%\pdfoutput=1\relax\pdfpagewidth=8.26in\pdfpageheight=11.69in\pdfcompresslevel=9

\documentclass[reqno]{amsart}
\usepackage{graphicx} \usepackage{hyperref}


\AtBeginDocument{{\noindent\small
{\em Electronic Journal of Differential Equations},
Vol. 2004(2004), No. 41, pp. 1--14.\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 2004 Texas State University - San Marcos.}
\vspace{9mm}}

\begin{document}

\title[\hfilneg EJDE-2004/41\hfil Existence of trivial and nontrivial solutions]
{Existence of trivial and nontrivial solutions of a fourth-order 
ordinary differential equation}

\author[Tihomir Gyulov \& Stepan Tersian\hfil EJDE-2004/41\hfilneg]
{Tihomir Gyulov \& Stepan Tersian} % in alphabetical order

\address{Tihomir Gyulov \hfill\break
C.A.M.I., University of Rousse,
 8 ``Studentska'' Str., 7017 Rousse, Bulgaria}
\email{tgyulov@ecs.ru.acad.bg}

\address{Stepan Tersian\hfill\break
C.A.M.I., University of Rousse,
 8 ``Studentska'' Str., 7017 Rousse, Bulgaria}
\email{tersian@ami.ru.acad.bg}

\date{}
\thanks{Submitted January 5, 2004. Published March 23, 2004.}
\subjclass[2000]{34B15, 34C25, 35K35}
\keywords{Fourth-order ordinary differential equation,
variational method, \hfill\break\indent Brezis-Nirenberg's theorem}

\begin{abstract}
 We study the multiplicity of nontrivial solutions for a
 semilinear fourth-order ordinary differential equation arising in
 spatial patterns for bistable systems. In the proof of our results,
 we use minimization theorems and Brezis-Nirenberg's linking theorem.
 We obtain also estimates on the minimizers of the
 corresponding functionals.
\end{abstract}

\maketitle
\numberwithin{equation}{section}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\numberwithin{equation}{section}

\section{Introduction}

In this paper, we study existence and multiplicity of solutions to the
boundary-value problem for the fourth-order ordinary differential equation
\begin{equation} \label{P}
 \begin{gathered}
u^{iv}+Au''+Bu+f(x,u)=0, \\ 
u(0)=u(L)=u''(0)=u''(L)=0,
\end{gathered}
\end{equation}
\noindent where $A$ and $B$ are constants and $f(x,u)$ is a
continuous function, defined in $\mathbb{R}^2$, whose potential
$F(x,u)=\int_0^uf(x,t)dt$ satisfies suitable
assumptions. The problem is motivated by the study of formation of spatial
periodic patterns in bistable systems. In the study of spatial patterns an
important role is played by a model equation, which is simpler than full
equations describing the process. Recently, interest has turned to
fourth-order parabolic differential equation, involving bistable dynamics,
such as the extended Fisher-Kolmogorov (EFK) equation
proposed by Coullet, Elphick \& Repaux in 1987 and Dee \& VanSaarlos in
1988. Another well known equation of this type is the Swift-Hohenberg (SH)
equation
proposed in 1977. With appropriate changes of variables, stationary
solutions of these equations lead to the equation
\begin{equation}
u^{iv}-pu''-u+u^{3}=0,  \label{e13}
\end{equation}
in which $p>0$ corresponds to EFK equation and $p<0$ to the SH equation.
Solutions of Eq. (\ref{e13}) which are bounded on the real line have been
recently studied by a variety of methods such as topological shooting method
and variational methods \cite{b1,c1,c2,p2,p3,p4,t1}.

When $f$ is an even $2L$ periodic function with respect to $x$, and odd with
respect to $u$, the $2L$ periodic extension $\bar{u}$ of the odd extension
of the solution $u$ of the problem \eqref{P} to the interval $[-L,L] $
yields a $2L$ periodic solution of \eqref{P}. The
solvability of \eqref{P} for some extension of (\ref{e13}) was
studied in \cite{c1,c2,p2,p3,t1} by variational methods.

We suppose that $f(x,0)=0$, $\forall x\in\mathbb{R}$ and the potential
\[
F(x,u)=\int_0^uf(x,s)\,ds
\]
satisfies following assumptions:
\begin{itemize}
\item[(H1)] There is a number $p>2$ and for each bounded
interval $I$ there is a constant $c>0$ such that
\[
F(x,u)\geq c|u| ^p,\ \forall x\in I , \forall u \in {\bf
R}
\]

\item[(H2)] $F(x,u)=o(u^2)$ as $u\rightarrow 0$, uniformly
with respect to $x $ in bounded intervals.
\end{itemize}
A typical function that satisfies (H1) and (H2) is
\[
f(x,u)=b(x)u|u| ^{p-2},\quad p>2,
\]
where $b(x)$ is a continuous, positive function.

Problem \eqref{P} has a variational structure and its solutions
can be found as critical points of the functional
\begin{equation}
I(u;L):=\frac 12\int_0^L(u''{}^2 -Au'^2+Bu^2)dx+\int_0^LF(x,u)dx  \label{e16}
\end{equation}
in the Sobolev space
\[
X(L):=H^2(0,L)\cap H_0^1(0,L).
\]

In this work we obtain nontrivial critical points of the functional $I$
using Brezis-Nirenberg's linking theorem \cite{b2,g1}. Recall its
statement. Let $E$ be a Banach space with a direct sum decomposition $
E=X\oplus Y$. The functional $J\in C^1(E,\mathbb{R})$ has a local
linking at $0$ if, for some $r>0$
\begin{gather*}
J(x)\leq 0, \quad x\in X, \quad \| x\| \leq r\,, \\
J(y)\geq 0, \quad y\in Y, \quad \| y\| \leq r\,.
\end{gather*}


\begin{theorem}[Brezis and Nirenberg \cite{b2}] \label{thm1}
Suppose that $J\in C^1(E,\mathbb{R})$
satisfies the $(PS)$ condition and has a local linking at 0.
Assume that $J$ is bounded below and $\inf_EJ<0$. Then $J$ has at least two
nontrivial critical points.
\end{theorem}

It is easy to see that if $4B\geq A^2$ and $f(x,u)u>0$ for $x\geq0$ and $
u\neq0$ the problem \eqref{P} has only the trivial solution. We shall assume $
4B<A^2$ and study separately the cases $A\leq0$ (EFK equation) and $A>0$ (SH
equation). Our main results are as follows.

\begin{theorem}[Nontrivial solutions] \label{thm2}
Let the function $F(x,u)$ satisfy (H1) and (H2).
\begin{itemize}
\item[(i)] Let $4B<A^2$, $A\leq 0$, $B<0$ and set
$L_1:=\pi \sqrt{2}/\sqrt{A+\sqrt{A^2-4B}}$. If $L>L_1$, then
problem \eqref{P} has at least two nontrivial solutions.

\item[(ii)] Let $4B<A^2$, $A>0$, and set
$L_1:=\pi \sqrt{2}/\sqrt{A+\sqrt{A^2-4B}}$. Then, problem \eqref{P}
has at least two nontrivial solutions if either
\begin{itemize}
\item[(a)] $B\leq 0$ and $L>L_1$, or
\item[(b)] $B>0 $, and $L\in ]nL_1,nM_1[$, where
$M_1:=\pi \sqrt{2}/\sqrt{A-\sqrt{A^2-4B}}$.
\end{itemize} \end{itemize}
\end{theorem}

\begin{theorem}[trivial solutions] \label{thm3}
Let the continuous function $f(x,u)$ satisfy the
assumption $f(x,0)=0$ and $f(x,u)u>0, u\neq 0$ for $x\in [0,L]$.
\begin{itemize}
\item[(i)] Let $4B<A^2$, $A\leq 0,$, set
$L_1:=\pi \sqrt{2}/\sqrt{A+\sqrt{A^2-4B}}$ for $B<0$.
Then problem \eqref{P} has only the trivial
solution provided that one of the following holds:
(a) $B\geq 0$, or (b) $B<0$ and $0<L\leq L_1$.

\item[(ii)] Let $4B<A^2$,  $A>0$, set
$L_1:=\pi \sqrt{2}/\sqrt{A+\sqrt{A^2-4B}}$ and
$$
h_n=\big(\frac{(n^2+n)A}{2n^2+2n+1}\big)^2,
$$
$n\in \mathbb{N}\cup \{0\}$.
Then problem \eqref{P} has only the
trivial solution provided that one of the following holds:
(a) $B \leq 0$ and $0<L<L_1$, or
(b) $h_n< B\leq h_{n+1}$ and $L\in T_{n+1}$, where $T_{n+1}$ is a finite
union of bounded intervals.
\end{itemize}
\end{theorem}

Next, we consider the problem
\begin{equation} \label{Q}
\begin{gathered}
u^{iv}+Au''+Bu+u^3=0,\quad 0<x<L, \\
u(0)=u(L)=u''(0)=u''(L)=0 ,
\end{gathered}
\end{equation}
which is related to stationary the EFK equation or to stationary the SH equation.
The corresponding energy functional is
\[
J(u;L)=\frac 1L\big\{ \frac 12\int_0^L(u''{}^2
-Au'^2+Bu^2)dx+\frac 14\int_0^Lu^4dx\big\}.
\]
By Theorem \ref{thm2} for $4B<A^2$, problem \eqref{Q} has at least two
nontrivial solutions if $L$ belongs to infinite interval $]L_1,+\infty[$ if
$A\leq 0,$or to a bounded interval $]nL_1,nM_1[$, if $A>0$, where $L_1$ and
$M_1$ are depended on $A$ and $B$. One of these nontrivial solutions is a
nontrivial minimizer $u_0$ of the functional $J$. In this section we will
estimate the average of $L^2$-norm of the minimizer $u_0$.

Let for $u\in X(L)$, let
\[
|u|^2:=\frac{1}{L}\int_0^{L}u^2(x)dx.
\]
Let $P(\xi )=\xi ^{4}-A\xi ^2+B$ be the symbol of the linear
operator $\mathcal{L}u=u^{iv}+Au''+Bu$. By the proof of
Theorem \ref{thm2}, if $L\in \Delta _n$, where $\Delta _n$ is an interval
which is the set of solutions of the inequality $P_n(L)<0$, \eqref{Q} has at
least two nontrivial solutions. Moreover if $L\in \Delta_n$, there exist
natural numbers $m,m+1,\ldots ,m+k$, $m\geq 1,k\geq 0$
depending on $L$ such that $P_{j}(L)<0$ if and only if
$j\in S=\{m,m+1,\ldots ,m+k\} $ and $P_{j}(L)\geq 0$ if and only if
$j\notin S$. Let $E_{k+1}(L)$ be the finite dimensional subspace of
$X(L)$
\[
E_{k+1}(L)=\mathop{\rm span}\big\{ \sin (\frac{m\pi x}{L})
,\ldots ,\sin (\frac{(m+k)\pi x}{L})\big\} ,
\]
and for $u\in X(L)$, $u=\overline{u}+\widetilde{u}$,
$\overline{u}\in E_{k+1}$, $\widetilde{u}\in E_{k+1}^{\bot }$ be the
orthogonal decomposition of $u$.

\begin{theorem} \label{thm4}
Let for a fixed $n\in \mathbb{N}$, let $\Delta _n$ be the set of solutions of the
inequality $P_n(L)<0$. For $L\in \Delta _n$ let
\[
P_{j}(L)<0\quad \mbox{if }j\in S=\{ m,m+1,\dots,m+k\} ,
\]
and
\[
p_n=P_{m_n}(L)=\min \{ P_{j}(L): j\in S\} <0.
\]
Then, if $L\in \Delta _n$, problem \eqref{Q} has a nontrivial
solution $u_0$, which is a minimizer of the functional $J$, and the following
estimates hold:
\begin{itemize}
\item[(i)] $-\frac 14p_n^2\leq J(u_0;L)\leq -\frac 16p_n^2$

\item[(ii)] $\frac 23|p_n|\leq |\overline{u}_0|^2\leq |u_0|^2\leq |p_n|$

\item[(iii)] $|\widetilde{u}_0|^2\leq (-\frac 23+\sqrt{\frac 23})|p_n|$

\item[(iv)] $J_1(\widetilde{u}_0;L)\leq \frac 29p_n^2$.
\end{itemize}
\end{theorem}

This paper is organized as follows: In Section 2 we prove some auxiliary
lemmas. In Section 3 we prove Theorem \ref{thm2}, and Theorem \ref{thm3}.
In Section 3 we prove Theorem \ref{thm4}.


\section{Preliminaries}

We study the nonautonomous fourth-order ordinary differential equation
\[
u^{iv}+Au''+Bu+f(x,u)=0,\quad 0<x<L,
\]
where $A\in \mathbb{R},\ B\in \mathbb{R}$ are constants and $f(x,u)$
is a continuous function, whose potential $F(x,u)=\int_0^uf(x,t)dt$ is a
nonnegative function which satisfies assumptions (H1) and (H2).

Let $X(L)$ be the Sobolev space
\[
X(L):=\{H^2(0,L):u(0)=u(
L)=0\}.
\]
A weak solution of the problem \eqref{P} is a function $u\in X(L)$, such that
\[
\int_0^L(u''v''-Au'v'+Buv+f(x,u)v)dx=0,\quad \forall v\in X(L).
\]
One can prove that a weak solution of \eqref{P} is a classical
solution of \eqref{P} (see \cite[Proposition 1]{t1}. Weak Solutions of
\eqref{P} are critical points of the functional $I:X(L)\rightarrow \mathbb{R}$,
\begin{equation}
I(u;L):=\frac 12\int_0^L(u''{}^2
-Au'^2 +Bu^2)dx+\int_0^LF(x,u)dx.  \label{e21}
\end{equation}

The following technical lemmata play an important role in further considerations.

\begin{lemma} \label{lm1}
We have the following: For $u\in X(L)$,
\begin{equation}
\int_0^Lu^2dx\leq \frac{L^{2k}}{\pi ^{2k}}\int_0^L(
u^{(k)})^2dx,\quad k=1,2 \,. \label{e22}
\end{equation}
The scalar product
\[
\langle u,v\rangle =\int_0^Lu''v''dx,
\quad u\in X(L),\ v\in X(L)
\]
\noindent induces an equivalent norm in $X(L)$. The set of
functions $\{ \sin (\frac{n\pi x}L):n\in \mathbb{N}\} $
is a complete orthogonal basis in $X(L)$.
\end{lemma}

\begin{proof} The Poinc\'{a}re type inequality (\ref{e22}) is proved in \cite{p1}.
For $u\in X(L)$, we have
\[
\int_0^Lu'^2 dx =\int_0^Lu' du=-\int_0^Luu''dx
\leq \frac 12\int_0^L(u^2+u''{}^2 )dx
\leq \frac 12\big(\frac{L^4}{\pi ^4}+1\big)\int_0^Lu''{}^2 dx,
\]
which shows that $[ u] =\langle u,u\rangle ^{1/2}$ is an
equivalent norm in $X(L)$. The set of functions $\{ \sin
\frac{n\pi x}L:n\in \mathbb{N}\} $ is clearly orthogonal in
$X(L)$ with respect to the scalar product $\langle \cdot ,\cdot \rangle $.
It is a complete orthogonal basis in $X(L)$. Indeed, let $v\in X(L)$ be such that
\[
\langle v,\sin \frac{n\pi x}L \rangle =0,\quad \forall n\in \mathbb{N}.
\]
Then
\[
0 =\int_0^{L}v''(\sin \frac{n\pi x}{L})''dx
=\int_0^{L}v(\sin \frac{n\pi x}{L})^{(4)}dx
=(\frac{n\pi }{L})^{4}\int_0^{L}v\sin \frac{n\pi x}{L}dx,
\]
and
\[
\int_0^Lv\sin \frac{n\pi x}Ldx=0,\quad \forall n\in \mathbb{N}.
\]
Since $X(L)\subset L^2(0,L)$ and $\{ \sin
\frac{n\pi x}L:n\in \mathbb{N}\} $ is an orthogonal basis in $L^2(
0,L)$ it follows that $v=0$, which means that the set $\{ \sin
\frac{n\pi x}L:n\in \mathbb{N}\} $ is a complete orthogonal basis in $
X(L)$.
\end{proof}

\begin{lemma} \label{lm2}
Let $A$, $B$ be constants and $f(x,u)$ be a continuous function
such that (H1) holds. Then the functional $I$ is bounded
from below and it satisfies the $(PS)$ condition.
\end{lemma}

\begin{proof}
Using Fourier series arguments and the previous lemma, we obtain
that for $u\in X(L)$,
\begin{equation} \label{e23}
\begin{gathered}
u=\sum_{k=1}^\infty c_k\sin \frac{k\pi x}L,\\
I(u;L)=\frac L4\sum_{k=1}^\infty c_k^2P(\frac{k\pi
}L)+\int_0^LF(x,u)dx,
\end{gathered}
\end{equation}
where $P(\xi )=\xi ^4-A\xi ^2+B$
is the symbol of the linear differential operator
\[
\mathcal{L}(u):=u^{(4)}+Au''+Bu.
\]
Observe that $P(\xi )$ is bounded from below for any $A$ and $B$
\[
P(\xi )\geq B-\frac{A^2}4.
\]
It follows from (\ref{e23}) and (H1) that $I(u;L)\geq 0$ if $4B\geq A^2$.
If $4B<A^2$ we have
\begin{equation}
I(u;L)>\frac 12\big(B-\frac{A^2}4\big)
\Vert u\Vert_{L^2}^2+C_1(L)\Vert u\Vert _{L^2}^p.  \label{e24}
\end{equation}
 From the elementary inequality
\[
-ax^2+bx^p\geq -a\frac{p-2}p(\frac{2a}{pb})^{\frac 2{p-2}}
\]
for $a>0$, $b>0$, $x>0$ and $p>2$, it follows that the right hand side of
(\ref{e24}) is bounded from below by a negative constant.

Suppose now that $(u_n)_n$ is a $(PS)$ sequence, i.e. there
exists $c_1>0$ such that
\begin{equation}
c_1>| I(u_n;L)| \quad \mbox{and} \quad I' (u_n;L)\rightarrow 0.  \label{e27}
\end{equation}
In what follows $c_{j}$ will denote various positive constants. We have
\[
I(u;L)=\frac 14\int_0^Lu''{}^2 dx+\frac 12
\bar{I}(u;L),
\]
where
\[
\bar{I}(u;L)=\frac 12\int_0^L(u''{}^2
-2Au'^2 +2Bu^2)dx+2\int_0^LF(x,u)dx.
\]
As before the functional $\bar{I}$ is bounded from below and we have
\[
c_1\geq \frac{1}{4}\int_0^{L}u_n''{}^2 dx-c_{2}.
\]
The sequence $(u_n)_n$ is a bounded sequence in $X(
L)$ in view of Lemma \ref{lm1}. There exists a subsequence still denoted by $
(u_n)_n$ and a function $u_0\in X(L)$ such
that
\begin{equation}
u_n\rightharpoonup u_0\quad \mbox{in} \quad X(L),
\label{e25}
\end{equation}
and by Sobolev's embedding theorem
\begin{equation} \label{e26}
\begin{gathered}
u_n \rightarrow u_0\quad \mbox{in} \quad C^{1}[0,L],   \\
u_n \rightarrow u_0\quad \mbox{in} \quad L^2(0,L).
\end{gathered}
\end{equation}
Since $f(x,u)$ is continuous and $\{ |u_n(
x)|\} $ uniformly bounded in $[0,L]$, and letting $
n\rightarrow \infty $ in
\[
(I'(u_n;L),u_0)=\int_0^L(
u_n''u_0''-Au_n'u_0' +Bu_nu_0+f(x,u_n)u_0)dx
\]
we obtain
\begin{equation}
\int_0^L(u_0''{}^2 -Au_0'^2
+Bu_0^2+f(x,u_0)u_0)dx=0.  \label{e28}
\end{equation}
 From the boundedness of $(u_n)_n$ in $X(L)$
and (\ref{e28}) it follows $(I'(u_n;L),u_n)\rightarrow 0$ and
\begin{align*}
\int_0^{L}u_n''{}^2 dx
&=(I'(u_n),u_n)+\int_0^{L}(Au_n'^2-Bu_n^2-f(x,u_n)u_n)dx \\
&\rightarrow \int_0^{L}(Au_0'^2-Bu_0^2-f(x,u_0)u_0)dx
=\int_0^{L}u_0''{}^2 dx,
\end{align*}
which implies that $\| u_n\|\rightarrow \| u_0\|$ and then
$\| u_n-u_0\|\rightarrow 0$, which completes the proof of Lemma \ref{lm2}.
\end{proof}

\section{Existence results}

The polynomial
\[
p(\xi )=\xi ^4-A\xi ^2  \nonumber
\]
and the real functions
\[
p_n(L)=p(\frac{n\pi }L) \nonumber
\]
play an important role in the sequel.

Let $A\leq 0$. The polynomial $p(\xi )$ is a positive
increasing and convex function for $\xi >0$. The functions
$p_n(L)$ are positive decreasing functions for every $n\in \mathbb{N}$ and
\begin{gather*}
p_n(L)\rightarrow +\infty ,\quad \mbox{as } L\rightarrow 0,\\
p_n(L)\rightarrow 0,\quad \mbox{as}{\quad }L\rightarrow +\infty .
\end{gather*}
These functions are ordered as
\[
0<p_1(L)<p_{2}(L)<\ldots <p_n(L) <\ldots
\]
for every $L>0$, and some of their graphs are showm in Figure 1.

\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.5\textwidth]{fig1.eps}  % figure=gst11.eps
\end{center}
\caption{Graphs of functions $p_n(L)= (\frac{n\pi }L)^4+ (\frac{n\pi }L)^2$, $n=1,2,3,4$}
\end{figure}

Let $B<0$. Then the  equation $p_n(L)+B=0$ has the unique
solution
\begin{equation}
L_n=nL_1,\quad L_1:=\frac{\pi \sqrt{2}}{\sqrt{A+\sqrt{A^2-4B}}}
\label{e32}
\end{equation}
and
\begin{gather}
p_n(L)+B \geq 0\quad \mbox{if }L\leq nL_1,\label{f1} \\
p_n(L)+B < 0\quad \mbox{if } L>nL_1.  \label{f2}
\end{gather}

Let $A>0$. Then the polynomial $p(\xi )=\xi ^{4}-A\xi ^2$ is
positive for $\xi >\sqrt{A}$ and it has a negative minimum
$p_0=-A^2/4$ at $\xi _0=\sqrt{A/2}$. The functions
$p_n(L)$ are decreasing if $0<L<n\pi \sqrt{2/A}$ and
increasing if $L>n\pi \sqrt{2/A}$, $p_n(L)>0$ if $0<L<n\pi /\sqrt{A}$
and $p_n(L)<0$ if $L>n\pi /\sqrt{A}$. The graphs of
functions $p_n(L)$ with $A=1$ and $n=1,2,3,4$ are presented
on Figure 2.

\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.5\textwidth]{fig2.eps} % gst22.eps
\end{center}
\caption{Graphs of functions $p_n(L)= (\frac{n\pi }L)^4- (\frac{n\pi }L)^2$, $n=1,2,3,4$}
\end{figure}

\begin{lemma} \label{lm3}
Let $l_n:=\frac \pi {\sqrt{A}}\sqrt{2n^2+2n+1}$ and
$L_1:=\pi \sqrt{2}/\sqrt{A+\sqrt{A^2-4B}}$. Then we have the following
results:
\begin{itemize}
\item[(a)]
\begin{equation} \label{e41}
\begin{aligned}
p_n(L)=p_{n+1}(L)\Leftrightarrow L=l_n,\\
p_n(L)<p_{n+1}(L)\Leftrightarrow L<l_n, \\
p_n(L)>p_{n+1}(L)\Leftrightarrow L>l_n,
\end{aligned}
\end{equation}
and
\begin{equation}
q(L)=\inf \{ p_n(L): n\in \mathbb{N} \} =
\begin{cases}
p_1(L),& 0<L\leq l_1, \\
p_{n+1}(L),& l_n<L\leq l_{n+1.}
\end{cases}  \label{e42}
\end{equation}

\item[(b)] Let $B\leq 0$. Then $p_n(L)+B<0$ if and only if $L>nL_1$.

\item[(c)] Let $B>0$ and
$M_1:=\pi \sqrt{2}/\sqrt{A-\sqrt{A^2-4B}}$.
Then $p_n(L)+B<0$ if and only if $nL_1<L<nM_1$.
\end{itemize}
 \end{lemma}

\begin{proof}
(a) The equation $p_n(L)=p_{n+1}(L)$ is equivalent to
$(nx)^4-A(nx)^2=((n+1)x)^4-A((n+1)x)^2$, where $x=\pi /L$. A direct calculation
shows that it is satisfied for $x^2=A/(2n^2+2n+1)$, i.e. for
$L=l_n=\frac \pi {\sqrt{A}}\sqrt{2n^2+2n+1}$.
We have
\[
p_n(l_n)=-\big(\frac{(n^2+n)A}{2n^2+2n+1}\big)^2=-h_n
\]
and (\ref{e41}) holds. From (\ref{e41}) it follows (\ref{e42}).

\noindent (b) To solve the inequality $p_n(L)+B<0$ we set $
x=\pi/ L$, and assuming that $B\leq0$ we are led to
\[
(nx)^2\in \big] 0,\frac{A+\sqrt{A^2-4B}}2\big[ ,
\]
equivalent to $L>nL_1$.

\noindent (c) To solve the same inequality in the case $B>0$ we compute
\[
(nx)^2\in \big] \frac{A-\sqrt{A^2-4B}}2,\frac{A+\sqrt{A^2-4B}}2\big[
\]
which is equivalent to $nL_1<L<nM_1$.
\end{proof}

\begin{proof}[Proof of Theorem \ref{thm2}]
\noindent Case (i). Let $L>L_1$. There exists a natural number $n$
such that $nL_1<L\leq (n+1)L_1$. Let
$\varphi_n\in E_n=\mathop{\rm span}\{ \sin \frac{
\pi x}L,\ldots ,\sin \frac{n\pi x}L\} $ such that
\[
\varphi _n(x)=\sum_{k=1}^nc_k\sin (\frac{k\pi x}
L),
\]
and set $c_1^2+\ldots +c_n^2=\rho ^2$. By (\ref{f2}) we have that
\[
\alpha _n=\max \{p_k(L)+B:k=1,\dots ,n\}<0.
\]
Let us take a small constant $\varepsilon $, such that
$0<\varepsilon <-\alpha _n/2$.

By assumption (H2), there exists $\delta >0$ such that
if $|u|\leq \delta $ then $F(x,u)\leq \varepsilon |u|^2$,
$x\in[ 0,L] $. Let us take $\rho ,0<\rho \leq \delta /\sqrt{n}$.
Then by
\[
|\varphi _n(x)|\leq \sum_{k=1}^n|c_k|\leq \sqrt{n}(
\sum_{k=1}^nc_k^2)^{1/2}=\sqrt{n}\rho \leq \delta
\]
it follows that
$F(x,\varphi _n(x))\leq \varepsilon |\varphi_n(x)|^2$
and
\begin{align*}
\int_0^LF(x,\varphi _n(x))dx
&\leq \varepsilon \int_0^L|\varphi _n(x)|^2dx\\
&=\varepsilon \int_0^L\sum_{k=1}^nc_k^2\sin ^2(\frac{k\pi x}L)dx\\
&=\varepsilon \frac L2\sum_{k=1}^nc_k^2=\varepsilon \frac L2\rho ^2.
\end{align*}
We have
\begin{equation} \label{e34}
\begin{aligned}
I(\varphi _n;L)
&=\frac L4\sum_{k=1}^n(p_k(L)+B)c_k^2
+\int_0^LF(x,\varphi _n(x))dx   \\
&\leq \frac L4\alpha _n\rho ^2+\varepsilon \frac L2\rho ^2   \\
&=\frac L2\rho ^2(\frac 12\alpha _n+\varepsilon )<0,
\end{aligned}
\end{equation}
if $0<\rho \leq \delta /\sqrt{n}$.
The functional $I$ has a local linking at $0$. Indeed, by (\ref{e34}), for
sufficiently small $\rho >0$ we have
\[
I(u;L)\leq0,\quad u\in E_n,\|u\|<\rho .
\]
Let $u\in E_n^{\perp }$ and $\|u\|\leq \rho $. It follows that $
p_{n+1}(L)+B\geq 0$ if $nL_1<L\leq (n+1)L_1$ by (\ref{f1}).
Since $p_{n+1}(L)<p_{n+2}(L)<\ldots ,$by
assumption (H1) there exists $C(L)>0$ such that
\begin{align*}
I(u;L)&\geq \frac 12\min ((p_k(L)+B):k\geq n+1)\Vert u\Vert_{L^2}^2
+C(L)\Vert u\Vert _{L^2(0,L)}^p \\
&\geq \frac 12(p_{n+1}(L)+B)\Vert u\Vert_{L^2}^2+C(L)\Vert u\Vert _{L^2(0,L)}^p
\geq 0,
\end{align*}
if $u\in E_n^{\perp }$. The functional $I$ satisfies the $(PS)$ condition.
In view of Theorem \ref{thm1}, for $L>L_1$ the functional $I$ has at least two
nontrivial critical points.

\noindent(ii). By Lemma \ref{lm3}, (b) $p_k(L)+B<0$ iff $
L>kL_1$. If $L>L_1$ there exists a natural number $n$ such that $nL_1<L\leq
(n+1)L_1$ and
\begin{gather*}
p_k(L)+B<0,\quad k=1,\dots ,n \\
p_k(L)+B\geq 0,\quad k\geq n+1.
\end{gather*}
In this case the proof is finished exactly as in the proof in the Case (i).


\noindent {\it Step 1. Nontrivial solutions in the case $B>0$.}
Let $\Delta _n=] nL_1,nM_1[ $. \medskip\ Observe that for a fixed
$L\in \Delta _n$ there exist finite number of intervals $\Delta _j$ numbered
as $\Delta _m, \Delta _{m+1},\dots ,\Delta _{m+k}$ such that $L\in \Delta _j
\cup \Delta _n $ if and only if  $j\in S:=\{m, m+1,\dots ,m+k \}$ and
\begin{gather*}
p_j(L)+B<0,\quad j\in S,\\
p_j(L)+B\geq 0,\quad j\notin S.
\end{gather*}
Let
\[
E_{k+1}:=\mathop{\rm span}\big\{ \sin \big(\frac{m\pi x}L\big),
\sin \big(\frac{(m+1)\pi x}L\big),\dots ,\sin \big(\frac{(m+k) \pi x}L\big)\big\} .
\]
With a computation similar to the one in the proof of Theorem \ref{thm3} we observe
that
\[
I(u,L)<0,\quad u\in E_{k+1},\; 0<\|u\|\leq r,
\]
if $r$ is sufficiently small and
\[
I(u,L)\geq 0,\quad u\in E_{k+1}^{\bot },
\]
which implies  that $I$ has a local linking at $0$. Then $I$ has at least two
nontrivial critical points.
\end{proof}

\begin{proof}[Proof of Theorem \ref{thm3}]
By Lemma \ref{lm1}, for $u\in X(L)$ we have:
\begin{gather*}
u=\sum_{k=1}^\infty c_k\sin \frac{k\pi x}L, \\
I(u;L)=\frac L4\sum_{k=1}^\infty c_k^2P(\frac{k\pi}L)+\int_0^LF(x,u)dx, \\
\mathcal{B}(u,u):=\langle I'(u;L),u\rangle
 =\frac L2 \sum_{k=1}^\infty c_k^2P(\frac{k\pi }L)
+\int_0^Lf(x,u)u \, dx,
\end{gather*}
where
$P(\xi )=\xi ^4-A\xi ^2+B=p(\xi)+B$
is the symbol of the linear differential operator
\[
\mathcal{L}(u):=u^{i v}+Au''+Bu.
\]

\noindent {Case (i).}
Let $B<0$ and $0<L\leq L_1$. We have seen that $
p_1(L)+B\geq 0$. As $P(\frac{k\pi }L)\geq
p_1(L)+B\geq0$ we infer that $\mathcal{B}(u,u)>0$ if $u\neq 0$ which means that
the functional $I$ has only the trivial critical point. If $B\geq0$ the same
argument applies for every $L>0$.

\noindent {Case (ii).}
 We consider the solvability of the inequality
\begin{equation}
q(L)+B\geq 0,  \label{e44}
\end{equation}
where $q(L)= \inf \{ p_n(L):n\in \mathbb{N}\} $. Let
$0<B\leq (\frac 4{25})A^2=h_1$ and
\begin{equation}
T_1:=\begin{cases}
] 0,L_1] , &  B<h_1 \\
] 0,L_1] \cup \{ l_1\} , & B=h_1.
\end{cases}
\end{equation}
By Lemma \ref{lm3}, (c) if $B\leq h_1$, inequality (\ref{e44}) holds if
and only if $L\in T_1$.
Let $l_0=0,h_n<B<h_{n+1}$ and
\[
D_{n+1}=] 0,L_1] \cup [ M_1,2L_1] \cup \dots \cup
[ nM_1,(n+1)L_1] .
\]
Let
\begin{equation}
T_{n+1}:=\begin{cases}
D_{n+1},&  h_n<B<h_{n+1}, \\
D_{n+1}\cup \{ l_{n+1}\} , & B=h_{n+1}.
\end{cases}
\end{equation}
By Lemma \ref{lm3}, (c) if $h_n<B\leq h_{n+1}$ the inequality (\ref{e44}) is
satisfied if and only if $L\in T_{n+1}$.

We estimate the quadratic term in $\mathcal{B}(u,u)$ as
\[
\frac L2\sum_{k=1}^\infty c_k^2P(\frac{k\pi }L)\geq
(q(L)+B)\|u\|_{L^2}^2\geq 0,
\]
and conclude that that $\mathcal{B}(u,u)>0$ if $u\neq 0$. Then the functional
$I$ has only the trivial critical point which completes the proof of Theorem
\ref{thm3}.
\end{proof}

\section{Bounds for the minimizer}

Let us consider the problem
\begin{gather*}
u^{iv}+Au''+Bu+u^3=0,\quad 0<x<L, \\
u(0)=u(L)=u''(0)=u''(L)=0 ,
\end{gather*}
and the corresponding energy functional
\[
J(u;L)=\frac 1L\big\{ \frac 12\int_0^L(u''{}^2
-Au'^2+Bu^2)dx+\frac 14\int_0^Lu^4dx\big\}.
\]
For $u\in X(L)$, let
\[
|u|^2:=\frac 1L\int_0^Lu^2(x)dx,
\]
and
\begin{equation}
J(u;L)=\frac 12J_1(u;L)+\frac 14J_2(
u;L) \label{ae1}
\end{equation}
where
\begin{gather*}
J_1(u;L):=\frac 1L\int_0^L(u''{}^2 -Au'^2+Bu^2)dx,\\
J_2(u;L):=\frac 1L\int_0^Lu^4dx.
\end{gather*}
Let $P(\xi )=\xi ^4-A\xi ^2+B$ be the symbol of the linear
operator $\mathcal{L}u=u^{iv}+Au''+Bu$ and
\[
P_n(L)=P(\frac{n\pi }L)=p(\frac{n\pi }
L)+B.
\]
We have that $u\in X(L)$
\begin{gather*}
u=\sum_{j=1}^\infty c_j\sin (\frac{j\pi x}L), \\
J(u;L)=\frac 14\sum_{j=1}^\infty c_j^2P_j(L)
+\frac 1{4L}\int_0^Lu^4dx.
\end{gather*}
If $u_0$ is the minimizer of  $J$, then
\begin{equation}
\frac{d}{dt}J(tu_0;L)\big|_{t=1}=J_1(u_0;L)+J_{2}(u_0;L)=0.  \label{ae2}
\end{equation}
By the proof of Theorem \ref{thm2}, if $L\in \Delta _n$, where $\Delta _n$ is an
interval which is the set of solutions of the inequality $P_n(L)
<0$, \eqref{Q} has at least two nontrivial solutions. Moreover, if $
L\in \Delta _n$, there exist natural numbers $m,m+1,\ldots ,m+k,\quad m\geq
1,k\geq 0$ depending on $L$ such that $P_j(L)<0$ if $j\in
S=\{ m,m+1,\ldots ,m+k\} $ and $P_j(L)\geq 0$ if $
j\notin S$.

Let $E_{k+1}(L)$ be the finite dimensional subspace of $X(L)$
\[
E_{k+1}(L)=\mathop{\rm span}\{ \sin (\frac{m\pi x}{L})
,\ldots ,\sin (\frac{(m+k)\pi x}{L})\} ,
\]
and for $u\in X(L),\ u=\overline{u}+\widetilde{u}\quad
\overline{u}\in E_{k+1},\ \widetilde{u}\in E_{k+1}^{\bot }$ be the
orthogonal decomposition of $u$. We have
\begin{gather}
J_1(\overline{u};L)=\sum_{j\in S}c_{j}^2P_{j}(L)\leq 0,  \label{ae21} \\
J_1(\widetilde{u};L)=\sum_{j\notin S}c_{j}^2P_{j}(L)\geq 0.  \label{ae22}
\end{gather}
Denote
\[
p_n=P_{m_n}(L)=\min \{ P_j(L): j\in S\} <0,
\]
where $m_n\in S$ depends on $n$ and $L$.

\begin{proof}[Proof of Theorem \ref{thm4}]
We have
\begin{equation} \label{ae3}
J(u;L)=\frac 14\sum_{j=1}^\infty c_j^2P_j(
L)+\frac 1{4L}\int_0^Lu^4dx
\geq \frac 12p_n|u|^2+\frac 14|u|^4
\geq -\frac 14p_n^2.
\end{equation}
The proof of the estimates (i)-(iv) is based
upon a selection of a suitable test function. We set
\begin{gather*}
E_n(L)=sp\{ \sin (\frac{m_n\pi x}{L}) \} \subset E(L),
u_1(x)=c\sin (\frac{m_n\pi x}{L})\in E_n(L),
\end{gather*}
where $c$ will be chosen later. We have
\[
J(u_0;L)\leq J(u_1;L)=\frac{1}{4}c^2p_n+
\frac{3c^{4}}{32}.
\]
Taking $c^2=c_0^2=-\frac 43p_n=\frac 43|p_n|$, we obtain
\begin{equation}
J(u_0;L)\leq J(u_1;L)=-\frac 16p_n^2,  \label{ae4}
\end{equation}
which together with (\ref{ae3}) proves (i).

We have by (\ref{ae1}), (\ref{ae2}), and (\ref{ae4})
\[
-\frac 16p_n^2\geq J(u_0;L)=\frac 14J_1(u_0;L)\geq
\frac 14p_n|\overline{u}_0|^2
\]
which implies
\begin{equation}
|\overline{u}_0|^2\geq -\frac 23p_n=\frac 23|p_n|
=|u_1|^2.  \label{ae5}
\end{equation}
However by (\ref{ae2}) and (\ref{ae22})
\begin{equation} \label{ae51}
\begin{aligned}
|u_0|^4 &\leq \frac 1L\int_0^Lu_0^4dx=-J_1(u_0;L)  \\
&=-J_1(\overline{u}_0;L)-J_1(\widetilde{u}_0;L)  \\
&\leq -J_1(\widetilde{u}_0;L)+|p_n\| \overline{u}_0|^2  \\
&\leq |p_n\|\overline{u}_0|^2
\leq |p_n\|u_0|^2
\end{aligned}
\end{equation}
and
\begin{equation}
|u_0|^2\leq |p_n|,  \label{ae6}
\end{equation}
which together with (\ref{ae5}) proves (ii).

Denote $p=|\overline{u}_0|^2$ and $q=|\widetilde{u}_0|^2$. 
Then we have $|u_0|^2=p+q$ and by (\ref{ae51}),
\[
(p+q)^2=|u_0|^{4}\leq -J_1(\widetilde{u}_0;L)+|p_n|p
\]
which is equivalent to
\begin{equation}
p^2+(2q-|p_n|)p+(q^2+J_1(\widetilde{u}_0;L))\leq 0.  \label{ae7}
\end{equation}
Then
\[
(2q-|p_n|)^2-4(q^2+J_1(\widetilde{u}_0;L))\geq 0
\]
which implies
\[
D_n^1:=|p_n|^2-4|p_n|q-4J_1(\widetilde{u}_0;L)\geq 0.
\]
It follows by (\ref{ae7}) that
\[
\frac 12(|p_n|-2q-\sqrt{D_n^1})\leq p\leq \frac
12(|p_n|-2q+\sqrt{D_n^1}).
\]
By (i), we have $\frac 23|p_n|\leq p$ and then
\[
\frac 23|p_n|\leq \frac 12(|p_n|-2q+\sqrt{
D_n^1})
\]
or
\[
\frac 16|p_n|+q\leq \frac 12\sqrt{D_n^1}.
\]
Then
\[
\frac 1{36}|p_n|^2+\frac 13|p_n|q+q^2\leq \frac
14(|p_n|^2-4|p_n|q-4J_1(\widetilde{u}_0;L)),
\]
or
\[
q^2+\frac 43|p_n|q+(J_1(\widetilde{u}_0;L)-\frac 29|p_n|^2)\leq 0.
\]
Then, we  have
\[
D_n^2=\frac 49|p_n|^2-\big(J_1(\widetilde{u}_0;L)
-\frac 29|p_n|^2\big)\geq 0
\]
or
\[
\frac 23|p_n|^2-J_1(\widetilde{u}_0;L)\geq 0.
\]
Moreover
\[
-\frac{2}{3}|p_n|-\sqrt{D_n^2}\leq q\leq -\frac{2}{3}
|p_n|+\sqrt{D_n^2}
\]
and since $J_1(\widetilde{u};L)\geq 0$,
\[
0\leq q\leq \Big(-\frac{2}{3}+\sqrt{\frac{2}{3}}\Big)|p_n|,
\]
which proves (iii). Observe that from
\[
0\leq -\frac{2}{3}|p_n|+\sqrt{\frac{2}{3}|p_n|
^2-J_1(\widetilde{u}_0;L)}
\]
it follows
$J_1(\widetilde{u}_0;L)\leq \frac{2}{9}|p_n|^2$,
which proves (iv).
\end{proof}

\subsection*{Acknowledgments}
The work in this paper was partially supported by Grant MM 904/99 from the
Bulgarian National Research Foundation,  and by Grant 2003-PF-03
from the University of Rousse.


\begin{thebibliography}{10}
\bibitem{b1}  G. J. B van den Berg, L. A. Peletier and W. C. Troy. {\it Global
branches of multy bump periodic solutions of Swift--Hohenberg equation,}
Arch. Rational Mech. Anal. 158 (2001), 91-153.

\bibitem{b2}  H. Brezis and L. Nirenberg, {\it Remarks on finding critical
points,} Comm. Pure and Appl. Math. XLIV (1991), 939-963.

\bibitem{c1}  J. Chaparova, {\it Existence and numerical approximations of
periodic solutions of semilinear fourth-order differential equations,} J.
Math. Anal. Appl. 273 (2002), 121-136.

\bibitem{c2}  J. Chaparova, L. Peletier and S. Tersian, {\it Existence and
nonexistence of nontrivial solutions of fourth and sixth order ODE,}
Differential and Integral Equations, 2003.

\bibitem{g1}  M. R. Grossinho, S. A. Tersian, ``An Introduction to minimax
theorems and their applications differential equations'', Kluwer Academic
Press, 2001.

\bibitem{p1}  L. A. Peletier and W. C. Troy, ``Spatial Patterns: Higher
Order Models in Physics and Mechanics'', Birkh\"{a}ser, Boston, 2001.

\bibitem{p2}  L. A. Peletier, W. C. Troy and V. der Vorst. {\it Stationary
solutions of a fourth-order nonlinear diffusion equation, } Differential
equations, 31, 2 (1995) 301-314.

\bibitem{p3}  L. A. Peletier and V. Rottsch\"{a}fer, {\it Large time
behavior of solutions of the Swift-Hohenberg equation, } C. R. Acad. Sc.
Paris, Ser. I, Math., 336 (2003), 225-230.

\bibitem{p4}  L. A. Peletier and V. Rottsch\"{a}fer, {\it Pattern selection
of solutions of the Swift-Hohenberg equation, } Preprint, 2003.

\bibitem{t1}  S. Tersian and J. Chaparova, {\it Periodic and homoclinic
solutions of extended Fisher-Kolmogorov equation, } J. Math Anal. Appl., 266
(2001), 490-506.
\end{thebibliography}

\end{document}
