\documentclass[reqno]{amsart} \usepackage{hyperref} \AtBeginDocument{{\noindent\small \emph{Electronic Journal of Differential Equations}, Vol. 2014 (2014), No. 55, pp. 1--13.\newline ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu \newline ftp ejde.math.txstate.edu} \thanks{\copyright 2014 Texas State University - San Marcos.} \vspace{9mm}} \begin{document} \title[\hfilneg EJDE-2014/55\hfil Emden-Fowler problem] {Emden-Fowler problem for discrete operators with variable exponent} \author[M. M\u{a}lin \hfil EJDE-2014/55\hfilneg] {Maria M\u{a}lin} % in alphabetical order \address{Maria M\u{a}lin \newline Department of Mathematics, University of Craiova, 200585, Romania} \email{amy.malin@yahoo.com} \thanks{Submitted December 8, 2013. Published February 25, 2014.} \subjclass[2000]{39A14, 44A55, 34C375} \keywords{Difference equations; discrete $p(\cdot)$-Laplacian; homoclinic solution, \hfill\break\indent mountain pass lemma} \begin{abstract} In this article, we prove the existence of homoclinic solutions for a $p(\cdot)$-Laplacian difference equation on the set of integers, involving a coercive weight function and a reaction term satisfying the Ambrosetti-Rabinowitz condition. The proof of the main result is obtained by using critical point theory combined with adequate variational techniques, which are mainly based on the mountain pass theorem. \end{abstract} \maketitle \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{definition}[theorem]{Definition} \newtheorem{remark}[theorem]{Remark} \allowdisplaybreaks \section{Introduction and main result} This article concerns the difference non-homogeneous equation \begin{equation}\label{eq1} \begin{gathered} -\Delta\phi_{p(k-1)}(\Delta u(k-1))+a(k)\phi_{p(k)}(u(k)) =f(k,u(k))\quad\text{for all }k\in\mathbb{Z} \\ u(k)\to 0 \quad\text{as }|k|\to \infty, \end{gathered} \end{equation} where $\phi_{p(k)}(t)=|t|^{p(k)-2}t$ for all $t\in\mathbb{R}$ and for each $k\in\mathbb{Z}$. We have denoted by $\Delta$ the difference operator, which is defined by $\Delta u(k-1)=u(k)-u(k-1)$ for each $k\in\mathbb{Z}$. Moreover, \begin{equation}\label{eq2} \Delta\phi_{p(k-1)}(\Delta u(k-1)) =|\Delta u(k)|^{p(k)-2}\Delta u(k)-|\Delta u(k-1)|^{p(k-1)-2}\Delta u(k-1) \end{equation} for each $k\in\mathbb{Z}$. The solutions of problem \eqref{eq1} are referred to as homoclinic solutions of the equations $$ -\Delta\phi_{p(k-1)}(\Delta u(k-1))+a(k)\phi_{p(k)}(u(k)) =f(k,u(k)),\quad k\in\mathbb{Z}. $$ The goal of this article is to use the critical point theory to establish the existence of nontrivial homoclinic solutions for \eqref{eq1}. Our idea is to transfer the problem of the existence of solutions for \eqref{eq1} into the problem of existence of critical points for some associated energy functional. To explain the notion of homoclinic solution we go back to the definition of homoclinic orbit, which was introduced by Poincar\'e \cite{12} for continuous Hamiltonian systems. In the theory of differential equations, a trajectory $x(t)$, which is asymptotic to a constant as $|t|\to\infty$, is called a doubly asymptotic or homoclinic orbit. Since we are seeking solutions $u(k)$ for \eqref{eq1} satisfying $\lim_{|k|\to\infty}u(k)=0$, then according to the notion of a homoclinic orbit by Poincar\'e \cite{12}, we are interested in finding homoclinic solutions for problem \eqref{eq1}. The presence of the nonconstant potential $p(\cdot)$ is an important feature of this paper. The study of difference equations involving non-homogeneous difference operators of type \eqref{eq2} was initiated by Mih\u ailescu, R\u adulescu and Tersian in \cite{10}, where some eigenvalue problems were investigated. The study of discrete boundary value problems has captured special attention in the previous decade. In this context we point out the results obtained in the papers of Agarwal, Perera and O'Regan \cite{1}, Cabada, Iannizzotto and Tersian \cite{3}, Fang and Zhao \cite{5}, Ma and Guo \cite{9}, Krist\'{a}ly, Mih\u ailescu, R\u adulescu and Tersian \cite{8}. The studies regarding such type of problems can be placed at the interface of certain mathematical fields such as nonlinear partial differential equations and numerical analysis. For the rest of this article, we will use the notation $$ p^{+}:=\sup_{k\in\mathbb{Z}}p(k),\quad p^{-}:=\inf_{k\in\mathbb{Z}}p(k). $$ We assume that \begin{equation}\label{eq3} p(\cdot):\mathbb{Z}\to (1,+\infty),\ 1
0$ for all $k\in\mathbb{Z}$; $a_0
p^{+}$ and $r>0$ such that $0<\alpha F(k,t)\leq f(k,t)t$ for all $k\in\mathbb{Z}$, $t\geq r>0$, where $F:\mathbb{Z}\times\mathbb{R}\to\mathbb{R}$ is defined by $$ F(k,t)=\int_0^t f(k,s) ds \quad\text{for all }k\in\mathbb{Z},\ t\in\mathbb{R}; $$ \end{itemize} The main result of this work is given by the following theorem. \begin{theorem}\label{th1} Suppose that $p(\cdot)$ satisfies \eqref{eq3}, $a(\cdot)$ satisfies {\rm (A1)} and $f$ satisfies {\rm (F1)-(F2)}. Then \eqref{eq1} admits at least a nontrivial homoclinic solution. \end{theorem} This article is organized as follows. In Section 2 we define the functional spaces and prove some of their useful properties and in Section 3 we prove the existence of homoclinic solutions of problem \eqref{eq1}, employing the mountain pass theorem of Ambrosetti $\&$ Rabinowitz \cite{2}. \section{Auxiliary results} For each $p(\cdot):\mathbb{Z}\to (1,\infty)$, we introduce the space $$ l^{p(\cdot)}:=\big\{u:\mathbb{Z}\to \mathbb{R}; \rho_{p(\cdot)}(u) :=\sum_{k\in\mathbb{Z}}|u(k)|^{p(k)}<\infty\big\}. $$ On $l^{p(\cdot)}$ we introduce the Luxemburg norm $$ |u|_{p(\cdot)}:=\inf\big\{\lambda>0: \sum_{k\in\mathbb{Z}} \big|\frac{u(k)}{\lambda}\big|^{p(k)}\leq 1\big\}. $$ We also consider the space $$ l^{\infty}=\big\{u:\mathbb{Z}\to\mathbb{R}; |u|_{\infty} :=\sup_{k\in\mathbb{Z}}|u(k)|<\infty\big\}. $$ We recall some useful properties of the space $l^{p(\cdot)}$. Firstly, by classical results of functional analysis we know that, for all $1
0$ such that \begin{equation}\label{fi1} (\phi_{p(k)}(x)-\phi_{p(k)}(y))(x-y)\geq c|x-y|^{p(k)}\quad \text{for all } x,y\in\mathbb{R},\; k\in\mathbb{Z}, \text{ if } p(k)\geq 2; \end{equation} or \begin{equation}\label{fi2} (\phi_{p(k)}(x)-\phi_{p(k)}(y))(x-y)\geq c(|x|+|y|)^{p(k)-2}|x-y|^2\quad \text{for all } x,y\in\mathbb{R},\; k\in\mathbb{Z}, \end{equation} if $1
1 \text{ implies } |u|_{p(\cdot)}^{p^{-}}\leq \rho_{p(\cdot)}(u)
\leq |u|_{p(\cdot)}^{p^{+}};
\\
\label{eq5}
|u|_{p(\cdot)}<1 \text{ implies } |u|_{p(\cdot)}^{p^{+}}
\leq \rho_{p(\cdot)}(u)\leq |u|_{p(\cdot)}^{p^{-}};
\\
\label{eq6}
|u_n|_{p(\cdot)}\to 0 \text{ if and only if } \rho_{p(\cdot)}(u_n)\to 0
\text{ as } n\to\infty;
\end{gather}
\end{proposition}
On the other hand, it is useful to introduce other space which is still
infinite dimensional but compactly embedded into $l^{p(\cdot)}$.
The main reason is that such compact embedding is a key tool to prove
Palais-Smale condition. We set
\begin{gather*}
X=\{u:\mathbb{Z}\to\mathbb{R}; \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
<\infty\};\\
\|u\|_{p(\cdot)}:=\inf\big\{\lambda>0: \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
|\frac{u(k)}{\lambda}|^{p(k)}\leq 1\big\}.
\end{gather*}
\begin{remark}\label{rem2} \rm
We deduce that
\begin{equation}\label{norm}
|u|_{p(\cdot)}\leq \big(\frac{a_0}{p^{+}}\big)^{-1}\|u\|_{p(\cdot)}
\quad \text{for all } u\in X.
\end{equation}
\end{remark}
Moreover, we have the following properties of the above norm,
which are similarly to the properties of the norm $|\cdot|_{p(\cdot)}$.
\begin{proposition}\label{prop2}
Let $u\in X$, $(u_n)\subset X$ and $p^{+}<+\infty$. Then the following properties
hold
\begin{gather}\label{eq7}
\|u\|_{p(\cdot)}>1 \text{ implies } \|u\|_{p(\cdot)}^{p^{-}}
\leq \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
\leq \|u\|_{p(\cdot)}^{p^{+}};
\\
\label{eq8}
\|u\|_{p(\cdot)}<1\text{ implies }\|u\|_{p(\cdot)}^{p^{+}}
\leq \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
\leq \|u\|_{p(\cdot)}^{p^{-}};
\\
\label{eq9}
\|u_n\|_{p(\cdot)}\to 0 \text{ if and only if }
\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)} \to 0 \text{ as } n\to\infty.
\end{gather}
\end{proposition}
\begin{proof}
For \eqref{eq7}, let $\|u\|_{p(\cdot)}>1$. Then
\begin{align*}
\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
&= \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u(k)}{\|u\|_{p(\cdot)}}\times \|u\|_{p(\cdot)}\big|^{p(k)}\\
&\geq \|u\|_{p(\cdot)}^{p^{-}}\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u(k)}{\|u\|_{p(\cdot)}}\big|^{p(k)}\\
&= \|u\|_{p(\cdot)}^{p^{-}}.
\end{align*}
On the other hand, we have
\begin{align*}
\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
&=\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u(k)}{\|u\|_{p(\cdot)}}\times \|u\|_{p(\cdot)}\big|^{p(k)}\\
&\leq \|u\|_{p(\cdot)}^{p^{+}}\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u(k)}{\|u\|_{p(\cdot)}}\big|^{p(k)}\\
&= \|u\|_{p(\cdot)}^{p^{+}}.
\end{align*}
Thus,
$$
\|u\|_{p(\cdot)}^{p^{-}}\leq \sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u(k)|^{p(k)}
\leq \|u\|_{p(\cdot)}^{p^{+}};
$$
The proof of \eqref{eq8} is similar to that for \eqref{eq7}.
For \eqref{eq9} we consider two cases.
Case 1. $ \|u_n\|_{p(\cdot)}\to 0$ as $n\to\infty$, then
\begin{align*}
\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)}
&=\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u_n(k)}{\|u_n\|_{p(\cdot)}}\times \|u_n\|_{p(\cdot)}\big|^{p(k)}\\
&\leq \|u_n\|_{p(\cdot)}^{\alpha}\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u_n(k)}{\|u_n\|_{p(\cdot)}}\big|^{p(k)}\\
&= \|u_n\|_{p(\cdot)}^{\alpha},
\end{align*}
where $\alpha= \begin{cases}
p^{+} &\text{if } \|u_n\|_{p(\cdot)}\geq 1 \\
p^{-} &\text{if } \|u_n\|_{p(\cdot)}<1.
\end{cases}$
So, $\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)} \to 0$
as $n\to\infty$.
Case 2. $\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)} \to 0$
as $n\to\infty$, then
\begin{align*}
\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)}
&=\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u_n(k)}{\|u_n\|_{p(\cdot)}}\times \|u_n\|_{p(\cdot)}\big|^{p(k)}\\
&\geq \|u_n\|_{p(\cdot)}^{\alpha}\sum_{k\in\mathbb{Z}}\frac{a(k)}{p(k)}
\big|\frac{u_n(k)}{\|u_n\|_{p(\cdot)}}\big|^{p(k)}\\
&= \|u_n\|_{p(\cdot)}^{\alpha},
\end{align*}
where $\alpha= \begin{cases}
p^{-} &\text{if } \|u_n\|_{p(\cdot)}\geq 1 \\
p^{+} &\text{if } \|u_n\|_{p(\cdot)}<1.
\end{cases}$
So, $ \|u_n\|_{p(\cdot)}\to 0$ as $n\to\infty$.
\end{proof}
Next we have a Discrete H\"{o}lder type inequality.
\begin{proposition}[{\cite[Theorem 2.1 ]{6}}] \label{prop6}
Let $u\in l^{p(\cdot)}$ and $v\in l^{q(\cdot)}$ be such that
$\frac{1}{p(k)}+\frac{1}{q(k)}=1$ for all $k\in\mathbb{Z}$. Then
\begin{equation}\label{hol}
\sum_{k\in\mathbb{Z}}|uv|
\leq\Big(\frac{1}{p^{-}}+\frac{1}{q^{-}}\Big)|u|_{p(\cdot)}|v|_{q(\cdot)}.
\end{equation}
\end{proposition}
Moreover, we have the following relations
\begin{equation}\label{rel}
\frac{1}{p^{-}}+\frac{1}{q^{+}}=1;\ \frac{1}{p^{+}}+\frac{1}{q^{-}}=1.
\end{equation}
\begin{proposition} \label{prop7}
$(X,\|\cdot\|_{p(\cdot)})$ is a reflexive Banach space and the embedding
$X\hookrightarrow l^{p(\cdot)}$ is compact.
\end{proposition}
\begin{proof}
Clearly, $X$ is a Banach space. Next, we prove that it is uniformly convex.
Indeed, we consider the $\sigma$-finite measure space $(\mathbb{Z},\mu)$
where the measure is defined by
$$
\mu(\emptyset)=0\quad \text{and}\quad \mu(S)
=\sum_{k\in S}\frac{a(k)}{p(k)}\quad
\text{for all}\ S\subseteq\mathbb{Z},\; S\neq\emptyset.
$$
Then $X$ is the usual $l^{p(\cdot)}$ space defined on $(\mathbb{Z},\mu)$.
The Clarkson theorem (see \cite{14}) implies that $X$ is uniformly convex,
hence reflexive by the Milman-Pettis theorem (see \cite{14}).
By Remark \ref{rem2} we have that the embedding $X\hookrightarrow l^{p(\cdot)}$
is continuous.
Finally, we prove that $X\hookrightarrow l^{p(\cdot)}$ is compact.
Let $(u_n)$ be a bounded sequence in $X$; i.e., there exists $M>0$ such that
\begin{equation}\label{emb}
\|u_n\|_{p(\cdot)}^{p^{+}}<\frac{M}{p^{+}}\quad \text{for all } n\in\mathbb{N},\; p^{+}>1.
\end{equation}
By reflexivity, passing to a subsequence we have $u_n\rightharpoonup u$
in $X$ for some $u\in X$. We may assume $u=0$. In particular, $u_n(k)\to 0$
as $n\to\infty$ for all $k\in\mathbb{Z}$. By the coercivity of $a(\cdot)$
and the relation \eqref{eq3}, for all $\varepsilon>0$, we can find $h\in\mathbb{N}$
such that
$$
\frac{a(k)}{p(k)}>\frac{1+M}{\varepsilon p^{+}}\quad \text{for all } |k|>h.
$$
By continuity of the finite sum, there exists $\nu\in\mathbb{N}$ such that
$$
\sum_{|k|\leq h}|u_n(k)|^{p(k)}<\frac{\varepsilon}{1+M}\ \text{for all}\ n>\nu.
$$
Hence, for all $n>\nu$ we have
\begin{align*}
\sum_{k\in \mathbb{Z}}|u_n(k)|^{p(k)}
&=\sum_{|k|\leq h}|u_n(k)|^{p(k)}+\sum_{|k|> h}|u_n(k)|^{p(k)}\\
&\leq\frac{\varepsilon}{1+M}+\frac{\varepsilon p^{+}}{1+M}
\sum_{|k|> h}\frac{a(k)}{p(k)}|u_n(k)|^{p(k)}\\
&\leq \frac{\varepsilon}{1+M}+\frac{\varepsilon p^{+}}{1+M}
\|u_n\|_{p(\cdot)}^{p^{+}}\quad (\text{see \eqref{eq7}})\\
&<\frac{\varepsilon}{1+M}\Big(1+p^{+}\frac{M}{p^{+}}\Big)\quad
(\text{see \eqref{emb}})\\
&= \varepsilon.
\end{align*}
Thus, $u_n\to 0$ in $l^{p(\cdot)}$.
\end{proof}
In the proof of our theorem, we will need the following technical result.
\begin{proposition}\label{c1}
If $S$ is a compact subset of $l^{p(\cdot)}$, then for all $\varepsilon>0$
there exists $h\in\mathbb{N}$ such that
$$
\Big(\sum_{|k|>h}|u(k)|^{p^{-}}\Big)^{1/p^{-}}<\varepsilon\quad \text{for all }
u\in l^{p(\cdot)}.
$$
\end{proposition}
\begin{proof}
Arguing by contradiction, assume that there exists $\varepsilon>0$ and a
sequence $(u_n)\subseteq S$ such that
$$
\Big(\sum_{|k|>n}|u_n(k)|^{p^{-}}\Big)^{1/p^{-}}>\varepsilon\quad
\text{for all } n\in\mathbb{N}.
$$
Due to compactness of $S$, passing if necessary to a subsequence we may
assume that $u_n\to u$ for some $u\in l^{p(\cdot)}$.
So, there exists $h\in\mathbb{N}$ such that
\[
\Big(\sum_{|k|>h}|u(k)|^{p^{-}}\Big)^{1/p^{-}}<\frac{\varepsilon}{2}.
\]
Moreover, there exists $n\geq h$ such that
\begin{align*}
|u_n-u|_{p(\cdot)}
&\leq \Big(\sum_{|k|> n}|u_n(k)-u(k)|^{p(k)}\Big)^{1/p^{-}}\\
&\leq \Big(\sum_{|k|> n}|u_n(k)-u(k)|^{p^{-}}\Big)^{1/p^{-}}\\
&< \frac{\varepsilon}{2}\quad (\text{see \eqref{eq4}}).
\end{align*}
By the classical Minkowski inequality and the above inequalities, we have
\begin{align*}
\varepsilon
&< \Big(\sum_{|k|>n}|u_n(k)|^{p^{-}}\Big)^{1/p^{-}}\\
&= \Big(\sum_{|k|>n}|u_n(k)-u(k)+u(k)|^{p^{-}}\Big)^{1/p^{-}}\\
&\leq \Big(\sum_{|k|> n}|u_n(k)-u(k)|^{p^{-}}\Big)^{1/p^{-}}
+\Big(\sum_{|k|>n}|u(k)|^{p^{-}}\Big)^{1/p^{-}}\\
&< \frac{\varepsilon}{2}+\frac{\varepsilon}{2}
= \varepsilon,
\end{align*}
which is a contradiction.
\end{proof}
We denote by $(X^{\ast},\|\cdot\|_{p^{\ast}(\cdot)}))$ the topological
dual of $(X,\|\cdot\|_{p(\cdot)})$.
We recall that a functional $J\in C^1(X,\mathbb{R})$, is said to satisfy
the \emph{Palais-Smale condition at the level c}, where $c$ is a given
real number, ($(PS)_c$ for short) if every sequence $(u_n)$ in $X$
satisfying $J(u_n)\to c$ and $J'(u_n)\to 0$ in $X^{\ast}$ contains
a convergent subsequence. Such condition is an essential hypothesis
in the following mountain pass theorem, due to Ambrosetti and Rabinowitz \cite{2}.
\begin{theorem}\label{th2}
Let $ X$ be a real Banach space and assume that $J\in C^1(X,\mathbb{R})$
satisfies the following geometric conditions:
\begin{itemize}
\item[(H1)] there exist two numbers $R>0$ and $c_0>0$ such that
$J(u)\geq c_0$ for all $u\in X$ with $\|u\|=R$;
\item[(H2)] $J(0) 0 \text{ and } J'(u_n)\to 0 \quad \text{in }
X^{\ast}, \text{ as } n\to\infty.
\]
We also we know by Lemma \ref{lem3} that $J$ satisfies $(PS)_c$ condition.
Then there exists a subsequence, still denoted by $(u_n)$, and $u_0\in X$
such that $(u_n)$ converges to $u_0$ in $X$. So, we have $J(u_0)=c>0$
and $J'(u_0)=0$ which implies that $u_0$ is a critical point of $J$
and consequently a solution of \eqref{eq1}(see Remark \ref{rem3}).
Moreover, since $J(u_0)>0$ we have $u_0\neq 0$.
Hence, we conclude that $u_0$ is a nontrivial solution of \eqref{eq1}.
\end{proof}
\begin{theorem}\label{th3}
Assume that the hypotheses of Theorem \ref{th1} are satisfied and in addition,
the following condition holds
\begin{equation}\label{eqth2}
f(k,t)\geq 0\ \text{for all}\ t<0\ \text{and all}\ k\in\mathbb{Z}\,.
\end{equation}
Then \eqref{eq1} has a positive homoclinic solution.
\end{theorem}
\begin{proof}
By Theorem \ref{th1} we know that \eqref{eq1} has a nontrivial homoclinic solution.
Now, arguing by contradiction, suppose that $u(k_0)<0$ for some
$k_0\in\mathbb{Z}$ and let $k_1$ be such that
$u(k_1)=\min\{u(k),k\in\mathbb{Z}\}<0$. In consequence
$\Delta\phi_{p(k_1-1)}(\Delta u(k_1-1))\geq 0$, which by equation \eqref{eq1}
implies that
$$
f(k_1,u(k_1))=-\Delta\phi_{p(k_1-1)}(\Delta u(k_1-1))+a(k_1)\phi_{p(k_1)}(u(k_1))<0,
$$
a contradiction with \eqref{eqth2}. So, $u(k)\geq 0$ for all $k\in\mathbb{Z}$.
Next, we prove that $u(k)>0$ for all $k\in\mathbb{Z}$.
Indeed, arguing by contradiction, assume that $u(k_2)=0$ for some
$k_2\in\mathbb{Z}$. By \eqref{eq1} we have
$$
\phi_{p(k_2)}(\Delta u(k_2))=\phi_{p(k_2-1)}(\Delta u(k_2-1))
$$
(recall that $f(k,0)=0)$. Note that if $u(k_2+1)=0$ or $u(k_2-1)=0$,
the solution is identically zero by a recursion, which is a contradiction
with $u\neq 0$. So, $u(k)>0$ for all $k\in\mathbb{Z}$.
\end{proof}
\begin{thebibliography}{99}
\bibitem{1} R. P. Agarwal, K. Perera, D. O'Regan;
\emph{Multiple positive solutions of singular and nonsingular discrete problems
via variational methods}, Nonlinear Anal. \textbf{58} (2004) 69-73.
\bibitem{2} A. Ambrosetti, P. H. Rabinowitz;
\emph{Dual variational methods in critical point theory
and applications}, J. Functional Analysis \textbf{14} (1973) 349-381.
\bibitem{3} A. Cabada, A. Iannizzotto, S. Tersian;
\emph{Multiple solutions for discrete boundary value problems},
J. Math. Anal. Appl. \textbf{356} (2009) 418-428.
\bibitem{4} A. Cabada, C. Li, S. Tersian;
\emph{On homoclinic solutions of a semilinear $p$-Laplacian difference
equation with periodic coefficients}, Adv. Difference Equ. \textbf{2010}
(2010) 17 pp.
\bibitem{14} M. Fabian, P. Habala, P. H\'{a}jek, V. Montesinos, V. Zizler;
\emph{Banach space theory}, Springer (2011).
\bibitem{5} H. Fang, D. Zhao;
\emph{Existence of nontrivial homoclinic orbits for fourth-order difference
equations}, Appl. Math. Comput. \textbf{214} (2009) 163-170.
\bibitem{6} A. Guiro, B. Kon\'e, S. Ouaro;
\emph{Weak homoclinic solutions of anisotropic difference equation with variable
exponents}, Adv. Differ. Equ. \textbf{154} (2012) 1-13.
\bibitem{7} A. Iannizzotto, V. R\u{a}dulescu;
\emph{Positive homoclinic solutions for the $p$-Laplacian with a coercive
weight function}, Differential and Integral Equations \textbf{27},
Numbers 1-2 (2014), 35-44.
\bibitem{13} A. Iannizzotto, S. Tersian;
\emph{Multiple homoclinic solutions for the discrete $p$-Laplacian via
critical point theory}, J. Math. Anal. Appl., \textbf{403} (2013), 173-182.
\bibitem{8} A. Krist\'{a}ly, M. Mih\u{a}ilescu, V. R\u{a}dulescu, S. Tersian;
\emph{Spectral estimates for a nonhomogeneous difference problem},
Comm. Contemp. Math. \textbf{12} (2010), 1015-1029.
\bibitem{9} M. Ma, Z. Guo;
\emph{Homoclinic orbits for second order self-adjoint difference equations},
J. Math. Anal. Appl. \textbf{323} (2005), 513-521.
\bibitem{10} M. Mih\u{a}ilescu, V. R\u{a}dulescu, S. Tersian;
\emph{Eigenvalue problems for anisotropic discrete boundary value problems},
J. Difference Equ. Appl \textbf{15} (2009), 557-567.
\bibitem{11} M. Mih\u{a}ilescu, V. R\u{a}dulescu, S. Tersian;
\emph{Homoclinic solutions of difference equations with variable exponents},
Topol. Methods Nonlinear Anal. \textbf{38} (2011), 277-289.
\bibitem{12} H. Poincar\'e;
\emph{Les M\'ethodes Nouvelles de la M\'ecanique C\'eleste},
Gauthier-Villars, Paris, 1899.
\end{thebibliography}
\end{document}