\documentclass[reqno]{amsart} \usepackage{hyperref} \AtBeginDocument{{\noindent\small 2005-Oujda International Conference on Nonlinear Analysis. \newline {\em Electronic Journal of Differential Equations}, Conference 14, 2006, pp. 231--240.\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 2006 Texas State University - San Marcos.} \vspace{9mm}} \setcounter{page}{231} \begin{document} \title[\hfilneg EJDE/Conf/14 \hfil Existence and uniqueness of positive solutions] {Existence and uniqueness of a positive solution for a non homogeneous problem of fourth order with weights} \author[M. Talbi, N. Tsouli \hfil EJDE/Conf/14 \hfilneg] {Mohamed Talbi, Najib Tsouli} % in alphabetical order \address{Mohamed Talbi \newline D\'epartement de Math\'ematiques et Informatique Facult\'e des Sciences, Universit\'e Mohamed 1, Oujda, Maroc} \email{talbi\_md@yahoo.fr} \address{Najib Tsouli \newline D\'epartement de Math\'ematiques et Informatique Facult\'e des Sciences, Universit\'e Mohamed 1, Oujda, Maroc} \email{tsouli@sciences.univ-oujda.ac.ma} \date{} \thanks{Published September 20, 2006.} \subjclass[2000]{35J60, 35J30, 35J65} \keywords{Ekeland's principle; p-biharmonic operator; Palais-Smale condition} \begin{abstract} In this work we study the existence of a positive solutions to the non homogeneous equation $$ \Delta( |\Delta u|^{p-2} \Delta u) = m |u|^{q-2}u $$ with Navier boundary conditions, where $1
q$ and $m\in \mathcal{C}(\overline{\Omega})$,
we prove the uniqueness of this solution.
\end{abstract}
\maketitle
\numberwithin{equation}{section}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{definition}{Definition}
\section{Introduction}
We consider the following problem with Navier boundary conditions
\begin{equation}\label{P1}
\begin{gathered}
\Delta^2_p u = m |u|^{q-2}u\quad \mbox{in } \Omega,\\
u>0\quad \mbox{in } \Omega,\\
u = \Delta u = 0 \quad \mbox{on }\partial \Omega.
\end{gathered}
\end{equation}
Here $\Omega$ is a smooth domain in $\mathbb{R}^N$ ($N\geq 1$),
$\Delta^2_p $ is the p-biharmonic operator defined by
$\Delta^2_p u=\Delta (|\Delta u|^{p-2}\Delta u)$,
$m\in L^{\infty}(\Omega)\setminus \{0\} , m\geq 0$ and
$p, q\in ]1,p^*_2[$, $p\neq q$ where
$$
p^*_2= \begin{cases}
\frac{Np}{N-2p} & \mbox{if } pq$ we prove the uniqueness of this
solution. Notice that our approach does not use the fundamental
multiplicity theorem and the mountain-pass lemma. We can refer
the reader to \cite{EL} for the existence of a positive solution
and to \cite{ID-OT} for the
uniqueness.
Similar results as ours, but with p-Laplacian operator, were
studied by authors \cite{ID-OT, DI-SA}.
\section{Preliminaries}
In this paper, we consider the transformation of Poisson problem
used by Dr\'abek and \^Otani \cite{DR-OT}.
We recall some properties of the Dirichlet problem for the Poisson
equation
\begin{equation}\label{pbf}
\begin{gathered}
-\Delta u = f \quad\mbox{in } \Omega, \\
u = 0 \quad\mbox{on } \partial \Omega.
\end{gathered}
\end{equation}
It is well known that \eqref{pbf} is uniquely solvable in
$W^{2,p}(\Omega)\cap W^{1,p}_0(\Omega)$ for all
$f\in L^p(\Omega)$ and for any $p\in ]1,+\infty[$.
We denote by:
$X=W^{2,p}(\Omega)\cap W^{1,p}_0(\Omega)$, \\
$\|u\|_p=(\int_{\Omega}|u|^pdx)^{1/p}$ the norm in $L^p(\Omega)$,\\
$ \|u\|_{2,p}=(\|\Delta u\|_p^p+\|u\|_p^p)^{1/p}$ the norm in $X$,\\
$\|u\|_{\infty}$ the norm in $L^{\infty}(\Omega)$, \\
and $\langle\cdot,\cdot\rangle$ is the duality bracket between
$L^p(\Omega)$ and $L^{p'}(\Omega)$, where $p'=p/(p-1)$. Denote by
$\Lambda$ the inverse operator of $-\Delta:X\to L^p(\Omega)$. The
following lemma gives us some properties of the operator $\Lambda$
(c.f. \cite{DR-OT,GI-TR}).
\begin{lemma}\label{lem2.1}
\begin{itemize}
\item[(i)] {\rm (Continuity):} There exists a constant $c_p > 0$ such
that
$$
\| \Lambda f \|_{2,p} \leq c_p \|f\|_p
$$
holds for all $p \in ]1, +\infty[$ and $f \in L^p (\Omega)$.
\item[(ii)] {\rm (Continuity)} Given $k \in \mathbb{N}^*$, there exists
a constant $c_{p,k}> 0$ such that
$$
\| \Lambda f\|_{W^{k+2,p}} \leq c_{p,k} \|f\|_{W^{k,p}}
$$
holds for all $p \in ]1, +\infty[$ and
$f \in W^{k,p} (\Omega)$.
\item[(iii)] {\rm (Symmetry)} The equality
$$
\int_{\Omega} \Lambda u \cdot v dx = \int_{\Omega} u \cdot \Lambda v dx
$$
holds for all $u \in L^p(\Omega)$ and $v \in
L^{p'} (\Omega)$ with $p \in ]1, +\infty[$.
\item[(iv)] {\rm (Regularity)}
Given $f \in L^{\infty} (\Omega)$, we have
$\Lambda f \in C^{1, \alpha} (\bar{\Omega})$ for all $\alpha \in ]0,1[$;
moreover, there exists $c_{\alpha} > 0$ such that
$$
\| \Lambda f\|_{C^{1,\alpha}} \leq c_{\alpha} \|f\|_{\infty}.
$$
\item[(v)] {\rm (Regularity and Hopf-type maximum principle)}
Let $f \in C (\bar{\Omega})$ and $f \geq 0$ then
$w = \Lambda f \in C^{1, \alpha}(\bar{\Omega})$, for all
$\alpha \in ]0,1[$ and $w$ satisfies:
$w> 0$ in $\Omega, \frac{\partial w}{\partial n} < 0$ on
$\partial \Omega$.
\item[(vi)] {\rm (Order preserving property)} Given
$f, g \in L^p (\Omega)$ if $f \leq g$ in $\Omega$, then
$\Lambda f < \Lambda g$ in $\Omega$.
\end{itemize}
\end{lemma}
Note that for all $u\in X$ and all $v \in L^p(\Omega)$,
we have $v=-\Delta u$ if and only if $u=\Lambda v$.
Let us denote $N_p$ the Nemytskii operator defined by
\[
N_p(v)(x) = \begin{cases} |v(x)|^{p-2}v(x) & \mbox{if } v(x)\neq 0\\
0 & \mbox{if } v(x)=0.
\end{cases}
\]
Then for all $v \in L^p(\Omega)$ and all
$w\in L^{p'}(\Omega)$, we have
$N_p(v)=w$ if and only if $v=N_{p'}(w)$.
For $v=-\Delta u$ which means that $u=\Lambda v$. As
$X\hookrightarrow L^q(\Omega)$, then $\Lambda v\in L^q(\Omega) \forall v\in
L^p(\Omega)$. We define the functionals $F,G: L^p(\Omega)\to
\mathbb{R}$ as follows:
$$
F(v)=\frac{1}{p}\| v\|_p^p \quad \text{and}\quad
G(v)=\frac{1}{q}\int_{\Omega}m|\Lambda v|^qdx.
$$
Then it is clear that $F$ and $G$ are well defined on $L^p(\Omega)$,
and are of class $\mathcal{C}^1$ on $L^p(\Omega)$ and for all
$v\in L^p(\Omega)$ we have $F'(v)=N_p(v)$ and
$G'(v)=\Lambda (mN_q(\Lambda v))$ in $L^{p'}(\Omega)$.
The operator $\Lambda$ enables us to transform problem \eqref{P1} to
another problem which we shall study in the space $L^p(\Omega)$.
\begin{definition} \label{def2.1} \rm
We say that $u\in X\setminus \{0\}$ is a solution of
problem \eqref{P1}, if $v=-\Delta u$ is a solution of the
problem: Find $v\in L^p(\Omega)\setminus \{0\}$, $v>0$, such that
\begin{equation}
N_p(v)=\Lambda(mN_q(\Lambda v)) \quad\text{in } L^{p'}(\Omega).\label{P2}
\end{equation}
\end{definition}
\section{Existence of a positive solution}
For solutions of \eqref{P2} we understand critical points of the
associated Euler-Lagrange functional $E\in
\mathcal{C}^1(L^p(\Omega))$, which are given by
$$
E(v)=F(v)-G(v).
$$
As in \cite{DR-PO, WI}, we introduce the
modified Euler-Lagrange functional defined on
$\mathbb{R}\times L^p(\Omega)$ by
$$
A(t,v)=E(tv).
$$
If $v$ is an arbitrary element of
$L^p(\Omega)$, $\partial_tA(.,v)$ (resp. $\partial_{tt}A(.,v)$)are
the first (resp. second) derivative of the real valued function:
$t\mapsto A(t,v)$.
Since the functional $A$ is even in $t$ and that we are
interested by the positive solutions, we limit our study for
$t>0$.
\begin{theorem}\label{solpos}
Problem \eqref{P1} has a positive solution.
\end{theorem}
To prove theorem \ref{solpos}, we need the following preliminary results.
\subsection*{Case $p>q$:} Let $v$ be an arbitrary
element of $L^p(\Omega)\setminus \{0\}$.
It is clair that the
real valued function $t\mapsto A(t,v)$ is decreasing on
$]0,t(v)[$, increasing on $]t(v),+\infty[$ and attains its unique
minimum for $t=t(v)$, where
\begin{equation}\label{minA}
t(v)=\big(\frac{qG(v)}{pF(v)}\big)^\frac{1}{p-q}.
\end{equation}
On the other hand, a direct computation gives
$$
A(t(v),v)=\big(\frac{1}{p}-\frac{1}{q}\big)
\frac{(qG(v))^\frac{p}{p-q}}{(pF(v))^\frac{q}{p-q}}<0.
$$
Furthermore we have proved in \cite{TA-TS} that $E$ is bounded
bellow and coercive. We deduce that $A$ is also bounded bellow and
if
\begin{equation}\label{inf}
\alpha =\inf_{v\in L^p(\Omega)\setminus\{0\}} A(t(v),v),
\end{equation}
we get $-\infty<\alpha<0$.
Let $(v_n)\subset L^p(\Omega)\setminus \{0\}$
be a minimizing sequence of (\ref{inf}). Put $V_n=t(v_n)v_n$. Since
$E$ is coercive the sequence $(V_n)$ is bounded.
\begin{lemma}\label{liminf} The sequence $(V_n)$ satisfies
$$
\liminf_{n\to +\infty}\| V_n\|_p>0.
$$
\end{lemma}
\begin{proof} Suppose that there is a subsequence of
$(V_n)$, still denoted by $(V_n)$ such that $\lim_{n\to +\infty}\|
V_n\|= 0$. It follows that $\lim_{n\to +\infty}E(V_n)= 0$; i.e.
$\alpha=0$, which is impossible since $A(t(v_n),v_n)<0$.
\end{proof}
\begin{lemma}\label{infpos}
If $\mathbb{S}$ is the unit sphere of $L^p(\Omega)$, we have
$$\alpha=\inf_{v\in\mathbb{S}, v\geq 0}A(t(v),v).$$
\end{lemma}
\begin{proof}
For every $v\in L^p(\Omega)$, we have $|\Lambda v|\leq \Lambda |v|$ and
since $p>q$, we get
$$
A(t(v),v)\geq\big(\frac{1}{p}-\frac{1}{q}\big)\frac{(qG(|v|))
^\frac{p}{p-q}}{(pF(|v|))^\frac{q}{p-q}}
=A(t(|v|),|v|).
$$
On the other hand the relation (\ref{minA})
implies that $\forall r>0\ and\ \forall v\in L^p(\Omega)\setminus
\{0\}$,
$t(v)=\frac{1}{r}t(\frac{v}{r})$.
We deduce that
\begin{equation}\label{infsph}
\alpha =\inf_{v\in\mathbb{S}, v\geq 0}A(t(v),v),
\end{equation}
where $\mathbb{S}$ is the unit sphere of $L^p(\Omega)$.
\end{proof}
Note that the
minimizing sequences considered up to here are in $\mathbb{S}$ and
are nonnegative.
\begin{lemma}\label{ps}
Let $(v_n)\subset \mathbb{S}$ be a minimizing sequence
of (\ref{infsph}), then $(V_n):=(t(v_n)v_n)$ is Palais-Smale
sequence for the functional $E$.
\end{lemma}
\begin{proof}
We have $E(V_n)\to \alpha$. We show
that
$$
E'(V_n)\to 0\quad\text{in } L^{p'}(\Omega).
$$
Note that for every
$v\in L^p(\Omega)\setminus \{0\}$, we have
$\partial_tA(t(v),v)=0$ and $\partial_{tt}A(t(v),v)\neq 0$. The
implicit function theorem implies that $v\to t(v)$ is
$\mathcal{C}^1$ since $A$ is. Let us introduce the $\mathcal{C}^1$
functional $B$ defined on $\mathbb{S}$ by
$$
B(v)=A(t(v),v)=E(t(v)v).
$$
Then
$$
\alpha=\inf_{v\in \mathbb{S},v\geq 0}B(v)\quad \text{and} \quad
\lim_{n\to +\infty}B(v_n)=\alpha
$$
Using the Ekeland variational principle on the complete manifold
$(\mathbb{S},\| \cdot\|_p)$ to the functional $B$, we conclude that
$$
| B'(v)(\varphi)|\leq \frac{1}{n}\|\varphi\|_p,
\quad \textrm{for every } \varphi\in T_{u_n}\mathbb{S},
$$
where
$T_{v_n}\mathbb{S}$ is the tangent space to $\mathbb{S}$ at the
point $v_n$. Moreover, for every $\varphi\in T_{v_n}\mathbb{S}$,
one has
\begin{align*}
B'(v_n)(\varphi) &= \partial_tA(t(v_n),v_n)t'(v_n)(\varphi)+\partial_vA(t(v),v)(\varphi)
\\
&= \partial_vA(t(v),v)(\varphi),
\end{align*}
since $\partial_tA(t(v),v)=0$, where $t'(v)$ denotes the
derivative of $v\mapsto t(v)$ at the point $v$.
Furthermore, let
$P: L^p(\Omega)\setminus \{0\}\to \mathbb{R}\times \mathbb{S}$,
$$
v\mapsto (P_1(v),P_2(v))=(\| v\|_p,\frac{v}{\|
v\|_p}).
$$
Applying H\"older's inequality,
for every $(v,\varphi)\in L^p(\Omega)\setminus \{0\}\times
L^p(\Omega)$ we have
$$
\| P'_2(v)(\varphi)\|_p\leq 2\frac{\| \varphi\|_p}{\| v\|_p}.
$$
From lemma \ref{liminf} and by the fact
that $\| V_n\|_p = t(v_n)$, there is a positive constant $C$ such
that
$$
t(v_n)\geq C, \quad \forall n\in \mathbb{N}.
$$
Then for every $\varphi\in L^p(\Omega)$ we get
\begin{align*}
|E'(V_n)(\varphi)|&=
|\partial_tA(P_1(V_n),P_2(V_n))P'_1(V_n)(\varphi)
+\partial_vA(P_1(V_n),P_2(V_n))P'_2(V_n)(\varphi)| \\
&= |\partial_vA(t(v_n),v_n)P'_2(V_n)(\varphi)| \\
&= |B'(v_n)P'_2(V_n)(\varphi)| \\
&\leq \frac{1}{n}\| P'_2(V_n)(\varphi)\|_p \\
&\leq \frac{2}{n}\frac{\|\varphi\|_p}{C}.
\end{align*}
We easily conclude that
$\lim_{n\to +\infty} E'(V_n)=0$ in $L^{p'}(\Omega)$.
\end{proof}
\subsection*{Case $p
0$.
\begin{lemma}\label{infpos2}
If $\mathbb{S}$ is the unit sphere of $L^p(\Omega)$, we have
$$
\alpha=\inf_{v\in\mathbb{S}, v\geq 0}A(t(v),v).
$$
\end{lemma}
\begin{proof}
For every $v\in L^p(\Omega)\setminus \{0\}$, we have
$$
A(t(v),v)=\big(\frac{1}{p}-\frac{1}{q}\big)
\frac{(pF(v))^\frac{q}{q-p}}{(qG(v))^\frac{p}{q-p}}.
$$
Since $|\Lambda v|\leq \Lambda|v|$, we get
\[
A(t(v),v)\geq
\big(\frac{1}{p}-\frac{1}{q}\big)
\frac{pF(|v|)^\frac{q}{q-p}}{qG(|v|)^\frac{p}{q-p}}
=A(t(|v|),|v|).
\]
On the other hand, the relation (\ref{maxA}) implies that for every
$r>0$ and for every $v\in L^p(\Omega)\setminus\{0\}$,
$t(v)=\frac{1}{r}t(\frac{v}{r})$.
Hence
\begin{equation}\label{infmax}
\alpha=\inf_{v\in\mathbb{S}, v\geq 0}A(t(v),v).
\end{equation}
\end{proof}
Let $(v_n)$ be a
minimizing sequence of (\ref{infmax}), as in the case $p>q$, we put
$$ V_n=t(v_n)v_n.
$$
The proof of the following lemmas can be done like in the
previous case.
\begin{lemma}\label{liminf2} $\liminf_{n\to +\infty}\|
V_n\|_p>0$.
\end{lemma}
\begin{lemma}\label{ps2}
Let $(v_n)\subset \mathbb{S}$ be a minimizing sequence
of (\ref{infsph}). Then $(V_n):=(t(v_n)v_n)$ is Palais-Smale
sequence for the functional $E$.
\end{lemma}
\begin{proof}[Proof of theorem \ref{solpos}]
In our paper \cite{TA-TS} we showed that $E$ verifies
the Palais-Smale condition. Then by lemma \ref{ps} and lemma \ref{ps2},
we deduce that there is a subsequence of $(V_n)$, still noted by $(V_n)$
such that $V_n\to V$, $V\in L^p(\Omega)\setminus\{0\}$ and
$V\geq 0$. Moreover, since $E'(V_n)\to 0$, then $E'(V)=0$. i.e. $V$ is a
nonnegative solution of problem \eqref{P2}.
Hence
\begin{equation}\label{Vsol}
N_p(V)=\Lambda (mN_q(\Lambda V)).
\end{equation}
The assertion (vi) of lemma \ref{lem2.1}, the relation
(\ref{Vsol}) and the fact that $m\in L^p(\Omega)\setminus \{0\}$,
$m\geq 0$ enable us to claim that $N_p(V)>0$ and $V>0$.
Furthermore $U=\Lambda V$ is a positive solution of problem
\eqref{P1}.
\end{proof}
\section{Uniqueness of the positive solution}
\begin{theorem}\label{uniq}
If $m\in \mathcal{C}(\overline{\Omega})$, $m\geq 0$ and
$p>q$, then \eqref{P1} has a unique nonnegative
solution.
\end{theorem}
Problem \eqref{P2} is equivalent to the problem:
Find $v\in L^p(\Omega)\setminus \{0\}$, $v>0$ such that
\begin{equation}\label{P3}
N_p(v)=\| m^{1/q}\Lambda v\|_q^{q-p}\| m^{1/q}\Lambda
v\|_q^{p-q}\Lambda(mN_q(\Lambda v)) \quad\text{in }
L^{p'}(\Omega).
\end{equation}
To prove that problem
\eqref{P2} has a unique nonnegative solution, we will
study the principal positive eigenvalue of the eigenvalue problem:
Find $v\in L^p(\Omega)\setminus \{0\}\times \mathbb{R_+^*}$
such that
\begin{equation}\label{VP}
N_p(v)=\lambda\| m^{1/q}\Lambda
v\|_q^{p-q}\Lambda(mN_q(\Lambda v)) \quad\textrm{in}\quad
L^{p'}(\Omega).
\end{equation}
Consider the functionals $f$ and $g$ defined on $L^p(\Omega)$ by
$$
f(v)=\frac{1}{p}\| v\|_p\quad \text{and}\quad
g(v)=\frac{1}{p}(\int_{\Omega}m|\Lambda v|^qdx)^\frac{p}{q}.
$$
Hence problem (\ref{VP}) is equivalent to the problem:
Find $(v,\lambda)\in L^p(\Omega)\setminus \{0\}\times \mathbb{R^*_+}$
such that
\begin{equation}
f'(v)=\lambda g'(v)\quad\textrm{in }
L^{p'}(\Omega).
\end{equation}
Define
$$
\lambda_1=\inf _{v\in M}f(v),
$$
where $M= \{v\in L^p(\Omega)/g(v)=1\}$.
We need the preliminary results.
\begin{lemma}\label{vp1}
\begin{itemize}
\item[(i)] $\lambda_1$ is the first positive eigenvalue of problem
(\ref{VP}). Moreover $v_1$ is an eigenfunction associated with
$\lambda_1$ if and only if
$$
f(v_1)-\lambda _1g(v_1)=0=\inf_{v\in L^p(\Omega)\setminus
\{0\}}f(v)-\lambda_1g(v).
$$
\item[(ii)] Every eigenfunction associated with $\lambda_1$
is positive or negative.
\end{itemize}\end{lemma}
\begin{proof} (i) The functional
$f$ is weakly semi-continuous below and coercive on $M$. Since $g$
is weakly continuous, then $M$ is weakly closed. Hence there is
$v_1\in M$ such that $f(v_1)=\lambda_1=\lambda_1g(v_1)$.
The p-homogeneity of $f$ and $g$ implies that $\lambda_1$ is an
eigenvalue of problem (\ref{VP}) if and only if
$$
\forall v\in L^p(\Omega)\setminus\{0\},\quad
\lambda_1\leq \frac{f(v)}{|g(v)|}
$$
if and only if for all $v\in L^p(\Omega)\setminus \{0\}$,
\[
f(v)-\lambda_1 g(v)\geq f(v)-\lambda_1 |g(v)|
\geq 0= f(v_1)-\lambda_1 g(v_1).
\]
Now we show that $\lambda_1$ is the first positive eigenvalue:
Suppose on the contrary that there exits
$\lambda\in ]0, \lambda_1[$ and $v\in L^p(\Omega)\setminus\{0\}$
such that
$f(v)-\lambda g(v)=0$. Then we get
$$
0=f(v_1)-\lambda_1g(v_1)\leq f(v)-\lambda_1g(v)