JAM Journal of Applied Mathematics 1687-0042 1110-757X Hindawi Publishing Corporation 705179 10.1155/2012/705179 705179 Research Article Nonstandard Finite Difference Variational Integrators for Multisymplectic PDEs Liao Cuicui Ding Xiaohua Zhdanov Renat 1 Department of Mathematics Harbin Institute of Technology 2 Wenhua West Road Shandong Weihai 264209 China hitwh.edu.cn 2012 14 11 2012 2012 12 08 2012 04 10 2012 2012 Copyright © 2012 Cuicui Liao and Xiaohua Ding. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We use the idea of nonstandard finite difference methods to derive the discrete variational integrators for multisymplectic PDEs. We obtain a nonstandard finite difference variational integrator for linear wave equation with a triangle discretization and two nonstandard finite difference variational integrators for the nonlinear Klein-Gordon equation with a triangle discretization and a square discretization, respectively. These methods are naturally multisymplectic. Their discrete multisymplectic structures are presented by the multisymplectic form formulas. The convergence of the discretization schemes is discussed. The effectiveness and efficiency of the proposed methods are verified by the numerical experiments.

1. Introduction

It is a fundamental approach to develop the discrete multisymplectic numerical methods based on the discrete Hamilton’s principle, because it leads in a natural way to multisymplectic integrators . The discrete Euler-Lagrange equation is produced in the discrete variational principle ; meanwhile, the discrete multisymplectic structure is also generated [5, 6]. In the other words, the discrete variational integrators are multisymplectic automatically.

1.1. Multisymplectic Structure of Discrete Variational Integrators

By the Hamilton’s principle , the discrete multisymplectic structure which is preserved by the discrete variational integrator, is described by Poincaré-Cartan forms, in a differential geometric language. In paper , Marsden et al. showed how to obtain this structure directly from the variational principle, on the Lagrangian side. They defined it as the multisymplectic form formula, and they showed that it was conserved by the discrete variational integrator.

Lemma 1.1.

If u is a solution of discrete Euler-Lagrange equation and V, W are first variations of u, then the following discrete multsisymplectic form formula holds: (1.1);U0(l:lU[(j1u)*(ij1Vij1WΩLl)]())=0.

The details of this conclusion could be referred to papers [5, 6]. This conclusion states that the discrete variational principles produce discrete variational integrators, and the multisymplecticity of these variational integrators is presented by the discrete multisymplectic form formula (1.1).

Vankerschaver et al.  revisited the multisymplectic form formula , showing that it could be obtained from the boundary Lagrangian that they defined in their paper. They presented an easy way to derive discrete multisymplectic form formula from discrete variational principle, using the notations of Poincaré-Cartan forms. In this paper, we follow the same way to derive the discrete multisymplectic form formulas of our discrete variational integrators.

When we use the discrete variational principle, we need to make a approximation of the Lagrangian. Here, in our paper, we would use nonstandard finite difference methods, instead of standard finite difference, to approximate the Lagrangian function, and derive the corresponding discrete variational integrators.

1.2. Nonstandard Finite Difference Methods

The nonstandard finite difference schemes are well developed by Mickens  in the past decades. These schemes are developed for compensating the weaknesses that may be caused by standard finite difference methods, for example, the numerical instabilities. Regarding the positivity of solutions, boundedness, and monotonicity of solutions, nonstandard finite difference schemes have a better performance than standard finite difference schemes, due to its flexibility to construct a nonstandard finite difference scheme that can preserve certain properties and structures, which are obeyed by the original equations. Also, the dynamic consistency could be presented well by nonstandard finite difference scheme. These advantages of nonstandard finite difference methods have been shown in many numerical applications. GonzLez-Parra et al.  developed nonstandard finite difference methods to solve population or biological models. The positivity condition and the conservation law of population dynamics are preserved by nonstandard finite difference schemes. Jordan  and Malek  constructed nonstandard finite difference schemes for heat transfer problems. For the symplectic systems, Mickens  derived the nonstandard finite difference variational integrator for symplectic ODEs. Ma et al.  developed the nonstandard finite difference variational integrator in stochastic ordinary differential equations.

The initial foundation of nonstandard finite difference methods is formed by the exact finite difference schemes . After generalizing these results, Mickens summarizes the following three basic rules to construct nonstandard finite difference schemes.

(1) The orders of the discrete derivatives should be equal to the orders of the corresponding derivatives appearing in the differential equations.

In our paper, we combine the advantages of nonstandard finite difference methods and discrete variational principles to construct nonstandard finite difference variational integrators, for two multisymplectic PDEs. These integrators are multysimplectic and their multysimplecticity are presented by their discrete multisymplectic form formulas, respectively.

In Section 2, we consider a simple linear wave equation. With the triangle discretization, we define the discrete Lagrangian using the idea of nonstandard finite difference and derive discrete variational integrator and the corresponding multisymplectic form formula, by discrete variational principle. The convergence of this method is analyzed. In Section 3, for the nonlinear Klein-Gordon equation, triangle discretization and square discretization are considered to obtain the nonstandard finite difference variational integrators. The discrete multisymplectic structures are presented, respectively. The convergence orders of the two methods are also discussed, and the convergence orders are shown in error tables in the numerical experiment section. Section 4 is devoted to showing the numerical behaviors of the developed nonstandard finite difference variational integrators.

2. Nonstandard Finite Difference Variational Integrator for Linear Wave Equation

We first consider a simple linear wave equation, (2.1)utt=uxx, where u(x,t) is a scalar field function with two independent variables, x and t.

This linear wave equation is actually a multisymplectic PDE. As a classical and simple multisymplectic example, wave equation and its multisymplectic structure have been studied from both Hamiltonian  and Lagrangian viewpoints [3, 5, 6, 10]. Based on the Lagrangian viewpoint, we could also obtain (2.1) from the Euler-Lagrange equation (2.2)Lu=ddtLut+ddxLux, with the Lagrangian function L(u,ut,ux), (2.3)L(u,ut,ux)=12ut2-12ux2.

Assume that we have a uniform quadrangular mesh in the base space, with mesh lengths Δx and Δt. The nodes in this mesh are denoted by (i,j)×, corresponding to the points (xi,tj):=(iΔx,jΔt) in 2. We denote the value of the field u at the node (i,j) by uij. We label the triangle at (i,j) with three ordered triple ((i,j),(i+1,j),(i,j+1)) as ij, and we define X to be the set of all such triangles. Then the discrete jet bundle [6, 10] is defined as follows: (2.4)J1Y:={(uij,ui+1j,uij+1)3:((i,j),(i+1,j),(i,j+1))X}, which is equal to X×3.

Now we use nonstandard finite difference to define the discrete Lagrangian Ld on J1Y, which is the discrete version of Lagrangian density [10, 29], (2.5)Ld(uij,ui+1j,uij+1):=12ΔtΔxL(uij+ui+1j+uij+13,uij+1-uijϕ(Δt),ui+1j-uijψ(Δx)), where denominator functions ϕ(Δt) and ψ(Δx) are defined according the exact solution of wave equation [5, 6, 12] (2.6)ϕ(Δt)=12sin(Δt2),ψ(Δx)=12sin(Δx2).

We have followed the rules of constructing nonstandard finite difference schemes in Mickens’ papers .

The discrete first-derivative is represented by (2.7)dudtuij+1-uijϕ(Δt),dudxui+1j-uijψ(Δx), where denominator functions ϕ(Δt), ψ(Δx) are defined in (2.6). Using Taylor series expansion, (2.8)sin(Δt2)=Δt2-148(Δt)3+. Then the denominator functions satisfy (2.9)ϕ(Δt)=Δt+𝒪((Δt)3),ψ(Δx)=Δx+𝒪((Δx)3).

Nonlocal representation on the discrete computational lattice are used here by (2.10)uuij+ui+1j+uij+13.

So, for the linear wave equation (2.1) with the Lagrangian (2.3), the discrete Lagrangian becomes (2.11)Ld(uij,ui+1j,uij+1)=12ΔtΔx(12(uij+1-uijϕ(Δt))2-12(ui+1j-uijψ(Δx))2)  .

By the discrete Hamilton’s principle [6, 10], we have the discrete Euler-Lagrange equation, (2.12)D1Ld(uij,ui+1j,uij+1)+D2Ld(ui-1j,uij,ui-1j+1)+D3Ld(uij-1,ui+1j-1,uij)=0, where Ld(ui-1j,uij,ui-1j+1) and Ld(uij-1,ui+1j-1,uij) are defined similarly as (2.11), which are (2.13)Ld(ui-1j,uij,ui-1j+1)=12ΔtΔx(12(ui-1j+1-ui-1jϕ(Δt))2-12(uij-ui-1jψ(Δx))2)  ,Ld(uij-1,ui+1j-1,uij)=12ΔtΔx(12(uij-uij-1ϕ(Δt))2-12(ui+1j-1-uij-1ψ(Δx))2).

After some simple calculations, the discrete Euler-Lagrange equation (2.12) becomes (2.14)uij+1-2uij+uij-1(ϕ(Δt))2-ui+1j-2uij+ui-1j(ψ(Δx))2=0.

We could find that this scheme is symmetric in (i,j+1) and (i,j-1), (i+1,j) and (i-1,j). This is the nonstandard finite difference variational integrator, for the linear wave equation.

As we mentioned in Section 1 and Lemma 1.1, the advantages of deriving the multisymplectic numerical schemes from discrete variational principle are that they are naturally multisymplectic and the discrete multisymplectic structures are also generated in the variational principle. Now it is meaningful to show the multisymplectic structure of this discrete variational integrator (2.14) based on nonstandard finite difference method.

Since we consider triangulation discretization here, we focus on three adjacent triangles around uij, denote this area by U. Following the idea in , the discrete boundary Lagrangian is given by (2.15)LU(uU):=extuin[Ld(uin,ui+1n,uin+1)+Ld(ui-1n,uin,ui-1n+1)+Ld(uin-1,ui+1n-1,uin)], where (2.16)uU:=(ui+1n,uin+1,ui-1n+1,ui-1n,uin-1,ui+1n-1).

Taking twice exterior derivative of both sides, we have, by the fact that d2LU0, the discrete multisymplectic form formula with following form : (2.17)k=13l=1;lk3ΩLk(Δ(l))=0, where ΩLk=-dΘLk (for k=1,2,3). The discrete Poincaré-Cartan forms ΘL1, ΘL2, and ΘL3 are defined by (2.18)ΘL1(uin,ui+1n,uin+1):=D1Ld(uin,ui+1n,uin+1)duin, and similarly for ΘL2 and ΘL3. Thus, for the linear wave equation (2.1), the multisymplectic form formula of this scheme (2.14) based on nonstandard finite difference method can be obtained as follows: (2.19)duij+1duij+duij-1duij(ϕ(Δt))2-dui+1jduij+dui-1jduij(ψ(Δx))2=0.

Now we have the first conclusion.

Theorem 2.1.

The nonstandard finite difference variational integrator (2.14), (2.20)uij+1-2uij+uij-1(ϕ(Δt))2-ui+1j-2uij+ui-1j(ψ(Δx))2=0, for linear wave equation (2.1) is multisymplectic, and the discrete multisymplectic structure is (2.21)duij+1duij+duij-1duij(ϕ(Δt))2-dui+1jduij+dui-1jduij(ψ(Δx))2=0.

We now discuss the convergence of this variational integrator (2.14) based on the nonstandard finite difference method. From the Lax equivalence theorem we know that, for a well-posed linear initial value problem, the consistent finite difference method is convergent if and only if it is stable.

By Taylor series expansion, we have (2.22)uij+1-2uij+uij-1(ϕ(Δt))2=uij+1-2uij+uij-1(Δt+𝒪((Δt)3))2=1-𝒪((Δt)2)(Δt)2(uij+1-2uij+uij-1)  =1-𝒪((Δt)2)(Δt)2((Δt)2utt(xi,tj)+𝒪(Δt)4)=utt(xi,tj)+𝒪((Δt)2).

Similarly, (2.23)ui+1j-2uij+ui-1j(ψ(Δx))2=uxx(xi,tj)+𝒪((Δx)2).

The above two equations show that the scheme is consistent and the truncation error for the integrator (2.14) is 𝒪((Δt)2+(Δx)2).

To explore the stability of the nonstandard finite difference variational integrator (2.14), we introduce the following notations: (2.24)vij=uij-uij-1ϕ(Δt),wi-1/2j=uij-ui-1jψ(Δt).

Then the three-level explicit integrator (2.14) is equivalent to the following two-level scheme: (2.25)vij+1-vijϕ(Δt)=wi+1/2j-wi-1/2jψ(Δx),wi-1/2j+1-wi-1/2jϕ(Δt)=vij+1-vi-1j+1ψ(Δx).

By using the Von Neumann method , we could get the amplification matrix of the above scheme, (2.26)G(β,Δt)=(12irsin(βΔx)2irsin(βΔx)1-4r2sin2(βΔx)), where r=ϕ(Δt)/ψ(Δx). Note that, in the above matrix, i=-1. Let η=4r2sin2(βΔx). We have the characteristic equation (2.27)λ2-(2-η)λ+1=0 and the eigenvalues (2.28)λ=1-12η±(14η2-η)1/2.

When |λ|1,  that  is,r1, the scheme (2.14) satisfies the Von Neumann conditions, which is a necessary condition of the stability of the scheme (2.14). If r<1, βhnπ, where n is an integer, then G has two different eigenvalues. If r<1, βh=nπ, then G is an identity matrix, but (d/dβh)G has two different eigenvalues . So r<1 is the sufficient condition of the stability for integrator (2.14). Note that, if r=1, there is an unbounded solution vij=(-1)i+j(1-2j), wi+1/2j=(-1)i+j2j. So the scheme (2.14) is not stable when r=1. Now, we find the necessary and sufficient condition of the stability for integrator (2.14), which is (2.29)r=ϕ(Δt)ψ(Δx)<1.

With the consistence and stability conditions, we have following conclusion.

Theorem 2.2.

The nonstandard finite difference variational integrator (2.14) is convergent, when the step sizes Δt and Δx satisfy ϕ(Δt)<ψ(Δx).

We have shown the idea of using the nonstandard finite difference method to get the discrete variational integrator and the corresponding discrete multisymplectic form formula. In the next section, we will consider the discrete variational integrators for a more complicated example, the nonlinear Klein-Gordon equation.

3. Nonstandard Finite Difference Variational Integrators for Nonlinear Klein-Gordon Equation

In this section, we consider the nonlinear Klein-Gordon equation , (3.1)utt=uxx-u3+u.

As known, this equation can be obtained by Euler-Lagrange equation (2.2) with the Lagrangian function (3.2)L(u,ut,ux)=12ut2-12ux2-14u4-12u2.

Now we consider the triangle discretization and square discretization, respectively, to get the nonstandard finite difference variational integrators.

3.1. Triangle Discretization

Following the steps in last section and using the idea of nonstandard finite difference, we define the discrete Lagrangian Ld as (3.3)Ld(uij,ui+1j,uij+1)  =12ΔtΔx(12(uij+1-uijϕ(Δt))2-12(ui+1j-uijψ(Δx))2-14(a1(uij4+ui+1j4+uij+14)+b1(uij2ui+1juij+1+ui+1j2uijuij+1+uij+12uijui+1j)3a1+3b1)+12(a2(uij2+ui+1j2+uij+12)+b2(uijui+1j+ui+1juij+1+uijuij+1)3a2+3b2)), based on the following constructing rules,

The discrete first derivative is represented by (3.4)dudtuij+1-uijϕ(Δt),dudxui+1j-uijψ(Δx), where the denominator functions are defined by (2.6), and (3.5)ϕ(Δt)=Δt+𝒪((Δt)3),ψ(Δx)=Δx+𝒪((Δx)3).

Nonlocal representations for u4 and u2 are given by (3.6)u4a1(uij4+ui+1j4+uij+14)+b1(uij2ui+1juij+1+ui+1j2uijuij+1+uij+12uijui+1j)3a1+3b1,u2a2(uij2+ui+1j2+uij+12)+b2(uijui+1j+ui+1juij+1+uijuij+1)3a2+3b2, where a1, b1, a2, and b2 are positive parameters. Such discretizations for u4 and u2 guarantee the symmetric property of the discrete Lagrangian function .

Similarly, we define discrete Lagrangians on other two adjoint triangles, (3.7)Ld(ui-1j,uij,ui-1j+1)=12ΔtΔx(12(ui-1j+1-ui-1jϕ(Δt))2-12(uij-ui-1jψ(Δx))2-14(a1(uij4+ui-1j4+ui-1j+14)+b1(ui-1j2uijui-1j+1+uij2ui-1jui-1j+1+ui-1j+12ui-1juij)3a1+3b1)+12(a2(ui-1j2+uij2+ui-1j+12)+b2(ui-1juij+uijui-1j+1+ui-1j+1ui-1j)3a2+3b2)),Ld(uij-1,ui+1j-1,uij)=12ΔtΔx(12(uij-uij-1ϕ(Δt))2-12(ui+1j-1-uij-1ψ(Δx))2-14(a1(uij-14+ui+1j-14+uij4)+b1(uij-12ui+1j-1uij+ui+1j-12uijuij-1+uij2uij-1ui+1j-1)3a1+3b1)+12(a2(uij-12+ui+1j-12+uij2)+b2(uij-1ui+1j-1+ui+1j-1uij+uijuij-1)3a2+3b2)).

Now, the discrete variational integrator with nonstandard finite difference methods could be obtained by discrete Euler-Lagrange equation (2.12): (3.8)D1Ld(uij,ui+1j,uij+1)+D2Ld(ui-1j,uij,ui-1j+1)+D3Ld(uij-1,ui+1j-1,uij)=0.

Substituting Ld(uij,ui+1j,uij+1), Ld(ui-1j,uij,ui-1j+1), and Ld(uij-1,ui+1j-1,uij) into above equation, we arrive at (3.9)-uij+1-2uij+uij-1(ϕ(Δt))2+ui+1j-2uij+ui-1j(ψ(Δx))2-1413a1+3b1[12a1uij3+2b1(uijui+1juij+1+uijui-1jui-1j+1+uijuij-1ui+1j-1)  +b1(ui+1j2uij+1+uij+12ui+1j+ui-1j2ui-1j+1+ui-1j+12ui-1j+uij-12ui+1j-1+ui+1j-12uij-1)]+1213a2+3b2[6a2uij+b2(ui+1j+uij+1+ui-1j+ui-1j+1+ui+1j-1+uij-1)]=0.

Using the definition of discrete Lagrangian functions, one can find that this scheme is symmetric with respect to (i,j+1) and (i,j-1), (i+1,j) and (i-1,j); that it is multisymplectic, and that it preserves the multisymplectic structure of the original equation.

Its corresponding discrete multisymplectic form formula can be obtained from (2.17), that is, (3.10)-duij+1duij-duijduij-1(ϕ(Δt))2+dui+1jduij-duijdui-1j(ψ(Δx))2-((C1uijuij+1+C1ui+1juij+1+12C1uij+12-C2)dui+1jduij+(C1uijui+1j+12C1ui+1j2+C1uij+1ui+1j-C2)duij+1duij+(C1uijui-1j+1+C1ui-1jui-1j+1+12C1ui-1j+12-C2)dui-1jduij+(C1uijui-1j+12C1ui-1j2+C1ui-1j+1ui-1j-C2)dui-1j+1duij+(C1uijui+1j-1+C1uij-1ui+1j-1+12C1ui+1j-12-C2)duij-1duij+(C1uijuij-1+12C1uij-12+C1ui+1j-1uij-1-C2)dui+1j-1duij)=0, where (3.11)C1=b12(3a1+3b1),C2=b23a2+3b2.

It shows the multisymplectic structure of scheme (3.9), and the relations between the field values on the three adjoint triangles are around uij.

We now analyze the truncation error of integrator (3.9). By Taylor series expansion [32, 33], we have (3.12)1413a1+3b1[12a1uij3+2b1(uijui+1juij+1+uijui-1jui-1j+1+uijuij-1ui+1j-1)  +b1(ui+1j2uij+1+uij+12ui+1j+ui-1j2ui-1j+1+ui-1j+12ui-1j+uij-12ui+1j-1+ui+1j-12uij-1)]=u3(xi,tj)+𝒪((Δx)2+ΔxΔt+(Δt)2),1213a2+3b2[6a2uij+b2(ui+1j+uij+1+ui-1j+ui-1j+1+ui+1j-1+uij-1)]=u(xi,tj)+𝒪((Δx)2+(Δt)2).

Combining the above two equations and (2.22), (2.23), we can observe that the nonstandard finite difference variational integrator (3.9) has the truncation error 𝒪((Δx)2+ΔxΔt+(Δt)2).

The above results are summarized in the following theorem.

Theorem 3.1.

The nonstandard finite difference variational integrator (3.9) for the nonlinear Klein-Gordon equation (3.1) is multisymplectic, and its truncation error is 𝒪((Δx)2+ΔxΔt+(Δt)2). The discrete multisymplectic structure of this scheme is presented by (3.10).

3.2. Square Discretization

In this case, we denote a square at (i,j) with four ordered quaternion ((i,j),(i+1,j),(i+1,j+1),and(i,j+1)) by ij and define X to be the set of all such squares. Then the discrete jet bundle [6, 10] is defined as (3.13)J1Y:={(uij,ui+1j,ui+1j+1,uij+1)4:((i,j),(i+1,j),(i+1,j+1),(i,j+1))X}, which is equal to X×4.

Following the philosophy of the nonstandard finite difference method, we define the discrete Lagrangian Ld on J1Y as (3.14)Ld(uij,ui+1j,ui+1j+1,uij+1)  =(12(uij+1-uij2ϕ(Δt)+ui+1j+1-ui+1j2ϕ(Δt))2-12(ui+1j+1-uij+12ψ(Δx)+ui+1j-uij2ψ(Δx))2-14uijui+1jui+1j+1uij+1        +12(uijui+1j+uijui+1j+1+uijuij+1+ui+1jui+1j+1+ui+1j+1uij+1+ui+1j+1uij+16))ΔtΔx.

In this case,

the discrete first-derivative is represented by (3.15)dudtuij+1-uij2ϕ(Δt)+ui+1j+1-ui+1j2ϕ(Δt),dudxui+1j+1-uij+12ψ(Δx)+ui+1j-uij2ψ(Δx), where the denominator functions are defined by (2.6), and (3.16)ϕ(Δt)=Δt+𝒪((Δt)3),ψ(Δx)=Δx+𝒪((Δx)3);

nonlocal representations for u4 and u2 are (3.17)u4uijui+1jui+1j+1uij+1,u2uijui+1j+uijui+1j+1+uijuij+1+ui+1jui+1j+1+ui+1j+1uij+1+ui+1j+1uij+16.

Similarly, we have the definitions of Ld on the other three squares adjoint to uij: (3.18)Ld(ui-1j,uij,uij+1,ui-1j+1)  =(12(uij+1-uij2ϕ(Δt)+ui-1j+1-ui-1j2ϕ(Δt))2-12(uij-ui-1j2ψ(Δx)+uij+1-ui-1j+12ψ(Δx))2-14ui-1juijuij+1ui-1j+1  +12(ui-1juij+ui-1juij+1+ui-1jui-1j+1+uijuij+1+uijui-1j+1+uij+1ui-1j+16))ΔtΔx,Ld(ui-1j-1,uij-1,uij,ui-1j)  =(12(ui-1j-ui-1j-12ϕ(Δt)+uij-uij-12ϕ(Δt))2-12(uij-ui-1j2ψ(Δx)+uij-1-ui-1j-12ψ(Δx))2-14ui-1j-1uij-1uijui-1j    +12(ui-1j-1uij-1+ui-1j-1uij+ui-1j-1ui-1j+uij-1uij+uij-1ui-1j+uijui-1j6))ΔtΔx,Ld(uij-1,ui+1j-1,ui+1j,uij)  =(12(uij-uij-12ϕ(Δt)+ui+1j-ui+1j-12ϕ(Δt))2-12(ui+1j-uij2ψ(Δx)+ui+1j-1-uij-12ψ(Δx))2-14uij-1ui+1j-1ui+1juij  +12(uij-1ui+1j-1+uij-1ui+1j+uij-1uij+ui+1j-1ui+1j+ui+1j-1uij+ui+1juij6))ΔtΔx.

Taking derivate of action functional with respect to uij, we have the discrete Euler-Lagrange equation in this square discretization [5, 6, 10, 34], which is (3.19)D1Ld(uij,ui+1j,ui+1j+1,uij+1)+D2Ld(ui-1j,uij,uij+1,ui-1j+1)+D3Ld(ui-1j-1,uij-1,uij,ui-1j)+D4Ld(uij-1,ui+1j-1,ui+1j,uij)=0.

Substituting the discrete Lagrangian Ld(uij,ui+1j,ui+1j+1,uij+1), Ld(ui-1j,uij,uij+1,ui-1j+1), Ld(ui-1j-1,uij-1,uij,ui-1j), and Ld(uij-1,ui+1j-1,ui+1j,uij) into the previous equation, we arrive at (3.20)(-12ϕ(Δt)2-12ψ(Δx)2+16)(uij+1+uij-1)+(12ϕ(Δt)2+12ψ(Δx)2+16)(ui+1j+ui-1j)+(1ϕ(Δt)2-1ψ(Δx)2)uij+(-14ϕ(Δt)2+14ψ(Δx)2+112)(ui+1j+1+ui-1j+1+ui-1j-1+ui+1j-1)-14(ui+1jui+1j+1uij+1+ui-1juij+1ui-1j+1+ui-1j-1uij-1ui-1j+uij-1ui+1j-1ui+1j)=0.

After simple calculations, it becomes (3.21)14(ui+1j+1-2ui+1j+ui+1j-1(ϕ(Δt))2+2uij+1-2uij+uij-1(ϕ(Δt))2+ui-1j+1-2ui-1j+ui-1j-1(ϕ(Δt))2)-14(ui+1j+1-2uij+1+ui-1j+1(ψ(Δx))2+2ui+1j-2uij+ui-1j(ψ(Δx))2+ui+1j-1-2uij-1+ui-1j-1(ψ(Δx))2)+14(ui+1jui+1j+1uij+1+ui-1juij+1ui-1j+1+ui-1j-1uij-1ui-1j+uij-1ui+1j-1ui+1j)  -112(ui+1j+1+ui-1j+1+ui-1j-1+ui+1j-1+2uij+1+2uij-1+2ui+1j+2ui-1j)=0.

It is multisymplectic and symmetric in (i,j+1) and (i,j-1), (i+1,j) and (i-1,j). Similarly, we have the discrete multisymplectic form formula: (3.22)(-12(ϕ(Δt))2-12(ψ(Δx))2+16-14ui+1jui+1j+1-14ui-1jui-1j+1)duij+1duij+(-12(ϕ(Δt))2-12(ψ(Δx))2+16-14ui-1j-1uij+1-14ui+1j-1ui+1j)duij-1duij+(12(ϕ(Δt))2+12(ψ(Δx))2+16-14ui+1j+1uij+1-14uij-1ui+1j-1)dui+1jduij+(12(ϕ(Δt))2+12(ψ(Δx))2+16-14uij+1ui-1j+1-14ui-1j-1uij-1)duij-1duij+(-14(ϕ(Δt))2+14(ψ(Δx))2+112-14ui+1juij+1)dui+1j+1duij+(-14(ϕ(Δt))2+14(ψ(Δx))2+112-14ui-1juij+1)dui-1j+1duij+(-14(ϕ(Δt))2+14(ψ(Δx))2+112-14uij-1ui-1j)dui-1j-1duij+(-14(ϕ(Δt))2+14(ψ(Δx))2+112-14uij-1ui+1j+1)dui+1j-1duij=0.

To study the truncation error of the integrator, we do the Taylor expansion which leads to (3.23)upj+1-2upj+upj-1(ϕ(Δt))2=utt(xp,tj)+𝒪((Δt)2),p=i-1,i,i+1,ui+1q-2uiq+ui-1q(ψ(Δx))2=uxx(xi,tq)+𝒪((Δx)2),q=j-1,j,j+1,14(ui+1jui+1j+1uij+1+ui-1juij+1ui-1j+1+ui-1j-1uij-1ui-1j+uij-1ui+1j-1ui+1j)=u3(xi,tj)+(Δx)2(2u(xi,tj)uxx(xi,tj)+ux2(xi,tj))+(Δt)2(2u(xi,tj)utt(xi,tj)+ut2(xi,tj))=u3(xi,tj)+𝒪((Δx)2+(Δt)2),112(ui+1j+1+ui-1j+1+ui-1j-1+ui+1j-1+2uij+1+2uij-1+2ui+1j+2ui-1j)  =u(xi,tj)+𝒪((Δx)2+(Δt)2).

Combing these equations, we can readily observe that the nonstandard finite difference variational integrator (3.21) has truncation error 𝒪((Δx)2+(Δt)2). To verify this conclusion, we investigate the numerical convergence order in our numerical experiments. See Section 4.

We summarize our conclusion in the following theorem.

Theorem 3.2.

The nonstandard finite difference variational integrator (3.21) for the nonlinear Klein-Gordon equation (3.1) is multisymplectic, and its truncation error is 𝒪((Δx)2+(Δt)2). The discrete multisymplectic structure of this scheme is presented by (3.22).

4. Numerical Simulations

In this section, we report the performance of the nonstandard finite difference variational integrator (2.14) for solving linear wave equation (2.1) and the nonstandard finite difference variational integrators (3.21) and (3.9) for the nonlinear Klein-Gordon equation (3.1).

4.1. Linear Wave Equation

For linear wave equation (2.1), we consider the initial conditions (4.1)u(x,0)=sechx,-10<x<10,ut(x,0)=0,-10<x<10, and the periodic boundary conditions (4.2)u(-10,t)=u(10,t),ux(-10,t)=ux(10,t).

The nonstandard finite difference variational integrator (2.14) is an explicit five points scheme. We choose the denominator functions ϕ and ψ in as ϕ(Δt)=2sin(Δt/2) and ψ(Δt)=2sin(Δx/2).

From Figure 1, we can see that the nonstandard finite difference variational integrator (2.14) for the linear wave equation performs very well and the periodicity of the linear wave equation is preserved accurately.

The waveforms of linear wave equation (2.1) by the nonstandard finite difference variational integrator (2.14) (Δt=0.1, Δx=0.2).

4.2. Nonlinear Klein-Gordon Equation

We now consider the nonlinear Klein-Gordon equation (3.1) with the initial condition (4.3)u(x,0)=A(1+cos(2πxL)),ut(x,0)=0, and periodic boundary conditions.

We use nonstandard finite difference variational integrator (3.21) to simulate this problem with amplitude A=5. The nonstandard finite difference variational integrator (2.14) is an implicit nine-points nonstandard finite difference scheme. The denominator functions ϕ and ψ are defined the same as before.

As depicted in Figure 2, the nonstandard finite difference variational integrator (3.21) simulates the wave propagation perfectly at the beginning. After a long time simulation, the integrator still performs very accurate and stable, without showing any blowup. With periodic boundary condition, the wave going out the computational domain shows up in the other direction periodically.

Waveforms of nonlinear Klein-Gordon equation by the nonstandard finite difference variational integrator (3.21) (Δt=0.01, Δx=0.01).

Waveforms at the beginning from t=0 to t=2

Waveforms from t=5 to t=8

4.3. Convergence Order of the Nonlinear Integrators <bold>(<xref ref-type="disp-formula" rid="EEq3.4">3.9</xref>)</bold> and <bold>(<xref ref-type="disp-formula" rid="EEq3.8">3.21</xref>)</bold>

To further investigate the numerical convergence of the proposed schemes, we conduct a series of numerical tests of our nonlinear integrators. In this example, we consider the nonlinear Klein-Gordon equation (3.1) with the initial boundary conditions as follows: (4.4)u(x,0)=2sech(-11-c2x),ut(x,0)=c-21-c2sech(-11-c2(x-ct))tanh(-11-c2(x-ct)),u(-10,t)=2sech(-11-c2(-10-ct)),u(10,t)=2sech(-11-c2(10-ct)), where c=1.2. The exact solution of the problem is (4.5)u(x,t)=2sech(-11-c2(10-ct)).

The nonstandard finite difference integrators (3.9) and (3.21) are applied to simulate the Klein-Gordon equation. In the integrator (3.9), we choose a1=b1=a2=b2=1 here. The l-norm errors at t=1, t=1.5, and t=2 are listed in Tables 1 and 2. The orders in the tables are calculated with the formula [35, 36] (4.6)Orderln(Error(Δx1)/Error(Δx2))  ln(Δx1/Δx2).

The error and convergence orders of (3.9) for Klein-Gordon problem with Δt=Δx.

Mesh size t = 1 t = 1.5 t = 2
Δ x = 0.1
Error 6.3 e - 3 8.9 e - 3 1.07 e - 2
Order

Δ x = 0.05
Error 1.6 e - 3 2.3 e - 3 2.8 e - 3
Order 1.9773 1.9522 1.9341

Δ x = 0.025
Error 4.1645 e - 4 5.8258 e - 4 7.0059 e - 4
Order 1.9419 1.9811 1.9988

Δ x = 0.0125
Error 1.0480 e - 4 1.4622 e - 4 1.7565 e - 4
Order 1.9905 1.9943 1.9959

The error and convergence orders of (3.21) for Klein-Gordon problem with Δt=Δx.

Mesh size t = 1 t = 1.5 t = 2
Δ x = 0.1
Error 1.28 e - 2 1.81 e - 2 2.25 e - 2
Order

Δ x = 0.05
Error 3.3 e - 3 4.6 e - 3 5.7 e - 3
Order 1.9556 1.9763 1.9809

Δ x = 0.025
Error 8.3445 e - 4 1.2 e - 3 1.4 e - 3
Order 1.9836 1.9386 2.0255

Δ x = 0.0125
Error 2.0990 e - 4 2.9458 e - 4 3.6187 e - 4
Order 1.9911 2.0263 1.9984

Overall, it is clear that the error decreases as the mesh size goes to zero, indicating the convergence of our nonlinear integrators (3.9) and (3.21). Moreover, numerical orders clearly exhibit second order convergence when the mesh size decreases with Δt=Δx, which further confirms our theoretical derivation of the truncation errors of the numerical schemes.

In all, the numerical tests verify that the nonstandard finite difference variational integrators that we developed are capable of preserving the characteristics of original equations. They are all accurate, effective, and suitable for solving multisymplectic systems.

5. Conclusion

In this paper, we have considered a linear wave equation and a nonlinear Klein-gordon equation. We have derived the nonstandard finite difference variational integrators and the corresponding multisymplectic form formulas from these two multisymplectic PDEs. We have shown that the nonstandard finite difference methods are flexible in constructing numerical schemes and can be employed to derive multisymiplectic schemes for multisymplectic systems. The convergence of our methods has been discussed. The numerical experiments have shown effectiveness and efficiency of these nonstandard finite difference variational integrators.

Acknowledgments

The authors are grateful to the editor and anonymous reviewers for their careful reading and many constructive suggestions which lead to a great improvement of this paper. This work is supported by the NNSF of China (no. 11271101) and the NNSF of Shandong Province (no. ZR2010AQ021).

Abraham R. Marsden J. E. Foundations of Mechanics 1978 2nd Reading, Mass, USA Benjamin/Cummings Publishing 515141 ZBL0454.62036 Chen J.-B. Variational integrators and the finite element method Applied Mathematics and Computation 2008 196 2 941 958 10.1016/j.amc.2007.07.028 2388747 ZBL1135.65406 Chen J.-B. Qin M.-Z. Scherer R. Multisymplectic and variational integrators International Journal of Pure and Applied Mathematics 2008 44 4 509 536 2418209 ZBL1147.65103 Ciarlet P. G. Iserles A. Kohn R. V. Wright M. H. Simulating Hamltonian Dynamics 2004 New York, NY, USA Cambridge University Press Cambridge Monographs on Applied and Computational Mathematics Marsden J. E. West M. Discrete mechanics and variational integrators Acta Numerica 2001 10 357 514 10.1017/S096249290100006X 2009697 ZBL1123.37327 Marsden J. E. Patrick G. W. Shkoller S. Multisymplectic geometry, variational integrators, and nonlinear PDEs Communications in Mathematical Physics 1998 199 2 351 395 10.1007/s002200050505 1666871 ZBL0951.70002 Leok M. Zhang J. Discrete Hamiltonian variational integrators IMA Journal of Numerical Analysis 2011 31 4 1497 1532 10.1093/imanum/drq027 2846764 ZBL1232.65177 Nawafleh K. I. Canonical quantization of higher-order Lagrangians Journal of Applied Mathematics 2011 2011 11 375838 10.1155/2011/375838 2844122 ZBL1226.81098 Pekarsky S. Marsden J. E. Abstract mechanical connection and abelian reconstruction for almost Kähler manifolds Journal of Applied Mathematics 2001 1 1 1 28 10.1155/S1110757X01000043 1844945 ZBL0998.53055 Vankerschaver J. Liao C. Leok M. Generating functionals and Lagrangian PDEs Journal of Nonlinear Science. In press 10.1080/00207161003743391 Mickens R. E. Applications of Nonstandard Finite Difference Schemes 2000 Singapore World Scientific Publishing 10.1142/9789812813251 1831401 Mickens R. E. Nonstandard finite difference schemes for differential equations Journal of Difference Equations and Applications 2002 8 9 823 847 10.1080/1023619021000000807 1919887 ZBL1010.65032 Mickens R. E. A nonstandard finite difference scheme for the diffusionless Burgers equation with logistic reaction Mathematics and Computers in Simulation 2003 62 1-2 117 124 10.1016/S0378-4754(02)00180-5 1983580 ZBL1015.65036 Mickens R. E. Dynamic consistency: a fundamental principle for constructing nonstandard finite difference schemes for differential equations Journal of Difference Equations and Applications 2005 11 7 645 653 10.1080/10236190412331334527 2173250 ZBL1073.65552 Mickens R. E. A numerical integration technique for conservative oscillators combining nonstandard finite-difference methods with a Hamilton's principle Journal of Sound and Vibration 2005 285 1-2 477 482 10.1016/j.jsv.2004.09.027 2213110 ZBL1237.70009 Arenas A. J. González-Parra G. Chen-Charpentier B. M. A nonstandard numerical scheme of predictor-corrector type for epidemic models Computers & Mathematics with Applications 2010 59 12 3740 3749 10.1016/j.camwa.2010.04.006 2651849 ZBL1198.65116 González-Parra G. Arenas A. J. Chen-Charpentier B. M. Combination of nonstandard schemes and Richardson's extrapolation to improve the numerical solution of population models Mathematical and Computer Modelling 2010 52 7-8 1030 1036 10.1016/j.mcm.2010.03.015 2718414 ZBL1205.65015 Villanueva R. J. Arenas A. J. Gonzlez-Parra G. A nonstandard dynamically consistent numerical scheme applied to obesity dynamics Journal of Applied Mathematics 2008 2008 14 640154 10.1155/2008/640154 ZBL1157.92028 Jordan P. M. A nonstandard finite difference scheme for nonlinear heat transfer in a thin finite rod Journal of Difference Equations and Applications 2003 9 11 1015 1021 10.1080/1023619031000146922 2027164 ZBL1037.65085 Malek A. Applications of nonstandard finite difference methods to nonlinear heat transfer problems heat transfer-mathematical modelling Numerical Methods and Information Technology In press 10. 5772/14439 Ma Q. Ding D. Ding X. A nonstandard fnite-difference method for a linear oscillator with additive noise Applied Mathematics and Information Science. In press 10.1007/s002200050505 Manning P. M. Margrave G. F. Introduction to non-standard fnite-difference modelling CREWES Research Report 2006 18 Bridges T. J. Multi-symplectic structures and wave propagation Mathematical Proceedings of the Cambridge Philosophical Society 1997 121 1 147 190 10.1017/S0305004196001429 1418367 ZBL0892.35123 Bridges T. J. Reich S. Numerical methods for Hamiltonian PDEs Journal of Physics A 2006 39 19 5287 5320 10.1088/0305-4470/39/19/S02 2220764 ZBL1090.65138 Bridges T. J. Reich S. Multi-symplectic integrators: numerical schemes for Hamiltonian PDEs that conserve symplecticity Physics Letters A 2001 284 4-5 184 193 10.1016/S0375-9601(01)00294-8 1854689 ZBL0984.37104 Hong J. Liu Y. A novel numerical approach to simulating nonlinear Schrödinger equations with varying coefficients Applied Mathematics Letters 2003 16 5 759 765 10.1016/S0893-9659(03)00079-X 1986047 Hong J. Liu Y. Munthe-Kaas H. Zanna A. Globally conservative properties and error estimation of a multi-symplectic scheme for Schrödinger equations with variable coefficients Applied Numerical Mathematics 2006 56 6 814 843 10.1016/j.apnum.2005.06.006 2222122 ZBL1110.65116 Li X.-C. Yao W.-A. Symplectic analytical solutions for the magnetoelectroelastic solids plane problem in rectangular domain Journal of Applied Mathematics 2011 2011 15 165160 10.1155/2011/165160 2773956 ZBL1213.74135 Vankerschaver J. Leok M. A variational, second-order integrator for point vortices on the sphere . In press 10.1080/00207161003743391 Meis T. Marcowitz U. Numerical Solution of Partial Differential Equations 1981 Berlin, Germany Springer 617910 El-Sayed A. M. A. Elsaid A. Hammad D. A reliable treatment of homotopy perturbation method for solving the nonlinear klein-gordon equation of arbitrary (fractional) orders Journal of Applied Mathematics 2012 2012 13 10.1155/2012/581481 581481 ZBL1235.65148 Chen J.-B. A multisymplectic integrator for the periodic nonlinear Schrödinger equation Applied Mathematics and Computation 2005 170 2 1394 1417 10.1016/j.amc.2005.01.031 2175292 ZBL1103.65130 Chen J.-B. Qin M.-Z. Tang Y.-F. Symplectic and multi-symplectic methods for the nonlinear Schrödinger equation Computers & Mathematics with Applications 2002 43 8-9 1095 1106 10.1016/S0898-1221(02)80015-3 1892487 ZBL1050.65127 Leok M. Shingel T. General techniques for constructing variational integrators Frontiers of Mathematics in China 2012 7 2 273 303 10.1007/s11464-012-0190-9 2897705 Zhou S. Cheng X. Numerical solution to coupled nonlinear Schrödinger equations on unbounded domains Mathematics and Computers in Simulation 2010 80 12 2362 2373 10.1016/j.matcom.2010.05.019 2721184 ZBL1197.65162 Zhou S. Cheng X. A linearly semi-implicit compact scheme for the Burgers-Huxley equation International Journal of Computer Mathematics 2011 88 4 795 804 10.1080/00207161003743391 2774513 ZBL1213.65122