Abstract
In this paper, with the help of potential function, we extend the classical Brezis–Lieb lemma on Euclidean space to graphs, which can be applied to the following Kirchhoff equation
on a connected locally finite graph \(G=(\mathbb {V}, \mathbb {E})\), where \(b, \lambda >0\), \(p>2\) and V(x) is a potential function defined on \(\mathbb {V}\). The purpose of this paper is four-fold. First of all, using the idea of the filtration Nehari manifold technique and a compactness result based on generalized Brezis–Lieb lemma on graphs, we prove that there admits a positive solution \(u_{\lambda , b} \in E_\lambda \) with positive energy for \(b \in (0, b^*)\) when \(2<p<4\). In the sequel, when \(p \geqslant 4\), a positive ground state solution \(w_{\lambda , b} \in E_\lambda \) is also obtained by using standard variational methods. What’s more, we explore various asymptotic behaviors of \(u_{\lambda , b}, w_{\lambda , b} \in E_\lambda \) by separately controlling the parameters \(\lambda \rightarrow \infty \) and \(b \rightarrow 0^{+}\), as well as jointly controlling both parameters. Finally, we utilize iteration to obtain the \(L^{\infty }\)-norm estimates of the solution.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Consider the following Kirchhoff type equation
with a bounded domain \(\Omega \subset \mathbb {R}^N\), which received much attention on mathematical studies, was originally introduced by Kirchhoff [18] to describe the transversal oscillations of a stretched string. Above equation is a general version of the Kirchhoff equation
with a strong physical background, where L is the length of the string, h is the area of cross section, \(\rho \) is the mass density, E is the Young’s modulus of the material, and \(P_0\) is the initial tension. Equation (1.2) can be also seen as a generalization of the classical D’Alembert’s wave equation for free vibrations of elastic strings. For more details in the physical and mathematical background of Kirchhoff type equations, one can refer to [23]. In addition, some early classical studies of Kirchhoff equations can be seen in Bernstein [4]. Afterwards, Lions [20] proposed an abstract functional analysis framework for the Kirchhoff type equation (1.1) in 1978, which attracted the attention of several researchers. Subsequently, based on variational methods, the solvability of (1.1) has been thoroughly investigated when nonlinear f satisfies various growth conditions, see [9, 24]. Moreover, there are numerous interesting results were built on the corresponding elliptic version, like
where \(N \geqslant 1\), \(a, b>0\), \(V \in C\left( \mathbb {R}^N, \mathbb {R}\right) \) and \(f \in C\left( \mathbb {R} \times \mathbb {R}^N, \mathbb {R}\right) \). To deal with (1.3) in the case where \(f(x, u):=f(u)\) with \(N=3\), via using the mountain pass theorem and Nehari manifold, He and Zou [17] established the existence of positive ground state solutions. Later, Li and Ye [19] obtained the existence of positive ground state solutions when \(f(x, u):=|u|^{p-2} u\) with \(3<p<6\) in (2.2). For more results about Kirchhoff type problems, we refer the readers to [7, 8, 21, 22, 27,28,29,30] and the references therein.
In recent years, analysis on graphs has begun to attract attention and has had several applications in different fields, data analysis, optimal transport, machine learning, etc [1, 10, 12]. Particularly, there are many interesting directions about graph in mathematics. It’s remarkable that, Grigor’yan et al. [13, 15, 16] established the variational framework on graphs for some nonlinear equations, and the existence of solution was considered, where mountain pass theorem and a method of upper and lower solutions have been uesd. Furthermore, they pointed out that the Sobolev space \(W^{m, p}(\mathbb {V} )\) on a finite graph \(G=(\mathbb {V}, \mathbb {E} )\) is pre-compact. Soon after, Zhang and Zhao [33] introduced two Sobolev embedding theorems on graphs and investigated the equation
where \(V(x) \geqslant 0\) satisfies following conditions:
\(\left( V_1\right) \ \) \(V(x) \geqslant 0\) on \(\mathbb {V} \), and the potential well \(\Omega =\left\{ x \in \mathbb {V}: V(x)=0\right\} \) is a non-empty, connected and bounded domain in \(\mathbb {V}\).
\(\left( V_2\right) \ \) There exists a vertex \(x_0 \in \mathbb {V}\) such that \(V(x) \rightarrow +\infty \) as \(d\left( x, x_0\right) \rightarrow +\infty \).
Specifically, they proved that there exists a ground state solution \(u_\lambda \) of (1.4) and it converges in \(W^{1,2}(\mathbb {V})\) to a ground state solution of the Dirichlet problem
as \(\lambda \rightarrow \infty \) along a subsequence. For more interesting resuts on graphs, we refer the readers to [2, 3, 6, 26].
Now, We turn our attention to the Kirchhoff equation. Pan and Ji [25] used the constrained variational method to study
on a locally finite graph, where f(u) satisfies super cubic growth condition and the other suitable assumptions. More precisely, the authors showed that (1.5) has a least energy sign-changing solution \(u_b\), and its energy is strictly larger than twice that of least energy solutions firstly. Then, they proved that \(u_b\) converges to a least energy sign-changing solution of the problem
as \(b \rightarrow 0^+\) along a subsequence.
Based on the above works, for the Kirchhoff equation on graphs, there is a lack of relevant results when the nonlinear terms are associated with subcubic growth. Thus, in the following article, we attempt to investigate the existence of positive solutions to the Kirchhoff type problem
on a connected locally finite graph \(G=(\mathbb {V}, \mathbb {E})\), where \(a>0\) is a constant, b and \(\lambda \) are positive parameters, \(p>2\) and the potential V(x) satisfies conditions \((V_1)\), \((V_2)\).
We shall consider (1.6) for \(2<p<4\) and \(p\geqslant 4\) separately. For the former, in order to overcome the barrier of lacking compactness on graphs, we introduce two embedding theorems based on steep potential. Moreover, to obtain compactness results for \(2<p<4\), we extend Brezis–Lieb lemma on graphs. For the latter, we use standard Nehari manifold to seek the positive ground state solution for the equation (2.3). What’s more, various asymptotic behaviors and the \(L^{\infty }\)-norm estimates of the solution are considered.
1.1 Notations
Let \(G=(\mathbb {V}, \mathbb { E})\) be a graph which is locally finite and connected, where \(\mathbb {V}\) and \(\mathbb { E}\) denote the vertex set and the edge set, respectively. Here, a graph G is said to be locally finite if for any \(x \in \mathbb {V}\), there are only finite \(y \in \mathbb {V}\) such that \(x y \in \mathbb { E}\). And a graph is connected if any two vertices x and y can be connected via finite edges. For any \(x, y \in \mathbb { V}\) with \(x y \in \mathbb { E}\), we assume it has a positive symmetric weight on G, namlely \(w_{x y}=w_{y x} >0\).
Then we define the measure \(\mu : \mathbb {V} \rightarrow \mathbb {R}^{+}\) on the graph, which is a finite positive function on \(\mathbb {V}\). And we call it a uniformly positive measure if there exists a constant \(\mu _{0 }>0\) such that \(\mu (x) \geqslant \mu _{0 }\) for all \(x \in \mathbb {V}\). For any function \(u: \mathbb {V} \rightarrow \mathbb {R}\), the \(\mu \)-Laplacian (or Laplacian for short) of u is defined as
where \(y \sim x\) means \(x y \in \mathbb { E}\) or y is adjacent to x. We denote the gradient form of two functions u and v on the graph by
Denote \(\Gamma (u)=\Gamma (u, u)\) and \(\nabla u \nabla v=\Gamma (u, v)\), then the length of gradient u is represented by
The integral of a function f over \(\mathbb {V}\) is given as
We denote the space of functions on \(\mathbb {V}\) by \(C(\mathbb {V})\). For \(u \in C(\mathbb {V})\), its support set is defined as \({\text {supt}}(u)= \{x \in \mathbb {V}: u(x) \ne 0\}\). Let \(C_c(\mathbb {V})\) be the set of all functions with finite support and \(W^{1,2}(\mathbb {V})\) be the completion of \(C_c(\mathbb {V})\) under the norm
Obviously, \(W^{1,2}(\mathbb {V})\) is a Hilbert space with the inner product
What’s more, the space \(L^p(\mathbb {V})\) is defined as
where
Consider a domain \(\Omega \subset \mathbb {V}\). The distance d(x, y) of two vertices \(x, y \in \Omega \) is defined by the minimal length of a path which connect x and y. \(\Omega \) is said a bounded domain in \(\mathbb {V}\), if the distance d(x, y) is bounded for any \(x, y \in \Omega \). The boundary of \(\Omega \) is defined as
and the interior of \(\Omega \) is denoted by \(\Omega ^{\circ }\). Obviously, we can see that \(\Omega ^{\circ }=\Omega \), which is different from the Euclidean case. Furthermore, the Hilbert space \(W_0^{1,2}(\Omega )\) is the completion of \(C_c(\Omega )\) under the norm
Because of the formula for integral by parts which will be introduced in Sect. 2, there is an extral integral on \(\partial \Omega \) for the gradient form of u.
And for more details about graphs, we refer the reader to [14].
Furthermore, if conditions \((V_1), (V_2)\) are satisfied, we denote \(m_\Omega \) the ground state energy of the equation
where \(p>2\) and \(\Omega =\left\{ x \in \mathbb {V}: V(x)=0\right\} \).
1.2 Main Results
After explaining the notations, we introduce the main results achieved. In this paper, we always assume that \(G=(\mathbb {V}, \mathbb { E})\) is a locally finite and connected graph with positive symmetric weight and uniformly positive measure. We consider Kirchhoff type problem (1.6) satisfying the conditions \((V_1)\) and \((V_2)\). To state our main results, without loss of generality, we may assume that \(a=1\). Consequently, we shall investigate the Kirchhoff type problem
with the associated energy functional
Then it is natural to consider a function space
with the norm
The space \(E_\lambda \) is also a Hilbert space and its inner product is
We now summarize our main results as follows. The first main result is concerned with the existence of positive solutions.
Theorem 1.1
(i) Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Then there exists \(b^*>0\) such that for every \(b\in (0,b^*)\) and \(\lambda > 0\), Eq. \((\mathcal {K}_{ \lambda ,b})\) has one positive solution \(u_{\lambda , b} \in E_\lambda \) satisfying
and
(ii) Suppose that \(p=4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold. Then there exists \({\hat{b}}>0\) such that for every \(b \in (0, {\hat{b}})\), \(\lambda >0,\) Eq. \((\mathcal {K}_{ \lambda ,b})\) has a positive ground state solution \(w_{\lambda , b}\in E_{\lambda }\) satisfying \(I_{\lambda , b}\left( w_{\lambda , b}\right) =l_{\lambda , b}>0\).
(iii) Suppose that \(p>4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold. Then for every \(\lambda >0\), \(b>0\), Eq. \((\mathcal {K}_{ \lambda ,b})\) has a positive ground state solution \(w_{\lambda , b}\in E_{\lambda }\) satisfying \(I_{\lambda , b}\left( w_{\lambda , b}\right) =l_{\lambda , b}>0\).
Remark 1.1
To illustrate that the condition \(\mu _0 \geqslant \frac{2 p}{p-2} m_{\Omega }\) can be achieved, we consider a finite connected graph \(G_6=(\mathbb {V}, \mathbb { E})\) as shown in the Fig. 1. The vertex set \(\mathbb {V} \) is \(\left\{ x_1, x_2, \cdots , x_6\right\} \) and the edge set \(\mathbb {E}\) is \(\left\{ x_{12}, x_{13}, x_{23}, x_{24}, x_{25}, x_{34}, x_{35}, x_{45}, x_{46}, x_{56}\right\} \), where \(x_{i j}\) represents the edge connecting vertices \(x_i\) and \(x_j\). For simplicity, we take the measure \(\mu \) satisfying \(\mu \left( x_i\right) =1\) for \(i=1,2, \cdots , 6\) and take the weight \(w_{x_i x_j}=w_{x_j x_i}=1\) for all \(x_{i j} \in \mathbb {E}\). Hence, \(G_6\) is a finite and connected graph with positive symmetric weight and uniformly positive measure.
Next, let V(x) be
We consider the equation (1.8) in a Hilbert space \(W_0^{1, 2}(\Omega )\), with corresponding functional
Clearly, the potential well is \(\Omega =\left\{ x_1\right\} \) with boundary \(\partial \Omega =\left\{ x_2, x_3\right\} \). Note that
satisfying \(u_1(x) \in \mathcal {N}_{\Omega }\), and so
where
Thus, we can take \(\mu _0=1\) satisfying \(\mu _0 \geqslant \frac{2 p}{p-2} m_{\Omega }\).
Note that Zhang and Du [32] applied the truncation technique to handle the similar Kirchhoff equation in \(\mathbb {R}^3\) space successfully, however, this technique is not applicable to graphs. Therefore, we need to identify an alternative approach. Inspired by Sun and Wu [29], we apply a novel constraint method, which will be introduced in Sect. 3, to obtain the boundedness of the Palais-Smale sequences. Furthermore, to overcome the compactness, we extend the classical Brezis–Lieb lemma [5] on Euclidean space to a connected locally finite graph in Sect. 2.2.
To study the asymptotic behavior of \(u_{\lambda , b}, w_{\lambda , b}\in E_\lambda \) as \(\lambda \rightarrow \infty \), it is advisable to consider the Dirichlet problem
Similar to Eq. \((\mathcal {K}_{ \lambda ,b})\), the Dirichlet problem \(\left( \mathcal {K}_{ \infty ,b}\right) \) also has a positive solution under some assumptions.
Theorem 1.2
-
(i)
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Then there exists \(b^*>0\) such that for every \( b \in (0, b^*)\), Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) has one positive solution \(u_{\infty , b} \in W_0^{1,2}\left( \Omega \right) \).
-
(ii)
Suppose that \(p=4\) and conditions \((V_1), (V_2)\) hold. Then there exists \({\hat{b}}>0\) such that for every \(b \in (0, {\hat{b}})\), Eq. \((\mathcal {K}_{ \infty ,b})\) has a positive ground state solution \(w_{\infty , b} \in W_0^{1,2}\left( \Omega \right) \).
-
(iii)
Suppose that \(p>4\) and conditions \((V_1), (V_2)\) hold. Then for every \(b>0\), Eq. \((\mathcal {K}_{ \infty ,b})\) has a positive ground state solution \(w_{\infty , b} \in W_0^{1,2}\left( \Omega \right) \).
Remark 1.2
For convenience, we use the same notations \(b^*, {\hat{b}}\) here as in Theorem 1.1.
Next, we have the following result:
Theorem 1.3
-
(i)
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Then there exists \(b_* \in \left( 0, b^*\right) \), for any sequence \(\lambda _n \rightarrow \infty \), pass to a subsequence, \(u_{\lambda _n,b}\in E_{\lambda _n}\) converges in \(W^{1,2}(\mathbb {V})\) to a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) with \(b\in \left( 0,b_*\right) \) fixed, where \(u_{\lambda _n,b}\in E_{\lambda _n}\) is the positive solution of Eq. \((\mathcal {K}_{ \lambda _n,b})\) obtained by Theorem 1.1(i).
-
(ii)
Suppose that \(p=4\), and conditions \((V_1), (V_2)\) hold. Then, for any sequence \(\lambda _n \rightarrow \infty \), pass to a subsequence, \(w_{\lambda _n,b}\in E_{\lambda _n}\) converges in \(W^{1,2}(\mathbb {V})\) to a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) with \(b \in (0, {\hat{b}})\) fixed, where \(w_{\lambda _n,b}\in E_{\lambda _n}\) is the positive ground state solution of Eq. \((\mathcal {K}_{ \lambda _n,b})\) obtained by Theorem 1.1(ii).
-
(iii)
Suppose that \(p>4\), and conditions \((V_1), (V_2)\) hold. Then, for any sequence \(\lambda _n \rightarrow \infty \), pass to a subsequence, \(w_{\lambda _n,b}\in E_{\lambda _n}\) converges in \(W^{1,2}(\mathbb {V})\) to a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) with \(b \in (0, +\infty )\) fixed, where \(w_{\lambda _n,b}\in E_{\lambda _n}\) is the positive ground state solution of Eq. \((\mathcal {K}_{ \lambda _n,b})\) obtained by Theorem 1.1(iii).
After exploring the asymptotic behavior of \(u_{ \lambda ,b}, w_{ \lambda ,b} \in E_\lambda \) as \(\lambda \rightarrow \infty \), we turn our attention to study asymptotic behavior of \(u_{ \lambda ,b}, w_{ \lambda ,b} \in E_\lambda \) as \(b\rightarrow 0^+\), thus
is considered naturally. According to [33], if conditions \(\left( V_1\right) ,\left( V_2\right) \) are satisfied, Eq. \((\mathcal {K}_{ \lambda ,0})\) has a positive solution for \(p>2\).
Theorem 1.4
-
(i)
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Let \(u_{ \lambda ,b}\in E_{\lambda }\) be the positive solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by theorem 1.1(i). Then for each \(\lambda \in \left( 0, \infty \right) \) fixed, up to a subsequence, \(u_{ \lambda ,b} \rightarrow u_{ \lambda ,0}\) in \(E_\lambda \) as \(b \rightarrow 0^+\), where \(u_{\lambda ,0} \in E_\lambda \) is a positive solution of Eq. \((\mathcal {K}_{ \lambda ,0})\).
-
(ii)
Suppose that \(p=4\) and conditions \((V_1), (V_2)\) hold. Let \(w_{ \lambda ,b}\in E_{\lambda }\) be the positive ground state solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by Theorem 1.1(ii). Then for each \(\lambda \in \left( 0, \infty \right) \) fixed, up to a subsequence, \(w_{ \lambda ,b} \rightarrow w_{ \lambda ,0}\) in \(E_\lambda \) as \(b \rightarrow 0^+\), where \(w_{\lambda ,0} \in E_\lambda \) is a positive solution of Eq. \((\mathcal {K}_{ \lambda ,0})\).
-
(iii)
Suppose that \(p>4\) and conditions \((V_1), (V_2)\) hold. Let \(w_{ \lambda ,b}\in E_{\lambda }\) be the positive ground state solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by Theorem 1.1(iii). Then for each \(\lambda \in \left( 0, \infty \right) \) fixed, up to a subsequence, \(w_{ \lambda ,b} \rightarrow w_{ \lambda ,0}\) in \(E_\lambda \) as \(b \rightarrow 0^+\), where \(w_{\lambda ,0} \in E_\lambda \) is a positive solution of Eq. \((\mathcal {K}_{ \lambda ,0})\).
Next, we present a theorem that describes the asymptotic behavior of \(u_{ \lambda ,b}, w_{ \lambda ,b} \in E_\lambda \) as \(\lambda \rightarrow \infty \) and \(b\rightarrow 0^+\). According to [33], if conditions \(\left( V_1\right) ,\left( V_2\right) \) are satisfied, then
has a positive solution for \(p>2\).
Theorem 1.5
-
(i)
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Let \(u_{ \lambda ,b}\in E_{\lambda }\) be the positive solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by Theorem 1.1(i). Then up to a subsequence, \(u_{ \lambda ,b} \rightarrow \) \(u_{ \infty ,0}\) in \(W^{1,2}(\mathbb {V})\) as \(b \rightarrow 0^+\) and \(\lambda \rightarrow \infty \), where \(u_{ \infty ,0} \in W_0^{1,2}\left( \Omega \right) \) is a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,0}\right) \).
-
(ii)
Suppose that \(p=4\), and conditions \((V_1), (V_2)\) hold. Let \(w_{ \lambda ,b}\in E_{\lambda }\) be the positive ground state solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by Theorem 1.1(ii). Then up to a subsequence, \(w_{ \lambda ,b} \rightarrow w_{ \infty ,0}\) in \(W^{1,2}(\mathbb {V})\) as \(b \rightarrow 0^+\) and \(\lambda \rightarrow \infty \), where \(w_{ \infty ,0} \in W_0^{1,2}\left( \Omega \right) \) is a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,0}\right) \).
-
(iii)
Suppose that \(p>4\), and conditions \((V_1), (V_2)\) hold. Let \(w_{ \lambda ,b}\in E_{\lambda }\) be the positive ground state solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b}\right) \) obtained by Theorem 1.1(iii). Then up to a subsequence, \(w_{ \lambda ,b} \rightarrow w_{ \infty ,0}\) in \(W^{1,2}(\mathbb {V})\) as \(b \rightarrow 0^+\) and \(\lambda \rightarrow \infty \), where \(w_{ \infty ,0} \in W_0^{1,2}\left( \Omega \right) \) is a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,0}\right) \).
Finally in this subsection, we give the estimate of solutions.
Theorem 1.6
Let \(u_{\lambda , b}, w_{\lambda , b} \in E_\lambda \) be obtained by Theorem 1.1. Then there exist \(C_1,C_2>0\) (independent of \(\lambda )\) satisfying
for all \(\lambda >0\).
The paper is organized as follows: in Sect. 1, we describe the development of Kirchhoff type equation and some research results about it on Euclidean Space. Following that, the research results of discrete equation in recent years are presented. What’s more, we provide an explanation of main notations and present the main results achieved, including the existence of positive solution and its asymptotic behavior for Eq. \((\mathcal {K}_{ \lambda , b})\). In Sect. 2, in order to derive the main results, we undertake essential preparations. This includes introducing the formula for integral by parts and the Sobolev embedding theorem on graphs. Subsequently, we introduce a generalized Brezis–Lieb lemma on graphs. In Sect. 3, we filtrate the Nehari manifold \({\textbf{N}}_{\lambda , b}\) associated with Eq. \((\mathcal {K}_{ \lambda , b})\) laying the foundation of the proof in next section. In Sect. 4, we show that the existence of the positive solution \(u_{\lambda , b} \in E_\lambda \) for Eq. \((\mathcal {K}_{ \lambda , b})\) when \(2<p<4\), where we have employed calculus of variations. In Sect. 5, we prove the multiple results regarding the asymptotic behavior of \(u_{\lambda , b} \in E_\lambda \) by controlling parameters \(\lambda \) and b. In Sect. 6, by applying standard variational methods, we show that the existence of the positive ground state solution \(w_{\lambda , b} \in E_\lambda \) for Eq. \((\mathcal {K}_{ \lambda , b})\) when \(p \geqslant 4\). In Sect. 7, we analyze the asymptotic behavior of the solution \(w_{\lambda , b} \in E_\lambda \) by controlling parameters \(\lambda \) and b. In Sect. 8, we give the estimate involving the \(L^{\infty }\)-norm of solutions by using iteration.
2 Preliminaries
We will introduce some useful results on graphs.
2.1 Formula for Integral by Parts and Embedding Theorems
In this subsection, we shall present two lemmas about integral by parts on graphs first. According to the work of Zhang and Zhao [33], we have the following lemmas.
Lemma 2.1
Suppose that \(u \in W^{1,2}(\mathbb {V})\) and its Laplacian \(\Delta u\) is defined by (1.7). Then for any \(v \in C_c(\mathbb {V})\), we have
Lemma 2.2
Suppose that \(u \in W^{1,2}(\mathbb {V})\) and its Laplacian \(\Delta u\) is well-defined. Let v be a function which belongs to \(C_c(\Omega )\), where \(\Omega \subset \mathbb {V}\) is a bounded domain. Then we have
After that, since the graph has no concept of dimension, the Sobolev embedding theorem becomes unusual. For this, we introduce two compactness results related to \(E_\lambda \).
Lemma 2.3
Assume that \(\lambda >0\) and V(x) satisfies conditions \(\left( V_1\right) \), \(\left( V_2\right) \). Then \(E_\lambda \) is weakly pre-compact and \(E_\lambda \) compactly embedded into \(L^q(\mathbb {V})\) for any \(q \in [2, \infty ]\) and the embedding is independent of \(\lambda \). Namely, there exists a constant C depending on \( \mu _0\) and q such that for any \(u \in E_\lambda ,\ \Vert u\Vert _{L^q(\mathbb {V})} \leqslant C\Vert u\Vert _\lambda \). Particularly, there holds
Moreover, for any bounded sequence \(\left\{ u_n\right\} \subset E_\lambda \), there exists \(u \in E_\lambda \) such that, going if necessary to a subsequence,
Proof
We assume that \(u \in E_\lambda \), take any vertex \(x_1 \in \mathbb {V}\) and fix it, there holds
which implies
Thus \(E_\lambda \hookrightarrow L^{\infty }(\mathbb {V})\) continuously and the embedding is independent of \(\lambda \). When \(2 \leqslant q<+\infty \), we have
so we get \(\Vert u\Vert _\lambda \geqslant \mu _0^{\frac{q-2}{2 q}}\Vert u\Vert _{L^q(\mathbb {V})}\). Then we obtain the continuous embedding \(E_\lambda \hookrightarrow L^q(\mathbb {V})\) for any \(2 \leqslant q \leqslant \infty \).
The rest of the proof is similar to the lemma 2.6 in [33], so we omit it here. \(\square \)
For the space \(W_0^{1,2}(\Omega )\), we have another embedding theorem as following.
Lemma 2.4
(See [33], Lemma 2.7) Assume that \(\Omega \) is a bounded domain in \(\mathbb {V}\). Then \(W_0^{1,2}(\Omega )\) is continuously embedded into \(L^q(\Omega )\) for any \(q \in [1, \infty ]\). Namely, there exists a constant C depending only on q and \(\Omega \) such that for any \(u \in W_0^{1,2}(\Omega ),\Vert u\Vert _{L^q ( \Omega )} \leqslant C\Vert u\Vert _{W_0^{1,2}(\Omega )}\). Moreover, for any bounded sequence \(\left\{ u_n\right\} \subset W_0^{1,2}(\Omega )\), there exists \(u \in W_0^{1,2}(\Omega )\) such that, going if necessary to a subsequence,
2.2 A Generalization of Brezis–Lieb Lemma
After introducing two embedding theorems, we establish a generalization of Brezis–Lieb lemma on graphs as following.
Lemma 2.5
Suppose that conditions \(\left( V_1\right) \) and \(\left( V_2\right) \) hold. If \(\left\{ u_n\right\} \subset E_\lambda \) is bounded and there exists \(u \in E_\lambda \) such that \( u_n \rightharpoonup u\), then there holds
for \(2 \leqslant q <\infty \).
Proof
We assume that \(\left\{ u_n\right\} \) has a upper bound M in \( E_\lambda \), then
Fix \(\varepsilon >0\), there exists \(C_\varepsilon >0\) such that, for all \(\alpha ,\beta \in \mathbb {R}\),
If we let
and take \(\alpha =u_{n}-u,\ \beta =u\), then
so there exists a constant C such that
In view of \(\left( V_2\right) \), let \(x_0 \in \mathbb {V}\) be fixed, there exists some \(R>0\) such that
Hence, we obatin
Since \(\left\{ u_n\right\} \subset E_\lambda \) is also bounded in \(L^2(\mathbb {V})\) and we have the weak convergence in \(L^2(\mathbb {V})\), then
for any \(\varphi \in L^2(\mathbb {V})\). Take any \(x_1 \in \mathbb {V}\) and let
which belongs to \(L^2(\mathbb {V})\). By substituting \(\varphi _1\) into (2.3), we get \(\lim \limits _{n \rightarrow \infty } \mu \left( x_1\right) [ u_n(x_1)-u(x_1)] =0\). Thus, \(\lim \limits _{n \rightarrow \infty } u_n(x)=u(x)\) for any \(x \in \mathbb {V}\), and so \(f_{n}^{\varepsilon }(x) \rightarrow 0\) for any \(x \in \mathbb {V}\). Moreover, because \(\left\{ x\in \mathbb {V}: {\text {dist}}\left( x, x_0\right) \leqslant R \right\} \) is a finite set, there holds
We conclude from (2.2) and (2.4) that \(\lim \limits _{n \rightarrow +\infty } \int _\mathbb {V} \left| f_{n}^{\varepsilon }\right| d \mu =0\), then
Note that
we obtain
Now let \(\varepsilon \rightarrow 0\), there exists
which implies that
We finish the proof now. \(\square \)
Remark 2.1
If \(\left\| u_n\right\| _{W_0^{1, 2}(\Omega )}\) is bounded and there exists \(u \in W_0^{1,2}(\Omega )\) satisfying \(u_n \rightharpoonup u\), then \(\lim \limits _{n \rightarrow \infty } u_n(x)=u(x)\) for all \(x \in \Omega \) is obviously, which implies that
for \(q \geqslant 1\).
3 The Construction of a Noval Constraint Manifold
To find the solutions of Eq. \((\mathcal {K}_{ \lambda ,b})\), it is naturally to consider the critical points of the functional \(I_{ \lambda ,b}(u)\). A direct calculation shows that
for all \(\varphi \in E_\lambda \), where \(I_{\lambda , b}^{\prime }\) denotes the Fréchet derivative of \(I_{\lambda , b}\). Next, we can define the Nehari manifold
then \(u \in {\textbf{N}}_{\lambda , b}\) if and only if \(\Vert u\Vert _\lambda ^2+b\left( \int _{\mathbb {V}}|\nabla u|^2 d \mu \right) ^2=\int _{\mathbb {V}} |u|^p d \mu \).
Motived by Sun and Wu [29], we filtrate the Nehari manifold \( {\textbf{N}}_{\lambda , b}\) with \(2<p<4\). For each \(r \in [2, +\infty ) \), suppose that conditions \(\left( V_1\right) \) and \(\left( V_2\right) \) hold, in view of (2.1), we find
Hence, for \(u \in {\textbf{N}}_{\lambda , b}\), there holds
which implies that
It’s noteworthy that the Nehari manifold \({\textbf{N}}_{\lambda , b}\) is closely linked to the behavior of fibrering map of the form \(h_{b, u}: t \rightarrow I_{\lambda , b}(t u)\) as
For \(u \in E_\lambda \), it is easy to see
and
Thus, for \(u \in E_\lambda \backslash \{0\}\) and \(t>0\), \( h_{b, u}^{\prime }(t)=0\) holds if and only if \(t u \in {\textbf{N}}_{\lambda , b}\). Especially, we can see that \(h_{b, u}^{\prime }(1)=0\) holds if and only if \(u \in {\textbf{N}}_{\lambda , b}\). According to [31], we may decompose \( {\textbf{N}}_{\lambda , b}\) into three disjoint parts
Then we have the following result.
Lemma 3.1
Suppose that \(u_0\) is a local minimizer for \(I_{\lambda , b}\) on \({\textbf{N}}_{\lambda , b}\) and \(u_0 \notin {\textbf{N}}_{\lambda , b}^0\). Then \(I_{\lambda , b}^{\prime }\left( u_0\right) =0\) in \({E_\lambda }^{-1}\).
Proof
If \(u_0\) is a local minimizer for \(I_{\lambda , b}\) on \({\textbf{N}}_{\lambda , b}\), then \(u_0\) is a solution of the optimization problem
where \(\gamma (u)=\Vert u\Vert _\lambda ^2+b\left( \int _{\mathbb {V}}|\nabla u|^2 d \mu \right) ^2-\int _{\mathbb {V}} |u|^p d \mu \). It follows from Lagrange multiplier rule that \(I_{\lambda , b}^{\prime }\left( u_0\right) =\theta \gamma ^{\prime }\left( u_0\right) \) for some \(\theta \in \mathbb {R}\). Hence, we have
For \(u_0 \in {\textbf{N}}_{\lambda , b}\), there holds
The condition \(u_0 \notin {\textbf{N}}_{\lambda , b}^0\) means that
So combining (3.2) and (3.3) gives
Thus, by (3.2) and (3.4), we have
which implies that \(\theta =0\), and so \(I_{\lambda , b}^{\prime }\left( u_0\right) =0\). The proof is completed. \(\square \)
Lemma 3.2
Suppose that \(2<p<4\) and conditions \(\left( V_1\right) \), \(\left( V_2\right) \) hold. Then \(I_{\lambda , b}\) is coercive and bounded below on \({\textbf{N}}_{\lambda , b}^{-}\). Furthermore, for all \(u \in {\textbf{N}}_{\lambda , b}^{-}\), there holds
Proof
Note that
for all \(u \in {\textbf{N}}_{\lambda , b}\). So combining (3.1) and (3.5) implies that
for all \(u \in {\textbf{N}}_{\lambda , b}^{-}\). This completes the proof. \(\square \)
Next, we attempt to do more analysis with \({\textbf{N}}_{\lambda , b}^{-}\). Suppose that conditions \((V_1), (V_2) \) hold. For any \(u \in {\textbf{N}}_{\lambda , b}\) with \(I_{\lambda , b}(u)< \frac{m_{\Omega }}{2}\left( \frac{2}{4-p}\right) ^{\frac{2}{p-2}} \), it is easy to obtain
Hence, if \(0<b<\frac{(p-2)^2}{2 p m_\Omega (4-p)}\left( \frac{4-p}{2}\right) ^{\frac{2}{p-2}}\), one has two positive numbers \(D_1\) and \(D_2\) satisfying
such that
Clearly, one can see that \(D_1 \rightarrow \infty \) as \(p \rightarrow 2^{+}\) or \(p \rightarrow 4^{-}\). Therefore, one has
where
and
Moreover, for \(0<b<\frac{(p-2)^2}{2 p m_\Omega (4-p)}\left( \frac{4-p}{2}\right) ^{\frac{2}{p-2}}\), there holds
If \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \), combining (3.5) and (3.7) gives
for all \( u \in {\textbf{N}}_{\lambda , b}^{(1)} \), and this shows that \( {\textbf{N}}_{\lambda , b}^{(1)} \subset {\textbf{N}}_{\lambda , b}^{-}\). Therefore, the following statement is true.
Lemma 3.3
Suppose that \(2<p<4\), \(0<b<\frac{(p-2)^2}{2 p m_\Omega (4-p)}\left( \frac{4-p}{2}\right) ^{\frac{2}{p-2}}\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2) \) hold, then \({\textbf{N}}_{\lambda , b}^{(1)} \subset {\textbf{N}}_{\lambda , b}^{-}\) is sub-manifold. Moreover, each local minimizer of the functional \(I_{\lambda , b}\) in the sub-manifold \({\textbf{N}}_{\lambda , b}^{(1)}\) is a critical point of \(I_{\lambda , b}\) in \(E_\lambda \).
For \(u \in E_\lambda \backslash \{0\}\), we define
for convenience.
Lemma 3.4
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. For every \(b>0\) and \(u \in E_\lambda \backslash \{0\}\) satisfying
there exist two constants \(t_{b}^{+}\)and \(t_{b}^{-}\)which satisfy
such that \(t_{b}^{ \pm } u \in {\textbf{N}}_{\lambda , b}^{ \pm }\).
Proof
For each \(u \in E_\lambda \backslash \{0\}\) and \(t>0\), we define
then
This implies that \(t u \in {\textbf{N}}_{\lambda , b}\) if and only if \(m(t)+b\left( \int _{\mathbb {V} }|\nabla u|^{2} d\mu \right) ^{2}=0\). Based on a calculation, there hold
and
Note that m(t) is decreasing when \(0<t<\left( \frac{2}{4-p}\right) ^{\frac{1}{p-2}} T_{\lambda } (u)\) and is increasing when \(t>\left( \frac{2}{4-p}\right) ^{\frac{1}{p-2}} T_{\lambda }(u)\). Hence, when
we have
here we have applied \(\left( \frac{p}{2}\right) ^{\frac{2}{p-2}}>1\). Then there exist two constants \(t_{b}^{+}, t_{b}^{-}>0\) satisfying
such that
and so \(t_{b}^{ \pm } u \in {\textbf{N}}_{\lambda , b}\). Note that
then through calculation, one has
Similarly, there holds \(h_{b, t_b^{+} u}^{\prime \prime }(1)>0\) and so \(t_b^{ \pm } u \in {\textbf{N}}_{\lambda , b}^{ \pm }\). Then the proof is completed. \(\square \)
Next we focus on whether \({\textbf{N}}_{\lambda , b}^{(1)}\) is non-empty. To see this, let’s consider the Dirichlet problem (1.8) in a Hilbert space \(W_0^{1, 2}(\Omega )\), with corresponding functional
According to [33], suppose that \(p>2\) and conditions \((V_1)\), \((V_2)\) hold, then there exists a ground solution \(w_{\Omega }\) satisfying
where
Furthermore, we obtain
and
Then for
it has
Consider lemma 3.4, it is easy to obtain positive constants \(t_{b}^{-}\) and \(t_{b}^{+}\) satisfying
such that \(t_{b}^{ \pm } w_{\Omega } \in {\textbf{N}}_{\lambda , b}^{ \pm }\), where
Moreover, when \( 1<t_b^{-}<\left( \frac{2}{4-p}\right) ^{\frac{1}{p-2}}, \) there hold
and
so \(t_b^{-} w_\Omega \in {\textbf{N}}_{\lambda , b}^{(1)}\). Thus, we obtain the following statement.
Lemma 3.5
Suppose that \(2<p<4\), \(\mu _0 > \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. When
the sub-manifold \({\textbf{N}}_{\lambda , b}^{(1)}\) is non-empty.
4 The Proof of Theorem 1.1(i)
Now, we are ready to investigate the compactness for the functional \(I_{\lambda , b}(u)\). In the following key proposition, we shall show that for c sufficiently small, the sequence \(\left\{ u_n\right\} \subset {\textbf{N}}_{\lambda , b}^{(1)}\) satisfying \(I_{\lambda , b}\left( u_n\right) \rightarrow c\) and \(\left\| I_{\lambda , b}^{\prime }\left( u_n\right) \right\| _{E_\lambda ^{-1}} \rightarrow 0\) has a convergent subsequence.
Proposition 4.1
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \(\left( V_1\right) , \left( V_2\right) \) hold. Then \(I_{\lambda , b}\) satisfies \((P S)_c\)-condition in \({\textbf{N}}_{\lambda , b}^{(1)}\) with \(c<\frac{m_{\Omega }}{2}\left( \frac{2}{4-p}\right) ^{\frac{2}{p-2}}\) for all \(\lambda >0\) and \(0<b<b^*\).
Proof
Let \(\left\{ u_n\right\} \subset {\textbf{N}}_{\lambda , b}^{(1)}\) be a \((P S)_c\)-sequence for \(I_{\lambda , b}\) with \(c<\frac{m_{\Omega }}{2}\left( \frac{2}{4-p}\right) ^{\frac{1}{p-2}}\). Then we have \(\left\| u_n\right\| _\lambda <D_1\) by (3.7). Passing to a subsequence if necessary, there exists \(u_0 \in E_\lambda \) such that
where we have used Lemma 2.3. Then, we aim to prove that \(v_n:=u_n-u_0 \rightarrow 0\) strongly in \(E_\lambda \). We conclude from Lemma 2.5 that
Since \(\left\| u_n\right\| _\lambda \) is bounded, we infer that, up to a subsequence,
where A is a positive constant.
For any \(\varphi \in E_\lambda \), by (4.1) and (4.2),
Then take \(\varphi =u_0\) in above equality, there holds
By \(u_n\in {\textbf{N}}_{\lambda , b}^{(1)} \subset {\textbf{N}}_{\lambda , b}\), we have
Combining (4.4), (4.5) and (4.6), we yield that
It’s easy to see \(\left\| v_n\right\| _\lambda \rightarrow 0\), where we have applied (4.3). Then the proof is completed. \(\square \)
According to lemma 3.2, 3.3 and 3.5, we can define
when \(0<b<b^*\). By using (3.6) and (3.8), we deduce that
Obviously, there exists a sequence \(\left\{ u_{n}\right\} \subset {\textbf{N}}_{\lambda , b}^{(1)}\) such that
and
from the Ekeland variational principle [11].
We have made sufficient preparations to prove the theorem 1.1(i) and we give the proof now.
It from (4.7), (4.8), (4.9) and proposition 4.1 that \(I_{\lambda , b}\) satisfies the \((PS)_{ m_{\lambda , b} }\) condition in \({E_\lambda }\) for all \(0<b<b^*\) and \(\lambda >0\). So we can find a subsequence \(\left\{ u_{n}\right\} \) and \(u_{\lambda , b} \in E_{\lambda }\) such that \(u_{n} \rightarrow u_{\lambda , b} \) strongly in \(E_{\lambda }\) for all \(0<b<b^{*}\) and \(\lambda > 0\). Thus, \(u_{\lambda , b}\in E_{\lambda } \) is a minimizer for \(I_{\lambda , b}\) in \({\textbf{N}}_{\lambda , b}^{(1)}\). Since \(\left| u_{\lambda , b} \right| \in {\textbf{N}}_{\lambda , b}^{(1)}\) and
we can derive that \(u_{\lambda , b}\in E_{\lambda } \) is a non-negative solution for Eq. \((\mathcal {K}_{\lambda ,b})\) easily, by using lemma 3.3. If \( u_{\lambda , b}\left( x_1\right) =0\) for some \(x_1 \in \mathbb {V}\), then \(\Delta u_{\lambda , b} \left( x_1\right) =0\), i.e.
which means \(\sum \limits _{x \sim x_1} w_{x x_1} u_{\lambda , b} (x)=0\) with \(w_{x x_1}>0, u_{\lambda ,b } (x) \geqslant 0\) for \(x \sim x_1\). Hence there holds \(u_{\lambda , b}(x)=0\) for all \(x \sim x_1\), we deduce from the arbitrariness of \(x_1\) that \(u_{\lambda , b} (x) \equiv 0\) in \(\mathbb {V}\), which implies that \(u_{\lambda , b} \) is a positive solution. Moreover, we deduce from (3.1) that any critical point \(u_{\lambda , b} \in {\textbf{N}}_{\lambda , b}^{(1)} \subset {\textbf{N}}_{\lambda , b}\) satisfying \(\Vert u_{\lambda , b}\Vert _\lambda \geqslant \mu _0^{\frac{1}{2}}>0\). The proof is now finished. \(\square \)
5 Asymptotic Behavior of Positive Solution \(u_{\lambda , b}\)
After obtaining the existence of the positive solution \(u_{\lambda , b} \in E_{\lambda }\) of Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \) for \(2<p<4\), we turn to study the asymptotic behavior of \(u_{\lambda , b} \in E_{\lambda }\) obtained by the theorem 1.1(i).
5.1 Asymptotic Behavior as \(\lambda \rightarrow \infty \)
In this subsection, we investigate the asymptotic behavior of \(u_{\lambda , b} \in E_{\lambda }\) of Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \) as \(\lambda \rightarrow \infty \) and give the proof of theorem 1.2(i) and 1.3(i).
To deal with Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \), it is naturally to consider
with corresponding Nehari manifold
Similar to the filtration of Nehari manifold \({\textbf{N}}_{\lambda , b}\), we can define
and
Just as what we have done for Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \), we can get a positive solution \(u_{\infty , b}\in W_0^{1,2}(\Omega ) \) of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) for \(2<p<4,\) which achieves \(m_{\infty , b}\) of the functional \( I_{\infty , b}(u)\) in \( {\textbf{N}}_{\infty , b}^{(1)}\). Thus, we prove the theorem 1.2(i). \(\square \)
Before proving the theorem 1.3(i), we establish the following lemma.
Lemma 5.1
Suppose that \(2<p<4\), \(\mu _0 \geqslant \frac{2p}{p-2} m_\Omega \) and conditions \((V_1), (V_2)\) hold. Then there exists \(b_* \in (0, b^*)\) such that \( m_{\lambda , b} \rightarrow m_{\infty , b}\) as \(\lambda \rightarrow \infty \) for all \(b \in \left( 0, b_* \right) \) fixed.
Proof
It is easy to see \(m_{\lambda , b} < m_{\infty , b}\) for any \(\lambda >0\). If not, we can find a nontrivial solution \(u_{\lambda , b} \in E_{\lambda }\) of Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \) which vanishes outside \(\Omega \). So we have \( u_{\lambda , b}\left( x_0\right) =0\) for some \(x_0 \in \mathbb {V}\), obviously there holds \(\Delta u_{\lambda , b} \left( x_0\right) =0\). In view of the maximum principle, we may deduce that \(u_{\lambda , b} (x) \equiv 0\) in \(\mathbb {V}\), which leads to a contradiction.
Taking a sequence \(\lambda _n \rightarrow \infty \) such that
where \(m_{\lambda _{n, b}}\) is associated with the positive solution \(u_{\lambda _n,b} \in {\textbf{N}}_{\lambda _{n, b}}^{(1)}\) of Eq. \(\left( \mathcal {K}_{\lambda _{n}, b}\right) \) obtained by theorem 1.1(i). We may deduce from (4.7) that \(M >0\). In addition, \(\left\{ u_{\lambda _n,b}\right\} \) is uniformly bounded in \(E_{\lambda }\). Going if necessary to a subsequence, we have \(u_0 \in E_{\lambda }\) such that
and for any \(q \geqslant 2\),
by applying lemma 2.3. Then we shall prove that \( u_0 |_{\Omega ^c}=0\), otherwise we can assume \(u_0\left( x_1\right) \ne 0\) for some \(x_1 \notin \Omega \). In view of \(u_{\lambda _n,b} \in {\textbf{N}}_{\lambda _{n, b}}^{(1)}\), there holds
and then \(\lim \limits _{n \rightarrow \infty } I_{\lambda _n, b}\left( u_{\lambda _n,b}\right) =+\infty \) by using \(V\left( x_1\right) >0\), \(u_{\lambda _n,b}\left( x_1\right) \rightarrow u_0(x_1) \ne 0\) and \(\mu \left( x_1\right) >0\). Hence, we get a contradiction to the fact that \(m_{\lambda _n,b}<m_{\infty , b}< +\infty \). Furthermore, in view of (4.7), \(u_{0} \not \equiv 0\) is clearly.
Consider that \(\Omega \) is a finite set and (5.2), we get
then
where we have used \(\left. u_0\right| _{\Omega ^c}=0\). Notice that
by (5.4) and \(\left. u_0\right| _{\Omega ^c}=0\). We obtain \(t \in (0,1]\) such that \(t u_0 \in {\textbf{N}}_{\infty , b}^{(1)}\) for b sufficiently small since \(\int _{\Omega \cup \partial \Omega }\left| \nabla u_0\right| ^2 d \mu \) has a bound independent of parameter b, i.e.
Hence, the following two cases are considered: \(0<t<1\) and \(t=1\).
For the former, by (5.4), we have
for b small enough, which contradicts \( \lim \limits _{n \rightarrow \infty } m_{\lambda _n, b}=M \leqslant m_{\infty , b} \).
For the latter, we deduce from (5.4) that
and
which implies
So the combination of (5.5) and (5.6) gives
and
According to (5.7) and (5.8), there holds
Consequently, we conclude from the above two cases that \(m_{\infty , b} =M\), and so \(\lim \limits _{n \rightarrow \infty } m_{\lambda _n, b}= m_{\infty , b}\). This completes the proof. \(\square \)
Based on the above discussion, the theorem 1.3(i) can be proved. Next, We need to prove that for any sequence \(\lambda _{n} \rightarrow \infty \), the positive solution \(u_{\lambda _n,b} \in {\textbf{N}}_{\lambda _{n},b} ^{(1)}\) of Eq. \(\left( \mathcal {K}_{ \lambda _n,b}\right) \) satisfying \(I_{\lambda _{n},b}\left( u_{\lambda _n,b}\right) = m_{\lambda _n, b}\) converges in \(W^{1,2}(\mathbb {V})\) to a positive solution \(u_{\infty , b}\in W_0^{1,2}(\Omega ) \) of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \) obtained by theorem 1.2(i) along a subsequence.
Similar to the discussion in the above lemma, there hold (5.1), (5.2), (5.3), \(u_0 |_{\Omega ^c}=0\) and \(u_{0} \not \equiv 0\) obviously. Then we aim to show that as \(n \rightarrow \infty \), up to a subsequence, there hold
and
If not, let’s analyze it in two cases that (5.9) and (5.10) aren’t valid respectively.
For the former, we assume that (5.9) doesn’t hold, then \( \lim \limits _{n \rightarrow \infty } \lambda _{n} \int _{\mathbb {V}} V(x) u_{\lambda _n,b}^{2} d \mu =\delta _1>0, \) and so
where we have used (5.4) and \(u_0|_{\Omega ^c}=0\).
For the latter, we assume that (5.10) doesn’t hold, then
with \(\delta _2>0,\) thus
by applying (5.4) and \(u_0|_{\Omega ^c}=0\).
We deduce from (5.11) and (5.12) that there exists \(t \in (0,1)\) such that \(t u_0 \in {\textbf{N}}_{\infty , b}^{(1)}\) with a sufficiently small parameter b, i.e.
Thus, in view of (5.9) and (5.10), we get
for b sufficiently small, which leads to a contradiction. So we find that \(u_{0}\) is a solution of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) .\) Furthermore, lemma 5.1 gives that \(u_{0}\) is a positive solution of Eq. \(\left( \mathcal {K}_{ \infty ,b}\right) \), which achieves \(m_{\infty , b}\) of the functional \( I_{\infty , b}(u)\) in \( {\textbf{N}}_{\infty , b}^{(1)}\). This completes the proof. \(\square \)
Remark 5.1
According to the proof of lemma 5.1 and the proof of this theorem, we can take the same \(b_* \in \left( 0, b^* \right) \) in lemma 5.1 and this theorem.
5.2 Asymptotic Behavior as \(b \rightarrow 0^+\)
In this subsection, we focus on reasearching asymptotic behavior of \(u_{ \lambda ,b} \in E_\lambda \) as \(b\rightarrow 0^+\). Based on the theorem 1.1(i), we give the proof of theorem 1.4(i).
Fix \(\lambda \in \left( 0, \infty \right) \), for any sequence \(b_n \rightarrow 0\), let \(u_{ \lambda ,b_n}\in E_\lambda \) be the positive solution of Eq. \(\left( \mathcal {K}_{ \lambda ,b_n}\right) \) obtained by theorem 1.1(i). Note that
going to a subsequence if necessary, there exists \(u_0 \in E_\lambda \) such that \( u_{\lambda , b_n} \rightharpoonup u_0 \) in \(E_\lambda \). As we have discussed in Sect. 4, we can deduce \( u_{\lambda , b_n} \rightarrow u_0 \) in \(E_\lambda \). Now we just have to prove that \(u_0\) is a positive solution of Eq. \(\left( \mathcal {K}_{ \lambda ,0}\right) \). Since \(\left\langle I_{ \lambda ,b_n}^{\prime }\left( u_{\lambda ,b_n}\right) , \varphi \right\rangle =0\) for all \(\varphi \in E_\lambda \), it is easy to check that
which implies that \(u_0\) is a nonnegative solution of Eq. \(\left( \mathcal {K}_{\lambda ,0}\right) \). And we have \(u_0 \ne 0\) by (5.13). Furthermore, consider the maximum principle, we may obtain \(u_0>0\) in \(\mathbb {V}\). The proof is now finished. \(\square \)
5.3 Asymptotic Behavior as \(\lambda \rightarrow \infty \) and \(b \rightarrow 0^+\)
Finally, we explore the asymptotic behavior of the positive solution \(u_{ \lambda ,b}\) as \(\lambda \rightarrow \infty \) and \(b \rightarrow 0^+\) in this subsection. And we are ready to give the proof of theorem 1.5(i).
Fix a sufficiently small b, for any sequence \(\lambda _n \rightarrow \infty \), we may obtain the positive solution \(u_{\lambda _n, b}\) of Eq. \(\left( \mathcal {K}_{\lambda _{n, b}}\right) \) by using theorem 1.1(i). Passing to a subsequence, \(u_{\lambda _n, b}\in E_{\lambda _n}\) converges to the positive solution \(u_{\infty , b}\in W_0^{1,2}(\Omega )\) of Eq. \(\left( \mathcal {K}_{\infty , b}\right) \) as \(\lambda _n \rightarrow \infty \) by applying theorem 1.3(i). Then, in view of the proof of theorem 1.4(i), we derive that \(u_{\infty , b}\in W_0^{1,2}(\Omega )\) converges to a positive solution of Eq. \(\left( \mathcal {K}_{\infty , 0}\right) \) as \(b \rightarrow 0^{+}\).
Alternatively, up to a subsequence, we can consider fixing \(\lambda \in (0, \infty )\) and letting \(b \rightarrow 0^{+}\) subsequently, after then letting \(\lambda \rightarrow \infty \). We may get similar result and the proof of theorem 1.5(i) is complete. \(\square \)
6 The Proof of Theorem 1.1(ii) and (iii)
For \(p \geqslant 4\), it is easy to prove that the existence of a ground state solution for Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \) by standard variational methods. First, we give several useful preliminary results about energy functional \(I_{\lambda , b}\) and Nehari manifold \({\textbf{N}}_{\lambda , b}\).
Lemma 6.1
-
(i)
Suppose that \(p=4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold. Then there exists \({\hat{b}}>0\) such that for any \(b \in (0, {\hat{b}})\) and any \(\lambda >0\), \( {\textbf{N}}_{\lambda , b}\) is non-empty.
-
(ii)
Suppose that \(p>4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold, Then \( {\textbf{N}}_{\lambda , b}\) is non-empty for any \(b>0, \lambda >0\).
Proof
-
(i)
For any \(u \in E_\lambda \backslash \{0\}\), there holds
$$\begin{aligned} \left\langle I_{\lambda , b}^{\prime }(t u), t u\right\rangle =t^2\Vert u\Vert _\lambda ^2+b t^4\left( \int _\mathbb {V}|\nabla u|^2 d \mu \right) ^2-t^p \int _\mathbb {V}|u|^p d \mu . \end{aligned}$$Hence, for sufficiently small b, it can be find \(t_0 \in (0,+\infty )\) such that \(\left\langle I_{\lambda , b}^{\prime }\left( t_0 u\right) , t_0 u\right\rangle =0\), which means that \( {\textbf{N}}_{\lambda , b}\) is non-empty.
-
(ii)
Similarly, we can prove (ii). \(\square \)
Lemma 6.2
-
(i)
Suppose that \(p=4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold, then \(l_{\lambda , b}:=\inf \limits _{u \in {\textbf{N}}_{\lambda , b}} I_{\lambda , b}(u)>0\) for any \(b \in (0, {\hat{b}})\) and \(\lambda >0\).
-
(ii)
Suppose that \(p>4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold, then \(l_{\lambda , b}>0\) for any \(b>0\) and \(\lambda >0\).
Proof
-
(i)
Consider (3.1), we deduce that
$$\begin{aligned} \begin{aligned} I_{\lambda , b}(u)&=\frac{1}{2}\Vert u\Vert _\lambda ^2+\frac{b}{4}\left( \int _\mathbb {V}|\nabla u|^2 d \mu \right) ^2-\frac{1}{p} \int _\mathbb {V}|u|^p d \mu \\&=\left( \frac{1}{2}-\frac{1}{4}\right) \Vert u\Vert _\lambda ^2+\left( \frac{1}{4}-\frac{1}{p}\right) \int _\mathbb {V}|u|^p d \mu \\&\geqslant \frac{1}{4}\Vert u\Vert _\lambda ^2 \geqslant \frac{1}{4} \mu _0 \end{aligned} \end{aligned}$$for any \(u \in {\textbf{N}}_{\lambda , b}\), so \(l_{\lambda , b} \geqslant \frac{1}{4} \mu _0>0\).
-
(ii)
We may prove (ii) similarly. \(\square \)
Lemma 6.3
Suppose that \(p \geqslant 4\) and conditions \(\left( V_1\right) ,\left( V_2\right) \) hold. Then for each \(\lambda > 0\), \(b>0\), \(I_{\lambda , b}\) satisfies the \((P S)_c\) condition in \({\textbf{N}}_{\lambda , b}\) for any \(c \in \mathbb {R}\).
Proof
To prove that \(I_{\lambda , b}\) satisfies \((P S)_c\) condition, we can assume \(\left\{ u_n\right\} \subset E_\lambda \) such that
and
for any \(\varphi \in E_\lambda \). Replacing \(\varphi \) by \(u_n\) in (6.2), there holds
Combining (6.1) and (6.3) gives
which implies that \(\left\{ u_n\right\} \) is bounded in \(E_\lambda \). Then we omit the rest of the proof due to its similarity with proposition 4.1. Now, the proof is completed. \(\square \)
Based on the above discussion, We may provide the proof of theorem 1.1(ii) and (iii). In view of the Ekeland variational principle [11], there exists a sequence \(\left\{ u_{n}\right\} \subset {\textbf{N}}_{\lambda , b}\) such that
and
obviously. We yield from lemmas 6.1, 6.2, 6.3 and (6.4), (6.5) that there exists \(w_{\lambda , b} \in E_{\lambda }\) such that \(u_{n} \rightarrow w_{\lambda , b} \) strongly in \(E_{\lambda }\). Hence, \(w_{\lambda , b} \in E_{\lambda }\) is a minimizer for \(I_{\lambda , b}\) in \({\textbf{N}}_{\lambda , b}\). Note that \(\left| w_{\lambda , b} \right| \in {\textbf{N}}_{\lambda , b}\) and
we obtain a positive ground solution \(w_{\lambda , b} \in E_{\lambda }\) for Eq. \((\mathcal {K}_{ \lambda ,b})\) satisfying \(I_{\lambda , b}\left( w_{\lambda , b}\right) =l_{\lambda , b}>0\) by the maximum principle. This ends the proof of theorem 1.1(ii) and (iii). \(\square \)
Similar to the above discussion, we can obtain the positive ground solution \(w_{\infty , b} \in W_0^{1,2}\left( \Omega \right) \) of Eq. \(\left( \mathcal {K}_{\infty , b}\right) \) for \(p \geqslant 4\). And the remaining proof of the theorem 1.2 is omitted. \(\square \)
7 Asymptotic Behavior of Positive Ground Solution \(w_{\lambda , b}\)
After studying the existence of positive ground state solution \(w_{\lambda , b} \in E_{\lambda }\) for Eq. \(\left( \mathcal {K}_{\lambda , b}\right) \), just as what we have discussed in Sect. 5, we may investigate the asymptotic behavior of \(w_{\lambda , b} \in E_{\lambda }\) analogously, so we omit the proof of remaining part of theorems 1.3, 1.4, 1.5 here.
8 The Estimate of the Solution \(u_{\lambda , b}, w_{\lambda , b} \in E_\lambda \)
After exploring the existence and the asymptotic behavior of the solution \(u_{\lambda , b}, w_{\lambda , b} \in E_\lambda \), we are interested in the estimate involving the \(L^{\infty }\)-norm of solutions. For this purpose, we have the following research. And the proof of theorem 1.6 is as following.
Let \(\Vert \cdot \Vert _q:=\Vert \cdot \Vert _{L^q(\mathbb {V})}\) for convenience. We assume \(m>0\) satisfying \(m>p\), then
for any \(\tau >1\), it follows
Set \(\sigma =\frac{m}{p}>1\). When \(\tau =\sigma \) in (8.1), there holds
Arguing by iteration, let \(\tau =\sigma ^j\) in (8.1), we can deduce that
Note that \(\left\| u_{\lambda , b}\right\| _{p+1} \leqslant \mu _0^{-\frac{1}{p(p+1)}}\left\| u_{\lambda , b}\right\| _p\), let \(j \rightarrow \infty \) and take \(m=p+1\) in (8.2), we find
Considering (2.1), (3.6) and \(m_{\lambda , b}<m_{\infty ,b}\), we obtain
Similarly, we may analyze the case of \(w_{\lambda , b} \in E_\lambda \). Hence, we finish the proof. \(\square \)
References
Abderrahim, E., Xavier, D., Zakaria, L., et al.: Nonlocal infinity Laplacian equation on graphs with applications in image processing and machine learning. Math. Comput. Simul. 102, 153–163 (2014)
Akduman, S., Pankov, A.: Nonlinear Schrödinger equation with growing potential on infinite metric graphs. Nonlinear Anal. 184, 258–272 (2019)
Bauer, F., Horn, P., Lin, Y., et al.: Li–Yau inequality on graphs. J. Differ. Geom. 99, 359–405 (2015)
Bernstein, S.: Sur une classe d’équations fonctionelles aux dérivées partielles. Bull. Acad. Sci. URSS. Sér. 4, 17–26 (1940)
Brézis, H., Lieb, E.: A relation between pointwise convergence of functions and convergence of functionals. Proc. Am. Math. Soc. 88, 486–490 (1983)
Chang, X., Rǎdulescu, V.D., Wang, R. et al: Convergence of least energy sign-changing solutions for logarithmic Schrödinger equations on locally finite graphs, Commun. Nonlinear Sci. Numer. Simul. 125, 107418 (2023)
Chen, W., Fu, Z., Wu, Y.: Positive solutions for nonlinear Schrödinger–Kirchhoff equations in \(R^3\). Appl. Math. Lett. 104, 106274 (2020)
Chen, W., Wu, Y.: Nontrivial solutions for 4-superlinear Schrödinger–Kirchhoff equations with indefinite potentials. J. Funct. Spaces 2021, 1–6 (2021)
D’Ancona, P., Shibata, Y.: On global solvability of non-linear viscoelastic equations in the analytic category. Math. Methods Appl. Sci. 17, 477–486 (1994)
Desquesnes, X., Elmoataz, A., Lézoray, O.: Eikonal equation adaptation on weighted graphs: fast geometric diffusion process for local and non-local image and data processing. J. Math. Imaging Vis. 46, 238–257 (2013)
Ekeland, I.: On the variational principle. J. Math. Anal. Appl. 47, 324–353 (1974)
Essid, M., Solomon, J.: Quadratically regularized optimal transport on graphs. SIAM J. Sci. Comput. 40, A1961–A1986 (2018)
Grigor’yan, A., Lin, Y., Yang, Y.: Yamabe type equations on graphs. J. Differ. Equ. 261, 4924–4943 (2016)
Grigor’yan, A.: Introduction to analysis on graphs. Am. Math. Soc. (2018)
Grigor’yan, A., Lin, Y., Yang, Y.: Kazdan–Warner equation on graph. Calc. Var. Partial Differ. Equ. 55, 1–13 (2016)
Grigor’yan, A., Lin, Y., Yang, Y.: Existence of positive solutions to some nonlinear equations on locally finite graphs. Sci. China Math. 60, 1311–1324 (2017)
He, X., Zou, W.: Existence and concentration behavior of positive solutions for a Kirchhoff equation in \(R^3\). J. Differ. Equ. 252, 1813–1834 (2012)
Kirchhoff, G.: Mechanik. Teubner, Leipzig (1883)
Li, G., Ye, H.: Existence of positive ground state solutions for the nonlinear Kirchhoff type equations in \(R^3\). J. Differ. Equ. 257, 566–600 (2014)
Lions, J.L.: On some questions in boundary value problems of mathematical physics. In: Contemporary Developments in Continuum Mechanics and Partial Differential Equations. North-Holl. Math. Stud., vol. 30, North-Holland, Amsterdam, New York, pp. 284–346 (1978)
Liu, Z., Guo, S.: Existence of positive ground state solutions for Kirchhoff type problems. Nonlinear Anal. 120, 1–13 (2015)
Liu, Z., Luo, H., Zhang, J.: Existence and multiplicity of bound state solutions to a Kirchhoff type equation with a general nonlinearity. J. Geom. Anal. 32, 125 (2022)
Mshimba, A.S.A., Tutschke, W.: Functional Analytic Methods in Complex Analysis and Applications to Partial Differential Equations. World Scientific (1995)
Nishihara, K.: On a global solution of some quasilinear hyperbolic equation. Tokyo J. Math. 7, 437–459 (1984)
Pan, G., Ji, C.: Existence and convergence of the least energy sign-changing solutions for nonlinear Kirchhoff equations on locally finite graphs. Asymptotic Anal. 133, 463–482 (2023)
Shao, M., Zhao, L., Yang, Y.: Sobolev spaces on locally finite graphs, arXiv preprint arXiv:2306.02262 (2023)
Shuai, W.: Sign-changing solutions for a class of Kirchhoff-type problem in bounded domains. J. Differ. Equ. 259, 1256–1274 (2015)
Sun, J., Wu, T.: Ground state solutions for an indefinite Kirchhoff type problem with steep potential well. J. Differ. Equ. 256, 1771–1792 (2014)
Sun, J., Wu, T.: Steep potential well may help Kirchhoff type equations to generate multiple solutions. Nonlinear Anal. 190, 111609 (2020)
Sun, J., Wu, T.: On the Kirchhoff type equations in \(\mathbb{R} ^{N}\). Adv. Differ. Equ. 27, 97–146 (2022)
Tarantello, G.: On nonhomogeneous elliptic equations involving critical Sobolev exponent. Ann. Inst. H. Poincaré Anal. Non Linéaire 9, 281–304 (1992)
Zhang, F., Du, M.: Existence and asymptotic behavior of positive solutions for Kirchhoff type problems with steep potential well. J. Differ. Equ. 269, 10085–10106 (2020)
Zhang, N., Zhao, L.: Convergence of ground state solutions for nonlinear Schrödinger equations on graphs. Sci. China Math. 61, 1481–1494 (2018)
Acknowledgements
H. Chen is supported by the National Natural Science Foundation of China (Grant No. 12071486). The authors are thankful for the reviewer who gives valuable comments and suggestions on this paper.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no Conflict of interest.
Additional information
Communicated by Rosihan M. Ali.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Cheng, S., Yao, S. & Chen, H. A Generalized Brezis–Lieb Lemma on Graphs and Its Application to Kirchhoff Type Equations. Bull. Malays. Math. Sci. Soc. 47, 141 (2024). https://doi.org/10.1007/s40840-024-01741-0
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40840-024-01741-0