\Bs} \wr \gen{\Cs}\)
\end{itemize}
In particular, each \(\As \ne \zero\) in \(\Rsig\) can be uniquely
expressed as a term in the operations \(+\), \(*\), and \(E\) and
the constant \(\one\) so that \(E(\one)\) does not occur as a subterm.
\end{prop}
For instance, the example in Figure \ref{osc_fig} is \(E(\one * \one+\one+\one)\).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Intermission}
\label{prelim:sec}
\subsection{Background and conventions}
We adopt the convention that the natural numbers include \(0\)
and in particular all counting starts at \(0\). We use
\(\Z\) to denote the set of integers. Unless otherwise stated,
\(i,j,k,l,m,n\) range over the natural numbers and \(p,q,r\)
range over the integers.
For instance we write
\(i < n\) to mean that \(i\) is a natural number less than \(n\).
If \(G\) and \(H\) are two groups, we use \(G + H\) to denote
their direct sum.
Even though we work primarily with nonabelian groups,
this additive notation and terminology fits better with the
correspondence we develop between
\(\Sfrak\) and the ordinals below \(\epsilon_0\).
%
Given \(f,g\in \HomeoI\), we denote by \(f^g\) the
conjugate \(g^{-1}fg\) and remark that under this notation
\(\supt(f^g)=\supt(f)g\).
Given
\(X\subset I\) we write \(Xf\) for
%
\(
\left\{xf\mid x\in X\right\}.
\)
%
In this paper we think of \(F\) as being a subgroup of
\(\PLoI\), although there typically is not a need to fix a
specific model. Representing \(F\) in \(\PLoI\) gives us access to
results about subgroups of \(\PLoI\) (specifically results about
centralizers). This representation also guarantees that any element
\(f\) of \(F\) has only finitely many components of its support and
hence is a product of the bumps of \(f\).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Fast generating sets}\label{FastGenSec}
We now give more details to some of the claims in Section
\ref{FastAttrSec} and more information from
\cite{fast_gen}.
If \(S \subset \HomeoI\) is a fast set of
\Marked{} functions,
then
the \emph{dynamical diagram} of \(S\) is the edge labeled, ordered,
directed graph \(D_S\) defined as follows: \begin{itemize}
\item the vertices of \(D_S\) are the feet of \(S\) with the order
induced from \(I\);
\item the directed edges of \(D_S\) are the bumps \(a\) which occur
in \(S\) with the endpoints of \(a\) being the feet of \(a\);
\item positive bumps are directed to the right and negative bumps
are directed to the left;
\item the label of a directed edge \(a\) in \(D_S\) is the unique
\(f\in S\) for which \(a\) is a bump of \(f\).
\end{itemize}
Note that such ordered directed graphs do not have multiple edges.
We sometimes refer to the constituents of a dynamical
diagram --- the vertices, the directed edges, the labels --- in
terms of the objects they are intended to represent --- the feet,
the bumps, the functions in \(S\).
For two fast sets of \Marked{} functions \(S_0\) and \(S_1\), an
isomorphism from \(D_{S_0}\) to \(D_{S_1}\) is an isomorphism
\(\theta:D_{S_0}\rightarrow D_{S_1}\) of ordered directed graphs
such that for bumps \(a\) and \(b\) which occur in \(S_0\), the
labels of \(a\) and \(b\) are equal if and only if the labels of
\(\theta(a)\) and \(\theta(b)\) are equal. Note that the choice of
\Marking{} affects the dynamical diagram but not its isomorphism type.
Also observe that the isomorphism \(\theta\) induces a well defined
order preserving bijection from \(S_0\) to \(S_1\). Lastly, there
is at most one isomorphism between any two dynamical diagrams.
If \(A\) is a totally ordered finite subset of a group, then the pair
\((\gen{A},A)\) is a \emph{marked group} that is \emph{marked by}
\(A\). A \emph{marked isomorphism} \(\theta:(\gen{A},A)\rightarrow
(\gen{B},B)\) between marked groups is an isomorphism from
\(\gen{A}\) to \(\gen{B}\) that restricts to an order preserving
bijection from \(A\) to \(B\). The next theorem asserts that the
dynamical diagram determines the marked isomorphism type of
\(\gen{S}\) whenever \(S \subset \HomeoI\) is a fast set of
\Marked{} functions. (The two uses of ``marked'' in the previous
sentence are not the same.)
\begin{thm} \cite{fast_gen} \label{comb_to_iso} If \(S_0\) and
\(S_1\) are fast sets of \Marked{} functions in \(\HomeoI\), and the
dynamical diagrams of \(S_0\) and \(S_1\) are isomorphic, then the
order preserving bijection from \(S_0\) to \(S_1\) extends to an
isomorphism \(\gen{S_0} \cong \gen{S_1}\). \end{thm}
We also need the following proposition which is closely related
to the results of \cite{fast_gen}.
\begin{prop} \label{faithful_prop}
Suppose that \(A\) is a finite geometrically fast set of bumps and \(X
\subset I\) intersects each orbital of \(A\). If \(g \in
\gen{A}\) is not the identity, then there is an \(x \in X \gen{A}\)
such that \(xg \ne x\).
\end{prop}
\begin{proof} Observe that the closure of \(X \gen{A}\) contains
the transition points of \(A\). The proof of Proposition 4.3 of
\cite{fast_gen} yields a \Marking{} of \(A\) which witnesses that it
is geometrically fast and has the property that every \Marker{} is in
the closure of \(X\gen{A}\); let \(M\) denote the set of these
\Markers{}. From \cite{fast_gen}, if \(g \in \gen{A}\) is not the
identity, then there is a
\(t \in M \gen{A} \subset \overline{X \gen{A}}\) such that \(tg \ne t\).
Continuity of \(g\) implies that there is an \(x \in X \gen{A}\) such that \(xg \ne x\).
\end{proof}
It is convenient to develop some conventions for drawing dynamical
diagrams. First, we arrange the vertices horizontally from left to
right in increasing order. We draw right directed edges as
over-arcs and left directed edges as underarcs, suppressing the
arrows. If \(f\) is a generator and a right foot \(J\) of \(f\) is
immediately followed by a left foot \(J'\) of \(f\), then the pair
of vertices \(\{J,J'\}\) is contracted to a single vertex when
drawing the diagram. This has the effect of simplifying the
dynamical diagram visually. It also has the feature that if \(f \in
S\) has connected extended support, then the edges with label \(f\)
form a connected component of the contracted diagram. Thus it is
sufficient to label only one bump of each such component.
This graphical representation of \(D_S\) can be derived from the
graphs of the elements of \(S\) drawn as in Section \ref{anat:sec}.
The drawing
%
\[
\xy
(0,0); (16,0)**\crv{(4,-4)&(12,-4)}; (32,0)*\crv{(20,-4)&(28,-4)};
(48,0)**\crv{(36,4)&(44,4)}; (64,0)**\crv{(52,4)&(60,4)}; (0,2.5)*{g_4};
(8,0); (24,0)*\crv{~*=<3pt>{.}(12,-4)&(20,-4)};
(40,0)**\crv{~*=<3pt>{.}(28,4)&(36,4)};
(56,0)**\crv{~*=<3pt>{.}(44,4)&(52,4)};
(8,3)*{f_4};
\endxy
\]
is the graph of \(\{f_4, g_4\}\) from Figure
\ref{Gt4Gt5_fig} as drawn in Section \ref{anat:sec}
and it is also a drawing of the dynamical diagram for \(\{f_4,
g_4\}\). In general, the dynamical diagram of a fast set of standard
functions \(S\) is a sketch of the graphs of the functions in \(S\).
Suppose that \(S \subset \HomeoI\) is a finite fast set of \Marked{}
functions with connected essential supports. A bump \(b\)
of \(S\) is \emph{isolated} in \(S\) if its support contains no
transition points of \(S\).
If \(E\) is a set of isolated bumps of \(S\) and for each \(f \in S\) there is a bump
of \(f\) which is not in \(E\), then
we say that \(E\) is an \emph{extraneous set of bumps} of \(S\).
We need a result of \cite{fast_gen} which
says that extraneous sets of bumps can be \emph{excised} without affecting
the marked isomorphism type of \(S\).
This is made precise as follows.
For \(g \in \HomeoI\) and \(E\) a set of bumps
(not necessarily bumps of \(g\)), we define \(g/E\in\HomeoI\) to be
the function which agrees with \(g\) on
%
\[
I\setminus\bigcup\{\supt(b) \mid b\in E\ \text{and}\ b\ \text{is a bump of }\
g\}
\]
%
and is the identity elsewhere.
We extend the definition above to a set \(S \subset \HomeoI\) by:
\[
S/E := \{g/E \mid g \in S\}.
\]
The next theorem is a special case of Theorem \(9.1\) of \cite{fast_gen}.
\begin{thm}\label{excision} \cite{fast_gen} If \(S \subset \HomeoI\)
is a fast set of \Marked{} functions and \(E\) is an extraneous set of
bumps of \(S\), then the map \(g \mapsto g/E\) extends to a
isomorphism from \(\gen{S}\) to \(\gen{S/E}\). \end{thm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Ordinals and their arithmetic}\label{ordinal:sec}
Recall that an \emph{ordinal} is the isomorphism type of a well-ordered set.
If \(\alpha\) and \(\beta\) are ordinals, then
\(\alpha < \beta\) is defined to mean that there is well-order of
type \(\alpha\) which is a proper initial part of a well-ordering of
type \(\beta\). For any two ordinals \(\alpha\) and \(\beta\),
precisely one of the following is true: \(\alpha < \beta\), \(\beta
< \alpha\), or \(\alpha = \beta\). We adopt von Neumann's convention that an
ordinal is the set of its predecessors and that \(\alpha < \beta\) means \(\alpha \in \beta\).
The least ordinal is \(0 :=\emptyset\) and the least infinite ordinal
is \(\omega\), which can be thought of as coinciding with the
natural numbers. If \(A\) is a set of ordinals, then there is
always a least ordinal \(\sup(A):= \bigcup A\) which is an upper bound for \(A\).
If \(\alpha\) is an ordinal, then \(\alpha+1:= \alpha \cup \{\alpha\}\) is
the least ordinal greater than \(\alpha\). Ordinals of the form
\(\alpha + 1\) are said to be \emph{successor ordinals}; all other
nonzero ordinals are said to be \emph{limit ordinals}.
It is possible to extend the usual arithmetic operations on the finite ordinals to all ordinals as follows:
\[
\alpha + \beta :=
\begin{cases}
\alpha & \textrm{ if } \beta =0 \\
(\alpha + \gamma) + 1 & \textrm{ if } \beta = \gamma + 1 \\
\sup_{\gamma < \beta} (\alpha + \gamma) & \textrm{ if } \beta \textrm{ is limit}
\end{cases}
\]
\[
\alpha \cdot \beta :=
\begin{cases}
0 & \textrm{ if } \beta =0 \\
(\alpha \cdot \gamma) + \alpha & \textrm{ if } \beta = \gamma + 1 \\
\sup_{\gamma < \beta} (\alpha \cdot \gamma) & \textrm{ if } \beta \textrm{ is limit}
\end{cases}
\]
\[
\alpha^\beta :=
\begin{cases}
1 & \textrm{ if } \beta =0 \\ %\textrm{ and } \alpha \ne 0\\
(\alpha^\beta) \cdot \alpha & \textrm{ if } \beta = \gamma + 1 \\
\sup_{\gamma < \beta} (\alpha^\gamma) & \textrm{ if } \beta \textrm{ is limit}
\end{cases}
\]
The reader is cautioned that while \(+\) and \(\cdot\) are associative,
neither \(+\) nor \(\cdot\) are commutative.
For instance:
%
\[
2 \cdot \omega = \sup_{n \in \omega}\ ( 2 \cdot n) = \omega < \omega \cdot 2 =
\sup_{n \in \omega}\ ( \omega + n) = \omega+\omega.
\]
%
Further, ordinal addition is not right cancellative, but is left
cancellative: \(\alpha+\beta = \alpha+\gamma\) implies \(\beta=\gamma\).
%
We adopt the standard binding conventions from ordinary
arithmetic (e.g. \(\alpha \cdot \beta + \gamma = (\alpha \cdot
\beta) + \gamma\)) and associate exponentiation to the right (as one
does in ordinary arithmetic): \(\alpha^{\beta^\gamma} =
\alpha^{(\beta^\gamma)}\), which typically does not coincide with
\((\alpha^\beta)^\gamma = \alpha^{\beta \cdot \gamma}\).
The ordinal \(\epsilon_0\) is the least ordinal solution to \(\omega^x = x\).
It also has the property that if \(\alpha,\beta < \epsilon_0\), then
\(\alpha+\beta\), \(\alpha\cdot \beta\), and \(\alpha^\beta\)
are all less than \(\epsilon_0\).
Further details on
ordinal arithmetic can be found in article II of \cite{MR0045635}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Elementary amenable groups}\label{EAClass:sec}
Consider the smallest class \(EG\) containing the finite and
abelian groups and closed under the following operations:
\begin{enumerate}
\item \label{extension}
taking an extension of one group
by another group;
\item \label{directed_union}
taking a directed union of a set of groups;
\item \label{subgroup}
taking a subgroup of a group;
\item \label{quotient}
taking a quotient of a group by a normal subgroup;
\end{enumerate}
This class of groups was first considered by Day \cite[P. 520]{MR19:1067c} under the name
\emph{elementary groups};
it is more common in the current literature to refer to them as the \emph{elementary amenable groups}.
This class was later studied by Chou \cite{Chou} who worked out much of the basic theory and showed that the operations
of extension and directed union are all that are needed to generate \(EG\).
Chou stratified \(EG\) by subclasses \(EG_\alpha\) with \(\alpha\) from the ordinals by setting:
\begin{itemize}
\item \(EG_0\) to be the class of all abelian and finite groups;
\item \(EG_{\alpha+1}\) to be those groups obtainable from
groups in \(EG_\alpha\) by a single application of operation (\ref{extension}) or
(\ref{directed_union}) above;
\item
\(EG_\alpha :=\bigcup_{\beta<\alpha}EG_\beta\) if \(\alpha\) is a limit ordinal.
\end{itemize}
It is proven in \cite{Chou} that each \(EG_\alpha\) is closed under
taking subgroups and quotients and that
every element of \(EG\) is in \(EG_\alpha\) for some ordinal \(\alpha\).
For \(G\in EG\), it has become customary to define the
\emph{elementary amenability class} \(\EA(G)\) of \(G\) as the smallest
\(\alpha\) for which \(G\in EG_\alpha\). It follows from the
definitions that for every limit \(\alpha\) there is no \(G\in EG\)
with \(\EA(G)=\alpha\), and there is no finitely generated \(G\in EG\)
with \(\EA(G)=\alpha+1\).
From Chou's result that the \(EG_\alpha\) are closed under
taking subgroups and quotients, it follows that for \(G\) and \(H\) in
\(EG\), if \(G\) is either a subgroup or a quotient of \(H\), then
\(\EA(G)\le \EA(H)\).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Wreath products}
\label{wreath:subsec}
Given a group of permutations \(G\) of \(X\) and a group of
permutations \(H\) of \(Y\), then the 1937 article
\cite{polya:wreath} defines \(G\wr H\), the wreath product of \(G\)
and \(H\), as a group of permutations of \(X\times Y\). Since the
1964 paper \cite{neumann:wreath}, the {\itshape standard wreath
product} obtained from \(G\) and \(H\) by regarding \(H\) as
permuting itself by (e.g.) right multiplication has become standard
and ``wreath product'' often means ``standard wreath product'';
``permutation wreath product'' has been used for the older notion.
Our focus is primarily on permutation wreath products in this
article and we proceed with the definition.
Given pairs \((G,X)\) and \((H,Y)\) where \(G\) is a group acting on
\(X\) and \(H\) is a group acting on \(Y\), then we write
\(G\wr H\), the \emph{permutation wreath product}
of \(G\) and \(H\) as shorthand for the pair
\((G\wr H, X\times Y)\) where the group and the action have to be defined.
We regard \(G^Y\), the set of functions from \(Y\) to \(G\), as a
group by multiplying pointwise. With 1 the identity of \(G\) and
for \(\phi\in G^Y\), we use \(\supt(\phi)\) to denote \(\{y\in
Y\mid \phi(y)\ne 1\}\), the \emph{support} of \(\phi\).
We use \(\sum_{Y}G\) to denote the direct sum of copies of \(G\)
indexed over \(Y\) which can be viewed concretely as
the group of finitely supported elements of \(G^Y\).
The group \(H\) also acts on
\(\sum_Y G\) on the
right by \(\phi^h(y) = \phi(yh^{-1})\).
We use this action to form the semidirect product \(\sum_Y G\rtimes H\)
on the set \((\sum_Y G)\times H\) with multiplication
\((\phi,h)(\theta,j) = (\phi \theta^{h^{-1}}, hj)\).
This semidirect product is the \emph{wreath product}
of \(G\) and \(H\) and is denoted \(G\wr H\).
The action of \(G\wr H\) on \(X\times Y\) is given by
\((x,y)(\phi,h) = (x\phi(y), yh)\).
In our setting, wreath products arise as in the next lemma.
A proof is given in the proof of \cite[Proposition 2.5]{BrinEG}.
The following definitions make the lemma easier to state.
Let \(H\) be a group acting on a set \(A\), let \(Y\) be a subset of \(A\),
and let \(\Ycal = \{Yh\mid h\in H\}\);
note that \(H\) also acts on \(\Ycal\).
We say that the action of \(H\) on \(\Ycal\) is
\emph{consistent} to mean that for all \(h\in H\) if
\(Yh\cap Y\ne\emptyset\), then \(h\) fixes \(Y\) pointwise.
We say that the action of \(H\) on \(\Ycal\) is \emph{faithful}
to mean that the only element of \(H\)
that fixes all elements of \(\Ycal\) is the identity of \(H\).
The lemma is now stated as follows (see Figure \ref{WreathDiag}):
\begin{figure}
\[
\xy
(11,11); (106,11)**@{-}; (106,49)**@{-}; (11,49)**@{-};
(11,11)**@{-};
(13,47)*{A};
(14,15); (36,15)**@{-}; (36,40)**@{-}; (14,40)**@{-}; (14,15)**@{-};
(19,37)*{Y\supseteq}; (22,33)*{\supt(G)};
(16,24); (24,24)**@{-}; (24,29)**@{-}; (16,29)**@{-}; (16,24)**@{-};
(20,26)*{X};
(26,24); (34,24)**@{-}; (34,29)**@{-}; (26,29)**@{-}; (26,24)**@{-};
(30,26)*{Xg_1};
(16,16); (24,16)**@{-}; (24,21)**@{-}; (16,21)**@{-}; (16,16)**@{-};
(20,18)*{Xg_2}; (30,18)*{\cdots};
(42,15); (64,15)**@{-}; (64,40)**@{-}; (42,40)**@{-}; (42,15)**@{-};
(47,37)*{Yh_1};
(70,15); (92,15)**@{-}; (92,40)**@{-}; (70,40)**@{-}; (70,15)**@{-};
(75,37)*{Yh_2};
(98,27)*{\cdots};
\endxy
\]
\caption{An illustration of the sets in Lemma \ref{PreWrLem}}\label{WreathDiag}
\end{figure}
\begin{lem}\label{PreWrLem}
Suppose that \(G\) and \(H\) act on a set \(A\) on the right.
Assume there are sets \(X \subset \supt(G) \subset Y \subset A\)
such that the action of \(G\) on
\(\Xcal = \{Xg\mid g\in G\}\) and the action of \(H\) on
\(\Ycal = \{Yh\mid h\in H\}\) are both consistent and faithful.
Then the
action of \(W=\langle G, H\rangle\) on \(\Wcal = \{Xw\mid w\in W\}\)
is also consistent and faithful and is isomorphic to the action of
the permutation wreath product \(G\wr H\) on \(\Xcal \times \Ycal\).
\end{lem}
We now detail how the EA-classes of groups
interact with certain permutation wreath products.
For convenience, we let \(\Sigma G\) denote the direct sum of
countably many copies of the group \(G\).
It is clear that \(\EA(\Sigma G)\ge \EA(G)\).
It is a straightforward inductive exercise to show that for all \(G\),
we have \(\EA(G + G)=\EA(G)\).
It follows that \(\EA(\Sigma G)\le \EA(G)+1\).
In the special case that \(\EA(\Sigma G)=\EA(G)\), we say that \(G\) has
property \(\Sigma\).
Notice, for instance, that every abelian group has \(\Sigma\) and that the groups with \(\Sigma\) are
closed under the elementary operations (\ref{extension})--(\ref{directed_union}).
The family of groups \(\Sfrak\) which we construct all satisfy \(\Sigma\).
The next proposition, mostly proven in \cite{BrinEG}, is very fruitful in
calculating and estimating the EA-classes of the groups we
consider later in the article.
\begin{prop}\label{EAGWrG}
For an infinite, finitely generated group \(G\in EG\)
acting faithfully on an infinite set \(Y\),
we have
\[
\EA(G)+1 \le \EA(G\wr G) \le \EA(G)+2.
\]
Further, if \(\EA(\Sigma G)=\EA(G)\), then
\(\EA(G\wr G) = \EA(G)+1\).
\end{prop}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The oscillation function}
\label{osc:sec}
In this section, we establish some further facts about the oscillation function first introduced in Section \ref{OscSig:sec}.
If \(A\) is a fast set of
\Marked{} functions, then we say that an orbital is \emph{active
with respect to \(A\)} if it contains a transition point of an
element of \(A\). Notice that an orbital \(J\) is active with
respect to \(A\) if and only if it contains a foot of \(A\) other
than those of \(J\).
Recall that if \(f < g\) are marked functions, then \(o(f,g)\) is
the number of orbitals of \(g\) which are active with respect to \(\{f,g\}\).
By convention \(o(f,g)\) is a symmetric function --- \(o(g,f) := o(f,g)\).
If \(f\) is a standard function with at least
one negative bump, we will call the unique common transition point
of a positive bump and negative bump of \(f\) the \emph{expansion
point} of \(f\). The following is a sketch of a standard function
where we have highlighted the expansion point with a bullet:
%
\[
\xy
(-2.3,3.5)*{f};
(15,0.8)*{\bullet};
(7,-0.85)*{\varnbump{4}{12}{16}};
(23,2.6)*{\varbump{4}{12}{16}};
(39,2.6)*{\varbump{4}{12}{16}};
\endxy
\]
The next two lemmas provide a number of useful characterizations of the oscillation function.
We omit the routine proofs.
\begin{lem}\label{Std_Base_Facts} Let \(f\) and \(g\) be standard
functions so that \(\{f,g\}\) is fast with \(f\sqsubset g\). Then
the following hold:
%
\begin{enumerate}
%
\item The pair \((f,g)\) is standard and \(o(f,g)=1\) if and only if
the support of \(f\) is contained in the rightmost orbital of \(g\).
%
\item The pair \((f,g)\) is standard and \(o(f,g)=2\) if and only if
\(g\) has an expansion point, and for the rightmost orbital
\((a,b)\) of \(f\), the endpoint \(a\) is contained in the leftmost
orbital of \(g\) and the endpoint \(b\) is contained in the
rightmost orbital of \(g\).
%
\item If the pair \((f,g)\) is standard and \(o(f,g)>2\), then \(f\)
has an expansion point, and the leftmost orbital of \(g\) and the
rightmost orbital of \(g\) each contain exactly one transition point
of \(f\) and neither contains the expansion point of \(f\).
%
\end{enumerate}
\end{lem}
\begin{lem} \label{o_lem}
If \(f \sqsubset g\) is a fast pair of standard functions, then
the following are true:
\begin{enumerate}
\item\label{o_sym_fact} \(o(f,g)\) is one greater than the number of
active orbitals of \(f\) with respect to \(\{f,g\}\);
\item\label{o_cut_one} if \((f,g)\) is a standard pair, then
\(o(f,g) = o(g\cut,f)+1\);
\item \label{o_cover_char} if \(n\) is the cardinality of the
smallest cover of the feet of \(\{f,g\}\) by disjoint intervals
each of which intersects the feet of at most one of \(f\) or \(g\),
then \(o(f,g) = (n-1)/2\);
\end{enumerate}
\end{lem}
Many arguments about \(\Sgen\) are inductive and take advantage of the recursive
nature of the definition of a standard pair.
Note that if \((f,g)\) is standard and \(o(f,g) >1\), then \(f \sqsubset f^g\).
Also observe that even without an assumption that
\(f\sqsubset g\), we have \(o(f,g) = o(f^p,g^q)\) for any nonzero
\(p,q \in \Z\) and that for every homeomorphism \(h\), \(o(f^h,g^h)
= o(f,g)\). The next lemma is useful in calculating values of
the oscillation function in subsequent sections.
\begin{lem} \label{o_conj_bound}
If \(f\sqsubset h\) and \(g\sqsubset h\) are marked
functions in \(\HomeoI\) with \(\{f,g,h\}\) fast (possibly \(f=g\)), then
%
\[
o(f,g^h) \leq \min\big(o(f,h),o(g,h)\big).
\]
Moreover if \((f,h)\) and \((g,h)\) are standard pairs, then
\(f < g^h\) and
\[
o(f,g^h) \leq \min\big(o(f,h) , o(g,h)-1\big).
\]
\end{lem}
\begin{proof} Observe that since the feet of \(g\) and \(h\) are
disjoint, every foot of \(g^h\) is contained in a foot of \(h\). By
(\ref{o_cover_char}) of Lemma \ref{o_lem}, \(o(f,g^h) \leq o(f,h)\).
Furthermore, we have \(o(f,g^h) = o(f^{h^{-1}},g) \leq o(g,h^{-1}) =
o(g,h)\).
Assume that \((f,h)\) and \((g,h)\) are standard pairs.
Since \(g\sqsubset h\), we have \(o(g,h)\ge1\).
If \(o(g,h)=1\), then Lemma \ref{Std_Base_Facts}
and the assumption that \(\{g,h\}\) is fast implies that
support of \(g^h\) is contained in the rightmost foot of \(h\) which
is to the right of the support of \(f\).
Thus \(f \ll g^h\) and
\(o(f,g^h) = 0 \le \min\big(o(f,h) , o(g,h)-1\big)\).
Now assume that \(o(g,h)>1\).
Let \(J\) be the union of the feet of \(h\)
other than the leftmost and rightmost feet of \(h\).
Notice that since \(o(g,h) > 1\), \(J\) is nonempty.
The feet of \(f^{h^{-1}}\) are contained in \(J\)
which in turn is contained in the support of \(g\) since \(o(g,h)>1\).
Thus \(f^{h^{-1}}\sqsubset g\) and all the feet, and thus all the
transition points, of \(f^{h^{-1}}\) are contained in the orbitals
of \(g\) that are active with respect to \(\{h\cut, g\}\).
By (\ref{o_sym_fact}) of Lemma \ref{o_lem}, the number of these
orbitals is \(o(h \cut,g) = o(g,h)-1\), and so
\(o(f,g^h) = o(f^{h^{-1}},g) \le o(g,h)-1\).
\end{proof}
For the following, recall that if \(A\) is in \(\Sgen\), then
\(A\) is linearly ordered by \(<\). If \(|A|=n\), we let
%
\(
a_0<a_1<\cdots <a_{n-1}
\)
be the elements of \(A\).
We also use \(a_{\max}\) to denote the
greatest element of \(A\).
Recall that if \(A \in \Sgen\),
then we write \(A = B + C\) if \(A = B \cup C\),
and \(b \ll c\) whenever \(b \in B\) and \(c \in C\).
If there are no nonempty \(B\) and \(C\) where
\(A=B+C\), then we say that \(A\) is \emph{indecomposable}. Many
proofs that follow argue the decomposable and indecomposable cases
separately. The next lemma gives a useful characterization of when
an element of \(\Sgen\) is decomposable.
\begin{lem} \label{dir_sum_char}
An \(A \in \Sgen\) is decomposable
if and only if there is an \(i < |A|-1\) such that
\(o(a_i,a_{\max}) = 0\).
\end{lem}
\begin{proof}
We begin by observing that \(b \ll c\) is equivalent to \(b < c\) and \(o(b,c) = 0\).
The forward implication in the lemma follows immediately from this equivalence.
To see the reverse implication, let \(j < |A|-1\) be maximal such that \(o(a_j,a_{\max}) = 0\).
Observe that every element of the support of \(a_j\) is less than every element of the support of \(a_{\max}\).
If \(i < j\), then \(\sup \supt(a_i) < \sup \supt(a_j)\) and consequently \(o(a_i,a_{\max}) = 0\).
We claim that if \(i \leq j < k < |A|-1\), then \(o(a_i,a_k) = 0\).
Since \(o(a_k,a_{\max}) > 0\), it follows that the extended support of \(a_k\) is contained in the extended
support of \(a_{\max}\).
Since \(\sup \supt(a_i) < \inf\supt(a_{\max})\), it follows that \(\sup \supt(a_i) < \inf \supt(a_k)\)
and hence that \(o(a_i,a_k) = 0\).
By the equivalence noted at the start of the proof, we therefore have that
if \(i \leq j < k <|A|\), then \(a_i \ll a_k\).
If we take \(B = \{a_i \mid i \leq j\}\) and \(C = \{a_k \mid j < k < |A|\}\),
then we've shown that \(A = B+C\).
\end{proof}
The following operation is useful in analyzing \(\Sgen\).
If \(A \in \Sgen\), define the \emph{rotation of \(A\)} to be
\(A\cut := A \cup \{a_{\max} \cut\} \setminus \{a_{\max}\}\) if there is an \(i < |A|-1\) such
that \(o(a_i,a_{\max}) > 0\);
otherwise set \(A \cut := A \setminus \{a_{\max}\}\).
Notice that it follows immediately from the definition of standard pair that \(A\cut\) is again in \(\Sgen\).
\begin{lem} \label{rotation_lem}
If \(A \in \Sgen\), then the following are true:
\begin{enumerate}
\item \label{sum_cut}
if \(A = B + C\) and \(C \ne \emptyset\), then \(A \cut = B + (C \cut)\).
\item \label{indecomp_cut}
If \(A\) is indecomposable,
then the least element of \(A\cut\) is \(a_{\max} \cut\).
\end{enumerate}
\end{lem}
\begin{proof}
To see (\ref{sum_cut}), observe first that \(a_{\max} = c_{\max}\).
It remains to show that for all \(i < |B|\), \(b_i < c_{\max} \cut\).
Because \(A = B + C\), we have that for all \(i < |B|\), \(b_i \ll c_{\max}\).
Since the support of \(c_{\max} \cut\) is contained in the support of \(c_{\max}\),
it follows that \(b_i \ll c_{\max} \cut\) for every \(i < |B|\).
To see (\ref{indecomp_cut}), observe that
\(o(a_0,a_{\max}) > 0\) by Lemma \ref{dir_sum_char}.
Thus \((a_{\max}\cut,a_0)\) is standard and hence \(a_{\max}\cut < a_0\).
\end{proof}
The following lemma gives a useful criteria for membership to \(\Sgen\).
\begin{lem} \label{Sgen_char}
If \(A\) is a fast set of standard functions which is totally ordered by \(<\),
then \(A\) is in \(\Sgen\) provided that the following conditions are satisfied:
\begin{itemize}
\item \(A\cut\) is in \(\Sgen\);
\item for all \(i < |A|-1\) if \(a_i \sqsubset a_{\max}\), then \(a_{\max}\cut < a_i\).
\end{itemize}
\end{lem}
\begin{proof} Let \(A \in \Sgen\) be given as in the statement of
the lemma. If \(|A| \leq 1\) there is nothing to show, so assume
that \(|A| > 1\). Observe that \(A \setminus \{a_{\max}\} \subseteq
A\cut\) is in \(\Sgen\) by assumption and therefore in order to
verify \(A \in \Sgen\), we need only to show that \((a_i,a_{\max})\)
is standard whenever \(i < |A|-1\). Let \(j < |A|-1\) be minimal
such that \(o(a_j,a_{\max}) > 0\). Since the support of \(a_{\max}
\cut\) is contained in the support of \(a_{\max}\), \(a_i \ll
a_{\max}\) whenever \(i < j\). If \(j \leq i\), then \(a_{\max}
\cut < a_i\) by hypothesis. Furthermore, \((a_{\max} \cut,a_i)\) is standard since
\(A\cut\) is assumed to be in \(\Sgen\).
Since \(o(a_i, a_{\max})>0\) and thus
\(a_i \sqsubset a_{\max}\), it follows that \((a_i,a_{\max})\) is
standard as desired. \end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Signatures}
\label{signature:sec}
In this section we expand on the notion of signature as defined in
Section \ref{OscSig:sec} and prove several properties.
We show that the signature \(\As\) of an \(A \in \Sgen\) completely encodes
the marked isomorphism type of \(\gen{A}\). Moreover, we give
a simple description of the family of all signatures.
Before
proving the main results, it will be helpful to define some
terminology and prove some lemmas.
If \(A \in \Sgen\), we
refer to pair \((|A|,\sum_{i < j} o(a_i,a_j))\) as the
\emph{complexity} of \(A\); the set of all complexities is ordered
lexicographically. Observe that if \(A \in \Sgen\),
then \(A \cut\) has strictly smaller complexity than \(A\). In what
follows, it is frequently useful to prove statements about the
elements of \(\Scal\) by induction on their complexity.
Theorem \ref{excision} allows us to remove extraneous bumps without changing the
isomorphism type of the group which is generated.
However if \(A\in \Sgen\) has extraneous bumps,
Theorem \ref{excision} does not ensure the modified generating set remains in \(\Sgen\).
The next two lemmas address this.
\begin{lem} \label{find_extraneous}
If \(A \in \Sgen\) has an
extraneous bump, then there is a nonempty set \(E\) of extraneous
bumps of \(A\) such that \(A/E\) has the same dynamical diagram as a
member of \(\Sgen\).
\end{lem}
\begin{proof} This is proved by induction on the complexity of
\(A\). First suppose that \(A = \{f\}\). Recall that \(f\) has at least 2 bumps, and has
either the same number of positive and negative bumps or one more
positive bump. If \(f\) has more positive bumps than negative
bumps, then let \(b\) be the rightmost bump of \(f\) and observe
that \(f/\{b\}\) is still a standard function. Similarly, if \(f\)
has the same number of positive an negative bumps and \(b\) is the
leftmost bump of \(f\), then \(f/\{b\}\) is a standard function. In
both cases \(A/\{b\}\) is in \(\Sgen\). If \(A = B + C\) for \(B,C
\ne \emptyset\), then either \(B\) or \(C\) has an extraneous bump
and we are finished by our induction hypothesis.
Suppose that \(|A| > 1\) is indecomposable with an extraneous
bump. If \(o(a_i, a_{\max})=1\) for all \(i<|A|-1\), then the
support of every \(a_i<a_{\max}\) is in the rightmost bump of
\(a_{\max}\). If the number of bumps of \(a_{\max}\) is at least 2,
we let \(E\) consist of all bumps of \(a_{\max}\) but the rightmost.
If \(a_{\max}\) has only one bump, then there is a bump extraneous
in \(A'=A\setminus \{a_{\max}\}\).
By induction there is an \(E\) consisting of extraneous bumps of \(A'\) with such that \(A'/E\) has the same dynamical diagram as
an element of \(\Sgen\).
Notice that that every element of \(E\) is also extraneous in \(A\) and
\(A/E=(A'/E)\cup \{a_{\max}\}\) has the same dynamical diagram as an element of
\(\Sgen\).
If there is an \(i<|A|-1\) with \(o(a_i, a_{\max})>1\), then both of
the outer orbitals of \(a_{\max}\) are active. Thus \(A \cut\) has
an extraneous bump and we can apply our induction hypothesis to find
a nonempty set of extraneous bumps \(E\) of \(A \cut\) such that \(A \cut / E\) has the same
dynamical diagram as a member of \(\Sgen\).
Observe that every element of \(E\) is also a
bump of \(A\): the only situation in which a bump of \(A\cut\) is
not a bump of \(A\) occurs when \(a_{\max}\) has at most two bumps
and in this case \(a_{\max} \cut\) is isolated and hence not extraneous.
Finally, it is easily checked that \(A/E\) has the same
dynamical diagram as a member of \(\Sgen\).
\end{proof}
\begin{lem} \label{excise_inactive}
If \(A\) is an element of \(\Sgen\), then there is an \(A' \in
\Sgen\) such that:
\begin{enumerate}
\item \(\gen{A'}\) is marked isomorphic to \(\gen{A}\);
\item \(A'\) and \(A\) have the same signature;
\item \(A'\) has no extraneous bumps.
\end{enumerate}
\end{lem}
\begin{proof}
Observe that extraneous bumps are not counted by signatures.
In particular, if \(E\) is a set of extraneous bumps of \(A \in \Sgen\),
then the signature of \(A/E\) coincides with the signature of \(A\).
The proof of the lemma is now by induction on the number of extraneous
bumps of \(A\), using Lemma \ref{find_extraneous} and Theorems
\ref{comb_to_iso} and \ref{excision}.
\end{proof}
\begin{lem} \label{no_extraneous_pres}
If \(A \in \Sgen\) and no bump of \(A\) is extraneous, then
no bump of \(A\cut\) is extraneous.
\end{lem}
\begin{proof}
First suppose that \(a_{\max}\) has at most two orbitals.
Notice that since \(a_{\max}\cut\) has only one orbital, its only bump is not extraneous.
If \(a_{\max}\) has one orbital, none of its transition points are in an orbital of another element of \(A\)
and hence \(a_{\max}\) plays no role in witnessing that other elements of \(A\) are not extraneous.
If \(a_{\max}\) has two orbitals, then the only transition point which it has to witness that
another element of \(A\) is not extraneous is its middle transition point, which remains a transition
point of \(a_{\max}\cut\).
This completes the proof under the assumption that \(a_{\max}\) has at most two orbitals.
Now suppose that \(a_{\max}\) more than two orbitals.
If \(b\) is an orbital of \(A\cut\), it is an orbital of \(A\) and
is not extraneous in \(A\). So \(b\) contains a transition point
\(t\) of some \(a_i\in A\). The only transition points of \(A\)
that are not transition points of \(A\cut\) are the rightmost and
leftmost transition points of \(a_{\max}\). But these are in no
orbital of \(A\) and so \(t\) is neither of these points. So \(t\)
is a transition point of \(A\cut\) and \(b\) is not extraneous in
\(A\cut\). \end{proof}
We now are ready to prove Theorem \ref{sig_thm} which asserts that
if \(A\) and \(B\) are elements of \(\Sgen\) which have the same signature, then
the order preserving bijection between \(A\) and \(B\) extends to an isomorphism \(\gen{A} \cong \gen{B}\).
\begin{proof}[Proof of Theorem \ref{sig_thm}]
The proof is by induction on the complexity of the common signature of \(A\) and \(B\);
let \(n\) denote \(|A| = |B|\).
By Lemma \ref{excise_inactive}, we may assume that \(A\) and \(B\) have no extraneous bumps.
By Theorem \ref{comb_to_iso} it suffices to show that \(A\) and \(B\) have isomorphic dynamical diagrams.
If \(A = A' + A''\) for some nonempty \(A'\) and \(A''\),
then \(B = B' + B''\) for some \(B'\) and \(B''\) having the same signatures as \(A'\) and \(A''\) respectively.
Notice that every orbital of \(A'\) is active (in \(A'\)) and similarly for \(A''\), \(B'\), and \(B''\).
Thus we can apply our induction hypothesis to conclude that
the dynamical diagrams of \(A'\) and \(B'\) are isomorphic and similarly for
\(A''\) and \(B''\).
Since the dynamical diagram of \(A\) is obtained by putting the diagram for \(A'\) to the left of the diagram for \(A''\)
--- and similarly for \(B\) --- we have that the dynamical diagrams of \(A\) and \(B\) are isomorphic.
Now suppose that neither \(A\) nor \(B\) decompose as a sum. By
Lemma \ref{dir_sum_char}, this means that \(o(a_i,a_{\max}) =
o(b_i,b_{\max}) > 0\) for all \(i < n-1\). If \(a_{\max}\) has a
single orbital, then \(a_{\max}\) is a positive bump, and
\(o(b_i,b_{\max}) = o(a_i,a_{\max}) = 1\) for all \(i < n-1\).
Notice that the definition of standard pair implies that if \(i <
n-1\), then the support of \(b_i\) is contained in the rightmost
orbital of \(b_{\max}\). Since no bump of \(b_{\max}\) is
extraneous, this must mean that \(b_{\max}\) has only one orbital
and must be a positive bump.
By our induction hypothesis, \(A \setminus \{a_{\max}\}\) and
\(B \setminus \{b_{\max}\}\) have isomorphic dynamical diagrams; let
\(D\) denote the common isomorphism type. Notice that the dynamical
diagram for \(A\) and for \(B\) are both obtained by a pair of new
vertices to \(D\) --- one to the far left and one to the far right
--- as well as a right directed edge between these new vertices.
This new edge is given a label distinct from the other labels.
Hence \(A\) and \(B\) have isomorphic dynamical diagrams.
Finally, we may now assume that both \(a_{\max}\) and \(b_{\max}\) have more than one orbital.
Observe that \(A\cut\) and \(B\cut\) have the same signature and lower complexity than \(A\) and \(B\):
if \(i < n-1\), then
\[
o(a_{\max} \cut,a_i) = o(a_i,a_{\max})-1 = o(b_i,b_{\max})-1 = o(b_{\max}\cut,b_i).
\]
By Lemma \ref{no_extraneous_pres}, \(A\cut\) and \(B\cut\) have no
extraneous bumps. By our induction hypothesis, \(A\cut\) and
\(B\cut\) have dynamical diagrams which are isomorphic to some common
\(D\). Notice that since every orbital of \(a_{\max}\) is active in
\(A\), \((a_{\max}) \cut\) is an isolated bump in \(A\cut\) if and
only if \(a_{\max}\) has exactly two orbitals. Since the former
equivalent condition is a property of the dynamical diagram of
\(A\), it follows that \(a_{\max}\) has exactly two orbitals if and
only if \(b_{\max}\) has exactly two orbitals. It is now easily
checked that in both cases, \(A\) and \(B\) must have isomorphic
dynamical diagrams. \end{proof}
We now turn to our characterization of the set of signatures. We start with the following proposition.
The function \(\varrho\) defined in its proof models the effects on oscillation numbers when
an indecomposable \(A \in \Sgen\) is replaced by \(A\cut\).
\begin{prop} \label{!_equiv}
The following are equivalent for an ordered triple \((p,q,r)\) of integers:
\begin{enumerate}
\item \label{rFocPoint}
\(r \geq \min (p-1,q)\), with equality holding unless \(p = q\);
\item \label{qFocPoint}
\(q \geq \min (p,r)\), with equality holding unless \(p = r+1\);
\item \label{pFocPoint}
\(p \geq \min (q,r+1)\), with equality holding unless \(q = r\);
\item \label{pqrConjunction} all of the following three inequalities
hold:
%
\[
p \geq \min (q,r+1), \quad
q \geq \min (p,r), \quad
r \geq \min (p-1,q).
\]
\end{enumerate}
\end{prop}
\begin{proof}
We first show that (\ref{rFocPoint}) implies (\ref{qFocPoint})
and then argue that (\ref{qFocPoint}) and (\ref{pFocPoint})
are equivalent to (\ref{rFocPoint}) by symmetry.
Let us assume that \(p,q,r\in\Z\) with \(r \geq \min (p-1,q)\) and with equality holding unless \(p = q\).
If \(p>q\) then we have \(r=q\) so that \(p>q=r\).
Therefore \(q=\min(p,r)\) and (\ref{qFocPoint}) holds.
If \(p<q\) then we have \(r=p-1<p<q\) so in particular \(p=r+1\) and \(q>\min(p,r)=r\) and again (\ref{qFocPoint}) holds.
Finally, if \(p=q\) then \(r\geq p-1=q-1\) by assumption.
We therefore have that \(\min(p,r) \leq p = q\) and again (\ref{qFocPoint}) holds.
Now consider the transformation \(\varrho:\Z^3 \to \Z^3\) defined by \(\varrho(p,q,r) = (r,p-1,q-1)\).
Observe that \((p,q,r)\) satisfies (\ref{rFocPoint}) if and only if \((\bar p, \bar q, \bar r):= \varrho(p,q,r)\) satisfies
(\ref{pFocPoint}):
the assertion
``\(r \geq \min (p-1,q)\), with equality holding unless \(p = q\)''
is the same as ``\(\bar p \geq \min (\bar q,\bar r + 1)\), with equality holding unless \(\bar q = \bar r\).''
Similarly, \((p,q,r)\) satisfies (\ref{qFocPoint}) if and only if \(\varrho(p,q,r)\) satisfies (\ref{rFocPoint}).
Similarly, \((p,q,r)\) satisfies (\ref{pFocPoint}) if and only if \(\varrho(p,q,r)\) satisfies (\ref{qFocPoint}).
It follows that (\ref{rFocPoint}), (\ref{qFocPoint}), and (\ref{pFocPoint}) are equivalent.
Next observe that the equivalence of (\ref{rFocPoint}), (\ref{qFocPoint}), and (\ref{pFocPoint}) immediately yields that
each implies (\ref{pqrConjunction}).
%
Lastly, we assume (\ref{pqrConjunction}) and show that
(\ref{rFocPoint}) holds.
If \(p=q\), then (\ref{rFocPoint}) just asserts \(r \geq \min (p-1,q)\) and there is nothing to show.
If \(p < q\), then since \(p \geq \min (q,r+1)\), it must be that \(q > r+1\) and hence
\(p \geq r+1\).
Since \(r \geq \min (p-1,q) = p-1\) we have \(r = p-1 = \min(p-1,q)\).
Similarly if \(q < p\), then \(q \geq \min (p,r)\) implies that \(q \geq r\).
Taken with \(r \geq \min (p-1,q)\), this implies
\(r = q = \min(p-1,q)\).
\end{proof}
When \((p,q,r) \in \Z^3\) satisfies any of the equivalent
relationships stated in the proposition, we write \(r = p !
q\). Using Proposition \ref{!_equiv} and \(\min(a-1,
b-1)=\min(a,b)-1\), we get the following.
\begin{cor}\label{!_equivCor} For all \((p,q,r)\in \Z^3\) we have
%
\[
r = p ! q \Longleftrightarrow
q-1 = r ! (p-1) \Longleftrightarrow
p-1 = q ! r.
\]
%
\end{cor}
If \(\As\) is a function from the pairs \(a<b\) of a finite linear
order \(A\) into \(\omega\), then we say that \(\As\) satisfies (!)
if whenever \(a < b < c\) are in \(A\), the relation \(\As(a,b) =
\As(b,c) ! \As(a,c)\) holds.
Recall that
each \(\As\) is equivalent, under the equivalence defined in Section
\ref{OscSig:sec}, to a map on pairs in which the base is \(\{0,\ldots,n-1\}\).
Define \(\Ssig\) to be the collection of all equivalence classes of
such maps satisfying (!). (Formally we view \(\Ssig\) as a set by
using the choice of canonical representatives from each equivalence
class noted above.) We often write that a function on pairs is
in \(\Ssig\) when we really mean that its equivalence class is in
\(\Ssig\).
Our next task is to prove that that \(\Ssig\) is exactly the set of
signatures of elements of \(\Sgen\), up to the above notion of equivalence.
Recall that \(\zero\) and \(\one\) denote the
elements of \(\Ssig\) with bases having cardinalities \(0\) and \(1\)
respectively.
Anticipating \(\Ssig\)'s relation to \(\Sgen\), we will often confuse the
distinction between an \(\As \in \Ssig\) and its base, writing things such as
``the cardinality of \(\As\)'' or ``the elements of \(\As\)'' when we are really referring
to the cardinality or elements of the base of \(\As\).
The operation \(+\) defined in Section \ref{BasicOp:sec} is well
defined on \(\Ssig\). Unlike with \(\Sgen\), \(+\) is a \emph{bona fide}
binary operation on \(\Ssig\). Observe that if \(A = B + C\), then
\(\As = \Bs + \Cs\) --- the signature of a sum is the sum of the
signatures.
The following application of (!) will be used often.
\begin{lem}\label{ZeroToEnd}
Suppose \(\As\in \Ssig\).
If \(i<j<k < |A|\) and \(\As(j,k)=0\), then
\(\As(i,k)=0\).
\end{lem}
\begin{proof}
If \(\As(i,k)\ne0\), then
%
\[
\As(i,j) = \min(\As(j,k)-1, \As(i,j)) = -1
\]
which is not possible.
\end{proof}
An element \(\As\) of \(\Ssig\) is \emph{indecomposable} if there do
not exist \(\Bs,\Cs \ne \zero\) such
that \(\As = \Bs + \Cs\).
We need the following analog of Lemma \ref{dir_sum_char}.
\begin{lem} \label{top_positive}
If \(\As \in \Ssig\) is indecomposable and \(i < n = |\As|-1\),
then \(\As(i,n) > 0\).
\end{lem}
\begin{proof}
We prove the contrapositive.
Let \(j\) be maximal such that \(\As(j,n) = 0\). By Lemma
\ref{ZeroToEnd}, \(\As(i,n)=0\) for all \(i<j\).
If \(i<j<k\le n\), we have \(\As(i,n)=0<\As(k,n)\) so
%
\[
\As(i,k) = \min(\As(j,k)-1, \As(i,n)) = 0.
\]
Thus setting \(\Bs := \{0,\ldots,j\}\)
and \(\Cs := \{j+1,\ldots,n\}\) we have \(\As=\Bs+\Cs\).
\end{proof}
We now define the analog of the rotation operation on \(\Ssig\).
Just as in the case of \(\Sgen\), we define the \emph{complexity} of an element \(\As \in \Ssig\)
to be the pair \((|\As|,\sum_{i < j} \As (i,j))\).
Set \(\zero \cut = \zero\) and \(\one \cut = \zero\).
If \(\As = \Bs + \Cs\) for \(\Bs,\Cs \ne \zero\) and \(\Cs\cut\) has been defined,
then we set \(\As \cut :=\Bs + (\Cs \cut)\).
If \(\As \in \Ssig\) is indecomposable and has
\(\{0,\ldots,n\}\) as its base for some \(n > 0\),
define \(n \cut := -1\) and \(A \cut := \{n\cut,0,\ldots,n-1\} = \{-1,0,\ldots,n-1\}\).
For \(n \cut < i < n\), set \(\As \cut(n\cut,i) = \As(i,n)-1\);
if \(n\cut < i < j < n\), set \(\As\cut(i,j) = \As (i,j)\).
By Lemma \ref{top_positive},
the values taken by \(\As \cut\) are nonnegative.
In order to verify that \(\As \cut\) is in \(\Ssig\), it suffices to show that
if \(n \cut < i < j < n\), then
\[
\As \cut(n\cut,i) \geq \min \big(\As \cut(i,j)-1, \As \cut(n\cut,j)\big)
\]
with equality holding unless
\(\As\cut(i,j) = \As\cut(n\cut,j)\).
But this is equivalent to
\(\As(i,n) \geq \min \big(\As(i,j),\As(j,n)\big)\) with equality unless
\(\As(i,j) = \As(j,n)-1\).
Since this is true by the equivalence of (\ref{rFocPoint}) and
(\ref{qFocPoint}) in Proposition \ref{!_equiv}, we have
that \(\As\cut\) is in \(\Ssig\).
Observe that if \(\As \ne \zero\), then the complexity of \(\As \cut\)
is strictly less than that of \(\As\).
Also notice that the map \(\As \mapsto \As\cut\) is one-to-one on the indecomposable
elements of \(\Ssig\).
\begin{thm}\label{sig_exclam}
\(\Ssig\) is the set of signatures of elements of \(\Sgen\).
Moreover, if \(A \in \Sgen\), then
the signature of the rotation of \(A\) is the rotation of the signature of \(A\).
\end{thm}
\begin{proof}
First we prove that every signature of an element \(A\) of \(\Sgen\) satisfies (!) and
hence is in \(\Ssig\).
The proof is by induction on the complexity of \(A\).
If \(|A| \leq 2\), there is nothing to show, so suppose that \(|A| \geq 3\).
Also, if \(A = B+C\) for nonempty \(B,C \in \Sgen\), then \(\Bs\) and \(\Cs\) are in \(\Ssig\)
by our induction hypothesis and since \(\As = \Bs + \Cs\), \(\As \in \Ssig\) by the closure
of \(\Ssig\) under sums.
Now suppose that \(A\) is indecomposable.
Then \(A\cut\) is in \(\Sgen\) and by Lemma \ref{rotation_lem},
\(a_{\max} \cut < a_i\) for all \(i < |A|\).
Since it has lower complexity than \(A\),
\(A\cut\) is subject to the induction hypothesis and hence its signature
is in \(\Ssig\).
It suffices to show that if \(i < j< k = |A|-1\),
then \(o(a_i,a_j) = o(a_j,a_k) ! o(a_i,a_k)\).
By our inductive assumption we know that
\[
o(a_k \cut,a_i) = o(a_i,a_j) ! o(a_k \cut, a_j)
\]
and thus
\[
o(a_i,a_k)-1 = o(a_i,a_j) ! (o(a_j , a_k) -1)
\]
By Corollary \ref{!_equivCor}, this is equivalent to \(o(a_i,a_j) =
o(a_j,a_k) ! o(a_i,a_k)\).
This completes the proof that signatures of elements of \(\Sgen\) are in \(\Ssig\).
Suppose now that \(\As\) is in \(\Ssig\); we need to prove that there is an \(A \in \Sgen\) whose signature is \(\As\).
This is proved by induction on the complexity of \(\As\).
We may assume \(\As\) has base \(\{0,\ldots,n\}\).
If \(n = 0\), there is nothing to show.
Also, if \(\As = \Bs + \Cs\), then let \(B\) and \(C\) be elements of \(\Sgen\) which have
\(\Bs\) and \(\Cs\) as their respective signatures.
By scaling and translating if necessary, we may assume that the elements of \(B\)
are supported on \((0,1/2)\) and the elements of \(C\) are supported on \((1/2,1)\).
It is now easy to check that every pair from \(A = B + C = B \cup C\) is standard
and thus \(A\) is in \(\Sgen\) and has \(\As = \Bs + \Cs\) as its signature.
Now suppose that \(n > 0\) and \(\As\) is indecomposable.
By Lemma \ref{top_positive}, \(\As(i,n) > 0\) for all \(i < n\).
Since the complexity of \(\As\cut\) is less than that of \(\As\),
our induction hypothesis implies that \(\As \cut\) is the signature of some ordered
sequence \(a_{n\cut},a_0, \ldots, a_{n-1}\) comprising an element of \(\Sgen\).
Without loss of generality, we may assume that the supports of each of these functions
is contained in \((1/3,2/3)\) and that moreover the greatest and least transition points of \(a_{n \cut}\) are
not in the closures of the feet of the \(a_i\)'s.
Let \(a_n\) be a standard function with extended support \((0,1)\)
such that \(a_n \cut = a_{n\cut}\) and whose feet are disjoint from those of \(a_i\) for all \(i < n\).
It follows that \((a_i,a_n)\) is standard for all \(i < n\).
Since the signature of \(A\) is \(\As\), we are finished with the proof of the first conclusion of the theorem.
That the signature and rotation maps commute follows from their definitions and
Lemmas \ref{o_lem} and \ref{rotation_lem}.
\end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The inflation operation}
\label{inflation:sec}
In this section we introduce a fundamental operation on \(\Sgen\) and
establish how it influences signatures.
This operation plays a central role in subsequent sections.
Let us begin with the observation that if \(A \in \Sgen\),
then
\[
N:=\gen{ (a_i)^{a_{\max}^p} \mid i < |A|-1 \textrm{ and } p \in \Z}
\]
is a normal subgroup of \(\gen{A}\), and \(\gen{A}\) is an extension of
\(N\) by \(\Z\).
If we define, for each \(k\),
\[
A_k := \{ (a_i)^{a_{\max}^p} \mid i < |A|-1 \textrm{ and } |p| \leq k \},
\]
then \(A_k\) is fast and \(N=\bigcup \gen{A_k}\). Each \(A_k\) has the
same dynamical diagram as
\[
B_k:=\{ (a_i)^{a_{\max}^p} \mid i < |A|-1 \textrm{ and } 0 \leq p
\leq 2k \}.
\]
The need to understand the relationships between the groups
\(\gen{A}\), \(N\), and the \(\gen{A_k}\) motivates a family of primitive
transformations which we term \emph{inflations}.
If \(A\) is in \(\Sgen\) and \(a \in A\), then the \emph{inflation} of
\(A\) by \(a\) is the set
\[
\infl{A}{a} := A \cup \{a^2\} \cup \{b^a \mid b \in A \textrm{ and }
b < a\} \setminus \{a\}.
\]
Observe if \(a = a_{\max}\), then by iterating this procedure we
obtain supersets of the \(B_k\).
Note that clearly \(\gen{\infl{A}{a}} \subseteq \gen{A}\) and since \(A\) is fast,
\(\gen{A}\) embeds into \(\gen{\infl{A}{a}}\);
see the end of the proof of Lemma \ref{inflate_biembedd} for details.
For \(A\), \(B\) in \(\Sgen\), we write \(A \leq B\)
if there is a sequence \((B_i \mid i \leq n)\) such that
\(B_0 = B\), \(B_n\) has the same dynamical diagram as \(A\), and such
that if \(i < n\), then \(B_{i+1}\) is contained in an inflation of \(B_i\).
Allowing \(n=0\) makes \(\le\) reflexive and \(\le\) is clearly transitive.
Define an equivalence relation \(\equiv\) on \(\Sgen\) by
\(A \equiv B\) if and only if \(A \le B \le A\).
A major aim of the rest of the paper is to show
that the relation \(\le\) coincides with
the embeddability relation on \(\Sgen\) and
that moreover \((\Sgen/\!\! \equiv\ ,\leq)\) is a well-order with order type \(\epsilon_0\).
Now we assign a \Marking{} to \(\infl{A}{a}\). The functions in
\(\infl{A}{a} \cap A = A \setminus \{a\}\) maintain their \Markings{}.
The \Markers{} of \(b^a\) are of the form \(s a\) where \(s\) is a
\Marker{} of \(b\). Finally, if \(s\) is a \Marker{} of a positive bump
of \(a\), then \(s\) is a \Marker{} of \(a^2\); if \(s\) is a \Marker{} of
a negative bump of \(a\), then \(sa\) is a \Marker{} of \(a^2\). This
\Marking{} has the property that if \(t\) is in the support of \(a\)
but not in one of its feet, then \(ta\) is not in a foot of \(a^2\).
In particular, \(\infl{A}{a}\) is fast. Notice that if \(A = B+C\),
then \(\infl{A}{b} = \infl{B}{b} + C\) for all \(b \in B\), and
\(\infl{A}{c} = B+ \infl{C}{c}\) for all \(c \in C\).
\begin{lem} \label{inflate_biembedd}
If \(A\) is in \(\Sgen\) and \(a \in A\), then \(\infl{A}{a}\) is in \(\Sgen\) and
\(\gen{\infl{A}{a}}\) is biembeddable with \(\gen{A}\).
\end{lem}
\begin{proof}
The proof of that \(\infl{A}{a}\) is in \(\Sgen\) is
by induction on the complexity of \(A\). There is nothing to show
if \(|A|\le1\). We have noted that if \(A \in \Sgen\) and \(a \in
A\), then \(\infl{A}{a}\) is fast. Furthermore, if \(b \in A\),
then \(\{b^a,a^2\} = \{b^a,(a^2)^a\}\) is a standard pair. Also, by
the previous observation we may assume that \(A\) is indecomposable.
If \(a < a_{\max}\), then observe that \((\infl{A}{a}) \cut \subset
\infl{(A\cut)}{a}\). By our inductive hypothesis, \(\infl{(A
\cut)}{a}\) is in \(\Sgen\) and thus \((\infl{A}{a}) \cut\) is in
\(\Sgen\). From the definition of standard pairs and the
indecomposability of \(A\), we have that \((a_{\max}\cut, b)\) is
standard for all \(b\in A\setminus\{a_{\max}\}\). We know \(j:=o(b,
a_{\max})>0\). If \(j=1\), then \(\supt(b)\) is contained in the
rightmost orbital of \(a_{\max}\) which implies that \(\supt(b^a)\)
is contained in that same orbital. Thus \(a_{\max}\cut\ll b^a\).
If \(j>1\), then the extreme orbitals of \(a_{\max}\) are active for
both \(\{a_{\max}, b\}\) and \(\{a_{\max}, b^a\}\), so
\(a_{\max}\cut \sqsubset b^a\).
By Lemma \ref{Sgen_char}, \(\infl{A}{a}\) is in \(\Sgen\).
Now suppose that \(a = a_{\max}\). Observe that it suffices to
assume that \(A = \{a,b,c\}\) with \(b<a\) and \(c < a\). Since
\(A\) is indecomposable, Lemma \ref{o_conj_bound} gives that
\(c<b^a\) and \(b<c^a\). Thus we need only to verify that that
\((c,b^a)\) and \((b,c^a)\) are standard pairs.
Assume first that \(o(c,a)\ge3\) and \(o(b,a)\ge3\), and consider
pairs \((c, b^a)\) and \((b, c^a)\).
By two applications of Lemma
\ref{o_lem}, we have \(o(c\cut, a\cut)\ge1\) and \(o(b\cut,
a\cut)\ge1\). Thus \(c\cut \sqsubset a\cut\) and \(b\cut\sqsubset
a\cut\) from which \((b\cut)^{a\cut} = (b\cut)^a = (b^a)\cut\) and
\((c\cut)^{a\cut} = (c\cut)^a = (c^a)\cut\) follows. By inductive
assumption, \((c\cut, (b\cut)^{a\cut}) = (c\cut, (b^a)\cut)\) is a
standard pair. From the definition, \((c, b^a)\) is standard.
Similarly, \((b, c^a)\) is standard as well.
Next suppose that either \(o(c,a)\le2\) or \(o(b,a)\le2\). By the
symmetry of \(b\) and \(c\), we can assume \(o(b,a)\le o(c,a)\).
Since we assume \(A\) is indecomposable, we have \(1\le o(b,a)\le
2\).
We first assume \(o(b,a)=2\).
From Lemma \ref{Std_Base_Facts}, we
have that the extreme orbitals of \(a\) are active in both
\(\{a,c\}\) and \(\{a,b\}\). Further the extreme orbitals of \(a\)
contain all the transition points of \(b\) with only one transition
point of \(b\) in the rightmost orbital of \(a\). This puts the
support of \(c\) in the rightmost orbital of \(b^a\). From Lemma
\ref{Std_Base_Facts}, this makes \((c, b^a)\) standard. If
\(o(c,a)=2\), a similar argument makes \((b,c^a)\) standard. If
\(o(c,a)\ge3\), then \(o(a\cut,c)\ge 2\) and
it follows from Lemma \ref{Std_Base_Facts} that
the extreme orbitals of \(c\) each contain exactly one transition
point of \(a\). This puts all of the transition points of
\(b^{a^{-1}}\) into the extreme orbitals of \(c\) with only one
transition point of \(b^{a^{-1}}\) in the rightmost orbital of
\(c\). From Lemma \ref{Std_Base_Facts}, this implies \((b^{a^{-1}},
c)\) standard, and thus \((b,c^a)\) is also standard.
Now assume \(o(b,a)=1\). From Lemma \ref{Std_Base_Facts}, this puts
the support of \(b\) in the rightmost orbital of \(a\). If
\(o(c,a)=1\), then \(c\ll b^a\) and \(b\ll c^a\) making both
\((c,b^a)\) and \((b,c^a)\) standard. If \(o(c,a)\ge2\), then from
Lemma \ref{Std_Base_Facts}, the only transition point of \(c\) in
rightmost orbital of \(a\) is the rightmost transition point of
\(c\). We still have \(c\ll b^a\) making \((c, b^a)\) standard.
But now the support of \(b^{a^{-1}}\) is in the rightmost orbital of
\(c\) which makes \((b^{a^{-1}}, c)\) standard by Lemma
\ref{Std_Base_Facts}.
This completes the proof that \(\Sgen\) is closed under inflation.
To see that \(\gen{\infl{A}{a}}\) is embeddable in \(\gen{A}\), first note that
\(\gen{\infl{A}{a}} \subset \gen{A}\).
Also, since \(A':=A\setminus \{a\} \cup \{a^2\}\) has the same dynamical diagram as
\(A\), \(\gen{A'} \subset \gen{\infl{A}{a}}\) is isomorphic to \(\gen{A}\) by Theorem \ref{comb_to_iso}.
\end{proof}
We can also define an operation of inflation and a relation \(\le\)
on \(\Ssig\) that corresponds to inflation and \(\le\) on \(\Sgen\).
If \(\As\) is in \(\Ssig\) and \(m \in \As\), then the
\emph{inflation} of \(\As\) by \(m\), denoted \(\infl{\As}{m}\)
has as its base (where \(i^m\) is a formal symbol)
%
\begin{equation}\label{InflUnderlying}
A \cup \{i^m \mid (i \in A) \mand ( i < m) \mand (\As(i,m) > 0)\}
\end{equation}
to which the linear order on \(A\) is extended by declaring
\(j < i^m < j^m < m \leq k\) whenever \(i < j < m \leq k < |A|\).
The function \(\infl{\As}{m}\) extends that of \(\As\) by defining
for each \(i,j < m \leq k < |\As|\):
\begin{align*}
\infl{\As}{m}(i^m,j^m) & := \As(i,j) \\
\infl{\As}{m}(i,j^m) & := \min\big(\As(j,m)-1,\As(i,m)\big) \\
\infl{\As}{m}(i^m,k) & := \min\big(\As(i,m),\As(m,k)\big).
\end{align*}
Here we adopt the notational convention that \(\As(m,m) = \infty\).
From the provision \(\As(i,m)>0\) in (\ref{InflUnderlying}), we get
the following fact that parallels the behavior of inflations in
\(\Sgen\): if \(\As=\Bs+\Cs\), then \(\infl{\As}{b} = \infl{\Bs}{b}
+ \Cs\) for all \(b\in \Bs\), and \(\infl{\As}{c} =
\Bs+\infl{\Cs}{c}\) for all \(c\in \Cs\).
\begin{lem} \label{infl_sig_commute}
If \(A\) is in \(\Sgen\) and \(m < |A|\), then the signature of
the inflation of \(A\) by \(a_m\) is the inflation of \(\As\) by \(m\).
\end{lem}
\begin{proof}
The proof is by induction on the complexity of \(A\).
Since the lemma is vacuous if \(|A| \leq 1\) and
by the remarks made just before Lemma \ref{inflate_biembedd},
we may assume that \(A\) is indecomposable.
First suppose that \(m = |A|-1\).
Since \(\infl{A}{{a_m}}\) is in \(\Sgen\), Theorem \ref{sig_exclam} implies that if \(i,j < m\) then
\begin{align*}
o(a_i,a_j^{a_m}) & \geq \min\big( o(a_j^{a_m},a_m)-1 , o(a_i,a_m) \big) \\
& = \min\big( o(a_j,a_m)-1 , o(a_i,a_m) \big).
\end{align*}
On the other hand, Lemma \ref{o_conj_bound} implies
\[
o(a_i,a_j^{a_m}) \leq \min\big( o(a_j,a_m)-1 , o(a_i,a_m) \big).
\]
Thus \(o(a_i,a_j^{a_m}) = \min\big( o(a_j,a_m)-1 , o(a_i,a_m) \big) = \infl{\As}{m}(i,j^m)\),
as desired.
If \(m < |A|-1\), then by our induction hypothesis the signature of the inflation \(\infl{(A\cut)}{m}\) is
the corresponding inflation \(\infl{(\As\cut)}{m}\) of the signature \(\As \cut\).
Since the signature and rotation maps commute, it therefore suffices to verify that whenever \(i < m\)
\[
o(a_i^m, a_{\max}) = \min \big( o(a_i,a_m),o(a_m,a_{\max}) \big).
\]
By our induction hypothesis
\begin{align*}
o(a_i^m,a_{\max}) & = o(a_{\max} \cut , a_i^m) + 1 \\
& =
\min
\big(
o(a_{\max} \cut ,a_m),
o(a_i, a_m) -1
\big) + 1 \\
&
= \min \big(
o(a_m, a_{\max}) -1,
o(a_i, a_m) -1
\big) + 1 \\
&
= \min \big( o(a_i,a_m),o(a_m,a_{\max}) \big).
\end{align*}
\end{proof}
\begin{remark}
Note the complementary role which the inequalities in (!) and those in
Lemma \ref{o_conj_bound} play in this proof.
\end{remark}
We now define the relation \(\le\) on \(\Ssig\) just as with
\(\Sgen\): \(\As \leq \Bs\) if and only if there is a sequence
\(\Bs_0 = \Bs\), \(\Bs_1\), ..., \(\Bs_n =\As\) such that if \(i <
n\) then \(\Bs_{i+1}\) is contained in (i.e., is a restriction of)
an inflation of \(\Bs_i\).
\begin{prop} \label{reduction_to_embedding}
If \(A, B \in \Sgen\) then
the following are true:
\begin{enumerate}
\item if \(A \leq B\), then \(\As \leq \Bs\);
\item if \(\As \leq \Bs\), then there exist \(A' \in \Sgen\) with
signature \(\As\) such that \(A' \leq B\).
\end{enumerate}
In particular, if \(\As \leq \Bs\) are in \(\Ssig\), then \(\gen{A}\) embeds into \(\gen{B}\).
\end{prop}
\begin{proof}
The first assertion is an immediate consequence of Lemma
\ref{infl_sig_commute}. The second assertion in the case \(\As =
\infl{\Bs}{m}\) for some \(m\) is also an immediate consequence of
Lemma \ref{infl_sig_commute}; the general case follows by induction.
The final assertion follows from the second assertion, Theorem
\ref{sig_thm} and Proposition \ref{inflate_biembedd}. \end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Wreath products of \(\Sgen\)-generated groups}
\label{wreath:sec}
In this section we show that the operation \(*\) on \(\Ssig\) introduced in Section \ref{BasicOp:sec}
corresponds to forming a permutation wreath product of the associated groups.
Recall that if \(A,B,C \in \Sgen\) then \(A = B * C\) asserts that \(A = B
\cup C\) and both \(b < c\) and \(o(b,c) =1\) hold whenever \(b \in
B\) and \(c \in C\).
Observe that if \(A \in \Sgen\) and \(A = B*C\) for
nonempty \(B,C\), then there is an open interval \(J \subset I\)
such that: \begin{itemize}
\item
\(J\) contains the supports of all elements of \(B\);
\item
\(J\) is contained in the rightmost orbital of each element of \(C\)
and is disjoint from the feet of \(C\).
\end{itemize}
If moreover \(\Bs(i,j) > 0\) for all \(i < j < |B|\),
then there is a \(t_0\) which is in the rightmost orbital of each \(b_i\).
Fix such \(J\) and \(t_0\) and define
\[
\Xcal := \{t_0 f \mid f \in \gen{B}\}
\qquad
\Ycal := \{ J f \mid f \in \gen{C}\}.
\]
The goal of this section is to prove the following proposition.
\begin{prop} \label{wreath_prop}
Suppose that \(A \in \Sgen\) and \(\As(i,j) > 0\) for all \(i < j <
|A|\). If \(A = B * C\) for nonempty \(B,C \subset A\), then the
actions of \(\gen{B}\) on \(\Xcal\) and \(\gen{C}\) on \(\Ycal\) are
faithful and consistent. In particular the action of \(\gen{A}\) on
\(\Xcal \times \Ycal\) is the permutation wreath product of these
actions: \(\gen{B * C} \cong \gen{B} \wr \gen{C}\).
\end{prop}
\begin{proof}
Observe that by Theorem \ref{excision} of this paper and by Theorem
7.1 of \cite{fast_gen}, we may assume that \(A\) has no extraneous
bumps. The proposition is an immediate consequence of the next two
lemmas using the assignments \(K = \{t_0\}\) with \(S=B\), and
\(K=J\) with \(S=C\).
\end{proof}
\begin{lem}\label{ComplIntLem}
Let \(S \in \Sgen\) and \(K\) be a singleton or an open interval
disjoint from the feet of \(S\). If \(g \in \gen{S}\) and \(K g\cap
K\ne \emptyset\), then \(g |_K\) is the identity.
\end{lem}
\begin{proof}
In the language of \cite{fast_gen}, every point in \(K\) has trivial
history. Lemma 5.6 of \cite{fast_gen} implies that every orbit of
\(\gen{S}\) intersects \(K\) in at most one point. This gives the
conclusion if \(K\) is a singleton. If \(K\) is open, observe that
if \(K g\cap K\ne\emptyset\), then some \(t\in K\) has \(tg\in K\)
which implies \(tg=t\). Let \(x\) be any fixed point of \(g\) in
\(K\). By continuity, there is an open subset \(U\) of \(K\) about
\(g\) with \(Ug\subset K\) implying that \(g\) is the identity on
\(U\). Thus the fixed set of \(g\) in \(K\) is open in \(K\).
Again by continuity, the fixed set of \(g\) in \(K\) is closed in
\(K\) and must equal \(K\).
\end{proof}
\begin{lem}\label{FaithfulLem}
Let \(S \in \Sgen\) have no extraneous bumps and satisfy that
\(\mathsf{S}(i,j) > 0\) for all \(i < j < |S|\). If \(J\) is a
singleton or an open interval contained in the rightmost orbital of
every \(f \in S\) and disjoint from the feet of \(S\), then the
action of \(\gen{S}\) on \(\{J g \mid g \in \gen{S} \}\) is
faithful.
\end{lem}
\begin{proof}
By Proposition \ref{faithful_prop}, it suffices to show that \(\bigcup \{J g \mid g \in \gen{S}\}\)
intersects each orbital of \(S\).
We prove this by induction on the complexity of \(S\).
Let \(f\) be the top element of \(S\).
If \(|S| \leq 1\), there is nothing to show.
By our inductive assumption,
\[
X:=\bigcup \{J g \mid g \in \gen{S \setminus \{f\}}\}
\]
intersects every orbital of \(S \setminus \{f\}\).
Moreover the closure of \(X\) contains the set of all transition points of \(S \setminus \{f\}\)
and hence intersects every active orbital of \(f\).
Since every orbital of \(f\) is active and open, \(X\) intersects every orbital of \(f\).
\end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The signatures are well-ordered by reduction}
\label{arithmetic:sec}
Our next task is to analyze the structure of the transitive,
reflexive relation \((\Ssig,\leq)\) and prove that it is a
\emph{pre-well-order} (i.e. a well-order on the equivalence classes
obtained by identifying \(\As\) and \(\Bs\) whenever \(\As \leq \Bs \leq \As\)).
We introduce a subcollection \(\Bsig \subset \Ssig\) of signatures in \emph{block form}.
We show that elements of \(\Ssig\)
are equivalent to elements of \(\Bsig\) and that elements of \(\Bsig\)
are equivalent to unique elements of \(\Rsig\) modulo permuting summands.
Our goal is to analyze the structure of \((\Rsig,\leq)\) and prove
Theorem \ref{R_thm} from the introduction.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Signatures in block and reduced form}
\label{Bcal_Rcal:subsec}
We use the notation
%
\[
\sum_{i < k+1} \As_i := \big(\sum_{i < k} \As_i
\big) + \As_k,
\quad
\text{and}
\quad
\As\cdot m := \sum_{i<m}\As.
\]
%
Recall that if \(\As\) is in \(\Ssig\), then \(\exp(\As)\)
has the same base as \(\As\) and
\(\exp(\As)(i,j) = \As(i,j)+1\) for all \(i < j\) in \(A\).
Note that \(\exp(\one) = \one\).
Observe that \(\exp\) maps \(\Ssig\) injectively into \(\Ssig\) and
that the range of \(\exp\) is exactly those elements of \(\Ssig\)
which take only positive values.
As noted earlier \(\As * \Bs\) is only defined if \(\Bs = \exp(\Xs)\) for some
\(\Xs \in \Ssig\).
Observe that \(+\) and \(*\) are associative
operations on \(\Ssig\) and that \(\exp(\As+\Bs) =
\exp(\As)*\exp(\Bs)\).
We now give a more detailed description of the family
\(\Rsig\) from Section \ref{BasicOp:sec}.
It is easiest to define \(\Rsig\) if we first define a class \(\Bsig\)
(signatures in block form), and an ordinal function \(\rho\) on \(\Bsig\).
The class \(\Bsig\) is the smallest class containing \(\{\zero,
\one\}\) so that whenever \((\Xs_i \mid i < n)\) is a sequence of
elements of \(\Bsig\), then \(\sum_{i<n} \exp(\Xs_i)\) is in
\(\Bsig\). Observe that if \(\Bs\ne \zero\) is in \(\Bsig\), then
there exists a unique sequence \((\As_i \mid i < n)\) of nonzero
elements of \(\Bsig\) such that \(\Bs = \sum_{i<n} \exp(\As_i)\).
Define \(\rho\) recursively on \(\Bsig\) by \(\rho(\zero) = 0\),
\(\rho(\one) = 1\), \(\rho\big(\exp(\As)\big) =
\omega^{-1+\rho(\As)}\), and \(\rho(\sum_{i<n} \exp\big(\As_i)\big)
= \sum_{i<n} \rho\big(\exp(\As_i)\big)\). Here \(-1 + \alpha =
\beta\) if \(\alpha = 1 + \beta\) (so \(-1 + \alpha = \alpha\) if
\(\alpha\) is infinite). This technicality exists because \(\one\)
is a fixed point of \(\exp\), but the ordinal 1 is not a fixed point
of \(\alpha\mapsto \omega^\alpha\).
Now we can define \(\Rsig\) to be the smallest subfamily of
\(\Bsig\) which contains \(\zero\) and \(\one\) so that
if \((\Xs_i \mid i < n)\) is a sequence from \(\Rsig\) and
\(\rho(\Xs_{i+1}) \leq \rho(\Xs_i)\) for all \(i < n-1\), then
\(\sum_{i<n} \exp(\Xs_i)\) is in \(\Rsig\). The reader can verify
that if \(\Xs\) is the signature in Figure \ref{osc_fig}, then
\(\Xs\in \Rsig\) and
%
\[
\rho(\Xs) = \omega^{\omega^{(\omega+2)}}.
\]
Observe that if \(\Bs \in \Bsig\) and \(\As \in \Ssig\)
is contained in \(\Bs\), then \(\As \in \Bsig\) and moreover
\(\rho(\As) \leq \rho(\Bs)\) (both facts are established by
induction on the complexity of \(\Bs\)). The next lemma has a
straightforward inductive proof and is left to the reader.
\begin{lem}\label{RhoIsIso} The restriction of \(\rho\) to \(\Rsig\)
is a bijective correspondence between
between the elements of \(\Rsig\) and the Cantor
normal forms for ordinals below \(\epsilon_0\). Further, the
bijection \(\rho\)
defines an isomorphism
%
\[
(\Rsig,\zero,\one,+,\exp) \cong (\epsilon_0,0,1,+,\zeta \mapsto
\omega^{-1 + \zeta}).
\]
provided \(+\) is restricted to those pairs in \(\Rsig\) whose sum remains in \(\Rsig\).
\end{lem}
Here we formally define \(\omega^{-1+0} = 0\).
Our first task in proving Theorem \ref{R_thm}
is to show \(\rho\) preserves the order which \(\Rsig\) inherits from \(\Ssig\).
The next lemma gives basic facts about the algebraic
operations and
their interaction with the relation \(\le\) on \(\Ssig\);
the proof is straightforward and left to the reader.
\begin{lem} \label{monotone}
Suppose that \(\As \leq \As'\) and \(\Bs \leq \Bs'\) are in \(\Ssig\).
The following are true:
\begin{enumerate}
\item \(\exp(\infl{\As}{m}) \leq \infl{(\exp(\As))}{m}\) for \(m < |A|\).
\item \(\exp (\As) \leq \exp (\As')\).
\item \(\As+\Bs \leq \As' + \Bs'\).
\item \(\As * \exp(\Bs) \leq \As' * \exp(\Bs')\).
\end{enumerate}
\end{lem}
For the following, recall that \(\As\equiv\Bs\) means \(\As\le
\Bs\le \As\).
\begin{lem} \label{duplication}
The following are true:
\begin{enumerate}
\item \label{gen_sum}
If \((\As_i \mid i < n)\) and \(\Bs\neq\zero\) are in \(\Ssig\) and
\(j < n\) is such that \(\As_i \leq \As_j\) for all \(i < n\), then
\((\sum_{i<n} \As_i) * \exp (\Bs) \equiv \As_j * \exp (\Bs)\).
\item \label{prod_case}
If \(\As \in \Ssig\) then for all \(m\), \(\exp(\As)\cdot m \le \exp(\As+\one)\).
\end{enumerate}
\end{lem}
\begin{proof} For (\ref{gen_sum}), first observe that either \(j > 0\) and
\(\As_0 \leq \sum_{k=1}^{n-1} \As_k\) or else \(\As_{n-1} \leq \As_0
\leq \sum_{k=0}^{n-2} \As_k\). Thus, by induction, it is sufficient
to prove the lemma when \(n=2\). Furthermore, by Lemma
\ref{monotone}, it suffices to prove that \((\As+\As) * \exp(\Bs)
\leq \As * \exp(\Bs)\).
In fact we show that \((\As+\As)*
\exp(\Bs) = \infl{\big(\As* \exp(\Bs)\big)}{m}\) where \(m\) is the
minimum element of \(\exp(\Bs)\) regarded as a suborder of \(\As*
\exp(\Bs)\). To see this, let \(i,j\) be elements of \(\As\) and
\(k\) be an element of \(\exp(\Bs)\) above \(m\). We have that
\begin{align*}
\infl{(\As* \exp(\Bs))}{m}(i,j^m) & = \min \big(\As*
\exp(\Bs)(i,m),\As* \exp(\Bs)(j,m)-1 \big)
\\
& =\min (1,0) = 0 \\
\infl{\big(\As* \exp(\Bs)\big)}{m}(j^m,k) & = \min\big(\As*
\exp(\Bs)(j,m),\As* \exp(\Bs)(m,k) \big)
\\
& =
\min\big(1,\exp(\Bs)(m,k)\big) = 1
\end{align*} which coincides
with the definition of \((\As+\As)*\exp(\Bs)\).
To see that (\ref{prod_case}) is true, let \(\As\) and \(m\) be given.
By (\ref{gen_sum}) we have \((\exp(\As)\cdot m)*\one \equiv \exp(\As)*\one\).
The conclusion follow by observing
\[
(\exp(\As)\cdot m)\le (\exp(\As)\cdot m)*\one,
\]
\[
\exp(\As)*\one = \exp(\As)*\exp(\one) = \exp(\As+\one).
\]
\end{proof}
Lemma \ref{duplication} is an early hint that the
arithmetic on \(\Ssig\) imitates the behavior of arithmetic on the ordinals.
For instance, in analogy to (\ref{gen_sum}), using \(1 + \omega = \omega\) we note
\[
(\omega+1 ) \cdot \omega\ =\ \sup_{n \in \omega}\ (\omega+1) \cdot n\ =\ \sup_{n \in \omega}\ \omega \cdot n + 1\ =\ \omega^2.
\]
Next we begin our analysis of \(\Rsig\).
\begin{lem} \label{R_ordering}
If \(\Bs\) is in \(\Rsig\) and \(\alpha < \rho(\Bs)\), then there an
\(\As\) in
\(\Rsig\) such that
\(\As \leq \Bs\) and \(\rho(\As) = \alpha\).
\end{lem}
\begin{proof}
The proof is by induction on \(\rho(\Bs)\).
If \(\rho(\Bs) = 0\), then the lemma is vacuously true.
Now suppose \(\rho(\Bs) \ge 1\) and let \(\Bs = \sum_{i<n} \exp(\Xs_i)\)
where \(\rho(\Xs_{i+1}) \leq \rho(\Xs_i)\) for each \(i < n-1\)
(note that
possibly \(n=1\) in which case
\(\Bs = \exp(\Xs_0)\)).
Let \(k < n\) be minimal such that \(\alpha < \rho\big(\sum_{i \leq k}
\exp(\Xs_i)\big)\).
If \(k=0\) and \(\Xs_0 = \one\), then \(\alpha < \exp(\Xs_0) = 1\)
and thus
\(\alpha = 0\).
In this case we take \(\As = \zero\).
If \(k=0\) and \(\Xs_0 = \Ys+\one\) for some \(\Ys \ne \zero\), then
\[
\alpha < \rho\big(\exp(\Xs_0)\big) = \rho\big(\exp(\Ys + 1)\big) =
\omega^{-1 + \rho(\Ys) + 1} = \rho\big(\exp(\Ys)\big) \cdot \omega
\]
and there is an \(m\) such that \(\alpha < \rho\big(\exp(\Ys)\big)
\cdot m\).
From (\ref{prod_case}) of Lemma \ref{duplication},
\[
\Bs' := \exp(\Ys) \cdot m \leq \exp(\Ys+\one) = \exp(\Xs_0) \leq \Bs.
\]
Since \(\Bs'\) is in \(\Rsig\), by our induction hypothesis
there is an \(\As\) in \(\Rsig\) such that \(\As \leq \exp(\Ys)
\cdot m = \Bs'
\leq \Bs\)
and such that \(\rho(\As) = \alpha\).
If \(k=0\) but we are not in the previous cases, then \(\rho(\Xs_0)
= \delta\)
is a limit ordinal.
Let \(0 < \gamma < \delta\) be such that \(\alpha < \omega^{-1+
\gamma}\).
By our induction hypothesis, there is a \(\Cs \leq \Xs_0\) in
\(\Rsig\) such
that
\(\rho(\Cs) = \gamma\) and hence \[
\rho\big(\exp(\Cs)\big) = \omega^{-1 + \gamma} < \omega^{-1 +
\delta} = \rho(\Bs).
\]
Applying our induction hypothesis again, there is an \(\As \leq
\exp(\Cs)\le \exp(\Xs_0)=B\) such that
\(\As\) is in \(\Rsig\) and \(\rho(\As) = \alpha\).
Finally, if \(0 < k \leq n\), then let \(0 < \alpha' <
\rho\big(\exp(\Xs_k)\big)\) be such that
\[
\alpha = \rho(\sum_{j < k} \exp\big(\Xs_j)\big) + \alpha'.
\]
Now
\[
\alpha' < \rho\big(\exp(\Xs_k)\big) <
\rho\big( \exp(\Xs_0) + \exp(\Xs_k)\big) \leq \rho(\Bs).
\]
By our induction hypothesis there is
an \(\As' \leq \exp(\Xs_k)\) such that \(\As' \in \Rsig\) and
\(\rho(\As') = \alpha'\).
Observe that \(\As' = \sum_{i < m} \exp(\Ys_i)\) for some \(m \geq 1\)
where the \(\Ys_i\)'s come from \(\Rsig\) and
\(\rho\big(\exp(\Ys_{i+1})\big) \leq
\rho\big(\exp(\Ys_i)\big)\) for all
\(i < m\).
In particular, \(\rho\big(\exp(\Ys_0)\big) \leq \alpha' <
\rho\big(\exp(\Xs_k)\big)\) and therefore
\[
\As = \sum_{i < k} \exp(\Xs_i) + \sum_{j < m} \exp(\Ys_j)
\]
is in \(\Rsig\) and satisfies that \(\rho(\As) = \alpha\).
\end{proof}
\begin{lem} \label{rank_to_reduction}
If \(\As\) and \(\Bs\) are in \(\Rsig\), then \(\rho(\As) \leq
\rho(\Bs)\) implies \(\As \leq \Bs\).
\end{lem}
\begin{proof}
If \(\rho(\As) < \rho(\Bs)\), then by Lemma \ref{R_ordering} there is an
\(\As' \leq \Bs\) in \(\Rsig\) such that
\(\rho(\As') = \rho(\As)\).
Since \(\rho\) is one-to-one on \(\Rsig\), we have that \(\As'=\As\).
Similarly, if \(\rho(\As) = \rho(\Bs)\), then \(\As = \Bs\).
\end{proof}
For \(\Xs \in \Ssig\), recall that
\(E(\Xs) = \exp\big(\exp(\Xs)\big)\).
\begin{lem} \label{proper_successor}
If \(A,B \in \Rsig\) and \(\As+1 \leq \Bs\), then
\(\gen{E(\Bs)}\) does
not embed into \(\gen{E(\As)}\).
\end{lem}
\begin{proof}
Let \(n\) be the maximum element of \(E(\As+\one)\) and note
that \(E(\As) * E(\As)\) is obtained from
\(\infl{E(\As + \one)}{n}\) by removing its maximum element.
Consequently
\[
E(\As) * E (\As) \leq E(\As+\one) \leq E(\Bs).
\]
By Propositions \ref{reduction_to_embedding}, \ref{wreath_prop}, and
\ref{rank_to_reduction}
\[
\gen{E(\As) * E(\As)} \cong \gen{E(\As)} \wr \gen{E(\As)}
\]
embeds into \(\gen{\exp(\Bs)}\).
By Theorem \ref{EAGWrG},
the EA-class of \(\gen{E(\As)}\) is less than that of
\(\gen{E(\As)} \wr \gen{E(\As)}\) which is at most
that of \(\gen{E (\Bs)}\).
Consequently \(\gen{E (\Bs)}\) is not embeddable into
\(\gen{E(\As)}\).
\end{proof}
\begin{lem} \label{embedding_to_rank}
If \(\As\) and \(\Bs\) are in \(\Rsig\) and \(\rho(\As) < \rho(\Bs)\),
then \(\gen{E(\Bs)}\) does not embed into \(\gen{E(\As)}\).
\end{lem}
\begin{proof}
If \(\rho(\As) < \rho(\Bs)\), then \(\rho(\As+\one) \leq \rho(\Bs)\).
By Lemma \ref{rank_to_reduction}, \(\As + \one \leq \Bs\) and
therefore we have the desired conclusion
by Lemma \ref{proper_successor}.
\end{proof}
This now leads to the following characterization of the restriction
of \(\leq\) to \(\Rsig\).
\begin{lem} \label{R_order_char}
If \(\As\) and \(\Bs\) are elements of \(\Rsig\), then the following
are equivalent:
\begin{enumerate}
\item \label{R_reduction}
\(\As \leq \Bs\).
\item \label{R_rank}
\(\rho(\As) \leq \rho(\Bs)\).
\end{enumerate}
\end{lem}
\begin{proof}
Lemma \ref{rank_to_reduction} establishes that (\ref{R_rank})
implies (\ref{R_reduction}).
If \(\As\le \Bs\), Lemma \ref{monotone} gives
\(E(\As)\le E(\Bs)\) which implies that
\(\gen{E(\As)}\) embeds in \(\gen{E(\Bs)}\) by Proposition
\ref{reduction_to_embedding}. The contrapositive of Lemma
\ref{embedding_to_rank} with the roles of \(\As\) and \(\Bs\)
reversed gives \(\rho(\As)\le \rho(\Bs)\).
\end{proof}
\begin{remark}
While Lemma \ref{R_order_char} does not mention groups or embeddings
between them, it is convenient to use the group theory concept of
the EA-class to establish the implication. This is a
matter of economy since we will eventually generalize the lemma to
\(\Ssig\) and add ``\(\gen{\As}\) embeds into \(\gen{\Bs}\)'' to the
list of equivalents.
On the other hand, it not obvious how to provide a
proof of Lemma \ref{R_order_char} which avoids group theory.
\end{remark}
We now broaden our analysis to \(\Bsig\) and then to \(\Ssig\).
\begin{lem} \label{regroup}
If \(\As\), \(\Bs\), and \(\Cs\) are in \(\Ssig\) with
\(\Bs,\Cs \ne\zero\), then
\[
\exp\big(\As + \Bs * \exp(\Cs)\big) \leq \exp \big((\As+\Bs) *
\exp(\Cs)\big).
\]
\end{lem}
\begin{proof}
Let \(m\) be the minimal element of \(\Bs\) in \(\exp\big((\As+\Bs)
* \exp(\Cs)\big)\).
It suffices to check that the map
\[
\pi: \exp\big(\As + \Bs * \exp(\Cs)\big) \to
\infl{\exp\big((\As+\Bs) * \exp(\Cs)\big)}{m}
\]
defined by
\[
\pi (i) =
\begin{cases}
i^m & \textrm{ if } i \in \As \\
i & \textrm{ if } i \in \Bs* \exp(\Cs)
\end{cases}
\]
is an embedding.
First observe that the restrictions of \(\pi\) to
\(\As\) and \(\Bs * \exp(\Cs)\) are embeddings.
Furthermore, if \(i \in \As\) and
\(j \in \Bs * \exp(\Cs)\), then \(i^m < m \leq j\) and hence \(\pi\)
is order preserving.
Finally suppose that \(i \in \As\) and \(j \in \Bs * \exp(\Cs)\).
By definition,
\[
\infl{\exp\big((\As+\Bs) * \exp(\Cs)\big)}{m} (\pi(i),\pi(j)) =
\]
\[
\min\big( \exp\big((\As+\Bs) * \exp(\Cs)\big)(i,m),
\exp\big((\As+\Bs) * \exp(\Cs)\big)(m,j) \big)
\]
\[
=\min \big(1,\exp\big((\As+\Bs) * \exp(\Cs)\big)(m,j) \big)
\]
\[
= 1 = \exp\big(\As+\Bs*\exp(\Cs)\big)(i,j).
\]
\end{proof}
\begin{lem} \label{weak_init_seg}
If \(\As\) is in \(\Bsig\) and \(\beta < \rho(\As)\),
then there exist \(\Bs,\Cs \in \Bsig\) such that \(\exp(\As)\) is
equivalent to \(\exp(\Bs+\Cs)\) and
\(\beta \leq \rho(\Bs) < \rho(\As)\).
Moreover, if \(\As\) is indecomposable, then we can take \(\Cs = \As\).
\end{lem}
\begin{proof}
The proof is by induction on \(\rho(\As)\) and then on the
cardinality of \(\As\).
Toward this end, let \(\As\) be given;
there are now several cases to consider.
If \(\As = \zero\), then the lemma is vacuously true and if
\(\As = \one\) then \(\beta = 0\) and we can take \(\Bs =\zero\) and \(\Cs = \As\).
Next suppose that \(\As = \Bs_0 + \Cs_0\) with both \(\Bs_0\) and
\(\Cs_0\) not \(\zero\).
If \(\beta \leq \rho(\Bs_0)\), then we are done.
Otherwise, let \(\gamma\) be such that
\(\beta = \rho(\Bs_0) + \gamma\), noting that \(\gamma < \rho(\Cs_0)\).
Since the cardinality of \(\Cs_0\) is smaller than that of \(\As\),
we can apply our induction hypothesis to find
\(\Bs_1\) and \(\Cs\) such that \(\exp(\Bs_1+\Cs)\) is equivalent to
\(\exp(\Cs_0)\) and
\(\gamma \leq \rho(\Bs_1) < \rho(\Cs_0)\).
Observe that \(\Bs = \Bs_0 + \Bs_1\) and \(\Cs\) satisfies the
conclusion of the lemma:
\[
\beta = \rho( \Bs_0 ) + \gamma \le \rho( \Bs_0 ) + \rho( \Bs_1 )
= \rho( \Bs_0 + \Bs_1 ) = \rho(\Bs)
\]
while basic manipulations with the arithmetic in \(\Bsig\) gives
\[
\exp(\Bs + \Cs) = \exp( \Bs_0 ) * \exp( \Bs_1 + \Cs) \equiv
\exp(\Bs_0) * \exp(\Cs_0) = \exp(\As).
\]
Next suppose that \(\As = \exp(\Ds + \one)\) for \(\Ds \ne \zero\).
Let \(n\) be such that \(\beta \leq \omega^{-1 + \rho(\Ds)} \cdot n\).
Set \(\Bs = \exp(\Ds) \cdot n\).
By our choice of \(n\) we have that \(\beta \leq \rho(\Bs) < \rho(\As)\).
Also, it is clear that \(\exp(\As) \leq \exp(\Bs+\As)\) and hence it
is sufficient in this case to show
that \(\exp(\Bs+\As) \leq \exp(\As)\).
This follows from Lemmas \ref{regroup} and \ref{duplication} respectively:
\[
\begin{split}
\exp\big(\exp(\Ds) \cdot n + &\exp(\Ds+\one)\big) =
\exp\big(\exp(\Ds) \cdot n + \exp(\Ds)* \exp(\one)\big) \\
\leq &\exp\big(\big(\exp(\Ds) \cdot (n+1)\big) * \exp(\one)\big) \\
\leq &\exp\big(\exp(\Ds) * \exp(\one)\big) =
\exp\big(\exp(\Ds+\one)\big).
\end{split}
\]
Finally, suppose that \(\As = \exp(\Ds)\) and \(\rho(\Ds)\) is a
limit ordinal \(\delta\).
Let \(\gamma < \delta\) be such that \(\beta \leq \omega^{-1 + \gamma}\).
By our induction hypothesis, there exist \(\Es,\Fs \in \Bsig\) such that
\(\exp(\Ds) \equiv \exp(\Es + \Fs)\) and \(\gamma \leq \rho(\Es) <
\delta\).
We set \(\Bs = \exp(\Es)\).
As in the previous case, it suffices to show that
\(\exp(\Bs+\As) \leq \exp(\As)\).
This again follows from Lemmas \ref{regroup} and \ref{duplication}
respectively:
\[
\begin{split}
\exp\big(\Bs + \As\big) \equiv &\exp\big(\exp(\Es) +
\exp(\Es+\Fs)\big) \\
= &\exp \big(\exp(\Es) + \exp(\Es)*\exp(\Fs)\big) \\
\leq &\exp\big(\big(\exp(\Es) + \exp(\Es)\big) * \exp(\Fs)\big) \\
\leq &\exp\big(\exp(\Es)* \exp(\Fs)\big) \equiv
\exp\big(\exp(\Ds)\big) = \exp(\As).
\end{split}
\]
\end{proof}
\begin{lem} \label{exact_init_seg}
If \(\As\) is in \(\Bsig\) and \(\beta < \rho(\As)\),
then there exist \(\Bs,\Cs \in \Bsig\) such that \(\exp(\As)\) is
equivalent to \(\exp(\Bs+\Cs)\) and
\(\beta = \rho(\Bs)\).
Moreover, if \(\As\) is indecomposable, then we can take \(\Cs = \As\).
\end{lem}
\begin{proof}
The proof is by induction on \(\rho(\As)\).
As before, the case \(\rho(\As) \leq 1\) is trivial.
If \(\beta < \rho(\As)\), then by Lemma \ref{weak_init_seg},
there exist \(\Xs,\Ys \in \Bsig\) such that
\(\exp(\Xs+\Ys) \equiv \exp (\As)\), \(\beta \leq \rho(\Xs) <
\rho(\As)\), and \(\Ys = \As\) if
\(\As\) is indecomposable.
If \(\rho(\Xs) = \beta\), then we are done.
Otherwise,
by our inductive assumption, there exist \(\Bs,\Rs \in \Bsig\) such
that
\(\rho(\Bs) = \beta\) and \(\exp (\Bs+\Rs) \equiv \exp(\Xs)\).
If we set \(\Cs := \Rs + \Ys\), we have that
\[
\begin{split}
\exp(\Bs + \Cs) = &\exp\big(\Bs + (\Rs+\Ys)\big) = \exp(\Bs+\Rs)
* \exp(\Ys) \\
\equiv &\exp(\Xs) * \exp(\Ys) = \exp(\Xs+\Ys) \equiv \exp(\As).
\end{split}
\]
If \(\As\) is indecomposable, then we get \(\exp(\Bs+\As)\equiv
\exp(\As)\) from
\[
\begin{split}
\exp(\As) \le &\exp(\Bs+\As) \le \exp(\Bs + \Rs + \As)
= \exp( \Bs + \Rs) * \exp(\As) \\
\equiv & \exp(\Xs)*\exp(\As) = \exp(\Xs+\As) \equiv \exp(\As).
\end{split}
\]
So \(\Cs=\As\) fits the conclusion of the lemma.
\end{proof}
\begin{lem} \label{block_to_R}
If \(\As\) is an indecomposable element of \(\Bsig\), then there is
a unique \(\Bs \in \Rsig\) such that
\(\As \equiv \Bs\).
\end{lem}
\begin{proof}
First observe that if \(\As \in \Bsig\) is indecomposable, then
\(\As = \exp(\Xs)\) for some \(\Xs\).
Uniqueness of \(\Bs\) follows immediately from Lemma
\ref{R_order_char}.
The proof of existence of \(\Bs\) is by induction on \(\rho(\As)\).
If \(\rho(\As) = 1\), then \(\As = \one\) and \(\As \in \Rsig\).
If \(\As = \exp(\Xs)\) and \(\Xs\) is indecomposable, then by our
induction hypothesis,
there is a \(\Ys \in \Rsig\) such that \(\Ys \equiv \Xs\).
Since \(\Bs = \exp(\Ys)\) is also in \(\Rsig\),
it follows from Lemma \ref{monotone} that \(\As = \exp(\Xs) \equiv
\exp(\Ys) = \Bs\).
If \(\As = \exp\big( \sum_{i<n} \exp(\Xs_i)\big)\), then by our
induction hypothesis there are \((\Ys_i \mid i < n)\) in \(\Rsig\)
such that
\(\Ys_i \equiv \Xs_i\) for each \(i < n\).
If there is no \(k < n-1\) such that \(\rho(\Ys_k) <
\rho(\Ys_{k+1})\), then
\(\exp\big( \sum_{i<n} \exp(\Ys_i)\big)\) is in \(\Rsig\) and is
equivalent to \(\As\).
Suppose now that there is an \(k < n-1\) such that \(\rho(\Ys_k) <
\rho(\Ys_{k+1})\) and note that in this case \(n > 1\).
We first claim that \(\exp\big(\exp(\Ys_k) + \exp(\Ys_{k+1})\big)
\equiv \exp\big(\exp(\Ys_{k+1})\big)\).
To see this, observe that by Lemma \ref{exact_init_seg} with
\(\As=\exp(\Ys_{k+1})\) and \(\beta=\rho(\exp(\Ys_k))\)
there exists
a \(\Rs \in \Bsig\) such that
\(\rho(\Rs) = \rho\big(\exp(\Ys_k)\big)\) and \(\exp\big(\Rs +
\exp(\Ys_{k+1})\big) \equiv \exp\big(\exp(\Ys_{k+1})\big)\).
Since
\[
\rho\big(\exp(\Ys_k)\big) < \rho\big(\exp(\Ys_{k+1})\big) \leq \rho(\As),
\] we can apply
our induction hypothesis to conclude that \(\Rs \equiv \exp(\Ys_k)\)
and thus
that \(\exp\big( \exp(\Ys_k) + \exp(\Ys_{k+1})\big) \equiv
\exp\big(\exp(\Ys_{k+1})\big)\).
Reindexing the remaining summands and repeating the process if
necessary, we eventually
arrive at an element of \(\Rsig\) which is equivalent to \(\As\).
\end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Representing elements of \(\Ssig\) in \(\Rsig\)}
\label{Scal_reduction:subsec}
Next we turn to the general analysis of elements of \(\Ssig\).
We need the following characterization of nontrivial products,
analogous to Lemma \ref{dir_sum_char}.
\begin{lem} \label{factor_lem}
Suppose \(\As\) is in \(\Ssig\) and has cardinality at least \(2\) and
maximum element \(n\).
If \(\As\) satisfies the following conditions:
\begin{itemize}
\item for all \(i < n\), \(\As(i,n) \geq 1\) and
\item
there exist \(m < n\) such that \(\As(i,n) = 1\) if and only if \(i \leq m\),
\end{itemize}
then \(\As = \Bs*\exp(\Cs)\) for some nonzero
\(\Bs,\Cs \in \Ssig\).
\end{lem}
\begin{proof}
If \(i \leq m < j < n\), then \(\As(i,n) = 1 < \As(j,n)\) and hence
\(\As(i,j) = \min \big(\As(j,n)-1,1\big) = 1\).
Furthermore, if \(m < i < j < n\) then \(\As(i,j) \geq \min (\As(j,n)-1, \As(i,n)) > 0\).
Thus \(\As = \Bs * \exp(\Cs)\), where \(\exp(\Cs)\) consists of the
elements of \(\As\) above
\(m\) and \(\Bs\) consists of the remaining elements. This holds
even if \(m=n-1\). In this case \(\As = \Bs*\one = \Bs *
\exp(\one)\).
\end{proof}
\begin{lem} \label{decomp_lem}
If \(\As \ne \zero,\one\) is in \(\Ssig\),
then one of the following is true for some nonzero \(\Bs,\Cs \in \Ssig\):
\begin{enumerate}
\item \(\As = \exp(\Bs)\),
\item \(\As = \Bs + \Cs\), or
\item \(\As\) is equivalent to \(\Bs * \exp(\Cs)\) and \(\Bs * \exp(\Cs)\) has the same
cardinality as \(\As\).
\end{enumerate}
\end{lem}
\begin{proof}
Let \(\As\) be given with maximum element \(n\).
We are done if there are no \(i < j \leq n\) such
that \(\As(i,j) = 0\) since then \(\As = \exp(\Bs)\) for some \(\Bs\).
If \(\As(i,n) = 0\) for some \(i < n\), then \(\As\) is decomposable
by Lemma \ref{top_positive}
and we are finished.
Now suppose the first two conclusions of the lemma do not occur.
In this case, there is an \(i < j < n\) such that \(\As(i,j) = 0\)
but there is no \(i' < n\) such
that \(\As(i',n) = 0\).
Since \(\As\) is in \(\Ssig\), it must be that \(\As(j,n) = 1\) and
in particular there is a
\(j < n\) such that \(\As(j,n) = 1\).
Define \(\Ds\) to be the restriction of \(\infl{\As}{n}\) such that
\[
\Ds = \{i^n \mid i<n \textrm{ and }\As(i,n) > 1\} \cup \{i \mid \As(i,n) =1\} \cup \{n\}.
\]
Clearly \(\Ds \leq \As\).
Observe that \(\Ds\) satisfies the hypothesis of Lemma
\ref{factor_lem} and thus has the desired form
\(\Bs * \exp(\Cs)\) for some nontrivial \(\Bs\) and \(\Cs\).
Thus it is sufficient to show that \(\As \leq \Ds\). For each
\(i<n\) with \(\As(i,n)=1\), let \(u_i\) be the unique least element
of \(\As\) with \(i<u_i\le n\) and \(\As(u_i,n)>1\). This always
exists since \(\As(n,n)=\infty\).
We now perform a sequence of inflations of \(D\) by the members of
\[
J := \{u_i^n \mid i<n \textrm{ and } \As(i,n)=1\}
\]
in increasing order and then remove some of the resulting conjugates
so that the
base of the resulting \(\Xs \in \Ssig\) consists of:
\begin{itemize}
\item all \(i^n\) such that \(i \le n\) and \(\As(i,n) > 1\) (where
\(n^n:=n\));
\item all \(i^{(u_i^n)}\) such that \(\As(i,n)=1\).
\end{itemize}
We claim that \(\As\) isomorphic to \(\Xs\).
Define \(\pi: A \to X\) by
\[
\pi(i) =
\begin{cases}
i^n, &i\le n,\ \As(i,n) > 1, \\
i^{(u_i^n)},
&i<n,\ \As(i,n) = 1.
\end{cases}
\]
The proof is completed by Claims \ref{PiOrder} and \ref{PiIso} below.
\end{proof}
\begin{claim} \label{PiOrder}
\(\pi\) is order preserving.
\end{claim}
\begin{proof}
Let \(i < j < n\).
If \(\As(i,n)\) and \(\As(j,n)\) are both greater than \(1\),
then \(\pi(i) = i^n < j^n = \pi(j)\).
If \(\As(i,n) = 1\) and \(\As(j,n) > 1\), then
\(i<u_i\le j\) and \(\pi(i) = i^{(u_i^n)} < u_i^n\le j^n = \pi(j)\).
If \(\As(i,n) > 1\) and \(\As(j,n) = 1\), then \(i<j<u_j\),
\(i^n<u_j^n\), and in the inflation by \(u_j^n\), we have
\(\pi (i) = i^n<j^{(u_j^n)} = \pi(j)\).
Finally if \(\As(i,n) = \As(j,n) = 1\), then either \(u_i=u_j\) or
\(u_i<u_j\). In the first case, \(\pi(i) =
i^{(u_j^n)}<j^{(u_j^n)} = \pi(j)\). In the second case,
\(i^{(u_i^n)} < u_i^n < u_j^n\), so \(\pi(i) = i^{(u_i^n)} < u_i^n <
j^{(u_j^n)} = \pi(j)\).
\end{proof}
\begin{claim} \label{LowOscSpefics}
If \(i<u_i<j<u_j\), then
\(\Ds(i,u_i^n) = 1\),
and \(\As(i,j) = 0\).
\end{claim}
\begin{proof}
The definition of inflation gives
%
\[
\Ds(i,u_i^n) = \min\big((\As(u_i,n)-1, \As(i,n)\big) = 1.
\]
For the second conclusion, we have \(\As(j,n) = 1< \As(u_i,n)\), so
%
\[
\As(u_i,j) = \min\big(\As(j,n)-1, \As(u_i,n)\big) = 0.
\]
Now \(i<u_i\) and Lemma \ref{ZeroToEnd} gives \(\As(i,j)=0\).
\end{proof}
\begin{claim}\label{PiIso}
If \(i < j \le n\), then \(\Xs\big(\pi(i),\pi(j)\big) = \As(i,j)\).
\end{claim}
\begin{proof}
We first consider \(j=n\).
If either \(\As(i,n)>1\) or both \(\As(i,n)=1\) and
\(u_i=n\),
then \(\Xs(\pi(i), \pi(n)) = \Ds(i^n,n)
= \As(i,n)\).
If \(\As(i,n)=1\) and \(u_i<n\), then by Claim
\ref{LowOscSpefics} we have
%
\[
\Xs(i^{(u_i^n)}, n) = \min(\Ds(i,u_i^n), \Ds(u_i^n,n)) =
\min(1, \Ds(u_i^n,n)) =
1.\]
For \(i<j<n\), we first assume \(\As(i,n)\ne \As(j,n)\).
Since \(\Xs\) is in \(\Ssig\), the previous case implies
\begin{align*}
\Xs(\pi(i),\pi(j)) & = \min\big( \Xs(\pi(j),n)-1 ,
\Xs(\pi(i),n) \big) \\
& = \min\big( \As(j,n)-1 , \As(i,n) \big) \\
& = \As(i,j).
\end{align*}
If \(\As(i,n) = \As(j,n) > 1\),
then \(\pi(i) = i^n\) and \(\pi(j) = j^n\).
In this case we have that
\(\Xs(i^n,j^n) = \Ds(i^n,j^n) = \As(i,j)\).
Finally we have the case \(\As(i,n) = 1 = \As(j,n)\). If
\(u_i = u_j\), then
\(\Xs(i^{(u_j^n)}, j^{(u_j^n)}) = \Ds(i,j) = \As(i,j)\). So we
assume
\(i<u_i<j<u_j\). From the beginning of the proof \(\Xs(j^{(u_j^n)},
n) = \As(j,n)=1\). Since \(\Xs(u_i^n,n) = \Ds(u_i^n,n) =
\As(u_i,n)>1\), we have
%
\[
\Xs(u_i^n, j^{(u_j^n)}) = \min(\Xs(j^{(u_j^n)}, n)-1, \Xs(u_i^n, n))
=0.
\]
Now \(i^{(u_i^n)}< u_i^n\), so by Lemma \ref{ZeroToEnd}, we have
\(\Xs(i^{(u_i^n)}, j^{(u_j^n)})=0\). By Claim \ref{LowOscSpefics},
\(\As(i,j)=0 = \Xs (\pi(i),\pi(j))\).
\end{proof}
\begin{lem} \label{standard_to_block}
If \(\As\) is in \(\Ssig\), then there is a \(\Bs \in \Bsig\) such
that \(\As \equiv \Bs\).
Moreover, if \(\As\) is indecomposable, then \(\Bs\) can be taken to
be indecomposable.
\end{lem}
\begin{proof}
The proof is by induction on the complexity of \(\As\).
If \(\As = \one\) or \(\As=\zero\), then \(\As\) is already in \(\Bsig\).
If \(\As = \exp(\Bs)\) for some \(\Bs\), then
by our induction hypothesis, there is a \(\Bs' \in \Bsig\) such that
\(\Bs' \equiv \Bs\).
Since \(\exp(\Bs') \in \Bsig\), by Lemma \ref{monotone}, \(\As
\equiv \exp(\Bs')\).
If \(\As = \Bs+\Cs\), then by our induction hypothesis, there are
\(\Bs',\Cs' \in \Bsig\) such
that \(\Bs' \equiv \Bs\) and \(\Cs' \equiv \Cs\).
Again, \(\Bs'+\Cs' \in \Bsig\) and \(\As = \Bs+\Cs \equiv \Bs' + \Cs'\).
If \(\As\) is not of these three forms, then Lemma \ref{decomp_lem}
implies
that there exist \(\Bs,\Cs \in \Ssig\) such that
\(\As \equiv \Bs*\exp(\Cs)\) and the cardinality of
\(\Bs*\exp(\Cs)\) is the same as that of \(\As\).
As above, by induction hypothesis, we may assume that \(\Bs\) and
\(\Cs\) are both in \(\Bsig\).
If \(\Bs = \exp(\Xs)\) for some \(\Xs\) (which would necessarily be
in \(\Bsig\)),
then \(\Bs*\exp(\Cs) = \exp(\Xs) * \exp(\Cs) = \exp(\Xs+\Cs)\),
which is in \(\Bsig\).
Thus we may assume that \(\Bs = \sum_{i=1}^n \exp(\Xs_i)\) where each
\(\Xs_i\) is in \(\Bsig\).
Let \(i \leq n\) be such that \(\rho(\Xs_i)\) is maximized.
By Lemmas \ref{R_order_char} and \ref{block_to_R},
we have that \(\Xs_j \leq \Xs_i\) for all \(j \neq n\).
By Lemma \ref{duplication},
\[
\big(\sum_{i=1}^n \exp(\Xs_i)\big) * \exp(\Cs) \equiv \exp(\Xs_j) *
\exp(\Cs) = \exp(\Xs_j + \Cs)\in \Bsig.
\]
\end{proof}
We are now in a position to extend the definition of \(\rho\) to \(\Ssig\):
if \(\As \in \Ssig\) set
\(\rho(\As) := \rho(\Bs)\) where \(\Bs \in \Rsig\) and \(\As \equiv \Bs\).
Such a \(\Bs\) exists by Lemmas \ref{standard_to_block} and
\ref{block_to_R} and
is unique by Lemma \ref{R_order_char}.
Lemma \ref{R_order_char} now immediately extends to all of \(\Ssig\).
\begin{lem} \label{S_order_char}
If \(\As\) and \(\Bs\) are indecomposable elements of \(\Ssig\), then the following
are equivalent:
\begin{enumerate}
\item \label{S_reduction}
\(\As \leq \Bs\).
\item \label{S_rank}
\(\rho(\As) \leq \rho(\Bs)\).
\end{enumerate}
\end{lem}
If \(\As \ne 0\) is in \(\Ssig\), let \(\As^-\) be the result of
removing the top element of
\(\As\).
\begin{lem} \label{decrease_rank}
If \(\As \ne 0\) is in \(\Ssig\),
then \(\rho(\As^-) < \rho(\As)\).
\end{lem}
\begin{proof}
We first show that \(\As^- + \one \leq \As\).
This is proved by induction on the complexity of \(\As\).
If \(\As = \As^- + \one\), there is nothing to show.
Let \(n = |\As| -1\) and fix an \(i < n\) such that \(\As(i,n) > 0\).
Define \(\Bs\) to be the restriction of \(\infl{\As}{n}\) to
\(\{0,\ldots,n-1,i^n\}\), noting that \(\Bs^- = \As^-\).
We now have that \(\Bs(i,i^n) = \As(i,n)-1\) and for
all \(j < n\),
\[
\Bs(j,i^n) = \min \big(\As(i,n)-1,\As(j,n)\big) \leq \As(j,n).
\]
Thus \(\Bs \leq \As\) and \(\Bs\) has smaller complexity than \(\As\).
By our induction hypothesis \(\As^- = \Bs^- \leq \As^- + \one =
\Bs^- + \one \leq \Bs \leq \As\).
Now suppose that \(\rho(\As^-) \geq \rho(\As)\).
Since \(\As^-+\one \leq \As\),
\[
\rho(\As^-) \leq \rho(\As^- + \one) \leq \rho(\As) \le \rho(\As^-)
\]
and it must be that \(\rho(\As^-) = \rho(\As^- + \one) = \rho(\As)\).
It follows from Lemma \ref{S_order_char} that \(E(\As^-) \equiv
E(\As^- + \one)\) and that
\(\gen{E(\As^-)}\) and \(\gen{E(\As^- + \one)}\) are in the same
biembeddability class.
This contradicts Lemma \ref{proper_successor}.
\end{proof}
\subsection{EA-class calculations}
If \(\xi < \epsilon_0\), then \(\Rs_\xi \in \Rsig\) is the
unique element with \(\rho(\Rs_\xi) = \xi\) and
\(G_\xi :=\gen{\Rs_\xi}\);
these are just restatements of the definitions of \(\Rs_\xi\) and \(G_\xi\) given
in Section \ref{BasicOp:sec} which made only implicit reference to \(\rho\).
With these definitions, it is immediate that \(G_{\xi+1} \cong G_\xi
+ \Z\) and by
Proposition \ref{order_char_prop}, \(G_\xi\) embeds into \(G_\eta\)
if and only if \(\xi \leq \eta\).
We now verify the EA-class calculations asserted in Theorem
\ref{main_thm}.
Observe that if \(\xi < \eta < \epsilon_0\), then \(\EA(G_\xi) \leq
\EA(G_\eta)\).
We first note the following lemma.
\begin{lem} \label{EA_calc_lem}
If \(A \in \Sgen\) and \(|A| > 1\), then
\[
\sup_{\Bs < \As} \EA(\gen{B}) \leq \EA(\gen{A}) \leq \big( \sup_{\Bs
< \As} \EA(\gen{B}) \big) + 2
\]
and \(\gen{A}\) has property \(\Sigma\).
In particular, if \(\sup_{\Bs < \As} \EA(\gen{B})\) is a limit ordinal,
then \(\EA(\gen{A}) = \big( \sup_{\Bs < \As} \EA(\gen{A}) \big) +2\).
\end{lem}
\begin{proof}
Since \(\Bs < \As\) implies \(\gen{B}\) embeds into \(\gen{A}\),
the first inequality follows from the monotonicity of EA-class.
To see the second inequality, define for each
\(k \in \omega\)
\[
B_k := \{a_i^{{a_{\max}}^p} \mid i < |A|-1 \textrm{ and } 0 \leq p
\leq 2k\}.
\]
%
As noted in the beginning of Section \ref{inflation:sec}, if \(A_k\) is obtained from
\(A\) by iteratively inflating by \(a_{\max}\), then \(B_k \leq A_k^-\).
In particular by Lemma \ref{decrease_rank},
\(\rho(\Bs_k) < \rho(\As)\).
By Lemma \ref{S_order_char}, \(\Bs_k < \As\).
Setting
\[
N:= \bigcup_{k=0}^\infty \gen{a_i^{a_{\max}^p} \mid i < |A|-1
\textrm{ and } -k \leq p \leq k }
\]
we have that \(\gen{A}\) is an extension of \(N\) by \(\Z\) and
\(N\) is an increasing union of groups isomorphic to ones of the
form \(\gen{B_k}\).
Thus
\[
\EA(\gen{A}) \leq \EA(N) +1 \leq \big( \sup_k \EA(\gen{B_k}) \big) +
2 \leq \big( \sup_{\Bs < \As} \EA(\gen{B}) \big ) + 2
\]
as desired.
The above argument in particular shows that every
\(\Sgen\)-generated group is in the smallest class
that contains the abelian groups and is closed under the elementary
operations of
\emph{extensions} and \emph{directed unions}.
Since the class of groups which has property \(\Sigma\) includes
this class, it follows that every \(\Sgen\)-generated
group has \(\Sigma\).
\end{proof}
\begin{lem}
If \(\xi = \omega^{(\omega^\alpha) \cdot ({2^n})}\) for
\(0 \le \alpha < \epsilon_0\) and \(n < \omega\),
then \(\EA(G_\xi) = \omega \cdot \alpha + n + 2\) if \(\alpha>0\)
and \(\EA(G_\xi) = n+1\) if \(\alpha=0\).
\end{lem}
\begin{proof}
Define
\[
\Xi:= \{ \omega^{(\omega^\alpha) \cdot ({2^n})} \mid 0 < \alpha <
\epsilon_0 \textrm{ and } n < \omega\}
\]
and set \(\theta (\omega^{(\omega^\alpha) \cdot (2^n)} ) = \omega
\cdot \alpha + n + 2\).
We verify by induction on \(\xi\)
that \(\EA(G_\xi) = \theta(\xi)\), which is what is asserted
in the first conclusion.
The least element of \(\Xi\) is \(\omega^\omega\).
In this case, \(\Rs_{\omega^\omega} = E (\one + \one)\)
is the signature of a standard pair with oscillation 2 and
\(G_{\omega^\omega}\) is the Brin-Navas group, which has EA-class
\(\omega+2 = \theta(\omega^\omega)\).
Next observe that if \(\omega^{(\omega^\alpha) \cdot (2^n)}\) is in
\(\Xi\), then the next element of \(\Xi\)
is \(\omega^{(\omega^\alpha) \cdot (2^{n+1})}\).
In particular, \(\xi \in \Xi\) is a limit point of
\(\Xi\) precisely when \(\xi = \omega^{\omega^\alpha}\)
for some \(\alpha > 1\).
If \(\xi \in \Xi\) is of the form \(\omega^{(\omega^\alpha)
\cdot (2^{n+1})}\),
then setting \(\xi' = \omega^{(\omega^\alpha) \cdot (2^n)}\) we have
\(\Rs_\xi = \Rs_{\xi'} * \Rs_{\xi'}\).
By Proposition \ref{wreath_prop}, \(G_\xi = G_{\xi'}
\wr G_{\xi'}\).
By Proposition \ref{EAGWrG} and our induction hypothesis,
\[
\EA(G_\xi) = \EA(G_{\xi'}) + 1 = \theta(\xi') +
1 = \theta (\xi).
\]
Now assume that \(\xi \in \Xi\) is a limit point
of \(\Xi\).
In what follows \(\xi'\) always represents an element of
\(\Xi\).
We first claim that
\begin{equation}\label{ThetaContinuous}
\theta (\xi) = \big (\sup_{\xi' < \xi} \theta (\xi') \big )+ 2.
\end{equation}
If \(\xi = \omega^{\omega^{\alpha + 1}}\), then
(\ref{ThetaContinuous}) follows from
the fact that
\(\xi = \sup_n \omega^{(\omega^\alpha) \cdot (2^n)}\) and
consequently that
%
\[
\begin{split}
\theta(\xi) = &\omega \cdot (\alpha + 1) + 2 = \omega \cdot \alpha +
\omega + 2 \\
= &\big ( \sup_n \omega \cdot \alpha + n \big) + 2 =
\big ( \sup_n \theta (\omega^{(\omega^\alpha) \cdot (2^n)}) \big ) + 2.
\end{split}
\]
%
If \(\xi = \omega^{\omega^\alpha}\) for a limit ordinal \(\alpha\),
then (\ref{ThetaContinuous}) follows
from the continuity of the maps \(\alpha \mapsto \omega^{\omega^\alpha}\)
and \(\alpha \mapsto \omega \cdot \alpha\).
Observe that in both cases \(\sup_{\xi' < \xi}
\theta(\xi') = \omega \cdot \alpha\) is a limit ordinal.
Now observe that by Lemma \ref{EA_calc_lem} and our induction hypothesis
\[
\EA(G_\lambda) = \big(\sup_{\xi' < \xi} \EA(G_{\xi'}) \big) + 2 =
\big (\sup_{\xi' < \xi} \theta
(\xi') \big )+ 2
= \theta (\xi).
\]
The first equality holds since \(\sup_{\xi' < \xi}
\EA(G_{\xi'}) = \sup_{\xi' < \xi} \theta (\xi')\)
is a limit ordinal.
If \(\alpha=n=0\), then \(\xi=\omega\), \(\Rs_\omega=\exp(\one+\one)
= \one * \one\), \(G_\omega = \Z\wr\Z\), and \(EA(G_\omega)=1\).
The last conclusion follows from Lemma \ref{EA_calc_lem} and
arguments similar to those above.
\end{proof}
We have established the following proposition.
\begin{prop} \label{F-less}
For each \(\xi < \epsilon_0\),
\(G_\xi\) is elementary amenable and in particular does not
contain a copy of Thompson's group \(F\).
\end{prop}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The embeddability relation on \(\Sgen\)-generated groups}
\label{main:sec}
We now complete the proof of Theorem \ref{main_thm}.
We have seen that that if \(\As\) and \(\Bs\) are indecomposable
elements of \(\Rsig\),
then \(\gen{\As}\) embeds into \(\gen{\Bs}\) if and only if
\(\rho(\As) \leq \rho(\Bs)\).
Moreover, we have seen that if \(\As\) is an indecomposable element
of \(\Ssig\), then there is a unique \(\Bs\)
in \(\Rsig\) such that \(\gen{\As}\) is biembeddable with \(\gen{\Bs}\).
To prove Theorem \ref{main_thm}, we must extend our analysis to
groups generated by decomposable elements of \(\Rsig\) and \(\Ssig\).
The following lemma is used in this section and the next.
\begin{lem}\label{FullWidth}
Let \(H\) be generated by a finite geometrically fast
system \(S\) of functions with \(|S|>1\) and a \(\sqsubset\)-maximum element \(h\) in that
\(g \sqsubset h\) for every \(g \in S \setminus \{h\}\).
Let \(N{}\) be the normal closure of \(S\setminus\{h\}\).
Then the following hold:
\begin{enumerate}
\item \label{H_emb}
There there is an embedding of \(H\) into \(\PLoI\) so that
the image of \(S\) has a \(\sqsubset\)-maximum element and so that every
element of \(H\setminus N{}\) has extended support identical to that of the image of \(h\).
\item \label{dont_commute}
For each \(f\in N{}\setminus\{1\}\) and each \(g\notin N{}\), we
have \([f,g]\ne1\).
Thus the center of \(H\) is trivial.
\item \label{central_cyclic}
The centralizer of every element of \(H\setminus N{}\) is
cyclic.
\end{enumerate}
\end{lem}
\begin{proof} In this proof, \(\gen2\) is the multiplicative
subgroup of \(\R\) generated by \(2\), and \(3\gen2\) is the coset
containing \(3\). We first adjust the elements of \(S\)
slightly so that after the adjustment it still satisfies the hypotheses,
the isomorphism class of \(\gen{S}\) has not changed,
and each element of \(S\) is piecewise linear
with all slopes used by \(h\) coming from \(3\gen 2\) and all
slopes used by elements of \(S\setminus\{h\}\)
coming from \(\gen 2\).
We can do this by keeping all transition points the same
and changing each bump so that its graph is two affine pieces. If
the pieces have slopes sufficiently close to 0 or \(+\infty\), as
appropriate, then the feet of each new bump are small enough to
be contained in the feet of the bump it replaces. The
dynamical diagram is unchanged and the group generated
isomorphic to the original.
All elements of \(H\) are of the form \(f=uh^i\) with \(u\in N{}\) a
product of elements of \(C=\{g^{h^j} \mid g\in S\setminus\{h\}, j\in
\Z\}\). It follows from the chain rule and the fact that \(h\) is
the identity on no open interval that every element of \(C\) has
slopes restricted to \(\langle 2\rangle\). The element \(f=uh^i\)
is outside \(N{}\) if and only if \(i\ne0\). Thus every element
\(f\) of \(H\setminus N{}\) has slope in \(3^i\langle 2\rangle\)
everywhere the slope of \(f\) is defined.
Hence every \(f\in H\setminus N{}\) has extended support equal to
that of \(h\), proving (\ref{H_emb}).
With \(f\in N{}\setminus \{1\}\) and \(g\in H\setminus N{}\), the
only way to have \([f,g]=1\) is for the orbitals of \(f\) to be
among the orbitals of \(g\). It then follows from Theorem 4.18 of
\cite{picric} that on each orbital \(J\) of \(g\), there would be a
piecewise linear bump \(b\) with support \(J\) with \(f|_J\) and \(g|_J\) each a
power of \(b|_J\). This is not possible because 3 to a nonzero,
rational power is never equal to such a power of 2, proving the
first part of (\ref{dont_commute}). The center of \(H\) is trivial because there is
no room for a nontrivial central element.
To prove (\ref{central_cyclic}), we add to the information in the
paragraph above. For \(g\in H\setminus N{}\) and \([f,g]=1\), we
now know \(f\notin N{}\), and the extended supports of \(f\) and
\(g\) equal that of \(h\). Thus the orbitals of \(f\) and \(g\) are
identical. Again, Theorem 4.18 of \cite{picric} makes each of
\(f\) and \(g\) powers of a common root on each orbital. By Theorem
4.15 of \cite{picric}, for each orbital \(J\) of \(g\) there is
a unique minimum root \(r_J\) so that all roots of \(f\) and \(g\)
on \(J\) are integral powers of \(r_J\). For each orbital \(J\) of
\(g\), let \(m_J\) and \(n_J\) be the integers so that
\(f|_J=r_J^{m_J}\) and \(g|_J=r_J^{n_J}\).
Assume by way of contradiction that there are orbitals \(J\ne K\) of
\(f\) for which \(m_J/n_J \ne m_K/n_K\). Now
%
\[
\begin{split}
\left(f|_J\right)^{n_J} &= r_J^{(m_J n_J)} = \left(g|_J\right)^{m_J},
\quad\mathrm{while} \\
\left(f|_K\right)^{n_J} &= r_K^{(m_K n_J)} \ne r_K^{(m_J n_K)} =
\left(g|_K\right)^{m_J}.
\end{split}
\]
%
Thus \(a=f^{n_J}g^{-m_J}\) is the identity on \(J\) and not on
\(K\). From (\ref{H_emb}) we have \(a\in N{}\). But this
contradicts the fact that \(a\) commutes with \(g\notin N{}\) which
by (\ref{dont_commute}) means that \(a\notin N{}\). Thus \(m_J/n_J
= m_K/n_K\) for all orbitals \(J\) and \(K\) of \(g\). Thus \(f\)
is determined by \(g\) and by the restriction of \(f\) to one
particular orbital of \(g\). By Theorem 5.5 of \cite{picric}, if
\(J\) is an orbital of \(g\), then the restriction of the
centralizer of \(g\) to \(J\) is cyclic. Thus the centralizer of
\(g\) in \(H\) is cyclic. \end{proof}
\begin{lem}\label{RsigEmbedOrder}
If \(\As,\Bs \in \Rsig\) and \(\rho(\As) < \rho(\Bs)\), then
\(\gen{\Bs}\) does not embed into \(\gen{\As}\).
\end{lem}
\begin{proof} Assume the lemma is false, and
let \((\beta, \alpha)\) be the lexicographically minimal pair
for which there are \(\As\) and \(\Bs\) in \(\Rsig\) so that
\(\rho(\As)=\alpha<\beta=\rho(\Bs)\) and \(\gen\Bs\) embeds in
\(\gen\As\). Since we know that \(\gen\Cs\) embeds in \(\gen\Bs\)
whenever \(\Cs\in \Rsig\) has \(\rho(\Cs)<\rho(\Bs)\), our choice of
\(\alpha\) and \(\beta\) must have \(\beta=\alpha+1\) and
\(\gen\As\) embeds in no \(\gen\Ds\) with \(\Ds\in \Rsig\) and
\(\rho(\Ds)<\rho(\As)\).
The result holds if \(\alpha\) is finite since then
\(\gen\As=\Z^\alpha\) and \(\gen\Bs=\Z^{\alpha+1}\) and an embedding
of \(\gen\Bs\) in \(\gen\As\) is not possible. We thus assume
that \(\alpha\) is infinite with Cantor normal form
%
\begin{equation}\label{AlphaForm}
\alpha=\omega^{\alpha_0}+\omega^{\alpha_1}+\cdots +
\omega^{\alpha_k} + n
\end{equation}
with \(\alpha_0\ge \alpha_1 \ge \cdots \ge \alpha_k>0\) and
\(n\ge0\).
We thus have
%
\[
\begin{split}
\As &= \As_0+\As_1+\cdots +\As_k+\one\cdot n,
\qquad\mathrm{and} \\
\Bs &= \As_0+\As_1+\cdots +\As_k+\one\cdot(n+1),
\end{split}
\]
%
where, for \(0\le i\le k\), each \(\As_i\) is in \(\Rsig\) and
indecomposable with \(|\As_i|>1\) and \(\rho(\As_i) =
\omega^{\alpha_i}\).
With \(G_i=\gen{\As_i}\), we have \(\gen\As\) representable as \(G_0
+ G_1 + \cdots + G_k + \Z^n\) and \(\gen\Bs\) as \(G_0+ G_1 + \cdots
+ G_k + \Z^{n+1}\). We assume a homomorphic embedding
\(\phi:\gen\Bs \rightarrow \gen\As\). For \(0\le i\le k\), we let
\(\pi_i:\gen\As \rightarrow G_i\) be the projection homomorphism,
and let \(\phi_i=\pi_i\circ \phi:\gen\Bs\rightarrow G_i\).
If for all \(i\) with \(0\le i\le k\), we have \(\phi_i(\Z^{n+1})\)
trivial, then \(\phi\) embeds \(\Z^{n+1}\) into \(\Z^n\) which is
not possible. Thus for some \(i\) with \(0\le i\le k\) and some
element \(x\in \Z^{n+1}\), we have that \(y=\phi_i(x)\)
is a nonidentity element of \(G_i\). Let \(A_i\in \Sgen\) have
signature \(\As_i\), let \(h_i\) be the maximum
element of \(A_i\), and let \(N_i\) be the normal closure in
\(G_i=\gen{A_i}\) of \(A_i\setminus\{h_i\}\).
We note that for all \(g\in \Bs\), we have \([g,x]=1\). Thus
\([\phi_i(g), y]=1\) in \(G_i\).
If \(y\notin N_i\), then by Lemma
\ref{FullWidth}(\ref{central_cyclic}) the centralizer \(C_i\) of
\(y\) in \(G_i\) is cyclic. From this, we have \(\phi_i(g)\in
C_i\). This puts the image of \(\phi\) in
%
\[
\gen{\As_0} + \cdots \gen{\As_{i-1}} + C_i + \gen{\As_{i+1}} +
\cdots \gen{\As_k} + \Z^n
\]
%
which is isomorphic to the group generated by
%
\[
\Es = \As_0 + \cdots + \As_{i-1} + \As_{i+1} + \cdots + \As_k +
\one\cdot(n+1)
\]
%
which is in \(\Rsig\). Since (\ref{AlphaForm}) is in normal form
and since \(\rho(\As_i)>\omega\), we have \(\rho(\Es)<\rho(\As) =
\alpha\). This contradicts our initial choice of \(\alpha\).
If \(y\in N_i\), then by Lemma \ref{FullWidth}(\ref{dont_commute})
the centralizer \(C_i\) of \(y\) in \(G_i\) is contained in
\(N_i\), and \(\phi_i(g)\) is in \(N_i\). This puts the image of
\(\phi\) in
%
\[
\gen{\As_0} + \cdots \gen{\As_{i-1}} + N_i + \gen{\As_{i+1}} +
\cdots \gen{\As_k} + \Z^n.
\]
%
Since \(\gen\Bs\) is finitely generated, we can replace \(N_i\) in
the above by a suitable finitely generated subgroup. There is an
\(l\) so that \(\phi_i(\gen\Bs)\) is contained in \(\gen{K_l}\)
where
%
\[
K_l := \{h^{{m_i}^j} \mid h\in \As_i,\, h<m_i,\, \mathrm{and}\,
-l\le j\le l\}.
\]
%
If \(K'_l\) is similarly defined with \(-l\le j\le l\) replaced
by \(0\le j\le 2k\), then \(\gen{K_l}\) is isomorphic to
\(\gen{K'_l}\). But \(K'_l\le K''_l\) where \(K''_l\) is an
appropriately chosen iterated inflation of \(A_i\). By Lemma
\ref{inflate_biembedd}, \(K''_l\) and thus also \(K'_l\) are in
\(\Sgen\) and \(\rho(K''_l) = \rho(\As_i)\). By Lemma
\ref{decrease_rank}, \(\rho(K'_l)<\rho(K''_l) = \rho(\As_i)\).
We now have that \(\gen{\Bs}\) embeds in
%
\[
\gen{\As_0} + \cdots \gen{\As_{i-1}} + \gen{K'_l} + \gen{\As_{i+1}} +
\cdots \gen{\As_k} + \Z^n.
\]
%
Once again, since (\ref{AlphaForm}) is in normal form, we get that
\(\rho(\Fs)< \rho(\As)=\alpha\). Again, this contradicts our choice
of \(\alpha\). This completes the proof.
\end{proof}
We are finally in a position to complete the proof of Theorem
\ref{main_thm}.
Observe that if \(\As\) and \(\Bs\) are in \(\Ssig\), then
\[
\gen{\As + \Bs} = \gen{\As} + \gen{\Bs} \cong \gen{\Bs} + \gen{\As}
= \gen{\As+\Bs}.
\]
On the other hand, the symbolic manipulations which define \(\leq\)
on \(\Ssig\)
do not typically yield \(\As+\Bs \leq \Bs+\As\).
We deal with this as follows.
Define \(\Ssig'\) to be all \(\As\) in \(\Ssig\) such that
\(\As = \sum_{i<n} \As_i\) where \(\As_i\) is indecomposable and
if \(\rho(\As_{i+1}) \leq \rho(\As_i)\) for all \(i < n\).
Clearly, if \(\As\) is in \(\Ssig\), then \(\gen{\As}\) is
isomorphic to \(\gen{\Bs}\) for some
\(\Bs \in \Ssig'\).
Putting this together with Lemma \ref{RsigEmbedOrder}
gives the following
proposition.
\begin{prop} \label{order_char_prop}
If \(\As\) and \(\Bs\) are in \(\Ssig'\), then the following are
equivalent:
\begin{enumerate}
\item \label{reduction} \(\As \leq \Bs\)
\item \label{embedding} \(\gen{\As}\) embeds into \(\gen{\Bs}\)
\item \label{lower_rank} \(\rho(\As) \leq \rho(\Bs)\).
\end{enumerate}
Moreover, for each \(\As\) in \(\Ssig'\), there is a unique element
\(\Bs\) of \(\Rsig\) such that
\(\gen{\As}\) is biembeddable with \(\Bs\).
\end{prop}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{\((\Ffrak,\emb)\) is not linear}
\label{nonlinear:sec}
We conclude this paper by showing that the class of those subgroups of \(F\)
admitting a finite geometrically fast generating set is not linearly
ordered by the embeddability relation.
Consider the groups \(B + \Z\) and \(G\) with
geometrically fast generating sets specified by
the dynamical diagrams in Figures \ref{BZ_fig} and \ref{G_fig} below.
We apply Lemma \ref{FullWidth}
embedding \(B+\Z\) and \(G\) in \(\PLoI\)
so that \(\{a,b\}\) and \(\{f,g,h\}\)
satisfy (\ref{H_emb}) of the Lemma.
Notice that \(\gen{a,b}\) is
the Brin-Navas group \(B\);
\(c\) generates a copy of \(\Z\) which commutes with the elements of
\(B\).
\begin{figure}[b]
\[
\xy
(24,10); (48,10)**\crv{(30,6)&(42,6)}; (23,13)*{a};
(48,10); (72,10)**\crv{(54,14)&(66,14)};
(36,0); (60,0)**\crv{(42,4)&(54,4)}; (34,3)*{b};
(84,0); (108,0)**\crv{(90,4)&(102,4)}; (84,3)*{c};
\endxy
\]
\caption{A diagram for \(B + \Z\)} \label{BZ_fig}
\end{figure}
\begin{figure}[b]
\[
\xy
(0,15); (24,15)**\crv{(6,11)&(18,11)}; (-1,18)*{f};
(24,15); (48,15)**\crv{(30,11)&(42,11)};
(48,15); (72,15)**\crv{(54,19)&(66,19)};
(72,15); (96,15)**\crv{(78,19)&(90,19)};
(36,0); (60,0)**\crv{(42,4)&(54,4)}; (34,2)*{h};
(12,0); (84,0)**\crv{(30,10)&(66,10)}; (10,2)*{g};
\endxy
\]
\caption{The group \(G = \gen{f,g,h}\)}\label{G_fig}
\end{figure}
\begin{thm}\label{NonEmbed} With \(G\) and \(B + \Z\) as above,
the following are true:
%
\begin{enumerate}
\item \label{EA-calc}
\(\EA(G)=\EA(B + \Z) = \omega+2\).
\item \label{BZ_nonemb}
There is no embedding of \(B + \Z\) into \(G\).
\item \label{nonemb_in_BZ}
There is no embedding of \(G\) into \(B + \Z\).
\end{enumerate}
\end{thm}
\begin{proof}
We begin by verifying (\ref{EA-calc}).
Define \(N_G\) to be the normal closure of \(\{g,h\}\) in \(G\) and
\(N_B\) to be the normal closure of \(b\) in \(B\).
The group \(N_G\) is generated by \(C=\{h^{f^i}, g^{f^i}\mid i\in \Z\}\).
If \(C_n = \{g^{f^i}, h^{f^i} \mid -n \leq i \leq n\}\),
then Theorem \ref{EAGWrG} implies \(\sup_n \EA(\gen{C_n}) = \omega\) and
hence that \(\EA (N_G) = \omega+1\).
Since \(G\) is finitely generated and \(G/N_G\cong \Z\), we have
\(\EA(G)=\omega+2\).
Similarly, the group \(N_B + \Z\) is generated by \(D=\{b^{a^i}\mid
i\in \Z\} \cup \{c\}\).
An analogous computation shows that \(\EA(N_B + \Z) = \omega+1\) and that
\(\EA(B + \Z) = \omega+2\).
Next we turn to (\ref{BZ_nonemb}).
Assume there is an embedding \(B + \Z\) into \(G\).
The centralizer of \(c\) in \(B + \Z\) includes \(B\) and is not abelian.
By (\ref{central_cyclic}) of Lemma \ref{FullWidth} the image of
\(c\) must be in \(N_G\).
By (\ref{dont_commute}) of Lemma \ref{FullWidth} the image of every
element of \(B\) must be in
\(N_G\) since otherwise that image would not commute with the image
of \(c\).
Thus an embedding of \(B + \Z\) into \(G\) has its image in \(N_G\).
But \(\EA(N_G)=\omega+1\) and \(\EA(B + \Z)=\omega+2\) so this is
not possible.
We will verify (\ref{nonemb_in_BZ}) through a series of claims.
\begin{claim}\label{GBZClaim}
If \(G\) embeds in \(B + \Z\), then
\(G\) embeds in \(B\).
\end{claim}
\begin{proof} If an embedding exists it can be composed with the
projection to \(B\).
The kernel of this projection consists of all \((1,c^k)\) in \(B + \Z\).
These are all in the center of
\(B + \Z\) and if the image of the embedding intersected this kernel,
then the image would have nontrivial center.
But by (\ref{dont_commute}) of Lemma \ref{FullWidth},
\(G\) has trivial center.
Thus the composition of the embedding with the projection is one-to-one.
\end{proof}
Next we introduce some tools used in \cite{ATaylor}.
Define the following predicates where the variables are intended to range over
elements of \(B + \Z\) and \(G\):
\[
\begin{split}
C(x,y) &:=\formula{[x,y]=1}, \\
D(x,y) &:= \formula{(\neg C(x,y)) \land C(x,x^y)}, \\
T(x,y,z) &:= \formula{D(x,y)\land D(x,z) \land D(y,z) \land C(x,y^z)}.
\end{split}
\]
We think of \(D(x,y)\) as saying that \(y\) ``dominates'' \(x\)
in that a typical pair that satisfies this is a fast pair of nested
one bump functions with the orbital of \(y\) as the larger of the two.
We think of \(T(x,y,z)\) as saying that \((x,y,z)\) forms
a ``tower'' in that a typical triple that satisfies \(T\) is a fast
triple of nested one bump functions with \(z\) the largest and \(x\)
the smallest.
While these are typical, they are not the only examples and we need
to know a little more about the functions that satisfy these
predicates.
\begin{claim}\label{DomClaim} If \(x,y \in F\) and
\(D(x,y)\) holds, then:
%
\begin{enumerate}
%
\item \label{J_conclusion}
For each orbital \(J\) of \(x\) one of the following holds:
\begin{enumerate}
\item \label{J_disjoint}
\(J\) is disjoint from
all orbitals of \(y\),
\item \label{J_coincide}
\(J\) equals an orbital of \(y\) with
\([x |_J, y |_J]=1\), or
\item \label{J_inside}
\(J y\) is disjoint from \(J\).
\end{enumerate}
\item \label{orbital_conclusion}
There is an orbital \(J\) of \(x\) such that \(Jy\) is disjoint from \(J\).
\end{enumerate}
\end{claim}
\begin{proof}
Toward proving (\ref{J_conclusion}), fix an orbital \(J\) of \(x\).
Observe that \(Jy\) is an orbital of \(x^y\).
If \(Jy\) intersects \(J\) but differs from \(J\),
then \(x\) and \(x^y\) can't commute.
If \(Jy\) is disjoint from \(J\), then (\ref{J_inside}) holds.
Now suppose \(Jy = J\).
The chain rule implies that \(x\) and \(x^y\) agree at the endpoints of \(J\).
If \(x |_J \ne x^y |_J\), then Theorem 4.18 of \cite{picric} implies \(x^y |_J\) cannot commute with
\(x |_J\), contradicting \(D(x,y)\).
It follows that \(x |_J\) commutes with \(y |_J\).
Moreover either \(y |_J\) is the identity
or else \(J\) is an orbital of \(y\).
Thus if \(Jy = J\) then either (\ref{J_disjoint}) or (\ref{J_coincide})
hold.
Finally observe that if Conclusion (\ref{orbital_conclusion}) fails, then \([x,y]=1\) by
Conclusion (\ref{J_conclusion}).
This would contradict \(D(x,y)\).
\end{proof}
\begin{claim}\label{TowerClaim}
If \(x, y, z \in F\)
and \(T(x,y,z)\) holds, then if \(J\) is an orbital of \(x\) properly
contained in an orbital \(K\) of \(y\), then there is an orbital
\(L\) of \(z\) that properly contains \(K\).
\end{claim}
\begin{proof}
By Claim \ref{DomClaim} applied to \(x\) and \(y\),
we have that \(Jy\) is disjoint from \(J\).
In particular, \(x |_K\) does not commute with \(y |_K\).
By Claim \ref{DomClaim} applied to \(y\), \(z\), and the orbital \(K\), we have
that either \(K\) is an orbital of \(z\) and \(y^z |_K = y |_K\) or else
\(Kz\) is disjoint from \(K\).
The former is impossible since it implies \(\neg C(x |_K , y^z |_K)\) which is contrary to \(C(x,y^z)\).
It follows that \(Kz\) is disjoint from \(K\).
Thus any orbital \(L\) of \(z\) which intersects \(K\) contains all of \(K \cup Kz\)
and hence properly contains \(K\).
\end{proof}
\begin{claim}\label{NoDomAtTop}
Let \(x\) and \(y\) be in \(B \setminus N_B\).
Then \(D(x,y)\) is false.
\end{claim}
\begin{proof}
Since \(\{a,b\}\) were chosen using (\ref{H_emb}) of Lemma \ref{FullWidth},
we know that the extended supports of \(x\) and \(y\) are connected
and identical.
If \(D(x,y)\) held, then \([x,y]\ne1\) implying that \(x^y\ne x\).
Thus \(x^y\) has connected
extended support equal to that of \(x\).
If \([x^y,x]=1\), then \(x^y\) and \(x\) have identical orbitals.
Thus \(y\) fixes all the transition points of \(x\).
This implies that the derivatives of \(x\) and \(x^y\) agree near the
ends of each orbital of \(y\).
But commuting bumps on the same
orbital on which the slopes agree near the ends of the orbital must
be identical bumps by Theorem 4.18 of \cite{picric}. Thus
\(x^y=x\) and \(D(x,y)\) cannot hold.
\end{proof}
We return to the task of proving (\ref{nonemb_in_BZ}) of Theorem \ref{NonEmbed}.
By Claim \ref{GBZClaim} it suffices to show that \(G\) does not embed in \(B\).
Suppose for contradiction that there is an embedding \(\phi : G \to B\).
Recall that \(G\) is generated by \(f < g < h\) as illustrated in Figure \ref{G_fig}.
\begin{claim}\label{GInBZLoc}
The \(\phi\)-image of \(N_G\) is contained in \(N_B\).
\end{claim}
\begin{proof}
Observe that if \(x\) is in \(B \setminus N_B\), then for all \(z\in B\), \(D(x,x^z)\) is false by Claim \ref{NoDomAtTop}:
since \(N_B\) is normal \(x^z\notin N_B\).
Since both \(D(h,h^f)\) and \(D(g,g^f)\) are true,
we have \(\phi(g) \in N_B\) and \(\phi(h) \in N_B\).
The conclusion of the claim now follows from the fact
that \(N_G\) is the normal closure of \(\{f,g\}\) in \(G\) and
\(N_B\) is normal in \(B\).
\end{proof}
For each \(i\), define \(f_i := f^{h^{-i}}\).
Observe that
\[
\supt (f_0) \subsetneq \supt (f_1) \subsetneq \supt (f_2) \subsetneq \cdots \subsetneq \supt (g).
\]
Define \(A:=\{f_i\mid i\ge0\}\cup\{g\}\) and note that all elements
of \(A\) are in \(N_G\) and must have \(\phi\)-images in \(N_B\).
For any triple \(x < y < z\) from \(A\), we have \(T(x,y,z)\) and hence \(T(\phi(x),\phi(y),\phi(z))\).
It follows from (\ref{orbital_conclusion}) of Claim \ref{DomClaim}
and Claim \ref{TowerClaim} that there are intervals
%
\(
I_0\subsetneq I_1\subsetneq I_2 \subsetneq \cdots \subsetneq J
\)
%
where each \(I_i\) is an orbital of \(\phi(f_i)\) and \(J\) is an
orbital of \(\phi(g)\).
However the orbitals used in \(N_B\) come in families \(\Jscr_n\)
indexed over \(\Z\) with each orbital in family \(\Jscr_n\) contained in
an orbital in family \(\Jscr_{n+1}\).
Since orbital \(I_0\) of \(\phi(f_0)\)
must come from some \(\Jscr_m\) and orbital \(J\) of \(\phi(g)\) must come
from some \(\Jscr_n\) with \(m<n\), there are only finitely many
different orbitals available in \(N_B\) between \(I_0\) and \(J\).
This contradicts the assumption that there is an embedding of \(N_G\) into \(N_B\).
Since this was shown to follow from the existence of an embedding
of \(G\) into \(B+\Z\), we have completed our proof of (\ref{nonemb_in_BZ}).
\end{proof}
\begin{remark}
Theorem \ref{NonEmbed} does not provide a counterexample to Conjecture \ref{univ_conj}.
If we let \(a=g_4\) and \(b=f_4\) be the generators of \(G_{\tau_4}\) as shown in
Figure \ref{Gt4Gt5_fig}, then the reader can check that \(A=\{a^2, b^{a^{-1}b^{-1}},
b^{ab}\}\) is fast and that after extraneous bumps are excised
from \(A\), the dynamical diagram for \(A\) is identical to that
\(\{f,g, h\}\), the generating set for \(G\).
Thus \(G\) embeds in \(G_{\tau_4}\).
\end{remark}
\def\cprime{$'$} \def\cprime{$'$} \def\cprime{$'$} \def\cprime{$'$}
\begin{thebibliography}{10}
\bibitem{MR0491063}
Proof theory and constructive mathematics.
\newblock pages 819--1142. Studies in Logic and the Foundations of Math., Vol.
90, 1977.
\newblock With contributions by C. Smory\'nski, Helmut Schwichtenberg, Richard
Statman, Solomon Feferman, A. S. Troelstra, Michael P. Fourman, Henk
Barendregt, Jeff Paris and Leo Harrington.
\bibitem{Bpasc}
Collin Bleak.
\newblock An algebraic classification of some solvable groups of
homeomorphisms.
\newblock {\em J. Algebra}, 319(4):1368--1397, 2008.
\bibitem{fast_gen}
Collin Bleak, Matthew~G. Brin, Martin Kassabov, Justin~Tatch Moore, and Matthew
Zaremsky.
\newblock Groups of fast piecewise linear homeomorphisms.
\newblock arXiv:1701.08321. 32 pages, 2017.
\bibitem{BrinU}
Matthew~G. Brin.
\newblock The ubiquity of {T}hompson's group {$F$} in groups of piecewise
linear homeomorphisms of the unit interval.
\newblock {\em J. London Math. Soc. (2)}, 60(2):449--460, 1999.
\bibitem{BrinEG}
\bysame.
\newblock Elementary amenable subgroups of {R}. {T}hompson's group {$F$}.
\newblock {\em Internat. J. Algebra Comput.}, 15(4):619--642, 2005.
\bibitem{MR2653970}
\bysame.
\newblock The free group of rank 2 is a limit of {T}hompson's group {$F$}.
\newblock {\em Groups Geom. Dyn.}, 4(3):433--454, 2010.
\bibitem{picric}
Matthew~G. Brin and Craig~C. Squier.
\newblock Presentations, conjugacy, roots, and centralizers in groups of
piecewise linear homeomorphisms of the real line.
\newblock {\em Comm. Algebra}, 29(10):4557--4596, 2001.
\bibitem{CFP}
J.~W. Cannon, W.~J. Floyd, and W.~R. Parry.
\newblock Introductory notes on {R}ichard {T}hompson's groups.
\newblock {\em Enseign. Math. (2)}, 42(3-4):215--256, 1996.
\bibitem{MR0045635}
Georg Cantor.
\newblock {\em Contributions to the founding of the theory of transfinite
numbers}.
\newblock Dover Publications, Inc., New York, N. Y., 1952.
\newblock Translated, and provided with an introduction and notes, by Philip E.
B. Jourdain.
\bibitem{Chou}
Ching Chou.
\newblock Elementary amenable groups.
\newblock {\em Illinois J. Math.}, 24(3):396--407, 1980.
\bibitem{MR19:1067c}
Mahlon~M. Day.
\newblock Amenable semigroups.
\newblock {\em Illinois J. Math.}, 1:509--544, 1957.
\bibitem{Gentzen_collected}
Gerhard Gentzen.
\newblock {\em The collected papers of {G}erhard {G}entzen}.
\newblock Edited by M. E. Szabo. Studies in Logic and the Foundations of
Mathematics. North-Holland Publishing Co., Amsterdam-London, 1969.
\bibitem{incompleteness}
Kurt G\"odel.
\newblock \"Uber formal unentscheidbare {S}\"atze der {P}rincipia {M}athematica
und verwandter {S}ysteme {I}.
\newblock {\em Monatsh. Math. Phys.}, 38(1):173--198, 1931.
\bibitem{Godel_collected1}
\bysame.
\newblock {\em Collected works. {V}ol. {I}}.
\newblock The Clarendon Press, Oxford University Press, New York, 1986.
\newblock Publications 1929--1936, Edited and with a preface by Solomon
Feferman.
\bibitem{MR0491157}
L.~A.~S. Kirby and J.~B. Paris.
\newblock Initial segments of models of {P}eano's axioms.
\newblock pages 211--226. Lecture Notes in Math., Vol. 619, 1977.
\bibitem{MR663480}
\bysame.
\newblock Accessible independence results for {P}eano arithmetic.
\newblock {\em Bull. London Math. Soc.}, 14(4):285--293, 1982.
\bibitem{Fraisse_conj}
Richard Laver.
\newblock On {F}ra\"\i ss\'e's order type conjecture.
\newblock {\em Ann. of Math. (2)}, 93:89--111, 1971.
\bibitem{MR2135961}
Andr{\'e}s Navas.
\newblock Quelques groupes moyennables de diff\'eomorphismes de l'intervalle.
\newblock {\em Bol. Soc. Mat. Mexicana (3)}, 10(2):219--244 (2005), 2004.
\bibitem{neumann:wreath}
Peter~M. Neumann.
\newblock On the structure of standard wreath products of groups.
\newblock {\em Math. Z.}, 84:343--373, 1964.
\bibitem{MR3079257}
Alexander~Yu. Ol'shanskii and Denis~V. Osin.
\newblock A quasi-isometric embedding theorem for groups.
\newblock {\em Duke Math. J.}, 162(9):1621--1648, 2013.
\bibitem{polya:wreath}
G.~Polya.
\newblock Kombinatorische {A}nzahlbestimmungen für {G}ruppen, {G}raphen und
chemische {V}erbindungen.
\newblock {\em Acta Mathematica}, 68:145--254, 1937.
\bibitem{SapirProblems}
Mark~V. Sapir.
\newblock Some group theory problems.
\newblock {\em Internat. J. Algebra Comput.}, 17(5-6):1189--1214, 2007.
\bibitem{ATaylor}
Amanda Taylor.
\newblock Locally solvable subgroups of {P}{L}o({I}), 2017.
\newblock Unpublished thesis.
\end{thebibliography}
\end{document}