Introduction
In this paper, we are interested in the following coupled fractional Schrödinger system with trapping potentials in R N \mathbb{R}^{N} R N :
{ ( − Δ ) s u 1 + V 1 ( x ) u 1 = μ u 1 + a 1 ∣ u 1 ∣ p − 2 u 1 + β ∣ u 1 ∣ q − 2 ∣ u 2 ∣ q u 1 , ( − Δ ) s u 2 + V 2 ( x ) u 2 = μ u 2 + a 2 ∣ u 2 ∣ p − 2 u 2 + β ∣ u 1 ∣ q ∣ u 2 ∣ q − 2 u 2 . (1) \left\{
\begin{aligned}
(-\Delta)^{s}u_{1}+V_{1}(x)u_{1} &= \mu u_{1}+a_{1}|u_{1}|^{p-2}u_{1}+\beta|u_{1}|^{q-2}|u_{2}|^{q}u_{1},\\
(-\Delta)^{s}u_{2}+V_{2}(x)u_{2} &= \mu u_{2}+a_{2}|u_{2}|^{p-2}u_{2}+\beta|u_{1}|^{q}|u_{2}|^{q-2}u_{2}.
\end{aligned}
\right. \tag*{(1)} { ( − Δ ) s u 1 + V 1 ( x ) u 1 ( − Δ ) s u 2 + V 2 ( x ) u 2 = μ u 1 + a 1 ∣ u 1 ∣ p − 2 u 1 + β ∣ u 1 ∣ q − 2 ∣ u 2 ∣ q u 1 , = μ u 2 + a 2 ∣ u 2 ∣ p − 2 u 2 + β ∣ u 1 ∣ q ∣ u 2 ∣ q − 2 u 2 . (1)
where 0 < s < 1 0 < s < 1 0 < s < 1 , N ≥ 1 N \ge1 N ≥ 1 , p = 2 + 4 s N p = 2 + \frac{4s}{N} p = 2 + N 4 s is the mass-critical exponent, and q = p 2 q = \frac{p}{2} q = 2 p . The parameter μ ∈ R \mu\in\mathbb{R} μ ∈ R is the chemical potential, which enters as the Lagrange multiplier associated with the mass constraint, whereas a i > 0 a_i > 0 a i > 0 (i = 1 , 2 i = 1, 2 i = 1 , 2 ) and β > 0 \beta> 0 β > 0 measure the strengths of the intraspecies and interspecies interactions, respectively; the positivity of these coefficients means that both types of interaction are attractive.
System (1.1) arises from a fundamental quantum mechanical model describing multicomponent Bose-Einstein condensates (BECs). Historically, the mathematical description of BECs, nonlinear optics, and many-body quantum physics has heavily relied on the classical (local) nonlinear Schrödinger and Gross-Pitaevskii (GP) systems [2 ] . In particular, two-component GP systems model binary BECs with intraspecies and interspecies interactions, where the sign and strength of the interactions determine whether the condensate is stable, collapses, or undergoes phase separation [13 ] . A natural nonlocal generalization is obtained when the particles in the condensate perform Lévy flights rather than Brownian motion: then the underlying kinematics are governed by the fractional Laplacian ( − Δ ) s (-\Delta)^s ( − Δ ) s [15 , 16 ] , which leads precisely to system (1.1). In the study of such physical models, seeking normalized solutions , namely solutions with a prescribed L 2 L^2 L 2 mass, has become a major theme because the mass directly corresponds to the number of particles, while the chemical potential naturally appears as a Lagrange multiplier. Before proceeding to the coupled fractional setting, it is instructive to recall the well-understood single-component theory, which provides the critical mass a ∗ a^* a ∗ and the ground state Q Q Q as basic building blocks.
The theory for a single component is by now well understood. For the GP equation in R 2 \mathbb{R}^2 R 2 with attractive interaction, Guo and Seiringer Guo and Seiringer [8 ] studied the mass concentration as the interaction strength a a a approaches the critical value a ∗ = ∥ Q ∥ 2 2 a^* = \lVert Q\rVert_2^2 a ∗ = ∥ Q ∥ 2 2 , where Q Q Q is the unique positive radial ground state of
− Δ u + u = u 3 in R 2 . (2) -\Delta u + u = u^3 \quad\text{in } \mathbb{R}^2. \tag*{(2)} − Δ u + u = u 3 in R 2 . (2)
They proved that minimizers exist if and only if 0 ≤ a < a ∗ 0 \le a < a^* 0 ≤ a < a ∗ , and that, as a ↗ a ∗ a \nearrow a^* a ↗ a ∗ , the minimizers blow up and concentrate at a global minimum of the trapping potential. Their proof relies on the sharp Gagliardo–Nirenberg inequality and delicate energy estimates. Later, Guo, Wang, Zeng and Zhou Guo, Wang, Zeng and Zhou [9 ] extended this analysis to multi-well potentials, showing that ground states concentrate and blow up at the center of the incircle of the well with the largest inradius, and they established symmetry breaking. For ring-shaped potentials, Guo, Zeng and Zhou Guo, Zeng and Zhou [10 ] proved that symmetry breaking occurs: for small a a a the minimizer is unique and radially symmetric, whereas for a a a close to a ∗ a^* a ∗ there are infinitely many non-radial minimizers concentrating at points on the circular bottom of the potential well. These works rely on the sharp Gagliardo–Nirenberg inequality in the local case, refined energy estimates, blow-up analysis, and the moving plane method.
These results have been extended to fractional Schrödinger equations. In the fractional setting, the corresponding ground state Q Q Q of
( − Δ ) s u + 2 s N u = ∣ u ∣ 4 s N u in R N (3) (-\Delta)^s u + \frac{2s}{N}u = |u|^{\frac{4s}{N}}u \quad\text{in } \mathbb{R}^N \tag*{(3)} ( − Δ ) s u + N 2 s u = ∣ u ∣ N 4 s u in R N (3)
decays only polynomially, Q ( x ) = O ( ∣ x ∣ − N − 2 s ) Q(x) = O(|x|^{-N-2s}) Q ( x ) = O ( ∣ x ∣ − N − 2 s ) , rather than exponentially, and the optimal constant in the fractional Gagliardo–Nirenberg inequality is more delicate. The uniqueness of the positive radial ground state Q Q Q was proved by Frank and Lenzmann Frank and Lenzmann [5 ] in one dimension and by Frank, Lenzmann and Silvestre Frank, Lenzmann and Silvestre [6 ] in higher dimensions, using nonlocal moving plane methods and the extension problem of Caffarelli and Silvestre. For fractional Schrödinger equations with trapping potentials, Du, Tian, Wang and Zhang [4 ] considered
( − Δ ) s u + V ( x ) u = μ u + a ∣ u ∣ p − 2 u , x ∈ R N , (4) (-\Delta)^s u + V(x)u = \mu u + a|u|^{p-2}u,\quad x \in\mathbb{R}^N, \tag*{(4)} ( − Δ ) s u + V ( x ) u = μu + a ∣ u ∣ p − 2 u , x ∈ R N , (4)
with 0 < s < 1 0 < s < 1 0 < s < 1 , N ≥ 2 N \ge2 N ≥ 2 , and p = 2 + 4 s / N p = 2 + 4s/N p = 2 + 4 s / N . They established the existence and non-existence of L 2 L^2 L 2 -normalized solutions by using a constrained variational method. They proved that minimizers exist if and only if 0 ≤ a < a ∗ 0 \le a < a^* 0 ≤ a < a ∗ , where a ∗ = ∥ Q ∥ 2 4 s / N a^* = \|Q\|_2^{4s/N} a ∗ = ∥ Q ∥ 2 4 s / N , and they analyzed the concentration behavior as a ↗ a ∗ a \nearrow a^* a ↗ a ∗ . The key tools are the fractional Gagliardo–Nirenberg inequality and compactness results in weighted Sobolev spaces. For fractional NLS with combined nonlinearities, Luo and Zhang [19 ] studied
( − Δ ) s u = λ u + μ ∣ u ∣ q − 2 u + ∣ u ∣ p − 2 u in R N , (5) (-\Delta)^s u = \lambda u + \mu|u|^{q-2}u + |u|^{p-2}u \quad\text{in } \mathbb{R}^N, \tag*{(5)} ( − Δ ) s u = λ u + μ ∣ u ∣ q − 2 u + ∣ u ∣ p − 2 u in R N , (5)
with prescribed mass ∫ ∣ u ∣ 2 = a 2 \int|u|^2 = a^2 ∫ ∣ u ∣ 2 = a 2 , and obtained existence and non-existence results by means of fiber maps, Pohozaev identities, and minimax arguments. For the mass-supercritical case, Peng and Xia [20 ] studied
( − Δ ) s u + V ( x ) u + ω u = ∣ u ∣ p − 2 u in R N , (6) (-\Delta)^s u + V(x)u + \omega u = |u|^{p-2}u \quad\text{in } \mathbb{R}^N, \tag*{(6)} ( − Δ ) s u + V ( x ) u + ω u = ∣ u ∣ p − 2 u in R N , (6)
with p ∈ ( 2 + 4 s / N , 2 s ∗ ) p \in(2 + 4s/N, 2_s^*) p ∈ ( 2 + 4 s / N , 2 s ∗ ) , and proved the existence of normalized solutions under suitable assumptions on the potential, using a new minimax argument and a splitting lemma for the nonlocal operator.
For systems, the situation is considerably richer. Two natural mass constraints appear in the literature. The first is the separate constraint
∫ R N ∣ u 1 ∣ 2 d x = ∫ R N ∣ u 2 ∣ 2 d x = 1 , (7) \int_{\mathbb{R}^N} |u_1|^2\,dx = \int_{\mathbb{R}^N} |u_2|^2\,dx = 1, \tag*{(7)} ∫ R N ∣ u 1 ∣ 2 d x = ∫ R N ∣ u 2 ∣ 2 d x = 1 , (7)
under which each component has a prescribed mass. The second is the total-mass constraint
∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 , (8) \int_{\mathbb{R}^N} (|u_1|^2 + |u_2|^2)\,dx = 1, \tag*{(8)} ∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 , (8)
under which only the sum of the masses is fixed, and mass may be transferred between the two components. These two constraints lead to substantially different variational structures, critical thresholds, and blow-up phenomena.
For two-component systems with separate mass constraints, Bartsch, Jeanjean and Soave [1 ] studied the coupled cubic Schrödinger system in R 3 \mathbb{R}^3 R 3 ,
{ − Δ u − λ 1 u = μ 1 u 3 + β u v 2 , − Δ v − λ 2 v = μ 2 v 3 + β u 2 v , (9) \begin{cases}
-\Delta u - \lambda_1 u = \mu_1 u^3 + \beta u v^2,\\
-\Delta v - \lambda_2 v = \mu_2 v^3 + \beta u^2 v, \tag*{(9)}
\end{cases} { − Δ u − λ 1 u = μ 1 u 3 + β u v 2 , − Δ v − λ 2 v = μ 2 v 3 + β u 2 v , (9)
with ∫ R 3 u 2 = a 1 2 \int_{\mathbb{R}^3} u^2 = a_1^2 ∫ R 3 u 2 = a 1 2 , ∫ R 3 v 2 = a 2 2 \int_{\mathbb{R}^3} v^2 = a_2^2 ∫ R 3 v 2 = a 2 2 . They proved the existence of positive normalized solutions for different ranges of the coupling parameter β > 0 \beta> 0 β > 0 , using a two-dimensional linking argument, a minimax principle, and a Pohozaev-type constraint. In R 2 \mathbb{R}^2 R 2 , Guo, Zeng and Zhou [11 ] studied the system
{ − Δ u 1 + V 1 ( x ) u 1 = μ 1 u 1 + b 1 u 1 3 + β u 2 2 u 1 , − Δ u 2 + V 2 ( x ) u 2 = μ 2 u 2 + b 2 u 2 3 + β u 1 2 u 2 , (10) \begin{cases}
-\Delta u_1 + V_1(x)u_1 = \mu_1 u_1 + b_1 u_1^3 + \beta u_2^2u_1,\\
-\Delta u_2 + V_2(x)u_2 = \mu_2 u_2 + b_2 u_2^3 + \beta u_1^2u_2, \tag*{(10)}
\end{cases} { − Δ u 1 + V 1 ( x ) u 1 = μ 1 u 1 + b 1 u 1 3 + β u 2 2 u 1 , − Δ u 2 + V 2 ( x ) u 2 = μ 2 u 2 + b 2 u 2 3 + β u 1 2 u 2 , (10)
with ∫ R 2 u 1 2 = ∫ R 2 u 2 2 = 1 \int_{\mathbb{R}^2} u_1^2 = \int_{\mathbb{R}^2} u_2^2 = 1 ∫ R 2 u 1 2 = ∫ R 2 u 2 2 = 1 . They obtained a partial classification of the existence and non-existence of minimizers: if 0 < b 1 , b 2 < a ∗ 0 < b_1,b_2 < a^* 0 < b 1 , b 2 < a ∗ and β < ( a ∗ − b 1 ) ( a ∗ − b 2 ) \beta< \sqrt{(a^* - b_1)(a^* - b_2)} β < ( a ∗ − b 1 ) ( a ∗ − b 2 ) , minimizers exist; if b 1 > a ∗ b_1>a^* b 1 > a ∗ or b 2 > a ∗ b_2>a^* b 2 > a ∗ or β > ( a ∗ − b 1 ) / 2 + ( a ∗ − b 2 ) / 2 \beta>(a^*-b_1)/2+(a^*-b_2)/2 β > ( a ∗ − b 1 ) /2 + ( a ∗ − b 2 ) /2 , no minimizer exists; the complete classification in the intermediate region ( a ∗ − b 1 ) ( a ∗ − b 2 ) ≤ β ≤ ( a ∗ − b 1 ) / 2 + ( a ∗ − b 2 ) / 2 \sqrt{(a^*-b_1)(a^*-b_2)}\le\beta\le(a^*-b_1)/2+(a^*-b_2)/2 ( a ∗ − b 1 ) ( a ∗ − b 2 ) ≤ β ≤ ( a ∗ − b 1 ) /2 + ( a ∗ − b 2 ) /2 remains open. They also proved that, as ( b 1 , b 2 ) ↗ ( a ∗ − β , a ∗ − β ) (b_1,b_2)\nearrow(a^*-\beta,a^*-\beta) ( b 1 , b 2 ) ↗ ( a ∗ − β , a ∗ − β ) , each component blows up at the same point and converges to the same profile after scaling, and they obtained the optimal blow-up rate. The proof uses an auxiliary minimization problem, refined energy estimates, and a blow-up analysis. In a subsequent paper [12 ] , they considered the case of attractive intraspecies and repulsive interspecies interactions (β < 0 \beta<0 β < 0 ) and obtained the blow-up behavior of ground states, showing that each component concentrates at a global minimum of the associated trapping potential.
When the constraint is the total mass, the situation changes significantly because mass can be transferred between the two components. Guo, Li, Wei and Zeng [7 ] studied the two-component attractive BEC system
{ − Δ u 1 + V 1 ( x ) u 1 = μ u 1 + a 1 u 1 3 + β u 2 2 u 1 , − Δ u 2 + V 2 ( x ) u 2 = μ u 2 + a 2 u 2 3 + β u 1 2 u 2 , (11) \begin{cases}
-\Delta u_1+V_1(x)u_1=\mu u_1+a_1u_1^3+\beta u_2^2u_1,\\
-\Delta u_2+V_2(x)u_2=\mu u_2+a_2u_2^3+\beta u_1^2u_2, \tag*{(11)}
\end{cases} { − Δ u 1 + V 1 ( x ) u 1 = μ u 1 + a 1 u 1 3 + β u 2 2 u 1 , − Δ u 2 + V 2 ( x ) u 2 = μ u 2 + a 2 u 2 3 + β u 1 2 u 2 , (11)
in R 2 \mathbb{R}^2 R 2 with ∫ R 2 ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 \int_{\mathbb{R}^2}(|u_1|^2+|u_2|^2)\,dx=1 ∫ R 2 ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 . They gave a complete classification of the existence and non-existence of ground states, and analyzed uniqueness and symmetry breaking as β ↗ β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) \beta\nearrow\beta^*=a^*+\sqrt{(a^*-a_1)(a^*-a_2)} β ↗ β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) . Their methods include a refined Gagliardo–Nirenberg inequality, energy estimates, and a local uniqueness argument based on the implicit function theorem.
For fractional systems with separate constraints, Liu, Zhang and Zuo [17 ] studied the one-dimensional half-Laplacian system
\begin{cases}
\sqrt{-\Delta}u_1+V_1(x)u_1=\mu_1u_1+a_1u_1^3+\beta u_2^2u_1,\\
\sqrt{-\Delta}u_2+V_2(x)u_2=\mu_2u_2+a_2u_2^3+\beta u_1^2u_2,
\end{cases}
\qquad x\in\mathbb{R}, \tag*{(12)}\end{cases}
with ∫ R u 1 2 = ∫ R u 2 2 = 1 \int_{\mathbb{R}}u_1^2=\int_{\mathbb{R}}u_2^2=1 ∫ R u 1 2 = ∫ R u 2 2 = 1 . They established a criterion for the existence and non-existence of minimizers by solving an auxiliary minimization problem, and they derived delicate estimates for the energy functional. Their results show that if 0 < a 1 , a 2 < a ∗ 0<a_1,a_2<a^* 0 < a 1 , a 2 < a ∗ and β < β − : = ( a ∗ − a 1 ) ( a ∗ − a 2 ) \beta<\beta_-:=\sqrt{(a^*-a_1)(a^*-a_2)} β < β − := ( a ∗ − a 1 ) ( a ∗ − a 2 ) , a minimizer exists, whereas if a 1 > a ∗ a_1>a^* a 1 > a ∗ or a 2 > a ∗ a_2>a^* a 2 > a ∗ or β > β + : = ( a ∗ − a 1 ) / 2 + ( a ∗ − a 2 ) / 2 \beta>\beta_+:=(a^*-a_1)/2+(a^*-a_2)/2 β > β + := ( a ∗ − a 1 ) /2 + ( a ∗ − a 2 ) /2 , no minimizer exists. The proof uses a refined fractional Gagliardo–Nirenberg inequality and a concentration-compactness argument. In a subsequent work [18 ] , they studied the uniqueness and concentration behavior of solutions for the same system under the separate constraint. They proved the uniqueness of nonnegative minimizers for small ∣ ( a 1 , a 2 , β ) ∣ |(a_1,a_2,\beta)| ∣ ( a 1 , a 2 , β ) ∣ via the implicit function theorem, and analyzed the blow-up behavior as ( a 1 , a 2 ) ↗ ( a ∗ − β , a ∗ − β ) (a_1,a_2)\nearrow(a^*-\beta,a^*-\beta) ( a 1 , a 2 ) ↗ ( a ∗ − β , a ∗ − β ) . They showed that each component blows up and concentrates at a flattest common minimum point of V 1 V_1 V 1 and V 2 V_2 V 2 , and obtained the optimal blow-up rate
ε k = ( a ∗ − a 1 k + a 2 k + 2 β 2 ) 1 p 0 + 1 , (13) \varepsilon_k=\left(a^*-\frac{a_{1k}+a_{2k}+2\beta}{2}\right)^{\frac{1}{p_0+1}}, \tag*{(13)} ε k = ( a ∗ − 2 a 1 k + a 2 k + 2 β ) p 0 + 1 1 , (13)
where p 0 p_0 p 0 is determined by the order of the flattest common minimum. The proof combines energy estimates, Pohozaev identities, and nonlocal regularity theory.
It is important to observe that the classification obtained in [17 ] for the separate constraint is not complete. Indeed, while the existence and non-existence of minimizers are settled when β < β − \beta<\beta_- β < β − and when β > β + \beta>\beta_+ β > β + , the intermediate region β − ≤ β ≤ β + \beta_-\le\beta\le\beta_+ β − ≤ β ≤ β + is only partially understood: for s = 1 2 s=\frac{1}{2} s = 2 1 , Theorems 1.5 and 1.6 of [17 ] provide some partial results under additional assumptions, whereas for s ≠ 1 2 s \ne\frac{1}{2} s = 2 1 the complete classification remains an open problem (see Remark 1.3 therein). Moreover, the blow-up analysis in [18 ] is restricted to the separate constraint, and the optimal rate is obtained under specific assumptions on the trapping potentials.
To the best of our knowledge, however, neither a complete classification nor a description of the concentration behavior of ground states of the coupled fractional system (1.1) under the total-mass constraint (1.8) is available in the literature. Inspired by the above works, this paper is devoted to two issues: the complete classification of ground states of (1.1) for arbitrary s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and N ≥ 1 N \ge1 N ≥ 1 , and the concentration behavior of the minimizers in the one-dimensional half-Laplacian case N = 1 N=1 N = 1 , s = 1 2 s=\frac{1}{2} s = 2 1 , for trapping potentials V i ( x ) V_i(x) V i ( x ) (i = 1 , 2 i=1,2 i = 1 , 2 ) satisfying
( Ω ) 0 ≤ V i ( x ) ∈ L l o c ∞ ( R N ) , lim ∣ x ∣ → ∞ V i ( x ) = ∞ , and inf x ∈ R N ( V 1 ( x ) + V 2 ( x ) ) = 0 is attained . (\Omega)\quad0 \le V_i(x) \in L^\infty_{\mathrm{loc}}(\mathbb{R}^N),\quad\lim_{|x|\to\infty}V_i(x)=\infty,\quad\text{and}\quad\inf_{x\in\mathbb{R}^N}\left(V_1(x)+V_2(x)\right)=0\text{ is attained}. ( Ω ) 0 ≤ V i ( x ) ∈ L loc ∞ ( R N ) , ∣ x ∣ → ∞ lim V i ( x ) = ∞ , and x ∈ R N inf ( V 1 ( x ) + V 2 ( x ) ) = 0 is attained .
We begin with some notation. Let the Sobolev space
H s ( R N ) : = { u ∈ L 2 ( R N ) : ∥ u ∥ H s ( R N ) 2 : = ∫ R N ( 1 + ∣ ξ ∣ 2 ) s ∣ F u ( ξ ) ∣ 2 d ξ < ∞ } . (14) H^s(\mathbb{R}^N):=\left\{u\in L^2(\mathbb{R}^N):\|u\|_{H^s(\mathbb{R}^N)}^2:=\int_{\mathbb{R}^N}(1+|\xi|^2)^s|\mathcal{F}u(\xi)|^2\,d\xi<\infty\right\}. \tag*{(14)} H s ( R N ) := { u ∈ L 2 ( R N ) : ∥ u ∥ H s ( R N ) 2 := ∫ R N ( 1 + ∣ ξ ∣ 2 ) s ∣ F u ( ξ ) ∣ 2 d ξ < ∞ } . (14)
Here F u \mathcal{F}u F u denotes the Fourier transform of u u u . Equipped with the inner product
( u , v ) H s ( R N ) : = ∫ R N ( 1 + ∣ ξ ∣ 2 ) s F u ( ξ ) F v ( ξ ) d ξ , (15) (u,v)_{H^s(\mathbb{R}^N)}:=\int_{\mathbb{R}^N}(1+|\xi|^2)^s\mathcal{F}u(\xi)\mathcal{F}v(\xi)\,d\xi, \tag*{(15)} ( u , v ) H s ( R N ) := ∫ R N ( 1 + ∣ ξ ∣ 2 ) s F u ( ξ ) F v ( ξ ) d ξ , (15)
the space H s ( R N ) H^s(\mathbb{R}^N) H s ( R N ) forms a Hilbert space. Moreover, the norm ∥ ⋅ ∥ H s ( R N ) \|\cdot\|_{H^s(\mathbb{R}^N)} ∥ ⋅ ∥ H s ( R N ) is equivalent to the one given by
∥ u ∥ H s ( R N ) 2 = ∫ R N [ u ( − Δ ) s u + u 2 ] d x = ∫ R N [ ∣ ( − Δ ) s 2 u ∣ 2 + u 2 ] d x . (16) \|u\|_{H^s(\mathbb{R}^N)}^2=\int_{\mathbb{R}^N}\left[u(-\Delta)^s u+u^2\right]\,dx=\int_{\mathbb{R}^N}\left[|(-\Delta)^{\frac{s}{2}}u|^2+u^2\right]\,dx. \tag*{(16)} ∥ u ∥ H s ( R N ) 2 = ∫ R N [ u ( − Δ ) s u + u 2 ] d x = ∫ R N [ ∣ ( − Δ ) 2 s u ∣ 2 + u 2 ] d x . (16)
Ground states of system (1.1) can be studied equivalently through the minimizers of the following constrained variational problem (see Proposition A.1 ):
e ( a 1 , a 2 , β ) : = inf ( u 1 , u 2 ) ∈ M E a 1 , a 2 , β ( u 1 , u 2 ) , (17) e(a_1,a_2,\beta):=\inf_{(u_1,u_2)\in\mathcal{M}}E_{a_1,a_2,\beta}(u_1,u_2), \tag*{(17)} e ( a 1 , a 2 , β ) := ( u 1 , u 2 ) ∈ M inf E a 1 , a 2 , β ( u 1 , u 2 ) , (17)
where the associated Gross-Pitaevskii (GP) energy functional is given by
E a 1 , a 2 , β ( u 1 , u 2 ) = ∑ i = 1 2 ∫ R N [ ∣ ( − Δ ) s 2 u i ∣ 2 + V i ( x ) u i 2 ] d x − 2 p ∑ i = 1 2 a i ∫ R N ∣ u i ∣ p d x − 4 p β ∫ R N ∣ u 1 ∣ q ∣ u 2 ∣ q d x , ( u 1 , u 2 ) ∈ H 1 × H 2 . (18) \begin{aligned}
E_{a_1,a_2,\beta}(u_1,u_2)
={}&\sum_{i=1}^{2}\int_{\mathbb{R}^N}\left[|(-\Delta)^{\frac{s}{2}}u_i|^2+V_i(x)u_i^2\right]\,dx-\frac{2}{p}\sum_{i=1}^{2}a_i\int_{\mathbb{R}^N}|u_i|^p\,dx\\
&-\frac{4}{p}\beta\int_{\mathbb{R}^N}|u_1|^q|u_2|^q\,dx,\qquad(u_1,u_2)\in\mathcal{H}_1\times\mathcal{H}_2. \tag*{(18)}
\end{aligned} E a 1 , a 2 , β ( u 1 , u 2 ) = i = 1 ∑ 2 ∫ R N [ ∣ ( − Δ ) 2 s u i ∣ 2 + V i ( x ) u i 2 ] d x − p 2 i = 1 ∑ 2 a i ∫ R N ∣ u i ∣ p d x − p 4 β ∫ R N ∣ u 1 ∣ q ∣ u 2 ∣ q d x , ( u 1 , u 2 ) ∈ H 1 × H 2 . (18)
Here, the mass constraint M \mathcal{M} M is defined by
M : = { ( u 1 , u 2 ) ∈ H 1 × H 2 : ∫ R N ( u 1 2 + u 2 2 ) d x = 1 } (19) \mathcal{M}:=\left\{(u_1,u_2)\in\mathcal{H}_1\times\mathcal{H}_2:\int_{\mathbb{R}^N}(u_1^2+u_2^2)\,dx=1\right\} \tag*{(19)} M := { ( u 1 , u 2 ) ∈ H 1 × H 2 : ∫ R N ( u 1 2 + u 2 2 ) d x = 1 } (19)
with
H i = { u ∈ H s ( R N ) : ∫ R N V i ( x ) u 2 d x < ∞ } , i = 1 , 2. (20) \mathcal{H}_i=\left\{u\in H^s(\mathbb{R}^N):\int_{\mathbb{R}^N}V_i(x)u^2\,dx<\infty\right\},\qquad i=1,2. \tag*{(20)} H i = { u ∈ H s ( R N ) : ∫ R N V i ( x ) u 2 d x < ∞ } , i = 1 , 2. (20)
and
∥ u ∥ H i 2 = ∫ R N [ ∣ ( − Δ ) s 2 u ∣ 2 + V i ( x ) u 2 ] d x . (21) \|u\|_{\mathcal{H}_i}^{2}=\int_{\mathbb{R}^{N}}\left[|(-\Delta)^{\frac{s}{2}}u|^{2}+V_i(x)u^{2}\right]\,dx. \tag*{(21)} ∥ u ∥ H i 2 = ∫ R N [ ∣ ( − Δ ) 2 s u ∣ 2 + V i ( x ) u 2 ] d x . (21)
In order to investigate the coupled minimization problem (17) , it is essential to recall several fundamental results concerning the single-component fractional equation. The classical scalar field equation associated with system (1) in the mass-critical case is known as
( − Δ ) s u + 2 s N u − ∣ u ∣ 4 s N u = 0 in R N , u ∈ H s ( R N ) . (22) (-\Delta)^s u+\frac{2s}{N}u-|u|^{\frac{4s}{N}}u=0\quad\text{in }\mathbb{R}^{N},\quad u\in H^s(\mathbb{R}^{N}). \tag*{(22)} ( − Δ ) s u + N 2 s u − ∣ u ∣ N 4 s u = 0 in R N , u ∈ H s ( R N ) . (22)
As is well known (see [5 , 6 ] ), there exists a unique (up to translations) positive radial ground state Q ∈ H s ( R N ) Q\in H^s(\mathbb{R}^{N}) Q ∈ H s ( R N ) of equation (22) . Moreover, Q Q Q has the decay estimate
Q ( x ) = O ( ∣ x ∣ − N − 2 s ) as ∣ x ∣ → ∞ . (23) Q(x)=O\left(|x|^{-N-2s}\right)\quad\text{as }|x|\to\infty. \tag*{(23)} Q ( x ) = O ( ∣ x ∣ − N − 2 s ) as ∣ x ∣ → ∞. (23)
The following Gagliardo-Nirenberg inequality holds with optimal constant:
∫ R N ∣ u ∣ p d x ≤ N + 2 s N ∥ Q ∥ 2 4 s N ∥ ( − Δ ) s 2 u ∥ 2 2 ∥ u ∥ 2 4 s N ∀ u ∈ H s ( R N ) . (24) \int_{\mathbb{R}^{N}}|u|^{p}\,dx\leq\frac{N+2s}{N\|Q\|_{2}^{\frac{4s}{N}}}\|(-\Delta)^{\frac{s}{2}}u\|_{2}^{2}\|u\|_{2}^{\frac{4s}{N}}\quad\forall u\in H^s(\mathbb{R}^{N}). \tag*{(24)} ∫ R N ∣ u ∣ p d x ≤ N ∥ Q ∥ 2 N 4 s N + 2 s ∥ ( − Δ ) 2 s u ∥ 2 2 ∥ u ∥ 2 N 4 s ∀ u ∈ H s ( R N ) . (24)
Define a ∗ : = ∥ Q ∥ 2 4 s N a^*:=\|Q\|_{2}^{\frac{4s}{N}} a ∗ := ∥ Q ∥ 2 N 4 s . This constant plays the role of a critical mass for the single equation. Thus, we have
∫ R N ∣ u ∣ p d x ≤ p 2 a ∗ ∥ ( − Δ ) s 2 u ∥ 2 2 ∥ u ∥ 2 4 s N ∀ u ∈ H s ( R N ) . (25) \int_{\mathbb{R}^{N}}|u|^{p}\,dx\leq\frac{p}{2a^*}\|(-\Delta)^{\frac{s}{2}}u\|_{2}^{2}\|u\|_{2}^{\frac{4s}{N}}\quad\forall u\in H^s(\mathbb{R}^{N}). \tag*{(25)} ∫ R N ∣ u ∣ p d x ≤ 2 a ∗ p ∥ ( − Δ ) 2 s u ∥ 2 2 ∥ u ∥ 2 N 4 s ∀ u ∈ H s ( R N ) . (25)
We now state our main results. The first result concerns the complete classification of existence and non-existence of the minimizers for (17) . To deal with the total-mass constraint, the main mathematical challenge lies in providing a lower bound for the energy functional. Instead of artificially constructing a new coupled Gagliardo-Nirenberg inequality, we introduce an auxiliary intensity function Λ ( a 1 , a 2 , β ) \Lambda(a_1,a_2,\beta) Λ ( a 1 , a 2 , β ) and characterize the exact critical coupling threshold β ∗ \beta^* β ∗ as follows:
Λ ( a 1 , a 2 , β ) : = max θ ∈ [ 0 , 1 ] ( a 1 θ q + a 2 ( 1 − θ ) q + 2 β θ q / 2 ( 1 − θ ) q / 2 ) . (26) \Lambda(a_1,a_2,\beta):=\max_{\theta\in[0,1]}\left(a_1\theta^q+a_2(1-\theta)^q+2\beta\theta^{q/2}(1-\theta)^{q/2}\right). \tag*{(26)} Λ ( a 1 , a 2 , β ) := θ ∈ [ 0 , 1 ] max ( a 1 θ q + a 2 ( 1 − θ ) q + 2 β θ q /2 ( 1 − θ ) q /2 ) . (26)
β ∗ : = min θ ∈ ( 0 , 1 ) a ∗ − a 1 θ q − a 2 ( 1 − θ ) q 2 θ q / 2 ( 1 − θ ) q / 2 . (27) \beta^*:=\min_{\theta\in(0,1)}\frac{a^*-a_1\theta^q-a_2(1-\theta)^q}{2\theta^{q/2}(1-\theta)^{q/2}}. \tag*{(27)} β ∗ := θ ∈ ( 0 , 1 ) min 2 θ q /2 ( 1 − θ ) q /2 a ∗ − a 1 θ q − a 2 ( 1 − θ ) q . (27)
Our first main result establishes the existence and non-existence of the minimizers of the problem (17) for arbitrary 0 < s < 1 0<s<1 0 < s < 1 and N ≥ 1 N\geq1 N ≥ 1 .
Theorem 1.2. Suppose that ( Ω ) (\Omega) ( Ω ) holds. Then the following statements hold.
(i) If 0 ≤ a 1 , a 2 < a ∗ 0\leq a_1,a_2<a^* 0 ≤ a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^* 0 < β < β ∗ , then problem (17) has at least one minimizer. Moreover, it holds
lim β → β ∗ e ( a 1 , a 2 , β ) = e ( a 1 , a 2 , β ∗ ) = 0 , 0 ≤ a 1 , a 2 < a ∗ . (28) \lim_{\beta\to\beta^*}e(a_1,a_2,\beta)=e(a_1,a_2,\beta^*)=0,\quad0\leq a_1,a_2<a^*. \tag*{(28)} β → β ∗ lim e ( a 1 , a 2 , β ) = e ( a 1 , a 2 , β ∗ ) = 0 , 0 ≤ a 1 , a 2 < a ∗ . (28) (ii) If a 1 ≥ a ∗ a_1\geq a^* a 1 ≥ a ∗ or a 2 ≥ a ∗ a_2\geq a^* a 2 ≥ a ∗ or β ≥ β ∗ \beta\geq\beta^* β ≥ β ∗ , then problem (17) has no minimizer.
Remark 1.3. We remark that the existence result also holds true for the system with repulsive interspecies interaction, i.e., β < 0 \beta< 0 β < 0 , and for the system with β = 0 \beta= 0 β = 0 ; this can be seen from the proof of Theorem 1.1 (i). This complete classification identifies the exact parameter regimes in which the total-mass-constrained fractional BEC system remains stable. In particular, it covers the degenerate cases a 1 = 0 a_1 = 0 a 1 = 0 or a 2 = 0 a_2 = 0 a 2 = 0 and yields a complete classification for all β ∈ R \beta\in\mathbb{R} β ∈ R , in contrast with the partial classification available in the separate-constraint setting of [17 ] .
Our second goal is to study the blow-up behavior and the optimal blow-up rate for the one-dimensional half-Laplacian system under the total-mass constraint, obtaining results analogous to the last two results of [7 ] . More precisely, we consider
{ − Δ u 1 + V 1 ( x ) u 1 = μ u 1 + a 1 u 1 3 + β u 2 2 u 1 , − Δ u 2 + V 2 ( x ) u 2 = μ u 2 + a 2 u 2 3 + β u 1 2 u 2 , x ∈ R , (29) \left\{\begin{aligned}
\sqrt{-\Delta}\,u_1 + V_1(x)u_1 &= \mu u_1 + a_1u_1^3 + \beta u_2^2u_1,\\
\sqrt{-\Delta}\,u_2 + V_2(x)u_2 &= \mu u_2 + a_2u_2^3 + \beta u_1^2u_2,
\end{aligned}\right.
\qquad x \in\mathbb{R}, \tag*{(29)} { − Δ u 1 + V 1 ( x ) u 1 − Δ u 2 + V 2 ( x ) u 2 = μ u 1 + a 1 u 1 3 + β u 2 2 u 1 , = μ u 2 + a 2 u 2 3 + β u 1 2 u 2 , x ∈ R , (29)
with ∫ R ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 \int_{\mathbb{R}} (|u_1|^2 + |u_2|^2)\,dx = 1 ∫ R ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) d x = 1 , which implies p = 4 p = 4 p = 4 and q = 2 q = 2 q = 2 . By (27) , one can easily verify that the minimum is attained at the optimal mass fraction
γ : = a ∗ − a 2 a ∗ − a 1 + a ∗ − a 2 , (30) \gamma:= \frac{\sqrt{a^* - a_2}}{\sqrt{a^* - a_1} + \sqrt{a^* - a_2}}, \tag*{(30)} γ := a ∗ − a 1 + a ∗ − a 2 a ∗ − a 2 , (30)
which directly yields β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) \beta^* = a^* + \sqrt{(a^* - a_1)(a^* - a_2)} β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) . The mass fraction γ \gamma γ and the threshold β ∗ \beta^* β ∗ play a crucial role in the proofs of the following theorems.
Theorem 1.4. Suppose ( Ω ) (\Omega) ( Ω ) holds and let ( u 1 β , u 2 β ) (u_{1\beta}, u_{2\beta}) ( u 1 β , u 2 β ) be a nonnegative minimizer of e ( a 1 , a 2 , β ) e(a_1,a_2,\beta) e ( a 1 , a 2 , β ) with N = 1 N = 1 N = 1 and s = 1 2 s = \frac{1}{2} s = 2 1 , where 0 < a 1 , a 2 < a ∗ 0 < a_1, a_2 < a^* 0 < a 1 , a 2 < a ∗ and 0 < β < β ∗ 0 < \beta< \beta^* 0 < β < β ∗ . Then for any sequence { β k } \{\beta_k\} { β k } with β k → β ∗ \beta_k \to\beta^* β k → β ∗ as k → ∞ k \to\infty k → ∞ , there exists a subsequence of { β k } \{\beta_k\} { β k } , still denoted by { β k } \{\beta_k\} { β k } , such that for i = 1 , 2 i = 1, 2 i = 1 , 2 and every k k k , u i β k u_{i\beta_k} u i β k has at least one global maximum point z i β k z_{i\beta_k} z i β k , and
lim k → ∞ z 1 β k = lim k → ∞ z 2 β k = x 0 for some x 0 ∈ V , (31) \lim_{k\to\infty} z_{1\beta_k} = \lim_{k\to\infty} z_{2\beta_k} = x_0 \quad\text{for some } x_0 \in\mathcal{V}, \tag*{(31)} k → ∞ lim z 1 β k = k → ∞ lim z 2 β k = x 0 for some x 0 ∈ V , (31) where
V : = { x ∈ R : V 1 ( x ) = V 2 ( x ) = 0 } . (32) \mathcal{V} := \{x \in\mathbb{R}: V_1(x) = V_2(x) = 0\}. \tag*{(32)} V := { x ∈ R : V 1 ( x ) = V 2 ( x ) = 0 } . (32) Moreover, we have
{ lim k → ∞ ε β k 1 2 u 1 β k ( ε β k x + z 1 β k ) = γ ∥ Q ∥ 2 Q ( x ) , lim k → ∞ ε β k 1 2 u 2 β k ( ε β k x + z 2 β k ) = 1 − γ ∥ Q ∥ 2 Q ( x ) , (33) \left\{\begin{aligned}
\lim_{k\to\infty} \varepsilon_{\beta_k}^{\frac{1}{2}}u_{1\beta_k}(\varepsilon_{\beta_k}x + z_{1\beta_k}) &= \frac{\sqrt{\gamma}}{\|Q\|_2}Q(x),\\
\lim_{k\to\infty} \varepsilon_{\beta_k}^{\frac{1}{2}}u_{2\beta_k}(\varepsilon_{\beta_k}x + z_{2\beta_k}) &= \frac{\sqrt{1-\gamma}}{\|Q\|_2}Q(x),
\end{aligned}\right. \tag*{(33)} ⎩ ⎨ ⎧ k → ∞ lim ε β k 2 1 u 1 β k ( ε β k x + z 1 β k ) k → ∞ lim ε β k 2 1 u 2 β k ( ε β k x + z 2 β k ) = ∥ Q ∥ 2 γ Q ( x ) , = ∥ Q ∥ 2 1 − γ Q ( x ) , (33) strongly in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) , where γ \gamma γ is given by (30) and Q Q Q is the positive radially symmetric ground state of (22) ; here ε β k > 0 \varepsilon_{\beta_k} > 0 ε β k > 0 satisfies
ε β k : = ( ∫ R ( ∣ ( − Δ ) 1 4 u 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β k ∣ 2 ) d x ) − 1 → 0 as k → ∞ , (34) \varepsilon_{\beta_k} := \left(\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}u_{1\beta_k}\right|^2 + \left|(-\Delta)^{\frac{1}{4}}u_{2\beta_k}\right|^2\right)\,dx\right)^{-1} \to0 \quad\text{as } k \to\infty, \tag*{(34)} ε β k := ( ∫ R ( ( − Δ ) 4 1 u 1 β k 2 + ( − Δ ) 4 1 u 2 β k 2 ) d x ) − 1 → 0 as k → ∞ , (34) lim k → ∞ ∣ z 1 β k − z 2 β k ∣ ε β k = 0 , (35) \lim_{k\to\infty}\frac{|z_{1\beta_k} - z_{2\beta_k}|}{\varepsilon_{\beta_k}} = 0, \tag*{(35)} k → ∞ lim ε β k ∣ z 1 β k − z 2 β k ∣ = 0 , (35) and
lim k → ∞ β ∗ − β k ε β k = 0. (36) \lim_{k\to\infty}\frac{\beta^* - \beta_k}{\varepsilon_{\beta_k}} = 0. \tag*{(36)} k → ∞ lim ε β k β ∗ − β k = 0. (36)
Remark 1.5. Unlike [7 ] , we do not establish the uniqueness of the global maximum point of each component; we only prove that at least one global maximum point exists. The uniqueness remains an open problem, mainly due to the lack of sufficient elliptic regularity for the fractional operator, which prevents us from proving the smoothness of the minimizer sequence (see also Remark 1.4 in [18 ] ).
To carry out a finer analysis of the concentration behavior of nonnegative minimizers, we henceforth assume that the trapping potentials V i ( x ) V_i(x) V i ( x ) have the form
V i ( x ) = h i ( x ) ∏ j = 1 m i ∣ x − x i j ∣ q i j , q i j > 0 and m i ∈ N + , i = 1 , 2 , (37) V_i(x) = h_i(x) \prod_{j=1}^{m_i} \lvert x-x_{ij}\rvert^{q_{ij}}, \qquad q_{ij}>0 \text{ and } m_i\in\mathbb{N}^{+},\ i=1,2, \tag*{(37)} V i ( x ) = h i ( x ) j = 1 ∏ m i ∣ x − x ij ∣ q ij , q ij > 0 and m i ∈ N + , i = 1 , 2 , (37)
where h i ( x ) ∈ C l o c α ( R ) h_i(x)\in C_{\mathrm{loc}}^{\alpha}(\mathbb{R}) h i ( x ) ∈ C loc α ( R ) satisfies 1 C < h i ( x ) < C \frac{1}{C}<h_i(x)<C C 1 < h i ( x ) < C for some positive constant C C C and x i j ≠ x i k x_{ij}\ne x_{ik} x ij = x ik for j ≠ k j\ne k j = k . We also assume, without loss of generality, that there is an integer b b b with 1 ≤ b ≤ min { m 1 , m 2 } 1\le b\le\min\{m_1,m_2\} 1 ≤ b ≤ min { m 1 , m 2 } such that
{ x 1 j = x 2 j = : x j , where j = 1 , … , b ; x 1 j ≠ x 2 k , where j ∈ { b + 1 , … , m 1 } and k ∈ { b + 1 , … , m 2 } . (38) \left\{
\begin{aligned}
x_{1j}=x_{2j}=:x_j, \qquad&\text{where } j=1,\ldots,b;\\
x_{1j}\ne x_{2k}, \qquad&\text{where } j\in\{b+1,\ldots,m_1\}\text{ and }k\in\{b+1,\ldots,m_2\}.
\end{aligned}
\right. \tag*{(38)} { x 1 j = x 2 j =: x j , x 1 j = x 2 k , where j = 1 , … , b ; where j ∈ { b + 1 , … , m 1 } and k ∈ { b + 1 , … , m 2 } . (38)
Notice that (38) implies
V = { x ∈ R : V 1 ( x ) = V 2 ( x ) = 0 } = { x 1 , … , x b } . (39) \mathcal{V}=\{x\in\mathbb{R}:V_1(x)=V_2(x)=0\}=\{x_1,\ldots,x_b\}. \tag*{(39)} V = { x ∈ R : V 1 ( x ) = V 2 ( x ) = 0 } = { x 1 , … , x b } . (39)
Additionally, we define λ j \lambda_j λ j for j = 1 , … , b j=1,\ldots,b j = 1 , … , b by
λ j : = { lim x → x j γ V 1 ( x ) ∣ x − x j ∣ q 1 j ∫ R ∣ y ∣ q 1 j Q 2 ( y ) d y if q 1 j < q 2 j , lim x → x j γ V 1 ( x ) + ( 1 − γ ) V 2 ( x ) ∣ x − x j ∣ q 1 j ∫ R ∣ y ∣ q 1 j Q 2 ( y ) d y if q 1 j = q 2 j , lim x → x j ( 1 − γ ) V 2 ( x ) ∣ x − x j ∣ q 2 j ∫ R ∣ y ∣ q 2 j Q 2 ( y ) d y if q 1 j > q 2 j , (40) \lambda_j:=\begin{cases}
\lim_{x\to x_j}\frac{\gamma V_1(x)}{\lvert x-x_j\rvert^{q_{1j}}}\int_{\mathbb{R}}\lvert y\rvert^{q_{1j}}Q^2(y)\,\mathrm{d}y & \text{if }q_{1j}<q_{2j},\\[1.2em]
\lim_{x\to x_j}\frac{\gamma V_1(x)+(1-\gamma)V_2(x)}{\lvert x-x_j\rvert^{q_{1j}}}\int_{\mathbb{R}}\lvert y\rvert^{q_{1j}}Q^2(y)\,\mathrm{d}y & \text{if }q_{1j}=q_{2j},\\[1.2em]
\lim_{x\to x_j}\frac{(1-\gamma)V_2(x)}{\lvert x-x_j\rvert^{q_{2j}}}\int_{\mathbb{R}}\lvert y\rvert^{q_{2j}}Q^2(y)\,\mathrm{d}y & \text{if }q_{1j}>q_{2j},
\end{cases} \tag*{(40)} λ j := ⎩ ⎨ ⎧ lim x → x j ∣ x − x j ∣ q 1 j γ V 1 ( x ) ∫ R ∣ y ∣ q 1 j Q 2 ( y ) d y lim x → x j ∣ x − x j ∣ q 1 j γ V 1 ( x ) + ( 1 − γ ) V 2 ( x ) ∫ R ∣ y ∣ q 1 j Q 2 ( y ) d y lim x → x j ∣ x − x j ∣ q 2 j ( 1 − γ ) V 2 ( x ) ∫ R ∣ y ∣ q 2 j Q 2 ( y ) d y if q 1 j < q 2 j , if q 1 j = q 2 j , if q 1 j > q 2 j , (40)
where γ \gamma γ is the optimal mass fraction given by (30) . Define
q j : = min { q 1 j , q 2 j } , j = 1 , … , b ; q 0 : = max 1 ≤ j ≤ b q j , (41) q_j:=\min\{q_{1j},q_{2j}\},\quad j=1,\ldots,b;\qquad q_0:=\max_{1\le j\le b}q_j, \tag*{(41)} q j := min { q 1 j , q 2 j } , j = 1 , … , b ; q 0 := 1 ≤ j ≤ b max q j , (41)
so that
Z : = { x j : q j = q 0 , j = 1 , … , b } ⊂ V . (42) Z:=\{x_j:q_j=q_0,\ j=1,\ldots,b\}\subset\mathcal{V}. \tag*{(42)} Z := { x j : q j = q 0 , j = 1 , … , b } ⊂ V . (42)
Set
λ 0 : = min j ∈ Γ λ j , where Γ : = { j : x j ∈ Z } , (43) \lambda_0:=\min_{j\in\Gamma}\lambda_j,\qquad\text{where }\Gamma:=\{j:x_j\in Z\}, \tag*{(43)} λ 0 := j ∈ Γ min λ j , where Γ := { j : x j ∈ Z } , (43)
and let
Z 0 : = { x j ∈ Z : λ j = λ 0 } (44) Z_0:=\{x_j\in Z:\lambda_j=\lambda_0\} \tag*{(44)} Z 0 := { x j ∈ Z : λ j = λ 0 } (44)
denote the set of the flattest common minimum points of V i ( x ) V_i(x) V i ( x ) for $i=1,2. Our third main result concerns the precise concentration behavior of nonnegative minimizers for (1.17) with a polynomial trapping potential defined above.
Theorem 1.6. Suppose that V i ( x ) V_i(x) V i ( x ) (i = 1 , 2 i = 1,2 i = 1 , 2 ) satisfy (37) and (38) , and let ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) be a nonnegative minimizer of e ( a 1 , a 2 , β ) e(a_1,a_2,\beta) e ( a 1 , a 2 , β ) with N = 1 N = 1 N = 1 and s = 1 2 s = \frac{1}{2} s = 2 1 , where 0 < a 1 , a 2 < a ∗ 0 < a_1,a_2 < a^* 0 < a 1 , a 2 < a ∗ and 0 < β < β ∗ 0 < \beta< \beta^* 0 < β < β ∗ . Moreover, if 0 < q 0 < 1 0 < q_0 < 1 0 < q 0 < 1 , then for any sequence { β k } \{\beta_k\} { β k } with β k → β ∗ \beta_k \to\beta^* β k → β ∗ as k → ∞ k \to\infty k → ∞ , there exists a subsequence of { β k } \{\beta_k\} { β k } , still denoted by { β k } \{\beta_k\} { β k } , such that for i = 1 , 2 i = 1,2 i = 1 , 2 and every k k k , u i β k u_{i\beta_k} u i β k has at least one global maximum point z i β k z_{i\beta_k} z i β k , and
lim k → ∞ z 1 β k − x j 0 ϵ β k = lim k → ∞ z 2 β k − x j 0 ϵ β k = 0 for some x j 0 ∈ Z 0 , (45) \lim_{k\to\infty}\frac{z_{1\beta_k}-x_{j_0}}{\epsilon_{\beta_k}}=\lim_{k\to\infty}\frac{z_{2\beta_k}-x_{j_0}}{\epsilon_{\beta_k}}=0 \quad\text{for some }x_{j_0}\in Z_0, \tag*{(45)} k → ∞ lim ϵ β k z 1 β k − x j 0 = k → ∞ lim ϵ β k z 2 β k − x j 0 = 0 for some x j 0 ∈ Z 0 , (45) where ϵ β k > 0 \epsilon_{\beta_k}>0 ϵ β k > 0 is given by
ϵ β k : = [ 2 γ ( 1 − γ ) q 0 λ 0 ( β ∗ − β k ) ] 1 q 0 + 1 . (46) \epsilon_{\beta_k}:=\left[\frac{2\gamma(1-\gamma)}{q_0\lambda_0}(\beta^*-\beta_k)\right]^{\frac{1}{q_0+1}}. \tag*{(46)} ϵ β k := [ q 0 λ 0 2 γ ( 1 − γ ) ( β ∗ − β k ) ] q 0 + 1 1 . (46) Moreover, we have
lim k → ∞ ϵ β k 1 2 u 1 β k ( ϵ β k x + z 1 β k ) = γ ∥ Q ∥ 2 Q ( x ) , lim k → ∞ ϵ β k 1 2 u 2 β k ( ϵ β k x + z 2 β k ) = 1 − γ ∥ Q ∥ 2 Q ( x ) , (47) \begin{aligned}
\lim_{k\to\infty}\epsilon_{\beta_k}^{\frac{1}{2}}u_{1\beta_k}(\epsilon_{\beta_k}x+z_{1\beta_k})&=\frac{\sqrt{\gamma}}{\lVert Q\rVert_2}Q(x),\\
\lim_{k\to\infty}\epsilon_{\beta_k}^{\frac{1}{2}}u_{2\beta_k}(\epsilon_{\beta_k}x+z_{2\beta_k})&=\frac{\sqrt{1-\gamma}}{\lVert Q\rVert_2}Q(x), \tag*{(47)}
\end{aligned} k → ∞ lim ϵ β k 2 1 u 1 β k ( ϵ β k x + z 1 β k ) k → ∞ lim ϵ β k 2 1 u 2 β k ( ϵ β k x + z 2 β k ) = ∥ Q ∥ 2 γ Q ( x ) , = ∥ Q ∥ 2 1 − γ Q ( x ) , (47) strongly in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) and
lim k → ∞ ϵ β k ∫ R ( ∣ ( − Δ ) 1 4 u 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β k ∣ 2 ) d x = 1. (48) \lim_{k\to\infty}\epsilon_{\beta_k}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}u_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}u_{2\beta_k}\right|^2\right)\,dx=1. \tag*{(48)} k → ∞ lim ϵ β k ∫ R ( ( − Δ ) 4 1 u 1 β k 2 + ( − Δ ) 4 1 u 2 β k 2 ) d x = 1. (48)
Theorem 1.6 shows that, despite the long-range algebraic interaction and the strong attractive coupling, both components retain the mass ratio γ : ( 1 − γ ) \gamma: (1-\gamma) γ : ( 1 − γ ) and concentrate simultaneously at the flattest common minimum.
The main difficulties of this work are three aspects. First, the passage from the classical Laplacian to the fractional operator ( − Δ ) s (-\Delta)^s ( − Δ ) s with arbitrary s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and dimension N ≥ 1 N \ge1 N ≥ 1 is not merely technical. The nonlocal nature of ( − Δ ) s (-\Delta)^s ( − Δ ) s destroys the usual maximum principle and ODE techniques; one has to rely on nonlocal elliptic regularity, De Giorgi–Nash–Moser theory, and the positivity of the resolvent kernel. Moreover, the critical exponents p = 2 + 4 s / N p = 2 + 4s/N p = 2 + 4 s / N and q = 1 + 2 s / N q = 1 + 2s/N q = 1 + 2 s / N depend on both s s s and N N N , so the structure of the nonlinearities changes with the dimension, and the ground state Q Q Q decays polynomially rather than exponentially, which makes the standard test-function estimates and the derivation of optimal energy bounds much more delicate. Second, the change from the separate constraint to the total-mass constraint is substantial: mass may be transferred between the two components, and the energy functional no longer admits an obvious lower bound; this difficulty is overcome by means of the auxiliary intensity function Λ ( a 1 , a 2 , β ) \Lambda(a_1,a_2,\beta) Λ ( a 1 , a 2 , β ) , which yields the exact critical coupling threshold β ∗ \beta^* β ∗ (defined in (27) ) for arbitrary s s s and N N N . It is only in the specific one-dimensional half-Laplacian case (N = 1 N = 1 N = 1 , s = 1 / 2 s = 1/2 s = 1/2 ) that this threshold simplifies to β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) \beta^* = a^*+\sqrt{(a^*-a_1)(a^*-a_2)} β ∗ = a ∗ + ( a ∗ − a 1 ) ( a ∗ − a 2 ) . This auxiliary method naturally covers the degenerate cases (a 1 = 0 a_1 = 0 a 1 = 0 or a 2 = 0 a_2 = 0 a 2 = 0 ) and the repulsive interspecies regime (β < 0 \beta< 0 β < 0 ), providing a complete classification for all β ∈ R \beta\in\mathbb{R} β ∈ R . Finally, the blow-up analysis requires careful treatment of the mutual interaction to show that both components concentrate simultaneously at the flattest common minimum.
Our analysis combines several ingredients: refined fractional Gagliardo–Nirenberg inequalities, constrained minimization, concentration-compactness arguments, Pohozaev identities, energy estimates, and nonlocal regularity theory. The paper is organized as follows. In Section 2, we establish the existence and non-existence criteria (Theorem 1.2 ). Section 3 is devoted to the proof of Theorem 1.4 on the general limit behavior, while Section 4 finishes the proof of Theorem 1.6 on the explicit blow-up profiles under polynomial trapping potentials.
Existence and non-existence results
This section is devoted to establishing the existence and non-existence of minimizers for problem (17) . We begin by recalling several preliminary results essential to our subsequent analysis.
Lemma 2.1. Under condition ( Ω ) (\Omega) ( Ω ) , the embedding H 1 × H 2 ↪ L r ( R N ) × L r ( R N ) \mathcal{H}_1 \times\mathcal{H}_2 \hookrightarrow L^r(\mathbb{R}^N) \times L^r(\mathbb{R}^N) H 1 × H 2 ↪ L r ( R N ) × L r ( R N ) is compact for any 2 ≤ r < 2 s ∗ 2 \le r < 2_s^* 2 ≤ r < 2 s ∗ , where 2 s ∗ = 2 N N − 2 s 2_s^* = \frac{2N}{N-2s} 2 s ∗ = N − 2 s 2 N for N > 2 s N > 2s N > 2 s and 2 s ∗ = ∞ 2_s^* = \infty 2 s ∗ = ∞ for N ≤ 2 s N \le2s N ≤ 2 s .
Proof. The proof is standard; see e.g. [4 Lemma 3.1] . The condition V i ( x ) → ∞ V_i(x) \to\infty V i ( x ) → ∞ ensures that bounded sequences in H i \mathcal{H}_i H i are tight, and local compactness of H s H^s H s gives strong convergence in L r L^r L r after extracting a subsequence. □ \square □
Lemma 2.3. Let u 1 , u 2 ∈ L p ( R N ) u_1, u_2 \in L^p(\mathbb{R}^N) u 1 , u 2 ∈ L p ( R N ) . Then,
∫ R N ∣ u 1 ∣ q ∣ u 2 ∣ q d x ≤ ( ∫ R N ∣ u 1 ∣ p d x ∫ R N ∣ u 2 ∣ p d x ) 1 / 2 , (49) \int_{\mathbb{R}^N} |u_1|^q |u_2|^q \, dx \le\left(\int_{\mathbb{R}^N} |u_1|^p \, dx \int_{\mathbb{R}^N} |u_2|^p \, dx\right)^{1/2}, \tag*{(49)} ∫ R N ∣ u 1 ∣ q ∣ u 2 ∣ q d x ≤ ( ∫ R N ∣ u 1 ∣ p d x ∫ R N ∣ u 2 ∣ p d x ) 1/2 , (49) and the equality holds if and only if ∣ u 1 ∣ p = κ ∣ u 2 ∣ p |u_1|^p = \kappa|u_2|^p ∣ u 1 ∣ p = κ ∣ u 2 ∣ p for some κ > 0 \kappa> 0 κ > 0 .
Lemma 2.4. If ( u 1 n , u 2 n ) ⇀ ( u 10 , u 20 ) (u_{1n}, u_{2n}) \rightharpoonup(u_{10}, u_{20}) ( u 1 n , u 2 n ) ⇀ ( u 10 , u 20 ) weakly in H 1 × H 2 \mathcal{H}_1 \times\mathcal{H}_2 H 1 × H 2 , then, by Lemma 2.1 , u i n → u i 0 u_{in} \to u_{i0} u in → u i 0 strongly in L r ( R N ) L^r(\mathbb{R}^N) L r ( R N ) for every 2 ≤ r < 2 s ∗ 2 \le r < 2_s^* 2 ≤ r < 2 s ∗ , and hence
lim n → ∞ ∫ R N ∣ u 1 n ∣ q ∣ u 2 n ∣ q d x = ∫ R N ∣ u 10 ∣ q ∣ u 20 ∣ q d x , (50) \lim_{n\to\infty} \int_{\mathbb{R}^N} |u_{1n}|^q |u_{2n}|^q \, dx = \int_{\mathbb{R}^N} |u_{10}|^q |u_{20}|^q \, dx, \tag*{(50)} n → ∞ lim ∫ R N ∣ u 1 n ∣ q ∣ u 2 n ∣ q d x = ∫ R N ∣ u 10 ∣ q ∣ u 20 ∣ q d x , (50) lim n → ∞ ∫ R N ∣ u i n ∣ p d x = ∫ R N ∣ u i 0 ∣ p d x , i = 1 , 2. (51) \lim_{n\to\infty} \int_{\mathbb{R}^N} |u_{in}|^p \, dx = \int_{\mathbb{R}^N} |u_{i0}|^p \, dx,\qquad i=1,2. \tag*{(51)} n → ∞ lim ∫ R N ∣ u in ∣ p d x = ∫ R N ∣ u i 0 ∣ p d x , i = 1 , 2. (51)
Proof. Since u i n → u i 0 u_{in} \to u_{i0} u in → u i 0 strongly in L p ( R N ) L^p(\mathbb{R}^N) L p ( R N ) , the sequences { u i n } \{u_{in}\} { u in } are uniformly bounded in L p ( R N ) L^p(\mathbb{R}^N) L p ( R N ) . Note that p / q = 2 p/q = 2 p / q = 2 , which allows us to work in L 2 L^2 L 2 . Using the elementary inequality
∣ ∣ a ∣ q − ∣ b ∣ q ∣ ≤ C q ( ∣ a ∣ q − 1 + ∣ b ∣ q − 1 ) ∣ a − b ∣ , ∀ a , b ∈ R , (52) \bigl||a|^q - |b|^q\bigr| \le C_q\bigl(|a|^{q-1} + |b|^{q-1}\bigr)|a-b|,\qquad\forall a,b \in\mathbb{R}, \tag*{(52)} ∣ a ∣ q − ∣ b ∣ q ≤ C q ( ∣ a ∣ q − 1 + ∣ b ∣ q − 1 ) ∣ a − b ∣ , ∀ a , b ∈ R , (52)
and Hölder’s inequality, we obtain
∥ ∣ u 1 n ∣ q − ∣ u 10 ∣ q ∥ 2 ≤ C q ( ∥ u 1 n ∥ p q − 1 + ∥ u 10 ∥ p q − 1 ) ∥ u 1 n − u 10 ∥ p → 0 as n → ∞ , (53) \bigl\||u_{1n}|^q - |u_{10}|^q\bigr\|_2 \le C_q\bigl(\|u_{1n}\|_p^{q-1} + \|u_{10}\|_p^{q-1}\bigr)\|u_{1n} - u_{10}\|_p \to0\quad\text{as }n\to\infty, \tag*{(53)} ∣ u 1 n ∣ q − ∣ u 10 ∣ q 2 ≤ C q ( ∥ u 1 n ∥ p q − 1 + ∥ u 10 ∥ p q − 1 ) ∥ u 1 n − u 10 ∥ p → 0 as n → ∞ , (53)
and similarly ∥ ∣ u 2 n ∣ q − ∣ u 20 ∣ q ∥ 2 → 0 \bigl\||u_{2n}|^q - |u_{20}|^q\bigr\|_2 \to0 ∣ u 2 n ∣ q − ∣ u 20 ∣ q 2 → 0 as n → ∞ n \to\infty n → ∞ .
Now estimate the difference of the integrals, by Hölder’s inequality, we have
∣ ∫ R N ∣ u 1 n ∣ q ∣ u 2 n ∣ q d x − ∫ R N ∣ u 10 ∣ q ∣ u 20 ∣ q d x ∣ ≤ ∫ R N ∣ ∣ u 1 n ∣ q − ∣ u 10 ∣ q ∣ ∣ u 2 n ∣ q d x + ∫ R N ∣ u 10 ∣ q ∣ ∣ u 2 n ∣ q − ∣ u 20 ∣ q ∣ d x ≤ ∥ ∣ u 1 n ∣ q − ∣ u 10 ∣ q ∥ 2 ∥ u 2 n ∥ p q + ∥ ∣ u 2 n ∣ q − ∣ u 20 ∣ q ∥ 2 ∥ u 10 ∥ p q . (54) \begin{aligned}
\left|\int_{\mathbb{R}^N}|u_{1n}|^q|u_{2n}|^q\,dx-\int_{\mathbb{R}^N}|u_{10}|^q|u_{20}|^q\,dx\right|
&\le\int_{\mathbb{R}^N}\bigl||u_{1n}|^q-|u_{10}|^q\bigr||u_{2n}|^q\,dx \\
&\quad+\int_{\mathbb{R}^N}|u_{10}|^q\bigl||u_{2n}|^q-|u_{20}|^q\bigr|\,dx \\
&\le\bigl\||u_{1n}|^q-|u_{10}|^q\bigr\|_2\|u_{2n}\|_p^q+\bigl\||u_{2n}|^q-|u_{20}|^q\bigr\|_2\|u_{10}\|_p^q. \tag*{(54)}
\end{aligned} ∫ R N ∣ u 1 n ∣ q ∣ u 2 n ∣ q d x − ∫ R N ∣ u 10 ∣ q ∣ u 20 ∣ q d x ≤ ∫ R N ∣ u 1 n ∣ q − ∣ u 10 ∣ q ∣ u 2 n ∣ q d x + ∫ R N ∣ u 10 ∣ q ∣ u 2 n ∣ q − ∣ u 20 ∣ q d x ≤ ∣ u 1 n ∣ q − ∣ u 10 ∣ q 2 ∥ u 2 n ∥ p q + ∣ u 2 n ∣ q − ∣ u 20 ∣ q 2 ∥ u 10 ∥ p q . (54)
The proof is finished by letting n → ∞ n \to\infty n → ∞ in (54) . □ \square □
Lemma 2.5. For any ( u 1 , u 2 ) ∈ H s ( R N ) × H s ( R N ) (u_1,u_2) \in H^s(\mathbb{R}^N) \times H^s(\mathbb{R}^N) ( u 1 , u 2 ) ∈ H s ( R N ) × H s ( R N ) , the following refined Gagliardo-Nirenberg inequality holds:
∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) q d x ≤ p 2 a ∗ ( ∥ ( − Δ ) s 2 u 1 ∥ 2 2 + ∥ ( − Δ ) s 2 u 2 ∥ 2 2 ) ( ∫ R N ( u 1 2 + u 2 2 ) d x ) 2 s N . (55) \int_{\mathbb{R}^N} \left(|u_1|^2+|u_2|^2\right)^q dx \leq\frac{p}{2a^*}\left(\left\|(-\Delta)^{\frac{s}{2}}u_1\right\|_2^2+\left\|(-\Delta)^{\frac{s}{2}}u_2\right\|_2^2\right)\left(\int_{\mathbb{R}^N}\left(u_1^2+u_2^2\right)dx\right)^{\frac{2s}{N}}. \tag*{(55)} ∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) q d x ≤ 2 a ∗ p ( ( − Δ ) 2 s u 1 2 2 + ( − Δ ) 2 s u 2 2 2 ) ( ∫ R N ( u 1 2 + u 2 2 ) d x ) N 2 s . (55)
Proof. Let v = u 1 2 + u 2 2 v=\sqrt{u_1^2+u_2^2} v = u 1 2 + u 2 2 . Then v ≥ 0 v\geq0 v ≥ 0 and ∥ v ∥ 2 2 = ∫ R N ( u 1 2 + u 2 2 ) d x \|v\|_2^2=\int_{\mathbb{R}^N}(u_1^2+u_2^2)dx ∥ v ∥ 2 2 = ∫ R N ( u 1 2 + u 2 2 ) d x . Combining (1.25) and (B.1), we obtain
∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) q d x = ∫ R N v p d x ≤ p 2 a ∗ ∥ ( − Δ ) s 2 v ∥ 2 2 ∥ v ∥ 2 4 s N ≤ p 2 a ∗ ( ∥ ( − Δ ) s 2 u 1 ∥ 2 2 + ∥ ( − Δ ) s 2 u 2 ∥ 2 2 ) ( ∫ R N ( u 1 2 + u 2 2 ) d x ) 2 s N . (56) \begin{aligned}
\int_{\mathbb{R}^N}\left(|u_1|^2+|u_2|^2\right)^q dx
&=\int_{\mathbb{R}^N}v^p dx\\
&\leq\frac{p}{2a^*}\left\|(-\Delta)^{\frac{s}{2}}v\right\|_2^2\|v\|_2^{\frac{4s}{N}}\\
&\leq\frac{p}{2a^*}\left(\left\|(-\Delta)^{\frac{s}{2}}u_1\right\|_2^2+\left\|(-\Delta)^{\frac{s}{2}}u_2\right\|_2^2\right)\left(\int_{\mathbb{R}^N}\left(u_1^2+u_2^2\right)dx\right)^{\frac{2s}{N}}. \tag*{(56)}
\end{aligned} ∫ R N ( ∣ u 1 ∣ 2 + ∣ u 2 ∣ 2 ) q d x = ∫ R N v p d x ≤ 2 a ∗ p ( − Δ ) 2 s v 2 2 ∥ v ∥ 2 N 4 s ≤ 2 a ∗ p ( ( − Δ ) 2 s u 1 2 2 + ( − Δ ) 2 s u 2 2 2 ) ( ∫ R N ( u 1 2 + u 2 2 ) d x ) N 2 s . (56)
This completes the proof of (2.7). □
Proof of Theorem 1.2. Our proof is divided into three parts.
In the case 0 ≤ a 1 , a 2 < a ∗ 0\leq a_1,a_2<a^* 0 ≤ a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^* 0 < β < β ∗ , we claim that (1.17) has at least one minimizer. For any ( u 1 , u 2 ) ∈ M (u_1,u_2)\in\mathcal{M} ( u 1 , u 2 ) ∈ M , we introduce the total amplitude function v ( x ) = u 1 2 ( x ) + u 2 2 ( x ) v(x)=\sqrt{u_1^2(x)+u_2^2(x)} v ( x ) = u 1 2 ( x ) + u 2 2 ( x ) . On the support where v ( x ) > 0 v(x)>0 v ( x ) > 0 , we define the mass distribution fraction θ ( x ) = u 2 2 ( x ) v 2 ( x ) ∈ [ 0 , 1 ] \theta(x)=\frac{u_2^2(x)}{v^2(x)}\in[0,1] θ ( x ) = v 2 ( x ) u 2 2 ( x ) ∈ [ 0 , 1 ] . Because the exponents satisfy p = 2 q p=2q p = 2 q , the nonlinear terms can be factorized pointwise:
N ( u 1 , u 2 ) : = a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q = v ( x ) p [ a 1 θ ( x ) q + a 2 ( 1 − θ ( x ) ) q + 2 β θ ( x ) q / 2 ( 1 − θ ( x ) ) q / 2 ] . (57) \begin{aligned}
\mathcal{N}(u_1,u_2)&:=a_1|u_1|^p+a_2|u_2|^p+2\beta|u_1|^q|u_2|^q\\
&=v(x)^p\left[a_1\theta(x)^q+a_2(1-\theta(x))^q+2\beta\theta(x)^{q/2}(1-\theta(x))^{q/2}\right]. \tag*{(57)}
\end{aligned} N ( u 1 , u 2 ) := a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q = v ( x ) p [ a 1 θ ( x ) q + a 2 ( 1 − θ ( x ) ) q + 2 β θ ( x ) q /2 ( 1 − θ ( x ) ) q /2 ] . (57)
By the definition of Λ ( a 1 , a 2 , β ) \Lambda(a_1,a_2,\beta) Λ ( a 1 , a 2 , β ) in (1.26), we have N ( u 1 , u 2 ) ≤ Λ ( a 1 , a 2 , β ) v ( x ) p \mathcal{N}(u_1,u_2)\leq\Lambda(a_1,a_2,\beta)v(x)^p N ( u 1 , u 2 ) ≤ Λ ( a 1 , a 2 , β ) v ( x ) p almost everywhere. Combining (1.19), (1.25) and (B.1), and recalling that ∥ v ∥ 2 = 1 \|v\|_2=1 ∥ v ∥ 2 = 1 on M \mathcal{M} M , we obtain
∫ R N N ( u 1 , u 2 ) d x ≤ Λ ( a 1 , a 2 , β ) ∫ R N v p d x ≤ Λ ( a 1 , a 2 , β ) p 2 a ∗ ∥ ( − Δ ) s 2 v ∥ 2 2 ∥ v ∥ 2 4 s N ≤ p Λ ( a 1 , a 2 , β ) 2 a ∗ ∑ i = 1 2 ∥ ( − Δ ) s 2 u i ∥ 2 2 . (58) \begin{aligned}
\int_{\mathbb{R}^N}\mathcal{N}(u_1,u_2)dx
&\leq\Lambda(a_1,a_2,\beta)\int_{\mathbb{R}^N}v^pdx\\
&\leq\Lambda(a_1,a_2,\beta)\frac{p}{2a^*}\left\|(-\Delta)^{\frac{s}{2}}v\right\|_2^2\|v\|_2^{\frac{4s}{N}}\\
&\leq\frac{p\Lambda(a_1,a_2,\beta)}{2a^*}\sum_{i=1}^{2}\left\|(-\Delta)^{\frac{s}{2}}u_i\right\|_2^2. \tag*{(58)}
\end{aligned} ∫ R N N ( u 1 , u 2 ) d x ≤ Λ ( a 1 , a 2 , β ) ∫ R N v p d x ≤ Λ ( a 1 , a 2 , β ) 2 a ∗ p ( − Δ ) 2 s v 2 2 ∥ v ∥ 2 N 4 s ≤ 2 a ∗ p Λ ( a 1 , a 2 , β ) i = 1 ∑ 2 ( − Δ ) 2 s u i 2 2 . (58)
It follows from (1.18), (2.10) and Remark 1.1 that
E a 1 , a 2 , β ( u 1 , u 2 ) = ∑ i = 1 2 ( ∥ ( − Δ ) s 2 u i ∥ 2 2 + ∫ R N V i u i 2 d x ) − 2 p ∫ R N N ( u 1 , u 2 ) d x ≥ ( 1 − Λ ( a 1 , a 2 , β ) a ∗ ) ∑ i = 1 2 ∥ ( − Δ ) s 2 u i ∥ 2 2 + ∑ i = 1 2 ∫ R N V i u i 2 d x ≥ 0. (59) \begin{aligned}
E_{a_1,a_2,\beta}(u_1,u_2)
&=\sum_{i=1}^{2}\left(\left\|(-\Delta)^{\frac{s}{2}}u_i\right\|_2^2+\int_{\mathbb{R}^N}V_i u_i^2dx\right)-\frac{2}{p}\int_{\mathbb{R}^N}\mathcal{N}(u_1,u_2)dx\\
&\geq\left(1-\frac{\Lambda(a_1,a_2,\beta)}{a^*}\right)\sum_{i=1}^{2}\left\|(-\Delta)^{\frac{s}{2}}u_i\right\|_2^2+\sum_{i=1}^{2}\int_{\mathbb{R}^N}V_i u_i^2dx\\
&\geq0. \tag*{(59)}
\end{aligned} E a 1 , a 2 , β ( u 1 , u 2 ) = i = 1 ∑ 2 ( ( − Δ ) 2 s u i 2 2 + ∫ R N V i u i 2 d x ) − p 2 ∫ R N N ( u 1 , u 2 ) d x ≥ ( 1 − a ∗ Λ ( a 1 , a 2 , β ) ) i = 1 ∑ 2 ( − Δ ) 2 s u i 2 2 + i = 1 ∑ 2 ∫ R N V i u i 2 d x ≥ 0. (59)
holds for any ( u 1 , u 2 ) ∈ M (u_1,u_2) \in\mathcal{M} ( u 1 , u 2 ) ∈ M . Consequently, E a 1 , a 2 , β ( u 1 , u 2 ) E_{a_1,a_2,\beta}(u_1,u_2) E a 1 , a 2 , β ( u 1 , u 2 ) is bounded from below in M \mathcal{M} M . Let { ( u 1 n , u 2 n ) } ⊂ M \{(u_{1n},u_{2n})\} \subset\mathcal{M} {( u 1 n , u 2 n )} ⊂ M be a minimizing sequence for problem (1.17), which satisfies
∫ R N ( u 1 n 2 + u 2 n 2 ) d x = 1 , lim n → ∞ E a 1 , a 2 , β ( u 1 n , u 2 n ) = e ( a 1 , a 2 , β ) . (60) \int_{\mathbb{R}^N}(u_{1n}^2+u_{2n}^2)\,dx=1,\qquad\lim_{n\to\infty}E_{a_1,a_2,\beta}(u_{1n},u_{2n})=e(a_1,a_2,\beta). \tag*{(60)} ∫ R N ( u 1 n 2 + u 2 n 2 ) d x = 1 , n → ∞ lim E a 1 , a 2 , β ( u 1 n , u 2 n ) = e ( a 1 , a 2 , β ) . (60)
Analogous to the derivation of (2.11), one can infer that
E a 1 , a 2 , β ( u 1 n , u 2 n ) ≥ ( 1 − Λ ( a 1 , a 2 , β ) a ∗ ) ∑ i = 1 2 ∥ ( − Δ ) s 2 u i n ∥ 2 2 + ∑ i = 1 2 ∫ R N V i ( x ) u i n 2 d x . (61) E_{a_1,a_2,\beta}(u_{1n},u_{2n})\geq\left(1-\frac{\Lambda(a_1,a_2,\beta)}{a^*}\right)\sum_{i=1}^{2}\left\|(-\Delta)^{\frac{s}{2}}u_{in}\right\|_2^2+\sum_{i=1}^{2}\int_{\mathbb{R}^N}V_i(x)u_{in}^2\,dx. \tag*{(61)} E a 1 , a 2 , β ( u 1 n , u 2 n ) ≥ ( 1 − a ∗ Λ ( a 1 , a 2 , β ) ) i = 1 ∑ 2 ( − Δ ) 2 s u in 2 2 + i = 1 ∑ 2 ∫ R N V i ( x ) u in 2 d x . (61)
Passing to the limit n → ∞ n\to\infty n → ∞ on both sides of the above inequality, we deduce from (2.12) that the minimizing sequence { ( u 1 n , u 2 n ) } \{(u_{1n},u_{2n})\} {( u 1 n , u 2 n )} is uniformly bounded in H 1 × H 2 \mathcal{H}_1\times\mathcal{H}_2 H 1 × H 2 . From Lemma 2.1, passing to a subsequence if necessary, there exists ( u 10 , u 20 ) ∈ H 1 × H 2 (u_{10},u_{20})\in\mathcal{H}_1\times\mathcal{H}_2 ( u 10 , u 20 ) ∈ H 1 × H 2 such that
{ ( u 1 n , u 2 n ) ⇀ ( u 10 , u 20 ) weakly in H 1 × H 2 , ( u 1 n , u 2 n ) → ( u 10 , u 20 ) strongly in L r ( R N ) × L r ( R N ) , 2 ≤ r < 2 s ∗ . (62) \begin{cases}
(u_{1n},u_{2n})\rightharpoonup(u_{10},u_{20}) & \text{weakly in }\mathcal{H}_1\times\mathcal{H}_2,\\
(u_{1n},u_{2n})\to(u_{10},u_{20}) & \text{strongly in }L^r(\mathbb{R}^N)\times L^r(\mathbb{R}^N),\quad2\leq r<2_s^*. \tag*{(62)}
\end{cases} { ( u 1 n , u 2 n ) ⇀ ( u 10 , u 20 ) ( u 1 n , u 2 n ) → ( u 10 , u 20 ) weakly in H 1 × H 2 , strongly in L r ( R N ) × L r ( R N ) , 2 ≤ r < 2 s ∗ . (62)
Hence, we can deduce that
lim n → ∞ ∫ R N ( u 1 n 2 + u 2 n 2 ) d x = 1 = ∫ R N ( u 10 2 + u 20 2 ) d x , (63) \lim_{n\to\infty}\int_{\mathbb{R}^N}(u_{1n}^2+u_{2n}^2)\,dx=1=\int_{\mathbb{R}^N}(u_{10}^2+u_{20}^2)\,dx, \tag*{(63)} n → ∞ lim ∫ R N ( u 1 n 2 + u 2 n 2 ) d x = 1 = ∫ R N ( u 10 2 + u 20 2 ) d x , (63)
i.e., ( u 10 , u 20 ) ∈ M (u_{10},u_{20})\in\mathcal{M} ( u 10 , u 20 ) ∈ M . Furthermore, it follows from the weak lower semi-continuity of the norms ∥ ⋅ ∥ H i \|\cdot\|_{\mathcal{H}_i} ∥ ⋅ ∥ H i and Lemma 2.4 that
e ( a 1 , a 2 , β ) ≤ E a 1 , a 2 , β ( u 10 , u 20 ) ≤ lim n → ∞ E a 1 , a 2 , β ( u 1 n , u 2 n ) = e ( a 1 , a 2 , β ) . (64) e(a_1,a_2,\beta)\leq E_{a_1,a_2,\beta}(u_{10},u_{20})\leq\lim_{n\to\infty}E_{a_1,a_2,\beta}(u_{1n},u_{2n})=e(a_1,a_2,\beta). \tag*{(64)} e ( a 1 , a 2 , β ) ≤ E a 1 , a 2 , β ( u 10 , u 20 ) ≤ n → ∞ lim E a 1 , a 2 , β ( u 1 n , u 2 n ) = e ( a 1 , a 2 , β ) . (64)
This implies that E a 1 , a 2 , β ( u 10 , u 20 ) = e ( a 1 , a 2 , β ) E_{a_1,a_2,\beta}(u_{10},u_{20})=e(a_1,a_2,\beta) E a 1 , a 2 , β ( u 10 , u 20 ) = e ( a 1 , a 2 , β ) , establishing ( u 10 , u 20 ) (u_{10},u_{20}) ( u 10 , u 20 ) as a minimizer of the problem (1.17).
(ii) In the case a 1 > a ∗ a_1>a^* a 1 > a ∗ or a 2 > a ∗ a_2>a^* a 2 > a ∗ or β > β ∗ \beta>\beta^* β > β ∗ , we infer that (1.17) has no minimizers. Let φ ∈ C 0 ∞ ( R N ) \varphi\in C_0^\infty(\mathbb{R}^N) φ ∈ C 0 ∞ ( R N ) be a cut-off function satisfying
φ ( x ) ≡ 1 for ∣ x ∣ ≤ 1 , φ ( x ) ≡ 0 for ∣ x ∣ ≥ 2 , 0 ≤ φ ( x ) ≤ 1 for all x ∈ R N . (65) \varphi(x)\equiv1\text{ for }|x|\leq1,\qquad\varphi(x)\equiv0\text{ for }|x|\geq2,\qquad0\leq\varphi(x)\leq1\text{ for all }x\in\mathbb{R}^N. \tag*{(65)} φ ( x ) ≡ 1 for ∣ x ∣ ≤ 1 , φ ( x ) ≡ 0 for ∣ x ∣ ≥ 2 , 0 ≤ φ ( x ) ≤ 1 for all x ∈ R N . (65)
Besides, for any σ > 0 \sigma>0 σ > 0 , set
{ ϕ 1 σ ( x ) : = θ A σ ∥ Q ∥ 2 σ N / 2 φ ( x − x 0 ) Q ( σ ( x − x 0 ) ) , ϕ 2 σ ( x ) : = 1 − θ A σ ∥ Q ∥ 2 σ N / 2 φ ( x − x 0 ) Q ( σ ( x − x 0 ) ) , (66) \begin{cases}
\phi_{1\sigma}(x):=\dfrac{\sqrt{\theta}A_\sigma}{\|Q\|_2}\sigma^{N/2}\varphi(x-x_0)Q(\sigma(x-x_0)),\\
\phi_{2\sigma}(x):=\dfrac{\sqrt{1-\theta}A_\sigma}{\|Q\|_2}\sigma^{N/2}\varphi(x-x_0)Q(\sigma(x-x_0)), \tag*{(66)}
\end{cases} ⎩ ⎨ ⎧ ϕ 1 σ ( x ) := ∥ Q ∥ 2 θ A σ σ N /2 φ ( x − x 0 ) Q ( σ ( x − x 0 )) , ϕ 2 σ ( x ) := ∥ Q ∥ 2 1 − θ A σ σ N /2 φ ( x − x 0 ) Q ( σ ( x − x 0 )) , (66)
where A σ > 0 A_\sigma>0 A σ > 0 is chosen such that ( ϕ 1 σ , ϕ 2 σ ) ∈ M (\phi_{1\sigma},\phi_{2\sigma})\in\mathcal{M} ( ϕ 1 σ , ϕ 2 σ ) ∈ M , θ ∈ [ 0 , 1 ] \theta\in[0,1] θ ∈ [ 0 , 1 ] and x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N will be determined later. Since ( ϕ 1 σ , ϕ 2 σ ) ∈ M (\phi_{1\sigma},\phi_{2\sigma})\in\mathcal{M} ( ϕ 1 σ , ϕ 2 σ ) ∈ M , we have
1 = ∫ R N A σ 2 ∥ Q ∥ 2 2 σ N φ 2 ( x − x 0 ) Q 2 ( σ ( x − x 0 ) ) d x = ∫ R N A σ 2 ∥ Q ∥ 2 2 φ 2 ( σ − 1 x ) Q 2 ( x ) d x . (67) 1=\int_{\mathbb{R}^N}\frac{A_\sigma^2}{\|Q\|_2^2}\sigma^N\varphi^2(x-x_0)Q^2(\sigma(x-x_0))\,dx=\int_{\mathbb{R}^N}\frac{A_\sigma^2}{\|Q\|_2^2}\varphi^2(\sigma^{-1}x)Q^2(x)\,dx. \tag*{(67)} 1 = ∫ R N ∥ Q ∥ 2 2 A σ 2 σ N φ 2 ( x − x 0 ) Q 2 ( σ ( x − x 0 )) d x = ∫ R N ∥ Q ∥ 2 2 A σ 2 φ 2 ( σ − 1 x ) Q 2 ( x ) d x . (67)
One can deduce from (1.23) and (2.19) that
∣ 1 − A σ 2 ∣ = ∣ ∫ R N [ A σ 2 ∥ Q ∥ 2 2 ( φ 2 ( σ − 1 x ) − 1 ) Q 2 ( x ) ] d x ∣ ≤ ∫ R N ∖ B σ ∣ A σ 2 ∥ Q ∥ 2 2 ( φ 2 ( σ − 1 x ) − 1 ) Q 2 ( x ) ∣ d x ≤ C ∫ R N ∖ B σ ∣ x ∣ − 2 N − 4 s d x ≤ C σ − N − 4 s . (68) \begin{aligned}
\left|1-A_\sigma^2\right|
&=\left|\int_{\mathbb{R}^N}\left[\frac{A_\sigma^2}{\|Q\|_2^2}\left(\varphi^2(\sigma^{-1}x)-1\right)Q^2(x)\right]\,dx\right|\\
&\leq\int_{\mathbb{R}^N\setminus B_\sigma}\left|\frac{A_\sigma^2}{\|Q\|_2^2}\left(\varphi^2(\sigma^{-1}x)-1\right)Q^2(x)\right|\,dx\\
&\leq C\int_{\mathbb{R}^N\setminus B_\sigma}|x|^{-2N-4s}\,dx\\
&\leq C\sigma^{-N-4s}. \tag*{(68)}
\end{aligned} 1 − A σ 2 = ∫ R N [ ∥ Q ∥ 2 2 A σ 2 ( φ 2 ( σ − 1 x ) − 1 ) Q 2 ( x ) ] d x ≤ ∫ R N ∖ B σ ∥ Q ∥ 2 2 A σ 2 ( φ 2 ( σ − 1 x ) − 1 ) Q 2 ( x ) d x ≤ C ∫ R N ∖ B σ ∣ x ∣ − 2 N − 4 s d x ≤ C σ − N − 4 s . (68)
for σ \sigma σ sufficiently large, where B σ : = { x ∈ R N ∣ ∣ x ∣ < σ } B_\sigma:= \{x \in\mathbb{R}^N \mid|x| < \sigma\} B σ := { x ∈ R N ∣ ∣ x ∣ < σ } . Thus, we can deduce that
1 − O ( σ − N − 4 s ) ≤ A σ 2 ≤ 1 + O ( σ − N − 4 s ) as σ → ∞ . (69) 1 - O(\sigma^{-N-4s}) \le A_\sigma^2 \le1 + O(\sigma^{-N-4s}) \quad\text{as } \sigma\to\infty. \tag*{(69)} 1 − O ( σ − N − 4 s ) ≤ A σ 2 ≤ 1 + O ( σ − N − 4 s ) as σ → ∞. (69)
Now, let us turn to estimate each term in E a 1 , a 2 , β ( φ 1 σ , φ 2 σ ) E_{a_1,a_2,\beta}(\varphi_{1\sigma},\varphi_{2\sigma}) E a 1 , a 2 , β ( φ 1 σ , φ 2 σ ) . Denoting φ σ = φ ( σ − 1 x ) \varphi_\sigma= \varphi(\sigma^{-1}x) φ σ = φ ( σ − 1 x ) and applying the scaling properties of the fractional Laplacian to the test function φ 1 σ \varphi_{1\sigma} φ 1 σ with scaling coefficient σ \sigma σ , we thus obtain
∫ R N ∣ ( − Δ ) s 2 φ 1 σ ∣ 2 d x = θ A σ 2 ∥ Q ∥ 2 2 σ 2 s ∫ R N ∣ ( − Δ ) s 2 ( φ σ Q ) ∣ 2 d x . (70) \int_{\mathbb{R}^N} \left|(-\Delta)^{\frac{s}{2}}\varphi_{1\sigma}\right|^2 dx = \frac{\theta A_\sigma^2}{\|Q\|_2^2}\sigma^{2s}\int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}(\varphi_\sigma Q)\right|^2dx. \tag*{(70)} ∫ R N ( − Δ ) 2 s φ 1 σ 2 d x = ∥ Q ∥ 2 2 θ A σ 2 σ 2 s ∫ R N ( − Δ ) 2 s ( φ σ Q ) 2 d x . (70)
By Lemma D.1, there exists C > 0 C>0 C > 0 such that for all σ \sigma σ sufficiently large,
∫ R N ∣ ( − Δ ) s 2 ( φ σ Q ) ∣ 2 d x ≤ ∫ R N ∣ ( − Δ ) s 2 Q ∣ 2 d x + C σ − 4 s . (71) \int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}(\varphi_\sigma Q)\right|^2 dx \le\int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}Q\right|^2 dx+C\sigma^{-4s}. \tag*{(71)} ∫ R N ( − Δ ) 2 s ( φ σ Q ) 2 d x ≤ ∫ R N ( − Δ ) 2 s Q 2 d x + C σ − 4 s . (71)
One can infer from (69) –(71) and (C.7) that
∫ R N ∣ ( − Δ ) s 2 φ 1 σ ∣ 2 d x ≤ θ A σ 2 ∥ Q ∥ 2 2 σ 2 s ( ∫ R N ∣ ( − Δ ) s 2 Q ∣ 2 d x + C σ − 4 s ) = θ A σ 2 ∥ Q ∥ 2 2 σ 2 s ∫ R N ∣ ( − Δ ) s 2 Q ∣ 2 d x + C θ A σ 2 ∥ Q ∥ 2 2 σ − 2 s ≤ θ σ 2 s + O ( σ − 2 s ) . (72) \begin{aligned}
\int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}\varphi_{1\sigma}\right|^2dx
&\le\frac{\theta A_\sigma^2}{\|Q\|_2^2}\sigma^{2s}\left(\int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}Q\right|^2dx+C\sigma^{-4s}\right)\\
&= \frac{\theta A_\sigma^2}{\|Q\|_2^2}\sigma^{2s}\int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}Q\right|^2dx+\frac{C\theta A_\sigma^2}{\|Q\|_2^2}\sigma^{-2s}\\
&\le\theta\sigma^{2s}+O(\sigma^{-2s}). \tag*{(72)}
\end{aligned} ∫ R N ( − Δ ) 2 s φ 1 σ 2 d x ≤ ∥ Q ∥ 2 2 θ A σ 2 σ 2 s ( ∫ R N ( − Δ ) 2 s Q 2 d x + C σ − 4 s ) = ∥ Q ∥ 2 2 θ A σ 2 σ 2 s ∫ R N ( − Δ ) 2 s Q 2 d x + ∥ Q ∥ 2 2 C θ A σ 2 σ − 2 s ≤ θ σ 2 s + O ( σ − 2 s ) . (72)
Apply the same steps to the test function φ 2 σ \varphi_{2\sigma} φ 2 σ , one can also obtain
∫ R N ∣ ( − Δ ) s 2 φ 2 σ ∣ 2 d x ≤ ( 1 − θ ) σ 2 s + O ( σ − 2 s ) . (73) \int_{\mathbb{R}^N}\left|(-\Delta)^{\frac{s}{2}}\varphi_{2\sigma}\right|^2dx\le(1-\theta)\sigma^{2s}+O(\sigma^{-2s}). \tag*{(73)} ∫ R N ( − Δ ) 2 s φ 2 σ 2 d x ≤ ( 1 − θ ) σ 2 s + O ( σ − 2 s ) . (73)
For the nonlinear terms, we first claim
∫ R N ∣ φ 1 σ ∣ p d x ≤ p θ p 2 2 a ∗ σ 2 s + O ( σ − N − 2 s ) . (74) \int_{\mathbb{R}^N}|\varphi_{1\sigma}|^pdx\le\frac{p\theta^{\frac{p}{2}}}{2a^*}\sigma^{2s}+O(\sigma^{-N-2s}). \tag*{(74)} ∫ R N ∣ φ 1 σ ∣ p d x ≤ 2 a ∗ p θ 2 p σ 2 s + O ( σ − N − 2 s ) . (74)
Indeed, by the scaling property, one can infer
∫ R N ∣ φ 1 σ ∣ p d x = θ p 2 A σ p ∥ Q ∥ 2 p σ N ( p 2 − 1 ) ∫ R N φ p ( σ − 1 x ) ∣ Q ( x ) ∣ p d x . (75) \int_{\mathbb{R}^N}|\varphi_{1\sigma}|^pdx=\frac{\theta^{\frac{p}{2}}A_\sigma^p}{\|Q\|_2^p}\sigma^{N(\frac{p}{2}-1)}\int_{\mathbb{R}^N}\varphi^p(\sigma^{-1}x)|Q(x)|^pdx. \tag*{(75)} ∫ R N ∣ φ 1 σ ∣ p d x = ∥ Q ∥ 2 p θ 2 p A σ p σ N ( 2 p − 1 ) ∫ R N φ p ( σ − 1 x ) ∣ Q ( x ) ∣ p d x . (75)
From the decay of Q Q Q , we have
∣ ∫ R N φ p ( σ − 1 x ) ∣ Q ( x ) ∣ p d x − ∫ R N ∣ Q ( x ) ∣ p d x ∣ = ∫ R N ( 1 − φ p ( σ − 1 x ) ) ∣ Q ( x ) ∣ p d x = ∫ ∣ x ∣ > σ ( 1 − φ p ( σ − 1 x ) ) ∣ Q ( x ) ∣ p d x ≤ ∫ ∣ x ∣ > σ ∣ Q ( x ) ∣ p d x ≤ C ∫ ∣ x ∣ > σ ∣ x ∣ − p ( N + 2 s ) d x ≤ C σ − p ( N + 2 s ) + N ≤ C σ − N − 4 s , (76) \begin{aligned}
\left|\int_{\mathbb{R}^N}\varphi^p(\sigma^{-1}x)|Q(x)|^pdx-\int_{\mathbb{R}^N}|Q(x)|^pdx\right|
&=\int_{\mathbb{R}^N}(1-\varphi^p(\sigma^{-1}x))|Q(x)|^pdx\\
&=\int_{|x|>\sigma}(1-\varphi^p(\sigma^{-1}x))|Q(x)|^pdx\\
&\le\int_{|x|>\sigma}|Q(x)|^pdx\\
&\le C\int_{|x|>\sigma}|x|^{-p(N+2s)}dx\\
&\le C\sigma^{-p(N+2s)+N}\\
&\le C\sigma^{-N-4s}, \tag*{(76)}
\end{aligned} ∫ R N φ p ( σ − 1 x ) ∣ Q ( x ) ∣ p d x − ∫ R N ∣ Q ( x ) ∣ p d x = ∫ R N ( 1 − φ p ( σ − 1 x )) ∣ Q ( x ) ∣ p d x = ∫ ∣ x ∣ > σ ( 1 − φ p ( σ − 1 x )) ∣ Q ( x ) ∣ p d x ≤ ∫ ∣ x ∣ > σ ∣ Q ( x ) ∣ p d x ≤ C ∫ ∣ x ∣ > σ ∣ x ∣ − p ( N + 2 s ) d x ≤ C σ − p ( N + 2 s ) + N ≤ C σ − N − 4 s , (76)
for σ \sigma σ sufficiently large. Combining (69) , (76) and (C.1), one can deduce from (75) that
∫ R N ∣ ϕ 1 σ ∣ p d x ≤ θ p 2 σ 2 s ∥ Q ∥ 2 p ∥ Q ∥ p p + O ( σ − N − 2 s ) = p θ p 2 σ 2 s 2 a ∗ + O ( σ − N − 2 s ) . (77) \begin{aligned}
\int_{\mathbb{R}^{N}}|\phi_{1\sigma}|^{p}\,dx
&\leq\frac{\theta^{\frac{p}{2}}\sigma^{2s}}{\|Q\|_{2}^{p}}\|Q\|_{p}^{p}+O(\sigma^{-N-2s}) \\
&=\frac{p\theta^{\frac{p}{2}}\sigma^{2s}}{2a^{*}}+O(\sigma^{-N-2s}). \tag*{(77)}
\end{aligned} ∫ R N ∣ ϕ 1 σ ∣ p d x ≤ ∥ Q ∥ 2 p θ 2 p σ 2 s ∥ Q ∥ p p + O ( σ − N − 2 s ) = 2 a ∗ p θ 2 p σ 2 s + O ( σ − N − 2 s ) . (77)
Thus, the claim (74) is proved. Note that, in a similar way, we can also obtain the following two estimates.
∫ R N ∣ ϕ 2 σ ∣ p d x ≤ p ( 1 − θ ) p 2 σ 2 s 2 a ∗ + O ( σ − N − 2 s ) , (78) \int_{\mathbb{R}^{N}}|\phi_{2\sigma}|^{p}\,dx\leq\frac{p(1-\theta)^{\frac{p}{2}}\sigma^{2s}}{2a^{*}}+O(\sigma^{-N-2s}), \tag*{(78)} ∫ R N ∣ ϕ 2 σ ∣ p d x ≤ 2 a ∗ p ( 1 − θ ) 2 p σ 2 s + O ( σ − N − 2 s ) , (78)
and
∫ R N ∣ ϕ 1 σ ∣ q ∣ ϕ 2 σ ∣ q d x ≤ p θ q 2 ( 1 − θ ) q 2 σ 2 s 2 a ∗ + O ( σ − N − 2 s ) . (79) \int_{\mathbb{R}^{N}}|\phi_{1\sigma}|^{q}|\phi_{2\sigma}|^{q}\,dx\leq\frac{p\theta^{\frac{q}{2}}(1-\theta)^{\frac{q}{2}}\sigma^{2s}}{2a^{*}}+O(\sigma^{-N-2s}). \tag*{(79)} ∫ R N ∣ ϕ 1 σ ∣ q ∣ ϕ 2 σ ∣ q d x ≤ 2 a ∗ p θ 2 q ( 1 − θ ) 2 q σ 2 s + O ( σ − N − 2 s ) . (79)
Furthermore, since the function x ↦ V i ( x ) φ 2 ( x − x 0 ) x\mapsto V_i(x)\varphi^2(x-x_0) x ↦ V i ( x ) φ 2 ( x − x 0 ) (i = 1 , 2 i=1,2 i = 1 , 2 ) is bounded, by the dominated convergence theorem, we infer
lim σ → ∞ ∫ R N V 1 ( x ) ϕ 1 σ 2 d x = lim σ → ∞ ∫ R N θ A σ 2 ∥ Q ∥ 2 2 σ N V 1 ( x ) φ 2 ( x − x 0 ) Q 2 ( σ ( x − x 0 ) ) d x = lim σ → ∞ ∫ R N θ A σ 2 ∥ Q ∥ 2 2 V 1 ( σ − 1 y + x 0 ) φ 2 ( σ − 1 y ) Q 2 ( y ) d y = θ ∥ Q ∥ 2 2 V 1 ( x 0 ) ∫ R N Q 2 ( y ) d y = θ V 1 ( x 0 ) , (80) \begin{aligned}
\lim_{\sigma\to\infty}\int_{\mathbb{R}^{N}}V_1(x)\phi_{1\sigma}^{2}\,dx
&=\lim_{\sigma\to\infty}\int_{\mathbb{R}^{N}}\frac{\theta A_{\sigma}^{2}}{\|Q\|_{2}^{2}}\sigma^{N}V_1(x)\varphi^2(x-x_0)Q^2(\sigma(x-x_0))\,dx \\
&=\lim_{\sigma\to\infty}\int_{\mathbb{R}^{N}}\frac{\theta A_{\sigma}^{2}}{\|Q\|_{2}^{2}}V_1(\sigma^{-1}y+x_0)\varphi^2(\sigma^{-1}y)Q^2(y)\,dy \\
&=\frac{\theta}{\|Q\|_{2}^{2}}V_1(x_0)\int_{\mathbb{R}^{N}}Q^2(y)\,dy=\theta V_1(x_0), \tag*{(80)}
\end{aligned} σ → ∞ lim ∫ R N V 1 ( x ) ϕ 1 σ 2 d x = σ → ∞ lim ∫ R N ∥ Q ∥ 2 2 θ A σ 2 σ N V 1 ( x ) φ 2 ( x − x 0 ) Q 2 ( σ ( x − x 0 )) d x = σ → ∞ lim ∫ R N ∥ Q ∥ 2 2 θ A σ 2 V 1 ( σ − 1 y + x 0 ) φ 2 ( σ − 1 y ) Q 2 ( y ) d y = ∥ Q ∥ 2 2 θ V 1 ( x 0 ) ∫ R N Q 2 ( y ) d y = θ V 1 ( x 0 ) , (80)
and similarly we can also obtain
lim σ → ∞ ∫ R N V 2 ( x ) ϕ 2 σ 2 d x = ( 1 − θ ) V 2 ( x 0 ) . (81) \lim_{\sigma\to\infty}\int_{\mathbb{R}^{N}}V_2(x)\phi_{2\sigma}^{2}\,dx=(1-\theta)V_2(x_0). \tag*{(81)} σ → ∞ lim ∫ R N V 2 ( x ) ϕ 2 σ 2 d x = ( 1 − θ ) V 2 ( x 0 ) . (81)
Then, it follows from (18) , (72) –(74) and (78) –(81) that
E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) = ∑ i = 1 2 ∫ R N ( ∣ ( − Δ ) s 2 ϕ i σ ∣ 2 + V i ( x ) ϕ i σ 2 ) d x − 2 p [ a 1 ∫ R N ∣ ϕ 1 σ ∣ p d x + a 2 ∫ R N ∣ ϕ 2 σ ∣ p d x + 2 β ∫ R N ∣ ϕ 1 σ ∣ q ∣ ϕ 2 σ ∣ q d x ] ≤ σ 2 s a ∗ [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ q 2 ( 1 − θ ) q 2 ] + O ( σ − 2 s ) + ∫ R N V 1 ( x ) ϕ 1 σ 2 d x + ∫ R N V 2 ( x ) ϕ 2 σ 2 d x = σ 2 s a ∗ [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ q 2 ( 1 − θ ) q 2 ] + O ( σ − 2 s ) + θ V 1 ( x 0 ) + ( 1 − θ ) V 2 ( x 0 ) . (82) \begin{aligned}
E_{a_1,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma})
={}&\sum_{i=1}^{2}\int_{\mathbb{R}^{N}}\left(|(-\Delta)^{\frac{s}{2}}\phi_{i\sigma}|^{2}+V_i(x)\phi_{i\sigma}^{2}\right)\,dx \\
&-\frac{2}{p}\left[a_1\int_{\mathbb{R}^{N}}|\phi_{1\sigma}|^{p}\,dx+a_2\int_{\mathbb{R}^{N}}|\phi_{2\sigma}|^{p}\,dx+2\beta\int_{\mathbb{R}^{N}}|\phi_{1\sigma}|^{q}|\phi_{2\sigma}|^{q}\,dx\right] \\
\leq{}&\frac{\sigma^{2s}}{a^{*}}\left[a^{*}-a_1\theta^{q}-a_2(1-\theta)^{q}-2\beta\theta^{\frac{q}{2}}(1-\theta)^{\frac{q}{2}}\right]+O(\sigma^{-2s}) \\
&+\int_{\mathbb{R}^{N}}V_1(x)\phi_{1\sigma}^{2}\,dx+\int_{\mathbb{R}^{N}}V_2(x)\phi_{2\sigma}^{2}\,dx \\
={}&\frac{\sigma^{2s}}{a^{*}}\left[a^{*}-a_1\theta^{q}-a_2(1-\theta)^{q}-2\beta\theta^{\frac{q}{2}}(1-\theta)^{\frac{q}{2}}\right]+O(\sigma^{-2s}) \\
&+\theta V_1(x_0)+(1-\theta)V_2(x_0). \tag*{(82)}
\end{aligned} E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) = ≤ = i = 1 ∑ 2 ∫ R N ( ∣ ( − Δ ) 2 s ϕ iσ ∣ 2 + V i ( x ) ϕ iσ 2 ) d x − p 2 [ a 1 ∫ R N ∣ ϕ 1 σ ∣ p d x + a 2 ∫ R N ∣ ϕ 2 σ ∣ p d x + 2 β ∫ R N ∣ ϕ 1 σ ∣ q ∣ ϕ 2 σ ∣ q d x ] a ∗ σ 2 s [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ 2 q ( 1 − θ ) 2 q ] + O ( σ − 2 s ) + ∫ R N V 1 ( x ) ϕ 1 σ 2 d x + ∫ R N V 2 ( x ) ϕ 2 σ 2 d x a ∗ σ 2 s [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ 2 q ( 1 − θ ) 2 q ] + O ( σ − 2 s ) + θ V 1 ( x 0 ) + ( 1 − θ ) V 2 ( x 0 ) . (82)
Next, we finish the proof of (ii) by considering the following three cases.
Case 1: a 1 > a ∗ a_1>a^{*} a 1 > a ∗ . One can obtain from (82) by taking θ = 1 \theta=1 θ = 1 and any x 0 ∈ R N x_0\in\mathbb{R}^{N} x 0 ∈ R N that
E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ σ 2 s a ∗ ( a ∗ − a 1 ) + V 1 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (83) E_{a_1,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma})\leq\frac{\sigma^{2s}}{a^{*}}(a^{*}-a_1)+V_1(x_0)+O(\sigma^{-2s})\to-\infty\quad\text{as }\sigma\to\infty, \tag*{(83)} E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ a ∗ σ 2 s ( a ∗ − a 1 ) + V 1 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (83)
which implies e ( a 1 , a 2 , β ) = − ∞ e(a_1,a_2,\beta)=-\infty e ( a 1 , a 2 , β ) = − ∞ in this case. Therefore, if a 1 > a ∗ a_1>a^* a 1 > a ∗ , we deduce that there is no minimizer of (17) .
Case 2: a 2 > a ∗ a_2>a^* a 2 > a ∗ . By taking θ = 0 \theta=0 θ = 0 and any x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N in (82) , one can infer
E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ σ 2 s a ∗ ( a ∗ − a 2 ) + V 2 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (84) E_{a_1,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma})\leq\frac{\sigma^{2s}}{a^*}(a^*-a_2)+V_2(x_0)+O(\sigma^{-2s})\to-\infty\quad\text{as }\sigma\to\infty, \tag*{(84)} E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ a ∗ σ 2 s ( a ∗ − a 2 ) + V 2 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (84)
which indicates e ( a 1 , a 2 , β ) = − ∞ e(a_1,a_2,\beta)=-\infty e ( a 1 , a 2 , β ) = − ∞ . Hence, if a 2 > a ∗ a_2>a^* a 2 > a ∗ , we conclude that there is no minimizer of (17) .
Case 3: β > β ∗ \beta>\beta^* β > β ∗ . Without loss of generality, one can suppose 0 ≤ a 1 , a 2 < a ∗ 0\leq a_1,a_2<a^* 0 ≤ a 1 , a 2 < a ∗ in this case. From the definition of β ∗ \beta^* β ∗ , there exists some θ 0 ∈ ( 0 , 1 ) \theta_0\in(0,1) θ 0 ∈ ( 0 , 1 ) such that
a ∗ − a 1 θ 0 q − a 2 ( 1 − θ 0 ) q = 2 β ∗ θ 0 q / 2 ( 1 − θ 0 ) q / 2 . (85) a^*-a_1\theta_0^q-a_2(1-\theta_0)^q=2\beta^*\theta_0^{q/2}(1-\theta_0)^{q/2}. \tag*{(85)} a ∗ − a 1 θ 0 q − a 2 ( 1 − θ 0 ) q = 2 β ∗ θ 0 q /2 ( 1 − θ 0 ) q /2 . (85)
Taking θ = θ 0 \theta=\theta_0 θ = θ 0 and any x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N in (82) , we deduce that
E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ 2 σ 2 s a ∗ θ 0 q / 2 ( 1 − θ 0 ) q / 2 ( β ∗ − β ) + θ 0 V 1 ( x 0 ) + ( 1 − θ 0 ) V 2 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (86) \begin{aligned}
E_{a_1,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma})\leq{}&\frac{2\sigma^{2s}}{a^*}\theta_0^{q/2}(1-\theta_0)^{q/2}(\beta^*-\beta)+\theta_0V_1(x_0)+(1-\theta_0)V_2(x_0)\\
&+O(\sigma^{-2s})\to-\infty\quad\text{as }\sigma\to\infty, \tag*{(86)}
\end{aligned} E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ a ∗ 2 σ 2 s θ 0 q /2 ( 1 − θ 0 ) q /2 ( β ∗ − β ) + θ 0 V 1 ( x 0 ) + ( 1 − θ 0 ) V 2 ( x 0 ) + O ( σ − 2 s ) → − ∞ as σ → ∞ , (86)
which shows e ( a 1 , a 2 , β ) = − ∞ e(a_1,a_2,\beta)=-\infty e ( a 1 , a 2 , β ) = − ∞ as well. We thus conclude that, if β > β ∗ \beta>\beta^* β > β ∗ , there is no minimizer of (17) .
(iii) Finally, we prove that there is no minimizer of (17) in the case a 1 = a ∗ a_1=a^* a 1 = a ∗ or a 2 = a ∗ a_2=a^* a 2 = a ∗ or β = β ∗ \beta=\beta^* β = β ∗ . By (ii), it remains to consider the parameters satisfying 0 ≤ a 1 , a 2 ≤ a ∗ 0\leq a_1,a_2\leq a^* 0 ≤ a 1 , a 2 ≤ a ∗ and Λ ( a 1 , a 2 , β ) ≤ a ∗ \Lambda(a_1,a_2,\beta)\leq a^* Λ ( a 1 , a 2 , β ) ≤ a ∗ ; for these parameters, (59) yields E a 1 , a 2 , β ( u 1 , u 2 ) ≥ 0 E_{a_1,a_2,\beta}(u_1,u_2)\geq0 E a 1 , a 2 , β ( u 1 , u 2 ) ≥ 0 for every ( u 1 , u 2 ) ∈ M (u_1,u_2)\in\mathcal{M} ( u 1 , u 2 ) ∈ M . From the assumption ( Ω ) (\Omega) ( Ω ) we know that
V = { x ∈ R N : V 1 ( x ) = V 2 ( x ) = 0 } ≠ ∅ . (87) \mathcal{V}=\{x\in\mathbb{R}^N:V_1(x)=V_2(x)=0\}\neq\varnothing. \tag*{(87)} V = { x ∈ R N : V 1 ( x ) = V 2 ( x ) = 0 } = ∅ . (87)
Therefore, we can choose x 0 ∈ V x_0\in\mathcal{V} x 0 ∈ V such that V 1 ( x 0 ) = V 2 ( x 0 ) = 0 V_1(x_0)=V_2(x_0)=0 V 1 ( x 0 ) = V 2 ( x 0 ) = 0 . It then follows from (59) and (82) that
0 ≤ e ( a 1 , a 2 , β ) ≤ E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ σ 2 s a ∗ [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ q / 2 ( 1 − θ ) q / 2 ] + O ( σ − 2 s ) . (88) \begin{aligned}
0\leq e(a_1,a_2,\beta)\leq E_{a_1,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma}) \\
&\leq\frac{\sigma^{2s}}{a^*}\left[a^*-a_1\theta^q-a_2(1-\theta)^q-2\beta\theta^{q/2}(1-\theta)^{q/2}\right]+O(\sigma^{-2s}). \tag*{(88)}
\end{aligned} 0 ≤ e ( a 1 , a 2 , β ) ≤ E a 1 , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) ≤ a ∗ σ 2 s [ a ∗ − a 1 θ q − a 2 ( 1 − θ ) q − 2 β θ q /2 ( 1 − θ ) q /2 ] + O ( σ − 2 s ) . (88)
We next discuss the following three cases similarly as part (ii).
Case 1: a 1 = a ∗ a_1=a^* a 1 = a ∗ . Let θ = 1 \theta=1 θ = 1 in (88) and σ → ∞ \sigma\to\infty σ → ∞ , we then obtain
0 ≤ e ( a ∗ , a 2 , β ) ≤ lim σ → ∞ E a ∗ , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) = 0. (89) 0\leq e(a^*,a_2,\beta)\leq\lim_{\sigma\to\infty}E_{a^*,a_2,\beta}(\phi_{1\sigma},\phi_{2\sigma})=0. \tag*{(89)} 0 ≤ e ( a ∗ , a 2 , β ) ≤ σ → ∞ lim E a ∗ , a 2 , β ( ϕ 1 σ , ϕ 2 σ ) = 0. (89)
Next, we argue by contradiction. Suppose e ( a ∗ , a 2 , β ) e(a^*,a_2,\beta) e ( a ∗ , a 2 , β ) has a minimizer ( u 10 , u 20 ) (u_{10},u_{20}) ( u 10 , u 20 ) . From (58) , (59) and (89) , we infer that
∫ R N ( ∣ ( − Δ ) s / 2 u 10 ∣ 2 + ∣ ( − Δ ) s / 2 u 20 ∣ 2 ) d x = 2 a ∗ p ∫ R N ( u 10 2 + u 20 2 ) q d x , (90) \int_{\mathbb{R}^N}\left(\left|(-\Delta)^{s/2}u_{10}\right|^2+\left|(-\Delta)^{s/2}u_{20}\right|^2\right)\,dx=\frac{2a^*}{p}\int_{\mathbb{R}^N}(u_{10}^2+u_{20}^2)^q\,dx, \tag*{(90)} ∫ R N ( ( − Δ ) s /2 u 10 2 + ( − Δ ) s /2 u 20 2 ) d x = p 2 a ∗ ∫ R N ( u 10 2 + u 20 2 ) q d x , (90)
∫ R N V 1 ( x ) u 10 2 d x = ∫ R N V 2 ( x ) u 20 2 d x = 0. (91) \int_{\mathbb{R}^N}V_1(x)u_{10}^2\,dx=\int_{\mathbb{R}^N}V_2(x)u_{20}^2\,dx=0. \tag*{(91)} ∫ R N V 1 ( x ) u 10 2 d x = ∫ R N V 2 ( x ) u 20 2 d x = 0. (91)
One can deduce from Lemma E.1 and the equality (90) that
( u 10 ( x ) , u 20 ( x ) ) = ( τ sin θ Q ( α x + η ) , τ cos θ Q ( α x + η ) ) (92) (u_{10}(x),u_{20}(x))=(\tau\sin\theta\,Q(\alpha x+\eta),\tau\cos\theta\,Q(\alpha x+\eta)) \tag*{(92)} ( u 10 ( x ) , u 20 ( x )) = ( τ sin θ Q ( α x + η ) , τ cos θ Q ( α x + η )) (92)
with some τ > 0 \tau>0 τ > 0 , α > 0 \alpha>0 α > 0 , θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) and η ∈ R N \eta\in\mathbb{R}^{N} η ∈ R N . However, (2.43) shows that the support of ( u 10 , u 20 ) (u_{10},u_{20}) ( u 10 , u 20 ) is compact, which contradicts (2.44). Therefore, in this case, there is no minimizer of e ( a ∗ , a 2 , β ) e(a^{*},a_{2},\beta) e ( a ∗ , a 2 , β ) .
Case 2: a 2 = a ∗ a_{2}=a^{*} a 2 = a ∗ . Let θ = 0 \theta=0 θ = 0 in (2.40) and σ → ∞ \sigma\to\infty σ → ∞ , we then have
0 ≤ e ( a 1 , a ∗ , β ) ≤ lim σ → ∞ E a 1 , a ∗ , β ( ϕ 1 σ , ϕ 2 σ ) = 0 , (93) 0\leq e(a_{1},a^{*},\beta)\leq\lim_{\sigma\to\infty}E_{a_{1},a^{*},\beta}(\phi_{1\sigma},\phi_{2\sigma})=0, \tag*{(93)} 0 ≤ e ( a 1 , a ∗ , β ) ≤ σ → ∞ lim E a 1 , a ∗ , β ( ϕ 1 σ , ϕ 2 σ ) = 0 , (93)
which shows that there is no minimizer of e ( a 1 , a ∗ , β ) e(a_{1},a^{*},\beta) e ( a 1 , a ∗ , β ) by arguing as above.
Case 3: β = β ∗ \beta=\beta^{*} β = β ∗ and 0 ≤ a 1 , a 2 < a ∗ 0\leq a_{1},a_{2}<a^{*} 0 ≤ a 1 , a 2 < a ∗ . Let θ = θ 0 \theta=\theta_{0} θ = θ 0 in (2.40) and σ → ∞ \sigma\to\infty σ → ∞ , we then get
0 ≤ e ( a 1 , a 2 , β ∗ ) ≤ lim σ → ∞ E a 1 , a 2 , β ∗ ( ϕ 1 σ , ϕ 2 σ ) = 0 , (94) 0\leq e(a_{1},a_{2},\beta^{*})\leq\lim_{\sigma\to\infty}E_{a_{1},a_{2},\beta^{*}}(\phi_{1\sigma},\phi_{2\sigma})=0, \tag*{(94)} 0 ≤ e ( a 1 , a 2 , β ∗ ) ≤ σ → ∞ lim E a 1 , a 2 , β ∗ ( ϕ 1 σ , ϕ 2 σ ) = 0 , (94)
which implies that there is no minimizer of e ( a 1 , a 2 , β ∗ ) e(a_{1},a_{2},\beta^{*}) e ( a 1 , a 2 , β ∗ ) by arguing as above.
Furthermore, in the case 0 ≤ a 1 , a 2 < a ∗ 0\leq a_{1},a_{2}<a^{*} 0 ≤ a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^{*} 0 < β < β ∗ , let θ = θ 0 \theta=\theta_{0} θ = θ 0 and σ = ( β ∗ − β ) − 1 4 s \sigma=(\beta^{*}-\beta)^{-\frac{1}{4s}} σ = ( β ∗ − β ) − 4 s 1 in (2.40), one can infer
0 ≤ e ( a 1 , a 2 , β ) ≤ σ 2 s a ∗ ⋅ 2 θ 0 q 2 ( 1 − θ 0 ) q 2 ( β ∗ − β ) + O ( σ − 2 s ) → 0 as β → β ∗ . (95) 0\leq e(a_{1},a_{2},\beta)\leq\frac{\sigma^{2s}}{a^{*}}\cdot2\theta_{0}^{\frac{q}{2}}(1-\theta_{0})^{\frac{q}{2}}(\beta^{*}-\beta)+O(\sigma^{-2s})\to0\quad\text{as }\beta\to\beta^{*}. \tag*{(95)} 0 ≤ e ( a 1 , a 2 , β ) ≤ a ∗ σ 2 s ⋅ 2 θ 0 2 q ( 1 − θ 0 ) 2 q ( β ∗ − β ) + O ( σ − 2 s ) → 0 as β → β ∗ . (95)
Combining this and (2.46), we thus deduce that
lim β → β ∗ e ( a 1 , a 2 , β ) = e ( a 1 , a 2 , β ∗ ) = 0 , 0 ≤ a 1 , a 2 < a ∗ , (96) \lim_{\beta\to\beta^{*}}e(a_{1},a_{2},\beta)=e(a_{1},a_{2},\beta^{*})=0,\quad0\leq a_{1},a_{2}<a^{*}, \tag*{(96)} β → β ∗ lim e ( a 1 , a 2 , β ) = e ( a 1 , a 2 , β ∗ ) = 0 , 0 ≤ a 1 , a 2 < a ∗ , (96)
which completes the proof of Theorem 1.2.
General concentration behavior
In this section, we investigate the concentration behavior of the minimizers for the problem (1.17) under general trapping potentials as β → β ∗ \beta\to\beta^{*} β → β ∗ . We specifically focus on the one-dimensional, half-order fractional Laplacian regime, which implies p = 4 p=4 p = 4 and q = 2 q=2 q = 2 . Since (B.1) gives ∥ ( − Δ ) s / 2 ∣ u i ∣ ∥ 2 ≤ ∥ ( − Δ ) s / 2 u i ∥ 2 \lVert(-\Delta)^{s/2}|u_{i}|\rVert_{2}\leq\lVert(-\Delta)^{s/2}u_{i}\rVert_{2} ∥( − Δ ) s /2 ∣ u i ∣ ∥ 2 ≤ ∥( − Δ ) s /2 u i ∥ 2 and E a 1 , a 2 , β E_{a_{1},a_{2},\beta} E a 1 , a 2 , β depends on u i u_{i} u i only through ∣ u i ∣ |u_{i}| ∣ u i ∣ , we may assume that the minimizers are nonnegative. For this purpose, we start with the following two lemmas.
Lemma 3.1. Suppose (Ω) holds and let ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) be a nonnegative minimizer of e ( a 1 , a 2 , β ) e(a_{1},a_{2},\beta) e ( a 1 , a 2 , β ) for 0 ≤ a 1 , a 2 < a ∗ 0\leq a_{1},a_{2}<a^{*} 0 ≤ a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^{*} 0 < β < β ∗ . Then the following hold:
(i) ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) blows up as β → β ∗ \beta\to\beta^{*} β → β ∗ , i.e., for i = 1 , 2 i=1,2 i = 1 , 2 ,
lim β → β ∗ ∫ R ∣ ( − Δ ) 1 4 u i β ∣ 2 d x = ∞ and lim β → β ∗ ∫ R u i β 4 d x = ∞ . (97) \lim_{\beta\to\beta^{*}}\int_{\mathbb{R}}\left|(-\Delta)^{\frac{1}{4}}u_{i\beta}\right|^{2}dx=\infty\quad\text{and}\quad\lim_{\beta\to\beta^{*}}\int_{\mathbb{R}}u_{i\beta}^{4}dx=\infty. \tag*{(97)} β → β ∗ lim ∫ R ( − Δ ) 4 1 u i β 2 d x = ∞ and β → β ∗ lim ∫ R u i β 4 d x = ∞. (97) (ii) ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) also satisfies
lim β → β ∗ ∫ R V 1 ( x ) u 1 β 2 d x = lim β → β ∗ ∫ R V 2 ( x ) u 2 β 2 d x = 0 , (98) \lim_{\beta\to\beta^{*}}\int_{\mathbb{R}}V_{1}(x)u_{1\beta}^{2}dx=\lim_{\beta\to\beta^{*}}\int_{\mathbb{R}}V_{2}(x)u_{2\beta}^{2}dx=0, \tag*{(98)} β → β ∗ lim ∫ R V 1 ( x ) u 1 β 2 d x = β → β ∗ lim ∫ R V 2 ( x ) u 2 β 2 d x = 0 , (98) lim β → β ∗ [ ( β ∗ − β ) ∫ R u 1 β 2 u 2 β 2 d x ] = 0 , (99) \lim_{\beta\to\beta^{*}}\left[(\beta^{*}-\beta)\int_{\mathbb{R}}u_{1\beta}^{2}u_{2\beta}^{2}dx\right]=0, \tag*{(99)} β → β ∗ lim [ ( β ∗ − β ) ∫ R u 1 β 2 u 2 β 2 d x ] = 0 , (99)
lim β → β ∗ [ a ∗ − a 1 ∫ R u 1 β 4 d x − a ∗ − a 2 ∫ R u 2 β 4 d x ] = 0 , (100) \lim_{\beta\to\beta^*}\left[\sqrt{a^*-a_1}\sqrt{\int_{\mathbb{R}}u_{1\beta}^4\,dx}-\sqrt{a^*-a_2}\sqrt{\int_{\mathbb{R}}u_{2\beta}^4\,dx}\right]=0, \tag*{(100)} β → β ∗ lim [ a ∗ − a 1 ∫ R u 1 β 4 d x − a ∗ − a 2 ∫ R u 2 β 4 d x ] = 0 , (100)
lim β → β ∗ [ ∫ R u 1 β 4 d x ∫ R u 2 β 4 d x − ∫ R u 1 β 2 u 2 β 2 d x ] = 0 , (101) \lim_{\beta\to\beta^*}\left[\sqrt{\int_{\mathbb{R}}u_{1\beta}^4\,dx}\sqrt{\int_{\mathbb{R}}u_{2\beta}^4\,dx}-\int_{\mathbb{R}}u_{1\beta}^2u_{2\beta}^2\,dx\right]=0, \tag*{(101)} β → β ∗ lim [ ∫ R u 1 β 4 d x ∫ R u 2 β 4 d x − ∫ R u 1 β 2 u 2 β 2 d x ] = 0 , (101)
as well as
lim β → β ∗ ∫ R ( ∣ ( − Δ ) 1 4 u 1 β ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β ∣ 2 ) d x ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x = a ∗ 2 , (102) \lim_{\beta\to\beta^*}\frac{\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac14}u_{1\beta}\right|^2+\left|(-\Delta)^{\frac14}u_{2\beta}\right|^2\right)\,dx}{\int_{\mathbb{R}}\left(u_{1\beta}^2+u_{2\beta}^2\right)^2\,dx}=\frac{a^*}{2}, \tag*{(102)} β → β ∗ lim ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x ∫ R ( ( − Δ ) 4 1 u 1 β 2 + ( − Δ ) 4 1 u 2 β 2 ) d x = 2 a ∗ , (102)
lim β → β ∗ ∫ R u 1 β 4 d x ∫ R u 2 β 4 d x = a ∗ − a 2 a ∗ − a 1 . (103) \lim_{\beta\to\beta^*}\frac{\int_{\mathbb{R}}u_{1\beta}^4\,dx}{\int_{\mathbb{R}}u_{2\beta}^4\,dx}=\frac{a^*-a_2}{a^*-a_1}. \tag*{(103)} β → β ∗ lim ∫ R u 2 β 4 d x ∫ R u 1 β 4 d x = a ∗ − a 1 a ∗ − a 2 . (103)
Proof. Combining the refined Gagliardo–Nirenberg inequality (55) and the Schwarz inequality, we deduce that
E a 1 , a 2 , β ( u 1 β , u 2 β ) = ∑ i = 1 2 ∫ R ∣ ( − Δ ) 1 4 u i β ∣ 2 d x − a ∗ 2 ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x + 1 2 [ a ∗ − a 1 ∫ R u 1 β 4 d x − a ∗ − a 2 ∫ R u 2 β 4 d x ] 2 + ( a ∗ − a 1 ) ( a ∗ − a 2 ) [ ∫ R u 1 β 4 d x ∫ R u 2 β 4 d x − ∫ R u 1 β 2 u 2 β 2 d x ] + ( β ∗ − β ) ∫ R u 1 β 2 u 2 β 2 d x + ∑ i = 1 2 ∫ R V i ( x ) u i β 2 d x ≥ 0. (104) \begin{aligned}
E_{a_1,a_2,\beta}(u_{1\beta},u_{2\beta})
={}&\sum_{i=1}^{2}\int_{\mathbb{R}}\left|(-\Delta)^{\frac14}u_{i\beta}\right|^2\,dx-\frac{a^*}{2}\int_{\mathbb{R}}\left(u_{1\beta}^2+u_{2\beta}^2\right)^2\,dx\\
&+\frac{1}{2}\left[\sqrt{a^*-a_1}\sqrt{\int_{\mathbb{R}}u_{1\beta}^4\,dx}-\sqrt{a^*-a_2}\sqrt{\int_{\mathbb{R}}u_{2\beta}^4\,dx}\right]^2\\
&+\sqrt{(a^*-a_1)(a^*-a_2)}\left[\sqrt{\int_{\mathbb{R}}u_{1\beta}^4\,dx}\sqrt{\int_{\mathbb{R}}u_{2\beta}^4\,dx}-\int_{\mathbb{R}}u_{1\beta}^2u_{2\beta}^2\,dx\right]\\
&+(\beta^*-\beta)\int_{\mathbb{R}}u_{1\beta}^2u_{2\beta}^2\,dx+\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(x)u_{i\beta}^2\,dx\\
\ge{}&0. \tag*{(104)}
\end{aligned} E a 1 , a 2 , β ( u 1 β , u 2 β ) = ≥ i = 1 ∑ 2 ∫ R ( − Δ ) 4 1 u i β 2 d x − 2 a ∗ ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x + 2 1 [ a ∗ − a 1 ∫ R u 1 β 4 d x − a ∗ − a 2 ∫ R u 2 β 4 d x ] 2 + ( a ∗ − a 1 ) ( a ∗ − a 2 ) [ ∫ R u 1 β 4 d x ∫ R u 2 β 4 d x − ∫ R u 1 β 2 u 2 β 2 d x ] + ( β ∗ − β ) ∫ R u 1 β 2 u 2 β 2 d x + i = 1 ∑ 2 ∫ R V i ( x ) u i β 2 d x 0. (104)
One can immediately deduce from (28) that (3.2)–(3.5) and the following hold,
lim β → β ∗ [ ∫ R ( ∣ ( − Δ ) 1 4 u 1 β ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β ∣ 2 ) d x − a ∗ 2 ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x ] = 0. (105) \lim_{\beta\to\beta^*}\left[\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac14}u_{1\beta}\right|^2+\left|(-\Delta)^{\frac14}u_{2\beta}\right|^2\right)\,dx-\frac{a^*}{2}\int_{\mathbb{R}}\left(u_{1\beta}^2+u_{2\beta}^2\right)^2\,dx\right]=0. \tag*{(105)} β → β ∗ lim [ ∫ R ( ( − Δ ) 4 1 u 1 β 2 + ( − Δ ) 4 1 u 2 β 2 ) d x − 2 a ∗ ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x ] = 0. (105)
Note that
lim β → β ∗ ∫ R ( ∣ ( − Δ ) 1 4 u 1 β ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β ∣ 2 ) d x = ∞ . (106) \lim_{\beta\to\beta^*}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac14}u_{1\beta}\right|^2+\left|(-\Delta)^{\frac14}u_{2\beta}\right|^2\right)\,dx=\infty. \tag*{(106)} β → β ∗ lim ∫ R ( ( − Δ ) 4 1 u 1 β 2 + ( − Δ ) 4 1 u 2 β 2 ) d x = ∞. (106)
In fact, suppose this does not hold, then we can infer that there exists a sequence { β k } \{\beta_k\} { β k } satisfying β k → β ∗ \beta_k\to\beta^* β k → β ∗ as k → ∞ k\to\infty k → ∞ , such that { ( u 1 β k , u 2 β k ) } \{(u_{1\beta_k},u_{2\beta_k})\} {( u 1 β k , u 2 β k )} is bounded uniformly in H 1 × H 2 \mathcal{H}_1\times\mathcal{H}_2 H 1 × H 2 . Therefore, from Lemma 2.1, we deduce that there exists a subsequence of { β k } \{\beta_k\} { β k } , still denoted by { β k } \{\beta_k\} { β k } , and ( u 10 , u 20 ) ∈ H 1 × H 2 (u_{10},u_{20})\in\mathcal{H}_1\times\mathcal{H}_2 ( u 10 , u 20 ) ∈ H 1 × H 2 such that
( u 1 β k , u 2 β k ) ⇀ ( u 10 , u 20 ) weakly in H 1 × H 2 , ( u 1 β k , u 2 β k ) → ( u 10 , u 20 ) strongly in L r ( R ) × L r ( R ) for 2 ≤ r < ∞ . (107) \begin{aligned}
(u_{1\beta_k},u_{2\beta_k})&\rightharpoonup(u_{10},u_{20})\quad\text{weakly in }\mathcal{H}_1\times\mathcal{H}_2,\\
(u_{1\beta_k},u_{2\beta_k})&\to(u_{10},u_{20})\quad\text{strongly in }L^r(\mathbb{R})\times L^r(\mathbb{R})\text{ for }2\le r<\infty. \tag*{(107)}
\end{aligned} ( u 1 β k , u 2 β k ) ( u 1 β k , u 2 β k ) ⇀ ( u 10 , u 20 ) weakly in H 1 × H 2 , → ( u 10 , u 20 ) strongly in L r ( R ) × L r ( R ) for 2 ≤ r < ∞. (107)
Then, it follows from Lemma 2.4 and the weak lower semi-continuity of the norm ∥ ⋅ ∥ H i \|\cdot\|_{\mathcal{H}_i} ∥ ⋅ ∥ H i that
0 = e ( a 1 , a 2 , β ∗ ) ≤ E a 1 , a 2 , β ∗ ( u 10 , u 20 ) ≤ lim k → ∞ E a 1 , a 2 , β k ( u 1 β k , u 2 β k ) = lim k → ∞ e ( a 1 , a 2 , β k ) = 0. (108) 0=e(a_1,a_2,\beta^*)\le E_{a_1,a_2,\beta^*}(u_{10},u_{20})\le\lim_{k\to\infty}E_{a_1,a_2,\beta_k}(u_{1\beta_k},u_{2\beta_k})=\lim_{k\to\infty}e(a_1,a_2,\beta_k)=0. \tag*{(108)} 0 = e ( a 1 , a 2 , β ∗ ) ≤ E a 1 , a 2 , β ∗ ( u 10 , u 20 ) ≤ k → ∞ lim E a 1 , a 2 , β k ( u 1 β k , u 2 β k ) = k → ∞ lim e ( a 1 , a 2 , β k ) = 0. (108)
It follows that ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) is a minimizer of E a 1 , a 2 , β ∗ E_{a_1,a_2,\beta^*} E a 1 , a 2 , β ∗ , which contradicts Theorem 1.2 and hence (106) holds. Combining (105) and (106) , (102) thus follows. Moreover, from (101) , (102) and (106) , we further conclude that
lim β → β ∗ ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x = lim β → β ∗ ( ∫ R u 1 β 4 d x + ∫ R u 2 β 4 d x ) 2 = ∞ . (109) \lim_{\beta\to\beta^*}\int_{\mathbb{R}}(u_{1\beta}^2+u_{2\beta}^2)^2\,\mathrm{d}x
=\lim_{\beta\to\beta^*}\left(\sqrt{\int_{\mathbb{R}}u_{1\beta}^4\,\mathrm{d}x}+\sqrt{\int_{\mathbb{R}}u_{2\beta}^4\,\mathrm{d}x}\right)^2=\infty. \tag*{(109)} β → β ∗ lim ∫ R ( u 1 β 2 + u 2 β 2 ) 2 d x = β → β ∗ lim ( ∫ R u 1 β 4 d x + ∫ R u 2 β 4 d x ) 2 = ∞. (109)
It then follows from (100) that for i = 1 , 2 i=1,2 i = 1 , 2 ,
lim β ↗ β ∗ ∫ R u i β 4 d x = ∞ . (110) \lim_{\beta\nearrow\beta^*}\int_{\mathbb{R}}u_{i\beta}^4\,\mathrm{d}x=\infty. \tag*{(110)} β ↗ β ∗ lim ∫ R u i β 4 d x = ∞. (110)
Therefore, we can deduce from the Gagliardo–Nirenberg inequality (25) that (97) holds. Finally, (97) and (100) show that (103) is true, which completes the proof.
For the nonnegative minimizer ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) of (17) , define
ε β : = ( ∫ R ( ∣ ( − Δ ) 1 4 u 1 β ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β ∣ 2 ) d x ) − 1 > 0 , (111) \varepsilon_\beta:=\left(\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}u_{1\beta}\right|^2+\left|(-\Delta)^{\frac{1}{4}}u_{2\beta}\right|^2\right)\,\mathrm{d}x\right)^{-1}>0, \tag*{(111)} ε β := ( ∫ R ( ( − Δ ) 4 1 u 1 β 2 + ( − Δ ) 4 1 u 2 β 2 ) d x ) − 1 > 0 , (111)
which satisfies ε β → 0 \varepsilon_\beta\to0 ε β → 0 as β → β ∗ \beta\to\beta^* β → β ∗ according to (106) . Next, we introduce another lemma that will play a fundamental role in the proof of Theorem 1.4 .
Lemma 3.2. Suppose ( Ω ) (\Omega) ( Ω ) holds and let ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) be a nonnegative minimizer of e ( a 1 , a 2 , β ) e(a_1,a_2,\beta) e ( a 1 , a 2 , β ) for 0 ≤ a 1 , a 2 < a ∗ 0\le a_1,a_2<a^* 0 ≤ a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^* 0 < β < β ∗ . Then for any sequence β k → β ∗ \beta_k\to\beta^* β k → β ∗ as k → ∞ k\to\infty k → ∞ , we have
(i) There exist a sequence { η β k } ⊂ R \{\eta_{\beta_k}\}\subset\mathbb{R} { η β k } ⊂ R and constants r 0 , δ > 0 r_0,\delta>0 r 0 , δ > 0 such that
w i β k ( x ) : = ε β k 1 2 u i β k ( ε β k x + ε β k η β k ) (112) w_{i\beta_k}(x):=\varepsilon_{\beta_k}^{\frac{1}{2}}u_{i\beta_k}\left(\varepsilon_{\beta_k}x+\varepsilon_{\beta_k}\eta_{\beta_k}\right) \tag*{(112)} w i β k ( x ) := ε β k 2 1 u i β k ( ε β k x + ε β k η β k ) (112) satisfies
lim inf k → ∞ ∫ B r 0 ( 0 ) ∣ w i β k ( x ) ∣ 2 d x ≥ δ > 0 , i = 1 , 2. (113) \liminf_{k\to\infty}\int_{B_{r_0}(0)}|w_{i\beta_k}(x)|^2\,\mathrm{d}x\ge\delta>0,\qquad i=1,2. \tag*{(113)} k → ∞ lim inf ∫ B r 0 ( 0 ) ∣ w i β k ( x ) ∣ 2 d x ≥ δ > 0 , i = 1 , 2. (113) Furthermore,
∫ R ( ∣ ( − Δ ) 1 4 w 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 w 2 β k ∣ 2 ) d x = ∫ R ( w 1 β k 2 + w 2 β k 2 ) d x = 1 , (114) \int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}w_{2\beta_k}\right|^2\right)\,\mathrm{d}x
=\int_{\mathbb{R}}\left(w_{1\beta_k}^2+w_{2\beta_k}^2\right)\,\mathrm{d}x=1, \tag*{(114)} ∫ R ( ( − Δ ) 4 1 w 1 β k 2 + ( − Δ ) 4 1 w 2 β k 2 ) d x = ∫ R ( w 1 β k 2 + w 2 β k 2 ) d x = 1 , (114) lim k → ∞ [ β ∗ − β k ε β k ∫ R w 1 β k 2 w 2 β k 2 d x ] = 0 , (115) \lim_{k\to\infty}\left[\frac{\beta^*-\beta_k}{\varepsilon_{\beta_k}}\int_{\mathbb{R}}w_{1\beta_k}^2w_{2\beta_k}^2\,\mathrm{d}x\right]=0, \tag*{(115)} k → ∞ lim [ ε β k β ∗ − β k ∫ R w 1 β k 2 w 2 β k 2 d x ] = 0 , (115) lim k → ∞ ∫ R ( w 1 β k 2 + w 2 β k 2 ) 2 d x = 2 a ∗ , (116) \lim_{k\to\infty}\int_{\mathbb{R}}\left(w_{1\beta_k}^2+w_{2\beta_k}^2\right)^2\,\mathrm{d}x=\frac{2}{a^*}, \tag*{(116)} k → ∞ lim ∫ R ( w 1 β k 2 + w 2 β k 2 ) 2 d x = a ∗ 2 , (116) lim k → ∞ ∫ R w 1 β k 4 d x = 2 γ 2 a ∗ , lim k → ∞ ∫ R w 2 β k 4 d x = 2 ( 1 − γ ) 2 a ∗ , \lim_{k\to\infty}\int_{\mathbb{R}}w_{1\beta_k}^4\,\mathrm{d}x=\frac{2\gamma^2}{a^*},\qquad
\lim_{k\to\infty}\int_{\mathbb{R}}w_{2\beta_k}^4\,\mathrm{d}x=\frac{2(1-\gamma)^2}{a^*}, k → ∞ lim ∫ R w 1 β k 4 d x = a ∗ 2 γ 2 , k → ∞ lim ∫ R w 2 β k 4 d x = a ∗ 2 ( 1 − γ ) 2 , lim k → ∞ ∫ R w 1 β k 2 w 2 β k 2 d x = 2 γ ( 1 − γ ) a ∗ . (117) \lim_{k\to\infty}\int_{\mathbb{R}}w_{1\beta_k}^2w_{2\beta_k}^2\,\mathrm{d}x=\frac{2\gamma(1-\gamma)}{a^*}. \tag*{(117)} k → ∞ lim ∫ R w 1 β k 2 w 2 β k 2 d x = a ∗ 2 γ ( 1 − γ ) . (117) (ii) There exists a subsequence of { β k } \{\beta_k\} { β k } , still denoted by { β k } \{\beta_k\} { β k } , such that
x β k : = ε β k η β k → x 0 as k → ∞ (118) x_{\beta_k}:=\varepsilon_{\beta_k}\eta_{\beta_k}\to x_0\qquad\text{as }k\to\infty \tag*{(118)} x β k := ε β k η β k → x 0 as k → ∞ (118) for some x 0 ∈ V x_0\in\mathcal{V} x 0 ∈ V .
(iii) There exists a subsequence of { β k } \{\beta_k\} { β k } , still denoted by { β k } \{\beta_k\} { β k } , and η 0 ∈ R \eta_0 \in\mathbb{R} η 0 ∈ R such that
lim k → ∞ w 1 β k = γ ∥ Q ∥ 2 Q ( x − η 0 ) , lim k → ∞ w 2 β k = 1 − γ ∥ Q ∥ 2 Q ( x − η 0 ) , (119) \begin{aligned}
\lim_{k\to\infty} w_{1\beta_k} &= \frac{\sqrt{\gamma}}{\lVert Q\rVert_2}Q(x-\eta_0),\\
\lim_{k\to\infty} w_{2\beta_k} &= \frac{\sqrt{1-\gamma}}{\lVert Q\rVert_2}Q(x-\eta_0), \tag*{(119)}
\end{aligned} k → ∞ lim w 1 β k k → ∞ lim w 2 β k = ∥ Q ∥ 2 γ Q ( x − η 0 ) , = ∥ Q ∥ 2 1 − γ Q ( x − η 0 ) , (119)
strongly in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) .
Proof. (i) Denote
v i β k : = ε β k 1 2 u i β k ( ε β k x ) , i = 1 , 2. (120) v_{i\beta_k}:=\varepsilon_{\beta_k}^{\frac{1}{2}}u_{i\beta_k}(\varepsilon_{\beta_k}x),\qquad i=1,2. \tag*{(120)} v i β k := ε β k 2 1 u i β k ( ε β k x ) , i = 1 , 2. (120)
By a simple calculation, one obtains from (3.15) and the mass constraint that
∫ R ( ∣ ( − Δ ) 1 4 v 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 v 2 β k ∣ 2 ) d x = ε β k ∫ R ( ∣ ( − Δ ) 1 4 u 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β k ∣ 2 ) d x = 1 = ∫ R ( ∣ u 1 β k ∣ 2 + ∣ u 2 β k ∣ 2 ) d x = ∫ R ( ∣ v 1 β k ∣ 2 + ∣ v 2 β k ∣ 2 ) d x . (121) \begin{aligned}
\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}v_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}v_{2\beta_k}\right|^2\right)\,dx
&=\varepsilon_{\beta_k}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}u_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}u_{2\beta_k}\right|^2\right)\,dx\\
&=1\\
&=\int_{\mathbb{R}}\left(|u_{1\beta_k}|^2+|u_{2\beta_k}|^2\right)\,dx\\
&=\int_{\mathbb{R}}\left(|v_{1\beta_k}|^2+|v_{2\beta_k}|^2\right)\,dx. \tag*{(121)}
\end{aligned} ∫ R ( ( − Δ ) 4 1 v 1 β k 2 + ( − Δ ) 4 1 v 2 β k 2 ) d x = ε β k ∫ R ( ( − Δ ) 4 1 u 1 β k 2 + ( − Δ ) 4 1 u 2 β k 2 ) d x = 1 = ∫ R ( ∣ u 1 β k ∣ 2 + ∣ u 2 β k ∣ 2 ) d x = ∫ R ( ∣ v 1 β k ∣ 2 + ∣ v 2 β k ∣ 2 ) d x . (121)
Besides, it comes from ∫ R v 1 β k 2 v 2 β k 2 d x = ε β k ∫ R u 1 β k 2 u 2 β k 2 d x \int_{\mathbb{R}}v_{1\beta_k}^2v_{2\beta_k}^2\,dx=\varepsilon_{\beta_k}\int_{\mathbb{R}}u_{1\beta_k}^2u_{2\beta_k}^2\,dx ∫ R v 1 β k 2 v 2 β k 2 d x = ε β k ∫ R u 1 β k 2 u 2 β k 2 d x and (3.3) that
lim k → ∞ [ β ∗ − β k ε β k ∫ R v 1 β k 2 v 2 β k 2 d x ] = lim k → ∞ [ ( β ∗ − β k ) ∫ R u 1 β k 2 u 2 β k 2 d x ] = 0. (122) \lim_{k\to\infty}\left[\frac{\beta^*-\beta_k}{\varepsilon_{\beta_k}}\int_{\mathbb{R}}v_{1\beta_k}^2v_{2\beta_k}^2\,dx\right]
=\lim_{k\to\infty}\left[(\beta^*-\beta_k)\int_{\mathbb{R}}u_{1\beta_k}^2u_{2\beta_k}^2\,dx\right]=0. \tag*{(122)} k → ∞ lim [ ε β k β ∗ − β k ∫ R v 1 β k 2 v 2 β k 2 d x ] = k → ∞ lim [ ( β ∗ − β k ) ∫ R u 1 β k 2 u 2 β k 2 d x ] = 0. (122)
Similarly from (3.4) and (3.5), we can obtain
lim k → ∞ [ a ∗ − a 1 ∫ R v 1 β k 4 d x − a ∗ − a 2 ∫ R v 2 β k 4 d x ] 2 = 0 (123) \lim_{k\to\infty}\left[\sqrt{a^*-a_1}\sqrt{\int_{\mathbb{R}}v_{1\beta_k}^4\,dx}-\sqrt{a^*-a_2}\sqrt{\int_{\mathbb{R}}v_{2\beta_k}^4\,dx}\right]^2=0 \tag*{(123)} k → ∞ lim [ a ∗ − a 1 ∫ R v 1 β k 4 d x − a ∗ − a 2 ∫ R v 2 β k 4 d x ] 2 = 0 (123)
and
lim k → ∞ [ ∫ R v 1 β k 4 d x ∫ R v 2 β k 4 d x − ∫ R v 1 β k 2 v 2 β k 2 d x ] = 0. (124) \lim_{k\to\infty}\left[\sqrt{\int_{\mathbb{R}}v_{1\beta_k}^4\,dx}\sqrt{\int_{\mathbb{R}}v_{2\beta_k}^4\,dx}-\int_{\mathbb{R}}v_{1\beta_k}^2v_{2\beta_k}^2\,dx\right]=0. \tag*{(124)} k → ∞ lim [ ∫ R v 1 β k 4 d x ∫ R v 2 β k 4 d x − ∫ R v 1 β k 2 v 2 β k 2 d x ] = 0. (124)
Moreover, from (3.6) and (3.25), one can easily conclude that
lim k → ∞ ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x = lim k → ∞ ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x ∫ R ( ∣ ( − Δ ) 1 4 v 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 v 2 β k ∣ 2 ) d x = lim k → ∞ ∫ R ( u 1 β k 2 + u 2 β k 2 ) 2 d x ∫ R ( ∣ ( − Δ ) 1 4 u 1 β k ∣ 2 + ∣ ( − Δ ) 1 4 u 2 β k ∣ 2 ) d x = 2 a ∗ . (125) \begin{aligned}
\lim_{k\to\infty}\int_{\mathbb{R}}(v_{1\beta_k}^2+v_{2\beta_k}^2)^2\,dx
&=\lim_{k\to\infty}\frac{\int_{\mathbb{R}}(v_{1\beta_k}^2+v_{2\beta_k}^2)^2\,dx}{\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}v_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}v_{2\beta_k}\right|^2\right)\,dx}\\
&=\lim_{k\to\infty}\frac{\int_{\mathbb{R}}(u_{1\beta_k}^2+u_{2\beta_k}^2)^2\,dx}{\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}u_{1\beta_k}\right|^2+\left|(-\Delta)^{\frac{1}{4}}u_{2\beta_k}\right|^2\right)\,dx}\\
&=\frac{2}{a^*}. \tag*{(125)}
\end{aligned} k → ∞ lim ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x = k → ∞ lim ∫ R ( ( − Δ ) 4 1 v 1 β k 2 + ( − Δ ) 4 1 v 2 β k 2 ) d x ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x = k → ∞ lim ∫ R ( ( − Δ ) 4 1 u 1 β k 2 + ( − Δ ) 4 1 u 2 β k 2 ) d x ∫ R ( u 1 β k 2 + u 2 β k 2 ) 2 d x = a ∗ 2 . (125)
Combining (123) –(125) and (30) , we can deduce that
2 a ∗ = lim k → ∞ ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x = lim k → ∞ ( ∫ R v 1 β k 4 d x + ∫ R v 2 β k 4 d x ) 2 = 1 γ 2 lim k → ∞ ∫ R v 1 β k 4 d x = 1 ( 1 − γ ) 2 lim k → ∞ ∫ R v 2 β k 4 d x , (126) \begin{aligned}
\frac{2}{a^{*}} &= \lim_{k\to\infty}\int_{\mathbb{R}}\left(v_{1\beta_k}^{2}+v_{2\beta_k}^{2}\right)^{2}\,dx \\
&= \lim_{k\to\infty}\left(\sqrt{\int_{\mathbb{R}}v_{1\beta_k}^{4}\,dx}+\sqrt{\int_{\mathbb{R}}v_{2\beta_k}^{4}\,dx}\right)^{2} \\
&= \frac{1}{\gamma^{2}}\lim_{k\to\infty}\int_{\mathbb{R}}v_{1\beta_k}^{4}\,dx \\
&= \frac{1}{(1-\gamma)^{2}}\lim_{k\to\infty}\int_{\mathbb{R}}v_{2\beta_k}^{4}\,dx, \tag*{(126)}
\end{aligned} a ∗ 2 = k → ∞ lim ∫ R ( v 1 β k 2 + v 2 β k 2 ) 2 d x = k → ∞ lim ( ∫ R v 1 β k 4 d x + ∫ R v 2 β k 4 d x ) 2 = γ 2 1 k → ∞ lim ∫ R v 1 β k 4 d x = ( 1 − γ ) 2 1 k → ∞ lim ∫ R v 2 β k 4 d x , (126)
where γ \gamma γ is defined by (30) . Therefore, we have
lim k → ∞ ∫ R v 1 β k 4 d x = 2 γ 2 a ∗ , lim k → ∞ ∫ R v 2 β k 4 d x = 2 ( 1 − γ ) 2 a ∗ , lim k → ∞ ∫ R v 1 β k 2 v 2 β k 2 d x = 2 γ ( 1 − γ ) a ∗ . (127) \lim_{k\to\infty}\int_{\mathbb{R}}v_{1\beta_k}^{4}\,dx=\frac{2\gamma^{2}}{a^{*}},\qquad
\lim_{k\to\infty}\int_{\mathbb{R}}v_{2\beta_k}^{4}\,dx=\frac{2(1-\gamma)^{2}}{a^{*}},\qquad
\lim_{k\to\infty}\int_{\mathbb{R}}v_{1\beta_k}^{2}v_{2\beta_k}^{2}\,dx=\frac{2\gamma(1-\gamma)}{a^{*}}. \tag*{(127)} k → ∞ lim ∫ R v 1 β k 4 d x = a ∗ 2 γ 2 , k → ∞ lim ∫ R v 2 β k 4 d x = a ∗ 2 ( 1 − γ ) 2 , k → ∞ lim ∫ R v 1 β k 2 v 2 β k 2 d x = a ∗ 2 γ ( 1 − γ ) . (127)
Note from (121) that { v i β k } \{v_{i\beta_k}\} { v i β k } is bounded in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) for i = 1 , 2 i=1,2 i = 1 , 2 . Next, we claim that there are a sequence { η β k } ⊂ R \{\eta_{\beta_k}\}\subset\mathbb{R} { η β k } ⊂ R and constants r 0 , δ > 0 r_0,\delta>0 r 0 , δ > 0 such that
lim inf k → ∞ ∫ B r 0 ( η β k ) ∣ v 1 β k ( x ) ∣ 2 d x ≥ δ > 0. (128) \liminf_{k\to\infty}\int_{B_{r_0}(\eta_{\beta_k})}|v_{1\beta_k}(x)|^{2}\,dx\geq\delta>0. \tag*{(128)} k → ∞ lim inf ∫ B r 0 ( η β k ) ∣ v 1 β k ( x ) ∣ 2 d x ≥ δ > 0. (128)
Arguing by contradiction, suppose that (128) is false. Then, for any r > 0 r>0 r > 0 , there exists a subsequence of { v 1 β k } \{v_{1\beta_k}\} { v 1 β k } , still denoted by { v 1 β k } \{v_{1\beta_k}\} { v 1 β k } , such that
lim sup k → ∞ sup y ∈ R ∫ B r ( y ) ∣ v 1 β k ( x ) ∣ 2 d x = 0 , (129) \limsup_{k\to\infty}\sup_{y\in\mathbb{R}}\int_{B_r(y)}|v_{1\beta_k}(x)|^{2}\,dx=0, \tag*{(129)} k → ∞ lim sup y ∈ R sup ∫ B r ( y ) ∣ v 1 β k ( x ) ∣ 2 d x = 0 , (129)
from which we can deduce that v 1 β k → 0 v_{1\beta_k}\to0 v 1 β k → 0 in L m ( R ) L^{m}(\mathbb{R}) L m ( R ) for all 2 < m < ∞ 2<m<\infty 2 < m < ∞ by Lions’ vanishing lemma (see [9 Lemma 2.4] ). However, this implies ∫ R v 1 β k 4 d x → 0 \int_{\mathbb{R}}v_{1\beta_k}^{4}\,dx\to0 ∫ R v 1 β k 4 d x → 0 as k → ∞ k\to\infty k → ∞ , which sharply contradicts (127) . Therefore, (128) follows. From (112) , we thus deduce
lim inf k → ∞ ∫ B r 0 ( 0 ) ∣ w 1 β k ( x ) ∣ 2 d x ≥ δ > 0. (130) \liminf_{k\to\infty}\int_{B_{r_0}(0)}|w_{1\beta_k}(x)|^{2}\,dx\geq\delta>0. \tag*{(130)} k → ∞ lim inf ∫ B r 0 ( 0 ) ∣ w 1 β k ( x ) ∣ 2 d x ≥ δ > 0. (130)
Comparing (112) with (120) , it follows from (121) –(127) that (114) –(117) hold. Furthermore, from (114) , one can assume that w i β k ⇀ w i 0 ≥ 0 w_{i\beta_k}\rightharpoonup w_{i0}\geq0 w i β k ⇀ w i 0 ≥ 0 in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) for some w i 0 ∈ H 1 2 ( R ) w_{i0}\in H^{\frac{1}{2}}(\mathbb{R}) w i 0 ∈ H 2 1 ( R ) , where i = 1 , 2 i=1,2 i = 1 , 2 . Moreover, from (130) , one can easily see that w 10 ≠ 0 w_{10}\ne0 w 10 = 0 .
To complete the proof, it remains to obtain a result similar to (130) for w 2 β k w_{2\beta_k} w 2 β k with the same sequence η β k \eta_{\beta_k} η β k . We first claim that { x β k } \{x_{\beta_k}\} { x β k } defined in (118) is bounded in R \mathbb{R} R . In fact, if it does not hold, there exists a subsequence of { x β k } \{x_{\beta_k}\} { x β k } , still denoted by { x β k } \{x_{\beta_k}\} { x β k } , such that ∣ x β k ∣ → ∞ |x_{\beta_k}|\to\infty ∣ x β k ∣ → ∞ as k → ∞ k\to\infty k → ∞ . From (98) , we infer
∑ i = 1 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ( x ) ∣ 2 d x = ∑ i = 1 2 ∫ R V i ( x ) ∣ u i β k ( x ) ∣ 2 d x → 0 as k → ∞ . (131) \sum_{i=1}^{2}\int_{\mathbb{R}}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})|w_{i\beta_k}(x)|^{2}\,dx
=\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(x)|u_{i\beta_k}(x)|^{2}\,dx\to0\quad\text{as }k\to\infty. \tag*{(131)} i = 1 ∑ 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ( x ) ∣ 2 d x = i = 1 ∑ 2 ∫ R V i ( x ) ∣ u i β k ( x ) ∣ 2 d x → 0 as k → ∞. (131)
Since V 1 ( x ) → ∞ V_1(x)\to\infty V 1 ( x ) → ∞ as ∣ x ∣ → ∞ |x|\to\infty ∣ x ∣ → ∞ , for any C > 0 C>0 C > 0 and sufficiently large k k k , we have V 1 ( ε β k x + x β k ) ≥ C V_1(\varepsilon_{\beta_k}x+x_{\beta_k})\geq C V 1 ( ε β k x + x β k ) ≥ C on B r 0 ( 0 ) B_{r_0}(0) B r 0 ( 0 ) . Combined with (130) , this yields
lim inf k → ∞ ∑ i = 1 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ( x ) ∣ 2 d x ≥ C lim inf k → ∞ ∫ B r 0 ( 0 ) ∣ w 1 β k ( x ) ∣ 2 d x ≥ C δ > 0 , \liminf_{k\to\infty}\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})|w_{i\beta_k}(x)|^{2}\,dx
\geq C\liminf_{k\to\infty}\int_{B_{r_0}(0)}|w_{1\beta_k}(x)|^{2}\,dx
\geq C\delta>0, k → ∞ lim inf i = 1 ∑ 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ( x ) ∣ 2 d x ≥ C k → ∞ lim inf ∫ B r 0 ( 0 ) ∣ w 1 β k ( x ) ∣ 2 d x ≥ C δ > 0 ,
which contradicts (131) and therefore the claim follows. Then, we can infer that, going if necessary to a subsequence, there exists some point x 0 ∈ R x_0 \in\mathbb{R} x 0 ∈ R such that
x β k → x 0 as k → ∞ . (132) x_{\beta_k} \to x_0 \quad\text{as } k \to\infty. \tag*{(132)} x β k → x 0 as k → ∞. (132)
Applying the translation and scaling properties of the fractional Laplacian ( − Δ ) 1 2 (-\Delta)^{\frac{1}{2}} ( − Δ ) 2 1 to w i β k w_{i\beta_k} w i β k , one can deduce from (112) that for i = 1 , 2 i=1,2 i = 1 , 2 ,
( − Δ ) 1 2 w i β k ( x ) = ε β k 3 2 ( − Δ ) 1 2 u i β k ( ε β k x + x β k ) . (133) (-\Delta)^{\frac{1}{2}}w_{i\beta_k}(x)=\varepsilon_{\beta_k}^{\frac{3}{2}}(-\Delta)^{\frac{1}{2}}u_{i\beta_k}(\varepsilon_{\beta_k}x+x_{\beta_k}). \tag*{(133)} ( − Δ ) 2 1 w i β k ( x ) = ε β k 2 3 ( − Δ ) 2 1 u i β k ( ε β k x + x β k ) . (133)
Moreover, from (112) , one can easily obtain
w 1 β k 3 ( x ) = ε β k 3 2 u 1 β k 3 ( ε β k x + x β k ) , (134) w_{1\beta_k}^{3}(x)=\varepsilon_{\beta_k}^{\frac{3}{2}}u_{1\beta_k}^{3}(\varepsilon_{\beta_k}x+x_{\beta_k}), \tag*{(134)} w 1 β k 3 ( x ) = ε β k 2 3 u 1 β k 3 ( ε β k x + x β k ) , (134)
w 2 β k 2 ( x ) w 1 β k ( x ) = ε β k 3 2 u 2 β k 2 ( ε β k x + x β k ) u 1 β k ( ε β k x + x β k ) , (135) w_{2\beta_k}^{2}(x)w_{1\beta_k}(x)=\varepsilon_{\beta_k}^{\frac{3}{2}}u_{2\beta_k}^{2}(\varepsilon_{\beta_k}x+x_{\beta_k})u_{1\beta_k}(\varepsilon_{\beta_k}x+x_{\beta_k}), \tag*{(135)} w 2 β k 2 ( x ) w 1 β k ( x ) = ε β k 2 3 u 2 β k 2 ( ε β k x + x β k ) u 1 β k ( ε β k x + x β k ) , (135)
w 1 β k 2 ( x ) w 2 β k ( x ) = ε β k 3 2 u 1 β k 2 ( ε β k x + x β k ) u 2 β k ( ε β k x + x β k ) . (136) w_{1\beta_k}^{2}(x)w_{2\beta_k}(x)=\varepsilon_{\beta_k}^{\frac{3}{2}}u_{1\beta_k}^{2}(\varepsilon_{\beta_k}x+x_{\beta_k})u_{2\beta_k}(\varepsilon_{\beta_k}x+x_{\beta_k}). \tag*{(136)} w 1 β k 2 ( x ) w 2 β k ( x ) = ε β k 2 3 u 1 β k 2 ( ε β k x + x β k ) u 2 β k ( ε β k x + x β k ) . (136)
Since the nonnegative minimizer ( u 1 β k , u 2 β k ) (u_{1\beta_k},u_{2\beta_k}) ( u 1 β k , u 2 β k ) satisfies system (1) with the Lagrange multiplier μ β k ∈ R \mu_{\beta_k} \in\mathbb{R} μ β k ∈ R for p = 4 p=4 p = 4 and q = 2 q=2 q = 2 , we then conclude from (112) and (133) –(136) that ( w 1 β k , w 2 β k ) (w_{1\beta_k},w_{2\beta_k}) ( w 1 β k , w 2 β k ) satisfies the following equations in R \mathbb{R} R ,
( − Δ ) 1 2 w 1 β k + ε β k V 1 ( ε β k x + x β k ) w 1 β k = ε β k μ β k w 1 β k + a 1 w 1 β k 3 + β k w 2 β k 2 w 1 β k , ( − Δ ) 1 2 w 2 β k + ε β k V 2 ( ε β k x + x β k ) w 2 β k = ε β k μ β k w 2 β k + a 2 w 2 β k 3 + β k w 1 β k 2 w 2 β k . (137) \begin{aligned}
(-\Delta)^{\frac{1}{2}}w_{1\beta_k}+\varepsilon_{\beta_k}V_1(\varepsilon_{\beta_k}x+x_{\beta_k})w_{1\beta_k}&=\varepsilon_{\beta_k}\mu_{\beta_k}w_{1\beta_k}+a_1w_{1\beta_k}^{3}+\beta_kw_{2\beta_k}^{2}w_{1\beta_k},\\
(-\Delta)^{\frac{1}{2}}w_{2\beta_k}+\varepsilon_{\beta_k}V_2(\varepsilon_{\beta_k}x+x_{\beta_k})w_{2\beta_k}&=\varepsilon_{\beta_k}\mu_{\beta_k}w_{2\beta_k}+a_2w_{2\beta_k}^{3}+\beta_kw_{1\beta_k}^{2}w_{2\beta_k}. \tag*{(137)}
\end{aligned} ( − Δ ) 2 1 w 1 β k + ε β k V 1 ( ε β k x + x β k ) w 1 β k ( − Δ ) 2 1 w 2 β k + ε β k V 2 ( ε β k x + x β k ) w 2 β k = ε β k μ β k w 1 β k + a 1 w 1 β k 3 + β k w 2 β k 2 w 1 β k , = ε β k μ β k w 2 β k + a 2 w 2 β k 3 + β k w 1 β k 2 w 2 β k . (137)
From (137) and (114) , one obtains that
ε β k μ β k = ∑ i = 1 2 ∫ R ( ∣ ( − Δ ) 1 4 w i β k ∣ 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x − ( a 1 ∫ R w 1 β k 4 d x + a 2 ∫ R w 2 β k 4 d x + 2 β k ∫ R w 1 β k 2 w 2 β k 2 d x ) . (138) \begin{aligned}
\varepsilon_{\beta_k}\mu_{\beta_k}
={}&\sum_{i=1}^{2}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{i\beta_k}\right|^{2}+\varepsilon_{\beta_k}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})w_{i\beta_k}^{2}\right)\,dx\\
&-\left(a_1\int_{\mathbb{R}}w_{1\beta_k}^{4}\,dx+a_2\int_{\mathbb{R}}w_{2\beta_k}^{4}\,dx+2\beta_k\int_{\mathbb{R}}w_{1\beta_k}^{2}w_{2\beta_k}^{2}\,dx\right). \tag*{(138)}
\end{aligned} ε β k μ β k = i = 1 ∑ 2 ∫ R ( ( − Δ ) 4 1 w i β k 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x − ( a 1 ∫ R w 1 β k 4 d x + a 2 ∫ R w 2 β k 4 d x + 2 β k ∫ R w 1 β k 2 w 2 β k 2 d x ) . (138)
On the other hand, one can deduce from (18) and (133) –(136) that
ε β k e ( a 1 , a 2 , β k ) = ε β k E a 1 , a 2 , β k ( u 1 β k , u 2 β k ) = ∑ i = 1 2 ∫ R ( ∣ ( − Δ ) 1 4 w i β k ∣ 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x − 1 2 ( a 1 ∫ R w 1 β k 4 d x + a 2 ∫ R w 2 β k 4 d x + 2 β k ∫ R w 1 β k 2 w 2 β k 2 d x ) . (139) \begin{aligned}
\varepsilon_{\beta_k}e(a_1,a_2,\beta_k)
={}&\varepsilon_{\beta_k}E_{a_1,a_2,\beta_k}(u_{1\beta_k},u_{2\beta_k})\\
={}&\sum_{i=1}^{2}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{i\beta_k}\right|^{2}+\varepsilon_{\beta_k}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})w_{i\beta_k}^{2}\right)\,dx\\
&-\frac{1}{2}\left(a_1\int_{\mathbb{R}}w_{1\beta_k}^{4}\,dx+a_2\int_{\mathbb{R}}w_{2\beta_k}^{4}\,dx+2\beta_k\int_{\mathbb{R}}w_{1\beta_k}^{2}w_{2\beta_k}^{2}\,dx\right). \tag*{(139)}
\end{aligned} ε β k e ( a 1 , a 2 , β k ) = = ε β k E a 1 , a 2 , β k ( u 1 β k , u 2 β k ) i = 1 ∑ 2 ∫ R ( ( − Δ ) 4 1 w i β k 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x − 2 1 ( a 1 ∫ R w 1 β k 4 d x + a 2 ∫ R w 2 β k 4 d x + 2 β k ∫ R w 1 β k 2 w 2 β k 2 d x ) . (139)
Combining (138) and (139) , we conclude from (111) and (114) that
ε β k μ β k = 2 ε β k e ( a 1 , a 2 , β k ) − ∑ i = 1 2 ∫ R ( ∣ ( − Δ ) 1 4 w i β k ∣ 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x → − 1 as k → ∞ . (140) \begin{aligned}
\varepsilon_{\beta_k}\mu_{\beta_k}=2\varepsilon_{\beta_k}e(a_1,a_2,\beta_k)-\sum_{i=1}^{2}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{i\beta_k}\right|^{2}+\varepsilon_{\beta_k}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})w_{i\beta_k}^{2}\right)\,dx \\
&\to-1 \quad\text{as } k \to\infty. \tag*{(140)}
\end{aligned} ε β k μ β k = 2 ε β k e ( a 1 , a 2 , β k ) − i = 1 ∑ 2 ∫ R ( ( − Δ ) 4 1 w i β k 2 + ε β k V i ( ε β k x + x β k ) w i β k 2 ) d x → − 1 as k → ∞. (140)
Since the right-hand side of (137) is bounded in L l o c 2 ( R ) L_{\mathrm{loc}}^{2}(\mathbb{R}) L loc 2 ( R ) , the sequence w i β k w_{i\beta_k} w i β k is bounded in H l o c 1 ( R ) H_{\mathrm{loc}}^{1}(\mathbb{R}) H loc 1 ( R ) ; hence, up to a subsequence, w i β k → w i 0 w_{i\beta_k}\to w_{i0} w i β k → w i 0 strongly in L l o c m ( R ) L_{\mathrm{loc}}^{m}(\mathbb{R}) L loc m ( R ) for every 2 ≤ m < ∞ 2\leq m<\infty 2 ≤ m < ∞ . Taking the weak limit of (137) , one can find that ( w 10 , w 20 ) (w_{10},w_{20}) ( w 10 , w 20 ) satisfies
( − Δ ) 1 2 w 10 + w 10 = a 1 w 10 3 + β ∗ w 20 2 w 10 , ( − Δ ) 1 2 w 20 + w 20 = a 2 w 20 3 + β ∗ w 10 2 w 20 . (141) \begin{aligned}
(-\Delta)^{\frac{1}{2}}w_{10}+w_{10}&=a_1w_{10}^{3}+\beta^{*}w_{20}^{2}w_{10},\\
(-\Delta)^{\frac{1}{2}}w_{20}+w_{20}&=a_2w_{20}^{3}+\beta^{*}w_{10}^{2}w_{20}. \tag*{(141)}
\end{aligned} ( − Δ ) 2 1 w 10 + w 10 ( − Δ ) 2 1 w 20 + w 20 = a 1 w 10 3 + β ∗ w 20 2 w 10 , = a 2 w 20 3 + β ∗ w 10 2 w 20 . (141)
We now show that there exist r 2 , δ 2 > 0 r_2,\delta_2>0 r 2 , δ 2 > 0 such that
lim inf k → ∞ ∫ B r 2 ( 0 ) ∣ w 2 β k ( x ) ∣ 2 d x ≥ δ 2 > 0. (142) \liminf_{k\to\infty}\int_{B_{r_2}(0)}|w_{2\beta_k}(x)|^2dx\ge\delta_2>0. \tag*{(142)} k → ∞ lim inf ∫ B r 2 ( 0 ) ∣ w 2 β k ( x ) ∣ 2 d x ≥ δ 2 > 0. (142)
Indeed, it suffices to prove w 20 ≠ 0 w_{20}\ne0 w 20 = 0 . Arguing by contradiction that w 20 = 0 w_{20}=0 w 20 = 0 , we then see from (141) that w 10 ≥ 0 w_{10}\ge0 w 10 ≥ 0 satisfies
( − Δ ) 1 2 w 10 + w 10 = a 1 w 10 3 in R . (143) (-\Delta)^{\frac{1}{2}}w_{10}+w_{10}=a_1w_{10}^3\quad\text{in }\mathbb{R}. \tag*{(143)} ( − Δ ) 2 1 w 10 + w 10 = a 1 w 10 3 in R . (143)
By a simple rescaling, a 1 w 10 \sqrt{a_1}\,w_{10} a 1 w 10 satisfies (22) ; together with (C.1) and (25) for p = 4 p=4 p = 4 , this yields
∥ w 10 ∥ 2 2 ≥ a ∗ a 1 > 1. (144) \|w_{10}\|_2^2\ge\frac{a^*}{a_1}>1. \tag*{(144)} ∥ w 10 ∥ 2 2 ≥ a 1 a ∗ > 1. (144)
However, it follows from (114) and weak lower semi-continuity of ∥ ⋅ ∥ 2 2 \|\cdot\|_2^2 ∥ ⋅ ∥ 2 2 that
∥ w 10 ∥ 2 2 ≤ lim inf k → ∞ ( ∥ w 1 β k ∥ 2 2 + ∥ w 2 β k ∥ 2 2 ) = 1 , (145) \|w_{10}\|_2^2\le\liminf_{k\to\infty}\left(\|w_{1\beta_k}\|_2^2+\|w_{2\beta_k}\|_2^2\right)=1, \tag*{(145)} ∥ w 10 ∥ 2 2 ≤ k → ∞ lim inf ( ∥ w 1 β k ∥ 2 2 + ∥ w 2 β k ∥ 2 2 ) = 1 , (145)
which leads to an immediate contradiction. Therefore, (142) holds. Combining (130) and (142) , we also obtain (113) , which finishes (i).
(ii) To prove (ii), from (132) , it suffices to show that x 0 ∈ V x_0\in\mathcal{V} x 0 ∈ V . Arguing by contradiction that x 0 ∉ V x_0\notin\mathcal{V} x 0 ∈ / V , without loss of generality, we suppose V 1 ( x 0 ) ≠ 0 V_1(x_0)\ne0 V 1 ( x 0 ) = 0 , which indicates V 1 ( x 0 ) > 0 V_1(x_0)>0 V 1 ( x 0 ) > 0 . We then deduce from (113) that
lim inf k → ∞ ∑ i = 1 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ∣ 2 d x ≥ V 1 ( x 0 ) lim inf k → ∞ ∫ B r 0 ( 0 ) ∣ w 1 β k ∣ 2 d x ≥ V 1 ( x 0 ) δ > 0 , (146) \liminf_{k\to\infty}\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(\varepsilon_{\beta_k}x+x_{\beta_k})|w_{i\beta_k}|^2dx\ge V_1(x_0)\liminf_{k\to\infty}\int_{B_{r_0}(0)}|w_{1\beta_k}|^2dx\ge V_1(x_0)\delta>0, \tag*{(146)} k → ∞ lim inf i = 1 ∑ 2 ∫ R V i ( ε β k x + x β k ) ∣ w i β k ∣ 2 d x ≥ V 1 ( x 0 ) k → ∞ lim inf ∫ B r 0 ( 0 ) ∣ w 1 β k ∣ 2 d x ≥ V 1 ( x 0 ) δ > 0 , (146)
which contradicts (131) so that (ii) follows.
(iii) According to Theorem 1.1 in [21 ] applied to the whole space R \mathbb{R} R , the boundary terms vanish. In our specific case N = 1 N=1 N = 1 and s = 1 2 s=\frac{1}{2} s = 2 1 , the coefficient ( 2 s − N ) (2s-N) ( 2 s − N ) vanishes identically. Thus, we can deduce that
∫ R ( w 10 2 + w 20 2 ) d x = 1 2 ∫ R ( a 1 w 10 4 + 2 β ∗ w 10 2 w 20 2 + a 2 w 20 4 ) d x . (147) \int_{\mathbb{R}}(w_{10}^2+w_{20}^2)dx=\frac{1}{2}\int_{\mathbb{R}}(a_1w_{10}^4+2\beta^*w_{10}^2w_{20}^2+a_2w_{20}^4)dx. \tag*{(147)} ∫ R ( w 10 2 + w 20 2 ) d x = 2 1 ∫ R ( a 1 w 10 4 + 2 β ∗ w 10 2 w 20 2 + a 2 w 20 4 ) d x . (147)
Linearly combining (147) with the standard Nehari identity of (141) , one directly deduces the following Pohozaev identity
∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x + 1 2 ∫ R ( w 10 2 + w 20 2 ) d x = 3 4 ∫ R ( a 1 w 10 4 + 2 β ∗ w 10 2 w 20 2 + a 2 w 20 4 ) d x . (148) \int_{\mathbb{R}}\left(|(-\Delta)^{\frac{1}{4}}w_{10}|^2+|(-\Delta)^{\frac{1}{4}}w_{20}|^2\right)dx+\frac{1}{2}\int_{\mathbb{R}}(w_{10}^2+w_{20}^2)dx
=\frac{3}{4}\int_{\mathbb{R}}(a_1w_{10}^4+2\beta^*w_{10}^2w_{20}^2+a_2w_{20}^4)dx. \tag*{(148)} ∫ R ( ∣ ( − Δ ) 4 1 w 10 ∣ 2 + ∣ ( − Δ ) 4 1 w 20 ∣ 2 ) d x + 2 1 ∫ R ( w 10 2 + w 20 2 ) d x = 4 3 ∫ R ( a 1 w 10 4 + 2 β ∗ w 10 2 w 20 2 + a 2 w 20 4 ) d x . (148)
Note from (114) that
∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x ≤ 1 and ∫ R ( w 10 2 + w 20 2 ) d x ≤ 1. (149) \int_{\mathbb{R}}\left(|(-\Delta)^{\frac{1}{4}}w_{10}|^2+|(-\Delta)^{\frac{1}{4}}w_{20}|^2\right)dx\le1
\quad\text{and}\quad
\int_{\mathbb{R}}(w_{10}^2+w_{20}^2)dx\le1. \tag*{(149)} ∫ R ( ∣ ( − Δ ) 4 1 w 10 ∣ 2 + ∣ ( − Δ ) 4 1 w 20 ∣ 2 ) d x ≤ 1 and ∫ R ( w 10 2 + w 20 2 ) d x ≤ 1. (149)
Moreover, we can infer from (141) that
∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x + ∫ R ( w 10 2 + w 20 2 ) d x = ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x . (150) \int_{\mathbb{R}}\left(|(-\Delta)^{\frac{1}{4}}w_{10}|^2+|(-\Delta)^{\frac{1}{4}}w_{20}|^2\right)dx+\int_{\mathbb{R}}(w_{10}^2+w_{20}^2)dx
=\int_{\mathbb{R}}(a_1w_{10}^4+a_2w_{20}^4+2\beta^*w_{10}^2w_{20}^2)dx. \tag*{(150)} ∫ R ( ∣ ( − Δ ) 4 1 w 10 ∣ 2 + ∣ ( − Δ ) 4 1 w 20 ∣ 2 ) d x + ∫ R ( w 10 2 + w 20 2 ) d x = ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x . (150)
Combining (3.52) with (3.54), we can infer that
2 ∫ R ( w 10 2 + w 20 2 ) d x = ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x . (151) 2\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})\,dx=\int_{\mathbb{R}}(a_{1}w_{10}^{4}+a_{2}w_{20}^{4}+2\beta^{*}w_{10}^{2}w_{20}^{2})\,dx. \tag*{(151)} 2 ∫ R ( w 10 2 + w 20 2 ) d x = ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x . (151)
Together with the refined Gagliardo–Nirenberg inequality (2.7) and Lemma 2.3 for p = 4 p=4 p = 4 and q = 2 q=2 q = 2 , we deduce from (3.53)–(3.55) that
∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x = ∫ R ( w 10 2 + w 20 2 ) d x = 1 2 ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x = a ∗ 2 ∫ R ( w 10 2 + w 20 2 ) 2 d x − 1 2 [ a ∗ − a 1 ∫ R w 10 4 d x − a ∗ − a 2 ∫ R w 20 4 d x ] 2 − ( a ∗ − a 1 ) ( a ∗ − a 2 ) [ ∫ R w 10 4 d x ∫ R w 20 4 d x − ∫ R w 10 2 w 20 2 d x ] ≤ a ∗ 2 ∫ R ( w 10 2 + w 20 2 ) 2 d x ≤ ∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x ∫ R ( w 10 2 + w 20 2 ) d x ≤ ∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x , (152) \begin{aligned}
\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{10}\right|^{2}+\left|(-\Delta)^{\frac{1}{4}}w_{20}\right|^{2}\right)\,dx
&=\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})\,dx\\
&=\frac{1}{2}\int_{\mathbb{R}}(a_{1}w_{10}^{4}+a_{2}w_{20}^{4}+2\beta^{*}w_{10}^{2}w_{20}^{2})\,dx\\
&=\frac{a^{*}}{2}\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})^{2}\,dx-\frac{1}{2}\left[\sqrt{a^{*}-a_{1}}\sqrt{\int_{\mathbb{R}}w_{10}^{4}\,dx}-\sqrt{a^{*}-a_{2}}\sqrt{\int_{\mathbb{R}}w_{20}^{4}\,dx}\right]^{2}\\
&\quad-\sqrt{(a^{*}-a_{1})(a^{*}-a_{2})}\left[\sqrt{\int_{\mathbb{R}}w_{10}^{4}\,dx}\sqrt{\int_{\mathbb{R}}w_{20}^{4}\,dx}-\int_{\mathbb{R}}w_{10}^{2}w_{20}^{2}\,dx\right]\\
&\leq\frac{a^{*}}{2}\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})^{2}\,dx\\
&\leq\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{10}\right|^{2}+\left|(-\Delta)^{\frac{1}{4}}w_{20}\right|^{2}\right)\,dx\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})\,dx\\
&\leq\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{10}\right|^{2}+\left|(-\Delta)^{\frac{1}{4}}w_{20}\right|^{2}\right)\,dx, \tag*{(152)}
\end{aligned} ∫ R ( ( − Δ ) 4 1 w 10 2 + ( − Δ ) 4 1 w 20 2 ) d x = ∫ R ( w 10 2 + w 20 2 ) d x = 2 1 ∫ R ( a 1 w 10 4 + a 2 w 20 4 + 2 β ∗ w 10 2 w 20 2 ) d x = 2 a ∗ ∫ R ( w 10 2 + w 20 2 ) 2 d x − 2 1 [ a ∗ − a 1 ∫ R w 10 4 d x − a ∗ − a 2 ∫ R w 20 4 d x ] 2 − ( a ∗ − a 1 ) ( a ∗ − a 2 ) [ ∫ R w 10 4 d x ∫ R w 20 4 d x − ∫ R w 10 2 w 20 2 d x ] ≤ 2 a ∗ ∫ R ( w 10 2 + w 20 2 ) 2 d x ≤ ∫ R ( ( − Δ ) 4 1 w 10 2 + ( − Δ ) 4 1 w 20 2 ) d x ∫ R ( w 10 2 + w 20 2 ) d x ≤ ∫ R ( ( − Δ ) 4 1 w 10 2 + ( − Δ ) 4 1 w 20 2 ) d x , (152)
which suggests that the aforementioned equalities hold. Therefore, based on Lemma 2.3 and Lemma E.1, we can infer that
( w 10 ( x ) , w 20 ( x ) ) = ( τ sin θ Q ( α x − η 0 ) , τ cos θ Q ( α x − η 0 ) ) , w 10 2 ( x ) = k w 20 2 ( x ) , (153) (w_{10}(x),w_{20}(x))=(\tau\sin\theta\,Q(\alpha x-\eta_{0}),\tau\cos\theta\,Q(\alpha x-\eta_{0})),\quad w_{10}^{2}(x)=kw_{20}^{2}(x), \tag*{(153)} ( w 10 ( x ) , w 20 ( x )) = ( τ sin θ Q ( α x − η 0 ) , τ cos θ Q ( α x − η 0 )) , w 10 2 ( x ) = k w 20 2 ( x ) , (153)
with some τ > 0 \tau>0 τ > 0 , α > 0 \alpha>0 α > 0 , θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) , η 0 ∈ R \eta_{0}\in\mathbb{R} η 0 ∈ R and k > 0 k>0 k > 0 . Actually, it follows from (3.56) that
∫ R ( ∣ ( − Δ ) 1 4 w 10 ∣ 2 + ∣ ( − Δ ) 1 4 w 20 ∣ 2 ) d x = ∫ R ( w 10 2 + w 20 2 ) d x = 1 , (154) \int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{10}\right|^{2}+\left|(-\Delta)^{\frac{1}{4}}w_{20}\right|^{2}\right)\,dx=\int_{\mathbb{R}}(w_{10}^{2}+w_{20}^{2})\,dx=1, \tag*{(154)} ∫ R ( ( − Δ ) 4 1 w 10 2 + ( − Δ ) 4 1 w 20 2 ) d x = ∫ R ( w 10 2 + w 20 2 ) d x = 1 , (154)
which implies α = 1 \alpha=1 α = 1 and τ = 1 ∥ Q ∥ 2 \tau=\frac{1}{\|Q\|_{2}} τ = ∥ Q ∥ 2 1 in (3.57). Besides, (3.56) also gives
a ∗ − a 1 ∫ R w 10 4 d x = a ∗ − a 2 ∫ R w 20 4 d x , (155) \sqrt{a^{*}-a_{1}}\sqrt{\int_{\mathbb{R}}w_{10}^{4}\,dx}=\sqrt{a^{*}-a_{2}}\sqrt{\int_{\mathbb{R}}w_{20}^{4}\,dx}, \tag*{(155)} a ∗ − a 1 ∫ R w 10 4 d x = a ∗ − a 2 ∫ R w 20 4 d x , (155)
which yields k = a ∗ − a 2 a ∗ − a 1 k=\sqrt{\frac{a^{*}-a_{2}}{a^{*}-a_{1}}} k = a ∗ − a 1 a ∗ − a 2 in (3.57) and through simple calculations we thus conclude that
w 10 ( x ) = γ ∥ Q ∥ 2 Q ( x − η 0 ) and w 20 ( x ) = 1 − γ ∥ Q ∥ 2 Q ( x − η 0 ) , (156) w_{10}(x)=\frac{\sqrt{\gamma}}{\|Q\|_{2}}Q(x-\eta_{0})\quad\text{and}\quad w_{20}(x)=\frac{\sqrt{1-\gamma}}{\|Q\|_{2}}Q(x-\eta_{0}), \tag*{(156)} w 10 ( x ) = ∥ Q ∥ 2 γ Q ( x − η 0 ) and w 20 ( x ) = ∥ Q ∥ 2 1 − γ Q ( x − η 0 ) , (156)
where γ \gamma γ is given by (1.30). Moreover, one can further derive from (3.18) and (3.58) that
lim k → ∞ ∫ R ( ∣ ( − Δ ) 1 4 w i β k ∣ 2 + w i β k 2 ) d x = ∫ R ( ∣ ( − Δ ) 1 4 w i 0 ∣ 2 + w i 0 2 ) d x , i = 1 , 2. (157) \lim_{k\to\infty}\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{i\beta_{k}}\right|^{2}+w_{i\beta_{k}}^{2}\right)\,dx=\int_{\mathbb{R}}\left(\left|(-\Delta)^{\frac{1}{4}}w_{i0}\right|^{2}+w_{i0}^{2}\right)\,dx,\quad i=1,2. \tag*{(157)} k → ∞ lim ∫ R ( ( − Δ ) 4 1 w i β k 2 + w i β k 2 ) d x = ∫ R ( ( − Δ ) 4 1 w i 0 2 + w i 0 2 ) d x , i = 1 , 2. (157)
Combining with the fact that w i β k ⇀ w i 0 w_{i\beta_k} \rightharpoonup w_{i0} w i β k ⇀ w i 0 in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) , we thus obtain that
w i β k → w i 0 strongly in H 1 2 ( R ) , i = 1 , 2. (158) w_{i\beta_k} \to w_{i0}\ \text{strongly in } H^{\frac{1}{2}}(\mathbb{R}),\quad i=1,2. \tag*{(158)} w i β k → w i 0 strongly in H 2 1 ( R ) , i = 1 , 2. (158)
By (156) , (119) holds, and the proof of the lemma is complete. □
Proof of Theorem 1.4. Notice that from (137) , one can obtain
( − Δ ) 1 2 w i β k ( x ) − C i β k ( x ) w i β k ( x ) ≤ 0 in R , i = 1 , 2 , (159) (-\Delta)^{\frac{1}{2}}w_{i\beta_k}(x)-C_{i\beta_k}(x)w_{i\beta_k}(x)\leq0\quad\text{in }\mathbb{R},\ i=1,2, \tag*{(159)} ( − Δ ) 2 1 w i β k ( x ) − C i β k ( x ) w i β k ( x ) ≤ 0 in R , i = 1 , 2 , (159)
where C 1 β k : = a 1 w 1 β k 2 + β k w 2 β k 2 C_{1\beta_k}:=a_1w_{1\beta_k}^2+\beta_kw_{2\beta_k}^2 C 1 β k := a 1 w 1 β k 2 + β k w 2 β k 2 and C 2 β k : = a 2 w 2 β k 2 + β k w 1 β k 2 C_{2\beta_k}:=a_2w_{2\beta_k}^2+\beta_kw_{1\beta_k}^2 C 2 β k := a 2 w 2 β k 2 + β k w 1 β k 2 . Indeed, the additional term ( ε β k μ β k − ε β k V i ( ε β k x + x β k ) ) w i β k (\varepsilon_{\beta_k}\mu_{\beta_k}-\varepsilon_{\beta_k}V_i(\varepsilon_{\beta_k}x+x_{\beta_k}))w_{i\beta_k} ( ε β k μ β k − ε β k V i ( ε β k x + x β k )) w i β k is nonpositive for k k k large, since ε β k μ β k → − 1 \varepsilon_{\beta_k}\mu_{\beta_k}\to-1 ε β k μ β k → − 1 and V i ≥ 0 V_i\geq0 V i ≥ 0 .
Thanks to the fractional Sobolev embedding in one dimension, H 1 2 ( R ) ↪ L r ( R ) H^{\frac{1}{2}}(\mathbb{R})\hookrightarrow L^r(\mathbb{R}) H 2 1 ( R ) ↪ L r ( R ) for all m ≥ 2 m\geq2 m ≥ 2 , which guarantees C i β k ∈ L r ( R ) C_{i\beta_k}\in L^r(\mathbb{R}) C i β k ∈ L r ( R ) for any r ≥ 1 r\geq1 r ≥ 1 . Thus, applying the non-local De Giorgi–Nash–Moser theory (see Theorem 1.1 in [14 ] ) to the globally nonnegative weak subsolution w i β k w_{i\beta_k} w i β k (i = 1 , 2 i=1,2 i = 1 , 2 ) of (159) ,
sup B 1 ( ξ 0 ) w i β k ≤ C ( ∫ B 2 ( ξ 0 ) ∣ w i β k ∣ 2 d x ) 1 2 + Tail ( w i β k ; ξ 0 , 1 ) , i = 1 , 2 , (160) \sup_{B_1(\xi_0)}w_{i\beta_k}\leq C\left(\int_{B_2(\xi_0)}|w_{i\beta_k}|^2\,dx\right)^{\frac{1}{2}}+\operatorname{Tail}(w_{i\beta_k};\xi_0,1),\quad i=1,2, \tag*{(160)} B 1 ( ξ 0 ) sup w i β k ≤ C ( ∫ B 2 ( ξ 0 ) ∣ w i β k ∣ 2 d x ) 2 1 + Tail ( w i β k ; ξ 0 , 1 ) , i = 1 , 2 , (160)
where C C C is a positive constant, ξ 0 \xi_0 ξ 0 is an arbitrary point in R \mathbb{R} R and
Tail ( w i β k ; ξ 0 , 1 ) = ∫ ∣ x − ξ 0 ∣ > 1 w i β k ( x ) ∣ x − ξ 0 ∣ 2 d x . (161) \operatorname{Tail}(w_{i\beta_k};\xi_0,1)=\int_{|x-\xi_0|>1}\frac{w_{i\beta_k}(x)}{|x-\xi_0|^2}\,dx. \tag*{(161)} Tail ( w i β k ; ξ 0 , 1 ) = ∫ ∣ x − ξ 0 ∣ > 1 ∣ x − ξ 0 ∣ 2 w i β k ( x ) d x . (161)
Moreover, from (119) we can deduce
∫ ∣ x ∣ > r ∣ w i β k ∣ 2 d x → 0 and Tail ( w i β k ; ξ 0 , r 2 ) → 0 as r → ∞ uniformly in k , i = 1 , 2. (162) \int_{|x|>r}|w_{i\beta_k}|^2\,dx\to0\quad\text{and}\quad\operatorname{Tail}\left(w_{i\beta_k};\xi_0,\frac{r}{2}\right)\to0\quad\text{as }r\to\infty\text{ uniformly in }k,\ i=1,2. \tag*{(162)} ∫ ∣ x ∣ > r ∣ w i β k ∣ 2 d x → 0 and Tail ( w i β k ; ξ 0 , 2 r ) → 0 as r → ∞ uniformly in k , i = 1 , 2. (162)
Combining (160) and (162) , we thus infer that
w i β k ( x ) → 0 as ∣ x ∣ → ∞ uniformly in k , i = 1 , 2. (163) w_{i\beta_k}(x)\to0\quad\text{as }|x|\to\infty\text{ uniformly in }k,\ i=1,2. \tag*{(163)} w i β k ( x ) → 0 as ∣ x ∣ → ∞ uniformly in k , i = 1 , 2. (163)
Consequently, each w i β k w_{i\beta_k} w i β k (i = 1 , 2 i=1,2 i = 1 , 2 ) has at least one global maximum point. Let z i β k z_{i\beta_k} z i β k be a global maximum point of u i β k u_{i\beta_k} u i β k , i = 1 , 2 i=1,2 i = 1 , 2 ; then, by (112) , w i β k w_{i\beta_k} w i β k attains its global maximum at the point z i β k − x β k ε β k \frac{z_{i\beta_k}-x_{\beta_k}}{\varepsilon_{\beta_k}} ε β k z i β k − x β k . Hence, one can derive from (113) and (163) that
lim sup k → ∞ ∣ z i β k − x β k ∣ ε β k < ∞ , i = 1 , 2. (164) \limsup_{k\to\infty}\frac{|z_{i\beta_k}-x_{\beta_k}|}{\varepsilon_{\beta_k}}<\infty,\quad i=1,2. \tag*{(164)} k → ∞ lim sup ε β k ∣ z i β k − x β k ∣ < ∞ , i = 1 , 2. (164)
Going if necessary to a subsequence, there exists η i ∈ R \eta_i\in\mathbb{R} η i ∈ R such that
lim k → ∞ z i β k − x β k ε β k = η i , i = 1 , 2. (165) \lim_{k\to\infty}\frac{z_{i\beta_k}-x_{\beta_k}}{\varepsilon_{\beta_k}}=\eta_i,\quad i=1,2. \tag*{(165)} k → ∞ lim ε β k z i β k − x β k = η i , i = 1 , 2. (165)
Define
w ‾ i β k ( x ) : = ε β k 1 2 u i β k ( ε β k x + z i β k ) = w i β k ( x + z i β k − x β k ε β k ) , i = 1 , 2. (166) \overline{w}_{i\beta_k}(x):=\varepsilon_{\beta_k}^{\frac{1}{2}}u_{i\beta_k}(\varepsilon_{\beta_k}x+z_{i\beta_k})=w_{i\beta_k}\left(x+\frac{z_{i\beta_k}-x_{\beta_k}}{\varepsilon_{\beta_k}}\right),\quad i=1,2. \tag*{(166)} w i β k ( x ) := ε β k 2 1 u i β k ( ε β k x + z i β k ) = w i β k ( x + ε β k z i β k − x β k ) , i = 1 , 2. (166)
From Lemma 3.2 (iii), we obtain
lim k → ∞ w ‾ 1 β k ( x ) = w ‾ 10 ( x ) = γ ∥ Q ∥ 2 Q ( x + η 1 − η 0 ) , lim k → ∞ w ‾ 2 β k ( x ) = w ‾ 20 ( x ) = 1 − γ ∥ Q ∥ 2 Q ( x + η 2 − η 0 ) . (167) \begin{aligned}
\lim_{k\to\infty}\overline{w}_{1\beta_k}(x)&=\overline{w}_{10}(x)=\frac{\sqrt{\gamma}}{\|Q\|_2}Q(x+\eta_1-\eta_0),\\
\lim_{k\to\infty}\overline{w}_{2\beta_k}(x)&=\overline{w}_{20}(x)=\frac{\sqrt{1-\gamma}}{\|Q\|_2}Q(x+\eta_2-\eta_0). \tag*{(167)}
\end{aligned} k → ∞ lim w 1 β k ( x ) k → ∞ lim w 2 β k ( x ) = w 10 ( x ) = ∥ Q ∥ 2 γ Q ( x + η 1 − η 0 ) , = w 20 ( x ) = ∥ Q ∥ 2 1 − γ Q ( x + η 2 − η 0 ) . (167)
strongly in H 1 2 ( R ) H^{\frac{1}{2}}(\mathbb{R}) H 2 1 ( R ) . Note from (166) that the origin is a critical point of w ˉ i β k \bar{w}_{i\beta_k} w ˉ i β k ; passing to the limit, it is also a critical point of w ˉ i 0 \bar{w}_{i0} w ˉ i 0 for i = 1 , 2 i=1,2 i = 1 , 2 . Since the ground state Q Q Q is radially symmetric and strictly decreasing in r = ∣ x ∣ r=|x| r = ∣ x ∣ , we deduce that w ˉ i 0 \bar{w}_{i0} w ˉ i 0 attains its maximum at the origin, which forces η 1 = η 2 = η 0 \eta_1=\eta_2=\eta_0 η 1 = η 2 = η 0 in (167) . Therefore, (33) and (35) immediately follow from (165) and (167) . Furthermore, (163) and Lemma 3.2 (ii) imply that (31) holds. Finally, (36) follows from (115) and (117) , which completes the proof of Theorem 1.4.
Explicit concentration behavior
In this section we study the concentration behavior of the minimizers of problem (17) for q = 2 q=2 q = 2 and p = 4 p=4 p = 4 under explicit trapping potentials, namely the polynomial potentials introduced in (37) ; some delicate estimates of the associated functionals will also be given.
Lemma 4.1. Suppose that V i ( x ) V_i(x) V i ( x ) (i = 1 , 2 i=1,2 i = 1 , 2 ) satisfy (37) and (38) . If 0 < q 0 < 1 0<q_0<1 0 < q 0 < 1 , then for any given 0 < a 1 , a 2 < a ∗ 0<a_1,a_2<a^* 0 < a 1 , a 2 < a ∗ and 0 < β < β ∗ 0<\beta<\beta^* 0 < β < β ∗ , we have
lim inf β → β ∗ e ( a 1 , a 2 , β ) ( β ∗ − β ) q 0 q 0 + 1 ≤ q 0 + 1 q 0 a ∗ ( q 0 λ 0 ) 1 q 0 + 1 [ 2 γ ( 1 − γ ) ] q 0 q 0 + 1 (168) \liminf_{\beta\to\beta^*}\frac{e(a_1,a_2,\beta)}{(\beta^*-\beta)^{\frac{q_0}{q_0+1}}}
\leq\frac{q_0+1}{q_0a^*}(q_0\lambda_0)^{\frac{1}{q_0+1}}[2\gamma(1-\gamma)]^{\frac{q_0}{q_0+1}} \tag*{(168)} β → β ∗ lim inf ( β ∗ − β ) q 0 + 1 q 0 e ( a 1 , a 2 , β ) ≤ q 0 a ∗ q 0 + 1 ( q 0 λ 0 ) q 0 + 1 1 [ 2 γ ( 1 − γ ) ] q 0 + 1 q 0 (168) where γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) , q 0 > 0 q_0>0 q 0 > 0 and λ 0 > 0 \lambda_0>0 λ 0 > 0 are given in (30) , (41) and (43) , respectively.
Proof. Let ( ϕ 1 σ , ϕ 2 σ ) (\phi_{1\sigma},\phi_{2\sigma}) ( ϕ 1 σ , ϕ 2 σ ) be the functions given by (66) for N = 1 N=1 N = 1 with
σ = [ q 0 λ 0 2 γ ( 1 − γ ) ( β ∗ − β ) ] 1 q 0 + 1 . (169) \sigma=\left[\frac{q_0\lambda_0}{2\gamma(1-\gamma)(\beta^*-\beta)}\right]^{\frac{1}{q_0+1}}. \tag*{(169)} σ = [ 2 γ ( 1 − γ ) ( β ∗ − β ) q 0 λ 0 ] q 0 + 1 1 . (169)
At the same time, set θ = γ \theta=\gamma θ = γ and take x 0 = x j 0 x_0=x_{j_0} x 0 = x j 0 , where x j 0 ∈ Z 0 x_{j_0}\in Z_0 x j 0 ∈ Z 0 with Z 0 Z_0 Z 0 given by (44) . Applying (82) , one can obtain
∑ i = 1 2 ∫ R ∣ ( − Δ ) 1 4 ϕ i σ ∣ 2 d x − 1 2 ∫ R ( a 1 ϕ 1 σ 4 + a 2 ϕ 2 σ 4 + 2 β ϕ 1 σ 2 ϕ 2 σ 2 ) d x ≤ 2 σ a ∗ γ ( 1 − γ ) ( β ∗ − β ) + O ( σ − 1 ) . (170) \begin{aligned}
\sum_{i=1}^{2}\int_{\mathbb{R}}\left|(-\Delta)^{\frac{1}{4}}\phi_{i\sigma}\right|^2dx-\frac{1}{2}\int_{\mathbb{R}}\left(a_1\phi_{1\sigma}^4+a_2\phi_{2\sigma}^4+2\beta\phi_{1\sigma}^2\phi_{2\sigma}^2\right)dx\\
\leq\frac{2\sigma}{a^*}\gamma(1-\gamma)(\beta^*-\beta)+O(\sigma^{-1}). \tag*{(170)}
\end{aligned} i = 1 ∑ 2 ∫ R ( − Δ ) 4 1 ϕ iσ 2 d x − 2 1 ∫ R ( a 1 ϕ 1 σ 4 + a 2 ϕ 2 σ 4 + 2 β ϕ 1 σ 2 ϕ 2 σ 2 ) d x ≤ a ∗ 2 σ γ ( 1 − γ ) ( β ∗ − β ) + O ( σ − 1 ) . (170)
Note that, from 0 < q 0 < 1 0<q_0<1 0 < q 0 < 1 , we can deduce that
2 σ a ∗ γ ( 1 − γ ) ( β ∗ − β ) + O ( σ − 1 ) = 2 σ a ∗ γ ( 1 − γ ) ( β ∗ − β ) + o ( σ − q 0 ) , (171) \frac{2\sigma}{a^*}\gamma(1-\gamma)(\beta^*-\beta)+O(\sigma^{-1})=\frac{2\sigma}{a^*}\gamma(1-\gamma)(\beta^*-\beta)+o(\sigma^{-q_0}), \tag*{(171)} a ∗ 2 σ γ ( 1 − γ ) ( β ∗ − β ) + O ( σ − 1 ) = a ∗ 2 σ γ ( 1 − γ ) ( β ∗ − β ) + o ( σ − q 0 ) , (171)
as σ → ∞ \sigma\to\infty σ → ∞ . Therefore, combining (170) and (171) , one can infer from (169) that
∑ i = 1 2 ∫ R ∣ ( − Δ ) 1 4 ϕ i σ ∣ 2 d x − 1 2 ∫ R ( a 1 ϕ 1 σ 4 + a 2 ϕ 2 σ 4 + 2 β ϕ 1 σ 2 ϕ 2 σ 2 ) d x ≤ 1 a ∗ ( q 0 λ 0 ) 1 q 0 + 1 [ 2 γ ( 1 − γ ) ] 1 q 0 + 1 ( β ∗ − β ) q 0 q 0 + 1 + o ( ( β ∗ − β ) q 0 q 0 + 1 ) as β → β ∗ . (172) \begin{aligned}
\sum_{i=1}^{2}\int_{\mathbb{R}}\left|(-\Delta)^{\frac{1}{4}}\phi_{i\sigma}\right|^2dx-\frac{1}{2}\int_{\mathbb{R}}\left(a_1\phi_{1\sigma}^4+a_2\phi_{2\sigma}^4+2\beta\phi_{1\sigma}^2\phi_{2\sigma}^2\right)dx\\
\leq\frac{1}{a^*}(q_0\lambda_0)^{\frac{1}{q_0+1}}[2\gamma(1-\gamma)]^{\frac{1}{q_0+1}}(\beta^*-\beta)^{\frac{q_0}{q_0+1}}+o\left((\beta^*-\beta)^{\frac{q_0}{q_0+1}}\right)\quad\text{as }\beta\to\beta^*. \tag*{(172)}
\end{aligned} i = 1 ∑ 2 ∫ R ( − Δ ) 4 1 ϕ iσ 2 d x − 2 1 ∫ R ( a 1 ϕ 1 σ 4 + a 2 ϕ 2 σ 4 + 2 β ϕ 1 σ 2 ϕ 2 σ 2 ) d x ≤ a ∗ 1 ( q 0 λ 0 ) q 0 + 1 1 [ 2 γ ( 1 − γ ) ] q 0 + 1 1 ( β ∗ − β ) q 0 + 1 q 0 + o ( ( β ∗ − β ) q 0 + 1 q 0 ) as β → β ∗ . (172)
On the other hand, without loss of generality, we suppose q 0 = q 1 j 0 < q 2 j 0 q_0=q_{1j_0}<q_{2j_0} q 0 = q 1 j 0 < q 2 j 0 (the case q 0 = q 2 j 0 < q 1 j 0 q_0=q_{2j_0}<q_{1j_0} q 0 = q 2 j 0 < q 1 j 0 is symmetric; if q 1 j 0 = q 2 j 0 = q 0 q_{1j_0}=q_{2j_0}=q_0 q 1 j 0 = q 2 j 0 = q 0 , both terms below are of order σ − q 0 \sigma^{-q_0} σ − q 0 and the second branch of (1.40) gives the same conclusion), hence we can deduce from (2.34) and (1.40)–(1.44) that
∫ R V 1 ( x ) ϕ 1 2 d x + ∫ R V 2 ( x ) ϕ 2 2 d x = σ − q 0 ∫ R A σ 2 γ V 1 ( σ − 1 x + x j 0 ) a ∗ σ q 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x + σ − q 0 ∫ R A σ 2 ( 1 − γ ) V 2 ( σ − 1 x + x j 0 ) a ∗ σ q 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x = σ − q 0 ∫ R A σ 2 γ V 1 ( σ − 1 x + x j 0 ) a ∗ ∣ σ − 1 x ∣ q 1 j 0 ∣ x ∣ q 1 j 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x + σ − q 0 ∫ R A σ 2 ( 1 − γ ) V 2 ( σ − 1 x + x j 0 ) a ∗ ∣ σ − 1 x ∣ q 2 j 0 σ q 0 − q 2 j 0 ∣ x ∣ q 2 j 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x = λ 0 a ∗ σ q 0 + o ( σ − q 0 ) as σ → ∞ . (173) \begin{aligned}
\int_{\mathbb{R}} V_1(x)\phi_1^2\,dx+\int_{\mathbb{R}}V_2(x)\phi_2^2\,dx
={}&\sigma^{-q_0}\int_{\mathbb{R}}\frac{A_\sigma^2\gamma V_1(\sigma^{-1}x+x_{j_0})}{a^*}\sigma^{q_0}\varphi^2(\sigma^{-1}x)Q^2(x)\,dx\\
&+\sigma^{-q_0}\int_{\mathbb{R}}\frac{A_\sigma^2(1-\gamma)V_2(\sigma^{-1}x+x_{j_0})}{a^*}\sigma^{q_0}\varphi^2(\sigma^{-1}x)Q^2(x)\,dx\\
={}&\sigma^{-q_0}\int_{\mathbb{R}}\frac{A_\sigma^2\gamma V_1(\sigma^{-1}x+x_{j_0})}{a^*|\sigma^{-1}x|^{q_{1j_0}}}|x|^{q_{1j_0}}\varphi^2(\sigma^{-1}x)Q^2(x)\,dx\\
&+\sigma^{-q_0}\int_{\mathbb{R}}\frac{A_\sigma^2(1-\gamma)V_2(\sigma^{-1}x+x_{j_0})}{a^*|\sigma^{-1}x|^{q_{2j_0}}}\sigma^{q_0-q_{2j_0}}|x|^{q_{2j_0}}\varphi^2(\sigma^{-1}x)Q^2(x)\,dx\\
={}&\frac{\lambda_0}{a^*\sigma^{q_0}}+o(\sigma^{-q_0})\quad\text{as }\sigma\to\infty. \tag*{(173)}
\end{aligned} ∫ R V 1 ( x ) ϕ 1 2 d x + ∫ R V 2 ( x ) ϕ 2 2 d x = = = σ − q 0 ∫ R a ∗ A σ 2 γ V 1 ( σ − 1 x + x j 0 ) σ q 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x + σ − q 0 ∫ R a ∗ A σ 2 ( 1 − γ ) V 2 ( σ − 1 x + x j 0 ) σ q 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x σ − q 0 ∫ R a ∗ ∣ σ − 1 x ∣ q 1 j 0 A σ 2 γ V 1 ( σ − 1 x + x j 0 ) ∣ x ∣ q 1 j 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x + σ − q 0 ∫ R a ∗ ∣ σ − 1 x ∣ q 2 j 0 A σ 2 ( 1 − γ ) V 2 ( σ − 1 x + x j 0 ) σ q 0 − q 2 j 0 ∣ x ∣ q 2 j 0 φ 2 ( σ − 1 x ) Q 2 ( x ) d x a ∗ σ q 0 λ 0 + o ( σ − q 0 ) as σ → ∞. (173)
Therefore, from (4.2), we thus infer
∫ R V 1 ( x ) ϕ 1 2 d x + ∫ R V 2 ( x ) ϕ 2 2 d x = 1 q 0 a ∗ ( q 0 λ 0 ) 1 q 0 + 1 [ 2 γ ( 1 − γ ) ] q 0 q 0 + 1 ( β ∗ − β ) q 0 q 0 + 1 + o ( ( β ∗ − β ) q 0 q 0 + 1 ) as β → β ∗ . (174) \begin{aligned}
\int_{\mathbb{R}}V_1(x)\phi_1^2\,dx+\int_{\mathbb{R}}V_2(x)\phi_2^2\,dx
={}&\frac{1}{q_0a^*}(q_0\lambda_0)^{\frac{1}{q_0+1}}[2\gamma(1-\gamma)]^{\frac{q_0}{q_0+1}}(\beta^*-\beta)^{\frac{q_0}{q_0+1}}\\
&+o\left((\beta^*-\beta)^{\frac{q_0}{q_0+1}}\right)\quad\text{as }\beta\to\beta^*. \tag*{(174)}
\end{aligned} ∫ R V 1 ( x ) ϕ 1 2 d x + ∫ R V 2 ( x ) ϕ 2 2 d x = q 0 a ∗ 1 ( q 0 λ 0 ) q 0 + 1 1 [ 2 γ ( 1 − γ ) ] q 0 + 1 q 0 ( β ∗ − β ) q 0 + 1 q 0 + o ( ( β ∗ − β ) q 0 + 1 q 0 ) as β → β ∗ . (174)
Combining (4.5) and (4.7), we then derive the estimate (4.1), as claimed. □
Proof of Theorem 1.6. Let { ( u 1 β k , u 2 β k ) } \{(u_{1\beta_k},u_{2\beta_k})\} {( u 1 β k , u 2 β k )} be the subsequence obtained in Theorem 1.4. Applying the refined Gagliardo–Nirenberg inequality (2.7) and Hölder’s inequality to (3.8), we then obtain for any k ∈ N k\in\mathbb{N} k ∈ N
e ( a 1 , a 2 , β k ) ≥ ( β ∗ − β k ) ∫ R u 1 β k 2 u 2 β k 2 d x + ∑ i = 1 2 ∫ R V i ( x ) u i β k 2 d x = β ∗ − β k ε β k ∫ R w ‾ 1 β k 2 w ‾ 2 β k 2 d x + ∑ i = 1 2 ∫ R V i ( ε β k x + z i β k ) w ‾ i β k 2 d x . (175) \begin{aligned}
e(a_1,a_2,\beta_k)\ge{}&(\beta^*-\beta_k)\int_{\mathbb{R}}u_{1\beta_k}^2u_{2\beta_k}^2\,dx+\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(x)u_{i\beta_k}^2\,dx\\
={}&\frac{\beta^*-\beta_k}{\varepsilon_{\beta_k}}\int_{\mathbb{R}}\overline{w}_{1\beta_k}^{\,2}\overline{w}_{2\beta_k}^{\,2}\,dx+\sum_{i=1}^{2}\int_{\mathbb{R}}V_i(\varepsilon_{\beta_k}x+z_{i\beta_k})\overline{w}_{i\beta_k}^{\,2}\,dx. \tag*{(175)}
\end{aligned} e ( a 1 , a 2 , β k ) ≥ = ( β ∗ − β k ) ∫ R u 1 β k 2 u 2 β k 2 d x + i = 1 ∑ 2 ∫ R V i ( x ) u i β k 2 d x ε β k β ∗ − β k ∫ R w 1 β k 2 w 2 β k 2 d x + i = 1 ∑ 2 ∫ R V i ( ε β k x + z i β k ) w i β k 2 d x . (175)
where the second equality comes from (3.70). Using (3.21) and (3.70), we find that
∫ R w ‾ 1 β k 2 w ‾ 2 β k 2 d x = 2 γ ( 1 − γ ) a ∗ + o ( 1 ) as k → ∞ . (176) \int_{\mathbb{R}}\overline{w}_{1\beta_k}^{\,2}\overline{w}_{2\beta_k}^{\,2}\,dx=\frac{2\gamma(1-\gamma)}{a^*}+o(1)\quad\text{as }k\to\infty. \tag*{(176)} ∫ R w 1 β k 2 w 2 β k 2 d x = a ∗ 2 γ ( 1 − γ ) + o ( 1 ) as k → ∞. (176)
Note that x 0 ∈ V x_0\in\mathcal{V} x 0 ∈ V in (1.31); hence there exists 1 ≤ j 0 ≤ b 1\le j_0\le b 1 ≤ j 0 ≤ b such that x 0 = x j 0 x_0=x_{j_0} x 0 = x j 0 . Without loss of generality, we may assume that q j 0 = q 1 j 0 < q 2 j 0 q_{j_0}=q_{1j_0}<q_{2j_0} q j 0 = q 1 j 0 < q 2 j 0 . Next, we claim that
z i β k − x j 0 ε β k is bounded as k → ∞ , (177) \frac{z_{i\beta_k}-x_{j_0}}{\varepsilon_{\beta_k}}\text{ is bounded as }k\to\infty, \tag*{(177)} ε β k z i β k − x j 0 is bounded as k → ∞ , (177)
and
q j 0 = q 0 . (178) q_{j_0}=q_0. \tag*{(178)} q j 0 = q 0 . (178)
Suppose by contradiction that either q j 0 < q 0 q_{j_0}<q_0 q j 0 < q 0 or (177) is false. Then, for any given M > 0 M>0 M > 0 , one can infer from Fatou’s Lemma that
lim inf k → ∞ ε β k − q 0 ∫ R V 1 ( ε β k x + z 1 β k ) w ‾ 1 β k 2 d x ≥ ∫ R lim inf k → ∞ ε β k q j 0 − q 0 ( V 1 ( ε β k x + z 1 β k ) ∣ ε β k x + z 1 β k − x j 0 ∣ q j 0 ∣ x + z 1 β k − x j 0 ε β k ∣ q j 0 w ‾ 1 β k 2 ) d x ≥ M . (179) \liminf_{k\to\infty}\varepsilon_{\beta_k}^{-q_0}\int_{\mathbb{R}}V_1(\varepsilon_{\beta_k}x+z_{1\beta_k})\overline{w}_{1\beta_k}^{2}\,dx
\geq\int_{\mathbb{R}}\liminf_{k\to\infty}\varepsilon_{\beta_k}^{q_{j_0}-q_0}\left(\frac{V_1(\varepsilon_{\beta_k}x+z_{1\beta_k})}{\lvert\varepsilon_{\beta_k}x+z_{1\beta_k}-x_{j_0}\rvert^{q_{j_0}}}\left|x+\frac{z_{1\beta_k}-x_{j_0}}{\varepsilon_{\beta_k}}\right|^{q_{j_0}}\overline{w}_{1\beta_k}^{2}\right)\,dx\geq M. \tag*{(179)} k → ∞ lim inf ε β k − q 0 ∫ R V 1 ( ε β k x + z 1 β k ) w 1 β k 2 d x ≥ ∫ R k → ∞ lim inf ε β k q j 0 − q 0 ( ∣ ε β k x + z 1 β k − x j 0 ∣ q j 0 V 1 ( ε β k x + z 1 β k ) x + ε β k z 1 β k − x j 0 q j 0 w 1 β k 2 ) d x ≥ M . (179)
From (175) , (176) and (179) , we thus deduce that
e ( a 1 , a 2 , β k ) ≥ 2 γ ( 1 − γ ) a ∗ ( β ∗ − β k ) ε β k − 1 + M ε β k q 0 ≥ C M 1 q 0 + 1 ( β ∗ − β k ) q 0 q 0 + 1 (180) e(a_1,a_2,\beta_k)\geq\frac{2\gamma(1-\gamma)}{a^*}(\beta^*-\beta_k)\varepsilon_{\beta_k}^{-1}+M\varepsilon_{\beta_k}^{q_0}\geq CM^{\frac{1}{q_0+1}}(\beta^*-\beta_k)^{\frac{q_0}{q_0+1}} \tag*{(180)} e ( a 1 , a 2 , β k ) ≥ a ∗ 2 γ ( 1 − γ ) ( β ∗ − β k ) ε β k − 1 + M ε β k q 0 ≥ C M q 0 + 1 1 ( β ∗ − β k ) q 0 + 1 q 0 (180)
for large k k k .
In fact, the second inequality above holds due to Young’s inequality, where C C C is a suitable positive constant. Since M M M is arbitrary, the claim follows from the contradiction between (168) and (180) . Following (35) and (177) , after passing to a subsequence, there exists z 0 ∈ R z_0\in\mathbb{R} z 0 ∈ R such that
lim k → ∞ z i β k − x j 0 ε β k = z 0 for i = 1 , 2. (181) \lim_{k\to\infty}\frac{z_{i\beta_k}-x_{j_0}}{\varepsilon_{\beta_k}}=z_0\quad\text{for }i=1,2. \tag*{(181)} k → ∞ lim ε β k z i β k − x j 0 = z 0 for i = 1 , 2. (181)
Combining this with (167) , we have
lim k → ∞ ε β k − q 0 ( ∫ R V 1 ( ε β k x + z 1 β k ) w ‾ 1 β k 2 d x + ∫ R V 2 ( ε β k x + z 2 β k ) w ‾ 2 β k 2 d x ) = lim k → ∞ ∫ R V 1 ( ε β k x + z 1 β k ) ∣ ε β k x + z 1 β k − x j 0 ∣ q 1 j 0 ∣ x + z 1 β k − x j 0 ε β k ∣ q 1 j 0 w ‾ 1 β k 2 d x + lim k → ∞ ε β k q 2 j 0 − q 0 ∫ R V 2 ( ε β k x + z 2 β k ) ∣ ε β k x + z 2 β k − x j 0 ∣ q 2 j 0 ∣ x + z 2 β k − x j 0 ε β k ∣ q 2 j 0 w ‾ 2 β k 2 d x = lim x → x j 0 γ V 1 ( x ) a ∗ ∣ x − x j 0 ∣ q 1 j 0 ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y . (182) \begin{aligned}
\lim_{k\to\infty}\varepsilon_{\beta_k}^{-q_0}\left(\int_{\mathbb{R}}V_1(\varepsilon_{\beta_k}x+z_{1\beta_k})\overline{w}_{1\beta_k}^{2}\,dx+\int_{\mathbb{R}}V_2(\varepsilon_{\beta_k}x+z_{2\beta_k})\overline{w}_{2\beta_k}^{2}\,dx\right)
={}&\lim_{k\to\infty}\int_{\mathbb{R}}\frac{V_1(\varepsilon_{\beta_k}x+z_{1\beta_k})}{\lvert\varepsilon_{\beta_k}x+z_{1\beta_k}-x_{j_0}\rvert^{q_{1j_0}}}\left|x+\frac{z_{1\beta_k}-x_{j_0}}{\varepsilon_{\beta_k}}\right|^{q_{1j_0}}\overline{w}_{1\beta_k}^{2}\,dx\\
&+\lim_{k\to\infty}\varepsilon_{\beta_k}^{q_{2j_0}-q_0}\int_{\mathbb{R}}\frac{V_2(\varepsilon_{\beta_k}x+z_{2\beta_k})}{\lvert\varepsilon_{\beta_k}x+z_{2\beta_k}-x_{j_0}\rvert^{q_{2j_0}}}\left|x+\frac{z_{2\beta_k}-x_{j_0}}{\varepsilon_{\beta_k}}\right|^{q_{2j_0}}\overline{w}_{2\beta_k}^{2}\,dx\\
={}&\lim_{x\to x_{j_0}}\frac{\gamma V_1(x)}{a^*\lvert x-x_{j_0}\rvert^{q_{1j_0}}}\int_{\mathbb{R}}\lvert y+z_0\rvert^{q_{1j_0}}Q^2(y)\,dy. \tag*{(182)}
\end{aligned} k → ∞ lim ε β k − q 0 ( ∫ R V 1 ( ε β k x + z 1 β k ) w 1 β k 2 d x + ∫ R V 2 ( ε β k x + z 2 β k ) w 2 β k 2 d x ) = = k → ∞ lim ∫ R ∣ ε β k x + z 1 β k − x j 0 ∣ q 1 j 0 V 1 ( ε β k x + z 1 β k ) x + ε β k z 1 β k − x j 0 q 1 j 0 w 1 β k 2 d x + k → ∞ lim ε β k q 2 j 0 − q 0 ∫ R ∣ ε β k x + z 2 β k − x j 0 ∣ q 2 j 0 V 2 ( ε β k x + z 2 β k ) x + ε β k z 2 β k − x j 0 q 2 j 0 w 2 β k 2 d x x → x j 0 lim a ∗ ∣ x − x j 0 ∣ q 1 j 0 γ V 1 ( x ) ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y . (182)
Note that
∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y ≥ ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y , (183) \int_{\mathbb{R}}\lvert y+z_0\rvert^{q_{1j_0}}Q^2(y)\,dy\geq\int_{\mathbb{R}}\lvert y\rvert^{q_{1j_0}}Q^2(y)\,dy, \tag*{(183)} ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y ≥ ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y , (183)
where the equality holds if and only if z 0 = 0 z_0=0 z 0 = 0 . Indeed, since Q Q Q is radially symmetric and decreasing, we have
∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y = ∫ R ∣ y − z 0 ∣ q 1 j 0 Q 2 ( y ) d y , (184) \int_{\mathbb{R}}\lvert y+z_0\rvert^{q_{1j_0}}Q^2(y)\,dy=\int_{\mathbb{R}}\lvert y-z_0\rvert^{q_{1j_0}}Q^2(y)\,dy, \tag*{(184)} ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y = ∫ R ∣ y − z 0 ∣ q 1 j 0 Q 2 ( y ) d y , (184)
then, we obtain
∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y − ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y = 1 2 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y + 1 2 ∫ R ( ∣ y − z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y = 1 2 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y + 1 2 ∫ R ( ∣ y ∣ q 1 j 0 − ∣ y + z 0 ∣ q 1 j 0 ) Q 2 ( y + z 0 ) d y = 1 2 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) ( Q 2 ( y ) − Q 2 ( y + z 0 ) ) d y ≥ 0. (185) \begin{aligned}
\int_{\mathbb{R}}\lvert y+z_0\rvert^{q_{1j_0}}Q^2(y)\,dy-\int_{\mathbb{R}}\lvert y\rvert^{q_{1j_0}}Q^2(y)\,dy
={}&\frac{1}{2}\int_{\mathbb{R}}\left(\lvert y+z_0\rvert^{q_{1j_0}}-\lvert y\rvert^{q_{1j_0}}\right)Q^2(y)\,dy\\
&+\frac{1}{2}\int_{\mathbb{R}}\left(\lvert y-z_0\rvert^{q_{1j_0}}-\lvert y\rvert^{q_{1j_0}}\right)Q^2(y)\,dy\\
={}&\frac{1}{2}\int_{\mathbb{R}}\left(\lvert y+z_0\rvert^{q_{1j_0}}-\lvert y\rvert^{q_{1j_0}}\right)Q^2(y)\,dy\\
&+\frac{1}{2}\int_{\mathbb{R}}\left(\lvert y\rvert^{q_{1j_0}}-\lvert y+z_0\rvert^{q_{1j_0}}\right)Q^2(y+z_0)\,dy\\
={}&\frac{1}{2}\int_{\mathbb{R}}\left(\lvert y+z_0\rvert^{q_{1j_0}}-\lvert y\rvert^{q_{1j_0}}\right)\left(Q^2(y)-Q^2(y+z_0)\right)\,dy\geq0. \tag*{(185)}
\end{aligned} ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y − ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y = = = 2 1 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y + 2 1 ∫ R ( ∣ y − z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y 2 1 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) Q 2 ( y ) d y + 2 1 ∫ R ( ∣ y ∣ q 1 j 0 − ∣ y + z 0 ∣ q 1 j 0 ) Q 2 ( y + z 0 ) d y 2 1 ∫ R ( ∣ y + z 0 ∣ q 1 j 0 − ∣ y ∣ q 1 j 0 ) ( Q 2 ( y ) − Q 2 ( y + z 0 ) ) d y ≥ 0. (185)
which implies that (183) holds. Therefore, from the definition of λ j 0 \lambda_{j_0} λ j 0 and λ 0 \lambda_0 λ 0 , one can find that
lim x → x j 0 γ V 1 ( x ) a ∗ ∣ x − x j 0 ∣ q 1 j 0 ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y ≥ lim x → x j 0 γ V 1 ( x ) a ∗ ∣ x − x j 0 ∣ q 1 j 0 ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y = λ j 0 a ∗ ≥ λ 0 a ∗ , (186) \begin{aligned}
\lim_{x\to x_{j_0}}\frac{\gamma V_1(x)}{a^*|x-x_{j_0}|^{q_{1j_0}}}\int_{\mathbb{R}}|y+z_0|^{q_{1j_0}}Q^2(y)\,dy
&\geq\lim_{x\to x_{j_0}}\frac{\gamma V_1(x)}{a^*|x-x_{j_0}|^{q_{1j_0}}}\int_{\mathbb{R}}|y|^{q_{1j_0}}Q^2(y)\,dy \\
&=\frac{\lambda_{j_0}}{a^*}\geq\frac{\lambda_0}{a^*}, \tag*{(186)}
\end{aligned} x → x j 0 lim a ∗ ∣ x − x j 0 ∣ q 1 j 0 γ V 1 ( x ) ∫ R ∣ y + z 0 ∣ q 1 j 0 Q 2 ( y ) d y ≥ x → x j 0 lim a ∗ ∣ x − x j 0 ∣ q 1 j 0 γ V 1 ( x ) ∫ R ∣ y ∣ q 1 j 0 Q 2 ( y ) d y = a ∗ λ j 0 ≥ a ∗ λ 0 , (186)
where the equalities hold if and only if z 0 = 0 z_0=0 z 0 = 0 and x j 0 ∈ Z 0 x_{j_0}\in Z_0 x j 0 ∈ Z 0 . Applying (175) , (176) , (182) and (186) , we further obtain
e ( a 1 , a 2 , β k ) ≥ ( 2 γ ( 1 − γ ) a ∗ + o ( 1 ) ) ( β ∗ − β k ) ε β k − 1 + ( λ 0 a ∗ + o ( 1 ) ) ε β k q 0 as k → ∞ . (187) e(a_1,a_2,\beta_k)\geq\left(\frac{2\gamma(1-\gamma)}{a^*}+o(1)\right)(\beta^*-\beta_k)\varepsilon_{\beta_k}^{-1}+\left(\frac{\lambda_0}{a^*}+o(1)\right)\varepsilon_{\beta_k}^{q_0}
\quad\text{as }k\to\infty. \tag*{(187)} e ( a 1 , a 2 , β k ) ≥ ( a ∗ 2 γ ( 1 − γ ) + o ( 1 ) ) ( β ∗ − β k ) ε β k − 1 + ( a ∗ λ 0 + o ( 1 ) ) ε β k q 0 as k → ∞. (187)
Therefore, one can verify by taking the infimum of the above inequality that
lim inf k → ∞ e ( a 1 , a 2 , β k ) ( β ∗ − β k ) q 0 q 0 + 1 ≥ q 0 + 1 q 0 a ∗ ( q 0 λ 0 ) 1 q 0 + 1 [ 2 γ ( 1 − γ ) ] q 0 q 0 + 1 , (188) \liminf_{k\to\infty}\frac{e(a_1,a_2,\beta_k)}{(\beta^*-\beta_k)^{\frac{q_0}{q_0+1}}}\geq\frac{q_0+1}{q_0a^*}(q_0\lambda_0)^{\frac{1}{q_0+1}}[2\gamma(1-\gamma)]^{\frac{q_0}{q_0+1}}, \tag*{(188)} k → ∞ lim inf ( β ∗ − β k ) q 0 + 1 q 0 e ( a 1 , a 2 , β k ) ≥ q 0 a ∗ q 0 + 1 ( q 0 λ 0 ) q 0 + 1 1 [ 2 γ ( 1 − γ ) ] q 0 + 1 q 0 , (188)
where the equality holds if and only if z 0 = 0 z_0=0 z 0 = 0 , x j 0 ∈ Z 0 x_{j_0}\in Z_0 x j 0 ∈ Z 0 and
lim k → ∞ ε β k ε β k = 1 (189) \lim_{k\to\infty}\frac{\varepsilon_{\beta_k}}{\varepsilon_{\beta_k}}=1 \tag*{(189)} k → ∞ lim ε β k ε β k = 1 (189)
with ε β k > 0 \varepsilon_{\beta_k}>0 ε β k > 0 given by (46) .
Note from (168) and (188) that the equality in (188) holds, hence z 0 = 0 z_0=0 z 0 = 0 , x j 0 ∈ Z 0 x_{j_0}\in Z_0 x j 0 ∈ Z 0 and (189) holds. Consequently, (47) and (48) follow from (33) , (34) and (189) . Moreover, using the fact that z 0 = 0 z_0=0 z 0 = 0 , x j 0 ∈ Z 0 x_{j_0}\in Z_0 x j 0 ∈ Z 0 , (45) follows from (181) and (189) . This completes the proof of Theorem 1.6.
Equivalence between ground states and constraint minimizers
In this appendix, we establish the equivalence between ground states of the coupled fractional Schrödinger system (1) and the constraint minimizers of the variational problem (17) .
For any given ( a 1 , a 2 , β ) ∈ R + × R + × R + (a_1,a_2,\beta)\in\mathbb{R}^+\times\mathbb{R}^+\times\mathbb{R}^+ ( a 1 , a 2 , β ) ∈ R + × R + × R + , the action functional of system (1) associated with μ ∈ R \mu\in\mathbb{R} μ ∈ R is defined by
I μ ( u 1 , u 2 ) : = 1 2 ∑ i = 1 2 ( ∥ ( − Δ ) s 2 u i ∥ 2 2 + ∫ R N ( V i ( x ) − μ ) u i 2 d x ) − 1 p ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x = 1 2 E a 1 , a 2 , β ( u 1 , u 2 ) − μ 2 ∫ R N ( u 1 2 + u 2 2 ) d x , ( u 1 , u 2 ) ∈ H 1 × H 2 , (190) \begin{aligned}
I_\mu(u_1,u_2)&:=\frac{1}{2}\sum_{i=1}^{2}\left(\left\|(-\Delta)^{\frac{s}{2}}u_i\right\|_2^2+\int_{\mathbb{R}^N}(V_i(x)-\mu)u_i^2\,dx\right)\\
&\quad-\frac{1}{p}\int_{\mathbb{R}^N}\left(a_1|u_1|^p+a_2|u_2|^p+2\beta|u_1|^q|u_2|^q\right)\,dx\\
&=\frac{1}{2}E_{a_1,a_2,\beta}(u_1,u_2)-\frac{\mu}{2}\int_{\mathbb{R}^N}(u_1^2+u_2^2)\,dx,\quad(u_1,u_2)\in\mathcal{H}_1\times\mathcal{H}_2, \tag*{(190)}
\end{aligned} I μ ( u 1 , u 2 ) := 2 1 i = 1 ∑ 2 ( ( − Δ ) 2 s u i 2 2 + ∫ R N ( V i ( x ) − μ ) u i 2 d x ) − p 1 ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x = 2 1 E a 1 , a 2 , β ( u 1 , u 2 ) − 2 μ ∫ R N ( u 1 2 + u 2 2 ) d x , ( u 1 , u 2 ) ∈ H 1 × H 2 , (190)
where p = 2 + 4 s N p=2+\frac{4s}{N} p = 2 + N 4 s , q = p 2 = 1 + 2 s N q=\frac{p}{2}=1+\frac{2s}{N} q = 2 p = 1 + N 2 s , and E a 1 , a 2 , β ( u 1 , u 2 ) E_{a_1,a_2,\beta}(u_1,u_2) E a 1 , a 2 , β ( u 1 , u 2 ) is given by (18) .
The set of all nontrivial weak solutions for system (1) is given by
S μ : = { ( u 1 , u 2 ) ∈ ( H 1 × H 2 ) ∖ { ( 0 , 0 ) } : ⟨ I μ ′ ( u 1 , u 2 ) , ( φ 1 , φ 2 ) ⟩ = 0 , ∀ ( φ 1 , φ 2 ) ∈ H 1 × H 2 } . (191) S_\mu:=\left\{(u_1,u_2)\in(\mathcal{H}_1\times\mathcal{H}_2)\setminus\{(0,0)\}:\left\langle I_\mu'(u_1,u_2),(\varphi_1,\varphi_2)\right\rangle=0,\ \forall(\varphi_1,\varphi_2)\in\mathcal{H}_1\times\mathcal{H}_2\right\}. \tag*{(191)} S μ := { ( u 1 , u 2 ) ∈ ( H 1 × H 2 ) ∖ {( 0 , 0 )} : ⟨ I μ ′ ( u 1 , u 2 ) , ( φ 1 , φ 2 ) ⟩ = 0 , ∀ ( φ 1 , φ 2 ) ∈ H 1 × H 2 } . (191)
and the set of all ground states for system (1) is defined as
G μ : = { ( u 1 , u 2 ) ∈ S μ : I μ ( u 1 , u 2 ) ≤ I μ ( u ˉ 1 , u ˉ 2 ) for all ( u ˉ 1 , u ˉ 2 ) ∈ S μ } . (192) G_\mu:=\{(u_1,u_2)\in S_\mu:I_\mu(u_1,u_2)\leq I_\mu(\bar u_1,\bar u_2)\text{ for all }(\bar u_1,\bar u_2)\in S_\mu\}. \tag*{(192)} G μ := {( u 1 , u 2 ) ∈ S μ : I μ ( u 1 , u 2 ) ≤ I μ ( u ˉ 1 , u ˉ 2 ) for all ( u ˉ 1 , u ˉ 2 ) ∈ S μ } . (192)
Proposition A.1. Suppose ( a 1 , a 2 , β ) ∈ R + × R + × R + (a_1,a_2,\beta) \in\mathbb{R}^{+} \times\mathbb{R}^{+} \times\mathbb{R}^{+} ( a 1 , a 2 , β ) ∈ R + × R + × R + is given, and let ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) be a minimizer of problem (1.17) with Lagrange multiplier μ ∈ R \mu\in\mathbb{R} μ ∈ R . Then every ground state in G μ \mathcal{G}_{\mu} G μ is a minimizer of problem (1.17); conversely, ( u 1 β , u 2 β ) ∈ G μ (u_{1\beta},u_{2\beta}) \in\mathcal{G}_{\mu} ( u 1 β , u 2 β ) ∈ G μ .
Proof. For any given ( a 1 , a 2 , β ) ∈ R + × R + × R + (a_1,a_2,\beta) \in\mathbb{R}^{+} \times\mathbb{R}^{+} \times\mathbb{R}^{+} ( a 1 , a 2 , β ) ∈ R + × R + × R + , assume that ( u 1 β , u 2 β ) (u_{1\beta},u_{2\beta}) ( u 1 β , u 2 β ) is a minimizer of (1.17). By the Lagrange multiplier theorem, there exists μ ∈ R \mu\in\mathbb{R} μ ∈ R such that ( u 1 β , u 2 β ) ∈ S μ (u_{1\beta},u_{2\beta}) \in S_{\mu} ( u 1 β , u 2 β ) ∈ S μ ; fix this μ \mu μ . Let ( u 1 , u 2 ) (u_1,u_2) ( u 1 , u 2 ) be an arbitrary ground state in G μ \mathcal{G}_{\mu} G μ . Set
u ~ i ( x ) : = u i ( x ) ρ , where ρ : = ∫ R N ( u 1 2 + u 2 2 ) d x > 0 , i = 1 , 2 , (193) \widetilde{u}_i(x):=\frac{u_i(x)}{\sqrt{\rho}}, \quad\text{where } \rho:=\int_{\mathbb{R}^{N}}(u_1^2+u_2^2)\,dx>0,\quad i=1,2, \tag*{(193)} u i ( x ) := ρ u i ( x ) , where ρ := ∫ R N ( u 1 2 + u 2 2 ) d x > 0 , i = 1 , 2 , (193)
so that ∫ R N ( u ~ 1 2 + u ~ 2 2 ) d x = 1 \int_{\mathbb{R}^{N}}(\widetilde{u}_1^2+\widetilde{u}_2^2)\,dx=1 ∫ R N ( u 1 2 + u 2 2 ) d x = 1 , which implies ( u ~ 1 , u ~ 2 ) ∈ M (\widetilde{u}_1,\widetilde{u}_2)\in\mathcal{M} ( u 1 , u 2 ) ∈ M . By (1.17) and (A.3), we obtain
E a 1 , a 2 , β ( u ~ 1 , u ~ 2 ) ≥ E a 1 , a 2 , β ( u 1 β , u 2 β ) and I μ ( u 1 β , u 2 β ) ≥ I μ ( u 1 , u 2 ) . (194) E_{a_1,a_2,\beta}(\widetilde{u}_1,\widetilde{u}_2)\ge E_{a_1,a_2,\beta}(u_{1\beta},u_{2\beta}) \quad\text{and}\quad I_{\mu}(u_{1\beta},u_{2\beta})\ge I_{\mu}(u_1,u_2). \tag*{(194)} E a 1 , a 2 , β ( u 1 , u 2 ) ≥ E a 1 , a 2 , β ( u 1 β , u 2 β ) and I μ ( u 1 β , u 2 β ) ≥ I μ ( u 1 , u 2 ) . (194)
Combining (A.1), (A.4) and (A.5) yields
I μ ( u ~ 1 , u ~ 2 ) ≥ I μ ( u 1 β , u 2 β ) ≥ I μ ( u 1 , u 2 ) . (195) I_{\mu}(\widetilde{u}_1,\widetilde{u}_2)\ge I_{\mu}(u_{1\beta},u_{2\beta})\ge I_{\mu}(u_1,u_2). \tag*{(195)} I μ ( u 1 , u 2 ) ≥ I μ ( u 1 β , u 2 β ) ≥ I μ ( u 1 , u 2 ) . (195)
On the other hand, one can infer from (1.1) and (A.1) that
I μ ( u 1 , u 2 ) = ( 1 2 − 1 p ) ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x . (196) I_{\mu}(u_1,u_2)=\left(\frac{1}{2}-\frac{1}{p}\right)\int_{\mathbb{R}^{N}}\left(a_1|u_1|^p+a_2|u_2|^p+2\beta|u_1|^q|u_2|^q\right)\,dx. \tag*{(196)} I μ ( u 1 , u 2 ) = ( 2 1 − p 1 ) ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x . (196)
Combining (1.1) and (A.4), we also have
I μ ( u ~ 1 , u ~ 2 ) = ( 1 2 ρ − 1 p ρ p / 2 ) ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x . (197) I_{\mu}(\widetilde{u}_1,\widetilde{u}_2)=\left(\frac{1}{2\rho}-\frac{1}{p\rho^{p/2}}\right)\int_{\mathbb{R}^{N}}\left(a_1|u_1|^p+a_2|u_2|^p+2\beta|u_1|^q|u_2|^q\right)\,dx. \tag*{(197)} I μ ( u 1 , u 2 ) = ( 2 ρ 1 − p ρ p /2 1 ) ∫ R N ( a 1 ∣ u 1 ∣ p + a 2 ∣ u 2 ∣ p + 2 β ∣ u 1 ∣ q ∣ u 2 ∣ q ) d x . (197)
Consider the scalar function h ( ρ ) : = 1 2 ρ − 1 p ρ p / 2 h(\rho):=\frac{1}{2\rho}-\frac{1}{p\rho^{p/2}} h ( ρ ) := 2 ρ 1 − p ρ p /2 1 for ρ > 0 \rho>0 ρ > 0 . Differentiating h ( ρ ) h(\rho) h ( ρ ) yields
h ′ ( ρ ) = − 1 2 ρ 2 + 1 2 ρ p / 2 + 1 = 1 2 ρ 2 ( ρ 1 − p / 2 − 1 ) . h'(\rho)=-\frac{1}{2\rho^2}+\frac{1}{2\rho^{p/2+1}}=\frac{1}{2\rho^2}\left(\rho^{1-p/2}-1\right). h ′ ( ρ ) = − 2 ρ 2 1 + 2 ρ p /2 + 1 1 = 2 ρ 2 1 ( ρ 1 − p /2 − 1 ) .
Since p = 2 + 4 s N > 2 p=2+\frac{4s}{N}>2 p = 2 + N 4 s > 2 , the exponent 1 − p / 2 = − 2 s N < 0 1-p/2=-\frac{2s}{N}<0 1 − p /2 = − N 2 s < 0 . Thus h ′ ( ρ ) > 0 h'(\rho)>0 h ′ ( ρ ) > 0 for ρ ∈ ( 0 , 1 ) \rho\in(0,1) ρ ∈ ( 0 , 1 ) and h ′ ( ρ ) < 0 h'(\rho)<0 h ′ ( ρ ) < 0 for ρ ∈ ( 1 , ∞ ) \rho\in(1,\infty) ρ ∈ ( 1 , ∞ ) . Therefore, h ( ρ ) h(\rho) h ( ρ ) attains its unique global maximum on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) strictly at ρ = 1 \rho=1 ρ = 1 , giving
h ( ρ ) ≤ h ( 1 ) = 1 2 − 1 p , (198) h(\rho)\le h(1)=\frac{1}{2}-\frac{1}{p}, \tag*{(198)} h ( ρ ) ≤ h ( 1 ) = 2 1 − p 1 , (198)
with equality holding if and only if ρ = 1 \rho=1 ρ = 1 . Together with (A.7), (A.8) and (A.9), we deduce that
I μ ( u ~ 1 , u ~ 2 ) ≤ I μ ( u 1 , u 2 ) . (199) I_{\mu}(\widetilde{u}_1,\widetilde{u}_2)\le I_{\mu}(u_1,u_2). \tag*{(199)} I μ ( u 1 , u 2 ) ≤ I μ ( u 1 , u 2 ) . (199)
One can easily see from (A.6) and (A.10) that I μ ( u ~ 1 , u ~ 2 ) = I μ ( u 1 , u 2 ) I_{\mu}(\widetilde{u}_1,\widetilde{u}_2)=I_{\mu}(u_1,u_2) I μ ( u 1 , u 2 ) = I μ ( u 1 , u 2 ) , and thus ρ = 1 \rho=1 ρ = 1 . This implies that all inequalities in (A.5) and (A.6) are equalities, i.e.,
I μ ( u 1 , u 2 ) = I μ ( u 1 β , u 2 β ) and E a 1 , a 2 , β ( u 1 , u 2 ) = E a 1 , a 2 , β ( u 1 β , u 2 β ) . (200) I_{\mu}(u_1,u_2)=I_{\mu}(u_{1\beta},u_{2\beta}) \quad\text{and}\quad E_{a_1,a_2,\beta}(u_1,u_2)=E_{a_1,a_2,\beta}(u_{1\beta},u_{2\beta}). \tag*{(200)} I μ ( u 1 , u 2 ) = I μ ( u 1 β , u 2 β ) and E a 1 , a 2 , β ( u 1 , u 2 ) = E a 1 , a 2 , β ( u 1 β , u 2 β ) . (200)
Consequently, ( u 1 , u 2 ) ∈ M (u_1,u_2)\in\mathcal{M} ( u 1 , u 2 ) ∈ M and, by (A.11),
E a 1 , a 2 , β ( u 1 , u 2 ) = E a 1 , a 2 , β ( u 1 β , u 2 β ) = e ( a 1 , a 2 , β ) , E_{a_1,a_2,\beta}(u_1,u_2)=E_{a_1,a_2,\beta}(u_{1\beta},u_{2\beta})=e(a_1,a_2,\beta), E a 1 , a 2 , β ( u 1 , u 2 ) = E a 1 , a 2 , β ( u 1 β , u 2 β ) = e ( a 1 , a 2 , β ) ,
that is, ( u 1 , u 2 ) (u_1,u_2) ( u 1 , u 2 ) is a constraint minimizer of problem (1.17). Since ( u 1 , u 2 ) ∈ G μ (u_1,u_2)\in\mathcal{G}_{\mu} ( u 1 , u 2 ) ∈ G μ was arbitrary, every ground state in G μ \mathcal{G}_{\mu} G μ is a minimizer of (1.17). It remains to prove the converse: ( u 1 β , u 2 β ) ∈ G μ (u_{1\beta},u_{2\beta})\in\mathcal{G}_{\mu} ( u 1 β , u 2 β ) ∈ G μ . For ( v 1 , v 2 ) ∈ S μ (v_1,v_2)\in S_{\mu} ( v 1 , v 2 ) ∈ S μ , set ρ v : = ∫ R N ( v 1 2 + v 2 2 ) d x \rho_v:=\int_{\mathbb{R}^{N}}(v_1^2+v_2^2)\,dx ρ v := ∫ R N ( v 1 2 + v 2 2 ) d x and v ~ i : = v i / ρ v \widetilde{v}_i:=v_i/\sqrt{\rho_v} v i := v i / ρ v . Since ( v ~ 1 , v ~ 2 ) ∈ M (\widetilde{v}_1,\widetilde{v}_2)\in\mathcal{M} ( v 1 , v 2 ) ∈ M , (1.17) and (A.1) yield I μ ( v ~ 1 , v ~ 2 ) ≥ I μ ( u 1 β , u 2 β ) I_{\mu}(\widetilde{v}_1,\widetilde{v}_2)\ge I_{\mu}(u_{1\beta},u_{2\beta}) I μ ( v 1 , v 2 ) ≥ I μ ( u 1 β , u 2 β ) ; on the other hand, (A.7)–(A.9) with ( v , ρ v ) (v,\rho_v) ( v , ρ v ) in place of ( u , ρ ) (u,\rho) ( u , ρ ) yield I μ ( v ~ 1 , v ~ 2 ) ≤ I μ ( v 1 , v 2 ) I_{\mu}(\widetilde{v}_1,\widetilde{v}_2)\le I_{\mu}(v_1,v_2) I μ ( v 1 , v 2 ) ≤ I μ ( v 1 , v 2 ) . Hence I μ ( u 1 β , u 2 β ) ≤ I μ ( v 1 , v 2 ) I_{\mu}(u_{1\beta},u_{2\beta})\le I_{\mu}(v_1,v_2) I μ ( u 1 β , u 2 β ) ≤ I μ ( v 1 , v 2 ) for every ( v 1 , v 2 ) ∈ S μ (v_1,v_2)\in S_{\mu} ( v 1 , v 2 ) ∈ S μ , that is, ( u 1 β , u 2 β ) ∈ G μ (u_{1\beta},u_{2\beta})\in\mathcal{G}_{\mu} ( u 1 β , u 2 β ) ∈ G μ . The proof is complete.
Convexity inequality for the fractional Laplacian
Lemma B.1. For any u 1 , u 2 ∈ H s ( R N ) u_1,u_2 \in H^s(\mathbb{R}^N) u 1 , u 2 ∈ H s ( R N ) , let v = u 1 2 + u 2 2 v=\sqrt{u_1^2+u_2^2} v = u 1 2 + u 2 2 . Then v ∈ H s ( R N ) v \in H^s(\mathbb{R}^N) v ∈ H s ( R N ) and
∥ ( − Δ ) s 2 v ∥ 2 2 ≤ ∥ ( − Δ ) s 2 u 1 ∥ 2 2 + ∥ ( − Δ ) s 2 u 2 ∥ 2 2 . (201) \left\|(-\Delta)^{\frac{s}{2}}v\right\|_2^2 \le\left\|(-\Delta)^{\frac{s}{2}}u_1\right\|_2^2+\left\|(-\Delta)^{\frac{s}{2}}u_2\right\|_2^2. \tag*{(201)} ( − Δ ) 2 s v 2 2 ≤ ( − Δ ) 2 s u 1 2 2 + ( − Δ ) 2 s u 2 2 2 . (201)
Proof. For any x , y ∈ R N x,y\in\mathbb{R}^N x , y ∈ R N , set a = ( u 1 ( x ) , u 2 ( x ) ) a=(u_1(x),u_2(x)) a = ( u 1 ( x ) , u 2 ( x )) , b = ( u 1 ( y ) , u 2 ( y ) ) b=(u_1(y),u_2(y)) b = ( u 1 ( y ) , u 2 ( y )) . Then v ( x ) = ∣ a ∣ v(x)=|a| v ( x ) = ∣ a ∣ , v ( y ) = ∣ b ∣ v(y)=|b| v ( y ) = ∣ b ∣ . By the reverse triangle inequality for vectors,
∣ v ( x ) − v ( y ) ∣ = ∣ ∣ a ∣ − ∣ b ∣ ∣ ≤ ∣ a − b ∣ = ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 + ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 . (202) |v(x)-v(y)|=\bigl||a|-|b|\bigr|\le|a-b|=\sqrt{|u_1(x)-u_1(y)|^2+|u_2(x)-u_2(y)|^2}. \tag*{(202)} ∣ v ( x ) − v ( y ) ∣ = ∣ a ∣ − ∣ b ∣ ≤ ∣ a − b ∣ = ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 + ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 . (202)
Squaring both sides,
∣ v ( x ) − v ( y ) ∣ 2 ≤ ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 + ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 . (203) |v(x)-v(y)|^2\le|u_1(x)-u_1(y)|^2+|u_2(x)-u_2(y)|^2. \tag*{(203)} ∣ v ( x ) − v ( y ) ∣ 2 ≤ ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 + ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 . (203)
Dividing by ∣ x − y ∣ N + 2 s |x-y|^{N+2s} ∣ x − y ∣ N + 2 s and integrating over R 2 N \mathbb{R}^{2N} R 2 N , we obtain
∫ R 2 N ∣ v ( x ) − v ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y ≤ ∫ R 2 N ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y + ∫ R 2 N ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 ∣ x − y ∣ N + 2 s d x d y . (204) \int_{\mathbb{R}^{2N}}\frac{|v(x)-v(y)|^2}{|x-y|^{N+2s}}\,dxdy\le\int_{\mathbb{R}^{2N}}\frac{|u_1(x)-u_1(y)|^2}{|x-y|^{N+2s}}\,dxdy+\int_{\mathbb{R}^{2N}}\frac{|u_2(x)-u_2(y)|^2}{|x-y|^{N+2s}}\,dxdy. \tag*{(204)} ∫ R 2 N ∣ x − y ∣ N + 2 s ∣ v ( x ) − v ( y ) ∣ 2 d x d y ≤ ∫ R 2 N ∣ x − y ∣ N + 2 s ∣ u 1 ( x ) − u 1 ( y ) ∣ 2 d x d y + ∫ R 2 N ∣ x − y ∣ N + 2 s ∣ u 2 ( x ) − u 2 ( y ) ∣ 2 d x d y . (204)
Multiplying by the normalization constant C N , s 2 \frac{C_{N,s}}{2} 2 C N , s and using Proposition 3.6 in [3 ] , which gives
[ u ] H s ( R N ) 2 = 2 C N , s − 1 ∥ ( − Δ ) s 2 u ∥ 2 2 , (205) [u]_{H^s(\mathbb{R}^N)}^2=2C_{N,s}^{-1}\left\|(-\Delta)^{\frac{s}{2}}u\right\|_2^2, \tag*{(205)} [ u ] H s ( R N ) 2 = 2 C N , s − 1 ( − Δ ) 2 s u 2 2 , (205)
we obtain the desired inequality. Moreover, v ∈ L 2 v\in L^2 v ∈ L 2 , since v 2 = u 1 2 + u 2 2 v^2=u_1^2+u_2^2 v 2 = u 1 2 + u 2 2 and u i ∈ L 2 u_i\in L^2 u i ∈ L 2 for i = 1 , 2 i=1,2 i = 1 , 2 . Since the Gagliardo seminorm of v v v is finite by the preceding inequality, we conclude that v ∈ H s ( R N ) v\in H^s(\mathbb{R}^N) v ∈ H s ( R N ) .
Pohozaev identity for the ground state Q
Lemma C.1. Let Q Q Q be the unique positive radial ground state of the single equation (1.22). Then
∥ ( − Δ ) s 2 Q ∥ 2 2 = ∣ Q ∣ 2 2 , ∣ Q ∣ p p = N + 2 s N ∣ Q ∣ 2 2 . (206) \left\|(-\Delta)^{\frac{s}{2}}Q\right\|_2^2=|Q|_2^2,\qquad|Q|_p^p=\frac{N+2s}{N}|Q|_2^2. \tag*{(206)} ( − Δ ) 2 s Q 2 2 = ∣ Q ∣ 2 2 , ∣ Q ∣ p p = N N + 2 s ∣ Q ∣ 2 2 . (206)
Proof. We derive the Pohozaev identity for a solution of the single equation ( − Δ ) s Q + 2 s N Q − ∣ Q ∣ 4 s N Q = 0 (-\Delta)^sQ+\frac{2s}{N}Q-|Q|^{\frac{4s}{N}}Q=0 ( − Δ ) s Q + N 2 s Q − ∣ Q ∣ N 4 s Q = 0 . Multiply the equation by x ⋅ ∇ Q x\cdot\nabla Q x ⋅ ∇ Q and integrate over R N \mathbb{R}^N R N . Using the identity (see e.g. [22 ] )
∫ R N ( − Δ ) s Q ( x ⋅ ∇ Q ) d x = 2 s − N 2 ∥ ( − Δ ) s 2 Q ∥ 2 2 , (207) \int_{\mathbb{R}^N}(-\Delta)^sQ(x\cdot\nabla Q)\,dx=\frac{2s-N}{2}\left\|(-\Delta)^{\frac{s}{2}}Q\right\|_2^2, \tag*{(207)} ∫ R N ( − Δ ) s Q ( x ⋅ ∇ Q ) d x = 2 2 s − N ( − Δ ) 2 s Q 2 2 , (207)
and
∫ R N 2 s N Q ( x ⋅ ∇ Q ) d x = − 2 s N ⋅ N 2 ∣ Q ∣ 2 2 = − s ∣ Q ∣ 2 2 , (208) \int_{\mathbb{R}^N}\frac{2s}{N}Q(x\cdot\nabla Q)\,dx=-\frac{2s}{N}\cdot\frac{N}{2}|Q|_2^2=-s|Q|_2^2, \tag*{(208)} ∫ R N N 2 s Q ( x ⋅ ∇ Q ) d x = − N 2 s ⋅ 2 N ∣ Q ∣ 2 2 = − s ∣ Q ∣ 2 2 , (208)
∫ R N ∣ Q ∣ 4 s N Q ( x ⋅ ∇ Q ) d x = − N 2 + 4 s N ∣ Q ∣ p p , (209) \int_{\mathbb{R}^N}|Q|^{\frac{4s}{N}}Q(x\cdot\nabla Q)\,dx=-\frac{N}{2+\frac{4s}{N}}|Q|_p^p, \tag*{(209)} ∫ R N ∣ Q ∣ N 4 s Q ( x ⋅ ∇ Q ) d x = − 2 + N 4 s N ∣ Q ∣ p p , (209)
we obtain
2 s − N 2 ∥ ( − Δ ) s 2 Q ∥ 2 2 − s ∣ Q ∣ 2 2 + N 2 + 4 s N ∣ Q ∣ p p = 0. (210) \frac{2s-N}{2}\left\|(-\Delta)^{\frac{s}{2}}Q\right\|_2^2-s|Q|_2^2+\frac{N}{2+\frac{4s}{N}}|Q|_p^p=0. \tag*{(210)} 2 2 s − N ( − Δ ) 2 s Q 2 2 − s ∣ Q ∣ 2 2 + 2 + N 4 s N ∣ Q ∣ p p = 0. (210)
On the other hand, multiplying the original equation by Q Q Q and integrating gives
∥ ( − Δ ) s 2 Q ∥ 2 2 + 2 s N ∥ Q ∥ 2 2 = ∥ Q ∥ p p . (211) \left\|(-\Delta)^{\frac{s}{2}}Q\right\|_{2}^{2}+\frac{2s}{N}\|Q\|_{2}^{2}=\|Q\|_{p}^{p}. \tag*{(211)} ( − Δ ) 2 s Q 2 2 + N 2 s ∥ Q ∥ 2 2 = ∥ Q ∥ p p . (211)
Eliminating ∥ Q ∥ p p \|Q\|_{p}^{p} ∥ Q ∥ p p from these two relations yields
∥ ( − Δ ) s 2 Q ∥ 2 2 = ∥ Q ∥ 2 2 , (212) \left\|(-\Delta)^{\frac{s}{2}}Q\right\|_{2}^{2}=\|Q\|_{2}^{2}, \tag*{(212)} ( − Δ ) 2 s Q 2 2 = ∥ Q ∥ 2 2 , (212)
then, we have
∥ ( − Δ ) s 2 Q ∥ 2 2 = N N + 2 s ∥ Q ∥ p p , (213) \left\|(-\Delta)^{\frac{s}{2}}Q\right\|_{2}^{2}=\frac{N}{N+2s}\|Q\|_{p}^{p}, \tag*{(213)} ( − Δ ) 2 s Q 2 2 = N + 2 s N ∥ Q ∥ p p , (213)
which is equivalent to
∥ Q ∥ p p = N + 2 s N ∥ Q ∥ 2 2 . (214) \|Q\|_{p}^{p}=\frac{N+2s}{N}\|Q\|_{2}^{2}. \tag*{(214)} ∥ Q ∥ p p = N N + 2 s ∥ Q ∥ 2 2 . (214)
The integration by parts for the fractional Laplacian used above can be justified by approximation and the decay of Q Q Q . This completes the proof. □ \square □
Estimate for the truncated ground state
Lemma D.1. Let Q Q Q be the ground state of (1.22) and let φ σ ( x ) = φ ( σ − 1 x ) \varphi_{\sigma}(x)=\varphi(\sigma^{-1}x) φ σ ( x ) = φ ( σ − 1 x ) be the cut-off function defined in (2.17). Then for sufficiently large σ \sigma σ , it holds that
∫ R N ∣ ( − Δ ) s 2 ( φ σ Q ) ∣ 2 d x ≤ ∫ R N ∣ ( − Δ ) s 2 Q ∣ 2 d x + O ( σ − 4 s ) . (215) \int_{\mathbb{R}^{N}}\left|(-\Delta)^{\frac{s}{2}}(\varphi_{\sigma}Q)\right|^{2}dx\leq\int_{\mathbb{R}^{N}}\left|(-\Delta)^{\frac{s}{2}}Q\right|^{2}dx+O(\sigma^{-4s}). \tag*{(215)} ∫ R N ( − Δ ) 2 s ( φ σ Q ) 2 d x ≤ ∫ R N ( − Δ ) 2 s Q 2 d x + O ( σ − 4 s ) . (215)
Proof. We first note that for dimensions N ≥ 2 N\geq2 N ≥ 2 , the desired estimate (D.1) has been established in [4 Lemma 3.2] . Therefore, it suffices to consider the one-dimensional case N = 1 N=1 N = 1 .
To this end, let φ σ ( x ) = φ ( σ − 1 x ) \varphi_{\sigma}(x)=\varphi(\sigma^{-1}x) φ σ ( x ) = φ ( σ − 1 x ) , where φ ∈ C 0 ∞ ( R ) \varphi\in C_{0}^{\infty}(\mathbb{R}) φ ∈ C 0 ∞ ( R ) satisfies (2.17). Set Q σ ( x ) : = φ σ ( x ) Q ( x ) Q_{\sigma}(x):=\varphi_{\sigma}(x)Q(x) Q σ ( x ) := φ σ ( x ) Q ( x ) . Denote
J ( u ) : = ∫ R 2 ∣ u ( x ) − u ( y ) ∣ 2 ∣ x − y ∣ 1 + 2 s d x d y . (216) J(u):=\int_{\mathbb{R}^{2}}\frac{|u(x)-u(y)|^{2}}{|x-y|^{1+2s}}\,dx\,dy. \tag*{(216)} J ( u ) := ∫ R 2 ∣ x − y ∣ 1 + 2 s ∣ u ( x ) − u ( y ) ∣ 2 d x d y . (216)
We use the decomposition in the proof of [4 Lemma 3.2] :
J ( Q σ ) = ∫ B σ × B σ + 2 ∫ D + 2 ∫ E + ∫ B σ c × B σ c , (217) J(Q_{\sigma})=\int_{B_{\sigma}\times B_{\sigma}}+2\int_{\mathbb{D}}+2\int_{\mathbb{E}}+\int_{B_{\sigma}^{c}\times B_{\sigma}^{c}}, \tag*{(217)} J ( Q σ ) = ∫ B σ × B σ + 2 ∫ D + 2 ∫ E + ∫ B σ c × B σ c , (217)
where
D = { ( x , y ) : x ∈ B σ , y ∈ B σ c , ∣ x − y ∣ > σ / 2 } , (218) \mathbb{D}=\{(x,y):x\in B_{\sigma},\ y\in B_{\sigma}^{c},\ |x-y|>\sigma/2\}, \tag*{(218)} D = {( x , y ) : x ∈ B σ , y ∈ B σ c , ∣ x − y ∣ > σ /2 } , (218)
E = { ( x , y ) : x ∈ B σ , y ∈ B σ c , ∣ x − y ∣ ≤ σ / 2 } . (219) \mathbb{E}=\{(x,y):x\in B_{\sigma},\ y\in B_{\sigma}^{c},\ |x-y|\leq\sigma/2\}. \tag*{(219)} E = {( x , y ) : x ∈ B σ , y ∈ B σ c , ∣ x − y ∣ ≤ σ /2 } . (219)
The integrals over B σ × B σ B_{\sigma}\times B_{\sigma} B σ × B σ , D \mathbb{D} D , and B σ c × B σ c B_{\sigma}^{c}\times B_{\sigma}^{c} B σ c × B σ c are estimated exactly as in [4 Lemma 3.2] ; their total contribution is at most
J ( Q ) + O ( σ − 4 s ) . (220) J(Q)+O(\sigma^{-4s}). \tag*{(220)} J ( Q ) + O ( σ − 4 s ) . (220)
Hence it remains to estimate the integral over E \mathbb{E} E . For ( x , y ) ∈ E (x,y) \in\mathbb{E} ( x , y ) ∈ E , we have y ∈ B σ c y \in B_\sigma^c y ∈ B σ c and ∣ x − y ∣ ≤ σ / 2 |x-y| \le\sigma/2 ∣ x − y ∣ ≤ σ /2 , so the segment joining x x x and y y y remains at distance at least c σ c\sigma c σ from the origin. By [4 Theorem 1.1] and the regularity and decay estimates for ground states of fractional Schrödinger equations [5 , 6 ] , one has
∣ ∇ Q ( z ) ∣ ≤ C ( 1 + ∣ z ∣ ) − 2 − 2 s . (221) |\nabla Q(z)| \le C(1+|z|)^{-2-2s}. \tag*{(221)} ∣∇ Q ( z ) ∣ ≤ C ( 1 + ∣ z ∣ ) − 2 − 2 s . (221)
Since φ ∈ C ∞ ( R ) \varphi\in C^\infty(\mathbb{R}) φ ∈ C ∞ ( R ) , there exists a constant C > 0 C>0 C > 0 , independent of σ \sigma σ , such that ∣ φ ′ ∣ ≤ C |\varphi'| \le C ∣ φ ′ ∣ ≤ C . Hence
∣ ∇ φ σ ( z ) ∣ ≤ C σ − 1 . (222) |\nabla\varphi_\sigma(z)| \le C\sigma^{-1}. \tag*{(222)} ∣∇ φ σ ( z ) ∣ ≤ C σ − 1 . (222)
Consequently, for ∣ z ∣ ≥ σ / 2 |z| \ge\sigma/2 ∣ z ∣ ≥ σ /2 ,
∣ ∇ Q σ ( z ) ∣ ≤ ∣ ∇ φ σ ( z ) ∣ ∣ Q ( z ) ∣ + φ σ ( z ) ∣ ∇ Q ( z ) ∣ ≤ C σ − 2 − 2 s . (223) |\nabla Q_\sigma(z)| \le|\nabla\varphi_\sigma(z)| |Q(z)|+\varphi_\sigma(z)|\nabla Q(z)| \le C\sigma^{-2-2s}. \tag*{(223)} ∣∇ Q σ ( z ) ∣ ≤ ∣∇ φ σ ( z ) ∣∣ Q ( z ) ∣ + φ σ ( z ) ∣∇ Q ( z ) ∣ ≤ C σ − 2 − 2 s . (223)
Therefore, for ( x , y ) ∈ E (x,y) \in\mathbb{E} ( x , y ) ∈ E ,
∣ Q σ ( x ) − Q σ ( y ) ∣ ≤ C σ − 2 − 2 s ∣ x − y ∣ . (224) |Q_\sigma(x)-Q_\sigma(y)| \le C\sigma^{-2-2s}|x-y|. \tag*{(224)} ∣ Q σ ( x ) − Q σ ( y ) ∣ ≤ C σ − 2 − 2 s ∣ x − y ∣. (224)
It follows that
∫ E ∣ Q σ ( x ) − Q σ ( y ) ∣ 2 ∣ x − y ∣ 1 + 2 s d x d y ≤ C σ − 4 − 4 s ∫ ∣ x ∣ ≤ σ d x ∫ ∣ ξ ∣ ≤ σ / 2 ∣ ξ ∣ 1 − 2 s d ξ = O ( σ − 1 − 6 s ) = O ( σ − 4 s ) , (225) \begin{aligned}
\int_{\mathbb{E}}\frac{|Q_\sigma(x)-Q_\sigma(y)|^2}{|x-y|^{1+2s}}\,dx\,dy
&\le C\sigma^{-4-4s}\int_{|x|\le\sigma}dx\int_{|\xi|\le\sigma/2}|\xi|^{1-2s}\,d\xi\\
&=O(\sigma^{-1-6s})=O(\sigma^{-4s}), \tag*{(225)}
\end{aligned} ∫ E ∣ x − y ∣ 1 + 2 s ∣ Q σ ( x ) − Q σ ( y ) ∣ 2 d x d y ≤ C σ − 4 − 4 s ∫ ∣ x ∣ ≤ σ d x ∫ ∣ ξ ∣ ≤ σ /2 ∣ ξ ∣ 1 − 2 s d ξ = O ( σ − 1 − 6 s ) = O ( σ − 4 s ) , (225)
where we used s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) and σ → ∞ \sigma\to\infty σ → ∞ . Combining (D.3), (D.6) and (D.11), one can infer that
J ( Q σ ) ≤ J ( Q ) + O ( σ − 4 s ) . (226) J(Q_\sigma)\le J(Q)+O(\sigma^{-4s}). \tag*{(226)} J ( Q σ ) ≤ J ( Q ) + O ( σ − 4 s ) . (226)
Finally, since
∫ R ∣ ( − Δ ) s 2 u ∣ 2 d x = c 1 , s J ( u ) (227) \int_{\mathbb{R}}\left|(-\Delta)^{\frac{s}{2}}u\right|^2\,dx=c_{1,s}J(u) \tag*{(227)} ∫ R ( − Δ ) 2 s u 2 d x = c 1 , s J ( u ) (227)
for a positive constant c 1 , s c_{1,s} c 1 , s depending only on s s s (see Proposition 3.6 in [3 ] ), we obtain
∫ R ∣ ( − Δ ) s 2 ( φ σ Q ) ∣ 2 d x ≤ ∫ R ∣ ( − Δ ) s 2 Q ∣ 2 d x + O ( σ − 4 s ) . (228) \int_{\mathbb{R}}\left|(-\Delta)^{\frac{s}{2}}(\varphi_\sigma Q)\right|^2\,dx
\le\int_{\mathbb{R}}\left|(-\Delta)^{\frac{s}{2}}Q\right|^2\,dx+O(\sigma^{-4s}). \tag*{(228)} ∫ R ( − Δ ) 2 s ( φ σ Q ) 2 d x ≤ ∫ R ( − Δ ) 2 s Q 2 d x + O ( σ − 4 s ) . (228)
This completes the proof for N = 1 N=1 N = 1 .
Optimizers of the coupled GNS inequality
Lemma E.1. Let ( u 1 , u 2 ) ∈ H s ( R N ) × H s ( R N ) ∖ { ( 0 , 0 ) } (u_1,u_2)\in H^s(\mathbb{R}^N)\times H^s(\mathbb{R}^N)\setminus\{(0,0)\} ( u 1 , u 2 ) ∈ H s ( R N ) × H s ( R N ) ∖ {( 0 , 0 )} be such that equality holds in both (49) and (58) . Then there exist constants κ > 0 \kappa>0 κ > 0 , α > 0 \alpha>0 α > 0 , η ∈ R N \eta\in\mathbb{R}^N η ∈ R N , and θ ∈ [ 0 , 2 π ) \theta\in[0,2\pi) θ ∈ [ 0 , 2 π ) such that
( u 1 ( x ) , u 2 ( x ) ) = ( κ sin θ Q ( α x + η ) , κ cos θ Q ( α x + η ) ) , (229) (u_1(x),u_2(x))=(\kappa\sin\theta\,Q(\alpha x+\eta),\kappa\cos\theta\,Q(\alpha x+\eta)), \tag*{(229)} ( u 1 ( x ) , u 2 ( x )) = ( κ sin θ Q ( α x + η ) , κ cos θ Q ( α x + η )) , (229) where Q Q Q is the ground state of (22) . Moreover, if the pair additionally satisfies ∫ R ( u 1 2 + u 2 2 ) d x = 1 \int_{\mathbb{R}}(u_1^2+u_2^2)\,dx=1 ∫ R ( u 1 2 + u 2 2 ) d x = 1 , then κ 2 α − N ∥ Q ∥ 2 2 = 1 \kappa^2\alpha^{-N}\|Q\|_2^2=1 κ 2 α − N ∥ Q ∥ 2 2 = 1 .
Proof. Step 1: Equality in (2.1) implies proportionality. By Hölder’s inequality, equality holds in (49) if and only if ∣ u 1 ∣ p |u_1|^p ∣ u 1 ∣ p and ∣ u 2 ∣ p |u_2|^p ∣ u 2 ∣ p are linearly dependent, i.e., there exists c ≥ 0 c\ge0 c ≥ 0 such that ∣ u 2 ∣ = c ∣ u 1 ∣ |u_2|=c|u_1| ∣ u 2 ∣ = c ∣ u 1 ∣ almost everywhere. Step 2: Equality in (58) forces v = u 1 2 + u 2 2 v=\sqrt{u_1^2+u_2^2} v = u 1 2 + u 2 2 to be an optimizer of the single GNS inequality. Indeed, equality in (58) requires that v v v saturates (25) and that equality holds in the convexity inequality (B.1), namely
∥ ( − Δ ) s 2 v ∥ 2 2 = ∥ ( − Δ ) s 2 u 1 ∥ 2 2 + ∥ ( − Δ ) s 2 u 2 ∥ 2 2 . (230) \left\|(-\Delta)^{\frac{s}{2}}v\right\|_2^2=\left\|(-\Delta)^{\frac{s}{2}}u_1\right\|_2^2+\left\|(-\Delta)^{\frac{s}{2}}u_2\right\|_2^2. \tag*{(230)} ( − Δ ) 2 s v 2 2 = ( − Δ ) 2 s u 1 2 2 + ( − Δ ) 2 s u 2 2 2 . (230)
Consequently, v ( x ) = β Q ( α x + η ) v(x)=\beta Q(\alpha x+\eta) v ( x ) = β Q ( α x + η ) for some β > 0 \beta>0 β > 0 , α > 0 \alpha>0 α > 0 and η ∈ R N \eta\in\mathbb{R}^N η ∈ R N .
Step 3: Form of u 1 u_1 u 1 and u 2 u_2 u 2 . From Step 1, ∣ u 2 ∣ = c ∣ u 1 ∣ |u_2|=c|u_1| ∣ u 2 ∣ = c ∣ u 1 ∣ , so v = 1 + c 2 ∣ u 1 ∣ v=\sqrt{1+c^2}|u_1| v = 1 + c 2 ∣ u 1 ∣ . Thus
∣ u 1 ∣ = β 1 + c 2 Q ( α x + η ) , ∣ u 2 ∣ = c β 1 + c 2 Q ( α x + η ) . (231) |u_1|=\frac{\beta}{\sqrt{1+c^2}}Q(\alpha x+\eta),\qquad|u_2|=\frac{c\beta}{\sqrt{1+c^2}}Q(\alpha x+\eta). \tag*{(231)} ∣ u 1 ∣ = 1 + c 2 β Q ( α x + η ) , ∣ u 2 ∣ = 1 + c 2 c β Q ( α x + η ) . (231)
Setting κ = β \kappa=\beta κ = β and defining θ \theta θ by sin θ = 1 / 1 + c 2 \sin\theta=1/\sqrt{1+c^2} sin θ = 1/ 1 + c 2 , cos θ = c / 1 + c 2 \cos\theta=c/\sqrt{1+c^2} cos θ = c / 1 + c 2 gives the desired form. The normalization condition follows from a direct computation.
Conflict of Interest
The authors declare no conflict of interest.
Data Availability Statement
Not applicable. No data were generated or analyzed in this study.