Introduction
The Born–Infeld theory [11 , 12 , 13 , 14 ] was proposed as a nonlinear alternative to Maxwell’s electrodynamics in which the field energy of a point charge is finite. In the electrostatic case the potential solves an equation involving the operator − div ( ∇ u / 1 − ∣ ∇ u ∣ 2 ) -\operatorname{div}(\nabla u/\sqrt{1-|\nabla u|^2}) − div ( ∇ u / 1 − ∣∇ u ∣ 2 ) , and the constraint ∣ ∇ u ∣ < 1 |\nabla u|<1 ∣∇ u ∣ < 1 expresses the fact that the electric field is bounded, the absolute field constant of Born being normalized to one. The same operator gives the mean curvature of spacelike graphs in the Lorentz–Minkowski space [3 ] . Even in the electrostatic case with a prescribed charge density, the regularity of the minimizers of the Born–Infeld energy, and in particular the question whether the gradient bound is strict, is a delicate issue, see Bonheure and Iacopetti [9 , 10 ] , and Byeon et al. [20 ] . The variational density of the operator
j B I ( r ) = 1 − 1 − r 2 = r 2 2 + r 4 8 + O ( r 6 ) ( r → 0 ) j_{\mathrm{BI}}(r)=1-\sqrt{1-r^2}=\frac{r^2}{2}+\frac{r^4}{8}+O(r^6)\qquad(r\to0) j BI ( r ) = 1 − 1 − r 2 = 2 r 2 + 8 r 4 + O ( r 6 ) ( r → 0 )
behaves like the Dirichlet density for small gradients, but it is not homogeneous and its derivative blows up as r → 1 − r\to1^- r → 1 − . These two features, the quadratic behavior for small gradients and the bound on the admissible gradients, are responsible for the phenomena studied in this paper, which is devoted to the normalized problem
{ − div ( ∇ u 1 − ∣ ∇ u ∣ 2 ) + μ u = f ( u ) in R N , N ≥ 3 , ∫ R N ∣ u ∣ 2 d x = 2 m 2 , (1) \begin{cases}
-\operatorname{div}\left(\frac{\nabla u}{\sqrt{1-\lvert\nabla u\rvert^2}}\right)+\mu u=f(u) & \text{in }\mathbb{R}^{N},\quad N\ge3,\\
\int_{\mathbb{R}^{N}}\lvert u\rvert^2\,dx=2m^2, \tag*{(1)}
\end{cases} ⎩ ⎨ ⎧ − div ( 1 − ∣ ∇ u ∣ 2 ∇ u ) + μu = f ( u ) ∫ R N ∣ u ∣ 2 d x = 2 m 2 , in R N , N ≥ 3 , (1)
where m > 0 m>0 m > 0 is prescribed and μ > 0 \mu>0 μ > 0 is an unknown Lagrange multiplier. Two questions are addressed: which masses are attained by the action ground states of the prescribed frequency problem, and which of these ground states minimize the energy on the mass sphere. The two questions have different answers: in the L 2 L^2 L 2 -supercritical case we find two different mass thresholds, one for each notion of ground state.
The time-dependent electromagnetic Born–Infeld equations have the structure of a system of conservation laws. Brenier [18 ] showed that, once the energy density and the Poynting vector are added to the unknowns, they form a hyperbolic system of ten conservation laws with a convex entropy which resembles the equations of magnetohydrodynamics. In the low field regime this augmented system reduces to the Maxwell equations and in the high field regime to pressureless magnetohydrodynamics. For variational results in the presence of a magnetic field we refer to [27 ] . The scalar problem (1) admits a different, formal Hamiltonian interpretation, which we now describe.
In nonlinear Schrödinger models the prescribed L 2 L^2 L 2 -norm represents the number of particles or the power of a beam, see [22 , 37 ] . In our scalar setting, a formal mass-preserving evolution is obtained by extending the energy to complex-valued functions ψ = q + i p \psi=q+ip ψ = q + i p . Setting
H ( ψ ) = ∫ R N ( 1 − 1 − ∣ ∇ ψ ∣ 2 − F ( ∣ ψ ∣ ) ) d x , ∣ ∇ ψ ∣ 2 = ∣ ∇ q ∣ 2 + ∣ ∇ p ∣ 2 , \mathcal{H}(\psi)=\int_{\mathbb{R}^{N}}\left(1-\sqrt{1-\lvert\nabla\psi\rvert^2}-F(\lvert\psi\rvert)\right)\,dx,\qquad\lvert\nabla\psi\rvert^2=\lvert\nabla q\rvert^2+\lvert\nabla p\rvert^2, H ( ψ ) = ∫ R N ( 1 − 1 − ∣ ∇ ψ ∣ 2 − F (∣ ψ ∣) ) d x , ∣ ∇ ψ ∣ 2 = ∣ ∇ q ∣ 2 + ∣ ∇ p ∣ 2 ,
where F ( s ) = ∫ 0 s f ( t ) d t F(s)=\int_0^s f(t)\,dt F ( s ) = ∫ 0 s f ( t ) d t , the canonical Hamilton equations q t = δ H / δ p q_t=\delta\mathcal{H}/\delta p q t = δ H / δ p , p t = − δ H / δ q p_t=-\delta\mathcal{H}/\delta q p t = − δ H / δ q take the form
i ∂ t ψ = − div ( ∇ ψ 1 − ∣ ∇ ψ ∣ 2 ) − f ( ∣ ψ ∣ ) ∣ ψ ∣ ψ , i\partial_t\psi=-\operatorname{div}\left(\frac{\nabla\psi}{\sqrt{1-\lvert\nabla\psi\rvert^2}}\right)-\frac{f(\lvert\psi\rvert)}{\lvert\psi\rvert}\psi, i ∂ t ψ = − div ( 1 − ∣ ∇ ψ ∣ 2 ∇ ψ ) − ∣ ψ ∣ f (∣ ψ ∣) ψ ,
where the nonlinear term is extended by continuity at ψ = 0 \psi=0 ψ = 0 . If f f f is odd, then H \mathcal{H} H coincides on real-valued functions with the energy E \mathcal{E} E introduced in Subsection 1.1. The Hamiltonian is invariant under the phase shifts ψ ↦ e i ϑ ψ \psi\mapsto e^{i\vartheta}\psi ψ ↦ e i ϑ ψ , and for smooth, sufficiently decaying solutions with ∥ ∇ ψ ∥ ∞ < 1 \lVert\nabla\psi\rVert_\infty<1 ∥ ∇ ψ ∥ ∞ < 1 a direct computation gives the local conservation law
∂ t ∣ ψ ∣ 2 2 + div ( Im ( ψ ‾ ∇ ψ ) 1 − ∣ ∇ ψ ∣ 2 ) = 0 , \partial_t\frac{\lvert\psi\rvert^2}{2}+\operatorname{div}\left(\frac{\operatorname{Im}(\overline{\psi}\nabla\psi)}{\sqrt{1-\lvert\nabla\psi\rvert^2}}\right)=0, ∂ t 2 ∣ ψ ∣ 2 + div ( 1 − ∣ ∇ ψ ∣ 2 Im ( ψ ∇ ψ ) ) = 0 ,
hence 1 2 ∥ ψ ∥ 2 2 \frac{1}{2}\lVert\psi\rVert_2^2 2 1 ∥ ψ ∥ 2 2 is conserved. If f f f is odd, the ansatz ψ ( t , x ) = e i μ t u ( x ) \psi(t,x)=e^{i\mu t}u(x) ψ ( t , x ) = e i μ t u ( x ) leads to the equation in (1) , and the constraint in (1) prescribes the conserved quantity. The same conclusion holds for the second order evolution
∂ t t ϕ − div ( ∇ ϕ 1 − ∣ ∇ ϕ ∣ 2 ) + κ 2 ϕ = f ( ∣ ϕ ∣ ) ∣ ϕ ∣ ϕ , κ > 0. (2) \partial_{tt}\phi-\operatorname{div}\left(\frac{\nabla\phi}{\sqrt{1-\lvert\nabla\phi\rvert^2}}\right)+\kappa^2\phi=\frac{f(\lvert\phi\rvert)}{\lvert\phi\rvert}\phi,\qquad\kappa>0. \tag*{(2)} ∂ tt ϕ − div ( 1 − ∣ ∇ ϕ ∣ 2 ∇ ϕ ) + κ 2 ϕ = ∣ ϕ ∣ f (∣ ϕ ∣) ϕ , κ > 0. (2)
which replaces the Laplacian in the nonlinear Klein–Gordon equation by the Born–Infeld operator and is the Hamiltonian system associated with the energy
∫ R N ( 1 2 ∣ ∂ t ϕ ∣ 2 + 1 − 1 − ∣ ∇ ϕ ∣ 2 + κ 2 2 ∣ ϕ ∣ 2 − F ( ∣ ϕ ∣ ) ) d x \int_{\mathbb{R}^{N}} \left(\frac{1}{2}|\partial_t\phi|^2+1-\sqrt{1-|\nabla\phi|^2}+\frac{\kappa^2}{2}|\phi|^2-F(|\phi|)\right)\,dx ∫ R N ( 2 1 ∣ ∂ t ϕ ∣ 2 + 1 − 1 − ∣∇ ϕ ∣ 2 + 2 κ 2 ∣ ϕ ∣ 2 − F ( ∣ ϕ ∣ ) ) d x
of a complex-valued field ϕ \phi ϕ . The invariance under the phase shifts gives the conserved charge Im ∫ R N ϕ ‾ ∂ t ϕ d x \operatorname{Im}\int_{\mathbb{R}^{N}}\overline{\phi}\partial_t\phi\,dx Im ∫ R N ϕ ∂ t ϕ d x , and, if f f f is odd, the standing waves ϕ ( t , x ) = e i ω t u ( x ) \phi(t,x)=e^{i\omega t}u(x) ϕ ( t , x ) = e iω t u ( x ) with 0 < ω < κ 0<\omega<\kappa 0 < ω < κ , the analogues of the Q Q Q -balls of [26 , 35 ] , solve ( B I μ ) (BI_{\mu}) ( B I μ ) with μ = κ 2 − ω 2 \mu=\kappa^2-\omega^2 μ = κ 2 − ω 2 and have charge ω ∥ u ∥ 2 2 = 2 ω m 2 \omega\|u\|_2^2=2\omega m^2 ω ∥ u ∥ 2 2 = 2 ω m 2 . Thus prescribing the charge at a given frequency amounts to prescribing the mass in (1.1). We stress that these are only formal motivations for the mass constraint, and not derivations from the electromagnetic system studied in [18 ] .
A further motivation comes from the geometry of spacelike hypersurfaces in the Lorentz–Minkowski space L N + 1 \mathbb{L}^{N+1} L N + 1 , that is, R N + 1 \mathbb{R}^{N+1} R N + 1 endowed with the metric d x 1 2 + ⋯ + d x N 2 − d z 2 dx_1^2+\cdots+dx_N^2-dz^2 d x 1 2 + ⋯ + d x N 2 − d z 2 . The connection between the prescribed mean curvature equation in L N + 1 \mathbb{L}^{N+1} L N + 1 and the Born–Infeld variational problem is developed in [3 , 9 , 10 , 15 ] . For u ∈ C 2 ( R N ) u\in C^2(\mathbb{R}^N) u ∈ C 2 ( R N ) with ∣ ∇ u ∣ < 1 |\nabla u|<1 ∣∇ u ∣ < 1 , the graph Γ u = { ( x , u ( x ) ) : x ∈ R N } \Gamma_u=\{(x,u(x)):x\in\mathbb{R}^N\} Γ u = {( x , u ( x )) : x ∈ R N } is spacelike and its induced area element is 1 − ∣ ∇ u ∣ 2 d x \sqrt{1-|\nabla u|^2}\,dx 1 − ∣∇ u ∣ 2 d x , and the Born–Infeld density 1 − 1 − ∣ ∇ u ∣ 2 1-\sqrt{1-|\nabla u|^2} 1 − 1 − ∣∇ u ∣ 2 therefore measures the local area deficit of Γ u \Gamma_u Γ u with respect to a horizontal hyperplane. If the mean curvature of Γ u \Gamma_u Γ u is normalized as
H u = 1 N div ( ∇ u 1 − ∣ ∇ u ∣ 2 ) , H_u=\frac{1}{N}\operatorname{div}\left(\frac{\nabla u}{\sqrt{1-|\nabla u|^2}}\right), H u = N 1 div ( 1 − ∣∇ u ∣ 2 ∇ u ) ,
then (1.1) is a prescribed mean curvature problem in which the curvature is a function of the height, namely H u = ( μ u − f ( u ) ) / N H_u=(\mu u-f(u))/N H u = ( μu − f ( u )) / N , and the parameter μ \mu μ is selected together with the graph by the constraint ∫ R N ∣ u ∣ 2 d x = 2 m 2 \int_{\mathbb{R}^N}|u|^2\,dx=2m^2 ∫ R N ∣ u ∣ 2 d x = 2 m 2 , which prescribes the L 2 L^2 L 2 -size of the height function and not the area of the graph. The regularity of the solutions and the strict gradient bound recalled below justify this interpretation for the solutions considered here.
Normalized solutions of nonlinear Schrödinger equations, that is, of (1.1) with the Born–Infeld operator replaced by − Δ -\Delta − Δ , have been intensively studied since [30 ] . We refer to [2 , 4 , 5 , 7 , 33 ] for the L 2 L^2 L 2 -supercritical case, to [31 , 32 , 39 , 40 , 42 ] for combined power nonlinearities, as in Example 1(i), to [25 , 34 ] for the cubic–quintic case, and to [23 , 24 , 25 , 28 , 29 ] for the Lagrangian and the action–energy points of view adopted here. For the Born–Infeld equation with prescribed L 2 L^2 L 2 -norm, Baldelli, the first author and Pomponio [1 ] obtained solutions in the L 2 L^2 L 2 -subcritical case with a pure power nonlinearity, by truncating the operator and removing the truncation afterwards by means of uniform estimates. The L 2 L^2 L 2 -supercritical case is not covered by this argument and was left open in [1 ] . In this paper we solve the L 2 L^2 L 2 -supercritical and the L 2 L^2 L 2 -critical problems for all masses above a threshold, which is sharp within the class of action ground states, and we establish a duality between action and energy ground states which covers the L 2 L^2 L 2 -critical case as well. We rely on the existence and regularity theory for the prescribed frequency problem developed in [6 , 21 ] , which provides, in particular, the strict gradient bound ∥ ∇ u ∥ ∞ < 1 \|\nabla u\|_{\infty}<1 ∥∇ u ∥ ∞ < 1 for its critical points. For the corresponding regularity theory of the electrostatic Born–Infeld equation with a prescribed charge density we refer to [8 , 9 , 10 , 20 ] .
For the nonlinearity f f f we make the following assumptions:
(F1) f ∈ C 1 ( R ) f\in C^1(\mathbb{R}) f ∈ C 1 ( R ) and f ( 0 ) = f ′ ( 0 ) = 0 f(0)=f'(0)=0 f ( 0 ) = f ′ ( 0 ) = 0 ;
(F2) s ↦ f ( s ) ∣ s ∣ s\mapsto\frac{f(s)}{|s|} s ↦ ∣ s ∣ f ( s ) is strictly increasing on R ∖ { 0 } \mathbb{R}\setminus\{0\} R ∖ { 0 } ;
(F3) lim ∣ s ∣ → ∞ f ( s ) s = + ∞ \lim_{|s|\to\infty}\frac{f(s)}{s}=+\infty lim ∣ s ∣ → ∞ s f ( s ) = + ∞ ;
(F4) for some 2 < p < 2 ∗ = 2 N N − 2 , \text{(F4) for some } 2<p<2^{*}=\frac{2N}{N-2}, (F4) for some 2 < p < 2 ∗ = N − 2 2 N ,
0 < lim inf s → 0 f ( s ) ∣ s ∣ p − 2 s ≤ lim sup s → 0 f ( s ) ∣ s ∣ p − 2 s < ∞ . 0<\liminf_{s\to0}\frac{f(s)}{|s|^{p-2}s}\leq\limsup_{s\to0}\frac{f(s)}{|s|^{p-2}s}<\infty. 0 < s → 0 lim inf ∣ s ∣ p − 2 s f ( s ) ≤ s → 0 lim sup ∣ s ∣ p − 2 s f ( s ) < ∞.
Condition (F3) is the superlinearity condition (G3) of [6 ] and does not require a power-like behavior of f f f at infinity. Condition (F4) only requires two-sided power bounds near the origin: it is equivalent to the existence of c 0 , C 0 , s 0 > 0 c_0,C_0,s_0>0 c 0 , C 0 , s 0 > 0 such that c 0 ∣ s ∣ p − 2 ≤ f ( s ) / s ≤ C 0 ∣ s ∣ p − 2 c_0|s|^{p-2}\leq f(s)/s\leq C_0|s|^{p-2} c 0 ∣ s ∣ p − 2 ≤ f ( s ) / s ≤ C 0 ∣ s ∣ p − 2 for 0 < ∣ s ∣ ≤ s 0 0<|s|\leq s_0 0 < ∣ s ∣ ≤ s 0 , and the quotient f ( s ) / ( ∣ s ∣ p − 2 s ) f(s)/(|s|^{p-2}s) f ( s ) / ( ∣ s ∣ p − 2 s ) need not have a limit as s → 0 s\to0 s → 0 . The exponent p p p decides whether the problem is L 2 L^2 L 2 -subcritical, L 2 L^2 L 2 -critical or L 2 L^2 L 2 -supercritical, according as p < 2 + 4 N p<2+\frac{4}{N} p < 2 + N 4 , p = 2 + 4 N p=2+\frac{4}{N} p = 2 + N 4 or p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 . No upper bound on the growth of f f f at infinity is required: the gradient constraint and the prescribed mass yield a uniform L ∞ L^\infty L ∞ bound on the admissible functions, see Lemma A.1. Subject to the remaining assumptions, nonlinearities growing faster than any power are therefore allowed, see Example 1(ii).
Variational framework
We briefly recall the results of [6 ] on the prescribed frequency problem:
{ − div ( ∇ u 1 − ∣ ∇ u ∣ 2 ) + μ u = f ( u ) , in R N , N ≥ 3 , u ( x ) → 0 , as ∣ x ∣ → + ∞ . \left\{
\begin{aligned}
-\operatorname{div}\left(\frac{\nabla u}{\sqrt{1-|\nabla u|^2}}\right)+\mu u&=f(u), &&\text{in }\mathbb{R}^{N},\ N\geq3,\\
u(x)&\to0, &&\text{as }|x|\to+\infty.
\end{aligned}
\right. ⎩ ⎨ ⎧ − div ( 1 − ∣∇ u ∣ 2 ∇ u ) + μu u ( x ) = f ( u ) , → 0 , in R N , N ≥ 3 , as ∣ x ∣ → + ∞.
The natural setting for the Born–Infeld operator, introduced in [6 , 21 ] , is the reflexive Banach space
X : = { u ∈ H 1 ( R N ) : ∇ u ∈ L 2 j 0 ( R N ) } , j 0 ∈ N , 2 j 0 > N , X:=\{u\in H^1(\mathbb{R}^{N}):\nabla u\in L^{2j_0}(\mathbb{R}^{N})\},\qquad j_0\in\mathbb{N},\ 2j_0>N, X := { u ∈ H 1 ( R N ) : ∇ u ∈ L 2 j 0 ( R N )} , j 0 ∈ N , 2 j 0 > N ,
equipped with the norm
∥ u ∥ X : = ( ∥ u ∥ H 1 2 + ∥ ∇ u ∥ 2 j 0 2 ) 1 / 2 , \|u\|_X:=\left(\|u\|_{H^1}^2+\|\nabla u\|_{2j_0}^2\right)^{1/2}, ∥ u ∥ X := ( ∥ u ∥ H 1 2 + ∥∇ u ∥ 2 j 0 2 ) 1/2 ,
where ∥ ⋅ ∥ r \|\cdot\|_r ∥ ⋅ ∥ r denotes the standard L r L^r L r -norm. Thanks to 2 j 0 > N 2j_0>N 2 j 0 > N , the following embedding holds (cf. [21 Lemma 2.15] ):
X ⊂ H 1 ( R N ) ∩ C 0 ( R N ) , C 0 ( R N ) = { u ∈ C ( R N ) : lim ∣ x ∣ → ∞ u ( x ) = 0 } . (3) X\subset H^1(\mathbb{R}^{N})\cap C_0(\mathbb{R}^{N}),\qquad C_0(\mathbb{R}^{N})=\{u\in C(\mathbb{R}^{N}):\lim_{|x|\to\infty}u(x)=0\}. \tag*{(3)} X ⊂ H 1 ( R N ) ∩ C 0 ( R N ) , C 0 ( R N ) = { u ∈ C ( R N ) : ∣ x ∣ → ∞ lim u ( x ) = 0 } . (3)
The action of ( B I μ ) (BI_\mu) ( B I μ ) , denoted by I μ I_\mu I μ , splits as the sum of the convex, lower semicontinuous term
Ψ ( u ) : = { ∫ R N ( 1 − 1 − ∣ ∇ u ∣ 2 ) d x , if ∥ ∇ u ∥ ∞ ≤ 1 , + ∞ , otherwise , \Psi(u):=
\begin{cases}
\int_{\mathbb{R}^{N}}\left(1-\sqrt{1-|\nabla u|^2}\right)\,dx,&\text{if }\|\nabla u\|_\infty\leq1,\\
+\infty,&\text{otherwise},
\end{cases} Ψ ( u ) := { ∫ R N ( 1 − 1 − ∣∇ u ∣ 2 ) d x , + ∞ , if ∥∇ u ∥ ∞ ≤ 1 , otherwise ,
and of a C 1 C^1 C 1 -term
Φ μ ( u ) : = μ 2 ∥ u ∥ 2 2 − ∫ R N F ( u ) d x , for every μ ∈ ( 0 , + ∞ ) . \Phi_\mu(u):=\frac{\mu}{2}\|u\|_2^2-\int_{\mathbb{R}^{N}}F(u)\,dx,\qquad\text{for every }\mu\in(0,+\infty). Φ μ ( u ) := 2 μ ∥ u ∥ 2 2 − ∫ R N F ( u ) d x , for every μ ∈ ( 0 , + ∞ ) .
We denote by D = { u ∈ X : ∥ ∇ u ∥ ∞ ≤ 1 } \mathcal{D}=\{u\in X:\|\nabla u\|_\infty\leq1\} D = { u ∈ X : ∥∇ u ∥ ∞ ≤ 1 } the effective domain and we study the energy and the action functionals
E ( u ) : = Ψ ( u ) − ∫ R N F ( u ) d x , I μ ( u ) : = Ψ ( u ) + μ 2 ∥ u ∥ 2 2 − ∫ R N F ( u ) d x , u ∈ D , μ > 0. (4) \mathcal{E}(u):=\Psi(u)-\int_{\mathbb{R}^{N}}F(u)\,dx,\qquad I_\mu(u):=\Psi(u)+\frac{\mu}{2}\|u\|_2^2-\int_{\mathbb{R}^{N}}F(u)\,dx,\qquad u\in\mathcal{D},\ \mu>0. \tag*{(4)} E ( u ) := Ψ ( u ) − ∫ R N F ( u ) d x , I μ ( u ) := Ψ ( u ) + 2 μ ∥ u ∥ 2 2 − ∫ R N F ( u ) d x , u ∈ D , μ > 0. (4)
The functions of prescribed mass form the set
S m : = { u ∈ D : ∥ u ∥ 2 2 = 2 m 2 } . S_m:=\{u\in\mathcal{D}:\|u\|_2^2=2m^2\}. S m := { u ∈ D : ∥ u ∥ 2 2 = 2 m 2 } .
Since the energy E \mathcal{E} E of (4) is not differentiable, we do not define the solutions of (1) as constrained critical points of E \mathcal{E} E on S m S_m S m : a solution of (1) is a pair ( μ , u ) ∈ ( 0 , + ∞ ) × S m (\mu,u)\in(0,+\infty)\times S_m ( μ , u ) ∈ ( 0 , + ∞ ) × S m such that u u u is a critical point of I μ I_\mu I μ in the sense of Szulkin [41 ] , see (5) below, the frequency μ \mu μ playing the role of a Lagrange multiplier. By the results recalled below, such a u u u is a classical solution of the equation in (1) with ∥ ∇ u ∥ ∞ < 1 \lVert\nabla u\rVert_\infty<1 ∥ ∇ u ∥ ∞ < 1 . When the frequency is specified or clear from the context, we also call u u u itself a solution.
The gradient constraint and the prescribed mass bound ∥ u ∥ ∞ \lVert u\rVert_\infty ∥ u ∥ ∞ uniformly on S m S_m S m , and consequently E \mathcal{E} E is bounded from below on every mass sphere, even for a purely focusing L 2 L^2 L 2 -supercritical nonlinearity. Both estimates are proved in Appendix A. Unlike the focusing pure power problem for the Laplacian, the Born–Infeld problem therefore allows one to minimize the energy globally on S m S_m S m also at supercritical masses. The main results below determine when this infimum is attained and show that the corresponding mass threshold differs from the threshold for normalized action ground states.
In [21 Propositions 2.10 and 2.13] (see also [6 Lemma 2.3] ), it is shown that critical points of I μ I_\mu I μ in the sense of Szulkin [41 ] are weak solutions, and in fact classical solutions with ∥ ∇ u ∥ ∞ < 1 \lVert\nabla u\rVert_\infty<1 ∥ ∇ u ∥ ∞ < 1 , of the prescribed frequency problem ( B I μ ) (BI_\mu) ( B I μ ) . Recall that u ∈ D u\in\mathcal{D} u ∈ D is a critical point of I μ I_\mu I μ if
Ψ ( v ) − Ψ ( u ) + ∂ u Φ μ ( u ) [ v − u ] = Ψ ( v ) − Ψ ( u ) + μ ∫ R N u ( v − u ) d x − ∫ R N f ( u ) ( v − u ) d x ≥ 0 (5) \Psi(v)-\Psi(u)+\partial_u\Phi_\mu(u)[v-u]=\Psi(v)-\Psi(u)+\mu\int_{\mathbb{R}^{N}}u(v-u)\,dx-\int_{\mathbb{R}^{N}}f(u)(v-u)\,dx\geq0 \tag*{(5)} Ψ ( v ) − Ψ ( u ) + ∂ u Φ μ ( u ) [ v − u ] = Ψ ( v ) − Ψ ( u ) + μ ∫ R N u ( v − u ) d x − ∫ R N f ( u ) ( v − u ) d x ≥ 0 (5)
for every v ∈ X v\in X v ∈ X . The Pohožaev-type (or combined Nehari–Pohožaev) functional related to ( B I μ ) (BI_\mu) ( B I μ ) is defined for any μ > 0 \mu>0 μ > 0 as
M μ ( u ) : = Ψ ( u ) − ∫ R N H μ ( u ) d x , M_\mu(u):=\Psi(u)-\int_{\mathbb{R}^{N}}H_\mu(u)\,dx, M μ ( u ) := Ψ ( u ) − ∫ R N H μ ( u ) d x ,
where
H μ ( s ) : = F ( s ) + 1 N f ( s ) s − μ ( 1 2 + 1 N ) ∣ s ∣ 2 . (6) H_\mu(s):=F(s)+\frac{1}{N}f(s)s-\mu\left(\frac{1}{2}+\frac{1}{N}\right)|s|^2. \tag*{(6)} H μ ( s ) := F ( s ) + N 1 f ( s ) s − μ ( 2 1 + N 1 ) ∣ s ∣ 2 . (6)
According to [6 Lemma 2.3] (see also [21 Proposition 2.6] ), every critical point u u u of I μ I_\mu I μ satisfies the Pohožaev identity M μ ( u ) = 0 M_\mu(u)=0 M μ ( u ) = 0 . Let
M μ : = { u ∈ D ∖ { 0 } : M μ ( u ) = 0 } . \mathcal{M}_\mu:=\{u\in\mathcal{D}\setminus\{0\}:M_\mu(u)=0\}. M μ := { u ∈ D ∖ { 0 } : M μ ( u ) = 0 } .
Hence M μ \mathcal{M}_\mu M μ contains all nontrivial critical points of I μ I_\mu I μ . Similarly to [28 Definition 1.1] , we define the action ground state level for a given μ \mu μ as
a ( μ ) : = inf u ∈ M μ I μ ( u ) . (7) a(\mu):=\inf_{u\in\mathcal{M}_\mu} I_\mu(u). \tag*{(7)} a ( μ ) := u ∈ M μ inf I μ ( u ) . (7)
Assumptions (F1)–(F3) coincide with the assumptions (G1)–(G3) of [6 ] for the equation with unit frequency ( μ = 1 ) (\mu=1) ( μ = 1 ) . For an arbitrary μ > 0 \mu>0 μ > 0 , the change of variables v ( x ) = μ u ( x / μ ) v(x)=\sqrt{\mu}\,u(x/\sqrt{\mu}) v ( x ) = μ u ( x / μ ) transforms ( B I μ ) (BI_\mu) ( B I μ ) into the equation with unit frequency and with f f f replaced by g μ ( s ) = μ − 1 / 2 f ( s / μ ) g_\mu(s)=\mu^{-1/2}f(s/\sqrt{\mu}) g μ ( s ) = μ − 1/2 f ( s / μ ) , which satisfies the same assumptions, see also Appendix B. In particular, a ( μ ) > 0 a(\mu)>0 a ( μ ) > 0 by [6 Lemma 5.4] , and [6 Theorem 1.6 and Section 5] apply to I μ I_\mu I μ : a ( μ ) a(\mu) a ( μ ) is attained by some u μ ∈ M μ u_\mu\in\mathcal{M}_\mu u μ ∈ M μ , every minimizer is a critical point of I μ I_\mu I μ in the sense of (5) (see [6 Lemma 5.5] , or apply [38 Lemma A.1] ), and a radial minimizer can be chosen. In Appendix B we supplement [6 ] by showing that every minimizer is radially symmetric about some point, see Theorem B.5 .
A minimizer u ∈ M μ u\in\mathcal{M}_\mu u ∈ M μ satisfying I μ ( u ) = a ( μ ) I_\mu(u)=a(\mu) I μ ( u ) = a ( μ ) is called an action ground state solution at frequency μ \mu μ . By the results just recalled, these are precisely the nontrivial critical points of I μ I_\mu I μ , in the sense of (5) , with least action among all nontrivial critical points. We denote their set by
C ( μ ) : = { u ∈ M μ : I μ ( u ) = a ( μ ) } . \mathcal{C}(\mu):=\{u\in\mathcal{M}_\mu:I_\mu(u)=a(\mu)\}. C ( μ ) := { u ∈ M μ : I μ ( u ) = a ( μ )} .
This set is nonempty, and every element is radial up to a translation. If in addition u ∈ S m u \in S_m u ∈ S m , we call u u u a normalized action ground state solution of mass 2 m 2 2m^2 2 m 2 . The associated pair ( μ , u ) (\mu,u) ( μ , u ) solves (1.1).
Main results
We introduce
b m ( μ ) : = a ( μ ) − μ m 2 . (8) b_m(\mu):=a(\mu)-\mu m^2. \tag*{(8)} b m ( μ ) := a ( μ ) − μ m 2 . (8)
Following [25 , 28 ] , we look for solutions of (1.1) among the action ground states of ( B I μ ) (BI_\mu) ( B I μ ) , choosing the frequency μ \mu μ in such a way that the mass is the prescribed one. Since neither the uniqueness of the action ground states nor the differentiability of a a a is known, we consider half of the smallest and of the largest mass in C ( μ ) \mathcal{C}(\mu) C ( μ ) ,
c − ( μ ) : = inf u ∈ C ( μ ) 1 2 ∥ u ∥ 2 2 , c + ( μ ) : = sup u ∈ C ( μ ) 1 2 ∥ u ∥ 2 2 , (9) c_-(\mu):=\inf_{u\in\mathcal{C}(\mu)}\frac{1}{2}\|u\|_2^2,\qquad c_+(\mu):=\sup_{u\in\mathcal{C}(\mu)}\frac{1}{2}\|u\|_2^2, \tag*{(9)} c − ( μ ) := u ∈ C ( μ ) inf 2 1 ∥ u ∥ 2 2 , c + ( μ ) := u ∈ C ( μ ) sup 2 1 ∥ u ∥ 2 2 , (9)
and we prove in Theorem 2.1 that a a a has one-sided derivatives, given by
a + ′ ( μ ) = c − ( μ ) , a − ′ ( μ ) = c + ( μ ) . (10) a_+'(\mu)=c_-(\mu),\qquad a_-'(\mu)=c_+(\mu). \tag*{(10)} a + ′ ( μ ) = c − ( μ ) , a − ′ ( μ ) = c + ( μ ) . (10)
If μ 0 > 0 \mu_0>0 μ 0 > 0 is a local minimum point of b m b_m b m , these formulas give
0 ≤ ( b m ) + ′ ( μ 0 ) = c − ( μ 0 ) − m 2 ≤ c + ( μ 0 ) − m 2 = ( b m ) − ′ ( μ 0 ) ≤ 0 , 0\leq(b_m)_+'(\mu_0)=c_-(\mu_0)-m^2\leq c_+(\mu_0)-m^2=(b_m)_-'(\mu_0)\leq0, 0 ≤ ( b m ) + ′ ( μ 0 ) = c − ( μ 0 ) − m 2 ≤ c + ( μ 0 ) − m 2 = ( b m ) − ′ ( μ 0 ) ≤ 0 ,
hence c − ( μ 0 ) = c + ( μ 0 ) = m 2 c_-(\mu_0)=c_+(\mu_0)=m^2 c − ( μ 0 ) = c + ( μ 0 ) = m 2 : every u ∈ C ( μ 0 ) u\in\mathcal{C}(\mu_0) u ∈ C ( μ 0 ) has mass 2 m 2 2m^2 2 m 2 , and ( μ 0 , u ) (\mu_0,u) ( μ 0 , u ) solves (1.1). Our first main result describes the masses for which this happens.
Theorem 1.1. Assume (F1)–(F4) with $2+\frac{4}{N}\leq p<2^$. Then c ∗ : = inf μ > 0 c − ( μ ) > 0 c_*:=\inf_{\mu>0}c_-(\mu)>0 c ∗ := inf μ > 0 c − ( μ ) > 0 and it is attained if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 . For every m 2 > c ∗ m^2>c_* m 2 > c ∗ , and for m 2 = c ∗ m^2=c_* m 2 = c ∗ whenever $c_ is attained, problem (1.1) admits a radial solution of constant sign with a positive Lagrange multiplier. More precisely:
(i) if m 2 < c ∗ m^2<c_* m 2 < c ∗ , or if m 2 = c ∗ m^2=c_* m 2 = c ∗ and c ∗ c_* c ∗ is not attained, then no action ground state u ∈ C ( μ ) u\in\mathcal{C}(\mu) u ∈ C ( μ ) , μ > 0 \mu>0 μ > 0 , satisfies ∥ u ∥ 2 2 = 2 m 2 \|u\|_2^2=2m^2 ∥ u ∥ 2 2 = 2 m 2 ;
(ii) if m 2 = c ∗ = c − ( μ 0 ) m^2=c_*=c_-(\mu_0) m 2 = c ∗ = c − ( μ 0 ) for some μ 0 > 0 \mu_0>0 μ 0 > 0 , then problem (1.1) has a radial solution ( μ 0 , u ) (\mu_0,u) ( μ 0 , u ) of constant sign with u ∈ C ( μ 0 ) u\in\mathcal{C}(\mu_0) u ∈ C ( μ 0 ) . Moreover, it holds that E ( u ) = b m ( μ 0 ) ≥ 0 E(u)=b_m(\mu_0)\geq0 E ( u ) = b m ( μ 0 ) ≥ 0 , and the inequality is strict if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 ;
(iii) if m 2 > c ∗ m^2>c_* m 2 > c ∗ , then there is μ > 0 \mu>0 μ > 0 such that every u ∈ C ( μ ) u\in\mathcal{C}(\mu) u ∈ C ( μ ) satisfies 1 2 ∥ u ∥ 2 2 = c − ( μ ) = c + ( μ ) = m 2 \frac{1}{2}\|u\|_2^2=c_-(\mu)=c_+(\mu)=m^2 2 1 ∥ u ∥ 2 2 = c − ( μ ) = c + ( μ ) = m 2 .
In particular, (1.1) has a radial solution ( μ , u ) (\mu,u) ( μ , u ) of constant sign with E ( u ) = b m ( μ ) E(u)=b_m(\mu) E ( u ) = b m ( μ ) . Moreover, if $f is odd, all the solutions in (ii) and (iii) can be taken positive.
Remark that whether problem (1.1) has a solution in the range m 2 < c ∗ m^2<c_* m 2 < c ∗ is still open. We next present the existence result in the L 2 L^2 L 2 -subcritical case.
Theorem 1.2. Assume (F1)–(F4) with $2<p<2+\frac{4}{N}$. Then for every m > 0 m>0 m > 0 there is μ > 0 \mu>0 μ > 0 such that every u ∈ C ( μ ) u\in\mathcal{C}(\mu) u ∈ C ( μ ) satisfies 1 2 ∥ u ∥ 2 2 = c − ( μ ) = c + ( μ ) = m 2 \frac{1}{2}\|u\|_2^2=c_-(\mu)=c_+(\mu)=m^2 2 1 ∥ u ∥ 2 2 = c − ( μ ) = c + ( μ ) = m 2 . In particular, (1.1) has a radial solution ( μ , u ) (\mu,u) ( μ , u ) of constant sign, positive if $f is odd, with E ( u ) = b m ( μ ) = min ( 0 , ∞ ) b m < 0 E(u)=b_m(\mu)=\min_{(0,\infty)}b_m<0 E ( u ) = b m ( μ ) = min ( 0 , ∞ ) b m < 0 .
In the L 2 L^2 L 2 -subcritical case, Theorem 1.2 recovers the existence result of [1 ] without truncating the operator and without any upper restriction on the growth of f f f at infinity. In the L 2 L^2 L 2 -supercritical case, Theorem 1.1 provides normalized solutions for every m 2 > c ∗ m^2>c_* m 2 > c ∗ : the solutions for m 2 > c ∗ m^2>c_* m 2 > c ∗ come from a local minimum point of b m b_m b m , while for m 2 = c ∗ m^2=c_* m 2 = c ∗ a ground state of minimum mass is selected. In the L 2 L^2 L 2 -critical case the same holds for every m 2 > c ∗ m^2>c_* m 2 > c ∗ , but Lemma 3.2 only bounds the masses of the action ground states away from zero at small frequencies, instead of forcing them to infinity. Hence the attainment of c ∗ c_* c ∗ is not automatic, and the mass 2 c ∗ 2c_* 2 c ∗ is attained by an action ground state exactly when c ∗ c_{*} c ∗ is attained. We stress that the nonexistence statement (i) concerns action ground states only, and not arbitrary solutions of (1.1).
The solutions of Theorems 1.1 and 1.2 are action ground states of ( B I μ ) (BI_{\mu}) ( B I μ ) , which have constant sign by Theorem B.5 , but for a non-odd f f f they need not be positive: if − f ( − s ) > f ( s ) -f(-s)>f(s) − f ( − s ) > f ( s ) for every s > 0 s>0 s > 0 , then F ( − s ) > F ( s ) F(-s)>F(s) F ( − s ) > F ( s ) for s > 0 s>0 s > 0 , hence Λ ( μ , − u ) < Λ ( μ , u ) \Lambda(\mu,-u)<\Lambda(\mu,u) Λ ( μ , − u ) < Λ ( μ , u ) for every u ≥ 0 u\geq0 u ≥ 0 , u ≠ 0 u\neq0 u = 0 , with Λ \Lambda Λ as in (15) , and all the action ground states are negative. When f f f is odd, the solutions can be taken positive. For a general f f f , positive solutions are obtained by applying Theorems 1.1–1.3 to the odd extension of f ∣ [ 0 , + ∞ ) f|_{[0,+\infty)} f ∣ [ 0 , + ∞ ) , which satisfies (F1)–(F4) as well. The mass thresholds in this application are those associated with the odd extension. The resulting positive solutions solve the original equation, but need not be action or energy ground states for the original non-odd nonlinearity.
We next investigate the relation between the normalized action ground states obtained in Theorems 1.1 and 1.2 and the minimization of the energy at prescribed mass. For m > 0 m>0 m > 0 , let
e m : = inf u ∈ S m E ( u ) , (11) e_m:=\inf_{u\in S_m}\mathcal{E}(u), \tag*{(11)} e m := u ∈ S m inf E ( u ) , (11)
which is finite by Lemma A.2 in all three mass regimes, in the L 2 L^{2} L 2 -critical and in the L 2 L^{2} L 2 -supercritical case as well. A minimizer u ∈ S m u\in S_m u ∈ S m satisfying E ( u ) = e m \mathcal{E}(u)=e_m E ( u ) = e m is called an energy ground state of mass 2 m 2 2m^{2} 2 m 2 . For the semilinear L 2 L^{2} L 2 -critical problem, a comparison between the minimization at fixed mass and the Lagrangian minimax levels is given in [24 Theorem 1.6 and Remark 1.11] . The following theorem characterizes the level e m e_m e m and its attainment, and it identifies the normalized action ground states which are energy ground states.
Theorem 1.3 (energy ground states). Assume (F1)–(F4) with 2 < p < 2 ∗ 2<p<2^{*} 2 < p < 2 ∗ and let m > 0 m>0 m > 0 .
(i) e m = inf μ > 0 b m ( μ ) e_m=\inf_{\mu>0}b_m(\mu) e m = inf μ > 0 b m ( μ ) , and m 2 ↦ e m m^{2}\mapsto e_m m 2 ↦ e m is concave and nonincreasing .
(ii) e m e_m e m is attained if and only if b m b_m b m attains its infimum on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . In this case, every global minimum point μ m \mu_m μ m of b m b_m b m satisfies c − ( μ m ) = c + ( μ m ) = m 2 c_{-}(\mu_m)=c_{+}(\mu_m)=m^{2} c − ( μ m ) = c + ( μ m ) = m 2 , and u ∈ S m u\in S_m u ∈ S m is an energy ground state if and only if u ∈ C ( μ m ) u\in C(\mu_m) u ∈ C ( μ m ) for some global minimum point μ m \mu_m μ m of b m b_m b m . In particular, ( μ m , u ) (\mu_m,u) ( μ m , u ) solves (1.1), and u u u has constant sign and is radially symmetric about some point .
(iii) If 2 < p < 2 + 4 N 2<p<2+\frac{4}{N} 2 < p < 2 + N 4 , then e m < 0 e_m<0 e m < 0 is attained for every m > 0 m>0 m > 0 . If 2 + 4 N ≤ p < 2 ∗ 2+\frac{4}{N}\leq p<2^{*} 2 + N 4 ≤ p < 2 ∗ , set
r e : = inf μ > 0 a ( μ ) μ . r_e:=\inf_{\mu>0}\frac{a(\mu)}{\mu}. r e := μ > 0 inf μ a ( μ ) . Then c ∗ ≤ r e < ∞ c_{*}\leq r_e<\infty c ∗ ≤ r e < ∞ , with c ∗ c_{*} c ∗ as in Theorem 1.1, and c ∗ < r e c_{*}<r_e c ∗ < r e if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 . The infimum e m = 0 e_m=0 e m = 0 is not attained for m 2 < r e m^{2}<r_e m 2 < r e , whereas e m < 0 e_m<0 e m < 0 is attained for m 2 > r e m^{2}>r_e m 2 > r e . For m 2 = r e m^{2}=r_e m 2 = r e , e m = 0 e_m=0 e m = 0 is attained if and only if μ ↦ a ( μ ) / μ \mu\mapsto a(\mu)/\mu μ ↦ a ( μ ) / μ attains its infimum, which is the case if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 .
Theorem 1.3(i) identifies − e m = sup μ > 0 ( μ m 2 − a ( μ ) ) -e_m=\sup_{\mu>0}(\mu m^{2}-a(\mu)) − e m = sup μ > 0 ( μ m 2 − a ( μ )) , as a function of m 2 m^{2} m 2 , with the Legendre–Fenchel transform of a a a . This is the Born–Infeld counterpart of [28 Theorem 1.2] , while (ii) corresponds to the rigidity result [28 Theorem 1.3] . The rays { t u : t > 0 } \{tu:t>0\} { t u : t > 0 } on which the Nehari approach of [28 ] is based leave D \mathcal{D} D for large t t t , and we use the combined Nehari–Pohožaev constraint instead. The resulting two-threshold structure appears even for a single focusing power: for c ∗ 2 < r e c_{*}^{2}<r_e c ∗ 2 < r e there are action ground states of positive energy but no energy ground state, whereas for m 2 > r e m^{2}>r_e m 2 > r e the energy minimum is attained. In the L 2 L^{2} L 2 -supercritical case c ∗ < r e c_{*}<r_e c ∗ < r e and both infima are attained. In the L 2 L^{2} L 2 -critical case our estimates decide neither whether c ∗ < r e c_{*}<r_e c ∗ < r e nor whether the two infima are attained. A similar distinction between normalized solutions and energy minimizers is known for focusing–defocusing nonlinearities: for the Schrödinger equation with a focusing cubic and a defocusing quintic term on R 3 \mathbb{R}^{3} R 3 , the infimum of the energy on the mass sphere is attained only for masses at and above a threshold, whereas normalized solutions of positive energy, which are not energy minimizers, exist as well, see [25 , 34 ] . In the present problem the boundedness of the energy and the separation of the two thresholds are caused by the gradient constraint alone, without any defocusing term.
For the L 2 L^{2} L 2 -critical pure power, the threshold can be computed explicitly, and the nonattainment at the threshold can be decided, by means of the sharp Gagliardo–Nirenberg inequality of Weinstein [43 ] .
Corollary 1.4 (L 2 L^{2} L 2 -critical pure power). Let f ( s ) = ∣ s ∣ 4 / N s f(s)=|s|^{4/N}s f ( s ) = ∣ s ∣ 4/ N s , and let Q Q Q be a positive ground state of
− Δ Q + Q = Q 1 + 4 / N in R N . -\Delta Q+Q=Q^{1+4/N}\quad\text{in }\mathbb{R}^{N}. − Δ Q + Q = Q 1 + 4/ N in R N . Set m c : = ∥ Q ∥ 2 / 2 m_{\mathrm{c}}:=\lVert Q\rVert_{2}/\sqrt{2} m c := ∥ Q ∥ 2 / 2 . Then r e = m c 2 r_{\mathrm{e}}=m_{\mathrm{c}}^{2} r e = m c 2 , with r e r_{\mathrm{e}} r e as in Theorem 1.3 (iii), and
(i) for 0 < m ≤ m c 0<m\le m_{\mathrm{c}} 0 < m ≤ m c , e m = 0 e_m=0 e m = 0 and the infimum is not attained;
(ii) for m > m c m>m_{\mathrm{c}} m > m c , − ∞ < e m < 0 -\infty<e_m<0 − ∞ < e m < 0 and the infimum is attained.
In particular, μ ↦ a ( μ ) / μ \mu\mapsto a(\mu)/\mu μ ↦ a ( μ ) / μ does not attain its infimum.
For comparison, if the Born–Infeld density is replaced by the Dirichlet density ∣ ∇ u ∣ 2 / 2 |\nabla u|^{2}/2 ∣∇ u ∣ 2 /2 and S m S_m S m by the mass sphere in H 1 ( R N ) H^{1}(\mathbb{R}^{N}) H 1 ( R N ) , without any gradient constraint, then the infimum of the energy is zero for 0 < m ≤ m c 0<m\le m_{\mathrm{c}} 0 < m ≤ m c , attained only when m = m c m=m_{\mathrm{c}} m = m c , and it is − ∞ -\infty − ∞ for m > m c m>m_{\mathrm{c}} m > m c , see e.g. [22 , 43 ] . Thus the Born–Infeld term changes both the behavior at the threshold and the existence of minimizers above it. Corollary 1.4 is proved at the end of Section 3.
Let us describe the strategy of the proofs. The dilations u t ( x ) = t u ( x / t ) u_t(x)=tu(x/t) u t ( x ) = t u ( x / t ) preserve the gradient constraint and provide a projection onto the combined Nehari–Pohožaev set M μ \mathcal{M}_{\mu} M μ . Comparing the maxima of I μ I_{\mu} I μ along these dilations at two frequencies, see Lemma 2.2 , we obtain
lim sup h → 0 + a ( μ + h ) − a ( μ ) h ≤ c − ( μ ) , lim inf h → 0 + a ( μ ) − a ( μ − h ) h ≥ c + ( μ ) for every μ > 0 , \limsup_{h\to0^{+}}\frac{a(\mu+h)-a(\mu)}{h}\le c_{-}(\mu),\qquad\liminf_{h\to0^{+}}\frac{a(\mu)-a(\mu-h)}{h}\ge c_{+}(\mu)\quad\text{for every }\mu>0, h → 0 + lim sup h a ( μ + h ) − a ( μ ) ≤ c − ( μ ) , h → 0 + lim inf h a ( μ ) − a ( μ − h ) ≥ c + ( μ ) for every μ > 0 ,
see (26) . These two inequalities already imply that c − ( μ ) = c + ( μ ) = m 2 c_{-}(\mu)=c_{+}(\mu)=m^{2} c − ( μ ) = c + ( μ ) = m 2 at every local minimum point μ \mu μ of b m b_m b m . Such a point exists in ( μ 1 , ∞ ) (\mu_1,\infty) ( μ 1 , ∞ ) whenever c − ( μ 1 ) < m 2 c_{-}(\mu_1)<m^{2} c − ( μ 1 ) < m 2 : in this case b m ( μ ) < b m ( μ 1 ) b_m(\mu)<b_m(\mu_1) b m ( μ ) < b m ( μ 1 ) for μ > μ 1 \mu>\mu_1 μ > μ 1 close to μ 1 \mu_1 μ 1 , and b m ( μ ) → + ∞ b_m(\mu)\to+\infty b m ( μ ) → + ∞ as μ → ∞ \mu\to\infty μ → ∞ , because the gradient constraint forces a ( μ ) / μ → ∞ a(\mu)/\mu\to\infty a ( μ ) / μ → ∞ , see Lemma 3.1. This gives the normalized action ground states of Theorem 1.1(iii), in the L 2 L^{2} L 2 -critical and in the L 2 L^{2} L 2 -supercritical case alike, while in the L 2 L^{2} L 2 -subcritical case b m b_m b m is negative at small frequencies and its minimum is global. We also prove that the ground states of the prescribed frequency problem obtained in [6 ] are compact, up to translations, when the frequency varies in a compact subset of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . This compactness gives the reverse inequalities, hence the one-sided derivative formulas (1.10), and the lower semicontinuity of c − c_{-} c − , on which the attainment of c ∗ c_{*} c ∗ for p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 relies. The relation between the action and the energy ground state levels does not depend on these formulas. It follows from two elementary observations: if an action ground state at the frequency μ \mu μ is dilated onto S m S_m S m , then its energy does not exceed b m ( μ ) b_m(\mu) b m ( μ ) , and conversely every u ∈ S m u\in S_m u ∈ S m with E ( u ) ≤ 0 \mathcal{E}(u)\le0 E ( u ) ≤ 0 belongs to M μ \mathcal{M}_{\mu} M μ for a suitable frequency μ > 0 \mu>0 μ > 0 .
The Lagrangian constructions of [23 , 24 , 25 ] are carried out in the radial space H r 1 ( R N ) H^{1}_{r}(\mathbb{R}^{N}) H r 1 ( R N ) , whereas here a ( μ ) a(\mu) a ( μ ) and C ( μ ) C(\mu) C ( μ ) are defined in the whole space X X X , without any symmetry restriction. In the L 2 L^{2} L 2 -supercritical construction of [1 ] , the boundedness of the Palais–Smale sequences satisfying an asymptotic Pohožaev identity relies on a coercivity condition, denoted there by (b3), which the truncated Born–Infeld operators do not satisfy near the singular gradient level, see [1 Remark 5.6] . For this reason the supercritical results of [1 ] do not pass to the Born–Infeld limit. Our approach avoids this difficulty: we work with the ground states of the untruncated prescribed frequency problem and select their mass afterwards, and no estimate at fixed mass for the truncated problems is needed. The following examples illustrate the scope of (F1)–(F4).
Example 1. Unless stated otherwise, f f f is given for s ≥ 0 s \ge0 s ≥ 0 and extended oddly to R \mathbb{R} R , with f ( 0 ) = 0 f(0)=0 f ( 0 ) = 0 . In (ii)–(iv), let 2 < p < 2 ∗ 2 < p < 2^* 2 < p < 2 ∗ .
Pure and combined powers. Let f ( s ) = ∑ i = 1 k a i s p i − 1 f(s)=\sum_{i=1}^{k}a_i s^{p_i-1} f ( s ) = ∑ i = 1 k a i s p i − 1 , where k ≥ 1 k \ge1 k ≥ 1 , a i > 0 a_i>0 a i > 0 , 2 < p 1 < ⋯ < p k 2<p_1<\cdots<p_k 2 < p 1 < ⋯ < p k and p 1 < 2 ∗ p_1<2^* p 1 < 2 ∗ . Then (F1)–(F4) hold with p = p 1 p=p_1 p = p 1 and f ( s ) / s p 1 − 1 → a 1 f(s)/s^{p_1-1}\to a_1 f ( s ) / s p 1 − 1 → a 1 as s → 0 + s\to0^+ s → 0 + . The higher powers may equal or exceed 2 ∗ 2^* 2 ∗ .
Non-power growth at infinity.
f ( s ) = s log ( 1 + s p − 2 ) and f ( s ) = s p − 1 e s 2 . f(s)=s\log(1+s^{p-2}) \qquad\text{and}\qquad f(s)=s^{p-1}e^{s^2}. f ( s ) = s log ( 1 + s p − 2 ) and f ( s ) = s p − 1 e s 2 . Both satisfy (F1)–(F4), with f ( s ) / s p − 1 → 1 f(s)/s^{p-1}\to1 f ( s ) / s p − 1 → 1 as s → 0 + s\to0^+ s → 0 + . At infinity , the first grows more slowly than any superlinear power, whereas the second grows faster than any power.
No asymptotic coefficient at zero.
f ( s ) = s p − 1 [ 2 + sin ( ( p − 2 ) log s ) ] , s > 0. f(s)=s^{p-1}\left[2+\sin((p-2)\log s)\right],\qquad s>0. f ( s ) = s p − 1 [ 2 + sin (( p − 2 ) log s ) ] , s > 0. The odd extension is of class C 1 C^1 C 1 with f ′ ( 0 ) = 0 f'(0)=0 f ′ ( 0 ) = 0 . Writing ϑ = ( p − 2 ) log s \vartheta=(p-2)\log s ϑ = ( p − 2 ) log s , we have
( f ( s ) s ) ′ = ( p − 2 ) s p − 3 ( 2 + sin ϑ + cos ϑ ) > 0. \left(\frac{f(s)}{s}\right)'=(p-2)s^{p-3}(2+\sin\vartheta+\cos\vartheta)>0. ( s f ( s ) ) ′ = ( p − 2 ) s p − 3 ( 2 + sin ϑ + cos ϑ ) > 0. Thus (F1)–(F4) hold, but the lower and upper limits in (F4) are 1 and 3.
Non-odd nonlinearities.
f ( s ) = { a + s p − 1 , s ≥ 0 , − a − ∣ s ∣ p − 1 , s < 0 , a + , a − > 0 , a + ≠ a − . f(s)=\begin{cases}a_+s^{p-1},&s\ge0,\\-a_-|s|^{p-1},&s<0,\end{cases}\qquad a_+,a_->0,\quad a_+\ne a_-. f ( s ) = { a + s p − 1 , − a − ∣ s ∣ p − 1 , s ≥ 0 , s < 0 , a + , a − > 0 , a + = a − . Here (F1)–(F4) hold with lower and upper limits min { a + , a − } \min\{a_+,a_-\} min { a + , a − } and max { a + , a − } \max\{a_+,a_-\} max { a + , a − } in (F4).
The symmetry of every action ground state, rather than merely the existence of a radial one, is established in Appendix B. The proof runs as follows. The minimality implies that for an action ground state u u u of constant sign ε ∈ { − 1 , 1 } \varepsilon\in\{-1,1\} ε ∈ { − 1 , 1 } , the rearrangement ε ∣ u ∣ ∗ \varepsilon|u|^* ε ∣ u ∣ ∗ is again an action ground state with the same kinetic energy, and since the Born–Infeld density is a series of even powers of ∣ ∇ u ∣ |\nabla u| ∣∇ u ∣ with positive coefficients, this reduces to the equality case of the Pólya–Szegő inequality for the Dirichlet integral, settled by Brothers and Ziemer [17 ] , which we use in the quantitative form of Burchard and Ferone [16 ] . A uniqueness argument for the radial ordinary differential equation then gives the strict monotonicity of the radial profile, see Theorem B.5 .
The paper is organized as follows. Section 2 proves the one-sided differentiability of a ( μ ) a(\mu) a ( μ ) . Section 3 derives from the gradient constraint the asymptotics of a ( μ ) a(\mu) a ( μ ) and c ± ( μ ) c_{\pm}(\mu) c ± ( μ ) and contains the proofs of Theorems 1.1–1.3 . Appendix A contains the two elementary estimates (A.1) and (A.2) on which Sections 2–3 rely. Appendix B proves the sign, the symmetry and the monotonicity of every Born–Infeld action ground state. Since this result applies to every ground state, we can center by a translation the ground states selected by their mass, and we obtain the symmetry statements for all the energy ground states.
Notation. ∥ ⋅ ∥ r \|\cdot\|_r ∥ ⋅ ∥ r is the norm of L r ( R N ) L^r(\mathbb{R}^N) L r ( R N ) , B R ( x ) B_R(x) B R ( x ) is the open ball of radius R R R centered at x x x , B R = B R ( 0 ) B_R=B_R(0) B R = B R ( 0 ) , ∣ E ∣ |E| ∣ E ∣ is the Lebesgue measure of E E E , ( v ) + = max { v , 0 } (v)_+=\max\{v,0\} ( v ) + = max { v , 0 } is the positive part of v v v , X r X_r X r is the subspace of radial functions in X X X , and C C C denotes positive constants which may change from line to line.
Action ground state level
In this section, we study the action ground state level as a function of the frequency. The following theorem is the Born–Infeld counterpart of [25 Theorem 1.3] and of [28 Theorem 1.5] . It is the first main ingredient of our approach. Throughout this section we assume (F1)–(F3).
Theorem 2.1. Let a a a , c − c_{-} c − and c + c_{+} c + be the functions defined in (7) and (9) . Then
a + ′ ( μ ) = c − ( μ ) , a − ′ ( μ ) = c + ( μ ) a_{+}^{\prime}(\mu)=c_{-}(\mu), \qquad a_{-}^{\prime}(\mu)=c_{+}(\mu) a + ′ ( μ ) = c − ( μ ) , a − ′ ( μ ) = c + ( μ ) for μ ∈ ( 0 , ∞ ) \mu\in(0,\infty) μ ∈ ( 0 , ∞ ) , where a + ′ ( μ ) a_{+}^{\prime}(\mu) a + ′ ( μ ) and a − ′ ( μ ) a_{-}^{\prime}(\mu) a − ′ ( μ ) are the right and left derivatives, respectively.
To prove Theorem 2.1, we need some preliminaries, most of which are taken from [6 ] . Let us point out that no upper bound on the growth of f f f at infinity is needed in what follows. Indeed, by (F1), for every R > 0 R>0 R > 0 and every compact interval J ⊂ ( 0 , ∞ ) J\subset(0,\infty) J ⊂ ( 0 , ∞ ) there is a constant C R , J C_{R,J} C R , J such that
∣ F ( s ) ∣ + ∣ f ( s ) s ∣ + ∣ H μ ( s ) ∣ ≤ C R , J s 2 , ∣ H μ ′ ( s ) ∣ ≤ C R , J ∣ s ∣ |F(s)|+|f(s)s|+|H_{\mu}(s)|\leq C_{R,J}s^{2}, \qquad|H_{\mu}^{\prime}(s)|\leq C_{R,J}|s| ∣ F ( s ) ∣ + ∣ f ( s ) s ∣ + ∣ H μ ( s ) ∣ ≤ C R , J s 2 , ∣ H μ ′ ( s ) ∣ ≤ C R , J ∣ s ∣
for ∣ s ∣ ≤ R |s|\leq R ∣ s ∣ ≤ R and μ ∈ J \mu\in J μ ∈ J . Since X ↪ L ∞ ( R N ) X\hookrightarrow L^{\infty}(\mathbb{R}^{N}) X ↪ L ∞ ( R N ) , these local estimates justify all the nonlinear integrals below and their differentiation on bounded subsets of X X X .
For u ∈ D u\in\mathcal{D} u ∈ D and θ > 0 \theta>0 θ > 0 , set u θ ( x ) : = θ u ( x / θ ) u_{\theta}(x):=\theta u(x/\theta) u θ ( x ) := θ u ( x / θ ) . This scaling preserves the gradient constraint and satisfies
Ψ ( u θ ) = θ N Ψ ( u ) , ∥ u θ ∥ 2 2 = θ N + 2 ∥ u ∥ 2 2 . \Psi(u_{\theta})=\theta^{N}\Psi(u), \qquad\|u_{\theta}\|_{2}^{2}=\theta^{N+2}\|u\|_{2}^{2}. Ψ ( u θ ) = θ N Ψ ( u ) , ∥ u θ ∥ 2 2 = θ N + 2 ∥ u ∥ 2 2 .
Hence, for every μ > 0 \mu>0 μ > 0 , u ∈ D ∖ { 0 } u\in\mathcal{D}\setminus\{0\} u ∈ D ∖ { 0 } and θ > 0 \theta>0 θ > 0 ,
I μ ( u θ ) = θ N ( Ψ ( u ) + μ θ 2 2 ∥ u ∥ 2 2 − ∫ R N F ( θ u ) d x ) , d d θ I μ ( u θ ) = N θ N − 1 φ ( θ , μ , u ) (12) I_{\mu}(u_{\theta})=\theta^{N}\left(\Psi(u)+\frac{\mu\theta^{2}}{2}\|u\|_{2}^{2}-\int_{\mathbb{R}^{N}}F(\theta u)\,dx\right), \qquad\frac{d}{d\theta}I_{\mu}(u_{\theta})=N\theta^{N-1}\varphi(\theta,\mu,u) \tag*{(12)} I μ ( u θ ) = θ N ( Ψ ( u ) + 2 μ θ 2 ∥ u ∥ 2 2 − ∫ R N F ( θ u ) d x ) , d θ d I μ ( u θ ) = N θ N − 1 φ ( θ , μ , u ) (12)
where
φ ( θ , μ , u ) : = θ − N M μ ( u θ ) = Ψ ( u ) − ∫ R N H μ ( θ u ) d x . \varphi(\theta,\mu,u):=\theta^{-N}M_{\mu}(u_{\theta})=\Psi(u)-\int_{\mathbb{R}^{N}}H_{\mu}(\theta u)\,dx. φ ( θ , μ , u ) := θ − N M μ ( u θ ) = Ψ ( u ) − ∫ R N H μ ( θ u ) d x .
In view of (F1)–(F3) and [6 Lemma 5.2] , for every fixed μ > 0 \mu>0 μ > 0 and u ∈ D ∖ { 0 } u\in\mathcal{D}\setminus\{0\} u ∈ D ∖ { 0 } there exists a unique θ μ ( u ) > 0 \theta_{\mu}(u)>0 θ μ ( u ) > 0 such that u θ μ ( u ) ∈ M μ u_{\theta_{\mu}(u)}\in\mathcal{M}_{\mu} u θ μ ( u ) ∈ M μ . Moreover,
φ ( θ , μ , u ) > 0 for 0 < θ < θ μ ( u ) , φ ( θ , μ , u ) < 0 for θ > θ μ ( u ) . (13) \varphi(\theta,\mu,u)>0 \quad\text{for }0<\theta<\theta_{\mu}(u), \qquad\varphi(\theta,\mu,u)<0 \quad\text{for }\theta>\theta_{\mu}(u). \tag*{(13)} φ ( θ , μ , u ) > 0 for 0 < θ < θ μ ( u ) , φ ( θ , μ , u ) < 0 for θ > θ μ ( u ) . (13)
Indeed, φ ( θ μ ( u ) , μ , u ) = 0 \varphi(\theta_{\mu}(u),\mu,u)=0 φ ( θ μ ( u ) , μ , u ) = 0 and, in view of (6) , the map θ ↦ θ − 2 φ ( θ , μ , u ) \theta\mapsto\theta^{-2}\varphi(\theta,\mu,u) θ ↦ θ − 2 φ ( θ , μ , u ) is strictly decreasing, because Ψ ( u ) > 0 \Psi(u)>0 Ψ ( u ) > 0 and, by (F2), the functions F ( s ) / s 2 F(s)/s^{2} F ( s ) / s 2 and f ( s ) / s f(s)/s f ( s ) / s are nondecreasing in ∣ s ∣ |s| ∣ s ∣ on each of the half-lines ( − ∞ , 0 ) (-\infty,0) ( − ∞ , 0 ) and ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . By (12) and (13) , the map θ ↦ I μ ( u θ ) \theta\mapsto I_{\mu}(u_{\theta}) θ ↦ I μ ( u θ ) is strictly increasing on ( 0 , θ μ ( u ) ] (0,\theta_{\mu}(u)] ( 0 , θ μ ( u )] and strictly decreasing on [ θ μ ( u ) , ∞ ) [\theta_{\mu}(u),\infty) [ θ μ ( u ) , ∞ ) , and it is positive on ( 0 , θ μ ( u ) ] (0,\theta_{\mu}(u)] ( 0 , θ μ ( u )] , because I μ ( u θ ) → 0 I_{\mu}(u_{\theta})\to0 I μ ( u θ ) → 0 as θ → 0 + \theta\to0^{+} θ → 0 + . In particular, θ μ ( u ) = 1 \theta_{\mu}(u)=1 θ μ ( u ) = 1 for u ∈ M μ u\in\mathcal{M}_{\mu} u ∈ M μ and
I μ ( u θ ) < I μ ( u ) for every u ∈ M μ and θ ∈ ( 0 , ∞ ) ∖ { 1 } . (14) I_{\mu}(u_{\theta})<I_{\mu}(u) \qquad\text{for every }u\in\mathcal{M}_{\mu}\text{ and }\theta\in(0,\infty)\setminus\{1\}. \tag*{(14)} I μ ( u θ ) < I μ ( u ) for every u ∈ M μ and θ ∈ ( 0 , ∞ ) ∖ { 1 } . (14)
We define the projection
m μ ( u ) : = u θ μ ( u ) , u ∈ D ∖ { 0 } , m_{\mu}(u):=u_{\theta_{\mu}(u)}, \qquad u\in\mathcal{D}\setminus\{0\}, m μ ( u ) := u θ μ ( u ) , u ∈ D ∖ { 0 } ,
and the maximal-energy map
Λ ( μ , u ) : = I μ ( m μ ( u ) ) = I μ ( u θ μ ( u ) ) = max θ > 0 I μ ( u θ ) . (15) \Lambda(\mu,u):=I_{\mu}(m_{\mu}(u))=I_{\mu}(u_{\theta_{\mu}(u)})=\max_{\theta>0}I_{\mu}(u_{\theta}). \tag*{(15)} Λ ( μ , u ) := I μ ( m μ ( u )) = I μ ( u θ μ ( u ) ) = θ > 0 max I μ ( u θ ) . (15)
Thus a ( μ ) = inf u ∈ D ∖ { 0 } Λ ( μ , u ) a(\mu)=\inf_{u\in\mathcal{D}\setminus\{0\}}\Lambda(\mu,u) a ( μ ) = inf u ∈ D ∖ { 0 } Λ ( μ , u ) , and Λ ( μ , u ) = a ( μ ) \Lambda(\mu,u)=a(\mu) Λ ( μ , u ) = a ( μ ) if u ∈ C ( μ ) u\in\mathcal{C}(\mu) u ∈ C ( μ ) . The next lemma shows that θ μ ( u ) \theta_{\mu}(u) θ μ ( u ) depends continuously on μ \mu μ and u u u , and it compares the action ground state levels at two frequencies.
Lemma 2.2. Let μ n → μ 0 ∈ ( 0 , ∞ ) \mu_n \to\mu_0 \in(0,\infty) μ n → μ 0 ∈ ( 0 , ∞ ) and let w n , w 0 ∈ D ∖ { 0 } w_n,w_0 \in\mathcal{D} \setminus\{0\} w n , w 0 ∈ D ∖ { 0 } be such that w n → w 0 w_n \to w_0 w n → w 0 in X X X . Then θ μ n ( w n ) → θ μ 0 ( w 0 ) \theta_{\mu_n}(w_n) \to\theta_{\mu_0}(w_0) θ μ n ( w n ) → θ μ 0 ( w 0 ) . Moreover,
a ( μ ) ≤ a ( ν ) + μ − ν 2 θ μ ( u ) N + 2 ∥ u ∥ 2 2 (16) a(\mu) \le a(\nu) + \frac{\mu-\nu}{2}\theta_\mu(u)^{N+2}\lVert u\rVert_2^2 \tag*{(16)} a ( μ ) ≤ a ( ν ) + 2 μ − ν θ μ ( u ) N + 2 ∥ u ∥ 2 2 (16) for every μ , ν > 0 \mu,\nu>0 μ , ν > 0 and u ∈ C ( ν ) u \in\mathcal{C}(\nu) u ∈ C ( ν ) .
Proof. Fix θ > 0 \theta>0 θ > 0 . Since Ψ ( w n ) → Ψ ( w 0 ) \Psi(w_n) \to\Psi(w_0) Ψ ( w n ) → Ψ ( w 0 ) by [6 Lemma 2.1] , and since w n → w 0 w_n \to w_0 w n → w 0 in L 2 ( R N ) L^2(\mathbb{R}^N) L 2 ( R N ) and ( w n ) (w_n) ( w n ) is bounded in L ∞ ( R N ) L^\infty(\mathbb{R}^N) L ∞ ( R N ) , the local estimates above yield
φ ( θ , μ n , w n ) → φ ( θ , μ 0 , w 0 ) as n → ∞ . \varphi(\theta,\mu_n,w_n) \to\varphi(\theta,\mu_0,w_0) \qquad\text{as } n \to\infty. φ ( θ , μ n , w n ) → φ ( θ , μ 0 , w 0 ) as n → ∞.
Let θ 0 : = θ μ 0 ( w 0 ) \theta_0 := \theta_{\mu_0}(w_0) θ 0 := θ μ 0 ( w 0 ) and 0 < ε < θ 0 0<\varepsilon<\theta_0 0 < ε < θ 0 . By (13) , we have φ ( θ 0 − ε , μ 0 , w 0 ) > 0 > φ ( θ 0 + ε , μ 0 , w 0 ) \varphi(\theta_0-\varepsilon,\mu_0,w_0)>0>\varphi(\theta_0+\varepsilon,\mu_0,w_0) φ ( θ 0 − ε , μ 0 , w 0 ) > 0 > φ ( θ 0 + ε , μ 0 , w 0 ) , and hence φ ( θ 0 − ε , μ n , w n ) > 0 > φ ( θ 0 + ε , μ n , w n ) \varphi(\theta_0-\varepsilon,\mu_n,w_n)>0>\varphi(\theta_0+\varepsilon,\mu_n,w_n) φ ( θ 0 − ε , μ n , w n ) > 0 > φ ( θ 0 + ε , μ n , w n ) for n n n large. Using (13) again, we get θ 0 − ε < θ μ n ( w n ) < θ 0 + ε \theta_0-\varepsilon<\theta_{\mu_n}(w_n)<\theta_0+\varepsilon θ 0 − ε < θ μ n ( w n ) < θ 0 + ε for n n n large, which proves the first assertion.
Let now μ , ν > 0 \mu,\nu>0 μ , ν > 0 , u ∈ C ( ν ) u \in\mathcal{C}(\nu) u ∈ C ( ν ) and θ : = θ μ ( u ) \theta:=\theta_\mu(u) θ := θ μ ( u ) . Since u θ ∈ M μ u_\theta\in\mathcal{M}_\mu u θ ∈ M μ and u ∈ M ν u \in\mathcal{M}_\nu u ∈ M ν , (14) gives
a ( μ ) ≤ I μ ( u θ ) = I ν ( u θ ) + μ − ν 2 ∥ u θ ∥ 2 2 ≤ I ν ( u ) + μ − ν 2 θ N + 2 ∥ u ∥ 2 2 , a(\mu) \le I_\mu(u_\theta)=I_\nu(u_\theta)+\frac{\mu-\nu}{2}\lVert u_\theta\rVert_2^2 \le I_\nu(u)+\frac{\mu-\nu}{2}\theta^{N+2}\lVert u\rVert_2^2, a ( μ ) ≤ I μ ( u θ ) = I ν ( u θ ) + 2 μ − ν ∥ u θ ∥ 2 2 ≤ I ν ( u ) + 2 μ − ν θ N + 2 ∥ u ∥ 2 2 ,
which is (16) , because I ν ( u ) = a ( ν ) I_\nu(u)=a(\nu) I ν ( u ) = a ( ν ) .
Lemma 2.3. Let μ n → μ 0 > 0 \mu_n \to\mu_0>0 μ n → μ 0 > 0 . Then lim sup n → ∞ a ( μ n ) ≤ a ( μ 0 ) \limsup_{n\to\infty}a(\mu_n)\le a(\mu_0) lim sup n → ∞ a ( μ n ) ≤ a ( μ 0 ) , and ( a ( μ n ) ) (a(\mu_n)) ( a ( μ n )) is bounded away from zero.
Proof. Fix u 0 ∈ C ( μ 0 ) u_0 \in\mathcal{C}(\mu_0) u 0 ∈ C ( μ 0 ) . By Lemma 2.2, we have θ μ n ( u 0 ) → θ μ 0 ( u 0 ) = 1 \theta_{\mu_n}(u_0)\to\theta_{\mu_0}(u_0)=1 θ μ n ( u 0 ) → θ μ 0 ( u 0 ) = 1 , and (16) with ν = μ 0 \nu=\mu_0 ν = μ 0 gives
a ( μ n ) ≤ a ( μ 0 ) + μ n − μ 0 2 θ μ n ( u 0 ) N + 2 ∥ u 0 ∥ 2 2 . a(\mu_n)\le a(\mu_0)+\frac{\mu_n-\mu_0}{2}\theta_{\mu_n}(u_0)^{N+2}\lVert u_0\rVert_2^2. a ( μ n ) ≤ a ( μ 0 ) + 2 μ n − μ 0 θ μ n ( u 0 ) N + 2 ∥ u 0 ∥ 2 2 .
Letting n → ∞ n\to\infty n → ∞ , we get the first assertion.
We now prove the lower bound. Choose μ − , μ + > 0 \mu_-,\mu_+>0 μ − , μ + > 0 such that μ n ∈ [ μ − , μ + ] \mu_n\in[\mu_-,\mu_+] μ n ∈ [ μ − , μ + ] for all sufficiently large n n n , and set G ( s ) : = F ( s ) + 1 N f ( s ) s G(s):=F(s)+\frac{1}{N}f(s)s G ( s ) := F ( s ) + N 1 f ( s ) s . In view of (F1), there exists δ > 0 \delta>0 δ > 0 such that
G ( s ) ≤ ( N + 2 ) μ − 4 N s 2 for ∣ s ∣ ≤ δ . (17) G(s)\le\frac{(N+2)\mu_-}{4N}s^2 \qquad\text{for } |s|\le\delta. \tag*{(17)} G ( s ) ≤ 4 N ( N + 2 ) μ − s 2 for ∣ s ∣ ≤ δ . (17)
We claim that every u ∈ M μ u\in\mathcal{M}_\mu u ∈ M μ , with μ ∈ [ μ − , μ + ] \mu\in[\mu_-,\mu_+] μ ∈ [ μ − , μ + ] , satisfies
∥ u ∥ ∞ > δ . (18) \lVert u\rVert_\infty>\delta. \tag*{(18)} ∥ u ∥ ∞ > δ . (18)
Otherwise, by (17) , one has
M μ ( u ) = Ψ ( u ) + N + 2 2 N μ ∥ u ∥ 2 2 − ∫ R N G ( u ) d x ≥ Ψ ( u ) + ( N + 2 ) μ − 4 N ∥ u ∥ 2 2 > 0 , \begin{aligned}
M_\mu(u)=\Psi(u)+\frac{N+2}{2N}\mu\lVert u\rVert_2^2-\int_{\mathbb{R}^N}G(u)\,dx \\
&\ge\Psi(u)+\frac{(N+2)\mu_-}{4N}\lVert u\rVert_2^2>0,
\end{aligned} M μ ( u ) = Ψ ( u ) + 2 N N + 2 μ ∥ u ∥ 2 2 − ∫ R N G ( u ) d x ≥ Ψ ( u ) + 4 N ( N + 2 ) μ − ∥ u ∥ 2 2 > 0 ,
which contradicts u ∈ M μ u\in\mathcal{M}_\mu u ∈ M μ .
For u ∈ D u\in\mathcal{D} u ∈ D , we have ∥ ∇ u ∥ ∞ ≤ 1 \lVert\nabla u\rVert_\infty\le1 ∥ ∇ u ∥ ∞ ≤ 1 and
Ψ ( u ) ≥ 1 2 ∥ ∇ u ∥ 2 2 . (19) \Psi(u)\ge\frac{1}{2}\lVert\nabla u\rVert_2^2. \tag*{(19)} Ψ ( u ) ≥ 2 1 ∥ ∇ u ∥ 2 2 . (19)
since 1 − 1 − t 2 ≥ 1 2 t 2 1-\sqrt{1-t^2}\ge\frac{1}{2}t^2 1 − 1 − t 2 ≥ 2 1 t 2 for 0 ≤ t ≤ 1 0\le t\le1 0 ≤ t ≤ 1 . Combining (18) , (19) and the second inequality in (A.1), we get
Ψ ( u ) ≥ η > 0 , \Psi(u)\ge\eta>0, Ψ ( u ) ≥ η > 0 ,
where η \eta η is independent of μ ∈ [ μ − , μ + ] \mu\in[\mu_{-},\mu_{+}] μ ∈ [ μ − , μ + ] . Moreover, (F2) implies
1 2 f ( s ) s − F ( s ) ≥ 0 for every s ∈ R . (20) \frac{1}{2}f(s)s-F(s)\ge0 \qquad\text{for every }s\in\mathbb{R}. \tag*{(20)} 2 1 f ( s ) s − F ( s ) ≥ 0 for every s ∈ R . (20)
For every u ∈ M μ u\in\mathcal{M}_{\mu} u ∈ M μ , the identity M μ ( u ) = 0 M_{\mu}(u)=0 M μ ( u ) = 0 yields
I μ ( u ) = 2 N + 2 [ Ψ ( u ) + ∫ R N ( 1 2 f ( u ) u − F ( u ) ) d x ] ≥ 2 η N + 2 . I_{\mu}(u)=\frac{2}{N+2}\left[\Psi(u)+\int_{\mathbb{R}^{N}}\left(\frac{1}{2}f(u)u-F(u)\right)\,dx\right]\ge\frac{2\eta}{N+2}. I μ ( u ) = N + 2 2 [ Ψ ( u ) + ∫ R N ( 2 1 f ( u ) u − F ( u ) ) d x ] ≥ N + 2 2 η .
Taking the infimum over u ∈ M μ u\in\mathcal{M}_{\mu} u ∈ M μ , we obtain the lower bound.
Lemma 2.4. If μ n → μ 0 ∈ ( 0 , ∞ ) \mu_{n}\to\mu_{0}\in(0,\infty) μ n → μ 0 ∈ ( 0 , ∞ ) and u n ∈ C ( μ n ) u_{n}\in C(\mu_{n}) u n ∈ C ( μ n ) for every n ∈ N n\in\mathbb{N} n ∈ N , then, after passing to a subsequence, there exist { y n } n ∈ N ⊂ R N \{y_{n}\}_{n\in\mathbb{N}}\subset\mathbb{R}^{N} { y n } n ∈ N ⊂ R N and u 0 ∈ C ( μ 0 ) u_{0}\in C(\mu_{0}) u 0 ∈ C ( μ 0 ) such that
u n ( ⋅ + y n ) → u 0 strongly in X . u_{n}(\cdot+y_{n})\to u_{0}\qquad\text{strongly in }X. u n ( ⋅ + y n ) → u 0 strongly in X . Proof. Set
K ( s ) : = 1 2 f ( s ) s − F ( s ) , G ( s ) : = F ( s ) + 1 N f ( s ) s . K(s):=\frac{1}{2}f(s)s-F(s),\qquad G(s):=F(s)+\frac{1}{N}f(s)s. K ( s ) := 2 1 f ( s ) s − F ( s ) , G ( s ) := F ( s ) + N 1 f ( s ) s . Since u n ∈ C ( μ n ) u_{n}\in C(\mu_{n}) u n ∈ C ( μ n ) , the identity M μ n ( u n ) = 0 M_{\mu_{n}}(u_{n})=0 M μ n ( u n ) = 0 and (2.9) give
a ( μ n ) = I μ n ( u n ) = 2 N + 2 [ Ψ ( u n ) + ∫ R N K ( u n ) d x ] ≥ 2 N + 2 Ψ ( u n ) . (21) a(\mu_{n})=I_{\mu_{n}}(u_{n})=\frac{2}{N+2}\left[\Psi(u_{n})+\int_{\mathbb{R}^{N}}K(u_{n})\,dx\right]
\ge\frac{2}{N+2}\Psi(u_{n}). \tag*{(21)} a ( μ n ) = I μ n ( u n ) = N + 2 2 [ Ψ ( u n ) + ∫ R N K ( u n ) d x ] ≥ N + 2 2 Ψ ( u n ) . (21) Lemma 2.3 therefore yields Ψ ( u n ) ≤ C \Psi(u_{n})\le C Ψ ( u n ) ≤ C . Since
1 − 1 − t 2 = ∑ j = 1 ∞ a j t 2 j , a j > 0 , 1-\sqrt{1-t^{2}}=\sum_{j=1}^{\infty}a_{j}t^{2j},\qquad a_{j}>0, 1 − 1 − t 2 = j = 1 ∑ ∞ a j t 2 j , a j > 0 , the sequences ( ∥ ∇ u n ∥ 2 ) (\|\nabla u_{n}\|_{2}) ( ∥∇ u n ∥ 2 ) and ( ∥ ∇ u n ∥ 2 j 0 ) (\|\nabla u_{n}\|_{2j_{0}}) ( ∥∇ u n ∥ 2 j 0 ) are bounded. The Sobolev and Morrey embeddings then imply
∥ u n ∥ 2 ∗ + ∥ u n ∥ ∞ ≤ C . \|u_{n}\|_{2^{*}}+\|u_{n}\|_{\infty}\le C. ∥ u n ∥ 2 ∗ + ∥ u n ∥ ∞ ≤ C . Let R : = sup n ∥ u n ∥ ∞ R:=\sup_{n}\|u_{n}\|_{\infty} R := sup n ∥ u n ∥ ∞ . By (F1), for every ε > 0 \varepsilon>0 ε > 0 there exists C ε > 0 C_{\varepsilon}>0 C ε > 0 such that
F ( s ) ≤ ε s 2 + C ε ∣ s ∣ 2 ∗ for ∣ s ∣ ≤ R . F(s)\le\varepsilon s^{2}+C_{\varepsilon}|s|^{2^{*}}\qquad\text{for }|s|\le R. F ( s ) ≤ ε s 2 + C ε ∣ s ∣ 2 ∗ for ∣ s ∣ ≤ R . Choose μ − > 0 \mu_{-}>0 μ − > 0 such that μ n ≥ μ − \mu_{n}\ge\mu_{-} μ n ≥ μ − for all sufficiently large n n n , and take ε = μ − / 4 \varepsilon=\mu_{-}/4 ε = μ − /4 . Since I μ n ( u n ) = a ( μ n ) I_{\mu_{n}}(u_{n})=a(\mu_{n}) I μ n ( u n ) = a ( μ n ) , we obtain
μ − 4 ∥ u n ∥ 2 2 ≤ a ( μ n ) + C ∥ u n ∥ 2 ∗ 2 ∗ ≤ C . \frac{\mu_{-}}{4}\|u_{n}\|_{2}^{2}\le a(\mu_{n})+C\|u_{n}\|_{2^{*}}^{2^{*}}\le C. 4 μ − ∥ u n ∥ 2 2 ≤ a ( μ n ) + C ∥ u n ∥ 2 ∗ 2 ∗ ≤ C . Thus, ( u n ) (u_{n}) ( u n ) is bounded in X X X . Set
α 0 : = N + 2 2 N μ 0 . \alpha_{0}:=\frac{N+2}{2N}\mu_{0}. α 0 := 2 N N + 2 μ 0 .
Applying the profile decomposition of [6 Proposition 3.3] to a subsequence, we find ( v j ) j ≥ 0 ⊂ X (v^j)_{j\geq0}\subset X ( v j ) j ≥ 0 ⊂ X and translations ( y n j ) n ≥ 1 ⊂ R N (y_n^j)_{n\geq1}\subset\mathbb{R}^N ( y n j ) n ≥ 1 ⊂ R N such that
u n ( ⋅ + y n j ) ⇀ v j in X , u n ( x + y n j ) → v j ( x ) a.e. in R N , lim inf n → ∞ [ Ψ ( u n ) + α 0 ∥ u n ∥ 2 2 ] ≥ ∑ j ≥ 0 [ Ψ ( v j ) + α 0 ∥ v j ∥ 2 2 ] , lim n → ∞ ∫ R N F ( u n ) d x = ∑ j ≥ 0 ∫ R N F ( v j ) d x , lim n → ∞ ∫ R N f ( u n ) u n d x = ∑ j ≥ 0 ∫ R N f ( v j ) v j d x . (22) \begin{aligned}
u_n(\cdot+y_n^j)&\rightharpoonup v^j &&\text{in }X,\\
u_n(x+y_n^j)&\to v^j(x) &&\text{a.e. in }\mathbb{R}^N,\\
\liminf_{n\to\infty}\left[\Psi(u_n)+\alpha_0\lVert u_n\rVert_2^2\right]&\geq\sum_{j\geq0}\left[\Psi(v^j)+\alpha_0\lVert v^j\rVert_2^2\right],\\
\lim_{n\to\infty}\int_{\mathbb{R}^N}F(u_n)\,dx&=\sum_{j\geq0}\int_{\mathbb{R}^N}F(v^j)\,dx,\\
\lim_{n\to\infty}\int_{\mathbb{R}^N}f(u_n)u_n\,dx&=\sum_{j\geq0}\int_{\mathbb{R}^N}f(v^j)v^j\,dx. \tag*{(22)}
\end{aligned} u n ( ⋅ + y n j ) u n ( x + y n j ) n → ∞ lim inf [ Ψ ( u n ) + α 0 ∥ u n ∥ 2 2 ] n → ∞ lim ∫ R N F ( u n ) d x n → ∞ lim ∫ R N f ( u n ) u n d x ⇀ v j → v j ( x ) ≥ j ≥ 0 ∑ [ Ψ ( v j ) + α 0 ∥ v j ∥ 2 2 ] , = j ≥ 0 ∑ ∫ R N F ( v j ) d x , = j ≥ 0 ∑ ∫ R N f ( v j ) v j d x . in X , a.e. in R N , (22)
Every profile v j v^j v j belongs to D D D , since D D D is convex and closed in X X X , hence weakly closed. The same profile decomposition also gives
lim inf n → ∞ Ψ ( u n ) ≥ ∑ j ≥ 0 Ψ ( v j ) . \liminf_{n\to\infty}\Psi(u_n)\geq\sum_{j\geq0}\Psi(v^j). n → ∞ lim inf Ψ ( u n ) ≥ j ≥ 0 ∑ Ψ ( v j ) .
Indeed, the inequality in (22) holds, with the same profiles, for every coefficient ε > 0 \varepsilon>0 ε > 0 in place of α 0 \alpha_0 α 0 , because the profiles are determined by the sequence ( u n ) (u_n) ( u n ) and the proof of [6 Proposition 3.3] does not use the value of the coefficient. Since ∥ u n ∥ 2 2 ≤ C \lVert u_n\rVert_2^2\leq C ∥ u n ∥ 2 2 ≤ C ,
lim inf n → ∞ Ψ ( u n ) + ε C ≥ lim inf n → ∞ [ Ψ ( u n ) + ε ∥ u n ∥ 2 2 ] ≥ ∑ j ≥ 0 [ Ψ ( v j ) + ε ∥ v j ∥ 2 2 ] ≥ ∑ j ≥ 0 Ψ ( v j ) , \liminf_{n\to\infty}\Psi(u_n)+\varepsilon C\geq\liminf_{n\to\infty}\left[\Psi(u_n)+\varepsilon\lVert u_n\rVert_2^2\right]\geq\sum_{j\geq0}\left[\Psi(v^j)+\varepsilon\lVert v^j\rVert_2^2\right]\geq\sum_{j\geq0}\Psi(v^j), n → ∞ lim inf Ψ ( u n ) + εC ≥ n → ∞ lim inf [ Ψ ( u n ) + ε ∥ u n ∥ 2 2 ] ≥ j ≥ 0 ∑ [ Ψ ( v j ) + ε ∥ v j ∥ 2 2 ] ≥ j ≥ 0 ∑ Ψ ( v j ) ,
and letting ε → 0 + \varepsilon\to0^+ ε → 0 + proves the assertion. Since ( u n ) (u_n) ( u n ) is bounded in L 2 L^2 L 2 and μ n → μ 0 \mu_n\to\mu_0 μ n → μ 0 , we have
M μ 0 ( u n ) = M μ n ( u n ) + N + 2 2 N ( μ 0 − μ n ) ∥ u n ∥ 2 2 → 0. M_{\mu_0}(u_n)=M_{\mu_n}(u_n)+\frac{N+2}{2N}(\mu_0-\mu_n)\lVert u_n\rVert_2^2\to0. M μ 0 ( u n ) = M μ n ( u n ) + 2 N N + 2 ( μ 0 − μ n ) ∥ u n ∥ 2 2 → 0.
Combining this fact with (22) , we obtain
0 ≥ ∑ j ≥ 0 M μ 0 ( v j ) . (23) 0\geq\sum_{j\geq0}M_{\mu_0}(v^j). \tag*{(23)} 0 ≥ j ≥ 0 ∑ M μ 0 ( v j ) . (23)
Let J : = { j : v j ≠ 0 } \mathcal{J}:=\{j:v^j\neq0\} J := { j : v j = 0 } . We claim that J ≠ ∅ \mathcal{J}\neq\varnothing J = ∅ . Indeed, if every profile vanished, then the last two identities in (22) would imply
∫ R N F ( u n ) d x → 0 , ∫ R N f ( u n ) u n d x → 0. \int_{\mathbb{R}^N}F(u_n)\,dx\to0,\qquad\int_{\mathbb{R}^N}f(u_n)u_n\,dx\to0. ∫ R N F ( u n ) d x → 0 , ∫ R N f ( u n ) u n d x → 0.
Since M μ 0 ( u n ) → 0 M_{\mu_0}(u_n)\to0 M μ 0 ( u n ) → 0 , it would follow that
Ψ ( u n ) + α 0 ∥ u n ∥ 2 2 → 0 , \Psi(u_n)+\alpha_0\lVert u_n\rVert_2^2\to0, Ψ ( u n ) + α 0 ∥ u n ∥ 2 2 → 0 ,
and then (2.10) would give a ( μ n ) → 0 a(\mu_n)\to0 a ( μ n ) → 0 . This contradicts Lemma 2.3.
By (2.12), there exists j 1 ∈ J j_1\in\mathcal{J} j 1 ∈ J such that M μ 0 ( v j 1 ) ≤ 0 M_{\mu_0}(v^{j_1})\leq0 M μ 0 ( v j 1 ) ≤ 0 . Put
θ 0 : = θ μ 0 ( v j 1 ) . \theta_0:=\theta_{\mu_0}(v^{j_1}). θ 0 := θ μ 0 ( v j 1 ) .
Since φ ( 1 , μ 0 , v j 1 ) = M μ 0 ( v j 1 ) ≤ 0 \varphi(1,\mu_0,v^{j_1})=M_{\mu_0}(v^{j_1})\leq0 φ ( 1 , μ 0 , v j 1 ) = M μ 0 ( v j 1 ) ≤ 0 , (2.2) shows that 0 < θ 0 ≤ 1 0<\theta_0\leq1 0 < θ 0 ≤ 1 . Moreover, (F2) implies
K ( s ) ≥ 0 , K ( θ s ) ≤ K ( s ) for s ∈ R and 0 < θ ≤ 1. K(s)\geq0,\qquad K(\theta s)\leq K(s)\quad\text{for }s\in\mathbb{R}\text{ and }0<\theta\leq1. K ( s ) ≥ 0 , K ( θ s ) ≤ K ( s ) for s ∈ R and 0 < θ ≤ 1.
Define
Q ( u ) : = 2 N + 2 [ Ψ ( u ) + ∫ R N K ( u ) d x ] . Q(u) := \frac{2}{N+2}\left[\Psi(u)+\int_{\mathbb{R}^{N}} K(u)\,dx\right]. Q ( u ) := N + 2 2 [ Ψ ( u ) + ∫ R N K ( u ) d x ] .
For u ∈ M μ u \in\mathcal{M}_{\mu} u ∈ M μ , one has I μ ( u ) = Q ( u ) I_{\mu}(u)=Q(u) I μ ( u ) = Q ( u ) . Consequently,
a ( μ 0 ) ≤ I μ 0 ( ( v j 1 ) θ 0 ) = 2 θ 0 N N + 2 [ Ψ ( v j 1 ) + ∫ R N K ( θ 0 v j 1 ) d x ] ≤ Q ( v j 1 ) . (24) a(\mu_{0}) \le I_{\mu_{0}}\left((v^{j_{1}})_{\theta_{0}}\right)
= \frac{2\theta_{0}^{N}}{N+2}\left[\Psi(v^{j_{1}})+\int_{\mathbb{R}^{N}}K(\theta_{0}v^{j_{1}})\,dx\right] \le Q(v^{j_{1}}). \tag*{(24)} a ( μ 0 ) ≤ I μ 0 ( ( v j 1 ) θ 0 ) = N + 2 2 θ 0 N [ Ψ ( v j 1 ) + ∫ R N K ( θ 0 v j 1 ) d x ] ≤ Q ( v j 1 ) . (24)
After passing to a further subsequence, Lemma 2.3 allows us to assume that a ( μ n ) → L a(\mu_{n}) \to L a ( μ n ) → L . From (2.11) and (2.10), we get
L ≥ ∑ j ∈ J Q ( v j ) ≥ Q ( v j 1 ) ≥ a ( μ 0 ) . (25) L \ge\sum_{j\in\mathcal{J}}Q(v^{j}) \ge Q(v^{j_{1}}) \ge a(\mu_{0}). \tag*{(25)} L ≥ j ∈ J ∑ Q ( v j ) ≥ Q ( v j 1 ) ≥ a ( μ 0 ) . (25)
On the other hand, L ≤ a ( μ 0 ) L \le a(\mu_{0}) L ≤ a ( μ 0 ) by Lemma 2.3. Therefore, equality holds throughout (25) . Equality in (2.13) forces θ 0 = 1 \theta_{0}=1 θ 0 = 1 , because Q ( v j 1 ) > 0 Q(v^{j_{1}})>0 Q ( v j 1 ) > 0 and the last inequality in (2.13) is strict when θ 0 < 1 \theta_{0}<1 θ 0 < 1 . It follows that
v j 1 ∈ M μ 0 , I μ 0 ( v j 1 ) = a ( μ 0 ) , v^{j_{1}}\in\mathcal{M}_{\mu_{0}},\qquad I_{\mu_{0}}(v^{j_{1}})=a(\mu_{0}), v j 1 ∈ M μ 0 , I μ 0 ( v j 1 ) = a ( μ 0 ) ,
and hence v j 1 ∈ C ( μ 0 ) v^{j_{1}}\in C(\mu_{0}) v j 1 ∈ C ( μ 0 ) . Furthermore, Q ( v ) > 0 Q(v)>0 Q ( v ) > 0 for every nonzero v ∈ X v\in X v ∈ X . Thus equality in (2.14) also shows that
# J = 1. \#\mathcal{J}=1. # J = 1.
Let u 0 : = v j 1 u_{0}:=v^{j_{1}} u 0 := v j 1 , y n : = y n j 1 y_{n}:=y_{n}^{j_{1}} y n := y n j 1 and w n : = u n ( ⋅ + y n ) w_{n}:=u_{n}(\cdot+y_{n}) w n := u n ( ⋅ + y n ) . Since there is exactly one nonzero profile, (2.11) gives
∫ R N G ( w n ) d x → ∫ R N G ( u 0 ) d x . \int_{\mathbb{R}^{N}}G(w_{n})\,dx\to\int_{\mathbb{R}^{N}}G(u_{0})\,dx. ∫ R N G ( w n ) d x → ∫ R N G ( u 0 ) d x .
Using M μ n ( w n ) = 0 M_{\mu_{n}}(w_{n})=0 M μ n ( w n ) = 0 , μ n → μ 0 \mu_{n}\to\mu_{0} μ n → μ 0 and M μ 0 ( u 0 ) = 0 M_{\mu_{0}}(u_{0})=0 M μ 0 ( u 0 ) = 0 , we conclude that
Ψ ( w n ) + α 0 ∥ w n ∥ 2 2 → Ψ ( u 0 ) + α 0 ∥ u 0 ∥ 2 2 . \Psi(w_{n})+\alpha_{0}\lVert w_{n}\rVert_{2}^{2}\to\Psi(u_{0})+\alpha_{0}\lVert u_{0}\rVert_{2}^{2}. Ψ ( w n ) + α 0 ∥ w n ∥ 2 2 → Ψ ( u 0 ) + α 0 ∥ u 0 ∥ 2 2 .
Together with w n ⇀ u 0 w_{n}\rightharpoonup u_{0} w n ⇀ u 0 in X X X , [6 Lemma 2.2] now yields w n → u 0 w_{n}\to u_{0} w n → u 0 strongly in X X X . □
Corollary 2.5. For every μ > 0 \mu>0 μ > 0 the infimum and the supremum in (1.9) are attained: there exist u + , u − ∈ C ( μ ) u_{+},u_{-}\in C(\mu) u + , u − ∈ C ( μ ) with 1 2 ∥ u + ∥ 2 2 = c + ( μ ) \frac{1}{2}\lVert u_{+}\rVert_{2}^{2}=c_{+}(\mu) 2 1 ∥ u + ∥ 2 2 = c + ( μ ) and 1 2 ∥ u − ∥ 2 2 = c − ( μ ) \frac{1}{2}\lVert u_{-}\rVert_{2}^{2}=c_{-}(\mu) 2 1 ∥ u − ∥ 2 2 = c − ( μ ) . Moreover, μ ↦ c − ( μ ) \mu\mapsto c_{-}(\mu) μ ↦ c − ( μ ) is lower semicontinuous and μ ↦ c + ( μ ) \mu\mapsto c_{+}(\mu) μ ↦ c + ( μ ) is upper semicontinuous on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) .
Proof. Let ( u n ) ⊂ C ( μ ) (u_{n})\subset C(\mu) ( u n ) ⊂ C ( μ ) be such that 1 2 ∥ u n ∥ 2 2 → c − ( μ ) \frac{1}{2}\lVert u_{n}\rVert_{2}^{2}\to c_{-}(\mu) 2 1 ∥ u n ∥ 2 2 → c − ( μ ) . By Lemma 2.4 with μ n ≡ μ \mu_{n}\equiv\mu μ n ≡ μ , a subsequence of u n ( ⋅ + y n ) u_{n}(\cdot+y_{n}) u n ( ⋅ + y n ) converges in X X X to some u − ∈ C ( μ ) u_{-}\in C(\mu) u − ∈ C ( μ ) , and the translation invariance of the L 2 L^{2} L 2 -norm gives 1 2 ∥ u − ∥ 2 2 = c − ( μ ) \frac{1}{2}\lVert u_{-}\rVert_{2}^{2}=c_{-}(\mu) 2 1 ∥ u − ∥ 2 2 = c − ( μ ) . In particular, c − ( μ ) > 0 c_{-}(\mu)>0 c − ( μ ) > 0 , since 0 ∉ C ( μ ) 0\notin C(\mu) 0 ∈ / C ( μ ) . The same argument, applied to a sequence ( u n ) ⊂ C ( μ ) (u_{n})\subset C(\mu) ( u n ) ⊂ C ( μ ) with 1 2 ∥ u n ∥ 2 2 → c + ( μ ) \frac{1}{2}\lVert u_{n}\rVert_{2}^{2}\to c_{+}(\mu) 2 1 ∥ u n ∥ 2 2 → c + ( μ ) , shows that c + ( μ ) < ∞ c_{+}(\mu)<\infty c + ( μ ) < ∞ and that it is attained. Now let μ n → μ 0 ∈ ( 0 , ∞ ) \mu_{n}\to\mu_{0}\in(0,\infty) μ n → μ 0 ∈ ( 0 , ∞ ) be such that c − ( μ n ) → lim inf μ → μ 0 c − ( μ ) c_{-}(\mu_{n})\to\liminf_{\mu\to\mu_{0}}c_{-}(\mu) c − ( μ n ) → lim inf μ → μ 0 c − ( μ ) , and choose u n ∈ C ( μ n ) u_{n}\in C(\mu_{n}) u n ∈ C ( μ n ) with 1 2 ∥ u n ∥ 2 2 = c − ( μ n ) \frac{1}{2}\lVert u_{n}\rVert_{2}^{2}=c_{-}(\mu_{n}) 2 1 ∥ u n ∥ 2 2 = c − ( μ n ) . By Lemma 2.4, a subsequence of u n ( ⋅ + y n ) u_{n}(\cdot+y_{n}) u n ( ⋅ + y n ) converges in X X X to some u 0 ∈ C ( μ 0 ) u_{0}\in C(\mu_{0}) u 0 ∈ C ( μ 0 ) , hence lim inf μ → μ 0 c − ( μ ) = lim n 1 2 ∥ u n ∥ 2 2 = 1 2 ∥ u 0 ∥ 2 2 ≥ c − ( μ 0 ) \liminf_{\mu\to\mu_{0}}c_{-}(\mu)=\lim_{n}\frac{1}{2}\lVert u_{n}\rVert_{2}^{2}=\frac{1}{2}\lVert u_{0}\rVert_{2}^{2}\ge c_{-}(\mu_{0}) lim inf μ → μ 0 c − ( μ ) = lim n 2 1 ∥ u n ∥ 2 2 = 2 1 ∥ u 0 ∥ 2 2 ≥ c − ( μ 0 ) . The upper semicontinuity of c + c_{+} c + is proved in the same way. □
Proof of Theorem 2.1. Fix μ 0 > 0 \mu_{0}>0 μ 0 > 0 . By Corollary 2.5 we may choose u − , u + ∈ C ( μ 0 ) u_{-},u_{+}\in C(\mu_{0}) u − , u + ∈ C ( μ 0 ) such that
1 2 ∥ u − ∥ 2 2 = c − ( μ 0 ) , 1 2 ∥ u + ∥ 2 2 = c + ( μ 0 ) . \frac{1}{2}\lVert u_{-}\rVert_{2}^{2}=c_{-}(\mu_{0}),\qquad
\frac{1}{2}\lVert u_{+}\rVert_{2}^{2}=c_{+}(\mu_{0}). 2 1 ∥ u − ∥ 2 2 = c − ( μ 0 ) , 2 1 ∥ u + ∥ 2 2 = c + ( μ 0 ) .
For 0 < h < μ 0 0<h<\mu_0 0 < h < μ 0 , inequality (2.5) with ν = μ 0 \nu=\mu_0 ν = μ 0 , applied to ( μ , u ) = ( μ 0 + h , u − ) (\mu,u)=(\mu_0+h,u_-) ( μ , u ) = ( μ 0 + h , u − ) and to ( μ , u ) = ( μ 0 − h , u + ) (\mu,u)=(\mu_0-h,u_+) ( μ , u ) = ( μ 0 − h , u + ) , gives
a ( μ 0 + h ) − a ( μ 0 ) h ≤ 1 2 θ μ 0 + h ( u − ) N + 2 ∥ u − ∥ 2 2 , a ( μ 0 ) − a ( μ 0 − h ) h ≥ 1 2 θ μ 0 − h ( u + ) N + 2 ∥ u + ∥ 2 2 . \frac{a(\mu_0+h)-a(\mu_0)}{h}\leq\frac{1}{2}\theta_{\mu_0+h}(u_-)^{N+2}\lVert u_-\rVert_2^2,\qquad\frac{a(\mu_0)-a(\mu_0-h)}{h}\geq\frac{1}{2}\theta_{\mu_0-h}(u_+)^{N+2}\lVert u_+\rVert_2^2. h a ( μ 0 + h ) − a ( μ 0 ) ≤ 2 1 θ μ 0 + h ( u − ) N + 2 ∥ u − ∥ 2 2 , h a ( μ 0 ) − a ( μ 0 − h ) ≥ 2 1 θ μ 0 − h ( u + ) N + 2 ∥ u + ∥ 2 2 .
Since θ μ 0 + h ( u − ) → θ μ 0 ( u − ) = 1 \theta_{\mu_0+h}(u_-)\to\theta_{\mu_0}(u_-)=1 θ μ 0 + h ( u − ) → θ μ 0 ( u − ) = 1 and θ μ 0 − h ( u + ) → θ μ 0 ( u + ) = 1 \theta_{\mu_0-h}(u_+)\to\theta_{\mu_0}(u_+)=1 θ μ 0 − h ( u + ) → θ μ 0 ( u + ) = 1 as h → 0 + h\to0^+ h → 0 + by Lemma 2.2, we obtain
lim sup h → 0 + a ( μ 0 + h ) − a ( μ 0 ) h ≤ c − ( μ 0 ) , lim inf h → 0 + a ( μ 0 ) − a ( μ 0 − h ) h ≥ c + ( μ 0 ) . (26) \limsup_{h\to0^+}\frac{a(\mu_0+h)-a(\mu_0)}{h}\leq c_-(\mu_0),\qquad\liminf_{h\to0^+}\frac{a(\mu_0)-a(\mu_0-h)}{h}\geq c_+(\mu_0). \tag*{(26)} h → 0 + lim sup h a ( μ 0 + h ) − a ( μ 0 ) ≤ c − ( μ 0 ) , h → 0 + lim inf h a ( μ 0 ) − a ( μ 0 − h ) ≥ c + ( μ 0 ) . (26)
For the reverse inequalities, let ( h n ) ⊂ ( 0 , μ 0 ) (h_n)\subset(0,\mu_0) ( h n ) ⊂ ( 0 , μ 0 ) be an arbitrary sequence such that h n → 0 h_n\to0 h n → 0 , and choose u n ∈ C ( μ 0 + h n ) u_n\in C(\mu_0+h_n) u n ∈ C ( μ 0 + h n ) and w n ∈ C ( μ 0 − h n ) w_n\in C(\mu_0-h_n) w n ∈ C ( μ 0 − h n ) . Inequality (2.5) with μ = μ 0 \mu=\mu_0 μ = μ 0 , applied to ( ν , u ) = ( μ 0 + h n , u n ) (\nu,u)=(\mu_0+h_n,u_n) ( ν , u ) = ( μ 0 + h n , u n ) and to ( ν , u ) = ( μ 0 − h n , w n ) (\nu,u)=(\mu_0-h_n,w_n) ( ν , u ) = ( μ 0 − h n , w n ) , gives
a ( μ 0 + h n ) − a ( μ 0 ) h n ≥ 1 2 θ μ 0 ( u n ) N + 2 ∥ u n ∥ 2 2 , a ( μ 0 ) − a ( μ 0 − h n ) h n ≤ 1 2 θ μ 0 ( w n ) N + 2 ∥ w n ∥ 2 2 . (27) \frac{a(\mu_0+h_n)-a(\mu_0)}{h_n}\geq\frac{1}{2}\theta_{\mu_0}(u_n)^{N+2}\lVert u_n\rVert_2^2,\qquad\frac{a(\mu_0)-a(\mu_0-h_n)}{h_n}\leq\frac{1}{2}\theta_{\mu_0}(w_n)^{N+2}\lVert w_n\rVert_2^2. \tag*{(27)} h n a ( μ 0 + h n ) − a ( μ 0 ) ≥ 2 1 θ μ 0 ( u n ) N + 2 ∥ u n ∥ 2 2 , h n a ( μ 0 ) − a ( μ 0 − h n ) ≤ 2 1 θ μ 0 ( w n ) N + 2 ∥ w n ∥ 2 2 . (27)
By Lemma 2.4, after passing to a subsequence, there exist ( y n ) , ( z n ) ⊂ R N (y_n),(z_n)\subset\mathbb{R}^N ( y n ) , ( z n ) ⊂ R N and u ∗ , w ∗ ∈ C ( μ 0 ) u_*,w_*\in C(\mu_0) u ∗ , w ∗ ∈ C ( μ 0 ) such that
u n ( ⋅ + y n ) → u ∗ , w n ( ⋅ + z n ) → w ∗ strongly in X . u_n(\cdot+y_n)\to u_*,\qquad w_n(\cdot+z_n)\to w_*\qquad\text{strongly in }X. u n ( ⋅ + y n ) → u ∗ , w n ( ⋅ + z n ) → w ∗ strongly in X .
Since θ μ 0 \theta_{\mu_0} θ μ 0 and the L 2 L^2 L 2 -norm are invariant under translations, Lemma 2.2 yields θ μ 0 ( u n ) → θ μ 0 ( u ∗ ) = 1 \theta_{\mu_0}(u_n)\to\theta_{\mu_0}(u_*)=1 θ μ 0 ( u n ) → θ μ 0 ( u ∗ ) = 1 and θ μ 0 ( w n ) → θ μ 0 ( w ∗ ) = 1 \theta_{\mu_0}(w_n)\to\theta_{\mu_0}(w_*)=1 θ μ 0 ( w n ) → θ μ 0 ( w ∗ ) = 1 . Hence, along this subsequence, the right-hand sides of (2.16) converge to 1 2 ∥ u ∗ ∥ 2 2 ≥ c − ( μ 0 ) \frac{1}{2}\lVert u_*\rVert_2^2\geq c_-(\mu_0) 2 1 ∥ u ∗ ∥ 2 2 ≥ c − ( μ 0 ) and to 1 2 ∥ w ∗ ∥ 2 2 ≤ c + ( μ 0 ) \frac{1}{2}\lVert w_*\rVert_2^2\leq c_+(\mu_0) 2 1 ∥ w ∗ ∥ 2 2 ≤ c + ( μ 0 ) , respectively. Since every such sequence ( h n ) (h_n) ( h n ) has a subsequence with these properties, we conclude that
lim inf h → 0 + a ( μ 0 + h ) − a ( μ 0 ) h ≥ c − ( μ 0 ) , lim sup h → 0 + a ( μ 0 ) − a ( μ 0 − h ) h ≤ c + ( μ 0 ) . \liminf_{h\to0^+}\frac{a(\mu_0+h)-a(\mu_0)}{h}\geq c_-(\mu_0),\qquad\limsup_{h\to0^+}\frac{a(\mu_0)-a(\mu_0-h)}{h}\leq c_+(\mu_0). h → 0 + lim inf h a ( μ 0 + h ) − a ( μ 0 ) ≥ c − ( μ 0 ) , h → 0 + lim sup h a ( μ 0 ) − a ( μ 0 − h ) ≤ c + ( μ 0 ) .
Together with (2.15), this proves a + ′ ( μ 0 ) = c − ( μ 0 ) a_+'(\mu_0)=c_-(\mu_0) a + ′ ( μ 0 ) = c − ( μ 0 ) and a − ′ ( μ 0 ) = c + ( μ 0 ) a_-'(\mu_0)=c_+(\mu_0) a − ′ ( μ 0 ) = c + ( μ 0 ) . □
Proofs of the main results
In this section we first derive the behavior of a ( μ ) a(\mu) a ( μ ) and c ± ( μ ) c_{\pm}(\mu) c ± ( μ ) as μ → ∞ \mu\to\infty μ → ∞ and as μ → 0 + \mu\to0^+ μ → 0 + , and then we prove the main results.
Behavior of a ( μ ) a(\mu) a ( μ ) , c ± ( μ ) c_{\pm}(\mu) c ± ( μ ) and b m ( μ ) b_m(\mu) b m ( μ )
Lemma 3.1. Assume (F1)–(F3) and set
κ ( 0 ) : = 0 , κ ( S ) : = max { f ( S ) S , f ( − S ) − S } ( S > 0 ) , σ : = κ − 1 . \kappa(0):=0,\qquad\kappa(S):=\max\left\{\frac{f(S)}{S},\frac{f(-S)}{-S}\right\}\ (S>0),\qquad\sigma:=\kappa^{-1}. κ ( 0 ) := 0 , κ ( S ) := max { S f ( S ) , − S f ( − S ) } ( S > 0 ) , σ := κ − 1 .
Then κ \kappa κ is a continuous, strictly increasing bijection of [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) onto itself and, with d N : = ∣ B 1 ∣ / 2 N + 3 d_N:=\lvert B_1\rvert/2^{N+3} d N := ∣ B 1 ∣ / 2 N + 3 , every μ > 0 \mu>0 μ > 0 and u ∈ M μ u\in M_\mu u ∈ M μ satisfy
∥ u ∥ ∞ ≥ σ ( μ ) , 1 2 ∥ u ∥ 2 2 ≥ d N σ ( μ ) N + 2 , a ( μ ) μ ≥ d N 2 σ ( μ / 2 ) N + 2 . (28) \lVert u\rVert_\infty\geq\sigma(\mu),\qquad\frac{1}{2}\lVert u\rVert_2^2\geq d_N\sigma(\mu)^{N+2},\qquad\frac{a(\mu)}{\mu}\geq\frac{d_N}{2}\sigma(\mu/2)^{N+2}. \tag*{(28)} ∥ u ∥ ∞ ≥ σ ( μ ) , 2 1 ∥ u ∥ 2 2 ≥ d N σ ( μ ) N + 2 , μ a ( μ ) ≥ 2 d N σ ( μ /2 ) N + 2 . (28)
*In particular, c − ( μ ) → ∞ c_-(\mu)\to\infty c − ( μ ) → ∞ and a ( μ ) / μ → ∞ a(\mu)/\mu\to\infty a ( μ ) / μ → ∞ as μ → ∞ \mu\to\infty μ → ∞ . *
Proof. The stated properties of κ \kappa κ follow from (F1)–(F3): κ \kappa κ is continuous and strictly increasing by (F1)–(F2), κ ( S ) → 0 \kappa(S) \to0 κ ( S ) → 0 as S → 0 + S \to0^{+} S → 0 + by (F1), and κ ( S ) → ∞ \kappa(S) \to\infty κ ( S ) → ∞ as S → ∞ S \to\infty S → ∞ by (F3). Let u ∈ M μ u \in\mathcal{M}_{\mu} u ∈ M μ and set A : = ∥ u ∥ ∞ A := \|u\|_{\infty} A := ∥ u ∥ ∞ . Recall that G ( s ) = F ( s ) + 1 N f ( s ) s G(s) = F(s) + \frac{1}{N}f(s)s G ( s ) = F ( s ) + N 1 f ( s ) s and that F ( s ) ≤ 1 2 f ( s ) s F(s) \le\frac{1}{2}f(s)s F ( s ) ≤ 2 1 f ( s ) s by (F2), hence G ( s ) ≤ N + 2 2 N κ ( A ) s 2 G(s) \le\frac{N+2}{2N}\kappa(A)s^{2} G ( s ) ≤ 2 N N + 2 κ ( A ) s 2 for ∣ s ∣ ≤ A |s| \le A ∣ s ∣ ≤ A . Since M μ ( u ) = 0 \mathcal{M}_{\mu}(u) = 0 M μ ( u ) = 0 and Ψ ( u ) ≥ 0 \Psi(u) \ge0 Ψ ( u ) ≥ 0 , we obtain
N + 2 2 N μ ∥ u ∥ 2 2 ≤ Ψ ( u ) + N + 2 2 N μ ∥ u ∥ 2 2 = ∫ R N G ( u ) d x ≤ N + 2 2 N κ ( A ) ∥ u ∥ 2 2 . \frac{N+2}{2N}\mu\|u\|_{2}^{2} \le\Psi(u) + \frac{N+2}{2N}\mu\|u\|_{2}^{2} = \int_{\mathbb{R}^{N}}G(u)\,dx \le\frac{N+2}{2N}\kappa(A)\|u\|_{2}^{2}. 2 N N + 2 μ ∥ u ∥ 2 2 ≤ Ψ ( u ) + 2 N N + 2 μ ∥ u ∥ 2 2 = ∫ R N G ( u ) d x ≤ 2 N N + 2 κ ( A ) ∥ u ∥ 2 2 .
Since u ≠ 0 u \ne0 u = 0 , this gives κ ( A ) ≥ μ \kappa(A) \ge\mu κ ( A ) ≥ μ , hence A ≥ σ ( μ ) A \ge\sigma(\mu) A ≥ σ ( μ ) . Estimate (A.1) now yields
1 2 ∥ u ∥ 2 2 ≥ ∣ B 1 ∣ 2 N + 3 A N + 2 ≥ d N σ ( μ ) N + 2 . \frac{1}{2}\|u\|_{2}^{2} \ge\frac{|B_{1}|}{2N+3}A^{N+2} \ge d_{N}\sigma(\mu)^{N+2}. 2 1 ∥ u ∥ 2 2 ≥ 2 N + 3 ∣ B 1 ∣ A N + 2 ≥ d N σ ( μ ) N + 2 .
Since C ( μ ) ⊂ M μ \mathcal{C}(\mu) \subset\mathcal{M}_{\mu} C ( μ ) ⊂ M μ , we obtain the same lower bound for c − ( μ ) c_{-}(\mu) c − ( μ ) .
Let now w ∈ C ( μ ) w \in\mathcal{C}(\mu) w ∈ C ( μ ) and v : = m μ / 2 ( w ) ∈ M μ / 2 v := m_{\mu/2}(w) \in\mathcal{M}_{\mu/2} v := m μ /2 ( w ) ∈ M μ /2 , so that ∥ v ∥ 2 2 = θ μ / 2 ( w ) N + 2 ∥ w ∥ 2 2 \|v\|_{2}^{2} = \theta_{\mu/2}(w)^{N+2}\|w\|_{2}^{2} ∥ v ∥ 2 2 = θ μ /2 ( w ) N + 2 ∥ w ∥ 2 2 . Inequality (2.5), with μ / 2 \mu/2 μ /2 and μ \mu μ in place of μ \mu μ and ν \nu ν , and the second inequality in (3.1), applied to v v v , give
a ( μ ) ≥ a ( μ / 2 ) + μ 4 ∥ v ∥ 2 2 ≥ a ( μ / 2 ) + d N μ 2 σ ( μ / 2 ) N + 2 . a(\mu) \ge a(\mu/2) + \frac{\mu}{4}\|v\|_{2}^{2} \ge a(\mu/2) + \frac{d_{N}\mu}{2}\sigma(\mu/2)^{N+2}. a ( μ ) ≥ a ( μ /2 ) + 4 μ ∥ v ∥ 2 2 ≥ a ( μ /2 ) + 2 d N μ σ ( μ /2 ) N + 2 .
Since a ( μ / 2 ) > 0 a(\mu/2) > 0 a ( μ /2 ) > 0 , the last inequality in (3.1) follows. Finally, σ ( μ ) → ∞ \sigma(\mu) \to\infty σ ( μ ) → ∞ as μ → ∞ \mu\to\infty μ → ∞ , which gives both limits. □ \square □
We set
β : = 2 p − 2 − N 2 . \beta:= \frac{2}{p-2} - \frac{N}{2}. β := p − 2 2 − 2 N .
and we note that β + 1 > 0 \beta+ 1 > 0 β + 1 > 0 if and only if p < 2 ∗ p < 2^{*} p < 2 ∗ , and that β < 0 \beta< 0 β < 0 if and only if p > 2 + 4 N p > 2 + \frac{4}{N} p > 2 + N 4 .
Lemma 3.2. Assume (F1)–(F4). Then there exist C , c > 0 C,c > 0 C , c > 0 such that for sufficiently small μ > 0 \mu> 0 μ > 0 and every u ∈ C ( μ ) u \in\mathcal{C}(\mu) u ∈ C ( μ )
a ( μ ) ≤ C μ β + 1 , (29) a(\mu) \le C\mu^{\beta+1}, \tag*{(29)} a ( μ ) ≤ C μ β + 1 , (29) and, if p ≥ 2 + 4 N p \ge2 + \frac{4}{N} p ≥ 2 + N 4 ,
1 2 ∥ u ∥ 2 2 ≥ c μ β , a ( μ ) ≥ c μ β + 1 . (30) \frac{1}{2}\|u\|_{2}^{2} \ge c\mu^{\beta}, \qquad a(\mu) \ge c\mu^{\beta+1}. \tag*{(30)} 2 1 ∥ u ∥ 2 2 ≥ c μ β , a ( μ ) ≥ c μ β + 1 . (30) Consequently,
(i) a ( μ ) → 0 a(\mu) \to0 a ( μ ) → 0 as μ → 0 + \mu\to0^{+} μ → 0 + ;
(ii) if 2 + 4 N < p < 2 ∗ 2 + \frac{4}{N} < p < 2^{*} 2 + N 4 < p < 2 ∗ , then a ( μ ) / μ → ∞ a(\mu)/\mu\to\infty a ( μ ) / μ → ∞ and c − ( μ ) → ∞ c_{-}(\mu) \to\infty c − ( μ ) → ∞ as μ → 0 + \mu\to0^{+} μ → 0 + ;
(iii) if 2 < p < 2 + 4 N 2 < p < 2 + \frac{4}{N} 2 < p < 2 + N 4 , then a ( μ ) / μ → 0 a(\mu)/\mu\to0 a ( μ ) / μ → 0 as μ → 0 + \mu\to0^{+} μ → 0 + .
Proof. Upper bound for a ( μ ) a(\mu) a ( μ ) . Fix a nonzero function ω ∈ C 0 ∞ ( R N ) \omega\in C_{0}^{\infty}(\mathbb{R}^{N}) ω ∈ C 0 ∞ ( R N ) with ω ≥ 0 \omega\ge0 ω ≥ 0 . For μ > 0 \mu> 0 μ > 0 put
w μ ( x ) : = μ 1 p − 2 ω ( μ 1 / 2 x ) . w_{\mu}(x) := \mu^{\frac{1}{p-2}}\omega(\mu^{1/2}x). w μ ( x ) := μ p − 2 1 ω ( μ 1/2 x ) .
Then ∥ ∇ w μ ∥ ∞ = μ 1 p − 2 + 1 2 ∥ ∇ ω ∥ ∞ ≤ 1 \|\nabla w_{\mu}\|_{\infty} = \mu^{\frac{1}{p-2}+\frac{1}{2}}\|\nabla\omega\|_{\infty} \le1 ∥∇ w μ ∥ ∞ = μ p − 2 1 + 2 1 ∥∇ ω ∥ ∞ ≤ 1 for small μ \mu μ , so w μ ∈ D w_{\mu} \in\mathcal{D} w μ ∈ D , and a direct computation gives
∥ ∇ w μ ∥ 2 2 = μ β + 1 ∥ ∇ ω ∥ 2 2 , ∥ w μ ∥ 2 2 = μ β ∥ ω ∥ 2 2 , ∫ R N w μ p d x = μ β + 1 ∫ R N ω p d x , ∥ w μ ∥ ∞ = μ 1 p − 2 ∥ ω ∥ ∞ . (31) \|\nabla w_{\mu}\|_{2}^{2} = \mu^{\beta+1}\|\nabla\omega\|_{2}^{2}, \qquad\|w_{\mu}\|_{2}^{2} = \mu^{\beta}\|\omega\|_{2}^{2}, \qquad\int_{\mathbb{R}^{N}}w_{\mu}^{p}\,dx = \mu^{\beta+1}\int_{\mathbb{R}^{N}}\omega^{p}\,dx, \qquad\|w_{\mu}\|_{\infty} = \mu^{\frac{1}{p-2}}\|\omega\|_{\infty}. \tag*{(31)} ∥∇ w μ ∥ 2 2 = μ β + 1 ∥∇ ω ∥ 2 2 , ∥ w μ ∥ 2 2 = μ β ∥ ω ∥ 2 2 , ∫ R N w μ p d x = μ β + 1 ∫ R N ω p d x , ∥ w μ ∥ ∞ = μ p − 2 1 ∥ ω ∥ ∞ . (31)
By (F4), there are 0 < c 0 ≤ C 0 0<c_0\le C_0 0 < c 0 ≤ C 0 and s 0 > 0 s_0>0 s 0 > 0 such that c 0 ∣ s ∣ p − 2 ≤ f ( s ) / s ≤ C 0 ∣ s ∣ p − 2 c_0|s|^{p-2}\le f(s)/s\le C_0|s|^{p-2} c 0 ∣ s ∣ p − 2 ≤ f ( s ) / s ≤ C 0 ∣ s ∣ p − 2 for 0 < ∣ s ∣ ≤ s 0 0<|s|\le s_0 0 < ∣ s ∣ ≤ s 0 . By integration, F ( s ) ≥ c 0 p ∣ s ∣ p F(s)\ge\frac{c_0}{p}|s|^p F ( s ) ≥ p c 0 ∣ s ∣ p for ∣ s ∣ ≤ s 0 |s|\le s_0 ∣ s ∣ ≤ s 0 . Using Ψ ( v ) ≤ ∥ ∇ v ∥ 2 2 \Psi(v)\le\|\nabla v\|_2^2 Ψ ( v ) ≤ ∥∇ v ∥ 2 2 on D \mathcal{D} D and (31) , for 0 < θ ≤ s 0 / ∥ w μ ∥ ∞ 0<\theta\le s_0/\|w_\mu\|_\infty 0 < θ ≤ s 0 /∥ w μ ∥ ∞ we get
I μ ( ( w μ ) θ ) ≤ θ N ( ∥ ∇ w μ ∥ 2 2 + μ θ 2 2 ∥ w μ ∥ 2 2 − c 0 θ p p ∫ R N w μ p d x ) = μ β + 1 θ N ( ∥ ∇ ω ∥ 2 2 + θ 2 2 ∥ ω ∥ 2 2 − c 0 θ p p ∫ R N ω p d x ) . I_\mu((w_\mu)_\theta)\le\theta^N\left(\|\nabla w_\mu\|_2^2+\frac{\mu\theta^2}{2}\|w_\mu\|_2^2-\frac{c_0\theta^p}{p}\int_{\mathbb{R}^N}w_\mu^p\,dx\right)
=\mu^{\beta+1}\theta^N\left(\|\nabla\omega\|_2^2+\frac{\theta^2}{2}\|\omega\|_2^2-\frac{c_0\theta^p}{p}\int_{\mathbb{R}^N}\omega^p\,dx\right). I μ (( w μ ) θ ) ≤ θ N ( ∥∇ w μ ∥ 2 2 + 2 μ θ 2 ∥ w μ ∥ 2 2 − p c 0 θ p ∫ R N w μ p d x ) = μ β + 1 θ N ( ∥∇ ω ∥ 2 2 + 2 θ 2 ∥ ω ∥ 2 2 − p c 0 θ p ∫ R N ω p d x ) .
Since p > 2 p>2 p > 2 , there is θ 1 > 0 \theta_1>0 θ 1 > 0 , independent of μ \mu μ , such that the last bracket is negative for θ ≥ θ 1 \theta\ge\theta_1 θ ≥ θ 1 . Moreover, by (31) we have θ 1 ≤ s 0 / ∥ w μ ∥ ∞ \theta_1\le s_0/\|w_\mu\|_\infty θ 1 ≤ s 0 /∥ w μ ∥ ∞ for μ \mu μ small, and therefore I μ ( ( w μ ) θ 1 ) ≤ 0 I_\mu((w_\mu)_{\theta_1})\le0 I μ (( w μ ) θ 1 ) ≤ 0 . Since I μ ( ( w μ ) θ ) > 0 I_\mu((w_\mu)_\theta)>0 I μ (( w μ ) θ ) > 0 for 0 < θ ≤ θ μ ( w μ ) 0<\theta\le\theta_\mu(w_\mu) 0 < θ ≤ θ μ ( w μ ) , see § 2 , we get θ μ ( w μ ) < θ 1 \theta_\mu(w_\mu)<\theta_1 θ μ ( w μ ) < θ 1 and
a ( μ ) ≤ Λ ( μ , w μ ) = max 0 < θ ≤ θ 1 I μ ( ( w μ ) θ ) ≤ μ β + 1 θ 1 N ( ∥ ∇ ω ∥ 2 2 + θ 1 2 2 ∥ ω ∥ 2 2 ) = C μ β + 1 . a(\mu)\le\Lambda(\mu,w_\mu)=\max_{0<\theta\le\theta_1}I_\mu((w_\mu)_\theta)\le\mu^{\beta+1}\theta_1^N\left(\|\nabla\omega\|_2^2+\frac{\theta_1^2}{2}\|\omega\|_2^2\right)=C\mu^{\beta+1}. a ( μ ) ≤ Λ ( μ , w μ ) = 0 < θ ≤ θ 1 max I μ (( w μ ) θ ) ≤ μ β + 1 θ 1 N ( ∥∇ ω ∥ 2 2 + 2 θ 1 2 ∥ ω ∥ 2 2 ) = C μ β + 1 .
Lower bound for a ( μ ) a(\mu) a ( μ ) and the mass when p ≥ 2 + 4 N p\ge2+\frac{4}{N} p ≥ 2 + N 4
Let u ∈ C ( μ ) u\in\mathcal{C}(\mu) u ∈ C ( μ ) . By (29) , (21) and (19) ,
C μ β + 1 ≥ a ( μ ) ≥ 2 N + 2 Ψ ( u ) ≥ 1 N + 2 ∥ ∇ u ∥ 2 2 . (32) C\mu^{\beta+1}\ge a(\mu)\ge\frac{2}{N+2}\Psi(u)\ge\frac{1}{N+2}\|\nabla u\|_2^2. \tag*{(32)} C μ β + 1 ≥ a ( μ ) ≥ N + 2 2 Ψ ( u ) ≥ N + 2 1 ∥∇ u ∥ 2 2 . (32)
Recall that G ( s ) : = F ( s ) + 1 N f ( s ) s G(s):=F(s)+\frac{1}{N}f(s)s G ( s ) := F ( s ) + N 1 f ( s ) s . The upper bound in (F4) gives G ( s ) ≤ C ∣ s ∣ p G(s)\le C|s|^p G ( s ) ≤ C ∣ s ∣ p for ∣ s ∣ ≤ s 0 |s|\le s_0 ∣ s ∣ ≤ s 0 . By (32) and (A.1), we have ∥ u ∥ ∞ ≤ s 0 \|u\|_\infty\le s_0 ∥ u ∥ ∞ ≤ s 0 for μ \mu μ small enough. Writing α : = N ( p − 2 ) 4 \alpha:=\frac{N(p-2)}{4} α := 4 N ( p − 2 ) and γ : = p 2 − α \gamma:=\frac{p}{2}-\alpha γ := 2 p − α , the identity M μ ( u ) = 0 M_\mu(u)=0 M μ ( u ) = 0 , (19) and the Gagliardo–Nirenberg inequality ∥ u ∥ p p ≤ C ∥ ∇ u ∥ 2 2 α ∥ u ∥ 2 2 γ \|u\|_p^p\le C\|\nabla u\|_2^{2\alpha}\|u\|_2^{2\gamma} ∥ u ∥ p p ≤ C ∥∇ u ∥ 2 2 α ∥ u ∥ 2 2 γ give
1 2 ∥ ∇ u ∥ 2 2 + N + 2 2 N μ ∥ u ∥ 2 2 ≤ Ψ ( u ) + N + 2 2 N μ ∥ u ∥ 2 2 = ∫ R N G ( u ) d x ≤ C ∫ R N ∣ u ∣ p d x ≤ C ∥ ∇ u ∥ 2 2 α ∥ u ∥ 2 2 γ . (33) \frac{1}{2}\|\nabla u\|_2^2+\frac{N+2}{2N}\mu\|u\|_2^2\le\Psi(u)+\frac{N+2}{2N}\mu\|u\|_2^2=\int_{\mathbb{R}^N}G(u)\,dx\le C\int_{\mathbb{R}^N}|u|^p\,dx\le C\|\nabla u\|_2^{2\alpha}\|u\|_2^{2\gamma}. \tag*{(33)} 2 1 ∥∇ u ∥ 2 2 + 2 N N + 2 μ ∥ u ∥ 2 2 ≤ Ψ ( u ) + 2 N N + 2 μ ∥ u ∥ 2 2 = ∫ R N G ( u ) d x ≤ C ∫ R N ∣ u ∣ p d x ≤ C ∥∇ u ∥ 2 2 α ∥ u ∥ 2 2 γ . (33)
If p ≥ 2 + 4 N p\ge2+\frac{4}{N} p ≥ 2 + N 4 , then α ≥ 1 \alpha\ge1 α ≥ 1 and γ > 0 \gamma>0 γ > 0 since p < 2 ∗ p<2^* p < 2 ∗ . Therefore, (32) and (33) give
∥ u ∥ 2 2 γ ≥ c ∥ ∇ u ∥ 2 2 ( 1 − α ) ≥ c ( C μ β + 1 ) 1 − α , \|u\|_2^{2\gamma}\ge c\|\nabla u\|_2^{2(1-\alpha)}\ge c(C\mu^{\beta+1})^{1-\alpha}, ∥ u ∥ 2 2 γ ≥ c ∥∇ u ∥ 2 2 ( 1 − α ) ≥ c ( C μ β + 1 ) 1 − α ,
which gives the first inequality in (30) , because ( β + 1 ) ( 1 − α ) / γ = β (\beta+1)(1-\alpha)/\gamma=\beta ( β + 1 ) ( 1 − α ) / γ = β . In particular, when p = 2 + 4 / N p=2+4/N p = 2 + 4/ N , one has α = 1 \alpha=1 α = 1 and β = 0 \beta=0 β = 0 , and this estimate gives a uniform positive lower bound for the mass.
Recall that K ( s ) : = 1 2 f ( s ) s − F ( s ) K(s):=\frac{1}{2}f(s)s-F(s) K ( s ) := 2 1 f ( s ) s − F ( s ) . We claim that K ( s ) ≥ c ∣ s ∣ p K(s)\ge c|s|^p K ( s ) ≥ c ∣ s ∣ p for ∣ s ∣ ≤ s 0 |s|\le s_0 ∣ s ∣ ≤ s 0 . For ϵ ∈ { − 1 , 1 } \epsilon\in\{-1,1\} ϵ ∈ { − 1 , 1 } let q ϵ ( r ) : = ϵ f ( ϵ r ) / r q_\epsilon(r):=\epsilon f(\epsilon r)/r q ϵ ( r ) := ϵ f ( ϵr ) / r , r > 0 r>0 r > 0 , which is increasing by (F2), and put ζ : = ( c 0 / ( 2 C 0 ) ) 1 / ( p − 2 ) ∈ ( 0 , 1 ) \zeta:=(c_0/(2C_0))^{1/(p-2)}\in(0,1) ζ := ( c 0 / ( 2 C 0 ) ) 1/ ( p − 2 ) ∈ ( 0 , 1 ) . For 0 < r ≤ s 0 0<r\le s_0 0 < r ≤ s 0 we have
K ( ϵ r ) = ∫ 0 r t ( q ϵ ( r ) − q ϵ ( t ) ) d t ≥ ∫ 0 ζ r t ( c 0 r p − 2 − C 0 ( ζ r ) p − 2 ) d t = c 0 ζ 2 4 r p , K(\epsilon r)=\int_0^r t\bigl(q_\epsilon(r)-q_\epsilon(t)\bigr)\,dt
\ge\int_0^{\zeta r}t\bigl(c_0r^{p-2}-C_0(\zeta r)^{p-2}\bigr)\,dt=\frac{c_0\zeta^2}{4}r^p, K ( ϵr ) = ∫ 0 r t ( q ϵ ( r ) − q ϵ ( t ) ) d t ≥ ∫ 0 ζ r t ( c 0 r p − 2 − C 0 ( ζ r ) p − 2 ) d t = 4 c 0 ζ 2 r p ,
which proves the claim. Since ∥ u ∥ ∞ ≤ s 0 \|u\|_\infty\le s_0 ∥ u ∥ ∞ ≤ s 0 , the energy identity (21) , (33) and the first inequality in (30) yield
a ( μ ) ≥ 2 N + 2 ∫ R N K ( u ) d x ≥ c ∫ R N ∣ u ∣ p d x ≥ c ∫ R N G ( u ) d x ≥ c μ ∥ u ∥ 2 2 ≥ c μ β + 1 , a(\mu)\ge\frac{2}{N+2}\int_{\mathbb{R}^N}K(u)\,dx\ge c\int_{\mathbb{R}^N}|u|^p\,dx\ge c\int_{\mathbb{R}^N}G(u)\,dx\ge c\mu\|u\|_2^2\ge c\mu^{\beta+1}, a ( μ ) ≥ N + 2 2 ∫ R N K ( u ) d x ≥ c ∫ R N ∣ u ∣ p d x ≥ c ∫ R N G ( u ) d x ≥ c μ ∥ u ∥ 2 2 ≥ c μ β + 1 ,
which is the second inequality in (30) . In the critical case p = 2 + 4 / N p=2+4/N p = 2 + 4/ N , the upper and lower bounds therefore give c μ ≤ a ( μ ) ≤ C μ c\mu\le a(\mu)\le C\mu c μ ≤ a ( μ ) ≤ C μ for all sufficiently small μ > 0 \mu>0 μ > 0 .
Since β + 1 > 0 \beta+1>0 β + 1 > 0 if and only if p < 2 ∗ p<2^* p < 2 ∗ , and β < 0 \beta<0 β < 0 if and only if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 , assertions (i)–(iii) follow from (29) and (30) . □
Recall that b m ( μ ) = a ( μ ) − μ m 2 b_m(\mu)=a(\mu)-\mu m^2 b m ( μ ) = a ( μ ) − μ m 2 , see (8) . The next two lemmas collect the properties of b m b_m b m , c − c_- c − and a a a which are used in the proofs of the main results.
Lemma 3.3. Assume (F1)–(F4) and let m > 0 m>0 m > 0 . Then b m b_m b m is continuous on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) ,
lim μ → 0 + b m ( μ ) = 0 and lim μ → ∞ b m ( μ ) = + ∞ . (34) \lim_{\mu\to0^+} b_m(\mu)=0 \qquad\text{and}\qquad\lim_{\mu\to\infty} b_m(\mu)=+\infty. \tag*{(34)} μ → 0 + lim b m ( μ ) = 0 and μ → ∞ lim b m ( μ ) = + ∞. (34) In particular, inf μ > 0 b m ( μ ) ≤ 0 \inf_{\mu>0}b_m(\mu)\le0 inf μ > 0 b m ( μ ) ≤ 0 , and this infimum is attained whenever it is negative. Moreover, for every μ > 0 \mu>0 μ > 0 the following assertions hold.
(i) If μ \mu μ is a local minimum point of b m b_m b m , then c − ( μ ) = c + ( μ ) = m 2 c_-(\mu)=c_+(\mu)=m^2 c − ( μ ) = c + ( μ ) = m 2 , that is, C ( μ ) ⊂ S m C(\mu)\subset S_m C ( μ ) ⊂ S m .
(ii) If u ∈ C ( μ ) ∩ S m u\in C(\mu)\cap S_m u ∈ C ( μ ) ∩ S m , then ( μ , u ) (\mu,u) ( μ , u ) is a solution of (1) with E ( u ) = b m ( μ ) E(u)=b_m(\mu) E ( u ) = b m ( μ ) . Moreover, u u u has constant sign, it is radially symmetric about some point x 0 ∈ R N x_0\in\mathbb{R}^N x 0 ∈ R N , and u ( ⋅ + x 0 ) ∈ C ( μ ) ∩ S m u(\cdot+x_0)\in C(\mu)\cap S_m u ( ⋅ + x 0 ) ∈ C ( μ ) ∩ S m . If f f f is odd, then also − u ∈ C ( μ ) ∩ S m -u\in C(\mu)\cap S_m − u ∈ C ( μ ) ∩ S m .
Proof. By Theorem 2.1 , a a a has finite one-sided derivatives at every point, hence a a a and b m b_m b m are continuous on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Since b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) b_m(\mu)=\mu(a(\mu)/\mu-m^2) b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) , the limits in (34) follow from a ( μ ) → 0 a(\mu)\to0 a ( μ ) → 0 as μ → 0 + \mu\to0^+ μ → 0 + (Lemma 3.2 ) and from a ( μ ) / μ → ∞ a(\mu)/\mu\to\infty a ( μ ) / μ → ∞ as μ → ∞ \mu\to\infty μ → ∞ (Lemma 3.1). Hence inf μ > 0 b m ( μ ) ≤ 0 \inf_{\mu>0}b_m(\mu)\le0 inf μ > 0 b m ( μ ) ≤ 0 , and if this infimum is negative, then by (34) it coincides with the infimum of b m b_m b m over a compact subinterval of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , where it is attained.
(i) Since μ \mu μ is a local minimum point of b m b_m b m , we have b m ( μ + h ) ≥ b m ( μ ) b_m(\mu+h)\ge b_m(\mu) b m ( μ + h ) ≥ b m ( μ ) and b m ( μ − h ) ≥ b m ( μ ) b_m(\mu-h)\ge b_m(\mu) b m ( μ − h ) ≥ b m ( μ ) for every sufficiently small h > 0 h>0 h > 0 , that is,
a ( μ + h ) − a ( μ ) h ≥ m 2 ≥ a ( μ ) − a ( μ − h ) h . \frac{a(\mu+h)-a(\mu)}{h}\ge m^2\ge\frac{a(\mu)-a(\mu-h)}{h}. h a ( μ + h ) − a ( μ ) ≥ m 2 ≥ h a ( μ ) − a ( μ − h ) .
Hence (26) , with μ 0 = μ \mu_0=\mu μ 0 = μ , gives c − ( μ ) ≥ m 2 ≥ c + ( μ ) c_-(\mu)\ge m^2\ge c_+(\mu) c − ( μ ) ≥ m 2 ≥ c + ( μ ) . Since c − ( μ ) ≤ c + ( μ ) c_-(\mu)\le c_+(\mu) c − ( μ ) ≤ c + ( μ ) , we conclude that c − ( μ ) = c + ( μ ) = m 2 c_-(\mu)=c_+(\mu)=m^2 c − ( μ ) = c + ( μ ) = m 2 .
(ii) Recall from Subsection 1.1 that u u u is a critical point of I μ I_\mu I μ . Since u ∈ S m u\in S_m u ∈ S m , the pair ( μ , u ) (\mu,u) ( μ , u ) is a solution of (1) , and
E ( u ) = I μ ( u ) − μ 2 ∥ u ∥ 2 2 = a ( μ ) − μ m 2 = b m ( μ ) . E(u)=I_\mu(u)-\frac{\mu}{2}\lVert u\rVert_2^2=a(\mu)-\mu m^2=b_m(\mu). E ( u ) = I μ ( u ) − 2 μ ∥ u ∥ 2 2 = a ( μ ) − μ m 2 = b m ( μ ) .
The sign and the symmetry of u u u are given by Theorem B.5 . Finally, if f f f is odd, then I μ I_\mu I μ and M μ M_\mu M μ are even, hence − u ∈ C ( μ ) ∩ S m -u\in C(\mu)\cap S_m − u ∈ C ( μ ) ∩ S m . □
Lemma 3.4. Assume (F1)–(F4) with 2 + 4 N ≤ p < 2 ∗ 2+\frac{4}{N}\le p<2^* 2 + N 4 ≤ p < 2 ∗ , and let
c ∗ = inf μ > 0 c − ( μ ) , r e = inf μ > 0 a ( μ ) μ c_*=\inf_{\mu>0}c_-(\mu),\qquad r_e=\inf_{\mu>0}\frac{a(\mu)}{\mu} c ∗ = μ > 0 inf c − ( μ ) , r e = μ > 0 inf μ a ( μ ) be as in Theorems 1.1 and 1.3 . Then c ∗ > 0 c_*>0 c ∗ > 0 , and the function μ ↦ a ( μ ) − c ∗ μ \mu\mapsto a(\mu)-c_*\mu μ ↦ a ( μ ) − c ∗ μ is nonnegative and nondecreasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . In particular, c ∗ ≤ r e < ∞ c_*\le r_e<\infty c ∗ ≤ r e < ∞ . Moreover, if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 , then c ∗ c_* c ∗ and r e r_e r e are attained, and a ( μ ) > c ∗ μ a(\mu)>c_*\mu a ( μ ) > c ∗ μ for every μ > 0 \mu>0 μ > 0 . In particular, c ∗ < r e c_*<r_e c ∗ < r e in this case.
Proof. Since β ≤ 0 \beta\le0 β ≤ 0 , by (30) there are c , δ > 0 c,\delta>0 c , δ > 0 such that c − ( μ ) ≥ c μ β ≥ c δ β c_-(\mu)\ge c\mu^\beta\ge c\delta^\beta c − ( μ ) ≥ c μ β ≥ c δ β for 0 < μ ≤ δ 0<\mu\le\delta 0 < μ ≤ δ . Moreover, (28) and the monotonicity of σ \sigma σ give c − ( μ ) ≥ d N σ ( δ ) N + 2 c_-(\mu)\ge d_N\sigma(\delta)^{N+2} c − ( μ ) ≥ d N σ ( δ ) N + 2 for μ ≥ δ \mu\ge\delta μ ≥ δ . Hence c ∗ > 0 c_*>0 c ∗ > 0 . The function g ( μ ) : = a ( μ ) − c ∗ μ g(\mu):=a(\mu)-c_*\mu g ( μ ) := a ( μ ) − c ∗ μ is continuous on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) by Lemma 3.3 , and the second inequality in (26) , together with c + ( μ ) ≥ c − ( μ ) ≥ c ∗ c_+(\mu)\ge c_-(\mu)\ge c_* c + ( μ ) ≥ c − ( μ ) ≥ c ∗ , gives
lim inf h → 0 + g ( μ ) − g ( μ − h ) h ≥ c + ( μ ) − c ∗ ≥ 0 for every μ > 0. \liminf_{h\to0^+}\frac{g(\mu)-g(\mu-h)}{h}\ge c_+(\mu)-c_*\ge0\qquad\text{for every }\mu>0. h → 0 + lim inf h g ( μ ) − g ( μ − h ) ≥ c + ( μ ) − c ∗ ≥ 0 for every μ > 0.
Therefore g g g is nondecreasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Indeed, otherwise there are ε > 0 \varepsilon>0 ε > 0 and 0 < μ 1 < μ 2 0<\mu_1<\mu_2 0 < μ 1 < μ 2 such that g ε ( μ ) : = g ( μ ) + ε μ g_\varepsilon(\mu):=g(\mu)+\varepsilon\mu g ε ( μ ) := g ( μ ) + ε μ satisfies g ε ( μ 1 ) > g ε ( μ 2 ) g_\varepsilon(\mu_1)>g_\varepsilon(\mu_2) g ε ( μ 1 ) > g ε ( μ 2 ) . Then μ 3 : = min { μ ∈ [ μ 1 , μ 2 ] : g ε ( μ ) ≤ g ε ( μ 2 ) } > μ 1 \mu_3:=\min\{\mu\in[\mu_1,\mu_2]:g_\varepsilon(\mu)\leq g_\varepsilon(\mu_2)\}>\mu_1 μ 3 := min { μ ∈ [ μ 1 , μ 2 ] : g ε ( μ ) ≤ g ε ( μ 2 )} > μ 1 , and g ε ( μ 3 − h ) > g ε ( μ 3 ) g_\varepsilon(\mu_3-h)>g_\varepsilon(\mu_3) g ε ( μ 3 − h ) > g ε ( μ 3 ) for 0 < h ≤ μ 3 − μ 1 0<h\leq\mu_3-\mu_1 0 < h ≤ μ 3 − μ 1 , which is impossible, because the lower limit of the left difference quotients of g ε g_\varepsilon g ε at μ 3 \mu_3 μ 3 is not smaller than ε \varepsilon ε . Since g ( μ ) → 0 g(\mu)\to0 g ( μ ) → 0 as μ → 0 + \mu\to0^+ μ → 0 + by Lemma 3.2, g g g is also nonnegative. Thus a ( μ ) / μ ≥ c ∗ a(\mu)/\mu\geq c_* a ( μ ) / μ ≥ c ∗ for every μ > 0 \mu>0 μ > 0 , and c ∗ ≤ r e < ∞ c_*\leq r_e<\infty c ∗ ≤ r e < ∞ .
Let now p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 . By Lemmas 3.1 and 3.2, c − ( μ ) → ∞ c_-(\mu)\to\infty c − ( μ ) → ∞ and a ( μ ) / μ → ∞ a(\mu)/\mu\to\infty a ( μ ) / μ → ∞ as μ → 0 + \mu\to0^+ μ → 0 + and as μ → ∞ \mu\to\infty μ → ∞ . Since c − c_- c − is lower semicontinuous by Corollary 2.5 and μ ↦ a ( μ ) / μ \mu\mapsto a(\mu)/\mu μ ↦ a ( μ ) / μ is continuous, both infima are attained. Moreover, a ( μ ) − c ∗ μ = μ ( a ( μ ) / μ − c ∗ ) > 0 a(\mu)-c_*\mu=\mu(a(\mu)/\mu-c_*)>0 a ( μ ) − c ∗ μ = μ ( a ( μ ) / μ − c ∗ ) > 0 for μ \mu μ small, and the monotonicity of μ ↦ a ( μ ) − c ∗ μ \mu\mapsto a(\mu)-c_*\mu μ ↦ a ( μ ) − c ∗ μ gives a ( μ ) > c ∗ μ a(\mu)>c_*\mu a ( μ ) > c ∗ μ for every μ > 0 \mu>0 μ > 0 . In particular, r e = min μ > 0 a ( μ ) / μ > c ∗ r_e=\min_{\mu>0}a(\mu)/\mu>c_* r e = min μ > 0 a ( μ ) / μ > c ∗ .
Proofs of Theorems 1.1–1.3
Proof of Theorem 1.1. By Lemma 3.4, c ∗ > 0 c_*>0 c ∗ > 0 , and c ∗ c_* c ∗ is attained if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 .
(i) If u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) has 1 2 ∥ u ∥ 2 2 = m 2 \frac{1}{2}\lVert u\rVert_2^2=m^2 2 1 ∥ u ∥ 2 2 = m 2 , then m 2 ≥ c − ( μ ) ≥ c ∗ m^2\geq c_-(\mu)\geq c_* m 2 ≥ c − ( μ ) ≥ c ∗ , which excludes m 2 < c ∗ m^2<c_* m 2 < c ∗ . If c ∗ c_* c ∗ is not attained, then c − ( μ ) > c ∗ c_-(\mu)>c_* c − ( μ ) > c ∗ for every μ > 0 \mu>0 μ > 0 , and the same inequality excludes m 2 = c ∗ m^2=c_* m 2 = c ∗ .
(ii) Let m 2 = c ∗ = c − ( μ 0 ) m^2=c_*=c_-(\mu_0) m 2 = c ∗ = c − ( μ 0 ) . By Corollary 2.5, there is u ∈ C ( μ 0 ) u\in C(\mu_0) u ∈ C ( μ 0 ) such that 1 2 ∥ u ∥ 2 2 = c − ( μ 0 ) = m 2 \frac{1}{2}\lVert u\rVert_2^2=c_-(\mu_0)=m^2 2 1 ∥ u ∥ 2 2 = c − ( μ 0 ) = m 2 , that is, u ∈ C ( μ 0 ) ∩ S m u\in C(\mu_0)\cap S_m u ∈ C ( μ 0 ) ∩ S m . By Lemma 3.3(ii), after a translation, ( μ 0 , u ) (\mu_0,u) ( μ 0 , u ) is a radial solution of (1.1) of constant sign, and
E ( u ) = b m ( μ 0 ) = a ( μ 0 ) − c ∗ μ 0 . \mathcal{E}(u)=b_m(\mu_0)=a(\mu_0)-c_*\mu_0. E ( u ) = b m ( μ 0 ) = a ( μ 0 ) − c ∗ μ 0 .
By Lemma 3.4, this quantity is nonnegative, and it is positive if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 .
(iii) Let m 2 > c ∗ m^2>c_* m 2 > c ∗ and choose μ 1 > 0 \mu_1>0 μ 1 > 0 with c − ( μ 1 ) < m 2 c_-(\mu_1)<m^2 c − ( μ 1 ) < m 2 . By Theorem 2.1, ( b m ) + ′ ( μ 1 ) = c − ( μ 1 ) − m 2 < 0 (b_m)'_+(\mu_1)=c_-(\mu_1)-m^2<0 ( b m ) + ′ ( μ 1 ) = c − ( μ 1 ) − m 2 < 0 , hence b m ( μ 1 + η ) < b m ( μ 1 ) b_m(\mu_1+\eta)<b_m(\mu_1) b m ( μ 1 + η ) < b m ( μ 1 ) for some η > 0 \eta>0 η > 0 . Since b m b_m b m is continuous and b m ( μ ) → + ∞ b_m(\mu)\to+\infty b m ( μ ) → + ∞ as μ → ∞ \mu\to\infty μ → ∞ by Lemma 3.3, the minimum of b m b_m b m over [ μ 1 , ∞ ) [\mu_1,\infty) [ μ 1 , ∞ ) is attained at some μ > μ 1 \mu>\mu_1 μ > μ 1 , which is a local minimum point of b m b_m b m . Lemma 3.3(i) then gives c − ( μ ) = c + ( μ ) = m 2 c_-(\mu)=c_+(\mu)=m^2 c − ( μ ) = c + ( μ ) = m 2 , that is, 1 2 ∥ u ∥ 2 2 = m 2 \frac{1}{2}\lVert u\rVert_2^2=m^2 2 1 ∥ u ∥ 2 2 = m 2 for every u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) . By Lemma 3.3(ii), after a translation, ( μ , u ) (\mu,u) ( μ , u ) is a radial solution of (1.1) of constant sign with E ( u ) = b m ( μ ) \mathcal{E}(u)=b_m(\mu) E ( u ) = b m ( μ ) .
Finally, if f f f is odd, then in (ii) and (iii) we may replace u u u by − u -u − u , see Lemma 3.3(ii), so that the solutions can be taken positive.
Remark 3.5. Let 2 + 4 N < p < 2 ∗ 2+\frac{4}{N}<p<2^* 2 + N 4 < p < 2 ∗ and m 2 > c ∗ m^2>c_* m 2 > c ∗ , and let μ 1 < μ \mu_1<\mu μ 1 < μ be as in the proof of Theorem 1.1(iii). Then b m b_m b m also has a local maximum point μ ˉ ∈ ( 0 , μ ) \bar{\mu}\in(0,\mu) μ ˉ ∈ ( 0 , μ ) . Indeed, b m ( ν ) = ν ( a ( ν ) / ν − m 2 ) > 0 b_m(\nu)=\nu(a(\nu)/\nu-m^2)>0 b m ( ν ) = ν ( a ( ν ) / ν − m 2 ) > 0 for ν \nu ν small by Lemma 3.2, b m ( ν ) → 0 b_m(\nu)\to0 b m ( ν ) → 0 as ν → 0 + \nu\to0^+ ν → 0 + by (34) , and b m ( μ ) < b m ( μ 1 ) b_m(\mu)<b_m(\mu_1) b m ( μ ) < b m ( μ 1 ) . Hence the maximum of b m b_m b m over ( 0 , μ ] (0,\mu] ( 0 , μ ] is positive, and it is attained at some μ ˉ ∈ ( 0 , μ ) \bar{\mu}\in(0,\mu) μ ˉ ∈ ( 0 , μ ) . In contrast with a local minimum point, Theorem 2.1 only gives c − ( μ ˉ ) ≤ m 2 ≤ c + ( μ ˉ ) c_-(\bar{\mu})\leq m^2\leq c_+(\bar{\mu}) c − ( μ ˉ ) ≤ m 2 ≤ c + ( μ ˉ ) at μ ˉ \bar{\mu} μ ˉ , which does not guarantee that some element of C ( μ ˉ ) C(\bar{\mu}) C ( μ ˉ ) has the prescribed mass. Whether μ ˉ \bar{\mu} μ ˉ gives rise to a normalized solution is an open question.
Proof of Theorem 1.2. Since a ( μ ) / μ → 0 a(\mu)/\mu\to0 a ( μ ) / μ → 0 as μ → 0 + \mu\to0^+ μ → 0 + by Lemma 3.2, we have b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) < 0 b_m(\mu)=\mu(a(\mu)/\mu-m^2)<0 b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) < 0 for μ \mu μ small. Hence, by Lemma 3.3, b m b_m b m attains a negative global minimum at some μ > 0 \mu>0 μ > 0 , and Lemma 3.3(i) gives c − ( μ ) = c + ( μ ) = m 2 c_-(\mu)=c_+(\mu)=m^2 c − ( μ ) = c + ( μ ) = m 2 , that is, 1 2 ∥ u ∥ 2 2 = m 2 \frac{1}{2}\lVert u\rVert_2^2=m^2 2 1 ∥ u ∥ 2 2 = m 2 for every u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) . By Lemma 3.3(ii), after a translation, ( μ , u ) (\mu,u) ( μ , u ) is a radial solution of (1.1) of constant sign with
E ( u ) = b m ( μ ) = min ( 0 , ∞ ) b m < 0 , \mathcal{E}(u)=b_m(\mu)=\min_{(0,\infty)}b_m<0, E ( u ) = b m ( μ ) = ( 0 , ∞ ) min b m < 0 ,
and if f f f is odd, we may replace u u u by − u -u − u , so that the solution can be taken positive.
Proof of Theorem 1.3. Step 1: e m ≤ b m ( μ ) e_m\leq b_m(\mu) e m ≤ b m ( μ ) for every μ > 0 \mu>0 μ > 0 . Let μ > 0 \mu>0 μ > 0 , u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) and θ : = ( 2 m 2 / ∥ u ∥ 2 2 ) 1 N + 2 \theta:=(2m^2/\lVert u\rVert_2^2)^{\frac{1}{N+2}} θ := ( 2 m 2 / ∥ u ∥ 2 2 ) N + 2 1 , so that u θ ∈ S m u_\theta\in S_m u θ ∈ S m . Since u ∈ M μ u\in\mathcal{M}_\mu u ∈ M μ , (2.3) gives
e m ≤ E ( u θ ) = I μ ( u θ ) − μ m 2 ≤ I μ ( u ) − μ m 2 = b m ( μ ) . (35) e_m\leq\mathcal{E}(u_\theta)=I_\mu(u_\theta)-\mu m^2\leq I_\mu(u)-\mu m^2=b_m(\mu). \tag*{(35)} e m ≤ E ( u θ ) = I μ ( u θ ) − μ m 2 ≤ I μ ( u ) − μ m 2 = b m ( μ ) . (35)
and the second inequality is strict unless θ = 1 \theta=1 θ = 1 , that is, unless u ∈ S m u\in S_m u ∈ S m .
Step 2: E ( u ) ≥ b m ( μ u ) \mathcal{E}(u)\ge b_m(\mu_u) E ( u ) ≥ b m ( μ u ) for every u ∈ S m u\in S_m u ∈ S m with E ( u ) ≤ 0 \mathcal{E}(u)\le0 E ( u ) ≤ 0 , where
μ u : = N ( N + 2 ) m 2 ( ∫ R N G ( u ) d x − Ψ ( u ) ) , G ( s ) = F ( s ) + 1 N f ( s ) s , \mu_u:=\frac{N}{(N+2)m^2}\left(\int_{\mathbb{R}^N}G(u)\,dx-\Psi(u)\right),\qquad G(s)=F(s)+\frac{1}{N}f(s)s, μ u := ( N + 2 ) m 2 N ( ∫ R N G ( u ) d x − Ψ ( u ) ) , G ( s ) = F ( s ) + N 1 f ( s ) s ,
and equality holds if and only if u ∈ C ( μ u ) u\in C(\mu_u) u ∈ C ( μ u ) . Indeed, E ( u ) ≤ 0 \mathcal{E}(u)\le0 E ( u ) ≤ 0 gives ∫ R N F ( u ) d x ≥ Ψ ( u ) \int_{\mathbb{R}^N}F(u)\,dx\ge\Psi(u) ∫ R N F ( u ) d x ≥ Ψ ( u ) , and ∫ R N f ( u ) u d x > 0 \int_{\mathbb{R}^N}f(u)u\,dx>0 ∫ R N f ( u ) u d x > 0 because f ( s ) s > 0 f(s)s>0 f ( s ) s > 0 for s ≠ 0 s\ne0 s = 0 by (F1)–(F2) and u ≠ 0 u\ne0 u = 0 . Thus μ u > 0 \mu_u>0 μ u > 0 . Since ∥ u ∥ 2 2 = 2 m 2 \lVert u\rVert_2^2=2m^2 ∥ u ∥ 2 2 = 2 m 2 , by (1.6)
M μ u ( u ) = Ψ ( u ) − ∫ R N G ( u ) d x + μ u ( 1 2 + 1 N ) ∥ u ∥ 2 2 = Ψ ( u ) − ∫ R N G ( u ) d x + ( N + 2 ) m 2 N μ u = 0 , M_{\mu_u}(u)=\Psi(u)-\int_{\mathbb{R}^N}G(u)\,dx+\mu_u\left(\frac{1}{2}+\frac{1}{N}\right)\lVert u\rVert_2^2=\Psi(u)-\int_{\mathbb{R}^N}G(u)\,dx+\frac{(N+2)m^2}{N}\mu_u=0, M μ u ( u ) = Ψ ( u ) − ∫ R N G ( u ) d x + μ u ( 2 1 + N 1 ) ∥ u ∥ 2 2 = Ψ ( u ) − ∫ R N G ( u ) d x + N ( N + 2 ) m 2 μ u = 0 ,
i.e. u ∈ M μ u u\in\mathcal{M}_{\mu_u} u ∈ M μ u . Hence I μ u ( u ) ≥ a ( μ u ) I_{\mu_u}(u)\ge a(\mu_u) I μ u ( u ) ≥ a ( μ u ) and E ( u ) = I μ u ( u ) − μ u m 2 ≥ a ( μ u ) − μ u m 2 = b m ( μ u ) \mathcal{E}(u)=I_{\mu_u}(u)-\mu_um^2\ge a(\mu_u)-\mu_um^2=b_m(\mu_u) E ( u ) = I μ u ( u ) − μ u m 2 ≥ a ( μ u ) − μ u m 2 = b m ( μ u ) , with equality if and only if I μ u ( u ) = a ( μ u ) I_{\mu_u}(u)=a(\mu_u) I μ u ( u ) = a ( μ u ) , that is, u ∈ C ( μ u ) u\in C(\mu_u) u ∈ C ( μ u ) .
Step 3: proof of (i) and (ii). By Step 1 and since b m ( μ ) → 0 b_m(\mu)\to0 b m ( μ ) → 0 as μ → 0 + \mu\to0^+ μ → 0 + by Lemma 3.2, we have e m ≤ inf μ > 0 b m ( μ ) ≤ 0 e_m\le\inf_{\mu>0}b_m(\mu)\le0 e m ≤ inf μ > 0 b m ( μ ) ≤ 0 . Conversely, let u ∈ S m u\in S_m u ∈ S m . If E ( u ) > 0 \mathcal{E}(u)>0 E ( u ) > 0 , then E ( u ) > inf μ > 0 b m ( μ ) \mathcal{E}(u)>\inf_{\mu>0}b_m(\mu) E ( u ) > inf μ > 0 b m ( μ ) , and if E ( u ) ≤ 0 \mathcal{E}(u)\le0 E ( u ) ≤ 0 , then E ( u ) ≥ b m ( μ u ) ≥ inf μ > 0 b m ( μ ) \mathcal{E}(u)\ge b_m(\mu_u)\ge\inf_{\mu>0}b_m(\mu) E ( u ) ≥ b m ( μ u ) ≥ inf μ > 0 b m ( μ ) by Step 2. Hence
e m = inf μ > 0 b m ( μ ) ≤ 0 , (36) e_m=\inf_{\mu>0}b_m(\mu)\le0, \tag*{(36)} e m = μ > 0 inf b m ( μ ) ≤ 0 , (36)
where e m > − ∞ e_m>-\infty e m > − ∞ by (A.2), and m 2 ↦ e m m^2\mapsto e_m m 2 ↦ e m is concave and nonincreasing as the infimum of the affine nonincreasing functions m 2 ↦ a ( μ ) − μ m 2 m^2\mapsto a(\mu)-\mu m^2 m 2 ↦ a ( μ ) − μ m 2 . This proves (i).
Let μ m \mu_m μ m be a global minimum point of b m b_m b m and let u ∈ C ( μ m ) u\in C(\mu_m) u ∈ C ( μ m ) . Then b m ( μ m ) = e m b_m(\mu_m)=e_m b m ( μ m ) = e m by (36) , so that equality holds throughout (3.8) with μ = μ m \mu=\mu_m μ = μ m . Hence u ∈ S m u\in S_m u ∈ S m and E ( u ) = e m \mathcal{E}(u)=e_m E ( u ) = e m . Thus C ( μ m ) ⊂ S m C(\mu_m)\subset S_m C ( μ m ) ⊂ S m , that is, c − ( μ m ) = c + ( μ m ) = m 2 c_-(\mu_m)=c_+(\mu_m)=m^2 c − ( μ m ) = c + ( μ m ) = m 2 , and every element of C ( μ m ) C(\mu_m) C ( μ m ) is an energy ground state. Conversely, let u ∈ S m u\in S_m u ∈ S m be such that E ( u ) = e m \mathcal{E}(u)=e_m E ( u ) = e m . Since e m ≤ 0 e_m\le0 e m ≤ 0 , Step 2 and (36) give
e m = E ( u ) ≥ b m ( μ u ) ≥ inf μ > 0 b m ( μ ) = e m . e_m=\mathcal{E}(u)\ge b_m(\mu_u)\ge\inf_{\mu>0}b_m(\mu)=e_m. e m = E ( u ) ≥ b m ( μ u ) ≥ μ > 0 inf b m ( μ ) = e m .
Hence μ u \mu_u μ u is a global minimum point of b m b_m b m , and u ∈ C ( μ u ) u\in C(\mu_u) u ∈ C ( μ u ) by the equality case in Step 2. In particular, e m e_m e m is attained if and only if b m b_m b m attains its infimum on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . The last assertion of (ii) follows from Lemma 3.3(ii).
Step 4: proof of (iii). If 2 < p < 2 + 4 N 2<p<2+\frac{4}{N} 2 < p < 2 + N 4 , then b m b_m b m attains a negative global minimum, as shown in the proof of Theorem 1.2, and (i)–(ii) give the claim. Let 2 + 4 N ≤ p < 2 ∗ 2+\frac{4}{N}\le p<2^* 2 + N 4 ≤ p < 2 ∗ . By Lemma 3.4, c ∗ ≤ r e < ∞ c_*\le r_e<\infty c ∗ ≤ r e < ∞ , and c ∗ < r e c_*<r_e c ∗ < r e if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 . Recall that b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) b_m(\mu)=\mu(a(\mu)/\mu-m^2) b m ( μ ) = μ ( a ( μ ) / μ − m 2 ) . If m 2 < r e m^2<r_e m 2 < r e , then b m > 0 b_m>0 b m > 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Hence, by (3.7), inf μ > 0 b m ( μ ) = 0 \inf_{\mu>0}b_m(\mu)=0 inf μ > 0 b m ( μ ) = 0 is not attained, and e m = 0 e_m=0 e m = 0 is not attained by (i)–(ii). If m 2 > r e m^2>r_e m 2 > r e , then b m b_m b m is negative somewhere. Hence inf μ > 0 b m ( μ ) < 0 \inf_{\mu>0}b_m(\mu)<0 inf μ > 0 b m ( μ ) < 0 is attained by Lemma 3.3, and e m < 0 e_m<0 e m < 0 is attained. If m 2 = r e m^2=r_e m 2 = r e , then b m ≥ 0 b_m\ge0 b m ≥ 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , so that e m = 0 e_m=0 e m = 0 , and b m ( μ ) = 0 b_m(\mu)=0 b m ( μ ) = 0 if and only if a ( μ ) / μ = r e a(\mu)/\mu=r_e a ( μ ) / μ = r e . By (ii), e m = 0 e_m=0 e m = 0 is attained if and only if μ ↦ a ( μ ) / μ \mu\mapsto a(\mu)/\mu μ ↦ a ( μ ) / μ attains its infimum, which is the case if p > 2 + 4 N p>2+\frac{4}{N} p > 2 + N 4 by Lemma 3.4. □
Remark 3.6. For m 2 > r e m^2>r_e m 2 > r e , and for m 2 = r e m^2=r_e m 2 = r e when the infimum r e r_e r e is attained, Theorem 1.3 provides a global minimum point μ m \mu_m μ m of b m b_m b m , and every action ground state at the frequency μ m \mu_m μ m is an energy ground state of mass 2 m 2 2m^2 2 m 2 . The frequency μ \mu μ selected in Theorem 1.1(iii), however, minimizes b m b_m b m only on a half-line [ μ 1 , ∞ ) [\mu_1,\infty) [ μ 1 , ∞ ) with μ 1 < μ \mu_1<\mu μ 1 < μ , as in its proof, and the above arguments do not tell whether this minimum is a global one on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Thus the elements of C ( μ ) C(\mu) C ( μ ) are energy ground states if and only if b m ( μ ) = inf ν > 0 b m ( ν ) b_m(\mu)=\inf_{\nu>0}b_m(\nu) b m ( μ ) = inf ν > 0 b m ( ν ) . For c ∗ < m 2 < r e c_*<m^2<r_e c ∗ < m 2 < r e they have positive energy, so they are not energy ground states, and in fact e m = 0 e_m=0 e m = 0 is not attained. At m 2 = c ∗ m^2=c_* m 2 = c ∗ , when c ∗ c_* c ∗ is attained, Theorem 1.1(ii) provides the elements of C ( μ 0 ) \mathcal{C}(\mu_0) C ( μ 0 ) of mass 2 c − ( μ 0 ) 2c_-(\mu_0) 2 c − ( μ 0 ) , but it does not assert that every element of C ( μ 0 ) \mathcal{C}(\mu_0) C ( μ 0 ) has this mass. The correspondence between the energy ground states and the global minimum points of b m b_m b m is the counterpart of the rigidity result [28 Theorem 1.3] . Finally, we point out that Steps 1–3 of the proof of Theorem 1.3 , and hence assertions (i) and (ii) of that theorem, use neither Lemma 2.4 nor Theorem 2.1 .
Proof of Corollary 1.4. Put p c : = 2 + 4 / N p_c:=2+4/N p c := 2 + 4/ N and define
E Δ ( v ) : = 1 2 ∥ ∇ v ∥ 2 2 − 1 p c ∥ v ∥ p c p c , v ∈ H 1 ( R N ) . \mathcal{E}_{\Delta}(v):=\frac{1}{2}\|\nabla v\|_2^2-\frac{1}{p_c}\|v\|_{p_c}^{p_c}, \qquad v\in H^1(\mathbb{R}^N). E Δ ( v ) := 2 1 ∥∇ v ∥ 2 2 − p c 1 ∥ v ∥ p c p c , v ∈ H 1 ( R N ) .
The sharp Gagliardo–Nirenberg inequality [43 ] states that
∥ v ∥ p c p c ≤ p c 2 ∥ Q ∥ 2 4 / N ∥ ∇ v ∥ 2 2 ∥ v ∥ 2 4 / N \|v\|_{p_c}^{p_c}\leq\frac{p_c}{2\|Q\|_2^{4/N}}\|\nabla v\|_2^2\|v\|_2^{4/N} ∥ v ∥ p c p c ≤ 2∥ Q ∥ 2 4/ N p c ∥∇ v ∥ 2 2 ∥ v ∥ 2 4/ N
for every v ∈ H 1 ( R N ) v\in H^1(\mathbb{R}^N) v ∈ H 1 ( R N ) .
Since 1 − 1 − r 2 > r 2 / 2 1-\sqrt{1-r^2}>r^2/2 1 − 1 − r 2 > r 2 /2 for 0 < r ≤ 1 0<r\leq1 0 < r ≤ 1 and a nonzero H 1 H^1 H 1 function cannot have zero gradient almost everywhere, for every u ∈ S m u\in\mathcal{S}_m u ∈ S m we obtain
E ( u ) > E Δ ( u ) ≥ 1 2 [ 1 − ( m m c ) 4 / N ] ∥ ∇ u ∥ 2 2 . \mathcal{E}(u)>\mathcal{E}_{\Delta}(u)\geq\frac{1}{2}\left[1-\left(\frac{m}{m_c}\right)^{4/N}\right]\|\nabla u\|_2^2. E ( u ) > E Δ ( u ) ≥ 2 1 [ 1 − ( m c m ) 4/ N ] ∥∇ u ∥ 2 2 .
Thus E ( u ) > 0 \mathcal{E}(u)>0 E ( u ) > 0 for 0 < m ≤ m c 0<m\leq m_c 0 < m ≤ m c . Since e m ≤ 0 e_m\leq0 e m ≤ 0 by (36) , it follows that e m = 0 e_m=0 e m = 0 and that the infimum is not attained in this range.
Let now m > m c m>m_c m > m c and put α : = m / m c > 1 \alpha:=m/m_c>1 α := m / m c > 1 . The Nehari and Pohožaev identities for Q Q Q give
∥ Q ∥ p c p c = p c 2 ∥ ∇ Q ∥ 2 2 . \|Q\|_{p_c}^{p_c}=\frac{p_c}{2}\|\nabla Q\|_2^2. ∥ Q ∥ p c p c = 2 p c ∥∇ Q ∥ 2 2 .
Consequently, ∥ α Q ∥ 2 2 = 2 m 2 \|\alpha Q\|_2^2=2m^2 ∥ α Q ∥ 2 2 = 2 m 2 and
E Δ ( α Q ) = α 2 − α p c 2 ∥ ∇ Q ∥ 2 2 < 0. \mathcal{E}_{\Delta}(\alpha Q)=\frac{\alpha^2-\alpha^{p_c}}{2}\|\nabla Q\|_2^2<0. E Δ ( α Q ) = 2 α 2 − α p c ∥∇ Q ∥ 2 2 < 0.
Approximating α Q \alpha Q α Q in H 1 H^1 H 1 by smooth compactly supported functions and normalizing their L 2 L^2 L 2 -norms, we may therefore choose v ∈ C ∞ ( R N ) v\in C^\infty(\mathbb{R}^N) v ∈ C ∞ ( R N ) with ∥ v ∥ 2 2 = 2 m 2 \|v\|_2^2=2m^2 ∥ v ∥ 2 2 = 2 m 2 and E Δ ( v ) < 0 \mathcal{E}_{\Delta}(v)<0 E Δ ( v ) < 0 . For λ > 0 \lambda>0 λ > 0 set v λ ( x ) = λ N / 2 v ( λ x ) v_\lambda(x)=\lambda^{N/2}v(\lambda x) v λ ( x ) = λ N /2 v ( λ x ) . For all sufficiently small λ \lambda λ ,
∥ v λ ∥ 2 2 = 2 m 2 , ∥ ∇ v λ ∥ ∞ = λ N / 2 + 1 ∥ ∇ v ∥ ∞ < 1 , \|v_\lambda\|_2^2=2m^2,\qquad\|\nabla v_\lambda\|_\infty=\lambda^{N/2+1}\|\nabla v\|_\infty<1, ∥ v λ ∥ 2 2 = 2 m 2 , ∥∇ v λ ∥ ∞ = λ N /2 + 1 ∥∇ v ∥ ∞ < 1 ,
hence v λ ∈ S m v_\lambda\in\mathcal{S}_m v λ ∈ S m . Rationalizing the kinetic density and changing variables gives
Ψ ( v λ ) λ 2 = ∫ R N ∣ ∇ v ∣ 2 1 + 1 − λ N + 2 ∣ ∇ v ∣ 2 d x ⟶ 1 2 ∥ ∇ v ∥ 2 2 as λ → 0 + . \frac{\Psi(v_\lambda)}{\lambda^2}=\int_{\mathbb{R}^N}\frac{|\nabla v|^2}{1+\sqrt{1-\lambda^{N+2}|\nabla v|^2}}\,dx\longrightarrow\frac{1}{2}\|\nabla v\|_2^2\quad\text{as }\lambda\to0^+. λ 2 Ψ ( v λ ) = ∫ R N 1 + 1 − λ N + 2 ∣∇ v ∣ 2 ∣∇ v ∣ 2 d x ⟶ 2 1 ∥∇ v ∥ 2 2 as λ → 0 + .
The integrand is bounded by ∣ ∇ v ∣ 2 |\nabla v|^2 ∣∇ v ∣ 2 , so the limit follows from dominated convergence. Since ∥ v λ ∥ p c p c = λ 2 ∥ v ∥ p c p c \|v_\lambda\|_{p_c}^{p_c}=\lambda^2\|v\|_{p_c}^{p_c} ∥ v λ ∥ p c p c = λ 2 ∥ v ∥ p c p c , we obtain
λ − 2 E ( v λ ) ⟶ E Δ ( v ) < 0. \lambda^{-2}\mathcal{E}(v_\lambda)\longrightarrow\mathcal{E}_{\Delta}(v)<0. λ − 2 E ( v λ ) ⟶ E Δ ( v ) < 0.
Hence e m < 0 e_m<0 e m < 0 for every m > m c m>m_c m > m c . The sign characterization in Theorem 1.3 (iii) now gives r e = m c 2 r_e=m_c^2 r e = m c 2 and attainment for m > m c m>m_c m > m c . Since e m c = 0 e_{m_c}=0 e m c = 0 is not attained, the same theorem shows that the infimum of a ( μ ) / μ a(\mu)/\mu a ( μ ) / μ is not attained either.
As for the comparison with the Laplacian, the same sharp inequality and the mass-preserving dilations v λ v_\lambda v λ show that the infimum of E Δ \mathcal{E}_{\Delta} E Δ on the mass sphere in H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) is zero for 0 < m ≤ m c 0<m\leq m_c 0 < m ≤ m c . It is not attained for m < m c m<m_c m < m c , by the displayed lower bound for E Δ \mathcal{E}_{\Delta} E Δ , and it is attained by Q Q Q at m = m c m=m_c m = m c . For m > m c m>m_c m > m c , a function v v v with negative energy as above satisfies E Δ ( v λ ) = λ 2 E Δ ( v ) → − ∞ \mathcal{E}_{\Delta}(v_\lambda)=\lambda^2\mathcal{E}_{\Delta}(v)\to-\infty E Δ ( v λ ) = λ 2 E Δ ( v ) → − ∞ as λ → ∞ \lambda\to\infty λ → ∞ . These concentrating dilations are admissible on the mass sphere in H 1 ( R N ) H^1(\mathbb{R}^N) H 1 ( R N ) , but not in S m \mathcal{S}_m S m .
Appendix A. Two estimates for functions in D \mathcal{D} D
In this appendix we prove the two estimates announced in § 1 , which express the fact that the gradient constraint ∥ ∇ u ∥ ∞ ≤ 1 \lVert\nabla u\rVert_{\infty}\le1 ∥ ∇ u ∥ ∞ ≤ 1 prevents concentration.
Lemma A.1. Let u ∈ D u\in\mathcal{D} u ∈ D . Then
∥ u ∥ ∞ ≤ 2 ( ∥ u ∥ 2 2 ∣ B 1 ∣ ) 1 N + 2 , ∥ u ∥ ∞ ≤ C N ′ ∥ ∇ u ∥ 2 2 N , (37) \lVert u\rVert_{\infty}\le2\left(\frac{\lVert u\rVert_{2}^{2}}{|B_{1}|}\right)^{\frac{1}{N+2}},\qquad\lVert u\rVert_{\infty}\le C_{N}^{\prime}\lVert\nabla u\rVert_{2}^{\frac{2}{N}}, \tag*{(37)} ∥ u ∥ ∞ ≤ 2 ( ∣ B 1 ∣ ∥ u ∥ 2 2 ) N + 2 1 , ∥ u ∥ ∞ ≤ C N ′ ∥ ∇ u ∥ 2 N 2 , (37) where C N ′ C_{N}^{\prime} C N ′ depends only on N N N .
Proof. By (1.3), u ∈ C 0 ( R N ) u\in C_{0}(\mathbb{R}^{N}) u ∈ C 0 ( R N ) , hence A : = ∥ u ∥ ∞ = ∣ u ( x 0 ) ∣ A:=\lVert u\rVert_{\infty}=|u(x_{0})| A := ∥ u ∥ ∞ = ∣ u ( x 0 ) ∣ for some x 0 ∈ R N x_{0}\in\mathbb{R}^{N} x 0 ∈ R N , and we may assume A > 0 A>0 A > 0 . Since u u u is 1-Lipschitz, ∣ u ∣ ≥ A / 2 |u|\ge A/2 ∣ u ∣ ≥ A /2 on B A / 2 ( x 0 ) B_{A/2}(x_{0}) B A /2 ( x 0 ) , hence
∥ u ∥ 2 2 ≥ ∫ B A / 2 ( x 0 ) ∣ u ∣ 2 d x ≥ ∣ B 1 ∣ ( A 2 ) N + 2 . \lVert u\rVert_{2}^{2}\ge\int_{B_{A/2}(x_{0})}|u|^{2}\,dx\ge|B_{1}|\left(\frac{A}{2}\right)^{N+2}. ∥ u ∥ 2 2 ≥ ∫ B A /2 ( x 0 ) ∣ u ∣ 2 d x ≥ ∣ B 1 ∣ ( 2 A ) N + 2 .
which is the first inequality in (A.1). For the second one, set w = ( ∣ u ∣ − A / 2 ) + w=(|u|-A/2)_{+} w = ( ∣ u ∣ − A /2 ) + . Then w ∈ H 1 ( R N ) w\in H^{1}(\mathbb{R}^{N}) w ∈ H 1 ( R N ) , ∥ ∇ w ∥ 2 ≤ ∥ ∇ u ∥ 2 \lVert\nabla w\rVert_{2}\le\lVert\nabla u\rVert_{2} ∥ ∇ w ∥ 2 ≤ ∥ ∇ u ∥ 2 and w ≥ A / 4 w\ge A/4 w ≥ A /4 on B A / 4 ( x 0 ) B_{A/4}(x_{0}) B A /4 ( x 0 ) . The Sobolev inequality then gives
A 4 ∣ B 1 ∣ 1 / 2 ∗ ( A 4 ) N / 2 ∗ ≤ ∥ w ∥ 2 ∗ ≤ S N ∥ ∇ w ∥ 2 . \frac{A}{4}|B_{1}|^{1/2^{*}}\left(\frac{A}{4}\right)^{N/2^{*}}\le\lVert w\rVert_{2^{*}}\le S_{N}\lVert\nabla w\rVert_{2}. 4 A ∣ B 1 ∣ 1/ 2 ∗ ( 4 A ) N / 2 ∗ ≤ ∥ w ∥ 2 ∗ ≤ S N ∥ ∇ w ∥ 2 .
Since 1 + N / 2 ∗ = N / 2 1+N/2^{*}=N/2 1 + N / 2 ∗ = N /2 , we get A N / 2 ≤ C ∥ ∇ u ∥ 2 A^{N/2}\le C\lVert\nabla u\rVert_{2} A N /2 ≤ C ∥ ∇ u ∥ 2 , and the second inequality in (A.1) follows. □
Lemma A.2. Assume (F1)–(F2) and let m > 0 m>0 m > 0 . Then
inf u ∈ S m E ( u ) ≥ − κ ( S ( m ) ) m 2 > − ∞ , (38) \inf_{u\in\mathcal{S}_{m}}\mathcal{E}(u)\ge-\kappa(S(m))m^{2}>-\infty, \tag*{(38)} u ∈ S m inf E ( u ) ≥ − κ ( S ( m )) m 2 > − ∞ , (38) where S ( m ) : = 2 ( 2 m 2 / ∣ B 1 ∣ ) 1 N + 2 S(m):=2(2m^{2}/|B_{1}|)^{\frac{1}{N+2}} S ( m ) := 2 ( 2 m 2 /∣ B 1 ∣ ) N + 2 1 and κ ( S ) : = max { f ( S ) / S , f ( − S ) / ( − S ) } \kappa(S):=\max\{f(S)/S,f(-S)/(-S)\} κ ( S ) := max { f ( S ) / S , f ( − S ) / ( − S )} .
Proof. Let u ∈ S m u\in\mathcal{S}_{m} u ∈ S m . By (A.1), ∥ u ∥ ∞ ≤ S ( m ) \lVert u\rVert_{\infty}\le S(m) ∥ u ∥ ∞ ≤ S ( m ) . Since s ↦ f ( s ) / s s\mapsto f(s)/s s ↦ f ( s ) / s is positive, nondecreasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) and nonincreasing on ( − ∞ , 0 ) (-\infty,0) ( − ∞ , 0 ) by (F1)–(F2), we have 0 < f ( s ) / s ≤ κ ( S ) 0<f(s)/s\le\kappa(S) 0 < f ( s ) / s ≤ κ ( S ) for 0 < ∣ s ∣ ≤ S 0<|s|\le S 0 < ∣ s ∣ ≤ S , hence
F ( s ) = ∫ 0 s f ( t ) t t d t ≤ κ ( S ) ⋅ s 2 2 (39) F(s)=\int_{0}^{s}\frac{f(t)}{t}\,t\,dt\le\kappa(S)\cdot\frac{s^{2}}{2} \tag*{(39)} F ( s ) = ∫ 0 s t f ( t ) t d t ≤ κ ( S ) ⋅ 2 s 2 (39)
for every S > 0 S>0 S > 0 and $|s|\le S. Applied with S = S ( m ) S=S(m) S = S ( m ) , this gives ∫ R N F ( u ) d x ≤ 1 2 κ ( S ( m ) ) ∥ u ∥ 2 2 = κ ( S ( m ) ) m 2 \int_{\mathbb{R}^{N}}F(u)\,dx\le\frac{1}{2}\kappa(S(m))\lVert u\rVert_{2}^{2}=\kappa(S(m))m^{2} ∫ R N F ( u ) d x ≤ 2 1 κ ( S ( m )) ∥ u ∥ 2 2 = κ ( S ( m )) m 2 , and (A.2) follows from Ψ ≥ 0 \Psi\ge0 Ψ ≥ 0 . □
Appendix B. Symmetry and monotonicity of the action ground states
In this appendix we prove that every action ground state of ( B I μ ) (BI_{\mu}) ( B I μ ) has constant sign and is radially symmetric about some point, with a strictly decreasing radial profile of its absolute value, see Theorem B.5 . In the application to the normalized problem, the symmetry of every action ground state is needed, and not only the existence of a radial minimizer at each frequency: for instance, at m 2 = c ∗ m^{2}=c_{*} m 2 = c ∗ we first select a ground state of minimum mass and then center it by a translation, which does not change its mass, and the same argument gives the radial symmetry of every energy ground state.
We combine the Schwarz rearrangement with the equality case of the Pólya–Szegő inequality due to Brothers and Ziemer [17 ] , in the form given by Burchard and Ferone [16 ] , and with the uniqueness for the radial ordinary differential equation. The point is that the Born–Infeld density is a series of even powers of ∣ ∇ u ∣ |\nabla u| ∣∇ u ∣ with positive coefficients, and hence the equality Ψ ( ∣ u ∣ ∗ ) = Ψ ( u ) \Psi(|u|^{*})=\Psi(u) Ψ ( ∣ u ∣ ∗ ) = Ψ ( u ) , which follows from
@page 22/27 starts_mid=0 ends_mid=1
Appendix A. Two estimates for functions in D \mathcal{D} D
In this appendix we prove the two estimates announced in § 1 , which express the fact that the gradient constraint ∥ ∇ u ∥ ∞ ≤ 1 \lVert\nabla u\rVert_{\infty}\le1 ∥ ∇ u ∥ ∞ ≤ 1 prevents concentration.
Lemma A.1. Let u ∈ D u\in\mathcal{D} u ∈ D . Then
∥ u ∥ ∞ ≤ 2 ( ∥ u ∥ 2 2 ∣ B 1 ∣ ) 1 N + 2 , ∥ u ∥ ∞ ≤ C N ′ ∥ ∇ u ∥ 2 2 N , (40) \lVert u\rVert_{\infty}\le2\left(\frac{\lVert u\rVert_{2}^{2}}{|B_{1}|}\right)^{\frac{1}{N+2}},\qquad\lVert u\rVert_{\infty}\le C_{N}^{\prime}\lVert\nabla u\rVert_{2}^{\frac{2}{N}}, \tag*{(40)} ∥ u ∥ ∞ ≤ 2 ( ∣ B 1 ∣ ∥ u ∥ 2 2 ) N + 2 1 , ∥ u ∥ ∞ ≤ C N ′ ∥ ∇ u ∥ 2 N 2 , (40) where C N ′ C_{N}^{\prime} C N ′ depends only on N N N .
Proof. By (1.3), u ∈ C 0 ( R N ) u\in C_{0}(\mathbb{R}^{N}) u ∈ C 0 ( R N ) , hence A : = ∥ u ∥ ∞ = ∣ u ( x 0 ) ∣ A:=\lVert u\rVert_{\infty}=|u(x_{0})| A := ∥ u ∥ ∞ = ∣ u ( x 0 ) ∣ for some x 0 ∈ R N x_{0}\in\mathbb{R}^{N} x 0 ∈ R N , and we may assume A > 0 A>0 A > 0 . Since u u u is 1-Lipschitz, ∣ u ∣ ≥ A / 2 |u|\ge A/2 ∣ u ∣ ≥ A /2 on B A / 2 ( x 0 ) B_{A/2}(x_{0}) B A /2 ( x 0 ) , hence
∥ u ∥ 2 2 ≥ ∫ B A / 2 ( x 0 ) ∣ u ∣ 2 d x ≥ ∣ B 1 ∣ ( A 2 ) N + 2 . \lVert u\rVert_{2}^{2}\ge\int_{B_{A/2}(x_{0})}|u|^{2}\,dx\ge|B_{1}|\left(\frac{A}{2}\right)^{N+2}. ∥ u ∥ 2 2 ≥ ∫ B A /2 ( x 0 ) ∣ u ∣ 2 d x ≥ ∣ B 1 ∣ ( 2 A ) N + 2 .
which is the first inequality in (A.1). For the second one, set w = ( ∣ u ∣ − A / 2 ) + w=(|u|-A/2)_{+} w = ( ∣ u ∣ − A /2 ) + . Then w ∈ H 1 ( R N ) w\in H^{1}(\mathbb{R}^{N}) w ∈ H 1 ( R N ) , ∥ ∇ w ∥ 2 ≤ ∥ ∇ u ∥ 2 \lVert\nabla w\rVert_{2}\le\lVert\nabla u\rVert_{2} ∥ ∇ w ∥ 2 ≤ ∥ ∇ u ∥ 2 and w ≥ A / 4 w\ge A/4 w ≥ A /4 on B A / 4 ( x 0 ) B_{A/4}(x_{0}) B A /4 ( x 0 ) . The Sobolev inequality then gives
A 4 ∣ B 1 ∣ 1 / 2 ∗ ( A 4 ) N / 2 ∗ ≤ ∥ w ∥ 2 ∗ ≤ S N ∥ ∇ w ∥ 2 . \frac{A}{4}|B_{1}|^{1/2^{*}}\left(\frac{A}{4}\right)^{N/2^{*}}\le\lVert w\rVert_{2^{*}}\le S_{N}\lVert\nabla w\rVert_{2}. 4 A ∣ B 1 ∣ 1/ 2 ∗ ( 4 A ) N / 2 ∗ ≤ ∥ w ∥ 2 ∗ ≤ S N ∥ ∇ w ∥ 2 .
Since 1 + N / 2 ∗ = N / 2 1+N/2^{*}=N/2 1 + N / 2 ∗ = N /2 , we get A N / 2 ≤ C ∥ ∇ u ∥ 2 A^{N/2}\le C\lVert\nabla u\rVert_{2} A N /2 ≤ C ∥ ∇ u ∥ 2 , and the second inequality in (A.1) follows. □
Lemma A.2. Assume (F1)–(F2) and let m > 0 m>0 m > 0 . Then
inf u ∈ S m E ( u ) ≥ − κ ( S ( m ) ) m 2 > − ∞ , (41) \inf_{u\in\mathcal{S}_{m}}\mathcal{E}(u)\ge-\kappa(S(m))m^{2}>-\infty, \tag*{(41)} u ∈ S m inf E ( u ) ≥ − κ ( S ( m )) m 2 > − ∞ , (41) where S ( m ) : = 2 ( 2 m 2 / ∣ B 1 ∣ ) 1 N + 2 S(m):=2(2m^{2}/|B_{1}|)^{\frac{1}{N+2}} S ( m ) := 2 ( 2 m 2 /∣ B 1 ∣ ) N + 2 1 and κ ( S ) : = max { f ( S ) / S , f ( − S ) / ( − S ) } \kappa(S):=\max\{f(S)/S,f(-S)/(-S)\} κ ( S ) := max { f ( S ) / S , f ( − S ) / ( − S )} .
Proof. Let u ∈ S m u\in\mathcal{S}_{m} u ∈ S m . By (A.1), ∥ u ∥ ∞ ≤ S ( m ) \lVert u\rVert_{\infty}\le S(m) ∥ u ∥ ∞ ≤ S ( m ) . Since s ↦ f ( s ) / s s\mapsto f(s)/s s ↦ f ( s ) / s is positive, nondecreasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) and nonincreasing on ( − ∞ , 0 ) (-\infty,0) ( − ∞ , 0 ) by (F1)–(F2), we have 0 < f ( s ) / s ≤ κ ( S ) 0<f(s)/s\le\kappa(S) 0 < f ( s ) / s ≤ κ ( S ) for 0 < ∣ s ∣ ≤ S 0<|s|\le S 0 < ∣ s ∣ ≤ S , hence
F ( s ) = ∫ 0 s f ( t ) t t d t ≤ κ ( S ) ⋅ s 2 2 (42) F(s)=\int_{0}^{s}\frac{f(t)}{t}\,t\,dt\le\kappa(S)\cdot\frac{s^{2}}{2} \tag*{(42)} F ( s ) = ∫ 0 s t f ( t ) t d t ≤ κ ( S ) ⋅ 2 s 2 (42)
for every S > 0 S>0 S > 0 and $|s|\le S. Applied with S = S ( m ) S=S(m) S = S ( m ) , this gives ∫ R N F ( u ) d x ≤ 1 2 κ ( S ( m ) ) ∥ u ∥ 2 2 = κ ( S ( m ) ) m 2 \int_{\mathbb{R}^{N}}F(u)\,dx\le\frac{1}{2}\kappa(S(m))\lVert u\rVert_{2}^{2}=\kappa(S(m))m^{2} ∫ R N F ( u ) d x ≤ 2 1 κ ( S ( m )) ∥ u ∥ 2 2 = κ ( S ( m )) m 2 , and (A.2) follows from Ψ ≥ 0 \Psi\ge0 Ψ ≥ 0 . □
Appendix B. Symmetry and monotonicity of the action ground states
In this appendix we prove that every action ground state of ( B I μ ) (BI_{\mu}) ( B I μ ) has constant sign and is radially symmetric about some point, with a strictly decreasing radial profile of its absolute value, see Theorem B.5 . In the application to the normalized problem, the symmetry of every action ground state is needed, and not only the existence of a radial minimizer at each frequency: for instance, at m 2 = c ∗ m^{2}=c_{*} m 2 = c ∗ we first select a ground state of minimum mass and then center it by a translation, which does not change its mass, and the same argument gives the radial symmetry of every energy ground state.
We combine the Schwarz rearrangement with the equality case of the Pólya–Szegő inequality due to Brothers and Ziemer [17 ] , in the form given by Burchard and Ferone [16 ] , and with the uniqueness for the radial ordinary differential equation. The point is that the Born–Infeld density is a series of even powers of ∣ ∇ u ∣ |\nabla u| ∣∇ u ∣ with positive coefficients, and hence the equality Ψ ( ∣ u ∣ ∗ ) = Ψ ( u ) \Psi(|u|^{*})=\Psi(u) Ψ ( ∣ u ∣ ∗ ) = Ψ ( u ) , which follows from the minimality, reduces to the equality ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 \lVert\nabla u^*\rVert_2 = \lVert\nabla u\rVert_2 ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 . Throughout, N ≥ 3 N \ge3 N ≥ 3 , μ > 0 \mu> 0 μ > 0 and f f f satisfies (F1)–(F3), and we use the notation of Subsection 1.1. For a nonnegative v ∈ D v \in\mathcal{D} v ∈ D we denote by v ∗ v^* v ∗ its Schwarz rearrangement, which is nonnegative, radial, radially nonincreasing and equimeasurable with v v v , see [36 Chapter 3] .
We recall from Subsection 1.1 that every u ∈ C ( μ ) u \in\mathcal{C}(\mu) u ∈ C ( μ ) is a critical point of I μ I_\mu I μ in the sense of (1.5), hence a solution of ( B I μ ) (BI_\mu) ( B I μ ) of class C 2 ( R N ) C^2(\mathbb{R}^N) C 2 ( R N ) with ∥ ∇ u ∥ ∞ < 1 \lVert\nabla u\rVert_\infty< 1 ∥ ∇ u ∥ ∞ < 1 , see [6 Lemmas 2.3 and 5.5] .
Lemma B.1. Every u ∈ C ( μ ) u \in\mathcal{C}(\mu) u ∈ C ( μ ) has constant sign, i.e., u > 0 u > 0 u > 0 or u < 0 u < 0 u < 0 a.e. on R N \mathbb{R}^N R N .
Proof. See [6 Proof of Theorem 1.6] . □
Lemma B.2. Let u ∈ C ( μ ) u \in\mathcal{C}(\mu) u ∈ C ( μ ) be nonnegative. Then u ∗ ∈ C ( μ ) u^* \in\mathcal{C}(\mu) u ∗ ∈ C ( μ ) , Ψ ( u ∗ ) = Ψ ( u ) \Psi(u^*) = \Psi(u) Ψ ( u ∗ ) = Ψ ( u ) and ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 \lVert\nabla u^*\rVert_2 = \lVert\nabla u\rVert_2 ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 .
Proof. By [6 Proof of Theorem 1.6] , applied to the equation with unit frequency and with f f f replaced by g μ g_\mu g μ as in Subsection 1.1, we have u ∗ ∈ C ( μ ) u^* \in\mathcal{C}(\mu) u ∗ ∈ C ( μ ) . As in that proof, see [6 Proof of Theorem 1.2] , the Pólya–Szegő inequality applies to every term of the expansion 1 − 1 − r 2 = ∑ k = 1 ∞ a k r 2 k 1-\sqrt{1-r^2}=\sum_{k=1}^{\infty}a_k r^{2k} 1 − 1 − r 2 = ∑ k = 1 ∞ a k r 2 k , 0 ≤ r ≤ 1 0 \le r \le1 0 ≤ r ≤ 1 , where a k > 0 a_k>0 a k > 0 and a 1 = 1 2 a_1=\frac{1}{2} a 1 = 2 1 . Since ∣ ∇ u ∣ ≤ 1 \lvert\nabla u\rvert\le1 ∣ ∇ u ∣ ≤ 1 and ∣ ∇ u ∗ ∣ ≤ 1 \lvert\nabla u^*\rvert\le1 ∣ ∇ u ∗ ∣ ≤ 1 a.e. on R N \mathbb{R}^N R N , the monotone convergence theorem gives
Ψ ( u ) − Ψ ( u ∗ ) = ∑ k = 1 ∞ a k ( ∫ R N ∣ ∇ u ∣ 2 k d x − ∫ R N ∣ ∇ u ∗ ∣ 2 k d x ) . (43) \Psi(u)-\Psi(u^*)=\sum_{k=1}^{\infty}a_k\left(\int_{\mathbb{R}^N}\lvert\nabla u\rvert^{2k}\,dx-\int_{\mathbb{R}^N}\lvert\nabla u^*\rvert^{2k}\,dx\right). \tag*{(43)} Ψ ( u ) − Ψ ( u ∗ ) = k = 1 ∑ ∞ a k ( ∫ R N ∣ ∇ u ∣ 2 k d x − ∫ R N ∣ ∇ u ∗ ∣ 2 k d x ) . (43)
where every term of the series is nonnegative, see also [19 Lemma 9] . On the other hand, since M μ ( u ) = M μ ( u ∗ ) = 0 M_\mu(u)=M_\mu(u^*)=0 M μ ( u ) = M μ ( u ∗ ) = 0 and since u u u and u ∗ u^* u ∗ are equimeasurable,
Ψ ( u ∗ ) = ∫ R N H μ ( u ∗ ) d x = ∫ R N H μ ( u ) d x = Ψ ( u ) . \Psi(u^*)=\int_{\mathbb{R}^N}H_\mu(u^*)\,dx=\int_{\mathbb{R}^N}H_\mu(u)\,dx=\Psi(u). Ψ ( u ∗ ) = ∫ R N H μ ( u ∗ ) d x = ∫ R N H μ ( u ) d x = Ψ ( u ) .
Hence all the terms in (B.1) vanish, and the term with k = 1 k=1 k = 1 gives ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 \lVert\nabla u^*\rVert_2=\lVert\nabla u\rVert_2 ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 . □
Lemma B.3. Let v ∈ C ( μ ) v \in\mathcal{C}(\mu) v ∈ C ( μ ) and ε ∈ { − 1 , 1 } \varepsilon\in\{-1,1\} ε ∈ { − 1 , 1 } be such that ε v ( x ) = W ( ∣ x ∣ ) \varepsilon v(x)=W(\lvert x\rvert) ε v ( x ) = W (∣ x ∣) for x ∈ R N x \in\mathbb{R}^N x ∈ R N , where W ≥ 0 W \ge0 W ≥ 0 is nonincreasing on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . Then W ∈ C 2 ( [ 0 , ∞ ) ) W \in C^2([0,\infty)) W ∈ C 2 ([ 0 , ∞ )) , W ( r ) > 0 W(r)>0 W ( r ) > 0 for every r ≥ 0 r \ge0 r ≥ 0 and W ′ ( r ) < 0 W'(r)<0 W ′ ( r ) < 0 for every r > 0 r>0 r > 0 .
Proof. Since v ∈ C 2 ( R N ) v \in C^2(\mathbb{R}^N) v ∈ C 2 ( R N ) is a solution of ( B I μ ) (BI_\mu) ( B I μ ) with ∥ ∇ v ∥ ∞ < 1 \lVert\nabla v\rVert_\infty<1 ∥ ∇ v ∥ ∞ < 1 , the function W ( r ) = ε v ( r e 1 ) W(r)=\varepsilon v(re_1) W ( r ) = ε v ( r e 1 ) belongs to C 2 ( [ 0 , ∞ ) ) C^2([0,\infty)) C 2 ([ 0 , ∞ )) , W ′ ( 0 ) = 0 W'(0)=0 W ′ ( 0 ) = 0 , ∣ W ′ ∣ ≤ ∥ ∇ v ∥ ∞ < 1 \lvert W'\rvert\le\lVert\nabla v\rVert_\infty<1 ∣ W ′ ∣ ≤ ∥ ∇ v ∥ ∞ < 1 and W ′ ≤ 0 W'\le0 W ′ ≤ 0 on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . Put h ε ( s ) : = ε f ( ε s ) − μ s h_\varepsilon(s):=\varepsilon f(\varepsilon s)-\mu s h ε ( s ) := ε f ( ε s ) − μ s , which is locally Lipschitz continuous by (F1) with h ε ( 0 ) = 0 h_\varepsilon(0)=0 h ε ( 0 ) = 0 . Writing the equation for ε v \varepsilon v ε v in polar coordinates, for r > 0 r>0 r > 0 the profile W W W solves
W ′ ′ = − N − 1 r W ′ ( 1 − W ′ 2 ) − h ε ( W ) ( 1 − W ′ 2 ) 3 / 2 . (44) W''=-\frac{N-1}{r}W'(1-W'^2)-h_\varepsilon(W)(1-W'^2)^{3/2}. \tag*{(44)} W ′′ = − r N − 1 W ′ ( 1 − W ′2 ) − h ε ( W ) ( 1 − W ′2 ) 3/2 . (44)
Since the right-hand side of (B.2) is locally Lipschitz continuous in ( W , W ′ ) (W,W') ( W , W ′ ) for ∣ W ′ ∣ < 1 \lvert W'\rvert<1 ∣ W ′ ∣ < 1 and r > 0 r>0 r > 0 , the solutions of (B.2) with ∣ W ′ ∣ < 1 \lvert W'\rvert<1 ∣ W ′ ∣ < 1 are uniquely determined by their data at any r 0 > 0 r_0>0 r 0 > 0 .
If W ( r 0 ) = 0 W(r_0)=0 W ( r 0 ) = 0 for some r 0 > 0 r_0>0 r 0 > 0 , then W = 0 W=0 W = 0 on [ r 0 , ∞ ) [r_0,\infty) [ r 0 , ∞ ) because W W W is nonnegative and nonincreasing, and W ≡ 0 W\equiv0 W ≡ 0 is the solution of (B.2) with the data ( 0 , 0 ) (0,0) ( 0 , 0 ) at r 0 r_0 r 0 , because h ε ( 0 ) = 0 h_\varepsilon(0)=0 h ε ( 0 ) = 0 . By uniqueness, W = 0 W=0 W = 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , which contradicts v ≠ 0 v\ne0 v = 0 . Hence W > 0 W>0 W > 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , and W ( 0 ) ≥ W ( r ) > 0 W(0)\ge W(r)>0 W ( 0 ) ≥ W ( r ) > 0 by monotonicity.
If W ′ ( r 0 ) = 0 W'(r_0)=0 W ′ ( r 0 ) = 0 for some r 0 > 0 r_0>0 r 0 > 0 , then r 0 r_0 r 0 is a maximum point of W ′ ≤ 0 W'\le0 W ′ ≤ 0 , hence W ′ ′ ( r 0 ) = 0 W''(r_0)=0 W ′′ ( r 0 ) = 0 , and (B.2) gives h ε ( W ( r 0 ) ) = 0 h_\varepsilon(W(r_0))=0 h ε ( W ( r 0 )) = 0 . The constant function W ( r 0 ) W(r_0) W ( r 0 ) then solves (B.2) with the same data at r 0 r_0 r 0 , and uniqueness forces W W W to be constant on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , which contradicts the positivity of W W W and W ( r ) → 0 W(r)\to0 W ( r ) → 0 as r → ∞ r\to\infty r → ∞ . Hence W ′ < 0 W'<0 W ′ < 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . □
We next recall the equality case of the Pólya–Szegő inequality, due to Brothers and Ziemer [17 Theorem 1.1] , in the quantitative form of Burchard and Ferone [16 ] . As in [16 ] , a nonnegative measurable function z z z on R N \mathbb{R}^{N} R N is said to vanish at infinity if the level sets { z > t } \{z>t\} { z > t } , t > 0 t>0 t > 0 , have finite measure, and ω k \omega_k ω k denotes the Lebesgue measure of the unit ball in R k \mathbb{R}^{k} R k .
Lemma B.4 ([16 Theorem 1.1] ). Let 1 < p < ∞ 1<p<\infty 1 < p < ∞ , and let z z z be a nonnegative function on R N \mathbb{R}^{N} R N which vanishes at infinity and whose distributional gradient belongs to L p ( R N ) L^{p}(\mathbb{R}^{N}) L p ( R N ) . If ∥ ∇ z ∥ p = ∥ ∇ z ∗ ∥ p \lVert\nabla z\rVert_p=\lVert\nabla z^*\rVert_p ∥ ∇ z ∥ p = ∥ ∇ z ∗ ∥ p , where z ∗ z^* z ∗ is the Schwarz rearrangement of z z z , then there exists a translation τ \tau τ of R N \mathbb{R}^{N} R N such that
∥ z − z ∗ ∘ τ ∥ q ≤ K N 1 / q ∥ z ∥ q 1 / N ′ ∥ z χ C ∥ q 1 / N (45) \lVert z-z^*\circ\tau\rVert_q\leq K_N^{1/q}\lVert z\rVert_q^{1/N'}\lVert z\chi_C\rVert_q^{1/N} \tag*{(45)} ∥ z − z ∗ ∘ τ ∥ q ≤ K N 1/ q ∥ z ∥ q 1/ N ′ ∥ z χ C ∥ q 1/ N (45) for every q ≥ 1 q\geq1 q ≥ 1 with z ∈ L q ( R N ) z\in L^q(\mathbb{R}^{N}) z ∈ L q ( R N ) . Here N ′ = N / ( N − 1 ) N'=N/(N-1) N ′ = N / ( N − 1 ) , K N = 2 ω N − 1 / ω N K_N=2\omega_{N-1}/\omega_N K N = 2 ω N − 1 / ω N and
C = { x ∈ R N : 0 < z ( x ) < ess sup z and ∣ ∇ z ( x ) ∣ = 0 } . C=\{x\in\mathbb{R}^{N}:0<z(x)<\operatorname{ess\,sup}z\ \text{and}\ |\nabla z(x)|=0\}. C = { x ∈ R N : 0 < z ( x ) < ess sup z and ∣∇ z ( x ) ∣ = 0 } .
As observed in [16 ] right after the statement of this theorem, see also [16 Lemma 3.3] , under the assumptions of Lemma B.4 the function z z z may be replaced by z ∗ z^* z ∗ on the right-hand side of (B.3) and in the definition of C C C . In particular, if z ∈ L q ( R N ) z\in L^q(\mathbb{R}^{N}) z ∈ L q ( R N ) for some q ≥ 1 q\geq1 q ≥ 1 and the set
C ∗ : = { x ∈ R N : 0 < z ∗ ( x ) < ess sup z and ∣ ∇ z ∗ ( x ) ∣ = 0 } C^*:=\{x\in\mathbb{R}^{N}:0<z^*(x)<\operatorname{ess\,sup}z\ \text{and}\ |\nabla z^*(x)|=0\} C ∗ := { x ∈ R N : 0 < z ∗ ( x ) < ess sup z and ∣∇ z ∗ ( x ) ∣ = 0 }
has measure zero, then z = z ∗ ∘ τ z=z^*\circ\tau z = z ∗ ∘ τ a.e. on R N \mathbb{R}^{N} R N .
Theorem B.5. Let N ≥ 3 N\geq3 N ≥ 3 , assume (F1)–(F3), and let μ > 0 \mu>0 μ > 0 . Then every u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) has the form
u = ε W ( ∣ ⋅ − x 0 ∣ ) , ε ∈ { − 1 , 1 } , x 0 ∈ R N , u=\varepsilon W(|\cdot-x_0|),\qquad\varepsilon\in\{-1,1\},\qquad x_0\in\mathbb{R}^{N}, u = ε W ( ∣ ⋅ − x 0 ∣ ) , ε ∈ { − 1 , 1 } , x 0 ∈ R N , where W ∈ C 2 ( [ 0 , ∞ ) ) W\in C^2([0,\infty)) W ∈ C 2 ([ 0 , ∞ )) , W ( r ) > 0 W(r)>0 W ( r ) > 0 for r ≥ 0 r\geq0 r ≥ 0 and W ′ ( r ) < 0 W'(r)<0 W ′ ( r ) < 0 for r > 0 r>0 r > 0 . Moreover, u ( ⋅ + x 0 ) ∈ C ( μ ) ∩ X r u(\cdot+x_0)\in C(\mu)\cap X_r u ( ⋅ + x 0 ) ∈ C ( μ ) ∩ X r and ∥ u ( ⋅ + x 0 ) ∥ 2 = ∥ u ∥ 2 \lVert u(\cdot+x_0)\rVert_2=\lVert u\rVert_2 ∥ u ( ⋅ + x 0 ) ∥ 2 = ∥ u ∥ 2 .
Proof. Let u ∈ C ( μ ) u\in C(\mu) u ∈ C ( μ ) . By Lemma B.1, we may assume that u > 0 u>0 u > 0 and prove the theorem with ε = 1 \varepsilon=1 ε = 1 (if u < 0 u<0 u < 0 , the same argument applies to − u -u − u and to the nonlinearity s ↦ − f ( − s ) s\mapsto-f(-s) s ↦ − f ( − s ) ). By Lemma B.2, u ∗ ∈ C ( μ ) u^*\in C(\mu) u ∗ ∈ C ( μ ) and
∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 . (46) \lVert\nabla u^*\rVert_2=\lVert\nabla u\rVert_2. \tag*{(46)} ∥ ∇ u ∗ ∥ 2 = ∥ ∇ u ∥ 2 . (46)
Write u ∗ ( x ) = W ( ∣ x ∣ ) u^*(x)=W(|x|) u ∗ ( x ) = W ( ∣ x ∣ ) , where W ≥ 0 W\geq0 W ≥ 0 is nonincreasing. Lemma B.3 applies to u ∗ u^* u ∗ with ε = 1 \varepsilon=1 ε = 1 and gives
W ∈ C 2 ( [ 0 , ∞ ) ) , W ( r ) > 0 for r ≥ 0 , W ′ ( r ) < 0 for r > 0. (47) W\in C^2([0,\infty)),\qquad W(r)>0\ \text{for }r\geq0,\qquad W'(r)<0\ \text{for }r>0. \tag*{(47)} W ∈ C 2 ([ 0 , ∞ )) , W ( r ) > 0 for r ≥ 0 , W ′ ( r ) < 0 for r > 0. (47)
We now show that u u u is a translate of u ∗ u^* u ∗ by means of Lemma B.4, applied to z = u z=u z = u with p = q = 2 p=q=2 p = q = 2 . Its hypotheses hold: u ∈ L 2 ( R N ) u\in L^2(\mathbb{R}^{N}) u ∈ L 2 ( R N ) is nonnegative, hence it vanishes at infinity, ∇ u ∈ L 2 ( R N ) \nabla u\in L^2(\mathbb{R}^{N}) ∇ u ∈ L 2 ( R N ) , and ∥ ∇ u ∥ 2 = ∥ ∇ u ∗ ∥ 2 \lVert\nabla u\rVert_2=\lVert\nabla u^*\rVert_2 ∥ ∇ u ∥ 2 = ∥ ∇ u ∗ ∥ 2 by (B.4). Moreover, ess sup u = ∥ u ∗ ∥ ∞ = W ( 0 ) \operatorname{ess\,sup}u=\lVert u^*\rVert_\infty=W(0) ess sup u = ∥ u ∗ ∥ ∞ = W ( 0 ) , and since W W W is strictly decreasing by (B.5),
{ x ∈ R N : 0 < u ∗ ( x ) < W ( 0 ) } = R N ∖ { 0 } \{x\in\mathbb{R}^{N}:0<u^*(x)<W(0)\}=\mathbb{R}^{N}\setminus\{0\} { x ∈ R N : 0 < u ∗ ( x ) < W ( 0 )} = R N ∖ { 0 }
and ∇ u ∗ ( x ) = W ′ ( ∣ x ∣ ) x / ∣ x ∣ ≠ 0 \nabla u^*(x)=W'(|x|)x/|x|\ne0 ∇ u ∗ ( x ) = W ′ ( ∣ x ∣ ) x /∣ x ∣ = 0 for x ≠ 0 x\ne0 x = 0 . Hence the set C ∗ C^* C ∗ associated with z = u z=u z = u is empty, and the observation following Lemma B.4 gives a translation τ \tau τ such that u = u ∗ ∘ τ u=u^*\circ\tau u = u ∗ ∘ τ a.e. on R N \mathbb{R}^{N} R N . Writing τ ( x ) = x − x 0 \tau(x)=x-x_0 τ ( x ) = x − x 0 with x 0 ∈ R N x_0\in\mathbb{R}^{N} x 0 ∈ R N , we get u = u ∗ ( ⋅ − x 0 ) u=u^*(\cdot-x_0) u = u ∗ ( ⋅ − x 0 ) a.e. on R N \mathbb{R}^{N} R N . Both sides are continuous, hence u ( x ) = u ∗ ( x − x 0 ) = W ( ∣ x − x 0 ∣ ) > 0 u(x)=u^*(x-x_0)=W(|x-x_0|)>0 u ( x ) = u ∗ ( x − x 0 ) = W ( ∣ x − x 0 ∣ ) > 0 for every x ∈ R N x\in\mathbb{R}^{N} x ∈ R N , with W W W as in (B.5).
Finally, the action, the constraint and the L 2 L^2 L 2 -norm are invariant under translations, hence u ( ⋅ + x 0 ) = W ( ∣ ⋅ ∣ ) ∈ C ( μ ) ∩ X r u(\cdot+x_0)=W(|\cdot|)\in C(\mu)\cap X_r u ( ⋅ + x 0 ) = W ( ∣ ⋅ ∣ ) ∈ C ( μ ) ∩ X r and ∥ u ( ⋅ + x 0 ) ∥ 2 = ∥ u ∥ 2 \lVert u(\cdot+x_0)\rVert_2=\lVert u\rVert_2 ∥ u ( ⋅ + x 0 ) ∥ 2 = ∥ u ∥ 2 .
The normalized action ground states of Theorems 1.1 and 1.2 and the energy ground states of Theorem 1.3 therefore have the profile described in Theorem B.5 , and centering them does not change their mass. More generally,
{ ∥ u ∥ 2 2 : u ∈ C ( μ ) } = { ∥ u ∥ 2 2 : u ∈ C ( μ ) ∩ X r } \left\{\lVert u\rVert_{2}^{2}: u \in\mathcal{C}(\mu)\right\} = \left\{\lVert u\rVert_{2}^{2}: u \in\mathcal{C}(\mu) \cap X_{r}\right\} { ∥ u ∥ 2 2 : u ∈ C ( μ ) } = { ∥ u ∥ 2 2 : u ∈ C ( μ ) ∩ X r }
for every μ > 0 \mu> 0 μ > 0 ,
which justifies the use of centered radial representatives in the analysis of the masses.
Acknowledgements
J. M. was partly supported by the National Science Centre, Poland, under the research grant 2025/57/B/ST1/04804 and the research grant 2023/51/B/ST1/00968.
Data availability. The manuscript has no associated data.
Conflict of interest. The authors declare no conflict of interest.
References
J. Mederski
Institute of Mathematics, Polish Academy of Sciences, ul. Śniadeckich 8, 00-656 Warsaw, Poland and Faculty of Mathematics and Computer Science, Nicolaus Copernicus University, ul. Chopina 12/18, 87-100 Toruń, Poland
Email address: jmederski@impan.pl
X. Zeng
Institute of Mathematics, Polish Academy of Sciences, ul. Śniadeckich 8, 00-656 Warsaw, Poland and Faculty of Mathematics, Informatics and Mechanics, University of Warsaw, ul. Banacha 2, 02-097 Warsaw, Poland
Email address: xzeng@impan.pl