Hybrid Legendre functions to solve differential equations with fractional derivatives

This paper introduces an approach based on hybrid operational matrix to obtain a numerical scheme to solve fractional differential equations. The idea is to convert the given equations into a system of equations, based on the block-pulse and Legendre polynomials. Also, we employ the Banach fixed-point theorem to analyze the problem on the Banach algebra C[0, b] for some fractional differential equations, which include many key functional differential equations that arise in linear and nonlinear analysis.


Introduction
The fractional differential equation (FDE) has received considerable interest and has constituted many applications in various scientific such as continuum and statistical mechanics [10], dynamical systems [8], and optimal control problems [2,7,9]. The most of FDEs cannot be solved analytically, and on the other hand, many applications of these problems motivate us to develop numerical schemes for their solutions. For this purpose, some techniques were suggested and there are some studies on the numerical method to solve FDEs, for example, see [9,11,12,14,15]. Another technique applied to solve FDE is to use the operational matrix of fractional order [2,6,11,17]. In this study, we present a numerical technique to solve FDEs where m 2 N and C D : denotes the Caputo fractional derivative [3,14]. Our suggested method is based upon the piecewise continuous functions and Legendre polynomials, depending on the operational matrices of fractional integration. The exclusivities of hybrid functions with the operational matrix are used to convert the FDE to an algebraic equation, and then, are utilized to evaluate the upon expanding unknown function by the basis functions with unknown coefficients.
The following is an overview of this article. In Sect. 2, briefly some definitions and mathematical preliminaries of the fractional calculus have been introduced. In Sect. 3, we consider existence and uniqueness theorems of the desired FDEs. Some proper properties of the hybrid basis consisting of the block-pulse functions and Legendre polynomials, and approximation of function by these basis are presented in Sect. 4. The relevant operational matrix is obtained in Sect. 5, end of this section is devoted to applying the hybrid functions method for solving FDEs. In Sect. 6, through the provided examples, our numerical finding reported and the reliability and performance of the proposed scheme is demonstrated.

Preliminaries and basic definitions
We give a few traits concerning of fractional calculus [3,14]. Let g 2 L 1 ½0; b (space of Lebesgue integrable real functionsÞ; and l 2 R þ ¼ ð0; 1Þ; be a fixed number.
for 0 s b, is denominated the Riemann-Liouville fractional integral operator of order l, where Cð:Þ denotes the Gamma function.
The operator J l transforms the space L 1 ½0; b into itself [14].
g Þis called the Riemann-Liouville fractional differential operator.
is the Caputo fractional derivative.
Note that J l C D l g s ð Þ ¼ g s ð Þ À P mÀ1 j¼0 g j ð Þ 0 þ ð Þ s j j! ; m À 1\l m; m 2 N: Let l ! 0. Assume that g is such that both C D l g and RL D l g exist. Then, Under the hypotheses of Lemma 1, C D l g s ð Þ ¼ RL D l g s ð Þ holds if and only if g has an l d e-fold zero at 0, i.e., if and only if D j y 0 ð Þ ¼ 0; for j ¼ 0; 1; . . .; l d e À 1:

Existence and uniqueness
We study the solvability of Problem () for g 2 C½0; b. In what follows, we suppose that Fðs; :Þ : ½0; b Â R À! R; be satisfied in the Lipschitz condition respect to the second component, with Lipschitz constant l, and there exist the constants k and g such that F s; gðsÞ ð Þ j j k þ g gðsÞ j j(sublinear nonlinearity), for all s 2 ½0; b and gðsÞ 2 R. Theorem 1 For 0\h ¼ lb l Cðlþ1Þ \1, Eq. (1a) and (1b) has a unique solution.
Proof To prove this result, we define operator K on the space We shall show that K : ½0; b Â R À! R is a contraction map. For g 1 ; g 2 2 C½0; b and s 2 ½0; b, we have Therefore, according to condition 0\h\1; the mapping K is contraction, so by the Banach's principle has a unique fixed-point, and there exists a unique solution to problem (1). h Proof Let s 1 ; s 2 2 ½0; b; s 1 \s 2 ; and g belong to a bounded set, then we have As s 1 ! s 2 the last term tends to zero. The equicontinuous for the cases s 1 \s 2 0 and s 1 0 s 2 is explicit. h

Basis functions
A set of block-pulse functions b p ðsÞ; p ¼ 1; 2; . . .; P for s 2 ½0; 1Þ is defined as follows [11,12]: These functions are disjoint and have the property of orthogonality on [0, 1). The hybrid functions h pq ðsÞ; p ¼ 1; 2; . . .; P; q ¼ 0; 1; . . .; Q À 1; on ½0; s f Þ are defined as where p is the order of the block-pulse functions and L q ðsÞs are the well-known Legendre polynomials of order q with the following recursive formula:

It is obvious that the set of hybrid functions is orthogonal.
A function g(s), defined on ½0; s f Þ can be expanded as where S ¼ PQ, Applying operational matrices where P l SÂS is obtained as the operational matrix of the fractional integration for hybrid functions, by the following formula: Also, matrix U SÂS is an invertible matrix and define using vector H S ðsÞ in collocation points s p ¼ 2pÀ1 2S ; p ¼ 1; 2; . . .; S as following: ; . . .; S À 1. Furthermore, using Eq. (2) and taking B S ðsÞ ¼ ½b 1 ðsÞ; b 2 ðsÞ; . . .; b S ðsÞ T ; hybrid functions can be expanded by Sterm of the block-pulse functions as and since F l SÂS is the operational matrix associated with the block-pulse functions, we get Finally, from Eqs. (6)- (8), one can conclude that Method implementation Consider a nonlinear differential equation of fractional order. We approximate C D l gðsÞ by the hybrid functions as where C S ¼ c 1 ; c 2 ; . . .; c S ½ T is an unknown vector. From Eq. (10), we get from Eq. (9), we have Substituting C D l gðsÞ and g(s) from relations (10), (11) in Eq. (1a) and (1b), we obtain a system of algebraic equations. Implementation of the proposed method is presented in the next section via numerical experiments.

Numerical experiments
We present some examples to comprehend overview and demonstrate the efficiency of the described method. where h T S is a known constant vector. Substituting these equations into Eq. (12), we give We solved the problem, applying the technique described in Sect. 4, the absolute errors for Q ¼ 3; P ¼ 2; 4; 6 are listed in Table 1. From Table 1, observed that, we have an acceptable approximation of the exact solution. Also, increasing the number of basis functions, provide improvement in the accuracy of the solutions.
Example 2 In [4,16,17], the FDE C D l g s ð Þ ¼ Àg s ð Þ; 0\l 2; has been solved by different methods. The exact solution is as follows [4]: is the Mittag-Leffler function of order l.
Since J l C D l gðsÞ ¼ gðsÞ À gð0Þ À tg 0 ð0Þ; we have the following algebraic system for Eq. (14): It is evident that as l approaches close to 1 or 2, the numerical solution by the presented hybrid method in previous sections converges to the exact solution. Table 2 shows the absolute errors for l ¼ 0:85; 1:2; 1:5 and S ¼ 8; 10; 24. Clearly, the approximations achieved through the hybrid scheme are in accordance with those established with other mentioned numerical schemes [16,17].   Let applying virtues of the block-pulse function, we get Substituting these equations into FDE (15), we have the following system of nonlinear equations: À 1; 1; . . .; 1 ½ B S ðsÞ ¼ 0: For l ¼ 1, the analytic solution of Eq. (15) is In Table 3, the results for Example 3 with l ¼ 0:5; 1 and S ¼ 16; by the hybrid method in some points s 2 ½0; 1; are given. Also, these outcomes are compared with Refs. [13,18]. Moreover, absolute errors of approximate solutions of Example 3 for S ¼ 48 are shown in Fig. 2.   Fig. 3.
Example 5 For FDE [4,17],  Table 4. therefore, From boundary condition (18) and Eq. (19), one concludes that Consequently, FDE (17) can be shown as the following algebraic system: reported in Table 5. Also, the approximate error of our proposed scheme for this example is illustrated in Fig. 4.