Orbital stability for NLS

Jordan Bell
April 3, 2014

Let n=3, and take p<43. Some of the material we will present for general n when it doesn’t simplify our work to use n=3.

The (defocusing) nonlinear Schrödinger equation is

i⁢ϕt+Δ⁢ϕ+|ϕ|p-1⁢ϕ=0.

ϕ⁢(x,0)=ϕ0∈H1.

For a function ψ on ℝn, the orbit of the function under the symmetries of NLS is

𝒢ψ={ψ(⋅+x0)ei⁢γ:(x0,γ)∈ℝn×𝕋}.

We say that ψ is orbitally stable if initial data being near it implies that the solution of NLS is near it always.

We define

ρ(ϕ(t),𝒢ψ)=inf(x0,γ)∈ℝn×𝕋∥ϕ(⋅+x0,t)ei⁢γ-ψ∥H1.

The ground state equation is

Δ⁢u-u+|u|p-1⁢u=0.

The ground state equation comes from the solution ϕ⁢(x,t)=ei⁢t⁢u⁢(x) of NLS. It is a fact that there is a positive bounded solution R of the ground state equation, which we call a ground state.

Theorem 1.

The ground state R is orbitally stable: for any ϵ>0 there is a δ⁢(ϵ)>0 such that if

ρ⁢(ϕ0,𝒢R)<δ⁢(ϵ)

then for all t>0

ρ⁢(ϕ⁢(t),𝒢R)<ϵ.

We define the energy functional ℰ by

ℰ⁢[ϕ]=∫|∇⁡ϕ|2+|ϕ|2-2p+1⁢|ϕ|p+1⁢d⁢x,

so ℰ⁢[ϕ] is a function of time but not of space.

It is a fact that for each t there are x0=x0⁢(t) and γ=γ⁢(t) such that

∥ϕ(⋅+x0,t)ei⁢γ-R∥H1=ρ(ϕ(t),𝒢R).

Let w=ϕ(⋅+x0,t)ei⁢γ-R; so ∥w⁢(t)∥H1=ρ⁢(ϕ⁢(t),𝒢R).

Let Δ⁢ℰ=ℰ⁢[ϕ0]-ℰ⁢[R]. We have

Δ⁢ℰ = ℰ[ϕ(⋅,t]-ℰ[R]
= ℰ[ϕ(⋅+x0,t)ei⁢γ]-ℰ[R]
= ℰ⁢[R+w]-ℰ⁢[R].

We shall express ℰ⁢[R+w] as a Taylor expansion about R. We compute the first variation as follows:

d⁢ℰ⁢[R]⁢w = ∫∇⁡w⁢∇⁡R¯+∇⁡R⁢∇⁡w¯+w⁢R¯+R⁢w¯-|R|p-1⁢(w⁢R¯+R⁢w¯)
= 2⁢ℜ⁢∫∇⁡w⁢∇⁡R+w⁢R-w⁢|R|p-1⁢R
= 2⁢ℜ⁢∫w⁢(-Δ⁢R+R-|R|p-1⁢R)
= 0,

where we used the fact that R is real valued, integration by parts, and the fact that R is a solution of the ground state equation. So the first variation of ℰ at R is 0.

We now compute the second variation of ℰ.

d2⁢ℰ⁢[R]⁢[w] = 2⁢ℜ⁢∫-w⁢Δ⁢w¯+|w|2-p-12⁢Rp-1⁢w2-p-12⁢Rp-1⁢|w|2
-Rp-1⁢|w|2
= 2⁢ℜ⁢∫-w⁢Δ⁢w¯+|w|2-p-12⁢Rp-1⁢w2-p+12⁢Rp-1⁢|w|2

Write w=u+i⁢v. Then we have

d2⁢ℰ⁢[R]⁢[w] = 2⁢∫-u⁢Δ⁢u-v⁢Δ⁢v+u2+v2-Rp-1⁢(p⁢u2+v2)

Define

L+=-Δ+1-p⁢Rp-1  L-=-Δ+1-Rp-1,

which gives

(L+⁢u,u)L2=∫-u⁢Δ⁢u+u2-p⁢u2⁢Rp-1

and

(L-⁢v,v)L2=∫-v⁢Δ⁢v+v2-v2⁢Rp-1.

Thus

d2⁢ℰ⁢[R]⁢[w]=2⁢(L+⁢u,u)L2+2⁢(L-⁢v,v)L2.

And we assert that the remainder term of the Taylor series is O⁢(∫|w|3), because R is bounded. Therefore

Δ⁢ℰ=(L+⁢u,u)L2+(L-⁢v,v)L2+O⁢(∫|w|3).

We can bound ∫|w|3 using the Gagliardo-Nirenberg inequality, which gives us (for n=3)

∥w∥L33≤C0⁢∥∇⁡w∥L23/2⁢∥w∥L23/2≤C0⁢∥w∥H13,

for some C0 that doesn’t depend on w. Therefore

Δ⁢ℰ=(L+⁢u,u)L2+(L-⁢v,v)L2+O⁢(∥w∥H13),

so there is some C such that

Δ⁢ℰ≥(L+⁢u,u)L2+(L-⁢v,v)L2-C⁢∥w∥H13.