The improved Boussinesq equation is solved with classical finite element method using the most basic Lagrange element k = 1, which leads us to a second order nonlinear ordinary differential equations system in time; this can be solved by any standard accurate numerical method for example Runge-Kutta-Fehlberg. The technique is validated with a typical example and a fourth order convergence in space is confirmed; the 1- and 2-soliton solutions are used to simulate wave travel, wave splitting and interaction; solution blow up is described graphically. The computer symbolic system MathLab is quite used for numerical simulation in this paper; the known results in the bibliography are confirmed.

Boussinesq Equation Soliton Finite Element Method Galerkin Method
1. Introduction

The improved Boussinesq equation (IBq) was proposed in Bogolyubsky’s work  , like a correct modification to solve the bad Boussinesq equation (BBq) which describes a large group of nonlinear dispersive wave phenomena, such as propagation of long waves on the surface of shallow water in both directions, propagation of ion-sound waves in a uniform isotropic plasma, and so on  . Bogolyubsky has also shown that the BBq equation describes an unphysical instability of short wave lengths and the Cauchy problem for this partial differential equation is incorrect. The BBq equation was first introduced in the 1870s by Joseph Boussinesq  , which is given by

where is a sufficiently differentiable real function, the correct modification to this partial dif- ferential equation is given by

which is the IBq and will be the principal study equation of this paper; it is convenient for computer simulation of the dynamics of different nonlinear waves with weak dispersion; in our case the IBq equation will help to formulate the finite element discretization in the spatial direction with the primal L2-Galerkin finite element formulation   ; this due to the correction in the fourth order derivative term which now leads us to the integral of a discontinuous function over a set of measure zero (for the Lagrange finite element). The IBq (2) has the 1-soliton solution  .

where, is the wave amplitude, is the wave speed and is the soliton center of symmetry. The initial displacement and velocity condition to the (2) equation are assumed to have the form.

where and are given functions in each example.

The boundary conditions at are assumed to be

Linearization techniques and finite differences are employed in most numerical works that solve the IBq  -  ; they need a relevant stability relation between the space and time discretization size, obtained by the Fourier method of analyzing stability and the Von Neumann’s necessary criterion for stability   ; in

contrast with the method proposed in this paper such a restriction is not needed. The nonlinear term

which for the finite difference method is a problem and needs to be linearized with the help of bounds solutions and/or iterative approach  is not a problem in our work which is treated formally by the L2-Galerkin finite element formulation and leads us due to the reduced support in the basis functions to a time dependent tridia- gonal antisymmetric square matrix for the Lagrange element case, so the only linearization is inherent to the finite element method; in this way the following Lagrange elements are expected to work with

better convergence properties in the x direction. The convergence in the proposed method is verified with the standard procedure  using -norm. Wave propagation, wave break up, inelastic and elastic

head-on collision, and blow-up solution are modeled and graphics representations are done  .

2. The Classical Finite Element Method

The classical finite element method relies over two basic ingredients   , the first is a weak or variational formulation for the IBq equation which is obtained for a fixed t by multiplying with a test function

where is the standard Sobolev space  defined by

the subindex and superindex 0, 1 refers to boundary conditions and to the derivative order that should belong to respectively, and integrate over to get after integrating by parts and applying the boundary conditions (5), the variational formulation for the IBq equation.

Find such that for

A classical (or conforming) approximation of u is obtained by looking for a function with such that

The second basic ingredient for the classical finite element method is to choose the finite dimensional subspace, in our case will be constructed with the finite element from the Lagrange family   , to

this end let, be a partition of the interval into subintervals of length, in our work a non adaptive mesh will be considered so

, we now let to be the set of functions v such that v is linear on each subinterval

is continuos on and As parameters to describe a function we may choose the values at the node points. Let us introduce the linear basis functions, defined by 

In this way for each a function has the unique representation

as a linear combination of the basis functions, and is a linear vector space of dimension with basis. The variational problem (9) is equivalent to the follow- ing L2-Galerkin space semi-discretization for the IBq equation. Find such that

If we substitute and take in turn in (10) we will obtain a second order in time nonlinear ordinary differential equations system to aproximate which in matrix notation is

where, are N by N matrices which will be calculated in the next section and as before.

3. The Finite Element Computational Aspects

As is usual all finite element computations like integration, interpolation are done over the master element over the following local basis functions are needed to integrate,

they have the property and see Figure 1.

The local finite element matrices are calculated over with the help of functions in (12) then multiplied

by the respective scale factor, to get the finite element matrix over, then we need to do the

Local basis functions associated to the points -1, +1

typical finite element assembly to get the global matrices. For instance the matrix M, using the scale factor is constructed from in the following way

transforms to

and after assembly from element 2 to N, M is given as follows

over and the basis functions should satisfy the respective boundary conditions (5), assembling these components to the last M we finally arrived to

analogously for K whose scale factor for integration is,

the matrix follows the same steps with scale factor

finally after assembly and putting the boundary conditions

this matrix represents the nonlinearity in the IBq Equation (2), the anti-symmetry structure is related to the Lagrange finite element and to the primal variational formulation and not to nonlinearity.

4. The Initial Value Problem

With the matrices and at hand it is possible to solve the nonlinear initial value problem (11), to get a unique solution it is necessary to impose the initial conditions (4), the system (11) is equivalent to

the matrix is a tridiagonal positive definite and therefore invertible  , as is usual with second order systems, one should introduce a new vector variable in this way the system (13) is equivalent to

the next first order nonlinear system of ordinary differential equations:

with initial conditions

the system (14) and (15), (16) is a standard initial value problem that can now be solved by integration algorithms like predictor corrector   and not by simply fourth order Runge-Kutta method. This paper will employ Runge-Kutta-Fehlberg of fourth and fifth order with variable time step size, the fifth order method will work like a predictor and the fourth order like a corrector  .

5. Numerical Examples

Firstly in Numerical Validation, the proposed method is used for the numerical wave propagation simulation, and comparing this simulation with the exact solution we validate the method, we are really approximating the soliton solution by a non-classical one, the compacton  .

1. Numerical Validation. We set and

the exact solution is given by (3), we discretize over and, the numerical results

are compared for t = 20 with the exact solution at some points in Table 1, where means the numerical solution, and the wave propagation numerical graphic is illustrated by Figure 2 and Figure 3 where the level curves are showed.

2. Wave brake-up. With the same as in Numerical Validation, and we will have an example of wave brake-up propagation, taking Figure 4 is plotted with the numerical solution and Figure 5 shows the level curves.

xError
15.11240.03580.03560.0002
20.05830.29440.29490.0005
22.90590.49900.49880.0002
25.00420.40160.40130.0003
30.09990.05670.05670.0000
35.04580.00500.00500.0000
Level curves for soliton propagation Soliton brake-up Level curves for soliton brake-up

3. The head-on wave collision. In this example we take with and

A negative speed indicate a wave traveling to the negative x side direction, so the two waves will have a head-on collision  . We obtain an inelastic collision, the Figure 6 shows the collision intercourse and Figure 7 the level curves where secondary solitons are visible, hence the collision is inelastic.

The next examples are done with different amplitudes.

If, the collision is inelastic.

If, the collision is inelastic.

If, the collision is elastic.

If, the collision is still elastic. The Figure 8 and Figure 9 show this case.

These results are in good agreement with those reported elsewhere   .

4. Blow-up solution. The blow-up solution is now simulated as discussed in   , the IBq (2) is solved

numerically on with 200 finite elements in the x direction and with initial conditions given by

It is know  the existence of such that exist unique local solution with as by the left side and as by the left side, Figure 10 and Figure 11 show the numerical results for some fixed times between and.

5. Convergence Order. For our technique, the convergence order will be calculated in the usual way using the results from Numerical Validation, as the following Table 2 shows the rate of convergence for Lagrange k = 1 finite element is in space.

6. Conclusion

A concrete development of a practical finite element scheme is used to make a semi discretization in the x direction and reduce the IBq equation to a system of ordinary differential equation with initial value; this de- velopment open the door to try the next Lagrange finite elements to get a better convergence property in the x direction, and the numerical results are highly accurate as Numerical Validation shows. A wave

Number of elementsError in -normC.O.
200.38524837209495
400.058645161585482.715704864
800.003612365237764.020996413