\documentclass[reqno]{amsart} \usepackage{hyperref} \AtBeginDocument{{\noindent\small {\em Electronic Journal of Differential Equations}, Vol. 2007(2007), No. 35, pp. 1--7.\newline ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu \newline ftp ejde.math.txstate.edu (login: ftp)} \thanks{\copyright 2007 Texas State University - San Marcos.} \vspace{9mm}} \begin{document} \title[\hfilneg EJDE-2007/35\hfil Eigencurves of the $p$-Laplacian] {Eigencurves of the $p$-Laplacian with weights and their asymptotic behavior} \author[A. Dakkak, M. Hadda\hfil EJDE-2007/35\hfilneg] {Ahmed Dakkak, Mohammed Hadda} % in alphabetical order \address{Ahmed Dakkak \newline Department of Mathematics, University Moulay Ismail, Faculty of Sciences and Techniques, B. P. 509 - Boutalamine, Errachidia, Morocco} \email{dakkakahmed@hotmail.com} \address{Mohammed Hadda \newline Department of Mathematics, University Moulay Ismail, Faculty of Sciences and Techniques, B. P. 509 - Boutalamine, Errachidia, Morocco} \email{mohammedhad@hotmail.com} \thanks{Submitted July 21, 2006. Published February 27, 2007.} \subjclass[2000]{35J20, 35J70, 35P05, 35P30} \keywords{Nonlinear eigenvalue problem; eigencurves; $p$-Laplacian; \hfil\break\indent indefinite weight; asymptotic behavior} \begin{abstract} In this paper we study the existence of the eigencurves of the $p$-Laplacian with indefinite weights. We obtain also their variational formulations and asymptotic behavior. \end{abstract} \maketitle \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \section{Introduction and preliminaries} We consider the nonlinear eigenvalue problem \begin{equation} \label{VP} \begin{gathered} -\Delta _pu = \lambda m(x)|u|^{p-2}u \quad \text{in }\Omega \\ u = 0 \quad \text{on }\partial \Omega \,, \end{gathered} \end{equation} where $\Omega $ is a smooth bounded domain in $\mathbb{R}^N$, $-\Delta _pu=-\mathop{\rm div}(|\nabla u|^{p-2}\nabla u)$ is the $p $-Laplacian, $1
0\}\neq 0 \,.
$$
We denote
\[
\mathcal{M}^{+}(\Omega )=\big\{m\in \mathcal{L}^\infty (\Omega ):
\mathop{\rm meas}\{x\in \Omega : m(x)>0\}\neq 0\big\}\,.
\]
We say that $\lambda $ is a eigenvalue of $p$-Laplacian with
weight $m$, when the problem \eqref{VP} has at least a nontrivial
solution $u$ in $\mathcal{W}_0^{1,p}(\Omega )$. The set of
positive eigenvalues constitutes the spectrum $\sigma ^{+}(-\Delta
_p,m,\Omega )$ of $p$-Laplacian with weight $m$ in the domain
$\Omega $. This spectrum contains an infinite sequence given by
$\lambda _1<\lambda _2\leq \dots \leq \lambda _n\to +\infty $ and
formulated as follows
\begin{equation}\label{1.1}
\frac 1{\lambda _n}=\frac 1{\lambda _n(m)}=\sup_{K\in \Gamma
_n}\min_{u\in K}\int_\Omega m|u|^p \,,
\end{equation}
where $\Gamma _n$ is defined by
\[
\Gamma _n=\{K\subset S: K \mbox{ is symetric, compact and }
\gamma (K)\geq n\} ,
\]
$S=\{u\in \mathcal{W}_0^{1,p}(\Omega ):\int_\Omega |\nabla
u|^p=1\}$ is the sphere unity of $\mathcal{W}_0^{1,p}(\Omega )$
and $\gamma $ is the genus function. We may also define the
negative spectrum when $-m\in \mathcal{M}^{+}(\Omega )$ by
$-\sigma ^{+}(-\Delta _p,-m,\Omega )$ which contains an infinite
sequence $\lambda _{-1}>\lambda _{-2}\geq \dots \geq \lambda
_{-n}\to -\infty $ such that
\begin{equation}\label{1.2}
\lambda _{-n}=\lambda _{-n}(m):=-\lambda _n(-m) \quad.
\end{equation}
The variational characterization $(\ref{1.1})$ and the properties
of $\lambda _n$ depending on weight $m$ was the subject of several
works of which we cite for example \cite{A1,A2,AT,T}.
In this note, we study the following problem: Find all the real
numbers $\alpha $, $\beta $ such that $\lambda_n(\alpha m_1+\beta
m_2)=1$.
This last equation comes from the problem of eigencurves
of Sturm-Liouville. Several applications of these problems can be
found in the bifurcation domain and other, making reference
\cite{BH2}. In \cite{BH1} we find the properties related to the
first eigencurve such as concavity, differentiability and the
asymptotic behavior. The authors wished to have information about
the other eigencurves, especially their asymptotic behavior. This
will be the object of our study. Let $m_1, m_2 \in
\mathcal{M}^{+}(\Omega )$ so that $\mathop{\rm ess\,inf}_{\Omega}
m_2>0$.
we define the graph of the $n^{th}$ eigencurve by
\begin{equation}\label{1.3}
C_n=\{(\alpha ,\beta )\in \mathbb{R}^{2}:\lambda _n(\alpha
m_1+\beta m_2)=1\} \,,
\end{equation}
We note that this definition differs from that given in
\cite{BH2}, which is
\begin{equation}\label{1.4}
\beta _n(\alpha )=\inf_{K\in \Gamma _n}\max_{u\in K}\frac{
\int_\Omega |\nabla u|^p-\alpha \int_\Omega m_1|u|^p}{\int_\Omega
m_2|u|^p} \quad.
\end{equation}
This paper is organized as follows. First, we are interested to
the existence of eigencurve $C_n$ . Then we show that
$(\alpha,\beta _n(\alpha ))\in $ $C_n$. This would allow us to
affirm the coincidence of the two definitions \eqref{1.3} et
\eqref{1.4} and also present the variational formulation of that
eigencurve. We would end up with the study of the asymptotic
behavior of the eigencurves $C_n$. And finally we affirm that all
eigencurves have the same asymptotic behavior.
\section{Existence of the eigencurve $C_n$}
We first recall the following
\begin{proposition}\label{prop2.1}
\begin{enumerate}
\item Let $m, m' \in \mathcal{M}^{+}(\Omega )$.
If $m\leq m'$ (resp. $m< m'$), then $\lambda _n(m)\geq \lambda
_n(m')$ (resp. $\lambda _n(m)> \lambda _n(m')$).
\item $\lambda _n : m\mapsto \lambda _n(m)$ is continuous in
$(\mathcal{M}^{+}(\Omega ),\| .\| _\infty )$.
\end{enumerate}
\end{proposition}
For the proof, see for example \cite{T}. \\
Next we can establish the following
\begin{proposition}\label{prop2.2}
Let $(m_k)_k$ be a sequence in $\mathcal{ M}^{+}(\Omega )$
such that $m_k\to m$ in $\mathcal{L}^\infty (\Omega )$. Then
$\lim_k \lambda _n(m_k)=+\infty$ if and only if $m\leq 0$ almost
everywhere in $\Omega$.
\end{proposition}
\begin{proof}
Let $(m_k)_k$ be a sequence in $\mathcal{ M}^{+}(\Omega )$ such
that $m_k$ converges to $m$ in $\mathcal{L}^\infty (\Omega )$.
Assume first that $\lim_k \lambda_n(m_k)=+\infty$, we claim that
$m\leq 0$ almost everywhere in $\Omega$;
otherwise
$$
\mathop{\rm meas}\{x\in \Omega : m(x)>0\}\neq 0
$$
it then follows that $\lim_k \lambda _n(m_k)=\lambda_n(m)$, is a
finite, a contradiction.
Inversely, if $m\leq 0$ almost everywhere in $\Omega $, suppose
that $\lim_k \lambda _n(m_k)$ is finite, then there exist
$\lambda >0$ such that
\begin{equation}\label{2.1}
\lambda _n(m_k)\leq \lambda \quad \mbox{for all } k \in \mathbb{N}
.
\end{equation}
We put $r=2\lambda/\lambda_n(2)$ and $\varepsilon=1/r$. Then there
exist $k=k(r)$, such that
$$
\|m_k-m\|_\infty <\varepsilon \,.
$$
We consider the following weights
$$
m_{k,r}(x)=\begin{cases}
m_k(x) & \text{if } x \in \Omega\setminus B_r\cap \Omega_k^- \\
\frac{1}{r} & \text{if } x \in B_r\cap \Omega_k^-
\end{cases}
$$
and
$$
m_r(x)=\begin{cases}
m(x) & \text{if } x \in \Omega\setminus B_r\cap \Omega_k^- \\
\frac{1}{r} & \text{if } x \in B_r\cap \Omega_k^-
\end{cases}
$$
where $B_r=B(x_k,\frac{1}{r})$ is a ball and $x_k \in \Omega
_k^-=\{x\in \Omega : m_k(x)<0\}$. It is clear that
$$
\|m_{k,r}-m_r\|_\infty \leq \|m_k-m\|_\infty
$$
so that
$$
m_{k,r}\leq m_r+\varepsilon \quad \mbox{almost everywhere in }
\Omega\,.
$$
Observe that $m_k\leq m_{k,r}$. Since $m\leq 0$ almost everywhere
in $\Omega$, we have $ m_r\leq 1/r$ almost everywhere in
$\Omega$, and
\begin{equation}\label{2.2}
m_k\leq\frac{1}{r}+\varepsilon \quad \mbox{almost everywhere in }
\Omega \,.
\end{equation}
It follows from \eqref{2.1} and \eqref{2.2} that
$$
\lambda\geq \lambda_n(m_k)\geq
\lambda_n(\frac{1}{r}+\varepsilon)\,.
$$
Since $\frac{1}{r}+\varepsilon=\frac{2}{r}$,
$$
2\lambda=r\lambda_n(2)=\lambda_n(\frac{1}{r}+\varepsilon)\leq
\lambda_n(m_k)\leq \lambda \;,
$$
which is a contradiction. So ${\lim_k } \lambda_n(m_k)=+\infty$.
\end{proof}
Now we can state the main theorem of this section.
\begin{theorem}\label{thm2.3}
Let $m_1, \, m_2\in \mathcal{M}^{+}(\Omega )$ be such that
$\mathop{\rm ess\,inf}_{\,\Omega} m_2>0$. So for all
$\alpha \in \mathbb{R}$ there exist a unique real $t_n(\alpha )$
which satisfies $\lambda _n(\alpha m_1+t_n(\alpha )m_2)=1$.
\end{theorem}
\begin{proof}
Let $\alpha \in \mathbb{R}$. We consider the function
$f_{\alpha}:t\mapsto \lambda _n(\alpha m_1+tm_2)$. According to
the proposition \ref{prop2.1} we affirm that $f_\alpha $ is
decreasing continuous. Consequently $f_\alpha $ is injective. To
show that the equation $f_{\alpha} (t)=1$ has a solution (hence
only one ), we distinguish three cases.
\noindent\textbf{Case 1:} $ \lambda _{-n}(m_1)<\alpha <\lambda
_n(m_1)$ It is clear that when $\alpha =0$, the unique real $t_0$
that verifies $\lambda _n(\alpha m_1+t_0m_2)=1$ is $t_0=\lambda
_n(m_2)$. Suppose that $\alpha $ is not nil, in this case we have
$$
f_\alpha (0)=\frac{\lambda _n(m_1)}{\alpha} \quad \mbox{if }
\alpha >0 \quad \mbox{and} \quad f_\alpha (0)=\frac{\lambda
_{-n}(m_1)}{\alpha} \quad \mbox{if } \alpha <0 \,;
$$
So that $f_{\alpha}(0)>1$. Now $\frac {\alpha} {t} m_1+m_2 \to
m_2$ in $\mathcal{L}^\infty (\Omega )$ as $t \to +\infty $;
$$
\lim_{t\to +\infty } f_{\alpha} (t)=\lim_{t\to +\infty }\frac
{1}{t} \lambda _n(\frac {\alpha} {t} m_1+m_2)=0 \,,
$$
so there exist a unique real $t_n(\alpha )\in ]0,+\infty [$ which
verifies $f_\alpha (t_n(\alpha ))=1 $.
\noindent\textbf{Case 2:} $\alpha >\lambda _n(m_1)$. In this case
$\alpha >0$ and
\begin{equation}\label{2.3}
0