Available online at www.sciencedirect.

com

ScienceDirect
Mathematics and Computers in Simulation 121 (2016) 133145
www.elsevier.com/locate/matcom

Original articles

A Greens function approach for the numerical solution of a class of

fractional reactiondiffusion equations
Eliseo Hernandez-Martinez a, , Francisco Valdes-Parada b , Jose Alvarez-Ramirez b ,
Hector Puebla c , Epifanio Morales-Zarate a
a Facultad de Ciencias Qumicas, Universidad Veracruzana, Xalapa Veracruz, Mexico
b Departamento de Ingeniera de Procesos e Hidraulica, Universidad Autonoma Metropolitana-Iztapalapa, Mexico D.F., Mexico
c Departamento de Energa, Universidad Autonoma Metropolitana-Azcapotzalco, Mexico D.F., Mexico

Received 30 July 2014; received in revised form 13 September 2015; accepted 14 September 2015
Available online 25 September 2015

Highlights
Numerical solutions of fractional RD systems are given using Greens function formulations.
The scheme proposed exhibit global approximation orders of O(h ).
Proposed scheme exhibit better numerical approximation than traditional schemes.

Abstract

Reactiondiffusion equations with spatial fractional derivatives are increasingly used in various science and engineering fields
to describe spatial patterns arising from the interaction of chemical or biochemical reactions and anomalous diffusive transport
mechanisms. Most numerical schemes to solve fractional reactiondiffusion equations use finite difference schemes based on the
GrunwaldLetnikov formula. This work introduces a new systematic approach based on Greens function formulations to obtain
numerical schemes for fractional reactiondiffusion equations. The idea is to pose an integral formulation of the equation in terms
of the underlying Greens function of the fractional operator to subsequently use numerical quadrature to obtain a set of ordinary
differential equations. To illustrate the numerical accuracy of the method, dynamic and steady-state situations are considered and
compared with analytical and numerical solutions via Grunwald finite differences schemes. Numerical simulations show that the
scheme proposed improves the performance and convergence of traditional finite differences schemes based on Grunwald formula.
c 2015 International Association for Mathematics and Computers in Simulation (IMACS). Published by Elsevier B.V. All rights
reserved.

Keywords: Anomalous diffusion; Reactiondiffusion equations; Greens function; Numerical solution

1. Introduction

Reactiondiffusion systems have been used in various science and engineering fields. However, recent research
indicates that the classical diffusion equation is inadequate to model many real situations. For example, in processes

Corresponding author.
E-mail address: [email protected] (E. Hernandez-Martinez).

http://dx.doi.org/10.1016/j.matcom.2015.09.004
0378-4754/ c 2015 International Association for Mathematics and Computers in Simulation (IMACS). Published by Elsevier B.V. All rights
reserved.
134 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

where the particles spread at a faster rate than that predicted by Fickian diffusion and can present asymmetry.
This phenomenon is commonly known as anomalous diffusion [23,27]. Fractional reactiondiffusion equations can
describe a large number of physical processes controlled by anomalous diffusion. Some example include, solute
transport [31], transport in plasma turbulence [7], in the pattern formation [13,12], transport at the Earth surface [30],
diffusion through porous materials [3,34]. Lenzi et al. [18] showed that the anomalous diffusion is a suitable
framework for modeling transport in catalytic particles. Recently, Zheng et al. [38] considered fractal theory for
modeling diffusion in porous media, finding good agreement between experimental data and theoretical predictions.
On the other hand, Das [6] computed approximate analytical solutions for fractional diffusion equations that can be
used in the modeling of reactiondiffusion in catalytic systems. Liu et al. [21] used a fractional differential equation
to describe anomalous diffusion in a mobile/immobile transport model. Friess et al. [11] showed that non-Fickian
diffusion phenomenon should be considered for describing transport of some alcohol vapors in teflon material.
Different numerical methods have been used to solve both linear and nonlinear fractional reactiondiffusion
equations such as, standard discretization of the fractional-in-space operator [16], Fourier methods [4], homotopy
methods [17], finite element [37] and Grunwald fractional difference (GrFD). Due to its simplicity of computational
implementation, the latter is the most commonly used [22,8,25,5]. However, it is well-known that the discretization
of fractional derivatives by means of GrFD is of first-order accurate O(h) and it may lead to instabilities [33]. For
this reason, several works are focused in the derivation of stable and high-order schemes [9,10,36]. In [19], Li and
Zeng presented a review about GrFD schemes where they discussed their application to different types of fractional
differential equations.
Due to the extensive applications of the fractional models, it is necessary to dispose of stable numerical
methodologies with high approximation order. In this sense, recent studies have demonstrated that numerical schemes
based on Greens functions are alternatives for solving reactiondiffusion equations [35]. In fact, systematic derivation
of nonstandard FD schemes can be obtained by means of Greens function formulations [1,15]. In this work a
new systematic approach for the discretization of fractional reactiondiffusion equations based on Greens function
formulations of the fractional operator is proposed. The numerical solution of proposed scheme is based on method
of lines (MOL), where a RungeKutta method is used for solving the ordinary differential equations (ODE) system
obtained. The paper is organized as follows. First we present the integral formulation of the fractional differential
equation [2,32]. Subsequently, a quadrature rule is used to obtain a set of ordinary differential equations. In addition,
depending on the discretization of the integral, a numerical scheme with equivalent structure to finite differences based
on Grunwald formula can be obtained. To illustrate the numerical accuracy of the method, the results are compared
with the analytical and numerical solutions via GrFD.

2. Preliminaries

For convenience, we start our derivations by presenting some definitions and properties of the fractional integral
and differential operators [2,32]. Later, we describe the derivation of integral formulation for solving fractional
reactiondiffusion systems.
For the study of fractional differential equations, different definitions of the derivatives and integrals have been
proposed [29,28,26]. The fractional integral operator of order > 0 of a function u(x) is defined as [2],
x
1
Ia u(x) = (x z)1 u(x)dz (1)
() a
where Ia is the so-called RiemannLiouville fractional integral operator of order . On the other hand, the fractional
differential operator for a function u(x) is defined as [32],
x
1 d n u(z)
Da u(x) = (x z)n1 dz (2)
(n ) a dz n
where Da is the Caputo derivative of order and n is an integer such that n 1 < n. Note that when = 2 one
recovers the classical diffusion operator. It is assumed that u(x) is a continuously differentiable function and piecewise
smooth, in order for the following considerations to be true [2,32],
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 135

Lemma 1. For > 1, if we assume that u(x) C(a, b) L(a, b), then the Caputo homogeneous differential equation
(Da u(x) = 0) has the following solution
u(x) = c0 + c1 x + c2 x 2 + + cn1 x n1 .

Remark 1. For > 1, if we assume u(x) C(a, b), then Da Ia u(x) = u(x) almost everywhere.

Lemma 2. If > 1 and assuming that u(x) C(a, b) L(a, b), then
Ia Da u(x) = u(x) + c1 + c2 x + + cn1 x n1
for some ci R, i = 1, . . . , n, where n is the smallest integer greater than or equal to .
Anomalous diffusion coupled with chemical reactions may be described by fractional diffusion operators such as,
u(x, t)
= Da u(x, t) f (u(x, t)), x [a, b] (3)
t
with initial condition as
u(x, 0) = (x) (4)
and the boundary conditions
u(a, t)
a + a u(a, t) + a = 0
x
u(b, t)
b + b u(b, t) + b = 0 (5)
x
where f (u(x, t)) is a continuously differentiable function of its arguments and (1, 2] is the order of the derivative.
Eq. (3) arises from a mass balance and a fractional Ficks law where the mass flux is proportional to the fractional gra-
dient 1 u(x, t) [34]. Note that = 2 corresponds to the classical diffusion equation. The case (1, 2) models
a superdiffusive flow in which a cloud of diffusing particles spreads at a faster rate than the classical diffusion model.
For developing the integral formulation, let us rewrite Eq. (3) as
Da u(x, t) = (x, t) x [a, b] (6)
where = u(x,t)
t + f (u(x, t)). For Lemma 2 with n = 2, if one multiplies Eq. (6) by the RiemannLiouville integral
operator, one obtains
Ia Da u(x, t) = u(x, t) + c0 + c1 x = Ia (u, t) (7)
or, equivalently
1
x
u(x, t) = (x z)1 (x, t)dz c0 c1 x. (8)
() a
Evaluating the boundary condition given by Eqs. (5) into Eq. (8) it results that
a a
c0 = c1 a c1 +
a a
b (b z)2 (b z)1 b a
b b
1
c1 = (z, t)dz + (z, t)dz +
b a + bb aa b a ( 1) a () b a
leading to
a
ax+ b b (b z)2 b (b z)1 b a

a
u(x, t) = b a (z, t)dz + (z, t)dz +
ba+ b a b a ( 1) a () b a
a
x
1
+ (x z)1 (z, t)dz . (9)
() a a
136 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

After some algebraic manipulations, Eq. (9) takes the form

a b
b
u(x, t) = ( (x) + 1) (x) + G C (z, x) (z, t)dz (10)
a b a
where
a
x a a
(x) = b a
ba+ b a

and G C (z, x) is the Caputo fractional Greens function,

b (b z)2 (b z)1 (x z)1

(x) (x) + if z < x
b ( 1) () ()

G C (z, x) = 2 1
(11)
b (b z) (x)
(b z)
(x)

if z x.

b ( 1) ()
Eq. (11) is the general Greens function, which, as a whole with Eq. (10), denote the Caputo Greens function
formulation (Caputo-GFF) for fractional reactiondiffusion equation given by Eq. (1). In order to observe the effect of
anomalous transport on Greens function, we consider Dirichlet-type boundary conditions (a = b = 0, a = b =
1), for x = 0.5 and z [0, 1], Fig. 1(a) shows the fractional Greens function distribution in the domain [0, 1], where
we can observe that for 1 < < 2 Greens function is not symmetrical (anomalous diffusion) whereas for = 2.0
the symmetrical profile is recovered (Fickian diffusion). Greens functions tend to zero only at z = 1.0, whereas at
z = 0.0 Greens functions are inversely proportional to , this behavior is compensated with the contribution of the
boundary conditions. For mixed-type boundary conditions (b = a = 0, a = b = 1), Fig. 1(b) shows that Greens
function distribution is dissipated at the Neumann-type boundary, which increase as decreases.
On the other hand, unlike traditional methods for the numerical solution of boundary value problems (e.g., finite
differences),the Caputo-GFF can provide an interpretation of how to carry out the transport and reaction processes.
Some comments regarding the structure of Greens function approach derived above are in order:
(i) Caputo Greens function G C (x, z) can be reduced to the standard Greens functions as 2 and according
to the parameters i and i , it is possible to obtain Greens function for different types of boundary conditions
(e.g., Dirichlet, Neumann, Robin). For instance, if we consider = 2 Greens function is
a b

z a x b if z < x
a b

1
G(z, x) = (12)
b a + bb aa b

x a a
zb if z x
a b
obtaining the integral formulation for the classical reactiondiffusion system with general-type boundary
conditions. This result was reported in [14], indicating that Eq. (11) is the general Greens function for
reactiondiffusion systems with general-type boundary conditions in a slab.
(ii) In the original boundary value problem the diffusive transport is described by the differential operator, while in
the integral formulation the transport phenomena is described through Greens function.
(iii) The proposed scheme identifies that the profile u(x, t) consists of the interaction of the phenomena involved
(accumulation, reaction and boundary conditions) and how they are distributed by Greens function.
(iv) The first two terms on the right side of Eq. (10) describe the influence of inhomogeneous boundary conditions.
(v) The integral contains two contributions: the memory of the system that indicates how the concentration profiles
change over the time and how fractional Greens function distributes the source in the domain (i.e., Greens
function describes the anomalous diffusion phenomena in time and space). When u(x,t) t = 0, the profile of
u(x, t) does not depend on time, then Eq. (10) describes the profiles of Eq. (1) in steady-state.
(vi) The proposed scheme is a systematic methodology that can be applied to different definitions of fractional
derivatives. For example, the RiemannLiouville derivative is given by Da u(x) = (n) dn
x
d x n a (x
1

z)+1n u(z)dz and its properties described in Lemmas 1 and 2 are u(x) = c1 x 1 + c2 x 2 + + cn1 x n1
and Ia Da u(x) = u(x) + c1 x 1 + c2 x 2 + + c N 1 x N 1 , respectively [2]. Following the procedure
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 137

Fig. 1. Distribution of fractional Greens function, for x = 0.5 and domain [0,1]. Greens function obtain of (a) Dirichlet-type and (b) Mixed-type
boundary conditions.

described in Eqs. (6)(11), one can obtain that the RiemannLiouville Greens function formulation (RL-GFF)
is given by

4 x 1 a b b

u(x, t) = (x) + (x) + G R L (z, x) (z, t)dz (13)
2 2 a b a

1 2
where (x) = 1 x1 4
2 x
2 3
, 1 = aa ( 2)a 3 + a 2 , 2 = aa ( 1)a 2 + a 1 , 3 =
b 3 + b2 , = b ( 1)b2 + b1 and G
b ( 2)b 4 b R L (z, x) is the RiemannLiouville fractional Greens
function,

b (b z)2 (b z)1 (x z)1

(x) (x) + if z < x
b ( 1) () ()

G R L (z, x) = 2 1
(14)
b (b z)
(x)
(b z)
(x)

if z x.

b ( 1) ()
138 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

3. Numerical scheme

In this section, the numerical methodologies, Grunwald finite differences scheme and Greens function formula-
tions, for solving fractional reactiondiffusion models given by Eqs. (3)(5) are presented. For this, we consider 1x =
h > 0 as the spatial grid size, 1x = (b a)/ (N + 1), with xi = a + i1x and u i (t) = u(xi , t), for i = 1, 2, . . . , N .

3.1. Grunwald finite differences scheme

Finite difference approximations are commonly used to numerically compute the solution of the fractional
boundary value problem given by Eq. (3). So, in order to compare the proposed methodology, we used a spatial
discretization of -order fractional derivative, by mean of right-shifted Grunwald formula [24], such as
i+1
1
Da u i (t) x,a u i (t) = g,k u ik+1 (t) (15)
h k=0
(k1)
where the weights g,k are given by g,k = () (k+1) . So, if the shifted Grunwald estimates are substituted in
Eq. (3), it is obtained the following
i+1
du i (t) 1
= g,k u ik+1 (t) f (u i (t)), i = 1, 2, . . . , N . (16)
dt h k=0
Note that for the classical diffusion equation, ( = 2), the weights g2,k are g2,0 = g2,2 = 1, g2,1 = 2 and for
k > 2 g2,k = 0. Eq. (16) is a set of N differential equations that can be solved numerically by means of numerical
integration methods to obtain the approximate solution of u(x, t).

3.2. Greens function formulation

For this case, we rewrite the Caputo-GFF of u(x, t) given by (10) as

b
u(x, t) = B(x, t) + G C (z, x) (z, t)dz (17)
a
where the boundary contribution B(x, t) is given by
a b
B(x, t) = ( (x) + 1) (x) (18)
a b
and Greens function G(z, x) is given by either Eq. (11). The proposed numerical scheme is obtained by forcing to
zero the residual of Eq. (17) at the grid points x = xi . That is,
b
u i (t) = B(xi , t) + G C (z, xi ) (z, t)dz, i = 1, . . . , N . (19)
a
In the second step, the integral in Eq. (19) is approximated by a numerical quadrature. For instance, the composite
trapezoidal rule gives
N

u i (t) = ri (t) + h G C (z j , xi ) j (t), i = 1, . . . , N (20)
j=1

where
h
ri (t) = B(xi , t) + [G C (a, xi )a (t) + G C (b, xi )b (t)] . (21)
2
Let u = (u 1 , u 2 , . . . , u N )T , r = (r1 , r2 , . . . , r N )T and = (1 , 2 , . . . , N )T . Then, Eq. (21) can be rewritten
as
u(t) = r(t) + hG(t) (22)
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 139

where the components of the N N -dimensional matrix G are gi j = G C (x j , xi ), i, j = 1, . . . , N . Substituting

(t) = du(t)
dt + f(u(t)), we obtain the following set of N initial-value differential equations

du(t) (u(t) r(t))

= G1 f(x, u(t)) (23)
dt h
with initial conditions u(t0 ) = ((x1 ), (x2 ), . . . , (x N ))T . Here, one can use a standard package for integration of
initial value problems to obtain the approximate solution u(t). The MOL technique is based on the discretization of
the spatial operator (using schemes discussed above), leading to a set of ODE. So, the system reduced is integrated
in time using an ODE code such as, RungeKutta methods. The advantage of the MOL technique is that temporal
accuracy can be specified by the users, therefore the error checking, robustness, order selection and time-step adaptive
features available in sophisticated ODE codes can be applied to the time integration of the partial differential equations.
For example, Liu et al. [20], proposed the MOL for solving of fractional FokkerPlanck equation, where backward
differentiation formulas were used. Their results showed a good agreement between the numerical solution and
analytic solution. Other alternative is to perform a full discretization that includes the temporal derivative. In this
way, let = 1t be a time step, tk = t0 + k and u ik = u i (tk ). For example, assuming an implicit Euler discretization
k)
k k1
of the time derivative in Eq. (22); namely, du(t
dt
u u
. This leads to the following expression:

h
uk = rk + G uk uk1 + f(x, uk ) , k = 1, 2, . . . , (24)

which, for each time tk , can be solved for the vector uk by means of, e.g., fixed-point or NewtonRaphson iterations.
On the other hand, considering the integral formulation given by Eq. (10) and following the ideas of the Grunwald
formula, one can propose a backward scheme with an equidistant spatial grid size, obtaining
b
xi xi+1 b a xi a aa b
xi+1
u i (t) = b + G C (z, x) (z, t)dz (25)
xi+1 a + b a
a a xi+1 a + bb a
a b a

with
xi a aa b (xi+1 z)2 (xi+1 z)1 (xi z)1

if z < x


xi+1 a + b +
b ( 1) () ()


b
G C (z, x) = a (26)
xi a a b (x i + 1 z)2 (xi+1 z)1



b if z x.
b ( 1) ()

x
i+1 a + b

Now, one can perform the discretization of the integral in (25) by means of numerical approximations. Using the
du (t)
trapezoidal rule to approximate the integral and substituting (z j , t) = dtj + f (u j (t)), one obtains the following
nonlinear differential system
N
du j (t)

u i (t) = ri (t) + h G C (z j , xi ) + f (u j (t)) , i = 1, . . . , N (27)
j=1
dt

where
b
xi xi+1 a
b xi a aa b
ri (t) = b a b a
xi+1 a + b a a xi+1 a + b a b
h du a (t) h du b (t)

+ + f (u a (t)) G C (a, xi ) + + f (u b (t)) G C (b, xi ).
2 dt 2 dt
Let u = (u 1 , . . . , u N )T , r = (r1 , . . . , r N )T and f(u(t)) = ( f (u 1 (t)), f (u 2 (t)), . . . , f (u N (t)))T . Then, Eq. (27)
can be rewritten as

du(t)
u(t) = r(t) + hG + f(u(t)) (28)
dt
140 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

Fig. 2. Exact and numerical solutions for Eq. (30), (a) profile u(x) and (b) maximum error vs. number of nodes.

where G is a lower triangular matrix of N N with components di, j = G C (x j , xi ), i, j = 1, . . . , N . Eq. (28)

can be presented as Eq. (23) and can be solved using numerical methods. Notice that the scheme (28) is equivalent
to the scheme (16), where, in order to calculate u i , both schemes consider the values of [u a , u 1 , . . . , u i , u b ] and the
weights, g,k / h in the GrFD scheme are replaced for G1 / h in Greens function.

4. Numerical examples

In this section we explore the ability of the proposed scheme to produce accurate solutions for fractional reaction
diffusion equations. To this end, we consider the integral formulation based on Caputo and RiemannLiouville
definitions, given by Eqs. (10)(11) and (13)(14), respectively. The ODE system was integrated by using
RungeKuttaFehlberg subroutine from Matlab package and the numerical results are compared with a classical
GrFD scheme and the corresponding analytical solution. As an index for measuring approximation errors, the absolute
maximum error E m was considered, which is defined as
E m = max |u e (xi ) u ap (xi )| (29)
where u ap (xi ) is the numerical approximation and u e (xi ) is the analytical solution.
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 141

Fig. 3. Solution for Eq. (30) with homogeneous Dirichlet-type boundary conditions; (a) Exact and approximate solutions and (b) maximum error
for integral and finite difference schemes.

Table 1
Approximation order for different values of in steady state.

Case Finite differences Caputo-GFF RiemannLiouville-GFF

1 1.8 0.9876 1.7154 1.7650
2 1.5 0.9950 1.4750 1.4707
3 1.2 0.9817 1.1957 1.1858

4.1. Steady state simulations

Consider the steady state fractional equation:

d u(x)
= f (u(x)) x [0, 1]. (30)
dx
To illustrate the effect of the fractional order () in the numerical accuracy of the proposed schemes, we study three
cases: = 1.8, = 1.5 and = 1.2.
142 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

6 x 1.8 +
Fig. 4. (a) Exact and numerical solutions for Eq. (30) with = 1.2 and f (x) = x 3 + x 2 (2.8) 2 0.8 u(x), (b) maximum error
(1.8) x
vs. number of nodes.

4.1.1. Case 1
Consider Eq. (30) with = 1.8, D(x) = (2.2) 6 x , f (x) = u(x) + x (x 1) D(x)
2.8 3 1 and Dirichlet-type

boundary conditions, u(0) = 0 and u(1) = 1. The exact solution for this problem is u(x) = x 3 . For 100 nodes,
Fig. 2(a) shows the profile u(x) obtained from Greens function formulation using Caputo and RiemannLiouville
definitions and traditional Grunwald finite differences. The numerical solutions compare well (E m < 102 ) with the
exact analytic solution to the fractional differential equation. Fig. 2(b) shows the absolute maximum error (E m ) as
function of the number of nodes (N ), the approximation errors are smaller for Greens function schemes than for
Grunwald finite differences. Notice that, for GrFD the approximation order is O(h), whereas the for Caputo-GFF is
O(h 1.71 ) and RiemannLiouville-GFF scheme is O(h 1.76 ), indicating better numerical accuracy for GrFD.

4.1.2. Case 2
Let us fix now = 1.5, f (x) = 128
x 3.5
7
64 2.5

5
x with non-homogeneous Neumann-type boundary conditions
(du(0)/du = 0 and u(1) = 2) and exact solution as u(x) = x 4 (x + 1). The numerical and analytical profiles u(x)
are shown in Fig. 3(a). The approximation errors are shown in Fig. 3(b); one can observe that the approximation order
calculated with GrFD converges to O(h), whereas for Greens function formulations it converges to O(h 1.5 ). As
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 143

Fig. 5. Dynamic solution for Eq. (3), (a) nonhomogeneous and (b) homogeneous boundary conditions.

in the previous case the minor E m is for Caputo-GFF scheme. However, the approximation order for RL-GFF and
Caputo-GFF schemes are the same.

4.1.3. Case 3
Consider Eq. (30) with = 1.2, f (x) = x 3 + x 2 (2.8)
6 2
x 1.8 + (1.8) x 0.8 u(x) with homogeneous boundary
conditions. The exact solution is u(x) = x (1 x). Similar to the previous examples the GFF schemes show major
2

performance than GrFD Fig. 4(a) and (b). However, it is appreciated that the approximation order calculated with
numerical schemes based on Greens functions decreases with the fractional order of the differential equation; in
other words, the approximation order for Greens function finite differences schemes is O(h ) (Table 1).

4.2. Dynamical simulations

Consider Eq. (3) with = 1.5, f (x, t) = et 128

7
with boundary conditions u(0, t) = 0, u(1, t) = 2et and
initial condition u(x, 0) = (x 1)x 4 . For this problem the analytical solution is u(x, t) = et x 4 (x 1). We used the
scheme given in (21) to obtain a set of ODEs. A RungeKuttaFehlberg method with 1t = 0.01 was used to integrate
the resulting set of equations. For N = 100, Fig. 5(a) shows the profile u(x, t) considering five different times. It is
144 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145

appreciated that both GFF schemes show better performance than Grunwald finite differences. For instance, the error
(E m ) in t = 1.0 is 0.019, 0.0044 and 0.0060 for GrFD, Caputo-GFFand RL-GFF schemes, respectively.
Next, let us consider the case in which = 1.8, f (x) = et (3.2)
24 6
x 2.2 (2.2) x 1.2 + x 4 x 3 with homo-
geneous boundary conditions. The exact solution for this problem is u(x) = et x 3 (x 1). We consider the same
conditions from the previous example, Fig. 5(b) shows the analytical and numerical solutions. It is noted that the GFF
formulations yield smaller approximation errors than the GrFD scheme. In fact, for t = 1.0 the maximum error is
smaller than 0.01% for the two GFF schemes whereas for the GrFD the error is four times smaller than that obtained
from GFF schemes.
In general, the numerical schemes based on Greens function formulations exhibit convergence O(h ), whereas
that GrFD exhibit an O(h). In addition, the approximation order of the proposed schemes can be increased if we
incorporate sequence acceleration methods (e.g. Richardson extrapolation), these methods have been recently applied
to proposed schemes to increase the approximation order [31,33].

5. Conclusions

In this paper we proposed a new approach for the numerical solution of fractional reactiondiffusion equations.
The numerical scheme is based on Greens function formulations of the fractional operator, yielding an integral form
of the equation solution. In a second step, the integral equation is approximated by means of standard quadratures
(e.g., composite trapezoidal rule), yielding a set of ordinary differential equations that is numerically solved by
efficient algorithms (e.g., RungeKuttaFehlberg). Numerical tests with fractional equations having unique analytical
solution were used to illustrate the performance of the numerical scheme, showing that Greens function approaches
are superior in approximation error and convergence rate to finite difference schemes based on Grunwald formula.

References

[1] J. Alvarez-Ramirez, F.J. Valdes-Parada, J. Alvarez, J.A. Ochoa-Tapia, A Greens function formulation for finite-differences schemes, Chem.
Eng. Sci. 62 (2009) 30833091.
[2] Z. Bai, H. Lu, Positive solutions for boundary value problems of nonlinear fractional differential equation, J. Math. Anal. Appl. 311 (2005)
495505.
[3] B. Berkowitz, H. Scher, Anomalous transport in random fracture networks, Phys. Rev. Lett. 79 (1997) 40384041.
[4] C.M. Chen, F. Liu, I. Turner, V. Anh, A Fourier method for the fractional diffusion equation describing subdiffusion, J. Comput. Phys. 227
(2007) 886897.
[5] M. Cui, Compact finite difference method for the fractional diffusion equation, J. Comput. Phys. 228 (2009) 77927804.
[6] S. Das, Approximate solution of fractional diffusion equation-revisited, Int. Rev. Chem. Eng. 4 (2012) 501504.
[7] D. del-Castillo-Negrete, B.A. Carreras, V.E. Lynch, Non-diffusive transport in plasma turbulence: A fractional diffusion approach, Phys. Rev.
Lett. 94 (2005) 065003.
[8] Z.Q. Deng, V.P. Singh, L. Bengtsson, Numerical solution of fractional advectiondispersion equation, J. Hydraul. Eng. 130 (2004) 422431.
[9] H. Ding, C. Li, Y. Chen, High-order algorithms for Riesz derivative and their applications (I), Abstr. Appl. Anal. (2014) 117. 653797.
[10] H. Ding, C. Li, Y. Chen, High-order algorithms for Riesz derivative and their applications (II), J. Comput. Phys. 293 (2015) 218237.
[11] K. Friess, J.C. Jansen, J. Pozivil, V. Hanta, V. Hynek, O. Vopicka, E. Drioli, Anomalous phenomena occurring during permeation and sorption
of C1 -C6 alcohol vapors in teflon AF 2400, Ind. Eng. Chem. Res. 52 (2013) 1040610417.
[12] V.V. Gafiychuk, B.Y. Datsko, Pattern formation in a fractional reactiondiffusion system, Physica A 365 (2006) 300306.
[13] B.I. Henry, S.L. Wearne, Existence of turing instabilities in a two-species fractional reactiondiffusion system, SIAM J. Appl. Math. 62 (2002)
870887.
[14] E. Hernandez-Martinez, J. Alvarez-Ramirez, F.J. Valdes-Parada, Formulaciones integrales para ecuaciones reaccion-difusion generalizadas,
Rev. Mex. Ing. Qui. 10 (2011) 363373.
[15] E. Hernandez-Martinez, F.J. Valdes-Parada, J. Alvarez-Ramirez, A Greens function formulation of nonlocal finite-difference schemes for
reactiondiffusion equations, J. Comput. Appl. Math. 235 (2011) 30963103.
[16] M. Ilic, I.W. Turner, F. Liu, V. Anh, Analytical and numerical solutions of a one-dimensional fractional-in-space diffusion equation in a
composite medium, Appl. Math. Comput. 216 (2010) 22482262.
[17] H. Jafari, S. Seifi, Homotopy analysis method for solving linear and nonlinear fractional diffusion-wave equation, Commun. Nonlinear Sci.
14 (2009) 20062012.
[18] E.K. Lenzi, H.V. Ribeiro, J. Martins, M.K. Lenzi, G.G. Lenzi, S. Specchia, Non-Markovian diffusion equation and diffusion in a porous
catalyst, Chem. Eng. J. 172 (2011) 10831087.
[19] C. Li, F. Zeng, Finite differences methods for fractional differential equations, Int. J. Bifurcation Chaos 22 (2012) 1230014.
[20] F. Liu, V. Anh, I. Turner, Numerical solution of the space fractional FokkerPlanck equation, J. Comput. Appl. Math. 166 (2004) 209219.
[21] Q. Liu, F. Liu, I. Turner, V. Anh, Y.T. Gu, A RBF meshless approach for modeling a fractal mobile/immobile transport model, Appl. Math.
Comput. 226 (2014) 336347.
 E. Hernandez-Martinez et al. / Mathematics and Computers in Simulation 121 (2016) 133145 145

[22] M.M. Meerschaert, C. Tadjeran, Finite difference approximations for two-sided space-fractional partial differential equations, Appl. Numer.
Math. 56 (2006) 8090.
[23] R. Metzler, J. Klafter, The random walks guide to anomalous diffusion: A fractional dynamics approach, Phys. Rep. 339 (2002) 177.
[24] K. Miller, B. Ross, An Introduction to the Fractional Calculus and Fractional Differential, Wiley, New York, 1993.
[25] S. Momani, Z. Odibat, A novel method for nonlinear fractional partial differential equations: Combination of DTM and generalized Taylors
formula, J. Comput. Appl. Math. 220 (2008) 8595.
[26] K.B. Oldham, J. Spanier, The Fractional Calculus, Dover Publications, New York, 2006.
[27] P. Paradisi, R. Cesari, F. Mainardi, F. Tampieri, The fractional Ficks law for non-local transport processes, Physica A 293 (2001) 130142.
[28] I. Podlubny, Fractional Differential Equations, Academic Press, San Diego, 1999.
[29] S.G. Samko, A.A. Kilbas, O.I. Marichev, Fractional Integrals and Derivatives: Theory and Dpplications, Gordon and Breach Science, New
York, 1993.
[30] R. Schumer, M.M. Meerschaert, B. Baeumer, Fractional advectiondispersion equations for modeling transport at the Earth surface,
J. Geophys. Res. 114 (2009) 115.
[31] C. Shen, M.S. Phanikumar, An efficient space-fractional dispersion approximation for stream solute transport modeling, Adv. Water Resour.
32 (2009) 14821494.
[32] X. Su, S. Zhang, Solution to boundary-value problems for nonlinear differential equations of fractional order, Electron. J. Differential
Equations 26 (2009) 115.
[33] C. Tadjeran, M.M. Meerschaert, H.P. Scheffler, A second-order accurate numerical approximation for the fractional diffusion equation,
J. Comput. Phys. 213 (2006) 205213.
[34] F.J. Valdes-Parada, J.A. Ochoa-Tapia, J. Alvarez-Ramirez, Effective medium equations for fractional Ficks law in porous media, Physica A
373 (2009) 339353.
[35] F.J. Valdes-Parada, M. Sales-Cruz, J.A. Ochoa-Tapia, J. Alvarez-Ramirez, On Greens function methods to solve nonlinear reactiondiffusion
systems, Comput. Chem. Eng. 32 (2008) 503511.
[36] Y. Zhang, Z. Sunb, H. Liao, Finite difference methods for the time fractional diffusion equation on non-uniform meshes, J. Comput. Phys.
265 (2014) 195210.
[37] Y. Zheng, C. Li, Z. Zhao, A note on the finite element method for the space-fractional advection diffusion equation, Comput. Math. Appl. 59
(2010) 17181726.
[38] Q. Zheng, B. Yu, S. Wang, L. Luo, A diffusivity model for gas diffusion through fractal porous media, Chem. Eng. Sci. 68 (2012) 650655.