Exponential convergence to the equilibrium for thestochastic Gross-Pitaevskii equation
R. Fukuizumi 1
A. de Bouard 2 and A. Debussche 3
1RCPAM, GSIS, Tohoku University, Japan2CMAP, Ecole Polytechnique, France
3IRMAR, ENS Cachan Bretagne, France
Mathematical and Physical models of Nonlinear Optics @ IMA
R. Fukuizumi (Tohoku Univ.) October 31, 2016 1 / 29
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 2 / 29
Introduction Physical Motivation
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 3 / 29
Introduction Physical Motivation
Bose-Einstein Condensation (BEC)
A state of matter of a dilute gas of bosons cooled to temperatures veryclose to absolute zero; a large amount of bosons occupy the lowestquantum state, where macroscopic quantum phenomena can be observed.Predicted in 1924-25, Realization in 1995.
87Rb, 23Na, 7Li, 1H, 85Rb, 41K, 4He, 133Cs, 174Yb,
52Cr, 40Ca, 84Sr : Example of the dilute gases of bosons
R. Fukuizumi (Tohoku Univ.) October 31, 2016 4 / 29
Introduction Physical Motivation
Dynamics of BEC at nite temperature
At the zero temperature T = 0, all the atoms are well-presented by asingle condensate function and the coherent evolution of the wavefunction is described by the standard GP equation
At higher temperature, all spontaneous and incoherent process (for ex.interaction with thermal cloud) may not be neglected
The eect of such incoherent elements are implemented by adding adisspation and a noise to the GP equation, called 'Stochastic GPequation,' which is well suited for vortex formulation in BEC (Weileret al. Nature (2008))
dψ = P− i
~LGPψdt +
G (x)
kBT(µ− LGP)ψdt + dWG (x , t)
where
LGP = − ~2
2m∇2 + V (x) + g |ψ|2, 〈dW ∗G (s, y), dWG (t, x)〉 = 2G (x)δt−sδx−y
R. Fukuizumi (Tohoku Univ.) October 31, 2016 5 / 29
Introduction Physical Motivation
P.B. Blakie et al., Advanced in Physics (2008); introduce moredetailed energy cut-o
The equation evolves the system to the grand-canonical equilibriumdistribution.
M.Kobayashi-L.Cugliandolo (arXiv:1606.03262); interested in a quenchdynamics in BEC
Phase transition leads to the formation of topological defects (i.e.vortices)Investigation of the thermal equilibrium (Gibbs equilibrium) gives theclassication (Universality class) of the type of phase transitionsStudy of the equilibrium properties around the phase transition pointsusing, as a model, Stochastic GP equation with V (x) = 0, G (x) = 1under a periodic boundary condition
Remark that essentially, nite dimension models are analyzed in Physics.Our aim: Study these models from a mathematical point of view, (i.e., ininnite dimension). In particular, interested in the Gibbs equilibrium (thecase of a space-time white noise).
R. Fukuizumi (Tohoku Univ.) October 31, 2016 6 / 29
Introduction Physical Motivation
We consider mathematically (for the moment in 1d):
(Ω,F ,P) : propability space endowed with ltration (Ft)t≥0The equation :
dX = (i + γ)(∂2xX − V (x)X − λ|X |2X )dt +√γdW ,
X (0) = X0, t > 0, x ∈ R
where γ > 0. Assume V (x) = x2 and λ = 1 (defocusing).
(∂2x − x2)hk = −λ2khk with λk =√2k + 1, k ∈ N. The eigenfunctions
hk(x) are known as the Hermite functions.
W (t) : cylindrical Wiener pocess on L2(R,C), i.e.
W (t, x) =∑k∈N
βk(t)hk(x), t > 0, x ∈ R
where βk(t)k∈N : a sequence of C-valued independent Brownianmotions.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 7 / 29
Gibbs equilibrium Gibbs measure
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 8 / 29
Gibbs equilibrium Gibbs measure
Hamiltonian for the case of γ = 0:
H(u) =1
2
∫R|(−∂2x + x2)1/2u|2dx +
1
4
∫R|u|4dx
The Gibbs measure:
ρ(du) = Γe−H(u)du
= Γe−14
∫R |u|
4dxe−12((−∂2x+x2)u,u)L2du = Γe−
14
∫R |u|
4dxµ(du)
where Γ is the normalizing constant, may be justied in Lp(R), p > 2(see later) taking the limit N →∞ of the Gaussian measure onR2(N+1) dened by
dµN :=N∏k=0
λ2k2π
e−λ2k2(a2
k+b2
k)dakdbk
if we write u =∑
k(ak + ibk)hk with (ak , bk) ∈ R2.
µ := limN→∞ µN can be interpreted as the law of random variable∑k∈N
√2
λkgk(ω)hk(x) with L(gk) = NC(0, 1)
R. Fukuizumi (Tohoku Univ.) October 31, 2016 9 / 29
Gibbs equilibrium Gibbs measure
Known results
Burq, Tzvetkov and Thomann, (γ = 0, λ = ±1, V (x) = x2, 1d)Ann. Inst. Fourier (Grenoble) (2013)
Construction of Gibbs measure, Global Cauchy theory in the negativeSobolev space.
Barton-Smith, (γ 6= 0, λ = ±1, V (x) = 0, bounded domain D, any d)Nonlinear dier. equ. appl.(2004)
Existence and Uniqueness of invariant measure on Lp(D) (p ∈ [2,∞))for a not too small γ 6= 0
E.A. Carlen, J. Fröhlich and J. Lebowitz, (regular noise, λ = −1,V (x) = 0, periodic boundary condi., 1d) (arXiv:1409.2327)
Construction of the grand-canonical Gibbs measure (i.e. with amodied Hamiltonian) and exponential convergence by Dirichlet formapproach under some assumptions on the noise
Our result is new on the points/no restriction on γ, the exp-convergencewith a space-time white noise
R. Fukuizumi (Tohoku Univ.) October 31, 2016 10 / 29
Gibbs equilibrium Main results
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 11 / 29
Gibbs equilibrium Main results
Recall that our equation isdX = (i + γ)(∂2xX − x2X − |X |2X )dt +
√γdW ,
X (0) = X0, t > 0, x ∈ R.
Let p ≥ 3, X0 ∈ Lp(R), and γ > 0.
Theorem
There exists a set O ⊂ Lp(R) such that ρ(O) = 1, and such that for
X0 ∈ O there exists a unique solution X (·) ∈ C ([0,∞), Lp(R)) a.s.
Let Pt be the transition semigroup for the equation, i.e.Ptφ(y) = E(φ(X (t, y))), y ∈ O, t ≥ 0.
Theorem
Let φ ∈ L2((Lp, dρ),R), and φ =∫Lpφ(y)dρ(y). Then Ptφ(·) converges
exponentially to φ in L2((Lp, dρ),R), as t →∞ ; more precisely,∫Lp|Ptφ(y)− φ|2dρ(y) ≤ e−γt
∫Lp|φ(y)− φ|2dρ(y).
R. Fukuizumi (Tohoku Univ.) October 31, 2016 12 / 29
Gibbs equilibrium Gibbs measure and the stationary solution
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 13 / 29
Gibbs equilibrium Gibbs measure and the stationary solution
Gibbs measure and the stationary solution
Let us denote by
Z∞(t) =√γ
∫ t
−∞e−(t−s)(i+γ)(−∂
2x+x
2)dW (s),
the solution of
dZ = (i + γ)(∂2x − x2)Zdt +√γdW
which is stationary, i.e. such that L(Z∞(t)) = L(Z∞(0)).
Write Z∞(t) using the basis hkk ,
Z∞(t) =√γ∑k∈N
(∫ t
−∞e−(t−s)(i+γ)λ
2kdβk(s)
)hk
The law of Z∞(t) equals to the Gaussian measure µ, since
L(√
γ
∫ 0
−∞es(i+γ)λ
2kdβk(s)
)= NC
(0,
2
λ2k
)R. Fukuizumi (Tohoku Univ.) October 31, 2016 14 / 29
Gibbs equilibrium Gibbs measure and the stationary solution
Support of the measure
Let p > 2. Note that by Minkowski inequality, for q ≥ p, one has
|Z∞(t)|Lqω(Lpx ) ≤ |Z∞(t)|Lpx (Lqω).
Since Z∞ is Gaussian, to compute |Z∞|Lqω when q = 2m, it suces tocompute the L2ω norm:
E(|Z∞(t, x)|2) ≤ γ∑k∈N
E∣∣∣ ∫ t
−∞e−λ
2k(t−s)(i+γ)dβk(s)
∣∣∣2|hk(x)|2
≤ 2γ∑k∈N
∫ t
−∞e−2λ
2kγ(t−s)ds |hk(x)|2 ≤
∑k∈N
|hk(x)|2
λ2k.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 15 / 29
Gibbs equilibrium Gibbs measure and the stationary solution
For q = 2m with 2m ≥ p, we have
E(|Z∞(t, x)|2m) ≤ CmE(|Z∞(t, x)|2)m ≤ Cm
(∑k
|hk(x)|2
λ2k
)mand
|Z∞(t)|L2mω (Lpx )≤ Cm
∣∣∣∑k∈N
|hk(x)|2
λ2k
∣∣∣1/2Lp/2x
≤ Cm
(∑k
|hk(x)|2Lpx
λ2k
)1/2It is known the decay of hk in Lp (Yajima-Zhang, CMP 2001): for
p ≥ 4, |hk |Lp(R) ≤ Cpλ−1/6k , and by interpolation,
if 2 ≤ p ≤ 4, |hk |Lp(R) ≤ Cpλ− 1
3(1− 2
p)
k .
Remind that λ2k = 2k + 1 and the series converges for p > 2, i.e.,Z∞ ∈ L2m(Ω; Lp) for any m ≥ p/2 > 1;i.e. Z∞ ∈ Lp a.s. i.e. ρ(Lp) = 1 for p > 2.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 16 / 29
Gibbs equilibrium Invariance of the Gibbs measure
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 17 / 29
Gibbs equilibrium Invariance of the Gibbs measure
Formal proof for the invariance of the Gibbs measure
We write the equation under the form
dX = −J∇xH(X )dt − γ∇xH(X )dt +√γdW , X (0) = y ∈ Lpx
where
J =
(0 −11 0
): R2 → R2.
Let Pt be the transition semigroup, i.e,Ptφ(y) = E(φ(X (t, y))) = 〈φ, µt,x〉 for φ ∈ Cb(Lp) withµt,x = L(X (t, y)).
〈Ptφ, ρ〉 = Ptφ(y) provided L(y) = ρ. Thus P∗t ρ = µt,x .i.e. ρ is invariant i P∗t ρ = ρ.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 18 / 29
Gibbs equilibrium Invariance of the Gibbs measure
Let L be the generater for the semigroup Pt :
Lφ(y) = limt→0
Ptφ(y)− φ(y)
t.
For the invariance of ρ it is enough to show∫Lp
(Lφ)(y)ρ(dy) = 0
On the other hand, the generator L is given by (cf. Kolmogorovequation)
(Lφ)(y) = γ∆yφ(y)−γ(∇yφ(y),∇yH(y))L2x − (∇yφ(y), J∇yH(y))L2x
R. Fukuizumi (Tohoku Univ.) October 31, 2016 19 / 29
Gibbs equilibrium Invariance of the Gibbs measure
∫Lp
(Lφ)(y)ρ(dy) =
∫Lp
[γ∆yφ(y)− γ(∇yφ(y),∇yH(y))L2
−(∇yφ(y), J∇yH(y))L2]e−H(y)dy
= γ
∫Lp
∆yφ(y)e−H(y)dy + γ
∫Lp
(∇yφ(y),∇y (e−H(y))
)L2dy
−∫Lp
(∇yφ(y), J∇y (e−H(y))
)L2dy = 0.
The last term is zero, and the second term is the opposite sign of therst term by I/P.
This calculation can be justied using a nite dimensionalapproximation.
This invariance holds for both purely dissipative case and mixed case.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 20 / 29
Local existence
Local existence
Write X = v + Z∞ with v satisfying the deterministic PDE:
∂tv = (i + γ)(∂2xv − x2v − |v + Z∞|2(v + Z∞)), t > 0, x ∈ R
Solve locally, then obtain a sol. X in C ([0,T ∗), Lp), p ≥ 3(Ginibre-Velo, CMP. 1997) using the estimates on the linear semigroupobtained by Mehler's formula:
|et(i+γ2)(∂2x−x2)f |Lr (R) ≤ Ct−12l |f |Ls , t > 0 small,
0 ≤ 1
r=
1
s− 1
l, 1 ≤ s, l ≤ ∞.
Energy methods give a global bound in Lp, but it requires somerestriction on the parameter γ. (not too small γ 6= 0)
R. Fukuizumi (Tohoku Univ.) October 31, 2016 21 / 29
Global existence for ρ a.e. initial data
ρ-a.e. Global existence (formal)
Let T ≥ T ∗, and p ≥ 3. Wish to show that there exists CT such that∫Lp
E(
supt∈[0,T∗)
|X (t,X0)|Lp)ρ(dX0) ≤ CT .
Mild form for X :
X (t) = et(i+γ)(∂2x−x2)X0 − et(i+γ)(∂
2x−x2)Z∞(0)
+
∫ t
0
e(t−s)(i+γ)(∂2x−x2)(|X |2X )(s)ds + Z∞(t).
By the regularity of Z∞: supt |Z∞(t)|Lp ≤ MT , we have∫Lp
E(
supt∈[0,T∗)
|X (t,X0)|Lp)dρ(X0)
≤ C
∫Lp
E|X0|Lpdρ(X0) + CE∫Lp
∫ T
0
|X (s)|3L3pdsdρ(X0) + MT .
R. Fukuizumi (Tohoku Univ.) October 31, 2016 22 / 29
Global existence for ρ a.e. initial data
Here the invariance of ρ implies∫LpPtφ(X0)dρ =
∫Lpφ(X0)dρ
with φ(X0) = |X0|3L3p . Therefore∫ T
0
∫Lp
E|X (s,X0)|3L3pdρ(X0)ds =
∫ T
0
∫LP|X0|3L3pdρ(X0)ds
≤∫ T
0
∫LP|X0|3L3pdµ(X0)ds
which is equivalent to∫ T0
E(|Z∞(s)|3L3p
)ds since L(Z∞(s)) = µ, andis bounded by CT .
Thusρ(X0 ∈ Lp; sup
t∈[0,T∗)|X (t,X0)|Lp < +∞) = 1.
This formal argument can be justied by an nite dimensionalapproximation.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 23 / 29
Convergence to the Gibbs equilibrium
Convergence to the equilibrium (formal)
The proof of the convergence is based on a Poincaré inequality: Let γ > 0.For any φ ∈ C 1
b (Lp), the following inequality is satised.∫Lp|∇yφ(y)|2L2ydρ(y) ≥ √γ
∫Lp|φ(y)− φ|2dρ(y), (1)
where φ =∫Lpφ(y)dρ(y).
Assume (1). Let u(t, y) = Ptφ(y) and φ = 0. Then, if L is thegenerator of Pt , then
dudt
= Lu.
L satises (again by the Kolmogorov equation)
Lu2(y) = 2√γ|∇yu|2L2 + 2uLu(y).
R. Fukuizumi (Tohoku Univ.) October 31, 2016 24 / 29
Convergence to the Gibbs equilibrium
Invariance of the measure ρ implies
0 =
∫LpLu2(y)dρ(y)
= 2√γ
∫Lp|∇yu|2L2dρ(y) + 2
∫Lpu(Lu)(y)dρ(y)
= 2√γ
∫Lp|∇yu|2L2dρ(y) +
d
dt
∫Lp|u(t)|2dρ(y).
hence by (1)d
dt|u(t)|2L2(Lp ,dρ) ≤ −2γ|u|
2L2(Lp ,dρ),
from which it follows
|u(t)|L2(Lp ,dρ) ≤ e−γt |φ|L2(Lp ,dρ).
R. Fukuizumi (Tohoku Univ.) October 31, 2016 25 / 29
Convergence to the Gibbs equilibrium Poincaré inequality
Outline
1 IntroductionPhysical Motivation
2 Gibbs equilibriumGibbs measureMain resultsGibbs measure and the stationary solutionInvariance of the Gibbs measure
3 Local existence
4 Global existence for ρ a.e. initial data
5 Convergence to the Gibbs equilibriumPoincaré inequality
6 Final remarks
R. Fukuizumi (Tohoku Univ.) October 31, 2016 26 / 29
Convergence to the Gibbs equilibrium Poincaré inequality
Proof of Poincaré inequality (formal)
This inequality is a property of the measure only, so that we can proveit by the purely dissipative case, for which ρ is also an invariantmeasure.
dY = γ(∂2xY − x2Y − |Y |2Y )dt +√γdW
This is again a formal argument, we need a nite dimensionalapproximation
Denote ηh := DyY (t, y).h; then ηh satisesdη = γ(∂2xη − x2η − γ|Y |2Y − 2γ(Re(Y η)Y )dt,η(0) = h ∈ L2(R).
Thanks to dissipative property, ηh satises
|ηh(t)|L2 ≤ e−γt |h|L2 . (2)
R. Fukuizumi (Tohoku Univ.) October 31, 2016 27 / 29
Convergence to the Gibbs equilibrium Poincaré inequality
let Rt be the transition semigroup, and let g(t, y) = Rtφ(y) forφ ∈ C 1
b (Lp) and y ∈ Lp. The above exp.decay of ηh and theinvariance of the measure ρ implies∫
Lp|∇yg(y)|2L2dρ(y) ≤ e−2γt
∫Lp
∣∣∣Rt |∇yφ|L2(y)∣∣∣2dρ(y)
≤ e−2γt∫LpRt(|∇yφ|2L2)(y)dρ(y)
= e−2γt∫Lp|∇yφ|2L2(y)dρ(y).
since (∇yg(t, y), h)L2 = E((∇yφ(Y (t, y)), ηh)L2)On the other hand, by the Kolmogorov equation and the invariance ofthe measure (as above),
d
dt|g(t, y)|2L2(Lp ,dρ) = −2√γ|∇yg(y)|2L2(Lp ,dρ)
Note that g(0, y) = φ(y) and that, thanks to dissipative property,|g(t)|L2(Lp ,dρ) → 0 as t →∞. Then integrating in time and tendingt →∞ concludes the Poincaré inequality.
R. Fukuizumi (Tohoku Univ.) October 31, 2016 28 / 29
Final remarks
Final remarks
Our results may be extended to
Stochastic complex Ginzburg-Landau eq. without the harmonicpotential (but with a periodic boundary condi., or on a boundeddomain with Dirichlet condi.)
with the chemical potential
global existence for all initial data
2d or 3d case, but for the renormalized Hamiltonian
with a rotating term; Lz := −i(x∂y − y∂x),
L(u) =1
2
∫R3
|(−H)1/2u|2 + Re
∫R3
(uΩLzu) +1
4
∫R3
|u|4
The focusing case (λ = −1)
R. Fukuizumi (Tohoku Univ.) October 31, 2016 29 / 29