Chaos, Solitons and Fractals 41 (2009) 1095–1104

Contents lists available at ScienceDirect

Chaos, Solitons and Fractals

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

Analysis of the solutions of coupled nonlinear fractional

reaction–diffusion equations
V. Gafiychuk a,b, B. Datsko b, V. Meleshko b, D. Blackmore c,*
a
Physics Department, New York City College of Technology, CUNY 300 Jay Street, Brooklyn, NY 11201, USA
b
Institute of Applied Problems of Mechanics and Mathematics, National Academy of Sciences of Ukraine, Naukova Street 3 B, Lviv 79053, Ukraine
c
Department of Mathematical Sciences, New Jersey Institute of Technology, Newark, NJ 07102, USA

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

Article history: This paper is concerned with analysis of coupled fractional reaction–diffusion equations. As
Accepted 28 April 2008 an example, the reaction–diffusion model with cubic nonlinearity and Brusselator model
are considered. It is shown that by combining the fractional derivatives index with the ratio
of characteristic times, it is possible to find the marginal value of the index where the oscil-
latory instability arises. Computer simulation and analytical methods are used to analyze
possible solutions for a linearized system. A computer simulation of the corresponding
nonlinear fractional ordinary differential equations is presented. It is shown that an
increase of the fractional derivative index leads to periodic solutions which become sto-
chastic as the index approaches the value of 2. It is established by computer simulation that
there exists a set of stable spatio-temporal structures of the one-dimensional system under
the Neumann and periodic boundary condition. The characteristic features of these solu-
tions consist in the transformation of the steady state dissipative structures to homoge-
neous oscillations or spatio-temporal structures at certain values of the fractional index.
Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction

Reaction–diffusion (RD) systems arise in many branches of physics, chemistry, biology, ecology, etc. Reviews of the theory
and applications of reaction–diffusion systems can be found in books and numerous articles (see, for example [1–4]). The
popularity of the RD system is driven by the underlying richness of the nonlinear phenomena, which include stationary
and spatio-temporal dissipative structures, oscillations, different types of waves, excitability, bistability, etc. The mechanism
of the formation of such nonlinear phenomena and the conditions for their emergence have been extensively studied during
the last couple of decades. Although the mathematical theory of these phenomena has not been fully developed yet due to
the essential nonlinearity of these systems, from the viewpoint of applied and experimental mathematics, the pattern of pos-
sible phenomena in RD system is more or less understood.
In recent years, there has been a great deal of interest in fractional reaction–diffusion (FRD) systems [5–14] which from
one side exhibit self-organization phenomena and from the other side introduce a new parameter to these systems, which is
a fractional derivative index, and it gives a greater degree of freedom for diversity of self-organization phenomena. At the
same time, the process of analyzing such FRD systems is much more complicated from the analytical and numerical point
of view.
We consider next a FRD system
oa u
s ¼ Kðu; AÞ; ð1Þ
ot a

* Corresponding author.
E-mail address: denis.l.blackmore@njit.edu (D. Blackmore).

0960-0779/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.chaos.2008.04.039
1096 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104

where u ¼ ðu1 ; u2 ÞT ; ui ðx; tÞ 2 R are activator and inhibitor variables, i ¼ 1; 2; the diagonal matrix s ¼ diag½si ; the operator
2 2
Kðu; AÞ ¼ ðl r2 u1  W; L2 r2 u2  Q ÞT ; r2 ¼ oxo 2 ; W; Q ; x; t 2 R are the nonlinear sources of the system modeling their
production rates, si ¼ ~sai ; ~si ; l; L; 2 R are characteristic times and lengths of the system, and A 2 R is an external parameter.
a
The fractional derivatives o uðx;tÞ
ot a
on the left-hand side of Eq. (1) instead of standard time derivatives are the Caputo fractional
derivatives in time of the order 0 < a < 2 and are represented as [16,17]
Z t
oa 1 uðmÞ ðsÞ
a uðtÞ :¼ ds; m  1 < a < m; m 2 N:
ot Cðm  aÞ 0 ðt  sÞaþ1m

In this paper, we always assume that the following conditions are fulfilled on the boundaries 0;lx : (i) Neumann:
 
oui  oui 
¼ ¼0 ð2Þ
ox x¼0 ox x¼lx

(ii) Periodic:
 
oui  oui 
ui ðt; 0Þ ¼ ui ðt; lx Þ;  ¼ ; ð3Þ
ox x¼0 ox x¼lx

with corresponding initial conditions depending on the value of the index a [16–18]. Eq. (1) at a ¼ 1 correspond to standard
RD system and at a–1 they describe phenomena of anomalous diffusion [5–14].

2. Linear stability analysis

Stability of the steady state constant solutions of the system (1) corresponding to homogeneous equilibrium states
Wðu  2 ; AÞ ¼ 0;
1 ; u 1 ; u
Q ðu  2 ; AÞ ¼ 0 ð4Þ
can be analyzed by linearization of the system nearby this solution. In this case the system (1) can be transformed to a linear
system and the solubility of this system leads to the characteristic equation det jkI  Aj ¼ 0; where
2 2
!
ðl k þ a11 Þ=s1 a12 =s1
A¼ 2
; ð5Þ
a21 =s2 ðL2 k þ a22 Þ=s2

k ¼ lpx j; j ¼ 1; 2; . . . ; I is the identity matrix, a11 ¼ W 0u1 , a12 ¼ W 0u2 , a21 ¼ Q 0u1 , and a22 ¼ Q 0u2 (all derivatives are taken at homo-
geneous equilibrium states ðu  2 Þ (condition (4)). As a result, the characteristic equation takes the form of a quadratic equa-
1 ; u
tion k2  tr Ak þ det A ¼ 0; which determines two eigenvalues k1 ; k2 :
In this case, the solution of the linearized vector equation is given by Mittag–Leffler functions [15–17,19,20]
X
1
ðki ta Þk
i ¼
Dui ðtÞ ¼ ui  u Dui ð0Þ ¼ Ea ðki t a ÞDui ð0Þ; i ¼ 1; 2: ð6Þ
k¼0
Cðka þ 1Þ

Using the result obtained in the paper [26], we can conclude that if for any of the roots j argðki Þj < ap=2 the solution has an
increasing function component then the system is asymptotically unstable (Figs. 1 and 2). Analyzing the roots of the char-
acteristic equations, we can see that at 4 det A  tr2 A > 0 eigenvalues are complex and can be represented as k1;2 ¼ kRe  ikIm :
The roots k1;2 are complex inside the parabola (Fig. 1a) and for the standard system the fixed points are spiral sources
ðtr A > 0Þ or spiral sinks ðtr A < 0Þ. The plot of the marginal value a :
2
a ¼ a0 ¼ j argðk1 ; k2 Þj ð7Þ
p
is presented in Fig. 1b. In contrast to the standard system here at Reki > 0 the system could be stable and moreover at
Reki < 0 could be unstable.

Fig. 1. (a) Shaded complex eigenvalue region. (b) Plot of the marginal value of a.
 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104 1097

Fig. 2. Schematic view of the instability domains for: (a) 0 < a 6 1. (b) 1 6 a < 2.

Let us analyze the system solution with the help of Fig. 1a. Consider the parameters which keep the system inside the
parabola. It is a well-known fact, that at a ¼ 1 the domain on the right-hand side of the parabola ðtr A > 0Þ is unstable with
the existing limit cycle, while the domain on the left-hand side ðtr A < 0Þ is stable. By crossing the axis tr A ¼ 0 the Hopf bifur-
cation conditions are seen to apply. In the general case of a : 0 < a < 2 for every point inside the parabola there exists a mar-
ginal value of a0 where the system changes its stability. The value of a is a certain bifurcation parameter which switches the
stable and unstable state of the system. When a < a0 ¼ 2p j argðki Þj the system has oscillatory modes but they are stable.
Increasing the value of a > a0 ¼ 2p j argðki Þj leads to instability. As a result, the domain below the curve a0 , as a function of
tr A is stable and the domain above the curve is unstable.
The mechanism of the system instability can be understood from Fig. 2. For example, for the case a0 > 1, having complex
number ki with Reki < 0 at a ! 2 it is always possible to satisfy the condition j argðki Þj < ap=2, and the system becomes
unstable according to homogeneous oscillations (Figs. 1b, 2b). The smaller the value of tr A, the easier it is to fulfill the insta-
bility conditions.
In contrast to this case, complex values of ki ; with Reki > 0 lead to the system instability for a regular system with a ¼ 1.
However fractional derivatives with a < 1 will stabilize the system if a < a0 ¼ 2p j argðki Þj: This makes it possible to conclude
that fractional derivative equations with a < 1 are more stable than their integer counterparts.

3. Solutions of the coupled fractional ordinary differential equations (FODEs)

Our particular interest here is the analysis of the specific nonlinear system of FRD equations. We consider two very well-
known examples. The first one is the RD system with cubical nonlinearity [2–4] which probably is the simplest one used in
RD systems modeling
Wðu1 ; u2 Þ ¼ u1 þ u31 =3 þ u2 ; Q ¼ u2  bu1  A: ð8Þ

The second example is known as Brusselator model [1] and it describes certain chemical reaction–diffusion processes with a
pair of variables whose concentrations are controlled by nonlinearities
Wðu1 ; u2 Þ ¼ A þ ðb þ 1Þu1  u21 u2 ; Q ðu1 ; u2 Þ ¼ bu1 þ u21 u2 : ð9Þ

Let us first consider the coupled fractional ordinary differential equations (FODEs) with nonlinearities (8) and analyze the
stability conditions for such systems. The plot of isoclines for the system (8) is represented in Fig. 3a. In this case, for the
homogeneous solution, which can be determined from the system of equations W ¼ Q ¼ 0, this is the solution of the cubic
algebraic equation ðb  1Þn  31 =3 þ A ¼ 0: Simple calculation makes it possible to write the expressions required for
1 þ u
analysis
!
ð1 þ u 21 Þ=s1 1=s1
A¼  21 Þ=s1  1=s2 ; det A ¼ ððb  1Þ þ u
; tr A ¼ ð1  u  21 Þ=s1 s2 :
b=s2 1=s2

It is easy to see that if the value of s1 =s2 , in certain cases, is smaller than 1, the instability conditions ðtr A > 0Þ lead to Hopf
bifurcation for a regular system ða ¼ 1Þ [1–3]. In this case, the plot of the characteristic domains, where instability exists, is
shown in Fig. 3b where the middle curve corresponds to a ¼ 1; the inside one to a ¼ 0:5 and the outside curve to a ¼ 1:5.
Linear analysis of the system for a ¼ 1 shows that, if s1 =s2 > 1, the solution corresponds to the intersection of two isoclines,
and it is stable. The smaller the ratio of s1 =s2 , the wider the instability region. Formally, at s1 =s2 ! 0, the instability region in
 1 coinsides with the interval (1, 1) where the null isocline Wðu1 ; u2 Þ ¼ 0 has its increasing part. The maximum value of the
u
curve s1 =s2 on ðu 1 Þ corresponds to the value s1 =s2 ¼ 1 where the system is neutrally stable. These results are widely known in
the theory of nonlinear dynamical systems [1–3].
In the FODEs the conditions of the instability change and we have to analyze the real and the imaginary part of the exist-
ing complex eigenvalues, especially the equation: 4 det A  tr2 A ¼ 4ððb  1Þ þ u  21 Þ=s1  1=s2 Þ2 > 0: In fact,
 21 Þ=s1 s2  ðð1  u
with the complex eigenvalues, it is possible to find the corresponding value of a where the instability condition holds. This
1098 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104

Fig. 3. (a) Null isoclines. (b) Instability domains (shaded regions) in coordinates ðu  1 ; s1 =s2 Þ (dependence of s1 =s2 on u
 1 ). (c) Dependence of a0 on
 21 ðs1 =s2 ¼ 1Þ. (d) Dependence of s1 =s2 on a0 (the domains in coordinates ðu
u  1 ; s1 =s2 Þ correspond to domains represented in (b) and (c).

interval is not correlated with the increasing part of the null isocline of the system. Indeed, omitting a simple calculation, we
2
can write an equation for marginal values of n 1 : u  21 þ s12  2 ss1 ð2b  1Þ þ 1 ¼ 0; and the expression
 41  2ð1 þ ss1 Þu
2 s2 2
qffiffiffiffiffiffiffi
 21 ¼ 1 þ ss1  2 b ss1 estimates the maximum and minimum values of u
u  1 where the system can be unstable at a certain value
2 2

of a ¼ a0 . Let us examine the domain of the FODEs where the eigenvalues are complex for fixed value s1 =s2 ; for example,
s1 ¼ s2 ¼ 1 and b ¼ 2: In this case, tr A ¼ u  21 , det A ¼ 1 þ u  21 , 4 det A  tr2 A ¼ 4 þ 4u  21  u
 41 > 0; which immediately leads to
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffi
the region of existing of complex roots  2 þ 2 2 6 u  1 6 2 þ 2 2 and we can conclude that the instability region,
due to the fractional order of the derivatives, can be much wider than the same region for a ¼ 1. The dependance of the value
of a on u 21 is presented in Fig. 3c.
In this case the plot is obtained at s1 =s2 ¼ 1 and that is why the marginal instability curve is determined for a > 1 (for
a < 1 the system is stable). In this figure the domain below the curve corresponds to stability and above it to instability.
Similar analysis can be provided for tr A > 0 where ðs1 =s2 Þ is smaller than unity. In this case, the instability conditions are
also not correlated with the increasing part of the null isocline W 1 ðu 1 ; u  2 Þ: In this case the plot of a will start not from 1 but
from a certain value smaller than unity. However, it is much better for understanding to get the marginal curve a0 as a func-
tion of s1 =s2 .
Let us analyze the dependence of a0 on the parameter s1 =s2 where the system changes its stability. As it is easy to see from
Fig. 3a and b that the easiest way to reach the instability domain is realized at A ¼ 0 when two isoclines intersect them-
selves at the point (0, 0) (Fig. 3a). Let us consider the dependence of this marginal value of a0 on the parameter s1 =s2 at
A ¼ 0 and determine the plot of this maximum as a functions of a0 on s1 =s2 : Such a curve for the given model is represented
in the Fig. 3d. This curve obtained from (7) corresponds to the dependence of a on tr A (Fig. 1a). Below this curve the system is
unstable, and above it is stable. We may therefore focus our attention on the general form of this curve. At a sufficiently small
value of s1 =s2 , oscillatory instability occurs even for small a < 1. In contrast, at a > 1, the instability conditions could have
place even for those cases when s1 =s2 is sufficiently large. This means that the fractional differential equations, by corre-
sponding combinations of the parameter s1 =s2 , can be stable or unstable practically in the whole region 0 < a < 2:
It should be noted that, even if the eigenvalues are not complex ðkIm ¼ 0Þ, the systems with fractional derivatives can have
oscillatory damping oscillations. Such a situation takes place when 4 det A  tr2 A < 0; tr A < 0; det A > 0 and two eigenvalues
are real and less than zero. In this case, at 1 < a < 2, steady state solutions of the system are stable and any perturbations are
damped. Such a system was considered, for example, in the article [25], where an analytical solution for a fractional oscillator
is obtained. We shall start analyzing possible solutions for this case.
Several examples of linear FODEs were solved analytically by the Adomnian decomposition method, as well as numeri-
cally. The obtained solutions were compared with the analytical solutions obtained by Mittag–Leffler functions (6) and the
results of the work [25].
 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104 1099

Three different solutions are plotted on the same Fig. 4. The results of the computer simulation of linear ordinary differ-
ential equations for (a) stable system and (b) unstable system of one variable u1 are given on the time interval [0, 30]. We can
see that in the stable domain ða < a0 Þ the oscillations are damped, and at a > a0 they grow exponentially. So, the fractional
ordinary differential equations (without space derivatives) of the system have two different modes. The solution is either
asymptotically stable or unstable.
The developed technique of Adomian decomposition [21–24,28] is a powerful method for solving the system of ordinary
differential equations. The effectiveness of this method was demonstrated for solving linear fractional differential equations,
and we used this method here to test the computer program and to compare these analytical results with Mittag–Leffler
functions.
First, we consider the case when the system is asymptotically stable at a < a0 ¼ 2, for example a ¼ 1:7 (Fig. 4a). Three
solutions obtained by these different methods differ little from each other and they show oscillatory damping time behavior
(the model described in [25]). The analytical result obtained by Adomian decomposition is given by the formula
X
1
t ka
u1 ¼ ð1Þk ; ð10Þ
k¼0
Cðka þ 1Þ

which is plotted in Fig. 4a by the empty circled curve (we truncate the series at k ¼ 60). For fairly small times (for example
t < 7) this series can be represented by

u1 ¼ 1  0:64738082  t1:7 þ 0:98657256  101  t 3:4  0:70199112  102  t 5:1 þ 0:296127  103  t 6:8  0:8382759
 105  t 8:5 þ 0:1718481  106  t10:2  0:2686528893  108  t11:9 þ 0:3324725330  1010  t 13:6
 0:3350625811  1012  t15:3 þ 0:2811457252  1014  t 17 :

The considered system, in this particular case, cannot be unstable (Fig. 4a) because the values of k are real and negative with
jargðk1 Þj ¼ p > ap=2. In the limit at a ¼ 2 we have a regular linear oscillator [25], analytical solution of which we get from
Adomian decomposition in the form of power series. The terms of this series completely coincide with the expansion of lin-
ear oscillator solution cos(t).
It should be noted that all damped oscillations solutions of the linear oscillator of fractional order have similar plots.
Quite different is the dynamics of FODEs for a > a0 when the oscillatory mode becomes unstable and leads to increasing
oscillations at a > a0 (Fig. 4b). In this case, the analytical solution obtained by Adomian decomposition for a ¼ 1:3 looks like
u1 ¼ 0:67  0:1716397314  t1:3  0:05663459553  t2:6 þ 0:008906834532  t 3:9 þ 0:000069919977  t5:2
 0:000046921632  t 6:5 þ 0:12923165  105  t7:8 þ 0:5298004324  107  t 9:1  0:2781417477  108  t10:4
þ 0:3895105489  1011  t11:7 þ 0:1811701048  1011  t 13 þ   

The result of taking into account 50 terms in Adomian decomposition expansion makes it possible to represent the solution
in the time interval [0, 30], which is presented in Fig. 4b. As a matter of fact, such a representation makes rather questionable
theoretical sense because our system is essentially nonlinear and does not allow this type of solution. For the initial stage
dynamics or for the linear system, the Adomian decomposition method is very effective. The application of Adomian decom-
position to the nonlinear FDOEs is not successful. Taking into account 10 terms shows that the solution for t up to 6 or 7 and
at higher values of t the discrepancy between the numerical solution and the one obtained by Adomian decomposition in-
creases rapidly.
To demonstrate the non-trivial properties of FODEs, here we consider the nonlinear dynamics of the above mentioned
nonlinear FODEs. For this we used a numerical simulation of the system. Some details of this approach are discussed in
the next section.
For a > a0 , small perturbations of the steady state solution, due to the memory inherent in fractional derivatives, survive
in the process of evolution and grow in amplitude while nonlinear terms of the system (8) restrict the size of these oscilla-
tions. In this case, time dependence of the variables corresponds to the oscillatory solution. (The phase portrait and isoclines
are presented in Fig. 5a–d).

Fig. 4. (a) Damped oscillator solution for oa u=ota ¼ u at a ¼ 1:7; uð0Þ ¼ 1. (b) Increasing time domain oscillations of the linearized system (1) for u1 and
1 1
a ¼ 1:3; A ¼ 0:1; b ¼ 1; s1 ¼ s2 ¼ 1; u1 ð0Þ ¼ ð3AÞ3 ; u2 ð0Þ ¼ ð3AÞ3 þ A. The small filled circled line denotes the Mittag–Leffler solution, the empty circle
line the Adomnian decomposition solution, and the solid gray line the numerical solution.
1100 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104

Fig. 5. Two-dimensional phase portrait (a)–(d) of system (1) with nonlinearities (8) for A ¼ 0:1; b ¼ 1; s1 ¼ s2 ¼ 1; l ¼ L ¼ 0. (a) a ¼ 1:3, (b) a ¼ 1:7,
(c) a ¼ 1:8, (d) a ¼ 1:90.

Analyzing the phase trajectory of the FODEs, we can see that the amplitude of the oscillations increases with increasing a:
At a approaching 2, the oscillations become more complicated and at a ¼ 2 they look quasi-chaotic.
The Brusselator system with nonlinearities (9) has quite similar behavior. In this case, the source terms (9) of the homo-
geneous solution of variables u  2 can be determined from the system of Eq. (4) and the null isoclines are represented in
 1 and u
Fig. 6a. The instability domain regions for different values of the fractional derivative index a in the coordinates ðB; AÞ are
represented in Fig. 6b. The calculation of the derivatives at the homogeneous state makes it possible to identify conditions
for different types of instability explicitly and to consider them in detail. The additional instability condition of the fractional
RD system is determined by

Fig. 6. (a) Nullclines for A ¼ 1; B ¼ 2. (b) Two-dimensional bifurcation diagram domains in coordinates ðA; BÞ for different values of a for k ¼ 0.
 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104 1101

4 det F  tr2 F ¼ 4A2  ðB  1  A2 Þ2 > 0: ð11Þ

As a result, a stationary state ðu  2 Þ becomes unstable, and limit cycle arises in the system. In our case tr F ¼ B  1  A2 ,
1 ; u
det F ¼ A2 > 0 and the stationary state is always unstable for B > 1 þ A2 . This instability is realized for different values of
a (Fig. 6b) Inside the parabola-like curves the system is unstable with wave numbers k ¼ 0, and outside – it is stable. The
domain of the darkest shade corresponds to a ¼ 1.
The phase portrait and isoclines for (9) are presented in Fig. 7a–d. At a K 1:5, we obtain the steady state solution and at
a J 1:5 the steady state oscillation. Increasing the value a leads to complication of the phase paths, and the two-dimensional
phase portrait looks much more complicated (Fig. 7b and d). The attractor of the system of the two coupled nonlinear dif-
ferential equations shows the features of a strange attractor and a ! 2 it corresponds to the attractor of the fourth order dif-
ferential equations determined by the nonlinearities (9).

4. Computer simulation of pattern formation

This section contains a discussion of the results of the numerical study of the initial value problem of the system (1). The
system with corresponding initial and boundary conditions was integrated numerically using explicit and implicit schemes
with respect to time and centered difference approximation for spatial derivatives. The time fractional derivatives were
approximated using two different schemes based on the Riemann–Liouville definition [27], as well as the scheme based
on the of Grünwald–Letnikov definition for 0 < a < 1 and 0 < a < 2 [17,18,28]. In other words, for the system of n fractional
RD equations
C a
o uj ðx; tÞ o2 uj ðx; tÞ
sj a ¼ dj þ fj ðu1 ; . . . ; un Þ; j ¼ 1; n; ð12Þ
ot ox2
where sj ; dj ; fj are certain parameters and nonlinearities of the RD system.
The applied numerical schemes are implicit, and for each time layer they are presented as the system of algebraic equa-
tions solved by the Newton–Raphson technique. Such an approach makes it possible to get the system of equations with

Fig. 7. Two-dimensional phase portraits (a)–(d) of the system (1) with nonlinearities (9) for A ¼ 2; b ¼ 2; s1 ¼ s2 ¼ 1; l ¼ L ¼ 0. (a) a ¼ 1:4, (b) a ¼ 1:5,
(c) a ¼ 1:6, (d) a ¼ 1:7.
1102 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104

band Jacobian for each node and to use the sweep method for the solution of the linear algebraic equations. Calculating the
values of the spatial derivatives and corresponding nonlinear terms on the previous layer, we obtained explicit schemes for
integration. Despite the fact that these algorithms are quite simple, they are very sensitive and require small steps of inte-
gration, and they often yield unsatisfactory numerical results. In contrast, the implicit schemes, in a certain sense, are similar
to the implicit Euler’s method, and they have produced good results in the modeling of fractional reaction–diffusion systems
for different step sizes of integration, as well as for nonlinear functions and the power function of fractional index. It should
be noted that the definition of the fractional derivative in Grünwald–Letnikov form is equivalent to the one in the Riemann–
Liouville method, but for numerical calculations it is much more flexible.
We have considered here the kinetics of formation of dissipative structures for different values of a. These results are pre-
sented in Figs. 8 and 9.
The simulations were carried out for a one-dimensional system on an equidistant grid with spatial step h varying from 0.1
to 0.01 and time step Dt varying from 0.001 to 0.1. We used imposed Neuman (2) or periodic boundary conditions (3). As the
initial condition, we used the uniform state which was superposed with a small spatially inhomogeneous perturbation.
The systems have rich dynamics, including steady state dissipative structures, homogeneous and non-homogeneous
oscillations, and spatio-temporal patterns. In this paper, we focus mainly on the study of general properties of the solutions
depending on the value of a. Similar effects can be achieved by changing the ratio of s1 =s2 at fixed value of a:
As discussed in Section 2, there are two different regions in the parameter A, where the system can be stable or unstable.
In the case of a ¼ 1 the steady state solutions in the form of non-homogeneous dissipative structures are inherent in the
unstable region u  1 2 ð1; 1Þ. Fig. 8a–c show the steady state dissipative structure formation and Fig. 8d–f present the spa-
tio-temporal evolution of dissipative structures, which eventually leads to homogeneous oscillations.
In Fig. 8a–d, the value a increases from 0.1 to 1.5 and on this whole interval the structures are in steady state. This is due
to the fact that a < a0 , where the oscillatory perturbations are damped, and we can see that small oscillations are at the tran-
sition period u1 : With increasing a, the steady state structures change to the spatio-temporal behavior (Fig. 8e and f).

Fig. 8. Numerical solution of the fractional reaction–diffusion Eq. (1) with nonlinearities (8). Dynamics of variable u1 on the time interval (0, 40) for
2
lx ¼ 8; A ¼ 0:1; b ¼ 1; s1 ¼ s2 ¼ 1; l ¼ 0:05; L2 ¼ 1; (a) a ¼ 0:1, (b) a ¼ 0:5, (c) a ¼ 0:99, (d) a ¼ 1:5, (e) a ¼ 1:6, (f) a ¼ 1:8.
 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104 1103

Fig. 9. Numerical solution of the fractional reaction–diffusion Eq. (1) with nonlinearities (9). Dynamics of variable u1 on the time interval (0,20) for
2
lx ¼ 10; A ¼ 2; b ¼ 2; s1 ¼ s2 ¼ 1; l ¼ 0:1; L2 ¼ 10; (a) a ¼ 0:1, (b) a ¼ 0:5, (c) a ¼ 0:99, (d) a ¼ 1:5, (e) a ¼ 1:6, (f) a ¼ 1:8.

The emergence of homogeneous oscillations, which destroy pattern formation (Fig. 8e and f), has deep physical meaning.
The matter is that the stationary dissipative structures consist of smooth and sharp regions of the variable u1 , and the smooth
shape of u2 . Linear system analysis shows that the homogeneous distribution of the variables is unstable according to oscil-
latory perturbations inside the interval of u  1 ; which is much wider than the interval (1, 1). At the same time, smooth dis-
pffiffiffi
tributions at the maximum and minimum values of u1 are  3, respectively. In the first approximation, these smooth
regions of the dissipative structures resemble homogeneous ones and are located inside the instability regions. As a result,
the unstable fluctuations lead to homogeneous oscillations, and the dissipative structures destroy themselves. We can con-
clude that oscillatory modes in such FODEs have a much wider attraction region than the corresponding region of the dis-
sipative structures.
For a wide range of the parameter a, the numerical solutions of the Brusselator problem show similar behavior (Fig. 9a–d).
The stationary solutions emerge practically in the same way. At small a, we see aperiodic formation of the structures, and
approaching a0 , the damping oscillations of the dissipative structures arise. At a0 ¼ 1:7 certain non-stationary structures
arise (Fig. 9e). The increase of a leads to a larger amplitude of pulsation. All these patterns are robust with respect to small
initial perturbations. The further increase of a leads to spatio-temporal chaos (Fig. 9f).
In contrast to the previous case, such non-homogeneous behavior is stable and does not lead to homogeneous oscillations.
The reason is that in the Brusselator model, the dissipative structures are much greater in amplitude and do not have smooth
distribution at the top.
It should be noted that the pulsation phenomena of the dissipative structures is closely related to the oscillation solutions
of the ODE (Figs. 5 and 7). Moreover, the fractional derivative of the first variable has the most impact on the emergence of
oscillations. This can be seen by performing a simulation where the first variable is a fractional derivative and the second one
is an integer. It should be emphasized that the distribution of u2 , within the solution, only shows a small deviation from the
stationary value (that is why this variable is not represented in the figures).

5. Conclusion

Using a linear theory of instability of reaction diffusion systems with fractional derivatives we introduced a new param-
eter – the marginal value a0 – which plays the role of bifurcation parameter. If the fractional derivative index a is smaller
than a0 , the system of FODEs is stable and has oscillatory damping solutions. When a > a0 , the FODEs become unstable,
1104 V. Gafiychuk et al. / Chaos, Solitons and Fractals 41 (2009) 1095–1104

and we obtain oscillatory or even more complex quasi-chaotic solutions. In addition, the stable and unstable domains of the
system were investigated.
By computer simulation of the fractional reaction–diffusion systems we provided evidence that pattern formation in the
fractional case, at a less than a certain value, is practically the same as in the regular case scenario a ¼ 1. At a > a0 , the kinet-
ics of formation becomes oscillatory. When a ¼ a0 , the oscillatory mode arises and can lead to homogeneous or non-homo-
geneous oscillations. In the last case scenario, depending on the parameters of the medium, we can see a rich variety of
spatio-temporal behavior.

References

[1] Nicolis G, Prigogine I. Self-organization in non-equilibrium systems. New York: Wiley; 1977.
[2] Mikhailov AS. Foundations of synergetics. Berlin: Springer-Verlag; 1990.
[3] Kerner BS, Osipov VV. Autosolitons. Dordrecht: Kluwer; 1994.
[4] Lubashevskii A, Gafiychuk VV. The projection dynamics of highly dissipative system. Phys Rev E 1994;50(1):171–81.
[5] Henry BI, Langlands TAM, Wearne SL. Turing pattern formation in fractional activator–inhibitor systems. Phys Rev E 2005;72:026101.
[6] Henry BI, Wearne SL. Fractional reaction–diffusion. Physica A 2000;276:448–55.
[7] Henry BI, Wearne SL. Existence of turing instabilities in a two-species fractional reaction–diffusion system. Siam J Appl Math 2002;62(3):870–87.
[8] Vlad MO, Ross J. Systematic derivation of reaction–diffusion equations with distributed delays and relations to fractional reaction–diffusion equations
and hyperbolic transport equations: application to the theory of Neolithic transition. Phys Rev E 2002;66:061908.
[9] Seki K, Wojcik M, Tachiya M. Fractional reaction–diffusion equation. J Chem Phys 2003;119:2165.
[10] Gafiychuk V, Datsko B. Pattern formation in a fractional reaction–diffusion system. Phys A Statist Mech Appl 2006;365:300–6.
[11] Saxena RK, Mathai AM, Haubold HJ. Fractional reaction–diffusion equations. arXiv:math.CA/0604473 v1 21, April 2006.
[12] Gafiychuk V, Datsko B. Stability analysis and oscillatory structures in time-fractional reaction–diffusion systems. Phys Rev E 2007;75:055201-1–1-4.
[13] Weitzner H, Zaslavsky GM. Some applications of fractional equations. Commun Nonlinear Sci Numer Simulat 2003;8:273–81.
[14] Varea C, Barrio RA. Travelling turing patterns with anomalous diffusion. J Phys Condens Matter 2004;16:5081–90.
[15] Tarasov VE, Zaslavsky GM. Nonholonomic constraints with fractional derivatives. J Phys A Math Gen 2006;39:9797–815.
[16] Samko SG, Kilbas AA, Marichev OI. Fractional integrals and derivatives: theory and applications. Newark, NJ: Gordon and Breach; 1993.
[17] Podlubny I. Fractional differential equations. Academic Press; 1999.
[18] Ciesielski M, Leszczynski J. Numerical treatment of an initial-boundary value problem for fractional partial differential equations. Signal Process
2006;86:2619–31.
[19] Odibat ZM, Shawagfeh NT. Generalized Taylor’s formula. Appl Math Comput 2006;186(1):286–93.
[20] Yu R, Zhang H. New function of Mittag–Leffler type and its application in the fractional diffusion-wave equation. Chaos, Solitons & Fractals
2006;30:946–55.
[21] Adomian G. Solving frontier problems of physics: the decomposition method. Dordrecht: Kluwer Academic Publishers; 1994.
[22] Biazar J, Babolian E, Islam R. Solution of the system of ordinary differential equations by Adomian decomposition method. Appl Math Comput
2004;147(3):713–9.
[23] Jafari H, Daftardar-Gejji V. Solving a system of nonlinear fractional differential equations using Adomian decomposition. J Comput Appl Math
2006;196:644–51.
[24] Daftardar-Gejji V, Babakhani A. Analysis of a system of fractional differential equations. J Math Anal Appl 2004;293:511–22.
[25] Zaslavsky GM, Stanislavsky AA, Edelman M. Chaotic and pseudochaotic attractors of perturbed fractional oscillator. arXiv:nlin.CD/0508018.
[26] Matignon D. Stability results for fractional differential equations with applications to control processing. Computational Eng in Sys Appl, vol. 2, Lille,
France 963, 1996.
[27] Oldham KD, Spanier J. The fractional calculus: theory and applications of differentiation and integration to arbitrary order. Mathematics in science and
engineering, vol.111. New York: Academic Press; 1974.
[28] Momani S, Odibat Z. Numerical comparison of methods for solving linear differential equations of fractional order. Chaos, Solitons & Fractals
2007;31:1248–55.