Introduction and main results
Coupled nonlinear Schrödinger equations arise in the study of interacting wave components, particularly in multicomponent Bose–Einstein condensates [16 , 24 ] and nonlinear optics [9 ] . The interactions between different components play an important role in the existence and structure of standing waves [2 , 25 ] . In normalized problems, the masses of the components are prescribed, while the corresponding frequencies are determined as Lagrange multipliers. This formulation connects the search for standing-wave profiles with critical points of the energy under mass constraints. For two components, the coupling must be treated together with two prescribed masses and two unknown multipliers.
The variational theory of normalized solutions for scalar nonlinear Schrödinger equations has attracted considerable attention. In the mass-subcritical regime, Cazenave and Lions [8 ] established the connection between constrained minimization, existence, and orbital stability of standing waves. When the energy is unbounded from below on the mass constraint, direct minimization is no longer applicable. Jeanjean [17 ] introduced a constrained mountain-pass framework based on a scaling parameter and the Pohozaev identity, which yields bounded Palais–Smale sequences. Further developments in the variational theory of normalized Schrödinger equations can be found in [1 , 3 , 18 , 20 , 21 , 26 ] .
For problems involving combined nonlinearities with Sobolev critical growth, a representative normalized model is
{ − Δ u = λ u + μ ∣ u ∣ q − 2 u + ∣ u ∣ 2 ∗ − 2 u , in R N , ∫ R N ∣ u ∣ 2 d x = ρ 2 , \left\{
\begin{aligned}
-\Delta u &= \lambda u+\mu|u|^{q-2}u+|u|^{2^{*}-2}u,\quad&&\text{in }\mathbb{R}^{N},\\
\int_{\mathbb{R}^{N}} |u|^{2}\,dx &= \rho^{2},
\end{aligned}
\right. ⎩ ⎨ ⎧ − Δ u ∫ R N ∣ u ∣ 2 d x = λ u + μ ∣ u ∣ q − 2 u + ∣ u ∣ 2 ∗ − 2 u , = ρ 2 , in R N ,
where N ≥ 3 N \ge3 N ≥ 3 , 2 < q < 2 ∗ = 2 N N − 2 2 < q < 2^{*} = \frac{2N}{N-2} 2 < q < 2 ∗ = N − 2 2 N , μ , ρ > 0 \mu,\rho> 0 μ , ρ > 0 , and λ ∈ R \lambda\in\mathbb{R} λ ∈ R is an unknown Lagrange multiplier. Soave [27 ] investigated normalized ground states for this type of problem and raised the question of the existence of a second solution of mountain-pass type in addition to the local minimizer. In the mass-subcritical regime and under suitable smallness assumptions, Jeanjean–Le [19 ] and Wei–Wu [28 ] subsequently established the corresponding multiplicity results for N ≥ 4 N \ge4 N ≥ 4 and N = 3 N = 3 N = 3 , respectively. Related results for coupled Schrödinger systems, including least-energy solutions and solutions with prescribed masses, can be found in [2 , 10 , 25 , 31 , 32 ] . These developments naturally motivate the study of normalized Schrödinger systems in weighted critical settings, where singular weights introduce additional variational difficulties.
For weighted critical problems, the Caffarelli–Kohn–Nirenberg (CKN) inequality provides a natural framework for the analysis of such singular and critical nonlinearities. The sharp CKN inequality takes the form
S ( a , b ) ( ∫ R N ∣ x ∣ − b p ∣ w ∣ p d x ) 2 / p ≤ ∫ R N ∣ x ∣ − 2 a ∣ ∇ w ∣ 2 d x . (1) S(a,b)\left(\int_{\mathbb{R}^{N}} |x|^{-bp}|w|^{p}\,dx\right)^{2/p}\leq\int_{\mathbb{R}^{N}} |x|^{-2a}|\nabla w|^{2}\,dx. \tag*{(1)} S ( a , b ) ( ∫ R N ∣ x ∣ − b p ∣ w ∣ p d x ) 2/ p ≤ ∫ R N ∣ x ∣ − 2 a ∣∇ w ∣ 2 d x . (1)
where S ( a , b ) S(a,b) S ( a , b ) denotes the best constant, p = 2 N N − 2 ( 1 + a − b ) p = \frac{2N}{N-2(1+a-b)} p = N − 2 ( 1 + a − b ) 2 N , 0 ≤ a < N − 4 2 0 \leq a < \frac{N-4}{2} 0 ≤ a < 2 N − 4 , and a < b < a + 1 a < b < a+1 a < b < a + 1 . The inequality was introduced by Caffarelli, Kohn and Nirenberg [5 ] . Catrina and Wang [7 ] investigated sharp constants and extremal functions, while Felli and Schneider [12 ] established symmetry-breaking results. Further developments concerning rigidity and stability can be found in [11 , 13 , 29 , 30 , 33 ] . The condition a ≥ 0 a \geq0 a ≥ 0 places the sharp extremal in the radial-symmetry range, while a < N − 4 2 a < \frac{N-4}{2} a < 2 N − 4 ensures its square integrability with respect to the weight ∣ x ∣ − 2 a |x|^{-2a} ∣ x ∣ − 2 a . This finite-mass property is used in the critical energy estimates.
For normalized inhomogeneous Schrödinger equations with critical Hardy–Sobolev-type nonlinearities, Cardoso, de Oliveira and Miyagaki [6 ] established related existence results. Their work provides a relevant scalar framework for comparison, particularly in the case a = 0 a = 0 a = 0 , where the diffusion operator reduces to the standard Laplacian. More recently, Goel and Rai [15 ] studied a scalar weighted equation with the CKN critical exponent in a radial setting. Under suitable parameter restrictions, they obtained two positive radial normalized solutions in the mass-subcritical regime and established the existence of a positive radial normalized ground state in the mass-critical and mass-supercritical regimes. Their approach uses mass-preserving fiber maps to analyze the Pohozaev geometry and construct constrained Palais–Smale sequences. Strict energy estimates are then used to establish compactness at the relevant variational levels.
The present study concerns the coupled CKN system, which we treat through synchronized reduction. The resulting scalar variational problem is analyzed in the full weighted energy space of real-valued functions, without imposing radial symmetry. The main contributions of this paper concern the existence and multiplicity of compatible synchronized solutions to the coupled CKN system, together with a strict comparison between the synchronized and full vector Pohozaev levels.
For prescribed masses ρ 1 , ρ 2 > 0 \rho_1,\rho_2>0 ρ 1 , ρ 2 > 0 , we consider the system
{ − div ( ∣ x ∣ − 2 a ∇ u ) = λ 1 ∣ x ∣ − 2 a u + β 1 ∣ x ∣ − b q ∣ u ∣ q − 2 u + ∣ x ∣ − b p ( μ 1 ∣ u ∣ p − 2 u + ν α ∣ u ∣ α − 2 u ∣ v ∣ γ ) , in R N , − div ( ∣ x ∣ − 2 a ∇ v ) = λ 2 ∣ x ∣ − 2 a v + β 2 ∣ x ∣ − b q ∣ v ∣ q − 2 v + ∣ x ∣ − b p ( μ 2 ∣ v ∣ p − 2 v + ν γ ∣ u ∣ α ∣ v ∣ γ − 2 v ) , in R N , ∫ R N ∣ x ∣ − 2 a u 2 d x = ρ 1 2 , ∫ R N ∣ x ∣ − 2 a v 2 d x = ρ 2 2 . (2) \left\{
\begin{aligned}
-\operatorname{div}(|x|^{-2a}\nabla u) &= \lambda_1|x|^{-2a}u+\beta_1|x|^{-bq}|u|^{q-2}u \\
&\quad+|x|^{-bp}\left(\mu_1|u|^{p-2}u+\nu\alpha|u|^{\alpha-2}u|v|^\gamma\right), &&\text{in }\mathbb{R}^N,\\
-\operatorname{div}(|x|^{-2a}\nabla v) &= \lambda_2|x|^{-2a}v+\beta_2|x|^{-bq}|v|^{q-2}v \\
&\quad+|x|^{-bp}\left(\mu_2|v|^{p-2}v+\nu\gamma|u|^\alpha|v|^{\gamma-2}v\right), &&\text{in }\mathbb{R}^N,\\
\int_{\mathbb{R}^N}|x|^{-2a}u^2\,dx &= \rho_1^2,\qquad\int_{\mathbb{R}^N}|x|^{-2a}v^2\,dx=\rho_2^2.
\end{aligned}
\right. \tag*{(2)} ⎩ ⎨ ⎧ − div ( ∣ x ∣ − 2 a ∇ u ) − div ( ∣ x ∣ − 2 a ∇ v ) ∫ R N ∣ x ∣ − 2 a u 2 d x = λ 1 ∣ x ∣ − 2 a u + β 1 ∣ x ∣ − b q ∣ u ∣ q − 2 u + ∣ x ∣ − b p ( μ 1 ∣ u ∣ p − 2 u + ν α ∣ u ∣ α − 2 u ∣ v ∣ γ ) , = λ 2 ∣ x ∣ − 2 a v + β 2 ∣ x ∣ − b q ∣ v ∣ q − 2 v + ∣ x ∣ − b p ( μ 2 ∣ v ∣ p − 2 v + ν γ ∣ u ∣ α ∣ v ∣ γ − 2 v ) , = ρ 1 2 , ∫ R N ∣ x ∣ − 2 a v 2 d x = ρ 2 2 . in R N , in R N , (2)
where N ≥ 5 N\ge5 N ≥ 5 , 0 ≤ a < N − 4 2 0\le a<\frac{N-4}{2} 0 ≤ a < 2 N − 4 , a < b < a + 1 a<b<a+1 a < b < a + 1 , d = 1 + a − b d=1+a-b d = 1 + a − b , p = 2 N N − 2 d p=\frac{2N}{N-2d} p = N − 2 d 2 N , and 2 < q < p 2<q<p 2 < q < p . Here λ 1 , λ 2 ∈ R \lambda_1,\lambda_2\in\mathbb{R} λ 1 , λ 2 ∈ R are unknown Lagrange multipliers. The coupling exponents satisfy
α > 1 , γ > 1 , α + γ = p , (3) \alpha>1,\qquad\gamma>1,\qquad\alpha+\gamma=p, \tag*{(3)} α > 1 , γ > 1 , α + γ = p , (3)
and the coefficients satisfy
μ 1 > 0 , μ 2 > 0 , ν > 0 , β 1 > 0 , β 2 > 0. (4) \mu_1>0,\qquad\mu_2>0,\qquad\nu>0,\qquad\beta_1>0,\qquad\beta_2>0. \tag*{(4)} μ 1 > 0 , μ 2 > 0 , ν > 0 , β 1 > 0 , β 2 > 0. (4)
To construct synchronized solutions, we impose the compatibility conditions
β 1 ρ 1 q − 2 = β 2 ρ 2 q − 2 = β > 0 (5) \beta_1\rho_1^{q-2}=\beta_2\rho_2^{q-2}=\beta>0 \tag*{(5)} β 1 ρ 1 q − 2 = β 2 ρ 2 q − 2 = β > 0 (5)
and
μ 1 ρ 1 p − 2 + ν α ρ 1 α − 2 ρ 2 γ = μ 2 ρ 2 p − 2 + ν γ ρ 1 α ρ 2 γ − 2 = κ > 0. (6) \mu_1\rho_1^{p-2}+\nu\alpha\rho_1^{\alpha-2}\rho_2^\gamma
=\mu_2\rho_2^{p-2}+\nu\gamma\rho_1^\alpha\rho_2^{\gamma-2}
=\kappa>0. \tag*{(6)} μ 1 ρ 1 p − 2 + ν α ρ 1 α − 2 ρ 2 γ = μ 2 ρ 2 p − 2 + ν γ ρ 1 α ρ 2 γ − 2 = κ > 0. (6)
Under these conditions, the synchronized ansatz
u = ρ 1 w , v = ρ 2 w u=\rho_1w,\qquad v=\rho_2w u = ρ 1 w , v = ρ 2 w
necessarily yields λ 1 = λ 2 = : λ \lambda_1=\lambda_2=:\lambda λ 1 = λ 2 =: λ . Consequently, the system (1.2) reduces, within the synchronized class, to the scalar normalized problem
{ − div ( ∣ x ∣ − 2 a ∇ w ) = λ ∣ x ∣ − 2 a w + β ∣ x ∣ − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ − b p ∣ w ∣ p − 2 w , ∫ R N ∣ x ∣ − 2 a w 2 d x = 1. (7) \left\{
\begin{aligned}
-\operatorname{div}(|x|^{-2a}\nabla w)&=\lambda|x|^{-2a}w+\beta|x|^{-bq}|w|^{q-2}w+\kappa|x|^{-bp}|w|^{p-2}w,\\
\int_{\mathbb{R}^N}|x|^{-2a}w^2\,dx&=1.
\end{aligned}
\right. \tag*{(7)} ⎩ ⎨ ⎧ − div ( ∣ x ∣ − 2 a ∇ w ) ∫ R N ∣ x ∣ − 2 a w 2 d x = λ ∣ x ∣ − 2 a w + β ∣ x ∣ − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ − b p ∣ w ∣ p − 2 w , = 1. (7)
where λ ∈ R \lambda\in\mathbb{R} λ ∈ R is an unknown Lagrange multiplier.
The main compactness difficulties arise from critical concentration and the possible loss of prescribed masses under weak convergence. Moreover, the vector Pohozaev identity alone does not determine the signs of the two Lagrange multipliers separately. The synchronized reduction allows us to work with a single multiplier, while recovering the prescribed mass and excluding critical concentration remain essential.
Define the interpolation exponent and the mass-critical exponent by
δ q = ( N − 2 a + 2 b ) q − 2 N 2 q , q c = 2 N + 4 N − 2 a + 2 b . \delta_q=\frac{(N-2a+2b)q-2N}{2q},\qquad q_c=\frac{2N+4}{N-2a+2b}. δ q = 2 q ( N − 2 a + 2 b ) q − 2 N , q c = N − 2 a + 2 b 2 N + 4 .
The exponent q c q_c q c is determined by the mass-preserving scaling (14) . Accordingly, 2 < q < q c 2<q<q_c 2 < q < q c , q = q c q=q_c q = q c , and q c < q < p q_c<q<p q c < q < p correspond to the mass-subcritical, mass-critical, and mass-supercritical regimes, respectively.
Let E E E denote the energy functional associated with (1.2), defined in (2.12), and let S s y n \mathcal{S}_{\mathrm{syn}} S syn and P s y n \mathcal{P}_{\mathrm{syn}} P syn be the synchronized mass constraint and Pohozaev set defined in (2.13). We denote by K q K_q K q the optimal constant in the weighted Gagliardo–Nirenberg inequality (2.6), and by T κ T_\kappa T κ the critical energy threshold for the reduced scalar problem, defined in Section 2. In the mass-subcritical regime, P s y n − \mathcal{P}_{\mathrm{syn}}^{-} P syn − denotes the upper branch of P s y n \mathcal{P}_{\mathrm{syn}} P syn , while the positive smallness threshold β 0 \beta_0 β 0 is specified in Section 3. A synchronized Pohozaev ground state is understood as a minimizer of E E E over P s y n \mathcal{P}_{\mathrm{syn}} P syn . Our first main result is as follows.
Theorem 1.1. Let N ≥ 5 N \ge5 N ≥ 5 , 0 ≤ a < N − 4 2 0 \le a < \frac{N-4}{2} 0 ≤ a < 2 N − 4 , a < b < a + 1 a < b < a + 1 a < b < a + 1 , 2 < q < p 2 < q < p 2 < q < p . Assume that (1.3), (1.4), and the compatibility conditions (1.5) and (1.6) hold.
(i) If 2 < q < q c 2 < q < q_c 2 < q < q c , there exists
β 0 = β 0 ( N , a , b , q , κ ) > 0 \beta_0=\beta_0(N,a,b,q,\kappa)>0 β 0 = β 0 ( N , a , b , q , κ ) > 0 such that, for 0 < β < β 0 0 < \beta< \beta_0 0 < β < β 0 , system (1.2) admits two positive synchronized normalized solutions
( u + , v + ) = ( ρ 1 w + , ρ 2 w + ) , ( u − , v − ) = ( ρ 1 w − , ρ 2 w − ) . (u^+,v^+)=(\rho_1w^+,\rho_2w^+),\qquad(u^-,v^-)=(\rho_1w^-,\rho_2w^-). ( u + , v + ) = ( ρ 1 w + , ρ 2 w + ) , ( u − , v − ) = ( ρ 1 w − , ρ 2 w − ) . The first is a local minimizer of E E E on S s y n \mathcal{S}_{\mathrm{syn}} S syn , while the second minimizes E E E on P s y n − \mathcal{P}_{\mathrm{syn}}^{-} P syn − . Their energies satisfy
E ( u + , v + ) < 0 < E ( u − , v − ) . (8) E(u^+,v^+)<0<E(u^-,v^-). \tag*{(8)} E ( u + , v + ) < 0 < E ( u − , v − ) . (8) (ii) Assume that either q = q c q=q_c q = q c and
0 < β < 1 δ q K q , (9) 0<\beta<\frac{1}{\delta_qK_q}, \tag*{(9)} 0 < β < δ q K q 1 , (9) or q c < q < p q_c<q<p q c < q < p and β > 0 \beta>0 β > 0 hold. Then system (1.2) admits a positive ground state on P s y n \mathcal{P}_{\mathrm{syn}} P syn of the form
( u , v ) = ( ρ 1 w , ρ 2 w ) . (u,v)=(\rho_1w,\rho_2w). ( u , v ) = ( ρ 1 w , ρ 2 w ) . Its energy satisfies
0 < E ( u , v ) < ( ρ 1 2 + ρ 2 2 ) T κ . (10) 0<E(u,v)<(\rho_1^2+\rho_2^2)T_\kappa. \tag*{(10)} 0 < E ( u , v ) < ( ρ 1 2 + ρ 2 2 ) T κ . (10) In all cases, the Lagrange multipliers satisfy
λ 1 = λ 2 = λ = ( δ q − 1 ) β ∫ R N ∣ x ∣ − b q ∣ w ∣ q d x < 0. (11) \lambda_1=\lambda_2=\lambda=(\delta_q-1)\beta\int_{\mathbb{R}^N}|x|^{-bq}|w|^q\,dx<0. \tag*{(11)} λ 1 = λ 2 = λ = ( δ q − 1 ) β ∫ R N ∣ x ∣ − b q ∣ w ∣ q d x < 0. (11)
Remark 1.1. In the mass-critical case q = q c q=q_c q = q c , δ q = 2 / q c \delta_q=2/q_c δ q = 2/ q c . Thus, condition (1.9) is equivalent to β < q c / ( 2 K q ) \beta<q_c/(2K_q) β < q c / ( 2 K q ) . For unit weighted mass and parameters in the common range, the corresponding threshold in Goel and Rai [15 ] has the same form, with their interpolation coefficient C a , b q c C_{a,b}^{q_c} C a , b q c in place of K q K_q K q .
Remark 1.2. The compatibility conditions (1.5) and (1.6) hold for the symmetric data ρ 1 = ρ 2 = ρ \rho_1=\rho_2=\rho ρ 1 = ρ 2 = ρ , β 1 = β 2 \beta_1=\beta_2 β 1 = β 2 , μ 1 = μ 2 \mu_1=\mu_2 μ 1 = μ 2 , and α = γ = p / 2 \alpha=\gamma=p/2 α = γ = p /2 , with β = β 1 ρ p − q \beta=\beta_1\rho^{p-q} β = β 1 ρ p − q , κ = ρ p − 2 ( μ 1 + p 2 ν ) \kappa=\rho^{p-2}(\mu_1+\frac{p}{2}\nu) κ = ρ p − 2 ( μ 1 + 2 p ν ) . More generally, these conditions also allow nonsymmetric data. With the masses, coupling exponents, and ν \nu ν fixed, a family of compatible positive coefficients is given by
β i = β ρ i 2 − q , i = 1 , 2 , μ 1 = ρ 1 2 − p ( κ − ν α ρ 1 α − 2 ρ 2 γ ) , μ 2 = ρ 2 2 − p ( κ − ν γ ρ 1 α ρ 2 γ − 2 ) , \beta_i=\beta\rho_i^{2-q},\quad i=1,2,\qquad
\mu_1=\rho_1^{2-p}\left(\kappa-\nu\alpha\rho_1^{\alpha-2}\rho_2^\gamma\right),\qquad
\mu_2=\rho_2^{2-p}\left(\kappa-\nu\gamma\rho_1^\alpha\rho_2^{\gamma-2}\right), β i = β ρ i 2 − q , i = 1 , 2 , μ 1 = ρ 1 2 − p ( κ − ν α ρ 1 α − 2 ρ 2 γ ) , μ 2 = ρ 2 2 − p ( κ − ν γ ρ 1 α ρ 2 γ − 2 ) ,
where β > 0 \beta>0 β > 0 and κ > ν max { α ρ 1 α − 2 ρ 2 γ , γ ρ 1 α ρ 2 γ − 2 } \kappa>\nu\max\left\{\alpha\rho_1^{\alpha-2}\rho_2^\gamma,\gamma\rho_1^\alpha\rho_2^{\gamma-2}\right\} κ > ν max { α ρ 1 α − 2 ρ 2 γ , γ ρ 1 α ρ 2 γ − 2 } .
The next result compares the synchronized and full vector Pohozaev levels. The levels m v e c m_{\mathrm{vec}} m vec , m s y n m_{\mathrm{syn}} m syn , and m χ m_\chi m χ are defined in Section 2.
Theorem 1.2 (Strict vector comparison). Assume that N ≥ 5 N \ge5 N ≥ 5 , 0 ≤ a < N − 4 2 0 \le a < \frac{N-4}{2} 0 ≤ a < 2 N − 4 , a < b < a + 1 a < b < a+1 a < b < a + 1 , and q = q c q=q_c q = q c , and that (1.9) holds. Let
ρ 1 = ρ 2 = 1 , β 1 = β 2 = β , μ 1 = μ 2 = μ > 0 , \rho_1=\rho_2=1,\qquad\beta_1=\beta_2=\beta,\qquad\mu_1=\mu_2=\mu>0, ρ 1 = ρ 2 = 1 , β 1 = β 2 = β , μ 1 = μ 2 = μ > 0 , α = γ = p / 2 , ν > 0 , κ = μ + p 2 ν . \alpha=\gamma=p/2,\qquad\nu>0,\qquad\kappa=\mu+\frac{p}{2}\nu. α = γ = p /2 , ν > 0 , κ = μ + 2 p ν . Then
0 < m v e c < m μ , m s y n = 2 m κ , m χ = χ − 2 / ( p − 2 ) m 1 ( χ > 0 ) . (12) 0<m_{\mathrm{vec}}<m_\mu,\qquad m_{\mathrm{syn}}=2m_\kappa,\qquad m_\chi=\chi^{-2/(p-2)}m_1\quad(\chi>0). \tag*{(12)} 0 < m vec < m μ , m syn = 2 m κ , m χ = χ − 2/ ( p − 2 ) m 1 ( χ > 0 ) . (12) In particular, if
0 < ν ≤ 2 μ p ( 2 ( p − 2 ) / 2 − 1 ) , (13) 0<\nu\le\frac{2\mu}{p}\left(2^{(p-2)/2}-1\right), \tag*{(13)} 0 < ν ≤ p 2 μ ( 2 ( p − 2 ) /2 − 1 ) , (13) then m v e c < m s y n m_{\mathrm{vec}}<m_{\mathrm{syn}} m vec < m syn . Every synchronized Pohozaev ground state provided by Theorem 1.1 has energy strictly greater than the full vector Pohozaev infimum.
The proof of Theorem 1.1 combines synchronized reduction with constrained variational methods. The mass-preserving fiber maps characterize the Pohozaev geometry and are used to construct bounded constrained Palais–Smale sequences. Compactness at the relevant energy levels is recovered through weighted subcritical embeddings and an energy-splitting argument in the spirit of Lions’ concentration–compactness principle [22 , 23 ] . More precisely, Brezis–Lieb splitting [4 ] and the sharp CKN inequality yield the critical energy threshold T κ T_\kappa T κ for nontrivial concentration, while the Lagrange multiplier identity recovers the prescribed mass of any nontrivial weak limit. To exclude critical concentration, we use a sharp CKN extremal with finite weighted mass to derive strict energy estimates without cutoff errors. In the mass-subcritical regime, the lower-branch solution is obtained by local minimization, while the upper-branch level lies strictly below the sum of the lower-branch level and T κ T_\kappa T κ . In the mass-critical and mass-supercritical regimes, the scalar Pohozaev level lies strictly below T κ T_\kappa T κ . These estimates yield the required compactness and establish the scalar existence results, which are then lifted to positive synchronized solutions of the original system.
The proof of Theorem 1.2 is based on a quantitative perturbation argument. We construct a vector test pair whose first component is a scalar ground state, while the second is a mass-normalized combination of a small multiple of the same ground state and a diffuse profile carrying the remaining mass. After mass normalization and a common mass-preserving dilation, an energy expansion yields m v e c < m μ m_{\mathrm{vec}}<m_\mu m vec < m μ . The scaling law for m χ m_\chi m χ and the identity m s y n = 2 m κ m_{\mathrm{syn}}=2m_\kappa m syn = 2 m κ then imply m v e c < m s y n m_{\mathrm{vec}}<m_{\mathrm{syn}} m vec < m syn under the stated weak-coupling condition. Attainment of m v e c m_{\mathrm{vec}} m vec requires a separate compactness argument for both components and is not asserted here.
The remainder of the paper is organized as follows. Section 2 introduces the functional framework and establishes the synchronized reduction. In Section 3, we analyze the geometry of the fiber maps and construct constrained Palais–Smale sequences. Regularity properties, Pohozaev identities, and compactness results are established in Section 4. The scalar existence results in the different mass regimes are proved in Sections 5 and 6. Finally, Section 7 completes the analysis of the vector system with a strict comparison between the synchronized and full vector Pohozaev levels.
Functional framework
We introduce the weighted functional framework and establish the estimates and synchronized reduction used in the proofs.
For real-valued functions, we use the following notation to facilitate subsequent computations:
A ( w ) = ∫ ∣ x ∣ − 2 a ∣ ∇ w ∣ 2 , M ( w ) = ∫ ∣ x ∣ − 2 a w 2 , A(w)=\int|x|^{-2a}|\nabla w|^2,\qquad M(w)=\int|x|^{-2a}w^2, A ( w ) = ∫ ∣ x ∣ − 2 a ∣∇ w ∣ 2 , M ( w ) = ∫ ∣ x ∣ − 2 a w 2 ,
B ( w ) = ∫ ∣ x ∣ − b q ∣ w ∣ q , C ( w ) = ∫ ∣ x ∣ − b p ∣ w ∣ p . B(w)=\int|x|^{-bq}|w|^q,\qquad C(w)=\int|x|^{-bp}|w|^p. B ( w ) = ∫ ∣ x ∣ − b q ∣ w ∣ q , C ( w ) = ∫ ∣ x ∣ − b p ∣ w ∣ p .
Let D a 1 , 2 D_a^{1,2} D a 1 , 2 be the completion of C c ∞ ( R N ) C_c^\infty(\mathbb{R}^N) C c ∞ ( R N ) with respect to the norm A 1 / 2 A^{1/2} A 1/2 , and set
X = D a 1 , 2 ( R N ) ∩ L 2 ( R N ; ∣ x ∣ − 2 a ) , S = { w ∈ X : M ( w ) = 1 } , X=D_a^{1,2}(\mathbb{R}^N)\cap L^2(\mathbb{R}^N;|x|^{-2a}),\qquad S=\{w\in X:M(w)=1\}, X = D a 1 , 2 ( R N ) ∩ L 2 ( R N ; ∣ x ∣ − 2 a ) , S = { w ∈ X : M ( w ) = 1 } ,
where ∥ w ∥ X 2 = A ( w ) + M ( w ) \|w\|_X^2=A(w)+M(w) ∥ w ∥ X 2 = A ( w ) + M ( w ) . Unless otherwise specified, all spatial integrals are taken over R N \mathbb{R}^N R N .
For t ∈ R t\in\mathbb{R} t ∈ R and w ∈ X w\in X w ∈ X , define the mass-preserving dilation
( t ⋆ w ) ( x ) = e ( N − 2 a ) t / 2 w ( e t x ) . (14) (t\star w)(x)=e^{(N-2a)t/2}w(e^t x). \tag*{(14)} ( t ⋆ w ) ( x ) = e ( N − 2 a ) t /2 w ( e t x ) . (14)
A change of variables yields
M ( t ⋆ w ) = M ( w ) , A ( t ⋆ w ) = e 2 t A ( w ) , M(t\star w)=M(w),\qquad A(t\star w)=e^{2t}A(w), M ( t ⋆ w ) = M ( w ) , A ( t ⋆ w ) = e 2 t A ( w ) ,
B ( t ⋆ w ) = e q δ q t B ( w ) , C ( t ⋆ w ) = e p t C ( w ) . (15) B(t\star w)=e^{q\delta_qt}B(w),\qquad C(t\star w)=e^{pt}C(w). \tag*{(15)} B ( t ⋆ w ) = e q δ q t B ( w ) , C ( t ⋆ w ) = e pt C ( w ) . (15)
In particular, t ⋆ w ∈ S t\star w\in S t ⋆ w ∈ S whenever w ∈ S w\in S w ∈ S .
For comparison with the scalar formulation in [15 ] , the rescaling y = κ 1 p − 2 w y=\kappa^{\frac{1}{p-2}}w y = κ p − 2 1 w , β ^ = β κ − q − 2 p − 2 \widehat{\beta}=\beta\kappa^{-\frac{q-2}{p-2}} β = β κ − p − 2 q − 2 transforms (1.7) into an equivalent scalar problem with unit critical coefficient, prescribed mass M ( y ) = κ 2 p − 2 M(y)=\kappa^{\frac{2}{p-2}} M ( y ) = κ p − 2 2 , and the same Lagrange multiplier λ \lambda λ . We retain κ \kappa κ to make the synchronization compatibility conditions explicit.
For later use, we record the following ground-state transform, which relates the weighted space X X X to the standard Sobolev space.
Lemma 2.1 (Ground-state transform). The map w ↦ ∣ x ∣ − a w w\mapsto|x|^{-a}w w ↦ ∣ x ∣ − a w is an isomorphism from X X X onto H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) . For z = ∣ x ∣ − a w z=|x|^{-a}w z = ∣ x ∣ − a w , we have
M ( w ) = ∥ z ∥ 2 2 , A ( w ) = ∫ ∣ ∇ z ∣ 2 − μ a ∫ ∣ x ∣ − 2 z 2 , μ a = a ( N − 2 − a ) . (16) M(w)=\|z\|_2^2,\qquad A(w)=\int|\nabla z|^2-\mu_a\int|x|^{-2}z^2,\qquad\mu_a=a(N-2-a). \tag*{(16)} M ( w ) = ∥ z ∥ 2 2 , A ( w ) = ∫ ∣∇ z ∣ 2 − μ a ∫ ∣ x ∣ − 2 z 2 , μ a = a ( N − 2 − a ) . (16) Smooth functions compactly supported away from the origin are dense in X X X .
Proof. For any w ∈ C c ∞ ( R N ∖ { 0 } ) w\in C_c^\infty(\mathbb{R}^N\setminus\{0\}) w ∈ C c ∞ ( R N ∖ { 0 }) , using
div ( ∣ x ∣ − 2 a − 2 x ) = ( N − 2 − 2 a ) ∣ x ∣ − 2 a − 2 , \operatorname{div}(|x|^{-2a-2}x)=(N-2-2a)|x|^{-2a-2}, div ( ∣ x ∣ − 2 a − 2 x ) = ( N − 2 − 2 a ) ∣ x ∣ − 2 a − 2 ,
integration by parts and the Hölder inequality yield
( N − 2 − 2 a ) ∫ R N ∣ x ∣ − 2 a − 2 w 2 d x = − 2 ∫ R N ∣ x ∣ − 2 a − 2 w x ⋅ ∇ w d x (N-2-2a)\int_{\mathbb{R}^N}|x|^{-2a-2}w^2\,dx=-2\int_{\mathbb{R}^N}|x|^{-2a-2}w x\cdot\nabla w\,dx ( N − 2 − 2 a ) ∫ R N ∣ x ∣ − 2 a − 2 w 2 d x = − 2 ∫ R N ∣ x ∣ − 2 a − 2 w x ⋅ ∇ w d x
≤ 2 ( ∫ R N ∣ x ∣ − 2 a − 2 w 2 d x ) 1 / 2 A ( w ) 1 / 2 . \leq2\left(\int_{\mathbb{R}^N}|x|^{-2a-2}w^2\,dx\right)^{1/2}A(w)^{1/2}. ≤ 2 ( ∫ R N ∣ x ∣ − 2 a − 2 w 2 d x ) 1/2 A ( w ) 1/2 .
Since N − 2 − 2 a > 0 N-2-2a>0 N − 2 − 2 a > 0 , it follows that
∫ R N ∣ x ∣ − 2 a − 2 w 2 d x ≤ 4 ( N − 2 − 2 a ) 2 A ( w ) . (17) \int_{\mathbb{R}^N}|x|^{-2a-2}w^2\,dx\leq\frac{4}{(N-2-2a)^2}A(w). \tag*{(17)} ∫ R N ∣ x ∣ − 2 a − 2 w 2 d x ≤ ( N − 2 − 2 a ) 2 4 A ( w ) . (17)
For any φ ∈ C c ∞ ( R N ) \varphi\in C_c^\infty(\mathbb{R}^N) φ ∈ C c ∞ ( R N ) , let η ε \eta_\varepsilon η ε be a smooth cutoff satisfying η ε = 0 \eta_\varepsilon=0 η ε = 0 on B ε B_\varepsilon B ε , η ε = 1 \eta_\varepsilon=1 η ε = 1 outside B 2 ε B_{2\varepsilon} B 2 ε , and ∣ ∇ η ε ∣ ≤ C / ε |\nabla\eta_\varepsilon|\leq C/\varepsilon ∣∇ η ε ∣ ≤ C / ε . Then η ε φ ∈ C c ∞ ( R N ∖ { 0 } ) \eta_\varepsilon\varphi\in C_c^\infty(\mathbb{R}^N\setminus\{0\}) η ε φ ∈ C c ∞ ( R N ∖ { 0 }) , and
A ( η ε φ − φ ) ≤ C φ ( ε N − 2 a + ε N − 2 − 2 a ) ⟶ 0 as ε → 0 + . A(\eta_\varepsilon\varphi-\varphi)\leq C_\varphi\left(\varepsilon^{N-2a}+\varepsilon^{N-2-2a}\right)\longrightarrow0\quad\text{as }\varepsilon\to0^+. A ( η ε φ − φ ) ≤ C φ ( ε N − 2 a + ε N − 2 − 2 a ) ⟶ 0 as ε → 0 + .
since N − 2 − 2 a > 0 N-2-2a>0 N − 2 − 2 a > 0 . Consequently, C ∞ ( R N ∖ { 0 } ) C^\infty(\mathbb{R}^N\setminus\{0\}) C ∞ ( R N ∖ { 0 }) is dense in D a 1 , 2 ( R N ) D_a^{1,2}(\mathbb{R}^N) D a 1 , 2 ( R N ) , and (17) extends to all w ∈ D a 1 , 2 ( R N ) w\in D_a^{1,2}(\mathbb{R}^N) w ∈ D a 1 , 2 ( R N ) by continuity.
We next establish the density in X X X . For any w ∈ X w\in X w ∈ X , choose a smooth radial cutoff χ ε , R \chi_{\varepsilon,R} χ ε , R such that χ ε , R = 1 \chi_{\varepsilon,R}=1 χ ε , R = 1 on { 2 ε ≤ ∣ x ∣ ≤ R } \{2\varepsilon\leq|x|\leq R\} { 2 ε ≤ ∣ x ∣ ≤ R } , χ ε , R = 0 \chi_{\varepsilon,R}=0 χ ε , R = 0 on { ∣ x ∣ ≤ ε } ∪ { ∣ x ∣ ≥ 2 R } \{|x|\leq\varepsilon\}\cup\{|x|\geq2R\} { ∣ x ∣ ≤ ε } ∪ { ∣ x ∣ ≥ 2 R } , and ∣ ∇ χ ε , R ∣ ≤ C / ∣ x ∣ |\nabla\chi_{\varepsilon,R}|\leq C/|x| ∣∇ χ ε , R ∣ ≤ C /∣ x ∣ . By the product rule,
∥ w − χ ε , R w ∥ X 2 ≤ C ∫ { ∣ x ∣ < 2 ε } ∪ { ∣ x ∣ > R } ∣ x ∣ − 2 a ( ∣ ∇ w ∣ 2 + w 2 ) d x + C ∫ { ε < ∣ x ∣ < 2 ε } ∪ { R < ∣ x ∣ < 2 R } ∣ x ∣ − 2 a − 2 w 2 d x ⟶ 0 \|w-\chi_{\varepsilon,R}w\|_X^2\leq C\int_{\{|x|<2\varepsilon\}\cup\{|x|>R\}}|x|^{-2a}(|\nabla w|^2+w^2)\,dx+C\int_{\{\varepsilon<|x|<2\varepsilon\}\cup\{R<|x|<2R\}}|x|^{-2a-2}w^2\,dx\longrightarrow0 ∥ w − χ ε , R w ∥ X 2 ≤ C ∫ { ∣ x ∣ < 2 ε } ∪ { ∣ x ∣ > R } ∣ x ∣ − 2 a ( ∣∇ w ∣ 2 + w 2 ) d x + C ∫ { ε < ∣ x ∣ < 2 ε } ∪ { R < ∣ x ∣ < 2 R } ∣ x ∣ − 2 a − 2 w 2 d x ⟶ 0
as ε → 0 + \varepsilon\to0^+ ε → 0 + and R → ∞ R\to\infty R → ∞ . Indeed, the convergence follows from w ∈ X w\in X w ∈ X and the integrability of ∣ x ∣ − 2 a − 2 w 2 |x|^{-2a-2}w^2 ∣ x ∣ − 2 a − 2 w 2 ensured by (17) . For fixed ε \varepsilon ε and R R R , the weights are bounded above and below on the annulus containing the support of χ ε , R w \chi_{\varepsilon,R}w χ ε , R w . Standard mollification therefore yields an approximation in C ∞ ( R N ∖ { 0 } ) C^\infty(\mathbb{R}^N\setminus\{0\}) C ∞ ( R N ∖ { 0 }) converging to χ ε , R w \chi_{\varepsilon,R}w χ ε , R w in X X X . Hence C ∞ ( R N ∖ { 0 } ) C^\infty(\mathbb{R}^N\setminus\{0\}) C ∞ ( R N ∖ { 0 }) is dense in X X X .
For w ∈ C ∞ ( R N ∖ { 0 } ) w\in C^\infty(\mathbb{R}^N\setminus\{0\}) w ∈ C ∞ ( R N ∖ { 0 }) , setting z = ∣ x ∣ − a w z=|x|^{-a}w z = ∣ x ∣ − a w , a direct computation followed by integration by parts yields (16) . By the classical Hardy inequality,
∫ R N ∣ x ∣ − 2 z 2 d x ≤ 4 ( N − 2 ) 2 ∫ R N ∣ ∇ z ∣ 2 d x \int_{\mathbb{R}^N}|x|^{-2}z^2\,dx\leq\frac{4}{(N-2)^2}\int_{\mathbb{R}^N}|\nabla z|^2\,dx ∫ R N ∣ x ∣ − 2 z 2 d x ≤ ( N − 2 ) 2 4 ∫ R N ∣∇ z ∣ 2 d x
we obtain
( 1 − 4 μ a ( N − 2 ) 2 ) ∥ ∇ z ∥ 2 2 ≤ A ( w ) ≤ ∥ ∇ z ∥ 2 2 . \left(1-\frac{4\mu_a}{(N-2)^2}\right)\|\nabla z\|_2^2\leq A(w)\leq\|\nabla z\|_2^2. ( 1 − ( N − 2 ) 2 4 μ a ) ∥∇ z ∥ 2 2 ≤ A ( w ) ≤ ∥∇ z ∥ 2 2 .
Since
1 − 4 μ a ( N − 2 ) 2 = ( N − 2 − 2 a N − 2 ) 2 > 0 , 1-\frac{4\mu_a}{(N-2)^2}=\left(\frac{N-2-2a}{N-2}\right)^2>0, 1 − ( N − 2 ) 2 4 μ a = ( N − 2 N − 2 − 2 a ) 2 > 0 ,
the gradient norms are equivalent. Together with the mass identity M ( w ) = ∥ z ∥ 2 2 M(w)=\|z\|_2^2 M ( w ) = ∥ z ∥ 2 2 and the density established above, this shows that the transformation T : w ↦ ∣ x ∣ − a w T:w\mapsto|x|^{-a}w T : w ↦ ∣ x ∣ − a w extends to a continuous linear map from X X X into H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) .
To establish surjectivity, let z ∈ H 1 ( R N ) z\in H^1(\mathbb{R}^N) z ∈ H 1 ( R N ) . Since N ≥ 3 N\geq3 N ≥ 3 , C ∞ ( R N ∖ { 0 } ) C^\infty(\mathbb{R}^N\setminus\{0\}) C ∞ ( R N ∖ { 0 }) is dense in H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) . Thus, there exists a sequence { z n } ⊂ C ∞ ( R N ∖ { 0 } ) \{z_n\}\subset C^\infty(\mathbb{R}^N\setminus\{0\}) { z n } ⊂ C ∞ ( R N ∖ { 0 }) such that z n → z z_n\to z z n → z in H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) . Setting w n = ∣ x ∣ a z n w_n=|x|^a z_n w n = ∣ x ∣ a z n , the preceding estimates imply that { w n } \{w_n\} { w n } is Cauchy in X X X . Hence w n → w w_n\to w w n → w in X X X for some w ∈ X w\in X w ∈ X . Since
∥ w n − ∣ x ∣ a z ∥ L 2 ( R N ; ∣ x ∣ − 2 a ) = ∥ z n − z ∥ 2 ⟶ 0 , \|w_n-|x|^a z\|_{L^2(\mathbb{R}^N;|x|^{-2a})}=\|z_n-z\|_2\longrightarrow0, ∥ w n − ∣ x ∣ a z ∥ L 2 ( R N ; ∣ x ∣ − 2 a ) = ∥ z n − z ∥ 2 ⟶ 0 ,
we have w = ∣ x ∣ a z w=|x|^a z w = ∣ x ∣ a z , and consequently T w = z Tw=z T w = z . Moreover, the same estimates yield
∥ T − 1 z ∥ X 2 ≤ ∥ z ∥ H 1 ( R N ) 2 . \|T^{-1}z\|_X^2\leq\|z\|_{H^1(\mathbb{R}^N)}^2. ∥ T − 1 z ∥ X 2 ≤ ∥ z ∥ H 1 ( R N ) 2 .
Therefore, T T T is a topological isomorphism from X X X onto H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) .
The dilation (14) satisfies
∥ t ⋆ h ∥ X 2 = e 2 t A ( h ) + M ( h ) ≤ e 2 ∣ t ∣ ∥ h ∥ X 2 , t ∈ R , h ∈ X . (18) \|t\star h\|_X^2=e^{2t}A(h)+M(h)\leq e^{2|t|}\|h\|_X^2,\qquad t\in\mathbb{R},\ h\in X. \tag*{(18)} ∥ t ⋆ h ∥ X 2 = e 2 t A ( h ) + M ( h ) ≤ e 2∣ t ∣ ∥ h ∥ X 2 , t ∈ R , h ∈ X . (18)
By the density result in Lemma 2.1 and the estimate (18) , the map t ↦ t ⋆ h t\mapsto t\star h t ↦ t ⋆ h is continuous from R \mathbb{R} R into X X X for every h ∈ X h\in X h ∈ X .
Lemma 2.2. For every q ∈ ( 2 , p ] q\in(2,p] q ∈ ( 2 , p ] , there exists an optimal constant K q > 0 K_q>0 K q > 0 such that
∫ R N ∣ x ∣ − b q ∣ w ∣ q d x ≤ K q A ( w ) δ q 2 M ( w ) q ( 1 − δ q ) 2 , w ∈ X , (19) \int_{\mathbb{R}^N}|x|^{-bq}|w|^q\,dx\leq K_qA(w)^{\frac{\delta_q}{2}}M(w)^{\frac{q(1-\delta_q)}{2}},\qquad w\in X, \tag*{(19)} ∫ R N ∣ x ∣ − b q ∣ w ∣ q d x ≤ K q A ( w ) 2 δ q M ( w ) 2 q ( 1 − δ q ) , w ∈ X , (19)
where
δ q = ( N − 2 a + 2 b ) q − 2 N 2 q . \delta_q=\frac{(N-2a+2b)q-2N}{2q}. δ q = 2 q ( N − 2 a + 2 b ) q − 2 N .
In particular, δ p = 1 \delta_p=1 δ p = 1 , and (19) coincides with the sharp Caffarelli–Kohn–Nirenberg inequality (1) when q = p q=p q = p .
Proof. For q ∈ ( 2 , p ] q \in(2,p] q ∈ ( 2 , p ] ,
δ q = b − a + N ( 1 2 − 1 q ) , b − a < δ q ≤ 1. \delta_q=b-a+N\left(\frac{1}{2}-\frac{1}{q}\right), \qquad b-a<\delta_q\leq1. δ q = b − a + N ( 2 1 − q 1 ) , b − a < δ q ≤ 1.
Moreover,
1 q − b N = δ q ( 1 2 − a + 1 N ) + ( 1 − δ q ) ( 1 2 − a N ) . \frac{1}{q}-\frac{b}{N}=\delta_q\left(\frac{1}{2}-\frac{a+1}{N}\right)+(1-\delta_q)\left(\frac{1}{2}-\frac{a}{N}\right). q 1 − N b = δ q ( 2 1 − N a + 1 ) + ( 1 − δ q ) ( 2 1 − N a ) .
Setting σ q = − a − b − a δ q \sigma_q=-a-\frac{b-a}{\delta_q} σ q = − a − δ q b − a , we obtain
− b = δ q σ q + ( 1 − δ q ) ( − a ) , 0 < − a − σ q = b − a δ q < 1. -b=\delta_q\sigma_q+(1-\delta_q)(-a), \qquad0<-a-\sigma_q=\frac{b-a}{\delta_q}<1. − b = δ q σ q + ( 1 − δ q ) ( − a ) , 0 < − a − σ q = δ q b − a < 1.
Since a < ( N − 2 ) / 2 a<(N-2)/2 a < ( N − 2 ) /2 and b q ≤ b p < N bq\leq bp<N b q ≤ b p < N , the Caffarelli–Kohn–Nirenberg inequality [5 ] yields
∥ ∣ x ∣ − b w ∥ q ≤ C q ∥ ∣ x ∣ − a ∇ w ∥ 2 δ q ∥ ∣ x ∣ − a w ∥ 2 1 − δ q . (20) \left\lVert|x|^{-b}w\right\rVert_q\leq C_q\left\lVert|x|^{-a}\nabla w\right\rVert_2^{\delta_q}\left\lVert|x|^{-a}w\right\rVert_2^{1-\delta_q}. \tag*{(20)} ∣ x ∣ − b w q ≤ C q ∣ x ∣ − a ∇ w 2 δ q ∣ x ∣ − a w 2 1 − δ q . (20)
for all w ∈ C c ∞ ( R N ) w\in C^\infty_c(\mathbb{R}^N) w ∈ C c ∞ ( R N ) . Raising both sides to the power q q q and using the density of C c ∞ ( R N ) C^\infty_c(\mathbb{R}^N) C c ∞ ( R N ) in X X X , we obtain (2.6). The optimal constant K q K_q K q is therefore positive and finite. □
The functionals B B B and C C C are of class C 1 C^1 C 1 on X X X . Indeed, for r > 2 r>2 r > 2 and real s , t s,t s , t ,
∣ ∣ s ∣ r − 2 s − ∣ t ∣ r − 2 t ∣ ≤ ( r − 1 ) ( ∣ s ∣ + ∣ t ∣ ) r − 2 ∣ s − t ∣ . \left||s|^{r-2}s-|t|^{r-2}t\right|\leq(r-1)(|s|+|t|)^{r-2}|s-t|. ∣ s ∣ r − 2 s − ∣ t ∣ r − 2 t ≤ ( r − 1 ) ( ∣ s ∣ + ∣ t ∣ ) r − 2 ∣ s − t ∣.
Applying this scalar inequality to ∣ x ∣ − b w |x|^{-b}w ∣ x ∣ − b w and using the Hölder inequality and Lemma 2.2, we find that the derivatives of B B B and C C C are locally Lipschitz from X X X to its dual. In particular, A , B , C A,B,C A , B , C are uniformly Lipschitz on bounded subsets of X X X .
Proposition 2.1. For every q ∈ ( 2 , p ) q\in(2,p) q ∈ ( 2 , p ) , the embedding
X ↪ L q ( R N ; ∣ x ∣ − b q ) X\hookrightarrow L^q(\mathbb{R}^N;|x|^{-bq}) X ↪ L q ( R N ; ∣ x ∣ − b q ) is compact.
Proof. Let w n ⇀ w w_n\rightharpoonup w w n ⇀ w in X X X . Put z n = w n − w z_n=w_n-w z n = w n − w . Fix 0 < R 0 < R 1 0<R_0<R_1 0 < R 0 < R 1 . The weights are bounded above and below on
Ω R 0 , R 1 = { x : R 0 < ∣ x ∣ < R 1 } . \Omega_{R_0,R_1}=\{x:R_0<|x|<R_1\}. Ω R 0 , R 1 = { x : R 0 < ∣ x ∣ < R 1 } .
Hence z n ⇀ 0 z_n\rightharpoonup0 z n ⇀ 0 in H 1 ( Ω R 0 , R 1 ) H^1(\Omega_{R_0,R_1}) H 1 ( Ω R 0 , R 1 ) . Since p < 2 ∗ = 2 N N − 2 p<2^*=\frac{2N}{N-2} p < 2 ∗ = N − 2 2 N , Rellich’s theorem gives
∫ Ω R 0 , R 1 ∣ x ∣ − b q ∣ z n ∣ q d x ⟶ 0. (21) \int_{\Omega_{R_0,R_1}}|x|^{-bq}|z_n|^q\,dx\longrightarrow0. \tag*{(21)} ∫ Ω R 0 , R 1 ∣ x ∣ − b q ∣ z n ∣ q d x ⟶ 0. (21)
For 0 < R < 1 0<R<1 0 < R < 1 , the Hölder inequality and (1.1) give
∫ ∣ x ∣ < R ∣ x ∣ − b q ∣ z n ∣ q d x = ∫ ∣ x ∣ < R ( ∣ x ∣ − b p ∣ z n ∣ p ) q p d x ≤ ∣ B R ∣ 1 − q p C ( z n ) q p ≤ C R N ( 1 − q p ) . (22) \int_{|x|<R}|x|^{-bq}|z_n|^q\,dx=\int_{|x|<R}\left(|x|^{-bp}|z_n|^p\right)^{\frac{q}{p}}dx\leq|B_R|^{1-\frac{q}{p}}C(z_n)^{\frac{q}{p}}\leq CR^{N\left(1-\frac{q}{p}\right)}. \tag*{(22)} ∫ ∣ x ∣ < R ∣ x ∣ − b q ∣ z n ∣ q d x = ∫ ∣ x ∣ < R ( ∣ x ∣ − b p ∣ z n ∣ p ) p q d x ≤ ∣ B R ∣ 1 − p q C ( z n ) p q ≤ C R N ( 1 − p q ) . (22)
Next, choose ϑ ∈ ( 0 , 1 ) \vartheta\in(0,1) ϑ ∈ ( 0 , 1 ) such that 1 q = ϑ 2 + 1 − ϑ p \frac{1}{q}=\frac{\vartheta}{2}+\frac{1-\vartheta}{p} q 1 = 2 ϑ + p 1 − ϑ . For R > 1 R>1 R > 1 , the Hölder inequality yields
∫ ∣ x ∣ > R ∣ x ∣ − b q ∣ z n ∣ q d x ≤ R − ϑ q ( b − a ) M ( z n ) ϑ q 2 C ( z n ) ( 1 − ϑ ) q p ≤ C R − ϑ q ( b − a ) . (23) \int_{|x|>R}|x|^{-bq}|z_n|^q\,dx\leq R^{-\vartheta q(b-a)}M(z_n)^{\frac{\vartheta q}{2}}C(z_n)^{\frac{(1-\vartheta)q}{p}}\leq CR^{-\vartheta q(b-a)}. \tag*{(23)} ∫ ∣ x ∣ > R ∣ x ∣ − b q ∣ z n ∣ q d x ≤ R − ϑq ( b − a ) M ( z n ) 2 ϑq C ( z n ) p ( 1 − ϑ ) q ≤ C R − ϑq ( b − a ) . (23)
The bounds in (2.9) and (2.10) are uniform in n n n . Letting R 0 → 0 + R_0\to0^+ R 0 → 0 + and R 1 → ∞ R_1\to\infty R 1 → ∞ , we conclude that
w n ⟶ w in L q ( R N ; ∣ x ∣ − b q ) . w_n\longrightarrow w\quad\text{in }L^q(\mathbb{R}^N;|x|^{-bq}). w n ⟶ w in L q ( R N ; ∣ x ∣ − b q ) .
□
The following lemma provides a sharp CKN extremal with finite weighted mass. This property is essential for the subsequent critical energy estimates.
Lemma 2.3. Let N ≥ 5 N \ge5 N ≥ 5 , 0 ≤ a < N − 4 2 0 \le a < \frac{N-4}{2} 0 ≤ a < 2 N − 4 , a < b < a + 1 a < b < a+1 a < b < a + 1 . Then, the sharp constant S ( a , b ) S(a,b) S ( a , b ) is attained by a positive radial function W ∈ X W \in X W ∈ X . Up to multiplication and dilation, this function has the form
W ( x ) = c 0 ( 1 + ∣ x ∣ σ ) − m , σ = 2 d ( N − 2 − 2 a ) N − 2 d , m = N − 2 d 2 d . (24) W(x) = c_0(1+|x|^\sigma)^{-m}, \qquad\sigma= \frac{2d(N-2-2a)}{N-2d}, \qquad m = \frac{N-2d}{2d}. \tag*{(24)} W ( x ) = c 0 ( 1 + ∣ x ∣ σ ) − m , σ = N − 2 d 2 d ( N − 2 − 2 a ) , m = 2 d N − 2 d . (24) In particular,
A ( W ) C ( W ) 2 / p = S ( a , b ) , M ( W ) < ∞ . \frac{A(W)}{C(W)^{2/p}} = S(a,b), \qquad M(W) < \infty. C ( W ) 2/ p A ( W ) = S ( a , b ) , M ( W ) < ∞. Proof. The existence and radiality follow from the sharp Caffarelli–Kohn–Nirenberg theory in the range a ≥ 0 a \ge0 a ≥ 0 ; see [7 ] . The Emden–Fowler reduction gives (24) . The profile is bounded near the origin. At infinity,
W ( x ) ≍ ∣ x ∣ − ( N − 2 − 2 a ) . W(x) \asymp|x|^{-(N-2-2a)}. W ( x ) ≍ ∣ x ∣ − ( N − 2 − 2 a ) . Thus the weighted mass density is comparable to
r N − 1 − 2 a r − 2 ( N − 2 − 2 a ) = r − N + 3 + 2 a . r^{N-1-2a}r^{-2(N-2-2a)} = r^{-N+3+2a}. r N − 1 − 2 a r − 2 ( N − 2 − 2 a ) = r − N + 3 + 2 a . It is integrable at infinity exactly when a < ( N − 4 ) / 2 a < (N-4)/2 a < ( N − 4 ) /2 . Hence W ∈ X W \in X W ∈ X .
For the scalar problem (7) , we define the critical energy threshold
T κ = d N S ( a , b ) N 2 d κ − N − 2 d 2 d . T_\kappa= \frac{d}{N}S(a,b)^{\frac{N}{2d}}\kappa^{-\frac{N-2d}{2d}}. T κ = N d S ( a , b ) 2 d N κ − 2 d N − 2 d .
and the energy functional
J ( w ) = 1 2 A ( w ) − β q B ( w ) − κ p C ( w ) . J(w) = \frac{1}{2}A(w) - \frac{\beta}{q}B(w) - \frac{\kappa}{p}C(w). J ( w ) = 2 1 A ( w ) − q β B ( w ) − p κ C ( w ) .
The corresponding Pohozaev functional is
P ( w ) = d d t J ( t ⋆ w ) ∣ t = 0 = A ( w ) − δ q β B ( w ) − κ C ( w ) , P(w) = \left.\frac{d}{dt}J(t\star w)\right|_{t=0} = A(w) - \delta_q\beta B(w) - \kappa C(w), P ( w ) = d t d J ( t ⋆ w ) t = 0 = A ( w ) − δ q β B ( w ) − κ C ( w ) ,
and the scalar Pohozaev set is
P = { w ∈ S : P ( w ) = 0 } . \mathcal{P} = \{w \in\mathcal{S}: P(w) = 0\}. P = { w ∈ S : P ( w ) = 0 } .
The energy functional of the original coupled system is
E ( u , v ) = 1 2 ∫ R N ∣ x ∣ − 2 a ( ∣ ∇ u ∣ 2 + ∣ ∇ v ∣ 2 ) d x − 1 q ∫ R N ∣ x ∣ − b q ( β 1 ∣ u ∣ q + β 2 ∣ v ∣ q ) d x − 1 p ∫ R N ∣ x ∣ − b p ( μ 1 ∣ u ∣ p + μ 2 ∣ v ∣ p + p ν ∣ u ∣ α ∣ v ∣ γ ) d x . (25) E(u,v) = \frac{1}{2}\int_{\mathbb{R}^N}|x|^{-2a}\left(|\nabla u|^2+|\nabla v|^2\right)\,dx-\frac{1}{q}\int_{\mathbb{R}^N}|x|^{-bq}\left(\beta_1|u|^q+\beta_2|v|^q\right)\,dx-\frac{1}{p}\int_{\mathbb{R}^N}|x|^{-bp}\left(\mu_1|u|^p+\mu_2|v|^p+p\nu|u|^\alpha|v|^\gamma\right)\,dx. \tag*{(25)} E ( u , v ) = 2 1 ∫ R N ∣ x ∣ − 2 a ( ∣∇ u ∣ 2 + ∣∇ v ∣ 2 ) d x − q 1 ∫ R N ∣ x ∣ − b q ( β 1 ∣ u ∣ q + β 2 ∣ v ∣ q ) d x − p 1 ∫ R N ∣ x ∣ − b p ( μ 1 ∣ u ∣ p + μ 2 ∣ v ∣ p + p ν ∣ u ∣ α ∣ v ∣ γ ) d x . (25)
We denote the mixed nonlinear term by
D ( u , v ) = ∫ R N ∣ x ∣ − b p ∣ u ∣ α ∣ v ∣ γ d x . D(u,v)=\int_{\mathbb{R}^N}|x|^{-bp}|u|^\alpha|v|^\gamma\,dx. D ( u , v ) = ∫ R N ∣ x ∣ − b p ∣ u ∣ α ∣ v ∣ γ d x .
By the Hölder inequality, D ( u , v ) ≤ C ( u ) α / p C ( v ) γ / p D(u,v) \le C(u)^{\alpha/p}C(v)^{\gamma/p} D ( u , v ) ≤ C ( u ) α / p C ( v ) γ / p . Since α , γ > 1 \alpha,\gamma>1 α , γ > 1 , the weighted Sobolev embeddings imply that E ∈ C 1 ( X 2 , R ) E \in C^1(X^2,\mathbb{R}) E ∈ C 1 ( X 2 , R ) . The corresponding mass constraint and Pohozaev set are defined by
S v e c = { ( u , v ) ∈ X 2 : M ( u ) = ρ 1 2 , M ( v ) = ρ 2 2 } , \mathcal{S}_{\mathrm{vec}}=\{(u,v)\in X^2:M(u)=\rho_1^2,\ M(v)=\rho_2^2\}, S vec = {( u , v ) ∈ X 2 : M ( u ) = ρ 1 2 , M ( v ) = ρ 2 2 } ,
and
P v e c = { ( u , v ) ∈ S v e c : d d t E ( t ⋆ u , t ⋆ v ) ∣ t = 0 = 0 } . \mathcal{P}_{\mathrm{vec}}=\left\{(u,v)\in\mathcal{S}_{\mathrm{vec}}:\left.\frac{d}{dt}E(t\star u,t\star v)\right|_{t=0}=0\right\}. P vec = { ( u , v ) ∈ S vec : d t d E ( t ⋆ u , t ⋆ v ) t = 0 = 0 } .
respectively. Finally, we introduce the synchronized mass constraint and Pohozaev set
S s y n = { ( ρ 1 w , ρ 2 w ) : w ∈ S } , P s y n = { ( ρ 1 w , ρ 2 w ) : w ∈ P } . (26) \mathcal{S}_{\mathrm{syn}}=\{(\rho_1w,\rho_2w):w\in\mathcal{S}\},\qquad\mathcal{P}_{\mathrm{syn}}=\{(\rho_1w,\rho_2w):w\in\mathcal{P}\}. \tag*{(26)} S syn = {( ρ 1 w , ρ 2 w ) : w ∈ S } , P syn = {( ρ 1 w , ρ 2 w ) : w ∈ P } . (26)
In the mass-subcritical regime, we further define
P s y n − = { ( ρ 1 w , ρ 2 w ) : w ∈ P − } , \mathcal{P}_{\mathrm{syn}}^{-}=\{(\rho_1w,\rho_2w):w\in\mathcal{P}^{-}\}, P syn − = {( ρ 1 w , ρ 2 w ) : w ∈ P − } ,
where P − \mathcal{P}^{-} P − denotes the upper scalar Pohozaev branch characterized in Lemma 3.1. Set
m v e c = inf P v e c E , m s y n = inf P s y n E . m_{\mathrm{vec}}=\inf_{\mathcal{P}_{\mathrm{vec}}}E,\qquad m_{\mathrm{syn}}=\inf_{\mathcal{P}_{\mathrm{syn}}}E. m vec = P vec inf E , m syn = P syn inf E .
For χ > 0 \chi>0 χ > 0 , write
J χ = 1 2 A − β q B − χ p C , P χ = { w ∈ S : A ( w ) − δ q β B ( w ) − χ C ( w ) = 0 } , J_{\chi}=\frac{1}{2}A-\frac{\beta}{q}B-\frac{\chi}{p}C,\qquad\mathcal{P}_{\chi}=\{w\in\mathcal{S}:A(w)-\delta_q\beta B(w)-\chi C(w)=0\}, J χ = 2 1 A − q β B − p χ C , P χ = { w ∈ S : A ( w ) − δ q β B ( w ) − χ C ( w ) = 0 } ,
and let m χ = inf P χ J χ m_{\chi}=\inf_{\mathcal{P}_{\chi}}J_{\chi} m χ = inf P χ J χ .
Proposition 2.2. Assume that (1.5) and (1.6) hold. If w ∈ S w\in\mathcal{S} w ∈ S is a constrained critical point of J J J , then
( u , v ) = ( ρ 1 w , ρ 2 w ) (u,v)=(\rho_1w,\rho_2w) ( u , v ) = ( ρ 1 w , ρ 2 w ) is a normalized solution of (1.2) with λ 1 = λ 2 = λ \lambda_1=\lambda_2=\lambda λ 1 = λ 2 = λ . Moreover,
E ( ρ 1 w , ρ 2 w ) = ( ρ 1 2 + ρ 2 2 ) J ( w ) . (27) E(\rho_1w,\rho_2w)=(\rho_1^2+\rho_2^2)J(w). \tag*{(27)} E ( ρ 1 w , ρ 2 w ) = ( ρ 1 2 + ρ 2 2 ) J ( w ) . (27)
Proof. Since w w w is a constrained critical point of J J J on S \mathcal{S} S , it satisfies (1.7) for some λ ∈ R \lambda\in\mathbb{R} λ ∈ R . The compatibility conditions (1.5) and (1.6) give
β ρ i = β i ρ i q − 1 , i = 1 , 2 , \beta\rho_i=\beta_i\rho_i^{q-1},\qquad i=1,2, β ρ i = β i ρ i q − 1 , i = 1 , 2 ,
and
κ ρ 1 = μ 1 ρ 1 p − 1 + ν α ρ 1 α − 1 ρ 2 γ , κ ρ 2 = μ 2 ρ 2 p − 1 + ν γ ρ 1 α ρ 2 γ − 1 . \kappa\rho_1=\mu_1\rho_1^{p-1}+\nu\alpha\rho_1^{\alpha-1}\rho_2^\gamma,\qquad
\kappa\rho_2=\mu_2\rho_2^{p-1}+\nu\gamma\rho_1^\alpha\rho_2^{\gamma-1}. κ ρ 1 = μ 1 ρ 1 p − 1 + ν α ρ 1 α − 1 ρ 2 γ , κ ρ 2 = μ 2 ρ 2 p − 1 + ν γ ρ 1 α ρ 2 γ − 1 .
Multiplying (1.7) by ρ 1 \rho_1 ρ 1 and ρ 2 \rho_2 ρ 2 , respectively, and using these identities, we obtain (1.2) with λ 1 = λ 2 = λ \lambda_1=\lambda_2=\lambda λ 1 = λ 2 = λ . The prescribed mass constraints follow from M ( w ) = 1 M(w)=1 M ( w ) = 1 .
To prove (2.14), set R = ρ 1 2 + ρ 2 2 R=\rho_1^2+\rho_2^2 R = ρ 1 2 + ρ 2 2 . The compatibility conditions yield
β 1 ρ 1 q + β 2 ρ 2 q = β R \beta_1\rho_1^q+\beta_2\rho_2^q=\beta R β 1 ρ 1 q + β 2 ρ 2 q = β R
and, since α + γ = p \alpha+\gamma=p α + γ = p ,
μ 1 ρ 1 p + μ 2 ρ 2 p + p ν ρ 1 α ρ 2 γ = κ R . \mu_1\rho_1^p+\mu_2\rho_2^p+p\nu\rho_1^\alpha\rho_2^\gamma=\kappa R. μ 1 ρ 1 p + μ 2 ρ 2 p + p ν ρ 1 α ρ 2 γ = κ R .
Consequently, by (2.12),
E ( ρ 1 w , ρ 2 w ) = R 2 A ( w ) − β R q B ( w ) − κ R p C ( w ) = R J ( w ) . E(\rho_1w,\rho_2w)=\frac{R}{2}A(w)-\frac{\beta R}{q}B(w)-\frac{\kappa R}{p}C(w)=RJ(w). E ( ρ 1 w , ρ 2 w ) = 2 R A ( w ) − q β R B ( w ) − p κ R C ( w ) = R J ( w ) .
The Pohozaev functional associated with the common mass-preserving dilation is given by
P v e c ( u , v ) = A ( u ) + A ( v ) − δ q ( β 1 B ( u ) + β 2 B ( v ) ) − μ 1 C ( u ) − μ 2 C ( v ) − p ν D ( u , v ) . P_{\mathrm{vec}}(u,v)=A(u)+A(v)-\delta_q\left(\beta_1B(u)+\beta_2B(v)\right)-\mu_1C(u)-\mu_2C(v)-p\nu D(u,v). P vec ( u , v ) = A ( u ) + A ( v ) − δ q ( β 1 B ( u ) + β 2 B ( v ) ) − μ 1 C ( u ) − μ 2 C ( v ) − p ν D ( u , v ) .
Indeed,
P v e c ( u , v ) = d d t E ( t ⋆ u , t ⋆ v ) ∣ t = 0 . P_{\mathrm{vec}}(u,v)=\left.\frac{d}{dt}E(t\star u,t\star v)\right|_{t=0}. P vec ( u , v ) = d t d E ( t ⋆ u , t ⋆ v ) t = 0 .
By the compatibility conditions and (2.14), we obtain
P v e c ( ρ 1 w , ρ 2 w ) = ( ρ 1 2 + ρ 2 2 ) P ( w ) , w ∈ S . P_{\mathrm{vec}}(\rho_1w,\rho_2w)=(\rho_1^2+\rho_2^2)P(w),\qquad w\in S. P vec ( ρ 1 w , ρ 2 w ) = ( ρ 1 2 + ρ 2 2 ) P ( w ) , w ∈ S .
Consequently,
P s y n = P v e c ∩ S s y n . P_{\mathrm{syn}}=P_{\mathrm{vec}}\cap S_{\mathrm{syn}}. P syn = P vec ∩ S syn .
Moreover, (2.14) shows that minimization of J J J over P \mathcal{P} P is equivalent to minimization of E E E over P s y n \mathcal{P}_{\mathrm{syn}} P syn . This correspondence is restricted to the synchronized class and does not characterize general vector solutions.
Fiber maps and constrained Palais–Smale sequences
In this section, we use the framework of Section 2 to establish the fiber geometry in Lemmas 3.1 and 3.2. We then develop the tools for constructing constrained Palais–Smale sequences.
For w ∈ S w\in S w ∈ S , the mass-preserving dilation (2.1) and (2.2) give the fiber map
Φ w ( t ) = J ( t ⋆ w ) = 1 2 e 2 t A ( w ) − β q e q δ q t B ( w ) − κ p e p t C ( w ) . \Phi_w(t)=J(t\star w)=\frac{1}{2}e^{2t}A(w)-\frac{\beta}{q}e^{q\delta_qt}B(w)-\frac{\kappa}{p}e^{pt}C(w). Φ w ( t ) = J ( t ⋆ w ) = 2 1 e 2 t A ( w ) − q β e q δ q t B ( w ) − p κ e pt C ( w ) .
Writing θ = q δ q \theta=q\delta_q θ = q δ q , its derivatives are
Φ w ′ ( t ) = e 2 t A ( w ) − δ q β e θ t B ( w ) − κ e p t C ( w ) . (28) \Phi'_w(t)=e^{2t}A(w)-\delta_q\beta e^{\theta t}B(w)-\kappa e^{pt}C(w). \tag*{(28)} Φ w ′ ( t ) = e 2 t A ( w ) − δ q β e θ t B ( w ) − κ e pt C ( w ) . (28)
and
Φ w ′ ′ ( t ) = 2 e 2 t A ( w ) − q δ q 2 β e θ t B ( w ) − p κ e p t C ( w ) . \Phi''_w(t)=2e^{2t}A(w)-q\delta_q^2\beta e^{\theta t}B(w)-p\kappa e^{pt}C(w). Φ w ′′ ( t ) = 2 e 2 t A ( w ) − q δ q 2 β e θ t B ( w ) − p κ e pt C ( w ) .
If θ < 2 \theta<2 θ < 2 , we partition the Pohozaev set P \mathcal{P} P according to the sign of Φ w ′ ′ ( 0 ) \Phi''_w(0) Φ w ′′ ( 0 ) as
P = P + ∪ P − ∪ P 0 , \mathcal{P}=\mathcal{P}^{+}\cup\mathcal{P}^{-}\cup\mathcal{P}^{0}, P = P + ∪ P − ∪ P 0 ,
where
P + = { w ∈ P : Φ w ′ ′ ( 0 ) > 0 } , \mathcal{P}^{+}=\{w\in\mathcal{P}:\Phi''_w(0)>0\}, P + = { w ∈ P : Φ w ′′ ( 0 ) > 0 } ,
P − = { w ∈ P : Φ w ′ ′ ( 0 ) < 0 } , \mathcal{P}^{-}=\{w\in\mathcal{P}:\Phi''_w(0)<0\}, P − = { w ∈ P : Φ w ′′ ( 0 ) < 0 } ,
P 0 = { w ∈ P : Φ w ′ ′ ( 0 ) = 0 } . \mathcal{P}^{0}=\{w\in\mathcal{P}:\Phi''_w(0)=0\}. P 0 = { w ∈ P : Φ w ′′ ( 0 ) = 0 } .
On P \mathcal{P} P ,
J ( w ) = d N A ( w ) − ( 1 q − δ q p ) β B ( w ) . (29) J(w)=\frac{d}{N}A(w)-\left(\frac{1}{q}-\frac{\delta_q}{p}\right)\beta B(w). \tag*{(29)} J ( w ) = N d A ( w ) − ( q 1 − p δ q ) β B ( w ) . (29)
and
J ( w ) = ( δ q 2 − 1 q ) β B ( w ) + d N κ C ( w ) . (30) J(w)=\left(\frac{\delta_q}{2}-\frac{1}{q}\right)\beta B(w)+\frac{d}{N}\kappa C(w). \tag*{(30)} J ( w ) = ( 2 δ q − q 1 ) β B ( w ) + N d κ C ( w ) . (30)
For s ≥ 0 s\geq0 s ≥ 0 , define
f β ( s ) = 1 2 s 2 − β K q q s θ − κ p S ( a , b ) p / 2 s p . f_{\beta}(s)=\frac{1}{2}s^2-\frac{\beta K_q}{q}s^\theta-\frac{\kappa}{pS(a,b)^{p/2}}s^p. f β ( s ) = 2 1 s 2 − q β K q s θ − pS ( a , b ) p /2 κ s p .
In the mass-subcritical case, 0 < θ < 2 0 < \theta< 2 0 < θ < 2 . To specify the smallness constant used below, set
c ∗ = p − 2 p − θ ( ( 2 − θ ) S ( a , b ) p / 2 ( p − θ ) κ ) 2 − θ p − 2 . c_* = \frac{p-2}{p-\theta}\left(\frac{(2-\theta)S(a,b)^{p/2}}{(p-\theta)\kappa}\right)^{\frac{2-\theta}{p-2}}. c ∗ = p − θ p − 2 ( ( p − θ ) κ ( 2 − θ ) S ( a , b ) p /2 ) p − 2 2 − θ .
and
β f = c ∗ 2 δ q K q . \beta_f = \frac{c_*}{2\delta_qK_q}. β f = 2 δ q K q c ∗ .
Choose
R ∗ = 1 2 ( p S ( a , b ) p / 2 8 κ ) 1 / ( p − 2 ) (31) R_* = \frac{1}{2}\left(\frac{pS(a,b)^{p/2}}{8\kappa}\right)^{1/(p-2)} \tag*{(31)} R ∗ = 2 1 ( 8 κ pS ( a , b ) p /2 ) 1/ ( p − 2 ) (31)
and set
β b = q 8 K q R ∗ 2 − θ . (32) \beta_b = \frac{q}{8K_q}R_*^{2-\theta}. \tag*{(32)} β b = 8 K q q R ∗ 2 − θ . (32)
We use the constant
β 0 = min { β f , β b } . \beta_0 = \min\{\beta_f,\beta_b\}. β 0 = min { β f , β b } .
For the mass-critical case, assumption (1.9) ensures that
c β : = 1 − δ q β K q > 0. c_\beta:= 1-\delta_q\beta K_q>0. c β := 1 − δ q β K q > 0.
Lemma 3.1. Let 2 < q < q c 2 < q < q_c 2 < q < q c and 0 < β < β 0 0 < \beta< \beta_0 0 < β < β 0 . Then P 0 = ∅ \mathcal{P}^0 = \varnothing P 0 = ∅ . Each fiber map has exactly two critical points
t 1 ( w ) < t 2 ( w ) . t_1(w)<t_2(w). t 1 ( w ) < t 2 ( w ) . The first is a strict local minimum with negative energy, and the second is a strict global maximum with positive energy. Moreover,
t 1 ( w ) ⋆ w ∈ P + , t 2 ( w ) ⋆ w ∈ P − . t_1(w)\star w\in\mathcal{P}^+,\qquad t_2(w)\star w\in\mathcal{P}^-. t 1 ( w ) ⋆ w ∈ P + , t 2 ( w ) ⋆ w ∈ P − . Both maps w ↦ t i ( w ) w\mapsto t_i(w) w ↦ t i ( w ) are of class C 1 C^1 C 1 .
Proof. Put s = e t s=e^t s = e t . Equation (3.1) is equivalent to
g ( s ) = δ q β B ( w ) , g ( s ) = A ( w ) s 2 − θ − κ C ( w ) s p − θ . (33) g(s)=\delta_q\beta B(w),\qquad g(s)=A(w)s^{2-\theta}-\kappa C(w)s^{p-\theta}. \tag*{(33)} g ( s ) = δ q β B ( w ) , g ( s ) = A ( w ) s 2 − θ − κ C ( w ) s p − θ . (33)
Direct differentiation shows that g g g has a unique maximum, given by
max s > 0 g ( s ) = p − 2 p − θ A ( w ) ( ( 2 − θ ) A ( w ) ( p − θ ) κ C ( w ) ) 2 − θ p − 2 . \max_{s>0}g(s)=\frac{p-2}{p-\theta}A(w)\left(\frac{(2-\theta)A(w)}{(p-\theta)\kappa C(w)}\right)^{\frac{2-\theta}{p-2}}. s > 0 max g ( s ) = p − θ p − 2 A ( w ) ( ( p − θ ) κ C ( w ) ( 2 − θ ) A ( w ) ) p − 2 2 − θ .
By (1.1),
C ( w ) ≤ S ( a , b ) − p / 2 A ( w ) p / 2 . C(w)\le S(a,b)^{-p/2}A(w)^{p/2}. C ( w ) ≤ S ( a , b ) − p /2 A ( w ) p /2 .
Therefore
max s > 0 g ( s ) ≥ c ∗ A ( w ) θ / 2 . \max_{s>0}g(s)\ge c_*A(w)^{\theta/2}. s > 0 max g ( s ) ≥ c ∗ A ( w ) θ /2 .
Since M ( w ) = 1 M(w)=1 M ( w ) = 1 , (2.6) gives
δ q β B ( w ) ≤ δ q β K q A ( w ) θ / 2 < 1 2 c ∗ A ( w ) θ / 2 . (34) \delta_q\beta B(w)\le\delta_q\beta K_qA(w)^{\theta/2}<\frac{1}{2}c_*A(w)^{\theta/2}. \tag*{(34)} δ q β B ( w ) ≤ δ q β K q A ( w ) θ /2 < 2 1 c ∗ A ( w ) θ /2 . (34)
Thus (3.6) has exactly two positive roots. The sign pattern of Φ w \Phi_w Φ w gives the local minimum and the global maximum. Since θ < 2 \theta<2 θ < 2 ,
Φ w ( t ) = − β q e θ t B ( w ) + o ( e θ t ) < 0 as t → − ∞ . \Phi_w(t)=-\frac{\beta}{q}e^{\theta t}B(w)+o(e^{\theta t})<0\quad\text{as }t\to-\infty. Φ w ( t ) = − q β e θ t B ( w ) + o ( e θ t ) < 0 as t → − ∞.
Hence the first critical value is negative. For the second value, applying (19) and (1) yields
Φ w ( t ) ≥ 1 2 y 2 − β K q q y θ − κ p S ( a , b ) p / 2 y p , y = e t A ( w ) 1 / 2 . \Phi_w(t) \ge\frac{1}{2}y^2-\frac{\beta K_q}{q}y^\theta-\frac{\kappa}{pS(a,b)^{p/2}}y^p,\qquad y=e^tA(w)^{1/2}. Φ w ( t ) ≥ 2 1 y 2 − q β K q y θ − pS ( a , b ) p /2 κ y p , y = e t A ( w ) 1/2 .
Taking y = R ∗ y=R_* y = R ∗ and using (32) , we obtain
Φ w ( t R ) ≥ f β ( R ∗ ) ≥ 1 4 R ∗ 2 > 0 , e t R A ( w ) 1 / 2 = R ∗ . (35) \Phi_w(t_R) \ge f_\beta(R_*) \ge\frac{1}{4}R_*^2>0,\qquad e^{t_R}A(w)^{1/2}=R_*. \tag*{(35)} Φ w ( t R ) ≥ f β ( R ∗ ) ≥ 4 1 R ∗ 2 > 0 , e t R A ( w ) 1/2 = R ∗ . (35)
Thus the global maximum is positive. The strict inequality in (34) rules out double roots of (33) , and hence P 0 = ∅ \mathcal{P}^0=\varnothing P 0 = ∅ . The final assertion follows from the implicit function theorem.
Lemma 3.2. Let q c ≤ q < p q_c\le q<p q c ≤ q < p . If q = q c q=q_c q = q c , assume in addition that (9) holds. Then, for every w ∈ S w\in S w ∈ S , the fiber map Φ w \Phi_w Φ w has exactly one critical point t ∗ ( w ) t_*(w) t ∗ ( w ) , which is a strict global maximum. Moreover, the map w ↦ t ∗ ( w ) w\mapsto t_*(w) w ↦ t ∗ ( w ) is of class C 1 C^1 C 1 on S S S .
Proof. If q > q c q>q_c q > q c , then θ > 2 \theta>2 θ > 2 . Dividing (28) by e 2 t e^{2t} e 2 t gives
A ( w ) = δ q β e ( θ − 2 ) t B ( w ) + κ e ( p − 2 ) t C ( w ) . A(w)=\delta_q\beta e^{(\theta-2)t}B(w)+\kappa e^{(p-2)t}C(w). A ( w ) = δ q β e ( θ − 2 ) t B ( w ) + κ e ( p − 2 ) t C ( w ) .
The right-hand side increases strictly from 0 0 0 to + ∞ +\infty + ∞ , so the equation has a unique root. If q = q c q=q_c q = q c , then θ = 2 \theta=2 θ = 2 . By (19) ,
A ( w ) − δ q β B ( w ) ≥ c β A ( w ) > 0. A(w)-\delta_q\beta B(w)\ge c_\beta A(w)>0. A ( w ) − δ q β B ( w ) ≥ c β A ( w ) > 0.
Equation (28) therefore admits a unique root t ∗ ( w ) t_*(w) t ∗ ( w ) . In both cases, Φ w ′ ( t ) \Phi'_w(t) Φ w ′ ( t ) changes sign from positive to negative at t ∗ ( w ) t_*(w) t ∗ ( w ) , so that t ∗ ( w ) t_*(w) t ∗ ( w ) is the unique strict global maximizer of Φ w \Phi_w Φ w . Moreover, the identity Φ w ′ ( t ∗ ( w ) ) = 0 \Phi'_w(t_*(w))=0 Φ w ′ ( t ∗ ( w )) = 0 gives
Φ w ′ ′ ( t ∗ ( w ) ) = − δ q β ( θ − 2 ) e θ t ∗ ( w ) B ( w ) − ( p − 2 ) κ e p t ∗ ( w ) C ( w ) < 0. \Phi''_w(t_*(w))=-\delta_q\beta(\theta-2)e^{\theta t_*(w)}B(w)-(p-2)\kappa e^{pt_*(w)}C(w)<0. Φ w ′′ ( t ∗ ( w )) = − δ q β ( θ − 2 ) e θ t ∗ ( w ) B ( w ) − ( p − 2 ) κ e p t ∗ ( w ) C ( w ) < 0.
The implicit function theorem then implies that the map w ↦ t ∗ ( w ) w\mapsto t_*(w) w ↦ t ∗ ( w ) is of class C 1 C^1 C 1 on S S S .
We now establish the variational properties common to the three mass regimes. Throughout the remainder of this section, we set
H = { P − , 2 < q < q c , P , q c ≤ q < p . \mathcal{H}=
\begin{cases}
\mathcal{P}^{-}, & 2<q<q_c,\\
\mathcal{P}, & q_c\le q<p.
\end{cases} H = { P − , P , 2 < q < q c , q c ≤ q < p .
We denote by t ∗ ( w ) t_*(w) t ∗ ( w ) the unique global maximizer of Φ w \Phi_w Φ w , with t ∗ ( w ) = t 2 ( w ) t_*(w)=t_2(w) t ∗ ( w ) = t 2 ( w ) when q < q c q<q_c q < q c . We assume β < β 0 \beta<\beta_0 β < β 0 in the mass-subcritical case and (9) in the mass-critical case.
Lemma 3.3 (Uniform fiber gap). There exists a constant c 1 > 0 c_1>0 c 1 > 0 such that
− Φ z ′ ′ ( 0 ) ≥ c 1 ( z ∈ H ) . (36) -\Phi''_z(0)\ge c_1\qquad(z\in\mathcal{H}). \tag*{(36)} − Φ z ′′ ( 0 ) ≥ c 1 ( z ∈ H ) . (36) Moreover, for every L > 0 L>0 L > 0 , there exist constants r L , C L > 0 r_L,C_L>0 r L , C L > 0 such that
∣ t ∗ ( y ) ∣ ≤ C L ∥ y − z ∥ X (37) |t_*(y)|\le C_L\lVert y-z\rVert_X \tag*{(37)} ∣ t ∗ ( y ) ∣ ≤ C L ∥ y − z ∥ X (37) whenever z ∈ H z\in\mathcal{H} z ∈ H , ∥ z ∥ X ≤ L \lVert z\rVert_X\le L ∥ z ∥ X ≤ L , y ∈ S y\in S y ∈ S , and ∥ y − z ∥ X < r L \lVert y-z\rVert_X<r_L ∥ y − z ∥ X < r L .
Proof. Suppose θ < 2 \theta< 2 θ < 2 . For z ∈ P − z \in\mathcal{P}^{-} z ∈ P − , the conditions P ( z ) = 0 P(z)=0 P ( z ) = 0 and Φ z ′ ′ ( 0 ) < 0 \Phi_z''(0)<0 Φ z ′′ ( 0 ) < 0 imply that
A ( z ) < p − θ 2 − θ κ C ( z ) . A(z)<\frac{p-\theta}{2-\theta}\kappa C(z). A ( z ) < 2 − θ p − θ κ C ( z ) .
Together with (1) , this implies
A ( z ) ≥ A ∗ : = ( ( 2 − θ ) S ( a , b ) p / 2 ( p − θ ) κ ) 2 / ( p − 2 ) . A(z)\geq A_*:=\left(\frac{(2-\theta)S(a,b)^{p/2}}{(p-\theta)\kappa}\right)^{2/(p-2)}. A ( z ) ≥ A ∗ := ( ( p − θ ) κ ( 2 − θ ) S ( a , b ) p /2 ) 2/ ( p − 2 ) .
Since c ∗ = p − 2 p − θ A ∗ ( 2 − θ ) / 2 c_*=\frac{p-2}{p-\theta}A_*^{(2-\theta)/2} c ∗ = p − θ p − 2 A ∗ ( 2 − θ ) /2 , the assumption β < β f \beta<\beta_f β < β f , together with Lemma 2.2, gives
− Φ z ′ ′ ( 0 ) = ( p − 2 ) A ( z ) − δ q ( p − θ ) β B ( z ) ≥ ( p − 2 − δ q ( p − θ ) β K q A ∗ ( θ − 2 ) / 2 ) A ( z ) ≥ p − 2 2 A ( z ) ≥ p − 2 2 A ∗ > 0. \begin{aligned}
-\Phi_z''(0)=(p-2)A(z)-\delta_q(p-\theta)\beta B(z) \\
&\geq\left(p-2-\delta_q(p-\theta)\beta K_qA_*^{(\theta-2)/2}\right)A(z) \\
&\geq\frac{p-2}{2}A(z)\geq\frac{p-2}{2}A_*>0.
\end{aligned} − Φ z ′′ ( 0 ) = ( p − 2 ) A ( z ) − δ q ( p − θ ) β B ( z ) ≥ ( p − 2 − δ q ( p − θ ) β K q A ∗ ( θ − 2 ) /2 ) A ( z ) ≥ 2 p − 2 A ( z ) ≥ 2 p − 2 A ∗ > 0.
Next, suppose θ > 2 \theta>2 θ > 2 . The identity P ( z ) = 0 P(z)=0 P ( z ) = 0 yields
− Φ z ′ ′ ( 0 ) = δ q ( θ − 2 ) β B ( z ) + ( p − 2 ) κ C ( z ) ≥ min { θ − 2 , p − 2 } A ( z ) . -\Phi_z''(0)=\delta_q(\theta-2)\beta B(z)+(p-2)\kappa C(z)\geq\min\{\theta-2,p-2\}A(z). − Φ z ′′ ( 0 ) = δ q ( θ − 2 ) β B ( z ) + ( p − 2 ) κ C ( z ) ≥ min { θ − 2 , p − 2 } A ( z ) .
By Lemma 2.2 and (1) , we obtain
1 ≤ δ q β K q A ( z ) ( θ − 2 ) / 2 + κ S ( a , b ) − p / 2 A ( z ) ( p − 2 ) / 2 . 1\leq\delta_q\beta K_qA(z)^{(\theta-2)/2}+\kappa S(a,b)^{-p/2}A(z)^{(p-2)/2}. 1 ≤ δ q β K q A ( z ) ( θ − 2 ) /2 + κ S ( a , b ) − p /2 A ( z ) ( p − 2 ) /2 .
Since θ > 2 \theta>2 θ > 2 and p > 2 p>2 p > 2 , this inequality implies that A ( z ) ≥ A 0 > 0 A(z)\geq A_0>0 A ( z ) ≥ A 0 > 0 for some constant A 0 A_0 A 0 independent of z z z . Consequently, − Φ z ′ ′ ( 0 ) -\Phi_z''(0) − Φ z ′′ ( 0 ) admits a uniform positive lower bound. Finally, suppose θ = 2 \theta=2 θ = 2 . Recalling that c β = 1 − δ q β K q > 0 c_\beta=1-\delta_q\beta K_q>0 c β = 1 − δ q β K q > 0 , we have
− Φ z ′ ′ ( 0 ) = ( p − 2 ) κ C ( z ) ≥ ( p − 2 ) c β A ( z ) . -\Phi_z''(0)=(p-2)\kappa C(z)\geq(p-2)c_\beta A(z). − Φ z ′′ ( 0 ) = ( p − 2 ) κ C ( z ) ≥ ( p − 2 ) c β A ( z ) .
Moreover, P ( z ) = 0 P(z)=0 P ( z ) = 0 and (1) imply
A ( z ) ≥ ( c β S ( a , b ) p / 2 κ ) 2 / ( p − 2 ) . A(z)\geq\left(\frac{c_\beta S(a,b)^{p/2}}{\kappa}\right)^{2/(p-2)}. A ( z ) ≥ ( κ c β S ( a , b ) p /2 ) 2/ ( p − 2 ) .
Thus (36) holds in all three cases.
We next prove (37) . Define
H ( t , y ) = Φ y ′ ( t ) = e 2 t A ( y ) − δ q β e θ t B ( y ) − κ e p t C ( y ) . H(t,y)=\Phi_y'(t)=e^{2t}A(y)-\delta_q\beta e^{\theta t}B(y)-\kappa e^{pt}C(y). H ( t , y ) = Φ y ′ ( t ) = e 2 t A ( y ) − δ q β e θ t B ( y ) − κ e pt C ( y ) .
Fix L > 0 L>0 L > 0 and let z ∈ H z\in\mathcal{H} z ∈ H satisfy ∥ z ∥ X ≤ L \lVert z\rVert_X\leq L ∥ z ∥ X ≤ L . Since t ∗ ( z ) = 0 t_*(z)=0 t ∗ ( z ) = 0 , we have
H ( 0 , z ) = 0 , H t ( 0 , z ) = Φ z ′ ′ ( 0 ) ≤ − c 1 . H(0,z)=0,\qquad H_t(0,z)=\Phi_z''(0)\leq-c_1. H ( 0 , z ) = 0 , H t ( 0 , z ) = Φ z ′′ ( 0 ) ≤ − c 1 .
Choose r L ≤ 1 r_L\leq1 r L ≤ 1 so that ∥ y ∥ X ≤ L + 1 \lVert y\rVert_X\leq L+1 ∥ y ∥ X ≤ L + 1 whenever ∥ y − z ∥ X < r L \lVert y-z\rVert_X<r_L ∥ y − z ∥ X < r L . Since A A A , B B B , and C C C are uniformly Lipschitz on bounded subsets of X X X , and H ( 0 , z ) = 0 H(0,z)=0 H ( 0 , z ) = 0 , there exists a constant K L > 0 K_L>0 K L > 0 , independent of y y y and z z z , such that
∣ H ( 0 , y ) ∣ + ∣ H t ( 0 , y ) − H t ( 0 , z ) ∣ ≤ K L ∥ y − z ∥ X . \lvert H(0,y)\rvert+\lvert H_t(0,y)-H_t(0,z)\rvert\leq K_L\lVert y-z\rVert_X. ∣ H ( 0 , y )∣ + ∣ H t ( 0 , y ) − H t ( 0 , z )∣ ≤ K L ∥ y − z ∥ X .
Furthermore, the explicit expression for H H H yields a constant M L > 0 M_L>0 M L > 0 such that
∣ H t t ( t , y ) ∣ ≤ M L \lvert H_{tt}(t,y)\rvert\leq M_L ∣ H tt ( t , y )∣ ≤ M L
for ∣ t ∣ ≤ 1 \lvert t\rvert\leq1 ∣ t ∣ ≤ 1 and ∥ y ∥ X ≤ L + 1 \lVert y\rVert_X\leq L+1 ∥ y ∥ X ≤ L + 1 . Consequently,
H t ( t , y ) ≤ − c 1 + M L ∣ t ∣ + K L ∥ y − z ∥ X . H_t(t,y)\leq-c_1+M_L\lvert t\rvert+K_L\lVert y-z\rVert_X. H t ( t , y ) ≤ − c 1 + M L ∣ t ∣ + K L ∥ y − z ∥ X .
Choose h L , r L ∈ ( 0 , 1 ) h_L,r_L \in(0,1) h L , r L ∈ ( 0 , 1 ) sufficiently small so that M L h L + K L r L ≤ c 1 2 M_Lh_L+K_Lr_L \le\frac{c_1}{2} M L h L + K L r L ≤ 2 c 1 , K L r L < c 1 h L 4 K_Lr_L < \frac{c_1h_L}{4} K L r L < 4 c 1 h L . Then, whenever ∥ y − z ∥ X < r L \lVert y-z\rVert_X<r_L ∥ y − z ∥ X < r L ,
H t ( t , y ) ≤ − c 1 2 ( ∣ t ∣ ≤ h L ) . H_t(t,y)\le-\frac{c_1}{2}\qquad(|t|\le h_L). H t ( t , y ) ≤ − 2 c 1 ( ∣ t ∣ ≤ h L ) .
It follows that
H ( − h L , y ) > 0 , H ( h L , y ) < 0. H(-h_L,y)>0,\qquad H(h_L,y)<0. H ( − h L , y ) > 0 , H ( h L , y ) < 0.
Therefore, H ( ⋅ , y ) H(\cdot,y) H ( ⋅ , y ) has a unique zero in ( − h L , h L ) (-h_L,h_L) ( − h L , h L ) . By the fiber geometry established in Lemma 3.1 and Lemma 3.2, this zero coincides with t ∗ ( y ) t_*(y) t ∗ ( y ) . Finally, the mean value theorem yields
c 1 2 ∣ t ∗ ( y ) ∣ ≤ ∣ H ( 0 , y ) ∣ ≤ K L ∥ y − z ∥ X . \frac{c_1}{2}|t_*(y)|\le|H(0,y)|\le K_L\lVert y-z\rVert_X. 2 c 1 ∣ t ∗ ( y ) ∣ ≤ ∣ H ( 0 , y ) ∣ ≤ K L ∥ y − z ∥ X .
Thus (3.10) follows with C L = 2 K L / c 1 C_L=2K_L/c_1 C L = 2 K L / c 1 .
Lemma 3.4 (Maximum functional). The functional
F ( w ) = max t ∈ R J ( t ⋆ w ) = 1 2 e 2 t ∗ ( w ) A ( w ) − β q e θ t ∗ ( w ) B ( w ) − κ p e p t ∗ ( w ) C ( w ) (38) F(w)=\max_{t\in\mathbb{R}}J(t\star w)=\frac{1}{2}e^{2t_*(w)}A(w)-\frac{\beta}{q}e^{\theta t_*(w)}B(w)-\frac{\kappa}{p}e^{pt_*(w)}C(w) \tag*{(38)} F ( w ) = t ∈ R max J ( t ⋆ w ) = 2 1 e 2 t ∗ ( w ) A ( w ) − q β e θ t ∗ ( w ) B ( w ) − p κ e p t ∗ ( w ) C ( w ) (38) is of class C 1 C^1 C 1 on S S S , and its derivative is given by
d F ( w ) [ h ] = J ′ ( t ∗ ( w ) ⋆ w ) [ t ∗ ( w ) ⋆ h ] ( h ∈ T w S ) . (39) dF(w)[h]=J'(t_*(w)\star w)[t_*(w)\star h]\qquad(h\in T_wS). \tag*{(39)} d F ( w ) [ h ] = J ′ ( t ∗ ( w ) ⋆ w ) [ t ∗ ( w ) ⋆ h ] ( h ∈ T w S ) . (39) Every local minimizer of J J J on H \mathcal{H} H is a constrained critical point of J J J on S S S .
Proof. Differentiate (3.11). The coefficient of d t ∗ ( w ) [ h ] dt_*(w)[h] d t ∗ ( w ) [ h ] vanishes because H ( t ∗ ( w ) , w ) = 0 H(t_*(w),w)=0 H ( t ∗ ( w ) , w ) = 0 . The remaining terms yield (3.12), since the dilation is bounded and linear on X X X . Since t ∗ t_* t ∗ is continuous on S S S , the strong continuity of the dilation and (2.5) imply that the map w ↦ t ∗ ( w ) ⋆ w w\mapsto t_*(w)\star w w ↦ t ∗ ( w ) ⋆ w is continuous from S S S into H \mathcal{H} H . Let z ∈ H z\in\mathcal{H} z ∈ H be a local minimizer of J J J on H \mathcal{H} H . Since t ∗ ( z ) = 0 t_*(z)=0 t ∗ ( z ) = 0 , it follows that z z z is a local minimizer of F F F on S S S . Consequently, d F ( z ) = 0 dF(z)=0 d F ( z ) = 0 , and (3.12) yields d ( J ∣ S ) ( z ) = 0 d(J|_S)(z)=0 d ( J ∣ S ) ( z ) = 0 .
Proposition 3.1 (Constrained Palais–Smale sequence). Let c = inf H J c=\inf_{\mathcal{H}}J c = inf H J . If there exists a bounded minimizing sequence for c c c in H \mathcal{H} H , then there exists a bounded Palais–Smale sequence { w n } ⊂ H \{w_n\}\subset\mathcal{H} { w n } ⊂ H for J ∣ S J|_S J ∣ S at level c c c satisfying
J ( w n ) → c , P ( w n ) = 0 , ∥ d ( J ∣ S ) ( w n ) ∥ → 0. (40) J(w_n)\to c,\qquad P(w_n)=0,\qquad\lVert d(J|_S)(w_n)\rVert\to0. \tag*{(40)} J ( w n ) → c , P ( w n ) = 0 , ∥ d ( J ∣ S ) ( w n )∥ → 0. (40)
Proof. The fiber lemmas give F > 0 F>0 F > 0 and inf S F = inf H J = c \inf_SF=\inf_{\mathcal{H}}J=c inf S F = inf H J = c . Let { z n } ⊂ H \{z_n\}\subset\mathcal{H} { z n } ⊂ H be a bounded minimizing sequence such that
J ( z n ) ≤ c + ε n 2 , ε n → 0 + . J(z_n)\le c+\varepsilon_n^2,\qquad\varepsilon_n\to0^+. J ( z n ) ≤ c + ε n 2 , ε n → 0 + .
Since S S S is closed in X X X , it is complete with respect to the induced metric. Applying Ekeland’s variational principle to F F F on S S S , we obtain { y n } ⊂ S \{y_n\}\subset S { y n } ⊂ S satisfying
∥ y n − z n ∥ X ≤ ε n , c ≤ F ( y n ) ≤ F ( z n ) , ∥ d F ( y n ) ∥ ( T y n S ) ∗ ≤ ε n . \lVert y_n-z_n\rVert_X\le\varepsilon_n,\qquad c\le F(y_n)\le F(z_n),\qquad\lVert dF(y_n)\rVert_{(T_{y_n}S)^*}\le\varepsilon_n. ∥ y n − z n ∥ X ≤ ε n , c ≤ F ( y n ) ≤ F ( z n ) , ∥ d F ( y n ) ∥ ( T y n S ) ∗ ≤ ε n .
The derivative estimate follows by applying Ekeland’s variational inequality along the normalized curves γ ( t ) = y n + t h M ( y n + t h ) 1 / 2 \gamma(t)=\frac{y_n+th}{M(y_n+th)^{1/2}} γ ( t ) = M ( y n + t h ) 1/2 y n + t h , h ∈ T y n S h\in T_{y_n}S h ∈ T y n S , which satisfy γ ( 0 ) = y n \gamma(0)=y_n γ ( 0 ) = y n and γ ′ ( 0 ) = h \gamma'(0)=h γ ′ ( 0 ) = h . Since { z n } \{z_n\} { z n } is bounded, there exists L > 0 L>0 L > 0 such that ∥ z n ∥ X ≤ L \lVert z_n\rVert_X\le L ∥ z n ∥ X ≤ L for all n n n . By (3.10), for all sufficiently large n n n ,
∣ s n ∣ : = ∣ t ∗ ( y n ) ∣ ≤ C L ∥ y n − z n ∥ X ≤ C L ε n → 0. |s_n|:=|t_*(y_n)|\le C_L\lVert y_n-z_n\rVert_X\le C_L\varepsilon_n\to0. ∣ s n ∣ := ∣ t ∗ ( y n ) ∣ ≤ C L ∥ y n − z n ∥ X ≤ C L ε n → 0.
Set w n = s n ⋆ y n w_n=s_n\star y_n w n = s n ⋆ y n . Then w n ∈ H w_n\in\mathcal{H} w n ∈ H and
J ( w n ) = F ( y n ) → c , P ( w n ) = 0. J(w_n)=F(y_n)\to c,\qquad P(w_n)=0. J ( w n ) = F ( y n ) → c , P ( w n ) = 0.
Moreover, since { y n } \{y_n\} { y n } is bounded and s n → 0 s_n \to0 s n → 0 , it follows from (18) that { w n } \{w_n\} { w n } is bounded in X X X . Finally, for any k ∈ T w n S k \in T_{w_n}S k ∈ T w n S , the mass-preserving dilation implies that ( − s n ) ⋆ k ∈ T y n S (-s_n)\star k \in T_{y_n}S ( − s n ) ⋆ k ∈ T y n S . Hence, by (3.12 and (18) ,
∣ J ′ ( w n ) [ k ] ∣ = ∣ d F ( y n ) [ ( − s n ) ⋆ k ] ∣ ≤ ∥ d F ( y n ) ∥ ( T y n S ) ∗ ∥ ( − s n ) ⋆ k ∥ X ≤ ε n e ∣ s n ∣ ∥ k ∥ X . |J'(w_n)[k]| = |dF(y_n)[(-s_n)\star k]| \leq\|dF(y_n)\|_{(T_{y_n}S)^*}\|(-s_n)\star k\|_X \leq\varepsilon_n e^{|s_n|}\|k\|_X. ∣ J ′ ( w n ) [ k ] ∣ = ∣ d F ( y n ) [( − s n ) ⋆ k ] ∣ ≤ ∥ d F ( y n ) ∥ ( T y n S ) ∗ ∥ ( − s n ) ⋆ k ∥ X ≤ ε n e ∣ s n ∣ ∥ k ∥ X .
Consequently,
∥ d ( J ∣ S ) ( w n ) ∥ ≤ ε n e ∣ s n ∣ → 0 , \|d(J|_S)(w_n)\| \leq\varepsilon_n e^{|s_n|} \to0, ∥ d ( J ∣ S ) ( w n ) ∥ ≤ ε n e ∣ s n ∣ → 0 ,
which completes the proof.
Compactness below the critical level
In this section, we establish compactness results below the critical energy level for use in the existence proofs in Sections 5 and 6.
Lemma 4.1 (Annular regularity and Pohozaev identity). Let w ∈ X w \in X w ∈ X be a nonzero weak solution of
− div ( ∣ x ∣ − 2 a ∇ w ) = λ ∣ x ∣ − 2 a w + β ∣ x ∣ − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ − b p ∣ w ∣ p − 2 w . (41) -\operatorname{div}(|x|^{-2a}\nabla w) = \lambda|x|^{-2a}w + \beta|x|^{-bq}|w|^{q-2}w + \kappa|x|^{-bp}|w|^{p-2}w. \tag*{(41)} − div ( ∣ x ∣ − 2 a ∇ w ) = λ ∣ x ∣ − 2 a w + β ∣ x ∣ − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ − b p ∣ w ∣ p − 2 w . (41) Then w ∈ L l o c ∞ ∩ H l o c 2 w \in L^\infty_{\mathrm{loc}} \cap H^2_{\mathrm{loc}} w ∈ L loc ∞ ∩ H loc 2 away from the origin, and
P ( w ) = 0 , λ M ( w ) = ( δ q − 1 ) β B ( w ) < 0. (42) P(w) = 0,\qquad\lambda M(w) = (\delta_q - 1)\beta B(w) < 0. \tag*{(42)} P ( w ) = 0 , λ M ( w ) = ( δ q − 1 ) β B ( w ) < 0. (42) If w ≥ 0 w \geq0 w ≥ 0 , then w > 0 w > 0 w > 0 on R N ∖ { 0 } \mathbb{R}^N \setminus\{0\} R N ∖ { 0 } .
Proof. Let Ω ′ \Omega' Ω ′ and Ω \Omega Ω be bounded annuli centered at the origin such that Ω ′ ⋐ Ω ⋐ R N ∖ { 0 } \Omega' \Subset\Omega\Subset\mathbb{R}^N \setminus\{0\} Ω ′ ⋐ Ω ⋐ R N ∖ { 0 } . Set
g ( x ) = ∣ x ∣ − 2 a , 2 ∗ = 2 N N − 2 . g(x) = |x|^{-2a},\qquad2^* = \frac{2N}{N-2}. g ( x ) = ∣ x ∣ − 2 a , 2 ∗ = N − 2 2 N .
There are constants g 0 , g 1 > 0 g_0,g_1 > 0 g 0 , g 1 > 0 such that
g 0 ≤ g ( x ) ≤ g 1 ( x ∈ Ω ) . g_0 \leq g(x) \leq g_1 \qquad(x \in\Omega). g 0 ≤ g ( x ) ≤ g 1 ( x ∈ Ω ) .
Since d = 1 + a − b ∈ ( 0 , 1 ) d = 1+a-b \in(0,1) d = 1 + a − b ∈ ( 0 , 1 ) , we have
p = 2 N N − 2 d < 2 ∗ , h : = 2 ∗ p − 2 = N ( N − 2 d ) 2 d ( N − 2 ) > N 2 . p = \frac{2N}{N-2d} < 2^*,\qquad h := \frac{2^*}{p-2} = \frac{N(N-2d)}{2d(N-2)} > \frac{N}{2}. p = N − 2 d 2 N < 2 ∗ , h := p − 2 2 ∗ = 2 d ( N − 2 ) N ( N − 2 d ) > 2 N .
On Ω \Omega Ω , write (41) as
− div ( g ∇ w ) = V w , V = λ ∣ x ∣ − 2 a + β ∣ x ∣ − b q ∣ w ∣ q − 2 + κ ∣ x ∣ − b p ∣ w ∣ p − 2 . -\operatorname{div}(g\nabla w) = Vw,\qquad V = \lambda|x|^{-2a} + \beta|x|^{-bq}|w|^{q-2} + \kappa|x|^{-bp}|w|^{p-2}. − div ( g ∇ w ) = V w , V = λ ∣ x ∣ − 2 a + β ∣ x ∣ − b q ∣ w ∣ q − 2 + κ ∣ x ∣ − b p ∣ w ∣ p − 2 .
The embedding H 1 ( Ω ) ↪ L 2 ∗ ( Ω ) H^1(\Omega) \hookrightarrow L^{2^*}(\Omega) H 1 ( Ω ) ↪ L 2 ∗ ( Ω ) gives
∣ w ∣ p − 2 ∈ L h ( Ω ) , ∣ w ∣ q − 2 ∈ L 2 ∗ q − 2 ( Ω ) ⊂ L h ( Ω ) . |w|^{p-2} \in L^h(\Omega),\qquad|w|^{q-2} \in L^{\frac{2^*}{q-2}}(\Omega) \subset L^h(\Omega). ∣ w ∣ p − 2 ∈ L h ( Ω ) , ∣ w ∣ q − 2 ∈ L q − 2 2 ∗ ( Ω ) ⊂ L h ( Ω ) .
Hence V ∈ L h ( Ω ) V \in L^h(\Omega) V ∈ L h ( Ω ) . Let s ≥ 1 s \geq1 s ≥ 1 and L > 0 L > 0 L > 0 . Choose η ∈ C ∞ ( Ω ) \eta\in C^\infty(\Omega) η ∈ C ∞ ( Ω ) such that 0 ≤ η ≤ 1 0 \leq\eta\leq1 0 ≤ η ≤ 1 and η ≡ 1 \eta\equiv1 η ≡ 1 on Ω ′ \Omega' Ω ′ . Set
w L = min { ∣ w ∣ , L } , f L = η w w L s − 1 . w_L = \min\{|w|,L\},\qquad f_L = \eta w w_L^{s-1}. w L = min { ∣ w ∣ , L } , f L = η w w L s − 1 .
Testing the weak equation with η 2 w w L 2 s − 2 \eta^2 w w_L^{2s-2} η 2 w w L 2 s − 2 gives
∫ g ∇ w ⋅ ∇ ( η 2 w w L 2 s − 2 ) = ∫ V η 2 w 2 w L 2 s − 2 . \int g\nabla w\cdot\nabla(\eta^2 w w_L^{2s-2}) = \int V\eta^2 w^2w_L^{2s-2}. ∫ g ∇ w ⋅ ∇ ( η 2 w w L 2 s − 2 ) = ∫ V η 2 w 2 w L 2 s − 2 .
Expanding the left-hand side and applying the Cauchy–Schwarz and Young inequalities, we obtain
∥ ∇ f L ∥ 2 2 ≤ C s 2 ∫ ∣ ∇ η ∣ 2 w L 2 s − 2 + C s 2 ∫ ∣ V ∣ f L 2 , (43) \|\nabla f_L\|_2^2 \le Cs^2 \int|\nabla\eta|^2 w_L^{2s-2} + Cs^2 \int|V|f_L^2, \tag*{(43)} ∥∇ f L ∥ 2 2 ≤ C s 2 ∫ ∣∇ η ∣ 2 w L 2 s − 2 + C s 2 ∫ ∣ V ∣ f L 2 , (43)
where C > 0 C > 0 C > 0 is independent of s s s and L L L . Set
r = 2 h h − 1 , ϑ = N 2 h ∈ ( 0 , 1 ) . r = \frac{2h}{h-1}, \qquad\vartheta= \frac{N}{2h} \in(0,1). r = h − 1 2 h , ϑ = 2 h N ∈ ( 0 , 1 ) .
Since
1 r = ϑ 2 ∗ + 1 − ϑ 2 , \frac{1}{r} = \frac{\vartheta}{2^*} + \frac{1-\vartheta}{2}, r 1 = 2 ∗ ϑ + 2 1 − ϑ ,
the Hölder inequality and the interpolation inequality yield
∫ ∣ V ∣ f L 2 ≤ ∥ V ∥ h ∥ f L ∥ r 2 ≤ C ∥ ∇ f L ∥ 2 2 ϑ ∥ f L ∥ 2 2 ( 1 − ϑ ) ≤ ε ∥ ∇ f L ∥ 2 2 + C ε − N 2 h − N ∥ f L ∥ 2 2 \int|V|f_L^2 \le\|V\|_h\|f_L\|_r^2 \le C\|\nabla f_L\|_2^{2\vartheta}\|f_L\|_2^{2(1-\vartheta)} \le\varepsilon\|\nabla f_L\|_2^2 + C\varepsilon^{-\frac{N}{2h-N}}\|f_L\|_2^2 ∫ ∣ V ∣ f L 2 ≤ ∥ V ∥ h ∥ f L ∥ r 2 ≤ C ∥∇ f L ∥ 2 2 ϑ ∥ f L ∥ 2 2 ( 1 − ϑ ) ≤ ε ∥∇ f L ∥ 2 2 + C ε − 2 h − N N ∥ f L ∥ 2 2
for every ε > 0 \varepsilon> 0 ε > 0 , where the last inequality follows from the Young inequality. Choosing ε = 1 2 C s 2 \varepsilon= \frac{1}{2Cs^2} ε = 2 C s 2 1 in (4.3) and absorbing the gradient term, we obtain
∥ ∇ f L ∥ 2 2 ≤ C s ℓ ∫ ( ∣ ∇ η ∣ 2 + η 2 ) w L 2 s − 2 , (44) \|\nabla f_L\|_2^2 \le Cs^\ell\int(|\nabla\eta|^2+\eta^2)w_L^{2s-2}, \tag*{(44)} ∥∇ f L ∥ 2 2 ≤ C s ℓ ∫ ( ∣∇ η ∣ 2 + η 2 ) w L 2 s − 2 , (44)
where ℓ = 2 + 2 N 2 h − N \ell= 2+\frac{2N}{2h-N} ℓ = 2 + 2 h − N 2 N . Here and below, C > 0 C > 0 C > 0 denotes a constant independent of s s s and L L L . Assume that w ∈ L 2 s ( Ω ) w \in L^{2s}(\Omega) w ∈ L 2 s ( Ω ) . Applying the Sobolev inequality to f L f_L f L and then letting L → ∞ L \to\infty L → ∞ , we deduce from (4.4) and Fatou’s lemma that
∥ w ∥ L 2 ∗ s ( Ω ′ ) ≤ ( C s ℓ ( 1 + ∥ ∇ η ∥ ∞ 2 ) ) 1 2 s ∥ w ∥ L 2 s ( Ω ) . (45) \|w\|_{L^{2^*s}(\Omega')} \le\left(Cs^\ell(1+\|\nabla\eta\|_\infty^2)\right)^{\frac{1}{2s}}\|w\|_{L^{2s}(\Omega)}. \tag*{(45)} ∥ w ∥ L 2 ∗ s ( Ω ′ ) ≤ ( C s ℓ ( 1 + ∥∇ η ∥ ∞ 2 ) ) 2 s 1 ∥ w ∥ L 2 s ( Ω ) . (45)
Choose a decreasing sequence of annuli { Ω j } j ≥ 0 \{\Omega_j\}_{j\ge0} { Ω j } j ≥ 0 such that
Ω ′ ⋐ Ω j + 1 ⋐ Ω j ⋐ Ω , ⋂ j = 0 ∞ Ω j = Ω ′ ‾ . \Omega' \Subset\Omega_{j+1} \Subset\Omega_j \Subset\Omega,\qquad\bigcap_{j=0}^{\infty}\Omega_j=\overline{\Omega'}. Ω ′ ⋐ Ω j + 1 ⋐ Ω j ⋐ Ω , j = 0 ⋂ ∞ Ω j = Ω ′ .
For each j ≥ 0 j\ge0 j ≥ 0 , let η j ∈ C c ∞ ( Ω j ) \eta_j\in C^\infty_c(\Omega_j) η j ∈ C c ∞ ( Ω j ) satisfy
0 ≤ η j ≤ 1 , η j ≡ 1 on Ω j + 1 , ∥ ∇ η j ∥ ∞ ≤ C 2 j . 0\le\eta_j\le1,\qquad\eta_j\equiv1\ \text{on }\Omega_{j+1},\qquad\|\nabla\eta_j\|_\infty\le C2^j. 0 ≤ η j ≤ 1 , η j ≡ 1 on Ω j + 1 , ∥∇ η j ∥ ∞ ≤ C 2 j .
Set
σ = N N − 2 > 1 , s j = σ j . \sigma=\frac{N}{N-2}>1,\qquad s_j=\sigma^j. σ = N − 2 N > 1 , s j = σ j .
Since 2 ∗ s j = 2 s j + 1 2^*s_j=2s_{j+1} 2 ∗ s j = 2 s j + 1 , applying (4.5) with s = s j s=s_j s = s j and η = η j \eta=\eta_j η = η j yields
∥ w ∥ L 2 s j + 1 ( Ω j + 1 ) ≤ ( C 4 j s j ℓ ) 1 / ( 2 s j ) ∥ w ∥ L 2 s j ( Ω j ) . \|w\|_{L^{2s_{j+1}}(\Omega_{j+1})}\le(C4^j s_j^\ell)^{1/(2s_j)}\|w\|_{L^{2s_j}(\Omega_j)}. ∥ w ∥ L 2 s j + 1 ( Ω j + 1 ) ≤ ( C 4 j s j ℓ ) 1/ ( 2 s j ) ∥ w ∥ L 2 s j ( Ω j ) .
Moreover,
∑ j = 0 ∞ log C + j log 4 + ℓ log s j 2 s j < ∞ , \sum_{j=0}^{\infty}\frac{\log C+j\log4+\ell\log s_j}{2s_j}<\infty, j = 0 ∑ ∞ 2 s j log C + j log 4 + ℓ log s j < ∞ ,
since s j = σ j s_j=\sigma^j s j = σ j with σ > 1 \sigma>1 σ > 1 . Consequently, iterating the preceding inequality and letting j → ∞ j\to\infty j → ∞ , we obtain
∥ w ∥ L ∞ ( Ω ′ ) ≤ C ∥ w ∥ L 2 ( Ω ) . \|w\|_{L^\infty(\Omega')}\le C\|w\|_{L^2(\Omega)}. ∥ w ∥ L ∞ ( Ω ′ ) ≤ C ∥ w ∥ L 2 ( Ω ) .
As Ω ′ \Omega' Ω ′ is arbitrary, it follows that
w ∈ L l o c ∞ ( R N ∖ { 0 } ) . w\in L^\infty_{\mathrm{loc}}(\mathbb{R}^N\setminus\{0\}). w ∈ L loc ∞ ( R N ∖ { 0 }) .
Dividing the equation by g ( x ) = ∣ x ∣ − 2 a g(x)=|x|^{-2a} g ( x ) = ∣ x ∣ − 2 a on Ω \Omega Ω gives
− Δ w + 2 a x ∣ x ∣ 2 ⋅ ∇ w = λ w + β ∣ x ∣ 2 a − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ 2 a − b p ∣ w ∣ p − 2 w . -\Delta w+2a\frac{x}{|x|^2}\cdot\nabla w=\lambda w+\beta|x|^{2a-bq}|w|^{q-2}w+\kappa|x|^{2a-bp}|w|^{p-2}w. − Δ w + 2 a ∣ x ∣ 2 x ⋅ ∇ w = λ w + β ∣ x ∣ 2 a − b q ∣ w ∣ q − 2 w + κ ∣ x ∣ 2 a − b p ∣ w ∣ p − 2 w .
Its right-hand side is locally bounded. Interior estimates give
w ∈ H l o c 2 ( R N ∖ { 0 } ) ∩ W l o c 2 , r ( R N ∖ { 0 } ) ( 1 < r < ∞ ) ; w\in H^2_{\mathrm{loc}}(\mathbb{R}^N\setminus\{0\})\cap W^{2,r}_{\mathrm{loc}}(\mathbb{R}^N\setminus\{0\})\qquad(1<r<\infty); w ∈ H loc 2 ( R N ∖ { 0 }) ∩ W loc 2 , r ( R N ∖ { 0 }) ( 1 < r < ∞ ) ;
see [14 Chapters 8–9] .
Let 0 < 2 ε < R 0<2\varepsilon<R 0 < 2 ε < R . Choose a cutoff function χ ε , R ∈ C ∞ ( R N ∖ { 0 } ) \chi_{\varepsilon,R}\in C^\infty(\mathbb{R}^N\setminus\{0\}) χ ε , R ∈ C ∞ ( R N ∖ { 0 }) satisfying 0 ≤ χ ε , R ≤ 1 0\leq\chi_{\varepsilon,R}\leq1 0 ≤ χ ε , R ≤ 1 and
χ ε , R = { 1 , if ε ≤ ∣ x ∣ ≤ R , 0 , if ∣ x ∣ ≤ ε / 2 or ∣ x ∣ ≥ 2 R , ∣ x ∣ ∣ ∇ χ ε , R ∣ ≤ C . \chi_{\varepsilon,R}=
\begin{cases}
1,&\text{if }\varepsilon\leq|x|\leq R,\\
0,&\text{if }|x|\leq\varepsilon/2\text{ or }|x|\geq2R,
\end{cases}
\qquad|x||\nabla\chi_{\varepsilon,R}|\leq C. χ ε , R = { 1 , 0 , if ε ≤ ∣ x ∣ ≤ R , if ∣ x ∣ ≤ ε /2 or ∣ x ∣ ≥ 2 R , ∣ x ∣∣∇ χ ε , R ∣ ≤ C .
The function
φ ε , R = χ ε , R x ⋅ ∇ w \varphi_{\varepsilon,R}=\chi_{\varepsilon,R}x\cdot\nabla w φ ε , R = χ ε , R x ⋅ ∇ w
belongs to H 0 1 ( B 2 R ∖ B ε / 2 ) H^1_0(B_{2R}\setminus B_{\varepsilon/2}) H 0 1 ( B 2 R ∖ B ε /2 ) and is an admissible test function for the weak formulation. Put χ = χ ε , R \chi=\chi_{\varepsilon,R} χ = χ ε , R and
A ε , R = { ε / 2 < ∣ x ∣ < ε } ∪ { R < ∣ x ∣ < 2 R } . A_{\varepsilon,R}=\{\varepsilon/2<|x|<\varepsilon\}\cup\{R<|x|<2R\}. A ε , R = { ε /2 < ∣ x ∣ < ε } ∪ { R < ∣ x ∣ < 2 R } .
Integration by parts gives
∫ ∣ x ∣ − 2 a ∇ w ⋅ ∇ ( χ x ⋅ ∇ w ) = − N − 2 − 2 a 2 ∫ χ ∣ x ∣ − 2 a ∣ ∇ w ∣ 2 + R A , \int|x|^{-2a}\nabla w\cdot\nabla(\chi x\cdot\nabla w)
=-\frac{N-2-2a}{2}\int\chi|x|^{-2a}|\nabla w|^2+\mathcal{R}_A, ∫ ∣ x ∣ − 2 a ∇ w ⋅ ∇ ( χ x ⋅ ∇ w ) = − 2 N − 2 − 2 a ∫ χ ∣ x ∣ − 2 a ∣∇ w ∣ 2 + R A ,
λ ∫ χ ∣ x ∣ − 2 a w ( x ⋅ ∇ w ) = − N − 2 a 2 λ ∫ χ ∣ x ∣ − 2 a w 2 + R M , \lambda\int\chi|x|^{-2a}w(x\cdot\nabla w)
=-\frac{N-2a}{2}\lambda\int\chi|x|^{-2a}w^2+\mathcal{R}_M, λ ∫ χ ∣ x ∣ − 2 a w ( x ⋅ ∇ w ) = − 2 N − 2 a λ ∫ χ ∣ x ∣ − 2 a w 2 + R M ,
β ∫ χ ∣ x ∣ − b q ∣ w ∣ q − 2 w ( x ⋅ ∇ w ) = − N − b q q β ∫ χ ∣ x ∣ − b q ∣ w ∣ q + R B , \beta\int\chi|x|^{-bq}|w|^{q-2}w(x\cdot\nabla w)
=-\frac{N-bq}{q}\beta\int\chi|x|^{-bq}|w|^q+\mathcal{R}_B, β ∫ χ ∣ x ∣ − b q ∣ w ∣ q − 2 w ( x ⋅ ∇ w ) = − q N − b q β ∫ χ ∣ x ∣ − b q ∣ w ∣ q + R B ,
κ ∫ χ ∣ x ∣ − b p ∣ w ∣ p − 2 w ( x ⋅ ∇ w ) = − N − b p p κ ∫ χ ∣ x ∣ − b p ∣ w ∣ p + R C , \kappa\int\chi|x|^{-bp}|w|^{p-2}w(x\cdot\nabla w)
=-\frac{N-bp}{p}\kappa\int\chi|x|^{-bp}|w|^p+\mathcal{R}_C, κ ∫ χ ∣ x ∣ − b p ∣ w ∣ p − 2 w ( x ⋅ ∇ w ) = − p N − b p κ ∫ χ ∣ x ∣ − b p ∣ w ∣ p + R C ,
where
R A = ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ w ) ( ∇ χ ⋅ ∇ w ) − 1 2 ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ χ ) ∣ ∇ w ∣ 2 , \mathcal{R}_A=\int|x|^{-2a}(x\cdot\nabla w)(\nabla\chi\cdot\nabla w)-\frac12\int|x|^{-2a}(x\cdot\nabla\chi)|\nabla w|^2, R A = ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ w ) ( ∇ χ ⋅ ∇ w ) − 2 1 ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ χ ) ∣∇ w ∣ 2 ,
R M = − λ 2 ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ χ ) w 2 , \mathcal{R}_M=-\frac{\lambda}{2}\int|x|^{-2a}(x\cdot\nabla\chi)w^2, R M = − 2 λ ∫ ∣ x ∣ − 2 a ( x ⋅ ∇ χ ) w 2 ,
R B = − β q ∫ ∣ x ∣ − b q ( x ⋅ ∇ χ ) ∣ w ∣ q , \mathcal{R}_B=-\frac{\beta}{q}\int|x|^{-bq}(x\cdot\nabla\chi)|w|^q, R B = − q β ∫ ∣ x ∣ − b q ( x ⋅ ∇ χ ) ∣ w ∣ q ,
R C = − κ p ∫ ∣ x ∣ − b p ( x ⋅ ∇ χ ) ∣ w ∣ p . \mathcal{R}_C=-\frac{\kappa}{p}\int|x|^{-bp}(x\cdot\nabla\chi)|w|^p. R C = − p κ ∫ ∣ x ∣ − b p ( x ⋅ ∇ χ ) ∣ w ∣ p .
Set
e w = ∣ x ∣ − 2 a ∣ ∇ w ∣ 2 + ∣ λ ∣ ∣ x ∣ − 2 a w 2 + β ∣ x ∣ − b q ∣ w ∣ q + κ ∣ x ∣ − b p ∣ w ∣ p . e_w=|x|^{-2a}|\nabla w|^2+|\lambda||x|^{-2a}w^2+\beta|x|^{-bq}|w|^q+\kappa|x|^{-bp}|w|^p. e w = ∣ x ∣ − 2 a ∣∇ w ∣ 2 + ∣ λ ∣∣ x ∣ − 2 a w 2 + β ∣ x ∣ − b q ∣ w ∣ q + κ ∣ x ∣ − b p ∣ w ∣ p .
Then
∣ R A ∣ + ∣ R M ∣ + ∣ R B ∣ + ∣ R C ∣ ≤ C ∫ A ε , R e w . |\mathcal{R}_A|+|\mathcal{R}_M|+|\mathcal{R}_B|+|\mathcal{R}_C|\leq C\int_{A_{\varepsilon,R}}e_w. ∣ R A ∣ + ∣ R M ∣ + ∣ R B ∣ + ∣ R C ∣ ≤ C ∫ A ε , R e w .
Lemma 2.2 and (1) imply e w ∈ L 1 ( R N ) e_w\in L^1(\mathbb{R}^N) e w ∈ L 1 ( R N ) . Therefore
lim ε → 0 + ∫ ∣ x ∣ < ε e w = 0 , lim R → + ∞ ∫ ∣ x ∣ > R e w = 0. \lim_{\varepsilon\to0^+}\int_{|x|<\varepsilon}e_w=0,\qquad
\lim_{R\to+\infty}\int_{|x|>R}e_w=0. ε → 0 + lim ∫ ∣ x ∣ < ε e w = 0 , R → + ∞ lim ∫ ∣ x ∣ > R e w = 0.
Letting ε → 0 + \varepsilon\to0^+ ε → 0 + and R → + ∞ R\to+\infty R → + ∞ gives
N − 2 − 2 a 2 A ( w ) = N − 2 a 2 λ M ( w ) + N − b q q β B ( w ) + N − b p p κ C ( w ) . (46) \frac{N-2-2a}{2}A(w)=\frac{N-2a}{2}\lambda M(w)+\frac{N-bq}{q}\beta B(w)+\frac{N-bp}{p}\kappa C(w). \tag*{(46)} 2 N − 2 − 2 a A ( w ) = 2 N − 2 a λ M ( w ) + q N − b q β B ( w ) + p N − b p κ C ( w ) . (46)
Testing (4.1) with w w w yields
A ( w ) = λ M ( w ) + β B ( w ) + κ C ( w ) . (47) A(w) = \lambda M(w) + \beta B(w) + \kappa C(w). \tag*{(47)} A ( w ) = λ M ( w ) + β B ( w ) + κ C ( w ) . (47)
Set
c a = N − 2 − 2 a 2 . c_a = \frac{N - 2 - 2a}{2}. c a = 2 N − 2 − 2 a .
By the definitions of p p p and δ q \delta_q δ q , we have
N − 2 a 2 = c a + 1 , N − b q q = c a + 1 − δ q , N − b p p = c a . \frac{N - 2a}{2} = c_a + 1, \qquad\frac{N - bq}{q} = c_a + 1 - \delta_q, \qquad\frac{N - bp}{p} = c_a. 2 N − 2 a = c a + 1 , q N − b q = c a + 1 − δ q , p N − b p = c a .
Subtracting c a c_a c a times (4.7) from (4.6), we obtain
λ M ( w ) = ( δ q − 1 ) β B ( w ) . (48) \lambda M(w) = (\delta_q - 1)\beta B(w). \tag*{(48)} λ M ( w ) = ( δ q − 1 ) β B ( w ) . (48)
Consequently,
P ( w ) = A ( w ) − δ q β B ( w ) − κ C ( w ) = λ M ( w ) + ( 1 − δ q ) β B ( w ) = 0. P(w) = A(w) - \delta_q\beta B(w) - \kappa C(w) = \lambda M(w) + (1 - \delta_q)\beta B(w) = 0. P ( w ) = A ( w ) − δ q β B ( w ) − κ C ( w ) = λ M ( w ) + ( 1 − δ q ) β B ( w ) = 0.
The function
r ↦ δ r = N − 2 a + 2 b 2 − N r r \mapsto\delta_r = \frac{N - 2a + 2b}{2} - \frac{N}{r} r ↦ δ r = 2 N − 2 a + 2 b − r N
is strictly increasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , so q < p q < p q < p implies δ q < δ p = 1 \delta_q < \delta_p = 1 δ q < δ p = 1 . Since w ≢ 0 w \not\equiv0 w ≡ 0 , both M ( w ) M(w) M ( w ) and B ( w ) B(w) B ( w ) are positive, so (4.8) yields λ < 0 \lambda< 0 λ < 0 .
Finally, assume w ≥ 0 w \geq0 w ≥ 0 . On every annulus,
− div ( ∣ x ∣ − 2 a ∇ w ) − λ ∣ x ∣ − 2 a w = β ∣ x ∣ − b q w q − 1 + κ ∣ x ∣ − b p w p − 1 ≥ 0. -\operatorname{div}(|x|^{-2a}\nabla w) - \lambda|x|^{-2a}w = \beta|x|^{-bq}w^{q-1} + \kappa|x|^{-bp}w^{p-1} \geq0. − div ( ∣ x ∣ − 2 a ∇ w ) − λ ∣ x ∣ − 2 a w = β ∣ x ∣ − b q w q − 1 + κ ∣ x ∣ − b p w p − 1 ≥ 0.
The operator is uniformly elliptic there, and its zero-order coefficient is positive. If w ( x 0 ) = 0 w(x_0) = 0 w ( x 0 ) = 0 for some x 0 ≠ 0 x_0 \ne0 x 0 = 0 , choose a bounded connected domain compactly contained in R N ∖ { 0 } \mathbb{R}^N \setminus\{0\} R N ∖ { 0 } that contains x 0 x_0 x 0 and a point where w > 0 w > 0 w > 0 . The strong maximum principle gives a contradiction. Thus w > 0 w > 0 w > 0 on R N ∖ { 0 } \mathbb{R}^N \setminus\{0\} R N ∖ { 0 } .
Lemma 4.2. Let w n ⇀ w w_n \rightharpoonup w w n ⇀ w in X X X and set z n = w n − w z_n = w_n - w z n = w n − w . Then, after passing to a subsequence,
A ( w n ) = A ( w ) + A ( z n ) + o ( 1 ) , A(w_n) = A(w) + A(z_n) + o(1), A ( w n ) = A ( w ) + A ( z n ) + o ( 1 ) , M ( w n ) = M ( w ) + M ( z n ) + o ( 1 ) , M(w_n) = M(w) + M(z_n) + o(1), M ( w n ) = M ( w ) + M ( z n ) + o ( 1 ) , B ( w n ) = B ( w ) + o ( 1 ) , (49) B(w_n) = B(w) + o(1), \tag*{(49)} B ( w n ) = B ( w ) + o ( 1 ) , (49) and
C ( w n ) = C ( w ) + C ( z n ) + o ( 1 ) . (50) C(w_n) = C(w) + C(z_n) + o(1). \tag*{(50)} C ( w n ) = C ( w ) + C ( z n ) + o ( 1 ) . (50) Proof. The first two identities follow from weak convergence in the two Hilbert structures. By Proposition 2.1, w n → w w_n \to w w n → w strongly in the weighted L q L^q L q space, which yields (4.9). Applying the Brezis–Lieb lemma [4 ] to ∣ x ∣ − b w n |x|^{-b}w_n ∣ x ∣ − b w n gives (4.10).
The following proposition establishes mass recovery for nontrivial weak limits of bounded constrained Palais–Smale sequences and quantifies the possible energy loss due to critical concentration.
Proposition 4.1 (Mass recovery and critical energy defect). Let { w n } ⊂ S \{w_n\}\subset S { w n } ⊂ S be a bounded Palais–Smale sequence for J ∣ S J|_S J ∣ S at level c c c , satisfying P ( w n ) → 0 P(w_n)\to0 P ( w n ) → 0 . Then, up to a subsequence, there exist w ∈ X w\in X w ∈ X and λ ∈ R \lambda\in\mathbb{R} λ ∈ R such that w n ⇀ w w_n\rightharpoonup w w n ⇀ w in X X X , λ n → λ \lambda_n\to\lambda λ n → λ , where λ n \lambda_n λ n denotes the associated Lagrange multiplier.
If w ≠ 0 w\ne0 w = 0 , then w w w satisfies (41) , and
M ( w ) = 1 , M ( w n − w ) → 0 , λ = ( δ q − 1 ) β B ( w ) < 0. (51) M(w)=1,\qquad M(w_n-w)\to0,\qquad\lambda=(\delta_q-1)\beta B(w)<0. \tag*{(51)} M ( w ) = 1 , M ( w n − w ) → 0 , λ = ( δ q − 1 ) β B ( w ) < 0. (51) Whether or not w = 0 w=0 w = 0 , there exists Λ ≥ 0 \Lambda\ge0 Λ ≥ 0 such that
A ( w n − w ) → Λ , κ C ( w n − w ) → Λ , c = J ( w ) + d N Λ . (52) A(w_n-w)\to\Lambda,\qquad\kappa C(w_n-w)\to\Lambda,\qquad c=J(w)+\frac{d}{N}\Lambda. \tag*{(52)} A ( w n − w ) → Λ , κ C ( w n − w ) → Λ , c = J ( w ) + N d Λ. (52) Furthermore, either Λ = 0 \Lambda=0 Λ = 0 , or
Λ ≥ S ( a , b ) N 2 d κ − N − 2 d 2 d , J ( w ) ≤ c − T κ . (53) \Lambda\ge S(a,b)^{\frac{N}{2d}}\kappa^{-\frac{N-2d}{2d}},\qquad J(w)\le c-T_\kappa. \tag*{(53)} Λ ≥ S ( a , b ) 2 d N κ − 2 d N − 2 d , J ( w ) ≤ c − T κ . (53) In particular, if Λ = 0 \Lambda=0 Λ = 0 and c ≠ 0 c\ne0 c = 0 , then w n → w w_n\to w w n → w strongly in X X X and $w\in S.
Proof. Since { w n } \{w_n\} { w n } is bounded in X X X , passing to a subsequence if necessary, we may assume that w n ⇀ w w_n\rightharpoonup w w n ⇀ w in X X X . By the constrained Palais–Smale condition and the standard Lagrange multiplier principle, there exists a sequence { λ n } ⊂ R \{\lambda_n\}\subset\mathbb{R} { λ n } ⊂ R such that, for every φ ∈ X \varphi\in X φ ∈ X ,
∫ ∣ x ∣ − 2 a ∇ w n ⋅ ∇ φ − λ n ∫ ∣ x ∣ − 2 a w n φ − β ∫ ∣ x ∣ − b q ∣ w n ∣ q − 2 w n φ − κ ∫ ∣ x ∣ − b p ∣ w n ∣ p − 2 w n φ = o ( 1 ) ∥ φ ∥ X . (54) \int|x|^{-2a}\nabla w_n\cdot\nabla\varphi-\lambda_n\int|x|^{-2a}w_n\varphi-\beta\int|x|^{-bq}|w_n|^{q-2}w_n\varphi-\kappa\int|x|^{-bp}|w_n|^{p-2}w_n\varphi=o(1)\|\varphi\|_X. \tag*{(54)} ∫ ∣ x ∣ − 2 a ∇ w n ⋅ ∇ φ − λ n ∫ ∣ x ∣ − 2 a w n φ − β ∫ ∣ x ∣ − b q ∣ w n ∣ q − 2 w n φ − κ ∫ ∣ x ∣ − b p ∣ w n ∣ p − 2 w n φ = o ( 1 ) ∥ φ ∥ X . (54)
Testing (54) with w n w_n w n and using M ( w n ) = 1 M(w_n)=1 M ( w n ) = 1 yields
λ n = P ( w n ) + ( δ q − 1 ) β B ( w n ) + o ( 1 ) ⟶ ( δ q − 1 ) β B ( w ) = : λ , (55) \lambda_n=P(w_n)+(\delta_q-1)\beta B(w_n)+o(1)\longrightarrow(\delta_q-1)\beta B(w)=:\lambda, \tag*{(55)} λ n = P ( w n ) + ( δ q − 1 ) β B ( w n ) + o ( 1 ) ⟶ ( δ q − 1 ) β B ( w ) =: λ , (55)
where the convergence follows from P ( w n ) → 0 P(w_n)\to0 P ( w n ) → 0 and Proposition 2.1.
Since { ∣ x ∣ − b w n } \{|x|^{-b}w_n\} { ∣ x ∣ − b w n } is bounded in L p L^p L p and converges almost everywhere to ∣ x ∣ − b w |x|^{-b}w ∣ x ∣ − b w , the corresponding critical nonlinearities converge weakly in X ∗ X^* X ∗ . Using this convergence, the strong convergence of the subcritical nonlinearities in X ∗ X^* X ∗ , and λ n → λ \lambda_n\to\lambda λ n → λ , we pass to the limit in (54) and conclude that w w w satisfies (41) .
If w ≠ 0 w\ne0 w = 0 , Lemma 4.1 yields
P ( w ) = 0 , λ M ( w ) = ( δ q − 1 ) β B ( w ) < 0. P(w)=0,\qquad\lambda M(w)=(\delta_q-1)\beta B(w)<0. P ( w ) = 0 , λ M ( w ) = ( δ q − 1 ) β B ( w ) < 0.
Comparing these identities with (55) , we obtain λ < 0 \lambda<0 λ < 0 and M ( w ) = 1 M(w)=1 M ( w ) = 1 . The Hilbert space splitting of the mass then gives (51) . If w = 0 w=0 w = 0 , the identity P ( w ) = 0 P(w)=0 P ( w ) = 0 also holds.
Set z n = w n − w z_n=w_n-w z n = w n − w . Lemma 4.2, together with P ( w n ) → 0 P(w_n)\to0 P ( w n ) → 0 and P ( w ) = 0 P(w)=0 P ( w ) = 0 , gives
A ( z n ) − κ C ( z n ) = o ( 1 ) , J ( w n ) = J ( w ) + 1 2 A ( z n ) − κ p C ( z n ) + o ( 1 ) . A(z_n)-\kappa C(z_n)=o(1),\qquad J(w_n)=J(w)+\frac{1}{2}A(z_n)-\frac{\kappa}{p}C(z_n)+o(1). A ( z n ) − κ C ( z n ) = o ( 1 ) , J ( w n ) = J ( w ) + 2 1 A ( z n ) − p κ C ( z n ) + o ( 1 ) .
Since { z n } \{z_n\} { z n } is bounded in X X X , passing to a subsequence, we may assume that
A ( z n ) → Λ ≥ 0 , κ C ( z n ) → Λ . A(z_n)\to\Lambda\ge0,\qquad\kappa C(z_n)\to\Lambda. A ( z n ) → Λ ≥ 0 , κ C ( z n ) → Λ.
Since J ( w n ) → c J(w_n)\to c J ( w n ) → c , it follows that
c = J ( w ) + d N Λ c=J(w)+\frac{d}{N}\Lambda c = J ( w ) + N d Λ
which proves (4.12). By (1.1), S ( a , b ) ( Λ κ ) 2 / p ≤ Λ S(a,b)\left(\frac{\Lambda}{\kappa}\right)^{2/p}\leq\Lambda S ( a , b ) ( κ Λ ) 2/ p ≤ Λ . Since p > 2 p>2 p > 2 , it follows that either Λ = 0 \Lambda=0 Λ = 0 or
Λ ≥ S ( a , b ) N 2 d κ − N − 2 d 2 d . \Lambda\geq S(a,b)\frac{N}{2d}\kappa^{-\frac{N-2d}{2d}}. Λ ≥ S ( a , b ) 2 d N κ − 2 d N − 2 d .
In the latter case, (4.12) gives
J ( w ) = c − d N Λ ≤ c − T κ , J(w)=c-\frac{d}{N}\Lambda\leq c-T_\kappa, J ( w ) = c − N d Λ ≤ c − T κ ,
which proves (4.13). Finally, if Λ = 0 \Lambda=0 Λ = 0 and w = 0 w=0 w = 0 , the energy splitting gives c = 0 c=0 c = 0 . Hence, when Λ = 0 \Lambda=0 Λ = 0 and c ≠ 0 c\ne0 c = 0 , we have w ≠ 0 w\ne0 w = 0 . The vanishing gradient defect, together with (4.11), then yields strong convergence in X X X .
Lemma 4.3 (Mass-dependent coercivity). Let 2 < q < q c 2<q<q_c 2 < q < q c . Set
a 0 = d N , b 0 = ( 1 q − δ q p ) K q , s 0 = θ 2 , C 0 = a 0 1 − s 0 s 0 ( b 0 s 0 a 0 ) 1 1 − s 0 . a_0=\frac{d}{N},\qquad b_0=\left(\frac{1}{q}-\frac{\delta_q}{p}\right)K_q,\qquad s_0=\frac{\theta}{2},\qquad C_0=a_0\frac{1-s_0}{s_0}\left(\frac{b_0s_0}{a_0}\right)^{\frac{1}{1-s_0}}. a 0 = N d , b 0 = ( q 1 − p δ q ) K q , s 0 = 2 θ , C 0 = a 0 s 0 1 − s 0 ( a 0 b 0 s 0 ) 1 − s 0 1 . Every w ∈ X w\in X w ∈ X with P ( w ) = 0 P(w)=0 P ( w ) = 0 satisfies
J ( w ) ≥ a 0 A ( w ) − b 0 β M ( w ) q ( 1 − δ q ) / 2 A ( w ) s 0 ≥ − C 0 β 2 / ( 2 − θ ) M ( w ) q ( 1 − δ q ) / ( 2 − θ ) . (56) J(w)\geq a_0A(w)-b_0\beta M(w)^{q(1-\delta_q)/2}A(w)^{s_0}\geq-C_0\beta^{2/(2-\theta)}M(w)^{q(1-\delta_q)/(2-\theta)}. \tag*{(56)} J ( w ) ≥ a 0 A ( w ) − b 0 β M ( w ) q ( 1 − δ q ) /2 A ( w ) s 0 ≥ − C 0 β 2/ ( 2 − θ ) M ( w ) q ( 1 − δ q ) / ( 2 − θ ) . (56) In particular , J J J is coercive as a function of A A A on { w : P ( w ) = 0 , M ( w ) ≤ 1 } \{w:P(w)=0,M(w)\leq1\} { w : P ( w ) = 0 , M ( w ) ≤ 1 } .
Proof. Combining (3.2) with (2.6), we obtain
J ( w ) ≥ a 0 A ( w ) − b 0 β M ( w ) q ( 1 − δ q ) / 2 A ( w ) s 0 . J(w)\geq a_0A(w)-b_0\beta M(w)^{q(1-\delta_q)/2}A(w)^{s_0}. J ( w ) ≥ a 0 A ( w ) − b 0 β M ( w ) q ( 1 − δ q ) /2 A ( w ) s 0 . Since 0 < s 0 < 1 0<s_0<1 0 < s 0 < 1 , minimizing the right-hand side over A ( w ) ≥ 0 A(w)\geq0 A ( w ) ≥ 0 yields the second estimate in (4.16). The first estimate also gives coercivity for M ( w ) ≤ 1 M(w)\leq1 M ( w ) ≤ 1 .
Two solutions in the mass-subcritical case {#sec:5} In this section, we prove the existence of two solutions with opposite energies in the mass-subcritical regime, corresponding to a local minimizer and an upper Pohozaev branch minimizer.
Throughout this section, we assume that
2 < q < q c , 0 < β < β 0 . 2<q<q_c,\qquad0<\beta<\beta_0. 2 < q < q c , 0 < β < β 0 . The inequalities (1.1) and (2.6) imply
J ( w ) ≥ f β ( A ( w ) 1 / 2 ) , w ∈ S . (57) J(w)\geq f_\beta(A(w)^{1/2}),\qquad w\in S. \tag*{(57)} J ( w ) ≥ f β ( A ( w ) 1/2 ) , w ∈ S . (57) Since β < β 0 ≤ β b \beta<\beta_0\leq\beta_b β < β 0 ≤ β b , the definitions of R ∗ R_* R ∗ and β b \beta_b β b yield
f β ( R ∗ ) ≥ 1 4 R ∗ 2 > 0. (58) f_\beta(R_*)\geq\frac{1}{4}R_*^2>0. \tag*{(58)} f β ( R ∗ ) ≥ 4 1 R ∗ 2 > 0. (58) Define
G = { w ∈ S : A ( w ) 1 / 2 < R ∗ } , m + = inf G J . \mathcal{G}=\{w\in S:A(w)^{1/2}<R_*\},\qquad m^+=\inf_{\mathcal{G}}J. G = { w ∈ S : A ( w ) 1/2 < R ∗ } , m + = G inf J . Proposition 5.1. Every minimizing sequence for m + m^+ m + has a strongly convergent subsequence in X X X . This level is attained by a nonnegative local minimizer w + w^+ w + of J ∣ S J|_S J ∣ S , and
J ( w + ) = m + < 0 , w + ∈ P + , m + = inf P + J . (59) J(w^+)=m^+<0,\qquad w^+\in\mathcal{P}^+,\qquad m^+=\inf_{\mathcal{P}^+}J. \tag*{(59)} J ( w + ) = m + < 0 , w + ∈ P + , m + = P + inf J . (59) Every point of P + \mathcal{P}^+ P + belongs to $\mathcal{G}.
Proof. By (57) and the definition of G \mathcal{G} G , we have m + ≥ min 0 ≤ s ≤ R ∗ f β ( s ) > − ∞ m^{+} \ge\min_{0\le s\le R_{*}} f_{\beta}(s)>-\infty m + ≥ min 0 ≤ s ≤ R ∗ f β ( s ) > − ∞ . For any w ∈ S w\in S w ∈ S , as t → − ∞ t\to-\infty t → − ∞ ,
J ( t ⋆ w ) = − β q e θ t B ( w ) + o ( e θ t ) < 0 , A ( t ⋆ w ) 1 / 2 → 0. J(t\star w)=-\frac{\beta}{q}e^{\theta t}B(w)+o(e^{\theta t})<0,\qquad A(t\star w)^{1/2}\to0. J ( t ⋆ w ) = − q β e θ t B ( w ) + o ( e θ t ) < 0 , A ( t ⋆ w ) 1/2 → 0.
Hence m + < 0 m^{+}<0 m + < 0 . By the continuity of f β f_{\beta} f β and (58) , there exists ε ∗ > 0 \varepsilon_{*}>0 ε ∗ > 0 such that
J ( w ) > 0 whenever R ∗ − ε ∗ ≤ A ( w ) 1 / 2 ≤ R ∗ , w ∈ S . J(w)>0\quad\text{whenever}\quad R_{*}-\varepsilon_{*}\le A(w)^{1/2}\le R_{*},\qquad w\in S. J ( w ) > 0 whenever R ∗ − ε ∗ ≤ A ( w ) 1/2 ≤ R ∗ , w ∈ S .
Thus, any minimizing sequence { w n } ⊂ G \{w_{n}\}\subset\mathcal{G} { w n } ⊂ G satisfies A ( w n ) 1 / 2 < R ∗ − ε ∗ A(w_{n})^{1/2}<R_{*}-\varepsilon_{*} A ( w n ) 1/2 < R ∗ − ε ∗ for all sufficiently large n n n . Since { w n } \{w_{n}\} { w n } is bounded in X X X , up to a subsequence, we have w n ⇀ w w_{n}\rightharpoonup w w n ⇀ w in X X X , A ( w ) 1 / 2 ≤ R ∗ − ε ∗ A(w)^{1/2}\le R_{*}-\varepsilon_{*} A ( w ) 1/2 ≤ R ∗ − ε ∗ . Set z n = w n − w z_{n}=w_{n}-w z n = w n − w . Then
A ( z n ) = A ( w n ) − A ( w ) + o ( 1 ) ≤ R ∗ 2 + o ( 1 ) . A(z_{n})=A(w_{n})-A(w)+o(1)\le R_{*}^{2}+o(1). A ( z n ) = A ( w n ) − A ( w ) + o ( 1 ) ≤ R ∗ 2 + o ( 1 ) .
It follows from Lemma 4.2 and (1) that
J ( w n ) − J ( w ) = 1 2 A ( z n ) − κ p C ( z n ) + o ( 1 ) ≥ ( 1 2 − κ R ∗ p − 2 p S ( a , b ) p / 2 ) A ( z n ) + o ( 1 ) = c 2 A ( z n ) + o ( 1 ) , (60) \begin{aligned}
J(w_{n})-J(w)
&=\frac{1}{2}A(z_{n})-\frac{\kappa}{p}C(z_{n})+o(1)\\
&\ge\left(\frac{1}{2}-\frac{\kappa R_{*}^{p-2}}{pS(a,b)^{p/2}}\right)A(z_{n})+o(1)\\
&=c_{2}A(z_{n})+o(1), \tag*{(60)}
\end{aligned} J ( w n ) − J ( w ) = 2 1 A ( z n ) − p κ C ( z n ) + o ( 1 ) ≥ ( 2 1 − pS ( a , b ) p /2 κ R ∗ p − 2 ) A ( z n ) + o ( 1 ) = c 2 A ( z n ) + o ( 1 ) , (60)
where c 2 > 0 c_{2}>0 c 2 > 0 by (31) . Consequently, J ( w ) ≤ m + < 0 J(w)\le m^{+}<0 J ( w ) ≤ m + < 0 , and hence w ≠ 0 w\ne0 w = 0 .
Write M ( w ) = σ 2 ≤ 1 M(w)=\sigma^{2}\le1 M ( w ) = σ 2 ≤ 1 . If σ < 1 \sigma<1 σ < 1 , define
w ^ ( x ) = σ ( N − 2 − 2 a ) / 2 w ( σ x ) . \widehat{w}(x)=\sigma^{(N-2-2a)/2}w(\sigma x). w ( x ) = σ ( N − 2 − 2 a ) /2 w ( σ x ) .
The critical scaling properties yield
M ( w ^ ) = 1 , A ( w ^ ) = A ( w ) , C ( w ^ ) = C ( w ) , B ( w ^ ) = σ − q ( 1 − δ q ) B ( w ) . M(\widehat{w})=1,\qquad A(\widehat{w})=A(w),\qquad C(\widehat{w})=C(w),\qquad B(\widehat{w})=\sigma^{-q(1-\delta_{q})}B(w). M ( w ) = 1 , A ( w ) = A ( w ) , C ( w ) = C ( w ) , B ( w ) = σ − q ( 1 − δ q ) B ( w ) .
Since q ( 1 − δ q ) > 0 q(1-\delta_{q})>0 q ( 1 − δ q ) > 0 and B ( w ) > 0 B(w)>0 B ( w ) > 0 , we have w ^ ∈ G \widehat{w}\in\mathcal{G} w ∈ G and
J ( w ^ ) = J ( w ) − β q ( σ − q ( 1 − δ q ) − 1 ) B ( w ) < J ( w ) ≤ m + , J(\widehat{w})=J(w)-\frac{\beta}{q}\left(\sigma^{-q(1-\delta_{q})}-1\right)B(w)<J(w)\le m^{+}, J ( w ) = J ( w ) − q β ( σ − q ( 1 − δ q ) − 1 ) B ( w ) < J ( w ) ≤ m + ,
a contradiction. Thus M ( w ) = 1 M(w)=1 M ( w ) = 1 and J ( w ) = m + J(w)=m^{+} J ( w ) = m + . Together with J ( w n ) → m + J(w_{n})\to m^{+} J ( w n ) → m + , (60) and the mass splitting yield A ( z n ) → 0 A(z_{n})\to0 A ( z n ) → 0 and M ( z n ) → 0 M(z_{n})\to0 M ( z n ) → 0 , respectively. Hence w n → w w_{n}\to w w n → w strongly in X X X .
As an interior minimizer of J J J on G \mathcal{G} G relative to S S S , w w w is a constrained critical point of J J J on S S S . Lemma 4.1 gives P ( w ) = 0 P(w)=0 P ( w ) = 0 , and J ( w ) = m + < 0 J(w)=m^{+}<0 J ( w ) = m + < 0 then implies w ∈ P + w\in P^{+} w ∈ P + by Lemma 3.1. Since replacing w w w by ∣ w ∣ |w| ∣ w ∣ preserves A A A , M M M , B B B , and C C C , the minimizer may be chosen nonnegative.
Finally, let z ∈ P + z\in P^{+} z ∈ P + . Since the lower fiber critical point occurs at t = 0 t=0 t = 0 , we have J ( t ⋆ z ) < 0 J(t\star z)<0 J ( t ⋆ z ) < 0 for all t ≤ 0 t\le0 t ≤ 0 . On the other hand, (35) gives J ( t R ⋆ z ) > 0 J(t_{R}\star z)>0 J ( t R ⋆ z ) > 0 , where e t R A ( z ) 1 / 2 = R ∗ e^{t_{R}}A(z)^{1/2}=R_{*} e t R A ( z ) 1/2 = R ∗ . Hence t R > 0 t_{R}>0 t R > 0 , which implies A ( z ) 1 / 2 < R ∗ A(z)^{1/2}<R_{*} A ( z ) 1/2 < R ∗ and thus P + ⊂ G P^{+}\subset\mathcal{G} P + ⊂ G . Together with the minimizer obtained above, this yields the last identity in (59) .
We define the upper-branch energy level by
m − : = inf w ∈ P − J ( w ) . m^{-}:=\inf_{w\in P^{-}}J(w). m − := w ∈ P − inf J ( w ) .
Lemma 5.1. The upper-branch energy level satisfies
m − ≥ R ∗ 2 4 > 0. m^{-}\ge\frac{R_{*}^{2}}{4}>0. m − ≥ 4 R ∗ 2 > 0.
Proof. For w ∈ P − w \in\mathcal{P}^{-} w ∈ P − , Lemma 3.1 and (35) give
J ( w ) = max t J ( t ⋆ w ) ≥ f β ( R ∗ ) ≥ R ∗ 2 / 4. J(w)=\max_{t}J(t\star w)\ge f_{\beta}(R_{*})\ge R_{*}^{2}/4. J ( w ) = t max J ( t ⋆ w ) ≥ f β ( R ∗ ) ≥ R ∗ 2 /4.
Take the infimum over P − \mathcal{P}^{-} P − .
For ε > 0 \varepsilon>0 ε > 0 , define the rescaling of the finite-mass extremal W W W by
W ε ( x ) = ε − ( N − 2 − 2 a ) / 2 W ( x / ε ) . W_{\varepsilon}(x)=\varepsilon^{-(N-2-2a)/2}W(x/\varepsilon). W ε ( x ) = ε − ( N − 2 − 2 a ) /2 W ( x / ε ) .
Then
A ( W ε ) = A ( W ) , C ( W ε ) = C ( W ) , M ( W ε ) = ε 2 M ( W ) , B ( W ε ) = ε q ( 1 − δ q ) B ( W ) . A(W_{\varepsilon})=A(W),\quad C(W_{\varepsilon})=C(W),\quad M(W_{\varepsilon})=\varepsilon^{2}M(W),\quad B(W_{\varepsilon})=\varepsilon^{q(1-\delta_q)}B(W). A ( W ε ) = A ( W ) , C ( W ε ) = C ( W ) , M ( W ε ) = ε 2 M ( W ) , B ( W ε ) = ε q ( 1 − δ q ) B ( W ) .
Let
J 0 ( w ) = 1 2 A ( w ) − κ p C ( w ) . J_{0}(w)=\frac{1}{2}A(w)-\frac{\kappa}{p}C(w). J 0 ( w ) = 2 1 A ( w ) − p κ C ( w ) .
The function W W W is an extremal for the sharp CKN inequality (1.1), so
max s > 0 J 0 ( s W ) = T κ . (61) \max_{s>0}J_{0}(sW)=T_{\kappa}. \tag*{(61)} s > 0 max J 0 ( s W ) = T κ . (61)
Lemma 5.2. Let w + w^{+} w + be the minimizer obtained in Proposition 5.1. Then the upper Pohozaev branch level satisfies
m − < m + + T κ . m^{-}<m^{+}+T_{\kappa}. m − < m + + T κ .
Proof. Recall that the Lagrange multiplier associated with w + w^{+} w + satisfies λ = ( δ q − 1 ) β B ( w + ) < 0 \lambda=(\delta_q-1)\beta B(w^{+})<0 λ = ( δ q − 1 ) β B ( w + ) < 0 . For ε > 0 \varepsilon>0 ε > 0 and s ≥ 0 s\ge0 s ≥ 0 , set
h ε , s = w + + s W ε , τ ε , s = M ( h ε , s ) 1 / 2 , h_{\varepsilon,s}=w^{+}+sW_{\varepsilon},\qquad\tau_{\varepsilon,s}=M(h_{\varepsilon,s})^{1/2}, h ε , s = w + + s W ε , τ ε , s = M ( h ε , s ) 1/2 ,
and define the normalized path
v ε , s ( x ) = τ ε , s ( N − 2 − 2 a ) / 2 h ε , s ( τ ε , s x ) . v_{\varepsilon,s}(x)=\tau_{\varepsilon,s}^{(N-2-2a)/2}h_{\varepsilon,s}(\tau_{\varepsilon,s}x). v ε , s ( x ) = τ ε , s ( N − 2 − 2 a ) /2 h ε , s ( τ ε , s x ) .
The scaling properties of the critical dilation yield
M ( v ε , s ) = 1 , A ( v ε , s ) = A ( h ε , s ) , C ( v ε , s ) = C ( h ε , s ) , B ( v ε , s ) = τ ε , s − q ( 1 − δ q ) B ( h ε , s ) . (62) M(v_{\varepsilon,s})=1,\quad A(v_{\varepsilon,s})=A(h_{\varepsilon,s}),\quad C(v_{\varepsilon,s})=C(h_{\varepsilon,s}),\quad B(v_{\varepsilon,s})=\tau_{\varepsilon,s}^{-q(1-\delta_q)}B(h_{\varepsilon,s}). \tag*{(62)} M ( v ε , s ) = 1 , A ( v ε , s ) = A ( h ε , s ) , C ( v ε , s ) = C ( h ε , s ) , B ( v ε , s ) = τ ε , s − q ( 1 − δ q ) B ( h ε , s ) . (62)
Set
m ε = ∫ ∣ x ∣ − 2 a w + W ε , Q ε = ∫ ∣ x ∣ − b q ( w + ) q − 1 W ε . m_{\varepsilon}=\int|x|^{-2a}w^{+}W_{\varepsilon},\qquad Q_{\varepsilon}=\int|x|^{-bq}(w^{+})^{q-1}W_{\varepsilon}. m ε = ∫ ∣ x ∣ − 2 a w + W ε , Q ε = ∫ ∣ x ∣ − b q ( w + ) q − 1 W ε .
The Hölder inequality and the scaling properties of W ε W_{\varepsilon} W ε give
0 ≤ m ε ≤ ε M ( W ) 1 / 2 , 0 ≤ Q ε ≤ ε 1 − δ q B ( w + ) ( q − 1 ) / q B ( W ) 1 / q . 0\le m_{\varepsilon}\le\varepsilon M(W)^{1/2},\qquad0\le Q_{\varepsilon}\le\varepsilon^{1-\delta_q}B(w^{+})^{(q-1)/q}B(W)^{1/q}. 0 ≤ m ε ≤ εM ( W ) 1/2 , 0 ≤ Q ε ≤ ε 1 − δ q B ( w + ) ( q − 1 ) / q B ( W ) 1/ q .
Moreover, expanding M ( h ε , s ) M(h_{\varepsilon,s}) M ( h ε , s ) gives
τ ε , s 2 = 1 + 2 s m ε + s 2 ε 2 M ( W ) . (63) \tau_{\varepsilon,s}^{2}=1+2sm_{\varepsilon}+s^{2}\varepsilon^{2}M(W). \tag*{(63)} τ ε , s 2 = 1 + 2 s m ε + s 2 ε 2 M ( W ) . (63)
For r ≥ 2 r\ge2 r ≥ 2 and x , y ≥ 0 x,y\ge0 x , y ≥ 0 ,
( x + y ) r ≥ x r + y r + r x r − 1 y . (64) (x+y)^{r}\ge x^{r}+y^{r}+rx^{r-1}y. \tag*{(64)} ( x + y ) r ≥ x r + y r + r x r − 1 y . (64)
Testing the equation for w + w^{+} w + with W ε W_{\varepsilon} W ε gives
∫ ∣ x ∣ − 2 a ∇ w + ⋅ ∇ W ε = λ m ε + β Q ε + κ ∫ ∣ x ∣ − b p ( w + ) p − 1 W ε . (65) \int|x|^{-2a}\nabla w^{+}\cdot\nabla W_{\varepsilon}
=\lambda m_{\varepsilon}+\beta Q_{\varepsilon}
+\kappa\int|x|^{-bp}(w^{+})^{p-1}W_{\varepsilon}. \tag*{(65)} ∫ ∣ x ∣ − 2 a ∇ w + ⋅ ∇ W ε = λ m ε + β Q ε + κ ∫ ∣ x ∣ − b p ( w + ) p − 1 W ε . (65)
Applying (64) to the q q q - and p p p -power terms and using (65) , we obtain, uniformly for s s s in bounded intervals,
J ( v ε , s ) − J ( w + ) − J 0 ( s W ) ≤ s λ m ε + β q ( 1 − τ ε , s − q ( 1 − δ q ) ) B ( w + ) + β s ( 1 − τ ε , s − q ( 1 − δ q ) ) Q ε − β q τ ε , s − q ( 1 − δ q ) s q ε q ( 1 − δ q ) B ( W ) . (66) J(v_{\varepsilon,s})-J(w^+)-J_0(sW)\le s\lambda m_\varepsilon+\frac{\beta}{q}(1-\tau_{\varepsilon,s}^{-q(1-\delta_q)})B(w^+)
+\beta s(1-\tau_{\varepsilon,s}^{-q(1-\delta_q)})Q_\varepsilon-\frac{\beta}{q}\tau_{\varepsilon,s}^{-q(1-\delta_q)}s^q\varepsilon^{q(1-\delta_q)}B(W). \tag*{(66)} J ( v ε , s ) − J ( w + ) − J 0 ( s W ) ≤ s λ m ε + q β ( 1 − τ ε , s − q ( 1 − δ q ) ) B ( w + ) + β s ( 1 − τ ε , s − q ( 1 − δ q ) ) Q ε − q β τ ε , s − q ( 1 − δ q ) s q ε q ( 1 − δ q ) B ( W ) . (66)
To estimate the first two terms, we use
1 − x − q ( 1 − δ q ) / 2 ≤ q ( 1 − δ q ) 2 ( x − 1 ) , x ≥ 1. 1-x^{-q(1-\delta_q)/2}\le\frac{q(1-\delta_q)}{2}(x-1),\qquad x\ge1. 1 − x − q ( 1 − δ q ) /2 ≤ 2 q ( 1 − δ q ) ( x − 1 ) , x ≥ 1.
Since τ ε , s 2 ≥ 1 \tau_{\varepsilon,s}^2\ge1 τ ε , s 2 ≥ 1 and
− λ = β ( 1 − δ q ) B ( w + ) , -\lambda=\beta(1-\delta_q)B(w^+), − λ = β ( 1 − δ q ) B ( w + ) ,
it follows from (63) that
s λ m ε + β q ( 1 − τ ε , s − q ( 1 − δ q ) ) B ( w + ) ≤ − λ 2 s 2 ε 2 M ( W ) . (67) s\lambda m_\varepsilon+\frac{\beta}{q}(1-\tau_{\varepsilon,s}^{-q(1-\delta_q)})B(w^+)\le-\frac{\lambda}{2}s^2\varepsilon^2M(W). \tag*{(67)} s λ m ε + q β ( 1 − τ ε , s − q ( 1 − δ q ) ) B ( w + ) ≤ − 2 λ s 2 ε 2 M ( W ) . (67)
Moreover, the estimates for m ε m_\varepsilon m ε and Q ε Q_\varepsilon Q ε , together with (63) , yield
0 ≤ β s ( 1 − τ ε , s − q ( 1 − δ q ) ) Q ε ≤ C ( ε 2 − δ q + ε 3 − δ q ) , (68) 0\le\beta s(1-\tau_{\varepsilon,s}^{-q(1-\delta_q)})Q_\varepsilon\le C\left(\varepsilon^{2-\delta_q}+\varepsilon^{3-\delta_q}\right), \tag*{(68)} 0 ≤ β s ( 1 − τ ε , s − q ( 1 − δ q ) ) Q ε ≤ C ( ε 2 − δ q + ε 3 − δ q ) , (68)
uniformly for s s s in bounded intervals. We next establish an estimate that holds uniformly down to s = 0 s=0 s = 0 . Since 2 < q < q c < p 2<q<q_c<p 2 < q < q c < p , we have
0 < q ( 1 − δ q ) = N − N − 2 d 2 q < 2 d < 2. 0<q(1-\delta_q)=N-\frac{N-2d}{2}q<2d<2. 0 < q ( 1 − δ q ) = N − 2 N − 2 d q < 2 d < 2.
Moreover,
2 − δ q − q ( 1 − δ q ) = 1 − d + ( q − 2 ) ( N ( q − 1 ) 2 q − d ) > 1 − d > 0 , 2-\delta_q-q(1-\delta_q)=1-d+(q-2)\left(\frac{N(q-1)}{2q}-d\right)>1-d>0, 2 − δ q − q ( 1 − δ q ) = 1 − d + ( q − 2 ) ( 2 q N ( q − 1 ) − d ) > 1 − d > 0 ,
since N ( q − 1 ) 2 q > N 4 > d \frac{N(q-1)}{2q}>\frac{N}{4}>d 2 q N ( q − 1 ) > 4 N > d . Consequently,
ζ 0 : = min { 2 − q ( 1 − δ q ) , 2 − δ q − q ( 1 − δ q ) } > 0. \zeta_0:=\min\{2-q(1-\delta_q),2-\delta_q-q(1-\delta_q)\}>0. ζ 0 := min { 2 − q ( 1 − δ q ) , 2 − δ q − q ( 1 − δ q )} > 0.
For any fixed L > 0 L>0 L > 0 , (63) and m ε = O ( ε ) m_\varepsilon=O(\varepsilon) m ε = O ( ε ) yield
1 ≤ τ ε , s 2 ≤ 1 + C L ε , 0 ≤ s ≤ L , 1\le\tau_{\varepsilon,s}^2\le1+C_L\varepsilon,\qquad0\le s\le L, 1 ≤ τ ε , s 2 ≤ 1 + C L ε , 0 ≤ s ≤ L ,
where C L > 0 C_L>0 C L > 0 is independent of s s s and ε \varepsilon ε . Consequently, τ ε , s − q ( 1 − δ q ) ≥ 1 2 \tau_{\varepsilon,s}^{-q(1-\delta_q)}\ge\frac{1}{2} τ ε , s − q ( 1 − δ q ) ≥ 2 1 for sufficiently small ε \varepsilon ε , uniformly in s ∈ [ 0 , L ] s\in[0,L] s ∈ [ 0 , L ] . Combining (66) , (67) , and (68) , we obtain
J ( v ε , s ) ≤ m + + J 0 ( s W ) − β 2 q s q B ( W ) ε q ( 1 − δ q ) + C L ε q ( 1 − δ q ) + ζ 0 , 0 ≤ s ≤ L . (69) J(v_{\varepsilon,s})\le m^++J_0(sW)-\frac{\beta}{2q}s^qB(W)\varepsilon^{q(1-\delta_q)}+C_L\varepsilon^{q(1-\delta_q)+\zeta_0},\qquad0\le s\le L. \tag*{(69)} J ( v ε , s ) ≤ m + + J 0 ( s W ) − 2 q β s q B ( W ) ε q ( 1 − δ q ) + C L ε q ( 1 − δ q ) + ζ 0 , 0 ≤ s ≤ L . (69)
Since (69) is valid only for bounded s s s and its negative correction degenerates as s → 0 s\to0 s → 0 , we first choose the path parameters independently of ε \varepsilon ε and then treat different ranges of s s s separately. By (62) and the nonnegativity of w + w^+ w + and W ε W_\varepsilon W ε , we have
J ( v ε , s ) ≤ H ( s ) : = A ( w + ) + s 2 A ( W ) − κ p s p C ( W ) , s ≥ 0. J(v_{\varepsilon,s})\le H(s):=A(w^+)+s^2A(W)-\frac{\kappa}{p}s^pC(W),\qquad s\ge0. J ( v ε , s ) ≤ H ( s ) := A ( w + ) + s 2 A ( W ) − p κ s p C ( W ) , s ≥ 0.
Let s ∗ > 0 s_*>0 s ∗ > 0 be the unique maximizer in (61) . Since H ( s ) → − ∞ H(s)\to-\infty H ( s ) → − ∞ as s → ∞ s\to\infty s → ∞ , we may fix s 1 > s ∗ s_1>s_* s 1 > s ∗ sufficiently large such that
H ( s ) ≤ m + + T κ − 1 , s ≥ 2 s 1 . H(s)\le m^++T_\kappa-1,\qquad s\ge2s_1. H ( s ) ≤ m + + T κ − 1 , s ≥ 2 s 1 .
Choose s 0 ∈ ( 0 , s ∗ ) s_0 \in(0,s_*) s 0 ∈ ( 0 , s ∗ ) and set
ζ = 1 4 min { 1 , T κ − J 0 ( s 0 W ) , T κ − J 0 ( s 1 W ) } > 0. \zeta=\frac{1}{4}\min\{1,T_\kappa-J_0(s_0W),T_\kappa-J_0(s_1W)\}>0. ζ = 4 1 min { 1 , T κ − J 0 ( s 0 W ) , T κ − J 0 ( s 1 W )} > 0.
By the monotonicity of the pure critical fiber,
J 0 ( s W ) ≤ T κ − 3 ζ , s ∈ [ 0 , s 0 ] ∪ [ s 1 , 2 s 1 ] . (70) J_0(sW)\leq T_\kappa-3\zeta,\qquad s\in[0,s_0]\cup[s_1,2s_1]. \tag*{(70)} J 0 ( s W ) ≤ T κ − 3 ζ , s ∈ [ 0 , s 0 ] ∪ [ s 1 , 2 s 1 ] . (70)
Applying (5.13) with L = 2 s 1 L=2s_1 L = 2 s 1 , we choose ε 0 ∈ ( 0 , 1 ) \varepsilon_0\in(0,1) ε 0 ∈ ( 0 , 1 ) sufficiently small such that the estimate holds for 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 and
C 2 s 1 ε 0 ζ 0 ≤ β 4 q s 0 q B ( W ) , C 2 s 1 ε 0 q ( 1 − δ q ) + ζ 0 ≤ ζ . C_{2s_1}\varepsilon_0^{\zeta_0}\leq\frac{\beta}{4q}s_0^qB(W),\qquad C_{2s_1}\varepsilon_0^{q(1-\delta_q)+\zeta_0}\leq\zeta. C 2 s 1 ε 0 ζ 0 ≤ 4 q β s 0 q B ( W ) , C 2 s 1 ε 0 q ( 1 − δ q ) + ζ 0 ≤ ζ .
For s ∈ [ s 0 , s 1 ] s\in[s_0,s_1] s ∈ [ s 0 , s 1 ] ,
J ( v ε , s ) ≤ m + + T κ − β 4 q s 0 q B ( W ) ε q ( 1 − δ q ) . J(v_{\varepsilon,s})\leq m^++T_\kappa-\frac{\beta}{4q}s_0^qB(W)\varepsilon^{q(1-\delta_q)}. J ( v ε , s ) ≤ m + + T κ − 4 q β s 0 q B ( W ) ε q ( 1 − δ q ) .
On [ 0 , s 0 ] ∪ [ s 1 , 2 s 1 ] [0,s_0]\cup[s_1,2s_1] [ 0 , s 0 ] ∪ [ s 1 , 2 s 1 ] , (5.13) and (5.14) give
J ( v ε , s ) ≤ m + + T κ − 2 ζ . J(v_{\varepsilon,s})\leq m^++T_\kappa-2\zeta. J ( v ε , s ) ≤ m + + T κ − 2 ζ .
For s ≥ 2 s 1 s\geq2s_1 s ≥ 2 s 1 , the bound involving H H H yields
J ( v ε , s ) ≤ m + + T κ − 1. J(v_{\varepsilon,s})\leq m^++T_\kappa-1. J ( v ε , s ) ≤ m + + T κ − 1.
Therefore, for every 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 ,
sup s ≥ 0 J ( v ε , s ) < m + + T κ . (71) \sup_{s\geq0}J(v_{\varepsilon,s})<m^++T_\kappa. \tag*{(71)} s ≥ 0 sup J ( v ε , s ) < m + + T κ . (71)
It remains to show that the path intersects P − \mathcal{P}^- P − . Let t 2 ( v ε , s ) t_2(v_{\varepsilon,s}) t 2 ( v ε , s ) be the upper fiber parameter given by Lemma 3.1. Since v ε , 0 = w + ∈ P + v_{\varepsilon,0}=w^+\in\mathcal{P}^+ v ε , 0 = w + ∈ P + , we have t 2 ( v ε , 0 ) > 0 t_2(v_{\varepsilon,0})>0 t 2 ( v ε , 0 ) > 0 . The Pohozaev identity at the upper fiber critical point gives
e ( p − 2 ) t 2 ( v ε , s ) ≤ A ( v ε , s ) κ C ( v ε , s ) . (72) e^{(p-2)t_2(v_{\varepsilon,s})}\leq\frac{A(v_{\varepsilon,s})}{\kappa C(v_{\varepsilon,s})}. \tag*{(72)} e ( p − 2 ) t 2 ( v ε , s ) ≤ κ C ( v ε , s ) A ( v ε , s ) . (72)
Moreover,
A ( v ε , s ) ≤ 2 A ( w + ) + 2 s 2 A ( W ) , C ( v ε , s ) ≥ s p C ( W ) . A(v_{\varepsilon,s})\leq2A(w^+)+2s^2A(W),\qquad C(v_{\varepsilon,s})\geq s^pC(W). A ( v ε , s ) ≤ 2 A ( w + ) + 2 s 2 A ( W ) , C ( v ε , s ) ≥ s p C ( W ) .
Consequently, for s ≥ 1 s\geq1 s ≥ 1 ,
e ( p − 2 ) t 2 ( v ε , s ) ≤ C s 2 − p , e^{(p-2)t_2(v_{\varepsilon,s})}\leq Cs^{2-p}, e ( p − 2 ) t 2 ( v ε , s ) ≤ C s 2 − p ,
where C > 0 C>0 C > 0 is independent of ε \varepsilon ε and s s s . Thus
t 2 ( v ε , s ) ≤ log C p − 2 − log s . t_2(v_{\varepsilon,s})\leq\frac{\log C}{p-2}-\log s. t 2 ( v ε , s ) ≤ p − 2 log C − log s .
In particular, there exists L 1 > 0 L_1>0 L 1 > 0 , independent of ε \varepsilon ε , such that t 2 ( v ε , L 1 ) < 0 t_2(v_{\varepsilon,L_1})<0 t 2 ( v ε , L 1 ) < 0 . By continuity, there exists s ε ∈ ( 0 , L 1 ) s_\varepsilon\in(0,L_1) s ε ∈ ( 0 , L 1 ) such that
t 2 ( v ε , s ε ) = 0 , v ε , s ε ∈ P − . t_2(v_{\varepsilon,s_\varepsilon})=0,\qquad v_{\varepsilon,s_\varepsilon}\in\mathcal{P}^-. t 2 ( v ε , s ε ) = 0 , v ε , s ε ∈ P − .
Combining this with (5.15), we conclude that
m − ≤ J ( v ε , s ε ) ≤ sup s ≥ 0 J ( v ε , s ) < m + + T κ . m^-\leq J(v_{\varepsilon,s_\varepsilon})\leq\sup_{s\geq0}J(v_{\varepsilon,s})<m^++T_\kappa. m − ≤ J ( v ε , s ε ) ≤ s ≥ 0 sup J ( v ε , s ) < m + + T κ .
Proposition 5.2. The level m − m^{-} m − is attained by a nonnegative constrained critical point w − ∈ P − w^{-} \in\mathcal{P}^{-} w − ∈ P − of J ∣ S J|_{\mathcal{S}} J ∣ S .
Proof. Let { w n } ⊂ P − \{w_n\} \subset\mathcal{P}^{-} { w n } ⊂ P − be a minimizing sequence for m − m^{-} m − . By (56) , the sequence { w n } \{w_n\} { w n } is bounded in X X X . Proposition 3.1 then yields a bounded Palais–Smale sequence, still denoted by { w n } \{w_n\} { w n } , satisfying (40) with c = m − c=m^{-} c = m − .
If critical loss occurs, Proposition 4.1 and Lemma 5.2 give
J ( w ) ≤ m − − T κ < m + < 0. J(w) \leq m^{-}-T_{\kappa}<m^{+}<0. J ( w ) ≤ m − − T κ < m + < 0.
Thus w ≠ 0 w\neq0 w = 0 . By (51) , we have M ( w ) = 1 M(w)=1 M ( w ) = 1 , while Lemma 4.1 gives P ( w ) = 0 P(w)=0 P ( w ) = 0 . Since J ( w ) < 0 J(w)<0 J ( w ) < 0 , the fiber geometry in Lemma 3.1 implies that w ∈ P + w\in\mathcal{P}^{+} w ∈ P + . Therefore, Proposition 5.1 yields
m + ≤ J ( w ) < m + , m^{+}\leq J(w)<m^{+}, m + ≤ J ( w ) < m + ,
which is impossible. Since m − > 0 m^{-}>0 m − > 0 , the compactness alternative yields, up to a subsequence, w n → w − w_n\to w^{-} w n → w − strongly in X X X for a nontrivial limit w − w^{-} w − . Moreover, P ( w − ) = 0 P(w^{-})=0 P ( w − ) = 0 , and (36) implies that w − ∈ P − w^{-}\in\mathcal{P}^{-} w − ∈ P − . Consequently,
J ( w − ) = m − . J(w^{-})=m^{-}. J ( w − ) = m − .
Finally, replacing w − w^{-} w − by ∣ w − ∣ |w^{-}| ∣ w − ∣ preserves the four scalar quantities appearing in J J J and P P P . Hence ∣ w − ∣ |w^{-}| ∣ w − ∣ is also a minimizer on P − \mathcal{P}^{-} P − . By Lemma 3.4 , it is a constrained critical point as well. □ \square □
Proof of Theorem 1.1 (i). Propositions 5.1 and 5.2 yield scalar solutions w + w^{+} w + and w − w^{-} w − with negative and positive energies, respectively. Lemma 4.1 gives their positivity away from the origin and the multiplier formula (11) . Proposition 2.2 lifts these scalar solutions to synchronized normalized solutions.
The map w ↦ ( ρ 1 w , ρ 2 w ) w\mapsto(\rho_1w,\rho_2w) w ↦ ( ρ 1 w , ρ 2 w ) is a homeomorphism from S \mathcal{S} S onto S s y n \mathcal{S}_{\mathrm{syn}} S syn . By (27) , it transfers the local minimizer on S \mathcal{S} S and the minimizer on P − \mathcal{P}^{-} P − to the corresponding minimizers on S s y n \mathcal{S}_{\mathrm{syn}} S syn and P s y n − \mathcal{P}_{\mathrm{syn}}^{-} P syn − , and yields (8) . □ \square □
The mass-critical and mass-supercritical cases
In this section, we consider the case
q c ≤ q < p . q_c\leq q<p. q c ≤ q < p .
By Lemma 3.2 , each fiber map admits a unique critical point, which is the global maximum. Hence the variational analysis is performed on the whole Pohozaev set P \mathcal{P} P .
In the mass-critical case, assume that (9) holds. Define
m = inf P J . m=\inf_{\mathcal{P}}J. m = P inf J .
Lemma 6.1. The variational level satisfies m > 0 m>0 m > 0 .
Proof. Suppose first that q > q c q>q_c q > q c . Then θ > 2 \theta>2 θ > 2 . For any w ∈ P w\in\mathcal{P} w ∈ P , the Pohozaev identity, Lemma 2.2 and (1) imply
1 ≤ δ q β K q A ( w ) ( θ − 2 ) / 2 + κ S ( a , b ) − p / 2 A ( w ) ( p − 2 ) / 2 . (73) 1\leq\delta_q\beta K_q A(w)^{(\theta-2)/2}+\kappa S(a,b)^{-p/2}A(w)^{(p-2)/2}. \tag*{(73)} 1 ≤ δ q β K q A ( w ) ( θ − 2 ) /2 + κ S ( a , b ) − p /2 A ( w ) ( p − 2 ) /2 . (73)
Since the right-hand side tends to zero as A ( w ) → 0 + A(w)\to0^{+} A ( w ) → 0 + , there exists a constant A 0 > 0 A_0>0 A 0 > 0 , independent of w ∈ P w\in\mathcal{P} w ∈ P , such that A ( w ) ≥ A 0 A(w)\geq A_0 A ( w ) ≥ A 0 . On P \mathcal{P} P , using the Pohozaev identity
A ( w ) = δ q β B ( w ) + κ C ( w ) , A(w)=\delta_q\beta B(w)+\kappa C(w), A ( w ) = δ q β B ( w ) + κ C ( w ) ,
we obtain
J ( w ) = θ − 2 2 q β B ( w ) + d N κ C ( w ) ≥ c 0 A ( w ) ≥ c 0 A 0 > 0 , w ∈ P , J(w)=\frac{\theta-2}{2q}\beta B(w)+\frac{d}{N}\kappa C(w)\geq c_0A(w)\geq c_0A_0>0,\qquad w\in\mathcal{P}, J ( w ) = 2 q θ − 2 β B ( w ) + N d κ C ( w ) ≥ c 0 A ( w ) ≥ c 0 A 0 > 0 , w ∈ P ,
where c 0 = min { θ − 2 2 q δ q , d N } > 0 c_0=\min\left\{\frac{\theta-2}{2q\delta^q},\frac{d}{N}\right\}>0 c 0 = min { 2 q δ q θ − 2 , N d } > 0 . Consequently, m = inf P J > 0 m=\inf_{\mathcal{P}}J>0 m = inf P J > 0 .
If q = q c q=q_c q = q c , then θ = 2 \theta=2 θ = 2 . The condition (1.9) gives
c β A ( w ) ≤ κ C ( w ) ≤ κ S ( a , b ) − p / 2 A ( w ) p / 2 . c_\beta A(w)\le\kappa C(w)\le\kappa S(a,b)^{-p/2}A(w)^{p/2}. c β A ( w ) ≤ κ C ( w ) ≤ κ S ( a , b ) − p /2 A ( w ) p /2 .
Therefore,
A ( w ) ≥ ( c β S ( a , b ) p / 2 κ ) 2 p − 2 > 0 , w ∈ P . A(w)\ge\left(\frac{c_\beta S(a,b)^{p/2}}{\kappa}\right)^{\frac{2}{p-2}}>0,\qquad w\in\mathcal{P}. A ( w ) ≥ ( κ c β S ( a , b ) p /2 ) p − 2 2 > 0 , w ∈ P .
Moreover, on P \mathcal{P} P ,
J ( w ) = d N κ C ( w ) ≥ d N c β A ( w ) > 0. J(w)=\frac{d}{N}\kappa C(w)\ge\frac{d}{N}c_\beta A(w)>0. J ( w ) = N d κ C ( w ) ≥ N d c β A ( w ) > 0.
Consequently,
m = inf P J ≥ d N c β ( c β S ( a , b ) p / 2 κ ) 2 p − 2 > 0. m=\inf_{\mathcal{P}}J\ge\frac{d}{N}c_\beta\left(\frac{c_\beta S(a,b)^{p/2}}{\kappa}\right)^{\frac{2}{p-2}}>0. m = P inf J ≥ N d c β ( κ c β S ( a , b ) p /2 ) p − 2 2 > 0.
Lemma 6.2. We have
0 < m < T κ . 0<m<T_\kappa. 0 < m < T κ . Proof. The lower bound follows from Lemma 6.1. For the upper bound, let W W W be the extremal from Lemma 2.3 and set
Z = W M ( W ) 1 / 2 ∈ S . Z=\frac{W}{M(W)^{1/2}}\in S. Z = M ( W ) 1/2 W ∈ S . For the pure critical functional, equality in the sharp CKN inequality (1.1) gives
max t ∈ R ( 1 2 e 2 t A ( Z ) − κ p e p t C ( Z ) ) = d N ( A ( Z ) ( κ C ( Z ) ) 2 / p ) N 2 d = T κ . \max_{t\in\mathbb{R}}\left(\frac{1}{2}e^{2t}A(Z)-\frac{\kappa}{p}e^{pt}C(Z)\right)=\frac{d}{N}\left(\frac{A(Z)}{(\kappa C(Z))^{2/p}}\right)^{\frac{N}{2d}}=T_\kappa. t ∈ R max ( 2 1 e 2 t A ( Z ) − p κ e pt C ( Z ) ) = N d ( ( κ C ( Z ) ) 2/ p A ( Z ) ) 2 d N = T κ . Since
J ( t ⋆ Z ) < 1 2 e 2 t A ( Z ) − κ p e p t C ( Z ) , t ∈ R , J(t\star Z)<\frac{1}{2}e^{2t}A(Z)-\frac{\kappa}{p}e^{pt}C(Z),\qquad t\in\mathbb{R}, J ( t ⋆ Z ) < 2 1 e 2 t A ( Z ) − p κ e pt C ( Z ) , t ∈ R , we have
max t ∈ R J ( t ⋆ Z ) < T κ . \max_{t\in\mathbb{R}}J(t\star Z)<T_\kappa. t ∈ R max J ( t ⋆ Z ) < T κ . Let t Z t_Z t Z be the unique maximizer of the fiber map t ↦ J ( t ⋆ Z ) t\mapsto J(t\star Z) t ↦ J ( t ⋆ Z ) , whose existence follows from Lemma 3.2 . Since P ( t Z ⋆ Z ) = 0 P(t_Z\star Z)=0 P ( t Z ⋆ Z ) = 0 , we have t Z ⋆ Z ∈ P t_Z\star Z\in\mathcal{P} t Z ⋆ Z ∈ P . Therefore,
m ≤ J ( t Z ⋆ Z ) = max t ∈ R J ( t ⋆ Z ) < T κ . m\le J(t_Z\star Z)=\max_{t\in\mathbb{R}}J(t\star Z)<T_\kappa. m ≤ J ( t Z ⋆ Z ) = t ∈ R max J ( t ⋆ Z ) < T κ .
Proposition 6.1. The level m m m is attained by a nonnegative function w ∈ P w\in\mathcal{P} w ∈ P , which is a constrained critical point of J ∣ S J|_S J ∣ S . Moreover,
J ( w ) = m ∈ ( 0 , T κ ) . J(w)=m\in(0,T_\kappa). J ( w ) = m ∈ ( 0 , T κ ) . The level admits the minimax characterization
m = inf z ∈ S max t ∈ R J ( t ⋆ z ) . (74) m=\inf_{z\in S}\max_{t\in\mathbb{R}}J(t\star z). \tag*{(74)} m = z ∈ S inf t ∈ R max J ( t ⋆ z ) . (74)
Proof. We first establish (74) . For every w ∈ S w \in S w ∈ S , (2.2) and (3.1) give t ⋆ w ∈ S t \star w \in S t ⋆ w ∈ S and Φ w ′ ( t ) = P ( t ⋆ w ) \Phi_w'(t) = P(t \star w) Φ w ′ ( t ) = P ( t ⋆ w ) for all t ∈ R t \in\mathbb{R} t ∈ R . By Lemma 3.2, t ∗ ( w ) t_*(w) t ∗ ( w ) is the unique critical point of Φ w \Phi_w Φ w and its strict global maximizer. Thus t ∗ ( w ) ⋆ w ∈ P t_*(w) \star w \in\mathcal{P} t ∗ ( w ) ⋆ w ∈ P and
m ≤ J ( t ∗ ( w ) ⋆ w ) = max t ∈ R J ( t ⋆ w ) . m \le J(t_*(w) \star w) = \max_{t \in\mathbb{R}} J(t \star w). m ≤ J ( t ∗ ( w ) ⋆ w ) = t ∈ R max J ( t ⋆ w ) .
Taking the infimum over w ∈ S w \in S w ∈ S , we obtain
m ≤ inf w ∈ S max t ∈ R J ( t ⋆ w ) . m \le\inf_{w \in S} \max_{t \in\mathbb{R}} J(t \star w). m ≤ w ∈ S inf t ∈ R max J ( t ⋆ w ) .
Conversely, for z ∈ P z \in\mathcal{P} z ∈ P , Φ z ′ ( 0 ) = P ( z ) = 0 \Phi_z'(0) = P(z) = 0 Φ z ′ ( 0 ) = P ( z ) = 0 , so the uniqueness of the critical point gives t ∗ ( z ) = 0 t_*(z) = 0 t ∗ ( z ) = 0 . Since P ⊂ S \mathcal{P} \subset S P ⊂ S ,
inf w ∈ S max t ∈ R J ( t ⋆ w ) ≤ inf z ∈ P max t ∈ R J ( t ⋆ z ) = inf z ∈ P J ( z ) = m . \inf_{w \in S} \max_{t \in\mathbb{R}} J(t \star w) \le\inf_{z \in\mathcal{P}} \max_{t \in\mathbb{R}} J(t \star z) = \inf_{z \in\mathcal{P}} J(z) = m. w ∈ S inf t ∈ R max J ( t ⋆ w ) ≤ z ∈ P inf t ∈ R max J ( t ⋆ z ) = z ∈ P inf J ( z ) = m .
Combining the two inequalities proves (74) .
To prove attainment of m m m , let { z n } ⊂ P \{z_n\} \subset\mathcal{P} { z n } ⊂ P be a minimizing sequence, so that J ( z n ) → m J(z_n) \to m J ( z n ) → m . If q > q c q > q_c q > q c , then δ q / 2 − 1 / q > 0 \delta_q/2 - 1/q > 0 δ q /2 − 1/ q > 0 , so (3.3) and the boundedness of { J ( z n ) } \{J(z_n)\} { J ( z n )} imply that both { B ( z n ) } \{B(z_n)\} { B ( z n )} and { C ( z n ) } \{C(z_n)\} { C ( z n )} are bounded. Since P ( z n ) = 0 P(z_n) = 0 P ( z n ) = 0 , we have
A ( z n ) = δ q β B ( z n ) + κ C ( z n ) . A(z_n) = \delta_q \beta B(z_n) + \kappa C(z_n). A ( z n ) = δ q β B ( z n ) + κ C ( z n ) .
which implies that { A ( z n ) } \{A(z_n)\} { A ( z n )} is bounded.
If q = q c q = q_c q = q c , then q δ q = 2 q\delta_q = 2 q δ q = 2 and c β = 1 − δ q β K q > 0 c_\beta= 1 - \delta_q \beta K_q > 0 c β = 1 − δ q β K q > 0 by (1.9). Using (2.6) and (3.3), together with M ( z n ) = 1 M(z_n) = 1 M ( z n ) = 1 and P ( z n ) = 0 P(z_n) = 0 P ( z n ) = 0 , we obtain
c β A ( z n ) ≤ A ( z n ) − δ q β B ( z n ) = κ C ( z n ) = N d J ( z n ) . c_\beta A(z_n) \le A(z_n) - \delta_q \beta B(z_n) = \kappa C(z_n) = \frac{N}{d}J(z_n). c β A ( z n ) ≤ A ( z n ) − δ q β B ( z n ) = κ C ( z n ) = d N J ( z n ) .
Thus, in both cases, { z n } \{z_n\} { z n } is bounded in X X X , since ∥ z n ∥ X 2 = A ( z n ) + 1 \lVert z_n\rVert_X^2 = A(z_n) + 1 ∥ z n ∥ X 2 = A ( z n ) + 1 .
Proposition 3.1 yields a bounded Palais–Smale sequence { w n } ⊂ P \{w_n\} \subset\mathcal{P} { w n } ⊂ P satisfying (3.13) with c = m c = m c = m . Passing to a subsequence, we may assume that w n ⇀ w w_n \rightharpoonup w w n ⇀ w in X X X . If critical loss occurs, Proposition 4.1 and Lemma 6.2 give
J ( w ) ≤ m − T κ < 0. J(w) \le m - T_\kappa< 0. J ( w ) ≤ m − T κ < 0.
Thus w ≢ 0 w \not\equiv0 w ≡ 0 . By (4.11), we have M ( w ) = 1 M(w) = 1 M ( w ) = 1 , while Lemma 4.1 gives P ( w ) = 0 P(w) = 0 P ( w ) = 0 . Hence w ∈ P w \in\mathcal{P} w ∈ P , and the definition of m m m yields
0 < m ≤ J ( w ) < 0 , 0 < m \le J(w) < 0, 0 < m ≤ J ( w ) < 0 ,
which is impossible. Since m > 0 m > 0 m > 0 , Proposition 4.1 then implies that, up to a subsequence, w n → w w_n \to w w n → w in X X X . By continuity, w ∈ P w \in\mathcal{P} w ∈ P and J ( w ) = m J(w) = m J ( w ) = m . Replacing w w w by ∣ w ∣ |w| ∣ w ∣ gives a nonnegative minimizer. By Lemma 3.4, it is a constrained critical point of J J J on S S S .
Proof of Theorem 1.1 (ii). Proposition 6.1 yields a scalar Pohozaev ground state with energy in ( 0 , T κ ) (0, T_\kappa) ( 0 , T κ ) . Lemma 4.1 gives its positivity away from the origin and the multiplier formula (1.11). Proposition 2.2 lifts this scalar solution to a synchronized normalized solution.
The map w ↦ ( ρ 1 w , ρ 2 w ) w \mapsto(\rho_1 w, \rho_2 w) w ↦ ( ρ 1 w , ρ 2 w ) is a homeomorphism from S S S onto S s y n S_{\mathrm{syn}} S syn . By (2.14), it transfers the minimizer of J J J on P \mathcal{P} P to a minimizer of E E E on P s y n \mathcal{P}_{\mathrm{syn}} P syn , hence to a synchronized Pohozaev ground state. Finally, (2.14), together with J ( w ) = m ∈ ( 0 , T κ ) J(w) = m \in(0, T_\kappa) J ( w ) = m ∈ ( 0 , T κ ) , yields (1.10).
Synchronized solutions and the full vector level
Proof of Theorem 1.2. Throughout this proof, q = q c q=q_c q = q c , both prescribed masses are one, and c β = 1 − δ q β K q > 0 c_\beta=1-\delta_q\beta K_q>0 c β = 1 − δ q β K q > 0 . For w ∈ S w\in S w ∈ S , set L ( w ) = A ( w ) − δ q β B ( w ) L(w)=A(w)-\delta_q\beta B(w) L ( w ) = A ( w ) − δ q β B ( w ) . Since δ q = 2 / q \delta_q=2/q δ q = 2/ q , we have L ( w ) ≥ c β A ( w ) > 0 L(w)\ge c_\beta A(w)>0 L ( w ) ≥ c β A ( w ) > 0 and
J χ ( t ⋆ w ) = 1 2 e 2 t L ( w ) − χ p e p t C ( w ) . J_\chi(t\star w)=\frac{1}{2}e^{2t}L(w)-\frac{\chi}{p}e^{pt}C(w). J χ ( t ⋆ w ) = 2 1 e 2 t L ( w ) − p χ e pt C ( w ) .
The unique maximum of this fiber is given by
max t J χ ( t ⋆ w ) = d N L ( w ) p / ( p − 2 ) ( χ C ( w ) ) 2 / ( p − 2 ) . (75) \max_t J_\chi(t\star w)=\frac{d}{N}\frac{L(w)^{p/(p-2)}}{(\chi C(w))^{2/(p-2)}}. \tag*{(75)} t max J χ ( t ⋆ w ) = N d ( χ C ( w ) ) 2/ ( p − 2 ) L ( w ) p / ( p − 2 ) . (75)
Taking the infimum over S S S proves m χ = χ − 2 / ( p − 2 ) m 1 m_\chi=\chi^{-2/(p-2)}m_1 m χ = χ − 2/ ( p − 2 ) m 1 by (6.2). The positivity and attainment of these levels follow from Proposition 6.1. The energy factorization for synchronized pairs gives m s y n = 2 m κ m_{\mathrm{syn}}=2m_\kappa m syn = 2 m κ .
For ( u , v ) ∈ S v e c (u,v)\in S_{\mathrm{vec}} ( u , v ) ∈ S vec , define
L 2 = L ( u ) + L ( v ) , K 2 = μ ( C ( u ) + C ( v ) ) + p ν D ( u , v ) . L_2=L(u)+L(v),\qquad K_2=\mu(C(u)+C(v))+p\nu D(u,v). L 2 = L ( u ) + L ( v ) , K 2 = μ ( C ( u ) + C ( v )) + p ν D ( u , v ) .
Since L 2 , K 2 > 0 L_2,K_2>0 L 2 , K 2 > 0 , the fiber map has a unique global maximum, given by
max t E ( t ⋆ u , t ⋆ v ) = d N L 2 p / ( p − 2 ) K 2 2 / ( p − 2 ) . (76) \max_t E(t\star u,t\star v)=\frac{d}{N}\frac{L_2^{p/(p-2)}}{K_2^{2/(p-2)}}. \tag*{(76)} t max E ( t ⋆ u , t ⋆ v ) = N d K 2 2/ ( p − 2 ) L 2 p / ( p − 2 ) . (76)
Each fiber intersects P v e c \mathcal{P}_{\mathrm{vec}} P vec at its maximizing point. To establish a positive lower bound, set A 2 = A ( u ) + A ( v ) A_2=A(u)+A(v) A 2 = A ( u ) + A ( v ) . The Hölder inequality gives
D ( u , v ) ≤ C ( u ) 1 / 2 C ( v ) 1 / 2 ≤ 1 2 ( C ( u ) + C ( v ) ) , D(u,v)\le C(u)^{1/2}C(v)^{1/2}\le\frac{1}{2}(C(u)+C(v)), D ( u , v ) ≤ C ( u ) 1/2 C ( v ) 1/2 ≤ 2 1 ( C ( u ) + C ( v )) ,
and hence
K 2 ≤ κ ( C ( u ) + C ( v ) ) ≤ κ S ( a , b ) − p / 2 A 2 p / 2 . K_2\le\kappa(C(u)+C(v))\le\kappa S(a,b)^{-p/2}A_2^{p/2}. K 2 ≤ κ ( C ( u ) + C ( v )) ≤ κ S ( a , b ) − p /2 A 2 p /2 .
On P v e c \mathcal{P}_{\mathrm{vec}} P vec , K 2 = L 2 ≥ c β A 2 K_2=L_2\ge c_\beta A_2 K 2 = L 2 ≥ c β A 2 . It follows that
A 2 ≥ ( c β S ( a , b ) p / 2 κ ) 2 / ( p − 2 ) , E ( u , v ) = d N K 2 ≥ d N c β A 2 > 0. A_2\ge\left(\frac{c_\beta S(a,b)^{p/2}}{\kappa}\right)^{2/(p-2)},\qquad E(u,v)=\frac{d}{N}K_2\ge\frac{d}{N}c_\beta A_2>0. A 2 ≥ ( κ c β S ( a , b ) p /2 ) 2/ ( p − 2 ) , E ( u , v ) = N d K 2 ≥ N d c β A 2 > 0.
Since this lower bound is uniform, m v e c > 0 m_{\mathrm{vec}}>0 m vec > 0 .
Let U ∈ P μ U\in\mathcal{P}_\mu U ∈ P μ be a positive minimizer with J μ ( U ) = m μ J_\mu(U)=m_\mu J μ ( U ) = m μ , so that L ( U ) = μ C ( U ) L(U)=\mu C(U) L ( U ) = μ C ( U ) . Choose a nonnegative function Z ∈ C ∞ ( R N ) ∩ S Z\in C^\infty(\mathbb{R}^N)\cap S Z ∈ C ∞ ( R N ) ∩ S with supp Z ⊂ { x ∈ R N : 1 < ∣ x ∣ < 2 } \operatorname{supp}Z\subset\{x\in\mathbb{R}^N:1<|x|<2\} supp Z ⊂ { x ∈ R N : 1 < ∣ x ∣ < 2 } , and define Z s = s ⋆ Z Z_s=s\star Z Z s = s ⋆ Z for s < 0 s<0 s < 0 . Since q = q c q=q_c q = q c , the scaling identities yield
M ( Z s ) = 1 , A ( Z s ) = e 2 s A ( Z ) , B ( Z s ) = e 2 s B ( Z ) , C ( Z s ) = e p s C ( Z ) . M(Z_s)=1,\qquad A(Z_s)=e^{2s}A(Z),\qquad B(Z_s)=e^{2s}B(Z),\qquad C(Z_s)=e^{ps}C(Z). M ( Z s ) = 1 , A ( Z s ) = e 2 s A ( Z ) , B ( Z s ) = e 2 s B ( Z ) , C ( Z s ) = e p s C ( Z ) .
Moreover,
supp Z s ⊂ { x ∈ R N : e − s < ∣ x ∣ < 2 e − s } . \operatorname{supp}Z_s\subset\{x\in\mathbb{R}^N:e^{-s}<|x|<2e^{-s}\}. supp Z s ⊂ { x ∈ R N : e − s < ∣ x ∣ < 2 e − s } .
Hence, by the Cauchy–Schwarz inequality and M ( Z s ) = 1 M(Z_s)=1 M ( Z s ) = 1 ,
0 ≤ m s : = ∫ R N ∣ x ∣ − 2 a U Z s d x ≤ ( ∫ ∣ x ∣ > e − s ∣ x ∣ − 2 a U 2 d x ) 1 / 2 ⟶ 0 as s → − ∞ , 0\le m_s:=\int_{\mathbb{R}^N}|x|^{-2a}UZ_s\,dx\le\left(\int_{|x|>e^{-s}}|x|^{-2a}U^2\,dx\right)^{1/2}\longrightarrow0\qquad\text{as }s\to-\infty, 0 ≤ m s := ∫ R N ∣ x ∣ − 2 a U Z s d x ≤ ( ∫ ∣ x ∣ > e − s ∣ x ∣ − 2 a U 2 d x ) 1/2 ⟶ 0 as s → − ∞ ,
where the convergence follows from U ∈ L 2 ( R N ; ∣ x ∣ − 2 a ) U\in L^2(\mathbb{R}^N;|x|^{-2a}) U ∈ L 2 ( R N ; ∣ x ∣ − 2 a ) .
For 0 < ε ≤ 1 / 2 0 < \varepsilon\le1/2 0 < ε ≤ 1/2 , define
h ε , s = ε U + ( 1 − ε 2 ) 1 / 2 Z s , σ ε , s = M ( h ε , s ) 1 / 2 , v ε , s = h ε , s σ ε , s . h_{\varepsilon,s} = \varepsilon U + (1-\varepsilon^2)^{1/2}Z_s,\qquad\sigma_{\varepsilon,s} = M(h_{\varepsilon,s})^{1/2},\qquad v_{\varepsilon,s} = \frac{h_{\varepsilon,s}}{\sigma_{\varepsilon,s}}. h ε , s = ε U + ( 1 − ε 2 ) 1/2 Z s , σ ε , s = M ( h ε , s ) 1/2 , v ε , s = σ ε , s h ε , s .
Both components of ( U , v ε , s ) (U,v_{\varepsilon,s}) ( U , v ε , s ) have unit mass. Moreover,
σ ε , s 2 = 1 + 2 ε ( 1 − ε 2 ) 1 / 2 m s , 1 ≤ σ ε , s , ∣ σ ε , s − 1 ∣ ≤ C m s . \sigma_{\varepsilon,s}^2 = 1 + 2\varepsilon(1-\varepsilon^2)^{1/2}m_s,\qquad1\le\sigma_{\varepsilon,s},\qquad|\sigma_{\varepsilon,s}-1|\le Cm_s. σ ε , s 2 = 1 + 2 ε ( 1 − ε 2 ) 1/2 m s , 1 ≤ σ ε , s , ∣ σ ε , s − 1∣ ≤ C m s .
In the estimates below, C C C may depend on the fixed quantities U U U , Z Z Z , β \beta β , μ \mu μ , ν \nu ν , p p p , q q q , but is independent of 0 < ε ≤ 1 / 2 0 < \varepsilon\le1/2 0 < ε ≤ 1/2 and s < 0 s < 0 s < 0 . Combining the identity
v ε , s − ε U = ( σ ε , s − 1 − 1 ) ε U + σ ε , s − 1 ( 1 − ε 2 ) 1 / 2 Z s v_{\varepsilon,s}-\varepsilon U=(\sigma_{\varepsilon,s}^{-1}-1)\varepsilon U+\sigma_{\varepsilon,s}^{-1}(1-\varepsilon^2)^{1/2}Z_s v ε , s − ε U = ( σ ε , s − 1 − 1 ) ε U + σ ε , s − 1 ( 1 − ε 2 ) 1/2 Z s
with the scaling formulas yields
∥ ∣ x ∣ − a ∇ ( v ε , s − ε U ) ∥ 2 ≤ C ( m s + ε s ) , \left\|\left|x\right|^{-a}\nabla(v_{\varepsilon,s}-\varepsilon U)\right\|_2\le C(m_s+\varepsilon^s), ∣ x ∣ − a ∇ ( v ε , s − ε U ) 2 ≤ C ( m s + ε s ) ,
∥ ∣ x ∣ − b ( v ε , s − ε U ) ∥ q ≤ C ( m s + ε 2 s / q ) , \left\|\left|x\right|^{-b}(v_{\varepsilon,s}-\varepsilon U)\right\|_q\le C(m_s+\varepsilon^{2s/q}), ∣ x ∣ − b ( v ε , s − ε U ) q ≤ C ( m s + ε 2 s / q ) ,
∥ ∣ x ∣ − b ( v ε , s − ε U ) ∥ p ≤ C ( m s + ε s ) . \left\|\left|x\right|^{-b}(v_{\varepsilon,s}-\varepsilon U)\right\|_p\le C(m_s+\varepsilon^s). ∣ x ∣ − b ( v ε , s − ε U ) p ≤ C ( m s + ε s ) .
The preceding estimates yield convergence in the weighted gradient and nonlinear norms as s → − ∞ s\to-\infty s → − ∞ . Convergence in the weighted L 2 L^2 L 2 -norm is precluded by M ( v ε , s ) = 1 M(v_{\varepsilon,s})=1 M ( v ε , s ) = 1 and M ( ε U ) = ε 2 M(\varepsilon U)=\varepsilon^2 M ( ε U ) = ε 2 .
Set f = ∣ x ∣ − b U f=|x|^{-b}U f = ∣ x ∣ − b U and g = ∣ x ∣ − b v ε , s g=|x|^{-b}v_{\varepsilon,s} g = ∣ x ∣ − b v ε , s . Then f , g ≥ 0 f,g\ge0 f , g ≥ 0 , and ∥ g ∥ p \|g\|_p ∥ g ∥ p is bounded uniformly for 0 < ε ≤ 1 / 2 0 < \varepsilon\le1/2 0 < ε ≤ 1/2 and s < 0 s<0 s < 0 . Using the elementary inequality
∣ ξ r − η r ∣ ≤ r ( ξ + η ) r − 1 ∣ ξ − η ∣ , ξ , η ≥ 0 , r > 1 , |\xi^r-\eta^r|\le r(\xi+\eta)^{r-1}|\xi-\eta|,\qquad\xi,\eta\ge0,\quad r>1, ∣ ξ r − η r ∣ ≤ r ( ξ + η ) r − 1 ∣ ξ − η ∣ , ξ , η ≥ 0 , r > 1 ,
with r = p / 2 r=p/2 r = p /2 , followed by the Hölder inequality with exponents 2 2 2 , 2 p / ( p − 2 ) 2p/(p-2) 2 p / ( p − 2 ) , and p p p , we obtain
∣ D ( U , v ε , s ) − ε p / 2 C ( U ) ∣ = ∣ ∫ R N ( f p / 2 − ( ε f ) p / 2 ) ( g p / 2 − ( ε f ) p / 2 ) d x ∣ ≤ p 2 ∫ R N f p / 2 ( g + ε f ) p / 2 − 1 ∣ g − ε f ∣ d x ≤ p 2 ∥ f ∥ p p / 2 ( ∥ g ∥ p + ε ∥ f ∥ p ) p / 2 − 1 ∥ g − ε f ∥ p ≤ C ∥ ∣ x ∣ − b ( v ε , s − ε U ) ∥ p ≤ C ( m s + ε s ) . \begin{aligned}
\left|D(U,v_{\varepsilon,s})-\varepsilon^{p/2}C(U)\right|
&=\left|\int_{\mathbb{R}^N} \left(f^{p/2}-(\varepsilon f)^{p/2}\right)(g^{p/2}-(\varepsilon f)^{p/2})\,dx\right|\\
&\le\frac{p}{2}\int_{\mathbb{R}^N}f^{p/2}(g+\varepsilon f)^{p/2-1}|g-\varepsilon f|\,dx\\
&\le\frac{p}{2}\|f\|_{p}^{p/2}(\|g\|_p+\varepsilon\|f\|_p)^{p/2-1}\|g-\varepsilon f\|_p\\
&\le C\left\||x|^{-b}(v_{\varepsilon,s}-\varepsilon U)\right\|_p\\
&\le C(m_s+\varepsilon^s).
\end{aligned} D ( U , v ε , s ) − ε p /2 C ( U ) = ∫ R N ( f p /2 − ( ε f ) p /2 ) ( g p /2 − ( ε f ) p /2 ) d x ≤ 2 p ∫ R N f p /2 ( g + ε f ) p /2 − 1 ∣ g − ε f ∣ d x ≤ 2 p ∥ f ∥ p p /2 ( ∥ g ∥ p + ε ∥ f ∥ p ) p /2 − 1 ∥ g − ε f ∥ p ≤ C ∣ x ∣ − b ( v ε , s − ε U ) p ≤ C ( m s + ε s ) .
Set
L ε = L ( U ) + ε 2 A ( U ) − δ q β ε q B ( U ) , L_\varepsilon=L(U)+\varepsilon^2A(U)-\delta_q\beta\varepsilon^qB(U), L ε = L ( U ) + ε 2 A ( U ) − δ q β ε q B ( U ) ,
K ε = μ C ( U ) ( 1 + ε p ) + p ν ε p / 2 C ( U ) , K_\varepsilon=\mu C(U)(1+\varepsilon^p)+p\nu\varepsilon^{p/2}C(U), K ε = μ C ( U ) ( 1 + ε p ) + p ν ε p /2 C ( U ) ,
E s = ( ∫ ∣ x ∣ > e − s ∣ x ∣ − 2 a U 2 ) 1 / 2 + ε s + ε 2 s / q ⟶ 0. E_s=\left(\int_{|x|>e^{-s}}|x|^{-2a}U^2\right)^{1/2}+\varepsilon^s+\varepsilon^{2s/q}\longrightarrow0. E s = ( ∫ ∣ x ∣ > e − s ∣ x ∣ − 2 a U 2 ) 1/2 + ε s + ε 2 s / q ⟶ 0.
The preceding norm estimates, together with the same power inequality, give
∣ L 2 ( U , v ε , s ) − L ε ∣ + ∣ K 2 ( U , v ε , s ) − K ε ∣ ≤ C E s . (77) |L_2(U,v_{\varepsilon,s})-L_\varepsilon|+|K_2(U,v_{\varepsilon,s})-K_\varepsilon|\le CE_s. \tag*{(77)} ∣ L 2 ( U , v ε , s ) − L ε ∣ + ∣ K 2 ( U , v ε , s ) − K ε ∣ ≤ C E s . (77)
All four coefficients are uniformly bounded above and bounded away from zero, with L 2 , L ε ≥ L ( U ) L_2,L_\varepsilon\ge L(U) L 2 , L ε ≥ L ( U ) and K 2 , K ε ≥ μ C ( U ) K_2,K_\varepsilon\ge\mu C(U) K 2 , K ε ≥ μ C ( U ) . For L ε L_\varepsilon L ε , the lower bound follows from q > 2 q>2 q > 2 , ε ≤ 1 \varepsilon\le1 ε ≤ 1 , and L ( U ) > 0 L(U)>0 L ( U ) > 0 . Consequently, the logarithmic and power functions appearing in the fiber formulas have bounded derivatives on the relevant coefficient ranges. Define
t ε , s = 1 p − 2 log L 2 ( U , v ε , s ) K 2 ( U , v ε , s ) , t ε = 1 p − 2 log L ε K ε , F ε = d N L ε p / ( p − 2 ) K ε 2 / ( p − 2 ) . t_{\varepsilon,s}=\frac{1}{p-2}\log\frac{L_2(U,v_{\varepsilon,s})}{K_2(U,v_{\varepsilon,s})},\qquad
t_\varepsilon=\frac{1}{p-2}\log\frac{L_\varepsilon}{K_\varepsilon},\qquad
\mathcal{F}_\varepsilon=\frac{d}{N}\frac{L_\varepsilon^{p/(p-2)}}{K_\varepsilon^{2/(p-2)}}. t ε , s = p − 2 1 log K 2 ( U , v ε , s ) L 2 ( U , v ε , s ) , t ε = p − 2 1 log K ε L ε , F ε = N d K ε 2/ ( p − 2 ) L ε p / ( p − 2 ) .
Thus, the mean value theorem, together with (76) and (77) , yields
∣ t ε , s − t ε ∣ + ∣ max t ∈ R E ( t ⋆ U , t ⋆ v ε , s ) − F ε ∣ ≤ C E s . (78) |t_{\varepsilon,s}-t_\varepsilon|+\left|\max_{t\in\mathbb{R}}E(t\star U,t\star v_{\varepsilon,s})-\mathcal{F}_\varepsilon\right|\leq C\mathcal{E}_s. \tag*{(78)} ∣ t ε , s − t ε ∣ + t ∈ R max E ( t ⋆ U , t ⋆ v ε , s ) − F ε ≤ C E s . (78)
Since 0 < d < 1 0<d<1 0 < d < 1 , N ≥ 5 N\geq5 N ≥ 5 , we have 2 < p = 2 N N − 2 d < 2 N N − 2 ≤ 10 3 < 4 2<p=\frac{2N}{N-2d}<\frac{2N}{N-2}\leq\frac{10}{3}<4 2 < p = N − 2 d 2 N < N − 2 2 N ≤ 3 10 < 4 . The identity L ( U ) = μ C ( U ) L(U)=\mu C(U) L ( U ) = μ C ( U ) and (75) give
F ε m μ = ( 1 + A ( U ) L ( U ) ε 2 − δ q β B ( U ) L ( U ) ε q ) p p − 2 ( 1 + p ν μ ε p / 2 + ε p ) − 2 p − 2 . \frac{\mathcal{F}_\varepsilon}{m_\mu}=\left(1+\frac{A(U)}{L(U)}\varepsilon^2-\frac{\delta_q\beta B(U)}{L(U)}\varepsilon^q\right)^{\frac{p}{p-2}}\left(1+\frac{p\nu}{\mu}\varepsilon^{p/2}+\varepsilon^p\right)^{-\frac{2}{p-2}}. m μ F ε = ( 1 + L ( U ) A ( U ) ε 2 − L ( U ) δ q β B ( U ) ε q ) p − 2 p ( 1 + μ p ν ε p /2 + ε p ) − p − 2 2 .
Since q > 2 q>2 q > 2 , the first factor equals 1 + O ( ε 2 ) 1+O(\varepsilon^2) 1 + O ( ε 2 ) as ε → 0 + \varepsilon\to0^+ ε → 0 + . Taylor’s formula applied to the second factor gives
( 1 + p ν μ ε p / 2 + ε p ) − 2 p − 2 = 1 − 2 p ν ( p − 2 ) μ ε p / 2 + O ( ε p ) . \left(1+\frac{p\nu}{\mu}\varepsilon^{p/2}+\varepsilon^p\right)^{-\frac{2}{p-2}}=1-\frac{2p\nu}{(p-2)\mu}\varepsilon^{p/2}+O(\varepsilon^p). ( 1 + μ p ν ε p /2 + ε p ) − p − 2 2 = 1 − ( p − 2 ) μ 2 p ν ε p /2 + O ( ε p ) .
Combining these expansions and using p > 2 p>2 p > 2 , we find constants C > 0 C>0 C > 0 and ε 0 ∈ ( 0 , 1 / 2 ] \varepsilon_0\in(0,1/2] ε 0 ∈ ( 0 , 1/2 ] such that
∣ F ε − m μ + c 3 ε p / 2 ∣ ≤ C ε 2 , c 3 = 2 p ν ( p − 2 ) μ m μ > 0 , (79) \left|\mathcal{F}_\varepsilon-m_\mu+c_3\varepsilon^{p/2}\right|\leq C\varepsilon^2,\qquad c_3=\frac{2p\nu}{(p-2)\mu}m_\mu>0, \tag*{(79)} F ε − m μ + c 3 ε p /2 ≤ C ε 2 , c 3 = ( p − 2 ) μ 2 p ν m μ > 0 , (79)
for every 0 < ε ≤ ε 0 0<\varepsilon\leq\varepsilon_0 0 < ε ≤ ε 0 .
Since p < 4 p<4 p < 4 , we may fix ε ∈ ( 0 , ε 0 ] \varepsilon\in(0,\varepsilon_0] ε ∈ ( 0 , ε 0 ] sufficiently small that C ε 2 − p / 2 ≤ c 3 4 C\varepsilon^{2-p/2}\leq\frac{c_3}{4} C ε 2 − p /2 ≤ 4 c 3 . With this ε \varepsilon ε fixed, the convergence E s → 0 \mathcal{E}_s\to0 E s → 0 as s → − ∞ s\to-\infty s → − ∞ allows us to choose s < 0 s<0 s < 0 sufficiently negative that the error term in (78) satisfies C E s ≤ c 3 4 ε p / 2 C\mathcal{E}_s\leq\frac{c_3}{4}\varepsilon^{p/2} C E s ≤ 4 c 3 ε p /2 . By construction, ( t ε , s ⋆ U , t ε , s ⋆ v ε , s ) ∈ P v e c (t_{\varepsilon,s}\star U,t_{\varepsilon,s}\star v_{\varepsilon,s})\in\mathcal{P}_{\mathrm{vec}} ( t ε , s ⋆ U , t ε , s ⋆ v ε , s ) ∈ P vec . Combining (78) with (79) , we obtain
m v e c ≤ max t ∈ R E ( t ⋆ U , t ⋆ v ε , s ) ≤ F ε + C E s ≤ m μ − c 3 2 ε p / 2 < m μ . m_{\mathrm{vec}}\leq\max_{t\in\mathbb{R}}E(t\star U,t\star v_{\varepsilon,s})\leq\mathcal{F}_\varepsilon+C\mathcal{E}_s\leq m_\mu-\frac{c_3}{2}\varepsilon^{p/2}<m_\mu. m vec ≤ t ∈ R max E ( t ⋆ U , t ⋆ v ε , s ) ≤ F ε + C E s ≤ m μ − 2 c 3 ε p /2 < m μ .
This establishes the strict comparison m v e c < m μ m_{\mathrm{vec}}<m_\mu m vec < m μ in (12) for every ν > 0 \nu>0 ν > 0 . Finally, (13) is equivalent to κ ≤ 2 ( p − 2 ) / 2 μ \kappa\leq2^{(p-2)/2}\mu κ ≤ 2 ( p − 2 ) /2 μ . The scaling relation m χ = χ − 2 / ( p − 2 ) m 1 m_\chi=\chi^{-2/(p-2)}m_1 m χ = χ − 2/ ( p − 2 ) m 1 gives
m v e c < m μ = μ − 2 / ( p − 2 ) m 1 ≤ 2 κ − 2 / ( p − 2 ) m 1 = m s y n . m_{\mathrm{vec}}<m_\mu=\mu^{-2/(p-2)}m_1\leq2\kappa^{-2/(p-2)}m_1=m_{\mathrm{syn}}. m vec < m μ = μ − 2/ ( p − 2 ) m 1 ≤ 2 κ − 2/ ( p − 2 ) m 1 = m syn .
Acknowledgements
This work has been supported by the National Natural Science Foundation of China under Grant No. 12371121 and the Fundamental Research Funds for the Central Universities under Grant No. SWU-KF26002.
Data availability statement
No datasets were generated or analysed during the present study.
Conflict of interest
The authors declare that they have no conflicts of interest.
References