%% This document created by Scientific Word (R) Version 3.5
% Please remove all other commands that change parameters such as
% margins or pagesizes.
% only use standard LaTeX packages
% only include packages that you actually need
% we recommend these ams packages
% we recommend the graphicx package for importing figures
% use this command to create hyperlinks (optional and recommended)
% \date{\dateline{submission date}{acceptance date}\\
% \small Mathematics Subject Classifications: comma separated list of
% MSC codes available from http://www.ams.org/mathscinet/freeTools.html}
\documentclass[12pt]{article}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\usepackage{e-jc}
\usepackage{amsthm,amsmath,amssymb}
\usepackage{graphicx}
\usepackage[colorlinks=true,citecolor=black,linkcolor=black,urlcolor=blue]{hyperref}
\setcounter{MaxMatrixCols}{10}
%TCIDATA{OutputFilter=Latex.dll}
%TCIDATA{Version=5.00.0.2552}
%TCIDATA{}
%TCIDATA{LastRevised=Saturday, January 11, 2014 17:02:49}
%TCIDATA{}
\theoremstyle{plain}
\newtheorem{theorem}{Theorem}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{fact}[theorem]{Fact}
\newtheorem{observation}[theorem]{Observation}
\newtheorem{claim}[theorem]{Claim}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{open}[theorem]{Open Problem}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{question}[theorem]{Question}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{note}[theorem]{Note}
\newcommand{\arxiv}[1]{\href{http://arxiv.org/abs/#1}{\texttt{arXiv:#1}}}
%\input{tcilatex}
\begin{document}
\title{\textbf{Some extremal problems for hereditary properties of graphs}}
\author{Vladimir Nikiforov \\
%EndAName
{\small Department of Mathematical Sciences }\\
[-0.8ex] {\small University of Memphis, Memphis, TN, USA }\\
{\small \texttt{vnikifrv@memphis.edu}}\\
}
\date{\dateline{Jan 1, 2012}{Jan 2, 2012}\\
{\small Mathematics Subject Classifications: 05C65, 05C35}}
\maketitle
\begin{abstract}
Given an infinite hereditary property of graphs $\mathcal{P},$ the principal
extremal parameter of $\mathcal{P}$ is the value%
\begin{equation*}
\pi\left( \mathcal{P}\right) =\lim_{n\rightarrow\infty}\binom{n}{2}%
^{-1}\max\{e\left( G\right) :\text{ }G\in\mathcal{P}\text{ and }v\left(
G\right) =n\}.
\end{equation*}
The Erd\H{o}s-Stone theorem gives $\pi\left( \mathcal{P}\right) $ if $%
\mathcal{P}$\ is monotone, but this result does not apply to hereditary $%
\mathcal{P}$. Thus, one of the results of this note is to establish $%
\pi\left( \mathcal{P}\right) $ for any hereditary property $\mathcal{P}.$
Similar questions are studied for the parameter $\lambda^{\left( p\right)
}\left( G\right) ,$ defined for every real number $p\geq1$ and every graph $%
G\ $of order $n$ as
\begin{equation*}
\lambda^{\left( p\right) }\left( G\right) =\max_{\left\vert x_{1}\right\vert
^{p}\text{ }+\text{ }\cdots\text{ }+\text{ }\left\vert x_{n}\right\vert ^{p}%
\text{ }=\text{ }1}2\sum_{\{u,v\}\in E\left( G\right) }x_{u}x_{v}.
\end{equation*}
It is shown that the limit%
\begin{equation*}
\lambda^{\left( p\right) }\left( \mathcal{P}\right) =\lim_{n\rightarrow
\infty}n^{2/p-2}\max\{\lambda^{\left( p\right) }\left( G\right) :\text{ }G\in%
\mathcal{P}\text{ and }v\left( G\right) =n\}
\end{equation*}
exists for every hereditary property $\mathcal{P}$.
A key result of the note is the equality%
\begin{equation*}
\lambda^{(p)}\left( \mathcal{P}\right) =\pi\left( \mathcal{P}\right) ,
\end{equation*}
which holds for all $p>1.$ In particular, edge extremal problems and
spectral extremal problems for graphs are asymptotically equivalent.
\bigskip\noindent \textbf{Keywords:} extremal problems; Tur\'{a}n problems;
hereditary property; largest eigenvalue.
\end{abstract}
\section{Introduction and main results}
In this note we study problems stemming from the following one:\
\begin{problem}
\label{p1}What is the maximum number of edges in a graph $G$ of order $n$ if
$G$ belongs to some hereditary property $P$.
\end{problem}
Let us recall the basics of graph properties: A \textbf{graph property} is
just a family of graphs closed under isomorphisms. A property is called
\textbf{monotone }if it is closed under taking subgraphs, and \textbf{%
hereditary }if it is closed under taking induced subgraphs. Given a set of
graphs $\mathcal{F},$ the family of all graphs that do not contain any $F\in
\mathcal{F}$ as a subgraph is a monotone property, denoted by $Mon\left(
\mathcal{F}\right) .$ Likewise, the family of all graphs that do not contain
any $F\in \mathcal{F}$ as an induced subgraph is a hereditary property,
denoted as $Her\left( \mathcal{F}\right) .$
It seems that the classically shaped Problem \ref{p1} has been disregarded
in the rich literature on hereditary properties, so in this paper we shall
fill in this gap. Note, however, that for monotone properties the theorem of
Erd\H{o}s and Stone provides a well-known solution, outlined in Proposition %
\ref{proES} below.
Writing $\mathcal{P}_{n}$ for the set of all graphs of order $n$ in a
property $\mathcal{P},$ now Problem \ref{p1} reads as: \medskip
\emph{Given a hereditary property }$\mathcal{P}$\emph{, find}
\begin{equation}
ex\left( \mathcal{P},n\right) =\max_{G\in\mathcal{P}_{n}}e\left( G\right) .
\label{maxed}
\end{equation}
Finding $ex\left( \mathcal{P},n\right) $ exactly seems hopeless for
arbitrary $\mathcal{P}$. A more feasible approach has been suggested by
Katona, Nemetz and Simonovits in \cite{KNS64} who proved the following fact:
\begin{proposition}
\label{proKNS}If $\mathcal{P}$ is a hereditary property, then the limit
\begin{equation*}
\pi\left( \mathcal{P}\right) =\lim_{n\rightarrow\infty}ex\left( \mathcal{P}%
,n\right) \binom{n}{2}^{-1}
\end{equation*}
exists.
\end{proposition}
In particular, for monotone properties Erd\H{o}s and Simonovits \cite{ErSi66}
observed the following consequence of the Erd\H{o}s-Stone theorem \cite%
{ErSt46}:
\begin{proposition}
\label{proES}If a monotone property $\mathcal{P}$ is given as $\mathcal{P}%
=Mon\left( \mathcal{F}\right) $ for some nonempty family $\mathcal{F},$ then
\begin{equation*}
\pi\left( Mon\left( \mathcal{F}\right) \right) =1-1/\chi,
\end{equation*}
where $\chi=\min\left\{ \chi\left( F\right) :F\in\mathcal{F}\right\} .$
\end{proposition}
Unfortunately, $\pi\left( \mathcal{P}\right) $ cannot be determined in the
same simple way for a general hereditary property $\mathcal{P}$, and so one
of the aims of this note is to establish $\pi\left( \mathcal{P}\right) $ for
such $\mathcal{P}$. However, our main focus is on extremal problems about a
more general graph parameter, denoted by $\lambda^{\left( p\right) }\left(
G\right) $ and defined as follows: \medskip
\emph{Given a graph }$G$\emph{\ and a real number }$p\geq1,$\emph{\ let}
\begin{equation*}
\lambda^{\left( p\right) }\left( G\right) =\max_{\left\vert x_{1}\right\vert
^{p}+\text{ }\cdots\text{ }+\left\vert x_{n}\right\vert
^{p}=1}2\sum_{\left\{ u,v\right\} \in E\left( G\right) }x_{u}x_{v}.
\end{equation*}
Note that $\lambda^{\left( 2\right) }\left( G\right) $ is the well-studied
spectral radius of $G,$ and also that $\lambda^{\left( 1\right) }\left(
G\right) $ is a another much studied graph parameter, known as the Lagrangian%
\footnote{%
Let us note that this use of the name \emph{Lagrangian} is at odds with the
tradition. Indeed, names as \emph{Laplacian, Hessian, Gramian, Grassmanian},
etc., usually denote a structured object like matrix, operator, or manifold,
and not just a single number.} of $G$. Moreover, letting $%
p\rightarrow\infty, $ one can show that $\lambda^{\left( p\right) }\left(
G\right) \rightarrow e\left( G\right) .$ So $\lambda^{\left( p\right)
}\left( G\right) $ is a common generalization of three central parameters in
extremal graph theory.
The parameter $\lambda^{\left( p\right) }\left( G\right) $ has been
introduced and studied for uniform hypergraphs first by Keevash, Lenz and
Mubayi in \cite{KLM13}, and next by the author in \cite{NikB} and \cite{NikC}%
. Here we shall study $\lambda^{\left( p\right) }\left( G\right) $ in the
same role as $e\left( G\right) $ in equation (\ref{maxed}), obtaining thus
the following problem.
\begin{problem}
\label{p2}Given a hereditary property $\mathcal{P},$ find%
\begin{equation}
\lambda^{\left( p\right) }\left( \mathcal{P},n\right) =\max_{G\in \mathcal{P}%
_{n}}\lambda^{\left( p\right) }\left( G\right) . \label{maxlam}
\end{equation}
\end{problem}
As for $ex\left( \mathcal{P},n\right) ,$ finding $\lambda^{\left( p\right)
}\left( \mathcal{P},n\right) $ seems hopeless for arbitrary $\mathcal{P},$
so we begin with an analog to Proposition \ref{proKNS}.
\begin{theorem}
\label{thlim}Let $p\geq1.$ If $\mathcal{P}$ is a hereditary property, then
the limit
\begin{equation*}
\lambda^{\left( p\right) }\left( \mathcal{P}\right) =\lim_{n\rightarrow
\infty}\lambda^{\left( p\right) }\left( \mathcal{P},n\right) n^{\left(
2/p\right) -2}
\end{equation*}
exists.
\end{theorem}
The main goal of this note is to find $\lambda^{\left( p\right) }\left(
\mathcal{P}\right) $ for every $\mathcal{P}$ and every $p\geq1.$ It turns
out that $\lambda^{\left( p\right) }\left( \mathcal{P}\right) $ and $%
\pi\left( \mathcal{P}\right) $ are almost identical. Indeed, general results
proved in \cite{NikB} imply that $\lambda^{\left( p\right) }\left( \mathcal{P%
}\right) =\pi\left( \mathcal{P}\right) $ for every $p>1,$ and also $%
\lambda^{\left( 1\right) }\left( \mathcal{P}\right) \geq\pi\left( \mathcal{P}%
\right) .$ In this note we give an alternative direct proof of these
results, and we find $\pi\left( \mathcal{P}\right) $ explicitly.\medskip
Before going further we need some definitions. Recall that a \textbf{%
complete }$r$\textbf{-partite graph} is a graph whose vertices can be split
into $r$ nonempty independent sets so that all edges between vertices of
different classes are present. In particular, an $1$-partite graph is just a
set of isolated vertices.
Further, note that every hereditary property $\mathcal{P}$ can be
represented as $\mathcal{P}=Her\left( \mathcal{F}\right) $ for some family $%
\mathcal{F},$ so hereafter we shall assume that every hereditary property is
given as $\mathcal{P}=Her\left( \mathcal{F}\right) $ for some explicit
family $\mathcal{F}.$
Next, for every family of graphs $\mathcal{F},$ define the parameters $%
\underline{\omega}\left( \mathcal{F}\right) $ and $\beta\left( \mathcal{F}%
\right) $ as
\begin{align*}
\underline{\omega}\left( \mathcal{F}\right) & =\left\{
\begin{array}{l}
0,\text{ if }\mathcal{F}\text{ contains no cliques;} \\
\min\left\{ r:K_{r}\in\mathcal{F}\right\} ,\text{ otherwise.}%
\end{array}
\right. \\
\beta\left( \mathcal{F}\right) & =\left\{
\begin{array}{l}
0,\text{ if }\mathcal{F}\text{ contains no complete partite graphs;} \\
\min\left\{ r:\mathcal{F}\text{ contains a complete }r\text{-partite graph}%
\right\} ,\text{ otherwise.}%
\end{array}
\right.
\end{align*}
The parameters $\underline{\omega}\left( \mathcal{F}\right) $ and $%
\beta\left( \mathcal{F}\right) $ are quite informative about the hereditary
property $Her\left( \mathcal{F}\right) $ as seen in the following
observation.
\begin{proposition}
\label{proinf}If the property $\mathcal{P}$ $=Her\left( \mathcal{F}\right) $
is infinite, then $\underline{\omega}\left( \mathcal{F}\right) =0$ or $%
\underline{\omega}\left( \mathcal{F}\right) \geq2$ and $\beta\left( \mathcal{%
F}\right) \geq2.$
\end{proposition}
\begin{proof}
Suppose that $\underline{\omega}\left( \mathcal{F}\right) \neq0.$ If
$\underline{\omega}\left( \mathcal{F}\right) =1,$ then $\mathcal{P}$ is
empty, so we shall suppose that $\underline{\omega}\left( \mathcal{F}\right)
\geq2.$ This implies that $\beta\left( \mathcal{F}\right) >0,$ as
$\mathcal{F}$ contains $K_{r}$ for some $r\geq2$ and $K_{r}$ is a complete
$r$-partite graph. If $\beta\left( \mathcal{F}\right) =1,$ then
$\mathcal{F}$ contains a graph $F$ consisting of isolated vertices, let say
$s$ be the order of $F.$ Choose a member $G\in\mathcal{P}$ with $v\left(
G\right) \geq r\left( K_{r},K_{s}\right) ,$ where $r\left( K_{r}%
,K_{s}\right) $ is the Ramsey number of $K_{r}$ vs. $K_{s}.$ Thus, $G$
contains either a $K_{r}$ or an independent set on $s$ vertices, both of which
are forbidden. This contradiction shows that $\beta\left( \mathcal{F}\right)
\geq2,$ proving Proposition \ref{proinf}.
\end{proof}
Clearly the study of (\ref{maxed}) and (\ref{maxlam}) makes sense only if $%
\mathcal{P}$ is infinite, and Proposition \ref{proinf} provides a necessary
condition for this feature of $\mathcal{P}$. Now, the following theorem
completely determines $\pi\left( \mathcal{P}\right) .$
\begin{theorem}
\label{thpi}Let $\mathcal{F}$ be a family of graphs. If the property $%
\mathcal{P}=Her\left( \mathcal{F}\right) $ is infinite, then
\begin{equation*}
\pi\left( \mathcal{P}\right) =\left\{
\begin{array}{l}
1,\text{ \ \ \ \ \ \ \ \ \ \ \ \ if }\underline{\omega}\left( \mathcal{F}%
\right) =0; \\
1-\frac{1}{\beta\left( \mathcal{F}\right) -1},\text{ otherwise.}%
\end{array}
\right. .
\end{equation*}
\end{theorem}
Let us turn now to the study of $\lambda^{\left( p\right) }\left( \mathcal{P}%
\right) .$ As mentioned above, in \cite{NikB} it has been proved that $%
\pi\left( \mathcal{P}\right) =\lambda^{\left( p\right) }\left( \mathcal{P}%
\right) $ for $p>1;$ however, for reader's sake we shall establish this
identity directly.
\begin{theorem}
\label{thlam} Let $p>1$ and let $\mathcal{F}$ be a family of graphs. If the
property $\mathcal{P}=Her\left( \mathcal{F}\right) $ is infinite, then
\begin{equation*}
\lambda^{\left( p\right) }\left( \mathcal{P}\right) =\left\{
\begin{array}{l}
1,\text{ \ \ \ \ \ \ \ \ \ \ \ \ if }\underline{\omega}\left( \mathcal{F}%
\right) =0; \\
1-\frac{1}{\beta\left( \mathcal{F}\right) -1},\text{ otherwise.}%
\end{array}
\right. .
\end{equation*}
\end{theorem}
To complete the description of $\lambda^{\left( p\right) }\left( \mathcal{P}%
\right) $ we need to determine the dependence of $\lambda^{\left( 1\right)
}\left( \mathcal{P}\right) $ on $\mathcal{P}$. Using the well-known idea of
Motzkin and Straus \cite{MoSt65}, we come up with the following theorem,
whose easy proof we omit.
\begin{theorem}
\label{thlam1}If $\mathcal{P}$ is an infinite hereditary property, then $%
\lambda^{\left( 1\right) }\left( \mathcal{P}\right) =1-1/r$ if $r$ is the
size of the largest clique in $\mathcal{P},$ and $\lambda^{\left( 1\right)
}\left( \mathcal{P}\right) =1$ if $\mathcal{P}$ contains arbitrary large
cliques.
\end{theorem}
\bigskip
The next section contains proofs of Theorems \ref{thlim}, \ref{thpi} and \ref%
{thlam}, and some auxiliary statements. In the final section we outline a
line of possible future research.
\section{\label{Pf}Proofs}
\subsection{Some preliminary results}
Below we state several results necessary for the proof of our key Theorem %
\ref{thlam}. The first one follows from a result in \cite{NikB}, but for
reader's sake we give its short proof here.
\begin{theorem}
\label{maxlK}Let $p\geq1.$ If $G$ is a graph with $m$ edges and $n$
vertices, with no $K_{r+1},$ then%
\begin{equation}
\lambda^{\left( p\right) }\left( G\right) \leq\left( 1-\frac{1}{r}\right)
^{1/p}\left( 2m\right) ^{1-1/p} \label{in1}
\end{equation}
and
\begin{equation}
\lambda^{\left( p\right) }\left( G\right) \leq\left( 1-\frac{1}{r}\right)
n^{2-2/p}. \label{in2}
\end{equation}
\end{theorem}
\begin{proof}
Indeed, let $\mathbf{x}=\left( x_{1},\ldots,x_{n}\right) $ be a vector such
that $\left\vert x_{1}\right\vert ^{p}+\cdots+\left\vert x_{n}\right\vert
^{p}=1$ and
\[
\lambda^{\left( p\right) }\left( G\right) =2\sum_{\left\{ u,v\right\}
\in E\left( G\right) }x_{u}x_{v}.
\]
Applying the Power Mean Inequality, we see that
\begin{align*}
\lambda^{\left( p\right) }\left( G\right) & =2\sum_{\left\{
u,v\right\} \in E\left( G\right) }x_{u}x_{v}\leq2\sum_{\left\{
u,v\right\} \in E\left( G\right) }\left\vert x_{u}\right\vert \left\vert
x_{v}\right\vert \\
& \leq\left( 2m\right) ^{1-1/p}\left( 2\sum_{\left\{ u,v\right\} \in
E\left( G\right) }\left\vert x_{u}\right\vert ^{p}\left\vert x_{v}%
\right\vert ^{p}\right) ^{1/p}.
\end{align*}
Now, the result of Motzkin and Straus \cite{MoSt65} impies that
\[
2\sum_{\left\{ u,v\right\} \in E\left( G\right) }\left\vert x_{u}%
\right\vert ^{p}\left\vert x_{v}\right\vert ^{p}\leq1-\frac{1}{r},
\]
and inequality (\ref{in1}) follows. Finally, inequality (\ref{in2}) follows
from (\ref{in1}) by Tur\'{a}n's inequality $2m\leq\left( 1-1/r\right)
n^{2}.$
\end{proof}
Note, in particular, that $\lambda^{\left( p\right) }\left( G\right)
\leq\left( 2m\right) ^{1-1/p}.$ This simple bound will be used in the proof
of the following proposition.
\begin{proposition}
\label{pro10}Let $p\geq1,$ $k>1$ and $G_{1}$ and $G_{2}$ be graphs on the
same vertex set. If $G_{1}$ and $G_{2}$ differ in at most $k$ edges, then%
\begin{equation*}
\left\vert \lambda^{\left( p\right) }\left( G_{1}\right) -\lambda^{\left(
p\right) }\left( G_{2}\right) \right\vert \leq\left( 2k\right) ^{1-1/p}.
\end{equation*}
\end{proposition}
\begin{proof}
Let $V=V\left( G_{1}\right) =V\left( G_{2}\right) $ and write $G_{12}$ for
the graph with $V\left( G_{12}\right) =V$ and $E\left( G_{12}\right)
=E\left( G_{1}\right) \cap E\left( G_{2}\right) .$ We may and shall assume
that $\lambda^{\left( p\right) }\left( G_{1}\right) \geq\lambda^{\left(
p\right) }\left( G_{2}\right) .$ Write $G_{3}$ for the graph with $V\left(
G_{3}\right) =V$ and $E\left( G_{3}\right) =E\left( G_{1}\right)
\backslash E\left( G_{2}\right) .$ In view of $G_{12}\subset G_{2},$ we have%
\begin{align*}
0 & \leq\lambda^{\left( p\right) }\left( G_{1}\right) -\lambda^{\left(
p\right) }\left( G_{2}\right) =\lambda^{\left( p\right) }\left(
G_{1}\right) -\lambda^{\left( p\right) }\left( G_{12}\right) -\left(
\lambda^{\left( p\right) }\left( G_{2}\right) -\lambda^{\left( p\right)
}\left( G_{12}\right) \right) \\
& \leq\lambda^{\left( p\right) }\left( G_{1}\right) -\lambda^{\left(
p\right) }\left( G_{12}\right) \leq\lambda^{\left( p\right) }\left(
G_{3}\right) \leq\left( 2e\left( G_{3}\right) \right) ^{1-1/p}\\
& \leq\left( 2k\right) ^{1-1/p},
\end{align*}
proving Proposition \ref{pro10}.
\end{proof}
Further, let us recall the following particular version of the Removal
Lemma, which is a consequence of the Szemer\'{e}di Regularity Lemma (\cite%
{Sze76}, \cite{Bol98}):\medskip
\textbf{Removal Lemma} \emph{For all }$r\geq3$\emph{\ and }$\varepsilon >0,$%
\emph{\ there exists }$\delta=\delta\left( r,\varepsilon\right) >0$\emph{\
such that if }$G$\emph{\ is a graph of order }$n,$\emph{\ with }$k_{r}\left(
G\right) <\delta n^{r},$\emph{\ then there is a graph }$G_{0}\subset G$\emph{%
\ such that }$e\left( G_{0}\right) \geq e\left( G\right) -\varepsilon n^{2}$%
\emph{\ and }$k_{r}\left( G_{0}\right) =0.\medskip$
Finally, we shall need the following theorem proved in \cite{Nik08}:\medskip
\textbf{Theorem A }\emph{For all }$r\geq2$\emph{\ and }$\varepsilon>0,$\emph{%
\ there exists }$\delta=\delta\left( r,\varepsilon\right) >0$\emph{\ such
that if }$G$\emph{\ a graph of order }$n$\emph{\ with }$k_{r}\left( G\right)
>\varepsilon n^{r},$\emph{\ then }$G$\emph{\ contains a }$K_{r}\left(
s\right) $\emph{\ with }$s=\left\lfloor \delta\log n\right\rfloor .$
\subsection{Proof of Theorem \protect\ref{thlim}}
\begin{proof}
Set for short $\lambda_{n}^{\left( p\right) }=\lambda^{\left( p\right)
}\left( \mathcal{P},n\right) .$ Let $G\in\mathcal{P}_{n}$ be such that
$\lambda_{n}^{\left( p\right) }=\lambda^{\left( p\right) }\left(
G\right) $ and let $\mathbf{x}=\left( x_{1},\ldots,x_{n}\right) $ be a
vector with $\left\vert x_{1}\right\vert ^{p}+\cdots+\left\vert x_{n}%
\right\vert ^{p}=1$ and
\[
\lambda_{n}^{\left( p\right) }=\lambda^{\left( p\right) }\left( G\right)
=2\sum_{\left\{ u,v\right\} \in E\left( G\right) }x_{u}x_{v}.
\]
If $p=1,$ we obviously have $\lambda_{n}^{\left( 1\right) }\geq\lambda
_{n-1}^{\left( 1\right) }$ and in view of
\[
\lambda_{n}^{\left( 1\right) }=2\sum_{\left\{ u,v\right\} \in E\left(
G\right) }x_{u}x_{v}\leq2\sum_{1\leq i1.$ Since $\left\vert x_{1}\right\vert ^{p}+\cdots
+\left\vert x_{n}\right\vert ^{p}=1,$ there is a vertex $k\in V\left(
G\right) $ such that $\left\vert x_{k}\right\vert ^{p}\leq1/n.$ Write $G-k$
for the graph obtained from $G$ by omitting the vertex $k,$ and let
$\mathbf{x}^{\prime}=\left( x_{1}^{\prime},\ldots,x_{n-1}^{\prime}\right) $
be the $\left( n-1\right) $-vector obtained from $\mathbf{x}$ by omitting
the entry $x_{k}.$ The eigenequation for $\lambda^{\left( p\right) }\left(
G\right) $ and the vertex $k$ is%
\[
\lambda^{\left( p\right) }\left( G\right) x_{k}\left\vert x_{k}\right\vert
^{p-2}=\sum_{\left\{ k,i\right\} \in E\left( G\right) }x_{i}.
\]
Hence, we see that%
\begin{align*}
2\sum_{\left\{ u,v\right\} \in E\left( G-k\right) }x_{u}^{\prime}%
x_{v}^{\prime} & =2\sum_{\left\{ u,v\right\} \in E\left( G\right) }%
x_{u}x_{v}-2x_{k}\sum_{\left\{ k,i\right\} \in E\left( G\right) }x_{i}\\
& =\lambda^{\left( p\right) }\left( G\right) -2x_{k}\left(
\lambda^{\left( p\right) }\left( G\right) x_{k}\left\vert x_{k}\right\vert
^{p-2}\right) =\lambda_{n}^{\left( p\right) }\left( 1-2\left\vert
x_{k}\right\vert ^{p}\right) .
\end{align*}
Since $\mathcal{P}$ is a hereditary property, $G-k\in\mathcal{P}_{n-1},$ and
therefore,%
\[
2\sum_{\left\{ u,v\right\} \in E\left( G-k\right) }x_{u}^{\prime}%
x_{v}^{\prime}\leq\lambda^{\left( p\right) }\left( G-k\right) \left\vert
\mathbf{x}^{\prime}\right\vert _{p}^{2}=\lambda^{\left( p\right) }\left(
G-k\right) \left( 1-\left\vert x_{k}\right\vert ^{p}\right) ^{2/p}%
\leq\lambda_{n-1}^{\left( p\right) }\left( 1-\left\vert x_{k}\right\vert
^{p}\right) ^{2/p}.
\]
Thus, we obtain%
\begin{equation}
\lambda_{n}^{\left( p\right) }\leq\lambda_{n-1}^{\left( p\right) }%
\frac{\left( 1-\left\vert x_{k}\right\vert ^{p}\right) ^{2/p}}{\left(
1-2\left\vert x_{k}\right\vert ^{p}\right) }. \label{in3}%
\end{equation}
Note that the function
\[
f\left( y\right) =\frac{\left( 1-y\right) ^{2/p}}{1-2y}%
\]
is nondecreasing in $y$ for $0\leq y\leq1/n$ and $n$ sufficiently large.
Indeed,%
\begin{align*}
\frac{df\left( y\right) }{dy} & =\frac{-\frac{2}{p}\left( 1-y\right)
^{2/p-1}\left( 1-2y\right) +2\left( 1-y\right) ^{2/p}}{\left(
1-2y\right) ^{2}}\\
& =\left( -\frac{1}{p}\left( 1-2y\right) +\left( 1-y\right) \right)
\frac{2\left( 1-y\right) ^{2/p-1}}{\left( 1-2y\right) ^{2}}\\
& =\left( -\left( \frac{1}{p}-1\right) +\left( \frac{2}{p}-1\right)
y\right) \frac{2\left( 1-y\right) ^{2/p-1}}{\left( 1-2y\right) ^{2}}\geq0
\end{align*}
Here we use the fact that $1/p-1>0$ and that $\left( 2/p-1\right) y$ tends
to $0$ when $n$ $\rightarrow\infty.$
Hence, in view of (\ref{in3}), we find that for $n$ large enough
\[
\lambda_{n}^{\left( p\right) }\leq\lambda_{n-1}^{\left( p\right) }f\left(
\left\vert x_{k}\right\vert ^{p}\right) \leq\lambda_{n-1}^{\left( p\right)
}f\left( \frac{1}{n}\right) =\lambda_{n-1}^{\left( p\right) }%
\frac{n\left( 1-1/n\right) ^{2/p}}{\left( n-2\right) },
\]
and so
\[
\frac{\lambda_{n}^{\left( p\right) }n^{2/p}}{n\left( n-1\right) }\leq
\frac{\lambda_{n-1}^{\left( p\right) }\left( n-1\right) ^{2/p}}{\left(
n-1\right) \left( n-2\right) }.
\]
Therefore, the sequence%
\[
\left\{ \frac{\lambda_{n}^{\left( p\right) }n^{2/p}}{n\left( n-1\right)
}\right\} _{n=1}^{\infty}%
\]
is nonincreasing, and so it is converging, completing the proof of Theorem
\ref{thlim}.
\end{proof}
\subsection{Proof of Theorem \protect\ref{thpi}}
\begin{proof}
Since $\mathcal{P}$ is infinite, Proposition \ref{proinf} implies that
$\underline{\omega}\left( \mathcal{F}\right) =0$ or $\underline{\omega
}\left( \mathcal{F}\right) \geq2$ and $\beta\left( \mathcal{F}\right)
\geq2.$ If $\underline{\omega}\left( \mathcal{F}\right) =0,$ then $K_{n}%
\in\mathcal{P}_{n},$ because all induced subgraphs of $K_{n}$ are complete and
therefore do not belong to $\mathcal{F}$. Hence,
\[
ex\left( \mathcal{P},n\right) =\binom{n}{2};
\]
and so, $\pi\left( \mathcal{P}\right) =1.$
Now assume that $r=\underline{\omega}\left( \mathcal{F}\right) \geq2$ and
$\beta=\beta\left( \mathcal{F}\right) \geq2.$ We shall prove that
$T_{\beta-1}\left( n\right) \in\mathcal{P}_{n},$ where $T_{\beta-1}\left(
n\right) $ is the complete $\left( \beta-1\right) $-partite Tur\'{a}n graph
of order $n.$ Indeed all induced subgraphs of $T_{\beta-1}\left( n\right) $
are complete $r$-partite graphs for some $r\leq\beta-1$, so should one of them
belong to $\mathcal{F},$ we would have $\beta\left( \mathcal{F}\right)
\leq\beta-1=\beta\left( \mathcal{F}\right) -1,$ which is a contradiction.
Therefore,
\[
ex\left( \mathcal{P},n\right) \geq e\left( T_{\beta-1}\left( n\right)
\right) =\left( 1-\frac{1}{\beta-1}+o\left( 1\right) \right) \binom{n}%
{2},
\]
and so%
\[
\pi\left( \mathcal{P}\right) \geq1-\frac{1}{\beta\left( \mathcal{F}\right)
-1}.
\]
To finish the proof we shall prove the opposite inequality. Let $F\in
\mathcal{F}$ be a complete $\beta$-partite graph, which exists by the
definition of $\beta\left( \mathcal{F}\right) ;$ let $s$ be the maximum of
the sizes of its vertex classes. Let $\varepsilon>0$ and set $t=r\left(
K_{r},K_{s}\right) ,$ where $r\left( K_{r},K_{s}\right) $ is the Ramsey
number of $K_{r}$ vs. $K_{s}.$ If $n$ is large enough and $G\in\mathcal{P}%
_{n}$ satisfies
\[
e\left( G\right) >\left( 1-\frac{1}{\beta\left( \mathcal{F}\right)
-1}+\varepsilon\right) \binom{n}{2},
\]
then, by the theorem of Erd\H{o}s and Stone \cite{ErSt46}, $G$ contains a
subgraph $G_{0}=K_{\beta}\left( t\right) ,$ that is to say, a complete
$\beta$-partite graph with $t$ vertices in each vertex class. Since $K_{r}%
\in\mathcal{F}$, we see that $G_{0}$ contains no $K_{r}.$ Hence each vertex
class of $G_{0}$ contains an independent set of size $s,$ and so $G$ contains
an induced subgraph $K_{\beta}\left( s\right) ,$ which in turn contains an
induced copy of $F.$ Thus, if $n$ is large enough and $G\in\mathcal{P}_{n},$
then
\[
e\left( G\right) \binom{n}{2}^{-1}\leq1-\frac{1}{\beta\left( \mathcal{F}%
\right) -1}+\varepsilon.
\]
This inequality implies that
\[
\pi\left( \mathcal{P}\right) \leq1-\frac{1}{\beta\left( \mathcal{F}\right)
-1},
\]
completing the proof of Theorem \ref{thpi}.
\end{proof}
\subsection{Proof of Theorem \protect\ref{thlam}}
\begin{proof}
First note the inequality
\[
\lambda^{\left( p\right) }\left( G\right) \geq2e\left( G\right)
/n^{2/p},
\]
which follows by taking $\left( x_{1},\ldots,x_{n}\right) =\left(
n^{-1/p},\ldots,n^{-1/p}\right) $ in (\ref{maxlam}). So we see that
\[
\lambda^{\left( p\right) }\left( \mathcal{P}\right) \geq\pi\left(
\mathcal{P}\right) ,
\]
and this inequality together with Theorem \ref{thpi} gives $\lambda^{\left(
p\right) }\left( \mathcal{P}\right) =1$ if $\underline{\omega}\left(
\mathcal{F}\right) =0,$ and
\[
\lambda^{\left( p\right) }\left( \mathcal{P}\right) \geq1-\frac{1}%
{\beta\left( \mathcal{F}\right) -1}%
\]
otherwise. To finish the proof we shall show that%
\[
\lambda^{\left( p\right) }\left( \mathcal{P}\right) \leq1-\frac{1}%
{\beta\left( \mathcal{F}\right) -1}.
\]
For this purpose write $k_{r}\left( G\right) $ for the number of $r$-cliques
of $G$. Let $F\in\mathcal{F}$ be a complete $\beta$-partite graph, which
exists by the definition of $\beta\left( \mathcal{F}\right) ;$ let $s$ be
the maximum of the sizes of its vertex classes.
Now if $\varepsilon>0$ and $\varepsilon$ is sufficiently small, choose
$\delta=\delta\left( \beta,\varepsilon\right) $ as in the Removal Lemma, and
set $t=r\left( K_{r},K_{s}\right) ,$ where $r\left( K_{r},K_{s}\right) $
is the Ramsey number of $K_{r}$ vs. $K_{s}.$ If $G\in\mathcal{P}_{n},$ then
$K_{\beta}\left( t\right) \nsubseteq G$ for otherwise, as in proof of
Theorem \ref{thpi}, we see that $G$ contains an induced copy of $F.$ So if $n$
is large enough Theorem A implies that $k_{\beta}\left( G\right) \leq\delta
n^{r}.$ Now, by the Removal Lemma, there is a graph $G_{0}\subset G$ such that
$e\left( G_{0}\right) \geq e\left( G\right) -\varepsilon n^{2}$ and
$k_{\beta}\left( G_{0}\right) =0.$
For $n$ sufficiently large Propositions \ref{maxlK} and \ref{pro10} imply
that
\[
\lambda^{\left( p\right) }\left( G\right) \leq\lambda^{\left( p\right)
}\left( G_{0}\right) +\left( 2\varepsilon n\right) ^{2-2/p}\leq\left(
1-\frac{1}{\beta-1}\right) n^{2-2/p}+\left( 2\varepsilon n\right) ^{2-2/p},
\]
and hence,
\[
\lambda^{\left( p\right) }\left( \mathcal{P},n\right) n^{2/p-2}\leq
1-\frac{1}{\beta-1}+\left( 2\varepsilon\right) ^{2-2/p}.
\]
Since $\varepsilon$ can be chosen arbitrarily small, we see that
\[
\lambda^{\left( p\right) }\left( \mathcal{P}\right) \leq1-\frac{1}%
{\beta-1},
\]
completing the proof of Theorem \ref{thlam}.
\end{proof}
\section{Concluding remarks}
In a sequence of papers the author has shown that many classical extremal
results like the Erd\H{o}s-Stone-Bollob\'{a}s theorem \cite{BoEr73}, the
Stability Theorem of Erd\H{o}s \cite{Erd66,Erd68} and Simonovits \cite{Sim68}%
, and various saturation problems can be strengthened by recasting them for
the largest eigenvalue instead of the number of edges; see \cite{Nik11} for
an overview and references.
The paper \cite{KLM13} and the present note show that some of these edge
extremal results can be extended further to $\lambda^{\left( p\right)
}\left( G\right) $ for any $p>1.$ A natural challenge here is to reprove all
of the above problems by substituting $\lambda^{\left( p\right) }\left(
G\right) $ for the number of edges.
Theorems \ref{thlim}, \ref{maxlK}, and Proposition \ref{pro10} have been
proved in \cite{NikB} for uniform hypergraphs. The streamlined proofs given
here are for reader's convenience.\bigskip
\subsection*{Acknowledgement}
Thanks are due to Bela Bollob\'{a}s and to Alex Sidorenko for useful
discussions.
\bigskip
\begin{thebibliography}{99}
\bibitem{Bol98} B. Bollob\'{a}s. \emph{Modern Graph Theory.} Graduate Texts
in Mathematics, 184, Springer-Verlag, New York (1998), xiv+394 pp.
\bibitem{BoEr73} B. Bollob\'{a}s and P. Erd\H{o}s. On the structure of edge
graphs. \emph{J. London Math. Soc.,} 5:317-321, 1973.
\bibitem{Erd66} P. Erd\H{o}s. Some recent results on extremal problems in
graph theory (Results). In \emph{Theory of Graphs (Internat. Sympos., Rome,
1966)}, pages 117--130. Gordon and Breach, New York; Dunod, Paris, 1967.
\bibitem{Erd68} P. Erd\H{o}s. On some new inequalities concerning extremal
properties of graphs. In \emph{Theory of Graphs (Proc. Colloq., Tihany,
1966),} pages 77--81. Academic Press, New York, 1968.
\bibitem{ErSt46} P. Erd\H{o}s, A.H. Stone. On the structure of linear
graphs. \emph{Bull. Amer. Math. Soc.,} 52:1087--1091, 1946.
\bibitem{ErSi66} P. Erd\H{o}s and M. Simonovits. A limit theorem in graph
theory. \emph{Studia Sci. Math. Hung.} 1:51-57, 1966.
\bibitem{KLM13} P. Keevash, J. Lenz, and D. Mubayi. Spectral extremal
problems for hypergraphs. \arxiv {1304.0050}.
\bibitem{KNS64} G. Katona, T. Nemetz, and M. Simonovits. On a problem of Tur%
\'{a}n in the theory of graphs. \emph{Mat. Lapok } 15:228--238, 1964.
\bibitem{MoSt65} T. Motzkin and E. Straus. Maxima for graphs and a new proof
of a theorem of Tur\'{a}n. \emph{Canad. J. Math.,} 17:533-540, 1965.
\bibitem{Nik08} V. Nikiforov. Graphs with many $r$-cliques have large
complete $r$-partite subgraphs. \emph{Bull. London Math. Soc.,} 40:23-25,
2008.
\bibitem{Nik11} V. Nikiforov. Some new results in extremal graph theory. In
\emph{Surveys in Combinatorics, }pages\emph{\ }141--181, Cambridge
University Press, 2011.
\bibitem{NikB} V. Nikiforov. An analytic theory of extremal hypergraph
problems. \arxiv {1305.1073v2}.
\bibitem{NikC} V. Nikiforov. Analytic methods for uniform hypergraphs.
\arxiv {1308.1654v3}.
\bibitem{Sim68} M. Simonovits. A method for solving extremal problems in
graph theory, stability problems. In \emph{Theory of Graphs (Proc. Colloq.,
Tihany, 1966),} pages 279--319, Academic Press, New York, 1968.
\bibitem{Sze76} E. Szemer\'{e}di. Regular partitions of graphs. In\emph{\
Colloques Internationaux C.N.R.S.} \emph{No 260 - Probl\`{e}mes
Combinatoires et Th\'{e}orie des Graphes,} pages 399-401, Orsay, 1976.
\bibitem{Tur41} P. Tur\'{a}n. On an extremal problem in graph theory (in
Hungarian). $\emph{Mat.}$ \emph{\'{e}s Fiz. Lapok,} 48:436-452, 1941.
\end{thebibliography}
\end{document}