Enhanced moving least square method for the solution of volterra integro-differential equation: an interpolating polynomial

Introduction Integro—differential equations (IDEs) are equations that take into account both integral and derivatives of an unknown function [30]. Mathematical modeling of real-life problems usually results in functional equations like ordinary or partial differential equations, integral and integro—differential equations, and stochastic equations. Many mathematical formulations of physical phenomena contain IDEs; those equations pop up in many fields namely physics, Astronomy potential theory, fluid dynamics, biological models, and chemical kinematics. Abstract

IDEs are usually difficult to solve analytically and as such, there is a need to obtain an efficient approximate solution. Recently, much interest from researchers in science and engineering has been given to non-traditional methods for non-linear IDEs. The Existence-uniqueness, stability, and application of integro-differential equations were presented by Lakshmikautham and Rao [19]. Armand and Gouyandeh discussed IDE of the first kind in [3] and nonlinear Fredholm Integral Equations of the second kind were discussed by Borzabadi, Kamyad, and Mehne in [7]. A comparison between Adomian Decomposition Method (ADM) and Wavelet-Galerkin Method for solving IDEs was considered in [11] . He's Homotopy Perturbation Method was applied to nth-order IDEs in [12] and [15]. Tau Numerical solution of Fredholm IDEs with arbitrary polynomial bases. Elaborate work on IDEs was discussed in [8,10,13,16,19,[22][23][24][25]31] and in [21] where Maleknejad and Mahmoudi applied Taylor polynomial to high-order nonlinear Volterra Fredholm Integro-differential Equations. Taylor Collocation Method was applied to linear IDEs in [18] by Karamete and Sezer. In [2]; Theory, Method, and Application of boundary value problems for higher-order integro-differential equations were considered. Wavelet-Galerkin method and Hybrid Fourier and Block-Pulse Function in [5] and [4] were applied to IDEs respectively. Numerical Approximation of nonlinear Fourth-Order IDEs by Spectral Methods were considered in [34][35][36][37][38] and in [32]. A New Algorithm was utilized in solving a class of nonlinear IDEs in the reproducing kernel space. In [30], a Comparison between Homotopy Perturbation Method and Sine-Cosine Wavelets Method was applied to linear IDEs while in [29], a new Homotopy Method was applied to First and Second Orders IDEs.
The pseudospectral method has been proposed by using shifted Chebyshev nested for solving the IDEs in [28] while [14] applied the Adomian Decomposition Method (ADM) for solving Fourth-Order Integro-differential Equations. In [30], the main objective was only to obtain the exact solution to Fourth-Order Integro-differential equations. The ADM in [14] and the Variational Method in [27] are applied to solve both linear and non-linear boundary value problems of fourth-order Integro-differential equation.
In recent years, meshless methods have gained more attention not only by mathematicians but also by researchers in other fields of sciences and engineering. During the past decades, the moving least square (MLS) method proposed in [20] has now become a very popular approximation scheme, especially when considering a mesh-free approximating function. In [17], MLS and Gauss Legendre were applied to solve Integral Equation of the second kind while [8] utilized MLS with Chebyshev polynomial as a basis function to solve IDEs and the basic MLS was adopted in [9] in the solution of IDEs. The work of [26] and [27] were on the application of a two -dimensional Interpolating Function to Irregularspaced data. A second kind chebyshev quadrature algorithm was developed for integral equations in [37] while a chebyshev collocation approached was adopted in the solution of IDEs in [33]. Many methodologies of IDEs in literature are popular with the use of regularspaced data, the disordered-spaced data approach of MLS requires great skill of computations and this has been a source of attraction to researchers over the years.
In this research work, we employ the MLS to solve fourth order integro-differential equation. The method is an effective approach for the approximation of an unknown function by using a set of disordered data. It consists of a local weighted least square fit, valid on a small neighborhood of a point, and does not require information about the where g(x) and h(x) are the limits of integration, is a constant parameter, k(x, t) is the kernel of the integral and u (n) (x) as defined in 1.1.1 above.   where F is a real non-linear continuous function, β, α i , i = 0, 1, 2, 3 are real constants, g(x), h(x) and f (x) are given.
Definition 1.1.5 [6]: The inverse of linear operator exists and it is linear L : P → Q.
This definition holds since if L −1 exists and its domain which is a vector space is Q then for any P 1 , P 2 ∈ P whose images are q 1 = LP 1 and q 2 = LP 2 we have P 1 = L −1 q 1 and P 2 = L −1 q 2 .L is linear implies that for any scalars α and β we have αq 1 + βq 2 = αLP 1 + βLP 2 = L(αP 1 + βP 2 ).
Thus for Y ∈ Q , there exists X in P such that L −1 : Y → X. In this paper, we consider a general n th order Volterra Integro-differential equation of the form: where F is a real non-linear continuous function, β, α i , i = 0, 1, 2, ..., n − 1 are real constants, g(x), h(x) and f (x) are given and can be approximated by the Taylor series. When n = 4 Eq. (5) reduces to fourth-order integro-differential equation with four conditions as proposed in this paper.

The conventional MLS scheme
This research is aimed at obtaining an efficient method for approximating voltterra integro-differential equations. The method was obtained by introducing an interpolation polynomial in the context of the moving least square method, thereby producing an enhanced form of the approach. The absolute difference between the true solutions and the approximated solutions obtained from the new approach was used to check how close the results are to the true solutions. This section comprises the basic idea of the conventional moving least square method and its convergence.

Overview of the conventional MLS
Consider a sub-domain x , the neighborhood of a point X , and the domain of definition of the MLS approximation for the trial function at X which is located in the problem domain . The approximation of the unknown function, u in x over some nodes, where P(x) is the basis function of the special coordinates, P T denotes the transpose of P,m is the number of basis function and a(x) is a vector containing coefficients a j (x), j = 0, 1, 2, ..., m which are functions of the space coordinate X . Also, 's a j (x) ′ s are the unknown coefficients to be determined. The coefficient vector a(x) is determined by minimizing a weighted discrete L 2 − norm , defined as: where U = (U 0 , U 1 , U 2 , ..., U n ) T is the exact solution and w i (x) is a new trigonometric weight function associated with the node i.n is the number of nodes for which the weight function, is always positive on [0, 1] and |.| denotes absolute value. The stationarity of J with respect to a j (x); j ≥ 0 gives: Selecting the values of x at the nodal points to ensure nonzero determinant of A and using the above inverse at each node, Eq. (10) becomes A simple Gram Schmidth algorithm that generates other polynomials:

Formulation of the proposed method
We wish to use the MLS method to obtain the numerical solution of (4): Suppose that the four-fold operator, exists. By applying (13) on both sides of (12) we have and To use the polynomials, we change the integral interval from [0, x] to a fixed interval [0, 1] using the translation t = xs; dt = xds : To apply the method, select the m + 1 polynomials (basis) with nodal points x i in [0, 1]. By using n j=0 U i ϕ i (x) instead of u(x) as the approximation of u(x) in (15) we have In compact form we have.
Finally, we introduce the use of interpolating polynomial, up(x) at all points in [0, 1]: Calculation of the unknowns requires 2N − 1 knots, 2N − 2 even steps, and MLS solution u(x) at the evaluation points The above equations constitute a solvable system of k equations in k unknowns.. In general, given z odd knots and N unknowns in Eq.

Numerical computations
In this section, we use the MLS Method to solve integro-differential equations in the interval [0, 1]. All computations were carried out with scripts written in 2015 MAT-LAB. The accuracy of this method is directly proportional to the number of basis functions (m) and the nodal points (n). To compute the integral part at the initial nodes, in the absence of an exact solution, we use a six-point Gauss Quadrature Rule (GQR). It involves the Gaussian nodes.
Using v as the number of nodes in the given evaluation points ( x ), initial condition: (1)) and j = 2 we estimate the corresponding values of u(x) through GQR: when the exact solution is unknown at the initial nodes.
The accuracy of MLS increases as the number of basis polynomials and nodal points increases.

Numerical examples
Example 1. Consider the following nonlinear fourth-order Integro-Differential Equation [1]: with initial conditions: U (i) (0) = 1, i = 0, 1, 2, 3. The exact solution is given by U (x) = e x and using the transformation in (15) with the given initial conditions, we have:

Solution of Example 1 with m = 5 and n points = 8:
Select initial nodal points xi, using dx = 0.25; and the corresponding approximate solution U (xi) . Following the outlined steps, we compute the values of u(x) at x = 0 to 1 in steps of 1/8 using the MLS method and five orthogonal polynomials. The following are the obtained results.
The optimal J (u) is quite close to zero, thus we expect a good approximation: The exact and Enhanced MLS solutions coincide at the knots (Fig. 1). All the interpolated values are close to the exact solution. An insignificant difference exists as shown in this figure. The next figure highlights this observation.  Table 4 are very close to the exact solution. From Fig. 3, the exact and approximate solutions coincide at the knots. The observed errors are insignificant as shown in Fig. 4. This implies perfect interpolation.

Solution of Example 2, using m = 5 polynomials and n = 15 nodes:
The exact solution is U (x) = x 5 . Applying (12) on Example 2, we have: Select initial nodal points and the corresponding approximate solution. Following the outlined steps, compute the values of u(x) in steps of 1/15 using the MLS method. The following are the obtained results.
U (x) = 1 55440 x 11 + 1 3024  The optimal J (u) is quite close to zero, thus we expect a good approximation: The exact and approximate solutions in Table 5   with initial conditions: The exact solution is given by U (x) = x 2 − 2 and using the transformation in (15) with the given initial conditions, we have
Following the procedure in example (1), the interpolating polynomial is Only the first and third coefficients are significant. Others are very close to zero and thus insignificant since their P-Values are greater than 0.05:      The Rsquare and Adjusted Rsquare are both 1.0. The statistics in Table 8 show that the chosen coefficients are the desired constants in The polynomial is a good fit for the MLS data. Any value, in.
[0, 1] interval can easily be evaluated with high precision.

Discussion of results
It is worthy to note that the computetions were carried out using MATLAB 9.2 on a personal computer of the following specifications.  Table 2 indicates a close proximity between the exact and MLS solutions. The following figure compares the obtained solutions.
All the P-Values, in Table 3, are less than 0.05. The computed Rsquare and Adjusted Rsquare are equal to 1.0. The statistics in Table 3 show that the estimated coefficients are the desired constants in The estimated polynomial is a good fit for the MLS data.  The observed errors are insignificant as shown in Fig. 6. This implies perfect interpolation.
Following the given procedure in example (1), the interpolating polynomial is All the computed coefficients are significant except the first which has zero value.
All parameters with a P-Value less than 0.05 are chosen. The Rsquare and Adjusted Rsquare are both one. The statistics in Table 6 show that the estimated coefficients are the desired constants of The polynomial is a good fit for the Enhanced MLS data. Any value of in [0, 1] interval can easily be evaluated with high precision. A high distinction of this method over existing methods is the significant interpolating polynomials obtained as a result of the constructed basis function which was then used to reproduce the solutions over the entire problem domain. The solutions produce a negligible magnitude of the error at each evaluation point and this demonstrates its reliability and effectiveness over existing methods.

Conclusion
An enhanced MLS method with smooth basis polynomials is used to solve the fourth order integro-differential equation of the Volterra type. At any arbitrary point, can be chosen to minimize the weight residual. Based on the results obtained, the value was given as a function of which accounts for the major difference between the Enhanced MLS, MLS method and the popular Least Square Method. Moreso, from the table of results, the error of the Enhanced MLS solution shows a tendency to increase as increases to the end boundary point. This behaviour is expected in any numerical method. Hence, we conclude that the proposed Enhanced Moving Least Square method is good for solving the class of equations described in this paper. Finally, a significant interpolating polynomial could be constructed and used to reproduce the solutions over the entire problem domain. The magnitude of the error at each evaluation knot demonstrates the reliability and effectiveness of this scheme. The application of the new weight function, svd, and orthogonal basis in the implementation of the conventional MLS method constitutes the said enhancement. The determinant of the moment matrix A(x) via SVD minimized the problem of near singularity and improved the accuracy of the results. The study concluded that enhanced MLS provides an alternative and efficient method of finding solutions to Volterra Integro-Differential equations and Fredholm-Volterra Integro-Differential equations. It is therefore recommended that the methods be used in solving the classes of problems considered.