Some questions related to free-by-cyclic groups and tubular groups
Extracted source text · arXiv 2504.14192
%2multibyte Version: 5.50.0.2960 CodePage: 65001
%\documentclass[openbib]{article}
\documentclass[12pt,a4paper]{amsart}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\usepackage{geometry}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{amsfonts}
\usepackage{mathrsfs}
\usepackage{graphicx}
\usepackage{xcolor}
\usepackage{enumitem}
\setcounter{MaxMatrixCols}{10}
\usepackage{bbm}
\usepackage[hidelinks,colorlinks=black,linkcolor=black,citecolor=black]{hyperref}
\usepackage[noabbrev,capitalize]{cleveref}
\geometry{left=2.5cm,right=2.5cm,top=2.5cm,bottom=2.5cm}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{acknowledgement}[theorem]{Acknowledgement}
\newtheorem{algorithm}[theorem]{Algorithm}
\newtheorem{axiom}[theorem]{Axiom}
\newtheorem{case}[theorem]{Case}
\newtheorem{claim}[theorem]{Claim}
\newtheorem{conclusion}[theorem]{Conclusion}
\newtheorem{condition}[theorem]{Condition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{criterion}[theorem]{Criterion}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{exercise}[theorem]{Exercise}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{notation}[theorem]{Notation}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{question}[theorem]{Question}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{solution}[theorem]{Solution}
\newtheorem{summary}[theorem]{Summary}
%\newenvironment{proof}[1][Proof]{\noindent\textbf{#1.} }{\ \rule{0.5em}{0.5em}}
%\input{tcilatex}
\begin{document}
%\title{Some virtually special free-by-cyclic groups}
\title{Some questions related to free-by-cyclic groups and tubular groups}
\author{Xiaolei Wu}
\address{Shanghai Center for Mathematical Sciences, Jiangwan Campus, Fudan University, No.2005 Songhu Road, Shanghai, 200438, P.R. China}
\email{xiaoleiwu@fudan.edu.cn}
\author{Shengkui Ye}
\address{NYU Shanghai, No.567 Yangsi West
Rd, Pudong New Area, Shanghai, 200124, P.R. China \\
NYU-ECNU Institute of Mathematical Sciences at NYU Shanghai, 3663 Zhongshan Road North, Shanghai, 200062, China}
\email{sy55@nyu.edu}
\subjclass[2010]{20E26, 20E06, 20F65}
\date{April, 2025}
\keywords{Free-by-cyclic groups, tubular groups, virtually special groups, Property (VRC), RFRS}
\maketitle
\begin{abstract}
We prove that a CAT(0) free-by-cyclic tubular group with one vertex is
virtually special, but many of them cannot virtually act freely and cocompactly on CAT(0) cube complexes. This partially confirms a question of
Brady--Soroko \cite[Section 9: Question 1]{BS} and answers a question of Lyman \cite[Question 1]{Ly} in the negative. Furthermore, we provide examples of free-by-cyclic groups
amalgamated along cyclic subgroups that are not virtually free-by-cyclic. This answers
negatively a question of Hagen--Wise \cite[Remark 3.6]{hw}. Lastly, we exhibit an example of a cyclic-subgroup-separable tubular group that does not have the property (VRC) (i.e. every cyclic subgroup is a virtual retract). This answers a question of Minasyan \cite[Question 11.6]{min} in the negative.
\end{abstract}
%\tableofcontents
\section{Introduction}
This paper aims to study some questions that lie at the intersection of free-by-cyclic groups and tubular groups. Recall that a group $G$ is called free-by-cyclic if it surjects to $\mathbb{Z}$ with a free kernel (of possibly infinite rank). If the kernel is a free group of rank $n,$ we call $G$ an $F_{n}$-by-$\mathbb{Z}$ group. A tubular group is a group that splits as a finite graph of groups with $\mathbb{Z}^2$ vertex groups and $\mathbb{Z}$ edge groups. There are many examples of free-by-cyclic groups that are tubular. For example, Gersten's famous non-$\mathrm{CAT(0)}$ example is tubular \cite{ge}. Button gives a criterion to determine whether a tubular group is free-by-cyclic \cite[Theorem 2.1]{bu}. Our study is motivated by the following two questions:
\begin{question} \cite[Section 9, Question 1]{BS}
\label{prob1}Does every $\mathrm{CAT(0)}$ free-by-cyclic group virtually embed in a right-angled Artin group $(\mathrm{RAAG})$?
\end{question}
\begin{question} \cite[Problem 5.3]{page}
\label{prob1.1} What’s the relationship between being $\mathrm{CAT(0)}$ and (virtually) $\mathrm{RFRS}$ for free-by-cyclic groups?
\end{question}
Since a virtually special group (i.e., a subgroup of a RAAG) is virtually RFRS \cite[Corollary 2.3]{Ag08}, a positive answer to Question \ref{prob1} would solve Question \ref{prob1.1} in one direction. Recall that it is already a hard question when a free-by-cyclic group is $\mathrm{CAT(0)}$ (see for example
Bridson's problem list \cite[Question 7.9]{Bri}). The far-reaching work of Agol \cite{Ag} and Wise \cite{wi12},
together with Hagen--Wise \cite{ha1,ha2} implies that a hyperbolic
$F_n$-by-$\mathbb{Z}$ group is virtually compact special. Button
and Kropholler \cite{bk} prove that every $F_{2}$-by-$\mathbb{Z}$ group is
the fundamental group of a non-positively curved cube complex of dimension
2. On the other hand, every $F_{2}$-by-$\mathbb{Z}$ group can be represented as a
non-positively curved punctured-torus bundle over the circle, implying that every $%
F_{2} $-by-$\mathbb{Z}$ group is $\mathrm{CAT(0)}$ and virtually special
by the work of Liu \cite[Theorem 1.1]{liu}. When $n\ge 3$, Samuelson \cite{sa} proves that the $F_{n}\rtimes
_{\phi }\mathbb{Z}$ groups are $\mathrm{CAT(0)}$ for some upper triangular
automorphism $\phi$. Lyman \cite{Ly} further provides several infinite families
of polynomially growing automorphisms of free groups whose mapping tori are
$\mathrm{CAT(0)}$. In a recent paper, Munro and Petyt \cite{mp} discuss the compact specialness
of free-by-cyclic groups using the median metric space. On the other hand, the study of cubulating tubular groups is in a much better shape. In fact, Wise in \cite[Theorem 1.1]{wi14} gives an easy criterion to determine when a tubular group acts freely on a $\mathrm{CAT(0)}$ cube complex. Woodhouse provides a way to produce free actions of tubular groups on finite dimensional $\mathrm{CAT(0)}$ cube complexes \cite[Theorem 1.2]{Wood2} and proves further in \cite[Theorem 1.1]{Wood} that a tubular group is virtually special if and only if it acts freely on a finite dimensional $\mathrm{CAT(0)}$ cube complex.
Let $F_3=\langle a,b,c \rangle$ be a free group of rank 3. Consider the following automorphism of $\Psi: F_3 \rightarrow F_3:$
$$
a \mapsto a, b \mapsto a^m b a^m, c \mapsto a^n b a^n $$
for integers $m,n$. Lyman \cite{Ly} proves that the corresponding group $F_{3}\rtimes _{\Psi }\mathbb{Z}$ (among others) are $\mathrm{CAT(0)}$.
Based on Question \ref{prob1}, Lyman \cite[Question 1.1]{Ly} further asks the
following question:
\begin{question}
\label{prob2} May these $\mathrm{CAT(0)}$ free-by-cyclic groups (eg. $F_{n}\rtimes _{\Psi }\mathbb{Z}$) be cocompactly
cubulated? Is the resulting cube complex virtually (co-)special?
\end{question}
We answer Question \ref{prob2} negatively by viewing it as a tubular group.
\begin{theorem}\label{thm:Lyman} [Example \ref{eg1}]
For any integers $m,n$, the group $F_{3}\rtimes _{\Psi }\mathbb{Z}$ is virtually special. But it virtually acts on a CAT(0) cube complex freely and cocompactly if and only if $|m| = |n|$.
\end{theorem}
Button (cf. \cite{but} remarks after Corollary 3) also notices that there exist non-compact virtually special tubular groups whose underlying graph is a
tree. More generally, we study free-by-cyclic groups $G=G(\{p_{i},q_{i}\}_{i=1}^{n-1})=F_{n}\rtimes _{\phi }%
\mathbb{Z}$ of the following form: let $F_{n}$ be a free group generated by $%
\{a_{0},a_{1},...,a_{n-1}\}$ and $\phi :F_{n}\rightarrow F_{n}$ be a
homomorphism given by
\begin{eqnarray*}
a_{0} &\mapsto &a_{0}, \\
a_{i} &\mapsto &a_0^{p_{i}}a_{i}a_0^{q_{i}},i=1,2,...,n-1,
\end{eqnarray*}%
for some integers $p_{1},\cdots ,p_{n-1},q_{1},\cdots ,q_{n-1}$. Note that
the semi-direct product $G=F_{n}\rtimes _{\phi }\mathbb{Z}$ has a
presentation:
\begin{equation*}
\langle
a_{0},a_{1},...,a_{n-1},t:ta_{0}t^{-1}=a_{0},ta_{i}t^{-1}=a_{0}^{p_{i}}a_{i}a_{0}^{q_{i}},i=1,2,...,n-1\rangle .
\end{equation*}%
Rewrite the relator $ta_{i}t^{-1}=a_{0}^{p_{i}}a_{i}a_{0}^{q_{i}}$ as
$a_{i}a_{0}^{q_{i}}ta_{i}^{-1} = a_{0}^{-p_{i}}t$ for each $i.$ One sees that $G$ can be viewed as a
tubular group with the vertex group $G_{v}=\mathbb{Z}^{2}=\langle a_{0},t\rangle
$ and stable letters $a_{i}$ for $1\leq i\leq n-1$. The group $G$ is
not hyperbolic since it contains $\mathbb{Z}^{2}$ as a subgroup. When $%
n=3,p_{1}=p_{2}=0$, the group $G$ is studied by Gersten \cite{ge}, who
proves that when $q_{1}\neq q_{2}$ are positive integers, the group $G$ is not $\mathrm{CAT(0)}$ (actually $G$
cannot act properly by semi-simple isometries on any $\mathrm{CAT(0)}$ metric
space)$.$ When $p_{i}= \pm q_{i}$ (generally symmetric automorphism $\phi $),
the group $G$ is studied by Lyman \cite{Ly} who proves that some of them
are $\mathrm{CAT(0)}$.
\begin{theorem} [Corollary \ref{cor1} and Theorem \ref{th7-1}]
\label{th7}Let $G=G(\{p_{i},q_{i}\}_{i=1}^{n-1})=F_{n}\rtimes _{\phi }\mathbb{Z}$. The following are
equivalent:
\begin{description}
\item[i)] $G$ is virtually special;
\item[ii)] $G$ is $\mathrm{CAT(0)}$;
\item[iii)] either
\end{description}
1) $p_{i}=-q_{i}$ for each $i=1,2,...,n-1$ or
2) $p_{s}\neq -q_{s}$ for some $s$ and
\begin{equation*}
q_{i}(q_{i}+p_{s}-q_{s})=p_{i}(p_{i}-p_{s}+q_{s})
\end{equation*}%
for each $i=1,2,...,n-1.$
\end{theorem}
\begin{remark}
Theorem \ref{th7} implies that when either of the two conditions in iii)
holds, the group $F_{n}\rtimes _{\phi }\mathbb{Z}$ is linear (actually $%
\mathbb{Z}$-linear). It is still an open question whether every $F_{n}$-by-$%
\mathbb{Z}$ group is linear; see for example \cite[Problem 5]{ds}.
\end{remark}
More generally, let
\begin{equation*}
v_{1},v_{2},...,v_{k},w_{1},w_{2},...,w_{k}\in \mathbb{Z}^{2}
\end{equation*}
be $2k$ nonzero vectors and
\begin{equation*}
G=\langle \mathbb{Z}^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle
\end{equation*}
be the multiple HNN extensions.
\begin{theorem} [Theorem \ref{thm:fr-cyc-vspel}]
\label{th4} Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle $ be a $\mathrm{CAT(0)}$
free-by-cyclic tubular group. Then $G$ is virtually special.
\end{theorem}
The proofs of Theorem \ref{th7} and Theorem \ref{th4} are based on a study
of a tubular group with a single vertex (see Theorem \ref{main}). We also prove that a tubular free-by-cyclic group is actually $F_{n}$-by-$\mathbb{Z}$ (see Theorem \ref{th10}).
\bigskip
The second part of our study considers the amalgamation of two free-by-cyclic groups. According
to Hagen--Wise \cite[Remark 3.6]{hw}, it is not clear whether the class of
free-by-cyclic groups is closed under amalgams along $\mathbb{Z}$ subgroups.
As Example \ref{eg2} will show that the answer to their question is negative, we therefore revise their question to the following.
\begin{question}
\label{prob3} Is the amalgamated product of two free-by-cyclic groups over a
cyclic subgroup virtually free-by-cyclic?
\end{question}
It turns out that the answer to Question \ref{prob3} is still
negative.
\begin{theorem}[\Cref{corlast}]
Let $G_1 =\langle a,b,c,t\mid tat^{-1}=a,tbt^{-1}=ba, tct^{-1} =ba^2\rangle $ be the Gersten group, $G_2 = \langle x,s\mid sxs^{-1} = x \rangle \cong \mathbb{Z}^2$. Then the tubular group $G = G_1 \ast_{a=x} G_2$ is not virtually free-by-cyclic.
\end{theorem}
\begin{remark}
Kielak and Linton in \cite[Theorem 1.1]{KL24} give a criterion for a hyperbolic and virtually compact special group to be virtually free-by-cyclic: it is virtually free-by-cyclic if and only if $b_2^{(2)}(G)=0$ and $\mathrm{cd}_{\mathbb{Q}}\leq 2$. Fisher improves their criterion by relaxing the condition ``hyperbolic and virtually compact special" to just ``$\mathrm{RFRS}$" \cite[Theorem A]{Fi24}. On the other hand, it is not too hard to see that tubular groups are $2$-dimensional, locally indicable, coherent \cite[Theorem 5.1]{Wi20}, and have vanishing $\ell^2$-Betti numbers. Also, our group $G$ does not have any Baumslag--Solitar subgroup of the form $\langle a,t\mid ta^mt^{-1} =a^n\rangle$ where $|m|\neq |n|$. This shows that the $\mathrm{RFRS}$ condition in Fisher's theorem cannot be easily further improved.
\end{remark}
The last part of our study considers the property $\mathrm{(VRC)}$. Recall that Minasyan \cite{min} following Long--Reid \cite{lr} defines a group $G$ to
be $\mathrm{(VRC)}$ (virtual retract onto any cyclic subgroup) if for every cyclic subgroup $C$,
there is a finite index subgroup $H\leq G$ such that $H\geq C$ and $H$ has
an epimorphism $H\rightarrow C$ satisfying $C\rightarrow
H\rightarrow C$ is the identity. A subgroup $K$ of a group $G$ is called separable if $K$ is the intersection of a family of finite index subgroups in $G$. The following question was asked by Minasyan.
\begin{question}\cite[Question 11.6]{min}
\label{Ques-Min}
Let $G$ be the fundamental group of a finite graph of groups with free abelian
vertex groups (of finite ranks) and cyclic edge groups. If $G$ is cyclic subgroup separable, does
it necessarily satisfy $\mathrm{(VRC)}$?
\end{question}
The answer to Question \ref{Ques-Min} is again negative.
\begin{theorem}[Example \ref{eg7.2}]
The Gersten group $$\langle a,b,c,t \mid tat^{-1}=a,tbt^{-1}=ba, tct^{-1}=ca^2 \rangle $$ is a cyclic-subgroup-separable tubular group that does not have the property $\mathrm{(VRC)}$.
\end{theorem}
In the last section, we discuss the subtle difference between RFRS and virtually RFRS. We remark that if one drops the virtualness assumption in Question \ref{prob1.1},
and a question of Agol in his ICM survey paper \cite[Question 11]{Ag14} (whether the braid groups are RFRS), there are counterexamples.
\subsection*{Structure of the Paper.} In Section 2, we study the CAT(0)ness of tubular groups. In Section 3, we give a sufficient and necessary condition for a tubular group to be free-by-cyclic or $F_n$-by-$\mathbb{Z}$ using Button's criterion. We then study the virtual specialness of tubular groups in Section 4. We prove that a CAT(0) free-by-cyclic tubular group over a single vertex is virtually special. In Section 5, we discuss when amalgamated products of two free-by-cyclic groups along cyclic subgroups are again (virtually) free-by-cyclic. In Section 6, we study the (VRC) property and subgroup separation of free-by-cyclic groups. In the final section, discuss the subtle difference between RFRS and virtually RFRS, then propose some questions.
\subsection*{Acknowledgments.}
Wu is currently a member of LMNS and is supported by NSFC No.12326601. Ye is
supported by NSFC No. 11971389. We thank Ian Agol, Sami Douba, Mark Hagen, Monika Kudlinska and Daniel Wise for their helpful communications. In particular, we learned the idea of proof for Lemma \ref{lem6.11} from Sami Douba. After we shared our preprint with Ashot Minasyan, he informed us that in his on-going joint with Jon Merladet, they also found many counterexamples to \Cref{Ques-Min}. We also thank him for many comments on our preprint.
\section{$\mathrm{CAT(0)}$ness}
We study when a tubular group with only one vertex is $\mathrm{CAT(0)}$ in this section.
Recall that a group $G$ is $\mathrm{CAT(0)}$ if it can act properly and cocompactly on a $\mathrm{CAT(0)}$ space by isometries. The proofs of Theorem \ref{th7} and Theorem \ref{th4} are based on a study
of tubular groups with a single vertex group. For this class of groups, we provide some concrete conditions so that they are $\mathrm{CAT(0)}$, free-by-cyclic, or virtually special.
\begin{theorem}
\label{main}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle .$
\begin{enumerate}[label=(\arabic*)]
\item Suppose that $\{v_{1},w_{1}\}$ are linearly independent columns of the invertible matrix $[v_1,w_1]$, and $%
[v_{1},w_{1}]^{-1}v_{i}=%
\begin{pmatrix}
x_{i} \\
y_{i}%
\end{pmatrix}%
,[v_{1},w_{1}]^{-1}w_{i}=%
\begin{pmatrix}
x_{i}^{\prime } \\
y_{i}^{\prime }%
\end{pmatrix}%
,i=2,...,k$. The group $G$ is $\mathrm{CAT(0)}$ if and only if
\begin{equation*}
|x_{i}|^{2}+|y_{i}|^{2}-2|x_{i}y_{i}|\cos \phi =|x_{i}^{\prime
}|^{2}+|y_{i}^{\prime }|^{2}-2|x_{i}^{\prime }y_{i}^{\prime }|\cos \phi
,i=2,...,k
\end{equation*}%
for some common $\phi \in (0,\pi ).$
\item The group $G$ is free-by-cyclic if and only if $v_{1}-w_{1}$, $%
v_{2}-w_{2},...,v_{k}-w_{k}$ lie in a common line $l$ which doesn't contain
any element of $\{v_{1},v_{2},...,v_{n}\}.$
\item The group $G$ is virtually special if $\{v_{1},w_{1}\}$ are
linearly independent and
\begin{eqnarray*}
\det [w_{1}-v_{1},v_{i}] &=&\pm \det [w_{1}-v_{1},w_{i}], \\
\det [w_{1}+v_{1},v_{i}] &=&\pm \det [w_{1}+v_{1},w_{i}],i=2,...,k.
\end{eqnarray*}
\end{enumerate}
\end{theorem}
\begin{remark}
A tubular free-by-cyclic group is actually $F_{n}$-by-$\mathbb{Z}$ (see Theorem \ref{th10}).
\end{remark}
The proof of Theorem \ref{main} will be spread in the following sections.
Let $v_{1},v_{2},...,v_{k},w_{1},w_{2},...,w_{k}\in \mathbb{Z}^{n}$ be $2k$
nonzero integer vectors and $G=\langle \mathbb{Z}%
^{n},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle $ be a multiple
HNN extension. The following is a starting point.
\begin{lemma}
\label{lem1}The group $G$ is $\mathrm{CAT(0)}$ if and only if there is an invertible
matrix $A_{n\times n}\in \mathrm{GL}_{n}(\mathbb{R})$ such that
\begin{equation*}
\parallel Av_{i}\parallel =\parallel Aw_{i}\parallel ,i=1,2,...,k.
\end{equation*}
\end{lemma}
\begin{proof}
If $G$ acts properly and cocompactly on a $\mathrm{CAT(0)}$ space $X,$ the flat torus
theorem implies that $\mathbb{Z}^{n}\leq G$ stabilize an Euclidean subspace $%
\mathbb{E}^n\subset X$ and $\mathbb{E}^n/\mathbb{Z}^{n}$ is a flat torus. Suppose that $%
\{e_{1},e_{2},...,e_{n}\}$ is the standard basis of $\mathbb{Z}^{n}\ $and
each $e_{i}$ acts on $\mathbb{E}^n$ by a translation of the vector $A_{i}\in
\mathbb{R}^{n}.$ Then $v_{i}\in \mathbb{Z}^{n}$ acts on $\mathbb{E}^n$ as the
translation corresponding to the vector $Av_{i}$ where $A=[A_{1},...,A_{n}]$. Since $s_{i}v_{i}s_{i}^{-1}=w_{i},$ we know that $%
\parallel Av_{i}\parallel =\parallel Aw_{i}\parallel $ for each $%
i=1,2,...,k. $
Conversely, suppose that there is an invertible matrix $A_{n\times
n} = [A_{1},...,A_{n}]\in \mathrm{GL}_{n}(\mathbb{R})$ such that
\begin{equation*}
\parallel Av_{i}\parallel =\parallel Aw_{i}\parallel ,i=1,2,...,k.
\end{equation*}%
Define the action of $\mathbb{Z}^{n}$ on the Euclidean space $\mathbb{E}^n$ by $%
e_{i}\rightarrow A_{i},i=1,2,...,k.$ Since $\parallel Av_{i}\parallel
=\parallel Aw_{i}\parallel ,$ the translation lengths of $v_{i}$ and $w_{i}$
are equal. Therefore, the multiple HNN extension is a $\mathrm{CAT(0)}$ group by \cite[Theorem II.11.18]{BH}.
\end{proof}
\begin{remark}
\label{remark}The necessary condition of the theorem holds for any
$\mathrm{CAT(0)}$ group containing $G.$ In particular, a tubular group is $\mathrm{CAT(0)}$ only
if for each vertex $v$ there is an invertible matrix $A_{2\times 2}\in
\mathrm{GL}_{2}(\mathbb{R})$ such that
\begin{equation*}
\parallel Av_{i}\parallel =\parallel Aw_{i}\parallel ,i=1,2,...,k.
\end{equation*}%
Here $v_i$ and $w_i$ come from the edge relation $
s_{i}v_{i}s_{i}^{-1}=w_{i}$ for each edge $s_{i}$ connecting $v$ to $v$.
\end{remark}
When $k=1,$ we have the following.
\begin{corollary}
The group $G=\langle \mathbb{Z}^{n},s|svs^{-1}=w\rangle $ is $\mathrm{CAT(0)}$ if and
only if $\{v,w\}$ is linearly independent or $v=\pm w.$
\end{corollary}
\begin{proof}
If $\{v,w\}$ is linearly independent, there is an invertible real matrix $%
A_{n\times n}$ containing $v,w$ as its first two columns. Then $%
A^{-1}v,A^{-1}w$ have the same length. If $v=\pm w,$ they already have the
same length. Lemma \ref{lem1} implies that $G$ is $\mathrm{CAT(0)}$.
If $G$ is $\mathrm{CAT(0)}$ and $\{v,w\}$ is linearly dependent, we have $av=bw$ for
some integers $a,b.$ Since $v,w$ are conjugate, they have the same
translation length. Therefore, we have $|av|=|bw|$ implying $v=\pm w.$
\end{proof}
Theorem \ref{main} (1) is the following result.
\begin{theorem}
\label{th1}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle .$ Suppose that $%
\{v_{1},w_{1}\}$ are linearly independent, and
\begin{equation*}
\lbrack v_{1},w_{1}]^{-1}v_{i}=%
\begin{pmatrix}
x_{i} \\
y_{i}%
\end{pmatrix}%
,[v_{1},w_{1}]^{-1}w_{i}=%
\begin{pmatrix}
x_{i}^{\prime } \\
y_{i}^{\prime }%
\end{pmatrix}%
,i=2,...,k.
\end{equation*}%
The group $G$ is $\mathrm{CAT(0)}$ if and only if
\begin{equation*}
|x_{i}|^{2}+|y_{i}|^{2}-2|x_{i}y_{i}|\cos \phi =|x_{i}^{\prime
}|^{2}+|y_{i}^{\prime }|^{2}-2|x_{i}^{\prime }y_{i}^{\prime }|\cos \phi
,i=2,...,k
\end{equation*}%
for some common $\phi \in (0,\pi ).$
\end{theorem}
\begin{proof}
If $G\ $is $\mathrm{CAT(0)}$, Lemma \ref{lem1} implies that there is $A\in \mathrm{GL}%
_{2}(\mathbb{R})$ such that $\parallel Av_{i}\parallel =\\\parallel
Aw_{i}\parallel ,i=1,2.$ Let
\begin{equation*}
B=A[v_{1},w_{1}]=[B_{1},B_{2}].
\end{equation*}%
Then%
\begin{eqnarray*}
Av_{1} &=&B[v_{1},w_{1}]^{-1}v_{1}=Be_{1}=B_{1}, \\
Aw_{1} &=&B[v_{1},w_{1}]^{-1}w_{1}=Be_{2}=B_{2}, \\
Av_{i} &=&B[v_{1},w_{1}]^{-1}v_{i}=B%
\begin{pmatrix}
x_i \\
y_i%
\end{pmatrix}%
=x_{i}B_{1}+y_{i}B_{2}, \\
Aw_{i} &=&B[v_{1},w_{1}]^{-1}w_{i}=B%
\begin{pmatrix}
x_i^{\prime } \\
y_i^{\prime }%
\end{pmatrix}%
=x_{i}^{\prime }B_{1}+y_{i}^{\prime }B_{2}.
\end{eqnarray*}%
Let $\phi $ be the angle between $B_{1}$ and $B_{2}.$ Since $B$ is invertible, we
know that $\phi \in (0,\pi ).$ The identities $\parallel Av_{i}\parallel
=\parallel Aw_{i}\parallel ,i=1,2,...,k$, and the law of cosine imply that%
\begin{eqnarray*}
\parallel B_{1}\parallel &=&\parallel B_{2}\parallel , \\
\parallel x_{i}B_{1}+y_{i}B_{2}\parallel &=&\parallel x_{i}^{\prime
}B_{1}+y_{i}^{\prime }B_{2}\parallel , \\
\parallel B_{1}\parallel (|x_{i}|^{2}+|y_{i}|^{2}-2|x_{i}y_{i}|\cos \phi
)&=&\parallel B_{2}\parallel (|x_{i}^{\prime }|^{2}+|y_{i}^{\prime
}|^{2}-2|x_{i}^{\prime }y_{i}^{\prime }|\cos \phi ),i=2,...,k.
\end{eqnarray*}
Conversely, choose two nonzero vectors $B_{1},B_{2}$ with the same length
and angle $\phi$. Take $A=[B_{1},B_{2}][v_{1},w_{1}]^{-1}$ to get that
$\parallel Av_{i}\parallel =\parallel Aw_{i}\parallel ,i=1,2,...,k,$
implying $G$ is $\mathrm{CAT(0)}$ by Lemma \ref{lem1}.
\end{proof}
\begin{example}\label{exam:Gpq-CAT(0)}
Consider the tubular group $G = G(\{p_1,q_1\},\{p_2,q_2\})$ in the Introduction. It has the following tubular presentation:
$$\langle a,b,c,t\mid [a,t]=1, b^{-1} a^{-p_1}tb = a^{q_1}t,c^{-1} a^{-p_2}tc = a^{q_2}t \rangle.$$
Let us first replace $t$ by $t' = a^{-p_1} t$, and get a slightly different tubular presentation for $G$:
$$\langle a,b,c,t\mid [a,t']=1, b^{-1} t' b = a^{p_1+ q_1}t',c^{-1} a^{p_1-p_2}t'c = a^{p_1+q_2}t' \rangle.$$
We will show the $\mathrm{CAT(0)}$ness of $G$ under the assumption that $p_1 - q_1 = p_2 -q_2$. Let $k= p_1 - q_1 = p_2 -q_2$, then $q_1 = p_1-k$ and $q_2 = p_2-k$. Let further $l=p_1-p_2$, then $p_2 = p_1 -l$. The presentation of $G$ is reduced to $$\langle a,b,c,t\mid [a,t']=1, b^{-1} t' b = a^{2p_1-k}t',c^{-1} a^{l}t'c = a^{2p_1-k-l}t' \rangle,$$
where $p_1,k,l$ can be any integer. Let $m=2p_1-k =p_1+q_1$, then the presentation of $G$ is further reduced to
$$\langle a,b,c,t\mid [a,t']=1, b^{-1} t' b = a^{m}t',c^{-1} a^{l}t'c = a^{m-l}t' \rangle,$$
for some integers $m,l$. Therefore, $G$ is the tubular group with a single vertex group $\mathbb{Z}^2$ and edge relations given by
\begin{eqnarray*}
b^{-1}%
\begin{bmatrix}
0 \\
1%
\end{bmatrix}%
b &=&%
\begin{bmatrix}
m \\
1%
\end{bmatrix}%
, \\
c^{-1}%
\begin{bmatrix}
l \\
1%
\end{bmatrix}%
c &=&%
\begin{bmatrix}
m-l\\
1%
\end{bmatrix}%
.
\end{eqnarray*}
We have
$$ \begin{pmatrix}
0 & m \\
1 & 1
\end{pmatrix}^{-1} \begin{pmatrix}
l \\
1
\end{pmatrix} = \frac{1}{m}\begin{pmatrix}
m-l \\
l
\end{pmatrix}, $$
$$ \begin{pmatrix}
0 & m \\
1 & 1
\end{pmatrix}^{-1} \begin{pmatrix}
m-l \\
1
\end{pmatrix} = \frac{1}{m}\begin{pmatrix}
l\\
m-l
\end{pmatrix}.$$
To show that $G$ is $\mathrm{CAT(0)}$, by Theorem \ref{main} (1), it suffices to find $\phi\in (0,\pi)$ such that
$$(m-l)^2 + l^2 -2\cos(\phi)|(m-l)l|= (l)^2 + (m-l)^2 -2\cos(\phi)|(l)(m-l)| .$$
Note now that the equality holds for any $\phi\in(0,\pi)$, implying that $G$ is indeed $\mathrm{CAT(0)}$.
\end{example}
\section{Free-by-cyclic groups}
In this section, we study when a tubular group is free-by-cyclic and $F_n$-by-$\mathbb{Z}$. Let us start by proving Theorem \ref{main} (2).
\begin{theorem}
\label{th2}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle$. Then $G$ is
free-by-cyclic if and only if $v_{1}-w_{1}$, $v_{2}-w_{2},...,v_{k}-w_{k}$
lie in a common line $l$ which doesn't contain any element of $%
\{v_{1},v_{2},...,v_{k}\}.$
\end{theorem}
\begin{proof}
It is already known by Button \cite[Theorem 2.1]{bu} that a tubular group is
free-by-cyclic if and only if there exists a homomorphism from $G$ to $%
\mathbb{Z}$, which is non-zero on every edge group.
If $v_{1}-w_{1}$ is parallel to any $v_{i}-w_{i},i=2,...,k,$ and one of them
is nonzero, suppose that $v=%
\begin{pmatrix}
a \\
b%
\end{pmatrix}%
$ is a primitive element in $\mathbb{Z}^{2}$ such that $%
v_{1}-w_{1},v_{2}-w_{2},...,v_{k}-w_{k}\in \langle v\rangle .$ Take $%
v^{\prime }=%
\begin{pmatrix}
c \\
d%
\end{pmatrix}%
$ be another primitive element such that $\{v,v^{\prime }\}$ is a basis for $%
\mathbb{Z}^{2}$ (for example, choose integers $c,d$ such that $ad-bc=1$). Let
\begin{equation*}
\mathrm{proj}:\mathbb{Z}^{2}\hookrightarrow \mathbb{R}^{2}\rightarrow \mathbb{Z}
\end{equation*}%
be the projection onto the component $\mathbb{R}v^{\prime }$ (i.e. for any $x=x_1 v+ x_2 v^{\prime}$, we have $\mathrm{proj}(x)=x_2 $). Define
\begin{eqnarray*}
f &:&G\rightarrow \mathbb{Z}, \\
s_{i} &\rightarrow &0, \\
x &\rightarrow &\mathrm{proj}(x),x\in \mathbb{Z}^{2}.
\end{eqnarray*}%
Since $\mathrm{proj}(v_{i})=\mathrm{proj}(w_{i}),i=1,...,k,$ we know that $f$
is a well-defined group homomorphism. Since the line $\mathbb{R}v$
containing $v_{1}-v_{2},v_{2}-w_{2},...,v_{k}-w_{k}$ does not contain any of
$\{v_{1},v_{2},...,v_{k}\},$ we know that $\mathrm{proj}(v_{i}),\mathrm{proj}%
(w_{i}),i=1,...,k,$ are nonzero. The criterion of Button proves that $G$ is
free-by-cyclic.
If all $v_{i}-w_{i}=0,$ let $\mathbb{R}%
\begin{pmatrix}
a \\
b%
\end{pmatrix}%
$ (for a primitive vector $v=
\begin{pmatrix}
a \\
b%
\end{pmatrix}%
\in \mathbb{Z}^{2}$) be a line not containing any $v_i$.
Take $v^{\prime }=%
\begin{pmatrix}
c \\
d%
\end{pmatrix}%
$ be another primitive element such that $\{v,v^{\prime }\}$ is a basis for $%
\mathbb{Z}^{2}.$ Let
\begin{equation*}
\mathrm{proj}:\mathbb{Z}^{2}\hookrightarrow \mathbb{R}^{2}\rightarrow \mathbb{Z}
\end{equation*}%
the projection onto the component $\mathbb{R}v^{\prime }.$ In this case, we
have nonzero $\mathrm{proj}(v_{i}),i=1,2,...,k.$ A similar argument finishes
the proof.
Conversely, if for some $i\neq j,$ the two vectors $v_{i}-w_{i}$, $%
v_{j}-w_{j}$ are not parallel, then $\mathrm{Span}_{\mathbb{Z}%
}(v_{i}-w_{i},v_{j}-w_{j}\}$ is a finite index subgroup of $\mathbb{Z}^{2}.$
For any group homomorphism $f:\mathbb{Z}^{2}\rightarrow \mathbb{Z}$ with $%
f(v_{i})=f(w_{i}),f(v_{j})=f(w_{j}),$ it has $\mathrm{Span}_{\mathbb{Z}%
}(v_{i}-w_{i},v_{j}-w_{j}\}\subset \ker f,$ implying $f$ is actually
trivial. This implies $f(v_{i})=0.$ Therefore, $G$ cannot have a
homomorphism to $\mathbb{Z}$ which is non-zero on every edge group.
If $v_{1}-w_{1}$, $v_{2}-w_{2},...,v_{k}-w_{k}$ lie in a common line $l$,
but $l$ contains some vector $v_{i}$ (or $w_{i}$), any group homomorphism $f:%
\mathbb{Z}^{2}\rightarrow \mathbb{Z}$ with $f(v_{j})=f(w_{j}),j=1,2,...,k,$
has $f(v_{i})=0.$ Again, $G$ cannot have a homomorphism to $\mathbb{Z}$
which is non-zero on every edge group.
\end{proof}
We now prove the following theorem.
\begin{theorem}
\label{th10} If a tubular
group is free-by-cyclic (resp. virtually free-by-cyclic), then it is $F_{n}$-by-$\mathbb{Z}$ (resp. virtually $F_{n}$-by-$\mathbb{Z}$).
\end{theorem}
To prove Theorem \ref{th10}, we need the following lemma.
\begin{lemma}
\label{leml2}Suppose that $G$ is finitely generated free-by-cyclic. The
group $G$ is $F_{n}$-by-$\mathbb{Z}$ if and only if the first $L^{2}$-Betti
number $b_{1}^{(2)}(G)=0.$
\end{lemma}
\begin{proof}
Suppose that $G=F_{n}\rtimes \mathbb{Z}$. A result of L\"{u}ck \cite[Theorem 2.1]{luck}
implies that a mapping torus of a finite CW complex has its
first $L^{2}$-Betti number vanishing. This gives that $b_{1}^{(2)}(G)=0.$
Conversely, let $G$ be a finitely generated free-by-cyclic group with $%
b_{1}^{(2)}(G)=0.$ Let us recall \cite[Proposition 2.3]{gkl} of Gardam--Kielak--Logan. Let $G$ be a torsion-free group that satisfies the Atiyah
conjecture. Let $\phi:G\rightarrow \mathbb{Z}$ be an epimorphism whose
kernel $K$ is a free product of finitely generated groups. If the first $%
L^{2}$-Betti number of $G$ is equal to zero, then $K$ is finitely generated.
In our situation, since $G$ is free-by-cyclic, it satisfies the Atiyah
conjecture (by the work of Linnell \cite{Lin}). Therefore, $K=\ker \phi $
is finitely generated.
\end{proof}
\begin{proof}[Proof of Theorem \protect\ref{th10}]
Chatterji--Hughes--Kropholler \cite[Theorem 1.5]{chk} prove that when a group
$G$ acts on a tree $T$ with $V,E$ as the $G$-orbits of vertices and edges$,$
we have
\begin{equation*}
b_{1}^{(G)}=\sum_{v\in V}(b_{1}^{(2)}(G_{v})-\frac{1}{|G_{v}|})+\sum_{e\in E}%
\frac{1}{|G_{e}|},
\end{equation*}%
if $b_{1}^{(2)}(G_{e})=0$ for each $e\in E.$ Let $G$ be a tubular group,
which is a finite graph of groups with vertex group $\mathbb{Z}^{2}$ and
edge group $\mathbb{Z}.$ The group $G$ acts on its Bass-Serre tree. Since $%
\mathbb{Z}^{2}$ has vanishing $L^{2}$-Betti numbers, we know that $%
b_{1}^{(2)}(G)=0$ by the above-mentioned result of
Chatterji--Hughes--Kropholler. Lemma \ref{leml2} implies that a
tubular group $G$ is free-by-cyclic if and only if it is $F_{n}$-by-$\mathbb{%
Z}$. Note that a finite index subgroup of a tubular group is tubular. The
case of virtually free-by-cyclic can be proved similarly.
\end{proof}
\section{Virtual specialness}
In this section, we study the virtual specialness of some tubular free-by-cyclic groups.
Let us first recall some results related to actions of tubular groups on CAT (0) cube complexes, see \cite{wi14, Wood2, Wood} for more information. Wise obtained in \cite{wi14} a criterion for a tubular group to act freely on
a (possibly infinite dimensional) $\mathrm{CAT(0)}$ cube complex.
Suppose that a tubular group $G$ is a group that splits as a finite graph $\Gamma$ of groups where each vertex group $G_v$ is isomorphic to $\mathbb{Z}^2$ and each edge group $G_e$ is isomorphic to $\mathbb{Z}$. For each edge $e$ of $\Gamma$, let $v(e_{\rightarrow })$ (resp. $v(e_{\leftarrow })$) denote the initial vertex (resp. the terminal vertex) and $e_{\rightarrow }$ (resp. $e_{\leftarrow}$) denote the generator of the edge
subgroup in $G_{v(e_{\rightarrow })}$ (resp. $G_{v(e_{\leftarrow })}$). For elements $s,v \in \mathbb{Z}^2$, let $\langle
s,v\rangle$ be the subgroup generated by $s,v$. Note that for $$s=%
\begin{bmatrix}
x \\
y%
\end{bmatrix}%
,v=%
\begin{bmatrix}
x^{\prime } \\
y^{\prime }%
\end{bmatrix}%
,$$ the index $[\mathbb{Z}^{2},\langle s,v\rangle ]=|\det [s,v]|=\#[s,v]$, which is the geometric intersection number of curves in the torus $T^2$ (with $\pi_1(T^2)=\mathbb{Z}^2$) represented by $s,v$. Abusing notation, we also use $s$ to denote the corresponding closed curve in $T^2$.
\begin{lemma}
A tubular group $G(\Gamma )$ acts freely on a $\mathrm{CAT(0)}$ cube complex if and only if
there exists a finite subset $S_{v}\subseteq G_{v}=\mathbb{Z}^{2}$ for each
vertex $v$ satisfying
1) For each edge $e$,
\begin{equation*}
\sum_{s\in S_{v(e_{\rightarrow })}}|[\mathbb{Z}^{2},\langle s,e_{\rightarrow
}\rangle ]|=\sum_{s\in S_{v(e_{\leftarrow })}}|[\mathbb{Z}^{2},\langle
s,e_{\leftarrow }\rangle ]|;
\end{equation*}
2) $\langle S_{v}\rangle $ is of finite index in $G_{v}.$
\end{lemma}
Such a collection of finite sets $\{
S_{v}\}_{v\in \Gamma^0}$ is called an equitable set. The tubular group $G(\Gamma)$ is the fundamental group of a graph of spaces $X(\Gamma)$, with vertex space a torus $X_v=S^1 \times S^1$ for each vertex $v$ and edge space $X_e=S^1$ for each edge $e$ of $\Gamma$. The space $X(\Gamma)$ is obtained from the union of $X_v=S^1 \times S^1$ by attaching cylinders $X_e \times [-1,1]$. An immersed wall is either of the form $X_e \times 0$
or a connected component of a graph built from the disjoint union of all the circles in $S_v=\{ \sigma_{vj} : v \in \Gamma^0, 1 \leq j \leq |S_v| \}$ together with some connecting arcs joining points in $e_{\leftarrow} \cap \sigma_{v(e_{\leftarrow}),i} $ bijectively with points in $e_{\rightarrow} \cap \sigma_{v(e_{\rightarrow}),i'}$ for each edge $e$. Here one uses the fact that $\{
S_{v}\}_{v\in \Gamma^0}$ is equitable, so the number of intersections is the same.
Woodhouse shows in \cite[Theorem 1.1]{Wood} that a tubular group $G$ is virtually special if and only if it acts freely on a finite-dimensional $\mathrm{CAT(0)}$ cube complex. He also proves in \cite[Theorem 1.2]{Wood2} that the dual cube complex $C(\tilde{X}, W)$ (the wall space defined by Wise \cite{wi14}) is finite dimensional if and only if each immersed wall is non-dilated (as described in the following).
Let $\Lambda $ be an immersed wall (not consisting of a single immersed circle) and let $q:\Lambda \rightarrow \Omega $
be the quotient map obtained by quotienting each circle in the equitable set $S_v$
to a vertex. The dilation function $R:\pi _{1}(\Lambda )\rightarrow \mathbb{Q%
}^{\ast }$ factors through $q_{\ast }:\pi _{1}(\Lambda )\rightarrow \pi
_{1}(\Omega )$ \cite[Section 5]{Wood2}:
\begin{equation*}
\begin{array}{ccc}
\pi _{1}(\Lambda ) & \overset{R}{\rightarrow } & \mathbb{Q}^{\ast } \\
q_{\ast }\downarrow & \nearrow R_{\ast } & \\
\pi _{1}(\Omega ) & &
\end{array}%
\end{equation*}%
Given an orientation to each edge in $\Omega $ such that all edges in the
same edge space have the same orientations. Let $X_{e}$ be an edge space in $%
X$, and let $\alpha $ be an arc in $X_{e}$ connecting the circles $%
C_{\rightarrow}\looparrowright X_{v({e_{\rightarrow})}},C_{\leftarrow}\looparrowright X_{v(e_{\leftarrow})}.$ Let $\alpha
_{\leftarrow},\alpha _{\rightarrow}$ be the corresponding elements of $C_{\leftarrow},C_{\rightarrow}$ in the
equitable set. Define%
\begin{equation*}
\omega (\alpha )=\frac{|\det [e_{\rightarrow },\alpha _{\rightarrow}]|}{|\det
[e_{\leftarrow },a_{\leftarrow}]|}.
\end{equation*}%
If $\gamma =\sigma _{1}^{\varepsilon _{1}}\sigma _{2}^{\varepsilon
_{2}}\cdots \sigma _{n}^{\varepsilon _{n}}$ is an edge path in $\Omega $
where $\sigma _{i}$ is an oriented arc in $\Omega $, and $\varepsilon
_{i}=\pm 1,$ then define $\omega (\alpha )=\omega (\sigma _{1})^{\varepsilon
_{1}}\omega (\sigma _{2})^{\varepsilon _{2}}\cdots \omega (\sigma
_{n})^{\varepsilon _{n}}$. Moreover, when $\alpha \in \pi_1(\Omega)$, we declare that $R^\ast(\alpha):= \omega(\alpha)$. The immersed wall $\Lambda $ is called
dilated if $\mathrm{Im}R=\mathrm{Im}R_{\ast }$ is infinite. Otherwise, an immersed wall is called non-dilated (see Woodhouse \cite[Definition 4.5]{Wood2}).
Theorem \ref{main} (3) is the following.
\begin{theorem}
\label{th3}When $\{v_{1},w_{1}\}$ are linearly independent and
\begin{eqnarray*}
\det [w_{1}-v_{1},v_{i}] &=&\pm \det [w_{1}-v_{1},w_{i}], \\
\det [w_{1}+v_{1},v_{i}] &=&\pm \det [w_{1}+v_{1},w_{i}],i=2,...,k,
\end{eqnarray*}%
the group $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle $ is virtually
special.
\end{theorem}
\begin{proof}
Choose
\begin{equation*}
S=\{z_{1}:=w_{1}-v_{1},z_{2}:=w_{1}+v_{1}\}.
\end{equation*}%
By Woodhouse \cite[Theorem 1.1]{Wood}, it is enough to prove that $G$ acts freely
on a finite-dimensional $\mathrm{CAT(0)}$ cube complex. According to \cite[Theorem 1.2]{Wood2}, it is enough to show that each immersed wall is non-dilated (so that the wall space $C(\tilde{X}, W)$ is finite dimensional).
Note that $S$ consists of two elements and an immersed wall is supported on the two
elements. For each stable letter $s_i$, denote by $s_i^{-},s_i^{+}$ the two curves in the vertex space represented by $v_i,w_i$. The intersection numbers are (assuming $i\geq 2$)%
\begin{eqnarray*}
\#[z_{1},s_{1}^{-}] &=&|\det [w_{1}-v_{1},v_{1}]|=|\det [w_{1},v_{1}]|=|\det
[w_{1}-v_{1},w_{1}]=\#[z_{1},s_{1}^{+}], \\
\#[z_{2},s_{1}^{-}] &=&|\det [w_{1}+v_{1},v_{1}]|=|\det [w_{1},v_{1}]|=|\det
[w_{1}+v_{1},w_{1}]=\#[z_{2},s_{1}^{+}], \\
\#[z_{1},s_{i}^{-}] &=&|\det [w_{1}-v_{1},v_{i}]|=|\det
[w_{1}-v_{1},w_{i}]|=\#[z_{1},s_{i}^{+}], \\
\#[z_{2},s_{i}^{-}] &=&|\det [w_{1}+v_{1},v_{i}]|=|\det
[w_{1}+v_{1},w_{i}]|=\#[z_{2},s_{i}^{+}].
\end{eqnarray*}%
Since the two ends of any edge space have the same intersection numbers
with $z_{i},i=1,2$, we can choose immersed walls $\Lambda $ connecting the
intersection points in $z_{i}\cap e_{s_{i}\rightarrow }$ to $z_{i}\cap
e_{s_{i}\leftarrow }$ as in Figure \ref{fig:graph}.
\begin{figure}
\centering
\includegraphics[width=0.8\linewidth]{pic1.pdf}
\caption{Immersed walls}
\label{fig:graph}
\end{figure}
Let $\Omega $ be the quotient obtained from $\Lambda $ by quotient each
circle in the equitable set to a vertex. Note that each closed loop in the
quotient $\Omega $ is a composition of edges labeled by $s_{1},$ $%
s_{i},i=2,...,k.$ But each single edge $e_{s_{1}},e_{s_{i}}$ labeled by $%
s_{1},s_{i}$ has the weight
\begin{eqnarray*}
w(e_{s_{1}}) &=&\frac{\#[z_{1},e_{s_{1}}^{-}]}{\#[z_{1},e_{s_{1}}^{+}]}=1=%
\frac{\#[z_{2},e_{s_{1}}^{-}]}{\#[z_{2},e_{s_{1}}^{+}]}, \\
w(e_{s_{i}}) &=&\frac{\#[z_{1},e_{c}^{-}]}{\#[z_{1},e_{c}^{+}]}=1=\frac{%
\#[z_{2},e_{s_{i}}^{-}]}{\#[z_{2},e_{s_{i}}^{+}]}.
\end{eqnarray*}%
Therefore, each immersed wall is non-dilated and the group $G$ is virtually
special. When there is
only one vertex space, it is also noted by Wise \cite[Remark 3.6]{wi14}) that it is enough to prove the geometric intersection numbers $\#[z_{1},s_{i}^{-}]=%
\#[z_{1},s_{i}^{+}]$ for any $i=1,2,...,k$ to have a finite-dimensional wall space.
\end{proof}
\begin{corollary}
Suppose that the group $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2\rangle $ is free-by-cyclic. When
$\{v_{1},w_{1}\}$ are linearly independent and $\det [v_{1},v_{2}]=\det
[v_{1},w_{2}],$ the group $G$ is virtually special.
\end{corollary}
\begin{proof}
When $G$ is free-by-cyclic, we know that $w_{1}-v_{1}$ is parrell to $%
w_{2}-v_{2}$ by Theorem \ref{th2}. This implies that $\det [w_{1}-v_{1},w_{2}-v_{2}]=0$ and $\det
[w_{1}-v_{1},v_{2}]=\det [w_{1}-v_{1},w_{2}].$ Now
\begin{eqnarray*}
\det [w_{1}+v_{1},v_{2}] &=&\det [w_{1}-v_{1},v_{2}]+\det
[2v_{1},v_{2}], \\
\det [w_{1}+v_{1},w_{2}]&=&\det [w_{1}-v_{1},w_{2}]+\det [2v_{1},w_{2}].
\end{eqnarray*}
When $\det [v_{1},v_{2}]=\det [v_{1},w_{2}],$ Theorem \ref{th3} finishes
the proof.
\end{proof}
Now we consider $\mathrm{CAT(0)}$ free-by-cyclic tubular groups.
\begin{lemma}
\label{lemcat}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle $ be a multiple HNN
extension of the $\mathbb{Z}^{2}$. Suppose that%
\begin{equation*}
v_{1}=%
\begin{pmatrix}
k_{1} \\
0%
\end{pmatrix}%
,w_{1}=%
\begin{pmatrix}
0 \\
k_{2}%
\end{pmatrix}%
\end{equation*}%
for some nonzero integers $k_{1},k_{2}$ and
\begin{equation*}
v_{i}=%
\begin{pmatrix}
a_{i} \\
b_{i}%
\end{pmatrix}%
,w_{i}=%
\begin{pmatrix}
c_{i} \\
d_{i}%
\end{pmatrix}%
,i=2,...,k.
\end{equation*}%
If $G$ is $\mathrm{CAT(0)}$ and free-by-cyclic, then
\begin{equation*}
a_{i}b_{i}=c_{i}d_{i}
\end{equation*}%
for each $i=2,...,k.$
\end{lemma}
\begin{proof}
When $G$ is free-by-cyclic, Theorem \ref{th2} implies that $v_{i}-w_{i}=%
\begin{pmatrix}
a_{i}-c_{i} \\
b_{i}-d_{i}%
\end{pmatrix}%
$ is parallel to $v_{1}-w_{1}=%
\begin{pmatrix}
k_{1} \\
-k_{2}%
\end{pmatrix}%
.$ Therefore, we have%
\begin{equation}
\frac{a_{i}-c_{i}}{k_{1}}=\frac{b_{i}-d_{i}}{-k_{2}} \tag{Eq1}
\end{equation}%
for each $i=2,...,k.$ When $G$ is $\mathrm{CAT(0)}$, Theorem \ref{th1} implies that
\begin{equation*}
(\frac{a_{i}}{k_{1}})^{2}+(\frac{b_{i}}{k_{2}})^{2}-|\frac{2a_{i}b_{i}}{%
k_{1}k_{2}}|\cos \phi =(\frac{c_{i}}{k_{1}})^{2}+(\frac{d_{i}}{k_{2}})^{2}-%
|\frac{2c_{i}d_{i}}{k_{1}k_{2}}|\cos \phi
\end{equation*}%
for some common $\phi \in (0,\pi )$. So we have
\begin{eqnarray*}
(\frac{a_{i}}{k_{1}})^{2}-(\frac{c_{i}}{k_{1}})^{2}+(\frac{b_{i}}{k_{2}}%
)^{2}-(\frac{d_{i}}{k_{2}})^{2} &=&(|\frac{2a_{i}b_{i}}{k_{1}k_{2}}|-|\frac{%
2c_{i}d_{i}}{k_{1}k_{2}}|)\cos \phi , \\
\frac{(a_{i}-c_{i})}{k_{1}}\frac{(a_{i}+c_{i})}{k_{1}}+\frac{(b_{i}-d_{i})}{%
k_{2}}\frac{(b_{i}+d_{i})}{k_{2}} &=&\frac{2(|a_{i}b_{i}|-|c_{i}d_{i}|)}{|k_{1}k_{2}|%
}\cos \phi , \\
-\frac{(b_{i}-d_{i})(a_{i}+c_{i})}{k_{1}k_{2}}-\frac{%
(a_{i}-c_{i})(b_{i}+d_{i})}{k_{1}k_{2}} &=&\frac{2(|a_{i}b_{i}|-|c_{i}d_{i}|)}{%
|k_{1}k_{2}|}\cos \phi \text{ (by (Eq1))}, \\
-(a_{i}b_{i}-c_{i}d_{i}) &=&\pm (|a_{i}b_{i}|-|c_{i}d_{i}|)\cos \phi .
\end{eqnarray*}%
Since $|x-y|\geq ||x|-|y||$ for any real numbers $x,y$ and $\cos \phi \neq \pm 1,$ we get that $a_{i}b_{i}=c_{i}d_{i}$ for each $%
i=2,...,k.$
\end{proof}
%The following is Theorem \ref{th4}.
\begin{theorem}\label{thm:fr-cyc-vspel}
Let $G=\langle \mathbb{Z}^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k%
\rangle $ be a $\mathrm{CAT(0)}$ free-by-cyclic tubular group. Then $G$ is virtually
special.
\end{theorem}
\begin{proof}
When $v_{i},w_{i}$ are linearly independent for some $i$, we assume $i=1$
without loss of generality. There are linearly independent primitive
elements $e_{1},e_{2}\in \mathbb{Z}^{2}$ such that $v_{1}=e_{1}^{k_{1}}$ and
$w_{1}=e_{2}^{k_{2}}$ for some positive integers $k_{1},k_{2}.$ After a
change of basis, we assume that
\begin{equation*}
v_{1}=%
\begin{pmatrix}
k_{1} \\
0%
\end{pmatrix}%
,w_{1}=%
\begin{pmatrix}
0 \\
k_{2}%
\end{pmatrix}%
.
\end{equation*}%
and
\begin{equation*}
v_{i}=%
\begin{pmatrix}
a_{i} \\
b_{i}%
\end{pmatrix}%
,w_{i}=%
\begin{pmatrix}
c_{i} \\
d_{i}%
\end{pmatrix}%
,i=2,...,k.
\end{equation*}%
By Lemma \ref{lemcat}, we have%
\begin{equation}
a_{i}b_{i}=c_{i}d_{i} \tag{Eq2}
\end{equation}%
for each $i=2,...,k.$ In order to prove that $G$ is virtually special, it is
enough to check the conditions of Theorem \ref{th3}. We assume
\begin{equation*}
v_{i}=%
\begin{pmatrix}
a_{i} \\
b_{i}%
\end{pmatrix}%
,w_{i}=%
\begin{pmatrix}
c_{i} \\
d_{i}%
\end{pmatrix}%
,
\end{equation*}%
for any $i=2,...,k.$ It is direct that
\begin{eqnarray*}
\det [w_{1}-v_{1},v_{i}] &=&\det
\begin{bmatrix}
-k_{1} & a_{i} \\
k_{2} & b_{i}%
\end{bmatrix}%
=-k_{1}b_{i}-k_{2}a_{i} \\
\det [w_{1}-v_{1},w_{i}] &=&\det
\begin{bmatrix}
-k_{1} & c_{i} \\
k_{2} & d_{i}%
\end{bmatrix}%
=-k_{1}d_{i}-k_{2}c_{i}.
\end{eqnarray*}%
By the identity (Eq1), $-k_{2}(a_{i}-c_{i})=k_{1}(b_{i}-d_{i}),$ implying $$%
-k_{1}b_{i}-k_{2}a_{i}=-k_{1}d_{i}-k_{2}c_{i}.$$
It remains to prove $\det [w_{1}+v_{1},v_{i}]=\pm \det [w_{1}+v_{1},w_{i}],$
which is equivalent to
\begin{equation}
k_{1}b_{i}-k_{2}a_{i}=\pm (k_{1}d_{i}-k_{2}c_{i}). \tag{Eq3}
\end{equation}%
Consider the identity (Eq1). If $b_{i}-d_{i}=0,$ then $a_{i}-c_{i}=0.$ This
implies that $$k_{1}b_{i}-k_{2}a_{i}=k_{1}d_{i}-k_{2}c_{i},$$ proving (Eq3)
with a positive sign.
If $b_{i}-d_{i}\neq 0,$ the identity (Eq1) gives%
\begin{eqnarray*}
\frac{k_{1}}{k_{2}} &=&\frac{c_{i}-a_{i}}{d_{i}-b_{i}} \\
&=&\frac{a_{i}c_{i}-a_{i}^{2}}{a_{i}b_{i}-a_{i}d_{i}}\text{ if }a_{i}\neq 0.
\end{eqnarray*}%
If $a_{i}=0,$ the identity (Eq1) gives $k_{2}c_{i}=k_{1}(b_{i}-d_{i}).$ The
identity (Eq2) gives that either $c_{i}=0$ or $d_{i}=0.$ For the former, we
have $b_{i}=d_{i},$ proving (Eq3) with positive sign. For the latter, we
have $k_{2}c_{i}=k_{1}b_{i}$, proving (Eq3) with negative sign.
If $a_{i}\neq 0,$ we continue to write $\frac{k_{1}}{k_{2}}$ as
\begin{equation*}
\frac{k_{1}}{k_{2}}=\frac{a_{i}c_{i}-a_{i}^{2}}{a_{i}b_{i}-a_{i}d_{i}}=\frac{%
a_{i}c_{i}-a_{i}^{2}}{c_{i}d_{i}-a_{i}d_{i}}
\end{equation*}%
using (Eq2). If $a_{i}-c_{i}=0,$ (Eq1) implies that $b_{i}=d_{i}$
proving (Eq3) with a positive sign. If $a_{i}-c_{i}\neq 0,$ we have%
\begin{equation*}
\frac{k_{1}}{k_{2}}=\frac{a_{i}}{d_{i}}.
\end{equation*}%
Consider the (Eq2). If $b_{i}=0,$ we have either $c_{i}=0$ or $d_{i}=0.$ For
the former, (Eq1) implies that $-k_{2}a_{i}=-k_{1}d_{i},$ proving the
(Eq3) with a negative sign. For the latter, (Eq1) implies that $%
-k_{2}(a_{i}-c_{i})=0,$ proving the (Eq3) with a negative sign. If $%
b_{i}\neq 0,$ the (Eq2) implies
\begin{equation*}
\frac{k_{1}}{k_{2}}=\frac{a_{i}}{d_{i}}=\frac{c_{i}}{b_{i}}.
\end{equation*}%
When $b_{i}+d_{i}= 0,$ we have $a_i +c_i =0$ by (Eq2), which proves (Eq3) with a negative sign. When $b_{i}+d_{i}\neq 0,$ we have%
\begin{eqnarray*}
\frac{k_{1}}{k_{2}} &=&\frac{c_{i}}{b_{i}}=\frac{c_{i}b_{i}+c_{i}d_{i}}{%
b_{i}^{2}+b_{i}d_{i}} \\
\overset{\text{(Eq2)}}{=}\frac{c_{i}b_{i}+a_{i}b_{i}}{b_{i}^{2}+b_{i}d_{i}}
&=&\frac{c_{i}+a_{i}}{b_{i}+d_{i}}.
\end{eqnarray*}%
This finally proves (Eq3) with a negative sign.
When $v_{i},w_{i}$ are parallel for any $i=1,2,...,k,$ there are primitive
elements $u_{i}\in \mathbb{Z}^{2}$ and integers $m_{i},n_{i}$ such that $%
v_{i}=u_{i}^{m_{i}},w_{i}=u_{i}^{n_{i}}.$ When $G$ is $\mathrm{CAT(0)}$, we have $%
m_{i}=\pm n_{i}.$ When $G$ is furthermore free-by-cyclic, Theorem \ref{th2}
implies that $m_i=n_i$ for any $i=2,...,k.$ Choose a primitive
element $z$ such that $\det [u_{1},z]=1.$ Consider the graph of spaces
associated to the tubular group $G.$ Then the set $S=\{u_{1},z\}$ is an equitable
set. The two ends of any edge space have the same intersection numbers
with $z_{i},i=1,2$, both $|m_{i}|=|n_{i}|.$ Therefore, each immersed wall is non-dilated. This proves that $G$ is virtually special by Woodhouse's criterion\cite[Theorem 1.1]{Wood}, using a similar argument as the proof of Theorem \ref{th3}.
\end{proof}
\begin{corollary}
\label{cor1}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle $ be a
free-by-cyclic tubular group. Then $G$ is $\mathrm{CAT(0)}$ if and only if $G$ is
virtually special.
\end{corollary}
\begin{proof}
Wise proves that a virtually special tubular group is $\mathrm{CAT(0)}$ (cf.
\cite[Lemma 4.4]{wi14}). Theorem \ref{th4} proves the other direction.
\end{proof}
The following lemma improves Wise's result \cite[Corollary 5.10]{wi14} to the virtual setting.
\begin{lemma}\label{lem:vir-cocom--cube}
Let $G$ be a tubular group.
\begin{enumerate}
\item Suppose that there are more than two parallelism classes of edge groups in some vertex group $G_v$. Then $G$ cannot virtually act freely and cocompactly on a $\mathrm{CAT(0)}$ cube complex.
\item Suppose that $G$ has no Baumslag--Solitar subgroup $\langle a,t\mid ta^mt^{-1} =a^n\rangle $ where $m\neq \pm n$. Then the following are equivalent:
\begin{enumerate}[label=(\roman*)]
\item $G$ act freely and cocompactly on a $\mathrm{CAT(0)}$ cube complex.
\item $G$ virtually act freely and cocompactly on a $\mathrm{CAT(0)}$ cube complex.
\item There are exactly one or two parallelism classes of edge groups in each vertex group.
\end{enumerate}
\end{enumerate}
\end{lemma}
\begin{proof}
Let us prove (1) first. By \cite[Corollary 5.10]{wi14}, $G$ cannot act freely and cocompactly on a $\mathrm{CAT(0)}$ cube
complex. Let $T$ be the corresponding Bass--Serre tree of $G$ and $H$ be any finite index subgroup of $G$. Then the finite-index subgroup $H$ also acts on $T$ cocompactly with $\mathbb{Z}^2$ as vertex groups and $\mathbb{Z}$ as edge groups, hence it is a tubular group. The corresponding Bass--Serre tree of $H$ is again $T$. Suppose $G_v$ is an edge group in $G$ that has more than $3$ parallelism classes of edge groups, say $G_{e_1}, G_{e_2}$ and $G_{e_3}$. View $[G_v] \in G/G_v$ as a vertex in $T$. Then $G_{e_1}, G_{e_2}$ and $G_{e_3}$ represent the stabilizers of three edges that connect to $G_v$. Consider now the action of $H$ on $T$. The stabilizer of $[G_v]$ under the action of $H$ is the finite index subgroup $G_v \cap H$, and the corresponding stabilizer $H_{e_i}$ of $e_i$ is the finite index subgroup $G_{e_i}\cap H$, $i=1,2,3$. Since $G_{e_1}, G_{e_2}$ and $G_{e_3}$ represent $3$ parallelism classes of edge groups in the vertex group $G_v$, we must also have $H_{e_1}$, $H_{e_2}$, $H_{e_3}$ represent $3$ parallelism classes of edge groups in $H_v$. Applying \cite[Corollary 5.10]{wi14} again, we have that $H$ cannot act freely and cocompactly on a $\mathrm{CAT(0)}$ cube complex.
For (2), $(i) $ implies $(ii)$ trivially, while $(ii)$ implies $(iii)$ following from $(1)$ and $(iii)$ implies $(i)$ following from \cite[Corollary 5.10]{wi14}.
\end{proof}
\begin{corollary}
\label{cor2}Let $G=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2\rangle .$ Suppose that $%
\{v_{1},w_{1}\}$ are linearly independent, and $[v_{1},w_{1}]^{-1}v_{2}=%
\begin{pmatrix}
x \\
y%
\end{pmatrix}%
,[v_{1},w_{1}]^{-1}w_{2}=%
\begin{pmatrix}
x^{\prime } \\
y^{\prime }%
\end{pmatrix}%
.$ Suppose that
\begin{enumerate}
\item $
|x|^{2}+|y|^{2}-2|xy|\cos \phi =|x^{\prime }|^{2}+|y^{\prime
}|^{2}-2|x^{\prime }y^{\prime }|\cos \phi
$
for some $\phi \in (0,\pi )$;
\item $v_{1}-w_{1}$, $v_{2}-w_{2}$ lie in a line $l$ which doesn't contain any
element of $\{v_{1},v_{2},w_{1},w_{2}\}$;
\item $v_{2}$ or $w_{2}$ does not lie in the union of the two lines $\mathbb{R}v_{1}\cup \mathbb{R}%
w_{1}.$
\end{enumerate}
Then $G$ is a $\mathrm{CAT(0)}$, free-by-cyclic, virtually non-cocompact special group.
\end{corollary}
\begin{proof}
According to Theorem \ref{main}, Conditions 1) and 2) imply that
the group $G$ is $\mathrm{CAT(0)}$ and free-by-cyclic. Theorem \ref{thm:fr-cyc-vspel} implies that it is virtually special. It remains to show that $G$ is not virtually cocompactly cubulated. By \Cref{lem:vir-cocom--cube} (1), Condition (3) implies that $G$ cannot be virtually cocompactly
special.
\end{proof}
Let now $G(\{p_i,q_i\}_{i=1}^{n-1})=F_{n}\rtimes _{\phi }\mathbb{Z}$ be as defined in the introduction. Theorem \ref{th7} follows from Corollary \ref{cor1} and the following theorem.
\begin{theorem}
\label{th7-1}The group $G(\{p_i,q_i\}_{i=1}^{n-1})$ is virtually
special if and only if either
1) $p_{i}=-q_{i}$ for each $i=1,2,...,n-1$, or
2) $p_{s}\neq -q_{s}$ for some $s$ and
\begin{equation*}
q_{i}(q_{i}+p_{s}-q_{s})=p_{i}(p_{i}-p_{s}+q_{s})
\end{equation*}%
for each $i=1,2,...,n-1.$
\end{theorem}
\begin{proof}
Note that $G(\{p_i,q_i\}_{i=1}^{n-1})$ is a tubular group \begin{equation*}
G(\{p_i,q_i\}_{i=1}^{n-1}) =\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle
\end{equation*}%
with
\begin{equation*}
v_{i}=%
\begin{pmatrix}
q_{i} \\
1%
\end{pmatrix}%
,w_{i}=%
\begin{pmatrix}
-p_{i} \\
1%
\end{pmatrix}%
,i=1,2,...,n-1.
\end{equation*}%
By Corollary \ref{cor1}, the group is virtually special if and only if it is $\mathrm{CAT(0)}$. If $p_{i}=-q_{i}$ for each $i=1,2,...,n-1,$ it is obvious that the multiple
HNN extension $G(\{p_i,q_i\}_{i=1}^{n-1})$ is $\mathrm{CAT(0)}$. When $p_{s}\neq
-q_{s}$ for some $s,$ without loss of generality we assume $s=1.$ The two
vectors $v_{1},w_{1}$ are linearly independent. Let $[v_{1},w_{1}]$ be the
matrix with columns $v_{1},w_{1}.$ Note that%
\begin{eqnarray*}
\lbrack v_{1},w_{1}]^{-1}v_{i} &=&\frac{1}{p_{1}+q_{1}}%
\begin{pmatrix}
p_{1}+q_{i} \\
q_{1}-q_{i}%
\end{pmatrix}%
, \\
\lbrack v_{1},w_{1}]^{-1}w_{i} &=&\frac{1}{p_{1}+q_{1}}%
\begin{pmatrix}
p_{1}-p_{i} \\
q_{1}+p_{i}%
\end{pmatrix}%
,i=2,...,n-1.
\end{eqnarray*}%
Theorem \ref{th1} implies that $G(\{p_i,q_i\}_{i=1}^{n-1})$ is $\mathrm{CAT(0)}$
if and only if there is a $\phi \in (0,\pi)$ such that%
\begin{eqnarray*}
&&(p_{1}+q_{i})^{2}+(q_{1}-q_{i})^{2}-2|(p_{1}+q_{i})(q_{1}-q_{i})|\cos \phi \\
&=&(p_{1}-p_{i})^{2}+(q_{1}+p_{i})^{2}-2|(p_{1}-p_{i})(q_{1}+p_{i})|\cos \phi ,
\end{eqnarray*}%
which is equivalent to
\begin{eqnarray*}
q_{i}(q_{i}+p_{1}-q_{1})-p_{i}(p_{i}-p_{1}+q_{1}) = (|(p_{1}+q_{i})(q_{1}-q_{i})|-|(p_{1}-p_{i})(q_{1}+p_{i})| ) cos \phi.
\end{eqnarray*}%
Note that the right hand side is $ \leq |-(q_{i}(q_{i}+p_{1}-q_{1})-p_{i}(p_{i}-p_{1}+q_{1}))\cos \phi |$. Therefore, $G(\{p_i,q_i\}_{i=1}^{n-1})$ is $\mathrm{CAT(0)}$ if and only $%
q_{i}(q_{i}+p_{1}-q_{1})-p_{i}(p_{i}-p_{1}+q_{1})=0$.
\end{proof}
\begin{corollary}\label{cor:vir-cpspe}
The group $G(\{p_i,q_i\}_{i=1}^{n-1})$ is virtually compact special if
and only if the set $\{-p_{i},q_{i}:i=1,2,...,n-1\}$ consists of at most two
elements and one of the two conditions of Theorem \ref{th7-1} holds.
\end{corollary}
\begin{proof}
Wise proves that a compact cubulated tubular group is virtually compact
special \cite[Corollary 5.9]{wi14}. Moreover, by \Cref{lem:vir-cocom--cube} a tubular group $G$ is virtually
compact cubulated if and only if there is exactly one or two parallelism
classes of edge groups in each vertex group and $G$ contains no the
Baumslag--Solitar group $BS(m,n)=\langle a,t:ta^{n}t^{-1}=a^{m}\rangle $ with
$n\neq \pm m.$ Note that the group has a presentation of the form
\begin{equation*}
G(\{p_i,q_i\}_{i=1}^{n-1})=\langle \mathbb{Z}%
^{2},s_{i}|s_{i}v_{i}s_{i}^{-1}=w_{i},i=1,2,...,k\rangle
\end{equation*}%
with
\begin{equation*}
v_{i}=%
\begin{pmatrix}
q_{i} \\
1%
\end{pmatrix}%
,w_{i}=%
\begin{pmatrix}
-p_{i} \\
1%
\end{pmatrix}%
,i=1,2,...,n-1.
\end{equation*}%
When the tubular group $G(\{p_i,q_i\}_{i=1}^{n-1})$ is virtually
compact special, there are at most two parallelism classes of $v_{i}$ and $%
w_{i},$ implying the set $\{-p_{i},q_{i}:i=1,2,...,n-1\}$ consists of at
most two elements. Since a compact special group is automatically $\mathrm{CAT(0)}$,
one of the two conditions of Theorem \ref{th7-1} holds. Conversely, when the
set $\{-p_{i},q_{i}:i=1,2,...,n-1\}$ consists of at most two elements, there
will be one or two parallelism classes of edge groups. When one of the two
conditions of Theorem \ref{th7-1} holds, the group $G(\{p_i,q_i\}_{i=1}^{n-1})$ is virtually special, hence $\mathrm{CAT(0)}$ by \Cref{cor1}. Since each element in a $\mathrm{CAT(0)}$ group is undistorted, $G(\{p_i,q_i\}_{i=1}^{n-1})$ contains no Baumslag--Solitar group $BS(m,n)$
with $n\neq \pm m.$ Therefore, $G(\{p_i,q_i\}_{i=1}^{n-1})$ is
virtually compact special by \Cref{lem:vir-cocom--cube} (2).
\end{proof}
The following example shows that the answer to Question \ref{prob2} is negative.
\begin{example}
\label{eg1} Let $\Psi$ be the automorphism of $F_3= \langle a,b,c\rangle$ given by $%
a\rightarrow a,b\rightarrow a^{m}ba^{m},c\rightarrow
a^{n}ca^{n}$ for some nonzero integers $m,n$. Let $G_1 = F_3\rtimes_\Psi \mathbb{Z}$ be the corresponding free-by-cyclic group. As shown in Example \ref{exam:Gpq-CAT(0)}, $G_1$ is a $\mathrm{CAT(0)}$ tubular group. By Corollary \ref{cor1}, it is virtually special. On the other hand, it has the following tubular representation:
$$\langle a,t, b,c\mid [a,t]=1, b^{-1} a^{-m}t b =a^mt, c^{-1} a^{-n}tc = a^nt \rangle.$$
When $|m|\neq |n|$, $a^{-m}t,a^mt, a^{-n}t, a^nt$ forms at least three parallelism
classes of edge groups inside the vertex group generated by $a$ and $t$, so
$G_1$ is not virtually cocompactly cubulated by \Cref{lem:vir-cocom--cube} (1).
When $|m| = |n|$, $a^{-m}t,a^mt, a^{-n}t, a^nt$ forms exactly two parallelism
classes of edge groups inside the vertex group generated by $a$ and $t$, so by \Cref{lem:vir-cocom--cube} (2), $G_1$ is a CAT(0) cube group. Moreover, it is virtually compact special by \cite[Corollary 5.9]{wi14} or \Cref{cor:vir-cpspe}.
\end{example}
The following free-by-cyclic group $F_3\rtimes_{\Phi} \mathbb{Z}$ is also from Lyman's paper \cite[Introduction]{Ly}.
\begin{example}
Let $\Phi$ be the automorphism of $F_3= \langle a,b,c\rangle$ given by $a \to a,b\to a^{-1}ba, c\to a^{-2} ca^2$. Let $G_2 = F_3\rtimes_{\Phi} \mathbb{Z}$ be the corresponding free-by-cyclic group. Up to the inner conjugation by $a$, we can replace $\Phi$ by the following automorphism:
$$a \to a,b\to b, c\to a^{-1} ca.$$
So $G_2$ has the following tubular presentation:
$$\langle a,t, b,c\mid [a,t]=1, btb^{-1}=t, c^{-1} atc = at \rangle.$$
Replace $a$ by $a'=at$, the tubular presentation of $G$ can be rewrite as:
$$\langle a',t, b,c\mid [a',t]=1, btb^{-1}=t, c^{-1} a'c = a' \rangle.$$
This is already a $\mathrm{RAAG}$. In particular, $G_2$ is the fundamental group of a compact special $\mathrm{CAT(0)}$ cube complex.
\end{example}
\section{Amalgamation of two free-by-cyclic groups}
In this section, we consider the amalgamation of two (virtually)
free-by-cyclic groups. We prove that the amalgamated product of two
free-by-cyclic groups along a cyclic
subgroup may not be virtually free-by-cyclic. Therefore, the answer to \Cref{prob3} is negative.
It is well known that the class of (virtually)
free-by-cyclic groups is closed under taking subgroups and free products \cite{bfm},
but not closed under taking HNN extensions along cyclic subgroups. It
was not clear whether the class of free-by-cyclic groups is closed under
amalgams along $\mathbb{Z}$ subgroups (see Hagen--Wise \cite[Remark
3.6]{hw}). Let us first provide a necessary condition on when the amalgamated product is again free-by-cyclic.
We need to introduce some terminology. Suppose that $G$ is decomposed as a semi-direct
product $H\rtimes \mathbb{\langle }a\mathbb{\rangle }$ for a free subgroup $%
H $ and an element $a.$ The free subgroup $H$ is called a free complement of
$a.$ An element $x\in G$ is primitive if $x$ is an element of a free basis
of some free complement $H.$ An element $g\in G$ is called a generalized retractor if there
exists a decomposition $G=H\rtimes \mathbb{\langle }a\mathbb{\rangle }$ for a
free subgroup $H$ and a \emph{nontrivial} element $a$ such that $g=ha^{i}$
for some $h\in H$ and a nonzero integer $i$. For example, when $G=\mathbb{Z}$%
, any non-zero element is a generalized retractor and $G$ itself is a free
complement (of the trivial element).
\begin{theorem}
\label{th8}Let $G_{1},G_{2}$ be two groups, $a\in G_{1},b\in
G_{2}$ be two elements of infinite order. Suppose that the amalgamated product $G=G_{1}\ast
_{a=b}G_{2}$ is free-by-cyclic. Then $G_1,G_2$ are free-by-cyclic, and one of the following holds
\begin{enumerate}
\item[1)] both $a,b$ are generalized retractors;
\item[2)] $a$ and $b$ lie in some free complements of $G_{1},G_{2}$ respectively.
Moreover, $a$ or $b$ is primitive.
\end{enumerate}
\end{theorem}
\begin{proof}
When $G_{1}\ast _{a=b}G_{2}$ is free-by-cyclic, there is an epimorphism
\begin{equation*}
\phi :G_{1}\ast _{a=b}G_{2}\rightarrow \mathbb{Z}
\end{equation*}%
with a free $\ker \phi .$ Since $\ker \phi |_{G_{i}}\leq\ker \phi ,i=1,2,$ we
know that $\ker \phi |_{G_{i}}$ is free as well. Therefore, both $G_1, G_2$ are free-by-cyclic (noting that a free group is free-by-cyclic).
If $\phi (a)\neq 0\in \mathbb{Z}$, we have $\phi |_{\langle a\rangle }$ is
injective. After passing to a finite index subgroup of $\mathbb{Z}$, we may
assume that $\phi |_{G_{1}}$ is surjective. Therefore,
\begin{equation*}
G_{1}=\ker \phi |_{G_{1}}\rtimes \langle x\rangle
\end{equation*}
for a preimage $x$ of the generator of $\mathrm{Im}(\phi |_{G_{1}})$. In this
case, we have $a=hx^{i}$ for the nonzero integer $i=\phi (a)$ and an element
$h\in \ker \phi |_{G_{1}}.$ Therefore, $a$ is a generalized retractor. The
same argument shows that $b$ is a generalized retractor.
If $\phi (a)=0\in \mathbb{Z}$, we have $\phi (b)=0\in \mathbb{Z}$. Since $%
\phi $ is surjective, $\phi |_{G_{1}},\phi |_{G_{2}}$ cannot be trivial
simultaneously. Without loss of generality, we assume that $\phi |_{G_{1}}$ is
surjective. Then
\begin{equation*}
G_{1}=\ker \phi |_{G_{1}}\rtimes \langle x\rangle
\end{equation*}
for a preimage $x$ of the generator of $\mathrm{Im}(\phi |_{G_{1}})\leq$ $\mathbb{%
Z}$. Consider the action of $\ker \phi $ on the Bass--Serre tree $T$ of the
amalgamated product $G_{1}\ast _{a=b}G_{2}.$ Note that each edge stabilizer
is $\mathbb{Z}$. The group $\ker \phi $ is a graph of free groups amalgamated over
cyclic subgroups. Let $v,w$ be the two vertices in the quotient $T/\ker \phi $, corresponding to $G_1,G_2$. The graph group $G([v,w])$ of the segment $[v,w]$ has vertex groups $\ker \phi |_{G_1}, \ker \phi |_{G_2}$, and edge group generated by $a=b$. Since $\ker \phi$ is free, the subgroup $G([v,w])$ is free.
It is well known that $G([v,w])$ is free only if
one of the subgroups $\langle a\rangle,\langle b\rangle $ is primitive ( cf.
\cite[Theorem 2.1]{bfr}).
\end{proof}
The following shows that condition 1) in Theorem \ref{th8} is also sufficient.
\begin{lemma}
\label{th8-cor}Let $G_{1},G_{2}$ be two free-by-cyclic groups, $a\in G_{1},b\in
G_{2}$ be two elements of infinite order. When both $a,b$ are generalized retractors, the amalgamated product $G=G_{1}\ast
_{a=b}G_{2}$ is free-by-cyclic.
\end{lemma}
\begin{proof}
When both $a,b$ are generalized retractors, there exist
decompositions $G_{1}=H_{1}\rtimes \mathbb{\langle }a_{1}\mathbb{\rangle }$,
$G_{2}=H_{2}\rtimes \mathbb{\langle }a_{2}\mathbb{\rangle }$ for free
subgroups $H_{1},H_{2}$ and $a=h_{1}a_{1}^{i_{1}},b=h_{2}a_{2}^{i_{2}}$ for
some elements $n_{1}\in H_{1},n_{2}\in H_{2}$ and some nonzero integers $%
i_{1},i_{2}.$ Let
\begin{equation*}
\phi :G_{1}\ast _{a=b}G_{2}\rightarrow \mathbb{Z}
\end{equation*}
be defined by
\begin{equation*}
\phi (H_{1})=0,\phi (a_{1})=i_{2},\phi (H_{2})=0,\phi (a_{2})=i_{1}.
\end{equation*}%
It is not hard to see that $\phi $ is a well-defined group homomorphism.
Considering the action of $\ker \phi $ on the Bass--Serre tree of $G_{1}\ast
_{a=b}G_{2},$ the edge stabilizer is trivial and the vertex stabilizers are
free (i.e. conjugates of $N_{1},N_{2}$). Therefore, the kernel $\ker \phi $
is a free product of free groups and thus free.
\end{proof}
\begin{remark}
When both $G_{1},G_{2}$ are free (then $G$ is called a cyclically pinched
one-relator group), Condition 1) in Theorem \ref{th8} is known to be sufficient by
Baumslag--Fine--Miller--Troeger \cite[Theorem 2]{bfm}. They also show that
the family of (virtually) free-by-cyclic groups is closed under taking
subgroups and free products.
\end{remark}
\begin{example}
Let $G_{1}$ be a free-by-cyclic group that is not abelian. Let $G_{2}=F_{2}\times \mathbb{Z}$ and $b$ be a generator of
the factor $\mathbb{Z}$. For any nontrivial element $a\in \lbrack
G_{1},G_{1}]$ (the commutator subgroup of $G_1$), the amalgamated product $G_{1}\ast
_{a=b}G_{2}$ is not free-by-cyclic.
\end{example}
\begin{proof}
Note that $b\in G_{2}$ commutes with any element in $F_{2}.$ This means that
$b$ does not lie in a free complement of any element. Suppose
that $G_{1}\ast _{a=b}G_{2}$ is free-by-cyclic. By Theorem \ref{th8}, the
element $a$ has to be a generalized retractor. However, since $a\in \lbrack
G_{1},G_{1}],$ any surjection $\phi :G_{1}\ast _{a=b}G_{2}\rightarrow
\mathbb{Z}$ must have $\phi (a)=0.$ This is a contradiction.
\end{proof}
\begin{example}
\label{eg2}Let $G_{1}=\langle a,b,s:ab=ba,sas^{-1}=b\rangle .$ The group $%
G_{1}\ast _{ab^{-1}=a}G_{1}$ is not free-by-cyclic.
\end{example}
\begin{proof}
By Theorem \ref{main} (2), the group $G_{1}$ is free-by-cyclic. It is even $F_n$-by-$\mathbb{Z}$ by \Cref{th10}. The element $%
a $ is a generalized retractor while $ab^{-1}$ is not, as any surjection $%
\phi :G_{1}\rightarrow \mathbb{Z}$ with a free kernel must have $\phi (ab^{-1})=0$. This implies any surjection from $G_{1}\ast _{ab^{-1}=a}G_{1} \to \mathbb{Z}$ must map $a$ in the right copy of $G_1$ to $0$. But since $a,b$ are conjugate, the element $b$ in the right copy of $G_1$ must also map to $0$. Therefore, both $a,b$ lie in the kernel and the kernel cannot be free.
\end{proof}
The following example shows that the condition (2) in \Cref{th8} is not sufficient.
\begin{example}
Let $G_1 =\langle a,s \mid sas^{-1} = a^{-1}\rangle $ and $G_2 = \langle b,t\mid tbt^{-1} =b\rangle $, and $G = G_1 \ast_{a=b} G_2$. By the Britton's lemma, the elements $s^2$ and $t$ generate a free subgroup of rank $2$ in $G$, and the intersection $\langle a\rangle \cap \langle s^2,t\rangle =\{1\}$. On the other hand, one checks that the element $a$ commutes with every element in $\langle s^2,t\rangle$, implying that elements $a$ and $[s^2,t]$ generate a free subgroup of rank $2$ in $G$. For any homomorphism $\phi$ from $G$ to $\mathbb{Z}$, we have $\phi(a)=1, \phi([s^2,t]) =1$. But this means that $\ker(\phi)$ contains a free abelian subgroup of rank $2$ and cannot be free.
\end{example}
For $F_n$-by-$\mathbb{Z}$ groups, we have a similar characterization.
\begin{corollary}
Let $G_{1},G_{2}$ be two finitely generated groups, $a\in G_{1},b\in
G_{2}$ be two infinite-order elements. If the amalgamated product $G=G_{1}\ast
_{a=b}G_{2}$ is $F_n$-by-$\mathbb{Z}$, then the groups $G_{1},G_{2}$ are $F_n$-by-$\mathbb{Z}$ and either both $a,b$ are generalized retractors, or $a,b$ lie in free complements of $G_{1},G_{2}$ respectively, with $a$ or $b$ is primitive.
\end{corollary}
\begin{proof}
It is enough to note that the first $L^2$-Betti number $b^{(2)}_1(G)=b^{(2)}_1(G_1)+b^{(2)}_1(G_2)$ by Chatterji--Hughes--Kropholler \cite[Theorem 1.5]{chk}. The finitely generated free-by-cyclic group $G$ is $F_n$-by-$\mathbb{Z}$ if and only if the first $L^2$-Betti number $b^{(2)}_1(G)=0$ by Lemma \ref{leml2}. If $G$ is $F_n$-by-$\mathbb{Z}$, we have $b^{(2)}_1(G_1)=b^{(2)}_1(G_2)=0$ and $G_1,G_2$ are $F_n$-by-$\mathbb{Z}$. The remaining proof can be finished using Theorem \ref{th8}.
\end{proof}
Now we consider the case of virtually free-by-cyclic groups. Let $G$ be a
virtually free-by-cyclic groups. An element $g\in G$ is called a virtual
generalized retractor if there is a finite index torsion-free subgroup $%
\Gamma \leq G$ and a decomposition $\Gamma =N\rtimes \mathbb{\langle }a\mathbb{%
\rangle }$ for a free subgroup $N$ and a nontrivial element $a,$ such that $%
g=na^{i}$ for some $n\in N$ and nonzero integer $i$. Similarly, the subgroup
$N$ is called a virtual free complement of the element $a$ if there is a
finite index torsion-free subgroup $\Gamma $ decomposed as $\Gamma =N\rtimes
\mathbb{\langle }a\mathbb{\rangle }$ for a free subgroup $N$ and an element $%
a.$
\begin{theorem}\label{thm:vir-free-nec}
Let $G_{1},G_{2}$ be two finitely generated groups, $a\in
G_{1},b\in G_{2}$ be two infinite-order elements. Suppose that the amalgamated
product $G_{1}\ast _{a=b}G_{2}$ is virtually free-by-cyclic. Then $G_1,G_2$ are virtually free-by-cyclic and there exists a
positive integer $m$ such that one of the following holds
\begin{enumerate}
\item[1)] both $a^{m},b^{m}$ are virtual generalized retractors;
\item[2)] $a^{m}$ and $b^{m}$ lie in virtual free complements of $%
G_{1},G_{2}$ respectively and $a^{m}$ or $b^{m}$ is primitive.
\end{enumerate}
\end{theorem}
\begin{proof}
Suppose that there is a finite index subgroup $\Gamma \leq G_{1}\ast _{a=b}G_{2}$
and an epimorphism $\phi :\Gamma \rightarrow \mathbb{Z}$ with a free $\ker
\phi .$ Note that $G_{1}\cap \Gamma ,G_{2}\cap \Gamma $ are free-by-cyclic
of finite indices in $G_{1},G_{2},$ respectively. Suppose that $\Gamma \cap
\langle a\rangle =\langle a^{m}\rangle $ for a positive integer $m.$
Moreover, we have
\begin{equation*}
\Gamma \geq(G_{1}\cap \Gamma )\ast _{a^{m}=b^{m}}(G_{1}\cap \Gamma )
\end{equation*}%
by Britton's lemma. Since $\Gamma $ is free-by-cyclic, the subgroup $%
(G_{1}\cap \Gamma )\ast _{a^{m}=b^{m}}(G_{1}\cap \Gamma )$ is free-by-cyclic
as well. Theorem \ref{th8} implies that either both $a^{m},b^{m}$ are
generalized retractors of $G_{1}\cap \Gamma ,G_{2}\cap \Gamma $ respectively,
or $a^{m}$ and $b^{m}$ both lie in free complements of $G_{1}\cap \Gamma
,G_{2}\cap \Gamma $ respectively and $a^{m}$ or $b^{n}$ is primitive.
\end{proof}
\begin{remark}
Some sufficient conditions for getting virtually free-by-cyclic groups are
obtained by Baumslag--Fine--Miller--Troeger \cite[Theorem 4]{bfm} and
Hagen--Wise \cite{hw}.
\end{remark}
\begin{definition}
Let $H\leq G$ be groups. $H$ is called a virtual retract of $G$, if there exists a finite index subgroup $K\leq G$ such that $H\leq K$ and $H$ is a retract of $K$.
\end{definition}
\begin{definition}
A group $G$ is called cyclically virtually retractible if for any infinite
cyclic subgroup $\langle a\rangle \leq G$ there is a finite-index subgroup $H\leq G$
and an epimorphism $f:H\rightarrow \mathbb{Z}$ such that $f|_{H\cap \langle
a\rangle }$ is an isomorphism.
\end{definition}
Clearly, a $\mathrm{(VRC)}$ group is cyclically virtually
retractable. Hagen and Wise \cite[Theorem 3.4]{hw} prove that a virtually
special group is cyclically virtually retractible.
\begin{lemma}
\label{lemresi}A torsion-free cyclically virtually retractible group $G$ is residually
finite.
\end{lemma}
\begin{proof}
For any $1\neq g\in G,$ there is a finite index $H\leq G$ and an epimorphism $%
f:H\rightarrow \mathbb{Z}$ such that $f(H\cap \langle g\rangle )=\mathbb{Z}.$
If $g\in H,$ the preimage $f^{-1}(2\mathbb{Z})$ is a finite index subgroup
of $H$ does not contain $g.$ If $g\notin H,$ we are done.
\end{proof}
\begin{lemma}
\label{lemvcr}A torsion-free group $G$ is cyclically virtually retractible if and only if $%
G$ is $\mathrm{(VRC)}$.
\end{lemma}
\begin{proof}
It is obvious that a $\mathrm{(VRC)}$ group is cyclically virtually retractible. Conversely, for a
torsion-free cyclically retractible group $G,$ it is residually finite by Lemma %
\ref{lemresi}. For any subgroup $\mathbb{Z},$ there is a finite index
subgroup $H\leq G$ and an epimorphism $f:H\rightarrow \mathbb{Z}$ such that $%
f|_{H\cap \mathbb{Z}}$ is an isomorphism. This implies that $H\cap \mathbb{Z}$
is a retract of $H.$ Minasyan \cite[Theorem 1.4]{min} shows that any
finite index extension $\mathbb{Z}\geq H\cap \mathbb{Z}$ is a virtually retract
of $G.$ Therefore, $G$ is $\mathrm{(VRC)}$.
\end{proof}
\begin{remark}
In general, a group has $\mathrm{(VRC)}$ if and only if it is residually finite and cyclically virtually retractible by \cite[Lemma 2.3 and Lemma 3.4]{min}.
\end{remark}
\begin{theorem}
\label{th9}
Let $G$ be a free-by-cyclic group and $a\in G$ is not a virtual retractor. Then the amalgamated product $%
G\ast _{ a =t }(F_{2}\times \mathbb{Z})$
(where $t$ is a generator of $\mathbb{Z})$ is not virtually free-by-cyclic.
\end{theorem}
\begin{proof}
Since $a$ is not a virtual retractor, we have for any finite index subgroup $H\leq G
$ and any epimorphism $\phi :H\rightarrow \mathbb{Z}$ the intersection $%
\langle a\rangle \cap H$ cannot be mapped $\mathbb{Z}$ by $\phi $ via isomorphism.
For any finite index subgroup $\Gamma \leq G\ast _{a =t }(F_{2}\times \mathbb{Z})$ and epimorphism $f:\Gamma \rightarrow
\mathbb{Z},$ we have $\Gamma \cap G$ is finite index in $G$. If $f|_{\Gamma
\cap G\cap \langle a\rangle }$ is injective, then $\mathrm{Im}f |_{\Gamma \cap
G\cap \langle a\rangle }$ is of finite index in the image $\mathbb{Z}$ and $%
f|_{G\cap \Gamma }^{-1}(\mathrm{Im}f |_{\Gamma \cap G\cap \langle a\rangle })\leq G$
would be of finite index, a contradiction to the choice of $a$. If $f_{\Gamma
\cap G\cap \langle a\rangle }$ is not injective, we have that $f(a^{k})=0$
for some nonzero integer $k.$ Then $f(t^{k})=0.$ But this means $\ker f$ contains the
subgroup $\ker f|_{F_{2}}\times \langle t^{k}\rangle ,$ which shows that $%
\ker f$ is not free. Therefore, any finite index subgroup $\Gamma $ is not
free-by-cyclic.
\end{proof}
\begin{example}
Let $G=F_{\infty }\rtimes \mathbb{Z}$ be the infinitely generated non-residually finite group
constructed by Baumslag \cite{baum}. By Lemma \ref{lemresi}, the group $G$ is not
cyclically virtually retractible. Theorem \ref{th9} implies that there is an
element $a\in G$ such that the amalgamated product $G\ast _{a =t }(F_{2}\times \mathbb{Z})$, is not virtually
free-by-cyclic.
\end{example}
The following is an observation about $\mathrm{(VRC)}$ groups.
\begin{lemma}
\label{vrc-euclid}
Let $G$ be a $\mathrm{(VRC)}$ group. For any infinite-order element $g\in G$, there is an Euclidean space $\mathbb{E}^k$ on which $G$ acts isometrically with $g$ acting as a nontrivial translation on some coordinate line.
\end{lemma}
\begin{proof}
Suppose that there is a finite index subgroup $\Gamma
\leq G$ and a retraction $r:\Gamma \rightarrow
\langle g\rangle .$ Let $\langle g\rangle \cong \mathbb{Z}$ act on
the real line $\mathbb{R}$ by translations. Let $\{g_{i}\}_{i=1}^{k}$ be a
system of representatives of left cosets $G/\Gamma $ (with $g_{1}=1$) and $%
\Pi _{G/\Gamma }\mathbb{R}$ the real vector space spanned by the basis $%
\{g_{i}\}_{i=1}^{k}.$ The group $G$ acts on the product $\mathbb{E}^k=\Pi _{G/\Gamma }%
\mathbb{R}$ by $g(\sum g_{i}x_{i})=\sum g_{j}h_{i}x_{i}$ assuming $%
gg_{i}=g_{j}h_{i}$ for some $h_{i}\in \Gamma .$ Since the action of each element $g$ is a composite of translations and permuting coordinates, the action is isometric.
\end{proof}
\begin{lemma}
\label{lem6.11}
Let $G = G(\{p_{i},q_{i}\}_{i=1}^{n-1})$ by the $F_n$-by-$\mathbb{Z}$ group
defined in the Introduction. Suppose that there are no vectors $%
v, w\in \mathbb{R}^{2}$ such that $w\neq 0$ and
\begin{equation*}
\Vert v-p_{i}w\Vert =\Vert v+q_{i}w\Vert ,i=1,2,...,n-1.
\end{equation*}%
Then $G(\{p_{i},q_{i}\}_{i=1}^{n-1})$ does not virtually retract onto the
cyclic subgroup $\langle a_{0}\rangle $. In particular, when $%
n=3, p_{1}=p_{2}=0,q_{1}\neq \pm q_{2}$ are nonzero integers, $a_{0}$ is not a virtual retractor of $G$.
\end{lemma}
\begin{proof}
Suppose that there is a finite index subgroup $\Gamma
\leq G({\{p_{i},q_{i}\}_{i=1}^{n-1}})$ and a retraction $r:\Gamma \rightarrow
\langle a_{0}\rangle .$ Let $\langle a_{0}\rangle \cong \mathbb{Z}$ act on
the real line $\mathbb{R}$ by translations. By Lemma \ref{vrc-euclid} (and its proof), there is an Euclidean space $\mathbb{E}^{k'}$ on which $G$ acts isometrically with $a_0$ acting as a nontrivial translation. The flat torus
theorem \cite[Theorem II.7.1]{BH} implies that there is an
Euclidean space $\mathbb{E}^{k}\subseteq \mathrm{Min}(\langle a_{0},t\rangle
)\subseteq \Pi _{G/\Gamma }\mathbb{R}$ ($k\leq 2$), on which $\langle
a_{0},t\rangle \cong \mathbb{Z}^{2}$ acts cocompactly by translations. Let $%
v,w$ be the translation vectors of $t,a$ respectively. Since $a_{0}$ acts by
non-trivial translation on $\mathbb{R}$, we see that $w\neq 0$ (and thus $%
k>0)$. Since $a_{0}^{-p_{i}}t=a_{i}a_{0}^{q_{i}}ta_{i}^{-1},$ we have $$\Vert
v-p_{i}w\Vert =\Vert v+q_{i}w\Vert $$ for each $i=1,2,...,n.$ (When $k=2,$
the vectors are from $\mathbb{R}^{2}.$ When $k=1,$ we view $\mathbb{R}$ as
a subspace of $\mathbb{R}^{2}$). This is a contradiction to the assumption. Since a circle (centered at the origin and of radius $|v|$) in the plane cannot intersect the line $v+\mathbb{R}w$ at three distinct points, $G$ cannot satisfy the condition when $
n=3, p_{1}=p_{2}=0,q_{1}\neq \pm q_{2}$.
\end{proof}
The following example answers Question \ref{prob3} in the negative.
\begin{example}\label{corlast}
Let $G=\langle a,b,c,t:tat^{-1}=a,tbt^{-1}=ba, tct^{-1} =ba^2\rangle $ be the Gersten group. Then $G\ast _{a=t}(F_{2}\times \mathbb{Z})$ (where $t$ is
the generator of $\mathbb{Z})$ is not virtually free-by-cyclic. In
particular, the family of virtually free-by-cyclic groups is not closed
under taking amalgamation along cyclic subgroups.
\end{example}
\begin{proof}
Lemma \ref{lem6.11} shows that $\langle a\rangle $ is not a virtual retractor of $G.$
Theorem \ref{th9} implies that $G\ast _{a=t}(F_{2}\times \mathbb{Z})$
is not virtually free-by-cyclic.
\end{proof}
We can also improve \Cref{corlast} to show that Condition (2) in \Cref{thm:vir-free-nec} is not sufficient.
\begin{example}
Let $G_1 =\langle a,b,c,t\mid tat^{-1}=a,tbt^{-1}=ba, tct^{-1} =ba^2\rangle $ be the Gersten group, $G_2 = \langle x,s\mid sxs^{-1} = x \rangle \cong \mathbb{Z}^2$. Then the tubular group $G = G_1 \ast_{a=x} G_2$ is not virtually free-by-cyclic.
\end{example}
\begin{proof}
Note first that $G$ has the following tubular presentation:
$$ \langle a,b,c,x, t,s \mid [a,t]=1,btb^{-1}=at, ctc^{-1} =a^2t, sas^{-1} =a\rangle .$$
Let $T$ be the corresponding Bass--Serre tree. Suppose that $H$ is a finite index subgroup of $G$. Since $H$ also acts cocompactly on $T$ with vertex stabilizers isomorphic to $\mathbb{Z}^2$ and edge stabilizers isomorphic to $\mathbb{Z}$, the group $H$ is tubular. Let $H\cap \langle a\rangle = \langle a^m\rangle$ for some $m\geq 1$. This means some edge group of the tubular group $H$ will be $\langle a^m\rangle$. Suppose that $H$ is a free-by-cyclic group. By \cite[Theorem 2.1]{bu}, there is a surjection $\phi: H \to \mathbb{Z}$ such that the restriction of $\phi$ to every edge group is injective. In particular $\phi\mid_{H\cap \langle a\rangle} = \phi\mid_{\langle a^m\rangle}$ is injective. But this means the restriction map $\phi: G_1\cap H \to \mathbb{Z}$ has $\phi(a^m) \neq 0$. Let $\bar{G}_1$ be the preimage of $\phi(a^m)$ in $G_1\cap H$. Then $\bar{G}_1$ is a finite index subgroup of $G_1\cap H$ hence $G_1$, and $a^m$ is a retractor of $\bar{G}_1$. By \cite[Theorem 1.4]{min}, we have that $a$ is also a virtual retractor of $G_1$. But this contradicts \Cref{lem6.11}.
\end{proof}
\section{$\mathrm{(VRC)}$ and cyclic subgroup separability}
The purpose of this section is to provide a counterexample to Question \ref{Ques-Min} of Minasyan \cite{min}. We first need a lemma on the separability of the cyclic subgroups of mapping tori. The proof of the following lemma is inspired by that of Hughes--Kudlinska \cite[Proposition 2.7]{hk}.
\begin{theorem}
\label{lastthm}
Let $K$ be a finitely generated group whose cyclic subgroups are separable
and $\phi :K\rightarrow K$ be an automorphism. Then any cyclic subgroup of $G=K\rtimes _{\phi }\mathbb{Z}$ is separable.
\end{theorem}
\begin{proof}
Let $t$ be a generator of $\mathbb{Z}$ and suppose that the cyclic subgroup is generated by $g=g_{0}t^{m}$. Now choose an arbitrary element $x=x_{0}t^{n}\in G$ such that $x\not\in \langle g\rangle$. It suffices now to prove that $x$ is separable from the cyclic subgroup $\langle g\rangle $.
If $m\neq 0$, $g$ is a generalized retractor since the natural projection $%
p:K\rtimes _{\phi }\mathbb{Z\rightarrow Z}$ has $p(g)\neq 0.$ Recall that a cyclic subgroup separable group is residually finite, so is $G$ (as a splitting cyclic extension of $K$). If $|m|$=1, $g$ is a retractor and $G$ is residually finite. So by \cite[Lemma 3.9]{hw99}), $\langle g\rangle $ is separable in $G$. If $|m|>1$, by \cite[Lemma 3.9]{hw99}), we have that $\langle g\rangle $ is separable in finite index subgroup $K\rtimes \langle t^m\rangle$ of $G$. This means $\langle g\rangle $ can be written as a family of finite index subgroups $\{H_i\}_{i\in I}$ of $K\rtimes \langle t^m\rangle$. Hence $\langle g\rangle = K\rtimes \langle t^m\rangle \cap (\cap_{i\in I} H_i)$. Since $K\rtimes \langle t^m\rangle$ is a finite index subgroup of $G$, $H_i$ are also finite index subgroups of $G$. This shows that $\langle g\rangle $ is separable in $G$.
If $m=0,$ and $n\neq 0.$ Take
\begin{equation*}
f:K\rtimes _{\phi }\mathbb{Z}~\overset{p}{\mathbb{\rightarrow }} ~\mathbb{%
Z\rightarrow Z}/2n
\end{equation*}%
to be the composition of $p$ with the natural epimorphism onto $\mathbb{Z}/2n$. We
have $f(g)=0$ but $f(x)\neq 0.$ The kernel $\ker f$ is a finite index
subgroup separating $g$ and $x.$
It remains to consider the case when $m=0$ and $n=0.$ Since $x\notin
\langle g\rangle $ and $K$ is cyclic subgroup separable, there is a finite index subgroup $H$ of $K$ such that $%
g\in H$ but $x\notin H.$ Since $K$ is finitely generated, the set $S$ of
subgroups in $K$ of index $[K,H]$ is finite. The automorphism $\phi $ acts
on this finite set $S$. So there is an integer $M$ such that $\phi ^{M}(H)=H.$ Now the
finite index subgroup $\langle H,t^{M}\rangle \leq G$ contains $g,$ but not $x.$
\end{proof}
Since finitely generated free groups are cyclic subgroup separable, we have the following corollary (see also \cite[Corollary 5.3]{BM06}, \cite[Proposition 2.7]{hk}).
\begin{corollary}\label{cor:css}
$F_n$-by-$\mathbb{Z}$ groups are cyclic subgroup separable.
\end{corollary}
The following example answers Question \ref{Ques-Min} in the negative, by combining Lemma \ref{lem6.11} and Corollary \ref{cor:css}.
\begin{example}
\label{eg7.2}
Let $G = \langle a,b,c,t \mid tat^{-1}=a,tbt^{-1}=ba, tct^{-1}=ca^2 \rangle$ be the Gersten group. It has a tubular presentation:
$$\langle a,b,c,t \mid [a,t]=1,b^{-1}tb=at, c^{-1}tc=a^2t \rangle.$$
It is cyclic subgroup separable by \Cref{cor:css} but does not have $\mathrm{(VRC)}$ by \Cref{lem6.11}.
\end{example}
\section{RFRS vs. virtually RFRS, comments and questions}
In this section, we first discuss the subtle difference between RFRF and virtually RFRS, then we provide some open questions. Note that the original version \cite[Problem 5.3]{page} of Question \ref{prob1.1} is stated in terms of RFRS instead of virtually RFRS. We first show that a $\mathrm{CAT(0)}$ free-by-cyclic may not be RFRS.
Recall that in \cite{Ag08} a finitely generated group $G$ is called RFRS if it has a sequence of finite index normal subgroups $G=G_{0}\geq G_{1}\geq \cdots$
such that: 1) $\cap G_{i}=1$; 2) $G_{i+1}\geq \ker (G_{i}\rightarrow
H_{1}(G_{i})\bigotimes \mathbb{Q})$ for each $i.$ A virtually special group
is virtually RFRS \cite[Theorem 2.2]{Ag08}.
Fisher recently proved that RFRS groups are locally indicable \cite[Proposition 2.7]{fish}. We improve this to the following
(one can get locally indicable by taking $H$ to be finitely generated all $g_{i}=1$).
\begin{lemma}
\label{8.1}
Let $G$ be a $\mathrm{RFRS}$ group. For any (possibly infinitely generated) non-trivial subgroup $H$ and
any finitely many elements $g_{1},...,g_{k}\in G,$ there are positive
integers $n_{1},n_{2},...,n_{k}$ such that there is a surjection $f:\langle
H,g_{1}^{n_{1}},...,g_{k}^{n_{k}}\rangle \rightarrow \mathbb{Z}$ satisfying $%
f(H)\neq 0.$
\end{lemma}
\begin{proof}
Let $n$ be the integer such that $H\leq G_{n}$ but $H$ is not a subgroup of $%
G_{n+1}.$ Since $G_{n}\leq G$ is finite index, there are integers $n_{i}$
satisfying $g_{i}^{n_{i}}\in G_{n}.$ The epimorphism $f:G_{n}\rightarrow
H_{1}(G_{n};\mathbb{Z})\rightarrow H_{1}(G_{n};\mathbb{Q})$ has $f(H)$
nontrivial, since $\ker f\leq G_{n+1}.$ Now a subgroup of the finitely generated
abelian group $$H_{1}(G_{n};\mathbb{Z})=H_{1}(G_{n};\mathbb{Z}%
)_{tor}\oplus H_{1}(G_{n};\mathbb{Z})_{free}$$ is finitely generated.
The image of $H$ in the free-part $H_{1}(G_{n};\mathbb{Z})_{free}$ is
nontrivial and thus has a surjection onto $\mathbb{Z}.$ The restriction $f|_{\langle
H,g_{1}^{n_{1}},...,g_{k}^{n_{k}}\rangle}$ satisfies the required property.
\end{proof}
\begin{example}
($\mathrm{CAT(0)}$ free-by-cyclic may not be $\mathrm{RFRS}$) Let $G=F_{2}\rtimes \mathbb{Z}=\langle
a,b,t:tat^{-1}=ab,tbt^{-1}=ab^{2}\rangle .$ Choose $H=F_{2},t.$ For any
positive integer $n,$ the subgroup $\langle H,t^{n}\rangle $ does not have a
surjection onto $\mathbb{Z}$ with nontrivial image of $H.$ Therefore, the group $G$ is not $\mathrm{RFRS}$ (but is indeed virtually $\mathrm{RFRS}$).
\end{example}
In the same vein, Agol asked in his ICM survey paper \cite[Question 11]{Ag14} whether the braid groups $B_n$ are RFRS. Note that any non-trivial subgroup of RFRS groups has a surjection onto $\mathbb{Z}$ by Lemma \ref{8.1}. On the other hand, the commutator subgroups of the braid groups with at least five strings are finitely presented and perfect \cite[Theorem 2.1]{GL69}, so $B_n$ cannot be RFRS if $n\geq 5$ by Lemma \ref{8.1} or \cite[Proposition 2.7]{fish}. It is still open though whether braid groups are virtually RFRS.
\begin{question}
\label{8.3}
Let $G$ be a $\mathrm{CAT(0)}$ free-by-cyclic group. Is every finitely generated subgroup of $G$ again $\mathrm{CAT(0)}$?
\end{question}
Since a subgroup of a virtually special (resp. virtually RFRS) group is virtually special (resp. virtually RFRS), a negative answer to Question \ref{8.3} would answer the other direction of Question \ref{prob1.1}.
At this point, it is hard to attack problems related to (infinitely generated free)-by-cyclic groups. It is natural to ask whether such problems can sometimes be reduced to the finite-rank case (we learned the first part of Question \ref{ques:embed} from Richard Wade in May 2023, see also \cite[Conjecture 1.2]{CW24}).
\begin{question}\label{ques:embed}
Does a finitely generated free-by-cyclic group embed into some $F_n$-by-$\mathbb{Z}$ group? Does a hyperbolic finitely generated free-by-cyclic group embed into some hyperbolic $F_n$-by-$\mathbb{Z}$ group?
\end{question}
It is well-known that we have the following chain of inclusions:
$\mathcal{F}_{vsp}:=$\{virtually special group\} $\subseteq $ $\mathcal{F}%
_{\mathrm{vRFRS}}:=$\{virtually RFRS group\} $\subseteq \mathcal{F}_{\mathrm{VRC}}:=$ \{VRC\} $%
\subseteq \mathcal{F}_{css}:=$ \{Cyclic subgroup separable group\}. Gersten's example
shows that a free-by-cyclic group may not be $\mathrm{(VRC)}$ (see Lemma \ref{lem6.11}), in particular it can not be virtually $\mathrm{RFRS}$. Minasyan \cite[Question 11.5]{min} asked whether the property $\mathrm{(VRC)}$ is stable under taking amalgamated free products over (virtually) cyclic subgroups. It is natural to ask the
following question.
\begin{question}
Are the inclusions $\mathcal{F}_{vsp}\subseteq $ $\mathcal{F}_{\mathrm{vRFRS}}\subseteq
\mathcal{F}_{\mathrm{VRC}}$ proper for the class of finitely generated free-by-cyclic groups? Is
the family $\mathcal{F}_{vsp}$ (resp. $\mathcal{F}_{\mathrm{vRFRS}},\mathcal{F}_{\mathrm{VRC}}$
) closed under taking amalgamated product over cyclic subgroup
among finitely generated free-by-cyclic groups?
\end{question}
%\begin{thebibliography}{9}
\begin{thebibliography}{BB}
\bibitem{page} AimPL: Rigidity properties of free-by-cyclic groups, available at \url{http://aimpl.org/freebycyclic}.
\bibitem{Ag} I. Agol, The virtual Haken conjecture, Doc. Math. 18 (2013),
1045--1087.
\bibitem{Ag08} I. Agol, Criteria for virtual fibering, J. Topol. 1 (2008), no. 2, 269–284.
\bibitem{Ag14} I. Agol, Virtual properties of 3-manifolds. Proceedings of the International Congress of Mathematicians—Seoul 2014. Vol. 1, 141–170, Kyung Moon Sa, Seoul, 2014.
\bibitem{baum}G. Baumslag. A non-cyclic, locally free, free-by-cyclic group all of whose finite factor
groups are cyclic. Bull. Austral. Math. Soc. 6:313–314, 1972.
\bibitem{bfm} G. Baumslag, B. Fine, C. F. Miller, III, and D. Troeger,
Virtual properties of cyclically pinched one-relator groups. Internat. J.
Algebra Comput. 19 (2009), 213-227.
\bibitem{bfr} G. Baumslag, B. Fine, and G. Rosenberger, One-relator groups:
an overview, Groups St Andrews 2017 in Birmingham, London Math. Soc. Lecture
Note Ser., vol. 455, Cambridge Univ. Press, Cambridge, 2019, pp. 119--157.
%\bibitem{BB} N. Brady and M. R. Bridson, %There is only one gap in the
%isoperimetric spectrum, Geom. Funct. Anal. %10(2000) 1053--1070.
\bibitem{BS} N. Brady and I. Soroko. Dehn functions of subgroups of
right-angled Artin groups. Geom. Dedicata, 200:197--239, 2019.
\bibitem{Bri} M. R. Bridson. Problems concerning hyperbolic and CAT(0)
groups. Available at
\url{https://docs.google.com/file/d/0B-tup63120-GVVZqNFlTcEJmMmc/edit}.
\bibitem{BH} M. R. Bridson and A. Haefliger, Metric spaces of nonpositive curvature, Grundlehren der Math. Wiss. 319, Springer-Verlag, Berlin, 1999.
%\bibitem{bks} R. G. Burns, A. Karrass, and %D. Solitar. A note on groups with
%separable finitely generated subgroups. %Bull. Austral. Math. Soc.,
%36(1):153--160, 1987.
\bibitem{BM06} J. Burillo, A. Martino,
Quasi-potency and cyclic subgroup separability.
J. Algebra 298 (2006), no. 1, 188–207.
\bibitem{bu} J. O. Button, Tubular Free by Cyclic Groups Act Freely on
CAT(0) Cube Complexes, Canad. Math. Bull. Vol. 60 (1), 2017 pp. 54--62.
\bibitem{but} J. O. Button, Tubular groups, 1-relator groups and nonpositive curvature, Internat. J.
Algebra Comput. 29 (2019), no. 8, 1367–1381.
\bibitem{bk} J. O. Button and R. P. Kropholler. Nonhyperbolic free-by-cyclic
and one-relator groups. New York J. Math., 22: 755--774, 2016.
%\bibitem{bm} J. Burillo, A. Martino, Quasi-potency and cyclic subgroup separability, J. Algebra 298 (2006), no. 1, 188–207.
\bibitem{chk} I. Chatterji, S. Hughes, P. Kropholler. The first $\ell^2$-Betti number and groups acting on trees, Proc. Edinb. Math. Soc. (2) 64
(2021), no. 4, 916-923.
\bibitem{CW24} H. K. Chong, D. T. Wise, Embedding Partial HNN Extensions In Ascending HNN Extensions, preprint 2024, arXiv:2408.00453.
\bibitem{ds} C. Drutu and M. Sapir. Non-linear residually finite groups,
Journal of Algebra 284 (2005), no. 1, 174--178.
\bibitem{gkl} G. Gardam, D. Kielak, and A. D. Logan. The surface group
conjectures for groups with two generators. Math. Res. Lett.,
30(1):109--123, 2023.
\bibitem{ge} S. M. Gersten. The automorphism group of a free group is not a
CAT(0) group. Proc. Amer. Math. Soc., 121(4):999--1002, 1994.
\bibitem{GL69} E. A. Gorin and V. Ja. Lin. Algebraic equations with continuous
coefficients and some problems of the algebraic theory of braids.
Mat. Sbornik., 78 (120)(4):569–596, 1969.
\bibitem{fish} S. Fisher, Improved algebraic fibrings. Compos. Math. 160, no. 9 (2021) 2203 - 2227.
\bibitem{Fi24} S. Fisher, On the cohomological dimension of kernels of maps to
$\mathbb{Z}$, preprint 2024 arXiv:2403.18758.
\bibitem{hw} M. Hagen and D. T. Wise, Special groups with an elementary
hierarchy are virtually free-by-$\mathbb{Z}$, Groups Geom. Dyn. 4 (2010),
no. 3, 597--603.
\bibitem{ha1} M. Hagen and D. T. Wise. Cubulating hyperbolic
free-by-cyclic groups: the general case. Geom. Funct. Anal., 25(1):134--179,
2015.
\bibitem{ha2} M. Hagen and D. T. Wise. Cubulating hyperbolic
free-by-cyclic groups: the irreducible case. Duke Math. J.,
165(9):1753--1813, 2016.
\bibitem{hw99} T. Hsu, D.T. Wise, On linear and residual properties of graph products. Michigan Math. J. 46 (1999), no. 2, 251-259.
\bibitem{hk} S. Hughes and M. Kudlinska, On profinite rigidity amongst free-by-cyclic groups I: the generic case, arXiv:2303.16834.
\bibitem{KL24} D. Kielak, M. Linton, Virtually free-by-cyclic groups. Geom. Funct. Anal. 34 (2024), no. 5, 1580–1608.
\bibitem{liu} Y. Liu, Virtual cubulation of nonpositively curved graph
manifolds, J. Topol. 6 (2013), no. 4, 793--822.
\bibitem{Lin} P. A. Linnell, Division rings and group von Neumann algebras,
Forum Math. 5 (1993), no. 6, 561--576.
\bibitem{lr} D. D. Long, A.W. Reid, Subgroup separability and virtual
retractions of groups. Topology 47 (2008), no. 3, 137--159.
\bibitem{luck} W. L\"{u}ck, L2-Betti numbers of mapping tori and groups.
Topology 33, 203-214 (1994).
\bibitem{Ly} R. A. Lyman, Some New CAT(0) Free-by-Cyclic Groups, Michigan
Mathematical Journal, 73 (2023).
\bibitem{min} A. Minasyan, Virtual retraction properties in groups, Int.
Math. Res. Not. IMRN 2021 (2021), 13434--13477.
\bibitem{mp} Z. Munro and H. Petyt, Coarse obstructions to cocompact
cubulation, arXiv:2407.09275.
\bibitem{sa} P. Samuelson, On CAT(0) structures for free-by-cyclic groups.
Topology Appl., 153(15):2823--2833, 2006.
\bibitem{wi12} D. T. Wise, From Riches to Raags: 3-Manifolds, Right-Angled
Artin Groups, and Cubical Geometry, CBMS Regional Conference Series in
Mathematics No. 117, American Mathematical Society, Providence, RI, 2012.
\bibitem{wi14} D. T. Wise, Cubular tubular groups. Transactions of the
American Mathematical Society, 366(10):5503--5521, 2014.
\bibitem{Wi20} D.T. Wise, An invitation to coherent groups. What's next?—the mathematical legacy of William P. Thurston, 326–414. Ann. of Math. Stud., 205 Princeton University Press, Princeton, NJ, 2020.
\bibitem{Wood2} D. Woodhouse, Classifying finite dimensional cubulations of
tubular groups. Michigan Math. J. 65 (2016), no. 3, 511--532.
\bibitem{Wood} D. J. Woodhouse. Classifying virtually special tubular
groups. Groups Geom. Dyn., 12(2):679--702, 2018.
\end{thebibliography}
\end{document}