Physica A 365 (2006) 300–306

Pattern formation in a fractional reaction–diffusion system
V.V. Gafiychuk
a,Ã
, B.Yo. Datsko
b
a
Institute of Computer Modeling, Cracow University of Technology, ul. Warszawska 24, Cracow 31-155, Poland
b
Institute for Applied Problems of Mechanics and Mathematics, National Academy of Sciences, Naukova St, 3b, Lviv 79601, Ukraine
Received 28 June 2005; received in revised form 13 September 2005
Available online 18 October 2005
Abstract
We investigate pattern formation in a fractional reaction–diffusion system. By the method of computer simulation of the
model of excitable media with cubic nonlinearity we are able to show structure formation in the system with time and space
fractional derivatives. We further compare the patterns obtained by computer simulation with those obtained by
simulation of the similar system without fractional derivatives. As a result, we are able to show that nonlinearity plays the
main role in structure formation and fractional derivative terms change the transient dynamics. So, when the order of time
derivative increases and approaches the value of 1.5, the special structure formation switches to homogeneous oscillations.
In the case of space fractional derivatives, the decrease of the order of these derivatives leads to more contrast dissipative
structures. The variational principle is used to find the approximate solution of such fractional reaction–diffusion model.
In addition, we provide a detailed analysis of the characteristic dissipative structures in the system under consideration.
r 2005 Elsevier B.V. All rights reserved.
Keywords: Fractional reaction–diffusion equations; Self-organization; Pattern formation; Dissipative structures; Traveling pulse
1. Introduction
Pattern formation in evolutionary dynamics has been studied as an example of phenomena inherent to
many natural systems. It is well known that the reaction–diffusion models are used to study self-organization
phenomena in physical, chemical and biological systems (see, for example, Refs. [1–4]). The interest in self-
organization phenomena dynamics as a means to explore diversity of the phenomena in nonlinear dissipative
systems far from equilibrium has grown exponentially over the past decade. There are several approaches that
we use to find solutions for nonlinear dissipative systems. Computer simulation of the nonlinear systems is the
most widely used method of investigating system dynamics. The variational method has been developed to
yield approximate solutions of the reaction–diffusion system. By this method, having obtained the nonlinear
solutions from the computer simulation, it is possible to find the analytical approximate solutions using the
variational principle [5–9]. As a result, we obtain an analytical expression for parameters approximating these
solutions. Such an approach was successfully used for the investigation of self-organization phenomena in
ARTICLE IN PRESS
www.elsevier.com/locate/physa
0378-4371/$ - see front matter r 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2005.09.046
Ã
Corresponding author.
E-mail address: viva@pancha.lviv.ua (V.V. Gafiychuk).
different reaction–diffusion systems and many new nonlinear phenomena were revealed by application of a
variational method [8,9].
In the recent years there has also been a great deal of interest in fractional diffusion models [10–14]. These
models are obtained by replacing in the classical diffusion equations the partial derivatives with respect to
space and time by integro-differential operators, namely by derivatives of noninteger order, in such a way that
the resulting Green function can still be interpreted as a probability density evolving in time differently from
the Gaussian type [15]. The diffusion processes of such type of behavior are called anomalous. The article [16]
provides a thorough review of the recent developments in the framework of fractional dynamics and its
applications.
In this paper, we study pattern formation in the fractional reaction–diffusion system with model
nonlinearity of cubic type and analyze peculiarities of this formation with respect to the corresponding system
with regular derivatives. We consider the situation with enhanced diffusion usually met in biological systems
and turbulent plasma [16], poly-crystals and heterogeneous media [10], etc.
The characteristic feature of fractional reaction–diffusion models is that not only can we not analyze
nonlinear solutions analytically, but also the linear theory of instability of uniform distributions of the
variables is not completely developed [11,17]. In this case, the computer simulation of the model under
consideration is difficult to overestimate. To study the self-organization phenomena, we consider the following
model that describes an excitable medium in terms of fractional reaction–diffusion equations

0
D
b
t
u ¼ l
2
D
a
x
u Àqðu; v; AÞ, (1)
0
D
b
t
v ¼ L
2
D
a
x
v ÀQðu; v; AÞ. (2)
Here l; L are the characteristic lengths of variations of u and v, respectively; ; A are certain parameters. The
general view of the reaction–diffusion equations is the same as considered in Refs. [1,2]; the only difference is
the form of the time and spatial operators. The operators on the left-hand side of Eqs. (1), (2), instead of time
derivatives, are the Caputo fractional derivatives in time, of the order 1obo2 [18],
0
D
b
t
f ðx; tÞ ¼
1
Gð2 ÀbÞ

t
0
q
2
t
f ðx; tÞ
ðt ÀtÞ
aÀ1
dt. (3)
The operators D
a
x
on the right-hand side of Eqs. (1), (2) are Riemman–Liouville fractional derivatives, and
they can be represented in the left and right form respectively, as Ref. [18]:
D
a
x
u
0
D
a
x
f ðx; tÞ ¼
1
GðmÀaÞ
q
m
x

x
0
f ðy; tÞ
ðx ÀyÞ
aþ1Àm
dy, (4)
D
a
x
u
x
D
a
L
x
f ðx; tÞ ¼
ðÀ1Þ
m
GðmÀaÞ
q
m
x

L
x
x
f ðy; tÞ
ðy ÀxÞ
aþ1Àm
dy, (5)
where mÀ1oaom with m being a positive integer, and we consider the system for 1oao2 on the interval
[0; L
x
].
This system is a subject to periodic boundary conditions uð0; tÞ ¼ uðL
x
; tÞ; vð0; tÞ ¼ vðL
x
; tÞ; u
x
ðx ¼ 0; tÞ ¼
u
x
ðx ¼ L
x
; tÞ; v
x
ðx ¼ 0; tÞ ¼ v
x
ðx ¼ L
x
; tÞ.
The specific form of the ‘‘sources’’, chosen here, is [6,7,19,20]:
q ¼ Àu þu
3
=3 þv; Q ¼ v Àu ÀA. (6)
Nullclines of these nonlinearities presented in Fig. 1(a) are the same as in the FitzHugh–Nagumo equations.
The variables u and v are the scalar real fields representing the activator and inhibitor, respectively. Depending
on the parameter A of the nullcline Q ¼ 0 the model (1), (2) has different solutions which we will try to find by
computer simulation. The reaction–diffusion system (1), (2) has a spatially homogeneous state u
0
; v
0
, which at
certain value of the parameter A is unstable. The type of this instability in the case of regular derivatives
(a ¼ 2; b ¼ 1) is very well known [1,2] and can lead to Turing or Hopf bifurcation. In fact, for the standard
reaction–diffusion system (a ¼ 2; b ¼ 1) with sources (6) when l=L51 the homogeneous state is unstable
under certain conditions ðÀ
1
3
oAo
1
3
Þ with respect to the wave numbers k
0
’ ðlLÞ
À1=2
ðq
0
y
Q
0
Z
ÀQ
0
y
q
0
Z
Þ
1=4
(Turing
ARTICLE IN PRESS
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 301
bifurcation). At the same values of the parameter A ðÀ
1
3
oAo
1
3
Þ the homogeneous distribution u
0
; v
0
is
unstable with respect to homogeneous (k ¼ 0) fluctuations o
0

À1=2
ðq
0
y
Q
0
Z
ÀQ
0
y
q
0
Z
Þ
1=2
if 51 (Hopf
Bifurcation).
In the case of fractional derivatives, we cannot implement such simple linear analysis of the system (1), (2)
[11,17] and there is a question about the possible patterns formation.
We perform numerical simulation of the system (1), (2) for establishing characteristic features of pattern
formation of the system under consideration. As a result, we show that the generated patterns in the case of
fractional derivatives are similar to the pattern we usually get from the regular reaction–diffusion system. The
obtained results confirm that the nonlinear mechanism, which is responsible for pattern formation, is mainly
the same as for the standard reaction–diffusion system. Fractional derivatives influence the dynamics of
pattern formation and their form. In this case, transient processes of pattern formation are different depending
on the value of the derivative.
Let us consider two different cases. In the first case, time derivatives are fractional and space derivatives are
integers. In the second case, time derivatives are integers and space derivatives are fractional.
2. The time fractional reaction–diffusion system
First of all we start with consideration of the reaction–diffusion equations (1), (2) provided that time
derivatives are fractional
0
D
b
t
u ¼ l
2
u
xx
þu Àu
3
=3 Àv, (7)
0
D
b
t
v ¼ L
2
v
xx
Àv þu ÀA. (8)
Now the original set of Eqs. (7), (8) will be the subject of numerical simulation. Using the classical expression
for the form of the second space derivative and the fractional differential scheme for the Grunwald–Letnikov
operator [21], which includes the initial condition, we developed numerical algorithms similar to [22]. In fact,
in a boundary—initial problem we define two kinds of grids: spatial and temporal. The spatial domain ½0; L
x
Š
is divided into the uniform mesh with N þ1 nodes x
i
¼ ih for i ¼ 0; N, where h ¼ L
x
=N. We used a classical
expression for the approximation of the second derivatives in the right-hand side of the system. The time-
interval ½0; TŠ is divided into F subintervals where the sub-interval length equals dt ¼ T=F and the time-nodes
are t
f
¼ f Âdt for f ¼ 0; F. In case where a is a fractional number the discrete forms of fractional derivatives
is exactly the same as derived in the article [22].
Here we present the results of computer simulation for the following characteristic examples. In Fig. 1(b) we
plotted the stationary distribution of the activator (variable u) and inhibitor (variable v) for b sufficiently close
to unity for two different values of the parameter l. In this case such distribution corresponds to the
distribution we usually obtain by the computer simulation of the regular system (b ¼ 1). This fact confirms
that the computation schema is reliable, while stationary distributions, when they exist, do not depend on the
parameter b. The figure shows that the profile of the dissipative structures is mainly influenced by the diffusion
ARTICLE IN PRESS
Fig. 1. (a) Nullclines of the system under consideration and (b) solutions at A ¼ 0. Solid lines correspond to variable u for b ¼ 1:1; l
2
¼
0:01; L ¼ 1; L
x
¼ 2:8 (black line) and for b ¼ 1:1; l
2
¼ 0:05; L ¼ 1 (gray line). Dotted line corresponds to variable v.
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 302
lengths l; L. Fig. 2 shows the kinetics of dissipative structures formation for the case A ¼ 0. The system was
simulated by changing the parameter b. For b approaching 1 the dynamics is quite similar to the case b ¼ 1.
Increasing the value b leads to the generation of small oscillations at the beginning of the structure formation,
and the transient process is not monotonic. Nevertheless, the dissipative structures emerge, and as a result we
have steady state dissipative structures when the parameter b ¼ 1:45. At some value near b ¼ 1:5, instead of
steady state dissipative structure formation, we have oscillations of the homogeneous state (Fig. 2d).
At A ¼ 0 we get u ¼ v ¼ 0 for homogeneous distributions given by (7), (8). In the linear limit the variables
u; v are correlated and the linear stage can be approximately represented by a single oscillator equation
0
D
b
t
u þu ¼ 0
which has recently been investigated in Ref. [23] and the solution of such oscillator for uð0Þ ¼ d; u
0
ð0Þ ¼ 0 is
given by the formula
uðtÞ ¼
d
p

1
0
e
Àrt
r
bÀ1
sinðpbÞ dr
r
2b
þ1 þ2r
b
cosðpbÞ
þ
d
p
2
b
e
t cosðp=bÞ
cos t sin
p
b
¸
. (9)
As it was analyzed in detail in Ref. [23], the solution represents damping oscillations when b41. These
oscillations are at initial stages of their paths of the order of d and that is probably the reason why the
nonhomogeneous dissipative structures are unstable at the increase of the parameter b.
We have also studied the dependence of such structures formation on the bifurcation parameter A. The
computer simulation shows that in all the region where nullcline q ¼ 0 is a growing function, the dissipative
structures exist and have the form presented in Figs. 1(b) and 2. It should be noted that the width (the distance
between two opposite walls on Fig. 1(b) of the structures decreases with the decrease of the parameter A, and
ARTICLE IN PRESS
Fig. 2. Kinetics of dissipative structure formation for different values of b: (a) b ¼ 1:01; (b) b ¼ 1:35; (c) b ¼ 1:45; (d) b ¼ 1:5. All other
parameters are: A ¼ 0; l
2
¼ 0:02; L ¼ 1; L
x
¼ 4.
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 303
increases when A increases. The kinetics in structure formation is practically the same for the whole region of
the existing dissipative structures ðÀ
1
3
oAo
1
3
Þ.
The next section serves to illustrate the qualitative features of the system under consideration when special
derivatives are of the fractional order.
3. The space fractional reaction–diffusion system
Let us investigate pattern formation features that emerge from considering the spatial fractional derivatives.
More precisely, we focus on the nonlinear fractional reaction–diffusion system:
u
t
¼ l
2
D
a
x
u þu Àu
3
=3 Àv, (10)
v
t
¼ L
2
D
a
x
v Àv þu ÀA. (11)
Numerical simulation of the system (10), (11) is based on the well-known technique for solving parabolic type
partial differential equations, namely, the methods of lines. For this Riemman–Liouville fractional derivatives
(at space) are represented at every point of the interval by corresponding difference convolution integrals,
while the derivatives along the time axis are untouched [24]. The system of partial differential equations is
thereby transformed into a system of ordinary differential equations (ODE) and then integrated in time using
an ODE software [25]. The choice of the value of discretization and the local error of integration were made so
that the necessary accuracy of the desired solution and the stability of calculations were ensured.
Some of the results of the computer simulation of the system are presented in Figs. 3 and 4. In the case when
¼ 1 we can get a stationary dissipative structures, the form of which is a little bit different from the structures
of the standard reaction–diffusion system. Dissipative structures have practically the same parameters (like the
same width), but exhibit much more contrast. The fact is that dispersion influenced by space fractional
derivatives is smaller than dispersion given by second order derivatives. Changing the parameter A changes the
general form of these structures, but almost does not change the region of their existence (Fig. 3).
It should be noted that in this case the effective way of investigation of the patterns is the use of variational
principles [5–7]. The matter is that contrasting distributions of the fast variable u and smooth distributions of
the variable v show that the variational principle is much more applicable than in the case of the standard
reaction–diffusion model (a ¼ 2). Such an approach makes it possible to find the approximate analytical
solution of dissipative structures. Let us write down the variational principle in the form [5–7]:
D ¼
1
2

1
À1
½ð
_
y Àl
2
D
a
x
y þqÞ
2
þð_ Z ÀL
2
D
a
x
Z þQÞ
2
Š dx. (12)
As an example, we will show how such a variational principle (12) can be used to find the solutions of Eqs. (1),
(2). To do this we consider the solution of the system for ¼ l=L51 of the fields y and Z as the combination of
sharp and smooth pieces for the distribution of y and smooth structures for Z at the interval ½0; L
x
=2Š. In
ARTICLE IN PRESS
Fig. 3. Stationary state for a ¼ 1:5 (solid line): (a) A ¼ 0:2 and (b) A ¼ 0. All other parameters are: ¼ 1; l
2
¼ 0:01; L ¼ 1; L
x
¼ 2:8.
Dotted line corresponds to standard reaction–diffusion system.
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 304
particular, we assume that the model solutions have the form [5–7]
y ¼
1
~
ðx ÀxÞ; jx Àxjp~ ;
y
h
; x þ ~ pxpL
x
=2;
Ày
h
; 0pxpx À ~ :

(13)
Z ¼ Z
0
¼ const. (14)
Substituting (13), (14) into (12) then we can determine their parameters by writing the following system of
equations of motion:
dD
d
_
x
¼ 0;
dD
d
_
y
h
¼ 0;
dD
d_ Z
0
¼ 0. (15)
Performing calculation we have, up to small quantities ~ ¼ l=L51, the following equations for the variables x,
Z
0
and y
h
:
_
x À ~ Z
0
¼ 0,
_ Z
0
þZ
0
ÀA À
L
2
Z
0
Gð2 ÀaÞð1 ÀaÞ
2
ðL
x
=2Þ
Àa
þy
h
2x ÀL
x
=2
L
x
=2
¼ 0,
_
y
h
þ
y
3
h
3
Ày
h
ÀZ
0
2x ÀL
x
=2
L
x
=2 À2~
¼ 0,
from which it follows for A ¼ 0, ~ ¼ l=L51 (Fig. 3) the stationary state of the original system is:
Z
0
¼ 0; y
h
¼
ffiffiffi
3
p
; x ¼ L
x
=4 (16)
and the solution (13), (14) corresponds to the one represented on the Fig. 3.
We illustrate also the existence of traveling structures which are presented at A ¼ 1:1 and A ¼ 0:5. At the
small value of , the classic reaction–diffusion system has a traveling pattern solution. In particular, for L ¼ 0
the reaction–diffusion model is known as the FitzHugh–Nagumo model. The results of computer simulation
of such model with fractional derivatives is presented in Fig. 4 a,b. The stimulated traveling pulses in this case
are not so wide and have a smaller velocity. Nevertheless, the main features of these traveling pulses are the
same as in the regular case.
4. Conclusion
The reaction–diffusion system has been extended in order to take into account fractional derivatives. It has
been demonstrated by computer simulation that dissipative structures, inherent to the standard system, exist
ARTICLE IN PRESS
Fig. 4. Traveling pulses for a ¼ 1:5 (solid line): (a) A ¼ 1:0, for (b) A ¼ 0:5. All other parameters are: ¼ 0:02; l
2
¼ 0:01;
L ¼ 0:0; L
x
¼ 8. Dotted line corresponds to standard reaction–diffusion system.
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 305
for this more general model also. We performed investigation of the kinetics of the structure formation
provided the order of the time derivatives increases bX1. We were able to show that at certain value of b, the
formation of dissipative structures switches to formation of homogeneous oscillations.
In the case of space fractional derivatives we investigated the form of the stationary and traveling structures
when the order of the derivatives decreases (ap2). Despite the fact that the form of the obtained structures is
different and the structures are much more contrast, they possess similar nonlinear properties. The obtained
steady state and traveling dissipative structures in the zeroth approximation represent the combination of
sharp walls and smooth distribution of variables. We used generalized Gauss variational principal to find an
analytical expression for approximate solution of the steady state dissipative structures.
References
[1] G. Nicolis, I. Prigogine, Self-organization in Non-equilibrium Systems, Wiley, New York, 1977.
[2] B.S. Kerner, V.V. Osipov, Autosolitons, Kluwer, Dordrecht, 1994.
[3] J. Buceta, K. Lindenberg, Patterns in reaction–diffusion systems generated by global alternation of dynamics, Physica A 325 (1–2)
(2003) 230–242.
[4] M. Mimura, H. Sakaguchi, M. Matsushita, Reaction–diffusion modelling of bacterial colony patterns, Physica A 282 (1–2) (2000)
283–303.
[5] V.V. Gafiychuk, I.A. Lubashevskii, Variational principles of dissipative systems, J. Sov. Math. 67 (2) (1993) 2943–2949.
[6] A. Lubashevskii, V.V. Gafiychuk, The projection dynamics of highly dissipative system, Phys. Rev. E 50 (1) (1994) 171–181.
[7] V.V. Gafiychuk, I.A. Lubashevskii, Variational representation of the projection dynamics and random motion of highly dissipative
systems, J. Math. Phys. 36 (10) (1995) 5735–5752.
[8] V.V. Osipov, Multifunctional variational method for description of evolution and dynamics of dissipative structures in
nonequilibrium systems, Phys. Rev. E 48 (1993) 88–100.
[9] C.B. Muratov, V.V. Osipov, Theory of spike spiral waves in a reaction–diffusion system, Phys. Rev. E 60 (1999) 242–246.
[10] I.A. Lubashevskii, V.L. Alatortsev, A.G. Keijan, Grain boundary random walks and diffusion in polycrystals, Physica A 193 (2)
(1993) 259–303.
[11] B.I. Henry, S.L. Wearne, Fractional reaction–diffusion, Physica A 276 (2000) 448–455.
[12] R. Hilfer, On fractional diffusion and continuous time random walks, Physica A 329 (2003) 35–40.
[13] L.R. da Silva, L.S. Lucena, E.K. Lenzi, R.S. Mendes, Kwok Sau Fa, Fractional and nonlinear diffusion equation: additional results,
Physica A 344 (2004) 671–676.
[14] H. Weitzner, G.M. Zaslavsky, Some applications of fractional equations, Commun. Nonlinear Sci. Numer. Simulat. 8 (2003)
273–281.
[15] R. Gorenflo, F. Mainardi, Simply and multiply scaled diffusion limits for continuous time random walks, J. Phys.: Conf. Ser. 7 (2005)
1–16.
[16] R. Metzler, J. Klafter, The restaurant at the end of the random walk: recent developments in the description of anomalous transport
by fractional dynamics, J. Phys. A 37 (2004) R161–R208.
[17] V. Daftardar-Gejji, H. Jafari, Adomian decomposition: a tool for solving a system of fractional differential equations, J. Math. Anal.
Appl. 301 (2005) 508–518.
[18] S.G. Samko, A.A. Kilbas, O.I. Marichev, Fractional Integrals and Derivatives: Theory and Applications, Gordon and Breach,
Newark, NJ, 1993.
[19] A. Hagberg, E. Meron, I. Rubinstein, B. Zaltzman, Order parameter equations for front transitions: planar and circular fronts, Phys.
Rev. E 55 (4) (1997) 4450–4457.
[20] A. Hagberg, E. Meron, Kinematic equations for front motion and spiral-wave nucleation, Physica A 249 (1998) 118.
[21] I. Podlubny, Fractional Differential Equations, Academic Press, San Diego, 1999.
[22] M. Ciesielski, J. Leszczynski, Numerical simulations of anomalous diffusion, Arxiv.org/abs/math-ph/0309007.
[23] G.M. Zaslavsky, A.A. Stanislavsky, M. Edelman, Chaotic and Pseudochaotic Attractors of Perturbed Fractional Oscillator,
arXiv:nlin.CD/0508018.
[24] F. Liu, V. Anh, I. Turner, Numerical solution of the space fractional Fokker–Planck equation, J. Comp. Appl. Math. 166 (2004)
209–219.
[25] J. Byrne, D.C. Hindmarch, Stiff ODE Solver: a review of current and coming attractions, J. Comput. Phys. 70 (1) (1987) 1–60.
ARTICLE IN PRESS
V.V. Gafiychuk, B.Y. Datsko / Physica A 365 (2006) 300–306 306