The Polynomial Least Squares Method for Nonlinear Fractional Volterra and Fredholm Integro-Differential Equations

: We present a relatively new and very efﬁcient method to ﬁnd approximate analytical solutions for a very general class of nonlinear fractional Volterra and Fredholm integro-differential equations. The test problems included and the comparison with previous results by other methods clearly illustrate the simplicity and accuracy of the method.


Introduction
The two mathematicians Vito Volterra and Erik Ivar Fredholm, through their works published in the early 1900s, laid the foundations of the modern theory of integro-differential equations. As the fractional Volterra and Fredholm integro-differential equations have multiple applications in various fields such as engineering, physics, mechanics, etc., they have aroused the interest of many researchers.
In recent years there have been numerous papers in which, for equations of this type, approximate analytical solutions or numerical solutions by various methods are presented, and this is because obtaining an exact solution is often impossible.
Among the methods used to compute approximate analytical and numerical solutions for fractional Volterra and Fredholm integro-differential equations we mention: • The Homotopy Perturbation Method, used by Ghasemi et al. in 2007 ([1])to solve nonlinear integro-differential equations and by Dheghan and Shakeri in 2008 ( [2]) to solve integro-differential equation with time-periodic coefficients, • The Bernoulli Matrix Method, used by Bhrawya et al. in 2012 ([3]) to find solutions for fractional Fredholm integro-differential equations, by Keshavarz et al. in 2019 ([4]) to find solutions for a class of nonlinear mixed Fredholm-Volterra integro-differential equations of fractional order and by Rajagopal et al. in 2020 ([5]) to find solutions for fractional-order Volterra integro-differential equations, • The Legendre Wavelets Method, employed by Meng et al. in 2014 ([6]) to solve Volterra-Fredholm integro-differential equations of fractional order, • The Legendre Spectral Element Method, employed by Lotfi and Alipanah in 2020 (( [7]) to solve Volterra-Fredholm integro-differential equations, • The Shifted Jacobi-Spectral Collocation Method, used by Al-Safi in 2018 ( [8]) to solve Volterra-Fredholm integro-differential equations of fractional order, • The Reproducing Kernel Hilbert Space Method, used by Arkub et al. in 2013 ( [9]) to solve Fredholm integro-differential equations, • The Sinc-Collocation Method, used by Alkan and Hatipoglu in 2017 ( [10]) to solve Volterra-Fredholm integro-differential equations of fractional order, • The Legendre Collocation Method, employed by Rohaninasab et al. in 2017 ([11]) to solve high-order linear Volterra-Fredholm integro-differential equations, • The Adomian Decomposition Method, used by Momani and Noor in 2006 ([12]) and by Farhood in 2015 ( [13]), • The Shifted Chebyshev Polynomials Method, also used by Farhood in 2015 ( [13]) to find solutions for nonlinear fractional integro-differential equations, • The Genocchi Polynomials Method, employed by Loh et al. in 2017 ([14]) to solve Volterra-Fredholm integro-differential equations of fractional order, • The Euler Wavelets Method, used by Wang and Zhu in 2017 ( [15]) to find approximate solutions for Fredholm-Volterra fractional integro-differential equations, • The Lucas Wavelets Method, employed by Dehestani [28]) to find solutions for fractional Fredholm-Volterra integro-differential equations.
The class of equations studied in this paper is: which, depending on the problem, may have attached a set of conditions of the type: Here, for q ∈ N * , D α denotes the Caputo fractional derivative of order α, namely: The kernel functions K f , K v and the function F are assumed to have suitable derivatives on the closed interval [0, 1] such that the problem consisting of the Equation (1) together with initial conditions (2) (if present) admits a solution.
This class of equations is evidently a very general one since it includes both Fredholm and Volterra-type equations, linear and nonlinear, and also both integro-differential and integral equations.
Unfortunately the exact solution of a nonlinear integro-differential equation of the type (1) cannot be found, with the exception of a relatively small number of simple cases (such as the test problems as examples). Thus, numerical solutions or (preferably) approximate analytical solutions must be computed.
The rest of the paper is structured as follows: in Section 2, we present the Polynomial Least Squares Method (denoted from this point forward as PLSM), in Section 3 we present the results of an extensive testing process involving most of the usual test problems included in similar studies, and in Section 4 we present the conclusions of the study.

The Method
In the following, we will denote the problems (1) + (2) and the Equation (1) together with the conditions (2).
For the problems (1) + (2), we consider the operator: If x app is an approximate solution of the Equation (1), the error obtained by replacing the exact solution x with the approximation x app is given by the remainder We will find approximate polynomial solutions u app of the problems (1) + (2) on the closed interval [0, 1], solutions which satisfy the following conditions: Definition 1. We call an ε-approximate polynomial solution of the problem (1) + (2) an approximate polynomial solution x app , satisfying the relations (6) and (7).

Definition 2.
We call a weak δ-approximate polynomial solution of the problems (1) + (2) an approximate polynomial solution x app , satisfying the relation: the initial conditions (6).

Definition 3.
We consider the sequence of polynomials P m (t) = a 0 + a 1 t + . . . + a m t m , a i ∈ R, i = 0, 1, . . . , m, satisfying the conditions: We call the sequence of polynomials P m (t) convergent to the solution of the problems (1) + (2) if lim m→∞ D(P m (t)) = 0.
The following convergence theorem holds: Theorem 1. The necessary condition for the problems (1) + (2) to admit a sequence of polynomials P m (t) convergent to the solution of this problem is: lim ε-approximate polynomial solution of the problem (1) + (2).
Proof. We will find a weak ε-polynomial solution of the type: where the constants c 0 , c 1 , . . . , c m are calculated using the steps outlined in the following.
• By substituting the approximate solution (8) into the Equation (1), we obtain the following expression: If we could find the constants c 0 0 , c 0 1 , . . . , c 0 m such that R(t, c 0 0 , c 0 1 , . . . , c 0 m ) = 0 for any t ∈ [0, 1] and the equivalents of (2) (if included in the problem): are also satisfied, then by substituting c 0 0 , c 0 1 , . . . , c 0 m in (7) we obtain the exact solution of the problems (1) + (2). In general this situation is rarely encountered in polynomial approximation methods. • Next, we attach to the problems (1) + (2) the following real functional: where c 0 , c 1 , . . . , c n−1 may be computed as functions of c n , c n+1 , . . . , c m by using the initial conditions (9) Based on the way the coefficients of polynomial T m (t) are computed and taking into account the relations (8)-(12), the following inequality holds: It follows that 0 ≤ lim We obtain lim From this limit we obtain that ∀ε > 0, ∃m 0 ∈ N satisfying the following property: ∀m ∈ N, m > m 0 ⇒ T m (t) is a weak ε -approximate polynomial solution of the problem (1) + (2).

Remark 1.
Any ε-approximate polynomial solution of the problem (1) + (2) is also a weak ε 2 · (b − a) approximate polynomial solution, but the opposite is not always true. It follows that the set of weak approximate solutions of the problem (1) + (2) also contains the approximate solutions of the problem.
Taking into account the above remark, in order to find ε-approximate polynomial solutions of the problems (1) + (2) by the Polynomial Least Squares Method, we will first determine weak approximate polynomial solutions,x app . If |R(t,x app )| < ε, thenx app is also an ε-approximate polynomial solution of the problem.

Application 1: Fredholm Nonlinear Fractional Integro-Differential Equation
The first application is the problem consisting of the equation ( [13,19]): together with the initial condition The exact solution of this problem is x e (t) = t. We will follow the steps of the algorithm described in the proof from the previous section. First, by choosing a first order polynomialx(t) = c 1 · t + c 0 , from the initial condition, it follows that c 0 = 0, sox(t) = c 1 · t.
The corresponding remainder (9) is and the corresponding functional (11) is By solving the equation J (c 1 ) = 0, we obtain three stationary (equilibrium) points and it is easy to show that the minimum of the functional corresponds, as expected, to the value c 1 = 1. This means that PLSM is able to find, in a very simple manner, the exact solution of the problem,x(t) = x e (t) = t. We remark that the previous methods in [13] (Shifted Chebyshev Polynomials Method and Adomian Decomposition Method) and [19] (Chebyshev Wavelets Method) were only able to find approximate solutions.

Application 2: Fredholm Nonlinear Fractional Integro-Differential Equation
The second application is ( [13,19]): where The exact solution of of the problem consisting of Equation (14) and the initial condition Again, using the PLSM steps outlined in the previous section, we choosex(t) = c 3 · t 3 + c 2 · t 2 + c 1 · t + c 0 , obtain c 0 = 0 from the initial condition and compute the corresponding remainder By minimizing the functional J(c 1 , c 2 , c 3 ) (with an expression too long to be included here), we obtain c 1 = 0, c 2 = 0 and c 3 = 1, thus finding the exact solutionx(t) = x e (t) = t 3 . We remark that the previous methods in [13,19] were only able to find approximate solutions.
Using PLSM for several fractional values for α, we obtain in the same manner: For α = 0.7:x 0.7 (t) = 1.1045793185308062 · t. Figure 1 presents the plots of these approximate solutions:

Application 4: Volterra Nonlinear Fractional Integro-Differential Equation
The next example consists of the Volterra equation ( [15,16,20]): where The exact solution of the problem is x e (t) = t 2 − t.
Using PLSM withx(t) = c 2 · t 2 + c 1 · t + c 0 we find from the condition that c 0 = 0 and the reminder (9) is Minimizing the corresponding functional J(c 1 , c 2 ) (too long to be included here) we get c 1 = −1 and c 2 = 1 which means that we obtain in fact the exact solutionx = x e = t 2 − 1. Again we remark that the previous methods in [15] (Euler Wavelets Method), [16] (Lucas Wavelets Method) and [20] (Chebyshev Wavelets Method) were only able to find approximate solutions.

Application 5: Volterra Fractional Integro-Differential Equation
We consider the Volterra equation ( [19,20,24]): where The problem consists of the Equation (17), together with the conditions Using PLSM, we choosex(t) = c 2 · t 2 + c 1 · t + c 0 . By using the initial conditions, we · (c 2 − 1) and the corresponding reminder and after minimizing the functional J(c 2 ), we find the minimum as c 2 = 1. Thus,x = t 2 , which is the exact solution of the problem while, again, the previous methods in [19] (Chebyshev Wavelets Method), Ref. [20] (also a Chebyshev Wavelets Method) and [24] (Newton-Kantorovitch Method) were only able to find approximate solutions.

Application 6: Volterra Nonlinear Fractional Integro-Differential Equation
We consider the problem consisting of the Volterra equation ( [2,16]): together with the condition x(0) = 0. For α = 1 by using the same PLSM steps as above we obtain the exact solution of the problemx(t) = x e = t.
Using PLSM for several fractional values for α, we obtain in the same manner: For α = 0.7:x 0.7 (t) = 1.0877449515729398 · t. Figure 2 presents the plots of these approximate solutions:

Application 7: Volterra Fractional Integro-Differential Equation
We consider the Volterra equation ( [2,16]): where To this equation, we attach the conditions x(0) = 1, x (0) = 0. The exact solution of this problem is only known in the case α = 1, when the exact solution is x e (t) = t · sin(t) + cos(t).

Application 9: Volterra-Fredholm Nonlinear Fractional Integro-Differential Equation
We consider the problem consisting of the Volterra-Fredholm equation ( [6,20]): where Again, while the methods in [6] (Legendre Wavelets Method) and [20] (Chebyshev Wavelets Method) are able to find good approximate solutions, by using PLSM we are able to find the exact solution of the problem, x e (t) = t 3 .

Application 13: Fredholm Fractional Integro-Differential Equation
The 13th application consists of the Fredholm equation ( [11,16,30,31]): together with the initial condition x(0) = 1. The exact solution is only known for the case α = 1, namely x e (t) = e t . For this case, approximate solutions were previously computed by using a hybrid Fourier and Block-pulse functions method in [30], by using a Taylor polynomials method in [31], by using a Legendre collocation method in [11] and by using a Lucas wavelets method in [16].
Since, unfortunately, the results in [11,16,30,31] were not presented in the same manner or for the same set of values of t, in Table 4 we present the order of the absolute errors corresponding to the most accurate approximations in these studies, together with the ones corresponding to the above approximate solutions computed by PLSM.
together with the boundary conditions x(0) = 1, x (0) = 1, x(1) = e, x (1) = e. The exact solution known for the case α = 4 is x e (t) = e t and approximate solutions in this case were previously computed by using the Adomian Decomposition Method in [12,32] and by using a CAS wavelets method in [33].
Again, the results in [12,32,33] were not presented as errors for a given set of values of t. The results obtained by the Adomian Decomposition Method were presented in [32] as a plot of the error function corresponding to the approximation, while in [33], the results obtained by the CAS wavelets method were presented by means of the approximate norm-2 ( er(t) 2 ) of the error functions corresponding to the approximations.
Computing the corresponding norm-2 for our approximations by PLSM, in Table 5 we present these results together with those corresponding to the most accurate approximations in the previous studies. Table 5. Comparison of the order of the absolute errors of the approximate solutions for problem (26) corresponding to α = 4. For fractional values of α we also computed by using PLSM several approximate solutions, plotted in Figure 5:

Conclusions
We presented the Polynomial Least Squares Method as a straightforward, efficient and accurate method to find approximate analytical solutions for a very general class of fractional nonlinear Volterra-Fredholm integro-differential equations.
The paper contains an extensive application list, including most of the usual test problems used for this type of equation and compare our results with previous results obtained by using other well-known methods. For the test problems where the exact solution is a polynomial one, PLSM is able to find the exact solution in a simple manner, while most of the other methods previously used were only able to compute approximate solutions. If the solution is not polynomial, PLSM is able to find approximate solutions, again in a very straightforward way, with errors usually smaller that the ones corresponding to the approximations computed by other methods.
Taking into the account the above considerations, the results of this paper recommend PLSM as a very useful tool in the study of fractional nonlinear integro-differential equations.