You are on page 1of 13

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267433204

Chaos in delay differential equations with


applications in population dynamics

Article in Discrete and Continuous Dynamical Systems · April 2013


DOI: 10.3934/dcds.2013.33.1633

CITATIONS READS

6 232

1 author:

Alfonso Ruiz-Herrera
University of Oviedo
34 PUBLICATIONS 134 CITATIONS

SEE PROFILE

All content following this page was uploaded by Alfonso Ruiz-Herrera on 29 October 2016.

The user has requested enhancement of the downloaded file.


DISCRETE AND CONTINUOUS doi:10.3934/dcds.2013.33.1633
DYNAMICAL SYSTEMS
Volume 33, Number 4, April 2013 pp. 1633–1644

CHAOS IN DELAY DIFFERENTIAL EQUATIONS WITH


APPLICATIONS IN POPULATION DYNAMICS

Alfonso Ruiz-Herrera
Departamento de Matemática Aplicada, Facultad de Ciencias
Universidad de Granada, 18071 Granada, Spain

(Communicated by Shouchuan Hu)

Abstract. We develop a geometrical method to detect the presence of chaotic


dynamics in delay differential equations. An application to the classical Lotka-
Volterra model with delay is given.

1. Introduction. For the last decades, delay differential equations have been con-
sidered as a natural framework to model many real world phenomena. These equa-
tions usually arise in models of population dynamics, neural networks or electrical
engineering since the use of time delay naturally appears to express the maturation
period of a concrete species, the transmission times along nerve axons, etc.
In this note we consider the system
 0
x (t) = x(t)(−a(t) − b(t)x(t) + c(t)y(t))
(1)
y 0 (t) = y(t)(d(t) − e(t)x(t) − f (t)y(t − τ ))
where τ > 0 and all the coefficients are positive and T -periodic. This system was
originally proposed by R. May in [25], the autonomous version, in order to describe
the evolution of two species, a herbivore (y) and a carnivore (x), sharing the same
environment. In system (1) we introduce the dependence on time to model the
seasonal effect of the environment, see [2] or [8]. Apart from this consideration, the
reader can find the biological interpretation of the parameters of model (1) in [25]
(also see [3]).
For system (1) we give sufficient conditions for the presence of chaotic dynamics.
Roughly speaking, by chaos in (1) we understand that the Poincaré operator, that
is,
P : D ⊂ R × C([−τ, 0], R) −→ R × C([−τ, 0], R)
z = (ξ, η) 7→ (x(T ; z), yT (z))
with yT (z)(t) := y(T + t; z), ((x(t; z), y(t; z)) is the solution of (1) with initial
condition at z) admits an invariant set Λ semi conjugate to the Bernoulli shift with
infinitely many periodic points and sensitive dependence on the initial conditions.
From the point of view of applications, these facts have deep consequences. For
instance, it is impossible to describe the dynamics in (1) from experimental data
since small errors in the initial values can produce great changes in the future

2010 Mathematics Subject Classification. 92D40, 37D50, 34K23, 34K60.


Key words and phrases. Topological horseshoes, delay, stretching along paths, global continu-
ation of horseshoes, predator-prey systems.
The author is supported by MTM 2008-02502, Ministerio de Educación y Ciencia, Spain.

1633
1634 ALFONSO RUIZ-HERRERA

(sensitive dependence). At this point it is important to observe that system (1) has
been extensively studied, see [2],[3], [16], [17], [22], [32]. However, to the best of
the author’s knowledge, it is the first time that the presence of complex dynamics
is detected analytically.
In relation to our method of proof, we combine the concept of topological horse-
shoes developed by Kennedy and Yorke in [12] (see also [37]) together with an
infinite dimensional variant of the notion of Stretching Along Paths developed by
Papini, Pireddu and Zanolin in [28], [29], [30]. This approach has a topological
character and involves the study of certain geometrical properties of the Poincaré
operator associated with (1). As an advantage, we point out that no hiperbolicity
conditions are required.
There is a broad literature concerning the presence of chaotic dynamics in delay
differential equations using different tools such as hyperbolicity and shadowing,
Conley index or Lefschetz fixed point theorem, see [5], [6] [18], [19], [20], [21],
[34],[35]. The first analytical results in this field were given by An der Heinden
and Walther in [6], [34]. In these results, the presence of chaotic dynamics was a
consequence of the classical Li-Yorke’s paper [23] since the dynamics in the infinite
dimensional space was essentially described by an interval map. In contrast, our
construction in this note is made directly in the infinite dimensional space and is
not based on any reduction of dimension.
Very recently, Wójcik and Zgliczyński in [36] also studied the presence of complex
dynamics in delay differential equations from a topological point of view. In this
interesting paper, the authors mainly proved that the notion of covering relation
introduced in [37] can be applied in delay differential equations with small delays (in
the sense of perturbation). In relation to [36], our results do not involve small/large
delays or any computer assisted proof.

Throughout the paper we employ the following notation:


L1T is the space of T -periodic integrable functions. Given two functions f (t), g(t) ∈
RT
L1T , their distance in this space is given by 0 |f (t) − g(t)| dt.
C([−τ, 0], R) denotes the space of continuous functions on [−τ, 0] with the max-
norm (k · k∞ ). In the space R × C([−τ, 0], R) we consider the norm k(x, y)k∞ :=
max{|x|, kyk∞ }.
Given a set A ⊂ R2 and a positive constant m,
XA,m := {(x, y) ∈ R × C([−τ, 0], R) : (x, y(0)) ∈ A, |y(t) − y(0)| ≤ m}.
Given P = (P1 , P2 ) : R × C([−τ, 0], R) −→ R × C([−τ, 0], R) an operator, we define
P ev as the map
P ev : R × C([−τ, 0], R) −→ R2
P ev (x, y) = (P1 (x, y), P2 (x, y)(0)).

2. Stretching along paths in infinite dimensional spaces. The aim of this


section is to present the notion of Stretching Along Paths in delay differential equa-
tions and to derive some consequences. First we describe precisely what we under-
stand by chaos or chaotic dynamics.
Definition 2.1. Consider (X, d) a metric space. Given ψ : Dψ → X a continuous
map and D ⊂ Dψ , we say that ψ induces chaotic dynamics on two symbols
on the set D if there exist two disjoint compact sets K0 , K1 ⊂ D such that, for
CHAOS IN DELAY DIFFERENTIAL EQUATIONS 1635

each two-sided sequence (si )i∈Z ∈ {0, 1}Z , there exists a corresponding sequence
(ωi )i∈Z ∈ DZ such that
ωi ∈ Ksi and ωi+1 = ψ(ωi ) for all i ∈ Z, (2)
and, whenever (si )i∈Z is a k-periodic sequence (that is, si+k = si , ∀ i ∈ Z) for some
k ≥ 1, there exists a k-periodic sequence (ωi )i∈Z ∈ DZ satisfying (2).
As mentioned in the introduction, our definition of chaos guarantees natural
properties of chaotic dynamics such as sensitive dependence on the initial conditions
or the presence of an invariant set Λ topologically transitive and semi-conjugate
with the Bernoulli shift. Besides these properties, it is important to observe that
our definition ensures the presence of infinitely many periodic points. Notice that
this property is not generally assumed in many definitions of chaos, (see Theorem
2.2 in [26] for the proofs of these properties and [1], [13] for the main definitions of
chaos).
Once these remarks have been done, we properly introduce the notion of Stretch-
ing Along Paths.
Definition 2.2. Let P : D ⊂ R × C([−τ, 0], R) −→ R × C([−τ, 0], R) be a compact
operator and consider a rectangle A = [a, b] × [c, d] ⊂ R2 , a constant m > 0, and
a closed set H ⊂ XA,m . We say that (H, P ) stretches XA,m along the paths and
write
(H, P ) : XA,m −→X
m A,m ,
if for every continuous path γ : [0, 1] −→ XA,m such that γ(0) ∈ {(x, y) ∈ XA,m :
x = a} and γ(1) ∈ {(x, y) ∈ XA,m : x = b}, there exists a subinterval [t0 , t00 ] ⊂ [0, 1]
satisfying that
• γ(t) ∈ H, P (γ(t)) ∈ XA,m for all t ∈ [t0 , t00 ],
• P (γ(t0 )) ∈ {(x, y) ∈ XA,m : x = a} and P (γ(t00 )) ∈ {(x, y) ∈ XA,m : x = b} or
viceversa.
The following result links the previous notion with our definition of chaos.
Theorem 2.3. Assume that there exist two disjoint compact sets K0 , K1 ⊂ A, a
constant m, and a compact operator P such that
(XK0 ,m , P ) : XA,m −→X
m A,m ,
(XK1 ,m , P ) : XA,m −→X
m A,m .
Then P induces chaotic dynamics on two symbols relatively to P (XA,m ) ∩ XK1 ,m
and P (XA,m ) ∩ XK0 ,m .
Before starting the proof of this theorem, we need the following results taken
from [26] and [28].
Lemma 2.4. Under the conditions of the previous theorem, given j ≥ 1 and a
(j + 1)-tuple (s0 , ..., sj ) with si ∈ {0, 1} and sj = s0 we have that
(I, P j ) : XA,m −→X
m A,m ,
where
I := {ω ∈ XKs0 ,m : P i (ω) ∈ XKsi ,m for all i = 0, ..., j − 1}.
Proof. This lemma is essentially Lemma A. 1 in [26].
1636 ALFONSO RUIZ-HERRERA

Theorem 2.5. (Theorem 8 in [28]) Let C 6= ∅ be a closed convex subset of the


normed space X. Let C := [−a, a] × C and define
Cl := {(−a, x) : x ∈ C}, Cr := {(a, x) : x ∈ C}.
Assume that
φ : C −→ φ(C)
is compact and there is a closed subset W ⊂ C such that the assumption
H: for every path σ ⊂ C with σ ∩ Cl 6= ∅ and σ ∩ Cr 6= ∅, there is a sub-path
γ ⊂ σ ∩ W with φ(γ) ⊂ C and φ(γ) ∩ Cl 6= ∅, φ(γ) ∩ Cr 6= ∅,
holds. Then there exists ze := (e e) ∈ W with φ(e
t, x z ) = ze.
Proof of Theorem 2.3. As a first step we prove that given j ≥ 1 and a (j + 1)-tuple
(s0 , ..., sj ) with si ∈ {0, 1} and sj = s0 , there exists a point ω ∈ XKs0 ,m such that
P i (ω) ∈ XKsi ,m and P j (ω) = ω.
Indeed, by Lemma 2.4, we deduce that
(I, P j ) : XA,m −→X
m A,m
where I is the set introduced in that lemma. Now, by Theorem 2.5, we conclude
the claim of this step.
Next we properly prove the assertion of our theorem. Fix a sequence (sj )j∈Z so
that sj ∈ {0, 1} for all j ∈ Z. After that, we define the compact sets Γi :=
{ω ∈ XKs0 ,m ∩ P (XA,m ) : P j (ω) ∈ XKsj ,m for all j = 0, ..., i} with i ∈ N. Clearly,
T∞
Γi+1 ⊂ Γi and by the previous step, Γi 6= ∅. Hence i=0 Γi is nonempty. The
assertion follows by a standard diagonal argument which enables us to extend the
result to bi-infinite sequences, see Theorem 2.2 in [11].

To finish this section we notice that the hypotheses of Theorem 2.3 can be mod-
ified to derive a result which is stable with respect to small perturbations of the
operator P . With this respect, we have the next result.
Corollary 1. Let A, K0 , K1 and m be as in Theorem 2.3. Suppose that there exists
a constant r > 0 so that (XKi ,m , P ) : XA,m −→X
m A,m , for i = 0, 1 with
K0 , K1 ⊂ ]a+r, b−r[×[c, d] , P ev (XK0 ,m )∩A, P ev (XK1 ,m )∩A ⊂ [a, b]×]c+r, d−r[

and P (XK0 ,m ), P (XK1 ,m ) ⊂ {(x, y) ∈ R × C([−τ, 0], R) : |y(t) − y(0)| < m − r}.
Then every compact operator Pε : XA,m → Pε (XA,m ) is chaotic provided that
||Pε (x, y) − P (x, y)|| ≤ ε, for all (x, y) ∈ XA,m
r
for all 0 < ε < 4.

Proof. We notice that, under the previous conditions,

e −→X
(XK0 ,m , Pε ) : XA,m e ,
m A,m

e −→X
(XK1 ,m , Pε ) : XA,m m A,m
e

with
Ae := [a + r, b − r] × [c, d].
Finally we use Theorem 2.3 to finish the proof.
CHAOS IN DELAY DIFFERENTIAL EQUATIONS 1637

3. A herbivore-carnivore system. The aim of this section is to apply the pre-


vious results to (1). With this purpose, firstly we consider the T -periodic system
 0
x (t) = g1 (t, x(t), y(t), y(t − τ ))
(3)
y 0 (t) = g2 (t, x(t), y(t), y(t − τ ))
defined as
x0 (t) = x(t)(−a1 + c1 y(t))

(4)
y 0 (t) = y(t)(d1 − e1 x(t)) for t ∈ [nT, nT + T1 [
x0 (t) = x(t)(−a2 − b2 x(t) + c2 y(t))

(5)
y 0 (t) = y(t)(d2 − f2 y(t − τ )) for t ∈ [nT + T1 , (n + 1)T [
where all parameters are strictly positive and 0 < T1 < T . For convenience we
employ the notation G to denote the map associated with the right-hand side of
system (3) and T2 := T − T1 .
From a mathematical point of view, the dynamics of (3) is described in the follow-
ing way: If (x(t; (p1 , p2 )), y(t; (p1 , p2 ))) denotes the maximal solution with initial
condition (p1 , p2 ) ∈ R × C([−τ, 0], R), we have that (x(t; (p1 , p2 )), y(t; (p1 , p2 ))) is
a solution of (4) for t ∈ [0, T1 [, it is a solution of (5) for t ∈ [T1 , T [ and the same
transition is repeated in a T -periodic manner. Hence, using that system (5) is
autonomous, the Poincaré operator associated with (3), that is
P : R × C([−τ, 0], R) −→ R × C([−τ, 0], R)
P (p1 , p2 ) = (x(T ; (p1 , p2 )), yT (p1 , p2 )),
satisfies that P = Φ ◦ Ψ with

Ψ : R × C([−τ, 0], R) −→ R × C([−τ, 0], R)
(6)
Ψ(p1 , p2 ) = (x1 (T1 ; (p1 , p2 )), (y1 )T1 (p1 , p2 ))

Φ : R × C([−τ, 0], R) −→ R × C([−τ, 0], R)
(7)
Φ(p1 , p2 ) = (x2 (T2 ; (p1 , p2 )), (y2 )T2 (p1 , p2 ))
where (x1 (t; (p1 , p2 )), y1 (t; (p1 , p2 ))) and (x2 (t; (p1 , p2 )), y2 (t; (p1 , p2 ))) are the solu-
tions of system (4) and (5) with initial condition at p = (p1 , p2 ) respectively.

Observe that T -periodic systems as (3), i.e. “piecewise autonomous systems” are
employed in many biological models see [8], [9], [10], [14], [15], [24], [27], [31]. In
particular, in population dynamics such a type of equations are known as systems
with seasonal succession, see [8] or [10].
Proposition 1. Consider system (3) with all parameters fixed except d1 , T1 , T2 , τ
and suppose that
3d2 a2 5d2
< < . (8)
4f2 c2 4f2
Then there exist a constant T2∗ and three maps d∗1 (Te2 ), T1∗ (de1 , Te2 ), and τ ∗ (de1 , Te1 , Te2 )
with the following properties:
The Poincaré map associated to (3) with parameters (T2 , d1 , T1 , τ ) is chaotic pro-
vided 0 < T2 < T2∗ , d1 > d∗1 (T2 ), T1 > T1∗ (d1 , T2 ), and 0 < τ < τ ∗ (d1 , T1 , T2 ).
Remark 1. In the proof we give precise estimates of T2∗ , d∗1 (T2 ), T1∗ (d1 , T2 ) and
τ ∗ (d1 , T1 , T2 ) depending on the coefficients of the system.
Next we state the main result of this paper.
1638 ALFONSO RUIZ-HERRERA

Theorem 3.1. Fix the parameters in (3) such that condition (8) is satisfied and
such that 0 < T2 < T2∗ , d1 > d∗1 (T2 ), T1 > T1∗ (d1 , T2 ) and 0 < τ < τ ∗ (d1 , T1 , T2 ).
Then there exists  > 0 such that if the distance in L1T between the previous pa-
rameters in (3) and the coefficients of (1) is smaller than , the Poincaré operator
associated to (1) is chaotic.
In our setting, the assumption of Theorem 3.1 means that the coefficient a(t)
satisfies Z T1 Z T
|a(t) − a1 |dt + |a(t) − a2 |dt < ,
0 T1
and so on (for the other coefficients).
There are some remarks to be made concerning Theorem 3.1. The coefficients in
(1) considered in the previous theorem are not necessarily “piecewise constant” and
can be taken as smooth as we like. The reader can see Theorem 3.1 as a result of
global continuation of topological horseshoes with respect to τ .

4. Proofs. Proof of Proposition 1. We split the proof into four steps.

Step 1: Construction of the rectangle and some estimates.


First we consider
  
3 d2 3 d2 7 d2 d2 d2
T2∗ := max{T2 : + T2 d2 − f2 − ≥ ,
2 f2 2 f2 4 f2 f2 4f2
  
1 d2 3 d2 1 d2 d2 d2
+ T2 d2 − f2 − ≥ }. (9)
2 f2 4 f2 4 f2 f2 4f2
After that we fix three constants T2 , α, and l satisfying that 0 < T2 ≤ T2∗ , α > l > 0
and
5 d2
− a2 − b2 (α + l) + c2 > 0, (10)
4 f2
5 d2
(α − l) + T2 (α − l)(−a2 − b2 (α + l) + c2 ) > α + l, (11)
4 f2
3 d2
− a2 − b2 (α − l) + c2 < 0, (12)
4 f2
3 d2
(α + l) + T2 (α − l)(−a2 − b2 (α − l) + c2 ) < α − l. (13)
4 f2
To prove the existence of these constants we observe that the previous inequalities
d
−a2 + 45 c2 f2
hold for l = 0 and 0 < α < b2 , (see (8)). Now we define the rectangle
2

 
d2 3d2
A = [α − l, α + l] × , .
2f2 2f2
Next we estimate d∗1 , (see (4) as a system without delay). Take d1 such that
d1
e1 > (α+l). It is clear that if (x1 (t), y1 (t)) is a solution of (4) with (x1 (t), y1 (t)) ∈ A
then we can compute the derivative of the first component with respect to the second
one. In such a case
dx1 x1 (−a1 + c1 y1 ) (α + l) max{| −a1 + 3c2f1 d2 2 |, | −a1 + c2f1 d2
2
|}
= ≤ d2
= K.
dy1 y1 (d1 − e1 x1 ) 2f2 (d1 − e1 (α + l))
Thus, if
d2 l
K < , (14)
f2 2
CHAOS IN DELAY DIFFERENTIAL EQUATIONS 1639

we deduce that the solutions of (4) with initial conditions at (α − 2l , dl22 ) and (α +
l d2 d2
2 , l2 ) leave the rectangle A across the sides [α − l, α + l] × { 2f2 } and [α − l, α +
3d2
l] × { 2f2 }. We illustrate this behavior with the following figure.

A
(d1 /e1 , a1 /c1 )

Figure 1. We draw the solutions of (4) with initial conditions at (α −


l d2
, )
2 l2
and (α + 2l , dl22 ) . Observe that these solutions leave the rectangle
d2
across the sides [α − l, α + l] × { 2f 2
} and [α − l, α + l] × { 3d
2f2
2
}

A straightforward computation shows that if d1 > d∗1 with


3c1 d2 c1 d2
4(α + l) max{| −a1 + 2f2 |, | −a1 + 2f2 |}
+ e1 (α + l) = d∗1 , (15)
l
condition (14) holds. Notice that d∗1 depends on the coefficients of system (4), ex-
cept d1 and T1 , and on the coefficients of system (5), including T2 . For the rest of
the proof we fix a constant d1 such that d1 > d∗1 .

Step 2: Construction of the compact sets K0 , K1 , and the maps T1∗ , τ ∗ .


First of all, we introduce the concept of rotation number associated to system
(4). Given q = (q1 , q2 ) ∈ IntR2+ \{( de11 , ac11 )}, the solution of (4) with this initial
condition, namely (x1 (t; q), y1 (t; q)), is a periodic orbit determined by the energy
function
E(x1 , y1 ) = c1 y1 + e1 x1 − a1 log y1 − d1 log x1 .
Using this fact, we can define the rotation number as
   
Z τ y1 (t; q) − a1 X1 (t) − x1 (t; q) − d1 X2 (t)
1 b1 e1
rot(q, τ ) := 2  2 dt,
2π 0

x1 (t; q) − de11 + y1 (t; q) − ab11
where
X1 (t) := x1 (t; q)(−a1 + c1 y1 (t; q)), X2 (t) := y1 (t; q)(d1 − e1 x1 (t; q))
The rotation number determines the number of winds around the equilibrium along
the interval [0, τ ] in the clockwise sense and has the following properties:
• rot(q, τ ) is a strictly increasing function of τ .
1640 ALFONSO RUIZ-HERRERA

• rot(q, τ ) ≶ m ⇐⇒ τ ≶ mP er(q) where P er(q) is the minimal period of the


solution (x1 (t; q), y1 (t; q)).
By applying Theorem 2 in [33], we obtain that the minimal period of (x1 (t; q),
y1 (t; q)) is a strictly increasing function with respect to the energy. Therefore using
that d1 > d∗1 , one can deduce that
   
l d2 l d2
P1 = P er α − , > P2 = P er α + , .
2 f2 2 f2
For convenience we introduce θ(τ, q) as the angular coordinate at time τ of the
solution of (4) departing from q ∈ A such that for all q ∈ A,
θ(τ, q) = θ(0, q) + 2πrot(q, τ ) ∈ [2πrot(q, τ ) − π/2, 2π rot(q, τ ) + π/2].
Now we are in a position to prove the aims of this step. Indeed, fix a constant T1
satisfying that
5P1 P2
T1 > = T1∗ .
P1 − P2
By the choice of T1 , there exist two integers m1 , m2 so that m1 ≤ PT11 , m2 ≥ PT12 and
m2 − m1 ≥ 3. From these inequalities, one can check that
rot(q, T1 ) ≤ m1 for all q ∈ A ∩ C1
rot(q, T1 ) ≥ m2 for all q ∈ A ∩ C2
where C1 = {(x, y) ∈ A : E(x, y) = E(α − l, df22 )} and C2 = {(x, y) ∈ A : E(x, y) =
E(α + l, df22 )}). Therefore, for k1 , k2 ∈ [m1 + 1, m2 − 1],
 
h π πi
2ki π − , 2ki π + ⊂ max θ(T1 , q), min θ(T1 , q) . (16)
2 2 q∈C1 q∈C2

After that, we define the sets


h π πi
K0 = {q ∈ A : θ(T1 , q) ∈ 2πk1 − , 2πk1 + ,
2 2
   
d2 d2
(x1 (T1 ; q), y1 (T1 ; q)) ∈ A, E α + l, ≤ E(q) ≤ E α − l, }
f2 f2
h π π i
K1 = {q ∈ A : θ(T1 , q) ∈ 2πk2 − , 2πk2 + ,
  2 2 
d2 d2
(x1 (T1 ; q), y1 (T1 ; q)) ∈ A, E α + l, ≤ E(q) ≤ E α − l, }.
f2 f2
To finish this step we estimate τ ∗ . Define
η = max{k(x1 (t; z)(−a1 + c1 y1 (t; z)), y1 (t; z)(d1 − e1 x1 (t; z)))k∞ : z ∈ A, t ∈ R},
3d22
   
3d2 7d2
R= := max{|y(d2 − f2 x)| : y ∈ 0, , x ∈ 0, },
2f2 2f2 4f2
d2 d2 1
τ ∗ = min{ , T1 , , T2 , }, (17)
4e2 R 4e2 η 6d2
we recall that (x1 (t; z), y1 (t; z)) denotes the solution of (4) with initial condition z.
In the sequel we fix a constant τ > 0 with 0 < τ < τ ∗ . Now, our main purpose is
d2
to prove that for m = 2f 2

(XK0 ,m , P ) : XA,m −→X


m A,m ,
(XK1 ,m , P ) : XA,m −→X
m A,m .
CHAOS IN DELAY DIFFERENTIAL EQUATIONS 1641

We deduce that by standard arguments P is a compact operator.

Step 3: Stretching property for Ψ.


First we notice that by (17), (specifically τ < min{ 4ed22η , T1 })
m
Ψ(XA,m ) ⊂ {(x, y) ∈ R × C([−τ, 0], R) : |y(t) − y(0)| ≤
}.
2
After this remark we prove the following stretching property: given a continuous
path
γ : [0, 1] −→ XA,m
with γ(0) ⊂ {(x, y) : x = α − l} and γ(1) ⊂ {(x, y) : x = α + l}, there exist two
subintervals [σ00 , σ10 ], [σ000 , σ100 ] verifying that
• γ([σ00 , σ10 ]) ⊂ XK0 ,m and γ([σ000 , σ100 ]) ⊂ XK1 ,m ,
• Ψ(γ([σ00 , σ10 ])), Ψ([γ(σ000 , σ100 ])) ⊂ XA,m ,
3d2
• Ψev (γ(σ00 )), Ψev (γ(σ000 )) ⊂ {(x, y) ∈ XA,m : y(0) = 2f2 },
d2
• Ψev (γ(σ10 )), Ψev (γ(σ100 )) ⊂ {(x, y) ∈ XA,m : y(0) = 2f2 }.
Indeed, take a path γ as above. Firstly we take a subinterval [Γ1 , Γ2 ] ⊂ [0, 1] with
γ([Γ1 , Γ2 ]) ⊂ {(x, y) ∈ XA,m : E(α + 2l , df22 ) ≤ E(x, y(0)) ≤ E(α − 2l , df22 )} and
 
l d2
γ(Γ1 ) ∈ {(x, y) ∈ XA,m : E(x, y(0)) = E α − , },
2 f2
 
l d2
γ(Γ2 ) ∈ {(x, y) ∈ XA,m : E(x, y(0)) = E α + , }.
2 f2
By continuity and using (16), we can find two subintervals [Γ00 , Γ01 ], [Γ000 , Γ001 ] satisfying
that
π π
− + 2k1 π ≤ θ(T1 , (x, y(0))) ≤ + 2k1 π for all (x, y) ∈ γ[Γ00 , Γ01 ],
2 2
π π
− + 2k2 π ≤ θ(T1 , (x, y(0))) ≤ + 2k2 π for all (x, y) ∈ γ[Γ000 , Γ001 ].
2 2
From these inequalities we obtain easily the desired subintervals [σ00 , σ10 ] ⊂ [Γ00 , Γ01 ]
and [σ000 , σ100 ] ⊂ [Γ000 , Γ001 ].

Step 4: Stretching property for Φ.


First we observe that
m
Φ(XA, m2 ) ⊂ {(x, y) ∈ R × C([−τ, 0], R) : |y(t) − y(0)| ≤ }. (18)
2
d2
Indeed, consider the equation (recall (17) and m = 2f2 )

y 0 (t) = y(t)(d2 − f2 y(t − τ )) (19)


and take y(t) a solution with initial condition in
 
d2 3d2 m
{y0 ∈ C([−τ, 0], R) : y0 (0) ∈ , , |y0 (t) − y0 (0)| ≤ }.
2f2 2f2 2
Clearly, y(t) can not be greater than df22 and increasing for all t ∈ [∆, ∞] with ∆ > 0.
Moreover, if y(0) = 3d 0
2f2 then y (0) < 0. By the previous comments and using the
2
1642 ALFONSO RUIZ-HERRERA

condition τ < 6d12 , one can prove that y(t) < 3d


2f2 for all t > 0. Indeed, assume by
2

contradiction that there exists t1 > 0 such that


3d2
t1 := min{t > 0 : y(t1 ) = }.
2f2
d2
Clearly, y 0 (t1 ) ≥ 0 and so y(t1 − τ ) ≤ f2 . Now easily one deduces that
3d2
y(t1 ) − y(t1 − τ ) ≤ τ d2 if t1 > τ
2f2

d2
y(t1 ) − y(t1 − τ ) ≤ |y(t1 ) − y(0)| + |y(t1 − τ ) − y(0)| ≤ if t1 < τ
2f2
and so y(t1 ) < 3d
2f2 . (Observe that y(t) > 0)
2

The previous property together with τ < min{T2 , 4fd22R } enables us to conclude (18).
d2
Notice that easily one can also obtain that y(t) > 2f 2
for all t > 0.
Next we prove the following stretching property: given a continuous path
γ : [0, 1] −→ XA,m
d2
with γ(0) ⊂ {(x, y) ∈ XA,m : y(0) = 2f 2
} and γ(1) ⊂ {(x, y) ∈ XA,m : y(0) =
3d2
2f2 }, there exists a subinterval [ζ1 , ζ2 ] ⊂ [0, 1] so that Φ(γ([ζ1 , ζ2 ])) ⊂ XA,m with
Φ(γ(ζ1 )) ⊂ {(x, y) ∈ XA,m : x = α − l} and Φ(γ(ζ2 )) ⊂ {(x, y) ∈ XA,m : x = α + l}.
d2
First of all we observe that Φ(γ([0, 1])) ⊂ {(x, y) ∈ R × C([−τ, 0], R) : 2f 2

3d2
y(0) ≤ f2 }. At this moment, it is enough to prove that Φ(γ(1)) ⊂ {(x, y) ∈
R × C([−τ, 0], R) : x > α + l} and Φ(γ(0)) ⊂ {(x, y) ∈ R × C([−τ, 0], R) : x < α − l}.
To see the first claim, we recall that by the definition of T2 , y(t, γ(1)) ⊂ [ 5d 2 3d2
4f2 , 2f2 ]
for all 0 < t < T2 . To finish the proof, we notice that by (10) and (11),
5d2 3d2
(α − l) + T2 min{x(−a2 − b2 x + c2 y) : (x, y) ∈ A ∩ { ≤y≤ }} > α + l.
4f2 2f2
The other claim is proved analogously.

To conclude the proof of Proposition 1, we apply Theorem 2.3 with XK0 ,m , XK1 ,m
and P .

Proof of Theorem 3.1. Notice that by the previous construction, the conditions
of Corollary 1 hold.

To finish this paper we illustrate how to apply our results in concrete examples.
Specifically we have to proceed in the following way:
• Fix the parameters in system (3) except d1 , T1 , T2 and τ and assume that (8)
holds.
• Take T2 , α, and l satisfying (4)-(13).
• Take d1 > d∗1 (see (15)).
• Compute P1 = P er(α − 2l , df22 ), P2 = P er(α + 2l , df22 ) and take T1 > P5P 1 P2
1 −P2
.

• Compute η and take 0 < τ < τ .
With these parameters we can use Proposition 1 and Theorem 3.1 . The compu-
tation of P1 = P er(α − 2l , df22 ), P2 = P er(α + 2l , df22 ) is not an easy task because
CHAOS IN DELAY DIFFERENTIAL EQUATIONS 1643

it involves an improper integral. However the reader can find properties and nu-
merical estimates in [4], [33] and [7]. Apart from the fourth step, the rest of the
parameters are easy to obtain.

Acknowledgments. The author is very grateful to two anonymous reviewers.


Their comments and insightful critique helped me very much to improve the paper.
REFERENCES

[1] B. Aulbach and B. Kieninger, On three definitions of chaos, Nonlinear Dyn. Syst. Theory, 1
(2001), 23–37.
[2] J. M. Cushing, Periodic time-dependent predator-prey systems, SIAM J. Appl. Math., 32
(1977), 82–95.
[3] T. Faria, Stability and bifurcation for a delayed predator-prey model and the effect of diffusion,
J. Math. Anal. Appl., 254 (2001), 433–463.
[4] J. Grasman and E. Veling, An asymptotic formula for the period of a Volterra-Lotka system,
Mathematical Biosciences, 18 (1973), 185–189.
[5] J. K. Hale and S. M. Tanaka, Square and pulse waves with two delays, J. Dynam. Differential
Equations, 12 (2000), 1–30.
[6] U. an der Heiden and H.-O. Walther, Existence of chaos in control systems with delayed
feedback , J. Differential Equations, 47 (1983), 273–295.
[7] S.-B. Hsu, A remark on the period of the periodic solution in the Lotka-Volterra system, J.
Math. Anal. Appl.,95 (1983), 428–436.
[8] S.-B. Hsu and X.-Q. Zhao, A Lotka-Volterra competition with seasonal sucession, J. Math.
Biol.
[9] A. Huppert, B. Blasius, R. Olinky and L. Stone, A model for seasonal phytoplankton blooms,
J. Theoret. Biol.,236 (2005), 276–290.
[10] M. Keeling, P. Rohani and B. T. Grenfell, Seasonally forced disease dynamics explored as
switching between attractors, Physica D, 148 (2001), 317–335.
[11] J. Kennedy, S. Koçak and J. A. Yorke, A chaos lemma, Amer. Math. Monthly, 108 (2001),
411–423.
[12] J. Kennedy and J. A. Yorke, Topological horseshoes, Trans. Amer. Math. Soc., 353 (2001),
2513–2530.
[13] U. Kirchgraber and D. Stoffer , On the definition of chaos, Z. Angew. Math. Mech., 69 (1989),
175–185.
[14] C. A. Klausmeier , Successional state dynamics: a novel approach to modeling nonequilibrium
foodweb dynamics, J. Theor. Biol., 262 (2010), 584–595.
[15] A. L. Koch , Coexistence resulting from an alternation of density dependent and density
independent growth, J. Theor. Biol., 44 (1974), 373–386.
[16] Y. Kuang, “Delay-differential Equations with Applications in Population Dynamics,” Aca-
demic, Boston, MA, 1993.
[17] Y. Kuang, Global stability in delay differential systems without dominanting instantaneous
negative feedbacks, J. Differential Equations, 119 (1995), 503–532.
[18] B. Lani-Wayda, Erratic solutions of simple delay equations, Trans. Amer. Math. Soc., 351
(1999), 901–945.
[19] B. Lani-Wayda and R. Srzednicki, A generalized Lefschetz fixed point theorem and symbolic
dynamics in delay equations, Ergodic Theory Dynam. Systems, 22 (2002), 1215–1232.
[20] B. Lani-Wayda and H.-O. Walther, Chaotic motion generated by delayed negative feedback.
I: A transversality criterion, Differential Integral Equations, 8 (1995), 1407–1452.
[21] B. Lani-Wayda and H.-O. Walther, Chaotic motion generated by delayed negative feedback.
II: Construction of nonlinearities, Math. Nachr., 180 (1996), 141–211.
[22] A. Leung, Conditions for global stability concerning a prey-predator model with delay effect,
SIAM J. Appl. Math., 36 (1979), 3602–3608.
[23] T. Y. Li and J. A. Yorke, Period three implies chaos, Amer. Math. Monthly, 82 (1975),
985–992.
[24] T. Malik and H. L. Smith, Does dormancy increase fitness of bacterial populations in time-
varying environments? , Bull. Math. Biol., 70 (2008), 1140–1162.
[25] R. May, Time-Delay Versus Stability in Population Models with Two and Three Trophic,
Ecology, 54 (1973), 315–325.
1644 ALFONSO RUIZ-HERRERA

[26] A. Medio, M. Pireddu and F. Zanolin, Chaotic dynamics for maps in one and two dimensions:
A geometrical method and applications to economics, Internat. J. Bifur. Chaos Appl. Sci.
Engrg., 19 (2009), 3283–3309.
[27] T. Namba and S. Takahashi, Competitive coexistence in a seasonally fluctuating environment
II: Multiple stable states and invasion succession, Theor. Popul. Biol., 44 (1995), 374–402.
[28] D. Papini and F. Zanolin, Some results on periodic points and chaotic dynamics arising from
the study of the nonlinear Hill equations, Rend. Semin. Mat. Univ. Politec. Torino, 65 (2007),
115–157.
[29] M. Pireddu and F. Zanolin, Cutting surfaces and applications to periodic points and chaotic-
like dynamics, Topol. Methods Nonlinear Anal., 30 (2007), 279–319.
[30] M. Pireddu and F. Zanolin, Fixed points for dissipative-repulsive systems and topological
dynamics of mappings defined on N -dimensional cells, Adv. Nonlinear Stud., 5 (2005), 411–
440.
[31] C. E. Steiner, A. S. Schwaderer, V. Huber, C. A. Klausmeier and E. Litchman, Periodi-
cally forced food chain dynamics: model predictions and experimental validation, Ecology, 90
(2009), 3099–3107.
[32] X. H. Tang and X. Zou, Global attractivity in a predator prey system with pure delay, Proc.
Edinb. Math. Soc., 51 (2008), 495–508.
[33] J. Waldvogel, The period in the Lotka-Volterra system is monotonic, J. Math. Anal. Appl.,
114 (1986), 178–184.
[34] H.-O. Walther, Homoclinic solution and chaos in ẋ(t) = f (x(t − 1)), Nonlinear Anal., 5
(1981), 775–788.
[35] H.-O. Walther, Hyperbolic periodic solutions, heteroclinic connections and transversal homo-
clinic points in autonomous differential delay equations, Mem. Amer. Math. Soc., 79 (1989)
iv+104 pp.
[36] K. Wójcik and P. Zgliczyński, Topological horseshoes and delay differential equations, Discrete
Contin. Dyn. Syst., 12 (2005), 827–852.
[37] P. Zgliczyński and M. Gidea, Covering relations for multidimensional dynamical systems, J.
Differential Equations, 202 (2004), 32–58.

Received October 2011; revised January 2012.


E-mail address: alfonsoruiz@ugr.es

View publication stats

You might also like