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

net/publication/243764052

Delay Differential Equation with Application in Population Dynamics

Article · January 1993

CITATIONS READS
2,779 12,422

1 author:

Yang Kuang
Arizona State University
197 PUBLICATIONS   9,861 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Wang Research Group View project

All content following this page was uploaded by Yang Kuang on 27 February 2019.

The user has requested enhancement of the downloaded file.

 Delay Differential Equations

Yang Kuang

Arizona State University, U.S.A.

All processes take time to complete. While physical processes such as

acceleration and deceleration may not take much time compare to the times

needed to travel most distances, biological process times such as gestation periods

and maturation times can be substantial when compared to the data collection

times in most population studies. Therefore, it is often imperative to explicitly

incorporate these process times in mathematical models of population dynamics.

These process times are often called delay times and the models that incorporate

such delay times are referred as delay differential equation (DDE) models.

Recent theoretical and computational advancements in delay differential

equations reveal that DDEs are capable of generating rich and plausible dynamics

with realistic parameter values. Naturally occurring complex dynamics are often

naturally generated by well formulated DDE models. This is simply due to the

fact that a DDE operates on an infinite dimensional space consisting of

continuous functions which accommodates high dimensional dynamics. For

example, the Lotka-Volterra predator-prey model with crowding effect does not

produce sustainable oscillatory solutions that describe population cycles, yet the

Nicholson’s blowflies model can generate rich and complex dynamics.

I. The concepts and notations

1
 DDEs are differential equations in which the derivatives of some unknown

functions at present time are dependent on the values of the functions at previous

times. Mathematically, a general delay differential equation for x(t ) ∈ R n takes

dx(t )
the form = f (t , xt ) where xt (θ ) = x(t + θ ) and − τ ≤ θ ≤ 0 . Observe that
dt

xt (θ ) with − τ ≤ θ ≤ 0 represents a portion of the solution trajectory in a recent

past. Here f is a functional operator that takes the input of a time and a

dx(t )
continuous function xt (θ ) with − τ ≤ θ ≤ 0 and generates a real number ( ) as
dt

its output. A well known example of delay differential equation is the Hutchinson

equation, or the discrete delay logistic equation, x ' = rx(1- x(t − τ ) / K ) . Some

DDEs can be conveniently solved in a stepwise fashion. In fact, the Hutchinson

equation can be rewritten as (ln x) ' = r (1- x(t − τ ) / K ) which can be used to

solve for x for 0 ≤ t ≤ τ . Some DDEs, such as

0
x '(t ) = rx(t )[1- a ∫ e as x(t + s )ds / K ] , are in fact a system of ordinary diffential
−∞

equations (ODEs), in disguise. This can be seen by letting

0
y (t ) = ∫
−∞
e as x(t + s )ds and noticing that y ' = a ( x(t ) − y (t )) which yield a system

of ODEs x '(t ) = rx(t )(1- y (t ) / K ); y ' = a ( x(t ) − y (t )) . Indeed, an integro-

0
differential equation of the form x '(t ) = f (t , x(t )) + ∫ k ( s ) g (x(t + s ))ds with
−∞

initial condition xt (θ ) where − ∞ < θ ≤ 0 is equivalent to a system of ODEs with

initial condition if k is a linear combination of functions eat , te at , t 2 e at ,..., t m e at ,

where a is a real number and m is a positive integer. The method of reducing such

2
a delay differential equation into a system of ODEs is called the linear chain

trick.

Numerically solving most delay differential equations or systems is almost

as simple as solving ODEs. The popular MATLAB based dde23 solver developed

by Shampine and Thompson for delay differential equation is well tested and

user-friendly. Interested readers can find many familiar and informative examples

at the website http://www.radford.edu/thompson/webddes/ddetutwhite.html and

more sophisticated users can find additional information at

http://www.radford.edu/thompson/webddes/.

As with linear ODEs, stability properties of linear DDEs can be

characterized and analyzed by studying their characteristic equations. For an

example, the characteristic equation for x '(t ) = ax(t ) + bx(t − τ ) is

λ − a − be − λτ = 0 . The roots λ of the characteristic equation are called

characteristic roots. Notice that the root appears in the exponent of the last term in

the characteristic equation, causing the characteristic equation to possess infinite

number of roots. However, there are only a finite number of roots located to the

right of any vertical line in the complex plane.

II. Some characteristics of DDEs

In most applications of delay differential equations in sciences, the need of

incorporating time delays is often due to the presence of process times or the

existence of some stage structures. In engineering applications, such time delay is

often modeled via high dimensional compartment models. In life science

3
applications, compartmental models can present additional challenges of

estimating some of the involved parameter values. In such cases, low dimensional

delay differential models with fewer parameters can be sensible alternatives.

Since the through-stage survival rate is often a function of such time

delays, it is easy to conceive that these models may involve some delay dependent

parameters. The ubiquitous presence of such parameters often greatly complicates

the task of a systematic study of such models. In some special cases, the stability

of a given steady state can be determined by the graphs of some functions of time

delay which can be expressed explicitly and thus can be depicted. The common

scenario is that as time delay increases, stability changes from stable to unstable

to stable, implying that a large delay can be stabilizing. This scenario often

contradicts the one provided by similar models with only delay independent

parameters.

In addition, a closer look at the cause of a time delay often suggests that

the time delay itself maybe dependent on some key model variables. In short, the

delays are state dependent. These state dependent delay differential equations are

notoriously difficult to study mathematically. However, they may possess some

surprising and more plausible dynamics.

III. Some simple delay differential equation models

Many consumer species go through two or more life stages as they

4
proceed from birth to death. In order to capture the oscillatory behavior often

observed in nature, various models are proposed. They include many difference

models and delay differential models. The Hutchinson equation

(1) x ' = rx(1- x(t − τ ) / K ) .

and its various variations, are among the ones that are most frequently employed

in theoretical ecology models. In (1), r is the growth rate, K is the carrying

capacity, and τ is a time delay that may have no real biological meaning. Like

logistic equation, these models are ad hoc and hence can be misleading. Indeed,

they produces artificially complex dynamics such as excessive volatility and huge

peak to valley ratios (Figure 1).

Solutions of x′=x(1−x(t−τ)) with τ=1, 3

10
τ=1
τ=3
9

x(t)
5

0
0 10 20 30 40 50 60 70 80 90 100
time t

5
 On the other hand, if we assume the adults have a constant birth rate of r

and the newborns mature in τ units of time, and the mortality rate is proportional

to the adult population density, then the following model may be a reasonable

model for the adult population

(2) x ' = rx(t - τ ) e− mτ - rx 2 / K .

However, the positive steady state of model (2) is always (globally) stable, similar

to the case when the delay is zero. On the other hand, the well known Nicholson’s

blowflies model

(3) x ' = px(t - τ ) e− ax (t −τ ) - mx

exhibits plausible and rich dynamics. In other words, the dynamics of delay

differential equations are extremely sensitive to model forms.

To further support the above statement, let us now examine some

predator-prey models with age structure. We assume that the prey or the

renewable resource, denoted by x, can be modeled by a logistic equation when the

consumer is absent. The predators or consumers is divided into two age groups:

juveniles and adults and they are denoted by y j and y respectively below. Only

adult predators are capable of preying on the prey species and that the juvenile

predators live on other resources. We have the following two-stage predator-prey

interaction model :

− d jτ 2
(4) x ' = rx(1 − x / K ) − yp ( x), y ' = be y (t − τ ) p ( x(t − τ )) − d a y − my .

With the aid of the geometric stability switch criteria that was specifically

developed to deal with models with delay dependent parameters, we can show

6
that this model generates increasingly more complex dynamics as its

characteristic equation produces more roots with positive real part when we

increase the time delay from 0.25 to 25 (Figure 2).

τ=0.25 τ=5
3 6
prey prey
preypredator preypredator
2.5 5
predator predator
2 4
x, y

x, y
1.5 3

1 2

0.5 1

0 0
0 100 200 300 0 100 200 300

τ=15 τ=25
6 6
prey prey
preypredator predator
5 5
predator
4 4
x, y

x, y

3 3

2 2

1 1

0 0
0 200 400 600 800 0 200 400 600 800
time t time t

If we assume that the maturation time delay in population dynamics is

determined by the resource uptake, then we may have

0
(5) ∫ τ p( x(s))ds = M ,
−

for some positive constant M which measures the resource requirement for a

newborn to mature. With this additional reality, solutions of model (4) tend to a

steady state or a limit cycle dynamics. In addition, the time to approach the limit

cycle is much shorter than a typical model without time delay or with constant

7
time delay, suggesting that the more realistic formulation of time delay (5)

satisfactorily describes the often observed short duration of transition dynamics

in nature (Figure 3).

3.5

2.5

2
y

1.5

0.5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
x

Delay differential equation models can be more effective and accurate

compare to ordinary differential equation based models when it is necessary to

capture oscillatory dynamics with specific periods and amplitudes. This

characteristic is successfully employed to explain why lemmings often have a 4-

year cycle whereas snowshoe hares have a 10-year cycle, and why the putative

cycles of the moose–wolf interactions on Isle Royale, Michigan, is 38-year long.

8
In addition, some simple and plausible models with two time delays can generate

the ubiquitous ultradian insulin secretory oscillations in the human glucose–

insulin regulatory system.

See Also the Following Articles

Difference equations
Ordinary differential equations
Partial differential equations
Integro-difference equations

Glossary

gestation period is the time in which a fetus develops, beginning with fertilization
and ending at birth. The duration of this period varies between species.

initial conditions refer to a set of initial dependent state variable values that must
be prescribed in order to identify the specific solution that that satisfies a given
differential system or a delay differential equation.

kernel is a weight function that describes the relative impacts of the past values of
some dependent variables on the current rate of change of a dependent variable.

linear chain trick is a method of transforming a system of DDEs to a larger

system of ODEs. This method can only be applied to systems of DDEs that
employ only a special classes of kernels.

population cycles are regular fluctuations in the size of a population, as seen in

some lemmings and snowshoe hare, for examples. Such cycles maybe caused by
density-dependent mortality which increases due to overcrowding induced food

9
shortage resulting dramatic decline in the population, the population then
gradually increase again.

transcendental functions are functions that do not satisfy polynomial equations

whose coefficients are themselves polynomials, in contrast to an algebraic
function, which does satisfy such an equation. Examples of transcendental
functions include the exponential function, the logarithm, and the trigonometric
functions.

Further Reading

W. G. Aiello and H. I. Freedman, 1990. A time-delay model of single-species

growth with stage structure, Math. Biosci. 101, 139--153.

E. Beretta and Y. Kuang, 2002. Geometric stability switch criteria in delay

differential systems with delay dependent parameters. SIAM J. Math. Anal.,

33, 1144-1165.

S. A. Gourley and Y. Kuang, 2004. A stage structured predator-prey model and its

dependence on maturation delay and death rate, J. Math. Biol., 49, 188-200.

S. A. Gourley and Y. Kuang, 2005. A Delay reaction-diffusion model of the

spread of bacteriophage infection, SIAM J. Appl. Math., 65, 550-566.

W. S. C. Gurney, S. P. Blythe and R. M. Nisbet, 1980. Nicholson's blowflies

revisited, Nature, 287, 17 – 21.

J. K. Hale and S. M. Verduyn Lunel, 1993. Introduction to Functional Differential

Equations, Springer-Verlag.

G. E. Hutchinson, 1948. Circular causal systems in ecology, Ann. N.Y. Acad. Sci.

50, 221-246.

10
Y. Kuang, 1993. Delay Differential Equations with Applications in Population

Dynamics, Academic Press, Boston.

J. Li, Y. Kuang and C. Mason, 2006. Modeling the glucose-insulin regulatory

system and ultradian insulin secretory oscillations with two time delays, J. of

Theor. Biol., 242, 722-735.

H. L. Smith, 2011. An introduction to delay differential equations with

applications to the life sciences. Texts in Applied Mathematics. Springer,

Berlin.

11
 Figure Captions

Figure 1. Solutions of the Hutchinson equation with delay values of 1 and 3.

Notice the peak to valley ratio is well over 2000 when the delay is 3.

Figure 2. A solution of model (4) with p(x) = px, where r=K=1, p = 1, b = 10,

d j = 0, d a = 0:5, m = 0:1 and τ varies from 0.25 to 25.

Figure 3. Maturation time delay does not generate complex dynamics other than

periodic solution. In addition, maturation time delay may significantly cut the

transition time from an initial point to an attracting limit cycle.

12

View publication stats