Applied Mathematics and Computation 256 (2015) 930–938

Contents lists available at ScienceDirect

Applied Mathematics and Computation

journal homepage: www.elsevier.com/locate/amc

A meshless method based on Point Interpolation Method (PIM)

for the space fractional diffusion equation
Q. Liu a,e, F. Liu b,⇑, Y.T. Gu c, P. Zhuang a,e, J. Chen d, I. Turner b
a
School of Mathematical Sciences, Xiamen University, Xiamen, Fujian, China
b
School of Mathematical Sciences, Queensland University of Technology, GPO Box 2434, Brisbane, Qld. 4001, Australia
c
School of Chemistry, Physics and Mechanical Engineering, Queensland University of Technology, GPO Box 2434, Brisbane, Qld. 4001, Australia
d
School of Sciences, Jimei University, Xiamen, Fujian, China
e
Fujian Provincial Key Laboratory of Mathematical Modeling and High-Performance Scientific Computation, Xiamen University, Xiamen 361005, China

a r t i c l e i n f o a b s t r a c t

Keywords: This paper aims to develop a meshless approach based on the Point Interpolation Method
Fractional diffusion equation (PIM) for numerical simulation of a space fractional diffusion equation. Two fully-discrete
Meshless approach schemes for the one-dimensional space fractional diffusion equation are obtained by using
Point interpolation method the PIM and the strong-forms of the space diffusion equation. Numerical examples with
Numerical simulation
different nodal distributions are studied to validate and investigate the accuracy and
efficiency of the newly developed meshless approach.
Ó 2015 Elsevier Inc. All rights reserved.

1. Introduction

In the last decades fractional differential equations have played a very important role in various fields such as mechanics,
signal processing, chemistry, electricity and economics. There is a growing realization that relaxation and diffusion phenom-
ena in complex materials cannot be expressed simply in terms of a sum of exponential decays, each characterized by a single
relaxation time or rate.
A physical–mathematical approach to anomalous diffusion may be based on generalized partial differential equations
(PDEs) containing derivatives of fractional order in space, or time, or space–time. Consequently many references discussed
the analytical and numerical solutions for space, time, and space–time fractional partial differential equations (FPDEs) [1–5].
For space fractional differential equations, most researchers considered the Finite Difference Method (FDM) to discretise the
space derivative. Liu et al. [6,7] proposed a computationally effective method of lines. They transformed the space fractional
partial differential equation into a system of ordinary differential equations (Method of Lines) that was then solved using
backward differentiation formulae. Meerschaert and Tadjeran [8] developed finite difference approximations for fractional
advection–dispersion flow equations. Meerschaert et al. [9] derived practical numerical methods to solve two-dimensional
fractional dispersion equations with variable coefficients on a finite domain and obtained first order accuracy in space and
time. Liu et al. [10] discussed an approximation of the Lévy–Feller advection–dispersion process by a random walk and finite
difference method. Zhuang et al. [11] proposed a new implicit numerical method and two solution techniques to improve its
convergence order for solving an anomalous subdiffusion equation. Liu et al. [12] presented the numerical simulation for the
3D seepage flow with fractional derivatives in porous media. Sun and Sun [13] studied heat and moisture transport through

⇑ Corresponding author.
E-mail address: [email protected] (F. Liu).

http://dx.doi.org/10.1016/j.amc.2015.01.092
0096-3003/Ó 2015 Elsevier Inc. All rights reserved.
 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938 931

porous textile materials and proposed an uncoupled finite difference method with a semi-implicit Euler scheme in time for
the solution of this system. They proved the existence and uniqueness of the solution of the finite difference system. Bolster
et al. proved a product rule for vector fractional derivatives using Fourier transforms [14]. Currently, the FDM is one of dom-
inant numerical methods for solving FPDEs and some high-order finite difference methods for fractional partial differential
equations attract increasing interests [15–18]. Some researchers considered the Finite Element Method (FEM) or Spectral
Method to handle the space derivative. However both FDM and FEM depend on pre-defined grids/meshes which will lead
to difficulties when solving many complex problems with irregular field domains.
More recently, many researchers have developed a group of meshless (or meshfree) methods and successfully applied
these meshless methods to many fields of computational engineering and science[19,20]. These meshless methods have
demonstrated distinguished advantages. First, meshless methods do not use a mesh for the field approximation. Second,
an adaptive analysis is easily achievable by a meshless method. Third, the meshless methods are usually more accurate than
FDM and FEM because of the use of higher order meshless trial functions. As a result of these unique advantages, meshless
methods have good potential for the simulation of FPDE. Nowadays, some works were reported to handle fractional partial
differential equations by the meshless techniques. Chen et al. [21] used the Kansa method for fractional diffusion equations.
Gu et al. [22] developed a meshless formulation for the non-linear anomalous sub-diffusion equation and investigated impli-
cit meshless approaches based on radial basis function (RBF) and the moving least squares (MLSs) [23,24] to simulate numer-
ically two time fractional diffusion equations. Their results are very encouraged. However, employing meshless methods to
the numerical simulation of FPDEs is still limited to time fractional partial differential equation. To the authors best knowl-
edge, no reference has been seen to his point for solving the space fractional differential equation with a meshless method.
In this paper we consider the following one-dimensional space fractional diffusion equation:
@uðx; tÞ @ a uðx; tÞ
¼j þ f ðx; tÞ; x 2 ½0; L; 0 < t < T; ð1Þ
@t @xa
with zero Dirichlet boundary condition

uð0; tÞ ¼ 0
; 0 < t < T; ð2Þ
uðL; tÞ ¼ 0
and initial condition
uðx; 0Þ ¼ xðxÞ; x 2 ½0; L; ð3Þ
@a
where denotes the Riemann–Liouville fractional derivative of order að1 < a < 2Þ with respect to the variable x, which is
@xa
defined by [25] as:
Z
@ a uðx; tÞ 1 @2 x
uðn; tÞ
¼ dn;
@x a Cð2  aÞ @x2 0 ðx  nÞa1
and CðÞ is the gamma function. In this paper we suppose that the solution of the space fractional differential Eq. (1)–(3) sat-
isfies uðx; Þ 2 C 2 ½ð0; TÞ.
The objective of this paper is to develop a new meshless approach based on the meshless Point Interpolation Method
(PIM) for the spatial discretization of space FPDEs. In Section 2, the governing equation of the space FPDE is given. In
Section 3, the PIM is briefly discussed. In Section 4 two full-schemes of the space fractional differential equation with
one-order accuracy or two-order accuracy for time are proposed. Section 5 discusses numerical results to testify our
approaches. The performance of the newly developed meshless approaches for space fractional differential equation is
investigated. Finally Section 6 presents the main conclusions of the work.

2. Meshless methods for space fractional differential equation

2.1. Construction of the PIM meshless shape function

Consider an unknown scalar function of a field variable uðxÞ in domain ½0; L. The PIM interpolation [19] of uðxÞ is defined
at x as
8 9
> a1 >
>
> >
>
Xm >
< a2 >
=
uðxÞ ¼ pj ðxÞaj ¼ fp1 ðxÞp2 ðxÞ . . . pm ðxÞg . ¼ pT ðxÞa; ð4Þ
>
> .
. >
>
j¼1 >
> >
: > ;
am
where pi ðxÞ is the given monomial of the spatial coordinate, m is the number of the monomials and ai is the coefficient of
pi ðxÞ to be confirmed, which will be given in (12) later. The one-dimensional complete polynomial of order p can be repre-
sented by the following form

pT ðxÞ ¼ f1xx2 . . . xp g; ð5Þ

932 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938

and the number of basis functions satisfies m ¼ p þ 1. To solve for the coefficient ai in (4), a local support domain of the point
x should be considered, which includes n nodes enclosing the point x. In the traditional PIM, the number of nodes selected in
the local support domain always equals to the number m of basis functions, i.e. n ¼ m. In Eq. (4), the coefficients ai can be
obtained by letting uðxÞ equal to the real values on the n nodes, i.e.
X
m
u1 ¼ ai pi ðx1 Þ ¼ a1 þ a2 x1 þ a2 x21 þ    þ am xp1
i¼1
X
m
u2 ¼ ai pi ðx2 Þ ¼ a1 þ a2 x2 þ a2 x22 þ    þ am xp2
i¼1
..
.
X
m
un ¼ ai pi ðxn Þ ¼ a1 þ a2 xn þ a2 x2n þ    þ am xpn :
i¼1

The above formula can be written in the following matrix form:

Us ¼ Pm a; ð6Þ
T T
where Us ¼ ðu1 u2 . . . un Þ is the vector of function values on the nodes, a ¼ ða1 a2 . . . am Þ is the vector of unknown coefficients,
and
2 3
1 x1 x21    xp1
6 7
61 x2 x22  xp2 7
6 7
Pm ¼ 6 . .. .. .. .. 7
6 .. . . . . 7
4 5
1 xn x2n    xpn
is the so called moment matrix. Because n ¼ m in PIM, the matrix Pm is square with dimensions n  n or m  m.
Solving Eq. (6), the coefficient vector a to be determined

a ¼ P1
m Us : ð7Þ
It should be noted here that the coefficients ai are constants for the point x. Because if the n interpolation nodes are
unchanged, even if the interpolation point is changed, the matrix Pm determined by the n field nodes is a constant matrix.
Substituting (7) into (4) and noting n ¼ m, it can be obtained that
X
n
uðxÞ ¼ pT ðxÞP1
m Us ¼ /i ðxÞui ¼ UT ðxÞUs ;
i¼1

where UðxÞ is the vector of shape functions defined by

T
UT ðxÞ ¼ pT ðxÞP1
m ¼ ð/1 ðxÞ/2 ðxÞ . . . /n ðxÞÞ : ð8Þ
The derivative of the shape functions of order l (l being a positive integer number or positive real number) can be easily
obtained
0 ðaÞ
1
/1 ðxÞ
B ðaÞ C
B / ðxÞ C @ a pT ðxÞ
UðaÞ ðxÞ ¼ B C
2
B. C¼ P1
m ;
B. C @xa
@. A
/ðnaÞ ðxÞ
ðaÞ
where /i ðxÞ means shape function /i ðxÞs Riemann–Liouville fractional derivative of order a when a is a positive real num-
ber. It has been proven that the PIM shape functions given in Eq. (8) satisfy the Kronecker delta condition [19], which makes
it easy to enforce the Dirichlet boundary conditions in the meshless method based on the PIM shape functions.

2.2. Meshless approaches

Firstly the problem domain is discretized by properly distributed field nodes. The PIM shape functions obtained in (8) are
used to approximate uðx; tÞ. Assume that there are M þ 1 field nodes x0 < x1 <    < xM , including M  1 internal (domain)
nodes and two boundary nodes. Suppose that fxl ; l ¼ i1 ; i2 ; . . . ; in g are the nodes selected in the support domain of xi to construct
h 2 R i
a x uðn;tÞ 
PIM shape functions, and Di ¼ fi1 ; k2 ; . . . ; in g. Denoting ui ðtÞ ¼ uðxi ; tÞ; f i ðtÞ ¼ f ðxi ; tÞ and @ @xuiaðtÞ ¼ Cð2
1 @
aÞ @x2 0 ðxnÞa1 dn  , the
x¼xi
following M þ 1 equations are satisfied,
 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938 933

8 @u ðtÞ @ a ui ðtÞ
< @t ¼ j @xa þ f i ðtÞ;
> i ¼ 1; 2; . . . ; M  1;
i

u0 ðtÞ ¼ 0;
>
:
uM ðtÞ ¼ 0;
@ a u ðtÞ P ðaÞ
where @xja ¼ i2Di ui ðtÞUi ðxj Þ. Because PIM shape functions given in Eq. (8) satisfy the Kronecker delta condition, then
P
i2Di ui ðtÞUi ðxj Þ ¼ uj ðtÞ.

3. Two fully-discrete schemes for time discretization

We discretize time by time instants as follows

T
tn ¼ ns; n ¼ 0; 1; 2; . . . ; N; s¼ ;
N
where s is the time step, and construct two full-discrete schemes for the space fractional differential Eq. (1) with boundary
conditions (2) and initial condition (3).

3.1. Scheme 1: first-order accuracy in time

Let unj be the approximation for uðxi ; tn Þ. Then applying the first difference quotient to approximate the time derivative,
the following fully-discrete scheme can be obtained
X
unþ1
j  js unþ1
i UiðaÞ ðxj Þ ¼ unj þ sf nþ1
j ; j ¼ 1; 2; . . . ; M  1: ð9Þ
i2Di

Using Taylor’s theorem it can be shown that scheme (9) is first-order accurate in time.

3.2. Scheme 2: second-order accuracy in time

Applying Crank–Nicolson scheme to approximate the time derivative and using Taylor’s theorem, the following full-
discrete scheme can be obtained
js X ðaÞ js X ðaÞ s nþ1

unþ1
j  unþ1
i Ui ðxj Þ ¼ unj þ uni Ui ðxj Þ þ fj þ t ni ; j ¼ 1; 2; . . . ; M  1; ð10Þ
2 i2Di
2 i2Di
2

which is with two-order accuracy for time.

4. Numerical results

The following formula about fractional order calculus of power function [25]:
Cðc þ 1Þ
a Dt
a
ðt  aÞc ¼ ðt  aÞca ; ða < 0; c > 1Þ or ð0 6 m 6 a < m þ 1; c > mÞ;
Cðc  a þ 1Þ
and the Leibniz rule for the fractional order derivative [25] are used in this section to obtain the following result.
X1  
@a a ðkÞ @ ak
a ð/ðtÞwðtÞÞ ¼ / ðtÞ ak wðtÞ:
@t k¼0
k @t

4.1. Example 1

Consider the following fractional differential equation

@uðx; tÞ @ 1:85 uðx; tÞ

¼ þ f ðx; tÞ; x 2 ½0; 1; t 2 ð0; 0:8; ð11Þ
@t @x1:85
with boundary conditions

uð0; tÞ ¼ 0
; t 2 ð0; 0:8; ð12Þ
uð1; tÞ ¼ 0
and initial condition
uðx; 0Þ ¼ 0; x 2 ½0; 1; ð13Þ
934 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938

where the source term is f ðx; tÞ ¼ 3:2t2:2 x3:2 ð1  xÞ2  t 3:2 ½Cð4:2Þ=Cð2:35Þð1  xÞ2 x1:35  3:7Cð4:2Þ=Cð3:35Þð1  xÞx2:35 þ
3:7Cð4:2Þ=Cð4:35Þx3:35 . It can be easily proven that the exact solution for this problem is uðx; tÞ ¼ t3:2 x3:2 ð1  xÞ2 .
Following error norms are defined as error indicators in this paper
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
.XNþ1
  XNþ1 2 2
e1 ¼ max uexact
i  unum
i
; e0 ¼
i¼1
ðuexact
i  unum i Þ i¼1
ðuexact
i Þ :
16i6Nþ1

In which e1 and e0 are the error norms for the solution, uexact
i and unum
i are exact and numerical solutions, respectively, for
interest point i.

4.1.1. Results for scheme 1

The problem domain ½0; 1 is firstly discretized by regularly distributed field nodes, i.e. xi ¼ ih; i ¼ 0; 1; 2; . . . ; 50. The pro-
posed full-scheme (9) is used to simulate this space fractional differential equation. Table 1 lists the detailed results of the
computational errors for different time steps. The convergence rates, R, of the two errors regarding to time steps are also
listed in the same table. The convergent rate is calculated as:
eðs1 Þ
Re ¼ :
eðs2 Þ
From Table 1, it can be seen that the convergence rates are close to
eðs1 Þ s1
Re ¼ ¼ :
eðs2 Þ s2
It can be concluded that the order of convergence in time of the present meshless method is OðsÞ, which is in good agreement
with the results obtained by the theoretical analysis.
The problem domain ½0; 1 is discretized using irregularly distributed field nodes as shown in Table 2 and the proposed
full-scheme (9) is used to simulate this space fractional differential equation. Table 3 lists the detailed results of the com-
putational errors for different time steps. The convergence rates, R, of the two errors regarding to time steps are also listed
in the same table.

4.1.2. Results for scheme 2

The problem domain ½0; 1 is firstly discretised by regularly distributed field nodes, i.e. xi ¼ ih; i ¼ 0; 1; 2; . . . ; 50. The pro-
posed full-scheme (10) is used to simulate this space fractional differential equation. Table 4 lists the detailed results of com-
putational errors for different time steps. From Table 4, it can be seen that the convergence rates are close to
 
eðs1 Þ s1 2
Re ¼ ¼ :
eðs2 Þ s2
It can be concluded that the order of convergences in time of the present meshless method is Oðs2 Þ.
The problem domain ½0; 1 is discretized using irregularly distributed field nodes as shown in Table 5 and the proposed
full-scheme (10) is used to simulate this space fractional differential equation. Table 6 lists the detailed results of the com-
putational errors for different time steps. The convergence rates, R, of the two errors regarding to time steps are also listed in
the same table.

4.2. Example 2.

Consider the following fractional differential equation

@uðx; tÞ @ a uðx; tÞ
¼j ; x 2 ½0; 1; t > 0; ð14Þ
@t @xa
with boundary conditions

uð0; tÞ ¼ 0
; t > 0; ð15Þ
uð1; tÞ ¼ 0
and initial condition

Table 1
The error obtained by meshless approach at t ¼ 0:8 (Regular nodal distribution).

s e1 Re max e0 Re0
0.2 1.289e3 – 9.783e2 –
0.1 6.719e4 1.918 5.082e2 1.925
0.05 3.368e4 1.995 2.531e2 2.008
 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938 935

Table 2
Coordinates for 51 irregular distributed field nodes.

0 0.025 0.037 0.06 0.083 0.110 0.127 0.144 0.165 0.178

0.205 0.221 0.247 0.263 0.281 0.310 0.325 0.343 0.361 0.387
0.410 0.422 0.441 0.467 0.476 0.507 0.523 0.542 0.568 0.583
0.610 0.624 0.647 0.662 0.688 0.710 0.723 0.746 0.763 0.782
0.810 0.824 0.842 0.868 0.883 0.910 0.923 0.944 0.963 0.982
1

Table 3
The error obtained by meshless approach at t ¼ 0:8 (Irregular nodal distribution).

s e1 Re max e0 Re0
0.2 1.313e3 – 9.986e2 –
0.1 6.952e4 1.889 5.287e2 1.889
0.05 3.599e4 1.932 2.727e2 1.939

Table 4
The error obtained by meshless approach at t ¼ 0:8 (Regular nodal distribution).

s e1 Re max e0 Re0
0.4 3.255e4 – 2.502e2 –
0.2 7.475e5 4.355 5.569e3 4.493
0.1 1.162e5 6.433 7.777e4 7.161

Table 5
Coordinates for 101 irregular distributed field nodes

0 0.011 0.022 0.033 0.044 0.056 0.064 0.073 0.085 0.093

0.102 0.111 0.123 0.131 0.144 0.153 0.166 0.171 0.183 0.197
0.205 0.212 0.221 0.235 0.243 0.256 0.261 0.282 0.288 0.292
0.303 0.311 0.327 0.333 0.341 0.359 0.364 0.371 0.386 0.392
0.407 0.412 0.425 0.431 0.448 0.455 0.461 0.478 0.483 0.493
0.510 0.514 0.521 0.538 0.543 0.552 0.568 0.571 0.585 0.592
0.610 0.617 0.629 0.632 0.644 0.651 0.667 0.672 0.684 0.691
0.708 0.715 0.727 0.733 0.741 0.753 0.768 0.772 0.789 0.792
0.803 0.818 0.822 0.833 0.845 0.857 0.863 0.877 0.883 0.891
0.904 0.914 0.922 0.933 0.946 0.952 0.967 0.978 0.987 0.992
1

Table 6
The error obtained by meshless approach at t ¼ 0:8 (Irregular nodal distribution).

s e1 Re max e0 Re0
0.4 3.343e4 – 2.583e2 –
0.3 8.440e5 3.961 6.455e3 4.002
0.1 2.028e5 4.162 1.540e3 4.192


1; x ¼ 0:5
uðx; 0Þ ¼ : ð16Þ
0; x – 0:5
Fig. 1 shows the response of the space fractional diffusion Eq. (14) with boundary and initial conditions (15,16) using the
proposed full-discrete scheme (9) for different diffusion coefficients j. It indicates that the solution decays more quickly
while the diffusion coefficient j increases.
Fig. 2 plots the numerical solutions’s diffusion behavior while the time increases.
In Fig. 3, the numerical solutions for different fractional order a are shown. It is well known that the space fractional
diffusion Eq. (1) becomes the traditional diffusion equation when a ¼ 2. The fractional derivative (called left fractional
936 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938

−6
x 10
7
κ=1.6
κ=1.7
κ=1.8
6 κ=1.9
κ=2.0

4
u(x,t=1)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Distance x

Fig. 1. Numerical solutions for a ¼ 1:8 with different diffusion coefficients at t ¼ 1.

−6
x 10
7
t=1.0
t=1.1
t=1.2
6 t=1.3
t=1.4

4
u(x,t)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Distance x
Fig. 2. Numerical solutions for a ¼ 1:8; j ¼ 1:6 while the time increases.

derivative in [25]) is not local operator and accumulates the diffusion process from the left end. From Fig. 3, the lowest line is
not skew as the other lines, which shows that there is not the memory property of fractional derivative, instead the locality
of integer derivative.
 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938 937

−5
x 10
1.6
α=1.6
α=1.7
α=1.8
1.4
α=1.9
α=2.0

1.2

1
u(x,t=1)

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Distance x
Fig. 3. Numerical solutions for j ¼ 1:6 with different fractional order a at t ¼ 1.

5. Conclusions

This paper aims to develop an implicit PIM meshless approach for the numerical simulation of a space fractional partial
differential equation. The first-order accurate and second-order accurate schemes for time are developed and compared. Sev-
eral numerical examples with different problem domains are used to investigate accuracy and efficiency of the newly devel-
oped meshless approaches. Both regular and irregular nodal distributions are employed in the studies. It has been found that
the convergence rates of the meshless approach using the first-order and the second-order schemes for time are OðsÞ and
Oðs2 Þ, respectively. It has proven that the newly developed meshless approach is effective for the numerical simulation of
the space FPDEs.

Acknowledgement

This work is supported by the ARC Future Fellowship grant (FT100100172), the National Natural Science Foundation of
China (Grant No. 11101344), the Natural Science Foundation of Fujian (Grant No. 2013J01021) and the Fundamental
Research Funds for the Central Universities (Grant Nos. 2012121001 and 2012121002).

References

[1] V.V. Anh, N.N. Leonenko, Spectral analysis of fractional kinetic equations with random data, J. Stat. Phys. 104 (2001) 1349–1387.
[2] V.V. Anh, N.N. Leonenko, Harmonic analysis of random fractional diffusion-wave equations. Advanced special functions and related topics in
differential equations (Melfi, 2001), Appl. Math. Comput. 141 (1) (2003) 77–85.
[3] A. Saadatmandi, M. Dehghan, A new operational matrix for solving fractional-order differential equations, Comput. Math. Appl. 59 (2010) 1326–1336.
[4] A. Saadatmandi, M. Delghan, A tau approach for solution of the space fractional diffusion equation, Comput. Math. Appl. 62 (2011) 1135–1142.
[5] A. Saadatmandi, M. Dehghan, M. Azizi, The Sinc–Legendre collocation method for a class of The Sinc–Legendre collocation method for a class of
fractional convection-diffusion equation with variable coefficients, Commun. Nonlinear Sci. Numer. Simul. 17 (2012) 4125–4136.
[6] F. Liu, V. Anh, I. Turner, P. Zhuang, Numerical solution for the solute transport in fractal porous media, ANZIAM J. 45 (E) (2004) 461–473.
[7] F. Liu, V. Anh, I. Turner, Numerical solution of the space fractional Fokker–Planck equation, J. Comput. Appl. Math. 166 (2004) 209–219.
[8] M.M. Meerschaert, C. Tadjeran, Finite difference approximations for fractional advection–dispersion flow equations, J. Comput. Appl. Math. 172 (2004)
65–77.
[9] M.M. Meerschaert, H. Scheffler, C. Tadjeran, Finite difference methods for two-dimensional fractional dispersion equation, J. Comput. Phys. 211 (2006)
249–261.
[10] Q. Liu, F. Liu, I. Turner, V. Anh, Approximation of the Lévy–Feller advection–dispersion process by random walk and finite difference method, J. Comput.
Phys. 222 (2007) 57–70.
[11] P. Zhuang, F. Liu, V. Anh, I. Turner, New solution and analytical techniques of the implicit numerical method for the anomalous subdiffusion equation,
SIAM J. Numer. Anal. 46 (2) (2008) 1079.
[12] Q. Liu, F. Liu, I. Turner, V. Anh, Numerical simulation for the 3D seepage flow with fractional derivatives in porous media, IMA J. Appl. Math. 74 (2009)
201–229.
[13] Wei Sun, Zhi-zhong Sun, Finite difference methods for a nonlinear and strongly coupled heat and moisture transport system in textile materials,
Numer. Math. 120 (1) (2012) 153–187.
938 Q. Liu et al. / Applied Mathematics and Computation 256 (2015) 930–938

[14] D. Bolster, M.M. Meerschaert, A. Sikorskii, A product rule for vector fractional derivatives, Fract. Calc. Appl. Anal. 15 (3) (2012) 463–478.
[15] H.F. Ding, C.P. Li, Y.Q. Chen, High-order algorithms for Riesz derivative and their applications (I), Abstr. Appl. Anal. (2014). Article ID 653797.
[16] H.F. Ding, C.P. Li, Y.Q. Chen, High-order algorithms for Riesz derivative and their applications (II), J. Comput. Phys. (2014). <http://dx.doi.org/10.1016/j.
jcp.2014.06.007>.
[17] C.P. Li, H.F. Ding, Higher order finite difference method for the reaction and anomalous-diffusion equation, Appl. Math. Model. 38 (15–16) (2014)
3802–3821.
[18] A. Mohebbi, M. Abbaszadeh, M. Dehghan, A high order and unconditionally stable scheme for the modified anomalous fractional subdiffusion equation
with a nonlinear source term, J. Comput. Phys. 240 (2013) 36–48.
[19] G.R. Liu, Y.T. Gu, Assessment and applications of point interpolation methods for computational mechanics, Int. J. Numer. Methods Eng. 59 (10) (2004)
1373–1397.
[20] M. Dehghan, A. Ghesmati, Numerical simulation of two-dimensional sine-Gordon solitons via a local weak meshless technique based on the radial
point interpolation method (RPIM), Comput. Phys. Commun. 181 (2010) 772–786.
[21] W. Chen, L. Ye, H. Sun, Fractional diffusion equations by the Kansa method, Comput. Math. Appl. 59 (5) (2010) 1614–1620.
[22] Y.T. Gu, P. Zhuang, F. Liu, An advanced implicit meshless approach for the non-linear anomalous subdiffusion equation, Comput. Model. Eng. Sci. 56 (3)
(2010) 303–334.
[23] Q. Liu, Y.T. Gu, P. Zhuang, F. Liu, Y. Nie, An implicit RBF meshless approach for time fractional diffusion equations, Comput. Mech. 48 (2010) 1–12.
[24] P. Zhuang, Y.T. Gu, F. Liu, I. Turner, P. Yarlagadda, Time-dependent fractional advection–diffusion equations by an implicit MLS meshless method, Int. J.
Numer. Methods Eng. 88 (2010) 1346–1362.
[25] I. Podlubny, Fractional Differential Equations, Academic Press, 1999.