Ground state homoclinic orbits of damped vibration problems

Abstract

In this paper, we consider a class of non-periodic damped vibration problems with superquadratic nonlinearities. We study the existence of nontrivial ground state homoclinic orbits for this class of damped vibration problems under some conditions weaker than those previously assumed. To the best of our knowledge, there has been no work focused on this case.

MSC: 49J40, 70H05.

1 Introduction and main results

We shall study the existence of ground state homoclinic orbits for the following non-periodic damped vibration system:

$$\ddot{u}(t)+M\dot{u}(t)-L(t)u(t)+H_u(t,u(t))=0,t\in \mathbb{R},$$

(1.1)

where M is an antisymmetric $N\times N$ constant matrix, $L(t)\in C(\mathbb{R},{\mathbb{R}}^{N\times N})$ is a symmetric matrix, $H(t,u)\in C^1(\mathbb{R}\times {\mathbb{R}}^N,\mathbb{R})$ and $H_u(t,u)$ denotes its gradient with respect to the u variable. We say that a solution $u(t)$ of (1.1) is homoclinic (to 0) if $u(t)\in C^2(\mathbb{R},{\mathbb{R}}^N)$ such that $u(t)\to 0$ and $\dot{u}(t)\to 0$ as $|t|\to \mathrm{\infty }$. If $u(t)\not\equiv 0$, then $u(t)$ is called a nontrivial homoclinic solution.

If $M=0$ (zero matrix), then (1.1) reduces to the following second-order Hamiltonian system:

$$\ddot{u}(t)-L(t)u(t)+H_u(t,u(t))=0,t\in \mathbb{R}.$$

(1.2)

This is a classical equation which can describe many mechanic systems such as a pendulum. In the past decades, the existence and multiplicity of periodic solutions and homoclinic orbits for (1.2) have been studied by many authors via variational methods; see [1–18] and the references therein.

The periodic assumptions are very important in the study of homoclinic orbits for (1.2) since periodicity is used to control the lack of compactness due to the fact that (1.2) is set on all ℝ. However, non-periodic problems are quite different from the ones described in periodic cases. Rabinowitz and Tanaka [11] introduced a type of coercive condition on the matrix $L(t)$,

$$l(t):=\underset{|u|=1}{inf}(L(t)u,u)\to +\mathrm{\infty }\text{as }|t|\to \mathrm{\infty },$$

(1.3)

and obtained the existence of a homoclinic orbit for non-periodic (1.2) under the Ambrosetti-Rabinowitz (AR) superquadratic condition:

$$0<\mu H(t,u)\le (H_u(t,u),u),\mathrm{\forall }t\in \mathbb{R},\mathrm{\forall }u\in {\mathbb{R}}^N\setminus \{0\},$$

where $\mu >2$ is a constant, $(\cdot ,\cdot )$ denotes the standard inner product in ${\mathbb{R}}^N$ and the associated norm is denoted by $|\cdot |$.

We should mention that in the case where $M\ne 0$, i.e., the damped vibration system (1.1), only a few authors have studied homoclinic orbits of (1.1); see [19–23]. Zhu [23] considered the periodic case of (1.1) (i.e., $L(t)$ and $H(t,u)$ are T-periodic in t with $T>0$) and obtained the existence of nontrivial homoclinic solutions of (1.1). The authors [19–22] considered the non-periodic case of (1.1): Zhang and Yuan [22] obtained the existence of at least one homoclinic orbit for (1.1) when H satisfies the subquadratic condition at infinity by using a standard minimizing argument; by a symmetric mountain pass theorem and a generalized mountain pass theorem, Wu and Zhang [21] obtained the existence and multiplicity of homoclinic orbits for (1.1) when H satisfies the local (AR) superquadratic growth condition:

$$0<\mu H(t,u)\le (H_u(t,u),u),\mathrm{\forall }t\in \mathbb{R},\mathrm{\forall }|u|\ge r,$$

(1.4)

where $\mu >2$ and $r>0$ are two constants. Notice that the authors [21, 22] all used condition (1.3). Recently, the author in [19, 20] obtained infinitely many homoclinic orbits for (1.1) when H satisfies the subquadratic[19] and asymptotically quadratic[20] condition at infinity by the following weaker conditions than (1.3):

(L₁) There is a constant $\beta >1$ such that

$$meas\{t\in \mathbb{R}:{|t|}^{-\beta }L(t)<b{I}_N\}<+\mathrm{\infty },\mathrm{\forall }b>0;$$

(L₂) There is a constant ${\gamma }_0\ge 0$ such that

$$l(t):=\underset{|u|=1}{inf}(L(t)u,u)\ge -{\gamma }_0,\mathrm{\forall }t\in \mathbb{R},$$

which were firstly used in [15]. It is not hard to check that the matrix-valued function $L(t):=(t^4{sin}^{2}t+1)I_N$ satisfies (L₁) and (L₂), but does not satisfy (1.3).

We define an operator $\mathrm{\Gamma }:H^1(\mathbb{R},{\mathbb{R}}^N)\to H^1(\mathbb{R},{\mathbb{R}}^N)$ by

$$(\mathrm{\Gamma }u,v):={\int }_{\mathbb{R}}(Mu(t),\dot{v}(t))dt,\mathrm{\forall }u,v\in H^1(\mathbb{R},{\mathbb{R}}^N).$$

Since M is an antisymmetric $N\times N$ constant matrix, Γ is self-adjoint on $H^1(\mathbb{R},{\mathbb{R}}^N)$. Let χ denote the self-adjoint extension of the operator $-\frac{d^2}{d{t}^2}+L(t)+\mathrm{\Gamma }$. We are interested in the indefinite case:

(J₁) $a:=sup(\sigma (\chi )\cap (-\mathrm{\infty },0))<0<b:=inf(\sigma (\chi )\cap (0,\mathrm{\infty }))$.

To state our main result, we still need the following assumptions:

(H₁) $|\mathrm{\nabla }H(t,u)|\le c(1+{|u|}^{p-1})$ for some $c>0$ and $p>2$, $\mathrm{\forall }t\in \mathbb{R}$ and $u\in {\mathbb{R}}^N$.

(H₂) $H(t,u)\ge \frac{1}{2}a{|u|}^2$, $\mathrm{\forall }t\in \mathbb{R}$ and $u\in {\mathbb{R}}^N$.

(H₃) For some $\delta >0$ and $\gamma \in (0,b)$,

$$|\mathrm{\nabla }H(t,u)|\le \gamma |u|,|H(t,u)|\le \frac{1}{2}|\mathrm{\nabla }H(t,u)|\cdot |u|,\mathrm{\forall }|u|<\delta ,\mathrm{\forall }t\in \mathbb{R}.$$

(H₄) $\frac{H(t,u)}{{|u|}^2}\to +\mathrm{\infty }$ as $|u|\to +\mathrm{\infty }$ and there exists $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$ such that

$$H(t,u)\ge -W_1(t),\mathrm{\forall }t\in \mathbb{R}\text{ and }u\in {\mathbb{R}}^N.$$

(1.5)

(H₅) For all $t\in \mathbb{R}$ and $u,z\in {\mathbb{R}}^N$, there holds

$$\begin{array}{c}H(t,u+z)-H(t,u)-r(\mathrm{\nabla }H(t,u),z)+\frac{{(r-1)}^2}{2}(\mathrm{\nabla }H(t,u),u)\hfill \\ \ge -W_1(t),\mathrm{\forall }r\in [0,1].\hfill \end{array}$$

Our main results read as follows.

Theorem 1.1If (L₁)-(L₂), (J₁) and (H₁)-(H₅) hold, then (1.1) has at least one nontrivial homoclinic orbit.

Theorem 1.2Let ℳ be the collection of solutions of (1.1), then there is a solution that minimizes the energy functional

$$I(u)=\frac{1}{2}{\int }_{\mathbb{R}}\left[\right|\dot{u}(t){|}^2+(Mu(t),\dot{u}(t))+(L(t)u(t),u(t))]dt-{\int }_{\mathbb{R}}H(t,u)dt,u\in E$$

over ℳ, where the spaceEis defined in Section  2. In addition, if

$$|\mathrm{\nabla }H(t,u)|=o(|u|)\mathit{\text{as }}|u|\to 0$$

uniformly int, then there is a nontrivial homoclinic orbit that minimizes the energy functional over$\mathcal{M}\setminus \{0\}$, i.e., a ground state homoclinic orbit.

Remark 1.1 Although the authors [21] have studied (1.1) with superquadratic nonlinearities, our superquadratic condition (H₄) is weaker than (1.4) in [21]. Moreover, we study the ground state homoclinic orbit of (1.1). To the best of our knowledge, there has been no result published concerning the ground state homoclinic orbit of (1.1).

Example 1.1

• (1) $H(t,u)={|u|}^p$,

• (2) $H(t,u)=g(t)({|u|}^p+(p-2){|u|}^{p-\epsilon }{sin}^2(\frac{{|u|}^{\epsilon }}{\epsilon }))$,

where $p>2$, $g(t)>0$ is continuous and $0<\epsilon <p-2$. It is easy to check that the above two functions satisfy assumptions (H₁)-(H₅) if we take $0\le W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$, where $W_1(t)$ is the function in (H₄)-(H₅).

The rest of the present paper is organized as follows. In Section 2, we establish the variational framework associated with (1.1), and we also give some preliminary lemmas, which are useful in the proofs of our main results. In Section 3, we give the detailed proofs of our main results.

2 Preliminary lemmas

In the following, we use ${\parallel \cdot \parallel }_{L^p}$ to denote the norm of $L^p(\mathbb{R},{\mathbb{R}}^N)$ for any $p\in [1,\mathrm{\infty }]$. Let $W:=H^1(\mathbb{R},{\mathbb{R}}^N)$ be a Hilbert space with the inner product and norm given respectively by

$${〈u,v〉}_W={\int }_{\mathbb{R}}[(\dot{u}(t),\dot{v}(t))+(u(t),v(t))]dt,{\parallel u\parallel }_W={〈u,u〉}_E^{1/2},\mathrm{\forall }u,v\in W.$$

It is well known that W is continuously embedded in $L^p(\mathbb{R},{\mathbb{R}}^N)$ for $p\in [2,\mathrm{\infty })$. We define an operator $\mathrm{\Gamma }:W\to W$ by

$$(\mathrm{\Gamma }u,v):={\int }_{\mathbb{R}}(Mu(t),\dot{v}(t))dt,\mathrm{\forall }u,v\in W.$$

Since M is an antisymmetric $N\times N$ constant matrix, Γ is self-adjoint on W. Moreover, we denote by χ the self-adjoint extension of the operator $-\frac{d^2}{d{t}^2}+L(t)+\mathrm{\Gamma }$ with the domain $\mathcal{D}(\chi )\subset L^2(\mathbb{R},{\mathbb{R}}^N)$.

Let $E:=\mathcal{D}({|\chi |}^{1/2})$, the domain of ${|\chi |}^{1/2}$. We define respectively on E the inner product and the norm

$${〈u,v〉}_E:={({|\chi |}^{1/2}u,{|\chi |}^{1/2}v)}_2+{(u,v)}_2\text{and}{\parallel u\parallel }_E={〈u,u〉}_E^{1/2},$$

where ${(\cdot ,\cdot )}_2$ denotes the inner product in $L^2(\mathbb{R},{\mathbb{R}}^N)$.

By a similar proof of Lemma 3.1 in [15], we can prove that if conditions (L₁) and (L₂) hold, then

$$E\text{ is compactly embedded into }{L}^p(\mathbb{R},{\mathbb{R}}^N),\mathrm{\forall }p\in [1,+\mathrm{\infty }].$$

(2.1)

Therefore, it is easy to prove that the spectrum $\sigma (\chi )$ has a sequence of eigenvalues (counted with their multiplicities)

$${\lambda }_1\le {\lambda }_2\le \cdots \le {\lambda }_k\le \cdots \to \mathrm{\infty },$$

and the corresponding system of eigenfunctions $\{e_k:k\in \mathbb{N}\}$ ($\chi e_k={\lambda }_k{e}_k$) forms an orthogonal basis in $L^2(\mathbb{R},{\mathbb{R}}^N)$.

By (J₁), we may let

$$k_1:=\mathrm{♯}\{j:{\lambda }_j<0\},E^{-}:=span\{e_1,\dots ,e_{k_1}\},E^{+}:={cl}_E(span\{e_{k_1+1},\dots \}).$$

Then one has the orthogonal decomposition

$$E=E^{-}\oplus E^{+}$$

with respect to the inner product ${〈\cdot ,\cdot 〉}_E$. Now, we introduce respectively on E the following new inner product and norm:

$$〈u,v〉:={({|\chi |}^{1/2}u,{|\chi |}^{1/2}v)}_2,\parallel u\parallel ={〈u,u〉}^{1/2},$$

(2.2)

where $u,v\in E=E^{-}\oplus E^{+}$ with $u=u^{-}+u^{+}$ and $v=v^{-}+v^{+}$. Clearly, the norms $\parallel \cdot \parallel$ and ${\parallel \cdot \parallel }_E$ are equivalent (see [4]), and the decomposition $E=E^{-}\oplus E^{+}$ is also orthogonal with respect to both inner products $〈\cdot ,\cdot 〉$ and ${(\cdot ,\cdot )}_2$. Hence, by (J₁), E with equivalent norms, besides, we have

$$-{\parallel u^{-}\parallel }^2={(\chi u^{-},u^{-})}_2\le a{\parallel u^{-}\parallel }_{L^2}^2,\mathrm{\forall }{u}^{-}\in E^{-}$$

(2.3)

and

$${\parallel u^{+}\parallel }^2={(\chi u^{+},u^{+})}_2\ge b{\parallel u^{+}\parallel }_{L^2}^2,\mathrm{\forall }{u}^{+}\in E^{+},$$

(2.4)

where a and b are defined in (J₁).

For problem (1.1), we consider the following functional:

$$I(u)=\frac{1}{2}{\int }_{\mathbb{R}}\left[\right|\dot{u}(t){|}^2+(Mu(t),\dot{u}(t))+(L(t)u(t),u(t))]dt-{\int }_{\mathbb{R}}H(t,u)dt,u\in E.$$

Then I can be rewritten as

$$I(u)=\frac{1}{2}{\parallel u^{+}\parallel }^2-\frac{1}{2}{\parallel u^{-}\parallel }^2-{\int }_{\mathbb{R}}H(t,u)dt,u=u^{-}+u^{+}\in E.$$

Let $\mathrm{\Psi }(u):={\int }_{\mathbb{R}}H(t,u)dt$. In view of the assumptions of H, we know $I,\mathrm{\Psi }\in C^1(E,\mathbb{R})$ and the derivatives are given by

$${\mathrm{\Psi }}^{\prime }(u)v={\int }_{\mathbb{R}}(H_u(t,u),v)dt,I^{\prime }(u)v=〈u^{+},v^{+}〉-〈u^{-},v^{-}〉-I^{\prime }(u)v$$

for any $u,v\in E=E^{-}\oplus E^{+}$ with $u=u^{-}+u^{+}$ and $v=v^{-}+v^{+}$. By the discussion of [24], the (weak) solutions of system (1.1) are the critical points of the $C^1$ functional $I:E\to \mathbb{R}$. Moreover, it is easy to verify that if $u\not\equiv 0$ is a solution of (1.1), then $u(t)\to 0$ and $\dot{u}(t)\to 0$ as $|t|\to \mathrm{\infty }$ (see Lemma 3.1 in [25]).

The following abstract critical point theorem plays an important role in proving our main result. Let E be a Hilbert space with the norm $\parallel \cdot \parallel$ and have an orthogonal decomposition $E=N\oplus N^{\mathrm{\perp }}$, $N\subset E$ is a closed and separable subspace. There exists a norm ${|v|}_{\omega }$ satisfying ${|v|}_{\omega }\le \parallel v\parallel$ for all $v\in N$ and inducing a topology equivalent to the weak topology of N on a bounded subset of N. For $u=v+w\in E=N\oplus N^{\mathrm{\perp }}$ with $v\in N$, $w\in N^{\mathrm{\perp }}$, we define ${|u|}_{\omega }^2={|v|}_{\omega }^2+{\parallel w\parallel }^2$. Particularly, if $u_n=v_n+w_n$ is $\parallel \cdot \parallel$-bounded and $u_n\stackrel{{|\cdot |}_{\omega }}{\to }u$, then $v_n\rightharpoonup v$ weakly in N, $w_n\to w$ strongly in $N^{\mathrm{\perp }}$, $u_n\rightharpoonup v+w$ weakly in E (cf.[26]).

Let $E:=E^{-}\oplus E^{+}$, $z_0\in E^{+}$ with $\parallel z_0\parallel =1$. Let $N:=E^{-}\oplus \mathbb{R}{z}_0$ and $E_1^{+}:=N^{\mathrm{\perp }}={(E^{-}\oplus \mathbb{R}{z}_0)}^{\mathrm{\perp }}$. For $R>0$, let

$$Q:=\{u:=u^{-}+s{z}_0:s\in {\mathbb{R}}^{+},u^{-}\in E^{-},\parallel u\parallel <R\}$$

with $p_0=s_0{z}_0\in Q$, $s_0>0$. We define

$$D:=\{u:=s{z}_0+w^{+}:s\in \mathbb{R},w^{+}\in E_1^{+},\parallel s{z}_0+w^{+}\parallel =s_0\}.$$

For $I\in C^1(E,\mathbb{R})$, define

$$\mathrm{\Gamma }:=\{h:\begin{array}{l}h:[0,1]\times \overline{Q}\mapsto E\text{ is }{|\cdot |}_{\omega }\text{-continuous};\\ h(0,u)=u\text{ and }I(h(s,u))\le I(u)\text{ for all }u\in \overline{Q};\\ \text{For any }(s_0,u_0)\in [0,1]\times \overline{Q},\text{there is a }{|\cdot |}_{\omega }\text{-neighborhood}\\ U_{(s_0,u_0)}\text{ s.t. }\{u-h(t,u):(t,u)\in U_{(s_0,u_0)}\cap ([0,1]\times \overline{Q})\}\subset E_{\mathrm{fin}}\end{array}\},$$

where $E_{\mathrm{fin}}$ denotes various finite-dimensional subspaces of E, $\mathrm{\Gamma }\ne 0$ since $\mathit{id}\in \mathrm{\Gamma }$.

The variant weak linking theorem is as follows.

Lemma 2.1 ([26])

The family of$C^1$-functionals$\{I_{\lambda }\}$has the form

$$I_{\lambda }(u):=\lambda K(u)-J(u),\mathrm{\forall }\lambda \in [1,{\lambda }_0],$$

where${\lambda }_0>1$. Assume that

• (a) $K(u)\ge 0$, $\mathrm{\forall }u\in E$, $I_1=I$;

• (b) $|J(u)|+K(u)\to \mathrm{\infty }$as$\parallel u\parallel \to \mathrm{\infty }$;

• (c) $I_{\lambda }$is${|\cdot |}_{\omega }$-upper semicontinuous, $I_{\lambda }^{\prime }$is weakly sequentially continuous onE. Moreover, $I_{\lambda }$maps bounded sets to bounded sets;

• (d) ${sup}_{\partial Q}{I}_{\lambda }<{inf}_D{I}_{\lambda }$, $\mathrm{\forall }\lambda \in [1,{\lambda }_0]$.

Then, for almost all$\lambda \in [1,{\lambda }_0]$, there exists a sequence$\{u_n\}$such that

$$\underset{n}{sup}\parallel u_n\parallel <\mathrm{\infty },I_{\lambda }^{\prime }(u_n)\to 0,I_{\lambda }(u_n)\to c_{\lambda },$$

where$c_{\lambda }:={inf}_{h\in \mathrm{\Gamma }}{sup}_{u\in \overline{Q}}{I}_{\lambda }(h(t,u))\in [{inf}_D{I}_{\lambda },{sup}_{\overline{Q}}{I}_{\lambda }]$.

In order to apply Lemma 2.1, we shall prove a few lemmas. We pick ${\lambda }_0$ such that $1<{\lambda }_0<min[2,b/\gamma ]$. For $1\le \lambda \le {\lambda }_0$, we consider

$$I_{\lambda }(u):=\frac{\lambda }{2}{\parallel u^{+}\parallel }^2-(\frac{1}{2}{\parallel u^{-}\parallel }^2+{\int }_{\mathbb{R}}H(t,u(t))dt):=\lambda K(u)-J(u).$$

It is easy to see that $I_{\lambda }$ satisfies condition (a) in Lemma 2.1. To see (c), if $u_n\stackrel{{|\cdot |}_{\omega }}{\to }u$ and $I_{\lambda }(u_n)\ge c$, then $u_n^{+}\to u^{+}$ and $u_n^{-}\rightharpoonup u^{-}$ in E, $u_n\to u$ a.e. on ℝ, going to a subsequence if necessary. It follows from the weak lower semicontinuity of the norm, Fatou’s lemma and the fact $H(t,u)+W_1(t)\ge 0$ for all $t\in \mathbb{R}$ and $u\in {\mathbb{R}}^N$ by (1.5) in (H₄) that

$$\begin{array}{rcl}c& \le & \underset{n\to \mathrm{\infty }}{lim\hspace{0.17em}sup}{I}_{\lambda }(u_n)\\ =& \underset{n\to \mathrm{\infty }}{lim\hspace{0.17em}sup}[\frac{\lambda }{2}{\parallel u_n^{+}\parallel }^2-(\frac{1}{2}{\parallel u_n^{-}\parallel }^2+{\int }_{\mathbb{R}}(H(t,u_n)+W_1(t))dt)+{\int }_{\mathbb{R}}{W}_1(t)dt]\\ \le & \frac{\lambda }{2}{\parallel u^{+}\parallel }^2-\underset{n\to \mathrm{\infty }}{lim\hspace{0.17em}inf}[\frac{1}{2}{\parallel u_n^{-}\parallel }^2+{\int }_{\mathbb{R}}(H(t,u_n)+W_1(t))dt]+{\int }_{\mathbb{R}}{W}_1(t)dt\\ \le & \frac{\lambda }{2}{\parallel u^{+}\parallel }^2-(\frac{1}{2}{\parallel u^{-}\parallel }^2+{\int }_{\mathbb{R}}H(t,u)dt)=I_{\lambda }(u).\end{array}$$

Thus we get $I_{\lambda }(u)\ge c$. It implies that $I_{\lambda }$ is ${|\cdot |}_{\omega }$-upper semicontinuous. $I_{\lambda }^{\prime }$ is weakly sequentially continuous on E due to [27].

Lemma 2.2Under assumptions of Theorem  1.1, then

$$J(u)+K(u)\to \mathrm{\infty }\mathit{\text{as }}\parallel u\parallel \to \mathrm{\infty }.$$

Proof By the definition of $I(u)$ and (H₄), we have

$$\begin{array}{rcl}J(u)+K(u)& =& \frac{1}{2}{\parallel u^{+}\parallel }^2+\frac{1}{2}{\parallel u^{-}\parallel }^2+{\int }_{\mathbb{R}}H(t,u(t))dt\\ \ge & \frac{1}{2}{\parallel u\parallel }^2-{\int }_{\mathbb{R}}{W}_1(t)dt\to +\mathrm{\infty }\text{as }\parallel u\parallel \to \mathrm{\infty },\end{array}$$

which is due to $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$. □

Therefore, Lemma 2.2 implies that condition (b) holds. To continue the discussion, we still need to verify condition (d), that is, the following two lemmas.

Lemma 2.3Under assumptions of Theorem  1.1, there are two positive constants$ϵ,\rho >0$such that

$$I_{\lambda }(u)\ge ϵ,u\in E^{+},\parallel u\parallel =\rho ,\lambda \in [1,{\lambda }_0].$$

Proof By (H₁), (H₃), (2.4) and the Sobolev embedding theorem, for all $u\in E^{+}$,

$$\begin{array}{rcl}{I}_{\lambda }(u)& \ge & \frac{1}{2}{\parallel u\parallel }^2-{\int }_{\mathbb{R}}H(t,u(t))dt\\ =& \frac{1}{2}{\parallel u\parallel }^2-{\int }_{\{t\in \mathbb{R}:|u|<\delta \}}H(t,u(t))dt-{\int }_{\{t\in \mathbb{R}:|u|\ge \delta \}}H(t,u(t))dt\\ \ge & \frac{1}{2}{\parallel u\parallel }^2-\frac{1}{2}\gamma {\int }_{\{t\in \mathbb{R}:|u|<\delta \}}{|u|}^{2}dt-c{\int }_{\{t\in \mathbb{R}:|u|\ge \delta \}}({|u|}^p+|u|)dt\\ \ge & \frac{1}{2}{\parallel u\parallel }^2-\frac{\gamma }{b}\frac{1}{2}{\parallel u\parallel }^2-C{\parallel u\parallel }^p\\ =& \frac{1}{2}{\parallel u\parallel }^2(1-\frac{\gamma }{b}-2C{\parallel u\parallel }^{p-2}),0\le \gamma <b,\end{array}$$

where C is a positive constant. It implies the conclusion if we take $\parallel u\parallel$ sufficiently small. □

Lemma 2.4Under assumptions of Theorem  1.1, then there is an$R>0$such that

$$I_{\lambda }(u)\le 0,u\in \partial Q_R,\lambda \in [1,{\lambda }_0],$$

where$Q_R:=\{u:=v+s{z}_0:s\ge 0,v\in E^{-},z_0\in E^{+}\mathit{\text{ with }}\parallel z_0\parallel =1,\parallel u\parallel \le R\}$.

Proof Suppose by contradiction that there exist $R_n\to \mathrm{\infty }$, ${\lambda }_n\in [1,{\lambda }_0]$ and $u_n=v_n+s_n{z}_0\in \partial Q_R$ such that $I_{{\lambda }_n}(u_n)>0$. If $s_n=0$, then by (H₂) and (2.3), we have

$$I_{{\lambda }_n}(v_n)=-\frac{1}{2}{\parallel v_n\parallel }^2-{\int }_{\mathbb{R}}H(t,v_n)dt\le -\frac{1}{2}{\parallel v_n\parallel }^2-\frac{1}{2}a{\parallel v_n\parallel }_{L^2}^2\le 0.$$

Therefore, $s_n\ne 0$ and ${\parallel u_n\parallel }^2={\parallel v_n\parallel }^2+s_n^2=R_n^2$. Let ${\tilde{u}}_n=\frac{u_n}{\parallel u_n\parallel }={\tilde{s}}_n{z}_0+{\tilde{v}}_n$, then

$${\parallel {\tilde{u}}_n\parallel }^2={\parallel {\tilde{v}}_n\parallel }^2+{\tilde{s}}_n^2=1.$$

It follows from $I_{{\lambda }_n}(u_n)>0$ and the definition of I that

$$\begin{array}{rcl}0& <& \frac{I_{{\lambda }_n}(u_n)}{{\parallel u_n\parallel }^2}=\frac{1}{2}({\lambda }_n{\tilde{s}}_n^2-{\parallel {\tilde{v}}_n\parallel }^2)-{\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt\\ =& \frac{1}{2}[({\lambda }_n+1){\tilde{s}}_n^2-1]-{\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt.\end{array}$$

(2.5)

There are renamed subsequences such that ${\tilde{s}}_n\to \tilde{s}$, ${\lambda }_n\to \lambda$, and there is a renamed subsequence such that ${\tilde{u}}_n=\frac{u_n}{\parallel u_n\parallel }={\tilde{s}}_n{z}_0+{\tilde{v}}_n\rightharpoonup \tilde{u}$ in E and ${\tilde{u}}_n\to \tilde{u}$ a.e. on ℝ.

We claim that

$$\underset{n\to \mathrm{\infty }}{lim\hspace{0.17em}inf}{\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt\ge 0.$$

(2.6)

Case 1. If $\tilde{u}\not\equiv 0$. Let ${\mathrm{\Omega }}_0$ be the subset of ℝ where $\tilde{u}\ne 0$, then for all $t\in {\mathrm{\Omega }}_0$ we have $|u_n|=|{\tilde{u}}_n|\cdot \parallel u_n\parallel \to \mathrm{\infty }$. It follows from (H₄) and $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$ that

$${\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt\ge {\int }_{{\mathrm{\Omega }}_0}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt-{\int }_{\mathbb{R}\setminus {\mathrm{\Omega }}_0}\frac{W_1(t)}{{\parallel u_n\parallel }^2}dt\to +\mathrm{\infty }\text{as }n\to \mathrm{\infty }.$$

Case 2. If $\tilde{u}\equiv 0$, then by (H₄) and $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$, we have

$${\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt={\int }_{\mathbb{R}}\frac{H(t,u_n)}{{\parallel u_n\parallel }^2}dt\ge -{\int }_{\mathbb{R}}\frac{W_1(t)}{{\parallel u_n\parallel }^2}dt\to 0\text{as }n\to \mathrm{\infty }.$$

Therefore, Cases 1 and 2 imply that (2.6) holds. Therefore, by (2.5), (2.6) and the facts ${\tilde{s}}_n\to \tilde{s}$, ${\lambda }_n\to \lambda$, we have

$$(\lambda +1){\tilde{s}}^2-1\ge 0,$$

that is, ${\tilde{s}}^2\ge \frac{1}{1+\lambda }\ge \frac{1}{1+{\lambda }_0}>0$. Thus, $\tilde{u}\ne 0$. It follows from (H₄) that

$${\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|{\tilde{u}}_n|}^{2}dt\to +\mathrm{\infty }\text{as }n\to \mathrm{\infty },$$

which contradicts (2.5). The proof is finished. □

Therefore, Lemmas 2.3 and 2.4 imply that condition (d) of Lemma 2.1 holds. Applying Lemma 2.1, we soon obtain the following fact.

Lemma 2.5Under assumptions of Theorem  1.1, for almost all$\lambda \in [1,{\lambda }_0]$, there exists a sequence$\{u_n\}$such that

$$\underset{n}{sup}\parallel u_n\parallel <\mathrm{\infty },I_{\lambda }^{\prime }(u_n)\to 0,I_{\lambda }(u_n)\to c_{\lambda },$$

where the definition of$c_{\lambda }$is given in Lemma  2.1.

Lemma 2.6Under assumptions of Theorem  1.1, for almost all$\lambda \in [1,{\lambda }_0]$, there exists a$u_{\lambda }\in E$such that

$$I_{\lambda }^{\prime }(u_{\lambda })=0,I_{\lambda }(u_{\lambda })=c_{\lambda }.$$

Proof Let $\{u_n\}$ be the sequence obtained in Lemma 2.5. Since $\{u_n\}$ is bounded, we can assume $u_n\rightharpoonup u_{\lambda }$ in E and $u_n\to u_{\lambda }$ a.e. on ℝ. By (H₁), (H₃), (2.1) and Theorem A.4 in [27], we have

$${\int }_{\mathbb{R}}\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)dt\to {\int }_{\mathbb{R}}\frac{1}{2}(\mathrm{\nabla }H(t,u_{\lambda }),u_{\lambda })dt$$

(2.7)

and

$${\int }_{\mathbb{R}}H(t,u_n)dt\to {\int }_{\mathbb{R}}H(t,u_{\lambda })dt.$$

(2.8)

By Lemma 2.5 and the fact that $I_{\lambda }^{\prime }$ is weakly sequentially continuous, we have

$$I_{\lambda }^{\prime }(u_{\lambda })\phi =\underset{n\to \mathrm{\infty }}{lim}{I}_{\lambda }^{\prime }(u_n)\phi =0,\mathrm{\forall }\phi \in E.$$

That is, $I_{\lambda }^{\prime }(u_{\lambda })=0$. By Lemma 2.5, we have

$$I_{\lambda }(u_n)-\frac{1}{2}{I}_{\lambda }^{\prime }(u_n)u_n={\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt\to c_{\lambda }.$$

It follows from (2.7), (2.8) and the fact $I_{\lambda }^{\prime }(u_{\lambda })=0$ that

$$I_{\lambda }(u_{\lambda })=I_{\lambda }(u_{\lambda })-\frac{1}{2}{I}_{\lambda }^{\prime }(u_{\lambda })u_{\lambda }={\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_{\lambda }),u_{\lambda })-H(t,u_{\lambda }))dt=c_{\lambda }.$$

The proof is finished. □

Applying Lemma 2.6, we soon obtain the following fact.

Lemma 2.7Under assumptions of Theorem  1.1, for every$\lambda \in [1,{\lambda }_0]$, there are sequences$\{u_n\}\subset E$and${\lambda }_n\in [1,{\lambda }_0]$with${\lambda }_n\to \lambda$such that

$$I_{{\lambda }_n}^{\prime }(u_n)=0,I_{{\lambda }_n}(u_n)=c_{{\lambda }_n}.$$

Lemma 2.8Under assumptions of Theorem  1.1, then

$${\int }_{\mathbb{R}}[H(t,u)-H(t,rw)+r^2(\mathrm{\nabla }H(t,u),w)-\frac{1+r^2}{2}(\mathrm{\nabla }H(t,u),u)]dt\le C,$$

where$u\in E$, $w\in E^{+}$, $0\le r\le 1$and the constant$C:={\int }_{\mathbb{R}}|W_1(t)|dt$does not depend onu, w, r.

Proof This follows from (H₅) if we take $u=u$ and $z=rw-u$. □

Lemma 2.9The sequences given in Lemma  2.7 are bounded.

Proof Write $u_n=u_n^{+}+u_n^{-}$, where $u_n^{\pm }\in E^{\pm }$. Suppose that

$$\parallel u_n\parallel \to \mathrm{\infty }\text{as }n\to \mathrm{\infty }.$$

Let $v_n:=\frac{u_n}{\parallel u_n\parallel }$, then $v_n^{+}=\frac{u_n^{+}}{\parallel u_n\parallel }$, $v_n^{-}=\frac{u_n^{-}}{\parallel u_n\parallel }$, ${\parallel v_n\parallel }^2={\parallel v_n^{+}\parallel }^2+{\parallel v_n^{-}\parallel }^2=1$ and $\parallel v_n^{+}\parallel \le 1$. Thus $v_n^{+}\rightharpoonup v^{+}$ in E and $v_n^{+}\to v^{+}$ a.e. on ℝ, after passing to a subsequence.

Case 1. If $v^{+}\not\equiv 0$. Let ${\mathrm{\Omega }}_1$ be the subset of ℝ where $v^{+}\ne 0$. Then $v\ne 0$ and $|u_n|=|v_n|\cdot \parallel u_n\parallel \to \mathrm{\infty }$ on ${\mathrm{\Omega }}_1$. It follows from (H₄) and $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$ that

$${\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|v_n|}^{2}dt\ge {\int }_{{\mathrm{\Omega }}_1}\frac{H(t,u_n)}{{|u_n|}^2}{|v_n|}^{2}dt-{\int }_{\mathbb{R}\setminus {\mathrm{\Omega }}_1}\frac{W_1(t)}{{\parallel u_n\parallel }^2}dt\to +\mathrm{\infty }\text{as }n\to \mathrm{\infty },$$

which together with Lemmas 2.3 and 2.7 and $v_n^{\pm }\to v^{\pm }$ in $L^q(\mathbb{R},{\mathbb{R}}^N)$ for all $1\le q\le \mathrm{\infty }$ (by (2.1)) implies that

$$0\le \frac{c_{{\lambda }_n}}{{\parallel u_n\parallel }^2}=\frac{I_{{\lambda }_n}(u_n)}{{\parallel u_n\parallel }^2}=\frac{{\lambda }_n}{2}{\parallel v_n^{+}\parallel }^2-\frac{1}{2}{\parallel v_n^{-}\parallel }^2-{\int }_{\mathbb{R}}\frac{H(t,u_n)}{{|u_n|}^2}{|v_n|}^{2}dt\to -\mathrm{\infty }\text{as }n\to \mathrm{\infty }.$$

It is a contradiction.

Case 2. If $v^{+}\equiv 0$. We claim that there is a constant C independent of $u_n$ and ${\lambda }_n$ such that

$$I_{{\lambda }_n}\left(r{u}_n^{+}\right)-I_{{\lambda }_n}(u_n)\le C,\mathrm{\forall }r\in [0,1].$$

(2.9)

Since

$$\frac{1}{2}{I}_{{\lambda }_n}^{\prime }(u_n)\phi =\frac{1}{2}{\lambda }_n〈u_n^{+},{\phi }^{+}〉-\frac{1}{2}〈u_n^{-},{\phi }^{-}〉-\frac{1}{2}{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),\phi )dt=0,\mathrm{\forall }\phi \in E,$$

it follows from the definition of I that

$$\begin{array}{rcl}{I}_{{\lambda }_n}\left(r{u}_n^{+}\right)-I_{{\lambda }_n}(u_n)& =& \frac{1}{2}{\lambda }_n(r^2-1){\parallel u_n^{+}\parallel }^2+\frac{1}{2}{\parallel u_n^{-}\parallel }^2+{\int }_{\mathbb{R}}(H(t,u_n)-H(t,r{u}_n^{+}))dt\\ +\frac{1}{2}{\lambda }_n〈u_n^{+},{\phi }^{+}〉-\frac{1}{2}〈u_n^{-},{\phi }^{-}〉-\frac{1}{2}{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),\phi )dt.\end{array}$$

(2.10)

Take $\phi :=(r^2+1)u_n^{-}-(r^2-1)u_n^{+}=(r^2+1)u_n-2r^2{u}_n^{+}$ in (2.10), then it follows from Lemma 2.8 that

$$\begin{array}{rcl}{I}_{{\lambda }_n}\left(r{u}_n^{+}\right)-I_{{\lambda }_n}(u_n)& =& -\frac{r^2}{2}{\parallel u_n^{-}\parallel }^2+{\int }_{\mathbb{R}}[H(t,u_n)-H(t,r{u}_n^{+})+r^2(\mathrm{\nabla }H(t,u_n),u_n^{+})\\ -\frac{1+r^2}{2}(\mathrm{\nabla }H(t,u_n),u_n)]dt\\ \le & C.\end{array}$$

Thus (2.9) holds.

Let $C_0>0$ be a fixed constant and take

$$r_n:=\frac{C_0}{\parallel u_n\parallel }\to 0\text{as }n\to \mathrm{\infty }.$$

Therefore, (2.9) implies that

$$I_{{\lambda }_n}\left(r_n{u}_n^{+}\right)-I_{{\lambda }_n}(u_n)\le C.$$

It follows from $v_n^{+}=\frac{u_n^{+}}{\parallel u_n\parallel }$ and Lemma 2.7 that

$$I_{{\lambda }_n}\left(C_0{v}_n^{+}\right)\le C^{\prime }.$$

(2.11)

Note that Lemmas 2.3 and 2.7 and (H₄) imply that

$$\begin{array}{rcl}0& \le & \frac{c_{{\lambda }_n}}{{\parallel u_n\parallel }^2}=\frac{I_{{\lambda }_n}(u_n)}{{\parallel u_n\parallel }^2}=\frac{{\lambda }_n}{2}{\parallel v_n^{+}\parallel }^2-\frac{1}{2}{\parallel v_n^{-}\parallel }^2-\frac{{\int }_{\mathbb{R}}H(t,u_n)}{{\parallel u_n\parallel }^2}dt\\ \le & \frac{{\lambda }_0}{2}{\parallel v_n^{+}\parallel }^2-\frac{1}{2}{\parallel v_n^{-}\parallel }^2+\frac{{\int }_{\mathbb{R}}{W}_1(t)dt}{{\parallel u_n\parallel }^2}.\end{array}$$

It follows from the fact $\frac{{\int }_{\mathbb{R}}{W}_1(t)dt}{{\parallel u_n\parallel }^2}\to 0$ as $n\to \mathrm{\infty }$ due to $W_1(t)\in L^1(\mathbb{R},{\mathbb{R}}^{+})$ that

$$\frac{{\lambda }_0}{2}{\parallel v_n^{+}\parallel }^2-\frac{1}{2}{\parallel v_n^{-}\parallel }^2+\epsilon \ge 0,\mathrm{\forall }\epsilon >0$$

(2.12)

for all sufficiently large n. We take $\epsilon =\frac{1}{4}$, by (2.12) and ${\parallel v_n\parallel }^2={\parallel v_n^{+}\parallel }^2+{\parallel v_n^{-}\parallel }^2=1$, we have

$${\parallel v_n^{+}\parallel }^2\ge \frac{1}{2(1+{\lambda }_0)}$$

(2.13)

for all sufficiently large n. By (H₁) and (H₃), we have

$$\begin{array}{c}{\int }_{\mathbb{R}}H(t,C_0{v}_n^{+})dt\hfill \\ \le \frac{1}{2}\gamma C_0^2{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|<\delta \}}|v_n^{+}{|}^{2}dt+\frac{1}{2}c{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|\ge \delta \}}\left(C_0\right|v_n^{+}|+C_0^p|v_n^{+}{|}^p)dt\hfill \\ \le \frac{1}{2}\gamma C_0^2{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|<\delta \}}|v_n^{+}{|}^{2}dt+C_1{C}_0^p{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|\ge \delta \}}|v_n^{+}{|}^{p}dt.\hfill \end{array}$$

(2.14)

For all sufficiently large n, by (2.13) and (2.14), it follows from ${\lambda }_n\to \lambda$ and $v_n^{+}\to v^{+}\equiv 0$ in $L^q(\mathbb{R},{\mathbb{R}}^N)$ for all $1\le q\le \mathrm{\infty }$ (by (2.1)) that

$$\begin{array}{rcl}{I}_{{\lambda }_n}\left(C_0{v}_n^{+}\right)& =& \frac{1}{2}{\lambda }_n{C}_0^2{\parallel v_n^{+}\parallel }^2-{\int }_{\mathbb{R}}H(t,C_0{v}_n^{+})dt\\ \ge & \frac{1}{2}{\lambda }_n{C}_0^2\frac{1}{2(1+{\lambda }_0)}-\frac{1}{2}\gamma C_0^2{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|<\delta \}}|v_n^{+}{|}^{2}dt\\ -C_1{C}_0^p{\int }_{\{t\in \mathbb{R}:|C_0{v}_n^{+}|\ge \delta \}}|v_n^{+}{|}^{p}dt\\ \to & \frac{\lambda C_0^2}{4(1+{\lambda }_0)}\text{as }n\to \mathrm{\infty }.\end{array}$$

This implies that $I_{{\lambda }_n}(C_0{v}_n^{+})\to \mathrm{\infty }$ as $C_0\to \mathrm{\infty }$, contrary to (2.11).

Therefore, $\{u_n\}$ are bounded. The proof is finished. □

3 Proofs of the main results

Proof of Theorem 1.1 From Lemma 2.7, there are sequences $1<{\lambda }_n\to 1$ and $\{u_n\}\subset E$ such that $I_{{\lambda }_n}^{\prime }(u_n)=0$ and $I_{{\lambda }_n}(u_n)=c_{{\lambda }_n}$. By Lemma 2.9, we know that $\{u_n\}$ is bounded in E. Thus we can assume $u_n\rightharpoonup u$ in E, $u_n\to u$ a.e. on ℝ. Therefore,

$$I_{{\lambda }_n}^{\prime }(u_n)\phi ={\lambda }_n〈u_n^{+},\phi 〉-〈u_n^{-},\phi 〉-{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),\phi )dt=0,\mathrm{\forall }\phi \in E.$$

Hence, in the limit,

$$I^{\prime }(u)\phi =〈u^{+},\phi 〉-〈u^{-},\phi 〉-{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u),\phi )dt=0,\mathrm{\forall }\phi \in E.$$

Thus $I^{\prime }(u)=0$. Note that

$$I_{{\lambda }_n}(u_n)-\frac{1}{2}{I}_{{\lambda }_n}^{\prime }(u_n)u_n={\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt=c_{{\lambda }_n}\ge c_1.$$

(3.1)

Similar to (2.7) and (2.8), we know

$${\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt\to {\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u),u)-H(t,u))dt\text{as }n\to \mathrm{\infty }.$$

It follows from $I^{\prime }(u)=0$, (3.1) and Lemma 2.3 that

$$\begin{array}{rcl}I(u)& =& I(u)-\frac{1}{2}{I}^{\prime }(u)u={\int }_{\mathbb{R}}[\frac{1}{2}(\mathrm{\nabla }H(t,u),u)-H(t,u)]dt\\ =& \underset{n\to \mathrm{\infty }}{lim}{\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt\ge c_1\ge ϵ>0.\end{array}$$

Therefore, $u\ne 0$. □

Proof of Theorem 1.2 By Theorem 1.1, $\mathcal{M}\ne \mathrm{\varnothing }$, where ℳ is the collection of solutions of (1.1). Let

$$\alpha :=\underset{u\in \mathcal{M}}{inf}I(u).$$

If u is a solution of (1.1), then by Lemma 2.8 (take $r=0$),

$$I(u)=I(u)-\frac{1}{2}{I}^{\prime }(u)u={\int }_{\mathbb{R}}[\frac{1}{2}(\mathrm{\nabla }H(t,u),u)-H(t,u)]dt\ge -C=-{\int }_{\mathbb{R}}|W_1(t)|dt.$$

Thus $\alpha >-\mathrm{\infty }$. Let $\{u_n\}$ be a sequence in ℳ such that

$$I(u_n)\to \alpha .$$

(3.2)

By Lemma 2.9, the sequence $\{u_n\}$ is bounded in E. Therefore, $u_n\rightharpoonup u$ in E, $u_n\to u$ a.e. on ℝ and $u_n\to u$ in $L^p(\mathbb{R},{\mathbb{R}}^N)$ for all $p\in [1,+\mathrm{\infty }]$ (by (2.1)), after passing to a subsequence. Therefore,

$$I^{\prime }(u_n)\phi =〈u_n^{+},\phi 〉-〈u_n^{-},\phi 〉-{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),\phi )dt=0,\mathrm{\forall }\phi \in E.$$

Hence, in the limit,

$$I^{\prime }(u)\phi =〈u^{+},\phi 〉-〈u^{-},\phi 〉-{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u),\phi )dt=0,\mathrm{\forall }\phi \in E.$$

Thus $I^{\prime }(u)=0$. Similar to (2.7) and (2.8), we have

$$\begin{array}{rcl}I(u_n)-\frac{1}{2}{I}^{\prime }(u_n)u_n& =& {\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt\\ \to & {\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u),u)-H(t,u))dt\text{as }n\to \mathrm{\infty }.\end{array}$$

It follows from $I^{\prime }(u)=0$ and (3.2) that

$$\begin{array}{rcl}I(u)& =& I(u)-\frac{1}{2}{I}^{\prime }(u)u={\int }_{\mathbb{R}}[\frac{1}{2}(\mathrm{\nabla }H(t,u),u)-H(t,u)]dt\\ =& \underset{n\to \mathrm{\infty }}{lim}{\int }_{\mathbb{R}}(\frac{1}{2}(\mathrm{\nabla }H(t,u_n),u_n)-H(t,u_n))dt\\ =& \underset{n\to \mathrm{\infty }}{lim}I(u_n)=\alpha .\end{array}$$

Now suppose that

$$|\mathrm{\nabla }H(t,u)|=o(|u|)\text{as }|u|\to 0.$$

It follows from (H₁) that for any $\epsilon >0$, there is a constant $C_{\epsilon }>0$ such that

$$|\mathrm{\nabla }H(t,u)|\le \epsilon |u|+C_{\epsilon }{|u|}^{p-1}.$$

(3.3)

Let

$$\beta :=\underset{u\in M^{\prime }}{inf}I(u),$$

where $M^{\prime }:=\mathcal{M}\setminus \{0\}$. Let $\{u_n\}$ be a sequence in $\mathcal{M}\setminus \{0\}$ such that

$$I(u_n)\to \beta .$$

(3.4)

Note that

$$0=I^{\prime }(u_n)u_n^{+}={\parallel u_n^{+}\parallel }^2-{\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),u_n^{+})dt,$$

which together with (3.3), Hölder’s inequality and the Sobolev embedding theorem implies

$$\begin{array}{rcl}{\parallel u_n^{+}\parallel }^2& =& {\int }_{\mathbb{R}}(\mathrm{\nabla }H(t,u_n),u_n^{+})dt\\ \le & \epsilon {\int }_0^T|u_n|\cdot \left|u_n^{+}\right|dt+C_{\epsilon }{\int }_0^T{|u_n|}^{p-1}\left|u_n^{+}\right|dt\\ \le & \epsilon \parallel u_n\parallel \cdot \parallel u_n^{+}\parallel +C_{\epsilon }^{\prime }{\parallel u_n\parallel }_{L^p}^{p-1}\parallel u_n^{+}\parallel \\ \le & \epsilon \parallel u_n\parallel \cdot \parallel u_n^{+}\parallel +C_{\epsilon }^{″}{\parallel u_n\parallel }_{L^p}^{p-2}\parallel u_n\parallel \cdot \parallel u_n^{+}\parallel \\ \le & \epsilon {\parallel u_n\parallel }^2+C_{\epsilon }^{″}{\parallel u_n\parallel }_{L^p}^{p-2}{\parallel u_n\parallel }^2.\end{array}$$

(3.5)

Similarly, we have

$${\parallel u_n^{-}\parallel }^2\le \epsilon {\parallel u_n\parallel }^2+C_{\epsilon }^{″}{\parallel u_n\parallel }_{L^p}^{p-2}{\parallel u_n\parallel }^2.$$

(3.6)

From (3.5) and (3.6), we get

$${\parallel u_n\parallel }^2\le 2\epsilon {\parallel u_n\parallel }^2+2C_{\epsilon }^{″}{\parallel u_n\parallel }_{L^p}^{p-2}{\parallel u_n\parallel }^2,$$

which means ${\parallel u_n\parallel }_{L^p}\ge C$ for some constant $C>0$. Since $u_n\to u$ in $L^p(\mathbb{R},{\mathbb{R}}^N)$, we know $u\ne 0$. As before, $I(u_n)\to I(u)=\beta$ as $n\to \mathrm{\infty }$. □