Sie sind auf Seite 1von 14

Numerical Solution of Nonlinear Schrödinger

Equation by Using Time-Space Pseudo-Spectral


Method
Mehdi Dehghan, Ameneh Taleei
Department of Applied Mathematics, Faculty of Mathematics and Computer Science,
Amirkabir University of Technology, Tehran 15914, Iran

Received 8 August 2008; accepted 5 February 2009


Published online 30 April 2009 in Wiley InterScience (www.interscience.wiley.com).
DOI 10.1002/num.20468

In this article, a time-space pseudo-spectral method is proposed for the numerical solution of nonlinear
Schrödinger equation. The employed method is based on Chebyshev-Gauss-Lobbato quadrature points.
Using the pseudo-spectral differentiation matrices the problem is reduced to a system of nonlinear algebraic
equations. However, this method is basically a spectral method, but a subdomain-in-time algorithm is used
which yields a smaller nonlinear system to study long-time numerical behavior. Because the time-space
pseudo-spectral method has spectral accuracy, we present numerical experiments which show high accuracy
of this method for the variant nonlinear Schrödinger equations and also particular attention is paid to the
conserved quantities as an indicator of the accuracy. © 2009 Wiley Periodicals, Inc. Numer Methods Partial
Differential Eq 26: 979–992, 2010

Keywords: nonlinear Schrödinger equation (NLS); pseudo-spectral method; soliton; time-space pseudo-
spectral (TSPS) method; Chebyshev-Gauss-Lobbato (CGL) quadrature points; Fourier pseudo-spectral time
splitting (FPTS) method; Gross-Pitaveskii (GP) equation

I. INTRODUCTION

The nonlinear Schrödinger (NLS) equation in its many versions is one of the most important
models of mathematical physics with several applications to different fields such as nonlinear
optic [1], models of protein dynamics [2], plasma physics [3], self-focussing in laser pulses [4]
and many other fields.
We consider the following nonlinear Schrödinger equation

it + axx + b||2  + F (x, t) = 0, (1.1)

with the initial condition

(x, 0) =  0 (x), x ∈ R,

Correspondence to: M. Dehghan, Department of Applied Mathematics, Faculty of Mathematics and Computer Science,
Amirkabir University of Technology, Tehran 15914, Iran (e-mail: mdehghan@aut.ac.ir)
© 2009 Wiley Periodicals, Inc.
980 DEHGHAN AND TALEEI

and the boundary conditions

lim (x, t) = 0, t ∈ [0, T ],


|x|−→∞

where a and b are constant, F (x, t) is a bounded real function and also (x, t) is the
complex-valued wave function.
In the numerical study of the NLS equation, there are two types of solutions which attract
much interest [5]. The first type is the bright soliton solutions, in which the solution and its spatial
derivatives vanish at |x| = ∞. The second type is L-periodic solutions defined by

(x + L, t) = (x, t), −∞ < x < ∞, t > 0.

An hypothesis about the soliton solution interval must be made. For it is impossible to compute
the solution on the whole real line. It must be therefore supposed that solution of Eq. (1.1) has a
compact support on a bounded interval [xL , xR ]. To compute a solution on the interval [xL , xR ],
artificial boundary conditions (xL , t) = (xR , t) = 0 are chosen to model the physical condi-
tions that  −→ 0 for |x| −→ ∞. Solutions decrease rapidly at infinity and it seems reasonable
to use such an approximation [6].
For the NLS equation, rapidly decaying solutions (or periodic solutions, in which case
integration is performed over one space period) are known to satisfy two conservation laws [7]:

(i) mass conservation:


 
E1 (t) = |(x, t)|2 dx = |(x, 0)|2 dx = E1 (0), t > 0.
R R

(ii) energy conservation:



E2 (t) = (a|x (x, t)|2 − b/2|(x, t)|4 − F (x, t)|(x, t)|2 ) dx
R

= (a|x (x, 0)|2 − b/2|(x, 0)|4 − F (x, 0)|(x, 0)|2 ) dx = E2 (0), t > 0.
R

Regarding, the importance for this kind of equations, numerous numerical methods have been
employed such as spectral methods [8–11], finite difference methods [12–14], finite element meth-
ods [6], and many other schemes. These numerical techniques are important tools to understand
the physical behavior of the equations [15].
It has been shown that spectral methods can provide a very useful tool for the solution of time-
dependent PDEs because of their spatially spectral accuracy, when the geometry of the problem
is smooth and regular [16]. The Fourier pseudo-spectral time splitting (FPTS) method is one of
spectral methods which is applied for variant NLS equation by many authors such as Weideman
and Herbst [5], Pathria and Morris [7], Muruganandam and Adhikari [17], and Bao et al [18]. But
this method is very efficient for NLS equation with periodic boundary conditions because it is
based on Fourier approximation in space. For the problems with nonperiodic boundary conditions,
Jacobi polynomials such as (Chebyshev, Legendre, Lagurre, Hermitte) are recommended [19]. In
traditional methods, spectral schemes are used to approximate the space derivatives and the well-
known time integration methods such as Runge-Kutta and multi-step are used in time. Fornberg
and Driscoll [20] used linearly implicit multi-step methods in time and a Fourier pseuo-spectral

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 981

in space for NLS equation with periodic boundary conditions. Javidi and Golbabai [21] stud-
ied pseudo-spectral Chebyshev method with preconditioning in space and Runge-Kutta in time.
But Shen and Wang [10] proposed a new time-space spectral method for this equation which is
based on a Legendre-Galerkin method in space and a dual Petrov Galerkin formulation in time.
The time-space spectral methods could be very effective for certain problems where very high
accuracy is needed over a long-time interval.
However, for equations with more complicated nonlinear terms it is better to use the pseudo-
spectral method than other spectral methods because this method is much easier to program and
the respective algorithms are easily verified. In this article, we apply the pseudo-spectral method
in both space and time variables that is based on Chebyshev-Gauss-Lobbato (CGL) points. Using
the pseudo-spectral differentiation matrices, we reduce the problem to a nonlinear algebraic sys-
tem but for studying long-time behavior, we use a subdomain-in-time algorithm. There is a long
history of space-time spectral methods. We can mention additional references [22–29]. For some
research works on numerical methods for nonlinear Schrödinger equation or the Gross-Pitaveski
equation, we refer the interested reader to published works [30–40] and the references therein.
Also note that the references [30–35] and [38, 40] have presented finite difference procedures.
The article is organized as follows: In Section II, we have presented the necessary tools for the
solution of PDE with time-space pseudo-spectral (TSPS) method. In Section III, we demonstrate
this method for the solution of NLS equation. In Section IV, we give some numerical experiments,
for estimation of numerical solution errors and studying conserved quantities, mass and energy,
by TSPS method. Although our main aim in this work is to study TSPS method for the soliton
solutions but we will also survey the periodic solution of the NLS equation. The conclusion of
this article is presented in Section V.

II. SOME PRELIMINARIES

In time-space pseudo-spectral method, a function of two variables u(x, t) can be approximated


as:


M 
N
u(x, t)  ũij hij (x, t), (2.1)
j =0 i=0

where the coefficients ũij = u(xi , tj ) are called the spectral coefficients and {{hi,j }Ni=0 }M
j =0 are the
basic functions associated with points {xi }i=0 and {tj }j =0 in the x and t directions, respectively.
N M

These basic functions are constructed as a set of Lagrange interpolant associated with the i-th and
j -th grid node as follows:

hi,j (x, t) = φi (x)φj (t),

where


N
x − xn
φi (x) = ,
n=0,n=i
xi − x n


M
t − tm
φj (t) = .
m=0,m=j
tj − t m

Numerical Methods for Partial Differential Equations DOI 10.1002/num


982 DEHGHAN AND TALEEI

In this work, we use Chebyshev polynomials Tk (ξ ) of degree at most k, which are orthogonal
on the interval ξ ∈ [−1, 1]. We interpolate function u(x, t) in the CGL points which are:

(xi , tj ) = (cos(iπ/N ), cos(j π/M)), i = 0(1)N , j = 0(1)M.

We also have the following [41]:

(−1)i (1 − ξ 2 ) dTL (ξ )
φi (ξ ) = ,
ci L2 (ξ − ξi ) dξ

where (ci = 1, i = 1(1)L − 1, c0 = cL = 2), L = M, N and ξ = x, t.


Eq. (2.1) can be rewritten in the following vector-matrix form:

u(x, t) = ([0:N ] (x) ⊗ [0:M] (t))T u, (2.2)

where

u = [ũ00 , . . . , ũ0M , ũ10 , . . . , ũ1M , . . . , ũN 0 , . . . , ũN M ]T ,


[0:L] (ξ ) = [φ0 (ξ ), . . . , φL (ξ )]T .

For the n-th derivatives of the function u(x, t) in x-direction and t-direction we can write [42]:

d n u(x, t)  (n) 
= ( [0:N ] (x) ⊗  [0:M] (t)) T
D [1:N +1,1:N +1] ⊗ I M+1 u, (2.3)
dx n
d n u(x, t)  (n) 
= ( [0:N ] (x) ⊗  [0:M] (t)) T
I N +1 ⊗ D [1:M+1,1:M+1] u, (2.4)
dt n

where D is the differential matrix and [D]ki = dφ dξ


i
(ξk ). For computation of the entries of
differentiation matrix D , n ≥ 1, we use the recursive formula [43, 44]:
(n)




 n ei (n−1)
[D ]k+1,k+1 − [D (n−1)
]k+1,i+1 , k  = i,
[D (n) ]k+1,i+1 = ξk − ξi ek

− N
j =0,j  =k [D ]k+1,j +1 , k = i,
(n)

where [D (0) ] is the identity matrix and the constants ek are defined by

ek = (−1)k , k = 1(1)L − 1, e0 = 1/2, eL = (−1)L /2.

For calculation of discrete conservative properties, mass and energy, we need numerical quad-
rature which is based on CGL points. These points can provide the approximate quadrature
1
−1
v(ξ )dξ = Ni=0 wi v(ξi ) where wi denotes the Chebyshev-Gauss weights and can be computed
in the following form [45]

1
w0 = wN = ,
N2 −1


4 
N /2
1 2π ki
wi = wN −i = cos , i = 1, . . . N /2,
N k=0 1 − 4k 2 N

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 983

when N is even and also


1
w0 = wN = ,
N2
N −1


4 2
1 2π ki N −1
wi = wN −i = cos , i = 1, . . . , ,
N k=0 1 − 4k 2 N 2

when N is odd.

III. THE TIME-SPACE PSEUDO-SPECTRAL METHOD FOR NLS EQUATION

In this section we consider Eq. (1.1) with homogeneous boundary conditions which by using the
affine transformations
xR − xL xR + xL
x −→ x+ ,
2 2
and
T T
t −→ t+ ,
2 2
is converted to the collocation coordinate system as follows:

iϕt + αϕxx + β|ϕ|2 ϕ + f (x, t)ϕ = 0, x, t ∈ [−1, 1],


ϕ(x, −1) = g(x), x ∈ [−1, 1],
ϕ(−1, t) = ϕ(1, t) = 0, t ∈ [−1, 1], (3.1)

xL −xR
where α = (x −x 2T
2 a, β = 2 b, f (x, t) = 2 F (x̂, t), x̂ =
T T
2
x + xL +x
2
R
.
R L)
Since for the numerical solution using time-space pseudo-spectral method, the equation with
homogeneous boundary, and initial conditions leads to more simple algebraic system, therefore
we reduce the problem to an equivalent problem with homogeneous boundary and initial condi-
tions. But the boundary conditions are homogeneous; thus, it is enough that we homogenize the
initial condition. For this work, a simple function (x, t) is used that satisfies the inhomogeneous
initial condition [16]. We can choose (x, t) as

1−t
(x, t) = g(x).
2
Now we define the new variable ψ,

ψ = ϕ − , (3.2)

which satisfies in the homogeneous initial condition. By using (3.1) and (3.2) we can write

iψt + αψxx + β(|ψ + |2 )(ψ + ) + i t + α xx + f (x, t)(ψ + ) = 0. (3.3)

Numerical Methods for Partial Differential Equations DOI 10.1002/num


984 DEHGHAN AND TALEEI

For our numerical solution, we decompose the complex functions ψ and in real and imaginary
parts by writing ψ(x, t) = u(x, t) + iv(x, t) and (x, t) = p(x, t) + iq(x, t), where u, v, p and
q are real functions. Substituting these in (3.3) yields the following system:

vt − αuxx − β((u + p)2 + (v + q)2 )(u + p) + qt − αpxx − f (x, t)(u + p) = 0,


ut + αvxx + β((u + p)2 + (v + q)2 )(v + q) + pt + αqxx + f (x, t)(v + q) = 0. (3.4)

In order to solve the coupled Eqs. (3.4) we consider approximations u(x, t) and v(x, t) as follows:


N −1 
M
u(x, t) = uij φi (x)φj (t) = ([1:N −1] (x) ⊗ [1:M] (t))T u,
i=1 j =1


N −1 
M
v(x, t) = vij φi (x)φj (t) = ([1:N −1] (x) ⊗ [1:M] (t))T v, (3.5)
i=1 j =1

where

u = [u1,1 , . . . , u1,M , u2,1 , . . . , u2,M , . . . , uN −1,1 , . . . , uN −1,M ]T ,


v = [v1,1 , . . . , v1,M , v2,1 , . . . , v2,M , . . . , vN −1,1 , . . . , vN −1,M ]T .

Note that in the above approximation, in contrast with (2.2), since the initial and boundary con-
ditions are homogeneous, the functions φ0 (x), φN (x), and φ0 (t) are omitted. Now by using (2.4)
and (3.5) we have

ut (x, t) = ([1:N −1] (x) ⊗ [1:M] (t))T (IN −1 ⊗ D[1:M,1:M] )u + ζ1 (x)φ0 (t),
vt (x, t) = ([1:N −1] (x) ⊗ [1:M] (t))T (IN −1 ⊗ D[1:M,1:M] )v + ζ2 (x)φ0 (t), (3.6)

where ζ1 and ζ2 are functions of x and also depend on some components of vectors u and v,
respectively [42]. Also by using (2.3) and (3.5), we can write
 2 
uxx (x, t) = ([1:N −1] (x) ⊗ [1:M] (t))T D[2:N ,2:N ] ⊗ IM u + γ1 (t)φ0 (x) + γ2 (t)φN (x),
 2 
vxx (x, t) = ([1:N −1] (x) ⊗ [1:M] (t))T D[2:N ,2:N ] ⊗ IM v + λ1 (t)φ0 (x) + λ2 (t)φN (x), (3.7)

where γi , i = 1, 2 and λi , i = 1, 2 are functions of t and depend to some components of vectors


u and v, respectively. By substituting (3.6) and (3.7) in (3.4) we will have:
   2  

 ([1:N −1] (x) ⊗ [1:M] (t))T (IN −1 ⊗ D[1:M,1:M] )v − α D[2:N ,2:N ] ⊗ IM u



 − β((u(x, t) + p(x, t))2 + (v(x, t) + q(x, t))2 )(u(x, t) + p(x, t)) + qt (x, t)



 − αpxx (x, t) − f (x, t)(u(x, t) + p(x, t)) + ζ2 φ0 (t) − α(γ1 (t)φ0 (x) + γ2 (t)φN (x)) = 0,
  2  

 ([1:N −1] (x) ⊗ [1:M] (t))T (IN −1 ⊗ D[1:M,1:M] )u + α D[2:N ,2:N ] ⊗ IM v





 + β((u(x, t) + p(x, t))2 + (v(x, t) + q(x, t))2 )(v(x, t) + q(x, t)) + pt (x, t)


+ αqxx (x, t) + f (x, t)(v(x, t) + q(x, t)) + ζ1 φ0 (t) + α(λ1 (t)φ0 (x) + λ2 (t)φN (x)) = 0.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 985

Now by collocating the above equations at the points (ξi , ηj ), i = 1(1)N − 1, j = 1(1)M, and
by noting that the values of φ0 (x), φN (x) and φ0 (t) at these collocation points are zero, finally we
get the following nonlinear system
  2 

 (IN −1 ⊗ D[1:M,1:M] )v − α D[2:N ,2:N ] ⊗ IM u


 − βIuv (u + p) + (qt − αpxx ) − If (u + p) = 0,
F (U , V ) =  2 

 (IN −1 ⊗ D[1:M,1:M] )u + α D[2:N ,2:N ] ⊗ IM v



− βIuv (v + q) + (pt + αqxx ) − If (v + q) = 0,
where
p = [p(ξ1 , η1 ), . . . , p(ξ1 , ηM ), . . . , p(ξN −1 , η1 ), . . . , p(ξN −1 , ηM )]T ,
q = [q(ξ1 , η1 ), . . . , q(ξ1 , ηM ), . . . , q(ξN −1 , η1 ), . . . , q(ξN −1 , ηM )]T ,
If = diag(f (ξ1 , η1 ), . . . , f (ξ1 , ηM ), . . . , f (ξN −1 , η1 ), . . . , f (ξN −1 , ηM )),
Iuv = diag((u + p)2 + (v + q)2 ).
The above nonlinear system can be solved by using the numerical methods such as the well-known
Newton method. The Jacobian matrix of this nonlinear system is as
  2   2  
−α D[2:N ,2:N ] ⊗ IM − β 3Iup + Ivq2
− If  2(IN −1 ⊗ D[1:M,1:M]
 ) − 2βIup Ivq
J = ,
(IN −1 ⊗ D[1:M,1:M] ) − 2βIup Ivq α D[2:N ,2:N ] ⊗ IM − β Iup2
+ 3Ivq
2
− If
where Iup and Ivq are diagonal matrices corresponding to vectors u + p and v + q respectively.
This Jacobian matrix J2MN ×2MN has MN(N +M −1) nonzero elements, thus this matrix is sparse.
We mention that the Jacobian matrix may be used by nonlinear equation solvers.
But this method is a fully implicit method; therefore, for studying function behavior in long-
time we use a subdomain-in-time algorithm which leads to solve smaller nonlinear systems. In
this technique first we partition the time interval [0, T ] into equidistant intervals of length l by
setting K = [T /l], tk = lk, Ik = (tk , tk+1 ) for 0 ≤ k ≤ K − 1 and thus for each interval Ik we
implement TSPS method. Note that if M be the number of points in interval Ik then nt = MK is
[10, 46] the number of points in interval [0, T ].

IV. NUMERICAL RESULTS

In this section some numerical experiments are made to verify efficiency and high accuracy of
TSPS method for variant NLS equations.
For estimating the error of the numerical solution we use the L∞ error norm defined by
 
L∞ = max jexact
approx 
,n − j ,n ,
j

where j ,n is the solution at the point (xj , tn ).


The conservation properties of the TSPS method are examined by calculating discrete ana-
logues of the conserved quantities E1 and E2 . For studying these conservation, Ė1 and Ė2 denote
the discrete quantities mass and energy, respectively and for the error of conservation we set
δ1 (t) = |Ė1 (t) − E1 (0)|,
δ2 (t) = |Ė2 (t) − E2 (0)|.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


986 DEHGHAN AND TALEEI

TABLE I. The values of errors at different times for the single soliton with s = 4 in Example 1.
N = 180, M = 5 N = 220, M = 10
Time (nx , nt ) L∞ (nx , nt ) L∞
0.25 (180, 5) 8.314 E −8 (220,10) 3.599 E −9
0.5 (180,10) 8.042 E −8 (220,20) 3.997 E −9
0.75 (180,15) 8.042 E −8 (220,30) 3.996 E −9
1 (180,20) 8.320 E −8 (220,40) 3.601 E −9
1.25 (180,25) 8.352 E −8 (220,50) 3.765 E −9

Recently, soliton solutions have attracted considerable attention due to their application in
different fields of modern physics. First we study TSPS method for the soliton solutions of cubic
nonlinear Schrödinger (CNLS) and generalized nonlinear Schrödinger (GNLS) equations and
then we see the periodic solution of NLS equation by this method.

Example 1. One-Soliton Solution for the CNLS.


Considering the Eq. (1.1) with parameters a = 1, b = 2, and F ≡ 0, we have

it + xx + 2||2  = 0,

which is known as the cubic nonlinear Schrödinger equation. This equation has important
applications in many fields of physics.
The analytic soliton solution of this problem is as follows [6]

21 

2 1 1 2
(x, t) = α expi sx − s − α t sechα(x − st),
2
q 2 4

where s is the speed of the soliton. For this test, we choose α = 1 and q = 1 with s = 4 and 1/4.
When the velocity is s = 4, in Table I we see the L∞ error of the numerical solution with respect
to the exact solution with (N = 180, M = 5) and (N = 220, M = 10). In our computation, to
use homogeneous boundary conditions up to time t = 1.25, it is enough to choose the interval
[xL , xR ] = [−20, 25]. For s = 1/4, Table II shows the L∞ error of the numerical solution with
respect to the exact solution until time t = 5 with (N = 140, M = 5) and (N = 160, M = 5).
In this case, the interval [−20, 20] is suitable. Tables I and II demonstrate the velocity of travel-
ing solitary wave in the studying numerical behavior is very efficient. For these kinds of waves
with high speed in long-time we usually need to larger intervals to make homogeneous bound-
ary conditions which lead to choose more number of points. Figures 1 and 2 show the curve

TABLE II. The values of errors at different times for the single soliton with s = 1/4 in Example 1.

N = 140, M = 5 N = 160, M = 5
Time (nx , nt ) L∞ (nx , nt ) L∞
1 (140, 5) 1.483 E −8 (160,5) 3.984 E −9
2 (140,10) 4.333 E −8 (160,10) 4.893 E −9
3 (140,15) 3.392 E −8 (160,30) 1.505 E −8
4 (140,20) 4.431 E −8 (160,40) 9.779 E −9
5 (140,25) 3.572 E −8 (160,50) 7.953 E −9

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 987

FIG. 1. The variation of the conservation errors of TSPS method for Example 1(s = 4, N = 220, M = 10):
left figure shows the error of mass conservation and right figure presents the error of energy conservation.

errors of the analogue mass and energy for Example 1 with (s = 4, N = 220, M = 10) and
(s = 1/4, N = 160, M = 5), respectively. These figures show the conserved quantities, mass
and energy, are very well preserved by TSPS method.

Example 2. One-Soliton Solution for the GNLS.


In this example we consider the equation (1.1) with parameters a = −1, b = 1, and

F (x, t) = ω + 2si(x − st) − 4(x − st)2 + 4ik(x − st) + k 2 + 2 + (exp(−2(x − st)2 ))2 ,

that the exact solution is as follows [30]

(x, t) = exp(−(x − st)2 + i(kx − ωt)).

FIG. 2. The variation of the conservation errors of TSPS method for Example 1(s = 1/4, N = 160, M =
5): left figure shows the error of mass conservation and right figure presents the error of energy conservation.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


988 DEHGHAN AND TALEEI

TABLE III. The values of errors at time t = 1 for the single soliton in Example 2.

(N, M) (50,8) (80,12) (100,14)


L∞ 5.295 E −3 7.361 E −9 3.236 E −12

TABLE IV. The values of errors at different times for the single soliton in Example 2.
N = 100, M = 10 N = 120, M = 12
Time (nx , nt ) L∞ (nx , nt ) L∞
1 (100,10) 2.182 E −7 (120,12) 2.603 E −9
2 (100,20) 2.539 E −7 (120,24) 3.500 E −9
3 (100,30) 2.313 E −7 (120,36) 3.496 E −9
4 (100,40) 2.157 E −7 (120,48) 2.594 E −9
5 (100,50) 3.310 E −7 (120,60) 3.187 E −9

For this test we choose s = 2, k = −1 and ω = −3, so that

F (x, t) = −4(x − 2t)2 + (exp{−2(x − 2t)2 })2 .

The velocity of the soliton solution is s = 2. In Table III the numerical solutions are com-
pared with the exact solution at t = 1 on the interval x ∈ [−10, 10] with different values of M
and N.
Table IV shows the L∞ error of the numerical solution with respect to the exact solution
with (N = 100, M = 10) and (N = 120, M = 12). This problem was solved on the interval
−10 ≤ x ≤ 20 up to time t = 5. In Fig. 3, we present the curve errors of the discrete conservation
properties, mass and energy, for Example 2 with s = 2, N = 120, M = 12. In this figure also we
observe that the conservation errors are very small.
The above results show the efficiency of this method for NLS equation with the soliton solution.
Now we employ this method for NLS equation with the periodic solution and the result is
compared with SCMP method [21].

FIG. 3. The variation of the conservation errors of TSPS method for Example 2(s = 2, N = 120, M = 12):
left figure shows the error of mass conservation and right figure presents the error of energy conservation.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 989

TABLE V. The values of errors at different times for GP equation in Example 3.


N = 10, M = 10 N = 14, M = 14
Time (nx , nt ) L∞ (nx , nt ) L∞
5 (10,20) 2.398 E −6 (14,24) 6.658 E −10
10 (10,40) 2.323 E −6 (14,48) 9.776 E −10
20 (10,60) 6.511 E −6 (14,72) 1.108 E −9
30 (10,80) 1.621 E −5 (14,96) 1.631 E −9

TABLE VI. Comparison of errors at different times for the GP equation in Example 3.
TSPS SCMP[21]
Method N = 16, M = 16 N = 16, dt = 0.001
Time (nx , nt ) L∞ (nx , nt ) L∞
8 (16,32) 6.073 E −12 (16,8000) 1.143 E −9
16 (16,64) 6.219 E −11 (16,16000) 9.790 E −10
24 (16,96) 1.834 E −10 (16,24000) 4.163 E −10
32 (16,128) 4.386 E −10 (16,32000) 2.642 E −9

Example 3. Gross-Pitaevskii (GP) equation.


For study nonlinear Schrödinger equation with periodic solution we can consider the GP equa-
tion. For this work, we set a = 21 , b = −1, and F (x, t) = −cos2 (x) in the Eq. (1.1) under the
periodic boundary condition. For this equation, the exact solution is [14]

(x, t) = sin(x)exp(−3/2it).

Table V gives the L∞ error of the numerical solution for Example 3 with respect to the exact
solution until time t = 30. Table VI shows the comparison of the accuracy by both the TSPS
and SCMP methods for this problem until time t = 32. The numerical errors of both methods
for this case are listed in Table VI. Tables V and VI show that the TSPS method gives a good
approximation for NLS equation with the periodic solution by using a sufficiently small number
of points. In Fig. 4 we present the conservation errors δ1 and δ2 , respectively. We observe the

FIG. 4. The variation of the conservation errors of TSPS method for Example 3(N = 16, M = 16): left
figure shows the error of mass conservation and right figure presents the error of energy conservation.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


990 DEHGHAN AND TALEEI

TSPS method preserves each of the conserved quantities very well (up to 10−11 for δ1 and up to
10−10 for δ2 ).
It is worth to point out that the new technique developed in the current paper can be employed
to solve the time-dependent partial differential equations investigated in [47–53].

V. CONCLUSION

In this article, we studied a pseudo-spectral method in time and space variables with Chebyshev-
Gauss-Lobbato points for nonlinear Schrodinger equation. For the solution of nonlinear Schro-
dinger equation via time-space pseudo-spectral method we need to solve a nonlinear system of
equations while the Jacobian matrix corresponding to this system is sparse. Also by applying
the subdomain-in-time algorithm we could survey function behavior in a long-time by using
smaller nonlinear systems. Using a sufficiently small number of points, this method gives a
good approximation. To show the efficiency of this method we presented several examples.
This method preserves conservation quantities and the conserved behavior provides a valuable
cheek on our numerical results. The time-step pseudo-spectral method could be very effec-
tive for these kinds of problems where very high-accurate solution is needed over a long-time
interval.

The authors are very grateful to the three reviewers for carefully reading the paper.

References
1. G. P. Agrawal, Nonlinear fiber optics, 3rd Ed., Academic Press, San Diego, 2001.
2. A. P. Fordy, Soliton theory: a survey of results, Manchester University Press, Manchester, 1990.
3. L. Stenflo and M. Y. Yu, Nonlinear wave modulation in a cylindrical plasma, IEEE Trans Plasma Sci
25 (1997), 1155–1157.
4. C. Sulem and P. L. Sulem, The nonlinear Schrödinger equation: self-focusing and wave collapse,
Springer, New York, 1999.
5. J. A. C. Weideman and B. M. Herbst, Split-step methods for the solution of the nonlinear Schrödinger
equation, SIAM J Numer Anal 23 (1986), 485–507.
6. I. Dag, A quadratic B-spline finite element method for solving nonlinear Schrödinger equation, Comput
Methods Appl Mech Eng 174 (1999), 247–258.
7. D. Pathria and J. L. Morris, Pseudo-spectral solution of nonlinear Schrödinger equations, J Comput
Phys 87 (1990), 108–125.
8. W. Bao, D. Jaksch, and P. A. Markowich, Numerical solution of the Gross-Pitaevskii equation for
Bose-Einstein condensation, J Comput Phys 187 (2003), 318–342.
9. W. Bao and J. Shen, A fourth-order time-splitting Lagurre-Hermite pseudo-spectral method for
Bose-Einstein condensates, SIAM J Sci Comput 26 (2005), 2010–2028.
10. J. Shen and L. L. Wang, Fourierization of the Legendre-Galerkin method and a new space-time spectral
method, Appl Numer Math 57 (2007), 710–720.
11. S. Yu, S. Zhao, and G. W. Wei, Local spectral time splitting method for first and second-order partial
differential equations, J Comput Phys 206 (2005), 727–780.
12. M. Delfour, M. Fortin, and G. Payre, Finite difference solutions of a nonlinear Schrödinger equation, J
Comput Phys 44 (1981), 277–288.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


NUMERICAL SOLUTION OF NONLINEAR SCHRÖDINGER EQUATION 991

13. S. R. K. Iyengar, G. Jayaraman, and V. Balasubramanian, Variable mesh difference schemes for solving a
nonlinear Schrödinger equation with a linear damping term, Comput Math Appl 40 (2000), 1375–1385.
14. H. Q. Wang, Numerical studies on the split-step finite difference method for nonlinear Schrödinger
equations, Appl Math Comput 170 (2005), 17–35.
15. M. Dehghan, Finite difference procedures for solving a problem arising in modeling and design of
certain optoelectronic devices, Math Comput Simul 71 (2006), 16–30.
16. J. P. Boyd, Chebyshev and Fourier spectral methods, Springer-Verlag (Berlin, New York), 1989.
17. P. Muruganandam and S. K. Adhikari, Bose-Einstein condensation dynamics in three dimensions by
the pseudo-spectral and finite difference methods, J Phys B 36 (2003), 2501–2513.
18. W. Bao, S. Jin, and P. A. Markowich, On time-splitting spectral approximation for the Schrödinger
equation in the semiclassical regime, J Comput Phys 175 (2002), 487–524.
19. B. Fornberg, A practical guide to pseudo-spectral methods, Cambridge University Press, 1996.
20. B. Fornberg and T. A. Driscoll, A fast spectral algorithm for nonlinear wave equation with linear
dispersion, J Comput Phys 155 (1999), 456–467.
21. M. Javidi and A. Golbabai, Numerical studies on nonlinear Schrödinger equations by spectral collocation
method with preconditioning, J Math Anal Appl 333 (2007), 1119–1127.
22. M. Dehghan and M. Shamsi, Numerical solution of two-dimensional parabolic equation subject to non-
standard boundary specifications using the pseudospectral Legendre method, Numer Methods Partial
Differential Eq 22 (2006), 1255–1266.
23. G. Delic, Spectral function methods for nonlinear diffusion equations, J Math Phys 28 (1987), 39–59.
24. M. Deville, L. Kleiser, and F. Montigny-Rannou, Pressure and time treatment for Chebyshev spectral
solution of a Stokes problem, Int J Numer Meth Fluids 4 (1984), 1149–1163.
25. P. Dutt, Spectral methods for initial boundary value problems, an alternative approach, SIAM J Numer
Anal 27 (1990), 885–903.
26. D. L. Lewis, J. Lund, and K. L. Bowers, The space-time sinc-Galerkin method for parabolic problems,
Int J Num Methods Eng 24 (1987), 1629–1644.
27. G. Ierley, B. J. Spencer, and R. A. Worthing, Spectral methods in time for a class of parabolic partial
differential equations, J Comput Phys 102 (1992), 88–97.
28. Y. Morchoisne, Pseudo-spectral space-time calculations of incompressible viscous flows, AIAA 81
(1981), 81–109.
29. Y. Morchoisne, Resolution of Navier-Stokes equations by a space-time pseudospectral method, Rech
Aerosp 5 (1979), 293–306.
30. Q. S. Chang, E. H. Jia, and W. Sun, Difference schemes for solving the generalized nonlinear Schrödinger
equation, J Comput Phys 148 (1999), 397–415.
31. M. M. Cerimele, M. L. Chiofalo, F. Pistella, S. Succi, and M. P. Tosi, Numerical solution of the Gross-
Pitaevskii equation using an explicit finite difference scheme: an application to trapped Bose-Einstein
condensates, Phys Rev E 62 (2000), 1382–1389.
32. J. B. Chen, M. Z. Qin, and Y. F. Tang, Symplectic and multi-symplectic methods for the nonlinear
Schrödinger equation, Comput Math Appl 43 (2002), 1095–1106.
33. W. Dai, Absolute stable explicit and semi-explicit schemes for Schrödinger equations, Math Numer
Sinica 11 (1989), 128–131.
34. F. Ivanauskas and M. Radziunas, On convergence and stability of the explicit difference method for
solution of nonlinear Schrödinger equations, SIAM J Numer Anal 36 (1999), 1466–1481.
35. M. C. Lai and C. Y. Huang, A simple Dufort-Frankel type scheme for the Gross-pitaevskii equation of
Bose-Einstein condensates on different geometries, Numer Methods Partial Differential Eq 20 (2004),
628–638.

Numerical Methods for Partial Differential Equations DOI 10.1002/num


992 DEHGHAN AND TALEEI

36. G. M. Muslu and H. A. Erbay, High-order split-step Fourier schemes for the generalized nonlinear
Schrödinger equation, Math Comput Simul 67 (2005), 581–592.
37. H. P. Pfeiffer, L. E. Kidder, M. A. Scheel, and S. A. Teukolsky, A multidomain spectral method for
solving elliptic equations, Comput Phys Commun 152 (2003), 253–273.
38. J. I. Ramos, Linearly implicit methods for the nonlinear Schrödinger equation in nonhomogeneous
media, Appl Math Comput 133 (2002), 1–28.
39. S. Reich, Multi-symplectic Runge-Kutta methods for Hamiltonian wave equations, J Comput Phys 157
(2000), 473–499.
40. L. Wu, Dufort-Frankel type methods for linear and nonlinear Schrödinger equations, SIAM J Numer
Anal 33 (1996), 1526–1533.
41. C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spectral methods in fluid dynamics,
Springer-Verlag, New York, 1988.
42. M. Shamsi and M. Dehghan, Recovering a time-dependent coefficient in a parabolic equation from over-
specified boundary data using the pseudo-spectral Legendre method, Numer Methods Partial Differential
Eq 23 (2007), 196–210.
43. R. Baltensperger and M. R. Trummer, Spectral differencing with a twist, SIAM J Sci Comput 24 (2003),
1465–1487.
44. B. D. Welfert, Generation of pseudo-spectral differentiation, SIAM J Numer Anal 34 (1997), 1640–1657.
45. L. N. Trefethen, Spectral methods in Matlab, SIAM, Philadelphia, 2000.
46. J. G. Tang and H. P. Ma, Single and multi-interval spectral methods in time for parabolic equations,
Numer Methods Partial Differen Equations 22 (2006), 1007–1034.
47. M. Dehghan, On the solution of an initial-boundary value problem that combines Neumann and integral
condition for the wave equation, Numer Methods Partial Differential Eq 21 (2005), 24–40.
48. M. Dehghan, Parameter determination in a partial differential equation from the overspecified data,
Math Comput Model 41 (2005), 196–213.
49. M. Dehghan, A computational study of the one-dimensional parabolic equation subject to nonclassical
boundary specifications, Numer Methods Partial Differential Eq 22 (2006), 220–257.
50. M. Dehghan, The one-dimensional heat equation subject to a boundary integral specification, Chaos,
Solitons Fractals 32 (2007), 661–675.
51. M. Dehghan, Implicit collocation technique for heat equation with non-classic initial condition, Int J
Nonlinear Sciences Numer Simul 7 (2006), 447–450.
52. M. Dehghan, Efficient techniques for the second-order parabolic equation subject to nonlocal
specifications, Appl Numer Math 52 (2005), 39–62.
53. M. Dehghan, Identification of a time-dependent coefficient in a partial differential equation subject to
an extra measurement, Numer Methods Partial Differential Eq 21 (2005), 611–622.

Numerical Methods for Partial Differential Equations DOI 10.1002/num

Das könnte Ihnen auch gefallen