fggroups.tex

\chapter{Group presentations}
\label{ch:fggroups}

\section{Brief overview of the chapter}
\label{sec:fggroups-overview}

TODO:
\begin{itemize}
\item Make a separate chapter on combinatorics? Actions and Burnside and counting colorings?
\item Cayley actions: $G$ acts on $\Gamma(G,S)$: Action on vertices is the left action of $G$ on itself: $t \mapsto (t=_{\BG} \pt)$, on vertices, for $s : S$, have edge $t = \pt$ to $t = \pt$
\item Recall universal property of free groups: If we have a map $\varphi : S \to H$, then we get a homomorphism $\bar\varphi : F(S) \to H$, represented by $BF(S) \to_\pt \BH$ defined by induction, sending $\pt$ to $\pt$ and $s$ to $\varphi(s)$.
\item define different types of graphs ($S$-digraphs, $\tilde S$-graphs,
  (partial) functional graphs, graph homomorphisms, quotients of graphs)
\item define (left/right) Cayley graphs of f.g.~groups
  -- $\Aut(\Gamma_G) = G$ (include $\alpha : F(S) \to G$ in notation?)
  -- Cayley graphs are vertex transitive
\item Cayley graphs and products, semi-direct products, homomorphisms
\item Some isomorphisms involving semi-direct products
  -- Exceptional automorphism of $\Sigma_6$:
  -- Exotic map $\Sigma_5 \to \Sigma_6$.
  (Conjugation action of $\Sigma_5$ on $6$ $5$-Sylow subgroups.)
  A set bundle $X : \BSigma_6 \to \BSigma_6$.
\item \url{https://math.ucr.edu/home/baez/six.html}
  Relating $\Sigma_6$ to the icosahedron.
  The icosahedron has $6$ axes. Two axes determines a golden rectangle (also known as a \emph{duad},\footnote{%
    These names come from Sylvester.}
  so there are $15$ such. A symmetry of the icosahedron can be described
  by knowing where a fixed rectangle goes, and a symmetry of that rectangle.
  Picking three rectangles not sharing a diagonal gives a \emph{syntheme}:
  three golden rectangles whose vertices make up the icosahedron.
  Some synthemes (known as \emph{true crosses}
  have the rectangles orthogonal to each other, as in
  \cref{fig:true-cross}.
  Fact: The symmetries of the icosahedron form the alternating symmetries of the $5$ true crosses.
  Of course, we get an action on the $6$ axes, thus a homomorphism $\constant{A}_5 \to \Sigma_6$.
  Every golden rectangle lies in one true cross and two skew crosses.
  The combinatorics of duads, synthemes, and synthematic totals are illustrated
  in the Cremona-Richardson configuration and the resulting Tutte-Coxeter graph.
  The automorphism group of the latter is in fact $\Aut(\Sigma_6)$.
  If we color the vertices according to duad/syntheme, we get $\Sigma_6$ itself.
\begin{marginfigure}
  \tdplotsetmaincoords{45}{135}
  \begin{tikzpicture}[tdplot_main_coords,scale=1.1]
    \begin{scope}[opacity=0.6]
      \draw (-1.00000, -1.61803, 0.00000) -- (-0.00000, -1.00000, -1.61803);
      \draw (-1.61803, 0.00000, -1.00000) -- (-0.00000, -1.00000, -1.61803);
      \draw (-1.00000, -1.61803, 0.00000) -- (-1.61803, -0.00000, -1.00000);
      \draw (1.00000, -1.61803, 0.00000) -- (-0.00000, -1.00000, -1.61803);
      \draw (1.00000, -1.61803, 0.00000) -- (-1.00000, -1.61803, 0.00000);
      \draw (-1.61803, 0.00000, 1.00000) -- (-1.00000, -1.61803, -0.00000);
      \draw (-1.61803, 0.00000, -1.00000) -- (0.00000, 1.00000, -1.61803);
      \draw (-1.61803, 0.00000, 1.00000) -- (-1.61803, 0.00000, -1.00000);
      \draw (0.00000, 1.00000, -1.61803) -- (0.00000, -1.00000, -1.61803);
      \fill[gray] (0.00000, 0.00000, 0.00000) -- (0.00000, -1.61803, 0.00000) -- (-1.00000, -1.61803, 0.00000) -- (-1.00000, 0.00000, 0.00000) -- cycle;
      \fill[casblue] (0.00000, 0.00000, 0.00000) -- (0.00000, 0.00000, -1.61803) -- (0.00000, -1.00000, -1.61803) -- (0.00000, -1.00000, 0.00000) -- cycle;
      \fill[casred] (0.00000, 0.00000, 0.00000) -- (-1.61803, 0.00000, 0.00000) -- (-1.61803, 0.00000, -1.00000) -- (0.00000, 0.00000, -1.00000) -- cycle;
      \draw (0.00000, -1.00000, 1.61803) -- (-1.00000, -1.61803, 0.00000);
      \draw (1.61803, 0.00000, -1.00000) -- (0.00000, -1.00000, -1.61803);
      \draw (-1.00000, 1.61803, 0.00000) -- (-1.61803, 0.00000, -1.00000);
      \fill[casred] (0.00000, 0.00000, 0.00000) -- (-1.61803, 0.00000, 0.00000) -- (-1.61803, 0.00000, 1.00000) -- (0.00000, 0.00000, 1.00000) -- cycle;
      \fill[casblue] (0.00000, 0.00000, 0.00000) -- (0.00000, 0.00000, -1.61803) -- (0.00000, 1.00000, -1.61803) -- (0.00000, 1.00000, 0.00000) -- cycle;
      \fill[gray] (0.00000, 0.00000, 0.00000) -- (0.00000, -1.61803, 0.00000) -- (1.00000, -1.61803, 0.00000) -- (1.00000, 0.00000, 0.00000) -- cycle;
      \draw (0.00000, -1.00000, 1.61803) -- (1.00000, -1.61803, 0.00000);
      \draw (0.00000, -1.00000, 1.61803) -- (-1.61803, 0.00000, 1.00000);
      \draw (1.61803, 0.00000, -1.00000) -- (1.00000, -1.61803, 0.00000);
      \draw (-1.61803, 0.00000, 1.00000) -- (-1.00000, 1.61803, 0.00000);
      \draw (0.00000, 1.00000, -1.61803) -- (1.61803, 0.00000, -1.00000);
      \draw (-1.00000, 1.61803, 0.00000) -- (0.00000, 1.00000, -1.61803);
      \fill[casred] (0.00000, 0.00000, 0.00000) -- (1.61803, 0.00000, 0.00000) -- (1.61803, 0.00000, -1.00000) -- (0.00000, 0.00000, -1.00000) -- cycle;
      \fill[gray] (0.00000, 0.00000, 0.00000) -- (0.00000, 1.61803, 0.00000) -- (-1.00000, 1.61803, 0.00000) -- (-1.00000, 0.00000, 0.00000) -- cycle;
      \fill[casblue] (0.00000, 0.00000, 0.00000) -- (0.00000, 0.00000, 1.61803) -- (0.00000, -1.00000, 1.61803) -- (0.00000, -1.00000, 0.00000) -- cycle;
      \draw (0.00000, 1.00000, 1.61803) -- (-1.61803, 0.00000, 1.00000);
      \draw (1.61803, 0.00000, 1.00000) -- (1.00000, -1.61803, 0.00000);
      \draw (1.00000, 1.61803, 0.00000) -- (0.00000, 1.00000, -1.61803);
      \fill[gray] (0.00000, 0.00000, 0.00000) -- (0.00000, 1.61803, 0.00000) -- (1.00000, 1.61803, 0.00000) -- (1.00000, 0.00000, 0.00000) -- cycle;
      \fill[casblue] (0.00000, 0.00000, 0.00000) -- (0.00000, 0.00000, 1.61803) -- (0.00000, 1.00000, 1.61803) -- (0.00000, 1.00000, 0.00000) -- cycle;
      \fill[casred] (0.00000, 0.00000, 0.00000) -- (1.61803, 0.00000, 0.00000) -- (1.61803, 0.00000, 1.00000) -- (0.00000, 0.00000, 1.00000) -- cycle;
      \draw (0.00000, -1.00000, 1.61803) -- (1.61803, 0.00000, 1.00000);
      \draw (0.00000, 1.00000, 1.61803) -- (0.00000, -1.00000, 1.61803);
      \draw (1.61803, 0.00000, 1.00000) -- (1.61803, 0.00000, -1.00000);
      \draw (1.00000, 1.61803, 0.00000) -- (1.61803, 0.00000, -1.00000);
      \draw (0.00000, 1.00000, 1.61803) -- (-1.00000, 1.61803, 0.00000);
      \draw (-1.00000, 1.61803, 0.00000) -- (1.00000, 1.61803, 0.00000);
      \draw (0.00000, 1.00000, 1.61803) -- (1.61803, 0.00000, 1.00000);
      \draw (0.00000, 1.00000, 1.61803) -- (1.00000, 1.61803, 0.00000);
      \draw (1.61803, 0.00000, 1.00000) -- (1.00000, 1.61803, 0.00000);
    \end{scope}
  \end{tikzpicture}
  \caption{Icosahedron with an inscribed true cross}
  \label{fig:true-cross}
\end{marginfigure}
\item define (left/right) presentation complex of group presentation
\item define Stallings folding
\item deduce Nielsen--Schreier and Nielsen basis
\item deduce algorithms for generalized word problem, conjugation, etc.
\item deduce Howson's theorem
\item think about 2-cell replacement for folding; better proofs in HoTT?
\item move decidability results to main flow
\item include undecidability of word problem in general
  -- doesn't depend on presentation (for classes closed under inverse images of monoid homomorphisms)
\item describe $F(S)/H$ in the case where $H$ has infinite index
\item describe normal closure of $R$ in $F(S)$ -- still f.g.? -- get Cayley graph of $F(S)/\langle R\rangle$. -- Todd-Coxeter algorithm?
\item in good cases we can recognize $\mathcal{S}(R)$ as a ``fundamental domain'' in Cayley graph of $\langle S\mid R\rangle$.
\end{itemize}

\begin{remark}
  In this chapter, we use letters from the
  beginning of the alphabet $a,b,c,\dots$
  to denote generators,
  and we use the corresponding capital letters
  $A,B,C$ to denote their inverses,
  so, e.g., $aA=Aa=1$.
  This cleans up the notational clutter significantly.
\end{remark}

Do we fix $S$, a finite set $S=\{a,b,\ldots\}$?
Mostly $F$ will denote the free group on $S$.
And for almost all examples, we take $S = \{a,b\}$.

\section{Graphs and Cayley graphs}
\label{sec:cayley-graphs}

We have seen in the previous chapter how cyclic groups
(those generated by a single generator)
have neatly described types of torsors.
Indeed, $\BCG_n \jdeq \Cyc_n$, where $\Cyc_n$ is the type of $n$-cycles,
and the classifying type of the integers, $\B\ZZ\jdeq\Sc$, \ie the circle,
is equivalent to the type of infinite cycles, $\Cyc_0$.
In \cref{cha:circle}, we defined the types of (finite or infinite)
cycles as certain components of $\sum_{X:\UU}(X\eqto X)$,
but we can equivalently consider components of $\sum_{X:\UU}(X\to X)$,
since the former is a subtype of the latter.
By thinking of functions in terms of their graphs,
we might as well look at components of $\sum_{X:\UU}(X \to X \to \UU)$.

In this section we shall generalize this story
to groups $G$ generated by a
(finite or just decidable)
set of generators $S$.

First recall from Cayley's \cref{lem:allgpsarepermutationgps}
that any group $G$ can be realized as a subgroup of the permutation group
on the underlying set of symmetries in $G$, $\USymG$.
In this description, a $G$-shape is a set $X$ equipped
a $G$-action that defines a $G$-torsor,
which in turn can be expressed as the structure of a map $\alpha:\USymG \to X \to X$
satisfying certain properties.

It may happen that already $\alpha$ restricted to a subset $S$ of $\USymG$
suffices to specify the action.
In that case we say that $S$ generates $G$, though we'll take the following
as the official definition.
\begin{definition}\label{def:gens-gp}
  Let $G$ be a group and $S$ be a subset of $\USymG$, given by an inclusion
  $\iota : S \to \USymG$. We say that \emph{$S$ generates $G$} if the induced
  homomorphism from the free group on $S$,
  \[
    \FG_S \to G,
  \]
  is an epimorphism.
\end{definition}
\begin{lemma}\label{lem:gens-gp-iff}
  Let $G$ be a group
  and $\iota : S \to \USymG$ an inclusion of a subset of the elements of $G$.
  Then $S$ generates $G$ if and only if the map
  \[
    \rho_S : \BG \to \sum_{X:\UU}(S \to X \to X) ,
    \quad
    \rho_S(t) \defeq \bigl(t \eqto \sh_G, s \mapsto \iota(s) \cdot \blank\bigr)
  \]
  is an embedding.\footnote{We use $t \eqto \sh_G$ rather than the equivalent
    $\sh_G \eqto t$ in order to conform to the representation from Cayley's theorem.}
\end{lemma}
In this case, then, $G$ can be identified with the automorphism group of $\rho_S(\sh_G)$
in the type $\sum_{X:\UU}(S \to X \to X)$, or even in the larger type (of which it's a subtype), $\sum_{X:\UU}(S \to X \to X \to \UU)$.

\begin{figure}
  \begin{sidecaption}%
    {Cayley graph for {$\protect\SG_3$} with respect to $S = \{(1\;2),(2\;3)\}$.}[fig:cayley-s3]
  \centering
  \begin{tikzpicture}
    \pgfmathsetmacro{\len}{2}
    \node[vertex,label=30:$(1\;3)$]   (n13)  at (30:\len)  {};
    \node[vertex,label=90:$(1\;3\;2)$]  (n132) at (90:\len)  {};
    \node[vertex,label=150:$(1\;2)$]  (n12)  at (150:\len) {};
    \node[vertex,label=210:$e$]     (ne)   at (210:\len) {};
    \node[vertex,label=270:$(2\;3)$]  (n23)  at (270:\len) {};
    \node[vertex,label=330:$(1\;2\;3)$] (n123) at (330:\len) {};
    \begin{scope}[every to/.style={bend right=22}]
      % generator a is (12)
      \draw[gena] (ne)   to (n12);
      \draw[gena] (n12)  to (ne);
      \draw[gena] (n13)  to (n132);
      \draw[gena] (n132) to (n13);
      \draw[gena] (n123) to (n23);
      \draw[gena] (n23)  to (n123);
      % generator b is (23)
      \draw[genb] (ne)   to (n23);
      \draw[genb] (n23)  to (ne);
      \draw[genb] (n13)  to (n123);
      \draw[genb] (n123) to (n13);
      \draw[genb] (n12)  to (n132);
      \draw[genb] (n132) to (n12);
    \end{scope}
  \end{tikzpicture}
  \end{sidecaption}
\end{figure}

Also note that $S$ generates $G$ if and only if the map on elements
$\UFG_S \to \USymG$ is surjective, meaning every element of $G$ can be expressed
as a product of the letters in a (reduced) word from $\mathcal R_S$, interpreted
according to the inclusion of $S$ into $\USymG$.
This is the case for example for $S$ consisting of the transpositions
$(1\;2)$, $(2\;3)$ in $\SG_3$, as illustrated in~\cref{fig:cayley-s3},
where the \textcolor{casblue}{blue} color represents $(1\;2)$
and the \textcolor{casred}{red} color represents $(2\;3)$.

Before we give the proof of~\cref{lem:gens-gp-iff}, let us study these types more closely.
\begin{definition}
  An $S$-labeled graph is an element $(V,E)$ of the type
  $\sum_{V:\UU}(S \to V \to V \to \UU)$.%
  \index{graph!labeled}
  The first component $V$ is called the type of \emph{vertices} of the graph,
  and the type $E(s,x,y)$ is called the type of $s$-colored \emph{edges}
  from $x$ (the source) to $y$ (the target).
\end{definition}
If for every vertex $x:V$ and every color $s:S$ there is unique $s$-colored edge out of $x$, \ie the type $\sum_{y:V}E(s,x,y)$ is contractible, then we say that the graph
is \emph{functional}. This means that the graph lives in the subtype $\sum_{V:\UU}(S \to V \to V)$,
as is the case for the graph $\rho_S(\sh_G)$ for a group $G$.
This graph is called the Cayley graph of $G$ with respect to the set $S$:
\begin{definition}\label{def:cayley-graph}
  The \emph{Cayley graph}\index{graph!Cayley graph}
  of a group $G$ with respect to a generating subset $S$
  is the graph $\Cay(G;S)$\glossary(Cay){$\protect\Cay(G;S)$}{Cayley graph of a group $G$
  with respect to $S$}
  is the $S$-colored graph with vertices $\USymG$
  and edges $S \times \USymG$ where the edge $(s,g)$ has source $g$, target $sg$,
  and color $s$.
\end{definition}
Convince yourself that this is really an equivalent description of $\rho_S(\sh_G)$
considered as an $S$-colored graph.

If $S$ is contractible (so there's only one color), then we just say \emph{graph},
and then we simplify the type of edges to $V \to V \to \UU$.
Of course, every $S$-labeled graph $(V,E)$ gives rise to such an unlabeled label
by summing over the colors, \ie the type of edges from $x$ to $y$ in this graph
is $\sum_{s:S}E(s,x,y)$.

Another way to represent a graph is to sum over all the sources and targets (and colors),
via~\cref{lem:typefamiliesandfibrations},
\ie as a tuple $(V,E,s,t,c)$, where $V:\UU$ is the type of vertices,
$E$ is the (total) type of edges,
$s,t : E \to V$ give the source and target of an edge,
while $c: E \to S$ gives the color (if we're talking about $S$-colored graphs).
In this description, to get the unlabeled graph we simply drop the last component.

Every graph $(V,E)$ (and thus every labeled graph) gives rise to a type
by ``gluing the edges to the vertices'' defined as follows.
\begin{definition}
  Fix an unlabeled graph $(V,E)$. The \emph{graph quotient}\footnote{%
    If the graph is represented by source and target maps
    $s,t: E \rightrightarrows V$, then the graph quotient is
  also called the \emph{coequalizer} of $s$ and $t$.} $V/E$ is
  the higher inductive type with constructors:
  \begin{enumerate}
  \item For every vertex $x : V$ a point $[x] : V/E$.
  \item For every edge $e : E(x,y)$ an identification $\qedge_e : [x] \eqto [y]$.
  \end{enumerate}

  Let $A(z)$ be a type for every element $z:V/E$. The induction principle
  for $V/E$ states that, in order to define an element of $A(z)$ for every $z:V/E$,
  it suffices to give elements $a_x : A([x])$ for every vertex $x:V$
  together with
  identifications $q_e : \pathover{a_x}{A}{\qedge_e}{a_y}$
  for every $e:E(x,y)$.
  The function $f$ thus defined satisfies $f([x])\jdeq a_x$ for $x:V$
  and we are provided identifications $\apd{f}(\qedge_e)\eqto q_e$ for each $e:E(x,y)$.
\end{definition}
\begin{remark}
  Note the similarity with the classifying type of a free group,
  \cf~\cref{def:bfree}. Indeed, if we form the (unlabeled!)
  graph $(\bn 1,S)$
  on one vertex with $S$ edges, then $\bn 1/S$ is essentially the same as $\BFG_S$.
\end{remark}
\begin{xca}
  An equivalence relation $R : A \to A \to \Prop$ on a set $A$
  can be regarded as a graph $(A,R)$.
  Construct an equivalence between set truncation of the graph quotient $\Trunc{A/R}_0$
  and the set quotient $A/R$ from~\cref{def:quotient-set} in this case.
  (So in the world of sets, the two notations agree.)
\end{xca}
While we're building up to the proof of~\cref{lem:gens-gp-iff} we need
a description of a sum type over a graph quotient.
By the above remark, this applies also to sum types over $\BFG_S$.
\begin{construction}\label{def:graph-quotient-flattening}
  Given a graph $(V,E)$ and a family of types $X : V/E \to \UU$.
  Define $V' \defeq \sum_{v:V}X([v])$ and
  $E'((v,x),(w,y)) \defeq \sum_{e:E(v,w)}\pathover{x}{X}{\qedge_e}{y}$.
  Then we have an equivalence\index{flattening construction}\footnote{%
    This is often called the \emph{flattening construction} (or flattening lemma),
    as it ``flattens'' a sum over a graph quotient into a single graph quotient.}
  \[
    \flt : \Bigl(\sum_{z:V/E}X(z)\Bigr) \equivto V'/E'
  \]
\end{construction}
\begin{implementation}{def:graph-quotient-flattening}
  We define functions $\varphi : V'/E' \to \sum_{z:V/E}X(z)$
  and $\psi : \prod_{z:V/E}\bigl(X(z) \to V'/E'\bigr)$
  using the induction principles:
  \begin{alignat*}2
    \varphi([(v,x)]) &\defeq ([v],x) &\quad
    \tilde\psi([v]) &\defeq (x \mapsto [(v,x)]) \\
    \ap\varphi(\qedge_{(e,q)}) &\defis \pathpair{\qedge_e}{q} &\quad
    \apd{\tilde\psi}(\qedge_e) &\defis h,
  \end{alignat*}
  where we need to construct\marginnote{%
    \begin{tikzcd}[column sep=small,ampersand replacement=\&]
      X([v]) \ar[rr,"{\trp[X]{\qedge_e}}","\simeq"']\ar[dr,"{\psi([v])}"']
      \& \& X([w]) \ar[dl,"{\psi([w])}"] \\
      \& V'/E' \&
    \end{tikzcd}}
  $h : \pathover{(x \mapsto [(v,x)])}{z \mapsto X(z) \to V'/E'}{\qedge_e}
  {(y \mapsto [(w,y)])}$ for all $e : E(v,w)$.
  By transporting in families of functions,
  it suffices to give an identification
  $[(v,x)] \eqto [(w,\trp[X]{\qedge_e}(x))]$ for all $x : X([v])$.
  We get this as the identification constructor $\qedge_{(e,q)}$ for $V'/E'$,
  where $q: \pathover{x}{X}{\qedge_e}{\trp[X]{\qedge_e}(x)}$
  is the identification over $\qedge_e$ corresponding
  to the reflexivity identification at
  $\trp[X]{\qedge_e}(x)$ via~\cref{def:pathover-trp}.
\end{implementation}
\begin{xca}
  Complete the implementation by giving identifications $\psi\circ\phi\eqto\id$
  and $\phi\circ\psi\eqto\id$, where $\psi : \bigl(\sum_{z:V/E}X(z)\bigr) \to V'/E'$
  is defined by $\psi((z,x)) \defeq \tilde\psi(z)(x)$.
\end{xca}

Later on we'll need also need the following results about graph quotients.
\begin{xca}\label{xca:graph-quotient-in-steps}
  Suppose the edges $E$ of a graph $(V,E)$ are expressed as a binary sum $E_0 \amalg E_1$.
  (Here, it doesn't matter whether $E$ is expressed as a type family $E : V \to V \to \UU$,
  in which case we have a family of equivalences $E(v,w) \equivto E_0(v,w) \amalg E_1(v,w)$,
  or $E$ is the total type of edges.)

  Then we can obtain the graph quotient $V/E$ by first gluing in the edges from $E_0$,
  and then gluing in the edges from $E_1$ to the resulting type $V/E_0$.
  Using the description of graphs with a total type of edges $E \equivto E_0 \amalg E_1$,
  we have corresponding source and target maps expressed as compositions:
  \[
    E_1 \hookrightarrow E_0\amalg E_1 \equivto E \rightrightarrows V \to V/E_0.
  \]
  Construct an equivalence $V/E \equivto V/(E_0 \amalg E_1) \equivto (V/E_0)/E_1$.
\end{xca}

\begin{xca}\label{xca:graph-quotient-whisker}
  Suppose we have any type $X$ with an element $x:X$.
  We can form a graph $(X\amalg\bn 1,\bn 1)$ with vertex type $X\amalg\bn 1$
  and a single edge from $\inl x$ to $\inr 0$.
  Construct an equivalence $X \equivto (X\amalg\bn 1)/\bn 1$.\footnote{%
    This equivalence can be visualized as follows, where $X$ ``grows a whisker''
    along the single edge.\\
    \begin{tikzpicture}
      \draw (0,0) ellipse (1 and 1.2);
      \node (X) at (0,1.5) {$X$};
      \node (t) at (1.7,0.6) {$\bn 1$};
      \node[dot,label=below:$x$] (x) at (.6,0.3) {};
      \node[dot] (pt) at (1.5,.4) {};
      \draw[->,bend left] (x) -- (pt) {};
    \end{tikzpicture}}
\end{xca}
\section{Examples}
\label{sec:fg-examples}



\begin{proof}[Proof of~\cref{lem:gens-gp-iff}]
  TBD (perhaps put in graph quotients first)
\end{proof}


\begin{figure}
  \begin{sidecaption}%
    {Cayley graph for $A_5$ with respect to $S = \{a,b\}$,
      where $a$ is a $1/5$-rotation about a vertex and
      $b$ is a $1/2$-rotation about an edge in an icosahedron.}[fig:cayley-a5]
  \centering
\tdplotsetmaincoords{45}{135}
\begin{tikzpicture}[tdplot_main_coords,scale=2]
  % cayley icosahedron
  \begin{scope}[fill=casblue,opacity=.2]
\fill (0.00000, 1.00000, 1.61803) -- (0.00000, -1.00000, 1.61803) -- (-1.61803, 0.00000, 1.00000) -- cycle;
\fill (-0.00000, 1.00000, 1.61803) -- (1.61803, 0.00000, 1.00000) -- (-0.00000, -1.00000, 1.61803) -- cycle;
\fill (0.00000, 1.00000, 1.61803) -- (-1.61803, 0.00000, 1.00000) -- (-1.00000, 1.61803, 0.00000) -- cycle;
\fill (-0.00000, 1.00000, 1.61803) -- (1.00000, 1.61803, 0.00000) -- (1.61803, 0.00000, 1.00000) -- cycle;
\fill (-0.00000, -1.00000, 1.61803) -- (1.61803, -0.00000, 1.00000) -- (1.00000, -1.61803, 0.00000) -- cycle;
\fill (0.00000, 1.00000, 1.61803) -- (-1.00000, 1.61803, 0.00000) -- (1.00000, 1.61803, 0.00000) -- cycle;
\fill (-0.00000, -1.00000, 1.61803) -- (-1.00000, -1.61803, 0.00000) -- (-1.61803, 0.00000, 1.00000) -- cycle;
\fill (1.61803, -0.00000, 1.00000) -- (1.00000, 1.61803, -0.00000) -- (1.61803, -0.00000, -1.00000) -- cycle;
\fill (-0.00000, -1.00000, 1.61803) -- (1.00000, -1.61803, 0.00000) -- (-1.00000, -1.61803, 0.00000) -- cycle;
\fill (-1.61803, 0.00000, 1.00000) -- (-1.61803, 0.00000, -1.00000) -- (-1.00000, 1.61803, 0.00000) -- cycle;
\fill (1.00000, 1.61803, 0.00000) -- (-1.00000, 1.61803, -0.00000) -- (0.00000, 1.00000, -1.61803) -- cycle;
\fill (-1.61803, 0.00000, 1.00000) -- (-1.00000, -1.61803, -0.00000) -- (-1.61803, 0.00000, -1.00000) -- cycle;
\fill (1.61803, 0.00000, 1.00000) -- (1.61803, -0.00000, -1.00000) -- (1.00000, -1.61803, 0.00000) -- cycle;
\fill (-1.00000, -1.61803, 0.00000) -- (1.00000, -1.61803, -0.00000) -- (-0.00000, -1.00000, -1.61803) -- cycle;
\fill (-1.00000, 1.61803, 0.00000) -- (-1.61803, 0.00000, -1.00000) -- (0.00000, 1.00000, -1.61803) -- cycle;
\fill (1.00000, 1.61803, 0.00000) -- (0.00000, 1.00000, -1.61803) -- (1.61803, 0.00000, -1.00000) -- cycle;
\fill (1.00000, -1.61803, 0.00000) -- (1.61803, -0.00000, -1.00000) -- (-0.00000, -1.00000, -1.61803) -- cycle;
\fill (-1.61803, 0.00000, -1.00000) -- (-1.00000, -1.61803, -0.00000) -- (-0.00000, -1.00000, -1.61803) -- cycle;
\fill (1.61803, 0.00000, -1.00000) -- (0.00000, 1.00000, -1.61803) -- (0.00000, -1.00000, -1.61803) -- cycle;
\fill (0.00000, 1.00000, -1.61803) -- (-1.61803, 0.00000, -1.00000) -- (0.00000, -1.00000, -1.61803) -- cycle;
\end{scope}
\node[vertex] (n) at (0.00000, 0.50000, 1.61803) {};
\node[vertex] (na) at (0.40451, 0.75000, 1.46353) {};
\node[vertex] (nb) at (-0.00000, -0.50000, 1.61803) {};
\node[vertex] (nA) at (-0.40451, 0.75000, 1.46353) {};
\node[vertex] (naa) at (0.25000, 1.15451, 1.21353) {};
\node[vertex] (nba) at (-0.40451, -0.75000, 1.46353) {};
\node[vertex] (nab) at (1.21353, 0.25000, 1.15451) {};
\node[vertex] (nAb) at (-1.21353, 0.25000, 1.15451) {};
\node[vertex] (nbA) at (0.40451, -0.75000, 1.46353) {};
\node[vertex] (nAA) at (-0.25000, 1.15451, 1.21353) {};
\node[vertex] (nbaa) at (-0.25000, -1.15451, 1.21353) {};
\node[vertex] (naba) at (1.21353, -0.25000, 1.15451) {};
\node[vertex] (nAba) at (-1.46353, 0.40451, 0.75000) {};
\node[vertex] (naab) at (0.75000, 1.46353, 0.40451) {};
\node[vertex] (nbab) at (-1.21353, -0.25000, 1.15451) {};
\node[vertex] (nAAb) at (-0.75000, 1.46353, 0.40451) {};
\node[vertex] (nabA) at (1.46353, 0.40451, 0.75000) {};
\node[vertex] (nbAA) at (0.25000, -1.15451, 1.21353) {};
\node[vertex] (nabaa) at (1.46353, -0.40451, 0.75000) {};
\node[vertex] (nAbaa) at (-1.61803, 0.00000, 0.50000) {};
\node[vertex] (naaba) at (1.15451, 1.21353, 0.25000) {};
\node[vertex] (nAAba) at (-0.50000, 1.61803, 0.00000) {};
\node[vertex] (nbaab) at (-0.75000, -1.46353, 0.40451) {};
\node[vertex] (nAbab) at (-1.15451, 1.21353, 0.25000) {};
\node[vertex] (nbAAb) at (0.75000, -1.46353, 0.40451) {};
\node[vertex] (naabA) at (0.50000, 1.61803, 0.00000) {};
\node[vertex] (nbabA) at (-1.46353, -0.40451, 0.75000) {};
\node[vertex] (nabAA) at (1.61803, -0.00000, 0.50000) {};
\node[vertex] (naabaa) at (1.15451, 1.21353, -0.25000) {};
\node[vertex] (nAAbaa) at (-0.75000, 1.46353, -0.40451) {};
\node[vertex] (nbaaba) at (-1.15451, -1.21353, 0.25000) {};
\node[vertex] (nbAAba) at (0.50000, -1.61803, 0.00000) {};
\node[vertex] (nabaab) at (1.15451, -1.21353, 0.25000) {};
\node[vertex] (nAbaab) at (-1.61803, 0.00000, -0.50000) {};
\node[vertex] (nabAAb) at (1.61803, -0.00000, -0.50000) {};
\node[vertex] (nbaabA) at (-0.50000, -1.61803, 0.00000) {};
\node[vertex] (nAbabA) at (-1.15451, 1.21353, -0.25000) {};
\node[vertex] (naabAA) at (0.75000, 1.46353, -0.40451) {};
\node[vertex] (nbaabaa) at (-1.15451, -1.21353, -0.25000) {};
\node[vertex] (nbAAbaa) at (0.75000, -1.46353, -0.40451) {};
\node[vertex] (nAbaaba) at (-1.46353, 0.40451, -0.75000) {};
\node[vertex] (nabAAba) at (1.46353, -0.40451, -0.75000) {};
\node[vertex] (naabaab) at (1.46353, 0.40451, -0.75000) {};
\node[vertex] (nAAbaab) at (-0.25000, 1.15451, -1.21353) {};
\node[vertex] (naabAAb) at (0.25000, 1.15451, -1.21353) {};
\node[vertex] (nabaabA) at (1.15451, -1.21353, -0.25000) {};
\node[vertex] (nAbaabA) at (-1.46353, -0.40451, -0.75000) {};
\node[vertex] (nbaabAA) at (-0.75000, -1.46353, -0.40451) {};
\node[vertex] (nAbaabaa) at (-1.21353, 0.25000, -1.15451) {};
\node[vertex] (nabAAbaa) at (1.21353, -0.25000, -1.15451) {};
\node[vertex] (naabAAba) at (0.40451, 0.75000, -1.46353) {};
\node[vertex] (nbAAbaab) at (0.25000, -1.15451, -1.21353) {};
\node[vertex] (nbaabAAb) at (-0.25000, -1.15451, -1.21353) {};
\node[vertex] (naabaabA) at (1.21353, 0.25000, -1.15451) {};
\node[vertex] (nAAbaabA) at (-0.40451, 0.75000, -1.46353) {};
\node[vertex] (nAbaabAA) at (-1.21353, -0.25000, -1.15451) {};
\node[vertex] (naabAAbaa) at (0.00000, 0.50000, -1.61803) {};
\node[vertex] (nbaabAAba) at (-0.40451, -0.75000, -1.46353) {};
\node[vertex] (nabAAbaab) at (0.40451, -0.75000, -1.46353) {};
\node[vertex] (nbaabAAbaa) at (-0.00000, -0.50000, -1.61803) {};
\draw[gena] (n) -- (nA);
\draw[genb] (n) -- (nb);
\draw[gena] (na) -- (n);
\draw[genb] (na) -- (nab);
\draw[gena] (nb) -- (nbA);
\draw[genb] (nb) -- (n);
\draw[gena] (nA) -- (nAA);
\draw[genb] (nA) -- (nAb);
\draw[gena] (naa) -- (na);
\draw[genb] (naa) -- (naab);
\draw[gena] (nba) -- (nb);
\draw[genb] (nba) -- (nbab);
\draw[gena] (nab) -- (nabA);
\draw[genb] (nab) -- (na);
\draw[gena] (nAb) -- (nbab);
\draw[genb] (nAb) -- (nA);
\draw[gena] (nbA) -- (nbAA);
\draw[genb] (nbA) -- (naba);
\draw[gena] (nAA) -- (naa);
\draw[genb] (nAA) -- (nAAb);
\draw[gena] (nbaa) -- (nba);
\draw[genb] (nbaa) -- (nbaab);
\draw[gena] (naba) -- (nab);
\draw[genb] (naba) -- (nbA);
\draw[gena] (nAba) -- (nAb);
\draw[genb] (nAba) -- (nAbab);
\draw[gena] (naab) -- (naabA);
\draw[genb] (naab) -- (naa);
\draw[gena] (nbab) -- (nbabA);
\draw[genb] (nbab) -- (nba);
\draw[gena] (nAAb) -- (nAbab);
\draw[genb] (nAAb) -- (nAA);
\draw[gena] (nabA) -- (nabAA);
\draw[genb] (nabA) -- (naaba);
\draw[gena] (nbAA) -- (nbaa);
\draw[genb] (nbAA) -- (nbAAb);
\draw[gena] (nabaa) -- (naba);
\draw[genb] (nabaa) -- (nabaab);
\draw[gena] (nAbaa) -- (nAba);
\draw[genb] (nAbaa) -- (nAbaab);
\draw[gena] (naaba) -- (naab);
\draw[genb] (naaba) -- (nabA);
\draw[gena] (nAAba) -- (nAAb);
\draw[genb] (nAAba) -- (naabA);
\draw[gena] (nbaab) -- (nbaabA);
\draw[genb] (nbaab) -- (nbaa);
\draw[gena] (nAbab) -- (nAbabA);
\draw[genb] (nAbab) -- (nAba);
\draw[gena] (nbAAb) -- (nabaab);
\draw[genb] (nbAAb) -- (nbAA);
\draw[gena] (naabA) -- (naabAA);
\draw[genb] (naabA) -- (nAAba);
\draw[gena] (nbabA) -- (nAbaa);
\draw[genb] (nbabA) -- (nbaaba);
\draw[gena] (nabAA) -- (nabaa);
\draw[genb] (nabAA) -- (nabAAb);
\draw[gena] (naabaa) -- (naaba);
\draw[genb] (naabaa) -- (naabaab);
\draw[gena] (nAAbaa) -- (nAAba);
\draw[genb] (nAAbaa) -- (nAAbaab);
\draw[gena] (nbaaba) -- (nbaab);
\draw[genb] (nbaaba) -- (nbabA);
\draw[gena] (nbAAba) -- (nbAAb);
\draw[genb] (nbAAba) -- (nbaabA);
\draw[gena] (nabaab) -- (nabaabA);
\draw[genb] (nabaab) -- (nabaa);
\draw[gena] (nAbaab) -- (nAbaabA);
\draw[genb] (nAbaab) -- (nAbaa);
\draw[gena] (nabAAb) -- (naabaab);
\draw[genb] (nabAAb) -- (nabAA);
\draw[gena] (nbaabA) -- (nbaabAA);
\draw[genb] (nbaabA) -- (nbAAba);
\draw[gena] (nAbabA) -- (nAAbaa);
\draw[genb] (nAbabA) -- (nAbaaba);
\draw[gena] (naabAA) -- (naabaa);
\draw[genb] (naabAA) -- (naabAAb);
\draw[gena] (nbaabaa) -- (nbaaba);
\draw[genb] (nbaabaa) -- (nAbaabA);
\draw[gena] (nbAAbaa) -- (nbAAba);
\draw[genb] (nbAAbaa) -- (nbAAbaab);
\draw[gena] (nAbaaba) -- (nAbaab);
\draw[genb] (nAbaaba) -- (nAbabA);
\draw[gena] (nabAAba) -- (nabAAb);
\draw[genb] (nabAAba) -- (nabaabA);
\draw[gena] (naabaab) -- (naabaabA);
\draw[genb] (naabaab) -- (naabaa);
\draw[gena] (nAAbaab) -- (nAAbaabA);
\draw[genb] (nAAbaab) -- (nAAbaa);
\draw[gena] (naabAAb) -- (nAAbaab);
\draw[genb] (naabAAb) -- (naabAA);
\draw[gena] (nabaabA) -- (nbAAbaa);
\draw[genb] (nabaabA) -- (nabAAba);
\draw[gena] (nAbaabA) -- (nAbaabAA);
\draw[genb] (nAbaabA) -- (nbaabaa);
\draw[gena] (nbaabAA) -- (nbaabaa);
\draw[genb] (nbaabAA) -- (nbaabAAb);
\draw[gena] (nAbaabaa) -- (nAbaaba);
\draw[genb] (nAbaabaa) -- (nAAbaabA);
\draw[gena] (nabAAbaa) -- (nabAAba);
\draw[genb] (nabAAbaa) -- (nabAAbaab);
\draw[gena] (naabAAba) -- (naabAAb);
\draw[genb] (naabAAba) -- (naabaabA);
\draw[gena] (nbAAbaab) -- (nabAAbaab);
\draw[genb] (nbAAbaab) -- (nbAAbaa);
\draw[gena] (nbaabAAb) -- (nbAAbaab);
\draw[genb] (nbaabAAb) -- (nbaabAA);
\draw[gena] (naabaabA) -- (nabAAbaa);
\draw[genb] (naabaabA) -- (naabAAba);
\draw[gena] (nAAbaabA) -- (naabAAbaa);
\draw[genb] (nAAbaabA) -- (nAbaabaa);
\draw[gena] (nAbaabAA) -- (nAbaabaa);
\draw[genb] (nAbaabAA) -- (nbaabAAba);
\draw[gena] (naabAAbaa) -- (naabAAba);
\draw[genb] (naabAAbaa) -- (nbaabAAbaa);
\draw[gena] (nbaabAAba) -- (nbaabAAb);
\draw[genb] (nbaabAAba) -- (nAbaabAA);
\draw[gena] (nabAAbaab) -- (nbaabAAbaa);
\draw[genb] (nabAAbaab) -- (nabAAbaa);
\draw[gena] (nbaabAAbaa) -- (nbaabAAba);
\draw[genb] (nbaabAAbaa) -- (naabAAbaa);
\end{tikzpicture}
  \end{sidecaption}
\end{figure}

\section{Subgroups of free groups}
\label{sec:subgroups-free}

We now study subgroups of free groups.\marginnote{%
  Our discussion follows the work of
  \citeauthor{Swan2022}\footnotemark{}.}\footcitetext{Swan2022}
We'll eventually prove the Nielsen--Schreier theorem,
which states that a finite index subgroup $H$ of a free group $\FG_S$ is itself a free group.
Furthermore, when $S$ is finite, the set of free generators of $H$ is itself finite.

Recall from~\cref{def:set-of-subgroups} that a subgroup is (or can be represented by)
a transitive $G$-set $X : \BG \to \Set$ along with an element of $X(\sh_G)$.
\begin{definition}\label{def:finite-index}
  A subgroup of a group $G$ has \emph{finite index} $m$ if the underlying
  transitive $G$-set, $X : \BG \to \Set$ is a family of finite sets of cardinality $m$.
\end{definition}
The is the case, of course, if and only if the set acted on, $X(\sh_G)$, is finite
of cardinality $m$.
Notice that the definition doesn't depend on the chosen element of $X(\sh_G)$,
so applies equally to all conjugacy classes of the subgroup.

Recall also that the classifying type of the subgroup is the total type $\sum_{t:\BG}X(t)$ (which is pointed via the chosen point of $X(\sh_G)$).
We'll use the Flattening~\cref{def:graph-quotient-flattening} to analyze
this in case where $G$ is the free group on a set $S$, $\FG_S$,
so we need to show that the quotient of the resulting graph
is equivalent to $\bn 1/T$ for some set $T$.

We do this by finding a ``spanning tree'' in the graph.

\begin{definition}
  A graph $(V,E)$ is \emph{connected}\index{connected!graph} if $V/E$ is a connected type
  and it's a \emph{tree} if $V/E$ is contractible.
\end{definition}

\begin{definition}
  A \emph{subgraph} of a graph $(V,E)$ consists of a subtype $h : U \hookrightarrow V$
  of the vertices along with, for every pair of vertices $v,w$ in $U$,
  a subtype $D(v,w)$ of the edges $E(v,w)$.
\end{definition}
If we represent graphs by source and target maps, then this amounts to embeddings
$h : U \hookrightarrow V$ and $k : D \hookrightarrow E$ along with witnesses that the
following squares commute:
\[
  \begin{tikzcd}
    D \ar[r,hook,"k"]\ar[d,"s"'] & E \ar[d,"s"] \\
    U \ar[r,hook,"h"'] & V
  \end{tikzcd}\quad
  \begin{tikzcd}
    D \ar[r,hook,"k"]\ar[d,"t"'] & E \ar[d,"t"] \\
    U \ar[r,hook,"h"'] & V
  \end{tikzcd}
\]
\begin{definition}
  A \emph{spanning tree} in a graph $(V,E)$ is a subgraph $(U,D)$
  such that $(U,D)$ is a tree, and the embedding of the vertices
  $U \hookrightarrow V$ is an equivalence.
\end{definition}
Equivalently, it's given by subtypes of the edges (leaving the vertices alone)
such that the underlying graph is a tree.
Very often we'll require that the edge embeddings are decidable,
\ie we can decide whether a given edge $e : E(v,w)$ is part of the tree.

\begin{marginfigure}
  \begin{tikzpicture}
    \draw (-.25,0.9) ellipse (.55 and 1.2);
    \node (X) at (0,2.4) {$V_0$};
    \draw (1.2,1.1) ellipse (.7 and 1);
    \node (Y) at (1,2.3) {$V_1$};
    \node[dot,label=below:$v_0$] (x) at (-.1,0.4) {};
    \node[dot,label=above:$v_1$] (y) at (1.6,1.3) {};
    \node[dot,label=below:$u_0$] (s) at (-.1,1.4) {};
    \node[dot,label=below:$u_1$] (t) at (1.0,1.3) {};
    \draw[casblue,dotted,<->] (s) -- (t) node[midway,label=above:$e$] {};
    \draw[dashed] (0.5,1.1) ellipse (1.5 and 1.9);
    \node (XY) at (-0.75,3.05) {$V_0 \amalg V_1$};
  \end{tikzpicture}
  \caption{A connected graph with a crossing edge}
  \label{fig:crossing-edge}
\end{marginfigure}
\begin{lemma}\label{lem:crossing-edge}
  Suppose we have a connected graph $(V,E)$ whose type of vertices
  decomposes as a binary sum $V \equivto V_0 \amalg V_1$
  and we have $v_0:V_0$ and $v_1:V_1$.
  Then there merely exists an edge $e$ either with source in $V_0$ and target in $V_1$
  or the other way round.
\end{lemma}
The situation is illustrated in~\cref{fig:crossing-edge}, where we assume
there is an edge relation on the binary sum that gives a connected graph,
and hence there must be a ``crossing edge'' $e$, going either from $V_0$ to $V_1$
or the other way.
\begin{proof}
  We may assume $V \jdeq V_0 \amalg V_1$ by path induction.
  The idea is then to define a family of propositions $P : V/E \to \Prop$
  that, on one hand is trivially true over $V_0$, and on the other hand
  expresses our desired goal, the existence of a ``crossing edge'', over $V_1$.

  We now define $P(z)$, for $z:V/E$,
  by the induction principle for the graph quotient $V/E$.
  We set $P([\inl v]) \defeq \true$ for $v : V_0$ and
  \[
    P([\inr v]) \defeq \Trunc*{\sum_{u_0:V_0}\sum_{u_1:V_1}
      \bigl(E(u_0,u_1) \amalg E(u_1,u_0)\bigr)}
  \]
  for $v : V_1$.
  We must then prove that the propositions $P([v])$ and $P([v'])$ are equivalent
  whenever there's an edge from $v$ to $v'$.
  This is the case by definition when $v,v'$ lie in the same summand,
  and it's also the case when they lie in different summands, since then
  we get a witness for the truth over $V_1$.

  Since $V/E$ is connected, $P$ must have a constant truth value,
  and since $P([\inl{v_0}])\jdeq\true$, every $P(z)$ is true.
  Hence also $P([\inr{v_1}])$ is true, which is exactly what we wanted.
\end{proof}

\begin{lemma}\label{lem:spanning-tree-step}
  Fix a connected graph $(V,E)$ where $V$ has decidable equality and $E$ is a family of sets.
  For any subgraph $(U,D)$, where the embedding $U\hookrightarrow V$ is decidable,
  and with vertices $u \in U$ and $v \in V \setminus U$,
  there merely exists\footnote{%
    Keep in mind that subgraphs consist not only of the vertices and edges,
    but also of the corresponding embeddings into the supergraph.
    It's for the sake of these that we only prove mere existence.\\
    \begin{tikzcd}[sep=small,ampersand replacement=\&]
      D \ar[r,hook]\ar[d,shift left]\ar[d,shift right] \&
      D\amalg\bn 1 \ar[r,hook]\ar[d,shift left]\ar[d,shift right] \&
      E\ar[d,shift left]\ar[d,shift right] \\
      U \ar[r,hook]\ar[d] \&
      U\amalg\bn 1 \ar[r,hook]\ar[d] \&
      V \ar[d] \\
      U/D \ar[r,"\equiv"] \&
      (U\amalg\bn 1)/(D\amalg\bn 1)\ar[r] \&
      V/E
    \end{tikzcd}}
  a larger subgraph with exactly one more vertex and one more edge,
  $(U\amalg\bn 1, D\amalg\bn 1)$ such that the induced map on graph quotients
  $U/D \to (U\amalg\bn 1)/(D\amalg\bn 1)$ is an equivalence.
\end{lemma}
\begin{proof}
  Since the embedding $U\hookrightarrow V$ is decidable,
  we can write $V$ as the binary sum $U \amalg (V\setminus U)$.
  Apply~\cref{lem:crossing-edge} to find a ``crossing edge'' $e$,
  and form the new subgraph $(U\amalg\bn 1, D\amalg\bn 1)$ by adding
  the incident vertex not in $U$ as well as the edge $e$ itself.
  The embedding $U\amalg\bn 1 \to V$ is still decidable, since $V$ has decidable equality.
  Finally, we have
  \[
    (U\amalg\bn 1) / (D\amalg\bn 1) \equivto \bigl((U\amalg\bn 1)/\bn 1\bigr)/D
    \equivto U/D,
  \]
  using~\cref{xca:graph-quotient-in-steps,xca:graph-quotient-whisker}, as desired.
\end{proof}

\begin{lemma}\label{lem:spanning-tree}
  Let $(V,E)$ be a connected graph where $V$ is an $n$-element set,
  and $E$ is a family of decidable sets. Then the graph merely has a spanning tree
  with exactly $n-1$ edges.
\end{lemma}

\begin{marginfigure}
  \begin{tikzpicture}
    \node[dot] (v1) at (0,0) {};
    \node[dot] (v2) at (1,0) {};
    \node[dot] (v3) at (-.4,1) {};
    \node[dot] (v4) at (.7,.9) {};
    \node[dot] (v5) at (.5,1.8) {};
    \node[dot] (v6) at (1.5,1.6) {};
    \draw[->,casred] (v1) -- (v2);
    \draw[->,casred] (v1) to[out=135,in=270] (v3);
    \draw[->,casred] (v3) -- (v4);
    \draw[->] (v4) -- (v2);
    \draw[->,casred] (v6) to[out=200,in=80] (v3);
    \draw[->] (v2) to[out=45,in=270] (v6);
    \draw[->] (v6) -- (v4);
    \draw[->,casred] (v6) to[out=90,in=45] (v5);
  \end{tikzpicture}
  \caption{A connected graph on $6$ vertices with a spanning tree indicated
    in~\textcolor{casred}{red}.}
  \label{fig:spanning-tree-example}
\end{marginfigure}

\begin{proof}
  We show by induction on $k$, with $1\le k\le n$, that there merely exists
  a subgraph $(U,D)$ with $k$ vertices, $k-1$ edges, and $U/D$ contractible,
  \ie the graph $(U,D)$ is a tree.

  For $k\jdeq 1$, we use that $V/E$ is connected to get that $V$ merely has a vertex $v$.
  This then defines the desired subgraph on one vertex with no edges,
  and this is clearly a tree.

  Suppose we have such a desired subgraph $(U,D)$
  with $k$ vertices and $k-1$ edges and $k<n$.
  Since $V$ is finite, there exists vertices $u \in U$ and $v \in V \setminus U$.
  Now apply~\cref{lem:spanning-tree-step} to get the next subgraph.

  Finally, the subgraph $(U,D)$ with $n$ vertices and $n-1$ edges gives
  the desired spanning tree, and any embedding of an $n$-element set
  in another $n$-element set is an equivalence.\footnote{%
    Assuming the Axiom of Choice, we can show the mere existence of a spanning tree
    in any graph $(V,E)$ with a sets of vertices and edges.
    See the above work by~\citeauthor{Swan2022}.}
\end{proof}

\begin{theorem}[Nielsen--Schreier Theorem]
  Suppose that $S$ is a set with decidable equality and
  $X : \BFG_S \to \Set$ defines a (conjugacy class of a)
  finite index subgroup of $\FG_S$.
  Then $\sum_{z:\BFG_S}X(z)$ is merely equivalent to $\BFG_T$ for some set $T$.

  Moreover, if $S$ is a finite set of cardinality $n$ and the subgroup has index $m$,
  then $T$ can be taken to be a finite set of cardinality $m(n-1)+1$.
\end{theorem}
\begin{proof}
  By the Flattening~\cref{def:graph-quotient-flattening},
  we have an equivalence $\flt:\bigl(\sum_{z:\BFG_S}X(z)\bigr) \equivto V/E$,
  with $V\defeq X(\base)$ and $E(x,y) \defeq \sum_{s:S}(\pathover{x}{X}{\Sloop_s}{y})$.
  By the finite index assumption, $V$ is a finite set,
  say, of cardinality $m>0$,
  and since both $S$ and $X(\base)$ are decidable, so is $E$.

  By~\cref{lem:spanning-tree}, the graph $(V,E)$ merely contains
  a spanning tree with
  $m-1$ edges $E_0$, and complementary edge set $E_1$. Hence,
  using \cref{xca:graph-quotient-in-steps}, we have a chain of equivalences:
  \[
    V/E \equivto (V/E_0)/E_1
    \equivto \bn 1/E_1 \equivto \BFG_{E_1}
  \]
  This establishes the first claim with $T \defeq E_1$.

  If, furthermore, $S$ has cardinality $n$, then the graph $(V,E)$
  has $mn$ edges, as there are precisely $n$ outgoing edges from each vertex.
  Since $E_0$ has $m-1$ edges,
  that leaves $mn - (m-1) = mn-m+1 = m(n-1)+1$ edges in $E_1$, as desired.
\end{proof}
(This also has an automata theoretic proof, see below.)

\section{Intersecting subgroups}
\label{sec:intersecting-subgroups}

Stallings folding\footcite{Stallings1991}.

\begin{theorem}
  Let $H$ be a finitely generated subgroup of $F(S)$ and let $u\in\tilde S^*$
  be a reduced word. Then $u$ represents an element of $H$ if and only if
  $u$ is recognized by the Stallings automaton $\mathcal{S}(H)$.
\end{theorem}
\begin{theorem}
  Let $H$ be a finitely generated subgroup of $F(S)$.
  Then $H$ has finite index if and only if $\mathcal{S}(H)$ is total.

  Furthermore, in this case the index equals the number of vertices of
  $\mathcal{S}(H)$.
\end{theorem}
\begin{corollary}
  If $H$ has index $n$ in $F(S)$, then $\casop{\constant{rk}} H = 1 + n(\casop{\constant{card}} S-1)$.
\end{corollary}

\begin{theorem}\label{thm:howson-neumann}
  Suppose $H_1,H_2$ are two subgroups of $F$ with finite indices $h_1,h_2$.
  Then the intersection $H_1\cap H_2$ has finite index at most $h_1h_2$.
\end{theorem}
\marginnote{The qualitative part of \cref{thm:howson-neumann}
  is known as \emph{Howson's theorem}, while the inequality
  is known as \emph{Hanna Neumann's inequality}.
  Hanna's son, Walter Neumann, conjectured that the $2$ could be removed,
  and this was later proved independently by Joel Friedman
  and Igor Mineyev.}

\section{Connections with automata (*)}

($S$ is still a fixed finite set.)

Let $\iota : F(S) \to \tilde S^*$ map an element of the free group to the corresponding reduced word.
The kernel of $\iota$ is the \emph{2-sided Dyck language} $\mathcal D_S$.

The following theorem is due to Benois.
\begin{theorem}
  A subset $X$ of $F(S)$ is rational if and only if
  $\iota(X)\subseteq \tilde S^*$ is a regular language.
\end{theorem}

\begin{lemma}
  Let $\rho : \tilde S^* \to \tilde S^*$ map a word to its reduction.
  Then $\rho$ maps regular languages to regular languages.
\end{lemma}

The following is due to Sénizergues:
\begin{theorem}
  A rational subset of $F(S)$ is either disjunctive or recognizable.
\end{theorem}

Given a surjective monoid homomorphism $\alpha : S^* \to G$,
we define the corresponding \emph{matched homomorphism} $\tilde\alpha : \tilde S^* \to G$ by $(\tilde\alpha(a^{-1}) \defeq \alpha(a)^{-1}$.
\begin{theorem}[?]
  Consider a f.g.\ group $G$ with a surjective homomorphism
  $\alpha : F(S) \to G$. A subset $X$ of $G$ is recognisable by a finite $G$-action
  if and only if $\tilde\alpha^{-1}(X) \subseteq \tilde S^*$ is rational (i.e., regular).
\end{theorem}
\begin{theorem}[Chomsky--Schützenberger]
  A language $L \subseteq T^*$ is context-free if and only if
  $L = h(R \cap D_S)$ for some finite $S$,
  where $h : T^* \to \tilde S^*$ is a homomorphism,
  $R \subseteq \tilde S^*$ is a regular language, and
  $D_S$ is the Dyck language for $S$.\footnote{%
    References TODO. The theorem is also true if we replace $D_S$
    by its one-sided variant, but in this case it reduces to
    the well-known equivalence between context-free languages
    and languages recognizable by pushdown automata.}
\end{theorem}
\begin{theorem}[Muller--Schupp, ?]
  Suppose $\tilde\alpha : \tilde S^* \to G$ is a surjective matched homomorphism
  onto a group $G$. Then $G$ is virtually free (i.e., $G$ has a normal free subgroup of finite index) if and only if $\ker(\tilde\alpha)$ is a context-free language.
\end{theorem}
\begin{theorem}

\end{theorem}
The Stallings automaton is an \emph{inverse automaton}:
it's deterministic,
and there's an edge $(p,a,q)$ if and only if there's one $(q,A,p)$.
We can always think of the latter as the \emph{reverse} edge.
(It's then also deterministic in the reverse direction.)

Two vertices $p,q$ get identified in the Stallings graph/automaton
if and only if there is a run from $p$ to $q$ with a word $w$
whose reduction is $1$. (So a word like $aAAaBBbb$.)

\begin{theorem}
  Let $X \subseteq F(S)$. Then $Y$ is a coset $Hw$ with $H$
  a finitely generated subgroup,
  if and only if
  there is a finite state inverse automaton whose language (after reduction)
  is $Y$.
\end{theorem}

\begin{corollary}
  The generalized word problem in $F(S)$ is solvable:
  Given a finitely generated subgroup $H$, and a word $u : \tilde S^*$,
  we can decide whether $u$ represents an element of $H$.
\end{corollary}
\marginnote{%
  The Stallings automaton for $H$ can be constructed in time
  $O(n \log^*n)$, where $n$ is the sum of the lengths of the generators for $H$.
  [Cite: Touikan: A fast algorithm for Stallings' folding process.]
  Once this has been constructed, we can solve membership in $H$ in linear time.}

As above, we get a basis for $H$ as a free group from a spanning tree in
$\mathcal{S}(H)$.

\begin{theorem}
  We can decide whether two f.g.\ subgroups of $F(S)$ are conjugate.
  Moreover, a f.g.\ subgroup $H$ is normal if and only if $\mathcal{S}(H)$
  is vertex-transitive.
\end{theorem}
\begin{proof}
  $G,H$ are conjugate of and only if their cores are equal.
\end{proof}

There are other connections between group theory and language theory:
\begin{theorem}[Anisimov and Seifert]
  A subgroup $H$ of $G$ is rational if and only if $H$ is finitely generated.
\end{theorem}
\begin{theorem}
  A subgroup $H$ of $G$ is recognizable if and only if it has finite index.
\end{theorem}
%%% Local Variables:
%%% mode: latex
%%% fill-column: 144
%%% TeX-master: "book"
%%% End: