Introduction and main results
In this paper we continue the study, started in [24 ] , of sign-changing blowing-up solutions of the Brezis–Nirenberg problem
{ − Δ u = ∣ u ∣ 2 ∗ − 2 u + ε u in Ω , u = 0 on ∂ Ω , (1) \begin{cases}
-\Delta u = |u|^{2^{*}-2}u+\varepsilon u & \text{in }\Omega,\\
u = 0 & \text{on }\partial\Omega, \tag*{(1)}
\end{cases} { − Δ u = ∣ u ∣ 2 ∗ − 2 u + ε u u = 0 in Ω , on ∂ Ω , (1)
where Ω ⊂ R N \Omega\subset\mathbb{R}^{N} Ω ⊂ R N with N ≥ 5 N \ge5 N ≥ 5 is a smooth bounded domain, ε > 0 \varepsilon> 0 ε > 0 is a small parameter and, as usual, 2 ∗ = 2 N N − 2 2^{*}=\frac{2N}{N-2} 2 ∗ = N − 2 2 N is the critical exponent for the Sobolev embedding. Problem (1) was introduced by Brezis and Nirenberg in their celebrated paper [5 ] .
A particular feature of (1) , due to the critical behaviour of the nonlinearity, is the possible existence of solutions which blow up at one or more points in the domain as the parameter ε \varepsilon ε approaches 0 in dimension N ≥ 4 N \ge4 N ≥ 4 or the parameter ε \varepsilon ε approaches a positive number λ ∗ \lambda_{*} λ ∗ in dimension N = 3 N = 3 N = 3 .
The profile of the positive blowing-up solutions and the construction of such positive solutions predicted by the asymptotic analysis have been the subject of a wide literature (see for instance [7 ] , [12 ] , [13 ] , [14 ] , [20 ] , [22 ] , [23 ] , [29 ] and references therein).
Instead, the profile of the sign-changing solutions is far from being completely understood. As far as we know the complete scenario of sign-changing solutions has been obtained only in the radial case by Esposito, Ghoussoub, Pistoia and Vaira [8 ] in dimensions N ≥ 7 N \ge7 N ≥ 7 and by Amadori, Gladiali, Grossi, Pistoia and Vaira [1 ] in lower dimensions.
Sign-changing solutions for (1) have been built by Micheletti and Pistoia [21 ] , Iacopetti and Vaira [15 , 16 ] , Li, Vaira, Wei and Wu [18 , 19 ] , Pistoia and Vaira in [27 ] and more recently in [24 ] and [25 ] .
Indeed, in [24 ] Pistoia, Rago and Vaira constructed sign-changing solutions with several concentration points in dimension N = 4 N=4 N = 4 , showing that the locations and the relative scales of the bubbles are governed by a signed Green–Robin interaction matrix. They formulated an abstract existence criterion in terms of a simple positive eigenvalue with a positive eigenvector and of a stable critical set and applied it to a positive-negative pair in a general domain and to several symmetric multi-peak configurations: alternating regular polygons, orthogonal polygons, one central peak surrounded by peaks of the opposite sign and aligned patterns with three, four and five peaks. For the two-peak solution they also proved that it has exactly two nodal domains and, under a natural balance condition and the connectedness of the boundary, that the closure of its nodal set meets the boundary.
Instead, in [25 ] the authors considered the configuration with a positive peak at the centre of the unit ball surrounded by k k k negative peaks at the vertices of a regular polygon. In dimension N = 5 N=5 N = 5 such solutions exist for every k ≥ 2 k\geq2 k ≥ 2 and the polygon approaches the boundary of the ball as k → + ∞ k\to+\infty k → + ∞ at distance O ( 1 k ) O\left(\frac{1}{k}\right) O ( k 1 ) .
On the other hand, for N ≥ 6 N\geq6 N ≥ 6 there are two distinct families whenever 2 ≤ k ≤ 6 2\leq k\leq6 2 ≤ k ≤ 6 , whereas for k k k large no solution of this type exists with parameters in a compact set. Moreover, for 2 ≤ k ≤ 6 2\leq k\leq6 2 ≤ k ≤ 6 the nodal set of these solutions always meets the boundary of the ball.
In dimension N ≥ 5 N\geq5 N ≥ 5 , sign-changing multi-bubble solutions for critical and supercritical problems were studied by Micheletti and Pistoia in [21 ] . In particular, for the Brezis-Nirenberg perturbation in dimensions N ≥ 5 N\geq5 N ≥ 5 , their reduction leads to a finite-dimensional functional depending simultaneously on the concentration points and on the individual scales. In order to understand the difficulties and also the main results of this paper, we introduce some notation used throughout the paper.
Let G Ω ( x , y ) G_{\Omega}(x,y) G Ω ( x , y ) be the Green function of − Δ -\Delta − Δ in Ω \Omega Ω with Dirichlet boundary conditions and let H H H be its regular part, i.e.
G Ω ( x , y ) : = 1 ω N ∣ x − y ∣ N − 2 − H ( x , y ) G_{\Omega}(x,y):=\frac{1}{\omega_N\lvert x-y\rvert^{N-2}}-H(x,y) G Ω ( x , y ) := ω N ∣ x − y ∣ N − 2 1 − H ( x , y )
where ω N : = ( N − 2 ) meas ( S N − 1 ) \omega_N:=(N-2)\operatorname{meas}(S^{N-1}) ω N := ( N − 2 ) meas ( S N − 1 ) , S N − 1 S^{N-1} S N − 1 is the unit sphere of R N \mathbb{R}^{N} R N and for every y ∈ Ω y\in\Omega y ∈ Ω ,
− Δ x H ( x , y ) = 0 in Ω , H ( x , y ) = 1 ω N ∣ x − y ∣ N − 2 on ∂ Ω . -\Delta_x H(x,y)=0\text{ in }\Omega,\qquad H(x,y)=\frac{1}{\omega_N\lvert x-y\rvert^{N-2}}\text{ on }\partial\Omega. − Δ x H ( x , y ) = 0 in Ω , H ( x , y ) = ω N ∣ x − y ∣ N − 2 1 on ∂ Ω.
The leading term of the regular part, τ Ω ( x ) = H ( x , x ) \tau_{\Omega}(x)=H(x,x) τ Ω ( x ) = H ( x , x ) , is called the Robin function of Ω \Omega Ω .
Given an integer k ≥ 2 k\geq2 k ≥ 2 we set
Ω k ∗ : = { ξ : = ( ξ 1 , … , ξ k ) ∈ Ω k : ξ i ≠ ξ j for i ≠ j } \Omega_k^*:=\left\{\xi:=(\xi_1,\ldots,\xi_k)\in\Omega^k:\ \xi_i\ne\xi_j\text{ for }i\ne j\right\} Ω k ∗ := { ξ := ( ξ 1 , … , ξ k ) ∈ Ω k : ξ i = ξ j for i = j }
and, for any ξ ∈ Ω k ∗ \xi\in\Omega_k^* ξ ∈ Ω k ∗ and σ = ( σ 1 , … , σ k ) ∈ { − 1 , 1 } k \sigma=(\sigma_1,\ldots,\sigma_k)\in\{-1,1\}^k σ = ( σ 1 , … , σ k ) ∈ { − 1 , 1 } k , we define the symmetric matrix
M σ ( ξ ) = ( τ Ω ( ξ 1 ) − σ 1 σ 2 G Ω ( ξ 1 , ξ 2 ) ⋯ − σ 1 σ k G Ω ( ξ 1 , ξ k ) − σ 1 σ 2 G Ω ( ξ 1 , ξ 2 ) τ Ω ( ξ 2 ) ⋯ − σ 2 σ k G Ω ( ξ 2 , ξ k ) ⋮ ⋮ ⋱ ⋮ − σ 1 σ k G Ω ( ξ 1 , ξ k ) − σ 2 σ k G Ω ( ξ 2 , ξ k ) ⋯ τ Ω ( ξ k ) ) . M^\sigma(\xi)=\begin{pmatrix}
\tau_{\Omega}(\xi_1) & -\sigma_1\sigma_2G_{\Omega}(\xi_1,\xi_2) & \cdots& -\sigma_1\sigma_kG_{\Omega}(\xi_1,\xi_k) \\
-\sigma_1\sigma_2G_{\Omega}(\xi_1,\xi_2) & \tau_{\Omega}(\xi_2) & \cdots& -\sigma_2\sigma_kG_{\Omega}(\xi_2,\xi_k) \\
\vdots& \vdots& \ddots& \vdots\\
-\sigma_1\sigma_kG_{\Omega}(\xi_1,\xi_k) & -\sigma_2\sigma_kG_{\Omega}(\xi_2,\xi_k) & \cdots& \tau_{\Omega}(\xi_k)
\end{pmatrix}. M σ ( ξ ) = τ Ω ( ξ 1 ) − σ 1 σ 2 G Ω ( ξ 1 , ξ 2 ) ⋮ − σ 1 σ k G Ω ( ξ 1 , ξ k ) − σ 1 σ 2 G Ω ( ξ 1 , ξ 2 ) τ Ω ( ξ 2 ) ⋮ − σ 2 σ k G Ω ( ξ 2 , ξ k ) ⋯ ⋯ ⋱ ⋯ − σ 1 σ k G Ω ( ξ 1 , ξ k ) − σ 2 σ k G Ω ( ξ 2 , ξ k ) ⋮ τ Ω ( ξ k ) .
The matrix M σ ( ξ ) M^\sigma(\xi) M σ ( ξ ) is the Green–Robin interaction matrix: its diagonal entries describe the interaction of each bubble with the boundary, while the off-diagonal terms describe the interaction between distinct bubbles.
In our notation, the leading part of the reduced energy in [21 ] is
F k σ ( d , ξ ) = a N 2 [ ∑ j = 1 k τ Ω ( ξ j ) d j N − 2 − ∑ i , j = 1 i ≠ j k σ j σ i G Ω ( ξ i , ξ j ) d i N − 2 2 d j N − 2 2 ] − b N ∑ j = 1 k d j 2 (2) F_k^\sigma(d,\xi)=\frac{a_N}{2}\left[\sum_{j=1}^k \tau_\Omega(\xi_j)d_j^{N-2}-\sum_{\substack{i,j=1\\i\ne j}}^k \sigma_j\sigma_iG_\Omega(\xi_i,\xi_j)d_i^{\frac{N-2}{2}}d_j^{\frac{N-2}{2}}\right]-b_N\sum_{j=1}^k d_j^2 \tag*{(2)} F k σ ( d , ξ ) = 2 a N j = 1 ∑ k τ Ω ( ξ j ) d j N − 2 − i , j = 1 i = j ∑ k σ j σ i G Ω ( ξ i , ξ j ) d i 2 N − 2 d j 2 N − 2 − b N j = 1 ∑ k d j 2 (2)
or equivalently, setting κ j : = d j N − 2 2 \kappa_j:=d_j^{\frac{N-2}{2}} κ j := d j 2 N − 2 ,
F k σ ( κ , ξ ) = a N 2 ⟨ M σ ( ξ ) κ , κ ⟩ − b N ∑ j = 1 k κ j 4 N − 2 (3) F_k^\sigma(\kappa,\xi)=\frac{a_N}{2}\langle M^\sigma(\xi)\kappa,\kappa\rangle-b_N\sum_{j=1}^k\kappa_j^{\frac{4}{N-2}} \tag*{(3)} F k σ ( κ , ξ ) = 2 a N ⟨ M σ ( ξ ) κ , κ ⟩ − b N j = 1 ∑ k κ j N − 2 4 (3)
where a N , b N > 0 a_N,b_N>0 a N , b N > 0 and explicitly given in (8) . Thus the geometry of the domain and the nodal pattern enter the reduced problem simultaneously.
The general reduction allows one to prescribe arbitrary signs, but the critical point analysis of (2) is rather delicate because the location variables and all the relative scales remain coupled. This is one of the reasons why explicit constructions in natural geometries were obtained only for particular configurations. For example, in the ball, in [21 ] Micheletti and Pistoia constructed a family with an even number of bubbles arranged on a regular polygon with alternating signs; the symmetry reduces all the scales to one parameter and all the locations to one radial variable. The same paper also produces many sign-changing patterns in specially constructed dumbbell-type domains and also the case of one positive and one negative bubble in an arbitrary smooth bounded domain is considered. Here the difficulty is not the general reduction itself, but the analysis of its finite-dimensional functional and the associated critical points. In dimension N = 4 N=4 N = 4 the situation is different: the concentration scales are exponentially small in 1 / ε 1/\varepsilon 1/ ε and the leading equations for the relative scales become spectral. Hence the scale variables can be eliminated through the spectral structure of the interaction matrix which makes accessible configurations that are considerably harder to detect through the higher-dimensional reduced functional. In this paper we show that two of the multi-peak configurations constructed in [24 ] survive in dimensions N ≥ 5 N\geq5 N ≥ 5 : alternating regular polygons in annuli and two regular polygons lying in orthogonal coordinate planes.
For r > 0 r>0 r > 0 and k ≥ 2 k\geq2 k ≥ 2 we set θ j : = 2 π ( j − 1 ) k \theta_j:=\frac{2\pi(j-1)}{k} θ j := k 2 π ( j − 1 ) and
ξ j ( r ) : = r ( cos θ j , sin θ j , 0 , … , 0 ) , η j ( r ) : = r ( 0 , 0 , cos θ j , sin θ j , 0 , … , 0 ) , j = 1 , … , k . (4) \begin{aligned}
\xi_j(r)&:=r(\cos\theta_j,\sin\theta_j,0,\ldots,0),\\
\eta_j(r)&:=r(0,0,\cos\theta_j,\sin\theta_j,0,\ldots,0),\qquad j=1,\ldots,k. \tag*{(4)}
\end{aligned} ξ j ( r ) η j ( r ) := r ( cos θ j , sin θ j , 0 , … , 0 ) , := r ( 0 , 0 , cos θ j , sin θ j , 0 , … , 0 ) , j = 1 , … , k . (4)
Our main results are the following.
Theorem 1.1 (Symmetric multi-peak configurations). Let N ≥ 5 N\geq5 N ≥ 5 . For ε > 0 \varepsilon>0 ε > 0 sufficiently small, problem (1) has the following sign-changing solutions.
(a) Alternating regular polygons in an annulus. Let
Ω ρ = { x ∈ R N : ρ < ∣ x ∣ < 1 } (5) \Omega_\rho=\{x\in\mathbb{R}^N:\rho<|x|<1\} \tag*{(5)} Ω ρ = { x ∈ R N : ρ < ∣ x ∣ < 1 } (5) with 0 < ρ < 1 0<\rho<1 0 < ρ < 1 . For every even k ≥ 2 k\geq2 k ≥ 2 there is a solution
u ε = ∑ j = 1 k ( − 1 ) j + 1 P U δ ε , ξ j ( r ε ) + ϕ ε , δ ε = d ε ε 1 N − 4 and ∥ ϕ ε ∥ H 0 1 ( Ω ρ ) → 0 u_\varepsilon=\sum_{j=1}^k(-1)^{j+1}PU_{\delta_\varepsilon,\xi_j(r_\varepsilon)}+\phi_\varepsilon,\qquad\delta_\varepsilon=d_\varepsilon\varepsilon^{\frac{1}{N-4}}\quad\text{and}\quad\|\phi_\varepsilon\|_{H_0^1(\Omega_\rho)}\to0 u ε = j = 1 ∑ k ( − 1 ) j + 1 P U δ ε , ξ j ( r ε ) + ϕ ε , δ ε = d ε ε N − 4 1 and ∥ ϕ ε ∥ H 0 1 ( Ω ρ ) → 0
with k k k concentration points of alternating sign at the vertices of a regular k k k -gon in the ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) -plane. It is even in x 2 x_2 x 2 and in x 3 , … , x N x_3,\ldots,x_N x 3 , … , x N , and anti-invariant under the rotation R 2 π / k R_{2\pi/k} R 2 π / k of angle 2 π / k 2\pi/k 2 π / k in the ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) -plane: u ( R 2 π / k x ) = − u ( x ) u(R_{2\pi/k}x)=-u(x) u ( R 2 π / k x ) = − u ( x ) . Moreover, up to a subsequence, r ε → r 0 r_\varepsilon\to r_0 r ε → r 0 and d ε → d 0 > 0 d_\varepsilon\to d_0>0 d ε → d 0 > 0 , where r 0 ∈ ( ρ , 1 ) r_0\in(\rho,1) r 0 ∈ ( ρ , 1 ) is a global minimum point of the function Ψ k \Psi_k Ψ k defined in (17) .
(b) Two orthogonal regular polygons in the ball. Let Ω = B 1 ( 0 ) \Omega=\mathbb{B}_1(0) Ω = B 1 ( 0 ) be the unit ball and let 2 ≤ k ≤ k ∗ ( N ) 2\leq k\leq k^*(N) 2 ≤ k ≤ k ∗ ( N ) , where
k ∗ ( N ) : = { 5 if N = 5 , 6 , 4 if N ≥ 7. k^*(N):=
\begin{cases}
5 & \text{if } N=5,6,\\
4 & \text{if } N\geq7.
\end{cases} k ∗ ( N ) := { 5 4 if N = 5 , 6 , if N ≥ 7.
There is a solution
u ε = ∑ j = 1 k P U δ ε , ξ j ( r ε ) − ∑ j = 1 k P U δ ε , η j ( r ε ) + ϕ ε , with δ ε = d ε ε 1 N − 4 and ∥ ϕ ε ∥ H 0 1 ( B 1 ( 0 ) ) → 0 u_\varepsilon=\sum_{j=1}^{k}PU_{\delta_\varepsilon,\xi_j}(r_\varepsilon)-\sum_{j=1}^{k}PU_{\delta_\varepsilon,\eta_j}(r_\varepsilon)+\phi_\varepsilon,\qquad\text{with }\delta_\varepsilon=d_\varepsilon\varepsilon^{\frac{1}{N-4}}\text{ and }\|\phi_\varepsilon\|_{H_0^1(\mathbb{B}_1(0))}\to0 u ε = j = 1 ∑ k P U δ ε , ξ j ( r ε ) − j = 1 ∑ k P U δ ε , η j ( r ε ) + ϕ ε , with δ ε = d ε ε N − 4 1 and ∥ ϕ ε ∥ H 0 1 ( B 1 ( 0 )) → 0
with k k k positive concentration points at the vertices of a regular k k k -gon in the ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) -plane and k k k negative concentration points at the vertices of a regular k k k -gon in the ( x 3 , x 4 ) (x_3,x_4) ( x 3 , x 4 ) -plane. It is invariant under the rotations of angle 2 π / k 2\pi/k 2 π / k in each of the two planes, even with respect to ( x 2 , x 4 ) → ( − x 2 , − x 4 ) (x_2,x_4)\to(-x_2,-x_4) ( x 2 , x 4 ) → ( − x 2 , − x 4 ) , radial in ( x 5 , … , x N ) (x_5,\ldots,x_N) ( x 5 , … , x N ) and antisymmetric under the exchange of the two planes:
u ( ω , y , z ) = − u ( y , ω , z ) , y , ω ∈ R 2 , z ∈ R N − 4 . u(\omega,y,z)=-u(y,\omega,z),\qquad y,\omega\in\mathbb{R}^2,\ z\in\mathbb{R}^{N-4}. u ( ω , y , z ) = − u ( y , ω , z ) , y , ω ∈ R 2 , z ∈ R N − 4 .
Moreover, r ε → r 0 r_\varepsilon\to r_0 r ε → r 0 and d ε → d 0 > 0 d_\varepsilon\to d_0>0 d ε → d 0 > 0 , where r 0 ∈ ( 0 , 1 ) r_0\in(0,1) r 0 ∈ ( 0 , 1 ) is the unique global minimum point of the function Φ k , N \Phi_{k,N} Φ k , N defined in (30) (see Theorem 1.2 ).
Theorem 1.2 (Sharpness of k ∗ ( N ) k^*(N) k ∗ ( N ) ). Let N ≥ 5 N\geq5 N ≥ 5 and Ω = B 1 ( 0 ) \Omega=\mathbb{B}_1(0) Ω = B 1 ( 0 ) .
(i) If 2 ≤ k ≤ k ∗ ( N ) 2\leq k\leq k^*(N) 2 ≤ k ≤ k ∗ ( N ) , the limiting reduced functional, restricted to the two-parameter symmetric family, has exactly one critical point, which is non-degenerate; in particular the solutions of Theorem 1.1 (b) satisfy r ε → r 0 r_\varepsilon\to r_0 r ε → r 0 and d ε → d 0 d_\varepsilon\to d_0 d ε → d 0 as ε → 0 \varepsilon\to0 ε → 0 , where r 0 r_0 r 0 is the unique critical point of Φ k , N \Phi_{k,N} Φ k , N .
(ii) If k > k ∗ ( N ) k>k^*(N) k > k ∗ ( N ) , for every compact set K ⊂ ( 0 , 1 ) × ( 0 , ∞ ) K\subset(0,1)\times(0,\infty) K ⊂ ( 0 , 1 ) × ( 0 , ∞ ) there is ε K > 0 \varepsilon_K>0 ε K > 0 such that, for ε ∈ ( 0 , ε K ) \varepsilon\in(0,\varepsilon_K) ε ∈ ( 0 , ε K ) , problem (1.1) has no solution of the form given in Theorem 1.1 (b) with ( r , d ) ∈ K (r,d)\in K ( r , d ) ∈ K (see Theorem 4.10 ).
The limit scale d 0 d_0 d 0 is given explicitly in Proposition 2.9 in terms of the minimum value of Ψ k \Psi_k Ψ k , respectively Φ k , N \Phi_{k,N} Φ k , N . Moreover, Theorem 1.1 (a) extends to annuli the alternating polygons constructed in the ball in [21 ] (see Theorem 2.4 ) and we point out that no restriction on the (even) number of peaks is needed.
Instead, in Theorem 1.1 (b), the number of peaks is bounded: the restriction k ≤ k ∗ ( N ) k\leq k^*(N) k ≤ k ∗ ( N ) comes from the sign of the constant c k , N c_{k,N} c k , N in (32) , which measures the competition between the mutual attraction of the peaks of the same polygon and the repulsion between the two polygons (see Proposition 4.4 ) and it is optimal by Theorem 1.2 . In dimension four the analogous condition gives k ≤ 6 k\leq6 k ≤ 6 (see Remark 4.5).
Organization of the paper
In Section 2 we recall the Lyapunov–Schmidt reduction and we perform it in a class of symmetric functions; we show that for symmetric configurations the reduced energy is governed by a single function of the radius and we prove an abstract existence result (Proposition 2.9 ). In Section 3 we analyze the case of the annulus, proving Theorem 1.1 (a) and finally, in Section 4 we study the case of the two orthogonal regular polygons in the ball, proving Theorem 1.1 (b) and Theorem 1.2 .
Acknowledgments. The author is deeply grateful to Professor Angela Pistoia and Professor Giusi Vaira for many stimulating discussions, valuable insights and helpful suggestions, which greatly contributed to the development of this work. Moreover the author acknowledges support from the INdAM-GNAMPA project “Singolarità, non linearità e interazioni non locali: nuovi approcci analitici per equazioni differenziali e sistemi” (CUP E53C25002010001). The author acknowledges also the use of ChatGPT (OpenAI) to assist selected calculations. He has independently verified all mathematical arguments and calculations to take full responsibility for the content of the paper.
Preliminaries
First we recall the definition of a stable critical point we use in this paper (see [17 ] ).
Definition 2.1. Let f : D → R f:D\to\mathbb{R} f : D → R be a C 1 C^1 C 1 -function, where D ⊂ R N D\subset\mathbb{R}^N D ⊂ R N is an open set. We say that ξ 0 \xi_0 ξ 0 is a stable critical point of f f f if ∇ f ( ξ 0 ) = 0 \nabla f(\xi_0)=0 ∇ f ( ξ 0 ) = 0 and there exists a neighbourhood V ⋐ D V\Subset D V ⋐ D of ξ 0 \xi_0 ξ 0 such that ∇ f ( ξ ) ≠ 0 \nabla f(\xi)\ne0 ∇ f ( ξ ) = 0 for any ξ ∈ ∂ V \xi\in\partial V ξ ∈ ∂ V ,
∇ f ( ξ ) = 0 , ξ ∈ V ⟹ f ( ξ ) = f ( ξ 0 ) \nabla f(\xi)=0,\qquad\xi\in V\Longrightarrow f(\xi)=f(\xi_0) ∇ f ( ξ ) = 0 , ξ ∈ V ⟹ f ( ξ ) = f ( ξ 0 ) and
deg ( ∇ f , V , 0 ) ≠ 0 , \deg(\nabla f,V,0)\ne0, deg ( ∇ f , V , 0 ) = 0 , where deg \deg deg denotes the Brouwer degree.
Next, let us define the bubble (see [2 , 6 , 32 ] )
U δ , ξ ( x ) = α N δ N − 2 2 ( δ 2 + ∣ x − ξ ∣ 2 ) N − 2 2 , x , ξ ∈ R N , δ > 0 (6) U_{\delta,\xi}(x)=\alpha_N\frac{\delta^{\frac{N-2}{2}}}{\left(\delta^2+\lvert x-\xi\rvert^2\right)^{\frac{N-2}{2}}},\qquad x,\xi\in\mathbb{R}^N,\ \delta>0 \tag*{(6)} U δ , ξ ( x ) = α N ( δ 2 + ∣ x − ξ ∣ 2 ) 2 N − 2 δ 2 N − 2 , x , ξ ∈ R N , δ > 0 (6)
where α N = [ N ( N − 2 ) ] N − 2 4 \alpha_N=[N(N-2)]^{\frac{N-2}{4}} α N = [ N ( N − 2 ) ] 4 N − 2 . As is well-known, these are all the positive solutions of the problem
− Δ U = U N + 2 N − 2 , in R N . (7) -\Delta U=U^{\frac{N+2}{N-2}},\qquad\text{in }\mathbb{R}^N. \tag*{(7)} − Δ U = U N − 2 N + 2 , in R N . (7)
Moreover, we denote by P U δ , ξ PU_{\delta,\xi} P U δ , ξ the projection of the bubble into H 0 1 ( Ω ) H^1_0(\Omega) H 0 1 ( Ω ) , i.e.
Δ P U δ , ξ = Δ U δ , ξ in Ω , P U δ , ξ = 0 on ∂ Ω \Delta PU_{\delta,\xi}=\Delta U_{\delta,\xi}\quad\text{in }\Omega,\qquad PU_{\delta,\xi}=0\quad\text{on }\partial\Omega Δ P U δ , ξ = Δ U δ , ξ in Ω , P U δ , ξ = 0 on ∂ Ω
and we set U : = U 1 , 0 U:=U_{1,0} U := U 1 , 0 and
a N : = ∫ R N U 2 ∗ − 1 ( x ) d x , b N : = 1 2 ∫ R N U 2 ( x ) d x (8) a_N:=\int_{\mathbb{R}^N}U^{2^*-1}(x)\,dx,\qquad b_N:=\frac{1}{2}\int_{\mathbb{R}^N}U^2(x)\,dx \tag*{(8)} a N := ∫ R N U 2 ∗ − 1 ( x ) d x , b N := 2 1 ∫ R N U 2 ( x ) d x (8)
which are finite since N ≥ 5 N\geq5 N ≥ 5 .
Moreover, in [21 ] , the following result is proved.
Theorem 2.2. ([21 Theorem 3.4] ) Let ( d 0 , ξ 0 ) ∈ ( 0 , + ∞ ) k × Ω k ∗ (d_0,\xi_0)\in(0,+\infty)^k\times\Omega_k^* ( d 0 , ξ 0 ) ∈ ( 0 , + ∞ ) k × Ω k ∗ be a stable critical point of the function F k 0 ( d , ξ ) F_k^0(d,\xi) F k 0 ( d , ξ ) defined in (2) in the sense of Definition 2.1. Then for ε \varepsilon ε small enough, problem (1) has a family of solutions
u ε = ∑ j = 1 k σ j P U δ ε , j , ξ ε , j + ϕ ε , ∥ ϕ ε ∥ H 0 1 ( Ω ) → 0 , u_\varepsilon=\sum_{j=1}^{k}\sigma_jPU_{\delta_{\varepsilon,j},\xi_{\varepsilon,j}}+\phi_\varepsilon,\qquad\lVert\phi_\varepsilon\rVert_{H^1_0(\Omega)}\to0, u ε = j = 1 ∑ k σ j P U δ ε , j , ξ ε , j + ϕ ε , ∥ ϕ ε ∥ H 0 1 ( Ω ) → 0 ,
such that δ j ε = d j ε 1 N − 4 \delta_j^\varepsilon=d_j\varepsilon^{\frac{1}{N-4}} δ j ε = d j ε N − 4 1 ; moreover, up to a subsequence d j ε → d j ∗ d_j^\varepsilon\to d_j^* d j ε → d j ∗ and ξ j ε → ξ j ∗ \xi_j^\varepsilon\to\xi_j^* ξ j ε → ξ j ∗ where ( d ∗ , ξ ∗ ) (d^*,\xi^*) ( d ∗ , ξ ∗ ) is a critical point of F k σ F_k^\sigma F k σ with F k σ ( d 0 , ξ 0 ) = F k σ ( d ∗ , ξ ∗ ) F_k^\sigma(d_0,\xi_0)=F_k^\sigma(d^*,\xi^*) F k σ ( d 0 , ξ 0 ) = F k σ ( d ∗ , ξ ∗ ) .
Theorem 2.2 establishes a general criterion to find a sign-changing solution of problem (1) in dimension N ≥ 5 N\geq5 N ≥ 5 . It remains to find domains and sign patterns for which the function F k σ F_k^\sigma F k σ has a stable critical point. In [21 ] this is done in two situations. The first one is a dumbbell-shaped domain with thin handles, in which problem (1) has many sign-changing solutions with many nodal domains.
Theorem 2.3. ([21 ] , Theorem 1.4) Assume N ≥ 5 N\geq5 N ≥ 5 and let k ≥ 2 k\geq2 k ≥ 2 be a fixed integer. There exists a contractible domain Ω \Omega Ω such that for ε \varepsilon ε small enough, problem (1) has at least 2 k − 1 − 1 2^{k-1}-1 2 k − 1 − 1 pairs of sign-changing solutions that blow up positively at h h h points and blow up negatively at k − h k-h k − h points for any h = 1 , … , k − 1 h=1,\ldots,k-1 h = 1 , … , k − 1 .
The second one is the ball, with peaks of alternating signs.
Theorem 2.4. ([21 ] , Theorem 1.5) Assume Ω \Omega Ω is the unit ball in R N \mathbb{R}^N R N with N ≥ 5 N\geq5 N ≥ 5 . Let h ≥ 1 h\geq1 h ≥ 1 . Then for ε \varepsilon ε small enough, problem (1) has at least a pair of sign-changing solutions that blow up positively at h h h points and blow up negatively at h h h points.
In both cases the symmetries are used to reduce the problem to a function in which the location and the scale variables are not fully coupled. Our aim is to find other configurations of sign-changing solutions, extending some of the results proved in dimension N = 4 N=4 N = 4 in [24 ] . We recall that the solutions of (1) are the critical points of the following functional
J ε ( u ) = 1 2 ∫ Ω ∣ ∇ u ∣ 2 d x − ε 2 ∫ Ω u 2 d x − 1 2 ∗ ∫ Ω ∣ u ∣ 2 ∗ d x , u ∈ H 0 1 ( Ω ) . J_\varepsilon(u)=\frac{1}{2}\int_\Omega|\nabla u|^2\,dx-\frac{\varepsilon}{2}\int_\Omega u^2\,dx-\frac{1}{2^*}\int_\Omega|u|^{2^*}\,dx,\qquad u\in H_0^1(\Omega). J ε ( u ) = 2 1 ∫ Ω ∣∇ u ∣ 2 d x − 2 ε ∫ Ω u 2 d x − 2 ∗ 1 ∫ Ω ∣ u ∣ 2 ∗ d x , u ∈ H 0 1 ( Ω ) .
The symmetric setting
Theorem 2.2 requires a stable critical point of F k σ F_k^\sigma F k σ in the whole space ( 0 , + ∞ ) k × Ω k ∗ (0,+\infty)^k\times\Omega_k^* ( 0 , + ∞ ) k × Ω k ∗ , whereas the configurations of Theorem 1.1 are critical points of F k σ F_k^\sigma F k σ restricted to a two-dimensional set of symmetric parameters and the stability, in the sense of Definition 2.1, is not inherited by the restriction. For this reason we perform the reduction directly in a space of symmetric functions, in which only two parameters survive. We treat the two parts of Theorem 1.1 at the same time.
The two configurations.
(A) Alternating regular polygons in an annulus. Let us consider the annulus Ω ρ \Omega_\rho Ω ρ as in (5) , an even integer k ≥ 2 k\geq2 k ≥ 2 and we set I : = ( ρ , 1 ) I:=(\rho,1) I := ( ρ , 1 ) and ℓ : = k \ell:=k ℓ := k .
Denoting by R R R the rotation of angle 2 π k \frac{2\pi}{k} k 2 π in the ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) -plane, the space of symmetric functions is
H s : = { u ∈ H 0 1 ( Ω ρ ) : u ( R x ) = − u ( x ) , u is even in x h for h = 2 , … , N } . \mathcal{H}_s:=\left\{u\in H_0^1(\Omega_\rho):u(Rx)=-u(x),\ u\text{ is even in }x_h\text{ for }h=2,\ldots,N\right\}. H s := { u ∈ H 0 1 ( Ω ρ ) : u ( R x ) = − u ( x ) , u is even in x h for h = 2 , … , N } .
(B) Two orthogonal regular polygons in the ball. Let Ω = B 1 ( 0 ) \Omega=B_1(0) Ω = B 1 ( 0 ) , k ≥ 2 k\geq2 k ≥ 2 , I : = ( 0 , 1 ) I:=(0,1) I := ( 0 , 1 ) and ℓ : = 2 k \ell:=2k ℓ := 2 k . We write
x = ( y , w , z ) ∈ R 2 × R 2 × R N − 4 , x=(y,w,z)\in\mathbb{R}^2\times\mathbb{R}^2\times\mathbb{R}^{N-4}, x = ( y , w , z ) ∈ R 2 × R 2 × R N − 4 ,
with y = ( x 1 , x 2 ) y=(x_1,x_2) y = ( x 1 , x 2 ) , w = ( x 3 , x 4 ) w=(x_3,x_4) w = ( x 3 , x 4 ) , z = ( x 5 , … , x N ) z=(x_5,\ldots,x_N) z = ( x 5 , … , x N ) . The bubbles are centred at ξ 1 ( r ) , … , ξ k ( r ) \xi_1(r),\ldots,\xi_k(r) ξ 1 ( r ) , … , ξ k ( r ) with a positive sign + 1 +1 + 1 and at ξ k + j ( r ) : = η j ( r ) \xi_{k+j}(r):=\eta_j(r) ξ k + j ( r ) := η j ( r ) , j = 1 , … , k j=1,\ldots,k j = 1 , … , k , with negative sign − 1 -1 − 1 ; that is, σ j = 1 \sigma_j=1 σ j = 1 for j ≤ k j\leq k j ≤ k and σ j = − 1 \sigma_j=-1 σ j = − 1 for j > k j>k j > k .
We introduce the following orthogonal maps
R 1 ( y , w , z ) = ( e 2 π i k y , w , z ) , R 2 ( y , w , z ) = ( y , e 2 π i k w , z ) , S ( y , w , z ) = ( y ‾ , w ‾ , z ) , R_1(y,w,z)=\left(e^{\frac{2\pi i}{k}}y,w,z\right),\qquad R_2(y,w,z)=\left(y,e^{\frac{2\pi i}{k}}w,z\right),\qquad S(y,w,z)=(\overline{y},\overline{w},z), R 1 ( y , w , z ) = ( e k 2 π i y , w , z ) , R 2 ( y , w , z ) = ( y , e k 2 π i w , z ) , S ( y , w , z ) = ( y , w , z ) ,
Q ^ ( y , w , z ) = ( y , w , Q z ) , with Q ∈ O ( N − 4 ) and T ( y , w , z ) = ( w , y , z ) . \widehat{Q}(y,w,z)=(y,w,Qz),\quad\text{with }Q\in O(N-4)\text{ and }T(y,w,z)=(w,y,z). Q ( y , w , z ) = ( y , w , Q z ) , with Q ∈ O ( N − 4 ) and T ( y , w , z ) = ( w , y , z ) .
Therefore, the space of symmetric functions is
H s : = { u ∈ H 0 1 ( B 1 ( 0 ) ) : u ∘ R 1 = u ∘ R 2 = u ∘ S = u ∘ Q ^ = u ∀ Q ∈ O ( N − 4 ) , u ∘ T = − u } . (9) \mathcal{H}_s := \left\{u \in H_0^1(\mathbb{B}_1(0)): u\circ R_1=u\circ R_2=u\circ S=u\circ\widehat{Q}=u\ \forall Q\in O(N-4),\ u\circ T=-u\right\}. \tag*{(9)} H s := { u ∈ H 0 1 ( B 1 ( 0 )) : u ∘ R 1 = u ∘ R 2 = u ∘ S = u ∘ Q = u ∀ Q ∈ O ( N − 4 ) , u ∘ T = − u } . (9)
In other words, u ∈ H s u\in\mathcal{H}_s u ∈ H s if and only if u u u is invariant under the rotations of angle 2 π k \frac{2\pi}{k} k 2 π in each of the two coordinate planes, even with respect to ( x 2 , x 4 ) ↦ ( − x 2 , − x 4 ) (x_2,x_4)\mapsto(-x_2,-x_4) ( x 2 , x 4 ) ↦ ( − x 2 , − x 4 ) , radial in z z z , and odd under the exchange of the two planes, u ( w , y , z ) = − u ( y , w , z ) u(w,y,z)=-u(y,w,z) u ( w , y , z ) = − u ( y , w , z ) .
In both cases the maps appearing in the definition of H s \mathcal{H}_s H s are orthogonal and leave Ω \Omega Ω invariant; hence, the critical points of J ε ∣ H s J_\varepsilon|_{\mathcal{H}_s} J ε ∣ H s are critical points of J ε J_\varepsilon J ε .
We look for solutions of the form
u ε = V y + ϕ , V y : = ∑ j = 1 ℓ σ j P U δ , ξ j ( r ) , δ = d ε 1 N − 4 , (10) u_\varepsilon=V_y+\phi,\qquad V_y:=\sum_{j=1}^{\ell}\sigma_jPU_{\delta,\xi_j}(r),\qquad\delta=d\varepsilon^{\frac{1}{N-4}}, \tag*{(10)} u ε = V y + ϕ , V y := j = 1 ∑ ℓ σ j P U δ , ξ j ( r ) , δ = d ε N − 4 1 , (10)
where y = ( r , d ) y=(r,d) y = ( r , d ) lies in a compact subset of M : = I × ( 0 , ∞ ) \mathcal{M}:=I\times(0,\infty) M := I × ( 0 , ∞ ) and ϕ ∈ H s \phi\in\mathcal{H}_s ϕ ∈ H s satisfies suitable orthogonality conditions and it is such that ∥ ϕ ∥ H 0 1 ( Ω ) → 0 \|\phi\|_{H_0^1(\Omega)}\to0 ∥ ϕ ∥ H 0 1 ( Ω ) → 0 as ε → 0 \varepsilon\to0 ε → 0 . For η > 0 \eta>0 η > 0 small we set
P η : = { r ∈ I : dist ( r , ∂ I ) ≥ η } × [ η , η − 1 ] ⊂ M . \mathcal{P}_\eta:=\{r\in I:\operatorname{dist}(r,\partial I)\ge\eta\}\times[\eta,\eta^{-1}]\subset\mathcal{M}. P η := { r ∈ I : dist ( r , ∂ I ) ≥ η } × [ η , η − 1 ] ⊂ M .
It is immediate to check that V y ∈ H s V_y\in\mathcal{H}_s V y ∈ H s . The Lyapunov–Schmidt reduction is by now standard and we do not repeat it here: we refer to [3 , 21 , 22 ] for the details and only recall its outcome. Because of the symmetries, as in [25 ] , the relevant kernel is two-dimensional:
Z : = span { ∑ j = 1 ℓ σ j ∂ δ P U δ , ξ j ( r ) ∣ δ = d ε 1 N − 4 , ∑ j = 1 ℓ σ j ∂ r P U δ , ξ j ( r ) } ⊂ H s . Z:=\operatorname{span}\left\{\left.\sum_{j=1}^{\ell}\sigma_j\partial_\delta PU_{\delta,\xi_j}(r)\right|_{\delta=d\varepsilon^{\frac{1}{N-4}}},\ \sum_{j=1}^{\ell}\sigma_j\partial_rPU_{\delta,\xi_j}(r)\right\}\subset\mathcal{H}_s. Z := span ⎩ ⎨ ⎧ j = 1 ∑ ℓ σ j ∂ δ P U δ , ξ j ( r ) δ = d ε N − 4 1 , j = 1 ∑ ℓ σ j ∂ r P U δ , ξ j ( r ) ⎭ ⎬ ⎫ ⊂ H s .
We denote by Z ⊥ Z^\perp Z ⊥ the orthogonal complement of Z Z Z in H s \mathcal{H}_s H s and by Π ⊥ \Pi^\perp Π ⊥ the orthogonal projection on it. Finally, we let μ = ε 1 N − 4 \mu=\varepsilon^{\frac{1}{N-4}} μ = ε N − 4 1 and
ϑ ( μ ) : = { μ 5 / 2 , N = 5 , μ 4 ∣ log μ ∣ 2 / 3 , N = 6 , μ N + 2 2 , N ≥ 7. \vartheta(\mu):=
\begin{cases}
\mu^{5/2}, & N=5,\\
\mu^4|\log\mu|^{2/3}, & N=6,\\
\mu^{\frac{N+2}{2}}, & N\ge7.
\end{cases} ϑ ( μ ) := ⎩ ⎨ ⎧ μ 5/2 , μ 4 ∣ log μ ∣ 2/3 , μ 2 N + 2 , N = 5 , N = 6 , N ≥ 7.
Proposition 2.5 ([3 , 22 ] ). There are ε 0 , C > 0 \varepsilon_0,C>0 ε 0 , C > 0 such that for ε ∈ ( 0 , ε 0 ) \varepsilon\in(0,\varepsilon_0) ε ∈ ( 0 , ε 0 ) and every y ∈ P η y\in\mathcal{P}_\eta y ∈ P η there is a unique ϕ y ∈ Z ⊥ \phi_y\in Z^\perp ϕ y ∈ Z ⊥ with ∥ ϕ y ∥ ≤ C ϑ ( μ ) \|\phi_y\|\le C\vartheta(\mu) ∥ ϕ y ∥ ≤ C ϑ ( μ ) such that
Π ⊥ [ V y + ϕ y − ( − Δ ) − 1 ( ∣ V y + ϕ y ∣ 2 ∗ − 2 ( V y + ϕ y ) + ε ( V y + ϕ y ) ) ] = 0. \Pi^\perp\left[V_y+\phi_y-(-\Delta)^{-1}\left(|V_y+\phi_y|^{2^*-2}(V_y+\phi_y)+\varepsilon(V_y+\phi_y)\right)\right]=0. Π ⊥ [ V y + ϕ y − ( − Δ ) − 1 ( ∣ V y + ϕ y ∣ 2 ∗ − 2 ( V y + ϕ y ) + ε ( V y + ϕ y ) ) ] = 0. Moreover, the map y ↦ ϕ y y\mapsto\phi_y y ↦ ϕ y is of class C 1 C^1 C 1 .
Hence the whole problem is reduced to the study of a finite-dimensional energy, described by the following results.
Lemma 2.6 ([3 , 22 ] ). Let J ~ ε ( y ) : = J ε ( V y + ϕ y ) \widetilde{J}_\varepsilon(y):=J_\varepsilon(V_y+\phi_y) J ε ( y ) := J ε ( V y + ϕ y ) . If y ∈ int P η y\in\operatorname{int}\mathcal{P}_\eta y ∈ int P η is a critical point of J ~ ε \widetilde{J}_\varepsilon J ε , then V y + ϕ y V_y+\phi_y V y + ϕ y is a solution of (1) .
Proposition 2.7 ([3 , 21 , 22 , 29 ] ). There are constants A , B > 0 A,B>0 A , B > 0 , depending only on N N N , such that, as ε → 0 \varepsilon\to0 ε → 0 ,
J ~ ε ( y ) = ℓ S N / 2 N + μ N − 2 ( ℓ Γ A , B ( y ) + ϖ ε ( y ) ) , ϖ ε → 0 in C 1 ( P η ) , \widetilde{J}_\varepsilon(y)=\ell\frac{S^{N/2}}{N}+\mu^{N-2}\left(\ell\Gamma_{A,B}(y)+\varpi_\varepsilon(y)\right),\qquad\varpi_\varepsilon\to0\text{ in }C^1(\mathcal{P}_\eta), J ε ( y ) = ℓ N S N /2 + μ N − 2 ( ℓ Γ A , B ( y ) + ϖ ε ( y ) ) , ϖ ε → 0 in C 1 ( P η ) ,
where S S S is the best Sobolev constant,
Γ A , B ( r , d ) = A 2 Φ ( r ) d N − 2 − B 2 d 2 (11) \Gamma_{A,B}(r,d)=\frac{A}{2}\Phi(r)d^{N-2}-\frac{B}{2}d^2 \tag*{(11)} Γ A , B ( r , d ) = 2 A Φ ( r ) d N − 2 − 2 B d 2 (11)
and
Φ ( r ) : = τ Ω ( ξ 1 ( r ) ) − ∑ j = 2 ℓ σ j G Ω ( ξ 1 ( r ) , ξ j ( r ) ) . (12) \Phi(r):=\tau_{\Omega}(\xi_1(r))-\sum_{j=2}^{\ell}\sigma_jG_{\Omega}(\xi_1(r),\xi_j(r)). \tag*{(12)} Φ ( r ) := τ Ω ( ξ 1 ( r )) − j = 2 ∑ ℓ σ j G Ω ( ξ 1 ( r ) , ξ j ( r )) . (12)
Let c : = ( B / A ) 1 N − 4 c:=(B/A)^{\frac{1}{N-4}} c := ( B / A ) N − 4 1 . Then Γ A , B ( r , d ) = B c 2 Γ ( r , d / c ) \Gamma_{A,B}(r,d)=Bc^2\Gamma(r,d/c) Γ A , B ( r , d ) = B c 2 Γ ( r , d / c ) , where
Γ ( r , d ) : = 1 2 ( Φ ( r ) d N − 2 − d 2 ) . (13) \Gamma(r,d):=\frac{1}{2}\left(\Phi(r)d^{N-2}-d^2\right). \tag*{(13)} Γ ( r , d ) := 2 1 ( Φ ( r ) d N − 2 − d 2 ) . (13)
Since this change of variables is a positive dilation in d d d , critical points of Γ A , B \Gamma_{A,B} Γ A , B and of Γ \Gamma Γ correspond.
For every r ∈ I r\in I r ∈ I with Φ ( r ) > 0 \Phi(r)>0 Φ ( r ) > 0 , we set
d ( r ) : = ( 2 ( N − 2 ) Φ ( r ) ) 1 N − 4 , (14) d(r):=\left(\frac{2}{(N-2)\Phi(r)}\right)^{\frac{1}{N-4}}, \tag*{(14)} d ( r ) := ( ( N − 2 ) Φ ( r ) 2 ) N − 4 1 , (14)
which is the unique positive zero of the function
d ↦ ∂ d Γ ( r , d ) = d ( N − 2 2 Φ ( r ) d N − 4 − 1 ) . d\mapsto\partial_d\Gamma(r,d)=d\left(\frac{N-2}{2}\Phi(r)d^{N-4}-1\right). d ↦ ∂ d Γ ( r , d ) = d ( 2 N − 2 Φ ( r ) d N − 4 − 1 ) .
Lemma 2.8. Let Φ ∈ C 2 ( I ) \Phi\in C^2(I) Φ ∈ C 2 ( I ) . Then ( r 0 , d 0 ) ∈ M (r_0,d_0)\in\mathcal{M} ( r 0 , d 0 ) ∈ M is a critical point of Γ \Gamma Γ if and only if
Φ ( r 0 ) > 0 , Φ ′ ( r 0 ) = 0 and d 0 = d ( r 0 ) . \Phi(r_0)>0,\qquad\Phi'(r_0)=0\qquad\text{and}\qquad d_0=d(r_0). Φ ( r 0 ) > 0 , Φ ′ ( r 0 ) = 0 and d 0 = d ( r 0 ) . Moreover, at such a point
D 2 Γ ( r 0 , d 0 ) = diag ( 1 2 Φ ′ ′ ( r 0 ) d 0 N − 2 , N − 4 ) . D^2\Gamma(r_0,d_0)=\operatorname{diag}\left(\frac{1}{2}\Phi''(r_0)d_0^{N-2},N-4\right). D 2 Γ ( r 0 , d 0 ) = diag ( 2 1 Φ ′′ ( r 0 ) d 0 N − 2 , N − 4 ) . In particular, ( r 0 , d 0 ) (r_0,d_0) ( r 0 , d 0 ) is non-degenerate if and only if Φ ′ ′ ( r 0 ) ≠ 0 \Phi''(r_0)\ne0 Φ ′′ ( r 0 ) = 0 .
Proof. Since
∂ r Γ ( r , d ) = 1 2 Φ ′ ( r ) d N − 2 , \partial_r\Gamma(r,d)=\frac{1}{2}\Phi'(r)d^{N-2}, ∂ r Γ ( r , d ) = 2 1 Φ ′ ( r ) d N − 2 ,
the first part follows from the definition of d ( r ) d(r) d ( r ) as in (14) .
At a critical point,
∂ r d 2 Γ ( r 0 , d 0 ) = N − 2 2 Φ ′ ( r 0 ) d 0 N − 3 = 0 \partial^2_{rd}\Gamma(r_0,d_0)=\frac{N-2}{2}\Phi'(r_0)d_0^{N-3}=0 ∂ r d 2 Γ ( r 0 , d 0 ) = 2 N − 2 Φ ′ ( r 0 ) d 0 N − 3 = 0
and, using that
N − 2 2 Φ ( r 0 ) d 0 N − 4 = 1 , (15) \frac{N-2}{2}\Phi(r_0)d_0^{N-4}=1, \tag*{(15)} 2 N − 2 Φ ( r 0 ) d 0 N − 4 = 1 , (15)
it is easy to check that
∂ d d 2 Γ = ( N − 2 ) ( N − 3 ) 2 Φ ( r 0 ) d 0 N − 4 − 1 = N − 4. \partial^2_{dd}\Gamma=\frac{(N-2)(N-3)}{2}\Phi(r_0)d_0^{N-4}-1=N-4. ∂ dd 2 Γ = 2 ( N − 2 ) ( N − 3 ) Φ ( r 0 ) d 0 N − 4 − 1 = N − 4.
We can now state the abstract existence result which we shall apply in § 3 and § 4 .
Proposition 2.9. Let us be in case (A) or (B) and assume that Φ \Phi Φ defined in (12) satisfies
Φ ( r ) > 0 for all r ∈ I and Φ ( r ) → + ∞ as r tends to the endpoints of I . (16) \Phi(r)>0 \text{ for all } r\in I \qquad\text{and}\qquad\Phi(r)\to+\infty\text{ as } r \text{ tends to the endpoints of } I. \tag*{(16)} Φ ( r ) > 0 for all r ∈ I and Φ ( r ) → + ∞ as r tends to the endpoints of I . (16) Then there exists ε 1 > 0 \varepsilon_1>0 ε 1 > 0 such that, for every ε ∈ ( 0 , ε 1 ) \varepsilon\in(0,\varepsilon_1) ε ∈ ( 0 , ε 1 ) , problem (1) has a sign-changing solution
u ε = V y ε + ϕ y ε ∈ H s , y ε = ( r ε , d ε ) , with ∥ ϕ y ε ∥ ≤ C ϑ ( μ ) . u_\varepsilon=V_{y_\varepsilon}+\phi_{y_\varepsilon}\in H^s,\qquad y_\varepsilon=(r_\varepsilon,d_\varepsilon),\qquad\text{with }\|\phi_{y_\varepsilon}\|\le C\vartheta(\mu). u ε = V y ε + ϕ y ε ∈ H s , y ε = ( r ε , d ε ) , with ∥ ϕ y ε ∥ ≤ C ϑ ( μ ) . Moreover, up to a subsequence, r ε → r 0 r_\varepsilon\to r_0 r ε → r 0 and d ε → d 0 d_\varepsilon\to d_0 d ε → d 0 , where r 0 r_0 r 0 is a global minimum point of Φ \Phi Φ and
d 0 = ( 2 B ( N − 2 ) A Φ ( r 0 ) ) 1 N − 4 . d_0=\left(\frac{2B}{(N-2)A\Phi(r_0)}\right)^{\frac{1}{N-4}}. d 0 = ( ( N − 2 ) A Φ ( r 0 ) 2 B ) N − 4 1 .
Proof. By (16) , the set K \mathcal{K} K of global minimum points of Φ \Phi Φ is a nonempty compact subset of I I I . For fixed r r r , the function d ↦ Γ ( r , d ) d\mapsto\Gamma(r,d) d ↦ Γ ( r , d ) has a unique minimum point at d ( r ) d(r) d ( r ) , and by (15) , we can consider
f ( Φ ( r ) ) : = Γ ( r , d ( r ) ) = − N − 4 2 ( N − 2 ) ( 2 ( N − 2 ) Φ ( r ) ) 2 N − 4 . f(\Phi(r)):=\Gamma(r,d(r))=-\frac{N-4}{2(N-2)}\left(\frac{2}{(N-2)\Phi(r)}\right)^{\frac{2}{N-4}}. f ( Φ ( r )) := Γ ( r , d ( r )) = − 2 ( N − 2 ) N − 4 ( ( N − 2 ) Φ ( r ) 2 ) N − 4 2 .
which is an increasing function of Φ ( r ) \Phi(r) Φ ( r ) . Hence Γ \Gamma Γ attains its global minimum on M \mathcal{M} M exactly on the compact set { ( r , d ( r ) ) : r ∈ K } \{(r,d(r)):r\in\mathcal{K}\} {( r , d ( r )) : r ∈ K } . Choosing
r − < min K ≤ max K < r + r_-<\min\mathcal{K}\le\max\mathcal{K}<r_+ r − < min K ≤ max K < r +
in I I I and 0 < d − < d ( r ) < d + 0<d_-<d(r)<d_+ 0 < d − < d ( r ) < d + for r ∈ [ r − , r + ] r\in[r_-,r_+] r ∈ [ r − , r + ] , the box
C : = [ r − , r + ] × [ d − , d + ] \mathcal{C}:=[r_-,r_+]\times[d_-,d_+] C := [ r − , r + ] × [ d − , d + ]
satisfies
min ∂ C Γ > min C Γ . \min_{\partial\mathcal{C}}\Gamma>\min_{\mathcal{C}}\Gamma. ∂ C min Γ > C min Γ.
Fix η > 0 \eta>0 η > 0 such that { ( r , c d ) : ( r , d ) ∈ C } ⊂ int P η \{(r,cd):(r,d)\in\mathcal{C}\}\subset\operatorname{int}P_\eta {( r , c d ) : ( r , d ) ∈ C } ⊂ int P η . By Proposition 2.7, the function
F ε ( r , d ) : = μ − ( N − 2 ) ℓ B c 2 ( J ~ ε ( r , c d ) − ℓ S N / 2 N ) F_\varepsilon(r,d):=\frac{\mu^{-(N-2)}}{\ell Bc^2}\left(\widetilde{J}_\varepsilon(r,cd)-\ell\frac{S^{N/2}}{N}\right) F ε ( r , d ) := ℓ B c 2 μ − ( N − 2 ) ( J ε ( r , c d ) − ℓ N S N /2 )
converges to Γ \Gamma Γ uniformly on C \mathcal{C} C . Hence, for ε \varepsilon ε small, min ∂ C F ε > min C F ε \min_{\partial\mathcal{C}}F_\varepsilon>\min_{\mathcal{C}}F_\varepsilon min ∂ C F ε > min C F ε and F ε F_\varepsilon F ε has a minimum point ( r ε , d ˉ ε ) (r_\varepsilon,\bar d_\varepsilon) ( r ε , d ˉ ε ) in the interior of C \mathcal{C} C . Then y ε : = ( r ε , c d ˉ ε ) y_\varepsilon:=(r_\varepsilon,c\bar d_\varepsilon) y ε := ( r ε , c d ˉ ε ) is a critical point of J ~ ε \widetilde{J}_\varepsilon J ε and, by Lemma 2.6 and Proposition 2.5, we can conclude that u ε : = V y ε + ϕ y ε u_\varepsilon:=V_{y_\varepsilon}+\phi_{y_\varepsilon} u ε := V y ε + ϕ y ε is a solution of (1) with the stated estimate. By uniform convergence, every limit point ( r 0 , d ˉ 0 ) (r_0,\bar d_0) ( r 0 , d ˉ 0 ) of ( r ε , d ˉ ε ) (r_\varepsilon,\bar d_\varepsilon) ( r ε , d ˉ ε ) is a minimum point of Γ \Gamma Γ on C \mathcal{C} C , i.e. r 0 ∈ K r_0\in\mathcal{K} r 0 ∈ K and d ˉ 0 = d ( r 0 ) \bar d_0=d(r_0) d ˉ 0 = d ( r 0 ) , which gives the value of d 0 = c d ( r 0 ) d_0=cd(r_0) d 0 = c d ( r 0 ) . Finally, u ε ≢ 0 u_\varepsilon\not\equiv0 u ε ≡ 0 and u ε ∘ R = − u ε u_\varepsilon\circ R=-u_\varepsilon u ε ∘ R = − u ε in case (A), u ε ∘ T = − u ε u_\varepsilon\circ T=-u_\varepsilon u ε ∘ T = − u ε in case (B), so that u ε u_\varepsilon u ε changes sign and this completes the proof. □
Remark 2.10. By Lemma 2.8, any critical point of Φ \Phi Φ at which Φ > 0 \Phi>0 Φ > 0 gives a critical point of Γ \Gamma Γ and a non-degenerate one gives a non-degenerate critical point. Since ϖ ε → 0 \varpi_\varepsilon\to0 ϖ ε → 0 in C 1 C^1 C 1 , a standard degree argument (as in [25 Section 3.1.3] ) gives a solution concentrating at the corresponding configuration. In this paper we only use global minima, which are C 0 C^0 C 0 -stable and do not require any non-degeneracy.
The case of the annulus
Let us consider the annulus Ω ρ \Omega_{\rho} Ω ρ as in (5) and k ≥ 2 k \ge2 k ≥ 2 be an even integer. We are in case (A) of Subsection 2.1, so that the function Φ ( r ) \Phi(r) Φ ( r ) in (12) is
Ψ k ( r ) : = τ Ω ρ ( ξ 1 ( r ) ) − ∑ j = 1 k − 1 ( − 1 ) j G Ω ρ ( ξ 1 ( r ) , ξ j + 1 ( r ) ) , r ∈ ( ρ , 1 ) . (17) \Psi_k(r) := \tau_{\Omega_{\rho}}(\xi_1(r))-\sum_{j=1}^{k-1}(-1)^jG_{\Omega_{\rho}}(\xi_1(r),\xi_{j+1}(r)), \qquad r \in(\rho,1). \tag*{(17)} Ψ k ( r ) := τ Ω ρ ( ξ 1 ( r )) − j = 1 ∑ k − 1 ( − 1 ) j G Ω ρ ( ξ 1 ( r ) , ξ j + 1 ( r )) , r ∈ ( ρ , 1 ) . (17)
By Proposition 2.9, Theorem 1.1 (a) follows once we show that Ψ k > 0 \Psi_k>0 Ψ k > 0 on ( ρ , 1 ) (\rho,1) ( ρ , 1 ) and Ψ k ( r ) → + ∞ \Psi_k(r)\to+\infty Ψ k ( r ) → + ∞ as r → ρ + r\to\rho^+ r → ρ + and as r → 1 − r\to1^- r → 1 − . It is standard the behavior of the function Ψ k \Psi_k Ψ k at the endpoints, instead it is much more involved the proof of Ψ k > 0 \Psi_k>0 Ψ k > 0 .
So, in order to show that, we write (17) by using the explicit expression of the Robin function given by (see [11 ] )
τ Ω ρ ( x ) = 1 ω N − 1 ∑ m = 0 ∞ d m Q m ( ∣ x ∣ ) , \tau_{\Omega_{\rho}}(x)=\frac{1}{\omega_{N-1}}\sum_{m=0}^{\infty}d_mQ_m(|x|), τ Ω ρ ( x ) = ω N − 1 1 m = 0 ∑ ∞ d m Q m ( ∣ x ∣ ) ,
where, for all m ≥ 0 m\ge0 m ≥ 0 ,
d m : = ( m + N − 2 N − 2 ) + ( m + N − 3 N − 2 ) d_m:=\binom{m+N-2}{N-2}+\binom{m+N-3}{N-2} d m := ( N − 2 m + N − 2 ) + ( N − 2 m + N − 3 )
and
Q m ( ∣ x ∣ ) : = ρ 2 m + N − 2 − 2 ρ 2 m + N − 2 ∣ x ∣ 2 m + N − 2 + ∣ x ∣ 4 m + 2 N − 4 ( 2 m + N − 2 ) ∣ x ∣ 2 N + 2 m − 4 ( 1 − ρ 2 m + N − 2 ) . (18) Q_m(|x|):=\frac{\rho^{2m+N-2}-2\rho^{2m+N-2}|x|^{2m+N-2}+|x|^{4m+2N-4}}{(2m+N-2)|x|^{2N+2m-4}(1-\rho^{2m+N-2})}. \tag*{(18)} Q m ( ∣ x ∣ ) := ( 2 m + N − 2 ) ∣ x ∣ 2 N + 2 m − 4 ( 1 − ρ 2 m + N − 2 ) ρ 2 m + N − 2 − 2 ρ 2 m + N − 2 ∣ x ∣ 2 m + N − 2 + ∣ x ∣ 4 m + 2 N − 4 . (18)
We point out that,
ρ 2 m + N − 2 − 2 ρ 2 m + N − 2 r 2 m + N − 2 + r 2 ( 2 m + N − 2 ) = ( r 2 m + N − 2 − ρ 2 m + N − 2 ) 2 + ρ 2 m + N − 2 ( 1 − ρ 2 m + N − 2 ) \rho^{2m+N-2}-2\rho^{2m+N-2}r^{2m+N-2}+r^{2(2m+N-2)}=(r^{2m+N-2}-\rho^{2m+N-2})^2+\rho^{2m+N-2}(1-\rho^{2m+N-2}) ρ 2 m + N − 2 − 2 ρ 2 m + N − 2 r 2 m + N − 2 + r 2 ( 2 m + N − 2 ) = ( r 2 m + N − 2 − ρ 2 m + N − 2 ) 2 + ρ 2 m + N − 2 ( 1 − ρ 2 m + N − 2 )
and so, we have that Q m ( r ) > 0 Q_m(r)>0 Q m ( r ) > 0 for all m ≥ 0 m\ge0 m ≥ 0 and r ∈ ( ρ , 1 ) r\in(\rho,1) r ∈ ( ρ , 1 ) . Moreover, the Green function of the annulus is given by
G Ω ρ ( x , y ) = 1 ω N − 1 [ 1 ( N − 2 ) ∣ x − y ∣ N − 2 − ∑ m = 0 ∞ Q m ( x , y ) Z m ( x ∣ x ∣ , y ∣ y ∣ ) ] , G_{\Omega_{\rho}}(x,y)=\frac{1}{\omega_{N-1}}\left[\frac{1}{(N-2)|x-y|^{N-2}}-\sum_{m=0}^{\infty}Q_m(x,y)Z_m\left(\frac{x}{|x|},\frac{y}{|y|}\right)\right], G Ω ρ ( x , y ) = ω N − 1 1 [ ( N − 2 ) ∣ x − y ∣ N − 2 1 − m = 0 ∑ ∞ Q m ( x , y ) Z m ( ∣ x ∣ x , ∣ y ∣ y ) ] ,
where
Q m ( x , y ) : = ρ 2 m + N − 2 − ρ 2 m + N − 2 ( ∣ x ∣ 2 m + N − 2 + ∣ y ∣ 2 m + N − 2 ) + ∣ x ∣ 2 m + N − 2 ∣ y ∣ 2 m + N − 2 ( 2 m + N − 2 ) ( ∣ x ∣ ∣ y ∣ ) m + N − 2 ( 1 − ρ 2 m + N − 2 ) , Q_m(x,y):=\frac{\rho^{2m+N-2}-\rho^{2m+N-2}(|x|^{2m+N-2}+|y|^{2m+N-2})+|x|^{2m+N-2}|y|^{2m+N-2}}{(2m+N-2)(|x||y|)^{m+N-2}(1-\rho^{2m+N-2})}, Q m ( x , y ) := ( 2 m + N − 2 ) ( ∣ x ∣∣ y ∣ ) m + N − 2 ( 1 − ρ 2 m + N − 2 ) ρ 2 m + N − 2 − ρ 2 m + N − 2 ( ∣ x ∣ 2 m + N − 2 + ∣ y ∣ 2 m + N − 2 ) + ∣ x ∣ 2 m + N − 2 ∣ y ∣ 2 m + N − 2 ,
where ω N − 1 = meas ( S N − 1 ) \omega_{N-1}=\operatorname{meas}(S^{N-1}) ω N − 1 = meas ( S N − 1 ) and Z m ( ζ , η ) Z_m(\zeta,\eta) Z m ( ζ , η ) represents the zonal harmonics of degree m m m which have a particularly simple expression in terms of the Gegenbauer (or ultra-spherical) polynomials C m λ C_m^\lambda C m λ . The latter can be defined in terms of generating functions. If we write (see [31 ] p. 148)
( 1 − 2 s t + s 2 ) − λ = ∑ m = 0 ∞ C m λ ( t ) s m , (19) (1-2st+s^2)^{-\lambda}=\sum_{m=0}^{\infty}C_m^\lambda(t)s^m, \tag*{(19)} ( 1 − 2 s t + s 2 ) − λ = m = 0 ∑ ∞ C m λ ( t ) s m , (19)
where 0 ≤ s < 1 0\le s<1 0 ≤ s < 1 , ∣ t ∣ ≤ 1 |t|\le1 ∣ t ∣ ≤ 1 and λ > 0 \lambda>0 λ > 0 , then the coefficient C m λ C_m^\lambda C m λ is called Gegenbauer polynomial of degree m m m associated with λ \lambda λ . We have, for all m ≥ 0 m\ge0 m ≥ 0 ,
d m = A N , m ⋅ c N , m , (20) d_m=A_{N,m}\cdot c_{N,m}, \tag*{(20)} d m = A N , m ⋅ c N , m , (20)
where
A N , m : = ( N + m − 3 N − 3 ) , c N , m = N + 2 m − 2 N − 2 . (21) A_{N,m} := \binom{N+m-3}{N-3}, \qquad c_{N,m} = \frac{N+2m-2}{N-2}. \tag*{(21)} A N , m := ( N − 3 N + m − 3 ) , c N , m = N − 2 N + 2 m − 2 . (21)
An important fact is to understand the link between the zonal harmonics and the Gegenbauer polynomials. Indeed (see Theorem 2.1 in [11 ] ) if N > 2 N>2 N > 2 and m = 0 , 1 , 2 , … m=0,1,2,\ldots m = 0 , 1 , 2 , … then we have that for all x ′ , y ′ x',y' x ′ , y ′ such that ∣ x ′ ∣ = ∣ y ′ ∣ = 1 |x'|=|y'|=1 ∣ x ′ ∣ = ∣ y ′ ∣ = 1 it holds
Z m ( x ′ , y ′ ) = c N , m C m N − 2 2 ( x ′ ⋅ y ′ ) . (22) Z_m(x',y')=c_{N,m}C_m^{\frac{N-2}{2}}(x'\cdot y'). \tag*{(22)} Z m ( x ′ , y ′ ) = c N , m C m 2 N − 2 ( x ′ ⋅ y ′ ) . (22)
Inserting these expressions in (17) , we have to study the sign of
Ψ k ( r ) = 1 ω N − 1 [ ∑ m = 0 + ∞ d m Q m ( r ) − ∑ j = 1 k − 1 ( − 1 ) j ( 1 ( N − 2 ) ∣ ξ 1 − ξ j + 1 ∣ N − 2 ∑ m = 0 + ∞ Q m ( r ) Z m ( ξ 1 ∣ ξ 1 ∣ , ξ j + 1 ∣ ξ j + 1 ∣ ) ) ] . (23) \Psi_k(r)=\frac{1}{\omega_{N-1}}\left[\sum_{m=0}^{+\infty}d_mQ_m(r)-\sum_{j=1}^{k-1}(-1)^j\left(\frac{1}{(N-2)|\xi_1-\xi_{j+1}|^{N-2}}\sum_{m=0}^{+\infty}Q_m(r)Z_m\left(\frac{\xi_1}{|\xi_1|},\frac{\xi_{j+1}}{|\xi_{j+1}|}\right)\right)\right]. \tag*{(23)} Ψ k ( r ) = ω N − 1 1 [ m = 0 ∑ + ∞ d m Q m ( r ) − j = 1 ∑ k − 1 ( − 1 ) j ( ( N − 2 ) ∣ ξ 1 − ξ j + 1 ∣ N − 2 1 m = 0 ∑ + ∞ Q m ( r ) Z m ( ∣ ξ 1 ∣ ξ 1 , ∣ ξ j + 1 ∣ ξ j + 1 ) ) ] . (23)
Firstly we point out that
∣ ξ 1 ( r ) − ξ j + 1 ( r ) ∣ N − 2 = r N − 2 ( 2 sin ( π j k ) ) N − 2 (24) |\xi_1(r)-\xi_{j+1}(r)|^{N-2}=r^{N-2}\left(2\sin\left(\frac{\pi j}{k}\right)\right)^{N-2} \tag*{(24)} ∣ ξ 1 ( r ) − ξ j + 1 ( r ) ∣ N − 2 = r N − 2 ( 2 sin ( k π j ) ) N − 2 (24)
for all j = 1 , ⋯ , k − 1 j=1,\cdots,k-1 j = 1 , ⋯ , k − 1 .
Hence
∑ j = 1 k − 1 ( − 1 ) j ∣ ξ 1 ( r ) − ξ j + 1 ( r ) ∣ N − 2 = E k , N r N − 2 , where E k , N : = ∑ j = 1 k − 1 ( − 1 ) j ( 2 sin ( π j k ) ) N − 2 . \sum_{j=1}^{k-1}\frac{(-1)^j}{|\xi_1(r)-\xi_{j+1}(r)|^{N-2}}=\frac{\mathcal{E}_{k,N}}{r^{N-2}}, \qquad\text{where } \mathcal{E}_{k,N}:=\sum_{j=1}^{k-1}\frac{(-1)^j}{\left(2\sin\left(\frac{\pi j}{k}\right)\right)^{N-2}}. j = 1 ∑ k − 1 ∣ ξ 1 ( r ) − ξ j + 1 ( r ) ∣ N − 2 ( − 1 ) j = r N − 2 E k , N , where E k , N := j = 1 ∑ k − 1 ( 2 sin ( k π j ) ) N − 2 ( − 1 ) j .
We can rewrite (23) as
Ψ k ( r ) ω N − 1 = − E k , N ( N − 2 ) r N − 2 + ∑ m = 0 + ∞ [ d m + ∑ j = 1 k − 1 ( − 1 ) j Z m ( ξ 1 ∣ ξ 1 ∣ , ξ j + 1 ∣ ξ j + 1 ∣ ) ] Q m ( r ) . (25) \Psi_k(r)\omega_{N-1}=-\frac{\mathcal{E}_{k,N}}{(N-2)r^{N-2}}+\sum_{m=0}^{+\infty}\left[d_m+\sum_{j=1}^{k-1}(-1)^jZ_m\left(\frac{\xi_1}{|\xi_1|},\frac{\xi_{j+1}}{|\xi_{j+1}|}\right)\right]Q_m(r). \tag*{(25)} Ψ k ( r ) ω N − 1 = − ( N − 2 ) r N − 2 E k , N + m = 0 ∑ + ∞ [ d m + j = 1 ∑ k − 1 ( − 1 ) j Z m ( ∣ ξ 1 ∣ ξ 1 , ∣ ξ j + 1 ∣ ξ j + 1 ) ] Q m ( r ) . (25)
Now, using (22) and (20) , (25) becomes
Ψ k ( r ) ω N − 1 = − E k , N ( N − 2 ) r N − 2 + ∑ m = 0 + ∞ c N , m α k , m , N Q m ( r ) , (26) \Psi_k(r)\omega_{N-1}=-\frac{\mathcal{E}_{k,N}}{(N-2)r^{N-2}}+\sum_{m=0}^{+\infty}c_{N,m}\alpha_{k,m,N}Q_m(r), \tag*{(26)} Ψ k ( r ) ω N − 1 = − ( N − 2 ) r N − 2 E k , N + m = 0 ∑ + ∞ c N , m α k , m , N Q m ( r ) , (26)
where
α k , m , N : = A N , m + S k , m , N with S k , m , N : = ∑ j = 1 k − 1 ( − 1 ) j C m λ ( cos ( 2 π j k ) ) (27) \alpha_{k,m,N}:=A_{N,m}+S_{k,m,N}\quad\text{with}\quad S_{k,m,N}:=\sum_{j=1}^{k-1}(-1)^jC_m^\lambda\left(\cos\left(\frac{2\pi j}{k}\right)\right) \tag*{(27)} α k , m , N := A N , m + S k , m , N with S k , m , N := j = 1 ∑ k − 1 ( − 1 ) j C m λ ( cos ( k 2 π j ) ) (27)
and λ = N − 2 2 \lambda=\frac{N-2}{2} λ = 2 N − 2 .
Now, since Q m ( r ) > 0 Q_m(r)>0 Q m ( r ) > 0 for all m ≥ 0 m\geq0 m ≥ 0 and for any ρ < r < 1 \rho<r<1 ρ < r < 1 , we have to show that
E k , N < 0 and α k , m , N ≥ 0 for all m ≥ 0. (28) \mathcal{E}_{k,N}<0\quad\text{and}\quad\alpha_{k,m,N}\geq0\quad\text{for all }m\geq0. \tag*{(28)} E k , N < 0 and α k , m , N ≥ 0 for all m ≥ 0. (28)
Aim 1: E k , N < 0 \mathcal{E}_{k,N}<0 E k , N < 0 .
First, since sin ( π j k ) = sin ( π ( k − j ) k ) \sin\left(\frac{\pi j}{k}\right)=\sin\left(\frac{\pi(k-j)}{k}\right) sin ( k π j ) = sin ( k π ( k − j ) ) and ( − 1 ) k − j = ( − 1 ) j (-1)^{k-j}=(-1)^j ( − 1 ) k − j = ( − 1 ) j , we can write
E k , N = 1 2 N − 2 ( 2 ∑ j = 1 k 2 − 1 ( − 1 ) j sin N − 2 ( π j k ) + ( − 1 ) k 2 ) . \mathcal{E}_{k,N}=\frac{1}{2^{N-2}}\left(2\sum_{j=1}^{\frac{k}{2}-1}\frac{(-1)^j}{\sin^{N-2}\left(\frac{\pi j}{k}\right)}+(-1)^{\frac{k}{2}}\right). E k , N = 2 N − 2 1 2 j = 1 ∑ 2 k − 1 sin N − 2 ( k π j ) ( − 1 ) j + ( − 1 ) 2 k .
If k 2 \frac{k}{2} 2 k is odd, we have that
E k , N = 1 2 N − 2 ( 2 ∑ j = 1 k 2 − 1 ( − 1 ) j sin N − 2 ( π j k ) − 1 ) = 1 2 N − 2 ( − 1 + 2 ∑ h = 1 k 4 − 1 2 − 1 sin N − 2 ( π k ( 2 h − 1 ) ) + 2 ∑ h = 1 k 4 − 1 2 1 sin N − 2 ( 2 π h k ) ) = 1 2 N − 2 [ − 1 + 2 ∑ h = 1 k 4 − 1 2 ( 1 sin N − 2 ( 2 π h k ) − 1 sin N − 2 ( π k ( 2 h − 1 ) ) ) ] . \begin{aligned}
\mathcal{E}_{k,N} &= \frac{1}{2^{N-2}}\left(2\sum_{j=1}^{\frac{k}{2}-1}\frac{(-1)^j}{\sin^{N-2}\left(\frac{\pi j}{k}\right)}-1\right)\\
&= \frac{1}{2^{N-2}}\left(-1+2\sum_{h=1}^{\frac{k}{4}-\frac{1}{2}}\frac{-1}{\sin^{N-2}\left(\frac{\pi}{k}(2h-1)\right)}+2\sum_{h=1}^{\frac{k}{4}-\frac{1}{2}}\frac{1}{\sin^{N-2}\left(\frac{2\pi h}{k}\right)}\right)\\
&= \frac{1}{2^{N-2}}\left[-1+2\sum_{h=1}^{\frac{k}{4}-\frac{1}{2}}\left(\frac{1}{\sin^{N-2}\left(\frac{2\pi h}{k}\right)}-\frac{1}{\sin^{N-2}\left(\frac{\pi}{k}(2h-1)\right)}\right)\right].
\end{aligned} E k , N = 2 N − 2 1 2 j = 1 ∑ 2 k − 1 sin N − 2 ( k π j ) ( − 1 ) j − 1 = 2 N − 2 1 − 1 + 2 h = 1 ∑ 4 k − 2 1 sin N − 2 ( k π ( 2 h − 1 ) ) − 1 + 2 h = 1 ∑ 4 k − 2 1 sin N − 2 ( k 2 π h ) 1 = 2 N − 2 1 − 1 + 2 h = 1 ∑ 4 k − 2 1 ( sin N − 2 ( k 2 π h ) 1 − sin N − 2 ( k π ( 2 h − 1 ) ) 1 ) .
Now, we can observe that, since h ≤ k 4 h \le\frac{k}{4} h ≤ 4 k ,
0 < π k ( 2 h − 1 ) < 2 π h k < π 2 0<\frac{\pi}{k}(2h-1)<\frac{2\pi h}{k}<\frac{\pi}{2} 0 < k π ( 2 h − 1 ) < k 2 π h < 2 π
and so, from the fact that sin x \sin x sin x is increasing in ( 0 , π 2 ) (0,\frac{\pi}{2}) ( 0 , 2 π ) we observe that
1 sin N − 2 ( π k ( 2 h − 1 ) ) > 1 sin N − 2 ( 2 π h k ) . \frac{1}{\sin^{N-2}\left(\frac{\pi}{k}(2h-1)\right)}>\frac{1}{\sin^{N-2}\left(\frac{2\pi h}{k}\right)}. sin N − 2 ( k π ( 2 h − 1 ) ) 1 > sin N − 2 ( k 2 π h ) 1 .
Hence,
E k , N = 1 2 N − 2 [ − 1 + 2 ∑ h = 1 k 4 − 1 2 ( 1 sin N − 2 ( 2 π h k ) − 1 sin N − 2 ( π k ( 2 h − 1 ) ) ) ] < 0 \mathcal{E}_{k,N}=\frac{1}{2^{N-2}}\left[-1+2\sum_{h=1}^{\frac{k}{4}-\frac{1}{2}}\left(\frac{1}{\sin^{N-2}\left(\frac{2\pi h}{k}\right)}-\frac{1}{\sin^{N-2}\left(\frac{\pi}{k}(2h-1)\right)}\right)\right]<0 E k , N = 2 N − 2 1 − 1 + 2 h = 1 ∑ 4 k − 2 1 ( sin N − 2 ( k 2 π h ) 1 − sin N − 2 ( k π ( 2 h − 1 ) ) 1 ) < 0
and so the claim follows.
Now, we want to consider the case in which k 2 \frac{k}{2} 2 k is even. Then we have that
E k , N = 1 2 N − 2 ( 1 + 2 A ) , with A : = ∑ j = 1 k 2 − 1 ( − 1 ) j sin N − 2 ( π j k ) \mathcal{E}_{k,N}=\frac{1}{2^{N-2}}(1+2\mathcal{A}),\quad\text{with }\mathcal{A}:=\sum_{j=1}^{\frac{k}{2}-1}\frac{(-1)^j}{\sin^{N-2}\left(\frac{\pi j}{k}\right)} E k , N = 2 N − 2 1 ( 1 + 2 A ) , with A := j = 1 ∑ 2 k − 1 sin N − 2 ( k π j ) ( − 1 ) j
and, so, in order to get E k , N < 0 \mathcal{E}_{k,N}<0 E k , N < 0 , we need to prove that A < − 1 2 \mathcal{A}<-\frac{1}{2} A < − 2 1 .
Now, we can write
A = ∑ j = 1 k 2 − 1 ( − 1 ) j a j , where a j : = 1 sin N − 2 ( π j k ) . \mathcal{A}=\sum_{j=1}^{\frac{k}{2}-1}(-1)^j a_j,\quad\text{where }a_j:=\frac{1}{\sin^{N-2}\left(\frac{\pi j}{k}\right)}. A = j = 1 ∑ 2 k − 1 ( − 1 ) j a j , where a j := sin N − 2 ( k π j ) 1 .
Now, since k 2 − 1 \frac{k}{2}-1 2 k − 1 is odd, we have that
A = − ( a 1 − a 2 ) − ( a 3 − a 4 ) − ⋯ − ( a k 2 − 3 − a k 2 − 2 ) − a k 2 − 1 < − a k 2 − 1 \mathcal{A}=-(a_1-a_2)-(a_3-a_4)-\cdots-\left(a_{\frac{k}{2}-3}-a_{\frac{k}{2}-2}\right)-a_{\frac{k}{2}-1}<-a_{\frac{k}{2}-1} A = − ( a 1 − a 2 ) − ( a 3 − a 4 ) − ⋯ − ( a 2 k − 3 − a 2 k − 2 ) − a 2 k − 1 < − a 2 k − 1
since a j > a j + 1 a_j>a_{j+1} a j > a j + 1 .
Moreover, we can observe that
a k 2 − 1 = 1 sin N − 2 ( π k ( k 2 − 1 ) ) = 1 cos N − 2 ( π k ) > 1. a_{\frac{k}{2}-1}=\frac{1}{\sin^{N-2}\left(\frac{\pi}{k}\left(\frac{k}{2}-1\right)\right)}=\frac{1}{\cos^{N-2}\left(\frac{\pi}{k}\right)}>1. a 2 k − 1 = sin N − 2 ( k π ( 2 k − 1 ) ) 1 = cos N − 2 ( k π ) 1 > 1.
Therefore, we obtain that A < − 1 \mathcal{A}<-1 A < − 1 . Hence the thesis follows.
Aim 2: α k , m , N ≥ 0 \alpha_{k,m,N}\geq0 α k , m , N ≥ 0 for all N ≥ 3 N\geq3 N ≥ 3 , k ≥ 2 k\geq2 k ≥ 2 even and m ≥ 0 m\geq0 m ≥ 0 .
Now, letting β j : = 2 π j k \beta_j := \frac{2\pi j}{k} β j := k 2 π j , we recall the following expression of Gegenbauer polynomials (see [10 8.934.2] )
C m λ ( cos β j ) = ∑ l = 0 m γ m , l cos ( ( m − 2 l ) β j ) , γ m , l : = Γ ( λ + l ) Γ ( λ + m − l ) Γ ( λ ) l ! ( m − l ) ! > 0. C_m^\lambda(\cos\beta_j) = \sum_{l=0}^{m} \gamma_{m,l} \cos((m-2l)\beta_j), \qquad\gamma_{m,l} := \frac{\Gamma(\lambda+l)\Gamma(\lambda+m-l)}{\Gamma(\lambda)l!(m-l)!} > 0. C m λ ( cos β j ) = l = 0 ∑ m γ m , l cos (( m − 2 l ) β j ) , γ m , l := Γ ( λ ) l ! ( m − l )! Γ ( λ + l ) Γ ( λ + m − l ) > 0.
Taking t = 1 t=1 t = 1 in (19) and using the binomial series, for ∣ s ∣ < 1 |s|<1 ∣ s ∣ < 1
∑ m = 0 ∞ C m λ ( 1 ) s m = ( 1 − s ) − 2 λ = ∑ m = 0 ∞ Γ ( 2 λ + m ) Γ ( 2 λ ) m ! s m , \sum_{m=0}^{\infty} C_m^\lambda(1)s^m = (1-s)^{-2\lambda} = \sum_{m=0}^{\infty} \frac{\Gamma(2\lambda+m)}{\Gamma(2\lambda)m!}s^m, m = 0 ∑ ∞ C m λ ( 1 ) s m = ( 1 − s ) − 2 λ = m = 0 ∑ ∞ Γ ( 2 λ ) m ! Γ ( 2 λ + m ) s m ,
so that, by the uniqueness of the coefficients of a power series and since 2 λ = N − 2 2\lambda=N-2 2 λ = N − 2 ,
C m λ ( 1 ) = ( N + m − 3 ) ! ( N − 3 ) ! m ! = A N , m . C_m^\lambda(1) = \frac{(N+m-3)!}{(N-3)!m!} = A_{N,m}. C m λ ( 1 ) = ( N − 3 )! m ! ( N + m − 3 )! = A N , m .
Moreover, taking β j = 0 \beta_j=0 β j = 0 in the expression of C m λ ( cos β j ) C_m^\lambda(\cos\beta_j) C m λ ( cos β j ) above, C m λ ( 1 ) = ∑ l = 0 m γ m , l C_m^\lambda(1)=\sum_{l=0}^{m}\gamma_{m,l} C m λ ( 1 ) = ∑ l = 0 m γ m , l .
Therefore
A N , m = C m λ ( 1 ) = ∑ l = 0 m γ m , l ; A_{N,m}=C_m^\lambda(1)=\sum_{l=0}^{m}\gamma_{m,l}; A N , m = C m λ ( 1 ) = l = 0 ∑ m γ m , l ;
so, since ( − 1 ) j cos ( ( m − 2 l ) β j ) = cos ( ( m − 2 l ) β j + π j ) (-1)^j\cos((m-2l)\beta_j)=\cos((m-2l)\beta_j+\pi j) ( − 1 ) j cos (( m − 2 l ) β j ) = cos (( m − 2 l ) β j + π j ) and π j = k 2 ⋅ 2 π j k \pi j=\frac{k}{2}\cdot\frac{2\pi j}{k} π j = 2 k ⋅ k 2 π j ,
α k , m , N = ∑ l = 0 m γ m , l [ 1 + ∑ j = 1 k − 1 cos ( ( m − 2 l + k 2 ) 2 π j k ) ] . \alpha_{k,m,N}=\sum_{l=0}^{m}\gamma_{m,l}\left[1+\sum_{j=1}^{k-1}\cos\left(\left(m-2l+\frac{k}{2}\right)\frac{2\pi j}{k}\right)\right]. α k , m , N = l = 0 ∑ m γ m , l [ 1 + j = 1 ∑ k − 1 cos ( ( m − 2 l + 2 k ) k 2 π j ) ] .
In addition, for every integer n n n we have (see also [24 (4.29)] )
∑ j = 0 k − 1 e 2 π i n j / k = { k if k ∣ n , 0 otherwise ; \sum_{j=0}^{k-1}e^{2\pi inj/k}=
\begin{cases}
k & \text{if } k\mid n,\\
0 & \text{otherwise};
\end{cases} j = 0 ∑ k − 1 e 2 π inj / k = { k 0 if k ∣ n , otherwise ;
hence
1 + ∑ j = 1 k − 1 cos ( 2 π n j k ) = { k if k divides n , 0 otherwise . (29) 1+\sum_{j=1}^{k-1}\cos\left(\frac{2\pi nj}{k}\right)=
\begin{cases}
k & \text{if } k\text{ divides }n,\\
0 & \text{otherwise}.
\end{cases} \tag*{(29)} 1 + j = 1 ∑ k − 1 cos ( k 2 π nj ) = { k 0 if k divides n , otherwise . (29)
Applying (29) with the integer n = m − 2 l + k 2 n=m-2l+\frac{k}{2} n = m − 2 l + 2 k we can conclude that
α k , m , N = k ∑ 0 ≤ l ≤ m k ∣ m − 2 l + k / 2 γ m , l ≥ 0. \alpha_{k,m,N}=k\sum_{\substack{0\le l\le m\\ k\mid m-2l+k/2}}\gamma_{m,l}\ge0. α k , m , N = k 0 ≤ l ≤ m k ∣ m − 2 l + k /2 ∑ γ m , l ≥ 0.
By (26) and (28) ,
ω N − 1 Ψ k ( r ) ≥ − C k , N ( N − 2 ) r N − 2 > 0 \omega_{N-1}\Psi_k(r)\ge-\frac{C_{k,N}}{(N-2)r^{N-2}}>0 ω N − 1 Ψ k ( r ) ≥ − ( N − 2 ) r N − 2 C k , N > 0
for every r ∈ ( ρ , 1 ) r\in(\rho,1) r ∈ ( ρ , 1 ) , which concludes the proof of Theorem 1.1 (a).
The case of the ball
Let us consider, for every N ≥ 5 N \ge5 N ≥ 5 , the unit ball
B 1 ( 0 ) : = { x ∈ R N : ∣ x ∣ < 1 } . B_{1}(0) := \{x \in\mathbb{R}^{N} : |x| < 1\}. B 1 ( 0 ) := { x ∈ R N : ∣ x ∣ < 1 } .
The Robin function and the Green function of B 1 ( 0 ) B_{1}(0) B 1 ( 0 ) are explicitly given by
τ ( x ) : = 1 ( 1 − ∣ x ∣ 2 ) N − 2 , G ( x , y ) : = ( 1 ∣ x − y ∣ N − 2 − 1 ( 1 − 2 ⟨ x , y ⟩ + ∣ x ∣ 2 ∣ y ∣ 2 ) N − 2 2 ) \tau(x) := \frac{1}{(1-|x|^{2})^{N-2}}, \qquad G(x,y) := \left(\frac{1}{|x-y|^{N-2}}-\frac{1}{(1-2\langle x,y\rangle+|x|^{2}|y|^{2})^{\frac{N-2}{2}}}\right) τ ( x ) := ( 1 − ∣ x ∣ 2 ) N − 2 1 , G ( x , y ) := ( ∣ x − y ∣ N − 2 1 − ( 1 − 2 ⟨ x , y ⟩ + ∣ x ∣ 2 ∣ y ∣ 2 ) 2 N − 2 1 )
up to the multiplicative constant ω N − 1 \omega_{N}^{-1} ω N − 1 , which we drop in order to simplify the computations (it can be absorbed in the constants A , B A,B A , B of Proposition 2.7 ).
The symmetric setting
We are in case ( B ) (B) ( B ) of Subsection 2.1, so that the function (12) is
Φ k , N ( r ) : = τ ( ξ 1 ) − ∑ j = 2 k G ( ξ 1 , ξ j ) + ∑ j = 1 k G ( ξ 1 , η j ) , r ∈ ( 0 , 1 ) , (30) \Phi_{k,N}(r):=\tau(\xi_{1})-\sum_{j=2}^{k}G(\xi_{1},\xi_{j})+\sum_{j=1}^{k}G(\xi_{1},\eta_{j}), \qquad r\in(0,1), \tag*{(30)} Φ k , N ( r ) := τ ( ξ 1 ) − j = 2 ∑ k G ( ξ 1 , ξ j ) + j = 1 ∑ k G ( ξ 1 , η j ) , r ∈ ( 0 , 1 ) , (30)
where ξ j = ξ j ( r ) \xi_j=\xi_j(r) ξ j = ξ j ( r ) and η j = η j ( r ) \eta_j=\eta_j(r) η j = η j ( r ) .
By Proposition 2.9 , Theorem 1.1 (b) follows once we prove that Φ k , N \Phi_{k,N} Φ k , N satisfies (16) on I = ( 0 , 1 ) I=(0,1) I = ( 0 , 1 ) for 2 ≤ k ≤ k ∗ ( N ) 2\le k\le k^{*}(N) 2 ≤ k ≤ k ∗ ( N ) .
Now, we let
Λ k : = ∑ j = 1 k − 1 1 sin N − 2 ( π j k ) (31) \Lambda_k:=\sum_{j=1}^{k-1}\frac{1}{\sin^{N-2}\left(\frac{\pi j}{k}\right)} \tag*{(31)} Λ k := j = 1 ∑ k − 1 sin N − 2 ( k π j ) 1 (31)
and we introduce the following constant
c k , N : = k 2 − N − 2 2 − 2 − ( N − 2 ) Λ k , (32) c_{k,N}:=k2^{-\frac{N-2}{2}}-2^{-(N-2)}\Lambda_k, \tag*{(32)} c k , N := k 2 − 2 N − 2 − 2 − ( N − 2 ) Λ k , (32)
or, equivalently, D k ( N ) = 2 N − 2 c k , N = k 2 N − 2 2 − Λ k D_k(N)=2^{N-2}c_{k,N}=k2^{\frac{N-2}{2}}-\Lambda_k D k ( N ) = 2 N − 2 c k , N = k 2 2 N − 2 − Λ k .
Structure of Φ k , N \Phi_{k,N} Φ k , N
In order to evaluate the Green and the Robin function we recall some useful facts.
In particular, for j = 1 , … , k j=1,\ldots,k j = 1 , … , k ,
∣ ξ 1 − ξ j ∣ 2 = 2 r 2 ( 1 − cos θ j ) = 4 r 2 sin 2 π ( j − 1 ) k , ξ 1 ⋅ ξ j = r 2 cos θ j , |\xi_1-\xi_j|^2=2r^2(1-\cos\theta_j)=4r^2\sin^2\frac{\pi(j-1)}{k}, \qquad\xi_1\cdot\xi_j=r^2\cos\theta_j, ∣ ξ 1 − ξ j ∣ 2 = 2 r 2 ( 1 − cos θ j ) = 4 r 2 sin 2 k π ( j − 1 ) , ξ 1 ⋅ ξ j = r 2 cos θ j ,
while, since the two planes are orthogonal,
∣ ξ 1 − η j ∣ 2 = 2 r 2 , ξ 1 ⋅ η j = 0. |\xi_1-\eta_j|^2=2r^2, \qquad\xi_1\cdot\eta_j=0. ∣ ξ 1 − η j ∣ 2 = 2 r 2 , ξ 1 ⋅ η j = 0.
Fix r ∈ ( 0 , 1 ) r\in(0,1) r ∈ ( 0 , 1 ) and define
h ( t ) = h r ( t ) : = ( 1 + r 4 − 2 r 2 t ) − N − 2 2 , t ∈ [ − 1 , 1 ] . (33) h(t)=h_r(t):=(1+r^4-2r^2t)^{-\frac{N-2}{2}}, \qquad t\in[-1,1]. \tag*{(33)} h ( t ) = h r ( t ) := ( 1 + r 4 − 2 r 2 t ) − 2 N − 2 , t ∈ [ − 1 , 1 ] . (33)
Since 1 + r 4 − 2 r 2 t ≥ ( 1 − r 2 ) 2 > 0 1+r^4-2r^2t\ge(1-r^2)^2>0 1 + r 4 − 2 r 2 t ≥ ( 1 − r 2 ) 2 > 0 on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , h h h is smooth and positive there, and
H ( ξ 1 , ξ j ) = h ( cos θ j ) , τ ( ξ 1 ) = h ( 1 ) = ( 1 − r 2 ) − ( N − 2 ) , H ( ξ 1 , η j ) = h ( 0 ) = ( 1 + r 4 ) − N − 2 2 . (34) H(\xi_1,\xi_j)=h(\cos\theta_j), \qquad\tau(\xi_1)=h(1)=(1-r^2)^{-(N-2)}, \qquad H(\xi_1,\eta_j)=h(0)=(1+r^4)^{-\frac{N-2}{2}}. \tag*{(34)} H ( ξ 1 , ξ j ) = h ( cos θ j ) , τ ( ξ 1 ) = h ( 1 ) = ( 1 − r 2 ) − ( N − 2 ) , H ( ξ 1 , η j ) = h ( 0 ) = ( 1 + r 4 ) − 2 N − 2 . (34)
Lemma 4.1 (Decomposition). For every k ≥ 2 k \ge2 k ≥ 2 and r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) ,
Φ k , N ( r ) = R k , N ( r ) + c k , N r N − 2 , R k , N ( r ) : = ∑ j = 1 k h ( cos θ j ) − k h ( 0 ) , (35) \Phi_{k,N}(r)=R_{k,N}(r)+\frac{c_{k,N}}{r^{N-2}},\qquad R_{k,N}(r):=\sum_{j=1}^{k}h(\cos\theta_j)-kh(0), \tag*{(35)} Φ k , N ( r ) = R k , N ( r ) + r N − 2 c k , N , R k , N ( r ) := j = 1 ∑ k h ( cos θ j ) − k h ( 0 ) , (35) with c k , N c_{k,N} c k , N given by (32) .
Proof. In the following we will treat separately the regular and the singular part of the Green function. In particular,
Φ k , N ( r ) = τ ( ξ 1 ) − ∑ j = 2 k [ 1 ∣ ξ 1 − ξ j ∣ N − 2 − H ( ξ 1 , ξ j ) ] + ∑ j = 1 k [ 1 ∣ ξ 1 − η j ∣ N − 2 − H ( ξ 1 , η j ) ] = ∑ j = 1 k H ( ξ 1 , ξ j ) − ∑ j = 1 k H ( ξ 1 , η j ) ⏟ ( I 1 ) + ∑ j = 1 k 1 ∣ ξ 1 − η j ∣ N − 2 − ∑ j = 2 k 1 ∣ ξ 1 − ξ j ∣ N − 2 ⏟ ( I 2 ) . (36) \begin{aligned}
\Phi_{k,N}(r)&=\tau(\xi_1)-\sum_{j=2}^{k}\left[\frac{1}{|\xi_1-\xi_j|^{N-2}}-H(\xi_1,\xi_j)\right]+\sum_{j=1}^{k}\left[\frac{1}{|\xi_1-\eta_j|^{N-2}}-H(\xi_1,\eta_j)\right]\\
&=\underbrace{\sum_{j=1}^{k}H(\xi_1,\xi_j)-\sum_{j=1}^{k}H(\xi_1,\eta_j)}_{(I_1)}+\underbrace{\sum_{j=1}^{k}\frac{1}{|\xi_1-\eta_j|^{N-2}}-\sum_{j=2}^{k}\frac{1}{|\xi_1-\xi_j|^{N-2}}}_{(I_2)}. \tag*{(36)}
\end{aligned} Φ k , N ( r ) = τ ( ξ 1 ) − j = 2 ∑ k [ ∣ ξ 1 − ξ j ∣ N − 2 1 − H ( ξ 1 , ξ j ) ] + j = 1 ∑ k [ ∣ ξ 1 − η j ∣ N − 2 1 − H ( ξ 1 , η j ) ] = ( I 1 ) j = 1 ∑ k H ( ξ 1 , ξ j ) − j = 1 ∑ k H ( ξ 1 , η j ) + ( I 2 ) j = 1 ∑ k ∣ ξ 1 − η j ∣ N − 2 1 − j = 2 ∑ k ∣ ξ 1 − ξ j ∣ N − 2 1 . (36) Therefore, from (34) , the regular parts give
( I 1 ) = τ ( ξ 1 ) + ∑ j = 2 k H ( ξ 1 , ξ j ) − ∑ j = 1 k H ( ξ 1 , η j ) = h ( cos θ 1 ) + ∑ j = 2 k h ( cos θ j ) − k h ( 0 ) = R k , N ( r ) . (I_1)=\tau(\xi_1)+\sum_{j=2}^{k}H(\xi_1,\xi_j)-\sum_{j=1}^{k}H(\xi_1,\eta_j)=h(\cos\theta_1)+\sum_{j=2}^{k}h(\cos\theta_j)-kh(0)=R_{k,N}(r). ( I 1 ) = τ ( ξ 1 ) + j = 2 ∑ k H ( ξ 1 , ξ j ) − j = 1 ∑ k H ( ξ 1 , η j ) = h ( cos θ 1 ) + j = 2 ∑ k h ( cos θ j ) − k h ( 0 ) = R k , N ( r ) . Instead, the singular parts give
( I 2 ) = − ∑ j = 2 k 1 ( 2 r sin π ( j − 1 ) k ) N − 2 + k ( 2 r 2 ) − N − 2 2 = r − ( N − 2 ) [ k 2 − N − 2 2 − 2 − ( N − 2 ) ∑ j = 1 k − 1 1 sin N − 2 ( π j k ) ] = c k , N r N − 2 . \begin{aligned}
(I_2)&=-\sum_{j=2}^{k}\frac{1}{\left(2r\sin\frac{\pi(j-1)}{k}\right)^{N-2}}+k(2r^2)^{-\frac{N-2}{2}}\\
&=r^{-(N-2)}\left[k2^{-\frac{N-2}{2}}-2^{-(N-2)}\sum_{j=1}^{k-1}\frac{1}{\sin^{N-2}\left(\frac{\pi j}{k}\right)}\right]=\frac{c_{k,N}}{r^{N-2}}.
\end{aligned} ( I 2 ) = − j = 2 ∑ k ( 2 r sin k π ( j − 1 ) ) N − 2 1 + k ( 2 r 2 ) − 2 N − 2 = r − ( N − 2 ) [ k 2 − 2 N − 2 − 2 − ( N − 2 ) j = 1 ∑ k − 1 sin N − 2 ( k π j ) 1 ] = r N − 2 c k , N . We now study the sign of Φ k , N \Phi_{k,N} Φ k , N , analyzing separately the sign of the regular part and of the singular part.
Lemma 4.2 (Positivity of the regular part). For every k ≥ 2 k \ge2 k ≥ 2 and every r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) ,
R k , N ( r ) > max { 0 , ( 1 − r 2 ) − ( N − 2 ) − k } . (37) R_{k,N}(r)>\max\left\{0,\,(1-r^2)^{-(N-2)}-k\right\}. \tag*{(37)} R k , N ( r ) > max { 0 , ( 1 − r 2 ) − ( N − 2 ) − k } . (37) In particular R k , N ( r ) → + ∞ R_{k,N}(r)\to+\infty R k , N ( r ) → + ∞ as r → 1 − r\to1^{-} r → 1 − .
Proof. We have
h ′ ( t ) = ( N − 2 ) r 2 ( 1 + r 4 − 2 r 2 t ) − N 2 > 0 h'(t)=(N-2)r^2(1+r^4-2r^2t)^{-\frac{N}{2}}>0 h ′ ( t ) = ( N − 2 ) r 2 ( 1 + r 4 − 2 r 2 t ) − 2 N > 0 and
h ′ ′ ( t ) = N ( N − 2 ) r 4 ( 1 + r 4 − 2 r 2 t ) − N 2 − 1 > 0 on [ − 1 , 1 ] , h''(t)=N(N-2)r^4(1+r^4-2r^2t)^{-\frac{N}{2}-1}>0\quad\text{on }[-1,1], h ′′ ( t ) = N ( N − 2 ) r 4 ( 1 + r 4 − 2 r 2 t ) − 2 N − 1 > 0 on [ − 1 , 1 ] , so h h h is strictly convex. Furthermore, it is known that
1 k ∑ j = 1 k cos 2 π ( j − 1 ) k = 1 k Re ∑ j = 0 k − 1 e 2 π i j / k = 0. \frac{1}{k}\sum_{j=1}^{k}\cos\frac{2\pi(j-1)}{k}=\frac{1}{k}\operatorname{Re}\sum_{j=0}^{k-1}e^{2\pi i j/k}=0. k 1 j = 1 ∑ k cos k 2 π ( j − 1 ) = k 1 Re j = 0 ∑ k − 1 e 2 π ij / k = 0.
By the strict Jensen inequality,
1 k ∑ j = 1 k h ( cos θ j ) > h ( 1 k ∑ j = 1 k cos θ j ) = h ( 0 ) \frac{1}{k}\sum_{j=1}^{k}h(\cos\theta_j)>h\left(\frac{1}{k}\sum_{j=1}^{k}\cos\theta_j\right)=h(0) k 1 j = 1 ∑ k h ( cos θ j ) > h ( k 1 j = 1 ∑ k cos θ j ) = h ( 0 )
and so, we can conclude that R k , N ( r ) > 0 R_{k,N}(r)>0 R k , N ( r ) > 0 .
Finally, since h > 0 h>0 h > 0 and h ( 0 ) = ( 1 + r 4 ) − N − 2 2 < 1 h(0)=(1+r^4)^{-\frac{N-2}{2}}<1 h ( 0 ) = ( 1 + r 4 ) − 2 N − 2 < 1 ,
R k , N ( r ) = ∑ j = 1 k h ( cos θ j ) − k h ( 0 ) = h ( cos θ 1 ) + ∑ j = 2 k h ( cos θ j ) − k h ( 0 ) = h ( 1 ) + ∑ j = 2 k h ( cos θ j ) − h ( 0 ) ⏟ > 0 > h ( 1 ) − k h ( 0 ) ≥ ( 1 − r 2 ) − ( N − 2 ) − k . \begin{aligned}
R_{k,N}(r)=\sum_{j=1}^{k}h(\cos\theta_j)-kh(0)=h(\cos\theta_1)+\sum_{j=2}^{k}h(\cos\theta_j)-kh(0) \\
&=h(1)+\sum_{j=2}^{k}\underbrace{h(\cos\theta_j)-h(0)}_{>0}>h(1)-kh(0)\geq(1-r^2)^{-(N-2)}-k.
\end{aligned} R k , N ( r ) = j = 1 ∑ k h ( cos θ j ) − k h ( 0 ) = h ( cos θ 1 ) + j = 2 ∑ k h ( cos θ j ) − k h ( 0 ) = h ( 1 ) + j = 2 ∑ k > 0 h ( cos θ j ) − h ( 0 ) > h ( 1 ) − k h ( 0 ) ≥ ( 1 − r 2 ) − ( N − 2 ) − k .
We also note that h r ( t ) → 1 h_r(t)\to1 h r ( t ) → 1 uniformly on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] as r → 0 + r\to0^+ r → 0 + , so that R k , N ( r ) → 0 R_{k,N}(r)\to0 R k , N ( r ) → 0 as r → 0 + r\to0^+ r → 0 + . □
Proposition 4.3. Let k ≥ 2 k\geq2 k ≥ 2 and assume c k , N > 0 c_{k,N}>0 c k , N > 0 . Then:
(i) Φ k , N ( r ) > c k , N r − ( N − 2 ) > 0 \Phi_{k,N}(r)>c_{k,N}r^{-(N-2)}>0 Φ k , N ( r ) > c k , N r − ( N − 2 ) > 0 for every r ∈ ( 0 , 1 ) r\in(0,1) r ∈ ( 0 , 1 ) ;
(ii) Φ k , N ( r ) → + ∞ \Phi_{k,N}(r)\to+\infty Φ k , N ( r ) → + ∞ as r → 0 + r\to0^+ r → 0 + and as r → 1 − r\to1^- r → 1 − .
In particular, Φ k , N \Phi_{k,N} Φ k , N satisfies (16) on I = ( 0 , 1 ) I=(0,1) I = ( 0 , 1 ) .
Proof. We can observe that (i) follows from Lemmas 4.1 and 4.2 . Instead, we get that, as r → 0 + r\to0^+ r → 0 + ,
Φ k , N ( r ) ≥ c k , N r N − 2 → + ∞ \Phi_{k,N}(r)\geq\frac{c_{k,N}}{r^{N-2}}\to+\infty Φ k , N ( r ) ≥ r N − 2 c k , N → + ∞
and, as r → 1 − r\to1^- r → 1 − , Φ k , N ( r ) ≥ R k , N ( r ) → + ∞ \Phi_{k,N}(r)\geq R_{k,N}(r)\to+\infty Φ k , N ( r ) ≥ R k , N ( r ) → + ∞ . Therefore (ii) easily follows. □
The sign of c k , N c_{k,N} c k , N
Proposition 4.4. Let N ≥ 5 N\geq5 N ≥ 5 and D k ( N ) = 2 N − 2 c k , N D_k(N)=2^{N-2}c_{k,N} D k ( N ) = 2 N − 2 c k , N as in (32) . Then
(a) D k ( N ) > 0 D_k(N)>0 D k ( N ) > 0 for k = 2 , 3 , 4 k=2,3,4 k = 2 , 3 , 4 and every N ≥ 5 N\geq5 N ≥ 5 ;
(b) D 5 ( 5 ) > 0 D_5(5)>0 D 5 ( 5 ) > 0 and D 5 ( 6 ) = 4 5 > 0 D_5(6)=\frac{4}{5}>0 D 5 ( 6 ) = 5 4 > 0 , while D 5 ( N ) < 0 D_5(N)<0 D 5 ( N ) < 0 for every N ≥ 7 N\geq7 N ≥ 7 ;
(c) D k ( N ) < 0 D_k(N)<0 D k ( N ) < 0 for every k ≥ 6 k\geq6 k ≥ 6 and every N ≥ 5 N\geq5 N ≥ 5 .
Consequently, for N ≥ 5 N\geq5 N ≥ 5 and k ≥ 2 k\geq2 k ≥ 2 , c k , N > 0 c_{k,N}>0 c k , N > 0 if and only if k ≤ k ∗ ( N ) k\leq k^*(N) k ≤ k ∗ ( N ) , with k ∗ ( N ) k^*(N) k ∗ ( N ) as in Theorem 1.1 .
Proof. (a) A direct computation gives
k = 2 k=2 k = 2 : D 2 ( N ) = 2 N 2 − 1 > 0 D_2(N)=2^{\frac{N}{2}}-1>0 D 2 ( N ) = 2 2 N − 1 > 0 ;
k = 3 k=3 k = 3 : D 3 ( N ) = ( 2 3 ) N − 2 2 ( 3 N 2 − 2 ⋅ 2 N 2 ) > 0 D_3(N)=\left(\frac{2}{3}\right)^{\frac{N-2}{2}}\left(3^{\frac{N}{2}}-2\cdot2^{\frac{N}{2}}\right)>0 D 3 ( N ) = ( 3 2 ) 2 N − 2 ( 3 2 N − 2 ⋅ 2 2 N ) > 0 ;
k = 4 k=4 k = 4 : D 4 ( N ) = 2 N 2 − 1 > 0 D_4(N)=2^{\frac{N}{2}}-1>0 D 4 ( N ) = 2 2 N − 1 > 0 .
(b) We use the classical values
sin 2 ( π 5 ) = 5 − 5 8 and sin 2 ( 2 π 5 ) = 5 + 5 8 , \sin^2\left(\frac{\pi}{5}\right)=\frac{5-\sqrt{5}}{8}\quad\text{and}\quad\sin^2\left(\frac{2\pi}{5}\right)=\frac{5+\sqrt{5}}{8}, sin 2 ( 5 π ) = 8 5 − 5 and sin 2 ( 5 2 π ) = 8 5 + 5 ,
and we recall that sin ( 3 π 5 ) = sin ( 2 π 5 ) \sin\left(\frac{3\pi}{5}\right)=\sin\left(\frac{2\pi}{5}\right) sin ( 5 3 π ) = sin ( 5 2 π ) and sin ( 4 π 5 ) = sin ( π 5 ) \sin\left(\frac{4\pi}{5}\right)=\sin\left(\frac{\pi}{5}\right) sin ( 5 4 π ) = sin ( 5 π ) .
Set
a : = ( 2 sin 2 ( π 5 ) ) − 1 = 4 5 − 5 and b : = ( 2 sin 2 ( 2 π 5 ) ) − 1 = 4 5 + 5 . \mathfrak{a}:=\left(2\sin^{2}\left(\frac{\pi}{5}\right)\right)^{-1}=\frac{4}{5-\sqrt{5}}
\quad\text{and}\quad
\mathfrak{b}:=\left(2\sin^{2}\left(\frac{2\pi}{5}\right)\right)^{-1}=\frac{4}{5+\sqrt{5}}. a := ( 2 sin 2 ( 5 π ) ) − 1 = 5 − 5 4 and b := ( 2 sin 2 ( 5 2 π ) ) − 1 = 5 + 5 4 .
hence
sin − ( N − 2 ) ( π j 5 ) = 2 N − 2 2 a N − 2 2 for j = 1 , 4 \sin^{-(N-2)}\left(\frac{\pi j}{5}\right)=2^{\frac{N-2}{2}}\mathfrak{a}^{\frac{N-2}{2}}\quad\text{for }j=1,4 sin − ( N − 2 ) ( 5 π j ) = 2 2 N − 2 a 2 N − 2 for j = 1 , 4
and
sin − ( N − 2 ) ( π j 5 ) = 2 N − 2 2 b N − 2 2 for j = 2 , 3. \sin^{-(N-2)}\left(\frac{\pi j}{5}\right)=2^{\frac{N-2}{2}}\mathfrak{b}^{\frac{N-2}{2}}\quad\text{for }j=2,3. sin − ( N − 2 ) ( 5 π j ) = 2 2 N − 2 b 2 N − 2 for j = 2 , 3.
Therefore
D 5 ( N ) = 2 N − 2 2 [ 5 − 2 ( a N − 2 2 + b N − 2 2 ) ] . D_{5}(N)=2^{\frac{N-2}{2}}\left[5-2\left(\mathfrak{a}^{\frac{N-2}{2}}+\mathfrak{b}^{\frac{N-2}{2}}\right)\right]. D 5 ( N ) = 2 2 N − 2 [ 5 − 2 ( a 2 N − 2 + b 2 N − 2 ) ] .
Note that
a 2 + b 2 = 2 ( 1 + 1 5 ) = 12 5 . \mathfrak{a}^{2}+\mathfrak{b}^{2}=2\left(1+\frac{1}{5}\right)=\frac{12}{5}. a 2 + b 2 = 2 ( 1 + 5 1 ) = 5 12 .
Moreover, for p > 0 p>0 p > 0 , let
M p : = ( a p + b p 2 ) 1 p . M_{p}:=\left(\frac{\mathfrak{a}^{p}+\mathfrak{b}^{p}}{2}\right)^{\frac{1}{p}}. M p := ( 2 a p + b p ) p 1 .
be the power mean of order p p p of a , b \mathfrak{a},\mathfrak{b} a , b ; it is well known that p ↦ M p p\mapsto M_{p} p ↦ M p is nondecreasing and a p + b p = 2 M p p \mathfrak{a}^{p}+\mathfrak{b}^{p}=2M_{p}^{p} a p + b p = 2 M p p and in particular M 2 = ( 6 5 ) 1 2 M_{2}=\left(\frac{6}{5}\right)^{\frac{1}{2}} M 2 = ( 5 6 ) 2 1 . Therefore, it is easy to check that
D 5 ( 6 ) = 4 [ 5 − 2 ( a 2 + b 2 ) ] = 4 5 . D_{5}(6)=4\left[5-2\left(\mathfrak{a}^{2}+\mathfrak{b}^{2}\right)\right]=\frac{4}{5}. D 5 ( 6 ) = 4 [ 5 − 2 ( a 2 + b 2 ) ] = 5 4 .
Instead, since M 3 2 ≤ M 2 M_{\frac{3}{2}}\leq M_{2} M 2 3 ≤ M 2 , we get that
a 3 2 + b 3 2 = 2 M 3 2 3 2 ≤ 2 M 2 3 2 = 2 ( 6 5 ) 3 4 . \mathfrak{a}^{\frac{3}{2}}+\mathfrak{b}^{\frac{3}{2}}=2M_{\frac{3}{2}}^{\frac{3}{2}}\leq2M_{2}^{\frac{3}{2}}=2\left(\frac{6}{5}\right)^{\frac{3}{4}}. a 2 3 + b 2 3 = 2 M 2 3 2 3 ≤ 2 M 2 2 3 = 2 ( 5 6 ) 4 3 .
Thus
D 5 ( 5 ) ≥ 2 3 2 [ 5 − 4 ( 6 5 ) 3 4 ] > 0. D_{5}(5)\geq2^{\frac{3}{2}}\left[5-4\left(\frac{6}{5}\right)^{\frac{3}{4}}\right]>0. D 5 ( 5 ) ≥ 2 2 3 [ 5 − 4 ( 5 6 ) 4 3 ] > 0.
Now, if we consider the case in which N ≥ 7 N\geq7 N ≥ 7 , using the fact that N − 2 2 ≥ 5 2 > 2 \frac{N-2}{2}\geq\frac{5}{2}>2 2 N − 2 ≥ 2 5 > 2 and so M N − 2 2 ≥ M 2 > 1 M_{\frac{N-2}{2}}\geq M_{2}>1 M 2 N − 2 ≥ M 2 > 1 , we have that
a N − 2 2 + b N − 2 2 = 2 M N − 2 2 N − 2 2 ≥ 2 M 2 N − 2 2 = 2 ( 6 5 ) N − 2 4 ≥ 2 ( 6 5 ) 5 4 \mathfrak{a}^{\frac{N-2}{2}}+\mathfrak{b}^{\frac{N-2}{2}}=2M_{\frac{N-2}{2}}^{\frac{N-2}{2}}\geq2M_{2}^{\frac{N-2}{2}}=2\left(\frac{6}{5}\right)^{\frac{N-2}{4}}\geq2\left(\frac{6}{5}\right)^{\frac{5}{4}} a 2 N − 2 + b 2 N − 2 = 2 M 2 N − 2 2 N − 2 ≥ 2 M 2 2 N − 2 = 2 ( 5 6 ) 4 N − 2 ≥ 2 ( 5 6 ) 4 5
and therefore
D 5 ( N ) ≤ 2 N − 2 2 [ 5 − 4 ( 6 5 ) 5 4 ] < 0. D_{5}(N)\leq2^{\frac{N-2}{2}}\left[5-4\left(\frac{6}{5}\right)^{\frac{5}{4}}\right]<0. D 5 ( N ) ≤ 2 2 N − 2 [ 5 − 4 ( 5 6 ) 4 5 ] < 0.
(c) We use the classical identity (see [9 ] )
∑ j = 1 k − 1 1 sin 2 ( π j k ) = k 2 − 1 3 . (38) \sum_{j=1}^{k-1}\frac{1}{\sin^{2}\left(\frac{\pi j}{k}\right)}=\frac{k^{2}-1}{3}. \tag*{(38)} j = 1 ∑ k − 1 sin 2 ( k π j ) 1 = 3 k 2 − 1 . (38)
The numbers y j : = sin − 2 ( π j k ) y_j := \sin^{-2}\left(\frac{\pi j}{k}\right) y j := sin − 2 ( k π j ) , j = 1 , … , k − 1 j=1,\ldots,k-1 j = 1 , … , k − 1 , are positive and y ↦ y N − 2 2 y \mapsto y^{\frac{N-2}{2}} y ↦ y 2 N − 2 is convex on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) since N − 2 2 ≥ 1 \frac{N-2}{2} \ge1 2 N − 2 ≥ 1 . By Jensen’s inequality and (38) ,
∑ j = 1 k − 1 1 sin N − 2 ( π j k ) = ∑ j = 1 k − 1 y j N − 2 2 ≥ ( k − 1 ) ( 1 k − 1 ∑ j = 1 k − 1 y j ) N − 2 2 = ( k − 1 ) ( k + 1 3 ) N − 2 2 . \sum_{j=1}^{k-1}\frac{1}{\sin^{N-2}\left(\frac{\pi j}{k}\right)}
=\sum_{j=1}^{k-1}y_j^{\frac{N-2}{2}}
\ge(k-1)\left(\frac{1}{k-1}\sum_{j=1}^{k-1}y_j\right)^{\frac{N-2}{2}}
=(k-1)\left(\frac{k+1}{3}\right)^{\frac{N-2}{2}}. j = 1 ∑ k − 1 sin N − 2 ( k π j ) 1 = j = 1 ∑ k − 1 y j 2 N − 2 ≥ ( k − 1 ) ( k − 1 1 j = 1 ∑ k − 1 y j ) 2 N − 2 = ( k − 1 ) ( 3 k + 1 ) 2 N − 2 .
Therefore
D k ( N ) ≤ k 2 N − 2 2 − ( k − 1 ) ( k + 1 3 ) N − 2 2 = k 2 N − 2 2 [ 1 − k − 1 k ( k + 1 6 ) N − 2 2 ] . (39) D_k(N)\le k2^{\frac{N-2}{2}}-(k-1)\left(\frac{k+1}{3}\right)^{\frac{N-2}{2}}
=k2^{\frac{N-2}{2}}\left[1-\frac{k-1}{k}\left(\frac{k+1}{6}\right)^{\frac{N-2}{2}}\right]. \tag*{(39)} D k ( N ) ≤ k 2 2 N − 2 − ( k − 1 ) ( 3 k + 1 ) 2 N − 2 = k 2 2 N − 2 [ 1 − k k − 1 ( 6 k + 1 ) 2 N − 2 ] . (39)
Let now k ≥ 6 k\ge6 k ≥ 6 . Then k + 1 6 > 1 \frac{k+1}{6}>1 6 k + 1 > 1 , so
( k + 1 6 ) N − 2 2 ≥ ( k + 1 6 ) 3 2 \left(\frac{k+1}{6}\right)^{\frac{N-2}{2}}\ge\left(\frac{k+1}{6}\right)^{\frac{3}{2}} ( 6 k + 1 ) 2 N − 2 ≥ ( 6 k + 1 ) 2 3
for N ≥ 5 N\ge5 N ≥ 5 ; moreover k − 1 k \frac{k-1}{k} k k − 1 and k + 1 6 \frac{k+1}{6} 6 k + 1 are increasing in k k k . Hence
k − 1 k ( k + 1 6 ) N − 2 2 ≥ 5 6 ( 7 6 ) 3 2 > 1. (40) \frac{k-1}{k}\left(\frac{k+1}{6}\right)^{\frac{N-2}{2}}
\ge\frac{5}{6}\left(\frac{7}{6}\right)^{\frac{3}{2}}>1. \tag*{(40)} k k − 1 ( 6 k + 1 ) 2 N − 2 ≥ 6 5 ( 6 7 ) 2 3 > 1. (40)
Thus, using (39) and (40) we get that D k ( N ) < 0 D_k(N)<0 D k ( N ) < 0 for every k ≥ 6 k\ge6 k ≥ 6 and every N ≥ 5 N\ge5 N ≥ 5 and this concludes the proof. □
Remark 4.5. For N = 4 N=4 N = 4 the identity (38) gives
D k ( 4 ) = 2 k − k 2 − 1 3 , D_k(4)=2k-\frac{k^2-1}{3}, D k ( 4 ) = 2 k − 3 k 2 − 1 ,
i.e. c k , 4 = − k 2 + 6 k + 1 12 c_{k,4}=\frac{-k^2+6k+1}{12} c k , 4 = 12 − k 2 + 6 k + 1 , which is positive exactly for k ≤ 6 k\le6 k ≤ 6 : this is the coefficient appearing in the 4 D 4D 4 D -case (see [24 ] ).
Proof of Theorem 1.1 (b). If 2 ≤ k ≤ k ∗ ( N ) 2\le k\le k^*(N) 2 ≤ k ≤ k ∗ ( N ) , then c k , N > 0 c_{k,N}>0 c k , N > 0 by Proposition 4.4; hence Φ k , N \Phi_{k,N} Φ k , N satisfies (16) by Proposition 4.3, and Proposition 2.9 applies in case (B). The convergence of the whole family follows from Corollary 4.9. □
The regular part as a power series
Throughout this subsection λ = N − 2 2 \lambda=\frac{N-2}{2} λ = 2 N − 2 , so that, with the notation of (33) ,
h r ( t ) = ( 1 − 2 r 2 t + r 4 ) − λ , h_r(t)=(1-2r^2t+r^4)^{-\lambda}, h r ( t ) = ( 1 − 2 r 2 t + r 4 ) − λ ,
with t ∈ [ − 1 , 1 ] t\in[-1,1] t ∈ [ − 1 , 1 ] and r ∈ ( 0 , 1 ) r\in(0,1) r ∈ ( 0 , 1 ) .
We recall from Section 3 the following expression of Gegenbauer polynomials, namely
C m λ ( cos φ ) = ∑ l = 0 m γ m , l cos ( ( m − 2 l ) φ ) , γ m , l = Γ ( λ + l ) Γ ( λ + m − l ) Γ ( λ ) 2 l ! ( m − l ) ! > 0. (41) C_m^\lambda(\cos\varphi)=\sum_{l=0}^{m}\gamma_{m,l}\cos((m-2l)\varphi),
\qquad
\gamma_{m,l}=\frac{\Gamma(\lambda+l)\Gamma(\lambda+m-l)}{\Gamma(\lambda)^2l!(m-l)!}>0. \tag*{(41)} C m λ ( cos φ ) = l = 0 ∑ m γ m , l cos (( m − 2 l ) φ ) , γ m , l = Γ ( λ ) 2 l ! ( m − l )! Γ ( λ + l ) Γ ( λ + m − l ) > 0. (41)
and we set, for n ≥ 0 n\ge0 n ≥ 0 ,
b n : = Γ ( λ + n ) Γ ( λ ) n ! , so that γ 2 n , n = b n 2 . (42) b_n:=\frac{\Gamma(\lambda+n)}{\Gamma(\lambda)n!},
\qquad\text{so that}\qquad\gamma_{2n,n}=b_n^2. \tag*{(42)} b n := Γ ( λ ) n ! Γ ( λ + n ) , so that γ 2 n , n = b n 2 . (42)
Lemma 4.6. Let N ≥ 4 N \ge4 N ≥ 4 and k ≥ 2 k \ge2 k ≥ 2 . Then, for every r ∈ ( 0 , 1 ) r \in(0,1) r ∈ ( 0 , 1 ) ,
R k , N ( r ) = ∑ m = 1 ∞ a m r 2 m , a m : = k ∑ 0 ≤ l ≤ m k ∣ m − 2 l γ m , l − k C m λ ( 0 ) , (43) R_{k,N}(r)=\sum_{m=1}^{\infty}a_m r^{2m},\qquad a_m:=k\sum_{\substack{0\le l\le m\\ k\mid m-2l}}\gamma_{m,l}-kC_m^\lambda(0), \tag*{(43)} R k , N ( r ) = m = 1 ∑ ∞ a m r 2 m , a m := k 0 ≤ l ≤ m k ∣ m − 2 l ∑ γ m , l − k C m λ ( 0 ) , (43) where the series converges absolutely, a m ≥ 0 a_m\ge0 a m ≥ 0 for every m ≥ 1 m\ge1 m ≥ 1 and a 2 > 0 a_2>0 a 2 > 0 .
Proof. By the generating function of the Gegenbauer polynomials, defined in (19) , with r 2 r^2 r 2 in place to s s s , we can write
h r ( t ) = ∑ m = 0 ∞ C m λ ( t ) r 2 m , t ∈ [ − 1 , 1 ] . h_r(t)=\sum_{m=0}^{\infty}C_m^\lambda(t)r^{2m},\qquad t\in[-1,1]. h r ( t ) = m = 0 ∑ ∞ C m λ ( t ) r 2 m , t ∈ [ − 1 , 1 ] .
Moreover, by (41) , we get that
∣ C m λ ( t ) ∣ ≤ ∑ l = 0 m γ m , l = C m λ ( 1 ) \lvert C_m^\lambda(t)\rvert\le\sum_{l=0}^{m}\gamma_{m,l}=C_m^\lambda(1) ∣ C m λ ( t )∣ ≤ l = 0 ∑ m γ m , l = C m λ ( 1 )
for ∣ t ∣ ≤ 1 \lvert t\rvert\le1 ∣ t ∣ ≤ 1 and
∑ m = 0 + ∞ C m λ ( 1 ) r 2 m = ( 1 − r 2 ) − 2 λ < ∞ ; \sum_{m=0}^{+\infty}C_m^\lambda(1)r^{2m}=(1-r^2)^{-2\lambda}<\infty; m = 0 ∑ + ∞ C m λ ( 1 ) r 2 m = ( 1 − r 2 ) − 2 λ < ∞ ;
hence the series converges absolutely and uniformly in t ∈ [ − 1 , 1 ] t\in[-1,1] t ∈ [ − 1 , 1 ] . Therefore
R k , N ( r ) = ∑ j = 1 k h r ( cos θ j ) − k h r ( 0 ) = ∑ m = 0 ∞ ( ∑ j = 1 k C m λ ( cos θ j ) − k C m λ ( 0 ) ) r 2 m . R_{k,N}(r)=\sum_{j=1}^{k}h_r(\cos\theta_j)-kh_r(0)=\sum_{m=0}^{\infty}\left(\sum_{j=1}^{k}C_m^\lambda(\cos\theta_j)-kC_m^\lambda(0)\right)r^{2m}. R k , N ( r ) = j = 1 ∑ k h r ( cos θ j ) − k h r ( 0 ) = m = 0 ∑ ∞ ( j = 1 ∑ k C m λ ( cos θ j ) − k C m λ ( 0 ) ) r 2 m .
By (41) and (29) (recalling the definition of θ j \theta_j θ j ),
∑ j = 1 k C m λ ( cos θ j ) = ∑ l = 0 m γ m , l ∑ j = 0 k − 1 cos ( 2 π ( m − 2 l ) j k ) = k ∑ 0 ≤ l ≤ m k ∣ m − 2 l γ m , l , \sum_{j=1}^{k}C_m^\lambda(\cos\theta_j)=\sum_{l=0}^{m}\gamma_{m,l}\sum_{j=0}^{k-1}\cos\left(\frac{2\pi(m-2l)j}{k}\right)=k\sum_{\substack{0\le l\le m\\ k\mid m-2l}}\gamma_{m,l}, j = 1 ∑ k C m λ ( cos θ j ) = l = 0 ∑ m γ m , l j = 0 ∑ k − 1 cos ( k 2 π ( m − 2 l ) j ) = k 0 ≤ l ≤ m k ∣ m − 2 l ∑ γ m , l ,
which gives (43) ; the term m = 0 m=0 m = 0 vanishes, since C 0 λ ( t ) ≡ 1 C_0^\lambda(t)\equiv1 C 0 λ ( t ) ≡ 1 (see 8.930, 1.10 of [10 ] ). Now, we have to compute C m λ ( 0 ) C_m^\lambda(0) C m λ ( 0 ) . Taking t = 0 t=0 t = 0 in the generating function of the Gegenbauer polynomials (see [31 p. 148] ) we get, for ∣ s ∣ < 1 \lvert s\rvert<1 ∣ s ∣ < 1 ,
∑ m = 0 ∞ C m λ ( 0 ) s m = ( 1 + s 2 ) − λ . (44) \sum_{m=0}^{\infty}C_m^\lambda(0)s^m=(1+s^2)^{-\lambda}. \tag*{(44)} m = 0 ∑ ∞ C m λ ( 0 ) s m = ( 1 + s 2 ) − λ . (44)
On the other hand, using standard results about the binomial series, we can write, for ∣ s ∣ < 1 \lvert s\rvert<1 ∣ s ∣ < 1 ,
( 1 + s 2 ) − λ = ∑ m = 0 ∞ c m s m , c 2 n : = ( − 1 ) n b n , c 2 n + 1 : = 0. (1+s^2)^{-\lambda}=\sum_{m=0}^{\infty}c_m s^m,\qquad c_{2n}:=(-1)^n b_n,\qquad c_{2n+1}:=0. ( 1 + s 2 ) − λ = m = 0 ∑ ∞ c m s m , c 2 n := ( − 1 ) n b n , c 2 n + 1 := 0.
Both this series and the one in (44) are power series converging on ( − 1 , 1 ) (-1,1) ( − 1 , 1 ) to the same function f ( s ) = ( 1 + s 2 ) − λ f(s)=(1+s^2)^{-\lambda} f ( s ) = ( 1 + s 2 ) − λ . Since the coefficients of a power series are determined by its sum, namely they are equal to f ( m ) ( 0 ) m ! \frac{f^{(m)}(0)}{m!} m ! f ( m ) ( 0 ) and so we conclude that C m λ ( 0 ) = c m C_m^\lambda(0)=c_m C m λ ( 0 ) = c m for every m m m , i.e.
C 2 n + 1 λ ( 0 ) = 0 , C 2 n λ ( 0 ) = ( − 1 ) n b n , n ≥ 0. (45) C_{2n+1}^\lambda(0)=0,\qquad C_{2n}^\lambda(0)=(-1)^n b_n,\qquad n\ge0. \tag*{(45)} C 2 n + 1 λ ( 0 ) = 0 , C 2 n λ ( 0 ) = ( − 1 ) n b n , n ≥ 0. (45)
Now we have to check that a m ≥ 0 a_m\ge0 a m ≥ 0 . Since all the γ m , l \gamma_{m,l} γ m , l are positive, the first term in the definition of a m a_m a m is nonnegative.
If m m m is odd, or m = 2 n m=2n m = 2 n with n n n odd, then C m λ ( 0 ) ≤ 0 C_m^\lambda(0)\le0 C m λ ( 0 ) ≤ 0 and a m ≥ 0 a_m\ge0 a m ≥ 0 .
If m = 2 n m=2n m = 2 n with n n n even, the index l = n l=n l = n appears in the sum, because m − 2 n = 0 m-2n=0 m − 2 n = 0 is divisible by k k k . Hence
a 2 n k ≥ γ 2 n , n − C 2 n λ ( 0 ) = b n 2 − b n = b n ( b n − 1 ) ≥ 0. \frac{a_{2n}}{k}\ge\gamma_{2n,n}-C_{2n}^\lambda(0)=b_n^2-b_n=b_n(b_n-1)\ge0. k a 2 n ≥ γ 2 n , n − C 2 n λ ( 0 ) = b n 2 − b n = b n ( b n − 1 ) ≥ 0.
since, by the functional equations
Γ ( x + 1 ) = x Γ ( x ) and Γ ( λ + n ) = Γ ( λ ) ∏ i = 0 n − 1 ( λ + i ) , \Gamma(x+1)=x\Gamma(x)\quad\text{and}\quad\Gamma(\lambda+n)=\Gamma(\lambda)\prod_{i=0}^{n-1}(\lambda+i), Γ ( x + 1 ) = x Γ ( x ) and Γ ( λ + n ) = Γ ( λ ) i = 0 ∏ n − 1 ( λ + i ) ,
easily follows
b n = ∏ i = 0 n − 1 λ + i 1 + i ≥ 1 as λ ≥ 1. b_n=\prod_{i=0}^{n-1}\frac{\lambda+i}{1+i}\ge1\quad\text{as }\lambda\ge1. b n = i = 0 ∏ n − 1 1 + i λ + i ≥ 1 as λ ≥ 1.
Finally, with a straightforward computation we obtain that
a 2 ≥ k ( γ 2 , 1 − C 2 λ ( 0 ) ) = k ( λ 2 + λ ) > 0. a_2\ge k\left(\gamma_{2,1}-C_2^\lambda(0)\right)=k(\lambda^2+\lambda)>0. a 2 ≥ k ( γ 2 , 1 − C 2 λ ( 0 ) ) = k ( λ 2 + λ ) > 0.
Remark 4.7. Lemma 4.6 gives a second proof of the positivity R k , N > 0 R_{k,N}>0 R k , N > 0 in Lemma 4.2, and shows moreover that R k , N R_{k,N} R k , N is strictly increasing on ( 0 , 1 ) (0,1) ( 0 , 1 ) .
The shape of Φ k , N \Phi_{k,N} Φ k , N
Proposition 4.8. Let N ≥ 5 N\ge5 N ≥ 5 and k ≥ 2 k\ge2 k ≥ 2 , and set
g ( r ) : = r N − 1 Φ k , N ′ ( r ) , r ∈ ( 0 , 1 ) . g(r):=r^{N-1}\Phi'_{k,N}(r),\qquad r\in(0,1). g ( r ) := r N − 1 Φ k , N ′ ( r ) , r ∈ ( 0 , 1 ) .
Then g g g is strictly increasing on ( 0 , 1 ) (0,1) ( 0 , 1 ) and
g ( r ) → − ( N − 2 ) c k , N as r → 0 + and g ( r ) → + ∞ as r → 1 − . g(r)\to-(N-2)c_{k,N}\quad\text{as }r\to0^+\quad\text{and}\quad g(r)\to+\infty\quad\text{as }r\to1^-. g ( r ) → − ( N − 2 ) c k , N as r → 0 + and g ( r ) → + ∞ as r → 1 − .
Consequently:
if c k , N < 0 c_{k,N}<0 c k , N < 0 , then Φ k , N ′ > 0 \Phi'_{k,N}>0 Φ k , N ′ > 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) ; in particular Φ k , N \Phi_{k,N} Φ k , N has no critical points in ( 0 , 1 ) (0,1) ( 0 , 1 ) ;
if c k , N > 0 c_{k,N}>0 c k , N > 0 , then Φ k , N \Phi_{k,N} Φ k , N has exactly one critical point r 0 ∈ ( 0 , 1 ) r_0\in(0,1) r 0 ∈ ( 0 , 1 ) , which is its unique global minimum point, and Φ k , N ′ ′ ( r 0 ) > 0 \Phi''_{k,N}(r_0)>0 Φ k , N ′′ ( r 0 ) > 0 .
Proof. By Lemmas 4.1 and 4.6, differentiating the power series term by term,
g ( r ) = ∑ m = 1 ∞ 2 m a m r 2 m + N − 2 − ( N − 2 ) c k , N . g(r)=\sum_{m=1}^{\infty}2m\,a_m r^{2m+N-2}-(N-2)c_{k,N}. g ( r ) = m = 1 ∑ ∞ 2 m a m r 2 m + N − 2 − ( N − 2 ) c k , N .
Since a m ≥ 0 a_m\ge0 a m ≥ 0 and a 2 > 0 a_2>0 a 2 > 0 , it follows that g g g is strictly increasing and g ( r ) → − ( N − 2 ) c k , N g(r)\to-(N-2)c_{k,N} g ( r ) → − ( N − 2 ) c k , N as r → 0 + r\to0^+ r → 0 + . Moreover, by monotone convergence,
g ( r ) → ∑ m = 1 + ∞ 2 m a m − ( N − 2 ) c k , N g(r)\to\sum_{m=1}^{+\infty}2m\,a_m-(N-2)c_{k,N} g ( r ) → m = 1 ∑ + ∞ 2 m a m − ( N − 2 ) c k , N
as r → 1 − r\to1^- r → 1 − and, by Lemma 4.2, we obtain that
∑ m = 1 + ∞ 2 m a m ≥ ∑ m = 1 + ∞ a m = lim r → 1 − R k , N ( r ) = + ∞ . \sum_{m=1}^{+\infty}2m\,a_m\ge\sum_{m=1}^{+\infty}a_m=\lim_{r\to1^-}R_{k,N}(r)=+\infty. m = 1 ∑ + ∞ 2 m a m ≥ m = 1 ∑ + ∞ a m = r → 1 − lim R k , N ( r ) = + ∞.
Moreover,
if c k , N < 0 c_{k,N}<0 c k , N < 0 , then g > − ( N − 2 ) c k , N > 0 g>-(N-2)c_{k,N}>0 g > − ( N − 2 ) c k , N > 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) ;
if c k , N > 0 c_{k,N}>0 c k , N > 0 , then g g g changes sign exactly once, at some r 0 ∈ ( 0 , 1 ) r_0\in(0,1) r 0 ∈ ( 0 , 1 ) , so Φ k , N \Phi_{k,N} Φ k , N is strictly decreasing on ( 0 , r 0 ) (0,r_0) ( 0 , r 0 ) and strictly increasing on ( r 0 , 1 ) (r_0,1) ( r 0 , 1 ) . Finally, since Φ k , N ′ ( r 0 ) = 0 \Phi'_{k,N}(r_0)=0 Φ k , N ′ ( r 0 ) = 0 ,
r 0 N − 1 Φ k , N ′ ′ ( r 0 ) = g ′ ( r 0 ) ≥ 4 a 2 ( N + 2 ) r 0 N + 1 > 0. r_0^{N-1}\Phi''_{k,N}(r_0)=g'(r_0)\ge4a_2(N+2)r_0^{N+1}>0. r 0 N − 1 Φ k , N ′′ ( r 0 ) = g ′ ( r 0 ) ≥ 4 a 2 ( N + 2 ) r 0 N + 1 > 0.
Corollary 4.9. Let N ≥ 5 N\ge5 N ≥ 5 and 2 ≤ k ≤ k ∗ ( N ) 2\le k\le k^*(N) 2 ≤ k ≤ k ∗ ( N ) . Then Γ \Gamma Γ has exactly one critical point in M = ( 0 , 1 ) × ( 0 , ∞ ) \mathcal{M}=(0,1)\times(0,\infty) M = ( 0 , 1 ) × ( 0 , ∞ ) , namely ( r 0 , c d ( r 0 ) ) (r_0,cd(r_0)) ( r 0 , c d ( r 0 )) , where r 0 r_0 r 0 is the unique global minimum point of Φ k , N \Phi_{k,N} Φ k , N , and it is non-degenerate. In particular, the solutions of Theorem 1.1 (b) satisfy r ε → r 0 r_\varepsilon\to r_0 r ε → r 0 and d ε → d 0 d_\varepsilon\to d_0 d ε → d 0 along the whole family ε → 0 \varepsilon\to0 ε → 0 .
Proof. It follows from Proposition 4.4 , Proposition 4.8 (ii) and Lemma 2.8 . The convergence of the whole family follows from the uniqueness of the global minimum point. □ \square □
Non-existence for k > k ∗ ( N ) k>k^*(N) k > k ∗ ( N )
Theorem 4.10. Let N ≥ 5 N\ge5 N ≥ 5 and k > k ∗ ( N ) k>k^*(N) k > k ∗ ( N ) , and let K ⊂ M = ( 0 , 1 ) × ( 0 , ∞ ) \mathcal{K}\subset\mathcal{M}=(0,1)\times(0,\infty) K ⊂ M = ( 0 , 1 ) × ( 0 , ∞ ) be compact. Then there exists ε K > 0 \varepsilon_{\mathcal{K}}>0 ε K > 0 such that, for every ε ∈ ( 0 , ε K ) \varepsilon\in(0,\varepsilon_{\mathcal{K}}) ε ∈ ( 0 , ε K ) , problem (1.1) in B 1 ( 0 ) B_1(0) B 1 ( 0 ) has no solution of the form
u = V y + ϕ , y ∈ K , ϕ ∈ Z ⊥ , ∥ ϕ ∥ H 0 1 ≤ C ϑ ( μ ) , u=V_y+\phi,\qquad y\in\mathcal{K},\qquad\phi\in Z^\perp,\qquad\|\phi\|_{H_0^1}\le C\vartheta(\mu), u = V y + ϕ , y ∈ K , ϕ ∈ Z ⊥ , ∥ ϕ ∥ H 0 1 ≤ C ϑ ( μ ) , where V y V_y V y is the ansatz (2.11) in case (B) and C C C is the constant of Proposition 2.5 .
Proof. Choose η > 0 \eta>0 η > 0 such that K ⊂ int P η \mathcal{K}\subset\operatorname{int}\mathcal{P}_\eta K ⊂ int P η . Assume by contradiction that there are ε n → 0 \varepsilon_n\to0 ε n → 0 , y n ∈ K y_n\in\mathcal{K} y n ∈ K and solutions u n = V y n + ϕ n u_n=V_{y_n}+\phi_n u n = V y n + ϕ n of (1.1) with ε = ε n \varepsilon=\varepsilon_n ε = ε n , ϕ n ∈ Z ⊥ \phi_n\in Z^\perp ϕ n ∈ Z ⊥ and ∥ ϕ n ∥ ≤ C ϑ ( μ n ) \|\phi_n\|\le C\vartheta(\mu_n) ∥ ϕ n ∥ ≤ C ϑ ( μ n ) . Since u n u_n u n solves (1.1), the projected equation of Proposition 2.5 holds, so that ϕ n = ϕ y n \phi_n=\phi_{y_n} ϕ n = ϕ y n by uniqueness. Hence ∇ J ~ ε n ( y n ) = 0 \nabla\widetilde{J}_{\varepsilon_n}(y_n)=0 ∇ J ε n ( y n ) = 0 , because ∂ y i J ~ ε ( y ) = J ε ′ ( V y + ϕ y ) [ ∂ y i ( V y + ϕ y ) ] \partial_{y_i}\widetilde{J}_\varepsilon(y)=J'_\varepsilon(V_y+\phi_y)[\partial_{y_i}(V_y+\phi_y)] ∂ y i J ε ( y ) = J ε ′ ( V y + ϕ y ) [ ∂ y i ( V y + ϕ y )] . By Proposition 2.7 ,
0 = μ n − ( N − 2 ) ∇ J ~ ε n ( y n ) = ℓ ∇ Γ A , B ( y n ) + ∇ ϖ ε n ( y n ) , ∇ ϖ ε n → 0 uniformly on P η . 0=\mu_n^{-(N-2)}\nabla\widetilde{J}_{\varepsilon_n}(y_n)=\ell\nabla\Gamma_{A,B}(y_n)+\nabla\varpi_{\varepsilon_n}(y_n),\qquad\nabla\varpi_{\varepsilon_n}\to0\text{ uniformly on }\mathcal{P}_\eta. 0 = μ n − ( N − 2 ) ∇ J ε n ( y n ) = ℓ ∇ Γ A , B ( y n ) + ∇ ϖ ε n ( y n ) , ∇ ϖ ε n → 0 uniformly on P η .
Up to a subsequence, y n → y ∗ ∈ K y_n\to y^*\in\mathcal{K} y n → y ∗ ∈ K , and y ∗ y^* y ∗ is a critical point of Γ A , B \Gamma_{A,B} Γ A , B , hence of Γ \Gamma Γ . By Lemma 2.8 ,
Φ k , N ′ ( r ∗ ) = 0 , \Phi'_{k,N}(r^*)=0, Φ k , N ′ ( r ∗ ) = 0 ,
which contradicts Proposition 4.8 (i), since c k , N < 0 c_{k,N}<0 c k , N < 0 by Proposition 4.4 . □ \square □
Proof of Theorem 1.2. Part (i) is Corollary 4.9, and part (ii) is Theorem 4.10 . □ \square □
Remark 4.11. Lemma 4.6 and Proposition 4.8 hold verbatim for N = 4 N=4 N = 4 , i.e. λ = 1 \lambda=1 λ = 1 . In particular Φ k , 4 \Phi_{k,4} Φ k , 4 is strictly increasing on ( 0 , 1 ) (0,1) ( 0 , 1 ) for every k ≥ 7 k\ge7 k ≥ 7 , since c k , 4 < 0 c_{k,4}<0 c k , 4 < 0 by Remark 4.5.
Data Availability Statements
All data generated or analysed during this study are included in this article.
Declarations
Conflicts of Interest: The author declares he has no financial interests.
References
DIPARTIMENTO DI MATEMATICA, UNIVERSITÁ DEGLI STUDI DI BARI ALDO MORO, ITALY
Email address: g.rago6@phd.uniba.it