L’explication que nous devons juger satisfaisante est celle qui adh`ere `a son objet : point de vide entre eux, pas d’interstice o`u une autre explication puisse aussi bien se loger ; elle ne convient qu’`a lui, il ne se prˆete qu’`a elle.
La pens´ee et le mouvant, Bergson
In this chapter we present two problems related to the non linear Schr¨odinger equation: (i) the semiclassical limit for the p-eigenvalues of the magnetic Laplacian,
(ii) the dimensional reduction of the time dependent non linear Schr¨odinger equation. 1. Non linear magnetic eigenvalues
1.1. Definition of the non linear eigenvalue. Let Ω be a bounded simply con-nected open set of R2. We introduce the following “nonlinear eigenvalue” (or optimal magnetic Sobolev constant):
(6.1.1) λ(Ω,A, p, h) = inf ψ∈H1 0(Ω),ψ6=0 Qh,A(ψ) R Ω|ψ|pdx 2 p = inf ψ∈H1 0(Ω), kψkLp(Ω)=1 Qh,A(ψ),
where the magnetic quadratic form is defined by
∀ψ ∈H10(Ω), Qh,A(ψ) =
Z
Ω
|(−ih∇+A)ψ|2dx.
Lemma 6.1. The infimum in (6.1.1) is attained.
Proof. Consider a minimizing sequence (ψj) that is normalized in Lp-norm. Then, by a H¨older inequality and using that Ω has bounded measure, (ψj) is bounded in L2. Since A ∈ L∞(Ω), we conclude that (ψj) is bounded in H1
0(Ω). By the Banach-Alaoglu Theorem there exists a subsequence (still denoted by (ψj)) and ψ∞ ∈ H1
0(Ω) such that
ψj * ψ∞ weakly inH1
0(Ω) andψj →ψ∞ in Lq(Ω) for all q ∈[2,+∞). This is enough to
conclude.
Lemma 6.2. The minimizers (which belong to H10(Ω)) of the Lp-normalized version of (6.1.1) satisfy the following equation in the sense of distributions:
In particular (by Sobolev embedding), the minimizers belong to the domain of Lh,A.
1.2. A result by Esteban and Lions. By using the famous concentration-compactness method, Esteban and Lions proved the following proposition in [66].
Proposition 6.3. Let A ∈ L(Rd,Rd) such that B 6= 0 and p ∈ (2,2∗), with 2∗ = d2−d2. We let (6.1.3) S = inf ψ∈Dom(QA),ψ6=0 QA(ψ) kψk2 Lp(Rd) .
Then, the infimum in (6.1.3) is attained.
Note that S >0. Indeed, if (φj)j≥1 is a minimizing sequence, normalized in Lp, such thatQA(φj)→0, we deduce that, by diamagnetism, |φj| →0 inH1(Rd). By the Sobolev embedding H1(Rd) ⊂Lp(Rd), we get that (φj)j≥1 goes to zero in Lp(Rd). We prove this proposition in Chapter 12, Section 1 by using an alternative method to the concentration-compactness principle. Moreover, it is possible to prove that the minimizers of (6.1.3) have an exponential decay.
Proposition 6.4. There exists α >0 such that, for any minimizerψ of (6.1.3), we have eα|x|ψ ∈L2(Rd).
We focus on the two dimensional case. Definition 6.5. For p∈(2,+∞), we define
(6.1.4) λ[0](p) =λ(R2,A[0], p,1) = inf ψ∈Dom(Q A[0]),ψ6=0 QA[0](ψ) kψk2 Lp , where A[0](x, y) = (0,−x). Here QA[0](ψ) = Z R2 |(−i∇+A[0])ψ|2dx, with domain Dom(QA[0]) = n ψ ∈L2(R2) : (−i∇+A[0])ψ ∈L2(R2) o .
Let us now state the main theorem of this section (the proof is given in Chapter 12, Section 2).
Theorem 6.6. Let p≥2. Let us assume that A is smooth on Ω, that B=∇ ×A does
not vanish on Ω and that its minimum b0 is attained in Ω. Then there exist C > 0 and
h0 >0 such that, for all h∈(0, h0),
(1−Ch18)λ[0](p)b 2 p 0h2h−2p ≤λ(Ω,A, p, h)≤(1 +Ch1/2)λ[0](p)b 2 p 0h2h−2p. 2. Non linear dynamics in waveguides
Let us now discuss another non linear problem.
With the same formalism, we will consider the case of unbounded curves and the case of closed curves. Consider a smooth, simple curve Γ in R2 defined by its normal parametrization γ :x1 7→γ(x1). For ε >0 we introduce the map
(6.2.1) Φε:S =M×(−1,1)3(x1, x2)7→γ(x1) +εx2ν(x1) =x,
whereν(x1) denotes the unit normal vector at the pointγ(x1) such that det(γ0(x1), ν(x1)) = 1 and where
M=
(
R for an unbounded curve,
T=R/(2πZ) for a closed curve.
We recall that the curvature at the pointγ(x1), denoted by κ(x1), is defined by
γ00(x1) = κ(x1)ν(x1).
The waveguide is Ωε = Φε(S) and we will work under the following assumption which states that waveguide does not overlap itself and that Φε is a smooth diffeomorphism.
Assumption 6.7. We assume that the function κ is bounded, as well as its derivatives
κ0 andκ00. Moreover, we assume that there existsε0 ∈(0,kκk1
L∞) such that, forε∈(0, ε0), Φε is injective.
We will denote by −∆DirΩε the Dirichlet Laplacian on Ωε. We are interested in the following equation:
(6.2.2) i∂tψε=−∆DirΩεψε+λεα|ψε|2ψε
on Ωε with a Cauchy conditionψε(0;·) = ψε
0 and whereα ≥1 andλ∈Rare parameters. By using the diffeomorphism Φε, we may rewrite (6.2.2) in the space coordinates (x1, x2) given by (6.2.1). For that purpose, let us introduce mε(x1, x2) = 1−εx2κ(x1) and consider the function ψε transported by Φε,
Uεψε(t;x1, x2) = φε(t;x1, x2) =ε1/2mε(x1, x2)1/2ψε(t; Φε(x1, x2)).
Note thatUεis unitary fromL2(Ωε, dx) toL2(S, dx1dx2) and mapsH10(Ωε) (resp. H2(Ωε)) to H10(S) (resp. to H2(S)). Moreover, the operator −∆DirΩε is unitarily equivalent to the self-adjoint operator onL2(S, dx1dx2),
Uε(−∆DirΩε)Uε−1 =Hε+Vε, with Hε=Pε,21+Pε,22,
where
Pε,1 =m−ε1/2Dx1m−ε1/2, Pε,2 =ε−1Dx2
and where the effective electricVε potential is defined by
Vε(x1, x2) = − κ(x1)
2
4(1−εx2κ(x1))2 .
Notice that, for all ε < ε0, we havemε≥1−ε0kκkL∞ >0. The problem (6.2.2) becomes (6.2.3) i∂tφε =Hεφε+Vεφε+λεα−1m−ε1|φε|2φε
with Dirichlet boundary conditionsφε(t;x1,±1) = 0 and the Cauchy condition φε(·; 0) =
φε0 =Uεψ0ε. We notice that the domains of Hε and Hε+Vε coincide with H2(S)∩H10(S). In order to study (6.2.8), it is natural to conjugate the equation by the unitary group
eitHε so that the problem (6.2.8) becomes (6.2.4) i∂tϕeε=eitHε(Vε−ε−2µ1)e−itHε
e
ϕε+λWε(t;ϕeε), ϕeε(0;·) = φε0,
where
(6.2.5) Wε(t;ϕ) = eitHεm−ε1|e−itHεϕ|2e−itHεϕ
and where ϕeε =eitHεϕε which satisfies ϕeε(t;x1,±1) = 0.
We will analyze the critical case α = 1 where the nonlinear term is of the same order as the parallel kinetic energy associated to D2x1. It is well-known that (6.2.2) (thus (6.2.3) also) has two conserved quantities: the L2 norm and the nonlinear energy. Let us introduce the first eigenvalue µ1 = π42 of D2x2 on (−1,1) with Dirichlet boundary conditions, associated to the eigenfunction e1(x2) = cos π2x2 and define the energy functional (6.2.6) Eε(φ) = 1 2 Z S |Pε,1φ|2dx1dx2+ 1 2 Z S |Pε,2φ|2dx1dx2+1 2 Z S Vε− µ1 ε2 |φ|2dx1dx2 +λ 4 Z S m−ε1|φ|4dx1dx2.
Notice that we have substracted the conserved quantity µ1
2ε2kφk2
L2 to the usual nonlinear energy, in order to deal with bounded energies. Indeed, we will consider initial conditions with bounded mass and energy, which means more precisely the following assumption. Assumption 6.8. There exists two constants M0 >0 and M1 > 0 such that the initial data φε
0 satisfies, for all ε∈(0, ε0),
kφε0kL2 ≤M0 and Eε(φε0)≤M1.
Let us define the projection Π1one1by letting Π1u=hu, e1iL2((−1,1))e1. A consequence of Assumption 6.8 is that φε0 has a boundedH1 norm and is close to its projection Π1φε0. Indeed, we will prove the following lemma (see Chapter 23, Section 2).
Lemma 6.9. Assume thatφε0 satisfies Assumption 6.8. Then there existsε1(M0)∈(0, ε0)
and a constant C > 0 independent of ε such that, for all ε ∈(0, ε1(M0)),
(6.2.7) kφε0kH1(S)≤C and kφε0−Π1φε0kL2(M,H1(−1,1)) ≤Cε .
It will be convenient to consider the following change of temporal gaugeφε(t;x1, x2) =
e−iµ1ε−2tϕε(t;x1, x2). This leads to the equation
(6.2.8) i∂tϕε =Hεϕε+ (Vε−ε−2µ1)ϕε+λm−ε1|ϕε|2ϕε
with conditions ϕε(t;x1,±1) = 0, ϕε(0;·) = φε
0.
We can now state the main theorem of this section (see Chapter 23, Section 2.2). Theorem 6.10. Assume that φε
0 ∈H2∩H1
0(S) and that there existM0 >0, M2 >0such
that, for all ε ∈(0, ε0),
(6.2.9) kφε0kL2 ≤M0, (Hε−µ1 ε2)φε0 L2 ≤M2. Then φε
0 satisfies Assumption 6.8 and, for all ε ∈ (0, ε1(M0)), (6.2.8) admits a unique
solution ϕε ∈ C(R+;H2∩H1
0(S))∩ C1(R+;L2(S)). Moreover, there exists C >0 such that,
for all ε∈(0, ε1(M0)) and t≥0, we have