Path Extendable Tournaments
Extracted source text ยท arXiv 2504.21653
%This is a draft version of the joint paper on path extendability tournaments
%%HB: this is a revised version; I have indicated all changes, suggestions, questions, remarks etc. by %%HB, so they are easy to find
%%Zanbo make some changes based on Hajo's version 2021/9/18
%%Zanbo make some changes according to feedback from Hajo 2021/9/23
%%Zanbo make revision after rejection from JCTB 2022/2/15
\documentclass{article}
\usepackage{latexsym}
\usepackage{mathrsfs}
\usepackage{indentfirst}
\usepackage{amsmath,amsthm, amssymb}
\usepackage[dvips]{graphicx}
\usepackage{graphicx}
\usepackage{epsfig}
\usepackage{verbatim}
\usepackage{color}
%For multiple lines in caption of a figure
\usepackage{caption}
%For different item styles
\usepackage{enumitem}
%The default font is computer modern, below line change it to times
%\usepackage{mathptmx}
%\usepackage{multirow}
%For compiling eps format figures in Texstudio
\usepackage{epstopdf}
%this package is used for adding strike-through when revising
%can be commented after revision
%It should be noted that this package will change emphasizing for italics to underline.
%We need \normalem command or the [normalem] option to disable the change
%\usepackage[normalem]{ulem}
%\normalem
%\begin{comment}
%These packages are for recalling a theorem
\usepackage{thmtools}
%\usepackage{thm-restate}
%\usepackage{hyperref}
%\usepackage{cleveref}
%\declaretheorem[name=Theorem,numberwithin=section]{thm}
%\end{comment}
%Definition of the end of proof
%\def\qed{\hfill \rule{4pt}{7pt}}
%
\def\pf{\noindent {\it Proof.} }
\topmargin -15mm
%\oddsidemargin -10mm
\oddsidemargin 0mm
%
%\evensidemargin -10mm
\evensidemargin 0mm
\textwidth 160mm
\textheight 220mm
%
%\textwidth 114mm
%
%\textheight 181mm
%space between lines
\linespread{1.1}
%no header and footer, specially removing the page number
%\pagestyle{empty}
%default is single spaced, below line is not needed
%\renewcommand{\baselinestretch}{1}
%
%\newtheorem*{theorem*}{Theorem}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{problem}{Problem}
\newtheorem{remark}{Remark}
\newtheorem{conjecture}{Conjecture}
\newtheorem{observation}{Observation}
\newtheorem{proposition}{Proposition}
\newtheorem{example}{Example}
\setlength{\parindent}{1em}
\begin{document}
%%HB: I suggest a short title since I think it is more attractive for the reader/referees; it is similar to the short title of Hendry's paper and of Thomassen's paper
\title{Path Extendable Tournaments
%\title{Path Extendability of Tournaments Close to Doubly Regular and Random Tournaments
%\thanks{}
}
\author{
Zan-Bo Zhang$^{1,2}$\thanks{Email address: eltonzhang2001@gmail.com, zanbozhang@gdufe.edu.cn. Partially supported by
Guangdong Basic and Applied Basic Research Foundation (2020B1515310009), and NSFC (U1811461).}
Weihua He$^{3}$\thanks{Email address: hwh12@gdut.edu.cn.
Supported by Guangdong Basic and Applied Basic Research Foundation (2021A1515012047),
and Science and Technology Program of Guangzhou (202002030401).}
Hajo Broersma$^4$\thanks{Email address: h.j.broersma@utwente.nl.}
Xiaoyan Zhang$^{5}$\thanks{Corresponding Author. Email address: royxyzhang@gmail.com, zhangxiaoyan@njnu.edu.cn.
Supported by NSFC (11871280) and QingLan Project.}
\\ \small $^1$ School of Statistics and Mathematics, Guangdong University of
\\ \small Finance and Economics, Guangzhou 510320, China
\\ \small $^2$ Institute of Artificial Intelligence and Deep Learning, Guangdong University
\\ \small of Finance and Economics, Guangzhou 510320, China
\\ \small $^3$ School of Mathematics and Statistics, Guangdong University
\\ \small of Technology, Guangzhou 510006, China
\\ \small $^4$ Faculty of Electrical Engineering, Mathematics and Computer Science,
\\ \small University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands
\\ \small $^5$Institute of Mathematics \& School of Mathematical Science,
\\ \small Nanjing Normal University, Nanjing, 210023, China
}
\date{}
\maketitle
%removing the first page number
%\thispagestyle{empty}
%abstract part
\begin{abstract}
%%ZZ: third version of the abstract
\noindent
A digraph $D$ is called \emph{path extendable} if for every nonhamiltonian (directed) path
$P$ in $D$, there exists another path $P^\prime$ with the same initial and
terminal vertices as $P$, and $V(P^\prime) = V (P)\cup \{w\}$ for a vertex $w \in V(D)\setminus V(P)$.
Hence, path extendability implies paths of continuous lengths between every vertex pair.
In earlier works of C. Thomassen and K. Zhang,
it was shown that the condition of small $i(T)$ or positive $\pi_2(T)$
implies paths of continuous lengths between every vertex pair in a tournament $T$,
where $i(T)$ is the irregularity of $T$ and
$\pi_2(T)$ denotes for the minimum number of paths of length $2$ from $u$ to $v$
among all vertex pairs $\{u,v\}$.
Motivated by these results, we study
sufficient conditions in terms of $i(T)$ and $\pi_2(T)$
that guarantee a tournament $T$ is path extendable.
We prove that (1) a tournament $T$ is path extendable
if $i(T)< 2\pi_2(T)-(|T|+8)/6$, and (2) a tournament $T$ is path extendable
if $\pi_2(T) > (7|T|-10)/36$.
As an application, we deduce that almost all random tournaments are path extendable.
%%ZZ: We choose to use Hajo's new abstract, and keep the old one in a comment environment
\begin{comment}
%%HB: I start by letting the reader know that the definitions/properties are known/have been studied since a long time
%%HB: I changed the next sentence as follows
%A digraph $D$ is called strongly panconnected (resp. completely strong path-connected)
Adopting the terminology introduced by Thomassen in 1980 and Zhang in 1982, we call a digraph $D$ strongly panconnected and completely strong path-connected, respectively,
%
if for every pair of distinct vertices $u,v\in V(D)$
%%HB: I put the restriction on k at the end, since then we can use respectively better
%and every integer $3 \le k \le |D|-1$ (resp. $2 \le k \le |D|-1$),
there exists a (directed)
%%HB: I added (directed) once to make clear what we mean and avoid confusion
path of length $k$ from $u$ to $v$ and one from $v$ to $u$, %
%%HB: I put the restriction on k at the end, since then we can use respectively better
for every integer $3 \le k \le |D|-1$, respectively $2 \le k \le |D|-1$.
%%HB: I also refer to Hendry's definition of path extendability in undirected graphs, by adding the following
Adapting the term introduced by Hendry for undirected graphs in 1990, a digraph
$D$ is called path extendable if for every nonhamiltonian
%%HB: again, just to be clear
(directed)
path
$P$ in $D$, there exists another path $P_0$ with the same initial vertex and
terminal vertex of $P$, and $V(P_0) = V (P)\cup \{w\}$ for %certain
%%HB: I think it is better to use `a' and the LaTeX `setminus'
a
%
vertex $w \in V(D)\setminus V(P)$.
%%HB: I think it is common to write the abstract as one paragraph
Let $i(T)$ denote the irregularity of a tournament $T$, %%HB: I added a `let'
let $p_2(u,v)$ denote
the number of
%%HB: I deleted $2$-paths and the () and added an `s' after `path' and separated u\in V
paths of length $2$ in $T$ from $u \in V(T)$ to $v \in V(T)$, % where $u,v \in V(T)$,
and
%%HB: another `let'
let
$\pi_2(T)=\min\{p_2(u,v) \mid u,v\in %%HB: corrected `D' to 'T'
V(T), u\ne v\}$.
Thomassen [J. Combin. Theory, Ser. B 28 (1980) 142-163]
proved that a tournament $T$ with small $i(T)$
%%HB: changed the above statement to involve T
is strongly panconnected,
while Zhang [J. Combin. Theory, Ser. B 33 (1982) 166-177]
proved that $\pi_2(T)\ge 1$ implies
%%HB: we can remove this: the tournament
$T$
%%HB: to be
is completely strong path-connected.
%%HB: I think it is better not to refer to the meta-conjecture since Bondy does not like it very much
%Extending Bondy's meta-conjecture that almost any nontrivial hamiltonian condition on a graph
%also implies the graph to be pancyclic,
%%HB: I made small changes to the formulation
%we seek
Motivated by the above results, we studied
%for
sufficient conditions
%
in terms of $i(T)$ and $\pi_2(T)$
%for
that guarantee a tournament $T$ is path extendable. %on
%involving $i(T)$ and $\pi_2(T)$.
We prove that
%%HB: again minor changes in the formulation
a tournament $T$ is path extendable
if $i(T)< 2\pi_2(T)-(|T|+8)/6$.
% then $T$ is path extendable.
As a corollary we
%%HB: again here
get that a tournament $T$ is path extendable
%prove that
if $\pi_2(T) > (7|T|-10)/36$.
% implies that $T$ is path extendable.
%%HB: I suggest to omit some details from the below statement
We also
%prove
show that
%in random tournaments $\pi_2(T)$ can be arbitrarily close
%to $|T|/4$, and thus
almost all random tournaments are path extendable.
%%HB: the below part was already omitted by you before, and I left it this way.
%A key idea in this work is the introduction of the concept of surplus
%with respect to $\pi_2$.
%The surplus of a vertex pair $\{u,v\}$ is defined as
%$s(u,v)=p_2(u,v)+p_2(v,u)-2\pi_2(T)$, and the surplus of a vertex subset $W$
%is defined as the sum of the surplus of all vertex pairs $\{u,v\}\subseteq W$.
%The concept of surplus is helpful in estimating precisely
%the number of $2$-paths between vertex pairs in a subgraph of $T$.
\end{comment}
\begin{comment}
%%HB: In fact, I prefer to replace the above abstract by the following one.
\noindent
Our work is motivated by the following two classic results on tournaments. In
[J.\ Combin.\ Theory, Ser.\ B 28 (1980) 142--163], Thomassen
proved that a tournament $T$ with small irregularity $i(T)$
is strongly panconnected, i.e., for every pair of distinct vertices $u,v\in V(T)$
there exists a (directed)
path of length $k$ from $u$ to $v$ and one from $v$ to $u$,
for every integer $k$ with $3 \le k \le |V(T)|-1$.
%
In [J.\ Combin.\ Theory, Ser.\ B 33 (1982) 166--177], Zhang
proved that $\pi_2(T)\ge 1$ implies
that the tournament
$T$
is completely strong path-connected, i.e., such paths exist for every integer $k$ with $2 \le k \le |V(T)|-1$.
%
Here $\pi_2(T)=\min\{p_2(u,v) \mid u,v\in V(T), u\ne v\}$, with $p_2(u,v)$ denoting the number of paths of length $2$ in $T$ from $u \in V(T)$ to $v \in V(T)$.
%
Adapting a term introduced by Hendry for undirected graphs, a digraph
$D$ is called path extendable if for every nonhamiltonian (directed) path
$P$ in $D$, there exists another path $P'$ with the same initial and
terminal vertices as $P$, and with $V(P') = V (P)\cup \{w\}$ for a vertex $w \in V(D)\setminus V(P)$.
%
Motivated by the above results, we studied sufficient conditions
in terms of $i(T)$ and $\pi_2(T)$
that guarantee a tournament $T$ is path extendable.
We prove that
a tournament $T$ is path extendable
if $i(T)< 2\pi_2(T)-(|V(T)|+8)/6$.
As a corollary we
get that a tournament $T$ is path extendable
if $\pi_2(T) > (7|V(T)|-10)/36$.
As a direct application of the last result, we use it to
deduce that
almost all random tournaments are path extendable.
\end{comment}
%key words
$\newline$\noindent\textbf{Key words}: path, path extendability, tournament,
random tournament
\end{abstract}
\section{Introduction, terminology and notation}
%%HB: I made several minor changes to the following paragraphs. For convenience I kept the old ones, with %-signs added, followed by the suggested ones; if you do not like my suggestions, you can easily revert to the old versions.
%
%Tournaments have rich cycle and path structures.
%Every tournament has a hamiltonian path. Further, with the additional condition that being strong,
%a tournament can be hamiltonian, pancyclic, vertex-pancyclic and cycle extendable
%(for cycle extendability a well-characterized class of exceptional graphs has to be excluded, see \cite{Moon1969, Hendry1989}).
%
%However, many so called path-connected properties, such as
%hamiltonian-connectedness, panconnectedness and path extendability,
%in which paths of specific lengths are involved,
%are not generally hold in tournaments.
%There have been a lot of research work to find conditions
%for these properties all along.
\noindent
It is commonly known that tournaments have a rich path and cycle structure. As an example,
every tournament has a hamiltonian path, and every strong tournament is hamiltonian, vertex-pancyclic,
and even cycle extendable
(for cycle extendability a well-characterized class of exceptional graphs has to be excluded, see \cite{Hendry1989,Moon1969}).
%Some recent works have delved deeper into the power of paths (\cite{DDF2021})
%and the number of Hamiltonian cycles (\cite{FK2005, Cuckler2007}) in a tournament.
%Zanbo "involving paths" are repeated here so I change the expression
%However, many stronger properties involving connecting paths, such as
%hamiltonian-connectedness, panconnectedness and path extendability,
%in which paths of specific lengths are involved,
However, many stronger properties involving connecting paths of specific
lengths, such as hamiltonian-connectedness, %arc-treacability (see \cite{BJR2006}),
panconnectedness and path extendability,
do not generally hold in tournaments.
Since the late 1960s, many researchers have devoted themselves to finding sufficient conditions
for these properties.
%%HB: again small changes in the below paragraph, which I kept intact for comparison and reverting back
%Let $D$ be a digraph and $u,v\in V(D)$ be two distinct vertices.
%We call a path of length $k$,
%i.e. containing $k$ arcs, a $k$-path.
%A $(u,v)$-$k$-path means a path from $u$ to $v$ of length $k$.
%Throughout the paper, we use $n$ to denote the number of vertices in a digraph.
%A digraph $D$ is said to be \emph{weakly hamiltonian-connected} (resp. \emph{weakly panconnected})
%if for any distinct $x,y \in V(D)$ there exists a hamiltonian path from $x$ to $y$ or from $y$ to $x$
%(resp. there exists an $(x,y)$-$k$-path or a $(y,x)$-$k$-path for every $3\le k \le n-1$),
%and $D$ is called \emph{strongly hamiltonian-connected} (resp. \emph{strongly panconnected})
%if for any distinct $x,y \in V(D)$ there exists a hamiltonian path from $x$ to $y$ and from $y$ to $x$
%(resp. there exists an $(x,y)$-$k$-path and a $(y,x)$-$k$-path for every $3\le k \le n-1$).
Throughout the paper, we use $n$ or $|D|$ to denote the number of vertices of a digraph $D$.
Let $D$ be a digraph on $n\ge 2$ vertices, and let $u,v\in V(D)$ be two distinct vertices.
We use the term $k$-path to indicate a directed path of length $k$,
i.e., containing $k$ arcs,
and $(u,v)$-$k$-path as shorthand for a $k$-path from $u$ to $v$.
Adopting the terminology of Thomassen \cite{Thomassen1980},
a digraph $D$ is said to be \emph{weakly hamiltonian-connected}
%(resp. \emph{weakly panconnected})
if for any distinct $x,y \in V(D)$ there exists a hamiltonian path from $x$ to $y$ or from $y$ to $x$,
%(resp. there exists an $(x,y)$-$k$-path or a $(y,x)$-$k$-path for every $3\le k \le n-1$),
and $D$ is called \emph{strongly hamiltonian-connected}
%(resp. \emph{strongly panconnected})
if for any distinct $x,y \in V(D)$ there exists a hamiltonian path from $x$ to $y$ and from $y$ to $x$.
%(resp. there exists an $(x,y)$-$k$-path and a $(y,x)$-$k$-path for every $3\le k \le n-1$).
Similarly, in \cite{Thomassen1980}, $D$ is called \emph{weakly panconnected} if for any distinct $x,y \in V(D)$ there exists an $(x,y)$-$k$-path or a $(y,x)$-$k$-path for every integer $k$ with $3\le k \le n-1$, and
\emph{strongly panconnected}
if for any distinct $x,y \in V(D)$ there exists an $(x,y)$-$k$-path and a $(y,x)$-$k$-path for every integer $k$ with $3\le k \le n-1$.
In \cite{Thomassen1980}, Thomassen characterized all weakly hamiltonian-connected tournaments,
and showed that for a tournament being weakly hamiltonian-connected is equivalent to being weakly
panconnected. He also proved that being $4$-strong is a sufficient condition for a
tournament to be strongly hamiltonian-connected.
Here a digraph $D$ is called \emph{$k$-strong} if $D-A$ is strongly connected for any set $A\subseteq V(D)$ with at most $k-1$ vertices.
Then, strengthening a result of Alspach et al., %(Theorem \ref{thm:PathOfAllLengthRegular} below),
he showed that a tournament $T$ that is close to regular, or to be precise, with small \emph{irregularity}
$i(T)\triangleq \max \{|d^+(u)-d^-(u)|\mid u\in V(T)\}$,
is strongly panconnected. %(Theorem \ref{thm:irregularitystronglypanconnected} below).
Note that in a tournament the condition of irregularity can be turned into a degree condition as
%%Zanbo change the fraction to vertical
\begin{equation} \label{eqn:i(T)andDegree}
\min\{\delta^+(T), \delta^-(T)\} =\frac{n-i(T)-1}{2}.
\end{equation}
%\begin{theorem} (Alspach et al. \cite{AlsRei1974}) \label{thm:PathOfAllLengthRegular}
%Let $uv$ be an arc of a regular tournament $T$ on $n\geq 7$ vertices.
%Then there exist paths from $u$ to $v$ of length $k$ in $T$, for every integer $k$ with $3\leq k \leq n-1$.
%\end{theorem}
\begin{theorem} \label{thm:irregularitystronglypanconnected} (Thomassen \cite{Thomassen1980})
If $T$ is a tournament on $n \geq 5k + 21$ vertices with $i(T) \leq k$, then $T$ is strongly panconnected.
\end{theorem}
%%HB: I made a few very minor changes above, and continue to do so below, where I kept the old version.
%In the definition of strongly panconnectedness, only paths of length at least three are required.
%A strengthened concept to include $2$-paths is \emph{completely strong path-connectedness},
%which means that for every distinct vertex pair $\{u,v\}$ in a digraph $D$
%and every integer $2\le k \le n-1$,
%there exists a $(u,v)$-$k$-path.
%Let $T$ be a tournament. We denote by $p_2(u,v)$ the number of $(u,v)$-$2$-paths in $T$,
%and let $\pi_2(T)=\min \{p_2(u,v): u,v \in V(T)\}$ be the global infimum of $p_2(u,v)$,
%where $u$, $v$ are distinct vertices in $T$.
%K. Zhang gave a condition for completely strong path-connectedness in terms of $\pi_2(T)$.
%\begin{theorem} \label{thm:2pathallpaths} (Zhang \cite{Zhang1982})
%A tournament $T$ with $n$ vertices is completely strong path-connected if and only if $\pi_2(T)\ge 1$,
%unless $T$ belongs to one well-characterized class of exceptional graphs.
%\end{theorem}
%Furthermore, one may pose the structural requirement of path extendability,
%which can imply the existence of paths of continuous lengths (but not necessary start from $2$ or $3$).
%A nonhamiltonian path $P$ in a digraph $D$ is called \emph{extendable},
%if there exists another path $P^\prime$ with
%the same initial vertex and terminal vertex of $P$,
%and $V(P^\prime)=V(P)\cup \{w\}$ for certain vertex $w\in V(D)\backslash V(P)$.
%A digraph $D$ is called \emph{$\{k+\}$-path extendable},
%if every nonhamiltonian path of length at least $k$ is extendable in $D$.
%A $\{1+\}$-path extendable digraph is called \emph{path extendable}.
%The current authors have studied path extendability in tournaments, and obtained Theorem \ref{thm:path3+Extendable},
%in which the exceptional graphs are defined as below.
Note that in the definition of strongly panconnectedness the existence of paths of length 2 is not required.
A more restrictive concept which includes the existence of $2$-paths was studied by Zhang in \cite{Zhang1982}. There he defines a digraph $D$ to be \emph{completely strong path-connected}
if there exists a $(u,v)$-$k$-path and a $(v,u)$-$k$-path in $D$ for every distinct $u\in V(D)$ and $v\in V(D)$
and every integer $k$ with $2\le k \le n-1$.
Let $T$ be a tournament. We denote %
%by $p_2(u,v)$
%%HB: this reads better
the number of distinct $(u,v)$-$2$-paths in $T$ by $p_2(u,v)$,
and we let $\pi_2(T)=\min \{p_2(u,v)\mid u,v \in V(T)\}$ be the global infimum of $p_2(u,v)$,
where $u$ and $v$ are distinct vertices of $T$.
In \cite{Zhang1982}, Zhang presented the following characterization of completely strong path-connected tournaments in terms of $\pi_2(T)$.
\begin{theorem} \label{thm:2pathallpaths} (Zhang \cite{Zhang1982})
A tournament $T$ on $n$ vertices is completely strong path-connected if and only if $\pi_2(T)\ge 1$,
unless $T$ belongs to one well-characterized class of exceptional graphs.
\end{theorem}
Motivated by the aforementioned structural results on tournaments and the results due to Hendry on path extendability in undirected graphs \cite{Hendry1990b}, we focus on the analogous concept of path extendability in tournaments.
For this purpose, we say that a
nonhamiltonian path $P$ in a digraph $D$ is \emph{extendable},
if there exists another path $P^\prime$ in $D$ with
the same initial vertex and terminal vertex as $P$,
and with $V(P^\prime)=V(P)\cup \{w\}$ for some vertex $w\in V(D)\setminus V(P)$
(note that Faudree and Gy\'{a}rf\'{a}s \cite{FG1996} study another kind of path extendability in tournaments,
which extends a path from one of its endpoints).
Moreover, by including the following specific length restriction,
a digraph $D$ is called \emph{$\{k+\}$-path extendable},
if every nonhamiltonian path of length at least $k$ is extendable in $D$.
And naturally, we use the shorter term \emph{path extendability} for $\{1+\}$-path extendability.
The current authors have studied path extendability in tournaments, and obtained
% Theorem~\ref{thm:path3+Extendable},
the following result,
in which the structure of the exceptional graphs will be described in Section \ref{sec:2PathsPathExt}
and be used to prove Theorem \ref{thm:2PathsPathExt}.
%are defined as below according to the illustrations in Figures~\ref{figure:TwoHybridVerticesAll}, \ref{figure:TwoHybridSets} and \ref{figure:Seven-Exception}, respectively.
%%HB: I think it is better to state the theorem here, before explaining the exceptional graphs
%%HB: I also think we should refer to the published paper only and not to the thesis, in order to avoid confusion
\begin{theorem} \label{thm:path3+Extendable} (Zhang et al. %\cite{Zhang2017}, Theorem 3.11 \&
\cite{ZZBL2017})
A regular tournament $T$ with at least $7$ vertices is $\{2+\}$-path extendable,
unless $T\in \mathcal{T}_1$, $T\in \mathcal{T}_2$ or {$T=T_0$}.
\end{theorem}
%\begin{theorem} \label{thm:path3+Extendable} (Zhang et al. \cite{Zhang2017}, Theorem 3.11 \& \cite{ZZBL2017})
%A regular tournament $T$ with at least $7$ vertices is $\{2+\}$-path extendable,
%unless $T\in \mathcal{T}_1$, $T\in \mathcal{T}_2$ or {$T=T_0$}.
%\end{theorem}
%%HB: I suggest to omit the following but mention Bondy's conjecture only in one sentence, just as a motivation
%In 1973, Bondy (\cite{Bondy1973}) raised the influential meta-conjecture that
%almost any nontrivial condition on a graph which implies that
%the graph is hamiltonian also implies that the graph is pancyclic.
%The first example that witnessed the conjecture,
%again due to Bondy (\cite{Bondy1971}),
%is that the Ore condition for hamiltonicity that
%$d(u)+d(v)\ge |G|$ for any pair of nonadjacent vertices $u$ and $v$
%in an (undirected) graph $G$ also implies that $G$ is pancyclic,
%unless $G$ is the balanced bipartite graph $K_{\frac{n}{2},\frac{n}{2}}$.
%Interestingly, Hendry showed that Ore's condition can further imply cycle extendability of $G$,
%and even path extendability of $G$ if we increase the lower bound by one
%(\cite{Hendry1990a,Hendry1990b}, where in both results there are
%some exceptional graphs that can be characterized).
%%HB: minor revisions of the below paragraph that I maintained below with %%
%%As we mentioned above, in tournaments strongness implies hamiltonicity, pancyclicity
%%and cycle extendability (with an exceptional class), but not path extendability.
Reflecting on Bondy's idea in his meta-conjecture on hamiltonian undirected graphs \cite{Bondy1973}, and considering results of Bondy on pancyclicity \cite{Bondy1971} and Hendry on cycle extendability \cite{Hendry1990a} and path extendability \cite{Hendry1990b}, one is tempted to expect that imposing similar conditions
as in
Theorem~\ref{thm:irregularitystronglypanconnected}
and Theorem~\ref{thm:2pathallpaths} on a tournament would in fact imply that it is path extendable.
With this in mind, the following problems arise quite naturally.
%As we have stated, in tournaments strongness implies hamiltonicity, pancyclicity
%and cycle extendability (with exception) but not path extendability.
%However, extending Bondy's idea,
%the current authors have expected conditions similar to the ones in
%Theorem \ref{thm:irregularitystronglypanconnected}
%and \ref{thm:2pathallpaths} to imply path extendability,
%and raised the following problems.
%%HB: minor changes in the formulations of the two problems, just for better grammar.
\begin{problem} \label{prb:pi_2} (Zhang et al. \cite{ZZBL2017}, Problem 2)
Does $\pi_2(T)\ge 1$ imply a tournament $T$ is path extendable?
\end{problem}
\begin{problem} \label{prb:i} (Zhang et al. \cite{ZZBL2017}, Problem 3)
Let $T$ be a tournament on $n$ vertices with $i(T) \le k$.
Does there exist a function $f(k)$, such that $n \ge f(k)$
implies $T$ is path extendable?
\end{problem}
%%HB: I made minor adjustments and added `by answering the two questions' below
In the next section, the two above problems will be solved in the negative.
%%HB: I also reformulated the essence of the next sentence
However, we derive several sufficient
conditions in terms of $i(T)$ and $\pi_2(T)$ for path extendability, including a condition that
involves both $i(T)$ and $\pi_2(T)$.
%%HB: I added the following
We also discuss the sharpness of our new results, and we deduce that almost all tournaments are path extendable.
%%Zanbo add some notation definition here
We explain some notation here.
Let $D$ be a digraph, and $U_1, U_2, U_3 \subseteq V(D)$ be three pairwise disjoint subsets of $D$.
The subdigraph induced by $U_1$ is denoted by $\langle U_1 \rangle$.
The number of arcs in $D$ from $U_1$ to $U_2$ is denoted by $d^+(U_1, U_2)$.
$U_1 \rightarrow U_2$ means that $U_1$ dominates $U_2$,
i.e. every vertex in $U_1$ sends an arc to every vertex in $U_2$.
For convenience, when $U_1=\{u\}$ we write $u$ instead of $\{u\}$, and
we use $U_1\rightarrow U_2\rightarrow U_3$ as an abbreviation for $U_1\rightarrow U_2$ and $U_2\rightarrow U_3$.
Let $P$ be a path and $u,v$ be two vertices on $P$. Then $P[u,v]$ stands for the subpath of $P$ from $u$ to $v$.
The other terms and notation in this paper are in accordance with the monograph \cite{BanGut2008}.
%%HB: title with only capital at the start, for consistency
\section{Main results} \label{sec:MainResults}
%%HB: I reformulated the first paragraph, and kept the original one with %-signs
Before we turn to the solutions of Problem~\ref{prb:pi_2} and Problem~\ref{prb:i},
let us first define what we mean by a \emph{doubly regular tournament}.
A regular tournament $T$ is called doubly regular if for every vertex pair
%%HB: I am assuming we use American English thoughout, so neigbor, characterize, etc.
$\{u,v\}$ in $T$, the number of common out-neighbors of
$u$ and $v$ is the same constant, which we denote by $\lambda$.
%%HB: I added a sentence:
Let $T$ be a doubly regular tournament on $n$ vertices, with $\lambda$ as above.
Then simple calculations show that $\lambda = (n-3)/4$, a result that can be found in an early paper due to Reid and Brown \cite{ReiBro1972}.
%%HB: I think there is a mistake here:
%Therefore, $n \equiv 4 \pmod 3$. I also added an expression for $n$ in terms of \lambda
Therefore, $n=4\lambda+3$, hence $n \equiv 3 \pmod 4$.
Furthermore, one also easily shows that for any $uv \in A(T)$,
the number of common in-neighbors of $u$ and $v$, $p_2(u,v)$,
and $p_2(v,u)$ are fixed to $\lambda$, $\lambda$, and $\lambda + 1$, respectively.
Hence, in a doubly regular tournament $T$ we have $\pi_2(T)=\lambda=(n-3)/4$.
This shows that the upper bound on $\pi_2(T)$ in the below theorem, which will be proved in Section~\ref{sec:Surplus},
is sharp.
%$(n-3)/4$ is exactly the upper bound of $\pi_2(T)$ in all tournaments.
\begin{theorem} \label{thm:pi_2Supremum}
In a tournament $T$, $\pi_2(T)\le (n-3)/4$.
\end{theorem}
%Before we go for the solutions of Problem \ref{prb:pi_2} and Problem \ref{prb:i},
%let's firstly define \emph{doubly regular tournaments}.
%A regular tournament $T$ is called doubly regular if for every vertex pair
%$\{u,v\}$ in $T$, the number of common out-neighbours of
%$u$ and $v$ is a constant, which we denote by $\lambda$.
%Simple calculations show that $\lambda = (n-3)/4$. Therefore, $n \equiv 4 \pmod 3$.
%Furthermore, for any $uv \in A(T)$,
%the number of common in-neighbours of $u$ and $v$, $p_2(u,v)$,
%and $p_2(v,u)$ are fixed to $\lambda$, $\lambda$ and $\lambda + 1$, respectively.
%Thus in a doubly regular tournament we have $\pi_2(T)=\lambda=(n-3)/4$.
%As shown in the below theorem, which will be proved in Section \ref{sec:Surplus},
%$(n-3)/4$ is exactly the upper bound of $\pi_2(T)$ in all tournaments.
%\begin{theorem} \label{thm:pi_2Supremum}
%In a tournament $T$, $\pi_2(T)\le (n-3)/4$.
%\end{theorem}
%%HB: I changed the wording below and added some space
%Now we are ready to state the solutions.
With the help of the above discussion on doubly regular tournaments, we can now provide negative answers to the questions in the aforementioned two problems.
\medskip
\noindent
%%HB: small changes in the below discussion
\textbf{Solution to Problem \ref{prb:pi_2}.}
Let $\mathcal{T}_3$ be the class of regular tournaments %$T$
%(thus $i(T)=0$)
whose vertex set can be partitioned as
$V=V_0\cup V_1 \cup V_2$, %where
with $|V_i|=4t+3$ for an integer $t\ge 1$. %
Let
the induced subgraph $\langle V_i \rangle$ %is
be a doubly regular tournament %,
%%HB: I added relevant information, without giving details on the orders, since this is what the reader needs to know; I think they will consult the paper of Reid and Brown if they want to know more
(these exist for infinitely many choices of $t$ by a result in \cite{ReiBro1972}),
and %
let
$V_i \rightarrow V_{i+1}$ for $i\in \{0,1,2\}$, with the subscripts %modulus $3$.
taken modulo $3$.
%%ZZ The figure is removed and we should remove the reference to it as well.
%See Figure~\ref{figure:PathExtCounterExample} %
% for a sketch of these tournaments.
%
%
%%HB: I have put the figure here; perhaps we should remove the figure, as it does not add much to the explanation. There is also a wrong arrow in the drawing!
%%ZZ: I have removed the figure.
% \begin{figure}[h]
%\centering
%\includegraphics[width=0.24\linewidth]{Figures/PathExtCounterExample.eps}
%\caption{A tournament in $\mathcal{T}_3$, where every $ V_i $ induces \\ a doubly regular tournament of the same order}
%\label{figure:PathExtCounterExample}
%\end{figure}
%
Consider any vertex pair $\{u,v\}$ with $uv\in A(T)$ in %
a tournament $T\in \mathcal{T}_3$
%%HB: I moved `with $uv\in A(T)$' and added the below order in terms of $t$.
on $n=12t+9$ vertices.
If $u,v \in V_i$ for any $i\in \{0,1,2\}$,
then all $2$-paths between $u$ and $v$ must be in $\langle V_i \rangle$.
By %
the
above discussion on doubly regular tournaments,
there are $t=(n-9)/12$ $(u,v)$-$2$-paths and $t+1=(n+3)/12$ $(v,u)$-$2$-paths in $\langle V_i \rangle$, %
hence
%and
in $T$ as well.
If $u$ and $v$ are in different $V_i$, then %there can be
there are clearly more than $t$
$(u,v)$-$2$-paths and $(v,u)$-$2$-paths in %$T_3$.
$T$. Hence $\pi_2(T)=(n-9)/12$.
But %
obviously
%
any hamiltonian path of $\langle V_i \rangle$ for $i\in \{0,1,2\}$ is not extendable.
So we have a negative
%solution to
answer to the question in Problem~\ref{prb:pi_2}.
%\begin{figure}[h]
%\centering
%\includegraphics[width=0.24\linewidth]{Figures/PathExtCounterExample.eps}
%\caption{A tournament in $\mathcal{T}_3$, where every $ V_i $ induced\\ a doubly regular tournament of the same order}
%\label{figure:PathExtCounterExample}
%\end{figure}
%%HB: I added space and \noindent and reformulated some of the following
\medskip
\noindent
\textbf{Solution to Problem \ref{prb:i}.}
%%HB: I think the reader/referee will not appreciate the details on the order
%By a result from \cite{ReiBro1972},
%there exist doubly regular tournaments of order $n_0=2^q k_0k_1\ldots k_{s-1}-1$,
%where each $k_i$ is of the form $p^r+1 \equiv 0 \pmod 4$.
By the main result in \cite{ReiBro1972}, there exist doubly regular tournaments of arbitrarily large (but specific) order. Using the above approach and notation,
we can construct a regular tournament $T\in \mathcal{T}_3$ of sufficiently large order $n$, such that each $\langle V_i \rangle$ is isomorphic to
a doubly regular tournament.
Since $i(T)=0$, for whatever nonnegative integer $k$ we choose, we always have $i(T) \le k$.
Moreover, for any function $f$, we can always construct $T$ as above with sufficiently large $n \ge f(k)$.
Recalling that $T$ is not path extendable, this way we obtain concrete evidence for a negative answer to the question in Problem~\ref{prb:i}.
$\newline$
%%HB: reformulated the following and kept the old paragraph below
To continue our exposition with some positive results on path extendability, we start by stating the following theorem which appeared in
%For the positive side, we seek for conditions on $i(T)$ and $\pi_2(T)$ for path extendability.
%The first one is cited from
Chapter 5 of \cite{Zhang2017}.%, the PhD thesis of the first author.
%It states that a doubly regular tournament $T$,
%that is, a tournaments $T$ with $i(T)=0$ and $\pi_2(T)=(n-3)/4$, is path extendable.
%For the positive side, we seek for conditions on $i(T)$ and $\pi_2(T)$ for path extendability.
%The first one is cited from Chapter 5 of \cite{Zhang2017}, the doctoral thesis of the first author.
%It states that a doubly regular tournament $T$,
%that is, a tournaments $T$ with $i(T)=0$ and $\pi_2(T)=(n-3)/4$, is path extendable.
\begin{theorem} \label{thm:DRTPathExt} (Zhang \cite{Zhang2017}, Theorem 5.17)
All doubly regular tournaments on at least $7$ vertices are path extendable.
\end{theorem}
%%HB: I added information on $T$ here instead of the above theorem, and gave some background question for motivation
Recall that a doubly regular tournament $T$ on $n$ vertices satisfies $i(T)=0$ and $\pi_2(T)=(n-3)/4$. A natural question is whether we can relax one or both of the conditions on $i(T)$ and $\pi_2(T)$ in order to obtain a more general result on path extendability of tournaments.
In Section~\ref{sec:2PathsPathExt},
we will study the exceptional graphs of Theorem~\ref{thm:path3+Extendable} carefully,
and prove the following result,
which improves the above %one
%%HB: I expanded the below sentence to reflect the improvement
result by reducing the lower bound on $\pi_2(T)$ to $(n-9)/12$ and considering regular instead of doubly regular tournaments.
\begin{theorem} \label{thm:2PathsPathExt}
Let $T$ be a regular tournament on $n\geq 9$ vertices. If $\pi_2(T)> (n-9)/12$, then $T$ is path extendable.
%%HB: I used `on' above and moved the next remark to below the statement, with some addition.
%Furthermore, the bound on $\pi_2(T)$ is sharp as shown by the graph class $\mathcal{T}_3$.
\end{theorem}
%Furthermore,
Note that the lower bound on $\pi_2(T)$ %
in the above result is sharp, as shown by the graph class $\mathcal{T}_3$, and recall that $i(T)=0$ for a regular tournament.
%%HB: I connected the sentences by using `Also note', and corrected some typos, and reformulated slightly
Also note that
the negative solution to Problem~\ref{prb:i} eliminates the possibility to %
establish a sufficient condition for path extendability involving $i(T)$ only.
%use $i(T)$ alone as
%a sufficient condition for path extendability.
%%HB: I reformulated the below paragraph and kept the old one with %-signs
However, by increasing the lower bound on $\pi_2(T)$ in Problem~\ref{prb:pi_2} and Theorem~\ref{thm:2PathsPathExt},
we can obtain the following sufficient condition for path extendability in general tournaments that is based on $\pi_2(T)$ only.
%%HB: I think it is better for the flow of the narrative to put the theorem here. I kept it also at the old place with %%.
\begin{theorem} \label{thm:pi_2(T)only}
Let $T$ be a tournament on $n$ vertices. If $\pi_2(T) > (7n-10)/36 $, then $T$ is path extendable.
\end{theorem}
%which presented in Theorem \ref{thm:pi_2(T)only} below.
Theorem~\ref{thm:pi_2(T)only} can be viewed as an analogue of Theorem~\ref{thm:2pathallpaths} for path extendability, although it does not characterize path extendable tournaments but only provides a sufficient condition in terms of $\pi_2(T)$.
%but also helps to prove path extendability of random tournaments subsequently. I
%Zanbo: From reviewer comments of JCTB, the result on random tournament is "not surprising", so I remove this statement.
%More importantly, Theorem~\ref{thm:pi_2(T)only} will turn out to be useful in proving path extendability of random tournaments.
%However, improving the lower bound of $\pi_2(T)$ in Problem~\ref{prb:pi_2},
%we can get a sufficient condition for path extendability,
%which presented in Theorem \ref{thm:pi_2(T)only} below.
%It can not only be viewed as an analog of Theorem \ref{thm:2pathallpaths},
%but also helps to prove path extendability of random tournaments subsequently.
%%HB: In view of the below suggested move of Theorem \ref{thm:pi_2(T)i(T)}, I suggest to replace the below text as well
%To prove Theorem \ref{thm:pi_2(T)only}, we firstly prove Theorem \ref{thm:pi_2(T)i(T)},
%a condition on both $i(T)$ and $\pi_2(T)$.
%%HB: I replaced it by:
In fact, Theorem~\ref{thm:pi_2(T)only} follows in a rather straightforward way from our next result, involving %Theorem~\ref{thm:pi_2(T)i(T)},
conditions on both $i(T)$ and $\pi_2(T)$.
%%HB: I also suggest to bring Theorem \ref{thm:pi_2(T)i(T)} here, for the convenience of the reader/referee; otherwise, they have to go up and down in the paper, and might get annoyed
\begin{theorem} \label{thm:pi_2(T)i(T)}
%Let $T$ be a tournament with $n\ge 9$ vertices, irregularity $i(T)$
%and at least $\pi_2(T)>(n-9)/12$ $(u,v)$-$2$-paths for every vertex pair $\{u,v\}$.
%If $i(T)< 2\pi_2(T)-(n+8)/6 $, then $T$ is path extendable.
%%HB: Am I wrong or can we streamline the above as follows? Am I missing something?
Let $T$ be a tournament on $n\ge 9$ vertices
%, irregularity $i(T)$
%and at least
with $\pi_2(T)>(n-9)/12$. %$(u,v)$-$2$-paths for every vertex pair $\{u,v\}$.
If $i(T)< 2\pi_2(T)-(n+8)/6 $, then $T$ is path extendable.
\end{theorem}
The idea of posing restrictions on both $i(T)$ and $\pi_2(T)$
%%HB: I suggest the following since the idea is from a trade-off, not just from the mentioned theorem
%comes from Theorem \ref{thm:2PathsPathExt},
reflects a trade-off between allowing $T$ to deviate more from regularity by
%where we may
increasing the upper bound on $i(T)$, and compensating it by % so that $T$ deviate from being regular,
%meanwhile we
increasing the lower bound on $\pi_2(T)$, thereby imposing a stronger condition on $\pi_2(T)$ than in (doubly) regular tournaments.
%maintain path extendability.
Theorem \ref{thm:pi_2(T)i(T)} %brings in
not only brings in a new %type
combination of conditions, but
%%HB: slight change
these conditions also turn out to be convenient in the structural analysis.
%also convenience in structural analysis.
%%HB: By the change of ordering, we can omit the next sentence.
%By a relation between $\pi_2(T)$ and $i(T)$ as shown below in Lemma \ref{lem:i(T)AndPi(T)},
%we obtain Theorem \ref{thm:pi_2(T)only} from Theorem \ref{thm:pi_2(T)i(T)} straightforward.
%%HB: I added the following
We will prove Theorem~\ref{thm:pi_2(T)i(T)} and the below lemma in Section~\ref{sec:pi_2(T)i(T)},
and show how Theorem~\ref{thm:pi_2(T)only} can be obtained from them.
\begin{lemma} \label{lem:i(T)AndPi(T)}
%%HB: For consistency I use `on' and omitted `we have' in the statement
In a tournament $T$ on $n$ vertices, $i(T)\leq n-4\pi_2(T)-3$.
\end{lemma}
%%HB: I moved the figure and added [h] in the hope it will end up where I like
%\begin{figure}[h]
%\centering
%\includegraphics[width=0.28\linewidth]{Figures/RegularNoTwoPath.eps}
%\caption{A regular tournament without $(u_0,u_1)$-$2$-paths}
%\label{figure:RegularNoTwoPath}
%\end{figure}
%Now we state the new sufficient conditions for path extendability in tournaments,
%which will be proved in Section \ref{sec:pi_2(T)i(T)}.
%\begin{theorem} \label{thm:pi_2(T)i(T)}
%Let $T$ be a tournament with $n\ge 9$ vertices, irregularity $i(T)$
%and at least $\pi_2(T)>(n-9)/12$ $(u,v)$-$2$-paths for every vertex pair $\{u,v\}$.
%If $i(T)< 2\pi_2(T)-(n+8)/6 $, then $T$ is path extendable.
%\end{theorem}
%\begin{theorem} \label{thm:pi_2(T)only}
%Let $T$ be a tournament. If $\pi_2(T) > (7n-10)/36 $, then $T$ is path extendable.
%\end{theorem}
%%HB: I suggest to skip the remarks below because they somehow repeat what has been said before, and the section is already quite long. I don't think the referee would like to see the repetition of similar remarks.
%\textbf{Remarks on Theorem \ref{thm:pi_2(T)i(T)} and Theorem \ref{thm:pi_2(T)only}.}
%As can be seen in Theorem \ref{thm:PathOfAllLengthRegular} and Theorem \ref{thm:irregularitystronglypanconnected},
%to generalize a result on regular tournaments, one way is to use irregularity as a measurement of
%how close a graph is to being regular.
%Given that doubly regular tournaments are path extendable,
%an analogue way to find more path extendable tournaments is to look at
%the tournaments that are close to doubly regular.
%The idea of Theorem \ref{thm:pi_2(T)i(T)} and Theorem \ref{thm:pi_2(T)only}
%is to use $\pi_2(T)$, or $i(T)$ and $\pi_2(T)$ together as a measurement of how close
%a tournament is to being doubly regular.
%We wish that such kinds of conditions would be generally applicable on more problems.
%%HB: Slight changes in the next paragraph
\medskip
%%Zanbo: JCTB reviewer said that the result on random tournament is not surprsing. So
%%let's state here that it is presented as a straightforward application of our former result.
Finally we show that almost all tournaments are path extendable.
%a result that may not come as a surprise, but the
Its proof we present here is a
straightforward application of Theorem \ref{thm:pi_2(T)only}.
%Finally, we
%look at
%turn to random tournaments.
In an early paper, Reid and Beineke proved that almost all tournaments are strong, and
%
they noted
%remarked
the following stronger result.
\begin{theorem} (Reid and Beineke \cite{ReiBei1978}) \label{thm:T3-Cycle}
In almost all tournaments, every pair of vertices
%lie
lies on a $3$-cycle.
\end{theorem}
The conclusion of Theorem \ref{thm:T3-Cycle} is equivalent to
%
the statement
that
for every arc $uv$, $p_2(v,u)\ge 1$.
%
%
%The
%authors have
%%HB: should it be `first author' according to the cite you give below?
%%HB: we can avoid this by changing the formulation, which I did
%first author of this paper
%proved a stronger result in the more general class of random oriented graphs.
%
We prove that a stronger result holds for the more general class of random oriented graphs.
Fixing $0 \le p \le 1/2$, consider the model $\overrightarrow{G}_{n,p}$
%%HB: I think we should add `random' and perhaps the following formulation is more accurate
%which contains
of
all %
random
oriented graphs $D$ on $n$ vertices,
%
in which
%and
for any vertex pair $\{u,v\}$
%in $D$,
with probability $p$ we have the arc $uv$,
with probability $p$ we have the arc $vu$,
and with probability $1-2p$ we have no arc between $u$ and $v$.
When $p=1/2$, $\overrightarrow{G}_{n,p}$ becomes the model for random tournaments on $n$ vertices.
%
%%HB: do we need to cite this explicitly or does it suffice to give the more informative explanation that follows later? I assume the latter and changed it here accordingly. As always, it is your choice if you want to revert back.
%\begin{theorem} (Zhang \cite{Zhang2017}, Theorem 5.9)
%In almost all oriented graphs there are linearly many $2$-paths between every pair of vertices.
%\end{theorem}
%
%The above theorem actually says that there exists certain constant $c=c(p)$
%such that almost all $D\in \overrightarrow{G}_{n,p}$
%satisfy that $\pi_2(D)\ge c n$. But an explicit $c(p)$ is not given.
%
In Theorem~5.9 of \cite{Zhang2017}, Zhang confirmed that there exists a constant $c=c(p)$
such that almost all $D\in \overrightarrow{G}_{n,p}$
satisfy that $\pi_2(D)\ge c n$. However, an explicit $c(p)$ is not given there.
%
In Section~\ref{sec:RandomTPi_2}, we use a Chernoff bound to prove the following theorem,
which
%%HB: small change and corrected the , to a .
%states
shows that $\pi_2(D)$
%can be
is arbitrarily close to its expectation.
\begin{theorem} \label{thm:RandomTPi_2}
For any $0< \varepsilon < p^2$, almost all oriented
%graph
graphs $D\in \overrightarrow{G}(n,p)$ satisfy that $\pi_2(D) \ge (p^2-\varepsilon)n$.
\end{theorem}
%By
And in the same section, using Theorem \ref{thm:pi_2(T)only} and Theorem \ref{thm:RandomTPi_2}, we conclude
that random tournaments are
%%HB: shouldn't we add `almost surely'?
almost surely
path extendable.
\begin{theorem} \label{thm:RandomTPathExt}
Almost all tournaments are path extendable.
\end{theorem}
%%HB: I suggest to put the short proof here.
%An
%%HB: I suggest `key' instead of `important' so also changed `An' into `A'
%important
A key ingredient in this work is the
%%HB: I moved `newly introduced' and omitted more precisely surplus on $\pi_2$
newly introduced
concept
of surplus %more precisely surplus on $\pi_2$,
(see Section~\ref{sec:Surplus}). % we newly introduce.
%%HB: I added a connecting sentence
This concept is based on the following considerations.
%
Note that even in a doubly regular tournament, $p_2(u,v) \ne p_2(v,u)$ for every vertex pair $\{u,v\}$.
And in a general tournament, $p_2(u,v)$
%%HB: Small changes below for `smoothness'
%changes more often when
can assume more different values as
$\{u, v\}$ runs over all vertex pairs.
That is to say, there
%
might
%
exist a lot of vertex pairs $\{u,v\}$ %such that
with
$p_2(u,v) > \pi_2(u,v)$.
It turns out that, in order to estimate the total number of $2$-paths in a tournament $T$, or in part of $T$,
it
%would be
is
easier to consider $p_2(u,v)$ and $p_2(v,u)$ together, rather than
%
to
consider them separately.
%Thus
Based on this,
we define
%
the
surplus
%on
of a vertex pair $\{u,v\}$, obtained by subtracting the global lower bound
$2\pi_2(T)$ from $p_2(u,v)+p_2(v,u)$.
The surplus of a vertex set $U$ is then defined to be the sum of the
%%HB: the plural of surplus is surpluses
%surplus
surpluses
of all vertex pairs
$\{u,v\}$ %where $u,v\in U$.
in $U$.
In Section~\ref{sec:Surplus}, we estimate the possible surplus %on
of a vertex pair $\{u,v\}$, and
obtain an infimum for the surplus of a vertex set.
%, both of which
Both are used several times in the proofs that follow.
The rest of the paper
%%HB: I changed this to a standard formulation, as well as minor changes
%consists mainly the proofs of the new results listed in this section.
is organized as follows.
In Section~\ref{sec:Surplus}, we
%
formally
introduce the concept of surplus
and prove Theorem~\ref{thm:pi_2Supremum}.
In Section~\ref{sec:2PathsPathExt}, we study path extendability in regular tournaments and prove Theorem~\ref{thm:2PathsPathExt}.
In Section~\ref{sec:pi_2(T)i(T)}, we study conditions on $i(T)$ and $\pi_2(T)$ for path extendability,
and prove Theorem~\ref{thm:pi_2(T)i(T)}, Lemma~\ref{lem:i(T)AndPi(T)} and Theorem~\ref{thm:pi_2(T)only}.
In Section~\ref{sec:RandomTPi_2}, we investigate $2$-paths and path extendability in random oriented graphs, and prove
Theorem~\ref{thm:RandomTPi_2} and Theorem~\ref{thm:RandomTPathExt}.
In Section~\ref{sec:Final}, we conclude the paper with some final discussion.
%%HB: I must emphasize that I did not check the proof sections in detail, only globally. I concentrated my efforts on spelling, grammar and flow of the narrative. I did not mark obvious and minor changes in the sections that follow.
%%HB: title with only capital at the start, for consistency
\section{Surplus and supremum of $\boldsymbol{\pi_2(T)}$; proof of Theorem~\ref{thm:pi_2Supremum}} \label{sec:Surplus}
%%HB: I think the figure is superfluous and even confusing. I suggest to replace it by an explanation; see later suggestion for a solution.
%\begin{figure}
%\centering
%\includegraphics[width=0.5\linewidth]{Figures/DegreeAndTwoPaths.eps}
%\caption{Out arcs from and $2$-paths between vertices $u$ and $v$}
%\label{figure:DegreeAndTwoPaths}
%\end{figure}
Let $u$ and $v$ be two distinct vertices in a tournament $T$.
We define the \emph{surplus of $\{u,v\}$} by:
$$s(u,v)=p_2(u,v)+p_2(v,u)-2\pi_2(T).$$
%Because the number of $2$-paths between every vertex pair on each direction is at least $\pi_2(T)$, the surplus
%of a vertex pair is always positive.
It is obvious that $s(u,v)\ge 0$ and that it measures
%It records
the excess %number
of $2$-paths between a vertex pair in both directions
%towards
with regard to the minimum %number of
$2\pi_2(T)$.
%As can be seen in Figure \ref{figure:DegreeAndTwoPaths},
%%HB: Here is my suggested solution
%%It is easy to obtain the following facts. First suppose $u\rightarrow v$. Then
%%ZB: Remove "it is easy ..."
Suppose $u\rightarrow v$. Then
%if $u\rightarrow v$, we have
\begin{equation} \label{eqn:UdominatesV}
p_2(u,v)-p_2(v,u)=d^+(u)-d^+(v)-1.
\end{equation}
%
This follows from counting the cardinalities of the four different sets of common neighborhoods of $u$ and $v$.
Let $x_1=p_2(u,v)$ denote the number of vertices in $V(T)\setminus\{u,v\}$ that are out-neighbors of $u$ and in-neighbors of $v$,
$x_2$ the number of vertices in $V(T)\setminus\{u,v\}$ that are common out-neighbors of $u$ and $v$,
$x_3$ the number of vertices in $V(T)\setminus\{u,v\}$ that are common in-neighbors of $u$ and $v$,
and $x_4=p_2(v,u)$ the number of vertices in $V(T)\setminus\{u,v\}$ that are in-neighbors of $u$ and out-neighbors of $v$.
Then, $p_2(u,v)-p_2(v,u)=x_1-x_4$, whereas $d^+(u)-d^+(v)=x_1+1-x_4$. Hence, the above equality.
Similarly, if $v\rightarrow u$, we get
\begin{equation} \label{eqn:UdominatedByV}
p_2(u,v)-p_2(v,u)=d^+(u)-d^+(v)+1.
\end{equation}
%%HB: small changes before the lemma
%Without loss of generality we may assume
Now, suppose $d^+(u)\ge d^+(v)$. We discuss three cases.
%
%Suppose that
Case 1. $d^+(u)=d^+(v)$. %and we may
Let us further assume that $u\rightarrow v$.
%By
Then, by (\ref{eqn:UdominatesV}), we have $p_2(v,u)=p_2(u,v)+1 \ge \pi_2(T)+1$.
Hence, $s(u,v)\ge 1$.
%
Case 2. %Suppose
%
%next that
$d^+(u)=d^+(v)+1$. If $u\rightarrow v$,
then by (\ref{eqn:UdominatesV}), we have $p_2(v,u)=p_2(u,v)$.
Thus if $p_2(u,v)=p_2(v,u)=\pi_2(T)$, we have $s(u,v)=0$.
If $v\rightarrow u$, then by (\ref{eqn:UdominatedByV}), we have
$p_2(v,u)=p_2(u,v)+2 \ge \pi_2(T)+2$, and $s(u,v)\ge 2$.
%
%Suppose
%Finally, suppose that
Case 3. $d^+(u)\ge d^+(v)+2$. If $u\rightarrow v$,
then by (\ref{eqn:UdominatesV}), we have $p_2(u,v)=p_2(v,u)+1 \ge \pi_2(T)+1$, and
$s(u,v) \ge 1$.
If $v\rightarrow u$, then by (\ref{eqn:UdominatedByV}), we have
$p_2(v,u)=p_2(u,v)+3 \ge \pi_2(T)+3$, and $s(u,v)\ge 3$.
%
%%HB: I reformulated the below sentence to indicate that we will not prove the lemma.
%Some of
Using the above facts,
% are summarized in the below
it is easy to prove the next lemma.
\begin{lemma} \label{lem:VSurplus0}
%%HB: minor changes
Let $u$ and $v$ be two distinct vertices in a tournament $T$ with $d^+(u) \ge d^+(v)$.
If $u\rightarrow v$, then $s(u,v)\ge |d^+(u)-d^+(v)-1| \ge 0$.
If $v\rightarrow u$, then $s(u,v)\ge |d^+(u)-d^+(v)+1| \ge 1$.
In particular, if $s(u,v)=0$, then $d^+(u)-d^+(v)=1$ and $u\rightarrow v$.
\end{lemma}
%%HB: Again, only minor changes in the sequel.
We define
%
the
\emph{surplus of a vertex set} $W\subseteq V(T)$, denoted by $s(W)$,
as the sum of the surpluses of all vertex pairs $\{u,v\}$ in $W$.
Thus the number of $2$-paths in $T$ with end-vertices in $W$ is
$$2{|W| \choose 2}\pi_2(T)+s(W).$$
For any $u,v\in W$, by Lemma~\ref{lem:VSurplus0},
only if the out-degree of $u$ and $v$ differ by $1$ it is possible that $s(u,v)=0$.
Thus, to estimate the number of vertex pairs $\{u,v\}$ with $s(u,v)=0$,
we firstly calculate the vertex pairs whose out-degrees differ by $1$.
Let $n_i$ be the order of the set $\{v\mid v\in W, d^+_T(v)=i\}$ for $0\le i \le n-1$.
Then the number of vertex pairs $\{u,v\}$ in $W$ whose out-degrees differ by $1$ is
\begin{equation} \label{eqn:NumberOfSurplus0AtMost}
\sum_{i=0}^{n-2} n_in_{i+1}
\le \sum_{\substack{i=0 \\ i \text{ is odd}}}^{n-1} n_i \sum_{\substack{i=0 \\ i \text{ is even}}}^{n-1} n_{i}
= \sum_{\substack{i=0 \\ i \text{ is odd}}}^{n-1} n_i (|W|-\sum_{\substack{i=0 \\ i \text{ is odd}}}^{n-1} n_i )
\le \Big \lfloor \frac{|W|}{2}\Big \rfloor \Big \lceil \frac{|W|}{2}\Big \rceil.
\end{equation}
For any of the other (at least ${|W| \choose 2} - \lfloor |W|/2\rfloor \lceil |W|/2\rceil$)
vertex pairs $\{u, v\}$ whose out-degrees are equal or differ by at least $2$, we have $s(u,v) \ge 1$. Therefore,
\begin{equation} \label{eqn:UpperBoundSurplus}
s(W)=\sum_{\substack{u,v\in W\\ u\ne v}}s(u,v) \ge {|W| \choose 2} - \Big\lfloor \frac{|W|}{2}\Big\rfloor \Big\lceil \frac{|W|}{2}\Big\rceil.
\end{equation}
Now we use (\ref{eqn:UpperBoundSurplus}) to prove Theorem~\ref{thm:pi_2Supremum}.
\begin{proof} [\textit{Proof of Theorem \ref{thm:pi_2Supremum}}]
\noindent
Let $n$ be the number of vertices in $T$.
When $n\le 2$, there is no $2$-path in $T$.
Therefore, we may assume that $n\ge 3$.
We distinguish two cases according to the parity of $n$.
\medskip
\noindent
\textbf{Case 1. $\boldsymbol{n}$ is odd.}
\noindent
Considering any vertex $w\in V(T)$, the number of $2$-paths that go through $w$ is
\begin{equation} \label{eqn:IndegreeoutdegreeOdd} \nonumber
N^+(w)\cdot N^-(w)\le \Big (\frac{n-1}{2}\Big )^2.
\end{equation}
Therefore, the total number of $2$-paths in $T$ is at most $n(n-1)^2/4$. By (\ref{eqn:UpperBoundSurplus}),
\begin{equation} \label{eqn:CalAll2Paths} \nonumber
\begin{split}
n(n-1)\pi_2(T) + {n \choose 2} - \Big \lfloor \frac{n}{2}\Big \rfloor \Big\lceil \frac{n}{2} \Big\rceil
&= n(n-1)\pi_2(T) + {n \choose 2} - \frac{(n-1)(n+1)}{4} \\
&\le n(n-1)\pi_2(T) +s(V(T)) \\
%%Zanbo changes \le to = here
&= \sum_{\substack{u,v\in V(T) \\ u\neq v}} p_2(u,v) \\
&\le \frac{n(n-1)^2}{4},
\end{split}
\end{equation}
from which we get
$$4\pi_2(T) \le n-3+\frac{n+1}{n}=n-2+\frac{1}{n}.$$
Since $1/n <1$, $4\pi_2(T)$ is even and $n$ is odd, we have $4\pi_2(T)\le n-3$ and thus $\pi_2(T)\le (n-3)/4$.
\medskip
\noindent
\textbf{Case 2. $\boldsymbol{n}$ is even.}
\noindent
Similarly to Case 1, considering any vertex $w\in V(T)$, the number of $2$-paths that go through $w$ is
\begin{equation} \label{eqn:IndegreeoutdegreeEven}
N^+(w)\cdot N^-(w)\le \frac{n}{2}\Big (\frac{n}{2}-1 \Big )=\frac{n(n-2)}{4},
\end{equation}
and the total number of $2$-paths in $T$ is at most $n^2(n-2)/4$.
By (\ref{eqn:UpperBoundSurplus}),
the total number of $2$-paths in $T$ is at least $$n(n-1)\pi_2(T)+{n \choose 2} - \Big(\frac{n}{2}\Big)^2.$$
Therefore
\begin{equation} \label{eqn:CallAll2PathsEven}
n(n-1)\pi_2(T)+{n \choose 2} - \Big(\frac{n}{2}\Big)^2
\le \sum_{\substack{u,v\in V(T) \\ u\neq v}} p_2(u,v) \le \frac{n^2(n-2)}{4},
\end{equation}
$\noindent$from which we obtain $\pi_2(T)\le (n-2)/4$.
If $\pi_2(T)= (n-2)/4$, then all inequalities in (\ref{eqn:IndegreeoutdegreeEven})
and (\ref{eqn:CallAll2PathsEven}) are in fact equalities.
By equality in (\ref{eqn:IndegreeoutdegreeEven}),
$T$ must be a tournament with $n/2$ vertices of out-degree $n/2$ and $n/2$ vertices of out-degree $n/2-1$
(such a tournament is called \emph{almost regular}).
And in order that the first equality holds in (\ref{eqn:CallAll2PathsEven}),
for any vertex pair $\{u,v\}$ with $d^+(u)=n/2$ and $d^+(v)=n/2-1$, we must have $s(u,v)=0$.
By Lemma \ref{lem:VSurplus0}, this implies $u\rightarrow v$.
Hence, every vertex of out-degree $n/2$
dominates every vertex of out-degree $n/2-1$, and $T$ is not strong, a contradiction to $\pi_2(T)\ge 1$.
So $\pi_2(T)<(n-2)/4$.
Since $n$ is even, $\pi_2(T)\le (n-4)/4 < (n-3)/4$. This completes the proof of Theorem \ref{thm:pi_2Supremum}.
\end{proof}
Let $\langle W \rangle$ be the subtournament of $T$ induced by $W$.
The below lemma is a structural result on $\langle W \rangle$ when equality holds in (\ref{eqn:UpperBoundSurplus}).
\begin{lemma} \label{lem:Surplus}
Let $W$ be a subset of $V(T)$ for a tournament $T$.
Then $$s(W)\ge {|W| \choose 2} - \Big\lfloor \frac{|W|}{2}\Big\rfloor \Big\lceil \frac{|W|}{2}\Big\rceil.$$
Furthermore, equality holds only when one of the following holds.
(i) $|W|\le (|T|+1)/2$,
$W$ can be partitioned into $W_0$ and $W_1$, such that
$\{|W_0|,|W_1|\}=\{\lfloor |W|/2\rfloor, \lceil |W|/2 \rceil\}$,
$W_0$ consists of vertices of degree $d$ in $T$, $W_1$
consists of vertices of degree $d+1$ in $T$ for a certain integer $d$,
and $W_1 \rightarrow W_0$.
(ii) $|W|\le |T|/3+2$,
$W$ can be partitioned into $W_0$, $W_1$ and $W_2$, such that $|W_0|, |W_1|, |W_2| > 0$,
$\{|W_0|+|W_2|,|W_1|\}=\{\lfloor |W|/2\rfloor, \lceil |W|/2 \rceil\}$,
$W_0$, $W_1$, and $W_2$ consist of vertices of degree $d-1$, $d$ and $d+1$ in $T$
respectively, for a certain integer $d$, and $W_i \rightarrow W_j$ for $i>j$.
\end{lemma}
\begin{proof}
By (\ref{eqn:UpperBoundSurplus}), $s(W)\ge {|W| \choose 2} - \lfloor |W|/2 \rfloor \lceil |W|/2 \rceil.$
For equality to hold, we must have equality in (\ref{eqn:NumberOfSurplus0AtMost}) as well.
%
In order that all equalities hold in (\ref{eqn:NumberOfSurplus0AtMost}) and (\ref{eqn:UpperBoundSurplus}),
there can be at most three consecutive $n_i$ which are nonzero. Furthermore, the number of vertices of
odd degree and the number of vertices of even degree must be equal or they differ by exactly $1$.
\medskip
\noindent
\textbf{Case 1. There are exactly two consecutive $\boldsymbol{n_i}$ which are nonzero.}
\noindent
Assume that only $n_d$ and $n_{d+1}$ are nonzero.
Let the set of vertices in $W$ with degree $d$ and $d+1$ be $W_0$ and $W_1$, respectively.
Equalities in (\ref{eqn:NumberOfSurplus0AtMost}) and (\ref{eqn:UpperBoundSurplus}) imply that
$n_d=n_{d+1}$ or $|n_d-n_{d+1}|=1$,
that is $\{n_d, n_{d+1}\}=\{\lceil W/2 \rceil, \lfloor W/2 \rfloor \}$.
Furthermore, in order that $s(u,v)=0$ for $u\in W_1$ and $v\in W_0$, we must have $u\rightarrow v$,
hence $W_1 \rightarrow W_0$.
For any arc $u_0u_1$ with $u_0, u_1 \in W_1$ and any $v \in W_0$, we have a $2$-path $u_0u_1v$.
Similarly, for any arc $v_0v_1$ with $v_0, v_1 \in W_0$ and any $u \in W_1$, we have a $2$-path $uv_0v_1$.
%%Zanbo change the expression
Furthermore, in order that equality holds we must have $s(u,v)=0$, and hence $p_2(u,v)=\pi_2(T)$ for
every $u\in W_1$ and $v\in W_0$.
%%Zanbo add an explanation on what we are counting
Hence, calculating the total number of $2$-paths from every $u\in W_1$ to every $v\in W_0$, we have
\begin{equation} \nonumber
n_dn_{d+1}\pi_2(T)=\sum_{u\in W_1, v\in W_0} p_2(u,v) \ge \frac{n_{d+1}(n_{d+1}-1)}{2}\cdot n_d + \frac{n_{d}(n_{d}-1)}{2}\cdot n_{d+1}
=n_d n_{d+1}\Big(\frac{|W|}{2}-1\Big).
\end{equation}
So, $\pi_2(T) \ge |W|/2-1$.
%
However, by Theorem~\ref{thm:pi_2Supremum},
$\pi_2(T) \le (n-3)/4$, so $|W|/2-1 \le (n-3)/4$ and we get $|W| \le (n+1)/2$.
Thus (i) holds.
\medskip
\noindent
\textbf{Case 2. There are three consecutive $\boldsymbol{n_i}$ which are nonzero.}
\noindent
Assume that $n_{d-1}$, $n_d$ and $n_{d+1}$ are nonzero. Let the set of vertices of degree $d-1$, $d$ and $d+1$
be $W_0$, $W_1$ and $W_2$, respectively.
By equalities in (\ref{eqn:NumberOfSurplus0AtMost}),
$n_d=n_{d+1}+n_{d-1}$ or $|n_d-(n_{d+1}+n_{d-1})|=1$.
%% Zanbo add some statements
Hence $\{|W_0|+|W_2|, |W_1|\}=\{\lfloor|W|/2\rfloor, \lceil|W|/2\rceil\}$.
Furthermore, in order that $s(u,v)=0$, for $u\in W_1$ and $v\in W_0$ or $u\in W_2$ and $v\in W_1$,
we must have $u\rightarrow v$, hence $W_2 \rightarrow W_1 \rightarrow W_0$.
Moreover, we need $s(u,v)=1$ for $u\in W_0$ and $v\in W_2$, and by Lemma~\ref{lem:VSurplus0}, $u \rightarrow v$.
So, $W_2\rightarrow W_0$.
For any arc $u_0u_1$ where $u_0,u_1\in W_2$ and $v\in W_0$, we have a $2$-path $u_0u_1v$.
Similarly, for any arc $v_0v_1$ where $v_0,v_1\in W_0$ and $u\in W_2$, we have a $2$-path $uv_0v_1$.
Furthermore, for any $u\in W_2$, $w\in W_1$ and $v\in W_0$, we have a $2$-path $uwv$.
%%Zanbo add detail explanation.
Finally, for every $u\in W_2$ and $v\in W_0$,
$d^+(u)=d^+(v)+2$, and by Lemma \ref{lem:VSurplus0}, $s(u,v)\ge 1$.
In order that equalities in (\ref{eqn:NumberOfSurplus0AtMost}) and (\ref{eqn:UpperBoundSurplus}) hold
%%Zanbo change some statements here
we need $s(u,v)=1$, which holds only when $p_2(u,v)=p_2(v,u)+1=\pi_2(T)+1$ .
Therefore, considering the total number of $2$-paths from vertices in $W_2$ to vertices in $W_0$, we have
\begin{equation} \nonumber
\begin{split}
n_{d-1}n_{d+1}(\pi_2(T)+1) &=
\sum_{u\in W_2, v\in W_0} p_2(u,v) \\
&\ge
\frac{n_{d-1}(n_{d-1}-1)}{2}\cdot n_{d+1} + n_{d-1}n_d n_{d+1} + \frac{n_{d+1}(n_{d+1}-1)}{2}\cdot n_{d-1} \\
&= n_{d-1}n_{d+1}\Big(n_d+\frac{n_{d-1}+n_{d+1}}{2}-1\Big) \\
&= n_{d-1}n_{d+1}\Big(n_d+\frac{|W|-n_d}{2}-1\Big) \\
&= n_{d-1}n_{d+1}\Big(\frac{|W|+n_d}{2}-1\Big).
\end{split}
\end{equation}
Therefore
$$\pi_2(T)+1\ge \frac{|W|+ n_d}{2}-1.$$
By Theorem \ref{thm:pi_2Supremum}, $ \pi_2(T) \le (n-3)/4$. Thus we have
$$(n-3)/4+1 \ge (|W|+n_d)/2-1 \ge (|W|+\lfloor |W|/2 \rfloor)/2-1 \ge (|W|+ (|W|-1)/2)/2-1,$$
from which we get
$$|W| \le \frac{n}{3}+2,$$
and (ii) holds. This completes the proof of Lemma~\ref{lem:Surplus}.
\end{proof}
In a regular tournament, the lower bound on $s(W)$ can be improved.
\begin{lemma} \label{lem:SurplusRegular}
Let $W$ be a subset of $V(T)$ for a regular tournament $T$. Then
$$s(W)\ge {|W| \choose 2}.$$
\end{lemma}
\begin{proof}
Since $d^+(u)=d^+(v)$ for every $u,v\in W$, we have $s(u,v)\ge 1$. Therefore,
$$s(W)=\sum_{\substack{u,v\in W\\ u\ne v}}s(u,v) \ge {|W| \choose 2}.$$
\end{proof}
%%HB: I suggest to shorten the title
\section{%Path extendability of regular tournaments;
Regular Tournaments; Proof of Theorem~\ref{thm:2PathsPathExt}} \label{sec:2PathsPathExt}
Recall that Theorem \ref{thm:path3+Extendable} states that
a regular tournament $T$ with at least $7$ vertices is $\{2+\}$-path extendable,
unless $T\in \mathcal{T}_1$, $T\in \mathcal{T}_2$ or {$T=T_0$}.
The current proof is based on the structure of the exceptional graphs $\mathcal{T}_1$, $\mathcal{T}_2$ and {$T_0$},
which is given below.
\begin{figure}[h]
\centering
\includegraphics[width=0.85\linewidth]{Figures/TwoHybridVerticesAll.eps}
%%HB: slightly adapted the caption \caption{$D_{1}$ and tournaments in $\mathcal{T}_1$}
\caption{The digraph $D_{1}$ (left) and the tournaments in $\mathcal{T}_1$}
\label{figure:TwoHybridVerticesAll}
\end{figure}
%%HB: I prefer to have the description close(r) to the illustration, for convenience of the reader/referee; I also made small changes, but kept the old description at the original place in the file, in case you want to revert to this.
Let $D_{1}$ be the digraph at the left in Figure~\ref{figure:TwoHybridVerticesAll}.
Then $\mathcal{T}_1$ is the class of regular tournaments whose vertex set can be partitioned
into four parts $V=\{v_0, v_1, v_2, v_3, v_4\}$, $\{ u_0, u_1\}$, $N_0^+$ and $N_0^-$,
where $\langle V\cup \{u_0, u_1\} \rangle=D_{1}+u_0u_1$ or $D_{1}+u_1u_0$,
$|N_0^+|=|N_0^-|=0$ or $|N_0^+|=|N_0^-|\geq 3$, $V\rightarrow N_0^+$ and $N_0^- \rightarrow V$.
%Here $\langle A \rangle$ denotes the digraph induced by the vertex set $A$, and $A\rightarrow B$ indicates that there are arcs from any vertex in $A$ to any vertex in $B$.
A sketch of the members of the class $\mathcal{T}_1$ is depicted at the right in Figure~\ref{figure:TwoHybridVerticesAll}.
%%HB: I have also added the following; otherwise the reader/referee will note that the digraphs are not (regular) tournaments; I hope the text below is indeed what you meant
In the sketch there are missing arcs that should be added in order to turn the digraphs into regular tournaments. This might be done in an arbitrary manner, as long as one respects the constraint that
the resulting tournament is regular.
\begin{figure}[h]
\centering
\includegraphics[width=0.5\linewidth]{Figures/TwoHybridSets.eps}
%%HB: Added `The' to the caption
\caption{The tournaments in $\mathcal{T}_2$}
\label{figure:TwoHybridSets}
\end{figure}
%%HB: Also here I moved the text referring to the figure and made minor changes
Let $p\ge 3$ be an odd integer, and let $\mathcal{T}_2$ be the set of regular tournaments whose vertex set can be partitioned into five sets $V=\{v_i, 0\leq i \leq p-1\}$, $N_0^+$, $N_1^+$, $N_0^-$ and $N_1^-$,
%where $p\geq 3$ is odd,
such that $ |N_0^+|=|N_0^-| = n_0$, $|N_1^+|=|N_1^-|=n_1$, for $n_1\leq (p-1)/2$ and $n_0+n_1 \geq p$, and let $v_0v_1\ldots v_{p-1}$ be a path. Adopting the above convention, let $V\rightarrow N_0^+$, $N_0^-\rightarrow V$, $V\setminus \{v_0\} \rightarrow N_1^+$, $N_1^+\rightarrow v_0$, $N_1^- \rightarrow V\setminus \{v_{p-1}\} $ and $v_{p-1} \rightarrow N_1^-$.
See Figure~\ref{figure:TwoHybridSets} for a sketch of the members of $\mathcal{T}_2$.
%%HB: I added something similar as above
As in Figure ~\ref{figure:TwoHybridVerticesAll}, the missing arcs should be added in one way or another in order to turn the digraphs into regular tournaments.
\begin{figure}[h]
\centering
\includegraphics[width=0.27\linewidth]{Figures/Seven-Tournament-Exception.eps}
%%HB: Added `The tournament' to the caption
\caption{The tournament $T_0$}
\label{figure:Seven-Exception}
\end{figure}
%Let $D_{1}$ be the digraph in the left side of Figure \ref{figure:TwoHybridVerticesAll}.
%Let $\mathcal{T}_1$ be the class of regular tournaments whose vertex set can be partitioned
%into four parts $V=\{v_0, v_1, v_2, v_3, v_4\}$, $\{ u_0, u_1\}$, $N_0^+$ and $N_0^-$,
%where $\langle V\cup \{u_0, u_1\} \rangle=D_{1}+u_0u_1$ or $D_{1}+u_1u_0$,
%$|N_0^+|=|N_0^-|=0$ or $|N_0^+|=|N_0^-|\geq 3$, $V\rightarrow N_0^+$ and $N_0^- \rightarrow V$.
%See the right side of Figure \ref{figure:TwoHybridVerticesAll}.
%Let $\mathcal{T}_2$ be the set of regular tournaments whose vertices can be partitioned into $5$ sets $V=\{v_i, 0\leq i \leq p-1\}$, $N_0^+$, $N_1^+$, $N_0^-$ and $N_1^-$, where $p\geq 3$ is odd, $ |N_0^+|=|N_0^-| = n_0$, $|N_1^+|=|N_1^-|=n_1$, $n_1\leq (p-1)/2$, $n_0+n_1 \geq p$, $v_0v_1\ldots v_{p-1}$ is a path, $V\rightarrow N_0^+$, $N_0^-\rightarrow V$, $V\backslash \{v_0\} \rightarrow N_1^+$, $N_1^+\rightarrow v_0$, $N_1^- \rightarrow V\backslash \{v_{p-1}\} $ and $v_{p-1} \rightarrow N_1^-$.
%See Figure \ref{figure:TwoHybridSets}.
%%HB: small change below; note that the figures will shift if one of the two possible abstracts has been chosen; we have to see how the layout looks in the final version; it is important to have a nice layout, in order to `please' the referees!
Let $T_0$ be the regular tournament on seven vertices depicted in Figure~\ref{figure:Seven-Exception}.
Let $T$ be a regular tournament with $n\ge 9$ vertices and $\pi_2(T)> (n-9)/12 \ge 0$. This
implies that every path of length $1$ is extendable.
Moreover, by Theorem~\ref{thm:path3+Extendable}, either $T$ is $\{2+\}$-path extendable,
or $T$ belongs to the exceptional graph class $\mathcal{T}_1\cup \mathcal{T}_2 \cup \{T_0\}$.
If $T$ is $\{2+\}$-path extendable, then together with $\pi_2(T)\ge 1$, $T$ is path extendable.
Therefore, we only need to discuss the case that $T$ belongs to the exceptional graphs.
Since $n\ge 9$, $T\ne T_0$.
%
If $T \in \mathcal{T}_1$, then $T$ contains a path $P=v_0v_1v_2v_3v_4$ as illustrated in Figure~\ref{figure:TwoHybridVerticesAll}.
However, by the definition of $T$ and as shown in Figure~\ref{figure:TwoHybridVerticesAll},
$\{v_0, v_1\} \rightarrow N_0^+ \cup \{v_3, v_4\}$ ,
$N_1^+ \rightarrow \{v_0, v_1\}$ and $v_1 \rightarrow v_2 \rightarrow v_0$,
thus $p_2(v_0,v_1)=0$, contradicting $\pi_2(T)> 0$.
So we may assume that $T\in \mathcal{T}_2$. Consider the path $P=v_0v_1\ldots v_{p-1}$ in $T$.
If $p=3$, then there is no $2$-path from $v_0$ to $v_1$, contradicting $\pi_2(T)> 1$.
Thus we may further assume that $p\ge 5$.
Let $P^\prime=P[v_1, v_{p-2}]$. Then $|V(P^\prime)|=p-2\ge 3$.
Since $V(P^\prime)$ dominates $N_0^+\cup N_1^+$ and is dominated by $N_0^-\cup N_1^-$,
any $2$-paths between vertices in $V(P^\prime)$ must either be in $\langle V(P^\prime) \rangle$ or
go through $v_0$ or $v_1$.
For every $v_i\in V(P^\prime)$, the number of $2$-paths that go through $v_i$
in $\langle V(P^\prime)\rangle$ is at most $((p-3)/2)^2$.
Thus the total number of $2$-paths in $P^\prime$ is at most $(p-2)((p-3)/2)^2$.
Since $p-2$ is odd, the number of $2$-paths between vertices in $V(P^\prime)$
that go through $v_0$ or $v_{p-1}$ is at most $2((p-1)/2)((p-3)/2)=(p-1)(p-3)/2$.
So, the total number of $2$-paths between vertices in $V(P^\prime)$ is at most
\begin{equation} \nonumber
(p-2)\Big(\frac{p-3}{2}\Big)^2+\frac{(p-1)(p-3)}{2}.
\end{equation}
Since $T$ is regular, by Lemma~\ref{lem:SurplusRegular},
$s(V(P^\prime))\ge {p-2 \choose 2}$, and hence
the total number of $2$-paths among vertices in $V(P^\prime)$ is at least
$(p-2)(p-3)\pi_2(T)+{p-2 \choose 2}$. So, we have
\begin{equation} \nonumber
(p-2)(p-3)\pi_2(T)+{p-2 \choose 2} \le (p-2)\Big(\frac{p-3}{2}\Big)^2+\frac{(p-1)(p-3)}{2},
\end{equation}
$\noindent$by which we get
$$4\pi_2(T)\le p-3+\frac{2}{p-2}.$$
Since $p\ge 5$, $2/(p-2) \le 1$, and since $\pi_2(T)$ and $p$ are integers, $4\pi_2(T)\le p-3.$
Furthermore, by $n_0+n_1\ge p$ in the definition of $T\in \mathcal{T}_2$, we have $p\le n/3$. Thus, we obtain
$$\pi_2(T)\le \frac{1}{4}(\frac{n}{3}-3)=\frac{n-9}{12},$$
contradicting $\pi_2(T)>(n-9)/12$.
Concluding, we have shown that under the condition $\pi_2(T)>(n-9)/12$, $T$ cannot
be in the class of exceptional graphs, and thereby completed the proof of Theorem~\ref{thm:2PathsPathExt}.
%%HB: similar
\section{Condition on $\boldsymbol{i(T)}$ and $\boldsymbol{\pi_2(T)}$; %for Path Extendability in Tournaments:
Proof of Theorem~\ref{thm:pi_2(T)i(T)} and Theorem~\ref{thm:pi_2(T)only}}% and Theorem \ref{thm:pi_2(T)only}}
\label{sec:pi_2(T)i(T)}
\subsection{Proof of Theorem~\ref{thm:pi_2(T)i(T)}}
Let $P = u_0 u_1 \ldots u_{p-1}$ be a non-extendable path on $p$ vertices in a
tournament $T$, and let $w \in V(T)\setminus V(P)$.
We classify $w$ into three categories. If $w\rightarrow V(P)$, we say that $w$ is a \emph{dominating vertex} of $P$.
If $V(P)\rightarrow w$, we say that $w$ is a \emph{dominated vertex} of $P$. In the other case, $w$ is called a \emph{hybrid
vertex} of $P$. We denote the set of hybrid vertices of $P$ by $H(P)$, and we let $|H(P)|=h(P)$.
Supposing $w$ is a hybrid vertex of $P$, there cannot exist $0\le i < j \le p-1$ such that
$u_i \rightarrow w \rightarrow u_j$, or there must be an integer $i^\prime$ such that $i\le i^\prime \le j-1$ and
$u_{i^\prime} \rightarrow w \rightarrow u_{i^\prime+1}$,
and then $P$ can be extended to $u_0\ldots u_{i^\prime} w u_{i^\prime+1} \ldots u_{p-1}$,
a contradiction. However, there must be arcs from $w$ to $P$ and arcs from $P$ to $w$.
Therefore, there exists $0\le k \le q-2$ such that
$\{u_{k+1}, \ldots, u_{p-1}\} \rightarrow w \rightarrow \{u_0, \ldots, u_k\}$.
We say that $w$ \emph{switches at} $k$.
The concepts of dominating vertex, dominated vertex and hybrid vertex are applicable to any subpath $P_0$ of $P$.
A dominating (dominated) vertex of $P$ remains dominating (dominated) for $P_0$. A hybrid vertex of $P$ may
become dominating, dominated or stay hybrid for $P_0$.
In our proof, we distinguish the cases that $P^\prime=u_1u_2\ldots u_{p-2}$ has a hybrid vertex or not, or
equivalently, whether there exists a vertex $w\in V(T)\setminus V(P)$
which switches at $k_0$ for some $1\le k_0 \le p-3$.
We have the following result when $P^\prime$ has any hybrid vertex.
\begin{lemma} \label{lem:h(P)}
Let $P = u_0 u_1 \ldots u_{p-1}$ be a non-extendable path
%%HB: I think this is a mistake:
%of length $p$ in a
on $p\ge 4$ vertices in a
tournament $T$.
% with $p\ge 4$.
Let $P^\prime=u_1u_2\ldots u_{p-2}$.
If $P^\prime$ has a hybrid vertex,
then
%$P$ has at most $i(T)+2$ hybrid vertices, i.e,
$h(P)\leq i(T)+2$.
\end{lemma}
\begin{proof}
Suppose $P^\prime$ has a hybrid vertex.
Then there is at least one hybrid vertex $w$ switching at $1\leq k_0 \leq p-3$,
that is $u_{k_0+1}\rightarrow w \rightarrow u_{k_0}$.
We have $u_j \rightarrow w \rightarrow u_i$, for all $i \in [1, k_0]$ and $j \in [k_0+1, p-2]$. In particular,
$u_{p-2} \rightarrow w \rightarrow u_1$. For any $1\leq i \leq p-3$, if $u_0u_{i+1} \in A(T)$ and
$u_i u_{p-1} \in A(T)$, then $P$ can be extended to
%%HB: $P_1$ usually denotes a path on one vertex, so I suggest to avoid using it here and changed it to $Q$
%$P_1
$Q=u_0u_{i+1}\ldots u_{p-2}wu_1 \ldots u_i u_{p-1}$,
contradicting that $P$ is not extendable. Hence $|\{u_0u_{i+1}, u_iu_{p-1}\} \cap A(T) |\leq 1$, and
\begin{equation} \nonumber \label{DegreeonP}
d_P^+(u_0)+d_P^-(u_{p-1}) \leq p-3+1+1+2 = p+1.
\end{equation}
%%HB: What is $F$? I think $F=T-V(P)$, so I added it here:
Let $F=T-V(P)$. Then,
%Therefore,
\begin{equation} \label{eqn:DegreetoF}
\begin{split}
d^+_F(u_0) + d^-_F(u_{p-1})
&=d^+(u_0)+d^-(u_{p-1})-d_p^+(u_0)-d_p^-(u_{p-1}) \\
&\geq \frac{n-1-i(T)}{2}+\frac{n-1-i(T)}{2}-(p+1) \\
&=n-p-i(T)-2.
\end{split}
\end{equation}
For any $w\in H(P)$, we have $u_{p-1}\rightarrow w \rightarrow u_0$,
and hence $H(P)\subseteq N_F^-(u_0)\cap N_F^+(u_{p-1})$. Since $T$ is a tournament,
\begin{equation}\label{eqn:Completeness}N_F^-(u_0)\cap N_F^+(u_{p-1}) = V(F)\setminus (N_F^+(u_0) \cup N_F^-(u_{p-1})).\end{equation}
Furthermore, since $P$ is not extendable,
\begin{equation}\label{eqn:Nonintersect} N_F^+(u_0) \cap N_F^-(u_{p-1}) = \emptyset.\end{equation}
By (\ref{eqn:DegreetoF}), (\ref{eqn:Completeness}) and (\ref{eqn:Nonintersect}),
\begin{equation} \nonumber
\begin{split}
h(P)
&=|H(P)| \\
&= |N_F^-(u_0)\cap N_F^+(u_{p-1})| \\
&= |V(F)\setminus (N_F^+(u_0) \cup N_F^-(u_{p-1}))| \\
&= |F|- |N_F^+(u_0)| - |N_F^-(u_{p-1}))|\\
&= |F|- d_F^+(u_0) - d_F^-(u_{p-1}) \\
&\leq |F|-(n-p-i(T)-2) \\
&= i(T)+2,
\end{split}
\end{equation}
$\noindent$which completes the proof of the lemma.
\end{proof}
In the proof of Theorem~\ref{thm:pi_2(T)i(T)}, we will need the following theorem which provides a lower bound on the length of $P$.
It has been proved in \cite{Zhang2017}. For the sake of completeness, we include the proof here.
\begin{theorem} (Zhang, Theorem 5.18 in \cite{Zhang2017}) \label{thm:LowerBoundP}
Let $T$ be a tournament with $\pi_2(T)\ge 1$. If $P$ is a non-extendable path in $T$ on $p$ vertices,
then $p \ge 3\pi_2(T)+3$.
\end{theorem}
\begin{proof}
Let $P=u_0u_1\ldots u_{p-1}$. Furthermore, let $F=T-V(P)$. For $u,v\in V(T)$, $u\neq v$,
we denote the set of the intermediate vertices of all $(u,v)$-$2$-paths in $T$ by $V_I(u,v)$.
By the condition of the theorem, $|V_I(u,v)|\geq \pi_2(T) \geq 1$.
Since $P$ is not extendable, any $v\in V(F)$ can only be a dominating, dominated or hybrid vertex of $P$.
Therefore, for any integer $i$ and $j$ such that $0\leq i < j \leq p-1$,
there cannot exist a vertex $v\in V(F)\cap V_I(u_i,u_j)$. Hence, $V_I(u_i,u_j)\subseteq V(P)$.
Since $V_I(u,v)\geq 1$ for all vertex pairs $\{u,v\}$, all arcs are extendable.
Therefore, we may assume $|V(P)|\geq 3$.
Since $|V_I(u_0,u_1)|\geq \pi_2(T)$ and $V_I(u_0, u_1)\subseteq V(P)$,
$u_1$ has at least $\pi_2(T)$ in-neighbors on $P$ besides $u_0$.
So, we have $$|N^-_P(u_1)|\geq \pi_2(T)+1.$$
Consider the set of out-neighbors $N^+_P(u_1)$ of $u_1$ on $P$.
For any $u_i\in N^+_P(u_1)$, we have $i > 1$.
Hence, we have $V_I(u_1, u_i)\subseteq V(P)$, and furthermore, $V_I(u_1, u_i)\subseteq N^+_P(u_1)$.
Since $|V_I(u_1,u_i)|\geq \pi_2(T)$, $u_i$ has at least $\pi_2(T)$ in-neighbors in $N^+_P(u_1)$.
This means that in $\langle N^+_P(u_1) \rangle$ the minimum in-degree is at least $\pi_2(T)$.
Therefore, $|N^+_P(u_1)|(|N^+_P(u_1)|-1)/2 \geq \pi_2(T) |N^+_P(u_1)|$,
from which we get $$|N^+_P(u_1)|\geq 2\pi_2(T)+1.$$
Since $T$ is a tournament, $N^-_P(u_1) \cap N^+_P(u_1) = \emptyset$. We conclude that $p\geq 1+(\pi_2(T)+1)+(2\pi_2(T)+1)=3\pi_2(T)+3$.
\end{proof}
Now we have all the ingredients to prove Theorem~\ref{thm:pi_2(T)i(T)}.
\begin{proof}[\textit{Proof of Theorem~\ref{thm:pi_2(T)i(T)}}]
Let $P=u_0u_1\ldots u_{p-1}$ be a non-extendable path in $T$.
By Theorem~\ref{thm:LowerBoundP}, we have $p \ge 3\pi_2(T)+3 \ge 6$.
Thus we may further let $P^\prime = u_1\ldots u_{p-2}$,
where $|V(P^\prime)|\ge 4$.
We discuss the following two cases.
In Case 1, $P^\prime$ does not have a hybrid vertex.
In Case 2, $P^\prime$ has at least one hybrid vertex,
and thus by Lemma~\ref{lem:h(P)}, $h(P)\leq i(T)+2$.
\medskip
\noindent
\textbf{Case 1. $\boldsymbol{P^\prime}$ does not have a hybrid vertex.}
\noindent
The vertices of $V(T)\setminus V(P)$ can be divided into four sets, i.e.,
those dominating $V(P)$,
those dominating $V(P^\prime)\cup \{u_0\}$ but dominated by $u_{p-1}$,
those dominated by $V(P^\prime)\cup \{u_{p-1}\}$ but dominating $u_{0}$,
and those dominated by $V(P)$, denoted by $N_0^-$, $N_1^-$, $N_1^+$ and
$N_0^+$, respectively. The orders of $N_0^-$, $N_1^-$, $N_1^+$ and
$N_0^+$ are denoted by $n_0^-$, $n_1^-$, $n_1^+$ and $n_0^+$, respectively.
%
Calculating the sum of the out-degrees of all vertices in $N_0^+\cup N_1^+$,
we have
$$(n_0^++n_1^+)\cdot \frac{n-i(T)-1}{2} \leq \sum_{v\in N_0^+\cup N_1^+}d^+(v)=
n_1^++\frac{(n_0^++n_1^+)(n_0^++n_1^+-1)}{2}+d^+(N_1^+\cup N_0^+, N_0^-\cup N_1^-).$$
Therefore,
$$\frac{n_0^++n_1^+}{2}(n-i(T)-n_0^+-n_1^+)-n_1^+
\leq d^+(N_1^+\cup N_0^+, N_0^-\cup N_1^-)
\leq (n_0^-+n_1^-)(n_0^++n_1^+).$$
\textcolor{black}{If $n_0^++n_1^+=0$, we must have $n_0^-=0$, for there can be no $2$-path from any
$x\in V(P)\setminus \{u_{p-1}\}$ to a $y\in N_0^-$.
%%Zanbo add the below statement
Then $n_1^- \ne 0$.
Moreover, there can be only one $2$-path $xu_{p-1}y$ from
any $x\in V(P)\setminus \{u_{p-1}\}$ to a $y\in N_1^-$, thus $\pi_2(T)=1$.
By the condition of the theorem, $2\pi_2(T)-(n+8)/6 > i(T) \ge 0$.
Thus we get $n<4$, contradicting $n\ge 9$.}
Thus we may assume $n_0^++n_1^+\neq 0$ and get
$$n-n_0^+-n_1^+-i(T)-\frac{2n_1^+}{n_0^++n_1^+}\leq 2n_0^-+2n_1^-.$$
%%Zanbo change the formula a little
Note that $n-n_0^+-n_1^+=p+n_0^-+n_1^-$, so we have
\begin{equation} \label{eqn:CalOutdegree}
p\leq n_0^-+n_1^-+i(T)+\frac{2n_1^+}{n_0^++n_1^+}.
\end{equation}
Similarly, if we consider the sum of the in-degrees of all vertices in $N_0^-\cup N_1^-$,
we obtain
\begin{equation} \label{eqn:CalIndegree}
p\leq n_0^++n_1^++i(T)+\frac{2n_1^-}{n_0^-+n_1^-}.
\end{equation}
Summing up (\ref{eqn:CalOutdegree}) and (\ref{eqn:CalIndegree}), we get
$$2p\leq n-p+2i(T)+\frac{2n_1^+}{n_0^++n_1^+}+\frac{2n_1^-}{n_0^-+n_1^-}.$$
Therefore,
$$3p\leq n+2i(T)+\frac{2n_1^+}{n_0^++n_1^+}+\frac{2n_1^-}{n_0^-+n_1^-}\leq n+2i(T)+4.$$
So,
\begin{equation} \label{eqn:UpperBoundForp}
p\leq \frac{n+2i(T)+4}{3}.
\end{equation}
Next, we consider $2$-paths between vertex pairs in $P^\prime$. By Lemma~\ref{lem:Surplus},
$$\sum_{u,v\in V(P^\prime)}p_2(u,v)= (p-2)(p-3)\pi_2(T)+s(V(P^\prime)) \ge
(p-2)(p-3)\pi_2(T)+{p-2 \choose 2}- \Big\lceil\frac{p-2}{2}\Big\rceil\Big\lfloor\frac{p-2}{2}\Big\rfloor.$$
Since $P^\prime$ does not have a hybrid vertex,
the intermediate vertex of a $2$-path between a vertex pair in $P^\prime$ must be in $V(P)$.
The number of $2$-paths with all vertices in $V(P^\prime)$ is
at most $(p-2)\lfloor (p-3)/2 \rfloor \lceil (p-3)/2 \rceil$.
The number of $2$-paths between a vertex pair in $P^\prime$
that go through $u_0$ or $u_{p-1}$ is at most
$\lfloor (p-2)/2 \rfloor \lceil (p-2)/2 \rceil$.
So,
$$\sum_{u,v\in V(P^\prime)}p_2(u,v) \le (p-2)\Big\lfloor \frac{p-3}{2} \Big\rfloor \Big\lceil \frac{p-3}{2} \Big\rceil
+\Big\lfloor \frac{p-2}{2} \Big\rfloor \Big\lceil \frac{p-2}{2} \Big\rceil.$$
Thus,
\begin{equation} \nonumber
(p-2)(p-3)\pi_2(T)+{p-2 \choose 2}- \Big\lceil\frac{p-2}{2}\Big\rceil\Big\lfloor\frac{p-2}{2}\Big\rfloor
\le (p-2)\Big\lfloor \frac{p-3}{2} \Big\rfloor \Big\lceil \frac{p-3}{2} \Big\rceil
+\Big\lfloor \frac{p-2}{2} \Big\rfloor \Big\lceil \frac{p-2}{2} \Big\rceil,
\end{equation}
from which we obtain
\begin{equation} \label{eqn:pAndPi_2}
\pi_2(T) \le \frac{1}{p-3}\Big\lfloor \frac{p-3}{2} \Big\rfloor \Big\lceil \frac{p-3}{2} \Big\rceil
+\frac{2}{(p-2)(p-3)}\Big\lfloor \frac{p}{2}-1 \Big\rfloor \Big\lceil \frac{p}{2}-1 \Big\rceil - \frac{1}{2}.
\end{equation}
First assume that $p$ is odd. Using (\ref{eqn:pAndPi_2}), we get
$$p\ge 4\pi_2(T)+3-\frac{2}{p-2}.$$
Since $p\ge 6$, clearly $2/(p-2) < 1$. Since $p$ is an odd integer,
$$p\ge 4\pi_2(T)+3.$$
Next assume $p$ is even. Using (\ref{eqn:pAndPi_2}), we get
$$p\ge 4\pi_2(T)+3-\frac{1}{p-3}.$$
Since $p$ is an even integer, similarly to the above
$$p\ge 4\pi_2(T)+4 > 4\pi_2(T)+3.$$
Combining the above with (\ref{eqn:UpperBoundForp}), we have
$$4\pi_2(T)+3\le \frac{n+2i(T)+4}{3},$$
which simplifies to
$$i(T)\ge 6\pi_2(T)-\frac{n-5}{2}.$$
$\newline$
\begin{comment}
If $|P^\prime|$ is odd, then the number of $2$-paths with all vertices in $V(P^\prime)$ is
at most $(p-3)^2(p-2)/4$, while the number of $2$-paths between a vertex pair in $P^\prime$
that go through $u_0$ or $u_{p-1}$ is at most $(p-1)(p-3)/2$.
If $|P^\prime|$ is even, then the above two numbers become $(p-2)^2(p-4)/4$ and $(p-2)^2/2$.
It can be checked that with $p\geq 4$, we always have
$(p-3)^2(p-2)/4+(p-1)(p-3)/2 \geq (p-2)^2(p-4)/4+(p-2)^2/2$. Therefore,
$$\sum_{u,v\in V(P^\prime)}p_2(u,v) \leq \frac{(p-3)^2(p-2)}{4}+\frac{(p-1)(p-3)}{2}.$$
Hence, we have
$$\frac{(p-3)^2(p-2)}{4}+\frac{(p-1)(p-3)}{2} \geq (p-2)(p-3)\pi_2(T).$$
That is,
\begin{equation} \nonumber
p\geq 4\pi_2(T)+1-\frac{2}{p-2}.
\end{equation}
Since $p\ge 6$, and $p$ and $\pi_2(T)$ are all integers, we get
\begin{equation} \label{eqn:LowerBoundforp}
p \ge 4\pi_2(T)+1.
\end{equation}
By (\ref{eqn:UpperBoundForp}) and (\ref{eqn:LowerBoundforp}),
$$4\pi_2(T)+1 \leq (n+2i(T)+4)/3,$$
$\noindent$ which simplifies to
\begin{equation}
i(T)\geq 6\pi_2(T)-\frac{n+1}{2}.
\end{equation}
\end{comment}
\medskip
\noindent
\textbf{Case 2. $\boldsymbol{P^\prime}$ has a hybrid vertex.}
\noindent
By Lemma~\ref{lem:h(P)}, we have $h(P)\leq i(T)+2$.
Consider the sum of the number of $2$-paths between vertex pairs of $V(P)$. By Lemma~\ref{lem:Surplus}, we have
\begin{equation} \label{eqn:LowerBoundHybrid}
\sum_{u,v\in V(P)}p_2(u,v)= p(p-1)\pi_2(T)+s(V(P))
\ge p(p-1)\pi_2(T)+{p \choose 2}-\Big\lceil \frac{p}{2} \Big\rceil \Big\lfloor \frac{p}{2} \Big\rfloor.
\end{equation}
All $2$-paths between a vertex pair of $V(P)$ must have their intermediate vertex in $V(P)\cup H(P)$.
The number of those with an intermediate vertex in $V(P)$ is at most
$p \lfloor (p-1)/2 \rfloor \lceil (p-1)/2 \rceil$,
and of those with an intermediate vertex in $H(P)$ is at most
$ \lfloor p/2 \rfloor \lceil p/2 \rceil h(p)$. Thus,
\begin{equation} \label{eqn:UpperBoundHybrid}
\sum_{u,v\in V(P)}p_2(u,v)\leq
p \Big\lfloor \frac{p-1}{2} \Big\rfloor \Big\lceil \frac{p-1}{2} \Big\rceil +
\Big\lfloor \frac{p}{2} \Big\rfloor \Big\lceil \frac{p}{2} \Big\rceil h(p).
\end{equation}
\begin{comment}
If $p$ is odd, then the number of such $2$-paths going through $V(P)$ is at most $p(p-1)^2/4$,
and the number of those going through $H(P)$ is at most $(p-1)(p+1)h(p)/4$.
If $p$ is even, then the number of such $2$-paths going through $V(P)$ is at most $p^2(p-2)/4$,
and the number of those going through $H(P)$ is at most $p^2h(p)/4$.
\begin{equation}
\begin{split}
p(p-1)^2/4+(p-1)(p+1)h(p)/4-(p^2(p-2)/4+p^2h(p)/4) &= p/4-h(p)/4.
\end{split}
\end{equation}
$$\sum_{u,v\in V(P)}p_2(u,v)\leq \frac{p(p-1)^2}{4}+\frac{p^2h(P)}{4}.$$
\begin{equation} \nonumber
p \geq 4\pi_2(T)-h(P)-\frac{h(P)}{p-1}+1.
\end{equation}
\end{comment}
By (\ref{eqn:LowerBoundHybrid}) and (\ref{eqn:UpperBoundHybrid}),
\begin{equation} \nonumber
p(p-1)\pi_2(T)+{p \choose 2}-\Big\lceil \frac{p}{2} \Big\rceil \Big\lfloor \frac{p}{2} \Big\rfloor \le
p \Big\lfloor \frac{p-1}{2} \Big\rfloor \Big\lceil \frac{p-1}{2} \Big\rceil + \Big\lfloor \frac{p}{2} \Big\rfloor \Big\lceil \frac{p}{2} \Big\rceil h(p),
\end{equation}
that is,
\begin{equation} \label{eqn:Pi_2Hybrid}
\pi_2(T)\le \frac{1}{p-1}\Big\lfloor \frac{p-1}{2} \Big\rfloor \Big\lceil \frac{p-1}{2} \Big\rceil
+\frac{h(p)+1}{p(p-1)}\Big\lfloor \frac{p}{2} \Big\rfloor \Big\lceil \frac{p}{2} \Big\rceil - \frac{1}{2}.
\end{equation}
From (\ref{eqn:Pi_2Hybrid}) we get
\begin{equation} \label{eqn:PAndPi_2Hybrid} \nonumber
p \ge \left\{
\begin{array}
{l l}
4\pi_2(T)-h(P)+2-(h(P)+1)/p, & \mbox{if\ } p \mbox{ is odd,} \\
4\pi_2(T)-h(P)+2-h(P)/(p-1), & \mbox{if\ } p \mbox{ is even.}
\end{array}
\right.
\end{equation}
By Theorem~\ref{thm:LowerBoundP},
$p\ge 3\pi_2(T)+3 > 3(n-9)/12+3 = (n+3)/4$,
and hence $h(P)\le n-p < (3n-3)/4$.
Therefore $(h(P)+1)/p < ((3n-3)/4+1) / ((n+3)/4) = (3n+1)/(n+3)< 3$ and
$h(P)/(p-1)<((3n-3)/4)/ ((n+3)/4-1)= (3n-3)/(n-1)=3$.
Since $p$, $\pi_2(T)$ and $h(P)$ are integers, we have
\begin{equation} \label{eqn:LowerBoundforpWithh(P)}
p \geq 4\pi_2(T)-h(P)+2-2=4\pi_2(T)-h(P).
\end{equation}
Considering the out-degrees of the vertices of $N_0^+$, we have
$$n_0^+\cdot \frac{n-1-i(T)}{2} \leq \sum_{u\in N_0^+}d^+(u)
\leq \frac{n_0^+(n_0^+-1)}{2}+n_0^+(h(P)+n_0^-), $$
that is,
$$n-n_0^+-n_0^--h(P) \leq n_0^-+i(T)+h(P).$$
Note that $p=n-n_0^+-n_0^--h(P)$, thus we have
\begin{equation} \label{eqn:UpperBoundforPwithn0-}
p\leq n_0^-+i(T)+h(P).
\end{equation}
Similarly, we have
\begin{equation} \label{eqn:UpperBoundforPwithn0+}
p\leq n_0^++i(T)+h(P).
\end{equation}
Summing up (\ref{eqn:UpperBoundforPwithn0-}) and (\ref{eqn:UpperBoundforPwithn0+}), we have
\begin{equation}\nonumber
\begin{split}
2p &\leq n_0^-+n_0^++2i(T)+2h(P) \\
&= n-p-h(P)+2i(T)+2h(P) \\
&= n-p+h(P)+2i(T).
\end{split}
\end{equation}
Therefore,
\begin{equation} \label{eqn:UpperBoundforpSum}
3p\leq n+h(P)+2i(T).
\end{equation}
Combining (\ref{eqn:LowerBoundforpWithh(P)}) and (\ref{eqn:UpperBoundforpSum}), we obtain
$$12\pi_2(T)-3h(P) \leq 3p \leq n+h(P)+2i(T).$$
Using that $h(P)\leq i(T)+2$, we get
\begin{equation} \nonumber
\begin{split}
12\pi_2(T) &\le n+4h(P)+2i(T) \\
&\le n+4(i(T)+2) + 2i(T) \\
&= n+6i(T)+8,
\end{split}
\end{equation}
that is,
\begin{equation}\label{i(T)andpi_2(T)} \nonumber
i(T)\ge 2\pi_2(T)-\frac{n+8}{6}.
\end{equation}
Using the condition of the theorem that $\pi_2(T)>(n-9)/12$, we get
$$ 6\pi_2(T)-\frac{n-5}{2}-\Big(2\pi_2(T)-\frac{n+8}{6}\Big)
= 4\pi_2(T)-\frac{4n-46}{12} > \frac{4n-36}{12} -\frac{4n-46}{12}=\frac{5}{6}>0.$$
Hence in either case, $i(T)\ge 2\pi_2(T)-(n+8)/6$, contradicting the condition of the theorem.
\end{proof}
%Then Theorem \ref{thm:pi_2(T)only} is a straight consequence of Theorem \ref{thm:pi_2(T)i(T)}
%and Lemma \ref{lem:i(T)AndPi(T)}.
%
%\begin{proof} [\textbf{Proof of Theorem \ref{thm:pi_2(T)only}}]
%Let $T$ be a tournament satisfying the condition of the theorem.
%Note that $\pi_2(T) > (7n-10)/36>(n-9)/12$.
%Furthermore, by Lemma \ref{lem:i(T)AndPi(T)}, $i(T)\le n-4\pi_2(T)-3$, hence
%\begin{equation} \nonumber
%\begin{split}
% 2\pi_2(T)-\frac{(n+8)}{6} - i(T)
% &\ge 2\pi_2(T)-\frac{(n+8)}{6} - (n-4\pi_2(T)-3) \\
% &= 6\pi_2(T)-\frac{7n-10}{6} \\
% &= 6\Big(\pi_2(T)-\frac{7n-10}{36}\Big) \\
% &> 0
%\end{split}
%\end{equation}
%Therefore $2\pi_2(T)-(n+8)/6 > i(T)$, and by Theorem \ref{thm:pi_2(T)i(T)},
%$T$ is path extendable.
%\end{proof}
\subsection{Proof of Lemma~\ref{lem:i(T)AndPi(T)}}
%Take
By (\ref{eqn:i(T)andDegree}), either $\delta^+(T)=(n-1-i(T))/2)$ or $\delta^-(T)=(n-1-i(T))/2)$.
Assume the first equation holds and let $u\in V(T)$ be a vertex in $T$ such that $d^+(u)=\delta^+(T)=(n-1-i(T))/2)$.
Consider the subdigraph $T_u=\langle N^+(u)\rangle$ and any vertex $v$ in it. Every
$\{u,v\}$-$2$-path in $T$ must have its intermediate vertex in $T_u$. Therefore,
%%HB: we have not defined the minimum in-degree notation, so I added some words
for the minimum in-degree $\delta^-_{T_u}$ of $T_u$, we have
$\delta^-_{T_u} \geq \pi_2(T)$.
Thus,
%%HB: I added the following since $|T_u|$ is not defined
using $|T_u|$ as shorthand for $|V(T_u)|$,
$$\pi_2(T)|T_u|\leq \sum_{v\in T_u}d_{T_u}^-(v) = |T_u|(|T_u|-1)/2,$$
that is $|T_u|\geq 2\pi_2(T)+1$, that is, $\delta^+(T)\geq 2\pi_2(T)+1$.
Therefore, $2\pi_2(T)+1\leq (n-1-i(T))/2$, we have that
$i(T)\leq n-4\pi_2(T)-3$.
For the case that $\delta^-(T)=(n-1-i(T))/2)$, we can have a similar proof
by considering a vertex $u$ with $d^-(u)=\delta^-(T)=(n-1-i(T))/2)$ and the sum of
the out-degree of the vertex in the subgraph $T_u=\langle N^-(u)\rangle$.
\subsection{Proof of Theorem~\ref{thm:pi_2(T)only}}
%%HB: I think it makes sense to put the short proof of Theorem \ref{thm:pi_2(T)only} here
%%HB: I use italics instead of boldface for consistency with the standard proof environment in LaTeX
%\begin{proof} [\textit{Proof of Theorem \ref{thm:pi_2(T)only}}]
Let $T$ be a tournament satisfying the condition of the theorem.
Note that $\pi_2(T) > (7n-10)/36>(n-9)/12$.
Furthermore, by Lemma \ref{lem:i(T)AndPi(T)}, $i(T)\le n-4\pi_2(T)-3$, hence
\begin{equation} \nonumber
\begin{split}
2\pi_2(T)-\frac{(n+8)}{6} - i(T)
&\ge 2\pi_2(T)-\frac{(n+8)}{6} - (n-4\pi_2(T)-3) \\
&= 6\pi_2(T)-\frac{7n-10}{6} \\
&= 6\Big(\pi_2(T)-\frac{7n-10}{36}\Big) \\
&> 0.
%%HB: I added a .
\end{split}
\end{equation}
Therefore $2\pi_2(T)-(n+8)/6 > i(T)$, and by Theorem \ref{thm:pi_2(T)i(T)},
$T$ is path extendable.
%\end{proof}
\section{%Path Extendability of Random Tournaments:
Random Tournament; Proof of Theorem~\ref{thm:RandomTPi_2} and Theorem~\ref{thm:RandomTPathExt}}
%and Theorem \ref{thm:RandomTPathExt}}
\label{sec:RandomTPi_2}
\subsection{Proof of Theorem~\ref{thm:RandomTPi_2}}
We will need the following Chernoff bound in our proof.
\begin{lemma} \label{lem:Chernoff}
(Multiplicative Chernoff bound, \cite{MitUpf2017}). Suppose $X_1$, $\ldots$, $X_n$ are independent random variables taking values in $\{0, 1\}$,
where $Pr\{X_i=1\}=p_i$.
Let $X$ denote their sum and let $\mu = E[X]$ denote the expectation of $X$. Then for any $ 0<\delta <1$,
$$Pr\{X<(1-\delta)\mu\}<\bigg( \frac{e^{-\delta}}{(1-\delta)^{1-\delta}}\bigg)^\mu.$$
\end{lemma}
\begin{proof}[\textit{Proof of Theorem \ref{thm:RandomTPi_2}}]
Let $u$, $v$ be two vertices in $D \in \overrightarrow{G}(n,p)$. Let the other $n-2$ vertices be denoted by $w_i$,
and define the indicator variable $X_i$ as
\begin{equation} \nonumber
X_i= \left \{
\begin{array}
{l l}
1 & \mbox{if } u \rightarrow w_i \rightarrow v, \\
0 & \mbox{otherwise,}
\end{array}
\right.
\end{equation}
for $ 0 \le i \le n-3$. We clearly have $E[X_i]=Pr\{X_i=1\}=p^2$.
Further define $X=\sum_{i=0}^{n-3}X_i$. Then the variable corresponding to $p_2(u,v)$ is $X$, and $\mu=E[X]=(n-2)p^2$.
Applying the Chernoff bound of Lemma~\ref{lem:Chernoff} to $X$, for $0< \varepsilon < p^2$ as defined in the condition of the theorem, we have
\begin{equation} \nonumber
Pr\{X<(1-\varepsilon)(n-2)p^2\}<\bigg( \frac{e^{-\varepsilon}}{(1-\varepsilon)^{1-\varepsilon}}\bigg)^{(n-2)p^2}.
\end{equation}
Let $N=2(1-\varepsilon)p^2/(\varepsilon(1-p^2))$. For any integer $n>N$, we have
\begin{equation} \nonumber
\begin{split}
(1-\varepsilon)(n-2)p^2-(p^2-\varepsilon)n
&= (1-p^2)\varepsilon n-2 (1-\varepsilon) p^2 \\
&> (1-p^2)\varepsilon N-2 (1-\varepsilon) p^2 \\
&= 0.
\end{split}
\end{equation}
Therefore, $(1-\varepsilon)(n-2)p^2 > (p^2-\varepsilon)n >0$, and
$$Pr\{p_2(u,v)<(p^2-\varepsilon)n\} =
Pr\{X<(p^2-\varepsilon)n\} < Pr\{X<(1-\varepsilon)(n-2)p^2\}
<\bigg( \frac{e^{-\varepsilon}}{(1-\varepsilon)^{1-\varepsilon}}\bigg)^{(n-2)p^2}.$$
Consider the event $\pi_2(D) < (p^2-\varepsilon)n$,
which happens if and only if there is at least a vertex pair $\{u, v\}$ in $D$
such that $p_2(u,v) < (p^2-\varepsilon)n$.
This event is the union of the events $p_2(u,v) < (p^2-\varepsilon)n$,
where $\{u, v\}$ runs over all vertex pairs in $D$.
Applying Boole's inequality that the probability of the union of events
is at most the sum of the probabilities of every event, we get
\begin{equation} \nonumber
\begin{split}
Pr\{\pi_2(D)<(p^2-\varepsilon)n \}
&= Pr\Big\{\bigcup_{\substack{u,v\in V(D) \\ u\ne v}}p_2(u,v)<(p^2-\varepsilon)n \Big\} \\
&\le \sum_{\substack{u,v\in V(D) \\ u\ne v}} Pr\{p_2(u,v)<(p^2-\varepsilon)n \}\\
&< n(n-1) \bigg( \frac{e^{-\varepsilon}}{(1-\varepsilon)^{1-\varepsilon}}\bigg)^{(n-2)p^2}.
\end{split}
\end{equation}
Let $g(x)=e^{-x}/(1-x)^{1-x}$, $0\le x \le 1/4$. Then $g'(x)=e^{-x}(1-x)^{x-1}\log (1-x)$.
Since $g'(x)<0$ for $0<x\le 1/4$, $g(x)$ is strictly decreasing in $(0,1/4)$, and $g(x)<g(0)=1$.
Therefore, $ e^{-\varepsilon}/(1-\varepsilon)^{1-\varepsilon} <1$, and for a fixed $p>0$,
$$ \lim_{n\rightarrow \infty} n(n-1)\bigg( \frac{e^{-\varepsilon}}{(1-\varepsilon)^{1-\varepsilon}}\bigg)^{(n-2)p^2}=0,$$
yielding
$$\lim_{n\rightarrow \infty} Pr\{\pi_2(D)\ge (p^2-\varepsilon)n \}
=1-\lim_{n\rightarrow \infty} Pr\{\pi_2(D)< (p^2-\varepsilon)n \}
=1.$$
We conclude that almost all oriented graphs $D\in \overrightarrow{G}(n,p)$ satisfy $\pi_2(D)\ge (p^2-\varepsilon)n$.
\end{proof}
%\begin{proof}[\textbf{Proof of Theorem \ref{thm:RandomTPathExt}}]
%Letting $p=1/2$ in $\overrightarrow{G}(n,p)$, we have a model for random tournament.
%By Theorem \ref{thm:RandomTPi_2}, for $0 < \varepsilon <1/4$, almost all tournament $T$ satisfy
%$\pi_2(T)\ge (1/4-\varepsilon)n$. For $\varepsilon < 1/18$ we have
%$(1/4- \varepsilon)n > 7n/36 > (7n-10)/36$, thus by Theorem \ref{thm:pi_2(T)only},
%almost all tournaments are path extendable.
%\end{proof}
\subsection{Proof of Theorem \ref{thm:RandomTPathExt}}
Letting $p=1/2$ in $\overrightarrow{G}(n,p)$, we
%have
obtain
a model for random tournaments.
By Theorem~\ref{thm:RandomTPi_2}, for $0 < \varepsilon <1/4$, almost all tournaments $T$ satisfy
$\pi_2(T)\ge (1/4-\varepsilon)n$. %For
Letting $\varepsilon < 1/18$, we %have
get
$(1/4- \varepsilon)n > 7n/36 > (7n-10)/36$. %thus
Hence, by Theorem~\ref{thm:pi_2(T)only},
almost all tournaments are path extendable.
\section{Conclusion and open problems} \label{sec:Final}
Earlier results on tournaments have revealed that in tournaments strongness implies hamiltonicity, pancyclicity
and cycle extendability (with an exceptional class), but not path extendability.
However, by considering analogous structural results on hamiltonian properties of undirected graphs, we were tempted to expect that imposing similar conditions as in
Theorem~\ref{thm:irregularitystronglypanconnected}
and Theorem~\ref{thm:2pathallpaths} on a tournament would imply that it is path extendable.
%With this in mind, the following problems arise quite naturally.
%%
%Bondy's meta-conjecture with the extension to cycle extendability is valid
%in tournaments (and bipartite tournaments as well, see \cite{ZZGL2021}).
Our results for path extendability show that the situation is in fact quite different.
%As we have seen, it is surprising that,
Although path extendability seems to be
asking only a little bit more than strong panconnectedness
and completely strong path-connectedness,
to guarantee path extendability we need to impose significantly %enhanced
stronger conditions,
in particular %, we are prompted to develop a new type of
conditions that
combine restrictions on $i(T)$ and $\pi_2(T)$.
%
%We
Based on our work, we leave the following two problems %below
for future research.
%%HB: I added a `connecting' sentence here. I only mention remarks on the lemma because I suggest to omit the remarks about the two theorems. See below.
%Before we continue with our results on random tournaments, let us first share some remarks about Lemma \ref{lem:i(T)AndPi(T)}.
%\medskip
%\noindent
%\textbf{Remarks on Lemma \ref{lem:i(T)AndPi(T)}.}
Lemma \ref{lem:i(T)AndPi(T)} shows that
%%HB: I changed this to a shorter statement, avoiding repetition of knows facts
%, when $\pi_2(T)$ is large, i.e., when there are many $2$-paths
%between every vertex pair, $i(T)$ becomes small, or equivalently, the tournament must be close to regular.
a larger $\pi_2(T)$ implies a smaller $i(T)$. In particular,
%%HB: Also some minor adjustments below
%Let $\pi_2(T)$ be its supermum $(n-3)/4$ we get $i(T)=0$ in this lemma.
if $\pi_2(T)$ attains its supremum $(n-3)/4$, we get $i(T)=0$, and $T$ is actually doubly regular.
In this sense, the result is tight.
It
%
is
worth noting that the reverse does not hold, that is,
a tournament $T$ with a small $i(T)$ need not
%to
have %
a
large $\pi_2(T)$.
Consider any tournament $T$ on $n=2k+1$ vertices
%%HB: small changes
for some integer
%with
$k\ge 3$, % presented in Figure \ref{figure:RegularNoTwoPath},
constructed as follows (see Figure~\ref{figure:RegularNoTwoPath}).
%
\begin{figure}[h]
\centering
\includegraphics[width=0.28\linewidth]{Figures/RegularNoTwoPath.eps}
\caption{A regular tournament without $(u_0,u_1)$-$2$-paths}
\label{figure:RegularNoTwoPath}
\end{figure}
%
Take a regular tournament $T^\prime$ on $2k-3$ vertices.
Let $V(T^\prime)=V_0 \cup V_1 \cup \{v\}$ be a partition of $V(T^\prime)$ with $|V_0|=|V_1|=k-2$.
$T$ is constructed by adding four new vertices $u_0$, $u_1$, $w_0$ and $w_1$ to $T^\prime$,
and adding arcs so that $V_1 \rightarrow \{u_0, u_1\} \rightarrow V_0$,
$V_0 \rightarrow \{w_0, w_1\} \rightarrow V_1$,
$w_1 \rightarrow \{u_0, u_1\} \rightarrow w_0$,
$u_0 \rightarrow u_1 \rightarrow v \rightarrow u_0$,
$w_0 \rightarrow v \rightarrow w_1$ and $w_0\rightarrow w_1$.
It can be verified that $T$ is regular but there is no $(u_0,u_1)$-$2$-path in $T$.
%This also supports our aforementioned statement that $\pi_2(T)=1$ seems to be a stronger conditon
%than being regular for path-connected properties.
It may be interesting to characterize the class of tournaments for which equality holds in
Lemma~\ref{lem:i(T)AndPi(T)}, which includes all doubly regular tournaments.
\begin{problem} Characterize the class of tournaments $T$ on $n$ vertices with $i(T)=n-4\pi_2(T)-3$. \end{problem}
If we let $i(T)=0$ in Theorem~\ref{thm:pi_2(T)i(T)},
then $T$ is a regular tournament and the condition becomes $\pi_2(T)>(n+8)/12$.
There is a clear gap between this bound and the sharp one $(n-9)/12$ of Theorem~\ref{thm:2PathsPathExt}.
Thus, there seems to be some space to further tighten the conditions in
Theorem~\ref{thm:pi_2(T)i(T)} and Theorem~\ref{thm:pi_2(T)only}.
However, we believe such improvements will require a different approach with a much more subtle and detailed structural analysis.
\begin{problem} Improve the bound $i(T)< 2\pi_2(T)-(n+8)/6$ for path extendability
of a tournament $T$ on $n$ vertices in Theorem~\ref{thm:pi_2(T)i(T)}. \end{problem}
\begin{thebibliography}{99}
%\bibitem{AlsRei1974}
%Alspach, B., Reid, K. B., \& Roselle, D. P. (1974). Bypasses in asymmetric digraphs. Journal of Combinatorial Theory, Series B, 17(1), 11-18.
%%Zanbo add the reference for terms and notation
\bibitem{BanGut2008}
Bang-Jensen, J., \& Gutin, G. (2008). Digraphs: theory, algorithms and applications. Springer Science \& Business Media.
%\bibitem{GuoVol1997}
%Guo, Y., \& Volkmann, L. (1997). Bypaths in tournaments. Discrete applied mathematics, 79(1), 127-135.
%\bibitem{McKay1990}
%McKay, B. D. (1990). The asymptotic numbers of regular tournaments, eulerian digraphs and eulerian oriented graphs. Combinatorica, 10(4), 367-377.
\bibitem{Bondy1971}
Bondy, J. A. (1971). Pancyclic graphs I. Journal of Combinatorial Theory, Series B, 11(1), 80-84.
\bibitem{Bondy1973}
Bondy, J.A. (1973). Pancyclic graphs: recent results.
In Colloquia Mathematica Societatis Janos Bolyai 10. Infinite and Finite Sets, Keszthely (Hungary).
%\bibitem{BJR2006}
%Busch, A. H., Jacobson, M. S., \& Reid, K. B. (2006).
%On arc--traceable tournaments. Journal of Graph Theory, 53(2), 157-166.
%\bibitem{Cuckler2007}
%Cuckler, B. (2007). Hamiltonian cycles in regular tournaments. Combinatorics,
%Probability and Computing, 16(2), 239-249.
%\bibitem{DDF2021}
%Dragani\'{c}, N., Dross, F., Fox, J., Gir\~{a}o, A., Havet, F., Kor\'{a}ndi, D.,
%William, L., Correia, D.M., Scott, A. \& Sudakov, B. (2021).
%Powers of paths in tournaments.
%Combinatorics, Probability and Computing, 30(6), 894-898.
\bibitem{FG1996}
Faudree, R. J., \& Gy\'{a}rf\'{a}s, A. (1996).
Endpoint extendable paths in tournaments. Journal of Graph Theory, 23(3), 303-307.
%\bibitem{FK2005}
%Friedgut, E., \& Kahn, J. (2005).
%On the number of Hamiltonian cycles in a tournament.
%Combinatorics, Probability and Computing, 14(5-6), 769-781.
\bibitem{Hendry1989}
Hendry, G. R. (1989). Extending cycles in directed graphs. Journal of Combinatorial Theory, Series B, 46(2), 162-172.
\bibitem{Hendry1990a}
Hendry, G. R. (1990). Extending cycles in graphs. Discrete Mathematics, 85(1), 59-72.
\bibitem{Hendry1990b}
Hendry, G. R. (1990). Path extendable graphs. Periodica Mathematica Hungarica, 21(3), 205-218.
\bibitem{MitUpf2017}
Mitzenmacher, M., \& Upfal, E. (2017).
Probability and computing: Randomization and probabilistic techniques in algorithms and data analysis.
Cambridge university press.
\bibitem{Moon1969}
Moon, J. W. (1969). On cycles in tournaments. Matematick\'{y} \v{c}asopis, 19(2), 121-125.
\bibitem{ReiBei1978}
Reid, K. B., \& Beineke, L. W. (1978). Tournaments. Selected topics in graph theory, 169-204.
\bibitem{ReiBro1972}
Reid, K. B., \& Brown, E. (1972). Doubly regular tournaments are equivalent to skew Hadamard matrices.
Journal of Combinatorial Theory, Series A, 12(3), 332--338.
%\bibitem{Spencer1974}
%Spencer, J. (1974). Random regular tournaments. Periodica Mathematica Hungarica, 5(2), 105-120.
\bibitem{Thomassen1980}
Thomassen, C. (1980). Hamiltonian-connected tournaments. Journal of Combinatorial Theory, Series B, 28(2), 142-163.
%\bibitem{TiaWuZha1982}
%Tian, F., Wu, Z. S., \& Zhang, C. Q. (1982). Cycles of each length in tournaments. Journal of Combinatorial Theory, Series B, 33(3), 245-255.
\bibitem{Zhang1982}
Zhang, K. (1982). Completely strong path-connected tournaments. Journal of Combinatorial Theory, Series B, 33(2), 166--177.
\bibitem{Zhang2017}
Zhang, Z. (2017). Paths, cycles and related partitioning problems in graphs. PhD Thesis, University of Twente.
\bibitem{ZZBL2017} Zhang, Z.-B., Zhang, X., Broersma, H. \& Lou, D. (2017).
Extremal and Degree Conditions for Path Extendability in Digraphs. SIAM J. Discrete Math., 31(3), 1990--2014.
%\bibitem{ZZGL2021}
%Zhang, Z. B., Zhang, X., Gutin, G., \& Lou, D. (2021).
%Hamiltonicity, pancyclicity, and full cycle extendability in multipartite tournaments.
%Journal of Graph Theory, 96(2), 171-191.
\end{thebibliography}
\end{document}