You are on page 1of 10

Vasyl Gafiychuk

SGT Inc.,
7701 Greenbelt Rd Suite 400,
Greenbelt, MD, 20770;
NASA Ames Research Center,
Moffett Field, CA, 94035-1000
Bohdan Datsko
1
Institute of Applied Problems of Mechanics
and Mathematics,
NAS of Ukraine,
Naukova Street 3B, Lviv, 79053, Ukraine
e-mail: b_datsko@yahoo.com
Different Types of Instabilities
and Complex Dynamics in
Reaction-Diffusion Systems With
Fractional Derivatives
In this article we analyze conditions for different types of instabilities and complex
dynamics that occur in nonlinear two-component fractional reaction-diffusion systems. It
is shown that the stability of steady state solutions and their evolution are mainly deter-
mined by the eigenvalue spectrum of a linearized system and the fractional derivative
order. The results of the linear stability analysis are conrmed by computer simulations
of the FitzHugh-Nahumo-like model. On the basis of this model, it is demonstrated that
the conditions of instability and the pattern formation dynamics in fractional activator-
inhibitor systems are different from the standard ones. As a result, a richer and a
more complicated spatiotemporal dynamics takes place in fractional reaction-diffusion
systems. A common picture of nonlinear solutions in time-fractional reaction-diffusion
systems and illustrative examples are presented. The results obtained in the article for
homogeneous perturbation have also been of interest for dynamical systems described by
fractional ordinary differential equations. [DOI: 10.1115/1.4005923]
1 Introduction
In recent years, there has been an increasing interest in the study
of dynamical mathematical models with fractional derivatives. This
interest is mainly determined by the attempts to understand phe-
nomena in fractal, irregular and hereditary media. The last investi-
gations have shown that anomalous behavior of many complex
heterogeneous systems is better described by fractional differential
equations (see for example Refs. [13] and references therein).
Among the fractional differential models, much attention has
been given to the fractional reaction-diffusion systems (RDS)
[49]. At the present time, fractional RDS (FRDS) describe a
large class of systems at different scales from the molecular [10]
to the space ones [11]. Due to this fact, the development of the
theory of such systems is important both from the scientic per-
spective and for its application in a set of new technologies. On
the basis of mathematical modeling of the standard reaction-
diffusion equations, a lot of amazing nonlinear self-organization
phenomena in physical, biological and chemical systems have
been explained [1214]. Moreover, the investigations of the
spatio-temporal order in such nonlinear systems and mechanisms
of pattern formation are a top-priority theme of present research
studies in many modern technological applications [14,15]. There-
fore, the development of fractional models for such media, as
granular and porous materials, various materials with memory,
complex chemical environments and biological tissues, where
self-organization phenomena are observed and diffusion has
essentially an anomalous character, represents a special interest.
A fractional reaction-diffusion equation is derived from a con-
tinuous time random walk model when the transport is dispersive
[16,17]. In Ref. [4], a general multi-species system undergoing
anomalous sub-diffusion with linear reaction dynamics is consid-
ered. The validity of such fractional reaction-diffusion (RD)
model by comparing solutions with Monte Carlo simulations
was conrmed. The extension to nonlinear reaction terms is
non-trivial. Several models with temporal order derivatives oper-
ating on both the Laplacian diffusion term and nonlinear reaction
kinetics from the law of mass action have been considered
[18,19]. The results for front propagation in such models are also
in reasonable agreement with Monte Carlo simulations [18].
These papers provide a useful platform for developing robust
models for multispecies fractional systems with nonlinear reac-
tions (the overview of different platforms for modeling reaction-
transport systems, including the ones with anomalous diffusion, is
presented in Ref. [9]).
For commensurate nonlinear time-fractional activator-inhibitor
RDS, the basic results are obtained for classical reaction kinetics.
In a series of articles [1820], through linear stability analysis and
numerical simulations, a stationary Turing pattern formation is
investigated. In Refs. [5,20] it is shown that the fractional deriva-
tive index is an additional bifurcation parameter which switches the
stable and unstable states of the system. Furthermore, in Ref. [5] it
was revealed that in FRDS, for a certain value of fractional deriva-
tive index, a new type of bifurcation takes place and the system can
be unstable towards perturbations of nite wave number. In these
articles, the limited cases are primarily investigated, when the
FRDS demonstrate relatively simple nonlinear dynamics.
In the present article, we have focused on FRD model under con-
ditions, when sufciently complex spatial-temporal patterns arise
in the system dynamics. By linear stability analysis and computer
simulation, it was shown that fractional derivative order can change
the stability of steady state solutions and signicantly enrich non-
linear system dynamics. The diversity of observed nonlinear phe-
nomena needs some classication in order for us to understand the
conditions for different pattern formation in such systems. In the
standard reaction-diffusion (RD) systems, such a classication has
been done a long time ago [13]. In our paper, we would like to
make the rst attempt for such classication for fractionalRDS. On
the basis of the time-fractional system with cubic nonlinearity, the
conditions for different types of instability in detail are analyzed
and an overall picture of different types of nonlinear solutions,
depending on parameters of the system, is presented.
2 Formulation of the Problem
In a general case, a reaction-diffusion model can be described
by a system of m nonlinear partial differential equations of para-
bolic type
1
Corresponding author.
Manuscript received August 16, 2009; nal manuscript received January 7, 2012;
published online March 13, 2012. Assoc. Editor: Om Parkash Agrawal.
Journal of Computational and Nonlinear Dynamics JULY 2012, Vol. 7 / 031001-1
Copyright VC
2012 by ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
su
t
DDu fu; A (1)
in space-time domain X W subject to certain boundary
conditions
l
@u
@n
1 lu on @X W (2)
and initial conditions
ur; 0 u
0
r for r 2 X (3)
Here, X is a bounded space domain in R
p
, p 2 f1; 2; 3g with
smooth boundary @X and W 0; T denes a time domain with
0 < T < 1. Vector u u
1
r; t; ; u
m
r; t
T
species the
variables of the model, positive diagonal matrices s
mm
and D
mm
represent the time and space scales of the system, respectively,
A 2 R R is an external bifurcation parameter and vector
f f
1
; f
2
; :::; f
m

T
denes the given smooth reaction-kinetics
functions f
i
u
1
r; t; u
2
r; t; :::; u
m
r; t : R
m
!R
m
Vector

1
; ;
m
, diagonal matrices l diagl
1
; ; l
m
,
1 diage
1
; ; e
m
with e
i
1, l
i
2 0; 1,
i
2 C
2
@X W
for all 2 1; ; m and normal vector n to the boundary @X deter-
mine the boundary conditions of the problem.
The starting point of our consideration is the fractional RDS
su
a
t
DDu fu; A (4)
Time derivatives u
a
t

@
a
u1r; t
@t
a
; ;
@
a
umr; t
@t
a
_ _
T
on the left-hand
side of the Eq. (4) instead of the standard time derivatives are the
Caputo fractional derivatives in time [2,3,21] of the order
0 < a < 2 and are represented as
@
a
c
u
i
r; t
@t
a
:
1
Cn a
_
t
0
u
n
i
r; s
t s
a1n
ds (5)
where n 1 < a < n; n 2 f1; 2g, i 2 f1; 2; ; mg, a 2 R.
Introduction of time fractional derivatives into differential
equations appreciably expands the family of dynamical models. In
particular, the time-fractional reaction-diffusion system sets a pos-
sibility of continuous transition between the parabolic, elliptic and
hyperbolical types of partial derivative equations, and we can con-
sider the fractional RDS as a generalized system which is a matter
of considerable theoretical interest [7,8].
For the study of the main characteristics of nonlinear dynamics,
the common RD model Eq. (1) in the simplest but quite general
case can be reduced to a system of two coupled one- dimensional
nonlinear equations of the activator-inhibitor type [12,13]:
s
1
@
a
c
u
1
x; t
@t
a
l
2
1
@
2
u
1
x; t
@x
2
Wu
1
; u
2
; A (6)
s
2
@
a
c
u
2
x; t
@t
a
l
2
2
@
2
u
2
x; t
@x
2
Qu
1
; u
2
; A (7)
Here u
1
x; t, u
2
x; t are the activator and inhibitor variables,
0 x L, s
1
; s
2
; l
1
; l
2
, as dened above, are the characteristic
times and lengths of the system, correspondingly, and A is an
external parameter. In the integer case a 1, the system (Eqs.
(6) and (7)) is the basis model for investigation of self-
organization phenomena in non-equilibrium media of different na-
ture [1214]. The positive feedback on the activator variable u
1
,
negative feedback on the inhibitor variable u
2
and the difference
in characteristic times and lengths lead to rich variety of qualita-
tively different solutions [12,13]. In the present article we investi-
gate the solutions of the system (Eqs. (6) and (7)) in a more
general case 0 < a < 2 and demonstrate the role of the frac-
tional derivative order on their stability and evolution.
It should be noted, that the question of global existence of
a solution for nonlinear RD systems even in the case of a two-
component ones with an integer derivative is an unresolved
problem. The review of the main results on global existence of
solutions in time for the family of m m RDS is well repre-
sented in a survey [22]. These RDS must satisfy the two main
properties: (a) the non-negativity of the solutions is preserved for
all time, and (b) the total mass of the components is a priori
bounded on all nite intervals. A lot of systems come naturally
with these two properties in applications. In our article, we
investigate RDS, which with a certain transformation can be
reduced to these properties. It is classical that for such type of
system a local solution exists and may be extended on interval
0; T. Moreover, for the system (Eq. (1)), the existence of spa-
tially homogeneous and stationary solutions, which can be
obtained from algebraic system fu; A 0 and realized at peri-
odic or neutral boundary conditions, is suggested. If the algebraic
system is consistent, it determines a set of solutions depending
on the bifurcation parameter. In this article, we consider that the
algebraic system has only one solution and in the next subsection
we will study the stability of this stationary solution for general-
ized system (Eq. (4)).
3 Linear Stability Analysis
Lets rst consider the formulation which allows us to deter-
mine the stability of such stationary solutions. The system
(Eq. (4)) can be presented in operator form
u
a
t
Nu; A (8)
where
Nu; A s
1
DDu fu; A (9)
Due to properties of the Caputo derivative, the stationary solution
ur is a particular solution of the Eq. (8). The stability of this
stationary solution, as in the case of integer time derivative, can
be estimated on the basis of the principle of linear stability, which
for fractional derivative evolutionary equation can be formulated
as:
Theorem 1. Lets assume ur is the stationary solution of
system Eq. (8) and ~ ur; t ur; t ur is the vector of devia-
tions from ur: j ~ u j=j u j << 1. If the operator N in the vicinity
of the stationary solution ur admits of expansion into a power
series of ~ u, then the stability of Eq. (8) can be reduced to the prob-
lem of stability of trivial solutions of the linear system
~ u
a
t
Lu; A~ u (10)
where Lu; A dN=du
u
is a Frechet derivative of the
operator N.
(1) If a trivial solution ~ u 0 of the linear problem (Eq. (10)) is
asymptotically stable, then the stationary solution u of the
nonlinear problem (Eq. (8)) is also asymptotically stable.
(2) If a trivial solution ~ u 0 of the linear problem (Eq. (10)) is
unstable, then stationary solution u of the nonlinear prob-
lem (Eq. (8)) is also unstable.
For a 1, this approach was considered by Nicolis and Prigo-
gine [23]. The constructive proof of this theorem is presented
below by considering of two-component reaction-diffusion system
with 0 < a < 2.
3.1 Linear Stability Analysis of Homogeneous Solution of
One-Dimensional Activator Inhibitor System. Let us now con-
sider in detail a special case of the system (Eq. (4)). For activator-
inhibitor RDS (Eqs. (6) and (7)) stationary solution can be
obtained from the system of algebraic equations
031001-2 / Vol. 7, JULY 2012 Transactions of the ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Wu
1
; u
2
; A 0; Qu
1
; u
2
; A 0 (11)
A simultaneous solution of the Eq. (11) leads to a spatially-
homogeneous solution u u
1
; u
2
, which is realized at periodic
u
i
j
x0
u
i
j
L
; @u
i
=@xj
x0
@u
i
=@xj
xL
; i 1; 2 (12)
or neutral boundary conditions
@u
i
=@xj
x0
@u
i
=@xj
xL
0; i 1; 2 (13)
The algebraic system (Eq. (11)) can have a set of real roots, but
we consider the case when the system (Eqs. (6) and (7)) has only
one stationary and spatially homogeneous solutions. The stability
of such steady-state solution of the system (Eqs. (6) and (7)) cor-
responding to the homogeneous equilibrium state Wu; A 0,
Qu; A 0 can be analyzed by linearization of the system
nearby this solution u u
1
; u
2

T
. As result, we can write the
expression for eigenvalues of the linearized system [9,12,13]
k
1; 2

1
2
trF6

tr
2
F 4 det F
p
(14)
which mainly denes the stability and nonlinear dynamics of the
system (Eqs. (6) and (7)). Here, the matrix
Fk Lj
u1; u2

a
11
k
2
l
2
1
=s
1
a
12
=s
1
a
21
=s
2
a
22
k
2
l
2
2
=s
2
_ _
(15)
is determined by a
11
W
0
u1
, a
12
W
0
u2
, a
21
Q
0
u1
, a
22
Q
0
u2
(all derivatives are taken at homogeneous equilibrium
states u u
1
; u
2

T
, trFk a
11
k
2
l
2
1
=s
1
a
22
k
2
l
2
2
=s
2
,
detFk a
11
k
2
l
2
1
a
22
k
2
l
2
2
=s
1
s
2
a
12
a
21
=s
1
s
2
, k
pj=L, j 1; 2; ..
For a 1, this stationary solution is unstable when the real part
of any eigenvalue Rek > 0.. In reaction-diffusion systems, a stable
homogeneous equilibrium solution ux const usually changes
spontaneously with the external parameters to the limit cycle by
Hopf bifurcation or stationary dissipative structures (DS) by
Turing bifurcation. As a result, we obtain nonlinear dynamics
leading to spatially-homogeneous oscillatory solutions or
spatially-inhomogeneous stationary DS. When conditions of both
instabilities arise, we can expect more complex dynamics. In the
case of a fractional reaction-diffusion (FRD) system, the dynamics
can be much more complex [5,19,20].
For a 6 1 we can also apply an analogous procedure to the
system (Eqs. (6) and (7)). Due to the property of Caputo deriva-
tive [3,21], we have a similar linear system with the same right-
hand side operator (Eq. (15)). The stability of this fractional linear
system is determined by the next theorem of Matignon [24]:
Theorem 2. The linear autonomous system:
d
a
ut
dt
a
: Au; u0 u
0
(16)
with 0 < a < 1; u 2 R
n
and A 2 R
nn
is asymptotically stable if
and only if
a <
2
p
jArgk
i
j (17)
is satised for all eigenvalues k
i
of matrix A.
Also, this system is stable if and only if a
2
p
jArgk
i
j is satis-
ed for all eigenvalues k
i
of matrix A with those critical eigenval-
ues satisfying a
2
p
jArgk
i
j with geometric multiplicity of one.
As a result of this theorem, we have additionally taken into
account the relationship between the imaginary and the real parts
of eigenvalues of the linearized system [20,24]. If the eigenvalues
k
1
; k
2
are real, then an instability of the system denes the
eigenvalue with a maximal real part. If the eigenvalues are com-
plex, then for 0 < a < 2 for every point inside the parabola
detF tr
2
F=4 (Fig. 1(a)), we can introduce a marginal value
a : a
0

2
p
jArgk
i
j given by the formula [5,20]
a
0

2
p
arctan

4 det F=tr
2
F 1
_
; trF > 0
2
2
p
arctan

4 det F=tr
2
F 1
_
; trF < 0
_

_
(18)
In this case, the value of a is a certain additional bifurcation
parameter which switches the stable and unstable states of the sys-
tem (Figs. 1(a) and 1(b)). At a lower a :a < a
0

2
p
jArgk
i
j, the
system has oscillatory modes, but they are stable. Increasing the
value of a > a
0

2
p
jArgk
i
j leads to oscillatory instability.
It is widely known for integer time derivatives [12,13] that the
system (Eqs. (6) and (7)) becomes unstable according to either
Hopf (k 0)
trF > 0; det F0 > 0 (19)
or Turing (k
0
6 0) bifurcations
trF < 0; det F0 > 0; det Fk
0
< 0 (20)
and these both types of instabilities are realized for positive feed-
back a
11
> 0) [12,13].
For fractional system, Hopf bifurcation is not connected with
the condition a
11
> 0 and can take a place for a
11
< 0 at a certain
values of a [20]. Moreover, in fractional RD systems at a > 1
when it is easier to satisfy conditions of Hopf bifurcation, we
meet a new type of instability [5]
trF < 0; 4 det F0 < tr
2
F0; 4 det Fk
0
> tr
2
Fk
0
(21)
It is worth analyzing inequalities (Eq. (21)) in detail. Taking into
account the explicit form of Fk, the last two conditions can be
rewritten as:
a
11
s
1
a
22
s
2

2
> 4a
12
a
21
s
1
s
2
(22)
4a
12
a
21
s
1
s
2
> a
11
k
2
l
2
1
s
2
a
22
k
2
l
2
2
s
1
_
2
(23)
The simplest way to satisfy the last condition is to estimate the
optimal value of k k
0
:
k
0
2
a
12
a
21
l
2
1
=s
2
l
2
2
=s
1
_ _
1=2
(24)
Having obtained Eq. (24), we can estimate the marginal value
of a
0
Fig. 1 The schematic view of the marginal curve of a
0
(solid
line) and the parabola detF tr
2
F=4 (pointed line) - (a). The
position of eigenvalue k corresponds to the marginal value of a
0
in the coordinate system (Rek; Imk) - (b). Shaded domains cor-
respond to the instability region.
Journal of Computational and Nonlinear Dynamics JULY 2012, Vol. 7 / 031001-3
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
a
0
2
2
p
arctan T (25)
where the expression T is given as
T
4a
12
a
21
s
1
s
2

1=2
a
11
s
2
a
22
s
1

l
2
1
s
2
l
2
2
s
1
l
2
2
s
2
l
2
2
s
1

a
11
s
2
a
22
s
1
(26)
The analysis of expressions in Eq. (21) shows that at k 0, we have
two real eigenvalues that are less than zero, and the system is cer-
tainly stable for the Hopf bifurcation. If the last inequality takes
place for a certain value of k
0
6 0, we can get two complex eigen-
values, and a new type of instability, connected with the interplay
between the determinant and the trace of Fk of the linearized sys-
tem, emerges. With such type of eigenvalues, it is possible to deter-
mine the value of fractional derivative index when the system
becomes unstable for Hopf bifurcation with this wave number k
0
[5].
3.2 Linear Stability Analysis Spatially-Nonhomogeneous
Solution. Stability analysis considered above can be applied also
to stationary spatially-nonhomogeneous solution ux u
1
x;
u
2
x
T
. In this case, the value of a
0

2
p
jArgk
i
j is determined by
Frechet operator for the system (Eqs. (6) and (7)):
Lj
ux

a
11
x r
2
l
2
1
=s
1
a
12
x=s
1
a
21
x=s
2
a
22
x r
2
l
2
2
=s
2
_ _
(27)
where a
11
x W
0
u1
, a
12
W
0
u2
, a
21
Q
0
u1
, a
22
Q
0
u2
(all deriva-
tives are taken at inhomogeneous equilibrium states ux
u
1
x; u
2
x
T
. The solution of this problem is sufciently
complicated but for practical application we can consider nite
difference approximation use the Theorem 2 for determination of
stability conditions.
4 Typical Example: Time-Fractional Van der
Pol-Fitzhugh- Nahumo-Like Model
To demonstrate the properties of FRD system, let us consider
the model with the cubic dependence for the activator variable
W u
1
u
3
1
=3 u
2
(28)
and the linear for the inhibitor variable
Q u
2
bu
1
A (29)
This model was proposed rstly by R. FitzHugh [13] for the descrip-
tion of the propagation of voltage impulse through a nerve axon and
is known as Bonhoeffer-van der Pol model. In RD systems this
model was considered in many books and articles (see for example
Refs. [9,12,13]). The homogeneous solution of variables u
1
and u
2
can be obtained from the system of equations Wu
1
; u
2
; A 0,
Qu
1
; u
2
; A 0, which in turn determines two null- clines u
2
u
1

(Fig. 3(b)). The intersection of these null-clines at the point


P u
1
; u
2
is determined by the equation: u
1
u
3
1
bu
1
A
0. In this case, the values of external parameters A, b determine
the value of u
1
and this makes it possible to investigate the condi-
tions of different types of instability explicitly considering the vari-
able u
1
as the main parameter for the system analysis.
4.1 Space Homogeneous Solutions.
4.1.1 Homogeneous Hopf Instability. For the investigation of
the Hopf bifurcation, let us consider homogeneous perturbation
with the wave number k 0. The linear analysis of the classical
FitzHugh-Nahumo system (a 1) shows that at s
1
=s
2
> 1 the
solution corresponding to any intersections of null- clines is sta-
ble. Formally, the condition trF > 0 is reduced to inequality
W
0
u1
> s
1
=s
2
Q
0
u2
, which in this case has a very simple form
1 u
2
1
> s
1
=s
2
. The smaller is the ratio of s
1
=s
2
, the wider is the
instability region (parabola-like solid lines with points on Fig.
1(b)). At s
1
=s
2
!0, the instability region for u
1
coincides with
the interval (1; 1) where the null cline Wu
1
; u
2
0 has its
increasing part (Fig. 3(b)). These results are very widely known in
the theory of nonlinear dynamical systems [9,12,13].
In the fractional system, the conditions of the instability depend
on the relationship between the real and the imaginary parts of the
complex eigenvalues. This requires an existence of the imaginary
part at least:
4 det F tr
2
F
4 b 1 u
2
1
_ _
s
1
s
2

1 u
2
1

s
1

1
s
2
_ _
2
> 0 (30)
In fact, with complex eigenvalues, it is always possible to nd the
corresponding value of a where the condition a > a
0
is true.
Omitting the simple calculation, we can write an equation for
marginal values of u
1
u
4
1
2 1
s
1
s
2
_ _
u
2
1

s
2
1
s
2
2
2
s
1
s
2
2b 1 1 0 (31)
The solution of this biquadratic equation
u
2
1
1
s
1
s
2
62

b
s
1
s
2
_
(32)
estimates the maximum and the minimum values of u
1
for which
homogeneous state u u
1
; u
2
is unstable at a marginal value of
a a
0
2 as a function of s
1
=s
2
and b.
The typical instability domains for the considered fractional
system, in the coordinates (u
1
; s
1
=s
2
) for different values a are
presented in Fig. 2(a)2(d), where curves corresponding to a 1
are denoted by thicker solid lines. For each particular value a in
the region between the corresponding curve and the horizontal
Fig. 2 Instability domains in coordinates u
1
; s
1
=s
2
for a frac-
tional order reaction-diffusion system with sources
W
1
u
1
u
3
1
u
2
, W
2
u
2
bu
1
A for different values of
a
0
0:25; 0:5; 0:75; 1:0; 1:25; 1:5; 1:75. The results of computer
simulation obtained at l
1
l
2
0 for: b 1:01 - (a), (b) and
b 10:0 - (c), (d).
031001-4 / Vol. 7, JULY 2012 Transactions of the ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
axis, the system is unstable with a wave numbers k 0, and out-
side - it is stable. The left-hand side plots correspond to a < 1 and
the right- hand side plots corresponds to a > 1. Figures 2(a) and
2(c) demonstrate that at a < 1 an increase of relation s
1
=s
2
leads
to decrease of the instability domain, which at s
1
=s
2
1 and
a 1 vanishes completely. The situation changes for a > 1. The
system is unstable not only for s
1
=s
2
< 1, but also for s
1
=s
2
> 1.
An increase in a makes the instability domain much wider with
respect to two coordinates (u
1
; s
1
=s
2
) and we obtain buttery-like
domains for a > 1. This means that, with increasing s
1
=s
2
(mov-
ing along the vertical axis), the system becomes stable in the cen-
ter and unstable at the greater values ju
1
j. In such a case, the
instability domain becomes symmetric along the vertical axis with
the minimum point at u
1
0. Figures 2(a), 2(b), 2(c) and 2(d)
present the plots for different values of b and show the same trend
with respect to a, but the region for a > 1 is much greater for
larger values of b. At the same time, for a < 1 and large values of
b the instability domain shrinks very sharply in comparison to the
small values of b.
It is possible to understand the mechanism of instability from
the plot of eigenvalue spectrum. Typical instability domains for
the same parameters as on Figs. 2(a) and 2(b) for k 0 are pre-
sented on the Fig. 3(a). The horizontal lines i, ii, iii on the upper
plot represent typical relations between s
1
and s
2
, when in the sys-
tem, depending on the value of a, qualitatively different instability
conditions can be realized. Eigenvalue spectrums for these typical
relations are presented on plots (i), (ii), (iii), correspondingly. Let
us analyze each of the possible situations in more detail.
Case (i): full type instability domain.
In this case, the instability domain consists of three sub-
domains with real and complex eigenvalues. The easiest way of
obtaining instability is realized at ju
1
j < u
E
1
when all eigenvalues
of the linearized system are real and positive (Fig. 3(a), (i)). This
region is represented by dark gray color. Positive eigenvalues
mean that the system is unstable practically for any value of
a > 0. Inside the domain ju
E
1
j < ju
1
j < ju
C
1
j the Hopf bifurcation
takes place for certain values of a from the interval 0 < a < 2.
Point D divides this region into two sub-domains where
Rek > 0 (gray color) and Rek < 0 (light gray color). In the
sub-domain Rek < 0 the system can be unstable for certain val-
ues of a > 1, which can be determined by the condition
a > a
0

2
p
jArgkj. In turn, for Rek > 0,, the system can be sta-
ble for a < 1 by the same reasoning. In other words, between
points C and E we have eigenvalues with an imaginary part, and
the value of a can change the stability of the FRD system. In the
domain ju
1
j > ju
C
1
j the eigenvalues are pure real and negative and;
as a result, the system is stable.
Case (ii): continuous instability domain with complex
eigenvalues.
For system parameters corresponding to the case (ii) the real
part of eigenvalues becomes less than zero for all values u
1
. At
the same time, for ju
1
j < ju
K
1
j the roots are complex and according
to the condition a > a
0

2
p
jArgkj instability takes place for
a > a
0
> 1. For ju
1
j > ju
K
1
j, the eigenvalues become real and
negative, and the system is stable.
Case (iii): separated instability domain with complex
eigenvalues.
In this case at the central part of the plot (ju
1
j < ju
G
1
j) the
eigenvalues are pure real and negative, and as a results the system
is stable. For ju
G
1
j < ju
1
j < ju
H
1
j eigenvalues are complex with neg-
ative real parts and like in the case (ii) instability takes place for
a > a
0
> 1. In other words, the instability domain consists of two
symmetrical regions of instability, separated by a stable region at
the center where the system is stable for any a. For ju
1
j > u
H
1
eigenvalues are real and negative again and the system is stable.
The presented analysis for homogeneous perturbation with the
wave number k 0 in a fractional RD system corresponds to an
investigation of a point system with l
1
0; l
2
0. Because of
this the same situations in the fractional systems of ordinary dif-
ferential equations can be realized.
4.1.2 Turing Instability. Conditions for Turing bifurcation
(k 6 0) can be analyzed on the basis of nonhomogeneous modes in
eigenvalue spectrum of the system. Eigenvalues for different val-
ues of k are presented in Fig. 3(b). The top plot represents the null-
clines of the system and determines the parameters b and A for
plot on Fig. 3(b), (iv). In Fig. 3(b), eigenvalues for k 1 and
k 2 and for comparison for k 0 are presented. It can be seen
from the picture (iv) that at intersection of null-clines in the neigh-
borhood of zero values of u
1
nonhomogeneous modes have much
greater values and we can expect a formation of stationary dissipa-
tive structures. If the ratio l
1
=l
2
is sufciently small, Turing bifur-
cation is dominant for all region ju
1
j < 1. Analyzing conditions,
(Eq. (20)) we can conclude that they are practically the same for
fractional and standard RDS. However, what is very important is
that the transient dynamics and the role of other modes in frac-
tional systems can be principally different. For this reason the nal
attractors in the fractional system can be also be different even
though the linear conditions of instability look the same.
The typical situation is presented in Fig. 3(b), (iv). Lets assume
the system parameters are close to the ones represented by point P
in the top plot in Fig. 3(b). For a given value Imk=Rek, we can
expect the next scenario of instability for homogeneous solution
corresponding to null-cline intersection in point P. The mode with
k 2 (Fig. 3(b), (iv)) is dominant and the stationary dissipative
structures with this wave number appear in the system for a wide
range of values of 0 < a < 2. Decreasing the value of a in point P
or moving the null-cline intersection to the coordinate center
enhances this tendency. In turn, increasing the value of a or mov-
ing the null-cline intersection from the coordinate center leads to
activation of other modes, including k 0 and stimulates
Fig. 3 Instability domains and the eigenvalues (Rek - black
lines, Imk - gray lines) for k 0; b 1:05 and different
proportions of s
1
=s
2
0:5 i; 1:0 ii; 2:0 iii - (a). The
null-clines for b 2:1; A 0:5 and eigenvalue spectrum for
different values of k (k 0 - hair-lines, k 1 - dash lines, k 2
- thick lines) - (b). The eigenvalues are presented for the follow-
ing parameters: l
2
1
=l
2
2
0:025; b 2:1; s
1
=s
2
0:1 iv,
l
2
1
=l
2
2
0:1; b 1:01; s
1
=s
2
0:6 v, l
2
1
=l
2
2
2:1; b 1:01; s
1
=s
2
3:5 vi.
Journal of Computational and Nonlinear Dynamics JULY 2012, Vol. 7 / 031001-5
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
oscillatory instabilities. Such a trend is quite general and by
changing the intersection point of null-clines or value of a we can
stimulate a different scenario of pattern formation. If the inuence
of eigenvalues for k 0 and k 6 0 is comparable we can expect
complex spatio-temporal dynamics.
4.1.3 Nonhomogeneous Hopf Instability. Above, we have
considered situations when a homogeneous solution is unstable
for Hopf or Turing bifurcations, which is typical for classical
RDS. In this subsection, the conditions for another type of insta-
bility nonhomogeneous Hopf bifurcation, which takes place
only in fractional RDS, will be demonstrated. The eigenvalue
spectrum for such instability is presented in Fig. 3(b), (v, vi).
From the plot we can see that outside the small domain in the cen-
ter the system is stable for k 0. At the same time, the spectrum
of the linearized system has the complex eigenvalues for k 6 0 in
wide range of null-cline intersections (Fig. 3(b), (v)). Moreover,
at l
1
=l
2
< 1 the plot consists of a separate domain for k 2 where
eigenvalues for other modes are pure real and negative. In this
domain at some critical values a
0
inhomogeneous oscillations
with this wave number can be expected.
Figure 3(b), (vi) presents the situation, where eigenvalues have
negative real part for all modes. In this case, standard RDS is
stable for any values of parameter ju
1
j > 1. In turn, for fractional
RD system we can satisfy conditions for nonhomogeneous Hopf
bifurcation (Eq. (21)) even for s
1
=s
2
> 1 and l
1
=l
2
> 1. This
situation can be predicted from a symmetrical view of expression
(26) for the system under consideration
T 2

b
_
= 1 u
2
1
_ _
l
2
1
s
2
l
2
2
s
1
l
2
1
s
2
l
2
2
s
1

1 u
2
1
_ _
s
2
=s
1
1
_ _
(33)
The plot of these surfaces, as a function of l
1
=l
2
and s
2
=s
1
, for dif-
ferent values of u
1
is presented in Fig. 4(a)4(d). The surfaces on
Figs. 4(a) and 4(b). demonstrate that at small and large values of
u
1
, the maximum of T is reached at the boundaries, where
l
1
=l
2
<< 1 (Fig. 4(a)) and s
2
=s
1
<< 1 (Fig. 4(b)). For values u
1
in neighborhoods of extremum points of null-cline Wu
1
; u
2
0
the optimal instability conditions are reached at a certain relation-
ship of parameters l
1
=l
2
and s
1
=s
2
(Figs. 4(c) and 4(d)).
Oscillatory structures for certain wave number k
0
6 0 can
appear, if at the given values of u
1
, l
1
=l
2
and s
1
=s
2
the fractional
derivative index a is greater than the one represented on the sur-
face and less than the one needed for Hopf bifurcations with
k 0. This means that only the perturbations with these wave
numbers k
0
are unstable, and they are unstable for oscillatory uc-
tuations. This situation is qualitatively different from the integer
RD system whether either Turing (k 6 0) or Hopf bifurcation
(k 0) takes place [9,13].
Moreover, the plot in Fig. 3(b), (vi) demonstrates that in the sys-
tem under consideration, we can choose the parameters when
Fig. 4 The view of the surface T in coordinates l
1
=l
2
; s
2
=s
1
for b 2 and different values of u
1
(u
1
0:1 - (a), u
1
5:0 -
(b), u
1
1:25 - (c), u
1
1:5 - (d))
031001-6 / Vol. 7, JULY 2012 Transactions of the ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
nonhomogeneous Hopf bifurcation at s
1
=s
2
> 1 and l
1
=l
2
> 1 can
be realized. In the central part of this plot we can see only pure real
and negative eigenvalues for k 0. At the same time for k 6 0
eigenvalues are complex. Thus, the conditions for Hopf bifurcation
can be realized only for a nonhomogeneous wave number. As a
result, perturbations with k 0 relax to the homogenous state, and
only the perturbations with k 1 or k > 1 become unstable and
the system exhibits inhomogeneous oscillations.
4.2 Space Nonhomogeneous Solutions. Lets assume the
homogeneous distribution of the variables u
1
; u
2
becomes unsta-
ble due to Turing bifurcation according to wavelength
l
1
l
2

1=2
$ L and the size of the domain L satises the inequal-
ities l
1
(L (l
2
. Then the system under consideration has the
stationary nonhomogeneous solution, which for A $ 0 and peri-
odic or neutral boundary conditions can be represented as [13,25]
u
1
x % tanh
L=2 x

2l
1
l
2
p
_ _
; u
2
% const; jj (1 (34)
or [12]
u
1
x %a

AA
c
_
cosmpx=L; u
2
x %b

AA
c
_
cosmpx=L
(35)
Fig. 5 Dynamics of pattern formation for u
1
variable. The results of computer simulations of the system at parameters:
a 0:8, A 0:25, b 2:1, l
2
1
0:025, l
2
2
1, s
1
=s
2
0:1 - (a); a 0:8, A 0:55, b 2:1, l
2
1
0:025, l
2
2
1, s
1
=s
2
0:1 - (b);
a 0:8, A 0:4, b 2:1, l
2
1
0:025, l
2
2
1, s
1
=s
2
0:1 - (c); a 0:8, A 0:45, b 2:1, l
2
1
0:025, l
2
2
1, s
1
=s
2
0:1 - (d);
a 1:6, A 0:01, b 1:05, l
2
1
0:05, l
2
2
1, s
1
=s
2
1:45 - (e); a 0:7, A 0:3, b 2:1, l
2
1
0:05, l
2
2
1, s
1
=s
2
0:2 - (f).
Journal of Computational and Nonlinear Dynamics JULY 2012, Vol. 7 / 031001-7
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
where A
c
corresponds to a critical value of bifurcation parameter
A, a; b are constants, m 1; 2; 3::., mp=L $ l
1
l
2

1=2
. The lin-
ear stability of the solutions Eqs. (34) or (35) can be determined
by Frechet operator which in this case is presented in the follow-
ing form
L
L
11
L
12
L
21
L
22
_ _

l
2
1
r
2
1 u
1
x
2
1
1 l
2
2
r
2
1
_ _
(36)
For a standard system (a 1) for s
1
s
2
the system is asymptoti-
cally stable and small perturbations d !0. That means that
all eigenvalues of the Frechet derivatives have Rek < 0. For the
solution (Eq. (34)) operator L
11
is a Schrodinger operator with
Poschl-Teller potential [26], eigenfunctions of which are known
and the maximum eigenvalue is greater than zero. The stability
condition of the whole Frechet derivative L is due to the damping
properties of second variable which lowers the maximum eigen-
value. For the solution (Eq. (35)) operators L
11
and L
22
form two
coupled Mathieu equations. These equations have the real parts of
all eigenvalues less than zero if the bifurcation is supercritical and
at least one real part of eigenvalues greater than zero if bifurcation
is subcritical [27]. In general, nding the spectrum of the operator
(Eq. (36)) is a sufciently complicated mathematical problem,
which in special cases can be solved by using approximate meth-
ods [13]. For the system with a 6 1 the stability conditions are
Fig. 6 Dynamics of pattern formation for u
1
(left column) and u
2
(right column) variables. The results of computer simula-
tions of the systems at parameters: A 0:01, a 1:8, b 1:01, l
2
1
0:02, l
2
2
1, s
1
=s
2
3:5 - (a)-(b); A 1:95, a 1:82,
b 1:01, l
2
1
0:1, l
2
2
1, s
1
=s
2
0:6 - (c)-(d); A 0:01, a 1:75, b 10, l
2
1
0:05, l
2
2
1, s
1
=s
2
0:05 - (e)-(f).
031001-8 / Vol. 7, JULY 2012 Transactions of the ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
completely different and depend on the values a. This means
that order of fractional derivative can also change the stability of
stationary spatially- nonhomogeneous solutions.
5 Spatio-Temporal Pattern Formation in Fractional
Van der Pol-Fitzhugh-Nahumo-Like System
The results of the numerical simulation of the fractional Van
der Pol-FitzHugh-Nahumo model are presented on Figs. 5 and 6.
From the pictures, we can see that the system demonstrates a rich
scenario of pattern formation: standard homogeneous oscillations,
Turing stable structures, interacting inhomogeneous structures
and inhomogeneous oscillatory structures.
Computer simulations conrm, that the variations in any system
parameter which qualitatively changes the eigenvalue spectrum
of the linearized system, can change also the system dynamics.
Spatiotemporal dynamics of the FRD system mainly by the maxi-
mum eigenvalues for the corresponding modes are determined. In
Figs. 5(a) and 5(c) we can see a formation of stationary dissipa-
tive structures as a result of inuence of maximal unstable mode
presented in Fig. 3(b), (iv) for a < 1. External parameters A; b
determine the intersection point, the slope of the isoclines in this
point and the power for each particular mode. In particular, for
A 0:25, null-clines intersect at the point where the dominant
value has an eigenvalue for mode with k 2 (thick lines on
Fig. 3(b), (iv)). This situation is preferable for Turing bifurcation.
For A 0:55 null-clines intersect at the point where inhomoge-
neous modes are not dominant and Hopf bifurcation takes place.
The characteristic feature for these two limit cases is the instanta-
neous formation of either dissipative structures (Fig. 5(a)) or ho-
mogeneous oscillations (Fig. 5(b)). Increasing the inuence of
Hopf bifurcation mode when the Turing one is dominant, or
increasing the Turing bifurcation mode when Hopf is dominant,
leads to more complicated transient dynamics (Figs. 5(c) and
5(d)). When conditions of these two instabilities coincide, we can
obtain either oscillatory inhomogeneous structures or space
modulated homogeneous oscillations (Figs. 5(e) and 5(f)). More-
over, when the real part of eigenvalues of linearized system is
close to zero, small variation of a can change the type of bifurca-
tion in the system. This trend is typical for any a 1.
For a > 1 the structure formation can be much more compli-
cated. Let us consider the bifurcation diagram presented in Figs.
2(b) and 2(d). It was already noted that for a given value a a
0
the region inside the corresponding curve is unstable for wave
numbers k 0 and outside - it is stable. From the viewpoint of
homogeneous oscillations, at s
1
=s
2
> 1 the system is stable near
u
1
0 (Fig. 2(b)). However, if we have l
1
<< l
2
, the system
becomes unstable according to Turing instability. As a result, we
expect the formation of stationary spatially-inhomogeneous
structures. In fact, rst only inhomogeneous uctuations grow in
amplitude and lead to inhomogeneous dissipative structures, simi-
lar to what is presented in Fig. 5(a). At the same time, at the
dynamics of pattern formation, the amplitude of the structures
increases, and at the maximum amplitude, the structures fall into
the domain where the homogenous solutions in the fractional
system are unstable. As result, we obtain a complex interaction of
Turing and Hopf bifurcations. Nonlinear dynamics of such an
interaction for activator and inhibitor variables are presented in
Figs. 6(a) and 6(b). Due to different evolution patterns for two
variables, the activator variable u
1
is presented in the left column
and the inhibitor one u
2
is presented in the right column.
In Figs. 6(c) and 6(d), the inhomogeneous oscillatory structures
are presented. Such struc- tures are obtained at eigenvalues with a
negative real part when the intersection of the null-clines for an
activator variable is located on the decreasing part of the null-
cline u
2
u
1
for the activator equation (Fig. 3(b)). The correspond-
ing eigenvalue spectrum for this situation is presented in Fig. 3(b),
(v). From the eigenvalue plots (Fig. 3) we can see corresponding
separated domains where inhomogeneous oscillatory modes with
k 2 are unstable. A successive increase of the fractional deriva-
tive index will increase the amplitude of the presented in Figs.
6(c) and 6(d) inhomogeneous oscillations. As a result, inhomoge-
neous oscillatory structures of large amplitude are realized in the
system.
Another type of sub-hyperbolical oscillatory structures with
more complicated dynamics are presented in Figs. 6(e) and 6(f)
for b 10, s
1
<< s
2
and a /2. As in the case considered above,
we have inhomogeneous oscillatory structures which in the region
of slow motion oscillate with a fast frequency of 1=s
1
. Such
behavior emerges due to hyperbolical properties of the activator-
inhibitor system at a approaching the value of 2.
6 Conclusion
In this article, a complex spatio-temporal pattern formation in
a basis fractional reaction-diffusion system is investigated. In con-
trast to a standard reaction-diffusion system with integer deriva-
tives, the fractional RDS possesses new properties connected with
the value of fractional derivative index. A fractional derivative
order can substantially change the eigenvalue spectrum and signif-
icantly enrich the nonlinear dynamics in RD systems. The frac-
tional derivative index plays a crucial role in pattern formation
because conditions of bifurcations depend substantially on its
value.
By the eigenvalue analysis we have studied instability condi-
tions for wave numbers k 0, and k 6 0 for different values of
external parameters of the fractional RDS. Nonlinear solutions,
obtained by computer simulation, show that the dynamics of the
system is mainly determined by most unstable modes of eigen-
value spectrum of the linearized system. When linear increments
are comparable for different types of bifurcation, we have an
interplay between Hopf and Turing modes. When a fractional de-
rivative index is changed from 0 to 1, the dissipative structures are
stationary if a limit cycle is damped at s
1
s
2
. Oscillatory dy-
namics at parameters these values of the fractional derivative
index can be realized only when s
1
<< s
2
. When a fractional de-
rivative index is changed from 1 to 2, the system demonstrate
more complex dynamics.
References
[1] Agrawal, O. P., Tenreiro Machado, J. A., and Sabatier, J., 2007, Advances in
Fractional Calculus: Theoretical Developments and Applications in Physics
and Engineering, Elsevier, Amsterdam.
[2] Kilbas, A. A., Srivastava, H. M., and Trujillo, J. J., 2006, Theory and Applica-
tions of Fractional Differential Equations, Elsevier, Amsterdam.
[3] Arteshock, Uchaikin, V. V., 2008, Fractional Derivative Method, Arteshock,
Ulyanovsk (in Russian).
[4] Henry, B. I., and Langlands, T. A. M., 2006, Anomalous Diffusion with Linear
Reaction Dynamics From Continuous Time Random Walks to Fractional
Reaction-Diffusion Equations, Phys. Rev. E, 74, 031116.
[5] Gaychuk, V., and Datsko, B., 2007, Stability Analysis and Oscillatory Structures
in Time- Fractional Reaction-Diffusion Systems, Phys. Rev. E, 75, 055201-4.
[6] Abad, E., Yuste, S. B., and Lindenberg, K., 2010, Reaction-Superdiffusion and
Reaction- Superdiffusion Equations for Evanescent Particles Performing
Continuous-Time Random Walks, Phys. Rev. E., 81, 031115.
[7] Kochubei, A. N., 2011, Fractional-Parabolic Systems, Potential Anal., (to be
published).
[8] Haubold, H. J., Mathai, A. M., and Saxena, R. K., 2011, Further Solutions of
Fractional Reaction-Diffusion Equations in Terms of the H-Function, J. Com-
put. Appl. Math., 235, pp. 13111316.
[9] Mendez, V., Fedotov, S., and Horsthemke, W., 2010, Reaction-Transport Systems:
Mesoscopic Foundations, Fronts, and Spatial Instabilities, Springer, New York.
[10] Weiss, M., and Nilsson, T., 2004, In a Mirror Dimly: Tracing the Movements
of Molecules in Living Cells, Trends Cell Biol., 14, pp. 267273.
[11] Zelenyi, L. M., and Milovanov, A. V., 2004, Fractal Topology and Strange
Kinetics: From Percolation Theory to Problems in Cosmic Electrodynamics,
Phys. Usp., 47, pp. 809852.
[12] Nicolis, G., and Prigogine, I., 1997, Self-Organization in Non-equilibrium
Systems, Wiley, New York.
[13] Kerner, B. S., and Osipov, V. V., 1994, Autosolitons, Kluwer, Dordrecht.
[14] Purwins, H.-G., Bodeker, H. U., and Amiranashvili, S., 2010, Dissipative
Solitons, Adv. Phys., 59, pp. 485701.
[15] Adamatzky, A., de Lacy Costello, B., and Asai, T., 2005, Reaction-Diffusion
Computers, Elsevier, Amsterdam.
[16] Henry, B. I., and Wearne, S. L., 2000, Fractional Reaction-Diffusion, Phys.
A, 276, pp. 448455.
Journal of Computational and Nonlinear Dynamics JULY 2012, Vol. 7 / 031001-9
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
[17] Seki, K., Wojcik, M., and Tachiya, M., 2003, Fractional Reaction-Diffusion
Equation, J. Chem. Phys., 119, 2165.
[18] Langlands, T., Henry, B. I., and Wearne, S. L., 2007, Turing Pattern Formation
with Fractional Diffusion and Fractional Reactions, J. Phys. Condens. Matter,
19, 065115.
[19] Gaychuk, V., Datsko, B., Meleshko, V., and Blackmore, D., 2009, Analysis
of the Solutions of Coupled Nonlinear Fractional Reaction-Diffusion Equa-
tions, Chaos, Solitons Fractals, 41, pp. 10951104.
[20] Gaychuk, V., Datsko, B., and Meleshko, V., 2008, Mathematical Modeling
of Time Fractional Reaction-Diffusion Systems, J. Comp. Appl. Math., 220,
pp. 215225.
[21] Podlubny, I., 1999, Fractional Differential Equations, Academic Press,
New York.
[22] Pierre, M., 2010, Global Existence in Reaction-Diffusion Systems with Con-
trol of Mass: a Survey, Milan J. Math., 78, pp. 417455.
[23] Nicolis, G., and Prigogine, I., 1989, Exploring Complexity: An Introduction,
Freeman & Co, New York.
[24] Matignon, D., 1996, Stability Results for Fractional Differential Equations
with Applications to Control Processing, Comput. Eng. Syst. Appl., 2, pp.
963970.
[25] Lubashevsky, I., and Gaychuk, V., 1994, Projection Dynamics of Highly
Dissipative Systems, Phys. Rev. E., 50, pp. 171181.
[26] Poschl, G., and Teller, E., 1933, Bemerkungen zur Quantenmechanik des
anharmonischen Oszillators, Z. Phys., 83, pp. 143151.
[27] Sattinger, D. H., 1973, Topics in Stability and Bifurcation Theory, Springer,
New York.
031001-10 / Vol. 7, JULY 2012 Transactions of the ASME
Downloaded 21 Dec 2012 to 216.91.96.130. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

You might also like