Finite Element Method of BBM-Burgers Equation with Dissipative Term Based on Adaptive Moving Mesh

A finite elementmodel is proposed for the Benjamin-Bona-Mahony-Burgers (BBM-Burgers) equationwith a high-order dissipative term; the scheme is based on adaptivemovingmeshes.Themodel can be applied to the equationswith spatial-timemixedderivatives and high-order derivative terms. In this scheme, new variables are needed to make the equation become a coupled system, and then the linear finite element method is used to discretize the spatial derivative and the fifth-order Radau IIA method is used to discretize the time derivative.The simulations of 1D and 2DBBM-Burgers equationswith high-order dissipative terms are presented in numerical examples. The numerical results show that the method keeps a second-order convergence in space and provides a smaller error than that based on the fixed mesh, which demonstrates the effectiveness and feasibility of the finite element method based on the moving mesh.We also study the effect of the dissipative terms with different coefficients in the equation; by numerical simulations, we find that the dissipative term uxx plays a more important role than uxxxx in dissipation.


Introduction
In mathematics, physics, and engineering, nonlinear partial differential equations play an important role due to their rich mathematical structures and features [1][2][3][4] as well as important applications in fluid dynamics, plasma physics, fiber optics, condensed matter physics, and chemistry [5][6][7][8][9][10][11]. It is important to search for solutions. These nonlinear partially differential equations can be solved exactly or numerically to understand the properties of solutions.
In this paper, we consider the nonlinear multidimensional Benjamin-Bona-Mahony-Burgers (BBM-Burgers) equation with a dissipative term and use the finite element method based on the moving mesh to solve its numerical solution. The equation has the following form: 2 +2 = , X ∈ Ω, ∈ (0, ] (X, ) = (X, ) , X ∈ Ω, ∈ (0, ] (X, 0) = 0 (X) , X ∈ Ω, where X = ( 1 , 2 , . . . , ), stands for the dimension, stands for the corresponding spatial vector, the coefficients , ∈ , , , > 0, Ω ⊆ , Δ is a Laplace operator, and is one constant to define the order of the dissipative term. In the equation, is a sufficiently smooth function and stands for the source term, and (X) and 0 (X) are given functions.
Equation (1) is related to the well-known Benjamin-Bona-Mahony (BBM) equation, which was advocated by Benjamin et al. [12] in 1972 as a refinement of the Korteweg-de-Vries In physical phenomena, the effect of dispersion in the BBM-Burgers equation with a high-order dissipative term is similar to the RLW equation; the effect of dissipation is similar to the Burgers equation.
To get the approximate solution of the BBM-Burgers equation with a high-order dissipative term, some numerical and analytical methods have been proposed in recent years. In the aspect of mathematical theory, Zhao et al. consider the existence of the generalized BBM-Burgers equation with a dissipative term and verify the corresponding convergence in H-measures [13,14]. Kondo and Webler apply kinetic decomposition for the generalized BBM-Burgers equations with a dissipative term to obtain the approximate transport equation, and then take advantage of the averaging lemma to get the convergence [15]. Seok states the existence and conservation laws of the generalized BBM-Burgers equation with a dissipative term [16]. For more theoretical analyses, refer to the references in [13][14][15][16]. In the aspect of numerical solutions, many methods have been used, like Painlev' e test, Darboux Transformation, bilinear method, symmetry method [17], and so on [18][19][20][21]. Hong et al. apply the homotopy perturbation method and He's variational iteration method to solve the BBM-Burgers equation. Ganji et al. put forward the notion that the method they used in [22] is better than the homotopy analysis method proposed in [23]. Finite element Galerkin methods have been discussed by Kadri et al. [24], Lee [25], Yin and Piao [26], and Dehghan et al. [27] to solve the BBM-Burgers equations with a dissipative term. Finite difference methods have also been widely used to solve the equations in [28][29][30]; as for other numerical methods, refer to [31][32][33][34].
The references introducing the numerical simulations above are mostly based on fixed meshes; in this paper, we consider the finite element method based on the adaptive moving mesh to solve the BBM-Burgers equation with a high-order dissipative term. The moving mesh method is a kind of adaptive mesh method which adds elements near the place where the numerical solution changes rapidly and decreases elements in the solution changing slowly; the total element number still remains unchanged. When we compare the finite element method based on adaptive moving mesh with methods based on fixed mesh or proposed in [26], we find that the moving mesh method takes fewer elements to get the same distinguishability and simulates steep waves and the transition of gap distinctly. In recent years, the moving mesh method has witnessed further development and extensive application [35]. This method was applied to approximate the anisotropic diffusion-convection-reaction problems to satisfy a discrete maximum principle, and it was adopted to study the porous medium equation and approximate the solution [36]. For more information about the moving mesh method, please refer to [37][38][39][40][41].
The remainder of this paper is divided into three sections. In Section 2, we introduce the method we used, the discretization of the BBM-Burgers equation with a fourth-order dissipative term in one dimension, and the implementation of the moving mesh method. Section 3 presents 1D and 2D numerical examples of BBM-Burgers equations with a dissipative term. Finally, Section 4 shows the conclusion we draw from the numerical examples.

Finite Element Method Based on Moving Mesh
In this section, we mainly introduce the linear finite element method based on moving mesh and make use of the method to solve the BBM-Burgers equation with a high-order dissipative term. For convenience, in Section 2.2, we introduce the finite element method of the one-dimensional equation with a fourth-order dissipative term ; the other parts of this section will focus on two dimensions.

Adaptive Moving Mesh Method.
In this subsection, we mainly want to show the generation of physical mesh T ℎ +1 at a new time level by an adaptive moving mesh strategy based on the Moving Mesh Partial Differential Equation (MMPDE) [42]; at the current time level, the physical mesh T ℎ = {X } and the numerical solution ( ℎ ) are known (the mesh and value will be introduced clearly in the following parts). During the strategy, one of the crucially important rules is to view the nonuniform mesh (i.e., the physical mesh in this paper) as a uniform mesh under the metric tensor M; the tensor M is a symmetric and positive definite matrix. Under the metric tensor M, the physical mesh T ℎ and the computational mesh T can be transformed into each other if and only if it satisfies (the following notations and concepts about the mesh are the same as those in [36]) where is an element of the physical mesh T ℎ , is an element of the computational mesh T corresponding to , and | | and | |, respectively, show the volume of the corresponding element. Meanwhile, |Ω | = ∑ ∈T | |, det(⋅) denotes the determinant of a matrix, and tr(⋅) stands for the trace of a matrix. ℎ = ∑ ∈T ℎ | | det(M ) 1/2 represents the total error of the numerical solution, is the Jacobian matrix of the affine mapping : → , and is the average of the metric tensor over the elements . In this paper, we take is an identity matrix and (⋅) stands for a Hessian matrix. Equations (4) and (5) also satisfy the equidistribution principle and alignment principle on moving mesh, which means that, under the metric tensor M, each element has the same numerical error and is similar to a reference element̂; the reference element̂is uniform. From (4) and (5), we find that the metric tensor M plays an important role in the adaptive moving mesh and controls the adaptation.
If we want to get the mesh which satisfies the two principles, we need to minimize the energy function that contains the metric tensor M: where and are dimensionless parameters. When we consider the equation to be just one-dimensional, (6) will have a relationship with (4) and = 0; when the equation is twodimensional, parameters , are usually defined as 1/3 and 2, and stands for the dimension in space. The energy function ℎ is also a function about the physical mesh coordinate X = ( , ) and the computational mesh coordinate = ( , ) ( = 0, . . . , ), where + 1 stands for the number of mesh vertices. The coordinate = (X, ) is an inverse transformation of X = X( , ); when = 0 or = , the coordinate is the same as X . The energy function is too long to write; we write it in a simple form: Assume that the current physical mesh T ℎ and ( ℎ ) are known; by minimizing the energy function, we can get the corresponding computational mesh of the next time level, which can be expressed by In order to solve the minimum of energy function in (6), we need to take the derivative of the coordinate with respect to the energy function; that is, It is not difficult to find that the equations are nonlinear and difficult to solve directly; in order to solve this problem, many scholars put forward a large number of methods to improve it. In this paper, we use the MMPDE [42] method to get the minimum value of the energy function (6). For convenience and efficiency, -formulation MMPDE is made use of rather than -formulation MMPDE; during the calculation, -formulation only involves the differentiation of with respect to its first two arguments in the energy function, and there is no need for the metric tensor M to update. The minimization process can be replaced as the derivative of the computational mesh coordinate with respect to , which is an ordinary differential equation and has the following form: where the derivative ℎ / is seen as a row vector, > 0 is a relaxation factor which is used for adjusting the time scale on moving mesh, and = { 0 , . . . , } is a positive function used to make (10) have some desired properties above; in this paper, = 10 −4 , = − det(M(X )) 1/2 .
At the current physical mesh T ℎ ( = 0, . . . , ) and computational solution ( ℎ ) , we define a reference computational meshT = {̂0, . . . ,̂}, which has uniform elements; the meshT also has the same relationship, grid connection, and number of mesh vertices as the physical mesh T ℎ and computational mesh T . The three kinds of meshes just differ in the location of the mesh coordinate. During the mesh generation, by integrating (10) from to +1 , we get the new computational mesh T +1 = { +1 }; there is no need for the metric tensor M to update. The relationship between the physical mesh and the computational mesh can be expressed by an affine mapping: T ℎ = Φ ℎ (T +1 ); the physical mesh in the next time level can take the reference computational mesh by using a form of linear interpolation to get the approximate result: T ℎ +1 = Φ ℎ (T ).

Linear Finite Element Discretization of the 1D BBM-Burgers Equation with a Dissipative Term.
In this subsection, we mainly introduce the linear finite element discretization of the 1D BBM-Burgers equation with a dissipative term. When ( ℎ ) , , and the discretization of the equation are known, we will get the value ( ℎ ) +1 at the new time level. The 1D BBM-Burgers equation with a fourth-order dissipative term has the formulation 4 Discrete Dynamics in Nature and Society Before solving (11) by the finite element method based on moving mesh, we have to introduce a new variable = rather than solving the equation directly; because a mixed derivative of the finite element approximation based on the moving mesh method is unable to be defined, then (11) becomes a coupled system. Before the discretization of the 2D equation with a fourth-order dissipative term, we need to introduce two new different variables to replace and ; the equation then becomes three coupled equations; the other steps in the 2D case are similar to the 1D case. The new formulation of (11) is as follows: We divide the domain Ω into uniform elements and + 1 mesh vertices; the initial physical mesh is T ℎ 0 = { 0 } ( = 0, 1, . . . , ). We define the linear basis function of the finite element method as . On a fixed mesh, the basis function is just related to ; on a moving mesh, it is also related to and . The basis function spans the space ℎ 0 = { 1 ( , ), . . . , −1 ( , )}, ∀ ∈ ℎ 0 . We need to find whether ℎ ( , ) ∈ ℎ 0 and ℎ ( , ) ∈ ℎ 0 with the Dirichlet boundary condition equal to zero can satisfy the following system: [̇̇] ( , , )), so ( ℎ ) can be expressed as Take the basis function = ( = 1, 2, . . . , − 1) and substitute (14) into (13); we can get We rewrite (15) into a matrix form: ( − ) + − = 0, where and stand for = ∫ Ω and = ∫ Ω ( / )( / ) and the vector can be expressed as . On a moving mesh, with the boundary condition, the solution of (11) can be obtained by solving an ordinary differential equation firstly: Only if the ODE (17) is solved can the solution of the BBM-Burgers equation with a dissipative term be expressed, where = − −1 . All the process of discretization is based on the homogeneous boundary condition 0 = ; when the boundary condition is nonhomogeneous, we just need to subtract the integration at the boundary.

The Implementation of the Method Combined with Finite Element and Moving Mesh.
In this subsection, we mainly introduce the implementation of the two-dimensional method. The ways to solve the equation usually can be divided into two kinds: a simultaneous solution procedure and an alternate solution procedure. The simultaneous solution procedure combines the finite element discretization with the mesh equation (10) into a large system and solves them together; the alternate solution procedure solves one by one. In this paper, we introduce a formulation of the second method which is called MP (the method solves the moving equation firstly and solves the physical equation later), which is more flexible and efficient. The method contains four important parts: initialize the mesh, move the mesh, discretize the equation, and integrate time from 0 to .
Discrete Dynamics in Nature and Society 5 Step 1 (initialize the mesh). In the whole computational domain, we first divide the domain into uniform elements; the corresponding initial physical mesh is T ℎ 0 , and the initial coordinate, initial value of the BBM-Burgers equation with a dissipative term, initial time step, and initial time are known; they are X 0 = ( 0 , 0 ), ( ℎ ) 0 , 0 , and 0 .
Step 3. On the new physical mesh T ℎ +1 , use the discrete version (16) based on the linear finite element method to solve (12) from time to +1 ; we get the new numerical solution at the new time level; the solution is ( ℎ ) +1 .
Step 4. When +1 < , ← + 1, go to Step 2; else, stop the computation and save the value of solution. We solve the numerical solution with the method we introduced; with the different element numbers, the corresponding error and the order of convergence for both moving mesh and fixed mesh are reported in Table 1. When the element number is the same, the error based on moving mesh is always smaller than that on fixed mesh, and the convergence order of the two kinds of meshes can lead to the second order in the spatial direction. Comparison between the exact solution and the numerical solution based on moving mesh at the final time = 10 is shown in Figure 1. From the figure, we can see that the grid points mainly focus on the domain, where the solution fluctuates greatly, and match the exact solution well. With time going on, the moving mesh trajectory is plotted in Figure 1; the feature of the moving process indicates that most of the points focus on nearby solitary waves and move with the propagation of the solitary wave, which also demonstrates the superiority of the moving mesh method.  In this example, the exact solution of the equation is unknown; we mainly want to show the superiority of the moving mesh. In Figure 2, the conservation laws (2) are plotted; we can see that Δ 1 ( ) is close to zero and Δ 2 ( ) is a little bigger than zero. The two conservation laws are not conserved by finite element approximation on the moving mesh, because, in this case,Ẋ ≡ 0 and both matrices and are time-dependent (if you want to know more, read [45]). Since the exact solution is unknown, we take the element number = 8000 on the fixed mesh as a reference solution and plot the numerical solutions of moving mesh and fixed mesh with element number = 400 at final time = 50 in Figure 3. It can be seen that the numerical solution of fixed mesh is on the left of the reference solution and the wave peaks are shorter than those in the reference solution; there are several small waves in the left of the first wave on the fixed mesh. The result also shows that the numerical solution of moving mesh can match the reference better than that based on the fixed mesh. If we consider the computer memory and computational efficiency, the moving mesh is better for that it moves its mesh points quickly to the place where the numerical solution changes rapidly to get higher distinguishability with fewer elements.  The propagation process of the three solitary waves is plotted in Figure 4.
In this example, we mainly want to talk about the dissipation effect of the dissipative terms and in the BBM equation. Numerical solutions obtained with different coefficients and one of the coefficients fixed at = 20 are shown in Figure 5. When two BBM equations with one dissipative term or have the same dissipation effect or similar wave peak at the final time, the corresponding coefficient of dissipative term is plotted in Figure 6. From these figures, we find that when the coefficient of the second-order dissipative term increases and the coefficient of the fourth-order dissipative term is fixed, the numerical solution decreases a lot; however, when the coefficient increases and the coefficient is fixed, the numerical solution changes a little. The numerical simulation provides three sets of data to show the result in and then the value drops slowly and steadily. We simulate a series of numerical examples with different coefficients; all of these data show us that the dissipative term plays a more important role in dissipation than the dissipative term .

Conclusion
In this paper, we study the numerical model based on adaptive moving mesh finite element method to simulate 1D and 2D BBM-Burgers equations with a high-order dissipative term. The method can be used in equations with mixed derivatives and high-order derivative terms; on the moving mesh, new variables need to be introduced to transform the equation into coupled equations. A series of numerical examples demonstrate that the finite element method based on moving mesh keeps a second-order convergence in space as that based on fixed mesh. With the same element number, the finite element moving mesh method improves the distinguishability and provides a smaller error than the fixed mesh. Meanwhile, numerical tests indicate that the dissipative term plays a more important role than in dissipation. We compare two BBM equations with different dissipative terms and ; to keep the maximum value max of the numerical solutions the same at the final time, the coefficient of is around 50 times that of in the same BBM equation with different dissipative terms in the dissipation aspect.

Conflicts of Interest
The authors declare that they have no conflicts of interest.