You are on page 1of 14

Applied Mathematical Modelling 45 (2017) 134–147

Contents lists available at ScienceDirect

Applied Mathematical Modelling


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

Bifurcations in a predator–prey model with general logistic


growth and exponential fading memoryR
Rodrigo Castro a,∗, Willy Sierra b, Eduardo Stange a
a
Instituto de Matemáticas. Universidad de Valparaíso, Casilla 5030, Valparaíso, Chile
b
Departamento de Matemáticas, Universidad del Cauca, Popayán, Colombia

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

Article history: A multiparameter predator–prey system generalizing the model introduced in [6] is con-
Received 23 December 2015 sidered. The system studied in this paper corresponds to the type of models with expo-
Revised 16 October 2016
nential fading memory where the logistic per capita rate growth of the prey is given by an
Accepted 8 December 2016
arbitrary function of class Ck , k ≥ 3. We prove that the model has a Hopf bifurcation and
Available online 19 December 2016
that there exist open sets in the parameter space such that the system exhibits singular
Keywords: attractors and asymptotically stable limit cycles. A numerical simulation is conducted in
Stability order to show the existence of critical attractor elements.
Logistic function As pointed out by Ayala et al. in [14], the Lotka–Volterra model of interspecific com-
Limit cycles petition, which is based on the logistic theory of population growth and assumes that the
Bifurcations intra and interspecific competitive interactions between species are linear, does not ex-
Predator–prey models plain satisfactorily the population dynamics of some species. This is due to fact that the
model does not take into account some important features of the population, which affect
its dynamics. The model introduced in this paper provides independent conditions of these
facts, for the existence of a Hopf bifurcation and the asymptotically stable limit cycles.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction

It is known that a generalized version of the classic Lotka–Volterra model is given by the system:
⎧  
⎪ dN N
⎨ dt =  N 1 − K − α NP,

⎩ dP = −γ P + β NP,
dt
where N and P correspond respectively to the population of prey and predator at each instant of time,  , γ , β > 0 and K
> 0 is the carrying capacity of the habitat. In the absence of predation (i.e. P = 0), the per capita rate growth of the prey is
linear and is given by:


 N

= 1−
N K

R
This paper is dedicated to the memory of Eduardo Stange Sturla.

Corresponding author. Tel.: +0056958739499.
E-mail addresses: rodrigo.castrom@uv.cl, rocacastro@hotmail.com (R. Castro), wsierra@unicauca.edu.co (W. Sierra), eduardo.stange@uv.cl (E. Stange).

http://dx.doi.org/10.1016/j.apm.2016.12.003
0307-904X/© 2016 Elsevier Inc. All rights reserved.
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 135

and it is said that the behavior of the growth rate of the prey population is of logistic type. A lot of models of interaction
of two species (predator–prey) are governed considering this behavior of the prey and a variety of growth curves have
been developed to model both unpredated, intraspecific population dynamics and more general biological growth. A good
comparison of several such models can be seen in [1–3].
In relation to the study of some models with competition between two species that exhibit bifurcations associated with
their dynamic behavior in which the logistic growth of the prey is not linear, known as θ -logistic growth, there are several
studies that show that such bifurcation is of Hopf type (see [3,4]). In this work, we see what we can get this type of
bifurcations generalizing the logistic growth function, associated to systems with exponential fading memory.
By the mid 1980, Farkas (see [5]) proposes a generalization of this model and Farkas et al. (see [6]) provide answers about
the behavior of the interaction of species indicating how it changes the dynamics of the system over time depending of the
parameters for which this is defined (existence of singular points and trajectories periodic attractors, bifurcations, etc.). This
model improves the above Lotka–Volterra model in the following two aspects. First, this takes into account intraspecific
competition of the prey, which has a saturation effect and therefore the prey does not grow at an exponential rate in the
absence of predator and tends to a finite limit. Second, they introduce a term memory: accepting that predator growth at
present depends on past quantities of prey and therefore a continuous weight (or density) function is introduced whose rol
and weigh moments of the past. Farkas’ model variations, particularly made in the differential equation that model the
behavior of the predator, have been studied in [7,8].
The model studied in [6] is given by:

⎪ dN
⎨ =  N f (N/K ) − α NP,
dt

⎩ dP = −γ P + β P  t N (ξ ) G(t − ξ ) dξ ,
−∞
dt
where f (N/K ) = 1 − N/K is the linear function associated with the logistic growth of prey and G : [0, +∞[→ R represents a
probability density function of class C1 .
The parameters of the model have the following meanings:

 = intrinsic growth rate of the prey;


α = maximun growth rate of the predator;
β = convertion growth rate of the prey;
γ = intrinsic growth rate of mortality by the predator;
K = environment carrying capacity of prey.

In that paper, is considered the probability density function G(s ) = as−as , a > 0 and, for this reason, we say that is a
system with exponentially fading memory because the largest weight is given to moments in the neighborhood of the present
and as we go back in time the weight is decreasing exponentially.
In this paper, we are interested in studying a generalization of this system and describe its dynamics maintaining the
characteristic of exponential fading memory but with a nonlinear function (general function) associated with logistic growth
rate of the prey. Our model is given by:

⎪ dN
⎨ =  N f (N/K ) − α NP,
dt
(1)

⎩ dP = −γ P + β P  t N (ξ ) G(t − ξ ) dξ ,
−∞
dt
γ
where G(s ) = as−as and f : [0, 1] → R+ is any function of class Ck , k ≥ 3, with f(0) > 0 and f  ( K β ) < 0. Here, we assume
that γ < Kβ .
We establish conditions for the existence of equilibrium points, periodic orbits and their type of stability and we show
that this family exhibits a Hopf bifurcation type.
In order to understand the dynamics of the model, we consider the auxiliary function
t
Q (t ) = N (ξ ) G(t − ξ ) dξ .
−∞

Its easy to see that dQ


dt
= a(N (t ) − Q (t )). Then the system (1) is C∞ -equivalent to the three dimensional system,
⎧ dN

⎪ dt =  N f (N/K ) − α NP,




dP
= −γ P + β P Q, (2)

⎪ dt

⎪ dQ

⎩ = a ( N − Q ).
dt
136 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

The equivalence between the systems (1) and (2) must be understood in the following sense. If we denote by C¯0 (−∞, 0] the
set of valued real and bounded functions in ] − ∞, 0] and (N, P ) : [0, ∞[→ [0, K] × R is the solution of (1) corresponding to
the initial function N̄ ∈ C¯0 (−∞, 0] and the initial value P0 = P (0 ), then (N, P, Q ) : [0, ∞[→ [0, K] × R × R+ is the solution of
t
(2) satisfying the initial condition N (0 ) = N̄ (0 ), P (0 ) = P0 and Q (0 ) = Q0 = −∞ N̄ (ξ ) eaξ dξ , and reciprocally.
To simplify the study of the dynamics of the model, we change the coordinates and the time in order to obtain a more
suitable system (normal form) as follows: {N → Kn, P → K p, Q → Kq, t → s }, where n, p, q denote the new coordinates,
respectively.
Then the system (2) is C∞ -equivalent to the system (or vector field):

⎪ Kα

⎪n˙ = n f (n ) − np,

⎪ 

Xμ : p˙ = − γ p + K β pq, (3)

⎪  



⎩q˙ = a (n − q ),

where the new parameters are μ = (a, K, α , β , γ ,  ) ∈ R6+ .
γ
Changing the parameters (using the same symbols for the new parameters) { α → α ,  → γ , β → β , a → a}, finally a
C∞ -equivalent five-parameters vector field is obtained :

⎨n˙ = n f (n ) − K α np,
Yη : p˙ = −γ p + K β pq, (4)

q˙ = a(n − q ),

where (n, p, q ) ∈ = {(n, p, q ) ∈ R3 : n, p, q ≥ 0} and the parameters are given by η = (a, K, α , β , γ ) ∈ R5+ .
γ
In order to show more clearly the results of the dynamic model associated with (4), let b = b(γ , β , K ) := K β and, in


the parameters space γ β a, we consider the open set (cylinder with triangular basis) C = (γ , β , a ) ∈ R3+ : 0 < b < 1 , the
parameter
f (b )
a = b f  (b ) − K β  ∈R
f (b )
and the open subsets

C1 = {(γ , β , a ) ∈ C : a < ā} and C2 = {(γ , β , a ) ∈ C : a > ā}.


Clearly, C1 is empty if a ≤ 0.

2. Main results

Lemma 1. For all (γ , β , a) ∈ C the vector field Yη has a unique singularity σ = (b, fK(αb) , b) ∈ = Int ( ). Furthermore, the
singularities (0, 0, 0) and (1, 0, 1) of Yη are hyperbolic saddle points.

Lemma 2. Let σ be the singularity of the vector field Yη of Lemma 1.

(i) If ā ≤ 0 then σ is an hyperbolic attractor point of Yη for all (γ , β , a) ∈ C.


(ii) If ā > 0 then:
(a) σ is a hyperbolic saddle point (or asymptotically unstable) of Yη if (γ , β , a) ∈ C1 .
(b) σ is an hyperbolic attractor point of Yη if (γ , β , a) ∈ C2 .
Moreover, in the parameter space γ β a there exists a two-dimensional submanifold Sā of class C k−1 which divides C in the
open connected components C1 and C2 .

In order to introduce the next result, consider the multiparameter polynomial

M (β , K, γ ) = 8F K 2 β 2 γ 10 G15 + 4F HK βγ 11 G14 − 4F K 3 βγ 10 G14 − 4F H 2 γ 12 G13 − 4F HK 2


γ 11 G13 − 16F 2 K 4 β 4 γ 9 G13 + 16K 4 β 4 γ 7 G13 − 6F 2 HK 3 β 3 γ 10 G12 + 8F 2 K 5 β 3 γ 9 G12 + 16HK 3 β 3
γ 8 G12 + 10F 2 H 2 K 2 β 2 γ 11 G11 + 8F 2 H K 4 β 2 γ 10 G11 + 4H 2 K 2 β 2 γ 9 G11 − 32F K 6 β 6 γ 6 G11 + 16K 6
β 6 γ 4 G11 − F 3 HK 5 β 5 γ 9 G10 − 8F 3 K 7 β 5 γ 8 G10 + 4F K 4 Sβ 4 γ 8 G10 − 4F HK 5 β 5 γ 7 G10 − 8F K 7 β 5
γ 6 G10 + 16HK 5 β 5 γ 5 G10 − F 3 H 2 K 4 β 4 γ 10 G9 + 4F 3 HK 6 β 4 γ 9 G9 + 4F 3 K 8 β 4 γ 8 G9 + 8F 4 K 8 β 8 γ 7
G9 − 12F HK 6 β 4 γ 7 G9 + 4F K 8 β 4 γ 6 G9 + 4H 2 K 4 β 4 γ 6 G9 + 16F 2 K 8 β 8 γ 5 G9 − 24F K 8 β 8 γ 3 G9
+F 4 HK 7 β 7 γ 8 G8 + 12F 4 K 9 β 7 γ 7 G8 − 4F 2 K 6 Sβ 6 γ 7 G8 − 12F 2 HK 7 β 7 γ 6 G8 + 2F 2 K 9 β 7 γ 5 G8
−20F HK 7 β 7 γ 4 G8 − 7F 4 H 2 K 6 β 6 γ 9 G7 − 14F 4 HK 8 β 6 γ 8 G7 − 8F 4 K 10 β 6 γ 7 G7 + 3F 2 H 2 K 6 β 6 γ 7
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 137

G7 + 12F 2 HK 8 β 6 γ 6 G7 − 8F 2 K 10 β 6 γ 5 G7 − 4F H 2 K 6 β 6 γ 5 G7 + 8F 3 K 10 β 10 γ 4 G7 − 4F 2 K 10 β 10
γ 2 G7 − F 5 HK 9 β 9 γ 7 G6 − 9F 3 K 8 Sβ 8 γ 6 G6 − 16F 3 HK 9 β 9 γ 5 G6 + 10F 3 K 11 β 9 γ 4 G6 − 12F 2 HK 9 β 9
γ 3 G6 − 8F 2 K 11 β 9 γ 2 G6 − F 5 H 2 K 8 β 8 γ 8 G5 + F 5 HK 10 β 8 γ 7 G5 − 6F 3 H 2 K 8 β 8 γ 6 G5 + F 3 HK 10
β 8 γ 5 G5 + 4F 3 K 12 β 8 γ 4 G5 − 5F 2 H 2 K 8 β 8 γ 4 G5 − 10F 4 K 12 β 12 γ 3 G5 − 4F 2 HK 10 β 8 γ 3 G5 + 10F 3
K 12 β 12 γ G5 − 8F 6 K 13 β 11 γ 5 G4 + 4F 4 K 10 Sβ 10 γ 5 G4 − 8F 4 HK 11 β 11 γ 4 G4 + 13F 3 HK 11 β 11 γ 2 G4
+8F 3 K 13 β 11 γ G4 + 2F 4 K 14 β 14 G3 + 7F 6 HK 12 β 10 γ 6 G3 + 4F 6 K 14 β 10 γ 5 G3 − 7F 4 H 2 K 10 β 10 γ 5 G3
+ 2F 4 HK 12 β 10 γ 4 G3 + 4F 3 H 2 K 10 β 10 γ 3 G3 + 4F 3 HK 12 β 10 γ 2 G3 + 2F 4 K 15 β 13 G2 + 6F 5 K 12 S
β γ G + 17F 5 HK 13 β 13 γ 3 G2 − 2F 5 K 15 β 13 γ 2 G2 + 3F 4 HK 13 β 13 γ G2 + F 7 HK 14 β 12 γ 5 G
12 4 2

−F 5 H 2 K 12 β 12 γ 4 G + F 5 HK 14 β 12 γ 3 G + F 4 H 2 K 12 β 12 γ 2 G + F 4 HK 14 β 12 γ G + F 6 K 14 Sβ 14 γ 3 + 4F 6 HK 15 β 15 γ 2 ,
where
F = f (b), G = f  (b), H = f  (b) and S = f  (b).
For δ > 0, we define the open subsets
C1,δ = {(γ , β , a ) ∈ C1 : a ∈ (ā − δ, ā )} and C2,δ = {(γ , β , a ) ∈ C2 : a ∈ (ā, ā + δ )}.
The next result shows that the vector field (4) (or system (1)) exhibits a bifurcation in the space of parameters R5+ for a = ā
and, taking into consideration Lemma 1 and according to the Poincaré–Andronov–Hopf theorem (see [9]), this will be a Hopf
bifurcation type.

Theorem 1. Let (γ , β , a) ∈ C and ā > 0. Then there exists δ > 0 and a neighborhood U ⊂ of σ = (b, fK(αb) , b) such that for all
a ∈ (ā − δ, ā ) ∪ (ā, ā + δ ) the vector field Yη has a unique periodic orbit = (a ) ⊂ U. Moreover:
(i) if (γ , β , a) ∈ C1, δ and M(β , K, γ ) < 0, then is an attracting hyperbolic periodic orbit of Yη .
(ii) if (γ , β , a) ∈ C2, δ and M(β , K, γ ) > 0 then is an hyperbolic periodic orbit of Yη of saddle-type (or unstable).

3. Proof of the main results

Proof of Lemma 1. Considering the system of equations



n f (n ) − K α np = 0
−γ p + K β pq = 0 ,
a (n − q ) = 0
it is clear that the only solutions are (0, 0, 0), (1, 0, 1) and
 1 
σ = b, f ( b ), b ,

γ
where b = K β . Since f is a positive function and b > 0, we have that σ ∈ .
Moreover, the linear part of Yη is given by:

f (n ) + n f  (n ) − K α p −K α n 0
DYη (n, p, q ) = 0 −γ + K β q Kβ p . (5)
a 0 −a
In particular,

f (0 ) 0 0
DYη (0, 0, 0 ) = 0 −γ 0 .
a 0 −a
It is clear that the singularity (0, 0, 0) is a saddle point. In addition, we can see that

f (1 ) + f  (1 ) −K α 0
DYη (1, 0, 1 ) = 0 −γ + K β 0 ,
a 0 −a
meaning that (1, 0, 1) is also a hyperbolic saddle point of Yη since the eigenvalues are λ1 = f (1 ) + f  (1 ) = 0, λ2 = K β − γ >
0, and λ3 = −a < 0. Note that b < 1 implies that K β − γ > 0. This proves Lemma 1. 

Proof of Lemma 2. From (5), we have that


⎛ ⎞
b f  (b ) − αγ
β 0
DYη (σ ) = ⎝ β ⎠.
0 0 α f (b )
a 0 −a
138 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

In order to determine the stability of the singularity σ of Yη , the characteristic equation associated with DYη (σ ) is given by:
   
Z 3 + a − b f  (b) Z 2 − ba f  (b) Z + aγ f (b) = 0. (6)

By the Routh–Hurwitz criterion (see [10]), it is known that σ is an hyperbolic attractor point (or asymptotically stable), if
and only if a0 , a1 , a2 > 0 and a1 a2 > a0 , where

a0 = aγ f (b),
a1 = −ba f  (b),
a2 = a − b f  (b).

Since f (b) < 0, the first condition is trivially satisfied. Now, to verify the second condition, σ is an hyperbolic attractor point,
if and only if,
  
a − b f  (b) −ba f  (b) > aγ f (b),
or equivalently,
γ f (b )
a > b f  (b ) − = ā.
b f  (b )
The above inequality is obvious if ā ≤ 0, because a > 0, which proves item (i) of Lemma 1. Next, if ā > 0 then σ is an
hyperbolic attractor point of Yη whenever (γ , β , a) ∈ C2 and σ is an hyperbolic saddle point (or asymptotically unstable) of
Yη whenever (γ , β , a) ∈ C1 . On the other hand, we define Sa = {(γ , β , a ) : F (γ , β , a ) = 0}, where
 γ  γ 2 γ  γ  γ 
F (γ , β , a ) = f −γ f −a f
Kβ Kβ Kβ Kβ Kβ
γ
is a function of class C k−1 , k ≥ 3. By direct calculation, ∂∂ Fa (γ , β , a ) = − K β f  (b) = 0. Then ∇ F(γ , β , a) = 0, which proves that
Sa is a submanifold two-dimensional of R3 of class C k−1 . Moreover, for all (γ , β , a) ∈ C, we have

F (γ , β , a ) = 0 ⇐⇒ (b f  (b))2 − γ f (b) − ab f  (b) = 0 ⇐⇒ a = a.


Therefore, Sa divides the open set C in the connected components C1 and C2 above defined. This concludes the proof of
Lemma 1. 

¯ we obtain the equation


Proof of Theorem 1. Introducing in (6) the parameter a,

Z 3 + (ā − b f  (b)) Z 2 − ā b f  (b)Z + āγ f (b) = 0,


whose roots are

Z0 (ā ) = b f  (b) − ā and Z1,2 (ā ) = ±wi,


where
 1/2
w = −āb f  (b) .
γ f (b )
It is clear that Z0 (ā ) = b f  (b) − ā = b f  (b )
< 0 and w > 0. Below, we will prove that the derivative with respect to the bifur-
cation parameter of the real part of the continuous extension of Z1 (ā ) is nonzero. In fact, we denote by Z1 (a) the root in
(6) that assumes the value wi in ā and we define
   
F (z, a ) = Z 3 + a − b f  (b) Z 2 − ab f  (b) Z + aγ f (b).
Now,
 
Fz (Z, a ) = 3Z 2 + 2 a − b f  (b) Z − ab f  (b).
Then,
γ
F (Z1 (ā ), ā ) = F (wi, ā ) = 0 and Fz (wi, ā ) = −2w2 − 2 b wi = 0.
By the Implicit Function theorem, there exist δ > 0 such that Z1 (ā ) = wi and F (Z1 (a ), a ) = 0 for all a ∈ (ā − δ, ā + δ ).
Moreover,
 2
d F  (wi, ā ) Z 2 − b f  (b)Z + γ f (b) w2 b2 f  (b)
(Z1 (ā ) ) = − a =− γ = − + iϒ
da Fz (wi, ā ) −2w2 − 2 b wi 2w2 b2 [ f  (b)] + 2γ 2 [ f (b)]
2 2

with ϒ ∈ R − {0}. Then,

d
[Re(Z1 (ā ))] < 0.
da
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 139

Introducing the parameter μ = 1a − 1a (or equivalently a(μ ) = 1+āāμ ), we have that a = a iff μ = 0. Also, μ < 0 (respec-
tively, μ > 0) iff (γ , β , a) ∈ C2 (respectively, (γ , β , a) ∈ C1 ). Now,

d d da
Re (Z1 (a(μ ) ) ) = Re (Z1 (a(μ ) ) )
dμ da dμ
 2
w2 b2 f  (b) −ā2
=− ·
2w2 b2 [ f  (b)] + 2γ 2 [ f (b)]
2 2
(1 + āμ )2
and therefore

d 
Re (Z1 (a(μ ) ) ) > 0. (7)

μ=0

We note also (and in correspondence with Lemma 1) that the singularity σ = (b, fK(αb) , b) ∈ is an hyperbolic attractor
point (respectively, an asymptotically unstable point) if μ < 0 (respectively, μ > 0).
Using (7) and δ > 0 above, by the theorem of Poincaré–Andronov–Hopf (see [9]) the vector field Yη has only one periodic

orbit = μ in a neighborhood of σ (whose radius is proportional to |μ|) for all μ ∈ (−δ, δ ) \ {0}.
Next, we will focus to establish the conditions for stability of the periodic orbit μ for different values of μ and this
deals with the fact prove that the vector field Yη admits a central manifold for μ = 0.
First, we consider the translation of the singularity σ to the origin (0, 0, 0) of the coordinate system {n → x + b, p →
y + K1β f (b), q → z + b}. For μ = 0 (or a = ā), the system (4) is C∞ -equivalent to the syste m
⎧  1


⎪x˙ = (x + b) f (x + b) − K α (x + b) y + f (b )

⎪ Kα

⎪  
⎨   1
Zη : y˙ = −γ y + K α f (b) + K β y + K α f (b) (z + b)
1
(8)



⎪  

⎩z˙ = (x − z ) b f  (b) − γ f(b) .

b f (b )

The Taylor expansion of the first component of Zη is given by:


 1

x˙ = (x + b) f (x + b) − K α (x + b) y + F

 1 2 1 3
 1
 
= (x + b) F + Gx + Hx + Sx − K α (x + b) y + F + O(|x|4 )
2 6 Kα
   
b 1 b
= (bG )x + G + H x2 + H + S x3 − K α xy − K α by + O(|x|4 ),
2 2 6

where

F = f (b), G = f  (b), H = f  (b) y S = f  (b).

Now, since the study of change of stability of the periodic orbit μ depends only on 3-jet of vector field Zη (see [11]), is
sufficient to consider the following local expression of (8):
⎧    
⎪ b 1 b
⎪x˙ = (bG )x + G + 2 H x + 2 H + 6 S x − K α xy − K α by
⎪ 2 3




K β (9)
⎪y˙ = yz + F z

⎪ β α



⎩ γ F
z˙ = (bG − )(x − z ).
bG
We observe that one basis of R3 of eigenvectors of DYη (σ ), for a = ā, is given by:
⎧ ⎛ b ⎞ ⎛ 2   2 ⎞⎫
⎪ ⎪


⎪  − f ( b ) ⎜ b f ( b ) ⎟⎪



⎨ 1 ⎜ w ⎟ ⎜ w2 ⎟⎬
⎜ β ⎟ ⎜ ⎟
E= 0 ,⎜ ⎟, bβ  ⎟⎪.

⎪ ⎝− α w f (b)⎠ ⎜⎝ − f ( b ) ⎠⎪


1
γα ⎪

⎩ ⎭
0 −1
140 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

In order to present the matrix of the linear part of system (9) as a Jordan matrix and using this basis, let us consider the
change of coordinates
⎧  2

⎪ b f  (b ) b2 f  (b)

⎪x = x1 − x2 + x3

⎨ w w2
β f (b ) bβ f  (b)

⎪y=− x2 − x3

⎪ wα αγ


z = x1 − x3 .
Thus, the system (9) in the new coordinate system has the form
⎧x˙ = wx + U x , x , x
⎪ 1 2 ( 1 2 3)


x˙ 2 = −wx1 + V (x1 , x2 , x3 ) (10)


⎩ Fγ
x˙ 3 = x3 + U (x1 , x2 , x3 ),
Gb
where the functions U and V are defined by:
 1
1
U ( x1 , x2 , x3 ) = C F w2 γ G + Hb x21 + C F Gγ (−2w2 + HGb3 )x22
2 2
γ bG2C   F γ wC
+ 2γ F 2 Gbβ 2 + HF G2 b4 β 2 + 2w4 x23 + (−2G2 b2 β 2 − HGb3 β 2 + γ Gb + F γ β 2 )x1 x2
2w2 β 2 bβ 2
γ F GC  3 3 2  Gγ C  
− G b β + HG2 b4 β 2 + γ w2 x2 x3 + 2F G2 b2 β 2 + Gbw2 + F w2 β 2 + F HGb3 β 2 x1 x3
wβ 2 β 2

1 1
+ F w γ C (3H + Sb)x1 + CF G b γ (3H + Sb)x1 x22
2 3 2 2
6 2
and
GbwC
 1
 1  
V ( x1 , x2 , x3 ) = − G+ Hb x21 − CG2 bw −2w2 + HGb3 x22
γ 2 2
GC  4 6 2 
− HG b β + 2F G3 b3 β 2 γ − 2F w2 γ 2 x23
2 wβ 2
C  2 3 
+ F γ − F Gbw2 β 2 γ + 2G3 b3 w2 β 2 + HG2 b4 w2 β 2 x1 x2
bβ 2
C  2 3  GwC  3 3 2 
+ −F γ + G5 b5 β 2 + HG4 b6 β 2 x3 x2 − 2G b β + Gbw2 β 2 − F γ 2 + HG2 b4 β 2 x1 x3
bβ 2 β2
1 4 4 1
+ CG b (3H + Sb)x32 + CG2 b2 w2 (3H + Sb)x21 x2 .
6 2
In order to determine the center manifold M of the system (10), tangent to the central space x1 x2 at the origin and
corresponding to the parameter μ = 0, suppose that its equation is given by:
M: x3 = h ( x1 , x2 ),
where h is Ck , k ≥ 3, and h(0, 0 ) = ∂∂xh (0, 0 ) = 0, for i = 1, 2.
i
If ∂∂xh (x, y ) = hi1 x1 + hi2 x2 + O(|x|2 ), where i = 1, 2 and x = (x1 , x2 ), by continuity hi1 = hi2 and we can assume that
i
h(x1 , x2 ) = 12 (h11 x21 + 2h12 x1 , x2 + h22 x22 ) + O(|x|2 ). Now, using (10) and the expansion in series of partial derivatives of h,
we can assume that the equation is given by the equality:
Fγ     
h ( x1 , x2 ) + U (x1 , x2 , h(x1 , x2 ) ) − h11 x1 + h12 x2 + O |x|2 wx2 + O |x|2
Gb
−(h12 x1 + h22 x2 + O(|x|2 ))(−wx1 + O(|x|2 )) = 0. (11)
In a neighborhood of (0, 0) small enough and omitting higher order terms we have that (11) can be can be expressed
as
1 Fγ 1  
h11 x21 + 2h12 x1 x2 + h22 x22 + CF Gγ −2w2 + HGb3 x22
2 Gb 2
 1
 CF γ w  
+ C F w2 γ G + Hb x21 + −2G2 b2 β 2 − HGb3 β 2 + γ Gb + F γ β 2 x1 x2
2 bβ 2

− wh11 x1 x2 − wh12 x22 + wh12 x21 + wh22 x1 x2 = 0.


R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 141

Hence,
CGb2 w2  2 2 2 
h11 = − HF β γ + 4F G3 bβ 2 γ + 4HF G2 b2 β 2 γ − 2F G2 γ 2
R
CF Gwγ  2 2 2 
h12 = − F β γ + F Gbγ 2 + 2G2 b2 w2 β 2 − 2G4 b4 β 2 − 2HG3 b5 β 2 (12)
R
CG2 b  
h22 = − 2F Gbw2 γ 2 − 4G2 b2 w4 β 2 − 4G4 b4 w2 β 2 + F 2 GHb3 β 2 γ 2 ,
R
where R = F 2 β 2 γ 2 + 4G2 b2 w2 β 2 .
To consider the system (10) restricted to the center manifold M, we introduce the new coordinates

u = x1 , v = x2 y3 = x3 − h(u, v ).
Hence, M = {(u, v, y3 ) :: y3 = 0} and the restriction of (10) to M corresponds to the system

u˙ = −wv + f (u, v, μ )
(13)
v˙ = wu + g(u, v, μ ).
For μ = 0, and using (12), this system has the form

u˙ = −wv + a20 u2 + a11 uv + a02 v2 + a30 u3 + a21 u2 v + a12 uv2 + a03 v3 ,
Wη : (14)
v˙ = wu + b20 u2 + b11 uv + b02 v2 + b30 u3 + b21 u2 v + b12 uv2 + b03 v3 ,
where the coefficients are given by:
  
a = CF w2 γ G + 1 Hb
 20
 2

a11 = CF γ w (−2G2 b2 β 2 − HGb3 β 2 + γ Gb + F γ β 2 )
 bβ 2


a02 = 1 CF Gγ (−2w2 + HGb3 )
 2

 CGγ 1
a30 = (2F G2 b2 β 2 + Gbw2 + F w2 β 2 + F HGb3 β 2 )h11 + CF w2 γ (3H + Sb)
 2β 2 6

a = − C γ F G  G3 b3 β 2 + H  G2 b4 β 2 + γ w2 h + CGγ  2F G2 b2 β 2 + Gbw2 + F w2 β 2 h + CGγ  F HGb3 β 2 h
 21 2 wβ 2
11
β2 12
β2 12


a12 = − C γ F G (G3 b3 β 2 + HG2 b4 β 2 + γ w2 )2h12 + 1 CF G2 b2 γ (3H + Sb)
 2 wβ 2
 2
 CG γ
 + (2F G b β + Gbw + F w β + F HGb β )h22
2 2 2 2 2 2 3 2
 2β 2

  
a03 = − C γ F G  G3 b3 β 2 + H  G2 b4 β 2 + γ w2 h22
 2 wβ 2


b20 = − CGbw (G + 1 Hb)
 γ 2

 C
b11 = 2 (F 2 γ 3 − F Gbw2 β 2 γ + 2G3 b3 w2 β 2 ) + HG2 b4 w2 β 2 )
 bβ

 1
b02 = − CG2 bw(−2w2 + HGb3 )
 2

b30 = − CGw 2G3 b3 β 2  2 + Gbw2 β 2  2 − F γ 2 + HG2 b4 β 2  2 h11
 2β 2

 1 1
b21 = (−F 2 γ 3 + G5 b5 β 2 + HG4 b6 β 2 )h11 + CG2 b2 w2 (3H + Sb)
 2bβ 2 2

 1  2 3   
b12 = 2 −F γ + G5 b5 β 2  3 + HG4 b6 β 2  3 h12 − CGw2 2G3 b3 β 2  2 + Gbw2 β 2  2 h22
 bβ  2β

 CGw  
 − −F γ 2 + HG2 b4 β 2  2 h22
 2β 2

 1 1
b03 = (−F 2 γ 3 + G5 b5 β 2 + HG4 b6 β 2 )h22 + CG4 b4 (3H + Sb).
2bβ 2 6
(15)
142 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

It is clear that, in the system of coordinates uv, the origin (0, 0) is a non-hyperbolic center-focus singularity of vector
field Wη because the trace T r (DWη )(0, 0 ) = 0 and the determinant Det (DWη )(0, 0 ) = w2 > 0.
In order to know the weakness of focus at the origin, we have to calculate the Lyapunov quantities and this are defined
through the focal values which are polynomials in the coefficients of the vector field (14). In fact, it is known that there
is an analytical function V, in a neighborhood of the origin, such that the rate of change along orbits, V, ˙ is of the form
η2 r2 + η4 r4 + · · ·, where r2 = x2 + y2 (see [12]). The focal values are the terms η2k . However, since they are polynomials, the
ideal they generate has a finite basis, so there exist ϑ ≥ 0 such that η2 = 0, for  ≤ ϑ, implies that η2 = 0 for all . The
value of ϑ is not known a priori so it is not clear in advance how many focal values should be calculated. The software
Mathematica (see [13]) is used to calculate the first few focal values. These are then “reduced” in the sense that each
is computed modulo the ideal generated by the previous ones: that is, the relations η2 = η4 = · · · = η2k = 0 are used to
eliminate some of the variables in η2k+2 . The reduced focal value η2k+2 , with strictly positive factors removed, is known as
the Lyapunov quantity.
Now, as T r (DWη )(0, 0 ) = 0, we have η1 = 0. Moreover, it is known that the focal value η2 depends only of the 3-jet of
vector field (14) and this is given by (see [11]):
η2 = (a02 a11 + a12 + a11 a20 + 3a30 + 2a02 b02 + 3b03 − b02 b11 )/8.
Using (12) and replacing the coefficients the system (14) in the expression of η2 we obtain:
M (β , K, γ )
η2 =   ,
8β 3 γ 2 k3 β 4 F 2 k4 + 4β 2 γ F G2 k2 − 4γ 2 G4 ( β 4 F 2 k4 + β 2 γ F G2 k2 − γ 2 G4 )2
where
M (β , K, γ ) = 8F K 2 β 2 γ 10 G15 + 4F HK βγ 11 G14 − 4F K 3 βγ 10 G14 − 4F H 2 γ 12 G13 − 4F HK 2
γ 11 G13 − 16F 2 K 4 β 4 γ 9 G13 + 16K 4 β 4 γ 7 G13 − 6F 2 HK 3 β 3 γ 10 G12 + 8F 2 K 5 β 3 γ 9 G12 + 16HK 3 β 3
γ 8 G12 + 10F 2 H 2 K 2 β 2 γ 11 G11 + 8F 2 H K 4 β 2 γ 10 G11 + 4H 2 K 2 β 2 γ 9 G11 − 32F K 6 β 6 γ 6 G11 + 16K 6
β 6 γ 4 G11 − F 3 HK 5 β 5 γ 9 G10 − 8F 3 K 7 β 5 γ 8 G10 + 4F K 4 Sβ 4 γ 8 G10 − 4F HK 5 β 5 γ 7 G10 − 8F K 7 β 5
γ 6 G10 + 16HK 5 β 5 γ 5 G10 − F 3 H 2 K 4 β 4 γ 10 G9 + 4F 3 HK 6 β 4 γ 9 G9 + 4F 3 K 8 β 4 γ 8 G9 + 8F 4 K 8 β 8 γ 7
G9 − 12F HK 6 β 4 γ 7 G9 + 4F K 8 β 4 γ 6 G9 + 4H 2 K 4 β 4 γ 6 G9 + 16F 2 K 8 β 8 γ 5 G9 − 24F K 8 β 8 γ 3 G9
+F 4 HK 7 β 7 γ 8 G8 + 12F 4 K 9 β 7 γ 7 G8 − 4F 2 K 6 Sβ 6 γ 7 G8 − 12F 2 HK 7 β 7 γ 6 G8 + 2F 2 K 9 β 7 γ 5 G8
−20F HK 7 β 7 γ 4 G8 − 7F 4 H 2 K 6 β 6 γ 9 G7 − 14F 4 HK 8 β 6 γ 8 G7 − 8F 4 K 10 β 6 γ 7 G7 + 3F 2 H 2 K 6 β 6 γ 7
G + 12F 2 HK 8 β 6 γ 6 G7 − 8F 2 K 10 β 6 γ 5 G7 − 4F H 2 K 6 β 6 γ 5 G7 + 8F 3 K 10 β 10 γ 4 G7 − 4F 2 K 10 β 10
7

γ 2 G7 − F 5 HK 9 β 9 γ 7 G6 − 9F 3 K 8 Sβ 8 γ 6 G6 − 16F 3 HK 9 β 9 γ 5 G6 + 10F 3 K 11 β 9 γ 4 G6 − 12F 2 HK 9 β 9


γ 3 G6 − 8F 2 K 11 β 9 γ 2 G6 − F 5 H 2 K 8 β 8 γ 8 G5 + F 5 HK 10 β 8 γ 7 G5 − 6F 3 H 2 K 8 β 8 γ 6 G5 + F 3 HK 10
β 8 γ 5 G5 + 4F 3 K 12 β 8 γ 4 G5 − 5F 2 H 2 K 8 β 8 γ 4 G5 − 10F 4 K 12 β 12 γ 3 G5 − 4F 2 HK 10 β 8 γ 3 G5 + 10F 3
K 12 β 12 γ G5 − 8F 6 K 13 β 11 γ 5 G4 + 4F 4 K 10 Sβ 10 γ 5 G4 − 8F 4 HK 11 β 11 γ 4 G4 + 13F 3 HK 11 β 11 γ 2 G4
+8F 3 K 13 β 11 γ G4 + 2F 4 K 14 β 14 G3 + 7F 6 HK 12 β 10 γ 6 G3 + 4F 6 K 14 β 10 γ 5 G3 − 7F 4 H 2 K 10 β 10 γ 5 G3
+2F 4 HK 12 β 10 γ 4 G3 + 4F 3 H 2 K 10 β 10 γ 3 G3 + 4F 3 HK 12 β 10 γ 2 G3 + 2F 4 K 15 β 13 G2 + 6F 5 K 12 S
β γ G + 17F 5 HK 13 β 13 γ 3 G2 − 2F 5 K 15 β 13 γ 2 G2 + 3F 4 HK 13 β 13 γ G2 + F 7 HK 14 β 12 γ 5 G − F 5 H 2 K 12 β 12 γ 4 G
12 4 2

+F 5 HK 14 β 12 γ 3 G + F 4 H 2 K 12 β 12 γ 2 G + F 4 HK 14 β 12 γ G + F 6 K 14 Sβ 14 γ 3 + 4F 6 HK 15 β 15 γ 2 . (16)
If we denote by ℵ = β 4 F 2 k4
+ 4β 2 γ F G 2 k2
= − 4γ 2 G4 β 4 F 2 k4 + 4γ G 2 ( β 2 F k2 − γ G2 ) the second factor of the denominator
of η2 , then ℵ > 0. In fact, by hypothesis ā > 0, then
 
γ2 γ2
γ ( f  (b))2 < f (b) ⇒ β Fk − γ G > β k −
2 2 2 2 2
f ( b ) = 0,
b2 b2
γ
since b = K β . From this, we can conclude that η2 and P0 (β , K, b, γ ) have the same sign.
Therefore, if M(β , K, γ ) < 0 ( respectively, M(β , K, γ ) > 0) the origin (0, 0) will be an attractor (respectively, a repelling)
hyperbolic focus of Wη corresponding to this, according to Lemma 1, to μ < 0 (respectively, μ > 0). Equivalently,

(i) if (γ , β , a) ∈ C1, δ and M(β , K, γ ) < 0, then = a is an attracting hyperbolic periodic orbit of Yη .
(ii) if (γ , β , a) ∈ C2, δ and M(β , K, γ ) > 0, then = a is an hyperbolic periodic orbit of Yη of saddle-type (or unstable).

This finishes the proof of theorem. 

Remark 1. The following table shows some models of interspecific competition, which are determined by eleven functions
depending on parameters (see [14] for more details). In the same way, as in the theta-logistic predator–prey models pro-
posed in [4,15,16], the functions studied in [14] give origin to different predator–prey models. Moreover, we can observe that,
from the eleven growth rates of the prey population size, five of them are of the form  N f (N/K ) − α NP, where the function
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 143

f is given explicitly in each case and N and P denote the prey and predator population, respectively, at time t. Therefore, the
theory developed in this paper shows that the predator–prey model with exponential fading memory associated to each one
of those five functions, have a Hopf bifurcation. Note that the first model is the classical Lotka–Volterra studied by Farkas
[5] and the seventh model gives origin to a new theta-logistic predator–prey model with exponential fading memory, which
we will analyze in the following section.
Model N˙ f (n )
1 ( N/K )(K − N − α P ) 1−n
2 ( N/ log K )(log K − log N − α log P ) Does not apply
  
3  N/K 1/2 K 1/2 − N1/2 − α P/K 1/2 1 − n 1/2
4 ( N/K )(K − N − α P − β NP ) Does not apply
 
5 ( N/K ) K − N − α P − β N2 1 − n − K β n2
 
6 ( N/K ) K − N − α P − β P2 Does not apply
  
7  N/K θ K θ − Nθ − α P/K 1−θ 1 − nθ
8 ( N/K )(K − N − α P − β (1 − exp (−γ N ) ) ) 1 − n − β (1 − exp (−γ Kn ) )
9 ( N/K )(K − N − α P − β (1 − exp (−γ P ) ) ) Does not apply
 
10 ( N/K ) K − N − α P − β NP − δ N2 Does not apply
 
11 ( N/K ) K − N − α P − β NP − δ N2 − γ p2 Does not apply
We remark that some other authors study a predator–prey model with Time Lag. For example, in [17], the model (17) below
is studied.
⎧ t
⎨N˙ (t ) =  N (t ) f (N/K ) − α N (t )P (t ) − K1 −∞ N (ξ )G(t − ξ )dξ ,
(17)
⎩ dP = −γ P (t ) + β P (t )  t N (ξ )G(t − ξ )dξ .
−∞
dt
This model is a modification of that studied by Farkas [5], who found that a Hopf bifurcation takes place at the value ā, of
“a”. In this regard, it is relevant to mention that the system (17) is obtained as a particular case of our model (1) by setting
t
f (n ) = 1 − n − n(ξ )G(t − ξ )dξ ,
−∞

and the Hopf bifurcation is guaranteed.

4. Application (Theta-logistic predator–prey model)

The following section, we present a theta-logistic predator–prey model with exponential fading memory. The theta-
logistic models have biological relevance as it has been shown in [14,15,18,19].
We consider the system (1), where f : [0, 1] −→ R+ is given by the theta-logistic growth function
f ( n ) = 1 − nθ , θ > 0.
Substituting f in (2), we obtain the system
⎧  

⎪n˙ = n 1 − nθ − K α np,

Yθ : p˙ = −γ p + K β pq, (18)



q˙ = a(n − q ).
γ
If Kβ
< 1, Lemma 1 implies that the system (18) has an equilibrium point located in the positive octant, which is given by:
   γ θ  γ 
γ 1
( n0 , p0 , q0 ) = , 1− , . (19)
K β αK Kβ Kβ

Furthermore, there exist the singularities (0, 0, 0) and (1, 0, 1) which are hyperbolic saddle points. Moreover, by defining ā
as
  γ θ 
1 γ
ā = − γ + θ2 +  θ ,
θ Kβ θ Kγβ
by Lemma 1, we have the following two cases:
144 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

If ā < 0, the system (18) has a stable equilibrium point at (n0 , p0 , q0 ), for all (γ , β , a) ∈ C, where
" γ #
C = (γ , β , a ) : 0 < <1 .

If ā > 0, in the parameter space, corresponding to γ , β and a, there exists a two-dimensional submanifold Sā of class
C k−1 which is defined implicitly by the equation:
 γ 2 θ  γ θ  γ θ
θ2 +γ + aθ −γ =0
Kβ Kβ Kβ
and divides C in the following two open connected components
C1 = {(γ , β , a ) ∈ C : a < ā} and C2 = {(γ , β , a ) ∈ C : a > ā}.
By Theorem 1, the stability of the equilibrium point depends on whether (γ , β , a) ∈ C1 or (γ , β , a) ∈ C2 . More precisely,
there exists δ > 0 and a neighborhood U ⊂ of (n0 , p0 , q0 ) such that for all a ∈ (ā − δ, ā ) ∪ (ā, ā + δ ) the vector field Yθ has
a unique periodic orbit = (a ) ⊂ U.
Moreover, the following theorem is obtained.

Theorem 2.
(i) if (γ , β , a) ∈ C1, δ and M(β , K, γ ) < 0, then is an attracting hyperbolic periodic orbit o f Yθ .
(ii) if (γ , β , a) ∈ C2, δ and M(β , K, γ ) > 0 then is an hyperbolic periodic orbit of Yθ of saddle-type (or unstable), where
C1,δ = {(γ , β , a ) ∈ C1 : a ∈ (ā − δ, ā )}, C2,δ = {(γ , β , a ) ∈ C2 : a ∈ (ā, ā + δ )}
and
M (β , b, γ , θ ) = −θ bθ −2 (β 4 (4θ 10 (θ + 1 )2 b10θ + γ 7 (θ 2 + θ − 2 )(bθ − 1 )6 + γ 6 (θ − 1 )θ 2 (bθ − 1 )5 ∗
(θ 5 b4θ − 5θ − 6 )b2θ − γ 5 θ 4 (bθ − 1 )4 (θ 5 (7θ − 15 )b4θ + 3θ (θ + 2 ) + 1 )b4θ
+γ 4 θ 3 (bθ − 1 )3 ((θ − 1 )θ 8 b8θ + θ 4 (15θ − 23 )b4θ + (θ + 1 )bθ − θ − 1 )b2θ + γ 3 θ 4 (bθ − 1 )2 ∗
(2θ (5θ − 13 )b8θ − θ 4 (θ (θ + 6 ) − 23 )b4θ − (θ + 1 )(4θ + 1 )bθ + (θ + 1 )(4θ + 1 ))b4θ
9

+γ 2 θ 6 (bθ − 1 )(4(θ − 3 )θ 9 b8θ − 4(θ − 5 )θ 4 (θ + 1 )b4θ − (θ + 1 )(5θ − 3 )bθ + (θ + 1 )(5θ − 3 ))b6θ


+4γ θ 8 (θ + 1 )(θ 4 (θ + 1 )b4θ + (θ + 2 )bθ − θ − 2 )b8θ + β 2 γ 3 θ (bθ − 1 )bθ (γ 6 (θ − 1 )(bθ − 1 )6
+γ 5 θ 2 (7θ − 15 )(−(bθ − 1 )5 )b2θ + γ 4 (bθ − 1 )4 ((θ − 1 )θ 4 b4θ + θ − 3 ) + 2γ 3 θ 2 (bθ − 1 )3
(θ 4 (7θ − 13 )b4θ − θ + 1 )b2θ + γ 2 (bθ − 1 )2 (4(θ − 3 )θ 8 b8θ + θ 4 (θ + 9 )b4θ − (θ + 1 )bθ + θ + 1 )
−2γ θ 2 (bθ − 1 )(4θ 9 b8θ + θ 4 (6θ − 5 )b4θ − 2(θ + 1 )bθ + 2(θ + 1 ))b2θ − 4θ 4 b4θ ∗
(θ b + θ 4 (3θ − 1 )b4θ − (θ + 1 )bθ + θ + 1 ) − 4γ 4 θ 2 (bθ − 1 )b2θ (θ 6 b6θ − γ 5 (bθ − 1 )5
9 8θ

+2γ 3 θ 4 (bθ − 1 )3 b4θ + γ 2 θ 2 (bθ − 1 )2 (θ 4 b4θ + 1 )b2θ + 2γ θ 4 (bθ − 1 )b4θ )))),


γ
with b = K β .

5. Numerical simulations

In this section, using PHASER software (see [20]), a numerical example will be presented. We establish the behavior and
stability of the trajectories of the system (4), also in the equivalent system at the base of eigenvectors (10) and finally its
restriction to the central manifold corresponding to the system (14). The logistic growth of the prey is not linear and given
by the hyperlogistic function (see [3])
f ( n ) = ( 1 − n )2 .
The following parameter values are chosen: K = 2, α = 1, β = 1,  = 1
2, γ = 1.
The system (4) is transformed into the system

⎨n˙ = n(1 − n ) − 2np,
2

Yη : p˙ = −p + 2 pq, (20)

q˙ = a(n − q ).
Then, b = the equilibrium point is σ = ( 12 , 16
1
2, , 2 ) and the bifurcation value is ā = 14 . Moreover, we obtain the following
1 1

values for the remaining parameters and coefficients which satisfy Lemmas 1 and 2 and Theorem 1:
F = 14 , G = −1, H = 2, S = 0, R = 18 , C = 64 1 7 3 11 1 6 2
5 , w = 2 , h11 = − 10 , h12 = − 10 , h22 = − 10 , a1 = − 5 , a2 = − 5 , a3 = 5 , c1 =
4 17 3 2 18 103 1
− 5 , c2 = 5 , c3 = 10 , b1 = 5 , b3 = 25 , d2 = 80 and d4 = 16 .
Now, considering that the value of bifurcation corresponds to μ = 0 (or ā = 1/4), the following figures (see Fig. 1, Fig. 2
and Fig. 3) show the behavior of the trayectories of Yη before and after this value and can see that the singular point σ is
asymptotically stable to negative values of μ.
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 145

Fig. 1. μ = −2 (or a = 1
2
).

Fig. 2. μ = 0 (or a = 14 ).

Fig. 3. μ = 6 (or a = 1
10
).

Furthermore, for these values the system (10) is rewritten in the form

⎧ 1 1 2 1 6 8 2 6

⎪x˙ 1 = x2 − x21 + x22 + x23 − x1 x2 + x2 x3 + x31 + x1 x22

⎨ 2 5 5 5 5 5 5 5
1 4 3 2 19 2 17 16 1 3
x˙ = − x1 − x21 + x − x + x1 x2 − x3 x2 − 2x1 x3 + x32 + x21 x2


2
2 5 10 2 10 3 5 5 5 5

⎩˙ 1 2 2 2 1 2 6 8 2 3 6
x3 = −x3 − x1 + x2 + x3 − x1 x2 + x2 x3 + x1 + x1 x2 , 2
5 5 5 5 5 5 5
146 R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147

Fig. 4. Traslation to the origin.

where their flow of trayectories is represented by Fig. 4. Finally, the system (14) restricted to the central manifold has the
form

⎨u˙ = 1 v − 1 u2 + 2 v2 − 6 uv − 14 u2 v − 309 uv2 + 1 u3 + 22 v3
2 5 5 5 25 800 32 25
⎩v˙ = − 1 u − 4 u2 + 3 v2 + 17 uv + 103 u2 v − 41 uv2 + 7 u3 + 1 v3
2 5 10 5 80 40 10 16

and computing the Lyapunov quantities we obtain that η1 = 0 and η2 = 1231800 > 0. Then, the origin is unstable equilibrium
point which implies that the periodic orbit = (a ) in Theorem 1 is asymptotically stable as shown in Fig. 5.

6. Discussion

In this work, we have studied a predator–prey model with general logistic growth and exponential fading memory. This
is the most general version for this type of interaction of two species proposed in [6] and we have proved that this model
has a Hopf bifurcation and that there are open sets in the space of parameters such that the system exhibits attractor
singular points and limit cycles asymptotically stable.
In order to understand dynamically the results obtained in this work, we prove that the system has a unique equilibrium
γ
in the interior of the positive orthant if b = K β < 1 (Lemma 1) and this inequality determines the admissible open region in
the parameter space where to focus the study of the dynamics of the system. The meaning of the condition γ < Kβ is: if the
mortality of the predator is low, and the carrying capacity and the conversion rate are large, then coexistence is possible.
If the converse is true, then the predator dies out. According to Lemma 1, if the parameters values lie in the admissible
region (the open set C ⊂ R3+ ) and the intrinsic growth rate of prey and the mortality of predator are large, and the delay
(1/a) and the limiting factor Kβ are small (i.e. (γ , β , a) ∈ C2 ), then the equilibrium is asymptotically stable. Furthermore, we
have determined a bifurcation surface (Sā ) which separates the dynamics of the system into two components. If the surface
is crossed by intrinsic birth rate of prey or the mortality of predator (keeping in all the cases the rest of the parameters
constant), then the equilibrium loses its stability by an Andronov–Hopf bifurcation (Theorem 1), i.e. periodic solutions, closed
trajectories occur in the neighborhood of the equilibrium. This implies that, in the case (i) of Theorem 1, for certain initial
conditions (near to the closed orbit) the interaction and coexistence between the two species will occur along the time and
there is no extinction phenomenon. The same conclusion is obtained in a neighborhood of the equilibrium point (which is
an hyperbolic attractor) in another open region of parameter space as it is shown in the case (ii) of Theorem 1.
R. Castro et al. / Applied Mathematical Modelling 45 (2017) 134–147 147

Fig. 5. Restriction to the center manifold.

Acknowledgment

The author wishes to thank the University of Cauca for providing time for this work through research project VRI ID
4387.

References

[1] A. Blumberg, Logistic growth rate functions, J. Theo. Biol. 21 (1968) 42–44.
[2] R. Buis, On the generalization of the logistic law of growth, Acta Biotheor. 39 (1991) 185–195.
[3] A. Tsoularis, Analysis of logistic growth models, Res. Lett. Inf. Math. Sci. 2 (2001) 23–46.
[4] S. Wang, Z. Ge, The Hopf bifurcation for a predator–prey system with θ -logistic growth and prey refuge, Abst. Appl. Anal. 2013 (2013) Article ID
168340, 13 pages, doi:10.1155/2013/168340.
[5] M. Farkas, Stable oscillations in a predator–prey model with time lag, J. Math. Anal. Appl. 102 (1984) 175–188.
[6] M. Farkas, A. Farkas, G. Szabo, Multiparameter bifurcation diagrams in predator–prey models with time lag, J. Math. Biol. 26 (1988) 93–103.
[7] S. Krise, S.R. Choudhury, Bifurcations and chaos in a predator–prey model with delay and a laser-diode system with self-sustained pulsations, Chaos,
Solitons and Fractals 16 (2003) 59–77.
[8] R. Shigui, Bifurcation analysis of a chemostat model with a distributed delay, J. Math. Anal. Appl. 204 (1996) 786–812.
[9] S. Wiggins, Introduction to applied nonlinear dynamical systems and chaos, Texts in Applied Mathematics, Springer-Verlag, New York, 1990.
[10] J.L. Willems, Stability Theory of Dynamical Systems, Wley, New York, 1970.
[11] F. Dumortier, Singularities of vector fields, J. Differential Equations 23 (1978) 53–106.
[12] N. Lloyd, Limit cycles of polynomial systems – some recent developments, in: New Directions in Dynamical Systems, Cambridge University Press,
Cambridge, 1988, pp. 192–234.
[13] S. Wolfram, The Mathematica Book, 5th ed., EE. UU. Wolfram Media, Incorporated, 2003.
[14] J. Ayala, E. Gilpin, J. Ehrenfled, Competition between species: theoretical models and experimental tests, Theor. Popul. Biol. 4 (1973) 331–356.
[15] D. Alstad, Basic Populus Models of Ecology, Preitice hall, Upper Saddle River, NJ, 2001. (Chapter 5)
[16] P. Stiling, Ecology: Theories and Applications, 2nd ed, Prentice Hall International, 1998.
[17] H. El-Owaidy, A.A. Ammar, Stable oscillations in a predator–prey model with time lag, J. Math. Anal. Appl. 130 (1988) 191–199.
[18] E. Gilpin, J. Ayala, Schoener’s model and Drosophila competition, Theor. Popul. Biol. 9 (1) (1976) 12–14.
[19] M. Sibly, B. Daniel, C. Denham, H. Jim, P. Mark, On the regulation of populations of mammals, birds, fish, and insects, Science 309 (5734) (2005)
607–610.
[20] J. Glick, H. Koçak, Phaser Scientific Software: A Universal Simulator for Dynamical Systems, LLC, 2003.

You might also like