IJMMSInternational Journal of Mathematics and Mathematical Sciences1687-04250161-1712Hindawi Publishing Corporation90745310.1155/2010/907453907453Research ArticleExistence of Multiple Solutions of a Second-Order Difference Boundary Value ProblemZhengBoXiaoHuafengBohnerMartinCollege of Mathematics and Information SciencesGuangzhou UniversityGuangzhou 510006Chinagzhu.edu.cn201093201020100307200905022010070320102010Copyright © 2010This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper studies the existence of multiple solutions of the second-order difference boundary value problem Δ2u(n1)+V(u(n))=0, n(1,T), u(0)=0=u(T+1). By applying Morse theory, critical groups, and the mountain pass theorem, we prove that the previous equation has at least three nontrivial solutions when the problem is resonant at the eigenvalue λk  (k2) of linear difference problem Δ2u(n1)+λu(n)=0, n(1,T), u(0)=0=u(T+1) near infinity and the trivial solution of the first equation is a local minimizer under some assumptions on V.

1. Introduction

Let , , and be the sets of real numbers, natural numbers, and integers, respectively. For any a,b, ab, define (a,b)={a,a+1,,b}.

Consider the second-order difference boundary value problem (BVP)

Δ2u(n-1)+V(u(n))=0,n(1,T),u(0)=0=u(T+1), where VC2(,) and Δ denotes the forward difference operator defined by Δu(n)=u(n+1)-u(n), Δ2u(n)=Δ(Δu(n)).

By a solution u of the BVP (1.1), we mean a real sequence {u(n)}n=0T+1(=(u(0),u(1),,u(T+1))) satisfying the BVP (1.1). For u={u(n)}n=0T+1 with u(0)=0=u(T+1), we say that u0 if there exists at least one n(1,T) such that u(n)0. We say that u is positive (and write u>0) if for all n(1,T), u(n)>0, and similarly, u is negative (u<0) if for all n(1,T), u(n)<0. The aim of this paper is to obtain the existence of multiple solutions of the BVP (1.1) and analyse the sign of solutions.

Recently, a few authors applied the minimax methods to examine the difference boundary value problems. For example, in , Agarwal et al. employed the Mountain Pass Lemma to study the following BVP:

Δ2u(n-1)+f(n,u(n))=0,n(1,T),u(0)=0=u(T+1) and obtained the existence of multiple positive solutions, where f may be singular at u=0. In , Jiang and Zhou employed the Mountain Pass Lemma together with strongly monotone operator principle, to study the following difference BVP:

Δ2u(n-1)+f(n,u(n))=0,n(1,T),u(0)=0=Δu(T) and obtained existence and uniqueness results, where f:(1,T)× is continuous. In , Cai and Yu employed the Linking Theorem and the Mountain Pass Lemma to study the following difference BVP:

Δ(p(n)(Δu(n-1))δ)+q(n)uδ(n)=f(n,u(n)),n(1,T),Δu(0)=A,u(T+1)=B and obtained the existence of multiple solutions, where δ>0 is the ratio of odd positive integers, {p(n)}n=1T+1 and {q(n)}n=1T are real sequences, p(n)0 for all n(1,T+1), and A, B are two given constants, f:(1,T)× is continuous.

Although applications of the minimax methods in the field of the difference BVP have attracted some scholarly attention in the recent years, efforts in applying Morse theory to the difference BVP are scarce. The main purpose of this paper is to develop a new approach to the BVP (1.1) by using Morse theory. To this end, we first consider the following linear difference eigenvalue problem:

Δ2u(n-1)+λu(n)=0,n(1,T),u(0)=0=u(T+1).

On the above eigenvalue problem, the following results hold; see .

Proposition 1.1.

The eigenvalues of (1.5) are λ=λl=4sin2lπ2(T+1),l=1,2,,T, and the corresponding eigenfunction with λl is ϕl(n)=sin(lπn/(T+1)), l=1,2,,T.

Remark 1.2.

(1) The set of functions {ϕl(n),l=1,2,,T} is orthogonal on (1,T) with respect to the weight function r(n)1, that is, n=1T(ϕl(n),ϕj(n))=0lj. Moreover, for each l(1,T), n=1Tsin2(lπn/(T+1))=(T+1)/2.

(2) It is easy to see that ϕ1 is positive and ϕl changes sign for each l(2,T), that is, {n:ϕl(n)>0} and {n:ϕl(n)<0}.

For (1.1), we assume that

V(0)=V(0)=0,V():=lim|t|V(t)t=λk, where λk is an eigenvalue of (1.5). Hence the BVP (1.1) has a trivial solution u0. And we say that BVP (1.1) is resonant at infinity if (1.9) holds.

Let

W-=span{ϕ1,ϕ2,,ϕk-1},W0=span{ϕk},W+=span{ϕk+1,ϕk+2,,ϕT}. Let G(t)=0tG(s)ds=V(t)-(λk/2)t2. By (1.9) we have

lim|t|G(t)t=0. Assume that the following conditions on G(t) hold.

If um such that vm/um1, then there exist δ>0 and M such that

±n=1T(G(um(n)),vm(n))δ,mM, where um=vm+wm, vmW0, wmW:=W+W-.

The main result of this paper is as follows.

Theorem 1.3.

Let (1.8), (1.9) hold and

V(t)>0 for all t,

V(0)<λ1

hold. Then the BVP (1.1) has at least three nontrivial solutions, with one positive solution and one negative solution, in each of the following cases:

(G+) and k2;

(G-) and k3.

To the author’s best knowledge, only Bin et al.  deal with the existence and multiplicity of nontrivial periodic solutions for asymptotically linear resonant difference problem by the aid of Su . In , G satisfies

|G(z)|c1|z|s+c2,limvinfvW01v2sG(z)4β2δT, where c1>0, c2>0, s(0,1),β=c1T(1-s)/2, δ>0. In , the authors obtained the existence of one nontrivial periodic solution. Notice that (1.13) implies that (1.11) holds; however, (G±) is not covered by (1.14). In fact, conditions (1.13) and (1.14) are borrowed from . The conditions in Theorem 1.3 coincide with the assumptions of Theorem 1 in . The aim of this paper is to develop a new approach to study the discrete systems by using Morse theory, minimax theorems, and some analysis technique. We wish to have some breakthrough points with the aid of the method of discretization.

The remaining part of this paper proceeds as follows. In the next section, we establish the variational framework of the BVP (1.1) and collect some results which will be used in the proof of Theorem 1.3. In Section 3, we give the proof of Theorem 1.3. Finally, in Section 4, we give an example to illustrate our main result and summarize conclusions and future directions.

2. Variational Framework and Auxiliary Results

Let

E={u:u={u(n)}n=0T+1  with  u(0)=0=u(T+1)}.E can be equipped with the norm · and the inner product ·,· as follows:

u=(n=0T|Δu(n)|2)1/2,uE,u,v=n=0T(Δu(n),Δv(n)),u,vE, where |·| denotes the Euclidean norm in and (·,·) denotes the usual scalar product in . It is easy to see that (E,·,·) is a Hilbert space. Consider the functional defined on E by

J(u)=12n=0T|Δu(n)|2-n=1TV(u(n)). We claim that if uE is a critical point of J, then u is precisely a solution of the BVP (1.1). Indeed, for every u,vE, we have

J(u),v=n=0T(Δu(n),Δv(n))-n=1T(V(u(n)),v(n))=-n=1T(Δ2u(n-1)+V(u(n)),v(n)).

So, if J(u)=0, then we have

n=1T(Δ2u(n-1)+V(u(n)),v(n))=0. Since vE is arbitrary, we obtain

Δ2u(n-1)+V(u(n))=0,n(1,T). Therefore, we reduce the problem of finding solutions of the BVP (1.1) to that of seeking critical points of the functional J in E.

According to Proposition 1.1 and Remark 1.2, E can be decomposed as E=W-W0W+. For all uE, denote u=w0+w++w- with w0W0, w+W+, and w-W-, then we have the following Wirtinger type inequalities:

λ1n=1T(u(n),u(n))u2λTn=1T(u(n),u(n)),uE,λ1n=1T(w-(n),w-(n))w-2λk-1n=1T(w-(n),w-(n)),w-W-,λk+1n=1T(w+(n),w+(n))w+2λTn=1T(w+(n),w+(n)),w+W+, see  for details.

Now we collect some results on Morse theory and the minimax methods.

Let E be a real Hilbert space and JC1(E,). Denote

Jc={uE:J(u)c},𝒦c={uE:J'(u)=0,J(u)=c} for c. The following is the definition of the Palais-Smale condition ((PS) condition).

Definition 2.1.

The functional J satisfies the (PS) condition if any sequence {um}E such that {J(um)} is bounded and J'(um)0 as m has a convergent subsequence.

In , Cerami introduced a weak version of the (PS) condition as follows.

Definition 2.2.

The functional J satisfies the Cerami condition ((C) condition) if any sequence {um}E such that {J(um)} is bounded and (1+um)J(um)0 as m has a convergent subsequence.

If J satisfies the (PS) condition or the (C) condition, then J satisfies the following deformation condition which is essential in Morse theory (cf. [9, 10]).

Definition 2.3.

The functional J satisfies the (Dc) condition at the level c if for any ϵ¯>0 and any neighborhood 𝒩 of 𝒦c, there are ϵ>0 and a continuous deformation η:[0,1]×EE such that

η(0,u)=u for all uE;

η(t,u)=u for all uJ-1([c-ϵ¯,c+ϵ¯]);

J(η(t,u)) is nonincreasing in t for any uE;

η(1,Jc+ϵ𝒩)Jc-ϵ.

J satisfies the (D) condition if J satisfies the (Dc) condition for all c.

Let u0 be an isolated critical point of J with J(u0)=c, and let U be a neighborhood of u0, the group

Cq(J,u0):=Hq(JcU,JcU{u0}),q, is called the qth critical group of J at u0, where Hq(A,B) denotes the qth singular relative homology group of the pair (A,B) over a field F, which is defined to be quotient Hq(A,B)=Zq(A,B)/Bq(A,B), where Zq(A,B) is the qth singular relative closed chain group and Bq(A,B) is the qth singular relative boundary chain group.

Let 𝒦={uE:J'(u)=0}. If J(𝒦) is bounded from below by a and J satisfies the (Dc) condition for all ca, then the group

Cq(J,):=Hq(E,Ja),q, is called the qth critical group of J at infinity .

Assume that #𝒦< and J satisfies the (D) condition. The Morse-type numbers of the pair (E,Ja) are defined by

Mq=Mq(E,Ja)=u𝒦dimCq(J,u), and the Betti numbers of the pair (E,Ja) are

βq:=dimCq(J,). By Morse theory [12, 13], the following relations hold:

j=0q(-1)q-jMjj=0q(-1)q-jβj,q,q=0Mq=q=0βq.

Thus, if Cq(J,)0, for some k, then there must exist a critical point u of J with Cq(J,u)0, which can be rephrased as follows.

Proposition 2.4.

Let E be a real Hilbert space and JC2(E,). Assume that #𝒦< and that J satisfies the (D) condition. If there exists some q such that Cq(J,)0, then J must have a critical point u with Cq(J,u)0.

In order to prove our main result, we need the following result about the critical group on Cq(J,).

Proposition 2.5.

Let the functional J:E be of the form J(u)=12Au,u+Q(u), where A:EE is a self-adjoint linear operator such that 0 is isolated in σ(A), the spectrum of A. Assume that QC1(E,) satisfies limuQ(u)u=0. Denote V:=kerA, W:=V=W+W-, where W+ (W-) is the subspace of E on which A is positive (negative) definite. Assume that μ:=dimW-, ν:=dimV0 are finite and that J satisfies the (D) condition. Then Cq(J,)δq,k±F,q, provided that J satisfies the angle conditions at infinity.

there exist M>0 and α(0,1) such that

±J'(u),v0for  u=v+w,uM,wαu, where k+=μ, k-=μ+ν, vV, and wW.

Remark 2.6.

Conditions (2.16) and (2.17) imply that J is asymptotically quadratic. Bartsch and Li  introduced the notion of critical groups at infinity and proved that if J satisfied some angle properties at infinity, the critical groups can be completely figured out. Proposition 2.5 is a slight improvement of [11, Proposition 3.10] by Su and Zhao . There are many other papers considering concrete problems by computing the critical groups at infinity with different methods, for example, see .

We will use the Mountain Pass Lemma (cf. [12, 18]) in our proof.

Let Bρ denote the open ball in E about 0 of radius r and let Bρ denote its boundary.

Theorem 2.7 (mountain pass lemma).

Let E be a real Banach space and JC1(E,) satisfying the (PS) condition. Suppose J(0)=0 and that

there are constants ρ>0,a>0 such that J|Bρa>0,

there is a u0EBρ such that J(u0)0,

then J possesses a critical value ca. Moreover c can be characterized as c=infhΓsups[0,1]J(h(s)), where Γ={hC([0,1],E)h(0)=0,h(1)=u0}.

Definition 2.8 (mountain pass point).

An isolated critical point u of J is called a mountain pass point, if C1(J,u)0.

The following result is useful in computing the critical group of a mountain pass point; see [13, 19] for details.

Theorem 2.9.

Let E be a real Hilbert space. Suppose that JC2(E,) has a mountain pass point u, and that J′′(u) is a Fredholm operator with finite Morse index, satisfying J′′(u0)0,0σ(J(u0))dimker(J′′(u0))=1, then Cq(J,u0)δq,1F,q.

3. Proof of Theorem <xref ref-type="statement" rid="thm1.1">1.3</xref>

We give the proof of Theorem 1.3 in this section. Firstly, we prove that the functional J satisfies the (C) condition (Lemma 3.1) and compute the critical group Cq(J,) (Lemma 3.2). Then, we employ the cut-off technique and the Mountain Pass Lemma to obtain two critical points u+,u- of J and compute the critical groups Cq(J,u+) and Cq(J,u-) (Lemmas 3.3 and 3.4). Finally, we prove Theorem 1.3.

Rewrite the functional J as

J(u)=12n=0T|Δu(n)|2-λk2n=1T|u(n)|2-n=1TG(u(n)),uE.

Lemma 3.1.

Let (1.8) and (1.9) hold. If G satisfies (G±), then the functional J satisfies the (C) condition.

Proof.

We only prove the case where (G+) holds. Let {um}E such that J(um)c,(1+um)J(um)0asm. Then for all φE, we have J(um),φ=um,φ-λkn=1T(um(n),φ(n))-n=1T(G(um(n)),φ(n)). Denote um=vm+wm++wm- with vmW0, wm+W+ and wm-W-. Since E is a finite-dimensional Hilbert space, it suffices to show that {um} is bounded. Suppose that {um} is unbounded. Passing to a subsequence we may assume that um as m.

By (1.11), for any ϵ>0, there exists b such that |G(t)|ϵ|t|+b,t. Let φ=wm+ in (3.3). Then by (2.7), (2.9), and (3.4), we have c1wm+2:=(1-λkλk+1)wm+2wm+2-λkn=1T(wm+(n),wm+(n))=n=1T(G'(um(n)),wm+(n))+J'(um),wm+wm++n=1T(ϵ|um(n)|+b)|wm+(n)|wm++ϵλ1λk+1umwm++bTλk+1wm+:=c2wm++c3umwm+, where c1=1-λkλk+1>0,c2=1+bTλk+1,c3=ϵλ1λk+1. And since ϵ>0 is arbitrary, we have wm+um0asm. Similarly, let φ=wm- in (3.3), by (2.8) and (3.4) we get -c4wm-2=:(1-λkλk-1)wm-2wm-2-λkn=1T(wm-(n),wm-(n))=n=1T(G(um(n)),wm-(n))+J(um),wm--wm--n=1T(ϵ|um(n)|+b)|wm-(n)|-wm--ϵλ1umwm--bTλ1wm-:=-c5wm--c6umwm-, where c4=λkλk-1-1>0,c5=1+bTλ1,c6=ϵλ1. And hence we also have wm-um0asm. By (3.7) and (3.10), we have wmum0,vmum1asm. By (G+), there exist δ>0 and M such that n=1T(G(um(n)),vm(n))δ,mM. This implies that J(um),vm=-n=1T(G(um(n)),vm(n))-δ,mM, and hence J(um)umJ(um)vm|J(um),vm|δ,mM, which is a contradiction to (3.2). Thus {um} is bounded. The proof is complete.

Lemma 3.2.

Let (1.8) and (1.9) hold. Then

Cq(J,)δq,kF provided that (G+) holds;

Cq(J,)δq,k-1F provided that (G-) holds.

Proof.

We only prove the case (1). Define a bilinear function a(u,v)=λkn=1T(u(n),v(n)),u,vE. Then by (2.7) we have |a(u,v)|λkλ1uv. And hence there exists a unique continuous bounded linear operator K:EE such that Ku,v=λkn=1T(u(n),v(n)). Since Ku,u for all uE, we can conclude that K is a self-adjoint operator and J(u)=12(I-K)u,u-n=1TG(u(n)). Then J has the form (2.16) with Q(u)=-n=1TG(u(n)), and (1.11) implies that (2.17) holds. Let A=I-K. Then kerA=W0=span{ϕk}. Next we show that (G+) implies that the angle condition (AC-) at infinity holds.

If not, then for any m and each αm=1/m, there exists um=vm+wmW0(W+W-) with vmW0, wmW+W- such that umm,wm1mum,J'(um),vm>0. On the other hand, (3.20) implies um,vmum1asm. Thus, by (G+) there exist δ>0 and M such that n=1T(G(um(n)),vm(n))δ,mM. Therefore, J(um),vm=-n=1T(G(um(n)),vm(n))-δ,mM, which is a contradiction to (3.21). Consequently (AC-) holds and by Lemma 3.1 and Proposition 2.5, Cq(J,)=δq,kF. Similarly, we can prove that (2) holds.

In order to obtain a mountain pass point, we need the following lemmas.

Lemma 3.3.

Let V+(t)={V(t),t0,0,t0,V-(t)={V(t),t0,0,t0, and V±(t)=0tV'±(s)ds. If lim|t|V(t)t=αλ1, then the functional J±(u)=12n=0T|Δu(n)|2-n=1TV±(u(n)) satisfies the (PS) condition.

Proof.

We only prove the case (J+). Let {um}E such that J(um)c,J'+(um)0     as m. Since E is a finite-dimensional space, it suffices to show that {um} is bounded in E. Suppose that {um} is unbounded. Passing to a subsequence we may assume that um and for each n, either |um(n)| or {um(n)} is bounded.

Noticing that for all φE, J'+(um),φ=um,φ-n=1T(V'+(um(n)),φ(n)).

Denote wm:=um/um, for a subsequence, wm converges to some w with w=1. By (3.29), we have J'+(um),φum=wm,φ-n=1T(V'+(um(n))um,φ(n)). If |um(n)|, then limmV'+(um(n))um(n)wm(n)=αw+(n), where w+(n)=max{w(n),0} with n(1,T). If {um(n)} is bounded, then limmV'+(um(n))um=0,w(n)=0. Since w0, there is an n for which |um(n)|. So passing to the limit in (3.30), we have n=0T(Δw(n),Δφ(n))-αn=1T(w+(n),φ(n))=0. This implies that w0 satisfies Δ2w(n-1)+αw+(n)=0,n(1,T),w(0)=0=w(T+1).

Now, we claim that w(n)>0,n(1,T).

In fact, let w(n0)=min{w(n):n(1,T)}. We only need to prove w(n0)>0. If not, assume that w(n0)0. Then by (3.34), we have Δ2w(n0-1)=0 and hence w(n0-1)=w(n0)=w(n0+1). By induction, it is easy to get w(n)=0 for all n(1,T) which is a contradiction to w0 and hence (3.35) holds.

On the other hand, by Proposition 1.1 and Remark 1.2, we see that only the eigenfunction corresponding to the eigenvalue λ1 is positive, which is a contradiction to αλ1. The proof is complete.

Lemma 3.4.

Under the conditions of Theorem 1.3, the functional J+ has a critical point u+>0 and Cq(J+,u+)δq,1F; the functional J- has a critical point u-<0 and Cq(J-,u-)δq,1F.

Proof.

We only prove the case of J+. Firstly, we prove that J+ satisfies the Mountain Pass Lemma and hence J+ has a nonzero critical point u+. In fact, J+C1(E,) and by Lemma 3.3 we see that J+ satisfies the (PS) condition. Clearly J+(0)=0. Thus we need to show that J+ satisfies (J1) and (J2). To verify (J1), by (1.8) and (V2), there exist ρ1>0 and ρ2>0 with V′′(0)<ρ2<λ1 such that V(t)12ρ2t2 for |t|ρ1. So, for all uE, if uλ1ρ1, then for each n(1,T), |u(n)|ρ1 and J+(u)=12u2-n=1TV+(u(n))=12u2-nN1V(u(n))12u2-12ρ2nN1(u(n),u(n))12u2-12ρ2n=1T(u(n),u(n))12u2-12ρ2λ1u2, where N1={n(1,T)u(n)0}. Let ρ=λ1ρ1,a=12(1-ρ2λ1)ρ2. Then J+(u)|Bρa>0 and hence (J1) holds. For (J2), by V()=λk(λk-1,λk+1), we claim that there exist γ¯>λk-1(λ1), b such that V(t)γ¯2t2+b,t.

In fact, by assumption (1.9), there exist M>0 and b1 such that V(t)(γ¯/2)t2+b1 for |t|M. Meanwhile, there exists b2 such that V(t)-(γ¯/2)t2b2 for |t|M by virtue of the continuity of V. Let b=min{b1,b2}, we get the conclusion.

Thus, if we choose espan{ϕ1} with e>0 and e=1, then J+(te)=t22-n=1TV(te(n))t22-γ¯t22(e,e)-bT=t22-γ¯t22λ1-bT- as 0<t+. Thus, we can choose a constant t large enough with t>ρ and u0=teE such that J+(u0)0. (J2) holds.

Therefore, by Theorem 2.7, J+ has a critical point u+0 and similar to the proof of Lemma 3.3, we can prove that u+>0. So u+ is also a critical point of J. In the following we compute the critical group Cq(J+,u+) by using Theorem 2.9.

Assume that J(u+)v,v=v,v-n=1T(V(u+(n))v(n),v(n))0,vE, and that there exists v00 such that J(u+)v0,v=0,vE. This implies that v0 satisfies Δ2v0(n-1)+V(u+(n))v0(n)=0,n(1,T),v0(0)=v0(T+1)=0. Hence the eigenvalue problem Δ2v(n-1)+λV(u+(n))v(n)=0,n(1,T),v(0)=v(T+1)=0 has an eigenvalue λ=1. (V1) implies that 1 must be a simple eigenvalue; see . So, dimker(J(u0))=1. Since E is a finite-dimensional Hilbert space, the Morse index of u+ must be finite and J(u+) must be a Fredholm operator. By Theorem 2.9, Cq(J+,u+)δq,1F. The proof is complete.

Remark 3.5.

We can choose the neighborhood U of u+ such that u>0 for all uU. Therefore, Cq(J,u+)Cq(J+,u+)δq,1F. Similarly, Cq(J,u-)Cq(J-,u-)δq,1F.

Now, we give the proof of Theorem 1.3.

Proof of Theorem <xref ref-type="statement" rid="thm1.1">1.3</xref>.

We only prove the case (i). By Lemma 3.2, Cq(J,)δq,kF. Hence by Proposition 2.4 the functional J has a critical point u1 satisfying Ck(J,u1)0. Since J(0)u,u(1-V(0)λ1)u2, by (V2) and J(0)=J(0)=0, we see that 0 is a local minimum of J. Hence Cq(J,0)δq,0F. By Remark 3.5, (3.49), (3.51), and k2 we get that u+, u-, and u1 are three nonzero critical points of J with u+>0 and u-<0. The proof is complete.

4. An Example and Future Directions

To illustrate the use of Theorem 1.3, we offer the following example.

Example 4.1.

Consider the BVP Δ2u(n-1)+V(u(n))=0,n(1,5),u(0)=0=u(6), where VC2(,) is defined as follows: V(t)={110t2,|t|1,12t2+34t4/3,|t|10,a  strictly  convex  function,otherwise. It is easy to verify that V satisfies (1.8), (1.9), (1.11), (V1), and (V2) with k=2. To verify the condition (G+), note that G(t)=t1/3 for |t|10, we claim that n=15(G(um(n)),vm(n))+,asm which implies that (G+) holds.

To this end, for any constant r>1, we introduce another norm in E(T=5) as follows: ur=(n=15|u(n)|r)1/r,uE. Since E is finite dimensional, there exist two constants C2C1>0 such that C1uurC2u,uE.

Now, by (G+), for any ϵ small enough, it is easy to see that wmϵum holds for m large enough.

Set Ω1={n(1,5):|um(n)|10},Ω2=(1,5)Ω1. Since um, Ω1, for m large enough. And for m large enough, we have n=15(G(um(n)),vm(n))=nΩ1(um1/3(n),vm(n))+nΩ2(G(um(n)),um(n))-nΩ2(G'(um(n)),wm(n))nΩ1(um1/3(n),vm(n))-c-cϵum=nΩ1(um1/3(n),um(n))-nΩ1(um1/3(n),wm(n))-c-cϵum. Here and below we denote by c various positive constants. Since nΩ1(um1/3(n),um(n))=n=15(um1/3(n),um(n))  -nΩ2(um1/3(n),um(n))=um4/34/3-c,nΩ1(um1/3(n),wm(n))n=15|um(n)|1/3|wm(n)|(n=15|um(n)|4/3)1/4(n=15|wm(n)|4/3)3/4=um4/31/3wm4/3cϵum4/34/3. Hence n=15(G(um(n)),vm(n))(1-cϵ)um4/34/3-cϵum-c. Since ϵ is small enough, we get (4.3) holds by the above and (4.5). Hence, by Theorem 1.3, BVP (4.1) has at least three nontrivial solutions.

Morse theory has been proved very useful in proving the existence and multiplicity of solutions of operator equations with variational frameworks. However, it is well known that the minimax methods is also a useful tool for the same purpose. The advantage of the minimax methods is that it provides an estimate of the critical value. But it is hard to distinguish critical points obtained by this methods with those by other methods, if the local behavior of the critical points is not very well known. However, critical groups serve as a topological tool in distinguishing isolated critical points. Hence, in order to obtain multiple solutions by using Morse theory, it is crucial to describe critical groups clearly.

A natural question is: can we use the same methods in this paper to other BVPs? Noticing that the key conditions which guarantee the multiplicity of solutions of the BVP (1.1) are as follows:

the BVP has a variational framework;

the eigenvalues of the corresponding linear BVP are nonzero and there is a one-sign eigenfunction,

hence, if the difference equation

Δ2u(n-1)+V'(u(n))=0,n(1,T) subject to some other boundary value conditions satisfying (1) and (2), then we can obtain similar results to Theorem 1.3.

Example 4.2.

Consider the BVP Δ2u(n-1)+V(u(n))=0,n(1,T),u(0)=0=Δu(T). Let E={u:u={u(n)}n=0T+1  with  u(0)=0=Δu(T)}. Then E is a T-dimensional Hilbert space with the inner product u,v=n=0T(Δu(n),Δv(n)). Define the functional J on E by J(u)=n=0T12|Δu(n)|2-n=1TV(u(n)). It is easy to see that u is a critical point of J in E if and only if u is a solution of the BVP (4.12).

The eigenvalues of the linear BVP Δ2u(n-1)+λu(n)=0,n(1,T),u(0)=0=Δu(T) are λ=λl=4sin2lπ2(2T+1),l=1,2,,T, and the corresponding eigenfunctions are ϕl(n)=sinlπn2T+1,l=1,2,,T. Hence, λl0 for all l(1,T) and ϕ1(n)>0 for all n(1,T). Therefore, the BVP (4.12) satisfies (1) and (2) and hence we can obtain similar results as in Theorem 1.3.

However, consider the following difference BVP:

Δ2u(n-1)+V'(u(n))=0,n(1,T),u(0)=u(T),Δu(0)=Δu(T). It is easy to verify that the variational functional of the BVP (4.19) is

J(u)=n=1T[12|Δu(n)|2-V(u(n))],uE1, where

E1={u:u={u(n)}n=0T+1  with  u(0)=u(T),Δu(0)=Δu(T+1)}. But, λ=0 is an eigenvalue of the linear BVP:

Δ2u(n-1)+λu(n)=0,n(1,T),u(0)=u(T),Δu(0)=Δu(T). So, for the BVP (4.19), we need to find other techniques (e.g., dual variational methods if possible) to study the BVP (4.19).

Acknowledgments

The authors would like to express their thanks to the referees for helpful suggestions. This research is supported by Guangdong College Yumiao Project (2009).

AgarwalR. P.PereraK.O'ReganD.Multiple positive solutions of singular and nonsingular discrete problems via variational methodsNonlinear Analysis: Theory, Methods & Applications2004581-2697310.1016/j.na.2003.11.012MR2070806ZBL1070.39005JiangL. Q.ZhouZ.Existence of nontrivial solutions for discrete nonlinear two point boundary value problemsApplied Mathematics and Computation2006180131832910.1016/j.amc.2005.12.018MR2263393ZBL1113.39023CaiX. C.cxchn@yahoo.com.cnYuJ. S.jsyu@gzhu.edu.cnExistence theorems for second-order discrete boundary value problemsJournal of Mathematical Analysis and Applications200632026496612-s2.0-003038004110.1016/j.jmaa.2005.07.029ZBL1113.39019AgarwalR. P.Difference Equations and Inequalities: Theory, Methods & Applications20002282ndNew York, NY, USAMarcel Dekkerxvi+971Monographs and Textbooks in Pure and Applied MathematicsMR1740241BinH.-H.binhonghua@163.comYuJ.-S.jsyu@gzhu.edu.cnGuoZ.-M.gzm100@21cn.comNontrivial periodic solutions for asymptotically linear resonant difference problemJournal of Mathematical Analysis and Applications200632214774882-s2.0-014204348110.1016/j.jmaa.2006.01.028ZBL1103.39005SuJ. B.sujb@mail.cnu.edu.cnMultiplicity results for asymptotically linear elliptic problems at resonanceJournal of Mathematical Analysis and Applications200327823974082-s2.0-003531411610.1016/S0022-247X(02)00707-2SuJ. B.sujb@mail.cnu.edu.cnZhaoL. G.An elliptic resonance problem with multiple solutionsJournal of Mathematical Analysis and Applications200631926046162-s2.0-003531411610.1016/j.jmaa.2005.10.059ZBL1155.35358CeramiG.An existence criterion for the critical points on unbounded manifoldsIstituto Lombardo. Accademia di Scienze e Lettere. Rendiconti. Scienze Matematiche e Applicazioni. A19781122332336MR581298BartoloP.BenciV.FortunatoD.Abstract critical point theorems and applications to some nonlinear problems with “ strong” resonance at infinityNonlinear Analysis: Theory, Methods & Applications19837998110122-s2.0-000917957810.1016/0362-546X(83)90115-3ZBL0522.58012ChangK.-C.Solutions of asymptotically linear operator equations via Morse theoryCommunications on Pure and Applied Mathematics19813456937122-s2.0-0009179578ZBL0444.58008BartschT.LiS.Critical point theory for asymptotically quadratic functionals and applications to problems with resonanceNonlinear Analysis: Theory, Methods & Applications19972834194412-s2.0-3824902512810.1016/0362-546X(95)00167-TZBL0872.58018MawhinJ.WillemM.Critical Point Theory and Hamiltonian Systems198974New York, NY, USASpringerxiv+277Applied Mathematical SciencesMR982267ChangK.-C.Infinite-Dimensional Morse Theory and Multiple Solution Problems1993Boston, Mass, USABirkhäuserx+312Progress in Nonlinear Differential Equations and Their Applications, 6MR1196690LiS. J.lisj@math03.math.ac.cnZouW. M.zouw@math03.math.ac.cnThe computations of the critical groups with an application to eliptic resonant problems at ahigher eigenvalueJournal of Mathematical Analysis and Applications199923512372592-s2.0-000294516610.1006/jmaa.1999.6396ZBL0935.35055PereraK.kperera@winnie.fit.eduSchechterM.mschecht@math.uci.eduNontrivial solutions of elliptic semilinear equations at resonanceManuscripta Mathematica200010133013112-s2.0-000007997710.1007/s002290050217ZBL0958.35051LiS. J.lisj@math03.math.ac.cnPereraK.kperera@winnie.fit.eduComputation of critical groups in resonance problems where the nonlinearity may not be sublinearNonlinear Analysis: Theory, Methods & Applications20014667777872-s2.0-003434725610.1016/S0362-546X(00)00133-4ZBL1088.58005HanZ.-Q.hanzhiq@dlut.edu.cnComputations of critical groups and applications to some differential equations at resonanceNonlinear Analysis: Theory, Methods & Applications2007676184718602-s2.0-003646647710.1016/j.na.2006.07.044ZBL1123.58007RabinowitzP. H.Minimax Methods in Critical Point Theory with Applications to Differential Equations1986Providence, RI, USAAmerican Mathematical SocietyCBMS Regional Conference Series in Mathematics no. 65ChangK.-C.AmbrosettiA.ChangK.-C.EkelandI.Morse theory in nonlinear analysisNonlinear Functional Analysis and Applications to Differential Equations (Trieste, 1997)1998River Edge, NJ, USAWorld Scientific60101MR1703528ZBL0960.58006