Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are encouraged to visit:
http://www.elsevier.com/copyright
Contents lists available atScienceDirect
Applied Numerical Mathematics
www.elsevier.com/locate/apnum
An analysis of the finite-difference method for one-dimensional Klein–Gordon equation on unbounded domain ✩
Houde Han
∗, Zhiwen Zhang
Department of Mathematical Sciences, Tsinghua University, Beijing 100084, PR China
a r t i c l e i n f o a b s t r a c t
Article history:
Received 4 September 2007
Received in revised form 21 October 2008 Accepted 24 October 2008
Available online 6 November 2008 Keywords:
Artificial Boundary Condition (ABC) Unbounded domain
Energy method Fast algorithm
Discrete Artificial Boundary Condition (DABC)
The numerical solution of the one-dimensional Klein–Gordon equation on an unbounded domain is analyzed in this paper. Two artificial boundary conditions are obtained to reduce the original problem to an initial boundary value problem on a bounded computational domain, which is discretized by an explicit difference scheme. The stability and convergence of the scheme are analyzed by the energy method. A fast algorithm is obtained to reduce the computational cost and a discrete artificial boundary condition (DABC) is derived by the Z-transform approach. Finally, we illustrate the efficiency of the proposed method by several numerical examples.
©2008 IMACS. Published by Elsevier B.V. All rights reserved.
1. Introduction
The Klein–Gordon equation arises in relativistic quantum mechanics and field theory, which is of great importance for the high energy physicists [15], and is used to model many different phenomena, including the propagation of dislocations in crystals and the behavior of elementary particles. The one-dimensional Klein–Gordon equation is given by the following partial differential equation:
¯ h2∂2u
∂t2 − ¯h2c2∂2u
∂x2 +m2c4u= f(x,t), ∀x∈R1, t>0, (1.1)
where u=u(x,t) represents the wave density at position x and time t, h¯ is the Planck constant, c and m are particle velocity and particle mass, respectively.
There are a lot of studies on the numerical solution of initial and initial-boundary problems of the linear or nonlinear Klein–Gordon equation. For example, Khalifa and Elgamal [22] developed a numerical scheme based on a finite element method for the nonlinear Klein–Gordon equation with Dirichlet boundary condition on a bounded domain, which shows the overflow solution as expected. Duncan [6] analyzed three finite difference approximations of the initial nonlinear Klein–
Gordon equation, showed they are directly related to symplectic mappings and tested the schemes on the traveling wave and periodic breather problems over long time intervals. However, when we wish to solve the Klein–Gordon numerically on an unbounded domain, these methods will face essential difficulties. Since the unboundedness of the physical domain in our problem, the standard finite element method or finite difference method can’t be used directly.
✩ This work was supported by The NSFC Project. No. 10471073.
*
Corresponding author.E-mail addresses:hhan@math.tsinghua.edu.cn(H. Han),zhangzhiwen02@mails.tsinghua.edu.cn(Z. Zhang).
0168-9274/$30.00©2008 IMACS. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.apnum.2008.10.005
The artificial boundary condition (ABC) method is a powerful approach to reduce the problems on the unbounded domain to a bounded computational domain. In the well-known paper of Engquist and Majda [12], absorbing boundary conditions using Padé approximation of a pseudo-differential operator on a line-type boundary for the wave equation are derived.
Higdon [20,21] developed radiation boundary conditions for the numerical modeling of dispersive waves. By specifying the wavenumber and frequency parameters, the boundary conditions based on compositions of simple first-order differen- tial operators was got. His formulas can be applied without modification to higher-dimensional problem. Low-order local ABCs may have low accuracy, whereas high-order local ABCs are usually hard to implement because they typically in- volve high-order derivatives [13]. The local ABCs may generate some nonphysical reflection at the artificial boundary and the well-posedness of the resulting truncated initial-boundary problem is still open in general. Nonlocal artificial bound- ary conditions have the potential of being more accurate than the local ones. Han and Zheng [19] obtained three kinds of exact nonreflecting boundary conditions for exterior problems of wave equations in two and three-dimensional space by an approach based on Duhamel’s principle. X. Antoine and C. Besse [2] obtained a nonreflecting boundary conditions for the one-dimensional Schrödinger equation. X. Antoine, C. Besse and V. Mouysset [3] also generalized their approach to simulate the two-dimensional Schrödinger equation using nonreflecting boundary conditions. Han and Huang [16], Han, Yin and Huang [18] derived the exact nonreflecting boundary conditions for two- and three-dimensional Schrödinger equations.
Han and Yin [17] also derived the exact nonreflecting boundary conditions for two- and three-dimensional Klein–Gordon equations. Generally speaking, the exact boundary conditions require more computational cost. In order to overcome this disadvantage, a fast algorithm is given in our paper.
The organization of this article is the following: In Section 2 we introduce two artificial boundaries and find the artificial boundary conditions, then reduce the original problem to an equivalent problem on the bounded computational domain. In Section 3, a finite-difference scheme for the reduced problem is given and its stability and convergence are analyzed. A fast algorithm is obtained in Section 4 to reduce the computational cost. In Section 5, a discrete artificial boundary condition (DABC) is derived by the Z-transform approach. Some numerical results will be given in Section 6 to demonstrate the accuracy and efficiency of the proposed methods.
2. The artificial boundary condition
In this section, we study the numerical approximation of a dispersive wave solutionu(x,t), to the equation with a source term on an unbounded domain. More precisely, we consider the following linear Klein–Gordon equation onR1× [0,T]:
∂2u
∂t2 −a2∂2u
∂x2 +b2u= f(x,t), ∀x∈R1, t∈ [0,T], (2.1)
u|t=0=ϕ0(x), ∀x∈R1, (2.2)
ut|t=0=ϕ1(x), ∀x∈R1. (2.3)
Herea,bare two real constants. Assumeϕ0(x),ϕ1(x)and f(x,t)satisfying: Supp{ϕ0(x)} ⊂ [xl,xr], Supp{ϕ1(x)} ⊂ [xl,xr]and Supp{f(x,t)} ⊂ [xl,xr] × [0,T]. For simplicity of the deduction, we takexl= −1 andxr=1.
In order to reduce the problem (2.1)–(2.3) to a bounded computational domain, we introduce two artificial boundaries, Σr=
(x,t)|x=1, 0tT , Σl=
(x,t)|x= −1, 0tT , which divide R1× [0,T]into three parts,
Dl=
(x,t)| −∞x−1, 0tT , Di=
(x,t)| −1x1, 0tT , Dr=
(x,t)|1x+∞, 0tT .
The bounded domain Di is our computational domain. Consider the restriction of u(x,t) on the unbounded domain Dr. u(x,t) satisfies,
∂2u
∂t2 −a2∂2u
∂x2 +b2u=0, ∀(x,t)∈Dr, (2.4)
u|Σr=u(1,t)≡g1(t), (2.5)
u|t=0=0, x1, (2.6)
ut|t=0=0, x1. (2.7)
Since u(1,t) is unknown, the problem (2.4)–(2.7) is incomplete, which cannot be solved independently. If u|Σr =u(1,t)≡ g1(t)is given, the problem above has a unique solution. Let
U(x,s)=L u(x,t)
= +∞
0
e−stu(x,t)dt, Res>0,
denotes the Laplace transform of the unknown solutionu(x,t). By (2.4)–(2.7) it satisfies,
s2U(x,s)−a2Uxx(x,s)+b2U(x,s)=0, 1<x<+∞, (2.8) U(1,s)=G(s)≡L
g1(t)
= +∞
0
e−stg1(t)dt, (2.9)
U(x,s)<+∞, x→ +∞. (2.10)
Eq. (2.8) is a second-order linear ODE with constant coefficient, its general solution is given by U(x,s)=c1(s)exp
−
√s2+b2
a (x−1) +c2(s)exp √
s2+b2
a (x−1) . The condition (2.10) implies c2(s)≡0, and we obtain
U(x,s)=c1(s)exp
−
√s2+b2
a (x−1) , (2.11)
here the roots with positive real parts are taken. The partial derivative with respect tox yields,
∂U(x,s)
∂x = −
√s2+b2
a c1(s)exp
−
√s2+b2
a (x−1) . (2.12)
Combining (2.11) and (2.12) on the artificial boundary Σr, we arrive at
∂U(1,s)
∂x = −
√s2+b2
a U(1,s)= − 1 a√
s2+b2
s2U(1,s)+b2U(1,s)
. (2.13)
By the table of Laplace transform (see page 1108 of [14]), we obtain L−1
1
√s2+b2 = J0(bt),
L−1
s2U(1,s)+b2U(1,s)
= ∂2u(1,t)
∂t2 +b2u(1,t).
Then, by the convolution theorem of Laplace transforms, from (2.13) we obtain:
∂u(1,t)
∂x = −1a t
0
J0(bt−bτ)
∂2u(1,τ)
∂τ2 +b
2u(1,τ)
dτ
= −1a∂u(1,t)
∂t −ba t
0
J0(bt−bτ)∂u(1,τ)
∂t dτ−b
2
a t
0
J0(bt−bτ)u(1,τ)dτ
= −1a∂u(1,t)
∂t −ba2 t
0
J0(bt−bτ)+ J0(bt−bτ)
u(1,τ)dτ.
We get the artificial boundary condition of the problem (2.1)–(2.3) on Σr. Similarly we can get the artificial boundary condition onΣl, witht∈ [0,T]. Hence, we can reduce the initial boundary value problem on the computational domain Di.
∂2u
∂t2 −a2∂2u
∂x2 +b2u= f(x,t), ∀(x,t)∈ [−1,1] × [0,T], (2.14)
u|t=0=ϕ0(x), ut|t=0=ϕ1(x), ∀x∈ [−1,1], (2.15)
∂u(1,t)
∂t +a∂u(1,t)
∂x = −b2 t
0
J0(bt−bτ)+ J0(bt−bτ)
u(1,τ)dτ, (2.16)
∂u(−1,t)
∂t −a∂u(−1,t)
∂x = −b2 t
0
J0(bt−bτ)+ J0(bt−bτ)
u(−1,τ)dτ. (2.17)
Let J(x)=J0(x)+J0(x), whereJ(x)is a special function. By some basic recursion formulas of Bessel functions (see page 242 of [1]), we have
J(x):=J0(x)+J0(x)=J0(x)−
J1(x)= 1 2
J0(x)+J2(x) .
The artificial boundary conditions (2.16), (2.17) will have two simple forms:
∂u(1,t)
∂t +a∂u(1,t)
∂x = −b2 t
0
J(bt−bτ)u(1,τ)dτ, (2.18)
∂u(−1,t)
∂t −a∂u(−1,t)
∂x = −b2 t
0
J(bt−bτ)u(−1,τ)dτ. (2.19)
Moreover, letF(x) denotes the primitive ofJ(x):
F(x)= x
0
J(s)ds= ∞ k=0
(−1)k k!k!(2k+2)
x 2
2k+1
+ ∞ k=0
(−1)kx k!(k+2)!(2k+3)
x 2
2k+3
.
Integrating by parts in Eqs. (2.18) and (2.19), we obtain anther two equivalent forms of boundary conditions. The new forms will bring some convenience for the proof of stability and convergence.
∂u(1,t)
∂t +a∂u(1,t)
∂x = −b t
0
F(bt−bτ)∂u(1,τ)
∂τ dτ. (2.20)
∂u(−1,t)
∂t −a∂u(−1,t)
∂x = −b
t
0
F(bt−bτ)∂u(−1,τ)
∂τ dτ. (2.21)
Next, we discuss the uniqueness and stability estimate of the reduced problem (2.14)–(2.17). Multiply Eq. (2.1) by ∂∂ut and integrate with respect tox∈ [−1,1]for fixedt∈(0,T], using the boundary condition (2.16), (2.17), we get
d dt
1 2 1
−1
∂u(x,t)
∂t
2
+a2
∂u(x,t)
∂x
2
+b2u2(x,t)
dx
+a2∂u(1,t)
∂t B u(1,t)
+a2∂u(−1,t)
∂t B
u(−1,t)
= 1
−1
f(x,t)∂u(x,t)
∂t dx. (2.22)
Here B
ν(t)
=1 a
dν(t) dt +b2
a t
0
J0(bt−bτ)+ J0(bt−bτ) ν(τ)dτ.
We introduce two auxiliary functions W(1)(x,t) and W(2)(x,t), which satisfy the following problems, respectively:
Wtt(1)−a2Wxx(1)+b2W(1)=0, ∀(x,t)∈Dr, W(1)|Σr=u(1,t), 0tT,
W(1)|t=0=0, 1<x<+∞, Wt(1)|t=0=0, 1<x<+∞
and
Wtt(2)−a2Wxx(2)+b2W(2)=0, ∀(x,t)∈Dl, W(2)|Σl=u(−1,t), 0tT,
W(2)|t=0=0, −∞<x<−1, Wt(2)|t=0=0, −∞<x<−1.
Through a similar analysis, we obtain,
a2∂u(1,t)
∂t B u(1,t)
=dtd 1
2 +∞
1
∂W(1)(x,t)
∂t
2
+a2
∂W(1)(x,t)
∂x
2
+b2
W(1)(x,t)2 dx
, (2.23)
and
a2∂u(−1,t)
∂t B
u(−1,t)
= d dt
1 2
−1
−∞
∂W(2)(x,t)
∂t
2
+a2
∂W(2)(x,t)
∂x
2
+b2
W(2)(x,t)2 dx
. (2.24)
We introduceE(t)and F(t) as following, E(t)= 1
2 1
−1
∂u(x,t)
∂t
2
+a2
∂u(x,t)
∂x
2
+b2u2(x,t)
dx
+ 12
+∞
1
∂W(1)(x,t)
∂t
2
+a2
∂W(1)(x,t)
∂x
2
+b2
W(1)(x,t)2 dx
+ 1 2
−1
−∞
∂W(2)(x,t)
∂t
2
+a2
∂W(2)(x,t)
∂x
2
+b2
W(2)(x,t)2
dx, (2.25)
F(t)= 12 1
−1
f(x,t)2
dx. (2.26)
Combining (2.22)–(2.26) and using Cauchy–Schwarz inequality, we obtain, d
dtE(t)E(t)+F(t), 0tT. (2.27)
Using the Gronwall inequality from inequality (2.27) and noticing that, E(0)=1
2 1
−1
ϕ1(x)2
+a2 ϕ1(x)2
+b2
ϕ0(x)2 dx.
we get the stability estimate for the solution of the reduced problem (2.14)–(2.17):
1
−1
∂u(x,t)
∂t
2
+a2
∂u(x,t)
∂x
2
+b2u2(x,t)
dx
1
−1
ϕ1(x)2
+a2 ϕ1(x)2
+b2
ϕ0(x)2 dx+
t
0
1
−1 et−τ
f(x,τ)2
dx dτ. (2.28)
Then, we obtain the following stability estimate of the reduced problem (2.14)–(2.17).
Theorem 2.1.The reduced problem(2.14)–(2.17)has at most one solution u(x,t)on the bounded computational domain Di, and u(x,t)continuously depends on the initial value{ϕ0(x),ϕ1(x)}, and f(x,t).
From Theorem 2.1, we know that the reduced problem (2.14)–(2.17) is equivalent to the original problem (2.1)–(2.3).
Namely, the solution of the reduced problem (2.14)–(2.17) is the restriction of the solution of original problem (2.1)–(2.3) on the bounded domain Di, vice versa.
3. Analysis of the difference scheme
In this section, we consider the finite difference approximation of the reduced problem (2.14)–(2.17) on the bounded domain Di. We divide the domain Di by a set of lines parallel to thex- and t-axes to form a grid. We writeh=1/I and τ=T/N for the line spacings, whereI and N are two positive integers. The crossing pointsΩhτ are called the grid points,
Ωhτ=
(xi,tn)|xi= −1+ih, i=0,1, . . . ,2I; tn=nτ, n=0,1, . . . ,T/τ.
SupposeU= {uni |0i2I,n0} is a grid function on Ωhτ. For the simplicity, assume the constantsa=b=1. Introduce the following notations [24]:
uni−1/2= 12
uni−1+uni
, uni−1/2=12
uni +uni−1
,
δxuni−1/2= 1 h
uni −uni−1
, δtuni−1/2= 1 τ
uni −uni−1 , δ0xuni = 1
2h
uni+1−uni−1
, δt0uni = 1 2τ
uni+1−uni−1 , δ2xuni = 1
h2
uni+1−2uni +uni−1
, δt2uni = 1 τ2
uni+1−2uni +uni−1 .
We denote the value of the solutionu(x,t) at the grid point(xi,tn)byUni and fin= f(xi,tn). Using the Taylor expansion, it follows from (2.14), (2.15), (2.21), (2.20) that:
δt2Uin−δ2xUni +Uni = fin+Rni, 1i2I−1, n1, (3.1)
U0i =φ0(ih), 0i2I, (3.2)
U1i =U0i +τφ1(ih)+S1i, 0i2I, (3.3)
δtUn2I+1/2+δxUn2I−1/2= − n
m=1
δ0tU2Im
mτ
(m−1)τ
F(nτ−s)ds+Pn2I, n2, (3.4)
δtUn0+1/2−δxUn1/2= − n
m=1
δt0Um0
mτ
(m−1)τ
F(nτ−s)ds+Q0n, n2. (3.5)
If the solutionu(x,t)is smooth enough, there exists a constant C, such that RniC
h2+τ2, S1iCτ2, Pn2IC(h+τ), Q0nC(h+τ).
Omitting the truncation errors in (3.1)–(3.5), we construct a finite difference scheme of the reduced problem (2.14)–(2.17):
δt2uni −δ2xuni +uni = fin, 1i2I−1, n1, (3.6)
u0i =φ0(ih), 0i2I, (3.7)
u1i =u0i +τφ1(ih), 0i2I, (3.8)
δtun2I+1/2+δxun2I−1/2= − n
m=1
δ0tum2I
mτ
(m−1)τ
F(nτ−s)ds, n2, (3.9)
δtun0+1/2−δxun1/2= − n
m=1
δt0um0
mτ
(m−1)τ
F(nτ−s)ds, n2. (3.10)
This is an explicit scheme with global boundary conditions. In the following, we consider the stability and convergence of the scheme. Multiplying (3.6) by 2hδ0tuni and summing up forifrom 1 to 2I−1, we obtain
2I−1
i=1
2hδt0uni
δt2uni −δx2uni +uni
=
2I−1
i=1
2hδ0tuni fin. (3.11)
After some calculation, we get
2I−1
i=1
2hδt0uniδt2uni = h τ
2I−1
i=1
δtuni+1/22
−
δtuni−1/22
. (3.12)
Using the summation by parts formula and the boundary condition (3.9)–(3.10), we have
−
2I−1
i=1
2hδ0tuniδ2xuni =h τ
2I−1
i=0
δxuni++11/2δxuni+1/2−h τ
2I−1
i=0
δxuni+1/2δxuni+−11/2+2δxun1/2δt0un0−2δxun2I−1/2δt0un2I
=h τ
2I−1
i=0
δxuni++11/2δxuni+1/2−h τ
2I−1
i=0
δxuni+1/2δxuni+−11/2
+2δ0tun0
δtun0+1/2+ n
m=1
δt0um0
mτ
(m−1)τ
F(nτ−s)ds
+2δ0tun2I
δtun2I+1/2+ n
m=1
δt0um2I
mτ
(m−1)τ
F(nτ−s)ds
≡I1+I2+I3. (3.13)
Notice that, I2=2δ0tun0
δtun0+1/2+ n
m=1
δt0um0
mτ
(m−1)τ
F(nτ−s)ds
=2δ0tun0
δt0un0+ n
m=1
δt0um0
mτ
(m−1)τ
F(nτ−s)ds
+2δt0un0
δtun0+1/2−δt0un0
≡I2A+I2B. Similarly,
I3≡I3A+I3B.
Using the linear interpolation to construct a continuous function u˜(1,t), which satisfies u˜t(1,t)=δt0um0, t∈ [mτ−τ,mτ). According to the stability estimate of the continuous case in Section 2, we obtain
I2A=2δt0un0
δ0tun0+ n
m=1
δ0tum0
mτ
(m−1)τ
F(nτ−s)ds
0. (3.14)
It is easy to check, I2B=2δt0un0
δtun0+1/2−δt0un0
=1 2
δtun0+1/22
−
δtun0−1/22 . By the same technique, we obtain that I3A is also nonnegative, and
I3B=2δt0un2I
δtun2I+1/2−δt0un2I
=12
δtun2I+1/22
−
δtun2I−1/22 . The third term in the left of Eq. (3.11) is easy,
2I−1
i=1
2hδt0uniuni = h τ
2I−1
i=1
uniuni+1− h τ
2I−1
i=1
uni−1uni. (3.15)
We introduce an auxiliary quantity E˜, E˜≡ s
2h
δtun0+1/22
+h
2I−1
i=1
δtuni+1/22
+ s 2h
δtun2I+1/22
+h
2I−1
i=0
δxuni++11/2δxuni+1/2+h
2I−1
i=1 uniuni+1
= s 2h
δtun0+1/22
+h
2I−1
i=1
δtuni+1/22
+ s 2h
δtun2I+1/22
+h
2I−1
i=0
δxuni++11//222
−h 4
2I−1
i=0
δxuni++11/2−δxuni+1/22
+h
2I−1
i=1
uni+1/22
−hτ2 4
2I−1
i=1
δtuni+1/22
=2sh
δtun0+1/22
+
1−s2−τ2 4 h
2I−1
i=1
δtuni+1/22
+2sh
δtun2I+1/22
+s2h
2I−1
i=1
δtuni+1/22
−hs42
2I−1
i=0
δtuni++11/2−δtuni+1/22
+h
2I−1
i=0
δxuni++11//222
+h
2I−1
i=1
uni+1/22
αh 2I
i=0
δtuni+1/22
+h
2I−1
i=0
δxuni++11//222
+h
2I−1
i=1
uni+1/22
>0, (3.16)