Existence of blowing-up solutions for a nonlinear elliptic equation with Hardy potential and critical
growth ∗
Veronica FELLI † and Angela PISTOIA ‡
January 8, 2004
Abstract We prove the existence of a positive solution to
−∆u − λ
|x| 2 u = k(x)u 2
∗−1 , u ∈ D 1,2 (R N ),
which blows-up at a suitable critical point of k as the positive parameter λ goes to zero.
Keywords: Hardy potential, Sobolev critical exponent, perturbation methods.
AMS subject classification: 35J70, 35J20, 35B33.
0 Introduction
Let us consider the following nonlinear elliptic equation
−∆u − |x| λ
2u = k(x)u 2
∗−1 in R N , u ∈ D 1,2 (R N ), u > 0 in R N ,
(0.1)
where N ≥ 3, 2 ∗ = N −2 2N and the space D 1,2 (R N ) is the closure of C 0 ∞ (R N ) with respect to the norm
kuk :=
Z
R
N|∇u| 2 dx
1/2
. (0.2)
∗
The first author is supported by M.U.R.S.T. project “Variational Methods and Nonlinear Differential Equa- tions”. The second author is supported by M.U.R.S.T. project “Metodi variazionali e topologici nello studio di fenomeni non lineari”.
†
Dipartimento di Matematica e Applicazioni, Universit` a di Milano Bicocca, Via Bicocca degli Arcimboldi 8, 20126 Milano, Italy. E-mail: [email protected].
‡
Dipartimento di Metodi e Modelli Matematici, Universit` a di Roma “La Sapienza”, 00100 Roma, Italy. E-mail:
[email protected].
We point out that one of the main features of problem (0.1) is a lack of compactness due to the critical growth of the nonlinearity. Non-existence results for (0.1) can be obtained using the Pohozaev type identity proved in [13] ; more precisely it turns that any positive solution to (0.1) in the space D 1,2 (R N ) satisfies the following identity
Z
R
Nh∇k(x), xiu 2
∗dx = 0
provided k is continuously differentiable. This implies that ther are no such solution if h∇k(x), xi does not change sign and k is not constant .
When λ = 0 and k ≡ 1, it is well known (see [5], [7] and [14]) that all the solutions to (0.1) are given by
U δ,ξ (x) := α N
δ
N −22(δ 2 + |x − ξ| 2 )
N −22, x ∈ R N , ξ ∈ R N , δ > 0, (0.3) where α N = [N (N − 2)] (N −2)/4 .
The case λ 6= 0 and k ≡ 1 has been studied by Terracini in [15], where a solution to (0.1) is found explicitly provided 0 < λ < (N − 2) 2 /4 , namely:
V λ (x) := c λ (N )
[|x| 1−a (1 + |x| 2a )]
N −22,
where a := p1 − 4λ/(N − 2) 2 and c λ (N ) is a suitable positive constant. This solution is also unique, up to a conformal transformation of the form V λ ε (x) := ε (2−N )/2 V λ (x/ε) for some ε > 0.
In [13] Smets considers the case N = 4 and proves that if k satisfies the “global” condition k(0) = lim |x|→+∞ k(x), then for any λ ∈ (0, 1) problem (0.1) has at least one solution.
In [8] Schneider and the first author consider the case N ≥ 3 (even for a more general class of differential operators related to Caffarelli-Kohn-Nirenberg inequalities) and assume that k is close to a constant, i.e. k(x) = 1 + εh(x), where h satisfies a suitable “global” condition and ε is a small real perturbation parameter. By using the perturbative method by Ambrosetti and Badiale [3], they find a solution to (0.1) which is close to a radial solution to the unperturbed problem, provided 0 < λ < (N − 2) 2 /4 and ε small enough.
In [1] Abdellaoui, Peral, and the first author prove the existence of solutions to problem (0.1) blowing-up at global maximum points of k as the parameter λ goes to zero, under some suitable assumption about the local behaviour of k close to such maximum points.
Let us also mention that some related singular equations with Hardy type potential were also studied in [2, 9, 10, 12].
In this paper we are interested in finding solutions to problem (0.1) blowing-up at a suitable critical point (not necessarily a maximum point) ξ 0 of the function k, as the parameter λ goes to zero (see Definition 0.1).
Before stating our main result it is useful to introduce some notation. Let ξ 0 ∈ R N be such that k(ξ 0 ) > 0 and set (see (0.3))
W δ,ξ (x) := k(ξ 0 ) −
2∗−21U δ,ξ (x). (0.4)
Then for δ > 0 and ξ ∈ R N the functions W δ,ξ ∈ D 1,2 (R N ) solve the limit equation
−∆W = k(ξ 0 )W 2
∗−1 in R N ,
W > 0 in R N .
We can introduce the following definition.
Definition 0.1 Let u λ be a solution to problem (0.1). We say that u λ blows-up at ξ 0 as λ goes to zero if there exist δ λ > 0 and ξ λ ∈ R N such that lim λ→0 δ λ = 0, lim λ→0 ξ λ = ξ 0 and u λ − W δ
λ,ξ
λ→ 0 in D 1,2 (R N ).
Let us assume k ∈ L ∞ (R N ) ∩ C 0 (R N ). We state below our main assumption on k and ξ 0 .
(∗)
Let ξ 0 ∈ R N \ {0} be a critical point of k with k(ξ 0 ) > 0 such that for some r > 0
k(x + ξ 0 ) = k(ξ 0 ) + Q ξ
0(x) + R ξ
0(x) ∀ x ∈ B(0, r) := {x ∈ R N : |x| < r}, (0.5) where Q ξ
0: R N −→ R is a continuous function such that for some θ ∈ (2, N)
Q ξ
0(tx) = t θ Q ξ
0(x) ∀ x ∈ R N , t > 0, (0.6) and R ξ
0: B(0, r) −→ R is a continuous function such that for some ∈ (0, N − θ)
|R ξ
0(x)| ≤ c|x| θ+ ∀ x ∈ B(0, r). (0.7) Condition (∗) should be compared to the so-called flatness assumptions in the problems of prescribing scalar curvature arising in differential geometry (see for example [4] and [11]). Our main result is the following.
Theorem 0.2 Let us assume that (∗) holds with Q ξ
0(x) = P N
i=1 a i x θ i , x ∈ R N , where θ is an even integer. Assume either
θ ∈ (2, N ) and a i < 0 for any i = 1, . . . , N (0.8) or
θ ∈
max
2, N − 2 4
, N
,
N
X
i=1
a i < 0, and a i 6= 0 for any i = 1, . . . , N. (0.9)
Then for λ small enough there exists a solution to problem (0.1) which blows-up at ξ 0 as λ goes to zero (see Definition 0.1).
We quote the fact that we prove the existence of solutions to (0.1) blowing-up at a point ξ 0
which is a critical point of k with k(ξ 0 ) > 0, by only requiring a suitable condition on k in a
neighborhood of ξ 0 .
The proof of Theorem 0.2 is based on a Ljapunov-Schmidt reduction method (see, for ex- ample, [6]). We would like to point out that our method could also allow to construct solutions blowing-up at m different critical points ξ 1 , . . . , ξ m of the function k, provided assumptions (0.5–0.7) hold and either (0.8) or (0.9) are satisfied at each point ξ i .
The paper is organized as follows. In Section 1 we reduce the problem to a finite dimensional one and in Section 2 we prove Theorem 0.2, by studying the reduced problem. In Appendix A and Appendix B we prove some technical results used in Section 1 and Section 2, respectively.
In Appendix C some useful estimates are given.
1 The finite dimensional reduction
Equation (0.1) is related to the Hardy inequality. For the reader’s convenience we recall it below referring for instance to [9] for a proof.
Lemma 1.1 (Hardy inequality) If u ∈ D 1,2 (R N ), then |x| u ∈ L 2 (R N ) and C N
Z
R
Nu 2
|x| 2 dx ≤ Z
R
N|∇u| 2 dx, (1.1)
where C N = N −2 2 2
is optimal and not attained.
For any λ ∈ (0, C N ) let us introduce the Hilbert space D λ := D 1,2 (R N ) equipped with the inner product
(u, v) λ :=
Z
R
N∇u∇v − λ
|x| 2 uv
dx, which induces the norm
kuk λ :=
Z
R
N|∇u| 2 − λ
|x| 2 u 2
dx
1/2
. By the Hardy inequality (1.1) we get that
1 − λ
C N
1/2
kuk ≤ kuk λ , (1.2)
hence k · k λ and k · k are equivalent norms. By (1.2) we deduce that kuk L
2∗(R
N) ≤
S
1 − λ
C N
−1/2
kuk λ , (1.3)
where S is the best Sobolev costant of the embedding D 1,2 (R N ) ,→ L 2
∗(R N ), i.e. the best constant in the Sobolev inequality
Skuk 2 L
2∗(R
N) ≤ kuk 2 D
1,2(R
N) . (1.4)
Let i λ : D λ ,→ L 2
∗(R N ) denote the embedding.
Definition 1.2 Let i ∗ λ : L
N +22N(R N ) → D λ be the adjoint of i λ , namely (i ∗ λ u, v) λ =
Z
R
Nuv dx ∀ v ∈ D λ , or equivalently w := i ∗ λ u is the unique solution to
−∆w − λ
|x| 2 w = u in R N , w ∈ D 1,2 (R N ).
From H¨ older inequality and (1.3) we obtain easily the following lemma.
Lemma 1.3 i ∗ λ : L
N +22N(R N ) → D λ is a continuous operator and
ki ∗ λ (u)k λ ≤
S
1 − λ
C N
−1/2
kuk
L
2N
N +2
for any u ∈ L
N +22N(R N ).
Problem (0.1) is equivalent to the following one
u = i ∗ λ k(x)f (u)), u ∈ D λ , (1.5)
where f (s) := (s + ) 2
∗−1 . We will look for solutions to (1.5) of the following form u(x) = W δ,ξ (x) + φ(x)
where the rest term φ is a lower order term belonging to a suitable space. Set for j = 1, . . . , N ψ δ,ξ j (x) := ∂W δ,ξ
∂ξ j (x) = α N (N − 2)k(ξ 0 ) −
12∗−2
δ
N −22x j − ξ j (δ 2 + |x − ξ| 2 )
N2(1.6) and
ψ δ,ξ 0 (x) := ∂W δ,ξ
∂δ (x) = α N (N − 2)
2 k(ξ 0 ) −
2∗−21δ
N −42|x − ξ| 2 − δ 2
(δ 2 + |x − ξ| 2 )
N2. (1.7) The N + 1 functions ψ δ,ξ j solve the linearized problem
−∆ψ = (2 ∗ − 1)k(ξ 0 )W δ,ξ 2
∗−2 ψ in R N , ψ ∈ D 1,2 (R N ).
(1.8)
Remark 1.4 It is known that the space of the solutions to problem (1.8) in D 1,2 (R N ) is the linear space spanned by the hψ j δ,ξ i j=0,1,...,N (see [4, Lemma 3.1]).
Let
K δ,ξ λ = n
φ ∈ D λ
ψ δ,ξ j , φ
λ = 0 ∀ j = 0, 1, . . . , N o
(1.9)
and let Π λ δ,ξ : D λ −→ K δ,ξ λ be the orthogonal projection.
Proposition 1.5 There exist Z ⊂⊂ R N such that ξ 0 ∈ Z, λ 0 > 0 and δ 0 > 0 such that for any λ ∈ (0, λ 0 ), δ ∈ (0, δ 0 ) and ξ ∈ Z there exists a unique φ λ δ,ξ ∈ K δ,ξ λ such that
Π λ δ,ξ n
W δ,ξ + φ λ δ,ξ − i ∗ λ h k(x)f
W δ,ξ + φ λ δ,ξ
io
= 0. (1.10)
Moreover
kφ λ δ,ξ k λ =
O |ξ − ξ 0 | θ + δ θ + λδ 2
if N ≥ 5, O |ξ − ξ 0 | θ + δ θ + λδ 2 log δ
if N = 4, O |ξ − ξ 0 | θ + δ θ + λδ
if N = 3.
(1.11)
Proof. See Appendix B.
Let us introduce the functional J λ : D 1,2 (R N ) −→ R defined by J λ (u) = 1
2 Z
R
N|∇u| 2 dx − λ 2
Z
R
Nu 2
|x| 2 dx − 1 2 ∗
Z
R
Nk(x)(u + ) 2
∗dx.
It is well known that u is a critical point of J λ if and only if u is a solution to problem (0.1).
The following result allows to reduce the problem of existence of solutions to (0.1) to the existence of critical points of a real-valued function of N + 1 variables.
Proposition 1.6 There exist Z ⊂⊂ R N such that ξ 0 ∈ Z, λ 0 > 0 and δ 0 > 0 such that for any λ ∈ (0, λ 0 ), δ ∈ (0, δ 0 ) and ξ ∈ Z the function W δ,ξ + φ λ δ,ξ is a solution to problem (0.1) if and only if (δ, ξ) is a critical point of the function
J e λ (δ, ξ) := J λ
W δ,ξ + φ λ δ,ξ
. (1.12)
Proof. See Appendix B.
2 The reduced problem
Let us introduce the function I λ : R N × R + −→ R defined by (see (1.12))
I λ (ζ, d) := e J λ δ(d, λ), ξ(ζ, d, λ), (2.1) where
δ(d, λ) :=
dλ
θ−11if N = 3, dλ
θ−21| log λ|
θ−21if N = 4, dλ
θ−21if N ≥ 5,
(2.2)
and
ξ(ζ, d, λ) := ξ 0 + δ(d, λ)ζ. (2.3)
Proposition 2.1 Let K ⊂ R N be a compact set and b > a > 0 be fixed. If (∗) is satisfied and θ ∈ (2, N ), there holds
I λ (d, ζ) =
a 3 (ξ 0 ) − Γ ξ
0(d, ζ)λ
θ−1θ+ o λ
θ−1θ, if N = 3, a 4 (ξ 0 ) − Γ ξ
0(d, ζ)λ
θ−2θ| log λ|
θ−2θ+ o λ
θ−2θ| log λ|
θ−2θ, if N = 4, a N (ξ 0 ) − Γ ξ
0(d, ζ)λ
θ−2θ+ o λ
θ−2θ, if N ≥ 5,
(2.4)
C 0 -uniformly with respect to ζ in K and to d in [a, b]. If, in addition, θ ∈ max 2, N −2 4 , N , estimate (2.4) holds C 1 -uniformly with respect to ζ in K and to d in [a, b].
The function Γ ξ
0: R + × R N → R in (2.4) is defined by
Γ ξ
0(d, ζ) = b N (ξ 0 )d θ Z
R
NQ ξ
0(y + ζ) (1 + |y| 2 ) N dy +
c 3 (ξ 0 )d if N = 3, c N (ξ 0 )d 2 if N ≥ 4,
(2.5)
and
a N (ξ 0 ) : = 1 2 − 1
2 ∗
α 2 N
∗[k(ξ 0 )] −
2∗−22Z
R
Ndx
(1 + |x| 2 ) N , (2.6)
b N (ξ 0 ) : = 1
2 ∗ α 2 N
∗[k(ξ 0 )] −
2∗−22∗, (2.7)
c N (ξ 0 ) : =
1
2 α 2 3 [k(ξ 0 )] −1/2 Z
R
N1
|x − ξ 0 | 2 1
|x| 2 dx, if N = 3, α 2 4 k(ξ 0 ) −1
2(θ − 2)|ξ 0 | 2 , if N = 4,
1
2 α 2 N [k(ξ 0 )] −
2∗−221
|ξ 0 | 2 Z
R
Ndx
(1 + |x| 2 ) N −2 , if N ≥ 5.
(2.8)
Proof. See Appendix C.
Definition 2.2 Let g : D → R be a C 1 -function, where D ⊂ R m is an open set, m ∈ IN. Let x 0 be a critical point of g.
(i) We say that x 0 is a C 0 -stable critical point of g if for any sequence g n : D → R with g n → g C 0 -uniformly in D there exists x n ∈ D such that ∇g n (x n ) = 0 and x n → x 0 . (ii) We say that x 0 is a C 1 -stable critical point of g if for any sequence g n : D → R with
g n → g C 1 -uniformly in D there exists x n ∈ D such that ∇g n (x n ) = 0 and x n → x 0 .
We point out that any strict local maximum point or strict local minimum point of g is a C 0 -
stable critical point of g. Moreover any non degenerate critical point of g is a C 1 -stable critical
point of g. Let us remark that a non degenerate critical point is not necessarily a C 0 -stable
critical point as the following example shows.
Example 2.3 Let g : R 2 → R, g(x, y) = xy. (0, 0) is a C 1 -stable critical point, but it is not a C 0 -stable critical point.
Proof. It is easy to verify that (0, 0) is a non degenerate critical point. Let us consider the sequence of functions g n (x, y) := xy + 1 n sin(nx), n ∈ IN. The sequence {g n } n∈IN converges to g uniformly in R 2 . If (x n , y n ) is a critical point of g n , then a direct computation shows that necessarily x n = 0 and y n = −1 for any n ∈ IN. Therefore there can not exist any sequence of critical points of g n converging to (0, 0). Hence (0, 0) is not a C 0 -stable critical point.
Theorem 2.4 Assume either θ ∈ (2, N ) and (d 0 , ζ 0 ) is a C 0 -stable critical point of Γ ξ
0(d, ζ) or θ ∈ max 2, N −2 4 , N and (d 0 , ζ 0 ) is a C 1 -stable critical point of Γ ξ
0(d, ζ). Then for λ small enough there exists a critical point (δ λ , ξ λ ) of the function e J defined in (1.12).
Proof. By Proposition 2.1 and Definition 2.2 it follows that there exists a critical point (d λ , ζ λ ) of I λ such that lim λ→0 d λ = d 0 and lim λ→0 ζ λ = ζ 0 . If δ λ is defined as in (2.2) and ξ λ is defined as in (2.3), it is easy to check that (δ λ , ξ λ ) is a critical point of e J .
Proposition 2.5 Assume Q ξ
0(x) = P N
i=1 a i x θ i , x ∈ R N , where θ ∈ (2, N ) is an even integer. If a i < 0 for any i = 1, . . . , N, there exists a strict maximum point of the function Γ ξ
0, which is a C 0 -stable critical point according to Definition 2.2.
Proof. Since necessarily N ≥ 5, we can write function Γ ξ
0as Γ(d, ζ) = b N d θ f (ζ)+c N d 2 , where
f (ζ) :=
Z
R
NQ ξ
0(y + ζ) (1 + |y| 2 ) N dy =
N
X
i=1
a i Z
R
N(y i + ζ i ) θ
(1 + |y| 2 ) N dy. (2.9) We claim that
f (ζ) < f (0) < 0 ∀ ζ ∈ R N \ {0}. (2.10) Indeed
Z
R
N(y i + ζ i ) θ (1 + |y| 2 ) N dy =
Z
R
Ny i θ
(1 + |y| 2 ) N dy +
θ
X
k=2 keven
θ k
Z
R
Ny θ−k i ζ i k
(1 + |y| 2 ) N dy ≥ Z
R
Ny i θ (1 + |y| 2 ) N dy with equality only for ζ i = 0 and claim (2.10) follows, since a i < 0 for any i. Let us set d 0 :=
h −2c
Nb
Nθf (0)
i
θ−21. Note that the definition of d 0 makes sense because of (2.10). We claim that (d 0 , 0) is a strict maximum point of Γ ξ
0. (2.11) Indeed by (2.10) we deduce that for any ζ it holds
Γ(d, ζ) ≤ Γ(d, 0) ≤ Γ(d 0 , 0) = max
d∈R
+Γ(d, 0).
If (d, ζ) 6= (d 0 , 0) then either d 6= d 0 and hence Γ(d, 0) < Γ(d 0 , 0) or ζ 6= 0 and hence from (2.10)
Γ(d, ζ) < Γ(d, 0). Therefore for any (d, ζ) 6= (d 0 , 0) we have Γ(d, ζ) < Γ(d 0 , 0).
Proposition 2.6 Assume Q ξ
0(x) = P N
i=1 a i x θ i , x ∈ R N , where θ ∈ (2, N ) is an even integer. If P N
i=1 a i < 0 and a i 6= 0 for any i = 1, . . . , N, there exists a non degenerate critical point of the function Γ ξ
0, which is a C 1 -stable critical point according to Definition 2.2.
Proof. As above, since N ≥ 5, we can write function Γ ξ
0as Γ(d, ζ) = b N d θ f (ζ) + c N d 2 , where f is defined in (2.9). We point out that
f (0) = d N N
X
i=1
a i < 0,
where d N := R
R
Ny 1 θ (1 + |y| 2 ) −N dy. Therefore it makes sense to set d 0 :=
h −2c
Nb
Nθf (0)
i
θ−21. We claim that
(d 0 , 0) is a nondegenerate critical point of Γ ξ
0. (2.12) Let us compute
∂f
∂ζ i
(0) = a i θ Z
R
Ny i θ−1
(1 + |y| 2 ) N dy = 0 (2.13)
and
∂ 2 f
∂ζ i ∂ζ j
(0) =
a i θ(θ − 1)e N if i = j,
0 if i 6= j,
(2.14)
where e N := R
R
Ny 1 θ−2 (1 + |y| 2 ) −N dy. By (2.13) we deduce that (d 0 , 0) is a critical point of Γ, since
∂Γ
∂d (d 0 , 0) = 0 and ∇ ζ Γ(d 0 , 0) = 0.
Moreover
∂ 2 Γ
∂d 2 (d 0 , 0) = b N θ(θ − 2)d θ−2 0 f (0) < 0 ,
∂ 2 Γ
∂d∂ζ i
(d 0 , 0) = b N θd θ−1 0 ∂f
∂ζ i
(0) = 0,
∂ 2 Γ
∂ζ i ∂ζ j (d 0 , 0) = b N d θ 0 ∂ 2 f
∂ζ i ∂ζ j (0). (2.15)
Finally (2.12) follows by (2.15) and (2.14).
Proof of Theorem 0.2. It follows from Proposition 2.6 and Proposition 2.5, taking into account Proposition 1.6 and Theorem 2.4.
3 Appendix A
In this appendix we prove Proposition 1.5 and Proposition 1.6. Let K δ,ξ λ be defined in (1.9) and
let Π λ δ,ξ : D λ −→ K δ,ξ λ denote the orthogonal projection.
Lemma 3.1 It holds
Π λ δ,ξ (u)
λ ≤ kuk λ ∀ u ∈ D λ . Let L λ δ,ξ : K δ,ξ λ −→ K δ,ξ λ be defined by
L λ δ,ξ (φ) = Π λ δ,ξ n
φ − i ∗ λ h
(2 ∗ − 1)k(ξ 0 )W δ,ξ 2
∗−2 φ io
.
Lemma 3.2 For any Z ⊂⊂ R N \ {0} there exist C > 0, λ 0 > 0 and δ 0 > 0 such that for any λ ∈ (0, λ 0 ), δ ∈ (0, δ 0 ) and ξ ∈ Z there holds
L λ δ,ξ (φ)
λ ≥ Ckφk ∀ φ ∈ K δ,ξ λ .
Proof. We argue by contradiction. Assume there exist sequences λ n → 0, δ n → 0, ξ n → ξ 6= 0, φ n , ψ n ∈ K δ λ
nn
,ξ
nsuch that
kφ n k λ
n= 1, ψ n = L λ δ
nn
,ξ
n(φ n ), kψ n k λ
n→ 0.
More precisely φ n and ψ n solve φ n − i ∗ λ
n
(2 ∗ − 1)k(ξ 0 )W δ 2
∗−2
n
,ξ
nφ n = ψ n + w n , (3.1) where w n = P N
i=0 c i n ψ δ i
n
,ξ
nfor some real numbers c i n . First of all we claim that
w n −→ 0 in D 1,2 (R N ). (3.2)
If we multiply (3.1) by w n we get for large n kw n k 2 λ
n
= −(2 ∗ − 1) Z
R
NU δ 2
∗−2
n
,ξ
nφ n w n ≤ (2 ∗ − 1)
Z
R
NU δ 2
n∗,ξ
n2/N
kφ n k L
2∗(R
N) kw n k L
2∗(R
N)
≤ const
S
1 − λ n
C N
−1
kw n k λ
nand hence kw n k λ
n≤ const . Therefore for large n
kw n k 2 = Z
R
N|∇w n | 2 =
N
X
i=0
c i n 2 kψ i δ
n,ξ
nk 2 ≤ const .
Multiplying (3.1) by ψ δ j
n,ξ
n, in view of (1.8) we get
w n , ψ δ j
n
,ξ
nλ
n= −(2 ∗ − 1) Z
R
NU δ 2
∗−2
n
,ξ
nφ n ψ δ j
n
,ξ
n= −λ n
Z
R
Nφ n ψ δ j
n
,ξ
n|x| 2 = O(λ n )kψ δ j
n
,ξ
nk. (3.3)
On the other hand
w n , ψ j δ
n
,ξ
nλ
n=
N
X
i=0
c i n
ψ i δ
n,ξ
n, ψ j δ
n
,ξ
nλ
n=
N
X
i=0
c i n
"
Z
R
N∇ψ i δ
n
,ξ
n· ∇ψ j δ
n
,ξ
n− λ n Z
R
Nψ δ i
n
,ξ
nψ j δ
n
,ξ
n|x| 2
#
= kψ j δ
n
,ξ
nk 2 c j n − λ n
N
X
i=0
c i n Z
R
Nψ i δ
n
,ξ
nψ δ j
n
,ξ
n|x| 2 and hence from Hardy’s inequality (1.1)
kψ j δ
n
,ξ
nk 2 c j n + C N −1 λ n kψ j δ
n
,ξ
nk
N
X
i=0
|c i n |kψ δ i
n
,ξ
nk
≥ w n , ψ j δ
n
,ξ
nλ
n≥ kψ j δ
n
,ξ
nk 2 c j n − C N −1 λ n kψ δ j
n
,ξ
nk
N
X
i=0
|c i n |kψ i δ
n,ξ
nk. (3.4)
From (3.3) and (3.4) we deduce that
|c j n |kψ δ j
n
,ξ
nk ≤ cλ n 1 +
N
X
i=0
|c i n |kψ i δ
n,ξ
nk
! ,
which implies that c j n kψ j δ
n
,ξ
nk → 0 for any j. Therefore (3.2) follows. Now let φ ˜ n (y) := δ
N −2
n
2φ n (δ n y + ξ n ), k ˜ φ n k = kφ n k ∈
1, 1 − λ n /C N
−1/2
(3.5) because of (1.3), and
ψ ˜ n (y) := δ
N −2
n
2ψ n (δ n y + ξ n ), k ˜ ψ n k = kψ n k → 0, (3.6)
˜
w n (y) := δ
N −2
n
2w n (δ n y + ξ n ), k ˜ w n k = kw n k → 0. (3.7) Up to a subsequence, we can assume that ˜ φ n → φ weakly in D 1,2 (R N ). Given ψ ∈ C c ∞ (R N ), let v n (x) = ψ x−ξ δ
nn
. Note that
kv n k = δ
N −2
n
2kψk.
Testing (3.1) with v n , we find Z
R
N∇ ˜ φ n ∇ψ − λ n δ 1−
N
n
2Z
R
Nφ n v n
|x| 2 − (2 ∗ − 1) Z
R
NU 1,0 2
∗−2 φ ˜ n ψ
= Z
R
N∇( ˜ ψ n + ˜ w n ) · ∇ψ − λ n δ 1−
N
n
2Z
R
N(ψ n + w n )v n
|x| 2 . (3.8)
From Hardy inequality we have that
δ 1−
N
n
2Z
R
N|φ n v n |
|x| 2 ≤ δ 1−
N
n
2Z
R
N|φ n | 2
|x| 2
12Z
R
N|v n | 2
|x| 2
12≤ C N −1
1 − λ n
C N
−
12kψk ≤ const (3.9)
and
δ 1−
N
n
2Z
R
N|ψ n + w n | |v n |
|x| 2 ≤ C N −1 kψ n + w n kkψk −→
n→+∞ 0. (3.10)
In view of (3.9) and (3.10), passing to the limit in (3.8) we get that φ is a (weak) solution to
−∆φ = (2 ∗ − 1)U 1,0 2
∗−2 φ in R N . (3.11) Moreover, since φ n ∈ K δ λ
nn
,ξ
n, it is easy to see that φ, ∂U δ,ξ
∂δ
(δ,ξ)=(1,0)
!
= φ, ∂U δ,ξ
∂ξ i
(δ,ξ)=(1,0)
!
= 0, i = 1, . . . , N. (3.12)
By (3.11), (3.12), and Remark 1.4 we deduce that φ = 0. On the other hand, if we multiply (3.1) by φ n we get
1 =kφ n k 2 λ
n
= (2 ∗ − 1) Z
R
NU δ 2
∗−2
n
,ξ
nφ 2 n + (ψ n + w n , φ n ) λ
n= (2 ∗ − 1) Z
R
NU 1,0 2
∗−2 φ ˜ 2 n + o(1) hence by letting n → ∞ we find 1 = (2 ∗ − 1) R
R
NU 1,0 2
∗−2 φ 2 and a contradiction arises.
Proof of Proposition 1.5. First of all we point out that φ solves equation (1.10) if and only if φ is a fixed point of the operator T δ,ξ λ : K δ,ξ λ −→ K δ,ξ λ defined by
T δ,ξ λ (φ) =
L λ δ,ξ −1
◦ Π λ δ,ξ ◦ i ∗ λ
k(x) f (W δ,ξ + φ) − f (W δ,ξ ) − f 0 (W δ,ξ ) φ + [k(x) − k(ξ 0 )] f 0 (W δ,ξ ) φ
+
L λ δ,ξ −1
◦ Π λ δ,ξ
{i ∗ λ [k(x)f (W δ,ξ )] − W δ,ξ } . (3.13) By Lemma 1.3, Lemma 3.1 and Lemma 3.2 we deduce that for some positive constant ¯ c and for any φ, φ 1 , φ 2 ∈ D λ it holds
kT δ,ξ λ (φ)k λ ≤ ¯ c kk(x) f (W δ,ξ + φ) − f (W δ,ξ ) − f 0 (W δ,ξ ) φ k
L
N +22N+ ¯ c k [k(x) − k(ξ 0 )] f 0 (W δ,ξ ) φk
L
2N
N +2
+ ¯ cki ∗ λ [k(x)f (W δ,ξ )] − W δ,ξ k λ (3.14) and
kT δ,ξ λ (φ 1 − φ 2 )k λ ≤ ¯ c kk(x) f (W δ,ξ + φ 1 ) − f (W δ,ξ + φ 2 ) − f 0 (W δ,ξ ) (φ 1 − φ 2 ) k
L
2N N +2
+ ¯ c k [k(x) − k(ξ 0 )] f 0 (W δ,ξ ) (φ 1 − φ 2 )k
L
2N
N +2
. (3.15)
Let us note that the following inequalities
(a + b) + s−1
−a s−1 − (s − 1)a s−2 b ≤
C(a s−3 |b| 2 + |b| s−1 ) if s ≥ 3 C |b| s−1 if 2 < s < 3,
(3.16) s
and
(a + b) + s−1
− a s−1
≤ C|b| s−1 if 1 < s ≤ 2 (3.17) s
where C = C(s) > 0, hold for any b ∈ R, a ∈ R + . From (3.16) s with s = 2 ∗ it follows that
f (W δ,ξ + φ) −f (W δ,ξ ) − f 0 (W δ,ξ ) φ ≤
C
W
6−N N −2
δ,ξ |φ| 2 + |φ|
N +2N −2if N ≤ 6 C |φ|
N +2N −2if N > 6.
(3.18)
From (3.18) and from the inequality
(α + β) γ ≤ 2 γ−1 (α γ + β γ ) for any α, β ≥ 0, γ ≥ 1, we find
f (W δ,ξ + φ) −f (W δ,ξ ) − f 0 (W δ,ξ ) φ
2N N +2
≤
C(W
2(6−N )N (N −2)(N +2)
δ,ξ |φ|
N +24N+ |φ|
N −22N) if N ≤ 6 C |φ|
N −22Nif N > 6.
(3.19)
From (3.19) and H¨ older’s inequality we have
k(x)f W δ,ξ + φ − f W δ,ξ − f 0 W δ,ξ φ
L
2N N +2
≤
C
kφk
4N N +2
λ + kφk
2N N −2
λ
N +22Nif N ≤ 6 Ckφk
N +2 N −2
λ if N > 6.
(3.20)
Moreover by using H¨ older’s inequality and Lemma 5.7 we get for some constant c ∗ > 0 k [k(x) − k(ξ 0 )] f 0 (W δ,ξ ) φk
L
2N
N +2
≤ c ∗ kφk λ |ξ − ξ 0 | θ + δ θ . (3.21) From (3.20), (3.21), and (5.18) we deduce that there exists a positive constant ˜ c such that for any φ ∈ ¯ B λ δ,ξ (0, ρ) := {v ∈ K δ,ξ λ : kvk λ ≤ ρ}, ρ ≤ 1,
kT δ,ξ λ (φ)k λ ≤ ˜ c
ρ min{2,
N +2N −2} + ρ |ξ − ξ 0 | θ + δ θ + |ξ − ξ 0 | θ + δ θ + h(λ, δ)
, (3.22) where
h(λ, δ) =
λδ 2 if N ≥ 5,
λδ 2 log δ if N = 4,
λδ if N = 3.
Since R
R
N(W δ,ξ +z) + 2
∗−2
−W δ,ξ 2
∗−2
N/2 goes to 0 as z → 0 in D λ by the Dominated Convergence Theorem, we can choose K, δ 0 , and λ 0 such that for any λ ∈ (0, λ 0 ), δ ∈ (0, δ 0 ), and ξ ∈ Z
1
4˜ c − |ξ − ξ 0 | θ − δ θ > 0 and
|ξ − ξ 0 | θ + δ θ + h(λ, δ) ≤ min n
1 4˜ c
1
4˜ c − |ξ − ξ 0 | θ − δ θ max { 1,
N −24} , 4c 1
∗¯ c
o sup
kzk
λ≤3ρ
λδ,ξR
R
N(W δ,ξ + z) + 2
∗−2
− W δ,ξ 2
∗−2
N/2 2/N
< [S(1−λ/C 4¯ c(2
∗−1)kkk
N)]
L∞1/2(3.23)
where
¯
ρ λ δ,ξ = 4˜ c |ξ − ξ 0 | θ + δ θ + h(λ, δ). (3.24) From (3.22) and (3.23) we find
kT δ,ξ λ (φ)k λ ≤ ¯ ρ λ δ,ξ for any φ ∈ ¯ B δ,ξ λ (0, ¯ ρ λ δ,ξ ).
In particular T δ,ξ λ maps B δ,ξ λ (0, ¯ ρ λ δ,ξ ) into itself; in order to prove that it is a contraction there we have also to estimate T δ,ξ λ (φ 1 − φ 2 ), i.e. the right hand side in (3.15). By H¨ older inequality and (3.23) we have
kk(x)f (W δ,ξ + φ 1 ) − f (W δ,ξ + φ 2 ) − f 0 (W δ,ξ ) (φ 1 − φ 2 )k
L
N +22N≤ kkk L
∞Z 1 0
d
dt f (W δ,ξ + φ 2 + t(φ 1 − φ 2 )
− f 0 (W δ,ξ )(φ 1 − φ 2 ) L
2N N +2
≤ kkk L
∞Z 1 0
f 0 (W δ,ξ + φ 2 + t(φ 1 − φ 2 )) − f 0 (W δ,ξ )(φ 1 − φ 2 ) L
N +22N≤ kkk L
∞kφ 1 − φ 2 k L
2∗sup
kzk
λ≤3 ¯ ρ
λδ,ξkf 0 (W δ,ξ + z) − f 0 (W δ,ξ )k L
N/2≤ 1
4¯ c kφ 1 − φ 2 k λ . (3.25) Moreover from (3.21) and (3.23) we have
k [k(x) − k(ξ 0 )] f 0 (W δ,ξ ) (φ 1 − φ 2 )k
L
2N
N +2
≤ c ∗ kφ 1 − φ 2 k λ |ξ − ξ 0 | θ + δ θ ≤ 1
4¯ c kφ 1 − φ 2 k λ . (3.26) From (3.15), (3.25), and (3.26) we obtain
kT δ,ξ λ (φ 1 − φ 2 )k λ ≤ 1
2 kφ 1 − φ 2 k λ for any φ 1 , φ 2 ∈ B λ δ,ξ (0, ¯ ρ λ δ,ξ )
namely T δ,ξ λ is a contraction in B δ,ξ λ (0, ¯ ρ λ δ,ξ ) and hence T δ,ξ λ has a unique fixed point φ λ δ,ξ in B δ,ξ λ (0, ¯ ρ λ δ,ξ ). In particular, in view of (3.24), φ λ δ,ξ satisfies estimate (1.11).
Lemma 3.3 There exist Z ⊂⊂ R N such that ξ 0 ∈ Z, λ 0 > 0 and δ 0 > 0 such that for any
λ ∈ (0, λ 0 ), the function (δ, ξ) 7→ φ λ δ,ξ given by Proposition 1.5 is of class C 1 on (0, δ 0 ) × Z.
Proof. From (1.10) it follows that for any v ∈ K δ,ξ λ
W δ,ξ + φ λ δ,ξ − i ∗ λ (k(x)f (W δ,ξ + φ λ δ,ξ )), v) λ = 0 i.e.
(W δ,ξ + φ λ δ,ξ , v) λ − Z
R
Nk(x)f (W δ,ξ + φ λ δ,ξ )v dx = 0, which can be written in term of the functional J λ as
(J λ 0 (W δ,ξ + φ λ δ,ξ ), v) λ = 0 for any v ∈ K δ,ξ λ
where J λ 0 (u) denotes the Fr´ echet derivative of J λ at u (identified with an element of D λ through the canonical identification of the Hilbert space D λ with its dual). Hence φ λ δ,ξ satisfies
Π λ δ,ξ J λ 0 (W δ,ξ + φ λ δ,ξ ) = 0. (3.27) For λ fixed consider the map
R + × R N × D λ −→ D λ
(δ, ξ, φ) 7−→ Π λ δ,ξ J λ 0 (W δ,ξ + φ). (3.28) Let us remark that J λ ∈ C 2 (D λ , R) and the map (δ, ξ) 7→ W δ,ξ which parametrizes the manifold of the solutions to the limit problem is of class C 2 . Morover the projection map Π λ δ,ξ can be written in the form
Π λ δ,ξ (u) = u −
N
X
j=0
u, Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
λ
Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ where
Υ 0 λ,δ,ξ = ψ δ,ξ 0 kψ δ,ξ 0 k λ , Υ j λ,δ,ξ = ψ δ,ξ j
kψ δ,ξ j k λ −
j−1
X
i=0
ψ δ,ξ j
kψ j δ,ξ k λ , Υ i λ,δ,ξ kΥ i λ,δ,ξ k λ
λ
Υ i λ,δ,ξ
kΥ i λ,δ,ξ k λ , j = 1, . . . , N. (3.29) Note that Υ j λ,δ,ξ , j = 0, 1, . . . , N generate the linear space span{ψ δ,ξ j : j = 0, 1, . . . , N }, solve equation (1.8), and satisfy
Υ j λ,δ,ξ , Υ i λ,δ,ξ
λ = 0, i 6= j. It is easy to verify that (δ, ξ) 7→ Υ j λ,δ,ξ is a C 1 −map, and consequently the projection map Π λ δ,ξ is of class C 1 . Hence we deduce that the map defined in (3.28) is of class C 1 , as it is shown in the diagram below
(δ, ξ, φ) C
2
7−→ W δ,ξ 7−→ W C
∞δ,ξ + φ C
1
7−→ J λ 0 (W δ,ξ + φ) C
1
7−→ Π λ δ,ξ J λ 0 (W δ,ξ + φ).
From (3.27), the regularity of the map defined in (3.28), and the Implicit Function Theorem the
statement follows.
Lemma 3.4 J λ 00 (W δ,ξ + φ λ δ,ξ )u = L λ δ,ξ (Π λ δ,ξ u) + kuk λ o(1) as δ, λ, |ξ − ξ 0 | → 0.
Proof. We have that for any v ∈ D λ
(J λ 00 (W δ,ξ + φ λ δ,ξ )u, v) λ − (L λ δ,ξ (Π λ δ,ξ u), v) λ = (J λ 00 (W δ,ξ + φ λ δ,ξ )Π λ δ,ξ u, Π λ δ,ξ v) λ
− (L λ δ,ξ (Π λ δ,ξ u), Π λ δ,ξ v) λ + (J λ 00 (W δ,ξ + φ λ δ,ξ )u, v − Π λ δ,ξ v) λ
+ (J λ 00 (W δ,ξ + φ λ δ,ξ )(u − Π λ δ,ξ u, Π λ δ,ξ v) λ . (3.30) For simplicity of notation, let us set f δ,ξ j := Υ
j λ,δ,ξ
kΥ
jλ,δ,ξk
λ. Note that by H¨ older inequality (J λ 00 (W δ,ξ + φ λ δ,ξ )u, v − Π λ δ,ξ v) λ
=
N
X
j=0
v, f δ,ξ j
λ
u, f δ,ξ j
λ − (2 ∗ − 1) Z
R
Nk(x)(W δ,ξ + φ λ δ,ξ ) 2
∗−2 f δ,ξ j u
=
N
X
j=0
v, f δ,ξ j
λ
− (2 ∗ − 1) Z
R
Nk(x)[(W δ,ξ + φ λ δ,ξ ) 2
∗−2 − W δ,ξ 2
∗−2 ]f δ,ξ j u
− (2 ∗ − 1) Z
R
N[k(x) − k(ξ 0 )]W δ,ξ 2
∗−2 f δ,ξ j u − λ Z
R
Nf δ,ξ j u
|x| 2
≤ constkuk λ kvk λ λ + k(k(x) − k(ξ 0 ))W δ,ξ 2
∗−2 k L
N/2+ k(W δ,ξ + φ λ δ,ξ ) 2
∗−2 − W δ,ξ 2
∗−2 k L
N/2.
Hence from (3.16) s and (3.17) s with s = 2 ∗ − 1 and Lemma 5.7, we get (J λ 00 (W δ,ξ + φ λ δ,ξ )u,v − Π λ δ,ξ v) λ
≤ constkuk λ kvk λ λ + |ξ − ξ 0 | ϑ + δ ϑ + kφ λ δ,ξ k λ + kφ λ δ,ξ k 4/(N −2) λ . (3.31) Arguing in a similar way we can show that
(J λ 00 (W δ,ξ + φ λ δ,ξ )(u − Π λ δ,ξ u), Π λ δ,ξ v) λ
≤ constkuk λ kvk λ λ + |ξ − ξ 0 | ϑ + δ ϑ + kφ λ δ,ξ k λ + kφ λ δ,ξ k 4/(N −2) λ
(3.32) and
(J λ 00 (W δ,ξ + φ λ δ,ξ )Π λ δ,ξ u,Π λ δ,ξ v) λ − (L λ δ,ξ (Π λ δ,ξ u), Π λ δ,ξ v) λ
= −(2 ∗ − 1)
Z
R
Nk(x)[(W δ,ξ + φ λ δ,ξ ) 2
∗−2 − W δ,ξ 2
∗−2 ]Π λ δ,ξ uΠ λ δ,ξ v +
Z
R
N(k(x) − k(ξ 0 ))W δ,ξ 2
∗−2 Π λ δ,ξ uΠ λ δ,ξ v
≤ constkuk λ kvk λ |ξ − ξ 0 | ϑ + δ ϑ + kφ λ δ,ξ k λ + kφ λ δ,ξ k 4/(N −2) λ . (3.33)
The lemma follows from (3.30), (3.31), (3.32), and (3.33).
Lemma 3.5 There holds
∂φ λ δ,ξ
∂δ λ
=
δ −1 O λδ + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k min{1,4/(N −2)}
λ
if N ≥ 5, δ −1 O λδp| log δ| + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k λ
if N = 4, δ −1 O λδ 1/2 + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k λ
if N = 3,
(3.34)
and for N ≥ 3 and i = 1, . . . , N
∂φ λ δ,ξ
∂ξ i
λ
= δ −1 O
λδ + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k min{1,4/(N −2)}
λ
. (3.35)
Proof. Let us first note that from standard calculations, it is easy to verify that for some positiver constants C 0 (N, k(ξ 0 )) and C 1 (N, k(ξ 0 ))
kψ δ,ξ 0 k = C 0 (N, k(ξ 0 ))δ −1 , kψ δ,ξ j k = C 1 (N, k(ξ 0 ))δ −1 , j = 1, . . . , N,
∂ψ 0
∂δ ,
∂ψ 0
∂ξ i ,
∂ψ j
∂δ ,
∂ψ j
∂ξ i
= O(δ −2 ). (3.36)
From (3.27) it follows that there exist N + 1 real valued C 1 -function α j (λ, δ, ξ) such that J λ 0 (W δ,ξ + φ λ δ,ξ ) =
N
X
j=0
α j (λ, δ, ξ) Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ , where Υ j λ,δ,ξ are defined in (3.29). Since
α j (λ, δ, ξ) +
N
X
i=0 i6=j
α i (λ, δ, ξ) Υ i λ,δ,ξ
kΥ i λ,δ,ξ k λ , Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
!
λ
=
J λ 0 (W δ,ξ + φ λ δ,ξ ), Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
λ
= − λ
kΥ j λ,δ,ξ k λ Z
R
NW δ,ξ Υ j λ,δ,ξ
|x| 2 dx
− Z
R
Nk(x)((W δ,ξ + φ λ δ,ξ ) 2
∗−1 − W δ,ξ 2
∗−1 ) Υ j λ,δ,ξ
kΥ j λ,δ,ξ k λ dx − Z
R
N(k(x) − k(ξ 0 ))W δ,ξ 2
∗−1 Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ dx, using (3.36), Lemma 5.2, Lemma 5.5, (3.29), and (3.20) we deduce that for any j = 0, 1, . . . , N
|α j (λ, δ, ξ)| =
O λδ 2 + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k λ
if N ≥ 5, O λδ 2 | log δ| + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k λ
if N = 4, O λδ + |ξ − ξ 0 | θ + δ θ + kφ λ δ,ξ k λ
if N = 3.
(3.37)
The function (λ, δ, ξ) 7→ (φ λ δ,ξ , α 0 (λ, δ, ξ), . . . , α N (λ, δ, ξ)) is implicitly given by H(δ, ξ, φ, α, λ) = 0 where
H : (0, δ 0 ) × R N × D λ × R N +1 × (0, λ 0 ) → D λ × R N +1 (δ, ξ, φ, α, λ) 7→
J λ 0 (W δ,ξ + φ) −
N
X
j=0
α j Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ ,
φ, Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
j=0,...,N λ
.
Since
∂H 1
∂(φ, α) (φ,α)
u β
= J λ 00 (W δ,ξ + φ)u −
N
X
j=0
β j
Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
∂H 2
∂(φ, α) (φ,α)
u β
=
u, Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
j=0,...,N λ
,
in view of Lemma 3.4 and (3.31) we find for any (u, β) ∈ D λ × R N +1
∂H
∂(φ, α)
(φ
λδ,ξ,α(λ,δ,ξ))
u β
2
=
J λ 00 (W δ,ξ + φ λ δ,ξ )u −
N
X
j=0
β j
Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
2
λ
+
N
X
j=0
u, Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
2 λ
=
J λ 00 (W δ,ξ + φ λ δ,ξ )u
2 λ +
N
X
j=0
β j 2 + ku − Π λ δ,ξ uk 2 − 2
N
X
j=0
β j
J λ 00 (W δ,ξ + φ λ δ,ξ )u, Υ j λ,δ,ξ kΥ j λ,δ,ξ k λ
λ
≥ (kL λ δ,ξ (Π λ δ,ξ u)k − kuk λ o(1)) 2 + |β| 2 + ku − Π λ δ,ξ uk 2 − 2
N
X
j=0
|β j |kuk λ o(1).
Hence from Lemma 3.2 we obtain for λ, δ, |ξ − ξ 0 | small
∂H
∂(φ, α)
(φ
λδ,ξ,α(λ,δ,ξ))
u β
2
≥ const
u β
2
.
Therefore there exist δ 0 > 0, λ 0 > 0, a small neighbourhood Z of ξ 0 , and a positive constant C ∗ , such that for any δ ∈ (0, δ 0 ), λ ∈ (0, λ 0 ), ξ ∈ Z,
∂H
∂(φ, α)
−1
(φ
λδ,ξ,α(λ,δ,ξ))
≤ C ∗ .
From the Implicit Function Theorem, we have that
∂ δ φ λ δ,ξ
∂ δ α(λ, δ, ξ)
= − ∂H
∂(φ, α)
−1
(φ
λδ,ξ,α(λ,δ,ξ))
× ∂H
∂δ (φ
λδ,ξ