PRIMUS

Problems, Resources, and Issues in Mathematics Undergraduate

Studies

ISSN: 1051-1970 (Print) 1935-4053 (Online) Journal homepage: http://www.tandfonline.com/loi/upri20

Discrete population models: Why they belong in a

differential equations course

Zsuzsanna M. Kadas

To cite this article: Zsuzsanna M. Kadas (2018): Discrete population models: Why they belong in a
differential equations course, PRIMUS, DOI: 10.1080/10511970.2018.1443532

To link to this article: https://doi.org/10.1080/10511970.2018.1443532

View supplementary material

Accepted author version posted online: 14

Jun 2018.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=upri20
ACCEPTED MANUSCRIPT

DISCRETE POPULATION
MODELS:
WHY THEY BELONG IN A
DIFFERENTIAL EQUATIONS
COURSE

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

DISCRETE POPULATION MODELS:

WHY THEY BELONG IN A
DIFFERENTIAL EQUATIONS COURSE

Abstract: We make a case for including difference equations, in particular

the discrete logistic equation, in basic differential equations courses. Con-
trasting the behavior of discrete and continuous models enriches students’
understanding of both modeling and differential equations. To facilitate shar-
ing discrete population models with students, some materials including class
handouts and an “exploratory” computer assignment are provided.

Keywords: discrete population models, discrete logistic equation, dif-

ferential equations, difference equations, iterated functions

1 INTRODUCTION

Basic population dynamics models are now staples of the standard sopho-
more level ordinary differential equations (ODE) course, with the focus
on continuous models such as the logistic differential equation. However,
many such courses could be significantly enriched by also introducing
discrete time models, an important related concept rarely encountered
in the mathematics curriculum up to this point. An awareness of dis-
crete models is crucial for aspiring mathematicians, particularly in this
digital age. While continuous models such as the logistic equation offer
an engaging setting for introducing useful analytical methods and rais-

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

2 Z. Kadas

ing a variety of questions around modeling, continuous models are not

the whole story. With continuous models, students can practice solv-
ing ODE’s explicitly, compare qualitative analysis with explicit solution
methods, and consider the challenges of fitting curves and estimating
parameters from data – all pertinent and important topics to address
in the ODE course. In addition, this is also an ideal moment to ex-
pose students to discrete models and develop an awareness of several
key points:

• A continuous model is not appropriate for every application; many

real-world phenomena are more naturally modeled by discrete sys-
tems.
• Continuous and discrete models that appear to be the same may
not predict the same behavior.
• The wrong type of model fails to capture the underlying dynamics
of the system and can produce erroneous predictions.
• Discretization automatically happens when computer methods are
used, and the unintended effects of this “hidden” process may not
be recognized.

Furthermore, the specific characterizations of steady states, stability,

and periodic solutions look a bit different for continuous and discrete
models. Presenting both models in parallel is sound pedagogy since it
encourages students to stop and think about the underlying concepts,
rather than reflexively applying a method. While the ultimate goal for
the modeler is to incorporate parameters based on data, the intent of
this article is only to lay out a path for introducing students to discrete
models that highlights mathematical distinctions between the two types
and illustrates the necessity for weighing which type is appropriate in a
given situation.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 3

Some current texts include material on difference equations. For

example, a discretionary section on the topic is included in Boyce and
DiPrima’s classic text [2], and the final chapter of Blanchard, Devaney
and Hall’s text [4] is titled “Discrete Dynamical Systems”. Although
most other texts omit difference equation models or mention them only
peripherally, this article advocates for including discrete models in the
introductory ODE course to offer students an alternate framework for
modeling through which they gain important mathematical insight.
Focusing on the logistic model, the “Continuous vs. Discrete” story
can be told succinctly, as done in Section 2. This could serve as a stand-
alone single class presentation. The subsequent section describes how I
have incorporated this material into my ODE course to give students a
first experience with discrete models.

2 THE LOGISTIC EQUATION: DISCRETE VS. CONTIN-

UOUS

Most treatments of population dynamics begin with the exponential

growth model, well-known to virtually all calculus students. The dif-
ferential equation and solution are

dP
= rP and P (t) = P0 ert (1)
dt

and the familiar graph for r > 0 is shown in Figure 1. While this model
is useful for the initial growth phase of a population with abundant
resources, the rapid and unlimited growth predicted is unrealistic for
biological populations. Looking at the equation differently reveals the
difficulty.
1 dP
The per capita growth rate for a population is defined as P dt , and
rewriting the exponential model as

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

4 Z. Kadas

Figure 1. The exponential growth model: P (t) = ert for r > 0.

1 dP
=r (2)
P dt
shows that its per capita growth rate is constant. To use a population
of fish in a pond as an example, the population grows at the same rate
regardless of how many fish there are in the pond. However, we ex-
pect that the population will grow more slowly as the fishpond becomes
more crowded, so the constant r should be replaced with a “density de-
pendent” per capita growth rate g(P ) that decreases as the population
increases. The simplest option is to use a linear function g(P ) = a − bP
with a, b > 0.

Figure 2. Per capita growth rate (PCG) for the logistic equation decreases

linearly as function of P .

This gives the well-known logistic equation,

dP
= P · (a − b · P ) with a, b > 0 (3)
dt

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 5

also known as the Verhulst equation, first published in 1838 [1].

The equation is often rewritten as

dP
= r · P · (1 − P/K) (4)
dt

so that K can be interpreted as the carrying capacity of the population,

and r can be thought of as the approximate per capita growth rate for
small P . All solutions of this differential equation with positive initial
values P (0) approach K as t increases as can be seen in Figure 3.

Figure 3. Slope field and solution curves for the logistic ODE (Eqn.4); r = 0.5

and K = 300.

For population models, we may wish to think of time as advancing

in steps of a fixed size, either because the species being modeled only
reproduces at fixed intervals (e.g. polar bears reproduce once a year) or
because we only have data collected at fixed intervals. We may also be
solving the equation numerically, with algorithms that necessarily intro-
duce discrete time steps, whether we intend to or not. We discretize the
dP
logistic equation by replacing the derivative dt by the forward difference
P (t+∆t)−P (t)
quotient ∆t and rearrange to obtain
 
P (t)
P (t + ∆t) ≈ P (t) + ∆t · r · P (t) · 1 − . (5)
K

Labeling the time steps tn and the corresponding populations Pn for

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

6 Z. Kadas

n = 0, 1, 2, ..., gives
 
Pn
Pn+1 = Pn + ∆t · r · Pn · 1 − . (6)
K
A solution of the difference equation (6) is a sequence of values {Pn }, n =
0, 1, 2.... Specific solutions can be found simply by choosing an initial
value and repeatedly applying (6), that is, iterating the function that
produces Pn+1 .
Comparing solutions of the discrete equation with those of the con-
tinuous equation, we find some startling results. For some values of
the parameters, the discrete solution closely tracks the continuous solu-
tion as in Figure 4(a). Certainly the outcome is qualitatively the same,
with both solutions approaching the carrying capacity as a stable steady
state. However, if the growth rate r is increased, qualitatively different
solutions may appear. For example, for the parameter values chosen for
Figure 4(b) the discrete equation has a periodic solution with period
2. If r is increased further, a period-4 solution appears. And if r is
increased a bit more, an irregular-looking solution emerges; in fact, this
is an example of a chaotic solution.
Why can the discrete logistic equation exhibit oscillations while the
logistic differential equation cannot? As the continuous solution ap-
proaches a steady state, the growth (or decay) rate is instantaneously
recalibrated: the closer the population is to the steady state, the more
slowly it approaches that value, so it can never reach or cross over an
equilibrium value. (If it did, the rate of change would become zero, and
it would remain constant forever after.) But in the discrete model, the
growth rate is only recalibrated at each time step, so a growing popula-
tion can overshoot the steady state; at the next time step, the growth
rate will be negative, and an oscillation begins.
For purposes of qualitative analysis, the discrete logistic equation is
usually written in a simpler form with only one parameter. To achieve

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 7

(a)

(b)

(c)

(d)

Figure 4. Oscillatory solutions of discrete logistic equation (6).

Parameter values used are ∆t = 1 and K = 300 throughout, while r is varied:

(a) r = 0.5 (b) r = 2 (c) r = 2.5 (d) r = 2.8. Points of the discrete solu-

tion {Pn } are connected by line segments. For comparison, solutions of the

continuous logistic equation (4) with K = 300 and r = 0.5 are also plotted.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

8 Z. Kadas

this form, the discretized equation (6) can be rearranged and the con-
stants re-labeled as follows:
  
Pn
Pn+1 = Pn · 1 + ∆t · r · 1 −
K
 
∆t · r
= Pn · 1 + ∆t · r − · Pn
K (7)
= Pn · (A − B · Pn )
 
B
= A · Pn · 1 − · Pn .
A
A
Rescaling the equation by letting Pn = B · xn gives the usual (dimen-
sionless) form of the discrete logistic equation:

xn+1 = A · xn · (1 − xn ). (8)

Observe that this rescaled form shows that the full range of possible
behaviors of the discrete logistic equation can be obtained by varying
the single parameter A, a one-dimensional parameter space. (Varying
both parameters K and r in (6) produces no additional qualitatively
different solutions.)
Cobweb diagrams (shown in Fig. 5) illustrate the iteration process
and reveal the qualitative nature of the solution for different parameter
values. (A handout for students to draw these provides an engaging
kinesthetic activity!) As the parameter A is increased from a number
slightly larger than 1 to 4, a surprising array of behavior is observed.
In particular, for smaller values of A, the solution {xn } approaches the
steady state monotonically. As A is increased, the approach to equilib-
rium becomes oscillatory. Then, as A is increased through 3, the steady
state is no longer the attractor; instead the solution approaches a two-
point periodic solution. When A is increased further, a period-4 solution
can be observed – although producing this one on a hand-drawn sketch
takes some care. For values of A in the range from approximately 3.57 to

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 9

Figure 5. Cobweb diagrams for the logistic equation xn+1 = A · xn · (1 − xn ).

4 irregular-looking chaotic solutions appear. (This provides an opening

to make the startling observation that a deterministic process can give
rise to outcomes that appear random - yet another point of interest to
modelers!)
A summary of the behavior of the solutions as the parameter A ranges
over the values of interest is given in Table 1. (The transition values have
been established through a combination of analytic and numerical means
[5], [6].)
The discrete logistic equation became an icon of chaos theory in the
1970s. Robert May (Baron May of Oxford), perhaps the most deco-
rated population biologist of his era, highlighted this model in papers
published in the journals Science [5] and Nature [6]. The Nature article
reflects his strong feelings on the topic :

The review ends with an evangelical plea for the introduction of

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

10 Z. Kadas

0<A<1 x = 0 is the attractor

A−1
1<A<3 x= A is the attractor; approach is monotonic (smaller A) or oscillatory
3 < A < 3.449 2-point periodic solution is the attractor
3.449 < A < 3.544 4-point periodic solution is the attractor
3.544 < A < 3.564 8-point periodic solution is the attractor
A > 3.570 Chaotic behavior begins
A = 3.82 3-point periodic solution appears
A=4 Chaotic behavior ends

Table 1. Summary of qualitative behavior for discrete logistic equation. (In-

teger values given are exact values for A; other values are approximations.)

these difference equations into elementary mathematics courses,

so that students’ intuition may be enriched by seeing the wild
things that simple nonlinear equations can do.

and

Not only in research, but also in the everyday world of politics

and economics, we would all be better off if more people realised
that simple nonlinear systems do not necessarily possess simple
dynamical properties.

I would add: to construct models appropriate to different situations,

it is necessary to be aware that the “same” continuous and discrete
equations can have very different solutions, and that discretization can
introduce unexpected behavior. Tread carefully and deliberately from
the continuous world into the discrete world, and vice-versa.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 11

3 DIFFERENCE EQUATIONS IN THE ODE COURSE: AN

OUTLINE

In an ODE course, a unit on difference equations fits nicely at the conclu-

sion of the treatment of first order differential equations, once students
have some experience with solving linear and nonlinear equations, and
are familiar with slope fields, steady states, stability, and the continuous
logistic equation as a population model. If approximation methods such
as Euler’s or Heun’s Method have been covered, these also provide a link
to the idea of time advancing in discrete steps. I usually spend three to
four one-hour class meetings of my ODE course on difference equations,
roughly following this path:

• Introduce linear difference equations via compound interest on sav-

ings accounts; periodic deposits; paying off loans. Relate to popu-
lation models with immigration/emigration.
• Solve the general first order constant coefficient linear difference
equation (by summing a geometric series) and catalogue the variety
of possible behaviors.
• Introduce nonlinear difference equations.
• Use handouts for cobweb diagrams: xn+1 = a · xn 2 as first exam-
ple; then the discrete logistic equation.(See Appendix for handouts.)
Summarize behaviors observed: steady states, two-point periodic
solutions, 4-point periodic solutions.
• Discuss steady states and stability (qualitatively) in the context of
difference equations.
• Emphasize the idea that solutions of difference equations are se-
quences, and that explicit solutions (formulas for xn ) usually cannot
be found for nonlinear ones.
• Optionally, show the classic video Chaos, Fractals, and Dynamics

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

12 Z. Kadas

starring Robert Devaney [3]. It begins with function iteration and

goes all the way to Julia Sets and the Mandelbrot Set. Although
the technology is dated, Devaney makes the points very clearly.
Students complete a short-answer handout during and after the
video (see Appendix).
• Do a more careful study of the discrete logistic equation. Recap be-
havior as the parameter varies. Use a computer algebra system to
plot cobweb graphs and display solutions. (See Appendix for a sam-
ple worksheet in Maple and links to other cobweb applets.) Discuss
finding steady states and periodic solutions of various periods.
• Assign the “exploratory” homework project on iterated functions,
including the discrete logistic equation and at least one other ex-
ample, such as xn+1 = A · cos(xn ). (Sample provided in Appendix.)
• Derive the stability condition for steady state. Contrast with the
stability condition for steady states of single ordinary differential
equation.
• Briefly describe the discretization process.
• Contrast the behavior of the continuous and the discrete logistic
equations, as presented in Section 2.

In fact, the concept of discrete time models could be presented to

students much sooner than the ODE course. Without any reliance on
calculus, one could motivate discrete-time population models that pre-
dict Pn+1 as a function of Pn for a fixed time increment, and explore some
illustrative examples. (The cobweb handouts could readily be used for a
high school audience.) Another possibility is to include discrete popula-
tion models as recursively defined sequences in second-semester calculus
and so introduce a modeling component into the unit on sequences and
series to link the “abstract” mathematics to “real world” phenomena.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 13

4 PEDAGOGICAL CONSIDERATIONS: CONNECT and CON-

TRAST

Students expand their mathematical knowledge by connecting new ideas

to concepts they already know, contrasting the new ideas with the fa-
miliar ones, and so building their repertoire. The ODE course already
affords many opportunities both for using concepts and tools students
have previously learned in their calculus courses and for broadening their
horizons by linking these to new ideas. Including an introduction to dif-
ference equations and, in particular, telling the story of the discrete lo-
gistic equation adds considerable richness to such a course. Exposure to
this famous and memorable example makes an important general point
about modeling: that the model used must fit the situation, and that
simply using the most familiar tool at hand may produce erroneous re-
sults. It broadens the students’ toolboxes by giving them an alternate
way to conceive of a model.
Including a discrete interlude in the ODE course is also an opportu-
nity to change up the style of mathematics after a few weeks of solving
differential equations. Cobweb plots are new to most students, they are
geometrically appealing and fun to draw by hand. Furthermore, graphi-
cally performing the function iteration provides hands-on reinforcement
of the graphic-algebraic connection. The shift of mindset in thinking
about a steady state P ∗ for continuous and discrete models (namely
F 0 (P ∗ ) = 0 and F (P ∗ ) = P ∗ respectively) forces students to pay closer
attention to the actual concept. Likewise, comparing periodicity for the
continuous and discrete cases reinforces the underlying notion. Seeking
periodic solutions in the discrete case requires using function composi-
tion and solving equations like F (F (x)) = x. Usually a computer is
needed, and finding these solutions computationally always raises the
question “how do you know you’ve found them all?” when students

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

14 Z. Kadas

inevitably discover different 4-point periodic solutions. The idea of ex-

ploring the behavior of a model by doing computations is new to most
students, so pitfalls like roundoff error and the challenge of organizing
their conclusions (on a homework assignment) offer additional learning
opportunities.

Anticipating more advanced topics in differential equations and mod-

eling, exploration of the one-dimensional parameter space for the discrete
logistic and related equations serves as an introduction to the concept of
mapping out the parameter space of a model to determine all possible
qualitatively different behaviors. While qualitative analysis is usually
introduced for single first-order differential equations, the richer vari-
ety of behaviors of the discrete equations (including periodic solutions
of various periods) paves the way for the analysis of phase portraits of
systems of two linear differential equations, with their two-dimensional
parameter space (see for example [2] Section 9.1 ).

Finally, using as many calculus concepts and methods as possible is

another not-so-subtle subtext of my ODE course. Difference equations
make use of ideas students find less comfortable such as sequences, con-
vergence, and difference quotients. And finally, I seize the opportunity
to use Taylor series to derive the stability criteria for steady states –
again slightly different for discrete and continuous cases. In addition
to looking back and appreciating the utility of past concepts learned,
we also raise the curtain for a peek at topics to be explored in future
mathematical study: more rigorous studies of convergence and analy-
sis in general, bifurcations and chaos, and computational methods. So
the foray into difference equations becomes an opportunity to connect,
contrast, and build the students’ mathematical knowledge.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

Discrete Models in DE Courses 15

5 CONCLUSION

In my experience, one of the main things that students appreciate about

the ODE course is the applicability of the mathematics to “real life”
situations, even though these are often idealized for simplicity. So stu-
dents begin to think that differential equations are a good modeling tool
for every situation. Consequently it seems important to illustrate that
continuity might itself be an idealization that could lead to modeling
mistakes. If reproduction in a population is synchronized and essen-
tially occurs only once in a chosen time period (such as polar bears or
birds that reproduce once a year), a discrete model is more appropri-
ate than a continuous-time model. On the other hand, if reproduction
in a population is asynchronous, occurring on an ongoing basis, then a
continuous model is a reasonable idealization, especially if the popula-
tion is relatively large (like bacteria in a culture or a large number of
asynchronously breeding fish in a pond).
Awareness of only continuous models, but not their discrete coun-
terparts, may mislead the novice modeler to draw unwarranted conclu-
sions. Knowing whether a population can be expected to oscillate and
understanding the causes of the oscillation is often an important goal
of a population model. Since continuous single-species models cannot
exhibit oscillations, the modeler might assume that oscillations in an
observed population are due to some other factors such as a fluctuation
environment or a predator-prey interaction, when they might simply
be a result of the “overshoot” phenomenon described above. On the
other hand, a perfectly good continuous model that would ordinarily
produce monotonic behavior might display oscillations as a result of dis-
cretization. Awareness of these phenomena will make the modeler more
cautious about drawing conclusions concerning the source of an observed
behavior.

ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT

16 Z. Kadas

The bottom line: introducing discrete models in the introductory

ODE course makes sense pedagogically because it enriches students’ un-
derstanding of both discrete and continuous mathematics and makes
them more savvy modelers in this digital age.

REFERENCES

[1] Bacaer, N. 2011.A Short History of Mathematical Population Dy-

namics . London: Springer-Verlag.

[2] Boyce, R. and W. Di Prima 2005. Elementary Differential Equations

and Boundary Value Problems Hoboken, NJ: Wiley.

[3] Devaney, R.Chaos, Fractals, and Dynamics (Video) 1992.Science

Television Company.

[4] Blanchard, P., R. Devaney, and G. Hall 2012. Differential Equations,

Fourth Edition Boston, MA: Brooks/Cole

[5] May, R.M. 1974. Biological populations with nonoverlapping gener-

ations. Science 186:645-647.

[6] May, R.M.1976. Simple mathematical models with very complex dy-
namics. Nature 261: 459-467.

ACCEPTED MANUSCRIPT