Available online at www.sciencedirect.

com

Applied Mathematics and Computation 198 (2008) 251–260

www.elsevier.com/locate/amc

Inhomogeneous oscillatory solutions in fractional

reaction–diffusion systems and their computer modeling
a,* b
V. Gafiychuk , B. Datsko
a
Physics Department, New York City College of Technology, CUNY 300 Jay Street, Brooklyn, NY 11201, United States
b
Institute of Applied Problems of Mechanics and Mathematics, National Academy of Sciences of Ukraine,
Naukova Street 3 B, Lviv 79053, Ukraine

Abstract

We investigate inhomogeneous oscillatory instability conditions in fractional reaction–diffusion systems. It is shown

that non-linear stationary solutions emerge as a result of Turing instability becoming unstable according to oscillatory per-
turbation and transform to inhomogeneous oscillatory structures. It is also shown that with the certain value of the frac-
tional derivatives index, a new type of instability takes place and the system becomes unstable towards perturbations of
finite wave number. As a result, oscillatory perturbations with this wave number become unstable and lead to non-linear
oscillations which result in spatial oscillatory structure formation. Computer simulation of a Bonhoeffer–van der Pol type
reaction–diffusion systems with fractional time derivatives is performed.
Ó 2007 Elsevier Inc. All rights reserved.

Keywords: Reaction–diffusion system; Fractional differential equations; Oscillations; Dissipative structures; Inhomogeneous oscillatory
structures

1. Introduction

Inhomogeneous non-linear oscillatory structures in reaction–diffusion (RD) systems have been investigated
analytically, numerically and qualitatively during last decades (see, for example, [1–9]). Although strict math-
ematical theory of such type of phenomena have not been developed yet due to the essential non-linearity of
these systems, from the viewpoint of the applied and experimental mathematics, these phenomena are under-
standable and are investigated in different autowave media.
In the recent years, there has been a great deal of interest in fractional reaction–diffusion (RD) systems
[10–15] which, on one hand, exhibit self-organization phenomena and, one the other hand, are sufficiently
complicated in investigation of their solutions.
In this article, we will show that in fractional RD systems we get several new types of instabilities. That is
not possible to find in RD systems with integer derivatives. We confirm the linear stability analysis by

*
Corresponding author.
E-mail address: vagaf@yahoo.com (V. Gafiychuk).

0096-3003/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2007.08.065
252 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260

computer simulation of a Bonhoeffer–van der Pol type fractional RD system which is probably the simplest
one used in RD modeling.
Let us consider the fractional RD system
sc uat ¼ Duxx þ W ðu; AÞ; ð1Þ
T
with two components u = (u1, u2) defined on the interval x 2 ½0; Lx  subject to (i) Neumann:
ux jx¼0 ¼ ux jx¼Lx ¼ 0 or (ii) Periodic: uðt; 0Þ ¼ uðt; Lx Þ; ux jx¼0 ¼ ux jx¼Lx , boundary conditions and with the cer-
tain initial conditions. Here A is a real parameter, W = (W1, W2)T – smooth reaction kinetics functions, s and
D are positive, diagonal matrices s = diag[si], D ¼ diag½l2i .
Fractional derivatives c uat ; on the left hand side of the Eq. (1), instead of the standard time derivatives, are
the Caputo fractional derivatives in time of the order 0 < a < 2 and are represented as [16,17]
Z t
a oac uðtÞ 1 uðmÞ ðsÞ
u
c t ¼ :¼ ds;
ot a Cðm  aÞ 0 ðt  sÞaþ1m
where m  1 < a < m; m 2 1; 2.

2. Linear stability analysis

2.1. Spectrum analysis

The stability of the steady-state constant solutions of the system (1) corresponding to homogeneous equi-
librium state
W ðu0 ; A0 Þ ¼ 0 ð2Þ

can be analyzed by linearization of the system nearby this solution u0 ¼ ðu01 ; u02 Þ. In this case, the system (1) can
be transformed to linear fractional RD system at this equilibrium point
sc vat ¼ Dvxx  W v ðuh ; A0 Þv ð3Þ
By substituting the solution in the form u(x, t)  cos kx, ðk ¼ Lpx j; j ¼ 1; 2; . . .Þ, into Eq. (3) we can get the
system of linear ordinary differential equations with right hand side matrix
ða11  k 2 l21 Þ=s1 a12 =s1
F ¼ , which can be transformed to diagonal form. Here the coefficients
a21 =s2 ða22  k 2 l22 Þ=s2
aij = {oWi/ovj} represent Jacobian. Simple calculations lead to the next eigenvalues
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 2
k1;2 ¼ 2 ðtrF  tr F  4 det F Þ.
In this case, the solution of the diagonal vector equation is given by Mittag–Leffler functions [16,17]
X
1 k
ðki ta Þ
Dui ðtÞ ¼ Dui ð0Þ ¼ Ea ðki ta ÞDui ð0Þ; i ¼ 1; 2: ð4Þ
k¼0
Cðka þ 1Þ
Dui are the new variables obtained as a result of change of the basis corresponding to diagonalization of the
matrix F.
Analyzing (4), we can conclude that if for any of the roots
jArgðki Þj < ap=2 ð5Þ
the solution has an increasing function component, then the system is asymptotically unstable [18,19].
For integer a = 1, the roots k1,2 are complex inside the parabola detF = tr2F/4, and the fixed points are the
spiral sources (trF > 0) or spiral sinks (trF < 0). In this case, the domain on the right hand side of the parabola
(trF > 0) is unstable with the existing limit cycle, while the domain on the left hand side (trF < 0) is stable. By
crossing the axis trF = 0, the Hopf bifurcation conditions become true.
For a: 0 < a < 2 for every point inside the parabola detF = tr2F/4, we can introduce a marginal value
a : a ¼ a0 ¼ p2 jArgðki Þj which follows from the equality conditions (5) and is given by the formula [13]
 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260 253
( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
p
arctan 4 det F =tr2 F  1; trF > 0;
a0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6Þ
2  p2 arctan 4 det F =tr2 F  1; trF < 0:
The value of a is a certain bifurcation parameter which switches the stable and unstable states of the system.
At lower a: a < a0 = p2 jArgðki Þj, the system has oscillatory modes, but they are stable. Increasing the value of
a > a0 = p2 jArgðki Þj leads to instability. In fact, having complex number ki with Reki < 0, Imki50, it is always
possible to satisfy the condition jArg(ki)j < ap/2, and the system becomes unstable according to a certain type
of oscillations. The smaller is the value of trF, the easier it is to fulfill the instability conditions. With k = 0,
this type of analysis was made in [13,14]. In this article, we would like to direct your attention to the fact that
we can have another type of instability with k 5 0 and which leads to non-linear inhomogeneous oscillations
of the system parameters.

2.2. Classifying instabilities in fractional reaction–diffusion systems

Let us consider stability conditions for different possible limits. It is widely known for integer time deriv-
atives that the system (1) becomes unstable according to either Turing or Hopf bifurcations.
Let consider the conditions for Hopf bifurcation which are held at k = 0 if
trF > 0; det F ðk ¼ 0Þ > 0: ð7Þ

When we rewrite the last expression explicitly, we get

2
4a12 a21 s1 s2 > ða11 s2  a22 s1 Þ ; ð8Þ

which holds at a11 > 0, a22 < 0, a12a21 < 0, s1 < s2 and leads to homogeneous oscillations.
In the case of fractional derivative index, Hopf bifurcation is not connected with the condition a11 > 0 and
can hold at certain value of a when the fractional derivative index is sufficiently large. In this case, the easiest
way to satisfy this condition is by bringing the right hand side of (8) close to zero. As a result, we have
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 a12 a21 s1 2 a12 a21 s2
a0  2  arctan  2  2  arctan  2 :
p a11 s2 p a22 s1

This limit was analyzed in detail in [14].

The conditions for the Turing instability are
trF < 0; det F ðk ¼ 0Þ > 0; det F ðk 0 Þ < 0: ð9Þ
When we rewrite the last condition, we have
 2
ða11  k 2 l21 Þs2  ða22  k 2 l22 Þs1 > 4a12 a21 s1 s2 ð10Þ
In this case, eigenvalues are real and at a11 > 0; a22 < 0; a12 a21 < 0; l21  l22 the conditions of Turing insta-
bility for k050 lead to spatial pattern formation.
Let us consider a new possible situation when
trF < 0; 4 det F ð0Þ < tr2 F ð0Þ; 4 det F ðk 0 Þ > tr2 F ðk 0 Þ: ð11Þ
The analysis of the expressions (11) shows that at k = 0 we have two real and less than zero eigenvalues, and
the system is certainly stable. If the last inequality takes place for a certain value of k0 5 0, we can get two
complex eigenvalues. As a result, in the case of fractional derivatives, a new type of instability, connected with
the interplay between the determinant and the trace of the linear system, emerges. With such type of eigen-
values, it is possible to determine the value of fractional derivative index when the system becomes unstable.
In fact, the last two conditions can be rewritten as
2
ða11 s1  a22 s2 Þ > 4a12 a21 s1 s2 ; ð12Þ
 2
 4a12 a21 s1 s2 > ða11  k 2 l21 Þs2  ða22  k 2 l22 Þs1 : ð13Þ
254 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260

The simplest way to satisfy the last condition, is to estimate the optimal value of k = k0
 
a s  a s 
2  11 2 22 1 
k0 ¼  : ð14Þ
 s1 l22  s2 l21 
Having obtained (14), we can estimate the marginal value of a0
2
a0 ¼ 2  arctan T ; ð15Þ
p
where the expression T is
1=2
ð4a12 a21 s1 s2 Þ
T ¼  :
 l2 s þl2 s 
ða11 s2  a22 s1 Þ l12 s2 l22 s1   a11 s2  a22 s1
1 2 2 1

The last expression determines the value of a0 as a function of all parameters of the system. The greater is
the expression, the smaller is the value a0. Trying to reach the maximum possible value of (15), we can see
that it goes to zero if either s1 or s 2 goes to zero and, as a result, a0 ! 2. In the intermediate situation,
when s1  s2 the expression reaches its maximum. Analyzing the last expression, we can see that at
l22  l21 the denominator is very large and the right hand side tends to zero. For different lengths l22  l21 ,
or l22  l21 , s1  s2, the expression looks sufficiently simple and it determines instability conditions for inho-
mogeneous wave number (14)
( )
2 ð4a12 a21 Þ1=2
a0 ¼ 2  arctan ;
p jða11  a22 Þj  a11  a22
In the case of equality in (12), we can estimate the last expression as a0 ¼ 2  p2 arctan 1=2  1:7. This seems to
be the value of a0 which is close to its minimum value.

3. Inhomogeneous oscillatory structures

This section contains a discussion of the general results applied for a specific model as well as confirmation
of these results by computer simulation of the system (1) at the parameters which lead to the outcome con-
sidered in previous section.

3.1. Bonhoeffer–van der Pol model

Here in this section we discuss a Bonhoeffer–van der Pol type RD system with cubical non-linearity (see, for
example [20,2,3,12]). In this case, the source term for activator variable is non-linear W 1 ¼ u1  u31  u2 and it’s
linear for the inhibitor one W 2 ¼ u2 þ bu1 þ A . The null isoclines of the system are represented in Fig. 1a.
The homogeneous solution of variables u1 and u2 can be determined from the system of equations
W1 = W2 = 0 and to determine u1 we have cubic algebraic equation
ðb  1Þu1 þ u31 =3 þ A ¼ 0: ð16Þ
Simple calculation of the derivatives at homogeneous state (16) makes it possible to write conditions of dif-
ferent types of instability explicitly.

3.2. Computation scheme and simulation

System (1) with corresponding initial and boundary conditions was integrated numerically using the explicit
and implicit schemes with respect to time and centered difference approximation for spatial derivatives. Frac-
tional derivatives were approximated using two different schemes on the basis of Riemann–Liouville defini-
tion: L1-scheme for 0 6 a < 1, L2-scheme for 1 6 a < 2 (see below), as well as the scheme on the basis of
Grunwald–Letnikov definition for 0 < a < 1 and 1 < a < 2 [17]. In other words, for the system of n fractional
RD equations
 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260 255

C aj
o uj ðx; tÞ o2 uj ðx; tÞ
sj ¼ d j þ fj ðu1 ; . . . ; un Þ; j ¼ 1; n; ð17Þ
otaj ox2
where sj, dj, fj – certain parameters and non-linearities of the RD system correspondingly, we used the next
numerical schemes:
L1-scheme
!
dj ða Þ
X
k1
ðaj Þ
a
ðkDtÞ j 0
dj ukj;i  ðukj;i1  2ukj;i þ ukj;iþ1 Þ  fj ðuk1;i ; . . . ; ukn;i Þ ¼ dj u0j;i wk j þ ulj;i bklþ1 þ sj u ;
ðDxÞ
2
l¼1
Cð1  aj Þ j;i
aj
sj ðDtÞ ða Þ 1  aj
dj ¼ ; wk j ¼  k 1aj  ðk  1Þ1aj ; bðajÞ
¼ s1aj  2ðs  1Þ1aj þ ðs  2Þ1aj ;
Cð2  aj Þ k aj s

s ¼ 2; k;

L2-scheme

d j kþ1
dj ukþ1
j;i  2
ðuj;i1  2ukþ1 kþ1 kþ1 kþ1
j;i þ uj;iþ1 Þ  fj ðu1;i ; . . . ; un;i Þ
Dx
" #
X
k1 X
m
ðkDtÞpaj op 0
0 aj 1 aj l ðaj Þ k 2aj
¼ dj uj;i w0;k þ uj;i w1;k þ uj;i bklþ2 þ uj;i ð2  3Þ þ sj u ;
l¼2 p¼0
Cðp  aj þ 1Þ otp j;i
aj
sj ðDtÞ a ð1  aj Þð2  aj Þ ð2  aj Þ
dj ¼ ; w0;kj ¼  aj 1 þ k 2aj  ðk  1Þ2aj ;
Cð3  aj Þ k aj k
aj ð2  aj Þ
w1;k ¼ aj 1  2k 2aj þ 3ðk  1Þ2aj  ðk  2Þ2aj ;
k
2a 2a 2a
bsðaj Þ ¼ s2aj  3ðs  1Þ j þ 3ðs  2Þ j  ðs  3Þ j ; s ¼ 3; k

and G-L scheme

d j ðDtÞaj  ðDtÞaj
ukj;i  2
ukj;i1  2ukj;i þ ukj;iþ1  fj ðuk1;i ; . . . ; ukn;i Þ
sj ðDxÞ sj

aj
X
m pa
ðkDtÞ j op 0 X k
ða Þ
¼ ðDtÞ p
u j;i  cl j ukl
j;i ;
p¼0
Cðp  aj þ 1Þ ot l¼1
 
ða Þ ða Þ ða Þ 1 þ aj
c0 j ¼ 1; cl j ¼ cl1j 1  ; l ¼ 1; 2; . . .
l

where ukj;i uj ðxi ; tk Þ uj ðiDx; kDtÞ; m ¼ ½a.

The applied numerical schemes are implicit, and for each time layer they are presented as the system of
algebraic equations solved by Newton–Raphson technique. Such approach makes it possible to get the sys-
tem of equations with band Jacobian for each node and to use the sweep method for the solution of linear
algebraic equations. Calculating the values of the spatial derivatives and corresponding non-linear terms on
the previous layer, we have obtained explicit schemes for integration. Despite the fact that these algorithms
are quite simple, they are very sensitive and require small steps of integration, and they often do not allow to
find numerical results. In contrast, the implicit schemes, in certain way, are similar to the implicit Euler’s
method, and they have shown very good behavior at the modeling of fractional RD systems for different
step size of integration, as well as for a non-linear function and the power function of fractional index.
Moreover, by modeling according to this algorithm system (1) we have observed that these results fully
match the results obtained prior.
It should be noted that the definition of the fractional derivative in Grunwald–Letnikow form is equivalent
to the one in Riemann–Liouville method, but for numerical calculations it is much more flexible.
256 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260

3.3. Hopf bifurcation

The conditions for a Bonhoeffer–van der Pol type RD system are

2
4 det F  tr2 F ¼ 4ððb  1Þ þ u21 Þ=s1 s2  ð1  u21 Þ=s1  1=s2 > 0: ð18Þ
As a result, at stationary state (u1 , u2 ) is no more stable and limit circle arises in the system. Hopf bifurcation,
in the case of fractional differential equations can be realized for any value of a. The typical stability domain in
the coordinates (u1 ; s1 =s2 ) is represented in Fig. 1b. Inside this closed ‘‘boomerang’’ curve the system is unsta-
ble with wave numbers k = 0, and outside it is stable. The typical view of the eigenvalues is given in Fig. 2a.
We see that at k = 0 the system can have substantial imaginary part and under condition (18) is unstable. A
detailed analysis of the non-linear dynamics can be found in the paper [14]. It should be noted that for the
system with integer derivatives instability conditions for Hopf bifurcation are very strict (1 < u1 < 1,
s1 << s2).

3.4. Turing bifurcation

The explicit conditions of Turing instability are

 2
ð1  u21  k 2 l21 Þs2 þ ð1 þ k 2 l22 Þs1 > 4bs1 s2 ð19Þ
The typical behavior of the real and imaginary parts of eigenvalues is given in Fig. 2. The growing modes
exp(ikx) with wave numbers k (kmax > k > kmin > 0) and Reki > 0 will grow exponentially until the non-lin-
earities bind the growth. The wave number k and the wave length L corresponding to unstable perturbation,
are related by L ¼ 2p=k: Analyzing (19) we can conclude that these conditions are practically the same for
fractional derivatives and a standard RD system. But what is very important the transient processes and
the dynamics of these systems are different, and for this reason final attractors can often be different even
though the linear conditions of instability look the same. Namely to this type of dynamics the next subsection
is devoted. Here we would like to direct your attention to specific phenomena which are imminent to fractional
RD system only.

Fig. 1. Null isoclines for A ¼ 1:0, b = 1.1 (a), two-dimensional diagram of bifurcation domains in coordinates (u1,s1/s2) for different
values of a: 1a = 1.0, 2a = 1.2, 3a = 1.4, 4-a = 1.6, 5-a = 1.8 (b).

Fig. 2. Dependencies of real (black) and imaginary (grey) parts of eigenvalues on k (dispersion relations) showing unstable Turing mode
and transition of Turing mode to unstable Hopf mode for (s1/s2 = 3, 5, a = 1.8, l1 = 0.05, l2 = 1) u1 ¼ 0 (a), u1 ¼ 2 (b).
 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260 257

3.5. Non-linear interaction of the Turing and Hopf bifurcations

Let us consider the bifurcation diagram presented in Fig. 1b. It was already noted that the region inside the
curve is unstable for wave numbers k = 0 and outside it is stable. Now, let us consider that the system param-
eters are close to the one represented by point A. From the viewpoint of homogeneous oscillations, system is
stable. But if we have l1 << l2 = 1, the system becomes unstable according to Turing instability. The plot of
eigenvalues for such case is represented in Fig. 2a. As a result, we expect the formation of stationary inhomo-
geneous structures. In fact, at the beginning, only inhomogeneous fluctuations grow in amplitude and lead to
inhomogeneous pattern formation. At the same time, at the dynamics of structure formation, the amplitude of
the structures increases, and at maximum and minimum amplitude, the structures fall into the domain
(Fig. 1b), where the homogenous structures are unstable. In this case, parameters are close to Hopf instability
domain and Turing instability conditions (Fig. 2a) changes to Hopf bifurcation (Fig. 2b). The system become
unstable according to Hopf bifurcation. As a result, the paths of non-linear structures being inside unstable
region, start to oscillate and leave this region becoming stable. But in this case homogeneous distribution sat-
isfies Turing bifurcations conditions and becomes unstable again. Then stable structures develop and eventu-
ally they reach the domain where Hopf bifurcation becomes significant. The cycle of the oscillations repeats
again.

Fig. 3. Pattern scenario with transformation from Turing structures to oscillatory patterns for variables n1 (a), n2 (b) (s1/s2 = 3.5, k = 2),
similar structures for n1 with initial conditions for (k = 1) (c), Destroying of the structures for smaller value of s1/s2 = 2.5 and formation of
homogeneous oscillations in subcritical region (d), scenario of formation of steady state dissipative structures by increasing value
s1/s2 = 7.0 (e), Another example of oscillatory structures at s1/s2 = 4 and l1 = 0.01 (f). Parameters of the computer simulations were
n1 ¼ n2 ¼ 0, a = 1.8, l1 = 0.05 l2 = 1, b = 1.1. Initial conditions are: n01 ¼ n1  0:05 cos ðk 0 xÞ; n02 ¼ n2  0:05 cos ðk 0 xÞ.
258 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260

This mechanism was confirmed by computer simulation of the system. In Fig. 3 we see the oscillations of
dissipative structures at a certain value of s1/s2. Fig. 3a and b correspond to the first and second variables, and
the Fig. 3c corresponds to the same mechanism of structure formations with smaller period of the structures.
Decreasing the value of s1/s2 the inhomogeneous structures totaly disappear, and we obtain homogenous
oscillations similar to the ones obtained in the article [14]. It should be noted that in this case system is stable
according to small perturbation. This means that Hopf bifurcation is subcritical and the oscillations emerge by
finite initial perturbation.
In the case of increasing parameter s1/s2 amplitude of the inhomogeneous structure oscillations decreases
(Fig. 3d), and we obtain steady state structures similar to those we have in the systems with integer derivative
index (Fig. 3e). The dynamics considered in Fig. 3a–e corresponds to a certain value of l1/l2. If we decrease the
ratio of l1/l2, then similar to standard RDS structures become more contrast, but the same non-linear dynam-
ics is inherent to the system. An example of such dynamical structure is presented in Fig. 3f.

3.6. Inhomogeneous oscillatory structures

In the sections considered above, the linearized system is unstable either Hopf or Turing bifurcation. Below,
we consider the case when we don’t have Turing or Hopf bifurcation. In fact, as follows from (12), (13)
 2  2
ð1  u21 Þs2 þ s1 > 4bs1 s2 > ð1  u21  k 2 l21 Þs2 þ ð1 þ k 2 l22 Þs1 : ð20Þ
we have instability for inhomogeneous wave vectors with eigenvalues with imaginary parts. The real and imag-
inary parts for such case are represented in Fig. 4a. We see that the real part of the roots is always less than
zero, and the imaginary one on some interval of wave number k becomes non-zero. In this case, when the frac-
tional derivative index becomes greater than some critical value a0, the instability condition holds true. So, as
this instability conditions are possible to realize for some interval kmin < k < kmax, this means that only the
perturbations with these wave numbers are unstable, and they are unstable for oscillatory fluctuations. This
situation is qualitatively different from the integer RD system whether either Turing (k 5 0) or Hopf bifurca-
tion (k = 0) takes place, and this depends on which condition is easier to realize. In the system under consid-
eration, we can choose the parameter when we don’t have Turing and Hopf bifurcations (for k = 0) at all.
Nevertheless, we obtain that conditions for Hopf bifurcation can be realized for non-homogeneous wave
number.
Taking into account calculations made above we can estimate the value of T
pffiffiffiffiffi
2 bf
T ¼  ;
 l2 þl2 f
ðð1  u21 Þf þ 1Þ l12 fl2 2  þ ðu21  1Þf þ 1
2 1

where s2/s1 = f and which determines the marginal value of a0 (15). In Fig. 4b, a two-dimensional plot dis-
plays the parameter ranges for the stability and existence of dynamical structures. In the largest ‘‘boomerang’’

Fig. 4. The plot of real(black) and imaginary(grey) parts of eigenvalues on k for (s1/s2 = 2.2, l1 = 0.1, l2 = 1, b = 2, u1 ¼ 3:6 (a). Two-
dimensional bifurcation diagram domain in coordinates (u1, s1/s2) for a = 1.9, l1 = 0.1, l2 = 1 (b) and zoomed region (c). Shaded region
corresponds to instability with k = 0, white closed domain with points A and B corresponds to Turing instability with k = 1 and domain
with indicated point C – to wave number k = 2.
 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260 259

Fig. 5. Oscillatory inhomogeneous patterns for u1 and u2 for homogeneous state in point C (A ¼ 21, a = 1.95 ) – (a), (b) (see points in
Fig. 4c), distributions of variable u1 for homogeneous state in point A (A ¼ 50, a = 1.94 ) – (c) and in point B (A ¼ 80, a = 1.94 ) –
(d). For all figures l1 = 0.1, l2 = 1, b = 2.

domain the system is unstable according to wave numbers k = 0 [13,14]. For the case k50, we find instability
conditions for different wave numbers k = 1, 2, 3 . . . (Solution of the equality jArg(ki)j = ap/2 at certain a (5)).
We can see that these regions overlap and at the same parameters, the instability conditions for different re-
gimes are fulfilled simultaneously (Fig. 4b). As it is seen from the figure, there are conditions where only insta-
bility according to non-homogeneous wave numbers holds. As a result, perturbations with k = 0 relax to the
homogenous state, and only the perturbations with a certain value of k become unstable and the system exhib-
its inhomogeneous oscillations.
The results of the numerical study of the initial value problem of the system (1) are represented in Fig. 5a–d.
We have obtained that with the increase in the parameter a, the amplitude of the oscillatory structures
increases. The emergence of inhomogeneous oscillations, which destroy the stationary state, leads to a new
type of pattern formation. The resulting structures are rather similar to standing waves, than to standard
structures already investigated in autowave media.

4. Conclusion

In this article we considered two different mechanisms of inhomogeneous oscillatory structure formation in
fractional RD systems. The first mechanism is non-linear, and in linear spectrum analysis we can not predict
non-linear behavior because system becomes unstable due to either Turing or Hopf bifurcation. But non-lin-
ear structures become so developed, that a new type of instability emerges, and it changes its previous state for
a new one. Combination of these two types of instabilities leads to non-linear inhomogeneous oscillations.
Second mechanism of instability is in some way a new one as we do not come across it in the standard RD
systems. We shown here that at a sufficient value of fractional derivative index a, the system becomes unstable
according to inhomogeneous perturbations (k 5 0) with eigenvalues with imaginary part. As a result of this
instability, pattern formation can be represented as oscillatory structures, similar to inhomogeneous standing
waves in linear systems.

References

[1] V.V. Gafiychuk, B.S. Kerner, I.M. Lazurchak, V.V. Osipov, Osillationarry dissipative structures in autowave distributed media,
Mikroelektronika 15 (1986) 180–189 (in Russian).
[2] B.S. Kerner, V.V. Osipov, Autosolitons, Kluwer, Dordrecht, 1994.
260 V. Gafiychuk, B. Datsko / Applied Mathematics and Computation 198 (2008) 251–260

[3] A. Lubashevskii, V.V. Gafiychuk, The projection dynamics of highly dissipative system, Phys. Rev. E. 50 (1) (1994) 171–181.
[4] 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.
[5] P. Schutz, M. Bode, V.V. Gafiychuk, Transition from stationary to travelling localized patterns in two-dimensional reaction–diffusion
system, Phys. Rev. E. 52 (4) (1995) 4465–4473.
[6] V.V. Gafiychuk, A.V. Demchuk, Analysis of the dissipative structures in Gierer–Meinhardt model, J. Math. Sci. 88 (4) (1998) 48–53.
[7] C.B. Muratov, V.V. Osipov, Stability of the static spike autosolitons in the Gray-scott model, Siam J. Appl. Math. 62 (5) (2002) 1463–
1487.
[8] S.D. Webb, Oscillatory reaction–diffusion equations with temporally varying parameters, Math. Comput. Model. 39 (2004) 45–60.
[9] L. Yang, A.M. Zhabotinsky, I.R. Epstein, Stable squares and other oscillatory turing patterns in a reaction–diffusion model, Phys.
Rev. Let. 92 (19) (2004) 198303-1.
[10] B.I. Henry, S.L. Wearne, Existence of turing instabilities in a two-species fractional reaction–diffusion system, Siam J. Appl. Math. 62
(3) (2002) 870–887.
[11] B.I. Henry, T.A.M. Langlands, S.L. Wearne, Turing pattern formation in fractional activator–inhibitor systems, Phys. Rev. E. 72
(026101) (2005).
[12] V. Gafiychuk, B. Datsko, Pattern formation in a fractional reaction–diffusion system, Physica A. 365 (2006) 300–306.
[13] V. Gafiychuk, B. Datsko, V. Meleshko, Mathematical modeling of pattern formation in sub- and supperdiffusive reaction–diffusion
systems, arXiv:nlin.AO/0611005.
[14] V. Gafiychuk, B. Datsko, V. Meleshko, Non-linear oscillations and stability domains in fractional reaction–diffusion systems,
arXiv:nlin.PS/0702013.
[15] G. Jumarie, New stochastic fractional models for Malthusian growth, the Poissonian birth process and optimal management of
populations, Math. Comput. Modell. 44 (2006) 231–254.
[16] S.G. Samko, A.A. Kilbas, O.I. Marichev, Fractional Integrals and Derivatives: Theory and Applications, Gordon and Breach,
Newark, NJ, 1993.
[17] I. Podlubny, Fractional Differential Equations, Academic Press, 1999.
[18] D. Matignon, Stability results for fractional differential equations with applications to control processing, Computational Eng. in Sys.
Appl. 2, Lille, France 963 (1996).
[19] M. Moze, J. Sabatier, LMI tools for stability analysis of fractional systems, in: Proceedings of ASME 2005. paper# DETC2005-
85182.
[20] R.A. FitzHugh, Impulses and physiological states in theoretical models of nerve membrane, Biophys. J. 1 (1961) 445–466.