Scattering for the 3D Gross-Pitaevskii equation
Zihua Guo Monash University
(joint with Zaher Hani, Kenji Nakanishi)
Jan 25, 2019
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Gross-Pitaevskii (GP) equation
Consider the Gross-Pitaevskii (GP) equation
i ψt+ ∆ψ = (|ψ|2− 1)ψ, ψ : R1+3→ C (1) with the boundary condition
|x|→∞lim ψ = 1. (2)
Formally, let φ = e−itψ, then φ solves
i φt+ ∆φ = |φ|2φ. (3)
The non-vanishing boundary condition: very different. Physical contexts: Bose-Einstein condensates, superfluids and nonlinear optics, or in the hydrodynamic interpretation of NLS
Gross-Pitaevskii (GP) equation
Consider the Gross-Pitaevskii (GP) equation
i ψt+ ∆ψ = (|ψ|2− 1)ψ, ψ : R1+3→ C (1) with the boundary condition
|x|→∞lim ψ = 1. (2)
Formally, let φ = e−itψ, then φ solves
i φt+ ∆φ = |φ|2φ. (3)
The non-vanishing boundary condition: very different. Physical contexts: Bose-Einstein condensates, superfluids and nonlinear optics, or in the hydrodynamic interpretation of NLS
Gross-Pitaevskii (GP) equation
Consider the Gross-Pitaevskii (GP) equation
i ψt+ ∆ψ = (|ψ|2− 1)ψ, ψ : R1+3→ C (1) with the boundary condition
|x|→∞lim ψ = 1. (2)
Formally, let φ = e−itψ, then φ solves
i φt+ ∆φ = |φ|2φ. (3)
The non-vanishing boundary condition: very different.
Physical contexts: Bose-Einstein condensates, superfluids and nonlinear optics, or in the hydrodynamic interpretation of NLS
Energy conservation
Let u = ψ − 1. Then u satisfies zero boundardy condition and
i ∂tu + ∆u − 2 Re u = u2+ 2|u|2+ |u|2u (4) which is equivalent to (writing u = u1+ iu2)
˙
u1 = −∆u2+ 2(u1+ |u|2/2)u2,
− ˙u2 = (2 − ∆)u1+ 3u12+ u22+ |u|2u1. (5)
Conservation of the energy: E (u) :=
Z
R3
|∇u|2+ (|u|2+ 2 Re u)2
2 dx
= Z
R3
|∇ψ|2+(|ψ|2− 1)2
2 dx = E (u0).
(6)
Note. We do not have conservation of kuk22, but ku(t)k2 ≤ CeCt by Gronwall inequality.
Energy conservation
Let u = ψ − 1. Then u satisfies zero boundardy condition and
i ∂tu + ∆u − 2 Re u = u2+ 2|u|2+ |u|2u (4) which is equivalent to (writing u = u1+ iu2)
˙
u1 = −∆u2+ 2(u1+ |u|2/2)u2,
− ˙u2 = (2 − ∆)u1+ 3u12+ u22+ |u|2u1. (5) Conservation of the energy:
E (u) :=
Z
R3
|∇u|2+ (|u|2 + 2 Re u)2
2 dx
= Z
R3
|∇ψ|2+(|ψ|2− 1)2
2 dx = E (u0).
(6)
Note. We do not have conservation of kuk22, but ku(t)k2 ≤ CeCt by Gronwall inequality.
Well-posedness
In Zhidkov space Xk = {u ∈ L∞ : ∂α ∈ L2, 1 ≤ |α| ≤ k}
d = 1: Zhidkov 1987;
d = 2, 3: Gallo 2004 In H1, GWP
d = 2, 3: B´ethuel and Saut 1999 In energy space, GWP
d = 1, 2, 3 and small data for d = 4: G´erard;
d = 4: Killip-Oh-Pocovnicu-Visan Remark. No scattering in the above works.
GWP in energy space
Energy space
E := {f ∈ ˙H1(R3) : 2 Re f + |f |2 ∈ L2(R3)} (7) with the distance dE(f , g ) defined by
dE(f , g )2 = k∇(f − g )k2L2 +1 2
|f |2+ 2 Re f − |g |2− 2 Re g
2 L2. Note that (E, dE) is a complete metric space.
Theorem (G´ erard, 2006)
Unconditional GWP of (4) in the energy space E. Remark. GWP in H1 ⊂ E, B´ethuel-Saut, 1999.
method: LWP by Strichartz analysis + energy conservation
GWP in energy space
Energy space
E := {f ∈ ˙H1(R3) : 2 Re f + |f |2 ∈ L2(R3)} (7) with the distance dE(f , g ) defined by
dE(f , g )2 = k∇(f − g )k2L2 +1 2
|f |2+ 2 Re f − |g |2− 2 Re g
2 L2. Note that (E, dE) is a complete metric space.
Theorem (G´ erard, 2006)
Unconditional GWP of (4) in the energy space E.
Remark. GWP in H1 ⊂ E, B´ethuel-Saut, 1999.
method: LWP by Strichartz analysis + energy conservation
Traveling wave solutions
There are a family of the solutions ψ(t, x ) = vc(x − ct) with finite energy for 0 < |c| <√
2.
Moreover, B´ethuel-Gravejat-Saut 2009 proved that in 3D E∗ := inf{E (ψ − 1)|1 6= ψ(t, x ) = v (x − ct) solves (1) for some c} > 0, conjecturedthat E∗ is the threshold for the global dispersive solutions.
Remark.
1. In 2D, there is no lower bound.
2. In the radial case, are there special non-scattering solutions in energy space? Not clear.
Traveling wave solutions
There are a family of the solutions ψ(t, x ) = vc(x − ct) with finite energy for 0 < |c| <√
2.
Moreover, B´ethuel-Gravejat-Saut 2009 proved that in 3D E∗ := inf{E (ψ − 1)|1 6= ψ(t, x ) = v (x − ct) solves (1) for some c} > 0, conjecturedthat E∗ is the threshold for the global dispersive solutions.
Remark.
1. In 2D, there is no lower bound.
2. In the radial case, are there special non-scattering solutions in energy space? Not clear.
Scattering
In 3D, scattering was proved by Gustafson-Nakanishi-Tsai (2006,2007,2009) for small data in weighted Sobolev space.
Question: What about the energy space?
Theorem (G.-Hani-Nakanishi, 2017)
For 3D GP equation, scattering holds for small radial data in E. Remark. Radial symmetry can be replaced by additional one order angular regularity.
Remaining questions: How about non-radial case? What about large data? Is smallness not needed?
Let’s state our theorem more precisely.
Scattering
In 3D, scattering was proved by Gustafson-Nakanishi-Tsai (2006,2007,2009) for small data in weighted Sobolev space.
Question: What about the energy space?
Theorem (G.-Hani-Nakanishi, 2017)
For 3D GP equation, scattering holds for small radial data in E.
Remark. Radial symmetry can be replaced by additional one order angular regularity.
Remaining questions: How about non-radial case? What about large data? Is smallness not needed?
Let’s state our theorem more precisely.
By the diagonalising transform
u = u1+ iu2 −→ v = v1 + iv2 := u1+ iUu2, (8) with
U :=p−∆(2 − ∆)−1, we can rewrite the GP equation for v :
i ∂tv − Hv = U(3u12+ u22+ |u|2u1) + i (2u1u2+ |u|2u2), (9) where (one can write u1 = v1, u2 = U−1v2)
H :=p−∆(2 − ∆). (10)
Note. H has symbol |ξ|p2 + |ξ|2. It behaves as Schr¨odinger equation for high frequency and as wave equation for low frequency.
Theorem
There exists δ > 0 such that for any u0 ∈ E, radial, with E (u0) ≤ δ, there exists a unique global solution u ∈ C (R : E) to (4). Moreover, there exists φ± ∈ H1 such that
t→±∞lim
u1+ (2 − ∆)−1u22 + iUu2− e−itHφ±
H1 = 0. (11) Remark.
We have the decay for quadratic terms of u1 (not true for u2):
t→±∞lim k(2 − ∆)−1u12kH1 = 0.
We can transfer the asymptotic behaviour in the energy space E.
Indeed, we can prove
t→±∞lim dE(u, T−1(e−itHφ±)) = 0, where T (u) = T (u1+ iu2) = u1+ (2 − ∆)−1u22+ iUu2.
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Difficulty 1: singularity at zero frequency
At low frequency U−1 ≈ |∇|−1. Idea. Nonlinear transform
This was treated by Gustafson-Nakanishi-Tsai using some nonlinear transform
z = z1 + iz2 = u1+u12+ u22
2 − ∆ + iUu2 (12)
Under the transform (12)
izt− Hz = − 2iU(u12) − 4h∇i−2∇ · (u1∇u2)
+ [−iU(|u|2u1) + U2(|u|2u2)]. (13) Note. Quadratic terms have no zero frequency singularity.
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Consider the 3D Schr¨odinger equation iut+ ∆u = |u|pu
For scattering, with small data in Hs, Strichartz analysis requires p ≥ 43, can not work for u2.
Scattering results for p < 43: 0 < p ≤ 23
Any nontrivial solution u, with φ ∈ S, does not scatter to linear solution in L2. (Glassey 1974, Strauss 1974 )
2
3 < p < 43
For any φ ∈ H1 with kx φk2 < ∞, the solution u scatters to linear solution in L2. (Tsutsumi-Yajima, 1984)
Nakanishi-Ozawa 2002, Masaki 2015 p = 23
Modified scatterings occur. Ozawa, Hayashi, Naumkin, etc
Consider the 3D Schr¨odinger equation iut+ ∆u = |u|pu
For scattering, with small data in Hs, Strichartz analysis requires p ≥ 43, can not work for u2.
Scattering results for p < 43: 0 < p ≤ 23
Any nontrivial solution u, with φ ∈ S, does not scatter to linear solution in L2. (Glassey 1974, Strauss 1974 )
2
3 < p < 43
For any φ ∈ H1 with kx φk2 < ∞, the solution u scatters to linear solution in L2. (Tsutsumi-Yajima, 1984)
Nakanishi-Ozawa 2002, Masaki 2015 p = 23
Modified scatterings occur. Ozawa, Hayashi, Naumkin, etc
Note. For quadratic terms 2/3 < 1 < 4/3
Weighted Hs is usually needed for 3D quadratic terms.
Our ideas. Replace weighted Hs by additional angular regularity. In particular, in the radial case, we can handle Hs.
Use earlier ideas in the works on 3D Zakharov system (G.-Nakanishi 2012, G.-Lee-Nakanishi-Wang 2015)
To illustrate our ideas, we consider the classical 3D quadratic Klein-Gordon equation
utt − ∆u + u = u2 (14)
To study the asymptotic problems, there are two well-known methods
Klainerman’s vector field method (1985)
Existence of enough vector fields + energy estimates Shatah’s normal form method (1985)
Non-resonant structure + normal form transform (transfer the quadratic term to a cubic or higher order term).
Key non-resonance: hξi + hηi − hξ + ηi&hηi1
Our approach: generalized Strichartz estimates+(partial) normal form
To illustrate our ideas, we consider the classical 3D quadratic Klein-Gordon equation
utt − ∆u + u = u2 (14)
To study the asymptotic problems, there are two well-known methods Klainerman’s vector field method (1985)
Existence of enough vector fields + energy estimates
Shatah’s normal form method (1985)
Non-resonant structure + normal form transform (transfer the quadratic term to a cubic or higher order term).
Key non-resonance: hξi + hηi − hξ + ηi&hηi1
Our approach: generalized Strichartz estimates+(partial) normal form
To illustrate our ideas, we consider the classical 3D quadratic Klein-Gordon equation
utt − ∆u + u = u2 (14)
To study the asymptotic problems, there are two well-known methods Klainerman’s vector field method (1985)
Existence of enough vector fields + energy estimates Shatah’s normal form method (1985)
Non-resonant structure + normal form transform (transfer the quadratic term to a cubic or higher order term).
Key non-resonance: hξi + hηi − hξ + ηi&hηi1
Our approach: generalized Strichartz estimates+(partial) normal form
To illustrate our ideas, we consider the classical 3D quadratic Klein-Gordon equation
utt − ∆u + u = u2 (14)
To study the asymptotic problems, there are two well-known methods Klainerman’s vector field method (1985)
Existence of enough vector fields + energy estimates Shatah’s normal form method (1985)
Non-resonant structure + normal form transform (transfer the quadratic term to a cubic or higher order term).
Key non-resonance: hξi + hηi − hξ + ηi&hηi1
Our approach: generalized Strichartz estimates+(partial) normal form
(i ∂t+ hDi)u = hDi−1u2 Let K (t) = eithDi. Then
Lemma 1 (radial case) (G.-Nakanishi-Wang 2013).
Let r > 10/3, for φ radial
keithDiPkφkL2
tLrx.2kβkkφkL2
x, where Pk ≈ F−11|ξ|∼2kF and
βk =
1
2 − 3r, k < 0;
1 −3r, k ≥ 0;103 < r < 4;
1
4 + , k ≥ 0; r = 4;
1
r, k ≥ 0; r > 4.
Remark. The classical best non-radial estimates keithDiPkφkL2
tL6x.25k/6kφkL2
x. Wave: Klainerman-Machedon, Sogge, Sterbenz 2005 Schr¨odinger: Sogge, Shao, G.-Wang 2010
Lemma 2 (non-radial case) (G.-Hani-Nakanishi 2016).
Let r > 10/3,
keithDiPkφkL2
tLrxL2σ.Bk(2, r )kφkL2
x, where Pk ≈ F−11|ξ|∼2kF and
Bk(2, r ) =
2k(12−3r), k < 0;
2k(1−3r), k ≥ 0;103 < r < 4;
hki2k4, k ≥ 0; r = 4;
2kr, k ≥ 0; r > 4.
Remark.
Wave: Sterbenz 2005
Schr¨odinger: G.-Lee-Nakanishi-Wang 2014, G. 2016
Theorem. Scattering for 3D quadratic KG with small data in H1× L2 and with 1-order angular regularity.
Generalized Strichartz for GP
Lemma 3 (G.-Hani-Nakanishi 2016). For GP, we have: for r > 10/3,
ke−itHPkφkL2
tLrxL2σ.Ck(2, r )kφkL2
x(R3), (15)
where
Ck(2, r ) =
2k(12−3r), k ≥ 0;
2k(2−7r), k < 0,103 < r < 4;
2k(1−3r), k < 0, r > 4;
hki2k4, k < 0, r = 4.
(16)
Note. H := p−∆(2 − ∆).
Intermediate Theorem
With this estimate and the nonlinear transform by (12), we get
Theorem
Scattering holds for small radial data u(0) ∈ H1.
Question: What about energy space E? For small data, we can think E = {u : Re u ∈ H1, Im u ∈ ˙H1∩ L4}
Difficulty. Im u 6∈ L2.
Outline
1 Introduction and results
2 Proof of the theorem
Difficulty 1: singularity at zero frequency Difficulty 2: 3D quadratic term
Difficulty 3: weak low-frequency component of u2
Difficulty. For u(0) ∈ E, we can only have ∇u2(0) ∈ L2. Recall under the nonlinear transform
z = z1 + iz2 = u1+u12+ u22
2 − ∆ + iUu2 (17)
izt− Hz = − 2iU(u12) − 4h∇i−2∇ · (u1∇u2)
+ [−iU(|u|2u1) + U2(|u|2u2)]. (18) Problematic terms: u22∆u2 or u22u1 when u2 has very low frequency.
Key new ingredients: “Null-structure” achieved by new nonlinear transform. Let
m = m1+ im2 = u1+ 2u12+ u22
2 − ∆ + iUu2 (19)
Then
i ∂tm − Hm = N2(m) + N3(m, u) + N4(m, u) + N5(m, u)
where
N2(m, u) = U(m12) + 2i
2 − ∆[−3m1∆u2− 2∇m1· ∇u2], N3(m, u) = U(2m1R) + iN31(u) + 2i
2 − ∆[4u1m1u2− m12u2], N4(m, u) = U(R2− |u|4/4) + 2i
2 − ∆[4u1Ru2− 2u2m1R], N5(m, u) = 2i
2 − ∆[−u2R2+ u2|u|4/4], with
R = −∆u22
2(2 − ∆) − (2 + ∆)u12 2(2 − ∆), N31(u) =(2 − ∆)−1
−2u2|∇u2|2+ 3∆u22
2 − ∆∆u2+ 2∇∆u22 2 − ∆ ∇u2
+ 4
2 − ∆
2u12 2 − ∆∆u2
− 2 + ∆ 2 − ∆u12
u2.
Remark
Our nonlinear transform can achieve cancellation and is actually natural.
The GP equation (4) for u = u1 + iu2 can be rewritten as follows
˙
u1 = −∆u2+ 2(u1+ |u|2/2)u2,
− ˙u2 = (2 − ∆)u1+ 3u12+ u22+ |u|2u1
= (2 − ∆)(u1) + 2u21+ u22+ (2u1+ |u|2)2/4 − |u|4/4.
Note that 2u1+ |u|2 ∈ L2 is bounded by the conserved energy.
In view of the equation of u2, we first make the following change of variables
z1 := u1+2u12+ u22
2 − ∆ , z2 = u2.
Thank you very much!