AAA Abstract and Applied Analysis 1687-0409 1085-3375 Hindawi Publishing Corporation 562140 10.1155/2013/562140 562140 Research Article Numerical Solution of the Fractional Partial Differential Equations by the Two-Dimensional Fractional-Order Legendre Functions http://orcid.org/0000-0002-8353-7000 Yin Fukang Song Junqiang Wu Yongwen Zhang Lilun Ray Santanu Saha 1 College of Computer National University of Defense Technology Changsha 410073 China nudt.edu.cn 2013 18 11 2013 2013 13 05 2013 08 09 2013 08 09 2013 2013 Copyright © 2013 Fukang Yin et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A numerical method is presented to obtain the approximate solutions of the fractional partial differential equations (FPDEs). The basic idea of this method is to achieve the approximate solutions in a generalized expansion form of two-dimensional fractional-order Legendre functions (2D-FLFs). The operational matrices of integration and derivative for 2D-FLFs are first derived. Then, by these matrices, a system of algebraic equations is obtained from FPDEs. Hence, by solving this system, the unknown 2D-FLFs coefficients can be computed. Three examples are discussed to demonstrate the validity and applicability of the proposed method.

1. Introduction

Fractional partial differential equations play a significant role in modeling physical and engineering processes. Therefore, there is an urgent need to develop efficient and fast convergent methods for FPDEs. Recently, several different techniques, including Adomian’s decomposition method (ADM) [1, 2], homotopy perturbation method (HPM) , variational iteration method (VIM) , spectral methods , orthogonal polynomials method , and wavelets method  have been presented and applied to solve FPDEs.

The method based on the orthogonal functions is a wonderful and powerful tool for solving the FDEs and has enjoyed many successes in this realm. The operational matrix of fractional integration has been determined for some types of orthogonal polynomials, such as Chebyshev polynomials , Legendre polynomials , Laguerre polynomials , and Jacobi polynomials . Moreover, the operational matrix of fractional derivative for Chebyshev polynomials  and Legendre polynomials [9, 14] also has been derived. However, since these polynomials using integer power series to approximate fractional ones, it cannot accurately represent properties of fractional calculus. Recently, Rida and Yousef  presented a fractional extension of the classical Legendre polynomials by replacing the integer order derivative in Rodrigues formula with fractional order derivatives. The defect is that the complexity of these functions made them unsuitable for solving FDEs. Subsequently, Kazem et al.  presented the orthogonal fractional order Legendre functions based on shifted Legendre polynomials to find the numerical solution of FDEs and drew a conclusion that their method is accurate, effective, and easy to implement.

Benefiting from their “exponential-convergence” property when smooth solutions are involved, spectral methods have been widely and effectively used for the numerical solution of partial differential equations. The basic idea of spectral methods is to expand a function into sets of smooth global functions, called the trial functions. Because of their special properties, the orthogonal polynomials are usually chosen to be trial functions. Spectral methods can obtain very accurate approximations for a smooth solution while only need a few degrees of freedom. Recently, Chebyshev spectral method , Legendre spectral method , and adaptive pseudospectral method  were proposed for solving fractional boundary value problems. Moreover, generalized Laguerre spectral algorithms and Legendre spectral Galerkin method were developed by Baleanu et al.  and Bhrawy and Alghamdi  for fractional initial value problems, respectively.

Motivated and inspired by the ongoing research in orthogonal polynomials methods and spectral methods, we construct two-dimensional fractional-order Legendre functions and derive the operational matrices of integration and derivative for the solution of FPDEs. To the best of the authors’ knowledge, such approach has not been employed for solving FPDEs.

The rest of the paper is organized as follows. In Section 2, we introduce some mathematical preliminaries of the fractional calculus theory and fractional-order Legendre functions. In Section 3, a basis of 2D-FLFs is defined and some properties are given. Section 4 is devoted to the operational matrices of fractional derivative and integration for 2D-FLFs. Some numerical examples are presented in Section 5. Finally, we conclude the paper with some remarks.

2. Preliminaries and Notations 2.1. Fractional Calculus Theory

Some necessary definitions and Lemma of the fractional calculus theory [29, 30] are listed here for our subsequent development.

Definition 1.

A real function h(t), t>0, is said to be in the space Cμ, μR, if there exists a real number p>μ, such that h(t)=tph1(t), where h1(t)C(0,), and it is said to be in the space Cμn if and only if h(n)Cμ, nN.

Definition 2.

Riemann-Liouville fractional integral operator (Jα) of order α0, of a function fCμ, μ-1 is defined as (1)Jαf(t)=1Γ(α)0t(t-τ)α-1f(τ)dτ,t>0,J0f(t)=f(t), where Γ(α) is the well-known Gamma function. Some properties of the operator Jα can be found, for example, in [29, 30].

Definition 3.

The fractional derivative of f(x) in the Caputo sense is defined as (2)(Dαf)(x)={1Γ(m-α)×0xf(m)(ξ)(x-ξ)α-m+1dξ,(α>0,m-1<α<m),dmf(x)dxm,α=m, where f:RR, xf(x) denotes a continuous (but not necessarily differentiable) function.

Lemma 4.

Let n-1<αn, nN, t>0, hCμn, μ-1. Then (3)(JαDα)h(t)=h(t)-k=0n-1h(k)(0+)tkk!.

2.2. Fractional-Order Legendre Functions

In this section, we introduce the fractional-order Legendre functions which were first proposed by Kazem et al. . The normalized eigenfunctions problem for FLFs is (4)((x-x1+α)Liα(x))+α2i(i+1)xα-1Liα(x)=0,((x-x1+α)Liα(x))+α2i(i+1)xα-1x(0,1), which is a singular Sturm-Liouville problem. The fractional-order Legendre polynomials, denoted by FLiα(x), are defined on the interval [0,1] and can be determined with the aid of following recurrence formulae: (5)FL0α(x)=1,FL1α(x)=2xα-1,FLi+1α(x)=(2i+1)(2xα-1)i+1FLiα(x)-ii+1FLi-1α(x),i=1,2,, and the analytic form of FLiα(x) of degree i is given by (6)FLiα(x)=s=0ibs,ixsα,bs,i=(-1)i+s(i+s)!(i-s)!(s!)2, where FLiα(0)=(-1)i and FLiα(1)=1. The orthogonality condition is (7)01FLnα(x)FLmα(x)ω(x)dx=1(2n+1)αδnm, where ω(x)=xα-1 is the weight function and δ is the Kronecker delta. For more details, please see .

3. 2D-FLFs

In this section, the definitions and theorems of 2D-FLFs are given by Liu’s method described in .

3.1. Definitions and Properties of the 2D-FLFs Definition 5.

Let {FLnα(x)}n=0 be the fractional Legendre polynomials on [0,1]; we call {FLiα(x)FLjβ(y)}i,j=0 the two-dimensional fractional Legendre polynomials on [0,1]×[0,1].

Theorem 6.

The basis {FLiα(x)FLjβ(y)}i,j=0 is orthogonal on [0,1]×[0,1] with the weight function ω(x,y)=ω(x)ω(y)=xα-1yβ-1.

Proof.

Let im or jn(8)0101ω(x,y)FLiα(x)FLjβ(y)FLmα(x)FLnβ(y)dxdy=01ω(x)FLiα(x)FLmα(x)dx×01ω(y)FLjβ(y)FLnβ(y)dy=0.

Theorem 7.

Consider (9)0101ω(x,y)[FLiα(x)FLjβ(y)]2dxdy=1(2i+1)α1(2j+1)β,0101ω(x,y)[FLiα(x)FLjβ(y)]2dxdy=01ω(x)[FLiα(x)]2dx01ω(y)[FLjβ(y)]2dy=1(2i+1)α1(2j+1)β.

3.2. 2D-FLFs Expansion Definition 8.

A function of two independent variables f(x,y) which is integrable in square [0,1]×[0,1] can be expanded as (10)f(x,y)=i=0j=0aijFLiα(x)FLjβ(y), where (11)aij=(2i+1)(2j+1)αβ×0101f(x,y)ω(x,y)FLiα(x)FLjβ(y)dxdy.

Theorem 9.

If the series i=0j=0aijFLiα(x)FLjβ(y) converges uniformly to f(x,y) on the square [0,1]×[0,1], then we have (12)aij=(2i+1)(2j+1)αβ×0101f(x,y)ω(x,y)FLiα(x)FLjβ(y)dxdy.

Proof.

By multiplying ω(x,y)FLnα(x)FLmβ(y) on both sides of (10), where n and m are fixed and integrating termwise with regard to x and y on [0,1]×[0,1], then (13)0101f(x,y)ω(x,y)FLnα(x)FLmβ(y)dxdy=i=0j=0aij0101ω(x,y)FLiα(x)FLjβ(y)000000000000000000×FLnα(x)FLmβ(y)dxdy=i=0j=0aij01ω(x)FLiα(x)FLnα(x)dx000000000000000×01ω(y)FLjβ(y)FLmβ(y)dy=anm01ω(x)[FLnα(x)]2dx01ω(y)[FLmβ(y)]2dy=anm1(2n+1)α1(2m+1)β. Finally one can get (11).

If the infinite series in (10) is truncated, then it can be written as (14)f(x,y)i=0mj=0maijFLiα(x)FLjβ(y)=CTΨ(xα,yα), where C and Ψ(xα,yβ) are given by (15)C=[c0,0,c0,1,,c0,m-1,c1,0,c1,1,,c1,m-1,,cm-1,0,cm-1,1,,cm-1,m-1]T,(16)Ψ(xα,yβ)=[ψ0,0,ψ0,1,,ψ0,m-1,ψ1,0,ψ1,1,,ψ1,m-1,,ψm-1,0,ψm-1,1,,ψm-1,m-1]T, where ψij=FLiα(x)FLjβ(y), i=0,1,,m, and j=0,1,,m.

According to the definition of FLFs, one can find that fractional Legendre polynomials are identical to Legendre polynomials shifted to [0,1] when using the transform xαx, yβy. Therefore, in a similar method described in , we can easily get the convergence and stability theorems of proposed method.

Lemma 10.

If the function f(x,y) is a continuous function on [0,1]×[0,1] and the series i=0j=0aijFLiα(x)FLjβ(y) converges uniformly to f(x,y), then i=0j=0aijFLiα(x)FLjβ(y) is the 2D-FLFs expansion of f(x,y).

Let (17)f(x,y)=i=0j=0bijFLiα(x)FLjβ(y),f(x,y)~i=0j=0aijFLiα(x)FLjβ(y). Then there is at least one coefficient such that anmbnm. However, (18)bnm=(2n+1)(2m+1)αβ×0101f(x,y)ω(x,y)FLnα(x)FLmβ(y)dxdy=anm.

Lemma 11.

If two continuous functions defined on [0,1]×[0,1] have the identical 2D-FLFs expansions, then these two functions are identical.

Proof.

Suppose that f(x,y) and g(x,y) can be expanded by 2D-FLFs as follows: (19)f(x,y)~i=0j=0aijFLiα(x)FLjβ(y),g(x,y)~i=0j=0aijFLiα(x)FLjβ(y). By subtracting the above two equations with each other, one has (20)f(x,y)-g(x,y)~i=0j=0(aij-aij)FLiα(x)FLjβ(y)=0=i=0j=00×FLiα(x)FLjβ(y). Then Lemma 11 can be proved.

Theorem 12.

If the 2D-FLFs expansion of a continuous function f(x,y) converges uniformly, then the 2D-FLFs expansion converges to the function f(x,y).

Proof.

Theorem 12 can be proved by Theorems 7 and 9.

Theorem 13.

If the sum of the absolute values of the 2D-FLFs coefficients of a continuous function f(x,y) forms a convergent series, then the 2D-FLFs expansion is absolutely uniformly convergent, and converges to the function f(x,y).

Proof.

Consider (21)|i=0j=0aijFLiα(x)FLjβ(y)|i=0j=0|aij||FLiα(x)||FLjβ(y)|i=0j=0|aij|. Then i=0j=0aijFLiα(x)FLjβ(y) converges uniformly to the function f(x,y).

Theorem 14.

If a continuous function f(x,y), defined on [0,1]×[0,1], has bounded mixed partial derivative Dx2αDy2βf(x,y), then the 2D-FLFs expansion of the function converges uniformly to the function.

Proof.

Let f(x,y) be a function defined on [0,1]×[0,1] such that (22)|Dx2αDy2βf(x,y)|M, where M is a positive constant and (23)aij=(2i+1)(2j+1)αβ×0101f(x,y)ω(x,y)FLiα(x)FLjβ(y)dxdy. By employing the transform X=2xα-1 and Y=2yβ-1, one can obtain (24)aij=2i+122j+12-11-11f(X,Y)pi(X)pj(Y)dXdY. Consequently, in a similar method described in , Theorem 14 can be proved.

4. Operational Matrices of 2D-FLFs 4.1. Integration Operational Matrices of 2D-FLFs Lemma 15.

The Riemann-Liouville fractional integration of order γ>0 of the 2D-FLFs ψij can be obtained in the form of (25)Jxγ{ψij(xα,yβ)}=FLjβ(y)s=0ibsiΓ(1+sα)Γ(1+sα+γ)xsα+γ.

Proof.

Consider (26)Jxγ{ψij(xα,yβ)}=Jxγ{FLiα(x)FLjβ(y)}=Jxγ{FLiα(x)}FLjβ(y)=Jxγ{s=0ibsixsα}FLjβ(y)=FLjβ(y)s=0ibsiΓ(1+sα)Γ(1+sα+γ)xsα+γ.

Lemma 16.

Let γ>0; then one has (27)0101Jxγ{ψij}ψijω(x,y)dxdy={s=0is=0ibsibsi(s+s+1)α+γ×Γ(1+sα)Γ(1+sα+γ)1(2j+1)β,j=j0,jj.

Proof.

Using previous Lemma 15 and (6), one can have (28)0101Jxγ{ψij(xα,yβ)}ψij(xα,yβ)ω(x,y)dxdy=0101ω(x,y)FLiα(xα)FLjβ(yβ)FLjβ(yβ)×s=0ibsiΓ(1+sα)Γ(1+sα+γ)xsα+γdxdy=0101ω(y)FLjβ(yβ)FLjβ(yβ)×s=0is=0ibsibsiΓ(1+sα)Γ(1+sα+γ)×x(s+s+1)α+γ-1dxdy=01ω(y)FLjβ(yβ)FLjβ(yβ)×(01s=0is=0ibsibsiΓ(1+sα)Γ(1+sα+γ)×x(s+s+1)α+γ-1dxs=0is=0ibsibsiΓ(1+sα)Γ(1+sα+γ))dy=s=0is=0ibsibsi(s+s+1)α+γΓ(1+sα)Γ(1+sα+γ)×01ω(y)FLjβ(yβ)FLjβ(yβ)dy={s=0is=0ibsibsi(s+s+1)α+γ×Γ(1+sα)Γ(1+sα+γ)1(2j+1)β,j=j0,jj.

Theorem 17.

Let Ψ(xα,yβ) be the 2D-FLFs vector defined in (16); then one has (29)JxγΨ(xα,yβ)PxγΨ(xα,yβ), where Pxγ is the mm×mm operational matrix of Riemann-Liouville fractional integration of order γ>0, and has the form as follows: (30)Pxγ=[E0,0E0,1E0,m-1E1,0E1,1E1,m-1Em-1,0Em-1,1Em-1,m-1], in which Ei,i is m×m matrix and the elements are defined as follows: (31)Ei,i=Is=0is=0ibsibsi(2i+1)α(s+s+1)α+γΓ(1+sα)Γ(1+sα+γ),00000000000000000000000i,i=0,1,,m-1, and I is m×m identity matrix.

Proof.

Using (29) and orthogonality property of FLFs, one can get (32)Pxγ=JxγΨ(xα,yβ),ΨT(xα,yβ)H-1, where JxγΨ(xα,yβ),ΨT(xα,yβ) and H-1 are two m×m matrices defined as (33)JxγΨ(xα,yβ),ΨT(xα,yβ)={0101Jxγ{Ψk(xα,yβ)}×Ψk(xα,yβ)ω(x,y)dxdy0101Jxγ{Ψk(xα,yβ)}}k,kmm={s=0is=0ibsibsiΓ(1+sα)Γ(1+sα+γ)×1(s+s+1)α+γ1(2j+1)βs=0is=0ibsibsiΓ(1+sα)Γ(1+sα+γ)}i,i;j=jm;m,H-1={(2i+1)(2j+1)αβ}i,i;j=jm;m.

Now by substituting above equations in (32), Theorem 12 can be proved.

In a similar way as previous, one can obtain the operational matrix of Riemann-Liouville fractional integration with respect to variable y.

Theorem 18.

Let Ψ(xα,yβ) be the 2D-FLFs vector defined in (16); one has (34)JyγΨ(xα,yβ)PyγΨ(xα,yβ), where Pyγ is the mm×mm operational matrix of Riemann-Liouville fractional integration of order γ>0, and has the form as follows: (35)Pyγ=[EOOOEOOOE], in which E is m×m matrix and the elements are defined as follows: (36)Ej,j=r=0jr=0jbrjbrj(2j+1)β(r+r+1)β+γΓ(1+rβ)Γ(1+rβ+γ),000000000000000000000j,j=0,1,,m-1.

4.2. Derivative Operational Matrices of 2D-FLFs Lemma 19.

The FLFs Caputo fractional derivative of γ>0 can be obtained in the form of (37)Dxγ{ψij(xα,yβ)}=FLjβ(yβ)s=0ibsiΓ(1+sα)Γ(1+sα-γ)xsα-γ, where bs,i=0 when sαN0 and sα<γ in other case bs,i=bs,i.

Proof.

Consider (38)Dxγ{ψij(xα,yβ)}=Dxγ{FLiα(xα)FLjβ(yβ)}=FLjβ(yβ)Dxγ{FLiα(xα)}=Dxγ{s=0ibsixsα}FLjβ(yβ)=FLjβ(yβ)s=0ibsiΓ(1+sα)Γ(1+sα-γ)xsα-γ.

Lemma 20.

Let γ>0, αN; then one has (39)0101Dxγ{ψij}ψijω(x,y)dxdy={s=0is=0ibsibsi(s+s+1)α-γ×Γ(1+sα)Γ(1+sα-γ)1(2j+1)β,j=j0,jj.

Proof.

Using previous Lemma 19 and (6), one can have (40)0101Dxγ{ψij(xα,yβ)}ψij(xα,yβ)ω(x,y)dxdy=0101ω(x,y)FLiα(xα)FLjβ(yβ)FLjβ(yβ)×s=0ibsiΓ(1+sα)Γ(1+sα-γ)xsα-γdxdy=0101ω(y)FLjβ(yβ)FLjβ(yβ)×s=0is=0ibsibsiΓ(1+sα)Γ(1+sα-γ)×x(s+s+1)α-γ-1dxdy=01ω(y)FLjβ(yβ)FLjβ(yβ)×(01s=0is=0ibsibsiΓ(1+sα)Γ(1+sα-γ)×x(s+s+1)α-γ-1dx01s=0is=0ibsibsiΓ(1+sα)Γ(1+sα-γ))dy=s=0is=0ibsibsi(s+s+1)α-γΓ(1+sα)Γ(1+sα-γ)×01ω(y)FLjβ(yβ)FLjβ(yβ)dy={s=0is=0ibsibsi(s+s+1)α-γ×Γ(1+sα)Γ(1+sα-γ)1(2j+1)β,j=j0,jj.

Theorem 21.

Let Ψ(xα,yβ) be the 2D-FLFs vector defined in (16); one has (41)DxγΨ(xα,yβ)DxγΨ(xα,yβ), where Dxγ is the mm×mm operational matrix of Caputo fractional derivative of order γ>0, and has the form as follows: (42)Dxγ=[OOOF1,0OOFm-1,0Fm-1,1O] in which Fi,i is m×m matrix and the elements are defined as follows: (43)Fi,i=Is=0is=0ibsibsi(2i+1)α(s+s+1)α-γΓ(1+sα)Γ(1+sα-γ),0000000000000000000000i,i=0,1,,m-1, and I is a m×m identity matrix.

Proof.

Using (41) and the orthogonality property of FLFs, one can have (44)Dxγ=DxγΨ(xα,yβ),ΨT(xα,yβ)H-1, where DxγΨ(xα,yβ),ΨT(xα,yβ) and H-1 are two mm×mm matrices defined as (45)DxγΨ(xα,yβ),ΨT(xα,yβ)={0101Dxγ{Ψk(xα,yβ)}×Ψk(xα,yβ)ω(x,y)dxdy0101Dxγ{Ψk(xα,yβ)}}k,kmm={s=0is=0ibsibsi(s+s+1)α-γ×Γ(1+sα)Γ(1+sα-γ)1(2j+1)βs=0is=0ibsibsi(s+s+1)α-γ}i,i;j=jm;mH-1={(2i+1)(2j+1)αβ}i,i;j=jm;m.

Now by substituting above equations in (44), Theorem 21 can be proved.

In a similar way as above, one can get Caputo fractional derivative of order γ>0 with respect to variable y.

Theorem 22.

Let Ψ(xα,yβ) be the 2D-FLFs vector defined in (16); one can have (46)DyγΨ(xα,yβ)DyγΨ(xα,yβ), where Dyγ is the mm×mm operational matrix of Caputo fractional derivative of order γ>0, and has the form as follows: (47)Dyγ=[FOOOFOOOF], in which F is m×m matrix and the elements are defined as follows: (48)Fj,j=r=0jr=0jbrjbrj(2j+1)β(r+r+1)β+γΓ(1+rβ)Γ(1+rβ+γ),0000000000000000000000j,j=0,1,,m-1.

5. Applications and Results

Consider the following FPDEs: (49)Dxαu(x,t)+Dtβu(x,t)+N[u(x,t)]+L[u(x,t)]=g(x,t),α,β(0,1], where L and N are linear operator and nonlinear operator; respectively. Dα and Dβ are the Caputo fractional derivatives of order α and β, respectively; g is a known analytic function.

By employing operator Jtβ on both sides of (49) and then using the Lemma 4, one can have (50)u(x,t)+Jtβ{Dxαu(x,t)+Nu(x,t)+Lu(x,t)}-k=0m-1u(k)(x,0)xkk!-Jtβg(x,t)=0.

We first express unknown function u(x,t) and derivative term Dxαu(x,t) as (51)u(x,t)=CTΨ(xα,tβ),Dxαu(x,t)=CTDxαΨ(xα,tβ).

Now for the nonlinear part, by employing the nonlinear term approximation method described in  and then by using transform xxα, ttβ, one can get the 2D-FLFs expansion of nonlinear term as (52)Nu(x,t)=NTΨ(xα,tβ), where NT is coefficient matrix of nonlinear term which must be computed and its order is mm×mm.

For the linear part, we have (53)Lu(x,t)=LTΨ(xα,tβ), where L is a matrix of order mm×mm.

After substituting (51)–(53) into (50), one can obtain (54)CT+(CTDxα+NT+LT)Pyβ-CguessT=0. According to the Wu’s  technology for determining the initial iteration value, the initial iteration value is chosen as uguess=k=0m-1u(k)(x,0)(xk/k!)+Jtβ{g(x,t)}=CguessTΨ(xα,tβ). The coefficient matrix CT can be computed by using the MATLAB function fsolve( ) or the method described in .

Now, the present method is applied to solve the linear and nonlinear FPDEs, and their results are compared with the solution of other methods. The accuracy of our approach is estimated by the following error functions: (55)ej=(uexact)j-(uapprox)j,e=uexact-uapprox,eL=max1jN|ej|,eL2=j=1N|(ej)2|,eRMS=1Nj=1N|(ej)2|.

Example 23.

Consider the one-dimensional linear inhomogeneous fractional Burger’s equation : (56)βu(x,t)tβ+u(x,t)x-2u(x,t)x2=2t2-βΓ(3-β)+2x-2,0<β1, with the initial condition u(x,0)=x2 and the exact solution being u(x,t)=x2+t2.

By employing 2D-FLFs method, one can get (57)CT[I+(Dxα-(Dxα)2)Ptβ]=CguessT, where α=1. Then we can get CT=CguessTinv(I+(Dxα-(Dxα)2)Ptβ).

Figures 1(a) and 1(b) show the numerical results for β=0.25 with m=3, m=9 and β=0.5 with m=3, m=5, respectively. It should be found that the accuracy of 2D-FLFs method is very high while only a small number of 2D-FLFs are needed.

Numerical results for Example 23 when β = 0.25, 0.50.

Example 24.

Consider nonlinear fractional Klein-Gordon equation [36, 37]: (58)Dtβu(x,t)-Dxαu(x)+u3(x)=g(x,t),0000000x0,t>0,α,β(1,2], subject to the initial conditions (59)u(x,0)=0,ut(x,0)=0, and g(x,t)=Γ(β+1)xα-Γ(α+1)tβ+x3αt3β. The exact solution of (58) is u(x,t)=xαtβ.

By employing 2D-FLFs method with m=3 and m=3, one can have (60)CT+(-CTDxα+NT)Ptβ-CguessT=0.

The numerical results of Example 24 for different values of α and β are shown in Figure 2. In addition, L2 and L errors are presented in Table 1. From Table 1, one can conclude that the solutions of 2D-FLFs method are in good agreement with the exact results. Compared with homotopy analysis method (HAM)  and homotopy perturbation method (HPM) , 2D-FLFs method can get high accuracy solution while only need a few terms of 2D-FLFs.

Errors of Example 24 for different values of α and β with M=M=4.

Error α = β = 1.25 α = β = 1.50 α = β = 1.75 α = β = 2.00
L 2 5.6437 e - 015 1.2075 e - 015 3.4584 e - 015 8.9917 e - 016
L 4.4409 e - 016 1.1102 e - 016 3.3307 e - 016 1.1102 e - 016

Numerical results of Example 24 for different values of α and β.

Example 25.

Consider the nonlinear time-fractional advection partial differential equation  (61)Dtβu(x,t)+u(x,t)ux(x,t)=x+xt2,00000000t>0,xR,0<β1, subject to the initial condition (62)u(x,0)=0.

Figure 3 gives the approximation solutions of (61) for β=0.50 with m=4, m=5 and β=0.75 with m=4, m=9. Moreover, Table 2 shows the approximate solutions for (61) obtained for different values of β using the fractional variational iteration method (FVIM)  and 2D-FLFs method. The values of β=1 are the only case for which we know the exact solution u(x,t)=xt. It should be noted that only the fourth-order term of the FVIM was used in evaluating the approximate solutions for Table 2. From Table 2, it clearly appears that 2D-FLFs method is more accurate than FVIM and the obtained results are in good agreement with exact solution.

Numerical values when β = 0.50, 0.75, and 1.0 for (61).

t x β = 0.50 β = 0.75 β = 1.00
FVIM 2D-FLFs FVIM 2D-FLFs FVIM 2D-FLFs Exact
0.25 0.25 0.12422501 0.12225461 0.09230374 0.09224583 0.06250058 0.062500 0.062500
0.50 0.24845002 0.24450922 0.18460748 0.18449165 0.12500117 0.125000 0.125000
0.75 0.37267504 0.36676383 0.27691122 0.27673748 0.18750175 0.187500 0.187500
1.00 0.49690005 0.48901844 0.36921496 0.36898331 0.25000234 0.250000 0.250000

0.50 0.25 0.18377520 0.16584130 0.15148283 0.14985508 0.12507592 0.125000 0.125000
0.50 0.36755040 0.33168259 0.30296566 0.29971016 0.25015184 0.250000 0.250000
0.75 0.55132559 0.49752389 0.45444848 0.44956524 0.37522776 0.375000 0.375000
1.00 0.73510079 0.66336518 0.60593131 0.59942032 0.50030368 0.500000 0.500000

0.75 0.25 0.27227270 0.20678964 0.21407798 0.20119503 0.18881843 0.187500 0.187500
0.50 0.54454540 0.41357929 0.42815596 0.40239005 0.37763687 0.375000 0.375000
0.75 0.81681810 0.62036893 0.64223394 0.60358508 0.56645530 0.562500 0.562500
1.00 1.08909080 0.82715857 0.85631192 0.80478011 0.75527373 0.750000 0.750000

Numerical results of Example 25 for different value of β.

Example 26.

We finally consider the linear time-fractional wave equation: (63)2βut2β=12x22ux2,t>0,xR,0.5<β1, subject to the initial conditions (64)u(x,0)=x,u(x,0)t=x2.

Table 3 gives a comparison of the approximate solutions at different values of β using the FVIM  and 2D-FLFs method. Figure 4 shows the numerical solutions of 2D-FLFs method for (63) at different values of β with m=3, m=9. The values of β=1 are the only case for which we know the exact solution u(x,t)=x+x2sinh(t). As previous, only the fourth-order term of the FVIM was used in evaluating the numerical solutions for Table 3. In the case of β=1, it can be found that absolute error of 2D-FLFs is not bigger than 1.0e-10 which is very small compared with that obtained by FVIM.

Numerical values when β = 0.750, 0.875, and 1.000 for (63).

t x β = 0.750 β = 0.875 β = 1.000
FVIM 2D-FLFs FVIM 2D-FLFs FVIM Exact
0.25 0.25 0.26622298 0.26622021 0.26593959 0.26594005 0.26578827 0.26578827
0.50 0.56489190 0.56488083 0.56375836 0.56376020 0.56315308 0.56315308
0.75 0.89600678 0.89598187 0.89345630 0.89346046 0.89209443 0.89209443
1.00 1.25956762 1.25952332 1.25503343 1.25504082 1.25261232 1.25261232

0.50 0.25 0.28474208 0.28474415 0.28340402 0.28340659 0.28256846 0.28256846
0.50 0.63896831 0.63897662 0.63361610 0.63362636 0.63027383 0.63027383
0.75 1.06267869 1.06269739 1.05063622 1.05065931 1.04311611 1.04311611
1.00 1.55587323 1.55590647 1.53446439 1.53450544 1.52109530 1.52109531

0.75 0.25 0.30690489 0.30690747 0.30361709 0.30361656 0.30139478 0.30139480
0.50 0.72761955 0.72762986 0.71446834 0.71446625 0.70557913 0.70557918
0.75 1.26214400 1.26216719 1.23255378 1.23254905 1.21255304 1.21255316
1.00 1.91047821 1.91051944 1.85787338 1.85786498 1.82231652 1.82231673

Numerical results of Example 26 for different value of β.

6. Conclusion

We define a basis of 2D-FLFs and derived its operational matrices of fractional derivative and integration, which are used to approximate the numerical solution of FPDEs. Compared with other numerical methods, 2D-FLFs method can accurately represent properties of fractional calculus. Moreover, only a small number of 2D-FLFs are needed to obtain a satisfactory result. The obtained results demonstrate the validity and applicability of proposed method for solving the FPFEs.

Acknowledgments

This work is supported by the National Natural Science Foundation of China (Grant no. 11272352). The authors are grateful to the anonymous referees for their comments which substantially improved the quality of this paper.