Numerical aspects of the nonlinear Schr¨odinger equation

48  Download (0)

Full text

(1)

Numerical aspects of the nonlinear Schr¨ odinger equation

by

Christophe Besse

Laboratoire Paul Painlev´e, Universit´e Lille 1, CNRS, Inria Simpaf Team

Lille

(2)

Outline of the talk

1

Motivation

2

Numerical methods for NLS

3

Absorbing boundary conditions for the Schr¨ odinger equations

4

Some numerical experiments.

(3)

Motivation

Nonlinear Schr¨ odinger Equation (NLS)

iε∂tψ=−ε2

2∆ψ+V(x)ψ+β|ψ|2ψ , (x, t)∈Rdx×[0;T], u(x,0) =ψ0(x), x∈Rdx.

ψ(x, t): complex-valued wave function V(x): real-valued external potential

β: interaction constant (= 0: linear,+1: repulsive interaction,−1: attractive interaction)

ε: scaled Planck constant

Numerical approximations

ε= 1: standard; 0< ε1andβ=±1→semi-classical regimes efficient, convergent and accurate numerical schemes

(4)

Application of NLS

In quantum physics

Interaction between particles with quantum effect

Bose-Einstein condensation (BEC): bosons at low temperature Superfluidity

In plasma physics

wave interaction between electron and ion In semiconductor industry

In quantum chemistry

chemical interaction based on the first principle In nonlinear optics & atom laser

In fluid mechanics (water waves)

NLS is the prototype for many dispersive systems àExamples :

Davey-Stewartson systems

(water waves)

( i∂tu+λ∂2xu+∂y2u = ν|u|2u+u∂xψ, α∂2ψ+∂2ψ = χ∂x(|u|2).

(5)

Properties of NLS

Time reversible

Time transverse invariant (gauge invariant)

V(x)→V(x) +α⇒ψ→ψe−itα/ε⇒ |ψ|unchanged Mass (wave energy) conservation

Nψ(t) :=

Z

Rd

|ψ(x, t)|2dx≡Nψ(0) :=

Z

Rd

|ψ(x,0)|2dx:=

Z

Rd

0(x)|2dx, t≥0

Energy ( or Hamiltonian) conservation

Eψ(t) :=

Z

Rd

ε2

2|∇ψ(x, t)|2+V(x)|ψ(x, t)|2

2|ψ(x, t)|4 dx≡Eψ(0), t≥0 Dispersion relation without external potential

ψ(x, t) =aei(k·x−ωt/ε) (plane wave solution) ⇒ω=ε2

|k|2+β|a|2

(6)

Numerical difficulties

Explicit vs Implicit (or computational cost) Spatial/temporal accuracy

Stability / Convergence

Keep the properties of NLS in the discretized level Time reversible & time transverse invariant Mass & energy conservation

Dispersion conservation

Resolution in the semiclassical regime: 0< ε1 Take BC into account

(7)

Outline of the talk

1

Motivation

2

Numerical methods for NLS

3

Absorbing boundary conditions for the Schr¨ odinger equations

4

Some numerical experiments.

(8)

Numerical methods for NLS

Classical schemes :

Crank-Nicolson type : Delfour-Fortin-Payre, D´uran-Sanz Serna, ...

Runge-Kutta type : Akrivis-Dougalis-Karakashian, ...

etc ...

àHigh order schemes, convergent, preserving of invariants,BUTsemi-implicit : need of a nonlinear step (Newton)

Example : Crank-Nicolson scheme (CNFD)

iεψn+1−ψn

δt =

−ε2

2∆ +β|ψn+1|2+|ψn|2

2 +V(·, tn+1/2)

ψn+1n 2 = 0, ψ0(x) =ψ0(x), x∈Rd.

whereunis the approximate value ofuat timetn=nδt.

Example : D´ uran-Sanz Serna scheme (SSFD)

iεψn+1−ψn

δt = −ε2 2∆ +β

ψn+1n 2

2

+V(·, tn+1/2)

n+1n 2 = 0,

Possible cures : to decouple linear and nonlinear parts of (NLS)

(9)

Numerical methods for NLS

Classical schemes :

Crank-Nicolson type : Delfour-Fortin-Payre, D´uran-Sanz Serna, ...

Runge-Kutta type : Akrivis-Dougalis-Karakashian, ...

etc ...

àHigh order schemes, convergent, preserving of invariants,BUTsemi-implicit : need of a nonlinear step (Newton)

Example : Crank-Nicolson scheme (CNFD)

iεψn+1−ψn

δt =

−ε2

2∆ +β|ψn+1|2+|ψn|2

2 +V(·, tn+1/2)

ψn+1n 2 = 0, ψ0(x) =ψ0(x), x∈Rd.

whereunis the approximate value ofuat timetn=nδt.

Example : D´ uran-Sanz Serna scheme (SSFD)

ψn+1−ψn ε2

ψn+1n2 !

ψn+1n

(10)

Numerical methods for NLS

Relaxation scheme

1 Relaxation scheme: to regard (NLS) as a Schr¨odinger-Poisson system where the Poisson equation for the potentialΥis replaced by the explicit formulaΥ =|ψ|2, that is,

( Υ =|ψ|2, (x, t)∈Rd×R+∗,

iε∂tψ=−ε22∆ψ+V(x)ψ+βψΥ, (x, t)∈Rd×R+∗,

Relaxation scheme (ReFD) CB ’04









Υn+1/2+ Υn−1/2

2 =|ψn|2, x∈Rd, iεψn+1−ψn

δt =

−ε2

2∆ +V(x) +βΥn+1/2

ψn+1n

2 = 0, x∈Rd,

Advantages convergent scheme, efficient, preserves the invariants, easily adaptable Drawbacks Υ6=|u|2 at the discrete level butΥ(x, t) =Rt

0s|u|2(x, s)ds

;make proof of convergence harder, second order, Υn>0∀n?.

(11)

Numerical methods for NLS

Relaxation scheme

1 Relaxation scheme: to regard (NLS) as a Schr¨odinger-Poisson system where the Poisson equation for the potentialΥis replaced by the explicit formulaΥ =|ψ|2, that is,

( Υ =|ψ|2, (x, t)∈Rd×R+∗,

iε∂tψ=−ε22∆ψ+V(x)ψ+βψΥ, (x, t)∈Rd×R+∗,

Relaxation scheme (ReFD) CB ’04









Υn+1/2+ Υn−1/2

2 =|ψn|2, x∈Rd, iεψn+1−ψn

δt =

−ε2

2∆ +V(x) +βΥn+1/2

ψn+1n

2 = 0, x∈Rd,

Advantages convergent scheme, efficient, preserves the invariants, easily adaptable Drawbacks Υ6=|u|2 at the discrete level butΥ(x, t) =Rt

0s|u|2(x, s)ds

;make proof of convergence harder, second order, Υn>0∀n?.

(12)

Numerical methods for NLS

Relaxation scheme

1 Relaxation scheme: to regard (NLS) as a Schr¨odinger-Poisson system where the Poisson equation for the potentialΥis replaced by the explicit formulaΥ =|ψ|2, that is,

( Υ =|ψ|2, (x, t)∈Rd×R+∗,

iε∂tψ=−ε22∆ψ+V(x)ψ+βψΥ, (x, t)∈Rd×R+∗,

Relaxation scheme (ReFD) CB ’04









Υn+1/2+ Υn−1/2

2 =|ψn|2, x∈Rd, iεψn+1−ψn

δt =

−ε2

2∆ +V(x) +βΥn+1/2

ψn+1n

2 = 0, x∈Rd,

Advantages convergent scheme, efficient, preserves the invariants, easily adaptable Drawbacks Υ6=|u|2 at the discrete level butΥ(x, t) =Rt

0s|u|2(x, s)ds

;make proof of convergence harder, second order, Υn>0∀n?.

(13)

Numerical methods for NLS

Splitting scheme

2 Splitting scheme : Splitting methods are based on a decomposition of the flow of (NLS).

We introduceStthe flow of(N LS):ψ(t,·) =Stψ0 and the two flowsXtandYt solutions to

tv=iε2∆v :v(·, y) =Xtv(·,0) iε∂tw=V(x)w+β|w|2w= 0 :w(·, y) =Ytw(·,0)

Solve nonlinear ODE analytically since∂t(|w(x, t)|2) = 0⇒ |w(x, t)|=|w(x,0)|. w(x, t) =e−it[V(x)+β|w(x,0)|2]/εw(x,0).

Idea :to approximate the flowStby combining the two flowsXtandYt Order 1 (Lie) ZLt =XtYtorZLt =YtXt

Order 2 (Strang) ZSt =Xt/2YtXt/2 orZSt =Yt/2XtYt/2 Higher order See Descombes, Thalhammer

(14)

Numerical methods for NLS

Splitting scheme

Convergence :

Theorem (CB, Bid´ egary, Descombes ’02)

∀u0∈H2,T >0,∃Candh0such that∀h∈(0, h0],∀nsuch thatnh≤T

ZLhn

u0−Snhu0

≤Chku0kH2. Moreover, ifu0∈H4, then

ZShn

u0−Snhu0

≤Ch2ku0kH4.

Advantages convergent scheme, efficient, easily adaptable, very well adapted in semi-classical regimes

Drawbacks energy is not preserved, periodic boundary conditions if Fourier for the space variable.

(15)

Numerical methods for NLS

Splitting scheme

Convergence :

Theorem (CB, Bid´ egary, Descombes ’02)

∀u0∈H2,T >0,∃Candh0such that∀h∈(0, h0],∀nsuch thatnh≤T

ZLhn

u0−Snhu0

≤Chku0kH2. Moreover, ifu0∈H4, then

ZShn

u0−Snhu0

≤Ch2ku0kH4.

Advantages convergent scheme, efficient, easily adaptable, very well adapted in semi-classical regimes

Drawbacks energy is not preserved, periodic boundary conditions if Fourier for the space variable.

(16)

Numerical methods for NLS

Method TSSP CNFD SSFD ReFD

Time Resersible 4 4 4 4

Inv. by Gauge Change 4 6 6 6

Mass Conservation 4 4 4 4

Energy Conservation 6 4 4 4

Dispersion conservation 4 6 6 6

UnconditionalL2 Stability 4 4 4 4

Convergence 4 4 4 4

Explicit Scheme 4 6 6 4

Time accuracy 2thor4th 2th 2th 2th

(17)

Outline of the talk

1

Motivation

2

Numerical methods for NLS

3

Absorbing boundary conditions for the Schr¨ odinger equations

4

Some numerical experiments.

(18)

ABCs for Schr¨ odinger equations

NLS in R

d

(S)









i∂tψ+ ∆ψ+V(x, t, ψ)ψ= 0, (x, t)∈Rd×[0;T] lim

|x|→+∞ψ(x, t) = 0, t∈[0;T] ψ(x,0) =ψ0(x), x∈Rd

whereV =V(x) +f(ψ).

Motivation : the numerical resolution of Schr¨odinger equations leads to introduce a finite computational domain Ω.

=⇒introduction of a regular fictive boundaryΓ

Natural questions :

1 which are the satisfactory conditions to set onΓ×[0, T]such that the approximate solution coincides with the restriction toΩof the true solution?

2 If these conditions exist, how do we handle them numerically?

Does the energy of the approximate solution decay inΩ(stability of the numerical

(19)

ABCs for Schr¨ odinger equations

NLS in R

d

(S)









i∂tψ+ ∆ψ+V(x, t, ψ)ψ= 0, (x, t)∈Rd×[0;T] lim

|x|→+∞ψ(x, t) = 0, t∈[0;T] ψ(x,0) =ψ0(x), x∈Rd

whereV =V(x) +f(ψ).

Motivation : the numerical resolution of Schr¨odinger equations leads to introduce a finite computational domain Ω.

=⇒introduction of a regular fictive boundaryΓ

Natural questions :

1 which are the satisfactory conditions to set onΓ×[0, T]such that the approximate solution coincides with the restriction toΩof the true solution?

If these conditions exist, how do we handle them numerically?

(20)

Domain truncation

Problem : Mesh an unbounded domain (here in 1D)Rd×[0;T] t

0 x

ψ

TruncationR×[0;T]−→ΩT:= ]xl, xr[×[0;T]

Introduction of a fictitious boundaryΣT :=Σ×[0;T]withΣ :=∂Ω ={xl, xr} ⇒ BC on the boundaryΣT

The boundary condition onxlmust represent the effect of the potential on]− ∞, xl]. Expression of the boundary condtion with the help of the Dirichlet-to-Neumann map:

nψ+iΛ+ψ= 0, onΣT

(21)

Domain truncation

Problem : Mesh an unbounded domain (here in 1D)Rd×[0;T]

T

T t

xl 0 xr

ΣT

x ψ

TruncationR×[0;T]−→ΩT:= ]xl, xr[×[0;T]

Introduction of a fictitious boundaryΣT:=Σ×[0;T]withΣ :=∂Ω ={xl, xr} ⇒ BC on the boundaryΣT

The boundary condition onxlmust represent the effect of the potential on]− ∞, xl]. Expression of the boundary condtion with the help of the Dirichlet-to-Neumann map:

nψ+iΛ+ψ= 0, onΣT

(22)

Domain truncation

Problem : Mesh an unbounded domain (here in 1D)Rd×[0;T]

T

T t

xl 0 xr

ΣT

x f(ψ, ∂nψ) = 0 ψ

TruncationR×[0;T]−→ΩT:= ]xl, xr[×[0;T]

Introduction of a fictitious boundaryΣT:=Σ×[0;T]withΣ :=∂Ω ={xl, xr} ⇒ BC on the boundaryΣT

The boundary condition onxlmust represent the effect of the potential on]− ∞, xl].

Expression of the boundary condtion with the help of the Dirichlet-to-Neumann map:

nψ+iΛ+ψ= 0, onΣT

(23)

What happen if we do not take BC into account

PotentialV(x) =x Initial datum:

gaussian function ψ0(x) =e−x2+10ix

Domain:Ω = [−5; 15]

−5 0 5 10 15

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x

| u(x,t) |

Evolution of|ψ|w.r.t time

∆t= 0.2

Homogeneous Dirichlet Boundary Conditions:

ψ= 0

⇒Parasistic reflexion

t

Exact Solution

t

0.6 0.8 1 1.2 1.4 1.6 1.8 2

0.3 0.4 0.5 0.6 0.7 0.8 0.9

(24)

What happen if we do not take BC into account

PotentialV(x) =x Initial datum:

gaussian function ψ0(x) =e−x2+10ix

Domain:Ω = [−5; 15]

−5 0 5 10 15

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x

| u(x,t) |

Evolution of|ψ|w.r.t time

∆t= 0.2

Homogeneous Dirichlet Boundary Conditions:

ψ= 0

⇒Parasistic reflexion

t

Approximated numerical solution

t

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

(25)

The 1D Eq. without potential

1 Splitting between interior and exterior problems Interior Problem





(i∂t+∂2x)v= 0, x∈Ω, t >0,

xv=∂xw, x∈Σ, t >0, v(x,0) =ψ0(x), x∈Ω.

Exterior problem









(i∂t+∂x2)w= 0, x∈Ω, t >0, w(x, t) =v(x, t), x=xl,r, t >0,

lim

|x|→+∞w(x, t) = 0, t >0,

w(x,0) =0, x∈Ω.

000 111 000 111

problem interior problem

left exterior problem

right exterior

(x,t)

xL xR

output: input:

(x ,t)L x(x ,t)L

Dirichlet data Neumann data

v

v w

2 Laplace transform w.r.t time (t→τ) onΩr: ODE inx

3 Argument: w∈L2(Ωr)to select the outgoing wave

4 Inverse Laplace transform

5 Exact boundary condition onxr: ∂xv(x, t)|x=xr=−e−iπ/41/2t v(xr, t) where ∂1/2f(x, t) = 1

√ ∂

Ztf(x, s)|x=xr

√ ds fractional derivative op-

(26)

The 1D Eq. without potential

The problem(S)is thus transformed in(Sapp)

The Schr¨ odinger Eq. in Ω

(Sapp)





i∂tψ+ ∆ψ= 0, (x, t)∈ΩT

nψ+e−iπ/4t1/2ψ= 0, onΣT, ψ(x,0) =ψ0(x), x∈ΩT

Remark:

x2+i∂t= (∂n+i√

i∂t)(∂n−i√ i∂t)

(27)

Extension to simple cases (1)

V = 0

Expression of the exact Dirichlet-to-Neumann operator

nψ+e−iπ/4t1/2ψ= 0, onΣT. (ABC0)

V = V

`,r

constant at the exterior of Ω

nψ+e−iπ/4eitV`,rt1/2

e−itV`,rψ

= 0, onΣT.

V = V (t) : Gauge change

We set: v(x, t) =ψ(x, t)e−iV(t) with V(t) = Z t

0

V(s)ds.

Then i∂tψ= (i∂tv−V(t)v)eiV(t), wherevis solution to the equation without potential

(28)

Non constant potentials

IfV =V(x, t), then the Laplace transform can not be used anymore.

One have to introduce a new tool:pseudodifferential calculus.

Pseudodifferential Operators in 1D

A pseudodifferential operatorP(x, t, ∂t)is described by its total symbolp(x, t, τ)in the Fourier space (τis the covariable oft)

P(x, t, ∂t)u(x, t) =Ft−1 p(x, t, τ)ˆu(x, τ)

= Z

R

p(x, t, τ)u(x, τˆ )eitτ

Notations: P=Op(p) , p(x, t, τ) =σ(P(x, t, ∂t))

(29)

Non constant potentials

IfV =V(x, t), then the Laplace transform can not be used anymore.

One have to introduce a new tool:pseudodifferential calculus.

Pseudodifferential Operators in 1D

A pseudodifferential operatorP(x, t, ∂t)is described by its total symbolp(x, t, τ)in the Fourier space (τis the covariable oft)

P(x, t, ∂t)u(x, t) =Ft−1 p(x, t, τ)u(x, τˆ )

= Z

R

p(x, t, τ)u(x, τˆ )eitτ

Notations: P=Op(p) , p(x, t, τ) =σ(P(x, t, ∂t))

(30)

Non constant potentials

Examples

The fractional operators ∂

t1/2

and I

tα/2

1/2t f(t) = 1

√π∂t

Z t 0

f(s)

√t−sds

Itα/2f(t) = 1 Γ(α/2)

Zt 0

(t−s)α/2−1f(s)ds

Nonlocal w.r.t time convolution operator

Operator ∂tt1/2 It1/2 It

↓ ↓ ↓ ↓

Symbol iτ e−iπ/4

−τ eiπ/4

√−τ

1 iτ

(31)

Non constant potentials

CaseV = 0

TBC: ∂nψ+e−iπ/4t1/2ψ= 0, onΣT.

nψ−iOp √

−τ

ψ= 0, onΣT.

Case constantV =V

TBC: ∂nψ+e−iπ/4eitVt1/2 e−itVψ

= 0, onΣT.

nψ−iOp√

−τ +V

(ψ) = 0, onΣT.

Lemma

Ifais a symbol belonging toSmindependent oft, andV =V(x), then Op(a(τ −V(x)))ψ=eitV(x)Op(a(τ))

e−itV(x)ψ

CaseV =V(t): Gauge change Antoine, Besse et Descombes, 2006

nψ−i eiV(t)Op √

−τ e−iV(t)ψ

= 0, onΣT.

(32)

Non constant potentials

CaseV = 0

TBC: ∂nψ+e−iπ/4t1/2ψ= 0, onΣT.

nψ−iOp √

−τ

ψ= 0, onΣT.

Case constantV =V TBC: ∂nψ−ieitVOp √

−τ

e−itVψ

= 0, onΣT.

nψ−iOp√

−τ +V

(ψ) = 0, onΣT.

Lemma

Ifais a symbol belonging toSmindependent oft, andV =V(x), then Op(a(τ −V(x)))ψ=eitV(x)Op(a(τ))

e−itV(x)ψ

CaseV =V(t): Gauge change Antoine, Besse et Descombes, 2006

nψ−i eiV(t)Op √

−τ e−iV(t)ψ

= 0, onΣT.

(33)

Non constant potentials

CaseV = 0

TBC: ∂nψ+e−iπ/4t1/2ψ= 0, onΣT.

nψ−iOp √

−τ

ψ= 0, onΣT.

Case constantV =V TBC: ∂nψ−ieitVOp √

−τ

e−itVψ

= 0, onΣT.

nψ−iOp√

−τ +V

(ψ) = 0, onΣT.

Lemma

Ifais a symbol belonging toSmindependent oft, andV =V(x), then Op(a(τ −V(x)))ψ=eitV(x)Op(a(τ))

e−itV(x)ψ

CaseV =V(t): Gauge change Antoine, Besse et Descombes, 2006

nψ−i eiV(t)Op √

−τ e−iV(t)ψ

= 0, onΣT.

(34)

Non constant potentials

CaseV = 0

TBC: ∂nψ+e−iπ/4t1/2ψ= 0, onΣT.

nψ−iOp √

−τ

ψ= 0, onΣT.

Case constantV =V TBC: ∂nψ−ieitVOp √

−τ

e−itVψ

= 0, onΣT.

nψ−iOp√

−τ +V

(ψ) = 0, onΣT.

Lemma

Ifais a symbol belonging toSmindependent oft, andV =V(x), then Op(a(τ −V(x)))ψ=eitV(x)Op(a(τ))

e−itV(x)ψ

CaseV =V(t): Gauge change Antoine, Besse et Descombes, 2006

(35)

With potential V = V (x, t) + f (|ψ|) : two strategies

Use of pseudodifferential calculus

No More Exact! : Artifical Boundary Condition

1) Gauge change

v(x, t) =e−iV(x,t)ψ(x, t), with V(x, t) = Z t

0 V(x, s, ψ)ds.

Involve operators

eiV(x,t)Op √

−τ

e−iV(x,t)ψ

2) Direct method

No Gauge change Involve operators

Opp

−τ +V(x, t, ψ) (ψ)

Equivalent strategies forV =V(x), non equivalent forV =V(x, t)

(36)

With potential V = V (x, t) + f (|ψ|) : two strategies

Nonlinearities and general repulsive potentialsx∂xV >0forx∈Ω

V = f (x, ψ)

V =f(x, ψ)andV(x, t) =Rt

0f(x, ψ(x, s))ds Absorbing boundary conditions (ABC) forM= 4:

ABC41: ∂nψ+e−iπ/4eiV1/2t

e−iVψ

−i∂nV 4 eiVIt

e−iVψ

= 0 ABC42: ∂nψ−ip

i∂t+Vψ+1

4∂nV(i∂t+V)−1ψ= 0

(37)

Extension to higher dimensions

Take into account the geometry: convex set with general boundary, smooth, with curvatureκ.

n Ω

Generalized coordinates systemof the bound- ary with respect to normal variablerand curvilinear abscissas

∆ =∂r2rr+h−1s h−1s

κr=h−1κ: curvature of a parallel surfaceΣr toΣ h(r, s) = 1 +rκ

⇒ L=∂r2rr+i∂t+h−1s h−1s

+V Schr¨odinger equation with variable coefficients:

pseudo-differential calculus

τ

s n

r M(s)

M(r,s)

(38)

Nonlocality

Nonlocal both in space and time.

Localizing in space: two approaches, valid for both strategies

”Taylor” approach: Taylor expansion of the symbols for|τ| ξ2

−τ−ξ2+V =−τ

1 +ξ2 τ −V

τ

Thereby:

p−τ−ξ2+V ≈√

−τ

1 + ξ2 2τ − V

=√

−τ−ξ2 2

√1

−τ +V 2

√1

−τ

=⇒Localizing in space only

Pad´e approximation approach:

Opp

−τ−ξ2+V

∼p

i∂t+ ∆Σ+V modOP S−1 formal approximation of√

·by Pad´e approximants=⇒Localizing both in spaceAND

(39)

Conclusion

Two possible approaches for each strategy, so 4 families of ABC.

”Taylor” approach ”Pad´e” approach

Gauge change ABC1,TM ABC1,PM

Direct method ABC2,TM ABC2,PM

↓ ↓

t1/2,I1/2t ,It Opp

−τ−ξ2 or v=e−iVu Opp

−τ−ξ2+V

(40)

ABC: Taylor approach

Gauge change

ABC21,Tnψ+e−iπ/4eiVt1/2 e−iVψ

+κ 2ψ

ABC31,T −eiπ/4eiV κ2

8 +∆Σ

2 +i∂sV∂s+1

2(i∂2sV −(∂sV)2)

I1/2t e−iVψ

ABC41,T +ieiV

s(κ∂s)

2 +κ3+∂2sκ

8 +i∂sκ∂sV 2

It

e−iVψ

−isg(∂nV) 4

p|∂nV|eiVIt

p

|∂nV|e−iVψ

= 0

Direct method

ABC22,Tnψ+e−iπ/4t1/2ψ+κ 2ψ

ABC32,T −eiπ/4 κ2

8 +∆Σ

2

It1/2ψ−eiπ/4sg(V) 2

p|V|It1/2p

|V|ψ ABC42,T +i

s(κ∂s)

2 +κ3+∂s2κ 8

Itψ−isg(∂nV) 4

p|∂nV|It

p|∂nV|ψ

= 0

(41)

ABC: Pad´ e approach

Gauge change

ABC11,Pnψ−ieiVp i∂t+ ∆Σ

e−iVψ

ABC21,P + κ

2ψ+∂sVeiVs(i∂t+ ∆Σ)−1/2 e−iVψ

− κ

2eiV(i∂t+ ∆Σ)−1Σ

e−iVψ

= 0

Direct method

ABC12,Pnψ−ip

i∂t+ ∆Σ+V ψ ABC22,P + κ

2ψ−κ

2(i∂t+ ∆Σ+V)−1Σψ= 0

(42)

Taylor approach conditions

Approximations of∂t1/2,It1/2,It by discrete convolutions, linked to the Crank-Nicolson scheme⇒trapezoidal rule[Schmidt - Yevick (97), Antoine - Besse (03)]

t1/2f(tn)≈ r 2

∆t

n

X

k=0

βn−kfk

It1/2f(tn)≈ r∆t

2

n

X

k=0

αn−kfk

Itf(tn)≈∆t 2

n

X

k=0

γn−kfk





0, α1, α2, . . .) =

1,1,1 2,1

2,3 8,3

8, . . .

βk= (−1)kαk,∀k≥0 (γ0, γ1, γ2, . . .) = (1,2,2,2, . . .)

(43)

Pad´ e approach conditions

[Bruneau - Di Menza (95), Szeftel (04)]

Approximation of√

i∂t+ ∆Σ+V

Rational approximation of the square root by Pad´e approximants

√z≈Rm(z) =

m

X

k=0

amk

m

X

k=1

amkdmk z+dmk

In theABC2,PM conditions:

pi∂t+ ∆Σ+V ; Rm(i∂t+ ∆Σ+V) Rm

⇒ p

i∂t+ ∆Σ+V ψ ≈

m

X

k=0

amk

! ψ−

m

X

k=1

amkdmk(i∂t+ ∆Σ+V +dmk)−1ψ

| {z }

ϕk

(44)

Outline of the talk

1

Motivation

2

Numerical methods for NLS

3

Absorbing boundary conditions for the Schr¨ odinger equations

4

Some numerical experiments.

(45)

Numerical application

1d linear caseu(x, t) = s

i

−4t+iexp

−ix2−k0x+k02t

−4t+i

k0= 8, Ω= [−5,5], N= 1024,∆t= 10−3

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

|u|

Evolution of |u|

t=0 t=0.15 t=0.30 t=0.45

(46)

2D linear case Explicit solution (2D)

u(x1, x2, t) = i i−4texp

−ix21+x22+ 5ix1+ 25it i−4t

.

Finite Element Approximation(P1):Ωi=D(0,10),3278triangles,∆t= 10−2.

(47)
(48)

2D linear and nonlinear case with potentials

Linear, Circular domain,

V(x, y) = 5(x2+y2) Linear, Mediator shaped domain, V(x, y) = 5p

x2+y2

Nonlinear, Circular domain, V(x, y) = 0

Nonlinear, Circular domain, V(x, y) =x2+y2

Figure

Updating...

References

Related subjects :