Numerical Solution of Second-Order Linear Fredholm Integro-Differential Equation Using Generalized Minimal Residual Method

Problem statement: This research purposely brought up to solve compli cated equations such as partial differential equations, integral eq uations, Integro-Differential Equations (IDE), stochastic equations and others. Many physical phen om a contain mathematical formulations such integro-differential equations which are arise in f luid dynamics, biological models and chemical kinetics. In fact, several formulations and numeri cal solutions of the linear Fredholm integrodifferential equation of second order currently hav e been proposed. This study presented the numerical solution of the linear Fredholm integro-differentia l equation of second order discretized by using fin ite difference and trapezoidal methods. Approach: The linear Fredholm integro-differential equation f second order will be discretized by using finite di fference and trapezoidal methods in order to derive an approximation equation. Later this approximation equation will be used to generate a dense linear system and solved by using the Generalized Minimal Residual (GMRES) method. Results: Several numerical experiments were conducted to examine the fficiency of GMRES method for solving linear system generated from the discretization of linear Fredholm integro-differential equation. For the comparison purpose, there are three parameters such a number of iterations, computational time and absolute error will be considered. Based on observa tion of numerical results, it can be seen that the number of iterations and computational time of GMRE S have declined much faster than Gauss-Seidel (GS) method. Conclusion: The efficiency of GMRES based on the proposed disc retization is superior as compared to GS iterative method.


INTRODUCTION
Integro-Differential Equation (IDE) is an important branch of modern mathematics and arises frequently in many applied areas which include engineering, mechanics, physics, chemistry, astronomy, biology, economics, potential theory and electrostatics (Kurt and Sezer, 2008). IDE is an equation that the unknown function appears under the sign of integration and it also contains the derivatives of the unknown function. It can be classified into Fredholm equations and Volterra equations. The upper bound of the region for integral part of Volterra type is variable, while it is a fixed number for that of Fredholm type. In this study, we focus on second order linear Fredholm integrodifferential equation. Generally, second-order linear Fredholm integro-differential equations can be defined as follows: with initial conditions: Where, the functions: p(x), q(x), r(x) = Constant matrices f(x) = A given vector function, the kernel k(x,t) = A given matrix function y(x) = The solution to be determined In the engineering field, numerical approaches were practiced to obtain an approximation solution for the problem (1). To solve a linear integro-differential equation numerically, discretization of integral equation to the solution of system of linear algebraic equations is the basic concept used by researchers to solve integro-differential problems. By considering numerical techniques, there are many methods can be used to discretize problem (1) such as compact finite difference (Zhao and Corless, 2006), Wavelet-Galerkin (Avudainayagam and Vani, 2000), variational iteration method (Sweilam, 2007) rationalized Haar functions (Maleknejad et al., 2004), Tau, (Hosseini and Shahmorad, 2003), Lagrange interpolation (Rashed, 2003), piecewise approximate solution (Hosseini and Shahmorad, 2005), conjugate gradient (Khosla and Rubin, 1981), quadraturedifference (Fedotov, 2009), variational (Saad and Schultz, 1986), collocation (Aguilar and Brunner, 1988), homotopy perturbation (Yildirim, 2008) and Euler-Chebyshev method (Van der Houwen and Sommeijer, 1997). Earlier numerical treatment has been done for first order integro-differential equation (Aruchunan and Sulaiman, 2009).
In this conjunction, there are many iterative methods under the category of Krylov subspaces have been proposed widely to be one of the feasible and successful classes of numerical algorithms for solving linear systems. Actually, there are several Krylov subspaces iterative methods can be considered such as Conjugate Gradient (CG) (Hestenes and Stiefel, 1952), Generalized Minimal Residual (GMRES) (Saad and Schultz, 1986), Conjugate Gradient Squared (Sonneveld, 1989), Bi-Conjugate Gradient Stabilized (Bi-CGSTAB) (Van der Vorst, 1992) and Orthogonal Minimization (ORTHOMIN) (Vinsome, 1976).
In this study, GMRES iterative method will be used for solving linear algebraic equations produced by the discretization of the second-order linear Fredholm integro-differential equations by using quadrature and finite difference methods. For differential part, second order central difference scheme was used for approximation whereas the integral term was discretized by quadrature method. In order to compare the efficiency of the GMRES method, Gauss-Seidel (GS) method was used for numerical comparison.

MATERIALS AND METHODS
Approximation equation: As afore-mentioned, a discretization method based on quadrature and finite difference methods was used to construct approximation equations for problem (1).

Quadrature method:
The formulas of quadrature method, in general have the form: where, j t ( j 0,1, , n) = … are the abscissas of the partition points of the integration interval [a,b] or quadrature (interpolation) nodes, A j (j = 0,1,…,n) are numerical coefficients that do not depend on the function y(t) and ε n (y) is the truncation error of Eq. 2. To facilitate in formulating the approximation equations for problem (1), further discussion will restrict onto Repeated Trapezoidal (RT) method, which is based on linear interpolation formulas with equally spaced data. Based on RT method, numerical coefficients A j are satisfied the following relation: j 1 h, j 0, n A 2 h, j 1,2, , n 1 where, the constant step size, h is defined as: In this study, interval [a,b] will be uniformly divided into m n 2 , m 2 = ≥ and then consider the discrete set of points be given as i x a ih = + .

Generalized Minimal Residual (GMRES) method:
The GMRES is an efficient algorithm for iteratively solving a general linear system in Eq. 4. The method is based upon the Arnoldi process (Saad and Schultz, 1986), which constructs an orthonormal basis of the Krylov subspace m algorithm may be described in Algorithm 1 (Saad, 2003).

;
Step 2 For j = 1, 2… until convergence Step 3 Step 4 If i 1 p + ≤ ε then End for i Step 5 If i 1 p + ≤ ε end for j else 0 x x =

RESULTS AND DISCUSSION
In this conjunction, the GMRES method was tested with the following problem (Hosseini and Shahmorad, 2005) and then compared its performances with GS method: ( ) In this research, parameters such as number of iterations, execution time and absolute error are considered as comparison. Throughout the simulations, the convergence test considered the tolerance error of the ε = 10 −16 . Figure 1 and 2 show number of iterations and execution time versus mesh size, respectively. Furthermore, the numerical results of this application presented with exact solution have been shown in Table 1.

CONCLUSION
Based on the results in Table 1, number of iterations of the GMRES methods has decreased approximately 44.00-47.82% compared to GS method as shown in Fig. 1. For the execution time, GMRES method is much faster about 39.17-45.85% compared to GS method, see in Fig. 2. As a conclusion, the numerical results have shown that the GMRES method is more advanced in term of number of iterations and the execution time than GS method.